xref: /freebsd/sys/net/if_vlan.c (revision 88640c0e8b6f503426cce9ea1337098c241d3801)
1 /*-
2  * Copyright 1998 Massachusetts Institute of Technology
3  * Copyright 2012 ADARA Networks, Inc.
4  * Copyright 2017 Dell EMC Isilon
5  *
6  * Portions of this software were developed by Robert N. M. Watson under
7  * contract to ADARA Networks, Inc.
8  *
9  * Permission to use, copy, modify, and distribute this software and
10  * its documentation for any purpose and without fee is hereby
11  * granted, provided that both the above copyright notice and this
12  * permission notice appear in all copies, that both the above
13  * copyright notice and this permission notice appear in all
14  * supporting documentation, and that the name of M.I.T. not be used
15  * in advertising or publicity pertaining to distribution of the
16  * software without specific, written prior permission.  M.I.T. makes
17  * no representations about the suitability of this software for any
18  * purpose.  It is provided "as is" without express or implied
19  * warranty.
20  *
21  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
22  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
23  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
24  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
25  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
28  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
29  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 /*
36  * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
37  * This is sort of sneaky in the implementation, since
38  * we need to pretend to be enough of an Ethernet implementation
39  * to make arp work.  The way we do this is by telling everyone
40  * that we are an Ethernet, and then catch the packets that
41  * ether_output() sends to us via if_transmit(), rewrite them for
42  * use by the real outgoing interface, and ask it to send them.
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_inet.h"
49 #include "opt_vlan.h"
50 #include "opt_ratelimit.h"
51 
52 #include <sys/param.h>
53 #include <sys/eventhandler.h>
54 #include <sys/kernel.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/mbuf.h>
58 #include <sys/module.h>
59 #include <sys/rmlock.h>
60 #include <sys/priv.h>
61 #include <sys/queue.h>
62 #include <sys/socket.h>
63 #include <sys/sockio.h>
64 #include <sys/sysctl.h>
65 #include <sys/systm.h>
66 #include <sys/sx.h>
67 #include <sys/taskqueue.h>
68 
69 #include <net/bpf.h>
70 #include <net/ethernet.h>
71 #include <net/if.h>
72 #include <net/if_var.h>
73 #include <net/if_clone.h>
74 #include <net/if_dl.h>
75 #include <net/if_types.h>
76 #include <net/if_vlan_var.h>
77 #include <net/vnet.h>
78 
79 #ifdef INET
80 #include <netinet/in.h>
81 #include <netinet/if_ether.h>
82 #endif
83 
84 #define	VLAN_DEF_HWIDTH	4
85 #define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
86 
87 #define	UP_AND_RUNNING(ifp) \
88     ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
89 
90 CK_SLIST_HEAD(ifvlanhead, ifvlan);
91 
92 struct ifvlantrunk {
93 	struct	ifnet   *parent;	/* parent interface of this trunk */
94 	struct	mtx	lock;
95 #ifdef VLAN_ARRAY
96 #define	VLAN_ARRAY_SIZE	(EVL_VLID_MASK + 1)
97 	struct	ifvlan	*vlans[VLAN_ARRAY_SIZE]; /* static table */
98 #else
99 	struct	ifvlanhead *hash;	/* dynamic hash-list table */
100 	uint16_t	hmask;
101 	uint16_t	hwidth;
102 #endif
103 	int		refcnt;
104 };
105 
106 /*
107  * This macro provides a facility to iterate over every vlan on a trunk with
108  * the assumption that none will be added/removed during iteration.
109  */
110 #ifdef VLAN_ARRAY
111 #define VLAN_FOREACH(_ifv, _trunk) \
112 	size_t _i; \
113 	for (_i = 0; _i < VLAN_ARRAY_SIZE; _i++) \
114 		if (((_ifv) = (_trunk)->vlans[_i]) != NULL)
115 #else /* VLAN_ARRAY */
116 #define VLAN_FOREACH(_ifv, _trunk) \
117 	struct ifvlan *_next; \
118 	size_t _i; \
119 	for (_i = 0; _i < (1 << (_trunk)->hwidth); _i++) \
120 		CK_SLIST_FOREACH_SAFE((_ifv), &(_trunk)->hash[_i], ifv_list, _next)
121 #endif /* VLAN_ARRAY */
122 
123 /*
124  * This macro provides a facility to iterate over every vlan on a trunk while
125  * also modifying the number of vlans on the trunk. The iteration continues
126  * until some condition is met or there are no more vlans on the trunk.
127  */
128 #ifdef VLAN_ARRAY
129 /* The VLAN_ARRAY case is simple -- just a for loop using the condition. */
130 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
131 	size_t _i; \
132 	for (_i = 0; !(_cond) && _i < VLAN_ARRAY_SIZE; _i++) \
133 		if (((_ifv) = (_trunk)->vlans[_i]))
134 #else /* VLAN_ARRAY */
135 /*
136  * The hash table case is more complicated. We allow for the hash table to be
137  * modified (i.e. vlans removed) while we are iterating over it. To allow for
138  * this we must restart the iteration every time we "touch" something during
139  * the iteration, since removal will resize the hash table and invalidate our
140  * current position. If acting on the touched element causes the trunk to be
141  * emptied, then iteration also stops.
142  */
143 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
144 	size_t _i; \
145 	bool _touch = false; \
146 	for (_i = 0; \
147 	    !(_cond) && _i < (1 << (_trunk)->hwidth); \
148 	    _i = (_touch && ((_trunk) != NULL) ? 0 : _i + 1), _touch = false) \
149 		if (((_ifv) = CK_SLIST_FIRST(&(_trunk)->hash[_i])) != NULL && \
150 		    (_touch = true))
151 #endif /* VLAN_ARRAY */
152 
153 struct vlan_mc_entry {
154 	struct sockaddr_dl		mc_addr;
155 	CK_SLIST_ENTRY(vlan_mc_entry)	mc_entries;
156 	struct epoch_context		mc_epoch_ctx;
157 };
158 
159 struct ifvlan {
160 	struct	ifvlantrunk *ifv_trunk;
161 	struct	ifnet *ifv_ifp;
162 #define	TRUNK(ifv)	((ifv)->ifv_trunk)
163 #define	PARENT(ifv)	((ifv)->ifv_trunk->parent)
164 	void	*ifv_cookie;
165 	int	ifv_pflags;	/* special flags we have set on parent */
166 	int	ifv_capenable;
167 	int	ifv_encaplen;	/* encapsulation length */
168 	int	ifv_mtufudge;	/* MTU fudged by this much */
169 	int	ifv_mintu;	/* min transmission unit */
170 	uint16_t ifv_proto;	/* encapsulation ethertype */
171 	uint16_t ifv_tag;	/* tag to apply on packets leaving if */
172 	uint16_t ifv_vid;	/* VLAN ID */
173 	uint8_t	ifv_pcp;	/* Priority Code Point (PCP). */
174 	struct task lladdr_task;
175 	CK_SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
176 #ifndef VLAN_ARRAY
177 	CK_SLIST_ENTRY(ifvlan) ifv_list;
178 #endif
179 };
180 
181 /* Special flags we should propagate to parent. */
182 static struct {
183 	int flag;
184 	int (*func)(struct ifnet *, int);
185 } vlan_pflags[] = {
186 	{IFF_PROMISC, ifpromisc},
187 	{IFF_ALLMULTI, if_allmulti},
188 	{0, NULL}
189 };
190 
191 extern int vlan_mtag_pcp;
192 
193 static const char vlanname[] = "vlan";
194 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
195 
196 static eventhandler_tag ifdetach_tag;
197 static eventhandler_tag iflladdr_tag;
198 
199 /*
200  * if_vlan uses two module-level synchronizations primitives to allow concurrent
201  * modification of vlan interfaces and (mostly) allow for vlans to be destroyed
202  * while they are being used for tx/rx. To accomplish this in a way that has
203  * acceptable performance and cooperation with other parts of the network stack
204  * there is a non-sleepable epoch(9) and an sx(9).
205  *
206  * The performance-sensitive paths that warrant using the epoch(9) are
207  * vlan_transmit and vlan_input. Both have to check for the vlan interface's
208  * existence using if_vlantrunk, and being in the network tx/rx paths the use
209  * of an epoch(9) gives a measureable improvement in performance.
