xref: /freebsd/sys/net/route.c (revision a2aef24aa3c8458e4036735dd6928b4ef77294e5)
1 /*-
2  * Copyright (c) 1980, 1986, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)route.c	8.3.1.1 (Berkeley) 2/23/95
30  * $FreeBSD$
31  */
32 /************************************************************************
33  * Note: In this file a 'fib' is a "forwarding information base"	*
34  * Which is the new name for an in kernel routing (next hop) table.	*
35  ***********************************************************************/
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_route.h"
40 #include "opt_sctp.h"
41 #include "opt_mrouting.h"
42 #include "opt_mpath.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/socket.h>
49 #include <sys/sysctl.h>
50 #include <sys/syslog.h>
51 #include <sys/sysproto.h>
52 #include <sys/proc.h>
53 #include <sys/domain.h>
54 #include <sys/kernel.h>
55 
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/route.h>
60 #include <net/route_var.h>
61 #include <net/vnet.h>
62 
63 #ifdef RADIX_MPATH
64 #include <net/radix_mpath.h>
65 #endif
66 
67 #include <netinet/in.h>
68 #include <netinet/ip_mroute.h>
69 
70 #include <vm/uma.h>
71 
72 #define	RT_MAXFIBS	UINT16_MAX
73 
74 /* Kernel config default option. */
75 #ifdef ROUTETABLES
76 #if ROUTETABLES <= 0
77 #error "ROUTETABLES defined too low"
78 #endif
79 #if ROUTETABLES > RT_MAXFIBS
80 #error "ROUTETABLES defined too big"
81 #endif
82 #define	RT_NUMFIBS	ROUTETABLES
83 #endif /* ROUTETABLES */
84 /* Initialize to default if not otherwise set. */
85 #ifndef	RT_NUMFIBS
86 #define	RT_NUMFIBS	1
87 #endif
88 
89 #if defined(INET) || defined(INET6)
90 #ifdef SCTP
91 extern void sctp_addr_change(struct ifaddr *ifa, int cmd);
92 #endif /* SCTP */
93 #endif
94 
95 
96 /* This is read-only.. */
97 u_int rt_numfibs = RT_NUMFIBS;
98 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RDTUN, &rt_numfibs, 0, "");
99 
100 /*
101  * By default add routes to all fibs for new interfaces.
102  * Once this is set to 0 then only allocate routes on interface
103  * changes for the FIB of the caller when adding a new set of addresses
104  * to an interface.  XXX this is a shotgun aproach to a problem that needs
105  * a more fine grained solution.. that will come.
106  * XXX also has the problems getting the FIB from curthread which will not
107  * always work given the fib can be overridden and prefixes can be added
108  * from the network stack context.
109  */
110 VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1;
111 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET,
112     &VNET_NAME(rt_add_addr_allfibs), 0, "");
113 
114 VNET_DEFINE(struct rtstat, rtstat);
115 #define	V_rtstat	VNET(rtstat)
116 
117 VNET_DEFINE(struct rib_head *, rt_tables);
118 #define	V_rt_tables	VNET(rt_tables)
119 
120 VNET_DEFINE(int, rttrash);		/* routes not in table but not freed */
121 #define	V_rttrash	VNET(rttrash)
122 
123 
124 /*
125  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
126  * The operation can be done safely (in this code) because a
127  * 'struct rtentry' starts with two 'struct radix_node''s, the first
128  * one representing leaf nodes in the routing tree, which is
129  * what the code in radix.c passes us as a 'struct radix_node'.
130  *
131  * But because there are a lot of assumptions in this conversion,
132  * do not cast explicitly, but always use the macro below.
133  */
134 #define RNTORT(p)	((struct rtentry *)(p))
135 
136 static VNET_DEFINE(uma_zone_t, rtzone);		/* Routing table UMA zone. */
137 #define	V_rtzone	VNET(rtzone)
138 
139 static int rtrequest1_fib_change(struct rib_head *, struct rt_addrinfo *,
140     struct rtentry **, u_int);
141 static void rt_setmetrics(const struct rt_addrinfo *, struct rtentry *);
142 static int rt_ifdelroute(const struct rtentry *rt, void *arg);
143 static struct rtentry *rt_unlinkrte(struct rib_head *rnh,
144     struct rt_addrinfo *info, int *perror);
145 static void rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info);
146 #ifdef RADIX_MPATH
147 static struct radix_node *rt_mpath_unlink(struct rib_head *rnh,
148     struct rt_addrinfo *info, struct rtentry *rto, int *perror);
149 #endif
150 static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info,
151     int flags);
152 
153 struct if_mtuinfo
154 {
155 	struct ifnet	*ifp;
156 	int		mtu;
157 };
158 
159 static int	if_updatemtu_cb(struct radix_node *, void *);
160 
161 /*
162  * handler for net.my_fibnum
163  */
164 static int
165 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
166 {
167         int fibnum;
168         int error;
169 
170         fibnum = curthread->td_proc->p_fibnum;
171         error = sysctl_handle_int(oidp, &fibnum, 0, req);
172         return (error);
173 }
174 
175 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
176             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
177 
178 static __inline struct rib_head **
179 rt_tables_get_rnh_ptr(int table, int fam)
180 {
181 	struct rib_head **rnh;
182 
183 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
184 	    __func__));
185 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
186 	    __func__));
187 
188 	/* rnh is [fib=0][af=0]. */
189 	rnh = (struct rib_head **)V_rt_tables;
190 	/* Get the offset to the requested table and fam. */
191 	rnh += table * (AF_MAX+1) + fam;
192 
193 	return (rnh);
194 }
195 
196 struct rib_head *
197 rt_tables_get_rnh(int table, int fam)
198 {
199 
200 	return (*rt_tables_get_rnh_ptr(table, fam));
201 }
202 
203 u_int
204 rt_tables_get_gen(int table, int fam)
205 {
206 	struct rib_head *rnh;
207 
208 	rnh = *rt_tables_get_rnh_ptr(table, fam);
209 	KASSERT(rnh != NULL, ("%s: NULL rib_head pointer table %d fam %d",
210 	    __func__, table, fam));
211 	return (rnh->rnh_gen);
212 }
213 
214 
215 /*
216  * route initialization must occur before ip6_init2(), which happenas at
217  * SI_ORDER_MIDDLE.
