xref: /freebsd/sys/net/route.c (revision 4b63a7c678a1291c8056de3770cfa393773b1f94)
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  * 4. 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/syslog.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/kernel.h>
56 
57 #include <net/if.h>
58 #include <net/if_var.h>
59 #include <net/if_dl.h>
60 #include <net/route.h>
61 #include <net/vnet.h>
62 #include <net/flowtable.h>
63 
64 #ifdef RADIX_MPATH
65 #include <net/radix_mpath.h>
66 #endif
67 
68 #include <netinet/in.h>
69 #include <netinet/ip_mroute.h>
70 
71 #include <vm/uma.h>
72 
73 #define	RT_MAXFIBS	UINT16_MAX
74 
75 /* Kernel config default option. */
76 #ifdef ROUTETABLES
77 #if ROUTETABLES <= 0
78 #error "ROUTETABLES defined too low"
79 #endif
80 #if ROUTETABLES > RT_MAXFIBS
81 #error "ROUTETABLES defined too big"
82 #endif
83 #define	RT_NUMFIBS	ROUTETABLES
84 #endif /* ROUTETABLES */
85 /* Initialize to default if not otherwise set. */
86 #ifndef	RT_NUMFIBS
87 #define	RT_NUMFIBS	1
88 #endif
89 
90 #if defined(INET) || defined(INET6)
91 #ifdef SCTP
92 extern void sctp_addr_change(struct ifaddr *ifa, int cmd);
93 #endif /* SCTP */
94 #endif
95 
96 
97 /* This is read-only.. */
98 u_int rt_numfibs = RT_NUMFIBS;
99 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RDTUN, &rt_numfibs, 0, "");
100 
101 /*
102  * By default add routes to all fibs for new interfaces.
103  * Once this is set to 0 then only allocate routes on interface
104  * changes for the FIB of the caller when adding a new set of addresses
105  * to an interface.  XXX this is a shotgun aproach to a problem that needs
106  * a more fine grained solution.. that will come.
107  * XXX also has the problems getting the FIB from curthread which will not
108  * always work given the fib can be overridden and prefixes can be added
109  * from the network stack context.
110  */
111 VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1;
112 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET,
113     &VNET_NAME(rt_add_addr_allfibs), 0, "");
114 
115 VNET_DEFINE(struct rtstat, rtstat);
116 #define	V_rtstat	VNET(rtstat)
117 
118 VNET_DEFINE(struct radix_node_head *, rt_tables);
119 #define	V_rt_tables	VNET(rt_tables)
120 
121 VNET_DEFINE(int, rttrash);		/* routes not in table but not freed */
122 #define	V_rttrash	VNET(rttrash)
123 
124 
125 /*
126  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
127  * The operation can be done safely (in this code) because a
128  * 'struct rtentry' starts with two 'struct radix_node''s, the first
129  * one representing leaf nodes in the routing tree, which is
130  * what the code in radix.c passes us as a 'struct radix_node'.
131  *
132  * But because there are a lot of assumptions in this conversion,
133  * do not cast explicitly, but always use the macro below.
134  */
135 #define RNTORT(p)	((struct rtentry *)(p))
136 
137 static VNET_DEFINE(uma_zone_t, rtzone);		/* Routing table UMA zone. */
138 #define	V_rtzone	VNET(rtzone)
139 
140 static int rtrequest1_fib_change(struct radix_node_head *, struct rt_addrinfo *,
141     struct rtentry **, u_int);
142 static void rt_setmetrics(const struct rt_addrinfo *, struct rtentry *);
143 
144 /*
145  * handler for net.my_fibnum
146  */
147 static int
148 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
149 {
150         int fibnum;
151         int error;
152 
153         fibnum = curthread->td_proc->p_fibnum;
154         error = sysctl_handle_int(oidp, &fibnum, 0, req);
155         return (error);
156 }
157 
158 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
159             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
160 
161 static __inline struct radix_node_head **
162 rt_tables_get_rnh_ptr(int table, int fam)
163 {
164 	struct radix_node_head **rnh;
165 
166 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
167 	    __func__));
168 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
169 	    __func__));
170 
171 	/* rnh is [fib=0][af=0]. */
172 	rnh = (struct radix_node_head **)V_rt_tables;
173 	/* Get the offset to the requested table and fam. */
174 	rnh += table * (AF_MAX+1) + fam;
175 
176 	return (rnh);
177 }
178 
179 struct radix_node_head *
180 rt_tables_get_rnh(int table, int fam)
181 {
182 
183 	return (*rt_tables_get_rnh_ptr(table, fam));
184 }
185 
186 /*
187  * route initialization must occur before ip6_init2(), which happenas at
188  * SI_ORDER_MIDDLE.
189  */
190 static void
191 route_init(void)
192 {
193 
194 	/* whack the tunable ints into  line. */
195 	if (rt_numfibs > RT_MAXFIBS)
196 		rt_numfibs = RT_MAXFIBS;
197 	if (rt_numfibs == 0)
198 		rt_numfibs = 1;
199 }
200 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
201 
202 static int
203 rtentry_zinit(void *mem, int size, int how)
204 {
205 	struct rtentry *rt = mem;
206 
207 	rt->rt_pksent = counter_u64_alloc(how);
208 	if (rt->rt_pksent == NULL)
209 		return (ENOMEM);
210 
211 	RT_LOCK_INIT(rt);
212 
213 	return (0);
214 }
215 
216 static void
217 rtentry_zfini(void *mem, int size)
218 {
219 	struct rtentry *rt = mem;
220 
221 	RT_LOCK_DESTROY(rt);
222 	counter_u64_free(rt->rt_pksent);
223 }
224 
225 static int
226 rtentry_ctor(void *mem, int size, void *arg, int how)
227 {
228 	struct rtentry *rt = mem;
229 
230 	bzero(rt, offsetof(struct rtentry, rt_endzero));
231 	counter_u64_zero(rt->rt_pksent);
232 
233 	return (0);
234 }
235 
236 static void
237 rtentry_dtor(void *mem, int size, void *arg)
238 {
239 	struct rtentry *rt = mem;
240 
241 	RT_UNLOCK_COND(rt);
242 }
243 
244 static void
245 vnet_route_init(const void *unused __unused)
246 {
247 	struct domain *dom;
248 	struct radix_node_head **rnh;
249 	int table;
250 	int fam;
251 
252 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
253 	    sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
254 
255 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
256 	    rtentry_ctor, rtentry_dtor,
257 	    rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
258 	for (dom = domains; dom; dom = dom->dom_next) {
259 		if (dom->dom_rtattach == NULL)
260 			continue;
261 
262 		for  (table = 0; table < rt_numfibs; table++) {
263 			fam = dom->dom_family;
264 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
265 				break;
266 
267 			rnh = rt_tables_get_rnh_ptr(table, fam);
268 			if (rnh == NULL)
269 				panic("%s: rnh NULL", __func__);
270 			dom->dom_rtattach((void **)rnh, 0);
271 		}
272 	}
273 }
274 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
275     vnet_route_init, 0);
276 
277 #ifdef VIMAGE
278 static void
279 vnet_route_uninit(const void *unused __unused)
280 {
281 	int table;
282 	int fam;
283 	struct domain *dom;
284 	struct radix_node_head **rnh;
285 
286 	for (dom = domains; dom; dom = dom->dom_next) {
287 		if (dom->dom_rtdetach == NULL)
288 			continue;
289 
290 		for (table = 0; table < rt_numfibs; table++) {
291 			fam = dom->dom_family;
292 
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_rtdetach((void **)rnh, 0);
300 		}
301 	}
302 
303 	free(V_rt_tables, M_RTABLE);
304 	uma_zdestroy(V_rtzone);
305 }
306 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
307     vnet_route_uninit, 0);
308 #endif
309 
310 #ifndef _SYS_SYSPROTO_H_
311 struct setfib_args {
312 	int     fibnum;
313 };
314 #endif
315 int
316 sys_setfib(struct thread *td, struct setfib_args *uap)
317 {
318 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
319 		return EINVAL;
320 	td->td_proc->p_fibnum = uap->fibnum;
321 	return (0);
322 }
323 
324 /*
325  * Packet routing routines.