210  *
211  * The reason for having an sx(9) is mostly because there are still areas that
212  * must be sleepable and also have safe concurrent access to a vlan interface.
213  * Since the sx(9) exists, it is used by default in most paths unless sleeping
214  * is not permitted, or if it is not clear whether sleeping is permitted.
215  *
216  */
217 #define _VLAN_SX_ID ifv_sx
218 
219 static struct sx _VLAN_SX_ID;
220 
221 #define VLAN_LOCKING_INIT() \
222 	sx_init(&_VLAN_SX_ID, "vlan_sx")
223 
224 #define VLAN_LOCKING_DESTROY() \
225 	sx_destroy(&_VLAN_SX_ID)
226 
227 #define	VLAN_SLOCK()			sx_slock(&_VLAN_SX_ID)
228 #define	VLAN_SUNLOCK()			sx_sunlock(&_VLAN_SX_ID)
229 #define	VLAN_XLOCK()			sx_xlock(&_VLAN_SX_ID)
230 #define	VLAN_XUNLOCK()			sx_xunlock(&_VLAN_SX_ID)
231 #define	VLAN_SLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_SLOCKED)
232 #define	VLAN_XLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_XLOCKED)
233 #define	VLAN_SXLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_LOCKED)
234 
235 
236 /*
237  * We also have a per-trunk mutex that should be acquired when changing
238  * its state.
239  */
240 #define	TRUNK_LOCK_INIT(trunk)		mtx_init(&(trunk)->lock, vlanname, NULL, MTX_DEF)
241 #define	TRUNK_LOCK_DESTROY(trunk)	mtx_destroy(&(trunk)->lock)
242 #define	TRUNK_WLOCK(trunk)		mtx_lock(&(trunk)->lock)
243 #define	TRUNK_WUNLOCK(trunk)		mtx_unlock(&(trunk)->lock)
244 #define	TRUNK_LOCK_ASSERT(trunk)	MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(trunk)->lock))
245 #define	TRUNK_WLOCK_ASSERT(trunk)	mtx_assert(&(trunk)->lock, MA_OWNED);
246 
247 /*
248  * The VLAN_ARRAY substitutes the dynamic hash with a static array
249  * with 4096 entries. In theory this can give a boost in processing,
250  * however in practice it does not. Probably this is because the array
251  * is too big to fit into CPU cache.
252  */
253 #ifndef VLAN_ARRAY
254 static	void vlan_inithash(struct ifvlantrunk *trunk);
255 static	void vlan_freehash(struct ifvlantrunk *trunk);
256 static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
257 static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
258 static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
259 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
260 	uint16_t vid);
261 #endif
262 static	void trunk_destroy(struct ifvlantrunk *trunk);
263 
264 static	void vlan_init(void *foo);
265 static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
266 static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
267 #ifdef RATELIMIT
268 static	int vlan_snd_tag_alloc(struct ifnet *,
269     union if_snd_tag_alloc_params *, struct m_snd_tag **);
270 #endif
271 static	void vlan_qflush(struct ifnet *ifp);
272 static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
273     int (*func)(struct ifnet *, int));
274 static	int vlan_setflags(struct ifnet *ifp, int status);
275 static	int vlan_setmulti(struct ifnet *ifp);
276 static	int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
277 static	void vlan_unconfig(struct ifnet *ifp);
278 static	void vlan_unconfig_locked(struct ifnet *ifp, int departing);
279 static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
280 static	void vlan_link_state(struct ifnet *ifp);
281 static	void vlan_capabilities(struct ifvlan *ifv);
282 static	void vlan_trunk_capabilities(struct ifnet *ifp);
283 
284 static	struct ifnet *vlan_clone_match_ethervid(const char *, int *);
285 static	int vlan_clone_match(struct if_clone *, const char *);
286 static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
287 static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
288 
289 static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
290 static  void vlan_iflladdr(void *arg, struct ifnet *ifp);
291 
292 static  void vlan_lladdr_fn(void *arg, int pending);
293 
294 static struct if_clone *vlan_cloner;
295 
296 #ifdef VIMAGE
297 VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner);
298 #define	V_vlan_cloner	VNET(vlan_cloner)
299 #endif
300 
301 static void
302 vlan_mc_free(struct epoch_context *ctx)
303 {
304 	struct vlan_mc_entry *mc = __containerof(ctx, struct vlan_mc_entry, mc_epoch_ctx);
305 	free(mc, M_VLAN);
306 }
307 
308 #ifndef VLAN_ARRAY
309 #define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
310 
311 static void
312 vlan_inithash(struct ifvlantrunk *trunk)
313 {
314 	int i, n;
315 
316 	/*
317 	 * The trunk must not be locked here since we call malloc(M_WAITOK).
318 	 * It is OK in case this function is called before the trunk struct
319 	 * gets hooked up and becomes visible from other threads.
320 	 */
321 
322 	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
323 	    ("%s: hash already initialized", __func__));
324 
325 	trunk->hwidth = VLAN_DEF_HWIDTH;
326 	n = 1 << trunk->hwidth;
327 	trunk->hmask = n - 1;
328 	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
329 	for (i = 0; i < n; i++)
330 		CK_SLIST_INIT(&trunk->hash[i]);
331 }
332 
333 static void
334 vlan_freehash(struct ifvlantrunk *trunk)
335 {
336 #ifdef INVARIANTS
337 	int i;
338 
339 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
340 	for (i = 0; i < (1 << trunk->hwidth); i++)
341 		KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]),
342 		    ("%s: hash table not empty", __func__));
343 #endif
344 	free(trunk->hash, M_VLAN);
345 	trunk->hash = NULL;
346 	trunk->hwidth = trunk->hmask = 0;
347 }
348 
349 static int
350 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
351 {
352 	int i, b;
353 	struct ifvlan *ifv2;
354 
355 	VLAN_XLOCK_ASSERT();
356 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
357 
358 	b = 1 << trunk->hwidth;
359 	i = HASH(ifv->ifv_vid, trunk->hmask);
360 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
361 		if (ifv->ifv_vid == ifv2->ifv_vid)
362 			return (EEXIST);
363 
364 	/*
365 	 * Grow the hash when the number of vlans exceeds half of the number of
366 	 * hash buckets squared. This will make the average linked-list length
367 	 * buckets/2.
368 	 */
369 	if (trunk->refcnt > (b * b) / 2) {
370 		vlan_growhash(trunk, 1);
371 		i = HASH(ifv->ifv_vid, trunk->hmask);
372 	}
373 	CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
374 	trunk->refcnt++;
375 
376 	return (0);
377 }
378 
379 static int
380 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
381 {
382 	int i, b;
383 	struct ifvlan *ifv2;
384 
385 	VLAN_XLOCK_ASSERT();
386 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
387 
388 	b = 1 << trunk->hwidth;
389 	i = HASH(ifv->ifv_vid, trunk->hmask);
390 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
391 		if (ifv2 == ifv) {
392 			trunk->refcnt--;
393 			CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list);
394 			if (trunk->refcnt < (b * b) / 2)
395 				vlan_growhash(trunk, -1);
396 			return (0);
397 		}
398 
399 	panic("%s: vlan not found\n", __func__);
400 	return (ENOENT); /*NOTREACHED*/
401 }
402 
403 /*
404  * Grow the hash larger or smaller if memory permits.
