xref: /freebsd/sys/net/if_vlan.c (revision 6b2c1e49da284f28ec7b52f7c031474087e37104)
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_kern_tls.h"
50 #include "opt_vlan.h"
51 #include "opt_ratelimit.h"
52 
53 #include <sys/param.h>
54 #include <sys/eventhandler.h>
55 #include <sys/kernel.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.h>
59 #include <sys/module.h>
60 #include <sys/rmlock.h>
61 #include <sys/priv.h>
62 #include <sys/queue.h>
63 #include <sys/socket.h>
64 #include <sys/sockio.h>
65 #include <sys/sysctl.h>
66 #include <sys/systm.h>
67 #include <sys/sx.h>
68 #include <sys/taskqueue.h>
69 
70 #include <net/bpf.h>
71 #include <net/ethernet.h>
72 #include <net/if.h>
73 #include <net/if_var.h>
74 #include <net/if_clone.h>
75 #include <net/if_dl.h>
76 #include <net/if_types.h>
77 #include <net/if_vlan_var.h>
78 #include <net/vnet.h>
79 
80 #ifdef INET
81 #include <netinet/in.h>
82 #include <netinet/if_ether.h>
83 #endif
84 
85 #define	VLAN_DEF_HWIDTH	4
86 #define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
87 
88 #define	UP_AND_RUNNING(ifp) \
89     ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
90 
91 CK_SLIST_HEAD(ifvlanhead, ifvlan);
92 
93 struct ifvlantrunk {
94 	struct	ifnet   *parent;	/* parent interface of this trunk */
95 	struct	mtx	lock;
96 #ifdef VLAN_ARRAY
97 #define	VLAN_ARRAY_SIZE	(EVL_VLID_MASK + 1)
98 	struct	ifvlan	*vlans[VLAN_ARRAY_SIZE]; /* static table */
99 #else
100 	struct	ifvlanhead *hash;	/* dynamic hash-list table */
101 	uint16_t	hmask;
102 	uint16_t	hwidth;
103 #endif
104 	int		refcnt;
105 };
106 
107 #if defined(KERN_TLS) || defined(RATELIMIT)
108 struct vlan_snd_tag {
109 	struct m_snd_tag com;
110 	struct m_snd_tag *tag;
111 };
112 
113 static inline struct vlan_snd_tag *
114 mst_to_vst(struct m_snd_tag *mst)
115 {
116 
117 	return (__containerof(mst, struct vlan_snd_tag, com));
118 }
119 #endif
120 
121 /*
122  * This macro provides a facility to iterate over every vlan on a trunk with
123  * the assumption that none will be added/removed during iteration.
124  */
125 #ifdef VLAN_ARRAY
126 #define VLAN_FOREACH(_ifv, _trunk) \
127 	size_t _i; \
128 	for (_i = 0; _i < VLAN_ARRAY_SIZE; _i++) \
129 		if (((_ifv) = (_trunk)->vlans[_i]) != NULL)
130 #else /* VLAN_ARRAY */
131 #define VLAN_FOREACH(_ifv, _trunk) \
132 	struct ifvlan *_next; \
133 	size_t _i; \
134 	for (_i = 0; _i < (1 << (_trunk)->hwidth); _i++) \
135 		CK_SLIST_FOREACH_SAFE((_ifv), &(_trunk)->hash[_i], ifv_list, _next)
136 #endif /* VLAN_ARRAY */
137 
138 /*
139  * This macro provides a facility to iterate over every vlan on a trunk while
140  * also modifying the number of vlans on the trunk. The iteration continues
141  * until some condition is met or there are no more vlans on the trunk.
142  */
143 #ifdef VLAN_ARRAY
144 /* The VLAN_ARRAY case is simple -- just a for loop using the condition. */
145 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
146 	size_t _i; \
147 	for (_i = 0; !(_cond) && _i < VLAN_ARRAY_SIZE; _i++) \
148 		if (((_ifv) = (_trunk)->vlans[_i]))
149 #else /* VLAN_ARRAY */
150 /*
151  * The hash table case is more complicated. We allow for the hash table to be
152  * modified (i.e. vlans removed) while we are iterating over it. To allow for
153  * this we must restart the iteration every time we "touch" something during
154  * the iteration, since removal will resize the hash table and invalidate our
155  * current position. If acting on the touched element causes the trunk to be
156  * emptied, then iteration also stops.
157  */
158 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
159 	size_t _i; \
160 	bool _touch = false; \
161 	for (_i = 0; \
162 	    !(_cond) && _i < (1 << (_trunk)->hwidth); \
163 	    _i = (_touch && ((_trunk) != NULL) ? 0 : _i + 1), _touch = false) \
164 		if (((_ifv) = CK_SLIST_FIRST(&(_trunk)->hash[_i])) != NULL && \
165 		    (_touch = true))
166 #endif /* VLAN_ARRAY */
167 
168 struct vlan_mc_entry {
169 	struct sockaddr_dl		mc_addr;
170 	CK_SLIST_ENTRY(vlan_mc_entry)	mc_entries;
171 	struct epoch_context		mc_epoch_ctx;
172 };
173 
174 struct ifvlan {
175 	struct	ifvlantrunk *ifv_trunk;
176 	struct	ifnet *ifv_ifp;
177 #define	TRUNK(ifv)	((ifv)->ifv_trunk)
178 #define	PARENT(ifv)	((ifv)->ifv_trunk->parent)
179 	void	*ifv_cookie;
180 	int	ifv_pflags;	/* special flags we have set on parent */
181 	int	ifv_capenable;
182 	int	ifv_encaplen;	/* encapsulation length */
183 	int	ifv_mtufudge;	/* MTU fudged by this much */
184 	int	ifv_mintu;	/* min transmission unit */
185 	uint16_t ifv_proto;	/* encapsulation ethertype */
186 	uint16_t ifv_tag;	/* tag to apply on packets leaving if */
187 	uint16_t ifv_vid;	/* VLAN ID */
188 	uint8_t	ifv_pcp;	/* Priority Code Point (PCP). */
189 	struct task lladdr_task;
190 	CK_SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
191 #ifndef VLAN_ARRAY
192 	CK_SLIST_ENTRY(ifvlan) ifv_list;
193 #endif
194 };
195 
196 /* Special flags we should propagate to parent. */
197 static struct {
198 	int flag;
199 	int (*func)(struct ifnet *, int);
200 } vlan_pflags[] = {
201 	{IFF_PROMISC, ifpromisc},
202 	{IFF_ALLMULTI, if_allmulti},
203 	{0, NULL}
204 };
205 
206 extern int vlan_mtag_pcp;
207 
208 static const char vlanname[] = "vlan";
209 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
210 
211 static eventhandler_tag ifdetach_tag;
212 static eventhandler_tag iflladdr_tag;
213 
214 /*
215  * if_vlan uses two module-level synchronizations primitives to allow concurrent
216  * modification of vlan interfaces and (mostly) allow for vlans to be destroyed
217  * while they are being used for tx/rx. To accomplish this in a way that has
218  * acceptable performance and cooperation with other parts of the network stack
219  * there is a non-sleepable epoch(9) and an sx(9).
220  *
221  * The performance-sensitive paths that warrant using the epoch(9) are
222  * vlan_transmit and vlan_input. Both have to check for the vlan interface's
223  * existence using if_vlantrunk, and being in the network tx/rx paths the use
224  * of an epoch(9) gives a measureable improvement in performance.
225  *
226  * The reason for having an sx(9) is mostly because there are still areas that
227  * must be sleepable and also have safe concurrent access to a vlan interface.
228  * Since the sx(9) exists, it is used by default in most paths unless sleeping
229  * is not permitted, or if it is not clear whether sleeping is permitted.
230  *
231  */
232 #define _VLAN_SX_ID ifv_sx
233 
234 static struct sx _VLAN_SX_ID;
235 
236 #define VLAN_LOCKING_INIT() \
237 	sx_init(&_VLAN_SX_ID, "vlan_sx")
238 
239 #define VLAN_LOCKING_DESTROY() \
240 	sx_destroy(&_VLAN_SX_ID)
241 
242 #define	VLAN_SLOCK()			sx_slock(&_VLAN_SX_ID)
243 #define	VLAN_SUNLOCK()			sx_sunlock(&_VLAN_SX_ID)
244 #define	VLAN_XLOCK()			sx_xlock(&_VLAN_SX_ID)
245 #define	VLAN_XUNLOCK()			sx_xunlock(&_VLAN_SX_ID)
246 #define	VLAN_SLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_SLOCKED)
247 #define	VLAN_XLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_XLOCKED)
248 #define	VLAN_SXLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_LOCKED)
249 
250 /*
251  * We also have a per-trunk mutex that should be acquired when changing
252  * its state.
