xref: /freebsd/sys/net/if_vlan.c (revision 78adacd4eab39a3508bd8c65f0aba94fc6b907ce)
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_inet6.h"
50 #include "opt_kern_tls.h"
51 #include "opt_vlan.h"
52 #include "opt_ratelimit.h"
53 
54 #include <sys/param.h>
55 #include <sys/eventhandler.h>
56 #include <sys/kernel.h>
57 #include <sys/lock.h>
58 #include <sys/malloc.h>
59 #include <sys/mbuf.h>
60 #include <sys/module.h>
61 #include <sys/rmlock.h>
62 #include <sys/priv.h>
63 #include <sys/queue.h>
64 #include <sys/socket.h>
65 #include <sys/sockio.h>
66 #include <sys/sysctl.h>
67 #include <sys/systm.h>
68 #include <sys/sx.h>
69 #include <sys/taskqueue.h>
70 
71 #include <net/bpf.h>
72 #include <net/ethernet.h>
73 #include <net/if.h>
74 #include <net/if_var.h>
75 #include <net/if_clone.h>
76 #include <net/if_dl.h>
77 #include <net/if_types.h>
78 #include <net/if_vlan_var.h>
79 #include <net/route.h>
80 #include <net/vnet.h>
81 
82 #ifdef INET
83 #include <netinet/in.h>
84 #include <netinet/if_ether.h>
85 #endif
86 
87 #ifdef INET6
88 /*
89  * XXX: declare here to avoid to include many inet6 related files..
90  * should be more generalized?
91  */
92 extern void	nd6_setmtu(struct ifnet *);
93 #endif
94 
95 #define	VLAN_DEF_HWIDTH	4
96 #define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
97 
98 #define	UP_AND_RUNNING(ifp) \
99     ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
100 
101 CK_SLIST_HEAD(ifvlanhead, ifvlan);
102 
103 struct ifvlantrunk {
104 	struct	ifnet   *parent;	/* parent interface of this trunk */
105 	struct	mtx	lock;
106 #ifdef VLAN_ARRAY
107 #define	VLAN_ARRAY_SIZE	(EVL_VLID_MASK + 1)
108 	struct	ifvlan	*vlans[VLAN_ARRAY_SIZE]; /* static table */
109 #else
110 	struct	ifvlanhead *hash;	/* dynamic hash-list table */
111 	uint16_t	hmask;
112 	uint16_t	hwidth;
113 #endif
114 	int		refcnt;
115 };
116 
117 #if defined(KERN_TLS) || defined(RATELIMIT)
118 struct vlan_snd_tag {
119 	struct m_snd_tag com;
120 	struct m_snd_tag *tag;
121 };
122 
123 static inline struct vlan_snd_tag *
124 mst_to_vst(struct m_snd_tag *mst)
125 {
126 
127 	return (__containerof(mst, struct vlan_snd_tag, com));
128 }
129 #endif
130 
131 /*
132  * This macro provides a facility to iterate over every vlan on a trunk with
133  * the assumption that none will be added/removed during iteration.
134  */
135 #ifdef VLAN_ARRAY
136 #define VLAN_FOREACH(_ifv, _trunk) \
137 	size_t _i; \
138 	for (_i = 0; _i < VLAN_ARRAY_SIZE; _i++) \
139 		if (((_ifv) = (_trunk)->vlans[_i]) != NULL)
140 #else /* VLAN_ARRAY */
141 #define VLAN_FOREACH(_ifv, _trunk) \
142 	struct ifvlan *_next; \
143 	size_t _i; \
144 	for (_i = 0; _i < (1 << (_trunk)->hwidth); _i++) \
145 		CK_SLIST_FOREACH_SAFE((_ifv), &(_trunk)->hash[_i], ifv_list, _next)
146 #endif /* VLAN_ARRAY */
147 
148 /*
149  * This macro provides a facility to iterate over every vlan on a trunk while
150  * also modifying the number of vlans on the trunk. The iteration continues
151  * until some condition is met or there are no more vlans on the trunk.
152  */
153 #ifdef VLAN_ARRAY
154 /* The VLAN_ARRAY case is simple -- just a for loop using the condition. */
155 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
156 	size_t _i; \
157 	for (_i = 0; !(_cond) && _i < VLAN_ARRAY_SIZE; _i++) \
158 		if (((_ifv) = (_trunk)->vlans[_i]))
159 #else /* VLAN_ARRAY */
160 /*
161  * The hash table case is more complicated. We allow for the hash table to be
162  * modified (i.e. vlans removed) while we are iterating over it. To allow for
163  * this we must restart the iteration every time we "touch" something during
164  * the iteration, since removal will resize the hash table and invalidate our
165  * current position. If acting on the touched element causes the trunk to be
166  * emptied, then iteration also stops.
167  */
168 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
169 	size_t _i; \
170 	bool _touch = false; \
171 	for (_i = 0; \
172 	    !(_cond) && _i < (1 << (_trunk)->hwidth); \
173 	    _i = (_touch && ((_trunk) != NULL) ? 0 : _i + 1), _touch = false) \
174 		if (((_ifv) = CK_SLIST_FIRST(&(_trunk)->hash[_i])) != NULL && \
175 		    (_touch = true))
176 #endif /* VLAN_ARRAY */
177 
178 struct vlan_mc_entry {
179 	struct sockaddr_dl		mc_addr;
180 	CK_SLIST_ENTRY(vlan_mc_entry)	mc_entries;
181 	struct epoch_context		mc_epoch_ctx;
182 };
183 
184 struct ifvlan {
185 	struct	ifvlantrunk *ifv_trunk;
186 	struct	ifnet *ifv_ifp;
187 #define	TRUNK(ifv)	((ifv)->ifv_trunk)
188 #define	PARENT(ifv)	(TRUNK(ifv)->parent)
189 	void	*ifv_cookie;
190 	int	ifv_pflags;	/* special flags we have set on parent */
191 	int	ifv_capenable;
192 	int	ifv_encaplen;	/* encapsulation length */
193 	int	ifv_mtufudge;	/* MTU fudged by this much */
194 	int	ifv_mintu;	/* min transmission unit */
195 	struct  ether_8021q_tag ifv_qtag;
196 #define ifv_proto	ifv_qtag.proto
197 #define ifv_vid		ifv_qtag.vid
198 #define ifv_pcp		ifv_qtag.pcp
199 	struct task lladdr_task;
200 	CK_SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
201 #ifndef VLAN_ARRAY
202 	CK_SLIST_ENTRY(ifvlan) ifv_list;
203 #endif
204 };
205 
206 /* Special flags we should propagate to parent. */
207 static struct {
208 	int flag;
209 	int (*func)(struct ifnet *, int);
210 } vlan_pflags[] = {
211 	{IFF_PROMISC, ifpromisc},
212 	{IFF_ALLMULTI, if_allmulti},
213 	{0, NULL}
214 };
215 
216 extern int vlan_mtag_pcp;
217 
218 static const char vlanname[] = "vlan";
219 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
220 
221 static eventhandler_tag ifdetach_tag;
222 static eventhandler_tag iflladdr_tag;
223 
224 /*
225  * if_vlan uses two module-level synchronizations primitives to allow concurrent
226  * modification of vlan interfaces and (mostly) allow for vlans to be destroyed
227  * while they are being used for tx/rx. To accomplish this in a way that has
228  * acceptable performance and cooperation with other parts of the network stack
229  * there is a non-sleepable epoch(9) and an sx(9).
230  *
231  * The performance-sensitive paths that warrant using the epoch(9) are
232  * vlan_transmit and vlan_input. Both have to check for the vlan interface's
233  * existence using if_vlantrunk, and being in the network tx/rx paths the use
234  * of an epoch(9) gives a measureable improvement in performance.
235  *
236  * The reason for having an sx(9) is mostly because there are still areas that
237  * must be sleepable and also have safe concurrent access to a vlan interface.
238  * Since the sx(9) exists, it is used by default in most paths unless sleeping
239  * is not permitted, or if it is not clear whether sleeping is permitted.
240  *
241  */
242 #define _VLAN_SX_ID ifv_sx
243 
244 static struct sx _VLAN_SX_ID;
245 
246 #define VLAN_LOCKING_INIT() \
247 	sx_init_flags(&_VLAN_SX_ID, "vlan_sx", SX_RECURSE)
248 
249 #define VLAN_LOCKING_DESTROY() \
250 	sx_destroy(&_VLAN_SX_ID)
251 
252 #define	VLAN_SLOCK()			sx_slock(&_VLAN_SX_ID)
253 #define	VLAN_SUNLOCK()			sx_sunlock(&_VLAN_SX_ID)
254 #define	VLAN_XLOCK()			sx_xlock(&_VLAN_SX_ID)
255 #define	VLAN_XUNLOCK()			sx_xunlock(&_VLAN_SX_ID)
256 #define	VLAN_SLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_SLOCKED)
257 #define	VLAN_XLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_XLOCKED)
258 #define	VLAN_SXLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_LOCKED)
259 
260 /*
261  * We also have a per-trunk mutex that should be acquired when changing
262  * its state.
