xref: /freebsd/sys/net/if_vlan.c (revision 10b59a9b4add0320d52c15ce057dd697261e7dfc)
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 static SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0,
135     "IEEE 802.1Q VLAN");
136 static SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0,
137     "for consistency");
138 
139 static int soft_pad = 0;
140 SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW, &soft_pad, 0,
141 	   "pad short frames before tagging");
142 
143 static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface");
144 
145 static eventhandler_tag ifdetach_tag;
146 static eventhandler_tag iflladdr_tag;
147 
148 /*
149  * We have a global mutex, that is used to serialize configuration
150  * changes and isn't used in normal packet delivery.
151  *
152  * We also have a per-trunk rwlock, that is locked shared on packet
153  * processing and exclusive when configuration is changed.
154  *
155  * The VLAN_ARRAY substitutes the dynamic hash with a static array
156  * with 4096 entries. In theory this can give a boost in processing,
157  * however on practice it does not. Probably this is because array
158  * is too big to fit into CPU cache.
159  */
160 static struct sx ifv_lock;
161 #define	VLAN_LOCK_INIT()	sx_init(&ifv_lock, "vlan_global")
162 #define	VLAN_LOCK_DESTROY()	sx_destroy(&ifv_lock)
163 #define	VLAN_LOCK_ASSERT()	sx_assert(&ifv_lock, SA_LOCKED)
164 #define	VLAN_LOCK()		sx_xlock(&ifv_lock)
165 #define	VLAN_UNLOCK()		sx_xunlock(&ifv_lock)
166 #define	TRUNK_LOCK_INIT(trunk)	rw_init(&(trunk)->rw, VLANNAME)
167 #define	TRUNK_LOCK_DESTROY(trunk) rw_destroy(&(trunk)->rw)
168 #define	TRUNK_LOCK(trunk)	rw_wlock(&(trunk)->rw)
169 #define	TRUNK_UNLOCK(trunk)	rw_wunlock(&(trunk)->rw)
170 #define	TRUNK_LOCK_ASSERT(trunk) rw_assert(&(trunk)->rw, RA_WLOCKED)
171 #define	TRUNK_RLOCK(trunk)	rw_rlock(&(trunk)->rw)
172 #define	TRUNK_RUNLOCK(trunk)	rw_runlock(&(trunk)->rw)
173 #define	TRUNK_LOCK_RASSERT(trunk) rw_assert(&(trunk)->rw, RA_RLOCKED)
174 
175 #ifndef VLAN_ARRAY
176 static	void vlan_inithash(struct ifvlantrunk *trunk);
177 static	void vlan_freehash(struct ifvlantrunk *trunk);
178 static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
179 static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
180 static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
181 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
182 	uint16_t tag);
183 #endif
184 static	void trunk_destroy(struct ifvlantrunk *trunk);
185 
186 static	void vlan_start(struct ifnet *ifp);
187 static	void vlan_init(void *foo);
188 static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
189 static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
190 static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
191     int (*func)(struct ifnet *, int));
192 static	int vlan_setflags(struct ifnet *ifp, int status);
193 static	int vlan_setmulti(struct ifnet *ifp);
194 static	void vlan_unconfig(struct ifnet *ifp);
195 static	void vlan_unconfig_locked(struct ifnet *ifp);
196 static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
197 static	void vlan_link_state(struct ifnet *ifp);
198 static	void vlan_capabilities(struct ifvlan *ifv);
199 static	void vlan_trunk_capabilities(struct ifnet *ifp);
200 
201 static	struct ifnet *vlan_clone_match_ethertag(struct if_clone *,
202     const char *, int *);
203 static	int vlan_clone_match(struct if_clone *, const char *);
204 static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
205 static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
206 
207 static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
208 static  void vlan_iflladdr(void *arg, struct ifnet *ifp);
209 
210 static	struct if_clone vlan_cloner = IFC_CLONE_INITIALIZER(VLANNAME, NULL,
211     IF_MAXUNIT, NULL, vlan_clone_match, vlan_clone_create, vlan_clone_destroy);
212 
213 #ifdef VIMAGE
214 static VNET_DEFINE(struct if_clone, vlan_cloner);
215 #define	V_vlan_cloner	VNET(vlan_cloner)
216 #endif
217 
218 #ifndef VLAN_ARRAY
219 #define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
220 
221 static void
222 vlan_inithash(struct ifvlantrunk *trunk)
223 {
224 	int i, n;
225 
226 	/*
227 	 * The trunk must not be locked here since we call malloc(M_WAITOK).
228 	 * It is OK in case this function is called before the trunk struct
229 	 * gets hooked up and becomes visible from other threads.
230 	 */
231 
232 	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
233 	    ("%s: hash already initialized", __func__));
234 
235 	trunk->hwidth = VLAN_DEF_HWIDTH;
236 	n = 1 << trunk->hwidth;
237 	trunk->hmask = n - 1;
238 	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
239 	for (i = 0; i < n; i++)
240 		LIST_INIT(&trunk->hash[i]);
241 }
242 
243 static void
244 vlan_freehash(struct ifvlantrunk *trunk)
245 {
246 #ifdef INVARIANTS
247 	int i;
248 
249 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
250 	for (i = 0; i < (1 << trunk->hwidth); i++)
251 		KASSERT(LIST_EMPTY(&trunk->hash[i]),
252 		    ("%s: hash table not empty", __func__));
253 #endif
254 	free(trunk->hash, M_VLAN);
255 	trunk->hash = NULL;
256 	trunk->hwidth = trunk->hmask = 0;
257 }
258 
259 static int
260 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
261 {
262 	int i, b;
263 	struct ifvlan *ifv2;
264 
265 	TRUNK_LOCK_ASSERT(trunk);
266 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
267 
268 	b = 1 << trunk->hwidth;
269 	i = HASH(ifv->ifv_tag, trunk->hmask);
270 	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
271 		if (ifv->ifv_tag == ifv2->ifv_tag)
272 			return (EEXIST);
273 
274 	/*
275 	 * Grow the hash when the number of vlans exceeds half of the number of
276 	 * hash buckets squared. This will make the average linked-list length
277 	 * buckets/2.
