xref: /freebsd/sys/net/route.c (revision c078c3fd69ead23e24c8328ccf05842f18ab3c2f)
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_ctl.h>
65 #include <net/route/route_var.h>
66 #include <net/route/nhop.h>
67 #include <net/route/shared.h>
68 #include <net/vnet.h>
69 
70 #ifdef RADIX_MPATH
71 #include <net/radix_mpath.h>
72 #endif
73 
74 #include <netinet/in.h>
75 #include <netinet/ip_mroute.h>
76 
77 /*
78  * By default add routes to all fibs for new interfaces.
79  * Once this is set to 0 then only allocate routes on interface
80  * changes for the FIB of the caller when adding a new set of addresses
81  * to an interface.  XXX this is a shotgun aproach to a problem that needs
82  * a more fine grained solution.. that will come.
83  * XXX also has the problems getting the FIB from curthread which will not
84  * always work given the fib can be overridden and prefixes can be added
85  * from the network stack context.
86  */
87 VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1;
88 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET,
89     &VNET_NAME(rt_add_addr_allfibs), 0, "");
90 
91 VNET_PCPUSTAT_DEFINE(struct rtstat, rtstat);
92 
93 VNET_PCPUSTAT_SYSINIT(rtstat);
94 #ifdef VIMAGE
95 VNET_PCPUSTAT_SYSUNINIT(rtstat);
96 #endif
97 
98 EVENTHANDLER_LIST_DEFINE(rt_addrmsg);
99 
100 static int rt_ifdelroute(const struct rtentry *rt, const struct nhop_object *,
101     void *arg);
102 static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info,
103     int flags);
104 
105 /*
106  * route initialization must occur before ip6_init2(), which happenas at
107  * SI_ORDER_MIDDLE.
108  */
109 static void
110 route_init(void)
111 {
112 
113 	nhops_init();
114 }
115 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, NULL);
116 
117 struct rib_head *
118 rt_table_init(int offset, int family, u_int fibnum)
119 {
120 	struct rib_head *rh;
121 
122 	rh = malloc(sizeof(struct rib_head), M_RTABLE, M_WAITOK | M_ZERO);
123 
124 	/* TODO: These details should be hidded inside radix.c */
125 	/* Init masks tree */
126 	rn_inithead_internal(&rh->head, rh->rnh_nodes, offset);
127 	rn_inithead_internal(&rh->rmhead.head, rh->rmhead.mask_nodes, 0);
128 	rh->head.rnh_masks = &rh->rmhead;
129 
130 	/* Save metadata associated with this routing table. */
131 	rh->rib_family = family;
132 	rh->rib_fibnum = fibnum;
133 #ifdef VIMAGE
134 	rh->rib_vnet = curvnet;
135 #endif
136 
137 	tmproutes_init(rh);
138 
139 	/* Init locks */
140 	RIB_LOCK_INIT(rh);
141 
142 	nhops_init_rib(rh);
143 
144 	/* Init subscription system */
145 	rib_init_subscriptions(rh);
146 
147 	/* Finally, set base callbacks */
148 	rh->rnh_addaddr = rn_addroute;
149 	rh->rnh_deladdr = rn_delete;
150 	rh->rnh_matchaddr = rn_match;
151 	rh->rnh_lookup = rn_lookup;
152 	rh->rnh_walktree = rn_walktree;
153 	rh->rnh_walktree_from = rn_walktree_from;
154 
155 	return (rh);
156 }
157 
158 static int
159 rt_freeentry(struct radix_node *rn, void *arg)
160 {
161 	struct radix_head * const rnh = arg;
162 	struct radix_node *x;
163 
164 	x = (struct radix_node *)rn_delete(rn + 2, NULL, rnh);
165 	if (x != NULL)
166 		R_Free(x);
167 	return (0);
168 }
169 
170 void
171 rt_table_destroy(struct rib_head *rh)
172 {
173 
174 	tmproutes_destroy(rh);
175 
176 	rn_walktree(&rh->rmhead.head, rt_freeentry, &rh->rmhead.head);
177 
178 	nhops_destroy_rib(rh);
179 
180 	rib_destroy_subscriptions(rh);
181 
182 	/* Assume table is already empty */
183 	RIB_LOCK_DESTROY(rh);
184 	free(rh, M_RTABLE);
185 }
186 
187 /*
188  * Adds a temporal redirect entry to the routing table.
189  * @fibnum: fib number
190  * @dst: destination to install redirect to
191  * @gateway: gateway to go via
192  * @author: sockaddr of originating router, can be NULL
193  * @ifp: interface to use for the redirected route
194  * @flags: set of flags to add. Allowed: RTF_GATEWAY
195  * @lifetime_sec: time in seconds to expire this redirect.
196  *
197  * Retuns 0 on success, errno otherwise.
