xref: /dflybsd-src/sys/net/if.c (revision b5523eac31a95e6876e05e20e6fe836ec3a45202)
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. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)if.c	8.3 (Berkeley) 1/4/94
30  * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
31  */
32 
33 #include "opt_compat.h"
34 #include "opt_inet6.h"
35 #include "opt_inet.h"
36 #include "opt_ifpoll.h"
37 
38 #include <sys/param.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/systm.h>
42 #include <sys/proc.h>
43 #include <sys/priv.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/socketops.h>
48 #include <sys/kernel.h>
49 #include <sys/ktr.h>
50 #include <sys/mutex.h>
51 #include <sys/sockio.h>
52 #include <sys/syslog.h>
53 #include <sys/sysctl.h>
54 #include <sys/domain.h>
55 #include <sys/thread.h>
56 #include <sys/serialize.h>
57 #include <sys/bus.h>
58 
59 #include <sys/thread2.h>
60 #include <sys/msgport2.h>
61 #include <sys/mutex2.h>
62 
63 #include <net/if.h>
64 #include <net/if_arp.h>
65 #include <net/if_dl.h>
66 #include <net/if_types.h>
67 #include <net/if_var.h>
68 #include <net/ifq_var.h>
69 #include <net/radix.h>
70 #include <net/route.h>
71 #include <net/if_clone.h>
72 #include <net/netisr2.h>
73 #include <net/netmsg2.h>
74 
75 #include <machine/atomic.h>
76 #include <machine/stdarg.h>
77 #include <machine/smp.h>
78 
79 #if defined(INET) || defined(INET6)
80 /*XXX*/
81 #include <netinet/in.h>
82 #include <netinet/in_var.h>
83 #include <netinet/if_ether.h>
84 #ifdef INET6
85 #include <netinet6/in6_var.h>
86 #include <netinet6/in6_ifattach.h>
87 #endif
88 #endif
89 
90 #if defined(COMPAT_43)
91 #include <emulation/43bsd/43bsd_socket.h>
92 #endif /* COMPAT_43 */
93 
94 struct netmsg_ifaddr {
95 	struct netmsg_base base;
96 	struct ifaddr	*ifa;
97 	struct ifnet	*ifp;
98 	int		tail;
99 };
100 
101 struct ifsubq_stage_head {
102 	TAILQ_HEAD(, ifsubq_stage)	stg_head;
103 } __cachealign;
104 
105 /*
106  * System initialization
107  */
108 static void	if_attachdomain(void *);
109 static void	if_attachdomain1(struct ifnet *);
110 static int	ifconf(u_long, caddr_t, struct ucred *);
111 static void	ifinit(void *);
112 static void	ifnetinit(void *);
113 static void	if_slowtimo(void *);
114 static void	link_rtrequest(int, struct rtentry *);
115 static int	if_rtdel(struct radix_node *, void *);
116 static void	if_slowtimo_dispatch(netmsg_t);
117 
118 /* Helper functions */
119 static void	ifsq_watchdog_reset(struct ifsubq_watchdog *);
120 static int	if_delmulti_serialized(struct ifnet *, struct sockaddr *);
121 static struct ifnet_array *ifnet_array_alloc(int);
122 static void	ifnet_array_free(struct ifnet_array *);
123 static struct ifnet_array *ifnet_array_add(struct ifnet *,
124 		    const struct ifnet_array *);
125 static struct ifnet_array *ifnet_array_del(struct ifnet *,
126 		    const struct ifnet_array *);
127 
128 #ifdef INET6
129 /*
130  * XXX: declare here to avoid to include many inet6 related files..
131  * should be more generalized?
132  */
133 extern void	nd6_setmtu(struct ifnet *);
134 #endif
135 
136 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
137 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
138 
139 static int ifsq_stage_cntmax = 4;
140 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
141 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
142     &ifsq_stage_cntmax, 0, "ifq staging packet count max");
143 
144 static int if_stats_compat = 0;
145 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
146     &if_stats_compat, 0, "Compat the old ifnet stats");
147 
148 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
149 /* Must be after netisr_init */
150 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
151 
152 static  if_com_alloc_t *if_com_alloc[256];
153 static  if_com_free_t *if_com_free[256];
154 
155 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
156 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
157 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
158 
159 int			ifqmaxlen = IFQ_MAXLEN;
160 struct ifnethead	ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
161 
162 static struct ifnet_array	ifnet_array0;
163 static struct ifnet_array	*ifnet_array = &ifnet_array0;
164 
165 static struct callout		if_slowtimo_timer;
166 static struct netmsg_base	if_slowtimo_netmsg;
167 
168 int			if_index = 0;
169 struct ifnet		**ifindex2ifnet = NULL;
170 static struct thread	ifnet_threads[MAXCPU];
171 static struct mtx	ifnet_mtx = MTX_INITIALIZER;
172 
173 static struct ifsubq_stage_head	ifsubq_stage_heads[MAXCPU];
174 
175 #ifdef notyet
176 #define IFQ_KTR_STRING		"ifq=%p"
177 #define IFQ_KTR_ARGS	struct ifaltq *ifq
178 #ifndef KTR_IFQ
179 #define KTR_IFQ			KTR_ALL
180 #endif
181 KTR_INFO_MASTER(ifq);
182 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
183 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
184 #define logifq(name, arg)	KTR_LOG(ifq_ ## name, arg)
185 
186 #define IF_START_KTR_STRING	"ifp=%p"
187 #define IF_START_KTR_ARGS	struct ifnet *ifp
188 #ifndef KTR_IF_START
189 #define KTR_IF_START		KTR_ALL
190 #endif
191 KTR_INFO_MASTER(if_start);
192 KTR_INFO(KTR_IF_START, if_start, run, 0,
193 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
194 KTR_INFO(KTR_IF_START, if_start, sched, 1,
195 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
196 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
197 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
198 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
199 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
200 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
201 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
202 #define logifstart(name, arg)	KTR_LOG(if_start_ ## name, arg)
203 #endif
204 
205 TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
206 
207 /*
208  * Network interface utility routines.
209  *
210  * Routines with ifa_ifwith* names take sockaddr *'s as
211  * parameters.
212  */
213 /* ARGSUSED*/
214 void
215 ifinit(void *dummy)
216 {
217 	struct ifnet *ifp;
218 
219 	callout_init_mp(&if_slowtimo_timer);
220 	netmsg_init(&if_slowtimo_netmsg, NULL, &netisr_adone_rport,
221 	    MSGF_PRIORITY, if_slowtimo_dispatch);
222 
223 	/* XXX is this necessary? */
224 	ifnet_lock();
225 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
226 		if (ifp->if_snd.altq_maxlen == 0) {
227 			if_printf(ifp, "XXX: driver didn't set altq_maxlen\n");
228 			ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
229 		}
230 	}
231 	ifnet_unlock();
232 
233 	/* Start if_slowtimo */
234 	lwkt_sendmsg(netisr_cpuport(0), &if_slowtimo_netmsg.lmsg);
235 }
236 
237 static void
238 ifsq_ifstart_ipifunc(void *arg)
239 {
240 	struct ifaltq_subque *ifsq = arg;
241 	struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
242 
243 	crit_enter();
244 	if (lmsg->ms_flags & MSGF_DONE)
245 		lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg);
246 	crit_exit();
247 }
248 
249 static __inline void
250 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
251 {
252 	KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
253 	TAILQ_REMOVE(&head->stg_head, stage, stg_link);
254 	stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
255 	stage->stg_cnt = 0;
256 	stage->stg_len = 0;
257 }
258 
259 static __inline void
260 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
261 {
262 	KKASSERT((stage->stg_flags &
263 	    (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
264 	stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
265 	TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
266 }
267 
268 /*
269  * Schedule ifnet.if_start on the subqueue owner CPU
270  */
271 static void
272 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
273 {
274 	int cpu;
275 
276 	if (!force && curthread->td_type == TD_TYPE_NETISR &&
277 	    ifsq_stage_cntmax > 0) {
278 		struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
279 
280 		stage->stg_cnt = 0;
281 		stage->stg_len = 0;
282 		if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
283 			ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
284 		stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
285 		return;
286 	}
287 
288 	cpu = ifsq_get_cpuid(ifsq);
289 	if (cpu != mycpuid)
290 		lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
291 	else
292 		ifsq_ifstart_ipifunc(ifsq);
293 }
294 
295 /*
296  * NOTE:
297  * This function will release ifnet.if_start subqueue interlock,
298  * if ifnet.if_start for the subqueue does not need to be scheduled
299  */
300 static __inline int
301 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
302 {
303 	if (!running || ifsq_is_empty(ifsq)
304 #ifdef ALTQ
305 	    || ifsq->ifsq_altq->altq_tbr != NULL
306 #endif
307 	) {
308 		ALTQ_SQ_LOCK(ifsq);
309 		/*
310 		 * ifnet.if_start subqueue interlock is released, if:
311 		 * 1) Hardware can not take any packets, due to
312 		 *    o  interface is marked down
313 		 *    o  hardware queue is full (ifsq_is_oactive)
314 		 *    Under the second situation, hardware interrupt
315 		 *    or polling(4) will call/schedule ifnet.if_start
316 		 *    on the subqueue when hardware queue is ready
317 		 * 2) There is no packet in the subqueue.
318 		 *    Further ifq_dispatch or ifq_handoff will call/
319 		 *    schedule ifnet.if_start on the subqueue.
320 		 * 3) TBR is used and it does not allow further
321 		 *    dequeueing.
322 		 *    TBR callout will call ifnet.if_start on the
323 		 *    subqueue.
324 		 */
325 		if (!running || !ifsq_data_ready(ifsq)) {
326 			ifsq_clr_started(ifsq);
327 			ALTQ_SQ_UNLOCK(ifsq);
328 			return 0;
329 		}
330 		ALTQ_SQ_UNLOCK(ifsq);
331 	}
332 	return 1;
333 }
334 
335 static void
336 ifsq_ifstart_dispatch(netmsg_t msg)
337 {
338 	struct lwkt_msg *lmsg = &msg->base.lmsg;
339 	struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
340 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
341 	struct globaldata *gd = mycpu;
342 	int running = 0, need_sched;
343 
344 	crit_enter_gd(gd);
345 
346 	lwkt_replymsg(lmsg, 0);	/* reply ASAP */
347 
348 	if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) {
349 		/*
350 		 * We need to chase the subqueue owner CPU change.
351 		 */
352 		ifsq_ifstart_schedule(ifsq, 1);
353 		crit_exit_gd(gd);
354 		return;
355 	}
356 
357 	ifsq_serialize_hw(ifsq);
358 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
359 		ifp->if_start(ifp, ifsq);
360 		if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
361 			running = 1;
362 	}
363 	need_sched = ifsq_ifstart_need_schedule(ifsq, running);
364 	ifsq_deserialize_hw(ifsq);
365 
366 	if (need_sched) {
367 		/*
368 		 * More data need to be transmitted, ifnet.if_start is
369 		 * scheduled on the subqueue owner CPU, and we keep going.
370 		 * NOTE: ifnet.if_start subqueue interlock is not released.
371 		 */
372 		ifsq_ifstart_schedule(ifsq, 0);
373 	}
374 
375 	crit_exit_gd(gd);
376 }
377 
378 /* Device driver ifnet.if_start helper function */
379 void
380 ifsq_devstart(struct ifaltq_subque *ifsq)
381 {
382 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
383 	int running = 0;
384 
385 	ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
386 
387 	ALTQ_SQ_LOCK(ifsq);
388 	if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
389 		ALTQ_SQ_UNLOCK(ifsq);
390 		return;
391 	}
392 	ifsq_set_started(ifsq);
393 	ALTQ_SQ_UNLOCK(ifsq);
394 
395 	ifp->if_start(ifp, ifsq);
396 
397 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
398 		running = 1;
399 
400 	if (ifsq_ifstart_need_schedule(ifsq, running)) {
401 		/*
402 		 * More data need to be transmitted, ifnet.if_start is
403 		 * scheduled on ifnet's CPU, and we keep going.
404 		 * NOTE: ifnet.if_start interlock is not released.
405 		 */
406 		ifsq_ifstart_schedule(ifsq, 0);
407 	}
408 }
409 
410 void
411 if_devstart(struct ifnet *ifp)
412 {
413 	ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
414 }
415 
416 /* Device driver ifnet.if_start schedule helper function */
417 void
418 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
419 {
420 	ifsq_ifstart_schedule(ifsq, 1);
421 }
422 
423 void
424 if_devstart_sched(struct ifnet *ifp)
425 {
426 	ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
427 }
428 
429 static void
430 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
431 {
432 	lwkt_serialize_enter(ifp->if_serializer);
433 }
434 
435 static void
436 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
437 {
438 	lwkt_serialize_exit(ifp->if_serializer);
439 }
440 
441 static int
442 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
443 {
444 	return lwkt_serialize_try(ifp->if_serializer);
445 }
446 
447 #ifdef INVARIANTS
448 static void
449 if_default_serialize_assert(struct ifnet *ifp,
450 			    enum ifnet_serialize slz __unused,
451 			    boolean_t serialized)
452 {
453 	if (serialized)
454 		ASSERT_SERIALIZED(ifp->if_serializer);
455 	else
456 		ASSERT_NOT_SERIALIZED(ifp->if_serializer);
457 }
458 #endif
459 
460 /*
461  * Attach an interface to the list of "active" interfaces.
462  *
463  * The serializer is optional.
