xref: /dflybsd-src/sys/net/route.c (revision 23c32883e759b0ea42fdaff39e661bd1a12e3b9f)
1 /*
2  * Copyright (c) 2004, 2005 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Jeffrey M. Hsu.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of The DragonFly Project nor the names of its
16  *    contributors may be used to endorse or promote products derived
17  *    from this software without specific, prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
22  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
23  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
25  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
27  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
28  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
29  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 2004, 2005 Jeffrey M. Hsu.  All rights reserved.
35  *
36  * License terms: all terms for the DragonFly license above plus the following:
37  *
38  * 4. All advertising materials mentioning features or use of this software
39  *    must display the following acknowledgement:
40  *
41  *	This product includes software developed by Jeffrey M. Hsu
42  *	for the DragonFly Project.
43  *
44  *    This requirement may be waived with permission from Jeffrey Hsu.
45  *    Permission will be granted to any DragonFly user for free.
46  *    This requirement will sunset and may be removed on Jan 31, 2006,
47  *    after which the standard DragonFly license (as shown above) will
48  *    apply.
49  */
50 
51 /*
52  * Copyright (c) 1980, 1986, 1991, 1993
53  *	The Regents of the University of California.  All rights reserved.
54  *
55  * Redistribution and use in source and binary forms, with or without
56  * modification, are permitted provided that the following conditions
57  * are met:
58  * 1. Redistributions of source code must retain the above copyright
59  *    notice, this list of conditions and the following disclaimer.
60  * 2. Redistributions in binary form must reproduce the above copyright
61  *    notice, this list of conditions and the following disclaimer in the
62  *    documentation and/or other materials provided with the distribution.
63  * 3. All advertising materials mentioning features or use of this software
64  *    must display the following acknowledgement:
65  *	This product includes software developed by the University of
66  *	California, Berkeley and its contributors.
67  * 4. Neither the name of the University nor the names of its contributors
68  *    may be used to endorse or promote products derived from this software
69  *    without specific prior written permission.
70  *
71  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
72  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
73  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
74  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
75  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
76  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
77  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
78  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
79  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
80  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
81  * SUCH DAMAGE.
82  *
83  *	@(#)route.c	8.3 (Berkeley) 1/9/95
84  * $FreeBSD: src/sys/net/route.c,v 1.59.2.10 2003/01/17 08:04:00 ru Exp $
85  * $DragonFly: src/sys/net/route.c,v 1.19 2005/03/04 03:34:47 hsu Exp $
86  */
87 
88 #include "opt_inet.h"
89 
90 #include <sys/param.h>
91 #include <sys/systm.h>
92 #include <sys/malloc.h>
93 #include <sys/mbuf.h>
94 #include <sys/socket.h>
95 #include <sys/domain.h>
96 #include <sys/kernel.h>
97 
98 #include <net/if.h>
99 #include <net/route.h>
100 
101 #include <netinet/in.h>
102 #include <net/ip_mroute/ip_mroute.h>
103 
104 static struct rtstat rtstat;
105 struct radix_node_head *rt_tables[AF_MAX+1];
106 
107 static void	rt_maskedcopy (struct sockaddr *, struct sockaddr *,
108 			       struct sockaddr *);
109 static void	rtable_init (void **);
110 
111 static void
112 rtable_init(void **table)
113 {
114 	struct domain *dom;
115 
116 	SLIST_FOREACH(dom, &domains, dom_next)
117 		if (dom->dom_rtattach)
118 			dom->dom_rtattach(&table[dom->dom_family],
119 					  dom->dom_rtoffset);
120 }
121 
122 void
123 route_init()
124 {
125 	rn_init();	/* initialize all zeroes, all ones, mask table */
126 	rtable_init((void **)rt_tables);
127 }
128 
129 /*
130  * Packet routing routines.
131  */
132 
133 /*
134  * Look up and fill in the "ro_rt" rtentry field in a route structure given
135  * an address in the "ro_dst" field.  Always send a report on a miss and
136  * always clone routes.
137  */
138 void
139 rtalloc(struct route *ro)
140 {
141 	rtalloc_ign(ro, 0UL);
142 }
143 
144 /*
145  * Look up and fill in the "ro_rt" rtentry field in a route structure given
146  * an address in the "ro_dst" field.  Always send a report on a miss and
147  * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being
148  * ignored.
149  */
150 void
151 rtalloc_ign(struct route *ro, u_long ignoreflags)
152 {
153 	if (ro->ro_rt != NULL) {
154 		if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP)
155 			return;
156 		rtfree(ro->ro_rt);
157 		ro->ro_rt = NULL;
158 	}
159 	ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags);
160 }
161 
162 /*
163  * Look up the route that matches the given "dst" address.
