xref: /freebsd/sys/net/if_vlan.c (revision 4ed925457ab06e83238a5db33e89ccc94b99a713)
1 /*-
2  * Copyright 1998 Massachusetts Institute of Technology
3  *
4  * Permission to use, copy, modify, and distribute this software and
5  * its documentation for any purpose and without fee is hereby
6  * granted, provided that both the above copyright notice and this
7  * permission notice appear in all copies, that both the above
8  * copyright notice and this permission notice appear in all
9  * supporting documentation, and that the name of M.I.T. not be used
10  * in advertising or publicity pertaining to distribution of the
11  * software without specific, written prior permission.  M.I.T. makes
12  * no representations about the suitability of this software for any
13  * purpose.  It is provided "as is" without express or implied
14  * warranty.
15  *
16  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
17  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
18  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
19  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
20  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
23  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
24  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
25  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
26  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * $FreeBSD$
30  */
31 
32 /*
33  * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
34  * Might be extended some day to also handle IEEE 802.1p priority
35  * tagging.  This is sort of sneaky in the implementation, since
36  * we need to pretend to be enough of an Ethernet implementation
37  * to make arp work.  The way we do this is by telling everyone
38  * that we are an Ethernet, and then catch the packets that
39  * ether_output() left on our output queue when it calls
40  * if_start(), rewrite them for use by the real outgoing interface,
41  * and ask it to send them.
42  */
43 
44 #include "opt_vlan.h"
45 
46 #include <sys/param.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/module.h>
52 #include <sys/rwlock.h>
53 #include <sys/queue.h>
54 #include <sys/socket.h>
55 #include <sys/sockio.h>
56 #include <sys/sysctl.h>
57 #include <sys/systm.h>
58 
59 #include <net/bpf.h>
60 #include <net/ethernet.h>
61 #include <net/if.h>
62 #include <net/if_clone.h>
63 #include <net/if_dl.h>
64 #include <net/if_types.h>
65 #include <net/if_vlan_var.h>
66 #include <net/vnet.h>
67 
68 #define VLANNAME	"vlan"
69 #define	VLAN_DEF_HWIDTH	4
70 #define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
71 
72 #define	UP_AND_RUNNING(ifp) \
73     ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
74 
75 LIST_HEAD(ifvlanhead, ifvlan);
76 
77 struct ifvlantrunk {
78 	struct	ifnet   *parent;	/* parent interface of this trunk */
79 	struct	rwlock	rw;
80 #ifdef VLAN_ARRAY
81 #define	VLAN_ARRAY_SIZE	(EVL_VLID_MASK + 1)
82 	struct	ifvlan	*vlans[VLAN_ARRAY_SIZE]; /* static table */
83 #else
84 	struct	ifvlanhead *hash;	/* dynamic hash-list table */
85 	uint16_t	hmask;
86 	uint16_t	hwidth;
87 #endif
88 	int		refcnt;
89 };
90 
91 struct vlan_mc_entry {
92 	struct ether_addr		mc_addr;
93 	SLIST_ENTRY(vlan_mc_entry)	mc_entries;
94 };
95 
96 struct	ifvlan {
97 	struct	ifvlantrunk *ifv_trunk;
98 	struct	ifnet *ifv_ifp;
99 #define	TRUNK(ifv)	((ifv)->ifv_trunk)
100 #define	PARENT(ifv)	((ifv)->ifv_trunk->parent)
101 	int	ifv_pflags;	/* special flags we have set on parent */
102 	struct	ifv_linkmib {
103 		int	ifvm_encaplen;	/* encapsulation length */
104 		int	ifvm_mtufudge;	/* MTU fudged by this much */
105 		int	ifvm_mintu;	/* min transmission unit */
106 		uint16_t ifvm_proto;	/* encapsulation ethertype */
107 		uint16_t ifvm_tag;	/* tag to apply on packets leaving if */
108 	}	ifv_mib;
109 	SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
110 #ifndef VLAN_ARRAY
111 	LIST_ENTRY(ifvlan) ifv_list;
112 #endif
113 };
114 #define	ifv_proto	ifv_mib.ifvm_proto
115 #define	ifv_tag		ifv_mib.ifvm_tag
116 #define	ifv_encaplen	ifv_mib.ifvm_encaplen
117 #define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
118 #define	ifv_mintu	ifv_mib.ifvm_mintu
119 
120 /* Special flags we should propagate to parent. */
121 static struct {
122 	int flag;
123 	int (*func)(struct ifnet *, int);
124 } vlan_pflags[] = {
125 	{IFF_PROMISC, ifpromisc},
126 	{IFF_ALLMULTI, if_allmulti},
127 	{0, NULL}
128 };
129 
130 SYSCTL_DECL(_net_link);
131 SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0, "IEEE 802.1Q VLAN");
132 SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0, "for consistency");
133 
134 static int soft_pad = 0;
135 SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW, &soft_pad, 0,
136 	   "pad short frames before tagging");
137 
138 static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface");
139 
140 static eventhandler_tag ifdetach_tag;
141 static eventhandler_tag iflladdr_tag;
142 
143 /*
144  * We have a global mutex, that is used to serialize configuration
145  * changes and isn't used in normal packet delivery.