326  */
327 void
328 rtalloc(struct route *ro)
329 {
330 
331 	rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
332 }
333 
334 void
335 rtalloc_fib(struct route *ro, u_int fibnum)
336 {
337 	rtalloc_ign_fib(ro, 0UL, fibnum);
338 }
339 
340 void
341 rtalloc_ign(struct route *ro, u_long ignore)
342 {
343 	struct rtentry *rt;
344 
345 	if ((rt = ro->ro_rt) != NULL) {
346 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
347 			return;
348 		RTFREE(rt);
349 		ro->ro_rt = NULL;
350 	}
351 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
352 	if (ro->ro_rt)
353 		RT_UNLOCK(ro->ro_rt);
354 }
355 
356 void
357 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
358 {
359 	struct rtentry *rt;
360 
361 	if ((rt = ro->ro_rt) != NULL) {
362 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
363 			return;
364 		RTFREE(rt);
365 		ro->ro_rt = NULL;
366 	}
367 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
368 	if (ro->ro_rt)
369 		RT_UNLOCK(ro->ro_rt);
370 }
371 
372 /*
373  * Look up the route that matches the address given
374  * Or, at least try.. Create a cloned route if needed.
375  *
376  * The returned route, if any, is locked.
377  */
378 struct rtentry *
379 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
380 {
381 
382 	return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
383 }
384 
385 struct rtentry *
386 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
387 		    u_int fibnum)
388 {
389 	struct radix_node_head *rnh;
390 	struct radix_node *rn;
391 	struct rtentry *newrt;
392 	struct rt_addrinfo info;
393 	int err = 0, msgtype = RTM_MISS;
394 	int needlock;
395 
396 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
397 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
398 	newrt = NULL;
399 	if (rnh == NULL)
400 		goto miss;
401 
402 	/*
403 	 * Look up the address in the table for that Address Family
404 	 */
405 	needlock = !(ignflags & RTF_RNH_LOCKED);
406 	if (needlock)
407 		RADIX_NODE_HEAD_RLOCK(rnh);
408 #ifdef INVARIANTS
409 	else
410 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
411 #endif
412 	rn = rnh->rnh_matchaddr(dst, rnh);
413 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
414 		newrt = RNTORT(rn);
415 		RT_LOCK(newrt);
416 		RT_ADDREF(newrt);
417 		if (needlock)
418 			RADIX_NODE_HEAD_RUNLOCK(rnh);
419 		goto done;
420 
421 	} else if (needlock)
422 		RADIX_NODE_HEAD_RUNLOCK(rnh);
423 
424 	/*
425 	 * Either we hit the root or couldn't find any match,
426 	 * Which basically means
427 	 * "caint get there frm here"
428 	 */
429 miss:
430 	V_rtstat.rts_unreach++;
431 
432 	if (report) {
433 		/*
434 		 * If required, report the failure to the supervising
435 		 * Authorities.
436 		 * For a delete, this is not an error. (report == 0)
437 		 */
438 		bzero(&info, sizeof(info));
439 		info.rti_info[RTAX_DST] = dst;
440 		rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
441 	}
442 done:
443 	if (newrt)
444 		RT_LOCK_ASSERT(newrt);
445 	return (newrt);
446 }
447 
448 /*
449  * Remove a reference count from an rtentry.
450  * If the count gets low enough, take it out of the routing table
451  */
452 void
453 rtfree(struct rtentry *rt)
454 {
455 	struct radix_node_head *rnh;
456 
457 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
458 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
459 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
460 
461 	RT_LOCK_ASSERT(rt);
462 
463 	/*
464 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
465 	 * we should come here exactly with the last reference.
466 	 */
467 	RT_REMREF(rt);
468 	if (rt->rt_refcnt > 0) {
469 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
470 		goto done;
471 	}
472 
473 	/*
474 	 * On last reference give the "close method" a chance
475 	 * to cleanup private state.  This also permits (for
476 	 * IPv4 and IPv6) a chance to decide if the routing table
477 	 * entry should be purged immediately or at a later time.
478 	 * When an immediate purge is to happen the close routine
479 	 * typically calls rtexpunge which clears the RTF_UP flag
480 	 * on the entry so that the code below reclaims the storage.
481 	 */
482 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
483 		rnh->rnh_close((struct radix_node *)rt, rnh);
484 
485 	/*
486 	 * If we are no longer "up" (and ref == 0)
487 	 * then we can free the resources associated
488 	 * with the route.
489 	 */
490 	if ((rt->rt_flags & RTF_UP) == 0) {
491 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
492 			panic("rtfree 2");
493 		/*
494 		 * the rtentry must have been removed from the routing table
495 		 * so it is represented in rttrash.. remove that now.
496 		 */
497 		V_rttrash--;
498 #ifdef	DIAGNOSTIC
499 		if (rt->rt_refcnt < 0) {
500 			printf("rtfree: %p not freed (neg refs)\n", rt);
501 			goto done;
502 		}
503 #endif
504 		/*
505 		 * release references on items we hold them on..
506 		 * e.g other routes and ifaddrs.
507 		 */
508 		if (rt->rt_ifa)
509 			ifa_free(rt->rt_ifa);
510 		/*
511 		 * The key is separatly alloc'd so free it (see rt_setgate()).
512 		 * This also frees the gateway, as they are always malloc'd
513 		 * together.
514 		 */
515 		Free(rt_key(rt));
516 
517 		/*
518 		 * and the rtentry itself of course
519 		 */
520 		uma_zfree(V_rtzone, rt);
521 		return;
522 	}
523 done:
524 	RT_UNLOCK(rt);
525 }
526 
527 
528 /*
529  * Force a routing table entry to the specified
530  * destination to go through the given gateway.
531  * Normally called as a result of a routing redirect
532  * message from the network layer.
533  */
534 void
535 rtredirect(struct sockaddr *dst,
536 	struct sockaddr *gateway,
537 	struct sockaddr *netmask,
538 	int flags,
539 	struct sockaddr *src)
540 {
541 
542 	rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
543 }
544 
545 void
546 rtredirect_fib(struct sockaddr *dst,
547 	struct sockaddr *gateway,
548 	struct sockaddr *netmask,
549 	int flags,
550 	struct sockaddr *src,
551 	u_int fibnum)
552 {
553 	struct rtentry *rt, *rt0 = NULL;
554 	int error = 0;
555 	short *stat = NULL;
556 	struct rt_addrinfo info;
557 	struct ifaddr *ifa;
558 	struct radix_node_head *rnh;
559 
560 	ifa = NULL;
561 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
562 	if (rnh == NULL) {
563 		error = EAFNOSUPPORT;
564 		goto out;
565 	}
566 
567 	/* verify the gateway is directly reachable */
568 	if ((ifa = ifa_ifwithnet(gateway, 0, fibnum)) == NULL) {
569 		error = ENETUNREACH;
570 		goto out;
571 	}
572 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
573 	/*
574 	 * If the redirect isn't from our current router for this dst,
575 	 * it's either old or wrong.  If it redirects us to ourselves,
576 	 * we have a routing loop, perhaps as a result of an interface
577 	 * going down recently.