198  */
199 int
200 rib_add_redirect(u_int fibnum, struct sockaddr *dst, struct sockaddr *gateway,
201     struct sockaddr *author, struct ifnet *ifp, int flags, int lifetime_sec)
202 {
203 	struct rib_cmd_info rc;
204 	int error;
205 	struct rt_addrinfo info;
206 	struct rt_metrics rti_rmx;
207 	struct ifaddr *ifa;
208 
209 	NET_EPOCH_ASSERT();
210 
211 	if (rt_tables_get_rnh(fibnum, dst->sa_family) == NULL)
212 		return (EAFNOSUPPORT);
213 
214 	/* Verify the allowed flag mask. */
215 	KASSERT(((flags & ~(RTF_GATEWAY)) == 0),
216 	    ("invalid redirect flags: %x", flags));
217 
218 	/* Get the best ifa for the given interface and gateway. */
219 	if ((ifa = ifaof_ifpforaddr(gateway, ifp)) == NULL)
220 		return (ENETUNREACH);
221 	ifa_ref(ifa);
222 
223 	bzero(&info, sizeof(info));
224 	info.rti_info[RTAX_DST] = dst;
225 	info.rti_info[RTAX_GATEWAY] = gateway;
226 	info.rti_ifa = ifa;
227 	info.rti_ifp = ifp;
228 	info.rti_flags = flags | RTF_HOST | RTF_DYNAMIC;
229 
230 	/* Setup route metrics to define expire time. */
231 	bzero(&rti_rmx, sizeof(rti_rmx));
232 	/* Set expire time as absolute. */
233 	rti_rmx.rmx_expire = lifetime_sec + time_second;
234 	info.rti_mflags |= RTV_EXPIRE;
235 	info.rti_rmx = &rti_rmx;
236 
237 	error = rib_action(fibnum, RTM_ADD, &info, &rc);
238 	ifa_free(ifa);
239 
240 	if (error != 0) {
241 		/* TODO: add per-fib redirect stats. */
242 		return (error);
243 	}
244 
245 	RT_LOCK(rc.rc_rt);
246 	flags = rc.rc_rt->rt_flags;
247 	RT_UNLOCK(rc.rc_rt);
248 
249 	RTSTAT_INC(rts_dynamic);
250 
251 	/* Send notification of a route addition to userland. */
252 	bzero(&info, sizeof(info));
253 	info.rti_info[RTAX_DST] = dst;
254 	info.rti_info[RTAX_GATEWAY] = gateway;
255 	info.rti_info[RTAX_AUTHOR] = author;
256 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
257 
258 	return (0);
259 }
260 
261 /*
262  * Routing table ioctl interface.
263  */
264 int
265 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
266 {
267 
268 	/*
269 	 * If more ioctl commands are added here, make sure the proper
270 	 * super-user checks are being performed because it is possible for
271 	 * prison-root to make it this far if raw sockets have been enabled
272 	 * in jails.
273 	 */
274 #ifdef INET
275 	/* Multicast goop, grrr... */
276 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
277 #else /* INET */
278 	return ENXIO;
279 #endif /* INET */
280 }
281 
282 struct ifaddr *
283 ifa_ifwithroute(int flags, const struct sockaddr *dst,
284     const struct sockaddr *gateway, u_int fibnum)
285 {
286 	struct ifaddr *ifa;
287 
288 	NET_EPOCH_ASSERT();
289 	if ((flags & RTF_GATEWAY) == 0) {
290 		/*
291 		 * If we are adding a route to an interface,
292 		 * and the interface is a pt to pt link
293 		 * we should search for the destination
294 		 * as our clue to the interface.  Otherwise
295 		 * we can use the local address.
296 		 */
297 		ifa = NULL;
298 		if (flags & RTF_HOST)
299 			ifa = ifa_ifwithdstaddr(dst, fibnum);
300 		if (ifa == NULL)
301 			ifa = ifa_ifwithaddr(gateway);
302 	} else {
303 		/*
304 		 * If we are adding a route to a remote net
305 		 * or host, the gateway may still be on the
306 		 * other end of a pt to pt link.
307 		 */
308 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
309 	}
310 	if (ifa == NULL)
311 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
312 	if (ifa == NULL) {
313 		struct nhop_object *nh;
314 
315 		nh = rib_lookup(fibnum, gateway, NHR_NONE, 0);
316 
317 		/*
318 		 * dismiss a gateway that is reachable only
319 		 * through the default router
320 		 */
321 		if ((nh == NULL) || (nh->nh_flags & NHF_DEFAULT))
322 			return (NULL);
323 		ifa = nh->nh_ifa;
324 	}
325 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
326 		struct ifaddr *oifa = ifa;
327 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
328 		if (ifa == NULL)
329 			ifa = oifa;
330 	}
331 
332 	return (ifa);
333 }
334 
335 
336 /*
337  * Copy most of @rt data into @info.
