xref: /dflybsd-src/sys/net/route.c (revision 10cf3cb7c694a65a355fa433dae9cbace95a62a2)
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) 1980, 1986, 1991, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. All advertising materials mentioning features or use of this software
46  *    must display the following acknowledgement:
47  *	This product includes software developed by the University of
48  *	California, Berkeley and its contributors.
49  * 4. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)route.c	8.3 (Berkeley) 1/9/95
66  * $FreeBSD: src/sys/net/route.c,v 1.59.2.10 2003/01/17 08:04:00 ru Exp $
67  * $DragonFly: src/sys/net/route.c,v 1.41 2008/11/09 10:50:15 sephe Exp $
68  */
69 
70 #include "opt_inet.h"
71 #include "opt_mpls.h"
72 
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/malloc.h>
76 #include <sys/mbuf.h>
77 #include <sys/socket.h>
78 #include <sys/domain.h>
79 #include <sys/kernel.h>
80 #include <sys/sysctl.h>
81 #include <sys/globaldata.h>
82 #include <sys/thread.h>
83 
84 #include <net/if.h>
85 #include <net/route.h>
86 #include <net/netisr.h>
87 
88 #include <netinet/in.h>
89 #include <net/ip_mroute/ip_mroute.h>
90 
91 #include <sys/thread2.h>
92 #include <sys/msgport2.h>
93 #include <net/netmsg2.h>
94 
95 #ifdef MPLS
96 #include <netproto/mpls/mpls.h>
97 #endif
98 
99 static struct rtstatistics rtstatistics_percpu[MAXCPU];
100 #ifdef SMP
101 #define rtstat	rtstatistics_percpu[mycpuid]
102 #else
103 #define rtstat	rtstatistics_percpu[0]
104 #endif
105 
106 struct radix_node_head *rt_tables[MAXCPU][AF_MAX+1];
107 struct lwkt_port *rt_ports[MAXCPU];
108 
109 static void	rt_maskedcopy (struct sockaddr *, struct sockaddr *,
110 			       struct sockaddr *);
111 static void rtable_init(void);
112 static void rtable_service_loop(void *dummy);
113 static void rtinit_rtrequest_callback(int, int, struct rt_addrinfo *,
114 				      struct rtentry *, void *);
115 
116 #ifdef SMP
117 static void rtredirect_msghandler(struct netmsg *netmsg);
118 static void rtrequest1_msghandler(struct netmsg *netmsg);
119 #endif
120 
121 static int rt_setshims(struct rtentry *, struct sockaddr **);
122 
123 SYSCTL_NODE(_net, OID_AUTO, route, CTLFLAG_RW, 0, "Routing");
124 
125 #ifdef ROUTE_DEBUG
126 static int route_debug = 1;
127 SYSCTL_INT(_net_route, OID_AUTO, route_debug, CTLFLAG_RW,
128            &route_debug, 0, "");
129 #endif
130 
131 int route_assert_owner_access = 0;
132 SYSCTL_INT(_net_route, OID_AUTO, assert_owner_access, CTLFLAG_RW,
133            &route_assert_owner_access, 0, "");
134 
135 /*
136  * Initialize the route table(s) for protocol domains and
137  * create a helper thread which will be responsible for updating
138  * route table entries on each cpu.
139  */
140 void
141 route_init(void)
142 {
143 	int cpu;
144 	thread_t rtd;
145 
146 	for (cpu = 0; cpu < ncpus; ++cpu)
147 		bzero(&rtstatistics_percpu[cpu], sizeof(struct rtstatistics));
148 	rn_init();      /* initialize all zeroes, all ones, mask table */
149 	rtable_init();	/* call dom_rtattach() on each cpu */
150 
151 	for (cpu = 0; cpu < ncpus; cpu++) {
152 		lwkt_create(rtable_service_loop, NULL, &rtd, NULL,
153 			    0, cpu, "rtable_cpu %d", cpu);
154 		rt_ports[cpu] = &rtd->td_msgport;
155 	}
156 }
157 
158 static void
159 rtable_init_oncpu(struct netmsg *nmsg)
160 {
161 	struct domain *dom;
162 	int cpu = mycpuid;
163 
164 	SLIST_FOREACH(dom, &domains, dom_next) {
165 		if (dom->dom_rtattach) {
166 			dom->dom_rtattach(
167 				(void **)&rt_tables[cpu][dom->dom_family],
168 			        dom->dom_rtoffset);
169 		}
170 	}
171 	ifnet_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
172 }
173 
174 static void
175 rtable_init(void)
176 {
177 	struct netmsg nmsg;
178 
179 	netmsg_init(&nmsg, &curthread->td_msgport, 0, rtable_init_oncpu);
180 	ifnet_domsg(&nmsg.nm_lmsg, 0);
181 }
182 
183 /*
184  * Our per-cpu table management protocol thread.  All route table operations
185  * are sequentially chained through all cpus starting at cpu #0 in order to
186  * maintain duplicate route tables on each cpu.  Having a spearate route
187  * table management thread allows the protocol and interrupt threads to
188  * issue route table changes.
189  */
190 static void
191 rtable_service_loop(void *dummy __unused)
192 {
193 	struct netmsg *netmsg;
194 	thread_t td = curthread;
195 
196 	while ((netmsg = lwkt_waitport(&td->td_msgport, 0)) != NULL) {
197 		netmsg->nm_dispatch(netmsg);
198 	}
199 }
200 
201 /*
202  * Routing statistics.
203  */
204 #ifdef SMP
205 static int
206 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS)
207 {
208 	int cpu, error = 0;
209 
210 	for (cpu = 0; cpu < ncpus; ++cpu) {
211 		if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu],
212 					sizeof(struct rtstatistics))))
213 				break;
214 		if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu],
215 					sizeof(struct rtstatistics))))
216 				break;
217 	}
218 
219 	return (error);
220 }
221 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW),
222 	0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics");
223 #else
224 SYSCTL_STRUCT(_net_route, OID_AUTO, stats, CTLFLAG_RW, &rtstat, rtstatistics,
225 "Routing statistics");
226 #endif
227 
228 /*
229  * Packet routing routines.
230  */
231 
232 /*
233  * Look up and fill in the "ro_rt" rtentry field in a route structure given
234  * an address in the "ro_dst" field.  Always send a report on a miss and
235  * always clone routes.
236  */
237 void
238 rtalloc(struct route *ro)
239 {
240 	rtalloc_ign(ro, 0UL);
241 }
242 
243 /*
244  * Look up and fill in the "ro_rt" rtentry field in a route structure given
245  * an address in the "ro_dst" field.  Always send a report on a miss and
246  * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being
247  * ignored.
