xref: /freebsd/sys/net/if.c (revision 44682688f038edbf34591b25ce36412a7f2d6d07)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2010 Bjoern A. Zeeb <bz@FreeBSD.org>
5  * Copyright (c) 1980, 1986, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)if.c	8.5 (Berkeley) 1/9/95
33  * $FreeBSD$
34  */
35 
36 #include "opt_bpf.h"
37 #include "opt_inet6.h"
38 #include "opt_inet.h"
39 #include "opt_ddb.h"
40 
41 #include <sys/param.h>
42 #include <sys/capsicum.h>
43 #include <sys/conf.h>
44 #include <sys/eventhandler.h>
45 #include <sys/malloc.h>
46 #include <sys/domainset.h>
47 #include <sys/sbuf.h>
48 #include <sys/bus.h>
49 #include <sys/epoch.h>
50 #include <sys/mbuf.h>
51 #include <sys/systm.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 #include <sys/kernel.h>
58 #include <sys/lock.h>
59 #include <sys/refcount.h>
60 #include <sys/module.h>
61 #include <sys/rwlock.h>
62 #include <sys/sockio.h>
63 #include <sys/syslog.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
66 #include <sys/taskqueue.h>
67 #include <sys/domain.h>
68 #include <sys/jail.h>
69 #include <sys/priv.h>
70 
71 #ifdef DDB
72 #include <ddb/ddb.h>
73 #endif
74 
75 #include <machine/stdarg.h>
76 #include <vm/uma.h>
77 
78 #include <net/bpf.h>
79 #include <net/ethernet.h>
80 #include <net/if.h>
81 #include <net/if_arp.h>
82 #include <net/if_clone.h>
83 #include <net/if_dl.h>
84 #include <net/if_types.h>
85 #include <net/if_var.h>
86 #include <net/if_media.h>
87 #include <net/if_mib.h>
88 #include <net/if_vlan_var.h>
89 #include <net/radix.h>
90 #include <net/route.h>
91 #include <net/route/route_ctl.h>
92 #include <net/vnet.h>
93 
94 #if defined(INET) || defined(INET6)
95 #include <net/ethernet.h>
96 #include <netinet/in.h>
97 #include <netinet/in_var.h>
98 #include <netinet/ip.h>
99 #include <netinet/ip_carp.h>
100 #ifdef INET
101 #include <net/debugnet.h>
102 #include <netinet/if_ether.h>
103 #endif /* INET */
104 #ifdef INET6
105 #include <netinet6/in6_var.h>
106 #include <netinet6/in6_ifattach.h>
107 #endif /* INET6 */
108 #endif /* INET || INET6 */
109 
110 #include <security/mac/mac_framework.h>
111 
112 /*
113  * Consumers of struct ifreq such as tcpdump assume no pad between ifr_name
114  * and ifr_ifru when it is used in SIOCGIFCONF.
115  */
116 _Static_assert(sizeof(((struct ifreq *)0)->ifr_name) ==
117     offsetof(struct ifreq, ifr_ifru), "gap between ifr_name and ifr_ifru");
118 
119 __read_mostly epoch_t net_epoch_preempt;
120 #ifdef COMPAT_FREEBSD32
121 #include <sys/mount.h>
122 #include <compat/freebsd32/freebsd32.h>
123 
124 struct ifreq_buffer32 {
125 	uint32_t	length;		/* (size_t) */
126 	uint32_t	buffer;		/* (void *) */
127 };
128 
129 /*
130  * Interface request structure used for socket
131  * ioctl's.  All interface ioctl's must have parameter
132  * definitions which begin with ifr_name.  The
133  * remainder may be interface specific.
134  */
135 struct ifreq32 {
136 	char	ifr_name[IFNAMSIZ];		/* if name, e.g. "en0" */
137 	union {
138 		struct sockaddr	ifru_addr;
139 		struct sockaddr	ifru_dstaddr;
140 		struct sockaddr	ifru_broadaddr;
141 		struct ifreq_buffer32 ifru_buffer;
142 		short		ifru_flags[2];
143 		short		ifru_index;
144 		int		ifru_jid;
145 		int		ifru_metric;
146 		int		ifru_mtu;
147 		int		ifru_phys;
148 		int		ifru_media;
149 		uint32_t	ifru_data;
150 		int		ifru_cap[2];
151 		u_int		ifru_fib;
152 		u_char		ifru_vlan_pcp;
153 	} ifr_ifru;
154 };
155 CTASSERT(sizeof(struct ifreq) == sizeof(struct ifreq32));
156 CTASSERT(__offsetof(struct ifreq, ifr_ifru) ==
157     __offsetof(struct ifreq32, ifr_ifru));
158 
159 struct ifconf32 {
160 	int32_t	ifc_len;
161 	union {
162 		uint32_t	ifcu_buf;
163 		uint32_t	ifcu_req;
164 	} ifc_ifcu;
165 };
166 #define	SIOCGIFCONF32	_IOWR('i', 36, struct ifconf32)
167 
168 struct ifdrv32 {
169 	char		ifd_name[IFNAMSIZ];
170 	uint32_t	ifd_cmd;
171 	uint32_t	ifd_len;
172 	uint32_t	ifd_data;
173 };
174 #define SIOCSDRVSPEC32	_IOC_NEWTYPE(SIOCSDRVSPEC, struct ifdrv32)
175 #define SIOCGDRVSPEC32	_IOC_NEWTYPE(SIOCGDRVSPEC, struct ifdrv32)
176 
177 struct ifgroupreq32 {
178 	char	ifgr_name[IFNAMSIZ];
179 	u_int	ifgr_len;
180 	union {
181 		char		ifgru_group[IFNAMSIZ];
182 		uint32_t	ifgru_groups;
183 	} ifgr_ifgru;
184 };
185 #define	SIOCAIFGROUP32	_IOC_NEWTYPE(SIOCAIFGROUP, struct ifgroupreq32)
186 #define	SIOCGIFGROUP32	_IOC_NEWTYPE(SIOCGIFGROUP, struct ifgroupreq32)
187 #define	SIOCDIFGROUP32	_IOC_NEWTYPE(SIOCDIFGROUP, struct ifgroupreq32)
188 #define	SIOCGIFGMEMB32	_IOC_NEWTYPE(SIOCGIFGMEMB, struct ifgroupreq32)
189 
190 struct ifmediareq32 {
191 	char		ifm_name[IFNAMSIZ];
192 	int		ifm_current;
193 	int		ifm_mask;
194 	int		ifm_status;
195 	int		ifm_active;
196 	int		ifm_count;
197 	uint32_t	ifm_ulist;	/* (int *) */
198 };
199 #define	SIOCGIFMEDIA32	_IOC_NEWTYPE(SIOCGIFMEDIA, struct ifmediareq32)
200 #define	SIOCGIFXMEDIA32	_IOC_NEWTYPE(SIOCGIFXMEDIA, struct ifmediareq32)
201 #endif /* COMPAT_FREEBSD32 */
202 
203 union ifreq_union {
204 	struct ifreq	ifr;
205 #ifdef COMPAT_FREEBSD32
206 	struct ifreq32	ifr32;
207 #endif
208 };
209 
210 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
211     "Link layers");
212 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
213     "Generic link-management");
214 
215 SYSCTL_INT(_net_link, OID_AUTO, ifqmaxlen, CTLFLAG_RDTUN,
216     &ifqmaxlen, 0, "max send queue size");
217 
218 /* Log link state change events */
219 static int log_link_state_change = 1;
220 
221 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
222 	&log_link_state_change, 0,
223 	"log interface link state change events");
224 
225 /* Log promiscuous mode change events */
226 static int log_promisc_mode_change = 1;
227 
228 SYSCTL_INT(_net_link, OID_AUTO, log_promisc_mode_change, CTLFLAG_RDTUN,
229 	&log_promisc_mode_change, 1,
230 	"log promiscuous mode change events");
231 
232 /* Interface description */
233 static unsigned int ifdescr_maxlen = 1024;
234 SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW,
235 	&ifdescr_maxlen, 0,
236 	"administrative maximum length for interface description");
237 
238 static MALLOC_DEFINE(M_IFDESCR, "ifdescr", "ifnet descriptions");
239 
240 /* global sx for non-critical path ifdescr */
241 static struct sx ifdescr_sx;
242 SX_SYSINIT(ifdescr_sx, &ifdescr_sx, "ifnet descr");
243 
244 void	(*ng_ether_link_state_p)(struct ifnet *ifp, int state);
245 void	(*lagg_linkstate_p)(struct ifnet *ifp, int state);
246 /* These are external hooks for CARP. */
247 void	(*carp_linkstate_p)(struct ifnet *ifp);
248 void	(*carp_demote_adj_p)(int, char *);
249 int	(*carp_master_p)(struct ifaddr *);
250 #if defined(INET) || defined(INET6)
251 int	(*carp_forus_p)(struct ifnet *ifp, u_char *dhost);
252 int	(*carp_output_p)(struct ifnet *ifp, struct mbuf *m,
253     const struct sockaddr *sa);
254 int	(*carp_ioctl_p)(struct ifreq *, u_long, struct thread *);
255 int	(*carp_attach_p)(struct ifaddr *, int);
256 void	(*carp_detach_p)(struct ifaddr *, bool);
257 #endif
258 #ifdef INET
259 int	(*carp_iamatch_p)(struct ifaddr *, uint8_t **);
260 #endif
261 #ifdef INET6
262 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6);
263 caddr_t	(*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m,
264     const struct in6_addr *taddr);
265 #endif
266 
267 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
268 
269 /*
270  * XXX: Style; these should be sorted alphabetically, and unprototyped
271  * static functions should be prototyped. Currently they are sorted by
272  * declaration order.
273  */
274 static void	if_attachdomain(void *);
275 static void	if_attachdomain1(struct ifnet *);
276 static int	ifconf(u_long, caddr_t);
277 static void	if_input_default(struct ifnet *, struct mbuf *);
278 static int	if_requestencap_default(struct ifnet *, struct if_encap_req *);
279 static void	if_route(struct ifnet *, int flag, int fam);
280 static int	if_setflag(struct ifnet *, int, int, int *, int);
281 static int	if_transmit(struct ifnet *ifp, struct mbuf *m);
282 static void	if_unroute(struct ifnet *, int flag, int fam);
283 static int	if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
284 static void	do_link_state_change(void *, int);
285 static int	if_getgroup(struct ifgroupreq *, struct ifnet *);
286 static int	if_getgroupmembers(struct ifgroupreq *);
287 static void	if_delgroups(struct ifnet *);
288 static void	if_attach_internal(struct ifnet *, int, struct if_clone *);
289 static int	if_detach_internal(struct ifnet *, int, struct if_clone **);
290 static void	if_siocaddmulti(void *, int);
291 static void	if_link_ifnet(struct ifnet *);
292 static bool	if_unlink_ifnet(struct ifnet *, bool);
293 #ifdef VIMAGE
294 static int	if_vmove(struct ifnet *, struct vnet *);
295 #endif
296 
297 #ifdef INET6
298 /*
299  * XXX: declare here to avoid to include many inet6 related files..
300  * should be more generalized?
301  */
302 extern void	nd6_setmtu(struct ifnet *);
303 #endif
304 
305 /* ipsec helper hooks */
306 VNET_DEFINE(struct hhook_head *, ipsec_hhh_in[HHOOK_IPSEC_COUNT]);
307 VNET_DEFINE(struct hhook_head *, ipsec_hhh_out[HHOOK_IPSEC_COUNT]);
308 
309 int	ifqmaxlen = IFQ_MAXLEN;
310 VNET_DEFINE(struct ifnethead, ifnet);	/* depend on static init XXX */
311 VNET_DEFINE(struct ifgrouphead, ifg_head);
312 
313 /* Table of ifnet by index. */
314 VNET_DEFINE_STATIC(int, if_index);
315 #define	V_if_index		VNET(if_index)
316 VNET_DEFINE_STATIC(int, if_indexlim) = 8;
317 #define	V_if_indexlim		VNET(if_indexlim)
318 VNET_DEFINE_STATIC(struct ifnet **, ifindex_table);
319 #define	V_ifindex_table		VNET(ifindex_table)
320 
321 SYSCTL_NODE(_net_link_generic, IFMIB_SYSTEM, system,
322     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
323     "Variables global to all interfaces");
324 SYSCTL_INT(_net_link_generic_system, IFMIB_IFCOUNT, ifcount,
325     CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(if_index), 0,
326     "Number of configured interfaces");
327 
328 /*
329  * The global network interface list (V_ifnet) and related state (such as
330  * if_index, if_indexlim, and ifindex_table) are protected by an sxlock.
331  * This may be acquired to stabilise the list, or we may rely on NET_EPOCH.
332  */
333 struct sx ifnet_sxlock;
334 SX_SYSINIT_FLAGS(ifnet_sx, &ifnet_sxlock, "ifnet_sx", SX_RECURSE);
335 
336 struct sx ifnet_detach_sxlock;
337 SX_SYSINIT_FLAGS(ifnet_detach, &ifnet_detach_sxlock, "ifnet_detach_sx",
338     SX_RECURSE);
339 
340 #ifdef VIMAGE
341 #define	VNET_IS_SHUTTING_DOWN(_vnet)					\
342     ((_vnet)->vnet_shutdown && (_vnet)->vnet_state < SI_SUB_VNET_DONE)
343 #endif
344 
345 static	if_com_alloc_t *if_com_alloc[256];
346 static	if_com_free_t *if_com_free[256];
347 
348 static MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
349 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
350 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
351 
352 struct ifnet *
353 ifnet_byindex(u_int idx)
354 {
355 
356 	NET_EPOCH_ASSERT();
357 
358 	if (__predict_false(idx > V_if_index))
359 		return (NULL);
360 
361 	return (ck_pr_load_ptr(&V_ifindex_table[idx]));
362 }
363 
364 struct ifnet *
365 ifnet_byindex_ref(u_int idx)
366 {
367 	struct ifnet *ifp;
368 
369 	ifp = ifnet_byindex(idx);
370 	if (ifp == NULL || (ifp->if_flags & IFF_DYING))
371 		return (NULL);
372 	if (!if_try_ref(ifp))
373 		return (NULL);
374 	return (ifp);
375 }
376 
377 /*
378  * Allocate an ifindex array entry.
379  */
380 static void
381 ifindex_alloc(struct ifnet *ifp)
382 {
383 	u_short idx;
384 
385 	IFNET_WLOCK();
386 	/*
387 	 * Try to find an empty slot below V_if_index.  If we fail, take the
388 	 * next slot.
389 	 */
390 	for (idx = 1; idx <= V_if_index; idx++) {
391 		if (V_ifindex_table[idx] == NULL)
392 			break;
393 	}
394 
395 	/* Catch if_index overflow. */
396 	if (idx >= V_if_indexlim) {
397 		struct ifnet **new, **old;
398 		int newlim;
399 
400 		newlim = V_if_indexlim * 2;
401 		new = malloc(newlim * sizeof(*new), M_IFNET, M_WAITOK | M_ZERO);
402 		memcpy(new, V_ifindex_table, V_if_indexlim * sizeof(*new));
403 		old = V_ifindex_table;
404 		ck_pr_store_ptr(&V_ifindex_table, new);
405 		V_if_indexlim = newlim;
406 		epoch_wait_preempt(net_epoch_preempt);
407 		free(old, M_IFNET);
408 	}
409 	if (idx > V_if_index)
410 		V_if_index = idx;
411 
412 	ifp->if_index = idx;
413 	ck_pr_store_ptr(&V_ifindex_table[idx], ifp);
414 	IFNET_WUNLOCK();
415 }
416 
417 static void
418 ifindex_free(u_short idx)
419 {
420 
421 	IFNET_WLOCK_ASSERT();
422 
423 	ck_pr_store_ptr(&V_ifindex_table[idx], NULL);
424 	while (V_if_index > 0 &&
425 	    V_ifindex_table[V_if_index] == NULL)
426 		V_if_index--;
427 }
428 
429 struct ifaddr *
430 ifaddr_byindex(u_short idx)
431 {
432 	struct ifnet *ifp;
433 	struct ifaddr *ifa = NULL;
434 
435 	NET_EPOCH_ASSERT();
436 
437 	ifp = ifnet_byindex(idx);
438 	if (ifp != NULL && (ifa = ifp->if_addr) != NULL)
439 		ifa_ref(ifa);
440 	return (ifa);
441 }
442 
443 /*
444  * Network interface utility routines.
445  *
446  * Routines with ifa_ifwith* names take sockaddr *'s as
447  * parameters.
448  */
449 
450 static void
451 vnet_if_init(const void *unused __unused)
452 {
453 
454 	CK_STAILQ_INIT(&V_ifnet);
455 	CK_STAILQ_INIT(&V_ifg_head);
456 	V_ifindex_table = malloc(V_if_indexlim * sizeof(*V_ifindex_table),
457 	    M_IFNET, M_WAITOK | M_ZERO);
458 	vnet_if_clone_init();
459 }
460 VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init,
461     NULL);
462 
463 #ifdef VIMAGE
464 static void
465 vnet_if_uninit(const void *unused __unused)
466 {
467 
468 	VNET_ASSERT(CK_STAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p "
469 	    "not empty", __func__, __LINE__, &V_ifnet));
470 	VNET_ASSERT(CK_STAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p "
471 	    "not empty", __func__, __LINE__, &V_ifg_head));
472 
473 	free((caddr_t)V_ifindex_table, M_IFNET);
474 }
475 VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
476     vnet_if_uninit, NULL);
477 #endif
478 
479 static void
480 if_link_ifnet(struct ifnet *ifp)
481 {
482 
483 	IFNET_WLOCK();
484 	CK_STAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
485 #ifdef VIMAGE
486 	curvnet->vnet_ifcnt++;
487 #endif
488 	IFNET_WUNLOCK();
489 }
490 
491 static bool
492 if_unlink_ifnet(struct ifnet *ifp, bool vmove)
493 {
494 	struct ifnet *iter;
495 	int found = 0;
496 
497 	IFNET_WLOCK();
498 	CK_STAILQ_FOREACH(iter, &V_ifnet, if_link)
499 		if (iter == ifp) {
500 			CK_STAILQ_REMOVE(&V_ifnet, ifp, ifnet, if_link);
501 			if (!vmove)
502 				ifp->if_flags |= IFF_DYING;
503 			found = 1;
504 			break;
505 		}
506 #ifdef VIMAGE
507 	curvnet->vnet_ifcnt--;
508 #endif
509 	IFNET_WUNLOCK();
510 
511 	return (found);
512 }
513 
514 #ifdef VIMAGE
515 static void
516 vnet_if_return(const void *unused __unused)
517 {
518 	struct ifnet *ifp, *nifp;
519 	struct ifnet **pending;
520 	int found __diagused;
521 	int i;
522 
523 	i = 0;
524 
525 	/*
526 	 * We need to protect our access to the V_ifnet tailq. Ordinarily we'd
527 	 * enter NET_EPOCH, but that's not possible, because if_vmove() calls
528 	 * if_detach_internal(), which waits for NET_EPOCH callbacks to
529 	 * complete. We can't do that from within NET_EPOCH.
530 	 *
531 	 * However, we can also use the IFNET_xLOCK, which is the V_ifnet
532 	 * read/write lock. We cannot hold the lock as we call if_vmove()
533 	 * though, as that presents LOR w.r.t ifnet_sx, in_multi_sx and iflib
534 	 * ctx lock.
535 	 */
536 	IFNET_WLOCK();
537 
538 	pending = malloc(sizeof(struct ifnet *) * curvnet->vnet_ifcnt,
539 	    M_IFNET, M_WAITOK | M_ZERO);
540 
541 	/* Return all inherited interfaces to their parent vnets. */
542 	CK_STAILQ_FOREACH_SAFE(ifp, &V_ifnet, if_link, nifp) {
543 		if (ifp->if_home_vnet != ifp->if_vnet) {
544 			found = if_unlink_ifnet(ifp, true);
545 			MPASS(found);
546 
547 			pending[i++] = ifp;
548 		}
549 	}
550 	IFNET_WUNLOCK();
551 
552 	for (int j = 0; j < i; j++) {
553 		if_vmove(pending[j], pending[j]->if_home_vnet);
554 	}
555 
556 	free(pending, M_IFNET);
557 }
558 VNET_SYSUNINIT(vnet_if_return, SI_SUB_VNET_DONE, SI_ORDER_ANY,
559     vnet_if_return, NULL);
560 #endif
561 
562 /*
563  * Allocate a struct ifnet and an index for an interface.  A layer 2
564  * common structure will also be allocated if an allocation routine is
565  * registered for the passed type.
