xref: /netbsd-src/sys/net/route.c (revision 627f7eb200a4419d89b531d55fccd2ee3ffdcde0)
1 /*	$NetBSD: route.c,v 1.229 2020/04/08 03:37:14 knakahara Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Kevin M. Lahey of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
35  * 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. Neither the name of the project nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  */
61 
62 /*
63  * Copyright (c) 1980, 1986, 1991, 1993
64  *	The Regents of the University of California.  All rights reserved.
65  *
66  * Redistribution and use in source and binary forms, with or without
67  * modification, are permitted provided that the following conditions
68  * are met:
69  * 1. Redistributions of source code must retain the above copyright
70  *    notice, this list of conditions and the following disclaimer.
71  * 2. Redistributions in binary form must reproduce the above copyright
72  *    notice, this list of conditions and the following disclaimer in the
73  *    documentation and/or other materials provided with the distribution.
74  * 3. Neither the name of the University nor the names of its contributors
75  *    may be used to endorse or promote products derived from this software
76  *    without specific prior written permission.
77  *
78  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
79  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
80  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
81  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
82  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
83  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
84  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
85  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
86  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
87  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
88  * SUCH DAMAGE.
89  *
90  *	@(#)route.c	8.3 (Berkeley) 1/9/95
91  */
92 
93 #ifdef _KERNEL_OPT
94 #include "opt_inet.h"
95 #include "opt_route.h"
96 #include "opt_net_mpsafe.h"
97 #endif
98 
99 #include <sys/cdefs.h>
100 __KERNEL_RCSID(0, "$NetBSD: route.c,v 1.229 2020/04/08 03:37:14 knakahara Exp $");
101 
102 #include <sys/param.h>
103 #ifdef RTFLUSH_DEBUG
104 #include <sys/sysctl.h>
105 #endif
106 #include <sys/systm.h>
107 #include <sys/callout.h>
108 #include <sys/proc.h>
109 #include <sys/mbuf.h>
110 #include <sys/socket.h>
111 #include <sys/socketvar.h>
112 #include <sys/domain.h>
113 #include <sys/kernel.h>
114 #include <sys/ioctl.h>
115 #include <sys/pool.h>
116 #include <sys/kauth.h>
117 #include <sys/workqueue.h>
118 #include <sys/syslog.h>
119 #include <sys/rwlock.h>
120 #include <sys/mutex.h>
121 #include <sys/cpu.h>
122 #include <sys/kmem.h>
123 
124 #include <net/if.h>
125 #include <net/if_dl.h>
126 #include <net/route.h>
127 #if defined(INET) || defined(INET6)
128 #include <net/if_llatbl.h>
129 #endif
130 
131 #include <netinet/in.h>
132 #include <netinet/in_var.h>
133 
134 #define	PRESERVED_RTF	(RTF_UP | RTF_GATEWAY | RTF_HOST | RTF_DONE | RTF_MASK)
135 
136 #ifdef RTFLUSH_DEBUG
137 #define	rtcache_debug() __predict_false(_rtcache_debug)
138 #else /* RTFLUSH_DEBUG */
139 #define	rtcache_debug() 0
140 #endif /* RTFLUSH_DEBUG */
141 
142 #ifdef RT_DEBUG
143 #define RT_REFCNT_TRACE(rt)	printf("%s:%d: rt=%p refcnt=%d\n", \
144 				    __func__, __LINE__, (rt), (rt)->rt_refcnt)
145 #else
146 #define RT_REFCNT_TRACE(rt)	do {} while (0)
147 #endif
148 
149 #ifdef RT_DEBUG
150 #define dlog(level, fmt, args...)	log(level, fmt, ##args)
151 #else
152 #define dlog(level, fmt, args...)	do {} while (0)
153 #endif
154 
155 struct rtstat		rtstat;
156 
157 static int		rttrash;	/* routes not in table but not freed */
158 
159 static struct pool	rtentry_pool;
160 static struct pool	rttimer_pool;
161 
162 static struct callout	rt_timer_ch; /* callout for rt_timer_timer() */
163 static struct workqueue	*rt_timer_wq;
164 static struct work	rt_timer_wk;
165 
166 static void	rt_timer_init(void);
167 static void	rt_timer_queue_remove_all(struct rttimer_queue *);
168 static void	rt_timer_remove_all(struct rtentry *);
169 static void	rt_timer_timer(void *);
170 
171 /*
172  * Locking notes:
173  * - The routing table is protected by a global rwlock
174  *   - API: RT_RLOCK and friends
175  * - rtcaches are NOT protected by the framework
176  *   - Callers must guarantee a rtcache isn't accessed simultaneously
177  *   - How the constraint is guranteed in the wild
178  *     - Protect a rtcache by a mutex (e.g., inp_route)
179  *     - Make rtcache per-CPU and allow only accesses from softint
180  *       (e.g., ipforward_rt_percpu)
181  * - References to a rtentry is managed by reference counting and psref
182  *   - Reference counting is used for temporal reference when a rtentry
183  *     is fetched from the routing table
184  *   - psref is used for temporal reference when a rtentry is fetched
185  *     from a rtcache
186  *     - struct route (rtcache) has struct psref, so we cannot obtain
187  *       a reference twice on the same struct route
188  *   - Before destroying or updating a rtentry, we have to wait for
189  *     all references left (see below for details)
190  *   - APIs
191  *     - An obtained rtentry via rtalloc1 or rtrequest* must be
192  *       unreferenced by rt_unref
193  *     - An obtained rtentry via rtcache_* must be unreferenced by
194  *       rtcache_unref
195  *   - TODO: once we get a lockless routing table, we should use only
196  *           psref for rtentries
197  * - rtentry destruction
198  *   - A rtentry is destroyed (freed) only when we call rtrequest(RTM_DELETE)
199  *   - If a caller of rtrequest grabs a reference of a rtentry, the caller
200  *     has a responsibility to destroy the rtentry by itself by calling
201  *     rt_free
202  *     - If not, rtrequest itself does that
203  *   - If rt_free is called in softint, the actual destruction routine is
204  *     deferred to a workqueue
205  * - rtentry update
206  *   - When updating a rtentry, RTF_UPDATING flag is set
207  *   - If a rtentry is set RTF_UPDATING, fetching the rtentry from
208  *     the routing table or a rtcache results in either of the following
209  *     cases:
210  *     - if the caller runs in softint, the caller fails to fetch
211  *     - otherwise, the caller waits for the update completed and retries
212  *       to fetch (probably succeed to fetch for the second time)
213  * - rtcache invalidation
214  *   - There is a global generation counter that is incremented when
215  *     any routes have been added or deleted
216  *   - When a rtcache caches a rtentry into itself, it also stores
217  *     a snapshot of the generation counter
218  *   - If the snapshot equals to the global counter, the cache is valid,
219  *     otherwise the cache is invalidated
220  */
221 
222 /*
223  * Global lock for the routing table.
224  */
225 static krwlock_t		rt_lock __cacheline_aligned;
226 #ifdef NET_MPSAFE
227 #define RT_RLOCK()		rw_enter(&rt_lock, RW_READER)
228 #define RT_WLOCK()		rw_enter(&rt_lock, RW_WRITER)
229 #define RT_UNLOCK()		rw_exit(&rt_lock)
230 #define RT_WLOCKED()		rw_write_held(&rt_lock)
231 #define	RT_ASSERT_WLOCK()	KASSERT(rw_write_held(&rt_lock))
232 #else
233 #define RT_RLOCK()		do {} while (0)
234 #define RT_WLOCK()		do {} while (0)
235 #define RT_UNLOCK()		do {} while (0)
236 #define RT_WLOCKED()		true
237 #define	RT_ASSERT_WLOCK()	do {} while (0)
238 #endif
239 
240 static uint64_t rtcache_generation;
241 
242 /*
243  * mutex and cv that are used to wait for references to a rtentry left
244  * before updating the rtentry.
245  */
246 static struct {
247 	kmutex_t		lock;
248 	kcondvar_t		cv;
249 	bool			ongoing;
250 	const struct lwp	*lwp;
251 } rt_update_global __cacheline_aligned;
252 
253 /*
254  * A workqueue and stuff that are used to defer the destruction routine
255  * of rtentries.
256  */
257 static struct {
258 	struct workqueue	*wq;
259 	struct work		wk;
260 	kmutex_t		lock;
261 	SLIST_HEAD(, rtentry)	queue;
262 	bool			enqueued;
263 } rt_free_global __cacheline_aligned;
264 
265 /* psref for rtentry */
266 static struct psref_class *rt_psref_class __read_mostly;
267 
268 #ifdef RTFLUSH_DEBUG
269 static int _rtcache_debug = 0;
270 #endif /* RTFLUSH_DEBUG */
271 
272 static kauth_listener_t route_listener;
273 
274 static int rtdeletemsg(struct rtentry *);
275 
276 static void rt_maskedcopy(const struct sockaddr *,
277     struct sockaddr *, const struct sockaddr *);
278 
279 static void rtcache_invalidate(void);
280 
281 static void rt_ref(struct rtentry *);
282 
283 static struct rtentry *
284     rtalloc1_locked(const struct sockaddr *, int, bool, bool);
285 
286 static struct ifaddr *rt_getifa(struct rt_addrinfo *, struct psref *);
287 static struct ifnet *rt_getifp(struct rt_addrinfo *, struct psref *);
288 static struct ifaddr *ifa_ifwithroute_psref(int, const struct sockaddr *,
289     const struct sockaddr *, struct psref *);
290 
291 static void rtcache_ref(struct rtentry *, struct route *);
292 
293 #ifdef NET_MPSAFE
294 static void rt_update_wait(void);
295 #endif
296 
297 static bool rt_wait_ok(void);
298 static void rt_wait_refcnt(const char *, struct rtentry *, int);
299 static void rt_wait_psref(struct rtentry *);
300 
301 #ifdef DDB
302 static void db_print_sa(const struct sockaddr *);
303 static void db_print_ifa(struct ifaddr *);
304 static int db_show_rtentry(struct rtentry *, void *);
305 #endif
306 
307 #ifdef RTFLUSH_DEBUG
308 static void sysctl_net_rtcache_setup(struct sysctllog **);
309 static void
310 sysctl_net_rtcache_setup(struct sysctllog **clog)
311 {
312 	const struct sysctlnode *rnode;
313 
314 	if (sysctl_createv(clog, 0, NULL, &rnode, CTLFLAG_PERMANENT,
315 	    CTLTYPE_NODE,
316 	    "rtcache", SYSCTL_DESCR("Route cache related settings"),
317 	    NULL, 0, NULL, 0, CTL_NET, CTL_CREATE, CTL_EOL) != 0)
318 		return;
319 	if (sysctl_createv(clog, 0, &rnode, &rnode,
320 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE, CTLTYPE_INT,
321 	    "debug", SYSCTL_DESCR("Debug route caches"),
322 	    NULL, 0, &_rtcache_debug, 0, CTL_CREATE, CTL_EOL) != 0)
323 		return;
324 }
325 #endif /* RTFLUSH_DEBUG */
326 
327 static inline void
328 rt_destroy(struct rtentry *rt)
329 {
330 	if (rt->_rt_key != NULL)
331 		sockaddr_free(rt->_rt_key);
332 	if (rt->rt_gateway != NULL)
333 		sockaddr_free(rt->rt_gateway);
334 	if (rt_gettag(rt) != NULL)
335 		sockaddr_free(rt_gettag(rt));
336 	rt->_rt_key = rt->rt_gateway = rt->rt_tag = NULL;
337 }
338 
339 static inline const struct sockaddr *
340 rt_setkey(struct rtentry *rt, const struct sockaddr *key, int flags)
341 {
342 	if (rt->_rt_key == key)
343 		goto out;
344 
345 	if (rt->_rt_key != NULL)
346 		sockaddr_free(rt->_rt_key);
347 	rt->_rt_key = sockaddr_dup(key, flags);
348 out:
349 	rt->rt_nodes->rn_key = (const char *)rt->_rt_key;
350 	return rt->_rt_key;
351 }
352 
353 struct ifaddr *
354 rt_get_ifa(struct rtentry *rt)
355 {
356 	struct ifaddr *ifa;
357 
358 	ifa = rt->rt_ifa;
359 	if (ifa->ifa_getifa == NULL)
360 		return ifa;
361 #if 0
362 	else if (ifa->ifa_seqno != NULL && *ifa->ifa_seqno == rt->rt_ifa_seqno)
363 		return ifa;
364 #endif
365 	else {
366 		ifa = (*ifa->ifa_getifa)(ifa, rt_getkey(rt));
367 		if (ifa == NULL)
368 			return NULL;
369 		rt_replace_ifa(rt, ifa);
370 		return ifa;
371 	}
372 }
373 
374 static void
375 rt_set_ifa1(struct rtentry *rt, struct ifaddr *ifa)
376 {
377 	rt->rt_ifa = ifa;
378 	if (ifa->ifa_seqno != NULL)
379 		rt->rt_ifa_seqno = *ifa->ifa_seqno;
380 }
381 
382 /*
383  * Is this route the connected route for the ifa?
