xref: /dflybsd-src/sys/net/if.c (revision 1682d0d9b38b9dfa89cf2360c842c2ebf748e352)
1 /*
2  * Copyright (c) 1980, 1986, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)if.c	8.3 (Berkeley) 1/4/94
34  * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
35  * $DragonFly: src/sys/net/if.c,v 1.84 2008/11/15 11:58:16 sephe Exp $
36  */
37 
38 #include "opt_compat.h"
39 #include "opt_inet6.h"
40 #include "opt_inet.h"
41 #include "opt_polling.h"
42 
43 #include <sys/param.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/systm.h>
47 #include <sys/proc.h>
48 #include <sys/priv.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/socketops.h>
53 #include <sys/protosw.h>
54 #include <sys/kernel.h>
55 #include <sys/ktr.h>
56 #include <sys/sockio.h>
57 #include <sys/syslog.h>
58 #include <sys/sysctl.h>
59 #include <sys/domain.h>
60 #include <sys/thread.h>
61 #include <sys/thread2.h>
62 #include <sys/serialize.h>
63 #include <sys/msgport2.h>
64 #include <sys/bus.h>
65 
66 #include <net/if.h>
67 #include <net/if_arp.h>
68 #include <net/if_dl.h>
69 #include <net/if_types.h>
70 #include <net/if_var.h>
71 #include <net/ifq_var.h>
72 #include <net/radix.h>
73 #include <net/route.h>
74 #include <net/if_clone.h>
75 #include <net/netisr.h>
76 #include <net/netmsg2.h>
77 
78 #include <machine/atomic.h>
79 #include <machine/stdarg.h>
80 #include <machine/smp.h>
81 
82 #if defined(INET) || defined(INET6)
83 /*XXX*/
84 #include <netinet/in.h>
85 #include <netinet/in_var.h>
86 #include <netinet/if_ether.h>
87 #ifdef INET6
88 #include <netinet6/in6_var.h>
89 #include <netinet6/in6_ifattach.h>
90 #endif
91 #endif
92 
93 #if defined(COMPAT_43)
94 #include <emulation/43bsd/43bsd_socket.h>
95 #endif /* COMPAT_43 */
96 
97 struct netmsg_ifaddr {
98 	struct netmsg	netmsg;
99 	struct ifaddr	*ifa;
100 	struct ifnet	*ifp;
101 	int		tail;
102 };
103 
104 /*
105  * System initialization
106  */
107 static void	if_attachdomain(void *);
108 static void	if_attachdomain1(struct ifnet *);
109 static int	ifconf(u_long, caddr_t, struct ucred *);
110 static void	ifinit(void *);
111 static void	ifnetinit(void *);
112 static void	if_slowtimo(void *);
113 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
114 static int	if_rtdel(struct radix_node *, void *);
115 
116 #ifdef INET6
117 /*
118  * XXX: declare here to avoid to include many inet6 related files..
119  * should be more generalized?
120  */
121 extern void	nd6_setmtu(struct ifnet *);
122 #endif
123 
124 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
125 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
126 
127 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
128 /* Must be after netisr_init */
129 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
130 
131 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
132 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
133 
134 int			ifqmaxlen = IFQ_MAXLEN;
135 struct ifnethead	ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
136 
137 /* In ifq_dispatch(), try to do direct ifnet.if_start first */
138 static int		ifq_dispatch_schedonly = 0;
139 SYSCTL_INT(_net_link_generic, OID_AUTO, ifq_dispatch_schedonly, CTLFLAG_RW,
140            &ifq_dispatch_schedonly, 0, "");
141 
142 /* In ifq_dispatch(), schedule ifnet.if_start without checking ifnet.if_snd */
143 static int		ifq_dispatch_schednochk = 0;
144 SYSCTL_INT(_net_link_generic, OID_AUTO, ifq_dispatch_schednochk, CTLFLAG_RW,
145            &ifq_dispatch_schednochk, 0, "");
146 
147 /* In if_devstart(), try to do direct ifnet.if_start first */
148 static int		if_devstart_schedonly = 0;
149 SYSCTL_INT(_net_link_generic, OID_AUTO, if_devstart_schedonly, CTLFLAG_RW,
150            &if_devstart_schedonly, 0, "");
151 
152 /* In if_devstart(), schedule ifnet.if_start without checking ifnet.if_snd */
153 static int		if_devstart_schednochk = 0;
154 SYSCTL_INT(_net_link_generic, OID_AUTO, if_devstart_schednochk, CTLFLAG_RW,
155            &if_devstart_schednochk, 0, "");
156 
157 #ifdef SMP
158 /* Schedule ifnet.if_start on the current CPU */
159 static int		if_start_oncpu_sched = 0;
160 SYSCTL_INT(_net_link_generic, OID_AUTO, if_start_oncpu_sched, CTLFLAG_RW,
161            &if_start_oncpu_sched, 0, "");
162 #endif
163 
164 struct callout		if_slowtimo_timer;
165 
166 int			if_index = 0;
167 struct ifnet		**ifindex2ifnet = NULL;
168 static struct thread	ifnet_threads[MAXCPU];
169 static int		ifnet_mpsafe_thread = NETMSG_SERVICE_MPSAFE;
170 
171 #define IFQ_KTR_STRING		"ifq=%p"
172 #define IFQ_KTR_ARG_SIZE	(sizeof(void *))
173 #ifndef KTR_IFQ
174 #define KTR_IFQ			KTR_ALL
175 #endif
176 KTR_INFO_MASTER(ifq);
177 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARG_SIZE);
178 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARG_SIZE);
179 #define logifq(name, arg)	KTR_LOG(ifq_ ## name, arg)
180 
181 #define IF_START_KTR_STRING	"ifp=%p"
182 #define IF_START_KTR_ARG_SIZE	(sizeof(void *))
183 #ifndef KTR_IF_START
184 #define KTR_IF_START		KTR_ALL
185 #endif
186 KTR_INFO_MASTER(if_start);
187 KTR_INFO(KTR_IF_START, if_start, run, 0,
188 	 IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
189 KTR_INFO(KTR_IF_START, if_start, sched, 1,
190 	 IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
191 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
192 	 IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
193 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
194 	 IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
195 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
196 	 IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
197 #define logifstart(name, arg)	KTR_LOG(if_start_ ## name, arg)
198 
199 /*
200  * Network interface utility routines.
201  *
202  * Routines with ifa_ifwith* names take sockaddr *'s as
203  * parameters.
204  */
205 /* ARGSUSED*/
206 void
207 ifinit(void *dummy)
208 {
209 	struct ifnet *ifp;
210 
211 	callout_init(&if_slowtimo_timer);
212 
213 	crit_enter();
214 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
215 		if (ifp->if_snd.ifq_maxlen == 0) {
216 			if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
217 			ifp->if_snd.ifq_maxlen = ifqmaxlen;
218 		}
219 	}
220 	crit_exit();
221 
222 	if_slowtimo(0);
223 }
224 
225 static int
226 if_start_cpuid(struct ifnet *ifp)
227 {
228 	return ifp->if_cpuid;
229 }
230 
231 #ifdef DEVICE_POLLING
232 static int
233 if_start_cpuid_poll(struct ifnet *ifp)
234 {
235 	int poll_cpuid = ifp->if_poll_cpuid;
236 
237 	if (poll_cpuid >= 0)
238 		return poll_cpuid;
239 	else
240 		return ifp->if_cpuid;
241 }
242 #endif
243 
244 static void
245 if_start_ipifunc(void *arg)
246 {
247 	struct ifnet *ifp = arg;
248 	struct lwkt_msg *lmsg = &ifp->if_start_nmsg[mycpuid].nm_lmsg;
249 
250 	crit_enter();
251 	if (lmsg->ms_flags & MSGF_DONE)
252 		lwkt_sendmsg(ifnet_portfn(mycpuid), lmsg);
253 	crit_exit();
254 }
255 
256 /*
257  * Schedule ifnet.if_start on ifnet's CPU
258  */
259 static void
260 if_start_schedule(struct ifnet *ifp)
261 {
262 #ifdef SMP
263 	int cpu;
264 
265 	if (if_start_oncpu_sched)
266 		cpu = mycpuid;
267 	else
268 		cpu = ifp->if_start_cpuid(ifp);
269 
270 	if (cpu != mycpuid)
271 		lwkt_send_ipiq(globaldata_find(cpu), if_start_ipifunc, ifp);
272 	else
273 #endif
274 	if_start_ipifunc(ifp);
275 }
276 
277 /*
278  * NOTE:
279  * This function will release ifnet.if_start interlock,
280  * if ifnet.if_start does not need to be scheduled
281  */
282 static __inline int
283 if_start_need_schedule(struct ifaltq *ifq, int running)
284 {
285 	if (!running || ifq_is_empty(ifq)
286 #ifdef ALTQ
287 	    || ifq->altq_tbr != NULL
288 #endif
289 	) {
290 		ALTQ_LOCK(ifq);
291 		/*
292 		 * ifnet.if_start interlock is released, if:
293 		 * 1) Hardware can not take any packets, due to
294 		 *    o  interface is marked down
295 		 *    o  hardware queue is full (IFF_OACTIVE)
296 		 *    Under the second situation, hardware interrupt
297 		 *    or polling(4) will call/schedule ifnet.if_start
298 		 *    when hardware queue is ready
299 		 * 2) There is not packet in the ifnet.if_snd.
300 		 *    Further ifq_dispatch or ifq_handoff will call/
301 		 *    schedule ifnet.if_start
302 		 * 3) TBR is used and it does not allow further
303 		 *    dequeueing.
