xref: /dflybsd-src/sys/net/if.c (revision edf2e6576f0ebea982f042a9f12c4a2ee1c89184)
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/priv.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/sockio.h>
51 #include <sys/syslog.h>
52 #include <sys/sysctl.h>
53 #include <sys/domain.h>
54 #include <sys/thread.h>
55 #include <sys/serialize.h>
56 #include <sys/bus.h>
57 
58 #include <sys/thread2.h>
59 #include <sys/msgport2.h>
60 #include <sys/mutex2.h>
61 
62 #include <net/if.h>
63 #include <net/if_arp.h>
64 #include <net/if_dl.h>
65 #include <net/if_types.h>
66 #include <net/if_var.h>
67 #include <net/if_ringmap.h>
68 #include <net/ifq_var.h>
69 #include <net/radix.h>
70 #include <net/route.h>
71 #include <net/if_clone.h>
72 #include <net/netisr2.h>
73 #include <net/netmsg2.h>
74 
75 #include <machine/atomic.h>
76 #include <machine/stdarg.h>
77 #include <machine/smp.h>
78 
79 #if defined(INET) || defined(INET6)
80 /*XXX*/
81 #include <netinet/in.h>
82 #include <netinet/in_var.h>
83 #include <netinet/if_ether.h>
84 #ifdef INET6
85 #include <netinet6/in6_var.h>
86 #include <netinet6/in6_ifattach.h>
87 #endif
88 #endif
89 
90 struct netmsg_ifaddr {
91 	struct netmsg_base base;
92 	struct ifaddr	*ifa;
93 	struct ifnet	*ifp;
94 	int		tail;
95 };
96 
97 struct ifsubq_stage_head {
98 	TAILQ_HEAD(, ifsubq_stage)	stg_head;
99 } __cachealign;
100 
101 struct if_ringmap {
102 	int		rm_cnt;
103 	int		rm_grid;
104 	int		rm_cpumap[];
105 };
106 
107 /*
108  * System initialization
109  */
110 static void	if_attachdomain(void *);
111 static void	if_attachdomain1(struct ifnet *);
112 static int	ifconf(u_long, caddr_t, struct ucred *);
113 static void	ifinit(void *);
114 static void	ifnetinit(void *);
115 static void	if_slowtimo(void *);
116 static void	link_rtrequest(int, struct rtentry *);
117 static int	if_rtdel(struct radix_node *, void *);
118 static void	if_slowtimo_dispatch(netmsg_t);
119 
120 /* Helper functions */
121 static void	ifsq_watchdog_reset(struct ifsubq_watchdog *);
122 static int	if_delmulti_serialized(struct ifnet *, struct sockaddr *);
123 static struct ifnet_array *ifnet_array_alloc(int);
124 static void	ifnet_array_free(struct ifnet_array *);
125 static struct ifnet_array *ifnet_array_add(struct ifnet *,
126 		    const struct ifnet_array *);
127 static struct ifnet_array *ifnet_array_del(struct ifnet *,
128 		    const struct ifnet_array *);
129 
130 #ifdef INET6
131 /*
132  * XXX: declare here to avoid to include many inet6 related files..
133  * should be more generalized?
134  */
135 extern void	nd6_setmtu(struct ifnet *);
136 #endif
137 
138 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
139 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
140 SYSCTL_NODE(_net_link, OID_AUTO, ringmap, CTLFLAG_RW, 0, "link ringmap");
141 
142 static int ifsq_stage_cntmax = 4;
143 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
144 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
145     &ifsq_stage_cntmax, 0, "ifq staging packet count max");
146 
147 static int if_stats_compat = 0;
148 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
149     &if_stats_compat, 0, "Compat the old ifnet stats");
150 
151 static int if_ringmap_dumprdr = 0;
152 SYSCTL_INT(_net_link_ringmap, OID_AUTO, dump_rdr, CTLFLAG_RW,
153     &if_ringmap_dumprdr, 0, "dump redirect table");
154 
155 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL);
156 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, ifnetinit, NULL);
157 
158 static  if_com_alloc_t *if_com_alloc[256];
159 static  if_com_free_t *if_com_free[256];
160 
161 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
162 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
163 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
164 
165 int			ifqmaxlen = IFQ_MAXLEN;
166 struct ifnethead	ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
167 
168 static struct ifnet_array	ifnet_array0;
169 static struct ifnet_array	*ifnet_array = &ifnet_array0;
170 
171 static struct callout		if_slowtimo_timer;
172 static struct netmsg_base	if_slowtimo_netmsg;
173 
174 int			if_index = 0;
175 struct ifnet		**ifindex2ifnet = NULL;
176 static struct mtx	ifnet_mtx = MTX_INITIALIZER("ifnet");
177 
178 static struct ifsubq_stage_head	ifsubq_stage_heads[MAXCPU];
179 
180 #ifdef notyet
181 #define IFQ_KTR_STRING		"ifq=%p"
182 #define IFQ_KTR_ARGS	struct ifaltq *ifq
183 #ifndef KTR_IFQ
184 #define KTR_IFQ			KTR_ALL
185 #endif
186 KTR_INFO_MASTER(ifq);
187 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
188 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
189 #define logifq(name, arg)	KTR_LOG(ifq_ ## name, arg)
190 
191 #define IF_START_KTR_STRING	"ifp=%p"
192 #define IF_START_KTR_ARGS	struct ifnet *ifp
193 #ifndef KTR_IF_START
194 #define KTR_IF_START		KTR_ALL
195 #endif
196 KTR_INFO_MASTER(if_start);
197 KTR_INFO(KTR_IF_START, if_start, run, 0,
198 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
199 KTR_INFO(KTR_IF_START, if_start, sched, 1,
200 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
201 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
202 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
203 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
204 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
205 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
206 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
207 #define logifstart(name, arg)	KTR_LOG(if_start_ ## name, arg)
208 #endif
209 
210 TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
211 
212 /*
213  * Network interface utility routines.
214  *
215  * Routines with ifa_ifwith* names take sockaddr *'s as
216  * parameters.
217  */
218 /* ARGSUSED*/
219 static void
220 ifinit(void *dummy)
221 {
222 	struct ifnet *ifp;
223 
224 	callout_init_mp(&if_slowtimo_timer);
225 	netmsg_init(&if_slowtimo_netmsg, NULL, &netisr_adone_rport,
226 	    MSGF_PRIORITY, if_slowtimo_dispatch);
227 
228 	/* XXX is this necessary? */
229 	ifnet_lock();
230 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
231 		if (ifp->if_snd.altq_maxlen == 0) {
232 			if_printf(ifp, "XXX: driver didn't set altq_maxlen\n");
233 			ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
234 		}
235 	}
236 	ifnet_unlock();
237 
238 	/* Start if_slowtimo */
239 	lwkt_sendmsg(netisr_cpuport(0), &if_slowtimo_netmsg.lmsg);
240 }
241 
242 static void
243 ifsq_ifstart_ipifunc(void *arg)
244 {
245 	struct ifaltq_subque *ifsq = arg;
246 	struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
247 
248 	crit_enter();
249 	if (lmsg->ms_flags & MSGF_DONE)
250 		lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg);
251 	crit_exit();
252 }
253 
254 static __inline void
255 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
256 {
257 	KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
258 	TAILQ_REMOVE(&head->stg_head, stage, stg_link);
259 	stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
260 	stage->stg_cnt = 0;
261 	stage->stg_len = 0;
262 }
263 
264 static __inline void
265 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
266 {
267 	KKASSERT((stage->stg_flags &
268 	    (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
269 	stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
270 	TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
271 }
272 
273 /*
274  * Schedule ifnet.if_start on the subqueue owner CPU
275  */
276 static void
277 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
278 {
279 	int cpu;
280 
281 	if (!force && curthread->td_type == TD_TYPE_NETISR &&
282 	    ifsq_stage_cntmax > 0) {
283 		struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
284 
285 		stage->stg_cnt = 0;
286 		stage->stg_len = 0;
287 		if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
288 			ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
289 		stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
290 		return;
291 	}
292 
293 	cpu = ifsq_get_cpuid(ifsq);
294 	if (cpu != mycpuid)
295 		lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
296 	else
297 		ifsq_ifstart_ipifunc(ifsq);
298 }
299 
300 /*
301  * NOTE:
302  * This function will release ifnet.if_start subqueue interlock,
303  * if ifnet.if_start for the subqueue does not need to be scheduled
304  */
305 static __inline int
306 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
307 {
308 	if (!running || ifsq_is_empty(ifsq)
309 #ifdef ALTQ
310 	    || ifsq->ifsq_altq->altq_tbr != NULL
311 #endif
312 	) {
313 		ALTQ_SQ_LOCK(ifsq);
314 		/*
315 		 * ifnet.if_start subqueue interlock is released, if:
316 		 * 1) Hardware can not take any packets, due to
317 		 *    o  interface is marked down
318 		 *    o  hardware queue is full (ifsq_is_oactive)
319 		 *    Under the second situation, hardware interrupt
320 		 *    or polling(4) will call/schedule ifnet.if_start
321 		 *    on the subqueue when hardware queue is ready
322 		 * 2) There is no packet in the subqueue.
323 		 *    Further ifq_dispatch or ifq_handoff will call/
324 		 *    schedule ifnet.if_start on the subqueue.
325 		 * 3) TBR is used and it does not allow further
326 		 *    dequeueing.
327 		 *    TBR callout will call ifnet.if_start on the
328 		 *    subqueue.
329 		 */
330 		if (!running || !ifsq_data_ready(ifsq)) {
331 			ifsq_clr_started(ifsq);
332 			ALTQ_SQ_UNLOCK(ifsq);
333 			return 0;
334 		}
335 		ALTQ_SQ_UNLOCK(ifsq);
336 	}
337 	return 1;
338 }
339 
340 static void
341 ifsq_ifstart_dispatch(netmsg_t msg)
342 {
343 	struct lwkt_msg *lmsg = &msg->base.lmsg;
344 	struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
345 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
346 	struct globaldata *gd = mycpu;
347 	int running = 0, need_sched;
348 
349 	crit_enter_gd(gd);
350 
351 	lwkt_replymsg(lmsg, 0);	/* reply ASAP */
352 
353 	if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) {
354 		/*
355 		 * We need to chase the subqueue owner CPU change.
356 		 */
357 		ifsq_ifstart_schedule(ifsq, 1);
358 		crit_exit_gd(gd);
359 		return;
360 	}
361 
362 	ifsq_serialize_hw(ifsq);
363 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
364 		ifp->if_start(ifp, ifsq);
365 		if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
366 			running = 1;
367 	}
368 	need_sched = ifsq_ifstart_need_schedule(ifsq, running);
369 	ifsq_deserialize_hw(ifsq);
370 
371 	if (need_sched) {
372 		/*
373 		 * More data need to be transmitted, ifnet.if_start is
374 		 * scheduled on the subqueue owner CPU, and we keep going.
375 		 * NOTE: ifnet.if_start subqueue interlock is not released.
376 		 */
377 		ifsq_ifstart_schedule(ifsq, 0);
378 	}
379 
380 	crit_exit_gd(gd);
381 }
382 
383 /* Device driver ifnet.if_start helper function */
384 void
385 ifsq_devstart(struct ifaltq_subque *ifsq)
386 {
387 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
388 	int running = 0;
389 
390 	ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
391 
392 	ALTQ_SQ_LOCK(ifsq);
393 	if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
394 		ALTQ_SQ_UNLOCK(ifsq);
395 		return;
396 	}
397 	ifsq_set_started(ifsq);
398 	ALTQ_SQ_UNLOCK(ifsq);
399 
400 	ifp->if_start(ifp, ifsq);
401 
402 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
403 		running = 1;
404 
405 	if (ifsq_ifstart_need_schedule(ifsq, running)) {
406 		/*
407 		 * More data need to be transmitted, ifnet.if_start is
408 		 * scheduled on ifnet's CPU, and we keep going.
409 		 * NOTE: ifnet.if_start interlock is not released.
410 		 */
411 		ifsq_ifstart_schedule(ifsq, 0);
412 	}
413 }
414 
415 void
416 if_devstart(struct ifnet *ifp)
417 {
418 	ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
419 }
420 
421 /* Device driver ifnet.if_start schedule helper function */
422 void
423 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
424 {
425 	ifsq_ifstart_schedule(ifsq, 1);
426 }
427 
428 void
429 if_devstart_sched(struct ifnet *ifp)
430 {
431 	ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
432 }
433 
434 static void
435 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
436 {
437 	lwkt_serialize_enter(ifp->if_serializer);
438 }
439 
440 static void
441 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
442 {
443 	lwkt_serialize_exit(ifp->if_serializer);
444 }
445 
446 static int
447 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
448 {
449 	return lwkt_serialize_try(ifp->if_serializer);
450 }
451 
452 #ifdef INVARIANTS
453 static void
454 if_default_serialize_assert(struct ifnet *ifp,
455 			    enum ifnet_serialize slz __unused,
456 			    boolean_t serialized)
457 {
458 	if (serialized)
459 		ASSERT_SERIALIZED(ifp->if_serializer);
460 	else
461 		ASSERT_NOT_SERIALIZED(ifp->if_serializer);
462 }
463 #endif
464 
465 /*
466  * Attach an interface to the list of "active" interfaces.
467  *
468  * The serializer is optional.
469  */
470 void
471 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
472 {
473 	unsigned socksize;
474 	int namelen, masklen;
475 	struct sockaddr_dl *sdl, *sdl_addr;
476 	struct ifaddr *ifa;
477 	struct ifaltq *ifq;
478 	struct ifnet **old_ifindex2ifnet = NULL;
479 	struct ifnet_array *old_ifnet_array;
480 	int i, q;
481 
482 	static int if_indexlim = 8;
483 
484 	if (ifp->if_serialize != NULL) {
485 		KASSERT(ifp->if_deserialize != NULL &&
486 			ifp->if_tryserialize != NULL &&
487 			ifp->if_serialize_assert != NULL,
488 			("serialize functions are partially setup"));
489 
490 		/*
491 		 * If the device supplies serialize functions,
492 		 * then clear if_serializer to catch any invalid
493 		 * usage of this field.
494 		 */
495 		KASSERT(serializer == NULL,
496 			("both serialize functions and default serializer "
497 			 "are supplied"));
498 		ifp->if_serializer = NULL;
499 	} else {
500 		KASSERT(ifp->if_deserialize == NULL &&
501 			ifp->if_tryserialize == NULL &&
502 			ifp->if_serialize_assert == NULL,
503 			("serialize functions are partially setup"));
504 		ifp->if_serialize = if_default_serialize;
505 		ifp->if_deserialize = if_default_deserialize;
506 		ifp->if_tryserialize = if_default_tryserialize;
507 #ifdef INVARIANTS
508 		ifp->if_serialize_assert = if_default_serialize_assert;
509 #endif
510 
511 		/*
512 		 * The serializer can be passed in from the device,
513 		 * allowing the same serializer to be used for both
514 		 * the interrupt interlock and the device queue.
