xref: /freebsd-src/sys/net/route.c (revision f0bd5302dd9e20355beadd0f260ffb926b6ac164)
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_RD, &rt_numfibs, 0, "");
100 /* and this can be set too big but will be fixed before it is used */
101 TUNABLE_INT("net.fibs", &rt_numfibs);
102 
103 /*
104  * By default add routes to all fibs for new interfaces.
105  * Once this is set to 0 then only allocate routes on interface
106  * changes for the FIB of the caller when adding a new set of addresses
107  * to an interface.  XXX this is a shotgun aproach to a problem that needs
108  * a more fine grained solution.. that will come.
109  * XXX also has the problems getting the FIB from curthread which will not
110  * always work given the fib can be overridden and prefixes can be added
111  * from the network stack context.
112  */
113 u_int rt_add_addr_allfibs = 1;
114 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RW,
115     &rt_add_addr_allfibs, 0, "");
116 TUNABLE_INT("net.add_addr_allfibs", &rt_add_addr_allfibs);
117 
118 VNET_DEFINE(struct rtstat, rtstat);
119 #define	V_rtstat	VNET(rtstat)
120 
121 VNET_DEFINE(struct radix_node_head *, rt_tables);
122 #define	V_rt_tables	VNET(rt_tables)
123 
124 VNET_DEFINE(int, rttrash);		/* routes not in table but not freed */
125 #define	V_rttrash	VNET(rttrash)
126 
127 
128 /* compare two sockaddr structures */
129 #define	sa_equal(a1, a2) (((a1)->sa_len == (a2)->sa_len) && \
130     (bcmp((a1), (a2), (a1)->sa_len) == 0))
131 
132 /*
133  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
134  * The operation can be done safely (in this code) because a
135  * 'struct rtentry' starts with two 'struct radix_node''s, the first
136  * one representing leaf nodes in the routing tree, which is
137  * what the code in radix.c passes us as a 'struct radix_node'.
138  *
139  * But because there are a lot of assumptions in this conversion,
140  * do not cast explicitly, but always use the macro below.
141  */
142 #define RNTORT(p)	((struct rtentry *)(p))
143 
144 static VNET_DEFINE(uma_zone_t, rtzone);		/* Routing table UMA zone. */
145 #define	V_rtzone	VNET(rtzone)
146 
147 /*
148  * handler for net.my_fibnum
149  */
150 static int
151 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
152 {
153         int fibnum;
154         int error;
155 
156         fibnum = curthread->td_proc->p_fibnum;
157         error = sysctl_handle_int(oidp, &fibnum, 0, req);
158         return (error);
159 }
160 
161 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
162             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
163 
164 static __inline struct radix_node_head **
165 rt_tables_get_rnh_ptr(int table, int fam)
166 {
167 	struct radix_node_head **rnh;
168 
169 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
170 	    __func__));
171 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
172 	    __func__));
173 
174 	/* rnh is [fib=0][af=0]. */
175 	rnh = (struct radix_node_head **)V_rt_tables;
176 	/* Get the offset to the requested table and fam. */
177 	rnh += table * (AF_MAX+1) + fam;
178 
179 	return (rnh);
180 }
181 
182 struct radix_node_head *
183 rt_tables_get_rnh(int table, int fam)
184 {
185 
186 	return (*rt_tables_get_rnh_ptr(table, fam));
187 }
188 
189 /*
190  * route initialization must occur before ip6_init2(), which happenas at
191  * SI_ORDER_MIDDLE.
192  */
193 static void
194 route_init(void)
195 {
196 
197 	/* whack the tunable ints into  line. */
198 	if (rt_numfibs > RT_MAXFIBS)
199 		rt_numfibs = RT_MAXFIBS;
200 	if (rt_numfibs == 0)
201 		rt_numfibs = 1;
202 }
203 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
204 
205 static int
206 rtentry_zinit(void *mem, int size, int how)
207 {
208 	struct rtentry *rt = mem;
209 
210 	rt->rt_pksent = counter_u64_alloc(how);
211 	if (rt->rt_pksent == NULL)
212 		return (ENOMEM);
213 
214 	RT_LOCK_INIT(rt);
215 
216 	return (0);
217 }
218 
219 static void
220 rtentry_zfini(void *mem, int size)
221 {
222 	struct rtentry *rt = mem;
223 
224 	RT_LOCK_DESTROY(rt);
225 	counter_u64_free(rt->rt_pksent);
226 }
227 
228 static int
229 rtentry_ctor(void *mem, int size, void *arg, int how)
230 {
231 	struct rtentry *rt = mem;
232 
233 	bzero(rt, offsetof(struct rtentry, rt_endzero));
234 	counter_u64_zero(rt->rt_pksent);
235 
236 	return (0);
237 }
238 
239 static void
240 vnet_route_init(const void *unused __unused)
241 {
242 	struct domain *dom;
243 	struct radix_node_head **rnh;
244 	int table;
245 	int fam;
246 
247 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
248 	    sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
249 
250 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
251 	    rtentry_ctor, NULL,
252 	    rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
253 	for (dom = domains; dom; dom = dom->dom_next) {
254 		if (dom->dom_rtattach == NULL)
255 			continue;
256 
257 		for  (table = 0; table < rt_numfibs; table++) {
258 			fam = dom->dom_family;
259 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
260 				break;
261 
262 			/*
263 			 * XXX MRT rtattach will be also called from
264 			 * vfs_export.c but the offset will be 0 (only for
265 			 * AF_INET and AF_INET6 which don't need it anyhow).
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, dom->dom_rtoffset);
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, dom->dom_rtoffset);
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 	switch (dst->sa_family) {
398 	case AF_INET6:
399 	case AF_INET:
400 		/* We support multiple FIBs. */
401 		break;
402 	default:
403 		fibnum = RT_DEFAULT_FIB;
404 		break;
405 	}
406 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
407 	newrt = NULL;
408 	if (rnh == NULL)
409 		goto miss;
410 
411 	/*
412 	 * Look up the address in the table for that Address Family
413 	 */
414 	needlock = !(ignflags & RTF_RNH_LOCKED);
415 	if (needlock)
416 		RADIX_NODE_HEAD_RLOCK(rnh);
417 #ifdef INVARIANTS
418 	else
419 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
420 #endif
421 	rn = rnh->rnh_matchaddr(dst, rnh);
422 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
423 		newrt = RNTORT(rn);
424 		RT_LOCK(newrt);
425 		RT_ADDREF(newrt);
426 		if (needlock)
427 			RADIX_NODE_HEAD_RUNLOCK(rnh);
428 		goto done;
429 
430 	} else if (needlock)
431 		RADIX_NODE_HEAD_RUNLOCK(rnh);
432 
433 	/*
434 	 * Either we hit the root or couldn't find any match,
435 	 * Which basically means
436 	 * "caint get there frm here"
437 	 */
438 miss:
439 	V_rtstat.rts_unreach++;
440 
441 	if (report) {
442 		/*
443 		 * If required, report the failure to the supervising
444 		 * Authorities.