253  */
254 #define	TRUNK_LOCK_INIT(trunk)		mtx_init(&(trunk)->lock, vlanname, NULL, MTX_DEF)
255 #define	TRUNK_LOCK_DESTROY(trunk)	mtx_destroy(&(trunk)->lock)
256 #define	TRUNK_WLOCK(trunk)		mtx_lock(&(trunk)->lock)
257 #define	TRUNK_WUNLOCK(trunk)		mtx_unlock(&(trunk)->lock)
258 #define	TRUNK_WLOCK_ASSERT(trunk)	mtx_assert(&(trunk)->lock, MA_OWNED);
259 
260 /*
261  * The VLAN_ARRAY substitutes the dynamic hash with a static array
262  * with 4096 entries. In theory this can give a boost in processing,
263  * however in practice it does not. Probably this is because the array
264  * is too big to fit into CPU cache.
265  */
266 #ifndef VLAN_ARRAY
267 static	void vlan_inithash(struct ifvlantrunk *trunk);
268 static	void vlan_freehash(struct ifvlantrunk *trunk);
269 static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
270 static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
271 static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
272 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
273 	uint16_t vid);
274 #endif
275 static	void trunk_destroy(struct ifvlantrunk *trunk);
276 
277 static	void vlan_init(void *foo);
278 static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
279 static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
280 #if defined(KERN_TLS) || defined(RATELIMIT)
281 static	int vlan_snd_tag_alloc(struct ifnet *,
282     union if_snd_tag_alloc_params *, struct m_snd_tag **);
283 static	int vlan_snd_tag_modify(struct m_snd_tag *,
284     union if_snd_tag_modify_params *);
285 static	int vlan_snd_tag_query(struct m_snd_tag *,
286     union if_snd_tag_query_params *);
287 static	void vlan_snd_tag_free(struct m_snd_tag *);
288 #endif
289 static	void vlan_qflush(struct ifnet *ifp);
290 static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
291     int (*func)(struct ifnet *, int));
292 static	int vlan_setflags(struct ifnet *ifp, int status);
293 static	int vlan_setmulti(struct ifnet *ifp);
294 static	int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
295 static	int vlan_output(struct ifnet *ifp, struct mbuf *m,
296     const struct sockaddr *dst, struct route *ro);
297 static	void vlan_unconfig(struct ifnet *ifp);
298 static	void vlan_unconfig_locked(struct ifnet *ifp, int departing);
299 static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
300 static	void vlan_link_state(struct ifnet *ifp);
301 static	void vlan_capabilities(struct ifvlan *ifv);
302 static	void vlan_trunk_capabilities(struct ifnet *ifp);
303 
304 static	struct ifnet *vlan_clone_match_ethervid(const char *, int *);
305 static	int vlan_clone_match(struct if_clone *, const char *);
306 static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
307 static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
308 
309 static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
310 static  void vlan_iflladdr(void *arg, struct ifnet *ifp);
311 
312 static  void vlan_lladdr_fn(void *arg, int pending);
313 
314 static struct if_clone *vlan_cloner;
315 
316 #ifdef VIMAGE
317 VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner);
318 #define	V_vlan_cloner	VNET(vlan_cloner)
319 #endif
320 
321 static void
322 vlan_mc_free(struct epoch_context *ctx)
323 {
324 	struct vlan_mc_entry *mc = __containerof(ctx, struct vlan_mc_entry, mc_epoch_ctx);
325 	free(mc, M_VLAN);
326 }
327 
328 #ifndef VLAN_ARRAY
329 #define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
330 
331 static void
332 vlan_inithash(struct ifvlantrunk *trunk)
333 {
334 	int i, n;
335 
336 	/*
337 	 * The trunk must not be locked here since we call malloc(M_WAITOK).
338 	 * It is OK in case this function is called before the trunk struct
339 	 * gets hooked up and becomes visible from other threads.
340 	 */
341 
342 	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
343 	    ("%s: hash already initialized", __func__));
344 
345 	trunk->hwidth = VLAN_DEF_HWIDTH;
346 	n = 1 << trunk->hwidth;
347 	trunk->hmask = n - 1;
348 	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
349 	for (i = 0; i < n; i++)
350 		CK_SLIST_INIT(&trunk->hash[i]);
351 }
352 
353 static void
354 vlan_freehash(struct ifvlantrunk *trunk)
355 {
356 #ifdef INVARIANTS
357 	int i;
358 
359 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
360 	for (i = 0; i < (1 << trunk->hwidth); i++)
361 		KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]),
362 		    ("%s: hash table not empty", __func__));
363 #endif
364 	free(trunk->hash, M_VLAN);
365 	trunk->hash = NULL;
366 	trunk->hwidth = trunk->hmask = 0;
367 }
368 
369 static int
370 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
371 {
372 	int i, b;
373 	struct ifvlan *ifv2;
374 
375 	VLAN_XLOCK_ASSERT();
376 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
377 
378 	b = 1 << trunk->hwidth;
379 	i = HASH(ifv->ifv_vid, trunk->hmask);
380 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
381 		if (ifv->ifv_vid == ifv2->ifv_vid)
382 			return (EEXIST);
383 
384 	/*
385 	 * Grow the hash when the number of vlans exceeds half of the number of
386 	 * hash buckets squared. This will make the average linked-list length
387 	 * buckets/2.
388 	 */
389 	if (trunk->refcnt > (b * b) / 2) {
390 		vlan_growhash(trunk, 1);
391 		i = HASH(ifv->ifv_vid, trunk->hmask);
392 	}
393 	CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
394 	trunk->refcnt++;
395 
396 	return (0);
397 }
398 
399 static int
400 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
401 {
402 	int i, b;
403 	struct ifvlan *ifv2;
404 
405 	VLAN_XLOCK_ASSERT();
406 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
407 
408 	b = 1 << trunk->hwidth;
409 	i = HASH(ifv->ifv_vid, trunk->hmask);
410 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
411 		if (ifv2 == ifv) {
412 			trunk->refcnt--;
413 			CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list);
414 			if (trunk->refcnt < (b * b) / 2)
415 				vlan_growhash(trunk, -1);
416 			return (0);
417 		}
418 
419 	panic("%s: vlan not found\n", __func__);
420 	return (ENOENT); /*NOTREACHED*/
421 }
422 
423 /*
424  * Grow the hash larger or smaller if memory permits.
425  */
426 static void
427 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
428 {
429 	struct ifvlan *ifv;
430 	struct ifvlanhead *hash2;
431 	int hwidth2, i, j, n, n2;
432 
433 	VLAN_XLOCK_ASSERT();
434 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
435 
436 	if (howmuch == 0) {
437 		/* Harmless yet obvious coding error */
438 		printf("%s: howmuch is 0\n", __func__);
439 		return;
440 	}
441 
442 	hwidth2 = trunk->hwidth + howmuch;
443 	n = 1 << trunk->hwidth;
444 	n2 = 1 << hwidth2;
445 	/* Do not shrink the table below the default */
446 	if (hwidth2 < VLAN_DEF_HWIDTH)
447 		return;
448 
449 	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK);
450 	if (hash2 == NULL) {
451 		printf("%s: out of memory -- hash size not changed\n",
452 		    __func__);
453 		return;		/* We can live with the old hash table */
454 	}
455 	for (j = 0; j < n2; j++)
456 		CK_SLIST_INIT(&hash2[j]);
457 	for (i = 0; i < n; i++)
458 		while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) {
459 			CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list);
460 			j = HASH(ifv->ifv_vid, n2 - 1);
461 			CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
462 		}
463 	NET_EPOCH_WAIT();
464 	free(trunk->hash, M_VLAN);
465 	trunk->hash = hash2;
466 	trunk->hwidth = hwidth2;
467 	trunk->hmask = n2 - 1;
468 
469 	if (bootverbose)
470 		if_printf(trunk->parent,
471 		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
472 }
473 
474 static __inline struct ifvlan *
475 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
476 {
477 	struct ifvlan *ifv;
478 
479 	NET_EPOCH_ASSERT();
480 
481 	CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
482 		if (ifv->ifv_vid == vid)
483 			return (ifv);
484 	return (NULL);
485 }
486 
487 #if 0
488 /* Debugging code to view the hashtables. */
489 static void
490 vlan_dumphash(struct ifvlantrunk *trunk)
491 {
492 	int i;
493 	struct ifvlan *ifv;
494 
495 	for (i = 0; i < (1 << trunk->hwidth); i++) {
496 		printf("%d: ", i);
497 		CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
498 			printf("%s ", ifv->ifv_ifp->if_xname);
499 		printf("\n");
500 	}
501 }
502 #endif /* 0 */
503 #else
504 
505 static __inline struct ifvlan *
506 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
507 {
508 
509 	return trunk->vlans[vid];
510 }
511 
512 static __inline int
513 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
514 {
515 
516 	if (trunk->vlans[ifv->ifv_vid] != NULL)
517 		return EEXIST;
518 	trunk->vlans[ifv->ifv_vid] = ifv;
519 	trunk->refcnt++;
520 
521 	return (0);
522 }
523 
524 static __inline int
525 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
526 {
527 
528 	trunk->vlans[ifv->ifv_vid] = NULL;
529 	trunk->refcnt--;
530 
531 	return (0);
532 }
533 
534 static __inline void
535 vlan_freehash(struct ifvlantrunk *trunk)
536 {
537 }
538 
539 static __inline void
540 vlan_inithash(struct ifvlantrunk *trunk)
541 {
542 }
543 
544 #endif /* !VLAN_ARRAY */
545 
546 static void
547 trunk_destroy(struct ifvlantrunk *trunk)
548 {
549 	VLAN_XLOCK_ASSERT();
550 
551 	vlan_freehash(trunk);
552 	trunk->parent->if_vlantrunk = NULL;
553 	TRUNK_LOCK_DESTROY(trunk);
554 	if_rele(trunk->parent);
555 	free(trunk, M_VLAN);
556 }
557 
558 /*
559  * Program our multicast filter. What we're actually doing is
560  * programming the multicast filter of the parent. This has the
561  * side effect of causing the parent interface to receive multicast
562  * traffic that it doesn't really want, which ends up being discarded
563  * later by the upper protocol layers. Unfortunately, there's no way
564  * to avoid this: there really is only one physical interface.
