xref: /freebsd/sys/net/route.c (revision 7790c8c1996ad89a22b8bd194a230cf23ee67f4b)
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_DEFINE(struct rtstat, rtstat);
112 #define	V_rtstat	VNET(rtstat)
113 
114 VNET_DEFINE(struct rib_head *, rt_tables);
115 #define	V_rt_tables	VNET(rt_tables)
116 
117 VNET_DEFINE(int, rttrash);		/* routes not in table but not freed */
118 #define	V_rttrash	VNET(rttrash)
119 
120 
121 /*
122  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
123  * The operation can be done safely (in this code) because a
124  * 'struct rtentry' starts with two 'struct radix_node''s, the first
125  * one representing leaf nodes in the routing tree, which is
126  * what the code in radix.c passes us as a 'struct radix_node'.
127  *
128  * But because there are a lot of assumptions in this conversion,
129  * do not cast explicitly, but always use the macro below.
130  */
131 #define RNTORT(p)	((struct rtentry *)(p))
132 
133 VNET_DEFINE_STATIC(uma_zone_t, rtzone);		/* Routing table UMA zone. */
134 #define	V_rtzone	VNET(rtzone)
135 
136 EVENTHANDLER_LIST_DEFINE(rt_addrmsg);
137 
138 static int rtrequest1_fib_change(struct rib_head *, struct rt_addrinfo *,
139     struct rtentry **, u_int);
140 static void rt_setmetrics(const struct rt_addrinfo *, struct rtentry *);
141 static int rt_ifdelroute(const struct rtentry *rt, void *arg);
142 static struct rtentry *rt_unlinkrte(struct rib_head *rnh,
143     struct rt_addrinfo *info, int *perror);
144 static void rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info);
145 #ifdef RADIX_MPATH
146 static struct radix_node *rt_mpath_unlink(struct rib_head *rnh,
147     struct rt_addrinfo *info, struct rtentry *rto, int *perror);
148 #endif
149 static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info,
150     int flags);
151 
152 struct if_mtuinfo
153 {
154 	struct ifnet	*ifp;
155 	int		mtu;
156 };
157 
158 static int	if_updatemtu_cb(struct radix_node *, void *);
159 
160 /*
161  * handler for net.my_fibnum
162  */
163 static int
164 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
165 {
166         int fibnum;
167         int error;
168 
169         fibnum = curthread->td_proc->p_fibnum;
170         error = sysctl_handle_int(oidp, &fibnum, 0, req);
171         return (error);
172 }
173 
174 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
175             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
176 
177 static __inline struct rib_head **
178 rt_tables_get_rnh_ptr(int table, int fam)
179 {
180 	struct rib_head **rnh;
181 
182 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
183 	    __func__));
184 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
185 	    __func__));
186 
187 	/* rnh is [fib=0][af=0]. */
188 	rnh = (struct rib_head **)V_rt_tables;
189 	/* Get the offset to the requested table and fam. */
190 	rnh += table * (AF_MAX+1) + fam;
191 
192 	return (rnh);
193 }
194 
195 struct rib_head *
196 rt_tables_get_rnh(int table, int fam)
197 {
198 
199 	return (*rt_tables_get_rnh_ptr(table, fam));
200 }
201 
202 u_int
203 rt_tables_get_gen(int table, int fam)
204 {
205 	struct rib_head *rnh;
206 
207 	rnh = *rt_tables_get_rnh_ptr(table, fam);
208 	KASSERT(rnh != NULL, ("%s: NULL rib_head pointer table %d fam %d",
209 	    __func__, table, fam));
210 	return (rnh->rnh_gen);
211 }
212 
213 
214 /*
215  * route initialization must occur before ip6_init2(), which happenas at
216  * SI_ORDER_MIDDLE.
217  */
218 static void
219 route_init(void)
220 {
221 
222 	/* whack the tunable ints into  line. */
223 	if (rt_numfibs > RT_MAXFIBS)
224 		rt_numfibs = RT_MAXFIBS;
225 	if (rt_numfibs == 0)
226 		rt_numfibs = 1;
227 }
228 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, NULL);
229 
230 static int
231 rtentry_zinit(void *mem, int size, int how)
232 {
233 	struct rtentry *rt = mem;
234 
235 	rt->rt_pksent = counter_u64_alloc(how);
236 	if (rt->rt_pksent == NULL)
237 		return (ENOMEM);
238 
239 	RT_LOCK_INIT(rt);
240 
241 	return (0);
242 }
243 
244 static void
245 rtentry_zfini(void *mem, int size)
246 {
247 	struct rtentry *rt = mem;
248 
249 	RT_LOCK_DESTROY(rt);
250 	counter_u64_free(rt->rt_pksent);
251 }
252 
253 static int
254 rtentry_ctor(void *mem, int size, void *arg, int how)
255 {
256 	struct rtentry *rt = mem;
257 
258 	bzero(rt, offsetof(struct rtentry, rt_endzero));
259 	counter_u64_zero(rt->rt_pksent);
260 	rt->rt_chain = NULL;
261 
262 	return (0);
263 }
264 
265 static void
266 rtentry_dtor(void *mem, int size, void *arg)
267 {
268 	struct rtentry *rt = mem;
269 
270 	RT_UNLOCK_COND(rt);
271 }
272 
273 static void
274 vnet_route_init(const void *unused __unused)
275 {
276 	struct domain *dom;
277 	struct rib_head **rnh;
278 	int table;
279 	int fam;
280 
281 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
282 	    sizeof(struct rib_head *), M_RTABLE, M_WAITOK|M_ZERO);
283 
284 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
285 	    rtentry_ctor, rtentry_dtor,
286 	    rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
287 	for (dom = domains; dom; dom = dom->dom_next) {
288 		if (dom->dom_rtattach == NULL)
289 			continue;
290 
291 		for  (table = 0; table < rt_numfibs; table++) {
292 			fam = dom->dom_family;
293 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
294 				break;
295 
296 			rnh = rt_tables_get_rnh_ptr(table, fam);
297 			if (rnh == NULL)
298 				panic("%s: rnh NULL", __func__);
299 			dom->dom_rtattach((void **)rnh, 0);
300 		}
301 	}
302 }
303 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
304     vnet_route_init, 0);
305 
306 #ifdef VIMAGE
307 static void
308 vnet_route_uninit(const void *unused __unused)
309 {
310 	int table;
311 	int fam;
312 	struct domain *dom;
313 	struct rib_head **rnh;
314 
315 	for (dom = domains; dom; dom = dom->dom_next) {
316 		if (dom->dom_rtdetach == NULL)
317 			continue;
318 
319 		for (table = 0; table < rt_numfibs; table++) {
320 			fam = dom->dom_family;
321 
322 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
323 				break;
324 
325 			rnh = rt_tables_get_rnh_ptr(table, fam);
326 			if (rnh == NULL)
327 				panic("%s: rnh NULL", __func__);
328 			dom->dom_rtdetach((void **)rnh, 0);
329 		}
330 	}
331 
332 	free(V_rt_tables, M_RTABLE);
333 	uma_zdestroy(V_rtzone);
334 }
335 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
336     vnet_route_uninit, 0);
337 #endif
338 
339 struct rib_head *
340 rt_table_init(int offset)
341 {
342 	struct rib_head *rh;
343 
344 	rh = malloc(sizeof(struct rib_head), M_RTABLE, M_WAITOK | M_ZERO);
345 
346 	/* TODO: These details should be hidded inside radix.c */
347 	/* Init masks tree */
348 	rn_inithead_internal(&rh->head, rh->rnh_nodes, offset);
349 	rn_inithead_internal(&rh->rmhead.head, rh->rmhead.mask_nodes, 0);
350 	rh->head.rnh_masks = &rh->rmhead;
351 
352 	/* Init locks */
353 	RIB_LOCK_INIT(rh);
354 
355 	/* Finally, set base callbacks */
356 	rh->rnh_addaddr = rn_addroute;
357 	rh->rnh_deladdr = rn_delete;
358 	rh->rnh_matchaddr = rn_match;
359 	rh->rnh_lookup = rn_lookup;
360 	rh->rnh_walktree = rn_walktree;
361 	rh->rnh_walktree_from = rn_walktree_from;
362 
363 	return (rh);
364 }
365 
366 static int
367 rt_freeentry(struct radix_node *rn, void *arg)
368 {
369 	struct radix_head * const rnh = arg;
370 	struct radix_node *x;
371 
372 	x = (struct radix_node *)rn_delete(rn + 2, NULL, rnh);
373 	if (x != NULL)
374 		R_Free(x);
375 	return (0);
376 }
377 
378 void
379 rt_table_destroy(struct rib_head *rh)
380 {
381 
382 	rn_walktree(&rh->rmhead.head, rt_freeentry, &rh->rmhead.head);
383 
384 	/* Assume table is already empty */
385 	RIB_LOCK_DESTROY(rh);
386 	free(rh, M_RTABLE);
387 }
388 
389 
390 #ifndef _SYS_SYSPROTO_H_
391 struct setfib_args {
392 	int     fibnum;
393 };
394 #endif
395 int
396 sys_setfib(struct thread *td, struct setfib_args *uap)
397 {
398 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
399 		return EINVAL;
400 	td->td_proc->p_fibnum = uap->fibnum;
401 	return (0);
402 }
403 
404 /*
405  * Packet routing routines.