218  */
219 static void
220 route_init(void)
221 {
222 
223 	/* whack the tunable ints into  line. */
224 	if (rt_numfibs > RT_MAXFIBS)
225 		rt_numfibs = RT_MAXFIBS;
226 	if (rt_numfibs == 0)
227 		rt_numfibs = 1;
228 }
229 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
230 
231 static int
232 rtentry_zinit(void *mem, int size, int how)
233 {
234 	struct rtentry *rt = mem;
235 
236 	rt->rt_pksent = counter_u64_alloc(how);
237 	if (rt->rt_pksent == NULL)
238 		return (ENOMEM);
239 
240 	RT_LOCK_INIT(rt);
241 
242 	return (0);
243 }
244 
245 static void
246 rtentry_zfini(void *mem, int size)
247 {
248 	struct rtentry *rt = mem;
249 
250 	RT_LOCK_DESTROY(rt);
251 	counter_u64_free(rt->rt_pksent);
252 }
253 
254 static int
255 rtentry_ctor(void *mem, int size, void *arg, int how)
256 {
257 	struct rtentry *rt = mem;
258 
259 	bzero(rt, offsetof(struct rtentry, rt_endzero));
260 	counter_u64_zero(rt->rt_pksent);
261 	rt->rt_chain = NULL;
262 
263 	return (0);
264 }
265 
266 static void
267 rtentry_dtor(void *mem, int size, void *arg)
268 {
269 	struct rtentry *rt = mem;
270 
271 	RT_UNLOCK_COND(rt);
272 }
273 
274 static void
275 vnet_route_init(const void *unused __unused)
276 {
277 	struct domain *dom;
278 	struct rib_head **rnh;
279 	int table;
280 	int fam;
281 
282 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
283 	    sizeof(struct rib_head *), M_RTABLE, M_WAITOK|M_ZERO);
284 
285 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
286 	    rtentry_ctor, rtentry_dtor,
287 	    rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
288 	for (dom = domains; dom; dom = dom->dom_next) {
289 		if (dom->dom_rtattach == NULL)
290 			continue;
291 
292 		for  (table = 0; table < rt_numfibs; table++) {
293 			fam = dom->dom_family;
294 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
295 				break;
296 
297 			rnh = rt_tables_get_rnh_ptr(table, fam);
298 			if (rnh == NULL)
299 				panic("%s: rnh NULL", __func__);
300 			dom->dom_rtattach((void **)rnh, 0);
301 		}
302 	}
303 }
304 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
305     vnet_route_init, 0);
306 
307 #ifdef VIMAGE
308 static void
309 vnet_route_uninit(const void *unused __unused)
310 {
311 	int table;
312 	int fam;
313 	struct domain *dom;
314 	struct rib_head **rnh;
315 
316 	for (dom = domains; dom; dom = dom->dom_next) {
317 		if (dom->dom_rtdetach == NULL)
318 			continue;
319 
320 		for (table = 0; table < rt_numfibs; table++) {
321 			fam = dom->dom_family;
322 
323 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
324 				break;
325 
326 			rnh = rt_tables_get_rnh_ptr(table, fam);
327 			if (rnh == NULL)
328 				panic("%s: rnh NULL", __func__);
329 			dom->dom_rtdetach((void **)rnh, 0);
330 		}
331 	}
332 
333 	free(V_rt_tables, M_RTABLE);
334 	uma_zdestroy(V_rtzone);
335 }
336 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
337     vnet_route_uninit, 0);
338 #endif
339 
340 struct rib_head *
341 rt_table_init(int offset)
342 {
343 	struct rib_head *rh;
344 
345 	rh = malloc(sizeof(struct rib_head), M_RTABLE, M_WAITOK | M_ZERO);
346 
347 	/* TODO: These details should be hidded inside radix.c */
348 	/* Init masks tree */
349 	rn_inithead_internal(&rh->head, rh->rnh_nodes, offset);
350 	rn_inithead_internal(&rh->rmhead.head, rh->rmhead.mask_nodes, 0);
351 	rh->head.rnh_masks = &rh->rmhead;
352 
353 	/* Init locks */
354 	RIB_LOCK_INIT(rh);
355 
356 	/* Finally, set base callbacks */
357 	rh->rnh_addaddr = rn_addroute;
358 	rh->rnh_deladdr = rn_delete;
359 	rh->rnh_matchaddr = rn_match;
360 	rh->rnh_lookup = rn_lookup;
361 	rh->rnh_walktree = rn_walktree;
362 	rh->rnh_walktree_from = rn_walktree_from;
363 
364 	return (rh);
365 }
366 
367 static int
368 rt_freeentry(struct radix_node *rn, void *arg)
369 {
370 	struct radix_head * const rnh = arg;
371 	struct radix_node *x;
372 
373 	x = (struct radix_node *)rn_delete(rn + 2, NULL, rnh);
374 	if (x != NULL)
375 		R_Free(x);
376 	return (0);
377 }
378 
379 void
380 rt_table_destroy(struct rib_head *rh)
381 {
382 
383 	rn_walktree(&rh->rmhead.head, rt_freeentry, &rh->rmhead.head);
384 
385 	/* Assume table is already empty */
386 	RIB_LOCK_DESTROY(rh);
387 	free(rh, M_RTABLE);
388 }
389 
390 
391 #ifndef _SYS_SYSPROTO_H_
392 struct setfib_args {
393 	int     fibnum;
394 };
395 #endif
396 int
397 sys_setfib(struct thread *td, struct setfib_args *uap)
398 {
399 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
400 		return EINVAL;
401 	td->td_proc->p_fibnum = uap->fibnum;
402 	return (0);
403 }
404 
405 /*
406  * Packet routing routines.
407  */
408 void
409 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
410 {
411 	struct rtentry *rt;
412 
413 	if ((rt = ro->ro_rt) != NULL) {
414 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
415 			return;
416 		RTFREE(rt);
417 		ro->ro_rt = NULL;
418 	}
419 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
420 	if (ro->ro_rt)
421 		RT_UNLOCK(ro->ro_rt);
422 }
423 
424 /*
425  * Look up the route that matches the address given
426  * Or, at least try.. Create a cloned route if needed.
427  *
428  * The returned route, if any, is locked.
429  */
430 struct rtentry *
431 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
432 {
433 
434 	return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
435 }
436 
437 struct rtentry *
438 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
439 		    u_int fibnum)
440 {
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 	ifa = NULL;
595 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
596 	if (rnh == NULL) {
597 		error = EAFNOSUPPORT;
598 		goto out;
599 	}
600 
601 	/* verify the gateway is directly reachable */
602 	if ((ifa = ifa_ifwithnet(gateway, 0, fibnum)) == NULL) {
603 		error = ENETUNREACH;
604 		goto out;
605 	}
606 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
607 	/*
608 	 * If the redirect isn't from our current router for this dst,
609 	 * it's either old or wrong.  If it redirects us to ourselves,
610 	 * we have a routing loop, perhaps as a result of an interface
611 	 * going down recently.
612 	 */
613 	if (!(flags & RTF_DONE) && rt) {
614 		if (!sa_equal(src, rt->rt_gateway)) {
615 			error = EINVAL;
616 			goto done;
617 		}
618 		if (rt->rt_ifa != ifa && ifa->ifa_addr->sa_family != AF_LINK) {
619 			error = EINVAL;
620 			goto done;
621 		}
622 	}
623 	if ((flags & RTF_GATEWAY) && ifa_ifwithaddr_check(gateway)) {
624 		error = EHOSTUNREACH;
625 		goto done;
626 	}
627 	/*
628 	 * Create a new entry if we just got back a wildcard entry
629 	 * or the lookup failed.  This is necessary for hosts
630 	 * which use routing redirects generated by smart gateways
631 	 * to dynamically build the routing tables.