338  *
339  * If @flags contains NHR_COPY, copies dst,netmask and gw to the
340  * pointers specified by @info structure. Assume such pointers
341  * are zeroed sockaddr-like structures with sa_len field initialized
342  * to reflect size of the provided buffer. if no NHR_COPY is specified,
343  * point dst,netmask and gw @info fields to appropriate @rt values.
344  *
345  * if @flags contains NHR_REF, do refcouting on rt_ifp and rt_ifa.
346  *
347  * Returns 0 on success.
348  */
349 int
350 rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags)
351 {
352 	struct rt_metrics *rmx;
353 	struct sockaddr *src, *dst;
354 	struct nhop_object *nh;
355 	int sa_len;
356 
357 	if (flags & NHR_COPY) {
358 		/* Copy destination if dst is non-zero */
359 		src = rt_key(rt);
360 		dst = info->rti_info[RTAX_DST];
361 		sa_len = src->sa_len;
362 		if (dst != NULL) {
363 			if (src->sa_len > dst->sa_len)
364 				return (ENOMEM);
365 			memcpy(dst, src, src->sa_len);
366 			info->rti_addrs |= RTA_DST;
367 		}
368 
369 		/* Copy mask if set && dst is non-zero */
370 		src = rt_mask(rt);
371 		dst = info->rti_info[RTAX_NETMASK];
372 		if (src != NULL && dst != NULL) {
373 
374 			/*
375 			 * Radix stores different value in sa_len,
376 			 * assume rt_mask() to have the same length
377 			 * as rt_key()
378 			 */
379 			if (sa_len > dst->sa_len)
380 				return (ENOMEM);
381 			memcpy(dst, src, src->sa_len);
382 			info->rti_addrs |= RTA_NETMASK;
383 		}
384 
385 		/* Copy gateway is set && dst is non-zero */
386 		src = &rt->rt_nhop->gw_sa;
387 		dst = info->rti_info[RTAX_GATEWAY];
388 		if ((rt->rt_flags & RTF_GATEWAY) && src != NULL && dst != NULL){
389 			if (src->sa_len > dst->sa_len)
390 				return (ENOMEM);
391 			memcpy(dst, src, src->sa_len);
392 			info->rti_addrs |= RTA_GATEWAY;
393 		}
394 	} else {
395 		info->rti_info[RTAX_DST] = rt_key(rt);
396 		info->rti_addrs |= RTA_DST;
397 		if (rt_mask(rt) != NULL) {
398 			info->rti_info[RTAX_NETMASK] = rt_mask(rt);
399 			info->rti_addrs |= RTA_NETMASK;
400 		}
401 		if (rt->rt_flags & RTF_GATEWAY) {
402 			info->rti_info[RTAX_GATEWAY] = &rt->rt_nhop->gw_sa;
403 			info->rti_addrs |= RTA_GATEWAY;
404 		}
405 	}
406 
407 	nh = rt->rt_nhop;
408 	rmx = info->rti_rmx;
409 	if (rmx != NULL) {
410 		info->rti_mflags |= RTV_MTU;
411 		rmx->rmx_mtu = nh->nh_mtu;
412 	}
413 
414 	info->rti_flags = rt->rt_flags | nhop_get_rtflags(nh);
415 	info->rti_ifp = nh->nh_ifp;
416 	info->rti_ifa = nh->nh_ifa;
417 	if (flags & NHR_REF) {
418 		if_ref(info->rti_ifp);
419 		ifa_ref(info->rti_ifa);
420 	}
421 
422 	return (0);
423 }
424 
425 /*
426  * Lookups up route entry for @dst in RIB database for fib @fibnum.
427  * Exports entry data to @info using rt_exportinfo().
428  *
429  * If @flags contains NHR_REF, refcouting is performed on rt_ifp and rt_ifa.
430  * All references can be released later by calling rib_free_info().
431  *
432  * Returns 0 on success.
433  * Returns ENOENT for lookup failure, ENOMEM for export failure.
434  */
435 int
436 rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags,
437     uint32_t flowid, struct rt_addrinfo *info)
438 {
439 	RIB_RLOCK_TRACKER;
440 	struct rib_head *rh;
441 	struct radix_node *rn;
442 	struct rtentry *rt;
443 	int error;
444 
445 	KASSERT((fibnum < rt_numfibs), ("rib_lookup_rte: bad fibnum"));
446 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
447 	if (rh == NULL)
448 		return (ENOENT);
449 
450 	RIB_RLOCK(rh);
451 	rn = rh->rnh_matchaddr(__DECONST(void *, dst), &rh->head);
452 	if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) {
453 		rt = RNTORT(rn);
454 		/* Ensure route & ifp is UP */
455 		if (RT_LINK_IS_UP(rt->rt_nhop->nh_ifp)) {
456 			flags = (flags & NHR_REF) | NHR_COPY;
457 			error = rt_exportinfo(rt, info, flags);
458 			RIB_RUNLOCK(rh);
459 
460 			return (error);
461 		}
462 	}
463 	RIB_RUNLOCK(rh);
464 
465 	return (ENOENT);
466 }
467 
468 /*
469  * Releases all references acquired by rib_lookup_info() when
470  * called with NHR_REF flags.