406  */
407 void
408 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
409 {
410 	struct rtentry *rt;
411 
412 	if ((rt = ro->ro_rt) != NULL) {
413 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
414 			return;
415 		RTFREE(rt);
416 		ro->ro_rt = NULL;
417 	}
418 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
419 	if (ro->ro_rt)
420 		RT_UNLOCK(ro->ro_rt);
421 }
422 
423 /*
424  * Look up the route that matches the address given
425  * Or, at least try.. Create a cloned route if needed.
426  *
427  * The returned route, if any, is locked.
428  */
429 struct rtentry *
430 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
431 {
432 
433 	return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
434 }
435 
436 struct rtentry *
437 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
438 		    u_int fibnum)
439 {
440 	RIB_RLOCK_TRACKER;
441 	struct rib_head *rh;
442 	struct radix_node *rn;
443 	struct rtentry *newrt;
444 	struct rt_addrinfo info;
445 	int err = 0, msgtype = RTM_MISS;
446 
447 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
448 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
449 	newrt = NULL;
450 	if (rh == NULL)
451 		goto miss;
452 
453 	/*
454 	 * Look up the address in the table for that Address Family
455 	 */
456 	if ((ignflags & RTF_RNH_LOCKED) == 0)
457 		RIB_RLOCK(rh);
458 #ifdef INVARIANTS
459 	else
460 		RIB_LOCK_ASSERT(rh);
461 #endif
462 	rn = rh->rnh_matchaddr(dst, &rh->head);
463 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
464 		newrt = RNTORT(rn);
465 		RT_LOCK(newrt);
466 		RT_ADDREF(newrt);
467 		if ((ignflags & RTF_RNH_LOCKED) == 0)
468 			RIB_RUNLOCK(rh);
469 		return (newrt);
470 
471 	} else if ((ignflags & RTF_RNH_LOCKED) == 0)
472 		RIB_RUNLOCK(rh);
473 	/*
474 	 * Either we hit the root or could not find any match,
475 	 * which basically means: "cannot get there from here".
476 	 */
477 miss:
478 	V_rtstat.rts_unreach++;
479 
480 	if (report) {
481 		/*
482 		 * If required, report the failure to the supervising
483 		 * Authorities.
484 		 * For a delete, this is not an error. (report == 0)
485 		 */
486 		bzero(&info, sizeof(info));
487 		info.rti_info[RTAX_DST] = dst;
488 		rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
489 	}
490 	return (newrt);
491 }
492 
493 /*
494  * Remove a reference count from an rtentry.
495  * If the count gets low enough, take it out of the routing table
496  */
497 void
498 rtfree(struct rtentry *rt)
499 {
500 	struct rib_head *rnh;
501 
502 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
503 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
504 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
505 
506 	RT_LOCK_ASSERT(rt);
507 
508 	/*
509 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
510 	 * we should come here exactly with the last reference.
511 	 */
512 	RT_REMREF(rt);
513 	if (rt->rt_refcnt > 0) {
514 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
515 		goto done;
516 	}
517 
518 	/*
519 	 * On last reference give the "close method" a chance
520 	 * to cleanup private state.  This also permits (for
521 	 * IPv4 and IPv6) a chance to decide if the routing table
522 	 * entry should be purged immediately or at a later time.
523 	 * When an immediate purge is to happen the close routine
524 	 * typically calls rtexpunge which clears the RTF_UP flag
525 	 * on the entry so that the code below reclaims the storage.
526 	 */
527 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
528 		rnh->rnh_close((struct radix_node *)rt, &rnh->head);
529 
530 	/*
531 	 * If we are no longer "up" (and ref == 0)
532 	 * then we can free the resources associated
533 	 * with the route.
534 	 */
535 	if ((rt->rt_flags & RTF_UP) == 0) {
536 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
537 			panic("rtfree 2");
538 		/*
539 		 * the rtentry must have been removed from the routing table
540 		 * so it is represented in rttrash.. remove that now.
541 		 */
542 		V_rttrash--;
543 #ifdef	DIAGNOSTIC
544 		if (rt->rt_refcnt < 0) {
545 			printf("rtfree: %p not freed (neg refs)\n", rt);
546 			goto done;
547 		}
548 #endif
549 		/*
550 		 * release references on items we hold them on..
551 		 * e.g other routes and ifaddrs.
552 		 */
553 		if (rt->rt_ifa)
554 			ifa_free(rt->rt_ifa);
555 		/*
556 		 * The key is separatly alloc'd so free it (see rt_setgate()).
557 		 * This also frees the gateway, as they are always malloc'd
558 		 * together.
559 		 */
560 		R_Free(rt_key(rt));
561 
562 		/*
563 		 * and the rtentry itself of course
564 		 */
565 		uma_zfree(V_rtzone, rt);
566 		return;
567 	}
568 done:
569 	RT_UNLOCK(rt);
570 }
571 
572 
573 /*
574  * Force a routing table entry to the specified
575  * destination to go through the given gateway.
576  * Normally called as a result of a routing redirect
577  * message from the network layer.
578  */
579 void
580 rtredirect_fib(struct sockaddr *dst,
581 	struct sockaddr *gateway,
582 	struct sockaddr *netmask,
583 	int flags,
584 	struct sockaddr *src,
585 	u_int fibnum)
586 {
587 	struct rtentry *rt;
588 	int error = 0;
589 	short *stat = NULL;
590 	struct rt_addrinfo info;
591 	struct ifaddr *ifa;
592 	struct rib_head *rnh;
593 
594 	NET_EPOCH_ASSERT();
595 
596 	ifa = NULL;
597 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
598 	if (rnh == NULL) {
599 		error = EAFNOSUPPORT;
600 		goto out;
601 	}
602 	/* verify the gateway is directly reachable */
603 	if ((ifa = ifa_ifwithnet(gateway, 0, fibnum)) == NULL) {
604 		error = ENETUNREACH;
605 		goto out;
606 	}
607 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
608 	/*
609 	 * If the redirect isn't from our current router for this dst,
610 	 * it's either old or wrong.  If it redirects us to ourselves,
611 	 * we have a routing loop, perhaps as a result of an interface
612 	 * going down recently.
613 	 */
614 	if (!(flags & RTF_DONE) && rt) {
615 		if (!sa_equal(src, rt->rt_gateway)) {
616 			error = EINVAL;
617 			goto done;
618 		}
619 		if (rt->rt_ifa != ifa && ifa->ifa_addr->sa_family != AF_LINK) {
620 			error = EINVAL;
621 			goto done;
622 		}
623 	}
624 	if ((flags & RTF_GATEWAY) && ifa_ifwithaddr_check(gateway)) {
625 		error = EHOSTUNREACH;
626 		goto done;
627 	}
628 	/*
629 	 * Create a new entry if we just got back a wildcard entry
630 	 * or the lookup failed.  This is necessary for hosts
631 	 * which use routing redirects generated by smart gateways
632 	 * to dynamically build the routing tables.
633 	 */
634 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
635 		goto create;
636 	/*
637 	 * Don't listen to the redirect if it's
638 	 * for a route to an interface.
639 	 */
640 	if (rt->rt_flags & RTF_GATEWAY) {
641 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
642 			/*
643 			 * Changing from route to net => route to host.
644 			 * Create new route, rather than smashing route to net.
645 			 */
646 		create:
647 			if (rt != NULL)
648 				RTFREE_LOCKED(rt);
649 
650 			flags |= RTF_DYNAMIC;
651 			bzero((caddr_t)&info, sizeof(info));
652 			info.rti_info[RTAX_DST] = dst;
653 			info.rti_info[RTAX_GATEWAY] = gateway;
654 			info.rti_info[RTAX_NETMASK] = netmask;
655 			ifa_ref(ifa);
656 			info.rti_ifa = ifa;
657 			info.rti_flags = flags;
658 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
659 			if (rt != NULL) {
660 				RT_LOCK(rt);
661 				flags = rt->rt_flags;
662 			}
663 
664 			stat = &V_rtstat.rts_dynamic;
665 		} else {
666 
667 			/*
668 			 * Smash the current notion of the gateway to
669 			 * this destination.  Should check about netmask!!!
670 			 */
671 			if ((flags & RTF_GATEWAY) == 0)
672 				rt->rt_flags &= ~RTF_GATEWAY;
673 			rt->rt_flags |= RTF_MODIFIED;
674 			flags |= RTF_MODIFIED;
675 			stat = &V_rtstat.rts_newgateway;
676 			/*
677 			 * add the key and gateway (in one malloc'd chunk).