578 	 */
579 	if (!(flags & RTF_DONE) && rt &&
580 	     (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
581 		error = EINVAL;
582 	else if (ifa_ifwithaddr_check(gateway))
583 		error = EHOSTUNREACH;
584 	if (error)
585 		goto done;
586 	/*
587 	 * Create a new entry if we just got back a wildcard entry
588 	 * or the lookup failed.  This is necessary for hosts
589 	 * which use routing redirects generated by smart gateways
590 	 * to dynamically build the routing tables.
591 	 */
592 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
593 		goto create;
594 	/*
595 	 * Don't listen to the redirect if it's
596 	 * for a route to an interface.
597 	 */
598 	if (rt->rt_flags & RTF_GATEWAY) {
599 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
600 			/*
601 			 * Changing from route to net => route to host.
602 			 * Create new route, rather than smashing route to net.
603 			 */
604 		create:
605 			rt0 = rt;
606 			rt = NULL;
607 
608 			flags |=  RTF_GATEWAY | RTF_DYNAMIC;
609 			bzero((caddr_t)&info, sizeof(info));
610 			info.rti_info[RTAX_DST] = dst;
611 			info.rti_info[RTAX_GATEWAY] = gateway;
612 			info.rti_info[RTAX_NETMASK] = netmask;
613 			info.rti_ifa = ifa;
614 			info.rti_flags = flags;
615 			if (rt0 != NULL)
616 				RT_UNLOCK(rt0);	/* drop lock to avoid LOR with RNH */
617 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
618 			if (rt != NULL) {
619 				RT_LOCK(rt);
620 				if (rt0 != NULL)
621 					EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
622 				flags = rt->rt_flags;
623 			}
624 			if (rt0 != NULL)
625 				RTFREE(rt0);
626 
627 			stat = &V_rtstat.rts_dynamic;
628 		} else {
629 			struct rtentry *gwrt;
630 
631 			/*
632 			 * Smash the current notion of the gateway to
633 			 * this destination.  Should check about netmask!!!
634 			 */
635 			rt->rt_flags |= RTF_MODIFIED;
636 			flags |= RTF_MODIFIED;
637 			stat = &V_rtstat.rts_newgateway;
638 			/*
639 			 * add the key and gateway (in one malloc'd chunk).
640 			 */
641 			RT_UNLOCK(rt);
642 			RADIX_NODE_HEAD_LOCK(rnh);
643 			RT_LOCK(rt);
644 			rt_setgate(rt, rt_key(rt), gateway);
645 			gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
646 			RADIX_NODE_HEAD_UNLOCK(rnh);
647 			EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
648 			RTFREE_LOCKED(gwrt);
649 		}
650 	} else
651 		error = EHOSTUNREACH;
652 done:
653 	if (rt)
654 		RTFREE_LOCKED(rt);
655 out:
656 	if (error)
657 		V_rtstat.rts_badredirect++;
658 	else if (stat != NULL)
659 		(*stat)++;
660 	bzero((caddr_t)&info, sizeof(info));
661 	info.rti_info[RTAX_DST] = dst;
662 	info.rti_info[RTAX_GATEWAY] = gateway;
663 	info.rti_info[RTAX_NETMASK] = netmask;
664 	info.rti_info[RTAX_AUTHOR] = src;
665 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
666 	if (ifa != NULL)
667 		ifa_free(ifa);
668 }
669 
670 int
671 rtioctl(u_long req, caddr_t data)
672 {
673 
674 	return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
675 }
676 
677 /*
678  * Routing table ioctl interface.
679  */
680 int
681 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
682 {
683 
684 	/*
685 	 * If more ioctl commands are added here, make sure the proper
686 	 * super-user checks are being performed because it is possible for
687 	 * prison-root to make it this far if raw sockets have been enabled
688 	 * in jails.
689 	 */
690 #ifdef INET
691 	/* Multicast goop, grrr... */
692 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
693 #else /* INET */
694 	return ENXIO;
695 #endif /* INET */
696 }
697 
698 struct ifaddr *
699 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway,
700 				u_int fibnum)
701 {
702 	struct ifaddr *ifa;
703 	int not_found = 0;
704 
705 	if ((flags & RTF_GATEWAY) == 0) {
706 		/*
707 		 * If we are adding a route to an interface,
708 		 * and the interface is a pt to pt link
709 		 * we should search for the destination
710 		 * as our clue to the interface.  Otherwise
711 		 * we can use the local address.
712 		 */
713 		ifa = NULL;
714 		if (flags & RTF_HOST)
715 			ifa = ifa_ifwithdstaddr(dst, fibnum);
716 		if (ifa == NULL)
717 			ifa = ifa_ifwithaddr(gateway);
718 	} else {
719 		/*
720 		 * If we are adding a route to a remote net
721 		 * or host, the gateway may still be on the
722 		 * other end of a pt to pt link.
723 		 */
724 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
725 	}
726 	if (ifa == NULL)
727 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
728 	if (ifa == NULL) {
729 		struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
730 		if (rt == NULL)
731 			return (NULL);
732 		/*
733 		 * dismiss a gateway that is reachable only
734 		 * through the default router
735 		 */
736 		switch (gateway->sa_family) {
737 		case AF_INET:
738 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
739 				not_found = 1;
740 			break;
741 		case AF_INET6:
742 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
743 				not_found = 1;
744 			break;
745 		default:
746 			break;
747 		}
748 		if (!not_found && rt->rt_ifa != NULL) {
749 			ifa = rt->rt_ifa;
750 			ifa_ref(ifa);
751 		}
752 		RT_REMREF(rt);
753 		RT_UNLOCK(rt);
754 		if (not_found || ifa == NULL)
755 			return (NULL);
756 	}
757 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
758 		struct ifaddr *oifa = ifa;
759 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
760 		if (ifa == NULL)
761 			ifa = oifa;
762 		else
763 			ifa_free(oifa);
764 	}
765 	return (ifa);
766 }
767 
768 /*
769  * Do appropriate manipulations of a routing tree given
770  * all the bits of info needed
771  */
772 int
773 rtrequest(int req,
774 	struct sockaddr *dst,
775 	struct sockaddr *gateway,
776 	struct sockaddr *netmask,
777 	int flags,
778 	struct rtentry **ret_nrt)
779 {
780 
781 	return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
782 	    RT_DEFAULT_FIB));
783 }
784 
785 int
786 rtrequest_fib(int req,
787 	struct sockaddr *dst,
788 	struct sockaddr *gateway,
789 	struct sockaddr *netmask,
790 	int flags,
791 	struct rtentry **ret_nrt,
792 	u_int fibnum)
793 {
794 	struct rt_addrinfo info;
795 
796 	if (dst->sa_len == 0)
797 		return(EINVAL);
798 
799 	bzero((caddr_t)&info, sizeof(info));
800 	info.rti_flags = flags;
801 	info.rti_info[RTAX_DST] = dst;
802 	info.rti_info[RTAX_GATEWAY] = gateway;
803 	info.rti_info[RTAX_NETMASK] = netmask;
804 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
805 }
806 
807 /*
808  * These (questionable) definitions of apparent local variables apply
809  * to the next two functions.  XXXXXX!!!