146  *
147  * We also have a per-trunk rwlock, that is locked shared on packet
148  * processing and exclusive when configuration is changed.
149  *
150  * The VLAN_ARRAY substitutes the dynamic hash with a static array
151  * with 4096 entries. In theory this can give a boost in processing,
152  * however on practice it does not. Probably this is because array
153  * is too big to fit into CPU cache.
154  */
155 static struct mtx ifv_mtx;
156 #define	VLAN_LOCK_INIT()	mtx_init(&ifv_mtx, "vlan_global", NULL, MTX_DEF)
157 #define	VLAN_LOCK_DESTROY()	mtx_destroy(&ifv_mtx)
158 #define	VLAN_LOCK_ASSERT()	mtx_assert(&ifv_mtx, MA_OWNED)
159 #define	VLAN_LOCK()		mtx_lock(&ifv_mtx)
160 #define	VLAN_UNLOCK()		mtx_unlock(&ifv_mtx)
161 #define	TRUNK_LOCK_INIT(trunk)	rw_init(&(trunk)->rw, VLANNAME)
162 #define	TRUNK_LOCK_DESTROY(trunk) rw_destroy(&(trunk)->rw)
163 #define	TRUNK_LOCK(trunk)	rw_wlock(&(trunk)->rw)
164 #define	TRUNK_UNLOCK(trunk)	rw_wunlock(&(trunk)->rw)
165 #define	TRUNK_LOCK_ASSERT(trunk) rw_assert(&(trunk)->rw, RA_WLOCKED)
166 #define	TRUNK_RLOCK(trunk)	rw_rlock(&(trunk)->rw)
167 #define	TRUNK_RUNLOCK(trunk)	rw_runlock(&(trunk)->rw)
168 #define	TRUNK_LOCK_RASSERT(trunk) rw_assert(&(trunk)->rw, RA_RLOCKED)
169 
170 #ifndef VLAN_ARRAY
171 static	void vlan_inithash(struct ifvlantrunk *trunk);
172 static	void vlan_freehash(struct ifvlantrunk *trunk);
173 static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
174 static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
175 static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
176 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
177 	uint16_t tag);
178 #endif
179 static	void trunk_destroy(struct ifvlantrunk *trunk);
180 
181 static	void vlan_start(struct ifnet *ifp);
182 static	void vlan_init(void *foo);
183 static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
184 static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
185 static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
186     int (*func)(struct ifnet *, int));
187 static	int vlan_setflags(struct ifnet *ifp, int status);
188 static	int vlan_setmulti(struct ifnet *ifp);
189 static	int vlan_unconfig(struct ifnet *ifp);
190 static	int vlan_unconfig_locked(struct ifnet *ifp);
191 static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
192 static	void vlan_link_state(struct ifnet *ifp);
193 static	void vlan_capabilities(struct ifvlan *ifv);
194 static	void vlan_trunk_capabilities(struct ifnet *ifp);
195 
196 static	struct ifnet *vlan_clone_match_ethertag(struct if_clone *,
197     const char *, int *);
198 static	int vlan_clone_match(struct if_clone *, const char *);
199 static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
200 static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
201 
202 static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
203 static  void vlan_iflladdr(void *arg, struct ifnet *ifp);
204 
205 static	struct if_clone vlan_cloner = IFC_CLONE_INITIALIZER(VLANNAME, NULL,
206     IF_MAXUNIT, NULL, vlan_clone_match, vlan_clone_create, vlan_clone_destroy);
207 
208 #ifndef VLAN_ARRAY
209 #define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
210 
211 static void
212 vlan_inithash(struct ifvlantrunk *trunk)
213 {
214 	int i, n;
215 
216 	/*
217 	 * The trunk must not be locked here since we call malloc(M_WAITOK).
