xref: /dflybsd-src/sys/net/route.c (revision 17ea22213f86a5c5966c1e6bf8e95f022ebb92b9)
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 1.14 2005/01/26 23:09:57 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 	for (dom = domains; dom; dom = dom->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 if "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 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 (report)
199 				rt_dstmsg(RTM_MISS, dst, error);
200 			rt->rt_refcnt++;
201 			return (rt);	/* return the uncloned route */
202 		}
203 		if (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 (report)
222 		rt_dstmsg(RTM_MISS, dst, 0);
223 	return (NULL);
224 }
225 
226 void
227 rtfree(struct rtentry *rt)
228 {
229 	struct radix_node_head *rnh = rt_tables[rt_key(rt)->sa_family];
230 
231 	--rt->rt_refcnt;
232 	if (rnh->rnh_close && rt->rt_refcnt == 0)
233 		rnh->rnh_close((struct radix_node *)rt, rnh);
234 	if (rt->rt_refcnt <= 0 && !(rt->rt_flags & RTF_UP)) {
235 		KASSERT(!(rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT)),
236 			("rtfree: rn_flags 0x%x ", rt->rt_nodes->rn_flags));
237 		KASSERT(rt->rt_refcnt == 0,
238 			("rtfree: rt_refcnt %ld", rt->rt_refcnt));
239 		if (rt->rt_ifa != NULL)
240 			IFAFREE(rt->rt_ifa);
241 		if (rt->rt_parent != NULL)
242 			RTFREE(rt->rt_parent);
243 		Free(rt_key(rt));  /* Also frees gateway.  See rt_setgate(). */
244 		Free(rt);
245 	}
246 }
247 
248 /*
249  * Force a routing table entry to the specified
250  * destination to go through the given gateway.
251  * 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(
258 	struct sockaddr *dst,
259 	struct sockaddr *gateway,
260 	struct sockaddr *netmask,
261 	int flags,
262 	struct sockaddr *src,
263 	struct rtentry **rtp)
264 {
265 	struct rtentry *rt;
266 	struct rt_addrinfo info;
267 	struct ifaddr *ifa;
268 	short *stat = NULL;
269 	int error;
270 
271 	/* verify the gateway is directly reachable */
272 	if ((ifa = ifa_ifwithnet(gateway)) == NULL) {
273 		error = ENETUNREACH;
274 		goto out;
275 	}
276 
277 	/*
278 	 * If the redirect isn't from our current router for this dst,
279 	 * it's either old or wrong.  If it redirects us to ourselves,
280 	 * we have a routing loop, perhaps as a result of an interface
281 	 * going down recently.
282 	 */
283 	if (!(flags & RTF_DONE) &&
284 	    (rt = rtpurelookup(dst)) != NULL &&
285 	    (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) {
286 		error = EINVAL;
287 		goto done;
288 	} else if (ifa_ifwithaddr(gateway)) {
289 		error = EHOSTUNREACH;
290 		goto done;
291 	}
292 
293 	/*
294 	 * Create a new entry if we just got back a wildcard entry
295 	 * or the the lookup failed.  This is necessary for hosts
296 	 * which use routing redirects generated by smart gateways
297 	 * to dynamically build the routing tables.
298 	 */
299 	if (rt == NULL || (rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2))
300 		goto create;
301 
302 	/*
303 	 * Don't listen to the redirect if it's for a route to an interface.
304 	 */
305 	if (rt->rt_flags & RTF_GATEWAY) {
306 		if ((!(rt->rt_flags & RTF_HOST)) && (flags & RTF_HOST)) {
307 			/*
308 			 * Changing from route to net => route to host.
309 			 * Create new route, rather than smashing route to net.
310 			 */
311 create:
312 			if (rt != NULL)
313 				rtfree(rt);
314 			flags |=  RTF_GATEWAY | RTF_DYNAMIC;
315 			bzero(&info, sizeof info);
316 			info.rti_info[RTAX_DST] = dst;
317 			info.rti_info[RTAX_GATEWAY] = gateway;
318 			info.rti_info[RTAX_NETMASK] = netmask;
319 			info.rti_ifa = ifa;
320 			info.rti_flags = flags;
321 			rt = NULL;
322 			error = rtrequest1(RTM_ADD, &info, &rt);
323 			if (rt != NULL)
324 				flags = rt->rt_flags;
325 			stat = &rtstat.rts_dynamic;
326 		} else {
327 			/*
328 			 * Smash the current notion of the gateway to
329 			 * this destination.  Should check about netmask!!!