566  */
567 static struct ifnet *
568 if_alloc_domain(u_char type, int numa_domain)
569 {
570 	struct ifnet *ifp;
571 
572 	KASSERT(numa_domain <= IF_NODOM, ("numa_domain too large"));
573 	if (numa_domain == IF_NODOM)
574 		ifp = malloc(sizeof(struct ifnet), M_IFNET,
575 		    M_WAITOK | M_ZERO);
576 	else
577 		ifp = malloc_domainset(sizeof(struct ifnet), M_IFNET,
578 		    DOMAINSET_PREF(numa_domain), M_WAITOK | M_ZERO);
579 	ifp->if_type = type;
580 	ifp->if_alloctype = type;
581 	ifp->if_numa_domain = numa_domain;
582 #ifdef VIMAGE
583 	ifp->if_vnet = curvnet;
584 #endif
585 	if (if_com_alloc[type] != NULL) {
586 		ifp->if_l2com = if_com_alloc[type](type, ifp);
587 		KASSERT(ifp->if_l2com, ("%s: if_com_alloc[%u] failed", __func__,
588 		    type));
589 	}
590 
591 	IF_ADDR_LOCK_INIT(ifp);
592 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
593 	TASK_INIT(&ifp->if_addmultitask, 0, if_siocaddmulti, ifp);
594 	ifp->if_afdata_initialized = 0;
595 	IF_AFDATA_LOCK_INIT(ifp);
596 	CK_STAILQ_INIT(&ifp->if_addrhead);
597 	CK_STAILQ_INIT(&ifp->if_multiaddrs);
598 	CK_STAILQ_INIT(&ifp->if_groups);
599 #ifdef MAC
600 	mac_ifnet_init(ifp);
601 #endif
602 	ifq_init(&ifp->if_snd, ifp);
603 
604 	refcount_init(&ifp->if_refcount, 1);	/* Index reference. */
605 	for (int i = 0; i < IFCOUNTERS; i++)
606 		ifp->if_counters[i] = counter_u64_alloc(M_WAITOK);
607 	ifp->if_get_counter = if_get_counter_default;
608 	ifp->if_pcp = IFNET_PCP_NONE;
609 
610 	ifindex_alloc(ifp);
611 
612 	return (ifp);
613 }
614 
615 struct ifnet *
616 if_alloc_dev(u_char type, device_t dev)
617 {
618 	int numa_domain;
619 
620 	if (dev == NULL || bus_get_domain(dev, &numa_domain) != 0)
621 		return (if_alloc_domain(type, IF_NODOM));
622 	return (if_alloc_domain(type, numa_domain));
623 }
624 
625 struct ifnet *
626 if_alloc(u_char type)
627 {
628 
629 	return (if_alloc_domain(type, IF_NODOM));
630 }
631 /*
632  * Do the actual work of freeing a struct ifnet, and layer 2 common
633  * structure.  This call is made when the network epoch guarantees
634  * us that nobody holds a pointer to the interface.
635  */
636 static void
637 if_free_deferred(epoch_context_t ctx)
638 {
639 	struct ifnet *ifp = __containerof(ctx, struct ifnet, if_epoch_ctx);
640 
641 	KASSERT((ifp->if_flags & IFF_DYING),
642 	    ("%s: interface not dying", __func__));
643 
644 	if (if_com_free[ifp->if_alloctype] != NULL)
645 		if_com_free[ifp->if_alloctype](ifp->if_l2com,
646 		    ifp->if_alloctype);
647 
648 #ifdef MAC
649 	mac_ifnet_destroy(ifp);
650 #endif /* MAC */
651 	IF_AFDATA_DESTROY(ifp);
652 	IF_ADDR_LOCK_DESTROY(ifp);
653 	ifq_delete(&ifp->if_snd);
654 
655 	for (int i = 0; i < IFCOUNTERS; i++)
656 		counter_u64_free(ifp->if_counters[i]);
657 
658 	free(ifp->if_description, M_IFDESCR);
659 	free(ifp->if_hw_addr, M_IFADDR);
660 	free(ifp, M_IFNET);
661 }
662 
663 /*
664  * Deregister an interface and free the associated storage.
665  */
666 void
667 if_free(struct ifnet *ifp)
668 {
669 
670 	ifp->if_flags |= IFF_DYING;			/* XXX: Locking */
671 
672 	/*
673 	 * XXXGL: An interface index is really an alias to ifp pointer.
674 	 * Why would we clear the alias now, and not in the deferred
675 	 * context?  Indeed there is nothing wrong with some network
676 	 * thread obtaining ifp via ifnet_byindex() inside the network
677 	 * epoch and then dereferencing ifp while we peform if_free(),
678 	 * and after if_free() finished, too.
679 	 *
680 	 * The reason is the VIMAGE.  For some reason it was designed
681 	 * to require all sockets drained before destroying, but not all
682 	 * ifnets.  A vnet destruction calls if_vmove() on ifnet, which
683 	 * causes ID change.  But ID change and a possible misidentification
684 	 * of an ifnet later is a lesser problem, as it doesn't crash kernel.
685 	 * A worse problem is that removed interface may outlive the vnet it
686 	 * belongs too!  The if_free_deferred() would see ifp->if_vnet freed.
687 	 */
688 	CURVNET_SET_QUIET(ifp->if_vnet);
689 	IFNET_WLOCK();
690 	MPASS(V_ifindex_table[ifp->if_index] == ifp);
691 	ifindex_free(ifp->if_index);
692 	IFNET_WUNLOCK();
693 
694 	if (refcount_release(&ifp->if_refcount))
695 		NET_EPOCH_CALL(if_free_deferred, &ifp->if_epoch_ctx);
696 	CURVNET_RESTORE();
697 }
698 
699 /*
700  * Interfaces to keep an ifnet type-stable despite the possibility of the
701  * driver calling if_free().  If there are additional references, we defer
702  * freeing the underlying data structure.
703  */
704 void
705 if_ref(struct ifnet *ifp)
706 {
707 	u_int old __diagused;
708 
709 	/* We don't assert the ifnet list lock here, but arguably should. */
710 	old = refcount_acquire(&ifp->if_refcount);
711 	KASSERT(old > 0, ("%s: ifp %p has 0 refs", __func__, ifp));
712 }
713 
714 bool
715 if_try_ref(struct ifnet *ifp)
716 {
717 	NET_EPOCH_ASSERT();
718 	return (refcount_acquire_if_not_zero(&ifp->if_refcount));
719 }
720 
721 void
722 if_rele(struct ifnet *ifp)
723 {
724 
725 	if (!refcount_release(&ifp->if_refcount))
726 		return;
727 	NET_EPOCH_CALL(if_free_deferred, &ifp->if_epoch_ctx);
728 }
729 
730 void
731 ifq_init(struct ifaltq *ifq, struct ifnet *ifp)
732 {
733 
734 	mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
735 
736 	if (ifq->ifq_maxlen == 0)
737 		ifq->ifq_maxlen = ifqmaxlen;
738 
739 	ifq->altq_type = 0;
740 	ifq->altq_disc = NULL;
741 	ifq->altq_flags &= ALTQF_CANTCHANGE;
742 	ifq->altq_tbr  = NULL;
743 	ifq->altq_ifp  = ifp;
744 }
745 
746 void
747 ifq_delete(struct ifaltq *ifq)
748 {
749 	mtx_destroy(&ifq->ifq_mtx);
750 }
751 
752 /*
753  * Perform generic interface initialization tasks and attach the interface
754  * to the list of "active" interfaces.  If vmove flag is set on entry
755  * to if_attach_internal(), perform only a limited subset of initialization
756  * tasks, given that we are moving from one vnet to another an ifnet which
757  * has already been fully initialized.
758  *
759  * Note that if_detach_internal() removes group membership unconditionally
760  * even when vmove flag is set, and if_attach_internal() adds only IFG_ALL.
761  * Thus, when if_vmove() is applied to a cloned interface, group membership
762  * is lost while a cloned one always joins a group whose name is
763  * ifc->ifc_name.  To recover this after if_detach_internal() and
764  * if_attach_internal(), the cloner should be specified to
765  * if_attach_internal() via ifc.  If it is non-NULL, if_attach_internal()
766  * attempts to join a group whose name is ifc->ifc_name.
767  *
768  * XXX:
769  *  - The decision to return void and thus require this function to
770  *    succeed is questionable.
771  *  - We should probably do more sanity checking.  For instance we don't
772  *    do anything to insure if_xname is unique or non-empty.
773  */
774 void
775 if_attach(struct ifnet *ifp)
776 {
777 
778 	if_attach_internal(ifp, 0, NULL);
779 }
780 
781 /*
782  * Compute the least common TSO limit.
783  */
784 void
785 if_hw_tsomax_common(if_t ifp, struct ifnet_hw_tsomax *pmax)
786 {
787 	/*
788 	 * 1) If there is no limit currently, take the limit from
789 	 * the network adapter.
790 	 *
791 	 * 2) If the network adapter has a limit below the current
792 	 * limit, apply it.
793 	 */
794 	if (pmax->tsomaxbytes == 0 || (ifp->if_hw_tsomax != 0 &&
795 	    ifp->if_hw_tsomax < pmax->tsomaxbytes)) {
796 		pmax->tsomaxbytes = ifp->if_hw_tsomax;
797 	}
798 	if (pmax->tsomaxsegcount == 0 || (ifp->if_hw_tsomaxsegcount != 0 &&
799 	    ifp->if_hw_tsomaxsegcount < pmax->tsomaxsegcount)) {
800 		pmax->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
801 	}
802 	if (pmax->tsomaxsegsize == 0 || (ifp->if_hw_tsomaxsegsize != 0 &&
803 	    ifp->if_hw_tsomaxsegsize < pmax->tsomaxsegsize)) {
804 		pmax->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
805 	}
806 }
807 
808 /*
809  * Update TSO limit of a network adapter.
810  *
811  * Returns zero if no change. Else non-zero.
812  */
813 int
814 if_hw_tsomax_update(if_t ifp, struct ifnet_hw_tsomax *pmax)
815 {
816 	int retval = 0;
817 	if (ifp->if_hw_tsomax != pmax->tsomaxbytes) {
818 		ifp->if_hw_tsomax = pmax->tsomaxbytes;
819 		retval++;
820 	}
821 	if (ifp->if_hw_tsomaxsegsize != pmax->tsomaxsegsize) {
822 		ifp->if_hw_tsomaxsegsize = pmax->tsomaxsegsize;
823 		retval++;
824 	}
825 	if (ifp->if_hw_tsomaxsegcount != pmax->tsomaxsegcount) {
826 		ifp->if_hw_tsomaxsegcount = pmax->tsomaxsegcount;
827 		retval++;
828 	}
829 	return (retval);
830 }
831 
832 static void
833 if_attach_internal(struct ifnet *ifp, int vmove, struct if_clone *ifc)
834 {
835 	unsigned socksize, ifasize;
836 	int namelen, masklen;
837 	struct sockaddr_dl *sdl;
838 	struct ifaddr *ifa;
839 
840 	MPASS(V_ifindex_table[ifp->if_index] == ifp);
841 
842 #ifdef VIMAGE
843 	ifp->if_vnet = curvnet;
844 	if (ifp->if_home_vnet == NULL)
845 		ifp->if_home_vnet = curvnet;
846 #endif
847 
848 	if_addgroup(ifp, IFG_ALL);
849 
850 	/* Restore group membership for cloned interfaces. */
851 	if (vmove && ifc != NULL)
852 		if_clone_addgroup(ifp, ifc);
853 
854 	getmicrotime(&ifp->if_lastchange);
855 	ifp->if_epoch = time_uptime;
856 
857 	KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
858 	    (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
859 	    ("transmit and qflush must both either be set or both be NULL"));
860 	if (ifp->if_transmit == NULL) {
861 		ifp->if_transmit = if_transmit;
862 		ifp->if_qflush = if_qflush;
863 	}
864 	if (ifp->if_input == NULL)
865 		ifp->if_input = if_input_default;
866 
867 	if (ifp->if_requestencap == NULL)
868 		ifp->if_requestencap = if_requestencap_default;
869 
870 	if (!vmove) {
871 #ifdef MAC
872 		mac_ifnet_create(ifp);
873 #endif
874 
875 		/*
876 		 * Create a Link Level name for this device.
877 		 */
878 		namelen = strlen(ifp->if_xname);
879 		/*
880 		 * Always save enough space for any possiable name so we
881 		 * can do a rename in place later.
882 		 */
883 		masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
884 		socksize = masklen + ifp->if_addrlen;
885 		if (socksize < sizeof(*sdl))
886 			socksize = sizeof(*sdl);
887 		socksize = roundup2(socksize, sizeof(long));
888 		ifasize = sizeof(*ifa) + 2 * socksize;
889 		ifa = ifa_alloc(ifasize, M_WAITOK);
890 		sdl = (struct sockaddr_dl *)(ifa + 1);
891 		sdl->sdl_len = socksize;
892 		sdl->sdl_family = AF_LINK;
893 		bcopy(ifp->if_xname, sdl->sdl_data, namelen);
894 		sdl->sdl_nlen = namelen;
895 		sdl->sdl_index = ifp->if_index;
896 		sdl->sdl_type = ifp->if_type;
897 		ifp->if_addr = ifa;
898 		ifa->ifa_ifp = ifp;
899 		ifa->ifa_addr = (struct sockaddr *)sdl;
900 		sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
901 		ifa->ifa_netmask = (struct sockaddr *)sdl;
902 		sdl->sdl_len = masklen;
903 		while (namelen != 0)
904 			sdl->sdl_data[--namelen] = 0xff;
905 		CK_STAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
906 		/* Reliably crash if used uninitialized. */
907 		ifp->if_broadcastaddr = NULL;
908 
909 		if (ifp->if_type == IFT_ETHER) {
910 			ifp->if_hw_addr = malloc(ifp->if_addrlen, M_IFADDR,
911 			    M_WAITOK | M_ZERO);
912 		}
913 
914 #if defined(INET) || defined(INET6)
915 		/* Use defaults for TSO, if nothing is set */
916 		if (ifp->if_hw_tsomax == 0 &&
917 		    ifp->if_hw_tsomaxsegcount == 0 &&
918 		    ifp->if_hw_tsomaxsegsize == 0) {
919 			/*
920 			 * The TSO defaults needs to be such that an
921 			 * NFS mbuf list of 35 mbufs totalling just
922 			 * below 64K works and that a chain of mbufs
923 			 * can be defragged into at most 32 segments:
924 			 */
925 			ifp->if_hw_tsomax = min(IP_MAXPACKET, (32 * MCLBYTES) -
926 			    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
927 			ifp->if_hw_tsomaxsegcount = 35;
928 			ifp->if_hw_tsomaxsegsize = 2048;	/* 2K */
929 
930 			/* XXX some drivers set IFCAP_TSO after ethernet attach */
931 			if (ifp->if_capabilities & IFCAP_TSO) {
932 				if_printf(ifp, "Using defaults for TSO: %u/%u/%u\n",
933 				    ifp->if_hw_tsomax,
934 				    ifp->if_hw_tsomaxsegcount,
935 				    ifp->if_hw_tsomaxsegsize);
936 			}
937 		}
938 #endif
939 	}
940 #ifdef VIMAGE
941 	else {
942 		/*
943 		 * Update the interface index in the link layer address
944 		 * of the interface.
945 		 */
946 		for (ifa = ifp->if_addr; ifa != NULL;
947 		    ifa = CK_STAILQ_NEXT(ifa, ifa_link)) {
948 			if (ifa->ifa_addr->sa_family == AF_LINK) {
949 				sdl = (struct sockaddr_dl *)ifa->ifa_addr;
950 				sdl->sdl_index = ifp->if_index;
951 			}
952 		}
953 	}
954 #endif
955 
956 	if_link_ifnet(ifp);
957 
958 	if (domain_init_status >= 2)
959 		if_attachdomain1(ifp);
960 
961 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
962 	if (IS_DEFAULT_VNET(curvnet))
963 		devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
964 
965 	/* Announce the interface. */
966 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
967 }
968 
969 static void
970 if_epochalloc(void *dummy __unused)
971 {
972 
973 	net_epoch_preempt = epoch_alloc("Net preemptible", EPOCH_PREEMPT);
974 }
975 SYSINIT(ifepochalloc, SI_SUB_EPOCH, SI_ORDER_ANY, if_epochalloc, NULL);
976 
977 static void
978 if_attachdomain(void *dummy)
979 {
980 	struct ifnet *ifp;
981 
982 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link)
983 		if_attachdomain1(ifp);
984 }
985 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
986     if_attachdomain, NULL);
987 
988 static void
989 if_attachdomain1(struct ifnet *ifp)
990 {
991 	struct domain *dp;
992 
993 	/*
994 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
995 	 * cannot lock ifp->if_afdata initialization, entirely.
996 	 */
997 	IF_AFDATA_LOCK(ifp);
998 	if (ifp->if_afdata_initialized >= domain_init_status) {
999 		IF_AFDATA_UNLOCK(ifp);
1000 		log(LOG_WARNING, "%s called more than once on %s\n",
1001 		    __func__, ifp->if_xname);
1002 		return;
1003 	}
1004 	ifp->if_afdata_initialized = domain_init_status;
1005 	IF_AFDATA_UNLOCK(ifp);
1006 
1007 	/* address family dependent data region */
1008 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
1009 	for (dp = domains; dp; dp = dp->dom_next) {
1010 		if (dp->dom_ifattach)
1011 			ifp->if_afdata[dp->dom_family] =
1012 			    (*dp->dom_ifattach)(ifp);
1013 	}
1014 }
1015 
1016 /*
1017  * Remove any unicast or broadcast network addresses from an interface.
1018  */
1019 void
1020 if_purgeaddrs(struct ifnet *ifp)
1021 {
1022 	struct ifaddr *ifa;
1023 
1024 	while (1) {
1025 		struct epoch_tracker et;
1026 
1027 		NET_EPOCH_ENTER(et);
1028 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1029 			if (ifa->ifa_addr->sa_family != AF_LINK)
1030 				break;
1031 		}
1032 		NET_EPOCH_EXIT(et);
1033 
1034 		if (ifa == NULL)
1035 			break;
1036 #ifdef INET
1037 		/* XXX: Ugly!! ad hoc just for INET */
1038 		if (ifa->ifa_addr->sa_family == AF_INET) {
1039 			struct ifaliasreq ifr;
1040 
1041 			bzero(&ifr, sizeof(ifr));
1042 			ifr.ifra_addr = *ifa->ifa_addr;
1043 			if (ifa->ifa_dstaddr)
1044 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
1045 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
1046 			    NULL) == 0)
1047 				continue;
1048 		}
1049 #endif /* INET */
1050 #ifdef INET6
1051 		if (ifa->ifa_addr->sa_family == AF_INET6) {
1052 			in6_purgeifaddr((struct in6_ifaddr *)ifa);
1053 			/* ifp_addrhead is already updated */
1054 			continue;
1055 		}
1056 #endif /* INET6 */
1057 		IF_ADDR_WLOCK(ifp);
1058 		CK_STAILQ_REMOVE(&ifp->if_addrhead, ifa, ifaddr, ifa_link);
1059 		IF_ADDR_WUNLOCK(ifp);
1060 		ifa_free(ifa);
1061 	}
1062 }
1063 
1064 /*
1065  * Remove any multicast network addresses from an interface when an ifnet
1066  * is going away.
1067  */
1068 static void
1069 if_purgemaddrs(struct ifnet *ifp)
1070 {
1071 	struct ifmultiaddr *ifma;
1072 
1073 	IF_ADDR_WLOCK(ifp);
1074 	while (!CK_STAILQ_EMPTY(&ifp->if_multiaddrs)) {
1075 		ifma = CK_STAILQ_FIRST(&ifp->if_multiaddrs);
1076 		CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
1077 		if_delmulti_locked(ifp, ifma, 1);
1078 	}
1079 	IF_ADDR_WUNLOCK(ifp);
1080 }
1081 
1082 /*
1083  * Detach an interface, removing it from the list of "active" interfaces.
1084  * If vmove flag is set on entry to if_detach_internal(), perform only a
1085  * limited subset of cleanup tasks, given that we are moving an ifnet from
1086  * one vnet to another, where it must be fully operational.
1087  *
1088  * XXXRW: There are some significant questions about event ordering, and
1089  * how to prevent things from starting to use the interface during detach.
1090  */
1091 void
1092 if_detach(struct ifnet *ifp)
1093 {
1094 	bool found;
1095 
1096 	CURVNET_SET_QUIET(ifp->if_vnet);
1097 	found = if_unlink_ifnet(ifp, false);
1098 	if (found) {
1099 		sx_xlock(&ifnet_detach_sxlock);
1100 		if_detach_internal(ifp, 0, NULL);
1101 		sx_xunlock(&ifnet_detach_sxlock);
1102 	}
1103 	CURVNET_RESTORE();
1104 }
1105 
1106 /*
1107  * The vmove flag, if set, indicates that we are called from a callpath
1108  * that is moving an interface to a different vnet instance.