164  *
165  * Route lookup can have the side-effect of creating and returning
166  * a cloned route instead when "dst" matches a cloning route and the
167  * RTF_CLONING and RTF_PRCLONING flags are not being ignored.
168  *
169  * Any route returned has its reference count incremented.
170  */
171 struct rtentry *
172 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore)
173 {
174 	struct radix_node_head *rnh = rt_tables[dst->sa_family];
175 	struct rtentry *rt;
176 
177 	if (rnh == NULL)
178 		goto unreach;
179 
180 	/*
181 	 * Look up route in the radix tree.
182 	 */
183 	rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh);
184 	if (rt == NULL)
185 		goto unreach;
186 
187 	/*
188 	 * Handle cloning routes.
189 	 */
190 	if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) {
191 		struct rtentry *clonedroute;
192 		int error;
193 
194 		clonedroute = rt;	/* copy in/copy out parameter */
195 		error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0,
196 				  &clonedroute);	/* clone the route */
197 		if (error != 0) {	/* cloning failed */
198 			if (generate_report)
199 				rt_dstmsg(RTM_MISS, dst, error);
200 			rt->rt_refcnt++;
201 			return (rt);	/* return the uncloned route */
202 		}
203 		if (generate_report) {
204 			if (clonedroute->rt_flags & RTF_XRESOLVE)
205 				rt_dstmsg(RTM_RESOLVE, dst, 0);
206 			else
207 				rt_rtmsg(RTM_ADD, clonedroute,
208 					 clonedroute->rt_ifp, 0);
209 		}
210 		return (clonedroute);	/* return cloned route */
211 	}
212 
213 	/*
214 	 * Increment the reference count of the matched route and return.
215 	 */
216 	rt->rt_refcnt++;
217 	return (rt);
218 
219 unreach:
220 	rtstat.rts_unreach++;
221 	if (generate_report)
222 		rt_dstmsg(RTM_MISS, dst, 0);
223 	return (NULL);
224 }
225 
226 void
227 rtfree(struct rtentry *rt)
228 {
229 	KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt));
230 
231 	--rt->rt_refcnt;
232 	if (rt->rt_refcnt == 0) {
233 		struct radix_node_head *rnh = rt_tables[rt_key(rt)->sa_family];
234 
235 		if (rnh->rnh_close)
236 			rnh->rnh_close((struct radix_node *)rt, rnh);
237 		if (!(rt->rt_flags & RTF_UP)) {
238 			/* deallocate route */
239 			if (rt->rt_ifa != NULL)
240 				IFAFREE(rt->rt_ifa);
241 			if (rt->rt_parent != NULL)
242 				RTFREE(rt->rt_parent);	/* recursive call! */
243 			Free(rt_key(rt));
244 			Free(rt);
245 		}
246 	}
247 }
248 
249 /*
250  * Force a routing table entry to the specified destination to go through
251  * the given gateway.  Normally called as a result of a routing redirect
252  * message from the network layer.
253  *
254  * N.B.: must be called at splnet
255  */
256 void
257 rtredirect(struct sockaddr *dst, struct sockaddr *gateway,
258 	   struct sockaddr *netmask, int flags, struct sockaddr *src)
259 {
260 	struct rtentry *rt = NULL;
261 	struct rt_addrinfo rtinfo;
262 	struct ifaddr *ifa;
263 	short *stat = NULL;
264 	int error;
265 
266 	/* verify the gateway is directly reachable */
267 	if ((ifa = ifa_ifwithnet(gateway)) == NULL) {
268 		error = ENETUNREACH;
269 		goto out;
270 	}
271 
272 	/*
273 	 * If the redirect isn't from our current router for this destination,
274 	 * it's either old or wrong.
275 	 */
276 	if (!(flags & RTF_DONE) &&		/* XXX JH */
277 	    (rt = rtpurelookup(dst)) != NULL &&
278 	    (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) {
279 		error = EINVAL;
280 		goto done;
281 	}
282 
283 	/*
284 	 * If it redirects us to ourselves, we have a routing loop,
285 	 * perhaps as a result of an interface going down recently.
286 	 */
287 	if (ifa_ifwithaddr(gateway)) {
288 		error = EHOSTUNREACH;
289 		goto done;
290 	}
291 
292 	/*
293 	 * Create a new entry if the lookup failed or if we got back
294 	 * a wildcard entry for the default route.  This is necessary
295 	 * for hosts which use routing redirects generated by smart
296 	 * gateways to dynamically build the routing tables.
297 	 */
298 	if (rt == NULL)
299 		goto create;
300 	if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) {
301 		rtfree(rt);
302 		goto create;
303 	}
304 
305 	/* Ignore redirects for directly connected hosts. */
306 	if (!(rt->rt_flags & RTF_GATEWAY)) {
307 		error = EHOSTUNREACH;
308 		goto done;
309 	}
310 
311 	if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) {
312 		/*
313 		 * Changing from a network route to a host route.