278 	 */
279 	if (trunk->refcnt > (b * b) / 2) {
280 		vlan_growhash(trunk, 1);
281 		i = HASH(ifv->ifv_tag, trunk->hmask);
282 	}
283 	LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
284 	trunk->refcnt++;
285 
286 	return (0);
287 }
288 
289 static int
290 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
291 {
292 	int i, b;
293 	struct ifvlan *ifv2;
294 
295 	TRUNK_LOCK_ASSERT(trunk);
296 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
297 
298 	b = 1 << trunk->hwidth;
299 	i = HASH(ifv->ifv_tag, trunk->hmask);
300 	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
301 		if (ifv2 == ifv) {
302 			trunk->refcnt--;
303 			LIST_REMOVE(ifv2, ifv_list);
304 			if (trunk->refcnt < (b * b) / 2)
305 				vlan_growhash(trunk, -1);
306 			return (0);
307 		}
308 
309 	panic("%s: vlan not found\n", __func__);
310 	return (ENOENT); /*NOTREACHED*/
311 }
312 
313 /*
314  * Grow the hash larger or smaller if memory permits.
315  */
316 static void
317 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
318 {
319 	struct ifvlan *ifv;
320 	struct ifvlanhead *hash2;
321 	int hwidth2, i, j, n, n2;
322 
323 	TRUNK_LOCK_ASSERT(trunk);
324 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
325 
326 	if (howmuch == 0) {
327 		/* Harmless yet obvious coding error */
328 		printf("%s: howmuch is 0\n", __func__);
329 		return;
330 	}
331 
332 	hwidth2 = trunk->hwidth + howmuch;
333 	n = 1 << trunk->hwidth;
334 	n2 = 1 << hwidth2;
335 	/* Do not shrink the table below the default */
336 	if (hwidth2 < VLAN_DEF_HWIDTH)
337 		return;
338 
339 	/* M_NOWAIT because we're called with trunk mutex held */
340 	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
341 	if (hash2 == NULL) {
342 		printf("%s: out of memory -- hash size not changed\n",
343 		    __func__);
344 		return;		/* We can live with the old hash table */
345 	}
346 	for (j = 0; j < n2; j++)
347 		LIST_INIT(&hash2[j]);
348 	for (i = 0; i < n; i++)
349 		while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
350 			LIST_REMOVE(ifv, ifv_list);
351 			j = HASH(ifv->ifv_tag, n2 - 1);
352 			LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
353 		}
354 	free(trunk->hash, M_VLAN);
355 	trunk->hash = hash2;
356 	trunk->hwidth = hwidth2;
357 	trunk->hmask = n2 - 1;
358 
359 	if (bootverbose)
360 		if_printf(trunk->parent,
361 		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
362 }
363 
364 static __inline struct ifvlan *
365 vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
366 {
367 	struct ifvlan *ifv;
368 
369 	TRUNK_LOCK_RASSERT(trunk);
370 
371 	LIST_FOREACH(ifv, &trunk->hash[HASH(tag, trunk->hmask)], ifv_list)
372 		if (ifv->ifv_tag == tag)
373 			return (ifv);
374 	return (NULL);
375 }
376 
377 #if 0
378 /* Debugging code to view the hashtables. */
379 static void
380 vlan_dumphash(struct ifvlantrunk *trunk)
381 {
382 	int i;
383 	struct ifvlan *ifv;
384 
385 	for (i = 0; i < (1 << trunk->hwidth); i++) {
386 		printf("%d: ", i);
387 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
388 			printf("%s ", ifv->ifv_ifp->if_xname);
389 		printf("\n");
390 	}
391 }
392 #endif /* 0 */
393 #else
394 
395 static __inline struct ifvlan *
396 vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
397 {
398 
399 	return trunk->vlans[tag];
400 }
401 
402 static __inline int
403 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
404 {
405 
406 	if (trunk->vlans[ifv->ifv_tag] != NULL)
407 		return EEXIST;
408 	trunk->vlans[ifv->ifv_tag] = ifv;
409 	trunk->refcnt++;
410 
411 	return (0);
412 }
413 
414 static __inline int
415 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
416 {
417 
418 	trunk->vlans[ifv->ifv_tag] = NULL;
419 	trunk->refcnt--;
420 
421 	return (0);
422 }
423 
424 static __inline void
425 vlan_freehash(struct ifvlantrunk *trunk)
426 {
427 }
428 
429 static __inline void
430 vlan_inithash(struct ifvlantrunk *trunk)
431 {
432 }
433 
434 #endif /* !VLAN_ARRAY */
435 
436 static void
437 trunk_destroy(struct ifvlantrunk *trunk)
438 {
439 	VLAN_LOCK_ASSERT();
440 
441 	TRUNK_LOCK(trunk);
442 	vlan_freehash(trunk);
443 	trunk->parent->if_vlantrunk = NULL;
444 	TRUNK_UNLOCK(trunk);
445 	TRUNK_LOCK_DESTROY(trunk);
446 	free(trunk, M_VLAN);
447 }
448 
449 /*
450  * Program our multicast filter. What we're actually doing is
451  * programming the multicast filter of the parent. This has the
452  * side effect of causing the parent interface to receive multicast
453  * traffic that it doesn't really want, which ends up being discarded
454  * later by the upper protocol layers. Unfortunately, there's no way
455  * to avoid this: there really is only one physical interface.
456  *
457  * XXX: There is a possible race here if more than one thread is
458  *      modifying the multicast state of the vlan interface at the same time.
459  */
460 static int
461 vlan_setmulti(struct ifnet *ifp)
462 {
463 	struct ifnet		*ifp_p;
464 	struct ifmultiaddr	*ifma, *rifma = NULL;
465 	struct ifvlan		*sc;
466 	struct vlan_mc_entry	*mc;
467 	int			error;
468 
469 	/*VLAN_LOCK_ASSERT();*/
470 
471 	/* Find the parent. */
472 	sc = ifp->if_softc;
473 	ifp_p = PARENT(sc);
474 
475 	CURVNET_SET_QUIET(ifp_p->if_vnet);
476 
477 	/* First, remove any existing filter entries. */
478 	while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
479 		error = if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
480 		if (error)
481 			return (error);
482 		SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
483 		free(mc, M_VLAN);
484 	}
485 
486 	/* Now program new ones. */
487 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
488 		if (ifma->ifma_addr->sa_family != AF_LINK)
489 			continue;
490 		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
491 		if (mc == NULL)
492 			return (ENOMEM);
493 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
494 		mc->mc_addr.sdl_index = ifp_p->if_index;
495 		SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
496 		error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
497 		    &rifma);
498 		if (error)
499 			return (error);
500 	}
501 
502 	CURVNET_RESTORE();
503 	return (0);
504 }
505 
506 /*
507  * A handler for parent interface link layer address changes.
