xref: /freebsd/sys/net/if_vlan.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
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 
30 /*
31  * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
32  * Might be extended some day to also handle IEEE 802.1p priority
33  * tagging.  This is sort of sneaky in the implementation, since
34  * we need to pretend to be enough of an Ethernet implementation
35  * to make arp work.  The way we do this is by telling everyone
36  * that we are an Ethernet, and then catch the packets that
37  * ether_output() left on our output queue when it calls
38  * if_start(), rewrite them for use by the real outgoing interface,
39  * and ask it to send them.
40  */
41 
42 #include <sys/cdefs.h>
43 __FBSDID("$FreeBSD$");
44 
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 #include <sys/sx.h>
60 
61 #include <net/bpf.h>
62 #include <net/ethernet.h>
63 #include <net/if.h>
64 #include <net/if_clone.h>
65 #include <net/if_dl.h>
66 #include <net/if_types.h>
67 #include <net/if_vlan_var.h>
68 #include <net/vnet.h>
69 
70 #define VLANNAME	"vlan"
71 #define	VLAN_DEF_HWIDTH	4
72 #define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
73 
74 #define	UP_AND_RUNNING(ifp) \
75     ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
76 
77 LIST_HEAD(ifvlanhead, ifvlan);
78 
79 struct ifvlantrunk {
80 	struct	ifnet   *parent;	/* parent interface of this trunk */
81 	struct	rwlock	rw;
82 #ifdef VLAN_ARRAY
83 #define	VLAN_ARRAY_SIZE	(EVL_VLID_MASK + 1)
84 	struct	ifvlan	*vlans[VLAN_ARRAY_SIZE]; /* static table */
85 #else
86 	struct	ifvlanhead *hash;	/* dynamic hash-list table */
87 	uint16_t	hmask;
88 	uint16_t	hwidth;
89 #endif
90 	int		refcnt;
91 };
92 
93 struct vlan_mc_entry {
94 	struct sockaddr_dl		mc_addr;
95 	SLIST_ENTRY(vlan_mc_entry)	mc_entries;
96 };
97 
98 struct	ifvlan {
99 	struct	ifvlantrunk *ifv_trunk;
100 	struct	ifnet *ifv_ifp;
101 	void	*ifv_cookie;
102 #define	TRUNK(ifv)	((ifv)->ifv_trunk)
103 #define	PARENT(ifv)	((ifv)->ifv_trunk->parent)
104 	int	ifv_pflags;	/* special flags we have set on parent */
105 	struct	ifv_linkmib {
106 		int	ifvm_encaplen;	/* encapsulation length */
107 		int	ifvm_mtufudge;	/* MTU fudged by this much */
108 		int	ifvm_mintu;	/* min transmission unit */
109 		uint16_t ifvm_proto;	/* encapsulation ethertype */
110 		uint16_t ifvm_tag;	/* tag to apply on packets leaving if */
111 	}	ifv_mib;
112 	SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
113 #ifndef VLAN_ARRAY
114 	LIST_ENTRY(ifvlan) ifv_list;
115 #endif
116 };
117 #define	ifv_proto	ifv_mib.ifvm_proto
118 #define	ifv_tag		ifv_mib.ifvm_tag
119 #define	ifv_encaplen	ifv_mib.ifvm_encaplen
120 #define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
121 #define	ifv_mintu	ifv_mib.ifvm_mintu
122 
123 /* Special flags we should propagate to parent. */
124 static struct {
125 	int flag;
126 	int (*func)(struct ifnet *, int);
127 } vlan_pflags[] = {
128 	{IFF_PROMISC, ifpromisc},
129 	{IFF_ALLMULTI, if_allmulti},
130 	{0, NULL}
131 };
132 
133 SYSCTL_DECL(_net_link);
134 SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0, "IEEE 802.1Q VLAN");
135 SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0, "for consistency");
136 
137 static int soft_pad = 0;
138 SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW, &soft_pad, 0,
139 	   "pad short frames before tagging");
140 
141 static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface");
142 
143 static eventhandler_tag ifdetach_tag;
144 static eventhandler_tag iflladdr_tag;
145 
146 /*
147  * We have a global mutex, that is used to serialize configuration
148  * changes and isn't used in normal packet delivery.
149  *
150  * We also have a per-trunk rwlock, that is locked shared on packet
151  * processing and exclusive when configuration is changed.
152  *
153  * The VLAN_ARRAY substitutes the dynamic hash with a static array
154  * with 4096 entries. In theory this can give a boost in processing,
155  * however on practice it does not. Probably this is because array
156  * is too big to fit into CPU cache.
157  */
158 static struct sx ifv_lock;
159 #define	VLAN_LOCK_INIT()	sx_init(&ifv_lock, "vlan_global")
160 #define	VLAN_LOCK_DESTROY()	sx_destroy(&ifv_lock)
161 #define	VLAN_LOCK_ASSERT()	sx_assert(&ifv_lock, SA_LOCKED)
162 #define	VLAN_LOCK()		sx_xlock(&ifv_lock)
163 #define	VLAN_UNLOCK()		sx_xunlock(&ifv_lock)
164 #define	TRUNK_LOCK_INIT(trunk)	rw_init(&(trunk)->rw, VLANNAME)
165 #define	TRUNK_LOCK_DESTROY(trunk) rw_destroy(&(trunk)->rw)
166 #define	TRUNK_LOCK(trunk)	rw_wlock(&(trunk)->rw)
167 #define	TRUNK_UNLOCK(trunk)	rw_wunlock(&(trunk)->rw)
168 #define	TRUNK_LOCK_ASSERT(trunk) rw_assert(&(trunk)->rw, RA_WLOCKED)
169 #define	TRUNK_RLOCK(trunk)	rw_rlock(&(trunk)->rw)
170 #define	TRUNK_RUNLOCK(trunk)	rw_runlock(&(trunk)->rw)
171 #define	TRUNK_LOCK_RASSERT(trunk) rw_assert(&(trunk)->rw, RA_RLOCKED)
172 
173 #ifndef VLAN_ARRAY
174 static	void vlan_inithash(struct ifvlantrunk *trunk);
175 static	void vlan_freehash(struct ifvlantrunk *trunk);
176 static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
177 static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
178 static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
179 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
180 	uint16_t tag);
181 #endif
182 static	void trunk_destroy(struct ifvlantrunk *trunk);
183 
184 static	void vlan_start(struct ifnet *ifp);
185 static	void vlan_init(void *foo);
186 static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
187 static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
188 static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
189     int (*func)(struct ifnet *, int));
190 static	int vlan_setflags(struct ifnet *ifp, int status);
191 static	int vlan_setmulti(struct ifnet *ifp);
192 static	void vlan_unconfig(struct ifnet *ifp);
193 static	void vlan_unconfig_locked(struct ifnet *ifp);
194 static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
195 static	void vlan_link_state(struct ifnet *ifp);
196 static	void vlan_capabilities(struct ifvlan *ifv);
197 static	void vlan_trunk_capabilities(struct ifnet *ifp);
198 
199 static	struct ifnet *vlan_clone_match_ethertag(struct if_clone *,
200     const char *, int *);
201 static	int vlan_clone_match(struct if_clone *, const char *);
202 static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
203 static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
204 
205 static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
206 static  void vlan_iflladdr(void *arg, struct ifnet *ifp);
207 
208 static	struct if_clone vlan_cloner = IFC_CLONE_INITIALIZER(VLANNAME, NULL,
209     IF_MAXUNIT, NULL, vlan_clone_match, vlan_clone_create, vlan_clone_destroy);
210 
211 #ifdef VIMAGE
212 static VNET_DEFINE(struct if_clone, vlan_cloner);
213 #define	V_vlan_cloner	VNET(vlan_cloner)
214 #endif
215 
216 #ifndef VLAN_ARRAY
217 #define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
218 
219 static void
220 vlan_inithash(struct ifvlantrunk *trunk)
221 {
222 	int i, n;
223 
224 	/*
225 	 * The trunk must not be locked here since we call malloc(M_WAITOK).
