xref: /dflybsd-src/sys/net/if.c (revision 1d06442f0168c72bd397c0c2b0763cb3a9f72217)
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  */
36 
37 #include "opt_compat.h"
38 #include "opt_inet6.h"
39 #include "opt_inet.h"
40 #include "opt_ifpoll.h"
41 
42 #include <sys/param.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/systm.h>
46 #include <sys/proc.h>
47 #include <sys/priv.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/socketops.h>
52 #include <sys/protosw.h>
53 #include <sys/kernel.h>
54 #include <sys/ktr.h>
55 #include <sys/mutex.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/serialize.h>
62 #include <sys/bus.h>
63 
64 #include <sys/thread2.h>
65 #include <sys/msgport2.h>
66 #include <sys/mutex2.h>
67 
68 #include <net/if.h>
69 #include <net/if_arp.h>
70 #include <net/if_dl.h>
71 #include <net/if_types.h>
72 #include <net/if_var.h>
73 #include <net/ifq_var.h>
74 #include <net/radix.h>
75 #include <net/route.h>
76 #include <net/if_clone.h>
77 #include <net/netisr.h>
78 #include <net/netmsg2.h>
79 
80 #include <machine/atomic.h>
81 #include <machine/stdarg.h>
82 #include <machine/smp.h>
83 
84 #if defined(INET) || defined(INET6)
85 /*XXX*/
86 #include <netinet/in.h>
87 #include <netinet/in_var.h>
88 #include <netinet/if_ether.h>
89 #ifdef INET6
90 #include <netinet6/in6_var.h>
91 #include <netinet6/in6_ifattach.h>
92 #endif
93 #endif
94 
95 #if defined(COMPAT_43)
96 #include <emulation/43bsd/43bsd_socket.h>
97 #endif /* COMPAT_43 */
98 
99 struct netmsg_ifaddr {
100 	struct netmsg_base base;
101 	struct ifaddr	*ifa;
102 	struct ifnet	*ifp;
103 	int		tail;
104 };
105 
106 struct ifsubq_stage_head {
107 	TAILQ_HEAD(, ifsubq_stage)	stg_head;
108 } __cachealign;
109 
110 /*
111  * System initialization
112  */
113 static void	if_attachdomain(void *);
114 static void	if_attachdomain1(struct ifnet *);
115 static int	ifconf(u_long, caddr_t, struct ucred *);
116 static void	ifinit(void *);
117 static void	ifnetinit(void *);
118 static void	if_slowtimo(void *);
119 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
120 static int	if_rtdel(struct radix_node *, void *);
121 
122 /* Helper functions */
123 static void	ifsq_watchdog_reset(struct ifsubq_watchdog *);
124 
125 #ifdef INET6
126 /*
127  * XXX: declare here to avoid to include many inet6 related files..
128  * should be more generalized?
129  */
130 extern void	nd6_setmtu(struct ifnet *);
131 #endif
132 
133 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
134 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
135 
136 static int ifsq_stage_cntmax = 4;
137 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
138 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
139     &ifsq_stage_cntmax, 0, "ifq staging packet count max");
140 
141 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
142 /* Must be after netisr_init */
143 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
144 
145 static  if_com_alloc_t *if_com_alloc[256];
146 static  if_com_free_t *if_com_free[256];
147 
148 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
149 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
150 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
151 
152 int			ifqmaxlen = IFQ_MAXLEN;
153 struct ifnethead	ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
154 
155 struct callout		if_slowtimo_timer;
156 
157 int			if_index = 0;
158 struct ifnet		**ifindex2ifnet = NULL;
159 static struct thread	ifnet_threads[MAXCPU];
160 
161 static struct ifsubq_stage_head	ifsubq_stage_heads[MAXCPU];
162 
163 #ifdef notyet
164 #define IFQ_KTR_STRING		"ifq=%p"
165 #define IFQ_KTR_ARGS	struct ifaltq *ifq
166 #ifndef KTR_IFQ
167 #define KTR_IFQ			KTR_ALL
168 #endif
169 KTR_INFO_MASTER(ifq);
170 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
171 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
172 #define logifq(name, arg)	KTR_LOG(ifq_ ## name, arg)
173 
174 #define IF_START_KTR_STRING	"ifp=%p"
175 #define IF_START_KTR_ARGS	struct ifnet *ifp
176 #ifndef KTR_IF_START
177 #define KTR_IF_START		KTR_ALL
178 #endif
179 KTR_INFO_MASTER(if_start);
180 KTR_INFO(KTR_IF_START, if_start, run, 0,
181 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
182 KTR_INFO(KTR_IF_START, if_start, sched, 1,
183 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
184 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
185 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
186 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
187 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
188 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
189 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
190 #define logifstart(name, arg)	KTR_LOG(if_start_ ## name, arg)
191 #endif
192 
193 TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
194 
195 /*
196  * Network interface utility routines.
197  *
198  * Routines with ifa_ifwith* names take sockaddr *'s as
199  * parameters.
200  */
201 /* ARGSUSED*/
202 void
203 ifinit(void *dummy)
204 {
205 	struct ifnet *ifp;
206 
207 	callout_init(&if_slowtimo_timer);
208 
209 	crit_enter();
210 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
211 		if (ifp->if_snd.altq_maxlen == 0) {
212 			if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
213 			ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
214 		}
215 	}
216 	crit_exit();
217 
218 	if_slowtimo(0);
219 }
220 
221 static void
222 ifsq_ifstart_ipifunc(void *arg)
223 {
224 	struct ifaltq_subque *ifsq = arg;
225 	struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
226 
227 	crit_enter();
228 	if (lmsg->ms_flags & MSGF_DONE)
229 		lwkt_sendmsg(netisr_portfn(mycpuid), lmsg);
230 	crit_exit();
231 }
232 
233 static __inline void
234 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
235 {
236 	KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
237 	TAILQ_REMOVE(&head->stg_head, stage, stg_link);
238 	stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
239 	stage->stg_cnt = 0;
240 	stage->stg_len = 0;
241 }
242 
243 static __inline void
244 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
245 {
246 	KKASSERT((stage->stg_flags &
247 	    (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
248 	stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
249 	TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
250 }
251 
252 /*
253  * Schedule ifnet.if_start on ifnet's CPU
254  */
255 static void
256 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
257 {
258 	int cpu;
259 
260 	if (!force && curthread->td_type == TD_TYPE_NETISR &&
261 	    ifsq_stage_cntmax > 0) {
262 		struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
263 
264 		stage->stg_cnt = 0;
265 		stage->stg_len = 0;
266 		if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
267 			ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
268 		stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
269 		return;
270 	}
271 
272 	cpu = ifsq_get_cpuid(ifsq);
273 	if (cpu != mycpuid)
274 		lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
275 	else
276 		ifsq_ifstart_ipifunc(ifsq);
277 }
278 
279 /*
280  * NOTE:
281  * This function will release ifnet.if_start interlock,
282  * if ifnet.if_start does not need to be scheduled
283  */
284 static __inline int
285 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
286 {
287 	if (!running || ifsq_is_empty(ifsq)
288 #ifdef ALTQ
289 	    || ifsq->ifsq_altq->altq_tbr != NULL
290 #endif
291 	) {
292 		ALTQ_SQ_LOCK(ifsq);
293 		/*
294 		 * ifnet.if_start interlock is released, if:
295 		 * 1) Hardware can not take any packets, due to
296 		 *    o  interface is marked down
297 		 *    o  hardware queue is full (ifq_is_oactive)
298 		 *    Under the second situation, hardware interrupt
299 		 *    or polling(4) will call/schedule ifnet.if_start
300 		 *    when hardware queue is ready
301 		 * 2) There is not packet in the ifnet.if_snd.
302 		 *    Further ifq_dispatch or ifq_handoff will call/
303 		 *    schedule ifnet.if_start
304 		 * 3) TBR is used and it does not allow further
305 		 *    dequeueing.
306 		 *    TBR callout will call ifnet.if_start
307 		 */
308 		if (!running || !ifsq_data_ready(ifsq)) {
309 			ifsq_clr_started(ifsq);
310 			ALTQ_SQ_UNLOCK(ifsq);
311 			return 0;
312 		}
313 		ALTQ_SQ_UNLOCK(ifsq);
314 	}
315 	return 1;
316 }
317 
318 static void
319 ifsq_ifstart_dispatch(netmsg_t msg)
320 {
321 	struct lwkt_msg *lmsg = &msg->base.lmsg;
322 	struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
323 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
324 	int running = 0, need_sched;
325 
326 	crit_enter();
327 	lwkt_replymsg(lmsg, 0);	/* reply ASAP */
328 	crit_exit();
329 
330 	if (mycpuid != ifsq_get_cpuid(ifsq)) {
331 		/*
332 		 * We need to chase the ifnet CPU change.
333 		 */
334 		ifsq_ifstart_schedule(ifsq, 1);
335 		return;
336 	}
337 
338 	ifnet_serialize_tx(ifp, ifsq);
339 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
340 		ifp->if_start(ifp, ifsq);
341 		if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
342 			running = 1;
343 	}
344 	need_sched = ifsq_ifstart_need_schedule(ifsq, running);
345 	ifnet_deserialize_tx(ifp, ifsq);
346 
347 	if (need_sched) {
348 		/*
349 		 * More data need to be transmitted, ifnet.if_start is
350 		 * scheduled on ifnet's CPU, and we keep going.
351 		 * NOTE: ifnet.if_start interlock is not released.
352 		 */
353 		ifsq_ifstart_schedule(ifsq, 0);
354 	}
355 }
356 
357 /* Device driver ifnet.if_start helper function */
358 void
359 ifsq_devstart(struct ifaltq_subque *ifsq)
360 {
361 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
362 	int running = 0;
363 
364 	ASSERT_IFNET_SERIALIZED_TX(ifp, ifsq);
365 
366 	ALTQ_SQ_LOCK(ifsq);
367 	if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
368 		ALTQ_SQ_UNLOCK(ifsq);
369 		return;
370 	}
371 	ifsq_set_started(ifsq);
372 	ALTQ_SQ_UNLOCK(ifsq);
373 
374 	ifp->if_start(ifp, ifsq);
375 
376 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
377 		running = 1;
378 
379 	if (ifsq_ifstart_need_schedule(ifsq, running)) {
380 		/*
381 		 * More data need to be transmitted, ifnet.if_start is
382 		 * scheduled on ifnet's CPU, and we keep going.
383 		 * NOTE: ifnet.if_start interlock is not released.
384 		 */
385 		ifsq_ifstart_schedule(ifsq, 0);
386 	}
387 }
388 
389 void
390 if_devstart(struct ifnet *ifp)
391 {
392 	ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
393 }
394 
395 /* Device driver ifnet.if_start schedule helper function */
396 void
397 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
398 {
399 	ifsq_ifstart_schedule(ifsq, 1);
400 }
401 
402 void
403 if_devstart_sched(struct ifnet *ifp)
404 {
405 	ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
406 }
407 
408 static void
409 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
410 {
411 	lwkt_serialize_enter(ifp->if_serializer);
412 }
413 
414 static void
415 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
416 {
417 	lwkt_serialize_exit(ifp->if_serializer);
418 }
419 
420 static int
421 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
422 {
423 	return lwkt_serialize_try(ifp->if_serializer);
424 }
425 
426 #ifdef INVARIANTS
427 static void
428 if_default_serialize_assert(struct ifnet *ifp,
429 			    enum ifnet_serialize slz __unused,
430 			    boolean_t serialized)
431 {
432 	if (serialized)
433 		ASSERT_SERIALIZED(ifp->if_serializer);
434 	else
435 		ASSERT_NOT_SERIALIZED(ifp->if_serializer);
436 }
437 #endif
438 
439 /*
440  * Attach an interface to the list of "active" interfaces.
441  *
442  * The serializer is optional.  If non-NULL access to the interface
443  * may be MPSAFE.