314 		 * Create a new host route rather than smashing the
315 		 * network route.
316 		 */
317 create:
318 		flags |=  RTF_GATEWAY | RTF_DYNAMIC;
319 		bzero(&rtinfo, sizeof(struct rt_addrinfo));
320 		rtinfo.rti_info[RTAX_DST] = dst;
321 		rtinfo.rti_info[RTAX_GATEWAY] = gateway;
322 		rtinfo.rti_info[RTAX_NETMASK] = netmask;
323 		rtinfo.rti_flags = flags;
324 		rtinfo.rti_ifa = ifa;
325 		rt = NULL;	/* copy-in/copy-out parameter */
326 		error = rtrequest1(RTM_ADD, &rtinfo, &rt);
327 		if (rt != NULL)
328 			flags = rt->rt_flags;
329 		stat = &rtstat.rts_dynamic;
330 	} else {
331 		/*
332 		 * Smash the current notion of the gateway to this destination.
333 		 * Should check about netmask!!!
334 		 */
335 		rt->rt_flags |= RTF_MODIFIED;
336 		flags |= RTF_MODIFIED;
337 		rt_setgate(rt, rt_key(rt), gateway);
338 		error = 0;
339 		stat = &rtstat.rts_newgateway;
340 	}
341 
342 done:
343 	if (rt != NULL)
344 		rtfree(rt);
345 out:
346 	if (error != 0)
347 		rtstat.rts_badredirect++;
348 	else if (stat != NULL)
349 		(*stat)++;
350 
351 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
352 	rtinfo.rti_info[RTAX_DST] = dst;
353 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
354 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
355 	rtinfo.rti_info[RTAX_AUTHOR] = src;
356 	rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error);
357 }
358 
359 /*
360 * Routing table ioctl interface.
361 */
362 int
363 rtioctl(u_long req, caddr_t data, struct thread *td)
364 {
365 #ifdef INET
366 	/* Multicast goop, grrr... */
367 	return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP;
368 #else
369 	return ENXIO;
370 #endif
371 }
372 
373 struct ifaddr *
374 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
375 {
376 	struct ifaddr *ifa;
377 
378 	if (!(flags & RTF_GATEWAY)) {
379 		/*
380 		 * If we are adding a route to an interface,
381 		 * and the interface is a point-to-point link,
382 		 * we should search for the destination
383 		 * as our clue to the interface.  Otherwise
384 		 * we can use the local address.
385 		 */
386 		ifa = NULL;
387 		if (flags & RTF_HOST) {
388 			ifa = ifa_ifwithdstaddr(dst);
389 		}
390 		if (ifa == NULL)
391 			ifa = ifa_ifwithaddr(gateway);
392 	} else {
393 		/*
394 		 * If we are adding a route to a remote net
395 		 * or host, the gateway may still be on the
396 		 * other end of a pt to pt link.
397 		 */
398 		ifa = ifa_ifwithdstaddr(gateway);
399 	}
400 	if (ifa == NULL)
401 		ifa = ifa_ifwithnet(gateway);
402 	if (ifa == NULL) {
403 		struct rtentry *rt;
404 
405 		rt = rtpurelookup(gateway);
406 		if (rt == NULL)
407 			return (NULL);
408 		rt->rt_refcnt--;
409 		if ((ifa = rt->rt_ifa) == NULL)
410 			return (NULL);
411 	}
412 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
413 		struct ifaddr *oldifa = ifa;
414 
415 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
416 		if (ifa == NULL)
417 			ifa = oldifa;
418 	}
419 	return (ifa);
420 }
421 
422 static int rt_fixdelete (struct radix_node *, void *);
423 static int rt_fixchange (struct radix_node *, void *);
424 
425 struct rtfc_arg {
426 	struct rtentry *rt0;
427 	struct radix_node_head *rnh;
428 };
429 
430 /*
431  * Set rtinfo->rti_ifa and rtinfo->rti_ifp.
432  */
433 int
434 rt_getifa(struct rt_addrinfo *rtinfo)
435 {
436 	struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY];
437 	struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
438 	struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA];
439 	int flags = rtinfo->rti_flags;
440 
441 	/*
442 	 * ifp may be specified by sockaddr_dl
443 	 * when protocol address is ambiguous.