263  */
264 #define	TRUNK_LOCK_INIT(trunk)		mtx_init(&(trunk)->lock, vlanname, NULL, MTX_DEF)
265 #define	TRUNK_LOCK_DESTROY(trunk)	mtx_destroy(&(trunk)->lock)
266 #define	TRUNK_WLOCK(trunk)		mtx_lock(&(trunk)->lock)
267 #define	TRUNK_WUNLOCK(trunk)		mtx_unlock(&(trunk)->lock)
268 #define	TRUNK_WLOCK_ASSERT(trunk)	mtx_assert(&(trunk)->lock, MA_OWNED);
269 
270 /*
271  * The VLAN_ARRAY substitutes the dynamic hash with a static array
272  * with 4096 entries. In theory this can give a boost in processing,
273  * however in practice it does not. Probably this is because the array
274  * is too big to fit into CPU cache.
275  */
276 #ifndef VLAN_ARRAY
277 static	void vlan_inithash(struct ifvlantrunk *trunk);
278 static	void vlan_freehash(struct ifvlantrunk *trunk);
279 static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
280 static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
281 static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
282 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
283 	uint16_t vid);
284 #endif
285 static	void trunk_destroy(struct ifvlantrunk *trunk);
286 
287 static	void vlan_init(void *foo);
288 static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
289 static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
290 #if defined(KERN_TLS) || defined(RATELIMIT)
291 static	int vlan_snd_tag_alloc(struct ifnet *,
292     union if_snd_tag_alloc_params *, struct m_snd_tag **);
293 static	int vlan_snd_tag_modify(struct m_snd_tag *,
294     union if_snd_tag_modify_params *);
295 static	int vlan_snd_tag_query(struct m_snd_tag *,
296     union if_snd_tag_query_params *);
297 static	void vlan_snd_tag_free(struct m_snd_tag *);
298 #endif
299 static	void vlan_qflush(struct ifnet *ifp);
300 static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
301     int (*func)(struct ifnet *, int));
302 static	int vlan_setflags(struct ifnet *ifp, int status);
303 static	int vlan_setmulti(struct ifnet *ifp);
304 static	int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
305 static	int vlan_output(struct ifnet *ifp, struct mbuf *m,
306     const struct sockaddr *dst, struct route *ro);
307 static	void vlan_unconfig(struct ifnet *ifp);
308 static	void vlan_unconfig_locked(struct ifnet *ifp, int departing);
309 static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag,
310 	uint16_t proto);
311 static	void vlan_link_state(struct ifnet *ifp);
312 static	void vlan_capabilities(struct ifvlan *ifv);
313 static	void vlan_trunk_capabilities(struct ifnet *ifp);
314 
315 static	struct ifnet *vlan_clone_match_ethervid(const char *, int *);
316 static	int vlan_clone_match(struct if_clone *, const char *);
317 static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
318 static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
319 
320 static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
321 static  void vlan_iflladdr(void *arg, struct ifnet *ifp);
322 
323 static  void vlan_lladdr_fn(void *arg, int pending);
324 
325 static struct if_clone *vlan_cloner;
326 
327 #ifdef VIMAGE
328 VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner);
329 #define	V_vlan_cloner	VNET(vlan_cloner)
330 #endif
331 
332 static void
333 vlan_mc_free(struct epoch_context *ctx)
334 {
335 	struct vlan_mc_entry *mc = __containerof(ctx, struct vlan_mc_entry, mc_epoch_ctx);
336 	free(mc, M_VLAN);
337 }
338 
339 #ifndef VLAN_ARRAY
340 #define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
341 
342 static void
343 vlan_inithash(struct ifvlantrunk *trunk)
344 {
345 	int i, n;
346 
347 	/*
348 	 * The trunk must not be locked here since we call malloc(M_WAITOK).
349 	 * It is OK in case this function is called before the trunk struct
350 	 * gets hooked up and becomes visible from other threads.
351 	 */
352 
353 	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
354 	    ("%s: hash already initialized", __func__));
355 
356 	trunk->hwidth = VLAN_DEF_HWIDTH;
357 	n = 1 << trunk->hwidth;
358 	trunk->hmask = n - 1;
359 	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
360 	for (i = 0; i < n; i++)
361 		CK_SLIST_INIT(&trunk->hash[i]);
362 }
363 
364 static void
365 vlan_freehash(struct ifvlantrunk *trunk)
366 {
367 #ifdef INVARIANTS
368 	int i;
369 
370 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
371 	for (i = 0; i < (1 << trunk->hwidth); i++)
372 		KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]),
373 		    ("%s: hash table not empty", __func__));
374 #endif
375 	free(trunk->hash, M_VLAN);
376 	trunk->hash = NULL;
377 	trunk->hwidth = trunk->hmask = 0;
378 }
379 
380 static int
381 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
382 {
383 	int i, b;
384 	struct ifvlan *ifv2;
385 
386 	VLAN_XLOCK_ASSERT();
387 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
388 
389 	b = 1 << trunk->hwidth;
390 	i = HASH(ifv->ifv_vid, trunk->hmask);
391 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
392 		if (ifv->ifv_vid == ifv2->ifv_vid)
393 			return (EEXIST);
394 
395 	/*
396 	 * Grow the hash when the number of vlans exceeds half of the number of
397 	 * hash buckets squared. This will make the average linked-list length
398 	 * buckets/2.
399 	 */
400 	if (trunk->refcnt > (b * b) / 2) {
401 		vlan_growhash(trunk, 1);
402 		i = HASH(ifv->ifv_vid, trunk->hmask);
403 	}
404 	CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
405 	trunk->refcnt++;
406 
407 	return (0);
408 }
409 
410 static int
411 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
412 {
413 	int i, b;
414 	struct ifvlan *ifv2;
415 
416 	VLAN_XLOCK_ASSERT();
417 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
418 
419 	b = 1 << trunk->hwidth;
420 	i = HASH(ifv->ifv_vid, trunk->hmask);
421 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
422 		if (ifv2 == ifv) {
423 			trunk->refcnt--;
424 			CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list);
425 			if (trunk->refcnt < (b * b) / 2)
426 				vlan_growhash(trunk, -1);
427 			return (0);
428 		}
429 
430 	panic("%s: vlan not found\n", __func__);
431 	return (ENOENT); /*NOTREACHED*/
432 }
433 
434 /*
435  * Grow the hash larger or smaller if memory permits.
436  */
437 static void
438 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
439 {
440 	struct ifvlan *ifv;
441 	struct ifvlanhead *hash2;
442 	int hwidth2, i, j, n, n2;
443 
444 	VLAN_XLOCK_ASSERT();
445 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
446 
447 	if (howmuch == 0) {
448 		/* Harmless yet obvious coding error */
449 		printf("%s: howmuch is 0\n", __func__);
450 		return;
451 	}
452 
453 	hwidth2 = trunk->hwidth + howmuch;
454 	n = 1 << trunk->hwidth;
455 	n2 = 1 << hwidth2;
456 	/* Do not shrink the table below the default */
457 	if (hwidth2 < VLAN_DEF_HWIDTH)
458 		return;
459 
460 	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK);
461 	if (hash2 == NULL) {
462 		printf("%s: out of memory -- hash size not changed\n",
463 		    __func__);
464 		return;		/* We can live with the old hash table */
465 	}
466 	for (j = 0; j < n2; j++)
467 		CK_SLIST_INIT(&hash2[j]);
468 	for (i = 0; i < n; i++)
469 		while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) {
470 			CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list);
471 			j = HASH(ifv->ifv_vid, n2 - 1);
472 			CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
473 		}
474 	NET_EPOCH_WAIT();
475 	free(trunk->hash, M_VLAN);
476 	trunk->hash = hash2;
477 	trunk->hwidth = hwidth2;
478 	trunk->hmask = n2 - 1;
479 
480 	if (bootverbose)
481 		if_printf(trunk->parent,
482 		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
483 }
484 
485 static __inline struct ifvlan *
486 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
487 {
488 	struct ifvlan *ifv;
489 
490 	NET_EPOCH_ASSERT();
491 
492 	CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
493 		if (ifv->ifv_vid == vid)
494 			return (ifv);
495 	return (NULL);
496 }
497 
498 #if 0
499 /* Debugging code to view the hashtables. */
500 static void
501 vlan_dumphash(struct ifvlantrunk *trunk)
502 {
503 	int i;
504 	struct ifvlan *ifv;
505 
506 	for (i = 0; i < (1 << trunk->hwidth); i++) {
507 		printf("%d: ", i);
508 		CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
509 			printf("%s ", ifv->ifv_ifp->if_xname);
510 		printf("\n");
511 	}
512 }
513 #endif /* 0 */
514 #else
515 
516 static __inline struct ifvlan *
517 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
518 {
519 
520 	return trunk->vlans[vid];
521 }
522 
523 static __inline int
524 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
525 {
526 
527 	if (trunk->vlans[ifv->ifv_vid] != NULL)
528 		return EEXIST;
529 	trunk->vlans[ifv->ifv_vid] = ifv;
530 	trunk->refcnt++;
531 
532 	return (0);
533 }
534 
535 static __inline int
536 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
537 {
538 
539 	trunk->vlans[ifv->ifv_vid] = NULL;
540 	trunk->refcnt--;
541 
542 	return (0);
543 }
544 
545 static __inline void
546 vlan_freehash(struct ifvlantrunk *trunk)
547 {
548 }
549 
550 static __inline void
551 vlan_inithash(struct ifvlantrunk *trunk)
552 {
553 }
554 
555 #endif /* !VLAN_ARRAY */
556 
557 static void
558 trunk_destroy(struct ifvlantrunk *trunk)
559 {
560 	VLAN_XLOCK_ASSERT();
561 
562 	vlan_freehash(trunk);
563 	trunk->parent->if_vlantrunk = NULL;
564 	TRUNK_LOCK_DESTROY(trunk);
565 	if_rele(trunk->parent);
566 	free(trunk, M_VLAN);
567 }
568 
569 /*
570  * Program our multicast filter. What we're actually doing is
571  * programming the multicast filter of the parent. This has the
572  * side effect of causing the parent interface to receive multicast
573  * traffic that it doesn't really want, which ends up being discarded
574  * later by the upper protocol layers. Unfortunately, there's no way
575  * to avoid this: there really is only one physical interface.