445 		 * For a delete, this is not an error. (report == 0)
446 		 */
447 		bzero(&info, sizeof(info));
448 		info.rti_info[RTAX_DST] = dst;
449 		rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
450 	}
451 done:
452 	if (newrt)
453 		RT_LOCK_ASSERT(newrt);
454 	return (newrt);
455 }
456 
457 /*
458  * Remove a reference count from an rtentry.
459  * If the count gets low enough, take it out of the routing table
460  */
461 void
462 rtfree(struct rtentry *rt)
463 {
464 	struct radix_node_head *rnh;
465 
466 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
467 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
468 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
469 
470 	RT_LOCK_ASSERT(rt);
471 
472 	/*
473 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
474 	 * we should come here exactly with the last reference.
475 	 */
476 	RT_REMREF(rt);
477 	if (rt->rt_refcnt > 0) {
478 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
479 		goto done;
480 	}
481 
482 	/*
483 	 * On last reference give the "close method" a chance
484 	 * to cleanup private state.  This also permits (for
485 	 * IPv4 and IPv6) a chance to decide if the routing table
486 	 * entry should be purged immediately or at a later time.
487 	 * When an immediate purge is to happen the close routine
488 	 * typically calls rtexpunge which clears the RTF_UP flag
489 	 * on the entry so that the code below reclaims the storage.
490 	 */
491 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
492 		rnh->rnh_close((struct radix_node *)rt, rnh);
493 
494 	/*
495 	 * If we are no longer "up" (and ref == 0)
496 	 * then we can free the resources associated
497 	 * with the route.
498 	 */
499 	if ((rt->rt_flags & RTF_UP) == 0) {
500 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
501 			panic("rtfree 2");
502 		/*
503 		 * the rtentry must have been removed from the routing table
504 		 * so it is represented in rttrash.. remove that now.
505 		 */
506 		V_rttrash--;
507 #ifdef	DIAGNOSTIC
508 		if (rt->rt_refcnt < 0) {
509 			printf("rtfree: %p not freed (neg refs)\n", rt);
510 			goto done;
511 		}
512 #endif
513 		/*
514 		 * release references on items we hold them on..
515 		 * e.g other routes and ifaddrs.
516 		 */
517 		if (rt->rt_ifa)
518 			ifa_free(rt->rt_ifa);
519 		/*
520 		 * The key is separatly alloc'd so free it (see rt_setgate()).
521 		 * This also frees the gateway, as they are always malloc'd
522 		 * together.
523 		 */
524 		Free(rt_key(rt));
525 
526 		/*
527 		 * and the rtentry itself of course
528 		 */
529 		uma_zfree(V_rtzone, rt);
530 		return;
531 	}
532 done:
533 	RT_UNLOCK(rt);
534 }
535 
536 
537 /*
538  * Force a routing table entry to the specified
539  * destination to go through the given gateway.
540  * Normally called as a result of a routing redirect
541  * message from the network layer.
542  */
543 void
544 rtredirect(struct sockaddr *dst,
545 	struct sockaddr *gateway,
546 	struct sockaddr *netmask,
547 	int flags,
548 	struct sockaddr *src)
549 {
550 
551 	rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
552 }
553 
554 void
555 rtredirect_fib(struct sockaddr *dst,
556 	struct sockaddr *gateway,
557 	struct sockaddr *netmask,
558 	int flags,
559 	struct sockaddr *src,
560 	u_int fibnum)
561 {
562 	struct rtentry *rt, *rt0 = NULL;
563 	int error = 0;
564 	short *stat = NULL;
565 	struct rt_addrinfo info;
566 	struct ifaddr *ifa;
567 	struct radix_node_head *rnh;
568 
569 	ifa = NULL;
570 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
571 	if (rnh == NULL) {
572 		error = EAFNOSUPPORT;
573 		goto out;
574 	}
575 
576 	/* verify the gateway is directly reachable */
577 	if ((ifa = ifa_ifwithnet(gateway, 0)) == NULL) {
578 		error = ENETUNREACH;
579 		goto out;
580 	}
581 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
582 	/*
583 	 * If the redirect isn't from our current router for this dst,
584 	 * it's either old or wrong.  If it redirects us to ourselves,
585 	 * we have a routing loop, perhaps as a result of an interface
586 	 * going down recently.
587 	 */
588 	if (!(flags & RTF_DONE) && rt &&
589 	     (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
590 		error = EINVAL;
591 	else if (ifa_ifwithaddr_check(gateway))
592 		error = EHOSTUNREACH;
593 	if (error)
594 		goto done;
595 	/*
596 	 * Create a new entry if we just got back a wildcard entry
597 	 * or the lookup failed.  This is necessary for hosts
598 	 * which use routing redirects generated by smart gateways
599 	 * to dynamically build the routing tables.
600 	 */
601 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
602 		goto create;
603 	/*
604 	 * Don't listen to the redirect if it's
605 	 * for a route to an interface.
606 	 */
607 	if (rt->rt_flags & RTF_GATEWAY) {
608 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
609 			/*
610 			 * Changing from route to net => route to host.
611 			 * Create new route, rather than smashing route to net.
612 			 */
613 		create:
614 			rt0 = rt;
615 			rt = NULL;
616 
617 			flags |=  RTF_GATEWAY | RTF_DYNAMIC;
618 			bzero((caddr_t)&info, sizeof(info));
619 			info.rti_info[RTAX_DST] = dst;
620 			info.rti_info[RTAX_GATEWAY] = gateway;
621 			info.rti_info[RTAX_NETMASK] = netmask;
622 			info.rti_ifa = ifa;
623 			info.rti_flags = flags;
624 			if (rt0 != NULL)
625 				RT_UNLOCK(rt0);	/* drop lock to avoid LOR with RNH */
626 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
627 			if (rt != NULL) {
628 				RT_LOCK(rt);
629 				if (rt0 != NULL)
630 					EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
631 				flags = rt->rt_flags;
632 			}
633 			if (rt0 != NULL)
634 				RTFREE(rt0);
635 
636 			stat = &V_rtstat.rts_dynamic;
637 		} else {
638 			struct rtentry *gwrt;
639 
640 			/*
641 			 * Smash the current notion of the gateway to
642 			 * this destination.  Should check about netmask!!!
