xref: /dflybsd-src/sys/net/if.c (revision 650ba6fbdd2d4b0bcc6a1976a6630ada8e96b53f)
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_inet6.h"
34 #include "opt_inet.h"
35 #include "opt_ifpoll.h"
36 
37 #include <sys/param.h>
38 #include <sys/malloc.h>
39 #include <sys/mbuf.h>
40 #include <sys/systm.h>
41 #include <sys/proc.h>
42 #include <sys/caps.h>
43 #include <sys/protosw.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/socketops.h>
47 #include <sys/kernel.h>
48 #include <sys/ktr.h>
49 #include <sys/mutex.h>
50 #include <sys/lock.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 #include <sys/jail.h>
59 
60 #include <sys/thread2.h>
61 #include <sys/msgport2.h>
62 #include <sys/mutex2.h>
63 
64 #include <net/if.h>
65 #include <net/if_arp.h>
66 #include <net/if_dl.h>
67 #include <net/if_types.h>
68 #include <net/if_var.h>
69 #include <net/if_ringmap.h>
70 #include <net/ifq_var.h>
71 #include <net/radix.h>
72 #include <net/route.h>
73 #include <net/if_clone.h>
74 #include <net/netisr2.h>
75 #include <net/netmsg2.h>
76 
77 #include <machine/atomic.h>
78 #include <machine/stdarg.h>
79 #include <machine/smp.h>
80 
81 #if defined(INET) || defined(INET6)
82 #include <netinet/in.h>
83 #include <netinet/in_var.h>
84 #include <netinet/if_ether.h>
85 #ifdef INET6
86 #include <netinet6/in6_var.h>
87 #include <netinet6/in6_ifattach.h>
88 #endif /* INET6 */
89 #endif /* INET || INET6 */
90 
91 struct netmsg_ifaddr {
92 	struct netmsg_base base;
93 	struct ifaddr	*ifa;
94 	struct ifnet	*ifp;
95 	int		tail;
96 };
97 
98 struct ifsubq_stage_head {
99 	TAILQ_HEAD(, ifsubq_stage)	stg_head;
100 } __cachealign;
101 
102 struct if_ringmap {
103 	int		rm_cnt;
104 	int		rm_grid;
105 	int		rm_cpumap[];
106 };
107 
108 #define RINGMAP_FLAG_NONE		0x0
109 #define RINGMAP_FLAG_POWEROF2		0x1
110 
111 /*
112  * System initialization
113  */
114 static void	if_attachdomain(void *);
115 static void	if_attachdomain1(struct ifnet *);
116 static int	ifconf(u_long, caddr_t, struct ucred *);
117 static void	ifinit(void *);
118 static void	ifnetinit(void *);
119 static void	if_slowtimo(void *);
120 static int	if_rtdel(struct radix_node *, void *);
121 static void	if_slowtimo_dispatch(netmsg_t);
122 
123 /* Helper functions */
124 static void	ifsq_watchdog_reset(struct ifsubq_watchdog *);
125 static int	if_delmulti_serialized(struct ifnet *, struct sockaddr *);
126 static struct ifnet_array *ifnet_array_alloc(int);
127 static void	ifnet_array_free(struct ifnet_array *);
128 static struct ifnet_array *ifnet_array_add(struct ifnet *,
129 		    const struct ifnet_array *);
130 static struct ifnet_array *ifnet_array_del(struct ifnet *,
131 		    const struct ifnet_array *);
132 static struct ifg_group *if_creategroup(const char *);
133 static int	if_destroygroup(struct ifg_group *);
134 static int	if_delgroup_locked(struct ifnet *, const char *);
135 static int	if_getgroups(struct ifgroupreq *, struct ifnet *);
136 static int	if_getgroupmembers(struct ifgroupreq *);
137 
138 #ifdef INET6
139 /*
140  * XXX: declare here to avoid to include many inet6 related files..
141  * should be more generalized?
142  */
143 extern void	nd6_setmtu(struct ifnet *);
144 #endif
145 
146 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
147 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
148 SYSCTL_NODE(_net_link, OID_AUTO, ringmap, CTLFLAG_RW, 0, "link ringmap");
149 
150 static int ifsq_stage_cntmax = 16;
151 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
152 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
153     &ifsq_stage_cntmax, 0, "ifq staging packet count max");
154 
155 static int if_stats_compat = 0;
156 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
157     &if_stats_compat, 0, "Compat the old ifnet stats");
158 
159 static int if_ringmap_dumprdr = 0;
160 SYSCTL_INT(_net_link_ringmap, OID_AUTO, dump_rdr, CTLFLAG_RW,
161     &if_ringmap_dumprdr, 0, "dump redirect table");
162 
163 /* Interface description */
164 static unsigned int ifdescr_maxlen = 1024;
165 SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW,
166 	&ifdescr_maxlen, 0,
167 	"administrative maximum length for interface description");
168 
169 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL);
170 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, ifnetinit, NULL);
171 
172 static if_com_alloc_t *if_com_alloc[256];
173 static if_com_free_t *if_com_free[256];
174 
175 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
176 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
177 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
178 MALLOC_DEFINE(M_IFDESCR, "ifdescr", "ifnet descriptions");
179 
180 int			ifqmaxlen = IFQ_MAXLEN;
181 struct ifnethead	ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
182 struct ifgrouphead	ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
183 static struct lock	ifgroup_lock;
184 
185 static struct ifnet_array	ifnet_array0;
186 static struct ifnet_array	*ifnet_array = &ifnet_array0;
187 
188 static struct callout		if_slowtimo_timer;
189 static struct netmsg_base	if_slowtimo_netmsg;
190 
191 int			if_index = 0;
192 struct ifnet		**ifindex2ifnet = NULL;
193 static struct mtx	ifnet_mtx = MTX_INITIALIZER("ifnet");
194 
195 static struct ifsubq_stage_head	ifsubq_stage_heads[MAXCPU];
196 
197 #ifdef notyet
198 #define IFQ_KTR_STRING		"ifq=%p"
199 #define IFQ_KTR_ARGS		struct ifaltq *ifq
200 #ifndef KTR_IFQ
201 #define KTR_IFQ			KTR_ALL
202 #endif
203 KTR_INFO_MASTER(ifq);
204 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
205 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
206 #define logifq(name, arg)	KTR_LOG(ifq_ ## name, arg)
207 
208 #define IF_START_KTR_STRING	"ifp=%p"
209 #define IF_START_KTR_ARGS	struct ifnet *ifp
210 #ifndef KTR_IF_START
211 #define KTR_IF_START		KTR_ALL
212 #endif
213 KTR_INFO_MASTER(if_start);
214 KTR_INFO(KTR_IF_START, if_start, run, 0,
215 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
216 KTR_INFO(KTR_IF_START, if_start, sched, 1,
217 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
218 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
219 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
220 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
221 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
222 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
223 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
224 #define logifstart(name, arg)	KTR_LOG(if_start_ ## name, arg)
225 #endif /* notyet */
226 
227 /*
228  * Network interface utility routines.
229  *
230  * Routines with ifa_ifwith* names take sockaddr *'s as
231  * parameters.
232  */
233 /* ARGSUSED */
234 static void
235 ifinit(void *dummy)
236 {
237 	lockinit(&ifgroup_lock, "ifgroup", 0, 0);
238 
239 	callout_init_mp(&if_slowtimo_timer);
240 	netmsg_init(&if_slowtimo_netmsg, NULL, &netisr_adone_rport,
241 	    MSGF_PRIORITY, if_slowtimo_dispatch);
242 
243 	/* Start if_slowtimo */
244 	lwkt_sendmsg(netisr_cpuport(0), &if_slowtimo_netmsg.lmsg);
245 }
246 
247 static void
248 ifsq_ifstart_ipifunc(void *arg)
249 {
250 	struct ifaltq_subque *ifsq = arg;
251 	struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
252 
253 	crit_enter();
254 	if (lmsg->ms_flags & MSGF_DONE)
255 		lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg);
256 	crit_exit();
257 }
258 
259 static __inline void
260 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
261 {
262 	KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
263 	TAILQ_REMOVE(&head->stg_head, stage, stg_link);
264 	stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
265 	stage->stg_cnt = 0;
266 	stage->stg_len = 0;
267 }
268 
269 static __inline void
270 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
271 {
272 	KKASSERT((stage->stg_flags &
273 	    (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
274 	stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
275 	TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
276 }
277 
278 /*
279  * Schedule ifnet.if_start on the subqueue owner CPU
280  */
281 static void
282 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
283 {
284 	int cpu;
285 
286 	if (!force && curthread->td_type == TD_TYPE_NETISR &&
287 	    ifsq_stage_cntmax > 0) {
288 		struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
289 
290 		stage->stg_cnt = 0;
291 		stage->stg_len = 0;
292 		if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
293 			ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
294 		stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
295 		return;
296 	}
297 
298 	cpu = ifsq_get_cpuid(ifsq);
299 	if (cpu != mycpuid)
300 		lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
301 	else
302 		ifsq_ifstart_ipifunc(ifsq);
303 }
304 
305 /*
306  * NOTE:
307  * This function will release ifnet.if_start subqueue interlock,
308  * if ifnet.if_start for the subqueue does not need to be scheduled
309  */
310 static __inline int
311 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
312 {
313 	if (!running || ifsq_is_empty(ifsq)
314 #ifdef ALTQ
315 	    || ifsq->ifsq_altq->altq_tbr != NULL
316 #endif
317 	) {
318 		ALTQ_SQ_LOCK(ifsq);
319 		/*
320 		 * ifnet.if_start subqueue interlock is released, if:
321 		 * 1) Hardware can not take any packets, due to
322 		 *    o  interface is marked down
323 		 *    o  hardware queue is full (ifsq_is_oactive)
324 		 *    Under the second situation, hardware interrupt
325 		 *    or polling(4) will call/schedule ifnet.if_start
326 		 *    on the subqueue when hardware queue is ready
327 		 * 2) There is no packet in the subqueue.
328 		 *    Further ifq_dispatch or ifq_handoff will call/
329 		 *    schedule ifnet.if_start on the subqueue.
330 		 * 3) TBR is used and it does not allow further
331 		 *    dequeueing.
332 		 *    TBR callout will call ifnet.if_start on the
333 		 *    subqueue.
334 		 */
335 		if (!running || !ifsq_data_ready(ifsq)) {
336 			ifsq_clr_started(ifsq);
337 			ALTQ_SQ_UNLOCK(ifsq);
338 			return 0;
339 		}
340 		ALTQ_SQ_UNLOCK(ifsq);
341 	}
342 	return 1;
343 }
344 
345 static void
346 ifsq_ifstart_dispatch(netmsg_t msg)
347 {
348 	struct lwkt_msg *lmsg = &msg->base.lmsg;
349 	struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
350 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
351 	struct globaldata *gd = mycpu;
352 	int running = 0, need_sched;
353 
354 	crit_enter_gd(gd);
355 
356 	lwkt_replymsg(lmsg, 0);	/* reply ASAP */
357 
358 	if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) {
359 		/*
360 		 * We need to chase the subqueue owner CPU change.
361 		 */
362 		ifsq_ifstart_schedule(ifsq, 1);
363 		crit_exit_gd(gd);
364 		return;
365 	}
366 
367 	ifsq_serialize_hw(ifsq);
368 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
369 		ifp->if_start(ifp, ifsq);
370 		if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
371 			running = 1;
372 	}
373 	need_sched = ifsq_ifstart_need_schedule(ifsq, running);
374 	ifsq_deserialize_hw(ifsq);
375 
376 	if (need_sched) {
377 		/*
378 		 * More data need to be transmitted, ifnet.if_start is
379 		 * scheduled on the subqueue owner CPU, and we keep going.
380 		 * NOTE: ifnet.if_start subqueue interlock is not released.
381 		 */
382 		ifsq_ifstart_schedule(ifsq, 0);
383 	}
384 
385 	crit_exit_gd(gd);
386 }
387 
388 /* Device driver ifnet.if_start helper function */
389 void
390 ifsq_devstart(struct ifaltq_subque *ifsq)
391 {
392 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
393 	int running = 0;
394 
395 	ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
396 
397 	ALTQ_SQ_LOCK(ifsq);
398 	if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
399 		ALTQ_SQ_UNLOCK(ifsq);
400 		return;
401 	}
402 	ifsq_set_started(ifsq);
403 	ALTQ_SQ_UNLOCK(ifsq);
404 
405 	ifp->if_start(ifp, ifsq);
406 
407 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
408 		running = 1;
409 
410 	if (ifsq_ifstart_need_schedule(ifsq, running)) {
411 		/*
412 		 * More data need to be transmitted, ifnet.if_start is
413 		 * scheduled on ifnet's CPU, and we keep going.
414 		 * NOTE: ifnet.if_start interlock is not released.
415 		 */
416 		ifsq_ifstart_schedule(ifsq, 0);
417 	}
418 }
419 
420 void
421 if_devstart(struct ifnet *ifp)
422 {
423 	ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
424 }
425 
426 /* Device driver ifnet.if_start schedule helper function */
427 void
428 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
429 {
430 	ifsq_ifstart_schedule(ifsq, 1);
431 }
432 
433 void
434 if_devstart_sched(struct ifnet *ifp)
435 {
436 	ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
437 }
438 
439 static void
440 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
441 {
442 	lwkt_serialize_enter(ifp->if_serializer);
443 }
444 
445 static void
446 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
447 {
448 	lwkt_serialize_exit(ifp->if_serializer);
449 }
450 
451 static int
452 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
453 {
454 	return lwkt_serialize_try(ifp->if_serializer);
455 }
456 
457 #ifdef INVARIANTS
458 static void
459 if_default_serialize_assert(struct ifnet *ifp,
460 			    enum ifnet_serialize slz __unused,
461 			    boolean_t serialized)
462 {
463 	if (serialized)
464 		ASSERT_SERIALIZED(ifp->if_serializer);
465 	else
466 		ASSERT_NOT_SERIALIZED(ifp->if_serializer);
467 }
468 #endif
469 
470 /*
471  * Attach an interface to the list of "active" interfaces.
472  *
473  * The serializer is optional.
474  */
475 void
476 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
477 {
478 	unsigned socksize;
479 	int namelen, masklen;
480 	struct sockaddr_dl *sdl, *sdl_addr;
481 	struct ifaddr *ifa;
482 	struct ifaltq *ifq;
483 	struct ifnet **old_ifindex2ifnet = NULL;
484 	struct ifnet_array *old_ifnet_array;
485 	int i, q, qlen;
486 	char qlenname[64];
487 
488 	static int if_indexlim = 8;
489 
490 	if (ifp->if_serialize != NULL) {
491 		KASSERT(ifp->if_deserialize != NULL &&
492 			ifp->if_tryserialize != NULL &&
493 			ifp->if_serialize_assert != NULL,
494 			("serialize functions are partially setup"));
495 
496 		/*
497 		 * If the device supplies serialize functions,
498 		 * then clear if_serializer to catch any invalid
499 		 * usage of this field.
500 		 */
501 		KASSERT(serializer == NULL,
502 			("both serialize functions and default serializer "
503 			 "are supplied"));
504 		ifp->if_serializer = NULL;
505 	} else {
506 		KASSERT(ifp->if_deserialize == NULL &&
507 			ifp->if_tryserialize == NULL &&
508 			ifp->if_serialize_assert == NULL,
509 			("serialize functions are partially setup"));
510 		ifp->if_serialize = if_default_serialize;
511 		ifp->if_deserialize = if_default_deserialize;
512 		ifp->if_tryserialize = if_default_tryserialize;
513 #ifdef INVARIANTS
514 		ifp->if_serialize_assert = if_default_serialize_assert;
515 #endif
516 
517 		/*
518 		 * The serializer can be passed in from the device,
519 		 * allowing the same serializer to be used for both
520 		 * the interrupt interlock and the device queue.
521 		 * If not specified, the netif structure will use an
522 		 * embedded serializer.
523 		 */
524 		if (serializer == NULL) {
525 			serializer = &ifp->if_default_serializer;
526 			lwkt_serialize_init(serializer);
527 		}
528 		ifp->if_serializer = serializer;
529 	}
530 
531 	/*
532 	 * Make if_addrhead available on all CPUs, since they
533 	 * could be accessed by any threads.