405  */
406 static void
407 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
408 {
409 	struct ifvlan *ifv;
410 	struct ifvlanhead *hash2;
411 	int hwidth2, i, j, n, n2;
412 
413 	VLAN_XLOCK_ASSERT();
414 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
415 
416 	if (howmuch == 0) {
417 		/* Harmless yet obvious coding error */
418 		printf("%s: howmuch is 0\n", __func__);
419 		return;
420 	}
421 
422 	hwidth2 = trunk->hwidth + howmuch;
423 	n = 1 << trunk->hwidth;
424 	n2 = 1 << hwidth2;
425 	/* Do not shrink the table below the default */
426 	if (hwidth2 < VLAN_DEF_HWIDTH)
427 		return;
428 
429 	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK);
430 	if (hash2 == NULL) {
431 		printf("%s: out of memory -- hash size not changed\n",
432 		    __func__);
433 		return;		/* We can live with the old hash table */
434 	}
435 	for (j = 0; j < n2; j++)
436 		CK_SLIST_INIT(&hash2[j]);
437 	for (i = 0; i < n; i++)
438 		while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) {
439 			CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list);
440 			j = HASH(ifv->ifv_vid, n2 - 1);
441 			CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
442 		}
443 	NET_EPOCH_WAIT();
444 	free(trunk->hash, M_VLAN);
445 	trunk->hash = hash2;
446 	trunk->hwidth = hwidth2;
447 	trunk->hmask = n2 - 1;
448 
449 	if (bootverbose)
450 		if_printf(trunk->parent,
451 		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
452 }
453 
454 static __inline struct ifvlan *
455 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
456 {
457 	struct ifvlan *ifv;
458 
459 	NET_EPOCH_ASSERT();
460 
461 	CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
462 		if (ifv->ifv_vid == vid)
463 			return (ifv);
464 	return (NULL);
465 }
466 
467 #if 0
468 /* Debugging code to view the hashtables. */
469 static void
470 vlan_dumphash(struct ifvlantrunk *trunk)
471 {
472 	int i;
473 	struct ifvlan *ifv;
474 
475 	for (i = 0; i < (1 << trunk->hwidth); i++) {
476 		printf("%d: ", i);
477 		CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
478 			printf("%s ", ifv->ifv_ifp->if_xname);
479 		printf("\n");
480 	}
481 }
482 #endif /* 0 */
483 #else
484 
485 static __inline struct ifvlan *
486 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
487 {
488 
489 	return trunk->vlans[vid];
490 }
491 
492 static __inline int
493 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
494 {
495 
496 	if (trunk->vlans[ifv->ifv_vid] != NULL)
497 		return EEXIST;
498 	trunk->vlans[ifv->ifv_vid] = ifv;
499 	trunk->refcnt++;
500 
501 	return (0);
502 }
503 
504 static __inline int
505 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
506 {
507 
508 	trunk->vlans[ifv->ifv_vid] = NULL;
509 	trunk->refcnt--;
510 
511 	return (0);
512 }
513 
514 static __inline void
515 vlan_freehash(struct ifvlantrunk *trunk)
516 {
517 }
518 
519 static __inline void
520 vlan_inithash(struct ifvlantrunk *trunk)
521 {
522 }
523 
524 #endif /* !VLAN_ARRAY */
525 
526 static void
527 trunk_destroy(struct ifvlantrunk *trunk)
528 {
529 	VLAN_XLOCK_ASSERT();
530 
531 	vlan_freehash(trunk);
532 	trunk->parent->if_vlantrunk = NULL;
533 	TRUNK_LOCK_DESTROY(trunk);
534 	if_rele(trunk->parent);
535 	free(trunk, M_VLAN);
536 }
537 
538 /*
539  * Program our multicast filter. What we're actually doing is
540  * programming the multicast filter of the parent. This has the
541  * side effect of causing the parent interface to receive multicast
542  * traffic that it doesn't really want, which ends up being discarded
543  * later by the upper protocol layers. Unfortunately, there's no way
544  * to avoid this: there really is only one physical interface.
545  */
546 static int
547 vlan_setmulti(struct ifnet *ifp)
548 {
549 	struct ifnet		*ifp_p;
550 	struct ifmultiaddr	*ifma;
551 	struct ifvlan		*sc;
552 	struct vlan_mc_entry	*mc;
553 	int			error;
554 
555 	VLAN_XLOCK_ASSERT();
556 
557 	/* Find the parent. */
558 	sc = ifp->if_softc;
559 	ifp_p = PARENT(sc);
560 
561 	CURVNET_SET_QUIET(ifp_p->if_vnet);
562 
563 	/* First, remove any existing filter entries. */
564 	while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
565 		CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
566 		(void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
567 		epoch_call(net_epoch_preempt, &mc->mc_epoch_ctx, vlan_mc_free);
568 	}
569 
570 	/* Now program new ones. */
571 	IF_ADDR_WLOCK(ifp);
572 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
573 		if (ifma->ifma_addr->sa_family != AF_LINK)
574 			continue;
575 		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
576 		if (mc == NULL) {
577 			IF_ADDR_WUNLOCK(ifp);
578 			return (ENOMEM);
579 		}
580 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
581 		mc->mc_addr.sdl_index = ifp_p->if_index;
582 		CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
583 	}
584 	IF_ADDR_WUNLOCK(ifp);
585 	CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
586 		error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
587 		    NULL);
588 		if (error)
589 			return (error);
590 	}
591 
592 	CURVNET_RESTORE();
593 	return (0);
594 }
595 
596 /*
597  * A handler for parent interface link layer address changes.
598  * If the parent interface link layer address is changed we
599  * should also change it on all children vlans.
600  */
601 static void
602 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
603 {
604 	struct epoch_tracker et;
605 	struct ifvlan *ifv;
606 	struct ifnet *ifv_ifp;
607 	struct ifvlantrunk *trunk;
608 	struct sockaddr_dl *sdl;
609 
610 	/* Need the epoch since this is run on taskqueue_swi. */
611 	NET_EPOCH_ENTER(et);
612 	trunk = ifp->if_vlantrunk;
613 	if (trunk == NULL) {
614 		NET_EPOCH_EXIT(et);
615 		return;
616 	}
617 
618 	/*
619 	 * OK, it's a trunk.  Loop over and change all vlan's lladdrs on it.
620 	 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
621 	 * ioctl calls on the parent garbling the lladdr of the child vlan.
622 	 */
623 	TRUNK_WLOCK(trunk);
624 	VLAN_FOREACH(ifv, trunk) {
625 		/*
626 		 * Copy new new lladdr into the ifv_ifp, enqueue a task
627 		 * to actually call if_setlladdr. if_setlladdr needs to
628 		 * be deferred to a taskqueue because it will call into
629 		 * the if_vlan ioctl path and try to acquire the global
630 		 * lock.
631 		 */
632 		ifv_ifp = ifv->ifv_ifp;
633 		bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
634 		    ifp->if_addrlen);
635 		sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
636 		sdl->sdl_alen = ifp->if_addrlen;
637 		taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
638 	}
639 	TRUNK_WUNLOCK(trunk);
640 	NET_EPOCH_EXIT(et);
641 }
642 
643 /*
644  * A handler for network interface departure events.
645  * Track departure of trunks here so that we don't access invalid
646  * pointers or whatever if a trunk is ripped from under us, e.g.,
647  * by ejecting its hot-plug card.  However, if an ifnet is simply
648  * being renamed, then there's no need to tear down the state.
649  */
650 static void
651 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
652 {
653 	struct ifvlan *ifv;
654 	struct ifvlantrunk *trunk;
655 
656 	/* If the ifnet is just being renamed, don't do anything. */
657 	if (ifp->if_flags & IFF_RENAMING)
658 		return;
659 	VLAN_XLOCK();
660 	trunk = ifp->if_vlantrunk;
661 	if (trunk == NULL) {
662 		VLAN_XUNLOCK();
663 		return;
664 	}
665 
666 	/*
667 	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
668 	 * Check trunk pointer after each vlan_unconfig() as it will
669 	 * free it and set to NULL after the last vlan was detached.
670 	 */
671 	VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
672 	    ifp->if_vlantrunk == NULL)
673 		vlan_unconfig_locked(ifv->ifv_ifp, 1);
674 
675 	/* Trunk should have been destroyed in vlan_unconfig(). */
676 	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
677 	VLAN_XUNLOCK();
678 }
679 
680 /*
681  * Return the trunk device for a virtual interface.
682  */
683 static struct ifnet  *
684 vlan_trunkdev(struct ifnet *ifp)
685 {
686 	struct epoch_tracker et;
687 	struct ifvlan *ifv;
688 
689 	if (ifp->if_type != IFT_L2VLAN)
690 		return (NULL);
691 
692 	NET_EPOCH_ENTER(et);
693 	ifv = ifp->if_softc;
694 	ifp = NULL;
695 	if (ifv->ifv_trunk)
696 		ifp = PARENT(ifv);
697 	NET_EPOCH_EXIT(et);
698 	return (ifp);
699 }
700 
701 /*
702  * Return the 12-bit VLAN VID for this interface, for use by external
703  * components such as Infiniband.
704  *
705  * XXXRW: Note that the function name here is historical; it should be named
706  * vlan_vid().