384  */
385 static int
386 rt_ifa_connected(const struct rtentry *rt, const struct ifaddr *ifa)
387 {
388 	const struct sockaddr *key, *dst, *odst;
389 	struct sockaddr_storage maskeddst;
390 
391 	key = rt_getkey(rt);
392 	dst = rt->rt_flags & RTF_HOST ? ifa->ifa_dstaddr : ifa->ifa_addr;
393 	if (dst == NULL ||
394 	    dst->sa_family != key->sa_family ||
395 	    dst->sa_len != key->sa_len)
396 		return 0;
397 	if ((rt->rt_flags & RTF_HOST) == 0 && ifa->ifa_netmask) {
398 		odst = dst;
399 		dst = (struct sockaddr *)&maskeddst;
400 		rt_maskedcopy(odst, (struct sockaddr *)&maskeddst,
401 		    ifa->ifa_netmask);
402 	}
403 	return (memcmp(dst, key, dst->sa_len) == 0);
404 }
405 
406 void
407 rt_replace_ifa(struct rtentry *rt, struct ifaddr *ifa)
408 {
409 	struct ifaddr *old;
410 
411 	if (rt->rt_ifa == ifa)
412 		return;
413 
414 	if (rt->rt_ifa != ifa &&
415 	    rt->rt_ifa->ifa_flags & IFA_ROUTE &&
416 	    rt_ifa_connected(rt, rt->rt_ifa))
417 	{
418 		RT_DPRINTF("rt->_rt_key = %p, ifa = %p, "
419 		    "replace deleted IFA_ROUTE\n",
420 		    (void *)rt->_rt_key, (void *)rt->rt_ifa);
421 		rt->rt_ifa->ifa_flags &= ~IFA_ROUTE;
422 		if (rt_ifa_connected(rt, ifa)) {
423 			RT_DPRINTF("rt->_rt_key = %p, ifa = %p, "
424 			    "replace added IFA_ROUTE\n",
425 			    (void *)rt->_rt_key, (void *)ifa);
426 			ifa->ifa_flags |= IFA_ROUTE;
427 		}
428 	}
429 
430 	ifaref(ifa);
431 	old = rt->rt_ifa;
432 	rt_set_ifa1(rt, ifa);
433 	ifafree(old);
434 }
435 
436 static void
437 rt_set_ifa(struct rtentry *rt, struct ifaddr *ifa)
438 {
439 	ifaref(ifa);
440 	rt_set_ifa1(rt, ifa);
441 }
442 
443 static int
444 route_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
445     void *arg0, void *arg1, void *arg2, void *arg3)
446 {
447 	struct rt_msghdr *rtm;
448 	int result;
449 
450 	result = KAUTH_RESULT_DEFER;
451 	rtm = arg1;
452 
453 	if (action != KAUTH_NETWORK_ROUTE)
454 		return result;
455 
456 	if (rtm->rtm_type == RTM_GET)
457 		result = KAUTH_RESULT_ALLOW;
458 
459 	return result;
460 }
461 
462 static void rt_free_work(struct work *, void *);
463 
464 void
465 rt_init(void)
466 {
467 	int error;
468 
469 #ifdef RTFLUSH_DEBUG
470 	sysctl_net_rtcache_setup(NULL);
471 #endif
472 
473 	mutex_init(&rt_free_global.lock, MUTEX_DEFAULT, IPL_SOFTNET);
474 	SLIST_INIT(&rt_free_global.queue);
475 	rt_free_global.enqueued = false;
476 
477 	rt_psref_class = psref_class_create("rtentry", IPL_SOFTNET);
478 
479 	error = workqueue_create(&rt_free_global.wq, "rt_free",
480 	    rt_free_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE);
481 	if (error)
482 		panic("%s: workqueue_create failed (%d)\n", __func__, error);
483 
484 	mutex_init(&rt_update_global.lock, MUTEX_DEFAULT, IPL_SOFTNET);
485 	cv_init(&rt_update_global.cv, "rt_update");
486 
487 	pool_init(&rtentry_pool, sizeof(struct rtentry), 0, 0, 0, "rtentpl",
488 	    NULL, IPL_SOFTNET);
489 	pool_init(&rttimer_pool, sizeof(struct rttimer), 0, 0, 0, "rttmrpl",
490 	    NULL, IPL_SOFTNET);
491 
492 	rn_init();	/* initialize all zeroes, all ones, mask table */
493 	rtbl_init();
494 
495 	route_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
496 	    route_listener_cb, NULL);
497 }
498 
499 static void
500 rtcache_invalidate(void)
501 {
502 
503 	RT_ASSERT_WLOCK();
504 
505 	if (rtcache_debug())
506 		printf("%s: enter\n", __func__);
507 
508 	rtcache_generation++;
509 }
510 
511 #ifdef RT_DEBUG
512 static void
513 dump_rt(const struct rtentry *rt)
514 {
515 	char buf[512];
516 
517 	log(LOG_DEBUG, "rt: ");
518 	log(LOG_DEBUG, "p=%p ", rt);
519 	if (rt->_rt_key == NULL) {
520 		log(LOG_DEBUG, "dst=(NULL) ");
521 	} else {
522 		sockaddr_format(rt->_rt_key, buf, sizeof(buf));
523 		log(LOG_DEBUG, "dst=%s ", buf);
524 	}
525 	if (rt->rt_gateway == NULL) {
526 		log(LOG_DEBUG, "gw=(NULL) ");
527 	} else {
528 		sockaddr_format(rt->_rt_key, buf, sizeof(buf));
529 		log(LOG_DEBUG, "gw=%s ", buf);
530 	}
531 	log(LOG_DEBUG, "flags=%x ", rt->rt_flags);
532 	if (rt->rt_ifp == NULL) {
533 		log(LOG_DEBUG, "if=(NULL) ");
534 	} else {
535 		log(LOG_DEBUG, "if=%s ", rt->rt_ifp->if_xname);
536 	}
537 	log(LOG_DEBUG, "\n");
538 }
539 #endif /* RT_DEBUG */
540 
541 /*
542  * Packet routing routines. If success, refcnt of a returned rtentry
543  * will be incremented. The caller has to rtfree it by itself.
544  */
545 struct rtentry *
546 rtalloc1_locked(const struct sockaddr *dst, int report, bool wait_ok,
547     bool wlock)
548 {
549 	rtbl_t *rtbl;
550 	struct rtentry *rt;
551 	int s;
552 
553 #ifdef NET_MPSAFE
554 retry:
555 #endif
556 	s = splsoftnet();
557 	rtbl = rt_gettable(dst->sa_family);
558 	if (rtbl == NULL)
559 		goto miss;
560 
561 	rt = rt_matchaddr(rtbl, dst);
562 	if (rt == NULL)
563 		goto miss;
564 
565 	if (!ISSET(rt->rt_flags, RTF_UP))
566 		goto miss;
567 
568 #ifdef NET_MPSAFE
569 	if (ISSET(rt->rt_flags, RTF_UPDATING) &&
570 	    /* XXX updater should be always able to acquire */
571 	    curlwp != rt_update_global.lwp) {
572 		if (!wait_ok || !rt_wait_ok())
573 			goto miss;
574 		RT_UNLOCK();
575 		splx(s);
576 
577 		/* We can wait until the update is complete */
578 		rt_update_wait();
579 
580 		if (wlock)
581 			RT_WLOCK();
582 		else
583 			RT_RLOCK();
584 		goto retry;
585 	}
586 #endif /* NET_MPSAFE */
587 
588 	rt_ref(rt);
589 	RT_REFCNT_TRACE(rt);
590 
591 	splx(s);
592 	return rt;
593 miss:
594 	rtstat.rts_unreach++;
595 	if (report) {
596 		struct rt_addrinfo info;
597 
598 		memset(&info, 0, sizeof(info));
599 		info.rti_info[RTAX_DST] = dst;
600 		rt_missmsg(RTM_MISS, &info, 0, 0);
601 	}
602 	splx(s);
603 	return NULL;
604 }
605 
606 struct rtentry *
607 rtalloc1(const struct sockaddr *dst, int report)
608 {
609 	struct rtentry *rt;
610 
611 	RT_RLOCK();
612 	rt = rtalloc1_locked(dst, report, true, false);
613 	RT_UNLOCK();
614 
615 	return rt;
616 }
617 
618 static void
619 rt_ref(struct rtentry *rt)
620 {
621 
622 	KASSERTMSG(rt->rt_refcnt >= 0, "rt_refcnt=%d", rt->rt_refcnt);
623 	atomic_inc_uint(&rt->rt_refcnt);
624 }
625 
626 void
627 rt_unref(struct rtentry *rt)
628 {
629 
630 	KASSERT(rt != NULL);
631 	KASSERTMSG(rt->rt_refcnt > 0, "refcnt=%d", rt->rt_refcnt);
632 
633 	atomic_dec_uint(&rt->rt_refcnt);
634 	if (!ISSET(rt->rt_flags, RTF_UP) || ISSET(rt->rt_flags, RTF_UPDATING)) {
635 		mutex_enter(&rt_free_global.lock);
636 		cv_broadcast(&rt->rt_cv);
637 		mutex_exit(&rt_free_global.lock);
638 	}
639 }
640 
641 static bool
642 rt_wait_ok(void)
643 {
644 
645 	KASSERT(!cpu_intr_p());
646 	return !cpu_softintr_p();
647 }
648 
649 void
650 rt_wait_refcnt(const char *title, struct rtentry *rt, int cnt)
651 {
652 	mutex_enter(&rt_free_global.lock);
653 	while (rt->rt_refcnt > cnt) {
654 		dlog(LOG_DEBUG, "%s: %s waiting (refcnt=%d)\n",
655 		    __func__, title, rt->rt_refcnt);
656 		cv_wait(&rt->rt_cv, &rt_free_global.lock);
657 		dlog(LOG_DEBUG, "%s: %s waited (refcnt=%d)\n",
658 		    __func__, title, rt->rt_refcnt);
659 	}
660 	mutex_exit(&rt_free_global.lock);
661 }
662 
663 void
664 rt_wait_psref(struct rtentry *rt)
665 {
666 
667 	psref_target_destroy(&rt->rt_psref, rt_psref_class);
668 	psref_target_init(&rt->rt_psref, rt_psref_class);
669 }
670 
671 static void
672 _rt_free(struct rtentry *rt)
673 {
674 	struct ifaddr *ifa;
675 
676 	/*
677 	 * Need to avoid a deadlock on rt_wait_refcnt of update
678 	 * and a conflict on psref_target_destroy of update.