678 			 */
679 			RT_UNLOCK(rt);
680 			RIB_WLOCK(rnh);
681 			RT_LOCK(rt);
682 			rt_setgate(rt, rt_key(rt), gateway);
683 			RIB_WUNLOCK(rnh);
684 		}
685 	} else
686 		error = EHOSTUNREACH;
687 done:
688 	if (rt)
689 		RTFREE_LOCKED(rt);
690  out:
691 	if (error)
692 		V_rtstat.rts_badredirect++;
693 	else if (stat != NULL)
694 		(*stat)++;
695 	bzero((caddr_t)&info, sizeof(info));
696 	info.rti_info[RTAX_DST] = dst;
697 	info.rti_info[RTAX_GATEWAY] = gateway;
698 	info.rti_info[RTAX_NETMASK] = netmask;
699 	info.rti_info[RTAX_AUTHOR] = src;
700 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
701 }
702 
703 /*
704  * Routing table ioctl interface.
705  */
706 int
707 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
708 {
709 
710 	/*
711 	 * If more ioctl commands are added here, make sure the proper
712 	 * super-user checks are being performed because it is possible for
713 	 * prison-root to make it this far if raw sockets have been enabled
714 	 * in jails.
715 	 */
716 #ifdef INET
717 	/* Multicast goop, grrr... */
718 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
719 #else /* INET */
720 	return ENXIO;
721 #endif /* INET */
722 }
723 
724 struct ifaddr *
725 ifa_ifwithroute(int flags, const struct sockaddr *dst, struct sockaddr *gateway,
726 				u_int fibnum)
727 {
728 	struct ifaddr *ifa;
729 	int not_found = 0;
730 
731 	MPASS(in_epoch(net_epoch_preempt));
732 	if ((flags & RTF_GATEWAY) == 0) {
733 		/*
734 		 * If we are adding a route to an interface,
735 		 * and the interface is a pt to pt link
736 		 * we should search for the destination
737 		 * as our clue to the interface.  Otherwise
738 		 * we can use the local address.
739 		 */
740 		ifa = NULL;
741 		if (flags & RTF_HOST)
742 			ifa = ifa_ifwithdstaddr(dst, fibnum);
743 		if (ifa == NULL)
744 			ifa = ifa_ifwithaddr(gateway);
745 	} else {
746 		/*
747 		 * If we are adding a route to a remote net
748 		 * or host, the gateway may still be on the
749 		 * other end of a pt to pt link.
750 		 */
751 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
752 	}
753 	if (ifa == NULL)
754 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
755 	if (ifa == NULL) {
756 		struct rtentry *rt;
757 
758 		rt = rtalloc1_fib(gateway, 0, flags, fibnum);
759 		if (rt == NULL)
760 			goto out;
761 		/*
762 		 * dismiss a gateway that is reachable only
763 		 * through the default router
764 		 */
765 		switch (gateway->sa_family) {
766 		case AF_INET:
767 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
768 				not_found = 1;
769 			break;
770 		case AF_INET6:
771 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
772 				not_found = 1;
773 			break;
774 		default:
775 			break;
776 		}
777 		if (!not_found && rt->rt_ifa != NULL) {
778 			ifa = rt->rt_ifa;
779 		}
780 		RT_REMREF(rt);
781 		RT_UNLOCK(rt);
782 		if (not_found || ifa == NULL)
783 			goto out;
784 	}
785 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
786 		struct ifaddr *oifa = ifa;
787 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
788 		if (ifa == NULL)
789 			ifa = oifa;
790 	}
791  out:
792 	return (ifa);
793 }
794 
795 /*
796  * Do appropriate manipulations of a routing tree given
797  * all the bits of info needed
798  */
799 int
800 rtrequest_fib(int req,
801 	struct sockaddr *dst,
802 	struct sockaddr *gateway,
803 	struct sockaddr *netmask,
804 	int flags,
805 	struct rtentry **ret_nrt,
806 	u_int fibnum)
807 {
808 	struct rt_addrinfo info;
809 
810 	if (dst->sa_len == 0)
811 		return(EINVAL);
812 
813 	bzero((caddr_t)&info, sizeof(info));
814 	info.rti_flags = flags;
815 	info.rti_info[RTAX_DST] = dst;
816 	info.rti_info[RTAX_GATEWAY] = gateway;
817 	info.rti_info[RTAX_NETMASK] = netmask;
818 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
819 }
820 
821 
822 /*
823  * Copy most of @rt data into @info.
824  *
825  * If @flags contains NHR_COPY, copies dst,netmask and gw to the
826  * pointers specified by @info structure. Assume such pointers
827  * are zeroed sockaddr-like structures with sa_len field initialized
828  * to reflect size of the provided buffer. if no NHR_COPY is specified,
829  * point dst,netmask and gw @info fields to appropriate @rt values.
830  *
831  * if @flags contains NHR_REF, do refcouting on rt_ifp.
832  *
833  * Returns 0 on success.
834  */
835 int
836 rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags)
837 {
838 	struct rt_metrics *rmx;
839 	struct sockaddr *src, *dst;
840 	int sa_len;
841 
842 	if (flags & NHR_COPY) {
843 		/* Copy destination if dst is non-zero */
844 		src = rt_key(rt);
845 		dst = info->rti_info[RTAX_DST];
846 		sa_len = src->sa_len;
847 		if (dst != NULL) {
848 			if (src->sa_len > dst->sa_len)
849 				return (ENOMEM);
850 			memcpy(dst, src, src->sa_len);
851 			info->rti_addrs |= RTA_DST;
852 		}
853 
854 		/* Copy mask if set && dst is non-zero */
855 		src = rt_mask(rt);
856 		dst = info->rti_info[RTAX_NETMASK];
857 		if (src != NULL && dst != NULL) {
858 
859 			/*
860 			 * Radix stores different value in sa_len,
861 			 * assume rt_mask() to have the same length
862 			 * as rt_key()
863 			 */
864 			if (sa_len > dst->sa_len)
865 				return (ENOMEM);
866 			memcpy(dst, src, src->sa_len);
867 			info->rti_addrs |= RTA_NETMASK;
868 		}
869 
870 		/* Copy gateway is set && dst is non-zero */
871 		src = rt->rt_gateway;
872 		dst = info->rti_info[RTAX_GATEWAY];
873 		if ((rt->rt_flags & RTF_GATEWAY) && src != NULL && dst != NULL){
874 			if (src->sa_len > dst->sa_len)
875 				return (ENOMEM);
876 			memcpy(dst, src, src->sa_len);
877 			info->rti_addrs |= RTA_GATEWAY;
878 		}
879 	} else {
880 		info->rti_info[RTAX_DST] = rt_key(rt);
881 		info->rti_addrs |= RTA_DST;
882 		if (rt_mask(rt) != NULL) {
883 			info->rti_info[RTAX_NETMASK] = rt_mask(rt);
884 			info->rti_addrs |= RTA_NETMASK;
885 		}
886 		if (rt->rt_flags & RTF_GATEWAY) {
887 			info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
888 			info->rti_addrs |= RTA_GATEWAY;
889 		}
890 	}
891 
892 	rmx = info->rti_rmx;
893 	if (rmx != NULL) {
894 		info->rti_mflags |= RTV_MTU;
895 		rmx->rmx_mtu = rt->rt_mtu;
896 	}
897 
898 	info->rti_flags = rt->rt_flags;
899 	info->rti_ifp = rt->rt_ifp;
900 	info->rti_ifa = rt->rt_ifa;
901 	ifa_ref(info->rti_ifa);
902 	if (flags & NHR_REF) {
903 		/* Do 'traditional' refcouting */
904 		if_ref(info->rti_ifp);
905 	}
906 
907 	return (0);
908 }
909 
910 /*
911  * Lookups up route entry for @dst in RIB database for fib @fibnum.
912  * Exports entry data to @info using rt_exportinfo().
913  *
914  * if @flags contains NHR_REF, refcouting is performed on rt_ifp.
915  *   All references can be released later by calling rib_free_info()
916  *
917  * Returns 0 on success.
918  * Returns ENOENT for lookup failure, ENOMEM for export failure.
919  */
920 int
921 rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags,
922     uint32_t flowid, struct rt_addrinfo *info)
923 {
924 	RIB_RLOCK_TRACKER;
925 	struct rib_head *rh;
926 	struct radix_node *rn;
927 	struct rtentry *rt;
928 	int error;
929 
930 	KASSERT((fibnum < rt_numfibs), ("rib_lookup_rte: bad fibnum"));
931 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
932 	if (rh == NULL)
933 		return (ENOENT);
934 
935 	RIB_RLOCK(rh);
936 	rn = rh->rnh_matchaddr(__DECONST(void *, dst), &rh->head);
937 	if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) {
938 		rt = RNTORT(rn);
939 		/* Ensure route & ifp is UP */
940 		if (RT_LINK_IS_UP(rt->rt_ifp)) {
941 			flags = (flags & NHR_REF) | NHR_COPY;
942 			error = rt_exportinfo(rt, info, flags);
943 			RIB_RUNLOCK(rh);
944 
945 			return (error);
946 		}
947 	}
948 	RIB_RUNLOCK(rh);
949 
950 	return (ENOENT);
951 }
952 
953 /*
954  * Releases all references acquired by rib_lookup_info() when
955  * called with NHR_REF flags.