304 		 *    TBR callout will call ifnet.if_start
305 		 */
306 		if (!running || !ifq_data_ready(ifq)) {
307 			ifq->altq_started = 0;
308 			ALTQ_UNLOCK(ifq);
309 			return 0;
310 		}
311 		ALTQ_UNLOCK(ifq);
312 	}
313 	return 1;
314 }
315 
316 static void
317 if_start_dispatch(struct netmsg *nmsg)
318 {
319 	struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
320 	struct ifnet *ifp = lmsg->u.ms_resultp;
321 	struct ifaltq *ifq = &ifp->if_snd;
322 	int running = 0;
323 
324 	crit_enter();
325 	lwkt_replymsg(lmsg, 0);	/* reply ASAP */
326 	crit_exit();
327 
328 #ifdef SMP
329 	if (!if_start_oncpu_sched && mycpuid != ifp->if_start_cpuid(ifp)) {
330 		/*
331 		 * If the ifnet is still up, we need to
332 		 * chase its CPU change.
333 		 */
334 		if (ifp->if_flags & IFF_UP) {
335 			logifstart(chase_sched, ifp);
336 			if_start_schedule(ifp);
337 			return;
338 		} else {
339 			goto check;
340 		}
341 	}
342 #endif
343 
344 	if (ifp->if_flags & IFF_UP) {
345 		lwkt_serialize_enter(ifp->if_serializer); /* XXX try? */
346 		if ((ifp->if_flags & IFF_OACTIVE) == 0) {
347 			logifstart(run, ifp);
348 			ifp->if_start(ifp);
349 			if ((ifp->if_flags &
350 			(IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
351 				running = 1;
352 		}
353 		lwkt_serialize_exit(ifp->if_serializer);
354 	}
355 #ifdef SMP
356 check:
357 #endif
358 	if (if_start_need_schedule(ifq, running)) {
359 		crit_enter();
360 		if (lmsg->ms_flags & MSGF_DONE)	{ /* XXX necessary? */
361 			logifstart(sched, ifp);
362 			lwkt_sendmsg(ifnet_portfn(mycpuid), lmsg);
363 		}
364 		crit_exit();
365 	}
366 }
367 
368 /* Device driver ifnet.if_start helper function */
369 void
370 if_devstart(struct ifnet *ifp)
371 {
372 	struct ifaltq *ifq = &ifp->if_snd;
373 	int running = 0;
374 
375 	ASSERT_SERIALIZED(ifp->if_serializer);
376 
377 	ALTQ_LOCK(ifq);
378 	if (ifq->altq_started || !ifq_data_ready(ifq)) {
379 		logifstart(avoid, ifp);
380 		ALTQ_UNLOCK(ifq);
381 		return;
382 	}
383 	ifq->altq_started = 1;
384 	ALTQ_UNLOCK(ifq);
385 
386 	if (if_devstart_schedonly) {
387 		/*
388 		 * Always schedule ifnet.if_start on ifnet's CPU,
389 		 * short circuit the rest of this function.
390 		 */
391 		logifstart(sched, ifp);
392 		if_start_schedule(ifp);
393 		return;
394 	}
395 
396 	logifstart(run, ifp);
397 	ifp->if_start(ifp);
398 
399 	if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
400 		running = 1;
401 
402 	if (if_devstart_schednochk || if_start_need_schedule(ifq, running)) {
403 		/*
404 		 * More data need to be transmitted, ifnet.if_start is
405 		 * scheduled on ifnet's CPU, and we keep going.
406 		 * NOTE: ifnet.if_start interlock is not released.
407 		 */
408 		logifstart(sched, ifp);
409 		if_start_schedule(ifp);
410 	}
411 }
412 
413 /*
414  * Attach an interface to the list of "active" interfaces.
415  *
416  * The serializer is optional.  If non-NULL access to the interface
417  * may be MPSAFE.
418  */
419 void
420 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
421 {
422 	unsigned socksize, ifasize;
423 	int namelen, masklen;
424 	struct sockaddr_dl *sdl;
425 	struct ifaddr *ifa;
426 	struct ifaltq *ifq;
427 	int i;
428 
429 	static int if_indexlim = 8;
430 
431 	/*
432 	 * The serializer can be passed in from the device, allowing the
433 	 * same serializer to be used for both the interrupt interlock and
434 	 * the device queue.  If not specified, the netif structure will
435 	 * use an embedded serializer.
436 	 */
437 	if (serializer == NULL) {
438 		serializer = &ifp->if_default_serializer;
439 		lwkt_serialize_init(serializer);
440 	}
441 	ifp->if_serializer = serializer;
442 
443 	ifp->if_start_cpuid = if_start_cpuid;
444 	ifp->if_cpuid = 0;
445 
446 #ifdef DEVICE_POLLING
447 	/* Device is not in polling mode by default */
448 	ifp->if_poll_cpuid = -1;
449 	if (ifp->if_poll != NULL)
450 		ifp->if_start_cpuid = if_start_cpuid_poll;
451 #endif
452 
453 	ifp->if_start_nmsg = kmalloc(ncpus * sizeof(struct netmsg),
454 				     M_LWKTMSG, M_WAITOK);
455 	for (i = 0; i < ncpus; ++i) {
456 		netmsg_init(&ifp->if_start_nmsg[i], &netisr_adone_rport, 0,
457 			    if_start_dispatch);
458 		ifp->if_start_nmsg[i].nm_lmsg.u.ms_resultp = ifp;
459 	}
460 
461 	TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
462 	ifp->if_index = ++if_index;
463 
464 	/*
465 	 * XXX -
466 	 * The old code would work if the interface passed a pre-existing
467 	 * chain of ifaddrs to this code.  We don't trust our callers to
468 	 * properly initialize the tailq, however, so we no longer allow
469 	 * this unlikely case.
470 	 */
471 	ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
472 				    M_IFADDR, M_WAITOK | M_ZERO);
473 	for (i = 0; i < ncpus; ++i)
474 		TAILQ_INIT(&ifp->if_addrheads[i]);
475 
476 	TAILQ_INIT(&ifp->if_prefixhead);
477 	LIST_INIT(&ifp->if_multiaddrs);
478 	getmicrotime(&ifp->if_lastchange);
479 	if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
480 		unsigned int n;
481 		struct ifnet **q;
482 
483 		if_indexlim <<= 1;
484 
485 		/* grow ifindex2ifnet */
486 		n = if_indexlim * sizeof(*q);
487 		q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
488 		if (ifindex2ifnet) {
489 			bcopy(ifindex2ifnet, q, n/2);
490 			kfree(ifindex2ifnet, M_IFADDR);
491 		}
492 		ifindex2ifnet = q;
493 	}
494 
495 	ifindex2ifnet[if_index] = ifp;
496 
497 	/*
498 	 * create a Link Level name for this device
499 	 */
500 	namelen = strlen(ifp->if_xname);
501 #define _offsetof(t, m) ((int)((caddr_t)&((t *)0)->m))
502 	masklen = _offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
503 	socksize = masklen + ifp->if_addrlen;
504 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
505 	if (socksize < sizeof(*sdl))
506 		socksize = sizeof(*sdl);
507 	socksize = ROUNDUP(socksize);
508 	ifasize = sizeof(struct ifaddr) + 2 * socksize;
509 	ifa = ifa_create(ifasize, M_WAITOK);
510 	sdl = (struct sockaddr_dl *)(ifa + 1);
511 	sdl->sdl_len = socksize;
512 	sdl->sdl_family = AF_LINK;
513 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
514 	sdl->sdl_nlen = namelen;
515 	sdl->sdl_index = ifp->if_index;
516 	sdl->sdl_type = ifp->if_type;
517 	ifp->if_lladdr = ifa;
518 	ifa->ifa_ifp = ifp;
519 	ifa->ifa_rtrequest = link_rtrequest;
520 	ifa->ifa_addr = (struct sockaddr *)sdl;
521 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
522 	ifa->ifa_netmask = (struct sockaddr *)sdl;
523 	sdl->sdl_len = masklen;
524 	while (namelen != 0)
525 		sdl->sdl_data[--namelen] = 0xff;
526 	ifa_iflink(ifa, ifp, 0 /* Insert head */);
527 
528 	EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
529 	devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
530 
531 	ifq = &ifp->if_snd;
532 	ifq->altq_type = 0;
533 	ifq->altq_disc = NULL;
534 	ifq->altq_flags &= ALTQF_CANTCHANGE;
535 	ifq->altq_tbr = NULL;
536 	ifq->altq_ifp = ifp;
537 	ifq->altq_started = 0;
538 	ifq->altq_prepended = NULL;
539 	ALTQ_LOCK_INIT(ifq);
540 	ifq_set_classic(ifq);
541 
542 	if (!SLIST_EMPTY(&domains))
543 		if_attachdomain1(ifp);
544 
545 	/* Announce the interface. */
546 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
547 }
548 
549 static void
550 if_attachdomain(void *dummy)
551 {
552 	struct ifnet *ifp;
553 
554 	crit_enter();
555 	TAILQ_FOREACH(ifp, &ifnet, if_list)
556 		if_attachdomain1(ifp);
557 	crit_exit();
558 }
559 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
560 	if_attachdomain, NULL);
561 
562 static void
563 if_attachdomain1(struct ifnet *ifp)
564 {
565 	struct domain *dp;
566 
567 	crit_enter();
568 
569 	/* address family dependent data region */
570 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
571 	SLIST_FOREACH(dp, &domains, dom_next)
572 		if (dp->dom_ifattach)
573 			ifp->if_afdata[dp->dom_family] =
574 				(*dp->dom_ifattach)(ifp);
575 	crit_exit();
576 }
577 
578 /*
579  * Purge all addresses whose type is _not_ AF_LINK
580  */
581 void
582 if_purgeaddrs_nolink(struct ifnet *ifp)
583 {
584 	struct ifaddr_container *ifac, *next;
585 
586 	TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
587 			      ifa_link, next) {
588 		struct ifaddr *ifa = ifac->ifa;
589 
590 		/* Leave link ifaddr as it is */
591 		if (ifa->ifa_addr->sa_family == AF_LINK)
592 			continue;
593 #ifdef INET
594 		/* XXX: Ugly!! ad hoc just for INET */
595 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
596 			struct ifaliasreq ifr;
597 #ifdef IFADDR_DEBUG_VERBOSE
598 			int i;
599 
600 			kprintf("purge in4 addr %p: ", ifa);
601 			for (i = 0; i < ncpus; ++i)
602 				kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
603 			kprintf("\n");
604 #endif
605 
606 			bzero(&ifr, sizeof ifr);
607 			ifr.ifra_addr = *ifa->ifa_addr;
608 			if (ifa->ifa_dstaddr)
609 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
610 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
611 				       NULL) == 0)
612 				continue;
613 		}
614 #endif /* INET */
615 #ifdef INET6
616 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
617 #ifdef IFADDR_DEBUG_VERBOSE
618 			int i;
619 
620 			kprintf("purge in6 addr %p: ", ifa);
621 			for (i = 0; i < ncpus; ++i)
622 				kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
623 			kprintf("\n");
624 #endif
625 
626 			in6_purgeaddr(ifa);
627 			/* ifp_addrhead is already updated */
628 			continue;
629 		}
630 #endif /* INET6 */
631 		ifa_ifunlink(ifa, ifp);
632 		ifa_destroy(ifa);
633 	}
634 }
635 
636 /*
637  * Detach an interface, removing it from the
638  * list of "active" interfaces.