444  */
445 void
446 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
447 {
448 	unsigned socksize, ifasize;
449 	int namelen, masklen;
450 	struct sockaddr_dl *sdl;
451 	struct ifaddr *ifa;
452 	struct ifaltq *ifq;
453 	int i, q;
454 
455 	static int if_indexlim = 8;
456 
457 	if (ifp->if_serialize != NULL) {
458 		KASSERT(ifp->if_deserialize != NULL &&
459 			ifp->if_tryserialize != NULL &&
460 			ifp->if_serialize_assert != NULL,
461 			("serialize functions are partially setup"));
462 
463 		/*
464 		 * If the device supplies serialize functions,
465 		 * then clear if_serializer to catch any invalid
466 		 * usage of this field.
467 		 */
468 		KASSERT(serializer == NULL,
469 			("both serialize functions and default serializer "
470 			 "are supplied"));
471 		ifp->if_serializer = NULL;
472 	} else {
473 		KASSERT(ifp->if_deserialize == NULL &&
474 			ifp->if_tryserialize == NULL &&
475 			ifp->if_serialize_assert == NULL,
476 			("serialize functions are partially setup"));
477 		ifp->if_serialize = if_default_serialize;
478 		ifp->if_deserialize = if_default_deserialize;
479 		ifp->if_tryserialize = if_default_tryserialize;
480 #ifdef INVARIANTS
481 		ifp->if_serialize_assert = if_default_serialize_assert;
482 #endif
483 
484 		/*
485 		 * The serializer can be passed in from the device,
486 		 * allowing the same serializer to be used for both
487 		 * the interrupt interlock and the device queue.
488 		 * If not specified, the netif structure will use an
489 		 * embedded serializer.
490 		 */
491 		if (serializer == NULL) {
492 			serializer = &ifp->if_default_serializer;
493 			lwkt_serialize_init(serializer);
494 		}
495 		ifp->if_serializer = serializer;
496 	}
497 
498 	mtx_init(&ifp->if_ioctl_mtx);
499 	mtx_lock(&ifp->if_ioctl_mtx);
500 
501 	TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
502 	ifp->if_index = ++if_index;
503 
504 	/*
505 	 * XXX -
506 	 * The old code would work if the interface passed a pre-existing
507 	 * chain of ifaddrs to this code.  We don't trust our callers to
508 	 * properly initialize the tailq, however, so we no longer allow
509 	 * this unlikely case.
510 	 */
511 	ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
512 				    M_IFADDR, M_WAITOK | M_ZERO);
513 	for (i = 0; i < ncpus; ++i)
514 		TAILQ_INIT(&ifp->if_addrheads[i]);
515 
516 	TAILQ_INIT(&ifp->if_prefixhead);
517 	TAILQ_INIT(&ifp->if_multiaddrs);
518 	TAILQ_INIT(&ifp->if_groups);
519 	getmicrotime(&ifp->if_lastchange);
520 	if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
521 		unsigned int n;
522 		struct ifnet **q;
523 
524 		if_indexlim <<= 1;
525 
526 		/* grow ifindex2ifnet */
527 		n = if_indexlim * sizeof(*q);
528 		q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
529 		if (ifindex2ifnet) {
530 			bcopy(ifindex2ifnet, q, n/2);
531 			kfree(ifindex2ifnet, M_IFADDR);
532 		}
533 		ifindex2ifnet = q;
534 	}
535 
536 	ifindex2ifnet[if_index] = ifp;
537 
538 	/*
539 	 * create a Link Level name for this device
540 	 */
541 	namelen = strlen(ifp->if_xname);
542 	masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
543 	socksize = masklen + ifp->if_addrlen;
544 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
545 	if (socksize < sizeof(*sdl))
546 		socksize = sizeof(*sdl);
547 	socksize = ROUNDUP(socksize);
548 #undef ROUNDUP
549 	ifasize = sizeof(struct ifaddr) + 2 * socksize;
550 	ifa = ifa_create(ifasize, M_WAITOK);
551 	sdl = (struct sockaddr_dl *)(ifa + 1);
552 	sdl->sdl_len = socksize;
553 	sdl->sdl_family = AF_LINK;
554 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
555 	sdl->sdl_nlen = namelen;
556 	sdl->sdl_index = ifp->if_index;
557 	sdl->sdl_type = ifp->if_type;
558 	ifp->if_lladdr = ifa;
559 	ifa->ifa_ifp = ifp;
560 	ifa->ifa_rtrequest = link_rtrequest;
561 	ifa->ifa_addr = (struct sockaddr *)sdl;
562 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
563 	ifa->ifa_netmask = (struct sockaddr *)sdl;
564 	sdl->sdl_len = masklen;
565 	while (namelen != 0)
566 		sdl->sdl_data[--namelen] = 0xff;
567 	ifa_iflink(ifa, ifp, 0 /* Insert head */);
568 
569 	ifp->if_data_pcpu = kmalloc_cachealign(
570 	    ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
571 
572 	EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
573 	devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
574 
575 	if (ifp->if_mapsubq == NULL)
576 		ifp->if_mapsubq = ifq_mapsubq_default;
577 
578 	ifq = &ifp->if_snd;
579 	ifq->altq_type = 0;
580 	ifq->altq_disc = NULL;
581 	ifq->altq_flags &= ALTQF_CANTCHANGE;
582 	ifq->altq_tbr = NULL;
583 	ifq->altq_ifp = ifp;
584 
585 	if (ifq->altq_subq_cnt <= 0)
586 		ifq->altq_subq_cnt = 1;
587 	ifq->altq_subq = kmalloc_cachealign(
588 	    ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
589 	    M_DEVBUF, M_WAITOK | M_ZERO);
590 
591 	if (ifq->altq_maxlen == 0) {
592 		if_printf(ifp, "driver didn't set ifq_maxlen\n");
593 		ifq_set_maxlen(ifq, ifqmaxlen);
594 	}
595 
596 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
597 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
598 
599 		ALTQ_SQ_LOCK_INIT(ifsq);
600 		ifsq->ifsq_index = q;
601 
602 		ifsq->ifsq_altq = ifq;
603 		ifsq->ifsq_ifp = ifp;
604 
605 		ifsq->ifq_maxlen = ifq->altq_maxlen;
606 		ifsq->ifsq_prepended = NULL;
607 		ifsq->ifsq_started = 0;
608 		ifsq->ifsq_hw_oactive = 0;
609 		ifsq_set_cpuid(ifsq, 0);
610 
611 		ifsq->ifsq_stage =
612 		    kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
613 		    M_DEVBUF, M_WAITOK | M_ZERO);
614 		for (i = 0; i < ncpus; ++i)
615 			ifsq->ifsq_stage[i].stg_subq = ifsq;
616 
617 		ifsq->ifsq_ifstart_nmsg =
618 		    kmalloc(ncpus * sizeof(struct netmsg_base),
619 		    M_LWKTMSG, M_WAITOK);
620 		for (i = 0; i < ncpus; ++i) {
621 			netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
622 			    &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
623 			ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
624 		}
625 	}
626 	ifq_set_classic(ifq);
627 
628 	if (!SLIST_EMPTY(&domains))
629 		if_attachdomain1(ifp);
630 
631 	/* Announce the interface. */
632 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
633 
634 	mtx_unlock(&ifp->if_ioctl_mtx);
635 }
636 
637 static void
638 if_attachdomain(void *dummy)
639 {
640 	struct ifnet *ifp;
641 
642 	crit_enter();
643 	TAILQ_FOREACH(ifp, &ifnet, if_list)
644 		if_attachdomain1(ifp);
645 	crit_exit();
646 }
647 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
648 	if_attachdomain, NULL);
649 
650 static void
651 if_attachdomain1(struct ifnet *ifp)
652 {
653 	struct domain *dp;
654 
655 	crit_enter();
656 
657 	/* address family dependent data region */
658 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
659 	SLIST_FOREACH(dp, &domains, dom_next)
660 		if (dp->dom_ifattach)
661 			ifp->if_afdata[dp->dom_family] =
662 				(*dp->dom_ifattach)(ifp);
663 	crit_exit();
664 }
665 
666 /*
667  * Purge all addresses whose type is _not_ AF_LINK
668  */
669 void
670 if_purgeaddrs_nolink(struct ifnet *ifp)
671 {
672 	struct ifaddr_container *ifac, *next;
673 
674 	TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
675 			      ifa_link, next) {
676 		struct ifaddr *ifa = ifac->ifa;
677 
678 		/* Leave link ifaddr as it is */
679 		if (ifa->ifa_addr->sa_family == AF_LINK)
680 			continue;
681 #ifdef INET
682 		/* XXX: Ugly!! ad hoc just for INET */
683 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
684 			struct ifaliasreq ifr;
685 #ifdef IFADDR_DEBUG_VERBOSE
686 			int i;
687 
688 			kprintf("purge in4 addr %p: ", ifa);
689 			for (i = 0; i < ncpus; ++i)
690 				kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
691 			kprintf("\n");
692 #endif
693 
694 			bzero(&ifr, sizeof ifr);
695 			ifr.ifra_addr = *ifa->ifa_addr;
696 			if (ifa->ifa_dstaddr)
697 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
698 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
699 				       NULL) == 0)
700 				continue;
701 		}
702 #endif /* INET */
703 #ifdef INET6
704 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
705 #ifdef IFADDR_DEBUG_VERBOSE
706 			int i;
707 
708 			kprintf("purge in6 addr %p: ", ifa);
709 			for (i = 0; i < ncpus; ++i)
710 				kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
711 			kprintf("\n");
712 #endif
713 
714 			in6_purgeaddr(ifa);
715 			/* ifp_addrhead is already updated */
716 			continue;
717 		}
718 #endif /* INET6 */
719 		ifa_ifunlink(ifa, ifp);
720 		ifa_destroy(ifa);
721 	}
722 }
723 
724 static void
725 ifq_stage_detach_handler(netmsg_t nmsg)
726 {
727 	struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
728 	int q;
729 
730 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
731 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
732 		struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
733 
734 		if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
735 			ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
736 	}
737 	lwkt_replymsg(&nmsg->lmsg, 0);
738 }
739 
740 static void
741 ifq_stage_detach(struct ifaltq *ifq)
742 {
743 	struct netmsg_base base;
744 	int cpu;
745 
746 	netmsg_init(&base, NULL, &curthread->td_msgport, 0,
747 	    ifq_stage_detach_handler);
748 	base.lmsg.u.ms_resultp = ifq;
749 
750 	for (cpu = 0; cpu < ncpus; ++cpu)
751 		lwkt_domsg(netisr_portfn(cpu), &base.lmsg, 0);
752 }
753 
754 /*
755  * Detach an interface, removing it from the
756  * list of "active" interfaces.
757  */
758 void
759 if_detach(struct ifnet *ifp)
760 {
761 	struct radix_node_head	*rnh;
762 	int i, q;
763 	int cpu, origcpu;
764 	struct domain *dp;
765 
766 	EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
767 
768 	/*
769 	 * Remove routes and flush queues.
770 	 */
771 	crit_enter();
772 #ifdef IFPOLL_ENABLE
773 	if (ifp->if_flags & IFF_NPOLLING)
774 		ifpoll_deregister(ifp);
775 #endif
776 	if_down(ifp);
777 
778 #ifdef ALTQ
779 	if (ifq_is_enabled(&ifp->if_snd))
780 		altq_disable(&ifp->if_snd);
781 	if (ifq_is_attached(&ifp->if_snd))
782 		altq_detach(&ifp->if_snd);
783 #endif
784 
785 	/*
786 	 * Clean up all addresses.
787 	 */
788 	ifp->if_lladdr = NULL;
789 
790 	if_purgeaddrs_nolink(ifp);
791 	if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
792 		struct ifaddr *ifa;
793 
794 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
795 		KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
796 			("non-link ifaddr is left on if_addrheads"));
797 
798 		ifa_ifunlink(ifa, ifp);
799 		ifa_destroy(ifa);
800 		KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
801 			("there are still ifaddrs left on if_addrheads"));
802 	}
803 
804 #ifdef INET
805 	/*
806 	 * Remove all IPv4 kernel structures related to ifp.