643 			 */
644 			rt->rt_flags |= RTF_MODIFIED;
645 			flags |= RTF_MODIFIED;
646 			stat = &V_rtstat.rts_newgateway;
647 			/*
648 			 * add the key and gateway (in one malloc'd chunk).
649 			 */
650 			RT_UNLOCK(rt);
651 			RADIX_NODE_HEAD_LOCK(rnh);
652 			RT_LOCK(rt);
653 			rt_setgate(rt, rt_key(rt), gateway);
654 			gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
655 			RADIX_NODE_HEAD_UNLOCK(rnh);
656 			EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
657 			RTFREE_LOCKED(gwrt);
658 		}
659 	} else
660 		error = EHOSTUNREACH;
661 done:
662 	if (rt)
663 		RTFREE_LOCKED(rt);
664 out:
665 	if (error)
666 		V_rtstat.rts_badredirect++;
667 	else if (stat != NULL)
668 		(*stat)++;
669 	bzero((caddr_t)&info, sizeof(info));
670 	info.rti_info[RTAX_DST] = dst;
671 	info.rti_info[RTAX_GATEWAY] = gateway;
672 	info.rti_info[RTAX_NETMASK] = netmask;
673 	info.rti_info[RTAX_AUTHOR] = src;
674 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
675 	if (ifa != NULL)
676 		ifa_free(ifa);
677 }
678 
679 int
680 rtioctl(u_long req, caddr_t data)
681 {
682 
683 	return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
684 }
685 
686 /*
687  * Routing table ioctl interface.
688  */
689 int
690 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
691 {
692 
693 	/*
694 	 * If more ioctl commands are added here, make sure the proper
695 	 * super-user checks are being performed because it is possible for
696 	 * prison-root to make it this far if raw sockets have been enabled
697 	 * in jails.
698 	 */
699 #ifdef INET
700 	/* Multicast goop, grrr... */
701 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
702 #else /* INET */
703 	return ENXIO;
704 #endif /* INET */
705 }
706 
707 /*
708  * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
709  */
710 struct ifaddr *
711 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
712 {
713 
714 	return (ifa_ifwithroute_fib(flags, dst, gateway, RT_DEFAULT_FIB));
715 }
716 
717 struct ifaddr *
718 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
719 				u_int fibnum)
720 {
721 	register struct ifaddr *ifa;
722 	int not_found = 0;
723 
724 	if ((flags & RTF_GATEWAY) == 0) {
725 		/*
726 		 * If we are adding a route to an interface,
727 		 * and the interface is a pt to pt link
728 		 * we should search for the destination
729 		 * as our clue to the interface.  Otherwise
730 		 * we can use the local address.
731 		 */
732 		ifa = NULL;
733 		if (flags & RTF_HOST)
734 			ifa = ifa_ifwithdstaddr(dst);
735 		if (ifa == NULL)
736 			ifa = ifa_ifwithaddr(gateway);
737 	} else {
738 		/*
739 		 * If we are adding a route to a remote net
740 		 * or host, the gateway may still be on the
741 		 * other end of a pt to pt link.
742 		 */
743 		ifa = ifa_ifwithdstaddr(gateway);
744 	}
745 	if (ifa == NULL)
746 		ifa = ifa_ifwithnet(gateway, 0);
747 	if (ifa == NULL) {
748 		struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
749 		if (rt == NULL)
750 			return (NULL);
751 		/*
752 		 * dismiss a gateway that is reachable only
753 		 * through the default router
754 		 */
755 		switch (gateway->sa_family) {
756 		case AF_INET:
757 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
758 				not_found = 1;
759 			break;
760 		case AF_INET6:
761 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
762 				not_found = 1;
763 			break;
764 		default:
765 			break;
766 		}
767 		if (!not_found && rt->rt_ifa != NULL) {
768 			ifa = rt->rt_ifa;
769 			ifa_ref(ifa);
770 		}
771 		RT_REMREF(rt);
772 		RT_UNLOCK(rt);
773 		if (not_found || ifa == NULL)
774 			return (NULL);
775 	}
776 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
777 		struct ifaddr *oifa = ifa;
778 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
779 		if (ifa == NULL)
780 			ifa = oifa;
781 		else
782 			ifa_free(oifa);
783 	}
784 	return (ifa);
785 }
786 
787 /*
788  * Do appropriate manipulations of a routing tree given
789  * all the bits of info needed
790  */
791 int
792 rtrequest(int req,
793 	struct sockaddr *dst,
794 	struct sockaddr *gateway,
795 	struct sockaddr *netmask,
796 	int flags,
797 	struct rtentry **ret_nrt)
798 {
799 
800 	return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
801 	    RT_DEFAULT_FIB));
802 }
803 
804 int
805 rtrequest_fib(int req,
806 	struct sockaddr *dst,
807 	struct sockaddr *gateway,
808 	struct sockaddr *netmask,
809 	int flags,
810 	struct rtentry **ret_nrt,
811 	u_int fibnum)
812 {
813 	struct rt_addrinfo info;
814 
815 	if (dst->sa_len == 0)
816 		return(EINVAL);
817 
818 	bzero((caddr_t)&info, sizeof(info));
819 	info.rti_flags = flags;
820 	info.rti_info[RTAX_DST] = dst;
821 	info.rti_info[RTAX_GATEWAY] = gateway;
822 	info.rti_info[RTAX_NETMASK] = netmask;
823 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
824 }
825 
826 /*
827  * These (questionable) definitions of apparent local variables apply
828  * to the next two functions.  XXXXXX!!!
829  */
830 #define	dst	info->rti_info[RTAX_DST]
831 #define	gateway	info->rti_info[RTAX_GATEWAY]
832 #define	netmask	info->rti_info[RTAX_NETMASK]
833 #define	ifaaddr	info->rti_info[RTAX_IFA]
834 #define	ifpaddr	info->rti_info[RTAX_IFP]
835 #define	flags	info->rti_flags
836 
837 int
838 rt_getifa(struct rt_addrinfo *info)
839 {
840 
841 	return (rt_getifa_fib(info, RT_DEFAULT_FIB));
842 }
843 
844 /*
845  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
846  * it will be referenced so the caller must free it.