444 	 */
445 	if (rtinfo->rti_ifp == NULL) {
446 		struct sockaddr *ifpaddr;
447 
448 		ifpaddr = rtinfo->rti_info[RTAX_IFP];
449 		if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) {
450 			struct ifaddr *ifa;
451 
452 			ifa = ifa_ifwithnet(ifpaddr);
453 			if (ifa != NULL)
454 				rtinfo->rti_ifp = ifa->ifa_ifp;
455 		}
456 	}
457 
458 	if (rtinfo->rti_ifa == NULL && ifaaddr != NULL)
459 		rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr);
460 	if (rtinfo->rti_ifa == NULL) {
461 		struct sockaddr *sa;
462 
463 		sa = ifaaddr != NULL ? ifaaddr :
464 		    (gateway != NULL ? gateway : dst);
465 		if (sa != NULL && rtinfo->rti_ifp != NULL)
466 			rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp);
467 		else if (dst != NULL && gateway != NULL)
468 			rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway);
469 		else if (sa != NULL)
470 			rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa);
471 	}
472 	if (rtinfo->rti_ifa == NULL)
473 		return (ENETUNREACH);
474 
475 	if (rtinfo->rti_ifp == NULL)
476 		rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp;
477 	return (0);
478 }
479 
480 /*
481  * Do appropriate manipulations of a routing tree given
482  * all the bits of info needed
483  */
484 int
485 rtrequest(
486 	int req,
487 	struct sockaddr *dst,
488 	struct sockaddr *gateway,
489 	struct sockaddr *netmask,
490 	int flags,
491 	struct rtentry **ret_nrt)
492 {
493 	struct rt_addrinfo rtinfo;
494 
495 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
496 	rtinfo.rti_info[RTAX_DST] = dst;
497 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
498 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
499 	rtinfo.rti_flags = flags;
500 	return rtrequest1(req, &rtinfo, ret_nrt);
501 }
502 
503 int
504 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt)
505 {
506 	struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
507 	struct rtentry *rt;
508 	struct radix_node *rn;
509 	struct radix_node_head *rnh;
510 	struct ifaddr *ifa;
511 	struct sockaddr *ndst;
512 	int error = 0;
513 	int s;
514 
515 #define gotoerr(x) { error = x ; goto bad; }
516 
517 	s = splnet();
518 	/*
519 	 * Find the correct routing tree to use for this Address Family
520 	 */
521 	if ((rnh = rt_tables[dst->sa_family]) == NULL)
522 		gotoerr(EAFNOSUPPORT);
523 
524 	/*
525 	 * If we are adding a host route then we don't want to put
526 	 * a netmask in the tree, nor do we want to clone it.
527 	 */
528 	if (rtinfo->rti_flags & RTF_HOST) {
529 		rtinfo->rti_info[RTAX_NETMASK] = NULL;
530 		rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING);
531 	}
532 
533 	switch (req) {
534 	case RTM_DELETE:
535 		/* Remove the item from the tree. */
536 		rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST],
537 				      (char *)rtinfo->rti_info[RTAX_NETMASK],
538 				      rnh);
539 		if (rn == NULL)
540 			gotoerr(ESRCH);
541 		KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)),
542 			("rnh_deladdr returned flags 0x%x", rn->rn_flags));
543 		rt = (struct rtentry *)rn;
544 
545 		/* Free any routes cloned from this one. */
546 		if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) &&
547 		    rt_mask(rt) != NULL) {
548 			rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
549 					       (char *)rt_mask(rt),
550 					       rt_fixdelete, rt);
551 		}
552 
553 		if (rt->rt_gwroute != NULL) {
554 			RTFREE(rt->rt_gwroute);
555 			rt->rt_gwroute = NULL;
556 		}
557 
558 		/*
559 		 * NB: RTF_UP must be set during the search above,
560 		 * because we might delete the last ref, causing
561 		 * rt to get freed prematurely.
562 		 */
563 		rt->rt_flags &= ~RTF_UP;
564 
565 		/* Give the protocol a chance to keep things in sync. */
566 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
567 			ifa->ifa_rtrequest(RTM_DELETE, rt, rtinfo);
568 
569 		/*
570 		 * If the caller wants it, then it can have it,
571 		 * but it's up to it to free the rtentry as we won't be
572 		 * doing it.