956  */
957 void
958 rib_free_info(struct rt_addrinfo *info)
959 {
960 
961 	if_rele(info->rti_ifp);
962 }
963 
964 /*
965  * Iterates over all existing fibs in system calling
966  *  @setwa_f function prior to traversing each fib.
967  *  Calls @wa_f function for each element in current fib.
968  * If af is not AF_UNSPEC, iterates over fibs in particular
969  * address family.
970  */
971 void
972 rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f,
973     void *arg)
974 {
975 	struct rib_head *rnh;
976 	uint32_t fibnum;
977 	int i;
978 
979 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
980 		/* Do we want some specific family? */
981 		if (af != AF_UNSPEC) {
982 			rnh = rt_tables_get_rnh(fibnum, af);
983 			if (rnh == NULL)
984 				continue;
985 			if (setwa_f != NULL)
986 				setwa_f(rnh, fibnum, af, arg);
987 
988 			RIB_WLOCK(rnh);
989 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
990 			RIB_WUNLOCK(rnh);
991 			continue;
992 		}
993 
994 		for (i = 1; i <= AF_MAX; i++) {
995 			rnh = rt_tables_get_rnh(fibnum, i);
996 			if (rnh == NULL)
997 				continue;
998 			if (setwa_f != NULL)
999 				setwa_f(rnh, fibnum, i, arg);
1000 
1001 			RIB_WLOCK(rnh);
1002 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
1003 			RIB_WUNLOCK(rnh);
1004 		}
1005 	}
1006 }
1007 
1008 struct rt_delinfo
1009 {
1010 	struct rt_addrinfo info;
1011 	struct rib_head *rnh;
1012 	struct rtentry *head;
1013 };
1014 
1015 /*
1016  * Conditionally unlinks @rn from radix tree based
1017  * on info data passed in @arg.
1018  */
1019 static int
1020 rt_checkdelroute(struct radix_node *rn, void *arg)
1021 {
1022 	struct rt_delinfo *di;
1023 	struct rt_addrinfo *info;
1024 	struct rtentry *rt;
1025 	int error;
1026 
1027 	di = (struct rt_delinfo *)arg;
1028 	rt = (struct rtentry *)rn;
1029 	info = &di->info;
1030 	error = 0;
1031 
1032 	info->rti_info[RTAX_DST] = rt_key(rt);
1033 	info->rti_info[RTAX_NETMASK] = rt_mask(rt);
1034 	info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1035 
1036 	rt = rt_unlinkrte(di->rnh, info, &error);
1037 	if (rt == NULL) {
1038 		/* Either not allowed or not matched. Skip entry */
1039 		return (0);
1040 	}
1041 
1042 	/* Entry was unlinked. Add to the list and return */
1043 	rt->rt_chain = di->head;
1044 	di->head = rt;
1045 
1046 	return (0);
1047 }
1048 
1049 /*
1050  * Iterates over all existing fibs in system.
1051  * Deletes each element for which @filter_f function returned
1052  * non-zero value.
1053  * If @af is not AF_UNSPEC, iterates over fibs in particular
1054  * address family.
1055  */
1056 void
1057 rt_foreach_fib_walk_del(int af, rt_filter_f_t *filter_f, void *arg)
1058 {
1059 	struct rib_head *rnh;
1060 	struct rt_delinfo di;
1061 	struct rtentry *rt;
1062 	uint32_t fibnum;
1063 	int i, start, end;
1064 
1065 	bzero(&di, sizeof(di));
1066 	di.info.rti_filter = filter_f;
1067 	di.info.rti_filterdata = arg;
1068 
1069 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
1070 		/* Do we want some specific family? */
1071 		if (af != AF_UNSPEC) {
1072 			start = af;
1073 			end = af;
1074 		} else {
1075 			start = 1;
1076 			end = AF_MAX;
1077 		}
1078 
1079 		for (i = start; i <= end; i++) {
1080 			rnh = rt_tables_get_rnh(fibnum, i);
1081 			if (rnh == NULL)
1082 				continue;
1083 			di.rnh = rnh;
1084 
1085 			RIB_WLOCK(rnh);
1086 			rnh->rnh_walktree(&rnh->head, rt_checkdelroute, &di);
1087 			RIB_WUNLOCK(rnh);
1088 
1089 			if (di.head == NULL)
1090 				continue;
1091 
1092 			/* We might have something to reclaim */
1093 			while (di.head != NULL) {
1094 				rt = di.head;
1095 				di.head = rt->rt_chain;
1096 				rt->rt_chain = NULL;
1097 
1098 				/* TODO std rt -> rt_addrinfo export */
1099 				di.info.rti_info[RTAX_DST] = rt_key(rt);
1100 				di.info.rti_info[RTAX_NETMASK] = rt_mask(rt);
1101 
1102 				rt_notifydelete(rt, &di.info);
1103 				RTFREE_LOCKED(rt);
1104 			}
1105 
1106 		}
1107 	}
1108 }
1109 
1110 /*
1111  * Delete Routes for a Network Interface
1112  *
1113  * Called for each routing entry via the rnh->rnh_walktree() call above
1114  * to delete all route entries referencing a detaching network interface.
1115  *
1116  * Arguments:
1117  *	rt	pointer to rtentry
1118  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1119  *
1120  * Returns:
1121  *	0	successful
1122  *	errno	failed - reason indicated
1123  */
1124 static int
1125 rt_ifdelroute(const struct rtentry *rt, void *arg)
1126 {
1127 	struct ifnet	*ifp = arg;
1128 
1129 	if (rt->rt_ifp != ifp)
1130 		return (0);
1131 
1132 	/*
1133 	 * Protect (sorta) against walktree recursion problems
1134 	 * with cloned routes
1135 	 */
1136 	if ((rt->rt_flags & RTF_UP) == 0)
1137 		return (0);
1138 
1139 	return (1);
1140 }
1141 
1142 /*
1143  * Delete all remaining routes using this interface
1144  * Unfortuneatly the only way to do this is to slog through
1145  * the entire routing table looking for routes which point
1146  * to this interface...oh well...
1147  */
1148 void
1149 rt_flushifroutes_af(struct ifnet *ifp, int af)
1150 {
1151 	KASSERT((af >= 1 && af <= AF_MAX), ("%s: af %d not >= 1 and <= %d",
1152 	    __func__, af, AF_MAX));
1153 
1154 	rt_foreach_fib_walk_del(af, rt_ifdelroute, ifp);
1155 }
1156 
1157 void
1158 rt_flushifroutes(struct ifnet *ifp)
1159 {
1160 
1161 	rt_foreach_fib_walk_del(AF_UNSPEC, rt_ifdelroute, ifp);
1162 }
1163 
1164 /*
1165  * Conditionally unlinks rtentry matching data inside @info from @rnh.
1166  * Returns unlinked, locked and referenced @rtentry on success,
1167  * Returns NULL and sets @perror to:
1168  * ESRCH - if prefix was not found,
1169  * EADDRINUSE - if trying to delete PINNED route without appropriate flag.
1170  * ENOENT - if supplied filter function returned 0 (not matched).
1171  */
1172 static struct rtentry *
1173 rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, int *perror)
1174 {
1175 	struct sockaddr *dst, *netmask;
1176 	struct rtentry *rt;
1177 	struct radix_node *rn;
1178 
1179 	dst = info->rti_info[RTAX_DST];
1180 	netmask = info->rti_info[RTAX_NETMASK];
1181 
1182 	rt = (struct rtentry *)rnh->rnh_lookup(dst, netmask, &rnh->head);
1183 	if (rt == NULL) {
1184 		*perror = ESRCH;
1185 		return (NULL);
1186 	}
1187 
1188 	if ((info->rti_flags & RTF_PINNED) == 0) {
1189 		/* Check if target route can be deleted */
1190 		if (rt->rt_flags & RTF_PINNED) {
1191 			*perror = EADDRINUSE;
1192 			return (NULL);
1193 		}
1194 	}
1195 
1196 	if (info->rti_filter != NULL) {
1197 		if (info->rti_filter(rt, info->rti_filterdata) == 0) {
1198 			/* Not matched */
1199 			*perror = ENOENT;
1200 			return (NULL);
1201 		}
1202 
1203 		/*
1204 		 * Filter function requested rte deletion.
1205 		 * Ease the caller work by filling in remaining info
1206 		 * from that particular entry.
1207 		 */
1208 		info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1209 	}
1210 
1211 	/*
1212 	 * Remove the item from the tree and return it.
1213 	 * Complain if it is not there and do no more processing.
1214 	 */
1215 	*perror = ESRCH;
1216 #ifdef RADIX_MPATH
1217 	if (rt_mpath_capable(rnh))
1218 		rn = rt_mpath_unlink(rnh, info, rt, perror);
1219 	else
1220 #endif
1221 	rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1222 	if (rn == NULL)
1223 		return (NULL);
1224 
1225 	if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1226 		panic ("rtrequest delete");
1227 
1228 	rt = RNTORT(rn);
1229 	RT_LOCK(rt);
1230 	RT_ADDREF(rt);
1231 	rt->rt_flags &= ~RTF_UP;
1232 
1233 	*perror = 0;
1234 
1235 	return (rt);
1236 }
1237 
1238 static void
1239 rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info)
1240 {
1241 	struct ifaddr *ifa;
1242 
1243 	/*
1244 	 * give the protocol a chance to keep things in sync.