226 	 * It is OK in case this function is called before the trunk struct
227 	 * gets hooked up and becomes visible from other threads.
228 	 */
229 
230 	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
231 	    ("%s: hash already initialized", __func__));
232 
233 	trunk->hwidth = VLAN_DEF_HWIDTH;
234 	n = 1 << trunk->hwidth;
235 	trunk->hmask = n - 1;
236 	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
237 	for (i = 0; i < n; i++)
238 		LIST_INIT(&trunk->hash[i]);
239 }
240 
241 static void
242 vlan_freehash(struct ifvlantrunk *trunk)
243 {
244 #ifdef INVARIANTS
245 	int i;
246 
247 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
248 	for (i = 0; i < (1 << trunk->hwidth); i++)
249 		KASSERT(LIST_EMPTY(&trunk->hash[i]),
250 		    ("%s: hash table not empty", __func__));
251 #endif
252 	free(trunk->hash, M_VLAN);
253 	trunk->hash = NULL;
254 	trunk->hwidth = trunk->hmask = 0;
255 }
256 
257 static int
258 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
259 {
260 	int i, b;
261 	struct ifvlan *ifv2;
262 
263 	TRUNK_LOCK_ASSERT(trunk);
264 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
265 
266 	b = 1 << trunk->hwidth;
267 	i = HASH(ifv->ifv_tag, trunk->hmask);
268 	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
269 		if (ifv->ifv_tag == ifv2->ifv_tag)
270 			return (EEXIST);
271 
272 	/*
273 	 * Grow the hash when the number of vlans exceeds half of the number of
274 	 * hash buckets squared. This will make the average linked-list length
275 	 * buckets/2.
276 	 */
277 	if (trunk->refcnt > (b * b) / 2) {
278 		vlan_growhash(trunk, 1);
279 		i = HASH(ifv->ifv_tag, trunk->hmask);
280 	}
281 	LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
282 	trunk->refcnt++;
283 
284 	return (0);
285 }
286 
287 static int
288 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
289 {
290 	int i, b;
291 	struct ifvlan *ifv2;
292 
293 	TRUNK_LOCK_ASSERT(trunk);
294 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
295 
296 	b = 1 << trunk->hwidth;
297 	i = HASH(ifv->ifv_tag, trunk->hmask);
298 	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
299 		if (ifv2 == ifv) {
300 			trunk->refcnt--;
301 			LIST_REMOVE(ifv2, ifv_list);
302 			if (trunk->refcnt < (b * b) / 2)
303 				vlan_growhash(trunk, -1);
304 			return (0);
305 		}
306 
307 	panic("%s: vlan not found\n", __func__);
308 	return (ENOENT); /*NOTREACHED*/
309 }
310 
311 /*
312  * Grow the hash larger or smaller if memory permits.
313  */
314 static void
315 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
316 {
317 	struct ifvlan *ifv;
318 	struct ifvlanhead *hash2;
319 	int hwidth2, i, j, n, n2;
320 
321 	TRUNK_LOCK_ASSERT(trunk);
322 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
323 
324 	if (howmuch == 0) {
325 		/* Harmless yet obvious coding error */
326 		printf("%s: howmuch is 0\n", __func__);
327 		return;
328 	}
329 
330 	hwidth2 = trunk->hwidth + howmuch;
331 	n = 1 << trunk->hwidth;
332 	n2 = 1 << hwidth2;
333 	/* Do not shrink the table below the default */
334 	if (hwidth2 < VLAN_DEF_HWIDTH)
335 		return;
336 
337 	/* M_NOWAIT because we're called with trunk mutex held */
338 	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
339 	if (hash2 == NULL) {
340 		printf("%s: out of memory -- hash size not changed\n",
341 		    __func__);
342 		return;		/* We can live with the old hash table */
343 	}
344 	for (j = 0; j < n2; j++)
345 		LIST_INIT(&hash2[j]);
346 	for (i = 0; i < n; i++)
347 		while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
348 			LIST_REMOVE(ifv, ifv_list);
349 			j = HASH(ifv->ifv_tag, n2 - 1);
350 			LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
351 		}
352 	free(trunk->hash, M_VLAN);
353 	trunk->hash = hash2;
354 	trunk->hwidth = hwidth2;
355 	trunk->hmask = n2 - 1;
356 
357 	if (bootverbose)
358 		if_printf(trunk->parent,
359 		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
360 }
361 
362 static __inline struct ifvlan *
363 vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
364 {
365 	struct ifvlan *ifv;
366 
367 	TRUNK_LOCK_RASSERT(trunk);
368 
369 	LIST_FOREACH(ifv, &trunk->hash[HASH(tag, trunk->hmask)], ifv_list)
370 		if (ifv->ifv_tag == tag)
371 			return (ifv);
372 	return (NULL);
373 }
374 
375 #if 0
376 /* Debugging code to view the hashtables. */
377 static void
378 vlan_dumphash(struct ifvlantrunk *trunk)
379 {
380 	int i;
381 	struct ifvlan *ifv;
382 
383 	for (i = 0; i < (1 << trunk->hwidth); i++) {
384 		printf("%d: ", i);
385 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
386 			printf("%s ", ifv->ifv_ifp->if_xname);
387 		printf("\n");
388 	}
389 }
390 #endif /* 0 */
391 #else
392 
393 static __inline struct ifvlan *
394 vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
395 {
396 
397 	return trunk->vlans[tag];
398 }
399 
400 static __inline int
401 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
402 {
403 
404 	if (trunk->vlans[ifv->ifv_tag] != NULL)
405 		return EEXIST;
406 	trunk->vlans[ifv->ifv_tag] = ifv;
407 	trunk->refcnt++;
408 
409 	return (0);
410 }
411 
412 static __inline int
413 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
414 {
415 
416 	trunk->vlans[ifv->ifv_tag] = NULL;
417 	trunk->refcnt--;
418 
419 	return (0);
420 }
421 
422 static __inline void
423 vlan_freehash(struct ifvlantrunk *trunk)
424 {
425 }
426 
427 static __inline void
428 vlan_inithash(struct ifvlantrunk *trunk)
429 {
430 }
431 
432 #endif /* !VLAN_ARRAY */
433 
434 static void
435 trunk_destroy(struct ifvlantrunk *trunk)
436 {
437 	VLAN_LOCK_ASSERT();
438 
439 	TRUNK_LOCK(trunk);
440 	vlan_freehash(trunk);
441 	trunk->parent->if_vlantrunk = NULL;
442 	TRUNK_UNLOCK(trunk);
443 	TRUNK_LOCK_DESTROY(trunk);
444 	free(trunk, M_VLAN);
445 }
446 
447 /*
448  * Program our multicast filter. What we're actually doing is
449  * programming the multicast filter of the parent. This has the
450  * side effect of causing the parent interface to receive multicast
451  * traffic that it doesn't really want, which ends up being discarded
452  * later by the upper protocol layers. Unfortunately, there's no way
453  * to avoid this: there really is only one physical interface.
