xref: /freebsd/sys/net/if_vlan.c (revision 8881d206f4e68b564c2c5f50fc717086fc3e827a)
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 VNET_DECLARE(int, vlan_mtag_pcp);
217 #define	V_vlan_mtag_pcp	VNET(vlan_mtag_pcp)
218 
219 static const char vlanname[] = "vlan";
220 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
221 
222 static eventhandler_tag ifdetach_tag;
223 static eventhandler_tag iflladdr_tag;
224 
225 /*
226  * if_vlan uses two module-level synchronizations primitives to allow concurrent
227  * modification of vlan interfaces and (mostly) allow for vlans to be destroyed
228  * while they are being used for tx/rx. To accomplish this in a way that has
229  * acceptable performance and cooperation with other parts of the network stack
230  * there is a non-sleepable epoch(9) and an sx(9).
231  *
232  * The performance-sensitive paths that warrant using the epoch(9) are
233  * vlan_transmit and vlan_input. Both have to check for the vlan interface's
234  * existence using if_vlantrunk, and being in the network tx/rx paths the use
235  * of an epoch(9) gives a measureable improvement in performance.
236  *
237  * The reason for having an sx(9) is mostly because there are still areas that
238  * must be sleepable and also have safe concurrent access to a vlan interface.
239  * Since the sx(9) exists, it is used by default in most paths unless sleeping
240  * is not permitted, or if it is not clear whether sleeping is permitted.
241  *
242  */
243 #define _VLAN_SX_ID ifv_sx
244 
245 static struct sx _VLAN_SX_ID;
246 
247 #define VLAN_LOCKING_INIT() \
248 	sx_init_flags(&_VLAN_SX_ID, "vlan_sx", SX_RECURSE)
249 
250 #define VLAN_LOCKING_DESTROY() \
251 	sx_destroy(&_VLAN_SX_ID)
252 
253 #define	VLAN_SLOCK()			sx_slock(&_VLAN_SX_ID)
254 #define	VLAN_SUNLOCK()			sx_sunlock(&_VLAN_SX_ID)
255 #define	VLAN_XLOCK()			sx_xlock(&_VLAN_SX_ID)
256 #define	VLAN_XUNLOCK()			sx_xunlock(&_VLAN_SX_ID)
257 #define	VLAN_SLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_SLOCKED)
258 #define	VLAN_XLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_XLOCKED)
259 #define	VLAN_SXLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_LOCKED)
260 
261 /*
262  * We also have a per-trunk mutex that should be acquired when changing
263  * its state.
264  */
265 #define	TRUNK_LOCK_INIT(trunk)		mtx_init(&(trunk)->lock, vlanname, NULL, MTX_DEF)
266 #define	TRUNK_LOCK_DESTROY(trunk)	mtx_destroy(&(trunk)->lock)
267 #define	TRUNK_WLOCK(trunk)		mtx_lock(&(trunk)->lock)
268 #define	TRUNK_WUNLOCK(trunk)		mtx_unlock(&(trunk)->lock)
269 #define	TRUNK_WLOCK_ASSERT(trunk)	mtx_assert(&(trunk)->lock, MA_OWNED);
270 
271 /*
272  * The VLAN_ARRAY substitutes the dynamic hash with a static array
273  * with 4096 entries. In theory this can give a boost in processing,
274  * however in practice it does not. Probably this is because the array
275  * is too big to fit into CPU cache.
276  */
277 #ifndef VLAN_ARRAY
278 static	void vlan_inithash(struct ifvlantrunk *trunk);
279 static	void vlan_freehash(struct ifvlantrunk *trunk);
280 static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
281 static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
282 static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
283 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
284 	uint16_t vid);
285 #endif
286 static	void trunk_destroy(struct ifvlantrunk *trunk);
287 
288 static	void vlan_init(void *foo);
289 static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
290 static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
291 #if defined(KERN_TLS) || defined(RATELIMIT)
292 static	int vlan_snd_tag_alloc(struct ifnet *,
293     union if_snd_tag_alloc_params *, struct m_snd_tag **);
294 static	int vlan_snd_tag_modify(struct m_snd_tag *,
295     union if_snd_tag_modify_params *);
296 static	int vlan_snd_tag_query(struct m_snd_tag *,
297     union if_snd_tag_query_params *);
298 static	void vlan_snd_tag_free(struct m_snd_tag *);
299 static struct m_snd_tag *vlan_next_snd_tag(struct m_snd_tag *);
300 static void vlan_ratelimit_query(struct ifnet *,
301     struct if_ratelimit_query_results *);
302 #endif
303 static	void vlan_qflush(struct ifnet *ifp);
304 static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
305     int (*func)(struct ifnet *, int));
306 static	int vlan_setflags(struct ifnet *ifp, int status);
307 static	int vlan_setmulti(struct ifnet *ifp);
308 static	int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
309 #ifdef ALTQ
310 static void vlan_altq_start(struct ifnet *ifp);
311 static	int vlan_altq_transmit(struct ifnet *ifp, struct mbuf *m);
312 #endif
313 static	int vlan_output(struct ifnet *ifp, struct mbuf *m,
314     const struct sockaddr *dst, struct route *ro);
315 static	void vlan_unconfig(struct ifnet *ifp);
316 static	void vlan_unconfig_locked(struct ifnet *ifp, int departing);
317 static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag,
318 	uint16_t proto);
319 static	void vlan_link_state(struct ifnet *ifp);
320 static	void vlan_capabilities(struct ifvlan *ifv);
321 static	void vlan_trunk_capabilities(struct ifnet *ifp);
322 
323 static	struct ifnet *vlan_clone_match_ethervid(const char *, int *);
324 static	int vlan_clone_match(struct if_clone *, const char *);
325 static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
326 static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
327 
328 static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
329 static  void vlan_iflladdr(void *arg, struct ifnet *ifp);
330 
331 static  void vlan_lladdr_fn(void *arg, int pending);
332 
333 static struct if_clone *vlan_cloner;
334 
335 #ifdef VIMAGE
336 VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner);
337 #define	V_vlan_cloner	VNET(vlan_cloner)
338 #endif
339 
340 #ifdef RATELIMIT
341 static const struct if_snd_tag_sw vlan_snd_tag_ul_sw = {
342 	.snd_tag_modify = vlan_snd_tag_modify,
343 	.snd_tag_query = vlan_snd_tag_query,
344 	.snd_tag_free = vlan_snd_tag_free,
345 	.next_snd_tag = vlan_next_snd_tag,
346 	.type = IF_SND_TAG_TYPE_UNLIMITED
347 };
348 
349 static const struct if_snd_tag_sw vlan_snd_tag_rl_sw = {
350 	.snd_tag_modify = vlan_snd_tag_modify,
351 	.snd_tag_query = vlan_snd_tag_query,
352 	.snd_tag_free = vlan_snd_tag_free,
353 	.next_snd_tag = vlan_next_snd_tag,
354 	.type = IF_SND_TAG_TYPE_RATE_LIMIT
355 };
356 #endif
357 
358 #ifdef KERN_TLS
359 static const struct if_snd_tag_sw vlan_snd_tag_tls_sw = {
360 	.snd_tag_modify = vlan_snd_tag_modify,
361 	.snd_tag_query = vlan_snd_tag_query,
362 	.snd_tag_free = vlan_snd_tag_free,
363 	.next_snd_tag = vlan_next_snd_tag,
364 	.type = IF_SND_TAG_TYPE_TLS
365 };
366 
367 #ifdef RATELIMIT
368 static const struct if_snd_tag_sw vlan_snd_tag_tls_rl_sw = {
369 	.snd_tag_modify = vlan_snd_tag_modify,
370 	.snd_tag_query = vlan_snd_tag_query,
371 	.snd_tag_free = vlan_snd_tag_free,
372 	.next_snd_tag = vlan_next_snd_tag,
373 	.type = IF_SND_TAG_TYPE_TLS_RATE_LIMIT
374 };
375 #endif
376 #endif
377 
378 static void
379 vlan_mc_free(struct epoch_context *ctx)
380 {
381 	struct vlan_mc_entry *mc = __containerof(ctx, struct vlan_mc_entry, mc_epoch_ctx);
382 	free(mc, M_VLAN);
383 }
384 
385 #ifndef VLAN_ARRAY
386 #define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
387 
388 static void
389 vlan_inithash(struct ifvlantrunk *trunk)
390 {
391 	int i, n;
392 
393 	/*
394 	 * The trunk must not be locked here since we call malloc(M_WAITOK).
395 	 * It is OK in case this function is called before the trunk struct
396 	 * gets hooked up and becomes visible from other threads.