1245 	 */
1246 	ifa = rt->rt_ifa;
1247 	if (ifa != NULL && ifa->ifa_rtrequest != NULL)
1248 		ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1249 
1250 	/*
1251 	 * One more rtentry floating around that is not
1252 	 * linked to the routing table. rttrash will be decremented
1253 	 * when RTFREE(rt) is eventually called.
1254 	 */
1255 	V_rttrash++;
1256 }
1257 
1258 
1259 /*
1260  * These (questionable) definitions of apparent local variables apply
1261  * to the next two functions.  XXXXXX!!!
1262  */
1263 #define	dst	info->rti_info[RTAX_DST]
1264 #define	gateway	info->rti_info[RTAX_GATEWAY]
1265 #define	netmask	info->rti_info[RTAX_NETMASK]
1266 #define	ifaaddr	info->rti_info[RTAX_IFA]
1267 #define	ifpaddr	info->rti_info[RTAX_IFP]
1268 #define	flags	info->rti_flags
1269 
1270 /*
1271  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
1272  * it will be referenced so the caller must free it.
1273  *
1274  * Assume basic consistency checks are executed by callers:
1275  * RTAX_DST exists, if RTF_GATEWAY is set, RTAX_GATEWAY exists as well.
1276  */
1277 int
1278 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
1279 {
1280 	struct epoch_tracker et;
1281 	int needref, error;
1282 
1283 	/*
1284 	 * ifp may be specified by sockaddr_dl
1285 	 * when protocol address is ambiguous.
1286 	 */
1287 	error = 0;
1288 	needref = (info->rti_ifa == NULL);
1289 	NET_EPOCH_ENTER(et);
1290 
1291 	/* If we have interface specified by the ifindex in the address, use it */
1292 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
1293 	    ifpaddr->sa_family == AF_LINK) {
1294 	    const struct sockaddr_dl *sdl = (const struct sockaddr_dl *)ifpaddr;
1295 	    if (sdl->sdl_index != 0)
1296 		    info->rti_ifp = ifnet_byindex(sdl->sdl_index);
1297 	}
1298 	/*
1299 	 * If we have source address specified, try to find it
1300 	 * TODO: avoid enumerating all ifas on all interfaces.
1301 	 */
1302 	if (info->rti_ifa == NULL && ifaaddr != NULL)
1303 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
1304 	if (info->rti_ifa == NULL) {
1305 		struct sockaddr *sa;
1306 
1307 		/*
1308 		 * Most common use case for the userland-supplied routes.
1309 		 *
1310 		 * Choose sockaddr to select ifa.
1311 		 * -- if ifp is set --
1312 		 * Order of preference:
1313 		 * 1) IFA address
1314 		 * 2) gateway address
1315 		 *   Note: for interface routes link-level gateway address
1316 		 *     is specified to indicate the interface index without
1317 		 *     specifying RTF_GATEWAY. In this case, ignore gateway
1318 		 *   Note: gateway AF may be different from dst AF. In this case,
1319 		 *   ignore gateway
1320 		 * 3) final destination.
1321 		 * 4) if all of these fails, try to get at least link-level ifa.
1322 		 * -- else --
1323 		 * try to lookup gateway or dst in the routing table to get ifa
1324 		 */
1325 		if (info->rti_info[RTAX_IFA] != NULL)
1326 			sa = info->rti_info[RTAX_IFA];
1327 		else if ((info->rti_flags & RTF_GATEWAY) != 0 &&
1328 		    gateway->sa_family == dst->sa_family)
1329 			sa = gateway;
1330 		else
1331 			sa = dst;
1332 		if (info->rti_ifp != NULL) {
1333 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
1334 			/* Case 4 */
1335 			if (info->rti_ifa == NULL && gateway != NULL)
1336 				info->rti_ifa = ifaof_ifpforaddr(gateway, info->rti_ifp);
1337 		} else if (dst != NULL && gateway != NULL)
1338 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
1339 							fibnum);
1340 		else if (sa != NULL)
1341 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
1342 							fibnum);
1343 	}
1344 	if (needref && info->rti_ifa != NULL) {
1345 		if (info->rti_ifp == NULL)
1346 			info->rti_ifp = info->rti_ifa->ifa_ifp;
1347 		ifa_ref(info->rti_ifa);
1348 	} else
1349 		error = ENETUNREACH;
1350 	NET_EPOCH_EXIT(et);
1351 	return (error);
1352 }
1353 
1354 static int
1355 if_updatemtu_cb(struct radix_node *rn, void *arg)
1356 {
1357 	struct rtentry *rt;
1358 	struct if_mtuinfo *ifmtu;
1359 
1360 	rt = (struct rtentry *)rn;
1361 	ifmtu = (struct if_mtuinfo *)arg;
1362 
1363 	if (rt->rt_ifp != ifmtu->ifp)
1364 		return (0);
1365 
1366 	if (rt->rt_mtu >= ifmtu->mtu) {
1367 		/* We have to decrease mtu regardless of flags */
1368 		rt->rt_mtu = ifmtu->mtu;
1369 		return (0);
1370 	}
1371 
1372 	/*
1373 	 * New MTU is bigger. Check if are allowed to alter it
1374 	 */
1375 	if ((rt->rt_flags & (RTF_FIXEDMTU | RTF_GATEWAY | RTF_HOST)) != 0) {
1376 
1377 		/*
1378 		 * Skip routes with user-supplied MTU and
1379 		 * non-interface routes
1380 		 */
1381 		return (0);
1382 	}
1383 
1384 	/* We are safe to update route MTU */
1385 	rt->rt_mtu = ifmtu->mtu;
1386 
1387 	return (0);
1388 }
1389 
1390 void
1391 rt_updatemtu(struct ifnet *ifp)
1392 {
1393 	struct if_mtuinfo ifmtu;
1394 	struct rib_head *rnh;
1395 	int i, j;
1396 
1397 	ifmtu.ifp = ifp;
1398 
1399 	/*
1400 	 * Try to update rt_mtu for all routes using this interface
1401 	 * Unfortunately the only way to do this is to traverse all
1402 	 * routing tables in all fibs/domains.
1403 	 */
1404 	for (i = 1; i <= AF_MAX; i++) {
1405 		ifmtu.mtu = if_getmtu_family(ifp, i);
1406 		for (j = 0; j < rt_numfibs; j++) {
1407 			rnh = rt_tables_get_rnh(j, i);
1408 			if (rnh == NULL)
1409 				continue;
1410 			RIB_WLOCK(rnh);
1411 			rnh->rnh_walktree(&rnh->head, if_updatemtu_cb, &ifmtu);
1412 			RIB_WUNLOCK(rnh);
1413 		}
1414 	}
1415 }
1416 
1417 
1418 #if 0
1419 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
1420 int rt_print(char *buf, int buflen, struct rtentry *rt);
1421 
1422 int
1423 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
1424 {
1425 	void *paddr = NULL;
1426 
1427 	switch (s->sa_family) {
1428 	case AF_INET:
1429 		paddr = &((struct sockaddr_in *)s)->sin_addr;
1430 		break;
1431 	case AF_INET6:
1432 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
1433 		break;
1434 	}
1435 
1436 	if (paddr == NULL)
1437 		return (0);
1438 
1439 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1440 		return (0);
1441 
1442 	return (strlen(buf));
1443 }
1444 
1445 int
1446 rt_print(char *buf, int buflen, struct rtentry *rt)
1447 {
1448 	struct sockaddr *addr, *mask;
1449 	int i = 0;
1450 
1451 	addr = rt_key(rt);
1452 	mask = rt_mask(rt);
1453 
1454 	i = p_sockaddr(buf, buflen, addr);
1455 	if (!(rt->rt_flags & RTF_HOST)) {
1456 		buf[i++] = '/';
1457 		i += p_sockaddr(buf + i, buflen - i, mask);
1458 	}
1459 
1460 	if (rt->rt_flags & RTF_GATEWAY) {
1461 		buf[i++] = '>';
1462 		i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1463 	}
1464 
1465 	return (i);
1466 }
1467 #endif
1468 
1469 #ifdef RADIX_MPATH
1470 /*
1471  * Deletes key for single-path routes, unlinks rtentry with
1472  * gateway specified in @info from multi-path routes.
1473  *
1474  * Returnes unlinked entry. In case of failure, returns NULL
1475  * and sets @perror to ESRCH.
1476  */
1477 static struct radix_node *
1478 rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info,
1479     struct rtentry *rto, int *perror)
1480 {
1481 	/*
1482 	 * if we got multipath routes, we require users to specify
1483 	 * a matching RTAX_GATEWAY.