454  *
455  * XXX: There is a possible race here if more than one thread is
456  *      modifying the multicast state of the vlan interface at the same time.
457  */
458 static int
459 vlan_setmulti(struct ifnet *ifp)
460 {
461 	struct ifnet		*ifp_p;
462 	struct ifmultiaddr	*ifma, *rifma = NULL;
463 	struct ifvlan		*sc;
464 	struct vlan_mc_entry	*mc;
465 	int			error;
466 
467 	/*VLAN_LOCK_ASSERT();*/
468 
469 	/* Find the parent. */
470 	sc = ifp->if_softc;
471 	ifp_p = PARENT(sc);
472 
473 	CURVNET_SET_QUIET(ifp_p->if_vnet);
474 
475 	/* First, remove any existing filter entries. */
476 	while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
477 		error = if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
478 		if (error)
479 			return (error);
480 		SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
481 		free(mc, M_VLAN);
482 	}
483 
484 	/* Now program new ones. */
485 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
486 		if (ifma->ifma_addr->sa_family != AF_LINK)
487 			continue;
488 		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
489 		if (mc == NULL)
490 			return (ENOMEM);
491 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
492 		mc->mc_addr.sdl_index = ifp_p->if_index;
493 		SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
494 		error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
495 		    &rifma);
496 		if (error)
497 			return (error);
498 	}
499 
500 	CURVNET_RESTORE();
501 	return (0);
502 }
503 
504 /*
505  * A handler for parent interface link layer address changes.
506  * If the parent interface link layer address is changed we
507  * should also change it on all children vlans.
508  */
509 static void
510 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
511 {
512 	struct ifvlan *ifv;
513 #ifndef VLAN_ARRAY
514 	struct ifvlan *next;
515 #endif
516 	int i;
517 
518 	/*
519 	 * Check if it's a trunk interface first of all
520 	 * to avoid needless locking.
521 	 */
522 	if (ifp->if_vlantrunk == NULL)
523 		return;
524 
525 	VLAN_LOCK();
526 	/*
527 	 * OK, it's a trunk.  Loop over and change all vlan's lladdrs on it.
528 	 */
529 #ifdef VLAN_ARRAY
530 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
531 		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
532 #else /* VLAN_ARRAY */
533 	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
534 		LIST_FOREACH_SAFE(ifv, &ifp->if_vlantrunk->hash[i], ifv_list, next) {
535 #endif /* VLAN_ARRAY */
536 			VLAN_UNLOCK();
537 			if_setlladdr(ifv->ifv_ifp, IF_LLADDR(ifp),
538 			    ifp->if_addrlen);
539 			VLAN_LOCK();
540 		}
541 	VLAN_UNLOCK();
542 
543 }
544 
545 /*
546  * A handler for network interface departure events.
547  * Track departure of trunks here so that we don't access invalid
548  * pointers or whatever if a trunk is ripped from under us, e.g.,
549  * by ejecting its hot-plug card.  However, if an ifnet is simply
550  * being renamed, then there's no need to tear down the state.
551  */
552 static void
553 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
554 {
555 	struct ifvlan *ifv;
556 	int i;
557 
558 	/*
559 	 * Check if it's a trunk interface first of all
560 	 * to avoid needless locking.
561 	 */
562 	if (ifp->if_vlantrunk == NULL)
563 		return;
564 
565 	/* If the ifnet is just being renamed, don't do anything. */
566 	if (ifp->if_flags & IFF_RENAMING)
567 		return;
568 
569 	VLAN_LOCK();
570 	/*
571 	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
572 	 * Check trunk pointer after each vlan_unconfig() as it will
573 	 * free it and set to NULL after the last vlan was detached.
574 	 */
575 #ifdef VLAN_ARRAY
576 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
577 		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
578 			vlan_unconfig_locked(ifv->ifv_ifp);
579 			if (ifp->if_vlantrunk == NULL)
580 				break;
581 		}
582 #else /* VLAN_ARRAY */
583 restart:
584 	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
585 		if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
586 			vlan_unconfig_locked(ifv->ifv_ifp);
587 			if (ifp->if_vlantrunk)
588 				goto restart;	/* trunk->hwidth can change */
589 			else
590 				break;
591 		}
592 #endif /* VLAN_ARRAY */
593 	/* Trunk should have been destroyed in vlan_unconfig(). */
594 	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
595 	VLAN_UNLOCK();
596 }
597 
598 /*
599  * Return the trunk device for a virtual interface.
600  */
601 static struct ifnet  *
602 vlan_trunkdev(struct ifnet *ifp)
603 {
604 	struct ifvlan *ifv;
605 
606 	if (ifp->if_type != IFT_L2VLAN)
607 		return (NULL);
608 	ifv = ifp->if_softc;
609 	ifp = NULL;
610 	VLAN_LOCK();
611 	if (ifv->ifv_trunk)
612 		ifp = PARENT(ifv);
613 	VLAN_UNLOCK();
614 	return (ifp);
615 }
616 
617 /*
618  * Return the 16bit vlan tag for this interface.
619  */
620 static int
621 vlan_tag(struct ifnet *ifp, uint16_t *tagp)
622 {
623 	struct ifvlan *ifv;
624 
625 	if (ifp->if_type != IFT_L2VLAN)
626 		return (EINVAL);
627 	ifv = ifp->if_softc;
628 	*tagp = ifv->ifv_tag;
629 	return (0);
630 }
631 
632 /*
633  * Return a driver specific cookie for this interface.  Synchronization
634  * with setcookie must be provided by the driver.
