xref: /dflybsd-src/sys/net/if.c (revision 2efb75f3055c1746efc358d68dbc2bf526faaf61)
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/lock.h>
51 #include <sys/sockio.h>
52 #include <sys/syslog.h>
53 #include <sys/sysctl.h>
54 #include <sys/domain.h>
55 #include <sys/thread.h>
56 #include <sys/serialize.h>
57 #include <sys/bus.h>
58 #include <sys/jail.h>
59 
60 #include <sys/thread2.h>
61 #include <sys/msgport2.h>
62 #include <sys/mutex2.h>
63 
64 #include <net/if.h>
65 #include <net/if_arp.h>
66 #include <net/if_dl.h>
67 #include <net/if_types.h>
68 #include <net/if_var.h>
69 #include <net/if_ringmap.h>
70 #include <net/ifq_var.h>
71 #include <net/radix.h>
72 #include <net/route.h>
73 #include <net/if_clone.h>
74 #include <net/netisr2.h>
75 #include <net/netmsg2.h>
76 
77 #include <machine/atomic.h>
78 #include <machine/stdarg.h>
79 #include <machine/smp.h>
80 
81 #if defined(INET) || defined(INET6)
82 #include <netinet/in.h>
83 #include <netinet/in_var.h>
84 #include <netinet/if_ether.h>
85 #ifdef INET6
86 #include <netinet6/in6_var.h>
87 #include <netinet6/in6_ifattach.h>
88 #endif /* INET6 */
89 #endif /* INET || INET6 */
90 
91 struct netmsg_ifaddr {
92 	struct netmsg_base base;
93 	struct ifaddr	*ifa;
94 	struct ifnet	*ifp;
95 	int		tail;
96 };
97 
98 struct ifsubq_stage_head {
99 	TAILQ_HEAD(, ifsubq_stage)	stg_head;
100 } __cachealign;
101 
102 struct if_ringmap {
103 	int		rm_cnt;
104 	int		rm_grid;
105 	int		rm_cpumap[];
106 };
107 
108 #define RINGMAP_FLAG_NONE		0x0
109 #define RINGMAP_FLAG_POWEROF2		0x1
110 
111 /*
112  * System initialization
113  */
114 static void	if_attachdomain(void *);
115 static void	if_attachdomain1(struct ifnet *);
116 static int	ifconf(u_long, caddr_t, struct ucred *);
117 static void	ifinit(void *);
118 static void	ifnetinit(void *);
119 static void	if_slowtimo(void *);
120 static void	link_rtrequest(int, struct rtentry *);
121 static int	if_rtdel(struct radix_node *, void *);
122 static void	if_slowtimo_dispatch(netmsg_t);
123 
124 /* Helper functions */
125 static void	ifsq_watchdog_reset(struct ifsubq_watchdog *);
126 static int	if_delmulti_serialized(struct ifnet *, struct sockaddr *);
127 static struct ifnet_array *ifnet_array_alloc(int);
128 static void	ifnet_array_free(struct ifnet_array *);
129 static struct ifnet_array *ifnet_array_add(struct ifnet *,
130 		    const struct ifnet_array *);
131 static struct ifnet_array *ifnet_array_del(struct ifnet *,
132 		    const struct ifnet_array *);
133 static struct ifg_group *if_creategroup(const char *);
134 static int	if_destroygroup(struct ifg_group *);
135 static int	if_delgroup_locked(struct ifnet *, const char *);
136 static int	if_getgroups(struct ifgroupreq *, struct ifnet *);
137 static int	if_getgroupmembers(struct ifgroupreq *);
138 
139 #ifdef INET6
140 /*
141  * XXX: declare here to avoid to include many inet6 related files..
142  * should be more generalized?
143  */
144 extern void	nd6_setmtu(struct ifnet *);
145 #endif
146 
147 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
148 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
149 SYSCTL_NODE(_net_link, OID_AUTO, ringmap, CTLFLAG_RW, 0, "link ringmap");
150 
151 static int ifsq_stage_cntmax = 16;
152 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
153 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
154     &ifsq_stage_cntmax, 0, "ifq staging packet count max");
155 
156 static int if_stats_compat = 0;
157 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
158     &if_stats_compat, 0, "Compat the old ifnet stats");
159 
160 static int if_ringmap_dumprdr = 0;
161 SYSCTL_INT(_net_link_ringmap, OID_AUTO, dump_rdr, CTLFLAG_RW,
162     &if_ringmap_dumprdr, 0, "dump redirect table");
163 
164 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL);
165 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, ifnetinit, NULL);
166 
167 static if_com_alloc_t *if_com_alloc[256];
168 static if_com_free_t *if_com_free[256];
169 
170 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
171 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
172 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
173 
174 int			ifqmaxlen = IFQ_MAXLEN;
175 struct ifnethead	ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
176 struct ifgrouphead	ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
177 static struct lock	ifgroup_lock;
178 
179 static struct ifnet_array	ifnet_array0;
180 static struct ifnet_array	*ifnet_array = &ifnet_array0;
181 
182 static struct callout		if_slowtimo_timer;
183 static struct netmsg_base	if_slowtimo_netmsg;
184 
185 int			if_index = 0;
186 struct ifnet		**ifindex2ifnet = NULL;
187 static struct mtx	ifnet_mtx = MTX_INITIALIZER("ifnet");
188 
189 static struct ifsubq_stage_head	ifsubq_stage_heads[MAXCPU];
190 
191 #ifdef notyet
192 #define IFQ_KTR_STRING		"ifq=%p"
193 #define IFQ_KTR_ARGS		struct ifaltq *ifq
194 #ifndef KTR_IFQ
195 #define KTR_IFQ			KTR_ALL
196 #endif
197 KTR_INFO_MASTER(ifq);
198 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
199 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
200 #define logifq(name, arg)	KTR_LOG(ifq_ ## name, arg)
201 
202 #define IF_START_KTR_STRING	"ifp=%p"
203 #define IF_START_KTR_ARGS	struct ifnet *ifp
204 #ifndef KTR_IF_START
205 #define KTR_IF_START		KTR_ALL
206 #endif
207 KTR_INFO_MASTER(if_start);
208 KTR_INFO(KTR_IF_START, if_start, run, 0,
209 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
210 KTR_INFO(KTR_IF_START, if_start, sched, 1,
211 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
212 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
213 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
214 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
215 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
216 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
217 	 IF_START_KTR_STRING, IF_START_KTR_ARGS);
218 #define logifstart(name, arg)	KTR_LOG(if_start_ ## name, arg)
219 #endif /* notyet */
220 
221 /*
222  * Network interface utility routines.
223  *
224  * Routines with ifa_ifwith* names take sockaddr *'s as
225  * parameters.
226  */
227 /* ARGSUSED */
228 static void
229 ifinit(void *dummy)
230 {
231 	lockinit(&ifgroup_lock, "ifgroup", 0, 0);
232 
233 	callout_init_mp(&if_slowtimo_timer);
234 	netmsg_init(&if_slowtimo_netmsg, NULL, &netisr_adone_rport,
235 	    MSGF_PRIORITY, if_slowtimo_dispatch);
236 
237 	/* Start if_slowtimo */
238 	lwkt_sendmsg(netisr_cpuport(0), &if_slowtimo_netmsg.lmsg);
239 }
240 
241 static void
242 ifsq_ifstart_ipifunc(void *arg)
243 {
244 	struct ifaltq_subque *ifsq = arg;
245 	struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
246 
247 	crit_enter();
248 	if (lmsg->ms_flags & MSGF_DONE)
249 		lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg);
250 	crit_exit();
251 }
252 
253 static __inline void
254 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
255 {
256 	KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
257 	TAILQ_REMOVE(&head->stg_head, stage, stg_link);
258 	stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
259 	stage->stg_cnt = 0;
260 	stage->stg_len = 0;
261 }
262 
263 static __inline void
264 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
265 {
266 	KKASSERT((stage->stg_flags &
267 	    (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
268 	stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
269 	TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
270 }
271 
272 /*
273  * Schedule ifnet.if_start on the subqueue owner CPU
274  */
275 static void
276 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
277 {
278 	int cpu;
279 
280 	if (!force && curthread->td_type == TD_TYPE_NETISR &&
281 	    ifsq_stage_cntmax > 0) {
282 		struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
283 
284 		stage->stg_cnt = 0;
285 		stage->stg_len = 0;
286 		if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
287 			ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
288 		stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
289 		return;
290 	}
291 
292 	cpu = ifsq_get_cpuid(ifsq);
293 	if (cpu != mycpuid)
294 		lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
295 	else
296 		ifsq_ifstart_ipifunc(ifsq);
297 }
298 
299 /*
300  * NOTE:
301  * This function will release ifnet.if_start subqueue interlock,
302  * if ifnet.if_start for the subqueue does not need to be scheduled
303  */
304 static __inline int
305 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
306 {
307 	if (!running || ifsq_is_empty(ifsq)
308 #ifdef ALTQ
309 	    || ifsq->ifsq_altq->altq_tbr != NULL
310 #endif
311 	) {
312 		ALTQ_SQ_LOCK(ifsq);
313 		/*
314 		 * ifnet.if_start subqueue interlock is released, if:
315 		 * 1) Hardware can not take any packets, due to
316 		 *    o  interface is marked down
317 		 *    o  hardware queue is full (ifsq_is_oactive)
318 		 *    Under the second situation, hardware interrupt
319 		 *    or polling(4) will call/schedule ifnet.if_start
320 		 *    on the subqueue when hardware queue is ready
321 		 * 2) There is no packet in the subqueue.
322 		 *    Further ifq_dispatch or ifq_handoff will call/
323 		 *    schedule ifnet.if_start on the subqueue.
324 		 * 3) TBR is used and it does not allow further
325 		 *    dequeueing.
326 		 *    TBR callout will call ifnet.if_start on the
327 		 *    subqueue.
328 		 */
329 		if (!running || !ifsq_data_ready(ifsq)) {
330 			ifsq_clr_started(ifsq);
331 			ALTQ_SQ_UNLOCK(ifsq);
332 			return 0;
333 		}
334 		ALTQ_SQ_UNLOCK(ifsq);
335 	}
336 	return 1;
337 }
338 
339 static void
340 ifsq_ifstart_dispatch(netmsg_t msg)
341 {
342 	struct lwkt_msg *lmsg = &msg->base.lmsg;
343 	struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
344 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
345 	struct globaldata *gd = mycpu;
346 	int running = 0, need_sched;
347 
348 	crit_enter_gd(gd);
349 
350 	lwkt_replymsg(lmsg, 0);	/* reply ASAP */
351 
352 	if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) {
353 		/*
354 		 * We need to chase the subqueue owner CPU change.
355 		 */
356 		ifsq_ifstart_schedule(ifsq, 1);
357 		crit_exit_gd(gd);
358 		return;
359 	}
360 
361 	ifsq_serialize_hw(ifsq);
362 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
363 		ifp->if_start(ifp, ifsq);
364 		if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
365 			running = 1;
366 	}
367 	need_sched = ifsq_ifstart_need_schedule(ifsq, running);
368 	ifsq_deserialize_hw(ifsq);
369 
370 	if (need_sched) {
371 		/*
372 		 * More data need to be transmitted, ifnet.if_start is
373 		 * scheduled on the subqueue owner CPU, and we keep going.
374 		 * NOTE: ifnet.if_start subqueue interlock is not released.
375 		 */
376 		ifsq_ifstart_schedule(ifsq, 0);
377 	}
378 
379 	crit_exit_gd(gd);
380 }
381 
382 /* Device driver ifnet.if_start helper function */
383 void
384 ifsq_devstart(struct ifaltq_subque *ifsq)
385 {
386 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
387 	int running = 0;
388 
389 	ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
390 
391 	ALTQ_SQ_LOCK(ifsq);
392 	if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
393 		ALTQ_SQ_UNLOCK(ifsq);
394 		return;
395 	}
396 	ifsq_set_started(ifsq);
397 	ALTQ_SQ_UNLOCK(ifsq);
398 
399 	ifp->if_start(ifp, ifsq);
400 
401 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
402 		running = 1;
403 
404 	if (ifsq_ifstart_need_schedule(ifsq, running)) {
405 		/*
406 		 * More data need to be transmitted, ifnet.if_start is
407 		 * scheduled on ifnet's CPU, and we keep going.
408 		 * NOTE: ifnet.if_start interlock is not released.
409 		 */
410 		ifsq_ifstart_schedule(ifsq, 0);
411 	}
412 }
413 
414 void
415 if_devstart(struct ifnet *ifp)
416 {
417 	ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
418 }
419 
420 /* Device driver ifnet.if_start schedule helper function */
421 void
422 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
423 {
424 	ifsq_ifstart_schedule(ifsq, 1);
425 }
426 
427 void
428 if_devstart_sched(struct ifnet *ifp)
429 {
430 	ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
431 }
432 
433 static void
434 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
435 {
436 	lwkt_serialize_enter(ifp->if_serializer);
437 }
438 
439 static void
440 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
441 {
442 	lwkt_serialize_exit(ifp->if_serializer);
443 }
444 
445 static int
446 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
447 {
448 	return lwkt_serialize_try(ifp->if_serializer);
449 }
450 
451 #ifdef INVARIANTS
452 static void
453 if_default_serialize_assert(struct ifnet *ifp,
454 			    enum ifnet_serialize slz __unused,
455 			    boolean_t serialized)
456 {
457 	if (serialized)
458 		ASSERT_SERIALIZED(ifp->if_serializer);
459 	else
460 		ASSERT_NOT_SERIALIZED(ifp->if_serializer);
461 }
462 #endif
463 
464 /*
465  * Attach an interface to the list of "active" interfaces.
466  *
467  * The serializer is optional.
468  */
469 void
470 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
471 {
472 	unsigned socksize;
473 	int namelen, masklen;
474 	struct sockaddr_dl *sdl, *sdl_addr;
475 	struct ifaddr *ifa;
476 	struct ifaltq *ifq;
477 	struct ifnet **old_ifindex2ifnet = NULL;
478 	struct ifnet_array *old_ifnet_array;
479 	int i, q, qlen;
480 	char qlenname[64];
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 	 * Make if_addrhead available on all CPUs, since they
527 	 * could be accessed by any threads.
528 	 */
529 	ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
530 				    M_IFADDR, M_WAITOK | M_ZERO);
531 	for (i = 0; i < ncpus; ++i)
532 		TAILQ_INIT(&ifp->if_addrheads[i]);
533 
534 	TAILQ_INIT(&ifp->if_multiaddrs);
535 	TAILQ_INIT(&ifp->if_groups);
536 	getmicrotime(&ifp->if_lastchange);
537 	if_addgroup(ifp, IFG_ALL);
538 
539 	/*
540 	 * create a Link Level name for this device
541 	 */
542 	namelen = strlen(ifp->if_xname);
543 	masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
544 	socksize = masklen + ifp->if_addrlen;
545 	if (socksize < sizeof(*sdl))
546 		socksize = sizeof(*sdl);
547 	socksize = RT_ROUNDUP(socksize);
548 	ifa = ifa_create(sizeof(struct ifaddr) + 2 * socksize);
549 	sdl = sdl_addr = (struct sockaddr_dl *)(ifa + 1);
550 	sdl->sdl_len = socksize;
551 	sdl->sdl_family = AF_LINK;
552 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
553 	sdl->sdl_nlen = namelen;
554 	sdl->sdl_type = ifp->if_type;
555 	ifp->if_lladdr = ifa;
556 	ifa->ifa_ifp = ifp;
557 	ifa->ifa_rtrequest = link_rtrequest;
558 	ifa->ifa_addr = (struct sockaddr *)sdl;
559 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
560 	ifa->ifa_netmask = (struct sockaddr *)sdl;
561 	sdl->sdl_len = masklen;
562 	while (namelen != 0)
563 		sdl->sdl_data[--namelen] = 0xff;
564 	ifa_iflink(ifa, ifp, 0 /* Insert head */);
565 
566 	/*
567 	 * Make if_data available on all CPUs, since they could
568 	 * be updated by hardware interrupt routing, which could
569 	 * be bound to any CPU.
