xref: /freebsd/sys/net/if_vlan.c (revision 2a01feab)
1 /*-
2  * Copyright 1998 Massachusetts Institute of Technology
3  * Copyright 2012 ADARA Networks, Inc.
4  * Copyright 2017 Dell EMC Isilon
5  *
6  * Portions of this software were developed by Robert N. M. Watson under
7  * contract to ADARA Networks, Inc.
8  *
9  * Permission to use, copy, modify, and distribute this software and
10  * its documentation for any purpose and without fee is hereby
11  * granted, provided that both the above copyright notice and this
12  * permission notice appear in all copies, that both the above
13  * copyright notice and this permission notice appear in all
14  * supporting documentation, and that the name of M.I.T. not be used
15  * in advertising or publicity pertaining to distribution of the
16  * software without specific, written prior permission.  M.I.T. makes
17  * no representations about the suitability of this software for any
18  * purpose.  It is provided "as is" without express or implied
19  * warranty.
20  *
21  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
22  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
23  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
24  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
25  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
28  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
29  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 /*
36  * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
37  * This is sort of sneaky in the implementation, since
38  * we need to pretend to be enough of an Ethernet implementation
39  * to make arp work.  The way we do this is by telling everyone
40  * that we are an Ethernet, and then catch the packets that
41  * ether_output() sends to us via if_transmit(), rewrite them for
42  * use by the real outgoing interface, and ask it to send them.
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_inet.h"
49 #include "opt_vlan.h"
50 #include "opt_ratelimit.h"
51 
52 #include <sys/param.h>
53 #include <sys/eventhandler.h>
54 #include <sys/kernel.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/mbuf.h>
58 #include <sys/module.h>
59 #include <sys/rmlock.h>
60 #include <sys/priv.h>
61 #include <sys/queue.h>
62 #include <sys/socket.h>
63 #include <sys/sockio.h>
64 #include <sys/sysctl.h>
65 #include <sys/systm.h>
66 #include <sys/sx.h>
67 #include <sys/taskqueue.h>
68 
69 #include <net/bpf.h>
70 #include <net/ethernet.h>
71 #include <net/if.h>
72 #include <net/if_var.h>
73 #include <net/if_clone.h>
74 #include <net/if_dl.h>
75 #include <net/if_types.h>
76 #include <net/if_vlan_var.h>
77 #include <net/vnet.h>
78 
79 #ifdef INET
80 #include <netinet/in.h>
81 #include <netinet/if_ether.h>
82 #endif
83 
84 #define	VLAN_DEF_HWIDTH	4
85 #define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
86 
87 #define	UP_AND_RUNNING(ifp) \
88     ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
89 
90 CK_SLIST_HEAD(ifvlanhead, ifvlan);
91 
92 struct ifvlantrunk {
93 	struct	ifnet   *parent;	/* parent interface of this trunk */
94 	struct	mtx	lock;
95 #ifdef VLAN_ARRAY
96 #define	VLAN_ARRAY_SIZE	(EVL_VLID_MASK + 1)
97 	struct	ifvlan	*vlans[VLAN_ARRAY_SIZE]; /* static table */
98 #else
99 	struct	ifvlanhead *hash;	/* dynamic hash-list table */
100 	uint16_t	hmask;
101 	uint16_t	hwidth;
102 #endif
103 	int		refcnt;
104 };
105 
106 /*
107  * This macro provides a facility to iterate over every vlan on a trunk with
108  * the assumption that none will be added/removed during iteration.
109  */
110 #ifdef VLAN_ARRAY
111 #define VLAN_FOREACH(_ifv, _trunk) \
112 	size_t _i; \
113 	for (_i = 0; _i < VLAN_ARRAY_SIZE; _i++) \
114 		if (((_ifv) = (_trunk)->vlans[_i]) != NULL)
115 #else /* VLAN_ARRAY */
116 #define VLAN_FOREACH(_ifv, _trunk) \
117 	struct ifvlan *_next; \
118 	size_t _i; \
119 	for (_i = 0; _i < (1 << (_trunk)->hwidth); _i++) \
120 		CK_SLIST_FOREACH_SAFE((_ifv), &(_trunk)->hash[_i], ifv_list, _next)
121 #endif /* VLAN_ARRAY */
122 
123 /*
124  * This macro provides a facility to iterate over every vlan on a trunk while
125  * also modifying the number of vlans on the trunk. The iteration continues
126  * until some condition is met or there are no more vlans on the trunk.
127  */
128 #ifdef VLAN_ARRAY
129 /* The VLAN_ARRAY case is simple -- just a for loop using the condition. */
130 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
131 	size_t _i; \
132 	for (_i = 0; !(_cond) && _i < VLAN_ARRAY_SIZE; _i++) \
133 		if (((_ifv) = (_trunk)->vlans[_i]))
134 #else /* VLAN_ARRAY */
135 /*
136  * The hash table case is more complicated. We allow for the hash table to be
137  * modified (i.e. vlans removed) while we are iterating over it. To allow for
138  * this we must restart the iteration every time we "touch" something during
139  * the iteration, since removal will resize the hash table and invalidate our
140  * current position. If acting on the touched element causes the trunk to be
141  * emptied, then iteration also stops.
142  */
143 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
144 	size_t _i; \
145 	bool _touch = false; \
146 	for (_i = 0; \
147 	    !(_cond) && _i < (1 << (_trunk)->hwidth); \
148 	    _i = (_touch && ((_trunk) != NULL) ? 0 : _i + 1), _touch = false) \
149 		if (((_ifv) = CK_SLIST_FIRST(&(_trunk)->hash[_i])) != NULL && \
150 		    (_touch = true))
151 #endif /* VLAN_ARRAY */
152 
153 struct vlan_mc_entry {
154 	struct sockaddr_dl		mc_addr;
155 	CK_SLIST_ENTRY(vlan_mc_entry)	mc_entries;
156 };
157 
158 struct	ifvlan {
159 	struct	ifvlantrunk *ifv_trunk;
160 	struct	ifnet *ifv_ifp;
161 #define	TRUNK(ifv)	((ifv)->ifv_trunk)
162 #define	PARENT(ifv)	((ifv)->ifv_trunk->parent)
163 	void	*ifv_cookie;
164 	int	ifv_pflags;	/* special flags we have set on parent */
165 	int	ifv_capenable;
166 	struct	ifv_linkmib {
167 		int	ifvm_encaplen;	/* encapsulation length */
168 		int	ifvm_mtufudge;	/* MTU fudged by this much */
169 		int	ifvm_mintu;	/* min transmission unit */
170 		uint16_t ifvm_proto;	/* encapsulation ethertype */
171 		uint16_t ifvm_tag;	/* tag to apply on packets leaving if */
172               	uint16_t ifvm_vid;	/* VLAN ID */
173 		uint8_t	ifvm_pcp;	/* Priority Code Point (PCP). */
174 	}	ifv_mib;
175 	struct task lladdr_task;
176 	CK_SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
177 #ifndef VLAN_ARRAY
178 	CK_SLIST_ENTRY(ifvlan) ifv_list;
179 #endif
180 };
181 #define	ifv_proto	ifv_mib.ifvm_proto
182 #define	ifv_tag		ifv_mib.ifvm_tag
183 #define	ifv_vid 	ifv_mib.ifvm_vid
184 #define	ifv_pcp		ifv_mib.ifvm_pcp
185 #define	ifv_encaplen	ifv_mib.ifvm_encaplen
186 #define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
187 #define	ifv_mintu	ifv_mib.ifvm_mintu
188 
189 /* Special flags we should propagate to parent. */
190 static struct {
191 	int flag;
192 	int (*func)(struct ifnet *, int);
193 } vlan_pflags[] = {
194 	{IFF_PROMISC, ifpromisc},
195 	{IFF_ALLMULTI, if_allmulti},
196 	{0, NULL}
197 };
198 
199 extern int vlan_mtag_pcp;
200 
201 static const char vlanname[] = "vlan";
202 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
203 
204 static eventhandler_tag ifdetach_tag;
205 static eventhandler_tag iflladdr_tag;
206 
207 /*
208  * if_vlan uses two module-level synchronizations primitives to allow concurrent
209  * modification of vlan interfaces and (mostly) allow for vlans to be destroyed
210  * while they are being used for tx/rx. To accomplish this in a way that has
211  * acceptable performance and cooperation with other parts of the network stack
212  * there is a non-sleepable epoch(9) and an sx(9).