847  */
848 int
849 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
850 {
851 	struct ifaddr *ifa;
852 	int error = 0;
853 
854 	/*
855 	 * ifp may be specified by sockaddr_dl
856 	 * when protocol address is ambiguous.
857 	 */
858 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
859 	    ifpaddr->sa_family == AF_LINK &&
860 	    (ifa = ifa_ifwithnet(ifpaddr, 0)) != NULL) {
861 		info->rti_ifp = ifa->ifa_ifp;
862 		ifa_free(ifa);
863 	}
864 	if (info->rti_ifa == NULL && ifaaddr != NULL)
865 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
866 	if (info->rti_ifa == NULL) {
867 		struct sockaddr *sa;
868 
869 		sa = ifaaddr != NULL ? ifaaddr :
870 		    (gateway != NULL ? gateway : dst);
871 		if (sa != NULL && info->rti_ifp != NULL)
872 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
873 		else if (dst != NULL && gateway != NULL)
874 			info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
875 							fibnum);
876 		else if (sa != NULL)
877 			info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
878 							fibnum);
879 	}
880 	if ((ifa = info->rti_ifa) != NULL) {
881 		if (info->rti_ifp == NULL)
882 			info->rti_ifp = ifa->ifa_ifp;
883 	} else
884 		error = ENETUNREACH;
885 	return (error);
886 }
887 
888 /*
889  * Expunges references to a route that's about to be reclaimed.
890  * The route must be locked.
891  */
892 int
893 rtexpunge(struct rtentry *rt)
894 {
895 #if !defined(RADIX_MPATH)
896 	struct radix_node *rn;
897 #else
898 	struct rt_addrinfo info;
899 	int fib;
900 	struct rtentry *rt0;
901 #endif
902 	struct radix_node_head *rnh;
903 	struct ifaddr *ifa;
904 	int error = 0;
905 
906 	/*
907 	 * Find the correct routing tree to use for this Address Family
908 	 */
909 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
910 	RT_LOCK_ASSERT(rt);
911 	if (rnh == NULL)
912 		return (EAFNOSUPPORT);
913 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
914 
915 #ifdef RADIX_MPATH
916 	fib = rt->rt_fibnum;
917 	bzero(&info, sizeof(info));
918 	info.rti_ifp = rt->rt_ifp;
919 	info.rti_flags = RTF_RNH_LOCKED;
920 	info.rti_info[RTAX_DST] = rt_key(rt);
921 	info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
922 
923 	RT_UNLOCK(rt);
924 	error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
925 
926 	if (error == 0 && rt0 != NULL) {
927 		rt = rt0;
928 		RT_LOCK(rt);
929 	} else if (error != 0) {
930 		RT_LOCK(rt);
931 		return (error);
932 	}
933 #else
934 	/*
935 	 * Remove the item from the tree; it should be there,
936 	 * but when callers invoke us blindly it may not (sigh).
937 	 */
938 	rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
939 	if (rn == NULL) {
940 		error = ESRCH;
941 		goto bad;
942 	}
943 	KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
944 		("unexpected flags 0x%x", rn->rn_flags));
945 	KASSERT(rt == RNTORT(rn),
946 		("lookup mismatch, rt %p rn %p", rt, rn));
947 #endif /* RADIX_MPATH */
948 
949 	rt->rt_flags &= ~RTF_UP;
950 
951 	/*
952 	 * Give the protocol a chance to keep things in sync.
953 	 */
954 	if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
955 		struct rt_addrinfo info;
956 
957 		bzero((caddr_t)&info, sizeof(info));
958 		info.rti_flags = rt->rt_flags;
959 		info.rti_info[RTAX_DST] = rt_key(rt);
960 		info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
961 		info.rti_info[RTAX_NETMASK] = rt_mask(rt);
962 		ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
963 	}
964 
965 	/*
966 	 * one more rtentry floating around that is not
967 	 * linked to the routing table.
968 	 */
969 	V_rttrash++;
970 #if !defined(RADIX_MPATH)
971 bad:
972 #endif
973 	return (error);
974 }
975 
976 #if 0
977 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
978 int rt_print(char *buf, int buflen, struct rtentry *rt);
979 
980 int
981 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
982 {
983 	void *paddr = NULL;
984 
985 	switch (s->sa_family) {
986 	case AF_INET:
987 		paddr = &((struct sockaddr_in *)s)->sin_addr;
988 		break;
989 	case AF_INET6:
990 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
991 		break;
992 	}
993 
994 	if (paddr == NULL)
995 		return (0);
996 
997 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
998 		return (0);
999 
1000 	return (strlen(buf));
1001 }
1002 
1003 int
1004 rt_print(char *buf, int buflen, struct rtentry *rt)
1005 {
1006 	struct sockaddr *addr, *mask;
1007 	int i = 0;
1008 
1009 	addr = rt_key(rt);
1010 	mask = rt_mask(rt);
1011 
1012 	i = p_sockaddr(buf, buflen, addr);
1013 	if (!(rt->rt_flags & RTF_HOST)) {
1014 		buf[i++] = '/';
1015 		i += p_sockaddr(buf + i, buflen - i, mask);
1016 	}
1017 
1018 	if (rt->rt_flags & RTF_GATEWAY) {
1019 		buf[i++] = '>';
1020 		i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1021 	}
1022 
1023 	return (i);
1024 }
1025 #endif
1026 
1027 #ifdef RADIX_MPATH
1028 static int
1029 rn_mpath_update(int req, struct rt_addrinfo *info,
1030     struct radix_node_head *rnh, struct rtentry **ret_nrt)
1031 {
1032 	/*
1033 	 * if we got multipath routes, we require users to specify
1034 	 * a matching RTAX_GATEWAY.
1035 	 */
1036 	struct rtentry *rt, *rto = NULL;
1037 	register struct radix_node *rn;
1038 	int error = 0;
1039 
1040 	rn = rnh->rnh_lookup(dst, netmask, rnh);
1041 	if (rn == NULL)
1042 		return (ESRCH);
1043 	rto = rt = RNTORT(rn);
1044 
1045 	rt = rt_mpath_matchgate(rt, gateway);
1046 	if (rt == NULL)
1047 		return (ESRCH);
1048 	/*
1049 	 * this is the first entry in the chain
1050 	 */
1051 	if (rto == rt) {
1052 		rn = rn_mpath_next((struct radix_node *)rt);
1053 		/*
1054 		 * there is another entry, now it's active
1055 		 */
1056 		if (rn) {
1057 			rto = RNTORT(rn);
1058 			RT_LOCK(rto);
1059 			rto->rt_flags |= RTF_UP;
1060 			RT_UNLOCK(rto);
1061 		} else if (rt->rt_flags & RTF_GATEWAY) {
1062 			/*
1063 			 * For gateway routes, we need to
1064 			 * make sure that we we are deleting
1065 			 * the correct gateway.