639  */
640 void
641 if_detach(struct ifnet *ifp)
642 {
643 	struct radix_node_head	*rnh;
644 	int i;
645 	int cpu, origcpu;
646 	struct domain *dp;
647 
648 	EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
649 
650 	/*
651 	 * Remove routes and flush queues.
652 	 */
653 	crit_enter();
654 #ifdef DEVICE_POLLING
655 	if (ifp->if_flags & IFF_POLLING)
656 		ether_poll_deregister(ifp);
657 #endif
658 	if_down(ifp);
659 
660 	if (ifq_is_enabled(&ifp->if_snd))
661 		altq_disable(&ifp->if_snd);
662 	if (ifq_is_attached(&ifp->if_snd))
663 		altq_detach(&ifp->if_snd);
664 
665 	/*
666 	 * Clean up all addresses.
667 	 */
668 	ifp->if_lladdr = NULL;
669 
670 	if_purgeaddrs_nolink(ifp);
671 	if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
672 		struct ifaddr *ifa;
673 
674 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
675 		KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
676 			("non-link ifaddr is left on if_addrheads"));
677 
678 		ifa_ifunlink(ifa, ifp);
679 		ifa_destroy(ifa);
680 		KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
681 			("there are still ifaddrs left on if_addrheads"));
682 	}
683 
684 #ifdef INET
685 	/*
686 	 * Remove all IPv4 kernel structures related to ifp.
687 	 */
688 	in_ifdetach(ifp);
689 #endif
690 
691 #ifdef INET6
692 	/*
693 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
694 	 * before removing routing entries below, since IPv6 interface direct
695 	 * routes are expected to be removed by the IPv6-specific kernel API.
696 	 * Otherwise, the kernel will detect some inconsistency and bark it.
697 	 */
698 	in6_ifdetach(ifp);
699 #endif
700 
701 	/*
702 	 * Delete all remaining routes using this interface
703 	 * Unfortuneatly the only way to do this is to slog through
704 	 * the entire routing table looking for routes which point
705 	 * to this interface...oh well...
706 	 */
707 	origcpu = mycpuid;
708 	for (cpu = 0; cpu < ncpus2; cpu++) {
709 		lwkt_migratecpu(cpu);
710 		for (i = 1; i <= AF_MAX; i++) {
711 			if ((rnh = rt_tables[cpu][i]) == NULL)
712 				continue;
713 			rnh->rnh_walktree(rnh, if_rtdel, ifp);
714 		}
715 	}
716 	lwkt_migratecpu(origcpu);
717 
718 	/* Announce that the interface is gone. */
719 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
720 	devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
721 
722 	SLIST_FOREACH(dp, &domains, dom_next)
723 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
724 			(*dp->dom_ifdetach)(ifp,
725 				ifp->if_afdata[dp->dom_family]);
726 
727 	/*
728 	 * Remove interface from ifindex2ifp[] and maybe decrement if_index.
729 	 */
730 	ifindex2ifnet[ifp->if_index] = NULL;
731 	while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
732 		if_index--;
733 
734 	TAILQ_REMOVE(&ifnet, ifp, if_link);
735 	kfree(ifp->if_addrheads, M_IFADDR);
736 	kfree(ifp->if_start_nmsg, M_LWKTMSG);
737 	crit_exit();
738 }
739 
740 /*
741  * Delete Routes for a Network Interface
742  *
743  * Called for each routing entry via the rnh->rnh_walktree() call above
744  * to delete all route entries referencing a detaching network interface.
745  *
746  * Arguments:
747  *	rn	pointer to node in the routing table
748  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
749  *
750  * Returns:
751  *	0	successful
752  *	errno	failed - reason indicated
753  *
754  */
755 static int
756 if_rtdel(struct radix_node *rn, void *arg)
757 {
758 	struct rtentry	*rt = (struct rtentry *)rn;
759 	struct ifnet	*ifp = arg;
760 	int		err;
761 
762 	if (rt->rt_ifp == ifp) {
763 
764 		/*
765 		 * Protect (sorta) against walktree recursion problems
766 		 * with cloned routes
767 		 */
768 		if (!(rt->rt_flags & RTF_UP))
769 			return (0);
770 
771 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
772 				rt_mask(rt), rt->rt_flags,
773 				(struct rtentry **) NULL);
774 		if (err) {
775 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
776 		}
777 	}
778 
779 	return (0);
780 }
781 
782 /*
783  * Locate an interface based on a complete address.
784  */
785 struct ifaddr *
786 ifa_ifwithaddr(struct sockaddr *addr)
787 {
788 	struct ifnet *ifp;
789 
790 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
791 		struct ifaddr_container *ifac;
792 
793 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
794 			struct ifaddr *ifa = ifac->ifa;
795 
796 			if (ifa->ifa_addr->sa_family != addr->sa_family)
797 				continue;
798 			if (sa_equal(addr, ifa->ifa_addr))
799 				return (ifa);
800 			if ((ifp->if_flags & IFF_BROADCAST) &&
801 			    ifa->ifa_broadaddr &&
802 			    /* IPv6 doesn't have broadcast */
803 			    ifa->ifa_broadaddr->sa_len != 0 &&
804 			    sa_equal(ifa->ifa_broadaddr, addr))
805 				return (ifa);
806 		}
807 	}
808 	return (NULL);
809 }
810 /*
811  * Locate the point to point interface with a given destination address.
812  */
813 struct ifaddr *
814 ifa_ifwithdstaddr(struct sockaddr *addr)
815 {
816 	struct ifnet *ifp;
817 
818 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
819 		struct ifaddr_container *ifac;
820 
821 		if (!(ifp->if_flags & IFF_POINTOPOINT))
822 			continue;
823 
824 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
825 			struct ifaddr *ifa = ifac->ifa;
826 
827 			if (ifa->ifa_addr->sa_family != addr->sa_family)
828 				continue;
829 			if (ifa->ifa_dstaddr &&
830 			    sa_equal(addr, ifa->ifa_dstaddr))
831 				return (ifa);
832 		}
833 	}
834 	return (NULL);
835 }
836 
837 /*
838  * Find an interface on a specific network.  If many, choice
839  * is most specific found.
840  */
841 struct ifaddr *
842 ifa_ifwithnet(struct sockaddr *addr)
843 {
844 	struct ifnet *ifp;
845 	struct ifaddr *ifa_maybe = NULL;
846 	u_int af = addr->sa_family;
847 	char *addr_data = addr->sa_data, *cplim;
848 
849 	/*
850 	 * AF_LINK addresses can be looked up directly by their index number,
851 	 * so do that if we can.
852 	 */
853 	if (af == AF_LINK) {
854 		struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
855 
856 		if (sdl->sdl_index && sdl->sdl_index <= if_index)
857 			return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
858 	}
859 
860 	/*
861 	 * Scan though each interface, looking for ones that have
862 	 * addresses in this address family.
863 	 */
864 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
865 		struct ifaddr_container *ifac;
866 
867 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
868 			struct ifaddr *ifa = ifac->ifa;
869 			char *cp, *cp2, *cp3;
870 
871 			if (ifa->ifa_addr->sa_family != af)
872 next:				continue;
873 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
874 				/*
875 				 * This is a bit broken as it doesn't
876 				 * take into account that the remote end may
877 				 * be a single node in the network we are
878 				 * looking for.
879 				 * The trouble is that we don't know the
880 				 * netmask for the remote end.
881 				 */
882 				if (ifa->ifa_dstaddr != NULL &&
883 				    sa_equal(addr, ifa->ifa_dstaddr))
884 					return (ifa);
885 			} else {
886 				/*
887 				 * if we have a special address handler,
888 				 * then use it instead of the generic one.
889 				 */
890 				if (ifa->ifa_claim_addr) {
891 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
892 						return (ifa);
893 					} else {
894 						continue;
895 					}
896 				}
897 
898 				/*
899 				 * Scan all the bits in the ifa's address.
900 				 * If a bit dissagrees with what we are
901 				 * looking for, mask it with the netmask
902 				 * to see if it really matters.