213  *
214  * The performance-sensitive paths that warrant using the epoch(9) are
215  * vlan_transmit and vlan_input. Both have to check for the vlan interface's
216  * existence using if_vlantrunk, and being in the network tx/rx paths the use
217  * of an epoch(9) gives a measureable improvement in performance.
218  *
219  * The reason for having an sx(9) is mostly because there are still areas that
220  * must be sleepable and also have safe concurrent access to a vlan interface.
221  * Since the sx(9) exists, it is used by default in most paths unless sleeping
222  * is not permitted, or if it is not clear whether sleeping is permitted.
223  *
224  */
225 #define _VLAN_SX_ID ifv_sx
226 
227 static struct sx _VLAN_SX_ID;
228 
229 #define VLAN_LOCKING_INIT() \
230 	sx_init(&_VLAN_SX_ID, "vlan_sx")
231 
232 #define VLAN_LOCKING_DESTROY() \
233 	sx_destroy(&_VLAN_SX_ID)
234 
235 #define	VLAN_RLOCK()			NET_EPOCH_ENTER();
236 #define	VLAN_RUNLOCK()			NET_EPOCH_EXIT();
237 #define	VLAN_RLOCK_ASSERT()		MPASS(in_epoch(net_epoch_preempt))
238 
239 #define	VLAN_SLOCK()			sx_slock(&_VLAN_SX_ID)
240 #define	VLAN_SUNLOCK()			sx_sunlock(&_VLAN_SX_ID)
241 #define	VLAN_XLOCK()			sx_xlock(&_VLAN_SX_ID)
242 #define	VLAN_XUNLOCK()			sx_xunlock(&_VLAN_SX_ID)
243 #define	VLAN_SLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_SLOCKED)
244 #define	VLAN_XLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_XLOCKED)
245 #define	VLAN_SXLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_LOCKED)
246 
247 
248 /*
249  * We also have a per-trunk mutex that should be acquired when changing
250  * its state.
251  */
252 #define	TRUNK_LOCK_INIT(trunk)		mtx_init(&(trunk)->lock, vlanname, NULL, MTX_DEF)
253 #define	TRUNK_LOCK_DESTROY(trunk)	mtx_destroy(&(trunk)->lock)
254 #define	TRUNK_RLOCK(trunk)		NET_EPOCH_ENTER()
255 #define	TRUNK_WLOCK(trunk)		mtx_lock(&(trunk)->lock)
256 #define	TRUNK_RUNLOCK(trunk)		NET_EPOCH_EXIT();
257 #define	TRUNK_WUNLOCK(trunk)		mtx_unlock(&(trunk)->lock)
258 #define	TRUNK_RLOCK_ASSERT(trunk)	MPASS(in_epoch(net_epoch_preempt))
259 #define	TRUNK_LOCK_ASSERT(trunk)	MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(trunk)->lock))
260 #define	TRUNK_WLOCK_ASSERT(trunk)	mtx_assert(&(trunk)->lock, MA_OWNED);
261 
262 /*
263  * The VLAN_ARRAY substitutes the dynamic hash with a static array
264  * with 4096 entries. In theory this can give a boost in processing,
265  * however in practice it does not. Probably this is because the array
266  * is too big to fit into CPU cache.
267  */
268 #ifndef VLAN_ARRAY
269 static	void vlan_inithash(struct ifvlantrunk *trunk);
270 static	void vlan_freehash(struct ifvlantrunk *trunk);
271 static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
272 static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
273 static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
274 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
275 	uint16_t vid);
276 #endif
277 static	void trunk_destroy(struct ifvlantrunk *trunk);
278 
279 static	void vlan_init(void *foo);
280 static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
281 static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
282 #ifdef RATELIMIT
283 static	int vlan_snd_tag_alloc(struct ifnet *,
284     union if_snd_tag_alloc_params *, struct m_snd_tag **);
285 #endif
286 static	void vlan_qflush(struct ifnet *ifp);
287 static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
288     int (*func)(struct ifnet *, int));
289 static	int vlan_setflags(struct ifnet *ifp, int status);
290 static	int vlan_setmulti(struct ifnet *ifp);
291 static	int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
292 static	void vlan_unconfig(struct ifnet *ifp);
293 static	void vlan_unconfig_locked(struct ifnet *ifp, int departing);
294 static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
295 static	void vlan_link_state(struct ifnet *ifp);
296 static	void vlan_capabilities(struct ifvlan *ifv);
297 static	void vlan_trunk_capabilities(struct ifnet *ifp);
298 
299 static	struct ifnet *vlan_clone_match_ethervid(const char *, int *);
300 static	int vlan_clone_match(struct if_clone *, const char *);
301 static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
302 static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
303 
304 static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
305 static  void vlan_iflladdr(void *arg, struct ifnet *ifp);
306 
307 static  void vlan_lladdr_fn(void *arg, int pending);
308 
309 static struct if_clone *vlan_cloner;
310 
311 #ifdef VIMAGE
312 VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner);
313 #define	V_vlan_cloner	VNET(vlan_cloner)
314 #endif
315 
316 #ifndef VLAN_ARRAY
317 #define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
318 
319 static void
320 vlan_inithash(struct ifvlantrunk *trunk)
321 {
322 	int i, n;
323 
324 	/*
325 	 * The trunk must not be locked here since we call malloc(M_WAITOK).
326 	 * It is OK in case this function is called before the trunk struct
327 	 * gets hooked up and becomes visible from other threads.
328 	 */
329 
330 	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
331 	    ("%s: hash already initialized", __func__));
332 
333 	trunk->hwidth = VLAN_DEF_HWIDTH;
334 	n = 1 << trunk->hwidth;
335 	trunk->hmask = n - 1;
336 	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
337 	for (i = 0; i < n; i++)
338 		CK_SLIST_INIT(&trunk->hash[i]);
339 }
340 
341 static void
342 vlan_freehash(struct ifvlantrunk *trunk)
343 {
344 #ifdef INVARIANTS
345 	int i;
346 
347 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
348 	for (i = 0; i < (1 << trunk->hwidth); i++)
349 		KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]),
350 		    ("%s: hash table not empty", __func__));
351 #endif
352 	free(trunk->hash, M_VLAN);
353 	trunk->hash = NULL;
354 	trunk->hwidth = trunk->hmask = 0;
355 }
356 
357 static int
358 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
359 {
360 	int i, b;
361 	struct ifvlan *ifv2;
362 
363 	VLAN_XLOCK_ASSERT();
364 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
365 
366 	b = 1 << trunk->hwidth;
367 	i = HASH(ifv->ifv_vid, trunk->hmask);
368 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
369 		if (ifv->ifv_vid == ifv2->ifv_vid)
370 			return (EEXIST);
371 
372 	/*
373 	 * Grow the hash when the number of vlans exceeds half of the number of
374 	 * hash buckets squared. This will make the average linked-list length
375 	 * buckets/2.
376 	 */
377 	if (trunk->refcnt > (b * b) / 2) {
378 		vlan_growhash(trunk, 1);
379 		i = HASH(ifv->ifv_vid, trunk->hmask);
380 	}
381 	CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
382 	trunk->refcnt++;
383 
384 	return (0);
385 }
386 
387 static int
388 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
389 {
390 	int i, b;
391 	struct ifvlan *ifv2;
392 
393 	VLAN_XLOCK_ASSERT();
394 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
395 
396 	b = 1 << trunk->hwidth;
397 	i = HASH(ifv->ifv_vid, trunk->hmask);
398 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
399 		if (ifv2 == ifv) {
400 			trunk->refcnt--;
401 			CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list);
402 			if (trunk->refcnt < (b * b) / 2)
403 				vlan_growhash(trunk, -1);
404 			return (0);
405 		}
406 
407 	panic("%s: vlan not found\n", __func__);
408 	return (ENOENT); /*NOTREACHED*/
409 }
410 
411 /*
412  * Grow the hash larger or smaller if memory permits.