515 		 * If not specified, the netif structure will use an
516 		 * embedded serializer.
517 		 */
518 		if (serializer == NULL) {
519 			serializer = &ifp->if_default_serializer;
520 			lwkt_serialize_init(serializer);
521 		}
522 		ifp->if_serializer = serializer;
523 	}
524 
525 	/*
526 	 * XXX -
527 	 * The old code would work if the interface passed a pre-existing
528 	 * chain of ifaddrs to this code.  We don't trust our callers to
529 	 * properly initialize the tailq, however, so we no longer allow
530 	 * this unlikely case.
531 	 */
532 	ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
533 				    M_IFADDR, M_WAITOK | M_ZERO);
534 	for (i = 0; i < ncpus; ++i)
535 		TAILQ_INIT(&ifp->if_addrheads[i]);
536 
537 	TAILQ_INIT(&ifp->if_multiaddrs);
538 	TAILQ_INIT(&ifp->if_groups);
539 	getmicrotime(&ifp->if_lastchange);
540 
541 	/*
542 	 * create a Link Level name for this device
543 	 */
544 	namelen = strlen(ifp->if_xname);
545 	masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
546 	socksize = masklen + ifp->if_addrlen;
547 	if (socksize < sizeof(*sdl))
548 		socksize = sizeof(*sdl);
549 	socksize = RT_ROUNDUP(socksize);
550 	ifa = ifa_create(sizeof(struct ifaddr) + 2 * socksize);
551 	sdl = sdl_addr = (struct sockaddr_dl *)(ifa + 1);
552 	sdl->sdl_len = socksize;
553 	sdl->sdl_family = AF_LINK;
554 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
555 	sdl->sdl_nlen = namelen;
556 	sdl->sdl_type = ifp->if_type;
557 	ifp->if_lladdr = ifa;
558 	ifa->ifa_ifp = ifp;
559 	ifa->ifa_rtrequest = link_rtrequest;
560 	ifa->ifa_addr = (struct sockaddr *)sdl;
561 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
562 	ifa->ifa_netmask = (struct sockaddr *)sdl;
563 	sdl->sdl_len = masklen;
564 	while (namelen != 0)
565 		sdl->sdl_data[--namelen] = 0xff;
566 	ifa_iflink(ifa, ifp, 0 /* Insert head */);
567 
568 	ifp->if_data_pcpu = kmalloc_cachealign(
569 	    ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
570 
571 	if (ifp->if_mapsubq == NULL)
572 		ifp->if_mapsubq = ifq_mapsubq_default;
573 
574 	ifq = &ifp->if_snd;
575 	ifq->altq_type = 0;
576 	ifq->altq_disc = NULL;
577 	ifq->altq_flags &= ALTQF_CANTCHANGE;
578 	ifq->altq_tbr = NULL;
579 	ifq->altq_ifp = ifp;
580 
581 	if (ifq->altq_subq_cnt <= 0)
582 		ifq->altq_subq_cnt = 1;
583 	ifq->altq_subq = kmalloc_cachealign(
584 	    ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
585 	    M_DEVBUF, M_WAITOK | M_ZERO);
586 
587 	if (ifq->altq_maxlen == 0) {
588 		if_printf(ifp, "driver didn't set altq_maxlen\n");
589 		ifq_set_maxlen(ifq, ifqmaxlen);
590 	}
591 
592 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
593 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
594 
595 		ALTQ_SQ_LOCK_INIT(ifsq);
596 		ifsq->ifsq_index = q;
597 
598 		ifsq->ifsq_altq = ifq;
599 		ifsq->ifsq_ifp = ifp;
600 
601 		ifsq->ifsq_maxlen = ifq->altq_maxlen;
602 		ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
603 		ifsq->ifsq_prepended = NULL;
604 		ifsq->ifsq_started = 0;
605 		ifsq->ifsq_hw_oactive = 0;
606 		ifsq_set_cpuid(ifsq, 0);
607 		if (ifp->if_serializer != NULL)
608 			ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
609 
610 		ifsq->ifsq_stage =
611 		    kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
612 		    M_DEVBUF, M_WAITOK | M_ZERO);
613 		for (i = 0; i < ncpus; ++i)
614 			ifsq->ifsq_stage[i].stg_subq = ifsq;
615 
616 		ifsq->ifsq_ifstart_nmsg =
617 		    kmalloc(ncpus * sizeof(struct netmsg_base),
618 		    M_LWKTMSG, M_WAITOK);
619 		for (i = 0; i < ncpus; ++i) {
620 			netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
621 			    &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
622 			ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
623 		}
624 	}
625 	ifq_set_classic(ifq);
626 
627 	/*
628 	 * Increase mbuf cluster/jcluster limits for the mbufs that
629 	 * could sit on the device queues for quite some time.
630 	 */
631 	if (ifp->if_nmbclusters > 0)
632 		mcl_inclimit(ifp->if_nmbclusters);
633 	if (ifp->if_nmbjclusters > 0)
634 		mjcl_inclimit(ifp->if_nmbjclusters);
635 
636 	/*
637 	 * Install this ifp into ifindex2inet, ifnet queue and ifnet
638 	 * array after it is setup.
639 	 *
640 	 * Protect ifindex2ifnet, ifnet queue and ifnet array changes
641 	 * by ifnet lock, so that non-netisr threads could get a
642 	 * consistent view.
643 	 */
644 	ifnet_lock();
645 
646 	/* Don't update if_index until ifindex2ifnet is setup */
647 	ifp->if_index = if_index + 1;
648 	sdl_addr->sdl_index = ifp->if_index;
649 
650 	/*
651 	 * Install this ifp into ifindex2ifnet
652 	 */
653 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
654 		unsigned int n;
655 		struct ifnet **q;
656 
657 		/*
658 		 * Grow ifindex2ifnet
659 		 */
660 		if_indexlim <<= 1;
661 		n = if_indexlim * sizeof(*q);
662 		q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
663 		if (ifindex2ifnet != NULL) {
664 			bcopy(ifindex2ifnet, q, n/2);
665 			/* Free old ifindex2ifnet after sync all netisrs */
666 			old_ifindex2ifnet = ifindex2ifnet;
667 		}
668 		ifindex2ifnet = q;
669 	}
670 	ifindex2ifnet[ifp->if_index] = ifp;
671 	/*
672 	 * Update if_index after this ifp is installed into ifindex2ifnet,
673 	 * so that netisrs could get a consistent view of ifindex2ifnet.
674 	 */
675 	cpu_sfence();
676 	if_index = ifp->if_index;
677 
678 	/*
679 	 * Install this ifp into ifnet array.
680 	 */
681 	/* Free old ifnet array after sync all netisrs */
682 	old_ifnet_array = ifnet_array;
683 	ifnet_array = ifnet_array_add(ifp, old_ifnet_array);
684 
685 	/*
686 	 * Install this ifp into ifnet queue.
687 	 */
688 	TAILQ_INSERT_TAIL(&ifnetlist, ifp, if_link);
689 
690 	ifnet_unlock();
691 
692 	/*
693 	 * Sync all netisrs so that the old ifindex2ifnet and ifnet array
694 	 * are no longer accessed and we can free them safely later on.
695 	 */
696 	netmsg_service_sync();
697 	if (old_ifindex2ifnet != NULL)
698 		kfree(old_ifindex2ifnet, M_IFADDR);
699 	ifnet_array_free(old_ifnet_array);
700 
701 	if (!SLIST_EMPTY(&domains))
702 		if_attachdomain1(ifp);
703 
704 	/* Announce the interface. */
705 	EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
706 	devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
707 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
708 }
709 
710 static void
711 if_attachdomain(void *dummy)
712 {
713 	struct ifnet *ifp;
714 
715 	ifnet_lock();
716 	TAILQ_FOREACH(ifp, &ifnetlist, if_list)
717 		if_attachdomain1(ifp);
718 	ifnet_unlock();
719 }
720 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
721 	if_attachdomain, NULL);
722 
723 static void
724 if_attachdomain1(struct ifnet *ifp)
725 {
726 	struct domain *dp;
727 
728 	crit_enter();
729 
730 	/* address family dependent data region */
731 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
732 	SLIST_FOREACH(dp, &domains, dom_next)
733 		if (dp->dom_ifattach)
734 			ifp->if_afdata[dp->dom_family] =
735 				(*dp->dom_ifattach)(ifp);
736 	crit_exit();
737 }
738 
739 /*
740  * Purge all addresses whose type is _not_ AF_LINK
741  */
742 static void
743 if_purgeaddrs_nolink_dispatch(netmsg_t nmsg)
744 {
745 	struct lwkt_msg *lmsg = &nmsg->lmsg;
746 	struct ifnet *ifp = lmsg->u.ms_resultp;
747 	struct ifaddr_container *ifac, *next;
748 
749 	ASSERT_IN_NETISR(0);
750 
751 	/*
752 	 * The ifaddr processing in the following loop will block,
753 	 * however, this function is called in netisr0, in which
754 	 * ifaddr list changes happen, so we don't care about the
755 	 * blockness of the ifaddr processing here.
756 	 */
757 	TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
758 			      ifa_link, next) {
759 		struct ifaddr *ifa = ifac->ifa;
760 
761 		/* Ignore marker */
762 		if (ifa->ifa_addr->sa_family == AF_UNSPEC)
763 			continue;
764 
765 		/* Leave link ifaddr as it is */
766 		if (ifa->ifa_addr->sa_family == AF_LINK)
767 			continue;
768 #ifdef INET
769 		/* XXX: Ugly!! ad hoc just for INET */
770 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
771 			struct ifaliasreq ifr;
772 #ifdef IFADDR_DEBUG_VERBOSE
773 			int i;
774 
775 			kprintf("purge in4 addr %p: ", ifa);
776 			for (i = 0; i < ncpus; ++i)
777 				kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
778 			kprintf("\n");
779 #endif
780 
781 			bzero(&ifr, sizeof ifr);
782 			ifr.ifra_addr = *ifa->ifa_addr;
783 			if (ifa->ifa_dstaddr)
784 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
785 			if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp,
786 				       NULL) == 0)
787 				continue;
788 		}
789 #endif /* INET */
790 #ifdef INET6
791 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
792 #ifdef IFADDR_DEBUG_VERBOSE
793 			int i;
794 
795 			kprintf("purge in6 addr %p: ", ifa);
796 			for (i = 0; i < ncpus; ++i)
797 				kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
798 			kprintf("\n");
799 #endif
800 
801 			in6_purgeaddr(ifa);
802 			/* ifp_addrhead is already updated */
803 			continue;
804 		}
805 #endif /* INET6 */
806 		ifa_ifunlink(ifa, ifp);
807 		ifa_destroy(ifa);
808 	}
809 
810 	lwkt_replymsg(lmsg, 0);
811 }
812 
813 void
814 if_purgeaddrs_nolink(struct ifnet *ifp)
815 {
816 	struct netmsg_base nmsg;
817 	struct lwkt_msg *lmsg = &nmsg.lmsg;
818 
819 	ASSERT_CANDOMSG_NETISR0(curthread);
820 
821 	netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
822 	    if_purgeaddrs_nolink_dispatch);
823 	lmsg->u.ms_resultp = ifp;
824 	lwkt_domsg(netisr_cpuport(0), lmsg, 0);
825 }
826 
827 static void
828 ifq_stage_detach_handler(netmsg_t nmsg)
829 {
830 	struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
831 	int q;
832 
833 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
834 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
835 		struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
836 
837 		if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
838 			ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
839 	}
840 	lwkt_replymsg(&nmsg->lmsg, 0);
841 }
842 
843 static void
844 ifq_stage_detach(struct ifaltq *ifq)
845 {
846 	struct netmsg_base base;
847 	int cpu;
848 
849 	netmsg_init(&base, NULL, &curthread->td_msgport, 0,
850 	    ifq_stage_detach_handler);
851 	base.lmsg.u.ms_resultp = ifq;
852 
853 	for (cpu = 0; cpu < ncpus; ++cpu)
854 		lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
855 }
856 
857 struct netmsg_if_rtdel {
858 	struct netmsg_base	base;
859 	struct ifnet		*ifp;
860 };
861 
862 static void
863 if_rtdel_dispatch(netmsg_t msg)
864 {
865 	struct netmsg_if_rtdel *rmsg = (void *)msg;
866 	int i, nextcpu, cpu;
867 
868 	cpu = mycpuid;
869 	for (i = 1; i <= AF_MAX; i++) {
870 		struct radix_node_head	*rnh;
871 
872 		if ((rnh = rt_tables[cpu][i]) == NULL)
873 			continue;
874 		rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
875 	}
876 
877 	nextcpu = cpu + 1;
878 	if (nextcpu < ncpus)
879 		lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg);
880 	else
881 		lwkt_replymsg(&rmsg->base.lmsg, 0);
882 }
883 
884 /*
885  * Detach an interface, removing it from the
886  * list of "active" interfaces.
887  */
888 void
889 if_detach(struct ifnet *ifp)
890 {
891 	struct ifnet_array *old_ifnet_array;
892 	struct netmsg_if_rtdel msg;
893 	struct domain *dp;
894 	int q;
895 
896 	/* Announce that the interface is gone. */
897 	EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
898 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
899 	devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
900 
901 	/*
902 	 * Remove this ifp from ifindex2inet, ifnet queue and ifnet
903 	 * array before it is whacked.
904 	 *
905 	 * Protect ifindex2ifnet, ifnet queue and ifnet array changes
906 	 * by ifnet lock, so that non-netisr threads could get a
907 	 * consistent view.
908 	 */
909 	ifnet_lock();
910 
911 	/*
912 	 * Remove this ifp from ifindex2ifnet and maybe decrement if_index.
913 	 */
914 	ifindex2ifnet[ifp->if_index] = NULL;
915 	while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
916 		if_index--;
917 
918 	/*
919 	 * Remove this ifp from ifnet queue.
920 	 */
921 	TAILQ_REMOVE(&ifnetlist, ifp, if_link);
922 
923 	/*
924 	 * Remove this ifp from ifnet array.
925 	 */
926 	/* Free old ifnet array after sync all netisrs */
927 	old_ifnet_array = ifnet_array;
928 	ifnet_array = ifnet_array_del(ifp, old_ifnet_array);
929 
930 	ifnet_unlock();
931 
932 	/*
933 	 * Sync all netisrs so that the old ifnet array is no longer
934 	 * accessed and we can free it safely later on.
935 	 */
936 	netmsg_service_sync();
937 	ifnet_array_free(old_ifnet_array);
938 
939 	/*
940 	 * Remove routes and flush queues.
