xref: /freebsd/sys/net/if_vlan.c (revision e4e9813eb92cd7c4d4b819a8fbed5cbd3d92f5d8)
1 /*-
2  * Copyright 1998 Massachusetts Institute of Technology
3  *
4  * Permission to use, copy, modify, and distribute this software and
5  * its documentation for any purpose and without fee is hereby
6  * granted, provided that both the above copyright notice and this
7  * permission notice appear in all copies, that both the above
8  * copyright notice and this permission notice appear in all
9  * supporting documentation, and that the name of M.I.T. not be used
10  * in advertising or publicity pertaining to distribution of the
11  * software without specific, written prior permission.  M.I.T. makes
12  * no representations about the suitability of this software for any
13  * purpose.  It is provided "as is" without express or implied
14  * warranty.
15  *
16  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
17  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
18  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
19  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
20  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
23  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
24  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
25  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
26  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * $FreeBSD$
30  */
31 
32 /*
33  * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
34  * Might be extended some day to also handle IEEE 802.1p priority
35  * tagging.  This is sort of sneaky in the implementation, since
36  * we need to pretend to be enough of an Ethernet implementation
37  * to make arp work.  The way we do this is by telling everyone
38  * that we are an Ethernet, and then catch the packets that
39  * ether_output() left on our output queue when it calls
40  * if_start(), rewrite them for use by the real outgoing interface,
41  * and ask it to send them.
42  */
43 
44 #include "opt_inet.h"
45 #include "opt_vlan.h"
46 
47 #include <sys/param.h>
48 #include <sys/kernel.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/module.h>
53 #include <sys/rwlock.h>
54 #include <sys/queue.h>
55 #include <sys/socket.h>
56 #include <sys/sockio.h>
57 #include <sys/sysctl.h>
58 #include <sys/systm.h>
59 
60 #include <net/bpf.h>
61 #include <net/ethernet.h>
62 #include <net/if.h>
63 #include <net/if_clone.h>
64 #include <net/if_arp.h>
65 #include <net/if_dl.h>
66 #include <net/if_types.h>
67 #include <net/if_vlan_var.h>
68 
69 #ifdef INET
70 #include <netinet/in.h>
71 #include <netinet/if_ether.h>
72 #endif
73 
74 #define VLANNAME	"vlan"
75 #define	VLAN_DEF_HWIDTH	4
76 #define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
77 
78 LIST_HEAD(ifvlanhead, ifvlan);
79 
80 struct ifvlantrunk {
81 	struct	ifnet   *parent;	/* parent interface of this trunk */
82 	struct	rwlock	rw;
83 #ifdef VLAN_ARRAY
84 #define	VLAN_ARRAY_SIZE	(EVL_VLID_MASK + 1)
85 	struct	ifvlan	*vlans[VLAN_ARRAY_SIZE]; /* static table */
86 #else
87 	struct	ifvlanhead *hash;	/* dynamic hash-list table */
88 	uint16_t	hmask;
89 	uint16_t	hwidth;
90 #endif
91 	int		refcnt;
92 	LIST_ENTRY(ifvlantrunk) trunk_entry;
93 };
94 static LIST_HEAD(, ifvlantrunk) trunk_list;
95 
96 struct vlan_mc_entry {
97 	struct ether_addr		mc_addr;
98 	SLIST_ENTRY(vlan_mc_entry)	mc_entries;
99 };
100 
101 struct	ifvlan {
102 	struct	ifvlantrunk *ifv_trunk;
103 	struct	ifnet *ifv_ifp;
104 #define	TRUNK(ifv)	((ifv)->ifv_trunk)
105 #define	PARENT(ifv)	((ifv)->ifv_trunk->parent)
106 	int	ifv_pflags;	/* special flags we have set on parent */
107 	struct	ifv_linkmib {
108 		int	ifvm_encaplen;	/* encapsulation length */
109 		int	ifvm_mtufudge;	/* MTU fudged by this much */
110 		int	ifvm_mintu;	/* min transmission unit */
111 		uint16_t ifvm_tag;	/* tag to apply on packets leaving if */
112 	}	ifv_mib;
113 	SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
114 	LIST_ENTRY(ifvlan) ifv_list;
115 };
116 #define	ifv_tag		ifv_mib.ifvm_tag
117 #define	ifv_encaplen	ifv_mib.ifvm_encaplen
118 #define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
119 #define	ifv_mintu	ifv_mib.ifvm_mintu
120 
121 /* Special flags we should propagate to parent. */
122 static struct {
123 	int flag;
124 	int (*func)(struct ifnet *, int);
125 } vlan_pflags[] = {
126 	{IFF_PROMISC, ifpromisc},
127 	{IFF_ALLMULTI, if_allmulti},
128 	{0, NULL}
129 };
130 
131 SYSCTL_DECL(_net_link);
132 SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0, "IEEE 802.1Q VLAN");
133 SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0, "for consistency");
134 
135 static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface");
136 
137 static eventhandler_tag ifdetach_tag;
138 
139 /*
140  * We have a global mutex, that is used to serialize configuration
141  * changes and isn't used in normal packet delivery.
142  *
143  * We also have a per-trunk rwlock, that is locked shared on packet
144  * processing and exclusive when configuration is changed.
145  *
146  * The VLAN_ARRAY substitutes the dynamic hash with a static array
147  * with 4096 entries. In theory this can give a boots in processing,
148  * however on practice it does not. Probably this is because array
149  * is too big to fit into CPU cache.