1109  *
1110  * The shutdown flag, if set, indicates that we are called in the
1111  * process of shutting down a vnet instance.  Currently only the
1112  * vnet_if_return SYSUNINIT function sets it.  Note: we can be called
1113  * on a vnet instance shutdown without this flag being set, e.g., when
1114  * the cloned interfaces are destoyed as first thing of teardown.
1115  */
1116 static int
1117 if_detach_internal(struct ifnet *ifp, int vmove, struct if_clone **ifcp)
1118 {
1119 	struct ifaddr *ifa;
1120 	int i;
1121 	struct domain *dp;
1122 #ifdef VIMAGE
1123 	bool shutdown;
1124 
1125 	shutdown = VNET_IS_SHUTTING_DOWN(ifp->if_vnet);
1126 #endif
1127 
1128 	/*
1129 	 * At this point we know the interface still was on the ifnet list
1130 	 * and we removed it so we are in a stable state.
1131 	 */
1132 	epoch_wait_preempt(net_epoch_preempt);
1133 
1134 	/*
1135 	 * Ensure all pending EPOCH(9) callbacks have been executed. This
1136 	 * fixes issues about late destruction of multicast options
1137 	 * which lead to leave group calls, which in turn access the
1138 	 * belonging ifnet structure:
1139 	 */
1140 	epoch_drain_callbacks(net_epoch_preempt);
1141 
1142 	/*
1143 	 * In any case (destroy or vmove) detach us from the groups
1144 	 * and remove/wait for pending events on the taskq.
1145 	 * XXX-BZ in theory an interface could still enqueue a taskq change?
1146 	 */
1147 	if_delgroups(ifp);
1148 
1149 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
1150 	taskqueue_drain(taskqueue_swi, &ifp->if_addmultitask);
1151 
1152 	/*
1153 	 * Check if this is a cloned interface or not. Must do even if
1154 	 * shutting down as a if_vmove_reclaim() would move the ifp and
1155 	 * the if_clone_addgroup() will have a corrupted string overwise
1156 	 * from a gibberish pointer.
1157 	 */
1158 	if (vmove && ifcp != NULL)
1159 		*ifcp = if_clone_findifc(ifp);
1160 
1161 	if_down(ifp);
1162 
1163 #ifdef VIMAGE
1164 	/*
1165 	 * On VNET shutdown abort here as the stack teardown will do all
1166 	 * the work top-down for us.
1167 	 */
1168 	if (shutdown) {
1169 		/* Give interface users the chance to clean up. */
1170 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
1171 
1172 		/*
1173 		 * In case of a vmove we are done here without error.
1174 		 * If we would signal an error it would lead to the same
1175 		 * abort as if we did not find the ifnet anymore.
1176 		 * if_detach() calls us in void context and does not care
1177 		 * about an early abort notification, so life is splendid :)
1178 		 */
1179 		goto finish_vnet_shutdown;
1180 	}
1181 #endif
1182 
1183 	/*
1184 	 * At this point we are not tearing down a VNET and are either
1185 	 * going to destroy or vmove the interface and have to cleanup
1186 	 * accordingly.
1187 	 */
1188 
1189 	/*
1190 	 * Remove routes and flush queues.
1191 	 */
1192 #ifdef ALTQ
1193 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1194 		altq_disable(&ifp->if_snd);
1195 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1196 		altq_detach(&ifp->if_snd);
1197 #endif
1198 
1199 	if_purgeaddrs(ifp);
1200 
1201 #ifdef INET
1202 	in_ifdetach(ifp);
1203 #endif
1204 
1205 #ifdef INET6
1206 	/*
1207 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
1208 	 * before removing routing entries below, since IPv6 interface direct
1209 	 * routes are expected to be removed by the IPv6-specific kernel API.
1210 	 * Otherwise, the kernel will detect some inconsistency and bark it.
1211 	 */
1212 	in6_ifdetach(ifp);
1213 #endif
1214 	if_purgemaddrs(ifp);
1215 
1216 	/* Announce that the interface is gone. */
1217 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1218 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
1219 	if (IS_DEFAULT_VNET(curvnet))
1220 		devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
1221 
1222 	if (!vmove) {
1223 		/*
1224 		 * Prevent further calls into the device driver via ifnet.
1225 		 */
1226 		if_dead(ifp);
1227 
1228 		/*
1229 		 * Clean up all addresses.
1230 		 */
1231 		IF_ADDR_WLOCK(ifp);
1232 		if (!CK_STAILQ_EMPTY(&ifp->if_addrhead)) {
1233 			ifa = CK_STAILQ_FIRST(&ifp->if_addrhead);
1234 			CK_STAILQ_REMOVE(&ifp->if_addrhead, ifa, ifaddr, ifa_link);
1235 			IF_ADDR_WUNLOCK(ifp);
1236 			ifa_free(ifa);
1237 		} else
1238 			IF_ADDR_WUNLOCK(ifp);
1239 	}
1240 
1241 	rt_flushifroutes(ifp);
1242 
1243 #ifdef VIMAGE
1244 finish_vnet_shutdown:
1245 #endif
1246 	/*
1247 	 * We cannot hold the lock over dom_ifdetach calls as they might
1248 	 * sleep, for example trying to drain a callout, thus open up the
1249 	 * theoretical race with re-attaching.
1250 	 */
1251 	IF_AFDATA_LOCK(ifp);
1252 	i = ifp->if_afdata_initialized;
1253 	ifp->if_afdata_initialized = 0;
1254 	IF_AFDATA_UNLOCK(ifp);
1255 	for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
1256 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family]) {
1257 			(*dp->dom_ifdetach)(ifp,
1258 			    ifp->if_afdata[dp->dom_family]);
1259 			ifp->if_afdata[dp->dom_family] = NULL;
1260 		}
1261 	}
1262 
1263 	return (0);
1264 }
1265 
1266 #ifdef VIMAGE
1267 /*
1268  * if_vmove() performs a limited version of if_detach() in current
1269  * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
1270  */
1271 static int
1272 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
1273 {
1274 	struct if_clone *ifc;
1275 #ifdef DEV_BPF
1276 	u_int bif_dlt, bif_hdrlen;
1277 #endif
1278 	int rc;
1279 
1280 #ifdef DEV_BPF
1281  	/*
1282 	 * if_detach_internal() will call the eventhandler to notify
1283 	 * interface departure.  That will detach if_bpf.  We need to
1284 	 * safe the dlt and hdrlen so we can re-attach it later.
1285 	 */
1286 	bpf_get_bp_params(ifp->if_bpf, &bif_dlt, &bif_hdrlen);
1287 #endif
1288 
1289 	/*
1290 	 * Detach from current vnet, but preserve LLADDR info, do not
1291 	 * mark as dead etc. so that the ifnet can be reattached later.
1292 	 * If we cannot find it, we lost the race to someone else.
1293 	 */
1294 	rc = if_detach_internal(ifp, 1, &ifc);
1295 	if (rc != 0)
1296 		return (rc);
1297 
1298 	/*
1299 	 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
1300 	 * the if_index for that vnet if possible.
1301 	 *
1302 	 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
1303 	 * or we'd lock on one vnet and unlock on another.
1304 	 */
1305 	IFNET_WLOCK();
1306 	ifindex_free(ifp->if_index);
1307 	IFNET_WUNLOCK();
1308 
1309 	/*
1310 	 * Perform interface-specific reassignment tasks, if provided by
1311 	 * the driver.
1312 	 */
1313 	if (ifp->if_reassign != NULL)
1314 		ifp->if_reassign(ifp, new_vnet, NULL);
1315 
1316 	/*
1317 	 * Switch to the context of the target vnet.
1318 	 */
1319 	CURVNET_SET_QUIET(new_vnet);
1320 	ifindex_alloc(ifp);
1321 	if_attach_internal(ifp, 1, ifc);
1322 
1323 #ifdef DEV_BPF
1324 	if (ifp->if_bpf == NULL)
1325 		bpfattach(ifp, bif_dlt, bif_hdrlen);
1326 #endif
1327 
1328 	CURVNET_RESTORE();
1329 	return (0);
1330 }
1331 
1332 /*
1333  * Move an ifnet to or from another child prison/vnet, specified by the jail id.
1334  */
1335 static int
1336 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
1337 {
1338 	struct prison *pr;
1339 	struct ifnet *difp;
1340 	int error;
1341 	bool found __diagused;
1342 	bool shutdown;
1343 
1344 	/* Try to find the prison within our visibility. */
1345 	sx_slock(&allprison_lock);
1346 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1347 	sx_sunlock(&allprison_lock);
1348 	if (pr == NULL)
1349 		return (ENXIO);
1350 	prison_hold_locked(pr);
1351 	mtx_unlock(&pr->pr_mtx);
1352 
1353 	/* Do not try to move the iface from and to the same prison. */
1354 	if (pr->pr_vnet == ifp->if_vnet) {
1355 		prison_free(pr);
1356 		return (EEXIST);
1357 	}
1358 
1359 	/* Make sure the named iface does not exists in the dst. prison/vnet. */
1360 	/* XXX Lock interfaces to avoid races. */
1361 	CURVNET_SET_QUIET(pr->pr_vnet);
1362 	difp = ifunit(ifname);
1363 	if (difp != NULL) {
1364 		CURVNET_RESTORE();
1365 		prison_free(pr);
1366 		return (EEXIST);
1367 	}
1368 
1369 	/* Make sure the VNET is stable. */
1370 	shutdown = VNET_IS_SHUTTING_DOWN(ifp->if_vnet);
1371 	if (shutdown) {
1372 		CURVNET_RESTORE();
1373 		prison_free(pr);
1374 		return (EBUSY);
1375 	}
1376 	CURVNET_RESTORE();
1377 
1378 	found = if_unlink_ifnet(ifp, true);
1379 	MPASS(found);
1380 
1381 	/* Move the interface into the child jail/vnet. */
1382 	error = if_vmove(ifp, pr->pr_vnet);
1383 
1384 	/* Report the new if_xname back to the userland on success. */
1385 	if (error == 0)
1386 		sprintf(ifname, "%s", ifp->if_xname);
1387 
1388 	prison_free(pr);
1389 	return (error);
1390 }
1391 
1392 static int
1393 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1394 {
1395 	struct prison *pr;
1396 	struct vnet *vnet_dst;
1397 	struct ifnet *ifp;
1398 	int error, found __diagused;
1399  	bool shutdown;
1400 
1401 	/* Try to find the prison within our visibility. */
1402 	sx_slock(&allprison_lock);
1403 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1404 	sx_sunlock(&allprison_lock);
1405 	if (pr == NULL)
1406 		return (ENXIO);
1407 	prison_hold_locked(pr);
1408 	mtx_unlock(&pr->pr_mtx);
1409 
1410 	/* Make sure the named iface exists in the source prison/vnet. */
1411 	CURVNET_SET(pr->pr_vnet);
1412 	ifp = ifunit(ifname);		/* XXX Lock to avoid races. */
1413 	if (ifp == NULL) {
1414 		CURVNET_RESTORE();
1415 		prison_free(pr);
1416 		return (ENXIO);
1417 	}
1418 
1419 	/* Do not try to move the iface from and to the same prison. */
1420 	vnet_dst = TD_TO_VNET(td);
1421 	if (vnet_dst == ifp->if_vnet) {
1422 		CURVNET_RESTORE();
1423 		prison_free(pr);
1424 		return (EEXIST);
1425 	}
1426 
1427 	/* Make sure the VNET is stable. */
1428 	shutdown = VNET_IS_SHUTTING_DOWN(ifp->if_vnet);
1429 	if (shutdown) {
1430 		CURVNET_RESTORE();
1431 		prison_free(pr);
1432 		return (EBUSY);
1433 	}
1434 
1435 	/* Get interface back from child jail/vnet. */
1436 	found = if_unlink_ifnet(ifp, true);
1437 	MPASS(found);
1438 	error = if_vmove(ifp, vnet_dst);
1439 	CURVNET_RESTORE();
1440 
1441 	/* Report the new if_xname back to the userland on success. */
1442 	if (error == 0)
1443 		sprintf(ifname, "%s", ifp->if_xname);
1444 
1445 	prison_free(pr);
1446 	return (error);
1447 }
1448 #endif /* VIMAGE */
1449 
1450 /*
1451  * Add a group to an interface
1452  */
1453 int
1454 if_addgroup(struct ifnet *ifp, const char *groupname)
1455 {
1456 	struct ifg_list		*ifgl;
1457 	struct ifg_group	*ifg = NULL;
1458 	struct ifg_member	*ifgm;
1459 	int 			 new = 0;
1460 
1461 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1462 	    groupname[strlen(groupname) - 1] <= '9')
1463 		return (EINVAL);
1464 
1465 	IFNET_WLOCK();
1466 	CK_STAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1467 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1468 			IFNET_WUNLOCK();
1469 			return (EEXIST);
1470 		}
1471 
1472 	if ((ifgl = malloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL) {
1473 	    	IFNET_WUNLOCK();
1474 		return (ENOMEM);
1475 	}
1476 
1477 	if ((ifgm = malloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
1478 		free(ifgl, M_TEMP);
1479 		IFNET_WUNLOCK();
1480 		return (ENOMEM);
1481 	}
1482 
1483 	CK_STAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1484 		if (!strcmp(ifg->ifg_group, groupname))
1485 			break;
1486 
1487 	if (ifg == NULL) {
1488 		if ((ifg = malloc(sizeof(*ifg), M_TEMP, M_NOWAIT)) == NULL) {
1489 			free(ifgl, M_TEMP);
1490 			free(ifgm, M_TEMP);
1491 			IFNET_WUNLOCK();
1492 			return (ENOMEM);
1493 		}
1494 		strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1495 		ifg->ifg_refcnt = 0;
1496 		CK_STAILQ_INIT(&ifg->ifg_members);
1497 		CK_STAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1498 		new = 1;
1499 	}
1500 
1501 	ifg->ifg_refcnt++;
1502 	ifgl->ifgl_group = ifg;
1503 	ifgm->ifgm_ifp = ifp;
1504 
1505 	IF_ADDR_WLOCK(ifp);
1506 	CK_STAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1507 	CK_STAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1508 	IF_ADDR_WUNLOCK(ifp);
1509 
1510 	IFNET_WUNLOCK();
1511 
1512 	if (new)
1513 		EVENTHANDLER_INVOKE(group_attach_event, ifg);
1514 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1515 
1516 	return (0);
1517 }
1518 
1519 /*
1520  * Helper function to remove a group out of an interface.  Expects the global
1521  * ifnet lock to be write-locked, and drops it before returning.
1522  */
1523 static void
1524 _if_delgroup_locked(struct ifnet *ifp, struct ifg_list *ifgl,
1525     const char *groupname)
1526 {
1527 	struct ifg_member *ifgm;
1528 	bool freeifgl;
1529 
1530 	IFNET_WLOCK_ASSERT();
1531 
1532 	IF_ADDR_WLOCK(ifp);
1533 	CK_STAILQ_REMOVE(&ifp->if_groups, ifgl, ifg_list, ifgl_next);
1534 	IF_ADDR_WUNLOCK(ifp);
1535 
1536 	CK_STAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next) {
1537 		if (ifgm->ifgm_ifp == ifp) {
1538 			CK_STAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1539 			    ifg_member, ifgm_next);
1540 			break;
1541 		}
1542 	}
1543 
1544 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1545 		CK_STAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_group,
1546 		    ifg_next);
1547 		freeifgl = true;
1548 	} else {
1549 		freeifgl = false;
1550 	}
1551 	IFNET_WUNLOCK();
1552 
1553 	epoch_wait_preempt(net_epoch_preempt);
1554 	if (freeifgl) {
1555 		EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1556 		free(ifgl->ifgl_group, M_TEMP);
1557 	}
1558 	free(ifgm, M_TEMP);
1559 	free(ifgl, M_TEMP);
1560 
1561 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1562 }
1563 
1564 /*
1565  * Remove a group from an interface
1566  */
1567 int
1568 if_delgroup(struct ifnet *ifp, const char *groupname)
1569 {
1570 	struct ifg_list *ifgl;
1571 
1572 	IFNET_WLOCK();
1573 	CK_STAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1574 		if (strcmp(ifgl->ifgl_group->ifg_group, groupname) == 0)
1575 			break;
1576 	if (ifgl == NULL) {
1577 		IFNET_WUNLOCK();
1578 		return (ENOENT);
1579 	}
1580 
1581 	_if_delgroup_locked(ifp, ifgl, groupname);
1582 
1583 	return (0);
1584 }
1585 
1586 /*
1587  * Remove an interface from all groups
1588  */
1589 static void
1590 if_delgroups(struct ifnet *ifp)
1591 {
1592 	struct ifg_list *ifgl;
1593 	char groupname[IFNAMSIZ];
1594 
1595 	IFNET_WLOCK();
1596 	while ((ifgl = CK_STAILQ_FIRST(&ifp->if_groups)) != NULL) {
1597 		strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1598 		_if_delgroup_locked(ifp, ifgl, groupname);
1599 		IFNET_WLOCK();
1600 	}
1601 	IFNET_WUNLOCK();
1602 }
1603 
1604 /*
1605  * Stores all groups from an interface in memory pointed to by ifgr.
1606  */
1607 static int
1608 if_getgroup(struct ifgroupreq *ifgr, struct ifnet *ifp)
1609 {
1610 	int			 len, error;
1611 	struct ifg_list		*ifgl;
1612 	struct ifg_req		 ifgrq, *ifgp;
1613 
1614 	NET_EPOCH_ASSERT();
1615 
1616 	if (ifgr->ifgr_len == 0) {
1617 		CK_STAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1618 			ifgr->ifgr_len += sizeof(struct ifg_req);
1619 		return (0);
1620 	}
1621 
1622 	len = ifgr->ifgr_len;
1623 	ifgp = ifgr->ifgr_groups;
1624 	/* XXX: wire */
1625 	CK_STAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1626 		if (len < sizeof(ifgrq))
1627 			return (EINVAL);
1628 		bzero(&ifgrq, sizeof ifgrq);
1629 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1630 		    sizeof(ifgrq.ifgrq_group));
1631 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req))))
1632 			return (error);
1633 		len -= sizeof(ifgrq);
1634 		ifgp++;
1635 	}
1636 
1637 	return (0);
1638 }
1639 
1640 /*
1641  * Stores all members of a group in memory pointed to by igfr
1642  */
1643 static int
1644 if_getgroupmembers(struct ifgroupreq *ifgr)
1645 {
1646 	struct ifg_group	*ifg;
1647 	struct ifg_member	*ifgm;
1648 	struct ifg_req		 ifgrq, *ifgp;
1649 	int			 len, error;
1650 
1651 	IFNET_RLOCK();
1652 	CK_STAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1653 		if (strcmp(ifg->ifg_group, ifgr->ifgr_name) == 0)
1654 			break;
1655 	if (ifg == NULL) {
1656 		IFNET_RUNLOCK();
1657 		return (ENOENT);
1658 	}
1659 
1660 	if (ifgr->ifgr_len == 0) {
1661 		CK_STAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1662 			ifgr->ifgr_len += sizeof(ifgrq);
1663 		IFNET_RUNLOCK();
1664 		return (0);
1665 	}
1666 
1667 	len = ifgr->ifgr_len;
1668 	ifgp = ifgr->ifgr_groups;
1669 	CK_STAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1670 		if (len < sizeof(ifgrq)) {
1671 			IFNET_RUNLOCK();
1672 			return (EINVAL);
1673 		}
1674 		bzero(&ifgrq, sizeof ifgrq);
1675 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1676 		    sizeof(ifgrq.ifgrq_member));
1677 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1678 			IFNET_RUNLOCK();
1679 			return (error);
1680 		}
1681 		len -= sizeof(ifgrq);
1682 		ifgp++;
1683 	}
1684 	IFNET_RUNLOCK();
1685 
1686 	return (0);
1687 }
1688 
1689 /*
1690  * Return counter values from counter(9)s stored in ifnet.
1691  */
1692 uint64_t
1693 if_get_counter_default(struct ifnet *ifp, ift_counter cnt)
1694 {
1695 
1696 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1697 
1698 	return (counter_u64_fetch(ifp->if_counters[cnt]));
1699 }
1700 
1701 /*
1702  * Increase an ifnet counter. Usually used for counters shared
1703  * between the stack and a driver, but function supports them all.
1704  */
1705 void
1706 if_inc_counter(struct ifnet *ifp, ift_counter cnt, int64_t inc)
1707 {
1708 
1709 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1710 
1711 	counter_u64_add(ifp->if_counters[cnt], inc);
1712 }
1713 
1714 /*
1715  * Copy data from ifnet to userland API structure if_data.