707  */
708 static int
709 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
710 {
711 	struct ifvlan *ifv;
712 
713 	if (ifp->if_type != IFT_L2VLAN)
714 		return (EINVAL);
715 	ifv = ifp->if_softc;
716 	*vidp = ifv->ifv_vid;
717 	return (0);
718 }
719 
720 static int
721 vlan_pcp(struct ifnet *ifp, uint16_t *pcpp)
722 {
723 	struct ifvlan *ifv;
724 
725 	if (ifp->if_type != IFT_L2VLAN)
726 		return (EINVAL);
727 	ifv = ifp->if_softc;
728 	*pcpp = ifv->ifv_pcp;
729 	return (0);
730 }
731 
732 /*
733  * Return a driver specific cookie for this interface.  Synchronization
734  * with setcookie must be provided by the driver.
735  */
736 static void *
737 vlan_cookie(struct ifnet *ifp)
738 {
739 	struct ifvlan *ifv;
740 
741 	if (ifp->if_type != IFT_L2VLAN)
742 		return (NULL);
743 	ifv = ifp->if_softc;
744 	return (ifv->ifv_cookie);
745 }
746 
747 /*
748  * Store a cookie in our softc that drivers can use to store driver
749  * private per-instance data in.
750  */
751 static int
752 vlan_setcookie(struct ifnet *ifp, void *cookie)
753 {
754 	struct ifvlan *ifv;
755 
756 	if (ifp->if_type != IFT_L2VLAN)
757 		return (EINVAL);
758 	ifv = ifp->if_softc;
759 	ifv->ifv_cookie = cookie;
760 	return (0);
761 }
762 
763 /*
764  * Return the vlan device present at the specific VID.
765  */
766 static struct ifnet *
767 vlan_devat(struct ifnet *ifp, uint16_t vid)
768 {
769 	struct epoch_tracker et;
770 	struct ifvlantrunk *trunk;
771 	struct ifvlan *ifv;
772 
773 	NET_EPOCH_ENTER(et);
774 	trunk = ifp->if_vlantrunk;
775 	if (trunk == NULL) {
776 		NET_EPOCH_EXIT(et);
777 		return (NULL);
778 	}
779 	ifp = NULL;
780 	ifv = vlan_gethash(trunk, vid);
781 	if (ifv)
782 		ifp = ifv->ifv_ifp;
783 	NET_EPOCH_EXIT(et);
784 	return (ifp);
785 }
786 
787 /*
788  * Recalculate the cached VLAN tag exposed via the MIB.
789  */
790 static void
791 vlan_tag_recalculate(struct ifvlan *ifv)
792 {
793 
794        ifv->ifv_tag = EVL_MAKETAG(ifv->ifv_vid, ifv->ifv_pcp, 0);
795 }
796 
797 /*
798  * VLAN support can be loaded as a module.  The only place in the
799  * system that's intimately aware of this is ether_input.  We hook
800  * into this code through vlan_input_p which is defined there and
801  * set here.  No one else in the system should be aware of this so
802  * we use an explicit reference here.
803  */
804 extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
805 
806 /* For if_link_state_change() eyes only... */
807 extern	void (*vlan_link_state_p)(struct ifnet *);
808 
809 static int
810 vlan_modevent(module_t mod, int type, void *data)
811 {
812 
813 	switch (type) {
814 	case MOD_LOAD:
815 		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
816 		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
817 		if (ifdetach_tag == NULL)
818 			return (ENOMEM);
819 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
820 		    vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
821 		if (iflladdr_tag == NULL)
822 			return (ENOMEM);
823 		VLAN_LOCKING_INIT();
824 		vlan_input_p = vlan_input;
825 		vlan_link_state_p = vlan_link_state;
826 		vlan_trunk_cap_p = vlan_trunk_capabilities;
827 		vlan_trunkdev_p = vlan_trunkdev;
828 		vlan_cookie_p = vlan_cookie;
829 		vlan_setcookie_p = vlan_setcookie;
830 		vlan_tag_p = vlan_tag;
831 		vlan_pcp_p = vlan_pcp;
832 		vlan_devat_p = vlan_devat;
833 #ifndef VIMAGE
834 		vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
835 		    vlan_clone_create, vlan_clone_destroy);
836 #endif
837 		if (bootverbose)
838 			printf("vlan: initialized, using "
839 #ifdef VLAN_ARRAY
840 			       "full-size arrays"
841 #else
842 			       "hash tables with chaining"
843 #endif
844 
845 			       "\n");
846 		break;
847 	case MOD_UNLOAD:
848 #ifndef VIMAGE
849 		if_clone_detach(vlan_cloner);
850 #endif
851 		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
852 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
853 		vlan_input_p = NULL;
854 		vlan_link_state_p = NULL;
855 		vlan_trunk_cap_p = NULL;
856 		vlan_trunkdev_p = NULL;
857 		vlan_tag_p = NULL;
858 		vlan_cookie_p = NULL;
859 		vlan_setcookie_p = NULL;
860 		vlan_devat_p = NULL;
861 		VLAN_LOCKING_DESTROY();
862 		if (bootverbose)
863 			printf("vlan: unloaded\n");
864 		break;
865 	default:
866 		return (EOPNOTSUPP);
867 	}
868 	return (0);
869 }
870 
871 static moduledata_t vlan_mod = {
872 	"if_vlan",
873 	vlan_modevent,
874 	0
875 };
876 
877 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
878 MODULE_VERSION(if_vlan, 3);
879 
880 #ifdef VIMAGE
881 static void
882 vnet_vlan_init(const void *unused __unused)
883 {
884 
885 	vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
886 		    vlan_clone_create, vlan_clone_destroy);
887 	V_vlan_cloner = vlan_cloner;
888 }
889 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
890     vnet_vlan_init, NULL);
891 
892 static void
893 vnet_vlan_uninit(const void *unused __unused)
894 {
895 
896 	if_clone_detach(V_vlan_cloner);
897 }
898 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
899     vnet_vlan_uninit, NULL);
900 #endif
901 
902 /*
903  * Check for <etherif>.<vlan> style interface names.
904  */
905 static struct ifnet *
906 vlan_clone_match_ethervid(const char *name, int *vidp)
907 {
908 	char ifname[IFNAMSIZ];
909 	char *cp;
910 	struct ifnet *ifp;
911 	int vid;
912 
913 	strlcpy(ifname, name, IFNAMSIZ);
914 	if ((cp = strchr(ifname, '.')) == NULL)
915 		return (NULL);
916 	*cp = '\0';
917 	if ((ifp = ifunit_ref(ifname)) == NULL)
918 		return (NULL);
919 	/* Parse VID. */
920 	if (*++cp == '\0') {
921 		if_rele(ifp);
922 		return (NULL);
923 	}
924 	vid = 0;
925 	for(; *cp >= '0' && *cp <= '9'; cp++)
926 		vid = (vid * 10) + (*cp - '0');
927 	if (*cp != '\0') {
928 		if_rele(ifp);
929 		return (NULL);
930 	}
931 	if (vidp != NULL)
932 		*vidp = vid;
933 
934 	return (ifp);
935 }
936 
937 static int
938 vlan_clone_match(struct if_clone *ifc, const char *name)
939 {
940 	const char *cp;
941 
942 	if (vlan_clone_match_ethervid(name, NULL) != NULL)
943 		return (1);
944 
945 	if (strncmp(vlanname, name, strlen(vlanname)) != 0)
946 		return (0);
947 	for (cp = name + 4; *cp != '\0'; cp++) {
948 		if (*cp < '0' || *cp > '9')
949 			return (0);
950 	}
951 
952 	return (1);
953 }
954 
955 static int
956 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
957 {
958 	char *dp;
959 	int wildcard;
960 	int unit;
961 	int error;
962 	int vid;
963 	struct ifvlan *ifv;
964 	struct ifnet *ifp;
965 	struct ifnet *p;
966 	struct ifaddr *ifa;
967 	struct sockaddr_dl *sdl;
968 	struct vlanreq vlr;
969 	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
970 
971 	/*
972 	 * There are 3 (ugh) ways to specify the cloned device:
973 	 * o pass a parameter block with the clone request.
974 	 * o specify parameters in the text of the clone device name
975 	 * o specify no parameters and get an unattached device that
976 	 *   must be configured separately.
977 	 * The first technique is preferred; the latter two are
978 	 * supported for backwards compatibility.
979 	 *
980 	 * XXXRW: Note historic use of the word "tag" here.  New ioctls may be
981 	 * called for.