534 	 */
535 	ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
536 				    M_IFADDR, M_WAITOK | M_ZERO);
537 	for (i = 0; i < ncpus; ++i)
538 		TAILQ_INIT(&ifp->if_addrheads[i]);
539 
540 	TAILQ_INIT(&ifp->if_multiaddrs);
541 	TAILQ_INIT(&ifp->if_groups);
542 	getmicrotime(&ifp->if_lastchange);
543 	if_addgroup(ifp, IFG_ALL);
544 
545 	/*
546 	 * create a Link Level name for this device
547 	 */
548 	namelen = strlen(ifp->if_xname);
549 	masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
550 	socksize = masklen + ifp->if_addrlen;
551 	if (socksize < sizeof(*sdl))
552 		socksize = sizeof(*sdl);
553 	socksize = RT_ROUNDUP(socksize);
554 	ifa = ifa_create(sizeof(struct ifaddr) + 2 * socksize);
555 	sdl = sdl_addr = (struct sockaddr_dl *)(ifa + 1);
556 	sdl->sdl_len = socksize;
557 	sdl->sdl_family = AF_LINK;
558 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
559 	sdl->sdl_nlen = namelen;
560 	sdl->sdl_type = ifp->if_type;
561 	ifp->if_lladdr = ifa;
562 	ifa->ifa_ifp = ifp;
563 	ifa->ifa_addr = (struct sockaddr *)sdl;
564 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
565 	ifa->ifa_netmask = (struct sockaddr *)sdl;
566 	sdl->sdl_len = masklen;
567 	while (namelen != 0)
568 		sdl->sdl_data[--namelen] = 0xff;
569 	ifa_iflink(ifa, ifp, 0 /* Insert head */);
570 
571 	/*
572 	 * Make if_data available on all CPUs, since they could
573 	 * be updated by hardware interrupt routing, which could
574 	 * be bound to any CPU.
575 	 */
576 	ifp->if_data_pcpu = kmalloc(ncpus * sizeof(struct ifdata_pcpu),
577 				    M_DEVBUF,
578 				    M_WAITOK | M_ZERO | M_CACHEALIGN);
579 
580 	if (ifp->if_mapsubq == NULL)
581 		ifp->if_mapsubq = ifq_mapsubq_default;
582 
583 	ifq = &ifp->if_snd;
584 	ifq->altq_type = 0;
585 	ifq->altq_disc = NULL;
586 	ifq->altq_flags &= ALTQF_CANTCHANGE;
587 	ifq->altq_tbr = NULL;
588 	ifq->altq_ifp = ifp;
589 
590 	if (ifq->altq_subq_cnt <= 0)
591 		ifq->altq_subq_cnt = 1;
592 	ifq->altq_subq =
593 		kmalloc(ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
594 			M_DEVBUF,
595 			M_WAITOK | M_ZERO | M_CACHEALIGN);
596 
597 	if (ifq->altq_maxlen == 0) {
598 		if_printf(ifp, "driver didn't set altq_maxlen\n");
599 		ifq_set_maxlen(ifq, ifqmaxlen);
600 	}
601 
602 	/* Allow user to override driver's setting. */
603 	ksnprintf(qlenname, sizeof(qlenname), "net.%s.qlenmax", ifp->if_xname);
604 	qlen = -1;
605 	TUNABLE_INT_FETCH(qlenname, &qlen);
606 	if (qlen > 0) {
607 		if_printf(ifp, "qlenmax -> %d\n", qlen);
608 		ifq_set_maxlen(ifq, qlen);
609 	}
610 
611 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
612 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
613 
614 		ALTQ_SQ_LOCK_INIT(ifsq);
615 		ifsq->ifsq_index = q;
616 
617 		ifsq->ifsq_altq = ifq;
618 		ifsq->ifsq_ifp = ifp;
619 
620 		ifsq->ifsq_maxlen = ifq->altq_maxlen;
621 		ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
622 		ifsq->ifsq_prepended = NULL;
623 		ifsq->ifsq_started = 0;
624 		ifsq->ifsq_hw_oactive = 0;
625 		ifsq_set_cpuid(ifsq, 0);
626 		if (ifp->if_serializer != NULL)
627 			ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
628 
629 		/* XXX: netisr_ncpus */
630 		ifsq->ifsq_stage =
631 			kmalloc(ncpus * sizeof(struct ifsubq_stage),
632 				M_DEVBUF,
633 				M_WAITOK | M_ZERO | M_CACHEALIGN);
634 		for (i = 0; i < ncpus; ++i)
635 			ifsq->ifsq_stage[i].stg_subq = ifsq;
636 
637 		/*
638 		 * Allocate one if_start message for each CPU, since
639 		 * the hardware TX ring could be assigned to any CPU.
640 		 *
641 		 * NOTE:
642 		 * If the hardware TX ring polling CPU and the hardware
643 		 * TX ring interrupt CPU are same, one if_start message
644 		 * should be enough.
645 		 */
646 		ifsq->ifsq_ifstart_nmsg =
647 		    kmalloc(ncpus * sizeof(struct netmsg_base),
648 		    M_LWKTMSG, M_WAITOK);
649 		for (i = 0; i < ncpus; ++i) {
650 			netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
651 			    &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
652 			ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
653 		}
654 	}
655 	ifq_set_classic(ifq);
656 
657 	/*
658 	 * Increase mbuf cluster/jcluster limits for the mbufs that
659 	 * could sit on the device queues for quite some time.
660 	 */
661 	if (ifp->if_nmbclusters > 0)
662 		mcl_inclimit(ifp->if_nmbclusters);
663 	if (ifp->if_nmbjclusters > 0)
664 		mjcl_inclimit(ifp->if_nmbjclusters);
665 
666 	/*
667 	 * Install this ifp into ifindex2inet, ifnet queue and ifnet
668 	 * array after it is setup.
669 	 *
670 	 * Protect ifindex2ifnet, ifnet queue and ifnet array changes
671 	 * by ifnet lock, so that non-netisr threads could get a
672 	 * consistent view.
673 	 */
674 	ifnet_lock();
675 
676 	/* Don't update if_index until ifindex2ifnet is setup */
677 	ifp->if_index = if_index + 1;
678 	sdl_addr->sdl_index = ifp->if_index;
679 
680 	/*
681 	 * Install this ifp into ifindex2ifnet
682 	 */
683 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
684 		unsigned int n;
685 		struct ifnet **q;
686 
687 		/*
688 		 * Grow ifindex2ifnet
689 		 */
690 		if_indexlim <<= 1;
691 		n = if_indexlim * sizeof(*q);
692 		q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
693 		if (ifindex2ifnet != NULL) {
694 			bcopy(ifindex2ifnet, q, n/2);
695 			/* Free old ifindex2ifnet after sync all netisrs */
696 			old_ifindex2ifnet = ifindex2ifnet;
697 		}
698 		ifindex2ifnet = q;
699 	}
700 	ifindex2ifnet[ifp->if_index] = ifp;
701 	/*
702 	 * Update if_index after this ifp is installed into ifindex2ifnet,
703 	 * so that netisrs could get a consistent view of ifindex2ifnet.
704 	 */
705 	cpu_sfence();
706 	if_index = ifp->if_index;
707 
708 	/*
709 	 * Install this ifp into ifnet array.
710 	 */
711 	/* Free old ifnet array after sync all netisrs */
712 	old_ifnet_array = ifnet_array;
713 	ifnet_array = ifnet_array_add(ifp, old_ifnet_array);
714 
715 	/*
716 	 * Install this ifp into ifnet queue.
717 	 */
718 	TAILQ_INSERT_TAIL(&ifnetlist, ifp, if_link);
719 
720 	ifnet_unlock();
721 
722 	/*
723 	 * Sync all netisrs so that the old ifindex2ifnet and ifnet array
724 	 * are no longer accessed and we can free them safely later on.
725 	 */
726 	netmsg_service_sync();
727 	if (old_ifindex2ifnet != NULL)
728 		kfree(old_ifindex2ifnet, M_IFADDR);
729 	ifnet_array_free(old_ifnet_array);
730 
731 	if (!SLIST_EMPTY(&domains))
732 		if_attachdomain1(ifp);
733 
734 	/* Announce the interface. */
735 	EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
736 	devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
737 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
738 }
739 
740 static void
741 if_attachdomain(void *dummy)
742 {
743 	struct ifnet *ifp;
744 
745 	ifnet_lock();
746 	TAILQ_FOREACH(ifp, &ifnetlist, if_list)
747 		if_attachdomain1(ifp);
748 	ifnet_unlock();
749 }
750 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
751 	if_attachdomain, NULL);
752 
753 static void
754 if_attachdomain1(struct ifnet *ifp)
755 {
756 	struct domain *dp;
757 
758 	crit_enter();
759 
760 	/* address family dependent data region */
761 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
762 	SLIST_FOREACH(dp, &domains, dom_next)
763 		if (dp->dom_ifattach)
764 			ifp->if_afdata[dp->dom_family] =
765 				(*dp->dom_ifattach)(ifp);
766 	crit_exit();
767 }
768 
769 /*
770  * Purge all addresses whose type is _not_ AF_LINK
771  */
772 static void
773 if_purgeaddrs_nolink_dispatch(netmsg_t nmsg)
774 {
775 	struct ifnet *ifp = nmsg->lmsg.u.ms_resultp;
776 	struct ifaddr_container *ifac, *next;
777 
778 	ASSERT_NETISR0;
779 
780 	/*
781 	 * The ifaddr processing in the following loop will block,
782 	 * however, this function is called in netisr0, in which
783 	 * ifaddr list changes happen, so we don't care about the
784 	 * blockness of the ifaddr processing here.
785 	 */
786 	TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
787 			      ifa_link, next) {
788 		struct ifaddr *ifa = ifac->ifa;
789 
790 		/* Ignore marker */
791 		if (ifa->ifa_addr->sa_family == AF_UNSPEC)
792 			continue;
793 
794 		/* Leave link ifaddr as it is */
795 		if (ifa->ifa_addr->sa_family == AF_LINK)
796 			continue;
797 #ifdef INET
798 		/* XXX: Ugly!! ad hoc just for INET */
799 		if (ifa->ifa_addr->sa_family == AF_INET) {
800 			struct ifaliasreq ifr;
801 			struct sockaddr_in saved_addr, saved_dst;
802 #ifdef IFADDR_DEBUG_VERBOSE
803 			int i;
804 
805 			kprintf("purge in4 addr %p: ", ifa);
806 			for (i = 0; i < ncpus; ++i) {
807 				kprintf("%d ",
808 				    ifa->ifa_containers[i].ifa_refcnt);
809 			}
810 			kprintf("\n");
811 #endif
812 
813 			/* Save information for panic. */
814 			memcpy(&saved_addr, ifa->ifa_addr, sizeof(saved_addr));
815 			if (ifa->ifa_dstaddr != NULL) {
816 				memcpy(&saved_dst, ifa->ifa_dstaddr,
817 				    sizeof(saved_dst));
818 			} else {
819 				memset(&saved_dst, 0, sizeof(saved_dst));
820 			}
821 
822 			bzero(&ifr, sizeof ifr);
823 			ifr.ifra_addr = *ifa->ifa_addr;
824 			if (ifa->ifa_dstaddr)
825 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
826 			if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp,
827 				       NULL) == 0)
828 				continue;
829 
830 			/* MUST NOT HAPPEN */
831 			panic("%s: in_control failed %x, dst %x", ifp->if_xname,
832 			    ntohl(saved_addr.sin_addr.s_addr),
833 			    ntohl(saved_dst.sin_addr.s_addr));
834 		}
835 #endif /* INET */
836 #ifdef INET6
837 		if (ifa->ifa_addr->sa_family == AF_INET6) {
838 #ifdef IFADDR_DEBUG_VERBOSE
839 			int i;
840 
841 			kprintf("purge in6 addr %p: ", ifa);
842 			for (i = 0; i < ncpus; ++i) {
843 				kprintf("%d ",
844 				    ifa->ifa_containers[i].ifa_refcnt);
845 			}
846 			kprintf("\n");
847 #endif
848 
849 			in6_purgeaddr(ifa);
850 			/* ifp_addrhead is already updated */
851 			continue;
852 		}
853 #endif /* INET6 */
854 		if_printf(ifp, "destroy ifaddr family %d\n",
855 		    ifa->ifa_addr->sa_family);
856 		ifa_ifunlink(ifa, ifp);
857 		ifa_destroy(ifa);
858 	}
859 
860 	netisr_replymsg(&nmsg->base, 0);
861 }
862 
863 void
864 if_purgeaddrs_nolink(struct ifnet *ifp)
865 {
866 	struct netmsg_base nmsg;
867 
868 	netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
869 	    if_purgeaddrs_nolink_dispatch);
870 	nmsg.lmsg.u.ms_resultp = ifp;
871 	netisr_domsg(&nmsg, 0);
872 }
873 
874 static void
875 ifq_stage_detach_handler(netmsg_t nmsg)
876 {
877 	struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
878 	int q;
879 
880 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
881 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
882 		struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
883 
884 		if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
885 			ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
886 	}
887 	lwkt_replymsg(&nmsg->lmsg, 0);
888 }
889 
890 static void
891 ifq_stage_detach(struct ifaltq *ifq)
892 {
893 	struct netmsg_base base;
894 	int cpu;
895 
896 	netmsg_init(&base, NULL, &curthread->td_msgport, 0,
897 	    ifq_stage_detach_handler);
898 	base.lmsg.u.ms_resultp = ifq;
899 
900 	/* XXX netisr_ncpus */
901 	for (cpu = 0; cpu < ncpus; ++cpu)
902 		lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
903 }
904 
905 struct netmsg_if_rtdel {
906 	struct netmsg_base	base;
907 	struct ifnet		*ifp;
908 };
909 
910 static void
911 if_rtdel_dispatch(netmsg_t msg)
912 {
913 	struct netmsg_if_rtdel *rmsg = (void *)msg;
914 	int i, cpu;
915 
916 	cpu = mycpuid;
917 	ASSERT_NETISR_NCPUS(cpu);
918 
919 	for (i = 1; i <= AF_MAX; i++) {
920 		struct radix_node_head	*rnh;
921 
922 		if ((rnh = rt_tables[cpu][i]) == NULL)
923 			continue;
924 		rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
925 	}
926 	netisr_forwardmsg(&msg->base, cpu + 1);
927 }
928 
929 /*
930  * Detach an interface, removing it from the
931  * list of "active" interfaces.
932  */
933 void
934 if_detach(struct ifnet *ifp)
935 {
936 	struct ifnet_array *old_ifnet_array;
937 	struct ifg_list *ifgl;
938 	struct netmsg_if_rtdel msg;
939 	struct domain *dp;
940 	int q;
941 
942 	/* Announce that the interface is gone. */
943 	EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
944 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
945 	devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
946 
947 	/*
948 	 * Remove this ifp from ifindex2inet, ifnet queue and ifnet
949 	 * array before it is whacked.
950 	 *
951 	 * Protect ifindex2ifnet, ifnet queue and ifnet array changes
952 	 * by ifnet lock, so that non-netisr threads could get a
953 	 * consistent view.
954 	 */
955 	ifnet_lock();
956 
957 	/*
958 	 * Remove this ifp from ifindex2ifnet and maybe decrement if_index.
959 	 */
960 	ifindex2ifnet[ifp->if_index] = NULL;
961 	while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
962 		if_index--;
963 
964 	/*
965 	 * Remove this ifp from ifnet queue.
966 	 */
967 	TAILQ_REMOVE(&ifnetlist, ifp, if_link);
968 
969 	/*
970 	 * Remove this ifp from ifnet array.
971 	 */
972 	/* Free old ifnet array after sync all netisrs */
973 	old_ifnet_array = ifnet_array;
974 	ifnet_array = ifnet_array_del(ifp, old_ifnet_array);
975 
976 	ifnet_unlock();
977 
978 	ifgroup_lockmgr(LK_EXCLUSIVE);
979 	while ((ifgl = TAILQ_FIRST(&ifp->if_groups)) != NULL)
980 		if_delgroup_locked(ifp, ifgl->ifgl_group->ifg_group);
981 	ifgroup_lockmgr(LK_RELEASE);
982 
983 	/*
984 	 * Sync all netisrs so that the old ifnet array is no longer
985 	 * accessed and we can free it safely later on.
986 	 */
987 	netmsg_service_sync();
988 	ifnet_array_free(old_ifnet_array);
989 
990 	/*
991 	 * Remove routes and flush queues.
992 	 */
993 	crit_enter();
994 #ifdef IFPOLL_ENABLE
995 	if (ifp->if_flags & IFF_NPOLLING)
996 		ifpoll_deregister(ifp);
997 #endif
998 	if_down(ifp);
999 
1000 	/* Decrease the mbuf clusters/jclusters limits increased by us */
1001 	if (ifp->if_nmbclusters > 0)
1002 		mcl_inclimit(-ifp->if_nmbclusters);
1003 	if (ifp->if_nmbjclusters > 0)
1004 		mjcl_inclimit(-ifp->if_nmbjclusters);
1005 
1006 #ifdef ALTQ
1007 	if (ifq_is_enabled(&ifp->if_snd))
1008 		altq_disable(&ifp->if_snd);
1009 	if (ifq_is_attached(&ifp->if_snd))
1010 		altq_detach(&ifp->if_snd);
1011 #endif
1012 
1013 	/*
1014 	 * Clean up all addresses.