248  */
249 void
250 rtalloc_ign(struct route *ro, u_long ignoreflags)
251 {
252 	if (ro->ro_rt != NULL) {
253 		if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP)
254 			return;
255 		rtfree(ro->ro_rt);
256 		ro->ro_rt = NULL;
257 	}
258 	ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags);
259 }
260 
261 /*
262  * Look up the route that matches the given "dst" address.
263  *
264  * Route lookup can have the side-effect of creating and returning
265  * a cloned route instead when "dst" matches a cloning route and the
266  * RTF_CLONING and RTF_PRCLONING flags are not being ignored.
267  *
268  * Any route returned has its reference count incremented.
269  */
270 struct rtentry *
271 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore)
272 {
273 	struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family];
274 	struct rtentry *rt;
275 
276 	if (rnh == NULL)
277 		goto unreach;
278 
279 	/*
280 	 * Look up route in the radix tree.
281 	 */
282 	rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh);
283 	if (rt == NULL)
284 		goto unreach;
285 
286 	/*
287 	 * Handle cloning routes.
288 	 */
289 	if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) {
290 		struct rtentry *clonedroute;
291 		int error;
292 
293 		clonedroute = rt;	/* copy in/copy out parameter */
294 		error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0,
295 				  &clonedroute);	/* clone the route */
296 		if (error != 0) {	/* cloning failed */
297 			if (generate_report)
298 				rt_dstmsg(RTM_MISS, dst, error);
299 			rt->rt_refcnt++;
300 			return (rt);	/* return the uncloned route */
301 		}
302 		if (generate_report) {
303 			if (clonedroute->rt_flags & RTF_XRESOLVE)
304 				rt_dstmsg(RTM_RESOLVE, dst, 0);
305 			else
306 				rt_rtmsg(RTM_ADD, clonedroute,
307 					 clonedroute->rt_ifp, 0);
308 		}
309 		return (clonedroute);	/* return cloned route */
310 	}
311 
312 	/*
313 	 * Increment the reference count of the matched route and return.
314 	 */
315 	rt->rt_refcnt++;
316 	return (rt);
317 
318 unreach:
319 	rtstat.rts_unreach++;
320 	if (generate_report)
321 		rt_dstmsg(RTM_MISS, dst, 0);
322 	return (NULL);
323 }
324 
325 void
326 rtfree(struct rtentry *rt)
327 {
328 	if (rt->rt_cpuid == mycpuid)
329 		rtfree_oncpu(rt);
330 	else
331 		rtfree_remote(rt, 1);
332 }
333 
334 void
335 rtfree_oncpu(struct rtentry *rt)
336 {
337 	KKASSERT(rt->rt_cpuid == mycpuid);
338 	KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt));
339 
340 	--rt->rt_refcnt;
341 	if (rt->rt_refcnt == 0) {
342 		struct radix_node_head *rnh =
343 		    rt_tables[mycpuid][rt_key(rt)->sa_family];
344 
345 		if (rnh->rnh_close)
346 			rnh->rnh_close((struct radix_node *)rt, rnh);
347 		if (!(rt->rt_flags & RTF_UP)) {
348 			/* deallocate route */
349 			if (rt->rt_ifa != NULL)
350 				IFAFREE(rt->rt_ifa);
351 			if (rt->rt_parent != NULL)
352 				RTFREE(rt->rt_parent);	/* recursive call! */
353 			Free(rt_key(rt));
354 			Free(rt);
355 		}
356 	}
357 }
358 
359 static void
360 rtfree_remote_dispatch(struct netmsg *nmsg)
361 {
362 	struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
363 	struct rtentry *rt = lmsg->u.ms_resultp;
364 
365 	rtfree_oncpu(rt);
366 	lwkt_replymsg(lmsg, 0);
367 }
368 
369 void
370 rtfree_remote(struct rtentry *rt, int allow_panic)
371 {
372 	struct netmsg nmsg;
373 	struct lwkt_msg *lmsg;
374 
375 	KKASSERT(rt->rt_cpuid != mycpuid);
376 
377 	if (route_assert_owner_access && allow_panic) {
378 		panic("rt remote free rt_cpuid %d, mycpuid %d\n",
379 		      rt->rt_cpuid, mycpuid);
380 	} else {
381 		kprintf("rt remote free rt_cpuid %d, mycpuid %d\n",
382 			rt->rt_cpuid, mycpuid);
383 		backtrace();
384 	}
385 
386 	netmsg_init(&nmsg, &curthread->td_msgport, 0, rtfree_remote_dispatch);
387 	lmsg = &nmsg.nm_lmsg;
388 	lmsg->u.ms_resultp = rt;
389 
390 	lwkt_domsg(rtable_portfn(rt->rt_cpuid), lmsg, 0);
391 }
392 
393 static int
394 rtredirect_oncpu(struct sockaddr *dst, struct sockaddr *gateway,
395 		 struct sockaddr *netmask, int flags, struct sockaddr *src)
396 {
397 	struct rtentry *rt = NULL;
398 	struct rt_addrinfo rtinfo;
399 	struct ifaddr *ifa;
400 	u_long *stat = NULL;
401 	int error;
402 
403 	/* verify the gateway is directly reachable */
404 	if ((ifa = ifa_ifwithnet(gateway)) == NULL) {
405 		error = ENETUNREACH;
406 		goto out;
407 	}
408 
409 	/*
410 	 * If the redirect isn't from our current router for this destination,
411 	 * it's either old or wrong.
412 	 */
413 	if (!(flags & RTF_DONE) &&		/* XXX JH */
414 	    (rt = rtpurelookup(dst)) != NULL &&
415 	    (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) {
416 		error = EINVAL;
417 		goto done;
418 	}
419 
420 	/*
421 	 * If it redirects us to ourselves, we have a routing loop,
422 	 * perhaps as a result of an interface going down recently.
423 	 */
424 	if (ifa_ifwithaddr(gateway)) {
425 		error = EHOSTUNREACH;
426 		goto done;
427 	}
428 
429 	/*
430 	 * Create a new entry if the lookup failed or if we got back
431 	 * a wildcard entry for the default route.  This is necessary
432 	 * for hosts which use routing redirects generated by smart
433 	 * gateways to dynamically build the routing tables.
434 	 */
435 	if (rt == NULL)
436 		goto create;
437 	if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) {
438 		rtfree(rt);
439 		goto create;
440 	}
441 
442 	/* Ignore redirects for directly connected hosts. */
443 	if (!(rt->rt_flags & RTF_GATEWAY)) {
444 		error = EHOSTUNREACH;
445 		goto done;
446 	}
447 
448 	if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) {
449 		/*
450 		 * Changing from a network route to a host route.
451 		 * Create a new host route rather than smashing the
452 		 * network route.