570 	 */
571 	ifp->if_data_pcpu = kmalloc_cachealign(
572 	    ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
573 
574 	if (ifp->if_mapsubq == NULL)
575 		ifp->if_mapsubq = ifq_mapsubq_default;
576 
577 	ifq = &ifp->if_snd;
578 	ifq->altq_type = 0;
579 	ifq->altq_disc = NULL;
580 	ifq->altq_flags &= ALTQF_CANTCHANGE;
581 	ifq->altq_tbr = NULL;
582 	ifq->altq_ifp = ifp;
583 
584 	if (ifq->altq_subq_cnt <= 0)
585 		ifq->altq_subq_cnt = 1;
586 	ifq->altq_subq = kmalloc_cachealign(
587 	    ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
588 	    M_DEVBUF, M_WAITOK | M_ZERO);
589 
590 	if (ifq->altq_maxlen == 0) {
591 		if_printf(ifp, "driver didn't set altq_maxlen\n");
592 		ifq_set_maxlen(ifq, ifqmaxlen);
593 	}
594 
595 	/* Allow user to override driver's setting. */
596 	ksnprintf(qlenname, sizeof(qlenname), "net.%s.qlenmax", ifp->if_xname);
597 	qlen = -1;
598 	TUNABLE_INT_FETCH(qlenname, &qlen);
599 	if (qlen > 0) {
600 		if_printf(ifp, "qlenmax -> %d\n", qlen);
601 		ifq_set_maxlen(ifq, qlen);
602 	}
603 
604 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
605 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
606 
607 		ALTQ_SQ_LOCK_INIT(ifsq);
608 		ifsq->ifsq_index = q;
609 
610 		ifsq->ifsq_altq = ifq;
611 		ifsq->ifsq_ifp = ifp;
612 
613 		ifsq->ifsq_maxlen = ifq->altq_maxlen;
614 		ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
615 		ifsq->ifsq_prepended = NULL;
616 		ifsq->ifsq_started = 0;
617 		ifsq->ifsq_hw_oactive = 0;
618 		ifsq_set_cpuid(ifsq, 0);
619 		if (ifp->if_serializer != NULL)
620 			ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
621 
622 		/* XXX: netisr_ncpus */
623 		ifsq->ifsq_stage =
624 		    kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
625 		    M_DEVBUF, M_WAITOK | M_ZERO);
626 		for (i = 0; i < ncpus; ++i)
627 			ifsq->ifsq_stage[i].stg_subq = ifsq;
628 
629 		/*
630 		 * Allocate one if_start message for each CPU, since
631 		 * the hardware TX ring could be assigned to any CPU.
632 		 *
633 		 * NOTE:
634 		 * If the hardware TX ring polling CPU and the hardware
635 		 * TX ring interrupt CPU are same, one if_start message
636 		 * should be enough.
637 		 */
638 		ifsq->ifsq_ifstart_nmsg =
639 		    kmalloc(ncpus * sizeof(struct netmsg_base),
640 		    M_LWKTMSG, M_WAITOK);
641 		for (i = 0; i < ncpus; ++i) {
642 			netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
643 			    &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
644 			ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
645 		}
646 	}
647 	ifq_set_classic(ifq);
648 
649 	/*
650 	 * Increase mbuf cluster/jcluster limits for the mbufs that
651 	 * could sit on the device queues for quite some time.
652 	 */
653 	if (ifp->if_nmbclusters > 0)
654 		mcl_inclimit(ifp->if_nmbclusters);
655 	if (ifp->if_nmbjclusters > 0)
656 		mjcl_inclimit(ifp->if_nmbjclusters);
657 
658 	/*
659 	 * Install this ifp into ifindex2inet, ifnet queue and ifnet
660 	 * array after it is setup.
661 	 *
662 	 * Protect ifindex2ifnet, ifnet queue and ifnet array changes
663 	 * by ifnet lock, so that non-netisr threads could get a
664 	 * consistent view.
665 	 */
666 	ifnet_lock();
667 
668 	/* Don't update if_index until ifindex2ifnet is setup */
669 	ifp->if_index = if_index + 1;
670 	sdl_addr->sdl_index = ifp->if_index;
671 
672 	/*
673 	 * Install this ifp into ifindex2ifnet
674 	 */
675 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
676 		unsigned int n;
677 		struct ifnet **q;
678 
679 		/*
680 		 * Grow ifindex2ifnet
681 		 */
682 		if_indexlim <<= 1;
683 		n = if_indexlim * sizeof(*q);
684 		q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
685 		if (ifindex2ifnet != NULL) {
686 			bcopy(ifindex2ifnet, q, n/2);
687 			/* Free old ifindex2ifnet after sync all netisrs */
688 			old_ifindex2ifnet = ifindex2ifnet;
689 		}
690 		ifindex2ifnet = q;
691 	}
692 	ifindex2ifnet[ifp->if_index] = ifp;
693 	/*
694 	 * Update if_index after this ifp is installed into ifindex2ifnet,
695 	 * so that netisrs could get a consistent view of ifindex2ifnet.
696 	 */
697 	cpu_sfence();
698 	if_index = ifp->if_index;
699 
700 	/*
701 	 * Install this ifp into ifnet array.
702 	 */
703 	/* Free old ifnet array after sync all netisrs */
704 	old_ifnet_array = ifnet_array;
705 	ifnet_array = ifnet_array_add(ifp, old_ifnet_array);
706 
707 	/*
708 	 * Install this ifp into ifnet queue.
709 	 */
710 	TAILQ_INSERT_TAIL(&ifnetlist, ifp, if_link);
711 
712 	ifnet_unlock();
713 
714 	/*
715 	 * Sync all netisrs so that the old ifindex2ifnet and ifnet array
716 	 * are no longer accessed and we can free them safely later on.
717 	 */
718 	netmsg_service_sync();
719 	if (old_ifindex2ifnet != NULL)
720 		kfree(old_ifindex2ifnet, M_IFADDR);
721 	ifnet_array_free(old_ifnet_array);
722 
723 	if (!SLIST_EMPTY(&domains))
724 		if_attachdomain1(ifp);
725 
726 	/* Announce the interface. */
727 	EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
728 	devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
729 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
730 }
731 
732 static void
733 if_attachdomain(void *dummy)
734 {
735 	struct ifnet *ifp;
736 
737 	ifnet_lock();
738 	TAILQ_FOREACH(ifp, &ifnetlist, if_list)
739 		if_attachdomain1(ifp);
740 	ifnet_unlock();
741 }
742 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
743 	if_attachdomain, NULL);
744 
745 static void
746 if_attachdomain1(struct ifnet *ifp)
747 {
748 	struct domain *dp;
749 
750 	crit_enter();
751 
752 	/* address family dependent data region */
753 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
754 	SLIST_FOREACH(dp, &domains, dom_next)
755 		if (dp->dom_ifattach)
756 			ifp->if_afdata[dp->dom_family] =
757 				(*dp->dom_ifattach)(ifp);
758 	crit_exit();
759 }
760 
761 /*
762  * Purge all addresses whose type is _not_ AF_LINK
763  */
764 static void
765 if_purgeaddrs_nolink_dispatch(netmsg_t nmsg)
766 {
767 	struct ifnet *ifp = nmsg->lmsg.u.ms_resultp;
768 	struct ifaddr_container *ifac, *next;
769 
770 	ASSERT_NETISR0;
771 
772 	/*
773 	 * The ifaddr processing in the following loop will block,
774 	 * however, this function is called in netisr0, in which
775 	 * ifaddr list changes happen, so we don't care about the
776 	 * blockness of the ifaddr processing here.
777 	 */
778 	TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
779 			      ifa_link, next) {
780 		struct ifaddr *ifa = ifac->ifa;
781 
782 		/* Ignore marker */
783 		if (ifa->ifa_addr->sa_family == AF_UNSPEC)
784 			continue;
785 
786 		/* Leave link ifaddr as it is */
787 		if (ifa->ifa_addr->sa_family == AF_LINK)
788 			continue;
789 #ifdef INET
790 		/* XXX: Ugly!! ad hoc just for INET */
791 		if (ifa->ifa_addr->sa_family == AF_INET) {
792 			struct ifaliasreq ifr;
793 			struct sockaddr_in saved_addr, saved_dst;
794 #ifdef IFADDR_DEBUG_VERBOSE
795 			int i;
796 
797 			kprintf("purge in4 addr %p: ", ifa);
798 			for (i = 0; i < ncpus; ++i) {
799 				kprintf("%d ",
800 				    ifa->ifa_containers[i].ifa_refcnt);
801 			}
802 			kprintf("\n");
803 #endif
804 
805 			/* Save information for panic. */
806 			memcpy(&saved_addr, ifa->ifa_addr, sizeof(saved_addr));
807 			if (ifa->ifa_dstaddr != NULL) {
808 				memcpy(&saved_dst, ifa->ifa_dstaddr,
809 				    sizeof(saved_dst));
810 			} else {
811 				memset(&saved_dst, 0, sizeof(saved_dst));
812 			}
813 
814 			bzero(&ifr, sizeof ifr);
815 			ifr.ifra_addr = *ifa->ifa_addr;
816 			if (ifa->ifa_dstaddr)
817 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
818 			if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp,
819 				       NULL) == 0)
820 				continue;
821 
822 			/* MUST NOT HAPPEN */
823 			panic("%s: in_control failed %x, dst %x", ifp->if_xname,
824 			    ntohl(saved_addr.sin_addr.s_addr),
825 			    ntohl(saved_dst.sin_addr.s_addr));
826 		}
827 #endif /* INET */
828 #ifdef INET6
829 		if (ifa->ifa_addr->sa_family == AF_INET6) {
830 #ifdef IFADDR_DEBUG_VERBOSE
831 			int i;
832 
833 			kprintf("purge in6 addr %p: ", ifa);
834 			for (i = 0; i < ncpus; ++i) {
835 				kprintf("%d ",
836 				    ifa->ifa_containers[i].ifa_refcnt);
837 			}
838 			kprintf("\n");
839 #endif
840 
841 			in6_purgeaddr(ifa);
842 			/* ifp_addrhead is already updated */
843 			continue;
844 		}
845 #endif /* INET6 */
846 		if_printf(ifp, "destroy ifaddr family %d\n",
847 		    ifa->ifa_addr->sa_family);
848 		ifa_ifunlink(ifa, ifp);
849 		ifa_destroy(ifa);
850 	}
851 
852 	netisr_replymsg(&nmsg->base, 0);
853 }
854 
855 void
856 if_purgeaddrs_nolink(struct ifnet *ifp)
857 {
858 	struct netmsg_base nmsg;
859 
860 	netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
861 	    if_purgeaddrs_nolink_dispatch);
862 	nmsg.lmsg.u.ms_resultp = ifp;
863 	netisr_domsg(&nmsg, 0);
864 }
865 
866 static void
867 ifq_stage_detach_handler(netmsg_t nmsg)
868 {
869 	struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
870 	int q;
871 
872 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
873 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
874 		struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
875 
876 		if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
877 			ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
878 	}
879 	lwkt_replymsg(&nmsg->lmsg, 0);
880 }
881 
882 static void
883 ifq_stage_detach(struct ifaltq *ifq)
884 {
885 	struct netmsg_base base;
886 	int cpu;
887 
888 	netmsg_init(&base, NULL, &curthread->td_msgport, 0,
889 	    ifq_stage_detach_handler);
890 	base.lmsg.u.ms_resultp = ifq;
891 
892 	/* XXX netisr_ncpus */
893 	for (cpu = 0; cpu < ncpus; ++cpu)
894 		lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
895 }
896 
897 struct netmsg_if_rtdel {
898 	struct netmsg_base	base;
899 	struct ifnet		*ifp;
900 };
901 
902 static void
903 if_rtdel_dispatch(netmsg_t msg)
904 {
905 	struct netmsg_if_rtdel *rmsg = (void *)msg;
906 	int i, cpu;
907 
908 	cpu = mycpuid;
909 	ASSERT_NETISR_NCPUS(cpu);
910 
911 	for (i = 1; i <= AF_MAX; i++) {
912 		struct radix_node_head	*rnh;
913 
914 		if ((rnh = rt_tables[cpu][i]) == NULL)
915 			continue;
916 		rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
917 	}
918 	netisr_forwardmsg(&msg->base, cpu + 1);
919 }
920 
921 /*
922  * Detach an interface, removing it from the
923  * list of "active" interfaces.
924  */
925 void
926 if_detach(struct ifnet *ifp)
927 {
928 	struct ifnet_array *old_ifnet_array;
929 	struct ifg_list *ifgl;
930 	struct netmsg_if_rtdel msg;
931 	struct domain *dp;
932 	int q;
933 
934 	/* Announce that the interface is gone. */
935 	EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
936 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
937 	devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
938 
939 	/*
940 	 * Remove this ifp from ifindex2inet, ifnet queue and ifnet
941 	 * array before it is whacked.
942 	 *
943 	 * Protect ifindex2ifnet, ifnet queue and ifnet array changes
944 	 * by ifnet lock, so that non-netisr threads could get a
945 	 * consistent view.
946 	 */
947 	ifnet_lock();
948 
949 	/*
950 	 * Remove this ifp from ifindex2ifnet and maybe decrement if_index.
951 	 */
952 	ifindex2ifnet[ifp->if_index] = NULL;
953 	while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
954 		if_index--;
955 
956 	/*
957 	 * Remove this ifp from ifnet queue.
958 	 */
959 	TAILQ_REMOVE(&ifnetlist, ifp, if_link);
960 
961 	/*
962 	 * Remove this ifp from ifnet array.
963 	 */
964 	/* Free old ifnet array after sync all netisrs */
965 	old_ifnet_array = ifnet_array;
966 	ifnet_array = ifnet_array_del(ifp, old_ifnet_array);
967 
968 	ifnet_unlock();
969 
970 	ifgroup_lockmgr(LK_EXCLUSIVE);
971 	while ((ifgl = TAILQ_FIRST(&ifp->if_groups)) != NULL)
972 		if_delgroup_locked(ifp, ifgl->ifgl_group->ifg_group);
973 	ifgroup_lockmgr(LK_RELEASE);
974 
975 	/*
976 	 * Sync all netisrs so that the old ifnet array is no longer
977 	 * accessed and we can free it safely later on.
978 	 */
979 	netmsg_service_sync();
980 	ifnet_array_free(old_ifnet_array);
981 
982 	/*
983 	 * Remove routes and flush queues.
984 	 */
985 	crit_enter();
986 #ifdef IFPOLL_ENABLE
987 	if (ifp->if_flags & IFF_NPOLLING)
988 		ifpoll_deregister(ifp);
989 #endif
990 	if_down(ifp);
991 
992 	/* Decrease the mbuf clusters/jclusters limits increased by us */
993 	if (ifp->if_nmbclusters > 0)
994 		mcl_inclimit(-ifp->if_nmbclusters);
995 	if (ifp->if_nmbjclusters > 0)
996 		mjcl_inclimit(-ifp->if_nmbjclusters);
997 
998 #ifdef ALTQ
999 	if (ifq_is_enabled(&ifp->if_snd))
1000 		altq_disable(&ifp->if_snd);
1001 	if (ifq_is_attached(&ifp->if_snd))
1002 		altq_detach(&ifp->if_snd);
1003 #endif
1004 
1005 	/*
1006 	 * Clean up all addresses.