1484 	 */
1485 	struct rtentry *rt; // *rto = NULL;
1486 	struct radix_node *rn;
1487 	struct sockaddr *gw;
1488 
1489 	gw = info->rti_info[RTAX_GATEWAY];
1490 	rt = rt_mpath_matchgate(rto, gw);
1491 	if (rt == NULL) {
1492 		*perror = ESRCH;
1493 		return (NULL);
1494 	}
1495 
1496 	/*
1497 	 * this is the first entry in the chain
1498 	 */
1499 	if (rto == rt) {
1500 		rn = rn_mpath_next((struct radix_node *)rt);
1501 		/*
1502 		 * there is another entry, now it's active
1503 		 */
1504 		if (rn) {
1505 			rto = RNTORT(rn);
1506 			RT_LOCK(rto);
1507 			rto->rt_flags |= RTF_UP;
1508 			RT_UNLOCK(rto);
1509 		} else if (rt->rt_flags & RTF_GATEWAY) {
1510 			/*
1511 			 * For gateway routes, we need to
1512 			 * make sure that we we are deleting
1513 			 * the correct gateway.
1514 			 * rt_mpath_matchgate() does not
1515 			 * check the case when there is only
1516 			 * one route in the chain.
1517 			 */
1518 			if (gw &&
1519 			    (rt->rt_gateway->sa_len != gw->sa_len ||
1520 				memcmp(rt->rt_gateway, gw, gw->sa_len))) {
1521 				*perror = ESRCH;
1522 				return (NULL);
1523 			}
1524 		}
1525 
1526 		/*
1527 		 * use the normal delete code to remove
1528 		 * the first entry
1529 		 */
1530 		rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1531 		*perror = 0;
1532 		return (rn);
1533 	}
1534 
1535 	/*
1536 	 * if the entry is 2nd and on up
1537 	 */
1538 	if (rt_mpath_deldup(rto, rt) == 0)
1539 		panic ("rtrequest1: rt_mpath_deldup");
1540 	*perror = 0;
1541 	rn = (struct radix_node *)rt;
1542 	return (rn);
1543 }
1544 #endif
1545 
1546 int
1547 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1548 				u_int fibnum)
1549 {
1550 	int error = 0;
1551 	struct rtentry *rt, *rt_old;
1552 	struct radix_node *rn;
1553 	struct rib_head *rnh;
1554 	struct ifaddr *ifa;
1555 	struct sockaddr *ndst;
1556 	struct sockaddr_storage mdst;
1557 
1558 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1559 	KASSERT((flags & RTF_RNH_LOCKED) == 0, ("rtrequest1_fib: locked"));
1560 	switch (dst->sa_family) {
1561 	case AF_INET6:
1562 	case AF_INET:
1563 		/* We support multiple FIBs. */
1564 		break;
1565 	default:
1566 		fibnum = RT_DEFAULT_FIB;
1567 		break;
1568 	}
1569 
1570 	/*
1571 	 * Find the correct routing tree to use for this Address Family
1572 	 */
1573 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1574 	if (rnh == NULL)
1575 		return (EAFNOSUPPORT);
1576 
1577 	/*
1578 	 * If we are adding a host route then we don't want to put
1579 	 * a netmask in the tree, nor do we want to clone it.
1580 	 */
1581 	if (flags & RTF_HOST)
1582 		netmask = NULL;
1583 
1584 	switch (req) {
1585 	case RTM_DELETE:
1586 		if (netmask) {
1587 			if (dst->sa_len > sizeof(mdst))
1588 				return (EINVAL);
1589 			rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1590 			dst = (struct sockaddr *)&mdst;
1591 		}
1592 
1593 		RIB_WLOCK(rnh);
1594 		rt = rt_unlinkrte(rnh, info, &error);
1595 		RIB_WUNLOCK(rnh);
1596 		if (error != 0)
1597 			return (error);
1598 
1599 		rt_notifydelete(rt, info);
1600 
1601 		/*
1602 		 * If the caller wants it, then it can have it,
1603 		 * but it's up to it to free the rtentry as we won't be
1604 		 * doing it.
1605 		 */
1606 		if (ret_nrt) {
1607 			*ret_nrt = rt;
1608 			RT_UNLOCK(rt);
1609 		} else
1610 			RTFREE_LOCKED(rt);
1611 		break;
1612 	case RTM_RESOLVE:
1613 		/*
1614 		 * resolve was only used for route cloning
1615 		 * here for compat
1616 		 */
1617 		break;
1618 	case RTM_ADD:
1619 		if ((flags & RTF_GATEWAY) && !gateway)
1620 			return (EINVAL);
1621 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1622 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1623 			return (EINVAL);
1624 
1625 		if (info->rti_ifa == NULL) {
1626 			error = rt_getifa_fib(info, fibnum);
1627 			if (error)
1628 				return (error);
1629 		}
1630 		rt = uma_zalloc(V_rtzone, M_NOWAIT);
1631 		if (rt == NULL) {
1632 			return (ENOBUFS);
1633 		}
1634 		rt->rt_flags = RTF_UP | flags;
1635 		rt->rt_fibnum = fibnum;
1636 		/*
1637 		 * Add the gateway. Possibly re-malloc-ing the storage for it.
1638 		 */
1639 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1640 			uma_zfree(V_rtzone, rt);
1641 			return (error);
1642 		}
1643 
1644 		/*
1645 		 * point to the (possibly newly malloc'd) dest address.
1646 		 */
1647 		ndst = (struct sockaddr *)rt_key(rt);
1648 
1649 		/*
1650 		 * make sure it contains the value we want (masked if needed).
1651 		 */
1652 		if (netmask) {
1653 			rt_maskedcopy(dst, ndst, netmask);
1654 		} else
1655 			bcopy(dst, ndst, dst->sa_len);
1656 
1657 		/*
1658 		 * We use the ifa reference returned by rt_getifa_fib().
1659 		 * This moved from below so that rnh->rnh_addaddr() can
1660 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1661 		 */
1662 		ifa = info->rti_ifa;
1663 		ifa_ref(ifa);
1664 		rt->rt_ifa = ifa;
1665 		rt->rt_ifp = ifa->ifa_ifp;
1666 		rt->rt_weight = 1;
1667 
1668 		rt_setmetrics(info, rt);
1669 
1670 		RIB_WLOCK(rnh);
1671 		RT_LOCK(rt);
1672 #ifdef RADIX_MPATH
1673 		/* do not permit exactly the same dst/mask/gw pair */
1674 		if (rt_mpath_capable(rnh) &&
1675 			rt_mpath_conflict(rnh, rt, netmask)) {
1676 			RIB_WUNLOCK(rnh);
1677 
1678 			ifa_free(rt->rt_ifa);
1679 			R_Free(rt_key(rt));
1680 			uma_zfree(V_rtzone, rt);
1681 			return (EEXIST);
1682 		}
1683 #endif
1684 
1685 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1686 		rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes);
1687 
1688 		rt_old = NULL;
1689 		if (rn == NULL && (info->rti_flags & RTF_PINNED) != 0) {
1690 
1691 			/*
1692 			 * Force removal and re-try addition
1693 			 * TODO: better multipath&pinned support
1694 			 */
1695 			struct sockaddr *info_dst = info->rti_info[RTAX_DST];
1696 			info->rti_info[RTAX_DST] = ndst;
1697 			/* Do not delete existing PINNED(interface) routes */
1698 			info->rti_flags &= ~RTF_PINNED;
1699 			rt_old = rt_unlinkrte(rnh, info, &error);
1700 			info->rti_flags |= RTF_PINNED;
1701 			info->rti_info[RTAX_DST] = info_dst;
1702 			if (rt_old != NULL)
1703 				rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head,
1704 				    rt->rt_nodes);
1705 		}
1706 		RIB_WUNLOCK(rnh);
1707 
1708 		if (rt_old != NULL)
1709 			RT_UNLOCK(rt_old);
1710 
1711 		/*
1712 		 * If it still failed to go into the tree,
1713 		 * then un-make it (this should be a function)
1714 		 */
1715 		if (rn == NULL) {
1716 			ifa_free(rt->rt_ifa);
1717 			R_Free(rt_key(rt));
1718 			uma_zfree(V_rtzone, rt);
1719 			return (EEXIST);
1720 		}
1721 
1722 		if (rt_old != NULL) {
1723 			rt_notifydelete(rt_old, info);
1724 			RTFREE(rt_old);
1725 		}
1726 
1727 		/*
1728 		 * If this protocol has something to add to this then
1729 		 * allow it to do that as well.
1730 		 */
1731 		if (ifa->ifa_rtrequest)
1732 			ifa->ifa_rtrequest(req, rt, info);
1733 
1734 		/*
1735 		 * actually return a resultant rtentry and
1736 		 * give the caller a single reference.