1066 			 * rt_mpath_matchgate() does not
1067 			 * check the case when there is only
1068 			 * one route in the chain.
1069 			 */
1070 			if (gateway &&
1071 			    (rt->rt_gateway->sa_len != gateway->sa_len ||
1072 				memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
1073 				error = ESRCH;
1074 			else {
1075 				/*
1076 				 * remove from tree before returning it
1077 				 * to the caller
1078 				 */
1079 				rn = rnh->rnh_deladdr(dst, netmask, rnh);
1080 				KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
1081 				goto gwdelete;
1082 			}
1083 
1084 		}
1085 		/*
1086 		 * use the normal delete code to remove
1087 		 * the first entry
1088 		 */
1089 		if (req != RTM_DELETE)
1090 			goto nondelete;
1091 
1092 		error = ENOENT;
1093 		goto done;
1094 	}
1095 
1096 	/*
1097 	 * if the entry is 2nd and on up
1098 	 */
1099 	if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1100 		panic ("rtrequest1: rt_mpath_deldup");
1101 gwdelete:
1102 	RT_LOCK(rt);
1103 	RT_ADDREF(rt);
1104 	if (req == RTM_DELETE) {
1105 		rt->rt_flags &= ~RTF_UP;
1106 		/*
1107 		 * One more rtentry floating around that is not
1108 		 * linked to the routing table. rttrash will be decremented
1109 		 * when RTFREE(rt) is eventually called.
1110 		 */
1111 		V_rttrash++;
1112 	}
1113 
1114 nondelete:
1115 	if (req != RTM_DELETE)
1116 		panic("unrecognized request %d", req);
1117 
1118 
1119 	/*
1120 	 * If the caller wants it, then it can have it,
1121 	 * but it's up to it to free the rtentry as we won't be
1122 	 * doing it.
1123 	 */
1124 	if (ret_nrt) {
1125 		*ret_nrt = rt;
1126 		RT_UNLOCK(rt);
1127 	} else
1128 		RTFREE_LOCKED(rt);
1129 done:
1130 	return (error);
1131 }
1132 #endif
1133 
1134 int
1135 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1136 				u_int fibnum)
1137 {
1138 	int error = 0, needlock = 0;
1139 	register struct rtentry *rt;
1140 #ifdef FLOWTABLE
1141 	register struct rtentry *rt0;
1142 #endif
1143 	register struct radix_node *rn;
1144 	register struct radix_node_head *rnh;
1145 	struct ifaddr *ifa;
1146 	struct sockaddr *ndst;
1147 	struct sockaddr_storage mdst;
1148 #define senderr(x) { error = x ; goto bad; }
1149 
1150 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1151 	switch (dst->sa_family) {
1152 	case AF_INET6:
1153 	case AF_INET:
1154 		/* We support multiple FIBs. */
1155 		break;
1156 	default:
1157 		fibnum = RT_DEFAULT_FIB;
1158 		break;
1159 	}
1160 
1161 	/*
1162 	 * Find the correct routing tree to use for this Address Family
1163 	 */
1164 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1165 	if (rnh == NULL)
1166 		return (EAFNOSUPPORT);
1167 	needlock = ((flags & RTF_RNH_LOCKED) == 0);
1168 	flags &= ~RTF_RNH_LOCKED;
1169 	if (needlock)
1170 		RADIX_NODE_HEAD_LOCK(rnh);
1171 	else
1172 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1173 	/*
1174 	 * If we are adding a host route then we don't want to put
1175 	 * a netmask in the tree, nor do we want to clone it.
1176 	 */
1177 	if (flags & RTF_HOST)
1178 		netmask = NULL;
1179 
1180 	switch (req) {
1181 	case RTM_DELETE:
1182 		if (netmask) {
1183 			rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1184 			dst = (struct sockaddr *)&mdst;
1185 		}
1186 #ifdef RADIX_MPATH
1187 		if (rn_mpath_capable(rnh)) {
1188 			error = rn_mpath_update(req, info, rnh, ret_nrt);
1189 			/*
1190 			 * "bad" holds true for the success case
1191 			 * as well
1192 			 */
1193 			if (error != ENOENT)
1194 				goto bad;
1195 			error = 0;
1196 		}
1197 #endif
1198 		if ((flags & RTF_PINNED) == 0) {
1199 			/* Check if target route can be deleted */
1200 			rt = (struct rtentry *)rnh->rnh_lookup(dst,
1201 			    netmask, rnh);
1202 			if ((rt != NULL) && (rt->rt_flags & RTF_PINNED))
1203 				senderr(EADDRINUSE);
1204 		}
1205 
1206 		/*
1207 		 * Remove the item from the tree and return it.
1208 		 * Complain if it is not there and do no more processing.
1209 		 */
1210 		rn = rnh->rnh_deladdr(dst, netmask, rnh);
1211 		if (rn == NULL)
1212 			senderr(ESRCH);
1213 		if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1214 			panic ("rtrequest delete");
1215 		rt = RNTORT(rn);
1216 		RT_LOCK(rt);
1217 		RT_ADDREF(rt);
1218 		rt->rt_flags &= ~RTF_UP;
1219 
1220 		/*
1221 		 * give the protocol a chance to keep things in sync.
1222 		 */
1223 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1224 			ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1225 
1226 		/*
1227 		 * One more rtentry floating around that is not
1228 		 * linked to the routing table. rttrash will be decremented
1229 		 * when RTFREE(rt) is eventually called.
1230 		 */
1231 		V_rttrash++;
1232 
1233 		/*
1234 		 * If the caller wants it, then it can have it,
1235 		 * but it's up to it to free the rtentry as we won't be
1236 		 * doing it.