679 	 */
680 #ifdef NET_MPSAFE
681 	rt_update_wait();
682 #endif
683 
684 	RT_REFCNT_TRACE(rt);
685 	KASSERTMSG(rt->rt_refcnt >= 0, "refcnt=%d", rt->rt_refcnt);
686 	rt_wait_refcnt("free", rt, 0);
687 #ifdef NET_MPSAFE
688 	psref_target_destroy(&rt->rt_psref, rt_psref_class);
689 #endif
690 
691 	rt_assert_inactive(rt);
692 	rttrash--;
693 	ifa = rt->rt_ifa;
694 	rt->rt_ifa = NULL;
695 	ifafree(ifa);
696 	rt->rt_ifp = NULL;
697 	cv_destroy(&rt->rt_cv);
698 	rt_destroy(rt);
699 	pool_put(&rtentry_pool, rt);
700 }
701 
702 static void
703 rt_free_work(struct work *wk, void *arg)
704 {
705 
706 	for (;;) {
707 		struct rtentry *rt;
708 
709 		mutex_enter(&rt_free_global.lock);
710 		if ((rt = SLIST_FIRST(&rt_free_global.queue)) == NULL) {
711 			rt_free_global.enqueued = false;
712 			mutex_exit(&rt_free_global.lock);
713 			return;
714 		}
715 		SLIST_REMOVE_HEAD(&rt_free_global.queue, rt_free);
716 		mutex_exit(&rt_free_global.lock);
717 		atomic_dec_uint(&rt->rt_refcnt);
718 		_rt_free(rt);
719 	}
720 }
721 
722 void
723 rt_free(struct rtentry *rt)
724 {
725 
726 	KASSERTMSG(rt->rt_refcnt > 0, "rt_refcnt=%d", rt->rt_refcnt);
727 	if (rt_wait_ok()) {
728 		atomic_dec_uint(&rt->rt_refcnt);
729 		_rt_free(rt);
730 		return;
731 	}
732 
733 	mutex_enter(&rt_free_global.lock);
734 	/* No need to add a reference here. */
735 	SLIST_INSERT_HEAD(&rt_free_global.queue, rt, rt_free);
736 	if (!rt_free_global.enqueued) {
737 		workqueue_enqueue(rt_free_global.wq, &rt_free_global.wk, NULL);
738 		rt_free_global.enqueued = true;
739 	}
740 	mutex_exit(&rt_free_global.lock);
741 }
742 
743 #ifdef NET_MPSAFE
744 static void
745 rt_update_wait(void)
746 {
747 
748 	mutex_enter(&rt_update_global.lock);
749 	while (rt_update_global.ongoing) {
750 		dlog(LOG_DEBUG, "%s: waiting lwp=%p\n", __func__, curlwp);
751 		cv_wait(&rt_update_global.cv, &rt_update_global.lock);
752 		dlog(LOG_DEBUG, "%s: waited lwp=%p\n", __func__, curlwp);
753 	}
754 	mutex_exit(&rt_update_global.lock);
755 }
756 #endif
757 
758 int
759 rt_update_prepare(struct rtentry *rt)
760 {
761 
762 	dlog(LOG_DEBUG, "%s: updating rt=%p lwp=%p\n", __func__, rt, curlwp);
763 
764 	RT_WLOCK();
765 	/* If the entry is being destroyed, don't proceed the update. */
766 	if (!ISSET(rt->rt_flags, RTF_UP)) {
767 		RT_UNLOCK();
768 		return ESRCH;
769 	}
770 	rt->rt_flags |= RTF_UPDATING;
771 	RT_UNLOCK();
772 
773 	mutex_enter(&rt_update_global.lock);
774 	while (rt_update_global.ongoing) {
775 		dlog(LOG_DEBUG, "%s: waiting ongoing updating rt=%p lwp=%p\n",
776 		    __func__, rt, curlwp);
777 		cv_wait(&rt_update_global.cv, &rt_update_global.lock);
778 		dlog(LOG_DEBUG, "%s: waited ongoing updating rt=%p lwp=%p\n",
779 		    __func__, rt, curlwp);
780 	}
781 	rt_update_global.ongoing = true;
782 	/* XXX need it to avoid rt_update_wait by updater itself. */
783 	rt_update_global.lwp = curlwp;
784 	mutex_exit(&rt_update_global.lock);
785 
786 	rt_wait_refcnt("update", rt, 1);
787 	rt_wait_psref(rt);
788 
789 	return 0;
790 }
791 
792 void
793 rt_update_finish(struct rtentry *rt)
794 {
795 
796 	RT_WLOCK();
797 	rt->rt_flags &= ~RTF_UPDATING;
798 	RT_UNLOCK();
799 
800 	mutex_enter(&rt_update_global.lock);
801 	rt_update_global.ongoing = false;
802 	rt_update_global.lwp = NULL;
803 	cv_broadcast(&rt_update_global.cv);
804 	mutex_exit(&rt_update_global.lock);
805 
806 	dlog(LOG_DEBUG, "%s: updated rt=%p lwp=%p\n", __func__, rt, curlwp);
807 }
808 
809 /*
810  * Force a routing table entry to the specified
811  * destination to go through the given gateway.
812  * Normally called as a result of a routing redirect
813  * message from the network layer.
814  *
815  * N.B.: must be called at splsoftnet
816  */
817 void
818 rtredirect(const struct sockaddr *dst, const struct sockaddr *gateway,
819 	const struct sockaddr *netmask, int flags, const struct sockaddr *src,
820 	struct rtentry **rtp)
821 {
822 	struct rtentry *rt;
823 	int error = 0;
824 	uint64_t *stat = NULL;
825 	struct rt_addrinfo info;
826 	struct ifaddr *ifa;
827 	struct psref psref;
828 
829 	/* verify the gateway is directly reachable */
830 	if ((ifa = ifa_ifwithnet_psref(gateway, &psref)) == NULL) {
831 		error = ENETUNREACH;
832 		goto out;
833 	}
834 	rt = rtalloc1(dst, 0);
835 	/*
836 	 * If the redirect isn't from our current router for this dst,
837 	 * it's either old or wrong.  If it redirects us to ourselves,
838 	 * we have a routing loop, perhaps as a result of an interface
839 	 * going down recently.
840 	 */
841 	if (!(flags & RTF_DONE) && rt &&
842 	     (sockaddr_cmp(src, rt->rt_gateway) != 0 || rt->rt_ifa != ifa))
843 		error = EINVAL;
844 	else {
845 		int s = pserialize_read_enter();
846 		struct ifaddr *_ifa;
847 
848 		_ifa = ifa_ifwithaddr(gateway);
849 		if (_ifa != NULL)
850 			error = EHOSTUNREACH;
851 		pserialize_read_exit(s);
852 	}
853 	if (error)
854 		goto done;
855 	/*
856 	 * Create a new entry if we just got back a wildcard entry
857 	 * or the lookup failed.  This is necessary for hosts
858 	 * which use routing redirects generated by smart gateways
859 	 * to dynamically build the routing tables.
860 	 */
861 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
862 		goto create;
863 	/*
864 	 * Don't listen to the redirect if it's
865 	 * for a route to an interface.
866 	 */
867 	if (rt->rt_flags & RTF_GATEWAY) {
868 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
869 			/*
870 			 * Changing from route to net => route to host.
871 			 * Create new route, rather than smashing route to net.
872 			 */
873 		create:
874 			if (rt != NULL)
875 				rt_unref(rt);
876 			flags |=  RTF_GATEWAY | RTF_DYNAMIC;
877 			memset(&info, 0, sizeof(info));
878 			info.rti_info[RTAX_DST] = dst;
879 			info.rti_info[RTAX_GATEWAY] = gateway;
880 			info.rti_info[RTAX_NETMASK] = netmask;
881 			info.rti_ifa = ifa;
882 			info.rti_flags = flags;
883 			rt = NULL;
884 			error = rtrequest1(RTM_ADD, &info, &rt);
885 			if (rt != NULL)
886 				flags = rt->rt_flags;
887 			stat = &rtstat.rts_dynamic;
888 		} else {
889 			/*
890 			 * Smash the current notion of the gateway to
891 			 * this destination.  Should check about netmask!!!
892 			 */
893 #ifdef NET_MPSAFE
894 			KASSERT(!cpu_softintr_p());
895 
896 			error = rt_update_prepare(rt);
897 			if (error == 0) {
898 #endif
899 				RT_WLOCK();
900 				error = rt_setgate(rt, gateway);
901 				if (error == 0) {
902 					rt->rt_flags |= RTF_MODIFIED;
903 					flags |= RTF_MODIFIED;
904 				}
905 				RT_UNLOCK();
906 #ifdef NET_MPSAFE
907 				rt_update_finish(rt);
908 			} else {
909 				/*
910 				 * If error != 0, the rtentry is being
911 				 * destroyed, so doing nothing doesn't
912 				 * matter.
913 				 */
914 			}
915 #endif
916 			stat = &rtstat.rts_newgateway;
917 		}
918 	} else
919 		error = EHOSTUNREACH;
920 done:
921 	if (rt) {
922 		if (rtp != NULL && !error)
923 			*rtp = rt;
924 		else
925 			rt_unref(rt);
926 	}
927 out:
928 	if (error)
929 		rtstat.rts_badredirect++;
930 	else if (stat != NULL)
931 		(*stat)++;
932 	memset(&info, 0, sizeof(info));
933 	info.rti_info[RTAX_DST] = dst;
934 	info.rti_info[RTAX_GATEWAY] = gateway;
935 	info.rti_info[RTAX_NETMASK] = netmask;
936 	info.rti_info[RTAX_AUTHOR] = src;
937 	rt_missmsg(RTM_REDIRECT, &info, flags, error);
938 	ifa_release(ifa, &psref);
939 }
940 
941 /*
942  * Delete a route and generate a message.
943  * It doesn't free a passed rt.
944  */
945 static int
946 rtdeletemsg(struct rtentry *rt)
947 {
948 	int error;
949 	struct rt_addrinfo info;
950 	struct rtentry *retrt;
951 
952 	/*
953 	 * Request the new route so that the entry is not actually
954 	 * deleted.  That will allow the information being reported to
955 	 * be accurate (and consistent with route_output()).
956 	 */
957 	memset(&info, 0, sizeof(info));
958 	info.rti_info[RTAX_DST] = rt_getkey(rt);
959 	info.rti_info[RTAX_NETMASK] = rt_mask(rt);
960 	info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
961 	info.rti_flags = rt->rt_flags;
962 	error = rtrequest1(RTM_DELETE, &info, &retrt);
963 
964 	rt_missmsg(RTM_DELETE, &info, info.rti_flags, error);
965 
966 	return error;
967 }
968 
969 static struct ifaddr *
970 ifa_ifwithroute_psref(int flags, const struct sockaddr *dst,
971     const struct sockaddr *gateway, struct psref *psref)
972 {
973 	struct ifaddr *ifa = NULL;
974 
975 	if ((flags & RTF_GATEWAY) == 0) {
976 		/*
977 		 * If we are adding a route to an interface,
978 		 * and the interface is a pt to pt link
979 		 * we should search for the destination
980 		 * as our clue to the interface.  Otherwise
981 		 * we can use the local address.
