xref: /freebsd/sys/net/route.c (revision 8fc257994d0ce2396196d7a06d50d20c8015f4b7)
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_route.h"
39 #include "opt_mrouting.h"
40 #include "opt_mpath.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/syslog.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/socket.h>
48 #include <sys/sysctl.h>
49 #include <sys/syslog.h>
50 #include <sys/sysproto.h>
51 #include <sys/proc.h>
52 #include <sys/domain.h>
53 #include <sys/kernel.h>
54 
55 #include <net/if.h>
56 #include <net/if_dl.h>
57 #include <net/route.h>
58 #include <net/vnet.h>
59 #include <net/flowtable.h>
60 
61 #ifdef RADIX_MPATH
62 #include <net/radix_mpath.h>
63 #endif
64 
65 #include <netinet/in.h>
66 #include <netinet/ip_mroute.h>
67 
68 #include <vm/uma.h>
69 
70 u_int rt_numfibs = RT_NUMFIBS;
71 SYSCTL_INT(_net, OID_AUTO, fibs, CTLFLAG_RD, &rt_numfibs, 0, "");
72 /*
73  * Allow the boot code to allow LESS than RT_MAXFIBS to be used.
74  * We can't do more because storage is statically allocated for now.
75  * (for compatibility reasons.. this will change).
76  */
77 TUNABLE_INT("net.fibs", &rt_numfibs);
78 
79 /*
80  * By default add routes to all fibs for new interfaces.
81  * Once this is set to 0 then only allocate routes on interface
82  * changes for the FIB of the caller when adding a new set of addresses
83  * to an interface.  XXX this is a shotgun aproach to a problem that needs
84  * a more fine grained solution.. that will come.
85  */
86 u_int rt_add_addr_allfibs = 1;
87 SYSCTL_INT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RW,
88     &rt_add_addr_allfibs, 0, "");
89 TUNABLE_INT("net.add_addr_allfibs", &rt_add_addr_allfibs);
90 
91 VNET_DEFINE(struct rtstat, rtstat);
92 #define	V_rtstat	VNET(rtstat)
93 
94 VNET_DEFINE(struct radix_node_head *, rt_tables);
95 #define	V_rt_tables	VNET(rt_tables)
96 
97 VNET_DEFINE(int, rttrash);		/* routes not in table but not freed */
98 #define	V_rttrash	VNET(rttrash)
99 
100 
101 /* compare two sockaddr structures */
102 #define	sa_equal(a1, a2) (bcmp((a1), (a2), (a1)->sa_len) == 0)
103 
104 /*
105  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
106  * The operation can be done safely (in this code) because a
107  * 'struct rtentry' starts with two 'struct radix_node''s, the first
108  * one representing leaf nodes in the routing tree, which is
109  * what the code in radix.c passes us as a 'struct radix_node'.
110  *
111  * But because there are a lot of assumptions in this conversion,
112  * do not cast explicitly, but always use the macro below.
113  */
114 #define RNTORT(p)	((struct rtentry *)(p))
115 
116 static VNET_DEFINE(uma_zone_t, rtzone);		/* Routing table UMA zone. */
117 #define	V_rtzone	VNET(rtzone)
118 
119 #if 0
120 /* default fib for tunnels to use */
121 u_int tunnel_fib = 0;
122 SYSCTL_INT(_net, OID_AUTO, tunnelfib, CTLFLAG_RD, &tunnel_fib, 0, "");
123 #endif
124 
125 /*
126  * handler for net.my_fibnum
127  */
128 static int
129 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
130 {
131         int fibnum;
132         int error;
133 
134         fibnum = curthread->td_proc->p_fibnum;
135         error = sysctl_handle_int(oidp, &fibnum, 0, req);
136         return (error);
137 }
138 
139 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
140             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
141 
142 static __inline struct radix_node_head **
143 rt_tables_get_rnh_ptr(int table, int fam)
144 {
145 	struct radix_node_head **rnh;
146 
147 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
148 	    __func__));
149 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
150 	    __func__));
151 
152 	/* rnh is [fib=0][af=0]. */
153 	rnh = (struct radix_node_head **)V_rt_tables;
154 	/* Get the offset to the requested table and fam. */
155 	rnh += table * (AF_MAX+1) + fam;
156 
157 	return (rnh);
158 }
159 
160 struct radix_node_head *
161 rt_tables_get_rnh(int table, int fam)
162 {
163 
164 	return (*rt_tables_get_rnh_ptr(table, fam));
165 }
166 
167 /*
168  * route initialization must occur before ip6_init2(), which happenas at
169  * SI_ORDER_MIDDLE.
170  */
171 static void
172 route_init(void)
173 {
174 	struct domain *dom;
175 	int max_keylen = 0;
176 
177 	/* whack the tunable ints into  line. */
178 	if (rt_numfibs > RT_MAXFIBS)
179 		rt_numfibs = RT_MAXFIBS;
180 	if (rt_numfibs == 0)
181 		rt_numfibs = 1;
182 
183 	for (dom = domains; dom; dom = dom->dom_next)
184 		if (dom->dom_maxrtkey > max_keylen)
185 			max_keylen = dom->dom_maxrtkey;
186 
187 	rn_init(max_keylen);	/* init all zeroes, all ones, mask table */
188 }
189 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
190 
191 static void
192 vnet_route_init(const void *unused __unused)
193 {
194 	struct domain *dom;
195 	struct radix_node_head **rnh;
196 	int table;
197 	int fam;
198 
199 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
200 	    sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
201 
202 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), NULL, NULL,
203 	    NULL, NULL, UMA_ALIGN_PTR, 0);
204 	for (dom = domains; dom; dom = dom->dom_next) {
205 		if (dom->dom_rtattach)  {
206 			for  (table = 0; table < rt_numfibs; table++) {
207 				if ( (fam = dom->dom_family) == AF_INET ||
208 				    table == 0) {
209  			        	/* for now only AF_INET has > 1 table */
210 					/* XXX MRT
211 					 * rtattach will be also called
212 					 * from vfs_export.c but the
213 					 * offset will be 0
214 					 * (only for AF_INET and AF_INET6
215 					 * which don't need it anyhow)
216 					 */
217 					rnh = rt_tables_get_rnh_ptr(table, fam);
218 					if (rnh == NULL)
219 						panic("%s: rnh NULL", __func__);
220 					dom->dom_rtattach((void **)rnh,
221 				    	    dom->dom_rtoffset);
222 				} else {
223 					break;
224 				}
225 			}
226 		}
227 	}
228 }
229 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
230     vnet_route_init, 0);
231 
232 #ifdef VIMAGE
233 static void
234 vnet_route_uninit(const void *unused __unused)
235 {
236 	int table;
237 	int fam;
238 	struct domain *dom;
239 	struct radix_node_head **rnh;
240 
241 	for (dom = domains; dom; dom = dom->dom_next) {
242 		if (dom->dom_rtdetach) {
243 			for (table = 0; table < rt_numfibs; table++) {
244 				if ( (fam = dom->dom_family) == AF_INET ||
245 				    table == 0) {
246 					/* For now only AF_INET has > 1 tbl. */
247 					rnh = rt_tables_get_rnh_ptr(table, fam);
248 					if (rnh == NULL)
249 						panic("%s: rnh NULL", __func__);
250 					dom->dom_rtdetach((void **)rnh,
251 					    dom->dom_rtoffset);
252 				} else {
253 					break;
254 				}
255 			}
256 		}
257 	}
258 }
259 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
260     vnet_route_uninit, 0);
261 #endif
262 
263 #ifndef _SYS_SYSPROTO_H_
264 struct setfib_args {
265 	int     fibnum;
266 };
267 #endif
268 int
269 setfib(struct thread *td, struct setfib_args *uap)
270 {
271 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
272 		return EINVAL;
273 	td->td_proc->p_fibnum = uap->fibnum;
274 	return (0);
275 }
276 
277 /*
278  * Packet routing routines.