810  */
811 #define	dst	info->rti_info[RTAX_DST]
812 #define	gateway	info->rti_info[RTAX_GATEWAY]
813 #define	netmask	info->rti_info[RTAX_NETMASK]
814 #define	ifaaddr	info->rti_info[RTAX_IFA]
815 #define	ifpaddr	info->rti_info[RTAX_IFP]
816 #define	flags	info->rti_flags
817 
818 int
819 rt_getifa(struct rt_addrinfo *info)
820 {
821 
822 	return (rt_getifa_fib(info, RT_DEFAULT_FIB));
823 }
824 
825 /*
826  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
827  * it will be referenced so the caller must free it.
828  */
829 int
830 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
831 {
832 	struct ifaddr *ifa;
833 	int error = 0;
834 
835 	/*
836 	 * ifp may be specified by sockaddr_dl
837 	 * when protocol address is ambiguous.
838 	 */
839 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
840 	    ifpaddr->sa_family == AF_LINK &&
841 	    (ifa = ifa_ifwithnet(ifpaddr, 0, fibnum)) != NULL) {
842 		info->rti_ifp = ifa->ifa_ifp;
843 		ifa_free(ifa);
844 	}
845 	if (info->rti_ifa == NULL && ifaaddr != NULL)
846 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
847 	if (info->rti_ifa == NULL) {
848 		struct sockaddr *sa;
849 
850 		sa = ifaaddr != NULL ? ifaaddr :
851 		    (gateway != NULL ? gateway : dst);
852 		if (sa != NULL && info->rti_ifp != NULL)
853 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
854 		else if (dst != NULL && gateway != NULL)
855 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
856 							fibnum);
857 		else if (sa != NULL)
858 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
859 							fibnum);
860 	}
861 	if ((ifa = info->rti_ifa) != NULL) {
862 		if (info->rti_ifp == NULL)
863 			info->rti_ifp = ifa->ifa_ifp;
864 	} else
865 		error = ENETUNREACH;
866 	return (error);
867 }
868 
869 /*
870  * Expunges references to a route that's about to be reclaimed.
871  * The route must be locked.
872  */
873 int
874 rt_expunge(struct radix_node_head *rnh, struct rtentry *rt)
875 {
876 #if !defined(RADIX_MPATH)
877 	struct radix_node *rn;
878 #else
879 	struct rt_addrinfo info;
880 	int fib;
881 	struct rtentry *rt0;
882 #endif
883 	struct ifaddr *ifa;
884 	int error = 0;
885 
886 	RT_LOCK_ASSERT(rt);
887 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
888 
889 #ifdef RADIX_MPATH
890 	fib = rt->rt_fibnum;
891 	bzero(&info, sizeof(info));
892 	info.rti_ifp = rt->rt_ifp;
893 	info.rti_flags = RTF_RNH_LOCKED;
894 	info.rti_info[RTAX_DST] = rt_key(rt);
895 	info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
896 
897 	RT_UNLOCK(rt);
898 	error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
899 
900 	if (error == 0 && rt0 != NULL) {
901 		rt = rt0;
902 		RT_LOCK(rt);
903 	} else if (error != 0) {
904 		RT_LOCK(rt);
905 		return (error);
906 	}
907 #else
908 	/*
909 	 * Remove the item from the tree; it should be there,
910 	 * but when callers invoke us blindly it may not (sigh).
911 	 */
912 	rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
913 	if (rn == NULL) {
914 		error = ESRCH;
915 		goto bad;
916 	}
917 	KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
918 		("unexpected flags 0x%x", rn->rn_flags));
919 	KASSERT(rt == RNTORT(rn),
920 		("lookup mismatch, rt %p rn %p", rt, rn));
921 #endif /* RADIX_MPATH */
922 
923 	rt->rt_flags &= ~RTF_UP;
924 
925 	/*
926 	 * Give the protocol a chance to keep things in sync.
927 	 */
928 	if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
929 		struct rt_addrinfo info;
930 
931 		bzero((caddr_t)&info, sizeof(info));
932 		info.rti_flags = rt->rt_flags;
933 		info.rti_info[RTAX_DST] = rt_key(rt);
934 		info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
935 		info.rti_info[RTAX_NETMASK] = rt_mask(rt);
936 		ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
937 	}
938 
939 	/*
940 	 * one more rtentry floating around that is not
941 	 * linked to the routing table.
942 	 */
943 	V_rttrash++;
944 #if !defined(RADIX_MPATH)
945 bad:
946 #endif
947 	return (error);
948 }
949 
950 #if 0
951 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
952 int rt_print(char *buf, int buflen, struct rtentry *rt);
953 
954 int
955 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
956 {
957 	void *paddr = NULL;
958 
959 	switch (s->sa_family) {
960 	case AF_INET:
961 		paddr = &((struct sockaddr_in *)s)->sin_addr;
962 		break;
963 	case AF_INET6:
964 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
965 		break;
966 	}
967 
968 	if (paddr == NULL)
969 		return (0);
970 
971 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
972 		return (0);
973 
974 	return (strlen(buf));
975 }
976 
977 int
978 rt_print(char *buf, int buflen, struct rtentry *rt)
979 {
980 	struct sockaddr *addr, *mask;
981 	int i = 0;
982 
983 	addr = rt_key(rt);
984 	mask = rt_mask(rt);
985 
986 	i = p_sockaddr(buf, buflen, addr);
987 	if (!(rt->rt_flags & RTF_HOST)) {
988 		buf[i++] = '/';
989 		i += p_sockaddr(buf + i, buflen - i, mask);
990 	}
991 
992 	if (rt->rt_flags & RTF_GATEWAY) {
993 		buf[i++] = '>';
994 		i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
995 	}
996 
997 	return (i);
998 }
999 #endif
1000 
1001 #ifdef RADIX_MPATH
1002 static int
1003 rn_mpath_update(int req, struct rt_addrinfo *info,
1004     struct radix_node_head *rnh, struct rtentry **ret_nrt)
1005 {
1006 	/*
1007 	 * if we got multipath routes, we require users to specify
1008 	 * a matching RTAX_GATEWAY.
1009 	 */
1010 	struct rtentry *rt, *rto = NULL;
1011 	struct radix_node *rn;
1012 	int error = 0;
1013 
1014 	rn = rnh->rnh_lookup(dst, netmask, rnh);
1015 	if (rn == NULL)
1016 		return (ESRCH);
1017 	rto = rt = RNTORT(rn);
1018 
1019 	rt = rt_mpath_matchgate(rt, gateway);
1020 	if (rt == NULL)
1021 		return (ESRCH);
1022 	/*
1023 	 * this is the first entry in the chain
1024 	 */
1025 	if (rto == rt) {
1026 		rn = rn_mpath_next((struct radix_node *)rt);
1027 		/*
1028 		 * there is another entry, now it's active
1029 		 */
1030 		if (rn) {
1031 			rto = RNTORT(rn);
1032 			RT_LOCK(rto);
1033 			rto->rt_flags |= RTF_UP;
1034 			RT_UNLOCK(rto);
1035 		} else if (rt->rt_flags & RTF_GATEWAY) {
1036 			/*
1037 			 * For gateway routes, we need to
1038 			 * make sure that we we are deleting
1039 			 * the correct gateway.