982 		 */
983 		if ((flags & RTF_HOST) && gateway->sa_family != AF_LINK)
984 			ifa = ifa_ifwithdstaddr_psref(dst, psref);
985 		if (ifa == NULL)
986 			ifa = ifa_ifwithaddr_psref(gateway, psref);
987 	} else {
988 		/*
989 		 * If we are adding a route to a remote net
990 		 * or host, the gateway may still be on the
991 		 * other end of a pt to pt link.
992 		 */
993 		ifa = ifa_ifwithdstaddr_psref(gateway, psref);
994 	}
995 	if (ifa == NULL)
996 		ifa = ifa_ifwithnet_psref(gateway, psref);
997 	if (ifa == NULL) {
998 		int s;
999 		struct rtentry *rt;
1000 
1001 		rt = rtalloc1_locked(gateway, 0, true, true);
1002 		if (rt == NULL)
1003 			return NULL;
1004 		if (rt->rt_flags & RTF_GATEWAY) {
1005 			rt_unref(rt);
1006 			return NULL;
1007 		}
1008 		/*
1009 		 * Just in case. May not need to do this workaround.
1010 		 * Revisit when working on rtentry MP-ification.
1011 		 */
1012 		s = pserialize_read_enter();
1013 		IFADDR_READER_FOREACH(ifa, rt->rt_ifp) {
1014 			if (ifa == rt->rt_ifa)
1015 				break;
1016 		}
1017 		if (ifa != NULL)
1018 			ifa_acquire(ifa, psref);
1019 		pserialize_read_exit(s);
1020 		rt_unref(rt);
1021 		if (ifa == NULL)
1022 			return NULL;
1023 	}
1024 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
1025 		struct ifaddr *nifa;
1026 		int s;
1027 
1028 		s = pserialize_read_enter();
1029 		nifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
1030 		if (nifa != NULL) {
1031 			ifa_release(ifa, psref);
1032 			ifa_acquire(nifa, psref);
1033 			ifa = nifa;
1034 		}
1035 		pserialize_read_exit(s);
1036 	}
1037 	return ifa;
1038 }
1039 
1040 /*
1041  * If it suceeds and ret_nrt isn't NULL, refcnt of ret_nrt is incremented.
1042  * The caller has to rtfree it by itself.
1043  */
1044 int
1045 rtrequest(int req, const struct sockaddr *dst, const struct sockaddr *gateway,
1046 	const struct sockaddr *netmask, int flags, struct rtentry **ret_nrt)
1047 {
1048 	struct rt_addrinfo info;
1049 
1050 	memset(&info, 0, sizeof(info));
1051 	info.rti_flags = flags;
1052 	info.rti_info[RTAX_DST] = dst;
1053 	info.rti_info[RTAX_GATEWAY] = gateway;
1054 	info.rti_info[RTAX_NETMASK] = netmask;
1055 	return rtrequest1(req, &info, ret_nrt);
1056 }
1057 
1058 /*
1059  * It's a utility function to add/remove a route to/from the routing table
1060  * and tell user processes the addition/removal on success.
1061  */
1062 int
1063 rtrequest_newmsg(const int req, const struct sockaddr *dst,
1064 	const struct sockaddr *gateway, const struct sockaddr *netmask,
1065 	const int flags)
1066 {
1067 	int error;
1068 	struct rtentry *ret_nrt = NULL;
1069 
1070 	KASSERT(req == RTM_ADD || req == RTM_DELETE);
1071 
1072 	error = rtrequest(req, dst, gateway, netmask, flags, &ret_nrt);
1073 	if (error != 0)
1074 		return error;
1075 
1076 	KASSERT(ret_nrt != NULL);
1077 
1078 	rt_newmsg(req, ret_nrt); /* tell user process */
1079 	if (req == RTM_DELETE)
1080 		rt_free(ret_nrt);
1081 	else
1082 		rt_unref(ret_nrt);
1083 
1084 	return 0;
1085 }
1086 
1087 static struct ifnet *
1088 rt_getifp(struct rt_addrinfo *info, struct psref *psref)
1089 {
1090 	const struct sockaddr *ifpaddr = info->rti_info[RTAX_IFP];
1091 
1092 	if (info->rti_ifp != NULL)
1093 		return NULL;
1094 	/*
1095 	 * ifp may be specified by sockaddr_dl when protocol address
1096 	 * is ambiguous
1097 	 */
1098 	if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) {
1099 		struct ifaddr *ifa;
1100 		int s = pserialize_read_enter();
1101 
1102 		ifa = ifa_ifwithnet(ifpaddr);
1103 		if (ifa != NULL)
1104 			info->rti_ifp = if_get_byindex(ifa->ifa_ifp->if_index,
1105 			    psref);
1106 		pserialize_read_exit(s);
1107 	}
1108 
1109 	return info->rti_ifp;
1110 }
1111 
1112 static struct ifaddr *
1113 rt_getifa(struct rt_addrinfo *info, struct psref *psref)
1114 {
1115 	struct ifaddr *ifa = NULL;
1116 	const struct sockaddr *dst = info->rti_info[RTAX_DST];
1117 	const struct sockaddr *gateway = info->rti_info[RTAX_GATEWAY];
1118 	const struct sockaddr *ifaaddr = info->rti_info[RTAX_IFA];
1119 	int flags = info->rti_flags;
1120 	const struct sockaddr *sa;
1121 
1122 	if (info->rti_ifa == NULL && ifaaddr != NULL) {
1123 		ifa = ifa_ifwithaddr_psref(ifaaddr, psref);
1124 		if (ifa != NULL)
1125 			goto got;
1126 	}
1127 
1128 	sa = ifaaddr != NULL ? ifaaddr :
1129 	    (gateway != NULL ? gateway : dst);
1130 	if (sa != NULL && info->rti_ifp != NULL)
1131 		ifa = ifaof_ifpforaddr_psref(sa, info->rti_ifp, psref);
1132 	else if (dst != NULL && gateway != NULL)
1133 		ifa = ifa_ifwithroute_psref(flags, dst, gateway, psref);
1134 	else if (sa != NULL)
1135 		ifa = ifa_ifwithroute_psref(flags, sa, sa, psref);
1136 	if (ifa == NULL)
1137 		return NULL;
1138 got:
1139 	if (ifa->ifa_getifa != NULL) {
1140 		/* FIXME ifa_getifa is NOMPSAFE */
1141 		ifa = (*ifa->ifa_getifa)(ifa, dst);
1142 		if (ifa == NULL)
1143 			return NULL;
1144 		ifa_acquire(ifa, psref);
1145 	}
1146 	info->rti_ifa = ifa;
1147 	if (info->rti_ifp == NULL)
1148 		info->rti_ifp = ifa->ifa_ifp;
1149 	return ifa;
1150 }
1151 
1152 /*
1153  * If it suceeds and ret_nrt isn't NULL, refcnt of ret_nrt is incremented.
1154  * The caller has to rtfree it by itself.
1155  */
1156 int
1157 rtrequest1(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt)
1158 {
1159 	int s = splsoftnet(), ss;
1160 	int error = 0, rc;
1161 	struct rtentry *rt;
1162 	rtbl_t *rtbl;
1163 	struct ifaddr *ifa = NULL;
1164 	struct sockaddr_storage maskeddst;
1165 	const struct sockaddr *dst = info->rti_info[RTAX_DST];
1166 	const struct sockaddr *gateway = info->rti_info[RTAX_GATEWAY];
1167 	const struct sockaddr *netmask = info->rti_info[RTAX_NETMASK];
1168 	int flags = info->rti_flags;
1169 	struct psref psref_ifp, psref_ifa;
1170 	int bound = 0;
1171 	struct ifnet *ifp = NULL;
1172 	bool need_to_release_ifa = true;
1173 	bool need_unlock = true;
1174 #define senderr(x) { error = x ; goto bad; }
1175 
1176 	RT_WLOCK();
1177 
1178 	bound = curlwp_bind();
1179 	if ((rtbl = rt_gettable(dst->sa_family)) == NULL)
1180 		senderr(ESRCH);
1181 	if (flags & RTF_HOST)
1182 		netmask = NULL;
1183 	switch (req) {
1184 	case RTM_DELETE:
1185 		if (netmask) {
1186 			rt_maskedcopy(dst, (struct sockaddr *)&maskeddst,
1187 			    netmask);
1188 			dst = (struct sockaddr *)&maskeddst;
1189 		}
1190 		if ((rt = rt_lookup(rtbl, dst, netmask)) == NULL)
1191 			senderr(ESRCH);
1192 		if ((rt = rt_deladdr(rtbl, dst, netmask)) == NULL)
1193 			senderr(ESRCH);
1194 		rt->rt_flags &= ~RTF_UP;
1195 		ifa = rt->rt_ifa;
1196 		if (ifa->ifa_flags & IFA_ROUTE &&
1197 		    rt_ifa_connected(rt, ifa)) {
1198 			RT_DPRINTF("rt->_rt_key = %p, ifa = %p, "
1199 			    "deleted IFA_ROUTE\n",
1200 			    (void *)rt->_rt_key, (void *)ifa);
1201 			ifa->ifa_flags &= ~IFA_ROUTE;
1202 		}
1203 		if (ifa->ifa_rtrequest)
1204 			ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1205 		ifa = NULL;
1206 		rttrash++;
1207 		if (ret_nrt) {
1208 			*ret_nrt = rt;
1209 			rt_ref(rt);
1210 			RT_REFCNT_TRACE(rt);
1211 		}
1212 		rtcache_invalidate();
1213 		RT_UNLOCK();
1214 		need_unlock = false;
1215 		rt_timer_remove_all(rt);
1216 #if defined(INET) || defined(INET6)
1217 		if (netmask != NULL)
1218 			lltable_prefix_free(dst->sa_family, dst, netmask, 0);
1219 #endif
1220 		if (ret_nrt == NULL) {
1221 			/* Adjust the refcount */
1222 			rt_ref(rt);
1223 			RT_REFCNT_TRACE(rt);
1224 			rt_free(rt);
1225 		}
1226 		break;
1227 
1228 	case RTM_ADD:
1229 		if (info->rti_ifa == NULL) {
1230 			ifp = rt_getifp(info, &psref_ifp);
1231 			ifa = rt_getifa(info, &psref_ifa);
1232 			if (ifa == NULL)
1233 				senderr(ENETUNREACH);
1234 		} else {
1235 			/* Caller should have a reference of ifa */
1236 			ifa = info->rti_ifa;
1237 			need_to_release_ifa = false;
1238 		}
1239 		rt = pool_get(&rtentry_pool, PR_NOWAIT);
1240 		if (rt == NULL)
1241 			senderr(ENOBUFS);
1242 		memset(rt, 0, sizeof(*rt));
1243 		rt->rt_flags = RTF_UP | (flags & ~RTF_DONTCHANGEIFA);
1244 		LIST_INIT(&rt->rt_timer);
1245 
1246 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1247 		if (netmask) {
1248 			rt_maskedcopy(dst, (struct sockaddr *)&maskeddst,
1249 			    netmask);
1250 			rt_setkey(rt, (struct sockaddr *)&maskeddst, M_NOWAIT);
1251 		} else {
1252 			rt_setkey(rt, dst, M_NOWAIT);
1253 		}
1254 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1255 		if (rt_getkey(rt) == NULL ||
1256 		    rt_setgate(rt, gateway) != 0) {
1257 			pool_put(&rtentry_pool, rt);
1258 			senderr(ENOBUFS);
1259 		}
1260 
1261 		rt_set_ifa(rt, ifa);
1262 		if (info->rti_info[RTAX_TAG] != NULL) {
1263 			const struct sockaddr *tag;
1264 			tag = rt_settag(rt, info->rti_info[RTAX_TAG]);
1265 			if (tag == NULL)
1266 				senderr(ENOBUFS);
1267 		}
1268 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1269 
1270 		ss = pserialize_read_enter();
1271 		if (info->rti_info[RTAX_IFP] != NULL) {
1272 			struct ifaddr *ifa2;
1273 			ifa2 = ifa_ifwithnet(info->rti_info[RTAX_IFP]);