330 			 */
331 			rt->rt_flags |= RTF_MODIFIED;
332 			flags |= RTF_MODIFIED;
333 			stat = &rtstat.rts_newgateway;
334 			/* Add the key and gateway (in one malloc'ed chunk). */
335 			rt_setgate(rt, rt_key(rt), gateway);
336 			error = 0;
337 		}
338 	} else {
339 		error = EHOSTUNREACH;
340 	}
341 
342 done:
343 	if (rt != NULL) {
344 		if (rtp != NULL && error == 0)
345 			*rtp = rt;
346 		else
347 			rtfree(rt);
348 	}
349 
350 out:
351 	if (error != 0)
352 		rtstat.rts_badredirect++;
353 	else if (stat != NULL)
354 		(*stat)++;
355 
356 	bzero(&info, sizeof info);
357 	info.rti_info[RTAX_DST] = dst;
358 	info.rti_info[RTAX_GATEWAY] = gateway;
359 	info.rti_info[RTAX_NETMASK] = netmask;
360 	info.rti_info[RTAX_AUTHOR] = src;
361 	rt_missmsg(RTM_REDIRECT, &info, flags, error);
362 }
363 
364 /*
365 * Routing table ioctl interface.
366 */
367 int
368 rtioctl(u_long req, caddr_t data, struct thread *td)
369 {
370 #ifdef INET
371 	/* Multicast goop, grrr... */
372 	return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP;
373 #else
374 	return ENXIO;
375 #endif
376 }
377 
378 struct ifaddr *
379 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
380 {
381 	struct ifaddr *ifa;
382 
383 	if (!(flags & RTF_GATEWAY)) {
384 		/*
385 		 * If we are adding a route to an interface,
386 		 * and the interface is a point-to-point link,
387 		 * we should search for the destination
388 		 * as our clue to the interface.  Otherwise
389 		 * we can use the local address.
390 		 */
391 		ifa = NULL;
392 		if (flags & RTF_HOST) {
393 			ifa = ifa_ifwithdstaddr(dst);
394 		}
395 		if (ifa == NULL)
396 			ifa = ifa_ifwithaddr(gateway);
397 	} else {
398 		/*
399 		 * If we are adding a route to a remote net
400 		 * or host, the gateway may still be on the
401 		 * other end of a pt to pt link.
402 		 */
403 		ifa = ifa_ifwithdstaddr(gateway);
404 	}
405 	if (ifa == NULL)
406 		ifa = ifa_ifwithnet(gateway);
407 	if (ifa == NULL) {
408 		struct rtentry *rt;
409 
410 		rt = rtpurelookup(gateway);
411 		if (rt == NULL)
412 			return (NULL);
413 		rt->rt_refcnt--;
414 		if ((ifa = rt->rt_ifa) == NULL)
415 			return (NULL);
416 	}
417 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
418 		struct ifaddr *oifa = ifa;
419 
420 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
421 		if (ifa == NULL)
422 			ifa = oifa;
423 	}
424 	return (ifa);
425 }
426 
427 static int rt_fixdelete (struct radix_node *, void *);
428 static int rt_fixchange (struct radix_node *, void *);
429 
430 struct rtfc_arg {
431 	struct rtentry *rt0;
432 	struct radix_node_head *rnh;
433 };
434 
435 /*
436  * Set rtinfo->rti_ifa and rtinfo->rti_ifp.
437  */
438 int
439 rt_getifa(struct rt_addrinfo *rtinfo)
440 {
441 	struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY];
442 	struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
443 	struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA];
444 	int flags = rtinfo->rti_flags;
445 
446 	/*
447 	 * ifp may be specified by sockaddr_dl
448 	 * when protocol address is ambiguous.