632 	 */
633 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
634 		goto create;
635 	/*
636 	 * Don't listen to the redirect if it's
637 	 * for a route to an interface.
638 	 */
639 	if (rt->rt_flags & RTF_GATEWAY) {
640 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
641 			/*
642 			 * Changing from route to net => route to host.
643 			 * Create new route, rather than smashing route to net.
644 			 */
645 		create:
646 			if (rt != NULL)
647 				RTFREE_LOCKED(rt);
648 
649 			flags |= RTF_DYNAMIC;
650 			bzero((caddr_t)&info, sizeof(info));
651 			info.rti_info[RTAX_DST] = dst;
652 			info.rti_info[RTAX_GATEWAY] = gateway;
653 			info.rti_info[RTAX_NETMASK] = netmask;
654 			info.rti_ifa = ifa;
655 			info.rti_flags = flags;
656 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
657 			if (rt != NULL) {
658 				RT_LOCK(rt);
659 				flags = rt->rt_flags;
660 			}
661 
662 			stat = &V_rtstat.rts_dynamic;
663 		} else {
664 
665 			/*
666 			 * Smash the current notion of the gateway to
667 			 * this destination.  Should check about netmask!!!
668 			 */
669 			if ((flags & RTF_GATEWAY) == 0)
670 				rt->rt_flags &= ~RTF_GATEWAY;
671 			rt->rt_flags |= RTF_MODIFIED;
672 			flags |= RTF_MODIFIED;
673 			stat = &V_rtstat.rts_newgateway;
674 			/*
675 			 * add the key and gateway (in one malloc'd chunk).
676 			 */
677 			RT_UNLOCK(rt);
678 			RIB_WLOCK(rnh);
679 			RT_LOCK(rt);
680 			rt_setgate(rt, rt_key(rt), gateway);
681 			RIB_WUNLOCK(rnh);
682 		}
683 	} else
684 		error = EHOSTUNREACH;
685 done:
686 	if (rt)
687 		RTFREE_LOCKED(rt);
688 out:
689 	if (error)
690 		V_rtstat.rts_badredirect++;
691 	else if (stat != NULL)
692 		(*stat)++;
693 	bzero((caddr_t)&info, sizeof(info));
694 	info.rti_info[RTAX_DST] = dst;
695 	info.rti_info[RTAX_GATEWAY] = gateway;
696 	info.rti_info[RTAX_NETMASK] = netmask;
697 	info.rti_info[RTAX_AUTHOR] = src;
698 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
699 	if (ifa != NULL)
700 		ifa_free(ifa);
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 	if ((flags & RTF_GATEWAY) == 0) {
732 		/*
733 		 * If we are adding a route to an interface,
734 		 * and the interface is a pt to pt link
735 		 * we should search for the destination
736 		 * as our clue to the interface.  Otherwise
737 		 * we can use the local address.
738 		 */
739 		ifa = NULL;
740 		if (flags & RTF_HOST)
741 			ifa = ifa_ifwithdstaddr(dst, fibnum);
742 		if (ifa == NULL)
743 			ifa = ifa_ifwithaddr(gateway);
744 	} else {
745 		/*
746 		 * If we are adding a route to a remote net
747 		 * or host, the gateway may still be on the
748 		 * other end of a pt to pt link.
749 		 */
750 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
751 	}
752 	if (ifa == NULL)
753 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
754 	if (ifa == NULL) {
755 		struct rtentry *rt;
756 
757 		rt = rtalloc1_fib(gateway, 0, flags, fibnum);
758 		if (rt == NULL)
759 			return (NULL);
760 		/*
761 		 * dismiss a gateway that is reachable only
762 		 * through the default router
763 		 */
764 		switch (gateway->sa_family) {
765 		case AF_INET:
766 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
767 				not_found = 1;
768 			break;
769 		case AF_INET6:
770 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
771 				not_found = 1;
772 			break;
773 		default:
774 			break;
775 		}
776 		if (!not_found && rt->rt_ifa != NULL) {
777 			ifa = rt->rt_ifa;
778 			ifa_ref(ifa);
779 		}
780 		RT_REMREF(rt);
781 		RT_UNLOCK(rt);
782 		if (not_found || ifa == NULL)
783 			return (NULL);
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 		else
791 			ifa_free(oifa);
792 	}
793 	return (ifa);
794 }
795 
796 /*
797  * Do appropriate manipulations of a routing tree given
798  * all the bits of info needed
799  */
800 int
801 rtrequest_fib(int req,
802 	struct sockaddr *dst,
803 	struct sockaddr *gateway,
804 	struct sockaddr *netmask,
805 	int flags,
806 	struct rtentry **ret_nrt,
807 	u_int fibnum)
808 {
809 	struct rt_addrinfo info;
810 
811 	if (dst->sa_len == 0)
812 		return(EINVAL);
813 
814 	bzero((caddr_t)&info, sizeof(info));
815 	info.rti_flags = flags;
816 	info.rti_info[RTAX_DST] = dst;
817 	info.rti_info[RTAX_GATEWAY] = gateway;
818 	info.rti_info[RTAX_NETMASK] = netmask;
819 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
820 }
821 
822 
823 /*
824  * Copy most of @rt data into @info.
825  *
826  * If @flags contains NHR_COPY, copies dst,netmask and gw to the
827  * pointers specified by @info structure. Assume such pointers
828  * are zeroed sockaddr-like structures with sa_len field initialized
829  * to reflect size of the provided buffer. if no NHR_COPY is specified,
830  * point dst,netmask and gw @info fields to appropriate @rt values.
831  *
832  * if @flags contains NHR_REF, do refcouting on rt_ifp.
833  *
834  * Returns 0 on success.