941 	 */
942 	crit_enter();
943 #ifdef IFPOLL_ENABLE
944 	if (ifp->if_flags & IFF_NPOLLING)
945 		ifpoll_deregister(ifp);
946 #endif
947 	if_down(ifp);
948 
949 	/* Decrease the mbuf clusters/jclusters limits increased by us */
950 	if (ifp->if_nmbclusters > 0)
951 		mcl_inclimit(-ifp->if_nmbclusters);
952 	if (ifp->if_nmbjclusters > 0)
953 		mjcl_inclimit(-ifp->if_nmbjclusters);
954 
955 #ifdef ALTQ
956 	if (ifq_is_enabled(&ifp->if_snd))
957 		altq_disable(&ifp->if_snd);
958 	if (ifq_is_attached(&ifp->if_snd))
959 		altq_detach(&ifp->if_snd);
960 #endif
961 
962 	/*
963 	 * Clean up all addresses.
964 	 */
965 	ifp->if_lladdr = NULL;
966 
967 	if_purgeaddrs_nolink(ifp);
968 	if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
969 		struct ifaddr *ifa;
970 
971 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
972 		KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
973 			("non-link ifaddr is left on if_addrheads"));
974 
975 		ifa_ifunlink(ifa, ifp);
976 		ifa_destroy(ifa);
977 		KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
978 			("there are still ifaddrs left on if_addrheads"));
979 	}
980 
981 #ifdef INET
982 	/*
983 	 * Remove all IPv4 kernel structures related to ifp.
984 	 */
985 	in_ifdetach(ifp);
986 #endif
987 
988 #ifdef INET6
989 	/*
990 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
991 	 * before removing routing entries below, since IPv6 interface direct
992 	 * routes are expected to be removed by the IPv6-specific kernel API.
993 	 * Otherwise, the kernel will detect some inconsistency and bark it.
994 	 */
995 	in6_ifdetach(ifp);
996 #endif
997 
998 	/*
999 	 * Delete all remaining routes using this interface
1000 	 */
1001 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1002 	    if_rtdel_dispatch);
1003 	msg.ifp = ifp;
1004 	rt_domsg_global(&msg.base);
1005 
1006 	SLIST_FOREACH(dp, &domains, dom_next)
1007 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
1008 			(*dp->dom_ifdetach)(ifp,
1009 				ifp->if_afdata[dp->dom_family]);
1010 
1011 	kfree(ifp->if_addrheads, M_IFADDR);
1012 
1013 	lwkt_synchronize_ipiqs("if_detach");
1014 	ifq_stage_detach(&ifp->if_snd);
1015 
1016 	for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
1017 		struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
1018 
1019 		kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
1020 		kfree(ifsq->ifsq_stage, M_DEVBUF);
1021 	}
1022 	kfree(ifp->if_snd.altq_subq, M_DEVBUF);
1023 
1024 	kfree(ifp->if_data_pcpu, M_DEVBUF);
1025 
1026 	crit_exit();
1027 }
1028 
1029 /*
1030  * Create interface group without members
1031  */
1032 struct ifg_group *
1033 if_creategroup(const char *groupname)
1034 {
1035         struct ifg_group        *ifg = NULL;
1036 
1037         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
1038             M_TEMP, M_NOWAIT)) == NULL)
1039                 return (NULL);
1040 
1041         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1042         ifg->ifg_refcnt = 0;
1043         ifg->ifg_carp_demoted = 0;
1044         TAILQ_INIT(&ifg->ifg_members);
1045 #if NPF > 0
1046         pfi_attach_ifgroup(ifg);
1047 #endif
1048         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
1049 
1050         return (ifg);
1051 }
1052 
1053 /*
1054  * Add a group to an interface
1055  */
1056 int
1057 if_addgroup(struct ifnet *ifp, const char *groupname)
1058 {
1059 	struct ifg_list		*ifgl;
1060 	struct ifg_group	*ifg = NULL;
1061 	struct ifg_member	*ifgm;
1062 
1063 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1064 	    groupname[strlen(groupname) - 1] <= '9')
1065 		return (EINVAL);
1066 
1067 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1068 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1069 			return (EEXIST);
1070 
1071 	if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
1072 		return (ENOMEM);
1073 
1074 	if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
1075 		kfree(ifgl, M_TEMP);
1076 		return (ENOMEM);
1077 	}
1078 
1079 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1080 		if (!strcmp(ifg->ifg_group, groupname))
1081 			break;
1082 
1083 	if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
1084 		kfree(ifgl, M_TEMP);
1085 		kfree(ifgm, M_TEMP);
1086 		return (ENOMEM);
1087 	}
1088 
1089 	ifg->ifg_refcnt++;
1090 	ifgl->ifgl_group = ifg;
1091 	ifgm->ifgm_ifp = ifp;
1092 
1093 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1094 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1095 
1096 #if NPF > 0
1097 	pfi_group_change(groupname);
1098 #endif
1099 
1100 	return (0);
1101 }
1102 
1103 /*
1104  * Remove a group from an interface
1105  */
1106 int
1107 if_delgroup(struct ifnet *ifp, const char *groupname)
1108 {
1109 	struct ifg_list		*ifgl;
1110 	struct ifg_member	*ifgm;
1111 
1112 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1113 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1114 			break;
1115 	if (ifgl == NULL)
1116 		return (ENOENT);
1117 
1118 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1119 
1120 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1121 		if (ifgm->ifgm_ifp == ifp)
1122 			break;
1123 
1124 	if (ifgm != NULL) {
1125 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1126 		kfree(ifgm, M_TEMP);
1127 	}
1128 
1129 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1130 		TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
1131 #if NPF > 0
1132 		pfi_detach_ifgroup(ifgl->ifgl_group);
1133 #endif
1134 		kfree(ifgl->ifgl_group, M_TEMP);
1135 	}
1136 
1137 	kfree(ifgl, M_TEMP);
1138 
1139 #if NPF > 0
1140 	pfi_group_change(groupname);
1141 #endif
1142 
1143 	return (0);
1144 }
1145 
1146 /*
1147  * Stores all groups from an interface in memory pointed
1148  * to by data
1149  */
1150 int
1151 if_getgroup(caddr_t data, struct ifnet *ifp)
1152 {
1153 	int			 len, error;
1154 	struct ifg_list		*ifgl;
1155 	struct ifg_req		 ifgrq, *ifgp;
1156 	struct ifgroupreq	*ifgr = (struct ifgroupreq *)data;
1157 
1158 	if (ifgr->ifgr_len == 0) {
1159 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1160 			ifgr->ifgr_len += sizeof(struct ifg_req);
1161 		return (0);
1162 	}
1163 
1164 	len = ifgr->ifgr_len;
1165 	ifgp = ifgr->ifgr_groups;
1166 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1167 		if (len < sizeof(ifgrq))
1168 			return (EINVAL);
1169 		bzero(&ifgrq, sizeof ifgrq);
1170 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1171 		    sizeof(ifgrq.ifgrq_group));
1172 		if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1173 		    sizeof(struct ifg_req))))
1174 			return (error);
1175 		len -= sizeof(ifgrq);
1176 		ifgp++;
1177 	}
1178 
1179 	return (0);
1180 }
1181 
1182 /*
1183  * Stores all members of a group in memory pointed to by data
1184  */
1185 int
1186 if_getgroupmembers(caddr_t data)
1187 {
1188 	struct ifgroupreq	*ifgr = (struct ifgroupreq *)data;
1189 	struct ifg_group	*ifg;
1190 	struct ifg_member	*ifgm;
1191 	struct ifg_req		 ifgrq, *ifgp;
1192 	int			 len, error;
1193 
1194 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1195 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1196 			break;
1197 	if (ifg == NULL)
1198 		return (ENOENT);
1199 
1200 	if (ifgr->ifgr_len == 0) {
1201 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1202 			ifgr->ifgr_len += sizeof(ifgrq);
1203 		return (0);
1204 	}
1205 
1206 	len = ifgr->ifgr_len;
1207 	ifgp = ifgr->ifgr_groups;
1208 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1209 		if (len < sizeof(ifgrq))
1210 			return (EINVAL);
1211 		bzero(&ifgrq, sizeof ifgrq);
1212 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1213 		    sizeof(ifgrq.ifgrq_member));
1214 		if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1215 		    sizeof(struct ifg_req))))
1216 			return (error);
1217 		len -= sizeof(ifgrq);
1218 		ifgp++;
1219 	}
1220 
1221 	return (0);
1222 }
1223 
1224 /*
1225  * Delete Routes for a Network Interface
1226  *
1227  * Called for each routing entry via the rnh->rnh_walktree() call above
1228  * to delete all route entries referencing a detaching network interface.
1229  *
1230  * Arguments:
1231  *	rn	pointer to node in the routing table
1232  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1233  *
1234  * Returns:
1235  *	0	successful
1236  *	errno	failed - reason indicated
1237  *
1238  */
1239 static int
1240 if_rtdel(struct radix_node *rn, void *arg)
1241 {
1242 	struct rtentry	*rt = (struct rtentry *)rn;
1243 	struct ifnet	*ifp = arg;
1244 	int		err;
1245 
1246 	if (rt->rt_ifp == ifp) {
1247 
1248 		/*
1249 		 * Protect (sorta) against walktree recursion problems
1250 		 * with cloned routes
1251 		 */
1252 		if (!(rt->rt_flags & RTF_UP))
1253 			return (0);
1254 
1255 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1256 				rt_mask(rt), rt->rt_flags,
1257 				NULL);
1258 		if (err) {
1259 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1260 		}
1261 	}
1262 
1263 	return (0);
1264 }
1265 
1266 static __inline boolean_t
1267 ifa_prefer(const struct ifaddr *cur_ifa, const struct ifaddr *old_ifa)
1268 {
1269 	if (old_ifa == NULL)
1270 		return TRUE;
1271 
1272 	if ((old_ifa->ifa_ifp->if_flags & IFF_UP) == 0 &&
1273 	    (cur_ifa->ifa_ifp->if_flags & IFF_UP))
1274 		return TRUE;
1275 	if ((old_ifa->ifa_flags & IFA_ROUTE) == 0 &&
1276 	    (cur_ifa->ifa_flags & IFA_ROUTE))
1277 		return TRUE;
1278 	return FALSE;
1279 }
1280 
1281 /*
1282  * Locate an interface based on a complete address.
1283  */
1284 struct ifaddr *
1285 ifa_ifwithaddr(struct sockaddr *addr)
1286 {
1287 	const struct ifnet_array *arr;
1288 	int i;
1289 
1290 	arr = ifnet_array_get();
1291 	for (i = 0; i < arr->ifnet_count; ++i) {
1292 		struct ifnet *ifp = arr->ifnet_arr[i];
1293 		struct ifaddr_container *ifac;
1294 
1295 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1296 			struct ifaddr *ifa = ifac->ifa;
1297 
1298 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1299 				continue;
1300 			if (sa_equal(addr, ifa->ifa_addr))
1301 				return (ifa);
1302 			if ((ifp->if_flags & IFF_BROADCAST) &&
1303 			    ifa->ifa_broadaddr &&
1304 			    /* IPv6 doesn't have broadcast */
1305 			    ifa->ifa_broadaddr->sa_len != 0 &&
1306 			    sa_equal(ifa->ifa_broadaddr, addr))
1307 				return (ifa);
1308 		}
1309 	}
1310 	return (NULL);
1311 }
1312 
1313 /*
1314  * Locate the point to point interface with a given destination address.
1315  */
1316 struct ifaddr *
1317 ifa_ifwithdstaddr(struct sockaddr *addr)
1318 {
1319 	const struct ifnet_array *arr;
1320 	int i;
1321 
1322 	arr = ifnet_array_get();
1323 	for (i = 0; i < arr->ifnet_count; ++i) {
1324 		struct ifnet *ifp = arr->ifnet_arr[i];
1325 		struct ifaddr_container *ifac;
1326 
1327 		if (!(ifp->if_flags & IFF_POINTOPOINT))
1328 			continue;
1329 
1330 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1331 			struct ifaddr *ifa = ifac->ifa;
1332 
1333 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1334 				continue;
1335 			if (ifa->ifa_dstaddr &&
1336 			    sa_equal(addr, ifa->ifa_dstaddr))
1337 				return (ifa);
1338 		}
1339 	}
1340 	return (NULL);
1341 }
1342 
1343 /*
1344  * Find an interface on a specific network.  If many, choice
1345  * is most specific found.
1346  */
1347 struct ifaddr *
1348 ifa_ifwithnet(struct sockaddr *addr)
1349 {
1350 	struct ifaddr *ifa_maybe = NULL;
1351 	u_int af = addr->sa_family;
1352 	char *addr_data = addr->sa_data, *cplim;
1353 	const struct ifnet_array *arr;
1354 	int i;
1355 
1356 	/*
1357 	 * AF_LINK addresses can be looked up directly by their index number,
1358 	 * so do that if we can.
1359 	 */
1360 	if (af == AF_LINK) {
1361 		struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1362 
1363 		if (sdl->sdl_index && sdl->sdl_index <= if_index)
1364 			return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1365 	}
1366 
1367 	/*
1368 	 * Scan though each interface, looking for ones that have
1369 	 * addresses in this address family.
1370 	 */
1371 	arr = ifnet_array_get();
1372 	for (i = 0; i < arr->ifnet_count; ++i) {
1373 		struct ifnet *ifp = arr->ifnet_arr[i];
1374 		struct ifaddr_container *ifac;
1375 
1376 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1377 			struct ifaddr *ifa = ifac->ifa;
1378 			char *cp, *cp2, *cp3;
1379 
1380 			if (ifa->ifa_addr->sa_family != af)
1381 next:				continue;
1382 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1383 				/*
1384 				 * This is a bit broken as it doesn't
1385 				 * take into account that the remote end may
1386 				 * be a single node in the network we are
1387 				 * looking for.
1388 				 * The trouble is that we don't know the
1389 				 * netmask for the remote end.
1390 				 */
1391 				if (ifa->ifa_dstaddr != NULL &&
1392 				    sa_equal(addr, ifa->ifa_dstaddr))
1393 					return (ifa);
1394 			} else {
1395 				/*
1396 				 * if we have a special address handler,
1397 				 * then use it instead of the generic one.
1398 				 */
1399 				if (ifa->ifa_claim_addr) {
1400 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1401 						return (ifa);
1402 					} else {
1403 						continue;
1404 					}
1405 				}
1406 
1407 				/*
1408 				 * Scan all the bits in the ifa's address.
1409 				 * If a bit dissagrees with what we are
1410 				 * looking for, mask it with the netmask
1411 				 * to see if it really matters.
1412 				 * (A byte at a time)
1413 				 */
1414 				if (ifa->ifa_netmask == 0)
1415 					continue;
1416 				cp = addr_data;
1417 				cp2 = ifa->ifa_addr->sa_data;
1418 				cp3 = ifa->ifa_netmask->sa_data;
1419 				cplim = ifa->ifa_netmask->sa_len +
1420 					(char *)ifa->ifa_netmask;
1421 				while (cp3 < cplim)
1422 					if ((*cp++ ^ *cp2++) & *cp3++)
1423 						goto next; /* next address! */
1424 				/*
1425 				 * If the netmask of what we just found
1426 				 * is more specific than what we had before
1427 				 * (if we had one) then remember the new one
1428 				 * before continuing to search for an even
1429 				 * better one.  If the netmasks are equal,
1430 				 * we prefer the this ifa based on the result
1431 				 * of ifa_prefer().