1040 			 * rt_mpath_matchgate() does not
1041 			 * check the case when there is only
1042 			 * one route in the chain.
1043 			 */
1044 			if (gateway &&
1045 			    (rt->rt_gateway->sa_len != gateway->sa_len ||
1046 				memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
1047 				error = ESRCH;
1048 			else {
1049 				/*
1050 				 * remove from tree before returning it
1051 				 * to the caller
1052 				 */
1053 				rn = rnh->rnh_deladdr(dst, netmask, rnh);
1054 				KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
1055 				goto gwdelete;
1056 			}
1057 
1058 		}
1059 		/*
1060 		 * use the normal delete code to remove
1061 		 * the first entry
1062 		 */
1063 		if (req != RTM_DELETE)
1064 			goto nondelete;
1065 
1066 		error = ENOENT;
1067 		goto done;
1068 	}
1069 
1070 	/*
1071 	 * if the entry is 2nd and on up
1072 	 */
1073 	if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1074 		panic ("rtrequest1: rt_mpath_deldup");
1075 gwdelete:
1076 	RT_LOCK(rt);
1077 	RT_ADDREF(rt);
1078 	if (req == RTM_DELETE) {
1079 		rt->rt_flags &= ~RTF_UP;
1080 		/*
1081 		 * One more rtentry floating around that is not
1082 		 * linked to the routing table. rttrash will be decremented
1083 		 * when RTFREE(rt) is eventually called.
1084 		 */
1085 		V_rttrash++;
1086 	}
1087 
1088 nondelete:
1089 	if (req != RTM_DELETE)
1090 		panic("unrecognized request %d", req);
1091 
1092 
1093 	/*
1094 	 * If the caller wants it, then it can have it,
1095 	 * but it's up to it to free the rtentry as we won't be
1096 	 * doing it.
1097 	 */
1098 	if (ret_nrt) {
1099 		*ret_nrt = rt;
1100 		RT_UNLOCK(rt);
1101 	} else
1102 		RTFREE_LOCKED(rt);
1103 done:
1104 	return (error);
1105 }
1106 #endif
1107 
1108 int
1109 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1110 				u_int fibnum)
1111 {
1112 	int error = 0, needlock = 0;
1113 	struct rtentry *rt;
1114 #ifdef FLOWTABLE
1115 	struct rtentry *rt0;
1116 #endif
1117 	struct radix_node *rn;
1118 	struct radix_node_head *rnh;
1119 	struct ifaddr *ifa;
1120 	struct sockaddr *ndst;
1121 	struct sockaddr_storage mdst;
1122 #define senderr(x) { error = x ; goto bad; }
1123 
1124 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1125 	switch (dst->sa_family) {
1126 	case AF_INET6:
1127 	case AF_INET:
1128 		/* We support multiple FIBs. */
1129 		break;
1130 	default:
1131 		fibnum = RT_DEFAULT_FIB;
1132 		break;
1133 	}
1134 
1135 	/*
1136 	 * Find the correct routing tree to use for this Address Family
1137 	 */
1138 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1139 	if (rnh == NULL)
1140 		return (EAFNOSUPPORT);
1141 	needlock = ((flags & RTF_RNH_LOCKED) == 0);
1142 	flags &= ~RTF_RNH_LOCKED;
1143 	if (needlock)
1144 		RADIX_NODE_HEAD_LOCK(rnh);
1145 	else
1146 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1147 	/*
1148 	 * If we are adding a host route then we don't want to put
1149 	 * a netmask in the tree, nor do we want to clone it.
1150 	 */
1151 	if (flags & RTF_HOST)
1152 		netmask = NULL;
1153 
1154 	switch (req) {
1155 	case RTM_DELETE:
1156 		if (netmask) {
1157 			rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1158 			dst = (struct sockaddr *)&mdst;
1159 		}
1160 #ifdef RADIX_MPATH
1161 		if (rn_mpath_capable(rnh)) {
1162 			error = rn_mpath_update(req, info, rnh, ret_nrt);
1163 			/*
1164 			 * "bad" holds true for the success case
1165 			 * as well
1166 			 */
1167 			if (error != ENOENT)
1168 				goto bad;
1169 			error = 0;
1170 		}
1171 #endif
1172 		if ((flags & RTF_PINNED) == 0) {
1173 			/* Check if target route can be deleted */
1174 			rt = (struct rtentry *)rnh->rnh_lookup(dst,
1175 			    netmask, rnh);
1176 			if ((rt != NULL) && (rt->rt_flags & RTF_PINNED))
1177 				senderr(EADDRINUSE);
1178 		}
1179 
1180 		/*
1181 		 * Remove the item from the tree and return it.
1182 		 * Complain if it is not there and do no more processing.
1183 		 */
1184 		rn = rnh->rnh_deladdr(dst, netmask, rnh);
1185 		if (rn == NULL)
1186 			senderr(ESRCH);
1187 		if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1188 			panic ("rtrequest delete");
1189 		rt = RNTORT(rn);
1190 		RT_LOCK(rt);
1191 		RT_ADDREF(rt);
1192 		rt->rt_flags &= ~RTF_UP;
1193 
1194 		/*
1195 		 * give the protocol a chance to keep things in sync.
1196 		 */
1197 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1198 			ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1199 
1200 		/*
1201 		 * One more rtentry floating around that is not
1202 		 * linked to the routing table. rttrash will be decremented
1203 		 * when RTFREE(rt) is eventually called.
1204 		 */
1205 		V_rttrash++;
1206 
1207 		/*
1208 		 * If the caller wants it, then it can have it,
1209 		 * but it's up to it to free the rtentry as we won't be
1210 		 * doing it.
1211 		 */
1212 		if (ret_nrt) {
1213 			*ret_nrt = rt;
1214 			RT_UNLOCK(rt);
1215 		} else
1216 			RTFREE_LOCKED(rt);
1217 		break;
1218 	case RTM_RESOLVE:
1219 		/*
1220 		 * resolve was only used for route cloning
1221 		 * here for compat
1222 		 */
1223 		break;
1224 	case RTM_ADD:
1225 		if ((flags & RTF_GATEWAY) && !gateway)
1226 			senderr(EINVAL);
1227 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1228 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1229 			senderr(EINVAL);
1230 
1231 		if (info->rti_ifa == NULL) {
1232 			error = rt_getifa_fib(info, fibnum);
1233 			if (error)
1234 				senderr(error);
1235 		} else
1236 			ifa_ref(info->rti_ifa);
1237 		ifa = info->rti_ifa;
1238 		rt = uma_zalloc(V_rtzone, M_NOWAIT);
1239 		if (rt == NULL) {
1240 			ifa_free(ifa);
1241 			senderr(ENOBUFS);
1242 		}
1243 		rt->rt_flags = RTF_UP | flags;
1244 		rt->rt_fibnum = fibnum;
1245 		/*
1246 		 * Add the gateway. Possibly re-malloc-ing the storage for it.
1247 		 */
1248 		RT_LOCK(rt);
1249 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1250 			ifa_free(ifa);
1251 			uma_zfree(V_rtzone, rt);
1252 			senderr(error);
1253 		}
1254 
1255 		/*
1256 		 * point to the (possibly newly malloc'd) dest address.
1257 		 */
1258 		ndst = (struct sockaddr *)rt_key(rt);
1259 
1260 		/*
1261 		 * make sure it contains the value we want (masked if needed).