464  */
465 void
466 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
467 {
468 	unsigned socksize, ifasize;
469 	int namelen, masklen;
470 	struct sockaddr_dl *sdl, *sdl_addr;
471 	struct ifaddr *ifa;
472 	struct ifaltq *ifq;
473 	struct ifnet **old_ifindex2ifnet = NULL;
474 	struct ifnet_array *old_ifnet_array;
475 	int i, q;
476 
477 	static int if_indexlim = 8;
478 
479 	if (ifp->if_serialize != NULL) {
480 		KASSERT(ifp->if_deserialize != NULL &&
481 			ifp->if_tryserialize != NULL &&
482 			ifp->if_serialize_assert != NULL,
483 			("serialize functions are partially setup"));
484 
485 		/*
486 		 * If the device supplies serialize functions,
487 		 * then clear if_serializer to catch any invalid
488 		 * usage of this field.
489 		 */
490 		KASSERT(serializer == NULL,
491 			("both serialize functions and default serializer "
492 			 "are supplied"));
493 		ifp->if_serializer = NULL;
494 	} else {
495 		KASSERT(ifp->if_deserialize == NULL &&
496 			ifp->if_tryserialize == NULL &&
497 			ifp->if_serialize_assert == NULL,
498 			("serialize functions are partially setup"));
499 		ifp->if_serialize = if_default_serialize;
500 		ifp->if_deserialize = if_default_deserialize;
501 		ifp->if_tryserialize = if_default_tryserialize;
502 #ifdef INVARIANTS
503 		ifp->if_serialize_assert = if_default_serialize_assert;
504 #endif
505 
506 		/*
507 		 * The serializer can be passed in from the device,
508 		 * allowing the same serializer to be used for both
509 		 * the interrupt interlock and the device queue.
510 		 * If not specified, the netif structure will use an
511 		 * embedded serializer.
512 		 */
513 		if (serializer == NULL) {
514 			serializer = &ifp->if_default_serializer;
515 			lwkt_serialize_init(serializer);
516 		}
517 		ifp->if_serializer = serializer;
518 	}
519 
520 	mtx_init(&ifp->if_ioctl_mtx);
521 
522 	/*
523 	 * XXX -
524 	 * The old code would work if the interface passed a pre-existing
525 	 * chain of ifaddrs to this code.  We don't trust our callers to
526 	 * properly initialize the tailq, however, so we no longer allow
527 	 * this unlikely case.
528 	 */
529 	ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
530 				    M_IFADDR, M_WAITOK | M_ZERO);
531 	for (i = 0; i < ncpus; ++i)
532 		TAILQ_INIT(&ifp->if_addrheads[i]);
533 
534 	TAILQ_INIT(&ifp->if_multiaddrs);
535 	TAILQ_INIT(&ifp->if_groups);
536 	getmicrotime(&ifp->if_lastchange);
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 	if (socksize < sizeof(*sdl))
545 		socksize = sizeof(*sdl);
546 	socksize = RT_ROUNDUP(socksize);
547 	ifasize = sizeof(struct ifaddr) + 2 * socksize;
548 	ifa = ifa_create(ifasize, M_WAITOK);
549 	sdl = sdl_addr = (struct sockaddr_dl *)(ifa + 1);
550 	sdl->sdl_len = socksize;
551 	sdl->sdl_family = AF_LINK;
552 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
553 	sdl->sdl_nlen = namelen;
554 	sdl->sdl_type = ifp->if_type;
555 	ifp->if_lladdr = ifa;
556 	ifa->ifa_ifp = ifp;
557 	ifa->ifa_rtrequest = link_rtrequest;
558 	ifa->ifa_addr = (struct sockaddr *)sdl;
559 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
560 	ifa->ifa_netmask = (struct sockaddr *)sdl;
561 	sdl->sdl_len = masklen;
562 	while (namelen != 0)
563 		sdl->sdl_data[--namelen] = 0xff;
564 	ifa_iflink(ifa, ifp, 0 /* Insert head */);
565 
566 	ifp->if_data_pcpu = kmalloc_cachealign(
567 	    ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
568 
569 	if (ifp->if_mapsubq == NULL)
570 		ifp->if_mapsubq = ifq_mapsubq_default;
571 
572 	ifq = &ifp->if_snd;
573 	ifq->altq_type = 0;
574 	ifq->altq_disc = NULL;
575 	ifq->altq_flags &= ALTQF_CANTCHANGE;
576 	ifq->altq_tbr = NULL;
577 	ifq->altq_ifp = ifp;
578 
579 	if (ifq->altq_subq_cnt <= 0)
580 		ifq->altq_subq_cnt = 1;
581 	ifq->altq_subq = kmalloc_cachealign(
582 	    ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
583 	    M_DEVBUF, M_WAITOK | M_ZERO);
584 
585 	if (ifq->altq_maxlen == 0) {
586 		if_printf(ifp, "driver didn't set altq_maxlen\n");
587 		ifq_set_maxlen(ifq, ifqmaxlen);
588 	}
589 
590 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
591 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
592 
593 		ALTQ_SQ_LOCK_INIT(ifsq);
594 		ifsq->ifsq_index = q;
595 
596 		ifsq->ifsq_altq = ifq;
597 		ifsq->ifsq_ifp = ifp;
598 
599 		ifsq->ifsq_maxlen = ifq->altq_maxlen;
600 		ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
601 		ifsq->ifsq_prepended = NULL;
602 		ifsq->ifsq_started = 0;
603 		ifsq->ifsq_hw_oactive = 0;
604 		ifsq_set_cpuid(ifsq, 0);
605 		if (ifp->if_serializer != NULL)
606 			ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
607 
608 		ifsq->ifsq_stage =
609 		    kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
610 		    M_DEVBUF, M_WAITOK | M_ZERO);
611 		for (i = 0; i < ncpus; ++i)
612 			ifsq->ifsq_stage[i].stg_subq = ifsq;
613 
614 		ifsq->ifsq_ifstart_nmsg =
615 		    kmalloc(ncpus * sizeof(struct netmsg_base),
616 		    M_LWKTMSG, M_WAITOK);
617 		for (i = 0; i < ncpus; ++i) {
618 			netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
619 			    &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
620 			ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
621 		}
622 	}
623 	ifq_set_classic(ifq);
624 
625 	/*
626 	 * Install this ifp into ifindex2inet, ifnet queue and ifnet
627 	 * array after it is setup.
628 	 *
629 	 * Protect ifindex2ifnet, ifnet queue and ifnet array changes
630 	 * by ifnet lock, so that non-netisr threads could get a
631 	 * consistent view.
632 	 */
633 	ifnet_lock();
634 
635 	/* Don't update if_index until ifindex2ifnet is setup */
636 	ifp->if_index = if_index + 1;
637 	sdl_addr->sdl_index = ifp->if_index;
638 
639 	/*
640 	 * Install this ifp into ifindex2ifnet
641 	 */
642 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
643 		unsigned int n;
644 		struct ifnet **q;
645 
646 		/*
647 		 * Grow ifindex2ifnet
648 		 */
649 		if_indexlim <<= 1;
650 		n = if_indexlim * sizeof(*q);
651 		q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
652 		if (ifindex2ifnet != NULL) {
653 			bcopy(ifindex2ifnet, q, n/2);
654 			/* Free old ifindex2ifnet after sync all netisrs */
655 			old_ifindex2ifnet = ifindex2ifnet;
656 		}
657 		ifindex2ifnet = q;
658 	}
659 	ifindex2ifnet[ifp->if_index] = ifp;
660 	/*
661 	 * Update if_index after this ifp is installed into ifindex2ifnet,
662 	 * so that netisrs could get a consistent view of ifindex2ifnet.
663 	 */
664 	cpu_sfence();
665 	if_index = ifp->if_index;
666 
667 	/*
668 	 * Install this ifp into ifnet array.
669 	 */
670 	/* Free old ifnet array after sync all netisrs */
671 	old_ifnet_array = ifnet_array;
672 	ifnet_array = ifnet_array_add(ifp, old_ifnet_array);
673 
674 	/*
675 	 * Install this ifp into ifnet queue.
676 	 */
677 	TAILQ_INSERT_TAIL(&ifnetlist, ifp, if_link);
678 
679 	ifnet_unlock();
680 
681 	/*
682 	 * Sync all netisrs so that the old ifindex2ifnet and ifnet array
683 	 * are no longer accessed and we can free them safely later on.
684 	 */
685 	netmsg_service_sync();
686 	if (old_ifindex2ifnet != NULL)
687 		kfree(old_ifindex2ifnet, M_IFADDR);
688 	ifnet_array_free(old_ifnet_array);
689 
690 	if (!SLIST_EMPTY(&domains))
691 		if_attachdomain1(ifp);
692 
693 	/* Announce the interface. */
694 	EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
695 	devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
696 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
697 }
698 
699 static void
700 if_attachdomain(void *dummy)
701 {
702 	struct ifnet *ifp;
703 
704 	ifnet_lock();
705 	TAILQ_FOREACH(ifp, &ifnetlist, if_list)
706 		if_attachdomain1(ifp);
707 	ifnet_unlock();
708 }
709 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
710 	if_attachdomain, NULL);
711 
712 static void
713 if_attachdomain1(struct ifnet *ifp)
714 {
715 	struct domain *dp;
716 
717 	crit_enter();
718 
719 	/* address family dependent data region */
720 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
721 	SLIST_FOREACH(dp, &domains, dom_next)
722 		if (dp->dom_ifattach)
723 			ifp->if_afdata[dp->dom_family] =
724 				(*dp->dom_ifattach)(ifp);
725 	crit_exit();
726 }
727 
728 /*
729  * Purge all addresses whose type is _not_ AF_LINK
730  */
731 void
732 if_purgeaddrs_nolink(struct ifnet *ifp)
733 {
734 	struct ifaddr_container *ifac, *next;
735 
736 	TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
737 			      ifa_link, next) {
738 		struct ifaddr *ifa = ifac->ifa;
739 
740 		/* Leave link ifaddr as it is */
741 		if (ifa->ifa_addr->sa_family == AF_LINK)
742 			continue;
743 #ifdef INET
744 		/* XXX: Ugly!! ad hoc just for INET */
745 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
746 			struct ifaliasreq ifr;
747 #ifdef IFADDR_DEBUG_VERBOSE
748 			int i;
749 
750 			kprintf("purge in4 addr %p: ", ifa);
751 			for (i = 0; i < ncpus; ++i)
752 				kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
753 			kprintf("\n");
754 #endif
755 
756 			bzero(&ifr, sizeof ifr);
757 			ifr.ifra_addr = *ifa->ifa_addr;
758 			if (ifa->ifa_dstaddr)
759 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
760 			if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp,
761 				       NULL) == 0)
762 				continue;
763 		}
764 #endif /* INET */
765 #ifdef INET6
766 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
767 #ifdef IFADDR_DEBUG_VERBOSE
768 			int i;
769 
770 			kprintf("purge in6 addr %p: ", ifa);
771 			for (i = 0; i < ncpus; ++i)
772 				kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
773 			kprintf("\n");
774 #endif
775 
776 			in6_purgeaddr(ifa);
777 			/* ifp_addrhead is already updated */
778 			continue;
779 		}
780 #endif /* INET6 */
781 		ifa_ifunlink(ifa, ifp);
782 		ifa_destroy(ifa);
783 	}
784 }
785 
786 static void
787 ifq_stage_detach_handler(netmsg_t nmsg)
788 {
789 	struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
790 	int q;
791 
792 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
793 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
794 		struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
795 
796 		if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
797 			ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
798 	}
799 	lwkt_replymsg(&nmsg->lmsg, 0);
800 }
801 
802 static void
803 ifq_stage_detach(struct ifaltq *ifq)
804 {
805 	struct netmsg_base base;
806 	int cpu;
807 
808 	netmsg_init(&base, NULL, &curthread->td_msgport, 0,
809 	    ifq_stage_detach_handler);
810 	base.lmsg.u.ms_resultp = ifq;
811 
812 	for (cpu = 0; cpu < ncpus; ++cpu)
813 		lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
814 }
815 
816 struct netmsg_if_rtdel {
817 	struct netmsg_base	base;
818 	struct ifnet		*ifp;
819 };
820 
821 static void
822 if_rtdel_dispatch(netmsg_t msg)
823 {
824 	struct netmsg_if_rtdel *rmsg = (void *)msg;
825 	int i, nextcpu, cpu;
826 
827 	cpu = mycpuid;
828 	for (i = 1; i <= AF_MAX; i++) {
829 		struct radix_node_head	*rnh;
830 
831 		if ((rnh = rt_tables[cpu][i]) == NULL)
832 			continue;
833 		rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
834 	}
835 
836 	nextcpu = cpu + 1;
837 	if (nextcpu < ncpus)
838 		lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg);
839 	else
840 		lwkt_replymsg(&rmsg->base.lmsg, 0);
841 }
842 
843 /*
844  * Detach an interface, removing it from the
845  * list of "active" interfaces.
846  */
847 void
848 if_detach(struct ifnet *ifp)
849 {
850 	struct ifnet_array *old_ifnet_array;
851 	struct netmsg_if_rtdel msg;
852 	struct domain *dp;
853 	int q;
854 
855 	/* Announce that the interface is gone. */
856 	EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
857 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
858 	devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
859 
860 	/*
861 	 * Remove this ifp from ifindex2inet, ifnet queue and ifnet
862 	 * array before it is whacked.