508  * If the parent interface link layer address is changed we
509  * should also change it on all children vlans.
510  */
511 static void
512 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
513 {
514 	struct ifvlan *ifv;
515 #ifndef VLAN_ARRAY
516 	struct ifvlan *next;
517 #endif
518 	int i;
519 
520 	/*
521 	 * Check if it's a trunk interface first of all
522 	 * to avoid needless locking.
523 	 */
524 	if (ifp->if_vlantrunk == NULL)
525 		return;
526 
527 	VLAN_LOCK();
528 	/*
529 	 * OK, it's a trunk.  Loop over and change all vlan's lladdrs on it.
530 	 */
531 #ifdef VLAN_ARRAY
532 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
533 		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
534 #else /* VLAN_ARRAY */
535 	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
536 		LIST_FOREACH_SAFE(ifv, &ifp->if_vlantrunk->hash[i], ifv_list, next) {
537 #endif /* VLAN_ARRAY */
538 			VLAN_UNLOCK();
539 			if_setlladdr(ifv->ifv_ifp, IF_LLADDR(ifp),
540 			    ifp->if_addrlen);
541 			VLAN_LOCK();
542 		}
543 	VLAN_UNLOCK();
544 
545 }
546 
547 /*
548  * A handler for network interface departure events.
549  * Track departure of trunks here so that we don't access invalid
550  * pointers or whatever if a trunk is ripped from under us, e.g.,
551  * by ejecting its hot-plug card.  However, if an ifnet is simply
552  * being renamed, then there's no need to tear down the state.
553  */
554 static void
555 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
556 {
557 	struct ifvlan *ifv;
558 	int i;
559 
560 	/*
561 	 * Check if it's a trunk interface first of all
562 	 * to avoid needless locking.
563 	 */
564 	if (ifp->if_vlantrunk == NULL)
565 		return;
566 
567 	/* If the ifnet is just being renamed, don't do anything. */
568 	if (ifp->if_flags & IFF_RENAMING)
569 		return;
570 
571 	VLAN_LOCK();
572 	/*
573 	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
574 	 * Check trunk pointer after each vlan_unconfig() as it will
575 	 * free it and set to NULL after the last vlan was detached.
576 	 */
577 #ifdef VLAN_ARRAY
578 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
579 		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
580 			vlan_unconfig_locked(ifv->ifv_ifp);
581 			if (ifp->if_vlantrunk == NULL)
582 				break;
583 		}
584 #else /* VLAN_ARRAY */
585 restart:
586 	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
587 		if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
588 			vlan_unconfig_locked(ifv->ifv_ifp);
589 			if (ifp->if_vlantrunk)
590 				goto restart;	/* trunk->hwidth can change */
591 			else
592 				break;
593 		}
594 #endif /* VLAN_ARRAY */
595 	/* Trunk should have been destroyed in vlan_unconfig(). */
596 	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
597 	VLAN_UNLOCK();
598 }
599 
600 /*
601  * Return the trunk device for a virtual interface.
602  */
603 static struct ifnet  *
604 vlan_trunkdev(struct ifnet *ifp)
605 {
606 	struct ifvlan *ifv;
607 
608 	if (ifp->if_type != IFT_L2VLAN)
609 		return (NULL);
610 	ifv = ifp->if_softc;
611 	ifp = NULL;
612 	VLAN_LOCK();
613 	if (ifv->ifv_trunk)
614 		ifp = PARENT(ifv);
615 	VLAN_UNLOCK();
616 	return (ifp);
617 }
618 
619 /*
620  * Return the 16bit vlan tag for this interface.
621  */
622 static int
623 vlan_tag(struct ifnet *ifp, uint16_t *tagp)
624 {
625 	struct ifvlan *ifv;
626 
627 	if (ifp->if_type != IFT_L2VLAN)
628 		return (EINVAL);
629 	ifv = ifp->if_softc;
630 	*tagp = ifv->ifv_tag;
631 	return (0);
632 }
633 
634 /*
635  * Return a driver specific cookie for this interface.  Synchronization
636  * with setcookie must be provided by the driver.
637  */
638 static void *
639 vlan_cookie(struct ifnet *ifp)
640 {
641 	struct ifvlan *ifv;
642 
643 	if (ifp->if_type != IFT_L2VLAN)
644 		return (NULL);
645 	ifv = ifp->if_softc;
646 	return (ifv->ifv_cookie);
647 }
648 
649 /*
650  * Store a cookie in our softc that drivers can use to store driver
651  * private per-instance data in.
652  */
653 static int
654 vlan_setcookie(struct ifnet *ifp, void *cookie)
655 {
656 	struct ifvlan *ifv;
657 
658 	if (ifp->if_type != IFT_L2VLAN)
659 		return (EINVAL);
660 	ifv = ifp->if_softc;
661 	ifv->ifv_cookie = cookie;
662 	return (0);
663 }
664 
665 /*
666  * Return the vlan device present at the specific tag.
667  */
668 static struct ifnet *
669 vlan_devat(struct ifnet *ifp, uint16_t tag)
670 {
671 	struct ifvlantrunk *trunk;
672 	struct ifvlan *ifv;
673 
674 	trunk = ifp->if_vlantrunk;
675 	if (trunk == NULL)
676 		return (NULL);
677 	ifp = NULL;
678 	TRUNK_RLOCK(trunk);
679 	ifv = vlan_gethash(trunk, tag);
680 	if (ifv)
681 		ifp = ifv->ifv_ifp;
682 	TRUNK_RUNLOCK(trunk);
683 	return (ifp);
684 }
685 
686 /*
687  * VLAN support can be loaded as a module.  The only place in the
688  * system that's intimately aware of this is ether_input.  We hook
689  * into this code through vlan_input_p which is defined there and
690  * set here.  Noone else in the system should be aware of this so
691  * we use an explicit reference here.