1737 		 */
1738 		if (ret_nrt) {
1739 			*ret_nrt = rt;
1740 			RT_ADDREF(rt);
1741 		}
1742 		rnh->rnh_gen++;		/* Routing table updated */
1743 		RT_UNLOCK(rt);
1744 		break;
1745 	case RTM_CHANGE:
1746 		RIB_WLOCK(rnh);
1747 		error = rtrequest1_fib_change(rnh, info, ret_nrt, fibnum);
1748 		RIB_WUNLOCK(rnh);
1749 		break;
1750 	default:
1751 		error = EOPNOTSUPP;
1752 	}
1753 
1754 	return (error);
1755 }
1756 
1757 #undef dst
1758 #undef gateway
1759 #undef netmask
1760 #undef ifaaddr
1761 #undef ifpaddr
1762 #undef flags
1763 
1764 static int
1765 rtrequest1_fib_change(struct rib_head *rnh, struct rt_addrinfo *info,
1766     struct rtentry **ret_nrt, u_int fibnum)
1767 {
1768 	struct rtentry *rt = NULL;
1769 	int error = 0;
1770 	int free_ifa = 0;
1771 	int family, mtu;
1772 	struct if_mtuinfo ifmtu;
1773 
1774 	RIB_WLOCK_ASSERT(rnh);
1775 
1776 	rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1777 	    info->rti_info[RTAX_NETMASK], &rnh->head);
1778 
1779 	if (rt == NULL)
1780 		return (ESRCH);
1781 
1782 #ifdef RADIX_MPATH
1783 	/*
1784 	 * If we got multipath routes,
1785 	 * we require users to specify a matching RTAX_GATEWAY.
1786 	 */
1787 	if (rt_mpath_capable(rnh)) {
1788 		rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]);
1789 		if (rt == NULL)
1790 			return (ESRCH);
1791 	}
1792 #endif
1793 
1794 	RT_LOCK(rt);
1795 
1796 	rt_setmetrics(info, rt);
1797 
1798 	/*
1799 	 * New gateway could require new ifaddr, ifp;
1800 	 * flags may also be different; ifp may be specified
1801 	 * by ll sockaddr when protocol address is ambiguous
1802 	 */
1803 	if (((rt->rt_flags & RTF_GATEWAY) &&
1804 	    info->rti_info[RTAX_GATEWAY] != NULL) ||
1805 	    info->rti_info[RTAX_IFP] != NULL ||
1806 	    (info->rti_info[RTAX_IFA] != NULL &&
1807 	     !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) {
1808 		/*
1809 		 * XXX: Temporarily set RTF_RNH_LOCKED flag in the rti_flags
1810 		 *	to avoid rlock in the ifa_ifwithroute().
1811 		 */
1812 		info->rti_flags |= RTF_RNH_LOCKED;
1813 		error = rt_getifa_fib(info, fibnum);
1814 		info->rti_flags &= ~RTF_RNH_LOCKED;
1815 		if (info->rti_ifa != NULL)
1816 			free_ifa = 1;
1817 
1818 		if (error != 0)
1819 			goto bad;
1820 	}
1821 
1822 	/* Check if outgoing interface has changed */
1823 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa &&
1824 	    rt->rt_ifa != NULL) {
1825 		if (rt->rt_ifa->ifa_rtrequest != NULL)
1826 			rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1827 		ifa_free(rt->rt_ifa);
1828 		rt->rt_ifa = NULL;
1829 	}
1830 	/* Update gateway address */
1831 	if (info->rti_info[RTAX_GATEWAY] != NULL) {
1832 		error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]);
1833 		if (error != 0)
1834 			goto bad;
1835 
1836 		rt->rt_flags &= ~RTF_GATEWAY;
1837 		rt->rt_flags |= (RTF_GATEWAY & info->rti_flags);
1838 	}
1839 
1840 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) {
1841 		ifa_ref(info->rti_ifa);
1842 		rt->rt_ifa = info->rti_ifa;
1843 		rt->rt_ifp = info->rti_ifp;
1844 	}
1845 	/* Allow some flags to be toggled on change. */
1846 	rt->rt_flags &= ~RTF_FMASK;
1847 	rt->rt_flags |= info->rti_flags & RTF_FMASK;
1848 
1849 	if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL)
1850 	       rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info);
1851 
1852 	/* Alter route MTU if necessary */
1853 	if (rt->rt_ifp != NULL) {
1854 		family = info->rti_info[RTAX_DST]->sa_family;
1855 		mtu = if_getmtu_family(rt->rt_ifp, family);
1856 		/* Set default MTU */
1857 		if (rt->rt_mtu == 0)
1858 			rt->rt_mtu = mtu;
1859 		if (rt->rt_mtu != mtu) {
1860 			/* Check if we really need to update */
1861 			ifmtu.ifp = rt->rt_ifp;
1862 			ifmtu.mtu = mtu;
1863 			if_updatemtu_cb(rt->rt_nodes, &ifmtu);
1864 		}
1865 	}
1866 
1867 	/*
1868 	 * This route change may have modified the route's gateway.  In that
1869 	 * case, any inpcbs that have cached this route need to invalidate their
1870 	 * llentry cache.
1871 	 */
1872 	rnh->rnh_gen++;
1873 
1874 	if (ret_nrt) {
1875 		*ret_nrt = rt;
1876 		RT_ADDREF(rt);
1877 	}
1878 bad:
1879 	RT_UNLOCK(rt);
1880 	if (free_ifa != 0) {
1881 		ifa_free(info->rti_ifa);
1882 		info->rti_ifa = NULL;
1883 	}
1884 	return (error);
1885 }
1886 
1887 static void
1888 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
1889 {
1890 
1891 	if (info->rti_mflags & RTV_MTU) {
1892 		if (info->rti_rmx->rmx_mtu != 0) {
1893 
1894 			/*
1895 			 * MTU was explicitly provided by user.
1896 			 * Keep it.
1897 			 */
1898 			rt->rt_flags |= RTF_FIXEDMTU;
1899 		} else {
1900 
1901 			/*
1902 			 * User explicitly sets MTU to 0.
1903 			 * Assume rollback to default.
1904 			 */
1905 			rt->rt_flags &= ~RTF_FIXEDMTU;
1906 		}
1907 		rt->rt_mtu = info->rti_rmx->rmx_mtu;
1908 	}
1909 	if (info->rti_mflags & RTV_WEIGHT)
1910 		rt->rt_weight = info->rti_rmx->rmx_weight;
1911 	/* Kernel -> userland timebase conversion. */
1912 	if (info->rti_mflags & RTV_EXPIRE)
1913 		rt->rt_expire = info->rti_rmx->rmx_expire ?
1914 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
1915 }
1916 
1917 int
1918 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1919 {
1920 	/* XXX dst may be overwritten, can we move this to below */
1921 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1922 
1923 	/*
1924 	 * Prepare to store the gateway in rt->rt_gateway.
1925 	 * Both dst and gateway are stored one after the other in the same
1926 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1927 	 * rt_gateway already points to the right place.
1928 	 * Otherwise, malloc a new block and update the 'dst' address.
1929 	 */
1930 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1931 		caddr_t new;
1932 
1933 		R_Malloc(new, caddr_t, dlen + glen);
1934 		if (new == NULL)
1935 			return ENOBUFS;
1936 		/*
1937 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1938 		 * rt_setgate() can be called to initialize a newly
1939 		 * allocated route entry, in which case rt_key(rt) == NULL
1940 		 * (and also rt->rt_gateway == NULL).
1941 		 * Free()/free() handle a NULL argument just fine.
1942 		 */
1943 		bcopy(dst, new, dlen);
1944 		R_Free(rt_key(rt));	/* free old block, if any */
1945 		rt_key(rt) = (struct sockaddr *)new;
1946 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1947 	}
1948 
1949 	/*
1950 	 * Copy the new gateway value into the memory chunk.
1951 	 */
1952 	bcopy(gate, rt->rt_gateway, glen);
1953 
1954 	return (0);
1955 }
1956 
1957 void
1958 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1959 {
1960 	u_char *cp1 = (u_char *)src;
1961 	u_char *cp2 = (u_char *)dst;
1962 	u_char *cp3 = (u_char *)netmask;
1963 	u_char *cplim = cp2 + *cp3;
1964 	u_char *cplim2 = cp2 + *cp1;
1965 
1966 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1967 	cp3 += 2;
1968 	if (cplim > cplim2)
1969 		cplim = cplim2;
1970 	while (cp2 < cplim)
1971 		*cp2++ = *cp1++ & *cp3++;
1972 	if (cp2 < cplim2)
1973 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1974 }
1975 
1976 /*
1977  * Set up a routing table entry, normally
1978  * for an interface.
1979  */
1980 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1981 static inline  int
1982 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1983 {
1984 	RIB_RLOCK_TRACKER;
1985 	struct sockaddr *dst;
1986 	struct sockaddr *netmask;
1987 	struct rtentry *rt = NULL;
1988 	struct rt_addrinfo info;
1989 	int error = 0;
1990 	int startfib, endfib;
1991 	char tempbuf[_SOCKADDR_TMPSIZE];
1992 	int didwork = 0;
1993 	int a_failure = 0;
1994 	struct sockaddr_dl *sdl = NULL;
1995 	struct rib_head *rnh;
1996 
1997 	if (flags & RTF_HOST) {
1998 		dst = ifa->ifa_dstaddr;
1999 		netmask = NULL;
2000 	} else {
2001 		dst = ifa->ifa_addr;
2002 		netmask = ifa->ifa_netmask;
2003 	}
2004 	if (dst->sa_len == 0)
2005 		return(EINVAL);
2006 	switch (dst->sa_family) {
2007 	case AF_INET6:
2008 	case AF_INET:
2009 		/* We support multiple FIBs. */
2010 		break;
2011 	default:
2012 		fibnum = RT_DEFAULT_FIB;
2013 		break;
2014 	}
2015 	if (fibnum == RT_ALL_FIBS) {
2016 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
2017 			startfib = endfib = ifa->ifa_ifp->if_fib;
2018 		else {
2019 			startfib = 0;
2020 			endfib = rt_numfibs - 1;
2021 		}
2022 	} else {
2023 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
2024 		startfib = fibnum;
2025 		endfib = fibnum;
2026 	}
2027 
2028 	/*
2029 	 * If it's a delete, check that if it exists,
2030 	 * it's on the correct interface or we might scrub
2031 	 * a route to another ifa which would
2032 	 * be confusing at best and possibly worse.