807 	 */
808 	in_ifdetach(ifp);
809 #endif
810 
811 #ifdef INET6
812 	/*
813 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
814 	 * before removing routing entries below, since IPv6 interface direct
815 	 * routes are expected to be removed by the IPv6-specific kernel API.
816 	 * Otherwise, the kernel will detect some inconsistency and bark it.
817 	 */
818 	in6_ifdetach(ifp);
819 #endif
820 
821 	/*
822 	 * Delete all remaining routes using this interface
823 	 * Unfortuneatly the only way to do this is to slog through
824 	 * the entire routing table looking for routes which point
825 	 * to this interface...oh well...
826 	 */
827 	origcpu = mycpuid;
828 	for (cpu = 0; cpu < ncpus; cpu++) {
829 		lwkt_migratecpu(cpu);
830 		for (i = 1; i <= AF_MAX; i++) {
831 			if ((rnh = rt_tables[cpu][i]) == NULL)
832 				continue;
833 			rnh->rnh_walktree(rnh, if_rtdel, ifp);
834 		}
835 	}
836 	lwkt_migratecpu(origcpu);
837 
838 	/* Announce that the interface is gone. */
839 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
840 	devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
841 
842 	SLIST_FOREACH(dp, &domains, dom_next)
843 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
844 			(*dp->dom_ifdetach)(ifp,
845 				ifp->if_afdata[dp->dom_family]);
846 
847 	/*
848 	 * Remove interface from ifindex2ifp[] and maybe decrement if_index.
849 	 */
850 	ifindex2ifnet[ifp->if_index] = NULL;
851 	while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
852 		if_index--;
853 
854 	TAILQ_REMOVE(&ifnet, ifp, if_link);
855 	kfree(ifp->if_addrheads, M_IFADDR);
856 
857 	lwkt_synchronize_ipiqs("if_detach");
858 	ifq_stage_detach(&ifp->if_snd);
859 
860 	for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
861 		struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
862 
863 		kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
864 		kfree(ifsq->ifsq_stage, M_DEVBUF);
865 	}
866 	kfree(ifp->if_snd.altq_subq, M_DEVBUF);
867 
868 	kfree(ifp->if_data_pcpu, M_DEVBUF);
869 
870 	crit_exit();
871 }
872 
873 /*
874  * Create interface group without members
875  */
876 struct ifg_group *
877 if_creategroup(const char *groupname)
878 {
879         struct ifg_group        *ifg = NULL;
880 
881         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
882             M_TEMP, M_NOWAIT)) == NULL)
883                 return (NULL);
884 
885         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
886         ifg->ifg_refcnt = 0;
887         ifg->ifg_carp_demoted = 0;
888         TAILQ_INIT(&ifg->ifg_members);
889 #if NPF > 0
890         pfi_attach_ifgroup(ifg);
891 #endif
892         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
893 
894         return (ifg);
895 }
896 
897 /*
898  * Add a group to an interface
899  */
900 int
901 if_addgroup(struct ifnet *ifp, const char *groupname)
902 {
903 	struct ifg_list		*ifgl;
904 	struct ifg_group	*ifg = NULL;
905 	struct ifg_member	*ifgm;
906 
907 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
908 	    groupname[strlen(groupname) - 1] <= '9')
909 		return (EINVAL);
910 
911 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
912 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
913 			return (EEXIST);
914 
915 	if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
916 		return (ENOMEM);
917 
918 	if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
919 		kfree(ifgl, M_TEMP);
920 		return (ENOMEM);
921 	}
922 
923 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
924 		if (!strcmp(ifg->ifg_group, groupname))
925 			break;
926 
927 	if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
928 		kfree(ifgl, M_TEMP);
929 		kfree(ifgm, M_TEMP);
930 		return (ENOMEM);
931 	}
932 
933 	ifg->ifg_refcnt++;
934 	ifgl->ifgl_group = ifg;
935 	ifgm->ifgm_ifp = ifp;
936 
937 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
938 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
939 
940 #if NPF > 0
941 	pfi_group_change(groupname);
942 #endif
943 
944 	return (0);
945 }
946 
947 /*
948  * Remove a group from an interface
949  */
950 int
951 if_delgroup(struct ifnet *ifp, const char *groupname)
952 {
953 	struct ifg_list		*ifgl;
954 	struct ifg_member	*ifgm;
955 
956 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
957 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
958 			break;
959 	if (ifgl == NULL)
960 		return (ENOENT);
961 
962 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
963 
964 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
965 		if (ifgm->ifgm_ifp == ifp)
966 			break;
967 
968 	if (ifgm != NULL) {
969 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
970 		kfree(ifgm, M_TEMP);
971 	}
972 
973 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
974 		TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
975 #if NPF > 0
976 		pfi_detach_ifgroup(ifgl->ifgl_group);
977 #endif
978 		kfree(ifgl->ifgl_group, M_TEMP);
979 	}
980 
981 	kfree(ifgl, M_TEMP);
982 
983 #if NPF > 0
984 	pfi_group_change(groupname);
985 #endif
986 
987 	return (0);
988 }
989 
990 /*
991  * Stores all groups from an interface in memory pointed
992  * to by data
993  */
994 int
995 if_getgroup(caddr_t data, struct ifnet *ifp)
996 {
997 	int			 len, error;
998 	struct ifg_list		*ifgl;
999 	struct ifg_req		 ifgrq, *ifgp;
1000 	struct ifgroupreq	*ifgr = (struct ifgroupreq *)data;
1001 
1002 	if (ifgr->ifgr_len == 0) {
1003 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1004 			ifgr->ifgr_len += sizeof(struct ifg_req);
1005 		return (0);
1006 	}
1007 
1008 	len = ifgr->ifgr_len;
1009 	ifgp = ifgr->ifgr_groups;
1010 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1011 		if (len < sizeof(ifgrq))
1012 			return (EINVAL);
1013 		bzero(&ifgrq, sizeof ifgrq);
1014 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1015 		    sizeof(ifgrq.ifgrq_group));
1016 		if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1017 		    sizeof(struct ifg_req))))
1018 			return (error);
1019 		len -= sizeof(ifgrq);
1020 		ifgp++;
1021 	}
1022 
1023 	return (0);
1024 }
1025 
1026 /*
1027  * Stores all members of a group in memory pointed to by data
1028  */
1029 int
1030 if_getgroupmembers(caddr_t data)
1031 {
1032 	struct ifgroupreq	*ifgr = (struct ifgroupreq *)data;
1033 	struct ifg_group	*ifg;
1034 	struct ifg_member	*ifgm;
1035 	struct ifg_req		 ifgrq, *ifgp;
1036 	int			 len, error;
1037 
1038 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1039 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1040 			break;
1041 	if (ifg == NULL)
1042 		return (ENOENT);
1043 
1044 	if (ifgr->ifgr_len == 0) {
1045 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1046 			ifgr->ifgr_len += sizeof(ifgrq);
1047 		return (0);
1048 	}
1049 
1050 	len = ifgr->ifgr_len;
1051 	ifgp = ifgr->ifgr_groups;
1052 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1053 		if (len < sizeof(ifgrq))
1054 			return (EINVAL);
1055 		bzero(&ifgrq, sizeof ifgrq);
1056 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1057 		    sizeof(ifgrq.ifgrq_member));
1058 		if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1059 		    sizeof(struct ifg_req))))
1060 			return (error);
1061 		len -= sizeof(ifgrq);
1062 		ifgp++;
1063 	}
1064 
1065 	return (0);
1066 }
1067 
1068 /*
1069  * Delete Routes for a Network Interface
1070  *
1071  * Called for each routing entry via the rnh->rnh_walktree() call above
1072  * to delete all route entries referencing a detaching network interface.
1073  *
1074  * Arguments:
1075  *	rn	pointer to node in the routing table
1076  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1077  *
1078  * Returns:
1079  *	0	successful
1080  *	errno	failed - reason indicated
1081  *
1082  */
1083 static int
1084 if_rtdel(struct radix_node *rn, void *arg)
1085 {
1086 	struct rtentry	*rt = (struct rtentry *)rn;
1087 	struct ifnet	*ifp = arg;
1088 	int		err;
1089 
1090 	if (rt->rt_ifp == ifp) {
1091 
1092 		/*
1093 		 * Protect (sorta) against walktree recursion problems
1094 		 * with cloned routes
1095 		 */
1096 		if (!(rt->rt_flags & RTF_UP))
1097 			return (0);
1098 
1099 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1100 				rt_mask(rt), rt->rt_flags,
1101 				NULL);
1102 		if (err) {
1103 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1104 		}
1105 	}
1106 
1107 	return (0);
1108 }
1109 
1110 /*
1111  * Locate an interface based on a complete address.
1112  */
1113 struct ifaddr *
1114 ifa_ifwithaddr(struct sockaddr *addr)
1115 {
1116 	struct ifnet *ifp;
1117 
1118 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1119 		struct ifaddr_container *ifac;
1120 
1121 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1122 			struct ifaddr *ifa = ifac->ifa;
1123 
1124 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1125 				continue;
1126 			if (sa_equal(addr, ifa->ifa_addr))
1127 				return (ifa);
1128 			if ((ifp->if_flags & IFF_BROADCAST) &&
1129 			    ifa->ifa_broadaddr &&
1130 			    /* IPv6 doesn't have broadcast */
1131 			    ifa->ifa_broadaddr->sa_len != 0 &&
1132 			    sa_equal(ifa->ifa_broadaddr, addr))
1133 				return (ifa);
1134 		}
1135 	}
1136 	return (NULL);
1137 }
1138 /*
1139  * Locate the point to point interface with a given destination address.
1140  */
1141 struct ifaddr *
1142 ifa_ifwithdstaddr(struct sockaddr *addr)
1143 {
1144 	struct ifnet *ifp;
1145 
1146 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1147 		struct ifaddr_container *ifac;
1148 
1149 		if (!(ifp->if_flags & IFF_POINTOPOINT))
1150 			continue;
1151 
1152 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1153 			struct ifaddr *ifa = ifac->ifa;
1154 
1155 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1156 				continue;
1157 			if (ifa->ifa_dstaddr &&
1158 			    sa_equal(addr, ifa->ifa_dstaddr))
1159 				return (ifa);
1160 		}
1161 	}
1162 	return (NULL);
1163 }
1164 
1165 /*
1166  * Find an interface on a specific network.  If many, choice
1167  * is most specific found.
1168  */
1169 struct ifaddr *
1170 ifa_ifwithnet(struct sockaddr *addr)
1171 {
1172 	struct ifnet *ifp;
1173 	struct ifaddr *ifa_maybe = NULL;
1174 	u_int af = addr->sa_family;
1175 	char *addr_data = addr->sa_data, *cplim;
1176 
1177 	/*
1178 	 * AF_LINK addresses can be looked up directly by their index number,
1179 	 * so do that if we can.
1180 	 */
1181 	if (af == AF_LINK) {
1182 		struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1183 
1184 		if (sdl->sdl_index && sdl->sdl_index <= if_index)
1185 			return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1186 	}
1187 
1188 	/*
1189 	 * Scan though each interface, looking for ones that have
1190 	 * addresses in this address family.
1191 	 */
1192 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1193 		struct ifaddr_container *ifac;
1194 
1195 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1196 			struct ifaddr *ifa = ifac->ifa;
1197 			char *cp, *cp2, *cp3;
1198 
1199 			if (ifa->ifa_addr->sa_family != af)
1200 next:				continue;
1201 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1202 				/*
1203 				 * This is a bit broken as it doesn't
1204 				 * take into account that the remote end may
1205 				 * be a single node in the network we are
1206 				 * looking for.
1207 				 * The trouble is that we don't know the
1208 				 * netmask for the remote end.