863 	 *
864 	 * Protect ifindex2ifnet, ifnet queue and ifnet array changes
865 	 * by ifnet lock, so that non-netisr threads could get a
866 	 * consistent view.
867 	 */
868 	ifnet_lock();
869 
870 	/*
871 	 * Remove this ifp from ifindex2ifnet and maybe decrement if_index.
872 	 */
873 	ifindex2ifnet[ifp->if_index] = NULL;
874 	while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
875 		if_index--;
876 
877 	/*
878 	 * Remove this ifp from ifnet queue.
879 	 */
880 	TAILQ_REMOVE(&ifnetlist, ifp, if_link);
881 
882 	/*
883 	 * Remove this ifp from ifnet array.
884 	 */
885 	/* Free old ifnet array after sync all netisrs */
886 	old_ifnet_array = ifnet_array;
887 	ifnet_array = ifnet_array_del(ifp, old_ifnet_array);
888 
889 	ifnet_unlock();
890 
891 	/*
892 	 * Sync all netisrs so that the old ifnet array is no longer
893 	 * accessed and we can free it safely later on.
894 	 */
895 	netmsg_service_sync();
896 	ifnet_array_free(old_ifnet_array);
897 
898 	/*
899 	 * Remove routes and flush queues.
900 	 */
901 	crit_enter();
902 #ifdef IFPOLL_ENABLE
903 	if (ifp->if_flags & IFF_NPOLLING)
904 		ifpoll_deregister(ifp);
905 #endif
906 	if_down(ifp);
907 
908 #ifdef ALTQ
909 	if (ifq_is_enabled(&ifp->if_snd))
910 		altq_disable(&ifp->if_snd);
911 	if (ifq_is_attached(&ifp->if_snd))
912 		altq_detach(&ifp->if_snd);
913 #endif
914 
915 	/*
916 	 * Clean up all addresses.
917 	 */
918 	ifp->if_lladdr = NULL;
919 
920 	if_purgeaddrs_nolink(ifp);
921 	if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
922 		struct ifaddr *ifa;
923 
924 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
925 		KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
926 			("non-link ifaddr is left on if_addrheads"));
927 
928 		ifa_ifunlink(ifa, ifp);
929 		ifa_destroy(ifa);
930 		KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
931 			("there are still ifaddrs left on if_addrheads"));
932 	}
933 
934 #ifdef INET
935 	/*
936 	 * Remove all IPv4 kernel structures related to ifp.
937 	 */
938 	in_ifdetach(ifp);
939 #endif
940 
941 #ifdef INET6
942 	/*
943 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
944 	 * before removing routing entries below, since IPv6 interface direct
945 	 * routes are expected to be removed by the IPv6-specific kernel API.
946 	 * Otherwise, the kernel will detect some inconsistency and bark it.
947 	 */
948 	in6_ifdetach(ifp);
949 #endif
950 
951 	/*
952 	 * Delete all remaining routes using this interface
953 	 */
954 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
955 	    if_rtdel_dispatch);
956 	msg.ifp = ifp;
957 	rt_domsg_global(&msg.base);
958 
959 	SLIST_FOREACH(dp, &domains, dom_next)
960 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
961 			(*dp->dom_ifdetach)(ifp,
962 				ifp->if_afdata[dp->dom_family]);
963 
964 	kfree(ifp->if_addrheads, M_IFADDR);
965 
966 	lwkt_synchronize_ipiqs("if_detach");
967 	ifq_stage_detach(&ifp->if_snd);
968 
969 	for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
970 		struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
971 
972 		kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
973 		kfree(ifsq->ifsq_stage, M_DEVBUF);
974 	}
975 	kfree(ifp->if_snd.altq_subq, M_DEVBUF);
976 
977 	kfree(ifp->if_data_pcpu, M_DEVBUF);
978 
979 	crit_exit();
980 }
981 
982 /*
983  * Create interface group without members
984  */
985 struct ifg_group *
986 if_creategroup(const char *groupname)
987 {
988         struct ifg_group        *ifg = NULL;
989 
990         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
991             M_TEMP, M_NOWAIT)) == NULL)
992                 return (NULL);
993 
994         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
995         ifg->ifg_refcnt = 0;
996         ifg->ifg_carp_demoted = 0;
997         TAILQ_INIT(&ifg->ifg_members);
998 #if NPF > 0
999         pfi_attach_ifgroup(ifg);
1000 #endif
1001         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
1002 
1003         return (ifg);
1004 }
1005 
1006 /*
1007  * Add a group to an interface
1008  */
1009 int
1010 if_addgroup(struct ifnet *ifp, const char *groupname)
1011 {
1012 	struct ifg_list		*ifgl;
1013 	struct ifg_group	*ifg = NULL;
1014 	struct ifg_member	*ifgm;
1015 
1016 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1017 	    groupname[strlen(groupname) - 1] <= '9')
1018 		return (EINVAL);
1019 
1020 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1021 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1022 			return (EEXIST);
1023 
1024 	if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
1025 		return (ENOMEM);
1026 
1027 	if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
1028 		kfree(ifgl, M_TEMP);
1029 		return (ENOMEM);
1030 	}
1031 
1032 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1033 		if (!strcmp(ifg->ifg_group, groupname))
1034 			break;
1035 
1036 	if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
1037 		kfree(ifgl, M_TEMP);
1038 		kfree(ifgm, M_TEMP);
1039 		return (ENOMEM);
1040 	}
1041 
1042 	ifg->ifg_refcnt++;
1043 	ifgl->ifgl_group = ifg;
1044 	ifgm->ifgm_ifp = ifp;
1045 
1046 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1047 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1048 
1049 #if NPF > 0
1050 	pfi_group_change(groupname);
1051 #endif
1052 
1053 	return (0);
1054 }
1055 
1056 /*
1057  * Remove a group from an interface
1058  */
1059 int
1060 if_delgroup(struct ifnet *ifp, const char *groupname)
1061 {
1062 	struct ifg_list		*ifgl;
1063 	struct ifg_member	*ifgm;
1064 
1065 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1066 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1067 			break;
1068 	if (ifgl == NULL)
1069 		return (ENOENT);
1070 
1071 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1072 
1073 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1074 		if (ifgm->ifgm_ifp == ifp)
1075 			break;
1076 
1077 	if (ifgm != NULL) {
1078 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1079 		kfree(ifgm, M_TEMP);
1080 	}
1081 
1082 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1083 		TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
1084 #if NPF > 0
1085 		pfi_detach_ifgroup(ifgl->ifgl_group);
1086 #endif
1087 		kfree(ifgl->ifgl_group, M_TEMP);
1088 	}
1089 
1090 	kfree(ifgl, M_TEMP);
1091 
1092 #if NPF > 0
1093 	pfi_group_change(groupname);
1094 #endif
1095 
1096 	return (0);
1097 }
1098 
1099 /*
1100  * Stores all groups from an interface in memory pointed
1101  * to by data
1102  */
1103 int
1104 if_getgroup(caddr_t data, struct ifnet *ifp)
1105 {
1106 	int			 len, error;
1107 	struct ifg_list		*ifgl;
1108 	struct ifg_req		 ifgrq, *ifgp;
1109 	struct ifgroupreq	*ifgr = (struct ifgroupreq *)data;
1110 
1111 	if (ifgr->ifgr_len == 0) {
1112 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1113 			ifgr->ifgr_len += sizeof(struct ifg_req);
1114 		return (0);
1115 	}
1116 
1117 	len = ifgr->ifgr_len;
1118 	ifgp = ifgr->ifgr_groups;
1119 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1120 		if (len < sizeof(ifgrq))
1121 			return (EINVAL);
1122 		bzero(&ifgrq, sizeof ifgrq);
1123 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1124 		    sizeof(ifgrq.ifgrq_group));
1125 		if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1126 		    sizeof(struct ifg_req))))
1127 			return (error);
1128 		len -= sizeof(ifgrq);
1129 		ifgp++;
1130 	}
1131 
1132 	return (0);
1133 }
1134 
1135 /*
1136  * Stores all members of a group in memory pointed to by data
1137  */
1138 int
1139 if_getgroupmembers(caddr_t data)
1140 {
1141 	struct ifgroupreq	*ifgr = (struct ifgroupreq *)data;
1142 	struct ifg_group	*ifg;
1143 	struct ifg_member	*ifgm;
1144 	struct ifg_req		 ifgrq, *ifgp;
1145 	int			 len, error;
1146 
1147 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1148 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1149 			break;
1150 	if (ifg == NULL)
1151 		return (ENOENT);
1152 
1153 	if (ifgr->ifgr_len == 0) {
1154 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1155 			ifgr->ifgr_len += sizeof(ifgrq);
1156 		return (0);
1157 	}
1158 
1159 	len = ifgr->ifgr_len;
1160 	ifgp = ifgr->ifgr_groups;
1161 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1162 		if (len < sizeof(ifgrq))
1163 			return (EINVAL);
1164 		bzero(&ifgrq, sizeof ifgrq);
1165 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1166 		    sizeof(ifgrq.ifgrq_member));
1167 		if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1168 		    sizeof(struct ifg_req))))
1169 			return (error);
1170 		len -= sizeof(ifgrq);
1171 		ifgp++;
1172 	}
1173 
1174 	return (0);
1175 }
1176 
1177 /*
1178  * Delete Routes for a Network Interface
1179  *
1180  * Called for each routing entry via the rnh->rnh_walktree() call above
1181  * to delete all route entries referencing a detaching network interface.
1182  *
1183  * Arguments:
1184  *	rn	pointer to node in the routing table
1185  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1186  *
1187  * Returns:
1188  *	0	successful
1189  *	errno	failed - reason indicated
1190  *
1191  */
1192 static int
1193 if_rtdel(struct radix_node *rn, void *arg)
1194 {
1195 	struct rtentry	*rt = (struct rtentry *)rn;
1196 	struct ifnet	*ifp = arg;
1197 	int		err;
1198 
1199 	if (rt->rt_ifp == ifp) {
1200 
1201 		/*
1202 		 * Protect (sorta) against walktree recursion problems
1203 		 * with cloned routes
1204 		 */
1205 		if (!(rt->rt_flags & RTF_UP))
1206 			return (0);
1207 
1208 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1209 				rt_mask(rt), rt->rt_flags,
1210 				NULL);
1211 		if (err) {
1212 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1213 		}
1214 	}
1215 
1216 	return (0);
1217 }
1218 
1219 /*
1220  * Locate an interface based on a complete address.
1221  */
1222 struct ifaddr *
1223 ifa_ifwithaddr(struct sockaddr *addr)
1224 {
1225 	const struct ifnet_array *arr;
1226 	int i;
1227 
1228 	arr = ifnet_array_get();
1229 	for (i = 0; i < arr->ifnet_count; ++i) {
1230 		struct ifnet *ifp = arr->ifnet_arr[i];
1231 		struct ifaddr_container *ifac;
1232 
1233 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1234 			struct ifaddr *ifa = ifac->ifa;
1235 
1236 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1237 				continue;
1238 			if (sa_equal(addr, ifa->ifa_addr))
1239 				return (ifa);
1240 			if ((ifp->if_flags & IFF_BROADCAST) &&
1241 			    ifa->ifa_broadaddr &&
1242 			    /* IPv6 doesn't have broadcast */
1243 			    ifa->ifa_broadaddr->sa_len != 0 &&
1244 			    sa_equal(ifa->ifa_broadaddr, addr))
1245 				return (ifa);
1246 		}
1247 	}
1248 	return (NULL);
1249 }
1250 /*
1251  * Locate the point to point interface with a given destination address.
1252  */
1253 struct ifaddr *
1254 ifa_ifwithdstaddr(struct sockaddr *addr)
1255 {
1256 	const struct ifnet_array *arr;
1257 	int i;
1258 
1259 	arr = ifnet_array_get();
1260 	for (i = 0; i < arr->ifnet_count; ++i) {
1261 		struct ifnet *ifp = arr->ifnet_arr[i];
1262 		struct ifaddr_container *ifac;
1263 
1264 		if (!(ifp->if_flags & IFF_POINTOPOINT))
1265 			continue;
1266 
1267 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1268 			struct ifaddr *ifa = ifac->ifa;
1269 
1270 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1271 				continue;
1272 			if (ifa->ifa_dstaddr &&
1273 			    sa_equal(addr, ifa->ifa_dstaddr))
1274 				return (ifa);
1275 		}
1276 	}
1277 	return (NULL);
1278 }
1279 
1280 /*
1281  * Find an interface on a specific network.  If many, choice
1282  * is most specific found.
1283  */
1284 struct ifaddr *
1285 ifa_ifwithnet(struct sockaddr *addr)
1286 {
1287 	struct ifaddr *ifa_maybe = NULL;
1288 	u_int af = addr->sa_family;
1289 	char *addr_data = addr->sa_data, *cplim;
1290 	const struct ifnet_array *arr;
1291 	int i;
1292 
1293 	/*
1294 	 * AF_LINK addresses can be looked up directly by their index number,
1295 	 * so do that if we can.
1296 	 */
1297 	if (af == AF_LINK) {
1298 		struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1299 
1300 		if (sdl->sdl_index && sdl->sdl_index <= if_index)
1301 			return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1302 	}
1303 
1304 	/*
1305 	 * Scan though each interface, looking for ones that have
1306 	 * addresses in this address family.