1237 		 */
1238 		if (ret_nrt) {
1239 			*ret_nrt = rt;
1240 			RT_UNLOCK(rt);
1241 		} else
1242 			RTFREE_LOCKED(rt);
1243 		break;
1244 	case RTM_RESOLVE:
1245 		/*
1246 		 * resolve was only used for route cloning
1247 		 * here for compat
1248 		 */
1249 		break;
1250 	case RTM_ADD:
1251 		if ((flags & RTF_GATEWAY) && !gateway)
1252 			senderr(EINVAL);
1253 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1254 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1255 			senderr(EINVAL);
1256 
1257 		if (info->rti_ifa == NULL) {
1258 			error = rt_getifa_fib(info, fibnum);
1259 			if (error)
1260 				senderr(error);
1261 		} else
1262 			ifa_ref(info->rti_ifa);
1263 		ifa = info->rti_ifa;
1264 		rt = uma_zalloc(V_rtzone, M_NOWAIT);
1265 		if (rt == NULL) {
1266 			ifa_free(ifa);
1267 			senderr(ENOBUFS);
1268 		}
1269 		rt->rt_flags = RTF_UP | flags;
1270 		rt->rt_fibnum = fibnum;
1271 		/*
1272 		 * Add the gateway. Possibly re-malloc-ing the storage for it.
1273 		 */
1274 		RT_LOCK(rt);
1275 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1276 			ifa_free(ifa);
1277 			uma_zfree(V_rtzone, rt);
1278 			senderr(error);
1279 		}
1280 
1281 		/*
1282 		 * point to the (possibly newly malloc'd) dest address.
1283 		 */
1284 		ndst = (struct sockaddr *)rt_key(rt);
1285 
1286 		/*
1287 		 * make sure it contains the value we want (masked if needed).
1288 		 */
1289 		if (netmask) {
1290 			rt_maskedcopy(dst, ndst, netmask);
1291 		} else
1292 			bcopy(dst, ndst, dst->sa_len);
1293 
1294 		/*
1295 		 * We use the ifa reference returned by rt_getifa_fib().
1296 		 * This moved from below so that rnh->rnh_addaddr() can
1297 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1298 		 */
1299 		rt->rt_ifa = ifa;
1300 		rt->rt_ifp = ifa->ifa_ifp;
1301 		rt->rt_weight = 1;
1302 
1303 #ifdef RADIX_MPATH
1304 		/* do not permit exactly the same dst/mask/gw pair */
1305 		if (rn_mpath_capable(rnh) &&
1306 			rt_mpath_conflict(rnh, rt, netmask)) {
1307 			ifa_free(rt->rt_ifa);
1308 			Free(rt_key(rt));
1309 			uma_zfree(V_rtzone, rt);
1310 			senderr(EEXIST);
1311 		}
1312 #endif
1313 
1314 #ifdef FLOWTABLE
1315 		rt0 = NULL;
1316 		/* "flow-table" only supports IPv6 and IPv4 at the moment. */
1317 		switch (dst->sa_family) {
1318 #ifdef INET6
1319 		case AF_INET6:
1320 #endif
1321 #ifdef INET
1322 		case AF_INET:
1323 #endif
1324 #if defined(INET6) || defined(INET)
1325 			rn = rnh->rnh_matchaddr(dst, rnh);
1326 			if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1327 				struct sockaddr *mask;
1328 				u_char *m, *n;
1329 				int len;
1330 
1331 				/*
1332 				 * compare mask to see if the new route is
1333 				 * more specific than the existing one
1334 				 */
1335 				rt0 = RNTORT(rn);
1336 				RT_LOCK(rt0);
1337 				RT_ADDREF(rt0);
1338 				RT_UNLOCK(rt0);
1339 				/*
1340 				 * A host route is already present, so
1341 				 * leave the flow-table entries as is.
1342 				 */
1343 				if (rt0->rt_flags & RTF_HOST) {
1344 					RTFREE(rt0);
1345 					rt0 = NULL;
1346 				} else if (!(flags & RTF_HOST) && netmask) {
1347 					mask = rt_mask(rt0);
1348 					len = mask->sa_len;
1349 					m = (u_char *)mask;
1350 					n = (u_char *)netmask;
1351 					while (len-- > 0) {
1352 						if (*n != *m)
1353 							break;
1354 						n++;
1355 						m++;
1356 					}
1357 					if (len == 0 || (*n < *m)) {
1358 						RTFREE(rt0);
1359 						rt0 = NULL;
1360 					}
1361 				}
1362 			}
1363 #endif/* INET6 || INET */
1364 		}
1365 #endif /* FLOWTABLE */
1366 
1367 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1368 		rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1369 		/*
1370 		 * If it still failed to go into the tree,
1371 		 * then un-make it (this should be a function)
1372 		 */
1373 		if (rn == NULL) {
1374 			ifa_free(rt->rt_ifa);
1375 			Free(rt_key(rt));
1376 			uma_zfree(V_rtzone, rt);
1377 #ifdef FLOWTABLE
1378 			if (rt0 != NULL)
1379 				RTFREE(rt0);
1380 #endif
1381 			senderr(EEXIST);
1382 		}
1383 #ifdef FLOWTABLE
1384 		else if (rt0 != NULL) {
1385 			flowtable_route_flush(dst->sa_family, rt0);
1386 			RTFREE(rt0);
1387 		}
1388 #endif
1389 
1390 		/*
1391 		 * If this protocol has something to add to this then
1392 		 * allow it to do that as well.
1393 		 */
1394 		if (ifa->ifa_rtrequest)
1395 			ifa->ifa_rtrequest(req, rt, info);
1396 
1397 		/*
1398 		 * actually return a resultant rtentry and
1399 		 * give the caller a single reference.
1400 		 */
1401 		if (ret_nrt) {
1402 			*ret_nrt = rt;
1403 			RT_ADDREF(rt);
1404 		}
1405 		RT_UNLOCK(rt);
1406 		break;
1407 	default:
1408 		error = EOPNOTSUPP;
1409 	}
1410 bad:
1411 	if (needlock)
1412 		RADIX_NODE_HEAD_UNLOCK(rnh);
1413 	return (error);
1414 #undef senderr
1415 }
1416 
1417 #undef dst
1418 #undef gateway
1419 #undef netmask
1420 #undef ifaaddr
1421 #undef ifpaddr
1422 #undef flags
1423 
1424 int
1425 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1426 {
1427 	/* XXX dst may be overwritten, can we move this to below */
1428 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1429 #ifdef INVARIANTS
1430 	struct radix_node_head *rnh;
1431 
1432 	rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1433 #endif
1434 
1435 	RT_LOCK_ASSERT(rt);
1436 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1437 
1438 	/*
1439 	 * Prepare to store the gateway in rt->rt_gateway.