471  */
472 void
473 rib_free_info(struct rt_addrinfo *info)
474 {
475 
476 	ifa_free(info->rti_ifa);
477 	if_rele(info->rti_ifp);
478 }
479 
480 /*
481  * Iterates over all existing fibs in system calling
482  *  @setwa_f function prior to traversing each fib.
483  *  Calls @wa_f function for each element in current fib.
484  * If af is not AF_UNSPEC, iterates over fibs in particular
485  * address family.
486  */
487 void
488 rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f,
489     void *arg)
490 {
491 	struct rib_head *rnh;
492 	uint32_t fibnum;
493 	int i;
494 
495 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
496 		/* Do we want some specific family? */
497 		if (af != AF_UNSPEC) {
498 			rnh = rt_tables_get_rnh(fibnum, af);
499 			if (rnh == NULL)
500 				continue;
501 			if (setwa_f != NULL)
502 				setwa_f(rnh, fibnum, af, arg);
503 
504 			RIB_WLOCK(rnh);
505 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
506 			RIB_WUNLOCK(rnh);
507 			continue;
508 		}
509 
510 		for (i = 1; i <= AF_MAX; i++) {
511 			rnh = rt_tables_get_rnh(fibnum, i);
512 			if (rnh == NULL)
513 				continue;
514 			if (setwa_f != NULL)
515 				setwa_f(rnh, fibnum, i, arg);
516 
517 			RIB_WLOCK(rnh);
518 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
519 			RIB_WUNLOCK(rnh);
520 		}
521 	}
522 }
523 
524 /*
525  * Iterates over all existing fibs in system and deletes each element
526  *  for which @filter_f function returns non-zero value.
527  * If @family is not AF_UNSPEC, iterates over fibs in particular
528  * address family.
529  */
530 void
531 rt_foreach_fib_walk_del(int family, rt_filter_f_t *filter_f, void *arg)
532 {
533 	u_int fibnum;
534 	int i, start, end;
535 
536 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
537 		/* Do we want some specific family? */
538 		if (family != AF_UNSPEC) {
539 			start = family;
540 			end = family;
541 		} else {
542 			start = 1;
543 			end = AF_MAX;
544 		}
545 
546 		for (i = start; i <= end; i++) {
547 			if (rt_tables_get_rnh(fibnum, i) == NULL)
548 				continue;
549 
550 			rib_walk_del(fibnum, i, filter_f, arg, 0);
551 		}
552 	}
553 }
554 
555 /*
556  * Delete Routes for a Network Interface
557  *
558  * Called for each routing entry via the rnh->rnh_walktree() call above
559  * to delete all route entries referencing a detaching network interface.
560  *
561  * Arguments:
562  *	rt	pointer to rtentry
563  *	nh	pointer to nhop
564  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
565  *
566  * Returns:
567  *	0	successful
568  *	errno	failed - reason indicated
569  */
570 static int
571 rt_ifdelroute(const struct rtentry *rt, const struct nhop_object *nh, void *arg)
572 {
573 	struct ifnet	*ifp = arg;
574 
575 	if (nh->nh_ifp != ifp)
576 		return (0);
577 
578 	/*
579 	 * Protect (sorta) against walktree recursion problems
580 	 * with cloned routes
581 	 */
582 	if ((rt->rt_flags & RTF_UP) == 0)
583 		return (0);
584 
585 	return (1);
586 }
587 
588 /*
589  * Delete all remaining routes using this interface
590  * Unfortuneatly the only way to do this is to slog through
591  * the entire routing table looking for routes which point
592  * to this interface...oh well...
593  */
594 void
595 rt_flushifroutes_af(struct ifnet *ifp, int af)
596 {
597 	KASSERT((af >= 1 && af <= AF_MAX), ("%s: af %d not >= 1 and <= %d",
598 	    __func__, af, AF_MAX));
599 
600 	rt_foreach_fib_walk_del(af, rt_ifdelroute, ifp);
601 }
602 
603 void
604 rt_flushifroutes(struct ifnet *ifp)
605 {
606 
607 	rt_foreach_fib_walk_del(AF_UNSPEC, rt_ifdelroute, ifp);
608 }
609 
610 /*
611  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
612  * it will be referenced so the caller must free it.