218 	 * It is OK in case this function is called before the trunk struct
219 	 * gets hooked up and becomes visible from other threads.
220 	 */
221 
222 	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
223 	    ("%s: hash already initialized", __func__));
224 
225 	trunk->hwidth = VLAN_DEF_HWIDTH;
226 	n = 1 << trunk->hwidth;
227 	trunk->hmask = n - 1;
228 	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
229 	for (i = 0; i < n; i++)
230 		LIST_INIT(&trunk->hash[i]);
231 }
232 
233 static void
234 vlan_freehash(struct ifvlantrunk *trunk)
235 {
236 #ifdef INVARIANTS
237 	int i;
238 
239 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
240 	for (i = 0; i < (1 << trunk->hwidth); i++)
241 		KASSERT(LIST_EMPTY(&trunk->hash[i]),
242 		    ("%s: hash table not empty", __func__));
243 #endif
244 	free(trunk->hash, M_VLAN);
245 	trunk->hash = NULL;
246 	trunk->hwidth = trunk->hmask = 0;
247 }
248 
249 static int
250 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
251 {
252 	int i, b;
253 	struct ifvlan *ifv2;
254 
255 	TRUNK_LOCK_ASSERT(trunk);
256 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
257 
258 	b = 1 << trunk->hwidth;
259 	i = HASH(ifv->ifv_tag, trunk->hmask);
260 	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
261 		if (ifv->ifv_tag == ifv2->ifv_tag)
262 			return (EEXIST);
263 
264 	/*
265 	 * Grow the hash when the number of vlans exceeds half of the number of
266 	 * hash buckets squared. This will make the average linked-list length
267 	 * buckets/2.
268 	 */
269 	if (trunk->refcnt > (b * b) / 2) {
270 		vlan_growhash(trunk, 1);
271 		i = HASH(ifv->ifv_tag, trunk->hmask);
272 	}
273 	LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
274 	trunk->refcnt++;
275 
276 	return (0);
277 }
278 
279 static int
280 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
281 {
282 	int i, b;
283 	struct ifvlan *ifv2;
284 
285 	TRUNK_LOCK_ASSERT(trunk);
286 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
287 
288 	b = 1 << trunk->hwidth;
289 	i = HASH(ifv->ifv_tag, trunk->hmask);
290 	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
291 		if (ifv2 == ifv) {
292 			trunk->refcnt--;
293 			LIST_REMOVE(ifv2, ifv_list);
294 			if (trunk->refcnt < (b * b) / 2)
295 				vlan_growhash(trunk, -1);
296 			return (0);
297 		}
298 
299 	panic("%s: vlan not found\n", __func__);
300 	return (ENOENT); /*NOTREACHED*/
301 }
302 
303 /*
304  * Grow the hash larger or smaller if memory permits.