1007 	 */
1008 	ifp->if_lladdr = NULL;
1009 
1010 	if_purgeaddrs_nolink(ifp);
1011 	if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
1012 		struct ifaddr *ifa;
1013 
1014 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1015 		KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
1016 			("non-link ifaddr is left on if_addrheads"));
1017 
1018 		ifa_ifunlink(ifa, ifp);
1019 		ifa_destroy(ifa);
1020 		KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
1021 			("there are still ifaddrs left on if_addrheads"));
1022 	}
1023 
1024 #ifdef INET
1025 	/*
1026 	 * Remove all IPv4 kernel structures related to ifp.
1027 	 */
1028 	in_ifdetach(ifp);
1029 #endif
1030 
1031 #ifdef INET6
1032 	/*
1033 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
1034 	 * before removing routing entries below, since IPv6 interface direct
1035 	 * routes are expected to be removed by the IPv6-specific kernel API.
1036 	 * Otherwise, the kernel will detect some inconsistency and bark it.
1037 	 */
1038 	in6_ifdetach(ifp);
1039 #endif
1040 
1041 	/*
1042 	 * Delete all remaining routes using this interface
1043 	 */
1044 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1045 	    if_rtdel_dispatch);
1046 	msg.ifp = ifp;
1047 	netisr_domsg_global(&msg.base);
1048 
1049 	SLIST_FOREACH(dp, &domains, dom_next) {
1050 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
1051 			(*dp->dom_ifdetach)(ifp,
1052 				ifp->if_afdata[dp->dom_family]);
1053 	}
1054 
1055 	kfree(ifp->if_addrheads, M_IFADDR);
1056 
1057 	lwkt_synchronize_ipiqs("if_detach");
1058 	ifq_stage_detach(&ifp->if_snd);
1059 
1060 	for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
1061 		struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
1062 
1063 		kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
1064 		kfree(ifsq->ifsq_stage, M_DEVBUF);
1065 	}
1066 	kfree(ifp->if_snd.altq_subq, M_DEVBUF);
1067 
1068 	kfree(ifp->if_data_pcpu, M_DEVBUF);
1069 
1070 	crit_exit();
1071 }
1072 
1073 int
1074 ifgroup_lockmgr(u_int flags)
1075 {
1076 	return lockmgr(&ifgroup_lock, flags);
1077 }
1078 
1079 /*
1080  * Create an empty interface group.
1081  */
1082 static struct ifg_group *
1083 if_creategroup(const char *groupname)
1084 {
1085 	struct ifg_group *ifg;
1086 
1087 	ifg = kmalloc(sizeof(*ifg), M_IFNET, M_WAITOK);
1088 	strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1089 	ifg->ifg_refcnt = 0;
1090 	ifg->ifg_carp_demoted = 0;
1091 	TAILQ_INIT(&ifg->ifg_members);
1092 
1093 	ifgroup_lockmgr(LK_EXCLUSIVE);
1094 	TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
1095 	ifgroup_lockmgr(LK_RELEASE);
1096 
1097 	EVENTHANDLER_INVOKE(group_attach_event, ifg);
1098 
1099 	return (ifg);
1100 }
1101 
1102 /*
1103  * Destroy an empty interface group.
1104  */
1105 static int
1106 if_destroygroup(struct ifg_group *ifg)
1107 {
1108 	KASSERT(ifg->ifg_refcnt == 0,
1109 		("trying to delete a non-empty interface group"));
1110 
1111 	ifgroup_lockmgr(LK_EXCLUSIVE);
1112 	TAILQ_REMOVE(&ifg_head, ifg, ifg_next);
1113 	ifgroup_lockmgr(LK_RELEASE);
1114 
1115 	EVENTHANDLER_INVOKE(group_detach_event, ifg);
1116 	kfree(ifg, M_IFNET);
1117 
1118 	return (0);
1119 }
1120 
1121 /*
1122  * Add the interface to a group.
1123  * The target group will be created if it doesn't exist.
1124  */
1125 int
1126 if_addgroup(struct ifnet *ifp, const char *groupname)
1127 {
1128 	struct ifg_list *ifgl;
1129 	struct ifg_group *ifg;
1130 	struct ifg_member *ifgm;
1131 
1132 	if (groupname[0] &&
1133 	    groupname[strlen(groupname) - 1] >= '0' &&
1134 	    groupname[strlen(groupname) - 1] <= '9')
1135 		return (EINVAL);
1136 
1137 	ifgroup_lockmgr(LK_SHARED);
1138 
1139 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1140 		if (strcmp(ifgl->ifgl_group->ifg_group, groupname) == 0) {
1141 			ifgroup_lockmgr(LK_RELEASE);
1142 			return (EEXIST);
1143 		}
1144 	}
1145 
1146 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next) {
1147 		if (strcmp(ifg->ifg_group, groupname) == 0)
1148 			break;
1149 	}
1150 
1151 	ifgroup_lockmgr(LK_RELEASE);
1152 
1153 	if (ifg == NULL)
1154 		ifg = if_creategroup(groupname);
1155 
1156 	ifgl = kmalloc(sizeof(*ifgl), M_IFNET, M_WAITOK);
1157 	ifgm = kmalloc(sizeof(*ifgm), M_IFNET, M_WAITOK);
1158 	ifgl->ifgl_group = ifg;
1159 	ifgm->ifgm_ifp = ifp;
1160 	ifg->ifg_refcnt++;
1161 
1162 	ifgroup_lockmgr(LK_EXCLUSIVE);
1163 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1164 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1165 	ifgroup_lockmgr(LK_RELEASE);
1166 
1167 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1168 
1169 	return (0);
1170 }
1171 
1172 /*
1173  * Remove the interface from a group.
1174  * The group will be destroyed if it becomes empty.
1175  *
1176  * The 'ifgroup_lock' must be hold exclusively when calling this.
1177  */
1178 static int
1179 if_delgroup_locked(struct ifnet *ifp, const char *groupname)
1180 {
1181 	struct ifg_list *ifgl;
1182 	struct ifg_member *ifgm;
1183 
1184 	KKASSERT(lockstatus(&ifgroup_lock, curthread) == LK_EXCLUSIVE);
1185 
1186 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1187 		if (strcmp(ifgl->ifgl_group->ifg_group, groupname) == 0)
1188 			break;
1189 	}
1190 	if (ifgl == NULL)
1191 		return (ENOENT);
1192 
1193 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1194 
1195 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next) {
1196 		if (ifgm->ifgm_ifp == ifp)
1197 			break;
1198 	}
1199 
1200 	if (ifgm != NULL) {
1201 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1202 
1203 		ifgroup_lockmgr(LK_RELEASE);
1204 		EVENTHANDLER_INVOKE(group_change_event, groupname);
1205 		ifgroup_lockmgr(LK_EXCLUSIVE);
1206 
1207 		kfree(ifgm, M_IFNET);
1208 		ifgl->ifgl_group->ifg_refcnt--;
1209 	}
1210 
1211 	if (ifgl->ifgl_group->ifg_refcnt == 0) {
1212 		ifgroup_lockmgr(LK_RELEASE);
1213 		if_destroygroup(ifgl->ifgl_group);
1214 		ifgroup_lockmgr(LK_EXCLUSIVE);
1215 	}
1216 
1217 	kfree(ifgl, M_IFNET);
1218 
1219 	return (0);
1220 }
1221 
1222 int
1223 if_delgroup(struct ifnet *ifp, const char *groupname)
1224 {
1225 	int error;
1226 
1227 	ifgroup_lockmgr(LK_EXCLUSIVE);
1228 	error = if_delgroup_locked(ifp, groupname);
1229 	ifgroup_lockmgr(LK_RELEASE);
1230 
1231 	return (error);
1232 }
1233 
1234 /*
1235  * Store all the groups that the interface belongs to in memory
1236  * pointed to by data.
1237  */
1238 static int
1239 if_getgroups(struct ifgroupreq *ifgr, struct ifnet *ifp)
1240 {
1241 	struct ifg_list *ifgl;
1242 	struct ifg_req *ifgrq, *p;
1243 	int len, error;
1244 
1245 	len = 0;
1246 	ifgroup_lockmgr(LK_SHARED);
1247 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1248 		len += sizeof(struct ifg_req);
1249 	ifgroup_lockmgr(LK_RELEASE);
1250 
1251 	if (ifgr->ifgr_len == 0) {
1252 		/*
1253 		 * Caller is asking how much memory should be allocated in
1254 		 * the next request in order to hold all the groups.
1255 		 */
1256 		ifgr->ifgr_len = len;
1257 		return (0);
1258 	} else if (ifgr->ifgr_len != len) {
1259 		return (EINVAL);
1260 	}
1261 
1262 	ifgrq = kmalloc(len, M_TEMP, M_INTWAIT | M_NULLOK | M_ZERO);
1263 	if (ifgrq == NULL)
1264 		return (ENOMEM);
1265 
1266 	ifgroup_lockmgr(LK_SHARED);
1267 	p = ifgrq;
1268 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1269 		if (len < sizeof(struct ifg_req)) {
1270 			ifgroup_lockmgr(LK_RELEASE);
1271 			return (EINVAL);
1272 		}
1273 
1274 		strlcpy(p->ifgrq_group, ifgl->ifgl_group->ifg_group,
1275 			sizeof(ifgrq->ifgrq_group));
1276 		len -= sizeof(struct ifg_req);
1277 		p++;
1278 	}
1279 	ifgroup_lockmgr(LK_RELEASE);
1280 
1281 	error = copyout(ifgrq, ifgr->ifgr_groups, ifgr->ifgr_len);
1282 	kfree(ifgrq, M_TEMP);
1283 	if (error)
1284 		return (error);
1285 
1286 	return (0);
1287 }
1288 
1289 /*
1290  * Store all the members of a group in memory pointed to by data.
1291  */
1292 static int
1293 if_getgroupmembers(struct ifgroupreq *ifgr)
1294 {
1295 	struct ifg_group *ifg;
1296 	struct ifg_member *ifgm;
1297 	struct ifg_req *ifgrq, *p;
1298 	int len, error;
1299 
1300 	ifgroup_lockmgr(LK_SHARED);
1301 
1302 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next) {
1303 		if (strcmp(ifg->ifg_group, ifgr->ifgr_name) == 0)
1304 			break;
1305 	}
1306 	if (ifg == NULL) {
1307 		ifgroup_lockmgr(LK_RELEASE);
1308 		return (ENOENT);
1309 	}
1310 
1311 	len = 0;
1312 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1313 		len += sizeof(struct ifg_req);
1314 
1315 	ifgroup_lockmgr(LK_RELEASE);
1316 
1317 	if (ifgr->ifgr_len == 0) {
1318 		ifgr->ifgr_len = len;
1319 		return (0);
1320 	} else if (ifgr->ifgr_len != len) {
1321 		return (EINVAL);
1322 	}
1323 
1324 	ifgrq = kmalloc(len, M_TEMP, M_INTWAIT | M_NULLOK | M_ZERO);
1325 	if (ifgrq == NULL)
1326 		return (ENOMEM);
1327 
1328 	ifgroup_lockmgr(LK_SHARED);
1329 	p = ifgrq;
1330 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1331 		if (len < sizeof(struct ifg_req)) {
1332 			ifgroup_lockmgr(LK_RELEASE);
1333 			return (EINVAL);
1334 		}
1335 
1336 		strlcpy(p->ifgrq_member, ifgm->ifgm_ifp->if_xname,
1337 			sizeof(p->ifgrq_member));
1338 		len -= sizeof(struct ifg_req);
1339 		p++;
1340 	}
1341 	ifgroup_lockmgr(LK_RELEASE);
1342 
1343 	error = copyout(ifgrq, ifgr->ifgr_groups, ifgr->ifgr_len);
1344 	kfree(ifgrq, M_TEMP);
1345 	if (error)
1346 		return (error);
1347 
1348 	return (0);
1349 }
1350 
1351 /*
1352  * Delete Routes for a Network Interface
1353  *
1354  * Called for each routing entry via the rnh->rnh_walktree() call above
1355  * to delete all route entries referencing a detaching network interface.
1356  *
1357  * Arguments:
1358  *	rn	pointer to node in the routing table
1359  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1360  *
1361  * Returns:
1362  *	0	successful
1363  *	errno	failed - reason indicated
1364  *
1365  */
1366 static int
1367 if_rtdel(struct radix_node *rn, void *arg)
1368 {
1369 	struct rtentry	*rt = (struct rtentry *)rn;
1370 	struct ifnet	*ifp = arg;
1371 	int		err;
1372 
1373 	if (rt->rt_ifp == ifp) {
1374 
1375 		/*
1376 		 * Protect (sorta) against walktree recursion problems
1377 		 * with cloned routes
1378 		 */
1379 		if (!(rt->rt_flags & RTF_UP))
1380 			return (0);
1381 
1382 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1383 				rt_mask(rt), rt->rt_flags,
1384 				NULL);
1385 		if (err) {
1386 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1387 		}
1388 	}
1389 
1390 	return (0);
1391 }
1392 
1393 static __inline boolean_t
1394 ifa_prefer(const struct ifaddr *cur_ifa, const struct ifaddr *old_ifa)
1395 {
1396 	if (old_ifa == NULL)
1397 		return TRUE;
1398 
1399 	if ((old_ifa->ifa_ifp->if_flags & IFF_UP) == 0 &&
1400 	    (cur_ifa->ifa_ifp->if_flags & IFF_UP))
1401 		return TRUE;
1402 	if ((old_ifa->ifa_flags & IFA_ROUTE) == 0 &&
1403 	    (cur_ifa->ifa_flags & IFA_ROUTE))
1404 		return TRUE;
1405 	return FALSE;
1406 }
1407 
1408 /*
1409  * Locate an interface based on a complete address.
1410  */
1411 struct ifaddr *
1412 ifa_ifwithaddr(struct sockaddr *addr)
1413 {
1414 	const struct ifnet_array *arr;
1415 	int i;
1416 
1417 	arr = ifnet_array_get();
1418 	for (i = 0; i < arr->ifnet_count; ++i) {
1419 		struct ifnet *ifp = arr->ifnet_arr[i];
1420 		struct ifaddr_container *ifac;
1421 
1422 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1423 			struct ifaddr *ifa = ifac->ifa;
1424 
1425 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1426 				continue;
1427 			if (sa_equal(addr, ifa->ifa_addr))
1428 				return (ifa);
1429 			if ((ifp->if_flags & IFF_BROADCAST) &&
1430 			    ifa->ifa_broadaddr &&
1431 			    /* IPv6 doesn't have broadcast */
1432 			    ifa->ifa_broadaddr->sa_len != 0 &&
1433 			    sa_equal(ifa->ifa_broadaddr, addr))
1434 				return (ifa);
1435 		}
1436 	}
1437 	return (NULL);
1438 }
1439 
1440 /*
1441  * Locate the point to point interface with a given destination address.
1442  */
1443 struct ifaddr *
1444 ifa_ifwithdstaddr(struct sockaddr *addr)
1445 {
1446 	const struct ifnet_array *arr;
1447 	int i;
1448 
1449 	arr = ifnet_array_get();
1450 	for (i = 0; i < arr->ifnet_count; ++i) {
1451 		struct ifnet *ifp = arr->ifnet_arr[i];
1452 		struct ifaddr_container *ifac;
1453 
1454 		if (!(ifp->if_flags & IFF_POINTOPOINT))
1455 			continue;
1456 
1457 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1458 			struct ifaddr *ifa = ifac->ifa;
1459 
1460 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1461 				continue;
1462 			if (ifa->ifa_dstaddr &&
1463 			    sa_equal(addr, ifa->ifa_dstaddr))
1464 				return (ifa);
1465 		}
1466 	}
1467 	return (NULL);
1468 }
1469 
1470 /*
1471  * Find an interface on a specific network.  If many, choice
1472  * is most specific found.