1209 				 */
1210 				if (ifa->ifa_dstaddr != NULL &&
1211 				    sa_equal(addr, ifa->ifa_dstaddr))
1212 					return (ifa);
1213 			} else {
1214 				/*
1215 				 * if we have a special address handler,
1216 				 * then use it instead of the generic one.
1217 				 */
1218 				if (ifa->ifa_claim_addr) {
1219 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1220 						return (ifa);
1221 					} else {
1222 						continue;
1223 					}
1224 				}
1225 
1226 				/*
1227 				 * Scan all the bits in the ifa's address.
1228 				 * If a bit dissagrees with what we are
1229 				 * looking for, mask it with the netmask
1230 				 * to see if it really matters.
1231 				 * (A byte at a time)
1232 				 */
1233 				if (ifa->ifa_netmask == 0)
1234 					continue;
1235 				cp = addr_data;
1236 				cp2 = ifa->ifa_addr->sa_data;
1237 				cp3 = ifa->ifa_netmask->sa_data;
1238 				cplim = ifa->ifa_netmask->sa_len +
1239 					(char *)ifa->ifa_netmask;
1240 				while (cp3 < cplim)
1241 					if ((*cp++ ^ *cp2++) & *cp3++)
1242 						goto next; /* next address! */
1243 				/*
1244 				 * If the netmask of what we just found
1245 				 * is more specific than what we had before
1246 				 * (if we had one) then remember the new one
1247 				 * before continuing to search
1248 				 * for an even better one.
1249 				 */
1250 				if (ifa_maybe == NULL ||
1251 				    rn_refines((char *)ifa->ifa_netmask,
1252 					       (char *)ifa_maybe->ifa_netmask))
1253 					ifa_maybe = ifa;
1254 			}
1255 		}
1256 	}
1257 	return (ifa_maybe);
1258 }
1259 
1260 /*
1261  * Find an interface address specific to an interface best matching
1262  * a given address.
1263  */
1264 struct ifaddr *
1265 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1266 {
1267 	struct ifaddr_container *ifac;
1268 	char *cp, *cp2, *cp3;
1269 	char *cplim;
1270 	struct ifaddr *ifa_maybe = NULL;
1271 	u_int af = addr->sa_family;
1272 
1273 	if (af >= AF_MAX)
1274 		return (0);
1275 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1276 		struct ifaddr *ifa = ifac->ifa;
1277 
1278 		if (ifa->ifa_addr->sa_family != af)
1279 			continue;
1280 		if (ifa_maybe == NULL)
1281 			ifa_maybe = ifa;
1282 		if (ifa->ifa_netmask == NULL) {
1283 			if (sa_equal(addr, ifa->ifa_addr) ||
1284 			    (ifa->ifa_dstaddr != NULL &&
1285 			     sa_equal(addr, ifa->ifa_dstaddr)))
1286 				return (ifa);
1287 			continue;
1288 		}
1289 		if (ifp->if_flags & IFF_POINTOPOINT) {
1290 			if (sa_equal(addr, ifa->ifa_dstaddr))
1291 				return (ifa);
1292 		} else {
1293 			cp = addr->sa_data;
1294 			cp2 = ifa->ifa_addr->sa_data;
1295 			cp3 = ifa->ifa_netmask->sa_data;
1296 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1297 			for (; cp3 < cplim; cp3++)
1298 				if ((*cp++ ^ *cp2++) & *cp3)
1299 					break;
1300 			if (cp3 == cplim)
1301 				return (ifa);
1302 		}
1303 	}
1304 	return (ifa_maybe);
1305 }
1306 
1307 /*
1308  * Default action when installing a route with a Link Level gateway.
1309  * Lookup an appropriate real ifa to point to.
1310  * This should be moved to /sys/net/link.c eventually.
1311  */
1312 static void
1313 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1314 {
1315 	struct ifaddr *ifa;
1316 	struct sockaddr *dst;
1317 	struct ifnet *ifp;
1318 
1319 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1320 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1321 		return;
1322 	ifa = ifaof_ifpforaddr(dst, ifp);
1323 	if (ifa != NULL) {
1324 		IFAFREE(rt->rt_ifa);
1325 		IFAREF(ifa);
1326 		rt->rt_ifa = ifa;
1327 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1328 			ifa->ifa_rtrequest(cmd, rt, info);
1329 	}
1330 }
1331 
1332 /*
1333  * Mark an interface down and notify protocols of
1334  * the transition.
1335  * NOTE: must be called at splnet or eqivalent.
1336  */
1337 void
1338 if_unroute(struct ifnet *ifp, int flag, int fam)
1339 {
1340 	struct ifaddr_container *ifac;
1341 
1342 	ifp->if_flags &= ~flag;
1343 	getmicrotime(&ifp->if_lastchange);
1344 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1345 		struct ifaddr *ifa = ifac->ifa;
1346 
1347 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1348 			kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1349 	}
1350 	ifq_purge_all(&ifp->if_snd);
1351 	rt_ifmsg(ifp);
1352 }
1353 
1354 /*
1355  * Mark an interface up and notify protocols of
1356  * the transition.
1357  * NOTE: must be called at splnet or eqivalent.
1358  */
1359 void
1360 if_route(struct ifnet *ifp, int flag, int fam)
1361 {
1362 	struct ifaddr_container *ifac;
1363 
1364 	ifq_purge_all(&ifp->if_snd);
1365 	ifp->if_flags |= flag;
1366 	getmicrotime(&ifp->if_lastchange);
1367 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1368 		struct ifaddr *ifa = ifac->ifa;
1369 
1370 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1371 			kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1372 	}
1373 	rt_ifmsg(ifp);
1374 #ifdef INET6
1375 	in6_if_up(ifp);
1376 #endif
1377 }
1378 
1379 /*
1380  * Mark an interface down and notify protocols of the transition.  An
1381  * interface going down is also considered to be a synchronizing event.
1382  * We must ensure that all packet processing related to the interface
1383  * has completed before we return so e.g. the caller can free the ifnet
1384  * structure that the mbufs may be referencing.
1385  *
1386  * NOTE: must be called at splnet or eqivalent.
1387  */
1388 void
1389 if_down(struct ifnet *ifp)
1390 {
1391 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1392 	netmsg_service_sync();
1393 }
1394 
1395 /*
1396  * Mark an interface up and notify protocols of
1397  * the transition.
1398  * NOTE: must be called at splnet or eqivalent.
1399  */
1400 void
1401 if_up(struct ifnet *ifp)
1402 {
1403 	if_route(ifp, IFF_UP, AF_UNSPEC);
1404 }
1405 
1406 /*
1407  * Process a link state change.
1408  * NOTE: must be called at splsoftnet or equivalent.
1409  */
1410 void
1411 if_link_state_change(struct ifnet *ifp)
1412 {
1413 	int link_state = ifp->if_link_state;
1414 
1415 	rt_ifmsg(ifp);
1416 	devctl_notify("IFNET", ifp->if_xname,
1417 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1418 }
1419 
1420 /*
1421  * Handle interface watchdog timer routines.  Called
1422  * from softclock, we decrement timers (if set) and
1423  * call the appropriate interface routine on expiration.
1424  */
1425 static void
1426 if_slowtimo(void *arg)
1427 {
1428 	struct ifnet *ifp;
1429 
1430 	crit_enter();
1431 
1432 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1433 		if (ifp->if_timer == 0 || --ifp->if_timer)
1434 			continue;
1435 		if (ifp->if_watchdog) {
1436 			if (ifnet_tryserialize_all(ifp)) {
1437 				(*ifp->if_watchdog)(ifp);
1438 				ifnet_deserialize_all(ifp);
1439 			} else {
1440 				/* try again next timeout */
1441 				++ifp->if_timer;
1442 			}
1443 		}
1444 	}
1445 
1446 	crit_exit();
1447 
1448 	callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1449 }
1450 
1451 /*
1452  * Map interface name to
1453  * interface structure pointer.
1454  */
1455 struct ifnet *
1456 ifunit(const char *name)
1457 {
1458 	struct ifnet *ifp;
1459 
1460 	/*
1461 	 * Search all the interfaces for this name/number
1462 	 */
1463 
1464 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1465 		if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1466 			break;
1467 	}
1468 	return (ifp);
1469 }
1470 
1471 
1472 /*
1473  * Map interface name in a sockaddr_dl to
1474  * interface structure pointer.
1475  */
1476 struct ifnet *
1477 if_withname(struct sockaddr *sa)
1478 {
1479 	char ifname[IFNAMSIZ+1];
1480 	struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1481 
1482 	if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1483 	     (sdl->sdl_nlen > IFNAMSIZ) )
1484 		return NULL;
1485 
1486 	/*
1487 	 * ifunit wants a null-terminated name.  It may not be null-terminated
1488 	 * in the sockaddr.  We don't want to change the caller's sockaddr,
1489 	 * and there might not be room to put the trailing null anyway, so we
1490 	 * make a local copy that we know we can null terminate safely.
1491 	 */
1492 
1493 	bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1494 	ifname[sdl->sdl_nlen] = '\0';
1495 	return ifunit(ifname);
1496 }
1497 
1498 
1499 /*
1500  * Interface ioctls.
1501  */
1502 int
1503 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1504 {
1505 	struct ifnet *ifp;
1506 	struct ifreq *ifr;
1507 	struct ifstat *ifs;
1508 	int error;
1509 	short oif_flags;
1510 	int new_flags;
1511 #ifdef COMPAT_43
1512 	int ocmd;
1513 #endif
1514 	size_t namelen, onamelen;
1515 	char new_name[IFNAMSIZ];
1516 	struct ifaddr *ifa;
1517 	struct sockaddr_dl *sdl;
1518 
1519 	switch (cmd) {
1520 	case SIOCGIFCONF:
1521 	case OSIOCGIFCONF:
1522 		return (ifconf(cmd, data, cred));
1523 	default:
1524 		break;
1525 	}
1526 
1527 	ifr = (struct ifreq *)data;
1528 
1529 	switch (cmd) {
1530 	case SIOCIFCREATE:
1531 	case SIOCIFCREATE2:
1532 		if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1533 			return (error);
1534 		return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1535 		    	cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1536 	case SIOCIFDESTROY:
1537 		if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1538 			return (error);
1539 		return (if_clone_destroy(ifr->ifr_name));
1540 	case SIOCIFGCLONERS:
1541 		return (if_clone_list((struct if_clonereq *)data));
1542 	default:
1543 		break;
1544 	}
1545 
1546 	/*
1547 	 * Nominal ioctl through interface, lookup the ifp and obtain a
1548 	 * lock to serialize the ifconfig ioctl operation.