565  */
566 static int
567 vlan_setmulti(struct ifnet *ifp)
568 {
569 	struct ifnet		*ifp_p;
570 	struct ifmultiaddr	*ifma;
571 	struct ifvlan		*sc;
572 	struct vlan_mc_entry	*mc;
573 	int			error;
574 
575 	VLAN_XLOCK_ASSERT();
576 
577 	/* Find the parent. */
578 	sc = ifp->if_softc;
579 	ifp_p = PARENT(sc);
580 
581 	CURVNET_SET_QUIET(ifp_p->if_vnet);
582 
583 	/* First, remove any existing filter entries. */
584 	while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
585 		CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
586 		(void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
587 		epoch_call(net_epoch_preempt, &mc->mc_epoch_ctx, vlan_mc_free);
588 	}
589 
590 	/* Now program new ones. */
591 	IF_ADDR_WLOCK(ifp);
592 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
593 		if (ifma->ifma_addr->sa_family != AF_LINK)
594 			continue;
595 		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
596 		if (mc == NULL) {
597 			IF_ADDR_WUNLOCK(ifp);
598 			return (ENOMEM);
599 		}
600 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
601 		mc->mc_addr.sdl_index = ifp_p->if_index;
602 		CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
603 	}
604 	IF_ADDR_WUNLOCK(ifp);
605 	CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
606 		error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
607 		    NULL);
608 		if (error)
609 			return (error);
610 	}
611 
612 	CURVNET_RESTORE();
613 	return (0);
614 }
615 
616 /*
617  * A handler for parent interface link layer address changes.
618  * If the parent interface link layer address is changed we
619  * should also change it on all children vlans.
620  */
621 static void
622 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
623 {
624 	struct epoch_tracker et;
625 	struct ifvlan *ifv;
626 	struct ifnet *ifv_ifp;
627 	struct ifvlantrunk *trunk;
628 	struct sockaddr_dl *sdl;
629 
630 	/* Need the epoch since this is run on taskqueue_swi. */
631 	NET_EPOCH_ENTER(et);
632 	trunk = ifp->if_vlantrunk;
633 	if (trunk == NULL) {
634 		NET_EPOCH_EXIT(et);
635 		return;
636 	}
637 
638 	/*
639 	 * OK, it's a trunk.  Loop over and change all vlan's lladdrs on it.
640 	 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
641 	 * ioctl calls on the parent garbling the lladdr of the child vlan.
642 	 */
643 	TRUNK_WLOCK(trunk);
644 	VLAN_FOREACH(ifv, trunk) {
645 		/*
646 		 * Copy new new lladdr into the ifv_ifp, enqueue a task
647 		 * to actually call if_setlladdr. if_setlladdr needs to
648 		 * be deferred to a taskqueue because it will call into
649 		 * the if_vlan ioctl path and try to acquire the global
650 		 * lock.
651 		 */
652 		ifv_ifp = ifv->ifv_ifp;
653 		bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
654 		    ifp->if_addrlen);
655 		sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
656 		sdl->sdl_alen = ifp->if_addrlen;
657 		taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
658 	}
659 	TRUNK_WUNLOCK(trunk);
660 	NET_EPOCH_EXIT(et);
661 }
662 
663 /*
664  * A handler for network interface departure events.
665  * Track departure of trunks here so that we don't access invalid
666  * pointers or whatever if a trunk is ripped from under us, e.g.,
667  * by ejecting its hot-plug card.  However, if an ifnet is simply
668  * being renamed, then there's no need to tear down the state.
669  */
670 static void
671 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
672 {
673 	struct ifvlan *ifv;
674 	struct ifvlantrunk *trunk;
675 
676 	/* If the ifnet is just being renamed, don't do anything. */
677 	if (ifp->if_flags & IFF_RENAMING)
678 		return;
679 	VLAN_XLOCK();
680 	trunk = ifp->if_vlantrunk;
681 	if (trunk == NULL) {
682 		VLAN_XUNLOCK();
683 		return;
684 	}
685 
686 	/*
687 	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
688 	 * Check trunk pointer after each vlan_unconfig() as it will
689 	 * free it and set to NULL after the last vlan was detached.
690 	 */
691 	VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
692 	    ifp->if_vlantrunk == NULL)
693 		vlan_unconfig_locked(ifv->ifv_ifp, 1);
694 
695 	/* Trunk should have been destroyed in vlan_unconfig(). */
696 	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
697 	VLAN_XUNLOCK();
698 }
699 
700 /*
701  * Return the trunk device for a virtual interface.
702  */
703 static struct ifnet  *
704 vlan_trunkdev(struct ifnet *ifp)
705 {
706 	struct ifvlan *ifv;
707 
708 	NET_EPOCH_ASSERT();
709 
710 	if (ifp->if_type != IFT_L2VLAN)
711 		return (NULL);
712 
713 	ifv = ifp->if_softc;
714 	ifp = NULL;
715 	if (ifv->ifv_trunk)
716 		ifp = PARENT(ifv);
717 	return (ifp);
718 }
719 
720 /*
721  * Return the 12-bit VLAN VID for this interface, for use by external
722  * components such as Infiniband.
723  *
724  * XXXRW: Note that the function name here is historical; it should be named
725  * vlan_vid().
726  */
727 static int
728 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
729 {
730 	struct ifvlan *ifv;
731 
732 	if (ifp->if_type != IFT_L2VLAN)
733 		return (EINVAL);
734 	ifv = ifp->if_softc;
735 	*vidp = ifv->ifv_vid;
736 	return (0);
737 }
738 
739 static int
740 vlan_pcp(struct ifnet *ifp, uint16_t *pcpp)
741 {
742 	struct ifvlan *ifv;
743 
744 	if (ifp->if_type != IFT_L2VLAN)
745 		return (EINVAL);
746 	ifv = ifp->if_softc;
747 	*pcpp = ifv->ifv_pcp;
748 	return (0);
749 }
750 
751 /*
752  * Return a driver specific cookie for this interface.  Synchronization
753  * with setcookie must be provided by the driver.
754  */
755 static void *
756 vlan_cookie(struct ifnet *ifp)
757 {
758 	struct ifvlan *ifv;
759 
760 	if (ifp->if_type != IFT_L2VLAN)
761 		return (NULL);
762 	ifv = ifp->if_softc;
763 	return (ifv->ifv_cookie);
764 }
765 
766 /*
767  * Store a cookie in our softc that drivers can use to store driver
768  * private per-instance data in.
769  */
770 static int
771 vlan_setcookie(struct ifnet *ifp, void *cookie)
772 {
773 	struct ifvlan *ifv;
774 
775 	if (ifp->if_type != IFT_L2VLAN)
776 		return (EINVAL);
777 	ifv = ifp->if_softc;
778 	ifv->ifv_cookie = cookie;
779 	return (0);
780 }
781 
782 /*
783  * Return the vlan device present at the specific VID.
784  */
785 static struct ifnet *
786 vlan_devat(struct ifnet *ifp, uint16_t vid)
787 {
788 	struct ifvlantrunk *trunk;
789 	struct ifvlan *ifv;
790 
791 	NET_EPOCH_ASSERT();
792 
793 	trunk = ifp->if_vlantrunk;
794 	if (trunk == NULL)
795 		return (NULL);
796 	ifp = NULL;
797 	ifv = vlan_gethash(trunk, vid);
798 	if (ifv)
799 		ifp = ifv->ifv_ifp;
800 	return (ifp);
801 }
802 
803 /*
804  * Recalculate the cached VLAN tag exposed via the MIB.