1473  */
1474 struct ifaddr *
1475 ifa_ifwithnet(struct sockaddr *addr)
1476 {
1477 	struct ifaddr *ifa_maybe = NULL;
1478 	u_int af = addr->sa_family;
1479 	char *addr_data = addr->sa_data, *cplim;
1480 	const struct ifnet_array *arr;
1481 	int i;
1482 
1483 	/*
1484 	 * AF_LINK addresses can be looked up directly by their index number,
1485 	 * so do that if we can.
1486 	 */
1487 	if (af == AF_LINK) {
1488 		struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1489 
1490 		if (sdl->sdl_index && sdl->sdl_index <= if_index)
1491 			return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1492 	}
1493 
1494 	/*
1495 	 * Scan though each interface, looking for ones that have
1496 	 * addresses in this address family.
1497 	 */
1498 	arr = ifnet_array_get();
1499 	for (i = 0; i < arr->ifnet_count; ++i) {
1500 		struct ifnet *ifp = arr->ifnet_arr[i];
1501 		struct ifaddr_container *ifac;
1502 
1503 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1504 			struct ifaddr *ifa = ifac->ifa;
1505 			char *cp, *cp2, *cp3;
1506 
1507 			if (ifa->ifa_addr->sa_family != af)
1508 next:				continue;
1509 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1510 				/*
1511 				 * This is a bit broken as it doesn't
1512 				 * take into account that the remote end may
1513 				 * be a single node in the network we are
1514 				 * looking for.
1515 				 * The trouble is that we don't know the
1516 				 * netmask for the remote end.
1517 				 */
1518 				if (ifa->ifa_dstaddr != NULL &&
1519 				    sa_equal(addr, ifa->ifa_dstaddr))
1520 					return (ifa);
1521 			} else {
1522 				/*
1523 				 * if we have a special address handler,
1524 				 * then use it instead of the generic one.
1525 				 */
1526 				if (ifa->ifa_claim_addr) {
1527 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1528 						return (ifa);
1529 					} else {
1530 						continue;
1531 					}
1532 				}
1533 
1534 				/*
1535 				 * Scan all the bits in the ifa's address.
1536 				 * If a bit dissagrees with what we are
1537 				 * looking for, mask it with the netmask
1538 				 * to see if it really matters.
1539 				 * (A byte at a time)
1540 				 */
1541 				if (ifa->ifa_netmask == 0)
1542 					continue;
1543 				cp = addr_data;
1544 				cp2 = ifa->ifa_addr->sa_data;
1545 				cp3 = ifa->ifa_netmask->sa_data;
1546 				cplim = ifa->ifa_netmask->sa_len +
1547 					(char *)ifa->ifa_netmask;
1548 				while (cp3 < cplim)
1549 					if ((*cp++ ^ *cp2++) & *cp3++)
1550 						goto next; /* next address! */
1551 				/*
1552 				 * If the netmask of what we just found
1553 				 * is more specific than what we had before
1554 				 * (if we had one) then remember the new one
1555 				 * before continuing to search for an even
1556 				 * better one.  If the netmasks are equal,
1557 				 * we prefer the this ifa based on the result
1558 				 * of ifa_prefer().
1559 				 */
1560 				if (ifa_maybe == NULL ||
1561 				    rn_refines((char *)ifa->ifa_netmask,
1562 				        (char *)ifa_maybe->ifa_netmask) ||
1563 				    (sa_equal(ifa_maybe->ifa_netmask,
1564 				        ifa->ifa_netmask) &&
1565 				     ifa_prefer(ifa, ifa_maybe)))
1566 					ifa_maybe = ifa;
1567 			}
1568 		}
1569 	}
1570 	return (ifa_maybe);
1571 }
1572 
1573 /*
1574  * Find an interface address specific to an interface best matching
1575  * a given address.
1576  */
1577 struct ifaddr *
1578 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1579 {
1580 	struct ifaddr_container *ifac;
1581 	char *cp, *cp2, *cp3;
1582 	char *cplim;
1583 	struct ifaddr *ifa_maybe = NULL;
1584 	u_int af = addr->sa_family;
1585 
1586 	if (af >= AF_MAX)
1587 		return (0);
1588 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1589 		struct ifaddr *ifa = ifac->ifa;
1590 
1591 		if (ifa->ifa_addr->sa_family != af)
1592 			continue;
1593 		if (ifa_maybe == NULL)
1594 			ifa_maybe = ifa;
1595 		if (ifa->ifa_netmask == NULL) {
1596 			if (sa_equal(addr, ifa->ifa_addr) ||
1597 			    (ifa->ifa_dstaddr != NULL &&
1598 			     sa_equal(addr, ifa->ifa_dstaddr)))
1599 				return (ifa);
1600 			continue;
1601 		}
1602 		if (ifp->if_flags & IFF_POINTOPOINT) {
1603 			if (sa_equal(addr, ifa->ifa_dstaddr))
1604 				return (ifa);
1605 		} else {
1606 			cp = addr->sa_data;
1607 			cp2 = ifa->ifa_addr->sa_data;
1608 			cp3 = ifa->ifa_netmask->sa_data;
1609 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1610 			for (; cp3 < cplim; cp3++)
1611 				if ((*cp++ ^ *cp2++) & *cp3)
1612 					break;
1613 			if (cp3 == cplim)
1614 				return (ifa);
1615 		}
1616 	}
1617 	return (ifa_maybe);
1618 }
1619 
1620 /*
1621  * Default action when installing a route with a Link Level gateway.
1622  * Lookup an appropriate real ifa to point to.
1623  * This should be moved to /sys/net/link.c eventually.
1624  */
1625 static void
1626 link_rtrequest(int cmd, struct rtentry *rt)
1627 {
1628 	struct ifaddr *ifa;
1629 	struct sockaddr *dst;
1630 	struct ifnet *ifp;
1631 
1632 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1633 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1634 		return;
1635 	ifa = ifaof_ifpforaddr(dst, ifp);
1636 	if (ifa != NULL) {
1637 		IFAFREE(rt->rt_ifa);
1638 		IFAREF(ifa);
1639 		rt->rt_ifa = ifa;
1640 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1641 			ifa->ifa_rtrequest(cmd, rt);
1642 	}
1643 }
1644 
1645 struct netmsg_ifroute {
1646 	struct netmsg_base	base;
1647 	struct ifnet		*ifp;
1648 	int			flag;
1649 	int			fam;
1650 };
1651 
1652 /*
1653  * Mark an interface down and notify protocols of the transition.
1654  */
1655 static void
1656 if_unroute_dispatch(netmsg_t nmsg)
1657 {
1658 	struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1659 	struct ifnet *ifp = msg->ifp;
1660 	int flag = msg->flag, fam = msg->fam;
1661 	struct ifaddr_container *ifac;
1662 
1663 	ASSERT_NETISR0;
1664 
1665 	ifp->if_flags &= ~flag;
1666 	getmicrotime(&ifp->if_lastchange);
1667 	/*
1668 	 * The ifaddr processing in the following loop will block,
1669 	 * however, this function is called in netisr0, in which
1670 	 * ifaddr list changes happen, so we don't care about the
1671 	 * blockness of the ifaddr processing here.
1672 	 */
1673 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1674 		struct ifaddr *ifa = ifac->ifa;
1675 
1676 		/* Ignore marker */
1677 		if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1678 			continue;
1679 
1680 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1681 			kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1682 	}
1683 	ifq_purge_all(&ifp->if_snd);
1684 	rt_ifmsg(ifp);
1685 
1686 	netisr_replymsg(&nmsg->base, 0);
1687 }
1688 
1689 void
1690 if_unroute(struct ifnet *ifp, int flag, int fam)
1691 {
1692 	struct netmsg_ifroute msg;
1693 
1694 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1695 	    if_unroute_dispatch);
1696 	msg.ifp = ifp;
1697 	msg.flag = flag;
1698 	msg.fam = fam;
1699 	netisr_domsg(&msg.base, 0);
1700 }
1701 
1702 /*
1703  * Mark an interface up and notify protocols of the transition.
1704  */
1705 static void
1706 if_route_dispatch(netmsg_t nmsg)
1707 {
1708 	struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1709 	struct ifnet *ifp = msg->ifp;
1710 	int flag = msg->flag, fam = msg->fam;
1711 	struct ifaddr_container *ifac;
1712 
1713 	ASSERT_NETISR0;
1714 
1715 	ifq_purge_all(&ifp->if_snd);
1716 	ifp->if_flags |= flag;
1717 	getmicrotime(&ifp->if_lastchange);
1718 	/*
1719 	 * The ifaddr processing in the following loop will block,
1720 	 * however, this function is called in netisr0, in which
1721 	 * ifaddr list changes happen, so we don't care about the
1722 	 * blockness of the ifaddr processing here.
1723 	 */
1724 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1725 		struct ifaddr *ifa = ifac->ifa;
1726 
1727 		/* Ignore marker */
1728 		if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1729 			continue;
1730 
1731 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1732 			kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1733 	}
1734 	rt_ifmsg(ifp);
1735 #ifdef INET6
1736 	in6_if_up(ifp);
1737 #endif
1738 
1739 	netisr_replymsg(&nmsg->base, 0);
1740 }
1741 
1742 void
1743 if_route(struct ifnet *ifp, int flag, int fam)
1744 {
1745 	struct netmsg_ifroute msg;
1746 
1747 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1748 	    if_route_dispatch);
1749 	msg.ifp = ifp;
1750 	msg.flag = flag;
1751 	msg.fam = fam;
1752 	netisr_domsg(&msg.base, 0);
1753 }
1754 
1755 /*
1756  * Mark an interface down and notify protocols of the transition.  An
1757  * interface going down is also considered to be a synchronizing event.
1758  * We must ensure that all packet processing related to the interface
1759  * has completed before we return so e.g. the caller can free the ifnet
1760  * structure that the mbufs may be referencing.
1761  *
1762  * NOTE: must be called at splnet or eqivalent.
1763  */
1764 void
1765 if_down(struct ifnet *ifp)
1766 {
1767 	EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_DOWN);
1768 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1769 	netmsg_service_sync();
1770 }
1771 
1772 /*
1773  * Mark an interface up and notify protocols of
1774  * the transition.
1775  * NOTE: must be called at splnet or eqivalent.
1776  */
1777 void
1778 if_up(struct ifnet *ifp)
1779 {
1780 	if_route(ifp, IFF_UP, AF_UNSPEC);
1781 	EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_UP);
1782 }
1783 
1784 /*
1785  * Process a link state change.
1786  * NOTE: must be called at splsoftnet or equivalent.
1787  */
1788 void
1789 if_link_state_change(struct ifnet *ifp)
1790 {
1791 	int link_state = ifp->if_link_state;
1792 
1793 	rt_ifmsg(ifp);
1794 	devctl_notify("IFNET", ifp->if_xname,
1795 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1796 
1797 	EVENTHANDLER_INVOKE(ifnet_link_event, ifp, link_state);
1798 }
1799 
1800 /*
1801  * Handle interface watchdog timer routines.  Called
1802  * from softclock, we decrement timers (if set) and
1803  * call the appropriate interface routine on expiration.
1804  */
1805 static void
1806 if_slowtimo_dispatch(netmsg_t nmsg)
1807 {
1808 	struct globaldata *gd = mycpu;
1809 	const struct ifnet_array *arr;
1810 	int i;
1811 
1812 	ASSERT_NETISR0;
1813 
1814 	crit_enter_gd(gd);
1815 	lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1816 	crit_exit_gd(gd);
1817 
1818 	arr = ifnet_array_get();
1819 	for (i = 0; i < arr->ifnet_count; ++i) {
1820 		struct ifnet *ifp = arr->ifnet_arr[i];
1821 
1822 		crit_enter_gd(gd);
1823 
1824 		if (if_stats_compat) {
1825 			IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1826 			IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1827 			IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1828 			IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1829 			IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1830 			IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1831 			IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1832 			IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1833 			IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1834 			IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1835 			IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1836 			IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
1837 		}
1838 
1839 		if (ifp->if_timer == 0 || --ifp->if_timer) {
1840 			crit_exit_gd(gd);
1841 			continue;
1842 		}
1843 		if (ifp->if_watchdog) {
1844 			if (ifnet_tryserialize_all(ifp)) {
1845 				(*ifp->if_watchdog)(ifp);
1846 				ifnet_deserialize_all(ifp);
1847 			} else {
1848 				/* try again next timeout */
1849 				++ifp->if_timer;
1850 			}
1851 		}
1852 
1853 		crit_exit_gd(gd);
1854 	}
1855 
1856 	callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1857 }
1858 
1859 static void
1860 if_slowtimo(void *arg __unused)
1861 {
1862 	struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1863 
1864 	KASSERT(mycpuid == 0, ("not on cpu0"));
1865 	crit_enter();
1866 	if (lmsg->ms_flags & MSGF_DONE)
1867 		lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1868 	crit_exit();
1869 }
1870 
1871 /*
1872  * Map interface name to
1873  * interface structure pointer.
1874  */
1875 struct ifnet *
1876 ifunit(const char *name)
1877 {
1878 	struct ifnet *ifp;
1879 
1880 	/*
1881 	 * Search all the interfaces for this name/number
1882 	 */
1883 	KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1884 
1885 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1886 		if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1887 			break;
1888 	}
1889 	return (ifp);
1890 }
1891 
1892 struct ifnet *
1893 ifunit_netisr(const char *name)
1894 {
1895 	const struct ifnet_array *arr;
1896 	int i;
1897 
1898 	/*
1899 	 * Search all the interfaces for this name/number
1900 	 */
1901 
1902 	arr = ifnet_array_get();
1903 	for (i = 0; i < arr->ifnet_count; ++i) {
1904 		struct ifnet *ifp = arr->ifnet_arr[i];
1905 
1906 		if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1907 			return ifp;
1908 	}
1909 	return NULL;
1910 }
1911 
1912 /*
1913  * Interface ioctls.
1914  */
1915 int
1916 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1917 {
1918 	struct ifnet *ifp;
1919 	struct ifgroupreq *ifgr;
1920 	struct ifreq *ifr;
1921 	struct ifstat *ifs;
1922 	int error, do_ifup = 0;
1923 	short oif_flags;
1924 	int new_flags;
1925 	size_t namelen, onamelen;
1926 	char new_name[IFNAMSIZ];
1927 	struct ifaddr *ifa;
1928 	struct sockaddr_dl *sdl;
1929 
1930 	switch (cmd) {
1931 	case SIOCGIFCONF:
1932 		return (ifconf(cmd, data, cred));
1933 	default:
1934 		break;
1935 	}
1936 
1937 	ifr = (struct ifreq *)data;
1938 
1939 	switch (cmd) {
1940 	case SIOCIFCREATE:
1941 	case SIOCIFCREATE2:
1942 		if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1943 			return (error);
1944 		return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1945 			cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1946 	case SIOCIFDESTROY:
1947 		if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1948 			return (error);
1949 		return (if_clone_destroy(ifr->ifr_name));
1950 	case SIOCIFGCLONERS:
1951 		return (if_clone_list((struct if_clonereq *)data));
1952 	case SIOCGIFGMEMB:
1953 		return (if_getgroupmembers((struct ifgroupreq *)data));
1954 	default:
1955 		break;
1956 	}
1957 
1958 	/*
1959 	 * Nominal ioctl through interface, lookup the ifp and obtain a
1960 	 * lock to serialize the ifconfig ioctl operation.