573 		 */
574 		KASSERT(rt->rt_refcnt >= 0,
575 			("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt));
576 		if (ret_nrt != NULL) {
577 			*ret_nrt = rt;
578 		} else if (rt->rt_refcnt == 0) {
579 			rt->rt_refcnt++;  /* refcnt > 0 required for rtfree() */
580 			rtfree(rt);
581 		}
582 		break;
583 
584 	case RTM_RESOLVE:
585 		if (ret_nrt == NULL || (rt = *ret_nrt) == NULL)
586 			gotoerr(EINVAL);
587 		ifa = rt->rt_ifa;
588 		rtinfo->rti_flags =
589 		    rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC);
590 		rtinfo->rti_flags |= RTF_WASCLONED;
591 		rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
592 		if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL)
593 			rtinfo->rti_flags |= RTF_HOST;
594 		goto makeroute;
595 
596 	case RTM_ADD:
597 		KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) ||
598 			rtinfo->rti_info[RTAX_GATEWAY] != NULL,
599 		    ("rtrequest: GATEWAY but no gateway"));
600 
601 		if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo)))
602 			gotoerr(error);
603 		ifa = rtinfo->rti_ifa;
604 makeroute:
605 		R_Malloc(rt, struct rtentry *, sizeof(struct rtentry));
606 		if (rt == NULL)
607 			gotoerr(ENOBUFS);
608 		bzero(rt, sizeof(struct rtentry));
609 		rt->rt_flags = RTF_UP | rtinfo->rti_flags;
610 		error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY]);
611 		if (error != 0) {
612 			Free(rt);
613 			gotoerr(error);
614 		}
615 
616 		ndst = rt_key(rt);
617 		if (rtinfo->rti_info[RTAX_NETMASK] != NULL)
618 			rt_maskedcopy(dst, ndst,
619 				      rtinfo->rti_info[RTAX_NETMASK]);
620 		else
621 			bcopy(dst, ndst, dst->sa_len);
622 
623 		/*
624 		 * Note that we now have a reference to the ifa.
625 		 * This moved from below so that rnh->rnh_addaddr() can
626 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
627 		 */
628 		IFAREF(ifa);
629 		rt->rt_ifa = ifa;
630 		rt->rt_ifp = ifa->ifa_ifp;
631 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
632 
633 		rn = rnh->rnh_addaddr((char *)ndst,
634 				      (char *)rtinfo->rti_info[RTAX_NETMASK],
635 				      rnh, rt->rt_nodes);
636 		if (rn == NULL) {
637 			struct rtentry *oldrt;
638 
639 			/*
640 			 * We already have one of these in the tree.
641 			 * We do a special hack: if the old route was
642 			 * cloned, then we blow it away and try
643 			 * re-inserting the new one.
644 			 */
645 			oldrt = rtpurelookup(ndst);
646 			if (oldrt != NULL) {
647 				--oldrt->rt_refcnt;
648 				if (oldrt->rt_flags & RTF_WASCLONED) {
649 					rtrequest(RTM_DELETE, rt_key(oldrt),
650 						  oldrt->rt_gateway,
651 						  rt_mask(oldrt),
652 						  oldrt->rt_flags, NULL);
653 					rn = rnh->rnh_addaddr((char *)ndst,
654 					    (char *)
655 						rtinfo->rti_info[RTAX_NETMASK],
656 					    rnh, rt->rt_nodes);
657 				}
658 			}
659 		}
660 
661 		/*
662 		 * If it still failed to go into the tree,
663 		 * then un-make it (this should be a function).
664 		 */
665 		if (rn == NULL) {
666 			if (rt->rt_gwroute != NULL)
667 				rtfree(rt->rt_gwroute);
668 			IFAFREE(ifa);
669 			Free(rt_key(rt));
670 			Free(rt);
671 			gotoerr(EEXIST);
672 		}
673 
674 		/*
675 		 * If we got here from RESOLVE, then we are cloning
676 		 * so clone the rest, and note that we
677 		 * are a clone (and increment the parent's references)
678 		 */
679 		if (req == RTM_RESOLVE) {
680 			rt->rt_rmx = (*ret_nrt)->rt_rmx;    /* copy metrics */
681 			rt->rt_rmx.rmx_pksent = 0;  /* reset packet counter */
682 			if ((*ret_nrt)->rt_flags &
683 				       (RTF_CLONING | RTF_PRCLONING)) {
684 				rt->rt_parent = *ret_nrt;
685 				(*ret_nrt)->rt_refcnt++;
686 			}
687 		}
688 
689 		/*
690 		 * if this protocol has something to add to this then
691 		 * allow it to do that as well.
692 		 */
693 		if (ifa->ifa_rtrequest != NULL)
694 			ifa->ifa_rtrequest(req, rt, rtinfo);
695 
696 		/*
697 		 * We repeat the same procedure from rt_setgate() here because
698 		 * it doesn't fire when we call it there because the node
699 		 * hasn't been added to the tree yet.
700 		 */
701 		if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) &&
702 		    rt_mask(rt) != NULL) {
703 			struct rtfc_arg arg = { rt, rnh };
704 
705 			rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
706 					       (char *)rt_mask(rt),
707 					       rt_fixchange, &arg);
708 		}
709 
710 		/*
711 		 * Return the resulting rtentry,
712 		 * increasing the number of references by one.
713 		 */
714 		if (ret_nrt != NULL) {
715 			rt->rt_refcnt++;
716 			*ret_nrt = rt;
717 		}
718 		break;
719 	default:
720 		error = EOPNOTSUPP;
721 	}
722 bad:
723 	splx(s);
724 	return (error);
725 }
726 
727 /*
728  * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family''
729  * (i.e., the routes related to it by the operation of cloning).  This
730  * routine is iterated over all potential former-child-routes by way of
731  * rnh->rnh_walktree_from() above, and those that actually are children of
732  * the late parent (passed in as VP here) are themselves deleted.