1274 			if (ifa2 != NULL)
1275 				rt->rt_ifp = ifa2->ifa_ifp;
1276 			else
1277 				rt->rt_ifp = ifa->ifa_ifp;
1278 		} else
1279 			rt->rt_ifp = ifa->ifa_ifp;
1280 		pserialize_read_exit(ss);
1281 		cv_init(&rt->rt_cv, "rtentry");
1282 		psref_target_init(&rt->rt_psref, rt_psref_class);
1283 
1284 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1285 		rc = rt_addaddr(rtbl, rt, netmask);
1286 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1287 		if (rc != 0) {
1288 			ifafree(ifa); /* for rt_set_ifa above */
1289 			cv_destroy(&rt->rt_cv);
1290 			rt_destroy(rt);
1291 			pool_put(&rtentry_pool, rt);
1292 			senderr(rc);
1293 		}
1294 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1295 		if (ifa->ifa_rtrequest)
1296 			ifa->ifa_rtrequest(req, rt, info);
1297 		if (need_to_release_ifa)
1298 			ifa_release(ifa, &psref_ifa);
1299 		ifa = NULL;
1300 		if_put(ifp, &psref_ifp);
1301 		ifp = NULL;
1302 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1303 		if (ret_nrt) {
1304 			*ret_nrt = rt;
1305 			rt_ref(rt);
1306 			RT_REFCNT_TRACE(rt);
1307 		}
1308 		rtcache_invalidate();
1309 		RT_UNLOCK();
1310 		need_unlock = false;
1311 		break;
1312 	case RTM_GET:
1313 		if (netmask != NULL) {
1314 			rt_maskedcopy(dst, (struct sockaddr *)&maskeddst,
1315 			    netmask);
1316 			dst = (struct sockaddr *)&maskeddst;
1317 		}
1318 		if ((rt = rt_lookup(rtbl, dst, netmask)) == NULL)
1319 			senderr(ESRCH);
1320 		if (ret_nrt != NULL) {
1321 			*ret_nrt = rt;
1322 			rt_ref(rt);
1323 			RT_REFCNT_TRACE(rt);
1324 		}
1325 		break;
1326 	}
1327 bad:
1328 	if (need_to_release_ifa)
1329 		ifa_release(ifa, &psref_ifa);
1330 	if_put(ifp, &psref_ifp);
1331 	curlwp_bindx(bound);
1332 	if (need_unlock)
1333 		RT_UNLOCK();
1334 	splx(s);
1335 	return error;
1336 }
1337 
1338 int
1339 rt_setgate(struct rtentry *rt, const struct sockaddr *gate)
1340 {
1341 	struct sockaddr *new, *old;
1342 
1343 	KASSERT(RT_WLOCKED());
1344 	KASSERT(rt->_rt_key != NULL);
1345 	RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1346 
1347 	new = sockaddr_dup(gate, M_ZERO | M_NOWAIT);
1348 	if (new == NULL)
1349 		return ENOMEM;
1350 
1351 	old = rt->rt_gateway;
1352 	rt->rt_gateway = new;
1353 	if (old != NULL)
1354 		sockaddr_free(old);
1355 
1356 	KASSERT(rt->_rt_key != NULL);
1357 	RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1358 
1359 	if (rt->rt_flags & RTF_GATEWAY) {
1360 		struct rtentry *gwrt;
1361 
1362 		gwrt = rtalloc1_locked(gate, 1, false, true);
1363 		/*
1364 		 * If we switched gateways, grab the MTU from the new
1365 		 * gateway route if the current MTU, if the current MTU is
1366 		 * greater than the MTU of gateway.
1367 		 * Note that, if the MTU of gateway is 0, we will reset the
1368 		 * MTU of the route to run PMTUD again from scratch. XXX
1369 		 */
1370 		if (gwrt != NULL) {
1371 			KASSERT(gwrt->_rt_key != NULL);
1372 			RT_DPRINTF("gwrt->_rt_key = %p\n", gwrt->_rt_key);
1373 			if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0 &&
1374 			    rt->rt_rmx.rmx_mtu &&
1375 			    rt->rt_rmx.rmx_mtu > gwrt->rt_rmx.rmx_mtu) {
1376 				rt->rt_rmx.rmx_mtu = gwrt->rt_rmx.rmx_mtu;
1377 			}
1378 			rt_unref(gwrt);
1379 		}
1380 	}
1381 	KASSERT(rt->_rt_key != NULL);
1382 	RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1383 	return 0;
1384 }
1385 
1386 static struct ifaddr *
1387 rt_update_get_ifa(const struct rt_addrinfo *info, const struct rtentry *rt,
1388     struct ifnet **ifp, struct psref *psref_ifp, struct psref *psref)
1389 {
1390 	struct ifaddr *ifa = NULL;
1391 
1392 	*ifp = NULL;
1393 	if (info->rti_info[RTAX_IFP] != NULL) {
1394 		ifa = ifa_ifwithnet_psref(info->rti_info[RTAX_IFP], psref);
1395 		if (ifa == NULL)
1396 			goto next;
1397 		*ifp = ifa->ifa_ifp;
1398 		if_acquire(*ifp, psref_ifp);
1399 		if (info->rti_info[RTAX_IFA] == NULL &&
1400 		    info->rti_info[RTAX_GATEWAY] == NULL)
1401 			goto out;
1402 		ifa_release(ifa, psref);
1403 		if (info->rti_info[RTAX_IFA] == NULL) {
1404 			/* route change <dst> <gw> -ifp <if> */
1405 			ifa = ifaof_ifpforaddr_psref(
1406 			    info->rti_info[RTAX_GATEWAY], *ifp, psref);
1407 		} else {
1408 			/* route change <dst> -ifp <if> -ifa <addr> */
1409 			ifa = ifa_ifwithaddr_psref(info->rti_info[RTAX_IFA],
1410 			    psref);
1411 			if (ifa != NULL)
1412 				goto out;
1413 			ifa = ifaof_ifpforaddr_psref(info->rti_info[RTAX_IFA],
1414 			    *ifp, psref);
1415 		}
1416 		goto out;
1417 	}
1418 next:
1419 	if (info->rti_info[RTAX_IFA] != NULL) {
1420 		/* route change <dst> <gw> -ifa <addr> */
1421 		ifa = ifa_ifwithaddr_psref(info->rti_info[RTAX_IFA], psref);
1422 		if (ifa != NULL)
1423 			goto out;
1424 	}
1425 	if (info->rti_info[RTAX_GATEWAY] != NULL) {
1426 		/* route change <dst> <gw> */
1427 		ifa = ifa_ifwithroute_psref(rt->rt_flags, rt_getkey(rt),
1428 		    info->rti_info[RTAX_GATEWAY], psref);
1429 	}
1430 out:
1431 	if (ifa != NULL && *ifp == NULL) {
1432 		*ifp = ifa->ifa_ifp;
1433 		if_acquire(*ifp, psref_ifp);
1434 	}
1435 	if (ifa == NULL && *ifp != NULL) {
1436 		if_put(*ifp, psref_ifp);
1437 		*ifp = NULL;
1438 	}
1439 	return ifa;
1440 }
1441 
1442 int
1443 rt_update(struct rtentry *rt, struct rt_addrinfo *info, void *rtm)
1444 {
1445 	int error = 0;
1446 	struct ifnet *ifp = NULL, *new_ifp = NULL;
1447 	struct ifaddr *ifa = NULL, *new_ifa;
1448 	struct psref psref_ifa, psref_new_ifa, psref_ifp, psref_new_ifp;
1449 	bool newgw, ifp_changed = false;
1450 
1451 	RT_WLOCK();
1452 	/*
1453 	 * New gateway could require new ifaddr, ifp;
1454 	 * flags may also be different; ifp may be specified
1455 	 * by ll sockaddr when protocol address is ambiguous
1456 	 */
1457 	newgw = info->rti_info[RTAX_GATEWAY] != NULL &&
1458 	    sockaddr_cmp(info->rti_info[RTAX_GATEWAY], rt->rt_gateway) != 0;
1459 
1460 	if (newgw || info->rti_info[RTAX_IFP] != NULL ||
1461 	    info->rti_info[RTAX_IFA] != NULL) {
1462 		ifp = rt_getifp(info, &psref_ifp);
1463 		/* info refers ifp so we need to keep a reference */
1464 		ifa = rt_getifa(info, &psref_ifa);
1465 		if (ifa == NULL) {
1466 			error = ENETUNREACH;
1467 			goto out;
1468 		}
1469 	}
1470 	if (newgw) {
1471 		error = rt_setgate(rt, info->rti_info[RTAX_GATEWAY]);
1472 		if (error != 0)
1473 			goto out;
1474 	}
1475 	if (info->rti_info[RTAX_TAG]) {
1476 		const struct sockaddr *tag;
1477 		tag = rt_settag(rt, info->rti_info[RTAX_TAG]);
1478 		if (tag == NULL) {
1479 			error = ENOBUFS;
1480 			goto out;
1481 		}
1482 	}
1483 	/*
1484 	 * New gateway could require new ifaddr, ifp;
1485 	 * flags may also be different; ifp may be specified
1486 	 * by ll sockaddr when protocol address is ambiguous
1487 	 */
1488 	new_ifa = rt_update_get_ifa(info, rt, &new_ifp, &psref_new_ifp,
1489 	    &psref_new_ifa);
1490 	if (new_ifa != NULL) {
1491 		ifa_release(ifa, &psref_ifa);
1492 		ifa = new_ifa;
1493 	}
1494 	if (ifa) {
1495 		struct ifaddr *oifa = rt->rt_ifa;
1496 		if (oifa != ifa && !ifa_is_destroying(ifa) &&
1497 		    new_ifp != NULL && !if_is_deactivated(new_ifp)) {
1498 			if (oifa && oifa->ifa_rtrequest)
1499 				oifa->ifa_rtrequest(RTM_DELETE, rt, info);
1500 			rt_replace_ifa(rt, ifa);
1501 			rt->rt_ifp = new_ifp;
1502 			ifp_changed = true;
1503 		}
1504 		if (new_ifa == NULL)
1505 			ifa_release(ifa, &psref_ifa);
1506 		/* To avoid ifa_release below */
1507 		ifa = NULL;
1508 	}
1509 	ifa_release(new_ifa, &psref_new_ifa);
1510 	if (new_ifp && rt->rt_ifp != new_ifp && !if_is_deactivated(new_ifp)) {
1511 		rt->rt_ifp = new_ifp;
1512 		ifp_changed = true;
1513 	}
1514 	rt_setmetrics(rtm, rt);
1515 	if (rt->rt_flags != info->rti_flags) {
1516 		rt->rt_flags = (info->rti_flags & ~PRESERVED_RTF) |
1517 		    (rt->rt_flags & PRESERVED_RTF);
1518 	}
1519 	if (rt->rt_ifa->ifa_rtrequest)
1520 		rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info);
1521 #if defined(INET) || defined(INET6)
1522 	if (ifp_changed && rt_mask(rt) != NULL)
1523 		lltable_prefix_free(rt_getkey(rt)->sa_family, rt_getkey(rt),
1524 		    rt_mask(rt), 0);
1525 #else
1526 	(void)ifp_changed; /* XXX gcc */
1527 #endif
1528 out:
1529 	ifa_release(ifa, &psref_ifa);
1530 	if_put(new_ifp, &psref_new_ifp);
1531 	if_put(ifp, &psref_ifp);
1532 
1533 	RT_UNLOCK();
1534 
1535 	return error;
1536 }
1537 
1538 static void
1539 rt_maskedcopy(const struct sockaddr *src, struct sockaddr *dst,
1540 	const struct sockaddr *netmask)
1541 {
1542 	const char *netmaskp = &netmask->sa_data[0],
1543 	           *srcp = &src->sa_data[0];
1544 	char *dstp = &dst->sa_data[0];
1545 	const char *maskend = (char *)dst + MIN(netmask->sa_len, src->sa_len);
1546 	const char *srcend = (char *)dst + src->sa_len;
1547 
1548 	dst->sa_len = src->sa_len;
1549 	dst->sa_family = src->sa_family;
1550 
1551 	while (dstp < maskend)
1552 		*dstp++ = *srcp++ & *netmaskp++;
1553 	if (dstp < srcend)
1554 		memset(dstp, 0, (size_t)(srcend - dstp));
1555 }
1556 
1557 /*
1558  * Inform the routing socket of a route change.