397 	 */
398 
399 	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
400 	    ("%s: hash already initialized", __func__));
401 
402 	trunk->hwidth = VLAN_DEF_HWIDTH;
403 	n = 1 << trunk->hwidth;
404 	trunk->hmask = n - 1;
405 	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
406 	for (i = 0; i < n; i++)
407 		CK_SLIST_INIT(&trunk->hash[i]);
408 }
409 
410 static void
411 vlan_freehash(struct ifvlantrunk *trunk)
412 {
413 #ifdef INVARIANTS
414 	int i;
415 
416 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
417 	for (i = 0; i < (1 << trunk->hwidth); i++)
418 		KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]),
419 		    ("%s: hash table not empty", __func__));
420 #endif
421 	free(trunk->hash, M_VLAN);
422 	trunk->hash = NULL;
423 	trunk->hwidth = trunk->hmask = 0;
424 }
425 
426 static int
427 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
428 {
429 	int i, b;
430 	struct ifvlan *ifv2;
431 
432 	VLAN_XLOCK_ASSERT();
433 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
434 
435 	b = 1 << trunk->hwidth;
436 	i = HASH(ifv->ifv_vid, trunk->hmask);
437 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
438 		if (ifv->ifv_vid == ifv2->ifv_vid)
439 			return (EEXIST);
440 
441 	/*
442 	 * Grow the hash when the number of vlans exceeds half of the number of
443 	 * hash buckets squared. This will make the average linked-list length
444 	 * buckets/2.
445 	 */
446 	if (trunk->refcnt > (b * b) / 2) {
447 		vlan_growhash(trunk, 1);
448 		i = HASH(ifv->ifv_vid, trunk->hmask);
449 	}
450 	CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
451 	trunk->refcnt++;
452 
453 	return (0);
454 }
455 
456 static int
457 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
458 {
459 	int i, b;
460 	struct ifvlan *ifv2;
461 
462 	VLAN_XLOCK_ASSERT();
463 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
464 
465 	b = 1 << trunk->hwidth;
466 	i = HASH(ifv->ifv_vid, trunk->hmask);
467 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
468 		if (ifv2 == ifv) {
469 			trunk->refcnt--;
470 			CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list);
471 			if (trunk->refcnt < (b * b) / 2)
472 				vlan_growhash(trunk, -1);
473 			return (0);
474 		}
475 
476 	panic("%s: vlan not found\n", __func__);
477 	return (ENOENT); /*NOTREACHED*/
478 }
479 
480 /*
481  * Grow the hash larger or smaller if memory permits.
482  */
483 static void
484 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
485 {
486 	struct ifvlan *ifv;
487 	struct ifvlanhead *hash2;
488 	int hwidth2, i, j, n, n2;
489 
490 	VLAN_XLOCK_ASSERT();
491 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
492 
493 	if (howmuch == 0) {
494 		/* Harmless yet obvious coding error */
495 		printf("%s: howmuch is 0\n", __func__);
496 		return;
497 	}
498 
499 	hwidth2 = trunk->hwidth + howmuch;
500 	n = 1 << trunk->hwidth;
501 	n2 = 1 << hwidth2;
502 	/* Do not shrink the table below the default */
503 	if (hwidth2 < VLAN_DEF_HWIDTH)
504 		return;
505 
506 	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK);
507 	if (hash2 == NULL) {
508 		printf("%s: out of memory -- hash size not changed\n",
509 		    __func__);
510 		return;		/* We can live with the old hash table */
511 	}
512 	for (j = 0; j < n2; j++)
513 		CK_SLIST_INIT(&hash2[j]);
514 	for (i = 0; i < n; i++)
515 		while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) {
516 			CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list);
517 			j = HASH(ifv->ifv_vid, n2 - 1);
518 			CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
519 		}
520 	NET_EPOCH_WAIT();
521 	free(trunk->hash, M_VLAN);
522 	trunk->hash = hash2;
523 	trunk->hwidth = hwidth2;
524 	trunk->hmask = n2 - 1;
525 
526 	if (bootverbose)
527 		if_printf(trunk->parent,
528 		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
529 }
530 
531 static __inline struct ifvlan *
532 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
533 {
534 	struct ifvlan *ifv;
535 
536 	NET_EPOCH_ASSERT();
537 
538 	CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
539 		if (ifv->ifv_vid == vid)
540 			return (ifv);
541 	return (NULL);
542 }
543 
544 #if 0
545 /* Debugging code to view the hashtables. */
546 static void
547 vlan_dumphash(struct ifvlantrunk *trunk)
548 {
549 	int i;
550 	struct ifvlan *ifv;
551 
552 	for (i = 0; i < (1 << trunk->hwidth); i++) {
553 		printf("%d: ", i);
554 		CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
555 			printf("%s ", ifv->ifv_ifp->if_xname);
556 		printf("\n");
557 	}
558 }
559 #endif /* 0 */
560 #else
561 
562 static __inline struct ifvlan *
563 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
564 {
565 
566 	return trunk->vlans[vid];
567 }
568 
569 static __inline int
570 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
571 {
572 
573 	if (trunk->vlans[ifv->ifv_vid] != NULL)
574 		return EEXIST;
575 	trunk->vlans[ifv->ifv_vid] = ifv;
576 	trunk->refcnt++;
577 
578 	return (0);
579 }
580 
581 static __inline int
582 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
583 {
584 
585 	trunk->vlans[ifv->ifv_vid] = NULL;
586 	trunk->refcnt--;
587 
588 	return (0);
589 }
590 
591 static __inline void
592 vlan_freehash(struct ifvlantrunk *trunk)
593 {
594 }
595 
596 static __inline void
597 vlan_inithash(struct ifvlantrunk *trunk)
598 {
599 }
600 
601 #endif /* !VLAN_ARRAY */
602 
603 static void
604 trunk_destroy(struct ifvlantrunk *trunk)
605 {
606 	VLAN_XLOCK_ASSERT();
607 
608 	vlan_freehash(trunk);
609 	trunk->parent->if_vlantrunk = NULL;
610 	TRUNK_LOCK_DESTROY(trunk);
611 	if_rele(trunk->parent);
612 	free(trunk, M_VLAN);
613 }
614 
615 /*
616  * Program our multicast filter. What we're actually doing is
617  * programming the multicast filter of the parent. This has the
618  * side effect of causing the parent interface to receive multicast
619  * traffic that it doesn't really want, which ends up being discarded
620  * later by the upper protocol layers. Unfortunately, there's no way
621  * to avoid this: there really is only one physical interface.
622  */
623 static int
624 vlan_setmulti(struct ifnet *ifp)
625 {
626 	struct ifnet		*ifp_p;
627 	struct ifmultiaddr	*ifma;
628 	struct ifvlan		*sc;
629 	struct vlan_mc_entry	*mc;
630 	int			error;
631 
632 	VLAN_XLOCK_ASSERT();
633 
634 	/* Find the parent. */
635 	sc = ifp->if_softc;
636 	ifp_p = PARENT(sc);
637 
638 	CURVNET_SET_QUIET(ifp_p->if_vnet);
639 
640 	/* First, remove any existing filter entries. */
641 	while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
642 		CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
643 		(void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
644 		NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
645 	}
646 
647 	/* Now program new ones. */
648 	IF_ADDR_WLOCK(ifp);
649 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
650 		if (ifma->ifma_addr->sa_family != AF_LINK)
651 			continue;
652 		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
653 		if (mc == NULL) {
654 			IF_ADDR_WUNLOCK(ifp);
655 			CURVNET_RESTORE();
656 			return (ENOMEM);
657 		}
658 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
659 		mc->mc_addr.sdl_index = ifp_p->if_index;
660 		CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
661 	}
662 	IF_ADDR_WUNLOCK(ifp);
663 	CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
664 		error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
665 		    NULL);
666 		if (error) {
667 			CURVNET_RESTORE();
668 			return (error);
669 		}
670 	}
671 
672 	CURVNET_RESTORE();
673 	return (0);
674 }
675 
676 /*
677  * A handler for parent interface link layer address changes.
678  * If the parent interface link layer address is changed we
679  * should also change it on all children vlans.
680  */
681 static void
682 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
683 {
684 	struct epoch_tracker et;
685 	struct ifvlan *ifv;
686 	struct ifnet *ifv_ifp;
687 	struct ifvlantrunk *trunk;
688 	struct sockaddr_dl *sdl;
689 
690 	/* Need the epoch since this is run on taskqueue_swi. */
691 	NET_EPOCH_ENTER(et);
692 	trunk = ifp->if_vlantrunk;
693 	if (trunk == NULL) {
694 		NET_EPOCH_EXIT(et);
695 		return;
696 	}
697 
698 	/*
699 	 * OK, it's a trunk.  Loop over and change all vlan's lladdrs on it.
700 	 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
701 	 * ioctl calls on the parent garbling the lladdr of the child vlan.
702 	 */
703 	TRUNK_WLOCK(trunk);
704 	VLAN_FOREACH(ifv, trunk) {
705 		/*
706 		 * Copy new new lladdr into the ifv_ifp, enqueue a task
707 		 * to actually call if_setlladdr. if_setlladdr needs to
708 		 * be deferred to a taskqueue because it will call into
709 		 * the if_vlan ioctl path and try to acquire the global
710 		 * lock.