449 	 */
450 	if (rtinfo->rti_ifp == NULL) {
451 		struct sockaddr *ifpaddr;
452 
453 		ifpaddr = rtinfo->rti_info[RTAX_IFP];
454 		if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) {
455 			struct ifaddr *ifa;
456 
457 			ifa = ifa_ifwithnet(ifpaddr);
458 			if (ifa != NULL)
459 				rtinfo->rti_ifp = ifa->ifa_ifp;
460 		}
461 	}
462 
463 	if (rtinfo->rti_ifa == NULL && ifaaddr != NULL)
464 		rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr);
465 	if (rtinfo->rti_ifa == NULL) {
466 		struct sockaddr *sa;
467 
468 		sa = ifaaddr != NULL ? ifaaddr :
469 		    (gateway != NULL ? gateway : dst);
470 		if (sa != NULL && rtinfo->rti_ifp != NULL)
471 			rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp);
472 		else if (dst != NULL && gateway != NULL)
473 			rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway);
474 		else if (sa != NULL)
475 			rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa);
476 	}
477 	if (rtinfo->rti_ifa == NULL)
478 		return (ENETUNREACH);
479 
480 	if (rtinfo->rti_ifp == NULL)
481 		rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp;
482 	return (0);
483 }
484 
485 /*
486  * Do appropriate manipulations of a routing tree given
487  * all the bits of info needed
488  */
489 int
490 rtrequest(
491 	int req,
492 	struct sockaddr *dst,
493 	struct sockaddr *gateway,
494 	struct sockaddr *netmask,
495 	int flags,
496 	struct rtentry **ret_nrt)
497 {
498 	struct rt_addrinfo info;
499 
500 	bzero(&info, sizeof info);
501 	info.rti_flags = flags;
502 	info.rti_info[RTAX_DST] = dst;
503 	info.rti_info[RTAX_GATEWAY] = gateway;
504 	info.rti_info[RTAX_NETMASK] = netmask;
505 	return rtrequest1(req, &info, ret_nrt);
506 }
507 
508 int
509 rtrequest1(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt)
510 {
511 	struct sockaddr *dst = info->rti_info[RTAX_DST];
512 	struct rtentry *rt;
513 	struct radix_node *rn;
514 	struct radix_node_head *rnh;
515 	struct ifaddr *ifa;
516 	struct sockaddr *ndst;
517 	int error = 0;
518 	int s;
519 
520 #define gotoerr(x) { error = x ; goto bad; }
521 
522 	s = splnet();
523 	/*
524 	 * Find the correct routing tree to use for this Address Family
525 	 */
526 	if ((rnh = rt_tables[dst->sa_family]) == NULL)
527 		gotoerr(EAFNOSUPPORT);
528 
529 	/*
530 	 * If we are adding a host route then we don't want to put
531 	 * a netmask in the tree, nor do we want to clone it.
532 	 */
533 	if (info->rti_flags & RTF_HOST) {
534 		info->rti_info[RTAX_NETMASK] = NULL;
535 		info->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING);
536 	}
537 
538 	switch (req) {
539 	case RTM_DELETE:
540 		/* Remove the item from the tree. */
541 		rn = rnh->rnh_deladdr((char *)info->rti_info[RTAX_DST],
542 				      (char *)info->rti_info[RTAX_NETMASK],
543 				      rnh);
544 		if (rn == NULL)
545 			gotoerr(ESRCH);
546 		KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)),
547 			("rnh_deladdr returned flags 0x%x", rn->rn_flags));
548 		rt = (struct rtentry *)rn;
549 
550 		/* Free any routes cloned from this one. */
551 		if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) &&
552 		    rt_mask(rt) != NULL) {
553 			rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
554 					       (char *)rt_mask(rt),
555 					       rt_fixdelete, rt);
556 		}
557 
558 		if (rt->rt_gwroute != NULL) {
559 			RTFREE(rt->rt_gwroute);
560 			rt->rt_gwroute = NULL;
561 		}
562 
563 		/*
564 		 * NB: RTF_UP must be set during the search above,
565 		 * because we might delete the last ref, causing
566 		 * rt to get freed prematurely.