1015 	 */
1016 	ifp->if_lladdr = NULL;
1017 
1018 	if_purgeaddrs_nolink(ifp);
1019 	if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
1020 		struct ifaddr *ifa;
1021 
1022 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1023 		KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
1024 			("non-link ifaddr is left on if_addrheads"));
1025 
1026 		ifa_ifunlink(ifa, ifp);
1027 		ifa_destroy(ifa);
1028 		KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
1029 			("there are still ifaddrs left on if_addrheads"));
1030 	}
1031 
1032 #ifdef INET
1033 	/*
1034 	 * Remove all IPv4 kernel structures related to ifp.
1035 	 */
1036 	in_ifdetach(ifp);
1037 #endif
1038 
1039 #ifdef INET6
1040 	/*
1041 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
1042 	 * before removing routing entries below, since IPv6 interface direct
1043 	 * routes are expected to be removed by the IPv6-specific kernel API.
1044 	 * Otherwise, the kernel will detect some inconsistency and bark it.
1045 	 */
1046 	in6_ifdetach(ifp);
1047 #endif
1048 
1049 	/*
1050 	 * Delete all remaining routes using this interface
1051 	 */
1052 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1053 	    if_rtdel_dispatch);
1054 	msg.ifp = ifp;
1055 	netisr_domsg_global(&msg.base);
1056 
1057 	SLIST_FOREACH(dp, &domains, dom_next) {
1058 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
1059 			(*dp->dom_ifdetach)(ifp,
1060 				ifp->if_afdata[dp->dom_family]);
1061 	}
1062 
1063 	kfree(ifp->if_addrheads, M_IFADDR);
1064 
1065 	lwkt_synchronize_ipiqs("if_detach");
1066 	ifq_stage_detach(&ifp->if_snd);
1067 
1068 	for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
1069 		struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
1070 
1071 		kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
1072 		kfree(ifsq->ifsq_stage, M_DEVBUF);
1073 	}
1074 	kfree(ifp->if_snd.altq_subq, M_DEVBUF);
1075 
1076 	kfree(ifp->if_data_pcpu, M_DEVBUF);
1077 
1078 	crit_exit();
1079 }
1080 
1081 int
1082 ifgroup_lockmgr(u_int flags)
1083 {
1084 	return lockmgr(&ifgroup_lock, flags);
1085 }
1086 
1087 /*
1088  * Create an empty interface group.
1089  */
1090 static struct ifg_group *
1091 if_creategroup(const char *groupname)
1092 {
1093 	struct ifg_group *ifg;
1094 
1095 	ifg = kmalloc(sizeof(*ifg), M_IFNET, M_WAITOK);
1096 	strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1097 	ifg->ifg_refcnt = 0;
1098 	ifg->ifg_carp_demoted = 0;
1099 	TAILQ_INIT(&ifg->ifg_members);
1100 
1101 	ifgroup_lockmgr(LK_EXCLUSIVE);
1102 	TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
1103 	ifgroup_lockmgr(LK_RELEASE);
1104 
1105 	EVENTHANDLER_INVOKE(group_attach_event, ifg);
1106 
1107 	return (ifg);
1108 }
1109 
1110 /*
1111  * Destroy an empty interface group.
1112  */
1113 static int
1114 if_destroygroup(struct ifg_group *ifg)
1115 {
1116 	KASSERT(ifg->ifg_refcnt == 0,
1117 		("trying to delete a non-empty interface group"));
1118 
1119 	ifgroup_lockmgr(LK_EXCLUSIVE);
1120 	TAILQ_REMOVE(&ifg_head, ifg, ifg_next);
1121 	ifgroup_lockmgr(LK_RELEASE);
1122 
1123 	EVENTHANDLER_INVOKE(group_detach_event, ifg);
1124 	kfree(ifg, M_IFNET);
1125 
1126 	return (0);
1127 }
1128 
1129 /*
1130  * Add the interface to a group.
1131  * The target group will be created if it doesn't exist.
1132  */
1133 int
1134 if_addgroup(struct ifnet *ifp, const char *groupname)
1135 {
1136 	struct ifg_list *ifgl;
1137 	struct ifg_group *ifg;
1138 	struct ifg_member *ifgm;
1139 
1140 	if (groupname[0] &&
1141 	    groupname[strlen(groupname) - 1] >= '0' &&
1142 	    groupname[strlen(groupname) - 1] <= '9')
1143 		return (EINVAL);
1144 
1145 	ifgroup_lockmgr(LK_SHARED);
1146 
1147 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1148 		if (strcmp(ifgl->ifgl_group->ifg_group, groupname) == 0) {
1149 			ifgroup_lockmgr(LK_RELEASE);
1150 			return (EEXIST);
1151 		}
1152 	}
1153 
1154 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next) {
1155 		if (strcmp(ifg->ifg_group, groupname) == 0)
1156 			break;
1157 	}
1158 
1159 	ifgroup_lockmgr(LK_RELEASE);
1160 
1161 	if (ifg == NULL)
1162 		ifg = if_creategroup(groupname);
1163 
1164 	ifgl = kmalloc(sizeof(*ifgl), M_IFNET, M_WAITOK);
1165 	ifgm = kmalloc(sizeof(*ifgm), M_IFNET, M_WAITOK);
1166 	ifgl->ifgl_group = ifg;
1167 	ifgm->ifgm_ifp = ifp;
1168 	ifg->ifg_refcnt++;
1169 
1170 	ifgroup_lockmgr(LK_EXCLUSIVE);
1171 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1172 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1173 	ifgroup_lockmgr(LK_RELEASE);
1174 
1175 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1176 
1177 	return (0);
1178 }
1179 
1180 /*
1181  * Remove the interface from a group.
1182  * The group will be destroyed if it becomes empty.
1183  *
1184  * The 'ifgroup_lock' must be hold exclusively when calling this.
1185  */
1186 static int
1187 if_delgroup_locked(struct ifnet *ifp, const char *groupname)
1188 {
1189 	struct ifg_list *ifgl;
1190 	struct ifg_member *ifgm;
1191 
1192 	KKASSERT(lockstatus(&ifgroup_lock, curthread) == LK_EXCLUSIVE);
1193 
1194 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1195 		if (strcmp(ifgl->ifgl_group->ifg_group, groupname) == 0)
1196 			break;
1197 	}
1198 	if (ifgl == NULL)
1199 		return (ENOENT);
1200 
1201 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1202 
1203 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next) {
1204 		if (ifgm->ifgm_ifp == ifp)
1205 			break;
1206 	}
1207 
1208 	if (ifgm != NULL) {
1209 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1210 
1211 		ifgroup_lockmgr(LK_RELEASE);
1212 		EVENTHANDLER_INVOKE(group_change_event, groupname);
1213 		ifgroup_lockmgr(LK_EXCLUSIVE);
1214 
1215 		kfree(ifgm, M_IFNET);
1216 		ifgl->ifgl_group->ifg_refcnt--;
1217 	}
1218 
1219 	if (ifgl->ifgl_group->ifg_refcnt == 0) {
1220 		ifgroup_lockmgr(LK_RELEASE);
1221 		if_destroygroup(ifgl->ifgl_group);
1222 		ifgroup_lockmgr(LK_EXCLUSIVE);
1223 	}
1224 
1225 	kfree(ifgl, M_IFNET);
1226 
1227 	return (0);
1228 }
1229 
1230 int
1231 if_delgroup(struct ifnet *ifp, const char *groupname)
1232 {
1233 	int error;
1234 
1235 	ifgroup_lockmgr(LK_EXCLUSIVE);
1236 	error = if_delgroup_locked(ifp, groupname);
1237 	ifgroup_lockmgr(LK_RELEASE);
1238 
1239 	return (error);
1240 }
1241 
1242 /*
1243  * Store all the groups that the interface belongs to in memory
1244  * pointed to by data.
1245  */
1246 static int
1247 if_getgroups(struct ifgroupreq *ifgr, struct ifnet *ifp)
1248 {
1249 	struct ifg_list *ifgl;
1250 	struct ifg_req *ifgrq, *p;
1251 	int len, error;
1252 
1253 	len = 0;
1254 	ifgroup_lockmgr(LK_SHARED);
1255 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1256 		len += sizeof(struct ifg_req);
1257 	ifgroup_lockmgr(LK_RELEASE);
1258 
1259 	if (ifgr->ifgr_len == 0) {
1260 		/*
1261 		 * Caller is asking how much memory should be allocated in
1262 		 * the next request in order to hold all the groups.
1263 		 */
1264 		ifgr->ifgr_len = len;
1265 		return (0);
1266 	} else if (ifgr->ifgr_len != len) {
1267 		return (EINVAL);
1268 	}
1269 
1270 	ifgrq = kmalloc(len, M_TEMP, M_INTWAIT | M_NULLOK | M_ZERO);
1271 	if (ifgrq == NULL)
1272 		return (ENOMEM);
1273 
1274 	ifgroup_lockmgr(LK_SHARED);
1275 	p = ifgrq;
1276 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1277 		if (len < sizeof(struct ifg_req)) {
1278 			ifgroup_lockmgr(LK_RELEASE);
1279 			error = EINVAL;
1280 			goto failed;
1281 		}
1282 
1283 		strlcpy(p->ifgrq_group, ifgl->ifgl_group->ifg_group,
1284 			sizeof(ifgrq->ifgrq_group));
1285 		len -= sizeof(struct ifg_req);
1286 		p++;
1287 	}
1288 	ifgroup_lockmgr(LK_RELEASE);
1289 
1290 	error = copyout(ifgrq, ifgr->ifgr_groups, ifgr->ifgr_len);
1291 failed:
1292 	kfree(ifgrq, M_TEMP);
1293 	return error;
1294 }
1295 
1296 /*
1297  * Store all the members of a group in memory pointed to by data.
1298  */
1299 static int
1300 if_getgroupmembers(struct ifgroupreq *ifgr)
1301 {
1302 	struct ifg_group *ifg;
1303 	struct ifg_member *ifgm;
1304 	struct ifg_req *ifgrq, *p;
1305 	int len, error;
1306 
1307 	ifgroup_lockmgr(LK_SHARED);
1308 
1309 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next) {
1310 		if (strcmp(ifg->ifg_group, ifgr->ifgr_name) == 0)
1311 			break;
1312 	}
1313 	if (ifg == NULL) {
1314 		ifgroup_lockmgr(LK_RELEASE);
1315 		return (ENOENT);
1316 	}
1317 
1318 	len = 0;
1319 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1320 		len += sizeof(struct ifg_req);
1321 
1322 	ifgroup_lockmgr(LK_RELEASE);
1323 
1324 	if (ifgr->ifgr_len == 0) {
1325 		ifgr->ifgr_len = len;
1326 		return (0);
1327 	} else if (ifgr->ifgr_len != len) {
1328 		return (EINVAL);
1329 	}
1330 
1331 	ifgrq = kmalloc(len, M_TEMP, M_INTWAIT | M_NULLOK | M_ZERO);
1332 	if (ifgrq == NULL)
1333 		return (ENOMEM);
1334 
1335 	ifgroup_lockmgr(LK_SHARED);
1336 	p = ifgrq;
1337 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1338 		if (len < sizeof(struct ifg_req)) {
1339 			ifgroup_lockmgr(LK_RELEASE);
1340 			error = EINVAL;
1341 			goto failed;
1342 		}
1343 
1344 		strlcpy(p->ifgrq_member, ifgm->ifgm_ifp->if_xname,
1345 			sizeof(p->ifgrq_member));
1346 		len -= sizeof(struct ifg_req);
1347 		p++;
1348 	}
1349 	ifgroup_lockmgr(LK_RELEASE);
1350 
1351 	error = copyout(ifgrq, ifgr->ifgr_groups, ifgr->ifgr_len);
1352 failed:
1353 	kfree(ifgrq, M_TEMP);
1354 	return error;
1355 }
1356 
1357 /*
1358  * Delete Routes for a Network Interface
1359  *
1360  * Called for each routing entry via the rnh->rnh_walktree() call above
1361  * to delete all route entries referencing a detaching network interface.
1362  *
1363  * Arguments:
1364  *	rn	pointer to node in the routing table
1365  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1366  *
1367  * Returns:
1368  *	0	successful
1369  *	errno	failed - reason indicated
1370  *
1371  */
1372 static int
1373 if_rtdel(struct radix_node *rn, void *arg)
1374 {
1375 	struct rtentry	*rt = (struct rtentry *)rn;
1376 	struct ifnet	*ifp = arg;
1377 	int		err;
1378 
1379 	if (rt->rt_ifp == ifp) {
1380 
1381 		/*
1382 		 * Protect (sorta) against walktree recursion problems
1383 		 * with cloned routes
1384 		 */
1385 		if (!(rt->rt_flags & RTF_UP))
1386 			return (0);
1387 
1388 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1389 				rt_mask(rt), rt->rt_flags,
1390 				NULL);
1391 		if (err) {
1392 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1393 		}
1394 	}
1395 
1396 	return (0);
1397 }
1398 
1399 static __inline boolean_t
1400 ifa_prefer(const struct ifaddr *cur_ifa, const struct ifaddr *old_ifa)
1401 {
1402 	if (old_ifa == NULL)
1403 		return TRUE;
1404 
1405 	if ((old_ifa->ifa_ifp->if_flags & IFF_UP) == 0 &&
1406 	    (cur_ifa->ifa_ifp->if_flags & IFF_UP))
1407 		return TRUE;
1408 	if ((old_ifa->ifa_flags & IFA_ROUTE) == 0 &&
1409 	    (cur_ifa->ifa_flags & IFA_ROUTE))
1410 		return TRUE;
1411 	return FALSE;
1412 }
1413 
1414 /*
1415  * Locate an interface based on a complete address.
1416  */
1417 struct ifaddr *
1418 ifa_ifwithaddr(struct sockaddr *addr)
1419 {
1420 	const struct ifnet_array *arr;
1421 	int i;
1422 
1423 	arr = ifnet_array_get();
1424 	for (i = 0; i < arr->ifnet_count; ++i) {
1425 		struct ifnet *ifp = arr->ifnet_arr[i];
1426 		struct ifaddr_container *ifac;
1427 
1428 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1429 			struct ifaddr *ifa = ifac->ifa;
1430 
1431 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1432 				continue;
1433 			if (sa_equal(addr, ifa->ifa_addr))
1434 				return (ifa);
1435 			if ((ifp->if_flags & IFF_BROADCAST) &&
1436 			    ifa->ifa_broadaddr &&
1437 			    /* IPv6 doesn't have broadcast */
1438 			    ifa->ifa_broadaddr->sa_len != 0 &&
1439 			    sa_equal(ifa->ifa_broadaddr, addr))
1440 				return (ifa);
1441 		}
1442 	}
1443 	return (NULL);
1444 }
1445 
1446 /*
1447  * Locate the point to point interface with a given destination address.
1448  */
1449 struct ifaddr *
1450 ifa_ifwithdstaddr(struct sockaddr *addr)
1451 {
1452 	const struct ifnet_array *arr;
1453 	int i;
1454 
1455 	arr = ifnet_array_get();
1456 	for (i = 0; i < arr->ifnet_count; ++i) {
1457 		struct ifnet *ifp = arr->ifnet_arr[i];
1458 		struct ifaddr_container *ifac;
1459 
1460 		if (!(ifp->if_flags & IFF_POINTOPOINT))
1461 			continue;
1462 
1463 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1464 			struct ifaddr *ifa = ifac->ifa;
1465 
1466 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1467 				continue;
1468 			if (ifa->ifa_dstaddr &&
1469 			    sa_equal(addr, ifa->ifa_dstaddr))
1470 				return (ifa);
1471 		}
1472 	}
1473 	return (NULL);
1474 }
1475 
1476 /*
1477  * Find an interface on a specific network.  If many, choice
1478  * is most specific found.
1479  */
1480 struct ifaddr *
1481 ifa_ifwithnet(struct sockaddr *addr)
1482 {
1483 	struct ifaddr *ifa_maybe = NULL;
1484 	u_int af = addr->sa_family;
1485 	char *addr_data = addr->sa_data, *cplim;
1486 	const struct ifnet_array *arr;
1487 	int i;
1488 
1489 	/*
1490 	 * AF_LINK addresses can be looked up directly by their index number,
1491 	 * so do that if we can.
1492 	 */
1493 	if (af == AF_LINK) {
1494 		struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1495 
1496 		if (sdl->sdl_index && sdl->sdl_index <= if_index)
1497 			return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1498 	}
1499 
1500 	/*
1501 	 * Scan though each interface, looking for ones that have
1502 	 * addresses in this address family.
