xref: /dflybsd-src/sys/net/route.c (revision 744c01d0dc2aa1481a40e5b0988d15691602f5c9)
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 1.27 2006/12/22 23:44:54 swildner 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 #include <sys/sysctl.h>
98 #include <sys/globaldata.h>
99 #include <sys/thread.h>
100 #include <sys/thread2.h>
101 #include <sys/msgport2.h>
102 
103 #include <net/if.h>
104 #include <net/route.h>
105 #include <net/netisr.h>
106 
107 #include <netinet/in.h>
108 #include <net/ip_mroute/ip_mroute.h>
109 
110 static struct rtstatistics rtstatistics_percpu[MAXCPU];
111 #ifdef SMP
112 #define rtstat	rtstatistics_percpu[mycpuid]
113 #else
114 #define rtstat	rtstatistics_percpu[0]
115 #endif
116 
117 struct radix_node_head *rt_tables[MAXCPU][AF_MAX+1];
118 struct lwkt_port *rt_ports[MAXCPU];
119 
120 static void	rt_maskedcopy (struct sockaddr *, struct sockaddr *,
121 			       struct sockaddr *);
122 static void rtable_init(void);
123 static void rtable_service_loop(void *dummy);
124 static void rtinit_rtrequest_callback(int, int, struct rt_addrinfo *,
125 				      struct rtentry *, void *);
126 
127 #ifdef SMP
128 static int rtredirect_msghandler(struct lwkt_msg *lmsg);
129 static int rtrequest1_msghandler(struct lwkt_msg *lmsg);
130 #endif
131 
132 SYSCTL_NODE(_net, OID_AUTO, route, CTLFLAG_RW, 0, "Routing");
133 
134 #ifdef ROUTE_DEBUG
135 static int route_debug = 1;
136 SYSCTL_INT(_net_route, OID_AUTO, route_debug, CTLFLAG_RW,
137            &route_debug, 0, "");
138 #endif
139 
140 /*
141  * Initialize the route table(s) for protocol domains and
142  * create a helper thread which will be responsible for updating
143  * route table entries on each cpu.
144  */
145 void
146 route_init(void)
147 {
148 	int cpu, origcpu;
149 	thread_t rtd;
150 
151 	for (cpu = 0; cpu < ncpus; ++cpu)
152 		bzero(&rtstatistics_percpu[cpu], sizeof(struct rtstatistics));
153 	rn_init();      /* initialize all zeroes, all ones, mask table */
154 	origcpu = mycpuid;
155 	for (cpu = 0; cpu < ncpus; cpu++) {
156 		lwkt_migratecpu(cpu);
157 		rtable_init();
158 		lwkt_create(rtable_service_loop, NULL, &rtd, NULL,
159 			    TDF_STOPREQ, cpu, "rtable_cpu %d", cpu);
160 		rt_ports[cpu] = &rtd->td_msgport;
161 		lwkt_schedule(rtd);
162 	}
163 	lwkt_migratecpu(origcpu);
164 }
165 
166 static void
167 rtable_init(void)
168 {
169 	struct domain *dom;
170 
171 	SLIST_FOREACH(dom, &domains, dom_next) {
172 		if (dom->dom_rtattach) {
173 			dom->dom_rtattach(
174 				(void **)&rt_tables[mycpuid][dom->dom_family],
175 			        dom->dom_rtoffset);
176 		}
177 	}
178 }
179 
180 /*
181  * Our per-cpu table management protocol thread.  All route table operations
182  * are chained through all cpus in order starting at cpu #0 in order to
183  * maintain duplicate route tables on each cpu.  Having a spearate route
184  * table management thread allows the protocol and interrupt threads to
185  * issue route table changes.
186  */
187 static void
188 rtable_service_loop(void *dummy __unused)
189 {
190 	struct lwkt_msg *lmsg;
191 	thread_t td = curthread;
192 
193 	while ((lmsg = lwkt_waitport(&td->td_msgport, NULL)) != NULL) {
194 		lmsg->ms_cmd.cm_func(lmsg);
195 	}
196 }
197 
198 /*
199  * Routing statistics.
200  */
201 #ifdef SMP
202 static int
203 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS)
204 {
205 	int cpu, error = 0;
206 
207 	for (cpu = 0; cpu < ncpus; ++cpu) {
208 		if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu],
209 					sizeof(struct rtstatistics))))
210 				break;
211 		if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu],
212 					sizeof(struct rtstatistics))))
213 				break;
214 	}
215 
216 	return (error);
217 }
218 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW),
219 	0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics");
220 #else
221 SYSCTL_STRUCT(_net_route, OID_AUTO, stats, CTLFLAG_RW, &rtstat, rtstatistics,
222 "Routing statistics");
223 #endif
224 
225 /*
226  * Packet routing routines.
227  */
228 
229 /*
230  * Look up and fill in the "ro_rt" rtentry field in a route structure given
231  * an address in the "ro_dst" field.  Always send a report on a miss and
232  * always clone routes.
233  */
234 void
235 rtalloc(struct route *ro)
236 {
237 	rtalloc_ign(ro, 0UL);
238 }
239 
240 /*
241  * Look up and fill in the "ro_rt" rtentry field in a route structure given
242  * an address in the "ro_dst" field.  Always send a report on a miss and
243  * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being
244  * ignored.
245  */
246 void
247 rtalloc_ign(struct route *ro, u_long ignoreflags)
248 {
249 	if (ro->ro_rt != NULL) {
250 		if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP)
251 			return;
252 		rtfree(ro->ro_rt);
253 		ro->ro_rt = NULL;
254 	}
255 	ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags);
256 }
257 
258 /*
259  * Look up the route that matches the given "dst" address.
260  *
261  * Route lookup can have the side-effect of creating and returning
262  * a cloned route instead when "dst" matches a cloning route and the
263  * RTF_CLONING and RTF_PRCLONING flags are not being ignored.
264  *
265  * Any route returned has its reference count incremented.
266  */
267 struct rtentry *
268 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore)
269 {
270 	struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family];
271 	struct rtentry *rt;
272 
273 	if (rnh == NULL)
274 		goto unreach;
275 
276 	/*
277 	 * Look up route in the radix tree.
278 	 */
279 	rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh);
280 	if (rt == NULL)
281 		goto unreach;
282 
283 	/*
284 	 * Handle cloning routes.