567 		 */
568 		rt->rt_flags &= ~RTF_UP;
569 
570 		/* Give the protocol a chance to keep things in sync. */
571 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
572 			ifa->ifa_rtrequest(RTM_DELETE, rt, info);
573 
574 		/*
575 		 * If the caller wants it, then it can have it,
576 		 * but it's up to it to free the rtentry as we won't be
577 		 * doing it.
578 		 */
579 		if (ret_nrt != NULL) {
580 			*ret_nrt = rt;
581 		} else if (rt->rt_refcnt <= 0) {
582 			rt->rt_refcnt++;  /* refcnt > 0 required for rtfree() */
583 			rtfree(rt);
584 		}
585 		break;
586 
587 	case RTM_RESOLVE:
588 		if (ret_nrt == NULL || (rt = *ret_nrt) == NULL)
589 			gotoerr(EINVAL);
590 		ifa = rt->rt_ifa;
591 		info->rti_flags =
592 		    rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC);
593 		info->rti_flags |= RTF_WASCLONED;
594 		info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
595 		if ((info->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL)
596 			info->rti_flags |= RTF_HOST;
597 		goto makeroute;
598 
599 	case RTM_ADD:
600 		KASSERT(!(info->rti_flags & RTF_GATEWAY) ||
601 			info->rti_info[RTAX_GATEWAY] != NULL,
602 		    ("rtrequest: GATEWAY but no gateway"));
603 
604 		if (info->rti_ifa == NULL && (error = rt_getifa(info)))
605 			gotoerr(error);
606 		ifa = info->rti_ifa;
607 makeroute:
608 		R_Malloc(rt, struct rtentry *, sizeof *rt);
609 		if (rt == NULL)
610 			gotoerr(ENOBUFS);
611 		bzero(rt, sizeof *rt);
612 		rt->rt_flags = RTF_UP | info->rti_flags;
613 		error = rt_setgate(rt, dst, info->rti_info[RTAX_GATEWAY]);
614 		if (error != 0) {
615 			Free(rt);
616 			gotoerr(error);
617 		}
618 
619 		ndst = rt_key(rt);
620 		if (info->rti_info[RTAX_NETMASK] != NULL)
621 			rt_maskedcopy(dst, ndst, info->rti_info[RTAX_NETMASK]);
622 		else
623 			bcopy(dst, ndst, dst->sa_len);
624 
625 		/*
626 		 * Note that we now have a reference to the ifa.
627 		 * This moved from below so that rnh->rnh_addaddr() can
628 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
629 		 */
630 		IFAREF(ifa);
631 		rt->rt_ifa = ifa;
632 		rt->rt_ifp = ifa->ifa_ifp;
633 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
634 
635 		rn = rnh->rnh_addaddr((char *)ndst,
636 				      (char *)info->rti_info[RTAX_NETMASK],
637 				      rnh, rt->rt_nodes);
638 		if (rn == NULL) {
639 			struct rtentry *oldrt;
640 
641 			/*
642 			 * We already have one of these in the tree.
643 			 * We do a special hack: if the old route was
644 			 * cloned, then we blow it away and try
645 			 * re-inserting the new one.
646 			 */
647 			oldrt = rtpurelookup(ndst);
648 			if (oldrt != NULL) {
649 				--oldrt->rt_refcnt;
650 				if (oldrt->rt_flags & RTF_WASCLONED) {
651 					rtrequest(RTM_DELETE, rt_key(oldrt),
652 						  oldrt->rt_gateway,
653 						  rt_mask(oldrt),
654 						  oldrt->rt_flags, NULL);
655 					rn = rnh->rnh_addaddr((char *)ndst,
656 						  (char *)
657 						  info->rti_info[RTAX_NETMASK],
658 						  rnh, rt->rt_nodes);
659 				}
660 			}
661 		}
662 
663 		/*
664 		 * If it still failed to go into the tree,
665 		 * then un-make it (this should be a function).