150  */
151 static struct mtx ifv_mtx;
152 #define	VLAN_LOCK_INIT()	mtx_init(&ifv_mtx, "vlan_global", NULL, MTX_DEF)
153 #define	VLAN_LOCK_DESTROY()	mtx_destroy(&ifv_mtx)
154 #define	VLAN_LOCK_ASSERT()	mtx_assert(&ifv_mtx, MA_OWNED)
155 #define	VLAN_LOCK()		mtx_lock(&ifv_mtx)
156 #define	VLAN_UNLOCK()		mtx_unlock(&ifv_mtx)
157 #define	TRUNK_LOCK_INIT(trunk)	rw_init(&(trunk)->rw, VLANNAME)
158 #define	TRUNK_LOCK_DESTROY(trunk) rw_destroy(&(trunk)->rw)
159 #define	TRUNK_LOCK(trunk)	rw_wlock(&(trunk)->rw)
160 #define	TRUNK_UNLOCK(trunk)	rw_wunlock(&(trunk)->rw)
161 #define	TRUNK_LOCK_ASSERT(trunk) rw_assert(&(trunk)->rw, RA_WLOCKED)
162 #define	TRUNK_RLOCK(trunk)	rw_rlock(&(trunk)->rw)
163 #define	TRUNK_RUNLOCK(trunk)	rw_runlock(&(trunk)->rw)
164 #define	TRUNK_LOCK_RASSERT(trunk) rw_assert(&(trunk)->rw, RA_RLOCKED)
165 
166 #ifndef VLAN_ARRAY
167 static	void vlan_inithash(struct ifvlantrunk *trunk);
168 static	void vlan_freehash(struct ifvlantrunk *trunk);
169 static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
170 static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
171 static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
172 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
173 	uint16_t tag);
174 #endif
175 static	void trunk_destroy(struct ifvlantrunk *trunk);
176 
177 static	void vlan_start(struct ifnet *ifp);
178 static	void vlan_init(void *foo);
179 static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
180 static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
181 static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
182     int (*func)(struct ifnet *, int));
183 static	int vlan_setflags(struct ifnet *ifp, int status);
184 static	int vlan_setmulti(struct ifnet *ifp);
185 static	int vlan_unconfig(struct ifnet *ifp);
186 static	int vlan_unconfig_locked(struct ifnet *ifp);
187 static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
188 static	void vlan_link_state(struct ifnet *ifp, int link);
189 static	void vlan_capabilities(struct ifvlan *ifv);
190 static	void vlan_trunk_capabilities(struct ifnet *ifp);
191 
192 static	struct ifnet *vlan_clone_match_ethertag(struct if_clone *,
193     const char *, int *);
194 static	int vlan_clone_match(struct if_clone *, const char *);
195 static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
196 static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
197 
198 static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
199 
200 static	struct if_clone vlan_cloner = IFC_CLONE_INITIALIZER(VLANNAME, NULL,
201     IF_MAXUNIT, NULL, vlan_clone_match, vlan_clone_create, vlan_clone_destroy);
202 
203 #ifndef VLAN_ARRAY
204 #define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
205 
206 static void
207 vlan_inithash(struct ifvlantrunk *trunk)
208 {
209 	int i, n;
210 
211 	/*
212 	 * The trunk must not be locked here since we call malloc(M_WAITOK).
213 	 * It is OK in case this function is called before the trunk struct
214 	 * gets hooked up and becomes visible from other threads.
215 	 */
216 
217 	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
218 	    ("%s: hash already initialized", __func__));
219 
220 	trunk->hwidth = VLAN_DEF_HWIDTH;
221 	n = 1 << trunk->hwidth;
222 	trunk->hmask = n - 1;
223 	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
224 	for (i = 0; i < n; i++)
225 		LIST_INIT(&trunk->hash[i]);
226 }
227 
228 static void
229 vlan_freehash(struct ifvlantrunk *trunk)
230 {
231 #ifdef INVARIANTS
232 	int i;
233 
234 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
235 	for (i = 0; i < (1 << trunk->hwidth); i++)
236 		KASSERT(LIST_EMPTY(&trunk->hash[i]),
237 		    ("%s: hash table not empty", __func__));
238 #endif
239 	free(trunk->hash, M_VLAN);
240 	trunk->hash = NULL;
241 	trunk->hwidth = trunk->hmask = 0;
242 }
243 
244 static int
245 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
246 {
247 	int i, b;
248 	struct ifvlan *ifv2;
249 
250 	TRUNK_LOCK_ASSERT(trunk);
251 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
252 
253 	b = 1 << trunk->hwidth;
254 	i = HASH(ifv->ifv_tag, trunk->hmask);
255 	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
256 		if (ifv->ifv_tag == ifv2->ifv_tag)
257 			return (EEXIST);
258 
259 	/*
260 	 * Grow the hash when the number of vlans exceeds half of the number of
261 	 * hash buckets squared. This will make the average linked-list length
262 	 * buckets/2.
263 	 */
264 	if (trunk->refcnt > (b * b) / 2) {
265 		vlan_growhash(trunk, 1);
266 		i = HASH(ifv->ifv_tag, trunk->hmask);
267 	}
268 	LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
269 	trunk->refcnt++;
270 
271 	return (0);
272 }
273 
274 static int
275 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
276 {
277 	int i, b;
278 	struct ifvlan *ifv2;
279 
280 	TRUNK_LOCK_ASSERT(trunk);
281 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
282 
283 	b = 1 << trunk->hwidth;
284 	i = HASH(ifv->ifv_tag, trunk->hmask);
285 	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
286 		if (ifv2 == ifv) {
287 			trunk->refcnt--;
288 			LIST_REMOVE(ifv2, ifv_list);
289 			if (trunk->refcnt < (b * b) / 2)
290 				vlan_growhash(trunk, -1);
291 			return (0);
292 		}
293 
294 	panic("%s: vlan not found\n", __func__);
295 	return (ENOENT); /*NOTREACHED*/
296 }
297 
298 /*
299  * Grow the hash larger or smaller if memory permits.