1559  */
1560 void
1561 rt_newmsg(const int cmd, const struct rtentry *rt)
1562 {
1563 	struct rt_addrinfo info;
1564 
1565 	memset((void *)&info, 0, sizeof(info));
1566 	info.rti_info[RTAX_DST] = rt_getkey(rt);
1567 	info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1568 	info.rti_info[RTAX_NETMASK] = rt_mask(rt);
1569 	if (rt->rt_ifp) {
1570 		info.rti_info[RTAX_IFP] = rt->rt_ifp->if_dl->ifa_addr;
1571 		info.rti_info[RTAX_IFA] = rt->rt_ifa->ifa_addr;
1572 	}
1573 
1574 	rt_missmsg(cmd, &info, rt->rt_flags, 0);
1575 }
1576 
1577 /*
1578  * Set up or tear down a routing table entry, normally
1579  * for an interface.
1580  */
1581 int
1582 rtinit(struct ifaddr *ifa, int cmd, int flags)
1583 {
1584 	struct rtentry *rt;
1585 	struct sockaddr *dst, *odst;
1586 	struct sockaddr_storage maskeddst;
1587 	struct rtentry *nrt = NULL;
1588 	int error;
1589 	struct rt_addrinfo info;
1590 
1591 	dst = flags & RTF_HOST ? ifa->ifa_dstaddr : ifa->ifa_addr;
1592 	if (cmd == RTM_DELETE) {
1593 		if ((flags & RTF_HOST) == 0 && ifa->ifa_netmask) {
1594 			/* Delete subnet route for this interface */
1595 			odst = dst;
1596 			dst = (struct sockaddr *)&maskeddst;
1597 			rt_maskedcopy(odst, dst, ifa->ifa_netmask);
1598 		}
1599 		if ((rt = rtalloc1(dst, 0)) != NULL) {
1600 			if (rt->rt_ifa != ifa) {
1601 				rt_unref(rt);
1602 				return (flags & RTF_HOST) ? EHOSTUNREACH
1603 							: ENETUNREACH;
1604 			}
1605 			rt_unref(rt);
1606 		}
1607 	}
1608 	memset(&info, 0, sizeof(info));
1609 	info.rti_ifa = ifa;
1610 	info.rti_flags = flags | ifa->ifa_flags | RTF_DONTCHANGEIFA;
1611 	info.rti_info[RTAX_DST] = dst;
1612 	info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1613 
1614 	/*
1615 	 * XXX here, it seems that we are assuming that ifa_netmask is NULL
1616 	 * for RTF_HOST.  bsdi4 passes NULL explicitly (via intermediate
1617 	 * variable) when RTF_HOST is 1.  still not sure if i can safely
1618 	 * change it to meet bsdi4 behavior.
1619 	 */
1620 	if (cmd != RTM_LLINFO_UPD)
1621 		info.rti_info[RTAX_NETMASK] = ifa->ifa_netmask;
1622 	error = rtrequest1((cmd == RTM_LLINFO_UPD) ? RTM_GET : cmd, &info,
1623 	    &nrt);
1624 	if (error != 0)
1625 		return error;
1626 
1627 	rt = nrt;
1628 	RT_REFCNT_TRACE(rt);
1629 	switch (cmd) {
1630 	case RTM_DELETE:
1631 		rt_newmsg(cmd, rt);
1632 		rt_free(rt);
1633 		break;
1634 	case RTM_LLINFO_UPD:
1635 		if (cmd == RTM_LLINFO_UPD && ifa->ifa_rtrequest != NULL)
1636 			ifa->ifa_rtrequest(RTM_LLINFO_UPD, rt, &info);
1637 		rt_newmsg(RTM_CHANGE, rt);
1638 		rt_unref(rt);
1639 		break;
1640 	case RTM_ADD:
1641 		KASSERT(rt->rt_ifa == ifa);
1642 		rt_newmsg(cmd, rt);
1643 		rt_unref(rt);
1644 		RT_REFCNT_TRACE(rt);
1645 		break;
1646 	}
1647 	return error;
1648 }
1649 
1650 /*
1651  * Create a local route entry for the address.
1652  * Announce the addition of the address and the route to the routing socket.
1653  */
1654 int
1655 rt_ifa_addlocal(struct ifaddr *ifa)
1656 {
1657 	struct rtentry *rt;
1658 	int e;
1659 
1660 	/* If there is no loopback entry, allocate one. */
1661 	rt = rtalloc1(ifa->ifa_addr, 0);
1662 #ifdef RT_DEBUG
1663 	if (rt != NULL)
1664 		dump_rt(rt);
1665 #endif
1666 	if (rt == NULL || (rt->rt_flags & RTF_HOST) == 0 ||
1667 	    (rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0)
1668 	{
1669 		struct rt_addrinfo info;
1670 		struct rtentry *nrt;
1671 
1672 		memset(&info, 0, sizeof(info));
1673 		info.rti_flags = RTF_HOST | RTF_LOCAL | RTF_DONTCHANGEIFA;
1674 		info.rti_info[RTAX_DST] = ifa->ifa_addr;
1675 		info.rti_info[RTAX_GATEWAY] =
1676 		    (const struct sockaddr *)ifa->ifa_ifp->if_sadl;
1677 		info.rti_ifa = ifa;
1678 		nrt = NULL;
1679 		e = rtrequest1(RTM_ADD, &info, &nrt);
1680 		rt_addrmsg_rt(RTM_ADD, ifa, e, nrt);
1681 		if (nrt != NULL) {
1682 			KASSERT(nrt->rt_ifa == ifa);
1683 #ifdef RT_DEBUG
1684 			dump_rt(nrt);
1685 #endif
1686 			rt_unref(nrt);
1687 			RT_REFCNT_TRACE(nrt);
1688 		}
1689 	} else {
1690 		e = 0;
1691 		rt_addrmsg(RTM_NEWADDR, ifa);
1692 	}
1693 	if (rt != NULL)
1694 		rt_unref(rt);
1695 	return e;
1696 }
1697 
1698 /*
1699  * Remove the local route entry for the address.
1700  * Announce the removal of the address and the route to the routing socket.
1701  */
1702 int
1703 rt_ifa_remlocal(struct ifaddr *ifa, struct ifaddr *alt_ifa)
1704 {
1705 	struct rtentry *rt;
1706 	int e = 0;
1707 
1708 	rt = rtalloc1(ifa->ifa_addr, 0);
1709 
1710 	/*
1711 	 * Before deleting, check if a corresponding loopbacked
1712 	 * host route surely exists.  With this check, we can avoid
1713 	 * deleting an interface direct route whose destination is
1714 	 * the same as the address being removed.  This can happen
1715 	 * when removing a subnet-router anycast address on an
1716 	 * interface attached to a shared medium.
1717 	 */
1718 	if (rt != NULL &&
1719 	    (rt->rt_flags & RTF_HOST) &&
1720 	    (rt->rt_ifp->if_flags & IFF_LOOPBACK))
1721 	{
1722 		/* If we cannot replace the route's ifaddr with the equivalent
1723 		 * ifaddr of another interface, I believe it is safest to
1724 		 * delete the route.
1725 		 */
1726 		if (alt_ifa == NULL) {
1727 			e = rtdeletemsg(rt);
1728 			if (e == 0) {
1729 				rt_unref(rt);
1730 				rt_free(rt);
1731 				rt = NULL;
1732 			}
1733 			rt_addrmsg(RTM_DELADDR, ifa);
1734 		} else {
1735 #ifdef NET_MPSAFE
1736 			int error = rt_update_prepare(rt);
1737 			if (error == 0) {
1738 				rt_replace_ifa(rt, alt_ifa);
1739 				rt_update_finish(rt);
1740 			} else {
1741 				/*
1742 				 * If error != 0, the rtentry is being
1743 				 * destroyed, so doing nothing doesn't
1744 				 * matter.
1745 				 */
1746 			}
1747 #else
1748 			rt_replace_ifa(rt, alt_ifa);
1749 #endif
1750 			rt_newmsg(RTM_CHANGE, rt);
1751 		}
1752 	} else
1753 		rt_addrmsg(RTM_DELADDR, ifa);
1754 	if (rt != NULL)
1755 		rt_unref(rt);
1756 	return e;
1757 }
1758 
1759 /*
1760  * Route timer routines.  These routes allow functions to be called
1761  * for various routes at any time.  This is useful in supporting
1762  * path MTU discovery and redirect route deletion.
1763  *
1764  * This is similar to some BSDI internal functions, but it provides
1765  * for multiple queues for efficiency's sake...
1766  */
1767 
1768 LIST_HEAD(, rttimer_queue) rttimer_queue_head;
1769 static int rt_init_done = 0;
1770 
1771 /*
1772  * Some subtle order problems with domain initialization mean that
1773  * we cannot count on this being run from rt_init before various
1774  * protocol initializations are done.  Therefore, we make sure
1775  * that this is run when the first queue is added...