635  */
636 static void *
637 vlan_cookie(struct ifnet *ifp)
638 {
639 	struct ifvlan *ifv;
640 
641 	if (ifp->if_type != IFT_L2VLAN)
642 		return (NULL);
643 	ifv = ifp->if_softc;
644 	return (ifv->ifv_cookie);
645 }
646 
647 /*
648  * Store a cookie in our softc that drivers can use to store driver
649  * private per-instance data in.
650  */
651 static int
652 vlan_setcookie(struct ifnet *ifp, void *cookie)
653 {
654 	struct ifvlan *ifv;
655 
656 	if (ifp->if_type != IFT_L2VLAN)
657 		return (EINVAL);
658 	ifv = ifp->if_softc;
659 	ifv->ifv_cookie = cookie;
660 	return (0);
661 }
662 
663 /*
664  * Return the vlan device present at the specific tag.
665  */
666 static struct ifnet *
667 vlan_devat(struct ifnet *ifp, uint16_t tag)
668 {
669 	struct ifvlantrunk *trunk;
670 	struct ifvlan *ifv;
671 
672 	trunk = ifp->if_vlantrunk;
673 	if (trunk == NULL)
674 		return (NULL);
675 	ifp = NULL;
676 	TRUNK_RLOCK(trunk);
677 	ifv = vlan_gethash(trunk, tag);
678 	if (ifv)
679 		ifp = ifv->ifv_ifp;
680 	TRUNK_RUNLOCK(trunk);
681 	return (ifp);
682 }
683 
684 /*
685  * VLAN support can be loaded as a module.  The only place in the
686  * system that's intimately aware of this is ether_input.  We hook
687  * into this code through vlan_input_p which is defined there and
688  * set here.  Noone else in the system should be aware of this so
689  * we use an explicit reference here.
690  */
691 extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
692 
693 /* For if_link_state_change() eyes only... */
694 extern	void (*vlan_link_state_p)(struct ifnet *);
695 
696 static int
697 vlan_modevent(module_t mod, int type, void *data)
698 {
699 
700 	switch (type) {
701 	case MOD_LOAD:
702 		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
703 		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
704 		if (ifdetach_tag == NULL)
705 			return (ENOMEM);
706 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
707 		    vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
708 		if (iflladdr_tag == NULL)
709 			return (ENOMEM);
710 		VLAN_LOCK_INIT();
711 		vlan_input_p = vlan_input;
712 		vlan_link_state_p = vlan_link_state;
713 		vlan_trunk_cap_p = vlan_trunk_capabilities;
714 		vlan_trunkdev_p = vlan_trunkdev;
715 		vlan_cookie_p = vlan_cookie;
716 		vlan_setcookie_p = vlan_setcookie;
717 		vlan_tag_p = vlan_tag;
718 		vlan_devat_p = vlan_devat;
719 #ifndef VIMAGE
720 		if_clone_attach(&vlan_cloner);
721 #endif
722 		if (bootverbose)
723 			printf("vlan: initialized, using "
724 #ifdef VLAN_ARRAY
725 			       "full-size arrays"
726 #else
727 			       "hash tables with chaining"
728 #endif
729 
730 			       "\n");
731 		break;
732 	case MOD_UNLOAD:
733 #ifndef VIMAGE
734 		if_clone_detach(&vlan_cloner);
735 #endif
736 		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
737 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
738 		vlan_input_p = NULL;
739 		vlan_link_state_p = NULL;
740 		vlan_trunk_cap_p = NULL;
741 		vlan_trunkdev_p = NULL;
742 		vlan_tag_p = NULL;
743 		vlan_cookie_p = vlan_cookie;
744 		vlan_setcookie_p = vlan_setcookie;
745 		vlan_devat_p = NULL;
746 		VLAN_LOCK_DESTROY();
747 		if (bootverbose)
748 			printf("vlan: unloaded\n");
749 		break;
750 	default:
751 		return (EOPNOTSUPP);
752 	}
753 	return (0);
754 }
755 
756 static moduledata_t vlan_mod = {
757 	"if_vlan",
758 	vlan_modevent,
759 	0
760 };
761 
762 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
763 MODULE_VERSION(if_vlan, 3);
764 
765 #ifdef VIMAGE
766 static void
767 vnet_vlan_init(const void *unused __unused)
768 {
769 
770 	V_vlan_cloner = vlan_cloner;
771 	if_clone_attach(&V_vlan_cloner);
772 }
773 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
774     vnet_vlan_init, NULL);
775 
776 static void
777 vnet_vlan_uninit(const void *unused __unused)
778 {
779 
780 	if_clone_detach(&V_vlan_cloner);
781 }
782 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
783     vnet_vlan_uninit, NULL);
784 #endif
785 
786 static struct ifnet *
787 vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
788 {
789 	const char *cp;
790 	struct ifnet *ifp;
791 	int t;
792 
793 	/* Check for <etherif>.<vlan> style interface names. */
794 	IFNET_RLOCK_NOSLEEP();
795 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
796 		/*
797 		 * We can handle non-ethernet hardware types as long as
798 		 * they handle the tagging and headers themselves.
799 		 */
800 		if (ifp->if_type != IFT_ETHER &&
801 		    (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
802 			continue;
803 		if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
804 			continue;
805 		cp = name + strlen(ifp->if_xname);
806 		if (*cp++ != '.')
807 			continue;
808 		if (*cp == '\0')
809 			continue;
810 		t = 0;
811 		for(; *cp >= '0' && *cp <= '9'; cp++)
812 			t = (t * 10) + (*cp - '0');
813 		if (*cp != '\0')
814 			continue;
815 		if (tag != NULL)
816 			*tag = t;
817 		break;
818 	}
819 	IFNET_RUNLOCK_NOSLEEP();
820 
821 	return (ifp);
822 }
823 
824 static int
825 vlan_clone_match(struct if_clone *ifc, const char *name)
826 {
827 	const char *cp;
828 
829 	if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
830 		return (1);
831 
832 	if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
833 		return (0);
834 	for (cp = name + 4; *cp != '\0'; cp++) {
835 		if (*cp < '0' || *cp > '9')
836 			return (0);
837 	}
838 
839 	return (1);
840 }
841 
842 static int
843 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
844 {
845 	char *dp;
846 	int wildcard;
847 	int unit;
848 	int error;
849 	int tag;
850 	int ethertag;
851 	struct ifvlan *ifv;
852 	struct ifnet *ifp;
853 	struct ifnet *p;
854 	struct ifaddr *ifa;
855 	struct sockaddr_dl *sdl;
856 	struct vlanreq vlr;
857 	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
858 
859 	/*
860 	 * There are 3 (ugh) ways to specify the cloned device:
861 	 * o pass a parameter block with the clone request.
862 	 * o specify parameters in the text of the clone device name
863 	 * o specify no parameters and get an unattached device that
864 	 *   must be configured separately.
865 	 * The first technique is preferred; the latter two are
866 	 * supported for backwards compatibilty.