903 				 * (A byte at a time)
904 				 */
905 				if (ifa->ifa_netmask == 0)
906 					continue;
907 				cp = addr_data;
908 				cp2 = ifa->ifa_addr->sa_data;
909 				cp3 = ifa->ifa_netmask->sa_data;
910 				cplim = ifa->ifa_netmask->sa_len +
911 					(char *)ifa->ifa_netmask;
912 				while (cp3 < cplim)
913 					if ((*cp++ ^ *cp2++) & *cp3++)
914 						goto next; /* next address! */
915 				/*
916 				 * If the netmask of what we just found
917 				 * is more specific than what we had before
918 				 * (if we had one) then remember the new one
919 				 * before continuing to search
920 				 * for an even better one.
921 				 */
922 				if (ifa_maybe == 0 ||
923 				    rn_refines((char *)ifa->ifa_netmask,
924 					       (char *)ifa_maybe->ifa_netmask))
925 					ifa_maybe = ifa;
926 			}
927 		}
928 	}
929 	return (ifa_maybe);
930 }
931 
932 /*
933  * Find an interface address specific to an interface best matching
934  * a given address.
935  */
936 struct ifaddr *
937 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
938 {
939 	struct ifaddr_container *ifac;
940 	char *cp, *cp2, *cp3;
941 	char *cplim;
942 	struct ifaddr *ifa_maybe = 0;
943 	u_int af = addr->sa_family;
944 
945 	if (af >= AF_MAX)
946 		return (0);
947 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
948 		struct ifaddr *ifa = ifac->ifa;
949 
950 		if (ifa->ifa_addr->sa_family != af)
951 			continue;
952 		if (ifa_maybe == 0)
953 			ifa_maybe = ifa;
954 		if (ifa->ifa_netmask == NULL) {
955 			if (sa_equal(addr, ifa->ifa_addr) ||
956 			    (ifa->ifa_dstaddr != NULL &&
957 			     sa_equal(addr, ifa->ifa_dstaddr)))
958 				return (ifa);
959 			continue;
960 		}
961 		if (ifp->if_flags & IFF_POINTOPOINT) {
962 			if (sa_equal(addr, ifa->ifa_dstaddr))
963 				return (ifa);
964 		} else {
965 			cp = addr->sa_data;
966 			cp2 = ifa->ifa_addr->sa_data;
967 			cp3 = ifa->ifa_netmask->sa_data;
968 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
969 			for (; cp3 < cplim; cp3++)
970 				if ((*cp++ ^ *cp2++) & *cp3)
971 					break;
972 			if (cp3 == cplim)
973 				return (ifa);
974 		}
975 	}
976 	return (ifa_maybe);
977 }
978 
979 /*
980  * Default action when installing a route with a Link Level gateway.
981  * Lookup an appropriate real ifa to point to.
982  * This should be moved to /sys/net/link.c eventually.
983  */
984 static void
985 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
986 {
987 	struct ifaddr *ifa;
988 	struct sockaddr *dst;
989 	struct ifnet *ifp;
990 
991 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
992 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
993 		return;
994 	ifa = ifaof_ifpforaddr(dst, ifp);
995 	if (ifa != NULL) {
996 		IFAFREE(rt->rt_ifa);
997 		IFAREF(ifa);
998 		rt->rt_ifa = ifa;
999 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1000 			ifa->ifa_rtrequest(cmd, rt, info);
1001 	}
1002 }
1003 
1004 /*
1005  * Mark an interface down and notify protocols of
1006  * the transition.
1007  * NOTE: must be called at splnet or eqivalent.
1008  */
1009 void
1010 if_unroute(struct ifnet *ifp, int flag, int fam)
1011 {
1012 	struct ifaddr_container *ifac;
1013 
1014 	ifp->if_flags &= ~flag;
1015 	getmicrotime(&ifp->if_lastchange);
1016 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1017 		struct ifaddr *ifa = ifac->ifa;
1018 
1019 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1020 			kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1021 	}
1022 	ifq_purge(&ifp->if_snd);
1023 	rt_ifmsg(ifp);
1024 }
1025 
1026 /*
1027  * Mark an interface up and notify protocols of
1028  * the transition.
1029  * NOTE: must be called at splnet or eqivalent.
1030  */
1031 void
1032 if_route(struct ifnet *ifp, int flag, int fam)
1033 {
1034 	struct ifaddr_container *ifac;
1035 
1036 	ifq_purge(&ifp->if_snd);
1037 	ifp->if_flags |= flag;
1038 	getmicrotime(&ifp->if_lastchange);
1039 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1040 		struct ifaddr *ifa = ifac->ifa;
1041 
1042 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1043 			kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1044 	}
1045 	rt_ifmsg(ifp);
1046 #ifdef INET6
1047 	in6_if_up(ifp);
1048 #endif
1049 }
1050 
1051 /*
1052  * Mark an interface down and notify protocols of the transition.  An
1053  * interface going down is also considered to be a synchronizing event.
1054  * We must ensure that all packet processing related to the interface
1055  * has completed before we return so e.g. the caller can free the ifnet
1056  * structure that the mbufs may be referencing.
1057  *
1058  * NOTE: must be called at splnet or eqivalent.
1059  */
1060 void
1061 if_down(struct ifnet *ifp)
1062 {
1063 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1064 	netmsg_service_sync();
1065 }
1066 
1067 /*
1068  * Mark an interface up and notify protocols of
1069  * the transition.
1070  * NOTE: must be called at splnet or eqivalent.
1071  */
1072 void
1073 if_up(struct ifnet *ifp)
1074 {
1075 	if_route(ifp, IFF_UP, AF_UNSPEC);
1076 }
1077 
1078 /*
1079  * Process a link state change.
1080  * NOTE: must be called at splsoftnet or equivalent.
1081  */
1082 void
1083 if_link_state_change(struct ifnet *ifp)
1084 {
1085 	int link_state = ifp->if_link_state;
1086 
1087 	rt_ifmsg(ifp);
1088 	devctl_notify("IFNET", ifp->if_xname,
1089 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1090 }
1091 
1092 /*
1093  * Handle interface watchdog timer routines.  Called
1094  * from softclock, we decrement timers (if set) and
1095  * call the appropriate interface routine on expiration.
1096  */
1097 static void
1098 if_slowtimo(void *arg)
1099 {
1100 	struct ifnet *ifp;
1101 
1102 	crit_enter();
1103 
1104 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1105 		if (ifp->if_timer == 0 || --ifp->if_timer)
1106 			continue;
1107 		if (ifp->if_watchdog) {
1108 			if (lwkt_serialize_try(ifp->if_serializer)) {
1109 				(*ifp->if_watchdog)(ifp);
1110 				lwkt_serialize_exit(ifp->if_serializer);
1111 			} else {
1112 				/* try again next timeout */
1113 				++ifp->if_timer;
1114 			}
1115 		}
1116 	}
1117 
1118 	crit_exit();
1119 
1120 	callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1121 }
1122 
1123 /*
1124  * Map interface name to
1125  * interface structure pointer.
1126  */
1127 struct ifnet *
1128 ifunit(const char *name)
1129 {
1130 	struct ifnet *ifp;
1131 
1132 	/*
1133 	 * Search all the interfaces for this name/number
1134 	 */
1135 
1136 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1137 		if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1138 			break;
1139 	}
1140 	return (ifp);
1141 }
1142 
1143 
1144 /*
1145  * Map interface name in a sockaddr_dl to
1146  * interface structure pointer.
1147  */
1148 struct ifnet *
1149 if_withname(struct sockaddr *sa)
1150 {
1151 	char ifname[IFNAMSIZ+1];
1152 	struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1153 
1154 	if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1155 	     (sdl->sdl_nlen > IFNAMSIZ) )
1156 		return NULL;
1157 
1158 	/*
1159 	 * ifunit wants a null-terminated name.  It may not be null-terminated
1160 	 * in the sockaddr.  We don't want to change the caller's sockaddr,
1161 	 * and there might not be room to put the trailing null anyway, so we
1162 	 * make a local copy that we know we can null terminate safely.
1163 	 */
1164 
1165 	bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1166 	ifname[sdl->sdl_nlen] = '\0';
1167 	return ifunit(ifname);
1168 }
1169 
1170 
1171 /*
1172  * Interface ioctls.
1173  */
1174 int
1175 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1176 {
1177 	struct ifnet *ifp;
1178 	struct ifreq *ifr;
1179 	struct ifstat *ifs;
1180 	int error;
1181 	short oif_flags;
1182 	int new_flags;
1183 	size_t namelen, onamelen;
1184 	char new_name[IFNAMSIZ];
1185 	struct ifaddr *ifa;
1186 	struct sockaddr_dl *sdl;
1187 
1188 	switch (cmd) {
1189 
1190 	case SIOCGIFCONF:
1191 	case OSIOCGIFCONF:
1192 		return (ifconf(cmd, data, cred));
1193 	}
1194 	ifr = (struct ifreq *)data;
1195 
1196 	switch (cmd) {
1197 	case SIOCIFCREATE:
1198 	case SIOCIFDESTROY:
1199 		if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1200 			return (error);
1201 		return ((cmd == SIOCIFCREATE) ?