1307 	 */
1308 	arr = ifnet_array_get();
1309 	for (i = 0; i < arr->ifnet_count; ++i) {
1310 		struct ifnet *ifp = arr->ifnet_arr[i];
1311 		struct ifaddr_container *ifac;
1312 
1313 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1314 			struct ifaddr *ifa = ifac->ifa;
1315 			char *cp, *cp2, *cp3;
1316 
1317 			if (ifa->ifa_addr->sa_family != af)
1318 next:				continue;
1319 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1320 				/*
1321 				 * This is a bit broken as it doesn't
1322 				 * take into account that the remote end may
1323 				 * be a single node in the network we are
1324 				 * looking for.
1325 				 * The trouble is that we don't know the
1326 				 * netmask for the remote end.
1327 				 */
1328 				if (ifa->ifa_dstaddr != NULL &&
1329 				    sa_equal(addr, ifa->ifa_dstaddr))
1330 					return (ifa);
1331 			} else {
1332 				/*
1333 				 * if we have a special address handler,
1334 				 * then use it instead of the generic one.
1335 				 */
1336 				if (ifa->ifa_claim_addr) {
1337 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1338 						return (ifa);
1339 					} else {
1340 						continue;
1341 					}
1342 				}
1343 
1344 				/*
1345 				 * Scan all the bits in the ifa's address.
1346 				 * If a bit dissagrees with what we are
1347 				 * looking for, mask it with the netmask
1348 				 * to see if it really matters.
1349 				 * (A byte at a time)
1350 				 */
1351 				if (ifa->ifa_netmask == 0)
1352 					continue;
1353 				cp = addr_data;
1354 				cp2 = ifa->ifa_addr->sa_data;
1355 				cp3 = ifa->ifa_netmask->sa_data;
1356 				cplim = ifa->ifa_netmask->sa_len +
1357 					(char *)ifa->ifa_netmask;
1358 				while (cp3 < cplim)
1359 					if ((*cp++ ^ *cp2++) & *cp3++)
1360 						goto next; /* next address! */
1361 				/*
1362 				 * If the netmask of what we just found
1363 				 * is more specific than what we had before
1364 				 * (if we had one) then remember the new one
1365 				 * before continuing to search
1366 				 * for an even better one.
1367 				 */
1368 				if (ifa_maybe == NULL ||
1369 				    rn_refines((char *)ifa->ifa_netmask,
1370 					       (char *)ifa_maybe->ifa_netmask))
1371 					ifa_maybe = ifa;
1372 			}
1373 		}
1374 	}
1375 	return (ifa_maybe);
1376 }
1377 
1378 /*
1379  * Find an interface address specific to an interface best matching
1380  * a given address.
1381  */
1382 struct ifaddr *
1383 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1384 {
1385 	struct ifaddr_container *ifac;
1386 	char *cp, *cp2, *cp3;
1387 	char *cplim;
1388 	struct ifaddr *ifa_maybe = NULL;
1389 	u_int af = addr->sa_family;
1390 
1391 	if (af >= AF_MAX)
1392 		return (0);
1393 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1394 		struct ifaddr *ifa = ifac->ifa;
1395 
1396 		if (ifa->ifa_addr->sa_family != af)
1397 			continue;
1398 		if (ifa_maybe == NULL)
1399 			ifa_maybe = ifa;
1400 		if (ifa->ifa_netmask == NULL) {
1401 			if (sa_equal(addr, ifa->ifa_addr) ||
1402 			    (ifa->ifa_dstaddr != NULL &&
1403 			     sa_equal(addr, ifa->ifa_dstaddr)))
1404 				return (ifa);
1405 			continue;
1406 		}
1407 		if (ifp->if_flags & IFF_POINTOPOINT) {
1408 			if (sa_equal(addr, ifa->ifa_dstaddr))
1409 				return (ifa);
1410 		} else {
1411 			cp = addr->sa_data;
1412 			cp2 = ifa->ifa_addr->sa_data;
1413 			cp3 = ifa->ifa_netmask->sa_data;
1414 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1415 			for (; cp3 < cplim; cp3++)
1416 				if ((*cp++ ^ *cp2++) & *cp3)
1417 					break;
1418 			if (cp3 == cplim)
1419 				return (ifa);
1420 		}
1421 	}
1422 	return (ifa_maybe);
1423 }
1424 
1425 /*
1426  * Default action when installing a route with a Link Level gateway.
1427  * Lookup an appropriate real ifa to point to.
1428  * This should be moved to /sys/net/link.c eventually.
1429  */
1430 static void
1431 link_rtrequest(int cmd, struct rtentry *rt)
1432 {
1433 	struct ifaddr *ifa;
1434 	struct sockaddr *dst;
1435 	struct ifnet *ifp;
1436 
1437 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1438 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1439 		return;
1440 	ifa = ifaof_ifpforaddr(dst, ifp);
1441 	if (ifa != NULL) {
1442 		IFAFREE(rt->rt_ifa);
1443 		IFAREF(ifa);
1444 		rt->rt_ifa = ifa;
1445 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1446 			ifa->ifa_rtrequest(cmd, rt);
1447 	}
1448 }
1449 
1450 /*
1451  * Mark an interface down and notify protocols of
1452  * the transition.
1453  * NOTE: must be called at splnet or eqivalent.
1454  */
1455 void
1456 if_unroute(struct ifnet *ifp, int flag, int fam)
1457 {
1458 	struct ifaddr_container *ifac;
1459 
1460 	ifp->if_flags &= ~flag;
1461 	getmicrotime(&ifp->if_lastchange);
1462 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1463 		struct ifaddr *ifa = ifac->ifa;
1464 
1465 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1466 			kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1467 	}
1468 	ifq_purge_all(&ifp->if_snd);
1469 	rt_ifmsg(ifp);
1470 }
1471 
1472 /*
1473  * Mark an interface up and notify protocols of
1474  * the transition.
1475  * NOTE: must be called at splnet or eqivalent.
1476  */
1477 void
1478 if_route(struct ifnet *ifp, int flag, int fam)
1479 {
1480 	struct ifaddr_container *ifac;
1481 
1482 	ifq_purge_all(&ifp->if_snd);
1483 	ifp->if_flags |= flag;
1484 	getmicrotime(&ifp->if_lastchange);
1485 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1486 		struct ifaddr *ifa = ifac->ifa;
1487 
1488 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1489 			kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1490 	}
1491 	rt_ifmsg(ifp);
1492 #ifdef INET6
1493 	in6_if_up(ifp);
1494 #endif
1495 }
1496 
1497 /*
1498  * Mark an interface down and notify protocols of the transition.  An
1499  * interface going down is also considered to be a synchronizing event.
1500  * We must ensure that all packet processing related to the interface
1501  * has completed before we return so e.g. the caller can free the ifnet
1502  * structure that the mbufs may be referencing.
1503  *
1504  * NOTE: must be called at splnet or eqivalent.
1505  */
1506 void
1507 if_down(struct ifnet *ifp)
1508 {
1509 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1510 	netmsg_service_sync();
1511 }
1512 
1513 /*
1514  * Mark an interface up and notify protocols of
1515  * the transition.
1516  * NOTE: must be called at splnet or eqivalent.
1517  */
1518 void
1519 if_up(struct ifnet *ifp)
1520 {
1521 	if_route(ifp, IFF_UP, AF_UNSPEC);
1522 }
1523 
1524 /*
1525  * Process a link state change.
1526  * NOTE: must be called at splsoftnet or equivalent.
1527  */
1528 void
1529 if_link_state_change(struct ifnet *ifp)
1530 {
1531 	int link_state = ifp->if_link_state;
1532 
1533 	rt_ifmsg(ifp);
1534 	devctl_notify("IFNET", ifp->if_xname,
1535 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1536 }
1537 
1538 /*
1539  * Handle interface watchdog timer routines.  Called
1540  * from softclock, we decrement timers (if set) and
1541  * call the appropriate interface routine on expiration.
1542  */
1543 static void
1544 if_slowtimo_dispatch(netmsg_t nmsg)
1545 {
1546 	struct globaldata *gd = mycpu;
1547 	const struct ifnet_array *arr;
1548 	int i;
1549 
1550 	KASSERT(&curthread->td_msgport == netisr_cpuport(0),
1551 	    ("not in netisr0"));
1552 
1553 	crit_enter_gd(gd);
1554 	lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1555 	crit_exit_gd(gd);
1556 
1557 	arr = ifnet_array_get();
1558 	for (i = 0; i < arr->ifnet_count; ++i) {
1559 		struct ifnet *ifp = arr->ifnet_arr[i];
1560 
1561 		crit_enter_gd(gd);
1562 
1563 		if (if_stats_compat) {
1564 			IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1565 			IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1566 			IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1567 			IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1568 			IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1569 			IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1570 			IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1571 			IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1572 			IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1573 			IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1574 			IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1575 		}
1576 
1577 		if (ifp->if_timer == 0 || --ifp->if_timer) {
1578 			crit_exit_gd(gd);
1579 			continue;
1580 		}
1581 		if (ifp->if_watchdog) {
1582 			if (ifnet_tryserialize_all(ifp)) {
1583 				(*ifp->if_watchdog)(ifp);
1584 				ifnet_deserialize_all(ifp);
1585 			} else {
1586 				/* try again next timeout */
1587 				++ifp->if_timer;
1588 			}
1589 		}
1590 
1591 		crit_exit_gd(gd);
1592 	}
1593 
1594 	callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1595 }
1596 
1597 static void
1598 if_slowtimo(void *arg __unused)
1599 {
1600 	struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1601 
1602 	KASSERT(mycpuid == 0, ("not on cpu0"));
1603 	crit_enter();
1604 	if (lmsg->ms_flags & MSGF_DONE)
1605 		lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1606 	crit_exit();
1607 }
1608 
1609 /*
1610  * Map interface name to
1611  * interface structure pointer.
1612  */
1613 struct ifnet *
1614 ifunit(const char *name)
1615 {
1616 	struct ifnet *ifp;
1617 
1618 	/*
1619 	 * Search all the interfaces for this name/number
1620 	 */
1621 	KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1622 
1623 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1624 		if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1625 			break;
1626 	}
1627 	return (ifp);
1628 }
1629 
1630 struct ifnet *
1631 ifunit_netisr(const char *name)
1632 {
1633 	const struct ifnet_array *arr;
1634 	int i;
1635 
1636 	/*
1637 	 * Search all the interfaces for this name/number
1638 	 */
1639 
1640 	arr = ifnet_array_get();
1641 	for (i = 0; i < arr->ifnet_count; ++i) {
1642 		struct ifnet *ifp = arr->ifnet_arr[i];
1643 
1644 		if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1645 			return ifp;
1646 	}
1647 	return NULL;
1648 }
1649 
1650 /*
1651  * Interface ioctls.
1652  */
1653 int
1654 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1655 {
1656 	struct ifnet *ifp;
1657 	struct ifreq *ifr;
1658 	struct ifstat *ifs;
1659 	int error;
1660 	short oif_flags;
1661 	int new_flags;
1662 #ifdef COMPAT_43
1663 	int ocmd;
1664 #endif
1665 	size_t namelen, onamelen;
1666 	char new_name[IFNAMSIZ];
1667 	struct ifaddr *ifa;
1668 	struct sockaddr_dl *sdl;
1669 
1670 	switch (cmd) {
1671 	case SIOCGIFCONF:
1672 	case OSIOCGIFCONF:
1673 		return (ifconf(cmd, data, cred));
1674 	default:
1675 		break;
1676 	}
1677 
1678 	ifr = (struct ifreq *)data;
1679 
1680 	switch (cmd) {
1681 	case SIOCIFCREATE:
1682 	case SIOCIFCREATE2:
1683 		if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1684 			return (error);
1685 		return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1686 		    	cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1687 	case SIOCIFDESTROY:
1688 		if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1689 			return (error);
1690 		return (if_clone_destroy(ifr->ifr_name));
1691 	case SIOCIFGCLONERS:
1692 		return (if_clone_list((struct if_clonereq *)data));
1693 	default:
1694 		break;
1695 	}
1696 
1697 	/*
1698 	 * Nominal ioctl through interface, lookup the ifp and obtain a
1699 	 * lock to serialize the ifconfig ioctl operation.