1432 				 */
1433 				if (ifa_maybe == NULL ||
1434 				    rn_refines((char *)ifa->ifa_netmask,
1435 				        (char *)ifa_maybe->ifa_netmask) ||
1436 				    (sa_equal(ifa_maybe->ifa_netmask,
1437 				        ifa->ifa_netmask) &&
1438 				     ifa_prefer(ifa, ifa_maybe)))
1439 					ifa_maybe = ifa;
1440 			}
1441 		}
1442 	}
1443 	return (ifa_maybe);
1444 }
1445 
1446 /*
1447  * Find an interface address specific to an interface best matching
1448  * a given address.
1449  */
1450 struct ifaddr *
1451 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1452 {
1453 	struct ifaddr_container *ifac;
1454 	char *cp, *cp2, *cp3;
1455 	char *cplim;
1456 	struct ifaddr *ifa_maybe = NULL;
1457 	u_int af = addr->sa_family;
1458 
1459 	if (af >= AF_MAX)
1460 		return (0);
1461 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1462 		struct ifaddr *ifa = ifac->ifa;
1463 
1464 		if (ifa->ifa_addr->sa_family != af)
1465 			continue;
1466 		if (ifa_maybe == NULL)
1467 			ifa_maybe = ifa;
1468 		if (ifa->ifa_netmask == NULL) {
1469 			if (sa_equal(addr, ifa->ifa_addr) ||
1470 			    (ifa->ifa_dstaddr != NULL &&
1471 			     sa_equal(addr, ifa->ifa_dstaddr)))
1472 				return (ifa);
1473 			continue;
1474 		}
1475 		if (ifp->if_flags & IFF_POINTOPOINT) {
1476 			if (sa_equal(addr, ifa->ifa_dstaddr))
1477 				return (ifa);
1478 		} else {
1479 			cp = addr->sa_data;
1480 			cp2 = ifa->ifa_addr->sa_data;
1481 			cp3 = ifa->ifa_netmask->sa_data;
1482 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1483 			for (; cp3 < cplim; cp3++)
1484 				if ((*cp++ ^ *cp2++) & *cp3)
1485 					break;
1486 			if (cp3 == cplim)
1487 				return (ifa);
1488 		}
1489 	}
1490 	return (ifa_maybe);
1491 }
1492 
1493 /*
1494  * Default action when installing a route with a Link Level gateway.
1495  * Lookup an appropriate real ifa to point to.
1496  * This should be moved to /sys/net/link.c eventually.
1497  */
1498 static void
1499 link_rtrequest(int cmd, struct rtentry *rt)
1500 {
1501 	struct ifaddr *ifa;
1502 	struct sockaddr *dst;
1503 	struct ifnet *ifp;
1504 
1505 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1506 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1507 		return;
1508 	ifa = ifaof_ifpforaddr(dst, ifp);
1509 	if (ifa != NULL) {
1510 		IFAFREE(rt->rt_ifa);
1511 		IFAREF(ifa);
1512 		rt->rt_ifa = ifa;
1513 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1514 			ifa->ifa_rtrequest(cmd, rt);
1515 	}
1516 }
1517 
1518 struct netmsg_ifroute {
1519 	struct netmsg_base	base;
1520 	struct ifnet		*ifp;
1521 	int			flag;
1522 	int			fam;
1523 };
1524 
1525 /*
1526  * Mark an interface down and notify protocols of the transition.
1527  */
1528 static void
1529 if_unroute_dispatch(netmsg_t nmsg)
1530 {
1531 	struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1532 	struct ifnet *ifp = msg->ifp;
1533 	int flag = msg->flag, fam = msg->fam;
1534 	struct ifaddr_container *ifac;
1535 
1536 	ifp->if_flags &= ~flag;
1537 	getmicrotime(&ifp->if_lastchange);
1538 	/*
1539 	 * The ifaddr processing in the following loop will block,
1540 	 * however, this function is called in netisr0, in which
1541 	 * ifaddr list changes happen, so we don't care about the
1542 	 * blockness of the ifaddr processing here.
1543 	 */
1544 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1545 		struct ifaddr *ifa = ifac->ifa;
1546 
1547 		/* Ignore marker */
1548 		if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1549 			continue;
1550 
1551 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1552 			kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1553 	}
1554 	ifq_purge_all(&ifp->if_snd);
1555 	rt_ifmsg(ifp);
1556 
1557 	lwkt_replymsg(&nmsg->lmsg, 0);
1558 }
1559 
1560 void
1561 if_unroute(struct ifnet *ifp, int flag, int fam)
1562 {
1563 	struct netmsg_ifroute msg;
1564 
1565 	ASSERT_CANDOMSG_NETISR0(curthread);
1566 
1567 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1568 	    if_unroute_dispatch);
1569 	msg.ifp = ifp;
1570 	msg.flag = flag;
1571 	msg.fam = fam;
1572 	lwkt_domsg(netisr_cpuport(0), &msg.base.lmsg, 0);
1573 }
1574 
1575 /*
1576  * Mark an interface up and notify protocols of the transition.
1577  */
1578 static void
1579 if_route_dispatch(netmsg_t nmsg)
1580 {
1581 	struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1582 	struct ifnet *ifp = msg->ifp;
1583 	int flag = msg->flag, fam = msg->fam;
1584 	struct ifaddr_container *ifac;
1585 
1586 	ifq_purge_all(&ifp->if_snd);
1587 	ifp->if_flags |= flag;
1588 	getmicrotime(&ifp->if_lastchange);
1589 	/*
1590 	 * The ifaddr processing in the following loop will block,
1591 	 * however, this function is called in netisr0, in which
1592 	 * ifaddr list changes happen, so we don't care about the
1593 	 * blockness of the ifaddr processing here.
1594 	 */
1595 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1596 		struct ifaddr *ifa = ifac->ifa;
1597 
1598 		/* Ignore marker */
1599 		if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1600 			continue;
1601 
1602 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1603 			kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1604 	}
1605 	rt_ifmsg(ifp);
1606 #ifdef INET6
1607 	in6_if_up(ifp);
1608 #endif
1609 
1610 	lwkt_replymsg(&nmsg->lmsg, 0);
1611 }
1612 
1613 void
1614 if_route(struct ifnet *ifp, int flag, int fam)
1615 {
1616 	struct netmsg_ifroute msg;
1617 
1618 	ASSERT_CANDOMSG_NETISR0(curthread);
1619 
1620 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1621 	    if_route_dispatch);
1622 	msg.ifp = ifp;
1623 	msg.flag = flag;
1624 	msg.fam = fam;
1625 	lwkt_domsg(netisr_cpuport(0), &msg.base.lmsg, 0);
1626 }
1627 
1628 /*
1629  * Mark an interface down and notify protocols of the transition.  An
1630  * interface going down is also considered to be a synchronizing event.
1631  * We must ensure that all packet processing related to the interface
1632  * has completed before we return so e.g. the caller can free the ifnet
1633  * structure that the mbufs may be referencing.
1634  *
1635  * NOTE: must be called at splnet or eqivalent.
1636  */
1637 void
1638 if_down(struct ifnet *ifp)
1639 {
1640 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1641 	netmsg_service_sync();
1642 }
1643 
1644 /*
1645  * Mark an interface up and notify protocols of
1646  * the transition.
1647  * NOTE: must be called at splnet or eqivalent.
1648  */
1649 void
1650 if_up(struct ifnet *ifp)
1651 {
1652 	if_route(ifp, IFF_UP, AF_UNSPEC);
1653 }
1654 
1655 /*
1656  * Process a link state change.
1657  * NOTE: must be called at splsoftnet or equivalent.
1658  */
1659 void
1660 if_link_state_change(struct ifnet *ifp)
1661 {
1662 	int link_state = ifp->if_link_state;
1663 
1664 	rt_ifmsg(ifp);
1665 	devctl_notify("IFNET", ifp->if_xname,
1666 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1667 }
1668 
1669 /*
1670  * Handle interface watchdog timer routines.  Called
1671  * from softclock, we decrement timers (if set) and
1672  * call the appropriate interface routine on expiration.
1673  */
1674 static void
1675 if_slowtimo_dispatch(netmsg_t nmsg)
1676 {
1677 	struct globaldata *gd = mycpu;
1678 	const struct ifnet_array *arr;
1679 	int i;
1680 
1681 	ASSERT_IN_NETISR(0);
1682 
1683 	crit_enter_gd(gd);
1684 	lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1685 	crit_exit_gd(gd);
1686 
1687 	arr = ifnet_array_get();
1688 	for (i = 0; i < arr->ifnet_count; ++i) {
1689 		struct ifnet *ifp = arr->ifnet_arr[i];
1690 
1691 		crit_enter_gd(gd);
1692 
1693 		if (if_stats_compat) {
1694 			IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1695 			IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1696 			IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1697 			IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1698 			IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1699 			IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1700 			IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1701 			IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1702 			IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1703 			IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1704 			IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1705 			IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
1706 		}
1707 
1708 		if (ifp->if_timer == 0 || --ifp->if_timer) {
1709 			crit_exit_gd(gd);
1710 			continue;
1711 		}
1712 		if (ifp->if_watchdog) {
1713 			if (ifnet_tryserialize_all(ifp)) {
1714 				(*ifp->if_watchdog)(ifp);
1715 				ifnet_deserialize_all(ifp);
1716 			} else {
1717 				/* try again next timeout */
1718 				++ifp->if_timer;
1719 			}
1720 		}
1721 
1722 		crit_exit_gd(gd);
1723 	}
1724 
1725 	callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1726 }
1727 
1728 static void
1729 if_slowtimo(void *arg __unused)
1730 {
1731 	struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1732 
1733 	KASSERT(mycpuid == 0, ("not on cpu0"));
1734 	crit_enter();
1735 	if (lmsg->ms_flags & MSGF_DONE)
1736 		lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1737 	crit_exit();
1738 }
1739 
1740 /*
1741  * Map interface name to
1742  * interface structure pointer.
1743  */
1744 struct ifnet *
1745 ifunit(const char *name)
1746 {
1747 	struct ifnet *ifp;
1748 
1749 	/*
1750 	 * Search all the interfaces for this name/number
1751 	 */
1752 	KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1753 
1754 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1755 		if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1756 			break;
1757 	}
1758 	return (ifp);
1759 }
1760 
1761 struct ifnet *
1762 ifunit_netisr(const char *name)
1763 {
1764 	const struct ifnet_array *arr;
1765 	int i;
1766 
1767 	/*
1768 	 * Search all the interfaces for this name/number
1769 	 */
1770 
1771 	arr = ifnet_array_get();
1772 	for (i = 0; i < arr->ifnet_count; ++i) {
1773 		struct ifnet *ifp = arr->ifnet_arr[i];
1774 
1775 		if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1776 			return ifp;
1777 	}
1778 	return NULL;
1779 }
1780 
1781 /*
1782  * Interface ioctls.
1783  */
1784 int
1785 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1786 {
1787 	struct ifnet *ifp;
1788 	struct ifreq *ifr;
1789 	struct ifstat *ifs;
1790 	int error, do_ifup = 0;
1791 	short oif_flags;
1792 	int new_flags;
1793 	size_t namelen, onamelen;
1794 	char new_name[IFNAMSIZ];
1795 	struct ifaddr *ifa;
1796 	struct sockaddr_dl *sdl;
1797 
1798 	switch (cmd) {
1799 	case SIOCGIFCONF:
1800 	case OSIOCGIFCONF:
1801 		return (ifconf(cmd, data, cred));
1802 	default:
1803 		break;
1804 	}
1805 
1806 	ifr = (struct ifreq *)data;
1807 
1808 	switch (cmd) {
1809 	case SIOCIFCREATE:
1810 	case SIOCIFCREATE2:
1811 		if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1812 			return (error);
1813 		return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1814 		    	cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1815 	case SIOCIFDESTROY:
1816 		if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1817 			return (error);
1818 		return (if_clone_destroy(ifr->ifr_name));
1819 	case SIOCIFGCLONERS:
1820 		return (if_clone_list((struct if_clonereq *)data));
1821 	default:
1822 		break;
1823 	}
1824 
1825 	/*
1826 	 * Nominal ioctl through interface, lookup the ifp and obtain a
1827 	 * lock to serialize the ifconfig ioctl operation.
1828 	 */
1829 	ifnet_lock();
1830 
1831 	ifp = ifunit(ifr->ifr_name);
1832 	if (ifp == NULL) {
1833 		ifnet_unlock();
1834 		return (ENXIO);
1835 	}
1836 	error = 0;
1837 
1838 	switch (cmd) {
1839 	case SIOCGIFINDEX:
1840 		ifr->ifr_index = ifp->if_index;
1841 		break;
1842 
1843 	case SIOCGIFFLAGS:
1844 		ifr->ifr_flags = ifp->if_flags;
1845 		ifr->ifr_flagshigh = ifp->if_flags >> 16;
1846 		break;
1847 
1848 	case SIOCGIFCAP:
1849 		ifr->ifr_reqcap = ifp->if_capabilities;
1850 		ifr->ifr_curcap = ifp->if_capenable;
1851 		break;
1852 
1853 	case SIOCGIFMETRIC:
1854 		ifr->ifr_metric = ifp->if_metric;
1855 		break;
1856 
1857 	case SIOCGIFMTU:
1858 		ifr->ifr_mtu = ifp->if_mtu;
1859 		break;
1860 
1861 	case SIOCGIFTSOLEN:
1862 		ifr->ifr_tsolen = ifp->if_tsolen;
1863 		break;
1864 
1865 	case SIOCGIFDATA:
1866 		error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1867 				sizeof(ifp->if_data));
1868 		break;
1869 
1870 	case SIOCGIFPHYS:
1871 		ifr->ifr_phys = ifp->if_physical;
1872 		break;
1873 
1874 	case SIOCGIFPOLLCPU:
1875 		ifr->ifr_pollcpu = -1;
1876 		break;
1877 
1878 	case SIOCSIFPOLLCPU:
1879 		break;
1880 
1881 	case SIOCSIFFLAGS:
1882 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1883 		if (error)
1884 			break;
1885 		new_flags = (ifr->ifr_flags & 0xffff) |
1886 		    (ifr->ifr_flagshigh << 16);
1887 		if (ifp->if_flags & IFF_SMART) {
1888 			/* Smart drivers twiddle their own routes */
1889 		} else if (ifp->if_flags & IFF_UP &&
1890 		    (new_flags & IFF_UP) == 0) {
1891 			if_down(ifp);
1892 		} else if (new_flags & IFF_UP &&
1893 		    (ifp->if_flags & IFF_UP) == 0) {
1894 			do_ifup = 1;
1895 		}
1896 
1897 #ifdef IFPOLL_ENABLE
1898 		if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1899 			if (new_flags & IFF_NPOLLING)
1900 				ifpoll_register(ifp);
1901 			else
1902 				ifpoll_deregister(ifp);
1903 		}
1904 #endif
1905 
1906 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1907 			(new_flags &~ IFF_CANTCHANGE);
1908 		if (new_flags & IFF_PPROMISC) {
1909 			/* Permanently promiscuous mode requested */
1910 			ifp->if_flags |= IFF_PROMISC;
1911 		} else if (ifp->if_pcount == 0) {
1912 			ifp->if_flags &= ~IFF_PROMISC;
1913 		}
1914 		if (ifp->if_ioctl) {
1915 			ifnet_serialize_all(ifp);
1916 			ifp->if_ioctl(ifp, cmd, data, cred);
1917 			ifnet_deserialize_all(ifp);
1918 		}
1919 		if (do_ifup)
1920 			if_up(ifp);
1921 		getmicrotime(&ifp->if_lastchange);
1922 		break;
1923 
1924 	case SIOCSIFCAP:
1925 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1926 		if (error)
1927 			break;
1928 		if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1929 			error = EINVAL;
1930 			break;
1931 		}
1932 		ifnet_serialize_all(ifp);
1933 		ifp->if_ioctl(ifp, cmd, data, cred);
1934 		ifnet_deserialize_all(ifp);
1935 		break;
1936 
1937 	case SIOCSIFNAME:
1938 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1939 		if (error)
1940 			break;
1941 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1942 		if (error)
1943 			break;
1944 		if (new_name[0] == '\0') {
1945 			error = EINVAL;
1946 			break;
1947 		}
1948 		if (ifunit(new_name) != NULL) {
1949 			error = EEXIST;
1950 			break;
1951 		}
1952 
1953 		EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1954 
1955 		/* Announce the departure of the interface. */
1956 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1957 
1958 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1959 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1960 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1961 		namelen = strlen(new_name);
1962 		onamelen = sdl->sdl_nlen;
1963 		/*
1964 		 * Move the address if needed.  This is safe because we
1965 		 * allocate space for a name of length IFNAMSIZ when we
1966 		 * create this in if_attach().