300  */
301 static void
302 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
303 {
304 
305 	struct ifvlan *ifv;
306 	struct ifvlanhead *hash2;
307 	int hwidth2, i, j, n, n2;
308 
309 	TRUNK_LOCK_ASSERT(trunk);
310 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
311 
312 	if (howmuch == 0) {
313 		/* Harmless yet obvious coding error */
314 		printf("%s: howmuch is 0\n", __func__);
315 		return;
316 	}
317 
318 	hwidth2 = trunk->hwidth + howmuch;
319 	n = 1 << trunk->hwidth;
320 	n2 = 1 << hwidth2;
321 	/* Do not shrink the table below the default */
322 	if (hwidth2 < VLAN_DEF_HWIDTH)
323 		return;
324 
325 	/* M_NOWAIT because we're called with trunk mutex held */
326 	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
327 	if (hash2 == NULL) {
328 		printf("%s: out of memory -- hash size not changed\n",
329 		    __func__);
330 		return;		/* We can live with the old hash table */
331 	}
332 	for (j = 0; j < n2; j++)
333 		LIST_INIT(&hash2[j]);
334 	for (i = 0; i < n; i++)
335 		while (!LIST_EMPTY(&trunk->hash[i])) {
336 			ifv = LIST_FIRST(&trunk->hash[i]);
337 			LIST_REMOVE(ifv, ifv_list);
338 			j = HASH(ifv->ifv_tag, n2 - 1);
339 			LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
340 		}
341 	free(trunk->hash, M_VLAN);
342 	trunk->hash = hash2;
343 	trunk->hwidth = hwidth2;
344 	trunk->hmask = n2 - 1;
345 }
346 
347 static __inline struct ifvlan *
348 vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
349 {
350 	struct ifvlan *ifv;
351 
352 	TRUNK_LOCK_RASSERT(trunk);
353 
354 	LIST_FOREACH(ifv, &trunk->hash[HASH(tag, trunk->hmask)], ifv_list)
355 		if (ifv->ifv_tag == tag)
356 			return (ifv);
357 	return (NULL);
358 }
359 
360 #if 0
361 /* Debugging code to view the hashtables. */
362 static void
363 vlan_dumphash(struct ifvlantrunk *trunk)
364 {
365 	int i;
366 	struct ifvlan *ifv;
367 
368 	for (i = 0; i < (1 << trunk->hwidth); i++) {
369 		printf("%d: ", i);
370 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
371 			printf("%s ", ifv->ifv_ifp->if_xname);
372 		printf("\n");
373 	}
374 }
375 #endif /* 0 */
376 #endif /* !VLAN_ARRAY */
377 
378 static void
379 trunk_destroy(struct ifvlantrunk *trunk)
380 {
381 	VLAN_LOCK_ASSERT();
382 
383 	TRUNK_LOCK(trunk);
384 #ifndef VLAN_ARRAY
385 	vlan_freehash(trunk);
386 #endif
387 	trunk->parent->if_vlantrunk = NULL;
388 	LIST_REMOVE(trunk, trunk_entry);
389 	TRUNK_UNLOCK(trunk);
390 	TRUNK_LOCK_DESTROY(trunk);
391 	free(trunk, M_VLAN);
392 }
393 
394 /*
395  * Program our multicast filter. What we're actually doing is
396  * programming the multicast filter of the parent. This has the
397  * side effect of causing the parent interface to receive multicast
398  * traffic that it doesn't really want, which ends up being discarded
399  * later by the upper protocol layers. Unfortunately, there's no way
400  * to avoid this: there really is only one physical interface.
401  *
402  * XXX: There is a possible race here if more than one thread is
403  *      modifying the multicast state of the vlan interface at the same time.
404  */
405 static int
406 vlan_setmulti(struct ifnet *ifp)
407 {
408 	struct ifnet		*ifp_p;
409 	struct ifmultiaddr	*ifma, *rifma = NULL;
410 	struct ifvlan		*sc;
411 	struct vlan_mc_entry	*mc = NULL;
412 	struct sockaddr_dl	sdl;
413 	int			error;
414 
415 	/*VLAN_LOCK_ASSERT();*/
416 
417 	/* Find the parent. */
418 	sc = ifp->if_softc;
419 	ifp_p = PARENT(sc);
420 
421 	bzero((char *)&sdl, sizeof(sdl));
422 	sdl.sdl_len = sizeof(sdl);
423 	sdl.sdl_family = AF_LINK;
424 	sdl.sdl_index = ifp_p->if_index;
425 	sdl.sdl_type = IFT_ETHER;
426 	sdl.sdl_alen = ETHER_ADDR_LEN;
427 
428 	/* First, remove any existing filter entries. */
429 	while (SLIST_FIRST(&sc->vlan_mc_listhead) != NULL) {
430 		mc = SLIST_FIRST(&sc->vlan_mc_listhead);
431 		bcopy((char *)&mc->mc_addr, LLADDR(&sdl), ETHER_ADDR_LEN);
432 		error = if_delmulti(ifp_p, (struct sockaddr *)&sdl);
433 		if (error)
434 			return (error);
435 		SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
436 		free(mc, M_VLAN);
437 	}
438 
439 	/* Now program new ones. */
440 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
441 		if (ifma->ifma_addr->sa_family != AF_LINK)
442 			continue;
443 		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
444 		if (mc == NULL)
445 			return (ENOMEM);
446 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
447 		    (char *)&mc->mc_addr, ETHER_ADDR_LEN);
448 		SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
449 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
450 		    LLADDR(&sdl), ETHER_ADDR_LEN);
451 		error = if_addmulti(ifp_p, (struct sockaddr *)&sdl, &rifma);
452 		if (error)
453 			return (error);
454 	}
455 
456 	return (0);
457 }
458 
459 /*
460  * A handler for network interface departure events.
461  * Track departure of trunks here so that we don't access invalid
462  * pointers or whatever if a trunk is ripped from under us, e.g.,
463  * by ejecting its hot-plug card.
464  */
465 static void
466 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
467 {
468 	struct ifvlan *ifv;
469 	int i;
470 
471 	/*
472 	 * Check if it's a trunk interface first of all
473 	 * to avoid needless locking.
474 	 */
475 	if (ifp->if_vlantrunk == NULL)
476 		return;
477 
478 	VLAN_LOCK();
479 	/*
480 	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
481 	 * Check trunk pointer after each vlan_unconfig() as it will
482 	 * free it and set to NULL after the last vlan was detached.
483 	 */
484 #ifdef VLAN_ARRAY
485 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
486 		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
487 			vlan_unconfig_locked(ifv->ifv_ifp);
488 			if (ifp->if_vlantrunk == NULL)
489 				break;
490 		}
491 #else /* VLAN_ARRAY */
492 restart:
493 	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
494 		if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
495 			vlan_unconfig_locked(ifv->ifv_ifp);
496 			if (ifp->if_vlantrunk)
497 				goto restart;	/* trunk->hwidth can change */
498 			else
499 				break;
500 		}
501 #endif /* VLAN_ARRAY */
502 	/* Trunk should have been destroyed in vlan_unconfig(). */
503 	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
504 	VLAN_UNLOCK();
505 }
506 
507 /*
508  * VLAN support can be loaded as a module.  The only place in the
509  * system that's intimately aware of this is ether_input.  We hook
510  * into this code through vlan_input_p which is defined there and
511  * set here.  Noone else in the system should be aware of this so
512  * we use an explicit reference here.