305  */
306 static void
307 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
308 {
309 	struct ifvlan *ifv;
310 	struct ifvlanhead *hash2;
311 	int hwidth2, i, j, n, n2;
312 
313 	TRUNK_LOCK_ASSERT(trunk);
314 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
315 
316 	if (howmuch == 0) {
317 		/* Harmless yet obvious coding error */
318 		printf("%s: howmuch is 0\n", __func__);
319 		return;
320 	}
321 
322 	hwidth2 = trunk->hwidth + howmuch;
323 	n = 1 << trunk->hwidth;
324 	n2 = 1 << hwidth2;
325 	/* Do not shrink the table below the default */
326 	if (hwidth2 < VLAN_DEF_HWIDTH)
327 		return;
328 
329 	/* M_NOWAIT because we're called with trunk mutex held */
330 	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
331 	if (hash2 == NULL) {
332 		printf("%s: out of memory -- hash size not changed\n",
333 		    __func__);
334 		return;		/* We can live with the old hash table */
335 	}
336 	for (j = 0; j < n2; j++)
337 		LIST_INIT(&hash2[j]);
338 	for (i = 0; i < n; i++)
339 		while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
340 			LIST_REMOVE(ifv, ifv_list);
341 			j = HASH(ifv->ifv_tag, n2 - 1);
342 			LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
343 		}
344 	free(trunk->hash, M_VLAN);
345 	trunk->hash = hash2;
346 	trunk->hwidth = hwidth2;
347 	trunk->hmask = n2 - 1;
348 
349 	if (bootverbose)
350 		if_printf(trunk->parent,
351 		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
352 }
353 
354 static __inline struct ifvlan *
355 vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
356 {
357 	struct ifvlan *ifv;
358 
359 	TRUNK_LOCK_RASSERT(trunk);
360 
361 	LIST_FOREACH(ifv, &trunk->hash[HASH(tag, trunk->hmask)], ifv_list)
362 		if (ifv->ifv_tag == tag)
363 			return (ifv);
364 	return (NULL);
365 }
366 
367 #if 0
368 /* Debugging code to view the hashtables. */
369 static void
370 vlan_dumphash(struct ifvlantrunk *trunk)
371 {
372 	int i;
373 	struct ifvlan *ifv;
374 
375 	for (i = 0; i < (1 << trunk->hwidth); i++) {
376 		printf("%d: ", i);
377 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
378 			printf("%s ", ifv->ifv_ifp->if_xname);
379 		printf("\n");
380 	}
381 }
382 #endif /* 0 */
383 #endif /* !VLAN_ARRAY */
384 
385 static void
386 trunk_destroy(struct ifvlantrunk *trunk)
387 {
388 	VLAN_LOCK_ASSERT();
389 
390 	TRUNK_LOCK(trunk);
391 #ifndef VLAN_ARRAY
392 	vlan_freehash(trunk);
393 #endif
394 	trunk->parent->if_vlantrunk = NULL;
395 	TRUNK_UNLOCK(trunk);
396 	TRUNK_LOCK_DESTROY(trunk);
397 	free(trunk, M_VLAN);
398 }
399 
400 /*
401  * Program our multicast filter. What we're actually doing is
402  * programming the multicast filter of the parent. This has the
403  * side effect of causing the parent interface to receive multicast
404  * traffic that it doesn't really want, which ends up being discarded
405  * later by the upper protocol layers. Unfortunately, there's no way
406  * to avoid this: there really is only one physical interface.
407  *
408  * XXX: There is a possible race here if more than one thread is
409  *      modifying the multicast state of the vlan interface at the same time.
410  */
411 static int
412 vlan_setmulti(struct ifnet *ifp)
413 {
414 	struct ifnet		*ifp_p;
415 	struct ifmultiaddr	*ifma, *rifma = NULL;
416 	struct ifvlan		*sc;
417 	struct vlan_mc_entry	*mc;
418 	struct sockaddr_dl	sdl;
419 	int			error;
420 
421 	/*VLAN_LOCK_ASSERT();*/
422 
423 	/* Find the parent. */
424 	sc = ifp->if_softc;
425 	ifp_p = PARENT(sc);
426 
427 	CURVNET_SET_QUIET(ifp_p->if_vnet);
428 
429 	bzero((char *)&sdl, sizeof(sdl));
430 	sdl.sdl_len = sizeof(sdl);
431 	sdl.sdl_family = AF_LINK;
432 	sdl.sdl_index = ifp_p->if_index;
433 	sdl.sdl_type = IFT_ETHER;
434 	sdl.sdl_alen = ETHER_ADDR_LEN;
435 
436 	/* First, remove any existing filter entries. */
437 	while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
438 		bcopy((char *)&mc->mc_addr, LLADDR(&sdl), ETHER_ADDR_LEN);
439 		error = if_delmulti(ifp_p, (struct sockaddr *)&sdl);
440 		if (error)
441 			return (error);
442 		SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
443 		free(mc, M_VLAN);
444 	}
445 
446 	/* Now program new ones. */
447 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
448 		if (ifma->ifma_addr->sa_family != AF_LINK)
449 			continue;
450 		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
451 		if (mc == NULL)
452 			return (ENOMEM);
453 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
454 		    (char *)&mc->mc_addr, ETHER_ADDR_LEN);
455 		SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
456 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
457 		    LLADDR(&sdl), ETHER_ADDR_LEN);
458 		error = if_addmulti(ifp_p, (struct sockaddr *)&sdl, &rifma);
459 		if (error)
460 			return (error);
461 	}
462 
463 	CURVNET_RESTORE();
464 	return (0);
465 }
466 
467 /*
468  * A handler for parent interface link layer address changes.