413  */
414 static void
415 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
416 {
417 	struct ifvlan *ifv;
418 	struct ifvlanhead *hash2;
419 	int hwidth2, i, j, n, n2;
420 
421 	VLAN_XLOCK_ASSERT();
422 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
423 
424 	if (howmuch == 0) {
425 		/* Harmless yet obvious coding error */
426 		printf("%s: howmuch is 0\n", __func__);
427 		return;
428 	}
429 
430 	hwidth2 = trunk->hwidth + howmuch;
431 	n = 1 << trunk->hwidth;
432 	n2 = 1 << hwidth2;
433 	/* Do not shrink the table below the default */
434 	if (hwidth2 < VLAN_DEF_HWIDTH)
435 		return;
436 
437 	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK);
438 	if (hash2 == NULL) {
439 		printf("%s: out of memory -- hash size not changed\n",
440 		    __func__);
441 		return;		/* We can live with the old hash table */
442 	}
443 	for (j = 0; j < n2; j++)
444 		CK_SLIST_INIT(&hash2[j]);
445 	for (i = 0; i < n; i++)
446 		while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) {
447 			CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list);
448 			j = HASH(ifv->ifv_vid, n2 - 1);
449 			CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
450 		}
451 	NET_EPOCH_WAIT();
452 	free(trunk->hash, M_VLAN);
453 	trunk->hash = hash2;
454 	trunk->hwidth = hwidth2;
455 	trunk->hmask = n2 - 1;
456 
457 	if (bootverbose)
458 		if_printf(trunk->parent,
459 		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
460 }
461 
462 static __inline struct ifvlan *
463 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
464 {
465 	struct ifvlan *ifv;
466 
467 	TRUNK_RLOCK_ASSERT(trunk);
468 
469 	CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
470 		if (ifv->ifv_vid == vid)
471 			return (ifv);
472 	return (NULL);
473 }
474 
475 #if 0
476 /* Debugging code to view the hashtables. */
477 static void
478 vlan_dumphash(struct ifvlantrunk *trunk)
479 {
480 	int i;
481 	struct ifvlan *ifv;
482 
483 	for (i = 0; i < (1 << trunk->hwidth); i++) {
484 		printf("%d: ", i);
485 		CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
486 			printf("%s ", ifv->ifv_ifp->if_xname);
487 		printf("\n");
488 	}
489 }
490 #endif /* 0 */
491 #else
492 
493 static __inline struct ifvlan *
494 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
495 {
496 
497 	return trunk->vlans[vid];
498 }
499 
500 static __inline int
501 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
502 {
503 
504 	if (trunk->vlans[ifv->ifv_vid] != NULL)
505 		return EEXIST;
506 	trunk->vlans[ifv->ifv_vid] = ifv;
507 	trunk->refcnt++;
508 
509 	return (0);
510 }
511 
512 static __inline int
513 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
514 {
515 
516 	trunk->vlans[ifv->ifv_vid] = NULL;
517 	trunk->refcnt--;
518 
519 	return (0);
520 }
521 
522 static __inline void
523 vlan_freehash(struct ifvlantrunk *trunk)
524 {
525 }
526 
527 static __inline void
528 vlan_inithash(struct ifvlantrunk *trunk)
529 {
530 }
531 
532 #endif /* !VLAN_ARRAY */
533 
534 static void
535 trunk_destroy(struct ifvlantrunk *trunk)
536 {
537 	VLAN_XLOCK_ASSERT();
538 
539 	vlan_freehash(trunk);
540 	trunk->parent->if_vlantrunk = NULL;
541 	TRUNK_LOCK_DESTROY(trunk);
542 	if_rele(trunk->parent);
543 	free(trunk, M_VLAN);
544 }
545 
546 /*
547  * Program our multicast filter. What we're actually doing is
548  * programming the multicast filter of the parent. This has the
549  * side effect of causing the parent interface to receive multicast
550  * traffic that it doesn't really want, which ends up being discarded
551  * later by the upper protocol layers. Unfortunately, there's no way
552  * to avoid this: there really is only one physical interface.
553  */
554 static int
555 vlan_setmulti(struct ifnet *ifp)
556 {
557 	struct ifnet		*ifp_p;
558 	struct ifmultiaddr	*ifma;
559 	struct ifvlan		*sc;
560 	struct vlan_mc_entry	*mc;
561 	int			error;
562 
563 	VLAN_XLOCK_ASSERT();
564 
565 	/* Find the parent. */
566 	sc = ifp->if_softc;
567 	ifp_p = PARENT(sc);
568 
569 	CURVNET_SET_QUIET(ifp_p->if_vnet);
570 
571 	/* First, remove any existing filter entries. */
572 	while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
573 		CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
574 		(void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
575 		NET_EPOCH_WAIT();
576 		free(mc, M_VLAN);
577 	}
578 
579 	/* Now program new ones. */
580 	IF_ADDR_WLOCK(ifp);
581 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
582 		if (ifma->ifma_addr->sa_family != AF_LINK)
583 			continue;
584 		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
585 		if (mc == NULL) {
586 			IF_ADDR_WUNLOCK(ifp);
587 			return (ENOMEM);
588 		}
589 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
590 		mc->mc_addr.sdl_index = ifp_p->if_index;
591 		CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
592 	}
593 	IF_ADDR_WUNLOCK(ifp);
594 	CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
595 		error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
596 		    NULL);
597 		if (error)
598 			return (error);
599 	}
600 
601 	CURVNET_RESTORE();
602 	return (0);
603 }
604 
605 /*
606  * A handler for parent interface link layer address changes.
607  * If the parent interface link layer address is changed we
608  * should also change it on all children vlans.
609  */
610 static void
611 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
612 {
613 	struct ifvlan *ifv;
614 	struct ifnet *ifv_ifp;
615 	struct ifvlantrunk *trunk;
616 	struct sockaddr_dl *sdl;
617 
618 	/* Need the rmlock since this is run on taskqueue_swi. */
619 	VLAN_RLOCK();
620 	trunk = ifp->if_vlantrunk;
621 	if (trunk == NULL) {
622 		VLAN_RUNLOCK();
623 		return;
624 	}
625 
626 	/*
627 	 * OK, it's a trunk.  Loop over and change all vlan's lladdrs on it.
628 	 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
629 	 * ioctl calls on the parent garbling the lladdr of the child vlan.
630 	 */
631 	TRUNK_WLOCK(trunk);
632 	VLAN_FOREACH(ifv, trunk) {
633 		/*
634 		 * Copy new new lladdr into the ifv_ifp, enqueue a task
635 		 * to actually call if_setlladdr. if_setlladdr needs to
636 		 * be deferred to a taskqueue because it will call into
637 		 * the if_vlan ioctl path and try to acquire the global
638 		 * lock.
639 		 */
640 		ifv_ifp = ifv->ifv_ifp;
641 		bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
642 		    ifp->if_addrlen);
643 		sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
644 		sdl->sdl_alen = ifp->if_addrlen;
645 		taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
646 	}
647 	TRUNK_WUNLOCK(trunk);
648 	VLAN_RUNLOCK();
649 }
650 
651 /*
652  * A handler for network interface departure events.
653  * Track departure of trunks here so that we don't access invalid
654  * pointers or whatever if a trunk is ripped from under us, e.g.,
655  * by ejecting its hot-plug card.  However, if an ifnet is simply
656  * being renamed, then there's no need to tear down the state.
657  */
658 static void
659 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
660 {
661 	struct ifvlan *ifv;
662 	struct ifvlantrunk *trunk;
663 
664 	/* If the ifnet is just being renamed, don't do anything. */
665 	if (ifp->if_flags & IFF_RENAMING)
666 		return;
667 	VLAN_XLOCK();
668 	trunk = ifp->if_vlantrunk;
669 	if (trunk == NULL) {
670 		VLAN_XUNLOCK();
671 		return;
672 	}
673 
674 	/*
675 	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
676 	 * Check trunk pointer after each vlan_unconfig() as it will
677 	 * free it and set to NULL after the last vlan was detached.
678 	 */
679 	VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
680 	    ifp->if_vlantrunk == NULL)
681 		vlan_unconfig_locked(ifv->ifv_ifp, 1);
682 
683 	/* Trunk should have been destroyed in vlan_unconfig(). */
684 	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
685 	VLAN_XUNLOCK();
686 }
687 
688 /*
689  * Return the trunk device for a virtual interface.
690  */
691 static struct ifnet  *
692 vlan_trunkdev(struct ifnet *ifp)
693 {
694 	struct ifvlan *ifv;
695 
696 	if (ifp->if_type != IFT_L2VLAN)
697 		return (NULL);
698 
699 	VLAN_RLOCK();
700 	ifv = ifp->if_softc;
701 	ifp = NULL;
702 	if (ifv->ifv_trunk)
703 		ifp = PARENT(ifv);
704 	VLAN_RUNLOCK();
705 	return (ifp);
706 }
707 
708 /*
709  * Return the 12-bit VLAN VID for this interface, for use by external
710  * components such as Infiniband.