513  */
514 extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
515 
516 /* For if_link_state_change() eyes only... */
517 extern	void (*vlan_link_state_p)(struct ifnet *, int);
518 
519 static int
520 vlan_modevent(module_t mod, int type, void *data)
521 {
522 
523 	switch (type) {
524 	case MOD_LOAD:
525 		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
526 		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
527 		if (ifdetach_tag == NULL)
528 			return (ENOMEM);
529 		LIST_INIT(&trunk_list);
530 		VLAN_LOCK_INIT();
531 		vlan_input_p = vlan_input;
532 		vlan_link_state_p = vlan_link_state;
533 		vlan_trunk_cap_p = vlan_trunk_capabilities;
534 		if_clone_attach(&vlan_cloner);
535 		break;
536 	case MOD_UNLOAD:
537 	    {
538 		struct ifvlantrunk *trunk, *trunk1;
539 
540 		if_clone_detach(&vlan_cloner);
541 		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
542 		vlan_input_p = NULL;
543 		vlan_link_state_p = NULL;
544 		vlan_trunk_cap_p = NULL;
545 		VLAN_LOCK();
546 		LIST_FOREACH_SAFE(trunk, &trunk_list, trunk_entry, trunk1)
547 			trunk_destroy(trunk);
548 		VLAN_UNLOCK();
549 		VLAN_LOCK_DESTROY();
550 		break;
551 	    }
552 	default:
553 		return (EOPNOTSUPP);
554 	}
555 	return (0);
556 }
557 
558 static moduledata_t vlan_mod = {
559 	"if_vlan",
560 	vlan_modevent,
561 	0
562 };
563 
564 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
565 MODULE_VERSION(if_vlan, 3);
566 MODULE_DEPEND(if_vlan, miibus, 1, 1, 1);
567 
568 static struct ifnet *
569 vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
570 {
571 	const char *cp;
572 	struct ifnet *ifp;
573 	int t = 0;
574 
575 	/* Check for <etherif>.<vlan> style interface names. */
576 	IFNET_RLOCK();
577 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
578 		if (ifp->if_type != IFT_ETHER)
579 			continue;
580 		if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
581 			continue;
582 		cp = name + strlen(ifp->if_xname);
583 		if (*cp != '.')
584 			continue;
585 		for(; *cp != '\0'; cp++) {
586 			if (*cp < '0' || *cp > '9')
587 				continue;
588 			t = (t * 10) + (*cp - '0');
589 		}
590 		if (tag != NULL)
591 			*tag = t;
592 		break;
593 	}
594 	IFNET_RUNLOCK();
595 
596 	return (ifp);
597 }
598 
599 static int
600 vlan_clone_match(struct if_clone *ifc, const char *name)
601 {
602 	const char *cp;
603 
604 	if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
605 		return (1);
606 
607 	if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
608 		return (0);
609 	for (cp = name + 4; *cp != '\0'; cp++) {
610 		if (*cp < '0' || *cp > '9')
611 			return (0);
612 	}
613 
614 	return (1);
615 }
616 
617 static int
618 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
619 {
620 	char *dp;
621 	int wildcard;
622 	int unit;
623 	int error;
624 	int tag;
625 	int ethertag;
626 	struct ifvlan *ifv;
627 	struct ifnet *ifp;
628 	struct ifnet *p;
629 	struct vlanreq vlr;
630 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
631 
632 	/*
633 	 * There are 3 (ugh) ways to specify the cloned device:
634 	 * o pass a parameter block with the clone request.
635 	 * o specify parameters in the text of the clone device name
636 	 * o specify no parameters and get an unattached device that
637 	 *   must be configured separately.
638 	 * The first technique is preferred; the latter two are
639 	 * supported for backwards compatibilty.
640 	 */
641 	if (params) {
642 		error = copyin(params, &vlr, sizeof(vlr));
643 		if (error)
644 			return error;
645 		p = ifunit(vlr.vlr_parent);
646 		if (p == NULL)
647 			return ENXIO;
648 		/*
649 		 * Don't let the caller set up a VLAN tag with
650 		 * anything except VLID bits.
651 		 */
652 		if (vlr.vlr_tag & ~EVL_VLID_MASK)
653 			return (EINVAL);
654 		error = ifc_name2unit(name, &unit);
655 		if (error != 0)
656 			return (error);
657 
658 		ethertag = 1;
659 		tag = vlr.vlr_tag;
660 		wildcard = (unit < 0);
661 	} else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
662 		ethertag = 1;
663 		unit = -1;
664 		wildcard = 0;
665 
666 		/*
667 		 * Don't let the caller set up a VLAN tag with
668 		 * anything except VLID bits.
669 		 */
670 		if (tag & ~EVL_VLID_MASK)
671 			return (EINVAL);
672 	} else {
673 		ethertag = 0;
674 
675 		error = ifc_name2unit(name, &unit);
676 		if (error != 0)
677 			return (error);
678 
679 		wildcard = (unit < 0);
680 	}
681 
682 	error = ifc_alloc_unit(ifc, &unit);
683 	if (error != 0)
684 		return (error);
685 
686 	/* In the wildcard case, we need to update the name. */
687 	if (wildcard) {
688 		for (dp = name; *dp != '\0'; dp++);
689 		if (snprintf(dp, len - (dp-name), "%d", unit) >
690 		    len - (dp-name) - 1) {
691 			panic("%s: interface name too long", __func__);
692 		}
693 	}
694 
695 	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
696 	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
697 	if (ifp == NULL) {
698 		ifc_free_unit(ifc, unit);
699 		free(ifv, M_VLAN);
700 		return (ENOSPC);
701 	}
702 	SLIST_INIT(&ifv->vlan_mc_listhead);
703 
704 	ifp->if_softc = ifv;
705 	/*
706 	 * Set the name manually rather than using if_initname because
707 	 * we don't conform to the default naming convention for interfaces.