576  */
577 static int
578 vlan_setmulti(struct ifnet *ifp)
579 {
580 	struct ifnet		*ifp_p;
581 	struct ifmultiaddr	*ifma;
582 	struct ifvlan		*sc;
583 	struct vlan_mc_entry	*mc;
584 	int			error;
585 
586 	VLAN_XLOCK_ASSERT();
587 
588 	/* Find the parent. */
589 	sc = ifp->if_softc;
590 	ifp_p = PARENT(sc);
591 
592 	CURVNET_SET_QUIET(ifp_p->if_vnet);
593 
594 	/* First, remove any existing filter entries. */
595 	while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
596 		CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
597 		(void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
598 		NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
599 	}
600 
601 	/* Now program new ones. */
602 	IF_ADDR_WLOCK(ifp);
603 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
604 		if (ifma->ifma_addr->sa_family != AF_LINK)
605 			continue;
606 		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
607 		if (mc == NULL) {
608 			IF_ADDR_WUNLOCK(ifp);
609 			return (ENOMEM);
610 		}
611 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
612 		mc->mc_addr.sdl_index = ifp_p->if_index;
613 		CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
614 	}
615 	IF_ADDR_WUNLOCK(ifp);
616 	CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
617 		error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
618 		    NULL);
619 		if (error)
620 			return (error);
621 	}
622 
623 	CURVNET_RESTORE();
624 	return (0);
625 }
626 
627 /*
628  * A handler for parent interface link layer address changes.
629  * If the parent interface link layer address is changed we
630  * should also change it on all children vlans.
631  */
632 static void
633 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
634 {
635 	struct epoch_tracker et;
636 	struct ifvlan *ifv;
637 	struct ifnet *ifv_ifp;
638 	struct ifvlantrunk *trunk;
639 	struct sockaddr_dl *sdl;
640 
641 	/* Need the epoch since this is run on taskqueue_swi. */
642 	NET_EPOCH_ENTER(et);
643 	trunk = ifp->if_vlantrunk;
644 	if (trunk == NULL) {
645 		NET_EPOCH_EXIT(et);
646 		return;
647 	}
648 
649 	/*
650 	 * OK, it's a trunk.  Loop over and change all vlan's lladdrs on it.
651 	 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
652 	 * ioctl calls on the parent garbling the lladdr of the child vlan.
653 	 */
654 	TRUNK_WLOCK(trunk);
655 	VLAN_FOREACH(ifv, trunk) {
656 		/*
657 		 * Copy new new lladdr into the ifv_ifp, enqueue a task
658 		 * to actually call if_setlladdr. if_setlladdr needs to
659 		 * be deferred to a taskqueue because it will call into
660 		 * the if_vlan ioctl path and try to acquire the global
661 		 * lock.
662 		 */
663 		ifv_ifp = ifv->ifv_ifp;
664 		bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
665 		    ifp->if_addrlen);
666 		sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
667 		sdl->sdl_alen = ifp->if_addrlen;
668 		taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
669 	}
670 	TRUNK_WUNLOCK(trunk);
671 	NET_EPOCH_EXIT(et);
672 }
673 
674 /*
675  * A handler for network interface departure events.
676  * Track departure of trunks here so that we don't access invalid
677  * pointers or whatever if a trunk is ripped from under us, e.g.,
678  * by ejecting its hot-plug card.  However, if an ifnet is simply
679  * being renamed, then there's no need to tear down the state.
680  */
681 static void
682 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
683 {
684 	struct ifvlan *ifv;
685 	struct ifvlantrunk *trunk;
686 
687 	/* If the ifnet is just being renamed, don't do anything. */
688 	if (ifp->if_flags & IFF_RENAMING)
689 		return;
690 	VLAN_XLOCK();
691 	trunk = ifp->if_vlantrunk;
692 	if (trunk == NULL) {
693 		VLAN_XUNLOCK();
694 		return;
695 	}
696 
697 	/*
698 	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
699 	 * Check trunk pointer after each vlan_unconfig() as it will
700 	 * free it and set to NULL after the last vlan was detached.
701 	 */
702 	VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
703 	    ifp->if_vlantrunk == NULL)
704 		vlan_unconfig_locked(ifv->ifv_ifp, 1);
705 
706 	/* Trunk should have been destroyed in vlan_unconfig(). */
707 	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
708 	VLAN_XUNLOCK();
709 }
710 
711 /*
712  * Return the trunk device for a virtual interface.
713  */
714 static struct ifnet  *
715 vlan_trunkdev(struct ifnet *ifp)
716 {
717 	struct ifvlan *ifv;
718 
719 	NET_EPOCH_ASSERT();
720 
721 	if (ifp->if_type != IFT_L2VLAN)
722 		return (NULL);
723 
724 	ifv = ifp->if_softc;
725 	ifp = NULL;
726 	if (ifv->ifv_trunk)
727 		ifp = PARENT(ifv);
728 	return (ifp);
729 }
730 
731 /*
732  * Return the 12-bit VLAN VID for this interface, for use by external
733  * components such as Infiniband.
734  *
735  * XXXRW: Note that the function name here is historical; it should be named
736  * vlan_vid().
737  */
738 static int
739 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
740 {
741 	struct ifvlan *ifv;
742 
743 	if (ifp->if_type != IFT_L2VLAN)
744 		return (EINVAL);
745 	ifv = ifp->if_softc;
746 	*vidp = ifv->ifv_vid;
747 	return (0);
748 }
749 
750 static int
751 vlan_pcp(struct ifnet *ifp, uint16_t *pcpp)
752 {
753 	struct ifvlan *ifv;
754 
755 	if (ifp->if_type != IFT_L2VLAN)
756 		return (EINVAL);
757 	ifv = ifp->if_softc;
758 	*pcpp = ifv->ifv_pcp;
759 	return (0);
760 }
761 
762 /*
763  * Return a driver specific cookie for this interface.  Synchronization
764  * with setcookie must be provided by the driver.
765  */
766 static void *
767 vlan_cookie(struct ifnet *ifp)
768 {
769 	struct ifvlan *ifv;
770 
771 	if (ifp->if_type != IFT_L2VLAN)
772 		return (NULL);
773 	ifv = ifp->if_softc;
774 	return (ifv->ifv_cookie);
775 }
776 
777 /*
778  * Store a cookie in our softc that drivers can use to store driver
779  * private per-instance data in.
780  */
781 static int
782 vlan_setcookie(struct ifnet *ifp, void *cookie)
783 {
784 	struct ifvlan *ifv;
785 
786 	if (ifp->if_type != IFT_L2VLAN)
787 		return (EINVAL);
788 	ifv = ifp->if_softc;
789 	ifv->ifv_cookie = cookie;
790 	return (0);
791 }
792 
793 /*
794  * Return the vlan device present at the specific VID.
