xref: /freebsd/sys/net/if_vlan.c (revision dc318a4ffabcbfa23bb56a33403aad36e6de30af)
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 static struct m_snd_tag *vlan_next_snd_tag(struct m_snd_tag *);
299 static void vlan_ratelimit_query(struct ifnet *,
300     struct if_ratelimit_query_results *);
301 #endif
302 static	void vlan_qflush(struct ifnet *ifp);
303 static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
304     int (*func)(struct ifnet *, int));
305 static	int vlan_setflags(struct ifnet *ifp, int status);
306 static	int vlan_setmulti(struct ifnet *ifp);
307 static	int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
308 static	int vlan_output(struct ifnet *ifp, struct mbuf *m,
309     const struct sockaddr *dst, struct route *ro);
310 static	void vlan_unconfig(struct ifnet *ifp);
311 static	void vlan_unconfig_locked(struct ifnet *ifp, int departing);
312 static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag,
313 	uint16_t proto);
314 static	void vlan_link_state(struct ifnet *ifp);
315 static	void vlan_capabilities(struct ifvlan *ifv);
316 static	void vlan_trunk_capabilities(struct ifnet *ifp);
317 
318 static	struct ifnet *vlan_clone_match_ethervid(const char *, int *);
319 static	int vlan_clone_match(struct if_clone *, const char *);
320 static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
321 static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
322 
323 static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
324 static  void vlan_iflladdr(void *arg, struct ifnet *ifp);
325 
326 static  void vlan_lladdr_fn(void *arg, int pending);
327 
328 static struct if_clone *vlan_cloner;
329 
330 #ifdef VIMAGE
331 VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner);
332 #define	V_vlan_cloner	VNET(vlan_cloner)
333 #endif
334 
335 static void
336 vlan_mc_free(struct epoch_context *ctx)
337 {
338 	struct vlan_mc_entry *mc = __containerof(ctx, struct vlan_mc_entry, mc_epoch_ctx);
339 	free(mc, M_VLAN);
340 }
341 
342 #ifndef VLAN_ARRAY
343 #define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
344 
345 static void
346 vlan_inithash(struct ifvlantrunk *trunk)
347 {
348 	int i, n;
349 
350 	/*
351 	 * The trunk must not be locked here since we call malloc(M_WAITOK).
352 	 * It is OK in case this function is called before the trunk struct
353 	 * gets hooked up and becomes visible from other threads.
354 	 */
355 
356 	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
357 	    ("%s: hash already initialized", __func__));
358 
359 	trunk->hwidth = VLAN_DEF_HWIDTH;
360 	n = 1 << trunk->hwidth;
361 	trunk->hmask = n - 1;
362 	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
363 	for (i = 0; i < n; i++)
364 		CK_SLIST_INIT(&trunk->hash[i]);
365 }
366 
367 static void
368 vlan_freehash(struct ifvlantrunk *trunk)
369 {
370 #ifdef INVARIANTS
371 	int i;
372 
373 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
374 	for (i = 0; i < (1 << trunk->hwidth); i++)
375 		KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]),
376 		    ("%s: hash table not empty", __func__));
377 #endif
378 	free(trunk->hash, M_VLAN);
379 	trunk->hash = NULL;
380 	trunk->hwidth = trunk->hmask = 0;
381 }
382 
383 static int
384 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
385 {
386 	int i, b;
387 	struct ifvlan *ifv2;
388 
389 	VLAN_XLOCK_ASSERT();
390 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
391 
392 	b = 1 << trunk->hwidth;
393 	i = HASH(ifv->ifv_vid, trunk->hmask);
394 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
395 		if (ifv->ifv_vid == ifv2->ifv_vid)
396 			return (EEXIST);
397 
398 	/*
399 	 * Grow the hash when the number of vlans exceeds half of the number of
400 	 * hash buckets squared. This will make the average linked-list length
401 	 * buckets/2.
402 	 */
403 	if (trunk->refcnt > (b * b) / 2) {
404 		vlan_growhash(trunk, 1);
405 		i = HASH(ifv->ifv_vid, trunk->hmask);
406 	}
407 	CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
408 	trunk->refcnt++;
409 
410 	return (0);
411 }
412 
413 static int
414 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
415 {
416 	int i, b;
417 	struct ifvlan *ifv2;
418 
419 	VLAN_XLOCK_ASSERT();
420 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
421 
422 	b = 1 << trunk->hwidth;
423 	i = HASH(ifv->ifv_vid, trunk->hmask);
424 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
425 		if (ifv2 == ifv) {
426 			trunk->refcnt--;
427 			CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list);
428 			if (trunk->refcnt < (b * b) / 2)
429 				vlan_growhash(trunk, -1);
430 			return (0);
431 		}
432 
433 	panic("%s: vlan not found\n", __func__);
434 	return (ENOENT); /*NOTREACHED*/
435 }
436 
437 /*
438  * Grow the hash larger or smaller if memory permits.
439  */
440 static void
441 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
442 {
443 	struct ifvlan *ifv;
444 	struct ifvlanhead *hash2;
445 	int hwidth2, i, j, n, n2;
446 
447 	VLAN_XLOCK_ASSERT();
448 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
449 
450 	if (howmuch == 0) {
451 		/* Harmless yet obvious coding error */
452 		printf("%s: howmuch is 0\n", __func__);
453 		return;
454 	}
455 
456 	hwidth2 = trunk->hwidth + howmuch;
457 	n = 1 << trunk->hwidth;
458 	n2 = 1 << hwidth2;
459 	/* Do not shrink the table below the default */
460 	if (hwidth2 < VLAN_DEF_HWIDTH)
461 		return;
462 
463 	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK);
464 	if (hash2 == NULL) {
465 		printf("%s: out of memory -- hash size not changed\n",
466 		    __func__);
467 		return;		/* We can live with the old hash table */
468 	}
469 	for (j = 0; j < n2; j++)
470 		CK_SLIST_INIT(&hash2[j]);
471 	for (i = 0; i < n; i++)
472 		while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) {
473 			CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list);
474 			j = HASH(ifv->ifv_vid, n2 - 1);
475 			CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
476 		}
477 	NET_EPOCH_WAIT();
478 	free(trunk->hash, M_VLAN);
479 	trunk->hash = hash2;
480 	trunk->hwidth = hwidth2;
481 	trunk->hmask = n2 - 1;
482 
483 	if (bootverbose)
484 		if_printf(trunk->parent,
485 		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
486 }
487 
488 static __inline struct ifvlan *
489 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
490 {
491 	struct ifvlan *ifv;
492 
493 	NET_EPOCH_ASSERT();
494 
495 	CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
496 		if (ifv->ifv_vid == vid)
497 			return (ifv);
498 	return (NULL);
499 }
500 
501 #if 0
502 /* Debugging code to view the hashtables. */
503 static void
504 vlan_dumphash(struct ifvlantrunk *trunk)
505 {
506 	int i;
507 	struct ifvlan *ifv;
508 
509 	for (i = 0; i < (1 << trunk->hwidth); i++) {
510 		printf("%d: ", i);
511 		CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
512 			printf("%s ", ifv->ifv_ifp->if_xname);
513 		printf("\n");
514 	}
515 }
516 #endif /* 0 */
517 #else
518 
519 static __inline struct ifvlan *
520 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
521 {
522 
523 	return trunk->vlans[vid];
524 }
525 
526 static __inline int
527 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
528 {
529 
530 	if (trunk->vlans[ifv->ifv_vid] != NULL)
531 		return EEXIST;
532 	trunk->vlans[ifv->ifv_vid] = ifv;
533 	trunk->refcnt++;
534 
535 	return (0);
536 }
537 
538 static __inline int
539 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
540 {
541 
542 	trunk->vlans[ifv->ifv_vid] = NULL;
543 	trunk->refcnt--;
544 
545 	return (0);
546 }
547 
548 static __inline void
549 vlan_freehash(struct ifvlantrunk *trunk)
550 {
551 }
552 
553 static __inline void
554 vlan_inithash(struct ifvlantrunk *trunk)
555 {
556 }
557 
558 #endif /* !VLAN_ARRAY */
559 
560 static void
561 trunk_destroy(struct ifvlantrunk *trunk)
562 {
563 	VLAN_XLOCK_ASSERT();
564 
565 	vlan_freehash(trunk);
566 	trunk->parent->if_vlantrunk = NULL;
567 	TRUNK_LOCK_DESTROY(trunk);
568 	if_rele(trunk->parent);
569 	free(trunk, M_VLAN);
570 }
571 
572 /*
573  * Program our multicast filter. What we're actually doing is
574  * programming the multicast filter of the parent. This has the
575  * side effect of causing the parent interface to receive multicast
576  * traffic that it doesn't really want, which ends up being discarded
577  * later by the upper protocol layers. Unfortunately, there's no way
578  * to avoid this: there really is only one physical interface.