279  */
280 void
281 rtalloc(struct route *ro)
282 {
283 	rtalloc_ign_fib(ro, 0UL, 0);
284 }
285 
286 void
287 rtalloc_fib(struct route *ro, u_int fibnum)
288 {
289 	rtalloc_ign_fib(ro, 0UL, fibnum);
290 }
291 
292 void
293 rtalloc_ign(struct route *ro, u_long ignore)
294 {
295 	struct rtentry *rt;
296 
297 	if ((rt = ro->ro_rt) != NULL) {
298 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
299 			return;
300 		RTFREE(rt);
301 		ro->ro_rt = NULL;
302 	}
303 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, 0);
304 	if (ro->ro_rt)
305 		RT_UNLOCK(ro->ro_rt);
306 }
307 
308 void
309 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
310 {
311 	struct rtentry *rt;
312 
313 	if ((rt = ro->ro_rt) != NULL) {
314 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
315 			return;
316 		RTFREE(rt);
317 		ro->ro_rt = NULL;
318 	}
319 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
320 	if (ro->ro_rt)
321 		RT_UNLOCK(ro->ro_rt);
322 }
323 
324 /*
325  * Look up the route that matches the address given
326  * Or, at least try.. Create a cloned route if needed.
327  *
328  * The returned route, if any, is locked.
329  */
330 struct rtentry *
331 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
332 {
333 	return (rtalloc1_fib(dst, report, ignflags, 0));
334 }
335 
336 struct rtentry *
337 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
338 		    u_int fibnum)
339 {
340 	struct radix_node_head *rnh;
341 	struct rtentry *rt;
342 	struct radix_node *rn;
343 	struct rtentry *newrt;
344 	struct rt_addrinfo info;
345 	int err = 0, msgtype = RTM_MISS;
346 	int needlock;
347 
348 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
349 	if (dst->sa_family != AF_INET)	/* Only INET supports > 1 fib now */
350 		fibnum = 0;
351 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
352 	newrt = NULL;
353 	/*
354 	 * Look up the address in the table for that Address Family
355 	 */
356 	if (rnh == NULL) {
357 		V_rtstat.rts_unreach++;
358 		goto miss;
359 	}
360 	needlock = !(ignflags & RTF_RNH_LOCKED);
361 	if (needlock)
362 		RADIX_NODE_HEAD_RLOCK(rnh);
363 #ifdef INVARIANTS
364 	else
365 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
366 #endif
367 	rn = rnh->rnh_matchaddr(dst, rnh);
368 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
369 		newrt = rt = RNTORT(rn);
370 		RT_LOCK(newrt);
371 		RT_ADDREF(newrt);
372 		if (needlock)
373 			RADIX_NODE_HEAD_RUNLOCK(rnh);
374 		goto done;
375 
376 	} else if (needlock)
377 		RADIX_NODE_HEAD_RUNLOCK(rnh);
378 
379 	/*
380 	 * Either we hit the root or couldn't find any match,
381 	 * Which basically means
382 	 * "caint get there frm here"
383 	 */
384 	V_rtstat.rts_unreach++;
385 miss:
386 	if (report) {
387 		/*
388 		 * If required, report the failure to the supervising
389 		 * Authorities.
390 		 * For a delete, this is not an error. (report == 0)
391 		 */
392 		bzero(&info, sizeof(info));
393 		info.rti_info[RTAX_DST] = dst;
394 		rt_missmsg(msgtype, &info, 0, err);
395 	}
396 done:
397 	if (newrt)
398 		RT_LOCK_ASSERT(newrt);
399 	return (newrt);
400 }
401 
402 /*
403  * Remove a reference count from an rtentry.
404  * If the count gets low enough, take it out of the routing table
405  */
406 void
407 rtfree(struct rtentry *rt)
408 {
409 	struct radix_node_head *rnh;
410 
411 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
412 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
413 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
414 
415 	RT_LOCK_ASSERT(rt);
416 
417 	/*
418 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
419 	 * we should come here exactly with the last reference.
420 	 */
421 	RT_REMREF(rt);
422 	if (rt->rt_refcnt > 0) {
423 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
424 		goto done;
425 	}
426 
427 	/*
428 	 * On last reference give the "close method" a chance
429 	 * to cleanup private state.  This also permits (for
430 	 * IPv4 and IPv6) a chance to decide if the routing table
431 	 * entry should be purged immediately or at a later time.
432 	 * When an immediate purge is to happen the close routine
433 	 * typically calls rtexpunge which clears the RTF_UP flag
434 	 * on the entry so that the code below reclaims the storage.
435 	 */
436 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
437 		rnh->rnh_close((struct radix_node *)rt, rnh);
438 
439 	/*
440 	 * If we are no longer "up" (and ref == 0)
441 	 * then we can free the resources associated
442 	 * with the route.