1700 	 */
1701 	ifnet_lock();
1702 
1703 	ifp = ifunit(ifr->ifr_name);
1704 	if (ifp == NULL) {
1705 		ifnet_unlock();
1706 		return (ENXIO);
1707 	}
1708 	error = 0;
1709 
1710 	switch (cmd) {
1711 	case SIOCGIFINDEX:
1712 		ifr->ifr_index = ifp->if_index;
1713 		break;
1714 
1715 	case SIOCGIFFLAGS:
1716 		ifr->ifr_flags = ifp->if_flags;
1717 		ifr->ifr_flagshigh = ifp->if_flags >> 16;
1718 		break;
1719 
1720 	case SIOCGIFCAP:
1721 		ifr->ifr_reqcap = ifp->if_capabilities;
1722 		ifr->ifr_curcap = ifp->if_capenable;
1723 		break;
1724 
1725 	case SIOCGIFMETRIC:
1726 		ifr->ifr_metric = ifp->if_metric;
1727 		break;
1728 
1729 	case SIOCGIFMTU:
1730 		ifr->ifr_mtu = ifp->if_mtu;
1731 		break;
1732 
1733 	case SIOCGIFTSOLEN:
1734 		ifr->ifr_tsolen = ifp->if_tsolen;
1735 		break;
1736 
1737 	case SIOCGIFDATA:
1738 		error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1739 				sizeof(ifp->if_data));
1740 		break;
1741 
1742 	case SIOCGIFPHYS:
1743 		ifr->ifr_phys = ifp->if_physical;
1744 		break;
1745 
1746 	case SIOCGIFPOLLCPU:
1747 		ifr->ifr_pollcpu = -1;
1748 		break;
1749 
1750 	case SIOCSIFPOLLCPU:
1751 		break;
1752 
1753 	case SIOCSIFFLAGS:
1754 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1755 		if (error)
1756 			break;
1757 		new_flags = (ifr->ifr_flags & 0xffff) |
1758 		    (ifr->ifr_flagshigh << 16);
1759 		if (ifp->if_flags & IFF_SMART) {
1760 			/* Smart drivers twiddle their own routes */
1761 		} else if (ifp->if_flags & IFF_UP &&
1762 		    (new_flags & IFF_UP) == 0) {
1763 			crit_enter();
1764 			if_down(ifp);
1765 			crit_exit();
1766 		} else if (new_flags & IFF_UP &&
1767 		    (ifp->if_flags & IFF_UP) == 0) {
1768 			crit_enter();
1769 			if_up(ifp);
1770 			crit_exit();
1771 		}
1772 
1773 #ifdef IFPOLL_ENABLE
1774 		if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1775 			if (new_flags & IFF_NPOLLING)
1776 				ifpoll_register(ifp);
1777 			else
1778 				ifpoll_deregister(ifp);
1779 		}
1780 #endif
1781 
1782 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1783 			(new_flags &~ IFF_CANTCHANGE);
1784 		if (new_flags & IFF_PPROMISC) {
1785 			/* Permanently promiscuous mode requested */
1786 			ifp->if_flags |= IFF_PROMISC;
1787 		} else if (ifp->if_pcount == 0) {
1788 			ifp->if_flags &= ~IFF_PROMISC;
1789 		}
1790 		if (ifp->if_ioctl) {
1791 			ifnet_serialize_all(ifp);
1792 			ifp->if_ioctl(ifp, cmd, data, cred);
1793 			ifnet_deserialize_all(ifp);
1794 		}
1795 		getmicrotime(&ifp->if_lastchange);
1796 		break;
1797 
1798 	case SIOCSIFCAP:
1799 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1800 		if (error)
1801 			break;
1802 		if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1803 			error = EINVAL;
1804 			break;
1805 		}
1806 		ifnet_serialize_all(ifp);
1807 		ifp->if_ioctl(ifp, cmd, data, cred);
1808 		ifnet_deserialize_all(ifp);
1809 		break;
1810 
1811 	case SIOCSIFNAME:
1812 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1813 		if (error)
1814 			break;
1815 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1816 		if (error)
1817 			break;
1818 		if (new_name[0] == '\0') {
1819 			error = EINVAL;
1820 			break;
1821 		}
1822 		if (ifunit(new_name) != NULL) {
1823 			error = EEXIST;
1824 			break;
1825 		}
1826 
1827 		EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1828 
1829 		/* Announce the departure of the interface. */
1830 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1831 
1832 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1833 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1834 		/* XXX IFA_LOCK(ifa); */
1835 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1836 		namelen = strlen(new_name);
1837 		onamelen = sdl->sdl_nlen;
1838 		/*
1839 		 * Move the address if needed.  This is safe because we
1840 		 * allocate space for a name of length IFNAMSIZ when we
1841 		 * create this in if_attach().
1842 		 */
1843 		if (namelen != onamelen) {
1844 			bcopy(sdl->sdl_data + onamelen,
1845 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1846 		}
1847 		bcopy(new_name, sdl->sdl_data, namelen);
1848 		sdl->sdl_nlen = namelen;
1849 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1850 		bzero(sdl->sdl_data, onamelen);
1851 		while (namelen != 0)
1852 			sdl->sdl_data[--namelen] = 0xff;
1853 		/* XXX IFA_UNLOCK(ifa) */
1854 
1855 		EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1856 
1857 		/* Announce the return of the interface. */
1858 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1859 		break;
1860 
1861 	case SIOCSIFMETRIC:
1862 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1863 		if (error)
1864 			break;
1865 		ifp->if_metric = ifr->ifr_metric;
1866 		getmicrotime(&ifp->if_lastchange);
1867 		break;
1868 
1869 	case SIOCSIFPHYS:
1870 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1871 		if (error)
1872 			break;
1873 		if (ifp->if_ioctl == NULL) {
1874 		        error = EOPNOTSUPP;
1875 			break;
1876 		}
1877 		ifnet_serialize_all(ifp);
1878 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1879 		ifnet_deserialize_all(ifp);
1880 		if (error == 0)
1881 			getmicrotime(&ifp->if_lastchange);
1882 		break;
1883 
1884 	case SIOCSIFMTU:
1885 	{
1886 		u_long oldmtu = ifp->if_mtu;
1887 
1888 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1889 		if (error)
1890 			break;
1891 		if (ifp->if_ioctl == NULL) {
1892 			error = EOPNOTSUPP;
1893 			break;
1894 		}
1895 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
1896 			error = EINVAL;
1897 			break;
1898 		}
1899 		ifnet_serialize_all(ifp);
1900 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1901 		ifnet_deserialize_all(ifp);
1902 		if (error == 0) {
1903 			getmicrotime(&ifp->if_lastchange);
1904 			rt_ifmsg(ifp);
1905 		}
1906 		/*
1907 		 * If the link MTU changed, do network layer specific procedure.
1908 		 */
1909 		if (ifp->if_mtu != oldmtu) {
1910 #ifdef INET6
1911 			nd6_setmtu(ifp);
1912 #endif
1913 		}
1914 		break;
1915 	}
1916 
1917 	case SIOCSIFTSOLEN:
1918 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1919 		if (error)
1920 			break;
1921 
1922 		/* XXX need driver supplied upper limit */
1923 		if (ifr->ifr_tsolen <= 0) {
1924 			error = EINVAL;
1925 			break;
1926 		}
1927 		ifp->if_tsolen = ifr->ifr_tsolen;
1928 		break;
1929 
1930 	case SIOCADDMULTI:
1931 	case SIOCDELMULTI:
1932 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1933 		if (error)
1934 			break;
1935 
1936 		/* Don't allow group membership on non-multicast interfaces. */
1937 		if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1938 			error = EOPNOTSUPP;
1939 			break;
1940 		}
1941 
1942 		/* Don't let users screw up protocols' entries. */
1943 		if (ifr->ifr_addr.sa_family != AF_LINK) {
1944 			error = EINVAL;
1945 			break;
1946 		}
1947 
1948 		if (cmd == SIOCADDMULTI) {
1949 			struct ifmultiaddr *ifma;
1950 			error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1951 		} else {
1952 			error = if_delmulti(ifp, &ifr->ifr_addr);
1953 		}
1954 		if (error == 0)
1955 			getmicrotime(&ifp->if_lastchange);
1956 		break;
1957 
1958 	case SIOCSIFPHYADDR:
1959 	case SIOCDIFPHYADDR:
1960 #ifdef INET6
1961 	case SIOCSIFPHYADDR_IN6:
1962 #endif
1963 	case SIOCSLIFPHYADDR:
1964         case SIOCSIFMEDIA:
1965 	case SIOCSIFGENERIC:
1966 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1967 		if (error)
1968 			break;
1969 		if (ifp->if_ioctl == 0) {
1970 			error = EOPNOTSUPP;
1971 			break;
1972 		}
1973 		ifnet_serialize_all(ifp);
1974 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1975 		ifnet_deserialize_all(ifp);
1976 		if (error == 0)
1977 			getmicrotime(&ifp->if_lastchange);
1978 		break;
1979 
1980 	case SIOCGIFSTATUS:
1981 		ifs = (struct ifstat *)data;
1982 		ifs->ascii[0] = '\0';
1983 		/* fall through */
1984 	case SIOCGIFPSRCADDR:
1985 	case SIOCGIFPDSTADDR:
1986 	case SIOCGLIFPHYADDR:
1987 	case SIOCGIFMEDIA:
1988 	case SIOCGIFGENERIC:
1989 		if (ifp->if_ioctl == NULL) {
1990 			error = EOPNOTSUPP;
1991 			break;
1992 		}
1993 		ifnet_serialize_all(ifp);
1994 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1995 		ifnet_deserialize_all(ifp);
1996 		break;
1997 
1998 	case SIOCSIFLLADDR:
1999 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2000 		if (error)
2001 			break;
2002 		error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2003 				     ifr->ifr_addr.sa_len);
2004 		EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2005 		break;
2006 
2007 	default:
2008 		oif_flags = ifp->if_flags;
2009 		if (so->so_proto == 0) {
2010 			error = EOPNOTSUPP;
2011 			break;
2012 		}
2013 #ifndef COMPAT_43
2014 		error = so_pru_control_direct(so, cmd, data, ifp);
2015 #else
2016 		ocmd = cmd;
2017 
2018 		switch (cmd) {
2019 		case SIOCSIFDSTADDR:
2020 		case SIOCSIFADDR:
2021 		case SIOCSIFBRDADDR:
2022 		case SIOCSIFNETMASK:
2023 #if BYTE_ORDER != BIG_ENDIAN
2024 			if (ifr->ifr_addr.sa_family == 0 &&
2025 			    ifr->ifr_addr.sa_len < 16) {
2026 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2027 				ifr->ifr_addr.sa_len = 16;
2028 			}
2029 #else
2030 			if (ifr->ifr_addr.sa_len == 0)
2031 				ifr->ifr_addr.sa_len = 16;
2032 #endif
2033 			break;
2034 		case OSIOCGIFADDR:
2035 			cmd = SIOCGIFADDR;
2036 			break;
2037 		case OSIOCGIFDSTADDR:
2038 			cmd = SIOCGIFDSTADDR;
2039 			break;
2040 		case OSIOCGIFBRDADDR:
2041 			cmd = SIOCGIFBRDADDR;
2042 			break;
2043 		case OSIOCGIFNETMASK:
2044 			cmd = SIOCGIFNETMASK;
2045 			break;
2046 		default:
2047 			break;
2048 		}
2049 
2050 		error = so_pru_control_direct(so, cmd, data, ifp);
2051 
2052 		switch (ocmd) {
2053 		case OSIOCGIFADDR:
2054 		case OSIOCGIFDSTADDR:
2055 		case OSIOCGIFBRDADDR:
2056 		case OSIOCGIFNETMASK:
2057 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2058 			break;
2059 		}
2060 #endif /* COMPAT_43 */
2061 
2062 		if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2063 #ifdef INET6
2064 			DELAY(100);/* XXX: temporary workaround for fxp issue*/
2065 			if (ifp->if_flags & IFF_UP) {
2066 				crit_enter();
2067 				in6_if_up(ifp);
2068 				crit_exit();
2069 			}
2070 #endif
2071 		}
2072 		break;
2073 	}
2074 
2075 	ifnet_unlock();
2076 	return (error);
2077 }
2078 
2079 /*
2080  * Set/clear promiscuous mode on interface ifp based on the truth value
2081  * of pswitch.  The calls are reference counted so that only the first
2082  * "on" request actually has an effect, as does the final "off" request.
2083  * Results are undefined if the "off" and "on" requests are not matched.
2084  */
2085 int
2086 ifpromisc(struct ifnet *ifp, int pswitch)
2087 {
2088 	struct ifreq ifr;
2089 	int error;
2090 	int oldflags;
2091 
2092 	oldflags = ifp->if_flags;
2093 	if (ifp->if_flags & IFF_PPROMISC) {
2094 		/* Do nothing if device is in permanently promiscuous mode */
2095 		ifp->if_pcount += pswitch ? 1 : -1;
2096 		return (0);
2097 	}
2098 	if (pswitch) {
2099 		/*
2100 		 * If the device is not configured up, we cannot put it in
2101 		 * promiscuous mode.
2102 		 */
2103 		if ((ifp->if_flags & IFF_UP) == 0)
2104 			return (ENETDOWN);
2105 		if (ifp->if_pcount++ != 0)
2106 			return (0);
2107 		ifp->if_flags |= IFF_PROMISC;
2108 		log(LOG_INFO, "%s: promiscuous mode enabled\n",
2109 		    ifp->if_xname);
2110 	} else {
2111 		if (--ifp->if_pcount > 0)
2112 			return (0);
2113 		ifp->if_flags &= ~IFF_PROMISC;
2114 		log(LOG_INFO, "%s: promiscuous mode disabled\n",
2115 		    ifp->if_xname);
2116 	}
2117 	ifr.ifr_flags = ifp->if_flags;
2118 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2119 	ifnet_serialize_all(ifp);
2120 	error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2121 	ifnet_deserialize_all(ifp);
2122 	if (error == 0)
2123 		rt_ifmsg(ifp);
2124 	else
2125 		ifp->if_flags = oldflags;
2126 	return error;
2127 }
2128 
2129 /*
2130  * Return interface configuration
2131  * of system.  List may be used
2132  * in later ioctl's (above) to get
2133  * other information.
2134  */
2135 static int
2136 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2137 {
2138 	struct ifconf *ifc = (struct ifconf *)data;
2139 	struct ifnet *ifp;
2140 	struct sockaddr *sa;
2141 	struct ifreq ifr, *ifrp;
2142 	int space = ifc->ifc_len, error = 0;
2143 
2144 	ifrp = ifc->ifc_req;
2145 
2146 	ifnet_lock();
2147 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2148 		struct ifaddr_container *ifac;
2149 		int addrs;
2150 
2151 		if (space <= sizeof ifr)
2152 			break;
2153 
2154 		/*
2155 		 * Zero the stack declared structure first to prevent
2156 		 * memory disclosure.