835  */
836 int
837 rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags)
838 {
839 	struct rt_metrics *rmx;
840 	struct sockaddr *src, *dst;
841 	int sa_len;
842 
843 	if (flags & NHR_COPY) {
844 		/* Copy destination if dst is non-zero */
845 		src = rt_key(rt);
846 		dst = info->rti_info[RTAX_DST];
847 		sa_len = src->sa_len;
848 		if (dst != NULL) {
849 			if (src->sa_len > dst->sa_len)
850 				return (ENOMEM);
851 			memcpy(dst, src, src->sa_len);
852 			info->rti_addrs |= RTA_DST;
853 		}
854 
855 		/* Copy mask if set && dst is non-zero */
856 		src = rt_mask(rt);
857 		dst = info->rti_info[RTAX_NETMASK];
858 		if (src != NULL && dst != NULL) {
859 
860 			/*
861 			 * Radix stores different value in sa_len,
862 			 * assume rt_mask() to have the same length
863 			 * as rt_key()
864 			 */
865 			if (sa_len > dst->sa_len)
866 				return (ENOMEM);
867 			memcpy(dst, src, src->sa_len);
868 			info->rti_addrs |= RTA_NETMASK;
869 		}
870 
871 		/* Copy gateway is set && dst is non-zero */
872 		src = rt->rt_gateway;
873 		dst = info->rti_info[RTAX_GATEWAY];
874 		if ((rt->rt_flags & RTF_GATEWAY) && src != NULL && dst != NULL){
875 			if (src->sa_len > dst->sa_len)
876 				return (ENOMEM);
877 			memcpy(dst, src, src->sa_len);
878 			info->rti_addrs |= RTA_GATEWAY;
879 		}
880 	} else {
881 		info->rti_info[RTAX_DST] = rt_key(rt);
882 		info->rti_addrs |= RTA_DST;
883 		if (rt_mask(rt) != NULL) {
884 			info->rti_info[RTAX_NETMASK] = rt_mask(rt);
885 			info->rti_addrs |= RTA_NETMASK;
886 		}
887 		if (rt->rt_flags & RTF_GATEWAY) {
888 			info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
889 			info->rti_addrs |= RTA_GATEWAY;
890 		}
891 	}
892 
893 	rmx = info->rti_rmx;
894 	if (rmx != NULL) {
895 		info->rti_mflags |= RTV_MTU;
896 		rmx->rmx_mtu = rt->rt_mtu;
897 	}
898 
899 	info->rti_flags = rt->rt_flags;
900 	info->rti_ifp = rt->rt_ifp;
901 	info->rti_ifa = rt->rt_ifa;
902 
903 	if (flags & NHR_REF) {
904 		/* Do 'traditional' refcouting */
905 		if_ref(info->rti_ifp);
906 	}
907 
908 	return (0);
909 }
910 
911 /*
912  * Lookups up route entry for @dst in RIB database for fib @fibnum.
913  * Exports entry data to @info using rt_exportinfo().
914  *
915  * if @flags contains NHR_REF, refcouting is performed on rt_ifp.
916  *   All references can be released later by calling rib_free_info()
917  *
918  * Returns 0 on success.
919  * Returns ENOENT for lookup failure, ENOMEM for export failure.
920  */
921 int
922 rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags,
923     uint32_t flowid, struct rt_addrinfo *info)
924 {
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 int
1275 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
1276 {
1277 	struct ifaddr *ifa;
1278 	int error = 0;
1279 
1280 	/*
1281 	 * ifp may be specified by sockaddr_dl
1282 	 * when protocol address is ambiguous.
1283 	 */
1284 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
1285 	    ifpaddr->sa_family == AF_LINK &&
1286 	    (ifa = ifa_ifwithnet(ifpaddr, 0, fibnum)) != NULL) {
1287 		info->rti_ifp = ifa->ifa_ifp;
1288 		ifa_free(ifa);
1289 	}
1290 	if (info->rti_ifa == NULL && ifaaddr != NULL)
1291 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
1292 	if (info->rti_ifa == NULL) {
1293 		struct sockaddr *sa;
1294 
1295 		sa = ifaaddr != NULL ? ifaaddr :
1296 		    (gateway != NULL ? gateway : dst);
1297 		if (sa != NULL && info->rti_ifp != NULL)
1298 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
1299 		else if (dst != NULL && gateway != NULL)
1300 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
1301 							fibnum);
1302 		else if (sa != NULL)
1303 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
1304 							fibnum);
1305 	}
1306 	if ((ifa = info->rti_ifa) != NULL) {
1307 		if (info->rti_ifp == NULL)
1308 			info->rti_ifp = ifa->ifa_ifp;
1309 	} else
1310 		error = ENETUNREACH;
1311 	return (error);
1312 }
1313 
1314 static int
1315 if_updatemtu_cb(struct radix_node *rn, void *arg)
1316 {
1317 	struct rtentry *rt;
1318 	struct if_mtuinfo *ifmtu;
1319 
1320 	rt = (struct rtentry *)rn;
1321 	ifmtu = (struct if_mtuinfo *)arg;
1322 
1323 	if (rt->rt_ifp != ifmtu->ifp)
1324 		return (0);
1325 
1326 	if (rt->rt_mtu >= ifmtu->mtu) {
1327 		/* We have to decrease mtu regardless of flags */
1328 		rt->rt_mtu = ifmtu->mtu;
1329 		return (0);
1330 	}
1331 
1332 	/*
1333 	 * New MTU is bigger. Check if are allowed to alter it
1334 	 */
1335 	if ((rt->rt_flags & (RTF_FIXEDMTU | RTF_GATEWAY | RTF_HOST)) != 0) {
1336 
1337 		/*
1338 		 * Skip routes with user-supplied MTU and
1339 		 * non-interface routes
1340 		 */
1341 		return (0);
1342 	}
1343 
1344 	/* We are safe to update route MTU */
1345 	rt->rt_mtu = ifmtu->mtu;
1346 
1347 	return (0);
1348 }
1349 
1350 void
1351 rt_updatemtu(struct ifnet *ifp)
1352 {
1353 	struct if_mtuinfo ifmtu;
1354 	struct rib_head *rnh;
1355 	int i, j;
1356 
1357 	ifmtu.ifp = ifp;
1358 
1359 	/*
1360 	 * Try to update rt_mtu for all routes using this interface
1361 	 * Unfortunately the only way to do this is to traverse all
1362 	 * routing tables in all fibs/domains.
1363 	 */
1364 	for (i = 1; i <= AF_MAX; i++) {
1365 		ifmtu.mtu = if_getmtu_family(ifp, i);
1366 		for (j = 0; j < rt_numfibs; j++) {
1367 			rnh = rt_tables_get_rnh(j, i);
1368 			if (rnh == NULL)
1369 				continue;
1370 			RIB_WLOCK(rnh);
1371 			rnh->rnh_walktree(&rnh->head, if_updatemtu_cb, &ifmtu);
1372 			RIB_WUNLOCK(rnh);
1373 		}
1374 	}
1375 }
1376 
1377 
1378 #if 0
1379 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
1380 int rt_print(char *buf, int buflen, struct rtentry *rt);
1381 
1382 int
1383 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
1384 {
1385 	void *paddr = NULL;
1386 
1387 	switch (s->sa_family) {
1388 	case AF_INET:
1389 		paddr = &((struct sockaddr_in *)s)->sin_addr;
1390 		break;
1391 	case AF_INET6:
1392 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
1393 		break;
1394 	}
1395 
1396 	if (paddr == NULL)
1397 		return (0);
1398 
1399 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1400 		return (0);
1401 
1402 	return (strlen(buf));
1403 }
1404 
1405 int
1406 rt_print(char *buf, int buflen, struct rtentry *rt)
1407 {
1408 	struct sockaddr *addr, *mask;
1409 	int i = 0;
1410 
1411 	addr = rt_key(rt);
1412 	mask = rt_mask(rt);
1413 
1414 	i = p_sockaddr(buf, buflen, addr);
1415 	if (!(rt->rt_flags & RTF_HOST)) {
1416 		buf[i++] = '/';
1417 		i += p_sockaddr(buf + i, buflen - i, mask);
1418 	}
1419 
1420 	if (rt->rt_flags & RTF_GATEWAY) {
1421 		buf[i++] = '>';
1422 		i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1423 	}
1424 
1425 	return (i);
1426 }
1427 #endif
1428 
1429 #ifdef RADIX_MPATH
1430 /*
1431  * Deletes key for single-path routes, unlinks rtentry with
1432  * gateway specified in @info from multi-path routes.