692  */
693 extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
694 
695 /* For if_link_state_change() eyes only... */
696 extern	void (*vlan_link_state_p)(struct ifnet *);
697 
698 static int
699 vlan_modevent(module_t mod, int type, void *data)
700 {
701 
702 	switch (type) {
703 	case MOD_LOAD:
704 		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
705 		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
706 		if (ifdetach_tag == NULL)
707 			return (ENOMEM);
708 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
709 		    vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
710 		if (iflladdr_tag == NULL)
711 			return (ENOMEM);
712 		VLAN_LOCK_INIT();
713 		vlan_input_p = vlan_input;
714 		vlan_link_state_p = vlan_link_state;
715 		vlan_trunk_cap_p = vlan_trunk_capabilities;
716 		vlan_trunkdev_p = vlan_trunkdev;
717 		vlan_cookie_p = vlan_cookie;
718 		vlan_setcookie_p = vlan_setcookie;
719 		vlan_tag_p = vlan_tag;
720 		vlan_devat_p = vlan_devat;
721 #ifndef VIMAGE
722 		if_clone_attach(&vlan_cloner);
723 #endif
724 		if (bootverbose)
725 			printf("vlan: initialized, using "
726 #ifdef VLAN_ARRAY
727 			       "full-size arrays"
728 #else
729 			       "hash tables with chaining"
730 #endif
731 
732 			       "\n");
733 		break;
734 	case MOD_UNLOAD:
735 #ifndef VIMAGE
736 		if_clone_detach(&vlan_cloner);
737 #endif
738 		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
739 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
740 		vlan_input_p = NULL;
741 		vlan_link_state_p = NULL;
742 		vlan_trunk_cap_p = NULL;
743 		vlan_trunkdev_p = NULL;
744 		vlan_tag_p = NULL;
745 		vlan_cookie_p = vlan_cookie;
746 		vlan_setcookie_p = vlan_setcookie;
747 		vlan_devat_p = NULL;
748 		VLAN_LOCK_DESTROY();
749 		if (bootverbose)
750 			printf("vlan: unloaded\n");
751 		break;
752 	default:
753 		return (EOPNOTSUPP);
754 	}
755 	return (0);
756 }
757 
758 static moduledata_t vlan_mod = {
759 	"if_vlan",
760 	vlan_modevent,
761 	0
762 };
763 
764 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
765 MODULE_VERSION(if_vlan, 3);
766 
767 #ifdef VIMAGE
768 static void
769 vnet_vlan_init(const void *unused __unused)
770 {
771 
772 	V_vlan_cloner = vlan_cloner;
773 	if_clone_attach(&V_vlan_cloner);
774 }
775 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
776     vnet_vlan_init, NULL);
777 
778 static void
779 vnet_vlan_uninit(const void *unused __unused)
780 {
781 
782 	if_clone_detach(&V_vlan_cloner);
783 }
784 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
785     vnet_vlan_uninit, NULL);
786 #endif
787 
788 static struct ifnet *
789 vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
790 {
791 	const char *cp;
792 	struct ifnet *ifp;
793 	int t;
794 
795 	/* Check for <etherif>.<vlan> style interface names. */
796 	IFNET_RLOCK_NOSLEEP();
797 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
798 		/*
799 		 * We can handle non-ethernet hardware types as long as
800 		 * they handle the tagging and headers themselves.
801 		 */
802 		if (ifp->if_type != IFT_ETHER &&
803 		    (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
804 			continue;
805 		if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
806 			continue;
807 		cp = name + strlen(ifp->if_xname);
808 		if (*cp++ != '.')
809 			continue;
810 		if (*cp == '\0')
811 			continue;
812 		t = 0;
813 		for(; *cp >= '0' && *cp <= '9'; cp++)
814 			t = (t * 10) + (*cp - '0');
815 		if (*cp != '\0')
816 			continue;
817 		if (tag != NULL)
818 			*tag = t;
819 		break;
820 	}
821 	IFNET_RUNLOCK_NOSLEEP();
822 
823 	return (ifp);
824 }
825 
826 static int
827 vlan_clone_match(struct if_clone *ifc, const char *name)
828 {
829 	const char *cp;
830 
831 	if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
832 		return (1);
833 
834 	if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
835 		return (0);
836 	for (cp = name + 4; *cp != '\0'; cp++) {
837 		if (*cp < '0' || *cp > '9')
838 			return (0);
839 	}
840 
841 	return (1);
842 }
843 
844 static int
845 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
846 {
847 	char *dp;
848 	int wildcard;
849 	int unit;
850 	int error;
851 	int tag;
852 	int ethertag;
853 	struct ifvlan *ifv;
854 	struct ifnet *ifp;
855 	struct ifnet *p;
856 	struct ifaddr *ifa;
857 	struct sockaddr_dl *sdl;
858 	struct vlanreq vlr;
859 	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
860 
861 	/*
862 	 * There are 3 (ugh) ways to specify the cloned device:
863 	 * o pass a parameter block with the clone request.
864 	 * o specify parameters in the text of the clone device name
865 	 * o specify no parameters and get an unattached device that
866 	 *   must be configured separately.
867 	 * The first technique is preferred; the latter two are
868 	 * supported for backwards compatibilty.
869 	 */
870 	if (params) {
871 		error = copyin(params, &vlr, sizeof(vlr));
872 		if (error)
873 			return error;
874 		p = ifunit(vlr.vlr_parent);
875 		if (p == NULL)
876 			return ENXIO;
877 		/*
878 		 * Don't let the caller set up a VLAN tag with
879 		 * anything except VLID bits.
880 		 */
881 		if (vlr.vlr_tag & ~EVL_VLID_MASK)
882 			return (EINVAL);
883 		error = ifc_name2unit(name, &unit);
884 		if (error != 0)
885 			return (error);
886 
887 		ethertag = 1;
888 		tag = vlr.vlr_tag;
889 		wildcard = (unit < 0);
890 	} else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
891 		ethertag = 1;
892 		unit = -1;
893 		wildcard = 0;
894 
895 		/*
896 		 * Don't let the caller set up a VLAN tag with
897 		 * anything except VLID bits.
898 		 */
899 		if (tag & ~EVL_VLID_MASK)
900 			return (EINVAL);
901 	} else {
902 		ethertag = 0;
903 
904 		error = ifc_name2unit(name, &unit);
905 		if (error != 0)
906 			return (error);
907 
908 		wildcard = (unit < 0);
909 	}
910 
911 	error = ifc_alloc_unit(ifc, &unit);
912 	if (error != 0)
913 		return (error);
914 
915 	/* In the wildcard case, we need to update the name. */
916 	if (wildcard) {
917 		for (dp = name; *dp != '\0'; dp++);
918 		if (snprintf(dp, len - (dp-name), "%d", unit) >
919 		    len - (dp-name) - 1) {
920 			panic("%s: interface name too long", __func__);
921 		}
922 	}
923 
924 	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
925 	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
926 	if (ifp == NULL) {
927 		ifc_free_unit(ifc, unit);
928 		free(ifv, M_VLAN);
929 		return (ENOSPC);
930 	}
931 	SLIST_INIT(&ifv->vlan_mc_listhead);
932 
933 	ifp->if_softc = ifv;
934 	/*
935 	 * Set the name manually rather than using if_initname because
936 	 * we don't conform to the default naming convention for interfaces.