285 	 */
286 	if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) {
287 		struct rtentry *clonedroute;
288 		int error;
289 
290 		clonedroute = rt;	/* copy in/copy out parameter */
291 		error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0,
292 				  &clonedroute);	/* clone the route */
293 		if (error != 0) {	/* cloning failed */
294 			if (generate_report)
295 				rt_dstmsg(RTM_MISS, dst, error);
296 			rt->rt_refcnt++;
297 			return (rt);	/* return the uncloned route */
298 		}
299 		if (generate_report) {
300 			if (clonedroute->rt_flags & RTF_XRESOLVE)
301 				rt_dstmsg(RTM_RESOLVE, dst, 0);
302 			else
303 				rt_rtmsg(RTM_ADD, clonedroute,
304 					 clonedroute->rt_ifp, 0);
305 		}
306 		return (clonedroute);	/* return cloned route */
307 	}
308 
309 	/*
310 	 * Increment the reference count of the matched route and return.
311 	 */
312 	rt->rt_refcnt++;
313 	return (rt);
314 
315 unreach:
316 	rtstat.rts_unreach++;
317 	if (generate_report)
318 		rt_dstmsg(RTM_MISS, dst, 0);
319 	return (NULL);
320 }
321 
322 void
323 rtfree(struct rtentry *rt)
324 {
325 	KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt));
326 
327 	--rt->rt_refcnt;
328 	if (rt->rt_refcnt == 0) {
329 		struct radix_node_head *rnh =
330 		    rt_tables[mycpuid][rt_key(rt)->sa_family];
331 
332 		if (rnh->rnh_close)
333 			rnh->rnh_close((struct radix_node *)rt, rnh);
334 		if (!(rt->rt_flags & RTF_UP)) {
335 			/* deallocate route */
336 			if (rt->rt_ifa != NULL)
337 				IFAFREE(rt->rt_ifa);
338 			if (rt->rt_parent != NULL)
339 				RTFREE(rt->rt_parent);	/* recursive call! */
340 			Free(rt_key(rt));
341 			Free(rt);
342 		}
343 	}
344 }
345 
346 static int
347 rtredirect_oncpu(struct sockaddr *dst, struct sockaddr *gateway,
348 		 struct sockaddr *netmask, int flags, struct sockaddr *src)
349 {
350 	struct rtentry *rt = NULL;
351 	struct rt_addrinfo rtinfo;
352 	struct ifaddr *ifa;
353 	u_long *stat = NULL;
354 	int error;
355 
356 	/* verify the gateway is directly reachable */
357 	if ((ifa = ifa_ifwithnet(gateway)) == NULL) {
358 		error = ENETUNREACH;
359 		goto out;
360 	}
361 
362 	/*
363 	 * If the redirect isn't from our current router for this destination,
364 	 * it's either old or wrong.
365 	 */
366 	if (!(flags & RTF_DONE) &&		/* XXX JH */
367 	    (rt = rtpurelookup(dst)) != NULL &&
368 	    (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) {
369 		error = EINVAL;
370 		goto done;
371 	}
372 
373 	/*
374 	 * If it redirects us to ourselves, we have a routing loop,
375 	 * perhaps as a result of an interface going down recently.
376 	 */
377 	if (ifa_ifwithaddr(gateway)) {
378 		error = EHOSTUNREACH;
379 		goto done;
380 	}
381 
382 	/*
383 	 * Create a new entry if the lookup failed or if we got back
384 	 * a wildcard entry for the default route.  This is necessary
385 	 * for hosts which use routing redirects generated by smart
386 	 * gateways to dynamically build the routing tables.
387 	 */
388 	if (rt == NULL)
389 		goto create;
390 	if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) {
391 		rtfree(rt);
392 		goto create;
393 	}
394 
395 	/* Ignore redirects for directly connected hosts. */
396 	if (!(rt->rt_flags & RTF_GATEWAY)) {
397 		error = EHOSTUNREACH;
398 		goto done;
399 	}
400 
401 	if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) {
402 		/*
403 		 * Changing from a network route to a host route.
404 		 * Create a new host route rather than smashing the
405 		 * network route.
406 		 */
407 create:
408 		flags |=  RTF_GATEWAY | RTF_DYNAMIC;
409 		bzero(&rtinfo, sizeof(struct rt_addrinfo));
410 		rtinfo.rti_info[RTAX_DST] = dst;
411 		rtinfo.rti_info[RTAX_GATEWAY] = gateway;
412 		rtinfo.rti_info[RTAX_NETMASK] = netmask;
413 		rtinfo.rti_flags = flags;
414 		rtinfo.rti_ifa = ifa;
415 		rt = NULL;	/* copy-in/copy-out parameter */
416 		error = rtrequest1(RTM_ADD, &rtinfo, &rt);
417 		if (rt != NULL)
418 			flags = rt->rt_flags;
419 		stat = &rtstat.rts_dynamic;
420 	} else {
421 		/*
422 		 * Smash the current notion of the gateway to this destination.
423 		 * Should check about netmask!!!
424 		 */
425 		rt->rt_flags |= RTF_MODIFIED;
426 		flags |= RTF_MODIFIED;
427 		rt_setgate(rt, rt_key(rt), gateway);
428 		error = 0;
429 		stat = &rtstat.rts_newgateway;
430 	}
431 
432 done:
433 	if (rt != NULL)
434 		rtfree(rt);
435 out:
436 	if (error != 0)
437 		rtstat.rts_badredirect++;
438 	else if (stat != NULL)
439 		(*stat)++;
440 
441 	return error;
442 }
443 
444 #ifdef SMP
445 
446 struct netmsg_rtredirect {
447 	struct lwkt_msg	lmsg;
448 	struct sockaddr *dst;
449 	struct sockaddr *gateway;
450 	struct sockaddr *netmask;
451 	int		flags;
452 	struct sockaddr *src;
453 };
454 
455 #endif
456 
457 /*
458  * Force a routing table entry to the specified
459  * destination to go through the given gateway.
460  * Normally called as a result of a routing redirect
461  * message from the network layer.