579  */
580 static int
581 vlan_setmulti(struct ifnet *ifp)
582 {
583 	struct ifnet		*ifp_p;
584 	struct ifmultiaddr	*ifma;
585 	struct ifvlan		*sc;
586 	struct vlan_mc_entry	*mc;
587 	int			error;
588 
589 	VLAN_XLOCK_ASSERT();
590 
591 	/* Find the parent. */
592 	sc = ifp->if_softc;
593 	ifp_p = PARENT(sc);
594 
595 	CURVNET_SET_QUIET(ifp_p->if_vnet);
596 
597 	/* First, remove any existing filter entries. */
598 	while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
599 		CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
600 		(void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
601 		NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
602 	}
603 
604 	/* Now program new ones. */
605 	IF_ADDR_WLOCK(ifp);
606 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
607 		if (ifma->ifma_addr->sa_family != AF_LINK)
608 			continue;
609 		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
610 		if (mc == NULL) {
611 			IF_ADDR_WUNLOCK(ifp);
612 			return (ENOMEM);
613 		}
614 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
615 		mc->mc_addr.sdl_index = ifp_p->if_index;
616 		CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
617 	}
618 	IF_ADDR_WUNLOCK(ifp);
619 	CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
620 		error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
621 		    NULL);
622 		if (error)
623 			return (error);
624 	}
625 
626 	CURVNET_RESTORE();
627 	return (0);
628 }
629 
630 /*
631  * A handler for parent interface link layer address changes.
632  * If the parent interface link layer address is changed we
633  * should also change it on all children vlans.
634  */
635 static void
636 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
637 {
638 	struct epoch_tracker et;
639 	struct ifvlan *ifv;
640 	struct ifnet *ifv_ifp;
641 	struct ifvlantrunk *trunk;
642 	struct sockaddr_dl *sdl;
643 
644 	/* Need the epoch since this is run on taskqueue_swi. */
645 	NET_EPOCH_ENTER(et);
646 	trunk = ifp->if_vlantrunk;
647 	if (trunk == NULL) {
648 		NET_EPOCH_EXIT(et);
649 		return;
650 	}
651 
652 	/*
653 	 * OK, it's a trunk.  Loop over and change all vlan's lladdrs on it.
654 	 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
655 	 * ioctl calls on the parent garbling the lladdr of the child vlan.
656 	 */
657 	TRUNK_WLOCK(trunk);
658 	VLAN_FOREACH(ifv, trunk) {
659 		/*
660 		 * Copy new new lladdr into the ifv_ifp, enqueue a task
661 		 * to actually call if_setlladdr. if_setlladdr needs to
662 		 * be deferred to a taskqueue because it will call into
663 		 * the if_vlan ioctl path and try to acquire the global
664 		 * lock.
665 		 */
666 		ifv_ifp = ifv->ifv_ifp;
667 		bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
668 		    ifp->if_addrlen);
669 		sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
670 		sdl->sdl_alen = ifp->if_addrlen;
671 		taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
672 	}
673 	TRUNK_WUNLOCK(trunk);
674 	NET_EPOCH_EXIT(et);
675 }
676 
677 /*
678  * A handler for network interface departure events.
679  * Track departure of trunks here so that we don't access invalid
680  * pointers or whatever if a trunk is ripped from under us, e.g.,
681  * by ejecting its hot-plug card.  However, if an ifnet is simply
682  * being renamed, then there's no need to tear down the state.
683  */
684 static void
685 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
686 {
687 	struct ifvlan *ifv;
688 	struct ifvlantrunk *trunk;
689 
690 	/* If the ifnet is just being renamed, don't do anything. */
691 	if (ifp->if_flags & IFF_RENAMING)
692 		return;
693 	VLAN_XLOCK();
694 	trunk = ifp->if_vlantrunk;
695 	if (trunk == NULL) {
696 		VLAN_XUNLOCK();
697 		return;
698 	}
699 
700 	/*
701 	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
702 	 * Check trunk pointer after each vlan_unconfig() as it will
703 	 * free it and set to NULL after the last vlan was detached.
704 	 */
705 	VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
706 	    ifp->if_vlantrunk == NULL)
707 		vlan_unconfig_locked(ifv->ifv_ifp, 1);
708 
709 	/* Trunk should have been destroyed in vlan_unconfig(). */
710 	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
711 	VLAN_XUNLOCK();
712 }
713 
714 /*
715  * Return the trunk device for a virtual interface.
716  */
717 static struct ifnet  *
718 vlan_trunkdev(struct ifnet *ifp)
719 {
720 	struct ifvlan *ifv;
721 
722 	NET_EPOCH_ASSERT();
723 
724 	if (ifp->if_type != IFT_L2VLAN)
725 		return (NULL);
726 
727 	ifv = ifp->if_softc;
728 	ifp = NULL;
729 	if (ifv->ifv_trunk)
730 		ifp = PARENT(ifv);
731 	return (ifp);
732 }
733 
734 /*
735  * Return the 12-bit VLAN VID for this interface, for use by external
736  * components such as Infiniband.
737  *
738  * XXXRW: Note that the function name here is historical; it should be named
739  * vlan_vid().
740  */
741 static int
742 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
743 {
744 	struct ifvlan *ifv;
745 
746 	if (ifp->if_type != IFT_L2VLAN)
747 		return (EINVAL);
748 	ifv = ifp->if_softc;
749 	*vidp = ifv->ifv_vid;
750 	return (0);
751 }
752 
753 static int
754 vlan_pcp(struct ifnet *ifp, uint16_t *pcpp)
755 {
756 	struct ifvlan *ifv;
757 
758 	if (ifp->if_type != IFT_L2VLAN)
759 		return (EINVAL);
760 	ifv = ifp->if_softc;
761 	*pcpp = ifv->ifv_pcp;
762 	return (0);
763 }
764 
765 /*
766  * Return a driver specific cookie for this interface.  Synchronization
767  * with setcookie must be provided by the driver.
768  */
769 static void *
770 vlan_cookie(struct ifnet *ifp)
771 {
772 	struct ifvlan *ifv;
773 
774 	if (ifp->if_type != IFT_L2VLAN)
775 		return (NULL);
776 	ifv = ifp->if_softc;
777 	return (ifv->ifv_cookie);
778 }
779 
780 /*
781  * Store a cookie in our softc that drivers can use to store driver
782  * private per-instance data in.
783  */
784 static int
785 vlan_setcookie(struct ifnet *ifp, void *cookie)
786 {
787 	struct ifvlan *ifv;
788 
789 	if (ifp->if_type != IFT_L2VLAN)
790 		return (EINVAL);
791 	ifv = ifp->if_softc;
792 	ifv->ifv_cookie = cookie;
793 	return (0);
794 }
795 
796 /*
797  * Return the vlan device present at the specific VID.
