xref: /dragonfly/sys/net/route.c (revision 9bb2a92d)
1 /*
2  * Copyright (c) 1980, 1986, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)route.c	8.3 (Berkeley) 1/9/95
34  * $FreeBSD: src/sys/net/route.c,v 1.59.2.10 2003/01/17 08:04:00 ru Exp $
35  * $DragonFly: src/sys/net/route.c,v 1.6 2003/08/26 20:49:47 rob Exp $
36  */
37 
38 #include "opt_inet.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/socket.h>
45 #include <sys/domain.h>
46 #include <sys/kernel.h>
47 
48 #include <net/if.h>
49 #include <net/route.h>
50 
51 #include <netinet/in.h>
52 #include <net/ip_mroute/ip_mroute.h>
53 
54 #define	SA(p) ((struct sockaddr *)(p))
55 
56 struct route_cb route_cb;
57 static struct rtstat rtstat;
58 struct radix_node_head *rt_tables[AF_MAX+1];
59 
60 static int	rttrash;		/* routes not in table but not freed */
61 
62 static void rt_maskedcopy (struct sockaddr *,
63 	    struct sockaddr *, struct sockaddr *);
64 static void rtable_init (void **);
65 
66 static void
67 rtable_init(table)
68 	void **table;
69 {
70 	struct domain *dom;
71 	for (dom = domains; dom; dom = dom->dom_next)
72 		if (dom->dom_rtattach)
73 			dom->dom_rtattach(&table[dom->dom_family],
74 			    dom->dom_rtoffset);
75 }
76 
77 void
78 route_init()
79 {
80 	rn_init();	/* initialize all zeroes, all ones, mask table */
81 	rtable_init((void **)rt_tables);
82 }
83 
84 /*
85  * Packet routing routines.
86  */
87 void
88 rtalloc(ro)
89 	struct route *ro;
90 {
91 	rtalloc_ign(ro, 0UL);
92 }
93 
94 void
95 rtalloc_ign(ro, ignore)
96 	struct route *ro;
97 	u_long ignore;
98 {
99 	struct rtentry *rt;
100 	int s;
101 
102 	if ((rt = ro->ro_rt) != NULL) {
103 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
104 			return;
105 		/* XXX - We are probably always at splnet here already. */
106 		s = splnet();
107 		RTFREE(rt);
108 		ro->ro_rt = NULL;
109 		splx(s);
110 	}
111 	ro->ro_rt = rtalloc1(&ro->ro_dst, 1, ignore);
112 }
113 
114 /*
115  * Look up the route that matches the address given
116  * Or, at least try.. Create a cloned route if needed.
117  */
118 struct rtentry *
119 rtalloc1(dst, report, ignflags)
120 	struct sockaddr *dst;
121 	int report;
122 	u_long ignflags;
123 {
124 	struct radix_node_head *rnh = rt_tables[dst->sa_family];
125 	struct rtentry *rt;
126 	struct radix_node *rn;
127 	struct rtentry *newrt = 0;
128 	struct rt_addrinfo info;
129 	u_long nflags;
130 	int  s = splnet(), err = 0, msgtype = RTM_MISS;
131 
132 	/*
133 	 * Look up the address in the table for that Address Family
134 	 */
135 	if (rnh && (rn = rnh->rnh_matchaddr((caddr_t)dst, rnh)) &&
136 	    ((rn->rn_flags & RNF_ROOT) == 0)) {
137 		/*
138 		 * If we find it and it's not the root node, then
139 		 * get a refernce on the rtentry associated.
140 		 */
141 		newrt = rt = (struct rtentry *)rn;
142 		nflags = rt->rt_flags & ~ignflags;
143 		if (report && (nflags & (RTF_CLONING | RTF_PRCLONING))) {
144 			/*
145 			 * We are apparently adding (report = 0 in delete).
146 			 * If it requires that it be cloned, do so.
147 			 * (This implies it wasn't a HOST route.)
148 			 */
149 			err = rtrequest(RTM_RESOLVE, dst, SA(0),
150 					      SA(0), 0, &newrt);
151 			if (err) {
152 				/*
153 				 * If the cloning didn't succeed, maybe
154 				 * what we have will do. Return that.
155 				 */
156 				newrt = rt;
157 				rt->rt_refcnt++;
158 				goto miss;
159 			}
160 			if ((rt = newrt) && (rt->rt_flags & RTF_XRESOLVE)) {
161 				/*
162 				 * If the new route specifies it be
163 				 * externally resolved, then go do that.
164 				 */
165 				msgtype = RTM_RESOLVE;
166 				goto miss;
167 			}
168 			/* Inform listeners of the new route. */
169 			bzero(&info, sizeof(info));
170 			info.rti_info[RTAX_DST] = rt_key(rt);
171 			info.rti_info[RTAX_NETMASK] = rt_mask(rt);
172 			info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
173 			if (rt->rt_ifp != NULL) {
174 				info.rti_info[RTAX_IFP] =
175 				    TAILQ_FIRST(&rt->rt_ifp->if_addrhead)->ifa_addr;
176 				info.rti_info[RTAX_IFA] = rt->rt_ifa->ifa_addr;
177 			}
178 			rt_missmsg(RTM_ADD, &info, rt->rt_flags, 0);
179 		} else
180 			rt->rt_refcnt++;
181 	} else {
182 		/*
183 		 * Either we hit the root or couldn't find any match,
184 		 * Which basically means
185 		 * "caint get there frm here"
186 		 */
187 		rtstat.rts_unreach++;
188 	miss:	if (report) {
189 			/*
190 			 * If required, report the failure to the supervising
191 			 * Authorities.