1716  */
1717 void
1718 if_data_copy(struct ifnet *ifp, struct if_data *ifd)
1719 {
1720 
1721 	ifd->ifi_type = ifp->if_type;
1722 	ifd->ifi_physical = 0;
1723 	ifd->ifi_addrlen = ifp->if_addrlen;
1724 	ifd->ifi_hdrlen = ifp->if_hdrlen;
1725 	ifd->ifi_link_state = ifp->if_link_state;
1726 	ifd->ifi_vhid = 0;
1727 	ifd->ifi_datalen = sizeof(struct if_data);
1728 	ifd->ifi_mtu = ifp->if_mtu;
1729 	ifd->ifi_metric = ifp->if_metric;
1730 	ifd->ifi_baudrate = ifp->if_baudrate;
1731 	ifd->ifi_hwassist = ifp->if_hwassist;
1732 	ifd->ifi_epoch = ifp->if_epoch;
1733 	ifd->ifi_lastchange = ifp->if_lastchange;
1734 
1735 	ifd->ifi_ipackets = ifp->if_get_counter(ifp, IFCOUNTER_IPACKETS);
1736 	ifd->ifi_ierrors = ifp->if_get_counter(ifp, IFCOUNTER_IERRORS);
1737 	ifd->ifi_opackets = ifp->if_get_counter(ifp, IFCOUNTER_OPACKETS);
1738 	ifd->ifi_oerrors = ifp->if_get_counter(ifp, IFCOUNTER_OERRORS);
1739 	ifd->ifi_collisions = ifp->if_get_counter(ifp, IFCOUNTER_COLLISIONS);
1740 	ifd->ifi_ibytes = ifp->if_get_counter(ifp, IFCOUNTER_IBYTES);
1741 	ifd->ifi_obytes = ifp->if_get_counter(ifp, IFCOUNTER_OBYTES);
1742 	ifd->ifi_imcasts = ifp->if_get_counter(ifp, IFCOUNTER_IMCASTS);
1743 	ifd->ifi_omcasts = ifp->if_get_counter(ifp, IFCOUNTER_OMCASTS);
1744 	ifd->ifi_iqdrops = ifp->if_get_counter(ifp, IFCOUNTER_IQDROPS);
1745 	ifd->ifi_oqdrops = ifp->if_get_counter(ifp, IFCOUNTER_OQDROPS);
1746 	ifd->ifi_noproto = ifp->if_get_counter(ifp, IFCOUNTER_NOPROTO);
1747 }
1748 
1749 /*
1750  * Initialization, destruction and refcounting functions for ifaddrs.
1751  */
1752 struct ifaddr *
1753 ifa_alloc(size_t size, int flags)
1754 {
1755 	struct ifaddr *ifa;
1756 
1757 	KASSERT(size >= sizeof(struct ifaddr),
1758 	    ("%s: invalid size %zu", __func__, size));
1759 
1760 	ifa = malloc(size, M_IFADDR, M_ZERO | flags);
1761 	if (ifa == NULL)
1762 		return (NULL);
1763 
1764 	if ((ifa->ifa_opackets = counter_u64_alloc(flags)) == NULL)
1765 		goto fail;
1766 	if ((ifa->ifa_ipackets = counter_u64_alloc(flags)) == NULL)
1767 		goto fail;
1768 	if ((ifa->ifa_obytes = counter_u64_alloc(flags)) == NULL)
1769 		goto fail;
1770 	if ((ifa->ifa_ibytes = counter_u64_alloc(flags)) == NULL)
1771 		goto fail;
1772 
1773 	refcount_init(&ifa->ifa_refcnt, 1);
1774 
1775 	return (ifa);
1776 
1777 fail:
1778 	/* free(NULL) is okay */
1779 	counter_u64_free(ifa->ifa_opackets);
1780 	counter_u64_free(ifa->ifa_ipackets);
1781 	counter_u64_free(ifa->ifa_obytes);
1782 	counter_u64_free(ifa->ifa_ibytes);
1783 	free(ifa, M_IFADDR);
1784 
1785 	return (NULL);
1786 }
1787 
1788 void
1789 ifa_ref(struct ifaddr *ifa)
1790 {
1791 	u_int old __diagused;
1792 
1793 	old = refcount_acquire(&ifa->ifa_refcnt);
1794 	KASSERT(old > 0, ("%s: ifa %p has 0 refs", __func__, ifa));
1795 }
1796 
1797 int
1798 ifa_try_ref(struct ifaddr *ifa)
1799 {
1800 
1801 	NET_EPOCH_ASSERT();
1802 	return (refcount_acquire_if_not_zero(&ifa->ifa_refcnt));
1803 }
1804 
1805 static void
1806 ifa_destroy(epoch_context_t ctx)
1807 {
1808 	struct ifaddr *ifa;
1809 
1810 	ifa = __containerof(ctx, struct ifaddr, ifa_epoch_ctx);
1811 	counter_u64_free(ifa->ifa_opackets);
1812 	counter_u64_free(ifa->ifa_ipackets);
1813 	counter_u64_free(ifa->ifa_obytes);
1814 	counter_u64_free(ifa->ifa_ibytes);
1815 	free(ifa, M_IFADDR);
1816 }
1817 
1818 void
1819 ifa_free(struct ifaddr *ifa)
1820 {
1821 
1822 	if (refcount_release(&ifa->ifa_refcnt))
1823 		NET_EPOCH_CALL(ifa_destroy, &ifa->ifa_epoch_ctx);
1824 }
1825 
1826 /*
1827  * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1828  * structs used to represent other address families, it is necessary
1829  * to perform a different comparison.
1830  */
1831 
1832 #define	sa_dl_equal(a1, a2)	\
1833 	((((const struct sockaddr_dl *)(a1))->sdl_len ==		\
1834 	 ((const struct sockaddr_dl *)(a2))->sdl_len) &&		\
1835 	 (bcmp(CLLADDR((const struct sockaddr_dl *)(a1)),		\
1836 	       CLLADDR((const struct sockaddr_dl *)(a2)),		\
1837 	       ((const struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1838 
1839 /*
1840  * Locate an interface based on a complete address.
1841  */
1842 /*ARGSUSED*/
1843 struct ifaddr *
1844 ifa_ifwithaddr(const struct sockaddr *addr)
1845 {
1846 	struct ifnet *ifp;
1847 	struct ifaddr *ifa;
1848 
1849 	NET_EPOCH_ASSERT();
1850 
1851 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1852 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1853 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1854 				continue;
1855 			if (sa_equal(addr, ifa->ifa_addr)) {
1856 				goto done;
1857 			}
1858 			/* IP6 doesn't have broadcast */
1859 			if ((ifp->if_flags & IFF_BROADCAST) &&
1860 			    ifa->ifa_broadaddr &&
1861 			    ifa->ifa_broadaddr->sa_len != 0 &&
1862 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1863 				goto done;
1864 			}
1865 		}
1866 	}
1867 	ifa = NULL;
1868 done:
1869 	return (ifa);
1870 }
1871 
1872 int
1873 ifa_ifwithaddr_check(const struct sockaddr *addr)
1874 {
1875 	struct epoch_tracker et;
1876 	int rc;
1877 
1878 	NET_EPOCH_ENTER(et);
1879 	rc = (ifa_ifwithaddr(addr) != NULL);
1880 	NET_EPOCH_EXIT(et);
1881 	return (rc);
1882 }
1883 
1884 /*
1885  * Locate an interface based on the broadcast address.
1886  */
1887 /* ARGSUSED */
1888 struct ifaddr *
1889 ifa_ifwithbroadaddr(const struct sockaddr *addr, int fibnum)
1890 {
1891 	struct ifnet *ifp;
1892 	struct ifaddr *ifa;
1893 
1894 	NET_EPOCH_ASSERT();
1895 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1896 		if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum))
1897 			continue;
1898 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1899 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1900 				continue;
1901 			if ((ifp->if_flags & IFF_BROADCAST) &&
1902 			    ifa->ifa_broadaddr &&
1903 			    ifa->ifa_broadaddr->sa_len != 0 &&
1904 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1905 				goto done;
1906 			}
1907 		}
1908 	}
1909 	ifa = NULL;
1910 done:
1911 	return (ifa);
1912 }
1913 
1914 /*
1915  * Locate the point to point interface with a given destination address.
1916  */
1917 /*ARGSUSED*/
1918 struct ifaddr *
1919 ifa_ifwithdstaddr(const struct sockaddr *addr, int fibnum)
1920 {
1921 	struct ifnet *ifp;
1922 	struct ifaddr *ifa;
1923 
1924 	NET_EPOCH_ASSERT();
1925 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1926 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1927 			continue;
1928 		if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum))
1929 			continue;
1930 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1931 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1932 				continue;
1933 			if (ifa->ifa_dstaddr != NULL &&
1934 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1935 				goto done;
1936 			}
1937 		}
1938 	}
1939 	ifa = NULL;
1940 done:
1941 	return (ifa);
1942 }
1943 
1944 /*
1945  * Find an interface on a specific network.  If many, choice
1946  * is most specific found.
1947  */
1948 struct ifaddr *
1949 ifa_ifwithnet(const struct sockaddr *addr, int ignore_ptp, int fibnum)
1950 {
1951 	struct ifnet *ifp;
1952 	struct ifaddr *ifa;
1953 	struct ifaddr *ifa_maybe = NULL;
1954 	u_int af = addr->sa_family;
1955 	const char *addr_data = addr->sa_data, *cplim;
1956 	const struct sockaddr_dl *sdl;
1957 
1958 	NET_EPOCH_ASSERT();
1959 	/*
1960 	 * AF_LINK addresses can be looked up directly by their index number,
1961 	 * so do that if we can.
1962 	 */
1963 	if (af == AF_LINK) {
1964 		sdl = (const struct sockaddr_dl *)addr;
1965 		if (sdl->sdl_index && sdl->sdl_index <= V_if_index) {
1966 			ifp = ifnet_byindex(sdl->sdl_index);
1967 			if (ifp == NULL)
1968 				return (NULL);
1969 
1970 			return (ifp->if_addr);
1971 		}
1972 	}
1973 
1974 	/*
1975 	 * Scan though each interface, looking for ones that have addresses
1976 	 * in this address family and the requested fib.
1977 	 */
1978 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1979 		if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum))
1980 			continue;
1981 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1982 			const char *cp, *cp2, *cp3;
1983 
1984 			if (ifa->ifa_addr->sa_family != af)
1985 next:				continue;
1986 			if (af == AF_INET &&
1987 			    ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1988 				/*
1989 				 * This is a bit broken as it doesn't
1990 				 * take into account that the remote end may
1991 				 * be a single node in the network we are
1992 				 * looking for.
1993 				 * The trouble is that we don't know the
1994 				 * netmask for the remote end.
1995 				 */
1996 				if (ifa->ifa_dstaddr != NULL &&
1997 				    sa_equal(addr, ifa->ifa_dstaddr)) {
1998 					goto done;
1999 				}
2000 			} else {
2001 				/*
2002 				 * Scan all the bits in the ifa's address.
2003 				 * If a bit dissagrees with what we are
2004 				 * looking for, mask it with the netmask
2005 				 * to see if it really matters.
2006 				 * (A byte at a time)
2007 				 */
2008 				if (ifa->ifa_netmask == 0)
2009 					continue;
2010 				cp = addr_data;
2011 				cp2 = ifa->ifa_addr->sa_data;
2012 				cp3 = ifa->ifa_netmask->sa_data;
2013 				cplim = ifa->ifa_netmask->sa_len
2014 					+ (char *)ifa->ifa_netmask;
2015 				while (cp3 < cplim)
2016 					if ((*cp++ ^ *cp2++) & *cp3++)
2017 						goto next; /* next address! */
2018 				/*
2019 				 * If the netmask of what we just found
2020 				 * is more specific than what we had before
2021 				 * (if we had one), or if the virtual status
2022 				 * of new prefix is better than of the old one,
2023 				 * then remember the new one before continuing
2024 				 * to search for an even better one.
2025 				 */
2026 				if (ifa_maybe == NULL ||
2027 				    ifa_preferred(ifa_maybe, ifa) ||
2028 				    rn_refines((caddr_t)ifa->ifa_netmask,
2029 				    (caddr_t)ifa_maybe->ifa_netmask)) {
2030 					ifa_maybe = ifa;
2031 				}
2032 			}
2033 		}
2034 	}
2035 	ifa = ifa_maybe;
2036 	ifa_maybe = NULL;
2037 done:
2038 	return (ifa);
2039 }
2040 
2041 /*
2042  * Find an interface address specific to an interface best matching
2043  * a given address.
2044  */
2045 struct ifaddr *
2046 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
2047 {
2048 	struct ifaddr *ifa;
2049 	const char *cp, *cp2, *cp3;
2050 	char *cplim;
2051 	struct ifaddr *ifa_maybe = NULL;
2052 	u_int af = addr->sa_family;
2053 
2054 	if (af >= AF_MAX)
2055 		return (NULL);
2056 
2057 	NET_EPOCH_ASSERT();
2058 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2059 		if (ifa->ifa_addr->sa_family != af)
2060 			continue;
2061 		if (ifa_maybe == NULL)
2062 			ifa_maybe = ifa;
2063 		if (ifa->ifa_netmask == 0) {
2064 			if (sa_equal(addr, ifa->ifa_addr) ||
2065 			    (ifa->ifa_dstaddr &&
2066 			    sa_equal(addr, ifa->ifa_dstaddr)))
2067 				goto done;
2068 			continue;
2069 		}
2070 		if (ifp->if_flags & IFF_POINTOPOINT) {
2071 			if (sa_equal(addr, ifa->ifa_dstaddr))
2072 				goto done;
2073 		} else {
2074 			cp = addr->sa_data;
2075 			cp2 = ifa->ifa_addr->sa_data;
2076 			cp3 = ifa->ifa_netmask->sa_data;
2077 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
2078 			for (; cp3 < cplim; cp3++)
2079 				if ((*cp++ ^ *cp2++) & *cp3)
2080 					break;
2081 			if (cp3 == cplim)
2082 				goto done;
2083 		}
2084 	}
2085 	ifa = ifa_maybe;
2086 done:
2087 	return (ifa);
2088 }
2089 
2090 /*
2091  * See whether new ifa is better than current one:
2092  * 1) A non-virtual one is preferred over virtual.
2093  * 2) A virtual in master state preferred over any other state.
2094  *
2095  * Used in several address selecting functions.
2096  */
2097 int
2098 ifa_preferred(struct ifaddr *cur, struct ifaddr *next)
2099 {
2100 
2101 	return (cur->ifa_carp && (!next->ifa_carp ||
2102 	    ((*carp_master_p)(next) && !(*carp_master_p)(cur))));
2103 }
2104 
2105 struct sockaddr_dl *
2106 link_alloc_sdl(size_t size, int flags)
2107 {
2108 
2109 	return (malloc(size, M_TEMP, flags));
2110 }
2111 
2112 void
2113 link_free_sdl(struct sockaddr *sa)
2114 {
2115 	free(sa, M_TEMP);
2116 }
2117 
2118 /*
2119  * Fills in given sdl with interface basic info.
2120  * Returns pointer to filled sdl.
2121  */
2122 struct sockaddr_dl *
2123 link_init_sdl(struct ifnet *ifp, struct sockaddr *paddr, u_char iftype)
2124 {
2125 	struct sockaddr_dl *sdl;
2126 
2127 	sdl = (struct sockaddr_dl *)paddr;
2128 	memset(sdl, 0, sizeof(struct sockaddr_dl));
2129 	sdl->sdl_len = sizeof(struct sockaddr_dl);
2130 	sdl->sdl_family = AF_LINK;
2131 	sdl->sdl_index = ifp->if_index;
2132 	sdl->sdl_type = iftype;
2133 
2134 	return (sdl);
2135 }
2136 
2137 /*
2138  * Mark an interface down and notify protocols of
2139  * the transition.
2140  */
2141 static void
2142 if_unroute(struct ifnet *ifp, int flag, int fam)
2143 {
2144 	struct ifaddr *ifa;
2145 	struct epoch_tracker et;
2146 
2147 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
2148 
2149 	ifp->if_flags &= ~flag;
2150 	getmicrotime(&ifp->if_lastchange);
2151 	NET_EPOCH_ENTER(et);
2152 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
2153 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
2154 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2155 	NET_EPOCH_EXIT(et);
2156 	ifp->if_qflush(ifp);
2157 
2158 	if (ifp->if_carp)
2159 		(*carp_linkstate_p)(ifp);
2160 	rt_ifmsg(ifp);
2161 }
2162 
2163 /*
2164  * Mark an interface up and notify protocols of
2165  * the transition.
2166  */
2167 static void
2168 if_route(struct ifnet *ifp, int flag, int fam)
2169 {
2170 	struct ifaddr *ifa;
2171 	struct epoch_tracker et;
2172 
2173 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
2174 
2175 	ifp->if_flags |= flag;
2176 	getmicrotime(&ifp->if_lastchange);
2177 	NET_EPOCH_ENTER(et);
2178 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
2179 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
2180 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
2181 	NET_EPOCH_EXIT(et);
2182 	if (ifp->if_carp)
2183 		(*carp_linkstate_p)(ifp);
2184 	rt_ifmsg(ifp);
2185 #ifdef INET6
2186 	in6_if_up(ifp);
2187 #endif
2188 }
2189 
2190 void	(*vlan_link_state_p)(struct ifnet *);	/* XXX: private from if_vlan */
2191 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
2192 struct ifnet *(*vlan_trunkdev_p)(struct ifnet *);
2193 struct	ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t);
2194 int	(*vlan_tag_p)(struct ifnet *, uint16_t *);
2195 int	(*vlan_pcp_p)(struct ifnet *, uint16_t *);
2196 int	(*vlan_setcookie_p)(struct ifnet *, void *);
2197 void	*(*vlan_cookie_p)(struct ifnet *);
2198 
2199 /*
2200  * Handle a change in the interface link state. To avoid LORs
2201  * between driver lock and upper layer locks, as well as possible
2202  * recursions, we post event to taskqueue, and all job
2203  * is done in static do_link_state_change().
2204  */
2205 void
2206 if_link_state_change(struct ifnet *ifp, int link_state)
2207 {
2208 	/* Return if state hasn't changed. */
2209 	if (ifp->if_link_state == link_state)
2210 		return;
2211 
2212 	ifp->if_link_state = link_state;
2213 
2214 	/* XXXGL: reference ifp? */
2215 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
2216 }
2217 
2218 static void
2219 do_link_state_change(void *arg, int pending)
2220 {
2221 	struct ifnet *ifp;
2222 	int link_state;
2223 
2224 	ifp = arg;
2225 	link_state = ifp->if_link_state;
2226 
2227 	CURVNET_SET(ifp->if_vnet);
2228 	rt_ifmsg(ifp);
2229 	if (ifp->if_vlantrunk != NULL)
2230 		(*vlan_link_state_p)(ifp);
2231 
2232 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
2233 	    ifp->if_l2com != NULL)
2234 		(*ng_ether_link_state_p)(ifp, link_state);
2235 	if (ifp->if_carp)
2236 		(*carp_linkstate_p)(ifp);
2237 	if (ifp->if_bridge)
2238 		ifp->if_bridge_linkstate(ifp);
2239 	if (ifp->if_lagg)
2240 		(*lagg_linkstate_p)(ifp, link_state);
2241 
2242 	if (IS_DEFAULT_VNET(curvnet))
2243 		devctl_notify("IFNET", ifp->if_xname,
2244 		    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
2245 		    NULL);
2246 	if (pending > 1)
2247 		if_printf(ifp, "%d link states coalesced\n", pending);
2248 	if (log_link_state_change)
2249 		if_printf(ifp, "link state changed to %s\n",
2250 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
2251 	EVENTHANDLER_INVOKE(ifnet_link_event, ifp, link_state);
2252 	CURVNET_RESTORE();
2253 }
2254 
2255 /*
2256  * Mark an interface down and notify protocols of
2257  * the transition.
2258  */
2259 void
2260 if_down(struct ifnet *ifp)
2261 {
2262 
2263 	EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_DOWN);
2264 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
2265 }
2266 
2267 /*
2268  * Mark an interface up and notify protocols of
2269  * the transition.
2270  */
2271 void
2272 if_up(struct ifnet *ifp)
2273 {
2274 
2275 	if_route(ifp, IFF_UP, AF_UNSPEC);
2276 	EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_UP);
2277 }
2278 
2279 /*
2280  * Flush an interface queue.