1776  */
1777 
1778 static void rt_timer_work(struct work *, void *);
1779 
1780 static void
1781 rt_timer_init(void)
1782 {
1783 	int error;
1784 
1785 	assert(rt_init_done == 0);
1786 
1787 	/* XXX should be in rt_init */
1788 	rw_init(&rt_lock);
1789 
1790 	LIST_INIT(&rttimer_queue_head);
1791 	callout_init(&rt_timer_ch, CALLOUT_MPSAFE);
1792 	error = workqueue_create(&rt_timer_wq, "rt_timer",
1793 	    rt_timer_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE);
1794 	if (error)
1795 		panic("%s: workqueue_create failed (%d)\n", __func__, error);
1796 	callout_reset(&rt_timer_ch, hz, rt_timer_timer, NULL);
1797 	rt_init_done = 1;
1798 }
1799 
1800 struct rttimer_queue *
1801 rt_timer_queue_create(u_int timeout)
1802 {
1803 	struct rttimer_queue *rtq;
1804 
1805 	if (rt_init_done == 0)
1806 		rt_timer_init();
1807 
1808 	R_Malloc(rtq, struct rttimer_queue *, sizeof *rtq);
1809 	if (rtq == NULL)
1810 		return NULL;
1811 	memset(rtq, 0, sizeof(*rtq));
1812 
1813 	rtq->rtq_timeout = timeout;
1814 	TAILQ_INIT(&rtq->rtq_head);
1815 	RT_WLOCK();
1816 	LIST_INSERT_HEAD(&rttimer_queue_head, rtq, rtq_link);
1817 	RT_UNLOCK();
1818 
1819 	return rtq;
1820 }
1821 
1822 void
1823 rt_timer_queue_change(struct rttimer_queue *rtq, long timeout)
1824 {
1825 
1826 	rtq->rtq_timeout = timeout;
1827 }
1828 
1829 static void
1830 rt_timer_queue_remove_all(struct rttimer_queue *rtq)
1831 {
1832 	struct rttimer *r;
1833 
1834 	RT_ASSERT_WLOCK();
1835 
1836 	while ((r = TAILQ_FIRST(&rtq->rtq_head)) != NULL) {
1837 		LIST_REMOVE(r, rtt_link);
1838 		TAILQ_REMOVE(&rtq->rtq_head, r, rtt_next);
1839 		rt_ref(r->rtt_rt); /* XXX */
1840 		RT_REFCNT_TRACE(r->rtt_rt);
1841 		RT_UNLOCK();
1842 		(*r->rtt_func)(r->rtt_rt, r);
1843 		pool_put(&rttimer_pool, r);
1844 		RT_WLOCK();
1845 		if (rtq->rtq_count > 0)
1846 			rtq->rtq_count--;
1847 		else
1848 			printf("rt_timer_queue_remove_all: "
1849 			    "rtq_count reached 0\n");
1850 	}
1851 }
1852 
1853 void
1854 rt_timer_queue_destroy(struct rttimer_queue *rtq)
1855 {
1856 
1857 	RT_WLOCK();
1858 	rt_timer_queue_remove_all(rtq);
1859 	LIST_REMOVE(rtq, rtq_link);
1860 	RT_UNLOCK();
1861 
1862 	/*
1863 	 * Caller is responsible for freeing the rttimer_queue structure.
1864 	 */
1865 }
1866 
1867 unsigned long
1868 rt_timer_count(struct rttimer_queue *rtq)
1869 {
1870 	return rtq->rtq_count;
1871 }
1872 
1873 static void
1874 rt_timer_remove_all(struct rtentry *rt)
1875 {
1876 	struct rttimer *r;
1877 
1878 	RT_WLOCK();
1879 	while ((r = LIST_FIRST(&rt->rt_timer)) != NULL) {
1880 		LIST_REMOVE(r, rtt_link);
1881 		TAILQ_REMOVE(&r->rtt_queue->rtq_head, r, rtt_next);
1882 		if (r->rtt_queue->rtq_count > 0)
1883 			r->rtt_queue->rtq_count--;
1884 		else
1885 			printf("rt_timer_remove_all: rtq_count reached 0\n");
1886 		pool_put(&rttimer_pool, r);
1887 	}
1888 	RT_UNLOCK();
1889 }
1890 
1891 int
1892 rt_timer_add(struct rtentry *rt,
1893 	void (*func)(struct rtentry *, struct rttimer *),
1894 	struct rttimer_queue *queue)
1895 {
1896 	struct rttimer *r;
1897 
1898 	KASSERT(func != NULL);
1899 	RT_WLOCK();
1900 	/*
1901 	 * If there's already a timer with this action, destroy it before
1902 	 * we add a new one.
1903 	 */
1904 	LIST_FOREACH(r, &rt->rt_timer, rtt_link) {
1905 		if (r->rtt_func == func)
1906 			break;
1907 	}
1908 	if (r != NULL) {
1909 		LIST_REMOVE(r, rtt_link);
1910 		TAILQ_REMOVE(&r->rtt_queue->rtq_head, r, rtt_next);
1911 		if (r->rtt_queue->rtq_count > 0)
1912 			r->rtt_queue->rtq_count--;
1913 		else
1914 			printf("rt_timer_add: rtq_count reached 0\n");
1915 	} else {
1916 		r = pool_get(&rttimer_pool, PR_NOWAIT);
1917 		if (r == NULL) {
1918 			RT_UNLOCK();
1919 			return ENOBUFS;
1920 		}
1921 	}
1922 
1923 	memset(r, 0, sizeof(*r));
1924 
1925 	r->rtt_rt = rt;
1926 	r->rtt_time = time_uptime;
1927 	r->rtt_func = func;
1928 	r->rtt_queue = queue;
1929 	LIST_INSERT_HEAD(&rt->rt_timer, r, rtt_link);
1930 	TAILQ_INSERT_TAIL(&queue->rtq_head, r, rtt_next);
1931 	r->rtt_queue->rtq_count++;
1932 
1933 	RT_UNLOCK();
1934 
1935 	return 0;
1936 }
1937 
1938 static void
1939 rt_timer_work(struct work *wk, void *arg)
1940 {
1941 	struct rttimer_queue *rtq;
1942 	struct rttimer *r;
1943 
1944 	RT_WLOCK();
1945 	LIST_FOREACH(rtq, &rttimer_queue_head, rtq_link) {
1946 		while ((r = TAILQ_FIRST(&rtq->rtq_head)) != NULL &&
1947 		    (r->rtt_time + rtq->rtq_timeout) < time_uptime) {
1948 			LIST_REMOVE(r, rtt_link);
1949 			TAILQ_REMOVE(&rtq->rtq_head, r, rtt_next);
1950 			/*
1951 			 * Take a reference to avoid the rtentry is freed
1952 			 * accidentally after RT_UNLOCK.  The callback
1953 			 * (rtt_func) must rt_unref it by itself.
1954 			 */
1955 			rt_ref(r->rtt_rt);
1956 			RT_REFCNT_TRACE(r->rtt_rt);
1957 			RT_UNLOCK();
1958 			(*r->rtt_func)(r->rtt_rt, r);
1959 			pool_put(&rttimer_pool, r);
1960 			RT_WLOCK();
1961 			if (rtq->rtq_count > 0)
1962 				rtq->rtq_count--;
1963 			else
1964 				printf("rt_timer_timer: rtq_count reached 0\n");
1965 		}
1966 	}
1967 	RT_UNLOCK();
1968 
1969 	callout_reset(&rt_timer_ch, hz, rt_timer_timer, NULL);
1970 }
1971 
1972 static void
1973 rt_timer_timer(void *arg)
1974 {
1975 
1976 	workqueue_enqueue(rt_timer_wq, &rt_timer_wk, NULL);
1977 }
1978 
1979 static struct rtentry *
1980 _rtcache_init(struct route *ro, int flag)
1981 {
1982 	struct rtentry *rt;
1983 
1984 	rtcache_invariants(ro);
1985 	KASSERT(ro->_ro_rt == NULL);
1986 
1987 	if (rtcache_getdst(ro) == NULL)
1988 		return NULL;
1989 	rt = rtalloc1(rtcache_getdst(ro), flag);
1990 	if (rt != NULL) {
1991 		RT_RLOCK();
1992 		if (ISSET(rt->rt_flags, RTF_UP)) {
1993 			ro->_ro_rt = rt;
1994 			ro->ro_rtcache_generation = rtcache_generation;
1995 			rtcache_ref(rt, ro);
1996 		}
1997 		RT_UNLOCK();
1998 		rt_unref(rt);
1999 	}
2000 
2001 	rtcache_invariants(ro);
2002 	return ro->_ro_rt;
2003 }
2004 
2005 struct rtentry *
2006 rtcache_init(struct route *ro)
2007 {
2008 
2009 	return _rtcache_init(ro, 1);
2010 }
2011 
2012 struct rtentry *
2013 rtcache_init_noclone(struct route *ro)
2014 {
2015 
2016 	return _rtcache_init(ro, 0);
2017 }
2018 
2019 struct rtentry *
2020 rtcache_update(struct route *ro, int clone)
2021 {
2022 
2023 	ro->_ro_rt = NULL;
2024 	return _rtcache_init(ro, clone);
2025 }
2026 
2027 void
2028 rtcache_copy(struct route *new_ro, struct route *old_ro)
2029 {
2030 	struct rtentry *rt;
2031 	int ret;
2032 
2033 	KASSERT(new_ro != old_ro);
2034 	rtcache_invariants(new_ro);
2035 	rtcache_invariants(old_ro);
2036 
2037 	rt = rtcache_validate(old_ro);
2038 
2039 	if (rtcache_getdst(old_ro) == NULL)
2040 		goto out;
2041 	ret = rtcache_setdst(new_ro, rtcache_getdst(old_ro));
2042 	if (ret != 0)
2043 		goto out;
2044 
2045 	RT_RLOCK();
2046 	new_ro->_ro_rt = rt;
2047 	new_ro->ro_rtcache_generation = rtcache_generation;
2048 	RT_UNLOCK();
2049 	rtcache_invariants(new_ro);
2050 out:
2051 	rtcache_unref(rt, old_ro);
2052 	return;
2053 }
2054 
2055 #if defined(RT_DEBUG) && defined(NET_MPSAFE)
2056 static void
2057 rtcache_trace(const char *func, struct rtentry *rt, struct route *ro)
2058 {
2059 	char dst[64];
2060 
2061 	sockaddr_format(ro->ro_sa, dst, 64);
2062 	printf("trace: %s:\tdst=%s cpu=%d lwp=%p psref=%p target=%p\n", func, dst,
2063 	    cpu_index(curcpu()), curlwp, &ro->ro_psref, &rt->rt_psref);
2064 }
2065 #define RTCACHE_PSREF_TRACE(rt, ro)	rtcache_trace(__func__, (rt), (ro))
2066 #else
2067 #define RTCACHE_PSREF_TRACE(rt, ro)	do {} while (0)
2068 #endif
2069 
2070 static void
2071 rtcache_ref(struct rtentry *rt, struct route *ro)
2072 {
2073 
2074 	KASSERT(rt != NULL);
2075 
2076 #ifdef NET_MPSAFE
2077 	RTCACHE_PSREF_TRACE(rt, ro);
2078 	ro->ro_bound = curlwp_bind();
2079 	/* XXX Use a real caller's address */
2080 	PSREF_DEBUG_FILL_RETURN_ADDRESS(&ro->ro_psref);
2081 	psref_acquire(&ro->ro_psref, &rt->rt_psref, rt_psref_class);
2082 #endif
2083 }
2084 
2085 void
2086 rtcache_unref(struct rtentry *rt, struct route *ro)
2087 {
2088 
2089 	if (rt == NULL)
2090 		return;
2091 
2092 #ifdef NET_MPSAFE