1961 	 */
1962 	ifnet_lock();
1963 
1964 	ifp = ifunit(ifr->ifr_name);
1965 	if (ifp == NULL) {
1966 		ifnet_unlock();
1967 		return (ENXIO);
1968 	}
1969 	error = 0;
1970 
1971 	switch (cmd) {
1972 	case SIOCGIFINDEX:
1973 		ifr->ifr_index = ifp->if_index;
1974 		break;
1975 
1976 	case SIOCGIFFLAGS:
1977 		ifr->ifr_flags = ifp->if_flags;
1978 		ifr->ifr_flagshigh = ifp->if_flags >> 16;
1979 		break;
1980 
1981 	case SIOCGIFCAP:
1982 		ifr->ifr_reqcap = ifp->if_capabilities;
1983 		ifr->ifr_curcap = ifp->if_capenable;
1984 		break;
1985 
1986 	case SIOCGIFMETRIC:
1987 		ifr->ifr_metric = ifp->if_metric;
1988 		break;
1989 
1990 	case SIOCGIFMTU:
1991 		ifr->ifr_mtu = ifp->if_mtu;
1992 		break;
1993 
1994 	case SIOCGIFTSOLEN:
1995 		ifr->ifr_tsolen = ifp->if_tsolen;
1996 		break;
1997 
1998 	case SIOCGIFDATA:
1999 		error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
2000 				sizeof(ifp->if_data));
2001 		break;
2002 
2003 	case SIOCGIFPHYS:
2004 		ifr->ifr_phys = ifp->if_physical;
2005 		break;
2006 
2007 	case SIOCGIFPOLLCPU:
2008 		ifr->ifr_pollcpu = -1;
2009 		break;
2010 
2011 	case SIOCSIFPOLLCPU:
2012 		break;
2013 
2014 	case SIOCSIFFLAGS:
2015 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2016 		if (error)
2017 			break;
2018 		new_flags = (ifr->ifr_flags & 0xffff) |
2019 		    (ifr->ifr_flagshigh << 16);
2020 		if (ifp->if_flags & IFF_SMART) {
2021 			/* Smart drivers twiddle their own routes */
2022 		} else if (ifp->if_flags & IFF_UP &&
2023 		    (new_flags & IFF_UP) == 0) {
2024 			if_down(ifp);
2025 		} else if (new_flags & IFF_UP &&
2026 		    (ifp->if_flags & IFF_UP) == 0) {
2027 			do_ifup = 1;
2028 		}
2029 
2030 #ifdef IFPOLL_ENABLE
2031 		if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
2032 			if (new_flags & IFF_NPOLLING)
2033 				ifpoll_register(ifp);
2034 			else
2035 				ifpoll_deregister(ifp);
2036 		}
2037 #endif
2038 
2039 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2040 			(new_flags &~ IFF_CANTCHANGE);
2041 		if (new_flags & IFF_PPROMISC) {
2042 			/* Permanently promiscuous mode requested */
2043 			ifp->if_flags |= IFF_PROMISC;
2044 		} else if (ifp->if_pcount == 0) {
2045 			ifp->if_flags &= ~IFF_PROMISC;
2046 		}
2047 		if (ifp->if_ioctl) {
2048 			ifnet_serialize_all(ifp);
2049 			ifp->if_ioctl(ifp, cmd, data, cred);
2050 			ifnet_deserialize_all(ifp);
2051 		}
2052 		if (do_ifup)
2053 			if_up(ifp);
2054 		getmicrotime(&ifp->if_lastchange);
2055 		break;
2056 
2057 	case SIOCSIFCAP:
2058 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2059 		if (error)
2060 			break;
2061 		if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
2062 			error = EINVAL;
2063 			break;
2064 		}
2065 		ifnet_serialize_all(ifp);
2066 		ifp->if_ioctl(ifp, cmd, data, cred);
2067 		ifnet_deserialize_all(ifp);
2068 		break;
2069 
2070 	case SIOCSIFNAME:
2071 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2072 		if (error)
2073 			break;
2074 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2075 		if (error)
2076 			break;
2077 		if (new_name[0] == '\0') {
2078 			error = EINVAL;
2079 			break;
2080 		}
2081 		if (ifunit(new_name) != NULL) {
2082 			error = EEXIST;
2083 			break;
2084 		}
2085 
2086 		EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
2087 
2088 		/* Announce the departure of the interface. */
2089 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2090 
2091 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2092 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
2093 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2094 		namelen = strlen(new_name);
2095 		onamelen = sdl->sdl_nlen;
2096 		/*
2097 		 * Move the address if needed.  This is safe because we
2098 		 * allocate space for a name of length IFNAMSIZ when we
2099 		 * create this in if_attach().
2100 		 */
2101 		if (namelen != onamelen) {
2102 			bcopy(sdl->sdl_data + onamelen,
2103 			    sdl->sdl_data + namelen, sdl->sdl_alen);
2104 		}
2105 		bcopy(new_name, sdl->sdl_data, namelen);
2106 		sdl->sdl_nlen = namelen;
2107 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2108 		bzero(sdl->sdl_data, onamelen);
2109 		while (namelen != 0)
2110 			sdl->sdl_data[--namelen] = 0xff;
2111 
2112 		EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
2113 
2114 		/* Announce the return of the interface. */
2115 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2116 		break;
2117 
2118 	case SIOCSIFMETRIC:
2119 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2120 		if (error)
2121 			break;
2122 		ifp->if_metric = ifr->ifr_metric;
2123 		getmicrotime(&ifp->if_lastchange);
2124 		break;
2125 
2126 	case SIOCSIFPHYS:
2127 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2128 		if (error)
2129 			break;
2130 		if (ifp->if_ioctl == NULL) {
2131 		        error = EOPNOTSUPP;
2132 			break;
2133 		}
2134 		ifnet_serialize_all(ifp);
2135 		error = ifp->if_ioctl(ifp, cmd, data, cred);
2136 		ifnet_deserialize_all(ifp);
2137 		if (error == 0)
2138 			getmicrotime(&ifp->if_lastchange);
2139 		break;
2140 
2141 	case SIOCSIFMTU:
2142 	{
2143 		u_long oldmtu = ifp->if_mtu;
2144 
2145 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2146 		if (error)
2147 			break;
2148 		if (ifp->if_ioctl == NULL) {
2149 			error = EOPNOTSUPP;
2150 			break;
2151 		}
2152 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2153 			error = EINVAL;
2154 			break;
2155 		}
2156 		ifnet_serialize_all(ifp);
2157 		error = ifp->if_ioctl(ifp, cmd, data, cred);
2158 		ifnet_deserialize_all(ifp);
2159 		if (error == 0) {
2160 			getmicrotime(&ifp->if_lastchange);
2161 			rt_ifmsg(ifp);
2162 		}
2163 		/*
2164 		 * If the link MTU changed, do network layer specific procedure.
2165 		 */
2166 		if (ifp->if_mtu != oldmtu) {
2167 #ifdef INET6
2168 			nd6_setmtu(ifp);
2169 #endif
2170 		}
2171 		break;
2172 	}
2173 
2174 	case SIOCSIFTSOLEN:
2175 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2176 		if (error)
2177 			break;
2178 
2179 		/* XXX need driver supplied upper limit */
2180 		if (ifr->ifr_tsolen <= 0) {
2181 			error = EINVAL;
2182 			break;
2183 		}
2184 		ifp->if_tsolen = ifr->ifr_tsolen;
2185 		break;
2186 
2187 	case SIOCADDMULTI:
2188 	case SIOCDELMULTI:
2189 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2190 		if (error)
2191 			break;
2192 
2193 		/* Don't allow group membership on non-multicast interfaces. */
2194 		if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2195 			error = EOPNOTSUPP;
2196 			break;
2197 		}
2198 
2199 		/* Don't let users screw up protocols' entries. */
2200 		if (ifr->ifr_addr.sa_family != AF_LINK) {
2201 			error = EINVAL;
2202 			break;
2203 		}
2204 
2205 		if (cmd == SIOCADDMULTI) {
2206 			struct ifmultiaddr *ifma;
2207 			error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2208 		} else {
2209 			error = if_delmulti(ifp, &ifr->ifr_addr);
2210 		}
2211 		if (error == 0)
2212 			getmicrotime(&ifp->if_lastchange);
2213 		break;
2214 
2215 	case SIOCSIFPHYADDR:
2216 	case SIOCDIFPHYADDR:
2217 #ifdef INET6
2218 	case SIOCSIFPHYADDR_IN6:
2219 #endif
2220 	case SIOCSLIFPHYADDR:
2221 	case SIOCSIFMEDIA:
2222 	case SIOCSIFGENERIC:
2223 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2224 		if (error)
2225 			break;
2226 		if (ifp->if_ioctl == 0) {
2227 			error = EOPNOTSUPP;
2228 			break;
2229 		}
2230 		ifnet_serialize_all(ifp);
2231 		error = ifp->if_ioctl(ifp, cmd, data, cred);
2232 		ifnet_deserialize_all(ifp);
2233 		if (error == 0)
2234 			getmicrotime(&ifp->if_lastchange);
2235 		break;
2236 
2237 	case SIOCGIFSTATUS:
2238 		ifs = (struct ifstat *)data;
2239 		ifs->ascii[0] = '\0';
2240 		/* fall through */
2241 	case SIOCGIFPSRCADDR:
2242 	case SIOCGIFPDSTADDR:
2243 	case SIOCGLIFPHYADDR:
2244 	case SIOCGIFMEDIA:
2245 	case SIOCGIFGENERIC:
2246 		if (ifp->if_ioctl == NULL) {
2247 			error = EOPNOTSUPP;
2248 			break;
2249 		}
2250 		ifnet_serialize_all(ifp);
2251 		error = ifp->if_ioctl(ifp, cmd, data, cred);
2252 		ifnet_deserialize_all(ifp);
2253 		break;
2254 
2255 	case SIOCSIFLLADDR:
2256 		error = priv_check_cred(cred, PRIV_ROOT, 0);
2257 		if (error)
2258 			break;
2259 		error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2260 				     ifr->ifr_addr.sa_len);
2261 		EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2262 		break;
2263 
2264 	case SIOCAIFGROUP:
2265 		ifgr = (struct ifgroupreq *)ifr;
2266 		if ((error = priv_check_cred(cred, PRIV_NET_ADDIFGROUP, 0)))
2267 			return (error);
2268 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2269 			return (error);
2270 		break;
2271 
2272 	case SIOCDIFGROUP:
2273 		ifgr = (struct ifgroupreq *)ifr;
2274 		if ((error = priv_check_cred(cred, PRIV_NET_DELIFGROUP, 0)))
2275 			return (error);
2276 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2277 			return (error);
2278 		break;
2279 
2280 	case SIOCGIFGROUP:
2281 		ifgr = (struct ifgroupreq *)ifr;
2282 		if ((error = if_getgroups(ifgr, ifp)))
2283 			return (error);
2284 		break;
2285 
2286 	default:
2287 		oif_flags = ifp->if_flags;
2288 		if (so->so_proto == 0) {
2289 			error = EOPNOTSUPP;
2290 			break;
2291 		}
2292 		error = so_pru_control_direct(so, cmd, data, ifp);
2293 
2294 		if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2295 #ifdef INET6
2296 			DELAY(100);/* XXX: temporary workaround for fxp issue*/
2297 			if (ifp->if_flags & IFF_UP) {
2298 				crit_enter();
2299 				in6_if_up(ifp);
2300 				crit_exit();
2301 			}
2302 #endif
2303 		}
2304 		break;
2305 	}
2306 
2307 	ifnet_unlock();
2308 	return (error);
2309 }
2310 
2311 /*
2312  * Set/clear promiscuous mode on interface ifp based on the truth value
2313  * of pswitch.  The calls are reference counted so that only the first
2314  * "on" request actually has an effect, as does the final "off" request.
2315  * Results are undefined if the "off" and "on" requests are not matched.
2316  */
2317 int
2318 ifpromisc(struct ifnet *ifp, int pswitch)
2319 {
2320 	struct ifreq ifr;
2321 	int error;
2322 	int oldflags;
2323 
2324 	oldflags = ifp->if_flags;
2325 	if (ifp->if_flags & IFF_PPROMISC) {
2326 		/* Do nothing if device is in permanently promiscuous mode */
2327 		ifp->if_pcount += pswitch ? 1 : -1;
2328 		return (0);
2329 	}
2330 	if (pswitch) {
2331 		/*
2332 		 * If the device is not configured up, we cannot put it in
2333 		 * promiscuous mode.
2334 		 */
2335 		if ((ifp->if_flags & IFF_UP) == 0)
2336 			return (ENETDOWN);
2337 		if (ifp->if_pcount++ != 0)
2338 			return (0);
2339 		ifp->if_flags |= IFF_PROMISC;
2340 		log(LOG_INFO, "%s: promiscuous mode enabled\n",
2341 		    ifp->if_xname);
2342 	} else {
2343 		if (--ifp->if_pcount > 0)
2344 			return (0);
2345 		ifp->if_flags &= ~IFF_PROMISC;
2346 		log(LOG_INFO, "%s: promiscuous mode disabled\n",
2347 		    ifp->if_xname);
2348 	}
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, NULL);
2353 	ifnet_deserialize_all(ifp);
2354 	if (error == 0)
2355 		rt_ifmsg(ifp);
2356 	else
2357 		ifp->if_flags = oldflags;
2358 	return error;
2359 }
2360 
2361 /*
2362  * Return interface configuration
2363  * of system.  List may be used
2364  * in later ioctl's (above) to get
2365  * other information.
2366  */
2367 static int
2368 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2369 {
2370 	struct ifconf *ifc = (struct ifconf *)data;
2371 	struct ifnet *ifp;
2372 	struct sockaddr *sa;
2373 	struct ifreq ifr, *ifrp;
2374 	int space = ifc->ifc_len, error = 0;
2375 
2376 	ifrp = ifc->ifc_req;
2377 
2378 	ifnet_lock();
2379 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2380 		struct ifaddr_container *ifac, *ifac_mark;
2381 		struct ifaddr_marker mark;
2382 		struct ifaddrhead *head;
2383 		int addrs;
2384 
2385 		if (space <= sizeof ifr)
2386 			break;
2387 
2388 		/*
2389 		 * Zero the stack declared structure first to prevent
2390 		 * memory disclosure.
2391 		 */
2392 		bzero(&ifr, sizeof(ifr));
2393 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2394 		    >= sizeof(ifr.ifr_name)) {
2395 			error = ENAMETOOLONG;
2396 			break;
2397 		}
2398 
2399 		/*
2400 		 * Add a marker, since copyout() could block and during that
2401 		 * period the list could be changed.  Inserting the marker to
2402 		 * the header of the list will not cause trouble for the code
2403 		 * assuming that the first element of the list is AF_LINK; the
2404 		 * marker will be moved to the next position w/o blocking.
2405 		 */
2406 		ifa_marker_init(&mark, ifp);
2407 		ifac_mark = &mark.ifac;
2408 		head = &ifp->if_addrheads[mycpuid];
2409 
2410 		addrs = 0;
2411 		TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2412 		while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2413 			struct ifaddr *ifa = ifac->ifa;
2414 
2415 			TAILQ_REMOVE(head, ifac_mark, ifa_link);
2416 			TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2417 
2418 			/* Ignore marker */
2419 			if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2420 				continue;
2421 
2422 			if (space <= sizeof ifr)
2423 				break;
2424 			sa = ifa->ifa_addr;
2425 			if (cred->cr_prison &&
2426 			    prison_if(cred, sa))
2427 				continue;
2428 			addrs++;
2429 			/*
2430 			 * Keep a reference on this ifaddr, so that it will
2431 			 * not be destroyed when its address is copied to
2432 			 * the userland, which could block.