453 		 */
454 create:
455 		flags |=  RTF_GATEWAY | RTF_DYNAMIC;
456 		bzero(&rtinfo, sizeof(struct rt_addrinfo));
457 		rtinfo.rti_info[RTAX_DST] = dst;
458 		rtinfo.rti_info[RTAX_GATEWAY] = gateway;
459 		rtinfo.rti_info[RTAX_NETMASK] = netmask;
460 		rtinfo.rti_flags = flags;
461 		rtinfo.rti_ifa = ifa;
462 		rt = NULL;	/* copy-in/copy-out parameter */
463 		error = rtrequest1(RTM_ADD, &rtinfo, &rt);
464 		if (rt != NULL)
465 			flags = rt->rt_flags;
466 		stat = &rtstat.rts_dynamic;
467 	} else {
468 		/*
469 		 * Smash the current notion of the gateway to this destination.
470 		 * Should check about netmask!!!
471 		 */
472 		rt->rt_flags |= RTF_MODIFIED;
473 		flags |= RTF_MODIFIED;
474 		rt_setgate(rt, rt_key(rt), gateway);
475 		error = 0;
476 		stat = &rtstat.rts_newgateway;
477 	}
478 
479 done:
480 	if (rt != NULL)
481 		rtfree(rt);
482 out:
483 	if (error != 0)
484 		rtstat.rts_badredirect++;
485 	else if (stat != NULL)
486 		(*stat)++;
487 
488 	return error;
489 }
490 
491 #ifdef SMP
492 
493 struct netmsg_rtredirect {
494 	struct netmsg	netmsg;
495 	struct sockaddr *dst;
496 	struct sockaddr *gateway;
497 	struct sockaddr *netmask;
498 	int		flags;
499 	struct sockaddr *src;
500 };
501 
502 #endif
503 
504 /*
505  * Force a routing table entry to the specified
506  * destination to go through the given gateway.
507  * Normally called as a result of a routing redirect
508  * message from the network layer.
509  *
510  * N.B.: must be called at splnet
511  */
512 void
513 rtredirect(struct sockaddr *dst, struct sockaddr *gateway,
514 	   struct sockaddr *netmask, int flags, struct sockaddr *src)
515 {
516 	struct rt_addrinfo rtinfo;
517 	int error;
518 #ifdef SMP
519 	struct netmsg_rtredirect msg;
520 
521 	netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
522 		    rtredirect_msghandler);
523 	msg.dst = dst;
524 	msg.gateway = gateway;
525 	msg.netmask = netmask;
526 	msg.flags = flags;
527 	msg.src = src;
528 	error = lwkt_domsg(rtable_portfn(0), &msg.netmsg.nm_lmsg, 0);
529 #else
530 	error = rtredirect_oncpu(dst, gateway, netmask, flags, src);
531 #endif
532 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
533 	rtinfo.rti_info[RTAX_DST] = dst;
534 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
535 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
536 	rtinfo.rti_info[RTAX_AUTHOR] = src;
537 	rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error);
538 }
539 
540 #ifdef SMP
541 
542 static void
543 rtredirect_msghandler(struct netmsg *netmsg)
544 {
545 	struct netmsg_rtredirect *msg = (void *)netmsg;
546 	int nextcpu;
547 
548 	rtredirect_oncpu(msg->dst, msg->gateway, msg->netmask,
549 			 msg->flags, msg->src);
550 	nextcpu = mycpuid + 1;
551 	if (nextcpu < ncpus)
552 		lwkt_forwardmsg(rtable_portfn(nextcpu), &netmsg->nm_lmsg);
553 	else
554 		lwkt_replymsg(&netmsg->nm_lmsg, 0);
555 }
556 
557 #endif
558 
559 /*
560 * Routing table ioctl interface.
561 */
562 int
563 rtioctl(u_long req, caddr_t data, struct ucred *cred)
564 {
565 #ifdef INET
566 	/* Multicast goop, grrr... */
567 	return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP;
568 #else
569 	return ENXIO;
570 #endif
571 }
572 
573 struct ifaddr *
574 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
575 {
576 	struct ifaddr *ifa;
577 
578 	if (!(flags & RTF_GATEWAY)) {
579 		/*
580 		 * If we are adding a route to an interface,
581 		 * and the interface is a point-to-point link,
582 		 * we should search for the destination
583 		 * as our clue to the interface.  Otherwise
584 		 * we can use the local address.
585 		 */
586 		ifa = NULL;
587 		if (flags & RTF_HOST) {
588 			ifa = ifa_ifwithdstaddr(dst);
589 		}
590 		if (ifa == NULL)
591 			ifa = ifa_ifwithaddr(gateway);
592 	} else {
593 		/*
594 		 * If we are adding a route to a remote net
595 		 * or host, the gateway may still be on the
596 		 * other end of a pt to pt link.
597 		 */
598 		ifa = ifa_ifwithdstaddr(gateway);
599 	}
600 	if (ifa == NULL)
601 		ifa = ifa_ifwithnet(gateway);
602 	if (ifa == NULL) {
603 		struct rtentry *rt;
604 
605 		rt = rtpurelookup(gateway);
606 		if (rt == NULL)
607 			return (NULL);
608 		rt->rt_refcnt--;
609 		if ((ifa = rt->rt_ifa) == NULL)
610 			return (NULL);
611 	}
612 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
613 		struct ifaddr *oldifa = ifa;
614 
615 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
616 		if (ifa == NULL)
617 			ifa = oldifa;
618 	}
619 	return (ifa);
620 }
621 
622 static int rt_fixdelete (struct radix_node *, void *);
623 static int rt_fixchange (struct radix_node *, void *);
624 
625 struct rtfc_arg {
626 	struct rtentry *rt0;
627 	struct radix_node_head *rnh;
628 };
629 
630 /*
631  * Set rtinfo->rti_ifa and rtinfo->rti_ifp.
632  */
633 int
634 rt_getifa(struct rt_addrinfo *rtinfo)
635 {
636 	struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY];
637 	struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
638 	struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA];
639 	int flags = rtinfo->rti_flags;
640 
641 	/*
642 	 * ifp may be specified by sockaddr_dl
643 	 * when protocol address is ambiguous.