708 	 */
709 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
710 	ifp->if_dname = ifc->ifc_name;
711 	ifp->if_dunit = unit;
712 	/* NB: flags are not set here */
713 	ifp->if_linkmib = &ifv->ifv_mib;
714 	ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
715 	/* NB: mtu is not set here */
716 
717 	ifp->if_init = vlan_init;
718 	ifp->if_start = vlan_start;
719 	ifp->if_ioctl = vlan_ioctl;
720 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
721 	ifp->if_flags = VLAN_IFFLAGS;
722 	ether_ifattach(ifp, eaddr);
723 	/* Now undo some of the damage... */
724 	ifp->if_baudrate = 0;
725 	ifp->if_type = IFT_L2VLAN;
726 	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
727 
728 	if (ethertag) {
729 		error = vlan_config(ifv, p, tag);
730 		if (error != 0) {
731 			/*
732 			 * Since we've partialy failed, we need to back
733 			 * out all the way, otherwise userland could get
734 			 * confused.  Thus, we destroy the interface.
735 			 */
736 			ether_ifdetach(ifp);
737 			vlan_unconfig(ifp);
738 			if_free_type(ifp, IFT_ETHER);
739 			free(ifv, M_VLAN);
740 
741 			return (error);
742 		}
743 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
744 
745 		/* Update flags on the parent, if necessary. */
746 		vlan_setflags(ifp, 1);
747 	}
748 
749 	return (0);
750 }
751 
752 static int
753 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
754 {
755 	struct ifvlan *ifv = ifp->if_softc;
756 	int unit = ifp->if_dunit;
757 
758 	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
759 	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
760 	if_free_type(ifp, IFT_ETHER);
761 	free(ifv, M_VLAN);
762 	ifc_free_unit(ifc, unit);
763 
764 	return (0);
765 }
766 
767 /*
768  * The ifp->if_init entry point for vlan(4) is a no-op.
769  */
770 static void
771 vlan_init(void *foo __unused)
772 {
773 }
774 
775 /*
776  * The if_start method for vlan(4) interface. It doesn't
777  * raises the IFF_DRV_OACTIVE flag, since it is called
778  * only from IFQ_HANDOFF() macro in ether_output_frame().
779  * If the interface queue is full, and vlan_start() is
780  * not called, the queue would never get emptied and
781  * interface would stall forever.
782  */
783 static void
784 vlan_start(struct ifnet *ifp)
785 {
786 	struct ifvlan *ifv;
787 	struct ifnet *p;
788 	struct mbuf *m;
789 	int error;
790 
791 	ifv = ifp->if_softc;
792 	p = PARENT(ifv);
793 
794 	for (;;) {
795 		IF_DEQUEUE(&ifp->if_snd, m);
796 		if (m == 0)
797 			break;
798 		BPF_MTAP(ifp, m);
799 
800 		/*
801 		 * Do not run parent's if_start() if the parent is not up,
802 		 * or parent's driver will cause a system crash.
803 		 */
804 		if (!((p->if_flags & IFF_UP) &&
805 		    (p->if_drv_flags & IFF_DRV_RUNNING))) {
806 			m_freem(m);
807 			ifp->if_collisions++;
808 			continue;
809 		}
810 
811 		/*
812 		 * If underlying interface can do VLAN tag insertion itself,
813 		 * just pass the packet along. However, we need some way to
814 		 * tell the interface where the packet came from so that it
815 		 * knows how to find the VLAN tag to use, so we attach a
816 		 * packet tag that holds it.
817 		 */
818 		if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
819 			struct m_tag *mtag = (struct m_tag *)
820 			    uma_zalloc(zone_mtag_vlan, M_NOWAIT);
821 			if (mtag == NULL) {
822 				ifp->if_oerrors++;
823 				m_freem(m);
824 				continue;
825 			}
826 			VLAN_TAG_VALUE(mtag) = ifv->ifv_tag;
827 			m_tag_prepend(m, mtag);
828 			m->m_flags |= M_VLANTAG;
829 		} else {
830 			struct ether_vlan_header *evl;
831 
832 			M_PREPEND(m, ifv->ifv_encaplen, M_DONTWAIT);
833 			if (m == NULL) {
834 				if_printf(ifp,
835 				    "unable to prepend VLAN header\n");
836 				ifp->if_oerrors++;
837 				continue;
838 			}
839 			/* M_PREPEND takes care of m_len, m_pkthdr.len for us */
840 
841 			if (m->m_len < sizeof(*evl)) {
842 				m = m_pullup(m, sizeof(*evl));
843 				if (m == NULL) {
844 					if_printf(ifp,
845 					    "cannot pullup VLAN header\n");
846 					ifp->if_oerrors++;
847 					continue;
848 				}
849 			}
850 
851 			/*
852 			 * Transform the Ethernet header into an Ethernet header
853 			 * with 802.1Q encapsulation.
854 			 */
855 			bcopy(mtod(m, char *) + ifv->ifv_encaplen,
856 			      mtod(m, char *), ETHER_HDR_LEN);
857 			evl = mtod(m, struct ether_vlan_header *);
858 			evl->evl_proto = evl->evl_encap_proto;
859 			evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
860 			evl->evl_tag = htons(ifv->ifv_tag);
861 #ifdef DEBUG
862 			printf("%s: %*D\n", __func__, (int)sizeof(*evl),
863 			    (unsigned char *)evl, ":");
864 #endif
865 		}
866 
867 		/*
868 		 * Send it, precisely as ether_output() would have.
869 		 * We are already running at splimp.
870 		 */
871 		IFQ_HANDOFF(p, m, error);
872 		if (!error)
873 			ifp->if_opackets++;
874 		else
875 			ifp->if_oerrors++;
876 	}
877 }
878 
879 static void
880 vlan_input(struct ifnet *ifp, struct mbuf *m)
881 {
882 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
883 	struct ifvlan *ifv;
884 	struct m_tag *mtag;
885 	uint16_t tag;
886 
887 	KASSERT(trunk != NULL, ("%s: no trunk", __func__));
888 
889 	if (m->m_flags & M_VLANTAG) {
890 		/*
891 		 * Packet is tagged, but m contains a normal
892 		 * Ethernet frame; the tag is stored out-of-band.
893 		 */
894 		mtag = m_tag_locate(m, MTAG_VLAN, MTAG_VLAN_TAG, NULL);
895 		KASSERT(mtag != NULL,
896 			("%s: M_VLANTAG without m_tag", __func__));
897 		tag = EVL_VLANOFTAG(VLAN_TAG_VALUE(mtag));
898 		m_tag_delete(m, mtag);
899 		m->m_flags &= ~M_VLANTAG;
900 	} else {
901 		struct ether_vlan_header *evl;
902 
903 		/*
904 		 * Packet is tagged in-band as specified by 802.1q.