798  */
799 static struct ifnet *
800 vlan_devat(struct ifnet *ifp, uint16_t vid)
801 {
802 	struct ifvlantrunk *trunk;
803 	struct ifvlan *ifv;
804 
805 	NET_EPOCH_ASSERT();
806 
807 	trunk = ifp->if_vlantrunk;
808 	if (trunk == NULL)
809 		return (NULL);
810 	ifp = NULL;
811 	ifv = vlan_gethash(trunk, vid);
812 	if (ifv)
813 		ifp = ifv->ifv_ifp;
814 	return (ifp);
815 }
816 
817 /*
818  * VLAN support can be loaded as a module.  The only place in the
819  * system that's intimately aware of this is ether_input.  We hook
820  * into this code through vlan_input_p which is defined there and
821  * set here.  No one else in the system should be aware of this so
822  * we use an explicit reference here.
823  */
824 extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
825 
826 /* For if_link_state_change() eyes only... */
827 extern	void (*vlan_link_state_p)(struct ifnet *);
828 
829 static int
830 vlan_modevent(module_t mod, int type, void *data)
831 {
832 
833 	switch (type) {
834 	case MOD_LOAD:
835 		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
836 		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
837 		if (ifdetach_tag == NULL)
838 			return (ENOMEM);
839 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
840 		    vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
841 		if (iflladdr_tag == NULL)
842 			return (ENOMEM);
843 		VLAN_LOCKING_INIT();
844 		vlan_input_p = vlan_input;
845 		vlan_link_state_p = vlan_link_state;
846 		vlan_trunk_cap_p = vlan_trunk_capabilities;
847 		vlan_trunkdev_p = vlan_trunkdev;
848 		vlan_cookie_p = vlan_cookie;
849 		vlan_setcookie_p = vlan_setcookie;
850 		vlan_tag_p = vlan_tag;
851 		vlan_pcp_p = vlan_pcp;
852 		vlan_devat_p = vlan_devat;
853 #ifndef VIMAGE
854 		vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
855 		    vlan_clone_create, vlan_clone_destroy);
856 #endif
857 		if (bootverbose)
858 			printf("vlan: initialized, using "
859 #ifdef VLAN_ARRAY
860 			       "full-size arrays"
861 #else
862 			       "hash tables with chaining"
863 #endif
864 
865 			       "\n");
866 		break;
867 	case MOD_UNLOAD:
868 #ifndef VIMAGE
869 		if_clone_detach(vlan_cloner);
870 #endif
871 		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
872 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
873 		vlan_input_p = NULL;
874 		vlan_link_state_p = NULL;
875 		vlan_trunk_cap_p = NULL;
876 		vlan_trunkdev_p = NULL;
877 		vlan_tag_p = NULL;
878 		vlan_cookie_p = NULL;
879 		vlan_setcookie_p = NULL;
880 		vlan_devat_p = NULL;
881 		VLAN_LOCKING_DESTROY();
882 		if (bootverbose)
883 			printf("vlan: unloaded\n");
884 		break;
885 	default:
886 		return (EOPNOTSUPP);
887 	}
888 	return (0);
889 }
890 
891 static moduledata_t vlan_mod = {
892 	"if_vlan",
893 	vlan_modevent,
894 	0
895 };
896 
897 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
898 MODULE_VERSION(if_vlan, 3);
899 
900 #ifdef VIMAGE
901 static void
902 vnet_vlan_init(const void *unused __unused)
903 {
904 
905 	vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
906 		    vlan_clone_create, vlan_clone_destroy);
907 	V_vlan_cloner = vlan_cloner;
908 }
909 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
910     vnet_vlan_init, NULL);
911 
912 static void
913 vnet_vlan_uninit(const void *unused __unused)
914 {
915 
916 	if_clone_detach(V_vlan_cloner);
917 }
918 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
919     vnet_vlan_uninit, NULL);
920 #endif
921 
922 /*
923  * Check for <etherif>.<vlan>[.<vlan> ...] style interface names.
924  */
925 static struct ifnet *
926 vlan_clone_match_ethervid(const char *name, int *vidp)
927 {
928 	char ifname[IFNAMSIZ];
929 	char *cp;
930 	struct ifnet *ifp;
931 	int vid;
932 
933 	strlcpy(ifname, name, IFNAMSIZ);
934 	if ((cp = strrchr(ifname, '.')) == NULL)
935 		return (NULL);
936 	*cp = '\0';
937 	if ((ifp = ifunit_ref(ifname)) == NULL)
938 		return (NULL);
939 	/* Parse VID. */
940 	if (*++cp == '\0') {
941 		if_rele(ifp);
942 		return (NULL);
943 	}
944 	vid = 0;
945 	for(; *cp >= '0' && *cp <= '9'; cp++)
946 		vid = (vid * 10) + (*cp - '0');
947 	if (*cp != '\0') {
948 		if_rele(ifp);
949 		return (NULL);
950 	}
951 	if (vidp != NULL)
952 		*vidp = vid;
953 
954 	return (ifp);
955 }
956 
957 static int
958 vlan_clone_match(struct if_clone *ifc, const char *name)
959 {
960 	struct ifnet *ifp;
961 	const char *cp;
962 
963 	ifp = vlan_clone_match_ethervid(name, NULL);
964 	if (ifp != NULL) {
965 		if_rele(ifp);
966 		return (1);
967 	}
968 
969 	if (strncmp(vlanname, name, strlen(vlanname)) != 0)
970 		return (0);
971 	for (cp = name + 4; *cp != '\0'; cp++) {
972 		if (*cp < '0' || *cp > '9')
973 			return (0);
974 	}
975 
976 	return (1);
977 }
978 
979 static int
980 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
981 {
982 	char *dp;
983 	bool wildcard = false;
984 	bool subinterface = false;
985 	int unit;
986 	int error;
987 	int vid = 0;
988 	uint16_t proto = ETHERTYPE_VLAN;
989 	struct ifvlan *ifv;
990 	struct ifnet *ifp;
991 	struct ifnet *p = NULL;
992 	struct ifaddr *ifa;
993 	struct sockaddr_dl *sdl;
994 	struct vlanreq vlr;
995 	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
996 
997 
998 	/*
999 	 * There are three ways to specify the cloned device:
1000 	 * o pass a parameter block with the clone request.
1001 	 * o specify parameters in the text of the clone device name
1002 	 * o specify no parameters and get an unattached device that
1003 	 *   must be configured separately.
1004 	 * The first technique is preferred; the latter two are supported
1005 	 * for backwards compatibility.
1006 	 *
1007 	 * XXXRW: Note historic use of the word "tag" here.  New ioctls may be
1008 	 * called for.