1433  *
1434  * Returnes unlinked entry. In case of failure, returns NULL
1435  * and sets @perror to ESRCH.
1436  */
1437 static struct radix_node *
1438 rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info,
1439     struct rtentry *rto, int *perror)
1440 {
1441 	/*
1442 	 * if we got multipath routes, we require users to specify
1443 	 * a matching RTAX_GATEWAY.
1444 	 */
1445 	struct rtentry *rt; // *rto = NULL;
1446 	struct radix_node *rn;
1447 	struct sockaddr *gw;
1448 
1449 	gw = info->rti_info[RTAX_GATEWAY];
1450 	rt = rt_mpath_matchgate(rto, gw);
1451 	if (rt == NULL) {
1452 		*perror = ESRCH;
1453 		return (NULL);
1454 	}
1455 
1456 	/*
1457 	 * this is the first entry in the chain
1458 	 */
1459 	if (rto == rt) {
1460 		rn = rn_mpath_next((struct radix_node *)rt);
1461 		/*
1462 		 * there is another entry, now it's active
1463 		 */
1464 		if (rn) {
1465 			rto = RNTORT(rn);
1466 			RT_LOCK(rto);
1467 			rto->rt_flags |= RTF_UP;
1468 			RT_UNLOCK(rto);
1469 		} else if (rt->rt_flags & RTF_GATEWAY) {
1470 			/*
1471 			 * For gateway routes, we need to
1472 			 * make sure that we we are deleting
1473 			 * the correct gateway.
1474 			 * rt_mpath_matchgate() does not
1475 			 * check the case when there is only
1476 			 * one route in the chain.
1477 			 */
1478 			if (gw &&
1479 			    (rt->rt_gateway->sa_len != gw->sa_len ||
1480 				memcmp(rt->rt_gateway, gw, gw->sa_len))) {
1481 				*perror = ESRCH;
1482 				return (NULL);
1483 			}
1484 		}
1485 
1486 		/*
1487 		 * use the normal delete code to remove
1488 		 * the first entry
1489 		 */
1490 		rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1491 		*perror = 0;
1492 		return (rn);
1493 	}
1494 
1495 	/*
1496 	 * if the entry is 2nd and on up
1497 	 */
1498 	if (rt_mpath_deldup(rto, rt) == 0)
1499 		panic ("rtrequest1: rt_mpath_deldup");
1500 	*perror = 0;
1501 	rn = (struct radix_node *)rt;
1502 	return (rn);
1503 }
1504 #endif
1505 
1506 int
1507 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1508 				u_int fibnum)
1509 {
1510 	int error = 0;
1511 	struct rtentry *rt, *rt_old;
1512 	struct radix_node *rn;
1513 	struct rib_head *rnh;
1514 	struct ifaddr *ifa;
1515 	struct sockaddr *ndst;
1516 	struct sockaddr_storage mdst;
1517 
1518 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1519 	KASSERT((flags & RTF_RNH_LOCKED) == 0, ("rtrequest1_fib: locked"));
1520 	switch (dst->sa_family) {
1521 	case AF_INET6:
1522 	case AF_INET:
1523 		/* We support multiple FIBs. */
1524 		break;
1525 	default:
1526 		fibnum = RT_DEFAULT_FIB;
1527 		break;
1528 	}
1529 
1530 	/*
1531 	 * Find the correct routing tree to use for this Address Family
1532 	 */
1533 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1534 	if (rnh == NULL)
1535 		return (EAFNOSUPPORT);
1536 
1537 	/*
1538 	 * If we are adding a host route then we don't want to put
1539 	 * a netmask in the tree, nor do we want to clone it.
1540 	 */
1541 	if (flags & RTF_HOST)
1542 		netmask = NULL;
1543 
1544 	switch (req) {
1545 	case RTM_DELETE:
1546 		if (netmask) {
1547 			rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1548 			dst = (struct sockaddr *)&mdst;
1549 		}
1550 
1551 		RIB_WLOCK(rnh);
1552 		rt = rt_unlinkrte(rnh, info, &error);
1553 		RIB_WUNLOCK(rnh);
1554 		if (error != 0)
1555 			return (error);
1556 
1557 		rt_notifydelete(rt, info);
1558 
1559 		/*
1560 		 * If the caller wants it, then it can have it,
1561 		 * but it's up to it to free the rtentry as we won't be
1562 		 * doing it.
1563 		 */
1564 		if (ret_nrt) {
1565 			*ret_nrt = rt;
1566 			RT_UNLOCK(rt);
1567 		} else
1568 			RTFREE_LOCKED(rt);
1569 		break;
1570 	case RTM_RESOLVE:
1571 		/*
1572 		 * resolve was only used for route cloning
1573 		 * here for compat
1574 		 */
1575 		break;
1576 	case RTM_ADD:
1577 		if ((flags & RTF_GATEWAY) && !gateway)
1578 			return (EINVAL);
1579 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1580 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1581 			return (EINVAL);
1582 
1583 		if (info->rti_ifa == NULL) {
1584 			error = rt_getifa_fib(info, fibnum);
1585 			if (error)
1586 				return (error);
1587 		} else
1588 			ifa_ref(info->rti_ifa);
1589 		ifa = info->rti_ifa;
1590 		rt = uma_zalloc(V_rtzone, M_NOWAIT);
1591 		if (rt == NULL) {
1592 			ifa_free(ifa);
1593 			return (ENOBUFS);
1594 		}
1595 		rt->rt_flags = RTF_UP | flags;
1596 		rt->rt_fibnum = fibnum;
1597 		/*
1598 		 * Add the gateway. Possibly re-malloc-ing the storage for it.
1599 		 */
1600 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1601 			ifa_free(ifa);
1602 			uma_zfree(V_rtzone, rt);
1603 			return (error);
1604 		}
1605 
1606 		/*
1607 		 * point to the (possibly newly malloc'd) dest address.
1608 		 */
1609 		ndst = (struct sockaddr *)rt_key(rt);
1610 
1611 		/*
1612 		 * make sure it contains the value we want (masked if needed).
1613 		 */
1614 		if (netmask) {
1615 			rt_maskedcopy(dst, ndst, netmask);
1616 		} else
1617 			bcopy(dst, ndst, dst->sa_len);
1618 
1619 		/*
1620 		 * We use the ifa reference returned by rt_getifa_fib().