1549 	 */
1550 	ifp = ifunit(ifr->ifr_name);
1551 	if (ifp == NULL)
1552 		return (ENXIO);
1553 	error = 0;
1554 	mtx_lock(&ifp->if_ioctl_mtx);
1555 
1556 	switch (cmd) {
1557 	case SIOCGIFINDEX:
1558 		ifr->ifr_index = ifp->if_index;
1559 		break;
1560 
1561 	case SIOCGIFFLAGS:
1562 		ifr->ifr_flags = ifp->if_flags;
1563 		ifr->ifr_flagshigh = ifp->if_flags >> 16;
1564 		break;
1565 
1566 	case SIOCGIFCAP:
1567 		ifr->ifr_reqcap = ifp->if_capabilities;
1568 		ifr->ifr_curcap = ifp->if_capenable;
1569 		break;
1570 
1571 	case SIOCGIFMETRIC:
1572 		ifr->ifr_metric = ifp->if_metric;
1573 		break;
1574 
1575 	case SIOCGIFMTU:
1576 		ifr->ifr_mtu = ifp->if_mtu;
1577 		break;
1578 
1579 	case SIOCGIFTSOLEN:
1580 		ifr->ifr_tsolen = ifp->if_tsolen;
1581 		break;
1582 
1583 	case SIOCGIFDATA:
1584 		error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1585 				sizeof(ifp->if_data));
1586 		break;
1587 
1588 	case SIOCGIFPHYS:
1589 		ifr->ifr_phys = ifp->if_physical;
1590 		break;
1591 
1592 	case SIOCGIFPOLLCPU:
1593 		ifr->ifr_pollcpu = -1;
1594 		break;
1595 
1596 	case SIOCSIFPOLLCPU:
1597 		break;
1598 
1599 	case SIOCSIFFLAGS:
1600 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1601 		if (error)
1602 			break;
1603 		new_flags = (ifr->ifr_flags & 0xffff) |
1604 		    (ifr->ifr_flagshigh << 16);
1605 		if (ifp->if_flags & IFF_SMART) {
1606 			/* Smart drivers twiddle their own routes */
1607 		} else if (ifp->if_flags & IFF_UP &&
1608 		    (new_flags & IFF_UP) == 0) {
1609 			crit_enter();
1610 			if_down(ifp);
1611 			crit_exit();
1612 		} else if (new_flags & IFF_UP &&
1613 		    (ifp->if_flags & IFF_UP) == 0) {
1614 			crit_enter();
1615 			if_up(ifp);
1616 			crit_exit();
1617 		}
1618 
1619 #ifdef IFPOLL_ENABLE
1620 		if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1621 			if (new_flags & IFF_NPOLLING)
1622 				ifpoll_register(ifp);
1623 			else
1624 				ifpoll_deregister(ifp);
1625 		}
1626 #endif
1627 
1628 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1629 			(new_flags &~ IFF_CANTCHANGE);
1630 		if (new_flags & IFF_PPROMISC) {
1631 			/* Permanently promiscuous mode requested */
1632 			ifp->if_flags |= IFF_PROMISC;
1633 		} else if (ifp->if_pcount == 0) {
1634 			ifp->if_flags &= ~IFF_PROMISC;
1635 		}
1636 		if (ifp->if_ioctl) {
1637 			ifnet_serialize_all(ifp);
1638 			ifp->if_ioctl(ifp, cmd, data, cred);
1639 			ifnet_deserialize_all(ifp);
1640 		}
1641 		getmicrotime(&ifp->if_lastchange);
1642 		break;
1643 
1644 	case SIOCSIFCAP:
1645 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1646 		if (error)
1647 			break;
1648 		if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1649 			error = EINVAL;
1650 			break;
1651 		}
1652 		ifnet_serialize_all(ifp);
1653 		ifp->if_ioctl(ifp, cmd, data, cred);
1654 		ifnet_deserialize_all(ifp);
1655 		break;
1656 
1657 	case SIOCSIFNAME:
1658 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1659 		if (error)
1660 			break;
1661 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1662 		if (error)
1663 			break;
1664 		if (new_name[0] == '\0') {
1665 			error = EINVAL;
1666 			break;
1667 		}
1668 		if (ifunit(new_name) != NULL) {
1669 			error = EEXIST;
1670 			break;
1671 		}
1672 
1673 		EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1674 
1675 		/* Announce the departure of the interface. */
1676 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1677 
1678 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1679 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1680 		/* XXX IFA_LOCK(ifa); */
1681 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1682 		namelen = strlen(new_name);
1683 		onamelen = sdl->sdl_nlen;
1684 		/*
1685 		 * Move the address if needed.  This is safe because we
1686 		 * allocate space for a name of length IFNAMSIZ when we
1687 		 * create this in if_attach().
1688 		 */
1689 		if (namelen != onamelen) {
1690 			bcopy(sdl->sdl_data + onamelen,
1691 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1692 		}
1693 		bcopy(new_name, sdl->sdl_data, namelen);
1694 		sdl->sdl_nlen = namelen;
1695 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1696 		bzero(sdl->sdl_data, onamelen);
1697 		while (namelen != 0)
1698 			sdl->sdl_data[--namelen] = 0xff;
1699 		/* XXX IFA_UNLOCK(ifa) */
1700 
1701 		EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1702 
1703 		/* Announce the return of the interface. */
1704 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1705 		break;
1706 
1707 	case SIOCSIFMETRIC:
1708 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1709 		if (error)
1710 			break;
1711 		ifp->if_metric = ifr->ifr_metric;
1712 		getmicrotime(&ifp->if_lastchange);
1713 		break;
1714 
1715 	case SIOCSIFPHYS:
1716 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1717 		if (error)
1718 			break;
1719 		if (ifp->if_ioctl == NULL) {
1720 		        error = EOPNOTSUPP;
1721 			break;
1722 		}
1723 		ifnet_serialize_all(ifp);
1724 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1725 		ifnet_deserialize_all(ifp);
1726 		if (error == 0)
1727 			getmicrotime(&ifp->if_lastchange);
1728 		break;
1729 
1730 	case SIOCSIFMTU:
1731 	{
1732 		u_long oldmtu = ifp->if_mtu;
1733 
1734 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1735 		if (error)
1736 			break;
1737 		if (ifp->if_ioctl == NULL) {
1738 			error = EOPNOTSUPP;
1739 			break;
1740 		}
1741 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
1742 			error = EINVAL;
1743 			break;
1744 		}
1745 		ifnet_serialize_all(ifp);
1746 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1747 		ifnet_deserialize_all(ifp);
1748 		if (error == 0) {
1749 			getmicrotime(&ifp->if_lastchange);
1750 			rt_ifmsg(ifp);
1751 		}
1752 		/*
1753 		 * If the link MTU changed, do network layer specific procedure.
1754 		 */
1755 		if (ifp->if_mtu != oldmtu) {
1756 #ifdef INET6
1757 			nd6_setmtu(ifp);
1758 #endif
1759 		}
1760 		break;
1761 	}
1762 
1763 	case SIOCSIFTSOLEN:
1764 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1765 		if (error)
1766 			break;
1767 
1768 		/* XXX need driver supplied upper limit */
1769 		if (ifr->ifr_tsolen <= 0) {
1770 			error = EINVAL;
1771 			break;
1772 		}
1773 		ifp->if_tsolen = ifr->ifr_tsolen;
1774 		break;
1775 
1776 	case SIOCADDMULTI:
1777 	case SIOCDELMULTI:
1778 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1779 		if (error)
1780 			break;
1781 
1782 		/* Don't allow group membership on non-multicast interfaces. */
1783 		if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1784 			error = EOPNOTSUPP;
1785 			break;
1786 		}
1787 
1788 		/* Don't let users screw up protocols' entries. */
1789 		if (ifr->ifr_addr.sa_family != AF_LINK) {
1790 			error = EINVAL;
1791 			break;
1792 		}
1793 
1794 		if (cmd == SIOCADDMULTI) {
1795 			struct ifmultiaddr *ifma;
1796 			error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1797 		} else {
1798 			error = if_delmulti(ifp, &ifr->ifr_addr);
1799 		}
1800 		if (error == 0)
1801 			getmicrotime(&ifp->if_lastchange);
1802 		break;
1803 
1804 	case SIOCSIFPHYADDR:
1805 	case SIOCDIFPHYADDR:
1806 #ifdef INET6
1807 	case SIOCSIFPHYADDR_IN6:
1808 #endif
1809 	case SIOCSLIFPHYADDR:
1810         case SIOCSIFMEDIA:
1811 	case SIOCSIFGENERIC:
1812 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1813 		if (error)
1814 			break;
1815 		if (ifp->if_ioctl == 0) {
1816 			error = EOPNOTSUPP;
1817 			break;
1818 		}
1819 		ifnet_serialize_all(ifp);
1820 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1821 		ifnet_deserialize_all(ifp);
1822 		if (error == 0)
1823 			getmicrotime(&ifp->if_lastchange);
1824 		break;
1825 
1826 	case SIOCGIFSTATUS:
1827 		ifs = (struct ifstat *)data;
1828 		ifs->ascii[0] = '\0';
1829 		/* fall through */
1830 	case SIOCGIFPSRCADDR:
1831 	case SIOCGIFPDSTADDR:
1832 	case SIOCGLIFPHYADDR:
1833 	case SIOCGIFMEDIA:
1834 	case SIOCGIFGENERIC:
1835 		if (ifp->if_ioctl == NULL) {
1836 			error = EOPNOTSUPP;
1837 			break;
1838 		}
1839 		ifnet_serialize_all(ifp);
1840 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1841 		ifnet_deserialize_all(ifp);
1842 		break;
1843 
1844 	case SIOCSIFLLADDR:
1845 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1846 		if (error)
1847 			break;
1848 		error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
1849 				     ifr->ifr_addr.sa_len);
1850 		EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1851 		break;
1852 
1853 	default:
1854 		oif_flags = ifp->if_flags;
1855 		if (so->so_proto == 0) {
1856 			error = EOPNOTSUPP;
1857 			break;
1858 		}
1859 #ifndef COMPAT_43
1860 		error = so_pru_control_direct(so, cmd, data, ifp);
1861 #else
1862 		ocmd = cmd;
1863 
1864 		switch (cmd) {
1865 		case SIOCSIFDSTADDR:
1866 		case SIOCSIFADDR:
1867 		case SIOCSIFBRDADDR:
1868 		case SIOCSIFNETMASK:
1869 #if BYTE_ORDER != BIG_ENDIAN
1870 			if (ifr->ifr_addr.sa_family == 0 &&
1871 			    ifr->ifr_addr.sa_len < 16) {
1872 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1873 				ifr->ifr_addr.sa_len = 16;
1874 			}
1875 #else
1876 			if (ifr->ifr_addr.sa_len == 0)
1877 				ifr->ifr_addr.sa_len = 16;
1878 #endif
1879 			break;
1880 		case OSIOCGIFADDR:
1881 			cmd = SIOCGIFADDR;
1882 			break;
1883 		case OSIOCGIFDSTADDR:
1884 			cmd = SIOCGIFDSTADDR;
1885 			break;
1886 		case OSIOCGIFBRDADDR:
1887 			cmd = SIOCGIFBRDADDR;
1888 			break;
1889 		case OSIOCGIFNETMASK:
1890 			cmd = SIOCGIFNETMASK;
1891 			break;
1892 		default:
1893 			break;
1894 		}
1895 
1896 		error = so_pru_control_direct(so, cmd, data, ifp);
1897 
1898 		switch (ocmd) {
1899 		case OSIOCGIFADDR:
1900 		case OSIOCGIFDSTADDR:
1901 		case OSIOCGIFBRDADDR:
1902 		case OSIOCGIFNETMASK:
1903 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1904 			break;
1905 		}
1906 #endif /* COMPAT_43 */
1907 
1908 		if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1909 #ifdef INET6
1910 			DELAY(100);/* XXX: temporary workaround for fxp issue*/
1911 			if (ifp->if_flags & IFF_UP) {
1912 				crit_enter();
1913 				in6_if_up(ifp);
1914 				crit_exit();
1915 			}
1916 #endif
1917 		}
1918 		break;
1919 	}
1920 
1921 	mtx_unlock(&ifp->if_ioctl_mtx);
1922 	return (error);
1923 }
1924 
1925 /*
1926  * Set/clear promiscuous mode on interface ifp based on the truth value
1927  * of pswitch.  The calls are reference counted so that only the first
1928  * "on" request actually has an effect, as does the final "off" request.
1929  * Results are undefined if the "off" and "on" requests are not matched.
1930  */
1931 int
1932 ifpromisc(struct ifnet *ifp, int pswitch)
1933 {
1934 	struct ifreq ifr;
1935 	int error;
1936 	int oldflags;
1937 
1938 	oldflags = ifp->if_flags;
1939 	if (ifp->if_flags & IFF_PPROMISC) {
1940 		/* Do nothing if device is in permanently promiscuous mode */
1941 		ifp->if_pcount += pswitch ? 1 : -1;
1942 		return (0);
1943 	}
1944 	if (pswitch) {
1945 		/*
1946 		 * If the device is not configured up, we cannot put it in
1947 		 * promiscuous mode.