192 			 * For a delete, this is not an error. (report == 0)
193 			 */
194 			bzero((caddr_t)&info, sizeof(info));
195 			info.rti_info[RTAX_DST] = dst;
196 			rt_missmsg(msgtype, &info, 0, err);
197 		}
198 	}
199 	splx(s);
200 	return (newrt);
201 }
202 
203 /*
204  * Remove a reference count from an rtentry.
205  * If the count gets low enough, take it out of the routing table
206  */
207 void
208 rtfree(rt)
209 	struct rtentry *rt;
210 {
211 	/*
212 	 * find the tree for that address family
213 	 */
214 	struct radix_node_head *rnh =
215 		rt_tables[rt_key(rt)->sa_family];
216 	struct ifaddr *ifa;
217 
218 	if (rt == 0 || rnh == 0)
219 		panic("rtfree");
220 
221 	/*
222 	 * decrement the reference count by one and if it reaches 0,
223 	 * and there is a close function defined, call the close function
224 	 */
225 	rt->rt_refcnt--;
226 	if(rnh->rnh_close && rt->rt_refcnt == 0) {
227 		rnh->rnh_close((struct radix_node *)rt, rnh);
228 	}
229 
230 	/*
231 	 * If we are no longer "up" (and ref == 0)
232 	 * then we can free the resources associated
233 	 * with the route.
234 	 */
235 	if (rt->rt_refcnt <= 0 && (rt->rt_flags & RTF_UP) == 0) {
236 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
237 			panic ("rtfree 2");
238 		/*
239 		 * the rtentry must have been removed from the routing table
240 		 * so it is represented in rttrash.. remove that now.
241 		 */
242 		rttrash--;
243 
244 #ifdef	DIAGNOSTIC
245 		if (rt->rt_refcnt < 0) {
246 			printf("rtfree: %p not freed (neg refs)\n", rt);
247 			return;
248 		}
249 #endif
250 
251 		/*
252 		 * release references on items we hold them on..
253 		 * e.g other routes and ifaddrs.
254 		 */
255 		if((ifa = rt->rt_ifa))
256 			IFAFREE(ifa);
257 		if (rt->rt_parent) {
258 			RTFREE(rt->rt_parent);
259 		}
260 
261 		/*
262 		 * The key is separatly alloc'd so free it (see rt_setgate()).
263 		 * This also frees the gateway, as they are always malloc'd
264 		 * together.
265 		 */
266 		Free(rt_key(rt));
267 
268 		/*
269 		 * and the rtentry itself of course
270 		 */
271 		Free(rt);
272 	}
273 }
274 
275 void
276 ifafree(ifa)
277 	struct ifaddr *ifa;
278 {
279 	if (ifa == NULL)
280 		panic("ifafree");
281 	if (ifa->ifa_refcnt == 0)
282 		free(ifa, M_IFADDR);
283 	else
284 		ifa->ifa_refcnt--;
285 }
286 
287 /*
288  * Force a routing table entry to the specified
289  * destination to go through the given gateway.
290  * Normally called as a result of a routing redirect
291  * message from the network layer.
292  *
293  * N.B.: must be called at splnet
294  *
295  */
296 void
297 rtredirect(dst, gateway, netmask, flags, src, rtp)
298 	struct sockaddr *dst, *gateway, *netmask, *src;
299 	int flags;
300 	struct rtentry **rtp;
301 {
302 	struct rtentry *rt;
303 	int error = 0;
304 	short *stat = 0;
305 	struct rt_addrinfo info;
306 	struct ifaddr *ifa;
307 
308 	/* verify the gateway is directly reachable */
309 	if ((ifa = ifa_ifwithnet(gateway)) == 0) {
310 		error = ENETUNREACH;
311 		goto out;
312 	}
313 	rt = rtalloc1(dst, 0, 0UL);
314 	/*
315 	 * If the redirect isn't from our current router for this dst,
316 	 * it's either old or wrong.  If it redirects us to ourselves,
317 	 * we have a routing loop, perhaps as a result of an interface
318 	 * going down recently.
319 	 */
320 #define	equal(a1, a2) (bcmp((caddr_t)(a1), (caddr_t)(a2), (a1)->sa_len) == 0)
321 	if (!(flags & RTF_DONE) && rt &&
322 	     (!equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
323 		error = EINVAL;
324 	else if (ifa_ifwithaddr(gateway))
325 		error = EHOSTUNREACH;
326 	if (error)
327 		goto done;
328 	/*
329 	 * Create a new entry if we just got back a wildcard entry
330 	 * or the the lookup failed.  This is necessary for hosts
331 	 * which use routing redirects generated by smart gateways
332 	 * to dynamically build the routing tables.