1202 			if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name)) :
1203 			if_clone_destroy(ifr->ifr_name));
1204 
1205 	case SIOCIFGCLONERS:
1206 		return (if_clone_list((struct if_clonereq *)data));
1207 	}
1208 
1209 	ifp = ifunit(ifr->ifr_name);
1210 	if (ifp == 0)
1211 		return (ENXIO);
1212 	switch (cmd) {
1213 
1214 	case SIOCGIFFLAGS:
1215 		ifr->ifr_flags = ifp->if_flags;
1216 		ifr->ifr_flagshigh = ifp->if_flags >> 16;
1217 		break;
1218 
1219 	case SIOCGIFCAP:
1220 		ifr->ifr_reqcap = ifp->if_capabilities;
1221 		ifr->ifr_curcap = ifp->if_capenable;
1222 		break;
1223 
1224 	case SIOCGIFMETRIC:
1225 		ifr->ifr_metric = ifp->if_metric;
1226 		break;
1227 
1228 	case SIOCGIFMTU:
1229 		ifr->ifr_mtu = ifp->if_mtu;
1230 		break;
1231 
1232 	case SIOCGIFPHYS:
1233 		ifr->ifr_phys = ifp->if_physical;
1234 		break;
1235 
1236 	case SIOCGIFPOLLCPU:
1237 #ifdef DEVICE_POLLING
1238 		ifr->ifr_pollcpu = ifp->if_poll_cpuid;
1239 #else
1240 		ifr->ifr_pollcpu = -1;
1241 #endif
1242 		break;
1243 
1244 	case SIOCSIFPOLLCPU:
1245 #ifdef DEVICE_POLLING
1246 		if ((ifp->if_flags & IFF_POLLING) == 0)
1247 			ether_pollcpu_register(ifp, ifr->ifr_pollcpu);
1248 #endif
1249 		break;
1250 
1251 	case SIOCSIFFLAGS:
1252 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1253 		if (error)
1254 			return (error);
1255 		new_flags = (ifr->ifr_flags & 0xffff) |
1256 		    (ifr->ifr_flagshigh << 16);
1257 		if (ifp->if_flags & IFF_SMART) {
1258 			/* Smart drivers twiddle their own routes */
1259 		} else if (ifp->if_flags & IFF_UP &&
1260 		    (new_flags & IFF_UP) == 0) {
1261 			crit_enter();
1262 			if_down(ifp);
1263 			crit_exit();
1264 		} else if (new_flags & IFF_UP &&
1265 		    (ifp->if_flags & IFF_UP) == 0) {
1266 			crit_enter();
1267 			if_up(ifp);
1268 			crit_exit();
1269 		}
1270 
1271 #ifdef DEVICE_POLLING
1272 		if ((new_flags ^ ifp->if_flags) & IFF_POLLING) {
1273 			if (new_flags & IFF_POLLING) {
1274 				ether_poll_register(ifp);
1275 			} else {
1276 				ether_poll_deregister(ifp);
1277 			}
1278 		}
1279 #endif
1280 
1281 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1282 			(new_flags &~ IFF_CANTCHANGE);
1283 		if (new_flags & IFF_PPROMISC) {
1284 			/* Permanently promiscuous mode requested */
1285 			ifp->if_flags |= IFF_PROMISC;
1286 		} else if (ifp->if_pcount == 0) {
1287 			ifp->if_flags &= ~IFF_PROMISC;
1288 		}
1289 		if (ifp->if_ioctl) {
1290 			lwkt_serialize_enter(ifp->if_serializer);
1291 			ifp->if_ioctl(ifp, cmd, data, cred);
1292 			lwkt_serialize_exit(ifp->if_serializer);
1293 		}
1294 		getmicrotime(&ifp->if_lastchange);
1295 		break;
1296 
1297 	case SIOCSIFCAP:
1298 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1299 		if (error)
1300 			return (error);
1301 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1302 			return (EINVAL);
1303 		lwkt_serialize_enter(ifp->if_serializer);
1304 		ifp->if_ioctl(ifp, cmd, data, cred);
1305 		lwkt_serialize_exit(ifp->if_serializer);
1306 		break;
1307 
1308 	case SIOCSIFNAME:
1309 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1310 		if (error != 0)
1311 			return (error);
1312 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1313 		if (error != 0)
1314 			return (error);
1315 		if (new_name[0] == '\0')
1316 			return (EINVAL);
1317 		if (ifunit(new_name) != NULL)
1318 			return (EEXIST);
1319 
1320 		EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1321 
1322 		/* Announce the departure of the interface. */
1323 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1324 
1325 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1326 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1327 		/* XXX IFA_LOCK(ifa); */
1328 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1329 		namelen = strlen(new_name);
1330 		onamelen = sdl->sdl_nlen;
1331 		/*
1332 		 * Move the address if needed.  This is safe because we
1333 		 * allocate space for a name of length IFNAMSIZ when we
1334 		 * create this in if_attach().
1335 		 */
1336 		if (namelen != onamelen) {
1337 			bcopy(sdl->sdl_data + onamelen,
1338 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1339 		}
1340 		bcopy(new_name, sdl->sdl_data, namelen);
1341 		sdl->sdl_nlen = namelen;
1342 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1343 		bzero(sdl->sdl_data, onamelen);
1344 		while (namelen != 0)
1345 			sdl->sdl_data[--namelen] = 0xff;
1346 		/* XXX IFA_UNLOCK(ifa) */
1347 
1348 		EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1349 
1350 		/* Announce the return of the interface. */
1351 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1352 		break;
1353 
1354 	case SIOCSIFMETRIC:
1355 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1356 		if (error)
1357 			return (error);
1358 		ifp->if_metric = ifr->ifr_metric;
1359 		getmicrotime(&ifp->if_lastchange);
1360 		break;
1361 
1362 	case SIOCSIFPHYS:
1363 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1364 		if (error)
1365 			return error;
1366 		if (!ifp->if_ioctl)
1367 		        return EOPNOTSUPP;
1368 		lwkt_serialize_enter(ifp->if_serializer);
1369 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1370 		lwkt_serialize_exit(ifp->if_serializer);
1371 		if (error == 0)
1372 			getmicrotime(&ifp->if_lastchange);
1373 		return (error);
1374 
1375 	case SIOCSIFMTU:
1376 	{
1377 		u_long oldmtu = ifp->if_mtu;
1378 
1379 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1380 		if (error)
1381 			return (error);
1382 		if (ifp->if_ioctl == NULL)
1383 			return (EOPNOTSUPP);
1384 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1385 			return (EINVAL);
1386 		lwkt_serialize_enter(ifp->if_serializer);
1387 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1388 		lwkt_serialize_exit(ifp->if_serializer);
1389 		if (error == 0) {
1390 			getmicrotime(&ifp->if_lastchange);
1391 			rt_ifmsg(ifp);
1392 		}
1393 		/*
1394 		 * If the link MTU changed, do network layer specific procedure.
1395 		 */
1396 		if (ifp->if_mtu != oldmtu) {
1397 #ifdef INET6
1398 			nd6_setmtu(ifp);
1399 #endif
1400 		}
1401 		return (error);
1402 	}
1403 
1404 	case SIOCADDMULTI:
1405 	case SIOCDELMULTI:
1406 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1407 		if (error)
1408 			return (error);
1409 
1410 		/* Don't allow group membership on non-multicast interfaces. */
1411 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
1412 			return EOPNOTSUPP;
1413 
1414 		/* Don't let users screw up protocols' entries. */
1415 		if (ifr->ifr_addr.sa_family != AF_LINK)
1416 			return EINVAL;
1417 
1418 		if (cmd == SIOCADDMULTI) {
1419 			struct ifmultiaddr *ifma;
1420 			error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1421 		} else {
1422 			error = if_delmulti(ifp, &ifr->ifr_addr);
1423 		}
1424 		if (error == 0)
1425 			getmicrotime(&ifp->if_lastchange);
1426 		return error;
1427 
1428 	case SIOCSIFPHYADDR:
1429 	case SIOCDIFPHYADDR:
1430 #ifdef INET6
1431 	case SIOCSIFPHYADDR_IN6:
1432 #endif
1433 	case SIOCSLIFPHYADDR:
1434         case SIOCSIFMEDIA:
1435 	case SIOCSIFGENERIC:
1436 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1437 		if (error)
1438 			return (error);
1439 		if (ifp->if_ioctl == 0)
1440 			return (EOPNOTSUPP);
1441 		lwkt_serialize_enter(ifp->if_serializer);
1442 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1443 		lwkt_serialize_exit(ifp->if_serializer);
1444 		if (error == 0)
1445 			getmicrotime(&ifp->if_lastchange);
1446 		return error;
1447 
1448 	case SIOCGIFSTATUS:
1449 		ifs = (struct ifstat *)data;
1450 		ifs->ascii[0] = '\0';
1451 
1452 	case SIOCGIFPSRCADDR:
1453 	case SIOCGIFPDSTADDR:
1454 	case SIOCGLIFPHYADDR:
1455 	case SIOCGIFMEDIA:
1456 	case SIOCGIFGENERIC:
1457 		if (ifp->if_ioctl == NULL)
1458 			return (EOPNOTSUPP);
1459 		lwkt_serialize_enter(ifp->if_serializer);
1460 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1461 		lwkt_serialize_exit(ifp->if_serializer);
1462 		return (error);
1463 
1464 	case SIOCSIFLLADDR:
1465 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1466 		if (error)
1467 			return (error);
1468 		return if_setlladdr(ifp,
1469 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1470 
1471 	default:
1472 		oif_flags = ifp->if_flags;
1473 		if (so->so_proto == 0)
1474 			return (EOPNOTSUPP);
1475 #ifndef COMPAT_43
1476 		error = so_pru_control(so, cmd, data, ifp);
1477 #else
1478 	    {
1479 		int ocmd = cmd;
1480 
1481 		switch (cmd) {
1482 
1483 		case SIOCSIFDSTADDR:
1484 		case SIOCSIFADDR:
1485 		case SIOCSIFBRDADDR:
1486 		case SIOCSIFNETMASK:
1487 #if BYTE_ORDER != BIG_ENDIAN
1488 			if (ifr->ifr_addr.sa_family == 0 &&
1489 			    ifr->ifr_addr.sa_len < 16) {
1490 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1491 				ifr->ifr_addr.sa_len = 16;
1492 			}
1493 #else
1494 			if (ifr->ifr_addr.sa_len == 0)
1495 				ifr->ifr_addr.sa_len = 16;
1496 #endif
1497 			break;
1498 
1499 		case OSIOCGIFADDR:
1500 			cmd = SIOCGIFADDR;
1501 			break;
1502 
1503 		case OSIOCGIFDSTADDR:
1504 			cmd = SIOCGIFDSTADDR;
1505 			break;
1506 
1507 		case OSIOCGIFBRDADDR:
1508 			cmd = SIOCGIFBRDADDR;
1509 			break;
1510 
1511 		case OSIOCGIFNETMASK:
1512 			cmd = SIOCGIFNETMASK;
1513 		}
1514 		error =  so_pru_control(so, cmd, data, ifp);
1515 		switch (ocmd) {
1516 
1517 		case OSIOCGIFADDR:
1518 		case OSIOCGIFDSTADDR:
1519 		case OSIOCGIFBRDADDR:
1520 		case OSIOCGIFNETMASK:
1521 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1522 
1523 		}
1524 	    }
1525 #endif /* COMPAT_43 */
1526 
1527 		if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1528 #ifdef INET6
1529 			DELAY(100);/* XXX: temporary workaround for fxp issue*/
1530 			if (ifp->if_flags & IFF_UP) {
1531 				crit_enter();
1532 				in6_if_up(ifp);
1533 				crit_exit();
1534 			}
1535 #endif
1536 		}
1537 		return (error);
1538 
1539 	}
1540 	return (0);
1541 }
1542 
1543 /*
1544  * Set/clear promiscuous mode on interface ifp based on the truth value
1545  * of pswitch.  The calls are reference counted so that only the first
1546  * "on" request actually has an effect, as does the final "off" request.