1967 		 */
1968 		if (namelen != onamelen) {
1969 			bcopy(sdl->sdl_data + onamelen,
1970 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1971 		}
1972 		bcopy(new_name, sdl->sdl_data, namelen);
1973 		sdl->sdl_nlen = namelen;
1974 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1975 		bzero(sdl->sdl_data, onamelen);
1976 		while (namelen != 0)
1977 			sdl->sdl_data[--namelen] = 0xff;
1978 
1979 		EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1980 
1981 		/* Announce the return of the interface. */
1982 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1983 		break;
1984 
1985 	case SIOCSIFMETRIC:
1986 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1987 		if (error)
1988 			break;
1989 		ifp->if_metric = ifr->ifr_metric;
1990 		getmicrotime(&ifp->if_lastchange);
1991 		break;
1992 
1993 	case SIOCSIFPHYS:
1994 		error = priv_check_cred(cred, PRIV_ROOT, 0);
1995 		if (error)
1996 			break;
1997 		if (ifp->if_ioctl == NULL) {
1998 		        error = EOPNOTSUPP;
1999 			break;
2000 		}
2001 		ifnet_serialize_all(ifp);
2002 		error = ifp->if_ioctl(ifp, cmd, data, cred);
2003 		ifnet_deserialize_all(ifp);
2004 		if (error == 0)
2005 			getmicrotime(&ifp->if_lastchange);
2006 		break;
2007 
2008 	case SIOCSIFMTU:
2009 	{
2010 		u_long oldmtu = ifp->if_mtu;
2011 
2012 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2013 		if (error)
2014 			break;
2015 		if (ifp->if_ioctl == NULL) {
2016 			error = EOPNOTSUPP;
2017 			break;
2018 		}
2019 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2020 			error = EINVAL;
2021 			break;
2022 		}
2023 		ifnet_serialize_all(ifp);
2024 		error = ifp->if_ioctl(ifp, cmd, data, cred);
2025 		ifnet_deserialize_all(ifp);
2026 		if (error == 0) {
2027 			getmicrotime(&ifp->if_lastchange);
2028 			rt_ifmsg(ifp);
2029 		}
2030 		/*
2031 		 * If the link MTU changed, do network layer specific procedure.
2032 		 */
2033 		if (ifp->if_mtu != oldmtu) {
2034 #ifdef INET6
2035 			nd6_setmtu(ifp);
2036 #endif
2037 		}
2038 		break;
2039 	}
2040 
2041 	case SIOCSIFTSOLEN:
2042 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2043 		if (error)
2044 			break;
2045 
2046 		/* XXX need driver supplied upper limit */
2047 		if (ifr->ifr_tsolen <= 0) {
2048 			error = EINVAL;
2049 			break;
2050 		}
2051 		ifp->if_tsolen = ifr->ifr_tsolen;
2052 		break;
2053 
2054 	case SIOCADDMULTI:
2055 	case SIOCDELMULTI:
2056 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2057 		if (error)
2058 			break;
2059 
2060 		/* Don't allow group membership on non-multicast interfaces. */
2061 		if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2062 			error = EOPNOTSUPP;
2063 			break;
2064 		}
2065 
2066 		/* Don't let users screw up protocols' entries. */
2067 		if (ifr->ifr_addr.sa_family != AF_LINK) {
2068 			error = EINVAL;
2069 			break;
2070 		}
2071 
2072 		if (cmd == SIOCADDMULTI) {
2073 			struct ifmultiaddr *ifma;
2074 			error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2075 		} else {
2076 			error = if_delmulti(ifp, &ifr->ifr_addr);
2077 		}
2078 		if (error == 0)
2079 			getmicrotime(&ifp->if_lastchange);
2080 		break;
2081 
2082 	case SIOCSIFPHYADDR:
2083 	case SIOCDIFPHYADDR:
2084 #ifdef INET6
2085 	case SIOCSIFPHYADDR_IN6:
2086 #endif
2087 	case SIOCSLIFPHYADDR:
2088         case SIOCSIFMEDIA:
2089 	case SIOCSIFGENERIC:
2090 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2091 		if (error)
2092 			break;
2093 		if (ifp->if_ioctl == 0) {
2094 			error = EOPNOTSUPP;
2095 			break;
2096 		}
2097 		ifnet_serialize_all(ifp);
2098 		error = ifp->if_ioctl(ifp, cmd, data, cred);
2099 		ifnet_deserialize_all(ifp);
2100 		if (error == 0)
2101 			getmicrotime(&ifp->if_lastchange);
2102 		break;
2103 
2104 	case SIOCGIFSTATUS:
2105 		ifs = (struct ifstat *)data;
2106 		ifs->ascii[0] = '\0';
2107 		/* fall through */
2108 	case SIOCGIFPSRCADDR:
2109 	case SIOCGIFPDSTADDR:
2110 	case SIOCGLIFPHYADDR:
2111 	case SIOCGIFMEDIA:
2112 	case SIOCGIFGENERIC:
2113 		if (ifp->if_ioctl == NULL) {
2114 			error = EOPNOTSUPP;
2115 			break;
2116 		}
2117 		ifnet_serialize_all(ifp);
2118 		error = ifp->if_ioctl(ifp, cmd, data, cred);
2119 		ifnet_deserialize_all(ifp);
2120 		break;
2121 
2122 	case SIOCSIFLLADDR:
2123 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2124 		if (error)
2125 			break;
2126 		error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2127 				     ifr->ifr_addr.sa_len);
2128 		EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2129 		break;
2130 
2131 	default:
2132 		oif_flags = ifp->if_flags;
2133 		if (so->so_proto == 0) {
2134 			error = EOPNOTSUPP;
2135 			break;
2136 		}
2137 		error = so_pru_control_direct(so, cmd, data, ifp);
2138 
2139 		if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2140 #ifdef INET6
2141 			DELAY(100);/* XXX: temporary workaround for fxp issue*/
2142 			if (ifp->if_flags & IFF_UP) {
2143 				crit_enter();
2144 				in6_if_up(ifp);
2145 				crit_exit();
2146 			}
2147 #endif
2148 		}
2149 		break;
2150 	}
2151 
2152 	ifnet_unlock();
2153 	return (error);
2154 }
2155 
2156 /*
2157  * Set/clear promiscuous mode on interface ifp based on the truth value
2158  * of pswitch.  The calls are reference counted so that only the first
2159  * "on" request actually has an effect, as does the final "off" request.
2160  * Results are undefined if the "off" and "on" requests are not matched.
2161  */
2162 int
2163 ifpromisc(struct ifnet *ifp, int pswitch)
2164 {
2165 	struct ifreq ifr;
2166 	int error;
2167 	int oldflags;
2168 
2169 	oldflags = ifp->if_flags;
2170 	if (ifp->if_flags & IFF_PPROMISC) {
2171 		/* Do nothing if device is in permanently promiscuous mode */
2172 		ifp->if_pcount += pswitch ? 1 : -1;
2173 		return (0);
2174 	}
2175 	if (pswitch) {
2176 		/*
2177 		 * If the device is not configured up, we cannot put it in
2178 		 * promiscuous mode.
2179 		 */
2180 		if ((ifp->if_flags & IFF_UP) == 0)
2181 			return (ENETDOWN);
2182 		if (ifp->if_pcount++ != 0)
2183 			return (0);
2184 		ifp->if_flags |= IFF_PROMISC;
2185 		log(LOG_INFO, "%s: promiscuous mode enabled\n",
2186 		    ifp->if_xname);
2187 	} else {
2188 		if (--ifp->if_pcount > 0)
2189 			return (0);
2190 		ifp->if_flags &= ~IFF_PROMISC;
2191 		log(LOG_INFO, "%s: promiscuous mode disabled\n",
2192 		    ifp->if_xname);
2193 	}
2194 	ifr.ifr_flags = ifp->if_flags;
2195 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2196 	ifnet_serialize_all(ifp);
2197 	error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2198 	ifnet_deserialize_all(ifp);
2199 	if (error == 0)
2200 		rt_ifmsg(ifp);
2201 	else
2202 		ifp->if_flags = oldflags;
2203 	return error;
2204 }
2205 
2206 /*
2207  * Return interface configuration
2208  * of system.  List may be used
2209  * in later ioctl's (above) to get
2210  * other information.
2211  */
2212 static int
2213 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2214 {
2215 	struct ifconf *ifc = (struct ifconf *)data;
2216 	struct ifnet *ifp;
2217 	struct sockaddr *sa;
2218 	struct ifreq ifr, *ifrp;
2219 	int space = ifc->ifc_len, error = 0;
2220 
2221 	ifrp = ifc->ifc_req;
2222 
2223 	ifnet_lock();
2224 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2225 		struct ifaddr_container *ifac, *ifac_mark;
2226 		struct ifaddr_marker mark;
2227 		struct ifaddrhead *head;
2228 		int addrs;
2229 
2230 		if (space <= sizeof ifr)
2231 			break;
2232 
2233 		/*
2234 		 * Zero the stack declared structure first to prevent
2235 		 * memory disclosure.
2236 		 */
2237 		bzero(&ifr, sizeof(ifr));
2238 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2239 		    >= sizeof(ifr.ifr_name)) {
2240 			error = ENAMETOOLONG;
2241 			break;
2242 		}
2243 
2244 		/*
2245 		 * Add a marker, since copyout() could block and during that
2246 		 * period the list could be changed.  Inserting the marker to
2247 		 * the header of the list will not cause trouble for the code
2248 		 * assuming that the first element of the list is AF_LINK; the
2249 		 * marker will be moved to the next position w/o blocking.
2250 		 */
2251 		ifa_marker_init(&mark, ifp);
2252 		ifac_mark = &mark.ifac;
2253 		head = &ifp->if_addrheads[mycpuid];
2254 
2255 		addrs = 0;
2256 		TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2257 		while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2258 			struct ifaddr *ifa = ifac->ifa;
2259 
2260 			TAILQ_REMOVE(head, ifac_mark, ifa_link);
2261 			TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2262 
2263 			/* Ignore marker */
2264 			if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2265 				continue;
2266 
2267 			if (space <= sizeof ifr)
2268 				break;
2269 			sa = ifa->ifa_addr;
2270 			if (cred->cr_prison &&
2271 			    prison_if(cred, sa))
2272 				continue;
2273 			addrs++;
2274 			/*
2275 			 * Keep a reference on this ifaddr, so that it will
2276 			 * not be destroyed when its address is copied to
2277 			 * the userland, which could block.
2278 			 */
2279 			IFAREF(ifa);
2280 			if (sa->sa_len <= sizeof(*sa)) {
2281 				ifr.ifr_addr = *sa;
2282 				error = copyout(&ifr, ifrp, sizeof ifr);
2283 				ifrp++;
2284 			} else {
2285 				if (space < (sizeof ifr) + sa->sa_len -
2286 					    sizeof(*sa)) {
2287 					IFAFREE(ifa);
2288 					break;
2289 				}
2290 				space -= sa->sa_len - sizeof(*sa);
2291 				error = copyout(&ifr, ifrp,
2292 						sizeof ifr.ifr_name);
2293 				if (error == 0)
2294 					error = copyout(sa, &ifrp->ifr_addr,
2295 							sa->sa_len);
2296 				ifrp = (struct ifreq *)
2297 					(sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2298 			}
2299 			IFAFREE(ifa);
2300 			if (error)
2301 				break;
2302 			space -= sizeof ifr;
2303 		}
2304 		TAILQ_REMOVE(head, ifac_mark, ifa_link);
2305 		if (error)
2306 			break;
2307 		if (!addrs) {
2308 			bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2309 			error = copyout(&ifr, ifrp, sizeof ifr);
2310 			if (error)
2311 				break;
2312 			space -= sizeof ifr;
2313 			ifrp++;
2314 		}
2315 	}
2316 	ifnet_unlock();
2317 
2318 	ifc->ifc_len -= space;
2319 	return (error);
2320 }
2321 
2322 /*
2323  * Just like if_promisc(), but for all-multicast-reception mode.
2324  */
2325 int
2326 if_allmulti(struct ifnet *ifp, int onswitch)
2327 {
2328 	int error = 0;
2329 	struct ifreq ifr;
2330 
2331 	crit_enter();
2332 
2333 	if (onswitch) {
2334 		if (ifp->if_amcount++ == 0) {
2335 			ifp->if_flags |= IFF_ALLMULTI;
2336 			ifr.ifr_flags = ifp->if_flags;
2337 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2338 			ifnet_serialize_all(ifp);
2339 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2340 					      NULL);
2341 			ifnet_deserialize_all(ifp);
2342 		}
2343 	} else {
2344 		if (ifp->if_amcount > 1) {
2345 			ifp->if_amcount--;
2346 		} else {
2347 			ifp->if_amcount = 0;
2348 			ifp->if_flags &= ~IFF_ALLMULTI;
2349 			ifr.ifr_flags = ifp->if_flags;
2350 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2351 			ifnet_serialize_all(ifp);
2352 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2353 					      NULL);
2354 			ifnet_deserialize_all(ifp);
2355 		}
2356 	}
2357 
2358 	crit_exit();
2359 
2360 	if (error == 0)
2361 		rt_ifmsg(ifp);
2362 	return error;
2363 }
2364 
2365 /*
2366  * Add a multicast listenership to the interface in question.