1503 	 */
1504 	arr = ifnet_array_get();
1505 	for (i = 0; i < arr->ifnet_count; ++i) {
1506 		struct ifnet *ifp = arr->ifnet_arr[i];
1507 		struct ifaddr_container *ifac;
1508 
1509 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1510 			struct ifaddr *ifa = ifac->ifa;
1511 			char *cp, *cp2, *cp3;
1512 
1513 			if (ifa->ifa_addr->sa_family != af)
1514 next:				continue;
1515 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1516 				/*
1517 				 * This is a bit broken as it doesn't
1518 				 * take into account that the remote end may
1519 				 * be a single node in the network we are
1520 				 * looking for.
1521 				 * The trouble is that we don't know the
1522 				 * netmask for the remote end.
1523 				 */
1524 				if (ifa->ifa_dstaddr != NULL &&
1525 				    sa_equal(addr, ifa->ifa_dstaddr))
1526 					return (ifa);
1527 			} else {
1528 				/*
1529 				 * if we have a special address handler,
1530 				 * then use it instead of the generic one.
1531 				 */
1532 				if (ifa->ifa_claim_addr) {
1533 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1534 						return (ifa);
1535 					} else {
1536 						continue;
1537 					}
1538 				}
1539 
1540 				/*
1541 				 * Scan all the bits in the ifa's address.
1542 				 * If a bit dissagrees with what we are
1543 				 * looking for, mask it with the netmask
1544 				 * to see if it really matters.
1545 				 * (A byte at a time)
1546 				 */
1547 				if (ifa->ifa_netmask == 0)
1548 					continue;
1549 				cp = addr_data;
1550 				cp2 = ifa->ifa_addr->sa_data;
1551 				cp3 = ifa->ifa_netmask->sa_data;
1552 				cplim = ifa->ifa_netmask->sa_len +
1553 					(char *)ifa->ifa_netmask;
1554 				while (cp3 < cplim)
1555 					if ((*cp++ ^ *cp2++) & *cp3++)
1556 						goto next; /* next address! */
1557 				/*
1558 				 * If the netmask of what we just found
1559 				 * is more specific than what we had before
1560 				 * (if we had one) then remember the new one
1561 				 * before continuing to search for an even
1562 				 * better one.  If the netmasks are equal,
1563 				 * we prefer the this ifa based on the result
1564 				 * of ifa_prefer().
1565 				 */
1566 				if (ifa_maybe == NULL ||
1567 				    rn_refines(ifa->ifa_netmask,
1568 					       ifa_maybe->ifa_netmask) ||
1569 				    (sa_equal(ifa_maybe->ifa_netmask,
1570 					      ifa->ifa_netmask) &&
1571 				     ifa_prefer(ifa, ifa_maybe)))
1572 					ifa_maybe = ifa;
1573 			}
1574 		}
1575 	}
1576 	return (ifa_maybe);
1577 }
1578 
1579 /*
1580  * Find an interface address specific to an interface best matching
1581  * a given address.
1582  */
1583 struct ifaddr *
1584 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1585 {
1586 	struct ifaddr_container *ifac;
1587 	char *cp, *cp2, *cp3;
1588 	char *cplim;
1589 	struct ifaddr *ifa_maybe = NULL;
1590 	u_int af = addr->sa_family;
1591 
1592 	if (af >= AF_MAX)
1593 		return (0);
1594 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1595 		struct ifaddr *ifa = ifac->ifa;
1596 
1597 		if (ifa->ifa_addr->sa_family != af)
1598 			continue;
1599 		if (ifa_maybe == NULL)
1600 			ifa_maybe = ifa;
1601 		if (ifa->ifa_netmask == NULL) {
1602 			if (sa_equal(addr, ifa->ifa_addr) ||
1603 			    (ifa->ifa_dstaddr != NULL &&
1604 			     sa_equal(addr, ifa->ifa_dstaddr)))
1605 				return (ifa);
1606 			continue;
1607 		}
1608 		if (ifp->if_flags & IFF_POINTOPOINT) {
1609 			if (sa_equal(addr, ifa->ifa_dstaddr))
1610 				return (ifa);
1611 		} else {
1612 			cp = addr->sa_data;
1613 			cp2 = ifa->ifa_addr->sa_data;
1614 			cp3 = ifa->ifa_netmask->sa_data;
1615 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1616 			for (; cp3 < cplim; cp3++)
1617 				if ((*cp++ ^ *cp2++) & *cp3)
1618 					break;
1619 			if (cp3 == cplim)
1620 				return (ifa);
1621 		}
1622 	}
1623 	return (ifa_maybe);
1624 }
1625 
1626 struct netmsg_if {
1627 	struct netmsg_base	base;
1628 	struct ifnet		*ifp;
1629 };
1630 
1631 /*
1632  * Mark an interface down and notify protocols of the transition.
1633  */
1634 static void
1635 if_down_dispatch(netmsg_t nmsg)
1636 {
1637 	struct netmsg_if *msg = (struct netmsg_if *)nmsg;
1638 	struct ifnet *ifp = msg->ifp;
1639 	struct ifaddr_container *ifac;
1640 	struct domain *dp;
1641 
1642 	ASSERT_NETISR0;
1643 
1644 	ifp->if_flags &= ~IFF_UP;
1645 	getmicrotime(&ifp->if_lastchange);
1646 	rt_ifmsg(ifp);
1647 
1648 	/*
1649 	 * The ifaddr processing in the following loop will block,
1650 	 * however, this function is called in netisr0, in which
1651 	 * ifaddr list changes happen, so we don't care about the
1652 	 * blockness of the ifaddr processing here.
1653 	 */
1654 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1655 		struct ifaddr *ifa = ifac->ifa;
1656 
1657 		/* Ignore marker */
1658 		if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1659 			continue;
1660 
1661 		kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1662 	}
1663 
1664 	SLIST_FOREACH(dp, &domains, dom_next)
1665 		if (dp->dom_if_down != NULL)
1666 			dp->dom_if_down(ifp);
1667 
1668 	ifq_purge_all(&ifp->if_snd);
1669 	netisr_replymsg(&nmsg->base, 0);
1670 }
1671 
1672 /*
1673  * Mark an interface up and notify protocols of the transition.
1674  */
1675 static void
1676 if_up_dispatch(netmsg_t nmsg)
1677 {
1678 	struct netmsg_if *msg = (struct netmsg_if *)nmsg;
1679 	struct ifnet *ifp = msg->ifp;
1680 	struct ifaddr_container *ifac;
1681 	struct domain *dp;
1682 
1683 	ASSERT_NETISR0;
1684 
1685 	ifq_purge_all(&ifp->if_snd);
1686 	ifp->if_flags |= IFF_UP;
1687 	getmicrotime(&ifp->if_lastchange);
1688 	rt_ifmsg(ifp);
1689 
1690 	/*
1691 	 * The ifaddr processing in the following loop will block,
1692 	 * however, this function is called in netisr0, in which
1693 	 * ifaddr list changes happen, so we don't care about the
1694 	 * blockness of the ifaddr processing here.
1695 	 */
1696 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1697 		struct ifaddr *ifa = ifac->ifa;
1698 
1699 		/* Ignore marker */
1700 		if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1701 			continue;
1702 
1703 		kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1704 	}
1705 
1706 	SLIST_FOREACH(dp, &domains, dom_next)
1707 		if (dp->dom_if_up != NULL)
1708 			dp->dom_if_up(ifp);
1709 
1710 	netisr_replymsg(&nmsg->base, 0);
1711 }
1712 
1713 /*
1714  * Mark an interface down and notify protocols of the transition.  An
1715  * interface going down is also considered to be a synchronizing event.
1716  * We must ensure that all packet processing related to the interface
1717  * has completed before we return so e.g. the caller can free the ifnet
1718  * structure that the mbufs may be referencing.
1719  *
1720  * NOTE: must be called at splnet or eqivalent.
1721  */
1722 void
1723 if_down(struct ifnet *ifp)
1724 {
1725 	struct netmsg_if msg;
1726 
1727 	EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_DOWN);
1728 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1729 	    if_down_dispatch);
1730 	msg.ifp = ifp;
1731 	netisr_domsg(&msg.base, 0);
1732 	netmsg_service_sync();
1733 }
1734 
1735 /*
1736  * Mark an interface up and notify protocols of
1737  * the transition.
1738  * NOTE: must be called at splnet or eqivalent.
1739  */
1740 void
1741 if_up(struct ifnet *ifp)
1742 {
1743 	struct netmsg_if msg;
1744 
1745 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1746 	    if_up_dispatch);
1747 	msg.ifp = ifp;
1748 	netisr_domsg(&msg.base, 0);
1749 	EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_UP);
1750 }
1751 
1752 /*
1753  * Process a link state change.
1754  * NOTE: must be called at splsoftnet or equivalent.
1755  */
1756 void
1757 if_link_state_change(struct ifnet *ifp)
1758 {
1759 	int link_state = ifp->if_link_state;
1760 
1761 	rt_ifmsg(ifp);
1762 	devctl_notify("IFNET", ifp->if_xname,
1763 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1764 
1765 	EVENTHANDLER_INVOKE(ifnet_link_event, ifp, link_state);
1766 }
1767 
1768 /*
1769  * Handle interface watchdog timer routines.  Called
1770  * from softclock, we decrement timers (if set) and
1771  * call the appropriate interface routine on expiration.
1772  */
1773 static void
1774 if_slowtimo_dispatch(netmsg_t nmsg)
1775 {
1776 	struct globaldata *gd = mycpu;
1777 	const struct ifnet_array *arr;
1778 	int i;
1779 
1780 	ASSERT_NETISR0;
1781 
1782 	crit_enter_gd(gd);
1783 	lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1784 	crit_exit_gd(gd);
1785 
1786 	arr = ifnet_array_get();
1787 	for (i = 0; i < arr->ifnet_count; ++i) {
1788 		struct ifnet *ifp = arr->ifnet_arr[i];
1789 
1790 		crit_enter_gd(gd);
1791 
1792 		if (if_stats_compat) {
1793 			IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1794 			IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1795 			IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1796 			IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1797 			IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1798 			IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1799 			IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1800 			IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1801 			IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1802 			IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1803 			IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1804 			IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
1805 		}
1806 
1807 		if (ifp->if_timer == 0 || --ifp->if_timer) {
1808 			crit_exit_gd(gd);
1809 			continue;
1810 		}
1811 		if (ifp->if_watchdog) {
1812 			if (ifnet_tryserialize_all(ifp)) {
1813 				(*ifp->if_watchdog)(ifp);
1814 				ifnet_deserialize_all(ifp);
1815 			} else {
1816 				/* try again next timeout */
1817 				++ifp->if_timer;
1818 			}
1819 		}
1820 
1821 		crit_exit_gd(gd);
1822 	}
1823 
1824 	callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1825 }
1826 
1827 static void
1828 if_slowtimo(void *arg __unused)
1829 {
1830 	struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1831 
1832 	KASSERT(mycpuid == 0, ("not on cpu0"));
1833 	crit_enter();
1834 	if (lmsg->ms_flags & MSGF_DONE)
1835 		lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1836 	crit_exit();
1837 }
1838 
1839 /*
1840  * Map interface name to
1841  * interface structure pointer.
1842  */
1843 struct ifnet *
1844 ifunit(const char *name)
1845 {
1846 	struct ifnet *ifp;
1847 
1848 	/*
1849 	 * Search all the interfaces for this name/number
1850 	 */
1851 	KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1852 
1853 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1854 		if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1855 			break;
1856 	}
1857 	return (ifp);
1858 }
1859 
1860 struct ifnet *
1861 ifunit_netisr(const char *name)
1862 {
1863 	const struct ifnet_array *arr;
1864 	int i;
1865 
1866 	/*
1867 	 * Search all the interfaces for this name/number
1868 	 */
1869 
1870 	arr = ifnet_array_get();
1871 	for (i = 0; i < arr->ifnet_count; ++i) {
1872 		struct ifnet *ifp = arr->ifnet_arr[i];
1873 
1874 		if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1875 			return ifp;
1876 	}
1877 	return NULL;
1878 }
1879 
1880 /*
1881  * Interface ioctls.
1882  */
1883 int
1884 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1885 {
1886 	struct ifnet *ifp;
1887 	struct ifgroupreq *ifgr;
1888 	struct ifreq *ifr;
1889 	struct ifstat *ifs;
1890 	int error, do_ifup = 0;
1891 	short oif_flags;
1892 	int new_flags;
1893 	size_t namelen, onamelen;
1894 	size_t descrlen;
1895 	char *descrbuf, *odescrbuf;
1896 	char new_name[IFNAMSIZ];
1897 	struct ifaddr *ifa;
1898 	struct sockaddr_dl *sdl;
1899 
1900 	switch (cmd) {
1901 	case SIOCGIFCONF:
1902 		return (ifconf(cmd, data, cred));
1903 	default:
1904 		break;
1905 	}
1906 
1907 	ifr = (struct ifreq *)data;
1908 
1909 	switch (cmd) {
1910 	case SIOCIFCREATE:
1911 	case SIOCIFCREATE2:
1912 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
1913 		if (error)
1914 			return (error);
1915 		return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1916 			(cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL), NULL));
1917 	case SIOCIFDESTROY:
1918 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
1919 		if (error)
1920 			return (error);
1921 		return (if_clone_destroy(ifr->ifr_name));
1922 	case SIOCIFGCLONERS:
1923 		return (if_clone_list((struct if_clonereq *)data));
1924 	case SIOCGIFGMEMB:
1925 		return (if_getgroupmembers((struct ifgroupreq *)data));
1926 	default:
1927 		break;
1928 	}
1929 
1930 	/*
1931 	 * Nominal ioctl through interface, lookup the ifp and obtain a
1932 	 * lock to serialize the ifconfig ioctl operation.
1933 	 */
1934 	ifnet_lock();
1935 
1936 	ifp = ifunit(ifr->ifr_name);
1937 	if (ifp == NULL) {
1938 		ifnet_unlock();
1939 		return (ENXIO);
1940 	}
1941 	error = 0;
1942 
1943 	switch (cmd) {
1944 	case SIOCGIFINDEX:
1945 		ifr->ifr_index = ifp->if_index;
1946 		break;
1947 
1948 	case SIOCGIFFLAGS:
1949 		ifr->ifr_flags = ifp->if_flags;
1950 		ifr->ifr_flagshigh = ifp->if_flags >> 16;
1951 		break;
1952 
1953 	case SIOCGIFCAP:
1954 		ifr->ifr_reqcap = ifp->if_capabilities;
1955 		ifr->ifr_curcap = ifp->if_capenable;
1956 		break;
1957 
1958 	case SIOCGIFMETRIC:
1959 		ifr->ifr_metric = ifp->if_metric;
1960 		break;
1961 
1962 	case SIOCGIFMTU:
1963 		ifr->ifr_mtu = ifp->if_mtu;
1964 		break;
1965 
1966 	case SIOCGIFTSOLEN:
1967 		ifr->ifr_tsolen = ifp->if_tsolen;
1968 		break;
1969 
1970 	case SIOCGIFDATA:
1971 		error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1972 				sizeof(ifp->if_data));
1973 		break;
1974 
1975 	case SIOCGIFPHYS:
1976 		ifr->ifr_phys = ifp->if_physical;
1977 		break;
1978 
1979 	case SIOCGIFPOLLCPU:
1980 		ifr->ifr_pollcpu = -1;
1981 		break;
1982 
1983 	case SIOCSIFPOLLCPU:
1984 		break;
1985 
1986 	case SIOCGIFDESCR:
1987 		error = 0;
1988 		ifnet_lock();
1989 		if (ifp->if_description == NULL) {
1990 			ifr->ifr_buffer.length = 0;
1991 			error = ENOMSG;
1992 		} else {
1993 			/* space for terminating nul */
1994 			descrlen = strlen(ifp->if_description) + 1;
1995 			if (ifr->ifr_buffer.length < descrlen)
1996 				error = ENAMETOOLONG;
1997 			else
1998 				error = copyout(ifp->if_description,
1999 				    ifr->ifr_buffer.buffer, descrlen);
2000 			ifr->ifr_buffer.length = descrlen;
2001 		}
2002 		ifnet_unlock();
2003 		break;
2004 
2005 	case SIOCSIFDESCR:
2006 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
2007 		if (error)
2008 			break;
2009 
2010 		/*
2011 		 * Copy only (length-1) bytes to make sure that
2012 		 * if_description is always nul terminated.  The
2013 		 * length parameter is supposed to count the
2014 		 * terminating nul in.