1547  * Results are undefined if the "off" and "on" requests are not matched.
1548  */
1549 int
1550 ifpromisc(struct ifnet *ifp, int pswitch)
1551 {
1552 	struct ifreq ifr;
1553 	int error;
1554 	int oldflags;
1555 
1556 	oldflags = ifp->if_flags;
1557 	if (ifp->if_flags & IFF_PPROMISC) {
1558 		/* Do nothing if device is in permanently promiscuous mode */
1559 		ifp->if_pcount += pswitch ? 1 : -1;
1560 		return (0);
1561 	}
1562 	if (pswitch) {
1563 		/*
1564 		 * If the device is not configured up, we cannot put it in
1565 		 * promiscuous mode.
1566 		 */
1567 		if ((ifp->if_flags & IFF_UP) == 0)
1568 			return (ENETDOWN);
1569 		if (ifp->if_pcount++ != 0)
1570 			return (0);
1571 		ifp->if_flags |= IFF_PROMISC;
1572 		log(LOG_INFO, "%s: promiscuous mode enabled\n",
1573 		    ifp->if_xname);
1574 	} else {
1575 		if (--ifp->if_pcount > 0)
1576 			return (0);
1577 		ifp->if_flags &= ~IFF_PROMISC;
1578 		log(LOG_INFO, "%s: promiscuous mode disabled\n",
1579 		    ifp->if_xname);
1580 	}
1581 	ifr.ifr_flags = ifp->if_flags;
1582 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
1583 	lwkt_serialize_enter(ifp->if_serializer);
1584 	error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1585 				 (struct ucred *)NULL);
1586 	lwkt_serialize_exit(ifp->if_serializer);
1587 	if (error == 0)
1588 		rt_ifmsg(ifp);
1589 	else
1590 		ifp->if_flags = oldflags;
1591 	return error;
1592 }
1593 
1594 /*
1595  * Return interface configuration
1596  * of system.  List may be used
1597  * in later ioctl's (above) to get
1598  * other information.
1599  */
1600 static int
1601 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
1602 {
1603 	struct ifconf *ifc = (struct ifconf *)data;
1604 	struct ifnet *ifp;
1605 	struct sockaddr *sa;
1606 	struct ifreq ifr, *ifrp;
1607 	int space = ifc->ifc_len, error = 0;
1608 
1609 	ifrp = ifc->ifc_req;
1610 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1611 		struct ifaddr_container *ifac;
1612 		int addrs;
1613 
1614 		if (space <= sizeof ifr)
1615 			break;
1616 
1617 		/*
1618 		 * Zero the stack declared structure first to prevent
1619 		 * memory disclosure.
1620 		 */
1621 		bzero(&ifr, sizeof(ifr));
1622 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
1623 		    >= sizeof(ifr.ifr_name)) {
1624 			error = ENAMETOOLONG;
1625 			break;
1626 		}
1627 
1628 		addrs = 0;
1629 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1630 			struct ifaddr *ifa = ifac->ifa;
1631 
1632 			if (space <= sizeof ifr)
1633 				break;
1634 			sa = ifa->ifa_addr;
1635 			if (cred->cr_prison &&
1636 			    prison_if(cred, sa))
1637 				continue;
1638 			addrs++;
1639 #ifdef COMPAT_43
1640 			if (cmd == OSIOCGIFCONF) {
1641 				struct osockaddr *osa =
1642 					 (struct osockaddr *)&ifr.ifr_addr;
1643 				ifr.ifr_addr = *sa;
1644 				osa->sa_family = sa->sa_family;
1645 				error = copyout(&ifr, ifrp, sizeof ifr);
1646 				ifrp++;
1647 			} else
1648 #endif
1649 			if (sa->sa_len <= sizeof(*sa)) {
1650 				ifr.ifr_addr = *sa;
1651 				error = copyout(&ifr, ifrp, sizeof ifr);
1652 				ifrp++;
1653 			} else {
1654 				if (space < (sizeof ifr) + sa->sa_len -
1655 					    sizeof(*sa))
1656 					break;
1657 				space -= sa->sa_len - sizeof(*sa);
1658 				error = copyout(&ifr, ifrp,
1659 						sizeof ifr.ifr_name);
1660 				if (error == 0)
1661 					error = copyout(sa, &ifrp->ifr_addr,
1662 							sa->sa_len);
1663 				ifrp = (struct ifreq *)
1664 					(sa->sa_len + (caddr_t)&ifrp->ifr_addr);
1665 			}
1666 			if (error)
1667 				break;
1668 			space -= sizeof ifr;
1669 		}
1670 		if (error)
1671 			break;
1672 		if (!addrs) {
1673 			bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
1674 			error = copyout(&ifr, ifrp, sizeof ifr);
1675 			if (error)
1676 				break;
1677 			space -= sizeof ifr;
1678 			ifrp++;
1679 		}
1680 	}
1681 	ifc->ifc_len -= space;
1682 	return (error);
1683 }
1684 
1685 /*
1686  * Just like if_promisc(), but for all-multicast-reception mode.
1687  */
1688 int
1689 if_allmulti(struct ifnet *ifp, int onswitch)
1690 {
1691 	int error = 0;
1692 	struct ifreq ifr;
1693 
1694 	crit_enter();
1695 
1696 	if (onswitch) {
1697 		if (ifp->if_amcount++ == 0) {
1698 			ifp->if_flags |= IFF_ALLMULTI;
1699 			ifr.ifr_flags = ifp->if_flags;
1700 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
1701 			lwkt_serialize_enter(ifp->if_serializer);
1702 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1703 					      (struct ucred *)NULL);
1704 			lwkt_serialize_exit(ifp->if_serializer);
1705 		}
1706 	} else {
1707 		if (ifp->if_amcount > 1) {
1708 			ifp->if_amcount--;
1709 		} else {
1710 			ifp->if_amcount = 0;
1711 			ifp->if_flags &= ~IFF_ALLMULTI;
1712 			ifr.ifr_flags = ifp->if_flags;
1713 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
1714 			lwkt_serialize_enter(ifp->if_serializer);
1715 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1716 					      (struct ucred *)NULL);
1717 			lwkt_serialize_exit(ifp->if_serializer);
1718 		}
1719 	}
1720 
1721 	crit_exit();
1722 
1723 	if (error == 0)
1724 		rt_ifmsg(ifp);
1725 	return error;
1726 }
1727 
1728 /*
1729  * Add a multicast listenership to the interface in question.
1730  * The link layer provides a routine which converts
1731  */
1732 int
1733 if_addmulti(
1734 	struct ifnet *ifp,	/* interface to manipulate */
1735 	struct sockaddr *sa,	/* address to add */
1736 	struct ifmultiaddr **retifma)
1737 {
1738 	struct sockaddr *llsa, *dupsa;
1739 	int error;
1740 	struct ifmultiaddr *ifma;
1741 
1742 	/*
1743 	 * If the matching multicast address already exists
1744 	 * then don't add a new one, just add a reference
1745 	 */
1746 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1747 		if (sa_equal(sa, ifma->ifma_addr)) {
1748 			ifma->ifma_refcount++;
1749 			if (retifma)
1750 				*retifma = ifma;
1751 			return 0;
1752 		}
1753 	}
1754 
1755 	/*
1756 	 * Give the link layer a chance to accept/reject it, and also
1757 	 * find out which AF_LINK address this maps to, if it isn't one
1758 	 * already.