937 	 */
938 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
939 	ifp->if_dname = ifc->ifc_name;
940 	ifp->if_dunit = unit;
941 	/* NB: flags are not set here */
942 	ifp->if_linkmib = &ifv->ifv_mib;
943 	ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
944 	/* NB: mtu is not set here */
945 
946 	ifp->if_init = vlan_init;
947 	ifp->if_start = vlan_start;
948 	ifp->if_ioctl = vlan_ioctl;
949 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
950 	ifp->if_flags = VLAN_IFFLAGS;
951 	ether_ifattach(ifp, eaddr);
952 	/* Now undo some of the damage... */
953 	ifp->if_baudrate = 0;
954 	ifp->if_type = IFT_L2VLAN;
955 	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
956 	ifa = ifp->if_addr;
957 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
958 	sdl->sdl_type = IFT_L2VLAN;
959 
960 	if (ethertag) {
961 		error = vlan_config(ifv, p, tag);
962 		if (error != 0) {
963 			/*
964 			 * Since we've partialy failed, we need to back
965 			 * out all the way, otherwise userland could get
966 			 * confused.  Thus, we destroy the interface.
967 			 */
968 			ether_ifdetach(ifp);
969 			vlan_unconfig(ifp);
970 			if_free(ifp);
971 			ifc_free_unit(ifc, unit);
972 			free(ifv, M_VLAN);
973 
974 			return (error);
975 		}
976 
977 		/* Update flags on the parent, if necessary. */
978 		vlan_setflags(ifp, 1);
979 	}
980 
981 	return (0);
982 }
983 
984 static int
985 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
986 {
987 	struct ifvlan *ifv = ifp->if_softc;
988 	int unit = ifp->if_dunit;
989 
990 	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
991 	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
992 	if_free(ifp);
993 	free(ifv, M_VLAN);
994 	ifc_free_unit(ifc, unit);
995 
996 	return (0);
997 }
998 
999 /*
1000  * The ifp->if_init entry point for vlan(4) is a no-op.
1001  */
1002 static void
1003 vlan_init(void *foo __unused)
1004 {
1005 }
1006 
1007 /*
1008  * The if_start method for vlan(4) interface. It doesn't
1009  * raises the IFF_DRV_OACTIVE flag, since it is called
1010  * only from IFQ_HANDOFF() macro in ether_output_frame().
1011  * If the interface queue is full, and vlan_start() is
1012  * not called, the queue would never get emptied and
1013  * interface would stall forever.
1014  */
1015 static void
1016 vlan_start(struct ifnet *ifp)
1017 {
1018 	struct ifvlan *ifv;
1019 	struct ifnet *p;
1020 	struct mbuf *m;
1021 	int error;
1022 
1023 	ifv = ifp->if_softc;
1024 	p = PARENT(ifv);
1025 
1026 	for (;;) {
1027 		IF_DEQUEUE(&ifp->if_snd, m);
1028 		if (m == NULL)
1029 			break;
1030 		BPF_MTAP(ifp, m);
1031 
1032 		/*
1033 		 * Do not run parent's if_start() if the parent is not up,
1034 		 * or parent's driver will cause a system crash.
1035 		 */
1036 		if (!UP_AND_RUNNING(p)) {
1037 			m_freem(m);
1038 			ifp->if_collisions++;
1039 			continue;
1040 		}
1041 
1042 		/*
1043 		 * Pad the frame to the minimum size allowed if told to.
1044 		 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
1045 		 * paragraph C.4.4.3.b.  It can help to work around buggy
1046 		 * bridges that violate paragraph C.4.4.3.a from the same
1047 		 * document, i.e., fail to pad short frames after untagging.
1048 		 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
1049 		 * untagging it will produce a 62-byte frame, which is a runt
1050 		 * and requires padding.  There are VLAN-enabled network
1051 		 * devices that just discard such runts instead or mishandle
1052 		 * them somehow.
1053 		 */
1054 		if (soft_pad && p->if_type == IFT_ETHER) {
1055 			static char pad[8];	/* just zeros */
1056 			int n;
1057 
1058 			for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
1059 			     n > 0; n -= sizeof(pad))
1060 				if (!m_append(m, min(n, sizeof(pad)), pad))
1061 					break;
1062 
1063 			if (n > 0) {
1064 				if_printf(ifp, "cannot pad short frame\n");
1065 				ifp->if_oerrors++;
1066 				m_freem(m);
1067 				continue;
1068 			}
1069 		}
1070 
1071 		/*
1072 		 * If underlying interface can do VLAN tag insertion itself,
1073 		 * just pass the packet along. However, we need some way to
1074 		 * tell the interface where the packet came from so that it
1075 		 * knows how to find the VLAN tag to use, so we attach a
1076 		 * packet tag that holds it.
1077 		 */
1078 		if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1079 			m->m_pkthdr.ether_vtag = ifv->ifv_tag;
1080 			m->m_flags |= M_VLANTAG;
1081 		} else {
1082 			m = ether_vlanencap(m, ifv->ifv_tag);
1083 			if (m == NULL) {
1084 				if_printf(ifp,
1085 				    "unable to prepend VLAN header\n");
1086 				ifp->if_oerrors++;
1087 				continue;
1088 			}
1089 		}
1090 
1091 		/*
1092 		 * Send it, precisely as ether_output() would have.
1093 		 * We are already running at splimp.
1094 		 */
1095 		error = (p->if_transmit)(p, m);
1096 		if (!error)
1097 			ifp->if_opackets++;
1098 		else
1099 			ifp->if_oerrors++;
1100 	}
1101 }
1102 
1103 static void
1104 vlan_input(struct ifnet *ifp, struct mbuf *m)
1105 {
1106 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1107 	struct ifvlan *ifv;
1108 	uint16_t tag;
1109 
1110 	KASSERT(trunk != NULL, ("%s: no trunk", __func__));
1111 
1112 	if (m->m_flags & M_VLANTAG) {
1113 		/*
1114 		 * Packet is tagged, but m contains a normal
1115 		 * Ethernet frame; the tag is stored out-of-band.