1009 	 */
1010 
1011 	if (params) {
1012 		error = copyin(params, &vlr, sizeof(vlr));
1013 		if (error)
1014 			return error;
1015 		vid = vlr.vlr_tag;
1016 		proto = vlr.vlr_proto;
1017 
1018 #ifdef COMPAT_FREEBSD12
1019 		if (proto == 0)
1020 			proto = ETHERTYPE_VLAN;
1021 #endif
1022 		p = ifunit_ref(vlr.vlr_parent);
1023 		if (p == NULL)
1024 			return (ENXIO);
1025 	}
1026 
1027 	if ((error = ifc_name2unit(name, &unit)) == 0) {
1028 
1029 		/*
1030 		 * vlanX interface. Set wildcard to true if the unit number
1031 		 * is not fixed (-1)
1032 		 */
1033 		wildcard = (unit < 0);
1034 	} else {
1035 		struct ifnet *p_tmp = vlan_clone_match_ethervid(name, &vid);
1036 		if (p_tmp != NULL) {
1037 			error = 0;
1038 			subinterface = true;
1039 			unit = IF_DUNIT_NONE;
1040 			wildcard = false;
1041 			if (p != NULL) {
1042 				if_rele(p_tmp);
1043 				if (p != p_tmp)
1044 					error = EINVAL;
1045 			} else
1046 				p = p_tmp;
1047 		} else
1048 			error = ENXIO;
1049 	}
1050 
1051 	if (error != 0) {
1052 		if (p != NULL)
1053 			if_rele(p);
1054 		return (error);
1055 	}
1056 
1057 	if (!subinterface) {
1058 		/* vlanX interface, mark X as busy or allocate new unit # */
1059 		error = ifc_alloc_unit(ifc, &unit);
1060 		if (error != 0) {
1061 			if (p != NULL)
1062 				if_rele(p);
1063 			return (error);
1064 		}
1065 	}
1066 
1067 	/* In the wildcard case, we need to update the name. */
1068 	if (wildcard) {
1069 		for (dp = name; *dp != '\0'; dp++);
1070 		if (snprintf(dp, len - (dp-name), "%d", unit) >
1071 		    len - (dp-name) - 1) {
1072 			panic("%s: interface name too long", __func__);
1073 		}
1074 	}
1075 
1076 	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
1077 	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
1078 	if (ifp == NULL) {
1079 		if (!subinterface)
1080 			ifc_free_unit(ifc, unit);
1081 		free(ifv, M_VLAN);
1082 		if (p != NULL)
1083 			if_rele(p);
1084 		return (ENOSPC);
1085 	}
1086 	CK_SLIST_INIT(&ifv->vlan_mc_listhead);
1087 	ifp->if_softc = ifv;
1088 	/*
1089 	 * Set the name manually rather than using if_initname because
1090 	 * we don't conform to the default naming convention for interfaces.
1091 	 */
1092 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
1093 	ifp->if_dname = vlanname;
1094 	ifp->if_dunit = unit;
1095 
1096 	ifp->if_init = vlan_init;
1097 	ifp->if_transmit = vlan_transmit;
1098 	ifp->if_qflush = vlan_qflush;
1099 	ifp->if_ioctl = vlan_ioctl;
1100 #if defined(KERN_TLS) || defined(RATELIMIT)
1101 	ifp->if_snd_tag_alloc = vlan_snd_tag_alloc;
1102 	ifp->if_snd_tag_modify = vlan_snd_tag_modify;
1103 	ifp->if_snd_tag_query = vlan_snd_tag_query;
1104 	ifp->if_snd_tag_free = vlan_snd_tag_free;
1105 	ifp->if_next_snd_tag = vlan_next_snd_tag;
1106 	ifp->if_ratelimit_query = vlan_ratelimit_query;
1107 #endif
1108 	ifp->if_flags = VLAN_IFFLAGS;
1109 	ether_ifattach(ifp, eaddr);
1110 	/* Now undo some of the damage... */
1111 	ifp->if_baudrate = 0;
1112 	ifp->if_type = IFT_L2VLAN;
1113 	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
1114 	ifa = ifp->if_addr;
1115 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1116 	sdl->sdl_type = IFT_L2VLAN;
1117 
1118 	if (p != NULL) {
1119 		error = vlan_config(ifv, p, vid, proto);
1120 		if_rele(p);
1121 		if (error != 0) {
1122 			/*
1123 			 * Since we've partially failed, we need to back
1124 			 * out all the way, otherwise userland could get
1125 			 * confused.  Thus, we destroy the interface.
1126 			 */
1127 			ether_ifdetach(ifp);
1128 			vlan_unconfig(ifp);
1129 			if_free(ifp);
1130 			if (!subinterface)
1131 				ifc_free_unit(ifc, unit);
1132 			free(ifv, M_VLAN);
1133 
1134 			return (error);
1135 		}
1136 	}
1137 
1138 	return (0);
1139 }
1140 
1141 static int
1142 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1143 {
1144 	struct ifvlan *ifv = ifp->if_softc;
1145 	int unit = ifp->if_dunit;
1146 
1147 	if (ifp->if_vlantrunk)
1148 		return (EBUSY);
1149 
1150 	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
1151 	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
1152 	/*
1153 	 * We should have the only reference to the ifv now, so we can now
1154 	 * drain any remaining lladdr task before freeing the ifnet and the
1155 	 * ifvlan.
1156 	 */
1157 	taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
1158 	NET_EPOCH_WAIT();
1159 	if_free(ifp);
1160 	free(ifv, M_VLAN);
1161 	if (unit != IF_DUNIT_NONE)
1162 		ifc_free_unit(ifc, unit);
1163 
1164 	return (0);
1165 }
1166 
1167 /*
1168  * The ifp->if_init entry point for vlan(4) is a no-op.
1169  */
1170 static void
1171 vlan_init(void *foo __unused)
1172 {
1173 }
1174 
1175 /*
1176  * The if_transmit method for vlan(4) interface.
1177  */
1178 static int
1179 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1180 {
1181 	struct ifvlan *ifv;
1182 	struct ifnet *p;
1183 	int error, len, mcast;
1184 
1185 	NET_EPOCH_ASSERT();
1186 
1187 	ifv = ifp->if_softc;
1188 	if (TRUNK(ifv) == NULL) {
1189 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1190 		m_freem(m);
1191 		return (ENETDOWN);
1192 	}
1193 	p = PARENT(ifv);
1194 	len = m->m_pkthdr.len;
1195 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1196 
1197 	BPF_MTAP(ifp, m);
1198 
1199 #if defined(KERN_TLS) || defined(RATELIMIT)
1200 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
1201 		struct vlan_snd_tag *vst;
1202 		struct m_snd_tag *mst;
1203 
1204 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
1205 		mst = m->m_pkthdr.snd_tag;
1206 		vst = mst_to_vst(mst);
1207 		if (vst->tag->ifp != p) {
1208 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1209 			m_freem(m);
1210 			return (EAGAIN);
1211 		}
1212 
1213 		m->m_pkthdr.snd_tag = m_snd_tag_ref(vst->tag);
1214 		m_snd_tag_rele(mst);
1215 	}
1216 #endif
1217 
1218 	/*
1219 	 * Do not run parent's if_transmit() if the parent is not up,
1220 	 * or parent's driver will cause a system crash.
1221 	 */
1222 	if (!UP_AND_RUNNING(p)) {
1223 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1224 		m_freem(m);
1225 		return (ENETDOWN);
1226 	}
1227 
1228 	if (!ether_8021q_frame(&m, ifp, p, &ifv->ifv_qtag)) {
1229 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1230 		return (0);
1231 	}
1232 
1233 	/*
1234 	 * Send it, precisely as ether_output() would have.
1235 	 */
1236 	error = (p->if_transmit)(p, m);
1237 	if (error == 0) {
1238 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1239 		if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1240 		if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1241 	} else
1242 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1243 	return (error);
1244 }
1245 
1246 static int
1247 vlan_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1248     struct route *ro)
1249 {
1250 	struct ifvlan *ifv;
1251 	struct ifnet *p;
1252 
1253 	NET_EPOCH_ASSERT();
1254 
1255 	/*
1256 	 * Find the first non-VLAN parent interface.
1257 	 */
1258 	ifv = ifp->if_softc;
1259 	do {
1260 		if (TRUNK(ifv) == NULL) {
1261 			m_freem(m);
1262 			return (ENETDOWN);
1263 		}
1264 		p = PARENT(ifv);
1265 		ifv = p->if_softc;
1266 	} while (p->if_type == IFT_L2VLAN);
1267 
1268 	return p->if_output(ifp, m, dst, ro);
1269 }
1270 
1271 /*
1272  * The ifp->if_qflush entry point for vlan(4) is a no-op.