711 		 */
712 		ifv_ifp = ifv->ifv_ifp;
713 		bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
714 		    ifp->if_addrlen);
715 		sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
716 		sdl->sdl_alen = ifp->if_addrlen;
717 		taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
718 	}
719 	TRUNK_WUNLOCK(trunk);
720 	NET_EPOCH_EXIT(et);
721 }
722 
723 /*
724  * A handler for network interface departure events.
725  * Track departure of trunks here so that we don't access invalid
726  * pointers or whatever if a trunk is ripped from under us, e.g.,
727  * by ejecting its hot-plug card.  However, if an ifnet is simply
728  * being renamed, then there's no need to tear down the state.
729  */
730 static void
731 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
732 {
733 	struct ifvlan *ifv;
734 	struct ifvlantrunk *trunk;
735 
736 	/* If the ifnet is just being renamed, don't do anything. */
737 	if (ifp->if_flags & IFF_RENAMING)
738 		return;
739 	VLAN_XLOCK();
740 	trunk = ifp->if_vlantrunk;
741 	if (trunk == NULL) {
742 		VLAN_XUNLOCK();
743 		return;
744 	}
745 
746 	/*
747 	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
748 	 * Check trunk pointer after each vlan_unconfig() as it will
749 	 * free it and set to NULL after the last vlan was detached.
750 	 */
751 	VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
752 	    ifp->if_vlantrunk == NULL)
753 		vlan_unconfig_locked(ifv->ifv_ifp, 1);
754 
755 	/* Trunk should have been destroyed in vlan_unconfig(). */
756 	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
757 	VLAN_XUNLOCK();
758 }
759 
760 /*
761  * Return the trunk device for a virtual interface.
762  */
763 static struct ifnet  *
764 vlan_trunkdev(struct ifnet *ifp)
765 {
766 	struct ifvlan *ifv;
767 
768 	NET_EPOCH_ASSERT();
769 
770 	if (ifp->if_type != IFT_L2VLAN)
771 		return (NULL);
772 
773 	ifv = ifp->if_softc;
774 	ifp = NULL;
775 	if (ifv->ifv_trunk)
776 		ifp = PARENT(ifv);
777 	return (ifp);
778 }
779 
780 /*
781  * Return the 12-bit VLAN VID for this interface, for use by external
782  * components such as Infiniband.
783  *
784  * XXXRW: Note that the function name here is historical; it should be named
785  * vlan_vid().
786  */
787 static int
788 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
789 {
790 	struct ifvlan *ifv;
791 
792 	if (ifp->if_type != IFT_L2VLAN)
793 		return (EINVAL);
794 	ifv = ifp->if_softc;
795 	*vidp = ifv->ifv_vid;
796 	return (0);
797 }
798 
799 static int
800 vlan_pcp(struct ifnet *ifp, uint16_t *pcpp)
801 {
802 	struct ifvlan *ifv;
803 
804 	if (ifp->if_type != IFT_L2VLAN)
805 		return (EINVAL);
806 	ifv = ifp->if_softc;
807 	*pcpp = ifv->ifv_pcp;
808 	return (0);
809 }
810 
811 /*
812  * Return a driver specific cookie for this interface.  Synchronization
813  * with setcookie must be provided by the driver.
814  */
815 static void *
816 vlan_cookie(struct ifnet *ifp)
817 {
818 	struct ifvlan *ifv;
819 
820 	if (ifp->if_type != IFT_L2VLAN)
821 		return (NULL);
822 	ifv = ifp->if_softc;
823 	return (ifv->ifv_cookie);
824 }
825 
826 /*
827  * Store a cookie in our softc that drivers can use to store driver
828  * private per-instance data in.
829  */
830 static int
831 vlan_setcookie(struct ifnet *ifp, void *cookie)
832 {
833 	struct ifvlan *ifv;
834 
835 	if (ifp->if_type != IFT_L2VLAN)
836 		return (EINVAL);
837 	ifv = ifp->if_softc;
838 	ifv->ifv_cookie = cookie;
839 	return (0);
840 }
841 
842 /*
843  * Return the vlan device present at the specific VID.
844  */
845 static struct ifnet *
846 vlan_devat(struct ifnet *ifp, uint16_t vid)
847 {
848 	struct ifvlantrunk *trunk;
849 	struct ifvlan *ifv;
850 
851 	NET_EPOCH_ASSERT();
852 
853 	trunk = ifp->if_vlantrunk;
854 	if (trunk == NULL)
855 		return (NULL);
856 	ifp = NULL;
857 	ifv = vlan_gethash(trunk, vid);
858 	if (ifv)
859 		ifp = ifv->ifv_ifp;
860 	return (ifp);
861 }
862 
863 /*
864  * VLAN support can be loaded as a module.  The only place in the
865  * system that's intimately aware of this is ether_input.  We hook
866  * into this code through vlan_input_p which is defined there and
867  * set here.  No one else in the system should be aware of this so
868  * we use an explicit reference here.
869  */
870 extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
871 
872 /* For if_link_state_change() eyes only... */
873 extern	void (*vlan_link_state_p)(struct ifnet *);
874 
875 static int
876 vlan_modevent(module_t mod, int type, void *data)
877 {
878 
879 	switch (type) {
880 	case MOD_LOAD:
881 		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
882 		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
883 		if (ifdetach_tag == NULL)
884 			return (ENOMEM);
885 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
886 		    vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
887 		if (iflladdr_tag == NULL)
888 			return (ENOMEM);
889 		VLAN_LOCKING_INIT();
890 		vlan_input_p = vlan_input;
891 		vlan_link_state_p = vlan_link_state;
892 		vlan_trunk_cap_p = vlan_trunk_capabilities;
893 		vlan_trunkdev_p = vlan_trunkdev;
894 		vlan_cookie_p = vlan_cookie;
895 		vlan_setcookie_p = vlan_setcookie;
896 		vlan_tag_p = vlan_tag;
897 		vlan_pcp_p = vlan_pcp;
898 		vlan_devat_p = vlan_devat;
899 #ifndef VIMAGE
900 		vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
901 		    vlan_clone_create, vlan_clone_destroy);
902 #endif
903 		if (bootverbose)
904 			printf("vlan: initialized, using "
905 #ifdef VLAN_ARRAY
906 			       "full-size arrays"
907 #else
908 			       "hash tables with chaining"
909 #endif
910 
911 			       "\n");
912 		break;
913 	case MOD_UNLOAD:
914 #ifndef VIMAGE
915 		if_clone_detach(vlan_cloner);
916 #endif
917 		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
918 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
919 		vlan_input_p = NULL;
920 		vlan_link_state_p = NULL;
921 		vlan_trunk_cap_p = NULL;
922 		vlan_trunkdev_p = NULL;
923 		vlan_tag_p = NULL;
924 		vlan_cookie_p = NULL;
925 		vlan_setcookie_p = NULL;
926 		vlan_devat_p = NULL;
927 		VLAN_LOCKING_DESTROY();
928 		if (bootverbose)
929 			printf("vlan: unloaded\n");
930 		break;
931 	default:
932 		return (EOPNOTSUPP);
933 	}
934 	return (0);
935 }
936 
937 static moduledata_t vlan_mod = {
938 	"if_vlan",
939 	vlan_modevent,
940 	0
941 };
942 
943 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
944 MODULE_VERSION(if_vlan, 3);
945 
946 #ifdef VIMAGE
947 static void
948 vnet_vlan_init(const void *unused __unused)
949 {
950 
951 	vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
952 		    vlan_clone_create, vlan_clone_destroy);
953 	V_vlan_cloner = vlan_cloner;
954 }
955 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
956     vnet_vlan_init, NULL);
957 
958 static void
959 vnet_vlan_uninit(const void *unused __unused)
960 {
961 
962 	if_clone_detach(V_vlan_cloner);
963 }
964 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
965     vnet_vlan_uninit, NULL);
966 #endif
967 
968 /*
969  * Check for <etherif>.<vlan>[.<vlan> ...] style interface names.