1116 		 */
1117 		tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
1118 		m->m_flags &= ~M_VLANTAG;
1119 	} else {
1120 		struct ether_vlan_header *evl;
1121 
1122 		/*
1123 		 * Packet is tagged in-band as specified by 802.1q.
1124 		 */
1125 		switch (ifp->if_type) {
1126 		case IFT_ETHER:
1127 			if (m->m_len < sizeof(*evl) &&
1128 			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
1129 				if_printf(ifp, "cannot pullup VLAN header\n");
1130 				return;
1131 			}
1132 			evl = mtod(m, struct ether_vlan_header *);
1133 			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1134 
1135 			/*
1136 			 * Remove the 802.1q header by copying the Ethernet
1137 			 * addresses over it and adjusting the beginning of
1138 			 * the data in the mbuf.  The encapsulated Ethernet
1139 			 * type field is already in place.
1140 			 */
1141 			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1142 			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
1143 			m_adj(m, ETHER_VLAN_ENCAP_LEN);
1144 			break;
1145 
1146 		default:
1147 #ifdef INVARIANTS
1148 			panic("%s: %s has unsupported if_type %u",
1149 			      __func__, ifp->if_xname, ifp->if_type);
1150 #endif
1151 			m_freem(m);
1152 			ifp->if_noproto++;
1153 			return;
1154 		}
1155 	}
1156 
1157 	TRUNK_RLOCK(trunk);
1158 	ifv = vlan_gethash(trunk, tag);
1159 	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1160 		TRUNK_RUNLOCK(trunk);
1161 		m_freem(m);
1162 		ifp->if_noproto++;
1163 		return;
1164 	}
1165 	TRUNK_RUNLOCK(trunk);
1166 
1167 	m->m_pkthdr.rcvif = ifv->ifv_ifp;
1168 	ifv->ifv_ifp->if_ipackets++;
1169 
1170 	/* Pass it back through the parent's input routine. */
1171 	(*ifp->if_input)(ifv->ifv_ifp, m);
1172 }
1173 
1174 static int
1175 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
1176 {
1177 	struct ifvlantrunk *trunk;
1178 	struct ifnet *ifp;
1179 	int error = 0;
1180 
1181 	/* VID numbers 0x0 and 0xFFF are reserved */
1182 	if (tag == 0 || tag == 0xFFF)
1183 		return (EINVAL);
1184 	if (p->if_type != IFT_ETHER &&
1185 	    (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1186 		return (EPROTONOSUPPORT);
1187 	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1188 		return (EPROTONOSUPPORT);
1189 	if (ifv->ifv_trunk)
1190 		return (EBUSY);
1191 
1192 	if (p->if_vlantrunk == NULL) {
1193 		trunk = malloc(sizeof(struct ifvlantrunk),
1194 		    M_VLAN, M_WAITOK | M_ZERO);
1195 		vlan_inithash(trunk);
1196 		VLAN_LOCK();
1197 		if (p->if_vlantrunk != NULL) {
1198 			/* A race that that is very unlikely to be hit. */
1199 			vlan_freehash(trunk);
1200 			free(trunk, M_VLAN);
1201 			goto exists;
1202 		}
1203 		TRUNK_LOCK_INIT(trunk);
1204 		TRUNK_LOCK(trunk);
1205 		p->if_vlantrunk = trunk;
1206 		trunk->parent = p;
1207 	} else {
1208 		VLAN_LOCK();
1209 exists:
1210 		trunk = p->if_vlantrunk;
1211 		TRUNK_LOCK(trunk);
1212 	}
1213 
1214 	ifv->ifv_tag = tag;	/* must set this before vlan_inshash() */
1215 	error = vlan_inshash(trunk, ifv);
1216 	if (error)
1217 		goto done;
1218 	ifv->ifv_proto = ETHERTYPE_VLAN;
1219 	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1220 	ifv->ifv_mintu = ETHERMIN;
1221 	ifv->ifv_pflags = 0;
1222 
1223 	/*
1224 	 * If the parent supports the VLAN_MTU capability,
1225 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1226 	 * use it.
1227 	 */
1228 	if (p->if_capenable & IFCAP_VLAN_MTU) {
1229 		/*
1230 		 * No need to fudge the MTU since the parent can
1231 		 * handle extended frames.
1232 		 */
1233 		ifv->ifv_mtufudge = 0;
1234 	} else {
1235 		/*
1236 		 * Fudge the MTU by the encapsulation size.  This
1237 		 * makes us incompatible with strictly compliant
1238 		 * 802.1Q implementations, but allows us to use
1239 		 * the feature with other NetBSD implementations,
1240 		 * which might still be useful.
1241 		 */
1242 		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1243 	}
1244 
1245 	ifv->ifv_trunk = trunk;
1246 	ifp = ifv->ifv_ifp;
1247 	/*
1248 	 * Initialize fields from our parent.  This duplicates some
1249 	 * work with ether_ifattach() but allows for non-ethernet
1250 	 * interfaces to also work.
1251 	 */
1252 	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1253 	ifp->if_baudrate = p->if_baudrate;
1254 	ifp->if_output = p->if_output;
1255 	ifp->if_input = p->if_input;
1256 	ifp->if_resolvemulti = p->if_resolvemulti;
1257 	ifp->if_addrlen = p->if_addrlen;
1258 	ifp->if_broadcastaddr = p->if_broadcastaddr;
1259 
1260 	/*
1261 	 * Copy only a selected subset of flags from the parent.
1262 	 * Other flags are none of our business.
1263 	 */
1264 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1265 	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1266 	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1267 #undef VLAN_COPY_FLAGS
1268 
1269 	ifp->if_link_state = p->if_link_state;
1270 
1271 	vlan_capabilities(ifv);
1272 
1273 	/*
1274 	 * Set up our interface address to reflect the underlying
1275 	 * physical interface's.
1276 	 */
1277 	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1278 	((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1279 	    p->if_addrlen;
1280 
1281 	/*
1282 	 * Configure multicast addresses that may already be
1283 	 * joined on the vlan device.