711  *
712  * XXXRW: Note that the function name here is historical; it should be named
713  * vlan_vid().
714  */
715 static int
716 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
717 {
718 	struct ifvlan *ifv;
719 
720 	if (ifp->if_type != IFT_L2VLAN)
721 		return (EINVAL);
722 	ifv = ifp->if_softc;
723 	*vidp = ifv->ifv_vid;
724 	return (0);
725 }
726 
727 static int
728 vlan_pcp(struct ifnet *ifp, uint16_t *pcpp)
729 {
730 	struct ifvlan *ifv;
731 
732 	if (ifp->if_type != IFT_L2VLAN)
733 		return (EINVAL);
734 	ifv = ifp->if_softc;
735 	*pcpp = ifv->ifv_pcp;
736 	return (0);
737 }
738 
739 /*
740  * Return a driver specific cookie for this interface.  Synchronization
741  * with setcookie must be provided by the driver.
742  */
743 static void *
744 vlan_cookie(struct ifnet *ifp)
745 {
746 	struct ifvlan *ifv;
747 
748 	if (ifp->if_type != IFT_L2VLAN)
749 		return (NULL);
750 	ifv = ifp->if_softc;
751 	return (ifv->ifv_cookie);
752 }
753 
754 /*
755  * Store a cookie in our softc that drivers can use to store driver
756  * private per-instance data in.
757  */
758 static int
759 vlan_setcookie(struct ifnet *ifp, void *cookie)
760 {
761 	struct ifvlan *ifv;
762 
763 	if (ifp->if_type != IFT_L2VLAN)
764 		return (EINVAL);
765 	ifv = ifp->if_softc;
766 	ifv->ifv_cookie = cookie;
767 	return (0);
768 }
769 
770 /*
771  * Return the vlan device present at the specific VID.
772  */
773 static struct ifnet *
774 vlan_devat(struct ifnet *ifp, uint16_t vid)
775 {
776 	struct ifvlantrunk *trunk;
777 	struct ifvlan *ifv;
778 
779 	VLAN_RLOCK();
780 	trunk = ifp->if_vlantrunk;
781 	if (trunk == NULL) {
782 		VLAN_RUNLOCK();
783 		return (NULL);
784 	}
785 	ifp = NULL;
786 	ifv = vlan_gethash(trunk, vid);
787 	if (ifv)
788 		ifp = ifv->ifv_ifp;
789 	VLAN_RUNLOCK();
790 	return (ifp);
791 }
792 
793 /*
794  * Recalculate the cached VLAN tag exposed via the MIB.
795  */
796 static void
797 vlan_tag_recalculate(struct ifvlan *ifv)
798 {
799 
800        ifv->ifv_tag = EVL_MAKETAG(ifv->ifv_vid, ifv->ifv_pcp, 0);
801 }
802 
803 /*
804  * VLAN support can be loaded as a module.  The only place in the
805  * system that's intimately aware of this is ether_input.  We hook
806  * into this code through vlan_input_p which is defined there and
807  * set here.  No one else in the system should be aware of this so
808  * we use an explicit reference here.
809  */
810 extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
811 
812 /* For if_link_state_change() eyes only... */
813 extern	void (*vlan_link_state_p)(struct ifnet *);
814 
815 static int
816 vlan_modevent(module_t mod, int type, void *data)
817 {
818 
819 	switch (type) {
820 	case MOD_LOAD:
821 		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
822 		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
823 		if (ifdetach_tag == NULL)
824 			return (ENOMEM);
825 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
826 		    vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
827 		if (iflladdr_tag == NULL)
828 			return (ENOMEM);
829 		VLAN_LOCKING_INIT();
830 		vlan_input_p = vlan_input;
831 		vlan_link_state_p = vlan_link_state;
832 		vlan_trunk_cap_p = vlan_trunk_capabilities;
833 		vlan_trunkdev_p = vlan_trunkdev;
834 		vlan_cookie_p = vlan_cookie;
835 		vlan_setcookie_p = vlan_setcookie;
836 		vlan_tag_p = vlan_tag;
837 		vlan_pcp_p = vlan_pcp;
838 		vlan_devat_p = vlan_devat;
839 #ifndef VIMAGE
840 		vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
841 		    vlan_clone_create, vlan_clone_destroy);
842 #endif
843 		if (bootverbose)
844 			printf("vlan: initialized, using "
845 #ifdef VLAN_ARRAY
846 			       "full-size arrays"
847 #else
848 			       "hash tables with chaining"
849 #endif
850 
851 			       "\n");
852 		break;
853 	case MOD_UNLOAD:
854 #ifndef VIMAGE
855 		if_clone_detach(vlan_cloner);
856 #endif
857 		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
858 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
859 		vlan_input_p = NULL;
860 		vlan_link_state_p = NULL;
861 		vlan_trunk_cap_p = NULL;
862 		vlan_trunkdev_p = NULL;
863 		vlan_tag_p = NULL;
864 		vlan_cookie_p = NULL;
865 		vlan_setcookie_p = NULL;
866 		vlan_devat_p = NULL;
867 		VLAN_LOCKING_DESTROY();
868 		if (bootverbose)
869 			printf("vlan: unloaded\n");
870 		break;
871 	default:
872 		return (EOPNOTSUPP);
873 	}
874 	return (0);
875 }
876 
877 static moduledata_t vlan_mod = {
878 	"if_vlan",
879 	vlan_modevent,
880 	0
881 };
882 
883 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
884 MODULE_VERSION(if_vlan, 3);
885 
886 #ifdef VIMAGE
887 static void
888 vnet_vlan_init(const void *unused __unused)
889 {
890 
891 	vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
892 		    vlan_clone_create, vlan_clone_destroy);
893 	V_vlan_cloner = vlan_cloner;
894 }
895 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
896     vnet_vlan_init, NULL);
897 
898 static void
899 vnet_vlan_uninit(const void *unused __unused)
900 {
901 
902 	if_clone_detach(V_vlan_cloner);
903 }
904 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
905     vnet_vlan_uninit, NULL);
906 #endif
907 
908 /*
909  * Check for <etherif>.<vlan> style interface names.
910  */
911 static struct ifnet *
912 vlan_clone_match_ethervid(const char *name, int *vidp)
913 {
914 	char ifname[IFNAMSIZ];
915 	char *cp;
916 	struct ifnet *ifp;
917 	int vid;
918 
919 	strlcpy(ifname, name, IFNAMSIZ);
920 	if ((cp = strchr(ifname, '.')) == NULL)
921 		return (NULL);
922 	*cp = '\0';
923 	if ((ifp = ifunit_ref(ifname)) == NULL)
924 		return (NULL);
925 	/* Parse VID. */
926 	if (*++cp == '\0') {
927 		if_rele(ifp);
928 		return (NULL);
929 	}
930 	vid = 0;
931 	for(; *cp >= '0' && *cp <= '9'; cp++)
932 		vid = (vid * 10) + (*cp - '0');
933 	if (*cp != '\0') {
934 		if_rele(ifp);
935 		return (NULL);
936 	}
937 	if (vidp != NULL)
938 		*vidp = vid;
939 
940 	return (ifp);
941 }
942 
943 static int
944 vlan_clone_match(struct if_clone *ifc, const char *name)
945 {
946 	const char *cp;
947 
948 	if (vlan_clone_match_ethervid(name, NULL) != NULL)
949 		return (1);
950 
951 	if (strncmp(vlanname, name, strlen(vlanname)) != 0)
952 		return (0);
953 	for (cp = name + 4; *cp != '\0'; cp++) {
954 		if (*cp < '0' || *cp > '9')
955 			return (0);
956 	}
957 
958 	return (1);
959 }
960 
961 static int
962 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
963 {
964 	char *dp;
965 	int wildcard;
966 	int unit;
967 	int error;
968 	int vid;
969 	struct ifvlan *ifv;
970 	struct ifnet *ifp;
971 	struct ifnet *p;
972 	struct ifaddr *ifa;
973 	struct sockaddr_dl *sdl;
974 	struct vlanreq vlr;
975 	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
976 
977 	/*
978 	 * There are 3 (ugh) ways to specify the cloned device:
979 	 * o pass a parameter block with the clone request.