2281  */
2282 void
2283 if_qflush(struct ifnet *ifp)
2284 {
2285 	struct mbuf *m, *n;
2286 	struct ifaltq *ifq;
2287 
2288 	ifq = &ifp->if_snd;
2289 	IFQ_LOCK(ifq);
2290 #ifdef ALTQ
2291 	if (ALTQ_IS_ENABLED(ifq))
2292 		ALTQ_PURGE(ifq);
2293 #endif
2294 	n = ifq->ifq_head;
2295 	while ((m = n) != NULL) {
2296 		n = m->m_nextpkt;
2297 		m_freem(m);
2298 	}
2299 	ifq->ifq_head = 0;
2300 	ifq->ifq_tail = 0;
2301 	ifq->ifq_len = 0;
2302 	IFQ_UNLOCK(ifq);
2303 }
2304 
2305 /*
2306  * Map interface name to interface structure pointer, with or without
2307  * returning a reference.
2308  */
2309 struct ifnet *
2310 ifunit_ref(const char *name)
2311 {
2312 	struct epoch_tracker et;
2313 	struct ifnet *ifp;
2314 
2315 	NET_EPOCH_ENTER(et);
2316 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2317 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2318 		    !(ifp->if_flags & IFF_DYING))
2319 			break;
2320 	}
2321 	if (ifp != NULL)
2322 		if_ref(ifp);
2323 	NET_EPOCH_EXIT(et);
2324 	return (ifp);
2325 }
2326 
2327 struct ifnet *
2328 ifunit(const char *name)
2329 {
2330 	struct epoch_tracker et;
2331 	struct ifnet *ifp;
2332 
2333 	NET_EPOCH_ENTER(et);
2334 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2335 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2336 			break;
2337 	}
2338 	NET_EPOCH_EXIT(et);
2339 	return (ifp);
2340 }
2341 
2342 void *
2343 ifr_buffer_get_buffer(void *data)
2344 {
2345 	union ifreq_union *ifrup;
2346 
2347 	ifrup = data;
2348 #ifdef COMPAT_FREEBSD32
2349 	if (SV_CURPROC_FLAG(SV_ILP32))
2350 		return ((void *)(uintptr_t)
2351 		    ifrup->ifr32.ifr_ifru.ifru_buffer.buffer);
2352 #endif
2353 	return (ifrup->ifr.ifr_ifru.ifru_buffer.buffer);
2354 }
2355 
2356 static void
2357 ifr_buffer_set_buffer_null(void *data)
2358 {
2359 	union ifreq_union *ifrup;
2360 
2361 	ifrup = data;
2362 #ifdef COMPAT_FREEBSD32
2363 	if (SV_CURPROC_FLAG(SV_ILP32))
2364 		ifrup->ifr32.ifr_ifru.ifru_buffer.buffer = 0;
2365 	else
2366 #endif
2367 		ifrup->ifr.ifr_ifru.ifru_buffer.buffer = NULL;
2368 }
2369 
2370 size_t
2371 ifr_buffer_get_length(void *data)
2372 {
2373 	union ifreq_union *ifrup;
2374 
2375 	ifrup = data;
2376 #ifdef COMPAT_FREEBSD32
2377 	if (SV_CURPROC_FLAG(SV_ILP32))
2378 		return (ifrup->ifr32.ifr_ifru.ifru_buffer.length);
2379 #endif
2380 	return (ifrup->ifr.ifr_ifru.ifru_buffer.length);
2381 }
2382 
2383 static void
2384 ifr_buffer_set_length(void *data, size_t len)
2385 {
2386 	union ifreq_union *ifrup;
2387 
2388 	ifrup = data;
2389 #ifdef COMPAT_FREEBSD32
2390 	if (SV_CURPROC_FLAG(SV_ILP32))
2391 		ifrup->ifr32.ifr_ifru.ifru_buffer.length = len;
2392 	else
2393 #endif
2394 		ifrup->ifr.ifr_ifru.ifru_buffer.length = len;
2395 }
2396 
2397 void *
2398 ifr_data_get_ptr(void *ifrp)
2399 {
2400 	union ifreq_union *ifrup;
2401 
2402 	ifrup = ifrp;
2403 #ifdef COMPAT_FREEBSD32
2404 	if (SV_CURPROC_FLAG(SV_ILP32))
2405 		return ((void *)(uintptr_t)
2406 		    ifrup->ifr32.ifr_ifru.ifru_data);
2407 #endif
2408 		return (ifrup->ifr.ifr_ifru.ifru_data);
2409 }
2410 
2411 /*
2412  * Hardware specific interface ioctls.
2413  */
2414 int
2415 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2416 {
2417 	struct ifreq *ifr;
2418 	int error = 0, do_ifup = 0;
2419 	int new_flags, temp_flags;
2420 	size_t namelen, onamelen;
2421 	size_t descrlen;
2422 	char *descrbuf, *odescrbuf;
2423 	char new_name[IFNAMSIZ];
2424 	char old_name[IFNAMSIZ], strbuf[IFNAMSIZ + 8];
2425 	struct ifaddr *ifa;
2426 	struct sockaddr_dl *sdl;
2427 
2428 	ifr = (struct ifreq *)data;
2429 	switch (cmd) {
2430 	case SIOCGIFINDEX:
2431 		ifr->ifr_index = ifp->if_index;
2432 		break;
2433 
2434 	case SIOCGIFFLAGS:
2435 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
2436 		ifr->ifr_flags = temp_flags & 0xffff;
2437 		ifr->ifr_flagshigh = temp_flags >> 16;
2438 		break;
2439 
2440 	case SIOCGIFCAP:
2441 		ifr->ifr_reqcap = ifp->if_capabilities;
2442 		ifr->ifr_curcap = ifp->if_capenable;
2443 		break;
2444 
2445 	case SIOCGIFDATA:
2446 	{
2447 		struct if_data ifd;
2448 
2449 		/* Ensure uninitialised padding is not leaked. */
2450 		memset(&ifd, 0, sizeof(ifd));
2451 
2452 		if_data_copy(ifp, &ifd);
2453 		error = copyout(&ifd, ifr_data_get_ptr(ifr), sizeof(ifd));
2454 		break;
2455 	}
2456 
2457 #ifdef MAC
2458 	case SIOCGIFMAC:
2459 		error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2460 		break;
2461 #endif
2462 
2463 	case SIOCGIFMETRIC:
2464 		ifr->ifr_metric = ifp->if_metric;
2465 		break;
2466 
2467 	case SIOCGIFMTU:
2468 		ifr->ifr_mtu = ifp->if_mtu;
2469 		break;
2470 
2471 	case SIOCGIFPHYS:
2472 		/* XXXGL: did this ever worked? */
2473 		ifr->ifr_phys = 0;
2474 		break;
2475 
2476 	case SIOCGIFDESCR:
2477 		error = 0;
2478 		sx_slock(&ifdescr_sx);
2479 		if (ifp->if_description == NULL)
2480 			error = ENOMSG;
2481 		else {
2482 			/* space for terminating nul */
2483 			descrlen = strlen(ifp->if_description) + 1;
2484 			if (ifr_buffer_get_length(ifr) < descrlen)
2485 				ifr_buffer_set_buffer_null(ifr);
2486 			else
2487 				error = copyout(ifp->if_description,
2488 				    ifr_buffer_get_buffer(ifr), descrlen);
2489 			ifr_buffer_set_length(ifr, descrlen);
2490 		}
2491 		sx_sunlock(&ifdescr_sx);
2492 		break;
2493 
2494 	case SIOCSIFDESCR:
2495 		error = priv_check(td, PRIV_NET_SETIFDESCR);
2496 		if (error)
2497 			return (error);
2498 
2499 		/*
2500 		 * Copy only (length-1) bytes to make sure that
2501 		 * if_description is always nul terminated.  The
2502 		 * length parameter is supposed to count the
2503 		 * terminating nul in.
2504 		 */
2505 		if (ifr_buffer_get_length(ifr) > ifdescr_maxlen)
2506 			return (ENAMETOOLONG);
2507 		else if (ifr_buffer_get_length(ifr) == 0)
2508 			descrbuf = NULL;
2509 		else {
2510 			descrbuf = malloc(ifr_buffer_get_length(ifr),
2511 			    M_IFDESCR, M_WAITOK | M_ZERO);
2512 			error = copyin(ifr_buffer_get_buffer(ifr), descrbuf,
2513 			    ifr_buffer_get_length(ifr) - 1);
2514 			if (error) {
2515 				free(descrbuf, M_IFDESCR);
2516 				break;
2517 			}
2518 		}
2519 
2520 		sx_xlock(&ifdescr_sx);
2521 		odescrbuf = ifp->if_description;
2522 		ifp->if_description = descrbuf;
2523 		sx_xunlock(&ifdescr_sx);
2524 
2525 		getmicrotime(&ifp->if_lastchange);
2526 		free(odescrbuf, M_IFDESCR);
2527 		break;
2528 
2529 	case SIOCGIFFIB:
2530 		ifr->ifr_fib = ifp->if_fib;
2531 		break;
2532 
2533 	case SIOCSIFFIB:
2534 		error = priv_check(td, PRIV_NET_SETIFFIB);
2535 		if (error)
2536 			return (error);
2537 		if (ifr->ifr_fib >= rt_numfibs)
2538 			return (EINVAL);
2539 
2540 		ifp->if_fib = ifr->ifr_fib;
2541 		break;
2542 
2543 	case SIOCSIFFLAGS:
2544 		error = priv_check(td, PRIV_NET_SETIFFLAGS);
2545 		if (error)
2546 			return (error);
2547 		/*
2548 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2549 		 * check, so we don't need special handling here yet.
2550 		 */
2551 		new_flags = (ifr->ifr_flags & 0xffff) |
2552 		    (ifr->ifr_flagshigh << 16);
2553 		if (ifp->if_flags & IFF_UP &&
2554 		    (new_flags & IFF_UP) == 0) {
2555 			if_down(ifp);
2556 		} else if (new_flags & IFF_UP &&
2557 		    (ifp->if_flags & IFF_UP) == 0) {
2558 			do_ifup = 1;
2559 		}
2560 		/* See if permanently promiscuous mode bit is about to flip */
2561 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2562 			if (new_flags & IFF_PPROMISC)
2563 				ifp->if_flags |= IFF_PROMISC;
2564 			else if (ifp->if_pcount == 0)
2565 				ifp->if_flags &= ~IFF_PROMISC;
2566 			if (log_promisc_mode_change)
2567                                 if_printf(ifp, "permanently promiscuous mode %s\n",
2568                                     ((new_flags & IFF_PPROMISC) ?
2569                                      "enabled" : "disabled"));
2570 		}
2571 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2572 			(new_flags &~ IFF_CANTCHANGE);
2573 		if (ifp->if_ioctl) {
2574 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
2575 		}
2576 		if (do_ifup)
2577 			if_up(ifp);
2578 		getmicrotime(&ifp->if_lastchange);
2579 		break;
2580 
2581 	case SIOCSIFCAP:
2582 		error = priv_check(td, PRIV_NET_SETIFCAP);
2583 		if (error)
2584 			return (error);
2585 		if (ifp->if_ioctl == NULL)
2586 			return (EOPNOTSUPP);
2587 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2588 			return (EINVAL);
2589 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2590 		if (error == 0)
2591 			getmicrotime(&ifp->if_lastchange);
2592 		break;
2593 
2594 #ifdef MAC
2595 	case SIOCSIFMAC:
2596 		error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2597 		break;
2598 #endif
2599 
2600 	case SIOCSIFNAME:
2601 		error = priv_check(td, PRIV_NET_SETIFNAME);
2602 		if (error)
2603 			return (error);
2604 		error = copyinstr(ifr_data_get_ptr(ifr), new_name, IFNAMSIZ,
2605 		    NULL);
2606 		if (error != 0)
2607 			return (error);
2608 		if (new_name[0] == '\0')
2609 			return (EINVAL);
2610 		if (strcmp(new_name, ifp->if_xname) == 0)
2611 			break;
2612 		if (ifunit(new_name) != NULL)
2613 			return (EEXIST);
2614 
2615 		/*
2616 		 * XXX: Locking.  Nothing else seems to lock if_flags,
2617 		 * and there are numerous other races with the
2618 		 * ifunit() checks not being atomic with namespace
2619 		 * changes (renames, vmoves, if_attach, etc).
2620 		 */
2621 		ifp->if_flags |= IFF_RENAMING;
2622 
2623 		/* Announce the departure of the interface. */
2624 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2625 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2626 
2627 		if_printf(ifp, "changing name to '%s'\n", new_name);
2628 
2629 		IF_ADDR_WLOCK(ifp);
2630 		strlcpy(old_name, ifp->if_xname, sizeof(old_name));
2631 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2632 		ifa = ifp->if_addr;
2633 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2634 		namelen = strlen(new_name);
2635 		onamelen = sdl->sdl_nlen;
2636 		/*
2637 		 * Move the address if needed.  This is safe because we
2638 		 * allocate space for a name of length IFNAMSIZ when we
2639 		 * create this in if_attach().
2640 		 */
2641 		if (namelen != onamelen) {
2642 			bcopy(sdl->sdl_data + onamelen,
2643 			    sdl->sdl_data + namelen, sdl->sdl_alen);
2644 		}
2645 		bcopy(new_name, sdl->sdl_data, namelen);
2646 		sdl->sdl_nlen = namelen;
2647 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2648 		bzero(sdl->sdl_data, onamelen);
2649 		while (namelen != 0)
2650 			sdl->sdl_data[--namelen] = 0xff;
2651 		IF_ADDR_WUNLOCK(ifp);
2652 
2653 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2654 		/* Announce the return of the interface. */
2655 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2656 
2657 		ifp->if_flags &= ~IFF_RENAMING;
2658 
2659 		snprintf(strbuf, sizeof(strbuf), "name=%s", new_name);
2660 		devctl_notify("IFNET", old_name, "RENAME", strbuf);
2661 		break;
2662 
2663 #ifdef VIMAGE
2664 	case SIOCSIFVNET:
2665 		error = priv_check(td, PRIV_NET_SETIFVNET);
2666 		if (error)
2667 			return (error);
2668 		error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2669 		break;
2670 #endif
2671 
2672 	case SIOCSIFMETRIC:
2673 		error = priv_check(td, PRIV_NET_SETIFMETRIC);
2674 		if (error)
2675 			return (error);
2676 		ifp->if_metric = ifr->ifr_metric;
2677 		getmicrotime(&ifp->if_lastchange);
2678 		break;
2679 
2680 	case SIOCSIFPHYS:
2681 		error = priv_check(td, PRIV_NET_SETIFPHYS);
2682 		if (error)
2683 			return (error);
2684 		if (ifp->if_ioctl == NULL)
2685 			return (EOPNOTSUPP);
2686 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2687 		if (error == 0)
2688 			getmicrotime(&ifp->if_lastchange);
2689 		break;
2690 
2691 	case SIOCSIFMTU:
2692 	{
2693 		u_long oldmtu = ifp->if_mtu;
2694 
2695 		error = priv_check(td, PRIV_NET_SETIFMTU);
2696 		if (error)
2697 			return (error);
2698 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2699 			return (EINVAL);
2700 		if (ifp->if_ioctl == NULL)
2701 			return (EOPNOTSUPP);
2702 		/* Disallow MTU changes on bridge member interfaces. */
2703 		if (ifp->if_bridge)
2704 			return (EOPNOTSUPP);
2705 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2706 		if (error == 0) {
2707 			getmicrotime(&ifp->if_lastchange);
2708 			rt_ifmsg(ifp);
2709 #ifdef INET
2710 			DEBUGNET_NOTIFY_MTU(ifp);
2711 #endif
2712 		}
2713 		/*
2714 		 * If the link MTU changed, do network layer specific procedure.
2715 		 */
2716 		if (ifp->if_mtu != oldmtu) {
2717 #ifdef INET6
2718 			nd6_setmtu(ifp);
2719 #endif
2720 			rt_updatemtu(ifp);
2721 		}
2722 		break;
2723 	}
2724 
2725 	case SIOCADDMULTI:
2726 	case SIOCDELMULTI:
2727 		if (cmd == SIOCADDMULTI)
2728 			error = priv_check(td, PRIV_NET_ADDMULTI);
2729 		else
2730 			error = priv_check(td, PRIV_NET_DELMULTI);
2731 		if (error)
2732 			return (error);
2733 
2734 		/* Don't allow group membership on non-multicast interfaces. */
2735 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2736 			return (EOPNOTSUPP);
2737 
2738 		/* Don't let users screw up protocols' entries. */
2739 		if (ifr->ifr_addr.sa_family != AF_LINK)
2740 			return (EINVAL);
2741 
2742 		if (cmd == SIOCADDMULTI) {
2743 			struct epoch_tracker et;
2744 			struct ifmultiaddr *ifma;
2745 
2746 			/*
2747 			 * Userland is only permitted to join groups once
2748 			 * via the if_addmulti() KPI, because it cannot hold
2749 			 * struct ifmultiaddr * between calls. It may also
2750 			 * lose a race while we check if the membership
2751 			 * already exists.
2752 			 */
2753 			NET_EPOCH_ENTER(et);
2754 			ifma = if_findmulti(ifp, &ifr->ifr_addr);
2755 			NET_EPOCH_EXIT(et);
2756 			if (ifma != NULL)
2757 				error = EADDRINUSE;
2758 			else
2759 				error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2760 		} else {
2761 			error = if_delmulti(ifp, &ifr->ifr_addr);
2762 		}
2763 		if (error == 0)
2764 			getmicrotime(&ifp->if_lastchange);
2765 		break;
2766 
2767 	case SIOCSIFPHYADDR:
2768 	case SIOCDIFPHYADDR:
2769 #ifdef INET6
2770 	case SIOCSIFPHYADDR_IN6:
2771 #endif
2772 	case SIOCSIFMEDIA:
2773 	case SIOCSIFGENERIC:
2774 		error = priv_check(td, PRIV_NET_HWIOCTL);
2775 		if (error)
2776 			return (error);
2777 		if (ifp->if_ioctl == NULL)
2778 			return (EOPNOTSUPP);
2779 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2780 		if (error == 0)
2781 			getmicrotime(&ifp->if_lastchange);
2782 		break;
2783 
2784 	case SIOCGIFSTATUS:
2785 	case SIOCGIFPSRCADDR:
2786 	case SIOCGIFPDSTADDR:
2787 	case SIOCGIFMEDIA:
2788 	case SIOCGIFXMEDIA:
2789 	case SIOCGIFGENERIC:
2790 	case SIOCGIFRSSKEY:
2791 	case SIOCGIFRSSHASH:
2792 	case SIOCGIFDOWNREASON:
2793 		if (ifp->if_ioctl == NULL)
2794 			return (EOPNOTSUPP);
2795 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2796 		break;
2797 
2798 	case SIOCSIFLLADDR:
2799 		error = priv_check(td, PRIV_NET_SETLLADDR);
2800 		if (error)
2801 			return (error);
2802 		error = if_setlladdr(ifp,
2803 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2804 		break;
2805 
2806 	case SIOCGHWADDR:
2807 		error = if_gethwaddr(ifp, ifr);
2808 		break;
2809 
2810 	case SIOCAIFGROUP:
2811 		error = priv_check(td, PRIV_NET_ADDIFGROUP);
2812 		if (error)
2813 			return (error);
2814 		error = if_addgroup(ifp,
2815 		    ((struct ifgroupreq *)data)->ifgr_group);
2816 		if (error != 0)
2817 			return (error);
2818 		break;
2819 
2820 	case SIOCGIFGROUP:
2821 	{
2822 		struct epoch_tracker et;
2823 
2824 		NET_EPOCH_ENTER(et);
2825 		error = if_getgroup((struct ifgroupreq *)data, ifp);
2826 		NET_EPOCH_EXIT(et);
2827 		break;
2828 	}
2829 
2830 	case SIOCDIFGROUP:
2831 		error = priv_check(td, PRIV_NET_DELIFGROUP);
2832 		if (error)
2833 			return (error);
2834 		error = if_delgroup(ifp,
2835 		    ((struct ifgroupreq *)data)->ifgr_group);
2836 		if (error != 0)
2837 			return (error);
2838 		break;
2839 
2840 	default:
2841 		error = ENOIOCTL;
2842 		break;
2843 	}
2844 	return (error);
2845 }
2846 
2847 /*
2848  * Interface ioctls.