333 	 */
334 	if ((rt == 0) || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
335 		goto create;
336 	/*
337 	 * Don't listen to the redirect if it's
338 	 * for a route to an interface.
339 	 */
340 	if (rt->rt_flags & RTF_GATEWAY) {
341 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
342 			/*
343 			 * Changing from route to net => route to host.
344 			 * Create new route, rather than smashing route to net.
345 			 */
346 		create:
347 			if (rt)
348 				rtfree(rt);
349 			flags |=  RTF_GATEWAY | RTF_DYNAMIC;
350 			bzero((caddr_t)&info, sizeof(info));
351 			info.rti_info[RTAX_DST] = dst;
352 			info.rti_info[RTAX_GATEWAY] = gateway;
353 			info.rti_info[RTAX_NETMASK] = netmask;
354 			info.rti_ifa = ifa;
355 			info.rti_flags = flags;
356 			rt = NULL;
357 			error = rtrequest1(RTM_ADD, &info, &rt);
358 			if (rt != NULL)
359 				flags = rt->rt_flags;
360 			stat = &rtstat.rts_dynamic;
361 		} else {
362 			/*
363 			 * Smash the current notion of the gateway to
364 			 * this destination.  Should check about netmask!!!
365 			 */
366 			rt->rt_flags |= RTF_MODIFIED;
367 			flags |= RTF_MODIFIED;
368 			stat = &rtstat.rts_newgateway;
369 			/*
370 			 * add the key and gateway (in one malloc'd chunk).
371 			 */
372 			rt_setgate(rt, rt_key(rt), gateway);
373 		}
374 	} else
375 		error = EHOSTUNREACH;
376 done:
377 	if (rt) {
378 		if (rtp && !error)
379 			*rtp = rt;
380 		else
381 			rtfree(rt);
382 	}
383 out:
384 	if (error)
385 		rtstat.rts_badredirect++;
386 	else if (stat != NULL)
387 		(*stat)++;
388 	bzero((caddr_t)&info, sizeof(info));
389 	info.rti_info[RTAX_DST] = dst;
390 	info.rti_info[RTAX_GATEWAY] = gateway;
391 	info.rti_info[RTAX_NETMASK] = netmask;
392 	info.rti_info[RTAX_AUTHOR] = src;
393 	rt_missmsg(RTM_REDIRECT, &info, flags, error);
394 }
395 
396 /*
397 * Routing table ioctl interface.
398 */
399 int
400 rtioctl(u_long req, caddr_t data, struct thread *td)
401 {
402 #ifdef INET
403 	/* Multicast goop, grrr... */
404 	return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP;
405 #else /* INET */
406 	return ENXIO;
407 #endif /* INET */
408 }
409 
410 struct ifaddr *
411 ifa_ifwithroute(flags, dst, gateway)
412 	int flags;
413 	struct sockaddr	*dst, *gateway;
414 {
415 	struct ifaddr *ifa;
416 	if ((flags & RTF_GATEWAY) == 0) {
417 		/*
418 		 * If we are adding a route to an interface,
419 		 * and the interface is a pt to pt link
420 		 * we should search for the destination
421 		 * as our clue to the interface.  Otherwise
422 		 * we can use the local address.
423 		 */
424 		ifa = 0;
425 		if (flags & RTF_HOST) {
426 			ifa = ifa_ifwithdstaddr(dst);
427 		}
428 		if (ifa == 0)
429 			ifa = ifa_ifwithaddr(gateway);
430 	} else {
431 		/*
432 		 * If we are adding a route to a remote net
433 		 * or host, the gateway may still be on the
434 		 * other end of a pt to pt link.
435 		 */
436 		ifa = ifa_ifwithdstaddr(gateway);
437 	}
438 	if (ifa == 0)
439 		ifa = ifa_ifwithnet(gateway);
440 	if (ifa == 0) {
441 		struct rtentry *rt = rtalloc1(gateway, 0, 0UL);
442 		if (rt == 0)
443 			return (0);
444 		rt->rt_refcnt--;
445 		if ((ifa = rt->rt_ifa) == 0)
446 			return (0);
447 	}
448 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
449 		struct ifaddr *oifa = ifa;
450 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
451 		if (ifa == 0)
452 			ifa = oifa;
453 	}
454 	return (ifa);
455 }
456 
457 static int rt_fixdelete (struct radix_node *, void *);
458 static int rt_fixchange (struct radix_node *, void *);
459 
460 struct rtfc_arg {
461 	struct rtentry *rt0;
462 	struct radix_node_head *rnh;
463 };
464 
465 /*
466  * Do appropriate manipulations of a routing tree given
467  * all the bits of info needed
468  */
469 int
470 rtrequest(req, dst, gateway, netmask, flags, ret_nrt)
471 	int req, flags;
472 	struct sockaddr *dst, *gateway, *netmask;
473 	struct rtentry **ret_nrt;
474 {
475 	struct rt_addrinfo info;
476 
477 	bzero((caddr_t)&info, sizeof(info));
478 	info.rti_flags = flags;
479 	info.rti_info[RTAX_DST] = dst;
480 	info.rti_info[RTAX_GATEWAY] = gateway;
481 	info.rti_info[RTAX_NETMASK] = netmask;
482 	return rtrequest1(req, &info, ret_nrt);
483 }
484 
485 /*
486  * These (questionable) definitions of apparent local variables apply
487  * to the next two functions.  XXXXXX!!!