1948 		 */
1949 		if ((ifp->if_flags & IFF_UP) == 0)
1950 			return (ENETDOWN);
1951 		if (ifp->if_pcount++ != 0)
1952 			return (0);
1953 		ifp->if_flags |= IFF_PROMISC;
1954 		log(LOG_INFO, "%s: promiscuous mode enabled\n",
1955 		    ifp->if_xname);
1956 	} else {
1957 		if (--ifp->if_pcount > 0)
1958 			return (0);
1959 		ifp->if_flags &= ~IFF_PROMISC;
1960 		log(LOG_INFO, "%s: promiscuous mode disabled\n",
1961 		    ifp->if_xname);
1962 	}
1963 	ifr.ifr_flags = ifp->if_flags;
1964 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
1965 	ifnet_serialize_all(ifp);
1966 	error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
1967 	ifnet_deserialize_all(ifp);
1968 	if (error == 0)
1969 		rt_ifmsg(ifp);
1970 	else
1971 		ifp->if_flags = oldflags;
1972 	return error;
1973 }
1974 
1975 /*
1976  * Return interface configuration
1977  * of system.  List may be used
1978  * in later ioctl's (above) to get
1979  * other information.
1980  */
1981 static int
1982 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
1983 {
1984 	struct ifconf *ifc = (struct ifconf *)data;
1985 	struct ifnet *ifp;
1986 	struct sockaddr *sa;
1987 	struct ifreq ifr, *ifrp;
1988 	int space = ifc->ifc_len, error = 0;
1989 
1990 	ifrp = ifc->ifc_req;
1991 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1992 		struct ifaddr_container *ifac;
1993 		int addrs;
1994 
1995 		if (space <= sizeof ifr)
1996 			break;
1997 
1998 		/*
1999 		 * Zero the stack declared structure first to prevent
2000 		 * memory disclosure.
2001 		 */
2002 		bzero(&ifr, sizeof(ifr));
2003 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2004 		    >= sizeof(ifr.ifr_name)) {
2005 			error = ENAMETOOLONG;
2006 			break;
2007 		}
2008 
2009 		addrs = 0;
2010 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2011 			struct ifaddr *ifa = ifac->ifa;
2012 
2013 			if (space <= sizeof ifr)
2014 				break;
2015 			sa = ifa->ifa_addr;
2016 			if (cred->cr_prison &&
2017 			    prison_if(cred, sa))
2018 				continue;
2019 			addrs++;
2020 #ifdef COMPAT_43
2021 			if (cmd == OSIOCGIFCONF) {
2022 				struct osockaddr *osa =
2023 					 (struct osockaddr *)&ifr.ifr_addr;
2024 				ifr.ifr_addr = *sa;
2025 				osa->sa_family = sa->sa_family;
2026 				error = copyout(&ifr, ifrp, sizeof ifr);
2027 				ifrp++;
2028 			} else
2029 #endif
2030 			if (sa->sa_len <= sizeof(*sa)) {
2031 				ifr.ifr_addr = *sa;
2032 				error = copyout(&ifr, ifrp, sizeof ifr);
2033 				ifrp++;
2034 			} else {
2035 				if (space < (sizeof ifr) + sa->sa_len -
2036 					    sizeof(*sa))
2037 					break;
2038 				space -= sa->sa_len - sizeof(*sa);
2039 				error = copyout(&ifr, ifrp,
2040 						sizeof ifr.ifr_name);
2041 				if (error == 0)
2042 					error = copyout(sa, &ifrp->ifr_addr,
2043 							sa->sa_len);
2044 				ifrp = (struct ifreq *)
2045 					(sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2046 			}
2047 			if (error)
2048 				break;
2049 			space -= sizeof ifr;
2050 		}
2051 		if (error)
2052 			break;
2053 		if (!addrs) {
2054 			bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2055 			error = copyout(&ifr, ifrp, sizeof ifr);
2056 			if (error)
2057 				break;
2058 			space -= sizeof ifr;
2059 			ifrp++;
2060 		}
2061 	}
2062 	ifc->ifc_len -= space;
2063 	return (error);
2064 }
2065 
2066 /*
2067  * Just like if_promisc(), but for all-multicast-reception mode.
2068  */
2069 int
2070 if_allmulti(struct ifnet *ifp, int onswitch)
2071 {
2072 	int error = 0;
2073 	struct ifreq ifr;
2074 
2075 	crit_enter();
2076 
2077 	if (onswitch) {
2078 		if (ifp->if_amcount++ == 0) {
2079 			ifp->if_flags |= IFF_ALLMULTI;
2080 			ifr.ifr_flags = ifp->if_flags;
2081 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2082 			ifnet_serialize_all(ifp);
2083 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2084 					      NULL);
2085 			ifnet_deserialize_all(ifp);
2086 		}
2087 	} else {
2088 		if (ifp->if_amcount > 1) {
2089 			ifp->if_amcount--;
2090 		} else {
2091 			ifp->if_amcount = 0;
2092 			ifp->if_flags &= ~IFF_ALLMULTI;
2093 			ifr.ifr_flags = ifp->if_flags;
2094 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2095 			ifnet_serialize_all(ifp);
2096 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2097 					      NULL);
2098 			ifnet_deserialize_all(ifp);
2099 		}
2100 	}
2101 
2102 	crit_exit();
2103 
2104 	if (error == 0)
2105 		rt_ifmsg(ifp);
2106 	return error;
2107 }
2108 
2109 /*
2110  * Add a multicast listenership to the interface in question.
2111  * The link layer provides a routine which converts
2112  */
2113 int
2114 if_addmulti(
2115 	struct ifnet *ifp,	/* interface to manipulate */
2116 	struct sockaddr *sa,	/* address to add */
2117 	struct ifmultiaddr **retifma)
2118 {
2119 	struct sockaddr *llsa, *dupsa;
2120 	int error;
2121 	struct ifmultiaddr *ifma;
2122 
2123 	/*
2124 	 * If the matching multicast address already exists
2125 	 * then don't add a new one, just add a reference
2126 	 */
2127 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2128 		if (sa_equal(sa, ifma->ifma_addr)) {
2129 			ifma->ifma_refcount++;
2130 			if (retifma)
2131 				*retifma = ifma;
2132 			return 0;
2133 		}
2134 	}
2135 
2136 	/*
2137 	 * Give the link layer a chance to accept/reject it, and also
2138 	 * find out which AF_LINK address this maps to, if it isn't one
2139 	 * already.
2140 	 */
2141 	if (ifp->if_resolvemulti) {
2142 		ifnet_serialize_all(ifp);
2143 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2144 		ifnet_deserialize_all(ifp);
2145 		if (error)
2146 			return error;
2147 	} else {
2148 		llsa = NULL;
2149 	}
2150 
2151 	ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2152 	dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_WAITOK);
2153 	bcopy(sa, dupsa, sa->sa_len);
2154 
2155 	ifma->ifma_addr = dupsa;
2156 	ifma->ifma_lladdr = llsa;
2157 	ifma->ifma_ifp = ifp;
2158 	ifma->ifma_refcount = 1;
2159 	ifma->ifma_protospec = 0;
2160 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2161 
2162 	/*
2163 	 * Some network interfaces can scan the address list at
2164 	 * interrupt time; lock them out.
2165 	 */
2166 	crit_enter();
2167 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2168 	crit_exit();
2169 	if (retifma)
2170 		*retifma = ifma;
2171 
2172 	if (llsa != NULL) {
2173 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2174 			if (sa_equal(ifma->ifma_addr, llsa))
2175 				break;
2176 		}
2177 		if (ifma) {
2178 			ifma->ifma_refcount++;
2179 		} else {
2180 			ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2181 			dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_WAITOK);
2182 			bcopy(llsa, dupsa, llsa->sa_len);
2183 			ifma->ifma_addr = dupsa;
2184 			ifma->ifma_ifp = ifp;
2185 			ifma->ifma_refcount = 1;
2186 			crit_enter();
2187 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2188 			crit_exit();
2189 		}
2190 	}
2191 	/*
2192 	 * We are certain we have added something, so call down to the
2193 	 * interface to let them know about it.
2194 	 */
2195 	crit_enter();
2196 	ifnet_serialize_all(ifp);
2197 	if (ifp->if_ioctl)
2198 		ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2199 	ifnet_deserialize_all(ifp);
2200 	crit_exit();
2201 
2202 	return 0;
2203 }
2204 
2205 /*
2206  * Remove a reference to a multicast address on this interface.  Yell
2207  * if the request does not match an existing membership.
2208  */
2209 int
2210 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2211 {
2212 	struct ifmultiaddr *ifma;
2213 
2214 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2215 		if (sa_equal(sa, ifma->ifma_addr))
2216 			break;
2217 	if (ifma == NULL)
2218 		return ENOENT;
2219 
2220 	if (ifma->ifma_refcount > 1) {
2221 		ifma->ifma_refcount--;
2222 		return 0;
2223 	}
2224 
2225 	rt_newmaddrmsg(RTM_DELMADDR, ifma);
2226 	sa = ifma->ifma_lladdr;
2227 	crit_enter();
2228 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2229 	/*
2230 	 * Make sure the interface driver is notified
2231 	 * in the case of a link layer mcast group being left.
2232 	 */
2233 	if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL) {
2234 		ifnet_serialize_all(ifp);
2235 		ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2236 		ifnet_deserialize_all(ifp);
2237 	}
2238 	crit_exit();
2239 	kfree(ifma->ifma_addr, M_IFMADDR);
2240 	kfree(ifma, M_IFMADDR);
2241 	if (sa == NULL)
2242 		return 0;
2243 
2244 	/*
2245 	 * Now look for the link-layer address which corresponds to
2246 	 * this network address.  It had been squirreled away in
2247 	 * ifma->ifma_lladdr for this purpose (so we don't have
2248 	 * to call ifp->if_resolvemulti() again), and we saved that
2249 	 * value in sa above.  If some nasty deleted the
2250 	 * link-layer address out from underneath us, we can deal because
2251 	 * the address we stored was is not the same as the one which was
2252 	 * in the record for the link-layer address.  (So we don't complain
2253 	 * in that case.)
2254 	 */
2255 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2256 		if (sa_equal(sa, ifma->ifma_addr))
2257 			break;
2258 	if (ifma == NULL)
2259 		return 0;
2260 
2261 	if (ifma->ifma_refcount > 1) {
2262 		ifma->ifma_refcount--;
2263 		return 0;
2264 	}
2265 
2266 	crit_enter();
2267 	ifnet_serialize_all(ifp);
2268 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2269 	ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2270 	ifnet_deserialize_all(ifp);
2271 	crit_exit();
2272 	kfree(ifma->ifma_addr, M_IFMADDR);
2273 	kfree(sa, M_IFMADDR);
2274 	kfree(ifma, M_IFMADDR);
2275 
2276 	return 0;
2277 }
2278 
2279 /*
2280  * Delete all multicast group membership for an interface.
2281  * Should be used to quickly flush all multicast filters.
2282  */
2283 void
2284 if_delallmulti(struct ifnet *ifp)
2285 {
2286 	struct ifmultiaddr *ifma;
2287 	struct ifmultiaddr *next;
2288 
2289 	TAILQ_FOREACH_MUTABLE(ifma, &ifp->if_multiaddrs, ifma_link, next)
2290 		if_delmulti(ifp, ifma->ifma_addr);
2291 }
2292 
2293 
2294 /*
2295  * Set the link layer address on an interface.
2296  *
2297  * At this time we only support certain types of interfaces,
2298  * and we don't allow the length of the address to change.
2299  */
2300 int
2301 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2302 {
2303 	struct sockaddr_dl *sdl;
2304 	struct ifreq ifr;
2305 
2306 	sdl = IF_LLSOCKADDR(ifp);
2307 	if (sdl == NULL)
2308 		return (EINVAL);
2309 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2310 		return (EINVAL);
2311 	switch (ifp->if_type) {
2312 	case IFT_ETHER:			/* these types use struct arpcom */
2313 	case IFT_XETHER:
2314 	case IFT_L2VLAN:
2315 		bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2316 		bcopy(lladdr, LLADDR(sdl), len);
2317 		break;
2318 	default:
2319 		return (ENODEV);
2320 	}
2321 	/*
2322 	 * If the interface is already up, we need
2323 	 * to re-init it in order to reprogram its
2324 	 * address filter.