666 		 */
667 		if (rn == NULL) {
668 			if (rt->rt_gwroute != NULL)
669 				rtfree(rt->rt_gwroute);
670 			IFAFREE(ifa);
671 			Free(rt_key(rt));
672 			Free(rt);
673 			gotoerr(EEXIST);
674 		}
675 
676 		/*
677 		 * If we got here from RESOLVE, then we are cloning
678 		 * so clone the rest, and note that we
679 		 * are a clone (and increment the parent's references)
680 		 */
681 		if (req == RTM_RESOLVE) {
682 			rt->rt_rmx = (*ret_nrt)->rt_rmx;    /* copy metrics */
683 			rt->rt_rmx.rmx_pksent = 0;  /* reset packet counter */
684 			if ((*ret_nrt)->rt_flags &
685 				       (RTF_CLONING | RTF_PRCLONING)) {
686 				rt->rt_parent = *ret_nrt;
687 				(*ret_nrt)->rt_refcnt++;
688 			}
689 		}
690 
691 		/*
692 		 * if this protocol has something to add to this then
693 		 * allow it to do that as well.
694 		 */
695 		if (ifa->ifa_rtrequest != NULL)
696 			ifa->ifa_rtrequest(req, rt, info);
697 
698 		/*
699 		 * We repeat the same procedure from rt_setgate() here because
700 		 * it doesn't fire when we call it there because the node
701 		 * hasn't been added to the tree yet.
702 		 */
703 		if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) &&
704 		    rt_mask(rt) != NULL) {
705 			struct rtfc_arg arg = { rt, rnh };
706 
707 			rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
708 					       (char *)rt_mask(rt),
709 					       rt_fixchange, &arg);
710 		}
711 
712 		/*
713 		 * Return the resulting rtentry,
714 		 * increasing the number of references by one.
715 		 */
716 		if (ret_nrt != NULL) {
717 			rt->rt_refcnt++;
718 			*ret_nrt = rt;
719 		}
720 		break;
721 	default:
722 		error = EOPNOTSUPP;
723 	}
724 bad:
725 	splx(s);
726 	return (error);
727 }
728 
729 /*
730  * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family''
731  * (i.e., the routes related to it by the operation of cloning).  This
732  * routine is iterated over all potential former-child-routes by way of
733  * rnh->rnh_walktree_from() above, and those that actually are children of
734  * the late parent (passed in as VP here) are themselves deleted.
735  */
736 static int
737 rt_fixdelete(struct radix_node *rn, void *vp)
738 {
739 	struct rtentry *rt = (struct rtentry *)rn;
740 	struct rtentry *rt0 = vp;
741 
742 	if (rt->rt_parent == rt0 &&
743 	    !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
744 		return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
745 				 rt->rt_flags, NULL);
746 	}
747 	return 0;
748 }
749 
750 /*
751  * This routine is called from rt_setgate() to do the analogous thing for
752  * adds and changes.  There is the added complication in this case of a
753  * middle insert; i.e., insertion of a new network route between an older
754  * network route and (cloned) host routes.  For this reason, a simple check
755  * of rt->rt_parent is insufficient; each candidate route must be tested
756  * against the (mask, value) of the new route (passed as before in vp)
757  * to see if the new route matches it.
758  *
759  * XXX - it may be possible to do fixdelete() for changes and reserve this
760  * routine just for adds.  I'm not sure why I thought it was necessary to do
761  * changes this way.
762  */
763 #ifdef DEBUG
764 static int rtfcdebug = 0;
765 #endif
766 
767 static int
768 rt_fixchange(struct radix_node *rn, void *vp)
769 {
770 	struct rtentry *rt = (struct rtentry *)rn;
771 	struct rtfc_arg *ap = vp;
772 	struct rtentry *rt0 = ap->rt0;
773 	struct radix_node_head *rnh = ap->rnh;
774 	u_char *xk1, *xm1, *xk2, *xmp;
775 	int i, len, mlen;
776 
777 #ifdef DEBUG
778 	if (rtfcdebug)
779 		printf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0);
780 #endif
781 
782 	if (rt->rt_parent == NULL ||
783 	    (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
784 #ifdef DEBUG
785 		if (rtfcdebug) printf("no parent, pinned or cloning\n");
786 #endif
787 		return 0;
788 	}
789 
790 	if (rt->rt_parent == rt0) {
791 #ifdef DEBUG
792 		if (rtfcdebug) printf("parent match\n");
793 #endif
794 		return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
795 				 rt->rt_flags, NULL);
796 	}
797 
798 	/*
799 	 * There probably is a function somewhere which does this...