2433 			 */
2434 			IFAREF(ifa);
2435 			if (sa->sa_len <= sizeof(*sa)) {
2436 				ifr.ifr_addr = *sa;
2437 				error = copyout(&ifr, ifrp, sizeof ifr);
2438 				ifrp++;
2439 			} else {
2440 				if (space < (sizeof ifr) + sa->sa_len -
2441 					    sizeof(*sa)) {
2442 					IFAFREE(ifa);
2443 					break;
2444 				}
2445 				space -= sa->sa_len - sizeof(*sa);
2446 				error = copyout(&ifr, ifrp,
2447 						sizeof ifr.ifr_name);
2448 				if (error == 0)
2449 					error = copyout(sa, &ifrp->ifr_addr,
2450 							sa->sa_len);
2451 				ifrp = (struct ifreq *)
2452 					(sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2453 			}
2454 			IFAFREE(ifa);
2455 			if (error)
2456 				break;
2457 			space -= sizeof ifr;
2458 		}
2459 		TAILQ_REMOVE(head, ifac_mark, ifa_link);
2460 		if (error)
2461 			break;
2462 		if (!addrs) {
2463 			bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2464 			error = copyout(&ifr, ifrp, sizeof ifr);
2465 			if (error)
2466 				break;
2467 			space -= sizeof ifr;
2468 			ifrp++;
2469 		}
2470 	}
2471 	ifnet_unlock();
2472 
2473 	ifc->ifc_len -= space;
2474 	return (error);
2475 }
2476 
2477 /*
2478  * Just like if_promisc(), but for all-multicast-reception mode.
2479  */
2480 int
2481 if_allmulti(struct ifnet *ifp, int onswitch)
2482 {
2483 	int error = 0;
2484 	struct ifreq ifr;
2485 
2486 	crit_enter();
2487 
2488 	if (onswitch) {
2489 		if (ifp->if_amcount++ == 0) {
2490 			ifp->if_flags |= IFF_ALLMULTI;
2491 			ifr.ifr_flags = ifp->if_flags;
2492 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2493 			ifnet_serialize_all(ifp);
2494 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2495 					      NULL);
2496 			ifnet_deserialize_all(ifp);
2497 		}
2498 	} else {
2499 		if (ifp->if_amcount > 1) {
2500 			ifp->if_amcount--;
2501 		} else {
2502 			ifp->if_amcount = 0;
2503 			ifp->if_flags &= ~IFF_ALLMULTI;
2504 			ifr.ifr_flags = ifp->if_flags;
2505 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2506 			ifnet_serialize_all(ifp);
2507 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2508 					      NULL);
2509 			ifnet_deserialize_all(ifp);
2510 		}
2511 	}
2512 
2513 	crit_exit();
2514 
2515 	if (error == 0)
2516 		rt_ifmsg(ifp);
2517 	return error;
2518 }
2519 
2520 /*
2521  * Add a multicast listenership to the interface in question.
2522  * The link layer provides a routine which converts
2523  */
2524 int
2525 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2526     struct ifmultiaddr **retifma)
2527 {
2528 	struct sockaddr *llsa, *dupsa;
2529 	int error;
2530 	struct ifmultiaddr *ifma;
2531 
2532 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
2533 
2534 	/*
2535 	 * If the matching multicast address already exists
2536 	 * then don't add a new one, just add a reference
2537 	 */
2538 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2539 		if (sa_equal(sa, ifma->ifma_addr)) {
2540 			ifma->ifma_refcount++;
2541 			if (retifma)
2542 				*retifma = ifma;
2543 			return 0;
2544 		}
2545 	}
2546 
2547 	/*
2548 	 * Give the link layer a chance to accept/reject it, and also
2549 	 * find out which AF_LINK address this maps to, if it isn't one
2550 	 * already.
2551 	 */
2552 	if (ifp->if_resolvemulti) {
2553 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2554 		if (error)
2555 			return error;
2556 	} else {
2557 		llsa = NULL;
2558 	}
2559 
2560 	ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2561 	dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2562 	bcopy(sa, dupsa, sa->sa_len);
2563 
2564 	ifma->ifma_addr = dupsa;
2565 	ifma->ifma_lladdr = llsa;
2566 	ifma->ifma_ifp = ifp;
2567 	ifma->ifma_refcount = 1;
2568 	ifma->ifma_protospec = NULL;
2569 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2570 
2571 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2572 	if (retifma)
2573 		*retifma = ifma;
2574 
2575 	if (llsa != NULL) {
2576 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2577 			if (sa_equal(ifma->ifma_addr, llsa))
2578 				break;
2579 		}
2580 		if (ifma) {
2581 			ifma->ifma_refcount++;
2582 		} else {
2583 			ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2584 			dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2585 			bcopy(llsa, dupsa, llsa->sa_len);
2586 			ifma->ifma_addr = dupsa;
2587 			ifma->ifma_ifp = ifp;
2588 			ifma->ifma_refcount = 1;
2589 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2590 		}
2591 	}
2592 	/*
2593 	 * We are certain we have added something, so call down to the
2594 	 * interface to let them know about it.
2595 	 */
2596 	if (ifp->if_ioctl)
2597 		ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2598 
2599 	return 0;
2600 }
2601 
2602 int
2603 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2604     struct ifmultiaddr **retifma)
2605 {
2606 	int error;
2607 
2608 	ifnet_serialize_all(ifp);
2609 	error = if_addmulti_serialized(ifp, sa, retifma);
2610 	ifnet_deserialize_all(ifp);
2611 
2612 	return error;
2613 }
2614 
2615 /*
2616  * Remove a reference to a multicast address on this interface.  Yell
2617  * if the request does not match an existing membership.
2618  */
2619 static int
2620 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2621 {
2622 	struct ifmultiaddr *ifma;
2623 
2624 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
2625 
2626 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2627 		if (sa_equal(sa, ifma->ifma_addr))
2628 			break;
2629 	if (ifma == NULL)
2630 		return ENOENT;
2631 
2632 	if (ifma->ifma_refcount > 1) {
2633 		ifma->ifma_refcount--;
2634 		return 0;
2635 	}
2636 
2637 	rt_newmaddrmsg(RTM_DELMADDR, ifma);
2638 	sa = ifma->ifma_lladdr;
2639 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2640 	/*
2641 	 * Make sure the interface driver is notified
2642 	 * in the case of a link layer mcast group being left.
2643 	 */
2644 	if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2645 		ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2646 	kfree(ifma->ifma_addr, M_IFMADDR);
2647 	kfree(ifma, M_IFMADDR);
2648 	if (sa == NULL)
2649 		return 0;
2650 
2651 	/*
2652 	 * Now look for the link-layer address which corresponds to
2653 	 * this network address.  It had been squirreled away in
2654 	 * ifma->ifma_lladdr for this purpose (so we don't have
2655 	 * to call ifp->if_resolvemulti() again), and we saved that
2656 	 * value in sa above.  If some nasty deleted the
2657 	 * link-layer address out from underneath us, we can deal because
2658 	 * the address we stored was is not the same as the one which was
2659 	 * in the record for the link-layer address.  (So we don't complain
2660 	 * in that case.)
2661 	 */
2662 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2663 		if (sa_equal(sa, ifma->ifma_addr))
2664 			break;
2665 	if (ifma == NULL)
2666 		return 0;
2667 
2668 	if (ifma->ifma_refcount > 1) {
2669 		ifma->ifma_refcount--;
2670 		return 0;
2671 	}
2672 
2673 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2674 	ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2675 	kfree(ifma->ifma_addr, M_IFMADDR);
2676 	kfree(sa, M_IFMADDR);
2677 	kfree(ifma, M_IFMADDR);
2678 
2679 	return 0;
2680 }
2681 
2682 int
2683 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2684 {
2685 	int error;
2686 
2687 	ifnet_serialize_all(ifp);
2688 	error = if_delmulti_serialized(ifp, sa);
2689 	ifnet_deserialize_all(ifp);
2690 
2691 	return error;
2692 }
2693 
2694 /*
2695  * Delete all multicast group membership for an interface.
2696  * Should be used to quickly flush all multicast filters.
2697  */
2698 void
2699 if_delallmulti_serialized(struct ifnet *ifp)
2700 {
2701 	struct ifmultiaddr *ifma, mark;
2702 	struct sockaddr sa;
2703 
2704 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
2705 
2706 	bzero(&sa, sizeof(sa));
2707 	sa.sa_family = AF_UNSPEC;
2708 	sa.sa_len = sizeof(sa);
2709 
2710 	bzero(&mark, sizeof(mark));
2711 	mark.ifma_addr = &sa;
2712 
2713 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2714 	while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2715 		TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2716 		TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2717 		    ifma_link);
2718 
2719 		if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2720 			continue;
2721 
2722 		if_delmulti_serialized(ifp, ifma->ifma_addr);
2723 	}
2724 	TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2725 }
2726 
2727 
2728 /*
2729  * Set the link layer address on an interface.
2730  *
2731  * At this time we only support certain types of interfaces,
2732  * and we don't allow the length of the address to change.
2733  */
2734 int
2735 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2736 {
2737 	struct sockaddr_dl *sdl;
2738 	struct ifreq ifr;
2739 
2740 	sdl = IF_LLSOCKADDR(ifp);
2741 	if (sdl == NULL)
2742 		return (EINVAL);
2743 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2744 		return (EINVAL);
2745 	switch (ifp->if_type) {
2746 	case IFT_ETHER:			/* these types use struct arpcom */
2747 	case IFT_XETHER:
2748 	case IFT_L2VLAN:
2749 	case IFT_IEEE8023ADLAG:
2750 		bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2751 		bcopy(lladdr, LLADDR(sdl), len);
2752 		break;
2753 	default:
2754 		return (ENODEV);
2755 	}
2756 	/*
2757 	 * If the interface is already up, we need
2758 	 * to re-init it in order to reprogram its
2759 	 * address filter.
2760 	 */
2761 	ifnet_serialize_all(ifp);
2762 	if ((ifp->if_flags & IFF_UP) != 0) {
2763 #ifdef INET
2764 		struct ifaddr_container *ifac;
2765 #endif
2766 
2767 		ifp->if_flags &= ~IFF_UP;
2768 		ifr.ifr_flags = ifp->if_flags;
2769 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
2770 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2771 			      NULL);
2772 		ifp->if_flags |= IFF_UP;
2773 		ifr.ifr_flags = ifp->if_flags;
2774 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
2775 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2776 				 NULL);
2777 #ifdef INET
2778 		/*
2779 		 * Also send gratuitous ARPs to notify other nodes about
2780 		 * the address change.
2781 		 */
2782 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2783 			struct ifaddr *ifa = ifac->ifa;
2784 
2785 			if (ifa->ifa_addr != NULL &&
2786 			    ifa->ifa_addr->sa_family == AF_INET)
2787 				arp_gratuitous(ifp, ifa);
2788 		}
2789 #endif
2790 	}
2791 	ifnet_deserialize_all(ifp);
2792 	return (0);
2793 }
2794 
2795 struct ifmultiaddr *
2796 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2797 {
2798 	struct ifmultiaddr *ifma;
2799 
2800 	/* TODO: need ifnet_serialize_main */
2801 	ifnet_serialize_all(ifp);
2802 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2803 		if (sa_equal(ifma->ifma_addr, sa))
2804 			break;
2805 	ifnet_deserialize_all(ifp);
2806 
2807 	return ifma;
2808 }
2809 
2810 /*
2811  * This function locates the first real ethernet MAC from a network
2812  * card and loads it into node, returning 0 on success or ENOENT if
2813  * no suitable interfaces were found.  It is used by the uuid code to
2814  * generate a unique 6-byte number.
2815  */
2816 int
2817 if_getanyethermac(uint16_t *node, int minlen)
2818 {
2819 	struct ifnet *ifp;
2820 	struct sockaddr_dl *sdl;
2821 
2822 	ifnet_lock();
2823 	TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2824 		if (ifp->if_type != IFT_ETHER)
2825 			continue;
2826 		sdl = IF_LLSOCKADDR(ifp);
2827 		if (sdl->sdl_alen < minlen)
2828 			continue;
2829 		bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2830 		      minlen);
2831 		ifnet_unlock();
2832 		return(0);
2833 	}
2834 	ifnet_unlock();
2835 	return (ENOENT);
2836 }
2837 
2838 /*
2839  * The name argument must be a pointer to storage which will last as
2840  * long as the interface does.  For physical devices, the result of
2841  * device_get_name(dev) is a good choice and for pseudo-devices a
2842  * static string works well.
2843  */
2844 void
2845 if_initname(struct ifnet *ifp, const char *name, int unit)
2846 {
2847 	ifp->if_dname = name;
2848 	ifp->if_dunit = unit;
2849 	if (unit != IF_DUNIT_NONE)
2850 		ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2851 	else
2852 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2853 }
2854 
2855 int
2856 if_printf(struct ifnet *ifp, const char *fmt, ...)
2857 {
2858 	__va_list ap;
2859 	int retval;
2860 
2861 	retval = kprintf("%s: ", ifp->if_xname);
2862 	__va_start(ap, fmt);
2863 	retval += kvprintf(fmt, ap);
2864 	__va_end(ap);
2865 	return (retval);
2866 }
2867 
2868 struct ifnet *
2869 if_alloc(uint8_t type)
2870 {
2871 	struct ifnet *ifp;
2872 	size_t size;
2873 
2874 	/*
2875 	 * XXX temporary hack until arpcom is setup in if_l2com
2876 	 */
2877 	if (type == IFT_ETHER)
2878 		size = sizeof(struct arpcom);
2879 	else
2880 		size = sizeof(struct ifnet);
2881 
2882 	ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2883 
2884 	ifp->if_type = type;
2885 
2886 	if (if_com_alloc[type] != NULL) {
2887 		ifp->if_l2com = if_com_alloc[type](type, ifp);
2888 		if (ifp->if_l2com == NULL) {
2889 			kfree(ifp, M_IFNET);
2890 			return (NULL);
2891 		}
2892 	}
2893 	return (ifp);
2894 }
2895 
2896 void
2897 if_free(struct ifnet *ifp)
2898 {
2899 	kfree(ifp, M_IFNET);
2900 }
2901 
2902 void
2903 ifq_set_classic(struct ifaltq *ifq)
2904 {
2905 	ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2906 	    ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2907 }
2908 
2909 void
2910 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2911     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2912 {
2913 	int q;
2914 
2915 	KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2916 	KASSERT(enqueue != NULL, ("enqueue is not specified"));
2917 	KASSERT(dequeue != NULL, ("dequeue is not specified"));
2918 	KASSERT(request != NULL, ("request is not specified"));
2919 
2920 	ifq->altq_mapsubq = mapsubq;
2921 	for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2922 		struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2923 
2924 		ifsq->ifsq_enqueue = enqueue;
2925 		ifsq->ifsq_dequeue = dequeue;
2926 		ifsq->ifsq_request = request;
2927 	}
2928 }
2929 
2930 static void
2931 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2932 {
2933 
2934 	classq_add(&ifsq->ifsq_norm, m);
2935 	ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2936 }
2937 
2938 static void
2939 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2940 {
2941 
2942 	classq_add(&ifsq->ifsq_prio, m);
2943 	ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2944 	ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2945 }
2946 
2947 static struct mbuf *
2948 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2949 {
2950 	struct mbuf *m;
2951 
2952 	m = classq_get(&ifsq->ifsq_norm);
2953 	if (m != NULL)
2954 		ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2955 	return (m);
2956 }
2957 
2958 static struct mbuf *
2959 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2960 {
2961 	struct mbuf *m;
2962 
2963 	m = classq_get(&ifsq->ifsq_prio);
2964 	if (m != NULL) {
2965 		ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2966 		ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2967 	}
2968 	return (m);
2969 }
2970 
2971 int
2972 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2973     struct altq_pktattr *pa __unused)
2974 {
2975 
2976 	M_ASSERTPKTHDR(m);
2977 again:
2978 	if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2979 	    ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2980 		struct mbuf *m_drop;
2981 
2982 		if (m->m_flags & M_PRIO) {
2983 			m_drop = NULL;
2984 			if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) &&
2985 			    ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) {
2986 				/* Try dropping some from normal queue. */
2987 				m_drop = ifsq_norm_dequeue(ifsq);
2988 			}
2989 			if (m_drop == NULL)
2990 				m_drop = ifsq_prio_dequeue(ifsq);
2991 		} else {
2992 			m_drop = ifsq_norm_dequeue(ifsq);
2993 		}
2994 		if (m_drop != NULL) {
2995 			IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1);
2996 			m_freem(m_drop);
2997 			goto again;
2998 		}
2999 		/*
3000 		 * No old packets could be dropped!