795  */
796 static struct ifnet *
797 vlan_devat(struct ifnet *ifp, uint16_t vid)
798 {
799 	struct ifvlantrunk *trunk;
800 	struct ifvlan *ifv;
801 
802 	NET_EPOCH_ASSERT();
803 
804 	trunk = ifp->if_vlantrunk;
805 	if (trunk == NULL)
806 		return (NULL);
807 	ifp = NULL;
808 	ifv = vlan_gethash(trunk, vid);
809 	if (ifv)
810 		ifp = ifv->ifv_ifp;
811 	return (ifp);
812 }
813 
814 /*
815  * VLAN support can be loaded as a module.  The only place in the
816  * system that's intimately aware of this is ether_input.  We hook
817  * into this code through vlan_input_p which is defined there and
818  * set here.  No one else in the system should be aware of this so
819  * we use an explicit reference here.
820  */
821 extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
822 
823 /* For if_link_state_change() eyes only... */
824 extern	void (*vlan_link_state_p)(struct ifnet *);
825 
826 static int
827 vlan_modevent(module_t mod, int type, void *data)
828 {
829 
830 	switch (type) {
831 	case MOD_LOAD:
832 		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
833 		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
834 		if (ifdetach_tag == NULL)
835 			return (ENOMEM);
836 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
837 		    vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
838 		if (iflladdr_tag == NULL)
839 			return (ENOMEM);
840 		VLAN_LOCKING_INIT();
841 		vlan_input_p = vlan_input;
842 		vlan_link_state_p = vlan_link_state;
843 		vlan_trunk_cap_p = vlan_trunk_capabilities;
844 		vlan_trunkdev_p = vlan_trunkdev;
845 		vlan_cookie_p = vlan_cookie;
846 		vlan_setcookie_p = vlan_setcookie;
847 		vlan_tag_p = vlan_tag;
848 		vlan_pcp_p = vlan_pcp;
849 		vlan_devat_p = vlan_devat;
850 #ifndef VIMAGE
851 		vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
852 		    vlan_clone_create, vlan_clone_destroy);
853 #endif
854 		if (bootverbose)
855 			printf("vlan: initialized, using "
856 #ifdef VLAN_ARRAY
857 			       "full-size arrays"
858 #else
859 			       "hash tables with chaining"
860 #endif
861 
862 			       "\n");
863 		break;
864 	case MOD_UNLOAD:
865 #ifndef VIMAGE
866 		if_clone_detach(vlan_cloner);
867 #endif
868 		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
869 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
870 		vlan_input_p = NULL;
871 		vlan_link_state_p = NULL;
872 		vlan_trunk_cap_p = NULL;
873 		vlan_trunkdev_p = NULL;
874 		vlan_tag_p = NULL;
875 		vlan_cookie_p = NULL;
876 		vlan_setcookie_p = NULL;
877 		vlan_devat_p = NULL;
878 		VLAN_LOCKING_DESTROY();
879 		if (bootverbose)
880 			printf("vlan: unloaded\n");
881 		break;
882 	default:
883 		return (EOPNOTSUPP);
884 	}
885 	return (0);
886 }
887 
888 static moduledata_t vlan_mod = {
889 	"if_vlan",
890 	vlan_modevent,
891 	0
892 };
893 
894 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
895 MODULE_VERSION(if_vlan, 3);
896 
897 #ifdef VIMAGE
898 static void
899 vnet_vlan_init(const void *unused __unused)
900 {
901 
902 	vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
903 		    vlan_clone_create, vlan_clone_destroy);
904 	V_vlan_cloner = vlan_cloner;
905 }
906 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
907     vnet_vlan_init, NULL);
908 
909 static void
910 vnet_vlan_uninit(const void *unused __unused)
911 {
912 
913 	if_clone_detach(V_vlan_cloner);
914 }
915 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
916     vnet_vlan_uninit, NULL);
917 #endif
918 
919 /*
920  * Check for <etherif>.<vlan>[.<vlan> ...] style interface names.
921  */
922 static struct ifnet *
923 vlan_clone_match_ethervid(const char *name, int *vidp)
924 {
925 	char ifname[IFNAMSIZ];
926 	char *cp;
927 	struct ifnet *ifp;
928 	int vid;
929 
930 	strlcpy(ifname, name, IFNAMSIZ);
931 	if ((cp = strrchr(ifname, '.')) == NULL)
932 		return (NULL);
933 	*cp = '\0';
934 	if ((ifp = ifunit_ref(ifname)) == NULL)
935 		return (NULL);
936 	/* Parse VID. */
937 	if (*++cp == '\0') {
938 		if_rele(ifp);
939 		return (NULL);
940 	}
941 	vid = 0;
942 	for(; *cp >= '0' && *cp <= '9'; cp++)
943 		vid = (vid * 10) + (*cp - '0');
944 	if (*cp != '\0') {
945 		if_rele(ifp);
946 		return (NULL);
947 	}
948 	if (vidp != NULL)
949 		*vidp = vid;
950 
951 	return (ifp);
952 }
953 
954 static int
955 vlan_clone_match(struct if_clone *ifc, const char *name)
956 {
957 	struct ifnet *ifp;
958 	const char *cp;
959 
960 	ifp = vlan_clone_match_ethervid(name, NULL);
961 	if (ifp != NULL) {
962 		if_rele(ifp);
963 		return (1);
964 	}
965 
966 	if (strncmp(vlanname, name, strlen(vlanname)) != 0)
967 		return (0);
968 	for (cp = name + 4; *cp != '\0'; cp++) {
969 		if (*cp < '0' || *cp > '9')
970 			return (0);
971 	}
972 
973 	return (1);
974 }
975 
976 static int
977 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
978 {
979 	char *dp;
980 	int wildcard;
981 	int unit;
982 	int error;
983 	int vid;
984 	uint16_t proto;
985 	struct ifvlan *ifv;
986 	struct ifnet *ifp;
987 	struct ifnet *p;
988 	struct ifaddr *ifa;
989 	struct sockaddr_dl *sdl;
990 	struct vlanreq vlr;
991 	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
992 
993 	proto = ETHERTYPE_VLAN;
994 
995 	/*
996 	 * There are two ways to specify the cloned device:
997 	 * o pass a parameter block with the clone request.
998 	 * o specify no parameters and get an unattached device that
999 	 *   must be configured separately.
1000 	 * The first technique is preferred; the latter is supported
1001 	 * for backwards compatibility.
1002 	 *
1003 	 * XXXRW: Note historic use of the word "tag" here.  New ioctls may be
1004 	 * called for.
1005 	 */
1006 	if (params) {
1007 		error = copyin(params, &vlr, sizeof(vlr));
1008 		if (error)
1009 			return error;
1010 		p = ifunit_ref(vlr.vlr_parent);
1011 		if (p == NULL)
1012 			return (ENXIO);
1013 		error = ifc_name2unit(name, &unit);
1014 		if (error != 0) {
1015 			if_rele(p);
1016 			return (error);
1017 		}
1018 		vid = vlr.vlr_tag;
1019 		proto = vlr.vlr_proto;
1020 		wildcard = (unit < 0);
1021 	} else {
1022 		p = NULL;
1023 		error = ifc_name2unit(name, &unit);
1024 		if (error != 0)
1025 			return (error);
1026 
1027 		wildcard = (unit < 0);
1028 	}
1029 
1030 	error = ifc_alloc_unit(ifc, &unit);
1031 	if (error != 0) {
1032 		if (p != NULL)
1033 			if_rele(p);
1034 		return (error);
1035 	}
1036 
1037 	/* In the wildcard case, we need to update the name. */
1038 	if (wildcard) {
1039 		for (dp = name; *dp != '\0'; dp++);
1040 		if (snprintf(dp, len - (dp-name), "%d", unit) >
1041 		    len - (dp-name) - 1) {
1042 			panic("%s: interface name too long", __func__);
1043 		}
1044 	}
1045 
1046 	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
1047 	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
1048 	if (ifp == NULL) {
1049 		ifc_free_unit(ifc, unit);
1050 		free(ifv, M_VLAN);
1051 		if (p != NULL)
1052 			if_rele(p);
1053 		return (ENOSPC);
1054 	}
1055 	CK_SLIST_INIT(&ifv->vlan_mc_listhead);
1056 	ifp->if_softc = ifv;
1057 	/*
1058 	 * Set the name manually rather than using if_initname because
1059 	 * we don't conform to the default naming convention for interfaces.