2367  * The link layer provides a routine which converts
2368  */
2369 int
2370 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2371     struct ifmultiaddr **retifma)
2372 {
2373 	struct sockaddr *llsa, *dupsa;
2374 	int error;
2375 	struct ifmultiaddr *ifma;
2376 
2377 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
2378 
2379 	/*
2380 	 * If the matching multicast address already exists
2381 	 * then don't add a new one, just add a reference
2382 	 */
2383 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2384 		if (sa_equal(sa, ifma->ifma_addr)) {
2385 			ifma->ifma_refcount++;
2386 			if (retifma)
2387 				*retifma = ifma;
2388 			return 0;
2389 		}
2390 	}
2391 
2392 	/*
2393 	 * Give the link layer a chance to accept/reject it, and also
2394 	 * find out which AF_LINK address this maps to, if it isn't one
2395 	 * already.
2396 	 */
2397 	if (ifp->if_resolvemulti) {
2398 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2399 		if (error)
2400 			return error;
2401 	} else {
2402 		llsa = NULL;
2403 	}
2404 
2405 	ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2406 	dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2407 	bcopy(sa, dupsa, sa->sa_len);
2408 
2409 	ifma->ifma_addr = dupsa;
2410 	ifma->ifma_lladdr = llsa;
2411 	ifma->ifma_ifp = ifp;
2412 	ifma->ifma_refcount = 1;
2413 	ifma->ifma_protospec = NULL;
2414 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2415 
2416 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2417 	if (retifma)
2418 		*retifma = ifma;
2419 
2420 	if (llsa != NULL) {
2421 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2422 			if (sa_equal(ifma->ifma_addr, llsa))
2423 				break;
2424 		}
2425 		if (ifma) {
2426 			ifma->ifma_refcount++;
2427 		} else {
2428 			ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2429 			dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2430 			bcopy(llsa, dupsa, llsa->sa_len);
2431 			ifma->ifma_addr = dupsa;
2432 			ifma->ifma_ifp = ifp;
2433 			ifma->ifma_refcount = 1;
2434 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2435 		}
2436 	}
2437 	/*
2438 	 * We are certain we have added something, so call down to the
2439 	 * interface to let them know about it.
2440 	 */
2441 	if (ifp->if_ioctl)
2442 		ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2443 
2444 	return 0;
2445 }
2446 
2447 int
2448 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2449     struct ifmultiaddr **retifma)
2450 {
2451 	int error;
2452 
2453 	ifnet_serialize_all(ifp);
2454 	error = if_addmulti_serialized(ifp, sa, retifma);
2455 	ifnet_deserialize_all(ifp);
2456 
2457 	return error;
2458 }
2459 
2460 /*
2461  * Remove a reference to a multicast address on this interface.  Yell
2462  * if the request does not match an existing membership.
2463  */
2464 static int
2465 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2466 {
2467 	struct ifmultiaddr *ifma;
2468 
2469 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
2470 
2471 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2472 		if (sa_equal(sa, ifma->ifma_addr))
2473 			break;
2474 	if (ifma == NULL)
2475 		return ENOENT;
2476 
2477 	if (ifma->ifma_refcount > 1) {
2478 		ifma->ifma_refcount--;
2479 		return 0;
2480 	}
2481 
2482 	rt_newmaddrmsg(RTM_DELMADDR, ifma);
2483 	sa = ifma->ifma_lladdr;
2484 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2485 	/*
2486 	 * Make sure the interface driver is notified
2487 	 * in the case of a link layer mcast group being left.
2488 	 */
2489 	if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2490 		ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2491 	kfree(ifma->ifma_addr, M_IFMADDR);
2492 	kfree(ifma, M_IFMADDR);
2493 	if (sa == NULL)
2494 		return 0;
2495 
2496 	/*
2497 	 * Now look for the link-layer address which corresponds to
2498 	 * this network address.  It had been squirreled away in
2499 	 * ifma->ifma_lladdr for this purpose (so we don't have
2500 	 * to call ifp->if_resolvemulti() again), and we saved that
2501 	 * value in sa above.  If some nasty deleted the
2502 	 * link-layer address out from underneath us, we can deal because
2503 	 * the address we stored was is not the same as the one which was
2504 	 * in the record for the link-layer address.  (So we don't complain
2505 	 * in that case.)
2506 	 */
2507 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2508 		if (sa_equal(sa, ifma->ifma_addr))
2509 			break;
2510 	if (ifma == NULL)
2511 		return 0;
2512 
2513 	if (ifma->ifma_refcount > 1) {
2514 		ifma->ifma_refcount--;
2515 		return 0;
2516 	}
2517 
2518 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2519 	ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2520 	kfree(ifma->ifma_addr, M_IFMADDR);
2521 	kfree(sa, M_IFMADDR);
2522 	kfree(ifma, M_IFMADDR);
2523 
2524 	return 0;
2525 }
2526 
2527 int
2528 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2529 {
2530 	int error;
2531 
2532 	ifnet_serialize_all(ifp);
2533 	error = if_delmulti_serialized(ifp, sa);
2534 	ifnet_deserialize_all(ifp);
2535 
2536 	return error;
2537 }
2538 
2539 /*
2540  * Delete all multicast group membership for an interface.
2541  * Should be used to quickly flush all multicast filters.
2542  */
2543 void
2544 if_delallmulti_serialized(struct ifnet *ifp)
2545 {
2546 	struct ifmultiaddr *ifma, mark;
2547 	struct sockaddr sa;
2548 
2549 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
2550 
2551 	bzero(&sa, sizeof(sa));
2552 	sa.sa_family = AF_UNSPEC;
2553 	sa.sa_len = sizeof(sa);
2554 
2555 	bzero(&mark, sizeof(mark));
2556 	mark.ifma_addr = &sa;
2557 
2558 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2559 	while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2560 		TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2561 		TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2562 		    ifma_link);
2563 
2564 		if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2565 			continue;
2566 
2567 		if_delmulti_serialized(ifp, ifma->ifma_addr);
2568 	}
2569 	TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2570 }
2571 
2572 
2573 /*
2574  * Set the link layer address on an interface.
2575  *
2576  * At this time we only support certain types of interfaces,
2577  * and we don't allow the length of the address to change.
2578  */
2579 int
2580 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2581 {
2582 	struct sockaddr_dl *sdl;
2583 	struct ifreq ifr;
2584 
2585 	sdl = IF_LLSOCKADDR(ifp);
2586 	if (sdl == NULL)
2587 		return (EINVAL);
2588 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2589 		return (EINVAL);
2590 	switch (ifp->if_type) {
2591 	case IFT_ETHER:			/* these types use struct arpcom */
2592 	case IFT_XETHER:
2593 	case IFT_L2VLAN:
2594 	case IFT_IEEE8023ADLAG:
2595 		bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2596 		bcopy(lladdr, LLADDR(sdl), len);
2597 		break;
2598 	default:
2599 		return (ENODEV);
2600 	}
2601 	/*
2602 	 * If the interface is already up, we need
2603 	 * to re-init it in order to reprogram its
2604 	 * address filter.
2605 	 */
2606 	ifnet_serialize_all(ifp);
2607 	if ((ifp->if_flags & IFF_UP) != 0) {
2608 #ifdef INET
2609 		struct ifaddr_container *ifac;
2610 #endif
2611 
2612 		ifp->if_flags &= ~IFF_UP;
2613 		ifr.ifr_flags = ifp->if_flags;
2614 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
2615 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2616 			      NULL);
2617 		ifp->if_flags |= IFF_UP;
2618 		ifr.ifr_flags = ifp->if_flags;
2619 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
2620 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2621 				 NULL);
2622 #ifdef INET
2623 		/*
2624 		 * Also send gratuitous ARPs to notify other nodes about
2625 		 * the address change.
2626 		 */
2627 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2628 			struct ifaddr *ifa = ifac->ifa;
2629 
2630 			if (ifa->ifa_addr != NULL &&
2631 			    ifa->ifa_addr->sa_family == AF_INET)
2632 				arp_gratuitous(ifp, ifa);
2633 		}
2634 #endif
2635 	}
2636 	ifnet_deserialize_all(ifp);
2637 	return (0);
2638 }
2639 
2640 struct ifmultiaddr *
2641 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2642 {
2643 	struct ifmultiaddr *ifma;
2644 
2645 	/* TODO: need ifnet_serialize_main */
2646 	ifnet_serialize_all(ifp);
2647 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2648 		if (sa_equal(ifma->ifma_addr, sa))
2649 			break;
2650 	ifnet_deserialize_all(ifp);
2651 
2652 	return ifma;
2653 }
2654 
2655 /*
2656  * This function locates the first real ethernet MAC from a network
2657  * card and loads it into node, returning 0 on success or ENOENT if
2658  * no suitable interfaces were found.  It is used by the uuid code to
2659  * generate a unique 6-byte number.
2660  */
2661 int
2662 if_getanyethermac(uint16_t *node, int minlen)
2663 {
2664 	struct ifnet *ifp;
2665 	struct sockaddr_dl *sdl;
2666 
2667 	ifnet_lock();
2668 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2669 		if (ifp->if_type != IFT_ETHER)
2670 			continue;
2671 		sdl = IF_LLSOCKADDR(ifp);
2672 		if (sdl->sdl_alen < minlen)
2673 			continue;
2674 		bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2675 		      minlen);
2676 		ifnet_unlock();
2677 		return(0);
2678 	}
2679 	ifnet_unlock();
2680 	return (ENOENT);
2681 }
2682 
2683 /*
2684  * The name argument must be a pointer to storage which will last as
2685  * long as the interface does.  For physical devices, the result of
2686  * device_get_name(dev) is a good choice and for pseudo-devices a
2687  * static string works well.
2688  */
2689 void
2690 if_initname(struct ifnet *ifp, const char *name, int unit)
2691 {
2692 	ifp->if_dname = name;
2693 	ifp->if_dunit = unit;
2694 	if (unit != IF_DUNIT_NONE)
2695 		ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2696 	else
2697 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2698 }
2699 
2700 int
2701 if_printf(struct ifnet *ifp, const char *fmt, ...)
2702 {
2703 	__va_list ap;
2704 	int retval;
2705 
2706 	retval = kprintf("%s: ", ifp->if_xname);
2707 	__va_start(ap, fmt);
2708 	retval += kvprintf(fmt, ap);
2709 	__va_end(ap);
2710 	return (retval);
2711 }
2712 
2713 struct ifnet *
2714 if_alloc(uint8_t type)
2715 {
2716         struct ifnet *ifp;
2717 	size_t size;
2718 
2719 	/*
2720 	 * XXX temporary hack until arpcom is setup in if_l2com
2721 	 */
2722 	if (type == IFT_ETHER)
2723 		size = sizeof(struct arpcom);
2724 	else
2725 		size = sizeof(struct ifnet);
2726 
2727 	ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2728 
2729 	ifp->if_type = type;
2730 
2731 	if (if_com_alloc[type] != NULL) {
2732 		ifp->if_l2com = if_com_alloc[type](type, ifp);
2733 		if (ifp->if_l2com == NULL) {
2734 			kfree(ifp, M_IFNET);
2735 			return (NULL);
2736 		}
2737 	}
2738 	return (ifp);
2739 }
2740 
2741 void
2742 if_free(struct ifnet *ifp)
2743 {
2744 	kfree(ifp, M_IFNET);
2745 }
2746 
2747 void
2748 ifq_set_classic(struct ifaltq *ifq)
2749 {
2750 	ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2751 	    ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2752 }
2753 
2754 void
2755 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2756     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2757 {
2758 	int q;
2759 
2760 	KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2761 	KASSERT(enqueue != NULL, ("enqueue is not specified"));
2762 	KASSERT(dequeue != NULL, ("dequeue is not specified"));
2763 	KASSERT(request != NULL, ("request is not specified"));
2764 
2765 	ifq->altq_mapsubq = mapsubq;
2766 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2767 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2768 
2769 		ifsq->ifsq_enqueue = enqueue;
2770 		ifsq->ifsq_dequeue = dequeue;
2771 		ifsq->ifsq_request = request;
2772 	}
2773 }
2774 
2775 static void
2776 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2777 {
2778 
2779 	classq_add(&ifsq->ifsq_norm, m);
2780 	ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2781 }
2782 
2783 static void
2784 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2785 {
2786 
2787 	classq_add(&ifsq->ifsq_prio, m);
2788 	ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2789 	ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2790 }
2791 
2792 static struct mbuf *
2793 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2794 {
2795 	struct mbuf *m;
2796 
2797 	m = classq_get(&ifsq->ifsq_norm);
2798 	if (m != NULL)
2799 		ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2800 	return (m);
2801 }
2802 
2803 static struct mbuf *
2804 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2805 {
2806 	struct mbuf *m;
2807 
2808 	m = classq_get(&ifsq->ifsq_prio);
2809 	if (m != NULL) {
2810 		ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2811 		ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2812 	}
2813 	return (m);
2814 }
2815 
2816 int
2817 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2818     struct altq_pktattr *pa __unused)
2819 {
2820 
2821 	M_ASSERTPKTHDR(m);
2822 again:
2823 	if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2824 	    ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2825 		struct mbuf *m_drop;
2826 
2827 		if (m->m_flags & M_PRIO) {
2828 			m_drop = NULL;
2829 			if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) &&
2830 			    ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) {
2831 				/* Try dropping some from normal queue. */
2832 				m_drop = ifsq_norm_dequeue(ifsq);
2833 			}
2834 			if (m_drop == NULL)
2835 				m_drop = ifsq_prio_dequeue(ifsq);
2836 		} else {
2837 			m_drop = ifsq_norm_dequeue(ifsq);
2838 		}
2839 		if (m_drop != NULL) {
2840 			IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1);
2841 			m_freem(m_drop);
2842 			goto again;
2843 		}
2844 		/*
2845 		 * No old packets could be dropped!
2846 		 * NOTE: Caller increases oqdrops.