613  *
614  * Assume basic consistency checks are executed by callers:
615  * RTAX_DST exists, if RTF_GATEWAY is set, RTAX_GATEWAY exists as well.
616  */
617 int
618 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
619 {
620 	const struct sockaddr *dst, *gateway, *ifpaddr, *ifaaddr;
621 	struct epoch_tracker et;
622 	int needref, error, flags;
623 
624 	dst = info->rti_info[RTAX_DST];
625 	gateway = info->rti_info[RTAX_GATEWAY];
626 	ifpaddr = info->rti_info[RTAX_IFP];
627 	ifaaddr = info->rti_info[RTAX_IFA];
628 	flags = info->rti_flags;
629 
630 	/*
631 	 * ifp may be specified by sockaddr_dl
632 	 * when protocol address is ambiguous.
633 	 */
634 	error = 0;
635 	needref = (info->rti_ifa == NULL);
636 	NET_EPOCH_ENTER(et);
637 
638 	/* If we have interface specified by the ifindex in the address, use it */
639 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
640 	    ifpaddr->sa_family == AF_LINK) {
641 	    const struct sockaddr_dl *sdl = (const struct sockaddr_dl *)ifpaddr;
642 	    if (sdl->sdl_index != 0)
643 		    info->rti_ifp = ifnet_byindex(sdl->sdl_index);
644 	}
645 	/*
646 	 * If we have source address specified, try to find it
647 	 * TODO: avoid enumerating all ifas on all interfaces.
648 	 */
649 	if (info->rti_ifa == NULL && ifaaddr != NULL)
650 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
651 	if (info->rti_ifa == NULL) {
652 		const struct sockaddr *sa;
653 
654 		/*
655 		 * Most common use case for the userland-supplied routes.
656 		 *
657 		 * Choose sockaddr to select ifa.
658 		 * -- if ifp is set --
659 		 * Order of preference:
660 		 * 1) IFA address
661 		 * 2) gateway address
662 		 *   Note: for interface routes link-level gateway address
663 		 *     is specified to indicate the interface index without
664 		 *     specifying RTF_GATEWAY. In this case, ignore gateway
665 		 *   Note: gateway AF may be different from dst AF. In this case,
666 		 *   ignore gateway
667 		 * 3) final destination.
668 		 * 4) if all of these fails, try to get at least link-level ifa.
669 		 * -- else --
670 		 * try to lookup gateway or dst in the routing table to get ifa
671 		 */
672 		if (info->rti_info[RTAX_IFA] != NULL)
673 			sa = info->rti_info[RTAX_IFA];
674 		else if ((info->rti_flags & RTF_GATEWAY) != 0 &&
675 		    gateway->sa_family == dst->sa_family)
676 			sa = gateway;
677 		else
678 			sa = dst;
679 		if (info->rti_ifp != NULL) {
680 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
681 			/* Case 4 */
682 			if (info->rti_ifa == NULL && gateway != NULL)
683 				info->rti_ifa = ifaof_ifpforaddr(gateway, info->rti_ifp);
684 		} else if (dst != NULL && gateway != NULL)
685 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
686 							fibnum);
687 		else if (sa != NULL)
688 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
689 							fibnum);
690 	}
691 	if (needref && info->rti_ifa != NULL) {
692 		if (info->rti_ifp == NULL)
693 			info->rti_ifp = info->rti_ifa->ifa_ifp;
694 		ifa_ref(info->rti_ifa);
695 	} else
696 		error = ENETUNREACH;
697 	NET_EPOCH_EXIT(et);
698 	return (error);
699 }
700 
701 void
702 rt_updatemtu(struct ifnet *ifp)
703 {
704 	struct rib_head *rnh;
705 	int mtu;
706 	int i, j;
707 
708 	/*
709 	 * Try to update rt_mtu for all routes using this interface
710 	 * Unfortunately the only way to do this is to traverse all
711 	 * routing tables in all fibs/domains.