462  *
463  * N.B.: must be called at splnet
464  */
465 void
466 rtredirect(struct sockaddr *dst, struct sockaddr *gateway,
467 	   struct sockaddr *netmask, int flags, struct sockaddr *src)
468 {
469 	struct rt_addrinfo rtinfo;
470 	int error;
471 #ifdef SMP
472 	struct netmsg_rtredirect msg;
473 
474 	lwkt_initmsg(&msg.lmsg, &curthread->td_msgport, 0,
475 		     lwkt_cmd_func(rtredirect_msghandler), lwkt_cmd_op_none);
476 	msg.dst = dst;
477 	msg.gateway = gateway;
478 	msg.netmask = netmask;
479 	msg.flags = flags;
480 	msg.src = src;
481 	error = lwkt_domsg(rtable_portfn(0), &msg.lmsg);
482 #else
483 	error = rtredirect_oncpu(dst, gateway, netmask, flags, src);
484 #endif
485 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
486 	rtinfo.rti_info[RTAX_DST] = dst;
487 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
488 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
489 	rtinfo.rti_info[RTAX_AUTHOR] = src;
490 	rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error);
491 }
492 
493 #ifdef SMP
494 
495 static int
496 rtredirect_msghandler(struct lwkt_msg *lmsg)
497 {
498 	struct netmsg_rtredirect *msg = (void *)lmsg;
499 	int nextcpu;
500 
501 	rtredirect_oncpu(msg->dst, msg->gateway, msg->netmask,
502 			 msg->flags, msg->src);
503 	nextcpu = mycpuid + 1;
504 	if (nextcpu < ncpus)
505 		lwkt_forwardmsg(rtable_portfn(nextcpu), &msg->lmsg);
506 	else
507 		lwkt_replymsg(&msg->lmsg, 0);
508 	return (0);
509 }
510 
511 #endif
512 
513 /*
514 * Routing table ioctl interface.
515 */
516 int
517 rtioctl(u_long req, caddr_t data, struct ucred *cred)
518 {
519 #ifdef INET
520 	/* Multicast goop, grrr... */
521 	return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP;
522 #else
523 	return ENXIO;
524 #endif
525 }
526 
527 struct ifaddr *
528 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
529 {
530 	struct ifaddr *ifa;
531 
532 	if (!(flags & RTF_GATEWAY)) {
533 		/*
534 		 * If we are adding a route to an interface,
535 		 * and the interface is a point-to-point link,
536 		 * we should search for the destination
537 		 * as our clue to the interface.  Otherwise
538 		 * we can use the local address.
539 		 */
540 		ifa = NULL;
541 		if (flags & RTF_HOST) {
542 			ifa = ifa_ifwithdstaddr(dst);
543 		}
544 		if (ifa == NULL)
545 			ifa = ifa_ifwithaddr(gateway);
546 	} else {
547 		/*
548 		 * If we are adding a route to a remote net
549 		 * or host, the gateway may still be on the
550 		 * other end of a pt to pt link.
551 		 */
552 		ifa = ifa_ifwithdstaddr(gateway);
553 	}
554 	if (ifa == NULL)
555 		ifa = ifa_ifwithnet(gateway);
556 	if (ifa == NULL) {
557 		struct rtentry *rt;
558 
559 		rt = rtpurelookup(gateway);
560 		if (rt == NULL)
561 			return (NULL);
562 		rt->rt_refcnt--;
563 		if ((ifa = rt->rt_ifa) == NULL)
564 			return (NULL);
565 	}
566 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
567 		struct ifaddr *oldifa = ifa;
568 
569 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
570 		if (ifa == NULL)
571 			ifa = oldifa;
572 	}
573 	return (ifa);
574 }
575 
576 static int rt_fixdelete (struct radix_node *, void *);
577 static int rt_fixchange (struct radix_node *, void *);
578 
579 struct rtfc_arg {
580 	struct rtentry *rt0;
581 	struct radix_node_head *rnh;
582 };
583 
584 /*
585  * Set rtinfo->rti_ifa and rtinfo->rti_ifp.
586  */
587 int
588 rt_getifa(struct rt_addrinfo *rtinfo)
589 {
590 	struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY];
591 	struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
592 	struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA];
593 	int flags = rtinfo->rti_flags;
594 
595 	/*
596 	 * ifp may be specified by sockaddr_dl
597 	 * when protocol address is ambiguous.
598 	 */
599 	if (rtinfo->rti_ifp == NULL) {
600 		struct sockaddr *ifpaddr;
601 
602 		ifpaddr = rtinfo->rti_info[RTAX_IFP];
603 		if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) {
604 			struct ifaddr *ifa;
605 
606 			ifa = ifa_ifwithnet(ifpaddr);
607 			if (ifa != NULL)
608 				rtinfo->rti_ifp = ifa->ifa_ifp;
609 		}
610 	}
611 
612 	if (rtinfo->rti_ifa == NULL && ifaaddr != NULL)
613 		rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr);
614 	if (rtinfo->rti_ifa == NULL) {
615 		struct sockaddr *sa;
616 
617 		sa = ifaaddr != NULL ? ifaaddr :
618 		    (gateway != NULL ? gateway : dst);
619 		if (sa != NULL && rtinfo->rti_ifp != NULL)
620 			rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp);
621 		else if (dst != NULL && gateway != NULL)
622 			rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway);
623 		else if (sa != NULL)
624 			rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa);
625 	}
626 	if (rtinfo->rti_ifa == NULL)
627 		return (ENETUNREACH);
628 
629 	if (rtinfo->rti_ifp == NULL)
630 		rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp;
631 	return (0);
632 }
633 
634 /*
635  * Do appropriate manipulations of a routing tree given
636  * all the bits of info needed
637  */
638 int
639 rtrequest(
640 	int req,
641 	struct sockaddr *dst,
642 	struct sockaddr *gateway,
643 	struct sockaddr *netmask,
644 	int flags,
645 	struct rtentry **ret_nrt)
646 {
647 	struct rt_addrinfo rtinfo;
648 
649 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
650 	rtinfo.rti_info[RTAX_DST] = dst;
651 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
652 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
653 	rtinfo.rti_flags = flags;
654 	return rtrequest1(req, &rtinfo, ret_nrt);
655 }
656 
657 int
658 rtrequest_global(
659 	int req,
660 	struct sockaddr *dst,
661 	struct sockaddr *gateway,
662 	struct sockaddr *netmask,
663 	int flags)
664 {
665 	struct rt_addrinfo rtinfo;
666 
667 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
668 	rtinfo.rti_info[RTAX_DST] = dst;
669 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
670 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
671 	rtinfo.rti_flags = flags;
672 	return rtrequest1_global(req, &rtinfo, NULL, NULL);
673 }
674 
675 #ifdef SMP
676 
677 struct netmsg_rtq {
678 	struct lwkt_msg		lmsg;
679 	int			req;
680 	struct rt_addrinfo	*rtinfo;
681 	rtrequest1_callback_func_t callback;
682 	void			*arg;
683 };
684 
685 #endif
686 
687 int
688 rtrequest1_global(int req, struct rt_addrinfo *rtinfo,
689 		  rtrequest1_callback_func_t callback, void *arg)
690 {
691 	int error;
692 #ifdef SMP
693 	struct netmsg_rtq msg;
694 
695 	lwkt_initmsg(&msg.lmsg, &curthread->td_msgport, 0,
696 		     lwkt_cmd_func(rtrequest1_msghandler), lwkt_cmd_op_none);
697 	msg.lmsg.ms_error = -1;
698 	msg.req = req;
699 	msg.rtinfo = rtinfo;
700 	msg.callback = callback;
701 	msg.arg = arg;
702 	error = lwkt_domsg(rtable_portfn(0), &msg.lmsg);
703 #else
704 	struct rtentry *rt = NULL;
705 
706 	error = rtrequest1(req, rtinfo, &rt);
707 	if (rt)
708 		--rt->rt_refcnt;
709 	if (callback)
710 		callback(req, error, rtinfo, rt, arg);
711 #endif
712 	return (error);
713 }
714 
715 /*
716  * Handle a route table request on the current cpu.  Since the route table's
717  * are supposed to be identical on each cpu, an error occuring later in the
718  * message chain is considered system-fatal.