970  */
971 static struct ifnet *
972 vlan_clone_match_ethervid(const char *name, int *vidp)
973 {
974 	char ifname[IFNAMSIZ];
975 	char *cp;
976 	struct ifnet *ifp;
977 	int vid;
978 
979 	strlcpy(ifname, name, IFNAMSIZ);
980 	if ((cp = strrchr(ifname, '.')) == NULL)
981 		return (NULL);
982 	*cp = '\0';
983 	if ((ifp = ifunit_ref(ifname)) == NULL)
984 		return (NULL);
985 	/* Parse VID. */
986 	if (*++cp == '\0') {
987 		if_rele(ifp);
988 		return (NULL);
989 	}
990 	vid = 0;
991 	for(; *cp >= '0' && *cp <= '9'; cp++)
992 		vid = (vid * 10) + (*cp - '0');
993 	if (*cp != '\0') {
994 		if_rele(ifp);
995 		return (NULL);
996 	}
997 	if (vidp != NULL)
998 		*vidp = vid;
999 
1000 	return (ifp);
1001 }
1002 
1003 static int
1004 vlan_clone_match(struct if_clone *ifc, const char *name)
1005 {
1006 	struct ifnet *ifp;
1007 	const char *cp;
1008 
1009 	ifp = vlan_clone_match_ethervid(name, NULL);
1010 	if (ifp != NULL) {
1011 		if_rele(ifp);
1012 		return (1);
1013 	}
1014 
1015 	if (strncmp(vlanname, name, strlen(vlanname)) != 0)
1016 		return (0);
1017 	for (cp = name + 4; *cp != '\0'; cp++) {
1018 		if (*cp < '0' || *cp > '9')
1019 			return (0);
1020 	}
1021 
1022 	return (1);
1023 }
1024 
1025 static int
1026 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
1027 {
1028 	char *dp;
1029 	bool wildcard = false;
1030 	bool subinterface = false;
1031 	int unit;
1032 	int error;
1033 	int vid = 0;
1034 	uint16_t proto = ETHERTYPE_VLAN;
1035 	struct ifvlan *ifv;
1036 	struct ifnet *ifp;
1037 	struct ifnet *p = NULL;
1038 	struct ifaddr *ifa;
1039 	struct sockaddr_dl *sdl;
1040 	struct vlanreq vlr;
1041 	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
1042 
1043 
1044 	/*
1045 	 * There are three ways to specify the cloned device:
1046 	 * o pass a parameter block with the clone request.
1047 	 * o specify parameters in the text of the clone device name
1048 	 * o specify no parameters and get an unattached device that
1049 	 *   must be configured separately.
1050 	 * The first technique is preferred; the latter two are supported
1051 	 * for backwards compatibility.
1052 	 *
1053 	 * XXXRW: Note historic use of the word "tag" here.  New ioctls may be
1054 	 * called for.
1055 	 */
1056 
1057 	if (params) {
1058 		error = copyin(params, &vlr, sizeof(vlr));
1059 		if (error)
1060 			return error;
1061 		vid = vlr.vlr_tag;
1062 		proto = vlr.vlr_proto;
1063 
1064 #ifdef COMPAT_FREEBSD12
1065 		if (proto == 0)
1066 			proto = ETHERTYPE_VLAN;
1067 #endif
1068 		p = ifunit_ref(vlr.vlr_parent);
1069 		if (p == NULL)
1070 			return (ENXIO);
1071 	}
1072 
1073 	if ((error = ifc_name2unit(name, &unit)) == 0) {
1074 
1075 		/*
1076 		 * vlanX interface. Set wildcard to true if the unit number
1077 		 * is not fixed (-1)
1078 		 */
1079 		wildcard = (unit < 0);
1080 	} else {
1081 		struct ifnet *p_tmp = vlan_clone_match_ethervid(name, &vid);
1082 		if (p_tmp != NULL) {
1083 			error = 0;
1084 			subinterface = true;
1085 			unit = IF_DUNIT_NONE;
1086 			wildcard = false;
1087 			if (p != NULL) {
1088 				if_rele(p_tmp);
1089 				if (p != p_tmp)
1090 					error = EINVAL;
1091 			} else
1092 				p = p_tmp;
1093 		} else
1094 			error = ENXIO;
1095 	}
1096 
1097 	if (error != 0) {
1098 		if (p != NULL)
1099 			if_rele(p);
1100 		return (error);
1101 	}
1102 
1103 	if (!subinterface) {
1104 		/* vlanX interface, mark X as busy or allocate new unit # */
1105 		error = ifc_alloc_unit(ifc, &unit);
1106 		if (error != 0) {
1107 			if (p != NULL)
1108 				if_rele(p);
1109 			return (error);
1110 		}
1111 	}
1112 
1113 	/* In the wildcard case, we need to update the name. */
1114 	if (wildcard) {
1115 		for (dp = name; *dp != '\0'; dp++);
1116 		if (snprintf(dp, len - (dp-name), "%d", unit) >
1117 		    len - (dp-name) - 1) {
1118 			panic("%s: interface name too long", __func__);
1119 		}
1120 	}
1121 
1122 	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
1123 	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
1124 	if (ifp == NULL) {
1125 		if (!subinterface)
1126 			ifc_free_unit(ifc, unit);
1127 		free(ifv, M_VLAN);
1128 		if (p != NULL)
1129 			if_rele(p);
1130 		return (ENOSPC);
1131 	}
1132 	CK_SLIST_INIT(&ifv->vlan_mc_listhead);
1133 	ifp->if_softc = ifv;
1134 	/*
1135 	 * Set the name manually rather than using if_initname because
1136 	 * we don't conform to the default naming convention for interfaces.
1137 	 */
1138 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
1139 	ifp->if_dname = vlanname;
1140 	ifp->if_dunit = unit;
1141 
1142 	ifp->if_init = vlan_init;
1143 #ifdef ALTQ
1144 	ifp->if_start = vlan_altq_start;
1145 	ifp->if_transmit = vlan_altq_transmit;
1146 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
1147 	ifp->if_snd.ifq_drv_maxlen = 0;
1148 	IFQ_SET_READY(&ifp->if_snd);
1149 #else
1150 	ifp->if_transmit = vlan_transmit;
1151 #endif
1152 	ifp->if_qflush = vlan_qflush;
1153 	ifp->if_ioctl = vlan_ioctl;
1154 #if defined(KERN_TLS) || defined(RATELIMIT)
1155 	ifp->if_snd_tag_alloc = vlan_snd_tag_alloc;
1156 	ifp->if_ratelimit_query = vlan_ratelimit_query;
1157 #endif
1158 	ifp->if_flags = VLAN_IFFLAGS;
1159 	ether_ifattach(ifp, eaddr);
1160 	/* Now undo some of the damage... */
1161 	ifp->if_baudrate = 0;
1162 	ifp->if_type = IFT_L2VLAN;
1163 	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
1164 	ifa = ifp->if_addr;
1165 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1166 	sdl->sdl_type = IFT_L2VLAN;
1167 
1168 	if (p != NULL) {
1169 		error = vlan_config(ifv, p, vid, proto);
1170 		if_rele(p);
1171 		if (error != 0) {
1172 			/*
1173 			 * Since we've partially failed, we need to back
1174 			 * out all the way, otherwise userland could get
1175 			 * confused.  Thus, we destroy the interface.
1176 			 */
1177 			ether_ifdetach(ifp);
1178 			vlan_unconfig(ifp);
1179 			if_free(ifp);
1180 			if (!subinterface)
1181 				ifc_free_unit(ifc, unit);
1182 			free(ifv, M_VLAN);
1183 
1184 			return (error);
1185 		}
1186 	}
1187 
1188 	return (0);
1189 }
1190 
1191 static int
1192 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1193 {
1194 	struct ifvlan *ifv = ifp->if_softc;
1195 	int unit = ifp->if_dunit;
1196 
1197 	if (ifp->if_vlantrunk)
1198 		return (EBUSY);
1199 
1200 #ifdef ALTQ
1201 	IFQ_PURGE(&ifp->if_snd);
1202 #endif
1203 	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
1204 	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
1205 	/*
1206 	 * We should have the only reference to the ifv now, so we can now
1207 	 * drain any remaining lladdr task before freeing the ifnet and the
1208 	 * ifvlan.
1209 	 */
1210 	taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
1211 	NET_EPOCH_WAIT();
1212 	if_free(ifp);
1213 	free(ifv, M_VLAN);
1214 	if (unit != IF_DUNIT_NONE)
1215 		ifc_free_unit(ifc, unit);
1216 
1217 	return (0);
1218 }
1219 
1220 /*
1221  * The ifp->if_init entry point for vlan(4) is a no-op.
1222  */
1223 static void
1224 vlan_init(void *foo __unused)
1225 {
1226 }
1227 
1228 /*
1229  * The if_transmit method for vlan(4) interface.
1230  */
1231 static int
1232 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1233 {
1234 	struct ifvlan *ifv;
1235 	struct ifnet *p;
1236 	int error, len, mcast;
1237 
1238 	NET_EPOCH_ASSERT();
1239 
1240 	ifv = ifp->if_softc;
1241 	if (TRUNK(ifv) == NULL) {
1242 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1243 		m_freem(m);
1244 		return (ENETDOWN);
1245 	}
1246 	p = PARENT(ifv);
1247 	len = m->m_pkthdr.len;
1248 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1249 
1250 	BPF_MTAP(ifp, m);
1251 
1252 #if defined(KERN_TLS) || defined(RATELIMIT)
1253 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
1254 		struct vlan_snd_tag *vst;
1255 		struct m_snd_tag *mst;
1256 
1257 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
1258 		mst = m->m_pkthdr.snd_tag;
1259 		vst = mst_to_vst(mst);
1260 		if (vst->tag->ifp != p) {
1261 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1262 			m_freem(m);
1263 			return (EAGAIN);
1264 		}
1265 
1266 		m->m_pkthdr.snd_tag = m_snd_tag_ref(vst->tag);
1267 		m_snd_tag_rele(mst);
1268 	}
1269 #endif
1270 
1271 	/*
1272 	 * Do not run parent's if_transmit() if the parent is not up,
1273 	 * or parent's driver will cause a system crash.