712 	 */
713 	for (i = 1; i <= AF_MAX; i++) {
714 		mtu = if_getmtu_family(ifp, i);
715 		for (j = 0; j < rt_numfibs; j++) {
716 			rnh = rt_tables_get_rnh(j, i);
717 			if (rnh == NULL)
718 				continue;
719 			nhops_update_ifmtu(rnh, ifp, mtu);
720 		}
721 	}
722 }
723 
724 
725 #if 0
726 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
727 int rt_print(char *buf, int buflen, struct rtentry *rt);
728 
729 int
730 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
731 {
732 	void *paddr = NULL;
733 
734 	switch (s->sa_family) {
735 	case AF_INET:
736 		paddr = &((struct sockaddr_in *)s)->sin_addr;
737 		break;
738 	case AF_INET6:
739 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
740 		break;
741 	}
742 
743 	if (paddr == NULL)
744 		return (0);
745 
746 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
747 		return (0);
748 
749 	return (strlen(buf));
750 }
751 
752 int
753 rt_print(char *buf, int buflen, struct rtentry *rt)
754 {
755 	struct sockaddr *addr, *mask;
756 	int i = 0;
757 
758 	addr = rt_key(rt);
759 	mask = rt_mask(rt);
760 
761 	i = p_sockaddr(buf, buflen, addr);
762 	if (!(rt->rt_flags & RTF_HOST)) {
763 		buf[i++] = '/';
764 		i += p_sockaddr(buf + i, buflen - i, mask);
765 	}
766 
767 	if (rt->rt_flags & RTF_GATEWAY) {
768 		buf[i++] = '>';
769 		i += p_sockaddr(buf + i, buflen - i, &rt->rt_nhop->gw_sa);
770 	}
771 
772 	return (i);
773 }
774 #endif
775 
776 #ifdef RADIX_MPATH
777 /*
778  * Deletes key for single-path routes, unlinks rtentry with
779  * gateway specified in @info from multi-path routes.
780  *
781  * Returnes unlinked entry. In case of failure, returns NULL
782  * and sets @perror to ESRCH.
783  */
784 struct radix_node *
785 rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info,
786     struct rtentry *rto, int *perror)
787 {
788 	/*
789 	 * if we got multipath routes, we require users to specify
790 	 * a matching RTAX_GATEWAY.
791 	 */
792 	struct rtentry *rt; // *rto = NULL;
793 	struct radix_node *rn;
794 	struct sockaddr *gw;
795 
796 	gw = info->rti_info[RTAX_GATEWAY];
797 	rt = rt_mpath_matchgate(rto, gw);
798 	if (rt == NULL) {
799 		*perror = ESRCH;
800 		return (NULL);
801 	}
802 
803 	/*
804 	 * this is the first entry in the chain
805 	 */
806 	if (rto == rt) {
807 		rn = rn_mpath_next((struct radix_node *)rt);
808 		/*
809 		 * there is another entry, now it's active
810 		 */
811 		if (rn) {
812 			rto = RNTORT(rn);
813 			RT_LOCK(rto);
814 			rto->rt_flags |= RTF_UP;
815 			RT_UNLOCK(rto);
816 		} else if (rt->rt_flags & RTF_GATEWAY) {
817 			/*
818 			 * For gateway routes, we need to
819 			 * make sure that we we are deleting
820 			 * the correct gateway.
821 			 * rt_mpath_matchgate() does not
822 			 * check the case when there is only
823 			 * one route in the chain.
824 			 */
825 			if (gw &&
826 			    (rt->rt_nhop->gw_sa.sa_len != gw->sa_len ||
827 				memcmp(&rt->rt_nhop->gw_sa, gw, gw->sa_len))) {
828 				*perror = ESRCH;
829 				return (NULL);
830 			}
831 		}
832 
833 		/*
834 		 * use the normal delete code to remove
835 		 * the first entry
836 		 */
837 		rn = rnh->rnh_deladdr(info->rti_info[RTAX_DST],
838 					info->rti_info[RTAX_NETMASK],
839 					&rnh->head);
840 		if (rn != NULL) {
841 			*perror = 0;
842 		} else {
843 			*perror = ESRCH;
844 		}
845 		return (rn);
846 	}
847 
848 	/*
849 	 * if the entry is 2nd and on up
850 	 */
851 	if (rt_mpath_deldup(rto, rt) == 0)
852 		panic ("rtrequest1: rt_mpath_deldup");
853 	*perror = 0;
854 	rn = (struct radix_node *)rt;
855 	return (rn);
856 }
857 #endif
858 
859 void
860 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
861 {
862 
863 	if (info->rti_mflags & RTV_WEIGHT)
864 		rt->rt_weight = info->rti_rmx->rmx_weight;
865 	/* Kernel -> userland timebase conversion. */
866 	if (info->rti_mflags & RTV_EXPIRE)
867 		rt->rt_expire = info->rti_rmx->rmx_expire ?
868 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
869 }
870 
871 void
872 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
873 {
874 	u_char *cp1 = (u_char *)src;
875 	u_char *cp2 = (u_char *)dst;
876 	u_char *cp3 = (u_char *)netmask;
877 	u_char *cplim = cp2 + *cp3;
878 	u_char *cplim2 = cp2 + *cp1;
879 
880 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
881 	cp3 += 2;
882 	if (cplim > cplim2)
883 		cplim = cplim2;
884 	while (cp2 < cplim)
885 		*cp2++ = *cp1++ & *cp3++;
886 	if (cp2 < cplim2)
887 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
888 }
889 
890 /*
891  * Set up a routing table entry, normally
892  * for an interface.