2849  */
2850 int
2851 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2852 {
2853 #ifdef COMPAT_FREEBSD32
2854 	union {
2855 		struct ifconf ifc;
2856 		struct ifdrv ifd;
2857 		struct ifgroupreq ifgr;
2858 		struct ifmediareq ifmr;
2859 	} thunk;
2860 	u_long saved_cmd;
2861 	struct ifconf32 *ifc32;
2862 	struct ifdrv32 *ifd32;
2863 	struct ifgroupreq32 *ifgr32;
2864 	struct ifmediareq32 *ifmr32;
2865 #endif
2866 	struct ifnet *ifp;
2867 	struct ifreq *ifr;
2868 	int error;
2869 	int oif_flags;
2870 #ifdef VIMAGE
2871 	bool shutdown;
2872 #endif
2873 
2874 	CURVNET_SET(so->so_vnet);
2875 #ifdef VIMAGE
2876 	/* Make sure the VNET is stable. */
2877 	shutdown = VNET_IS_SHUTTING_DOWN(so->so_vnet);
2878 	if (shutdown) {
2879 		CURVNET_RESTORE();
2880 		return (EBUSY);
2881 	}
2882 #endif
2883 
2884 #ifdef COMPAT_FREEBSD32
2885 	saved_cmd = cmd;
2886 	switch (cmd) {
2887 	case SIOCGIFCONF32:
2888 		ifc32 = (struct ifconf32 *)data;
2889 		thunk.ifc.ifc_len = ifc32->ifc_len;
2890 		thunk.ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2891 		data = (caddr_t)&thunk.ifc;
2892 		cmd = SIOCGIFCONF;
2893 		break;
2894 	case SIOCGDRVSPEC32:
2895 	case SIOCSDRVSPEC32:
2896 		ifd32 = (struct ifdrv32 *)data;
2897 		memcpy(thunk.ifd.ifd_name, ifd32->ifd_name,
2898 		    sizeof(thunk.ifd.ifd_name));
2899 		thunk.ifd.ifd_cmd = ifd32->ifd_cmd;
2900 		thunk.ifd.ifd_len = ifd32->ifd_len;
2901 		thunk.ifd.ifd_data = PTRIN(ifd32->ifd_data);
2902 		data = (caddr_t)&thunk.ifd;
2903 		cmd = _IOC_NEWTYPE(cmd, struct ifdrv);
2904 		break;
2905 	case SIOCAIFGROUP32:
2906 	case SIOCGIFGROUP32:
2907 	case SIOCDIFGROUP32:
2908 	case SIOCGIFGMEMB32:
2909 		ifgr32 = (struct ifgroupreq32 *)data;
2910 		memcpy(thunk.ifgr.ifgr_name, ifgr32->ifgr_name,
2911 		    sizeof(thunk.ifgr.ifgr_name));
2912 		thunk.ifgr.ifgr_len = ifgr32->ifgr_len;
2913 		switch (cmd) {
2914 		case SIOCAIFGROUP32:
2915 		case SIOCDIFGROUP32:
2916 			memcpy(thunk.ifgr.ifgr_group, ifgr32->ifgr_group,
2917 			    sizeof(thunk.ifgr.ifgr_group));
2918 			break;
2919 		case SIOCGIFGROUP32:
2920 		case SIOCGIFGMEMB32:
2921 			thunk.ifgr.ifgr_groups = PTRIN(ifgr32->ifgr_groups);
2922 			break;
2923 		}
2924 		data = (caddr_t)&thunk.ifgr;
2925 		cmd = _IOC_NEWTYPE(cmd, struct ifgroupreq);
2926 		break;
2927 	case SIOCGIFMEDIA32:
2928 	case SIOCGIFXMEDIA32:
2929 		ifmr32 = (struct ifmediareq32 *)data;
2930 		memcpy(thunk.ifmr.ifm_name, ifmr32->ifm_name,
2931 		    sizeof(thunk.ifmr.ifm_name));
2932 		thunk.ifmr.ifm_current = ifmr32->ifm_current;
2933 		thunk.ifmr.ifm_mask = ifmr32->ifm_mask;
2934 		thunk.ifmr.ifm_status = ifmr32->ifm_status;
2935 		thunk.ifmr.ifm_active = ifmr32->ifm_active;
2936 		thunk.ifmr.ifm_count = ifmr32->ifm_count;
2937 		thunk.ifmr.ifm_ulist = PTRIN(ifmr32->ifm_ulist);
2938 		data = (caddr_t)&thunk.ifmr;
2939 		cmd = _IOC_NEWTYPE(cmd, struct ifmediareq);
2940 		break;
2941 	}
2942 #endif
2943 
2944 	switch (cmd) {
2945 	case SIOCGIFCONF:
2946 		error = ifconf(cmd, data);
2947 		goto out_noref;
2948 	}
2949 
2950 	ifr = (struct ifreq *)data;
2951 	switch (cmd) {
2952 #ifdef VIMAGE
2953 	case SIOCSIFRVNET:
2954 		error = priv_check(td, PRIV_NET_SETIFVNET);
2955 		if (error == 0)
2956 			error = if_vmove_reclaim(td, ifr->ifr_name,
2957 			    ifr->ifr_jid);
2958 		goto out_noref;
2959 #endif
2960 	case SIOCIFCREATE:
2961 	case SIOCIFCREATE2:
2962 		error = priv_check(td, PRIV_NET_IFCREATE);
2963 		if (error == 0)
2964 			error = if_clone_create(ifr->ifr_name,
2965 			    sizeof(ifr->ifr_name), cmd == SIOCIFCREATE2 ?
2966 			    ifr_data_get_ptr(ifr) : NULL);
2967 		goto out_noref;
2968 	case SIOCIFDESTROY:
2969 		error = priv_check(td, PRIV_NET_IFDESTROY);
2970 
2971 		if (error == 0) {
2972 			sx_xlock(&ifnet_detach_sxlock);
2973 			error = if_clone_destroy(ifr->ifr_name);
2974 			sx_xunlock(&ifnet_detach_sxlock);
2975 		}
2976 		goto out_noref;
2977 
2978 	case SIOCIFGCLONERS:
2979 		error = if_clone_list((struct if_clonereq *)data);
2980 		goto out_noref;
2981 
2982 	case SIOCGIFGMEMB:
2983 		error = if_getgroupmembers((struct ifgroupreq *)data);
2984 		goto out_noref;
2985 
2986 #if defined(INET) || defined(INET6)
2987 	case SIOCSVH:
2988 	case SIOCGVH:
2989 		if (carp_ioctl_p == NULL)
2990 			error = EPROTONOSUPPORT;
2991 		else
2992 			error = (*carp_ioctl_p)(ifr, cmd, td);
2993 		goto out_noref;
2994 #endif
2995 	}
2996 
2997 	ifp = ifunit_ref(ifr->ifr_name);
2998 	if (ifp == NULL) {
2999 		error = ENXIO;
3000 		goto out_noref;
3001 	}
3002 
3003 	error = ifhwioctl(cmd, ifp, data, td);
3004 	if (error != ENOIOCTL)
3005 		goto out_ref;
3006 
3007 	oif_flags = ifp->if_flags;
3008 	if (so->so_proto == NULL) {
3009 		error = EOPNOTSUPP;
3010 		goto out_ref;
3011 	}
3012 
3013 	/*
3014 	 * Pass the request on to the socket control method, and if the
3015 	 * latter returns EOPNOTSUPP, directly to the interface.
3016 	 *
3017 	 * Make an exception for the legacy SIOCSIF* requests.  Drivers
3018 	 * trust SIOCSIFADDR et al to come from an already privileged
3019 	 * layer, and do not perform any credentials checks or input
3020 	 * validation.
3021 	 */
3022 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd, data,
3023 	    ifp, td));
3024 	if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL &&
3025 	    cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR &&
3026 	    cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK)
3027 		error = (*ifp->if_ioctl)(ifp, cmd, data);
3028 
3029 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
3030 #ifdef INET6
3031 		if (ifp->if_flags & IFF_UP)
3032 			in6_if_up(ifp);
3033 #endif
3034 	}
3035 
3036 out_ref:
3037 	if_rele(ifp);
3038 out_noref:
3039 	CURVNET_RESTORE();
3040 #ifdef COMPAT_FREEBSD32
3041 	if (error != 0)
3042 		return (error);
3043 	switch (saved_cmd) {
3044 	case SIOCGIFCONF32:
3045 		ifc32->ifc_len = thunk.ifc.ifc_len;
3046 		break;
3047 	case SIOCGDRVSPEC32:
3048 		/*
3049 		 * SIOCGDRVSPEC is IOWR, but nothing actually touches
3050 		 * the struct so just assert that ifd_len (the only
3051 		 * field it might make sense to update) hasn't
3052 		 * changed.
3053 		 */
3054 		KASSERT(thunk.ifd.ifd_len == ifd32->ifd_len,
3055 		    ("ifd_len was updated %u -> %zu", ifd32->ifd_len,
3056 			thunk.ifd.ifd_len));
3057 		break;
3058 	case SIOCGIFGROUP32:
3059 	case SIOCGIFGMEMB32:
3060 		ifgr32->ifgr_len = thunk.ifgr.ifgr_len;
3061 		break;
3062 	case SIOCGIFMEDIA32:
3063 	case SIOCGIFXMEDIA32:
3064 		ifmr32->ifm_current = thunk.ifmr.ifm_current;
3065 		ifmr32->ifm_mask = thunk.ifmr.ifm_mask;
3066 		ifmr32->ifm_status = thunk.ifmr.ifm_status;
3067 		ifmr32->ifm_active = thunk.ifmr.ifm_active;
3068 		ifmr32->ifm_count = thunk.ifmr.ifm_count;
3069 		break;
3070 	}
3071 #endif
3072 	return (error);
3073 }
3074 
3075 /*
3076  * The code common to handling reference counted flags,
3077  * e.g., in ifpromisc() and if_allmulti().
3078  * The "pflag" argument can specify a permanent mode flag to check,
3079  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
3080  *
3081  * Only to be used on stack-owned flags, not driver-owned flags.
3082  */
3083 static int
3084 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
3085 {
3086 	struct ifreq ifr;
3087 	int error;
3088 	int oldflags, oldcount;
3089 
3090 	/* Sanity checks to catch programming errors */
3091 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
3092 	    ("%s: setting driver-owned flag %d", __func__, flag));
3093 
3094 	if (onswitch)
3095 		KASSERT(*refcount >= 0,
3096 		    ("%s: increment negative refcount %d for flag %d",
3097 		    __func__, *refcount, flag));
3098 	else
3099 		KASSERT(*refcount > 0,
3100 		    ("%s: decrement non-positive refcount %d for flag %d",
3101 		    __func__, *refcount, flag));
3102 
3103 	/* In case this mode is permanent, just touch refcount */
3104 	if (ifp->if_flags & pflag) {
3105 		*refcount += onswitch ? 1 : -1;
3106 		return (0);
3107 	}
3108 
3109 	/* Save ifnet parameters for if_ioctl() may fail */
3110 	oldcount = *refcount;
3111 	oldflags = ifp->if_flags;
3112 
3113 	/*
3114 	 * See if we aren't the only and touching refcount is enough.
3115 	 * Actually toggle interface flag if we are the first or last.
3116 	 */
3117 	if (onswitch) {
3118 		if ((*refcount)++)
3119 			return (0);
3120 		ifp->if_flags |= flag;
3121 	} else {
3122 		if (--(*refcount))
3123 			return (0);
3124 		ifp->if_flags &= ~flag;
3125 	}
3126 
3127 	/* Call down the driver since we've changed interface flags */
3128 	if (ifp->if_ioctl == NULL) {
3129 		error = EOPNOTSUPP;
3130 		goto recover;
3131 	}
3132 	ifr.ifr_flags = ifp->if_flags & 0xffff;
3133 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
3134 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3135 	if (error)
3136 		goto recover;
3137 	/* Notify userland that interface flags have changed */
3138 	rt_ifmsg(ifp);
3139 	return (0);
3140 
3141 recover:
3142 	/* Recover after driver error */
3143 	*refcount = oldcount;
3144 	ifp->if_flags = oldflags;
3145 	return (error);
3146 }
3147 
3148 /*
3149  * Set/clear promiscuous mode on interface ifp based on the truth value
3150  * of pswitch.  The calls are reference counted so that only the first
3151  * "on" request actually has an effect, as does the final "off" request.
3152  * Results are undefined if the "off" and "on" requests are not matched.
3153  */
3154 int
3155 ifpromisc(struct ifnet *ifp, int pswitch)
3156 {
3157 	int error;
3158 	int oldflags = ifp->if_flags;
3159 
3160 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
3161 			   &ifp->if_pcount, pswitch);
3162 	/* If promiscuous mode status has changed, log a message */
3163 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC) &&
3164             log_promisc_mode_change)
3165 		if_printf(ifp, "promiscuous mode %s\n",
3166 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
3167 	return (error);
3168 }
3169 
3170 /*
3171  * Return interface configuration
3172  * of system.  List may be used
3173  * in later ioctl's (above) to get
3174  * other information.
3175  */
3176 /*ARGSUSED*/
3177 static int
3178 ifconf(u_long cmd, caddr_t data)
3179 {
3180 	struct ifconf *ifc = (struct ifconf *)data;
3181 	struct ifnet *ifp;
3182 	struct ifaddr *ifa;
3183 	struct ifreq ifr;
3184 	struct sbuf *sb;
3185 	int error, full = 0, valid_len, max_len;
3186 
3187 	/* Limit initial buffer size to maxphys to avoid DoS from userspace. */
3188 	max_len = maxphys - 1;
3189 
3190 	/* Prevent hostile input from being able to crash the system */
3191 	if (ifc->ifc_len <= 0)
3192 		return (EINVAL);
3193 
3194 again:
3195 	if (ifc->ifc_len <= max_len) {
3196 		max_len = ifc->ifc_len;
3197 		full = 1;
3198 	}
3199 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
3200 	max_len = 0;
3201 	valid_len = 0;
3202 
3203 	IFNET_RLOCK();
3204 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
3205 		struct epoch_tracker et;
3206 		int addrs;
3207 
3208 		/*
3209 		 * Zero the ifr to make sure we don't disclose the contents
3210 		 * of the stack.
3211 		 */
3212 		memset(&ifr, 0, sizeof(ifr));
3213 
3214 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
3215 		    >= sizeof(ifr.ifr_name)) {
3216 			sbuf_delete(sb);
3217 			IFNET_RUNLOCK();
3218 			return (ENAMETOOLONG);
3219 		}
3220 
3221 		addrs = 0;
3222 		NET_EPOCH_ENTER(et);
3223 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3224 			struct sockaddr *sa = ifa->ifa_addr;
3225 
3226 			if (prison_if(curthread->td_ucred, sa) != 0)
3227 				continue;
3228 			addrs++;
3229 			if (sa->sa_len <= sizeof(*sa)) {
3230 				if (sa->sa_len < sizeof(*sa)) {
3231 					memset(&ifr.ifr_ifru.ifru_addr, 0,
3232 					    sizeof(ifr.ifr_ifru.ifru_addr));
3233 					memcpy(&ifr.ifr_ifru.ifru_addr, sa,
3234 					    sa->sa_len);
3235 				} else
3236 					ifr.ifr_ifru.ifru_addr = *sa;
3237 				sbuf_bcat(sb, &ifr, sizeof(ifr));
3238 				max_len += sizeof(ifr);
3239 			} else {
3240 				sbuf_bcat(sb, &ifr,
3241 				    offsetof(struct ifreq, ifr_addr));
3242 				max_len += offsetof(struct ifreq, ifr_addr);
3243 				sbuf_bcat(sb, sa, sa->sa_len);
3244 				max_len += sa->sa_len;
3245 			}
3246 
3247 			if (sbuf_error(sb) == 0)
3248 				valid_len = sbuf_len(sb);
3249 		}
3250 		NET_EPOCH_EXIT(et);
3251 		if (addrs == 0) {
3252 			sbuf_bcat(sb, &ifr, sizeof(ifr));
3253 			max_len += sizeof(ifr);
3254 
3255 			if (sbuf_error(sb) == 0)
3256 				valid_len = sbuf_len(sb);
3257 		}
3258 	}
3259 	IFNET_RUNLOCK();
3260 
3261 	/*
3262 	 * If we didn't allocate enough space (uncommon), try again.  If
3263 	 * we have already allocated as much space as we are allowed,
3264 	 * return what we've got.
3265 	 */
3266 	if (valid_len != max_len && !full) {
3267 		sbuf_delete(sb);
3268 		goto again;
3269 	}
3270 
3271 	ifc->ifc_len = valid_len;
3272 	sbuf_finish(sb);
3273 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
3274 	sbuf_delete(sb);
3275 	return (error);
3276 }
3277 
3278 /*
3279  * Just like ifpromisc(), but for all-multicast-reception mode.
3280  */
3281 int
3282 if_allmulti(struct ifnet *ifp, int onswitch)
3283 {
3284 
3285 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
3286 }
3287 
3288 struct ifmultiaddr *
3289 if_findmulti(struct ifnet *ifp, const struct sockaddr *sa)
3290 {
3291 	struct ifmultiaddr *ifma;
3292 
3293 	IF_ADDR_LOCK_ASSERT(ifp);
3294 
3295 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3296 		if (sa->sa_family == AF_LINK) {
3297 			if (sa_dl_equal(ifma->ifma_addr, sa))
3298 				break;
3299 		} else {
3300 			if (sa_equal(ifma->ifma_addr, sa))
3301 				break;
3302 		}
3303 	}
3304 
3305 	return ifma;
3306 }
3307 
3308 /*
3309  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
3310  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
3311  * the ifnet multicast address list here, so the caller must do that and
3312  * other setup work (such as notifying the device driver).  The reference
3313  * count is initialized to 1.
3314  */
3315 static struct ifmultiaddr *
3316 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
3317     int mflags)
3318 {
3319 	struct ifmultiaddr *ifma;
3320 	struct sockaddr *dupsa;
3321 
3322 	ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
3323 	    M_ZERO);
3324 	if (ifma == NULL)
3325 		return (NULL);
3326 
3327 	dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
3328 	if (dupsa == NULL) {
3329 		free(ifma, M_IFMADDR);
3330 		return (NULL);
3331 	}
3332 	bcopy(sa, dupsa, sa->sa_len);
3333 	ifma->ifma_addr = dupsa;
3334 
3335 	ifma->ifma_ifp = ifp;
3336 	ifma->ifma_refcount = 1;
3337 	ifma->ifma_protospec = NULL;
3338 
3339 	if (llsa == NULL) {
3340 		ifma->ifma_lladdr = NULL;
3341 		return (ifma);
3342 	}
3343 
3344 	dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
3345 	if (dupsa == NULL) {
3346 		free(ifma->ifma_addr, M_IFMADDR);
3347 		free(ifma, M_IFMADDR);
3348 		return (NULL);
3349 	}
3350 	bcopy(llsa, dupsa, llsa->sa_len);
3351 	ifma->ifma_lladdr = dupsa;
3352 
3353 	return (ifma);
3354 }
3355 
3356 /*
3357  * if_freemulti: free ifmultiaddr structure and possibly attached related
3358  * addresses.  The caller is responsible for implementing reference
3359  * counting, notifying the driver, handling routing messages, and releasing
3360  * any dependent link layer state.
3361  */
3362 #ifdef MCAST_VERBOSE
3363 extern void kdb_backtrace(void);
3364 #endif
3365 static void
3366 if_freemulti_internal(struct ifmultiaddr *ifma)
3367 {
3368 
3369 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
3370 	    ifma->ifma_refcount));
3371 
3372 	if (ifma->ifma_lladdr != NULL)
3373 		free(ifma->ifma_lladdr, M_IFMADDR);
3374 #ifdef MCAST_VERBOSE
3375 	kdb_backtrace();
3376 	printf("%s freeing ifma: %p\n", __func__, ifma);
3377 #endif
3378 	free(ifma->ifma_addr, M_IFMADDR);
3379 	free(ifma, M_IFMADDR);
3380 }
3381 
3382 static void
3383 if_destroymulti(epoch_context_t ctx)
3384 {
3385 	struct ifmultiaddr *ifma;
3386 
3387 	ifma = __containerof(ctx, struct ifmultiaddr, ifma_epoch_ctx);
3388 	if_freemulti_internal(ifma);
3389 }
3390 
3391 void
3392 if_freemulti(struct ifmultiaddr *ifma)
3393 {
3394 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti_epoch: refcount %d",
3395 	    ifma->ifma_refcount));
3396 
3397 	NET_EPOCH_CALL(if_destroymulti, &ifma->ifma_epoch_ctx);
3398 }
3399 
3400 /*
3401  * Register an additional multicast address with a network interface.
3402  *
3403  * - If the address is already present, bump the reference count on the
3404  *   address and return.
3405  * - If the address is not link-layer, look up a link layer address.
3406  * - Allocate address structures for one or both addresses, and attach to the
3407  *   multicast address list on the interface.  If automatically adding a link
3408  *   layer address, the protocol address will own a reference to the link
3409  *   layer address, to be freed when it is freed.
3410  * - Notify the network device driver of an addition to the multicast address
3411  *   list.
3412  *
3413  * 'sa' points to caller-owned memory with the desired multicast address.