2015 		 */
2016 		if (ifr->ifr_buffer.length > ifdescr_maxlen)
2017 			return (ENAMETOOLONG);
2018 		else if (ifr->ifr_buffer.length == 0)
2019 			descrbuf = NULL;
2020 		else {
2021 			descrbuf = kmalloc(ifr->ifr_buffer.length, M_IFDESCR,
2022 			    M_WAITOK | M_ZERO);
2023 			error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2024 			    ifr->ifr_buffer.length - 1);
2025 			if (error) {
2026 				kfree(descrbuf, M_IFDESCR);
2027 				break;
2028 			}
2029 		}
2030 
2031 		ifnet_lock();
2032 		odescrbuf = ifp->if_description;
2033 		ifp->if_description = descrbuf;
2034 		ifnet_unlock();
2035 
2036 		if (odescrbuf)
2037 			kfree(odescrbuf, M_IFDESCR);
2038 
2039 	case SIOCSIFFLAGS:
2040 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
2041 		if (error)
2042 			break;
2043 		new_flags = (ifr->ifr_flags & 0xffff) |
2044 		    (ifr->ifr_flagshigh << 16);
2045 		if (ifp->if_flags & IFF_SMART) {
2046 			/* Smart drivers twiddle their own routes */
2047 		} else if (ifp->if_flags & IFF_UP &&
2048 		    (new_flags & IFF_UP) == 0) {
2049 			if_down(ifp);
2050 		} else if (new_flags & IFF_UP &&
2051 		    (ifp->if_flags & IFF_UP) == 0) {
2052 			do_ifup = 1;
2053 		}
2054 
2055 #ifdef IFPOLL_ENABLE
2056 		if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
2057 			if (new_flags & IFF_NPOLLING)
2058 				ifpoll_register(ifp);
2059 			else
2060 				ifpoll_deregister(ifp);
2061 		}
2062 #endif
2063 
2064 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2065 			(new_flags &~ IFF_CANTCHANGE);
2066 		if (new_flags & IFF_PPROMISC) {
2067 			/* Permanently promiscuous mode requested */
2068 			ifp->if_flags |= IFF_PROMISC;
2069 		} else if (ifp->if_pcount == 0) {
2070 			ifp->if_flags &= ~IFF_PROMISC;
2071 		}
2072 		if (ifp->if_ioctl) {
2073 			ifnet_serialize_all(ifp);
2074 			ifp->if_ioctl(ifp, cmd, data, cred);
2075 			ifnet_deserialize_all(ifp);
2076 		}
2077 		if (do_ifup)
2078 			if_up(ifp);
2079 		getmicrotime(&ifp->if_lastchange);
2080 		break;
2081 
2082 	case SIOCSIFCAP:
2083 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
2084 		if (error)
2085 			break;
2086 		if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
2087 			error = EINVAL;
2088 			break;
2089 		}
2090 		ifnet_serialize_all(ifp);
2091 		ifp->if_ioctl(ifp, cmd, data, cred);
2092 		ifnet_deserialize_all(ifp);
2093 		break;
2094 
2095 	case SIOCSIFNAME:
2096 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
2097 		if (error)
2098 			break;
2099 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2100 		if (error)
2101 			break;
2102 		if (new_name[0] == '\0') {
2103 			error = EINVAL;
2104 			break;
2105 		}
2106 		if (ifunit(new_name) != NULL) {
2107 			error = EEXIST;
2108 			break;
2109 		}
2110 
2111 		EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
2112 
2113 		/* Announce the departure of the interface. */
2114 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2115 
2116 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2117 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
2118 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2119 		namelen = strlen(new_name);
2120 		onamelen = sdl->sdl_nlen;
2121 		/*
2122 		 * Move the address if needed.  This is safe because we
2123 		 * allocate space for a name of length IFNAMSIZ when we
2124 		 * create this in if_attach().
2125 		 */
2126 		if (namelen != onamelen) {
2127 			bcopy(sdl->sdl_data + onamelen,
2128 			    sdl->sdl_data + namelen, sdl->sdl_alen);
2129 		}
2130 		bcopy(new_name, sdl->sdl_data, namelen);
2131 		sdl->sdl_nlen = namelen;
2132 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2133 		bzero(sdl->sdl_data, onamelen);
2134 		while (namelen != 0)
2135 			sdl->sdl_data[--namelen] = 0xff;
2136 
2137 		EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
2138 
2139 		/* Announce the return of the interface. */
2140 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2141 		break;
2142 
2143 	case SIOCSIFMETRIC:
2144 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
2145 		if (error)
2146 			break;
2147 		ifp->if_metric = ifr->ifr_metric;
2148 		getmicrotime(&ifp->if_lastchange);
2149 		break;
2150 
2151 	case SIOCSIFPHYS:
2152 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
2153 		if (error)
2154 			break;
2155 		if (ifp->if_ioctl == NULL) {
2156 		        error = EOPNOTSUPP;
2157 			break;
2158 		}
2159 		ifnet_serialize_all(ifp);
2160 		error = ifp->if_ioctl(ifp, cmd, data, cred);
2161 		ifnet_deserialize_all(ifp);
2162 		if (error == 0)
2163 			getmicrotime(&ifp->if_lastchange);
2164 		break;
2165 
2166 	case SIOCSIFMTU:
2167 	{
2168 		u_long oldmtu = ifp->if_mtu;
2169 
2170 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
2171 		if (error)
2172 			break;
2173 		if (ifp->if_ioctl == NULL) {
2174 			error = EOPNOTSUPP;
2175 			break;
2176 		}
2177 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2178 			error = EINVAL;
2179 			break;
2180 		}
2181 		ifnet_serialize_all(ifp);
2182 		error = ifp->if_ioctl(ifp, cmd, data, cred);
2183 		ifnet_deserialize_all(ifp);
2184 		if (error == 0) {
2185 			getmicrotime(&ifp->if_lastchange);
2186 			rt_ifmsg(ifp);
2187 		}
2188 		/*
2189 		 * If the link MTU changed, do network layer specific procedure.
2190 		 */
2191 		if (ifp->if_mtu != oldmtu) {
2192 #ifdef INET6
2193 			nd6_setmtu(ifp);
2194 #endif
2195 		}
2196 		break;
2197 	}
2198 
2199 	case SIOCSIFTSOLEN:
2200 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
2201 		if (error)
2202 			break;
2203 
2204 		/* XXX need driver supplied upper limit */
2205 		if (ifr->ifr_tsolen <= 0) {
2206 			error = EINVAL;
2207 			break;
2208 		}
2209 		ifp->if_tsolen = ifr->ifr_tsolen;
2210 		break;
2211 
2212 	case SIOCADDMULTI:
2213 	case SIOCDELMULTI:
2214 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
2215 		if (error)
2216 			break;
2217 
2218 		/* Don't allow group membership on non-multicast interfaces. */
2219 		if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2220 			error = EOPNOTSUPP;
2221 			break;
2222 		}
2223 
2224 		/* Don't let users screw up protocols' entries. */
2225 		if (ifr->ifr_addr.sa_family != AF_LINK) {
2226 			error = EINVAL;
2227 			break;
2228 		}
2229 
2230 		if (cmd == SIOCADDMULTI) {
2231 			struct ifmultiaddr *ifma;
2232 			error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2233 		} else {
2234 			error = if_delmulti(ifp, &ifr->ifr_addr);
2235 		}
2236 		if (error == 0)
2237 			getmicrotime(&ifp->if_lastchange);
2238 		break;
2239 
2240 	case SIOCSIFPHYADDR:
2241 	case SIOCDIFPHYADDR:
2242 #ifdef INET6
2243 	case SIOCSIFPHYADDR_IN6:
2244 #endif
2245 	case SIOCSLIFPHYADDR:
2246 	case SIOCSIFMEDIA:
2247 	case SIOCSIFGENERIC:
2248 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
2249 		if (error)
2250 			break;
2251 		if (ifp->if_ioctl == NULL) {
2252 			error = EOPNOTSUPP;
2253 			break;
2254 		}
2255 		ifnet_serialize_all(ifp);
2256 		error = ifp->if_ioctl(ifp, cmd, data, cred);
2257 		ifnet_deserialize_all(ifp);
2258 		if (error == 0)
2259 			getmicrotime(&ifp->if_lastchange);
2260 		break;
2261 
2262 	case SIOCGIFSTATUS:
2263 		ifs = (struct ifstat *)data;
2264 		ifs->ascii[0] = '\0';
2265 		/* fall through */
2266 	case SIOCGIFPSRCADDR:
2267 	case SIOCGIFPDSTADDR:
2268 	case SIOCGLIFPHYADDR:
2269 	case SIOCGIFMEDIA:
2270 	case SIOCGIFXMEDIA:
2271 	case SIOCGIFGENERIC:
2272 		if (ifp->if_ioctl == NULL) {
2273 			error = EOPNOTSUPP;
2274 			break;
2275 		}
2276 		ifnet_serialize_all(ifp);
2277 		error = ifp->if_ioctl(ifp, cmd, data, cred);
2278 		ifnet_deserialize_all(ifp);
2279 		break;
2280 
2281 	case SIOCSIFLLADDR:
2282 		error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT);
2283 		if (error)
2284 			break;
2285 		error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2286 				     ifr->ifr_addr.sa_len);
2287 		EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2288 		break;
2289 
2290 	case SIOCAIFGROUP:
2291 		ifgr = (struct ifgroupreq *)ifr;
2292 		error = caps_priv_check(cred, SYSCAP_NONET_IFCONFIG);
2293 		if (error)
2294 			return (error);
2295 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2296 			return (error);
2297 		break;
2298 
2299 	case SIOCDIFGROUP:
2300 		ifgr = (struct ifgroupreq *)ifr;
2301 		error = caps_priv_check(cred, SYSCAP_NONET_IFCONFIG);
2302 		if (error)
2303 			return (error);
2304 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2305 			return (error);
2306 		break;
2307 
2308 	case SIOCGIFGROUP:
2309 		ifgr = (struct ifgroupreq *)ifr;
2310 		if ((error = if_getgroups(ifgr, ifp)))
2311 			return (error);
2312 		break;
2313 
2314 	default:
2315 		oif_flags = ifp->if_flags;
2316 		if (so->so_proto == 0) {
2317 			error = EOPNOTSUPP;
2318 			break;
2319 		}
2320 		error = so_pru_control_direct(so, cmd, data, ifp);
2321 
2322 		/*
2323 		 * If the socket control method returns EOPNOTSUPP, pass the
2324 		 * request directly to the interface.
2325 		 *
2326 		 * Exclude the SIOCSIF{ADDR,BRDADDR,DSTADDR,NETMASK} ioctls,
2327 		 * because drivers may trust these ioctls to come from an
2328 		 * already privileged layer and thus do not perform credentials
2329 		 * checks or input validation.
2330 		 */
2331 		if (error == EOPNOTSUPP &&
2332 		    ifp->if_ioctl != NULL &&
2333 		    cmd != SIOCSIFADDR &&
2334 		    cmd != SIOCSIFBRDADDR &&
2335 		    cmd != SIOCSIFDSTADDR &&
2336 		    cmd != SIOCSIFNETMASK) {
2337 			ifnet_serialize_all(ifp);
2338 			error = ifp->if_ioctl(ifp, cmd, data, cred);
2339 			ifnet_deserialize_all(ifp);
2340 		}
2341 
2342 		if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2343 #ifdef INET6
2344 			DELAY(100);/* XXX: temporary workaround for fxp issue*/
2345 			if (ifp->if_flags & IFF_UP) {
2346 				crit_enter();
2347 				in6_if_up(ifp);
2348 				crit_exit();
2349 			}
2350 #endif
2351 		}
2352 		break;
2353 	}
2354 
2355 	ifnet_unlock();
2356 	return (error);
2357 }
2358 
2359 /*
2360  * Set/clear promiscuous mode on interface ifp based on the truth value
2361  * of pswitch.  The calls are reference counted so that only the first
2362  * "on" request actually has an effect, as does the final "off" request.
2363  * Results are undefined if the "off" and "on" requests are not matched.
2364  */
2365 int
2366 ifpromisc(struct ifnet *ifp, int pswitch)
2367 {
2368 	struct ifreq ifr;
2369 	int error;
2370 	int oldflags;
2371 
2372 	oldflags = ifp->if_flags;
2373 	if (ifp->if_flags & IFF_PPROMISC) {
2374 		/* Do nothing if device is in permanently promiscuous mode */
2375 		ifp->if_pcount += pswitch ? 1 : -1;
2376 		return (0);
2377 	}
2378 	if (pswitch) {
2379 		/*
2380 		 * If the device is not configured up, we cannot put it in
2381 		 * promiscuous mode.
2382 		 */
2383 		if ((ifp->if_flags & IFF_UP) == 0)
2384 			return (ENETDOWN);
2385 		if (ifp->if_pcount++ != 0)
2386 			return (0);
2387 		ifp->if_flags |= IFF_PROMISC;
2388 		log(LOG_INFO, "%s: promiscuous mode enabled\n",
2389 		    ifp->if_xname);
2390 	} else {
2391 		if (--ifp->if_pcount > 0)
2392 			return (0);
2393 		ifp->if_flags &= ~IFF_PROMISC;
2394 		log(LOG_INFO, "%s: promiscuous mode disabled\n",
2395 		    ifp->if_xname);
2396 	}
2397 	ifr.ifr_flags = ifp->if_flags;
2398 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2399 	ifnet_serialize_all(ifp);
2400 	error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2401 	ifnet_deserialize_all(ifp);
2402 	if (error == 0)
2403 		rt_ifmsg(ifp);
2404 	else
2405 		ifp->if_flags = oldflags;
2406 	return error;
2407 }
2408 
2409 /*
2410  * Return interface configuration
2411  * of system.  List may be used
2412  * in later ioctl's (above) to get
2413  * other information.
2414  */
2415 static int
2416 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2417 {
2418 	struct ifconf *ifc = (struct ifconf *)data;
2419 	struct ifnet *ifp;
2420 	struct sockaddr *sa;
2421 	struct ifreq ifr, *ifrp;
2422 	int space = ifc->ifc_len, error = 0;
2423 
2424 	ifrp = ifc->ifc_req;
2425 
2426 	ifnet_lock();
2427 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2428 		struct ifaddr_container *ifac, *ifac_mark;
2429 		struct ifaddr_marker mark;
2430 		struct ifaddrhead *head;
2431 		int addrs;
2432 
2433 		if (space <= sizeof ifr)
2434 			break;
2435 
2436 		/*
2437 		 * Zero the stack declared structure first to prevent
2438 		 * memory disclosure.
2439 		 */
2440 		bzero(&ifr, sizeof(ifr));
2441 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2442 		    >= sizeof(ifr.ifr_name)) {
2443 			error = ENAMETOOLONG;
2444 			break;
2445 		}
2446 
2447 		/*
2448 		 * Add a marker, since copyout() could block and during that
2449 		 * period the list could be changed.  Inserting the marker to
2450 		 * the header of the list will not cause trouble for the code
2451 		 * assuming that the first element of the list is AF_LINK; the
2452 		 * marker will be moved to the next position w/o blocking.
2453 		 */
2454 		ifa_marker_init(&mark, ifp);
2455 		ifac_mark = &mark.ifac;
2456 		head = &ifp->if_addrheads[mycpuid];
2457 
2458 		addrs = 0;
2459 		TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2460 		while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2461 			struct ifaddr *ifa = ifac->ifa;
2462 
2463 			TAILQ_REMOVE(head, ifac_mark, ifa_link);
2464 			TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2465 
2466 			/* Ignore marker */
2467 			if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2468 				continue;
2469 
2470 			if (space <= sizeof ifr)
2471 				break;
2472 			sa = ifa->ifa_addr;
2473 			if (cred->cr_prison && prison_if(cred, sa))
2474 				continue;
2475 			addrs++;
2476 			/*
2477 			 * Keep a reference on this ifaddr, so that it will
2478 			 * not be destroyed when its address is copied to
2479 			 * the userland, which could block.
2480 			 */
2481 			IFAREF(ifa);
2482 			if (sa->sa_len <= sizeof(*sa)) {
2483 				ifr.ifr_addr = *sa;
2484 				error = copyout(&ifr, ifrp, sizeof ifr);
2485 				ifrp++;
2486 			} else {
2487 				if (space < (sizeof ifr) + sa->sa_len -
2488 					    sizeof(*sa)) {
2489 					IFAFREE(ifa);
2490 					break;
2491 				}
2492 				space -= sa->sa_len - sizeof(*sa);
2493 				error = copyout(&ifr, ifrp,
2494 						sizeof ifr.ifr_name);
2495 				if (error == 0)
2496 					error = copyout(sa, &ifrp->ifr_addr,
2497 							sa->sa_len);
2498 				ifrp = (struct ifreq *)
2499 					(sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2500 			}
2501 			IFAFREE(ifa);
2502 			if (error)
2503 				break;
2504 			space -= sizeof ifr;
2505 		}
2506 		TAILQ_REMOVE(head, ifac_mark, ifa_link);
2507 		if (error)
2508 			break;
2509 		if (!addrs) {
2510 			bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2511 			error = copyout(&ifr, ifrp, sizeof ifr);
2512 			if (error)
2513 				break;
2514 			space -= sizeof ifr;
2515 			ifrp++;
2516 		}
2517 	}
2518 	ifnet_unlock();
2519 
2520 	ifc->ifc_len -= space;
2521 	return (error);
2522 }
2523 
2524 /*
2525  * Just like if_promisc(), but for all-multicast-reception mode.