1273  */
1274 static void
1275 vlan_qflush(struct ifnet *ifp __unused)
1276 {
1277 }
1278 
1279 static void
1280 vlan_input(struct ifnet *ifp, struct mbuf *m)
1281 {
1282 	struct ifvlantrunk *trunk;
1283 	struct ifvlan *ifv;
1284 	struct m_tag *mtag;
1285 	uint16_t vid, tag;
1286 
1287 	NET_EPOCH_ASSERT();
1288 
1289 	trunk = ifp->if_vlantrunk;
1290 	if (trunk == NULL) {
1291 		m_freem(m);
1292 		return;
1293 	}
1294 
1295 	if (m->m_flags & M_VLANTAG) {
1296 		/*
1297 		 * Packet is tagged, but m contains a normal
1298 		 * Ethernet frame; the tag is stored out-of-band.
1299 		 */
1300 		tag = m->m_pkthdr.ether_vtag;
1301 		m->m_flags &= ~M_VLANTAG;
1302 	} else {
1303 		struct ether_vlan_header *evl;
1304 
1305 		/*
1306 		 * Packet is tagged in-band as specified by 802.1q.
1307 		 */
1308 		switch (ifp->if_type) {
1309 		case IFT_ETHER:
1310 			if (m->m_len < sizeof(*evl) &&
1311 			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
1312 				if_printf(ifp, "cannot pullup VLAN header\n");
1313 				return;
1314 			}
1315 			evl = mtod(m, struct ether_vlan_header *);
1316 			tag = ntohs(evl->evl_tag);
1317 
1318 			/*
1319 			 * Remove the 802.1q header by copying the Ethernet
1320 			 * addresses over it and adjusting the beginning of
1321 			 * the data in the mbuf.  The encapsulated Ethernet
1322 			 * type field is already in place.
1323 			 */
1324 			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1325 			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
1326 			m_adj(m, ETHER_VLAN_ENCAP_LEN);
1327 			break;
1328 
1329 		default:
1330 #ifdef INVARIANTS
1331 			panic("%s: %s has unsupported if_type %u",
1332 			      __func__, ifp->if_xname, ifp->if_type);
1333 #endif
1334 			if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1335 			m_freem(m);
1336 			return;
1337 		}
1338 	}
1339 
1340 	vid = EVL_VLANOFTAG(tag);
1341 
1342 	ifv = vlan_gethash(trunk, vid);
1343 	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1344 		if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1345 		m_freem(m);
1346 		return;
1347 	}
1348 
1349 	if (vlan_mtag_pcp) {
1350 		/*
1351 		 * While uncommon, it is possible that we will find a 802.1q
1352 		 * packet encapsulated inside another packet that also had an
1353 		 * 802.1q header.  For example, ethernet tunneled over IPSEC
1354 		 * arriving over ethernet.  In that case, we replace the
1355 		 * existing 802.1q PCP m_tag value.
1356 		 */
1357 		mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1358 		if (mtag == NULL) {
1359 			mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1360 			    sizeof(uint8_t), M_NOWAIT);
1361 			if (mtag == NULL) {
1362 				if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1363 				m_freem(m);
1364 				return;
1365 			}
1366 			m_tag_prepend(m, mtag);
1367 		}
1368 		*(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1369 	}
1370 
1371 	m->m_pkthdr.rcvif = ifv->ifv_ifp;
1372 	if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1373 
1374 	/* Pass it back through the parent's input routine. */
1375 	(*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
1376 }
1377 
1378 static void
1379 vlan_lladdr_fn(void *arg, int pending __unused)
1380 {
1381 	struct ifvlan *ifv;
1382 	struct ifnet *ifp;
1383 
1384 	ifv = (struct ifvlan *)arg;
1385 	ifp = ifv->ifv_ifp;
1386 
1387 	CURVNET_SET(ifp->if_vnet);
1388 
1389 	/* The ifv_ifp already has the lladdr copied in. */
1390 	if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1391 
1392 	CURVNET_RESTORE();
1393 }
1394 
1395 static int
1396 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid,
1397 	uint16_t proto)
1398 {
1399 	struct epoch_tracker et;
1400 	struct ifvlantrunk *trunk;
1401 	struct ifnet *ifp;
1402 	int error = 0;
1403 
1404 	/*
1405 	 * We can handle non-ethernet hardware types as long as
1406 	 * they handle the tagging and headers themselves.
1407 	 */
1408 	if (p->if_type != IFT_ETHER &&
1409 	    p->if_type != IFT_L2VLAN &&
1410 	    (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1411 		return (EPROTONOSUPPORT);
1412 	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1413 		return (EPROTONOSUPPORT);
1414 	/*
1415 	 * Don't let the caller set up a VLAN VID with
1416 	 * anything except VLID bits.
1417 	 * VID numbers 0x0 and 0xFFF are reserved.
1418 	 */
1419 	if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1420 		return (EINVAL);
1421 	if (ifv->ifv_trunk)
1422 		return (EBUSY);
1423 
1424 	VLAN_XLOCK();
1425 	if (p->if_vlantrunk == NULL) {
1426 		trunk = malloc(sizeof(struct ifvlantrunk),
1427 		    M_VLAN, M_WAITOK | M_ZERO);
1428 		vlan_inithash(trunk);
1429 		TRUNK_LOCK_INIT(trunk);
1430 		TRUNK_WLOCK(trunk);
1431 		p->if_vlantrunk = trunk;
1432 		trunk->parent = p;
1433 		if_ref(trunk->parent);
1434 		TRUNK_WUNLOCK(trunk);
1435 	} else {
1436 		trunk = p->if_vlantrunk;
1437 	}
1438 
1439 	ifv->ifv_vid = vid;	/* must set this before vlan_inshash() */
1440 	ifv->ifv_pcp = 0;       /* Default: best effort delivery. */
1441 	error = vlan_inshash(trunk, ifv);
1442 	if (error)
1443 		goto done;
1444 	ifv->ifv_proto = proto;
1445 	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1446 	ifv->ifv_mintu = ETHERMIN;
1447 	ifv->ifv_pflags = 0;
1448 	ifv->ifv_capenable = -1;
1449 
1450 	/*
1451 	 * If the parent supports the VLAN_MTU capability,
1452 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1453 	 * use it.
1454 	 */
1455 	if (p->if_capenable & IFCAP_VLAN_MTU) {
1456 		/*
1457 		 * No need to fudge the MTU since the parent can
1458 		 * handle extended frames.
1459 		 */
1460 		ifv->ifv_mtufudge = 0;
1461 	} else {
1462 		/*
1463 		 * Fudge the MTU by the encapsulation size.  This
1464 		 * makes us incompatible with strictly compliant
1465 		 * 802.1Q implementations, but allows us to use
1466 		 * the feature with other NetBSD implementations,
1467 		 * which might still be useful.
1468 		 */
1469 		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1470 	}
1471 
1472 	ifv->ifv_trunk = trunk;
1473 	ifp = ifv->ifv_ifp;
1474 	/*
1475 	 * Initialize fields from our parent.  This duplicates some
1476 	 * work with ether_ifattach() but allows for non-ethernet
1477 	 * interfaces to also work.
1478 	 */
1479 	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1480 	ifp->if_baudrate = p->if_baudrate;
1481 	ifp->if_input = p->if_input;
1482 	ifp->if_resolvemulti = p->if_resolvemulti;
1483 	ifp->if_addrlen = p->if_addrlen;
1484 	ifp->if_broadcastaddr = p->if_broadcastaddr;
1485 	ifp->if_pcp = ifv->ifv_pcp;
1486 
1487 	/*
1488 	 * We wrap the parent's if_output using vlan_output to ensure that it
1489 	 * can't become stale.
1490 	 */
1491 	ifp->if_output = vlan_output;
1492 
1493 	/*
1494 	 * Copy only a selected subset of flags from the parent.
1495 	 * Other flags are none of our business.