2325 	 */
2326 	ifnet_serialize_all(ifp);
2327 	if ((ifp->if_flags & IFF_UP) != 0) {
2328 #ifdef INET
2329 		struct ifaddr_container *ifac;
2330 #endif
2331 
2332 		ifp->if_flags &= ~IFF_UP;
2333 		ifr.ifr_flags = ifp->if_flags;
2334 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
2335 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2336 			      NULL);
2337 		ifp->if_flags |= IFF_UP;
2338 		ifr.ifr_flags = ifp->if_flags;
2339 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
2340 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2341 				 NULL);
2342 #ifdef INET
2343 		/*
2344 		 * Also send gratuitous ARPs to notify other nodes about
2345 		 * the address change.
2346 		 */
2347 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2348 			struct ifaddr *ifa = ifac->ifa;
2349 
2350 			if (ifa->ifa_addr != NULL &&
2351 			    ifa->ifa_addr->sa_family == AF_INET)
2352 				arp_gratuitous(ifp, ifa);
2353 		}
2354 #endif
2355 	}
2356 	ifnet_deserialize_all(ifp);
2357 	return (0);
2358 }
2359 
2360 struct ifmultiaddr *
2361 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2362 {
2363 	struct ifmultiaddr *ifma;
2364 
2365 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2366 		if (sa_equal(ifma->ifma_addr, sa))
2367 			break;
2368 
2369 	return ifma;
2370 }
2371 
2372 /*
2373  * This function locates the first real ethernet MAC from a network
2374  * card and loads it into node, returning 0 on success or ENOENT if
2375  * no suitable interfaces were found.  It is used by the uuid code to
2376  * generate a unique 6-byte number.
2377  */
2378 int
2379 if_getanyethermac(uint16_t *node, int minlen)
2380 {
2381 	struct ifnet *ifp;
2382 	struct sockaddr_dl *sdl;
2383 
2384 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
2385 		if (ifp->if_type != IFT_ETHER)
2386 			continue;
2387 		sdl = IF_LLSOCKADDR(ifp);
2388 		if (sdl->sdl_alen < minlen)
2389 			continue;
2390 		bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2391 		      minlen);
2392 		return(0);
2393 	}
2394 	return (ENOENT);
2395 }
2396 
2397 /*
2398  * The name argument must be a pointer to storage which will last as
2399  * long as the interface does.  For physical devices, the result of
2400  * device_get_name(dev) is a good choice and for pseudo-devices a
2401  * static string works well.
2402  */
2403 void
2404 if_initname(struct ifnet *ifp, const char *name, int unit)
2405 {
2406 	ifp->if_dname = name;
2407 	ifp->if_dunit = unit;
2408 	if (unit != IF_DUNIT_NONE)
2409 		ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2410 	else
2411 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2412 }
2413 
2414 int
2415 if_printf(struct ifnet *ifp, const char *fmt, ...)
2416 {
2417 	__va_list ap;
2418 	int retval;
2419 
2420 	retval = kprintf("%s: ", ifp->if_xname);
2421 	__va_start(ap, fmt);
2422 	retval += kvprintf(fmt, ap);
2423 	__va_end(ap);
2424 	return (retval);
2425 }
2426 
2427 struct ifnet *
2428 if_alloc(uint8_t type)
2429 {
2430         struct ifnet *ifp;
2431 	size_t size;
2432 
2433 	/*
2434 	 * XXX temporary hack until arpcom is setup in if_l2com
2435 	 */
2436 	if (type == IFT_ETHER)
2437 		size = sizeof(struct arpcom);
2438 	else
2439 		size = sizeof(struct ifnet);
2440 
2441 	ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2442 
2443 	ifp->if_type = type;
2444 
2445 	if (if_com_alloc[type] != NULL) {
2446 		ifp->if_l2com = if_com_alloc[type](type, ifp);
2447 		if (ifp->if_l2com == NULL) {
2448 			kfree(ifp, M_IFNET);
2449 			return (NULL);
2450 		}
2451 	}
2452 	return (ifp);
2453 }
2454 
2455 void
2456 if_free(struct ifnet *ifp)
2457 {
2458 	kfree(ifp, M_IFNET);
2459 }
2460 
2461 void
2462 ifq_set_classic(struct ifaltq *ifq)
2463 {
2464 	ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2465 	    ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2466 }
2467 
2468 void
2469 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2470     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2471 {
2472 	int q;
2473 
2474 	KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2475 	KASSERT(enqueue != NULL, ("enqueue is not specified"));
2476 	KASSERT(dequeue != NULL, ("dequeue is not specified"));
2477 	KASSERT(request != NULL, ("request is not specified"));
2478 
2479 	ifq->altq_mapsubq = mapsubq;
2480 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2481 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2482 
2483 		ifsq->ifsq_enqueue = enqueue;
2484 		ifsq->ifsq_dequeue = dequeue;
2485 		ifsq->ifsq_request = request;
2486 	}
2487 }
2488 
2489 int
2490 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2491     struct altq_pktattr *pa __unused)
2492 {
2493 	if (IF_QFULL(ifsq)) {
2494 		m_freem(m);
2495 		return(ENOBUFS);
2496 	} else {
2497 		IF_ENQUEUE(ifsq, m);
2498 		return(0);
2499 	}
2500 }
2501 
2502 struct mbuf *
2503 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, struct mbuf *mpolled, int op)
2504 {
2505 	struct mbuf *m;
2506 
2507 	switch (op) {
2508 	case ALTDQ_POLL:
2509 		IF_POLL(ifsq, m);
2510 		break;
2511 	case ALTDQ_REMOVE:
2512 		IF_DEQUEUE(ifsq, m);
2513 		break;
2514 	default:
2515 		panic("unsupported ALTQ dequeue op: %d", op);
2516 	}
2517 	KKASSERT(mpolled == NULL || mpolled == m);
2518 	return(m);
2519 }
2520 
2521 int
2522 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2523 {
2524 	switch (req) {
2525 	case ALTRQ_PURGE:
2526 		IF_DRAIN(ifsq);
2527 		break;
2528 	default:
2529 		panic("unsupported ALTQ request: %d", req);
2530 	}
2531 	return(0);
2532 }
2533 
2534 static void
2535 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2536 {
2537 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
2538 	int running = 0, need_sched;
2539 
2540 	/*
2541 	 * Try to do direct ifnet.if_start first, if there is
2542 	 * contention on ifnet's serializer, ifnet.if_start will
2543 	 * be scheduled on ifnet's CPU.
2544 	 */
2545 	if (!ifnet_tryserialize_tx(ifp, ifsq)) {
2546 		/*
2547 		 * ifnet serializer contention happened,
2548 		 * ifnet.if_start is scheduled on ifnet's
2549 		 * CPU, and we keep going.
2550 		 */
2551 		ifsq_ifstart_schedule(ifsq, 1);
2552 		return;
2553 	}
2554 
2555 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2556 		ifp->if_start(ifp, ifsq);
2557 		if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2558 			running = 1;
2559 	}
2560 	need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2561 
2562 	ifnet_deserialize_tx(ifp, ifsq);
2563 
2564 	if (need_sched) {
2565 		/*
2566 		 * More data need to be transmitted, ifnet.if_start is
2567 		 * scheduled on ifnet's CPU, and we keep going.
2568 		 * NOTE: ifnet.if_start interlock is not released.
2569 		 */
2570 		ifsq_ifstart_schedule(ifsq, force_sched);
2571 	}
2572 }
2573 
2574 /*
2575  * IFSUBQ packets staging mechanism:
2576  *
2577  * The packets enqueued into IFSUBQ are staged to a certain amount before the
2578  * ifnet's if_start is called.  In this way, the driver could avoid writing
2579  * to hardware registers upon every packet, instead, hardware registers
2580  * could be written when certain amount of packets are put onto hardware
2581  * TX ring.  The measurement on several modern NICs (emx(4), igb(4), bnx(4),
2582  * bge(4), jme(4)) shows that the hardware registers writing aggregation
2583  * could save ~20% CPU time when 18bytes UDP datagrams are transmitted at
2584  * 1.48Mpps.  The performance improvement by hardware registers writing
2585  * aggeregation is also mentioned by Luigi Rizzo's netmap paper
2586  * (http://info.iet.unipi.it/~luigi/netmap/).
2587  *
2588  * IFSUBQ packets staging is performed for two entry points into drivers's
2589  * transmission function:
2590  * - Direct ifnet's if_start calling, i.e. ifsq_ifstart_try()
2591  * - ifnet's if_start scheduling, i.e. ifsq_ifstart_schedule()
2592  *
2593  * IFSUBQ packets staging will be stopped upon any of the following conditions:
2594  * - If the count of packets enqueued on the current CPU is great than or
2595  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2596  * - If the total length of packets enqueued on the current CPU is great
2597  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
2598  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
2599  *   is usually less than hardware's MTU.
2600  * - ifsq_ifstart_schedule() is not pending on the current CPU and if_start
2601  *   interlock (if_snd.altq_started) is not released.
2602  * - The if_start_rollup(), which is registered as low priority netisr
2603  *   rollup function, is called; probably because no more work is pending
2604  *   for netisr.
2605  *
2606  * NOTE:
2607  * Currently IFSUBQ packet staging is only performed in netisr threads.