800 	 * if not, there should be.
801 	 */
802 	len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len);
803 
804 	xk1 = (u_char *)rt_key(rt0);
805 	xm1 = (u_char *)rt_mask(rt0);
806 	xk2 = (u_char *)rt_key(rt);
807 
808 	/* avoid applying a less specific route */
809 	xmp = (u_char *)rt_mask(rt->rt_parent);
810 	mlen = rt_key(rt->rt_parent)->sa_len;
811 	if (mlen > rt_key(rt0)->sa_len) {
812 #ifdef DEBUG
813 		if (rtfcdebug)
814 			printf("rt_fixchange: inserting a less "
815 			       "specific route\n");
816 #endif
817 		return 0;
818 	}
819 	for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) {
820 		if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) {
821 #ifdef DEBUG
822 			if (rtfcdebug)
823 				printf("rt_fixchange: inserting a less "
824 				       "specific route\n");
825 #endif
826 			return 0;
827 		}
828 	}
829 
830 	for (i = rnh->rnh_treetop->rn_offset; i < len; i++) {
831 		if ((xk2[i] & xm1[i]) != xk1[i]) {
832 #ifdef DEBUG
833 			if (rtfcdebug) printf("no match\n");
834 #endif
835 			return 0;
836 		}
837 	}
838 
839 	/*
840 	 * OK, this node is a clone, and matches the node currently being
841 	 * changed/added under the node's mask.  So, get rid of it.
842 	 */
843 #ifdef DEBUG
844 	if (rtfcdebug) printf("deleting\n");
845 #endif
846 	return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
847 			 rt->rt_flags, NULL);
848 }
849 
850 #define ROUNDUP(a) (a>0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))
851 
852 int
853 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate)
854 {
855 	char *space, *oldspace;
856 	int dlen = ROUNDUP(dst->sa_len), glen = ROUNDUP(gate->sa_len);
857 	struct rtentry *rt = rt0;
858 	struct radix_node_head *rnh = rt_tables[dst->sa_family];
859 
860 	/*
861 	 * A host route with the destination equal to the gateway
862 	 * will interfere with keeping LLINFO in the routing
863 	 * table, so disallow it.
864 	 */
865 	if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) ==
866 			      (RTF_HOST | RTF_GATEWAY)) &&
867 	    dst->sa_len == gate->sa_len &&
868 	    sa_equal(dst, gate)) {
869 		/*
870 		 * The route might already exist if this is an RTM_CHANGE
871 		 * or a routing redirect, so try to delete it.
872 		 */
873 		if (rt_key(rt0) != NULL)
874 			rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway,
875 				  rt_mask(rt0), rt0->rt_flags, NULL);
876 		return EADDRNOTAVAIL;
877 	}
878 
879 	/*
880 	 * Both dst and gateway are stored in the same malloc'ed chunk
881 	 * (If I ever get my hands on....)
882 	 * if we need to malloc a new chunk, then keep the old one around
883 	 * till we don't need it any more.
884 	 */
885 	if (rt->rt_gateway == NULL || glen > ROUNDUP(rt->rt_gateway->sa_len)) {
886 		oldspace = (char *)rt_key(rt);
887 		R_Malloc(space, char *, dlen + glen);
888 		if (space == NULL)
889 			return ENOBUFS;
890 		rt->rt_nodes->rn_key = space;
891 	} else {
892 		space = (char *)rt_key(rt);	/* Just use the old space. */
893 		oldspace = NULL;
894 	}
895 
896 	/* Set the gateway value. */
897 	rt->rt_gateway = (struct sockaddr *)(space + dlen);
898 	bcopy(gate, rt->rt_gateway, glen);
899 
900 	if (oldspace != NULL) {
901 		/*
902 		 * If we allocated a new chunk, preserve the original dst.