1621 		 * This moved from below so that rnh->rnh_addaddr() can
1622 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1623 		 */
1624 		rt->rt_ifa = ifa;
1625 		rt->rt_ifp = ifa->ifa_ifp;
1626 		rt->rt_weight = 1;
1627 
1628 		rt_setmetrics(info, rt);
1629 
1630 		RIB_WLOCK(rnh);
1631 		RT_LOCK(rt);
1632 #ifdef RADIX_MPATH
1633 		/* do not permit exactly the same dst/mask/gw pair */
1634 		if (rt_mpath_capable(rnh) &&
1635 			rt_mpath_conflict(rnh, rt, netmask)) {
1636 			RIB_WUNLOCK(rnh);
1637 
1638 			ifa_free(rt->rt_ifa);
1639 			R_Free(rt_key(rt));
1640 			uma_zfree(V_rtzone, rt);
1641 			return (EEXIST);
1642 		}
1643 #endif
1644 
1645 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1646 		rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes);
1647 
1648 		rt_old = NULL;
1649 		if (rn == NULL && (info->rti_flags & RTF_PINNED) != 0) {
1650 
1651 			/*
1652 			 * Force removal and re-try addition
1653 			 * TODO: better multipath&pinned support
1654 			 */
1655 			struct sockaddr *info_dst = info->rti_info[RTAX_DST];
1656 			info->rti_info[RTAX_DST] = ndst;
1657 			/* Do not delete existing PINNED(interface) routes */
1658 			info->rti_flags &= ~RTF_PINNED;
1659 			rt_old = rt_unlinkrte(rnh, info, &error);
1660 			info->rti_flags |= RTF_PINNED;
1661 			info->rti_info[RTAX_DST] = info_dst;
1662 			if (rt_old != NULL)
1663 				rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head,
1664 				    rt->rt_nodes);
1665 		}
1666 		RIB_WUNLOCK(rnh);
1667 
1668 		if (rt_old != NULL)
1669 			RT_UNLOCK(rt_old);
1670 
1671 		/*
1672 		 * If it still failed to go into the tree,
1673 		 * then un-make it (this should be a function)
1674 		 */
1675 		if (rn == NULL) {
1676 			ifa_free(rt->rt_ifa);
1677 			R_Free(rt_key(rt));
1678 			uma_zfree(V_rtzone, rt);
1679 			return (EEXIST);
1680 		}
1681 
1682 		if (rt_old != NULL) {
1683 			rt_notifydelete(rt_old, info);
1684 			RTFREE(rt_old);
1685 		}
1686 
1687 		/*
1688 		 * If this protocol has something to add to this then
1689 		 * allow it to do that as well.
1690 		 */
1691 		if (ifa->ifa_rtrequest)
1692 			ifa->ifa_rtrequest(req, rt, info);
1693 
1694 		/*
1695 		 * actually return a resultant rtentry and
1696 		 * give the caller a single reference.
1697 		 */
1698 		if (ret_nrt) {
1699 			*ret_nrt = rt;
1700 			RT_ADDREF(rt);
1701 		}
1702 		rnh->rnh_gen++;		/* Routing table updated */
1703 		RT_UNLOCK(rt);
1704 		break;
1705 	case RTM_CHANGE:
1706 		RIB_WLOCK(rnh);
1707 		error = rtrequest1_fib_change(rnh, info, ret_nrt, fibnum);
1708 		RIB_WUNLOCK(rnh);
1709 		break;
1710 	default:
1711 		error = EOPNOTSUPP;
1712 	}
1713 
1714 	return (error);
1715 }
1716 
1717 #undef dst
1718 #undef gateway
1719 #undef netmask
1720 #undef ifaaddr
1721 #undef ifpaddr
1722 #undef flags
1723 
1724 static int
1725 rtrequest1_fib_change(struct rib_head *rnh, struct rt_addrinfo *info,
1726     struct rtentry **ret_nrt, u_int fibnum)
1727 {
1728 	struct rtentry *rt = NULL;
1729 	int error = 0;
1730 	int free_ifa = 0;
1731 	int family, mtu;
1732 	struct if_mtuinfo ifmtu;
1733 
1734 	rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1735 	    info->rti_info[RTAX_NETMASK], &rnh->head);
1736 
1737 	if (rt == NULL)
1738 		return (ESRCH);
1739 
1740 #ifdef RADIX_MPATH
1741 	/*
1742 	 * If we got multipath routes,
1743 	 * we require users to specify a matching RTAX_GATEWAY.
1744 	 */
1745 	if (rt_mpath_capable(rnh)) {
1746 		rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]);
1747 		if (rt == NULL)
1748 			return (ESRCH);
1749 	}
1750 #endif
1751 
1752 	RT_LOCK(rt);
1753 
1754 	rt_setmetrics(info, rt);
1755 
1756 	/*
1757 	 * New gateway could require new ifaddr, ifp;
1758 	 * flags may also be different; ifp may be specified
1759 	 * by ll sockaddr when protocol address is ambiguous
1760 	 */
1761 	if (((rt->rt_flags & RTF_GATEWAY) &&
1762 	    info->rti_info[RTAX_GATEWAY] != NULL) ||
1763 	    info->rti_info[RTAX_IFP] != NULL ||
1764 	    (info->rti_info[RTAX_IFA] != NULL &&
1765 	     !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) {
1766 		/*
1767 		 * XXX: Temporarily set RTF_RNH_LOCKED flag in the rti_flags
1768 		 *	to avoid rlock in the ifa_ifwithroute().
1769 		 */
1770 		info->rti_flags |= RTF_RNH_LOCKED;
1771 		error = rt_getifa_fib(info, fibnum);
1772 		info->rti_flags &= ~RTF_RNH_LOCKED;
1773 		if (info->rti_ifa != NULL)
1774 			free_ifa = 1;
1775 
1776 		if (error != 0)
1777 			goto bad;
1778 	}
1779 
1780 	/* Check if outgoing interface has changed */
1781 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa &&
1782 	    rt->rt_ifa != NULL && rt->rt_ifa->ifa_rtrequest != NULL) {
1783 		rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1784 		ifa_free(rt->rt_ifa);
1785 	}
1786 	/* Update gateway address */
1787 	if (info->rti_info[RTAX_GATEWAY] != NULL) {
1788 		error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]);
1789 		if (error != 0)
1790 			goto bad;
1791 
1792 		rt->rt_flags &= ~RTF_GATEWAY;
1793 		rt->rt_flags |= (RTF_GATEWAY & info->rti_flags);
1794 	}
1795 
1796 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) {
1797 		ifa_ref(info->rti_ifa);
1798 		rt->rt_ifa = info->rti_ifa;
1799 		rt->rt_ifp = info->rti_ifp;
1800 	}
1801 	/* Allow some flags to be toggled on change. */
1802 	rt->rt_flags &= ~RTF_FMASK;
1803 	rt->rt_flags |= info->rti_flags & RTF_FMASK;
1804 
1805 	if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL)
1806 	       rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info);
1807 
1808 	/* Alter route MTU if necessary */
1809 	if (rt->rt_ifp != NULL) {
1810 		family = info->rti_info[RTAX_DST]->sa_family;
1811 		mtu = if_getmtu_family(rt->rt_ifp, family);
1812 		/* Set default MTU */
1813 		if (rt->rt_mtu == 0)
1814 			rt->rt_mtu = mtu;
1815 		if (rt->rt_mtu != mtu) {
1816 			/* Check if we really need to update */
1817 			ifmtu.ifp = rt->rt_ifp;
1818 			ifmtu.mtu = mtu;
1819 			if_updatemtu_cb(rt->rt_nodes, &ifmtu);
1820 		}
1821 	}
1822 
1823 	if (ret_nrt) {
1824 		*ret_nrt = rt;
1825 		RT_ADDREF(rt);
1826 	}
1827 bad:
1828 	RT_UNLOCK(rt);
1829 	if (free_ifa != 0)
1830 		ifa_free(info->rti_ifa);
1831 	return (error);
1832 }
1833 
1834 static void
1835 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
1836 {
1837 
1838 	if (info->rti_mflags & RTV_MTU) {
1839 		if (info->rti_rmx->rmx_mtu != 0) {
1840 
1841 			/*
1842 			 * MTU was explicitly provided by user.