1262 		 */
1263 		if (netmask) {
1264 			rt_maskedcopy(dst, ndst, netmask);
1265 		} else
1266 			bcopy(dst, ndst, dst->sa_len);
1267 
1268 		/*
1269 		 * We use the ifa reference returned by rt_getifa_fib().
1270 		 * This moved from below so that rnh->rnh_addaddr() can
1271 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1272 		 */
1273 		rt->rt_ifa = ifa;
1274 		rt->rt_ifp = ifa->ifa_ifp;
1275 		rt->rt_weight = 1;
1276 
1277 		rt_setmetrics(info, rt);
1278 
1279 #ifdef RADIX_MPATH
1280 		/* do not permit exactly the same dst/mask/gw pair */
1281 		if (rn_mpath_capable(rnh) &&
1282 			rt_mpath_conflict(rnh, rt, netmask)) {
1283 			ifa_free(rt->rt_ifa);
1284 			Free(rt_key(rt));
1285 			uma_zfree(V_rtzone, rt);
1286 			senderr(EEXIST);
1287 		}
1288 #endif
1289 
1290 #ifdef FLOWTABLE
1291 		rt0 = NULL;
1292 		/* "flow-table" only supports IPv6 and IPv4 at the moment. */
1293 		switch (dst->sa_family) {
1294 #ifdef INET6
1295 		case AF_INET6:
1296 #endif
1297 #ifdef INET
1298 		case AF_INET:
1299 #endif
1300 #if defined(INET6) || defined(INET)
1301 			rn = rnh->rnh_matchaddr(dst, rnh);
1302 			if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1303 				struct sockaddr *mask;
1304 				u_char *m, *n;
1305 				int len;
1306 
1307 				/*
1308 				 * compare mask to see if the new route is
1309 				 * more specific than the existing one
1310 				 */
1311 				rt0 = RNTORT(rn);
1312 				RT_LOCK(rt0);
1313 				RT_ADDREF(rt0);
1314 				RT_UNLOCK(rt0);
1315 				/*
1316 				 * A host route is already present, so
1317 				 * leave the flow-table entries as is.
1318 				 */
1319 				if (rt0->rt_flags & RTF_HOST) {
1320 					RTFREE(rt0);
1321 					rt0 = NULL;
1322 				} else if (!(flags & RTF_HOST) && netmask) {
1323 					mask = rt_mask(rt0);
1324 					len = mask->sa_len;
1325 					m = (u_char *)mask;
1326 					n = (u_char *)netmask;
1327 					while (len-- > 0) {
1328 						if (*n != *m)
1329 							break;
1330 						n++;
1331 						m++;
1332 					}
1333 					if (len == 0 || (*n < *m)) {
1334 						RTFREE(rt0);
1335 						rt0 = NULL;
1336 					}
1337 				}
1338 			}
1339 #endif/* INET6 || INET */
1340 		}
1341 #endif /* FLOWTABLE */
1342 
1343 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1344 		rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1345 		/*
1346 		 * If it still failed to go into the tree,
1347 		 * then un-make it (this should be a function)
1348 		 */
1349 		if (rn == NULL) {
1350 			ifa_free(rt->rt_ifa);
1351 			Free(rt_key(rt));
1352 			uma_zfree(V_rtzone, rt);
1353 #ifdef FLOWTABLE
1354 			if (rt0 != NULL)
1355 				RTFREE(rt0);
1356 #endif
1357 			senderr(EEXIST);
1358 		}
1359 #ifdef FLOWTABLE
1360 		else if (rt0 != NULL) {
1361 			flowtable_route_flush(dst->sa_family, rt0);
1362 			RTFREE(rt0);
1363 		}
1364 #endif
1365 
1366 		/*
1367 		 * If this protocol has something to add to this then
1368 		 * allow it to do that as well.
1369 		 */
1370 		if (ifa->ifa_rtrequest)
1371 			ifa->ifa_rtrequest(req, rt, info);
1372 
1373 		/*
1374 		 * actually return a resultant rtentry and
1375 		 * give the caller a single reference.
1376 		 */
1377 		if (ret_nrt) {
1378 			*ret_nrt = rt;
1379 			RT_ADDREF(rt);
1380 		}
1381 		RT_UNLOCK(rt);
1382 		break;
1383 	case RTM_CHANGE:
1384 		error = rtrequest1_fib_change(rnh, info, ret_nrt, fibnum);
1385 		break;
1386 	default:
1387 		error = EOPNOTSUPP;
1388 	}
1389 bad:
1390 	if (needlock)
1391 		RADIX_NODE_HEAD_UNLOCK(rnh);
1392 	return (error);
1393 #undef senderr
1394 }
1395 
1396 #undef dst
1397 #undef gateway
1398 #undef netmask
1399 #undef ifaaddr
1400 #undef ifpaddr
1401 #undef flags
1402 
1403 static int
1404 rtrequest1_fib_change(struct radix_node_head *rnh, struct rt_addrinfo *info,
1405     struct rtentry **ret_nrt, u_int fibnum)
1406 {
1407 	struct rtentry *rt = NULL;
1408 	int error = 0;
1409 	int free_ifa = 0;
1410 	int family, mtu;
1411 
1412 	rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1413 	    info->rti_info[RTAX_NETMASK], rnh);
1414 
1415 	if (rt == NULL)
1416 		return (ESRCH);
1417 
1418 #ifdef RADIX_MPATH
1419 	/*
1420 	 * If we got multipath routes,
1421 	 * we require users to specify a matching RTAX_GATEWAY.
1422 	 */
1423 	if (rn_mpath_capable(rnh)) {
1424 		rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]);
1425 		if (rt == NULL)
1426 			return (ESRCH);
1427 	}
1428 #endif
1429 
1430 	RT_LOCK(rt);
1431 
1432 	rt_setmetrics(info, rt);
1433 
1434 	/*
1435 	 * New gateway could require new ifaddr, ifp;
1436 	 * flags may also be different; ifp may be specified
1437 	 * by ll sockaddr when protocol address is ambiguous
1438 	 */
1439 	if (((rt->rt_flags & RTF_GATEWAY) &&
1440 	    info->rti_info[RTAX_GATEWAY] != NULL) ||
1441 	    info->rti_info[RTAX_IFP] != NULL ||
1442 	    (info->rti_info[RTAX_IFA] != NULL &&
1443 	     !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) {
1444 
1445 		error = rt_getifa_fib(info, fibnum);
1446 		if (info->rti_ifa != NULL)
1447 			free_ifa = 1;
1448 
1449 		if (error != 0)
1450 			goto bad;
1451 	}
1452 
1453 	/* Check if outgoing interface has changed */
1454 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa &&
1455 	    rt->rt_ifa != NULL && rt->rt_ifa->ifa_rtrequest != NULL) {
1456 		rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1457 		ifa_free(rt->rt_ifa);
1458 	}
1459 	/* Update gateway address */
1460 	if (info->rti_info[RTAX_GATEWAY] != NULL) {
1461 		error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]);
1462 		if (error != 0)
1463 			goto bad;
1464 
1465 		rt->rt_flags &= ~RTF_GATEWAY;
1466 		rt->rt_flags |= (RTF_GATEWAY & info->rti_flags);
1467 	}
1468 
1469 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) {
1470 		ifa_ref(info->rti_ifa);
1471 		rt->rt_ifa = info->rti_ifa;
1472 		rt->rt_ifp = info->rti_ifp;
1473 	}
1474 	/* Allow some flags to be toggled on change. */
1475 	rt->rt_flags &= ~RTF_FMASK;
1476 	rt->rt_flags |= info->rti_flags & RTF_FMASK;
1477 
1478 	if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL)
1479 	       rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info);
1480 
1481 	/* Ensure route MTU is not bigger than interface MTU */
1482 	if (rt->rt_ifp != NULL) {
1483 		family = info->rti_info[RTAX_DST]->sa_family;
1484 		mtu = if_getmtu_family(rt->rt_ifp, family);
1485 		if (rt->rt_mtu > mtu)
1486 			rt->rt_mtu = mtu;
1487 	}
1488 
1489 	if (ret_nrt) {
1490 		*ret_nrt = rt;
1491 		RT_ADDREF(rt);
1492 	}
1493 bad:
1494 	RT_UNLOCK(rt);
1495 	if (free_ifa != 0)
1496 		ifa_free(info->rti_ifa);
1497 	return (error);
1498 }
1499 
1500 static void
1501 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
1502 {
1503 
1504 	if (info->rti_mflags & RTV_MTU)
1505 		rt->rt_mtu = info->rti_rmx->rmx_mtu;
1506 	if (info->rti_mflags & RTV_WEIGHT)
1507 		rt->rt_weight = info->rti_rmx->rmx_weight;
1508 	/* Kernel -> userland timebase conversion. */
1509 	if (info->rti_mflags & RTV_EXPIRE)
1510 		rt->rt_expire = info->rti_rmx->rmx_expire ?