719  */
720 #ifdef SMP
721 
722 static int
723 rtrequest1_msghandler(struct lwkt_msg *lmsg)
724 {
725 	struct netmsg_rtq *msg = (void *)lmsg;
726 	struct rtentry *rt = NULL;
727 	int nextcpu;
728 	int error;
729 
730 	error = rtrequest1(msg->req, msg->rtinfo, &rt);
731 	if (rt)
732 		--rt->rt_refcnt;
733 	if (msg->callback)
734 		msg->callback(msg->req, error, msg->rtinfo, rt, msg->arg);
735 
736 	/*
737 	 * RTM_DELETE's are propogated even if an error occurs, since a
738 	 * cloned route might be undergoing deletion and cloned routes
739 	 * are not necessarily replicated.  An overall error is returned
740 	 * only if no cpus have the route in question.
741 	 */
742 	if (msg->lmsg.ms_error < 0 || error == 0)
743 		msg->lmsg.ms_error = error;
744 
745 	nextcpu = mycpuid + 1;
746 	if (error && msg->req != RTM_DELETE) {
747 		if (mycpuid != 0) {
748 			panic("rtrequest1_msghandler: rtrequest table "
749 			      "error was not on cpu #0: %p", msg->rtinfo);
750 		}
751 		lwkt_replymsg(&msg->lmsg, error);
752 	} else if (nextcpu < ncpus) {
753 		lwkt_forwardmsg(rtable_portfn(nextcpu), &msg->lmsg);
754 	} else {
755 		lwkt_replymsg(&msg->lmsg, msg->lmsg.ms_error);
756 	}
757 	return (0);
758 }
759 
760 #endif
761 
762 int
763 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt)
764 {
765 	struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
766 	struct rtentry *rt;
767 	struct radix_node *rn;
768 	struct radix_node_head *rnh;
769 	struct ifaddr *ifa;
770 	struct sockaddr *ndst;
771 	int error = 0;
772 
773 #define gotoerr(x) { error = x ; goto bad; }
774 
775 #ifdef ROUTE_DEBUG
776 	if (route_debug)
777 		rt_addrinfo_print(req, rtinfo);
778 #endif
779 
780 	crit_enter();
781 	/*
782 	 * Find the correct routing tree to use for this Address Family
783 	 */
784 	if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL)
785 		gotoerr(EAFNOSUPPORT);
786 
787 	/*
788 	 * If we are adding a host route then we don't want to put
789 	 * a netmask in the tree, nor do we want to clone it.
790 	 */
791 	if (rtinfo->rti_flags & RTF_HOST) {
792 		rtinfo->rti_info[RTAX_NETMASK] = NULL;
793 		rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING);
794 	}
795 
796 	switch (req) {
797 	case RTM_DELETE:
798 		/* Remove the item from the tree. */
799 		rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST],
800 				      (char *)rtinfo->rti_info[RTAX_NETMASK],
801 				      rnh);
802 		if (rn == NULL)
803 			gotoerr(ESRCH);
804 		KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)),
805 			("rnh_deladdr returned flags 0x%x", rn->rn_flags));
806 		rt = (struct rtentry *)rn;
807 
808 		/* ref to prevent a deletion race */
809 		++rt->rt_refcnt;
810 
811 		/* Free any routes cloned from this one. */
812 		if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) &&
813 		    rt_mask(rt) != NULL) {
814 			rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
815 					       (char *)rt_mask(rt),
816 					       rt_fixdelete, rt);
817 		}
818 
819 		if (rt->rt_gwroute != NULL) {
820 			RTFREE(rt->rt_gwroute);
821 			rt->rt_gwroute = NULL;
822 		}
823 
824 		/*
825 		 * NB: RTF_UP must be set during the search above,
826 		 * because we might delete the last ref, causing
827 		 * rt to get freed prematurely.
828 		 */
829 		rt->rt_flags &= ~RTF_UP;
830 
831 #ifdef ROUTE_DEBUG
832 		if (route_debug)
833 			rt_print(rtinfo, rt);
834 #endif
835 
836 		/* Give the protocol a chance to keep things in sync. */
837 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
838 			ifa->ifa_rtrequest(RTM_DELETE, rt, rtinfo);
839 
840 		/*
841 		 * If the caller wants it, then it can have it,
842 		 * but it's up to it to free the rtentry as we won't be
843 		 * doing it.