443 	 */
444 	if ((rt->rt_flags & RTF_UP) == 0) {
445 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
446 			panic("rtfree 2");
447 		/*
448 		 * the rtentry must have been removed from the routing table
449 		 * so it is represented in rttrash.. remove that now.
450 		 */
451 		V_rttrash--;
452 #ifdef	DIAGNOSTIC
453 		if (rt->rt_refcnt < 0) {
454 			printf("rtfree: %p not freed (neg refs)\n", rt);
455 			goto done;
456 		}
457 #endif
458 		/*
459 		 * release references on items we hold them on..
460 		 * e.g other routes and ifaddrs.
461 		 */
462 		if (rt->rt_ifa)
463 			ifa_free(rt->rt_ifa);
464 		/*
465 		 * The key is separatly alloc'd so free it (see rt_setgate()).
466 		 * This also frees the gateway, as they are always malloc'd
467 		 * together.
468 		 */
469 		Free(rt_key(rt));
470 
471 		/*
472 		 * and the rtentry itself of course
473 		 */
474 		RT_LOCK_DESTROY(rt);
475 		uma_zfree(V_rtzone, rt);
476 		return;
477 	}
478 done:
479 	RT_UNLOCK(rt);
480 }
481 
482 
483 /*
484  * Force a routing table entry to the specified
485  * destination to go through the given gateway.
486  * Normally called as a result of a routing redirect
487  * message from the network layer.
488  */
489 void
490 rtredirect(struct sockaddr *dst,
491 	struct sockaddr *gateway,
492 	struct sockaddr *netmask,
493 	int flags,
494 	struct sockaddr *src)
495 {
496 	rtredirect_fib(dst, gateway, netmask, flags, src, 0);
497 }
498 
499 void
500 rtredirect_fib(struct sockaddr *dst,
501 	struct sockaddr *gateway,
502 	struct sockaddr *netmask,
503 	int flags,
504 	struct sockaddr *src,
505 	u_int fibnum)
506 {
507 	struct rtentry *rt, *rt0 = NULL;
508 	int error = 0;
509 	short *stat = NULL;
510 	struct rt_addrinfo info;
511 	struct ifaddr *ifa;
512 	struct radix_node_head *rnh;
513 
514 	ifa = NULL;
515 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
516 	if (rnh == NULL) {
517 		error = EAFNOSUPPORT;
518 		goto out;
519 	}
520 
521 	/* verify the gateway is directly reachable */
522 	if ((ifa = ifa_ifwithnet(gateway, 0)) == NULL) {
523 		error = ENETUNREACH;
524 		goto out;
525 	}
526 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
527 	/*
528 	 * If the redirect isn't from our current router for this dst,
529 	 * it's either old or wrong.  If it redirects us to ourselves,
530 	 * we have a routing loop, perhaps as a result of an interface
531 	 * going down recently.
532 	 */
533 	if (!(flags & RTF_DONE) && rt &&
534 	     (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
535 		error = EINVAL;
536 	else if (ifa_ifwithaddr_check(gateway))
537 		error = EHOSTUNREACH;
538 	if (error)
539 		goto done;
540 	/*
541 	 * Create a new entry if we just got back a wildcard entry
542 	 * or the the lookup failed.  This is necessary for hosts
543 	 * which use routing redirects generated by smart gateways
544 	 * to dynamically build the routing tables.
545 	 */
546 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
547 		goto create;
548 	/*
549 	 * Don't listen to the redirect if it's
550 	 * for a route to an interface.
551 	 */
552 	if (rt->rt_flags & RTF_GATEWAY) {
553 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
554 			/*
555 			 * Changing from route to net => route to host.
556 			 * Create new route, rather than smashing route to net.
557 			 */
558 		create:
559 			rt0 = rt;
560 			rt = NULL;
561 
562 			flags |=  RTF_GATEWAY | RTF_DYNAMIC;
563 			bzero((caddr_t)&info, sizeof(info));
564 			info.rti_info[RTAX_DST] = dst;
565 			info.rti_info[RTAX_GATEWAY] = gateway;
566 			info.rti_info[RTAX_NETMASK] = netmask;
567 			info.rti_ifa = ifa;
568 			info.rti_flags = flags;
569 			if (rt0 != NULL)
570 				RT_UNLOCK(rt0);	/* drop lock to avoid LOR with RNH */
571 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
572 			if (rt != NULL) {
573 				RT_LOCK(rt);
574 				if (rt0 != NULL)
575 					EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
576 				flags = rt->rt_flags;
577 			}
578 			if (rt0 != NULL)
579 				RTFREE(rt0);
580 
581 			stat = &V_rtstat.rts_dynamic;
582 		} else {
583 			struct rtentry *gwrt;
584 
585 			/*
586 			 * Smash the current notion of the gateway to
587 			 * this destination.  Should check about netmask!!!
588 			 */
589 			rt->rt_flags |= RTF_MODIFIED;
590 			flags |= RTF_MODIFIED;
591 			stat = &V_rtstat.rts_newgateway;
592 			/*
593 			 * add the key and gateway (in one malloc'd chunk).
594 			 */
595 			RT_UNLOCK(rt);
596 			RADIX_NODE_HEAD_LOCK(rnh);
597 			RT_LOCK(rt);
598 			rt_setgate(rt, rt_key(rt), gateway);
599 			gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
600 			RADIX_NODE_HEAD_UNLOCK(rnh);
601 			EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
602 			RTFREE_LOCKED(gwrt);
603 		}
604 	} else
605 		error = EHOSTUNREACH;
606 done:
607 	if (rt)
608 		RTFREE_LOCKED(rt);
609 out:
610 	if (error)
611 		V_rtstat.rts_badredirect++;
612 	else if (stat != NULL)
613 		(*stat)++;
614 	bzero((caddr_t)&info, sizeof(info));
615 	info.rti_info[RTAX_DST] = dst;
616 	info.rti_info[RTAX_GATEWAY] = gateway;
617 	info.rti_info[RTAX_NETMASK] = netmask;
618 	info.rti_info[RTAX_AUTHOR] = src;
619 	rt_missmsg(RTM_REDIRECT, &info, flags, error);
620 	if (ifa != NULL)
621 		ifa_free(ifa);
622 }
623 
624 int
625 rtioctl(u_long req, caddr_t data)
626 {
627 	return (rtioctl_fib(req, data, 0));
628 }
629 
630 /*
631  * Routing table ioctl interface.