2157 		 */
2158 		bzero(&ifr, sizeof(ifr));
2159 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2160 		    >= sizeof(ifr.ifr_name)) {
2161 			error = ENAMETOOLONG;
2162 			break;
2163 		}
2164 
2165 		addrs = 0;
2166 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2167 			struct ifaddr *ifa = ifac->ifa;
2168 
2169 			if (space <= sizeof ifr)
2170 				break;
2171 			sa = ifa->ifa_addr;
2172 			if (cred->cr_prison &&
2173 			    prison_if(cred, sa))
2174 				continue;
2175 			addrs++;
2176 #ifdef COMPAT_43
2177 			if (cmd == OSIOCGIFCONF) {
2178 				struct osockaddr *osa =
2179 					 (struct osockaddr *)&ifr.ifr_addr;
2180 				ifr.ifr_addr = *sa;
2181 				osa->sa_family = sa->sa_family;
2182 				error = copyout(&ifr, ifrp, sizeof ifr);
2183 				ifrp++;
2184 			} else
2185 #endif
2186 			if (sa->sa_len <= sizeof(*sa)) {
2187 				ifr.ifr_addr = *sa;
2188 				error = copyout(&ifr, ifrp, sizeof ifr);
2189 				ifrp++;
2190 			} else {
2191 				if (space < (sizeof ifr) + sa->sa_len -
2192 					    sizeof(*sa))
2193 					break;
2194 				space -= sa->sa_len - sizeof(*sa);
2195 				error = copyout(&ifr, ifrp,
2196 						sizeof ifr.ifr_name);
2197 				if (error == 0)
2198 					error = copyout(sa, &ifrp->ifr_addr,
2199 							sa->sa_len);
2200 				ifrp = (struct ifreq *)
2201 					(sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2202 			}
2203 			if (error)
2204 				break;
2205 			space -= sizeof ifr;
2206 		}
2207 		if (error)
2208 			break;
2209 		if (!addrs) {
2210 			bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2211 			error = copyout(&ifr, ifrp, sizeof ifr);
2212 			if (error)
2213 				break;
2214 			space -= sizeof ifr;
2215 			ifrp++;
2216 		}
2217 	}
2218 	ifnet_unlock();
2219 
2220 	ifc->ifc_len -= space;
2221 	return (error);
2222 }
2223 
2224 /*
2225  * Just like if_promisc(), but for all-multicast-reception mode.
2226  */
2227 int
2228 if_allmulti(struct ifnet *ifp, int onswitch)
2229 {
2230 	int error = 0;
2231 	struct ifreq ifr;
2232 
2233 	crit_enter();
2234 
2235 	if (onswitch) {
2236 		if (ifp->if_amcount++ == 0) {
2237 			ifp->if_flags |= IFF_ALLMULTI;
2238 			ifr.ifr_flags = ifp->if_flags;
2239 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2240 			ifnet_serialize_all(ifp);
2241 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2242 					      NULL);
2243 			ifnet_deserialize_all(ifp);
2244 		}
2245 	} else {
2246 		if (ifp->if_amcount > 1) {
2247 			ifp->if_amcount--;
2248 		} else {
2249 			ifp->if_amcount = 0;
2250 			ifp->if_flags &= ~IFF_ALLMULTI;
2251 			ifr.ifr_flags = ifp->if_flags;
2252 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2253 			ifnet_serialize_all(ifp);
2254 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2255 					      NULL);
2256 			ifnet_deserialize_all(ifp);
2257 		}
2258 	}
2259 
2260 	crit_exit();
2261 
2262 	if (error == 0)
2263 		rt_ifmsg(ifp);
2264 	return error;
2265 }
2266 
2267 /*
2268  * Add a multicast listenership to the interface in question.
2269  * The link layer provides a routine which converts
2270  */
2271 int
2272 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2273     struct ifmultiaddr **retifma)
2274 {
2275 	struct sockaddr *llsa, *dupsa;
2276 	int error;
2277 	struct ifmultiaddr *ifma;
2278 
2279 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
2280 
2281 	/*
2282 	 * If the matching multicast address already exists
2283 	 * then don't add a new one, just add a reference
2284 	 */
2285 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2286 		if (sa_equal(sa, ifma->ifma_addr)) {
2287 			ifma->ifma_refcount++;
2288 			if (retifma)
2289 				*retifma = ifma;
2290 			return 0;
2291 		}
2292 	}
2293 
2294 	/*
2295 	 * Give the link layer a chance to accept/reject it, and also
2296 	 * find out which AF_LINK address this maps to, if it isn't one
2297 	 * already.
2298 	 */
2299 	if (ifp->if_resolvemulti) {
2300 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2301 		if (error)
2302 			return error;
2303 	} else {
2304 		llsa = NULL;
2305 	}
2306 
2307 	ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2308 	dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_WAITOK);
2309 	bcopy(sa, dupsa, sa->sa_len);
2310 
2311 	ifma->ifma_addr = dupsa;
2312 	ifma->ifma_lladdr = llsa;
2313 	ifma->ifma_ifp = ifp;
2314 	ifma->ifma_refcount = 1;
2315 	ifma->ifma_protospec = NULL;
2316 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2317 
2318 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2319 	if (retifma)
2320 		*retifma = ifma;
2321 
2322 	if (llsa != NULL) {
2323 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2324 			if (sa_equal(ifma->ifma_addr, llsa))
2325 				break;
2326 		}
2327 		if (ifma) {
2328 			ifma->ifma_refcount++;
2329 		} else {
2330 			ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2331 			dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_WAITOK);
2332 			bcopy(llsa, dupsa, llsa->sa_len);
2333 			ifma->ifma_addr = dupsa;
2334 			ifma->ifma_ifp = ifp;
2335 			ifma->ifma_refcount = 1;
2336 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2337 		}
2338 	}
2339 	/*
2340 	 * We are certain we have added something, so call down to the
2341 	 * interface to let them know about it.
2342 	 */
2343 	if (ifp->if_ioctl)
2344 		ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2345 
2346 	return 0;
2347 }
2348 
2349 int
2350 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2351     struct ifmultiaddr **retifma)
2352 {
2353 	int error;
2354 
2355 	ifnet_serialize_all(ifp);
2356 	error = if_addmulti_serialized(ifp, sa, retifma);
2357 	ifnet_deserialize_all(ifp);
2358 
2359 	return error;
2360 }
2361 
2362 /*
2363  * Remove a reference to a multicast address on this interface.  Yell
2364  * if the request does not match an existing membership.
2365  */
2366 static int
2367 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2368 {
2369 	struct ifmultiaddr *ifma;
2370 
2371 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
2372 
2373 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2374 		if (sa_equal(sa, ifma->ifma_addr))
2375 			break;
2376 	if (ifma == NULL)
2377 		return ENOENT;
2378 
2379 	if (ifma->ifma_refcount > 1) {
2380 		ifma->ifma_refcount--;
2381 		return 0;
2382 	}
2383 
2384 	rt_newmaddrmsg(RTM_DELMADDR, ifma);
2385 	sa = ifma->ifma_lladdr;
2386 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2387 	/*
2388 	 * Make sure the interface driver is notified
2389 	 * in the case of a link layer mcast group being left.
2390 	 */
2391 	if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2392 		ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2393 	kfree(ifma->ifma_addr, M_IFMADDR);
2394 	kfree(ifma, M_IFMADDR);
2395 	if (sa == NULL)
2396 		return 0;
2397 
2398 	/*
2399 	 * Now look for the link-layer address which corresponds to
2400 	 * this network address.  It had been squirreled away in
2401 	 * ifma->ifma_lladdr for this purpose (so we don't have
2402 	 * to call ifp->if_resolvemulti() again), and we saved that
2403 	 * value in sa above.  If some nasty deleted the
2404 	 * link-layer address out from underneath us, we can deal because
2405 	 * the address we stored was is not the same as the one which was
2406 	 * in the record for the link-layer address.  (So we don't complain
2407 	 * in that case.)
2408 	 */
2409 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2410 		if (sa_equal(sa, ifma->ifma_addr))
2411 			break;
2412 	if (ifma == NULL)
2413 		return 0;
2414 
2415 	if (ifma->ifma_refcount > 1) {
2416 		ifma->ifma_refcount--;
2417 		return 0;
2418 	}
2419 
2420 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2421 	ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2422 	kfree(ifma->ifma_addr, M_IFMADDR);
2423 	kfree(sa, M_IFMADDR);
2424 	kfree(ifma, M_IFMADDR);
2425 
2426 	return 0;
2427 }
2428 
2429 int
2430 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2431 {
2432 	int error;
2433 
2434 	ifnet_serialize_all(ifp);
2435 	error = if_delmulti_serialized(ifp, sa);
2436 	ifnet_deserialize_all(ifp);
2437 
2438 	return error;
2439 }
2440 
2441 /*
2442  * Delete all multicast group membership for an interface.
2443  * Should be used to quickly flush all multicast filters.
2444  */
2445 void
2446 if_delallmulti_serialized(struct ifnet *ifp)
2447 {
2448 	struct ifmultiaddr *ifma, mark;
2449 	struct sockaddr sa;
2450 
2451 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
2452 
2453 	bzero(&sa, sizeof(sa));
2454 	sa.sa_family = AF_UNSPEC;
2455 	sa.sa_len = sizeof(sa);
2456 
2457 	bzero(&mark, sizeof(mark));
2458 	mark.ifma_addr = &sa;
2459 
2460 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2461 	while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2462 		TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2463 		TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2464 		    ifma_link);
2465 
2466 		if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2467 			continue;
2468 
2469 		if_delmulti_serialized(ifp, ifma->ifma_addr);
2470 	}
2471 	TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2472 }
2473 
2474 
2475 /*
2476  * Set the link layer address on an interface.
2477  *
2478  * At this time we only support certain types of interfaces,
2479  * and we don't allow the length of the address to change.
2480  */
2481 int
2482 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2483 {
2484 	struct sockaddr_dl *sdl;
2485 	struct ifreq ifr;
2486 
2487 	sdl = IF_LLSOCKADDR(ifp);
2488 	if (sdl == NULL)
2489 		return (EINVAL);
2490 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2491 		return (EINVAL);
2492 	switch (ifp->if_type) {
2493 	case IFT_ETHER:			/* these types use struct arpcom */
2494 	case IFT_XETHER:
2495 	case IFT_L2VLAN:
2496 	case IFT_IEEE8023ADLAG:
2497 		bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2498 		bcopy(lladdr, LLADDR(sdl), len);
2499 		break;
2500 	default:
2501 		return (ENODEV);
2502 	}
2503 	/*
2504 	 * If the interface is already up, we need
2505 	 * to re-init it in order to reprogram its
2506 	 * address filter.
2507 	 */
2508 	ifnet_serialize_all(ifp);
2509 	if ((ifp->if_flags & IFF_UP) != 0) {
2510 #ifdef INET
2511 		struct ifaddr_container *ifac;
2512 #endif
2513 
2514 		ifp->if_flags &= ~IFF_UP;
2515 		ifr.ifr_flags = ifp->if_flags;
2516 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
2517 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2518 			      NULL);
2519 		ifp->if_flags |= IFF_UP;
2520 		ifr.ifr_flags = ifp->if_flags;
2521 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
2522 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2523 				 NULL);
2524 #ifdef INET
2525 		/*
2526 		 * Also send gratuitous ARPs to notify other nodes about
2527 		 * the address change.
2528 		 */
2529 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2530 			struct ifaddr *ifa = ifac->ifa;
2531 
2532 			if (ifa->ifa_addr != NULL &&
2533 			    ifa->ifa_addr->sa_family == AF_INET)
2534 				arp_gratuitous(ifp, ifa);
2535 		}
2536 #endif
2537 	}
2538 	ifnet_deserialize_all(ifp);
2539 	return (0);
2540 }
2541 
2542 struct ifmultiaddr *
2543 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2544 {
2545 	struct ifmultiaddr *ifma;
2546 
2547 	/* TODO: need ifnet_serialize_main */
2548 	ifnet_serialize_all(ifp);
2549 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2550 		if (sa_equal(ifma->ifma_addr, sa))
2551 			break;
2552 	ifnet_deserialize_all(ifp);
2553 
2554 	return ifma;
2555 }
2556 
2557 /*
2558  * This function locates the first real ethernet MAC from a network
2559  * card and loads it into node, returning 0 on success or ENOENT if
2560  * no suitable interfaces were found.  It is used by the uuid code to
2561  * generate a unique 6-byte number.
2562  */
2563 int
2564 if_getanyethermac(uint16_t *node, int minlen)
2565 {
2566 	struct ifnet *ifp;
2567 	struct sockaddr_dl *sdl;
2568 
2569 	ifnet_lock();
2570 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2571 		if (ifp->if_type != IFT_ETHER)
2572 			continue;
2573 		sdl = IF_LLSOCKADDR(ifp);
2574 		if (sdl->sdl_alen < minlen)
2575 			continue;
2576 		bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2577 		      minlen);
2578 		ifnet_unlock();
2579 		return(0);
2580 	}
2581 	ifnet_unlock();
2582 	return (ENOENT);
2583 }
2584 
2585 /*
2586  * The name argument must be a pointer to storage which will last as
2587  * long as the interface does.  For physical devices, the result of
2588  * device_get_name(dev) is a good choice and for pseudo-devices a
2589  * static string works well.