844 		 */
845 		KASSERT(rt->rt_refcnt >= 0,
846 			("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt));
847 		if (ret_nrt != NULL) {
848 			/* leave ref intact for return */
849 			*ret_nrt = rt;
850 		} else {
851 			/* deref / attempt to destroy */
852 			rtfree(rt);
853 		}
854 		break;
855 
856 	case RTM_RESOLVE:
857 		if (ret_nrt == NULL || (rt = *ret_nrt) == NULL)
858 			gotoerr(EINVAL);
859 		ifa = rt->rt_ifa;
860 		rtinfo->rti_flags =
861 		    rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC);
862 		rtinfo->rti_flags |= RTF_WASCLONED;
863 		rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
864 		if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL)
865 			rtinfo->rti_flags |= RTF_HOST;
866 		goto makeroute;
867 
868 	case RTM_ADD:
869 		KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) ||
870 			rtinfo->rti_info[RTAX_GATEWAY] != NULL,
871 		    ("rtrequest: GATEWAY but no gateway"));
872 
873 		if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo)))
874 			gotoerr(error);
875 		ifa = rtinfo->rti_ifa;
876 makeroute:
877 		R_Malloc(rt, struct rtentry *, sizeof(struct rtentry));
878 		if (rt == NULL)
879 			gotoerr(ENOBUFS);
880 		bzero(rt, sizeof(struct rtentry));
881 		rt->rt_flags = RTF_UP | rtinfo->rti_flags;
882 		error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY]);
883 		if (error != 0) {
884 			Free(rt);
885 			gotoerr(error);
886 		}
887 
888 		ndst = rt_key(rt);
889 		if (rtinfo->rti_info[RTAX_NETMASK] != NULL)
890 			rt_maskedcopy(dst, ndst,
891 				      rtinfo->rti_info[RTAX_NETMASK]);
892 		else
893 			bcopy(dst, ndst, dst->sa_len);
894 
895 		/*
896 		 * Note that we now have a reference to the ifa.
897 		 * This moved from below so that rnh->rnh_addaddr() can
898 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
899 		 */
900 		IFAREF(ifa);
901 		rt->rt_ifa = ifa;
902 		rt->rt_ifp = ifa->ifa_ifp;
903 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
904 
905 		rn = rnh->rnh_addaddr((char *)ndst,
906 				      (char *)rtinfo->rti_info[RTAX_NETMASK],
907 				      rnh, rt->rt_nodes);
908 		if (rn == NULL) {
909 			struct rtentry *oldrt;
910 
911 			/*
912 			 * We already have one of these in the tree.
913 			 * We do a special hack: if the old route was
914 			 * cloned, then we blow it away and try
915 			 * re-inserting the new one.
916 			 */
917 			oldrt = rtpurelookup(ndst);
918 			if (oldrt != NULL) {
919 				--oldrt->rt_refcnt;
920 				if (oldrt->rt_flags & RTF_WASCLONED) {
921 					rtrequest(RTM_DELETE, rt_key(oldrt),
922 						  oldrt->rt_gateway,
923 						  rt_mask(oldrt),
924 						  oldrt->rt_flags, NULL);
925 					rn = rnh->rnh_addaddr((char *)ndst,
926 					    (char *)
927 						rtinfo->rti_info[RTAX_NETMASK],
928 					    rnh, rt->rt_nodes);
929 				}
930 			}
931 		}
932 
933 		/*
934 		 * If it still failed to go into the tree,
935 		 * then un-make it (this should be a function).
936 		 */
937 		if (rn == NULL) {
938 			if (rt->rt_gwroute != NULL)
939 				rtfree(rt->rt_gwroute);
940 			IFAFREE(ifa);
941 			Free(rt_key(rt));
942 			Free(rt);
943 			gotoerr(EEXIST);
944 		}
945 
946 		/*
947 		 * If we got here from RESOLVE, then we are cloning
948 		 * so clone the rest, and note that we
949 		 * are a clone (and increment the parent's references)
950 		 */
951 		if (req == RTM_RESOLVE) {
952 			rt->rt_rmx = (*ret_nrt)->rt_rmx;    /* copy metrics */
953 			rt->rt_rmx.rmx_pksent = 0;  /* reset packet counter */
954 			if ((*ret_nrt)->rt_flags &
955 				       (RTF_CLONING | RTF_PRCLONING)) {
956 				rt->rt_parent = *ret_nrt;
957 				(*ret_nrt)->rt_refcnt++;
958 			}
959 		}
960 
961 		/*
962 		 * if this protocol has something to add to this then
963 		 * allow it to do that as well.
964 		 */
965 		if (ifa->ifa_rtrequest != NULL)
966 			ifa->ifa_rtrequest(req, rt, rtinfo);
967 
968 		/*
969 		 * We repeat the same procedure from rt_setgate() here because
970 		 * it doesn't fire when we call it there because the node
971 		 * hasn't been added to the tree yet.
972 		 */
973 		if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) &&
974 		    rt_mask(rt) != NULL) {
975 			struct rtfc_arg arg = { rt, rnh };
976 
977 			rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
978 					       (char *)rt_mask(rt),
979 					       rt_fixchange, &arg);
980 		}
981 
982 #ifdef ROUTE_DEBUG
983 		if (route_debug)
984 			rt_print(rtinfo, rt);
985 #endif
986 		/*
987 		 * Return the resulting rtentry,
988 		 * increasing the number of references by one.
989 		 */
990 		if (ret_nrt != NULL) {
991 			rt->rt_refcnt++;
992 			*ret_nrt = rt;
993 		}
994 		break;
995 	default:
996 		error = EOPNOTSUPP;
997 	}
998 bad:
999 #ifdef ROUTE_DEBUG
1000 	if (route_debug) {
1001 		if (error)
1002 			kprintf("rti %p failed error %d\n", rtinfo, error);
1003 		else
1004 			kprintf("rti %p succeeded\n", rtinfo);
1005 	}
1006 #endif
1007 	crit_exit();
1008 	return (error);
1009 }
1010 
1011 /*
1012  * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family''
1013  * (i.e., the routes related to it by the operation of cloning).  This
1014  * routine is iterated over all potential former-child-routes by way of
1015  * rnh->rnh_walktree_from() above, and those that actually are children of
1016  * the late parent (passed in as VP here) are themselves deleted.