632  */
633 int
634 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
635 {
636 
637 	/*
638 	 * If more ioctl commands are added here, make sure the proper
639 	 * super-user checks are being performed because it is possible for
640 	 * prison-root to make it this far if raw sockets have been enabled
641 	 * in jails.
642 	 */
643 #ifdef INET
644 	/* Multicast goop, grrr... */
645 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
646 #else /* INET */
647 	return ENXIO;
648 #endif /* INET */
649 }
650 
651 /*
652  * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
653  */
654 struct ifaddr *
655 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
656 {
657 	return (ifa_ifwithroute_fib(flags, dst, gateway, 0));
658 }
659 
660 struct ifaddr *
661 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
662 				u_int fibnum)
663 {
664 	register struct ifaddr *ifa;
665 	int not_found = 0;
666 
667 	if ((flags & RTF_GATEWAY) == 0) {
668 		/*
669 		 * If we are adding a route to an interface,
670 		 * and the interface is a pt to pt link
671 		 * we should search for the destination
672 		 * as our clue to the interface.  Otherwise
673 		 * we can use the local address.
674 		 */
675 		ifa = NULL;
676 		if (flags & RTF_HOST)
677 			ifa = ifa_ifwithdstaddr(dst);
678 		if (ifa == NULL)
679 			ifa = ifa_ifwithaddr(gateway);
680 	} else {
681 		/*
682 		 * If we are adding a route to a remote net
683 		 * or host, the gateway may still be on the
684 		 * other end of a pt to pt link.
685 		 */
686 		ifa = ifa_ifwithdstaddr(gateway);
687 	}
688 	if (ifa == NULL)
689 		ifa = ifa_ifwithnet(gateway, 0);
690 	if (ifa == NULL) {
691 		struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
692 		if (rt == NULL)
693 			return (NULL);
694 		/*
695 		 * dismiss a gateway that is reachable only
696 		 * through the default router
697 		 */
698 		switch (gateway->sa_family) {
699 		case AF_INET:
700 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
701 				not_found = 1;
702 			break;
703 		case AF_INET6:
704 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
705 				not_found = 1;
706 			break;
707 		default:
708 			break;
709 		}
710 		if (!not_found && rt->rt_ifa != NULL) {
711 			ifa = rt->rt_ifa;
712 			ifa_ref(ifa);
713 		}
714 		RT_REMREF(rt);
715 		RT_UNLOCK(rt);
716 		if (not_found || ifa == NULL)
717 			return (NULL);
718 	}
719 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
720 		struct ifaddr *oifa = ifa;
721 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
722 		if (ifa == NULL)
723 			ifa = oifa;
724 		else
725 			ifa_free(oifa);
726 	}
727 	return (ifa);
728 }
729 
730 /*
731  * Do appropriate manipulations of a routing tree given
732  * all the bits of info needed
733  */
734 int
735 rtrequest(int req,
736 	struct sockaddr *dst,
737 	struct sockaddr *gateway,
738 	struct sockaddr *netmask,
739 	int flags,
740 	struct rtentry **ret_nrt)
741 {
742 	return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt, 0));
743 }
744 
745 int
746 rtrequest_fib(int req,
747 	struct sockaddr *dst,
748 	struct sockaddr *gateway,
749 	struct sockaddr *netmask,
750 	int flags,
751 	struct rtentry **ret_nrt,
752 	u_int fibnum)
753 {
754 	struct rt_addrinfo info;
755 
756 	if (dst->sa_len == 0)
757 		return(EINVAL);
758 
759 	bzero((caddr_t)&info, sizeof(info));
760 	info.rti_flags = flags;
761 	info.rti_info[RTAX_DST] = dst;
762 	info.rti_info[RTAX_GATEWAY] = gateway;
763 	info.rti_info[RTAX_NETMASK] = netmask;
764 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
765 }
766 
767 /*
768  * These (questionable) definitions of apparent local variables apply
769  * to the next two functions.  XXXXXX!!!
770  */
771 #define	dst	info->rti_info[RTAX_DST]
772 #define	gateway	info->rti_info[RTAX_GATEWAY]
773 #define	netmask	info->rti_info[RTAX_NETMASK]
774 #define	ifaaddr	info->rti_info[RTAX_IFA]
775 #define	ifpaddr	info->rti_info[RTAX_IFP]
776 #define	flags	info->rti_flags
777 
778 int
779 rt_getifa(struct rt_addrinfo *info)
780 {
781 	return (rt_getifa_fib(info, 0));
782 }
783 
784 /*
785  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
786  * it will be referenced so the caller must free it.
787  */
788 int
789 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
790 {
791 	struct ifaddr *ifa;
792 	int error = 0;
793 
794 	/*
795 	 * ifp may be specified by sockaddr_dl
796 	 * when protocol address is ambiguous.
797 	 */
798 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
799 	    ifpaddr->sa_family == AF_LINK &&
800 	    (ifa = ifa_ifwithnet(ifpaddr, 0)) != NULL) {
801 		info->rti_ifp = ifa->ifa_ifp;
802 		ifa_free(ifa);
803 	}
804 	if (info->rti_ifa == NULL && ifaaddr != NULL)
805 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
806 	if (info->rti_ifa == NULL) {
807 		struct sockaddr *sa;
808 
809 		sa = ifaaddr != NULL ? ifaaddr :
810 		    (gateway != NULL ? gateway : dst);
811 		if (sa != NULL && info->rti_ifp != NULL)
812 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
813 		else if (dst != NULL && gateway != NULL)
814 			info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
815 							fibnum);
816 		else if (sa != NULL)
817 			info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
818 							fibnum);
819 	}
820 	if ((ifa = info->rti_ifa) != NULL) {
821 		if (info->rti_ifp == NULL)
822 			info->rti_ifp = ifa->ifa_ifp;
823 	} else
824 		error = ENETUNREACH;
825 	return (error);
826 }
827 
828 /*
829  * Expunges references to a route that's about to be reclaimed.
830  * The route must be locked.