644 	 */
645 	if (rtinfo->rti_ifp == NULL) {
646 		struct sockaddr *ifpaddr;
647 
648 		ifpaddr = rtinfo->rti_info[RTAX_IFP];
649 		if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) {
650 			struct ifaddr *ifa;
651 
652 			ifa = ifa_ifwithnet(ifpaddr);
653 			if (ifa != NULL)
654 				rtinfo->rti_ifp = ifa->ifa_ifp;
655 		}
656 	}
657 
658 	if (rtinfo->rti_ifa == NULL && ifaaddr != NULL)
659 		rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr);
660 	if (rtinfo->rti_ifa == NULL) {
661 		struct sockaddr *sa;
662 
663 		sa = ifaaddr != NULL ? ifaaddr :
664 		    (gateway != NULL ? gateway : dst);
665 		if (sa != NULL && rtinfo->rti_ifp != NULL)
666 			rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp);
667 		else if (dst != NULL && gateway != NULL)
668 			rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway);
669 		else if (sa != NULL)
670 			rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa);
671 	}
672 	if (rtinfo->rti_ifa == NULL)
673 		return (ENETUNREACH);
674 
675 	if (rtinfo->rti_ifp == NULL)
676 		rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp;
677 	return (0);
678 }
679 
680 /*
681  * Do appropriate manipulations of a routing tree given
682  * all the bits of info needed
683  */
684 int
685 rtrequest(
686 	int req,
687 	struct sockaddr *dst,
688 	struct sockaddr *gateway,
689 	struct sockaddr *netmask,
690 	int flags,
691 	struct rtentry **ret_nrt)
692 {
693 	struct rt_addrinfo rtinfo;
694 
695 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
696 	rtinfo.rti_info[RTAX_DST] = dst;
697 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
698 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
699 	rtinfo.rti_flags = flags;
700 	return rtrequest1(req, &rtinfo, ret_nrt);
701 }
702 
703 int
704 rtrequest_global(
705 	int req,
706 	struct sockaddr *dst,
707 	struct sockaddr *gateway,
708 	struct sockaddr *netmask,
709 	int flags)
710 {
711 	struct rt_addrinfo rtinfo;
712 
713 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
714 	rtinfo.rti_info[RTAX_DST] = dst;
715 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
716 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
717 	rtinfo.rti_flags = flags;
718 	return rtrequest1_global(req, &rtinfo, NULL, NULL);
719 }
720 
721 #ifdef SMP
722 
723 struct netmsg_rtq {
724 	struct netmsg		netmsg;
725 	int			req;
726 	struct rt_addrinfo	*rtinfo;
727 	rtrequest1_callback_func_t callback;
728 	void			*arg;
729 };
730 
731 #endif
732 
733 int
734 rtrequest1_global(int req, struct rt_addrinfo *rtinfo,
735 		  rtrequest1_callback_func_t callback, void *arg)
736 {
737 	int error;
738 #ifdef SMP
739 	struct netmsg_rtq msg;
740 
741 	netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
742 		    rtrequest1_msghandler);
743 	msg.netmsg.nm_lmsg.ms_error = -1;
744 	msg.req = req;
745 	msg.rtinfo = rtinfo;
746 	msg.callback = callback;
747 	msg.arg = arg;
748 	error = lwkt_domsg(rtable_portfn(0), &msg.netmsg.nm_lmsg, 0);
749 #else
750 	struct rtentry *rt = NULL;
751 
752 	error = rtrequest1(req, rtinfo, &rt);
753 	if (rt)
754 		--rt->rt_refcnt;
755 	if (callback)
756 		callback(req, error, rtinfo, rt, arg);
757 #endif
758 	return (error);
759 }
760 
761 /*
762  * Handle a route table request on the current cpu.  Since the route table's
763  * are supposed to be identical on each cpu, an error occuring later in the
764  * message chain is considered system-fatal.
765  */
766 #ifdef SMP
767 
768 static void
769 rtrequest1_msghandler(struct netmsg *netmsg)
770 {
771 	struct netmsg_rtq *msg = (void *)netmsg;
772 	struct rtentry *rt = NULL;
773 	int nextcpu;
774 	int error;
775 
776 	error = rtrequest1(msg->req, msg->rtinfo, &rt);
777 	if (rt)
778 		--rt->rt_refcnt;
779 	if (msg->callback)
780 		msg->callback(msg->req, error, msg->rtinfo, rt, msg->arg);
781 
782 	/*
783 	 * RTM_DELETE's are propogated even if an error occurs, since a
784 	 * cloned route might be undergoing deletion and cloned routes
785 	 * are not necessarily replicated.  An overall error is returned
786 	 * only if no cpus have the route in question.
787 	 */
788 	if (msg->netmsg.nm_lmsg.ms_error < 0 || error == 0)
789 		msg->netmsg.nm_lmsg.ms_error = error;
790 
791 	nextcpu = mycpuid + 1;
792 	if (error && msg->req != RTM_DELETE) {
793 		if (mycpuid != 0) {
794 			panic("rtrequest1_msghandler: rtrequest table "
795 			      "error was not on cpu #0: %p", msg->rtinfo);
796 		}
797 		lwkt_replymsg(&msg->netmsg.nm_lmsg, error);
798 	} else if (nextcpu < ncpus) {
799 		lwkt_forwardmsg(rtable_portfn(nextcpu), &msg->netmsg.nm_lmsg);
800 	} else {
801 		lwkt_replymsg(&msg->netmsg.nm_lmsg,
802 			      msg->netmsg.nm_lmsg.ms_error);
803 	}
804 }
805 
806 #endif
807 
808 int
809 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt)
810 {
811 	struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
812 	struct rtentry *rt;
813 	struct radix_node *rn;
814 	struct radix_node_head *rnh;
815 	struct ifaddr *ifa;
816 	struct sockaddr *ndst;
817 	int error = 0;
818 
819 #define gotoerr(x) { error = x ; goto bad; }
820 
821 #ifdef ROUTE_DEBUG
822 	if (route_debug)
823 		rt_addrinfo_print(req, rtinfo);
824 #endif
825 
826 	crit_enter();
827 	/*
828 	 * Find the correct routing tree to use for this Address Family
829 	 */
830 	if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL)
831 		gotoerr(EAFNOSUPPORT);
832 
833 	/*
834 	 * If we are adding a host route then we don't want to put
835 	 * a netmask in the tree, nor do we want to clone it.
836 	 */
837 	if (rtinfo->rti_flags & RTF_HOST) {
838 		rtinfo->rti_info[RTAX_NETMASK] = NULL;
839 		rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING);
840 	}
841 
842 	switch (req) {
843 	case RTM_DELETE:
844 		/* Remove the item from the tree. */
845 		rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST],
846 				      (char *)rtinfo->rti_info[RTAX_NETMASK],
847 				      rnh);
848 		if (rn == NULL)
849 			gotoerr(ESRCH);
850 		KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)),
851 			("rnh_deladdr returned flags 0x%x", rn->rn_flags));
852 		rt = (struct rtentry *)rn;
853 
854 		/* ref to prevent a deletion race */
855 		++rt->rt_refcnt;
856 
857 		/* Free any routes cloned from this one. */
858 		if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) &&
859 		    rt_mask(rt) != NULL) {
860 			rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
861 					       (char *)rt_mask(rt),
862 					       rt_fixdelete, rt);
863 		}
864 
865 		if (rt->rt_gwroute != NULL) {
866 			RTFREE(rt->rt_gwroute);
867 			rt->rt_gwroute = NULL;
868 		}
869 
870 		/*
871 		 * NB: RTF_UP must be set during the search above,
872 		 * because we might delete the last ref, causing
873 		 * rt to get freed prematurely.