893  */
894 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
895 static inline  int
896 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
897 {
898 	RIB_RLOCK_TRACKER;
899 	struct epoch_tracker et;
900 	struct sockaddr *dst;
901 	struct sockaddr *netmask;
902 	struct rib_cmd_info rc;
903 	struct rt_addrinfo info;
904 	int error = 0;
905 	int startfib, endfib;
906 	char tempbuf[_SOCKADDR_TMPSIZE];
907 	int didwork = 0;
908 	int a_failure = 0;
909 	struct sockaddr_dl_short *sdl = NULL;
910 	struct rib_head *rnh;
911 
912 	if (flags & RTF_HOST) {
913 		dst = ifa->ifa_dstaddr;
914 		netmask = NULL;
915 	} else {
916 		dst = ifa->ifa_addr;
917 		netmask = ifa->ifa_netmask;
918 	}
919 	if (dst->sa_len == 0)
920 		return(EINVAL);
921 	switch (dst->sa_family) {
922 	case AF_INET6:
923 	case AF_INET:
924 		/* We support multiple FIBs. */
925 		break;
926 	default:
927 		fibnum = RT_DEFAULT_FIB;
928 		break;
929 	}
930 	if (fibnum == RT_ALL_FIBS) {
931 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
932 			startfib = endfib = ifa->ifa_ifp->if_fib;
933 		else {
934 			startfib = 0;
935 			endfib = rt_numfibs - 1;
936 		}
937 	} else {
938 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
939 		startfib = fibnum;
940 		endfib = fibnum;
941 	}
942 
943 	/*
944 	 * If it's a delete, check that if it exists,
945 	 * it's on the correct interface or we might scrub
946 	 * a route to another ifa which would
947 	 * be confusing at best and possibly worse.
948 	 */
949 	if (cmd == RTM_DELETE) {
950 		/*
951 		 * It's a delete, so it should already exist..
952 		 * If it's a net, mask off the host bits
953 		 * (Assuming we have a mask)
954 		 * XXX this is kinda inet specific..
955 		 */
956 		if (netmask != NULL) {
957 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
958 			dst = (struct sockaddr *)tempbuf;
959 		}
960 	} else if (cmd == RTM_ADD) {
961 		sdl = (struct sockaddr_dl_short *)tempbuf;
962 		bzero(sdl, sizeof(struct sockaddr_dl_short));
963 		sdl->sdl_family = AF_LINK;
964 		sdl->sdl_len = sizeof(struct sockaddr_dl_short);
965 		sdl->sdl_type = ifa->ifa_ifp->if_type;
966 		sdl->sdl_index = ifa->ifa_ifp->if_index;
967         }
968 	/*
969 	 * Now go through all the requested tables (fibs) and do the
970 	 * requested action. Realistically, this will either be fib 0
971 	 * for protocols that don't do multiple tables or all the
972 	 * tables for those that do.
973 	 */
974 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
975 		if (cmd == RTM_DELETE) {
976 			struct radix_node *rn;
977 			/*
978 			 * Look up an rtentry that is in the routing tree and
979 			 * contains the correct info.
980 			 */
981 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
982 			if (rnh == NULL)
983 				/* this table doesn't exist but others might */
984 				continue;
985 			RIB_RLOCK(rnh);
986 			rn = rnh->rnh_lookup(dst, netmask, &rnh->head);
987 #ifdef RADIX_MPATH
988 			if (rt_mpath_capable(rnh)) {
989 
990 				if (rn == NULL)
991 					error = ESRCH;
992 				else {
993 					struct rtentry *rt = RNTORT(rn);
994 					/*
995 					 * for interface route the gateway
996 					 * gateway is sockaddr_dl, so
997 					 * rt_mpath_matchgate must use the
998 					 * interface address
999 					 */
1000 					rt = rt_mpath_matchgate(rt,
1001 					    ifa->ifa_addr);
1002 					if (rt == NULL)
1003 						error = ESRCH;
1004 				}
1005 			}
1006 #endif
1007 			error = (rn == NULL ||
1008 			    (rn->rn_flags & RNF_ROOT) ||
1009 			    RNTORT(rn)->rt_nhop->nh_ifa != ifa);
1010 			RIB_RUNLOCK(rnh);
1011 			if (error) {
1012 				/* this is only an error if bad on ALL tables */
1013 				continue;
1014 			}
1015 		}
1016 		/*
1017 		 * Do the actual request
1018 		 */
1019 		bzero((caddr_t)&info, sizeof(info));
1020 		info.rti_ifa = ifa;
1021 		info.rti_flags = flags |
1022 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1023 		info.rti_info[RTAX_DST] = dst;
1024 		/*
1025 		 * doing this for compatibility reasons
1026 		 */
1027 		if (cmd == RTM_ADD)
1028 			info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)sdl;
1029 		else
1030 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1031 		info.rti_info[RTAX_NETMASK] = netmask;
1032 		NET_EPOCH_ENTER(et);
1033 		error = rib_action(fibnum, cmd, &info, &rc);
1034 		if (error == 0 && rc.rc_rt != NULL) {
1035 			/*
1036 			 * notify any listening routing agents of the change
1037 			 */
1038 
1039 			/* TODO: interface routes/aliases */
1040 			rt_newaddrmsg_fib(cmd, ifa, rc.rc_rt, fibnum);
1041 			didwork = 1;
1042 		}
1043 		NET_EPOCH_EXIT(et);
1044 		if (error)
1045 			a_failure = error;
1046 	}
1047 	if (cmd == RTM_DELETE) {
1048 		if (didwork) {
1049 			error = 0;
1050 		} else {
1051 			/* we only give an error if it wasn't in any table */
1052 			error = ((flags & RTF_HOST) ?