805  */
806 static void
807 vlan_tag_recalculate(struct ifvlan *ifv)
808 {
809 
810        ifv->ifv_tag = EVL_MAKETAG(ifv->ifv_vid, ifv->ifv_pcp, 0);
811 }
812 
813 /*
814  * VLAN support can be loaded as a module.  The only place in the
815  * system that's intimately aware of this is ether_input.  We hook
816  * into this code through vlan_input_p which is defined there and
817  * set here.  No one else in the system should be aware of this so
818  * we use an explicit reference here.
819  */
820 extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
821 
822 /* For if_link_state_change() eyes only... */
823 extern	void (*vlan_link_state_p)(struct ifnet *);
824 
825 static int
826 vlan_modevent(module_t mod, int type, void *data)
827 {
828 
829 	switch (type) {
830 	case MOD_LOAD:
831 		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
832 		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
833 		if (ifdetach_tag == NULL)
834 			return (ENOMEM);
835 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
836 		    vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
837 		if (iflladdr_tag == NULL)
838 			return (ENOMEM);
839 		VLAN_LOCKING_INIT();
840 		vlan_input_p = vlan_input;
841 		vlan_link_state_p = vlan_link_state;
842 		vlan_trunk_cap_p = vlan_trunk_capabilities;
843 		vlan_trunkdev_p = vlan_trunkdev;
844 		vlan_cookie_p = vlan_cookie;
845 		vlan_setcookie_p = vlan_setcookie;
846 		vlan_tag_p = vlan_tag;
847 		vlan_pcp_p = vlan_pcp;
848 		vlan_devat_p = vlan_devat;
849 #ifndef VIMAGE
850 		vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
851 		    vlan_clone_create, vlan_clone_destroy);
852 #endif
853 		if (bootverbose)
854 			printf("vlan: initialized, using "
855 #ifdef VLAN_ARRAY
856 			       "full-size arrays"
857 #else
858 			       "hash tables with chaining"
859 #endif
860 
861 			       "\n");
862 		break;
863 	case MOD_UNLOAD:
864 #ifndef VIMAGE
865 		if_clone_detach(vlan_cloner);
866 #endif
867 		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
868 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
869 		vlan_input_p = NULL;
870 		vlan_link_state_p = NULL;
871 		vlan_trunk_cap_p = NULL;
872 		vlan_trunkdev_p = NULL;
873 		vlan_tag_p = NULL;
874 		vlan_cookie_p = NULL;
875 		vlan_setcookie_p = NULL;
876 		vlan_devat_p = NULL;
877 		VLAN_LOCKING_DESTROY();
878 		if (bootverbose)
879 			printf("vlan: unloaded\n");
880 		break;
881 	default:
882 		return (EOPNOTSUPP);
883 	}
884 	return (0);
885 }
886 
887 static moduledata_t vlan_mod = {
888 	"if_vlan",
889 	vlan_modevent,
890 	0
891 };
892 
893 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
894 MODULE_VERSION(if_vlan, 3);
895 
896 #ifdef VIMAGE
897 static void
898 vnet_vlan_init(const void *unused __unused)
899 {
900 
901 	vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
902 		    vlan_clone_create, vlan_clone_destroy);
903 	V_vlan_cloner = vlan_cloner;
904 }
905 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
906     vnet_vlan_init, NULL);
907 
908 static void
909 vnet_vlan_uninit(const void *unused __unused)
910 {
911 
912 	if_clone_detach(V_vlan_cloner);
913 }
914 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
915     vnet_vlan_uninit, NULL);
916 #endif
917 
918 /*
919  * Check for <etherif>.<vlan> style interface names.
920  */
921 static struct ifnet *
922 vlan_clone_match_ethervid(const char *name, int *vidp)
923 {
924 	char ifname[IFNAMSIZ];
925 	char *cp;
926 	struct ifnet *ifp;
927 	int vid;
928 
929 	strlcpy(ifname, name, IFNAMSIZ);
930 	if ((cp = strchr(ifname, '.')) == NULL)
931 		return (NULL);
932 	*cp = '\0';
933 	if ((ifp = ifunit_ref(ifname)) == NULL)
934 		return (NULL);
935 	/* Parse VID. */
936 	if (*++cp == '\0') {
937 		if_rele(ifp);
938 		return (NULL);
939 	}
940 	vid = 0;
941 	for(; *cp >= '0' && *cp <= '9'; cp++)
942 		vid = (vid * 10) + (*cp - '0');
943 	if (*cp != '\0') {
944 		if_rele(ifp);
945 		return (NULL);
946 	}
947 	if (vidp != NULL)
948 		*vidp = vid;
949 
950 	return (ifp);
951 }
952 
953 static int
954 vlan_clone_match(struct if_clone *ifc, const char *name)
955 {
956 	const char *cp;
957 
958 	if (vlan_clone_match_ethervid(name, NULL) != NULL)
959 		return (1);
960 
961 	if (strncmp(vlanname, name, strlen(vlanname)) != 0)
962 		return (0);
963 	for (cp = name + 4; *cp != '\0'; cp++) {
964 		if (*cp < '0' || *cp > '9')
965 			return (0);
966 	}
967 
968 	return (1);
969 }
970 
971 static int
972 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
973 {
974 	char *dp;
975 	int wildcard;
976 	int unit;
977 	int error;
978 	int vid;
979 	struct ifvlan *ifv;
980 	struct ifnet *ifp;
981 	struct ifnet *p;
982 	struct ifaddr *ifa;
983 	struct sockaddr_dl *sdl;
984 	struct vlanreq vlr;
985 	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
986 
987 	/*
988 	 * There are 3 (ugh) ways to specify the cloned device:
989 	 * o pass a parameter block with the clone request.
990 	 * o specify parameters in the text of the clone device name
991 	 * o specify no parameters and get an unattached device that
992 	 *   must be configured separately.
993 	 * The first technique is preferred; the latter two are
994 	 * supported for backwards compatibility.
995 	 *
996 	 * XXXRW: Note historic use of the word "tag" here.  New ioctls may be
997 	 * called for.
998 	 */
999 	if (params) {
1000 		error = copyin(params, &vlr, sizeof(vlr));
1001 		if (error)
1002 			return error;
1003 		p = ifunit_ref(vlr.vlr_parent);
1004 		if (p == NULL)
1005 			return (ENXIO);
1006 		error = ifc_name2unit(name, &unit);
1007 		if (error != 0) {
1008 			if_rele(p);
1009 			return (error);
1010 		}
1011 		vid = vlr.vlr_tag;
1012 		wildcard = (unit < 0);
1013 	} else if ((p = vlan_clone_match_ethervid(name, &vid)) != NULL) {
1014 		unit = -1;
1015 		wildcard = 0;
1016 	} else {
1017 		p = NULL;
1018 		error = ifc_name2unit(name, &unit);
1019 		if (error != 0)
1020 			return (error);
1021 
1022 		wildcard = (unit < 0);
1023 	}
1024 
1025 	error = ifc_alloc_unit(ifc, &unit);
1026 	if (error != 0) {
1027 		if (p != NULL)
1028 			if_rele(p);
1029 		return (error);
1030 	}
1031 
1032 	/* In the wildcard case, we need to update the name. */
1033 	if (wildcard) {
1034 		for (dp = name; *dp != '\0'; dp++);
1035 		if (snprintf(dp, len - (dp-name), "%d", unit) >
1036 		    len - (dp-name) - 1) {
1037 			panic("%s: interface name too long", __func__);
1038 		}
1039 	}
1040 
1041 	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
1042 	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
1043 	if (ifp == NULL) {
1044 		ifc_free_unit(ifc, unit);
1045 		free(ifv, M_VLAN);
1046 		if (p != NULL)
1047 			if_rele(p);
1048 		return (ENOSPC);
1049 	}
1050 	CK_SLIST_INIT(&ifv->vlan_mc_listhead);
1051 	ifp->if_softc = ifv;
1052 	/*
1053 	 * Set the name manually rather than using if_initname because
1054 	 * we don't conform to the default naming convention for interfaces.