1274 	 */
1275 	if (!UP_AND_RUNNING(p)) {
1276 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1277 		m_freem(m);
1278 		return (ENETDOWN);
1279 	}
1280 
1281 	if (!ether_8021q_frame(&m, ifp, p, &ifv->ifv_qtag)) {
1282 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1283 		return (0);
1284 	}
1285 
1286 	/*
1287 	 * Send it, precisely as ether_output() would have.
1288 	 */
1289 	error = (p->if_transmit)(p, m);
1290 	if (error == 0) {
1291 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1292 		if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1293 		if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1294 	} else
1295 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1296 	return (error);
1297 }
1298 
1299 static int
1300 vlan_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1301     struct route *ro)
1302 {
1303 	struct ifvlan *ifv;
1304 	struct ifnet *p;
1305 
1306 	NET_EPOCH_ASSERT();
1307 
1308 	/*
1309 	 * Find the first non-VLAN parent interface.
1310 	 */
1311 	ifv = ifp->if_softc;
1312 	do {
1313 		if (TRUNK(ifv) == NULL) {
1314 			m_freem(m);
1315 			return (ENETDOWN);
1316 		}
1317 		p = PARENT(ifv);
1318 		ifv = p->if_softc;
1319 	} while (p->if_type == IFT_L2VLAN);
1320 
1321 	return p->if_output(ifp, m, dst, ro);
1322 }
1323 
1324 #ifdef ALTQ
1325 static void
1326 vlan_altq_start(if_t ifp)
1327 {
1328 	struct ifaltq *ifq = &ifp->if_snd;
1329 	struct mbuf *m;
1330 
1331 	IFQ_LOCK(ifq);
1332 	IFQ_DEQUEUE_NOLOCK(ifq, m);
1333 	while (m != NULL) {
1334 		vlan_transmit(ifp, m);
1335 		IFQ_DEQUEUE_NOLOCK(ifq, m);
1336 	}
1337 	IFQ_UNLOCK(ifq);
1338 }
1339 
1340 static int
1341 vlan_altq_transmit(if_t ifp, struct mbuf *m)
1342 {
1343 	int err;
1344 
1345 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
1346 		IFQ_ENQUEUE(&ifp->if_snd, m, err);
1347 		if (err == 0)
1348 			vlan_altq_start(ifp);
1349 	} else
1350 		err = vlan_transmit(ifp, m);
1351 
1352 	return (err);
1353 }
1354 #endif	/* ALTQ */
1355 
1356 /*
1357  * The ifp->if_qflush entry point for vlan(4) is a no-op.
1358  */
1359 static void
1360 vlan_qflush(struct ifnet *ifp __unused)
1361 {
1362 }
1363 
1364 static void
1365 vlan_input(struct ifnet *ifp, struct mbuf *m)
1366 {
1367 	struct ifvlantrunk *trunk;
1368 	struct ifvlan *ifv;
1369 	struct m_tag *mtag;
1370 	uint16_t vid, tag;
1371 
1372 	NET_EPOCH_ASSERT();
1373 
1374 	trunk = ifp->if_vlantrunk;
1375 	if (trunk == NULL) {
1376 		m_freem(m);
1377 		return;
1378 	}
1379 
1380 	if (m->m_flags & M_VLANTAG) {
1381 		/*
1382 		 * Packet is tagged, but m contains a normal
1383 		 * Ethernet frame; the tag is stored out-of-band.
1384 		 */
1385 		tag = m->m_pkthdr.ether_vtag;
1386 		m->m_flags &= ~M_VLANTAG;
1387 	} else {
1388 		struct ether_vlan_header *evl;
1389 
1390 		/*
1391 		 * Packet is tagged in-band as specified by 802.1q.
1392 		 */
1393 		switch (ifp->if_type) {
1394 		case IFT_ETHER:
1395 			if (m->m_len < sizeof(*evl) &&
1396 			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
1397 				if_printf(ifp, "cannot pullup VLAN header\n");
1398 				return;
1399 			}
1400 			evl = mtod(m, struct ether_vlan_header *);
1401 			tag = ntohs(evl->evl_tag);
1402 
1403 			/*
1404 			 * Remove the 802.1q header by copying the Ethernet
1405 			 * addresses over it and adjusting the beginning of
1406 			 * the data in the mbuf.  The encapsulated Ethernet
1407 			 * type field is already in place.
1408 			 */
1409 			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1410 			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
1411 			m_adj(m, ETHER_VLAN_ENCAP_LEN);
1412 			break;
1413 
1414 		default:
1415 #ifdef INVARIANTS
1416 			panic("%s: %s has unsupported if_type %u",
1417 			      __func__, ifp->if_xname, ifp->if_type);
1418 #endif
1419 			if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1420 			m_freem(m);
1421 			return;
1422 		}
1423 	}
1424 
1425 	vid = EVL_VLANOFTAG(tag);
1426 
1427 	ifv = vlan_gethash(trunk, vid);
1428 	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1429 		if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1430 		m_freem(m);
1431 		return;
1432 	}
1433 
1434 	if (V_vlan_mtag_pcp) {
1435 		/*
1436 		 * While uncommon, it is possible that we will find a 802.1q
1437 		 * packet encapsulated inside another packet that also had an
1438 		 * 802.1q header.  For example, ethernet tunneled over IPSEC
1439 		 * arriving over ethernet.  In that case, we replace the
1440 		 * existing 802.1q PCP m_tag value.
1441 		 */
1442 		mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1443 		if (mtag == NULL) {
1444 			mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1445 			    sizeof(uint8_t), M_NOWAIT);
1446 			if (mtag == NULL) {
1447 				if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1448 				m_freem(m);
1449 				return;
1450 			}
1451 			m_tag_prepend(m, mtag);
1452 		}
1453 		*(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1454 	}
1455 
1456 	m->m_pkthdr.rcvif = ifv->ifv_ifp;
1457 	if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1458 
1459 	/* Pass it back through the parent's input routine. */
1460 	(*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
1461 }
1462 
1463 static void
1464 vlan_lladdr_fn(void *arg, int pending __unused)
1465 {
1466 	struct ifvlan *ifv;
1467 	struct ifnet *ifp;
1468 
1469 	ifv = (struct ifvlan *)arg;
1470 	ifp = ifv->ifv_ifp;
1471 
1472 	CURVNET_SET(ifp->if_vnet);
1473 
1474 	/* The ifv_ifp already has the lladdr copied in. */
1475 	if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1476 
1477 	CURVNET_RESTORE();
1478 }
1479 
1480 static int
1481 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid,
1482 	uint16_t proto)
1483 {
1484 	struct epoch_tracker et;
1485 	struct ifvlantrunk *trunk;
1486 	struct ifnet *ifp;
1487 	int error = 0;
1488 
1489 	/*
1490 	 * We can handle non-ethernet hardware types as long as
1491 	 * they handle the tagging and headers themselves.
1492 	 */
1493 	if (p->if_type != IFT_ETHER &&
1494 	    p->if_type != IFT_L2VLAN &&
1495 	    (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1496 		return (EPROTONOSUPPORT);
1497 	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1498 		return (EPROTONOSUPPORT);
1499 	/*
1500 	 * Don't let the caller set up a VLAN VID with
1501 	 * anything except VLID bits.
1502 	 * VID numbers 0x0 and 0xFFF are reserved.
1503 	 */
1504 	if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1505 		return (EINVAL);
1506 	if (ifv->ifv_trunk)
1507 		return (EBUSY);
1508 
1509 	VLAN_XLOCK();
1510 	if (p->if_vlantrunk == NULL) {
1511 		trunk = malloc(sizeof(struct ifvlantrunk),
1512 		    M_VLAN, M_WAITOK | M_ZERO);
1513 		vlan_inithash(trunk);
1514 		TRUNK_LOCK_INIT(trunk);
1515 		TRUNK_WLOCK(trunk);
1516 		p->if_vlantrunk = trunk;
1517 		trunk->parent = p;
1518 		if_ref(trunk->parent);
1519 		TRUNK_WUNLOCK(trunk);
1520 	} else {
1521 		trunk = p->if_vlantrunk;
1522 	}
1523 
1524 	ifv->ifv_vid = vid;	/* must set this before vlan_inshash() */
1525 	ifv->ifv_pcp = 0;       /* Default: best effort delivery. */
1526 	error = vlan_inshash(trunk, ifv);
1527 	if (error)
1528 		goto done;
1529 	ifv->ifv_proto = proto;
1530 	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1531 	ifv->ifv_mintu = ETHERMIN;
1532 	ifv->ifv_pflags = 0;
1533 	ifv->ifv_capenable = -1;
1534 
1535 	/*
1536 	 * If the parent supports the VLAN_MTU capability,
1537 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1538 	 * use it.