831  */
832 int
833 rtexpunge(struct rtentry *rt)
834 {
835 #if !defined(RADIX_MPATH)
836 	struct radix_node *rn;
837 #else
838 	struct rt_addrinfo info;
839 	int fib;
840 	struct rtentry *rt0;
841 #endif
842 	struct radix_node_head *rnh;
843 	struct ifaddr *ifa;
844 	int error = 0;
845 
846 	/*
847 	 * Find the correct routing tree to use for this Address Family
848 	 */
849 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
850 	RT_LOCK_ASSERT(rt);
851 	if (rnh == NULL)
852 		return (EAFNOSUPPORT);
853 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
854 
855 #ifdef RADIX_MPATH
856 	fib = rt->rt_fibnum;
857 	bzero(&info, sizeof(info));
858 	info.rti_ifp = rt->rt_ifp;
859 	info.rti_flags = RTF_RNH_LOCKED;
860 	info.rti_info[RTAX_DST] = rt_key(rt);
861 	info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
862 
863 	RT_UNLOCK(rt);
864 	error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
865 
866 	if (error == 0 && rt0 != NULL) {
867 		rt = rt0;
868 		RT_LOCK(rt);
869 	} else if (error != 0) {
870 		RT_LOCK(rt);
871 		return (error);
872 	}
873 #else
874 	/*
875 	 * Remove the item from the tree; it should be there,
876 	 * but when callers invoke us blindly it may not (sigh).
877 	 */
878 	rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
879 	if (rn == NULL) {
880 		error = ESRCH;
881 		goto bad;
882 	}
883 	KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
884 		("unexpected flags 0x%x", rn->rn_flags));
885 	KASSERT(rt == RNTORT(rn),
886 		("lookup mismatch, rt %p rn %p", rt, rn));
887 #endif /* RADIX_MPATH */
888 
889 	rt->rt_flags &= ~RTF_UP;
890 
891 	/*
892 	 * Give the protocol a chance to keep things in sync.
893 	 */
894 	if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
895 		struct rt_addrinfo info;
896 
897 		bzero((caddr_t)&info, sizeof(info));
898 		info.rti_flags = rt->rt_flags;
899 		info.rti_info[RTAX_DST] = rt_key(rt);
900 		info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
901 		info.rti_info[RTAX_NETMASK] = rt_mask(rt);
902 		ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
903 	}
904 
905 	/*
906 	 * one more rtentry floating around that is not
907 	 * linked to the routing table.
908 	 */
909 	V_rttrash++;
910 #if !defined(RADIX_MPATH)
911 bad:
912 #endif
913 	return (error);
914 }
915 
916 #ifdef RADIX_MPATH
917 static int
918 rn_mpath_update(int req, struct rt_addrinfo *info,
919     struct radix_node_head *rnh, struct rtentry **ret_nrt)
920 {
921 	/*
922 	 * if we got multipath routes, we require users to specify
923 	 * a matching RTAX_GATEWAY.
924 	 */
925 	struct rtentry *rt, *rto = NULL;
926 	register struct radix_node *rn;
927 	int error = 0;
928 
929 	rn = rnh->rnh_matchaddr(dst, rnh);
930 	if (rn == NULL)
931 		return (ESRCH);
932 	rto = rt = RNTORT(rn);
933 	rt = rt_mpath_matchgate(rt, gateway);
934 	if (rt == NULL)
935 		return (ESRCH);
936 	/*
937 	 * this is the first entry in the chain
938 	 */
939 	if (rto == rt) {
940 		rn = rn_mpath_next((struct radix_node *)rt);
941 		/*
942 		 * there is another entry, now it's active
943 		 */
944 		if (rn) {
945 			rto = RNTORT(rn);
946 			RT_LOCK(rto);
947 			rto->rt_flags |= RTF_UP;
948 			RT_UNLOCK(rto);
949 		} else if (rt->rt_flags & RTF_GATEWAY) {
950 			/*
951 			 * For gateway routes, we need to
952 			 * make sure that we we are deleting
953 			 * the correct gateway.
954 			 * rt_mpath_matchgate() does not
955 			 * check the case when there is only
956 			 * one route in the chain.
957 			 */
958 			if (gateway &&
959 			    (rt->rt_gateway->sa_len != gateway->sa_len ||
960 				memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
961 				error = ESRCH;
962 			else {
963 				/*
964 				 * remove from tree before returning it
965 				 * to the caller
966 				 */
967 				rn = rnh->rnh_deladdr(dst, netmask, rnh);
968 				KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
969 				goto gwdelete;
970 			}
971 
972 		}
973 		/*
974 		 * use the normal delete code to remove
975 		 * the first entry
976 		 */
977 		if (req != RTM_DELETE)
978 			goto nondelete;
979 
980 		error = ENOENT;
981 		goto done;
982 	}
983 
984 	/*
985 	 * if the entry is 2nd and on up
986 	 */
987 	if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
988 		panic ("rtrequest1: rt_mpath_deldup");
989 gwdelete:
990 	RT_LOCK(rt);
991 	RT_ADDREF(rt);
992 	if (req == RTM_DELETE) {
993 		rt->rt_flags &= ~RTF_UP;
994 		/*
995 		 * One more rtentry floating around that is not
996 		 * linked to the routing table. rttrash will be decremented
997 		 * when RTFREE(rt) is eventually called.
998 		 */
999 		V_rttrash++;
1000 	}
1001 
1002 nondelete:
1003 	if (req != RTM_DELETE)
1004 		panic("unrecognized request %d", req);
1005 
1006 
1007 	/*
1008 	 * If the caller wants it, then it can have it,
1009 	 * but it's up to it to free the rtentry as we won't be
1010 	 * doing it.
1011 	 */
1012 	if (ret_nrt) {
1013 		*ret_nrt = rt;
1014 		RT_UNLOCK(rt);
1015 	} else
1016 		RTFREE_LOCKED(rt);
1017 done:
1018 	return (error);
1019 }
1020 #endif
1021 
1022 int
1023 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1024 				u_int fibnum)
1025 {
1026 	int error = 0, needlock = 0;
1027 	register struct rtentry *rt;
1028 #ifdef FLOWTABLE
1029 	register struct rtentry *rt0;
1030 #endif
1031 	register struct radix_node *rn;
1032 	register struct radix_node_head *rnh;
1033 	struct ifaddr *ifa;
1034 	struct sockaddr *ndst;
1035 #define senderr(x) { error = x ; goto bad; }
1036 
1037 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1038 	if (dst->sa_family != AF_INET)	/* Only INET supports > 1 fib now */
1039 		fibnum = 0;
1040 	/*
1041 	 * Find the correct routing tree to use for this Address Family
1042 	 */
1043 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1044 	if (rnh == NULL)
1045 		return (EAFNOSUPPORT);
1046 	needlock = ((flags & RTF_RNH_LOCKED) == 0);
1047 	flags &= ~RTF_RNH_LOCKED;
1048 	if (needlock)
1049 		RADIX_NODE_HEAD_LOCK(rnh);
1050 	else
1051 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1052 	/*
1053 	 * If we are adding a host route then we don't want to put
1054 	 * a netmask in the tree, nor do we want to clone it.