1055 	 */
1056 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
1057 	ifp->if_dname = vlanname;
1058 	ifp->if_dunit = unit;
1059 
1060 	ifp->if_init = vlan_init;
1061 	ifp->if_transmit = vlan_transmit;
1062 	ifp->if_qflush = vlan_qflush;
1063 	ifp->if_ioctl = vlan_ioctl;
1064 #if defined(KERN_TLS) || defined(RATELIMIT)
1065 	ifp->if_snd_tag_alloc = vlan_snd_tag_alloc;
1066 	ifp->if_snd_tag_modify = vlan_snd_tag_modify;
1067 	ifp->if_snd_tag_query = vlan_snd_tag_query;
1068 	ifp->if_snd_tag_free = vlan_snd_tag_free;
1069 #endif
1070 	ifp->if_flags = VLAN_IFFLAGS;
1071 	ether_ifattach(ifp, eaddr);
1072 	/* Now undo some of the damage... */
1073 	ifp->if_baudrate = 0;
1074 	ifp->if_type = IFT_L2VLAN;
1075 	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
1076 	ifa = ifp->if_addr;
1077 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1078 	sdl->sdl_type = IFT_L2VLAN;
1079 
1080 	if (p != NULL) {
1081 		error = vlan_config(ifv, p, vid);
1082 		if_rele(p);
1083 		if (error != 0) {
1084 			/*
1085 			 * Since we've partially failed, we need to back
1086 			 * out all the way, otherwise userland could get
1087 			 * confused.  Thus, we destroy the interface.
1088 			 */
1089 			ether_ifdetach(ifp);
1090 			vlan_unconfig(ifp);
1091 			if_free(ifp);
1092 			ifc_free_unit(ifc, unit);
1093 			free(ifv, M_VLAN);
1094 
1095 			return (error);
1096 		}
1097 	}
1098 
1099 	return (0);
1100 }
1101 
1102 static int
1103 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1104 {
1105 	struct ifvlan *ifv = ifp->if_softc;
1106 	int unit = ifp->if_dunit;
1107 
1108 	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
1109 	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
1110 	/*
1111 	 * We should have the only reference to the ifv now, so we can now
1112 	 * drain any remaining lladdr task before freeing the ifnet and the
1113 	 * ifvlan.
1114 	 */
1115 	taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
1116 	NET_EPOCH_WAIT();
1117 	if_free(ifp);
1118 	free(ifv, M_VLAN);
1119 	ifc_free_unit(ifc, unit);
1120 
1121 	return (0);
1122 }
1123 
1124 /*
1125  * The ifp->if_init entry point for vlan(4) is a no-op.
1126  */
1127 static void
1128 vlan_init(void *foo __unused)
1129 {
1130 }
1131 
1132 /*
1133  * The if_transmit method for vlan(4) interface.
1134  */
1135 static int
1136 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1137 {
1138 	struct ifvlan *ifv;
1139 	struct ifnet *p;
1140 	int error, len, mcast;
1141 
1142 	NET_EPOCH_ASSERT();
1143 
1144 	ifv = ifp->if_softc;
1145 	if (TRUNK(ifv) == NULL) {
1146 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1147 		m_freem(m);
1148 		return (ENETDOWN);
1149 	}
1150 	p = PARENT(ifv);
1151 	len = m->m_pkthdr.len;
1152 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1153 
1154 	BPF_MTAP(ifp, m);
1155 
1156 #if defined(KERN_TLS) || defined(RATELIMIT)
1157 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
1158 		struct vlan_snd_tag *vst;
1159 		struct m_snd_tag *mst;
1160 
1161 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
1162 		mst = m->m_pkthdr.snd_tag;
1163 		vst = mst_to_vst(mst);
1164 		if (vst->tag->ifp != p) {
1165 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1166 			m_freem(m);
1167 			return (EAGAIN);
1168 		}
1169 
1170 		m->m_pkthdr.snd_tag = m_snd_tag_ref(vst->tag);
1171 		m_snd_tag_rele(mst);
1172 	}
1173 #endif
1174 
1175 	/*
1176 	 * Do not run parent's if_transmit() if the parent is not up,
1177 	 * or parent's driver will cause a system crash.
1178 	 */
1179 	if (!UP_AND_RUNNING(p)) {
1180 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1181 		m_freem(m);
1182 		return (ENETDOWN);
1183 	}
1184 
1185 	if (!ether_8021q_frame(&m, ifp, p, ifv->ifv_vid, ifv->ifv_pcp)) {
1186 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1187 		return (0);
1188 	}
1189 
1190 	/*
1191 	 * Send it, precisely as ether_output() would have.
1192 	 */
1193 	error = (p->if_transmit)(p, m);
1194 	if (error == 0) {
1195 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1196 		if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1197 		if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1198 	} else
1199 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1200 	return (error);
1201 }
1202 
1203 static int
1204 vlan_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1205     struct route *ro)
1206 {
1207 	struct ifvlan *ifv;
1208 	struct ifnet *p;
1209 
1210 	NET_EPOCH_ASSERT();
1211 
1212 	ifv = ifp->if_softc;
1213 	if (TRUNK(ifv) == NULL) {
1214 		m_freem(m);
1215 		return (ENETDOWN);
1216 	}
1217 	p = PARENT(ifv);
1218 	return p->if_output(ifp, m, dst, ro);
1219 }
1220 
1221 
1222 /*
1223  * The ifp->if_qflush entry point for vlan(4) is a no-op.
1224  */
1225 static void
1226 vlan_qflush(struct ifnet *ifp __unused)
1227 {
1228 }
1229 
1230 static void
1231 vlan_input(struct ifnet *ifp, struct mbuf *m)
1232 {
1233 	struct ifvlantrunk *trunk;
1234 	struct ifvlan *ifv;
1235 	struct m_tag *mtag;
1236 	uint16_t vid, tag;
1237 
1238 	NET_EPOCH_ASSERT();
1239 
1240 	trunk = ifp->if_vlantrunk;
1241 	if (trunk == NULL) {
1242 		m_freem(m);
1243 		return;
1244 	}
1245 
1246 	if (m->m_flags & M_VLANTAG) {
1247 		/*
1248 		 * Packet is tagged, but m contains a normal
1249 		 * Ethernet frame; the tag is stored out-of-band.
1250 		 */
1251 		tag = m->m_pkthdr.ether_vtag;
1252 		m->m_flags &= ~M_VLANTAG;
1253 	} else {
1254 		struct ether_vlan_header *evl;
1255 
1256 		/*
1257 		 * Packet is tagged in-band as specified by 802.1q.
1258 		 */
1259 		switch (ifp->if_type) {
1260 		case IFT_ETHER:
1261 			if (m->m_len < sizeof(*evl) &&
1262 			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
1263 				if_printf(ifp, "cannot pullup VLAN header\n");
1264 				return;
1265 			}
1266 			evl = mtod(m, struct ether_vlan_header *);
1267 			tag = ntohs(evl->evl_tag);
1268 
1269 			/*
1270 			 * Remove the 802.1q header by copying the Ethernet
1271 			 * addresses over it and adjusting the beginning of
1272 			 * the data in the mbuf.  The encapsulated Ethernet
1273 			 * type field is already in place.
1274 			 */
1275 			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1276 			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
1277 			m_adj(m, ETHER_VLAN_ENCAP_LEN);
1278 			break;
1279 
1280 		default:
1281 #ifdef INVARIANTS
1282 			panic("%s: %s has unsupported if_type %u",
1283 			      __func__, ifp->if_xname, ifp->if_type);
1284 #endif
1285 			if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1286 			m_freem(m);
1287 			return;
1288 		}
1289 	}
1290 
1291 	vid = EVL_VLANOFTAG(tag);
1292 
1293 	ifv = vlan_gethash(trunk, vid);
1294 	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1295 		if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1296 		m_freem(m);
1297 		return;
1298 	}
1299 
1300 	if (vlan_mtag_pcp) {
1301 		/*
1302 		 * While uncommon, it is possible that we will find a 802.1q
1303 		 * packet encapsulated inside another packet that also had an
1304 		 * 802.1q header.  For example, ethernet tunneled over IPSEC
1305 		 * arriving over ethernet.  In that case, we replace the
1306 		 * existing 802.1q PCP m_tag value.
1307 		 */
1308 		mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1309 		if (mtag == NULL) {
1310 			mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1311 			    sizeof(uint8_t), M_NOWAIT);
1312 			if (mtag == NULL) {
1313 				if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1314 				m_freem(m);
1315 				return;
1316 			}
1317 			m_tag_prepend(m, mtag);
1318 		}
1319 		*(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1320 	}
1321 
1322 	m->m_pkthdr.rcvif = ifv->ifv_ifp;
1323 	if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1324 
1325 	/* Pass it back through the parent's input routine. */
1326 	(*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
1327 }
1328 
1329 static void
1330 vlan_lladdr_fn(void *arg, int pending __unused)
1331 {
1332 	struct ifvlan *ifv;
1333 	struct ifnet *ifp;
1334 
1335 	ifv = (struct ifvlan *)arg;
1336 	ifp = ifv->ifv_ifp;
1337 
1338 	CURVNET_SET(ifp->if_vnet);
1339 
1340 	/* The ifv_ifp already has the lladdr copied in. */
1341 	if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1342 
1343 	CURVNET_RESTORE();
1344 }
1345 
1346 static int
1347 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid)
1348 {
1349 	struct epoch_tracker et;
1350 	struct ifvlantrunk *trunk;
1351 	struct ifnet *ifp;
1352 	int error = 0;
1353 
1354 	/*
1355 	 * We can handle non-ethernet hardware types as long as
1356 	 * they handle the tagging and headers themselves.