733  */
734 static int
735 rt_fixdelete(struct radix_node *rn, void *vp)
736 {
737 	struct rtentry *rt = (struct rtentry *)rn;
738 	struct rtentry *rt0 = vp;
739 
740 	if (rt->rt_parent == rt0 &&
741 	    !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
742 		return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
743 				 rt->rt_flags, NULL);
744 	}
745 	return 0;
746 }
747 
748 /*
749  * This routine is called from rt_setgate() to do the analogous thing for
750  * adds and changes.  There is the added complication in this case of a
751  * middle insert; i.e., insertion of a new network route between an older
752  * network route and (cloned) host routes.  For this reason, a simple check
753  * of rt->rt_parent is insufficient; each candidate route must be tested
754  * against the (mask, value) of the new route (passed as before in vp)
755  * to see if the new route matches it.
756  *
757  * XXX - it may be possible to do fixdelete() for changes and reserve this
758  * routine just for adds.  I'm not sure why I thought it was necessary to do
759  * changes this way.
760  */
761 #ifdef DEBUG
762 static int rtfcdebug = 0;
763 #endif
764 
765 static int
766 rt_fixchange(struct radix_node *rn, void *vp)
767 {
768 	struct rtentry *rt = (struct rtentry *)rn;
769 	struct rtfc_arg *ap = vp;
770 	struct rtentry *rt0 = ap->rt0;
771 	struct radix_node_head *rnh = ap->rnh;
772 	u_char *xk1, *xm1, *xk2, *xmp;
773 	int i, len, mlen;
774 
775 #ifdef DEBUG
776 	if (rtfcdebug)
777 		printf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0);
778 #endif
779 
780 	if (rt->rt_parent == NULL ||
781 	    (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
782 #ifdef DEBUG
783 		if (rtfcdebug) printf("no parent, pinned or cloning\n");
784 #endif
785 		return 0;
786 	}
787 
788 	if (rt->rt_parent == rt0) {
789 #ifdef DEBUG
790 		if (rtfcdebug) printf("parent match\n");
791 #endif
792 		return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
793 				 rt->rt_flags, NULL);
794 	}
795 
796 	/*
797 	 * There probably is a function somewhere which does this...
798 	 * if not, there should be.
799 	 */
800 	len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len);
801 
802 	xk1 = (u_char *)rt_key(rt0);
803 	xm1 = (u_char *)rt_mask(rt0);
804 	xk2 = (u_char *)rt_key(rt);
805 
806 	/* avoid applying a less specific route */
807 	xmp = (u_char *)rt_mask(rt->rt_parent);
808 	mlen = rt_key(rt->rt_parent)->sa_len;
809 	if (mlen > rt_key(rt0)->sa_len) {
810 #ifdef DEBUG
811 		if (rtfcdebug)
812 			printf("rt_fixchange: inserting a less "
813 			       "specific route\n");
814 #endif
815 		return 0;
816 	}
817 	for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) {
818 		if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) {
819 #ifdef DEBUG
820 			if (rtfcdebug)
821 				printf("rt_fixchange: inserting a less "
822 				       "specific route\n");
823 #endif
824 			return 0;
825 		}
826 	}
827 
828 	for (i = rnh->rnh_treetop->rn_offset; i < len; i++) {
829 		if ((xk2[i] & xm1[i]) != xk1[i]) {
830 #ifdef DEBUG
831 			if (rtfcdebug) printf("no match\n");
832 #endif
833 			return 0;
834 		}
835 	}
836 
837 	/*
838 	 * OK, this node is a clone, and matches the node currently being
839 	 * changed/added under the node's mask.  So, get rid of it.
840 	 */
841 #ifdef DEBUG
842 	if (rtfcdebug) printf("deleting\n");
843 #endif
844 	return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
845 			 rt->rt_flags, NULL);
846 }
847 
848 #define ROUNDUP(a) (a>0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))
849 
850 int
851 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate)
852 {
853 	char *space, *oldspace;
854 	int dlen = ROUNDUP(dst->sa_len), glen = ROUNDUP(gate->sa_len);
855 	struct rtentry *rt = rt0;
856 	struct radix_node_head *rnh = rt_tables[dst->sa_family];
857 
858 	/*
859 	 * A host route with the destination equal to the gateway
860 	 * will interfere with keeping LLINFO in the routing
861 	 * table, so disallow it.