867 	 */
868 	if (params) {
869 		error = copyin(params, &vlr, sizeof(vlr));
870 		if (error)
871 			return error;
872 		p = ifunit(vlr.vlr_parent);
873 		if (p == NULL)
874 			return ENXIO;
875 		/*
876 		 * Don't let the caller set up a VLAN tag with
877 		 * anything except VLID bits.
878 		 */
879 		if (vlr.vlr_tag & ~EVL_VLID_MASK)
880 			return (EINVAL);
881 		error = ifc_name2unit(name, &unit);
882 		if (error != 0)
883 			return (error);
884 
885 		ethertag = 1;
886 		tag = vlr.vlr_tag;
887 		wildcard = (unit < 0);
888 	} else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
889 		ethertag = 1;
890 		unit = -1;
891 		wildcard = 0;
892 
893 		/*
894 		 * Don't let the caller set up a VLAN tag with
895 		 * anything except VLID bits.
896 		 */
897 		if (tag & ~EVL_VLID_MASK)
898 			return (EINVAL);
899 	} else {
900 		ethertag = 0;
901 
902 		error = ifc_name2unit(name, &unit);
903 		if (error != 0)
904 			return (error);
905 
906 		wildcard = (unit < 0);
907 	}
908 
909 	error = ifc_alloc_unit(ifc, &unit);
910 	if (error != 0)
911 		return (error);
912 
913 	/* In the wildcard case, we need to update the name. */
914 	if (wildcard) {
915 		for (dp = name; *dp != '\0'; dp++);
916 		if (snprintf(dp, len - (dp-name), "%d", unit) >
917 		    len - (dp-name) - 1) {
918 			panic("%s: interface name too long", __func__);
919 		}
920 	}
921 
922 	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
923 	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
924 	if (ifp == NULL) {
925 		ifc_free_unit(ifc, unit);
926 		free(ifv, M_VLAN);
927 		return (ENOSPC);
928 	}
929 	SLIST_INIT(&ifv->vlan_mc_listhead);
930 
931 	ifp->if_softc = ifv;
932 	/*
933 	 * Set the name manually rather than using if_initname because
934 	 * we don't conform to the default naming convention for interfaces.
935 	 */
936 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
937 	ifp->if_dname = ifc->ifc_name;
938 	ifp->if_dunit = unit;
939 	/* NB: flags are not set here */
940 	ifp->if_linkmib = &ifv->ifv_mib;
941 	ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
942 	/* NB: mtu is not set here */
943 
944 	ifp->if_init = vlan_init;
945 	ifp->if_start = vlan_start;
946 	ifp->if_ioctl = vlan_ioctl;
947 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
948 	ifp->if_flags = VLAN_IFFLAGS;
949 	ether_ifattach(ifp, eaddr);
950 	/* Now undo some of the damage... */
951 	ifp->if_baudrate = 0;
952 	ifp->if_type = IFT_L2VLAN;
953 	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
954 	ifa = ifp->if_addr;
955 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
956 	sdl->sdl_type = IFT_L2VLAN;
957 
958 	if (ethertag) {
959 		error = vlan_config(ifv, p, tag);
960 		if (error != 0) {
961 			/*
962 			 * Since we've partialy failed, we need to back
963 			 * out all the way, otherwise userland could get
964 			 * confused.  Thus, we destroy the interface.
965 			 */
966 			ether_ifdetach(ifp);
967 			vlan_unconfig(ifp);
968 			if_free_type(ifp, IFT_ETHER);
969 			ifc_free_unit(ifc, unit);
970 			free(ifv, M_VLAN);
971 
972 			return (error);
973 		}
974 
975 		/* Update flags on the parent, if necessary. */
976 		vlan_setflags(ifp, 1);
977 	}
978 
979 	return (0);
980 }
981 
982 static int
983 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
984 {
985 	struct ifvlan *ifv = ifp->if_softc;
986 	int unit = ifp->if_dunit;
987 
988 	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
989 	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
990 	if_free_type(ifp, IFT_ETHER);
991 	free(ifv, M_VLAN);
992 	ifc_free_unit(ifc, unit);
993 
994 	return (0);
995 }
996 
997 /*
998  * The ifp->if_init entry point for vlan(4) is a no-op.
999  */
1000 static void
1001 vlan_init(void *foo __unused)
1002 {
1003 }
1004 
1005 /*
1006  * The if_start method for vlan(4) interface. It doesn't
1007  * raises the IFF_DRV_OACTIVE flag, since it is called
1008  * only from IFQ_HANDOFF() macro in ether_output_frame().
1009  * If the interface queue is full, and vlan_start() is
1010  * not called, the queue would never get emptied and
1011  * interface would stall forever.
1012  */
1013 static void
1014 vlan_start(struct ifnet *ifp)
1015 {
1016 	struct ifvlan *ifv;
1017 	struct ifnet *p;
1018 	struct mbuf *m;
1019 	int error;
1020 
1021 	ifv = ifp->if_softc;
1022 	p = PARENT(ifv);
1023 
1024 	for (;;) {
1025 		IF_DEQUEUE(&ifp->if_snd, m);
1026 		if (m == NULL)
1027 			break;
1028 		BPF_MTAP(ifp, m);
1029 
1030 		/*
1031 		 * Do not run parent's if_start() if the parent is not up,
1032 		 * or parent's driver will cause a system crash.
1033 		 */
1034 		if (!UP_AND_RUNNING(p)) {
1035 			m_freem(m);
1036 			ifp->if_collisions++;
1037 			continue;
1038 		}
1039 
1040 		/*
1041 		 * Pad the frame to the minimum size allowed if told to.
1042 		 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
1043 		 * paragraph C.4.4.3.b.  It can help to work around buggy
1044 		 * bridges that violate paragraph C.4.4.3.a from the same
1045 		 * document, i.e., fail to pad short frames after untagging.
1046 		 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
1047 		 * untagging it will produce a 62-byte frame, which is a runt
1048 		 * and requires padding.  There are VLAN-enabled network
1049 		 * devices that just discard such runts instead or mishandle
1050 		 * them somehow.
1051 		 */
1052 		if (soft_pad && p->if_type == IFT_ETHER) {
1053 			static char pad[8];	/* just zeros */
1054 			int n;
1055 
1056 			for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
1057 			     n > 0; n -= sizeof(pad))
1058 				if (!m_append(m, min(n, sizeof(pad)), pad))
1059 					break;
1060 
1061 			if (n > 0) {
1062 				if_printf(ifp, "cannot pad short frame\n");
1063 				ifp->if_oerrors++;
1064 				m_freem(m);
1065 				continue;
1066 			}
1067 		}
1068 
1069 		/*
1070 		 * If underlying interface can do VLAN tag insertion itself,
1071 		 * just pass the packet along. However, we need some way to
1072 		 * tell the interface where the packet came from so that it
1073 		 * knows how to find the VLAN tag to use, so we attach a
1074 		 * packet tag that holds it.