3414  *
3415  * 'retifma' will be used to return a pointer to the resulting multicast
3416  * address reference, if desired.
3417  */
3418 int
3419 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
3420     struct ifmultiaddr **retifma)
3421 {
3422 	struct ifmultiaddr *ifma, *ll_ifma;
3423 	struct sockaddr *llsa;
3424 	struct sockaddr_dl sdl;
3425 	int error;
3426 
3427 #ifdef INET
3428 	IN_MULTI_LIST_UNLOCK_ASSERT();
3429 #endif
3430 #ifdef INET6
3431 	IN6_MULTI_LIST_UNLOCK_ASSERT();
3432 #endif
3433 	/*
3434 	 * If the address is already present, return a new reference to it;
3435 	 * otherwise, allocate storage and set up a new address.
3436 	 */
3437 	IF_ADDR_WLOCK(ifp);
3438 	ifma = if_findmulti(ifp, sa);
3439 	if (ifma != NULL) {
3440 		ifma->ifma_refcount++;
3441 		if (retifma != NULL)
3442 			*retifma = ifma;
3443 		IF_ADDR_WUNLOCK(ifp);
3444 		return (0);
3445 	}
3446 
3447 	/*
3448 	 * The address isn't already present; resolve the protocol address
3449 	 * into a link layer address, and then look that up, bump its
3450 	 * refcount or allocate an ifma for that also.
3451 	 * Most link layer resolving functions returns address data which
3452 	 * fits inside default sockaddr_dl structure. However callback
3453 	 * can allocate another sockaddr structure, in that case we need to
3454 	 * free it later.
3455 	 */
3456 	llsa = NULL;
3457 	ll_ifma = NULL;
3458 	if (ifp->if_resolvemulti != NULL) {
3459 		/* Provide called function with buffer size information */
3460 		sdl.sdl_len = sizeof(sdl);
3461 		llsa = (struct sockaddr *)&sdl;
3462 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
3463 		if (error)
3464 			goto unlock_out;
3465 	}
3466 
3467 	/*
3468 	 * Allocate the new address.  Don't hook it up yet, as we may also
3469 	 * need to allocate a link layer multicast address.
3470 	 */
3471 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
3472 	if (ifma == NULL) {
3473 		error = ENOMEM;
3474 		goto free_llsa_out;
3475 	}
3476 
3477 	/*
3478 	 * If a link layer address is found, we'll need to see if it's
3479 	 * already present in the address list, or allocate is as well.
3480 	 * When this block finishes, the link layer address will be on the
3481 	 * list.
3482 	 */
3483 	if (llsa != NULL) {
3484 		ll_ifma = if_findmulti(ifp, llsa);
3485 		if (ll_ifma == NULL) {
3486 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
3487 			if (ll_ifma == NULL) {
3488 				--ifma->ifma_refcount;
3489 				if_freemulti(ifma);
3490 				error = ENOMEM;
3491 				goto free_llsa_out;
3492 			}
3493 			ll_ifma->ifma_flags |= IFMA_F_ENQUEUED;
3494 			CK_STAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
3495 			    ifma_link);
3496 		} else
3497 			ll_ifma->ifma_refcount++;
3498 		ifma->ifma_llifma = ll_ifma;
3499 	}
3500 
3501 	/*
3502 	 * We now have a new multicast address, ifma, and possibly a new or
3503 	 * referenced link layer address.  Add the primary address to the
3504 	 * ifnet address list.
3505 	 */
3506 	ifma->ifma_flags |= IFMA_F_ENQUEUED;
3507 	CK_STAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3508 
3509 	if (retifma != NULL)
3510 		*retifma = ifma;
3511 
3512 	/*
3513 	 * Must generate the message while holding the lock so that 'ifma'
3514 	 * pointer is still valid.
3515 	 */
3516 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3517 	IF_ADDR_WUNLOCK(ifp);
3518 
3519 	/*
3520 	 * We are certain we have added something, so call down to the
3521 	 * interface to let them know about it.
3522 	 */
3523 	if (ifp->if_ioctl != NULL) {
3524 		if (THREAD_CAN_SLEEP())
3525 			(void )(*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3526 		else
3527 			taskqueue_enqueue(taskqueue_swi, &ifp->if_addmultitask);
3528 	}
3529 
3530 	if ((llsa != NULL) && (llsa != (struct sockaddr *)&sdl))
3531 		link_free_sdl(llsa);
3532 
3533 	return (0);
3534 
3535 free_llsa_out:
3536 	if ((llsa != NULL) && (llsa != (struct sockaddr *)&sdl))
3537 		link_free_sdl(llsa);
3538 
3539 unlock_out:
3540 	IF_ADDR_WUNLOCK(ifp);
3541 	return (error);
3542 }
3543 
3544 static void
3545 if_siocaddmulti(void *arg, int pending)
3546 {
3547 	struct ifnet *ifp;
3548 
3549 	ifp = arg;
3550 #ifdef DIAGNOSTIC
3551 	if (pending > 1)
3552 		if_printf(ifp, "%d SIOCADDMULTI coalesced\n", pending);
3553 #endif
3554 	CURVNET_SET(ifp->if_vnet);
3555 	(void )(*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3556 	CURVNET_RESTORE();
3557 }
3558 
3559 /*
3560  * Delete a multicast group membership by network-layer group address.
3561  *
3562  * Returns ENOENT if the entry could not be found. If ifp no longer
3563  * exists, results are undefined. This entry point should only be used
3564  * from subsystems which do appropriate locking to hold ifp for the
3565  * duration of the call.
3566  * Network-layer protocol domains must use if_delmulti_ifma().
3567  */
3568 int
3569 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3570 {
3571 	struct ifmultiaddr *ifma;
3572 	int lastref;
3573 
3574 	KASSERT(ifp, ("%s: NULL ifp", __func__));
3575 
3576 	IF_ADDR_WLOCK(ifp);
3577 	lastref = 0;
3578 	ifma = if_findmulti(ifp, sa);
3579 	if (ifma != NULL)
3580 		lastref = if_delmulti_locked(ifp, ifma, 0);
3581 	IF_ADDR_WUNLOCK(ifp);
3582 
3583 	if (ifma == NULL)
3584 		return (ENOENT);
3585 
3586 	if (lastref && ifp->if_ioctl != NULL) {
3587 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3588 	}
3589 
3590 	return (0);
3591 }
3592 
3593 /*
3594  * Delete all multicast group membership for an interface.
3595  * Should be used to quickly flush all multicast filters.
3596  */
3597 void
3598 if_delallmulti(struct ifnet *ifp)
3599 {
3600 	struct ifmultiaddr *ifma;
3601 	struct ifmultiaddr *next;
3602 
3603 	IF_ADDR_WLOCK(ifp);
3604 	CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3605 		if_delmulti_locked(ifp, ifma, 0);
3606 	IF_ADDR_WUNLOCK(ifp);
3607 }
3608 
3609 void
3610 if_delmulti_ifma(struct ifmultiaddr *ifma)
3611 {
3612 	if_delmulti_ifma_flags(ifma, 0);
3613 }
3614 
3615 /*
3616  * Delete a multicast group membership by group membership pointer.
3617  * Network-layer protocol domains must use this routine.
3618  *
3619  * It is safe to call this routine if the ifp disappeared.
3620  */
3621 void
3622 if_delmulti_ifma_flags(struct ifmultiaddr *ifma, int flags)
3623 {
3624 	struct ifnet *ifp;
3625 	int lastref;
3626 	MCDPRINTF("%s freeing ifma: %p\n", __func__, ifma);
3627 #ifdef INET
3628 	IN_MULTI_LIST_UNLOCK_ASSERT();
3629 #endif
3630 	ifp = ifma->ifma_ifp;
3631 #ifdef DIAGNOSTIC
3632 	if (ifp == NULL) {
3633 		printf("%s: ifma_ifp seems to be detached\n", __func__);
3634 	} else {
3635 		struct epoch_tracker et;
3636 		struct ifnet *oifp;
3637 
3638 		NET_EPOCH_ENTER(et);
3639 		CK_STAILQ_FOREACH(oifp, &V_ifnet, if_link)
3640 			if (ifp == oifp)
3641 				break;
3642 		NET_EPOCH_EXIT(et);
3643 		if (ifp != oifp)
3644 			ifp = NULL;
3645 	}
3646 #endif
3647 	/*
3648 	 * If and only if the ifnet instance exists: Acquire the address lock.
3649 	 */
3650 	if (ifp != NULL)
3651 		IF_ADDR_WLOCK(ifp);
3652 
3653 	lastref = if_delmulti_locked(ifp, ifma, flags);
3654 
3655 	if (ifp != NULL) {
3656 		/*
3657 		 * If and only if the ifnet instance exists:
3658 		 *  Release the address lock.
3659 		 *  If the group was left: update the hardware hash filter.
3660 		 */
3661 		IF_ADDR_WUNLOCK(ifp);
3662 		if (lastref && ifp->if_ioctl != NULL) {
3663 			(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3664 		}
3665 	}
3666 }
3667 
3668 /*
3669  * Perform deletion of network-layer and/or link-layer multicast address.
3670  *
3671  * Return 0 if the reference count was decremented.
3672  * Return 1 if the final reference was released, indicating that the
3673  * hardware hash filter should be reprogrammed.
3674  */
3675 static int
3676 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3677 {
3678 	struct ifmultiaddr *ll_ifma;
3679 
3680 	if (ifp != NULL && ifma->ifma_ifp != NULL) {
3681 		KASSERT(ifma->ifma_ifp == ifp,
3682 		    ("%s: inconsistent ifp %p", __func__, ifp));
3683 		IF_ADDR_WLOCK_ASSERT(ifp);
3684 	}
3685 
3686 	ifp = ifma->ifma_ifp;
3687 	MCDPRINTF("%s freeing %p from %s \n", __func__, ifma, ifp ? ifp->if_xname : "");
3688 
3689 	/*
3690 	 * If the ifnet is detaching, null out references to ifnet,
3691 	 * so that upper protocol layers will notice, and not attempt
3692 	 * to obtain locks for an ifnet which no longer exists. The
3693 	 * routing socket announcement must happen before the ifnet
3694 	 * instance is detached from the system.
3695 	 */
3696 	if (detaching) {
3697 #ifdef DIAGNOSTIC
3698 		printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3699 #endif
3700 		/*
3701 		 * ifp may already be nulled out if we are being reentered
3702 		 * to delete the ll_ifma.
3703 		 */
3704 		if (ifp != NULL) {
3705 			rt_newmaddrmsg(RTM_DELMADDR, ifma);
3706 			ifma->ifma_ifp = NULL;
3707 		}
3708 	}
3709 
3710 	if (--ifma->ifma_refcount > 0)
3711 		return 0;
3712 
3713 	if (ifp != NULL && detaching == 0 && (ifma->ifma_flags & IFMA_F_ENQUEUED)) {
3714 		CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
3715 		ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
3716 	}
3717 	/*
3718 	 * If this ifma is a network-layer ifma, a link-layer ifma may
3719 	 * have been associated with it. Release it first if so.
3720 	 */
3721 	ll_ifma = ifma->ifma_llifma;
3722 	if (ll_ifma != NULL) {
3723 		KASSERT(ifma->ifma_lladdr != NULL,
3724 		    ("%s: llifma w/o lladdr", __func__));
3725 		if (detaching)
3726 			ll_ifma->ifma_ifp = NULL;	/* XXX */
3727 		if (--ll_ifma->ifma_refcount == 0) {
3728 			if (ifp != NULL) {
3729 				if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
3730 					CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr,
3731 						ifma_link);
3732 					ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
3733 				}
3734 			}
3735 			if_freemulti(ll_ifma);
3736 		}
3737 	}
3738 #ifdef INVARIANTS
3739 	if (ifp) {
3740 		struct ifmultiaddr *ifmatmp;
3741 
3742 		CK_STAILQ_FOREACH(ifmatmp, &ifp->if_multiaddrs, ifma_link)
3743 			MPASS(ifma != ifmatmp);
3744 	}
3745 #endif
3746 	if_freemulti(ifma);
3747 	/*
3748 	 * The last reference to this instance of struct ifmultiaddr
3749 	 * was released; the hardware should be notified of this change.
3750 	 */
3751 	return 1;
3752 }
3753 
3754 /*
3755  * Set the link layer address on an interface.
3756  *
3757  * At this time we only support certain types of interfaces,
3758  * and we don't allow the length of the address to change.
3759  *
3760  * Set noinline to be dtrace-friendly
3761  */
3762 __noinline int
3763 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3764 {
3765 	struct sockaddr_dl *sdl;
3766 	struct ifaddr *ifa;
3767 	struct ifreq ifr;
3768 
3769 	ifa = ifp->if_addr;
3770 	if (ifa == NULL)
3771 		return (EINVAL);
3772 
3773 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3774 	if (sdl == NULL)
3775 		return (EINVAL);
3776 
3777 	if (len != sdl->sdl_alen)	/* don't allow length to change */
3778 		return (EINVAL);
3779 
3780 	switch (ifp->if_type) {
3781 	case IFT_ETHER:
3782 	case IFT_XETHER:
3783 	case IFT_L2VLAN:
3784 	case IFT_BRIDGE:
3785 	case IFT_IEEE8023ADLAG:
3786 		bcopy(lladdr, LLADDR(sdl), len);
3787 		break;
3788 	default:
3789 		return (ENODEV);
3790 	}
3791 
3792 	/*
3793 	 * If the interface is already up, we need
3794 	 * to re-init it in order to reprogram its
3795 	 * address filter.
3796 	 */
3797 	if ((ifp->if_flags & IFF_UP) != 0) {
3798 		if (ifp->if_ioctl) {
3799 			ifp->if_flags &= ~IFF_UP;
3800 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3801 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3802 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3803 			ifp->if_flags |= IFF_UP;
3804 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3805 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3806 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3807 		}
3808 	}
3809 	EVENTHANDLER_INVOKE(iflladdr_event, ifp);
3810 
3811 	return (0);
3812 }
3813 
3814 /*
3815  * Compat function for handling basic encapsulation requests.
3816  * Not converted stacks (FDDI, IB, ..) supports traditional
3817  * output model: ARP (and other similar L2 protocols) are handled
3818  * inside output routine, arpresolve/nd6_resolve() returns MAC
3819  * address instead of full prepend.
3820  *
3821  * This function creates calculated header==MAC for IPv4/IPv6 and
3822  * returns EAFNOSUPPORT (which is then handled in ARP code) for other
3823  * address families.
3824  */
3825 static int
3826 if_requestencap_default(struct ifnet *ifp, struct if_encap_req *req)
3827 {
3828 
3829 	if (req->rtype != IFENCAP_LL)
3830 		return (EOPNOTSUPP);
3831 
3832 	if (req->bufsize < req->lladdr_len)
3833 		return (ENOMEM);
3834 
3835 	switch (req->family) {
3836 	case AF_INET:
3837 	case AF_INET6:
3838 		break;
3839 	default:
3840 		return (EAFNOSUPPORT);
3841 	}
3842 
3843 	/* Copy lladdr to storage as is */
3844 	memmove(req->buf, req->lladdr, req->lladdr_len);
3845 	req->bufsize = req->lladdr_len;
3846 	req->lladdr_off = 0;
3847 
3848 	return (0);
3849 }
3850 
3851 /*
3852  * Tunnel interfaces can nest, also they may cause infinite recursion
3853  * calls when misconfigured. We'll prevent this by detecting loops.
3854  * High nesting level may cause stack exhaustion. We'll prevent this
3855  * by introducing upper limit.
3856  *
3857  * Return 0, if tunnel nesting count is equal or less than limit.
3858  */
3859 int
3860 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, uint32_t cookie,
3861     int limit)
3862 {
3863 	struct m_tag *mtag;
3864 	int count;
3865 
3866 	count = 1;
3867 	mtag = NULL;
3868 	while ((mtag = m_tag_locate(m, cookie, 0, mtag)) != NULL) {
3869 		if (*(struct ifnet **)(mtag + 1) == ifp) {
3870 			log(LOG_NOTICE, "%s: loop detected\n", if_name(ifp));
3871 			return (EIO);
3872 		}
3873 		count++;
3874 	}
3875 	if (count > limit) {
3876 		log(LOG_NOTICE,
3877 		    "%s: if_output recursively called too many times(%d)\n",
3878 		    if_name(ifp), count);
3879 		return (EIO);
3880 	}
3881 	mtag = m_tag_alloc(cookie, 0, sizeof(struct ifnet *), M_NOWAIT);
3882 	if (mtag == NULL)
3883 		return (ENOMEM);
3884 	*(struct ifnet **)(mtag + 1) = ifp;
3885 	m_tag_prepend(m, mtag);
3886 	return (0);
3887 }
3888 
3889 /*
3890  * Get the link layer address that was read from the hardware at attach.
3891  *
3892  * This is only set by Ethernet NICs (IFT_ETHER), but laggX interfaces re-type
3893  * their component interfaces as IFT_IEEE8023ADLAG.
3894  */
3895 int
3896 if_gethwaddr(struct ifnet *ifp, struct ifreq *ifr)
3897 {
3898 
3899 	if (ifp->if_hw_addr == NULL)
3900 		return (ENODEV);
3901 
3902 	switch (ifp->if_type) {
3903 	case IFT_ETHER:
3904 	case IFT_IEEE8023ADLAG:
3905 		bcopy(ifp->if_hw_addr, ifr->ifr_addr.sa_data, ifp->if_addrlen);
3906 		return (0);
3907 	default:
3908 		return (ENODEV);
3909 	}
3910 }
3911 
3912 /*
3913  * The name argument must be a pointer to storage which will last as
3914  * long as the interface does.  For physical devices, the result of
3915  * device_get_name(dev) is a good choice and for pseudo-devices a
3916  * static string works well.
3917  */
3918 void
3919 if_initname(struct ifnet *ifp, const char *name, int unit)
3920 {
3921 	ifp->if_dname = name;
3922 	ifp->if_dunit = unit;
3923 	if (unit != IF_DUNIT_NONE)
3924 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3925 	else
3926 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
3927 }
3928 
3929 static int
3930 if_vlog(struct ifnet *ifp, int pri, const char *fmt, va_list ap)
3931 {
3932 	char if_fmt[256];
3933 
3934 	snprintf(if_fmt, sizeof(if_fmt), "%s: %s", ifp->if_xname, fmt);
3935 	vlog(pri, if_fmt, ap);
3936 	return (0);
3937 }
3938 
3939 
3940 int
3941 if_printf(struct ifnet *ifp, const char *fmt, ...)
3942 {
3943 	va_list ap;
3944 
3945 	va_start(ap, fmt);
3946 	if_vlog(ifp, LOG_INFO, fmt, ap);
3947 	va_end(ap);
3948 	return (0);
3949 }
3950 
3951 int
3952 if_log(struct ifnet *ifp, int pri, const char *fmt, ...)
3953 {
3954 	va_list ap;
3955 
3956 	va_start(ap, fmt);
3957 	if_vlog(ifp, pri, fmt, ap);
3958 	va_end(ap);
3959 	return (0);
3960 }
3961 
3962 void
3963 if_start(struct ifnet *ifp)
3964 {
3965 
3966 	(*(ifp)->if_start)(ifp);
3967 }
3968 
3969 /*
3970  * Backwards compatibility interface for drivers
3971  * that have not implemented it
3972  */
3973 static int
3974 if_transmit(struct ifnet *ifp, struct mbuf *m)
3975 {
3976 	int error;
3977 
3978 	IFQ_HANDOFF(ifp, m, error);
3979 	return (error);
3980 }
3981 
3982 static void
3983 if_input_default(struct ifnet *ifp __unused, struct mbuf *m)
3984 {
3985 
3986 	m_freem(m);
3987 }
3988 
3989 int
3990 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3991 {
3992 	int active = 0;
3993 
3994 	IF_LOCK(ifq);
3995 	if (_IF_QFULL(ifq)) {
3996 		IF_UNLOCK(ifq);
3997 		if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
3998 		m_freem(m);
3999 		return (0);
4000 	}
4001 	if (ifp != NULL) {
4002 		if_inc_counter(ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len + adjust);
4003 		if (m->m_flags & (M_BCAST|M_MCAST))
4004 			if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
4005 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
4006 	}
4007 	_IF_ENQUEUE(ifq, m);
4008 	IF_UNLOCK(ifq);
4009 	if (ifp != NULL && !active)
4010 		(*(ifp)->if_start)(ifp);
4011 	return (1);
4012 }
4013 
4014 void
4015 if_register_com_alloc(u_char type,
4016     if_com_alloc_t *a, if_com_free_t *f)
4017 {
4018 
4019 	KASSERT(if_com_alloc[type] == NULL,
4020 	    ("if_register_com_alloc: %d already registered", type));
4021 	KASSERT(if_com_free[type] == NULL,
4022 	    ("if_register_com_alloc: %d free already registered", type));
4023 
4024 	if_com_alloc[type] = a;
4025 	if_com_free[type] = f;
4026 }
4027 
4028 void
4029 if_deregister_com_alloc(u_char type)
4030 {
4031 
4032 	KASSERT(if_com_alloc[type] != NULL,
4033 	    ("if_deregister_com_alloc: %d not registered", type));
4034 	KASSERT(if_com_free[type] != NULL,
4035 	    ("if_deregister_com_alloc: %d free not registered", type));
4036 
4037 	/*
4038 	 * Ensure all pending EPOCH(9) callbacks have been executed. This
4039 	 * fixes issues about late invocation of if_destroy(), which leads
4040 	 * to memory leak from if_com_alloc[type] allocated if_l2com.