905 		 */
906 		mtag = NULL;
907 		switch (ifp->if_type) {
908 		case IFT_ETHER:
909 			if (m->m_len < sizeof(*evl) &&
910 			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
911 				if_printf(ifp, "cannot pullup VLAN header\n");
912 				return;
913 			}
914 			evl = mtod(m, struct ether_vlan_header *);
915 			KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN,
916 				("%s: bad encapsulation protocol (%u)",
917 				 __func__, ntohs(evl->evl_encap_proto)));
918 
919 			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
920 
921 			/*
922 			 * Restore the original ethertype.  We'll remove
923 			 * the encapsulation after we've found the vlan
924 			 * interface corresponding to the tag.
925 			 */
926 			evl->evl_encap_proto = evl->evl_proto;
927 			break;
928 		default:
929 #ifdef INVARIANTS
930 			panic("%s: %s has unsupported if_type %u",
931 			      __func__, ifp->if_xname, ifp->if_type);
932 #endif
933 			m_freem(m);
934 			ifp->if_noproto++;
935 			return;
936 		}
937 	}
938 
939 	TRUNK_RLOCK(trunk);
940 #ifdef VLAN_ARRAY
941 	ifv = trunk->vlans[tag];
942 #else
943 	ifv = vlan_gethash(trunk, tag);
944 #endif
945 	if (ifv == NULL || (ifv->ifv_ifp->if_flags & IFF_UP) == 0) {
946 		TRUNK_RUNLOCK(trunk);
947 		m_freem(m);
948 		ifp->if_noproto++;
949 		return;
950 	}
951 	TRUNK_RUNLOCK(trunk);
952 
953 	if (mtag == NULL) {
954 		/*
955 		 * Packet had an in-line encapsulation header;
956 		 * remove it.  The original header has already
957 		 * been fixed up above.
958 		 */
959 		bcopy(mtod(m, caddr_t),
960 		      mtod(m, caddr_t) + ETHER_VLAN_ENCAP_LEN,
961 		      ETHER_HDR_LEN);
962 		m_adj(m, ETHER_VLAN_ENCAP_LEN);
963 	}
964 
965 	m->m_pkthdr.rcvif = ifv->ifv_ifp;
966 	ifv->ifv_ifp->if_ipackets++;
967 
968 	/* Pass it back through the parent's input routine. */
969 	(*ifp->if_input)(ifv->ifv_ifp, m);
970 }
971 
972 static int
973 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
974 {
975 	struct ifvlantrunk *trunk;
976 	struct ifnet *ifp;
977 	int error = 0;
978 
979 	/* VID numbers 0x0 and 0xFFF are reserved */
980 	if (tag == 0 || tag == 0xFFF)
981 		return (EINVAL);
982 	if (p->if_type != IFT_ETHER)
983 		return (EPROTONOSUPPORT);
984 	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
985 		return (EPROTONOSUPPORT);
986 	if (ifv->ifv_trunk)
987 		return (EBUSY);
988 
989 	if (p->if_vlantrunk == NULL) {
990 		trunk = malloc(sizeof(struct ifvlantrunk),
991 		    M_VLAN, M_WAITOK | M_ZERO);
992 #ifndef VLAN_ARRAY
993 		vlan_inithash(trunk);
994 #endif
995 		VLAN_LOCK();
996 		if (p->if_vlantrunk != NULL) {
997 			/* A race that that is very unlikely to be hit. */
998 #ifndef VLAN_ARRAY
999 			vlan_freehash(trunk);
1000 #endif
1001 			free(trunk, M_VLAN);
1002 			goto exists;
1003 		}
1004 		TRUNK_LOCK_INIT(trunk);
1005 		LIST_INSERT_HEAD(&trunk_list, trunk, trunk_entry);
1006 		TRUNK_LOCK(trunk);
1007 		p->if_vlantrunk = trunk;
1008 		trunk->parent = p;
1009 	} else {
1010 		VLAN_LOCK();
1011 exists:
1012 		trunk = p->if_vlantrunk;
1013 		TRUNK_LOCK(trunk);
1014 	}
1015 
1016 	ifv->ifv_tag = tag;	/* must set this before vlan_inshash() */
1017 #ifdef VLAN_ARRAY
1018 	if (trunk->vlans[tag] != NULL) {
1019 		error = EEXIST;
1020 		goto done;
1021 	}
1022 	trunk->vlans[tag] = ifv;
1023 	trunk->refcnt++;
1024 #else
1025 	error = vlan_inshash(trunk, ifv);
1026 	if (error)
1027 		goto done;
1028 #endif
1029 	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1030 	ifv->ifv_mintu = ETHERMIN;
1031 	ifv->ifv_pflags = 0;
1032 
1033 	/*
1034 	 * If the parent supports the VLAN_MTU capability,
1035 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1036 	 * use it.
1037 	 */
1038 	if (p->if_capenable & IFCAP_VLAN_MTU) {
1039 		/*
1040 		 * No need to fudge the MTU since the parent can
1041 		 * handle extended frames.
1042 		 */
1043 		ifv->ifv_mtufudge = 0;
1044 	} else {
1045 		/*
1046 		 * Fudge the MTU by the encapsulation size.  This
1047 		 * makes us incompatible with strictly compliant
1048 		 * 802.1Q implementations, but allows us to use
1049 		 * the feature with other NetBSD implementations,
1050 		 * which might still be useful.
1051 		 */
1052 		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1053 	}
1054 
1055 	ifv->ifv_trunk = trunk;
1056 	ifp = ifv->ifv_ifp;
1057 	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1058 	ifp->if_baudrate = p->if_baudrate;
1059 	/*
1060 	 * Copy only a selected subset of flags from the parent.
1061 	 * Other flags are none of our business.
1062 	 */
1063 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1064 	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1065 	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1066 #undef VLAN_COPY_FLAGS
1067 
1068 	ifp->if_link_state = p->if_link_state;
1069 
1070 	vlan_capabilities(ifv);
1071 
1072 	/*
1073 	 * Set up our ``Ethernet address'' to reflect the underlying
1074 	 * physical interface's.
1075 	 */
1076 	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), ETHER_ADDR_LEN);
1077 
1078 	/*
1079 	 * Configure multicast addresses that may already be
1080 	 * joined on the vlan device.