982 	 */
983 	if (params) {
984 		error = copyin(params, &vlr, sizeof(vlr));
985 		if (error)
986 			return error;
987 		p = ifunit_ref(vlr.vlr_parent);
988 		if (p == NULL)
989 			return (ENXIO);
990 		error = ifc_name2unit(name, &unit);
991 		if (error != 0) {
992 			if_rele(p);
993 			return (error);
994 		}
995 		vid = vlr.vlr_tag;
996 		wildcard = (unit < 0);
997 	} else if ((p = vlan_clone_match_ethervid(name, &vid)) != NULL) {
998 		unit = -1;
999 		wildcard = 0;
1000 	} else {
1001 		p = NULL;
1002 		error = ifc_name2unit(name, &unit);
1003 		if (error != 0)
1004 			return (error);
1005 
1006 		wildcard = (unit < 0);
1007 	}
1008 
1009 	error = ifc_alloc_unit(ifc, &unit);
1010 	if (error != 0) {
1011 		if (p != NULL)
1012 			if_rele(p);
1013 		return (error);
1014 	}
1015 
1016 	/* In the wildcard case, we need to update the name. */
1017 	if (wildcard) {
1018 		for (dp = name; *dp != '\0'; dp++);
1019 		if (snprintf(dp, len - (dp-name), "%d", unit) >
1020 		    len - (dp-name) - 1) {
1021 			panic("%s: interface name too long", __func__);
1022 		}
1023 	}
1024 
1025 	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
1026 	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
1027 	if (ifp == NULL) {
1028 		ifc_free_unit(ifc, unit);
1029 		free(ifv, M_VLAN);
1030 		if (p != NULL)
1031 			if_rele(p);
1032 		return (ENOSPC);
1033 	}
1034 	CK_SLIST_INIT(&ifv->vlan_mc_listhead);
1035 	ifp->if_softc = ifv;
1036 	/*
1037 	 * Set the name manually rather than using if_initname because
1038 	 * we don't conform to the default naming convention for interfaces.
1039 	 */
1040 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
1041 	ifp->if_dname = vlanname;
1042 	ifp->if_dunit = unit;
1043 
1044 	ifp->if_init = vlan_init;
1045 	ifp->if_transmit = vlan_transmit;
1046 	ifp->if_qflush = vlan_qflush;
1047 	ifp->if_ioctl = vlan_ioctl;
1048 #ifdef RATELIMIT
1049 	ifp->if_snd_tag_alloc = vlan_snd_tag_alloc;
1050 #endif
1051 	ifp->if_flags = VLAN_IFFLAGS;
1052 	ether_ifattach(ifp, eaddr);
1053 	/* Now undo some of the damage... */
1054 	ifp->if_baudrate = 0;
1055 	ifp->if_type = IFT_L2VLAN;
1056 	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
1057 	ifa = ifp->if_addr;
1058 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1059 	sdl->sdl_type = IFT_L2VLAN;
1060 
1061 	if (p != NULL) {
1062 		error = vlan_config(ifv, p, vid);
1063 		if_rele(p);
1064 		if (error != 0) {
1065 			/*
1066 			 * Since we've partially failed, we need to back
1067 			 * out all the way, otherwise userland could get
1068 			 * confused.  Thus, we destroy the interface.
1069 			 */
1070 			ether_ifdetach(ifp);
1071 			vlan_unconfig(ifp);
1072 			if_free(ifp);
1073 			ifc_free_unit(ifc, unit);
1074 			free(ifv, M_VLAN);
1075 
1076 			return (error);
1077 		}
1078 	}
1079 
1080 	return (0);
1081 }
1082 
1083 static int
1084 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1085 {
1086 	struct ifvlan *ifv = ifp->if_softc;
1087 	int unit = ifp->if_dunit;
1088 
1089 	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
1090 	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
1091 	/*
1092 	 * We should have the only reference to the ifv now, so we can now
1093 	 * drain any remaining lladdr task before freeing the ifnet and the
1094 	 * ifvlan.
1095 	 */
1096 	taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
1097 	NET_EPOCH_WAIT();
1098 	if_free(ifp);
1099 	free(ifv, M_VLAN);
1100 	ifc_free_unit(ifc, unit);
1101 
1102 	return (0);
1103 }
1104 
1105 /*
1106  * The ifp->if_init entry point for vlan(4) is a no-op.
1107  */
1108 static void
1109 vlan_init(void *foo __unused)
1110 {
1111 }
1112 
1113 /*
1114  * The if_transmit method for vlan(4) interface.
1115  */
1116 static int
1117 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1118 {
1119 	struct epoch_tracker et;
1120 	struct ifvlan *ifv;
1121 	struct ifnet *p;
1122 	int error, len, mcast;
1123 
1124 	NET_EPOCH_ENTER(et);
1125 	ifv = ifp->if_softc;
1126 	if (TRUNK(ifv) == NULL) {
1127 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1128 		NET_EPOCH_EXIT(et);
1129 		m_freem(m);
1130 		return (ENETDOWN);
1131 	}
1132 	p = PARENT(ifv);
1133 	len = m->m_pkthdr.len;
1134 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1135 
1136 	BPF_MTAP(ifp, m);
1137 
1138 	/*
1139 	 * Do not run parent's if_transmit() if the parent is not up,
1140 	 * or parent's driver will cause a system crash.
1141 	 */
1142 	if (!UP_AND_RUNNING(p)) {
1143 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1144 		NET_EPOCH_EXIT(et);
1145 		m_freem(m);
1146 		return (ENETDOWN);
1147 	}
1148 
1149 	if (!ether_8021q_frame(&m, ifp, p, ifv->ifv_vid, ifv->ifv_pcp)) {
1150 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1151 		NET_EPOCH_EXIT(et);
1152 		return (0);
1153 	}
1154 
1155 	/*
1156 	 * Send it, precisely as ether_output() would have.
1157 	 */
1158 	error = (p->if_transmit)(p, m);
1159 	if (error == 0) {
1160 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1161 		if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1162 		if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1163 	} else
1164 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1165 	NET_EPOCH_EXIT(et);
1166 	return (error);
1167 }
1168 
1169 /*
1170  * The ifp->if_qflush entry point for vlan(4) is a no-op.
1171  */
1172 static void
1173 vlan_qflush(struct ifnet *ifp __unused)
1174 {
1175 }
1176 
1177 static void
1178 vlan_input(struct ifnet *ifp, struct mbuf *m)
1179 {
1180 	struct epoch_tracker et;
1181 	struct ifvlantrunk *trunk;
1182 	struct ifvlan *ifv;
1183 	struct m_tag *mtag;
1184 	uint16_t vid, tag;
1185 
1186 	NET_EPOCH_ENTER(et);
1187 	trunk = ifp->if_vlantrunk;
1188 	if (trunk == NULL) {
1189 		NET_EPOCH_EXIT(et);
1190 		m_freem(m);
1191 		return;
1192 	}
1193 
1194 	if (m->m_flags & M_VLANTAG) {
1195 		/*
1196 		 * Packet is tagged, but m contains a normal
1197 		 * Ethernet frame; the tag is stored out-of-band.
1198 		 */
1199 		tag = m->m_pkthdr.ether_vtag;
1200 		m->m_flags &= ~M_VLANTAG;
1201 	} else {
1202 		struct ether_vlan_header *evl;
1203 
1204 		/*
1205 		 * Packet is tagged in-band as specified by 802.1q.
1206 		 */
1207 		switch (ifp->if_type) {
1208 		case IFT_ETHER:
1209 			if (m->m_len < sizeof(*evl) &&
1210 			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
1211 				if_printf(ifp, "cannot pullup VLAN header\n");
1212 				NET_EPOCH_EXIT(et);
1213 				return;
1214 			}
1215 			evl = mtod(m, struct ether_vlan_header *);
1216 			tag = ntohs(evl->evl_tag);
1217 
1218 			/*
1219 			 * Remove the 802.1q header by copying the Ethernet
1220 			 * addresses over it and adjusting the beginning of
1221 			 * the data in the mbuf.  The encapsulated Ethernet
1222 			 * type field is already in place.