488  */
489 #define	dst	info->rti_info[RTAX_DST]
490 #define	gateway	info->rti_info[RTAX_GATEWAY]
491 #define	netmask	info->rti_info[RTAX_NETMASK]
492 #define	ifaaddr	info->rti_info[RTAX_IFA]
493 #define	ifpaddr	info->rti_info[RTAX_IFP]
494 #define	flags	info->rti_flags
495 
496 int
497 rt_getifa(info)
498 	struct rt_addrinfo *info;
499 {
500 	struct ifaddr *ifa;
501 	int error = 0;
502 
503 	/*
504 	 * ifp may be specified by sockaddr_dl
505 	 * when protocol address is ambiguous.
506 	 */
507 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
508 	    ifpaddr->sa_family == AF_LINK &&
509 	    (ifa = ifa_ifwithnet(ifpaddr)) != NULL)
510 		info->rti_ifp = ifa->ifa_ifp;
511 	if (info->rti_ifa == NULL && ifaaddr != NULL)
512 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
513 	if (info->rti_ifa == NULL) {
514 		struct sockaddr *sa;
515 
516 		sa = ifaaddr != NULL ? ifaaddr :
517 		    (gateway != NULL ? gateway : dst);
518 		if (sa != NULL && info->rti_ifp != NULL)
519 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
520 		else if (dst != NULL && gateway != NULL)
521 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway);
522 		else if (sa != NULL)
523 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa);
524 	}
525 	if ((ifa = info->rti_ifa) != NULL) {
526 		if (info->rti_ifp == NULL)
527 			info->rti_ifp = ifa->ifa_ifp;
528 	} else
529 		error = ENETUNREACH;
530 	return (error);
531 }
532 
533 int
534 rtrequest1(req, info, ret_nrt)
535 	int req;
536 	struct rt_addrinfo *info;
537 	struct rtentry **ret_nrt;
538 {
539 	int s = splnet(); int error = 0;
540 	struct rtentry *rt;
541 	struct radix_node *rn;
542 	struct radix_node_head *rnh;
543 	struct ifaddr *ifa;
544 	struct sockaddr *ndst;
545 #define senderr(x) { error = x ; goto bad; }
546 
547 	/*
548 	 * Find the correct routing tree to use for this Address Family
549 	 */
550 	if ((rnh = rt_tables[dst->sa_family]) == 0)
551 		senderr(EAFNOSUPPORT);
552 	/*
553 	 * If we are adding a host route then we don't want to put
554 	 * a netmask in the tree, nor do we want to clone it.
555 	 */
556 	if (flags & RTF_HOST) {
557 		netmask = 0;
558 		flags &= ~(RTF_CLONING | RTF_PRCLONING);
559 	}
560 	switch (req) {
561 	case RTM_DELETE:
562 		/*
563 		 * Remove the item from the tree and return it.
564 		 * Complain if it is not there and do no more processing.
565 		 */
566 		if ((rn = rnh->rnh_deladdr(dst, netmask, rnh)) == 0)
567 			senderr(ESRCH);
568 		if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
569 			panic ("rtrequest delete");
570 		rt = (struct rtentry *)rn;
571 
572 		/*
573 		 * Now search what's left of the subtree for any cloned
574 		 * routes which might have been formed from this node.
575 		 */
576 		if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) &&
577 		    rt_mask(rt)) {
578 			rnh->rnh_walktree_from(rnh, dst, rt_mask(rt),
579 					       rt_fixdelete, rt);
580 		}
581 
582 		/*
583 		 * Remove any external references we may have.
584 		 * This might result in another rtentry being freed if
585 		 * we held its last reference.
586 		 */
587 		if (rt->rt_gwroute) {
588 			rt = rt->rt_gwroute;
589 			RTFREE(rt);
590 			(rt = (struct rtentry *)rn)->rt_gwroute = 0;
591 		}
592 
593 		/*
594 		 * NB: RTF_UP must be set during the search above,
595 		 * because we might delete the last ref, causing
596 		 * rt to get freed prematurely.
597 		 *  eh? then why not just add a reference?
598 		 * I'm not sure how RTF_UP helps matters. (JRE)
599 		 */
600 		rt->rt_flags &= ~RTF_UP;
601 
602 		/*
603 		 * give the protocol a chance to keep things in sync.
604 		 */
605 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
606 			ifa->ifa_rtrequest(RTM_DELETE, rt, info);
607 
608 		/*
609 		 * one more rtentry floating around that is not
610 		 * linked to the routing table.