874 		 */
875 		rt->rt_flags &= ~RTF_UP;
876 
877 #ifdef ROUTE_DEBUG
878 		if (route_debug)
879 			rt_print(rtinfo, rt);
880 #endif
881 
882 		/* Give the protocol a chance to keep things in sync. */
883 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
884 			ifa->ifa_rtrequest(RTM_DELETE, rt, rtinfo);
885 
886 		/*
887 		 * If the caller wants it, then it can have it,
888 		 * but it's up to it to free the rtentry as we won't be
889 		 * doing it.
890 		 */
891 		KASSERT(rt->rt_refcnt >= 0,
892 			("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt));
893 		if (ret_nrt != NULL) {
894 			/* leave ref intact for return */
895 			*ret_nrt = rt;
896 		} else {
897 			/* deref / attempt to destroy */
898 			rtfree(rt);
899 		}
900 		break;
901 
902 	case RTM_RESOLVE:
903 		if (ret_nrt == NULL || (rt = *ret_nrt) == NULL)
904 			gotoerr(EINVAL);
905 		ifa = rt->rt_ifa;
906 		rtinfo->rti_flags =
907 		    rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC);
908 		rtinfo->rti_flags |= RTF_WASCLONED;
909 		rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
910 		if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL)
911 			rtinfo->rti_flags |= RTF_HOST;
912 		rtinfo->rti_info[RTAX_MPLS1] = rt->rt_shim[0];
913 		rtinfo->rti_info[RTAX_MPLS2] = rt->rt_shim[1];
914 		rtinfo->rti_info[RTAX_MPLS3] = rt->rt_shim[2];
915 		goto makeroute;
916 
917 	case RTM_ADD:
918 		KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) ||
919 			rtinfo->rti_info[RTAX_GATEWAY] != NULL,
920 		    ("rtrequest: GATEWAY but no gateway"));
921 
922 		if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo)))
923 			gotoerr(error);
924 		ifa = rtinfo->rti_ifa;
925 makeroute:
926 		R_Malloc(rt, struct rtentry *, sizeof(struct rtentry));
927 		if (rt == NULL)
928 			gotoerr(ENOBUFS);
929 		bzero(rt, sizeof(struct rtentry));
930 		rt->rt_flags = RTF_UP | rtinfo->rti_flags;
931 		rt->rt_cpuid = mycpuid;
932 		error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY]);
933 		if (error != 0) {
934 			Free(rt);
935 			gotoerr(error);
936 		}
937 
938 		ndst = rt_key(rt);
939 		if (rtinfo->rti_info[RTAX_NETMASK] != NULL)
940 			rt_maskedcopy(dst, ndst,
941 				      rtinfo->rti_info[RTAX_NETMASK]);
942 		else
943 			bcopy(dst, ndst, dst->sa_len);
944 
945 		if (rtinfo->rti_info[RTAX_MPLS1] != NULL)
946 			rt_setshims(rt, rtinfo->rti_info);
947 
948 		/*
949 		 * Note that we now have a reference to the ifa.
950 		 * This moved from below so that rnh->rnh_addaddr() can
951 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
952 		 */
953 		IFAREF(ifa);
954 		rt->rt_ifa = ifa;
955 		rt->rt_ifp = ifa->ifa_ifp;
956 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
957 
958 		rn = rnh->rnh_addaddr((char *)ndst,
959 				      (char *)rtinfo->rti_info[RTAX_NETMASK],
960 				      rnh, rt->rt_nodes);
961 		if (rn == NULL) {
962 			struct rtentry *oldrt;
963 
964 			/*
965 			 * We already have one of these in the tree.
966 			 * We do a special hack: if the old route was
967 			 * cloned, then we blow it away and try
968 			 * re-inserting the new one.
969 			 */
970 			oldrt = rtpurelookup(ndst);
971 			if (oldrt != NULL) {
972 				--oldrt->rt_refcnt;
973 				if (oldrt->rt_flags & RTF_WASCLONED) {
974 					rtrequest(RTM_DELETE, rt_key(oldrt),
975 						  oldrt->rt_gateway,
976 						  rt_mask(oldrt),
977 						  oldrt->rt_flags, NULL);
978 					rn = rnh->rnh_addaddr((char *)ndst,
979 					    (char *)
980 						rtinfo->rti_info[RTAX_NETMASK],
981 					    rnh, rt->rt_nodes);
982 				}
983 			}
984 		}
985 
986 		/*
987 		 * If it still failed to go into the tree,
988 		 * then un-make it (this should be a function).
989 		 */
990 		if (rn == NULL) {
991 			if (rt->rt_gwroute != NULL)
992 				rtfree(rt->rt_gwroute);
993 			IFAFREE(ifa);
994 			Free(rt_key(rt));
995 			Free(rt);
996 			gotoerr(EEXIST);
997 		}
998 
999 		/*
1000 		 * If we got here from RESOLVE, then we are cloning
1001 		 * so clone the rest, and note that we
1002 		 * are a clone (and increment the parent's references)
1003 		 */
1004 		if (req == RTM_RESOLVE) {
1005 			rt->rt_rmx = (*ret_nrt)->rt_rmx;    /* copy metrics */
1006 			rt->rt_rmx.rmx_pksent = 0;  /* reset packet counter */
1007 			if ((*ret_nrt)->rt_flags &
1008 				       (RTF_CLONING | RTF_PRCLONING)) {
1009 				rt->rt_parent = *ret_nrt;
1010 				(*ret_nrt)->rt_refcnt++;
1011 			}
1012 		}
1013 
1014 		/*
1015 		 * if this protocol has something to add to this then
1016 		 * allow it to do that as well.
1017 		 */
1018 		if (ifa->ifa_rtrequest != NULL)
1019 			ifa->ifa_rtrequest(req, rt, rtinfo);
1020 
1021 		/*
1022 		 * We repeat the same procedure from rt_setgate() here because
1023 		 * it doesn't fire when we call it there because the node
1024 		 * hasn't been added to the tree yet.
1025 		 */
1026 		if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) &&
1027 		    rt_mask(rt) != NULL) {
1028 			struct rtfc_arg arg = { rt, rnh };
1029 
1030 			rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
1031 					       (char *)rt_mask(rt),
1032 					       rt_fixchange, &arg);
1033 		}
1034 
1035 #ifdef ROUTE_DEBUG
1036 		if (route_debug)
1037 			rt_print(rtinfo, rt);
1038 #endif
1039 		/*
1040 		 * Return the resulting rtentry,
1041 		 * increasing the number of references by one.