1511 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
1512 }
1513 
1514 int
1515 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1516 {
1517 	/* XXX dst may be overwritten, can we move this to below */
1518 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1519 #ifdef INVARIANTS
1520 	struct radix_node_head *rnh;
1521 
1522 	rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1523 #endif
1524 
1525 	RT_LOCK_ASSERT(rt);
1526 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1527 
1528 	/*
1529 	 * Prepare to store the gateway in rt->rt_gateway.
1530 	 * Both dst and gateway are stored one after the other in the same
1531 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1532 	 * rt_gateway already points to the right place.
1533 	 * Otherwise, malloc a new block and update the 'dst' address.
1534 	 */
1535 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1536 		caddr_t new;
1537 
1538 		R_Malloc(new, caddr_t, dlen + glen);
1539 		if (new == NULL)
1540 			return ENOBUFS;
1541 		/*
1542 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1543 		 * rt_setgate() can be called to initialize a newly
1544 		 * allocated route entry, in which case rt_key(rt) == NULL
1545 		 * (and also rt->rt_gateway == NULL).
1546 		 * Free()/free() handle a NULL argument just fine.
1547 		 */
1548 		bcopy(dst, new, dlen);
1549 		Free(rt_key(rt));	/* free old block, if any */
1550 		rt_key(rt) = (struct sockaddr *)new;
1551 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1552 	}
1553 
1554 	/*
1555 	 * Copy the new gateway value into the memory chunk.
1556 	 */
1557 	bcopy(gate, rt->rt_gateway, glen);
1558 
1559 	return (0);
1560 }
1561 
1562 void
1563 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1564 {
1565 	u_char *cp1 = (u_char *)src;
1566 	u_char *cp2 = (u_char *)dst;
1567 	u_char *cp3 = (u_char *)netmask;
1568 	u_char *cplim = cp2 + *cp3;
1569 	u_char *cplim2 = cp2 + *cp1;
1570 
1571 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1572 	cp3 += 2;
1573 	if (cplim > cplim2)
1574 		cplim = cplim2;
1575 	while (cp2 < cplim)
1576 		*cp2++ = *cp1++ & *cp3++;
1577 	if (cp2 < cplim2)
1578 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1579 }
1580 
1581 /*
1582  * Set up a routing table entry, normally
1583  * for an interface.
1584  */
1585 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1586 static inline  int
1587 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1588 {
1589 	struct sockaddr *dst;
1590 	struct sockaddr *netmask;
1591 	struct rtentry *rt = NULL;
1592 	struct rt_addrinfo info;
1593 	int error = 0;
1594 	int startfib, endfib;
1595 	char tempbuf[_SOCKADDR_TMPSIZE];
1596 	int didwork = 0;
1597 	int a_failure = 0;
1598 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1599 	struct radix_node_head *rnh;
1600 
1601 	if (flags & RTF_HOST) {
1602 		dst = ifa->ifa_dstaddr;
1603 		netmask = NULL;
1604 	} else {
1605 		dst = ifa->ifa_addr;
1606 		netmask = ifa->ifa_netmask;
1607 	}
1608 	if (dst->sa_len == 0)
1609 		return(EINVAL);
1610 	switch (dst->sa_family) {
1611 	case AF_INET6:
1612 	case AF_INET:
1613 		/* We support multiple FIBs. */
1614 		break;
1615 	default:
1616 		fibnum = RT_DEFAULT_FIB;
1617 		break;
1618 	}
1619 	if (fibnum == RT_ALL_FIBS) {
1620 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
1621 			startfib = endfib = ifa->ifa_ifp->if_fib;
1622 		else {
1623 			startfib = 0;
1624 			endfib = rt_numfibs - 1;
1625 		}
1626 	} else {
1627 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1628 		startfib = fibnum;
1629 		endfib = fibnum;
1630 	}
1631 
1632 	/*
1633 	 * If it's a delete, check that if it exists,
1634 	 * it's on the correct interface or we might scrub
1635 	 * a route to another ifa which would
1636 	 * be confusing at best and possibly worse.
1637 	 */
1638 	if (cmd == RTM_DELETE) {
1639 		/*
1640 		 * It's a delete, so it should already exist..
1641 		 * If it's a net, mask off the host bits
1642 		 * (Assuming we have a mask)
1643 		 * XXX this is kinda inet specific..
1644 		 */
1645 		if (netmask != NULL) {
1646 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1647 			dst = (struct sockaddr *)tempbuf;
1648 		}
1649 	}
1650 	/*
1651 	 * Now go through all the requested tables (fibs) and do the
1652 	 * requested action. Realistically, this will either be fib 0
1653 	 * for protocols that don't do multiple tables or all the
1654 	 * tables for those that do.
1655 	 */
1656 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1657 		if (cmd == RTM_DELETE) {
1658 			struct radix_node *rn;
1659 			/*
1660 			 * Look up an rtentry that is in the routing tree and
1661 			 * contains the correct info.