611 		 */
612 		rttrash++;
613 
614 		/*
615 		 * If the caller wants it, then it can have it,
616 		 * but it's up to it to free the rtentry as we won't be
617 		 * doing it.
618 		 */
619 		if (ret_nrt)
620 			*ret_nrt = rt;
621 		else if (rt->rt_refcnt <= 0) {
622 			rt->rt_refcnt++; /* make a 1->0 transition */
623 			rtfree(rt);
624 		}
625 		break;
626 
627 	case RTM_RESOLVE:
628 		if (ret_nrt == 0 || (rt = *ret_nrt) == 0)
629 			senderr(EINVAL);
630 		ifa = rt->rt_ifa;
631 		flags = rt->rt_flags &
632 		    ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC);
633 		flags |= RTF_WASCLONED;
634 		gateway = rt->rt_gateway;
635 		if ((netmask = rt->rt_genmask) == 0)
636 			flags |= RTF_HOST;
637 		goto makeroute;
638 
639 	case RTM_ADD:
640 		if ((flags & RTF_GATEWAY) && !gateway)
641 			panic("rtrequest: GATEWAY but no gateway");
642 
643 		if (info->rti_ifa == NULL && (error = rt_getifa(info)))
644 			senderr(error);
645 		ifa = info->rti_ifa;
646 
647 	makeroute:
648 		R_Malloc(rt, struct rtentry *, sizeof(*rt));
649 		if (rt == 0)
650 			senderr(ENOBUFS);
651 		Bzero(rt, sizeof(*rt));
652 		rt->rt_flags = RTF_UP | flags;
653 		/*
654 		 * Add the gateway. Possibly re-malloc-ing the storage for it
655 		 * also add the rt_gwroute if possible.
656 		 */
657 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
658 			Free(rt);
659 			senderr(error);
660 		}
661 
662 		/*
663 		 * point to the (possibly newly malloc'd) dest address.
664 		 */
665 		ndst = rt_key(rt);
666 
667 		/*
668 		 * make sure it contains the value we want (masked if needed).
669 		 */
670 		if (netmask) {
671 			rt_maskedcopy(dst, ndst, netmask);
672 		} else
673 			Bcopy(dst, ndst, dst->sa_len);
674 
675 		/*
676 		 * Note that we now have a reference to the ifa.
677 		 * This moved from below so that rnh->rnh_addaddr() can
678 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
679 		 */
680 		ifa->ifa_refcnt++;
681 		rt->rt_ifa = ifa;
682 		rt->rt_ifp = ifa->ifa_ifp;
683 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
684 
685 		rn = rnh->rnh_addaddr((caddr_t)ndst, (caddr_t)netmask,
686 					rnh, rt->rt_nodes);
687 		if (rn == 0) {
688 			struct rtentry *rt2;
689 			/*
690 			 * Uh-oh, we already have one of these in the tree.
691 			 * We do a special hack: if the route that's already
692 			 * there was generated by the protocol-cloning
693 			 * mechanism, then we just blow it away and retry
694 			 * the insertion of the new one.
695 			 */
696 			rt2 = rtalloc1(dst, 0, RTF_PRCLONING);
697 			if (rt2 && rt2->rt_parent) {
698 				rtrequest(RTM_DELETE,
699 					  (struct sockaddr *)rt_key(rt2),
700 					  rt2->rt_gateway,
701 					  rt_mask(rt2), rt2->rt_flags, 0);
702 				RTFREE(rt2);
703 				rn = rnh->rnh_addaddr((caddr_t)ndst,
704 						      (caddr_t)netmask,
705 						      rnh, rt->rt_nodes);
706 			} else if (rt2) {
707 				/* undo the extra ref we got */
708 				RTFREE(rt2);
709 			}
710 		}
711 
712 		/*
713 		 * If it still failed to go into the tree,
714 		 * then un-make it (this should be a function)
715 		 */
716 		if (rn == 0) {
717 			if (rt->rt_gwroute)
718 				rtfree(rt->rt_gwroute);
719 			if (rt->rt_ifa) {
720 				IFAFREE(rt->rt_ifa);
721 			}
722 			Free(rt_key(rt));
723 			Free(rt);
724 			senderr(EEXIST);
725 		}
726 
727 		rt->rt_parent = 0;
728 
729 		/*
730 		 * If we got here from RESOLVE, then we are cloning
731 		 * so clone the rest, and note that we
732 		 * are a clone (and increment the parent's references)
733 		 */
734 		if (req == RTM_RESOLVE) {
735 			rt->rt_rmx = (*ret_nrt)->rt_rmx; /* copy metrics */
736 			rt->rt_rmx.rmx_pksent = 0; /* reset packet counter */
737 			if ((*ret_nrt)->rt_flags & (RTF_CLONING | RTF_PRCLONING)) {
738 				rt->rt_parent = (*ret_nrt);
739 				(*ret_nrt)->rt_refcnt++;
740 			}
741 		}
742 
743 		/*
744 		 * if this protocol has something to add to this then
745 		 * allow it to do that as well.