1042 		 */
1043 		if (ret_nrt != NULL) {
1044 			rt->rt_refcnt++;
1045 			*ret_nrt = rt;
1046 		}
1047 		break;
1048 	default:
1049 		error = EOPNOTSUPP;
1050 	}
1051 bad:
1052 #ifdef ROUTE_DEBUG
1053 	if (route_debug) {
1054 		if (error)
1055 			kprintf("rti %p failed error %d\n", rtinfo, error);
1056 		else
1057 			kprintf("rti %p succeeded\n", rtinfo);
1058 	}
1059 #endif
1060 	crit_exit();
1061 	return (error);
1062 }
1063 
1064 /*
1065  * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family''
1066  * (i.e., the routes related to it by the operation of cloning).  This
1067  * routine is iterated over all potential former-child-routes by way of
1068  * rnh->rnh_walktree_from() above, and those that actually are children of
1069  * the late parent (passed in as VP here) are themselves deleted.
1070  */
1071 static int
1072 rt_fixdelete(struct radix_node *rn, void *vp)
1073 {
1074 	struct rtentry *rt = (struct rtentry *)rn;
1075 	struct rtentry *rt0 = vp;
1076 
1077 	if (rt->rt_parent == rt0 &&
1078 	    !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
1079 		return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1080 				 rt->rt_flags, NULL);
1081 	}
1082 	return 0;
1083 }
1084 
1085 /*
1086  * This routine is called from rt_setgate() to do the analogous thing for
1087  * adds and changes.  There is the added complication in this case of a
1088  * middle insert; i.e., insertion of a new network route between an older
1089  * network route and (cloned) host routes.  For this reason, a simple check
1090  * of rt->rt_parent is insufficient; each candidate route must be tested
1091  * against the (mask, value) of the new route (passed as before in vp)
1092  * to see if the new route matches it.
1093  *
1094  * XXX - it may be possible to do fixdelete() for changes and reserve this
1095  * routine just for adds.  I'm not sure why I thought it was necessary to do
1096  * changes this way.
1097  */
1098 #ifdef DEBUG
1099 static int rtfcdebug = 0;
1100 #endif
1101 
1102 static int
1103 rt_fixchange(struct radix_node *rn, void *vp)
1104 {
1105 	struct rtentry *rt = (struct rtentry *)rn;
1106 	struct rtfc_arg *ap = vp;
1107 	struct rtentry *rt0 = ap->rt0;
1108 	struct radix_node_head *rnh = ap->rnh;
1109 	u_char *xk1, *xm1, *xk2, *xmp;
1110 	int i, len, mlen;
1111 
1112 #ifdef DEBUG
1113 	if (rtfcdebug)
1114 		kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0);
1115 #endif
1116 
1117 	if (rt->rt_parent == NULL ||
1118 	    (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
1119 #ifdef DEBUG
1120 		if (rtfcdebug) kprintf("no parent, pinned or cloning\n");
1121 #endif
1122 		return 0;
1123 	}
1124 
1125 	if (rt->rt_parent == rt0) {
1126 #ifdef DEBUG
1127 		if (rtfcdebug) kprintf("parent match\n");
1128 #endif
1129 		return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1130 				 rt->rt_flags, NULL);
1131 	}
1132 
1133 	/*
1134 	 * There probably is a function somewhere which does this...
1135 	 * if not, there should be.
1136 	 */
1137 	len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len);
1138 
1139 	xk1 = (u_char *)rt_key(rt0);
1140 	xm1 = (u_char *)rt_mask(rt0);
1141 	xk2 = (u_char *)rt_key(rt);
1142 
1143 	/* avoid applying a less specific route */
1144 	xmp = (u_char *)rt_mask(rt->rt_parent);
1145 	mlen = rt_key(rt->rt_parent)->sa_len;
1146 	if (mlen > rt_key(rt0)->sa_len) {
1147 #ifdef DEBUG
1148 		if (rtfcdebug)
1149 			kprintf("rt_fixchange: inserting a less "
1150 			       "specific route\n");
1151 #endif
1152 		return 0;
1153 	}
1154 	for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) {
1155 		if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) {
1156 #ifdef DEBUG
1157 			if (rtfcdebug)
1158 				kprintf("rt_fixchange: inserting a less "
1159 				       "specific route\n");
1160 #endif
1161 			return 0;
1162 		}
1163 	}
1164 
1165 	for (i = rnh->rnh_treetop->rn_offset; i < len; i++) {
1166 		if ((xk2[i] & xm1[i]) != xk1[i]) {
1167 #ifdef DEBUG
1168 			if (rtfcdebug) kprintf("no match\n");
1169 #endif
1170 			return 0;
1171 		}
1172 	}
1173 
1174 	/*
1175 	 * OK, this node is a clone, and matches the node currently being
1176 	 * changed/added under the node's mask.  So, get rid of it.
1177 	 */
1178 #ifdef DEBUG
1179 	if (rtfcdebug) kprintf("deleting\n");
1180 #endif
1181 	return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1182 			 rt->rt_flags, NULL);
1183 }
1184 
1185 #define ROUNDUP(a) (a>0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))
1186 
1187 int
1188 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate)
1189 {
1190 	char *space, *oldspace;
1191 	int dlen = ROUNDUP(dst->sa_len), glen = ROUNDUP(gate->sa_len);
1192 	struct rtentry *rt = rt0;
1193 	struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family];
1194 
1195 	/*
1196 	 * A host route with the destination equal to the gateway
1197 	 * will interfere with keeping LLINFO in the routing
1198 	 * table, so disallow it.
1199 	 */
1200 	if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) ==
1201 			      (RTF_HOST | RTF_GATEWAY)) &&
1202 	    dst->sa_len == gate->sa_len &&
1203 	    sa_equal(dst, gate)) {
1204 		/*
1205 		 * The route might already exist if this is an RTM_CHANGE
1206 		 * or a routing redirect, so try to delete it.
1207 		 */
1208 		if (rt_key(rt0) != NULL)
1209 			rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway,
1210 				  rt_mask(rt0), rt0->rt_flags, NULL);
1211 		return EADDRNOTAVAIL;
1212 	}
1213 
1214 	/*
1215 	 * Both dst and gateway are stored in the same malloc'ed chunk
1216 	 * (If I ever get my hands on....)
1217 	 * if we need to malloc a new chunk, then keep the old one around
1218 	 * till we don't need it any more.
1219 	 */
1220 	if (rt->rt_gateway == NULL || glen > ROUNDUP(rt->rt_gateway->sa_len)) {
1221 		oldspace = (char *)rt_key(rt);
1222 		R_Malloc(space, char *, dlen + glen);
1223 		if (space == NULL)
1224 			return ENOBUFS;
1225 		rt->rt_nodes->rn_key = space;
1226 	} else {
1227 		space = (char *)rt_key(rt);	/* Just use the old space. */
1228 		oldspace = NULL;
1229 	}
1230 
1231 	/* Set the gateway value. */
1232 	rt->rt_gateway = (struct sockaddr *)(space + dlen);
1233 	bcopy(gate, rt->rt_gateway, glen);
1234 
1235 	if (oldspace != NULL) {
1236 		/*
1237 		 * If we allocated a new chunk, preserve the original dst.