1075 		 */
1076 		if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1077 			m->m_pkthdr.ether_vtag = ifv->ifv_tag;
1078 			m->m_flags |= M_VLANTAG;
1079 		} else {
1080 			m = ether_vlanencap(m, ifv->ifv_tag);
1081 			if (m == NULL) {
1082 				if_printf(ifp,
1083 				    "unable to prepend VLAN header\n");
1084 				ifp->if_oerrors++;
1085 				continue;
1086 			}
1087 		}
1088 
1089 		/*
1090 		 * Send it, precisely as ether_output() would have.
1091 		 * We are already running at splimp.
1092 		 */
1093 		error = (p->if_transmit)(p, m);
1094 		if (!error)
1095 			ifp->if_opackets++;
1096 		else
1097 			ifp->if_oerrors++;
1098 	}
1099 }
1100 
1101 static void
1102 vlan_input(struct ifnet *ifp, struct mbuf *m)
1103 {
1104 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1105 	struct ifvlan *ifv;
1106 	uint16_t tag;
1107 
1108 	KASSERT(trunk != NULL, ("%s: no trunk", __func__));
1109 
1110 	if (m->m_flags & M_VLANTAG) {
1111 		/*
1112 		 * Packet is tagged, but m contains a normal
1113 		 * Ethernet frame; the tag is stored out-of-band.
1114 		 */
1115 		tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
1116 		m->m_flags &= ~M_VLANTAG;
1117 	} else {
1118 		struct ether_vlan_header *evl;
1119 
1120 		/*
1121 		 * Packet is tagged in-band as specified by 802.1q.
1122 		 */
1123 		switch (ifp->if_type) {
1124 		case IFT_ETHER:
1125 			if (m->m_len < sizeof(*evl) &&
1126 			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
1127 				if_printf(ifp, "cannot pullup VLAN header\n");
1128 				return;
1129 			}
1130 			evl = mtod(m, struct ether_vlan_header *);
1131 			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1132 
1133 			/*
1134 			 * Remove the 802.1q header by copying the Ethernet
1135 			 * addresses over it and adjusting the beginning of
1136 			 * the data in the mbuf.  The encapsulated Ethernet
1137 			 * type field is already in place.
1138 			 */
1139 			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1140 			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
1141 			m_adj(m, ETHER_VLAN_ENCAP_LEN);
1142 			break;
1143 
1144 		default:
1145 #ifdef INVARIANTS
1146 			panic("%s: %s has unsupported if_type %u",
1147 			      __func__, ifp->if_xname, ifp->if_type);
1148 #endif
1149 			m_freem(m);
1150 			ifp->if_noproto++;
1151 			return;
1152 		}
1153 	}
1154 
1155 	TRUNK_RLOCK(trunk);
1156 	ifv = vlan_gethash(trunk, tag);
1157 	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1158 		TRUNK_RUNLOCK(trunk);
1159 		m_freem(m);
1160 		ifp->if_noproto++;
1161 		return;
1162 	}
1163 	TRUNK_RUNLOCK(trunk);
1164 
1165 	m->m_pkthdr.rcvif = ifv->ifv_ifp;
1166 	ifv->ifv_ifp->if_ipackets++;
1167 
1168 	/* Pass it back through the parent's input routine. */
1169 	(*ifp->if_input)(ifv->ifv_ifp, m);
1170 }
1171 
1172 static int
1173 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
1174 {
1175 	struct ifvlantrunk *trunk;
1176 	struct ifnet *ifp;
1177 	int error = 0;
1178 
1179 	/* VID numbers 0x0 and 0xFFF are reserved */
1180 	if (tag == 0 || tag == 0xFFF)
1181 		return (EINVAL);
1182 	if (p->if_type != IFT_ETHER &&
1183 	    (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1184 		return (EPROTONOSUPPORT);
1185 	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1186 		return (EPROTONOSUPPORT);
1187 	if (ifv->ifv_trunk)
1188 		return (EBUSY);
1189 
1190 	if (p->if_vlantrunk == NULL) {
1191 		trunk = malloc(sizeof(struct ifvlantrunk),
1192 		    M_VLAN, M_WAITOK | M_ZERO);
1193 		vlan_inithash(trunk);
1194 		VLAN_LOCK();
1195 		if (p->if_vlantrunk != NULL) {
1196 			/* A race that that is very unlikely to be hit. */
1197 			vlan_freehash(trunk);
1198 			free(trunk, M_VLAN);
1199 			goto exists;
1200 		}
1201 		TRUNK_LOCK_INIT(trunk);
1202 		TRUNK_LOCK(trunk);
1203 		p->if_vlantrunk = trunk;
1204 		trunk->parent = p;
1205 	} else {
1206 		VLAN_LOCK();
1207 exists:
1208 		trunk = p->if_vlantrunk;
1209 		TRUNK_LOCK(trunk);
1210 	}
1211 
1212 	ifv->ifv_tag = tag;	/* must set this before vlan_inshash() */
1213 	error = vlan_inshash(trunk, ifv);
1214 	if (error)
1215 		goto done;
1216 	ifv->ifv_proto = ETHERTYPE_VLAN;
1217 	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1218 	ifv->ifv_mintu = ETHERMIN;
1219 	ifv->ifv_pflags = 0;
1220 
1221 	/*
1222 	 * If the parent supports the VLAN_MTU capability,
1223 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1224 	 * use it.
1225 	 */
1226 	if (p->if_capenable & IFCAP_VLAN_MTU) {
1227 		/*
1228 		 * No need to fudge the MTU since the parent can
1229 		 * handle extended frames.
1230 		 */
1231 		ifv->ifv_mtufudge = 0;
1232 	} else {
1233 		/*
1234 		 * Fudge the MTU by the encapsulation size.  This
1235 		 * makes us incompatible with strictly compliant
1236 		 * 802.1Q implementations, but allows us to use
1237 		 * the feature with other NetBSD implementations,
1238 		 * which might still be useful.
1239 		 */
1240 		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1241 	}
1242 
1243 	ifv->ifv_trunk = trunk;
1244 	ifp = ifv->ifv_ifp;
1245 	/*
1246 	 * Initialize fields from our parent.  This duplicates some
1247 	 * work with ether_ifattach() but allows for non-ethernet
1248 	 * interfaces to also work.
1249 	 */
1250 	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1251 	ifp->if_baudrate = p->if_baudrate;
1252 	ifp->if_output = p->if_output;
1253 	ifp->if_input = p->if_input;
1254 	ifp->if_resolvemulti = p->if_resolvemulti;
1255 	ifp->if_addrlen = p->if_addrlen;
1256 	ifp->if_broadcastaddr = p->if_broadcastaddr;
1257 
1258 	/*
1259 	 * Copy only a selected subset of flags from the parent.
1260 	 * Other flags are none of our business.