1055 	 */
1056 	if (flags & RTF_HOST)
1057 		netmask = NULL;
1058 
1059 	switch (req) {
1060 	case RTM_DELETE:
1061 #ifdef RADIX_MPATH
1062 		if (rn_mpath_capable(rnh)) {
1063 			error = rn_mpath_update(req, info, rnh, ret_nrt);
1064 			/*
1065 			 * "bad" holds true for the success case
1066 			 * as well
1067 			 */
1068 			if (error != ENOENT)
1069 				goto bad;
1070 			error = 0;
1071 		}
1072 #endif
1073 		/*
1074 		 * Remove the item from the tree and return it.
1075 		 * Complain if it is not there and do no more processing.
1076 		 */
1077 		rn = rnh->rnh_deladdr(dst, netmask, rnh);
1078 		if (rn == NULL)
1079 			senderr(ESRCH);
1080 		if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1081 			panic ("rtrequest delete");
1082 		rt = RNTORT(rn);
1083 		RT_LOCK(rt);
1084 		RT_ADDREF(rt);
1085 		rt->rt_flags &= ~RTF_UP;
1086 
1087 		/*
1088 		 * give the protocol a chance to keep things in sync.
1089 		 */
1090 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1091 			ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1092 
1093 		/*
1094 		 * One more rtentry floating around that is not
1095 		 * linked to the routing table. rttrash will be decremented
1096 		 * when RTFREE(rt) is eventually called.
1097 		 */
1098 		V_rttrash++;
1099 
1100 		/*
1101 		 * If the caller wants it, then it can have it,
1102 		 * but it's up to it to free the rtentry as we won't be
1103 		 * doing it.
1104 		 */
1105 		if (ret_nrt) {
1106 			*ret_nrt = rt;
1107 			RT_UNLOCK(rt);
1108 		} else
1109 			RTFREE_LOCKED(rt);
1110 		break;
1111 	case RTM_RESOLVE:
1112 		/*
1113 		 * resolve was only used for route cloning
1114 		 * here for compat
1115 		 */
1116 		break;
1117 	case RTM_ADD:
1118 		if ((flags & RTF_GATEWAY) && !gateway)
1119 			senderr(EINVAL);
1120 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1121 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1122 			senderr(EINVAL);
1123 
1124 		if (info->rti_ifa == NULL) {
1125 			error = rt_getifa_fib(info, fibnum);
1126 			if (error)
1127 				senderr(error);
1128 		} else
1129 			ifa_ref(info->rti_ifa);
1130 		ifa = info->rti_ifa;
1131 		rt = uma_zalloc(V_rtzone, M_NOWAIT | M_ZERO);
1132 		if (rt == NULL) {
1133 			if (ifa != NULL)
1134 				ifa_free(ifa);
1135 			senderr(ENOBUFS);
1136 		}
1137 		RT_LOCK_INIT(rt);
1138 		rt->rt_flags = RTF_UP | flags;
1139 		rt->rt_fibnum = fibnum;
1140 		/*
1141 		 * Add the gateway. Possibly re-malloc-ing the storage for it
1142 		 *
1143 		 */
1144 		RT_LOCK(rt);
1145 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1146 			RT_LOCK_DESTROY(rt);
1147 			if (ifa != NULL)
1148 				ifa_free(ifa);
1149 			uma_zfree(V_rtzone, rt);
1150 			senderr(error);
1151 		}
1152 
1153 		/*
1154 		 * point to the (possibly newly malloc'd) dest address.
1155 		 */
1156 		ndst = (struct sockaddr *)rt_key(rt);
1157 
1158 		/*
1159 		 * make sure it contains the value we want (masked if needed).
1160 		 */
1161 		if (netmask) {
1162 			rt_maskedcopy(dst, ndst, netmask);
1163 		} else
1164 			bcopy(dst, ndst, dst->sa_len);
1165 
1166 		/*
1167 		 * We use the ifa reference returned by rt_getifa_fib().
1168 		 * This moved from below so that rnh->rnh_addaddr() can
1169 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1170 		 */
1171 		rt->rt_ifa = ifa;
1172 		rt->rt_ifp = ifa->ifa_ifp;
1173 		rt->rt_rmx.rmx_weight = 1;
1174 
1175 #ifdef RADIX_MPATH
1176 		/* do not permit exactly the same dst/mask/gw pair */
1177 		if (rn_mpath_capable(rnh) &&
1178 			rt_mpath_conflict(rnh, rt, netmask)) {
1179 			if (rt->rt_ifa) {
1180 				ifa_free(rt->rt_ifa);
1181 			}
1182 			Free(rt_key(rt));
1183 			RT_LOCK_DESTROY(rt);
1184 			uma_zfree(V_rtzone, rt);
1185 			senderr(EEXIST);
1186 		}
1187 #endif
1188 
1189 #ifdef FLOWTABLE
1190 		rt0 = NULL;
1191 		/* XXX
1192 		 * "flow-table" only support IPv4 at the moment.
1193 		 */
1194 #ifdef INET
1195 		if (dst->sa_family == AF_INET) {
1196 			rn = rnh->rnh_matchaddr(dst, rnh);
1197 			if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1198 				struct sockaddr *mask;
1199 				u_char *m, *n;
1200 				int len;
1201 
1202 				/*
1203 				 * compare mask to see if the new route is
1204 				 * more specific than the existing one
1205 				 */
1206 				rt0 = RNTORT(rn);
1207 				RT_LOCK(rt0);
1208 				RT_ADDREF(rt0);
1209 				RT_UNLOCK(rt0);
1210 				/*
1211 				 * A host route is already present, so
1212 				 * leave the flow-table entries as is.