1238 		 * This way, rt_setgate() really just sets the gate
1239 		 * and leaves the dst field alone.
1240 		 */
1241 		bcopy(dst, space, dlen);
1242 		Free(oldspace);
1243 	}
1244 
1245 	/*
1246 	 * If there is already a gwroute, it's now almost definitely wrong
1247 	 * so drop it.
1248 	 */
1249 	if (rt->rt_gwroute != NULL) {
1250 		RTFREE(rt->rt_gwroute);
1251 		rt->rt_gwroute = NULL;
1252 	}
1253 	if (rt->rt_flags & RTF_GATEWAY) {
1254 		/*
1255 		 * Cloning loop avoidance: In the presence of
1256 		 * protocol-cloning and bad configuration, it is
1257 		 * possible to get stuck in bottomless mutual recursion
1258 		 * (rtrequest rt_setgate rtlookup).  We avoid this
1259 		 * by not allowing protocol-cloning to operate for
1260 		 * gateways (which is probably the correct choice
1261 		 * anyway), and avoid the resulting reference loops
1262 		 * by disallowing any route to run through itself as
1263 		 * a gateway.  This is obviously mandatory when we
1264 		 * get rt->rt_output().
1265 		 *
1266 		 * This breaks TTCP for hosts outside the gateway!  XXX JH
1267 		 */
1268 		rt->rt_gwroute = _rtlookup(gate, RTL_REPORTMSG, RTF_PRCLONING);
1269 		if (rt->rt_gwroute == rt) {
1270 			rt->rt_gwroute = NULL;
1271 			--rt->rt_refcnt;
1272 			return EDQUOT; /* failure */
1273 		}
1274 	}
1275 
1276 	/*
1277 	 * This isn't going to do anything useful for host routes, so
1278 	 * don't bother.  Also make sure we have a reasonable mask
1279 	 * (we don't yet have one during adds).
1280 	 */
1281 	if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) {
1282 		struct rtfc_arg arg = { rt, rnh };
1283 
1284 		rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
1285 				       (char *)rt_mask(rt),
1286 				       rt_fixchange, &arg);
1287 	}
1288 
1289 	return 0;
1290 }
1291 
1292 static void
1293 rt_maskedcopy(
1294 	struct sockaddr *src,
1295 	struct sockaddr *dst,
1296 	struct sockaddr *netmask)
1297 {
1298 	u_char *cp1 = (u_char *)src;
1299 	u_char *cp2 = (u_char *)dst;
1300 	u_char *cp3 = (u_char *)netmask;
1301 	u_char *cplim = cp2 + *cp3;
1302 	u_char *cplim2 = cp2 + *cp1;
1303 
1304 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1305 	cp3 += 2;
1306 	if (cplim > cplim2)
1307 		cplim = cplim2;
1308 	while (cp2 < cplim)
1309 		*cp2++ = *cp1++ & *cp3++;
1310 	if (cp2 < cplim2)
1311 		bzero(cp2, cplim2 - cp2);
1312 }
1313 
1314 int
1315 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt)
1316 {
1317 	struct rtentry *up_rt, *rt;
1318 
1319 	if (!(rt0->rt_flags & RTF_UP)) {
1320 		up_rt = rtlookup(dst);
1321 		if (up_rt == NULL)
1322 			return (EHOSTUNREACH);
1323 		up_rt->rt_refcnt--;
1324 	} else
1325 		up_rt = rt0;
1326 	if (up_rt->rt_flags & RTF_GATEWAY) {
1327 		if (up_rt->rt_gwroute == NULL) {
1328 			up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
1329 			if (up_rt->rt_gwroute == NULL)
1330 				return (EHOSTUNREACH);
1331 		} else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) {
1332 			rtfree(up_rt->rt_gwroute);
1333 			up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
1334 			if (up_rt->rt_gwroute == NULL)
1335 				return (EHOSTUNREACH);
1336 		}
1337 		rt = up_rt->rt_gwroute;
1338 	} else
1339 		rt = up_rt;
1340 	if (rt->rt_flags & RTF_REJECT &&
1341 	    (rt->rt_rmx.rmx_expire == 0 ||		/* rt doesn't expire */
1342 	     time_second < rt->rt_rmx.rmx_expire))	/* rt not expired */
1343 		return (rt->rt_flags & RTF_HOST ?  EHOSTDOWN : EHOSTUNREACH);
1344 	*drt = rt;
1345 	return 0;
1346 }
1347 
1348 static int
1349 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){
1350 	int i;
1351 
1352 	for (i=0; i<3; i++) {
1353 		struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i];
1354 		int shimlen;
1355 
1356 		if (shim == NULL)
1357 			break;
1358 
1359 		shimlen = ROUNDUP(shim->sa_len);
1360 		R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen);
1361 		bcopy(shim, rt->rt_shim[i], shimlen);
1362 	}
1363 
1364 	return 0;
1365 }
1366 
1367 #ifdef ROUTE_DEBUG
1368 
1369 /*
1370  * Print out a route table entry
1371  */
1372 void
1373 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn)
1374 {
1375 	kprintf("rti %p cpu %d route %p flags %08lx: ",
1376 		rtinfo, mycpuid, rn, rn->rt_flags);
1377 	sockaddr_print(rt_key(rn));
1378 	kprintf(" mask ");
1379 	sockaddr_print(rt_mask(rn));
1380 	kprintf(" gw ");
1381 	sockaddr_print(rn->rt_gateway);
1382 	kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?");
1383 	kprintf(" ifa %p\n", rn->rt_ifa);
1384 }
1385 
1386 void
1387 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti)
1388 {
1389 	int didit = 0;
1390 	int i;
1391 
1392 #ifdef ROUTE_DEBUG
1393 	if (cmd == RTM_DELETE && route_debug > 1)
1394 		backtrace();
1395 #endif
1396 
1397 	switch(cmd) {
1398 	case RTM_ADD:
1399 		kprintf("ADD ");
1400 		break;
1401 	case RTM_RESOLVE:
1402 		kprintf("RES ");
1403 		break;
1404 	case RTM_DELETE:
1405 		kprintf("DEL ");
1406 		break;
1407 	default:
1408 		kprintf("C%02d ", cmd);
1409 		break;
1410 	}
1411 	kprintf("rti %p cpu %d ", rti, mycpuid);
1412 	for (i = 0; i < rti->rti_addrs; ++i) {
1413 		if (rti->rti_info[i] == NULL)
1414 			continue;
1415 		if (didit)
1416 			kprintf(" ,");
1417 		switch(i) {
1418 		case RTAX_DST:
1419 			kprintf("(DST ");
1420 			break;
1421 		case RTAX_GATEWAY:
1422 			kprintf("(GWY ");
1423 			break;