1223 			 */
1224 			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1225 			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
1226 			m_adj(m, ETHER_VLAN_ENCAP_LEN);
1227 			break;
1228 
1229 		default:
1230 #ifdef INVARIANTS
1231 			panic("%s: %s has unsupported if_type %u",
1232 			      __func__, ifp->if_xname, ifp->if_type);
1233 #endif
1234 			if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1235 			NET_EPOCH_EXIT(et);
1236 			m_freem(m);
1237 			return;
1238 		}
1239 	}
1240 
1241 	vid = EVL_VLANOFTAG(tag);
1242 
1243 	ifv = vlan_gethash(trunk, vid);
1244 	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1245 		NET_EPOCH_EXIT(et);
1246 		if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1247 		m_freem(m);
1248 		return;
1249 	}
1250 
1251 	if (vlan_mtag_pcp) {
1252 		/*
1253 		 * While uncommon, it is possible that we will find a 802.1q
1254 		 * packet encapsulated inside another packet that also had an
1255 		 * 802.1q header.  For example, ethernet tunneled over IPSEC
1256 		 * arriving over ethernet.  In that case, we replace the
1257 		 * existing 802.1q PCP m_tag value.
1258 		 */
1259 		mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1260 		if (mtag == NULL) {
1261 			mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1262 			    sizeof(uint8_t), M_NOWAIT);
1263 			if (mtag == NULL) {
1264 				if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1265 				NET_EPOCH_EXIT(et);
1266 				m_freem(m);
1267 				return;
1268 			}
1269 			m_tag_prepend(m, mtag);
1270 		}
1271 		*(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1272 	}
1273 
1274 	m->m_pkthdr.rcvif = ifv->ifv_ifp;
1275 	if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1276 	NET_EPOCH_EXIT(et);
1277 
1278 	/* Pass it back through the parent's input routine. */
1279 	(*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
1280 }
1281 
1282 static void
1283 vlan_lladdr_fn(void *arg, int pending __unused)
1284 {
1285 	struct ifvlan *ifv;
1286 	struct ifnet *ifp;
1287 
1288 	ifv = (struct ifvlan *)arg;
1289 	ifp = ifv->ifv_ifp;
1290 
1291 	CURVNET_SET(ifp->if_vnet);
1292 
1293 	/* The ifv_ifp already has the lladdr copied in. */
1294 	if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1295 
1296 	CURVNET_RESTORE();
1297 }
1298 
1299 static int
1300 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid)
1301 {
1302 	struct epoch_tracker et;
1303 	struct ifvlantrunk *trunk;
1304 	struct ifnet *ifp;
1305 	int error = 0;
1306 
1307 	/*
1308 	 * We can handle non-ethernet hardware types as long as
1309 	 * they handle the tagging and headers themselves.
1310 	 */
1311 	if (p->if_type != IFT_ETHER &&
1312 	    (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1313 		return (EPROTONOSUPPORT);
1314 	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1315 		return (EPROTONOSUPPORT);
1316 	/*
1317 	 * Don't let the caller set up a VLAN VID with
1318 	 * anything except VLID bits.
1319 	 * VID numbers 0x0 and 0xFFF are reserved.
1320 	 */
1321 	if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1322 		return (EINVAL);
1323 	if (ifv->ifv_trunk)
1324 		return (EBUSY);
1325 
1326 	VLAN_XLOCK();
1327 	if (p->if_vlantrunk == NULL) {
1328 		trunk = malloc(sizeof(struct ifvlantrunk),
1329 		    M_VLAN, M_WAITOK | M_ZERO);
1330 		vlan_inithash(trunk);
1331 		TRUNK_LOCK_INIT(trunk);
1332 		TRUNK_WLOCK(trunk);
1333 		p->if_vlantrunk = trunk;
1334 		trunk->parent = p;
1335 		if_ref(trunk->parent);
1336 		TRUNK_WUNLOCK(trunk);
1337 	} else {
1338 		trunk = p->if_vlantrunk;
1339 	}
1340 
1341 	ifv->ifv_vid = vid;	/* must set this before vlan_inshash() */
1342 	ifv->ifv_pcp = 0;       /* Default: best effort delivery. */
1343 	vlan_tag_recalculate(ifv);
1344 	error = vlan_inshash(trunk, ifv);
1345 	if (error)
1346 		goto done;
1347 	ifv->ifv_proto = ETHERTYPE_VLAN;
1348 	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1349 	ifv->ifv_mintu = ETHERMIN;
1350 	ifv->ifv_pflags = 0;
1351 	ifv->ifv_capenable = -1;
1352 
1353 	/*
1354 	 * If the parent supports the VLAN_MTU capability,
1355 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1356 	 * use it.
1357 	 */
1358 	if (p->if_capenable & IFCAP_VLAN_MTU) {
1359 		/*
1360 		 * No need to fudge the MTU since the parent can
1361 		 * handle extended frames.
1362 		 */
1363 		ifv->ifv_mtufudge = 0;
1364 	} else {
1365 		/*
1366 		 * Fudge the MTU by the encapsulation size.  This
1367 		 * makes us incompatible with strictly compliant
1368 		 * 802.1Q implementations, but allows us to use
1369 		 * the feature with other NetBSD implementations,
1370 		 * which might still be useful.
1371 		 */
1372 		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1373 	}
1374 
1375 	ifv->ifv_trunk = trunk;
1376 	ifp = ifv->ifv_ifp;
1377 	/*
1378 	 * Initialize fields from our parent.  This duplicates some
1379 	 * work with ether_ifattach() but allows for non-ethernet
1380 	 * interfaces to also work.
1381 	 */
1382 	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1383 	ifp->if_baudrate = p->if_baudrate;
1384 	ifp->if_output = p->if_output;
1385 	ifp->if_input = p->if_input;
1386 	ifp->if_resolvemulti = p->if_resolvemulti;
1387 	ifp->if_addrlen = p->if_addrlen;
1388 	ifp->if_broadcastaddr = p->if_broadcastaddr;
1389 	ifp->if_pcp = ifv->ifv_pcp;
1390 
1391 	/*
1392 	 * Copy only a selected subset of flags from the parent.
1393 	 * Other flags are none of our business.
1394 	 */
1395 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1396 	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1397 	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1398 #undef VLAN_COPY_FLAGS
1399 
1400 	ifp->if_link_state = p->if_link_state;
1401 
1402 	NET_EPOCH_ENTER(et);
1403 	vlan_capabilities(ifv);
1404 	NET_EPOCH_EXIT(et);
1405 
1406 	/*
1407 	 * Set up our interface address to reflect the underlying
1408 	 * physical interface's.
1409 	 */
1410 	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1411 	((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1412 	    p->if_addrlen;
1413 
1414 	TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
1415 
1416 	/* We are ready for operation now. */
1417 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1418 
1419 	/* Update flags on the parent, if necessary. */
1420 	vlan_setflags(ifp, 1);
1421 
1422 	/*
1423 	 * Configure multicast addresses that may already be
1424 	 * joined on the vlan device.
1425 	 */
1426 	(void)vlan_setmulti(ifp);
1427 
1428 done:
1429 	if (error == 0)
1430 		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1431 	VLAN_XUNLOCK();
1432 
1433 	return (error);
1434 }
1435 
1436 static void
1437 vlan_unconfig(struct ifnet *ifp)
1438 {
1439 
1440 	VLAN_XLOCK();
1441 	vlan_unconfig_locked(ifp, 0);
1442 	VLAN_XUNLOCK();
1443 }
1444 
1445 static void
1446 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1447 {
1448 	struct ifvlantrunk *trunk;
1449 	struct vlan_mc_entry *mc;
1450 	struct ifvlan *ifv;
1451 	struct ifnet  *parent;
1452 	int error;
1453 
1454 	VLAN_XLOCK_ASSERT();
1455 
1456 	ifv = ifp->if_softc;
1457 	trunk = ifv->ifv_trunk;
1458 	parent = NULL;
1459 
1460 	if (trunk != NULL) {
1461 		parent = trunk->parent;
1462 
1463 		/*
1464 		 * Since the interface is being unconfigured, we need to
1465 		 * empty the list of multicast groups that we may have joined
1466 		 * while we were alive from the parent's list.
1467 		 */
1468 		while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1469 			/*
1470 			 * If the parent interface is being detached,
1471 			 * all its multicast addresses have already
1472 			 * been removed.  Warn about errors if
1473 			 * if_delmulti() does fail, but don't abort as
1474 			 * all callers expect vlan destruction to
1475 			 * succeed.