1213 				 */
1214 				if (rt0->rt_flags & RTF_HOST) {
1215 					RTFREE(rt0);
1216 					rt0 = NULL;
1217 				} else if (!(flags & RTF_HOST) && netmask) {
1218 					mask = rt_mask(rt0);
1219 					len = mask->sa_len;
1220 					m = (u_char *)mask;
1221 					n = (u_char *)netmask;
1222 					while (len-- > 0) {
1223 						if (*n != *m)
1224 							break;
1225 						n++;
1226 						m++;
1227 					}
1228 					if (len == 0 || (*n < *m)) {
1229 						RTFREE(rt0);
1230 						rt0 = NULL;
1231 					}
1232 				}
1233 			}
1234 		}
1235 #endif
1236 #endif
1237 
1238 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1239 		rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1240 		/*
1241 		 * If it still failed to go into the tree,
1242 		 * then un-make it (this should be a function)
1243 		 */
1244 		if (rn == NULL) {
1245 			if (rt->rt_ifa)
1246 				ifa_free(rt->rt_ifa);
1247 			Free(rt_key(rt));
1248 			RT_LOCK_DESTROY(rt);
1249 			uma_zfree(V_rtzone, rt);
1250 #ifdef FLOWTABLE
1251 			if (rt0 != NULL)
1252 				RTFREE(rt0);
1253 #endif
1254 			senderr(EEXIST);
1255 		}
1256 #ifdef FLOWTABLE
1257 		else if (rt0 != NULL) {
1258 #ifdef INET
1259 			flowtable_route_flush(V_ip_ft, rt0);
1260 #endif
1261 			RTFREE(rt0);
1262 		}
1263 #endif
1264 
1265 		/*
1266 		 * If this protocol has something to add to this then
1267 		 * allow it to do that as well.
1268 		 */
1269 		if (ifa->ifa_rtrequest)
1270 			ifa->ifa_rtrequest(req, rt, info);
1271 
1272 		/*
1273 		 * actually return a resultant rtentry and
1274 		 * give the caller a single reference.
1275 		 */
1276 		if (ret_nrt) {
1277 			*ret_nrt = rt;
1278 			RT_ADDREF(rt);
1279 		}
1280 		RT_UNLOCK(rt);
1281 		break;
1282 	default:
1283 		error = EOPNOTSUPP;
1284 	}
1285 bad:
1286 	if (needlock)
1287 		RADIX_NODE_HEAD_UNLOCK(rnh);
1288 	return (error);
1289 #undef senderr
1290 }
1291 
1292 #undef dst
1293 #undef gateway
1294 #undef netmask
1295 #undef ifaaddr
1296 #undef ifpaddr
1297 #undef flags
1298 
1299 int
1300 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1301 {
1302 	/* XXX dst may be overwritten, can we move this to below */
1303 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1304 #ifdef INVARIANTS
1305 	struct radix_node_head *rnh;
1306 
1307 	rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1308 #endif
1309 
1310 	RT_LOCK_ASSERT(rt);
1311 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1312 
1313 	/*
1314 	 * Prepare to store the gateway in rt->rt_gateway.
1315 	 * Both dst and gateway are stored one after the other in the same
1316 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1317 	 * rt_gateway already points to the right place.
1318 	 * Otherwise, malloc a new block and update the 'dst' address.
1319 	 */
1320 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1321 		caddr_t new;
1322 
1323 		R_Malloc(new, caddr_t, dlen + glen);
1324 		if (new == NULL)
1325 			return ENOBUFS;
1326 		/*
1327 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1328 		 * rt_setgate() can be called to initialize a newly
1329 		 * allocated route entry, in which case rt_key(rt) == NULL
1330 		 * (and also rt->rt_gateway == NULL).
1331 		 * Free()/free() handle a NULL argument just fine.
1332 		 */
1333 		bcopy(dst, new, dlen);
1334 		Free(rt_key(rt));	/* free old block, if any */
1335 		rt_key(rt) = (struct sockaddr *)new;
1336 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1337 	}
1338 
1339 	/*
1340 	 * Copy the new gateway value into the memory chunk.
1341 	 */
1342 	bcopy(gate, rt->rt_gateway, glen);
1343 
1344 	return (0);
1345 }
1346 
1347 void
1348 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1349 {
1350 	register u_char *cp1 = (u_char *)src;
1351 	register u_char *cp2 = (u_char *)dst;
1352 	register u_char *cp3 = (u_char *)netmask;
1353 	u_char *cplim = cp2 + *cp3;
1354 	u_char *cplim2 = cp2 + *cp1;
1355 
1356 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1357 	cp3 += 2;
1358 	if (cplim > cplim2)
1359 		cplim = cplim2;
1360 	while (cp2 < cplim)
1361 		*cp2++ = *cp1++ & *cp3++;
1362 	if (cp2 < cplim2)
1363 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1364 }
1365 
1366 /*
1367  * Set up a routing table entry, normally
1368  * for an interface.
1369  */
1370 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1371 static inline  int
1372 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1373 {
1374 	struct sockaddr *dst;
1375 	struct sockaddr *netmask;
1376 	struct rtentry *rt = NULL;
1377 	struct rt_addrinfo info;
1378 	int error = 0;
1379 	int startfib, endfib;
1380 	char tempbuf[_SOCKADDR_TMPSIZE];
1381 	int didwork = 0;
1382 	int a_failure = 0;
1383 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1384 
1385 	if (flags & RTF_HOST) {
1386 		dst = ifa->ifa_dstaddr;
1387 		netmask = NULL;
1388 	} else {
1389 		dst = ifa->ifa_addr;
1390 		netmask = ifa->ifa_netmask;
1391 	}
1392 	if ( dst->sa_family != AF_INET)
1393 		fibnum = 0;
1394 	if (fibnum == -1) {
1395 		if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1396 			startfib = endfib = curthread->td_proc->p_fibnum;
1397 		} else {
1398 			startfib = 0;
1399 			endfib = rt_numfibs - 1;
1400 		}
1401 	} else {
1402 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1403 		startfib = fibnum;
1404 		endfib = fibnum;
1405 	}
1406 	if (dst->sa_len == 0)
1407 		return(EINVAL);
1408 
1409 	/*
1410 	 * If it's a delete, check that if it exists,
1411 	 * it's on the correct interface or we might scrub
1412 	 * a route to another ifa which would
1413 	 * be confusing at best and possibly worse.
1414 	 */
1415 	if (cmd == RTM_DELETE) {
1416 		/*
1417 		 * It's a delete, so it should already exist..
1418 		 * If it's a net, mask off the host bits
1419 		 * (Assuming we have a mask)
1420 		 * XXX this is kinda inet specific..