862 	 */
863 	if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) ==
864 			      (RTF_HOST | RTF_GATEWAY)) &&
865 	    dst->sa_len == gate->sa_len &&
866 	    sa_equal(dst, gate)) {
867 		/*
868 		 * The route might already exist if this is an RTM_CHANGE
869 		 * or a routing redirect, so try to delete it.
870 		 */
871 		if (rt_key(rt0) != NULL)
872 			rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway,
873 				  rt_mask(rt0), rt0->rt_flags, NULL);
874 		return EADDRNOTAVAIL;
875 	}
876 
877 	/*
878 	 * Both dst and gateway are stored in the same malloc'ed chunk
879 	 * (If I ever get my hands on....)
880 	 * if we need to malloc a new chunk, then keep the old one around
881 	 * till we don't need it any more.
882 	 */
883 	if (rt->rt_gateway == NULL || glen > ROUNDUP(rt->rt_gateway->sa_len)) {
884 		oldspace = (char *)rt_key(rt);
885 		R_Malloc(space, char *, dlen + glen);
886 		if (space == NULL)
887 			return ENOBUFS;
888 		rt->rt_nodes->rn_key = space;
889 	} else {
890 		space = (char *)rt_key(rt);	/* Just use the old space. */
891 		oldspace = NULL;
892 	}
893 
894 	/* Set the gateway value. */
895 	rt->rt_gateway = (struct sockaddr *)(space + dlen);
896 	bcopy(gate, rt->rt_gateway, glen);
897 
898 	if (oldspace != NULL) {
899 		/*
900 		 * If we allocated a new chunk, preserve the original dst.
901 		 * This way, rt_setgate() really just sets the gate
902 		 * and leaves the dst field alone.
903 		 */
904 		bcopy(dst, space, dlen);
905 		Free(oldspace);
906 	}
907 
908 	/*
909 	 * If there is already a gwroute, it's now almost definitely wrong
910 	 * so drop it.
911 	 */
912 	if (rt->rt_gwroute != NULL) {
913 		RTFREE(rt->rt_gwroute);
914 		rt->rt_gwroute = NULL;
915 	}
916 	if (rt->rt_flags & RTF_GATEWAY) {
917 		/*
918 		 * Cloning loop avoidance: In the presence of
919 		 * protocol-cloning and bad configuration, it is
920 		 * possible to get stuck in bottomless mutual recursion
921 		 * (rtrequest rt_setgate rtlookup).  We avoid this
922 		 * by not allowing protocol-cloning to operate for
923 		 * gateways (which is probably the correct choice
924 		 * anyway), and avoid the resulting reference loops
925 		 * by disallowing any route to run through itself as
926 		 * a gateway.  This is obviously mandatory when we
927 		 * get rt->rt_output().
928 		 *
929 		 * This breaks TTCP for hosts outside the gateway!  XXX JH
930 		 */
931 		rt->rt_gwroute = _rtlookup(gate, RTL_REPORTMSG, RTF_PRCLONING);
932 		if (rt->rt_gwroute == rt) {
933 			rt->rt_gwroute = NULL;
934 			--rt->rt_refcnt;
935 			return EDQUOT; /* failure */
936 		}
937 	}
938 
939 	/*
940 	 * This isn't going to do anything useful for host routes, so
941 	 * don't bother.  Also make sure we have a reasonable mask
942 	 * (we don't yet have one during adds).
943 	 */
944 	if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) {
945 		struct rtfc_arg arg = { rt, rnh };
946 
947 		rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
948 				       (char *)rt_mask(rt),
949 				       rt_fixchange, &arg);
950 	}
951 
952 	return 0;
953 }
954 
955 static void
956 rt_maskedcopy(
957 	struct sockaddr *src,
958 	struct sockaddr *dst,
959 	struct sockaddr *netmask)
960 {
961 	u_char *cp1 = (u_char *)src;
962 	u_char *cp2 = (u_char *)dst;
963 	u_char *cp3 = (u_char *)netmask;
964 	u_char *cplim = cp2 + *cp3;
965 	u_char *cplim2 = cp2 + *cp1;
966 
967 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
968 	cp3 += 2;
969 	if (cplim > cplim2)
970 		cplim = cplim2;
971 	while (cp2 < cplim)
972 		*cp2++ = *cp1++ & *cp3++;
973 	if (cp2 < cplim2)
974 		bzero(cp2, cplim2 - cp2);
975 }
976 
977 int
978 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt)
979 {
980 	struct rtentry *up_rt, *rt;
981 
982 	if (!(rt0->rt_flags & RTF_UP)) {
983 		up_rt = rtlookup(dst);
984 		if (up_rt == NULL)
985 			return (EHOSTUNREACH);
986 		up_rt->rt_refcnt--;
987 	} else
988 		up_rt = rt0;
989 	if (up_rt->rt_flags & RTF_GATEWAY) {
990 		if (up_rt->rt_gwroute == NULL) {
991 			up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
992 			if (up_rt->rt_gwroute == NULL)
993 				return (EHOSTUNREACH);
994 		} else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) {
995 			rtfree(up_rt->rt_gwroute);
996 			up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
997 			if (up_rt->rt_gwroute == NULL)
998 				return (EHOSTUNREACH);
999 		}
1000 		rt = up_rt->rt_gwroute;
1001 	} else
1002 		rt = up_rt;
1003 	if (rt->rt_flags & RTF_REJECT &&
1004 	    (rt->rt_rmx.rmx_expire == 0 ||		/* rt doesn't expire */
1005 	     time_second < rt->rt_rmx.rmx_expire))	/* rt not expired */
1006 		return (rt->rt_flags & RTF_HOST ?  EHOSTDOWN : EHOSTUNREACH);
1007 	*drt = rt;
1008 	return 0;
1009 }
1010 
1011 /*
1012  * Set up a routing table entry, normally for an interface.