746 		 */
747 		if (ifa->ifa_rtrequest)
748 			ifa->ifa_rtrequest(req, rt, info);
749 
750 		/*
751 		 * We repeat the same procedure from rt_setgate() here because
752 		 * it doesn't fire when we call it there because the node
753 		 * hasn't been added to the tree yet.
754 		 */
755 		if (req == RTM_ADD &&
756 		    !(rt->rt_flags & RTF_HOST) && rt_mask(rt) != 0) {
757 			struct rtfc_arg arg;
758 			arg.rnh = rnh;
759 			arg.rt0 = rt;
760 			rnh->rnh_walktree_from(rnh, rt_key(rt), rt_mask(rt),
761 					       rt_fixchange, &arg);
762 		}
763 
764 		/*
765 		 * actually return a resultant rtentry and
766 		 * give the caller a single reference.
767 		 */
768 		if (ret_nrt) {
769 			*ret_nrt = rt;
770 			rt->rt_refcnt++;
771 		}
772 		break;
773 	default:
774 		error = EOPNOTSUPP;
775 	}
776 bad:
777 	splx(s);
778 	return (error);
779 #undef dst
780 #undef gateway
781 #undef netmask
782 #undef ifaaddr
783 #undef ifpaddr
784 #undef flags
785 }
786 
787 /*
788  * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family''
789  * (i.e., the routes related to it by the operation of cloning).  This
790  * routine is iterated over all potential former-child-routes by way of
791  * rnh->rnh_walktree_from() above, and those that actually are children of
792  * the late parent (passed in as VP here) are themselves deleted.
793  */
794 static int
795 rt_fixdelete(rn, vp)
796 	struct radix_node *rn;
797 	void *vp;
798 {
799 	struct rtentry *rt = (struct rtentry *)rn;
800 	struct rtentry *rt0 = vp;
801 
802 	if (rt->rt_parent == rt0 &&
803 	    !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
804 		return rtrequest(RTM_DELETE, rt_key(rt),
805 				 (struct sockaddr *)0, rt_mask(rt),
806 				 rt->rt_flags, (struct rtentry **)0);
807 	}
808 	return 0;
809 }
810 
811 /*
812  * This routine is called from rt_setgate() to do the analogous thing for
813  * adds and changes.  There is the added complication in this case of a
814  * middle insert; i.e., insertion of a new network route between an older
815  * network route and (cloned) host routes.  For this reason, a simple check
816  * of rt->rt_parent is insufficient; each candidate route must be tested
817  * against the (mask, value) of the new route (passed as before in vp)
818  * to see if the new route matches it.
819  *
820  * XXX - it may be possible to do fixdelete() for changes and reserve this
821  * routine just for adds.  I'm not sure why I thought it was necessary to do
822  * changes this way.
823  */
824 #ifdef DEBUG
825 static int rtfcdebug = 0;
826 #endif
827 
828 static int
829 rt_fixchange(rn, vp)
830 	struct radix_node *rn;
831 	void *vp;
832 {
833 	struct rtentry *rt = (struct rtentry *)rn;
834 	struct rtfc_arg *ap = vp;
835 	struct rtentry *rt0 = ap->rt0;
836 	struct radix_node_head *rnh = ap->rnh;
837 	u_char *xk1, *xm1, *xk2, *xmp;
838 	int i, len, mlen;
839 
840 #ifdef DEBUG
841 	if (rtfcdebug)
842 		printf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0);
843 #endif
844 
845 	if (!rt->rt_parent ||
846 	    (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
847 #ifdef DEBUG
848 		if(rtfcdebug) printf("no parent, pinned or cloning\n");
849 #endif
850 		return 0;
851 	}
852 
853 	if (rt->rt_parent == rt0) {
854 #ifdef DEBUG
855 		if(rtfcdebug) printf("parent match\n");
856 #endif
857 		return rtrequest(RTM_DELETE, rt_key(rt),
858 				 (struct sockaddr *)0, rt_mask(rt),
859 				 rt->rt_flags, (struct rtentry **)0);
860 	}
861 
862 	/*
863 	 * There probably is a function somewhere which does this...
864 	 * if not, there should be.
865 	 */
866 	len = imin(((struct sockaddr *)rt_key(rt0))->sa_len,
867 		   ((struct sockaddr *)rt_key(rt))->sa_len);
868 
869 	xk1 = (u_char *)rt_key(rt0);
870 	xm1 = (u_char *)rt_mask(rt0);
871 	xk2 = (u_char *)rt_key(rt);
872 
873 	/* avoid applying a less specific route */
874 	xmp = (u_char *)rt_mask(rt->rt_parent);
875 	mlen = ((struct sockaddr *)rt_key(rt->rt_parent))->sa_len;
876 	if (mlen > ((struct sockaddr *)rt_key(rt0))->sa_len) {
877 #ifdef DEBUG
878 		if (rtfcdebug)
879 			printf("rt_fixchange: inserting a less "
880 			       "specific route\n");
881 #endif
882 		return 0;
883 	}
884 	for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) {
885 		if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) {
886 #ifdef DEBUG
887 			if (rtfcdebug)
888 				printf("rt_fixchange: inserting a less "
889 				       "specific route\n");
890 #endif
891 			return 0;
892 		}
893 	}
894 
895 	for (i = rnh->rnh_treetop->rn_offset; i < len; i++) {
896 		if ((xk2[i] & xm1[i]) != xk1[i]) {
897 #ifdef DEBUG
898 			if(rtfcdebug) printf("no match\n");
899 #endif
900 			return 0;
901 		}
902 	}
903 
904 	/*
905 	 * OK, this node is a clone, and matches the node currently being
906 	 * changed/added under the node's mask.  So, get rid of it.