1017  */
1018 static int
1019 rt_fixdelete(struct radix_node *rn, void *vp)
1020 {
1021 	struct rtentry *rt = (struct rtentry *)rn;
1022 	struct rtentry *rt0 = vp;
1023 
1024 	if (rt->rt_parent == rt0 &&
1025 	    !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
1026 		return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1027 				 rt->rt_flags, NULL);
1028 	}
1029 	return 0;
1030 }
1031 
1032 /*
1033  * This routine is called from rt_setgate() to do the analogous thing for
1034  * adds and changes.  There is the added complication in this case of a
1035  * middle insert; i.e., insertion of a new network route between an older
1036  * network route and (cloned) host routes.  For this reason, a simple check
1037  * of rt->rt_parent is insufficient; each candidate route must be tested
1038  * against the (mask, value) of the new route (passed as before in vp)
1039  * to see if the new route matches it.
1040  *
1041  * XXX - it may be possible to do fixdelete() for changes and reserve this
1042  * routine just for adds.  I'm not sure why I thought it was necessary to do
1043  * changes this way.
1044  */
1045 #ifdef DEBUG
1046 static int rtfcdebug = 0;
1047 #endif
1048 
1049 static int
1050 rt_fixchange(struct radix_node *rn, void *vp)
1051 {
1052 	struct rtentry *rt = (struct rtentry *)rn;
1053 	struct rtfc_arg *ap = vp;
1054 	struct rtentry *rt0 = ap->rt0;
1055 	struct radix_node_head *rnh = ap->rnh;
1056 	u_char *xk1, *xm1, *xk2, *xmp;
1057 	int i, len, mlen;
1058 
1059 #ifdef DEBUG
1060 	if (rtfcdebug)
1061 		kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0);
1062 #endif
1063 
1064 	if (rt->rt_parent == NULL ||
1065 	    (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
1066 #ifdef DEBUG
1067 		if (rtfcdebug) kprintf("no parent, pinned or cloning\n");
1068 #endif
1069 		return 0;
1070 	}
1071 
1072 	if (rt->rt_parent == rt0) {
1073 #ifdef DEBUG
1074 		if (rtfcdebug) kprintf("parent match\n");
1075 #endif
1076 		return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1077 				 rt->rt_flags, NULL);
1078 	}
1079 
1080 	/*
1081 	 * There probably is a function somewhere which does this...
1082 	 * if not, there should be.
1083 	 */
1084 	len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len);
1085 
1086 	xk1 = (u_char *)rt_key(rt0);
1087 	xm1 = (u_char *)rt_mask(rt0);
1088 	xk2 = (u_char *)rt_key(rt);
1089 
1090 	/* avoid applying a less specific route */
1091 	xmp = (u_char *)rt_mask(rt->rt_parent);
1092 	mlen = rt_key(rt->rt_parent)->sa_len;
1093 	if (mlen > rt_key(rt0)->sa_len) {
1094 #ifdef DEBUG
1095 		if (rtfcdebug)
1096 			kprintf("rt_fixchange: inserting a less "
1097 			       "specific route\n");
1098 #endif
1099 		return 0;
1100 	}
1101 	for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) {
1102 		if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) {
1103 #ifdef DEBUG
1104 			if (rtfcdebug)
1105 				kprintf("rt_fixchange: inserting a less "
1106 				       "specific route\n");
1107 #endif
1108 			return 0;
1109 		}
1110 	}
1111 
1112 	for (i = rnh->rnh_treetop->rn_offset; i < len; i++) {
1113 		if ((xk2[i] & xm1[i]) != xk1[i]) {
1114 #ifdef DEBUG
1115 			if (rtfcdebug) kprintf("no match\n");
1116 #endif
1117 			return 0;
1118 		}
1119 	}
1120 
1121 	/*
1122 	 * OK, this node is a clone, and matches the node currently being
1123 	 * changed/added under the node's mask.  So, get rid of it.
1124 	 */
1125 #ifdef DEBUG
1126 	if (rtfcdebug) kprintf("deleting\n");
1127 #endif
1128 	return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1129 			 rt->rt_flags, NULL);
1130 }
1131 
1132 #define ROUNDUP(a) (a>0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))
1133 
1134 int
1135 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate)
1136 {
1137 	char *space, *oldspace;
1138 	int dlen = ROUNDUP(dst->sa_len), glen = ROUNDUP(gate->sa_len);
1139 	struct rtentry *rt = rt0;
1140 	struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family];
1141 
1142 	/*
1143 	 * A host route with the destination equal to the gateway
1144 	 * will interfere with keeping LLINFO in the routing
1145 	 * table, so disallow it.
1146 	 */
1147 	if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) ==
1148 			      (RTF_HOST | RTF_GATEWAY)) &&
1149 	    dst->sa_len == gate->sa_len &&
1150 	    sa_equal(dst, gate)) {
1151 		/*
1152 		 * The route might already exist if this is an RTM_CHANGE
1153 		 * or a routing redirect, so try to delete it.
1154 		 */
1155 		if (rt_key(rt0) != NULL)
1156 			rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway,
1157 				  rt_mask(rt0), rt0->rt_flags, NULL);
1158 		return EADDRNOTAVAIL;
1159 	}
1160 
1161 	/*
1162 	 * Both dst and gateway are stored in the same malloc'ed chunk
1163 	 * (If I ever get my hands on....)
1164 	 * if we need to malloc a new chunk, then keep the old one around
1165 	 * till we don't need it any more.
1166 	 */
1167 	if (rt->rt_gateway == NULL || glen > ROUNDUP(rt->rt_gateway->sa_len)) {
1168 		oldspace = (char *)rt_key(rt);
1169 		R_Malloc(space, char *, dlen + glen);
1170 		if (space == NULL)
1171 			return ENOBUFS;
1172 		rt->rt_nodes->rn_key = space;
1173 	} else {
1174 		space = (char *)rt_key(rt);	/* Just use the old space. */
1175 		oldspace = NULL;
1176 	}
1177 
1178 	/* Set the gateway value. */
1179 	rt->rt_gateway = (struct sockaddr *)(space + dlen);
1180 	bcopy(gate, rt->rt_gateway, glen);
1181 
1182 	if (oldspace != NULL) {
1183 		/*
1184 		 * If we allocated a new chunk, preserve the original dst.
1185 		 * This way, rt_setgate() really just sets the gate
1186 		 * and leaves the dst field alone.
1187 		 */
1188 		bcopy(dst, space, dlen);
1189 		Free(oldspace);
1190 	}
1191 
1192 	/*
1193 	 * If there is already a gwroute, it's now almost definitely wrong
1194 	 * so drop it.