469  * If the parent interface link layer address is changed we
470  * should also change it on all children vlans.
471  */
472 static void
473 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
474 {
475 	struct ifvlan *ifv;
476 #ifndef VLAN_ARRAY
477 	struct ifvlan *next;
478 #endif
479 	int i;
480 
481 	/*
482 	 * Check if it's a trunk interface first of all
483 	 * to avoid needless locking.
484 	 */
485 	if (ifp->if_vlantrunk == NULL)
486 		return;
487 
488 	VLAN_LOCK();
489 	/*
490 	 * OK, it's a trunk.  Loop over and change all vlan's lladdrs on it.
491 	 */
492 #ifdef VLAN_ARRAY
493 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
494 		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
495 #else /* VLAN_ARRAY */
496 	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
497 		LIST_FOREACH_SAFE(ifv, &ifp->if_vlantrunk->hash[i], ifv_list, next) {
498 #endif /* VLAN_ARRAY */
499 			VLAN_UNLOCK();
500 			if_setlladdr(ifv->ifv_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN);
501 			VLAN_LOCK();
502 		}
503 	VLAN_UNLOCK();
504 
505 }
506 
507 /*
508  * A handler for network interface departure events.
509  * Track departure of trunks here so that we don't access invalid
510  * pointers or whatever if a trunk is ripped from under us, e.g.,
511  * by ejecting its hot-plug card.  However, if an ifnet is simply
512  * being renamed, then there's no need to tear down the state.
513  */
514 static void
515 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
516 {
517 	struct ifvlan *ifv;
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 	/* If the ifnet is just being renamed, don't do anything. */
528 	if (ifp->if_flags & IFF_RENAMING)
529 		return;
530 
531 	VLAN_LOCK();
532 	/*
533 	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
534 	 * Check trunk pointer after each vlan_unconfig() as it will
535 	 * free it and set to NULL after the last vlan was detached.
536 	 */
537 #ifdef VLAN_ARRAY
538 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
539 		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
540 			vlan_unconfig_locked(ifv->ifv_ifp);
541 			if (ifp->if_vlantrunk == NULL)
542 				break;
543 		}
544 #else /* VLAN_ARRAY */
545 restart:
546 	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
547 		if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
548 			vlan_unconfig_locked(ifv->ifv_ifp);
549 			if (ifp->if_vlantrunk)
550 				goto restart;	/* trunk->hwidth can change */
551 			else
552 				break;
553 		}
554 #endif /* VLAN_ARRAY */
555 	/* Trunk should have been destroyed in vlan_unconfig(). */
556 	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
557 	VLAN_UNLOCK();
558 }
559 
560 /*
561  * VLAN support can be loaded as a module.  The only place in the
562  * system that's intimately aware of this is ether_input.  We hook
563  * into this code through vlan_input_p which is defined there and
564  * set here.  Noone else in the system should be aware of this so
565  * we use an explicit reference here.
566  */
567 extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
568 
569 /* For if_link_state_change() eyes only... */
570 extern	void (*vlan_link_state_p)(struct ifnet *);
571 
572 static int
573 vlan_modevent(module_t mod, int type, void *data)
574 {
575 
576 	switch (type) {
577 	case MOD_LOAD:
578 		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
579 		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
580 		if (ifdetach_tag == NULL)
581 			return (ENOMEM);
582 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
583 		    vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
584 		if (iflladdr_tag == NULL)
585 			return (ENOMEM);
586 		VLAN_LOCK_INIT();
587 		vlan_input_p = vlan_input;
588 		vlan_link_state_p = vlan_link_state;
589 		vlan_trunk_cap_p = vlan_trunk_capabilities;
590 		if_clone_attach(&vlan_cloner);
591 		if (bootverbose)
592 			printf("vlan: initialized, using "
593 #ifdef VLAN_ARRAY
594 			       "full-size arrays"
595 #else
596 			       "hash tables with chaining"
597 #endif
598 
599 			       "\n");
600 		break;
601 	case MOD_UNLOAD:
602 		if_clone_detach(&vlan_cloner);
603 		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
604 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
605 		vlan_input_p = NULL;
606 		vlan_link_state_p = NULL;
607 		vlan_trunk_cap_p = NULL;
608 		VLAN_LOCK_DESTROY();
609 		if (bootverbose)
610 			printf("vlan: unloaded\n");
611 		break;
612 	default:
613 		return (EOPNOTSUPP);
614 	}
615 	return (0);
616 }
617 
618 static moduledata_t vlan_mod = {
619 	"if_vlan",
620 	vlan_modevent,
621 	0
622 };
623 
624 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
625 MODULE_VERSION(if_vlan, 3);
626 
627 static struct ifnet *
628 vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
629 {
630 	const char *cp;
631 	struct ifnet *ifp;
632 	int t;
633 
634 	/* Check for <etherif>.<vlan> style interface names. */
635 	IFNET_RLOCK_NOSLEEP();
636 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
637 		if (ifp->if_type != IFT_ETHER)
638 			continue;
639 		if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
640 			continue;
641 		cp = name + strlen(ifp->if_xname);
642 		if (*cp++ != '.')