980 	 * o specify parameters in the text of the clone device name
981 	 * o specify no parameters and get an unattached device that
982 	 *   must be configured separately.
983 	 * The first technique is preferred; the latter two are
984 	 * supported for backwards compatibility.
985 	 *
986 	 * XXXRW: Note historic use of the word "tag" here.  New ioctls may be
987 	 * called for.
988 	 */
989 	if (params) {
990 		error = copyin(params, &vlr, sizeof(vlr));
991 		if (error)
992 			return error;
993 		p = ifunit_ref(vlr.vlr_parent);
994 		if (p == NULL)
995 			return (ENXIO);
996 		error = ifc_name2unit(name, &unit);
997 		if (error != 0) {
998 			if_rele(p);
999 			return (error);
1000 		}
1001 		vid = vlr.vlr_tag;
1002 		wildcard = (unit < 0);
1003 	} else if ((p = vlan_clone_match_ethervid(name, &vid)) != NULL) {
1004 		unit = -1;
1005 		wildcard = 0;
1006 	} else {
1007 		p = NULL;
1008 		error = ifc_name2unit(name, &unit);
1009 		if (error != 0)
1010 			return (error);
1011 
1012 		wildcard = (unit < 0);
1013 	}
1014 
1015 	error = ifc_alloc_unit(ifc, &unit);
1016 	if (error != 0) {
1017 		if (p != NULL)
1018 			if_rele(p);
1019 		return (error);
1020 	}
1021 
1022 	/* In the wildcard case, we need to update the name. */
1023 	if (wildcard) {
1024 		for (dp = name; *dp != '\0'; dp++);
1025 		if (snprintf(dp, len - (dp-name), "%d", unit) >
1026 		    len - (dp-name) - 1) {
1027 			panic("%s: interface name too long", __func__);
1028 		}
1029 	}
1030 
1031 	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
1032 	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
1033 	if (ifp == NULL) {
1034 		ifc_free_unit(ifc, unit);
1035 		free(ifv, M_VLAN);
1036 		if (p != NULL)
1037 			if_rele(p);
1038 		return (ENOSPC);
1039 	}
1040 	CK_SLIST_INIT(&ifv->vlan_mc_listhead);
1041 	ifp->if_softc = ifv;
1042 	/*
1043 	 * Set the name manually rather than using if_initname because
1044 	 * we don't conform to the default naming convention for interfaces.
1045 	 */
1046 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
1047 	ifp->if_dname = vlanname;
1048 	ifp->if_dunit = unit;
1049 	/* NB: flags are not set here */
1050 	ifp->if_linkmib = &ifv->ifv_mib;
1051 	ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
1052 	/* NB: mtu is not set here */
1053 
1054 	ifp->if_init = vlan_init;
1055 	ifp->if_transmit = vlan_transmit;
1056 	ifp->if_qflush = vlan_qflush;
1057 	ifp->if_ioctl = vlan_ioctl;
1058 #ifdef RATELIMIT
1059 	ifp->if_snd_tag_alloc = vlan_snd_tag_alloc;
1060 #endif
1061 	ifp->if_flags = VLAN_IFFLAGS;
1062 	ether_ifattach(ifp, eaddr);
1063 	/* Now undo some of the damage... */
1064 	ifp->if_baudrate = 0;
1065 	ifp->if_type = IFT_L2VLAN;
1066 	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
1067 	ifa = ifp->if_addr;
1068 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1069 	sdl->sdl_type = IFT_L2VLAN;
1070 
1071 	if (p != NULL) {
1072 		error = vlan_config(ifv, p, vid);
1073 		if_rele(p);
1074 		if (error != 0) {
1075 			/*
1076 			 * Since we've partially failed, we need to back
1077 			 * out all the way, otherwise userland could get
1078 			 * confused.  Thus, we destroy the interface.
1079 			 */
1080 			ether_ifdetach(ifp);
1081 			vlan_unconfig(ifp);
1082 			if_free(ifp);
1083 			ifc_free_unit(ifc, unit);
1084 			free(ifv, M_VLAN);
1085 
1086 			return (error);
1087 		}
1088 	}
1089 
1090 	return (0);
1091 }
1092 
1093 static int
1094 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1095 {
1096 	struct ifvlan *ifv = ifp->if_softc;
1097 	int unit = ifp->if_dunit;
1098 
1099 	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
1100 	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
1101 	/*
1102 	 * We should have the only reference to the ifv now, so we can now
1103 	 * drain any remaining lladdr task before freeing the ifnet and the
1104 	 * ifvlan.
1105 	 */
1106 	taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
1107 	NET_EPOCH_WAIT();
1108 	if_free(ifp);
1109 	free(ifv, M_VLAN);
1110 	ifc_free_unit(ifc, unit);
1111 
1112 	return (0);
1113 }
1114 
1115 /*
1116  * The ifp->if_init entry point for vlan(4) is a no-op.
1117  */
1118 static void
1119 vlan_init(void *foo __unused)
1120 {
1121 }
1122 
1123 /*
1124  * The if_transmit method for vlan(4) interface.
1125  */
1126 static int
1127 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1128 {
1129 	struct ifvlan *ifv;
1130 	struct ifnet *p;
1131 	int error, len, mcast;
1132 
1133 	VLAN_RLOCK();
1134 	ifv = ifp->if_softc;
1135 	if (TRUNK(ifv) == NULL) {
1136 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1137 		VLAN_RUNLOCK();
1138 		m_freem(m);
1139 		return (ENETDOWN);
1140 	}
1141 	p = PARENT(ifv);
1142 	len = m->m_pkthdr.len;
1143 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1144 
1145 	BPF_MTAP(ifp, m);
1146 
1147 	/*
1148 	 * Do not run parent's if_transmit() if the parent is not up,
1149 	 * or parent's driver will cause a system crash.
1150 	 */
1151 	if (!UP_AND_RUNNING(p)) {
1152 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1153 		VLAN_RUNLOCK();
1154 		m_freem(m);
1155 		return (ENETDOWN);
1156 	}
1157 
1158 	if (!ether_8021q_frame(&m, ifp, p, ifv->ifv_vid, ifv->ifv_pcp)) {
1159 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1160 		VLAN_RUNLOCK();
1161 		return (0);
1162 	}
1163 
1164 	/*
1165 	 * Send it, precisely as ether_output() would have.
1166 	 */
1167 	error = (p->if_transmit)(p, m);
1168 	if (error == 0) {
1169 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1170 		if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1171 		if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1172 	} else
1173 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1174 	VLAN_RUNLOCK();
1175 	return (error);
1176 }
1177 
1178 /*
1179  * The ifp->if_qflush entry point for vlan(4) is a no-op.
1180  */
1181 static void
1182 vlan_qflush(struct ifnet *ifp __unused)
1183 {
1184 }
1185 
1186 static void
1187 vlan_input(struct ifnet *ifp, struct mbuf *m)
1188 {
1189 	struct ifvlantrunk *trunk;
1190 	struct ifvlan *ifv;
1191 	struct m_tag *mtag;
1192 	uint16_t vid, tag;
1193 
1194 	VLAN_RLOCK();
1195 	trunk = ifp->if_vlantrunk;
1196 	if (trunk == NULL) {
1197 		VLAN_RUNLOCK();
1198 		m_freem(m);
1199 		return;
1200 	}
1201 
1202 	if (m->m_flags & M_VLANTAG) {
1203 		/*
1204 		 * Packet is tagged, but m contains a normal
1205 		 * Ethernet frame; the tag is stored out-of-band.
1206 		 */
1207 		tag = m->m_pkthdr.ether_vtag;
1208 		m->m_flags &= ~M_VLANTAG;
1209 	} else {
1210 		struct ether_vlan_header *evl;
1211 
1212 		/*
1213 		 * Packet is tagged in-band as specified by 802.1q.
1214 		 */
1215 		switch (ifp->if_type) {
1216 		case IFT_ETHER:
1217 			if (m->m_len < sizeof(*evl) &&
1218 			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
1219 				if_printf(ifp, "cannot pullup VLAN header\n");
1220 				VLAN_RUNLOCK();
1221 				return;
1222 			}
1223 			evl = mtod(m, struct ether_vlan_header *);
1224 			tag = ntohs(evl->evl_tag);
1225 
1226 			/*
1227 			 * Remove the 802.1q header by copying the Ethernet
1228 			 * addresses over it and adjusting the beginning of
1229 			 * the data in the mbuf.  The encapsulated Ethernet
1230 			 * type field is already in place.