1261 	 */
1262 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1263 	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1264 	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1265 #undef VLAN_COPY_FLAGS
1266 
1267 	ifp->if_link_state = p->if_link_state;
1268 
1269 	vlan_capabilities(ifv);
1270 
1271 	/*
1272 	 * Set up our interface address to reflect the underlying
1273 	 * physical interface's.
1274 	 */
1275 	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1276 	((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1277 	    p->if_addrlen;
1278 
1279 	/*
1280 	 * Configure multicast addresses that may already be
1281 	 * joined on the vlan device.
1282 	 */
1283 	(void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1284 
1285 	/* We are ready for operation now. */
1286 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1287 done:
1288 	TRUNK_UNLOCK(trunk);
1289 	if (error == 0)
1290 		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_tag);
1291 	VLAN_UNLOCK();
1292 
1293 	return (error);
1294 }
1295 
1296 static void
1297 vlan_unconfig(struct ifnet *ifp)
1298 {
1299 
1300 	VLAN_LOCK();
1301 	vlan_unconfig_locked(ifp);
1302 	VLAN_UNLOCK();
1303 }
1304 
1305 static void
1306 vlan_unconfig_locked(struct ifnet *ifp)
1307 {
1308 	struct ifvlantrunk *trunk;
1309 	struct vlan_mc_entry *mc;
1310 	struct ifvlan *ifv;
1311 	struct ifnet  *parent;
1312 
1313 	VLAN_LOCK_ASSERT();
1314 
1315 	ifv = ifp->if_softc;
1316 	trunk = ifv->ifv_trunk;
1317 	parent = NULL;
1318 
1319 	if (trunk != NULL) {
1320 
1321 		TRUNK_LOCK(trunk);
1322 		parent = trunk->parent;
1323 
1324 		/*
1325 		 * Since the interface is being unconfigured, we need to
1326 		 * empty the list of multicast groups that we may have joined
1327 		 * while we were alive from the parent's list.
1328 		 */
1329 		while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1330 			/*
1331 			 * This may fail if the parent interface is
1332 			 * being detached.  Regardless, we should do a
1333 			 * best effort to free this interface as much
1334 			 * as possible as all callers expect vlan
1335 			 * destruction to succeed.
1336 			 */
1337 			(void)if_delmulti(parent,
1338 			    (struct sockaddr *)&mc->mc_addr);
1339 			SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1340 			free(mc, M_VLAN);
1341 		}
1342 
1343 		vlan_setflags(ifp, 0); /* clear special flags on parent */
1344 		vlan_remhash(trunk, ifv);
1345 		ifv->ifv_trunk = NULL;
1346 
1347 		/*
1348 		 * Check if we were the last.
1349 		 */
1350 		if (trunk->refcnt == 0) {
1351 			trunk->parent->if_vlantrunk = NULL;
1352 			/*
1353 			 * XXXGL: If some ithread has already entered
1354 			 * vlan_input() and is now blocked on the trunk
1355 			 * lock, then it should preempt us right after
1356 			 * unlock and finish its work. Then we will acquire
1357 			 * lock again in trunk_destroy().
1358 			 */
1359 			TRUNK_UNLOCK(trunk);
1360 			trunk_destroy(trunk);
1361 		} else
1362 			TRUNK_UNLOCK(trunk);
1363 	}
1364 
1365 	/* Disconnect from parent. */
1366 	if (ifv->ifv_pflags)
1367 		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1368 	ifp->if_mtu = ETHERMTU;
1369 	ifp->if_link_state = LINK_STATE_UNKNOWN;
1370 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1371 
1372 	/*
1373 	 * Only dispatch an event if vlan was
1374 	 * attached, otherwise there is nothing
1375 	 * to cleanup anyway.
1376 	 */
1377 	if (parent != NULL)
1378 		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_tag);
1379 }
1380 
1381 /* Handle a reference counted flag that should be set on the parent as well */
1382 static int
1383 vlan_setflag(struct ifnet *ifp, int flag, int status,
1384 	     int (*func)(struct ifnet *, int))
1385 {
1386 	struct ifvlan *ifv;
1387 	int error;
1388 
1389 	/* XXX VLAN_LOCK_ASSERT(); */
1390 
1391 	ifv = ifp->if_softc;
1392 	status = status ? (ifp->if_flags & flag) : 0;
1393 	/* Now "status" contains the flag value or 0 */
1394 
1395 	/*
1396 	 * See if recorded parent's status is different from what
1397 	 * we want it to be.  If it is, flip it.  We record parent's
1398 	 * status in ifv_pflags so that we won't clear parent's flag
1399 	 * we haven't set.  In fact, we don't clear or set parent's
1400 	 * flags directly, but get or release references to them.
1401 	 * That's why we can be sure that recorded flags still are
1402 	 * in accord with actual parent's flags.
1403 	 */
1404 	if (status != (ifv->ifv_pflags & flag)) {
1405 		error = (*func)(PARENT(ifv), status);
1406 		if (error)
1407 			return (error);
1408 		ifv->ifv_pflags &= ~flag;
1409 		ifv->ifv_pflags |= status;
1410 	}
1411 	return (0);
1412 }
1413 
1414 /*
1415  * Handle IFF_* flags that require certain changes on the parent:
1416  * if "status" is true, update parent's flags respective to our if_flags;
1417  * if "status" is false, forcedly clear the flags set on parent.
1418  */
1419 static int
1420 vlan_setflags(struct ifnet *ifp, int status)
1421 {
1422 	int error, i;
1423 
1424 	for (i = 0; vlan_pflags[i].flag; i++) {
1425 		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1426 				     status, vlan_pflags[i].func);
1427 		if (error)
1428 			return (error);
1429 	}
1430 	return (0);
1431 }
1432 
1433 /* Inform all vlans that their parent has changed link state */
1434 static void
1435 vlan_link_state(struct ifnet *ifp)
1436 {
1437 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1438 	struct ifvlan *ifv;
1439 	int i;
1440 
1441 	TRUNK_LOCK(trunk);
1442 #ifdef VLAN_ARRAY
1443 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1444 		if (trunk->vlans[i] != NULL) {
1445 			ifv = trunk->vlans[i];
1446 #else
1447 	for (i = 0; i < (1 << trunk->hwidth); i++)
1448 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1449 #endif
1450 			ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1451 			if_link_state_change(ifv->ifv_ifp,
1452 			    trunk->parent->if_link_state);
1453 		}
1454 	TRUNK_UNLOCK(trunk);
1455 }
1456 
1457 static void
1458 vlan_capabilities(struct ifvlan *ifv)
1459 {
1460 	struct ifnet *p = PARENT(ifv);
1461 	struct ifnet *ifp = ifv->ifv_ifp;
1462 
1463 	TRUNK_LOCK_ASSERT(TRUNK(ifv));
1464 
1465 	/*
1466 	 * If the parent interface can do checksum offloading
1467 	 * on VLANs, then propagate its hardware-assisted
1468 	 * checksumming flags. Also assert that checksum
1469 	 * offloading requires hardware VLAN tagging.