643 			continue;
644 		if (*cp == '\0')
645 			continue;
646 		t = 0;
647 		for(; *cp >= '0' && *cp <= '9'; cp++)
648 			t = (t * 10) + (*cp - '0');
649 		if (*cp != '\0')
650 			continue;
651 		if (tag != NULL)
652 			*tag = t;
653 		break;
654 	}
655 	IFNET_RUNLOCK_NOSLEEP();
656 
657 	return (ifp);
658 }
659 
660 static int
661 vlan_clone_match(struct if_clone *ifc, const char *name)
662 {
663 	const char *cp;
664 
665 	if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
666 		return (1);
667 
668 	if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
669 		return (0);
670 	for (cp = name + 4; *cp != '\0'; cp++) {
671 		if (*cp < '0' || *cp > '9')
672 			return (0);
673 	}
674 
675 	return (1);
676 }
677 
678 static int
679 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
680 {
681 	char *dp;
682 	int wildcard;
683 	int unit;
684 	int error;
685 	int tag;
686 	int ethertag;
687 	struct ifvlan *ifv;
688 	struct ifnet *ifp;
689 	struct ifnet *p;
690 	struct vlanreq vlr;
691 	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
692 
693 	/*
694 	 * There are 3 (ugh) ways to specify the cloned device:
695 	 * o pass a parameter block with the clone request.
696 	 * o specify parameters in the text of the clone device name
697 	 * o specify no parameters and get an unattached device that
698 	 *   must be configured separately.
699 	 * The first technique is preferred; the latter two are
700 	 * supported for backwards compatibilty.
701 	 */
702 	if (params) {
703 		error = copyin(params, &vlr, sizeof(vlr));
704 		if (error)
705 			return error;
706 		p = ifunit(vlr.vlr_parent);
707 		if (p == NULL)
708 			return ENXIO;
709 		/*
710 		 * Don't let the caller set up a VLAN tag with
711 		 * anything except VLID bits.
712 		 */
713 		if (vlr.vlr_tag & ~EVL_VLID_MASK)
714 			return (EINVAL);
715 		error = ifc_name2unit(name, &unit);
716 		if (error != 0)
717 			return (error);
718 
719 		ethertag = 1;
720 		tag = vlr.vlr_tag;
721 		wildcard = (unit < 0);
722 	} else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
723 		ethertag = 1;
724 		unit = -1;
725 		wildcard = 0;
726 
727 		/*
728 		 * Don't let the caller set up a VLAN tag with
729 		 * anything except VLID bits.
730 		 */
731 		if (tag & ~EVL_VLID_MASK)
732 			return (EINVAL);
733 	} else {
734 		ethertag = 0;
735 
736 		error = ifc_name2unit(name, &unit);
737 		if (error != 0)
738 			return (error);
739 
740 		wildcard = (unit < 0);
741 	}
742 
743 	error = ifc_alloc_unit(ifc, &unit);
744 	if (error != 0)
745 		return (error);
746 
747 	/* In the wildcard case, we need to update the name. */
748 	if (wildcard) {
749 		for (dp = name; *dp != '\0'; dp++);
750 		if (snprintf(dp, len - (dp-name), "%d", unit) >
751 		    len - (dp-name) - 1) {
752 			panic("%s: interface name too long", __func__);
753 		}
754 	}
755 
756 	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
757 	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
758 	if (ifp == NULL) {
759 		ifc_free_unit(ifc, unit);
760 		free(ifv, M_VLAN);
761 		return (ENOSPC);
762 	}
763 	SLIST_INIT(&ifv->vlan_mc_listhead);
764 
765 	ifp->if_softc = ifv;
766 	/*
767 	 * Set the name manually rather than using if_initname because
768 	 * we don't conform to the default naming convention for interfaces.