907 	 */
908 #ifdef DEBUG
909 	if(rtfcdebug) printf("deleting\n");
910 #endif
911 	return rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
912 			 rt_mask(rt), rt->rt_flags, (struct rtentry **)0);
913 }
914 
915 #define ROUNDUP(a) (a>0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))
916 
917 int
918 rt_setgate(rt0, dst, gate)
919 	struct rtentry *rt0;
920 	struct sockaddr *dst, *gate;
921 {
922 	caddr_t new, old;
923 	int dlen = ROUNDUP(dst->sa_len), glen = ROUNDUP(gate->sa_len);
924 	struct rtentry *rt = rt0;
925 	struct radix_node_head *rnh = rt_tables[dst->sa_family];
926 
927 	/*
928 	 * A host route with the destination equal to the gateway
929 	 * will interfere with keeping LLINFO in the routing
930 	 * table, so disallow it.
931 	 */
932 	if (((rt0->rt_flags & (RTF_HOST|RTF_GATEWAY|RTF_LLINFO)) ==
933 					(RTF_HOST|RTF_GATEWAY)) &&
934 	    (dst->sa_len == gate->sa_len) &&
935 	    (bcmp(dst, gate, dst->sa_len) == 0)) {
936 		/*
937 		 * The route might already exist if this is an RTM_CHANGE
938 		 * or a routing redirect, so try to delete it.
939 		 */
940 		if (rt_key(rt0))
941 			rtrequest(RTM_DELETE, (struct sockaddr *)rt_key(rt0),
942 			    rt0->rt_gateway, rt_mask(rt0), rt0->rt_flags, 0);
943 		return EADDRNOTAVAIL;
944 	}
945 
946 	/*
947 	 * Both dst and gateway are stored in the same malloc'd chunk
948 	 * (If I ever get my hands on....)
949 	 * if we need to malloc a new chunk, then keep the old one around
950 	 * till we don't need it any more.
951 	 */
952 	if (rt->rt_gateway == 0 || glen > ROUNDUP(rt->rt_gateway->sa_len)) {
953 		old = (caddr_t)rt_key(rt);
954 		R_Malloc(new, caddr_t, dlen + glen);
955 		if (new == 0)
956 			return ENOBUFS;
957 		rt->rt_nodes->rn_key = new;
958 	} else {
959 		/*
960 		 * otherwise just overwrite the old one
961 		 */
962 		new = rt->rt_nodes->rn_key;
963 		old = 0;
964 	}
965 
966 	/*
967 	 * copy the new gateway value into the memory chunk
968 	 */
969 	Bcopy(gate, (rt->rt_gateway = (struct sockaddr *)(new + dlen)), glen);
970 
971 	/*
972 	 * if we are replacing the chunk (or it's new) we need to
973 	 * replace the dst as well
974 	 */
975 	if (old) {
976 		Bcopy(dst, new, dlen);
977 		Free(old);
978 	}
979 
980 	/*
981 	 * If there is already a gwroute, it's now almost definitly wrong
982 	 * so drop it.
983 	 */
984 	if (rt->rt_gwroute != NULL) {
985 		RTFREE(rt->rt_gwroute);
986 		rt->rt_gwroute = NULL;
987 	}
988 	/*
989 	 * Cloning loop avoidance:
990 	 * In the presence of protocol-cloning and bad configuration,
991 	 * it is possible to get stuck in bottomless mutual recursion
992 	 * (rtrequest rt_setgate rtalloc1).  We avoid this by not allowing
993 	 * protocol-cloning to operate for gateways (which is probably the
994 	 * correct choice anyway), and avoid the resulting reference loops
995 	 * by disallowing any route to run through itself as a gateway.
996 	 * This is obviously mandatory when we get rt->rt_output().
997 	 */
998 	if (rt->rt_flags & RTF_GATEWAY) {
999 		rt->rt_gwroute = rtalloc1(gate, 1, RTF_PRCLONING);
1000 		if (rt->rt_gwroute == rt) {
1001 			RTFREE(rt->rt_gwroute);
1002 			rt->rt_gwroute = 0;
1003 			return EDQUOT; /* failure */
1004 		}
1005 	}
1006 
1007 	/*
1008 	 * This isn't going to do anything useful for host routes, so
1009 	 * don't bother.  Also make sure we have a reasonable mask
1010 	 * (we don't yet have one during adds).