1195 	 */
1196 	if (rt->rt_gwroute != NULL) {
1197 		RTFREE(rt->rt_gwroute);
1198 		rt->rt_gwroute = NULL;
1199 	}
1200 	if (rt->rt_flags & RTF_GATEWAY) {
1201 		/*
1202 		 * Cloning loop avoidance: In the presence of
1203 		 * protocol-cloning and bad configuration, it is
1204 		 * possible to get stuck in bottomless mutual recursion
1205 		 * (rtrequest rt_setgate rtlookup).  We avoid this
1206 		 * by not allowing protocol-cloning to operate for
1207 		 * gateways (which is probably the correct choice
1208 		 * anyway), and avoid the resulting reference loops
1209 		 * by disallowing any route to run through itself as
1210 		 * a gateway.  This is obviously mandatory when we
1211 		 * get rt->rt_output().
1212 		 *
1213 		 * This breaks TTCP for hosts outside the gateway!  XXX JH
1214 		 */
1215 		rt->rt_gwroute = _rtlookup(gate, RTL_REPORTMSG, RTF_PRCLONING);
1216 		if (rt->rt_gwroute == rt) {
1217 			rt->rt_gwroute = NULL;
1218 			--rt->rt_refcnt;
1219 			return EDQUOT; /* failure */
1220 		}
1221 	}
1222 
1223 	/*
1224 	 * This isn't going to do anything useful for host routes, so
1225 	 * don't bother.  Also make sure we have a reasonable mask
1226 	 * (we don't yet have one during adds).
1227 	 */
1228 	if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) {
1229 		struct rtfc_arg arg = { rt, rnh };
1230 
1231 		rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
1232 				       (char *)rt_mask(rt),
1233 				       rt_fixchange, &arg);
1234 	}
1235 
1236 	return 0;
1237 }
1238 
1239 static void
1240 rt_maskedcopy(
1241 	struct sockaddr *src,
1242 	struct sockaddr *dst,
1243 	struct sockaddr *netmask)
1244 {
1245 	u_char *cp1 = (u_char *)src;
1246 	u_char *cp2 = (u_char *)dst;
1247 	u_char *cp3 = (u_char *)netmask;
1248 	u_char *cplim = cp2 + *cp3;
1249 	u_char *cplim2 = cp2 + *cp1;
1250 
1251 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1252 	cp3 += 2;
1253 	if (cplim > cplim2)
1254 		cplim = cplim2;
1255 	while (cp2 < cplim)
1256 		*cp2++ = *cp1++ & *cp3++;
1257 	if (cp2 < cplim2)
1258 		bzero(cp2, cplim2 - cp2);
1259 }
1260 
1261 int
1262 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt)
1263 {
1264 	struct rtentry *up_rt, *rt;
1265 
1266 	if (!(rt0->rt_flags & RTF_UP)) {
1267 		up_rt = rtlookup(dst);
1268 		if (up_rt == NULL)
1269 			return (EHOSTUNREACH);
1270 		up_rt->rt_refcnt--;
1271 	} else
1272 		up_rt = rt0;
1273 	if (up_rt->rt_flags & RTF_GATEWAY) {
1274 		if (up_rt->rt_gwroute == NULL) {
1275 			up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
1276 			if (up_rt->rt_gwroute == NULL)
1277 				return (EHOSTUNREACH);
1278 		} else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) {
1279 			rtfree(up_rt->rt_gwroute);
1280 			up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
1281 			if (up_rt->rt_gwroute == NULL)
1282 				return (EHOSTUNREACH);
1283 		}
1284 		rt = up_rt->rt_gwroute;
1285 	} else
1286 		rt = up_rt;
1287 	if (rt->rt_flags & RTF_REJECT &&
1288 	    (rt->rt_rmx.rmx_expire == 0 ||		/* rt doesn't expire */
1289 	     time_second < rt->rt_rmx.rmx_expire))	/* rt not expired */
1290 		return (rt->rt_flags & RTF_HOST ?  EHOSTDOWN : EHOSTUNREACH);
1291 	*drt = rt;
1292 	return 0;
1293 }
1294 
1295 #ifdef ROUTE_DEBUG
1296 
1297 /*
1298  * Print out a route table entry
1299  */
1300 void
1301 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn)
1302 {
1303 	kprintf("rti %p cpu %d route %p flags %08lx: ",
1304 		rtinfo, mycpuid, rn, rn->rt_flags);
1305 	sockaddr_print(rt_key(rn));
1306 	kprintf(" mask ");
1307 	sockaddr_print(rt_mask(rn));
1308 	kprintf(" gw ");
1309 	sockaddr_print(rn->rt_gateway);
1310 	kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?");
1311 	kprintf(" ifa %p\n", rn->rt_ifa);
1312 }
1313 
1314 void
1315 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti)
1316 {
1317 	int didit = 0;
1318 	int i;
1319 
1320 #ifdef ROUTE_DEBUG
1321 	if (cmd == RTM_DELETE && route_debug > 1)
1322 		db_print_backtrace();
1323 #endif
1324 
1325 	switch(cmd) {
1326 	case RTM_ADD:
1327 		kprintf("ADD ");
1328 		break;
1329 	case RTM_RESOLVE:
1330 		kprintf("RES ");
1331 		break;
1332 	case RTM_DELETE:
1333 		kprintf("DEL ");
1334 		break;
1335 	default:
1336 		kprintf("C%02d ", cmd);
1337 		break;
1338 	}
1339 	kprintf("rti %p cpu %d ", rti, mycpuid);
1340 	for (i = 0; i < rti->rti_addrs; ++i) {
1341 		if (rti->rti_info[i] == NULL)
1342 			continue;
1343 		if (didit)
1344 			kprintf(" ,");
1345 		switch(i) {
1346 		case RTAX_DST:
1347 			kprintf("(DST ");
1348 			break;
1349 		case RTAX_GATEWAY:
1350 			kprintf("(GWY ");
1351 			break;
1352 		case RTAX_NETMASK:
1353 			kprintf("(MSK ");
1354 			break;
1355 		case RTAX_GENMASK:
1356 			kprintf("(GEN ");
1357 			break;
1358 		case RTAX_IFP:
1359 			kprintf("(IFP ");
1360 			break;
1361 		case RTAX_IFA:
1362 			kprintf("(IFA ");
1363 			break;
1364 		case RTAX_AUTHOR:
1365 			kprintf("(AUT ");
1366 			break;
1367 		case RTAX_BRD:
1368 			kprintf("(BRD ");
1369 			break;
1370 		default:
1371 			kprintf("(?%02d ", i);
1372 			break;
1373 		}
1374 		sockaddr_print(rti->rti_info[i]);
1375 		kprintf(")");
1376 		didit = 1;
1377 	}
1378 	kprintf("\n");
1379 }
1380 
1381 void
1382 sockaddr_print(struct sockaddr *sa)
1383 {
1384 	struct sockaddr_in *sa4;
1385 	struct sockaddr_in6 *sa6;
1386 	int len;
1387 	int i;
1388 
1389 	if (sa == NULL) {
1390 		kprintf("NULL");
1391 		return;
1392 	}
1393 
1394 	len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]);
1395 
1396 	switch(sa->sa_family) {
1397 	case AF_INET:
1398 	case AF_INET6:
1399 	default:
1400 		switch(sa->sa_family) {
1401 		case AF_INET:
1402 			sa4 = (struct sockaddr_in *)sa;
1403 			kprintf("INET %d %d.%d.%d.%d",
1404 				ntohs(sa4->sin_port),
1405 				(ntohl(sa4->sin_addr.s_addr) >> 24) & 255,
1406 				(ntohl(sa4->sin_addr.s_addr) >> 16) & 255,
1407 				(ntohl(sa4->sin_addr.s_addr) >> 8) & 255,
1408 				(ntohl(sa4->sin_addr.s_addr) >> 0) & 255
1409 			);
1410 			break;
1411 		case AF_INET6:
1412 			sa6 = (struct sockaddr_in6 *)sa;
1413 			kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x",
1414 				ntohs(sa6->sin6_port),
1415 				sa6->sin6_addr.s6_addr16[0],
1416 				sa6->sin6_addr.s6_addr16[1],
1417 				sa6->sin6_addr.s6_addr16[2],
1418 				sa6->sin6_addr.s6_addr16[3],
1419 				sa6->sin6_addr.s6_addr16[4],
1420 				sa6->sin6_addr.s6_addr16[5],
1421 				sa6->sin6_addr.s6_addr16[6],
1422 				sa6->sin6_addr.s6_addr16[7]
1423 			);
1424 			break;
1425 		default:
1426 			kprintf("AF%d ", sa->sa_family);
1427 			while (len > 0 && sa->sa_data[len-1] == 0)
1428 				--len;
1429 
1430 			for (i = 0; i < len; ++i) {
1431 				if (i)
1432 					kprintf(".");
1433 				kprintf("%d", (unsigned char)sa->sa_data[i]);
1434 			}
1435 			break;
1436 		}
1437 	}
1438 }
1439 
1440 #endif
1441 
1442 /*
1443  * Set up a routing table entry, normally for an interface.
1444  */
1445 int
1446 rtinit(struct ifaddr *ifa, int cmd, int flags)
1447 {
1448 	struct sockaddr *dst, *deldst, *netmask;
1449 	struct mbuf *m = NULL;
1450 	struct radix_node_head *rnh;
1451 	struct radix_node *rn;
1452 	struct rt_addrinfo rtinfo;
1453 	int error;
1454 
1455 	if (flags & RTF_HOST) {
1456 		dst = ifa->ifa_dstaddr;
1457 		netmask = NULL;
1458 	} else {
1459 		dst = ifa->ifa_addr;
1460 		netmask = ifa->ifa_netmask;
1461 	}
1462 	/*
1463 	 * If it's a delete, check that if it exists, it's on the correct
1464 	 * interface or we might scrub a route to another ifa which would
1465 	 * be confusing at best and possibly worse.
1466 	 */
1467 	if (cmd == RTM_DELETE) {
1468 		/*
1469 		 * It's a delete, so it should already exist..
1470 		 * If it's a net, mask off the host bits
1471 		 * (Assuming we have a mask)
1472 		 */
1473 		if (netmask != NULL) {
1474 			m = m_get(MB_DONTWAIT, MT_SONAME);
1475 			if (m == NULL)
1476 				return (ENOBUFS);
1477 			deldst = mtod(m, struct sockaddr *);
1478 			rt_maskedcopy(dst, deldst, netmask);
1479 			dst = deldst;
1480 		}
1481 		/*
1482 		 * Look up an rtentry that is in the routing tree and
1483 		 * contains the correct info.
1484 		 */
1485 		if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL ||
1486 		    (rn = rnh->rnh_lookup((char *)dst,
1487 					  (char *)netmask, rnh)) == NULL ||
1488 		    ((struct rtentry *)rn)->rt_ifa != ifa ||
1489 		    !sa_equal((struct sockaddr *)rn->rn_key, dst)) {
1490 			if (m != NULL)
1491 				m_free(m);
1492 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1493 		}
1494 		/* XXX */
1495 #if 0
1496 		else {
1497 			/*
1498 			 * One would think that as we are deleting, and we know
1499 			 * it doesn't exist, we could just return at this point
1500 			 * with an "ELSE" clause, but apparently not..
1501 			 */
1502 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1503 		}
1504 #endif
1505 	}
1506 	/*
1507 	 * Do the actual request
1508 	 */
1509 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
1510 	rtinfo.rti_info[RTAX_DST] = dst;
1511 	rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1512 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
1513 	rtinfo.rti_flags = flags | ifa->ifa_flags;
1514 	rtinfo.rti_ifa = ifa;
1515 	error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa);
1516 	if (m != NULL)
1517 		m_free(m);
1518 	return (error);
1519 }
1520 
1521 static void
1522 rtinit_rtrequest_callback(int cmd, int error,
1523 			  struct rt_addrinfo *rtinfo, struct rtentry *rt,
1524 			  void *arg)
1525 {
1526 	struct ifaddr *ifa = arg;
1527 
1528 	if (error == 0 && rt) {
1529 		if (mycpuid == 0) {
1530 			++rt->rt_refcnt;
1531 			rt_newaddrmsg(cmd, ifa, error, rt);
1532 			--rt->rt_refcnt;
1533 		}
1534 		if (cmd == RTM_DELETE) {
1535 			if (rt->rt_refcnt == 0) {
1536 				++rt->rt_refcnt;
1537 				rtfree(rt);
1538 			}
1539 		}
1540 	}
1541 }
1542 
1543 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */
1544 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
1545