1357 	 */
1358 	if (p->if_type != IFT_ETHER &&
1359 	    (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1360 		return (EPROTONOSUPPORT);
1361 	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1362 		return (EPROTONOSUPPORT);
1363 	/*
1364 	 * Don't let the caller set up a VLAN VID with
1365 	 * anything except VLID bits.
1366 	 * VID numbers 0x0 and 0xFFF are reserved.
1367 	 */
1368 	if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1369 		return (EINVAL);
1370 	if (ifv->ifv_trunk)
1371 		return (EBUSY);
1372 
1373 	VLAN_XLOCK();
1374 	if (p->if_vlantrunk == NULL) {
1375 		trunk = malloc(sizeof(struct ifvlantrunk),
1376 		    M_VLAN, M_WAITOK | M_ZERO);
1377 		vlan_inithash(trunk);
1378 		TRUNK_LOCK_INIT(trunk);
1379 		TRUNK_WLOCK(trunk);
1380 		p->if_vlantrunk = trunk;
1381 		trunk->parent = p;
1382 		if_ref(trunk->parent);
1383 		TRUNK_WUNLOCK(trunk);
1384 	} else {
1385 		trunk = p->if_vlantrunk;
1386 	}
1387 
1388 	ifv->ifv_vid = vid;	/* must set this before vlan_inshash() */
1389 	ifv->ifv_pcp = 0;       /* Default: best effort delivery. */
1390 	vlan_tag_recalculate(ifv);
1391 	error = vlan_inshash(trunk, ifv);
1392 	if (error)
1393 		goto done;
1394 	ifv->ifv_proto = ETHERTYPE_VLAN;
1395 	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1396 	ifv->ifv_mintu = ETHERMIN;
1397 	ifv->ifv_pflags = 0;
1398 	ifv->ifv_capenable = -1;
1399 
1400 	/*
1401 	 * If the parent supports the VLAN_MTU capability,
1402 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1403 	 * use it.
1404 	 */
1405 	if (p->if_capenable & IFCAP_VLAN_MTU) {
1406 		/*
1407 		 * No need to fudge the MTU since the parent can
1408 		 * handle extended frames.
1409 		 */
1410 		ifv->ifv_mtufudge = 0;
1411 	} else {
1412 		/*
1413 		 * Fudge the MTU by the encapsulation size.  This
1414 		 * makes us incompatible with strictly compliant
1415 		 * 802.1Q implementations, but allows us to use
1416 		 * the feature with other NetBSD implementations,
1417 		 * which might still be useful.
1418 		 */
1419 		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1420 	}
1421 
1422 	ifv->ifv_trunk = trunk;
1423 	ifp = ifv->ifv_ifp;
1424 	/*
1425 	 * Initialize fields from our parent.  This duplicates some
1426 	 * work with ether_ifattach() but allows for non-ethernet
1427 	 * interfaces to also work.
1428 	 */
1429 	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1430 	ifp->if_baudrate = p->if_baudrate;
1431 	ifp->if_input = p->if_input;
1432 	ifp->if_resolvemulti = p->if_resolvemulti;
1433 	ifp->if_addrlen = p->if_addrlen;
1434 	ifp->if_broadcastaddr = p->if_broadcastaddr;
1435 	ifp->if_pcp = ifv->ifv_pcp;
1436 
1437 	/*
1438 	 * We wrap the parent's if_output using vlan_output to ensure that it
1439 	 * can't become stale.
1440 	 */
1441 	ifp->if_output = vlan_output;
1442 
1443 	/*
1444 	 * Copy only a selected subset of flags from the parent.
1445 	 * Other flags are none of our business.
1446 	 */
1447 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1448 	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1449 	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1450 #undef VLAN_COPY_FLAGS
1451 
1452 	ifp->if_link_state = p->if_link_state;
1453 
1454 	NET_EPOCH_ENTER(et);
1455 	vlan_capabilities(ifv);
1456 	NET_EPOCH_EXIT(et);
1457 
1458 	/*
1459 	 * Set up our interface address to reflect the underlying
1460 	 * physical interface's.
1461 	 */
1462 	TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
1463 	((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1464 	    p->if_addrlen;
1465 
1466 	/*
1467 	 * Do not schedule link address update if it was the same
1468 	 * as previous parent's. This helps avoid updating for each
1469 	 * associated llentry.
1470 	 */
1471 	if (memcmp(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen) != 0) {
1472 		bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1473 		taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
1474 	}
1475 
1476 	/* We are ready for operation now. */
1477 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1478 
1479 	/* Update flags on the parent, if necessary. */
1480 	vlan_setflags(ifp, 1);
1481 
1482 	/*
1483 	 * Configure multicast addresses that may already be
1484 	 * joined on the vlan device.
1485 	 */
1486 	(void)vlan_setmulti(ifp);
1487 
1488 done:
1489 	if (error == 0)
1490 		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1491 	VLAN_XUNLOCK();
1492 
1493 	return (error);
1494 }
1495 
1496 static void
1497 vlan_unconfig(struct ifnet *ifp)
1498 {
1499 
1500 	VLAN_XLOCK();
1501 	vlan_unconfig_locked(ifp, 0);
1502 	VLAN_XUNLOCK();
1503 }
1504 
1505 static void
1506 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1507 {
1508 	struct ifvlantrunk *trunk;
1509 	struct vlan_mc_entry *mc;
1510 	struct ifvlan *ifv;
1511 	struct ifnet  *parent;
1512 	int error;
1513 
1514 	VLAN_XLOCK_ASSERT();
1515 
1516 	ifv = ifp->if_softc;
1517 	trunk = ifv->ifv_trunk;
1518 	parent = NULL;
1519 
1520 	if (trunk != NULL) {
1521 		parent = trunk->parent;
1522 
1523 		/*
1524 		 * Since the interface is being unconfigured, we need to
1525 		 * empty the list of multicast groups that we may have joined
1526 		 * while we were alive from the parent's list.
1527 		 */
1528 		while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1529 			/*
1530 			 * If the parent interface is being detached,
1531 			 * all its multicast addresses have already
1532 			 * been removed.  Warn about errors if
1533 			 * if_delmulti() does fail, but don't abort as
1534 			 * all callers expect vlan destruction to
1535 			 * succeed.
1536 			 */
1537 			if (!departing) {
1538 				error = if_delmulti(parent,
1539 				    (struct sockaddr *)&mc->mc_addr);
1540 				if (error)
1541 					if_printf(ifp,
1542 		    "Failed to delete multicast address from parent: %d\n",
1543 					    error);
1544 			}
1545 			CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1546 			epoch_call(net_epoch_preempt, &mc->mc_epoch_ctx, vlan_mc_free);
1547 		}
1548 
1549 		vlan_setflags(ifp, 0); /* clear special flags on parent */
1550 
1551 		vlan_remhash(trunk, ifv);
1552 		ifv->ifv_trunk = NULL;
1553 
1554 		/*
1555 		 * Check if we were the last.
1556 		 */
1557 		if (trunk->refcnt == 0) {
1558 			parent->if_vlantrunk = NULL;
1559 			NET_EPOCH_WAIT();
1560 			trunk_destroy(trunk);
1561 		}
1562 	}
1563 
1564 	/* Disconnect from parent. */
1565 	if (ifv->ifv_pflags)
1566 		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1567 	ifp->if_mtu = ETHERMTU;
1568 	ifp->if_link_state = LINK_STATE_UNKNOWN;
1569 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1570 
1571 	/*
1572 	 * Only dispatch an event if vlan was
1573 	 * attached, otherwise there is nothing
1574 	 * to cleanup anyway.
1575 	 */
1576 	if (parent != NULL)
1577 		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1578 }
1579 
1580 /* Handle a reference counted flag that should be set on the parent as well */
1581 static int
1582 vlan_setflag(struct ifnet *ifp, int flag, int status,
1583 	     int (*func)(struct ifnet *, int))
1584 {
1585 	struct ifvlan *ifv;
1586 	int error;
1587 
1588 	VLAN_SXLOCK_ASSERT();
1589 
1590 	ifv = ifp->if_softc;
1591 	status = status ? (ifp->if_flags & flag) : 0;
1592 	/* Now "status" contains the flag value or 0 */
1593 
1594 	/*
1595 	 * See if recorded parent's status is different from what
1596 	 * we want it to be.  If it is, flip it.  We record parent's
1597 	 * status in ifv_pflags so that we won't clear parent's flag
1598 	 * we haven't set.  In fact, we don't clear or set parent's
1599 	 * flags directly, but get or release references to them.