1759 	 */
1760 	if (ifp->if_resolvemulti) {
1761 		lwkt_serialize_enter(ifp->if_serializer);
1762 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
1763 		lwkt_serialize_exit(ifp->if_serializer);
1764 		if (error)
1765 			return error;
1766 	} else {
1767 		llsa = 0;
1768 	}
1769 
1770 	MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, M_WAITOK);
1771 	MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, M_WAITOK);
1772 	bcopy(sa, dupsa, sa->sa_len);
1773 
1774 	ifma->ifma_addr = dupsa;
1775 	ifma->ifma_lladdr = llsa;
1776 	ifma->ifma_ifp = ifp;
1777 	ifma->ifma_refcount = 1;
1778 	ifma->ifma_protospec = 0;
1779 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
1780 
1781 	/*
1782 	 * Some network interfaces can scan the address list at
1783 	 * interrupt time; lock them out.
1784 	 */
1785 	crit_enter();
1786 	LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1787 	crit_exit();
1788 	*retifma = ifma;
1789 
1790 	if (llsa != 0) {
1791 		LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1792 			if (sa_equal(ifma->ifma_addr, llsa))
1793 				break;
1794 		}
1795 		if (ifma) {
1796 			ifma->ifma_refcount++;
1797 		} else {
1798 			MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma,
1799 			       M_IFMADDR, M_WAITOK);
1800 			MALLOC(dupsa, struct sockaddr *, llsa->sa_len,
1801 			       M_IFMADDR, M_WAITOK);
1802 			bcopy(llsa, dupsa, llsa->sa_len);
1803 			ifma->ifma_addr = dupsa;
1804 			ifma->ifma_ifp = ifp;
1805 			ifma->ifma_refcount = 1;
1806 			crit_enter();
1807 			LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1808 			crit_exit();
1809 		}
1810 	}
1811 	/*
1812 	 * We are certain we have added something, so call down to the
1813 	 * interface to let them know about it.
1814 	 */
1815 	crit_enter();
1816 	lwkt_serialize_enter(ifp->if_serializer);
1817 	ifp->if_ioctl(ifp, SIOCADDMULTI, 0, (struct ucred *)NULL);
1818 	lwkt_serialize_exit(ifp->if_serializer);
1819 	crit_exit();
1820 
1821 	return 0;
1822 }
1823 
1824 /*
1825  * Remove a reference to a multicast address on this interface.  Yell
1826  * if the request does not match an existing membership.
1827  */
1828 int
1829 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
1830 {
1831 	struct ifmultiaddr *ifma;
1832 
1833 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1834 		if (sa_equal(sa, ifma->ifma_addr))
1835 			break;
1836 	if (ifma == 0)
1837 		return ENOENT;
1838 
1839 	if (ifma->ifma_refcount > 1) {
1840 		ifma->ifma_refcount--;
1841 		return 0;
1842 	}
1843 
1844 	rt_newmaddrmsg(RTM_DELMADDR, ifma);
1845 	sa = ifma->ifma_lladdr;
1846 	crit_enter();
1847 	LIST_REMOVE(ifma, ifma_link);
1848 	/*
1849 	 * Make sure the interface driver is notified
1850 	 * in the case of a link layer mcast group being left.
1851 	 */
1852 	if (ifma->ifma_addr->sa_family == AF_LINK && sa == 0) {
1853 		lwkt_serialize_enter(ifp->if_serializer);
1854 		ifp->if_ioctl(ifp, SIOCDELMULTI, 0, (struct ucred *)NULL);
1855 		lwkt_serialize_exit(ifp->if_serializer);
1856 	}
1857 	crit_exit();
1858 	kfree(ifma->ifma_addr, M_IFMADDR);
1859 	kfree(ifma, M_IFMADDR);
1860 	if (sa == 0)
1861 		return 0;
1862 
1863 	/*
1864 	 * Now look for the link-layer address which corresponds to
1865 	 * this network address.  It had been squirreled away in
1866 	 * ifma->ifma_lladdr for this purpose (so we don't have
1867 	 * to call ifp->if_resolvemulti() again), and we saved that
1868 	 * value in sa above.  If some nasty deleted the
1869 	 * link-layer address out from underneath us, we can deal because
1870 	 * the address we stored was is not the same as the one which was
1871 	 * in the record for the link-layer address.  (So we don't complain
1872 	 * in that case.)
1873 	 */
1874 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1875 		if (sa_equal(sa, ifma->ifma_addr))
1876 			break;
1877 	if (ifma == 0)
1878 		return 0;
1879 
1880 	if (ifma->ifma_refcount > 1) {
1881 		ifma->ifma_refcount--;
1882 		return 0;
1883 	}
1884 
1885 	crit_enter();
1886 	lwkt_serialize_enter(ifp->if_serializer);
1887 	LIST_REMOVE(ifma, ifma_link);
1888 	ifp->if_ioctl(ifp, SIOCDELMULTI, 0, (struct ucred *)NULL);
1889 	lwkt_serialize_exit(ifp->if_serializer);
1890 	crit_exit();
1891 	kfree(ifma->ifma_addr, M_IFMADDR);
1892 	kfree(sa, M_IFMADDR);
1893 	kfree(ifma, M_IFMADDR);
1894 
1895 	return 0;
1896 }
1897 
1898 /*
1899  * Set the link layer address on an interface.
1900  *
1901  * At this time we only support certain types of interfaces,
1902  * and we don't allow the length of the address to change.
1903  */
1904 int
1905 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
1906 {
1907 	struct sockaddr_dl *sdl;
1908 	struct ifreq ifr;
1909 
1910 	sdl = IF_LLSOCKADDR(ifp);
1911 	if (sdl == NULL)
1912 		return (EINVAL);
1913 	if (len != sdl->sdl_alen)	/* don't allow length to change */
1914 		return (EINVAL);
1915 	switch (ifp->if_type) {
1916 	case IFT_ETHER:			/* these types use struct arpcom */
1917 	case IFT_XETHER:
1918 	case IFT_L2VLAN:
1919 		bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
1920 		bcopy(lladdr, LLADDR(sdl), len);
1921 		break;
1922 	default:
1923 		return (ENODEV);
1924 	}
1925 	/*
1926 	 * If the interface is already up, we need
1927 	 * to re-init it in order to reprogram its
1928 	 * address filter.
1929 	 */
1930 	lwkt_serialize_enter(ifp->if_serializer);
1931 	if ((ifp->if_flags & IFF_UP) != 0) {
1932 		struct ifaddr_container *ifac;
1933 
1934 		ifp->if_flags &= ~IFF_UP;
1935 		ifr.ifr_flags = ifp->if_flags;
1936 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
1937 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1938 			      (struct ucred *)NULL);
1939 		ifp->if_flags |= IFF_UP;
1940 		ifr.ifr_flags = ifp->if_flags;
1941 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
1942 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1943 				 (struct ucred *)NULL);
1944 #ifdef INET
1945 		/*
1946 		 * Also send gratuitous ARPs to notify other nodes about
1947 		 * the address change.
1948 		 */
1949 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1950 			struct ifaddr *ifa = ifac->ifa;
1951 
1952 			if (ifa->ifa_addr != NULL &&
1953 			    ifa->ifa_addr->sa_family == AF_INET)
1954 				arp_ifinit(ifp, ifa);
1955 		}
1956 #endif
1957 	}
1958 	lwkt_serialize_exit(ifp->if_serializer);
1959 	return (0);
1960 }
1961 
1962 struct ifmultiaddr *
1963 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
1964 {
1965 	struct ifmultiaddr *ifma;
1966 
1967 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1968 		if (sa_equal(ifma->ifma_addr, sa))
1969 			break;
1970 
1971 	return ifma;
1972 }
1973 
1974 /*
1975  * This function locates the first real ethernet MAC from a network
1976  * card and loads it into node, returning 0 on success or ENOENT if
1977  * no suitable interfaces were found.  It is used by the uuid code to
1978  * generate a unique 6-byte number.
1979  */
1980 int
1981 if_getanyethermac(uint16_t *node, int minlen)
1982 {
1983 	struct ifnet *ifp;
1984 	struct sockaddr_dl *sdl;
1985 
1986 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1987 		if (ifp->if_type != IFT_ETHER)
1988 			continue;
1989 		sdl = IF_LLSOCKADDR(ifp);
1990 		if (sdl->sdl_alen < minlen)
1991 			continue;
1992 		bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
1993 		      minlen);
1994 		return(0);
1995 	}
1996 	return (ENOENT);
1997 }
1998 
1999 /*
2000  * The name argument must be a pointer to storage which will last as
2001  * long as the interface does.  For physical devices, the result of
2002  * device_get_name(dev) is a good choice and for pseudo-devices a
2003  * static string works well.
2004  */
2005 void
2006 if_initname(struct ifnet *ifp, const char *name, int unit)
2007 {
2008 	ifp->if_dname = name;
2009 	ifp->if_dunit = unit;
2010 	if (unit != IF_DUNIT_NONE)
2011 		ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2012 	else
2013 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2014 }
2015 
2016 int
2017 if_printf(struct ifnet *ifp, const char *fmt, ...)