903 		 * This way, rt_setgate() really just sets the gate
904 		 * and leaves the dst field alone.
905 		 */
906 		bcopy(dst, space, dlen);
907 		Free(oldspace);
908 	}
909 
910 	/*
911 	 * If there is already a gwroute, it's now almost definitely wrong
912 	 * so drop it.
913 	 */
914 	if (rt->rt_gwroute != NULL) {
915 		RTFREE(rt->rt_gwroute);
916 		rt->rt_gwroute = NULL;
917 	}
918 	if (rt->rt_flags & RTF_GATEWAY) {
919 		/*
920 		 * Cloning loop avoidance: In the presence of
921 		 * protocol-cloning and bad configuration, it is
922 		 * possible to get stuck in bottomless mutual recursion
923 		 * (rtrequest rt_setgate rtlookup).  We avoid this
924 		 * by not allowing protocol-cloning to operate for
925 		 * gateways (which is probably the correct choice
926 		 * anyway), and avoid the resulting reference loops
927 		 * by disallowing any route to run through itself as
928 		 * a gateway.  This is obviously mandatory when we
929 		 * get rt->rt_output().
930 		 *
931 		 * This breaks TTCP for hosts outside the gateway!  XXX JH
932 		 */
933 		rt->rt_gwroute = _rtlookup(gate, RTL_REPORTMSG, RTF_PRCLONING);
934 		if (rt->rt_gwroute == rt) {
935 			rt->rt_gwroute = NULL;
936 			--rt->rt_refcnt;
937 			return EDQUOT; /* failure */
938 		}
939 	}
940 
941 	/*
942 	 * This isn't going to do anything useful for host routes, so
943 	 * don't bother.  Also make sure we have a reasonable mask
944 	 * (we don't yet have one during adds).
945 	 */
946 	if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) {
947 		struct rtfc_arg arg = { rt, rnh };
948 
949 		rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
950 				       (char *)rt_mask(rt),
951 				       rt_fixchange, &arg);
952 	}
953 
954 	return 0;
955 }
956 
957 static void
958 rt_maskedcopy(
959 	struct sockaddr *src,
960 	struct sockaddr *dst,
961 	struct sockaddr *netmask)
962 {
963 	u_char *cp1 = (u_char *)src;
964 	u_char *cp2 = (u_char *)dst;
965 	u_char *cp3 = (u_char *)netmask;
966 	u_char *cplim = cp2 + *cp3;
967 	u_char *cplim2 = cp2 + *cp1;
968 
969 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
970 	cp3 += 2;
971 	if (cplim > cplim2)
972 		cplim = cplim2;
973 	while (cp2 < cplim)
974 		*cp2++ = *cp1++ & *cp3++;
975 	if (cp2 < cplim2)
976 		bzero(cp2, cplim2 - cp2);
977 }
978 
979 int
980 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt)
981 {
982 	struct rtentry *up_rt, *rt;
983 
984 	if (!(rt0->rt_flags & RTF_UP)) {
985 		up_rt = rtlookup(dst);
986 		if (up_rt == NULL)
987 			return (EHOSTUNREACH);
988 		up_rt->rt_refcnt--;
989 	} else
990 		up_rt = rt0;
991 	if (up_rt->rt_flags & RTF_GATEWAY) {
992 		if (up_rt->rt_gwroute == NULL) {
993 			up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
994 			if (up_rt->rt_gwroute == NULL)
995 				return (EHOSTUNREACH);
996 		} else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) {
997 			rtfree(up_rt->rt_gwroute);
998 			up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
999 			if (up_rt->rt_gwroute == NULL)
1000 				return (EHOSTUNREACH);
1001 		}
1002 		rt = up_rt->rt_gwroute;
1003 	} else
1004 		rt = up_rt;
1005 	if (rt->rt_flags & RTF_REJECT &&
1006 	    (rt->rt_rmx.rmx_expire == 0 ||		/* rt doesn't expire */
1007 	     time_second < rt->rt_rmx.rmx_expire))	/* rt not expired */
1008 		return (rt->rt_flags & RTF_HOST ?  EHOSTDOWN : EHOSTUNREACH);
1009 	*drt = rt;
1010 	return 0;
1011 }
1012 
1013 /*
1014  * Set up a routing table entry, normally
1015  * for an interface.