1440 	 * Both dst and gateway are stored one after the other in the same
1441 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1442 	 * rt_gateway already points to the right place.
1443 	 * Otherwise, malloc a new block and update the 'dst' address.
1444 	 */
1445 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1446 		caddr_t new;
1447 
1448 		R_Malloc(new, caddr_t, dlen + glen);
1449 		if (new == NULL)
1450 			return ENOBUFS;
1451 		/*
1452 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1453 		 * rt_setgate() can be called to initialize a newly
1454 		 * allocated route entry, in which case rt_key(rt) == NULL
1455 		 * (and also rt->rt_gateway == NULL).
1456 		 * Free()/free() handle a NULL argument just fine.
1457 		 */
1458 		bcopy(dst, new, dlen);
1459 		Free(rt_key(rt));	/* free old block, if any */
1460 		rt_key(rt) = (struct sockaddr *)new;
1461 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1462 	}
1463 
1464 	/*
1465 	 * Copy the new gateway value into the memory chunk.
1466 	 */
1467 	bcopy(gate, rt->rt_gateway, glen);
1468 
1469 	return (0);
1470 }
1471 
1472 void
1473 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1474 {
1475 	register u_char *cp1 = (u_char *)src;
1476 	register u_char *cp2 = (u_char *)dst;
1477 	register u_char *cp3 = (u_char *)netmask;
1478 	u_char *cplim = cp2 + *cp3;
1479 	u_char *cplim2 = cp2 + *cp1;
1480 
1481 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1482 	cp3 += 2;
1483 	if (cplim > cplim2)
1484 		cplim = cplim2;
1485 	while (cp2 < cplim)
1486 		*cp2++ = *cp1++ & *cp3++;
1487 	if (cp2 < cplim2)
1488 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1489 }
1490 
1491 /*
1492  * Set up a routing table entry, normally
1493  * for an interface.
1494  */
1495 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1496 static inline  int
1497 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1498 {
1499 	struct sockaddr *dst;
1500 	struct sockaddr *netmask;
1501 	struct rtentry *rt = NULL;
1502 	struct rt_addrinfo info;
1503 	int error = 0;
1504 	int startfib, endfib;
1505 	char tempbuf[_SOCKADDR_TMPSIZE];
1506 	int didwork = 0;
1507 	int a_failure = 0;
1508 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1509 	struct radix_node_head *rnh;
1510 
1511 	if (flags & RTF_HOST) {
1512 		dst = ifa->ifa_dstaddr;
1513 		netmask = NULL;
1514 	} else {
1515 		dst = ifa->ifa_addr;
1516 		netmask = ifa->ifa_netmask;
1517 	}
1518 	if (dst->sa_len == 0)
1519 		return(EINVAL);
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 	if (fibnum == RT_ALL_FIBS) {
1530 		if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1531 			startfib = endfib = curthread->td_proc->p_fibnum;
1532 		} else {
1533 			startfib = 0;
1534 			endfib = rt_numfibs - 1;
1535 		}
1536 	} else {
1537 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1538 		startfib = fibnum;
1539 		endfib = fibnum;
1540 	}
1541 
1542 	/*
1543 	 * If it's a delete, check that if it exists,
1544 	 * it's on the correct interface or we might scrub
1545 	 * a route to another ifa which would
1546 	 * be confusing at best and possibly worse.
1547 	 */
1548 	if (cmd == RTM_DELETE) {
1549 		/*
1550 		 * It's a delete, so it should already exist..
1551 		 * If it's a net, mask off the host bits
1552 		 * (Assuming we have a mask)
1553 		 * XXX this is kinda inet specific..
1554 		 */
1555 		if (netmask != NULL) {
1556 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1557 			dst = (struct sockaddr *)tempbuf;
1558 		}
1559 	}
1560 	/*
1561 	 * Now go through all the requested tables (fibs) and do the
1562 	 * requested action. Realistically, this will either be fib 0
1563 	 * for protocols that don't do multiple tables or all the
1564 	 * tables for those that do.
1565 	 */
1566 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1567 		if (cmd == RTM_DELETE) {
1568 			struct radix_node *rn;
1569 			/*
1570 			 * Look up an rtentry that is in the routing tree and
1571 			 * contains the correct info.
1572 			 */
1573 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1574 			if (rnh == NULL)
1575 				/* this table doesn't exist but others might */
1576 				continue;
1577 			RADIX_NODE_HEAD_RLOCK(rnh);
1578 			rn = rnh->rnh_lookup(dst, netmask, rnh);
1579 #ifdef RADIX_MPATH
1580 			if (rn_mpath_capable(rnh)) {
1581 
1582 				if (rn == NULL)
1583 					error = ESRCH;
1584 				else {
1585 					rt = RNTORT(rn);
1586 					/*
1587 					 * for interface route the
1588 					 * rt->rt_gateway is sockaddr_intf
1589 					 * for cloning ARP entries, so
1590 					 * rt_mpath_matchgate must use the
1591 					 * interface address
1592 					 */
1593 					rt = rt_mpath_matchgate(rt,
1594 					    ifa->ifa_addr);
1595 					if (rt == NULL)
1596 						error = ESRCH;
1597 				}
1598 			}
1599 #endif
1600 			error = (rn == NULL ||
1601 			    (rn->rn_flags & RNF_ROOT) ||
1602 			    RNTORT(rn)->rt_ifa != ifa);
1603 			RADIX_NODE_HEAD_RUNLOCK(rnh);
1604 			if (error) {
1605 				/* this is only an error if bad on ALL tables */
1606 				continue;
1607 			}
1608 		}
1609 		/*
1610 		 * Do the actual request
1611 		 */
1612 		bzero((caddr_t)&info, sizeof(info));
1613 		info.rti_ifa = ifa;
1614 		info.rti_flags = flags |
1615 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1616 		info.rti_info[RTAX_DST] = dst;
1617 		/*
1618 		 * doing this for compatibility reasons
1619 		 */
1620 		if (cmd == RTM_ADD)
1621 			info.rti_info[RTAX_GATEWAY] =
1622 			    (struct sockaddr *)&null_sdl;
1623 		else
1624 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1625 		info.rti_info[RTAX_NETMASK] = netmask;
1626 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1627 
1628 		if ((error == EEXIST) && (cmd == RTM_ADD)) {
1629 			/*
1630 			 * Interface route addition failed.