1231 			 */
1232 			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1233 			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
1234 			m_adj(m, ETHER_VLAN_ENCAP_LEN);
1235 			break;
1236 
1237 		default:
1238 #ifdef INVARIANTS
1239 			panic("%s: %s has unsupported if_type %u",
1240 			      __func__, ifp->if_xname, ifp->if_type);
1241 #endif
1242 			if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1243 			VLAN_RUNLOCK();
1244 			m_freem(m);
1245 			return;
1246 		}
1247 	}
1248 
1249 	vid = EVL_VLANOFTAG(tag);
1250 
1251 	ifv = vlan_gethash(trunk, vid);
1252 	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1253 		VLAN_RUNLOCK();
1254 		if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1255 		m_freem(m);
1256 		return;
1257 	}
1258 
1259 	if (vlan_mtag_pcp) {
1260 		/*
1261 		 * While uncommon, it is possible that we will find a 802.1q
1262 		 * packet encapsulated inside another packet that also had an
1263 		 * 802.1q header.  For example, ethernet tunneled over IPSEC
1264 		 * arriving over ethernet.  In that case, we replace the
1265 		 * existing 802.1q PCP m_tag value.
1266 		 */
1267 		mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1268 		if (mtag == NULL) {
1269 			mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1270 			    sizeof(uint8_t), M_NOWAIT);
1271 			if (mtag == NULL) {
1272 				if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1273 				VLAN_RUNLOCK();
1274 				m_freem(m);
1275 				return;
1276 			}
1277 			m_tag_prepend(m, mtag);
1278 		}
1279 		*(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1280 	}
1281 
1282 	m->m_pkthdr.rcvif = ifv->ifv_ifp;
1283 	if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1284 	VLAN_RUNLOCK();
1285 
1286 	/* Pass it back through the parent's input routine. */
1287 	(*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
1288 }
1289 
1290 static void
1291 vlan_lladdr_fn(void *arg, int pending __unused)
1292 {
1293 	struct ifvlan *ifv;
1294 	struct ifnet *ifp;
1295 
1296 	ifv = (struct ifvlan *)arg;
1297 	ifp = ifv->ifv_ifp;
1298 	/* The ifv_ifp already has the lladdr copied in. */
1299 	if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1300 }
1301 
1302 static int
1303 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid)
1304 {
1305 	struct ifvlantrunk *trunk;
1306 	struct ifnet *ifp;
1307 	int error = 0;
1308 
1309 	/*
1310 	 * We can handle non-ethernet hardware types as long as
1311 	 * they handle the tagging and headers themselves.
1312 	 */
1313 	if (p->if_type != IFT_ETHER &&
1314 	    (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1315 		return (EPROTONOSUPPORT);
1316 	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1317 		return (EPROTONOSUPPORT);
1318 	/*
1319 	 * Don't let the caller set up a VLAN VID with
1320 	 * anything except VLID bits.
1321 	 * VID numbers 0x0 and 0xFFF are reserved.
1322 	 */
1323 	if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1324 		return (EINVAL);
1325 	if (ifv->ifv_trunk)
1326 		return (EBUSY);
1327 
1328 	VLAN_XLOCK();
1329 	if (p->if_vlantrunk == NULL) {
1330 		trunk = malloc(sizeof(struct ifvlantrunk),
1331 		    M_VLAN, M_WAITOK | M_ZERO);
1332 		vlan_inithash(trunk);
1333 		TRUNK_LOCK_INIT(trunk);
1334 		TRUNK_WLOCK(trunk);
1335 		p->if_vlantrunk = trunk;
1336 		trunk->parent = p;
1337 		if_ref(trunk->parent);
1338 		TRUNK_WUNLOCK(trunk);
1339 	} else {
1340 		trunk = p->if_vlantrunk;
1341 	}
1342 
1343 	ifv->ifv_vid = vid;	/* must set this before vlan_inshash() */
1344 	ifv->ifv_pcp = 0;       /* Default: best effort delivery. */
1345 	vlan_tag_recalculate(ifv);
1346 	error = vlan_inshash(trunk, ifv);
1347 	if (error)
1348 		goto done;
1349 	ifv->ifv_proto = ETHERTYPE_VLAN;
1350 	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1351 	ifv->ifv_mintu = ETHERMIN;
1352 	ifv->ifv_pflags = 0;
1353 	ifv->ifv_capenable = -1;
1354 
1355 	/*
1356 	 * If the parent supports the VLAN_MTU capability,
1357 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1358 	 * use it.
1359 	 */
1360 	if (p->if_capenable & IFCAP_VLAN_MTU) {
1361 		/*
1362 		 * No need to fudge the MTU since the parent can
1363 		 * handle extended frames.
1364 		 */
1365 		ifv->ifv_mtufudge = 0;
1366 	} else {
1367 		/*
1368 		 * Fudge the MTU by the encapsulation size.  This
1369 		 * makes us incompatible with strictly compliant
1370 		 * 802.1Q implementations, but allows us to use
1371 		 * the feature with other NetBSD implementations,
1372 		 * which might still be useful.
1373 		 */
1374 		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1375 	}
1376 
1377 	ifv->ifv_trunk = trunk;
1378 	ifp = ifv->ifv_ifp;
1379 	/*
1380 	 * Initialize fields from our parent.  This duplicates some
1381 	 * work with ether_ifattach() but allows for non-ethernet
1382 	 * interfaces to also work.
1383 	 */
1384 	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1385 	ifp->if_baudrate = p->if_baudrate;
1386 	ifp->if_output = p->if_output;
1387 	ifp->if_input = p->if_input;
1388 	ifp->if_resolvemulti = p->if_resolvemulti;
1389 	ifp->if_addrlen = p->if_addrlen;
1390 	ifp->if_broadcastaddr = p->if_broadcastaddr;
1391 	ifp->if_pcp = ifv->ifv_pcp;
1392 
1393 	/*
1394 	 * Copy only a selected subset of flags from the parent.
1395 	 * Other flags are none of our business.
1396 	 */
1397 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1398 	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1399 	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1400 #undef VLAN_COPY_FLAGS
1401 
1402 	ifp->if_link_state = p->if_link_state;
1403 
1404 	TRUNK_RLOCK(TRUNK(ifv));
1405 	vlan_capabilities(ifv);
1406 	TRUNK_RUNLOCK(TRUNK(ifv));
1407 
1408 	/*
1409 	 * Set up our interface address to reflect the underlying
1410 	 * physical interface's.
1411 	 */
1412 	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1413 	((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1414 	    p->if_addrlen;
1415 
1416 	TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
1417 
1418 	/* We are ready for operation now. */
1419 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1420 
1421 	/* Update flags on the parent, if necessary. */
1422 	vlan_setflags(ifp, 1);
1423 
1424 	/*
1425 	 * Configure multicast addresses that may already be
1426 	 * joined on the vlan device.
1427 	 */
1428 	(void)vlan_setmulti(ifp);
1429 
1430 done:
1431 	if (error == 0)
1432 		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1433 	VLAN_XUNLOCK();
1434 
1435 	return (error);
1436 }
1437 
1438 static void
1439 vlan_unconfig(struct ifnet *ifp)
1440 {
1441 
1442 	VLAN_XLOCK();
1443 	vlan_unconfig_locked(ifp, 0);
1444 	VLAN_XUNLOCK();
1445 }
1446 
1447 static void
1448 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1449 {
1450 	struct ifvlantrunk *trunk;
1451 	struct vlan_mc_entry *mc;
1452 	struct ifvlan *ifv;
1453 	struct ifnet  *parent;
1454 	int error;
1455 
1456 	VLAN_XLOCK_ASSERT();
1457 
1458 	ifv = ifp->if_softc;
1459 	trunk = ifv->ifv_trunk;
1460 	parent = NULL;
1461 
1462 	if (trunk != NULL) {
1463 		parent = trunk->parent;
1464 
1465 		/*
1466 		 * Since the interface is being unconfigured, we need to
1467 		 * empty the list of multicast groups that we may have joined
1468 		 * while we were alive from the parent's list.