2093 	psref_release(&ro->ro_psref, &rt->rt_psref, rt_psref_class);
2094 	curlwp_bindx(ro->ro_bound);
2095 	RTCACHE_PSREF_TRACE(rt, ro);
2096 #endif
2097 }
2098 
2099 struct rtentry *
2100 rtcache_validate(struct route *ro)
2101 {
2102 	struct rtentry *rt = NULL;
2103 
2104 #ifdef NET_MPSAFE
2105 retry:
2106 #endif
2107 	rtcache_invariants(ro);
2108 	RT_RLOCK();
2109 	if (ro->ro_rtcache_generation != rtcache_generation) {
2110 		/* The cache is invalidated */
2111 		rt = NULL;
2112 		goto out;
2113 	}
2114 
2115 	rt = ro->_ro_rt;
2116 	if (rt == NULL)
2117 		goto out;
2118 
2119 	if ((rt->rt_flags & RTF_UP) == 0) {
2120 		rt = NULL;
2121 		goto out;
2122 	}
2123 #ifdef NET_MPSAFE
2124 	if (ISSET(rt->rt_flags, RTF_UPDATING)) {
2125 		if (rt_wait_ok()) {
2126 			RT_UNLOCK();
2127 
2128 			/* We can wait until the update is complete */
2129 			rt_update_wait();
2130 			goto retry;
2131 		} else {
2132 			rt = NULL;
2133 		}
2134 	} else
2135 #endif
2136 		rtcache_ref(rt, ro);
2137 out:
2138 	RT_UNLOCK();
2139 	return rt;
2140 }
2141 
2142 struct rtentry *
2143 rtcache_lookup2(struct route *ro, const struct sockaddr *dst,
2144     int clone, int *hitp)
2145 {
2146 	const struct sockaddr *odst;
2147 	struct rtentry *rt = NULL;
2148 
2149 	odst = rtcache_getdst(ro);
2150 	if (odst == NULL)
2151 		goto miss;
2152 
2153 	if (sockaddr_cmp(odst, dst) != 0) {
2154 		rtcache_free(ro);
2155 		goto miss;
2156 	}
2157 
2158 	rt = rtcache_validate(ro);
2159 	if (rt == NULL) {
2160 		ro->_ro_rt = NULL;
2161 		goto miss;
2162 	}
2163 
2164 	rtcache_invariants(ro);
2165 
2166 	if (hitp != NULL)
2167 		*hitp = 1;
2168 	return rt;
2169 miss:
2170 	if (hitp != NULL)
2171 		*hitp = 0;
2172 	if (rtcache_setdst(ro, dst) == 0)
2173 		rt = _rtcache_init(ro, clone);
2174 
2175 	rtcache_invariants(ro);
2176 
2177 	return rt;
2178 }
2179 
2180 void
2181 rtcache_free(struct route *ro)
2182 {
2183 
2184 	ro->_ro_rt = NULL;
2185 	if (ro->ro_sa != NULL) {
2186 		sockaddr_free(ro->ro_sa);
2187 		ro->ro_sa = NULL;
2188 	}
2189 	rtcache_invariants(ro);
2190 }
2191 
2192 int
2193 rtcache_setdst(struct route *ro, const struct sockaddr *sa)
2194 {
2195 	KASSERT(sa != NULL);
2196 
2197 	rtcache_invariants(ro);
2198 	if (ro->ro_sa != NULL) {
2199 		if (ro->ro_sa->sa_family == sa->sa_family) {
2200 			ro->_ro_rt = NULL;
2201 			sockaddr_copy(ro->ro_sa, ro->ro_sa->sa_len, sa);
2202 			rtcache_invariants(ro);
2203 			return 0;
2204 		}
2205 		/* free ro_sa, wrong family */
2206 		rtcache_free(ro);
2207 	}
2208 
2209 	KASSERT(ro->_ro_rt == NULL);
2210 
2211 	if ((ro->ro_sa = sockaddr_dup(sa, M_ZERO | M_NOWAIT)) == NULL) {
2212 		rtcache_invariants(ro);
2213 		return ENOMEM;
2214 	}
2215 	rtcache_invariants(ro);
2216 	return 0;
2217 }
2218 
2219 static void
2220 rtcache_percpu_init_cpu(void *p, void *arg __unused, struct cpu_info *ci __unused)
2221 {
2222 	struct route **rop = p;
2223 
2224 	/*
2225 	 * We can't have struct route as percpu data because it can be destroyed
2226 	 * over a memory enlargement processing of percpu.
2227 	 */
2228 	*rop = kmem_zalloc(sizeof(**rop), KM_SLEEP);
2229 }
2230 
2231 percpu_t *
2232 rtcache_percpu_alloc(void)
2233 {
2234 
2235 	return percpu_create(sizeof(struct route *),
2236 	    rtcache_percpu_init_cpu, NULL, NULL);
2237 }
2238 
2239 const struct sockaddr *
2240 rt_settag(struct rtentry *rt, const struct sockaddr *tag)
2241 {
2242 	if (rt->rt_tag != tag) {
2243 		if (rt->rt_tag != NULL)
2244 			sockaddr_free(rt->rt_tag);
2245 		rt->rt_tag = sockaddr_dup(tag, M_ZERO | M_NOWAIT);
2246 	}
2247 	return rt->rt_tag;
2248 }
2249 
2250 struct sockaddr *
2251 rt_gettag(const struct rtentry *rt)
2252 {
2253 	return rt->rt_tag;
2254 }
2255 
2256 int
2257 rt_check_reject_route(const struct rtentry *rt, const struct ifnet *ifp)
2258 {
2259 
2260 	if ((rt->rt_flags & RTF_REJECT) != 0) {
2261 		/* Mimic looutput */
2262 		if (ifp->if_flags & IFF_LOOPBACK)
2263 			return (rt->rt_flags & RTF_HOST) ?
2264 			    EHOSTUNREACH : ENETUNREACH;
2265 		else if (rt->rt_rmx.rmx_expire == 0 ||
2266 		    time_uptime < rt->rt_rmx.rmx_expire)
2267 			return (rt->rt_flags & RTF_GATEWAY) ?
2268 			    EHOSTUNREACH : EHOSTDOWN;
2269 	}
2270 
2271 	return 0;
2272 }
2273 
2274 void
2275 rt_delete_matched_entries(sa_family_t family, int (*f)(struct rtentry *, void *),
2276     void *v)
2277 {
2278 
2279 	for (;;) {
2280 		int s;
2281 		int error;
2282 		struct rtentry *rt, *retrt = NULL;
2283 
2284 		RT_RLOCK();
2285 		s = splsoftnet();
2286 		rt = rtbl_search_matched_entry(family, f, v);
2287 		if (rt == NULL) {
2288 			splx(s);
2289 			RT_UNLOCK();
2290 			return;
2291 		}
2292 		rt_ref(rt);
2293 		splx(s);
2294 		RT_UNLOCK();
2295 
2296 		error = rtrequest(RTM_DELETE, rt_getkey(rt), rt->rt_gateway,
2297 		    rt_mask(rt), rt->rt_flags, &retrt);
2298 		if (error == 0) {
2299 			KASSERT(retrt == rt);
2300 			KASSERT((retrt->rt_flags & RTF_UP) == 0);
2301 			retrt->rt_ifp = NULL;
2302 			rt_unref(rt);
2303 			rt_free(retrt);
2304 		} else if (error == ESRCH) {
2305 			/* Someone deleted the entry already. */
2306 			rt_unref(rt);
2307 		} else {
2308 			log(LOG_ERR, "%s: unable to delete rtentry @ %p, "
2309 			    "error = %d\n", rt->rt_ifp->if_xname, rt, error);
2310 			/* XXX how to treat this case? */
2311 		}
2312 	}
2313 }
2314 
2315 static int
2316 rt_walktree_locked(sa_family_t family, int (*f)(struct rtentry *, void *),
2317     void *v)
2318 {
2319 
2320 	return rtbl_walktree(family, f, v);
2321 }
2322 
2323 int
2324 rt_walktree(sa_family_t family, int (*f)(struct rtentry *, void *), void *v)
2325 {
2326 	int error;
2327 
2328 	RT_RLOCK();
2329 	error = rt_walktree_locked(family, f, v);
2330 	RT_UNLOCK();
2331 
2332 	return error;
2333 }
2334 
2335 #ifdef DDB
2336 
2337 #include <machine/db_machdep.h>
2338 #include <ddb/db_interface.h>
2339 #include <ddb/db_output.h>
2340 
2341 #define	rt_expire rt_rmx.rmx_expire
2342 
2343 static void
2344 db_print_sa(const struct sockaddr *sa)
2345 {
2346 	int len;
2347 	const u_char *p;
2348 
2349 	if (sa == NULL) {
2350 		db_printf("[NULL]");
2351 		return;
2352 	}
2353 
2354 	p = (const u_char *)sa;
2355 	len = sa->sa_len;
2356 	db_printf("[");
2357 	while (len > 0) {
2358 		db_printf("%d", *p);
2359 		p++; len--;
2360 		if (len) db_printf(",");
2361 	}
2362 	db_printf("]\n");
2363 }
2364 
2365 static void
2366 db_print_ifa(struct ifaddr *ifa)
2367 {
2368 	if (ifa == NULL)
2369 		return;
2370 	db_printf("  ifa_addr=");
2371 	db_print_sa(ifa->ifa_addr);
2372 	db_printf("  ifa_dsta=");
2373 	db_print_sa(ifa->ifa_dstaddr);
2374 	db_printf("  ifa_mask=");
2375 	db_print_sa(ifa->ifa_netmask);
2376 	db_printf("  flags=0x%x,refcnt=%d,metric=%d\n",
2377 			  ifa->ifa_flags,
2378 			  ifa->ifa_refcnt,
2379 			  ifa->ifa_metric);
2380 }
2381 
2382 /*
2383  * Function to pass to rt_walktree().
2384  * Return non-zero error to abort walk.
2385  */
2386 static int
2387 db_show_rtentry(struct rtentry *rt, void *w)
2388 {
2389 	db_printf("rtentry=%p", rt);
2390 
2391 	db_printf(" flags=0x%x refcnt=%d use=%"PRId64" expire=%"PRId64"\n",
2392 			  rt->rt_flags, rt->rt_refcnt,
2393 			  rt->rt_use, (uint64_t)rt->rt_expire);
2394 
2395 	db_printf(" key="); db_print_sa(rt_getkey(rt));
2396 	db_printf(" mask="); db_print_sa(rt_mask(rt));
2397 	db_printf(" gw="); db_print_sa(rt->rt_gateway);
2398 
2399 	db_printf(" ifp=%p ", rt->rt_ifp);
2400 	if (rt->rt_ifp)
2401 		db_printf("(%s)", rt->rt_ifp->if_xname);
2402 	else
2403 		db_printf("(NULL)");
2404 
2405 	db_printf(" ifa=%p\n", rt->rt_ifa);
2406 	db_print_ifa(rt->rt_ifa);
2407 
2408 	db_printf(" gwroute=%p llinfo=%p\n",
2409 			  rt->rt_gwroute, rt->rt_llinfo);
2410 
2411 	return 0;
2412 }
2413 
2414 /*
2415  * Function to print all the route trees.
2416  * Use this from ddb:  "show routes"
2417  */
2418 void
2419 db_show_routes(db_expr_t addr, bool have_addr,
2420     db_expr_t count, const char *modif)
2421 {
2422 
2423 	/* Taking RT_LOCK will fail if LOCKDEBUG is enabled. */
2424 	rt_walktree_locked(AF_INET, db_show_rtentry, NULL);
2425 }
2426 #endif
2427