1496 	 */
1497 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1498 	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1499 	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1500 #undef VLAN_COPY_FLAGS
1501 
1502 	ifp->if_link_state = p->if_link_state;
1503 
1504 	NET_EPOCH_ENTER(et);
1505 	vlan_capabilities(ifv);
1506 	NET_EPOCH_EXIT(et);
1507 
1508 	/*
1509 	 * Set up our interface address to reflect the underlying
1510 	 * physical interface's.
1511 	 */
1512 	TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
1513 	((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1514 	    p->if_addrlen;
1515 
1516 	/*
1517 	 * Do not schedule link address update if it was the same
1518 	 * as previous parent's. This helps avoid updating for each
1519 	 * associated llentry.
1520 	 */
1521 	if (memcmp(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen) != 0) {
1522 		bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1523 		taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
1524 	}
1525 
1526 	/* We are ready for operation now. */
1527 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1528 
1529 	/* Update flags on the parent, if necessary. */
1530 	vlan_setflags(ifp, 1);
1531 
1532 	/*
1533 	 * Configure multicast addresses that may already be
1534 	 * joined on the vlan device.
1535 	 */
1536 	(void)vlan_setmulti(ifp);
1537 
1538 done:
1539 	if (error == 0)
1540 		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1541 	VLAN_XUNLOCK();
1542 
1543 	return (error);
1544 }
1545 
1546 static void
1547 vlan_unconfig(struct ifnet *ifp)
1548 {
1549 
1550 	VLAN_XLOCK();
1551 	vlan_unconfig_locked(ifp, 0);
1552 	VLAN_XUNLOCK();
1553 }
1554 
1555 static void
1556 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1557 {
1558 	struct ifvlantrunk *trunk;
1559 	struct vlan_mc_entry *mc;
1560 	struct ifvlan *ifv;
1561 	struct ifnet  *parent;
1562 	int error;
1563 
1564 	VLAN_XLOCK_ASSERT();
1565 
1566 	ifv = ifp->if_softc;
1567 	trunk = ifv->ifv_trunk;
1568 	parent = NULL;
1569 
1570 	if (trunk != NULL) {
1571 		parent = trunk->parent;
1572 
1573 		/*
1574 		 * Since the interface is being unconfigured, we need to
1575 		 * empty the list of multicast groups that we may have joined
1576 		 * while we were alive from the parent's list.
1577 		 */
1578 		while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1579 			/*
1580 			 * If the parent interface is being detached,
1581 			 * all its multicast addresses have already
1582 			 * been removed.  Warn about errors if
1583 			 * if_delmulti() does fail, but don't abort as
1584 			 * all callers expect vlan destruction to
1585 			 * succeed.
1586 			 */
1587 			if (!departing) {
1588 				error = if_delmulti(parent,
1589 				    (struct sockaddr *)&mc->mc_addr);
1590 				if (error)
1591 					if_printf(ifp,
1592 		    "Failed to delete multicast address from parent: %d\n",
1593 					    error);
1594 			}
1595 			CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1596 			NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
1597 		}
1598 
1599 		vlan_setflags(ifp, 0); /* clear special flags on parent */
1600 
1601 		vlan_remhash(trunk, ifv);
1602 		ifv->ifv_trunk = NULL;
1603 
1604 		/*
1605 		 * Check if we were the last.
1606 		 */
1607 		if (trunk->refcnt == 0) {
1608 			parent->if_vlantrunk = NULL;
1609 			NET_EPOCH_WAIT();
1610 			trunk_destroy(trunk);
1611 		}
1612 	}
1613 
1614 	/* Disconnect from parent. */
1615 	if (ifv->ifv_pflags)
1616 		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1617 	ifp->if_mtu = ETHERMTU;
1618 	ifp->if_link_state = LINK_STATE_UNKNOWN;
1619 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1620 
1621 	/*
1622 	 * Only dispatch an event if vlan was
1623 	 * attached, otherwise there is nothing
1624 	 * to cleanup anyway.
1625 	 */
1626 	if (parent != NULL)
1627 		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1628 }
1629 
1630 /* Handle a reference counted flag that should be set on the parent as well */
1631 static int
1632 vlan_setflag(struct ifnet *ifp, int flag, int status,
1633 	     int (*func)(struct ifnet *, int))
1634 {
1635 	struct ifvlan *ifv;
1636 	int error;
1637 
1638 	VLAN_SXLOCK_ASSERT();
1639 
1640 	ifv = ifp->if_softc;
1641 	status = status ? (ifp->if_flags & flag) : 0;
1642 	/* Now "status" contains the flag value or 0 */
1643 
1644 	/*
1645 	 * See if recorded parent's status is different from what
1646 	 * we want it to be.  If it is, flip it.  We record parent's
1647 	 * status in ifv_pflags so that we won't clear parent's flag
1648 	 * we haven't set.  In fact, we don't clear or set parent's
1649 	 * flags directly, but get or release references to them.
1650 	 * That's why we can be sure that recorded flags still are
1651 	 * in accord with actual parent's flags.
1652 	 */
1653 	if (status != (ifv->ifv_pflags & flag)) {
1654 		error = (*func)(PARENT(ifv), status);
1655 		if (error)
1656 			return (error);
1657 		ifv->ifv_pflags &= ~flag;
1658 		ifv->ifv_pflags |= status;
1659 	}
1660 	return (0);
1661 }
1662 
1663 /*
1664  * Handle IFF_* flags that require certain changes on the parent:
1665  * if "status" is true, update parent's flags respective to our if_flags;
1666  * if "status" is false, forcedly clear the flags set on parent.
1667  */
1668 static int
1669 vlan_setflags(struct ifnet *ifp, int status)
1670 {
1671 	int error, i;
1672 
1673 	for (i = 0; vlan_pflags[i].flag; i++) {
1674 		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1675 				     status, vlan_pflags[i].func);
1676 		if (error)
1677 			return (error);
1678 	}
1679 	return (0);
1680 }
1681 
1682 /* Inform all vlans that their parent has changed link state */
1683 static void
1684 vlan_link_state(struct ifnet *ifp)
1685 {
1686 	struct epoch_tracker et;
1687 	struct ifvlantrunk *trunk;
1688 	struct ifvlan *ifv;
1689 
1690 	NET_EPOCH_ENTER(et);
1691 	trunk = ifp->if_vlantrunk;
1692 	if (trunk == NULL) {
1693 		NET_EPOCH_EXIT(et);
1694 		return;
1695 	}
1696 
1697 	TRUNK_WLOCK(trunk);
1698 	VLAN_FOREACH(ifv, trunk) {
1699 		ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1700 		if_link_state_change(ifv->ifv_ifp,
1701 		    trunk->parent->if_link_state);
1702 	}
1703 	TRUNK_WUNLOCK(trunk);
1704 	NET_EPOCH_EXIT(et);
1705 }
1706 
1707 static void
1708 vlan_capabilities(struct ifvlan *ifv)
1709 {
1710 	struct ifnet *p;
1711 	struct ifnet *ifp;
1712 	struct ifnet_hw_tsomax hw_tsomax;
1713 	int cap = 0, ena = 0, mena;
1714 	u_long hwa = 0;
1715 
1716 	NET_EPOCH_ASSERT();
1717 	VLAN_SXLOCK_ASSERT();
1718 
1719 	p = PARENT(ifv);
1720 	ifp = ifv->ifv_ifp;
1721 
1722 	/* Mask parent interface enabled capabilities disabled by user. */
1723 	mena = p->if_capenable & ifv->ifv_capenable;
1724 
1725 	/*
1726 	 * If the parent interface can do checksum offloading
1727 	 * on VLANs, then propagate its hardware-assisted
1728 	 * checksumming flags. Also assert that checksum
1729 	 * offloading requires hardware VLAN tagging.