3001 		 * NOTE: Caller increases oqdrops.
3002 		 */
3003 		m_freem(m);
3004 		return (ENOBUFS);
3005 	} else {
3006 		if (m->m_flags & M_PRIO)
3007 			ifsq_prio_enqueue(ifsq, m);
3008 		else
3009 			ifsq_norm_enqueue(ifsq, m);
3010 		return (0);
3011 	}
3012 }
3013 
3014 struct mbuf *
3015 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
3016 {
3017 	struct mbuf *m;
3018 
3019 	switch (op) {
3020 	case ALTDQ_POLL:
3021 		m = classq_head(&ifsq->ifsq_prio);
3022 		if (m == NULL)
3023 			m = classq_head(&ifsq->ifsq_norm);
3024 		break;
3025 
3026 	case ALTDQ_REMOVE:
3027 		m = ifsq_prio_dequeue(ifsq);
3028 		if (m == NULL)
3029 			m = ifsq_norm_dequeue(ifsq);
3030 		break;
3031 
3032 	default:
3033 		panic("unsupported ALTQ dequeue op: %d", op);
3034 	}
3035 	return m;
3036 }
3037 
3038 int
3039 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
3040 {
3041 	switch (req) {
3042 	case ALTRQ_PURGE:
3043 		for (;;) {
3044 			struct mbuf *m;
3045 
3046 			m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
3047 			if (m == NULL)
3048 				break;
3049 			m_freem(m);
3050 		}
3051 		break;
3052 
3053 	default:
3054 		panic("unsupported ALTQ request: %d", req);
3055 	}
3056 	return 0;
3057 }
3058 
3059 static void
3060 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
3061 {
3062 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
3063 	int running = 0, need_sched;
3064 
3065 	/*
3066 	 * Try to do direct ifnet.if_start on the subqueue first, if there is
3067 	 * contention on the subqueue hardware serializer, ifnet.if_start on
3068 	 * the subqueue will be scheduled on the subqueue owner CPU.
3069 	 */
3070 	if (!ifsq_tryserialize_hw(ifsq)) {
3071 		/*
3072 		 * Subqueue hardware serializer contention happened,
3073 		 * ifnet.if_start on the subqueue is scheduled on
3074 		 * the subqueue owner CPU, and we keep going.
3075 		 */
3076 		ifsq_ifstart_schedule(ifsq, 1);
3077 		return;
3078 	}
3079 
3080 	if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
3081 		ifp->if_start(ifp, ifsq);
3082 		if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
3083 			running = 1;
3084 	}
3085 	need_sched = ifsq_ifstart_need_schedule(ifsq, running);
3086 
3087 	ifsq_deserialize_hw(ifsq);
3088 
3089 	if (need_sched) {
3090 		/*
3091 		 * More data need to be transmitted, ifnet.if_start on the
3092 		 * subqueue is scheduled on the subqueue owner CPU, and we
3093 		 * keep going.
3094 		 * NOTE: ifnet.if_start subqueue interlock is not released.
3095 		 */
3096 		ifsq_ifstart_schedule(ifsq, force_sched);
3097 	}
3098 }
3099 
3100 /*
3101  * Subqeue packets staging mechanism:
3102  *
3103  * The packets enqueued into the subqueue are staged to a certain amount
3104  * before the ifnet.if_start on the subqueue is called.  In this way, the
3105  * driver could avoid writing to hardware registers upon every packet,
3106  * instead, hardware registers could be written when certain amount of
3107  * packets are put onto hardware TX ring.  The measurement on several modern
3108  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
3109  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
3110  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
3111  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
3112  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
3113  *
3114  * Subqueue packets staging is performed for two entry points into drivers'
3115  * transmission function:
3116  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
3117  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
3118  *
3119  * Subqueue packets staging will be stopped upon any of the following
3120  * conditions:
3121  * - If the count of packets enqueued on the current CPU is great than or
3122  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
3123  * - If the total length of packets enqueued on the current CPU is great
3124  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
3125  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
3126  *   is usually less than hardware's MTU.
3127  * - ifsq_ifstart_schedule() is not pending on the current CPU and
3128  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
3129  *   released.
3130  * - The if_start_rollup(), which is registered as low priority netisr
3131  *   rollup function, is called; probably because no more work is pending
3132  *   for netisr.
3133  *
3134  * NOTE:
3135  * Currently subqueue packet staging is only performed in netisr threads.
3136  */
3137 int
3138 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
3139 {
3140 	struct ifaltq *ifq = &ifp->if_snd;
3141 	struct ifaltq_subque *ifsq;
3142 	int error, start = 0, len, mcast = 0, avoid_start = 0;
3143 	struct ifsubq_stage_head *head = NULL;
3144 	struct ifsubq_stage *stage = NULL;
3145 	struct globaldata *gd = mycpu;
3146 	struct thread *td = gd->gd_curthread;
3147 
3148 	crit_enter_quick(td);
3149 
3150 	ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
3151 	ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3152 
3153 	len = m->m_pkthdr.len;
3154 	if (m->m_flags & M_MCAST)
3155 		mcast = 1;
3156 
3157 	if (td->td_type == TD_TYPE_NETISR) {
3158 		head = &ifsubq_stage_heads[mycpuid];
3159 		stage = ifsq_get_stage(ifsq, mycpuid);
3160 
3161 		stage->stg_cnt++;
3162 		stage->stg_len += len;
3163 		if (stage->stg_cnt < ifsq_stage_cntmax &&
3164 		    stage->stg_len < (ifp->if_mtu - max_protohdr))
3165 			avoid_start = 1;
3166 	}
3167 
3168 	ALTQ_SQ_LOCK(ifsq);
3169 	error = ifsq_enqueue_locked(ifsq, m, pa);
3170 	if (error) {
3171 		IFNET_STAT_INC(ifp, oqdrops, 1);
3172 		if (!ifsq_data_ready(ifsq)) {
3173 			ALTQ_SQ_UNLOCK(ifsq);
3174 			crit_exit_quick(td);
3175 			return error;
3176 		}
3177 		avoid_start = 0;
3178 	}
3179 	if (!ifsq_is_started(ifsq)) {
3180 		if (avoid_start) {
3181 			ALTQ_SQ_UNLOCK(ifsq);
3182 
3183 			KKASSERT(!error);
3184 			if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3185 				ifsq_stage_insert(head, stage);
3186 
3187 			IFNET_STAT_INC(ifp, obytes, len);
3188 			if (mcast)
3189 				IFNET_STAT_INC(ifp, omcasts, 1);
3190 			crit_exit_quick(td);
3191 			return error;
3192 		}
3193 
3194 		/*
3195 		 * Hold the subqueue interlock of ifnet.if_start
3196 		 */
3197 		ifsq_set_started(ifsq);
3198 		start = 1;
3199 	}
3200 	ALTQ_SQ_UNLOCK(ifsq);
3201 
3202 	if (!error) {
3203 		IFNET_STAT_INC(ifp, obytes, len);
3204 		if (mcast)
3205 			IFNET_STAT_INC(ifp, omcasts, 1);
3206 	}
3207 
3208 	if (stage != NULL) {
3209 		if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3210 			KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3211 			if (!avoid_start) {
3212 				ifsq_stage_remove(head, stage);
3213 				ifsq_ifstart_schedule(ifsq, 1);
3214 			}
3215 			crit_exit_quick(td);
3216 			return error;
3217 		}
3218 
3219 		if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3220 			ifsq_stage_remove(head, stage);
3221 		} else {
3222 			stage->stg_cnt = 0;
3223 			stage->stg_len = 0;
3224 		}
3225 	}
3226 
3227 	if (!start) {
3228 		crit_exit_quick(td);
3229 		return error;
3230 	}
3231 
3232 	ifsq_ifstart_try(ifsq, 0);
3233 
3234 	crit_exit_quick(td);
3235 	return error;
3236 }
3237 
3238 void *
3239 ifa_create(int size)
3240 {
3241 	struct ifaddr *ifa;
3242 	int i;
3243 
3244 	KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3245 
3246 	ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3247 
3248 	/*
3249 	 * Make ifa_container availabel on all CPUs, since they
3250 	 * could be accessed by any threads.
3251 	 */
3252 	ifa->ifa_containers =
3253 	    kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
3254 	        M_IFADDR, M_INTWAIT | M_ZERO);
3255 
3256 	ifa->ifa_ncnt = ncpus;
3257 	for (i = 0; i < ncpus; ++i) {
3258 		struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3259 
3260 		ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3261 		ifac->ifa = ifa;
3262 		ifac->ifa_refcnt = 1;
3263 	}
3264 #ifdef IFADDR_DEBUG
3265 	kprintf("alloc ifa %p %d\n", ifa, size);
3266 #endif
3267 	return ifa;
3268 }
3269 
3270 void
3271 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3272 {
3273 	struct ifaddr *ifa = ifac->ifa;
3274 
3275 	KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3276 	KKASSERT(ifac->ifa_refcnt == 0);
3277 	KASSERT(ifac->ifa_listmask == 0,
3278 		("ifa is still on %#x lists", ifac->ifa_listmask));
3279 
3280 	ifac->ifa_magic = IFA_CONTAINER_DEAD;
3281 
3282 #ifdef IFADDR_DEBUG_VERBOSE
3283 	kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3284 #endif
3285 
3286 	KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3287 		("invalid # of ifac, %d", ifa->ifa_ncnt));
3288 	if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3289 #ifdef IFADDR_DEBUG
3290 		kprintf("free ifa %p\n", ifa);
3291 #endif
3292 		kfree(ifa->ifa_containers, M_IFADDR);
3293 		kfree(ifa, M_IFADDR);
3294 	}
3295 }
3296 
3297 static void
3298 ifa_iflink_dispatch(netmsg_t nmsg)
3299 {
3300 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3301 	struct ifaddr *ifa = msg->ifa;
3302 	struct ifnet *ifp = msg->ifp;
3303 	int cpu = mycpuid;
3304 	struct ifaddr_container *ifac;
3305 
3306 	crit_enter();
3307 
3308 	ifac = &ifa->ifa_containers[cpu];
3309 	ASSERT_IFAC_VALID(ifac);
3310 	KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3311 		("ifaddr is on if_addrheads"));
3312 
3313 	ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3314 	if (msg->tail)
3315 		TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3316 	else
3317 		TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3318 
3319 	crit_exit();
3320 
3321 	netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3322 }
3323 
3324 void
3325 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3326 {
3327 	struct netmsg_ifaddr msg;
3328 
3329 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3330 		    0, ifa_iflink_dispatch);
3331 	msg.ifa = ifa;
3332 	msg.ifp = ifp;
3333 	msg.tail = tail;
3334 
3335 	netisr_domsg(&msg.base, 0);
3336 }
3337 
3338 static void
3339 ifa_ifunlink_dispatch(netmsg_t nmsg)
3340 {
3341 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3342 	struct ifaddr *ifa = msg->ifa;
3343 	struct ifnet *ifp = msg->ifp;
3344 	int cpu = mycpuid;
3345 	struct ifaddr_container *ifac;
3346 
3347 	crit_enter();
3348 
3349 	ifac = &ifa->ifa_containers[cpu];
3350 	ASSERT_IFAC_VALID(ifac);
3351 	KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3352 		("ifaddr is not on if_addrhead"));
3353 
3354 	TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3355 	ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3356 
3357 	crit_exit();
3358 
3359 	netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3360 }
3361 
3362 void
3363 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3364 {
3365 	struct netmsg_ifaddr msg;
3366 
3367 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3368 		    0, ifa_ifunlink_dispatch);
3369 	msg.ifa = ifa;
3370 	msg.ifp = ifp;
3371 
3372 	netisr_domsg(&msg.base, 0);
3373 }
3374 
3375 static void
3376 ifa_destroy_dispatch(netmsg_t nmsg)
3377 {
3378 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3379 
3380 	IFAFREE(msg->ifa);
3381 	netisr_forwardmsg_all(&nmsg->base, mycpuid + 1);
3382 }
3383 
3384 void
3385 ifa_destroy(struct ifaddr *ifa)
3386 {
3387 	struct netmsg_ifaddr msg;
3388 
3389 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3390 		    0, ifa_destroy_dispatch);
3391 	msg.ifa = ifa;
3392 
3393 	netisr_domsg(&msg.base, 0);
3394 }
3395 
3396 static void
3397 if_start_rollup(void)
3398 {
3399 	struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3400 	struct ifsubq_stage *stage;
3401 
3402 	crit_enter();
3403 
3404 	while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3405 		struct ifaltq_subque *ifsq = stage->stg_subq;
3406 		int is_sched = 0;
3407 
3408 		if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3409 			is_sched = 1;
3410 		ifsq_stage_remove(head, stage);
3411 
3412 		if (is_sched) {
3413 			ifsq_ifstart_schedule(ifsq, 1);
3414 		} else {
3415 			int start = 0;
3416 
3417 			ALTQ_SQ_LOCK(ifsq);
3418 			if (!ifsq_is_started(ifsq)) {
3419 				/*
3420 				 * Hold the subqueue interlock of
3421 				 * ifnet.if_start
3422 				 */
3423 				ifsq_set_started(ifsq);
3424 				start = 1;
3425 			}
3426 			ALTQ_SQ_UNLOCK(ifsq);
3427 
3428 			if (start)
3429 				ifsq_ifstart_try(ifsq, 1);
3430 		}
3431 		KKASSERT((stage->stg_flags &
3432 		    (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3433 	}
3434 
3435 	crit_exit();
3436 }
3437 
3438 static void
3439 ifnetinit(void *dummy __unused)
3440 {
3441 	int i;
3442 
3443 	/* XXX netisr_ncpus */
3444 	for (i = 0; i < ncpus; ++i)
3445 		TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3446 	netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3447 }
3448 
3449 void
3450 if_register_com_alloc(u_char type,
3451     if_com_alloc_t *a, if_com_free_t *f)
3452 {
3453 
3454         KASSERT(if_com_alloc[type] == NULL,
3455             ("if_register_com_alloc: %d already registered", type));
3456         KASSERT(if_com_free[type] == NULL,
3457             ("if_register_com_alloc: %d free already registered", type));
3458 
3459         if_com_alloc[type] = a;
3460         if_com_free[type] = f;
3461 }
3462 
3463 void
3464 if_deregister_com_alloc(u_char type)
3465 {
3466 
3467         KASSERT(if_com_alloc[type] != NULL,
3468             ("if_deregister_com_alloc: %d not registered", type));
3469         KASSERT(if_com_free[type] != NULL,
3470             ("if_deregister_com_alloc: %d free not registered", type));
3471         if_com_alloc[type] = NULL;
3472         if_com_free[type] = NULL;
3473 }
3474 
3475 void
3476 ifq_set_maxlen(struct ifaltq *ifq, int len)
3477 {
3478 	ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3479 }
3480 
3481 int
3482 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3483 {
3484 	return ALTQ_SUBQ_INDEX_DEFAULT;
3485 }
3486 
3487 int
3488 ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid)
3489 {
3490 
3491 	return (cpuid % ifq->altq_subq_mappriv);
3492 }
3493 
3494 static void
3495 ifsq_watchdog(void *arg)
3496 {
3497 	struct ifsubq_watchdog *wd = arg;
3498 	struct ifnet *ifp;
3499 
3500 	if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3501 		goto done;
3502 
3503 	ifp = ifsq_get_ifp(wd->wd_subq);
3504 	if (ifnet_tryserialize_all(ifp)) {
3505 		wd->wd_watchdog(wd->wd_subq);
3506 		ifnet_deserialize_all(ifp);
3507 	} else {
3508 		/* try again next timeout */
3509 		wd->wd_timer = 1;
3510 	}
3511 done:
3512 	ifsq_watchdog_reset(wd);
3513 }
3514 
3515 static void
3516 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3517 {
3518 	callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3519 	    ifsq_get_cpuid(wd->wd_subq));
3520 }
3521 
3522 void
3523 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3524     ifsq_watchdog_t watchdog)
3525 {
3526 	callout_init_mp(&wd->wd_callout);
3527 	wd->wd_timer = 0;
3528 	wd->wd_subq = ifsq;
3529 	wd->wd_watchdog = watchdog;
3530 }
3531 
3532 void
3533 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3534 {
3535 	wd->wd_timer = 0;
3536 	ifsq_watchdog_reset(wd);
3537 }
3538 
3539 void
3540 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3541 {
3542 	wd->wd_timer = 0;
3543 	callout_stop(&wd->wd_callout);
3544 }
3545 
3546 void
3547 ifnet_lock(void)
3548 {
3549 	KASSERT(curthread->td_type != TD_TYPE_NETISR,
3550 	    ("try holding ifnet lock in netisr"));
3551 	mtx_lock(&ifnet_mtx);
3552 }
3553 
3554 void
3555 ifnet_unlock(void)
3556 {
3557 	KASSERT(curthread->td_type != TD_TYPE_NETISR,
3558 	    ("try holding ifnet lock in netisr"));
3559 	mtx_unlock(&ifnet_mtx);
3560 }
3561 
3562 static struct ifnet_array *
3563 ifnet_array_alloc(int count)
3564 {
3565 	struct ifnet_array *arr;
3566 
3567 	arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3568 	    M_IFNET, M_WAITOK);
3569 	arr->ifnet_count = count;
3570 
3571 	return arr;
3572 }
3573 
3574 static void
3575 ifnet_array_free(struct ifnet_array *arr)
3576 {
3577 	if (arr == &ifnet_array0)
3578 		return;
3579 	kfree(arr, M_IFNET);
3580 }
3581 
3582 static struct ifnet_array *
3583 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3584 {
3585 	struct ifnet_array *arr;
3586 	int count, i;
3587 
3588 	KASSERT(old_arr->ifnet_count >= 0,
3589 	    ("invalid ifnet array count %d", old_arr->ifnet_count));
3590 	count = old_arr->ifnet_count + 1;
3591 	arr = ifnet_array_alloc(count);
3592 
3593 	/*
3594 	 * Save the old ifnet array and append this ifp to the end of
3595 	 * the new ifnet array.