1631 			 * Atomically delete current prefix generating
1632 			 * RTM_DELETE message, and retry adding
1633 			 * interface prefix.
1634 			 */
1635 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1636 			RADIX_NODE_HEAD_LOCK(rnh);
1637 
1638 			/* Delete old prefix */
1639 			info.rti_ifa = NULL;
1640 			info.rti_flags = RTF_RNH_LOCKED;
1641 
1642 			error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1643 			if (error == 0) {
1644 				info.rti_ifa = ifa;
1645 				info.rti_flags = flags | RTF_RNH_LOCKED |
1646 				    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1647 				error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1648 			}
1649 
1650 			RADIX_NODE_HEAD_UNLOCK(rnh);
1651 		}
1652 
1653 
1654 		if (error == 0 && rt != NULL) {
1655 			/*
1656 			 * notify any listening routing agents of the change
1657 			 */
1658 			RT_LOCK(rt);
1659 #ifdef RADIX_MPATH
1660 			/*
1661 			 * in case address alias finds the first address
1662 			 * e.g. ifconfig bge0 192.0.2.246/24
1663 			 * e.g. ifconfig bge0 192.0.2.247/24
1664 			 * the address set in the route is 192.0.2.246
1665 			 * so we need to replace it with 192.0.2.247
1666 			 */
1667 			if (memcmp(rt->rt_ifa->ifa_addr,
1668 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1669 				ifa_free(rt->rt_ifa);
1670 				ifa_ref(ifa);
1671 				rt->rt_ifp = ifa->ifa_ifp;
1672 				rt->rt_ifa = ifa;
1673 			}
1674 #endif
1675 			/*
1676 			 * doing this for compatibility reasons
1677 			 */
1678 			if (cmd == RTM_ADD) {
1679 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1680 				rt->rt_ifp->if_type;
1681 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1682 				rt->rt_ifp->if_index;
1683 			}
1684 			RT_ADDREF(rt);
1685 			RT_UNLOCK(rt);
1686 			rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1687 			RT_LOCK(rt);
1688 			RT_REMREF(rt);
1689 			if (cmd == RTM_DELETE) {
1690 				/*
1691 				 * If we are deleting, and we found an entry,
1692 				 * then it's been removed from the tree..
1693 				 * now throw it away.
1694 				 */
1695 				RTFREE_LOCKED(rt);
1696 			} else {
1697 				if (cmd == RTM_ADD) {
1698 					/*
1699 					 * We just wanted to add it..
1700 					 * we don't actually need a reference.
1701 					 */
1702 					RT_REMREF(rt);
1703 				}
1704 				RT_UNLOCK(rt);
1705 			}
1706 			didwork = 1;
1707 		}
1708 		if (error)
1709 			a_failure = error;
1710 	}
1711 	if (cmd == RTM_DELETE) {
1712 		if (didwork) {
1713 			error = 0;
1714 		} else {
1715 			/* we only give an error if it wasn't in any table */
1716 			error = ((flags & RTF_HOST) ?
1717 			    EHOSTUNREACH : ENETUNREACH);
1718 		}
1719 	} else {
1720 		if (a_failure) {
1721 			/* return an error if any of them failed */
1722 			error = a_failure;
1723 		}
1724 	}
1725 	return (error);
1726 }
1727 
1728 #ifndef BURN_BRIDGES
1729 /* special one for inet internal use. may not use. */
1730 int
1731 rtinit_fib(struct ifaddr *ifa, int cmd, int flags)
1732 {
1733 	return (rtinit1(ifa, cmd, flags, RT_ALL_FIBS));
1734 }
1735 #endif
1736 
1737 /*
1738  * Set up a routing table entry, normally
1739  * for an interface.
1740  */
1741 int
1742 rtinit(struct ifaddr *ifa, int cmd, int flags)
1743 {
1744 	struct sockaddr *dst;
1745 	int fib = RT_DEFAULT_FIB;
1746 
1747 	if (flags & RTF_HOST) {
1748 		dst = ifa->ifa_dstaddr;
1749 	} else {
1750 		dst = ifa->ifa_addr;
1751 	}
1752 
1753 	switch (dst->sa_family) {
1754 	case AF_INET6:
1755 	case AF_INET:
1756 		/* We do support multiple FIBs. */
1757 		fib = RT_ALL_FIBS;
1758 		break;
1759 	}
1760 	return (rtinit1(ifa, cmd, flags, fib));
1761 }
1762 
1763 /*
1764  * Announce interface address arrival/withdraw
1765  * Returns 0 on success.
1766  */
1767 int
1768 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
1769 {
1770 
1771 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1772 	    ("unexpected cmd %d", cmd));
1773 
1774 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1775 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1776 
1777 	return (rtsock_addrmsg(cmd, ifa, fibnum));
1778 }
1779 
1780 /*
1781  * Announce route addition/removal.
1782  * Users of this function MUST validate input data BEFORE calling.
1783  * However we have to be able to handle invalid data:
1784  * if some userland app sends us "invalid" route message (invalid mask,
1785  * no dst, wrong address families, etc...) we need to pass it back
1786  * to app (and any other rtsock consumers) with rtm_errno field set to
1787  * non-zero value.
1788  * Returns 0 on success.
1789  */
1790 int
1791 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
1792     int fibnum)
1793 {
1794 
1795 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1796 	    ("unexpected cmd %d", cmd));
1797 
1798 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1799 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1800 
1801 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
1802 
1803 	return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
1804 }
1805 
1806 void
1807 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
1808 {
1809 
1810 	rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
1811 }
1812 
1813 /*
1814  * This is called to generate messages from the routing socket
1815  * indicating a network interface has had addresses associated with it.
1816  */
1817 void
1818 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
1819     int fibnum)
1820 {
1821 
1822 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1823 		("unexpected cmd %u", cmd));
1824 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1825 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1826 
1827 #if defined(INET) || defined(INET6)
1828 #ifdef SCTP
1829 	/*
1830 	 * notify the SCTP stack
1831 	 * this will only get called when an address is added/deleted
1832 	 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
1833 	 */
1834 	sctp_addr_change(ifa, cmd);
1835 #endif /* SCTP */
1836 #endif
1837 	if (cmd == RTM_ADD) {
1838 		rt_addrmsg(cmd, ifa, fibnum);
1839 		if (rt != NULL)
1840 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1841 	} else {
1842 		if (rt != NULL)
1843 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1844 		rt_addrmsg(cmd, ifa, fibnum);
1845 	}
1846 }
1847 
1848