1539 	 */
1540 	if (p->if_capenable & IFCAP_VLAN_MTU) {
1541 		/*
1542 		 * No need to fudge the MTU since the parent can
1543 		 * handle extended frames.
1544 		 */
1545 		ifv->ifv_mtufudge = 0;
1546 	} else {
1547 		/*
1548 		 * Fudge the MTU by the encapsulation size.  This
1549 		 * makes us incompatible with strictly compliant
1550 		 * 802.1Q implementations, but allows us to use
1551 		 * the feature with other NetBSD implementations,
1552 		 * which might still be useful.
1553 		 */
1554 		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1555 	}
1556 
1557 	ifv->ifv_trunk = trunk;
1558 	ifp = ifv->ifv_ifp;
1559 	/*
1560 	 * Initialize fields from our parent.  This duplicates some
1561 	 * work with ether_ifattach() but allows for non-ethernet
1562 	 * interfaces to also work.
1563 	 */
1564 	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1565 	ifp->if_baudrate = p->if_baudrate;
1566 	ifp->if_input = p->if_input;
1567 	ifp->if_resolvemulti = p->if_resolvemulti;
1568 	ifp->if_addrlen = p->if_addrlen;
1569 	ifp->if_broadcastaddr = p->if_broadcastaddr;
1570 	ifp->if_pcp = ifv->ifv_pcp;
1571 
1572 	/*
1573 	 * We wrap the parent's if_output using vlan_output to ensure that it
1574 	 * can't become stale.
1575 	 */
1576 	ifp->if_output = vlan_output;
1577 
1578 	/*
1579 	 * Copy only a selected subset of flags from the parent.
1580 	 * Other flags are none of our business.
1581 	 */
1582 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1583 	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1584 	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1585 #undef VLAN_COPY_FLAGS
1586 
1587 	ifp->if_link_state = p->if_link_state;
1588 
1589 	NET_EPOCH_ENTER(et);
1590 	vlan_capabilities(ifv);
1591 	NET_EPOCH_EXIT(et);
1592 
1593 	/*
1594 	 * Set up our interface address to reflect the underlying
1595 	 * physical interface's.
1596 	 */
1597 	TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
1598 	((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1599 	    p->if_addrlen;
1600 
1601 	/*
1602 	 * Do not schedule link address update if it was the same
1603 	 * as previous parent's. This helps avoid updating for each
1604 	 * associated llentry.
1605 	 */
1606 	if (memcmp(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen) != 0) {
1607 		bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1608 		taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
1609 	}
1610 
1611 	/* We are ready for operation now. */
1612 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1613 
1614 	/* Update flags on the parent, if necessary. */
1615 	vlan_setflags(ifp, 1);
1616 
1617 	/*
1618 	 * Configure multicast addresses that may already be
1619 	 * joined on the vlan device.
1620 	 */
1621 	(void)vlan_setmulti(ifp);
1622 
1623 done:
1624 	if (error == 0)
1625 		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1626 	VLAN_XUNLOCK();
1627 
1628 	return (error);
1629 }
1630 
1631 static void
1632 vlan_unconfig(struct ifnet *ifp)
1633 {
1634 
1635 	VLAN_XLOCK();
1636 	vlan_unconfig_locked(ifp, 0);
1637 	VLAN_XUNLOCK();
1638 }
1639 
1640 static void
1641 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1642 {
1643 	struct ifvlantrunk *trunk;
1644 	struct vlan_mc_entry *mc;
1645 	struct ifvlan *ifv;
1646 	struct ifnet  *parent;
1647 	int error;
1648 
1649 	VLAN_XLOCK_ASSERT();
1650 
1651 	ifv = ifp->if_softc;
1652 	trunk = ifv->ifv_trunk;
1653 	parent = NULL;
1654 
1655 	if (trunk != NULL) {
1656 		parent = trunk->parent;
1657 
1658 		/*
1659 		 * Since the interface is being unconfigured, we need to
1660 		 * empty the list of multicast groups that we may have joined
1661 		 * while we were alive from the parent's list.
1662 		 */
1663 		while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1664 			/*
1665 			 * If the parent interface is being detached,
1666 			 * all its multicast addresses have already
1667 			 * been removed.  Warn about errors if
1668 			 * if_delmulti() does fail, but don't abort as
1669 			 * all callers expect vlan destruction to
1670 			 * succeed.
1671 			 */
1672 			if (!departing) {
1673 				error = if_delmulti(parent,
1674 				    (struct sockaddr *)&mc->mc_addr);
1675 				if (error)
1676 					if_printf(ifp,
1677 		    "Failed to delete multicast address from parent: %d\n",
1678 					    error);
1679 			}
1680 			CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1681 			NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
1682 		}
1683 
1684 		vlan_setflags(ifp, 0); /* clear special flags on parent */
1685 
1686 		vlan_remhash(trunk, ifv);
1687 		ifv->ifv_trunk = NULL;
1688 
1689 		/*
1690 		 * Check if we were the last.
1691 		 */
1692 		if (trunk->refcnt == 0) {
1693 			parent->if_vlantrunk = NULL;
1694 			NET_EPOCH_WAIT();
1695 			trunk_destroy(trunk);
1696 		}
1697 	}
1698 
1699 	/* Disconnect from parent. */
1700 	if (ifv->ifv_pflags)
1701 		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1702 	ifp->if_mtu = ETHERMTU;
1703 	ifp->if_link_state = LINK_STATE_UNKNOWN;
1704 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1705 
1706 	/*
1707 	 * Only dispatch an event if vlan was
1708 	 * attached, otherwise there is nothing
1709 	 * to cleanup anyway.
1710 	 */
1711 	if (parent != NULL)
1712 		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1713 }
1714 
1715 /* Handle a reference counted flag that should be set on the parent as well */
1716 static int
1717 vlan_setflag(struct ifnet *ifp, int flag, int status,
1718 	     int (*func)(struct ifnet *, int))
1719 {
1720 	struct ifvlan *ifv;
1721 	int error;
1722 
1723 	VLAN_SXLOCK_ASSERT();
1724 
1725 	ifv = ifp->if_softc;
1726 	status = status ? (ifp->if_flags & flag) : 0;
1727 	/* Now "status" contains the flag value or 0 */
1728 
1729 	/*
1730 	 * See if recorded parent's status is different from what
1731 	 * we want it to be.  If it is, flip it.  We record parent's
1732 	 * status in ifv_pflags so that we won't clear parent's flag
1733 	 * we haven't set.  In fact, we don't clear or set parent's
1734 	 * flags directly, but get or release references to them.
1735 	 * That's why we can be sure that recorded flags still are
1736 	 * in accord with actual parent's flags.
1737 	 */
1738 	if (status != (ifv->ifv_pflags & flag)) {
1739 		error = (*func)(PARENT(ifv), status);
1740 		if (error)
1741 			return (error);
1742 		ifv->ifv_pflags &= ~flag;
1743 		ifv->ifv_pflags |= status;
1744 	}
1745 	return (0);
1746 }
1747 
1748 /*
1749  * Handle IFF_* flags that require certain changes on the parent:
1750  * if "status" is true, update parent's flags respective to our if_flags;
1751  * if "status" is false, forcedly clear the flags set on parent.
1752  */
1753 static int
1754 vlan_setflags(struct ifnet *ifp, int status)
1755 {
1756 	int error, i;
1757 
1758 	for (i = 0; vlan_pflags[i].flag; i++) {
1759 		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1760 				     status, vlan_pflags[i].func);
1761 		if (error)
1762 			return (error);
1763 	}
1764 	return (0);
1765 }
1766 
1767 /* Inform all vlans that their parent has changed link state */
1768 static void
1769 vlan_link_state(struct ifnet *ifp)
1770 {
1771 	struct epoch_tracker et;
1772 	struct ifvlantrunk *trunk;
1773 	struct ifvlan *ifv;
1774 
1775 	NET_EPOCH_ENTER(et);
1776 	trunk = ifp->if_vlantrunk;
1777 	if (trunk == NULL) {
1778 		NET_EPOCH_EXIT(et);
1779 		return;
1780 	}
1781 
1782 	TRUNK_WLOCK(trunk);
1783 	VLAN_FOREACH(ifv, trunk) {
1784 		ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1785 		if_link_state_change(ifv->ifv_ifp,
1786 		    trunk->parent->if_link_state);
1787 	}
1788 	TRUNK_WUNLOCK(trunk);
1789 	NET_EPOCH_EXIT(et);
1790 }
1791 
1792 static void
1793 vlan_capabilities(struct ifvlan *ifv)
1794 {
1795 	struct ifnet *p;
1796 	struct ifnet *ifp;
1797 	struct ifnet_hw_tsomax hw_tsomax;
1798 	int cap = 0, ena = 0, mena;
1799 	u_long hwa = 0;
1800 
1801 	NET_EPOCH_ASSERT();
1802 	VLAN_SXLOCK_ASSERT();
1803 
1804 	p = PARENT(ifv);
1805 	ifp = ifv->ifv_ifp;
1806 
1807 	/* Mask parent interface enabled capabilities disabled by user. */
1808 	mena = p->if_capenable & ifv->ifv_capenable;
1809 
1810 	/*
1811 	 * If the parent interface can do checksum offloading
1812 	 * on VLANs, then propagate its hardware-assisted
1813 	 * checksumming flags. Also assert that checksum
1814 	 * offloading requires hardware VLAN tagging.