1843 			 * Keep it.
1844 			 */
1845 			rt->rt_flags |= RTF_FIXEDMTU;
1846 		} else {
1847 
1848 			/*
1849 			 * User explicitly sets MTU to 0.
1850 			 * Assume rollback to default.
1851 			 */
1852 			rt->rt_flags &= ~RTF_FIXEDMTU;
1853 		}
1854 		rt->rt_mtu = info->rti_rmx->rmx_mtu;
1855 	}
1856 	if (info->rti_mflags & RTV_WEIGHT)
1857 		rt->rt_weight = info->rti_rmx->rmx_weight;
1858 	/* Kernel -> userland timebase conversion. */
1859 	if (info->rti_mflags & RTV_EXPIRE)
1860 		rt->rt_expire = info->rti_rmx->rmx_expire ?
1861 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
1862 }
1863 
1864 int
1865 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1866 {
1867 	/* XXX dst may be overwritten, can we move this to below */
1868 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1869 
1870 	/*
1871 	 * Prepare to store the gateway in rt->rt_gateway.
1872 	 * Both dst and gateway are stored one after the other in the same
1873 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1874 	 * rt_gateway already points to the right place.
1875 	 * Otherwise, malloc a new block and update the 'dst' address.
1876 	 */
1877 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1878 		caddr_t new;
1879 
1880 		R_Malloc(new, caddr_t, dlen + glen);
1881 		if (new == NULL)
1882 			return ENOBUFS;
1883 		/*
1884 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1885 		 * rt_setgate() can be called to initialize a newly
1886 		 * allocated route entry, in which case rt_key(rt) == NULL
1887 		 * (and also rt->rt_gateway == NULL).
1888 		 * Free()/free() handle a NULL argument just fine.
1889 		 */
1890 		bcopy(dst, new, dlen);
1891 		R_Free(rt_key(rt));	/* free old block, if any */
1892 		rt_key(rt) = (struct sockaddr *)new;
1893 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1894 	}
1895 
1896 	/*
1897 	 * Copy the new gateway value into the memory chunk.
1898 	 */
1899 	bcopy(gate, rt->rt_gateway, glen);
1900 
1901 	return (0);
1902 }
1903 
1904 void
1905 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1906 {
1907 	u_char *cp1 = (u_char *)src;
1908 	u_char *cp2 = (u_char *)dst;
1909 	u_char *cp3 = (u_char *)netmask;
1910 	u_char *cplim = cp2 + *cp3;
1911 	u_char *cplim2 = cp2 + *cp1;
1912 
1913 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1914 	cp3 += 2;
1915 	if (cplim > cplim2)
1916 		cplim = cplim2;
1917 	while (cp2 < cplim)
1918 		*cp2++ = *cp1++ & *cp3++;
1919 	if (cp2 < cplim2)
1920 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1921 }
1922 
1923 /*
1924  * Set up a routing table entry, normally
1925  * for an interface.
1926  */
1927 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1928 static inline  int
1929 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1930 {
1931 	struct sockaddr *dst;
1932 	struct sockaddr *netmask;
1933 	struct rtentry *rt = NULL;
1934 	struct rt_addrinfo info;
1935 	int error = 0;
1936 	int startfib, endfib;
1937 	char tempbuf[_SOCKADDR_TMPSIZE];
1938 	int didwork = 0;
1939 	int a_failure = 0;
1940 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1941 	struct rib_head *rnh;
1942 
1943 	if (flags & RTF_HOST) {
1944 		dst = ifa->ifa_dstaddr;
1945 		netmask = NULL;
1946 	} else {
1947 		dst = ifa->ifa_addr;
1948 		netmask = ifa->ifa_netmask;
1949 	}
1950 	if (dst->sa_len == 0)
1951 		return(EINVAL);
1952 	switch (dst->sa_family) {
1953 	case AF_INET6:
1954 	case AF_INET:
1955 		/* We support multiple FIBs. */
1956 		break;
1957 	default:
1958 		fibnum = RT_DEFAULT_FIB;
1959 		break;
1960 	}
1961 	if (fibnum == RT_ALL_FIBS) {
1962 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
1963 			startfib = endfib = ifa->ifa_ifp->if_fib;
1964 		else {
1965 			startfib = 0;
1966 			endfib = rt_numfibs - 1;
1967 		}
1968 	} else {
1969 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1970 		startfib = fibnum;
1971 		endfib = fibnum;
1972 	}
1973 
1974 	/*
1975 	 * If it's a delete, check that if it exists,
1976 	 * it's on the correct interface or we might scrub
1977 	 * a route to another ifa which would
1978 	 * be confusing at best and possibly worse.
1979 	 */
1980 	if (cmd == RTM_DELETE) {
1981 		/*
1982 		 * It's a delete, so it should already exist..
1983 		 * If it's a net, mask off the host bits
1984 		 * (Assuming we have a mask)
1985 		 * XXX this is kinda inet specific..
1986 		 */
1987 		if (netmask != NULL) {
1988 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1989 			dst = (struct sockaddr *)tempbuf;
1990 		}
1991 	}
1992 	/*
1993 	 * Now go through all the requested tables (fibs) and do the
1994 	 * requested action. Realistically, this will either be fib 0
1995 	 * for protocols that don't do multiple tables or all the
1996 	 * tables for those that do.
1997 	 */
1998 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1999 		if (cmd == RTM_DELETE) {
2000 			struct radix_node *rn;
2001 			/*
2002 			 * Look up an rtentry that is in the routing tree and
2003 			 * contains the correct info.