1469 		 */
1470 		while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1471 			/*
1472 			 * If the parent interface is being detached,
1473 			 * all its multicast addresses have already
1474 			 * been removed.  Warn about errors if
1475 			 * if_delmulti() does fail, but don't abort as
1476 			 * all callers expect vlan destruction to
1477 			 * succeed.
1478 			 */
1479 			if (!departing) {
1480 				error = if_delmulti(parent,
1481 				    (struct sockaddr *)&mc->mc_addr);
1482 				if (error)
1483 					if_printf(ifp,
1484 		    "Failed to delete multicast address from parent: %d\n",
1485 					    error);
1486 			}
1487 			CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1488 			NET_EPOCH_WAIT();
1489 			free(mc, M_VLAN);
1490 		}
1491 
1492 		vlan_setflags(ifp, 0); /* clear special flags on parent */
1493 
1494 		vlan_remhash(trunk, ifv);
1495 		ifv->ifv_trunk = NULL;
1496 
1497 		/*
1498 		 * Check if we were the last.
1499 		 */
1500 		if (trunk->refcnt == 0) {
1501 			parent->if_vlantrunk = NULL;
1502 			NET_EPOCH_WAIT();
1503 			trunk_destroy(trunk);
1504 		}
1505 	}
1506 
1507 	/* Disconnect from parent. */
1508 	if (ifv->ifv_pflags)
1509 		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1510 	ifp->if_mtu = ETHERMTU;
1511 	ifp->if_link_state = LINK_STATE_UNKNOWN;
1512 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1513 
1514 	/*
1515 	 * Only dispatch an event if vlan was
1516 	 * attached, otherwise there is nothing
1517 	 * to cleanup anyway.
1518 	 */
1519 	if (parent != NULL)
1520 		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1521 }
1522 
1523 /* Handle a reference counted flag that should be set on the parent as well */
1524 static int
1525 vlan_setflag(struct ifnet *ifp, int flag, int status,
1526 	     int (*func)(struct ifnet *, int))
1527 {
1528 	struct ifvlan *ifv;
1529 	int error;
1530 
1531 	VLAN_SXLOCK_ASSERT();
1532 
1533 	ifv = ifp->if_softc;
1534 	status = status ? (ifp->if_flags & flag) : 0;
1535 	/* Now "status" contains the flag value or 0 */
1536 
1537 	/*
1538 	 * See if recorded parent's status is different from what
1539 	 * we want it to be.  If it is, flip it.  We record parent's
1540 	 * status in ifv_pflags so that we won't clear parent's flag
1541 	 * we haven't set.  In fact, we don't clear or set parent's
1542 	 * flags directly, but get or release references to them.
1543 	 * That's why we can be sure that recorded flags still are
1544 	 * in accord with actual parent's flags.
1545 	 */
1546 	if (status != (ifv->ifv_pflags & flag)) {
1547 		error = (*func)(PARENT(ifv), status);
1548 		if (error)
1549 			return (error);
1550 		ifv->ifv_pflags &= ~flag;
1551 		ifv->ifv_pflags |= status;
1552 	}
1553 	return (0);
1554 }
1555 
1556 /*
1557  * Handle IFF_* flags that require certain changes on the parent:
1558  * if "status" is true, update parent's flags respective to our if_flags;
1559  * if "status" is false, forcedly clear the flags set on parent.
1560  */
1561 static int
1562 vlan_setflags(struct ifnet *ifp, int status)
1563 {
1564 	int error, i;
1565 
1566 	for (i = 0; vlan_pflags[i].flag; i++) {
1567 		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1568 				     status, vlan_pflags[i].func);
1569 		if (error)
1570 			return (error);
1571 	}
1572 	return (0);
1573 }
1574 
1575 /* Inform all vlans that their parent has changed link state */
1576 static void
1577 vlan_link_state(struct ifnet *ifp)
1578 {
1579 	struct ifvlantrunk *trunk;
1580 	struct ifvlan *ifv;
1581 
1582 	/* Called from a taskqueue_swi task, so we cannot sleep. */
1583 	VLAN_RLOCK();
1584 	trunk = ifp->if_vlantrunk;
1585 	if (trunk == NULL) {
1586 		VLAN_RUNLOCK();
1587 		return;
1588 	}
1589 
1590 	TRUNK_WLOCK(trunk);
1591 	VLAN_FOREACH(ifv, trunk) {
1592 		ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1593 		if_link_state_change(ifv->ifv_ifp,
1594 		    trunk->parent->if_link_state);
1595 	}
1596 	TRUNK_WUNLOCK(trunk);
1597 	VLAN_RUNLOCK();
1598 }
1599 
1600 static void
1601 vlan_capabilities(struct ifvlan *ifv)
1602 {
1603 	struct ifnet *p;
1604 	struct ifnet *ifp;
1605 	struct ifnet_hw_tsomax hw_tsomax;
1606 	int cap = 0, ena = 0, mena;
1607 	u_long hwa = 0;
1608 
1609 	VLAN_SXLOCK_ASSERT();
1610 	TRUNK_RLOCK_ASSERT(TRUNK(ifv));
1611 	p = PARENT(ifv);
1612 	ifp = ifv->ifv_ifp;
1613 
1614 	/* Mask parent interface enabled capabilities disabled by user. */
1615 	mena = p->if_capenable & ifv->ifv_capenable;
1616 
1617 	/*
1618 	 * If the parent interface can do checksum offloading
1619 	 * on VLANs, then propagate its hardware-assisted
1620 	 * checksumming flags. Also assert that checksum
1621 	 * offloading requires hardware VLAN tagging.
1622 	 */
1623 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1624 		cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1625 	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1626 	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1627 		ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1628 		if (ena & IFCAP_TXCSUM)
1629 			hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
1630 			    CSUM_UDP | CSUM_SCTP);
1631 		if (ena & IFCAP_TXCSUM_IPV6)
1632 			hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
1633 			    CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
1634 	}
1635 
1636 	/*
1637 	 * If the parent interface can do TSO on VLANs then
1638 	 * propagate the hardware-assisted flag. TSO on VLANs
1639 	 * does not necessarily require hardware VLAN tagging.
1640 	 */
1641 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1642 	if_hw_tsomax_common(p, &hw_tsomax);
1643 	if_hw_tsomax_update(ifp, &hw_tsomax);
1644 	if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1645 		cap |= p->if_capabilities & IFCAP_TSO;
1646 	if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1647 		ena |= mena & IFCAP_TSO;
1648 		if (ena & IFCAP_TSO)
1649 			hwa |= p->if_hwassist & CSUM_TSO;
1650 	}
1651 
1652 	/*
1653 	 * If the parent interface can do LRO and checksum offloading on
1654 	 * VLANs, then guess it may do LRO on VLANs.  False positive here
1655 	 * cost nothing, while false negative may lead to some confusions.
1656 	 */
1657 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1658 		cap |= p->if_capabilities & IFCAP_LRO;
1659 	if (p->if_capenable & IFCAP_VLAN_HWCSUM)
1660 		ena |= p->if_capenable & IFCAP_LRO;
1661 
1662 	/*
1663 	 * If the parent interface can offload TCP connections over VLANs then
1664 	 * propagate its TOE capability to the VLAN interface.
1665 	 *
1666 	 * All TOE drivers in the tree today can deal with VLANs.  If this
1667 	 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1668 	 * with its own bit.
1669 	 */
1670 #define	IFCAP_VLAN_TOE IFCAP_TOE
1671 	if (p->if_capabilities & IFCAP_VLAN_TOE)
1672 		cap |= p->if_capabilities & IFCAP_TOE;
1673 	if (p->if_capenable & IFCAP_VLAN_TOE) {
1674 		TOEDEV(ifp) = TOEDEV(p);
1675 		ena |= mena & IFCAP_TOE;
1676 	}
1677 
1678 	/*
1679 	 * If the parent interface supports dynamic link state, so does the
1680 	 * VLAN interface.
1681 	 */
1682 	cap |= (p->if_capabilities & IFCAP_LINKSTATE);
1683 	ena |= (mena & IFCAP_LINKSTATE);
1684 
1685 #ifdef RATELIMIT
1686 	/*
1687 	 * If the parent interface supports ratelimiting, so does the
1688 	 * VLAN interface.