1815 	 */
1816 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1817 		cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1818 	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1819 	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1820 		ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1821 		if (ena & IFCAP_TXCSUM)
1822 			hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
1823 			    CSUM_UDP | CSUM_SCTP);
1824 		if (ena & IFCAP_TXCSUM_IPV6)
1825 			hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
1826 			    CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
1827 	}
1828 
1829 	/*
1830 	 * If the parent interface can do TSO on VLANs then
1831 	 * propagate the hardware-assisted flag. TSO on VLANs
1832 	 * does not necessarily require hardware VLAN tagging.
1833 	 */
1834 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1835 	if_hw_tsomax_common(p, &hw_tsomax);
1836 	if_hw_tsomax_update(ifp, &hw_tsomax);
1837 	if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1838 		cap |= p->if_capabilities & IFCAP_TSO;
1839 	if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1840 		ena |= mena & IFCAP_TSO;
1841 		if (ena & IFCAP_TSO)
1842 			hwa |= p->if_hwassist & CSUM_TSO;
1843 	}
1844 
1845 	/*
1846 	 * If the parent interface can do LRO and checksum offloading on
1847 	 * VLANs, then guess it may do LRO on VLANs.  False positive here
1848 	 * cost nothing, while false negative may lead to some confusions.
1849 	 */
1850 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1851 		cap |= p->if_capabilities & IFCAP_LRO;
1852 	if (p->if_capenable & IFCAP_VLAN_HWCSUM)
1853 		ena |= p->if_capenable & IFCAP_LRO;
1854 
1855 	/*
1856 	 * If the parent interface can offload TCP connections over VLANs then
1857 	 * propagate its TOE capability to the VLAN interface.
1858 	 *
1859 	 * All TOE drivers in the tree today can deal with VLANs.  If this
1860 	 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1861 	 * with its own bit.
1862 	 */
1863 #define	IFCAP_VLAN_TOE IFCAP_TOE
1864 	if (p->if_capabilities & IFCAP_VLAN_TOE)
1865 		cap |= p->if_capabilities & IFCAP_TOE;
1866 	if (p->if_capenable & IFCAP_VLAN_TOE) {
1867 		TOEDEV(ifp) = TOEDEV(p);
1868 		ena |= mena & IFCAP_TOE;
1869 	}
1870 
1871 	/*
1872 	 * If the parent interface supports dynamic link state, so does the
1873 	 * VLAN interface.
1874 	 */
1875 	cap |= (p->if_capabilities & IFCAP_LINKSTATE);
1876 	ena |= (mena & IFCAP_LINKSTATE);
1877 
1878 #ifdef RATELIMIT
1879 	/*
1880 	 * If the parent interface supports ratelimiting, so does the
1881 	 * VLAN interface.
1882 	 */
1883 	cap |= (p->if_capabilities & IFCAP_TXRTLMT);
1884 	ena |= (mena & IFCAP_TXRTLMT);
1885 #endif
1886 
1887 	/*
1888 	 * If the parent interface supports unmapped mbufs, so does
1889 	 * the VLAN interface.  Note that this should be fine even for
1890 	 * interfaces that don't support hardware tagging as headers
1891 	 * are prepended in normal mbufs to unmapped mbufs holding
1892 	 * payload data.
1893 	 */
1894 	cap |= (p->if_capabilities & IFCAP_MEXTPG);
1895 	ena |= (mena & IFCAP_MEXTPG);
1896 
1897 	/*
1898 	 * If the parent interface can offload encryption and segmentation
1899 	 * of TLS records over TCP, propagate it's capability to the VLAN
1900 	 * interface.
1901 	 *
1902 	 * All TLS drivers in the tree today can deal with VLANs.  If
1903 	 * this ever changes, then a new IFCAP_VLAN_TXTLS can be
1904 	 * defined.
1905 	 */
1906 	if (p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
1907 		cap |= p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
1908 	if (p->if_capenable & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
1909 		ena |= mena & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
1910 
1911 	ifp->if_capabilities = cap;
1912 	ifp->if_capenable = ena;
1913 	ifp->if_hwassist = hwa;
1914 }
1915 
1916 static void
1917 vlan_trunk_capabilities(struct ifnet *ifp)
1918 {
1919 	struct epoch_tracker et;
1920 	struct ifvlantrunk *trunk;
1921 	struct ifvlan *ifv;
1922 
1923 	VLAN_SLOCK();
1924 	trunk = ifp->if_vlantrunk;
1925 	if (trunk == NULL) {
1926 		VLAN_SUNLOCK();
1927 		return;
1928 	}
1929 	NET_EPOCH_ENTER(et);
1930 	VLAN_FOREACH(ifv, trunk)
1931 		vlan_capabilities(ifv);
1932 	NET_EPOCH_EXIT(et);
1933 	VLAN_SUNLOCK();
1934 }
1935 
1936 static int
1937 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1938 {
1939 	struct ifnet *p;
1940 	struct ifreq *ifr;
1941 	struct ifaddr *ifa;
1942 	struct ifvlan *ifv;
1943 	struct ifvlantrunk *trunk;
1944 	struct vlanreq vlr;
1945 	int error = 0, oldmtu;
1946 
1947 	ifr = (struct ifreq *)data;
1948 	ifa = (struct ifaddr *) data;
1949 	ifv = ifp->if_softc;
1950 
1951 	switch (cmd) {
1952 	case SIOCSIFADDR:
1953 		ifp->if_flags |= IFF_UP;
1954 #ifdef INET
1955 		if (ifa->ifa_addr->sa_family == AF_INET)
1956 			arp_ifinit(ifp, ifa);
1957 #endif
1958 		break;
1959 	case SIOCGIFADDR:
1960 		bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
1961 		    ifp->if_addrlen);
1962 		break;
1963 	case SIOCGIFMEDIA:
1964 		VLAN_SLOCK();
1965 		if (TRUNK(ifv) != NULL) {
1966 			p = PARENT(ifv);
1967 			if_ref(p);
1968 			error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1969 			if_rele(p);
1970 			/* Limit the result to the parent's current config. */
1971 			if (error == 0) {
1972 				struct ifmediareq *ifmr;
1973 
1974 				ifmr = (struct ifmediareq *)data;
1975 				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1976 					ifmr->ifm_count = 1;
1977 					error = copyout(&ifmr->ifm_current,
1978 						ifmr->ifm_ulist,
1979 						sizeof(int));
1980 				}
1981 			}
1982 		} else {
1983 			error = EINVAL;
1984 		}
1985 		VLAN_SUNLOCK();
1986 		break;
1987 
1988 	case SIOCSIFMEDIA:
1989 		error = EINVAL;
1990 		break;
1991 
1992 	case SIOCSIFMTU:
1993 		/*
1994 		 * Set the interface MTU.
1995 		 */
1996 		VLAN_SLOCK();
1997 		trunk = TRUNK(ifv);
1998 		if (trunk != NULL) {
1999 			TRUNK_WLOCK(trunk);
2000 			if (ifr->ifr_mtu >
2001 			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
2002 			    ifr->ifr_mtu <
2003 			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
2004 				error = EINVAL;
2005 			else
2006 				ifp->if_mtu = ifr->ifr_mtu;
2007 			TRUNK_WUNLOCK(trunk);
2008 		} else
2009 			error = EINVAL;
2010 		VLAN_SUNLOCK();
2011 		break;
2012 
2013 	case SIOCSETVLAN:
2014 #ifdef VIMAGE
2015 		/*
2016 		 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
2017 		 * interface to be delegated to a jail without allowing the
2018 		 * jail to change what underlying interface/VID it is
2019 		 * associated with.  We are not entirely convinced that this
2020 		 * is the right way to accomplish that policy goal.