2608  */
2609 int
2610 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2611 {
2612 	struct ifaltq *ifq = &ifp->if_snd;
2613 	struct ifaltq_subque *ifsq;
2614 	int error, start = 0, len, mcast = 0, avoid_start = 0;
2615 	struct ifsubq_stage_head *head = NULL;
2616 	struct ifsubq_stage *stage = NULL;
2617 
2618 	ifsq = ifq_map_subq(ifq, mycpuid);
2619 	ASSERT_IFNET_NOT_SERIALIZED_TX(ifp, ifsq);
2620 
2621 	len = m->m_pkthdr.len;
2622 	if (m->m_flags & M_MCAST)
2623 		mcast = 1;
2624 
2625 	if (curthread->td_type == TD_TYPE_NETISR) {
2626 		head = &ifsubq_stage_heads[mycpuid];
2627 		stage = ifsq_get_stage(ifsq, mycpuid);
2628 
2629 		stage->stg_cnt++;
2630 		stage->stg_len += len;
2631 		if (stage->stg_cnt < ifsq_stage_cntmax &&
2632 		    stage->stg_len < (ifp->if_mtu - max_protohdr))
2633 			avoid_start = 1;
2634 	}
2635 
2636 	ALTQ_SQ_LOCK(ifsq);
2637 	error = ifsq_enqueue_locked(ifsq, m, pa);
2638 	if (error) {
2639 		if (!ifsq_data_ready(ifsq)) {
2640 			ALTQ_SQ_UNLOCK(ifsq);
2641 			return error;
2642 		}
2643 		avoid_start = 0;
2644 	}
2645 	if (!ifsq_is_started(ifsq)) {
2646 		if (avoid_start) {
2647 			ALTQ_SQ_UNLOCK(ifsq);
2648 
2649 			KKASSERT(!error);
2650 			if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
2651 				ifsq_stage_insert(head, stage);
2652 
2653 			IFNET_STAT_INC(ifp, obytes, len);
2654 			if (mcast)
2655 				IFNET_STAT_INC(ifp, omcasts, 1);
2656 			return error;
2657 		}
2658 
2659 		/*
2660 		 * Hold the interlock of ifnet.if_start
2661 		 */
2662 		ifsq_set_started(ifsq);
2663 		start = 1;
2664 	}
2665 	ALTQ_SQ_UNLOCK(ifsq);
2666 
2667 	if (!error) {
2668 		IFNET_STAT_INC(ifp, obytes, len);
2669 		if (mcast)
2670 			IFNET_STAT_INC(ifp, omcasts, 1);
2671 	}
2672 
2673 	if (stage != NULL) {
2674 		if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
2675 			KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
2676 			if (!avoid_start) {
2677 				ifsq_stage_remove(head, stage);
2678 				ifsq_ifstart_schedule(ifsq, 1);
2679 			}
2680 			return error;
2681 		}
2682 
2683 		if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
2684 			ifsq_stage_remove(head, stage);
2685 		} else {
2686 			stage->stg_cnt = 0;
2687 			stage->stg_len = 0;
2688 		}
2689 	}
2690 
2691 	if (!start)
2692 		return error;
2693 
2694 	ifsq_ifstart_try(ifsq, 0);
2695 	return error;
2696 }
2697 
2698 void *
2699 ifa_create(int size, int flags)
2700 {
2701 	struct ifaddr *ifa;
2702 	int i;
2703 
2704 	KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
2705 
2706 	ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2707 	if (ifa == NULL)
2708 		return NULL;
2709 
2710 	ifa->ifa_containers =
2711 	    kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
2712 	        M_IFADDR, M_WAITOK | M_ZERO);
2713 	ifa->ifa_ncnt = ncpus;
2714 	for (i = 0; i < ncpus; ++i) {
2715 		struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2716 
2717 		ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2718 		ifac->ifa = ifa;
2719 		ifac->ifa_refcnt = 1;
2720 	}
2721 #ifdef IFADDR_DEBUG
2722 	kprintf("alloc ifa %p %d\n", ifa, size);
2723 #endif
2724 	return ifa;
2725 }
2726 
2727 void
2728 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2729 {
2730 	struct ifaddr *ifa = ifac->ifa;
2731 
2732 	KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2733 	KKASSERT(ifac->ifa_refcnt == 0);
2734 	KASSERT(ifac->ifa_listmask == 0,
2735 		("ifa is still on %#x lists", ifac->ifa_listmask));
2736 
2737 	ifac->ifa_magic = IFA_CONTAINER_DEAD;
2738 
2739 #ifdef IFADDR_DEBUG_VERBOSE
2740 	kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2741 #endif
2742 
2743 	KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2744 		("invalid # of ifac, %d", ifa->ifa_ncnt));
2745 	if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2746 #ifdef IFADDR_DEBUG
2747 		kprintf("free ifa %p\n", ifa);
2748 #endif
2749 		kfree(ifa->ifa_containers, M_IFADDR);
2750 		kfree(ifa, M_IFADDR);
2751 	}
2752 }
2753 
2754 static void
2755 ifa_iflink_dispatch(netmsg_t nmsg)
2756 {
2757 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2758 	struct ifaddr *ifa = msg->ifa;
2759 	struct ifnet *ifp = msg->ifp;
2760 	int cpu = mycpuid;
2761 	struct ifaddr_container *ifac;
2762 
2763 	crit_enter();
2764 
2765 	ifac = &ifa->ifa_containers[cpu];
2766 	ASSERT_IFAC_VALID(ifac);
2767 	KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2768 		("ifaddr is on if_addrheads"));
2769 
2770 	ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2771 	if (msg->tail)
2772 		TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2773 	else
2774 		TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2775 
2776 	crit_exit();
2777 
2778 	ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2779 }
2780 
2781 void
2782 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2783 {
2784 	struct netmsg_ifaddr msg;
2785 
2786 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2787 		    0, ifa_iflink_dispatch);
2788 	msg.ifa = ifa;
2789 	msg.ifp = ifp;
2790 	msg.tail = tail;
2791 
2792 	ifa_domsg(&msg.base.lmsg, 0);
2793 }
2794 
2795 static void
2796 ifa_ifunlink_dispatch(netmsg_t nmsg)
2797 {
2798 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2799 	struct ifaddr *ifa = msg->ifa;
2800 	struct ifnet *ifp = msg->ifp;
2801 	int cpu = mycpuid;
2802 	struct ifaddr_container *ifac;
2803 
2804 	crit_enter();
2805 
2806 	ifac = &ifa->ifa_containers[cpu];
2807 	ASSERT_IFAC_VALID(ifac);
2808 	KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2809 		("ifaddr is not on if_addrhead"));
2810 
2811 	TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2812 	ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2813 
2814 	crit_exit();
2815 
2816 	ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2817 }
2818 
2819 void
2820 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2821 {
2822 	struct netmsg_ifaddr msg;
2823 
2824 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2825 		    0, ifa_ifunlink_dispatch);
2826 	msg.ifa = ifa;
2827 	msg.ifp = ifp;
2828 
2829 	ifa_domsg(&msg.base.lmsg, 0);
2830 }
2831 
2832 static void
2833 ifa_destroy_dispatch(netmsg_t nmsg)
2834 {
2835 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2836 
2837 	IFAFREE(msg->ifa);
2838 	ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2839 }
2840 
2841 void
2842 ifa_destroy(struct ifaddr *ifa)
2843 {
2844 	struct netmsg_ifaddr msg;
2845 
2846 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2847 		    0, ifa_destroy_dispatch);
2848 	msg.ifa = ifa;
2849 
2850 	ifa_domsg(&msg.base.lmsg, 0);
2851 }
2852 
2853 struct lwkt_port *
2854 ifnet_portfn(int cpu)
2855 {
2856 	return &ifnet_threads[cpu].td_msgport;
2857 }
2858 
2859 void
2860 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2861 {
2862 	KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2863 
2864 	if (next_cpu < ncpus)
2865 		lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2866 	else
2867 		lwkt_replymsg(lmsg, 0);
2868 }
2869 
2870 int
2871 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
2872 {
2873 	KKASSERT(cpu < ncpus);
2874 	return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
2875 }
2876 
2877 void
2878 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
2879 {
2880 	KKASSERT(cpu < ncpus);
2881 	lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
2882 }
2883 
2884 /*
2885  * Generic netmsg service loop.  Some protocols may roll their own but all
2886  * must do the basic command dispatch function call done here.
2887  */
2888 static void
2889 ifnet_service_loop(void *arg __unused)
2890 {
2891 	netmsg_t msg;
2892 
2893 	while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
2894 		KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
2895 		msg->base.nm_dispatch(msg);
2896 	}
2897 }
2898 
2899 static void
2900 if_start_rollup(void)
2901 {
2902 	struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
2903 	struct ifsubq_stage *stage;
2904 
2905 	while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
2906 		struct ifaltq_subque *ifsq = stage->stg_subq;
2907 		int is_sched = 0;
2908 
2909 		if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
2910 			is_sched = 1;
2911 		ifsq_stage_remove(head, stage);
2912 
2913 		if (is_sched) {
2914 			ifsq_ifstart_schedule(ifsq, 1);
2915 		} else {
2916 			int start = 0;
2917 
2918 			ALTQ_SQ_LOCK(ifsq);
2919 			if (!ifsq_is_started(ifsq)) {
2920 				/*
2921 				 * Hold the interlock of ifnet.if_start
2922 				 */
2923 				ifsq_set_started(ifsq);
2924 				start = 1;
2925 			}
2926 			ALTQ_SQ_UNLOCK(ifsq);
2927 
2928 			if (start)
2929 				ifsq_ifstart_try(ifsq, 1);
2930 		}
2931 		KKASSERT((stage->stg_flags &
2932 		    (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
2933 	}
2934 }
2935 
2936 static void
2937 ifnetinit(void *dummy __unused)
2938 {
2939 	int i;
2940 
2941 	for (i = 0; i < ncpus; ++i) {
2942 		struct thread *thr = &ifnet_threads[i];
2943 
2944 		lwkt_create(ifnet_service_loop, NULL, NULL,
2945 			    thr, TDF_NOSTART|TDF_FORCE_SPINPORT,
2946 			    i, "ifnet %d", i);
2947 		netmsg_service_port_init(&thr->td_msgport);
2948 		lwkt_schedule(thr);
2949 	}
2950 
2951 	for (i = 0; i < ncpus; ++i)
2952 		TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
2953 	netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
2954 }
2955 
2956 struct ifnet *
2957 ifnet_byindex(unsigned short idx)
2958 {
2959 	if (idx > if_index)
2960 		return NULL;
2961 	return ifindex2ifnet[idx];
2962 }
2963 
2964 struct ifaddr *
2965 ifaddr_byindex(unsigned short idx)
2966 {
2967 	struct ifnet *ifp;
2968 
2969 	ifp = ifnet_byindex(idx);
2970 	if (!ifp)
2971 		return NULL;
2972 	return TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
2973 }
2974 
2975 void
2976 if_register_com_alloc(u_char type,
2977     if_com_alloc_t *a, if_com_free_t *f)
2978 {
2979 
2980         KASSERT(if_com_alloc[type] == NULL,
2981             ("if_register_com_alloc: %d already registered", type));
2982         KASSERT(if_com_free[type] == NULL,
2983             ("if_register_com_alloc: %d free already registered", type));
2984 
2985         if_com_alloc[type] = a;
2986         if_com_free[type] = f;
2987 }
2988 
2989 void
2990 if_deregister_com_alloc(u_char type)
2991 {
2992 
2993         KASSERT(if_com_alloc[type] != NULL,
2994             ("if_deregister_com_alloc: %d not registered", type));
2995         KASSERT(if_com_free[type] != NULL,
2996             ("if_deregister_com_alloc: %d free not registered", type));
2997         if_com_alloc[type] = NULL;
2998         if_com_free[type] = NULL;
2999 }
3000 
3001 int
3002 if_ring_count2(int cnt, int cnt_max)
3003 {
3004 	int shift = 0;
3005 
3006 	KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3007 	    ("invalid ring count max %d", cnt_max));
3008 
3009 	if (cnt <= 0)
3010 		cnt = cnt_max;
3011 	if (cnt > ncpus2)
3012 		cnt = ncpus2;
3013 	if (cnt > cnt_max)
3014 		cnt = cnt_max;
3015 
3016 	while ((1 << (shift + 1)) <= cnt)
3017 		++shift;
3018 	cnt = 1 << shift;
3019 
3020 	KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3021 	    ("calculate cnt %d, ncpus2 %d, cnt max %d",
3022 	     cnt, ncpus2, cnt_max));
3023 	return cnt;
3024 }
3025 
3026 void
3027 ifq_set_maxlen(struct ifaltq *ifq, int len)
3028 {
3029 	ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3030 }
3031 
3032 int
3033 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3034 {
3035 	return ALTQ_SUBQ_INDEX_DEFAULT;
3036 }
3037 
3038 int
3039 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3040 {
3041 	return (cpuid & ifq->altq_subq_mask);
3042 }
3043 
3044 static void
3045 ifsq_watchdog(void *arg)
3046 {
3047 	struct ifsubq_watchdog *wd = arg;
3048 	struct ifnet *ifp;
3049 
3050 	if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3051 		goto done;
3052 
3053 	ifp = ifsq_get_ifp(wd->wd_subq);
3054 	if (ifnet_tryserialize_all(ifp)) {
3055 		wd->wd_watchdog(wd->wd_subq);
3056 		ifnet_deserialize_all(ifp);
3057 	} else {
3058 		/* try again next timeout */
3059 		wd->wd_timer = 1;
3060 	}
3061 done:
3062 	ifsq_watchdog_reset(wd);
3063 }
3064 
3065 static void
3066 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3067 {
3068 	callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3069 	    ifsq_get_cpuid(wd->wd_subq));
3070 }
3071 
3072 void
3073 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3074     ifsq_watchdog_t watchdog)
3075 {
3076 	callout_init_mp(&wd->wd_callout);
3077 	wd->wd_timer = 0;
3078 	wd->wd_subq = ifsq;
3079 	wd->wd_watchdog = watchdog;
3080 }
3081 
3082 void
3083 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3084 {
3085 	wd->wd_timer = 0;
3086 	ifsq_watchdog_reset(wd);
3087 }
3088 
3089 void
3090 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3091 {
3092 	wd->wd_timer = 0;
3093 	callout_stop(&wd->wd_callout);
3094 }
3095