1421 		 */
1422 		if (netmask != NULL) {
1423 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1424 			dst = (struct sockaddr *)tempbuf;
1425 		}
1426 	}
1427 	/*
1428 	 * Now go through all the requested tables (fibs) and do the
1429 	 * requested action. Realistically, this will either be fib 0
1430 	 * for protocols that don't do multiple tables or all the
1431 	 * tables for those that do. XXX For this version only AF_INET.
1432 	 * When that changes code should be refactored to protocol
1433 	 * independent parts and protocol dependent parts.
1434 	 */
1435 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1436 		if (cmd == RTM_DELETE) {
1437 			struct radix_node_head *rnh;
1438 			struct radix_node *rn;
1439 			/*
1440 			 * Look up an rtentry that is in the routing tree and
1441 			 * contains the correct info.
1442 			 */
1443 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1444 			if (rnh == NULL)
1445 				/* this table doesn't exist but others might */
1446 				continue;
1447 			RADIX_NODE_HEAD_LOCK(rnh);
1448 #ifdef RADIX_MPATH
1449 			if (rn_mpath_capable(rnh)) {
1450 
1451 				rn = rnh->rnh_matchaddr(dst, rnh);
1452 				if (rn == NULL)
1453 					error = ESRCH;
1454 				else {
1455 					rt = RNTORT(rn);
1456 					/*
1457 					 * for interface route the
1458 					 * rt->rt_gateway is sockaddr_intf
1459 					 * for cloning ARP entries, so
1460 					 * rt_mpath_matchgate must use the
1461 					 * interface address
1462 					 */
1463 					rt = rt_mpath_matchgate(rt,
1464 					    ifa->ifa_addr);
1465 					if (!rt)
1466 						error = ESRCH;
1467 				}
1468 			}
1469 			else
1470 #endif
1471 			rn = rnh->rnh_lookup(dst, netmask, rnh);
1472 			error = (rn == NULL ||
1473 			    (rn->rn_flags & RNF_ROOT) ||
1474 			    RNTORT(rn)->rt_ifa != ifa ||
1475 			    !sa_equal((struct sockaddr *)rn->rn_key, dst));
1476 			RADIX_NODE_HEAD_UNLOCK(rnh);
1477 			if (error) {
1478 				/* this is only an error if bad on ALL tables */
1479 				continue;
1480 			}
1481 		}
1482 		/*
1483 		 * Do the actual request
1484 		 */
1485 		bzero((caddr_t)&info, sizeof(info));
1486 		info.rti_ifa = ifa;
1487 		info.rti_flags = flags | ifa->ifa_flags;
1488 		info.rti_info[RTAX_DST] = dst;
1489 		/*
1490 		 * doing this for compatibility reasons
1491 		 */
1492 		if (cmd == RTM_ADD)
1493 			info.rti_info[RTAX_GATEWAY] =
1494 			    (struct sockaddr *)&null_sdl;
1495 		else
1496 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1497 		info.rti_info[RTAX_NETMASK] = netmask;
1498 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1499 		if (error == 0 && rt != NULL) {
1500 			/*
1501 			 * notify any listening routing agents of the change
1502 			 */
1503 			RT_LOCK(rt);
1504 #ifdef RADIX_MPATH
1505 			/*
1506 			 * in case address alias finds the first address
1507 			 * e.g. ifconfig bge0 192.103.54.246/24
1508 			 * e.g. ifconfig bge0 192.103.54.247/24
1509 			 * the address set in the route is 192.103.54.246
1510 			 * so we need to replace it with 192.103.54.247
1511 			 */
1512 			if (memcmp(rt->rt_ifa->ifa_addr,
1513 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1514 				ifa_free(rt->rt_ifa);
1515 				ifa_ref(ifa);
1516 				rt->rt_ifp = ifa->ifa_ifp;
1517 				rt->rt_ifa = ifa;
1518 			}
1519 #endif
1520 			/*
1521 			 * doing this for compatibility reasons
1522 			 */
1523 			if (cmd == RTM_ADD) {
1524 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1525 				rt->rt_ifp->if_type;
1526 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1527 				rt->rt_ifp->if_index;
1528 			}
1529 			RT_ADDREF(rt);
1530 			RT_UNLOCK(rt);
1531 			rt_newaddrmsg(cmd, ifa, error, rt);
1532 			RT_LOCK(rt);
1533 			RT_REMREF(rt);
1534 			if (cmd == RTM_DELETE) {
1535 				/*
1536 				 * If we are deleting, and we found an entry,
1537 				 * then it's been removed from the tree..
1538 				 * now throw it away.
1539 				 */
1540 				RTFREE_LOCKED(rt);
1541 			} else {
1542 				if (cmd == RTM_ADD) {
1543 					/*
1544 					 * We just wanted to add it..
1545 					 * we don't actually need a reference.
1546 					 */
1547 					RT_REMREF(rt);
1548 				}
1549 				RT_UNLOCK(rt);
1550 			}
1551 			didwork = 1;
1552 		}
1553 		if (error)
1554 			a_failure = error;
1555 	}
1556 	if (cmd == RTM_DELETE) {
1557 		if (didwork) {
1558 			error = 0;
1559 		} else {
1560 			/* we only give an error if it wasn't in any table */
1561 			error = ((flags & RTF_HOST) ?
1562 			    EHOSTUNREACH : ENETUNREACH);
1563 		}
1564 	} else {
1565 		if (a_failure) {
1566 			/* return an error if any of them failed */
1567 			error = a_failure;
1568 		}
1569 	}
1570 	return (error);
1571 }
1572 
1573 /* special one for inet internal use. may not use. */
1574 int
1575 rtinit_fib(struct ifaddr *ifa, int cmd, int flags)
1576 {
1577 	return (rtinit1(ifa, cmd, flags, -1));
1578 }
1579 
1580 /*
1581  * Set up a routing table entry, normally
1582  * for an interface.
1583  */
1584 int
1585 rtinit(struct ifaddr *ifa, int cmd, int flags)
1586 {
1587 	struct sockaddr *dst;
1588 	int fib = 0;
1589 
1590 	if (flags & RTF_HOST) {
1591 		dst = ifa->ifa_dstaddr;
1592 	} else {
1593 		dst = ifa->ifa_addr;
1594 	}
1595 
1596 	if (dst->sa_family == AF_INET)
1597 		fib = -1;
1598 	return (rtinit1(ifa, cmd, flags, fib));
1599 }
1600