1081 	 */
1082 	(void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1083 done:
1084 	TRUNK_UNLOCK(trunk);
1085 	VLAN_UNLOCK();
1086 
1087 	return (error);
1088 }
1089 
1090 static int
1091 vlan_unconfig(struct ifnet *ifp)
1092 {
1093 	int ret;
1094 
1095 	VLAN_LOCK();
1096 	ret = vlan_unconfig_locked(ifp);
1097 	VLAN_UNLOCK();
1098 	return (ret);
1099 }
1100 
1101 static int
1102 vlan_unconfig_locked(struct ifnet *ifp)
1103 {
1104 	struct ifvlantrunk *trunk;
1105 	struct vlan_mc_entry *mc;
1106 	struct ifvlan *ifv;
1107 	int error;
1108 
1109 	VLAN_LOCK_ASSERT();
1110 
1111 	ifv = ifp->if_softc;
1112 	trunk = ifv->ifv_trunk;
1113 
1114 	if (trunk) {
1115 		struct sockaddr_dl sdl;
1116 		struct ifnet *p = trunk->parent;
1117 
1118 		TRUNK_LOCK(trunk);
1119 
1120 		/*
1121 		 * Since the interface is being unconfigured, we need to
1122 		 * empty the list of multicast groups that we may have joined
1123 		 * while we were alive from the parent's list.
1124 		 */
1125 		bzero((char *)&sdl, sizeof(sdl));
1126 		sdl.sdl_len = sizeof(sdl);
1127 		sdl.sdl_family = AF_LINK;
1128 		sdl.sdl_index = p->if_index;
1129 		sdl.sdl_type = IFT_ETHER;
1130 		sdl.sdl_alen = ETHER_ADDR_LEN;
1131 
1132 		while(SLIST_FIRST(&ifv->vlan_mc_listhead) != NULL) {
1133 			mc = SLIST_FIRST(&ifv->vlan_mc_listhead);
1134 			bcopy((char *)&mc->mc_addr, LLADDR(&sdl),
1135 			    ETHER_ADDR_LEN);
1136 			error = if_delmulti(p, (struct sockaddr *)&sdl);
1137 			if (error)
1138 				return (error);
1139 			SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1140 			free(mc, M_VLAN);
1141 		}
1142 
1143 		vlan_setflags(ifp, 0); /* clear special flags on parent */
1144 #ifdef VLAN_ARRAY
1145 		trunk->vlans[ifv->ifv_tag] = NULL;
1146 		trunk->refcnt--;
1147 #else
1148 		vlan_remhash(trunk, ifv);
1149 #endif
1150 		ifv->ifv_trunk = NULL;
1151 
1152 		/*
1153 		 * Check if we were the last.
1154 		 */
1155 		if (trunk->refcnt == 0) {
1156 			trunk->parent->if_vlantrunk = NULL;
1157 			/*
1158 			 * XXXGL: If some ithread has already entered
1159 			 * vlan_input() and is now blocked on the trunk
1160 			 * lock, then it should preempt us right after
1161 			 * unlock and finish its work. Then we will acquire
1162 			 * lock again in trunk_destroy().
1163 			 * XXX: not true in case of VLAN_ARRAY
1164 			 */
1165 			TRUNK_UNLOCK(trunk);
1166 			trunk_destroy(trunk);
1167 		} else
1168 			TRUNK_UNLOCK(trunk);
1169 	}
1170 
1171 	/* Disconnect from parent. */
1172 	if (ifv->ifv_pflags)
1173 		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1174 	ifp->if_mtu = ETHERMTU;
1175 	ifp->if_link_state = LINK_STATE_UNKNOWN;
1176 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1177 
1178 	return (0);
1179 }
1180 
1181 /* Handle a reference counted flag that should be set on the parent as well */
1182 static int
1183 vlan_setflag(struct ifnet *ifp, int flag, int status,
1184 	     int (*func)(struct ifnet *, int))
1185 {
1186 	struct ifvlan *ifv;
1187 	int error;
1188 
1189 	/* XXX VLAN_LOCK_ASSERT(); */
1190 
1191 	ifv = ifp->if_softc;
1192 	status = status ? (ifp->if_flags & flag) : 0;
1193 	/* Now "status" contains the flag value or 0 */
1194 
1195 	/*
1196 	 * See if recorded parent's status is different from what
1197 	 * we want it to be.  If it is, flip it.  We record parent's
1198 	 * status in ifv_pflags so that we won't clear parent's flag
1199 	 * we haven't set.  In fact, we don't clear or set parent's
1200 	 * flags directly, but get or release references to them.
1201 	 * That's why we can be sure that recorded flags still are
1202 	 * in accord with actual parent's flags.
1203 	 */
1204 	if (status != (ifv->ifv_pflags & flag)) {
1205 		error = (*func)(PARENT(ifv), status);
1206 		if (error)
1207 			return (error);
1208 		ifv->ifv_pflags &= ~flag;
1209 		ifv->ifv_pflags |= status;
1210 	}
1211 	return (0);
1212 }
1213 
1214 /*
1215  * Handle IFF_* flags that require certain changes on the parent:
1216  * if "status" is true, update parent's flags respective to our if_flags;
1217  * if "status" is false, forcedly clear the flags set on parent.
1218  */
1219 static int
1220 vlan_setflags(struct ifnet *ifp, int status)
1221 {
1222 	int error, i;
1223 
1224 	for (i = 0; vlan_pflags[i].flag; i++) {
1225 		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1226 				     status, vlan_pflags[i].func);
1227 		if (error)
1228 			return (error);
1229 	}
1230 	return (0);
1231 }
1232 
1233 /* Inform all vlans that their parent has changed link state */
1234 static void
1235 vlan_link_state(struct ifnet *ifp, int link)
1236 {
1237 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1238 	struct ifvlan *ifv;
1239 	int i;
1240 
1241 	TRUNK_LOCK(trunk);
1242 #ifdef VLAN_ARRAY
1243 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1244 		if (trunk->vlans[i] != NULL) {
1245 			ifv = trunk->vlans[i];
1246 #else
1247 	for (i = 0; i < (1 << trunk->hwidth); i++) {
1248 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1249 #endif
1250 			if_link_state_change(ifv->ifv_ifp,
1251 			    trunk->parent->if_link_state);
1252 	}
1253 	TRUNK_UNLOCK(trunk);
1254 }
1255 
1256 static void
1257 vlan_capabilities(struct ifvlan *ifv)
1258 {
1259 	struct ifnet *p = PARENT(ifv);
1260 	struct ifnet *ifp = ifv->ifv_ifp;
1261 
1262 	TRUNK_LOCK_ASSERT(TRUNK(ifv));
1263 
1264 	/*
1265 	 * If the parent interface can do checksum offloading
1266 	 * on VLANs, then propagate its hardware-assisted
1267 	 * checksumming flags. Also assert that checksum
1268 	 * offloading requires hardware VLAN tagging.