2021 		 */
2022 		if (ifp->if_vnet != ifp->if_home_vnet) {
2023 			error = EPERM;
2024 			break;
2025 		}
2026 #endif
2027 		error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
2028 		if (error)
2029 			break;
2030 		if (vlr.vlr_parent[0] == '\0') {
2031 			vlan_unconfig(ifp);
2032 			break;
2033 		}
2034 		p = ifunit_ref(vlr.vlr_parent);
2035 		if (p == NULL) {
2036 			error = ENOENT;
2037 			break;
2038 		}
2039 #ifdef COMPAT_FREEBSD12
2040 		if (vlr.vlr_proto == 0)
2041 			vlr.vlr_proto = ETHERTYPE_VLAN;
2042 #endif
2043 		oldmtu = ifp->if_mtu;
2044 		error = vlan_config(ifv, p, vlr.vlr_tag, vlr.vlr_proto);
2045 		if_rele(p);
2046 
2047 		/*
2048 		 * VLAN MTU may change during addition of the vlandev.
2049 		 * If it did, do network layer specific procedure.
2050 		 */
2051 		if (ifp->if_mtu != oldmtu) {
2052 #ifdef INET6
2053 			nd6_setmtu(ifp);
2054 #endif
2055 			rt_updatemtu(ifp);
2056 		}
2057 		break;
2058 
2059 	case SIOCGETVLAN:
2060 #ifdef VIMAGE
2061 		if (ifp->if_vnet != ifp->if_home_vnet) {
2062 			error = EPERM;
2063 			break;
2064 		}
2065 #endif
2066 		bzero(&vlr, sizeof(vlr));
2067 		VLAN_SLOCK();
2068 		if (TRUNK(ifv) != NULL) {
2069 			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
2070 			    sizeof(vlr.vlr_parent));
2071 			vlr.vlr_tag = ifv->ifv_vid;
2072 			vlr.vlr_proto = ifv->ifv_proto;
2073 		}
2074 		VLAN_SUNLOCK();
2075 		error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
2076 		break;
2077 
2078 	case SIOCSIFFLAGS:
2079 		/*
2080 		 * We should propagate selected flags to the parent,
2081 		 * e.g., promiscuous mode.
2082 		 */
2083 		VLAN_XLOCK();
2084 		if (TRUNK(ifv) != NULL)
2085 			error = vlan_setflags(ifp, 1);
2086 		VLAN_XUNLOCK();
2087 		break;
2088 
2089 	case SIOCADDMULTI:
2090 	case SIOCDELMULTI:
2091 		/*
2092 		 * If we don't have a parent, just remember the membership for
2093 		 * when we do.
2094 		 *
2095 		 * XXX We need the rmlock here to avoid sleeping while
2096 		 * holding in6_multi_mtx.
2097 		 */
2098 		VLAN_XLOCK();
2099 		trunk = TRUNK(ifv);
2100 		if (trunk != NULL)
2101 			error = vlan_setmulti(ifp);
2102 		VLAN_XUNLOCK();
2103 
2104 		break;
2105 	case SIOCGVLANPCP:
2106 #ifdef VIMAGE
2107 		if (ifp->if_vnet != ifp->if_home_vnet) {
2108 			error = EPERM;
2109 			break;
2110 		}
2111 #endif
2112 		ifr->ifr_vlan_pcp = ifv->ifv_pcp;
2113 		break;
2114 
2115 	case SIOCSVLANPCP:
2116 #ifdef VIMAGE
2117 		if (ifp->if_vnet != ifp->if_home_vnet) {
2118 			error = EPERM;
2119 			break;
2120 		}
2121 #endif
2122 		error = priv_check(curthread, PRIV_NET_SETVLANPCP);
2123 		if (error)
2124 			break;
2125 		if (ifr->ifr_vlan_pcp > VLAN_PCP_MAX) {
2126 			error = EINVAL;
2127 			break;
2128 		}
2129 		ifv->ifv_pcp = ifr->ifr_vlan_pcp;
2130 		ifp->if_pcp = ifv->ifv_pcp;
2131 		/* broadcast event about PCP change */
2132 		EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
2133 		break;
2134 
2135 	case SIOCSIFCAP:
2136 		VLAN_SLOCK();
2137 		ifv->ifv_capenable = ifr->ifr_reqcap;
2138 		trunk = TRUNK(ifv);
2139 		if (trunk != NULL) {
2140 			struct epoch_tracker et;
2141 
2142 			NET_EPOCH_ENTER(et);
2143 			vlan_capabilities(ifv);
2144 			NET_EPOCH_EXIT(et);
2145 		}
2146 		VLAN_SUNLOCK();
2147 		break;
2148 
2149 	default:
2150 		error = EINVAL;
2151 		break;
2152 	}
2153 
2154 	return (error);
2155 }
2156 
2157 #if defined(KERN_TLS) || defined(RATELIMIT)
2158 static int
2159 vlan_snd_tag_alloc(struct ifnet *ifp,
2160     union if_snd_tag_alloc_params *params,
2161     struct m_snd_tag **ppmt)
2162 {
2163 	struct epoch_tracker et;
2164 	const struct if_snd_tag_sw *sw;
2165 	struct vlan_snd_tag *vst;
2166 	struct ifvlan *ifv;
2167 	struct ifnet *parent;
2168 	int error;
2169 
2170 	switch (params->hdr.type) {
2171 #ifdef RATELIMIT
2172 	case IF_SND_TAG_TYPE_UNLIMITED:
2173 		sw = &vlan_snd_tag_ul_sw;
2174 		break;
2175 	case IF_SND_TAG_TYPE_RATE_LIMIT:
2176 		sw = &vlan_snd_tag_rl_sw;
2177 		break;
2178 #endif
2179 #ifdef KERN_TLS
2180 	case IF_SND_TAG_TYPE_TLS:
2181 		sw = &vlan_snd_tag_tls_sw;
2182 		break;
2183 #ifdef RATELIMIT
2184 	case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
2185 		sw = &vlan_snd_tag_tls_rl_sw;
2186 		break;
2187 #endif
2188 #endif
2189 	default:
2190 		return (EOPNOTSUPP);
2191 	}
2192 
2193 	NET_EPOCH_ENTER(et);
2194 	ifv = ifp->if_softc;
2195 	if (ifv->ifv_trunk != NULL)
2196 		parent = PARENT(ifv);
2197 	else
2198 		parent = NULL;
2199 	if (parent == NULL) {
2200 		NET_EPOCH_EXIT(et);
2201 		return (EOPNOTSUPP);
2202 	}
2203 	if_ref(parent);
2204 	NET_EPOCH_EXIT(et);
2205 
2206 	vst = malloc(sizeof(*vst), M_VLAN, M_NOWAIT);
2207 	if (vst == NULL) {
2208 		if_rele(parent);
2209 		return (ENOMEM);
2210 	}
2211 
2212 	error = m_snd_tag_alloc(parent, params, &vst->tag);
2213 	if_rele(parent);
2214 	if (error) {
2215 		free(vst, M_VLAN);
2216 		return (error);
2217 	}
2218 
2219 	m_snd_tag_init(&vst->com, ifp, sw);
2220 
2221 	*ppmt = &vst->com;
2222 	return (0);
2223 }
2224 
2225 static struct m_snd_tag *
2226 vlan_next_snd_tag(struct m_snd_tag *mst)
2227 {
2228 	struct vlan_snd_tag *vst;
2229 
2230 	vst = mst_to_vst(mst);
2231 	return (vst->tag);
2232 }
2233 
2234 static int
2235 vlan_snd_tag_modify(struct m_snd_tag *mst,
2236     union if_snd_tag_modify_params *params)
2237 {
2238 	struct vlan_snd_tag *vst;
2239 
2240 	vst = mst_to_vst(mst);
2241 	return (vst->tag->sw->snd_tag_modify(vst->tag, params));
2242 }
2243 
2244 static int
2245 vlan_snd_tag_query(struct m_snd_tag *mst,
2246     union if_snd_tag_query_params *params)
2247 {
2248 	struct vlan_snd_tag *vst;
2249 
2250 	vst = mst_to_vst(mst);
2251 	return (vst->tag->sw->snd_tag_query(vst->tag, params));
2252 }
2253 
2254 static void
2255 vlan_snd_tag_free(struct m_snd_tag *mst)
2256 {
2257 	struct vlan_snd_tag *vst;
2258 
2259 	vst = mst_to_vst(mst);
2260 	m_snd_tag_rele(vst->tag);
2261 	free(vst, M_VLAN);
2262 }
2263 
2264 static void
2265 vlan_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
2266 {
2267 	/*
2268 	 * For vlan, we have an indirect
2269 	 * interface. The caller needs to
2270 	 * get a ratelimit tag on the actual
2271 	 * interface the flow will go on.
2272 	 */
2273 	q->rate_table = NULL;
2274 	q->flags = RT_IS_INDIRECT;
2275 	q->max_flows = 0;
2276 	q->number_of_rates = 0;
2277 }
2278 
2279 #endif
2280