1600 	 * That's why we can be sure that recorded flags still are
1601 	 * in accord with actual parent's flags.
1602 	 */
1603 	if (status != (ifv->ifv_pflags & flag)) {
1604 		error = (*func)(PARENT(ifv), status);
1605 		if (error)
1606 			return (error);
1607 		ifv->ifv_pflags &= ~flag;
1608 		ifv->ifv_pflags |= status;
1609 	}
1610 	return (0);
1611 }
1612 
1613 /*
1614  * Handle IFF_* flags that require certain changes on the parent:
1615  * if "status" is true, update parent's flags respective to our if_flags;
1616  * if "status" is false, forcedly clear the flags set on parent.
1617  */
1618 static int
1619 vlan_setflags(struct ifnet *ifp, int status)
1620 {
1621 	int error, i;
1622 
1623 	for (i = 0; vlan_pflags[i].flag; i++) {
1624 		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1625 				     status, vlan_pflags[i].func);
1626 		if (error)
1627 			return (error);
1628 	}
1629 	return (0);
1630 }
1631 
1632 /* Inform all vlans that their parent has changed link state */
1633 static void
1634 vlan_link_state(struct ifnet *ifp)
1635 {
1636 	struct epoch_tracker et;
1637 	struct ifvlantrunk *trunk;
1638 	struct ifvlan *ifv;
1639 
1640 	NET_EPOCH_ENTER(et);
1641 	trunk = ifp->if_vlantrunk;
1642 	if (trunk == NULL) {
1643 		NET_EPOCH_EXIT(et);
1644 		return;
1645 	}
1646 
1647 	TRUNK_WLOCK(trunk);
1648 	VLAN_FOREACH(ifv, trunk) {
1649 		ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1650 		if_link_state_change(ifv->ifv_ifp,
1651 		    trunk->parent->if_link_state);
1652 	}
1653 	TRUNK_WUNLOCK(trunk);
1654 	NET_EPOCH_EXIT(et);
1655 }
1656 
1657 static void
1658 vlan_capabilities(struct ifvlan *ifv)
1659 {
1660 	struct ifnet *p;
1661 	struct ifnet *ifp;
1662 	struct ifnet_hw_tsomax hw_tsomax;
1663 	int cap = 0, ena = 0, mena;
1664 	u_long hwa = 0;
1665 
1666 	NET_EPOCH_ASSERT();
1667 	VLAN_SXLOCK_ASSERT();
1668 
1669 	p = PARENT(ifv);
1670 	ifp = ifv->ifv_ifp;
1671 
1672 	/* Mask parent interface enabled capabilities disabled by user. */
1673 	mena = p->if_capenable & ifv->ifv_capenable;
1674 
1675 	/*
1676 	 * If the parent interface can do checksum offloading
1677 	 * on VLANs, then propagate its hardware-assisted
1678 	 * checksumming flags. Also assert that checksum
1679 	 * offloading requires hardware VLAN tagging.
1680 	 */
1681 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1682 		cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1683 	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1684 	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1685 		ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1686 		if (ena & IFCAP_TXCSUM)
1687 			hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
1688 			    CSUM_UDP | CSUM_SCTP);
1689 		if (ena & IFCAP_TXCSUM_IPV6)
1690 			hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
1691 			    CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
1692 	}
1693 
1694 	/*
1695 	 * If the parent interface can do TSO on VLANs then
1696 	 * propagate the hardware-assisted flag. TSO on VLANs
1697 	 * does not necessarily require hardware VLAN tagging.
1698 	 */
1699 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1700 	if_hw_tsomax_common(p, &hw_tsomax);
1701 	if_hw_tsomax_update(ifp, &hw_tsomax);
1702 	if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1703 		cap |= p->if_capabilities & IFCAP_TSO;
1704 	if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1705 		ena |= mena & IFCAP_TSO;
1706 		if (ena & IFCAP_TSO)
1707 			hwa |= p->if_hwassist & CSUM_TSO;
1708 	}
1709 
1710 	/*
1711 	 * If the parent interface can do LRO and checksum offloading on
1712 	 * VLANs, then guess it may do LRO on VLANs.  False positive here
1713 	 * cost nothing, while false negative may lead to some confusions.
1714 	 */
1715 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1716 		cap |= p->if_capabilities & IFCAP_LRO;
1717 	if (p->if_capenable & IFCAP_VLAN_HWCSUM)
1718 		ena |= p->if_capenable & IFCAP_LRO;
1719 
1720 	/*
1721 	 * If the parent interface can offload TCP connections over VLANs then
1722 	 * propagate its TOE capability to the VLAN interface.
1723 	 *
1724 	 * All TOE drivers in the tree today can deal with VLANs.  If this
1725 	 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1726 	 * with its own bit.
1727 	 */
1728 #define	IFCAP_VLAN_TOE IFCAP_TOE
1729 	if (p->if_capabilities & IFCAP_VLAN_TOE)
1730 		cap |= p->if_capabilities & IFCAP_TOE;
1731 	if (p->if_capenable & IFCAP_VLAN_TOE) {
1732 		TOEDEV(ifp) = TOEDEV(p);
1733 		ena |= mena & IFCAP_TOE;
1734 	}
1735 
1736 	/*
1737 	 * If the parent interface supports dynamic link state, so does the
1738 	 * VLAN interface.
1739 	 */
1740 	cap |= (p->if_capabilities & IFCAP_LINKSTATE);
1741 	ena |= (mena & IFCAP_LINKSTATE);
1742 
1743 #ifdef RATELIMIT
1744 	/*
1745 	 * If the parent interface supports ratelimiting, so does the
1746 	 * VLAN interface.
1747 	 */
1748 	cap |= (p->if_capabilities & IFCAP_TXRTLMT);
1749 	ena |= (mena & IFCAP_TXRTLMT);
1750 #endif
1751 
1752 	/*
1753 	 * If the parent interface supports unmapped mbufs, so does
1754 	 * the VLAN interface.  Note that this should be fine even for
1755 	 * interfaces that don't support hardware tagging as headers
1756 	 * are prepended in normal mbufs to unmapped mbufs holding
1757 	 * payload data.
1758 	 */
1759 	cap |= (p->if_capabilities & IFCAP_NOMAP);
1760 	ena |= (mena & IFCAP_NOMAP);
1761 
1762 	/*
1763 	 * If the parent interface can offload encryption and segmentation
1764 	 * of TLS records over TCP, propagate it's capability to the VLAN
1765 	 * interface.
1766 	 *
1767 	 * All TLS drivers in the tree today can deal with VLANs.  If
1768 	 * this ever changes, then a new IFCAP_VLAN_TXTLS can be
1769 	 * defined.
1770 	 */
1771 	if (p->if_capabilities & IFCAP_TXTLS)
1772 		cap |= p->if_capabilities & IFCAP_TXTLS;
1773 	if (p->if_capenable & IFCAP_TXTLS)
1774 		ena |= mena & IFCAP_TXTLS;
1775 
1776 	ifp->if_capabilities = cap;
1777 	ifp->if_capenable = ena;
1778 	ifp->if_hwassist = hwa;
1779 }
1780 
1781 static void
1782 vlan_trunk_capabilities(struct ifnet *ifp)
1783 {
1784 	struct epoch_tracker et;
1785 	struct ifvlantrunk *trunk;
1786 	struct ifvlan *ifv;
1787 
1788 	VLAN_SLOCK();
1789 	trunk = ifp->if_vlantrunk;
1790 	if (trunk == NULL) {
1791 		VLAN_SUNLOCK();
1792 		return;
1793 	}
1794 	NET_EPOCH_ENTER(et);
1795 	VLAN_FOREACH(ifv, trunk)
1796 		vlan_capabilities(ifv);
1797 	NET_EPOCH_EXIT(et);
1798 	VLAN_SUNLOCK();
1799 }
1800 
1801 static int
1802 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1803 {
1804 	struct ifnet *p;
1805 	struct ifreq *ifr;
1806 	struct ifaddr *ifa;
1807 	struct ifvlan *ifv;
1808 	struct ifvlantrunk *trunk;
1809 	struct vlanreq vlr;
1810 	int error = 0;
1811 
1812 	ifr = (struct ifreq *)data;
1813 	ifa = (struct ifaddr *) data;
1814 	ifv = ifp->if_softc;
1815 
1816 	switch (cmd) {
1817 	case SIOCSIFADDR:
1818 		ifp->if_flags |= IFF_UP;
1819 #ifdef INET
1820 		if (ifa->ifa_addr->sa_family == AF_INET)
1821 			arp_ifinit(ifp, ifa);
1822 #endif
1823 		break;
1824 	case SIOCGIFADDR:
1825 		bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
1826 		    ifp->if_addrlen);
1827 		break;
1828 	case SIOCGIFMEDIA:
1829 		VLAN_SLOCK();
1830 		if (TRUNK(ifv) != NULL) {
1831 			p = PARENT(ifv);
1832 			if_ref(p);
1833 			error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1834 			if_rele(p);
1835 			/* Limit the result to the parent's current config. */
1836 			if (error == 0) {
1837 				struct ifmediareq *ifmr;
1838 
1839 				ifmr = (struct ifmediareq *)data;
1840 				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1841 					ifmr->ifm_count = 1;
1842 					error = copyout(&ifmr->ifm_current,
1843 						ifmr->ifm_ulist,
1844 						sizeof(int));
1845 				}
1846 			}
1847 		} else {
1848 			error = EINVAL;
1849 		}
1850 		VLAN_SUNLOCK();
1851 		break;
1852 
1853 	case SIOCSIFMEDIA:
1854 		error = EINVAL;
1855 		break;
1856 
1857 	case SIOCSIFMTU:
1858 		/*
1859 		 * Set the interface MTU.