1013  */
1014 int
1015 rtinit(struct ifaddr *ifa, int cmd, int flags)
1016 {
1017 	struct sockaddr *dst, *deldst, *netmask;
1018 	struct rtentry *rt;
1019 	struct mbuf *m = NULL;
1020 	struct radix_node_head *rnh;
1021 	struct radix_node *rn;
1022 	struct rt_addrinfo rtinfo;
1023 	int error;
1024 
1025 	if (flags & RTF_HOST) {
1026 		dst = ifa->ifa_dstaddr;
1027 		netmask = NULL;
1028 	} else {
1029 		dst = ifa->ifa_addr;
1030 		netmask = ifa->ifa_netmask;
1031 	}
1032 	/*
1033 	 * If it's a delete, check that if it exists, it's on the correct
1034 	 * interface or we might scrub a route to another ifa which would
1035 	 * be confusing at best and possibly worse.
1036 	 */
1037 	if (cmd == RTM_DELETE) {
1038 		/*
1039 		 * It's a delete, so it should already exist..
1040 		 * If it's a net, mask off the host bits
1041 		 * (Assuming we have a mask)
1042 		 */
1043 		if (netmask != NULL) {
1044 			m = m_get(MB_DONTWAIT, MT_SONAME);
1045 			if (m == NULL)
1046 				return (ENOBUFS);
1047 			deldst = mtod(m, struct sockaddr *);
1048 			rt_maskedcopy(dst, deldst, netmask);
1049 			dst = deldst;
1050 		}
1051 		/*
1052 		 * Look up an rtentry that is in the routing tree and
1053 		 * contains the correct info.
1054 		 */
1055 		if ((rnh = rt_tables[dst->sa_family]) == NULL ||
1056 		    (rn = rnh->rnh_lookup((char *)dst,
1057 					  (char *)netmask, rnh)) == NULL ||
1058 		    ((struct rtentry *)rn)->rt_ifa != ifa ||
1059 		    !sa_equal((struct sockaddr *)rn->rn_key, dst)) {
1060 			if (m != NULL)
1061 				m_free(m);
1062 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1063 		}
1064 		/* XXX */
1065 #if 0
1066 		else {
1067 			/*
1068 			 * One would think that as we are deleting, and we know
1069 			 * it doesn't exist, we could just return at this point
1070 			 * with an "ELSE" clause, but apparently not..
1071 			 */
1072 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1073 		}
1074 #endif
1075 	}
1076 	/*
1077 	 * Do the actual request
1078 	 */
1079 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
1080 	rtinfo.rti_info[RTAX_DST] = dst;
1081 	rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1082 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
1083 	rtinfo.rti_flags = flags | ifa->ifa_flags;
1084 	rtinfo.rti_ifa = ifa;
1085 	error = rtrequest1(cmd, &rtinfo, &rt);
1086 	if (error == 0 && rt != NULL) {
1087 		/*
1088 		 * notify any listening routing agents of the change
1089 		 */
1090 		rt_newaddrmsg(cmd, ifa, error, rt);
1091 		if (cmd == RTM_DELETE) {
1092 			/*
1093 			 * If we are deleting, and we found an entry, then
1094 			 * it's been removed from the tree.. now throw it away.
1095 			 */
1096 			if (rt->rt_refcnt == 0) {
1097 				rt->rt_refcnt++; /* make a 1->0 transition */
1098 				rtfree(rt);
1099 			}
1100 		} else if (cmd == RTM_ADD) {
1101 			/*
1102 			 * We just wanted to add it.. we don't actually
1103 			 * need a reference.
1104 			 */
1105 			rt->rt_refcnt--;
1106 		}
1107 	}
1108 	if (m != NULL)
1109 		m_free(m);
1110 	return (error);
1111 }
1112 
1113 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */
1114 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
1115