1016  */
1017 int
1018 rtinit(struct ifaddr *ifa, int cmd, int flags)
1019 {
1020 	struct sockaddr *dst, *deldst, *netmask;
1021 	struct rtentry *rt;
1022 	struct rtentry *nrt = NULL;
1023 	struct mbuf *m = NULL;
1024 	struct radix_node_head *rnh;
1025 	struct radix_node *rn;
1026 	struct rt_addrinfo info;
1027 	int error;
1028 
1029 	if (flags & RTF_HOST) {
1030 		dst = ifa->ifa_dstaddr;
1031 		netmask = NULL;
1032 	} else {
1033 		dst = ifa->ifa_addr;
1034 		netmask = ifa->ifa_netmask;
1035 	}
1036 	/*
1037 	 * If it's a delete, check that if it exists, it's on the correct
1038 	 * interface or we might scrub a route to another ifa which would
1039 	 * be confusing at best and possibly worse.
1040 	 */
1041 	if (cmd == RTM_DELETE) {
1042 		/*
1043 		 * It's a delete, so it should already exist..
1044 		 * If it's a net, mask off the host bits
1045 		 * (Assuming we have a mask)
1046 		 */
1047 		if (netmask != NULL) {
1048 			m = m_get(MB_DONTWAIT, MT_SONAME);
1049 			if (m == NULL)
1050 				return (ENOBUFS);
1051 			deldst = mtod(m, struct sockaddr *);
1052 			rt_maskedcopy(dst, deldst, netmask);
1053 			dst = deldst;
1054 		}
1055 		/*
1056 		 * Look up an rtentry that is in the routing tree and
1057 		 * contains the correct info.
1058 		 */
1059 		if ((rnh = rt_tables[dst->sa_family]) == NULL ||
1060 		    (rn = rnh->rnh_lookup((char *)dst,
1061 					  (char *)netmask, rnh)) == NULL ||
1062 		    ((struct rtentry *)rn)->rt_ifa != ifa ||
1063 		    !sa_equal((struct sockaddr *)rn->rn_key, dst)) {
1064 			if (m != NULL)
1065 				m_free(m);
1066 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1067 		}
1068 		/* XXX */
1069 #if 0
1070 		else {
1071 			/*
1072 			 * One would think that as we are deleting, and we know
1073 			 * it doesn't exist, we could just return at this point
1074 			 * with an "ELSE" clause, but apparently not..
1075 			 */
1076 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1077 		}
1078 #endif
1079 	}
1080 	/*
1081 	 * Do the actual request
1082 	 */
1083 	bzero(&info, sizeof info);
1084 	info.rti_ifa = ifa;
1085 	info.rti_flags = flags | ifa->ifa_flags;
1086 	info.rti_info[RTAX_DST] = dst;
1087 	info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1088 	info.rti_info[RTAX_NETMASK] = netmask;
1089 	error = rtrequest1(cmd, &info, &nrt);
1090 	if (error == 0 && (rt = nrt) != NULL) {
1091 		/*
1092 		 * notify any listening routing agents of the change
1093 		 */
1094 		rt_newaddrmsg(cmd, ifa, error, rt);
1095 		if (cmd == RTM_DELETE) {
1096 			/*
1097 			 * If we are deleting, and we found an entry, then
1098 			 * it's been removed from the tree.. now throw it away.
1099 			 */
1100 			if (rt->rt_refcnt <= 0) {
1101 				rt->rt_refcnt++; /* make a 1->0 transition */
1102 				rtfree(rt);
1103 			}
1104 		} else if (cmd == RTM_ADD) {
1105 			/*
1106 			 * We just wanted to add it.. we don't actually
1107 			 * need a reference.
1108 			 */
1109 			rt->rt_refcnt--;
1110 		}
1111 	}
1112 	if (m != NULL)
1113 		m_free(m);
1114 	return (error);
1115 }
1116 
1117 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */
1118 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
1119