2526  */
2527 int
2528 if_allmulti(struct ifnet *ifp, int onswitch)
2529 {
2530 	int error = 0;
2531 	struct ifreq ifr;
2532 
2533 	crit_enter();
2534 
2535 	if (onswitch) {
2536 		if (ifp->if_amcount++ == 0) {
2537 			ifp->if_flags |= IFF_ALLMULTI;
2538 			ifr.ifr_flags = ifp->if_flags;
2539 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2540 			ifnet_serialize_all(ifp);
2541 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2542 					      NULL);
2543 			ifnet_deserialize_all(ifp);
2544 		}
2545 	} else {
2546 		if (ifp->if_amcount > 1) {
2547 			ifp->if_amcount--;
2548 		} else {
2549 			ifp->if_amcount = 0;
2550 			ifp->if_flags &= ~IFF_ALLMULTI;
2551 			ifr.ifr_flags = ifp->if_flags;
2552 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2553 			ifnet_serialize_all(ifp);
2554 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2555 					      NULL);
2556 			ifnet_deserialize_all(ifp);
2557 		}
2558 	}
2559 
2560 	crit_exit();
2561 
2562 	if (error == 0)
2563 		rt_ifmsg(ifp);
2564 	return error;
2565 }
2566 
2567 /*
2568  * Add a multicast listenership to the interface in question.
2569  * The link layer provides a routine which converts
2570  */
2571 int
2572 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2573     struct ifmultiaddr **retifma)
2574 {
2575 	struct sockaddr *llsa, *dupsa;
2576 	int error;
2577 	struct ifmultiaddr *ifma;
2578 
2579 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
2580 
2581 	/*
2582 	 * If the matching multicast address already exists
2583 	 * then don't add a new one, just add a reference
2584 	 */
2585 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2586 		if (sa_equal(sa, ifma->ifma_addr)) {
2587 			ifma->ifma_refcount++;
2588 			if (retifma)
2589 				*retifma = ifma;
2590 			return 0;
2591 		}
2592 	}
2593 
2594 	/*
2595 	 * Give the link layer a chance to accept/reject it, and also
2596 	 * find out which AF_LINK address this maps to, if it isn't one
2597 	 * already.
2598 	 */
2599 	if (ifp->if_resolvemulti) {
2600 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2601 		if (error)
2602 			return error;
2603 	} else {
2604 		llsa = NULL;
2605 	}
2606 
2607 	ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2608 	dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2609 	bcopy(sa, dupsa, sa->sa_len);
2610 
2611 	ifma->ifma_addr = dupsa;
2612 	ifma->ifma_lladdr = llsa;
2613 	ifma->ifma_ifp = ifp;
2614 	ifma->ifma_refcount = 1;
2615 	ifma->ifma_protospec = NULL;
2616 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2617 
2618 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2619 	if (retifma)
2620 		*retifma = ifma;
2621 
2622 	if (llsa != NULL) {
2623 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2624 			if (sa_equal(ifma->ifma_addr, llsa))
2625 				break;
2626 		}
2627 		if (ifma) {
2628 			ifma->ifma_refcount++;
2629 		} else {
2630 			ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2631 			dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2632 			bcopy(llsa, dupsa, llsa->sa_len);
2633 			ifma->ifma_addr = dupsa;
2634 			ifma->ifma_ifp = ifp;
2635 			ifma->ifma_refcount = 1;
2636 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2637 		}
2638 	}
2639 	/*
2640 	 * We are certain we have added something, so call down to the
2641 	 * interface to let them know about it.
2642 	 */
2643 	if (ifp->if_ioctl)
2644 		ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2645 
2646 	return 0;
2647 }
2648 
2649 int
2650 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2651     struct ifmultiaddr **retifma)
2652 {
2653 	int error;
2654 
2655 	ifnet_serialize_all(ifp);
2656 	error = if_addmulti_serialized(ifp, sa, retifma);
2657 	ifnet_deserialize_all(ifp);
2658 
2659 	return error;
2660 }
2661 
2662 /*
2663  * Remove a reference to a multicast address on this interface.  Yell
2664  * if the request does not match an existing membership.
2665  */
2666 static int
2667 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2668 {
2669 	struct ifmultiaddr *ifma;
2670 
2671 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
2672 
2673 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2674 		if (sa_equal(sa, ifma->ifma_addr))
2675 			break;
2676 	if (ifma == NULL)
2677 		return ENOENT;
2678 
2679 	if (ifma->ifma_refcount > 1) {
2680 		ifma->ifma_refcount--;
2681 		return 0;
2682 	}
2683 
2684 	rt_newmaddrmsg(RTM_DELMADDR, ifma);
2685 	sa = ifma->ifma_lladdr;
2686 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2687 	/*
2688 	 * Make sure the interface driver is notified
2689 	 * in the case of a link layer mcast group being left.
2690 	 */
2691 	if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2692 		ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2693 	kfree(ifma->ifma_addr, M_IFMADDR);
2694 	kfree(ifma, M_IFMADDR);
2695 	if (sa == NULL)
2696 		return 0;
2697 
2698 	/*
2699 	 * Now look for the link-layer address which corresponds to
2700 	 * this network address.  It had been squirreled away in
2701 	 * ifma->ifma_lladdr for this purpose (so we don't have
2702 	 * to call ifp->if_resolvemulti() again), and we saved that
2703 	 * value in sa above.  If some nasty deleted the
2704 	 * link-layer address out from underneath us, we can deal because
2705 	 * the address we stored was is not the same as the one which was
2706 	 * in the record for the link-layer address.  (So we don't complain
2707 	 * in that case.)
2708 	 */
2709 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2710 		if (sa_equal(sa, ifma->ifma_addr))
2711 			break;
2712 	if (ifma == NULL)
2713 		return 0;
2714 
2715 	if (ifma->ifma_refcount > 1) {
2716 		ifma->ifma_refcount--;
2717 		return 0;
2718 	}
2719 
2720 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2721 	ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2722 	kfree(ifma->ifma_addr, M_IFMADDR);
2723 	kfree(sa, M_IFMADDR);
2724 	kfree(ifma, M_IFMADDR);
2725 
2726 	return 0;
2727 }
2728 
2729 int
2730 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2731 {
2732 	int error;
2733 
2734 	ifnet_serialize_all(ifp);
2735 	error = if_delmulti_serialized(ifp, sa);
2736 	ifnet_deserialize_all(ifp);
2737 
2738 	return error;
2739 }
2740 
2741 /*
2742  * Delete all multicast group membership for an interface.
2743  * Should be used to quickly flush all multicast filters.
2744  */
2745 void
2746 if_delallmulti_serialized(struct ifnet *ifp)
2747 {
2748 	struct ifmultiaddr *ifma, mark;
2749 	struct sockaddr sa;
2750 
2751 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
2752 
2753 	bzero(&sa, sizeof(sa));
2754 	sa.sa_family = AF_UNSPEC;
2755 	sa.sa_len = sizeof(sa);
2756 
2757 	bzero(&mark, sizeof(mark));
2758 	mark.ifma_addr = &sa;
2759 
2760 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2761 	while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2762 		TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2763 		TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2764 		    ifma_link);
2765 
2766 		if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2767 			continue;
2768 
2769 		if_delmulti_serialized(ifp, ifma->ifma_addr);
2770 	}
2771 	TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2772 }
2773 
2774 
2775 /*
2776  * Set the link layer address on an interface.
2777  *
2778  * At this time we only support certain types of interfaces,
2779  * and we don't allow the length of the address to change.
2780  */
2781 int
2782 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2783 {
2784 	struct sockaddr_dl *sdl;
2785 	struct ifreq ifr;
2786 
2787 	sdl = IF_LLSOCKADDR(ifp);
2788 	if (sdl == NULL)
2789 		return (EINVAL);
2790 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2791 		return (EINVAL);
2792 	switch (ifp->if_type) {
2793 	case IFT_ETHER:			/* these types use struct arpcom */
2794 	case IFT_XETHER:
2795 	case IFT_L2VLAN:
2796 	case IFT_IEEE8023ADLAG:
2797 		bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2798 		bcopy(lladdr, LLADDR(sdl), len);
2799 		break;
2800 	default:
2801 		return (ENODEV);
2802 	}
2803 	/*
2804 	 * If the interface is already up, we need
2805 	 * to re-init it in order to reprogram its
2806 	 * address filter.
2807 	 */
2808 	ifnet_serialize_all(ifp);
2809 	if ((ifp->if_flags & IFF_UP) != 0) {
2810 #ifdef INET
2811 		struct ifaddr_container *ifac;
2812 #endif
2813 
2814 		ifp->if_flags &= ~IFF_UP;
2815 		ifr.ifr_flags = ifp->if_flags;
2816 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
2817 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2818 			      NULL);
2819 		ifp->if_flags |= IFF_UP;
2820 		ifr.ifr_flags = ifp->if_flags;
2821 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
2822 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2823 				 NULL);
2824 #ifdef INET
2825 		/*
2826 		 * Also send gratuitous ARPs to notify other nodes about
2827 		 * the address change.
2828 		 */
2829 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2830 			struct ifaddr *ifa = ifac->ifa;
2831 
2832 			if (ifa->ifa_addr != NULL &&
2833 			    ifa->ifa_addr->sa_family == AF_INET)
2834 				arp_gratuitous(ifp, ifa);
2835 		}
2836 #endif
2837 	}
2838 	ifnet_deserialize_all(ifp);
2839 	return (0);
2840 }
2841 
2842 
2843 /*
2844  * Tunnel interfaces can nest, also they may cause infinite recursion
2845  * calls when misconfigured.  Introduce an upper limit to prevent infinite
2846  * recursions, as well as to constrain the nesting depth.
2847  *
2848  * Return 0, if tunnel nesting count is equal or less than limit.
2849  */
2850 int
2851 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, uint32_t cookie,
2852 			int limit)
2853 {
2854 	struct m_tag *mtag;
2855 	int count;
2856 
2857 	count = 1;
2858 	mtag = m_tag_locate(m, cookie, 0 /* type */, NULL);
2859 	if (mtag != NULL)
2860 		count += *(int *)(mtag + 1);
2861 	if (count > limit) {
2862 		log(LOG_NOTICE,
2863 		    "%s: packet looped too many times (%d), limit %d\n",
2864 		    ifp->if_xname, count, limit);
2865 		return (ELOOP);
2866 	}
2867 
2868 	if (mtag == NULL) {
2869 		mtag = m_tag_alloc(cookie, 0, sizeof(int), M_NOWAIT);
2870 		if (mtag == NULL)
2871 			return (ENOMEM);
2872 		m_tag_prepend(m, mtag);
2873 	}
2874 
2875 	*(int *)(mtag + 1) = count;
2876 	return (0);
2877 }
2878 
2879 
2880 /*
2881  * Locate an interface based on a complete address.
2882  */
2883 struct ifnet *
2884 if_bylla(const void *lla, unsigned char lla_len)
2885 {
2886 	const struct ifnet_array *arr;
2887 	struct ifnet *ifp;
2888 	struct sockaddr_dl *sdl;
2889 	int i;
2890 
2891 	arr = ifnet_array_get();
2892 	for (i = 0; i < arr->ifnet_count; ++i) {
2893 		ifp = arr->ifnet_arr[i];
2894 		if (ifp->if_addrlen != lla_len)
2895 			continue;
2896 
2897 		sdl = IF_LLSOCKADDR(ifp);
2898 		if (memcmp(lla, LLADDR(sdl), lla_len) == 0)
2899 			return (ifp);
2900 	}
2901 	return (NULL);
2902 }
2903 
2904 struct ifmultiaddr *
2905 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2906 {
2907 	struct ifmultiaddr *ifma;
2908 
2909 	/* TODO: need ifnet_serialize_main */
2910 	ifnet_serialize_all(ifp);
2911 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2912 		if (sa_equal(ifma->ifma_addr, sa))
2913 			break;
2914 	ifnet_deserialize_all(ifp);
2915 
2916 	return ifma;
2917 }
2918 
2919 /*
2920  * This function locates the first real ethernet MAC from a network
2921  * card and loads it into node, returning 0 on success or ENOENT if
2922  * no suitable interfaces were found.  It is used by the uuid code to
2923  * generate a unique 6-byte number.
2924  */
2925 int
2926 if_getanyethermac(uint16_t *node, int minlen)
2927 {
2928 	struct ifnet *ifp;
2929 	struct sockaddr_dl *sdl;
2930 
2931 	ifnet_lock();
2932 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2933 		if (ifp->if_type != IFT_ETHER)
2934 			continue;
2935 		sdl = IF_LLSOCKADDR(ifp);
2936 		if (sdl->sdl_alen < minlen)
2937 			continue;
2938 		bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2939 		      minlen);
2940 		ifnet_unlock();
2941 		return(0);
2942 	}
2943 	ifnet_unlock();
2944 	return (ENOENT);
2945 }
2946 
2947 /*
2948  * The name argument must be a pointer to storage which will last as
2949  * long as the interface does.  For physical devices, the result of
2950  * device_get_name(dev) is a good choice and for pseudo-devices a
2951  * static string works well.
2952  */
2953 void
2954 if_initname(struct ifnet *ifp, const char *name, int unit)
2955 {
2956 	ifp->if_dname = name;
2957 	ifp->if_dunit = unit;
2958 	if (unit != IF_DUNIT_NONE)
2959 		ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2960 	else
2961 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2962 }
2963 
2964 int
2965 if_printf(struct ifnet *ifp, const char *fmt, ...)
2966 {
2967 	__va_list ap;
2968 	int retval;
2969 
2970 	retval = kprintf("%s: ", ifp->if_xname);
2971 	__va_start(ap, fmt);
2972 	retval += kvprintf(fmt, ap);
2973 	__va_end(ap);
2974 	return (retval);
2975 }
2976 
2977 struct ifnet *
2978 if_alloc(uint8_t type)
2979 {
2980 	struct ifnet *ifp;
2981 	size_t size;
2982 
2983 	/*
2984 	 * XXX temporary hack until arpcom is setup in if_l2com
2985 	 */
2986 	if (type == IFT_ETHER)
2987 		size = sizeof(struct arpcom);
2988 	else
2989 		size = sizeof(struct ifnet);
2990 
2991 	ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2992 
2993 	ifp->if_type = type;
2994 
2995 	if (if_com_alloc[type] != NULL) {
2996 		ifp->if_l2com = if_com_alloc[type](type, ifp);
2997 		if (ifp->if_l2com == NULL) {
2998 			kfree(ifp, M_IFNET);
2999 			return (NULL);
3000 		}
3001 	}
3002 	return (ifp);
3003 }
3004 
3005 void
3006 if_free(struct ifnet *ifp)
3007 {
3008 	if (ifp->if_description != NULL)
3009 		kfree(ifp->if_description, M_IFDESCR);
3010 	kfree(ifp, M_IFNET);
3011 }
3012 
3013 void
3014 ifq_set_classic(struct ifaltq *ifq)
3015 {
3016 	ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
3017 	    ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
3018 }
3019 
3020 void
3021 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
3022     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
3023 {
3024 	int q;
3025 
3026 	KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
3027 	KASSERT(enqueue != NULL, ("enqueue is not specified"));
3028 	KASSERT(dequeue != NULL, ("dequeue is not specified"));
3029 	KASSERT(request != NULL, ("request is not specified"));
3030 
3031 	ifq->altq_mapsubq = mapsubq;
3032 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
3033 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
3034 
3035 		ifsq->ifsq_enqueue = enqueue;
3036 		ifsq->ifsq_dequeue = dequeue;
3037 		ifsq->ifsq_request = request;
3038 	}
3039 }
3040 
3041 static void
3042 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
3043 {
3044 
3045 	classq_add(&ifsq->ifsq_norm, m);
3046 	ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
3047 }
3048 
3049 static void
3050 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
3051 {
3052 
3053 	classq_add(&ifsq->ifsq_prio, m);
3054 	ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
3055 	ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
3056 }
3057 
3058 static struct mbuf *
3059 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
3060 {
3061 	struct mbuf *m;
3062 
3063 	m = classq_get(&ifsq->ifsq_norm);
3064 	if (m != NULL)
3065 		ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
3066 	return (m);
3067 }
3068 
3069 static struct mbuf *
3070 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
3071 {
3072 	struct mbuf *m;
3073 
3074 	m = classq_get(&ifsq->ifsq_prio);
3075 	if (m != NULL) {
3076 		ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
3077 		ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
3078 	}
3079 	return (m);
3080 }
3081 
3082 int
3083 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
3084     struct altq_pktattr *pa __unused)
3085 {
3086 
3087 	M_ASSERTPKTHDR(m);
3088 again:
3089 	if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
3090 	    ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
3091 		struct mbuf *m_drop;
3092 
3093 		if (m->m_flags & M_PRIO) {
3094 			m_drop = NULL;
3095 			if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) &&
3096 			    ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) {
3097 				/* Try dropping some from normal queue. */
3098 				m_drop = ifsq_norm_dequeue(ifsq);
3099 			}
3100 			if (m_drop == NULL)
3101 				m_drop = ifsq_prio_dequeue(ifsq);
3102 		} else {
3103 			m_drop = ifsq_norm_dequeue(ifsq);
3104 		}
3105 		if (m_drop != NULL) {
3106 			IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1);
3107 			m_freem(m_drop);
3108 			goto again;
3109 		}
3110 		/*
3111 		 * No old packets could be dropped!