2590  */
2591 void
2592 if_initname(struct ifnet *ifp, const char *name, int unit)
2593 {
2594 	ifp->if_dname = name;
2595 	ifp->if_dunit = unit;
2596 	if (unit != IF_DUNIT_NONE)
2597 		ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2598 	else
2599 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2600 }
2601 
2602 int
2603 if_printf(struct ifnet *ifp, const char *fmt, ...)
2604 {
2605 	__va_list ap;
2606 	int retval;
2607 
2608 	retval = kprintf("%s: ", ifp->if_xname);
2609 	__va_start(ap, fmt);
2610 	retval += kvprintf(fmt, ap);
2611 	__va_end(ap);
2612 	return (retval);
2613 }
2614 
2615 struct ifnet *
2616 if_alloc(uint8_t type)
2617 {
2618         struct ifnet *ifp;
2619 	size_t size;
2620 
2621 	/*
2622 	 * XXX temporary hack until arpcom is setup in if_l2com
2623 	 */
2624 	if (type == IFT_ETHER)
2625 		size = sizeof(struct arpcom);
2626 	else
2627 		size = sizeof(struct ifnet);
2628 
2629 	ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2630 
2631 	ifp->if_type = type;
2632 
2633 	if (if_com_alloc[type] != NULL) {
2634 		ifp->if_l2com = if_com_alloc[type](type, ifp);
2635 		if (ifp->if_l2com == NULL) {
2636 			kfree(ifp, M_IFNET);
2637 			return (NULL);
2638 		}
2639 	}
2640 	return (ifp);
2641 }
2642 
2643 void
2644 if_free(struct ifnet *ifp)
2645 {
2646 	kfree(ifp, M_IFNET);
2647 }
2648 
2649 void
2650 ifq_set_classic(struct ifaltq *ifq)
2651 {
2652 	ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2653 	    ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2654 }
2655 
2656 void
2657 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2658     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2659 {
2660 	int q;
2661 
2662 	KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2663 	KASSERT(enqueue != NULL, ("enqueue is not specified"));
2664 	KASSERT(dequeue != NULL, ("dequeue is not specified"));
2665 	KASSERT(request != NULL, ("request is not specified"));
2666 
2667 	ifq->altq_mapsubq = mapsubq;
2668 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2669 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2670 
2671 		ifsq->ifsq_enqueue = enqueue;
2672 		ifsq->ifsq_dequeue = dequeue;
2673 		ifsq->ifsq_request = request;
2674 	}
2675 }
2676 
2677 static void
2678 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2679 {
2680 	m->m_nextpkt = NULL;
2681 	if (ifsq->ifsq_norm_tail == NULL)
2682 		ifsq->ifsq_norm_head = m;
2683 	else
2684 		ifsq->ifsq_norm_tail->m_nextpkt = m;
2685 	ifsq->ifsq_norm_tail = m;
2686 	ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2687 }
2688 
2689 static void
2690 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2691 {
2692 	m->m_nextpkt = NULL;
2693 	if (ifsq->ifsq_prio_tail == NULL)
2694 		ifsq->ifsq_prio_head = m;
2695 	else
2696 		ifsq->ifsq_prio_tail->m_nextpkt = m;
2697 	ifsq->ifsq_prio_tail = m;
2698 	ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2699 	ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2700 }
2701 
2702 static struct mbuf *
2703 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2704 {
2705 	struct mbuf *m;
2706 
2707 	m = ifsq->ifsq_norm_head;
2708 	if (m != NULL) {
2709 		if ((ifsq->ifsq_norm_head = m->m_nextpkt) == NULL)
2710 			ifsq->ifsq_norm_tail = NULL;
2711 		m->m_nextpkt = NULL;
2712 		ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2713 	}
2714 	return m;
2715 }
2716 
2717 static struct mbuf *
2718 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2719 {
2720 	struct mbuf *m;
2721 
2722 	m = ifsq->ifsq_prio_head;
2723 	if (m != NULL) {
2724 		if ((ifsq->ifsq_prio_head = m->m_nextpkt) == NULL)
2725 			ifsq->ifsq_prio_tail = NULL;
2726 		m->m_nextpkt = NULL;
2727 		ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2728 		ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2729 	}
2730 	return m;
2731 }
2732 
2733 int
2734 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2735     struct altq_pktattr *pa __unused)
2736 {
2737 	M_ASSERTPKTHDR(m);
2738 	if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2739 	    ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2740 		if ((m->m_flags & M_PRIO) &&
2741 		    ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen / 2) &&
2742 		    ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt / 2)) {
2743 			struct mbuf *m_drop;
2744 
2745 			/*
2746 			 * Perform drop-head on normal queue
2747 			 */
2748 			m_drop = ifsq_norm_dequeue(ifsq);
2749 			if (m_drop != NULL) {
2750 				m_freem(m_drop);
2751 				ifsq_prio_enqueue(ifsq, m);
2752 				return 0;
2753 			}
2754 			/* XXX nothing could be dropped? */
2755 		}
2756 		m_freem(m);
2757 		return ENOBUFS;
2758 	} else {
2759 		if (m->m_flags & M_PRIO)
2760 			ifsq_prio_enqueue(ifsq, m);
2761 		else
2762 			ifsq_norm_enqueue(ifsq, m);
2763 		return 0;
2764 	}
2765 }
2766 
2767 struct mbuf *
2768 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2769 {
2770 	struct mbuf *m;
2771 
2772 	switch (op) {
2773 	case ALTDQ_POLL:
2774 		m = ifsq->ifsq_prio_head;
2775 		if (m == NULL)
2776 			m = ifsq->ifsq_norm_head;
2777 		break;
2778 
2779 	case ALTDQ_REMOVE:
2780 		m = ifsq_prio_dequeue(ifsq);
2781 		if (m == NULL)
2782 			m = ifsq_norm_dequeue(ifsq);
2783 		break;
2784 
2785 	default:
2786 		panic("unsupported ALTQ dequeue op: %d", op);
2787 	}
2788 	return m;
2789 }
2790 
2791 int
2792 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2793 {
2794 	switch (req) {
2795 	case ALTRQ_PURGE:
2796 		for (;;) {
2797 			struct mbuf *m;
2798 
2799 			m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2800 			if (m == NULL)
2801 				break;
2802 			m_freem(m);
2803 		}
2804 		break;
2805 
2806 	default:
2807 		panic("unsupported ALTQ request: %d", req);
2808 	}
2809 	return 0;
2810 }
2811 
2812 static void
2813 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2814 {
2815 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
2816 	int running = 0, need_sched;
2817 
2818 	/*
2819 	 * Try to do direct ifnet.if_start on the subqueue first, if there is
2820 	 * contention on the subqueue hardware serializer, ifnet.if_start on
2821 	 * the subqueue will be scheduled on the subqueue owner CPU.
2822 	 */
2823 	if (!ifsq_tryserialize_hw(ifsq)) {
2824 		/*
2825 		 * Subqueue hardware serializer contention happened,
2826 		 * ifnet.if_start on the subqueue is scheduled on
2827 		 * the subqueue owner CPU, and we keep going.
2828 		 */
2829 		ifsq_ifstart_schedule(ifsq, 1);
2830 		return;
2831 	}
2832 
2833 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2834 		ifp->if_start(ifp, ifsq);
2835 		if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2836 			running = 1;
2837 	}
2838 	need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2839 
2840 	ifsq_deserialize_hw(ifsq);
2841 
2842 	if (need_sched) {
2843 		/*
2844 		 * More data need to be transmitted, ifnet.if_start on the
2845 		 * subqueue is scheduled on the subqueue owner CPU, and we
2846 		 * keep going.
2847 		 * NOTE: ifnet.if_start subqueue interlock is not released.
2848 		 */
2849 		ifsq_ifstart_schedule(ifsq, force_sched);
2850 	}
2851 }
2852 
2853 /*
2854  * Subqeue packets staging mechanism:
2855  *
2856  * The packets enqueued into the subqueue are staged to a certain amount
2857  * before the ifnet.if_start on the subqueue is called.  In this way, the
2858  * driver could avoid writing to hardware registers upon every packet,
2859  * instead, hardware registers could be written when certain amount of
2860  * packets are put onto hardware TX ring.  The measurement on several modern
2861  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
2862  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
2863  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
2864  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
2865  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
2866  *
2867  * Subqueue packets staging is performed for two entry points into drivers'
2868  * transmission function:
2869  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
2870  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
2871  *
2872  * Subqueue packets staging will be stopped upon any of the following
2873  * conditions:
2874  * - If the count of packets enqueued on the current CPU is great than or
2875  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2876  * - If the total length of packets enqueued on the current CPU is great
2877  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
2878  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
2879  *   is usually less than hardware's MTU.
2880  * - ifsq_ifstart_schedule() is not pending on the current CPU and
2881  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
2882  *   released.
2883  * - The if_start_rollup(), which is registered as low priority netisr
2884  *   rollup function, is called; probably because no more work is pending
2885  *   for netisr.
2886  *
2887  * NOTE:
2888  * Currently subqueue packet staging is only performed in netisr threads.
2889  */
2890 int
2891 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2892 {
2893 	struct ifaltq *ifq = &ifp->if_snd;
2894 	struct ifaltq_subque *ifsq;
2895 	int error, start = 0, len, mcast = 0, avoid_start = 0;
2896 	struct ifsubq_stage_head *head = NULL;
2897 	struct ifsubq_stage *stage = NULL;
2898 	struct globaldata *gd = mycpu;
2899 	struct thread *td = gd->gd_curthread;
2900 
2901 	crit_enter_quick(td);
2902 
2903 	ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
2904 	ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
2905 
2906 	len = m->m_pkthdr.len;
2907 	if (m->m_flags & M_MCAST)
2908 		mcast = 1;
2909 
2910 	if (td->td_type == TD_TYPE_NETISR) {
2911 		head = &ifsubq_stage_heads[mycpuid];
2912 		stage = ifsq_get_stage(ifsq, mycpuid);
2913 
2914 		stage->stg_cnt++;
2915 		stage->stg_len += len;
2916 		if (stage->stg_cnt < ifsq_stage_cntmax &&
2917 		    stage->stg_len < (ifp->if_mtu - max_protohdr))
2918 			avoid_start = 1;
2919 	}
2920 
2921 	ALTQ_SQ_LOCK(ifsq);
2922 	error = ifsq_enqueue_locked(ifsq, m, pa);
2923 	if (error) {
2924 		if (!ifsq_data_ready(ifsq)) {
2925 			ALTQ_SQ_UNLOCK(ifsq);
2926 			crit_exit_quick(td);
2927 			return error;
2928 		}
2929 		avoid_start = 0;
2930 	}
2931 	if (!ifsq_is_started(ifsq)) {
2932 		if (avoid_start) {
2933 			ALTQ_SQ_UNLOCK(ifsq);
2934 
2935 			KKASSERT(!error);
2936 			if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
2937 				ifsq_stage_insert(head, stage);
2938 
2939 			IFNET_STAT_INC(ifp, obytes, len);
2940 			if (mcast)
2941 				IFNET_STAT_INC(ifp, omcasts, 1);
2942 			crit_exit_quick(td);
2943 			return error;
2944 		}
2945 
2946 		/*
2947 		 * Hold the subqueue interlock of ifnet.if_start
2948 		 */
2949 		ifsq_set_started(ifsq);
2950 		start = 1;
2951 	}
2952 	ALTQ_SQ_UNLOCK(ifsq);
2953 
2954 	if (!error) {
2955 		IFNET_STAT_INC(ifp, obytes, len);
2956 		if (mcast)
2957 			IFNET_STAT_INC(ifp, omcasts, 1);
2958 	}
2959 
2960 	if (stage != NULL) {
2961 		if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
2962 			KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
2963 			if (!avoid_start) {
2964 				ifsq_stage_remove(head, stage);
2965 				ifsq_ifstart_schedule(ifsq, 1);
2966 			}
2967 			crit_exit_quick(td);
2968 			return error;
2969 		}
2970 
2971 		if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
2972 			ifsq_stage_remove(head, stage);
2973 		} else {
2974 			stage->stg_cnt = 0;
2975 			stage->stg_len = 0;
2976 		}
2977 	}
2978 
2979 	if (!start) {
2980 		crit_exit_quick(td);
2981 		return error;
2982 	}
2983 
2984 	ifsq_ifstart_try(ifsq, 0);
2985 
2986 	crit_exit_quick(td);
2987 	return error;
2988 }
2989 
2990 void *
2991 ifa_create(int size, int flags)
2992 {
2993 	struct ifaddr *ifa;
2994 	int i;
2995 
2996 	KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
2997 
2998 	ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2999 	if (ifa == NULL)
3000 		return NULL;
3001 
3002 	ifa->ifa_containers =
3003 	    kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
3004 	        M_IFADDR, M_WAITOK | M_ZERO);
3005 	ifa->ifa_ncnt = ncpus;
3006 	for (i = 0; i < ncpus; ++i) {
3007 		struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3008 
3009 		ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3010 		ifac->ifa = ifa;
3011 		ifac->ifa_refcnt = 1;
3012 	}
3013 #ifdef IFADDR_DEBUG
3014 	kprintf("alloc ifa %p %d\n", ifa, size);
3015 #endif
3016 	return ifa;
3017 }
3018 
3019 void