769 	 */
770 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
771 	ifp->if_dname = ifc->ifc_name;
772 	ifp->if_dunit = unit;
773 	/* NB: flags are not set here */
774 	ifp->if_linkmib = &ifv->ifv_mib;
775 	ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
776 	/* NB: mtu is not set here */
777 
778 	ifp->if_init = vlan_init;
779 	ifp->if_start = vlan_start;
780 	ifp->if_ioctl = vlan_ioctl;
781 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
782 	ifp->if_flags = VLAN_IFFLAGS;
783 	ether_ifattach(ifp, eaddr);
784 	/* Now undo some of the damage... */
785 	ifp->if_baudrate = 0;
786 	ifp->if_type = IFT_L2VLAN;
787 	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
788 
789 	if (ethertag) {
790 		error = vlan_config(ifv, p, tag);
791 		if (error != 0) {
792 			/*
793 			 * Since we've partialy failed, we need to back
794 			 * out all the way, otherwise userland could get
795 			 * confused.  Thus, we destroy the interface.
796 			 */
797 			ether_ifdetach(ifp);
798 			vlan_unconfig(ifp);
799 			if_free_type(ifp, IFT_ETHER);
800 			ifc_free_unit(ifc, unit);
801 			free(ifv, M_VLAN);
802 
803 			return (error);
804 		}
805 
806 		/* Update flags on the parent, if necessary. */
807 		vlan_setflags(ifp, 1);
808 	}
809 
810 	return (0);
811 }
812 
813 static int
814 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
815 {
816 	struct ifvlan *ifv = ifp->if_softc;
817 	int unit = ifp->if_dunit;
818 
819 	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
820 	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
821 	if_free_type(ifp, IFT_ETHER);
822 	free(ifv, M_VLAN);
823 	ifc_free_unit(ifc, unit);
824 
825 	return (0);
826 }
827 
828 /*
829  * The ifp->if_init entry point for vlan(4) is a no-op.
830  */
831 static void
832 vlan_init(void *foo __unused)
833 {
834 }
835 
836 /*
837  * The if_start method for vlan(4) interface. It doesn't
838  * raises the IFF_DRV_OACTIVE flag, since it is called
839  * only from IFQ_HANDOFF() macro in ether_output_frame().
840  * If the interface queue is full, and vlan_start() is
841  * not called, the queue would never get emptied and
842  * interface would stall forever.
843  */
844 static void
845 vlan_start(struct ifnet *ifp)
846 {
847 	struct ifvlan *ifv;
848 	struct ifnet *p;
849 	struct mbuf *m;
850 	int error;
851 
852 	ifv = ifp->if_softc;
853 	p = PARENT(ifv);
854 
855 	for (;;) {
856 		IF_DEQUEUE(&ifp->if_snd, m);
857 		if (m == NULL)
858 			break;
859 		BPF_MTAP(ifp, m);
860 
861 		/*
862 		 * Do not run parent's if_start() if the parent is not up,
863 		 * or parent's driver will cause a system crash.
864 		 */
865 		if (!UP_AND_RUNNING(p)) {
866 			m_freem(m);
867 			ifp->if_collisions++;
868 			continue;
869 		}
870 
871 		/*
872 		 * Pad the frame to the minimum size allowed if told to.
873 		 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
874 		 * paragraph C.4.4.3.b.  It can help to work around buggy
875 		 * bridges that violate paragraph C.4.4.3.a from the same
876 		 * document, i.e., fail to pad short frames after untagging.
877 		 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
878 		 * untagging it will produce a 62-byte frame, which is a runt
879 		 * and requires padding.  There are VLAN-enabled network
880 		 * devices that just discard such runts instead or mishandle
881 		 * them somehow.