1060 	 */
1061 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
1062 	ifp->if_dname = vlanname;
1063 	ifp->if_dunit = unit;
1064 
1065 	ifp->if_init = vlan_init;
1066 	ifp->if_transmit = vlan_transmit;
1067 	ifp->if_qflush = vlan_qflush;
1068 	ifp->if_ioctl = vlan_ioctl;
1069 #if defined(KERN_TLS) || defined(RATELIMIT)
1070 	ifp->if_snd_tag_alloc = vlan_snd_tag_alloc;
1071 	ifp->if_snd_tag_modify = vlan_snd_tag_modify;
1072 	ifp->if_snd_tag_query = vlan_snd_tag_query;
1073 	ifp->if_snd_tag_free = vlan_snd_tag_free;
1074 #endif
1075 	ifp->if_flags = VLAN_IFFLAGS;
1076 	ether_ifattach(ifp, eaddr);
1077 	/* Now undo some of the damage... */
1078 	ifp->if_baudrate = 0;
1079 	ifp->if_type = IFT_L2VLAN;
1080 	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
1081 	ifa = ifp->if_addr;
1082 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1083 	sdl->sdl_type = IFT_L2VLAN;
1084 
1085 	if (p != NULL) {
1086 		error = vlan_config(ifv, p, vid, proto);
1087 		if_rele(p);
1088 		if (error != 0) {
1089 			/*
1090 			 * Since we've partially failed, we need to back
1091 			 * out all the way, otherwise userland could get
1092 			 * confused.  Thus, we destroy the interface.
1093 			 */
1094 			ether_ifdetach(ifp);
1095 			vlan_unconfig(ifp);
1096 			if_free(ifp);
1097 			ifc_free_unit(ifc, unit);
1098 			free(ifv, M_VLAN);
1099 
1100 			return (error);
1101 		}
1102 	}
1103 
1104 	return (0);
1105 }
1106 
1107 static int
1108 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1109 {
1110 	struct ifvlan *ifv = ifp->if_softc;
1111 
1112 	if (ifp->if_vlantrunk)
1113 		return (EBUSY);
1114 
1115 	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
1116 	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
1117 	/*
1118 	 * We should have the only reference to the ifv now, so we can now
1119 	 * drain any remaining lladdr task before freeing the ifnet and the
1120 	 * ifvlan.
1121 	 */
1122 	taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
1123 	NET_EPOCH_WAIT();
1124 	if_free(ifp);
1125 	free(ifv, M_VLAN);
1126 	ifc_free_unit(ifc, ifp->if_dunit);
1127 
1128 	return (0);
1129 }
1130 
1131 /*
1132  * The ifp->if_init entry point for vlan(4) is a no-op.
1133  */
1134 static void
1135 vlan_init(void *foo __unused)
1136 {
1137 }
1138 
1139 /*
1140  * The if_transmit method for vlan(4) interface.
1141  */
1142 static int
1143 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1144 {
1145 	struct ifvlan *ifv;
1146 	struct ifnet *p;
1147 	int error, len, mcast;
1148 
1149 	NET_EPOCH_ASSERT();
1150 
1151 	ifv = ifp->if_softc;
1152 	if (TRUNK(ifv) == NULL) {
1153 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1154 		m_freem(m);
1155 		return (ENETDOWN);
1156 	}
1157 	p = PARENT(ifv);
1158 	len = m->m_pkthdr.len;
1159 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1160 
1161 	BPF_MTAP(ifp, m);
1162 
1163 #if defined(KERN_TLS) || defined(RATELIMIT)
1164 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
1165 		struct vlan_snd_tag *vst;
1166 		struct m_snd_tag *mst;
1167 
1168 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
1169 		mst = m->m_pkthdr.snd_tag;
1170 		vst = mst_to_vst(mst);
1171 		if (vst->tag->ifp != p) {
1172 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1173 			m_freem(m);
1174 			return (EAGAIN);
1175 		}
1176 
1177 		m->m_pkthdr.snd_tag = m_snd_tag_ref(vst->tag);
1178 		m_snd_tag_rele(mst);
1179 	}
1180 #endif
1181 
1182 	/*
1183 	 * Do not run parent's if_transmit() if the parent is not up,
1184 	 * or parent's driver will cause a system crash.
1185 	 */
1186 	if (!UP_AND_RUNNING(p)) {
1187 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1188 		m_freem(m);
1189 		return (ENETDOWN);
1190 	}
1191 
1192 	if (!ether_8021q_frame(&m, ifp, p, &ifv->ifv_qtag)) {
1193 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1194 		return (0);
1195 	}
1196 
1197 	/*
1198 	 * Send it, precisely as ether_output() would have.
1199 	 */
1200 	error = (p->if_transmit)(p, m);
1201 	if (error == 0) {
1202 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1203 		if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1204 		if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1205 	} else
1206 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1207 	return (error);
1208 }
1209 
1210 static int
1211 vlan_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1212     struct route *ro)
1213 {
1214 	struct ifvlan *ifv;
1215 	struct ifnet *p;
1216 
1217 	NET_EPOCH_ASSERT();
1218 
1219 	/*
1220 	 * Find the first non-VLAN parent interface.
1221 	 */
1222 	ifv = ifp->if_softc;
1223 	do {
1224 		if (TRUNK(ifv) == NULL) {
1225 			m_freem(m);
1226 			return (ENETDOWN);
1227 		}
1228 		p = PARENT(ifv);
1229 		ifv = p->if_softc;
1230 	} while (p->if_type == IFT_L2VLAN);
1231 
1232 	return p->if_output(ifp, m, dst, ro);
1233 }
1234 
1235 /*
1236  * The ifp->if_qflush entry point for vlan(4) is a no-op.
1237  */
1238 static void
1239 vlan_qflush(struct ifnet *ifp __unused)
1240 {
1241 }
1242 
1243 static void
1244 vlan_input(struct ifnet *ifp, struct mbuf *m)
1245 {
1246 	struct ifvlantrunk *trunk;
1247 	struct ifvlan *ifv;
1248 	struct m_tag *mtag;
1249 	uint16_t vid, tag;
1250 
1251 	NET_EPOCH_ASSERT();
1252 
1253 	trunk = ifp->if_vlantrunk;
1254 	if (trunk == NULL) {
1255 		m_freem(m);
1256 		return;
1257 	}
1258 
1259 	if (m->m_flags & M_VLANTAG) {
1260 		/*
1261 		 * Packet is tagged, but m contains a normal
1262 		 * Ethernet frame; the tag is stored out-of-band.
1263 		 */
1264 		tag = m->m_pkthdr.ether_vtag;
1265 		m->m_flags &= ~M_VLANTAG;
1266 	} else {
1267 		struct ether_vlan_header *evl;
1268 
1269 		/*
1270 		 * Packet is tagged in-band as specified by 802.1q.
1271 		 */
1272 		switch (ifp->if_type) {
1273 		case IFT_ETHER:
1274 			if (m->m_len < sizeof(*evl) &&
1275 			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
1276 				if_printf(ifp, "cannot pullup VLAN header\n");
1277 				return;
1278 			}
1279 			evl = mtod(m, struct ether_vlan_header *);
1280 			tag = ntohs(evl->evl_tag);
1281 
1282 			/*
1283 			 * Remove the 802.1q header by copying the Ethernet
1284 			 * addresses over it and adjusting the beginning of
1285 			 * the data in the mbuf.  The encapsulated Ethernet
1286 			 * type field is already in place.
1287 			 */
1288 			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1289 			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
1290 			m_adj(m, ETHER_VLAN_ENCAP_LEN);
1291 			break;
1292 
1293 		default:
1294 #ifdef INVARIANTS
1295 			panic("%s: %s has unsupported if_type %u",
1296 			      __func__, ifp->if_xname, ifp->if_type);
1297 #endif
1298 			if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1299 			m_freem(m);
1300 			return;
1301 		}
1302 	}
1303 
1304 	vid = EVL_VLANOFTAG(tag);
1305 
1306 	ifv = vlan_gethash(trunk, vid);
1307 	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1308 		if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1309 		m_freem(m);
1310 		return;
1311 	}
1312 
1313 	if (vlan_mtag_pcp) {
1314 		/*
1315 		 * While uncommon, it is possible that we will find a 802.1q
1316 		 * packet encapsulated inside another packet that also had an
1317 		 * 802.1q header.  For example, ethernet tunneled over IPSEC
1318 		 * arriving over ethernet.  In that case, we replace the
1319 		 * existing 802.1q PCP m_tag value.
1320 		 */
1321 		mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1322 		if (mtag == NULL) {
1323 			mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1324 			    sizeof(uint8_t), M_NOWAIT);
1325 			if (mtag == NULL) {
1326 				if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1327 				m_freem(m);
1328 				return;
1329 			}
1330 			m_tag_prepend(m, mtag);
1331 		}
1332 		*(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1333 	}
1334 
1335 	m->m_pkthdr.rcvif = ifv->ifv_ifp;
1336 	if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1337 
1338 	/* Pass it back through the parent's input routine. */
1339 	(*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
1340 }
1341 
1342 static void
1343 vlan_lladdr_fn(void *arg, int pending __unused)
1344 {
1345 	struct ifvlan *ifv;
1346 	struct ifnet *ifp;
1347 
1348 	ifv = (struct ifvlan *)arg;
1349 	ifp = ifv->ifv_ifp;
1350 
1351 	CURVNET_SET(ifp->if_vnet);
1352 
1353 	/* The ifv_ifp already has the lladdr copied in. */
1354 	if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1355 
1356 	CURVNET_RESTORE();
1357 }
1358 
1359 static int
1360 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid,
1361 	uint16_t proto)
1362 {
1363 	struct epoch_tracker et;
1364 	struct ifvlantrunk *trunk;
1365 	struct ifnet *ifp;
1366 	int error = 0;
1367 
1368 	/*
1369 	 * We can handle non-ethernet hardware types as long as
1370 	 * they handle the tagging and headers themselves.