2033 	 */
2034 	if (cmd == RTM_DELETE) {
2035 		/*
2036 		 * It's a delete, so it should already exist..
2037 		 * If it's a net, mask off the host bits
2038 		 * (Assuming we have a mask)
2039 		 * XXX this is kinda inet specific..
2040 		 */
2041 		if (netmask != NULL) {
2042 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
2043 			dst = (struct sockaddr *)tempbuf;
2044 		}
2045 	} else if (cmd == RTM_ADD) {
2046 		sdl = (struct sockaddr_dl *)tempbuf;
2047 		bzero(sdl, sizeof(struct sockaddr_dl));
2048 		sdl->sdl_family = AF_LINK;
2049 		sdl->sdl_len = sizeof(struct sockaddr_dl);
2050 		sdl->sdl_type = ifa->ifa_ifp->if_type;
2051 		sdl->sdl_index = ifa->ifa_ifp->if_index;
2052         }
2053 	/*
2054 	 * Now go through all the requested tables (fibs) and do the
2055 	 * requested action. Realistically, this will either be fib 0
2056 	 * for protocols that don't do multiple tables or all the
2057 	 * tables for those that do.
2058 	 */
2059 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
2060 		if (cmd == RTM_DELETE) {
2061 			struct radix_node *rn;
2062 			/*
2063 			 * Look up an rtentry that is in the routing tree and
2064 			 * contains the correct info.
2065 			 */
2066 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
2067 			if (rnh == NULL)
2068 				/* this table doesn't exist but others might */
2069 				continue;
2070 			RIB_RLOCK(rnh);
2071 			rn = rnh->rnh_lookup(dst, netmask, &rnh->head);
2072 #ifdef RADIX_MPATH
2073 			if (rt_mpath_capable(rnh)) {
2074 
2075 				if (rn == NULL)
2076 					error = ESRCH;
2077 				else {
2078 					rt = RNTORT(rn);
2079 					/*
2080 					 * for interface route the
2081 					 * rt->rt_gateway is sockaddr_intf
2082 					 * for cloning ARP entries, so
2083 					 * rt_mpath_matchgate must use the
2084 					 * interface address
2085 					 */
2086 					rt = rt_mpath_matchgate(rt,
2087 					    ifa->ifa_addr);
2088 					if (rt == NULL)
2089 						error = ESRCH;
2090 				}
2091 			}
2092 #endif
2093 			error = (rn == NULL ||
2094 			    (rn->rn_flags & RNF_ROOT) ||
2095 			    RNTORT(rn)->rt_ifa != ifa);
2096 			RIB_RUNLOCK(rnh);
2097 			if (error) {
2098 				/* this is only an error if bad on ALL tables */
2099 				continue;
2100 			}
2101 		}
2102 		/*
2103 		 * Do the actual request
2104 		 */
2105 		bzero((caddr_t)&info, sizeof(info));
2106 		ifa_ref(ifa);
2107 		info.rti_ifa = ifa;
2108 		info.rti_flags = flags |
2109 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
2110 		info.rti_info[RTAX_DST] = dst;
2111 		/*
2112 		 * doing this for compatibility reasons
2113 		 */
2114 		if (cmd == RTM_ADD)
2115 			info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)sdl;
2116 		else
2117 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
2118 		info.rti_info[RTAX_NETMASK] = netmask;
2119 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
2120 
2121 		if (error == 0 && rt != NULL) {
2122 			/*
2123 			 * notify any listening routing agents of the change
2124 			 */
2125 			RT_LOCK(rt);
2126 #ifdef RADIX_MPATH
2127 			/*
2128 			 * in case address alias finds the first address
2129 			 * e.g. ifconfig bge0 192.0.2.246/24
2130 			 * e.g. ifconfig bge0 192.0.2.247/24
2131 			 * the address set in the route is 192.0.2.246
2132 			 * so we need to replace it with 192.0.2.247
2133 			 */
2134 			if (memcmp(rt->rt_ifa->ifa_addr,
2135 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
2136 				ifa_free(rt->rt_ifa);
2137 				ifa_ref(ifa);
2138 				rt->rt_ifp = ifa->ifa_ifp;
2139 				rt->rt_ifa = ifa;
2140 			}
2141 #endif
2142 			RT_ADDREF(rt);
2143 			RT_UNLOCK(rt);
2144 			rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
2145 			RT_LOCK(rt);
2146 			RT_REMREF(rt);
2147 			if (cmd == RTM_DELETE) {
2148 				/*
2149 				 * If we are deleting, and we found an entry,
2150 				 * then it's been removed from the tree..
2151 				 * now throw it away.
2152 				 */
2153 				RTFREE_LOCKED(rt);
2154 			} else {
2155 				if (cmd == RTM_ADD) {
2156 					/*
2157 					 * We just wanted to add it..
2158 					 * we don't actually need a reference.
2159 					 */
2160 					RT_REMREF(rt);
2161 				}
2162 				RT_UNLOCK(rt);
2163 			}
2164 			didwork = 1;
2165 		}
2166 		if (error)
2167 			a_failure = error;
2168 	}
2169 	if (cmd == RTM_DELETE) {
2170 		if (didwork) {
2171 			error = 0;
2172 		} else {
2173 			/* we only give an error if it wasn't in any table */
2174 			error = ((flags & RTF_HOST) ?
2175 			    EHOSTUNREACH : ENETUNREACH);
2176 		}
2177 	} else {
2178 		if (a_failure) {
2179 			/* return an error if any of them failed */
2180 			error = a_failure;
2181 		}
2182 	}
2183 	return (error);
2184 }
2185 
2186 /*
2187  * Set up a routing table entry, normally
2188  * for an interface.
2189  */
2190 int
2191 rtinit(struct ifaddr *ifa, int cmd, int flags)
2192 {
2193 	struct sockaddr *dst;
2194 	int fib = RT_DEFAULT_FIB;
2195 
2196 	if (flags & RTF_HOST) {
2197 		dst = ifa->ifa_dstaddr;
2198 	} else {
2199 		dst = ifa->ifa_addr;
2200 	}
2201 
2202 	switch (dst->sa_family) {
2203 	case AF_INET6:
2204 	case AF_INET:
2205 		/* We do support multiple FIBs. */
2206 		fib = RT_ALL_FIBS;
2207 		break;
2208 	}
2209 	return (rtinit1(ifa, cmd, flags, fib));
2210 }
2211 
2212 /*
2213  * Announce interface address arrival/withdraw
2214  * Returns 0 on success.
2215  */
2216 int
2217 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
2218 {
2219 
2220 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2221 	    ("unexpected cmd %d", cmd));
2222 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2223 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2224 
2225 	EVENTHANDLER_DIRECT_INVOKE(rt_addrmsg, ifa, cmd);
2226 	return (rtsock_addrmsg(cmd, ifa, fibnum));
2227 }
2228 
2229 /*
2230  * Announce route addition/removal.
2231  * Users of this function MUST validate input data BEFORE calling.
2232  * However we have to be able to handle invalid data:
2233  * if some userland app sends us "invalid" route message (invalid mask,
2234  * no dst, wrong address families, etc...) we need to pass it back
2235  * to app (and any other rtsock consumers) with rtm_errno field set to
2236  * non-zero value.
2237  * Returns 0 on success.
2238  */
2239 int
2240 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
2241     int fibnum)
2242 {
2243 
2244 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2245 	    ("unexpected cmd %d", cmd));
2246 
2247 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2248 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2249 
2250 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
2251 
2252 	return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
2253 }
2254 
2255 void
2256 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
2257 {
2258 
2259 	rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
2260 }
2261 
2262 /*
2263  * This is called to generate messages from the routing socket
2264  * indicating a network interface has had addresses associated with it.
2265  */
2266 void
2267 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
2268     int fibnum)
2269 {
2270 
2271 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2272 		("unexpected cmd %u", cmd));
2273 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2274 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2275 
2276 	if (cmd == RTM_ADD) {
2277 		rt_addrmsg(cmd, ifa, fibnum);
2278 		if (rt != NULL)
2279 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2280 	} else {
2281 		if (rt != NULL)
2282 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2283 		rt_addrmsg(cmd, ifa, fibnum);
2284 	}
2285 }
2286 
2287