1689 	 */
1690 	cap |= (p->if_capabilities & IFCAP_TXRTLMT);
1691 	ena |= (mena & IFCAP_TXRTLMT);
1692 #endif
1693 
1694 	ifp->if_capabilities = cap;
1695 	ifp->if_capenable = ena;
1696 	ifp->if_hwassist = hwa;
1697 }
1698 
1699 static void
1700 vlan_trunk_capabilities(struct ifnet *ifp)
1701 {
1702 	struct ifvlantrunk *trunk;
1703 	struct ifvlan *ifv;
1704 
1705 	VLAN_SLOCK();
1706 	trunk = ifp->if_vlantrunk;
1707 	if (trunk == NULL) {
1708 		VLAN_SUNLOCK();
1709 		return;
1710 	}
1711 	TRUNK_RLOCK(trunk);
1712 	VLAN_FOREACH(ifv, trunk) {
1713 		vlan_capabilities(ifv);
1714 	}
1715 	TRUNK_RUNLOCK(trunk);
1716 	VLAN_SUNLOCK();
1717 }
1718 
1719 static int
1720 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1721 {
1722 	struct ifnet *p;
1723 	struct ifreq *ifr;
1724 	struct ifaddr *ifa;
1725 	struct ifvlan *ifv;
1726 	struct ifvlantrunk *trunk;
1727 	struct vlanreq vlr;
1728 	int error = 0;
1729 
1730 	ifr = (struct ifreq *)data;
1731 	ifa = (struct ifaddr *) data;
1732 	ifv = ifp->if_softc;
1733 
1734 	switch (cmd) {
1735 	case SIOCSIFADDR:
1736 		ifp->if_flags |= IFF_UP;
1737 #ifdef INET
1738 		if (ifa->ifa_addr->sa_family == AF_INET)
1739 			arp_ifinit(ifp, ifa);
1740 #endif
1741 		break;
1742 	case SIOCGIFADDR:
1743 		bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
1744 		    ifp->if_addrlen);
1745 		break;
1746 	case SIOCGIFMEDIA:
1747 		VLAN_SLOCK();
1748 		if (TRUNK(ifv) != NULL) {
1749 			p = PARENT(ifv);
1750 			if_ref(p);
1751 			error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1752 			if_rele(p);
1753 			/* Limit the result to the parent's current config. */
1754 			if (error == 0) {
1755 				struct ifmediareq *ifmr;
1756 
1757 				ifmr = (struct ifmediareq *)data;
1758 				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1759 					ifmr->ifm_count = 1;
1760 					error = copyout(&ifmr->ifm_current,
1761 						ifmr->ifm_ulist,
1762 						sizeof(int));
1763 				}
1764 			}
1765 		} else {
1766 			error = EINVAL;
1767 		}
1768 		VLAN_SUNLOCK();
1769 		break;
1770 
1771 	case SIOCSIFMEDIA:
1772 		error = EINVAL;
1773 		break;
1774 
1775 	case SIOCSIFMTU:
1776 		/*
1777 		 * Set the interface MTU.
1778 		 */
1779 		VLAN_SLOCK();
1780 		trunk = TRUNK(ifv);
1781 		if (trunk != NULL) {
1782 			TRUNK_WLOCK(trunk);
1783 			if (ifr->ifr_mtu >
1784 			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1785 			    ifr->ifr_mtu <
1786 			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1787 				error = EINVAL;
1788 			else
1789 				ifp->if_mtu = ifr->ifr_mtu;
1790 			TRUNK_WUNLOCK(trunk);
1791 		} else
1792 			error = EINVAL;
1793 		VLAN_SUNLOCK();
1794 		break;
1795 
1796 	case SIOCSETVLAN:
1797 #ifdef VIMAGE
1798 		/*
1799 		 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1800 		 * interface to be delegated to a jail without allowing the
1801 		 * jail to change what underlying interface/VID it is
1802 		 * associated with.  We are not entirely convinced that this
1803 		 * is the right way to accomplish that policy goal.
1804 		 */
1805 		if (ifp->if_vnet != ifp->if_home_vnet) {
1806 			error = EPERM;
1807 			break;
1808 		}
1809 #endif
1810 		error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
1811 		if (error)
1812 			break;
1813 		if (vlr.vlr_parent[0] == '\0') {
1814 			vlan_unconfig(ifp);
1815 			break;
1816 		}
1817 		p = ifunit_ref(vlr.vlr_parent);
1818 		if (p == NULL) {
1819 			error = ENOENT;
1820 			break;
1821 		}
1822 		error = vlan_config(ifv, p, vlr.vlr_tag);
1823 		if_rele(p);
1824 		break;
1825 
1826 	case SIOCGETVLAN:
1827 #ifdef VIMAGE
1828 		if (ifp->if_vnet != ifp->if_home_vnet) {
1829 			error = EPERM;
1830 			break;
1831 		}
1832 #endif
1833 		bzero(&vlr, sizeof(vlr));
1834 		VLAN_SLOCK();
1835 		if (TRUNK(ifv) != NULL) {
1836 			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1837 			    sizeof(vlr.vlr_parent));
1838 			vlr.vlr_tag = ifv->ifv_vid;
1839 		}
1840 		VLAN_SUNLOCK();
1841 		error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
1842 		break;
1843 
1844 	case SIOCSIFFLAGS:
1845 		/*
1846 		 * We should propagate selected flags to the parent,
1847 		 * e.g., promiscuous mode.
1848 		 */
1849 		VLAN_XLOCK();
1850 		if (TRUNK(ifv) != NULL)
1851 			error = vlan_setflags(ifp, 1);
1852 		VLAN_XUNLOCK();
1853 		break;
1854 
1855 	case SIOCADDMULTI:
1856 	case SIOCDELMULTI:
1857 		/*
1858 		 * If we don't have a parent, just remember the membership for
1859 		 * when we do.
1860 		 *
1861 		 * XXX We need the rmlock here to avoid sleeping while
1862 		 * holding in6_multi_mtx.
1863 		 */
1864 		VLAN_XLOCK();
1865 		trunk = TRUNK(ifv);
1866 		if (trunk != NULL)
1867 			error = vlan_setmulti(ifp);
1868 		VLAN_XUNLOCK();
1869 
1870 		break;
1871 	case SIOCGVLANPCP:
1872 #ifdef VIMAGE
1873 		if (ifp->if_vnet != ifp->if_home_vnet) {
1874 			error = EPERM;
1875 			break;
1876 		}
1877 #endif
1878 		ifr->ifr_vlan_pcp = ifv->ifv_pcp;
1879 		break;
1880 
1881 	case SIOCSVLANPCP:
1882 #ifdef VIMAGE
1883 		if (ifp->if_vnet != ifp->if_home_vnet) {
1884 			error = EPERM;
1885 			break;
1886 		}
1887 #endif
1888 		error = priv_check(curthread, PRIV_NET_SETVLANPCP);
1889 		if (error)
1890 			break;
1891 		if (ifr->ifr_vlan_pcp > 7) {
1892 			error = EINVAL;
1893 			break;
1894 		}
1895 		ifv->ifv_pcp = ifr->ifr_vlan_pcp;
1896 		ifp->if_pcp = ifv->ifv_pcp;
1897 		vlan_tag_recalculate(ifv);
1898 		/* broadcast event about PCP change */
1899 		EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
1900 		break;
1901 
1902 	case SIOCSIFCAP:
1903 		VLAN_SLOCK();
1904 		ifv->ifv_capenable = ifr->ifr_reqcap;
1905 		trunk = TRUNK(ifv);
1906 		if (trunk != NULL) {
1907 			TRUNK_RLOCK(trunk);
1908 			vlan_capabilities(ifv);
1909 			TRUNK_RUNLOCK(trunk);
1910 		}
1911 		VLAN_SUNLOCK();
1912 		break;
1913 
1914 	default:
1915 		error = EINVAL;
1916 		break;
1917 	}
1918 
1919 	return (error);
1920 }
1921 
1922 #ifdef RATELIMIT
1923 static int
1924 vlan_snd_tag_alloc(struct ifnet *ifp,
1925     union if_snd_tag_alloc_params *params,
1926     struct m_snd_tag **ppmt)
1927 {
1928 
1929 	/* get trunk device */
1930 	ifp = vlan_trunkdev(ifp);
1931 	if (ifp == NULL || (ifp->if_capenable & IFCAP_TXRTLMT) == 0)
1932 		return (EOPNOTSUPP);
1933 	/* forward allocation request */
1934 	return (ifp->if_snd_tag_alloc(ifp, params, ppmt));
1935 }
1936 #endif
1937