1371 	 */
1372 	if (p->if_type != IFT_ETHER &&
1373 	    p->if_type != IFT_L2VLAN &&
1374 	    (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1375 		return (EPROTONOSUPPORT);
1376 	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1377 		return (EPROTONOSUPPORT);
1378 	/*
1379 	 * Don't let the caller set up a VLAN VID with
1380 	 * anything except VLID bits.
1381 	 * VID numbers 0x0 and 0xFFF are reserved.
1382 	 */
1383 	if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1384 		return (EINVAL);
1385 	if (ifv->ifv_trunk)
1386 		return (EBUSY);
1387 
1388 	VLAN_XLOCK();
1389 	if (p->if_vlantrunk == NULL) {
1390 		trunk = malloc(sizeof(struct ifvlantrunk),
1391 		    M_VLAN, M_WAITOK | M_ZERO);
1392 		vlan_inithash(trunk);
1393 		TRUNK_LOCK_INIT(trunk);
1394 		TRUNK_WLOCK(trunk);
1395 		p->if_vlantrunk = trunk;
1396 		trunk->parent = p;
1397 		if_ref(trunk->parent);
1398 		TRUNK_WUNLOCK(trunk);
1399 	} else {
1400 		trunk = p->if_vlantrunk;
1401 	}
1402 
1403 	ifv->ifv_vid = vid;	/* must set this before vlan_inshash() */
1404 	ifv->ifv_pcp = 0;       /* Default: best effort delivery. */
1405 	error = vlan_inshash(trunk, ifv);
1406 	if (error)
1407 		goto done;
1408 	ifv->ifv_proto = proto;
1409 	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1410 	ifv->ifv_mintu = ETHERMIN;
1411 	ifv->ifv_pflags = 0;
1412 	ifv->ifv_capenable = -1;
1413 
1414 	/*
1415 	 * If the parent supports the VLAN_MTU capability,
1416 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1417 	 * use it.
1418 	 */
1419 	if (p->if_capenable & IFCAP_VLAN_MTU) {
1420 		/*
1421 		 * No need to fudge the MTU since the parent can
1422 		 * handle extended frames.
1423 		 */
1424 		ifv->ifv_mtufudge = 0;
1425 	} else {
1426 		/*
1427 		 * Fudge the MTU by the encapsulation size.  This
1428 		 * makes us incompatible with strictly compliant
1429 		 * 802.1Q implementations, but allows us to use
1430 		 * the feature with other NetBSD implementations,
1431 		 * which might still be useful.
1432 		 */
1433 		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1434 	}
1435 
1436 	ifv->ifv_trunk = trunk;
1437 	ifp = ifv->ifv_ifp;
1438 	/*
1439 	 * Initialize fields from our parent.  This duplicates some
1440 	 * work with ether_ifattach() but allows for non-ethernet
1441 	 * interfaces to also work.
1442 	 */
1443 	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1444 	ifp->if_baudrate = p->if_baudrate;
1445 	ifp->if_input = p->if_input;
1446 	ifp->if_resolvemulti = p->if_resolvemulti;
1447 	ifp->if_addrlen = p->if_addrlen;
1448 	ifp->if_broadcastaddr = p->if_broadcastaddr;
1449 	ifp->if_pcp = ifv->ifv_pcp;
1450 
1451 	/*
1452 	 * We wrap the parent's if_output using vlan_output to ensure that it
1453 	 * can't become stale.
1454 	 */
1455 	ifp->if_output = vlan_output;
1456 
1457 	/*
1458 	 * Copy only a selected subset of flags from the parent.
1459 	 * Other flags are none of our business.
1460 	 */
1461 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1462 	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1463 	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1464 #undef VLAN_COPY_FLAGS
1465 
1466 	ifp->if_link_state = p->if_link_state;
1467 
1468 	NET_EPOCH_ENTER(et);
1469 	vlan_capabilities(ifv);
1470 	NET_EPOCH_EXIT(et);
1471 
1472 	/*
1473 	 * Set up our interface address to reflect the underlying
1474 	 * physical interface's.
1475 	 */
1476 	TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
1477 	((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1478 	    p->if_addrlen;
1479 
1480 	/*
1481 	 * Do not schedule link address update if it was the same
1482 	 * as previous parent's. This helps avoid updating for each
1483 	 * associated llentry.
1484 	 */
1485 	if (memcmp(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen) != 0) {
1486 		bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1487 		taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
1488 	}
1489 
1490 	/* We are ready for operation now. */
1491 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1492 
1493 	/* Update flags on the parent, if necessary. */
1494 	vlan_setflags(ifp, 1);
1495 
1496 	/*
1497 	 * Configure multicast addresses that may already be
1498 	 * joined on the vlan device.
1499 	 */
1500 	(void)vlan_setmulti(ifp);
1501 
1502 done:
1503 	if (error == 0)
1504 		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1505 	VLAN_XUNLOCK();
1506 
1507 	return (error);
1508 }
1509 
1510 static void
1511 vlan_unconfig(struct ifnet *ifp)
1512 {
1513 
1514 	VLAN_XLOCK();
1515 	vlan_unconfig_locked(ifp, 0);
1516 	VLAN_XUNLOCK();
1517 }
1518 
1519 static void
1520 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1521 {
1522 	struct ifvlantrunk *trunk;
1523 	struct vlan_mc_entry *mc;
1524 	struct ifvlan *ifv;
1525 	struct ifnet  *parent;
1526 	int error;
1527 
1528 	VLAN_XLOCK_ASSERT();
1529 
1530 	ifv = ifp->if_softc;
1531 	trunk = ifv->ifv_trunk;
1532 	parent = NULL;
1533 
1534 	if (trunk != NULL) {
1535 		parent = trunk->parent;
1536 
1537 		/*
1538 		 * Since the interface is being unconfigured, we need to
1539 		 * empty the list of multicast groups that we may have joined
1540 		 * while we were alive from the parent's list.
1541 		 */
1542 		while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1543 			/*
1544 			 * If the parent interface is being detached,
1545 			 * all its multicast addresses have already
1546 			 * been removed.  Warn about errors if
1547 			 * if_delmulti() does fail, but don't abort as
1548 			 * all callers expect vlan destruction to
1549 			 * succeed.
1550 			 */
1551 			if (!departing) {
1552 				error = if_delmulti(parent,
1553 				    (struct sockaddr *)&mc->mc_addr);
1554 				if (error)
1555 					if_printf(ifp,
1556 		    "Failed to delete multicast address from parent: %d\n",
1557 					    error);
1558 			}
1559 			CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1560 			NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
1561 		}
1562 
1563 		vlan_setflags(ifp, 0); /* clear special flags on parent */
1564 
1565 		vlan_remhash(trunk, ifv);
1566 		ifv->ifv_trunk = NULL;
1567 
1568 		/*
1569 		 * Check if we were the last.
1570 		 */
1571 		if (trunk->refcnt == 0) {
1572 			parent->if_vlantrunk = NULL;
1573 			NET_EPOCH_WAIT();
1574 			trunk_destroy(trunk);
1575 		}
1576 	}
1577 
1578 	/* Disconnect from parent. */
1579 	if (ifv->ifv_pflags)
1580 		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1581 	ifp->if_mtu = ETHERMTU;
1582 	ifp->if_link_state = LINK_STATE_UNKNOWN;
1583 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1584 
1585 	/*
1586 	 * Only dispatch an event if vlan was
1587 	 * attached, otherwise there is nothing
1588 	 * to cleanup anyway.
1589 	 */
1590 	if (parent != NULL)
1591 		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1592 }
1593 
1594 /* Handle a reference counted flag that should be set on the parent as well */
1595 static int
1596 vlan_setflag(struct ifnet *ifp, int flag, int status,
1597 	     int (*func)(struct ifnet *, int))
1598 {
1599 	struct ifvlan *ifv;
1600 	int error;
1601 
1602 	VLAN_SXLOCK_ASSERT();
1603 
1604 	ifv = ifp->if_softc;
1605 	status = status ? (ifp->if_flags & flag) : 0;
1606 	/* Now "status" contains the flag value or 0 */
1607 
1608 	/*
1609 	 * See if recorded parent's status is different from what
1610 	 * we want it to be.  If it is, flip it.  We record parent's
1611 	 * status in ifv_pflags so that we won't clear parent's flag
1612 	 * we haven't set.  In fact, we don't clear or set parent's
1613 	 * flags directly, but get or release references to them.