1730 	 */
1731 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1732 		cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1733 	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1734 	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1735 		ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1736 		if (ena & IFCAP_TXCSUM)
1737 			hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
1738 			    CSUM_UDP | CSUM_SCTP);
1739 		if (ena & IFCAP_TXCSUM_IPV6)
1740 			hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
1741 			    CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
1742 	}
1743 
1744 	/*
1745 	 * If the parent interface can do TSO on VLANs then
1746 	 * propagate the hardware-assisted flag. TSO on VLANs
1747 	 * does not necessarily require hardware VLAN tagging.
1748 	 */
1749 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1750 	if_hw_tsomax_common(p, &hw_tsomax);
1751 	if_hw_tsomax_update(ifp, &hw_tsomax);
1752 	if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1753 		cap |= p->if_capabilities & IFCAP_TSO;
1754 	if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1755 		ena |= mena & IFCAP_TSO;
1756 		if (ena & IFCAP_TSO)
1757 			hwa |= p->if_hwassist & CSUM_TSO;
1758 	}
1759 
1760 	/*
1761 	 * If the parent interface can do LRO and checksum offloading on
1762 	 * VLANs, then guess it may do LRO on VLANs.  False positive here
1763 	 * cost nothing, while false negative may lead to some confusions.
1764 	 */
1765 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1766 		cap |= p->if_capabilities & IFCAP_LRO;
1767 	if (p->if_capenable & IFCAP_VLAN_HWCSUM)
1768 		ena |= p->if_capenable & IFCAP_LRO;
1769 
1770 	/*
1771 	 * If the parent interface can offload TCP connections over VLANs then
1772 	 * propagate its TOE capability to the VLAN interface.
1773 	 *
1774 	 * All TOE drivers in the tree today can deal with VLANs.  If this
1775 	 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1776 	 * with its own bit.
1777 	 */
1778 #define	IFCAP_VLAN_TOE IFCAP_TOE
1779 	if (p->if_capabilities & IFCAP_VLAN_TOE)
1780 		cap |= p->if_capabilities & IFCAP_TOE;
1781 	if (p->if_capenable & IFCAP_VLAN_TOE) {
1782 		TOEDEV(ifp) = TOEDEV(p);
1783 		ena |= mena & IFCAP_TOE;
1784 	}
1785 
1786 	/*
1787 	 * If the parent interface supports dynamic link state, so does the
1788 	 * VLAN interface.
1789 	 */
1790 	cap |= (p->if_capabilities & IFCAP_LINKSTATE);
1791 	ena |= (mena & IFCAP_LINKSTATE);
1792 
1793 #ifdef RATELIMIT
1794 	/*
1795 	 * If the parent interface supports ratelimiting, so does the
1796 	 * VLAN interface.
1797 	 */
1798 	cap |= (p->if_capabilities & IFCAP_TXRTLMT);
1799 	ena |= (mena & IFCAP_TXRTLMT);
1800 #endif
1801 
1802 	/*
1803 	 * If the parent interface supports unmapped mbufs, so does
1804 	 * the VLAN interface.  Note that this should be fine even for
1805 	 * interfaces that don't support hardware tagging as headers
1806 	 * are prepended in normal mbufs to unmapped mbufs holding
1807 	 * payload data.
1808 	 */
1809 	cap |= (p->if_capabilities & IFCAP_MEXTPG);
1810 	ena |= (mena & IFCAP_MEXTPG);
1811 
1812 	/*
1813 	 * If the parent interface can offload encryption and segmentation
1814 	 * of TLS records over TCP, propagate it's capability to the VLAN
1815 	 * interface.
1816 	 *
1817 	 * All TLS drivers in the tree today can deal with VLANs.  If
1818 	 * this ever changes, then a new IFCAP_VLAN_TXTLS can be
1819 	 * defined.
1820 	 */
1821 	if (p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
1822 		cap |= p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
1823 	if (p->if_capenable & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
1824 		ena |= mena & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
1825 
1826 	ifp->if_capabilities = cap;
1827 	ifp->if_capenable = ena;
1828 	ifp->if_hwassist = hwa;
1829 }
1830 
1831 static void
1832 vlan_trunk_capabilities(struct ifnet *ifp)
1833 {
1834 	struct epoch_tracker et;
1835 	struct ifvlantrunk *trunk;
1836 	struct ifvlan *ifv;
1837 
1838 	VLAN_SLOCK();
1839 	trunk = ifp->if_vlantrunk;
1840 	if (trunk == NULL) {
1841 		VLAN_SUNLOCK();
1842 		return;
1843 	}
1844 	NET_EPOCH_ENTER(et);
1845 	VLAN_FOREACH(ifv, trunk)
1846 		vlan_capabilities(ifv);
1847 	NET_EPOCH_EXIT(et);
1848 	VLAN_SUNLOCK();
1849 }
1850 
1851 static int
1852 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1853 {
1854 	struct ifnet *p;
1855 	struct ifreq *ifr;
1856 	struct ifaddr *ifa;
1857 	struct ifvlan *ifv;
1858 	struct ifvlantrunk *trunk;
1859 	struct vlanreq vlr;
1860 	int error = 0, oldmtu;
1861 
1862 	ifr = (struct ifreq *)data;
1863 	ifa = (struct ifaddr *) data;
1864 	ifv = ifp->if_softc;
1865 
1866 	switch (cmd) {
1867 	case SIOCSIFADDR:
1868 		ifp->if_flags |= IFF_UP;
1869 #ifdef INET
1870 		if (ifa->ifa_addr->sa_family == AF_INET)
1871 			arp_ifinit(ifp, ifa);
1872 #endif
1873 		break;
1874 	case SIOCGIFADDR:
1875 		bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
1876 		    ifp->if_addrlen);
1877 		break;
1878 	case SIOCGIFMEDIA:
1879 		VLAN_SLOCK();
1880 		if (TRUNK(ifv) != NULL) {
1881 			p = PARENT(ifv);
1882 			if_ref(p);
1883 			error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1884 			if_rele(p);
1885 			/* Limit the result to the parent's current config. */
1886 			if (error == 0) {
1887 				struct ifmediareq *ifmr;
1888 
1889 				ifmr = (struct ifmediareq *)data;
1890 				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1891 					ifmr->ifm_count = 1;
1892 					error = copyout(&ifmr->ifm_current,
1893 						ifmr->ifm_ulist,
1894 						sizeof(int));
1895 				}
1896 			}
1897 		} else {
1898 			error = EINVAL;
1899 		}
1900 		VLAN_SUNLOCK();
1901 		break;
1902 
1903 	case SIOCSIFMEDIA:
1904 		error = EINVAL;
1905 		break;
1906 
1907 	case SIOCSIFMTU:
1908 		/*
1909 		 * Set the interface MTU.
1910 		 */
1911 		VLAN_SLOCK();
1912 		trunk = TRUNK(ifv);
1913 		if (trunk != NULL) {
1914 			TRUNK_WLOCK(trunk);
1915 			if (ifr->ifr_mtu >
1916 			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1917 			    ifr->ifr_mtu <
1918 			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1919 				error = EINVAL;
1920 			else
1921 				ifp->if_mtu = ifr->ifr_mtu;
1922 			TRUNK_WUNLOCK(trunk);
1923 		} else
1924 			error = EINVAL;
1925 		VLAN_SUNLOCK();
1926 		break;
1927 
1928 	case SIOCSETVLAN:
1929 #ifdef VIMAGE
1930 		/*
1931 		 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1932 		 * interface to be delegated to a jail without allowing the
1933 		 * jail to change what underlying interface/VID it is
1934 		 * associated with.  We are not entirely convinced that this
1935 		 * is the right way to accomplish that policy goal.