1662 			 */
1663 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1664 			if (rnh == NULL)
1665 				/* this table doesn't exist but others might */
1666 				continue;
1667 			RADIX_NODE_HEAD_RLOCK(rnh);
1668 			rn = rnh->rnh_lookup(dst, netmask, rnh);
1669 #ifdef RADIX_MPATH
1670 			if (rn_mpath_capable(rnh)) {
1671 
1672 				if (rn == NULL)
1673 					error = ESRCH;
1674 				else {
1675 					rt = RNTORT(rn);
1676 					/*
1677 					 * for interface route the
1678 					 * rt->rt_gateway is sockaddr_intf
1679 					 * for cloning ARP entries, so
1680 					 * rt_mpath_matchgate must use the
1681 					 * interface address
1682 					 */
1683 					rt = rt_mpath_matchgate(rt,
1684 					    ifa->ifa_addr);
1685 					if (rt == NULL)
1686 						error = ESRCH;
1687 				}
1688 			}
1689 #endif
1690 			error = (rn == NULL ||
1691 			    (rn->rn_flags & RNF_ROOT) ||
1692 			    RNTORT(rn)->rt_ifa != ifa);
1693 			RADIX_NODE_HEAD_RUNLOCK(rnh);
1694 			if (error) {
1695 				/* this is only an error if bad on ALL tables */
1696 				continue;
1697 			}
1698 		}
1699 		/*
1700 		 * Do the actual request
1701 		 */
1702 		bzero((caddr_t)&info, sizeof(info));
1703 		info.rti_ifa = ifa;
1704 		info.rti_flags = flags |
1705 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1706 		info.rti_info[RTAX_DST] = dst;
1707 		/*
1708 		 * doing this for compatibility reasons
1709 		 */
1710 		if (cmd == RTM_ADD)
1711 			info.rti_info[RTAX_GATEWAY] =
1712 			    (struct sockaddr *)&null_sdl;
1713 		else
1714 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1715 		info.rti_info[RTAX_NETMASK] = netmask;
1716 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1717 
1718 		if ((error == EEXIST) && (cmd == RTM_ADD)) {
1719 			/*
1720 			 * Interface route addition failed.
1721 			 * Atomically delete current prefix generating
1722 			 * RTM_DELETE message, and retry adding
1723 			 * interface prefix.
1724 			 */
1725 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1726 			RADIX_NODE_HEAD_LOCK(rnh);
1727 
1728 			/* Delete old prefix */
1729 			info.rti_ifa = NULL;
1730 			info.rti_flags = RTF_RNH_LOCKED;
1731 
1732 			error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1733 			if (error == 0) {
1734 				info.rti_ifa = ifa;
1735 				info.rti_flags = flags | RTF_RNH_LOCKED |
1736 				    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1737 				error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1738 			}
1739 
1740 			RADIX_NODE_HEAD_UNLOCK(rnh);
1741 		}
1742 
1743 
1744 		if (error == 0 && rt != NULL) {
1745 			/*
1746 			 * notify any listening routing agents of the change
1747 			 */
1748 			RT_LOCK(rt);
1749 #ifdef RADIX_MPATH
1750 			/*
1751 			 * in case address alias finds the first address
1752 			 * e.g. ifconfig bge0 192.0.2.246/24
1753 			 * e.g. ifconfig bge0 192.0.2.247/24
1754 			 * the address set in the route is 192.0.2.246
1755 			 * so we need to replace it with 192.0.2.247
1756 			 */
1757 			if (memcmp(rt->rt_ifa->ifa_addr,
1758 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1759 				ifa_free(rt->rt_ifa);
1760 				ifa_ref(ifa);
1761 				rt->rt_ifp = ifa->ifa_ifp;
1762 				rt->rt_ifa = ifa;
1763 			}
1764 #endif
1765 			/*
1766 			 * doing this for compatibility reasons
1767 			 */
1768 			if (cmd == RTM_ADD) {
1769 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1770 				rt->rt_ifp->if_type;
1771 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1772 				rt->rt_ifp->if_index;
1773 			}
1774 			RT_ADDREF(rt);
1775 			RT_UNLOCK(rt);
1776 			rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1777 			RT_LOCK(rt);
1778 			RT_REMREF(rt);
1779 			if (cmd == RTM_DELETE) {
1780 				/*
1781 				 * If we are deleting, and we found an entry,
1782 				 * then it's been removed from the tree..
1783 				 * now throw it away.
1784 				 */
1785 				RTFREE_LOCKED(rt);
1786 			} else {
1787 				if (cmd == RTM_ADD) {
1788 					/*
1789 					 * We just wanted to add it..
1790 					 * we don't actually need a reference.
1791 					 */
1792 					RT_REMREF(rt);
1793 				}
1794 				RT_UNLOCK(rt);
1795 			}
1796 			didwork = 1;
1797 		}
1798 		if (error)
1799 			a_failure = error;
1800 	}
1801 	if (cmd == RTM_DELETE) {
1802 		if (didwork) {
1803 			error = 0;
1804 		} else {
1805 			/* we only give an error if it wasn't in any table */
1806 			error = ((flags & RTF_HOST) ?
1807 			    EHOSTUNREACH : ENETUNREACH);
1808 		}
1809 	} else {
1810 		if (a_failure) {
1811 			/* return an error if any of them failed */
1812 			error = a_failure;
1813 		}
1814 	}
1815 	return (error);
1816 }
1817 
1818 /*
1819  * Set up a routing table entry, normally
1820  * for an interface.
1821  */
1822 int
1823 rtinit(struct ifaddr *ifa, int cmd, int flags)
1824 {
1825 	struct sockaddr *dst;
1826 	int fib = RT_DEFAULT_FIB;
1827 
1828 	if (flags & RTF_HOST) {
1829 		dst = ifa->ifa_dstaddr;
1830 	} else {
1831 		dst = ifa->ifa_addr;
1832 	}
1833 
1834 	switch (dst->sa_family) {
1835 	case AF_INET6:
1836 	case AF_INET:
1837 		/* We do support multiple FIBs. */
1838 		fib = RT_ALL_FIBS;
1839 		break;
1840 	}
1841 	return (rtinit1(ifa, cmd, flags, fib));
1842 }
1843 
1844 /*
1845  * Announce interface address arrival/withdraw
1846  * Returns 0 on success.
1847  */
1848 int
1849 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
1850 {
1851 
1852 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1853 	    ("unexpected cmd %d", cmd));
1854 
1855 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1856 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1857 
1858 #if defined(INET) || defined(INET6)
1859 #ifdef SCTP
1860 	/*
1861 	 * notify the SCTP stack
1862 	 * this will only get called when an address is added/deleted
1863 	 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
1864 	 */
1865 	sctp_addr_change(ifa, cmd);
1866 #endif /* SCTP */
1867 #endif
1868 	return (rtsock_addrmsg(cmd, ifa, fibnum));
1869 }
1870 
1871 /*
1872  * Announce route addition/removal.
1873  * Users of this function MUST validate input data BEFORE calling.
1874  * However we have to be able to handle invalid data:
1875  * if some userland app sends us "invalid" route message (invalid mask,
1876  * no dst, wrong address families, etc...) we need to pass it back
1877  * to app (and any other rtsock consumers) with rtm_errno field set to
1878  * non-zero value.
1879  * Returns 0 on success.
1880  */
1881 int
1882 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
1883     int fibnum)
1884 {
1885 
1886 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1887 	    ("unexpected cmd %d", cmd));
1888 
1889 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1890 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1891 
1892 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
1893 
1894 	return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
1895 }
1896 
1897 void
1898 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
1899 {
1900 
1901 	rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
1902 }
1903 
1904 /*
1905  * This is called to generate messages from the routing socket
1906  * indicating a network interface has had addresses associated with it.
1907  */
1908 void
1909 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
1910     int fibnum)
1911 {
1912 
1913 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1914 		("unexpected cmd %u", cmd));
1915 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1916 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1917 
1918 	if (cmd == RTM_ADD) {
1919 		rt_addrmsg(cmd, ifa, fibnum);
1920 		if (rt != NULL)
1921 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1922 	} else {
1923 		if (rt != NULL)
1924 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1925 		rt_addrmsg(cmd, ifa, fibnum);
1926 	}
1927 }
1928 
1929