1614 	 * That's why we can be sure that recorded flags still are
1615 	 * in accord with actual parent's flags.
1616 	 */
1617 	if (status != (ifv->ifv_pflags & flag)) {
1618 		error = (*func)(PARENT(ifv), status);
1619 		if (error)
1620 			return (error);
1621 		ifv->ifv_pflags &= ~flag;
1622 		ifv->ifv_pflags |= status;
1623 	}
1624 	return (0);
1625 }
1626 
1627 /*
1628  * Handle IFF_* flags that require certain changes on the parent:
1629  * if "status" is true, update parent's flags respective to our if_flags;
1630  * if "status" is false, forcedly clear the flags set on parent.
1631  */
1632 static int
1633 vlan_setflags(struct ifnet *ifp, int status)
1634 {
1635 	int error, i;
1636 
1637 	for (i = 0; vlan_pflags[i].flag; i++) {
1638 		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1639 				     status, vlan_pflags[i].func);
1640 		if (error)
1641 			return (error);
1642 	}
1643 	return (0);
1644 }
1645 
1646 /* Inform all vlans that their parent has changed link state */
1647 static void
1648 vlan_link_state(struct ifnet *ifp)
1649 {
1650 	struct epoch_tracker et;
1651 	struct ifvlantrunk *trunk;
1652 	struct ifvlan *ifv;
1653 
1654 	NET_EPOCH_ENTER(et);
1655 	trunk = ifp->if_vlantrunk;
1656 	if (trunk == NULL) {
1657 		NET_EPOCH_EXIT(et);
1658 		return;
1659 	}
1660 
1661 	TRUNK_WLOCK(trunk);
1662 	VLAN_FOREACH(ifv, trunk) {
1663 		ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1664 		if_link_state_change(ifv->ifv_ifp,
1665 		    trunk->parent->if_link_state);
1666 	}
1667 	TRUNK_WUNLOCK(trunk);
1668 	NET_EPOCH_EXIT(et);
1669 }
1670 
1671 static void
1672 vlan_capabilities(struct ifvlan *ifv)
1673 {
1674 	struct ifnet *p;
1675 	struct ifnet *ifp;
1676 	struct ifnet_hw_tsomax hw_tsomax;
1677 	int cap = 0, ena = 0, mena;
1678 	u_long hwa = 0;
1679 
1680 	NET_EPOCH_ASSERT();
1681 	VLAN_SXLOCK_ASSERT();
1682 
1683 	p = PARENT(ifv);
1684 	ifp = ifv->ifv_ifp;
1685 
1686 	/* Mask parent interface enabled capabilities disabled by user. */
1687 	mena = p->if_capenable & ifv->ifv_capenable;
1688 
1689 	/*
1690 	 * If the parent interface can do checksum offloading
1691 	 * on VLANs, then propagate its hardware-assisted
1692 	 * checksumming flags. Also assert that checksum
1693 	 * offloading requires hardware VLAN tagging.
1694 	 */
1695 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1696 		cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1697 	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1698 	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1699 		ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1700 		if (ena & IFCAP_TXCSUM)
1701 			hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
1702 			    CSUM_UDP | CSUM_SCTP);
1703 		if (ena & IFCAP_TXCSUM_IPV6)
1704 			hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
1705 			    CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
1706 	}
1707 
1708 	/*
1709 	 * If the parent interface can do TSO on VLANs then
1710 	 * propagate the hardware-assisted flag. TSO on VLANs
1711 	 * does not necessarily require hardware VLAN tagging.
1712 	 */
1713 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1714 	if_hw_tsomax_common(p, &hw_tsomax);
1715 	if_hw_tsomax_update(ifp, &hw_tsomax);
1716 	if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1717 		cap |= p->if_capabilities & IFCAP_TSO;
1718 	if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1719 		ena |= mena & IFCAP_TSO;
1720 		if (ena & IFCAP_TSO)
1721 			hwa |= p->if_hwassist & CSUM_TSO;
1722 	}
1723 
1724 	/*
1725 	 * If the parent interface can do LRO and checksum offloading on
1726 	 * VLANs, then guess it may do LRO on VLANs.  False positive here
1727 	 * cost nothing, while false negative may lead to some confusions.
1728 	 */
1729 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1730 		cap |= p->if_capabilities & IFCAP_LRO;
1731 	if (p->if_capenable & IFCAP_VLAN_HWCSUM)
1732 		ena |= p->if_capenable & IFCAP_LRO;
1733 
1734 	/*
1735 	 * If the parent interface can offload TCP connections over VLANs then
1736 	 * propagate its TOE capability to the VLAN interface.
1737 	 *
1738 	 * All TOE drivers in the tree today can deal with VLANs.  If this
1739 	 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1740 	 * with its own bit.
1741 	 */
1742 #define	IFCAP_VLAN_TOE IFCAP_TOE
1743 	if (p->if_capabilities & IFCAP_VLAN_TOE)
1744 		cap |= p->if_capabilities & IFCAP_TOE;
1745 	if (p->if_capenable & IFCAP_VLAN_TOE) {
1746 		TOEDEV(ifp) = TOEDEV(p);
1747 		ena |= mena & IFCAP_TOE;
1748 	}
1749 
1750 	/*
1751 	 * If the parent interface supports dynamic link state, so does the
1752 	 * VLAN interface.
1753 	 */
1754 	cap |= (p->if_capabilities & IFCAP_LINKSTATE);
1755 	ena |= (mena & IFCAP_LINKSTATE);
1756 
1757 #ifdef RATELIMIT
1758 	/*
1759 	 * If the parent interface supports ratelimiting, so does the
1760 	 * VLAN interface.
1761 	 */
1762 	cap |= (p->if_capabilities & IFCAP_TXRTLMT);
1763 	ena |= (mena & IFCAP_TXRTLMT);
1764 #endif
1765 
1766 	/*
1767 	 * If the parent interface supports unmapped mbufs, so does
1768 	 * the VLAN interface.  Note that this should be fine even for
1769 	 * interfaces that don't support hardware tagging as headers
1770 	 * are prepended in normal mbufs to unmapped mbufs holding
1771 	 * payload data.
1772 	 */
1773 	cap |= (p->if_capabilities & IFCAP_NOMAP);
1774 	ena |= (mena & IFCAP_NOMAP);
1775 
1776 	/*
1777 	 * If the parent interface can offload encryption and segmentation
1778 	 * of TLS records over TCP, propagate it's capability to the VLAN
1779 	 * interface.
1780 	 *
1781 	 * All TLS drivers in the tree today can deal with VLANs.  If
1782 	 * this ever changes, then a new IFCAP_VLAN_TXTLS can be
1783 	 * defined.
1784 	 */
1785 	if (p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
1786 		cap |= p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
1787 	if (p->if_capenable & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
1788 		ena |= mena & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
1789 
1790 	ifp->if_capabilities = cap;
1791 	ifp->if_capenable = ena;
1792 	ifp->if_hwassist = hwa;
1793 }
1794 
1795 static void
1796 vlan_trunk_capabilities(struct ifnet *ifp)
1797 {
1798 	struct epoch_tracker et;
1799 	struct ifvlantrunk *trunk;
1800 	struct ifvlan *ifv;
1801 
1802 	VLAN_SLOCK();
1803 	trunk = ifp->if_vlantrunk;
1804 	if (trunk == NULL) {
1805 		VLAN_SUNLOCK();
1806 		return;
1807 	}
1808 	NET_EPOCH_ENTER(et);
1809 	VLAN_FOREACH(ifv, trunk)
1810 		vlan_capabilities(ifv);
1811 	NET_EPOCH_EXIT(et);
1812 	VLAN_SUNLOCK();
1813 }
1814 
1815 static int
1816 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1817 {
1818 	struct ifnet *p;
1819 	struct ifreq *ifr;
1820 	struct ifaddr *ifa;
1821 	struct ifvlan *ifv;
1822 	struct ifvlantrunk *trunk;
1823 	struct vlanreq vlr;
1824 	int error = 0, oldmtu;
1825 
1826 	ifr = (struct ifreq *)data;
1827 	ifa = (struct ifaddr *) data;
1828 	ifv = ifp->if_softc;
1829 
1830 	switch (cmd) {
1831 	case SIOCSIFADDR:
1832 		ifp->if_flags |= IFF_UP;
1833 #ifdef INET
1834 		if (ifa->ifa_addr->sa_family == AF_INET)
1835 			arp_ifinit(ifp, ifa);
1836 #endif
1837 		break;
1838 	case SIOCGIFADDR:
1839 		bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
1840 		    ifp->if_addrlen);
1841 		break;
1842 	case SIOCGIFMEDIA:
1843 		VLAN_SLOCK();
1844 		if (TRUNK(ifv) != NULL) {
1845 			p = PARENT(ifv);
1846 			if_ref(p);
1847 			error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1848 			if_rele(p);
1849 			/* Limit the result to the parent's current config. */
1850 			if (error == 0) {
1851 				struct ifmediareq *ifmr;
1852 
1853 				ifmr = (struct ifmediareq *)data;
1854 				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1855 					ifmr->ifm_count = 1;
1856 					error = copyout(&ifmr->ifm_current,
1857 						ifmr->ifm_ulist,
1858 						sizeof(int));
1859 				}
1860 			}
1861 		} else {
1862 			error = EINVAL;
1863 		}
1864 		VLAN_SUNLOCK();
1865 		break;
1866 
1867 	case SIOCSIFMEDIA:
1868 		error = EINVAL;
1869 		break;
1870 
1871 	case SIOCSIFMTU:
1872 		/*
1873 		 * Set the interface MTU.