1053 			    EHOSTUNREACH : ENETUNREACH);
1054 		}
1055 	} else {
1056 		if (a_failure) {
1057 			/* return an error if any of them failed */
1058 			error = a_failure;
1059 		}
1060 	}
1061 	return (error);
1062 }
1063 
1064 /*
1065  * Set up a routing table entry, normally
1066  * for an interface.
1067  */
1068 int
1069 rtinit(struct ifaddr *ifa, int cmd, int flags)
1070 {
1071 	struct sockaddr *dst;
1072 	int fib = RT_DEFAULT_FIB;
1073 
1074 	if (flags & RTF_HOST) {
1075 		dst = ifa->ifa_dstaddr;
1076 	} else {
1077 		dst = ifa->ifa_addr;
1078 	}
1079 
1080 	switch (dst->sa_family) {
1081 	case AF_INET6:
1082 	case AF_INET:
1083 		/* We do support multiple FIBs. */
1084 		fib = RT_ALL_FIBS;
1085 		break;
1086 	}
1087 	return (rtinit1(ifa, cmd, flags, fib));
1088 }
1089 
1090 /*
1091  * Announce interface address arrival/withdraw
1092  * Returns 0 on success.
1093  */
1094 int
1095 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
1096 {
1097 
1098 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1099 	    ("unexpected cmd %d", cmd));
1100 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1101 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1102 
1103 	EVENTHANDLER_DIRECT_INVOKE(rt_addrmsg, ifa, cmd);
1104 	return (rtsock_addrmsg(cmd, ifa, fibnum));
1105 }
1106 
1107 /*
1108  * Announce kernel-originated route addition/removal to rtsock based on @rt data.
1109  * cmd: RTM_ cmd
1110  * @rt: valid rtentry
1111  * @ifp: target route interface
1112  * @fibnum: fib id or RT_ALL_FIBS
1113  *
1114  * Returns 0 on success.
1115  */
1116 int
1117 rt_routemsg(int cmd, struct rtentry *rt, struct ifnet *ifp, int rti_addrs,
1118     int fibnum)
1119 {
1120 
1121 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1122 	    ("unexpected cmd %d", cmd));
1123 
1124 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1125 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1126 
1127 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
1128 
1129 	return (rtsock_routemsg(cmd, rt, ifp, 0, fibnum));
1130 }
1131 
1132 /*
1133  * Announce kernel-originated route addition/removal to rtsock based on @rt data.
1134  * cmd: RTM_ cmd
1135  * @info: addrinfo structure with valid data.
1136  * @fibnum: fib id or RT_ALL_FIBS
1137  *
1138  * Returns 0 on success.
1139  */
1140 int
1141 rt_routemsg_info(int cmd, struct rt_addrinfo *info, int fibnum)
1142 {
1143 
1144 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE || cmd == RTM_CHANGE,
1145 	    ("unexpected cmd %d", cmd));
1146 
1147 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1148 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1149 
1150 	KASSERT(info->rti_info[RTAX_DST] != NULL, (":%s: RTAX_DST must be supplied", __func__));
1151 
1152 	return (rtsock_routemsg_info(cmd, info, fibnum));
1153 }
1154 
1155 
1156 /*
1157  * This is called to generate messages from the routing socket
1158  * indicating a network interface has had addresses associated with it.
1159  */
1160 void
1161 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, struct rtentry *rt, int fibnum)
1162 {
1163 
1164 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1165 		("unexpected cmd %u", cmd));
1166 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1167 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1168 
1169 	if (cmd == RTM_ADD) {
1170 		rt_addrmsg(cmd, ifa, fibnum);
1171 		if (rt != NULL)
1172 			rt_routemsg(cmd, rt, ifa->ifa_ifp, 0, fibnum);
1173 	} else {
1174 		if (rt != NULL)
1175 			rt_routemsg(cmd, rt, ifa->ifa_ifp, 0, fibnum);
1176 		rt_addrmsg(cmd, ifa, fibnum);
1177 	}
1178 }
1179 
1180