4041 	 */
4042 	epoch_drain_callbacks(net_epoch_preempt);
4043 
4044 	if_com_alloc[type] = NULL;
4045 	if_com_free[type] = NULL;
4046 }
4047 
4048 /* API for driver access to network stack owned ifnet.*/
4049 uint64_t
4050 if_setbaudrate(struct ifnet *ifp, uint64_t baudrate)
4051 {
4052 	uint64_t oldbrate;
4053 
4054 	oldbrate = ifp->if_baudrate;
4055 	ifp->if_baudrate = baudrate;
4056 	return (oldbrate);
4057 }
4058 
4059 uint64_t
4060 if_getbaudrate(if_t ifp)
4061 {
4062 
4063 	return (((struct ifnet *)ifp)->if_baudrate);
4064 }
4065 
4066 int
4067 if_setcapabilities(if_t ifp, int capabilities)
4068 {
4069 	((struct ifnet *)ifp)->if_capabilities = capabilities;
4070 	return (0);
4071 }
4072 
4073 int
4074 if_setcapabilitiesbit(if_t ifp, int setbit, int clearbit)
4075 {
4076 	((struct ifnet *)ifp)->if_capabilities |= setbit;
4077 	((struct ifnet *)ifp)->if_capabilities &= ~clearbit;
4078 
4079 	return (0);
4080 }
4081 
4082 int
4083 if_getcapabilities(if_t ifp)
4084 {
4085 	return ((struct ifnet *)ifp)->if_capabilities;
4086 }
4087 
4088 int
4089 if_setcapenable(if_t ifp, int capabilities)
4090 {
4091 	((struct ifnet *)ifp)->if_capenable = capabilities;
4092 	return (0);
4093 }
4094 
4095 int
4096 if_setcapenablebit(if_t ifp, int setcap, int clearcap)
4097 {
4098 	if(setcap)
4099 		((struct ifnet *)ifp)->if_capenable |= setcap;
4100 	if(clearcap)
4101 		((struct ifnet *)ifp)->if_capenable &= ~clearcap;
4102 
4103 	return (0);
4104 }
4105 
4106 const char *
4107 if_getdname(if_t ifp)
4108 {
4109 	return ((struct ifnet *)ifp)->if_dname;
4110 }
4111 
4112 int
4113 if_togglecapenable(if_t ifp, int togglecap)
4114 {
4115 	((struct ifnet *)ifp)->if_capenable ^= togglecap;
4116 	return (0);
4117 }
4118 
4119 int
4120 if_getcapenable(if_t ifp)
4121 {
4122 	return ((struct ifnet *)ifp)->if_capenable;
4123 }
4124 
4125 /*
4126  * This is largely undesirable because it ties ifnet to a device, but does
4127  * provide flexiblity for an embedded product vendor. Should be used with
4128  * the understanding that it violates the interface boundaries, and should be
4129  * a last resort only.
4130  */
4131 int
4132 if_setdev(if_t ifp, void *dev)
4133 {
4134 	return (0);
4135 }
4136 
4137 int
4138 if_setdrvflagbits(if_t ifp, int set_flags, int clear_flags)
4139 {
4140 	((struct ifnet *)ifp)->if_drv_flags |= set_flags;
4141 	((struct ifnet *)ifp)->if_drv_flags &= ~clear_flags;
4142 
4143 	return (0);
4144 }
4145 
4146 int
4147 if_getdrvflags(if_t ifp)
4148 {
4149 	return ((struct ifnet *)ifp)->if_drv_flags;
4150 }
4151 
4152 int
4153 if_setdrvflags(if_t ifp, int flags)
4154 {
4155 	((struct ifnet *)ifp)->if_drv_flags = flags;
4156 	return (0);
4157 }
4158 
4159 int
4160 if_setflags(if_t ifp, int flags)
4161 {
4162 
4163 	ifp->if_flags = flags;
4164 	return (0);
4165 }
4166 
4167 int
4168 if_setflagbits(if_t ifp, int set, int clear)
4169 {
4170 	((struct ifnet *)ifp)->if_flags |= set;
4171 	((struct ifnet *)ifp)->if_flags &= ~clear;
4172 
4173 	return (0);
4174 }
4175 
4176 int
4177 if_getflags(if_t ifp)
4178 {
4179 	return ((struct ifnet *)ifp)->if_flags;
4180 }
4181 
4182 int
4183 if_clearhwassist(if_t ifp)
4184 {
4185 	((struct ifnet *)ifp)->if_hwassist = 0;
4186 	return (0);
4187 }
4188 
4189 int
4190 if_sethwassistbits(if_t ifp, int toset, int toclear)
4191 {
4192 	((struct ifnet *)ifp)->if_hwassist |= toset;
4193 	((struct ifnet *)ifp)->if_hwassist &= ~toclear;
4194 
4195 	return (0);
4196 }
4197 
4198 int
4199 if_sethwassist(if_t ifp, int hwassist_bit)
4200 {
4201 	((struct ifnet *)ifp)->if_hwassist = hwassist_bit;
4202 	return (0);
4203 }
4204 
4205 int
4206 if_gethwassist(if_t ifp)
4207 {
4208 	return ((struct ifnet *)ifp)->if_hwassist;
4209 }
4210 
4211 int
4212 if_setmtu(if_t ifp, int mtu)
4213 {
4214 	((struct ifnet *)ifp)->if_mtu = mtu;
4215 	return (0);
4216 }
4217 
4218 int
4219 if_getmtu(if_t ifp)
4220 {
4221 	return ((struct ifnet *)ifp)->if_mtu;
4222 }
4223 
4224 int
4225 if_getmtu_family(if_t ifp, int family)
4226 {
4227 	struct domain *dp;
4228 
4229 	for (dp = domains; dp; dp = dp->dom_next) {
4230 		if (dp->dom_family == family && dp->dom_ifmtu != NULL)
4231 			return (dp->dom_ifmtu((struct ifnet *)ifp));
4232 	}
4233 
4234 	return (((struct ifnet *)ifp)->if_mtu);
4235 }
4236 
4237 /*
4238  * Methods for drivers to access interface unicast and multicast
4239  * link level addresses.  Driver shall not know 'struct ifaddr' neither
4240  * 'struct ifmultiaddr'.
4241  */
4242 u_int
4243 if_lladdr_count(if_t ifp)
4244 {
4245 	struct epoch_tracker et;
4246 	struct ifaddr *ifa;
4247 	u_int count;
4248 
4249 	count = 0;
4250 	NET_EPOCH_ENTER(et);
4251 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
4252 		if (ifa->ifa_addr->sa_family == AF_LINK)
4253 			count++;
4254 	NET_EPOCH_EXIT(et);
4255 
4256 	return (count);
4257 }
4258 
4259 u_int
4260 if_foreach_lladdr(if_t ifp, iflladdr_cb_t cb, void *cb_arg)
4261 {
4262 	struct epoch_tracker et;
4263 	struct ifaddr *ifa;
4264 	u_int count;
4265 
4266 	MPASS(cb);
4267 
4268 	count = 0;
4269 	NET_EPOCH_ENTER(et);
4270 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
4271 		if (ifa->ifa_addr->sa_family != AF_LINK)
4272 			continue;
4273 		count += (*cb)(cb_arg, (struct sockaddr_dl *)ifa->ifa_addr,
4274 		    count);
4275 	}
4276 	NET_EPOCH_EXIT(et);
4277 
4278 	return (count);
4279 }
4280 
4281 u_int
4282 if_llmaddr_count(if_t ifp)
4283 {
4284 	struct epoch_tracker et;
4285 	struct ifmultiaddr *ifma;
4286 	int count;
4287 
4288 	count = 0;
4289 	NET_EPOCH_ENTER(et);
4290 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
4291 		if (ifma->ifma_addr->sa_family == AF_LINK)
4292 			count++;
4293 	NET_EPOCH_EXIT(et);
4294 
4295 	return (count);
4296 }
4297 
4298 u_int
4299 if_foreach_llmaddr(if_t ifp, iflladdr_cb_t cb, void *cb_arg)
4300 {
4301 	struct epoch_tracker et;
4302 	struct ifmultiaddr *ifma;
4303 	u_int count;
4304 
4305 	MPASS(cb);
4306 
4307 	count = 0;
4308 	NET_EPOCH_ENTER(et);
4309 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4310 		if (ifma->ifma_addr->sa_family != AF_LINK)
4311 			continue;
4312 		count += (*cb)(cb_arg, (struct sockaddr_dl *)ifma->ifma_addr,
4313 		    count);
4314 	}
4315 	NET_EPOCH_EXIT(et);
4316 
4317 	return (count);
4318 }
4319 
4320 int
4321 if_setsoftc(if_t ifp, void *softc)
4322 {
4323 	((struct ifnet *)ifp)->if_softc = softc;
4324 	return (0);
4325 }
4326 
4327 void *
4328 if_getsoftc(if_t ifp)
4329 {
4330 	return ((struct ifnet *)ifp)->if_softc;
4331 }
4332 
4333 void
4334 if_setrcvif(struct mbuf *m, if_t ifp)
4335 {
4336 
4337 	MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0);
4338 	m->m_pkthdr.rcvif = (struct ifnet *)ifp;
4339 }
4340 
4341 void
4342 if_setvtag(struct mbuf *m, uint16_t tag)
4343 {
4344 	m->m_pkthdr.ether_vtag = tag;
4345 }
4346 
4347 uint16_t
4348 if_getvtag(struct mbuf *m)
4349 {
4350 
4351 	return (m->m_pkthdr.ether_vtag);
4352 }
4353 
4354 int
4355 if_sendq_empty(if_t ifp)
4356 {
4357 	return IFQ_DRV_IS_EMPTY(&((struct ifnet *)ifp)->if_snd);
4358 }
4359 
4360 struct ifaddr *
4361 if_getifaddr(if_t ifp)
4362 {
4363 	return ((struct ifnet *)ifp)->if_addr;
4364 }
4365 
4366 int
4367 if_getamcount(if_t ifp)
4368 {
4369 	return ((struct ifnet *)ifp)->if_amcount;
4370 }
4371 
4372 int
4373 if_setsendqready(if_t ifp)
4374 {
4375 	IFQ_SET_READY(&((struct ifnet *)ifp)->if_snd);
4376 	return (0);
4377 }
4378 
4379 int
4380 if_setsendqlen(if_t ifp, int tx_desc_count)
4381 {
4382 	IFQ_SET_MAXLEN(&((struct ifnet *)ifp)->if_snd, tx_desc_count);
4383 	((struct ifnet *)ifp)->if_snd.ifq_drv_maxlen = tx_desc_count;
4384 
4385 	return (0);
4386 }
4387 
4388 int
4389 if_vlantrunkinuse(if_t ifp)
4390 {
4391 	return ((struct ifnet *)ifp)->if_vlantrunk != NULL?1:0;
4392 }
4393 
4394 int
4395 if_input(if_t ifp, struct mbuf* sendmp)
4396 {
4397 	(*((struct ifnet *)ifp)->if_input)((struct ifnet *)ifp, sendmp);
4398 	return (0);
4399 
4400 }
4401 
4402 struct mbuf *
4403 if_dequeue(if_t ifp)
4404 {
4405 	struct mbuf *m;
4406 	IFQ_DRV_DEQUEUE(&((struct ifnet *)ifp)->if_snd, m);
4407 
4408 	return (m);
4409 }
4410 
4411 int
4412 if_sendq_prepend(if_t ifp, struct mbuf *m)
4413 {
4414 	IFQ_DRV_PREPEND(&((struct ifnet *)ifp)->if_snd, m);
4415 	return (0);
4416 }
4417 
4418 int
4419 if_setifheaderlen(if_t ifp, int len)
4420 {
4421 	((struct ifnet *)ifp)->if_hdrlen = len;
4422 	return (0);
4423 }
4424 
4425 caddr_t
4426 if_getlladdr(if_t ifp)
4427 {
4428 	return (IF_LLADDR((struct ifnet *)ifp));
4429 }
4430 
4431 void *
4432 if_gethandle(u_char type)
4433 {
4434 	return (if_alloc(type));
4435 }
4436 
4437 void
4438 if_bpfmtap(if_t ifh, struct mbuf *m)
4439 {
4440 	struct ifnet *ifp = (struct ifnet *)ifh;
4441 
4442 	BPF_MTAP(ifp, m);
4443 }
4444 
4445 void
4446 if_etherbpfmtap(if_t ifh, struct mbuf *m)
4447 {
4448 	struct ifnet *ifp = (struct ifnet *)ifh;
4449 
4450 	ETHER_BPF_MTAP(ifp, m);
4451 }
4452 
4453 void
4454 if_vlancap(if_t ifh)
4455 {
4456 	struct ifnet *ifp = (struct ifnet *)ifh;
4457 	VLAN_CAPABILITIES(ifp);
4458 }
4459 
4460 int
4461 if_sethwtsomax(if_t ifp, u_int if_hw_tsomax)
4462 {
4463 
4464 	((struct ifnet *)ifp)->if_hw_tsomax = if_hw_tsomax;
4465         return (0);
4466 }
4467 
4468 int
4469 if_sethwtsomaxsegcount(if_t ifp, u_int if_hw_tsomaxsegcount)
4470 {
4471 
4472 	((struct ifnet *)ifp)->if_hw_tsomaxsegcount = if_hw_tsomaxsegcount;
4473         return (0);
4474 }
4475 
4476 int
4477 if_sethwtsomaxsegsize(if_t ifp, u_int if_hw_tsomaxsegsize)
4478 {
4479 
4480 	((struct ifnet *)ifp)->if_hw_tsomaxsegsize = if_hw_tsomaxsegsize;
4481         return (0);
4482 }
4483 
4484 u_int
4485 if_gethwtsomax(if_t ifp)
4486 {
4487 
4488 	return (((struct ifnet *)ifp)->if_hw_tsomax);
4489 }
4490 
4491 u_int
4492 if_gethwtsomaxsegcount(if_t ifp)
4493 {
4494 
4495 	return (((struct ifnet *)ifp)->if_hw_tsomaxsegcount);
4496 }
4497 
4498 u_int
4499 if_gethwtsomaxsegsize(if_t ifp)
4500 {
4501 
4502 	return (((struct ifnet *)ifp)->if_hw_tsomaxsegsize);
4503 }
4504 
4505 void
4506 if_setinitfn(if_t ifp, void (*init_fn)(void *))
4507 {
4508 	((struct ifnet *)ifp)->if_init = init_fn;
4509 }
4510 
4511 void
4512 if_setioctlfn(if_t ifp, int (*ioctl_fn)(if_t, u_long, caddr_t))
4513 {
4514 	((struct ifnet *)ifp)->if_ioctl = (void *)ioctl_fn;
4515 }
4516 
4517 void
4518 if_setstartfn(if_t ifp, void (*start_fn)(if_t))
4519 {
4520 	((struct ifnet *)ifp)->if_start = (void *)start_fn;
4521 }
4522 
4523 void
4524 if_settransmitfn(if_t ifp, if_transmit_fn_t start_fn)
4525 {
4526 	((struct ifnet *)ifp)->if_transmit = start_fn;
4527 }
4528 
4529 void if_setqflushfn(if_t ifp, if_qflush_fn_t flush_fn)
4530 {
4531 	((struct ifnet *)ifp)->if_qflush = flush_fn;
4532 
4533 }
4534 
4535 void
4536 if_setgetcounterfn(if_t ifp, if_get_counter_t fn)
4537 {
4538 
4539 	ifp->if_get_counter = fn;
4540 }
4541 
4542 #ifdef DDB
4543 static void
4544 if_show_ifnet(struct ifnet *ifp)
4545 {
4546 
4547 	if (ifp == NULL)
4548 		return;
4549 	db_printf("%s:\n", ifp->if_xname);
4550 #define	IF_DB_PRINTF(f, e)	db_printf("   %s = " f "\n", #e, ifp->e);
4551 	IF_DB_PRINTF("%s", if_dname);
4552 	IF_DB_PRINTF("%d", if_dunit);
4553 	IF_DB_PRINTF("%s", if_description);
4554 	IF_DB_PRINTF("%u", if_index);
4555 	IF_DB_PRINTF("%u", if_refcount);
4556 	IF_DB_PRINTF("%d", if_index_reserved);
4557 	IF_DB_PRINTF("%p", if_softc);
4558 	IF_DB_PRINTF("%p", if_l2com);
4559 	IF_DB_PRINTF("%p", if_llsoftc);
4560 	IF_DB_PRINTF("%d", if_amcount);
4561 	IF_DB_PRINTF("%p", if_addr);
4562 	IF_DB_PRINTF("%p", if_broadcastaddr);
4563 	IF_DB_PRINTF("%p", if_afdata);
4564 	IF_DB_PRINTF("%d", if_afdata_initialized);
4565 	IF_DB_PRINTF("%u", if_fib);
4566 	IF_DB_PRINTF("%p", if_vnet);
4567 	IF_DB_PRINTF("%p", if_home_vnet);
4568 	IF_DB_PRINTF("%p", if_vlantrunk);
4569 	IF_DB_PRINTF("%p", if_bpf);
4570 	IF_DB_PRINTF("%u", if_pcount);
4571 	IF_DB_PRINTF("%p", if_bridge);
4572 	IF_DB_PRINTF("%p", if_lagg);
4573 	IF_DB_PRINTF("%p", if_pf_kif);
4574 	IF_DB_PRINTF("%p", if_carp);
4575 	IF_DB_PRINTF("%p", if_label);
4576 	IF_DB_PRINTF("%p", if_netmap);
4577 	IF_DB_PRINTF("0x%08x", if_flags);
4578 	IF_DB_PRINTF("0x%08x", if_drv_flags);
4579 	IF_DB_PRINTF("0x%08x", if_capabilities);
4580 	IF_DB_PRINTF("0x%08x", if_capenable);
4581 	IF_DB_PRINTF("%p", if_snd.ifq_head);
4582 	IF_DB_PRINTF("%p", if_snd.ifq_tail);
4583 	IF_DB_PRINTF("%d", if_snd.ifq_len);
4584 	IF_DB_PRINTF("%d", if_snd.ifq_maxlen);
4585 	IF_DB_PRINTF("%p", if_snd.ifq_drv_head);
4586 	IF_DB_PRINTF("%p", if_snd.ifq_drv_tail);
4587 	IF_DB_PRINTF("%d", if_snd.ifq_drv_len);
4588 	IF_DB_PRINTF("%d", if_snd.ifq_drv_maxlen);
4589 	IF_DB_PRINTF("%d", if_snd.altq_type);
4590 	IF_DB_PRINTF("%x", if_snd.altq_flags);
4591 #undef IF_DB_PRINTF
4592 }
4593 
4594 DB_SHOW_COMMAND(ifnet, db_show_ifnet)
4595 {
4596 
4597 	if (!have_addr) {
4598 		db_printf("usage: show ifnet <struct ifnet *>\n");
4599 		return;
4600 	}
4601 
4602 	if_show_ifnet((struct ifnet *)addr);
4603 }
4604 
4605 DB_SHOW_ALL_COMMAND(ifnets, db_show_all_ifnets)
4606 {
4607 	VNET_ITERATOR_DECL(vnet_iter);
4608 	struct ifnet *ifp;
4609 	u_short idx;
4610 
4611 	VNET_FOREACH(vnet_iter) {
4612 		CURVNET_SET_QUIET(vnet_iter);
4613 #ifdef VIMAGE
4614 		db_printf("vnet=%p\n", curvnet);
4615 #endif
4616 		for (idx = 1; idx <= V_if_index; idx++) {
4617 			ifp = V_ifindex_table[idx];
4618 			if (ifp == NULL)
4619 				continue;
4620 			db_printf( "%20s ifp=%p\n", ifp->if_xname, ifp);
4621 			if (db_pager_quit)
4622 				break;
4623 		}
4624 		CURVNET_RESTORE();
4625 	}
4626 }
4627 #endif	/* DDB */
4628