1874 		 */
1875 		VLAN_SLOCK();
1876 		trunk = TRUNK(ifv);
1877 		if (trunk != NULL) {
1878 			TRUNK_WLOCK(trunk);
1879 			if (ifr->ifr_mtu >
1880 			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1881 			    ifr->ifr_mtu <
1882 			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1883 				error = EINVAL;
1884 			else
1885 				ifp->if_mtu = ifr->ifr_mtu;
1886 			TRUNK_WUNLOCK(trunk);
1887 		} else
1888 			error = EINVAL;
1889 		VLAN_SUNLOCK();
1890 		break;
1891 
1892 	case SIOCSETVLAN:
1893 #ifdef VIMAGE
1894 		/*
1895 		 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1896 		 * interface to be delegated to a jail without allowing the
1897 		 * jail to change what underlying interface/VID it is
1898 		 * associated with.  We are not entirely convinced that this
1899 		 * is the right way to accomplish that policy goal.
1900 		 */
1901 		if (ifp->if_vnet != ifp->if_home_vnet) {
1902 			error = EPERM;
1903 			break;
1904 		}
1905 #endif
1906 		error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
1907 		if (error)
1908 			break;
1909 		if (vlr.vlr_parent[0] == '\0') {
1910 			vlan_unconfig(ifp);
1911 			break;
1912 		}
1913 		p = ifunit_ref(vlr.vlr_parent);
1914 		if (p == NULL) {
1915 			error = ENOENT;
1916 			break;
1917 		}
1918 		oldmtu = ifp->if_mtu;
1919 		error = vlan_config(ifv, p, vlr.vlr_tag, vlr.vlr_proto);
1920 		if_rele(p);
1921 
1922 		/*
1923 		 * VLAN MTU may change during addition of the vlandev.
1924 		 * If it did, do network layer specific procedure.
1925 		 */
1926 		if (ifp->if_mtu != oldmtu) {
1927 #ifdef INET6
1928 			nd6_setmtu(ifp);
1929 #endif
1930 			rt_updatemtu(ifp);
1931 		}
1932 		break;
1933 
1934 	case SIOCGETVLAN:
1935 #ifdef VIMAGE
1936 		if (ifp->if_vnet != ifp->if_home_vnet) {
1937 			error = EPERM;
1938 			break;
1939 		}
1940 #endif
1941 		bzero(&vlr, sizeof(vlr));
1942 		VLAN_SLOCK();
1943 		if (TRUNK(ifv) != NULL) {
1944 			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1945 			    sizeof(vlr.vlr_parent));
1946 			vlr.vlr_tag = ifv->ifv_vid;
1947 			vlr.vlr_proto = ifv->ifv_proto;
1948 		}
1949 		VLAN_SUNLOCK();
1950 		error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
1951 		break;
1952 
1953 	case SIOCSIFFLAGS:
1954 		/*
1955 		 * We should propagate selected flags to the parent,
1956 		 * e.g., promiscuous mode.
1957 		 */
1958 		VLAN_XLOCK();
1959 		if (TRUNK(ifv) != NULL)
1960 			error = vlan_setflags(ifp, 1);
1961 		VLAN_XUNLOCK();
1962 		break;
1963 
1964 	case SIOCADDMULTI:
1965 	case SIOCDELMULTI:
1966 		/*
1967 		 * If we don't have a parent, just remember the membership for
1968 		 * when we do.
1969 		 *
1970 		 * XXX We need the rmlock here to avoid sleeping while
1971 		 * holding in6_multi_mtx.
1972 		 */
1973 		VLAN_XLOCK();
1974 		trunk = TRUNK(ifv);
1975 		if (trunk != NULL)
1976 			error = vlan_setmulti(ifp);
1977 		VLAN_XUNLOCK();
1978 
1979 		break;
1980 	case SIOCGVLANPCP:
1981 #ifdef VIMAGE
1982 		if (ifp->if_vnet != ifp->if_home_vnet) {
1983 			error = EPERM;
1984 			break;
1985 		}
1986 #endif
1987 		ifr->ifr_vlan_pcp = ifv->ifv_pcp;
1988 		break;
1989 
1990 	case SIOCSVLANPCP:
1991 #ifdef VIMAGE
1992 		if (ifp->if_vnet != ifp->if_home_vnet) {
1993 			error = EPERM;
1994 			break;
1995 		}
1996 #endif
1997 		error = priv_check(curthread, PRIV_NET_SETVLANPCP);
1998 		if (error)
1999 			break;
2000 		if (ifr->ifr_vlan_pcp > 7) {
2001 			error = EINVAL;
2002 			break;
2003 		}
2004 		ifv->ifv_pcp = ifr->ifr_vlan_pcp;
2005 		ifp->if_pcp = ifv->ifv_pcp;
2006 		/* broadcast event about PCP change */
2007 		EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
2008 		break;
2009 
2010 	case SIOCSIFCAP:
2011 		VLAN_SLOCK();
2012 		ifv->ifv_capenable = ifr->ifr_reqcap;
2013 		trunk = TRUNK(ifv);
2014 		if (trunk != NULL) {
2015 			struct epoch_tracker et;
2016 
2017 			NET_EPOCH_ENTER(et);
2018 			vlan_capabilities(ifv);
2019 			NET_EPOCH_EXIT(et);
2020 		}
2021 		VLAN_SUNLOCK();
2022 		break;
2023 
2024 	default:
2025 		error = EINVAL;
2026 		break;
2027 	}
2028 
2029 	return (error);
2030 }
2031 
2032 #if defined(KERN_TLS) || defined(RATELIMIT)
2033 static int
2034 vlan_snd_tag_alloc(struct ifnet *ifp,
2035     union if_snd_tag_alloc_params *params,
2036     struct m_snd_tag **ppmt)
2037 {
2038 	struct epoch_tracker et;
2039 	struct vlan_snd_tag *vst;
2040 	struct ifvlan *ifv;
2041 	struct ifnet *parent;
2042 	int error;
2043 
2044 	NET_EPOCH_ENTER(et);
2045 	ifv = ifp->if_softc;
2046 	if (ifv->ifv_trunk != NULL)
2047 		parent = PARENT(ifv);
2048 	else
2049 		parent = NULL;
2050 	if (parent == NULL) {
2051 		NET_EPOCH_EXIT(et);
2052 		return (EOPNOTSUPP);
2053 	}
2054 	if_ref(parent);
2055 	NET_EPOCH_EXIT(et);
2056 
2057 	vst = malloc(sizeof(*vst), M_VLAN, M_NOWAIT);
2058 	if (vst == NULL) {
2059 		if_rele(parent);
2060 		return (ENOMEM);
2061 	}
2062 
2063 	error = m_snd_tag_alloc(parent, params, &vst->tag);
2064 	if_rele(parent);
2065 	if (error) {
2066 		free(vst, M_VLAN);
2067 		return (error);
2068 	}
2069 
2070 	m_snd_tag_init(&vst->com, ifp, vst->tag->type);
2071 
2072 	*ppmt = &vst->com;
2073 	return (0);
2074 }
2075 
2076 static int
2077 vlan_snd_tag_modify(struct m_snd_tag *mst,
2078     union if_snd_tag_modify_params *params)
2079 {
2080 	struct vlan_snd_tag *vst;
2081 
2082 	vst = mst_to_vst(mst);
2083 	return (vst->tag->ifp->if_snd_tag_modify(vst->tag, params));
2084 }
2085 
2086 static int
2087 vlan_snd_tag_query(struct m_snd_tag *mst,
2088     union if_snd_tag_query_params *params)
2089 {
2090 	struct vlan_snd_tag *vst;
2091 
2092 	vst = mst_to_vst(mst);
2093 	return (vst->tag->ifp->if_snd_tag_query(vst->tag, params));
2094 }
2095 
2096 static void
2097 vlan_snd_tag_free(struct m_snd_tag *mst)
2098 {
2099 	struct vlan_snd_tag *vst;
2100 
2101 	vst = mst_to_vst(mst);
2102 	m_snd_tag_rele(vst->tag);
2103 	free(vst, M_VLAN);
2104 }
2105 #endif
2106