1860 		 */
1861 		VLAN_SLOCK();
1862 		trunk = TRUNK(ifv);
1863 		if (trunk != NULL) {
1864 			TRUNK_WLOCK(trunk);
1865 			if (ifr->ifr_mtu >
1866 			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1867 			    ifr->ifr_mtu <
1868 			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1869 				error = EINVAL;
1870 			else
1871 				ifp->if_mtu = ifr->ifr_mtu;
1872 			TRUNK_WUNLOCK(trunk);
1873 		} else
1874 			error = EINVAL;
1875 		VLAN_SUNLOCK();
1876 		break;
1877 
1878 	case SIOCSETVLAN:
1879 #ifdef VIMAGE
1880 		/*
1881 		 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1882 		 * interface to be delegated to a jail without allowing the
1883 		 * jail to change what underlying interface/VID it is
1884 		 * associated with.  We are not entirely convinced that this
1885 		 * is the right way to accomplish that policy goal.
1886 		 */
1887 		if (ifp->if_vnet != ifp->if_home_vnet) {
1888 			error = EPERM;
1889 			break;
1890 		}
1891 #endif
1892 		error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
1893 		if (error)
1894 			break;
1895 		if (vlr.vlr_parent[0] == '\0') {
1896 			vlan_unconfig(ifp);
1897 			break;
1898 		}
1899 		p = ifunit_ref(vlr.vlr_parent);
1900 		if (p == NULL) {
1901 			error = ENOENT;
1902 			break;
1903 		}
1904 		error = vlan_config(ifv, p, vlr.vlr_tag);
1905 		if_rele(p);
1906 		break;
1907 
1908 	case SIOCGETVLAN:
1909 #ifdef VIMAGE
1910 		if (ifp->if_vnet != ifp->if_home_vnet) {
1911 			error = EPERM;
1912 			break;
1913 		}
1914 #endif
1915 		bzero(&vlr, sizeof(vlr));
1916 		VLAN_SLOCK();
1917 		if (TRUNK(ifv) != NULL) {
1918 			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1919 			    sizeof(vlr.vlr_parent));
1920 			vlr.vlr_tag = ifv->ifv_vid;
1921 		}
1922 		VLAN_SUNLOCK();
1923 		error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
1924 		break;
1925 
1926 	case SIOCSIFFLAGS:
1927 		/*
1928 		 * We should propagate selected flags to the parent,
1929 		 * e.g., promiscuous mode.
1930 		 */
1931 		VLAN_XLOCK();
1932 		if (TRUNK(ifv) != NULL)
1933 			error = vlan_setflags(ifp, 1);
1934 		VLAN_XUNLOCK();
1935 		break;
1936 
1937 	case SIOCADDMULTI:
1938 	case SIOCDELMULTI:
1939 		/*
1940 		 * If we don't have a parent, just remember the membership for
1941 		 * when we do.
1942 		 *
1943 		 * XXX We need the rmlock here to avoid sleeping while
1944 		 * holding in6_multi_mtx.
1945 		 */
1946 		VLAN_XLOCK();
1947 		trunk = TRUNK(ifv);
1948 		if (trunk != NULL)
1949 			error = vlan_setmulti(ifp);
1950 		VLAN_XUNLOCK();
1951 
1952 		break;
1953 	case SIOCGVLANPCP:
1954 #ifdef VIMAGE
1955 		if (ifp->if_vnet != ifp->if_home_vnet) {
1956 			error = EPERM;
1957 			break;
1958 		}
1959 #endif
1960 		ifr->ifr_vlan_pcp = ifv->ifv_pcp;
1961 		break;
1962 
1963 	case SIOCSVLANPCP:
1964 #ifdef VIMAGE
1965 		if (ifp->if_vnet != ifp->if_home_vnet) {
1966 			error = EPERM;
1967 			break;
1968 		}
1969 #endif
1970 		error = priv_check(curthread, PRIV_NET_SETVLANPCP);
1971 		if (error)
1972 			break;
1973 		if (ifr->ifr_vlan_pcp > 7) {
1974 			error = EINVAL;
1975 			break;
1976 		}
1977 		ifv->ifv_pcp = ifr->ifr_vlan_pcp;
1978 		ifp->if_pcp = ifv->ifv_pcp;
1979 		vlan_tag_recalculate(ifv);
1980 		/* broadcast event about PCP change */
1981 		EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
1982 		break;
1983 
1984 	case SIOCSIFCAP:
1985 		VLAN_SLOCK();
1986 		ifv->ifv_capenable = ifr->ifr_reqcap;
1987 		trunk = TRUNK(ifv);
1988 		if (trunk != NULL) {
1989 			struct epoch_tracker et;
1990 
1991 			NET_EPOCH_ENTER(et);
1992 			vlan_capabilities(ifv);
1993 			NET_EPOCH_EXIT(et);
1994 		}
1995 		VLAN_SUNLOCK();
1996 		break;
1997 
1998 	default:
1999 		error = EINVAL;
2000 		break;
2001 	}
2002 
2003 	return (error);
2004 }
2005 
2006 #if defined(KERN_TLS) || defined(RATELIMIT)
2007 static int
2008 vlan_snd_tag_alloc(struct ifnet *ifp,
2009     union if_snd_tag_alloc_params *params,
2010     struct m_snd_tag **ppmt)
2011 {
2012 	struct epoch_tracker et;
2013 	struct vlan_snd_tag *vst;
2014 	struct ifvlan *ifv;
2015 	struct ifnet *parent;
2016 	int error;
2017 
2018 	NET_EPOCH_ENTER(et);
2019 	ifv = ifp->if_softc;
2020 	if (ifv->ifv_trunk != NULL)
2021 		parent = PARENT(ifv);
2022 	else
2023 		parent = NULL;
2024 	if (parent == NULL || parent->if_snd_tag_alloc == NULL) {
2025 		NET_EPOCH_EXIT(et);
2026 		return (EOPNOTSUPP);
2027 	}
2028 	if_ref(parent);
2029 	NET_EPOCH_EXIT(et);
2030 
2031 	vst = malloc(sizeof(*vst), M_VLAN, M_NOWAIT);
2032 	if (vst == NULL) {
2033 		if_rele(parent);
2034 		return (ENOMEM);
2035 	}
2036 
2037 	error = parent->if_snd_tag_alloc(parent, params, &vst->tag);
2038 	if_rele(parent);
2039 	if (error) {
2040 		free(vst, M_VLAN);
2041 		return (error);
2042 	}
2043 
2044 	m_snd_tag_init(&vst->com, ifp);
2045 
2046 	*ppmt = &vst->com;
2047 	return (0);
2048 }
2049 
2050 static int
2051 vlan_snd_tag_modify(struct m_snd_tag *mst,
2052     union if_snd_tag_modify_params *params)
2053 {
2054 	struct vlan_snd_tag *vst;
2055 
2056 	vst = mst_to_vst(mst);
2057 	return (vst->tag->ifp->if_snd_tag_modify(vst->tag, params));
2058 }
2059 
2060 static int
2061 vlan_snd_tag_query(struct m_snd_tag *mst,
2062     union if_snd_tag_query_params *params)
2063 {
2064 	struct vlan_snd_tag *vst;
2065 
2066 	vst = mst_to_vst(mst);
2067 	return (vst->tag->ifp->if_snd_tag_query(vst->tag, params));
2068 }
2069 
2070 static void
2071 vlan_snd_tag_free(struct m_snd_tag *mst)
2072 {
2073 	struct vlan_snd_tag *vst;
2074 
2075 	vst = mst_to_vst(mst);
2076 	m_snd_tag_rele(vst->tag);
2077 	free(vst, M_VLAN);
2078 }
2079 #endif
2080