1011 	 */
1012 	if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != 0) {
1013 		struct rtfc_arg arg;
1014 		arg.rnh = rnh;
1015 		arg.rt0 = rt;
1016 		rnh->rnh_walktree_from(rnh, rt_key(rt), rt_mask(rt),
1017 				       rt_fixchange, &arg);
1018 	}
1019 
1020 	return 0;
1021 }
1022 
1023 static void
1024 rt_maskedcopy(src, dst, netmask)
1025 	struct sockaddr *src, *dst, *netmask;
1026 {
1027 	u_char *cp1 = (u_char *)src;
1028 	u_char *cp2 = (u_char *)dst;
1029 	u_char *cp3 = (u_char *)netmask;
1030 	u_char *cplim = cp2 + *cp3;
1031 	u_char *cplim2 = cp2 + *cp1;
1032 
1033 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1034 	cp3 += 2;
1035 	if (cplim > cplim2)
1036 		cplim = cplim2;
1037 	while (cp2 < cplim)
1038 		*cp2++ = *cp1++ & *cp3++;
1039 	if (cp2 < cplim2)
1040 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1041 }
1042 
1043 /*
1044  * Set up a routing table entry, normally
1045  * for an interface.
1046  */
1047 int
1048 rtinit(ifa, cmd, flags)
1049 	struct ifaddr *ifa;
1050 	int cmd, flags;
1051 {
1052 	struct rtentry *rt;
1053 	struct sockaddr *dst;
1054 	struct sockaddr *deldst;
1055 	struct sockaddr *netmask;
1056 	struct mbuf *m = 0;
1057 	struct rtentry *nrt = 0;
1058 	struct radix_node_head *rnh;
1059 	struct radix_node *rn;
1060 	int error;
1061 	struct rt_addrinfo info;
1062 
1063 	if (flags & RTF_HOST) {
1064 		dst = ifa->ifa_dstaddr;
1065 		netmask = NULL;
1066 	} else {
1067 		dst = ifa->ifa_addr;
1068 		netmask = ifa->ifa_netmask;
1069 	}
1070 	/*
1071 	 * If it's a delete, check that if it exists, it's on the correct
1072 	 * interface or we might scrub a route to another ifa which would
1073 	 * be confusing at best and possibly worse.
1074 	 */
1075 	if (cmd == RTM_DELETE) {
1076 		/*
1077 		 * It's a delete, so it should already exist..
1078 		 * If it's a net, mask off the host bits
1079 		 * (Assuming we have a mask)
1080 		 */
1081 		if (netmask != NULL) {
1082 			m = m_get(M_DONTWAIT, MT_SONAME);
1083 			if (m == NULL)
1084 				return(ENOBUFS);
1085 			deldst = mtod(m, struct sockaddr *);
1086 			rt_maskedcopy(dst, deldst, netmask);
1087 			dst = deldst;
1088 		}
1089 		/*
1090 		 * Look up an rtentry that is in the routing tree and
1091 		 * contains the correct info.
1092 		 */
1093 		if ((rnh = rt_tables[dst->sa_family]) == NULL ||
1094 		    (rn = rnh->rnh_lookup(dst, netmask, rnh)) == NULL ||
1095 		    (rn->rn_flags & RNF_ROOT) ||
1096 		    ((struct rtentry *)rn)->rt_ifa != ifa ||
1097 		    !equal(SA(rn->rn_key), dst)) {
1098 			if (m)
1099 				(void) m_free(m);
1100 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1101 		}
1102 		/* XXX */
1103 #if 0
1104 		else {
1105 			/*
1106 			 * One would think that as we are deleting, and we know
1107 			 * it doesn't exist, we could just return at this point
1108 			 * with an "ELSE" clause, but apparently not..
1109 			 */
1110 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1111 		}
1112 #endif
1113 	}
1114 	/*
1115 	 * Do the actual request
1116 	 */
1117 	bzero((caddr_t)&info, sizeof(info));
1118 	info.rti_ifa = ifa;
1119 	info.rti_flags = flags | ifa->ifa_flags;
1120 	info.rti_info[RTAX_DST] = dst;
1121 	info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1122 	info.rti_info[RTAX_NETMASK] = netmask;
1123 	error = rtrequest1(cmd, &info, &nrt);
1124 	if (error == 0 && (rt = nrt) != NULL) {
1125 		/*
1126 		 * notify any listening routing agents of the change
1127 		 */
1128 		rt_newaddrmsg(cmd, ifa, error, rt);
1129 		if (cmd == RTM_DELETE) {
1130 			/*
1131 			 * If we are deleting, and we found an entry, then
1132 			 * it's been removed from the tree.. now throw it away.
1133 			 */
1134 			if (rt->rt_refcnt <= 0) {
1135 				rt->rt_refcnt++; /* make a 1->0 transition */
1136 				rtfree(rt);
1137 			}
1138 		} else if (cmd == RTM_ADD) {
1139 			/*
1140 			 * We just wanted to add it.. we don't actually
1141 			 * need a reference.
1142 			 */
1143 			rt->rt_refcnt--;
1144 		}
1145 	}
1146 	if (m)
1147 		(void) m_free(m);
1148 	return (error);
1149 }
1150 
1151 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */
1152 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
1153