1936 		 */
1937 		if (ifp->if_vnet != ifp->if_home_vnet) {
1938 			error = EPERM;
1939 			break;
1940 		}
1941 #endif
1942 		error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
1943 		if (error)
1944 			break;
1945 		if (vlr.vlr_parent[0] == '\0') {
1946 			vlan_unconfig(ifp);
1947 			break;
1948 		}
1949 		p = ifunit_ref(vlr.vlr_parent);
1950 		if (p == NULL) {
1951 			error = ENOENT;
1952 			break;
1953 		}
1954 #ifdef COMPAT_FREEBSD12
1955 		if (vlr.vlr_proto == 0)
1956 			vlr.vlr_proto = ETHERTYPE_VLAN;
1957 #endif
1958 		oldmtu = ifp->if_mtu;
1959 		error = vlan_config(ifv, p, vlr.vlr_tag, vlr.vlr_proto);
1960 		if_rele(p);
1961 
1962 		/*
1963 		 * VLAN MTU may change during addition of the vlandev.
1964 		 * If it did, do network layer specific procedure.
1965 		 */
1966 		if (ifp->if_mtu != oldmtu) {
1967 #ifdef INET6
1968 			nd6_setmtu(ifp);
1969 #endif
1970 			rt_updatemtu(ifp);
1971 		}
1972 		break;
1973 
1974 	case SIOCGETVLAN:
1975 #ifdef VIMAGE
1976 		if (ifp->if_vnet != ifp->if_home_vnet) {
1977 			error = EPERM;
1978 			break;
1979 		}
1980 #endif
1981 		bzero(&vlr, sizeof(vlr));
1982 		VLAN_SLOCK();
1983 		if (TRUNK(ifv) != NULL) {
1984 			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1985 			    sizeof(vlr.vlr_parent));
1986 			vlr.vlr_tag = ifv->ifv_vid;
1987 			vlr.vlr_proto = ifv->ifv_proto;
1988 		}
1989 		VLAN_SUNLOCK();
1990 		error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
1991 		break;
1992 
1993 	case SIOCSIFFLAGS:
1994 		/*
1995 		 * We should propagate selected flags to the parent,
1996 		 * e.g., promiscuous mode.
1997 		 */
1998 		VLAN_XLOCK();
1999 		if (TRUNK(ifv) != NULL)
2000 			error = vlan_setflags(ifp, 1);
2001 		VLAN_XUNLOCK();
2002 		break;
2003 
2004 	case SIOCADDMULTI:
2005 	case SIOCDELMULTI:
2006 		/*
2007 		 * If we don't have a parent, just remember the membership for
2008 		 * when we do.
2009 		 *
2010 		 * XXX We need the rmlock here to avoid sleeping while
2011 		 * holding in6_multi_mtx.
2012 		 */
2013 		VLAN_XLOCK();
2014 		trunk = TRUNK(ifv);
2015 		if (trunk != NULL)
2016 			error = vlan_setmulti(ifp);
2017 		VLAN_XUNLOCK();
2018 
2019 		break;
2020 	case SIOCGVLANPCP:
2021 #ifdef VIMAGE
2022 		if (ifp->if_vnet != ifp->if_home_vnet) {
2023 			error = EPERM;
2024 			break;
2025 		}
2026 #endif
2027 		ifr->ifr_vlan_pcp = ifv->ifv_pcp;
2028 		break;
2029 
2030 	case SIOCSVLANPCP:
2031 #ifdef VIMAGE
2032 		if (ifp->if_vnet != ifp->if_home_vnet) {
2033 			error = EPERM;
2034 			break;
2035 		}
2036 #endif
2037 		error = priv_check(curthread, PRIV_NET_SETVLANPCP);
2038 		if (error)
2039 			break;
2040 		if (ifr->ifr_vlan_pcp > 7) {
2041 			error = EINVAL;
2042 			break;
2043 		}
2044 		ifv->ifv_pcp = ifr->ifr_vlan_pcp;
2045 		ifp->if_pcp = ifv->ifv_pcp;
2046 		/* broadcast event about PCP change */
2047 		EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
2048 		break;
2049 
2050 	case SIOCSIFCAP:
2051 		VLAN_SLOCK();
2052 		ifv->ifv_capenable = ifr->ifr_reqcap;
2053 		trunk = TRUNK(ifv);
2054 		if (trunk != NULL) {
2055 			struct epoch_tracker et;
2056 
2057 			NET_EPOCH_ENTER(et);
2058 			vlan_capabilities(ifv);
2059 			NET_EPOCH_EXIT(et);
2060 		}
2061 		VLAN_SUNLOCK();
2062 		break;
2063 
2064 	default:
2065 		error = EINVAL;
2066 		break;
2067 	}
2068 
2069 	return (error);
2070 }
2071 
2072 #if defined(KERN_TLS) || defined(RATELIMIT)
2073 static int
2074 vlan_snd_tag_alloc(struct ifnet *ifp,
2075     union if_snd_tag_alloc_params *params,
2076     struct m_snd_tag **ppmt)
2077 {
2078 	struct epoch_tracker et;
2079 	struct vlan_snd_tag *vst;
2080 	struct ifvlan *ifv;
2081 	struct ifnet *parent;
2082 	int error;
2083 
2084 	NET_EPOCH_ENTER(et);
2085 	ifv = ifp->if_softc;
2086 	if (ifv->ifv_trunk != NULL)
2087 		parent = PARENT(ifv);
2088 	else
2089 		parent = NULL;
2090 	if (parent == NULL) {
2091 		NET_EPOCH_EXIT(et);
2092 		return (EOPNOTSUPP);
2093 	}
2094 	if_ref(parent);
2095 	NET_EPOCH_EXIT(et);
2096 
2097 	vst = malloc(sizeof(*vst), M_VLAN, M_NOWAIT);
2098 	if (vst == NULL) {
2099 		if_rele(parent);
2100 		return (ENOMEM);
2101 	}
2102 
2103 	error = m_snd_tag_alloc(parent, params, &vst->tag);
2104 	if_rele(parent);
2105 	if (error) {
2106 		free(vst, M_VLAN);
2107 		return (error);
2108 	}
2109 
2110 	m_snd_tag_init(&vst->com, ifp, vst->tag->type);
2111 
2112 	*ppmt = &vst->com;
2113 	return (0);
2114 }
2115 
2116 static struct m_snd_tag *
2117 vlan_next_snd_tag(struct m_snd_tag *mst)
2118 {
2119 	struct vlan_snd_tag *vst;
2120 
2121 	vst = mst_to_vst(mst);
2122 	return (vst->tag);
2123 }
2124 
2125 static int
2126 vlan_snd_tag_modify(struct m_snd_tag *mst,
2127     union if_snd_tag_modify_params *params)
2128 {
2129 	struct vlan_snd_tag *vst;
2130 
2131 	vst = mst_to_vst(mst);
2132 	return (vst->tag->ifp->if_snd_tag_modify(vst->tag, params));
2133 }
2134 
2135 static int
2136 vlan_snd_tag_query(struct m_snd_tag *mst,
2137     union if_snd_tag_query_params *params)
2138 {
2139 	struct vlan_snd_tag *vst;
2140 
2141 	vst = mst_to_vst(mst);
2142 	return (vst->tag->ifp->if_snd_tag_query(vst->tag, params));
2143 }
2144 
2145 static void
2146 vlan_snd_tag_free(struct m_snd_tag *mst)
2147 {
2148 	struct vlan_snd_tag *vst;
2149 
2150 	vst = mst_to_vst(mst);
2151 	m_snd_tag_rele(vst->tag);
2152 	free(vst, M_VLAN);
2153 }
2154 
2155 static void
2156 vlan_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
2157 {
2158 	/*
2159 	 * For vlan, we have an indirect
2160 	 * interface. The caller needs to
2161 	 * get a ratelimit tag on the actual
2162 	 * interface the flow will go on.
2163 	 */
2164 	q->rate_table = NULL;
2165 	q->flags = RT_IS_INDIRECT;
2166 	q->max_flows = 0;
2167 	q->number_of_rates = 0;
2168 }
2169 
2170 #endif
2171