3020 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3021 {
3022 	struct ifaddr *ifa = ifac->ifa;
3023 
3024 	KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3025 	KKASSERT(ifac->ifa_refcnt == 0);
3026 	KASSERT(ifac->ifa_listmask == 0,
3027 		("ifa is still on %#x lists", ifac->ifa_listmask));
3028 
3029 	ifac->ifa_magic = IFA_CONTAINER_DEAD;
3030 
3031 #ifdef IFADDR_DEBUG_VERBOSE
3032 	kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3033 #endif
3034 
3035 	KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3036 		("invalid # of ifac, %d", ifa->ifa_ncnt));
3037 	if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3038 #ifdef IFADDR_DEBUG
3039 		kprintf("free ifa %p\n", ifa);
3040 #endif
3041 		kfree(ifa->ifa_containers, M_IFADDR);
3042 		kfree(ifa, M_IFADDR);
3043 	}
3044 }
3045 
3046 static void
3047 ifa_iflink_dispatch(netmsg_t nmsg)
3048 {
3049 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3050 	struct ifaddr *ifa = msg->ifa;
3051 	struct ifnet *ifp = msg->ifp;
3052 	int cpu = mycpuid;
3053 	struct ifaddr_container *ifac;
3054 
3055 	crit_enter();
3056 
3057 	ifac = &ifa->ifa_containers[cpu];
3058 	ASSERT_IFAC_VALID(ifac);
3059 	KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3060 		("ifaddr is on if_addrheads"));
3061 
3062 	ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3063 	if (msg->tail)
3064 		TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3065 	else
3066 		TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3067 
3068 	crit_exit();
3069 
3070 	ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
3071 }
3072 
3073 void
3074 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3075 {
3076 	struct netmsg_ifaddr msg;
3077 
3078 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3079 		    0, ifa_iflink_dispatch);
3080 	msg.ifa = ifa;
3081 	msg.ifp = ifp;
3082 	msg.tail = tail;
3083 
3084 	ifa_domsg(&msg.base.lmsg, 0);
3085 }
3086 
3087 static void
3088 ifa_ifunlink_dispatch(netmsg_t nmsg)
3089 {
3090 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3091 	struct ifaddr *ifa = msg->ifa;
3092 	struct ifnet *ifp = msg->ifp;
3093 	int cpu = mycpuid;
3094 	struct ifaddr_container *ifac;
3095 
3096 	crit_enter();
3097 
3098 	ifac = &ifa->ifa_containers[cpu];
3099 	ASSERT_IFAC_VALID(ifac);
3100 	KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3101 		("ifaddr is not on if_addrhead"));
3102 
3103 	TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3104 	ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3105 
3106 	crit_exit();
3107 
3108 	ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
3109 }
3110 
3111 void
3112 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3113 {
3114 	struct netmsg_ifaddr msg;
3115 
3116 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3117 		    0, ifa_ifunlink_dispatch);
3118 	msg.ifa = ifa;
3119 	msg.ifp = ifp;
3120 
3121 	ifa_domsg(&msg.base.lmsg, 0);
3122 }
3123 
3124 static void
3125 ifa_destroy_dispatch(netmsg_t nmsg)
3126 {
3127 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3128 
3129 	IFAFREE(msg->ifa);
3130 	ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3131 }
3132 
3133 void
3134 ifa_destroy(struct ifaddr *ifa)
3135 {
3136 	struct netmsg_ifaddr msg;
3137 
3138 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3139 		    0, ifa_destroy_dispatch);
3140 	msg.ifa = ifa;
3141 
3142 	ifa_domsg(&msg.base.lmsg, 0);
3143 }
3144 
3145 struct lwkt_port *
3146 ifnet_portfn(int cpu)
3147 {
3148 	return &ifnet_threads[cpu].td_msgport;
3149 }
3150 
3151 void
3152 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
3153 {
3154 	KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
3155 
3156 	if (next_cpu < ncpus)
3157 		lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
3158 	else
3159 		lwkt_replymsg(lmsg, 0);
3160 }
3161 
3162 int
3163 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
3164 {
3165 	KKASSERT(cpu < ncpus);
3166 	return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
3167 }
3168 
3169 void
3170 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
3171 {
3172 	KKASSERT(cpu < ncpus);
3173 	lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
3174 }
3175 
3176 /*
3177  * Generic netmsg service loop.  Some protocols may roll their own but all
3178  * must do the basic command dispatch function call done here.
3179  */
3180 static void
3181 ifnet_service_loop(void *arg __unused)
3182 {
3183 	netmsg_t msg;
3184 
3185 	while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
3186 		KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
3187 		msg->base.nm_dispatch(msg);
3188 	}
3189 }
3190 
3191 static void
3192 if_start_rollup(void)
3193 {
3194 	struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3195 	struct ifsubq_stage *stage;
3196 
3197 	crit_enter();
3198 
3199 	while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3200 		struct ifaltq_subque *ifsq = stage->stg_subq;
3201 		int is_sched = 0;
3202 
3203 		if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3204 			is_sched = 1;
3205 		ifsq_stage_remove(head, stage);
3206 
3207 		if (is_sched) {
3208 			ifsq_ifstart_schedule(ifsq, 1);
3209 		} else {
3210 			int start = 0;
3211 
3212 			ALTQ_SQ_LOCK(ifsq);
3213 			if (!ifsq_is_started(ifsq)) {
3214 				/*
3215 				 * Hold the subqueue interlock of
3216 				 * ifnet.if_start
3217 				 */
3218 				ifsq_set_started(ifsq);
3219 				start = 1;
3220 			}
3221 			ALTQ_SQ_UNLOCK(ifsq);
3222 
3223 			if (start)
3224 				ifsq_ifstart_try(ifsq, 1);
3225 		}
3226 		KKASSERT((stage->stg_flags &
3227 		    (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3228 	}
3229 
3230 	crit_exit();
3231 }
3232 
3233 static void
3234 ifnetinit(void *dummy __unused)
3235 {
3236 	int i;
3237 
3238 	for (i = 0; i < ncpus; ++i) {
3239 		struct thread *thr = &ifnet_threads[i];
3240 
3241 		lwkt_create(ifnet_service_loop, NULL, NULL,
3242 			    thr, TDF_NOSTART|TDF_FORCE_SPINPORT|TDF_FIXEDCPU,
3243 			    i, "ifnet %d", i);
3244 		netmsg_service_port_init(&thr->td_msgport);
3245 		lwkt_schedule(thr);
3246 	}
3247 
3248 	for (i = 0; i < ncpus; ++i)
3249 		TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3250 	netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3251 }
3252 
3253 void
3254 if_register_com_alloc(u_char type,
3255     if_com_alloc_t *a, if_com_free_t *f)
3256 {
3257 
3258         KASSERT(if_com_alloc[type] == NULL,
3259             ("if_register_com_alloc: %d already registered", type));
3260         KASSERT(if_com_free[type] == NULL,
3261             ("if_register_com_alloc: %d free already registered", type));
3262 
3263         if_com_alloc[type] = a;
3264         if_com_free[type] = f;
3265 }
3266 
3267 void
3268 if_deregister_com_alloc(u_char type)
3269 {
3270 
3271         KASSERT(if_com_alloc[type] != NULL,
3272             ("if_deregister_com_alloc: %d not registered", type));
3273         KASSERT(if_com_free[type] != NULL,
3274             ("if_deregister_com_alloc: %d free not registered", type));
3275         if_com_alloc[type] = NULL;
3276         if_com_free[type] = NULL;
3277 }
3278 
3279 int
3280 if_ring_count2(int cnt, int cnt_max)
3281 {
3282 	int shift = 0;
3283 
3284 	KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3285 	    ("invalid ring count max %d", cnt_max));
3286 
3287 	if (cnt <= 0)
3288 		cnt = cnt_max;
3289 	if (cnt > ncpus2)
3290 		cnt = ncpus2;
3291 	if (cnt > cnt_max)
3292 		cnt = cnt_max;
3293 
3294 	while ((1 << (shift + 1)) <= cnt)
3295 		++shift;
3296 	cnt = 1 << shift;
3297 
3298 	KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3299 	    ("calculate cnt %d, ncpus2 %d, cnt max %d",
3300 	     cnt, ncpus2, cnt_max));
3301 	return cnt;
3302 }
3303 
3304 void
3305 ifq_set_maxlen(struct ifaltq *ifq, int len)
3306 {
3307 	ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3308 }
3309 
3310 int
3311 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3312 {
3313 	return ALTQ_SUBQ_INDEX_DEFAULT;
3314 }
3315 
3316 int
3317 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3318 {
3319 	return (cpuid & ifq->altq_subq_mask);
3320 }
3321 
3322 static void
3323 ifsq_watchdog(void *arg)
3324 {
3325 	struct ifsubq_watchdog *wd = arg;
3326 	struct ifnet *ifp;
3327 
3328 	if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3329 		goto done;
3330 
3331 	ifp = ifsq_get_ifp(wd->wd_subq);
3332 	if (ifnet_tryserialize_all(ifp)) {
3333 		wd->wd_watchdog(wd->wd_subq);
3334 		ifnet_deserialize_all(ifp);
3335 	} else {
3336 		/* try again next timeout */
3337 		wd->wd_timer = 1;
3338 	}
3339 done:
3340 	ifsq_watchdog_reset(wd);
3341 }
3342 
3343 static void
3344 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3345 {
3346 	callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3347 	    ifsq_get_cpuid(wd->wd_subq));
3348 }
3349 
3350 void
3351 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3352     ifsq_watchdog_t watchdog)
3353 {
3354 	callout_init_mp(&wd->wd_callout);
3355 	wd->wd_timer = 0;
3356 	wd->wd_subq = ifsq;
3357 	wd->wd_watchdog = watchdog;
3358 }
3359 
3360 void
3361 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3362 {
3363 	wd->wd_timer = 0;
3364 	ifsq_watchdog_reset(wd);
3365 }
3366 
3367 void
3368 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3369 {
3370 	wd->wd_timer = 0;
3371 	callout_stop(&wd->wd_callout);
3372 }
3373 
3374 void
3375 ifnet_lock(void)
3376 {
3377 	KASSERT(curthread->td_type != TD_TYPE_NETISR,
3378 	    ("try holding ifnet lock in netisr"));
3379 	mtx_lock(&ifnet_mtx);
3380 }
3381 
3382 void
3383 ifnet_unlock(void)
3384 {
3385 	KASSERT(curthread->td_type != TD_TYPE_NETISR,
3386 	    ("try holding ifnet lock in netisr"));
3387 	mtx_unlock(&ifnet_mtx);
3388 }
3389 
3390 static struct ifnet_array *
3391 ifnet_array_alloc(int count)
3392 {
3393 	struct ifnet_array *arr;
3394 
3395 	arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3396 	    M_IFNET, M_WAITOK);
3397 	arr->ifnet_count = count;
3398 
3399 	return arr;
3400 }
3401 
3402 static void
3403 ifnet_array_free(struct ifnet_array *arr)
3404 {
3405 	if (arr == &ifnet_array0)
3406 		return;
3407 	kfree(arr, M_IFNET);
3408 }
3409 
3410 static struct ifnet_array *
3411 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3412 {
3413 	struct ifnet_array *arr;
3414 	int count, i;
3415 
3416 	KASSERT(old_arr->ifnet_count >= 0,
3417 	    ("invalid ifnet array count %d", old_arr->ifnet_count));
3418 	count = old_arr->ifnet_count + 1;
3419 	arr = ifnet_array_alloc(count);
3420 
3421 	/*
3422 	 * Save the old ifnet array and append this ifp to the end of
3423 	 * the new ifnet array.
3424 	 */
3425 	for (i = 0; i < old_arr->ifnet_count; ++i) {
3426 		KASSERT(old_arr->ifnet_arr[i] != ifp,
3427 		    ("%s is already in ifnet array", ifp->if_xname));
3428 		arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3429 	}
3430 	KASSERT(i == count - 1,
3431 	    ("add %s, ifnet array index mismatch, should be %d, but got %d",
3432 	     ifp->if_xname, count - 1, i));
3433 	arr->ifnet_arr[i] = ifp;
3434 
3435 	return arr;
3436 }
3437 
3438 static struct ifnet_array *
3439 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3440 {
3441 	struct ifnet_array *arr;
3442 	int count, i, idx, found = 0;
3443 
3444 	KASSERT(old_arr->ifnet_count > 0,
3445 	    ("invalid ifnet array count %d", old_arr->ifnet_count));
3446 	count = old_arr->ifnet_count - 1;
3447 	arr = ifnet_array_alloc(count);
3448 
3449 	/*
3450 	 * Save the old ifnet array, but skip this ifp.
3451 	 */
3452 	idx = 0;
3453 	for (i = 0; i < old_arr->ifnet_count; ++i) {
3454 		if (old_arr->ifnet_arr[i] == ifp) {
3455 			KASSERT(!found,
3456 			    ("dup %s is in ifnet array", ifp->if_xname));
3457 			found = 1;
3458 			continue;
3459 		}
3460 		KASSERT(idx < count,
3461 		    ("invalid ifnet array index %d, count %d", idx, count));
3462 		arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3463 		++idx;
3464 	}
3465 	KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3466 	KASSERT(idx == count,
3467 	    ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3468 	     ifp->if_xname, count, idx));
3469 
3470 	return arr;
3471 }
3472 
3473 const struct ifnet_array *
3474 ifnet_array_get(void)
3475 {
3476 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3477 	return ifnet_array;
3478 }
3479 
3480 int
3481 ifnet_array_isempty(void)
3482 {
3483 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3484 	if (ifnet_array->ifnet_count == 0)
3485 		return 1;
3486 	else
3487 		return 0;
3488 }
3489