1284 	 */
1285 	(void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1286 
1287 	/* We are ready for operation now. */
1288 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1289 done:
1290 	TRUNK_UNLOCK(trunk);
1291 	if (error == 0)
1292 		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_tag);
1293 	VLAN_UNLOCK();
1294 
1295 	return (error);
1296 }
1297 
1298 static void
1299 vlan_unconfig(struct ifnet *ifp)
1300 {
1301 
1302 	VLAN_LOCK();
1303 	vlan_unconfig_locked(ifp);
1304 	VLAN_UNLOCK();
1305 }
1306 
1307 static void
1308 vlan_unconfig_locked(struct ifnet *ifp)
1309 {
1310 	struct ifvlantrunk *trunk;
1311 	struct vlan_mc_entry *mc;
1312 	struct ifvlan *ifv;
1313 	struct ifnet  *parent;
1314 
1315 	VLAN_LOCK_ASSERT();
1316 
1317 	ifv = ifp->if_softc;
1318 	trunk = ifv->ifv_trunk;
1319 	parent = NULL;
1320 
1321 	if (trunk != NULL) {
1322 
1323 		TRUNK_LOCK(trunk);
1324 		parent = trunk->parent;
1325 
1326 		/*
1327 		 * Since the interface is being unconfigured, we need to
1328 		 * empty the list of multicast groups that we may have joined
1329 		 * while we were alive from the parent's list.
1330 		 */
1331 		while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1332 			/*
1333 			 * This may fail if the parent interface is
1334 			 * being detached.  Regardless, we should do a
1335 			 * best effort to free this interface as much
1336 			 * as possible as all callers expect vlan
1337 			 * destruction to succeed.
1338 			 */
1339 			(void)if_delmulti(parent,
1340 			    (struct sockaddr *)&mc->mc_addr);
1341 			SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1342 			free(mc, M_VLAN);
1343 		}
1344 
1345 		vlan_setflags(ifp, 0); /* clear special flags on parent */
1346 		vlan_remhash(trunk, ifv);
1347 		ifv->ifv_trunk = NULL;
1348 
1349 		/*
1350 		 * Check if we were the last.
1351 		 */
1352 		if (trunk->refcnt == 0) {
1353 			trunk->parent->if_vlantrunk = NULL;
1354 			/*
1355 			 * XXXGL: If some ithread has already entered
1356 			 * vlan_input() and is now blocked on the trunk
1357 			 * lock, then it should preempt us right after
1358 			 * unlock and finish its work. Then we will acquire
1359 			 * lock again in trunk_destroy().
1360 			 */
1361 			TRUNK_UNLOCK(trunk);
1362 			trunk_destroy(trunk);
1363 		} else
1364 			TRUNK_UNLOCK(trunk);
1365 	}
1366 
1367 	/* Disconnect from parent. */
1368 	if (ifv->ifv_pflags)
1369 		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1370 	ifp->if_mtu = ETHERMTU;
1371 	ifp->if_link_state = LINK_STATE_UNKNOWN;
1372 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1373 
1374 	/*
1375 	 * Only dispatch an event if vlan was
1376 	 * attached, otherwise there is nothing
1377 	 * to cleanup anyway.
1378 	 */
1379 	if (parent != NULL)
1380 		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_tag);
1381 }
1382 
1383 /* Handle a reference counted flag that should be set on the parent as well */
1384 static int
1385 vlan_setflag(struct ifnet *ifp, int flag, int status,
1386 	     int (*func)(struct ifnet *, int))
1387 {
1388 	struct ifvlan *ifv;
1389 	int error;
1390 
1391 	/* XXX VLAN_LOCK_ASSERT(); */
1392 
1393 	ifv = ifp->if_softc;
1394 	status = status ? (ifp->if_flags & flag) : 0;
1395 	/* Now "status" contains the flag value or 0 */
1396 
1397 	/*
1398 	 * See if recorded parent's status is different from what
1399 	 * we want it to be.  If it is, flip it.  We record parent's
1400 	 * status in ifv_pflags so that we won't clear parent's flag
1401 	 * we haven't set.  In fact, we don't clear or set parent's
1402 	 * flags directly, but get or release references to them.
1403 	 * That's why we can be sure that recorded flags still are
1404 	 * in accord with actual parent's flags.
1405 	 */
1406 	if (status != (ifv->ifv_pflags & flag)) {
1407 		error = (*func)(PARENT(ifv), status);
1408 		if (error)
1409 			return (error);
1410 		ifv->ifv_pflags &= ~flag;
1411 		ifv->ifv_pflags |= status;
1412 	}
1413 	return (0);
1414 }
1415 
1416 /*
1417  * Handle IFF_* flags that require certain changes on the parent:
1418  * if "status" is true, update parent's flags respective to our if_flags;
1419  * if "status" is false, forcedly clear the flags set on parent.
1420  */
1421 static int
1422 vlan_setflags(struct ifnet *ifp, int status)
1423 {
1424 	int error, i;
1425 
1426 	for (i = 0; vlan_pflags[i].flag; i++) {
1427 		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1428 				     status, vlan_pflags[i].func);
1429 		if (error)
1430 			return (error);
1431 	}
1432 	return (0);
1433 }
1434 
1435 /* Inform all vlans that their parent has changed link state */
1436 static void
1437 vlan_link_state(struct ifnet *ifp)
1438 {
1439 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1440 	struct ifvlan *ifv;
1441 	int i;
1442 
1443 	TRUNK_LOCK(trunk);
1444 #ifdef VLAN_ARRAY
1445 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1446 		if (trunk->vlans[i] != NULL) {
1447 			ifv = trunk->vlans[i];
1448 #else
1449 	for (i = 0; i < (1 << trunk->hwidth); i++)
1450 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1451 #endif
1452 			ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1453 			if_link_state_change(ifv->ifv_ifp,
1454 			    trunk->parent->if_link_state);
1455 		}
1456 	TRUNK_UNLOCK(trunk);
1457 }
1458 
1459 static void
1460 vlan_capabilities(struct ifvlan *ifv)
1461 {
1462 	struct ifnet *p = PARENT(ifv);
1463 	struct ifnet *ifp = ifv->ifv_ifp;
1464 
1465 	TRUNK_LOCK_ASSERT(TRUNK(ifv));
1466 
1467 	/*
1468 	 * If the parent interface can do checksum offloading
1469 	 * on VLANs, then propagate its hardware-assisted
1470 	 * checksumming flags. Also assert that checksum
1471 	 * offloading requires hardware VLAN tagging.
1472 	 */
1473 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1474 		ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1475 
1476 	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1477 	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1478 		ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1479 		ifp->if_hwassist = p->if_hwassist & (CSUM_IP | CSUM_TCP |
1480 		    CSUM_UDP | CSUM_SCTP | CSUM_IP_FRAGS | CSUM_FRAGMENT);
1481 	} else {
1482 		ifp->if_capenable = 0;
1483 		ifp->if_hwassist = 0;
1484 	}
1485 	/*
1486 	 * If the parent interface can do TSO on VLANs then
1487 	 * propagate the hardware-assisted flag. TSO on VLANs
1488 	 * does not necessarily require hardware VLAN tagging.