3596 	 */
3597 	for (i = 0; i < old_arr->ifnet_count; ++i) {
3598 		KASSERT(old_arr->ifnet_arr[i] != ifp,
3599 		    ("%s is already in ifnet array", ifp->if_xname));
3600 		arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3601 	}
3602 	KASSERT(i == count - 1,
3603 	    ("add %s, ifnet array index mismatch, should be %d, but got %d",
3604 	     ifp->if_xname, count - 1, i));
3605 	arr->ifnet_arr[i] = ifp;
3606 
3607 	return arr;
3608 }
3609 
3610 static struct ifnet_array *
3611 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3612 {
3613 	struct ifnet_array *arr;
3614 	int count, i, idx, found = 0;
3615 
3616 	KASSERT(old_arr->ifnet_count > 0,
3617 	    ("invalid ifnet array count %d", old_arr->ifnet_count));
3618 	count = old_arr->ifnet_count - 1;
3619 	arr = ifnet_array_alloc(count);
3620 
3621 	/*
3622 	 * Save the old ifnet array, but skip this ifp.
3623 	 */
3624 	idx = 0;
3625 	for (i = 0; i < old_arr->ifnet_count; ++i) {
3626 		if (old_arr->ifnet_arr[i] == ifp) {
3627 			KASSERT(!found,
3628 			    ("dup %s is in ifnet array", ifp->if_xname));
3629 			found = 1;
3630 			continue;
3631 		}
3632 		KASSERT(idx < count,
3633 		    ("invalid ifnet array index %d, count %d", idx, count));
3634 		arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3635 		++idx;
3636 	}
3637 	KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3638 	KASSERT(idx == count,
3639 	    ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3640 	     ifp->if_xname, count, idx));
3641 
3642 	return arr;
3643 }
3644 
3645 const struct ifnet_array *
3646 ifnet_array_get(void)
3647 {
3648 	const struct ifnet_array *ret;
3649 
3650 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3651 	ret = ifnet_array;
3652 	/* Make sure 'ret' is really used. */
3653 	cpu_ccfence();
3654 	return (ret);
3655 }
3656 
3657 int
3658 ifnet_array_isempty(void)
3659 {
3660 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3661 	if (ifnet_array->ifnet_count == 0)
3662 		return 1;
3663 	else
3664 		return 0;
3665 }
3666 
3667 void
3668 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3669 {
3670 	struct ifaddr *ifa;
3671 
3672 	memset(mark, 0, sizeof(*mark));
3673 	ifa = &mark->ifa;
3674 
3675 	mark->ifac.ifa = ifa;
3676 
3677 	ifa->ifa_addr = &mark->addr;
3678 	ifa->ifa_dstaddr = &mark->dstaddr;
3679 	ifa->ifa_netmask = &mark->netmask;
3680 	ifa->ifa_ifp = ifp;
3681 }
3682 
3683 static int
3684 if_ringcnt_fixup(int ring_cnt, int ring_cntmax)
3685 {
3686 
3687 	KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax));
3688 
3689 	if (ring_cnt <= 0 || ring_cnt > ring_cntmax)
3690 		ring_cnt = ring_cntmax;
3691 	if (ring_cnt > netisr_ncpus)
3692 		ring_cnt = netisr_ncpus;
3693 	return (ring_cnt);
3694 }
3695 
3696 static void
3697 if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid)
3698 {
3699 	int i, offset;
3700 
3701 	KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid));
3702 	KASSERT(grid >= rm->rm_cnt, ("invalid if_ringmap grid %d, count %d",
3703 	    grid, rm->rm_cnt));
3704 	rm->rm_grid = grid;
3705 
3706 	offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus;
3707 	for (i = 0; i < rm->rm_cnt; ++i) {
3708 		rm->rm_cpumap[i] = offset + i;
3709 		KASSERT(rm->rm_cpumap[i] < netisr_ncpus,
3710 		    ("invalid cpumap[%d] = %d, offset %d", i,
3711 		     rm->rm_cpumap[i], offset));
3712 	}
3713 }
3714 
3715 static struct if_ringmap *
3716 if_ringmap_alloc_flags(device_t dev, int ring_cnt, int ring_cntmax,
3717     uint32_t flags)
3718 {
3719 	struct if_ringmap *rm;
3720 	int i, grid = 0, prev_grid;
3721 
3722 	ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3723 	rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]),
3724 	    M_DEVBUF, M_WAITOK | M_ZERO);
3725 
3726 	rm->rm_cnt = ring_cnt;
3727 	if (flags & RINGMAP_FLAG_POWEROF2)
3728 		rm->rm_cnt = 1 << (fls(rm->rm_cnt) - 1);
3729 
3730 	prev_grid = netisr_ncpus;
3731 	for (i = 0; i < netisr_ncpus; ++i) {
3732 		if (netisr_ncpus % (i + 1) != 0)
3733 			continue;
3734 
3735 		grid = netisr_ncpus / (i + 1);
3736 		if (rm->rm_cnt > grid) {
3737 			grid = prev_grid;
3738 			break;
3739 		}
3740 
3741 		if (rm->rm_cnt > netisr_ncpus / (i + 2))
3742 			break;
3743 		prev_grid = grid;
3744 	}
3745 	if_ringmap_set_grid(dev, rm, grid);
3746 
3747 	return (rm);
3748 }
3749 
3750 struct if_ringmap *
3751 if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax)
3752 {
3753 
3754 	return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3755 	    RINGMAP_FLAG_NONE));
3756 }
3757 
3758 struct if_ringmap *
3759 if_ringmap_alloc2(device_t dev, int ring_cnt, int ring_cntmax)
3760 {
3761 
3762 	return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3763 	    RINGMAP_FLAG_POWEROF2));
3764 }
3765 
3766 void
3767 if_ringmap_free(struct if_ringmap *rm)
3768 {
3769 
3770 	kfree(rm, M_DEVBUF);
3771 }
3772 
3773 /*
3774  * Align the two ringmaps.
3775  *
3776  * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3777  *
3778  * Before:
3779  *
3780  * CPU      0  1  2  3   4  5  6  7
3781  * NIC_RX               n0 n1 n2 n3
3782  * NIC_TX        N0 N1
3783  *
3784  * After:
3785  *
3786  * CPU      0  1  2  3   4  5  6  7
3787  * NIC_RX               n0 n1 n2 n3
3788  * NIC_TX               N0 N1
3789  */
3790 void
3791 if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3792 {
3793 
3794 	if (rm0->rm_grid > rm1->rm_grid)
3795 		if_ringmap_set_grid(dev, rm1, rm0->rm_grid);
3796 	else if (rm0->rm_grid < rm1->rm_grid)
3797 		if_ringmap_set_grid(dev, rm0, rm1->rm_grid);
3798 }
3799 
3800 void
3801 if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3802 {
3803 	int subset_grid, cnt, divisor, mod, offset, i;
3804 	struct if_ringmap *subset_rm, *rm;
3805 	int old_rm0_grid, old_rm1_grid;
3806 
3807 	if (rm0->rm_grid == rm1->rm_grid)
3808 		return;
3809 
3810 	/* Save grid for later use */
3811 	old_rm0_grid = rm0->rm_grid;
3812 	old_rm1_grid = rm1->rm_grid;
3813 
3814 	if_ringmap_align(dev, rm0, rm1);
3815 
3816 	/*
3817 	 * Re-shuffle rings to get more even distribution.
3818 	 *
3819 	 * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3820 	 *
3821 	 * CPU       0  1  2  3   4  5  6  7   8  9 10 11
3822 	 *
3823 	 * NIC_RX   a0 a1 a2 a3  b0 b1 b2 b3  c0 c1 c2 c3
3824 	 * NIC_TX   A0 A1        B0 B1        C0 C1
3825 	 *
3826 	 * NIC_RX   d0 d1 d2 d3  e0 e1 e2 e3  f0 f1 f2 f3
3827 	 * NIC_TX         D0 D1        E0 E1        F0 F1
3828 	 */
3829 
3830 	if (rm0->rm_cnt >= (2 * old_rm1_grid)) {
3831 		cnt = rm0->rm_cnt;
3832 		subset_grid = old_rm1_grid;
3833 		subset_rm = rm1;
3834 		rm = rm0;
3835 	} else if (rm1->rm_cnt > (2 * old_rm0_grid)) {
3836 		cnt = rm1->rm_cnt;
3837 		subset_grid = old_rm0_grid;
3838 		subset_rm = rm0;
3839 		rm = rm1;
3840 	} else {
3841 		/* No space to shuffle. */
3842 		return;
3843 	}
3844 
3845 	mod = cnt / subset_grid;
3846 	KKASSERT(mod >= 2);
3847 	divisor = netisr_ncpus / rm->rm_grid;
3848 	offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid;
3849 
3850 	for (i = 0; i < subset_rm->rm_cnt; ++i) {
3851 		subset_rm->rm_cpumap[i] += offset;
3852 		KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus,
3853 		    ("match: invalid cpumap[%d] = %d, offset %d",
3854 		     i, subset_rm->rm_cpumap[i], offset));
3855 	}
3856 #ifdef INVARIANTS
3857 	for (i = 0; i < subset_rm->rm_cnt; ++i) {
3858 		int j;
3859 
3860 		for (j = 0; j < rm->rm_cnt; ++j) {
3861 			if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i])
3862 				break;
3863 		}
3864 		KASSERT(j < rm->rm_cnt,
3865 		    ("subset cpumap[%d] = %d not found in superset",
3866 		     i, subset_rm->rm_cpumap[i]));
3867 	}
3868 #endif
3869 }
3870 
3871 int
3872 if_ringmap_count(const struct if_ringmap *rm)
3873 {
3874 
3875 	return (rm->rm_cnt);
3876 }
3877 
3878 int
3879 if_ringmap_cpumap(const struct if_ringmap *rm, int ring)
3880 {
3881 
3882 	KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring));
3883 	return (rm->rm_cpumap[ring]);
3884 }
3885 
3886 void
3887 if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent)
3888 {
3889 	int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy;
3890 
3891 	KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0,
3892 	    ("invalid redirect table entries %d", table_nent));
3893 
3894 	grid_idx = 0;
3895 	for (i = 0; i < NETISR_CPUMAX; ++i) {
3896 		table[i] = grid_idx++ % rm->rm_cnt;
3897 
3898 		if (grid_idx == rm->rm_grid)
3899 			grid_idx = 0;
3900 	}
3901 
3902 	/*
3903 	 * Make the ring distributed more evenly for the remainder
3904 	 * of each grid.
3905 	 *
3906 	 * e.g. 12 netisrs, rm contains 8 rings.
3907 	 *
3908 	 * Redirect table before:
3909 	 *
3910 	 *  0  1  2  3  4  5  6  7  0  1  2  3  0  1  2  3
3911 	 *  4  5  6  7  0  1  2  3  0  1  2  3  4  5  6  7
3912 	 *  0  1  2  3  0  1  2  3  4  5  6  7  0  1  2  3
3913 	 *  ....
3914 	 *
3915 	 * Redirect table after being patched (pX, patched entries):
3916 	 *
3917 	 *  0  1  2  3  4  5  6  7 p0 p1 p2 p3  0  1  2  3
3918 	 *  4  5  6  7 p4 p5 p6 p7  0  1  2  3  4  5  6  7
3919 	 * p0 p1 p2 p3  0  1  2  3  4  5  6  7 p4 p5 p6 p7
3920 	 *  ....
3921 	 */
3922 	patch_cnt = rm->rm_grid % rm->rm_cnt;
3923 	if (patch_cnt == 0)
3924 		goto done;
3925 	patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt);
3926 
3927 	grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid;
3928 	grid_idx = 0;
3929 	for (i = 0; i < grid_cnt; ++i) {
3930 		int j;
3931 
3932 		for (j = 0; j < patch_cnt; ++j) {
3933 			int fix_idx;
3934 
3935 			fix_idx = (i * rm->rm_grid) + patch_off + j;
3936 			if (fix_idx >= NETISR_CPUMAX)
3937 				goto done;
3938 			table[fix_idx] = grid_idx++ % rm->rm_cnt;
3939 		}
3940 	}
3941 done:
3942 	/*
3943 	 * If the device supports larger redirect table, duplicate
3944 	 * the first NETISR_CPUMAX entries to the rest of the table,
3945 	 * so that it matches upper layer's expectation:
3946 	 * (hash & NETISR_CPUMASK) % netisr_ncpus
3947 	 */
3948 	ncopy = table_nent / NETISR_CPUMAX;
3949 	for (i = 1; i < ncopy; ++i) {
3950 		memcpy(&table[i * NETISR_CPUMAX], table,
3951 		    NETISR_CPUMAX * sizeof(table[0]));
3952 	}
3953 	if (if_ringmap_dumprdr) {
3954 		for (i = 0; i < table_nent; ++i) {
3955 			if (i != 0 && i % 16 == 0)
3956 				kprintf("\n");
3957 			kprintf("%03d ", table[i]);
3958 		}
3959 		kprintf("\n");
3960 	}
3961 }
3962 
3963 int
3964 if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS)
3965 {
3966 	struct if_ringmap *rm = arg1;
3967 	int i, error = 0;
3968 
3969 	for (i = 0; i < rm->rm_cnt; ++i) {
3970 		int cpu = rm->rm_cpumap[i];
3971 
3972 		error = SYSCTL_OUT(req, &cpu, sizeof(cpu));
3973 		if (error)
3974 			break;
3975 	}
3976 	return (error);
3977 }
3978