2847 		 */
2848 		m_freem(m);
2849 		return (ENOBUFS);
2850 	} else {
2851 		if (m->m_flags & M_PRIO)
2852 			ifsq_prio_enqueue(ifsq, m);
2853 		else
2854 			ifsq_norm_enqueue(ifsq, m);
2855 		return (0);
2856 	}
2857 }
2858 
2859 struct mbuf *
2860 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2861 {
2862 	struct mbuf *m;
2863 
2864 	switch (op) {
2865 	case ALTDQ_POLL:
2866 		m = classq_head(&ifsq->ifsq_prio);
2867 		if (m == NULL)
2868 			m = classq_head(&ifsq->ifsq_norm);
2869 		break;
2870 
2871 	case ALTDQ_REMOVE:
2872 		m = ifsq_prio_dequeue(ifsq);
2873 		if (m == NULL)
2874 			m = ifsq_norm_dequeue(ifsq);
2875 		break;
2876 
2877 	default:
2878 		panic("unsupported ALTQ dequeue op: %d", op);
2879 	}
2880 	return m;
2881 }
2882 
2883 int
2884 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2885 {
2886 	switch (req) {
2887 	case ALTRQ_PURGE:
2888 		for (;;) {
2889 			struct mbuf *m;
2890 
2891 			m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2892 			if (m == NULL)
2893 				break;
2894 			m_freem(m);
2895 		}
2896 		break;
2897 
2898 	default:
2899 		panic("unsupported ALTQ request: %d", req);
2900 	}
2901 	return 0;
2902 }
2903 
2904 static void
2905 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2906 {
2907 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
2908 	int running = 0, need_sched;
2909 
2910 	/*
2911 	 * Try to do direct ifnet.if_start on the subqueue first, if there is
2912 	 * contention on the subqueue hardware serializer, ifnet.if_start on
2913 	 * the subqueue will be scheduled on the subqueue owner CPU.
2914 	 */
2915 	if (!ifsq_tryserialize_hw(ifsq)) {
2916 		/*
2917 		 * Subqueue hardware serializer contention happened,
2918 		 * ifnet.if_start on the subqueue is scheduled on
2919 		 * the subqueue owner CPU, and we keep going.
2920 		 */
2921 		ifsq_ifstart_schedule(ifsq, 1);
2922 		return;
2923 	}
2924 
2925 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2926 		ifp->if_start(ifp, ifsq);
2927 		if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2928 			running = 1;
2929 	}
2930 	need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2931 
2932 	ifsq_deserialize_hw(ifsq);
2933 
2934 	if (need_sched) {
2935 		/*
2936 		 * More data need to be transmitted, ifnet.if_start on the
2937 		 * subqueue is scheduled on the subqueue owner CPU, and we
2938 		 * keep going.
2939 		 * NOTE: ifnet.if_start subqueue interlock is not released.
2940 		 */
2941 		ifsq_ifstart_schedule(ifsq, force_sched);
2942 	}
2943 }
2944 
2945 /*
2946  * Subqeue packets staging mechanism:
2947  *
2948  * The packets enqueued into the subqueue are staged to a certain amount
2949  * before the ifnet.if_start on the subqueue is called.  In this way, the
2950  * driver could avoid writing to hardware registers upon every packet,
2951  * instead, hardware registers could be written when certain amount of
2952  * packets are put onto hardware TX ring.  The measurement on several modern
2953  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
2954  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
2955  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
2956  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
2957  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
2958  *
2959  * Subqueue packets staging is performed for two entry points into drivers'
2960  * transmission function:
2961  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
2962  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
2963  *
2964  * Subqueue packets staging will be stopped upon any of the following
2965  * conditions:
2966  * - If the count of packets enqueued on the current CPU is great than or
2967  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2968  * - If the total length of packets enqueued on the current CPU is great
2969  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
2970  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
2971  *   is usually less than hardware's MTU.
2972  * - ifsq_ifstart_schedule() is not pending on the current CPU and
2973  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
2974  *   released.
2975  * - The if_start_rollup(), which is registered as low priority netisr
2976  *   rollup function, is called; probably because no more work is pending
2977  *   for netisr.
2978  *
2979  * NOTE:
2980  * Currently subqueue packet staging is only performed in netisr threads.
2981  */
2982 int
2983 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2984 {
2985 	struct ifaltq *ifq = &ifp->if_snd;
2986 	struct ifaltq_subque *ifsq;
2987 	int error, start = 0, len, mcast = 0, avoid_start = 0;
2988 	struct ifsubq_stage_head *head = NULL;
2989 	struct ifsubq_stage *stage = NULL;
2990 	struct globaldata *gd = mycpu;
2991 	struct thread *td = gd->gd_curthread;
2992 
2993 	crit_enter_quick(td);
2994 
2995 	ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
2996 	ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
2997 
2998 	len = m->m_pkthdr.len;
2999 	if (m->m_flags & M_MCAST)
3000 		mcast = 1;
3001 
3002 	if (td->td_type == TD_TYPE_NETISR) {
3003 		head = &ifsubq_stage_heads[mycpuid];
3004 		stage = ifsq_get_stage(ifsq, mycpuid);
3005 
3006 		stage->stg_cnt++;
3007 		stage->stg_len += len;
3008 		if (stage->stg_cnt < ifsq_stage_cntmax &&
3009 		    stage->stg_len < (ifp->if_mtu - max_protohdr))
3010 			avoid_start = 1;
3011 	}
3012 
3013 	ALTQ_SQ_LOCK(ifsq);
3014 	error = ifsq_enqueue_locked(ifsq, m, pa);
3015 	if (error) {
3016 		IFNET_STAT_INC(ifp, oqdrops, 1);
3017 		if (!ifsq_data_ready(ifsq)) {
3018 			ALTQ_SQ_UNLOCK(ifsq);
3019 			crit_exit_quick(td);
3020 			return error;
3021 		}
3022 		avoid_start = 0;
3023 	}
3024 	if (!ifsq_is_started(ifsq)) {
3025 		if (avoid_start) {
3026 			ALTQ_SQ_UNLOCK(ifsq);
3027 
3028 			KKASSERT(!error);
3029 			if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3030 				ifsq_stage_insert(head, stage);
3031 
3032 			IFNET_STAT_INC(ifp, obytes, len);
3033 			if (mcast)
3034 				IFNET_STAT_INC(ifp, omcasts, 1);
3035 			crit_exit_quick(td);
3036 			return error;
3037 		}
3038 
3039 		/*
3040 		 * Hold the subqueue interlock of ifnet.if_start
3041 		 */
3042 		ifsq_set_started(ifsq);
3043 		start = 1;
3044 	}
3045 	ALTQ_SQ_UNLOCK(ifsq);
3046 
3047 	if (!error) {
3048 		IFNET_STAT_INC(ifp, obytes, len);
3049 		if (mcast)
3050 			IFNET_STAT_INC(ifp, omcasts, 1);
3051 	}
3052 
3053 	if (stage != NULL) {
3054 		if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3055 			KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3056 			if (!avoid_start) {
3057 				ifsq_stage_remove(head, stage);
3058 				ifsq_ifstart_schedule(ifsq, 1);
3059 			}
3060 			crit_exit_quick(td);
3061 			return error;
3062 		}
3063 
3064 		if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3065 			ifsq_stage_remove(head, stage);
3066 		} else {
3067 			stage->stg_cnt = 0;
3068 			stage->stg_len = 0;
3069 		}
3070 	}
3071 
3072 	if (!start) {
3073 		crit_exit_quick(td);
3074 		return error;
3075 	}
3076 
3077 	ifsq_ifstart_try(ifsq, 0);
3078 
3079 	crit_exit_quick(td);
3080 	return error;
3081 }
3082 
3083 void *
3084 ifa_create(int size)
3085 {
3086 	struct ifaddr *ifa;
3087 	int i;
3088 
3089 	KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3090 
3091 	ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3092 	ifa->ifa_containers =
3093 	    kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
3094 	        M_IFADDR, M_INTWAIT | M_ZERO);
3095 
3096 	ifa->ifa_ncnt = ncpus;
3097 	for (i = 0; i < ncpus; ++i) {
3098 		struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3099 
3100 		ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3101 		ifac->ifa = ifa;
3102 		ifac->ifa_refcnt = 1;
3103 	}
3104 #ifdef IFADDR_DEBUG
3105 	kprintf("alloc ifa %p %d\n", ifa, size);
3106 #endif
3107 	return ifa;
3108 }
3109 
3110 void
3111 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3112 {
3113 	struct ifaddr *ifa = ifac->ifa;
3114 
3115 	KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3116 	KKASSERT(ifac->ifa_refcnt == 0);
3117 	KASSERT(ifac->ifa_listmask == 0,
3118 		("ifa is still on %#x lists", ifac->ifa_listmask));
3119 
3120 	ifac->ifa_magic = IFA_CONTAINER_DEAD;
3121 
3122 #ifdef IFADDR_DEBUG_VERBOSE
3123 	kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3124 #endif
3125 
3126 	KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3127 		("invalid # of ifac, %d", ifa->ifa_ncnt));
3128 	if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3129 #ifdef IFADDR_DEBUG
3130 		kprintf("free ifa %p\n", ifa);
3131 #endif
3132 		kfree(ifa->ifa_containers, M_IFADDR);
3133 		kfree(ifa, M_IFADDR);
3134 	}
3135 }
3136 
3137 static void
3138 ifa_iflink_dispatch(netmsg_t nmsg)
3139 {
3140 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3141 	struct ifaddr *ifa = msg->ifa;
3142 	struct ifnet *ifp = msg->ifp;
3143 	int cpu = mycpuid;
3144 	struct ifaddr_container *ifac;
3145 
3146 	crit_enter();
3147 
3148 	ifac = &ifa->ifa_containers[cpu];
3149 	ASSERT_IFAC_VALID(ifac);
3150 	KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3151 		("ifaddr is on if_addrheads"));
3152 
3153 	ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3154 	if (msg->tail)
3155 		TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3156 	else
3157 		TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3158 
3159 	crit_exit();
3160 
3161 	netisr_forwardmsg(&nmsg->base, cpu + 1);
3162 }
3163 
3164 void
3165 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3166 {
3167 	struct netmsg_ifaddr msg;
3168 
3169 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3170 		    0, ifa_iflink_dispatch);
3171 	msg.ifa = ifa;
3172 	msg.ifp = ifp;
3173 	msg.tail = tail;
3174 
3175 	netisr_domsg(&msg.base, 0);
3176 }
3177 
3178 static void
3179 ifa_ifunlink_dispatch(netmsg_t nmsg)
3180 {
3181 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3182 	struct ifaddr *ifa = msg->ifa;
3183 	struct ifnet *ifp = msg->ifp;
3184 	int cpu = mycpuid;
3185 	struct ifaddr_container *ifac;
3186 
3187 	crit_enter();
3188 
3189 	ifac = &ifa->ifa_containers[cpu];
3190 	ASSERT_IFAC_VALID(ifac);
3191 	KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3192 		("ifaddr is not on if_addrhead"));
3193 
3194 	TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3195 	ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3196 
3197 	crit_exit();
3198 
3199 	netisr_forwardmsg(&nmsg->base, cpu + 1);
3200 }
3201 
3202 void
3203 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3204 {
3205 	struct netmsg_ifaddr msg;
3206 
3207 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3208 		    0, ifa_ifunlink_dispatch);
3209 	msg.ifa = ifa;
3210 	msg.ifp = ifp;
3211 
3212 	netisr_domsg(&msg.base, 0);
3213 }
3214 
3215 static void
3216 ifa_destroy_dispatch(netmsg_t nmsg)
3217 {
3218 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3219 
3220 	IFAFREE(msg->ifa);
3221 	netisr_forwardmsg(&nmsg->base, mycpuid + 1);
3222 }
3223 
3224 void
3225 ifa_destroy(struct ifaddr *ifa)
3226 {
3227 	struct netmsg_ifaddr msg;
3228 
3229 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3230 		    0, ifa_destroy_dispatch);
3231 	msg.ifa = ifa;
3232 
3233 	netisr_domsg(&msg.base, 0);
3234 }
3235 
3236 static void
3237 if_start_rollup(void)
3238 {
3239 	struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3240 	struct ifsubq_stage *stage;
3241 
3242 	crit_enter();
3243 
3244 	while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3245 		struct ifaltq_subque *ifsq = stage->stg_subq;
3246 		int is_sched = 0;
3247 
3248 		if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3249 			is_sched = 1;
3250 		ifsq_stage_remove(head, stage);
3251 
3252 		if (is_sched) {
3253 			ifsq_ifstart_schedule(ifsq, 1);
3254 		} else {
3255 			int start = 0;
3256 
3257 			ALTQ_SQ_LOCK(ifsq);
3258 			if (!ifsq_is_started(ifsq)) {
3259 				/*
3260 				 * Hold the subqueue interlock of
3261 				 * ifnet.if_start
3262 				 */
3263 				ifsq_set_started(ifsq);
3264 				start = 1;
3265 			}
3266 			ALTQ_SQ_UNLOCK(ifsq);
3267 
3268 			if (start)
3269 				ifsq_ifstart_try(ifsq, 1);
3270 		}
3271 		KKASSERT((stage->stg_flags &
3272 		    (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3273 	}
3274 
3275 	crit_exit();
3276 }
3277 
3278 static void
3279 ifnetinit(void *dummy __unused)
3280 {
3281 	int i;
3282 
3283 	for (i = 0; i < ncpus; ++i)
3284 		TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3285 	netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3286 }
3287 
3288 void
3289 if_register_com_alloc(u_char type,
3290     if_com_alloc_t *a, if_com_free_t *f)
3291 {
3292 
3293         KASSERT(if_com_alloc[type] == NULL,
3294             ("if_register_com_alloc: %d already registered", type));
3295         KASSERT(if_com_free[type] == NULL,
3296             ("if_register_com_alloc: %d free already registered", type));
3297 
3298         if_com_alloc[type] = a;
3299         if_com_free[type] = f;
3300 }
3301 
3302 void
3303 if_deregister_com_alloc(u_char type)
3304 {
3305 
3306         KASSERT(if_com_alloc[type] != NULL,
3307             ("if_deregister_com_alloc: %d not registered", type));
3308         KASSERT(if_com_free[type] != NULL,
3309             ("if_deregister_com_alloc: %d free not registered", type));
3310         if_com_alloc[type] = NULL;
3311         if_com_free[type] = NULL;
3312 }
3313 
3314 void
3315 ifq_set_maxlen(struct ifaltq *ifq, int len)
3316 {
3317 	ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3318 }
3319 
3320 int
3321 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3322 {
3323 	return ALTQ_SUBQ_INDEX_DEFAULT;
3324 }
3325 
3326 int
3327 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3328 {
3329 
3330 	return (cpuid & ifq->altq_subq_mappriv);
3331 }
3332 
3333 int
3334 ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid)
3335 {
3336 
3337 	return (cpuid % ifq->altq_subq_mappriv);
3338 }
3339 
3340 static void
3341 ifsq_watchdog(void *arg)
3342 {
3343 	struct ifsubq_watchdog *wd = arg;
3344 	struct ifnet *ifp;
3345 
3346 	if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3347 		goto done;
3348 
3349 	ifp = ifsq_get_ifp(wd->wd_subq);
3350 	if (ifnet_tryserialize_all(ifp)) {
3351 		wd->wd_watchdog(wd->wd_subq);
3352 		ifnet_deserialize_all(ifp);
3353 	} else {
3354 		/* try again next timeout */
3355 		wd->wd_timer = 1;
3356 	}
3357 done:
3358 	ifsq_watchdog_reset(wd);
3359 }
3360 
3361 static void
3362 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3363 {
3364 	callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3365 	    ifsq_get_cpuid(wd->wd_subq));
3366 }
3367 
3368 void
3369 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3370     ifsq_watchdog_t watchdog)
3371 {
3372 	callout_init_mp(&wd->wd_callout);
3373 	wd->wd_timer = 0;
3374 	wd->wd_subq = ifsq;
3375 	wd->wd_watchdog = watchdog;
3376 }
3377 
3378 void
3379 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3380 {
3381 	wd->wd_timer = 0;
3382 	ifsq_watchdog_reset(wd);
3383 }
3384 
3385 void
3386 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3387 {
3388 	wd->wd_timer = 0;
3389 	callout_stop(&wd->wd_callout);
3390 }
3391 
3392 void
3393 ifnet_lock(void)
3394 {
3395 	KASSERT(curthread->td_type != TD_TYPE_NETISR,
3396 	    ("try holding ifnet lock in netisr"));
3397 	mtx_lock(&ifnet_mtx);
3398 }
3399 
3400 void
3401 ifnet_unlock(void)
3402 {
3403 	KASSERT(curthread->td_type != TD_TYPE_NETISR,
3404 	    ("try holding ifnet lock in netisr"));
3405 	mtx_unlock(&ifnet_mtx);
3406 }
3407 
3408 static struct ifnet_array *
3409 ifnet_array_alloc(int count)
3410 {
3411 	struct ifnet_array *arr;
3412 
3413 	arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3414 	    M_IFNET, M_WAITOK);
3415 	arr->ifnet_count = count;
3416 
3417 	return arr;
3418 }
3419 
3420 static void
3421 ifnet_array_free(struct ifnet_array *arr)
3422 {
3423 	if (arr == &ifnet_array0)
3424 		return;
3425 	kfree(arr, M_IFNET);
3426 }
3427 
3428 static struct ifnet_array *
3429 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3430 {
3431 	struct ifnet_array *arr;
3432 	int count, i;
3433 
3434 	KASSERT(old_arr->ifnet_count >= 0,
3435 	    ("invalid ifnet array count %d", old_arr->ifnet_count));
3436 	count = old_arr->ifnet_count + 1;
3437 	arr = ifnet_array_alloc(count);
3438 
3439 	/*
3440 	 * Save the old ifnet array and append this ifp to the end of
3441 	 * the new ifnet array.