2018 {
2019 	__va_list ap;
2020 	int retval;
2021 
2022 	retval = kprintf("%s: ", ifp->if_xname);
2023 	__va_start(ap, fmt);
2024 	retval += kvprintf(fmt, ap);
2025 	__va_end(ap);
2026 	return (retval);
2027 }
2028 
2029 void
2030 ifq_set_classic(struct ifaltq *ifq)
2031 {
2032 	ifq->altq_enqueue = ifq_classic_enqueue;
2033 	ifq->altq_dequeue = ifq_classic_dequeue;
2034 	ifq->altq_request = ifq_classic_request;
2035 }
2036 
2037 int
2038 ifq_classic_enqueue(struct ifaltq *ifq, struct mbuf *m,
2039 		    struct altq_pktattr *pa __unused)
2040 {
2041 	logifq(enqueue, ifq);
2042 	if (IF_QFULL(ifq)) {
2043 		m_freem(m);
2044 		return(ENOBUFS);
2045 	} else {
2046 		IF_ENQUEUE(ifq, m);
2047 		return(0);
2048 	}
2049 }
2050 
2051 struct mbuf *
2052 ifq_classic_dequeue(struct ifaltq *ifq, struct mbuf *mpolled, int op)
2053 {
2054 	struct mbuf *m;
2055 
2056 	switch (op) {
2057 	case ALTDQ_POLL:
2058 		IF_POLL(ifq, m);
2059 		break;
2060 	case ALTDQ_REMOVE:
2061 		logifq(dequeue, ifq);
2062 		IF_DEQUEUE(ifq, m);
2063 		break;
2064 	default:
2065 		panic("unsupported ALTQ dequeue op: %d", op);
2066 	}
2067 	KKASSERT(mpolled == NULL || mpolled == m);
2068 	return(m);
2069 }
2070 
2071 int
2072 ifq_classic_request(struct ifaltq *ifq, int req, void *arg)
2073 {
2074 	switch (req) {
2075 	case ALTRQ_PURGE:
2076 		IF_DRAIN(ifq);
2077 		break;
2078 	default:
2079 		panic("unsupported ALTQ request: %d", req);
2080 	}
2081 	return(0);
2082 }
2083 
2084 int
2085 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2086 {
2087 	struct ifaltq *ifq = &ifp->if_snd;
2088 	int running = 0, error, start = 0;
2089 
2090 	ASSERT_NOT_SERIALIZED(ifp->if_serializer);
2091 
2092 	ALTQ_LOCK(ifq);
2093 	error = ifq_enqueue_locked(ifq, m, pa);
2094 	if (error) {
2095 		ALTQ_UNLOCK(ifq);
2096 		return error;
2097 	}
2098 	if (!ifq->altq_started) {
2099 		/*
2100 		 * Hold the interlock of ifnet.if_start
2101 		 */
2102 		ifq->altq_started = 1;
2103 		start = 1;
2104 	}
2105 	ALTQ_UNLOCK(ifq);
2106 
2107 	ifp->if_obytes += m->m_pkthdr.len;
2108 	if (m->m_flags & M_MCAST)
2109 		ifp->if_omcasts++;
2110 
2111 	if (!start) {
2112 		logifstart(avoid, ifp);
2113 		return 0;
2114 	}
2115 
2116 	if (ifq_dispatch_schedonly) {
2117 		/*
2118 		 * Always schedule ifnet.if_start on ifnet's CPU,
2119 		 * short circuit the rest of this function.
2120 		 */
2121 		logifstart(sched, ifp);
2122 		if_start_schedule(ifp);
2123 		return 0;
2124 	}
2125 
2126 	/*
2127 	 * Try to do direct ifnet.if_start first, if there is
2128 	 * contention on ifnet's serializer, ifnet.if_start will
2129 	 * be scheduled on ifnet's CPU.
2130 	 */
2131 	if (!lwkt_serialize_try(ifp->if_serializer)) {
2132 		/*
2133 		 * ifnet serializer contention happened,
2134 		 * ifnet.if_start is scheduled on ifnet's
2135 		 * CPU, and we keep going.
2136 		 */
2137 		logifstart(contend_sched, ifp);
2138 		if_start_schedule(ifp);
2139 		return 0;
2140 	}
2141 
2142 	if ((ifp->if_flags & IFF_OACTIVE) == 0) {
2143 		logifstart(run, ifp);
2144 		ifp->if_start(ifp);
2145 		if ((ifp->if_flags &
2146 		     (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
2147 			running = 1;
2148 	}
2149 
2150 	lwkt_serialize_exit(ifp->if_serializer);
2151 
2152 	if (ifq_dispatch_schednochk || if_start_need_schedule(ifq, running)) {
2153 		/*
2154 		 * More data need to be transmitted, ifnet.if_start is
2155 		 * scheduled on ifnet's CPU, and we keep going.
2156 		 * NOTE: ifnet.if_start interlock is not released.
2157 		 */
2158 		logifstart(sched, ifp);
2159 		if_start_schedule(ifp);
2160 	}
2161 	return 0;
2162 }
2163 
2164 void *
2165 ifa_create(int size, int flags)
2166 {
2167 	struct ifaddr *ifa;
2168 	int i;
2169 
2170 	KASSERT(size >= sizeof(*ifa), ("ifaddr size too small\n"));
2171 
2172 	ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2173 	if (ifa == NULL)
2174 		return NULL;
2175 
2176 	ifa->ifa_containers = kmalloc(ncpus * sizeof(struct ifaddr_container),
2177 				      M_IFADDR, M_WAITOK | M_ZERO);
2178 	ifa->ifa_ncnt = ncpus;
2179 	for (i = 0; i < ncpus; ++i) {
2180 		struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2181 
2182 		ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2183 		ifac->ifa = ifa;
2184 		ifac->ifa_refcnt = 1;
2185 	}
2186 #ifdef IFADDR_DEBUG
2187 	kprintf("alloc ifa %p %d\n", ifa, size);
2188 #endif
2189 	return ifa;
2190 }
2191 
2192 void
2193 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2194 {
2195 	struct ifaddr *ifa = ifac->ifa;
2196 
2197 	KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2198 	KKASSERT(ifac->ifa_refcnt == 0);
2199 	KASSERT(ifac->ifa_listmask == 0,
2200 		("ifa is still on %#x lists\n", ifac->ifa_listmask));
2201 
2202 	ifac->ifa_magic = IFA_CONTAINER_DEAD;
2203 
2204 #ifdef IFADDR_DEBUG_VERBOSE
2205 	kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2206 #endif
2207 
2208 	KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2209 		("invalid # of ifac, %d\n", ifa->ifa_ncnt));
2210 	if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2211 #ifdef IFADDR_DEBUG
2212 		kprintf("free ifa %p\n", ifa);
2213 #endif
2214 		kfree(ifa->ifa_containers, M_IFADDR);
2215 		kfree(ifa, M_IFADDR);
2216 	}
2217 }
2218 
2219 static void
2220 ifa_iflink_dispatch(struct netmsg *nmsg)
2221 {
2222 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2223 	struct ifaddr *ifa = msg->ifa;
2224 	struct ifnet *ifp = msg->ifp;
2225 	int cpu = mycpuid;
2226 	struct ifaddr_container *ifac;
2227 
2228 	crit_enter();
2229 
2230 	ifac = &ifa->ifa_containers[cpu];
2231 	ASSERT_IFAC_VALID(ifac);
2232 	KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2233 		("ifaddr is on if_addrheads\n"));
2234 
2235 	ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2236 	if (msg->tail)
2237 		TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2238 	else
2239 		TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2240 
2241 	crit_exit();
2242 
2243 	ifa_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
2244 }
2245 
2246 void
2247 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2248 {
2249 	struct netmsg_ifaddr msg;
2250 
2251 	netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2252 		    ifa_iflink_dispatch);
2253 	msg.ifa = ifa;
2254 	msg.ifp = ifp;
2255 	msg.tail = tail;
2256 
2257 	ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2258 }
2259 
2260 static void
2261 ifa_ifunlink_dispatch(struct netmsg *nmsg)
2262 {
2263 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2264 	struct ifaddr *ifa = msg->ifa;
2265 	struct ifnet *ifp = msg->ifp;
2266 	int cpu = mycpuid;
2267 	struct ifaddr_container *ifac;
2268 
2269 	crit_enter();
2270 
2271 	ifac = &ifa->ifa_containers[cpu];
2272 	ASSERT_IFAC_VALID(ifac);
2273 	KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2274 		("ifaddr is not on if_addrhead\n"));
2275 
2276 	TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2277 	ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2278 
2279 	crit_exit();
2280 
2281 	ifa_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
2282 }
2283 
2284 void
2285 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2286 {
2287 	struct netmsg_ifaddr msg;
2288 
2289 	netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2290 		    ifa_ifunlink_dispatch);
2291 	msg.ifa = ifa;
2292 	msg.ifp = ifp;
2293 
2294 	ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2295 }
2296 
2297 static void
2298 ifa_destroy_dispatch(struct netmsg *nmsg)
2299 {
2300 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2301 
2302 	IFAFREE(msg->ifa);
2303 	ifa_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2304 }
2305 
2306 void
2307 ifa_destroy(struct ifaddr *ifa)
2308 {
2309 	struct netmsg_ifaddr msg;
2310 
2311 	netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2312 		    ifa_destroy_dispatch);
2313 	msg.ifa = ifa;
2314 
2315 	ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2316 }
2317 
2318 struct lwkt_port *
2319 ifnet_portfn(int cpu)
2320 {
2321 	return &ifnet_threads[cpu].td_msgport;
2322 }
2323 
2324 void
2325 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2326 {
2327 	KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2328 
2329 	if (next_cpu < ncpus)
2330 		lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2331 	else
2332 		lwkt_replymsg(lmsg, 0);
2333 }
2334 
2335 int
2336 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
2337 {
2338 	KKASSERT(cpu < ncpus);
2339 	return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
2340 }
2341 
2342 void
2343 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
2344 {
2345 	KKASSERT(cpu < ncpus);
2346 	lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
2347 }
2348 
2349 static void
2350 ifnetinit(void *dummy __unused)
2351 {
2352 	int i;
2353 
2354 	for (i = 0; i < ncpus; ++i) {
2355 		struct thread *thr = &ifnet_threads[i];
2356 
2357 		lwkt_create(netmsg_service_loop, &ifnet_mpsafe_thread, NULL,
2358 			    thr, TDF_NETWORK | TDF_MPSAFE, i, "ifnet %d", i);
2359 		netmsg_service_port_init(&thr->td_msgport);
2360 	}
2361 }
2362