1470 	 */
1471 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1472 		ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1473 
1474 	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1475 	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1476 		ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1477 		ifp->if_hwassist = p->if_hwassist & (CSUM_IP | CSUM_TCP |
1478 		    CSUM_UDP | CSUM_SCTP | CSUM_IP_FRAGS | CSUM_FRAGMENT);
1479 	} else {
1480 		ifp->if_capenable = 0;
1481 		ifp->if_hwassist = 0;
1482 	}
1483 	/*
1484 	 * If the parent interface can do TSO on VLANs then
1485 	 * propagate the hardware-assisted flag. TSO on VLANs
1486 	 * does not necessarily require hardware VLAN tagging.
1487 	 */
1488 	if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1489 		ifp->if_capabilities |= p->if_capabilities & IFCAP_TSO;
1490 	if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1491 		ifp->if_capenable |= p->if_capenable & IFCAP_TSO;
1492 		ifp->if_hwassist |= p->if_hwassist & CSUM_TSO;
1493 	} else {
1494 		ifp->if_capenable &= ~(p->if_capenable & IFCAP_TSO);
1495 		ifp->if_hwassist &= ~(p->if_hwassist & CSUM_TSO);
1496 	}
1497 }
1498 
1499 static void
1500 vlan_trunk_capabilities(struct ifnet *ifp)
1501 {
1502 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1503 	struct ifvlan *ifv;
1504 	int i;
1505 
1506 	TRUNK_LOCK(trunk);
1507 #ifdef VLAN_ARRAY
1508 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1509 		if (trunk->vlans[i] != NULL) {
1510 			ifv = trunk->vlans[i];
1511 #else
1512 	for (i = 0; i < (1 << trunk->hwidth); i++) {
1513 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1514 #endif
1515 			vlan_capabilities(ifv);
1516 	}
1517 	TRUNK_UNLOCK(trunk);
1518 }
1519 
1520 static int
1521 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1522 {
1523 	struct ifnet *p;
1524 	struct ifreq *ifr;
1525 	struct ifaddr *ifa;
1526 	struct ifvlan *ifv;
1527 	struct vlanreq vlr;
1528 	int error = 0;
1529 
1530 	ifr = (struct ifreq *)data;
1531 	ifa = (struct ifaddr *) data;
1532 	ifv = ifp->if_softc;
1533 
1534 	switch (cmd) {
1535 	case SIOCSIFADDR:
1536 		ifp->if_flags |= IFF_UP;
1537 #ifdef INET
1538 		if (ifa->ifa_addr->sa_family == AF_INET)
1539 			arp_ifinit(ifp, ifa);
1540 #endif
1541 		break;
1542 	case SIOCGIFADDR:
1543                 {
1544 			struct sockaddr *sa;
1545 
1546 			sa = (struct sockaddr *)&ifr->ifr_data;
1547 			bcopy(IF_LLADDR(ifp), sa->sa_data, ifp->if_addrlen);
1548                 }
1549 		break;
1550 	case SIOCGIFMEDIA:
1551 		VLAN_LOCK();
1552 		if (TRUNK(ifv) != NULL) {
1553 			p = PARENT(ifv);
1554 			VLAN_UNLOCK();
1555 			error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1556 			/* Limit the result to the parent's current config. */
1557 			if (error == 0) {
1558 				struct ifmediareq *ifmr;
1559 
1560 				ifmr = (struct ifmediareq *)data;
1561 				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1562 					ifmr->ifm_count = 1;
1563 					error = copyout(&ifmr->ifm_current,
1564 						ifmr->ifm_ulist,
1565 						sizeof(int));
1566 				}
1567 			}
1568 		} else {
1569 			VLAN_UNLOCK();
1570 			error = EINVAL;
1571 		}
1572 		break;
1573 
1574 	case SIOCSIFMEDIA:
1575 		error = EINVAL;
1576 		break;
1577 
1578 	case SIOCSIFMTU:
1579 		/*
1580 		 * Set the interface MTU.
1581 		 */
1582 		VLAN_LOCK();
1583 		if (TRUNK(ifv) != NULL) {
1584 			if (ifr->ifr_mtu >
1585 			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1586 			    ifr->ifr_mtu <
1587 			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1588 				error = EINVAL;
1589 			else
1590 				ifp->if_mtu = ifr->ifr_mtu;
1591 		} else
1592 			error = EINVAL;
1593 		VLAN_UNLOCK();
1594 		break;
1595 
1596 	case SIOCSETVLAN:
1597 #ifdef VIMAGE
1598 		if (ifp->if_vnet != ifp->if_home_vnet) {
1599 			error = EPERM;
1600 			break;
1601 		}
1602 #endif
1603 		error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1604 		if (error)
1605 			break;
1606 		if (vlr.vlr_parent[0] == '\0') {
1607 			vlan_unconfig(ifp);
1608 			break;
1609 		}
1610 		p = ifunit(vlr.vlr_parent);
1611 		if (p == NULL) {
1612 			error = ENOENT;
1613 			break;
1614 		}
1615 		/*
1616 		 * Don't let the caller set up a VLAN tag with
1617 		 * anything except VLID bits.
1618 		 */
1619 		if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1620 			error = EINVAL;
1621 			break;
1622 		}
1623 		error = vlan_config(ifv, p, vlr.vlr_tag);
1624 		if (error)
1625 			break;
1626 
1627 		/* Update flags on the parent, if necessary. */
1628 		vlan_setflags(ifp, 1);
1629 		break;
1630 
1631 	case SIOCGETVLAN:
1632 #ifdef VIMAGE
1633 		if (ifp->if_vnet != ifp->if_home_vnet) {
1634 			error = EPERM;
1635 			break;
1636 		}
1637 #endif
1638 		bzero(&vlr, sizeof(vlr));
1639 		VLAN_LOCK();
1640 		if (TRUNK(ifv) != NULL) {
1641 			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1642 			    sizeof(vlr.vlr_parent));
1643 			vlr.vlr_tag = ifv->ifv_tag;
1644 		}
1645 		VLAN_UNLOCK();
1646 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1647 		break;
1648 
1649 	case SIOCSIFFLAGS:
1650 		/*
1651 		 * We should propagate selected flags to the parent,
1652 		 * e.g., promiscuous mode.
1653 		 */
1654 		if (TRUNK(ifv) != NULL)
1655 			error = vlan_setflags(ifp, 1);
1656 		break;
1657 
1658 	case SIOCADDMULTI:
1659 	case SIOCDELMULTI:
1660 		/*
1661 		 * If we don't have a parent, just remember the membership for
1662 		 * when we do.
1663 		 */
1664 		if (TRUNK(ifv) != NULL)
1665 			error = vlan_setmulti(ifp);
1666 		break;
1667 
1668 	default:
1669 		error = EINVAL;
1670 		break;
1671 	}
1672 
1673 	return (error);
1674 }
1675