3442 	 */
3443 	for (i = 0; i < old_arr->ifnet_count; ++i) {
3444 		KASSERT(old_arr->ifnet_arr[i] != ifp,
3445 		    ("%s is already in ifnet array", ifp->if_xname));
3446 		arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3447 	}
3448 	KASSERT(i == count - 1,
3449 	    ("add %s, ifnet array index mismatch, should be %d, but got %d",
3450 	     ifp->if_xname, count - 1, i));
3451 	arr->ifnet_arr[i] = ifp;
3452 
3453 	return arr;
3454 }
3455 
3456 static struct ifnet_array *
3457 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3458 {
3459 	struct ifnet_array *arr;
3460 	int count, i, idx, found = 0;
3461 
3462 	KASSERT(old_arr->ifnet_count > 0,
3463 	    ("invalid ifnet array count %d", old_arr->ifnet_count));
3464 	count = old_arr->ifnet_count - 1;
3465 	arr = ifnet_array_alloc(count);
3466 
3467 	/*
3468 	 * Save the old ifnet array, but skip this ifp.
3469 	 */
3470 	idx = 0;
3471 	for (i = 0; i < old_arr->ifnet_count; ++i) {
3472 		if (old_arr->ifnet_arr[i] == ifp) {
3473 			KASSERT(!found,
3474 			    ("dup %s is in ifnet array", ifp->if_xname));
3475 			found = 1;
3476 			continue;
3477 		}
3478 		KASSERT(idx < count,
3479 		    ("invalid ifnet array index %d, count %d", idx, count));
3480 		arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3481 		++idx;
3482 	}
3483 	KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3484 	KASSERT(idx == count,
3485 	    ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3486 	     ifp->if_xname, count, idx));
3487 
3488 	return arr;
3489 }
3490 
3491 const struct ifnet_array *
3492 ifnet_array_get(void)
3493 {
3494 	const struct ifnet_array *ret;
3495 
3496 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3497 	ret = ifnet_array;
3498 	/* Make sure 'ret' is really used. */
3499 	cpu_ccfence();
3500 	return (ret);
3501 }
3502 
3503 int
3504 ifnet_array_isempty(void)
3505 {
3506 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3507 	if (ifnet_array->ifnet_count == 0)
3508 		return 1;
3509 	else
3510 		return 0;
3511 }
3512 
3513 void
3514 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3515 {
3516 	struct ifaddr *ifa;
3517 
3518 	memset(mark, 0, sizeof(*mark));
3519 	ifa = &mark->ifa;
3520 
3521 	mark->ifac.ifa = ifa;
3522 
3523 	ifa->ifa_addr = &mark->addr;
3524 	ifa->ifa_dstaddr = &mark->dstaddr;
3525 	ifa->ifa_netmask = &mark->netmask;
3526 	ifa->ifa_ifp = ifp;
3527 }
3528 
3529 static int
3530 if_ringcnt_fixup(int ring_cnt, int ring_cntmax)
3531 {
3532 
3533 	KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax));
3534 	if (ring_cnt == 1 || ring_cntmax == 1 || netisr_ncpus == 1)
3535 		return (1);
3536 
3537 	if (ring_cnt <= 0 || ring_cnt > ring_cntmax)
3538 		ring_cnt = ring_cntmax;
3539 	if (ring_cnt > netisr_ncpus)
3540 		ring_cnt = netisr_ncpus;
3541 	return (ring_cnt);
3542 }
3543 
3544 static void
3545 if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid)
3546 {
3547 	int i, offset;
3548 
3549 	KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid));
3550 	rm->rm_grid = grid;
3551 
3552 	offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus;
3553 	for (i = 0; i < rm->rm_cnt; ++i) {
3554 		rm->rm_cpumap[i] = offset + i;
3555 		KASSERT(rm->rm_cpumap[i] < netisr_ncpus,
3556 		    ("invalid cpumap[%d] = %d, offset %d", i,
3557 		     rm->rm_cpumap[i], offset));
3558 	}
3559 }
3560 
3561 struct if_ringmap *
3562 if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax)
3563 {
3564 	struct if_ringmap *rm;
3565 	int i, grid = 0;
3566 
3567 	ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3568 	rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]),
3569 	    M_DEVBUF, M_WAITOK | M_ZERO);
3570 
3571 	rm->rm_cnt = ring_cnt;
3572 	for (i = 0; i < netisr_ncpus; ++i) {
3573 		if (netisr_ncpus % (i + 1) != 0)
3574 			continue;
3575 
3576 		if (rm->rm_cnt > netisr_ncpus / (i + 2)) {
3577 			grid = netisr_ncpus / (i + 1);
3578 			if (rm->rm_cnt > grid)
3579 				rm->rm_cnt = grid;
3580 			break;
3581 		}
3582 	}
3583 	if_ringmap_set_grid(dev, rm, grid);
3584 
3585 	return (rm);
3586 }
3587 
3588 void
3589 if_ringmap_free(struct if_ringmap *rm)
3590 {
3591 
3592 	kfree(rm, M_DEVBUF);
3593 }
3594 
3595 /*
3596  * Align the two ringmaps.
3597  *
3598  * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3599  *
3600  * Before:
3601  *
3602  * CPU      0  1  2  3   4  5  6  7
3603  * NIC_RX               n0 n1 n2 n3
3604  * NIC_TX        N0 N1
3605  *
3606  * After:
3607  *
3608  * CPU      0  1  2  3   4  5  6  7
3609  * NIC_RX               n0 n1 n2 n3
3610  * NIC_TX               N0 N1
3611  */
3612 void
3613 if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3614 {
3615 
3616 	if (rm0->rm_grid > rm1->rm_grid)
3617 		if_ringmap_set_grid(dev, rm1, rm0->rm_grid);
3618 	else if (rm0->rm_grid < rm1->rm_grid)
3619 		if_ringmap_set_grid(dev, rm0, rm1->rm_grid);
3620 }
3621 
3622 void
3623 if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3624 {
3625 	int subset_grid, cnt, divisor, mod, offset, i;
3626 	struct if_ringmap *subset_rm, *rm;
3627 	int old_rm0_grid, old_rm1_grid;
3628 
3629 	if (rm0->rm_grid == rm1->rm_grid)
3630 		return;
3631 
3632 	/* Save grid for later use */
3633 	old_rm0_grid = rm0->rm_grid;
3634 	old_rm1_grid = rm1->rm_grid;
3635 
3636 	if_ringmap_align(dev, rm0, rm1);
3637 
3638 	/*
3639 	 * Re-shuffle rings to get more even distribution.
3640 	 *
3641 	 * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3642 	 *
3643 	 * CPU       0  1  2  3   4  5  6  7   8  9 10 11
3644 	 *
3645 	 * NIC_RX   a0 a1 a2 a3  b0 b1 b2 b3  c0 c1 c2 c3
3646 	 * NIC_TX   A0 A1        B0 B1        C0 C1
3647 	 *
3648 	 * NIC_RX   d0 d1 d2 d3  e0 e1 e2 e3  f0 f1 f2 f3
3649 	 * NIC_TX         D0 D1        E0 E1        F0 F1
3650 	 */
3651 
3652 	if (rm0->rm_cnt >= (2 * old_rm1_grid)) {
3653 		cnt = rm0->rm_cnt;
3654 		subset_grid = old_rm1_grid;
3655 		subset_rm = rm1;
3656 		rm = rm0;
3657 	} else if (rm1->rm_cnt > (2 * old_rm0_grid)) {
3658 		cnt = rm1->rm_cnt;
3659 		subset_grid = old_rm0_grid;
3660 		subset_rm = rm0;
3661 		rm = rm1;
3662 	} else {
3663 		/* No space to shuffle. */
3664 		return;
3665 	}
3666 
3667 	mod = cnt / subset_grid;
3668 	KKASSERT(mod >= 2);
3669 	divisor = netisr_ncpus / rm->rm_grid;
3670 	offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid;
3671 
3672 	for (i = 0; i < subset_rm->rm_cnt; ++i) {
3673 		subset_rm->rm_cpumap[i] += offset;
3674 		KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus,
3675 		    ("match: invalid cpumap[%d] = %d, offset %d",
3676 		     i, subset_rm->rm_cpumap[i], offset));
3677 	}
3678 #ifdef INVARIANTS
3679 	for (i = 0; i < subset_rm->rm_cnt; ++i) {
3680 		int j;
3681 
3682 		for (j = 0; j < rm->rm_cnt; ++j) {
3683 			if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i])
3684 				break;
3685 		}
3686 		KASSERT(j < rm->rm_cnt,
3687 		    ("subset cpumap[%d] = %d not found in superset",
3688 		     i, subset_rm->rm_cpumap[i]));
3689 	}
3690 #endif
3691 }
3692 
3693 int
3694 if_ringmap_count(const struct if_ringmap *rm)
3695 {
3696 
3697 	return (rm->rm_cnt);
3698 }
3699 
3700 int
3701 if_ringmap_cpumap(const struct if_ringmap *rm, int ring)
3702 {
3703 
3704 	KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring));
3705 	return (rm->rm_cpumap[ring]);
3706 }
3707 
3708 void
3709 if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent)
3710 {
3711 	int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy;
3712 
3713 	KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0,
3714 	    ("invalid redirect table entries %d", table_nent));
3715 
3716 	grid_idx = 0;
3717 	for (i = 0; i < NETISR_CPUMAX; ++i) {
3718 		table[i] = grid_idx++ % rm->rm_cnt;
3719 
3720 		if (grid_idx == rm->rm_grid)
3721 			grid_idx = 0;
3722 	}
3723 
3724 	/*
3725 	 * Make the ring distributed more evenly for the remainder
3726 	 * of each grid.
3727 	 *
3728 	 * e.g. 12 netisrs, rm contains 8 rings.
3729 	 *
3730 	 * Redirect table before:
3731 	 *
3732 	 *  0  1  2  3  4  5  6  7  0  1  2  3  0  1  2  3
3733 	 *  4  5  6  7  0  1  2  3  0  1  2  3  4  5  6  7
3734 	 *  0  1  2  3  0  1  2  3  4  5  6  7  0  1  2  3
3735 	 *  ....
3736 	 *
3737 	 * Redirect table after being patched (pX, patched entries):
3738 	 *
3739 	 *  0  1  2  3  4  5  6  7 p0 p1 p2 p3  0  1  2  3
3740 	 *  4  5  6  7 p4 p5 p6 p7  0  1  2  3  4  5  6  7
3741 	 * p0 p1 p2 p3  0  1  2  3  4  5  6  7 p4 p5 p6 p7
3742 	 *  ....
3743 	 */
3744 	patch_cnt = rm->rm_grid % rm->rm_cnt;
3745 	if (patch_cnt == 0)
3746 		goto done;
3747 	patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt);
3748 
3749 	grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid;
3750 	grid_idx = 0;
3751 	for (i = 0; i < grid_cnt; ++i) {
3752 		int j;
3753 
3754 		for (j = 0; j < patch_cnt; ++j) {
3755 			int fix_idx;
3756 
3757 			fix_idx = (i * rm->rm_grid) + patch_off + j;
3758 			if (fix_idx >= NETISR_CPUMAX)
3759 				goto done;
3760 			table[fix_idx] = grid_idx++ % rm->rm_cnt;
3761 		}
3762 	}
3763 done:
3764 	/*
3765 	 * If the device supports larger redirect table, duplicate
3766 	 * the first NETISR_CPUMAX entries to the rest of the table,
3767 	 * so that it matches upper layer's expectation:
3768 	 * (hash & NETISR_CPUMASK) % netisr_ncpus
3769 	 */
3770 	ncopy = table_nent / NETISR_CPUMAX;
3771 	for (i = 1; i < ncopy; ++i) {
3772 		memcpy(&table[i * NETISR_CPUMAX], table,
3773 		    NETISR_CPUMAX * sizeof(table[0]));
3774 	}
3775 	if (if_ringmap_dumprdr) {
3776 		for (i = 0; i < table_nent; ++i) {
3777 			if (i != 0 && i % 16 == 0)
3778 				kprintf("\n");
3779 			kprintf("%03d ", table[i]);
3780 		}
3781 		kprintf("\n");
3782 	}
3783 }
3784 
3785 int
3786 if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS)
3787 {
3788 	struct if_ringmap *rm = arg1;
3789 	int i, error = 0;
3790 
3791 	for (i = 0; i < rm->rm_cnt; ++i) {
3792 		int cpu = rm->rm_cpumap[i];
3793 
3794 		error = SYSCTL_OUT(req, &cpu, sizeof(cpu));
3795 		if (error)
3796 			break;
3797 	}
3798 	return (error);
3799 }
3800 
3801 int
3802 if_ring_count2(int ring_cnt, int ring_cntmax)
3803 {
3804 
3805 	ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3806 	return (1 << (fls(ring_cnt) - 1));
3807 }
3808