2004 			 */
2005 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
2006 			if (rnh == NULL)
2007 				/* this table doesn't exist but others might */
2008 				continue;
2009 			RIB_RLOCK(rnh);
2010 			rn = rnh->rnh_lookup(dst, netmask, &rnh->head);
2011 #ifdef RADIX_MPATH
2012 			if (rt_mpath_capable(rnh)) {
2013 
2014 				if (rn == NULL)
2015 					error = ESRCH;
2016 				else {
2017 					rt = RNTORT(rn);
2018 					/*
2019 					 * for interface route the
2020 					 * rt->rt_gateway is sockaddr_intf
2021 					 * for cloning ARP entries, so
2022 					 * rt_mpath_matchgate must use the
2023 					 * interface address
2024 					 */
2025 					rt = rt_mpath_matchgate(rt,
2026 					    ifa->ifa_addr);
2027 					if (rt == NULL)
2028 						error = ESRCH;
2029 				}
2030 			}
2031 #endif
2032 			error = (rn == NULL ||
2033 			    (rn->rn_flags & RNF_ROOT) ||
2034 			    RNTORT(rn)->rt_ifa != ifa);
2035 			RIB_RUNLOCK(rnh);
2036 			if (error) {
2037 				/* this is only an error if bad on ALL tables */
2038 				continue;
2039 			}
2040 		}
2041 		/*
2042 		 * Do the actual request
2043 		 */
2044 		bzero((caddr_t)&info, sizeof(info));
2045 		info.rti_ifa = ifa;
2046 		info.rti_flags = flags |
2047 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
2048 		info.rti_info[RTAX_DST] = dst;
2049 		/*
2050 		 * doing this for compatibility reasons
2051 		 */
2052 		if (cmd == RTM_ADD)
2053 			info.rti_info[RTAX_GATEWAY] =
2054 			    (struct sockaddr *)&null_sdl;
2055 		else
2056 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
2057 		info.rti_info[RTAX_NETMASK] = netmask;
2058 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
2059 
2060 		if (error == 0 && rt != NULL) {
2061 			/*
2062 			 * notify any listening routing agents of the change
2063 			 */
2064 			RT_LOCK(rt);
2065 #ifdef RADIX_MPATH
2066 			/*
2067 			 * in case address alias finds the first address
2068 			 * e.g. ifconfig bge0 192.0.2.246/24
2069 			 * e.g. ifconfig bge0 192.0.2.247/24
2070 			 * the address set in the route is 192.0.2.246
2071 			 * so we need to replace it with 192.0.2.247
2072 			 */
2073 			if (memcmp(rt->rt_ifa->ifa_addr,
2074 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
2075 				ifa_free(rt->rt_ifa);
2076 				ifa_ref(ifa);
2077 				rt->rt_ifp = ifa->ifa_ifp;
2078 				rt->rt_ifa = ifa;
2079 			}
2080 #endif
2081 			/*
2082 			 * doing this for compatibility reasons
2083 			 */
2084 			if (cmd == RTM_ADD) {
2085 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
2086 				rt->rt_ifp->if_type;
2087 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
2088 				rt->rt_ifp->if_index;
2089 			}
2090 			RT_ADDREF(rt);
2091 			RT_UNLOCK(rt);
2092 			rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
2093 			RT_LOCK(rt);
2094 			RT_REMREF(rt);
2095 			if (cmd == RTM_DELETE) {
2096 				/*
2097 				 * If we are deleting, and we found an entry,
2098 				 * then it's been removed from the tree..
2099 				 * now throw it away.
2100 				 */
2101 				RTFREE_LOCKED(rt);
2102 			} else {
2103 				if (cmd == RTM_ADD) {
2104 					/*
2105 					 * We just wanted to add it..
2106 					 * we don't actually need a reference.
2107 					 */
2108 					RT_REMREF(rt);
2109 				}
2110 				RT_UNLOCK(rt);
2111 			}
2112 			didwork = 1;
2113 		}
2114 		if (error)
2115 			a_failure = error;
2116 	}
2117 	if (cmd == RTM_DELETE) {
2118 		if (didwork) {
2119 			error = 0;
2120 		} else {
2121 			/* we only give an error if it wasn't in any table */
2122 			error = ((flags & RTF_HOST) ?
2123 			    EHOSTUNREACH : ENETUNREACH);
2124 		}
2125 	} else {
2126 		if (a_failure) {
2127 			/* return an error if any of them failed */
2128 			error = a_failure;
2129 		}
2130 	}
2131 	return (error);
2132 }
2133 
2134 /*
2135  * Set up a routing table entry, normally
2136  * for an interface.
2137  */
2138 int
2139 rtinit(struct ifaddr *ifa, int cmd, int flags)
2140 {
2141 	struct sockaddr *dst;
2142 	int fib = RT_DEFAULT_FIB;
2143 
2144 	if (flags & RTF_HOST) {
2145 		dst = ifa->ifa_dstaddr;
2146 	} else {
2147 		dst = ifa->ifa_addr;
2148 	}
2149 
2150 	switch (dst->sa_family) {
2151 	case AF_INET6:
2152 	case AF_INET:
2153 		/* We do support multiple FIBs. */
2154 		fib = RT_ALL_FIBS;
2155 		break;
2156 	}
2157 	return (rtinit1(ifa, cmd, flags, fib));
2158 }
2159 
2160 /*
2161  * Announce interface address arrival/withdraw
2162  * Returns 0 on success.
2163  */
2164 int
2165 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
2166 {
2167 
2168 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2169 	    ("unexpected cmd %d", cmd));
2170 
2171 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2172 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2173 
2174 #if defined(INET) || defined(INET6)
2175 #ifdef SCTP
2176 	/*
2177 	 * notify the SCTP stack
2178 	 * this will only get called when an address is added/deleted
2179 	 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
2180 	 */
2181 	sctp_addr_change(ifa, cmd);
2182 #endif /* SCTP */
2183 #endif
2184 	return (rtsock_addrmsg(cmd, ifa, fibnum));
2185 }
2186 
2187 /*
2188  * Announce route addition/removal.
2189  * Users of this function MUST validate input data BEFORE calling.
2190  * However we have to be able to handle invalid data:
2191  * if some userland app sends us "invalid" route message (invalid mask,
2192  * no dst, wrong address families, etc...) we need to pass it back
2193  * to app (and any other rtsock consumers) with rtm_errno field set to
2194  * non-zero value.
2195  * Returns 0 on success.
2196  */
2197 int
2198 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
2199     int fibnum)
2200 {
2201 
2202 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2203 	    ("unexpected cmd %d", cmd));
2204 
2205 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2206 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2207 
2208 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
2209 
2210 	return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
2211 }
2212 
2213 void
2214 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
2215 {
2216 
2217 	rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
2218 }
2219 
2220 /*
2221  * This is called to generate messages from the routing socket
2222  * indicating a network interface has had addresses associated with it.
2223  */
2224 void
2225 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
2226     int fibnum)
2227 {
2228 
2229 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2230 		("unexpected cmd %u", cmd));
2231 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2232 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2233 
2234 	if (cmd == RTM_ADD) {
2235 		rt_addrmsg(cmd, ifa, fibnum);
2236 		if (rt != NULL)
2237 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2238 	} else {
2239 		if (rt != NULL)
2240 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2241 		rt_addrmsg(cmd, ifa, fibnum);
2242 	}
2243 }
2244 
2245