3112 		 * NOTE: Caller increases oqdrops.
3113 		 */
3114 		m_freem(m);
3115 		return (ENOBUFS);
3116 	} else {
3117 		if (m->m_flags & M_PRIO)
3118 			ifsq_prio_enqueue(ifsq, m);
3119 		else
3120 			ifsq_norm_enqueue(ifsq, m);
3121 		return (0);
3122 	}
3123 }
3124 
3125 struct mbuf *
3126 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
3127 {
3128 	struct mbuf *m;
3129 
3130 	switch (op) {
3131 	case ALTDQ_POLL:
3132 		m = classq_head(&ifsq->ifsq_prio);
3133 		if (m == NULL)
3134 			m = classq_head(&ifsq->ifsq_norm);
3135 		break;
3136 
3137 	case ALTDQ_REMOVE:
3138 		m = ifsq_prio_dequeue(ifsq);
3139 		if (m == NULL)
3140 			m = ifsq_norm_dequeue(ifsq);
3141 		break;
3142 
3143 	default:
3144 		panic("unsupported ALTQ dequeue op: %d", op);
3145 	}
3146 	return m;
3147 }
3148 
3149 int
3150 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
3151 {
3152 	switch (req) {
3153 	case ALTRQ_PURGE:
3154 		for (;;) {
3155 			struct mbuf *m;
3156 
3157 			m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
3158 			if (m == NULL)
3159 				break;
3160 			m_freem(m);
3161 		}
3162 		break;
3163 
3164 	default:
3165 		panic("unsupported ALTQ request: %d", req);
3166 	}
3167 	return 0;
3168 }
3169 
3170 static void
3171 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
3172 {
3173 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
3174 	int running = 0, need_sched;
3175 
3176 	/*
3177 	 * Try to do direct ifnet.if_start on the subqueue first, if there is
3178 	 * contention on the subqueue hardware serializer, ifnet.if_start on
3179 	 * the subqueue will be scheduled on the subqueue owner CPU.
3180 	 */
3181 	if (!ifsq_tryserialize_hw(ifsq)) {
3182 		/*
3183 		 * Subqueue hardware serializer contention happened,
3184 		 * ifnet.if_start on the subqueue is scheduled on
3185 		 * the subqueue owner CPU, and we keep going.
3186 		 */
3187 		ifsq_ifstart_schedule(ifsq, 1);
3188 		return;
3189 	}
3190 
3191 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
3192 		ifp->if_start(ifp, ifsq);
3193 		if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
3194 			running = 1;
3195 	}
3196 	need_sched = ifsq_ifstart_need_schedule(ifsq, running);
3197 
3198 	ifsq_deserialize_hw(ifsq);
3199 
3200 	if (need_sched) {
3201 		/*
3202 		 * More data need to be transmitted, ifnet.if_start on the
3203 		 * subqueue is scheduled on the subqueue owner CPU, and we
3204 		 * keep going.
3205 		 * NOTE: ifnet.if_start subqueue interlock is not released.
3206 		 */
3207 		ifsq_ifstart_schedule(ifsq, force_sched);
3208 	}
3209 }
3210 
3211 /*
3212  * Subqeue packets staging mechanism:
3213  *
3214  * The packets enqueued into the subqueue are staged to a certain amount
3215  * before the ifnet.if_start on the subqueue is called.  In this way, the
3216  * driver could avoid writing to hardware registers upon every packet,
3217  * instead, hardware registers could be written when certain amount of
3218  * packets are put onto hardware TX ring.  The measurement on several modern
3219  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
3220  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
3221  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
3222  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
3223  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
3224  *
3225  * Subqueue packets staging is performed for two entry points into drivers'
3226  * transmission function:
3227  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
3228  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
3229  *
3230  * Subqueue packets staging will be stopped upon any of the following
3231  * conditions:
3232  * - If the count of packets enqueued on the current CPU is great than or
3233  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
3234  * - If the total length of packets enqueued on the current CPU is great
3235  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
3236  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
3237  *   is usually less than hardware's MTU.
3238  * - ifsq_ifstart_schedule() is not pending on the current CPU and
3239  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
3240  *   released.
3241  * - The if_start_rollup(), which is registered as low priority netisr
3242  *   rollup function, is called; probably because no more work is pending
3243  *   for netisr.
3244  *
3245  * NOTE:
3246  * Currently subqueue packet staging is only performed in netisr threads.
3247  */
3248 int
3249 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
3250 {
3251 	struct ifaltq *ifq = &ifp->if_snd;
3252 	struct ifaltq_subque *ifsq;
3253 	int error, start = 0, len, mcast = 0, avoid_start = 0;
3254 	struct ifsubq_stage_head *head = NULL;
3255 	struct ifsubq_stage *stage = NULL;
3256 	struct globaldata *gd = mycpu;
3257 	struct thread *td = gd->gd_curthread;
3258 
3259 	crit_enter_quick(td);
3260 
3261 	ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
3262 	ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3263 
3264 	len = m->m_pkthdr.len;
3265 	if (m->m_flags & M_MCAST)
3266 		mcast = 1;
3267 
3268 	if (td->td_type == TD_TYPE_NETISR) {
3269 		head = &ifsubq_stage_heads[mycpuid];
3270 		stage = ifsq_get_stage(ifsq, mycpuid);
3271 
3272 		stage->stg_cnt++;
3273 		stage->stg_len += len;
3274 		if (stage->stg_cnt < ifsq_stage_cntmax &&
3275 		    stage->stg_len < (ifp->if_mtu - max_protohdr))
3276 			avoid_start = 1;
3277 	}
3278 
3279 	ALTQ_SQ_LOCK(ifsq);
3280 	error = ifsq_enqueue_locked(ifsq, m, pa);
3281 	if (error) {
3282 		IFNET_STAT_INC(ifp, oqdrops, 1);
3283 		if (!ifsq_data_ready(ifsq)) {
3284 			ALTQ_SQ_UNLOCK(ifsq);
3285 			goto done;
3286 		}
3287 		avoid_start = 0;
3288 	} else {
3289 		IFNET_STAT_INC(ifp, obytes, len);
3290 		if (mcast)
3291 			IFNET_STAT_INC(ifp, omcasts, 1);
3292 	}
3293 	if (!ifsq_is_started(ifsq)) {
3294 		if (avoid_start) {
3295 			ALTQ_SQ_UNLOCK(ifsq);
3296 
3297 			KKASSERT(!error);
3298 			if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3299 				ifsq_stage_insert(head, stage);
3300 
3301 			goto done;
3302 		}
3303 
3304 		/*
3305 		 * Hold the subqueue interlock of ifnet.if_start
3306 		 */
3307 		ifsq_set_started(ifsq);
3308 		start = 1;
3309 	}
3310 	ALTQ_SQ_UNLOCK(ifsq);
3311 
3312 	if (stage != NULL) {
3313 		if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3314 			KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3315 			if (!avoid_start) {
3316 				ifsq_stage_remove(head, stage);
3317 				ifsq_ifstart_schedule(ifsq, 1);
3318 			}
3319 			goto done;
3320 		}
3321 
3322 		if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3323 			ifsq_stage_remove(head, stage);
3324 		} else {
3325 			stage->stg_cnt = 0;
3326 			stage->stg_len = 0;
3327 		}
3328 	}
3329 
3330 	if (start)
3331 		ifsq_ifstart_try(ifsq, 0);
3332 
3333 done:
3334 	crit_exit_quick(td);
3335 	return error;
3336 }
3337 
3338 void *
3339 ifa_create(int size)
3340 {
3341 	struct ifaddr *ifa;
3342 	int i;
3343 
3344 	KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3345 
3346 	ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3347 
3348 	/*
3349 	 * Make ifa_container availabel on all CPUs, since they
3350 	 * could be accessed by any threads.
3351 	 */
3352 	ifa->ifa_containers =
3353 		kmalloc(ncpus * sizeof(struct ifaddr_container),
3354 			M_IFADDR,
3355 			M_INTWAIT | M_ZERO | M_CACHEALIGN);
3356 
3357 	ifa->ifa_ncnt = ncpus;
3358 	for (i = 0; i < ncpus; ++i) {
3359 		struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3360 
3361 		ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3362 		ifac->ifa = ifa;
3363 		ifac->ifa_refcnt = 1;
3364 	}
3365 #ifdef IFADDR_DEBUG
3366 	kprintf("alloc ifa %p %d\n", ifa, size);
3367 #endif
3368 	return ifa;
3369 }
3370 
3371 void
3372 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3373 {
3374 	struct ifaddr *ifa = ifac->ifa;
3375 
3376 	KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3377 	KKASSERT(ifac->ifa_refcnt == 0);
3378 	KASSERT(ifac->ifa_listmask == 0,
3379 		("ifa is still on %#x lists", ifac->ifa_listmask));
3380 
3381 	ifac->ifa_magic = IFA_CONTAINER_DEAD;
3382 
3383 #ifdef IFADDR_DEBUG_VERBOSE
3384 	kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3385 #endif
3386 
3387 	KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3388 		("invalid # of ifac, %d", ifa->ifa_ncnt));
3389 	if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3390 #ifdef IFADDR_DEBUG
3391 		kprintf("free ifa %p\n", ifa);
3392 #endif
3393 		kfree(ifa->ifa_containers, M_IFADDR);
3394 		kfree(ifa, M_IFADDR);
3395 	}
3396 }
3397 
3398 static void
3399 ifa_iflink_dispatch(netmsg_t nmsg)
3400 {
3401 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3402 	struct ifaddr *ifa = msg->ifa;
3403 	struct ifnet *ifp = msg->ifp;
3404 	int cpu = mycpuid;
3405 	struct ifaddr_container *ifac;
3406 
3407 	crit_enter();
3408 
3409 	ifac = &ifa->ifa_containers[cpu];
3410 	ASSERT_IFAC_VALID(ifac);
3411 	KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3412 		("ifaddr is on if_addrheads"));
3413 
3414 	ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3415 	if (msg->tail)
3416 		TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3417 	else
3418 		TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3419 
3420 	crit_exit();
3421 
3422 	netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3423 }
3424 
3425 void
3426 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3427 {
3428 	struct netmsg_ifaddr msg;
3429 
3430 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3431 		    0, ifa_iflink_dispatch);
3432 	msg.ifa = ifa;
3433 	msg.ifp = ifp;
3434 	msg.tail = tail;
3435 
3436 	netisr_domsg(&msg.base, 0);
3437 }
3438 
3439 static void
3440 ifa_ifunlink_dispatch(netmsg_t nmsg)
3441 {
3442 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3443 	struct ifaddr *ifa = msg->ifa;
3444 	struct ifnet *ifp = msg->ifp;
3445 	int cpu = mycpuid;
3446 	struct ifaddr_container *ifac;
3447 
3448 	crit_enter();
3449 
3450 	ifac = &ifa->ifa_containers[cpu];
3451 	ASSERT_IFAC_VALID(ifac);
3452 	KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3453 		("ifaddr is not on if_addrhead"));
3454 
3455 	TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3456 	ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3457 
3458 	crit_exit();
3459 
3460 	netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3461 }
3462 
3463 void
3464 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3465 {
3466 	struct netmsg_ifaddr msg;
3467 
3468 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3469 		    0, ifa_ifunlink_dispatch);
3470 	msg.ifa = ifa;
3471 	msg.ifp = ifp;
3472 
3473 	netisr_domsg(&msg.base, 0);
3474 }
3475 
3476 static void
3477 ifa_destroy_dispatch(netmsg_t nmsg)
3478 {
3479 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3480 
3481 	IFAFREE(msg->ifa);
3482 	netisr_forwardmsg_all(&nmsg->base, mycpuid + 1);
3483 }
3484 
3485 void
3486 ifa_destroy(struct ifaddr *ifa)
3487 {
3488 	struct netmsg_ifaddr msg;
3489 
3490 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3491 		    0, ifa_destroy_dispatch);
3492 	msg.ifa = ifa;
3493 
3494 	netisr_domsg(&msg.base, 0);
3495 }
3496 
3497 static void
3498 if_start_rollup(void)
3499 {
3500 	struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3501 	struct ifsubq_stage *stage;
3502 
3503 	crit_enter();
3504 
3505 	while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3506 		struct ifaltq_subque *ifsq = stage->stg_subq;
3507 		int is_sched = 0;
3508 
3509 		if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3510 			is_sched = 1;
3511 		ifsq_stage_remove(head, stage);
3512 
3513 		if (is_sched) {
3514 			ifsq_ifstart_schedule(ifsq, 1);
3515 		} else {
3516 			int start = 0;
3517 
3518 			ALTQ_SQ_LOCK(ifsq);
3519 			if (!ifsq_is_started(ifsq)) {
3520 				/*
3521 				 * Hold the subqueue interlock of
3522 				 * ifnet.if_start
3523 				 */
3524 				ifsq_set_started(ifsq);
3525 				start = 1;
3526 			}
3527 			ALTQ_SQ_UNLOCK(ifsq);
3528 
3529 			if (start)
3530 				ifsq_ifstart_try(ifsq, 1);
3531 		}
3532 		KKASSERT((stage->stg_flags &
3533 		    (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3534 	}
3535 
3536 	crit_exit();
3537 }
3538 
3539 static void
3540 ifnetinit(void *dummy __unused)
3541 {
3542 	int i;
3543 
3544 	/* XXX netisr_ncpus */
3545 	for (i = 0; i < ncpus; ++i)
3546 		TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3547 	netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3548 }
3549 
3550 void
3551 if_register_com_alloc(u_char type,
3552     if_com_alloc_t *a, if_com_free_t *f)
3553 {
3554 
3555         KASSERT(if_com_alloc[type] == NULL,
3556             ("if_register_com_alloc: %d already registered", type));
3557         KASSERT(if_com_free[type] == NULL,
3558             ("if_register_com_alloc: %d free already registered", type));
3559 
3560         if_com_alloc[type] = a;
3561         if_com_free[type] = f;
3562 }
3563 
3564 void
3565 if_deregister_com_alloc(u_char type)
3566 {
3567 
3568         KASSERT(if_com_alloc[type] != NULL,
3569             ("if_deregister_com_alloc: %d not registered", type));
3570         KASSERT(if_com_free[type] != NULL,
3571             ("if_deregister_com_alloc: %d free not registered", type));
3572         if_com_alloc[type] = NULL;
3573         if_com_free[type] = NULL;
3574 }
3575 
3576 void
3577 ifq_set_maxlen(struct ifaltq *ifq, int len)
3578 {
3579 	ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3580 }
3581 
3582 int
3583 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3584 {
3585 	return ALTQ_SUBQ_INDEX_DEFAULT;
3586 }
3587 
3588 int
3589 ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid)
3590 {
3591 
3592 	return (cpuid % ifq->altq_subq_mappriv);
3593 }
3594 
3595 /*
3596  * Watchdog timeout.  Process callback as appropriate.  If we cannot
3597  * serialize the ifnet just try again on the next timeout.
3598  *
3599  * NOTE: The ifnet can adjust wd_timer while holding the serializer.  We
3600  *	 can only safely adjust it under the same circumstances.
3601  */
3602 static void
3603 ifsq_watchdog(void *arg)
3604 {
3605 	struct ifsubq_watchdog *wd = arg;
3606 	struct ifnet *ifp;
3607 	int count;
3608 
3609 	/*
3610 	 * Fast track.  Try to avoid acquiring the serializer when not
3611 	 * near the terminal count, unless asked to.  If the atomic op
3612 	 * to decrement the count fails just retry on the next callout.
3613 	 */
3614 	count = wd->wd_timer;
3615 	cpu_ccfence();
3616 	if (count == 0)
3617 		goto done;
3618 	if (count > 2 && (wd->wd_flags & IF_WDOG_ALLTICKS) == 0) {
3619 		(void)atomic_cmpset_int(&wd->wd_timer, count, count - 1);
3620 		goto done;
3621 	}
3622 
3623 	/*
3624 	 * Obtain the serializer and then re-test all wd_timer conditions
3625 	 * as it may have changed.  NICs do not mess with wd_timer without
3626 	 * holding the serializer.
3627 	 *
3628 	 * If we are unable to obtain the serializer just retry the same
3629 	 * count on the next callout.