1269 	 */
1270 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1271 		ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1272 
1273 	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1274 	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1275 		ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1276 		ifp->if_hwassist = p->if_hwassist;
1277 	} else {
1278 		ifp->if_capenable = 0;
1279 		ifp->if_hwassist = 0;
1280 	}
1281 }
1282 
1283 static void
1284 vlan_trunk_capabilities(struct ifnet *ifp)
1285 {
1286 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1287 	struct ifvlan *ifv;
1288 	int i;
1289 
1290 	TRUNK_LOCK(trunk);
1291 #ifdef VLAN_ARRAY
1292 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1293 		if (trunk->vlans[i] != NULL) {
1294 			ifv = trunk->vlans[i];
1295 #else
1296 	for (i = 0; i < (1 << trunk->hwidth); i++) {
1297 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1298 #endif
1299 			vlan_capabilities(ifv);
1300 	}
1301 	TRUNK_UNLOCK(trunk);
1302 }
1303 
1304 static int
1305 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1306 {
1307 	struct ifaddr *ifa;
1308 	struct ifnet *p;
1309 	struct ifreq *ifr;
1310 	struct ifvlan *ifv;
1311 	struct vlanreq vlr;
1312 	int error = 0;
1313 
1314 	ifr = (struct ifreq *)data;
1315 	ifa = (struct ifaddr *)data;
1316 	ifv = ifp->if_softc;
1317 
1318 	switch (cmd) {
1319 	case SIOCSIFADDR:
1320 		ifp->if_flags |= IFF_UP;
1321 
1322 		switch (ifa->ifa_addr->sa_family) {
1323 #ifdef INET
1324 		case AF_INET:
1325 			arp_ifinit(ifv->ifv_ifp, ifa);
1326 			break;
1327 #endif
1328 		default:
1329 			break;
1330 		}
1331 		break;
1332 
1333 	case SIOCGIFADDR:
1334 		{
1335 			struct sockaddr *sa;
1336 
1337 			sa = (struct sockaddr *) &ifr->ifr_data;
1338 			bcopy(IF_LLADDR(ifp), (caddr_t)sa->sa_data,
1339 			    ETHER_ADDR_LEN);
1340 		}
1341 		break;
1342 
1343 	case SIOCGIFMEDIA:
1344 		VLAN_LOCK();
1345 		if (TRUNK(ifv) != NULL) {
1346 			error = (*PARENT(ifv)->if_ioctl)(PARENT(ifv),
1347 					SIOCGIFMEDIA, data);
1348 			VLAN_UNLOCK();
1349 			/* Limit the result to the parent's current config. */
1350 			if (error == 0) {
1351 				struct ifmediareq *ifmr;
1352 
1353 				ifmr = (struct ifmediareq *)data;
1354 				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1355 					ifmr->ifm_count = 1;
1356 					error = copyout(&ifmr->ifm_current,
1357 						ifmr->ifm_ulist,
1358 						sizeof(int));
1359 				}
1360 			}
1361 		} else {
1362 			VLAN_UNLOCK();
1363 			error = EINVAL;
1364 		}
1365 		break;
1366 
1367 	case SIOCSIFMEDIA:
1368 		error = EINVAL;
1369 		break;
1370 
1371 	case SIOCSIFMTU:
1372 		/*
1373 		 * Set the interface MTU.
1374 		 */
1375 		VLAN_LOCK();
1376 		if (TRUNK(ifv) != NULL) {
1377 			if (ifr->ifr_mtu >
1378 			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1379 			    ifr->ifr_mtu <
1380 			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1381 				error = EINVAL;
1382 			else
1383 				ifp->if_mtu = ifr->ifr_mtu;
1384 		} else
1385 			error = EINVAL;
1386 		VLAN_UNLOCK();
1387 		break;
1388 
1389 	case SIOCSETVLAN:
1390 		error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1391 		if (error)
1392 			break;
1393 		if (vlr.vlr_parent[0] == '\0') {
1394 			vlan_unconfig(ifp);
1395 			break;
1396 		}
1397 		p = ifunit(vlr.vlr_parent);
1398 		if (p == 0) {
1399 			error = ENOENT;
1400 			break;
1401 		}
1402 		/*
1403 		 * Don't let the caller set up a VLAN tag with
1404 		 * anything except VLID bits.
1405 		 */
1406 		if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1407 			error = EINVAL;
1408 			break;
1409 		}
1410 		error = vlan_config(ifv, p, vlr.vlr_tag);
1411 		if (error)
1412 			break;
1413 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
1414 
1415 		/* Update flags on the parent, if necessary. */
1416 		vlan_setflags(ifp, 1);
1417 		break;
1418 
1419 	case SIOCGETVLAN:
1420 		bzero(&vlr, sizeof(vlr));
1421 		VLAN_LOCK();
1422 		if (TRUNK(ifv) != NULL) {
1423 			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1424 			    sizeof(vlr.vlr_parent));
1425 			vlr.vlr_tag = ifv->ifv_tag;
1426 		}
1427 		VLAN_UNLOCK();
1428 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1429 		break;
1430 
1431 	case SIOCSIFFLAGS:
1432 		/*
1433 		 * We should propagate selected flags to the parent,
1434 		 * e.g., promiscuous mode.
1435 		 */
1436 		if (TRUNK(ifv) != NULL)
1437 			error = vlan_setflags(ifp, 1);
1438 		break;
1439 
1440 	case SIOCADDMULTI:
1441 	case SIOCDELMULTI:
1442 		/*
1443 		 * If we don't have a parent, just remember the membership for
1444 		 * when we do.
1445 		 */
1446 		if (TRUNK(ifv) != NULL)
1447 			error = vlan_setmulti(ifp);
1448 		break;
1449 
1450 	default:
1451 		error = EINVAL;
1452 	}
1453 
1454 	return (error);
1455 }
1456