1476 			 */
1477 			if (!departing) {
1478 				error = if_delmulti(parent,
1479 				    (struct sockaddr *)&mc->mc_addr);
1480 				if (error)
1481 					if_printf(ifp,
1482 		    "Failed to delete multicast address from parent: %d\n",
1483 					    error);
1484 			}
1485 			CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1486 			epoch_call(net_epoch_preempt, &mc->mc_epoch_ctx, vlan_mc_free);
1487 		}
1488 
1489 		vlan_setflags(ifp, 0); /* clear special flags on parent */
1490 
1491 		vlan_remhash(trunk, ifv);
1492 		ifv->ifv_trunk = NULL;
1493 
1494 		/*
1495 		 * Check if we were the last.
1496 		 */
1497 		if (trunk->refcnt == 0) {
1498 			parent->if_vlantrunk = NULL;
1499 			NET_EPOCH_WAIT();
1500 			trunk_destroy(trunk);
1501 		}
1502 	}
1503 
1504 	/* Disconnect from parent. */
1505 	if (ifv->ifv_pflags)
1506 		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1507 	ifp->if_mtu = ETHERMTU;
1508 	ifp->if_link_state = LINK_STATE_UNKNOWN;
1509 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1510 
1511 	/*
1512 	 * Only dispatch an event if vlan was
1513 	 * attached, otherwise there is nothing
1514 	 * to cleanup anyway.
1515 	 */
1516 	if (parent != NULL)
1517 		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1518 }
1519 
1520 /* Handle a reference counted flag that should be set on the parent as well */
1521 static int
1522 vlan_setflag(struct ifnet *ifp, int flag, int status,
1523 	     int (*func)(struct ifnet *, int))
1524 {
1525 	struct ifvlan *ifv;
1526 	int error;
1527 
1528 	VLAN_SXLOCK_ASSERT();
1529 
1530 	ifv = ifp->if_softc;
1531 	status = status ? (ifp->if_flags & flag) : 0;
1532 	/* Now "status" contains the flag value or 0 */
1533 
1534 	/*
1535 	 * See if recorded parent's status is different from what
1536 	 * we want it to be.  If it is, flip it.  We record parent's
1537 	 * status in ifv_pflags so that we won't clear parent's flag
1538 	 * we haven't set.  In fact, we don't clear or set parent's
1539 	 * flags directly, but get or release references to them.
1540 	 * That's why we can be sure that recorded flags still are
1541 	 * in accord with actual parent's flags.
1542 	 */
1543 	if (status != (ifv->ifv_pflags & flag)) {
1544 		error = (*func)(PARENT(ifv), status);
1545 		if (error)
1546 			return (error);
1547 		ifv->ifv_pflags &= ~flag;
1548 		ifv->ifv_pflags |= status;
1549 	}
1550 	return (0);
1551 }
1552 
1553 /*
1554  * Handle IFF_* flags that require certain changes on the parent:
1555  * if "status" is true, update parent's flags respective to our if_flags;
1556  * if "status" is false, forcedly clear the flags set on parent.
1557  */
1558 static int
1559 vlan_setflags(struct ifnet *ifp, int status)
1560 {
1561 	int error, i;
1562 
1563 	for (i = 0; vlan_pflags[i].flag; i++) {
1564 		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1565 				     status, vlan_pflags[i].func);
1566 		if (error)
1567 			return (error);
1568 	}
1569 	return (0);
1570 }
1571 
1572 /* Inform all vlans that their parent has changed link state */
1573 static void
1574 vlan_link_state(struct ifnet *ifp)
1575 {
1576 	struct epoch_tracker et;
1577 	struct ifvlantrunk *trunk;
1578 	struct ifvlan *ifv;
1579 
1580 	/* Called from a taskqueue_swi task, so we cannot sleep. */
1581 	NET_EPOCH_ENTER(et);
1582 	trunk = ifp->if_vlantrunk;
1583 	if (trunk == NULL) {
1584 		NET_EPOCH_EXIT(et);
1585 		return;
1586 	}
1587 
1588 	TRUNK_WLOCK(trunk);
1589 	VLAN_FOREACH(ifv, trunk) {
1590 		ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1591 		if_link_state_change(ifv->ifv_ifp,
1592 		    trunk->parent->if_link_state);
1593 	}
1594 	TRUNK_WUNLOCK(trunk);
1595 	NET_EPOCH_EXIT(et);
1596 }
1597 
1598 static void
1599 vlan_capabilities(struct ifvlan *ifv)
1600 {
1601 	struct ifnet *p;
1602 	struct ifnet *ifp;
1603 	struct ifnet_hw_tsomax hw_tsomax;
1604 	int cap = 0, ena = 0, mena;
1605 	u_long hwa = 0;
1606 
1607 	VLAN_SXLOCK_ASSERT();
1608 	NET_EPOCH_ASSERT();
1609 	p = PARENT(ifv);
1610 	ifp = ifv->ifv_ifp;
1611 
1612 	/* Mask parent interface enabled capabilities disabled by user. */
1613 	mena = p->if_capenable & ifv->ifv_capenable;
1614 
1615 	/*
1616 	 * If the parent interface can do checksum offloading
1617 	 * on VLANs, then propagate its hardware-assisted
1618 	 * checksumming flags. Also assert that checksum
1619 	 * offloading requires hardware VLAN tagging.
1620 	 */
1621 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1622 		cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1623 	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1624 	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1625 		ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1626 		if (ena & IFCAP_TXCSUM)
1627 			hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
1628 			    CSUM_UDP | CSUM_SCTP);
1629 		if (ena & IFCAP_TXCSUM_IPV6)
1630 			hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
1631 			    CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
1632 	}
1633 
1634 	/*
1635 	 * If the parent interface can do TSO on VLANs then
1636 	 * propagate the hardware-assisted flag. TSO on VLANs
1637 	 * does not necessarily require hardware VLAN tagging.
1638 	 */
1639 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1640 	if_hw_tsomax_common(p, &hw_tsomax);
1641 	if_hw_tsomax_update(ifp, &hw_tsomax);
1642 	if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1643 		cap |= p->if_capabilities & IFCAP_TSO;
1644 	if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1645 		ena |= mena & IFCAP_TSO;
1646 		if (ena & IFCAP_TSO)
1647 			hwa |= p->if_hwassist & CSUM_TSO;
1648 	}
1649 
1650 	/*
1651 	 * If the parent interface can do LRO and checksum offloading on
1652 	 * VLANs, then guess it may do LRO on VLANs.  False positive here
1653 	 * cost nothing, while false negative may lead to some confusions.
1654 	 */
1655 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1656 		cap |= p->if_capabilities & IFCAP_LRO;
1657 	if (p->if_capenable & IFCAP_VLAN_HWCSUM)
1658 		ena |= p->if_capenable & IFCAP_LRO;
1659 
1660 	/*
1661 	 * If the parent interface can offload TCP connections over VLANs then
1662 	 * propagate its TOE capability to the VLAN interface.
1663 	 *
1664 	 * All TOE drivers in the tree today can deal with VLANs.  If this
1665 	 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1666 	 * with its own bit.
1667 	 */
1668 #define	IFCAP_VLAN_TOE IFCAP_TOE
1669 	if (p->if_capabilities & IFCAP_VLAN_TOE)
1670 		cap |= p->if_capabilities & IFCAP_TOE;
1671 	if (p->if_capenable & IFCAP_VLAN_TOE) {
1672 		TOEDEV(ifp) = TOEDEV(p);
1673 		ena |= mena & IFCAP_TOE;
1674 	}
1675 
1676 	/*
1677 	 * If the parent interface supports dynamic link state, so does the
1678 	 * VLAN interface.
1679 	 */
1680 	cap |= (p->if_capabilities & IFCAP_LINKSTATE);
1681 	ena |= (mena & IFCAP_LINKSTATE);
1682 
1683 #ifdef RATELIMIT
1684 	/*
1685 	 * If the parent interface supports ratelimiting, so does the
1686 	 * VLAN interface.