3630 	 *
3631 	 * - call watchdog in terminal count (0)
3632 	 * - call watchdog on last tick (1) if requested
3633 	 * - call watchdog on all ticks if requested
3634 	 */
3635 	ifp = ifsq_get_ifp(wd->wd_subq);
3636 	if (ifnet_tryserialize_all(ifp) == 0)
3637 		goto done;
3638 	if (atomic_cmpset_int(&wd->wd_timer, count, count - 1)) {
3639 		--count;
3640 		if (count == 0 ||
3641 		    (wd->wd_flags & IF_WDOG_ALLTICKS) ||
3642 		    ((wd->wd_flags & IF_WDOG_LASTTICK) && count == 1)) {
3643 			wd->wd_watchdog(wd->wd_subq);
3644 		}
3645 	}
3646 	ifnet_deserialize_all(ifp);
3647 done:
3648 	ifsq_watchdog_reset(wd);
3649 }
3650 
3651 static void
3652 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3653 {
3654 	callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3655 	    ifsq_get_cpuid(wd->wd_subq));
3656 }
3657 
3658 void
3659 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3660 		   ifsq_watchdog_t watchdog, int flags)
3661 {
3662 	callout_init_mp(&wd->wd_callout);
3663 	wd->wd_timer = 0;
3664 	wd->wd_flags = flags;
3665 	wd->wd_subq = ifsq;
3666 	wd->wd_watchdog = watchdog;
3667 }
3668 
3669 void
3670 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3671 {
3672 	atomic_swap_int(&wd->wd_timer, 0);
3673 	ifsq_watchdog_reset(wd);
3674 }
3675 
3676 void
3677 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3678 {
3679 	atomic_swap_int(&wd->wd_timer, 0);
3680 	callout_stop(&wd->wd_callout);
3681 }
3682 
3683 void
3684 ifsq_watchdog_set_count(struct ifsubq_watchdog *wd, int count)
3685 {
3686 	atomic_swap_int(&wd->wd_timer, count);
3687 }
3688 
3689 void
3690 ifnet_lock(void)
3691 {
3692 	KASSERT(curthread->td_type != TD_TYPE_NETISR,
3693 	    ("try holding ifnet lock in netisr"));
3694 	mtx_lock(&ifnet_mtx);
3695 }
3696 
3697 void
3698 ifnet_unlock(void)
3699 {
3700 	KASSERT(curthread->td_type != TD_TYPE_NETISR,
3701 	    ("try holding ifnet lock in netisr"));
3702 	mtx_unlock(&ifnet_mtx);
3703 }
3704 
3705 static struct ifnet_array *
3706 ifnet_array_alloc(int count)
3707 {
3708 	struct ifnet_array *arr;
3709 
3710 	arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3711 	    M_IFNET, M_WAITOK);
3712 	arr->ifnet_count = count;
3713 
3714 	return arr;
3715 }
3716 
3717 static void
3718 ifnet_array_free(struct ifnet_array *arr)
3719 {
3720 	if (arr == &ifnet_array0)
3721 		return;
3722 	kfree(arr, M_IFNET);
3723 }
3724 
3725 static struct ifnet_array *
3726 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3727 {
3728 	struct ifnet_array *arr;
3729 	int count, i;
3730 
3731 	KASSERT(old_arr->ifnet_count >= 0,
3732 	    ("invalid ifnet array count %d", old_arr->ifnet_count));
3733 	count = old_arr->ifnet_count + 1;
3734 	arr = ifnet_array_alloc(count);
3735 
3736 	/*
3737 	 * Save the old ifnet array and append this ifp to the end of
3738 	 * the new ifnet array.
3739 	 */
3740 	for (i = 0; i < old_arr->ifnet_count; ++i) {
3741 		KASSERT(old_arr->ifnet_arr[i] != ifp,
3742 		    ("%s is already in ifnet array", ifp->if_xname));
3743 		arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3744 	}
3745 	KASSERT(i == count - 1,
3746 	    ("add %s, ifnet array index mismatch, should be %d, but got %d",
3747 	     ifp->if_xname, count - 1, i));
3748 	arr->ifnet_arr[i] = ifp;
3749 
3750 	return arr;
3751 }
3752 
3753 static struct ifnet_array *
3754 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3755 {
3756 	struct ifnet_array *arr;
3757 	int count, i, idx, found = 0;
3758 
3759 	KASSERT(old_arr->ifnet_count > 0,
3760 	    ("invalid ifnet array count %d", old_arr->ifnet_count));
3761 	count = old_arr->ifnet_count - 1;
3762 	arr = ifnet_array_alloc(count);
3763 
3764 	/*
3765 	 * Save the old ifnet array, but skip this ifp.
3766 	 */
3767 	idx = 0;
3768 	for (i = 0; i < old_arr->ifnet_count; ++i) {
3769 		if (old_arr->ifnet_arr[i] == ifp) {
3770 			KASSERT(!found,
3771 			    ("dup %s is in ifnet array", ifp->if_xname));
3772 			found = 1;
3773 			continue;
3774 		}
3775 		KASSERT(idx < count,
3776 		    ("invalid ifnet array index %d, count %d", idx, count));
3777 		arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3778 		++idx;
3779 	}
3780 	KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3781 	KASSERT(idx == count,
3782 	    ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3783 	     ifp->if_xname, count, idx));
3784 
3785 	return arr;
3786 }
3787 
3788 const struct ifnet_array *
3789 ifnet_array_get(void)
3790 {
3791 	const struct ifnet_array *ret;
3792 
3793 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3794 	ret = ifnet_array;
3795 	/* Make sure 'ret' is really used. */
3796 	cpu_ccfence();
3797 	return (ret);
3798 }
3799 
3800 int
3801 ifnet_array_isempty(void)
3802 {
3803 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3804 	if (ifnet_array->ifnet_count == 0)
3805 		return 1;
3806 	else
3807 		return 0;
3808 }
3809 
3810 void
3811 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3812 {
3813 	struct ifaddr *ifa;
3814 
3815 	memset(mark, 0, sizeof(*mark));
3816 	ifa = &mark->ifa;
3817 
3818 	mark->ifac.ifa = ifa;
3819 
3820 	ifa->ifa_addr = &mark->addr;
3821 	ifa->ifa_dstaddr = &mark->dstaddr;
3822 	ifa->ifa_netmask = &mark->netmask;
3823 	ifa->ifa_ifp = ifp;
3824 }
3825 
3826 static int
3827 if_ringcnt_fixup(int ring_cnt, int ring_cntmax)
3828 {
3829 
3830 	KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax));
3831 
3832 	if (ring_cnt <= 0 || ring_cnt > ring_cntmax)
3833 		ring_cnt = ring_cntmax;
3834 	if (ring_cnt > netisr_ncpus)
3835 		ring_cnt = netisr_ncpus;
3836 	return (ring_cnt);
3837 }
3838 
3839 static void
3840 if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid)
3841 {
3842 	int i, offset;
3843 
3844 	KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid));
3845 	KASSERT(grid >= rm->rm_cnt, ("invalid if_ringmap grid %d, count %d",
3846 	    grid, rm->rm_cnt));
3847 	rm->rm_grid = grid;
3848 
3849 	offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus;
3850 	for (i = 0; i < rm->rm_cnt; ++i) {
3851 		rm->rm_cpumap[i] = offset + i;
3852 		KASSERT(rm->rm_cpumap[i] < netisr_ncpus,
3853 		    ("invalid cpumap[%d] = %d, offset %d", i,
3854 		     rm->rm_cpumap[i], offset));
3855 	}
3856 }
3857 
3858 static struct if_ringmap *
3859 if_ringmap_alloc_flags(device_t dev, int ring_cnt, int ring_cntmax,
3860     uint32_t flags)
3861 {
3862 	struct if_ringmap *rm;
3863 	int i, grid = 0, prev_grid;
3864 
3865 	ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3866 	rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]),
3867 	    M_DEVBUF, M_WAITOK | M_ZERO);
3868 
3869 	rm->rm_cnt = ring_cnt;
3870 	if (flags & RINGMAP_FLAG_POWEROF2)
3871 		rm->rm_cnt = 1 << (fls(rm->rm_cnt) - 1);
3872 
3873 	prev_grid = netisr_ncpus;
3874 	for (i = 0; i < netisr_ncpus; ++i) {
3875 		if (netisr_ncpus % (i + 1) != 0)
3876 			continue;
3877 
3878 		grid = netisr_ncpus / (i + 1);
3879 		if (rm->rm_cnt > grid) {
3880 			grid = prev_grid;
3881 			break;
3882 		}
3883 
3884 		if (rm->rm_cnt > netisr_ncpus / (i + 2))
3885 			break;
3886 		prev_grid = grid;
3887 	}
3888 	if_ringmap_set_grid(dev, rm, grid);
3889 
3890 	return (rm);
3891 }
3892 
3893 struct if_ringmap *
3894 if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax)
3895 {
3896 
3897 	return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3898 	    RINGMAP_FLAG_NONE));
3899 }
3900 
3901 struct if_ringmap *
3902 if_ringmap_alloc2(device_t dev, int ring_cnt, int ring_cntmax)
3903 {
3904 
3905 	return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3906 	    RINGMAP_FLAG_POWEROF2));
3907 }
3908 
3909 void
3910 if_ringmap_free(struct if_ringmap *rm)
3911 {
3912 
3913 	kfree(rm, M_DEVBUF);
3914 }
3915 
3916 /*
3917  * Align the two ringmaps.
3918  *
3919  * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3920  *
3921  * Before:
3922  *
3923  * CPU      0  1  2  3   4  5  6  7
3924  * NIC_RX               n0 n1 n2 n3
3925  * NIC_TX        N0 N1
3926  *
3927  * After:
3928  *
3929  * CPU      0  1  2  3   4  5  6  7
3930  * NIC_RX               n0 n1 n2 n3
3931  * NIC_TX               N0 N1
3932  */
3933 void
3934 if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3935 {
3936 
3937 	if (rm0->rm_grid > rm1->rm_grid)
3938 		if_ringmap_set_grid(dev, rm1, rm0->rm_grid);
3939 	else if (rm0->rm_grid < rm1->rm_grid)
3940 		if_ringmap_set_grid(dev, rm0, rm1->rm_grid);
3941 }
3942 
3943 void
3944 if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3945 {
3946 	int subset_grid, cnt, divisor, mod, offset, i;
3947 	struct if_ringmap *subset_rm, *rm;
3948 	int old_rm0_grid, old_rm1_grid;
3949 
3950 	if (rm0->rm_grid == rm1->rm_grid)
3951 		return;
3952 
3953 	/* Save grid for later use */
3954 	old_rm0_grid = rm0->rm_grid;
3955 	old_rm1_grid = rm1->rm_grid;
3956 
3957 	if_ringmap_align(dev, rm0, rm1);
3958 
3959 	/*
3960 	 * Re-shuffle rings to get more even distribution.
3961 	 *
3962 	 * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3963 	 *
3964 	 * CPU       0  1  2  3   4  5  6  7   8  9 10 11
3965 	 *
3966 	 * NIC_RX   a0 a1 a2 a3  b0 b1 b2 b3  c0 c1 c2 c3
3967 	 * NIC_TX   A0 A1        B0 B1        C0 C1
3968 	 *
3969 	 * NIC_RX   d0 d1 d2 d3  e0 e1 e2 e3  f0 f1 f2 f3
3970 	 * NIC_TX         D0 D1        E0 E1        F0 F1
3971 	 */
3972 
3973 	if (rm0->rm_cnt >= (2 * old_rm1_grid)) {
3974 		cnt = rm0->rm_cnt;
3975 		subset_grid = old_rm1_grid;
3976 		subset_rm = rm1;
3977 		rm = rm0;
3978 	} else if (rm1->rm_cnt > (2 * old_rm0_grid)) {
3979 		cnt = rm1->rm_cnt;
3980 		subset_grid = old_rm0_grid;
3981 		subset_rm = rm0;
3982 		rm = rm1;
3983 	} else {
3984 		/* No space to shuffle. */
3985 		return;
3986 	}
3987 
3988 	mod = cnt / subset_grid;
3989 	KKASSERT(mod >= 2);
3990 	divisor = netisr_ncpus / rm->rm_grid;
3991 	offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid;
3992 
3993 	for (i = 0; i < subset_rm->rm_cnt; ++i) {
3994 		subset_rm->rm_cpumap[i] += offset;
3995 		KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus,
3996 		    ("match: invalid cpumap[%d] = %d, offset %d",
3997 		     i, subset_rm->rm_cpumap[i], offset));
3998 	}
3999 #ifdef INVARIANTS
4000 	for (i = 0; i < subset_rm->rm_cnt; ++i) {
4001 		int j;
4002 
4003 		for (j = 0; j < rm->rm_cnt; ++j) {
4004 			if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i])
4005 				break;
4006 		}
4007 		KASSERT(j < rm->rm_cnt,
4008 		    ("subset cpumap[%d] = %d not found in superset",
4009 		     i, subset_rm->rm_cpumap[i]));
4010 	}
4011 #endif
4012 }
4013 
4014 int
4015 if_ringmap_count(const struct if_ringmap *rm)
4016 {
4017 
4018 	return (rm->rm_cnt);
4019 }
4020 
4021 int
4022 if_ringmap_cpumap(const struct if_ringmap *rm, int ring)
4023 {
4024 
4025 	KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring));
4026 	return (rm->rm_cpumap[ring]);
4027 }
4028 
4029 void
4030 if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent)
4031 {
4032 	int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy;
4033 
4034 	KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0,
4035 	    ("invalid redirect table entries %d", table_nent));
4036 
4037 	grid_idx = 0;
4038 	for (i = 0; i < NETISR_CPUMAX; ++i) {
4039 		table[i] = grid_idx++ % rm->rm_cnt;
4040 
4041 		if (grid_idx == rm->rm_grid)
4042 			grid_idx = 0;
4043 	}
4044 
4045 	/*
4046 	 * Make the ring distributed more evenly for the remainder
4047 	 * of each grid.
4048 	 *
4049 	 * e.g. 12 netisrs, rm contains 8 rings.
4050 	 *
4051 	 * Redirect table before:
4052 	 *
4053 	 *  0  1  2  3  4  5  6  7  0  1  2  3  0  1  2  3
4054 	 *  4  5  6  7  0  1  2  3  0  1  2  3  4  5  6  7
4055 	 *  0  1  2  3  0  1  2  3  4  5  6  7  0  1  2  3
4056 	 *  ....
4057 	 *
4058 	 * Redirect table after being patched (pX, patched entries):
4059 	 *
4060 	 *  0  1  2  3  4  5  6  7 p0 p1 p2 p3  0  1  2  3
4061 	 *  4  5  6  7 p4 p5 p6 p7  0  1  2  3  4  5  6  7
4062 	 * p0 p1 p2 p3  0  1  2  3  4  5  6  7 p4 p5 p6 p7
4063 	 *  ....
4064 	 */
4065 	patch_cnt = rm->rm_grid % rm->rm_cnt;
4066 	if (patch_cnt == 0)
4067 		goto done;
4068 	patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt);
4069 
4070 	grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid;
4071 	grid_idx = 0;
4072 	for (i = 0; i < grid_cnt; ++i) {
4073 		int j;
4074 
4075 		for (j = 0; j < patch_cnt; ++j) {
4076 			int fix_idx;
4077 
4078 			fix_idx = (i * rm->rm_grid) + patch_off + j;
4079 			if (fix_idx >= NETISR_CPUMAX)
4080 				goto done;
4081 			table[fix_idx] = grid_idx++ % rm->rm_cnt;
4082 		}
4083 	}
4084 done:
4085 	/*
4086 	 * If the device supports larger redirect table, duplicate
4087 	 * the first NETISR_CPUMAX entries to the rest of the table,
4088 	 * so that it matches upper layer's expectation:
4089 	 * (hash & NETISR_CPUMASK) % netisr_ncpus
4090 	 */
4091 	ncopy = table_nent / NETISR_CPUMAX;
4092 	for (i = 1; i < ncopy; ++i) {
4093 		memcpy(&table[i * NETISR_CPUMAX], table,
4094 		    NETISR_CPUMAX * sizeof(table[0]));
4095 	}
4096 	if (if_ringmap_dumprdr) {
4097 		for (i = 0; i < table_nent; ++i) {
4098 			if (i != 0 && i % 16 == 0)
4099 				kprintf("\n");
4100 			kprintf("%03d ", table[i]);
4101 		}
4102 		kprintf("\n");
4103 	}
4104 }
4105 
4106 int
4107 if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS)
4108 {
4109 	struct if_ringmap *rm = arg1;
4110 	int i, error = 0;
4111 
4112 	for (i = 0; i < rm->rm_cnt; ++i) {
4113 		int cpu = rm->rm_cpumap[i];
4114 
4115 		error = SYSCTL_OUT(req, &cpu, sizeof(cpu));
4116 		if (error)
4117 			break;
4118 	}
4119 	return (error);
4120 }
4121