1424 		case RTAX_NETMASK:
1425 			kprintf("(MSK ");
1426 			break;
1427 		case RTAX_GENMASK:
1428 			kprintf("(GEN ");
1429 			break;
1430 		case RTAX_IFP:
1431 			kprintf("(IFP ");
1432 			break;
1433 		case RTAX_IFA:
1434 			kprintf("(IFA ");
1435 			break;
1436 		case RTAX_AUTHOR:
1437 			kprintf("(AUT ");
1438 			break;
1439 		case RTAX_BRD:
1440 			kprintf("(BRD ");
1441 			break;
1442 		default:
1443 			kprintf("(?%02d ", i);
1444 			break;
1445 		}
1446 		sockaddr_print(rti->rti_info[i]);
1447 		kprintf(")");
1448 		didit = 1;
1449 	}
1450 	kprintf("\n");
1451 }
1452 
1453 void
1454 sockaddr_print(struct sockaddr *sa)
1455 {
1456 	struct sockaddr_in *sa4;
1457 	struct sockaddr_in6 *sa6;
1458 	int len;
1459 	int i;
1460 
1461 	if (sa == NULL) {
1462 		kprintf("NULL");
1463 		return;
1464 	}
1465 
1466 	len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]);
1467 
1468 	switch(sa->sa_family) {
1469 	case AF_INET:
1470 	case AF_INET6:
1471 	default:
1472 		switch(sa->sa_family) {
1473 		case AF_INET:
1474 			sa4 = (struct sockaddr_in *)sa;
1475 			kprintf("INET %d %d.%d.%d.%d",
1476 				ntohs(sa4->sin_port),
1477 				(ntohl(sa4->sin_addr.s_addr) >> 24) & 255,
1478 				(ntohl(sa4->sin_addr.s_addr) >> 16) & 255,
1479 				(ntohl(sa4->sin_addr.s_addr) >> 8) & 255,
1480 				(ntohl(sa4->sin_addr.s_addr) >> 0) & 255
1481 			);
1482 			break;
1483 		case AF_INET6:
1484 			sa6 = (struct sockaddr_in6 *)sa;
1485 			kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x",
1486 				ntohs(sa6->sin6_port),
1487 				sa6->sin6_addr.s6_addr16[0],
1488 				sa6->sin6_addr.s6_addr16[1],
1489 				sa6->sin6_addr.s6_addr16[2],
1490 				sa6->sin6_addr.s6_addr16[3],
1491 				sa6->sin6_addr.s6_addr16[4],
1492 				sa6->sin6_addr.s6_addr16[5],
1493 				sa6->sin6_addr.s6_addr16[6],
1494 				sa6->sin6_addr.s6_addr16[7]
1495 			);
1496 			break;
1497 		default:
1498 			kprintf("AF%d ", sa->sa_family);
1499 			while (len > 0 && sa->sa_data[len-1] == 0)
1500 				--len;
1501 
1502 			for (i = 0; i < len; ++i) {
1503 				if (i)
1504 					kprintf(".");
1505 				kprintf("%d", (unsigned char)sa->sa_data[i]);
1506 			}
1507 			break;
1508 		}
1509 	}
1510 }
1511 
1512 #endif
1513 
1514 /*
1515  * Set up a routing table entry, normally for an interface.
1516  */
1517 int
1518 rtinit(struct ifaddr *ifa, int cmd, int flags)
1519 {
1520 	struct sockaddr *dst, *deldst, *netmask;
1521 	struct mbuf *m = NULL;
1522 	struct radix_node_head *rnh;
1523 	struct radix_node *rn;
1524 	struct rt_addrinfo rtinfo;
1525 	int error;
1526 
1527 	if (flags & RTF_HOST) {
1528 		dst = ifa->ifa_dstaddr;
1529 		netmask = NULL;
1530 	} else {
1531 		dst = ifa->ifa_addr;
1532 		netmask = ifa->ifa_netmask;
1533 	}
1534 	/*
1535 	 * If it's a delete, check that if it exists, it's on the correct
1536 	 * interface or we might scrub a route to another ifa which would
1537 	 * be confusing at best and possibly worse.
1538 	 */
1539 	if (cmd == RTM_DELETE) {
1540 		/*
1541 		 * It's a delete, so it should already exist..
1542 		 * If it's a net, mask off the host bits
1543 		 * (Assuming we have a mask)
1544 		 */
1545 		if (netmask != NULL) {
1546 			m = m_get(MB_DONTWAIT, MT_SONAME);
1547 			if (m == NULL)
1548 				return (ENOBUFS);
1549 			mbuftrackid(m, 34);
1550 			deldst = mtod(m, struct sockaddr *);
1551 			rt_maskedcopy(dst, deldst, netmask);
1552 			dst = deldst;
1553 		}
1554 		/*
1555 		 * Look up an rtentry that is in the routing tree and
1556 		 * contains the correct info.
1557 		 */
1558 		if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL ||
1559 		    (rn = rnh->rnh_lookup((char *)dst,
1560 					  (char *)netmask, rnh)) == NULL ||
1561 		    ((struct rtentry *)rn)->rt_ifa != ifa ||
1562 		    !sa_equal((struct sockaddr *)rn->rn_key, dst)) {
1563 			if (m != NULL)
1564 				m_free(m);
1565 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1566 		}
1567 		/* XXX */
1568 #if 0
1569 		else {
1570 			/*
1571 			 * One would think that as we are deleting, and we know
1572 			 * it doesn't exist, we could just return at this point
1573 			 * with an "ELSE" clause, but apparently not..
1574 			 */
1575 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1576 		}
1577 #endif
1578 	}
1579 	/*
1580 	 * Do the actual request
1581 	 */
1582 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
1583 	rtinfo.rti_info[RTAX_DST] = dst;
1584 	rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1585 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
1586 	rtinfo.rti_flags = flags | ifa->ifa_flags;
1587 	rtinfo.rti_ifa = ifa;
1588 	error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa);
1589 	if (m != NULL)
1590 		m_free(m);
1591 	return (error);
1592 }
1593 
1594 static void
1595 rtinit_rtrequest_callback(int cmd, int error,
1596 			  struct rt_addrinfo *rtinfo, struct rtentry *rt,
1597 			  void *arg)
1598 {
1599 	struct ifaddr *ifa = arg;
1600 
1601 	if (error == 0 && rt) {
1602 		if (mycpuid == 0) {
1603 			++rt->rt_refcnt;
1604 			rt_newaddrmsg(cmd, ifa, error, rt);
1605 			--rt->rt_refcnt;
1606 		}
1607 		if (cmd == RTM_DELETE) {
1608 			if (rt->rt_refcnt == 0) {
1609 				++rt->rt_refcnt;
1610 				rtfree(rt);
1611 			}
1612 		}
1613 	}
1614 }
1615 
1616 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */
1617 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
1618