1687 	 */
1688 	cap |= (p->if_capabilities & IFCAP_TXRTLMT);
1689 	ena |= (mena & IFCAP_TXRTLMT);
1690 #endif
1691 
1692 	ifp->if_capabilities = cap;
1693 	ifp->if_capenable = ena;
1694 	ifp->if_hwassist = hwa;
1695 }
1696 
1697 static void
1698 vlan_trunk_capabilities(struct ifnet *ifp)
1699 {
1700 	struct epoch_tracker et;
1701 	struct ifvlantrunk *trunk;
1702 	struct ifvlan *ifv;
1703 
1704 	VLAN_SLOCK();
1705 	trunk = ifp->if_vlantrunk;
1706 	if (trunk == NULL) {
1707 		VLAN_SUNLOCK();
1708 		return;
1709 	}
1710 	NET_EPOCH_ENTER(et);
1711 	VLAN_FOREACH(ifv, trunk) {
1712 		vlan_capabilities(ifv);
1713 	}
1714 	NET_EPOCH_EXIT(et);
1715 	VLAN_SUNLOCK();
1716 }
1717 
1718 static int
1719 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1720 {
1721 	struct ifnet *p;
1722 	struct ifreq *ifr;
1723 	struct ifaddr *ifa;
1724 	struct ifvlan *ifv;
1725 	struct ifvlantrunk *trunk;
1726 	struct vlanreq vlr;
1727 	int error = 0;
1728 
1729 	ifr = (struct ifreq *)data;
1730 	ifa = (struct ifaddr *) data;
1731 	ifv = ifp->if_softc;
1732 
1733 	switch (cmd) {
1734 	case SIOCSIFADDR:
1735 		ifp->if_flags |= IFF_UP;
1736 #ifdef INET
1737 		if (ifa->ifa_addr->sa_family == AF_INET)
1738 			arp_ifinit(ifp, ifa);
1739 #endif
1740 		break;
1741 	case SIOCGIFADDR:
1742 		bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
1743 		    ifp->if_addrlen);
1744 		break;
1745 	case SIOCGIFMEDIA:
1746 		VLAN_SLOCK();
1747 		if (TRUNK(ifv) != NULL) {
1748 			p = PARENT(ifv);
1749 			if_ref(p);
1750 			error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1751 			if_rele(p);
1752 			/* Limit the result to the parent's current config. */
1753 			if (error == 0) {
1754 				struct ifmediareq *ifmr;
1755 
1756 				ifmr = (struct ifmediareq *)data;
1757 				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1758 					ifmr->ifm_count = 1;
1759 					error = copyout(&ifmr->ifm_current,
1760 						ifmr->ifm_ulist,
1761 						sizeof(int));
1762 				}
1763 			}
1764 		} else {
1765 			error = EINVAL;
1766 		}
1767 		VLAN_SUNLOCK();
1768 		break;
1769 
1770 	case SIOCSIFMEDIA:
1771 		error = EINVAL;
1772 		break;
1773 
1774 	case SIOCSIFMTU:
1775 		/*
1776 		 * Set the interface MTU.
1777 		 */
1778 		VLAN_SLOCK();
1779 		trunk = TRUNK(ifv);
1780 		if (trunk != NULL) {
1781 			TRUNK_WLOCK(trunk);
1782 			if (ifr->ifr_mtu >
1783 			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1784 			    ifr->ifr_mtu <
1785 			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1786 				error = EINVAL;
1787 			else
1788 				ifp->if_mtu = ifr->ifr_mtu;
1789 			TRUNK_WUNLOCK(trunk);
1790 		} else
1791 			error = EINVAL;
1792 		VLAN_SUNLOCK();
1793 		break;
1794 
1795 	case SIOCSETVLAN:
1796 #ifdef VIMAGE
1797 		/*
1798 		 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1799 		 * interface to be delegated to a jail without allowing the
1800 		 * jail to change what underlying interface/VID it is
1801 		 * associated with.  We are not entirely convinced that this
1802 		 * is the right way to accomplish that policy goal.
1803 		 */
1804 		if (ifp->if_vnet != ifp->if_home_vnet) {
1805 			error = EPERM;
1806 			break;
1807 		}
1808 #endif
1809 		error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
1810 		if (error)
1811 			break;
1812 		if (vlr.vlr_parent[0] == '\0') {
1813 			vlan_unconfig(ifp);
1814 			break;
1815 		}
1816 		p = ifunit_ref(vlr.vlr_parent);
1817 		if (p == NULL) {
1818 			error = ENOENT;
1819 			break;
1820 		}
1821 		error = vlan_config(ifv, p, vlr.vlr_tag);
1822 		if_rele(p);
1823 		break;
1824 
1825 	case SIOCGETVLAN:
1826 #ifdef VIMAGE
1827 		if (ifp->if_vnet != ifp->if_home_vnet) {
1828 			error = EPERM;
1829 			break;
1830 		}
1831 #endif
1832 		bzero(&vlr, sizeof(vlr));
1833 		VLAN_SLOCK();
1834 		if (TRUNK(ifv) != NULL) {
1835 			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1836 			    sizeof(vlr.vlr_parent));
1837 			vlr.vlr_tag = ifv->ifv_vid;
1838 		}
1839 		VLAN_SUNLOCK();
1840 		error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
1841 		break;
1842 
1843 	case SIOCSIFFLAGS:
1844 		/*
1845 		 * We should propagate selected flags to the parent,
1846 		 * e.g., promiscuous mode.
1847 		 */
1848 		VLAN_XLOCK();
1849 		if (TRUNK(ifv) != NULL)
1850 			error = vlan_setflags(ifp, 1);
1851 		VLAN_XUNLOCK();
1852 		break;
1853 
1854 	case SIOCADDMULTI:
1855 	case SIOCDELMULTI:
1856 		/*
1857 		 * If we don't have a parent, just remember the membership for
1858 		 * when we do.
1859 		 *
1860 		 * XXX We need the rmlock here to avoid sleeping while
1861 		 * holding in6_multi_mtx.
1862 		 */
1863 		VLAN_XLOCK();
1864 		trunk = TRUNK(ifv);
1865 		if (trunk != NULL)
1866 			error = vlan_setmulti(ifp);
1867 		VLAN_XUNLOCK();
1868 
1869 		break;
1870 	case SIOCGVLANPCP:
1871 #ifdef VIMAGE
1872 		if (ifp->if_vnet != ifp->if_home_vnet) {
1873 			error = EPERM;
1874 			break;
1875 		}
1876 #endif
1877 		ifr->ifr_vlan_pcp = ifv->ifv_pcp;
1878 		break;
1879 
1880 	case SIOCSVLANPCP:
1881 #ifdef VIMAGE
1882 		if (ifp->if_vnet != ifp->if_home_vnet) {
1883 			error = EPERM;
1884 			break;
1885 		}
1886 #endif
1887 		error = priv_check(curthread, PRIV_NET_SETVLANPCP);
1888 		if (error)
1889 			break;
1890 		if (ifr->ifr_vlan_pcp > 7) {
1891 			error = EINVAL;
1892 			break;
1893 		}
1894 		ifv->ifv_pcp = ifr->ifr_vlan_pcp;
1895 		ifp->if_pcp = ifv->ifv_pcp;
1896 		vlan_tag_recalculate(ifv);
1897 		/* broadcast event about PCP change */
1898 		EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
1899 		break;
1900 
1901 	case SIOCSIFCAP:
1902 		VLAN_SLOCK();
1903 		ifv->ifv_capenable = ifr->ifr_reqcap;
1904 		trunk = TRUNK(ifv);
1905 		if (trunk != NULL) {
1906 			struct epoch_tracker et;
1907 
1908 			NET_EPOCH_ENTER(et);
1909 			vlan_capabilities(ifv);
1910 			NET_EPOCH_EXIT(et);
1911 		}
1912 		VLAN_SUNLOCK();
1913 		break;
1914 
1915 	default:
1916 		error = EINVAL;
1917 		break;
1918 	}
1919 
1920 	return (error);
1921 }
1922 
1923 #ifdef RATELIMIT
1924 static int
1925 vlan_snd_tag_alloc(struct ifnet *ifp,
1926     union if_snd_tag_alloc_params *params,
1927     struct m_snd_tag **ppmt)
1928 {
1929 
1930 	/* get trunk device */
1931 	ifp = vlan_trunkdev(ifp);
1932 	if (ifp == NULL || (ifp->if_capenable & IFCAP_TXRTLMT) == 0)
1933 		return (EOPNOTSUPP);
1934 	/* forward allocation request */
1935 	return (ifp->if_snd_tag_alloc(ifp, params, ppmt));
1936 }
1937 #endif
1938