1489 	 */
1490 	if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1491 		ifp->if_capabilities |= p->if_capabilities & IFCAP_TSO;
1492 	if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1493 		ifp->if_capenable |= p->if_capenable & IFCAP_TSO;
1494 		ifp->if_hwassist |= p->if_hwassist & CSUM_TSO;
1495 	} else {
1496 		ifp->if_capenable &= ~(p->if_capenable & IFCAP_TSO);
1497 		ifp->if_hwassist &= ~(p->if_hwassist & CSUM_TSO);
1498 	}
1499 }
1500 
1501 static void
1502 vlan_trunk_capabilities(struct ifnet *ifp)
1503 {
1504 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1505 	struct ifvlan *ifv;
1506 	int i;
1507 
1508 	TRUNK_LOCK(trunk);
1509 #ifdef VLAN_ARRAY
1510 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1511 		if (trunk->vlans[i] != NULL) {
1512 			ifv = trunk->vlans[i];
1513 #else
1514 	for (i = 0; i < (1 << trunk->hwidth); i++) {
1515 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1516 #endif
1517 			vlan_capabilities(ifv);
1518 	}
1519 	TRUNK_UNLOCK(trunk);
1520 }
1521 
1522 static int
1523 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1524 {
1525 	struct ifnet *p;
1526 	struct ifreq *ifr;
1527 	struct ifaddr *ifa;
1528 	struct ifvlan *ifv;
1529 	struct vlanreq vlr;
1530 	int error = 0;
1531 
1532 	ifr = (struct ifreq *)data;
1533 	ifa = (struct ifaddr *) data;
1534 	ifv = ifp->if_softc;
1535 
1536 	switch (cmd) {
1537 	case SIOCSIFADDR:
1538 		ifp->if_flags |= IFF_UP;
1539 #ifdef INET
1540 		if (ifa->ifa_addr->sa_family == AF_INET)
1541 			arp_ifinit(ifp, ifa);
1542 #endif
1543 		break;
1544 	case SIOCGIFADDR:
1545                 {
1546 			struct sockaddr *sa;
1547 
1548 			sa = (struct sockaddr *)&ifr->ifr_data;
1549 			bcopy(IF_LLADDR(ifp), sa->sa_data, ifp->if_addrlen);
1550                 }
1551 		break;
1552 	case SIOCGIFMEDIA:
1553 		VLAN_LOCK();
1554 		if (TRUNK(ifv) != NULL) {
1555 			p = PARENT(ifv);
1556 			VLAN_UNLOCK();
1557 			error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1558 			/* Limit the result to the parent's current config. */
1559 			if (error == 0) {
1560 				struct ifmediareq *ifmr;
1561 
1562 				ifmr = (struct ifmediareq *)data;
1563 				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1564 					ifmr->ifm_count = 1;
1565 					error = copyout(&ifmr->ifm_current,
1566 						ifmr->ifm_ulist,
1567 						sizeof(int));
1568 				}
1569 			}
1570 		} else {
1571 			VLAN_UNLOCK();
1572 			error = EINVAL;
1573 		}
1574 		break;
1575 
1576 	case SIOCSIFMEDIA:
1577 		error = EINVAL;
1578 		break;
1579 
1580 	case SIOCSIFMTU:
1581 		/*
1582 		 * Set the interface MTU.
1583 		 */
1584 		VLAN_LOCK();
1585 		if (TRUNK(ifv) != NULL) {
1586 			if (ifr->ifr_mtu >
1587 			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1588 			    ifr->ifr_mtu <
1589 			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1590 				error = EINVAL;
1591 			else
1592 				ifp->if_mtu = ifr->ifr_mtu;
1593 		} else
1594 			error = EINVAL;
1595 		VLAN_UNLOCK();
1596 		break;
1597 
1598 	case SIOCSETVLAN:
1599 #ifdef VIMAGE
1600 		if (ifp->if_vnet != ifp->if_home_vnet) {
1601 			error = EPERM;
1602 			break;
1603 		}
1604 #endif
1605 		error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1606 		if (error)
1607 			break;
1608 		if (vlr.vlr_parent[0] == '\0') {
1609 			vlan_unconfig(ifp);
1610 			break;
1611 		}
1612 		p = ifunit(vlr.vlr_parent);
1613 		if (p == NULL) {
1614 			error = ENOENT;
1615 			break;
1616 		}
1617 		/*
1618 		 * Don't let the caller set up a VLAN tag with
1619 		 * anything except VLID bits.
1620 		 */
1621 		if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1622 			error = EINVAL;
1623 			break;
1624 		}
1625 		error = vlan_config(ifv, p, vlr.vlr_tag);
1626 		if (error)
1627 			break;
1628 
1629 		/* Update flags on the parent, if necessary. */
1630 		vlan_setflags(ifp, 1);
1631 		break;
1632 
1633 	case SIOCGETVLAN:
1634 #ifdef VIMAGE
1635 		if (ifp->if_vnet != ifp->if_home_vnet) {
1636 			error = EPERM;
1637 			break;
1638 		}
1639 #endif
1640 		bzero(&vlr, sizeof(vlr));
1641 		VLAN_LOCK();
1642 		if (TRUNK(ifv) != NULL) {
1643 			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1644 			    sizeof(vlr.vlr_parent));
1645 			vlr.vlr_tag = ifv->ifv_tag;
1646 		}
1647 		VLAN_UNLOCK();
1648 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1649 		break;
1650 
1651 	case SIOCSIFFLAGS:
1652 		/*
1653 		 * We should propagate selected flags to the parent,
1654 		 * e.g., promiscuous mode.
1655 		 */
1656 		if (TRUNK(ifv) != NULL)
1657 			error = vlan_setflags(ifp, 1);
1658 		break;
1659 
1660 	case SIOCADDMULTI:
1661 	case SIOCDELMULTI:
1662 		/*
1663 		 * If we don't have a parent, just remember the membership for
1664 		 * when we do.
1665 		 */
1666 		if (TRUNK(ifv) != NULL)
1667 			error = vlan_setmulti(ifp);
1668 		break;
1669 
1670 	default:
1671 		error = EINVAL;
1672 		break;
1673 	}
1674 
1675 	return (error);
1676 }
1677