882 		 */
883 		if (soft_pad) {
884 			static char pad[8];	/* just zeros */
885 			int n;
886 
887 			for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
888 			     n > 0; n -= sizeof(pad))
889 				if (!m_append(m, min(n, sizeof(pad)), pad))
890 					break;
891 
892 			if (n > 0) {
893 				if_printf(ifp, "cannot pad short frame\n");
894 				ifp->if_oerrors++;
895 				m_freem(m);
896 				continue;
897 			}
898 		}
899 
900 		/*
901 		 * If underlying interface can do VLAN tag insertion itself,
902 		 * just pass the packet along. However, we need some way to
903 		 * tell the interface where the packet came from so that it
904 		 * knows how to find the VLAN tag to use, so we attach a
905 		 * packet tag that holds it.
906 		 */
907 		if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
908 			m->m_pkthdr.ether_vtag = ifv->ifv_tag;
909 			m->m_flags |= M_VLANTAG;
910 		} else {
911 			m = ether_vlanencap(m, ifv->ifv_tag);
912 			if (m == NULL) {
913 				if_printf(ifp,
914 				    "unable to prepend VLAN header\n");
915 				ifp->if_oerrors++;
916 				continue;
917 			}
918 		}
919 
920 		/*
921 		 * Send it, precisely as ether_output() would have.
922 		 * We are already running at splimp.
923 		 */
924 		error = (p->if_transmit)(p, m);
925 		if (!error)
926 			ifp->if_opackets++;
927 		else
928 			ifp->if_oerrors++;
929 	}
930 }
931 
932 static void
933 vlan_input(struct ifnet *ifp, struct mbuf *m)
934 {
935 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
936 	struct ifvlan *ifv;
937 	uint16_t tag;
938 
939 	KASSERT(trunk != NULL, ("%s: no trunk", __func__));
940 
941 	if (m->m_flags & M_VLANTAG) {
942 		/*
943 		 * Packet is tagged, but m contains a normal
944 		 * Ethernet frame; the tag is stored out-of-band.
945 		 */
946 		tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
947 		m->m_flags &= ~M_VLANTAG;
948 	} else {
949 		struct ether_vlan_header *evl;
950 
951 		/*
952 		 * Packet is tagged in-band as specified by 802.1q.
953 		 */
954 		switch (ifp->if_type) {
955 		case IFT_ETHER:
956 			if (m->m_len < sizeof(*evl) &&
957 			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
958 				if_printf(ifp, "cannot pullup VLAN header\n");
959 				return;
960 			}
961 			evl = mtod(m, struct ether_vlan_header *);
962 			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
963 
964 			/*
965 			 * Remove the 802.1q header by copying the Ethernet
966 			 * addresses over it and adjusting the beginning of
967 			 * the data in the mbuf.  The encapsulated Ethernet
968 			 * type field is already in place.
969 			 */
970 			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
971 			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
972 			m_adj(m, ETHER_VLAN_ENCAP_LEN);
973 			break;
974 
975 		default:
976 #ifdef INVARIANTS
977 			panic("%s: %s has unsupported if_type %u",
978 			      __func__, ifp->if_xname, ifp->if_type);
979 #endif
980 			m_freem(m);
981 			ifp->if_noproto++;
982 			return;
983 		}
984 	}
985 
986 	TRUNK_RLOCK(trunk);
987 #ifdef VLAN_ARRAY
988 	ifv = trunk->vlans[tag];
989 #else
990 	ifv = vlan_gethash(trunk, tag);
991 #endif
992 	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
993 		TRUNK_RUNLOCK(trunk);
994 		m_freem(m);
995 		ifp->if_noproto++;
996 		return;
997 	}
998 	TRUNK_RUNLOCK(trunk);
999 
1000 	m->m_pkthdr.rcvif = ifv->ifv_ifp;
1001 	ifv->ifv_ifp->if_ipackets++;
1002 
1003 	/* Pass it back through the parent's input routine. */
1004 	(*ifp->if_input)(ifv->ifv_ifp, m);
1005 }
1006 
1007 static int
1008 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
1009 {
1010 	struct ifvlantrunk *trunk;
1011 	struct ifnet *ifp;
1012 	int error = 0;
1013 
1014 	/* VID numbers 0x0 and 0xFFF are reserved */
1015 	if (tag == 0 || tag == 0xFFF)
1016 		return (EINVAL);
1017 	if (p->if_type != IFT_ETHER)
1018 		return (EPROTONOSUPPORT);
1019 	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1020 		return (EPROTONOSUPPORT);
1021 	if (ifv->ifv_trunk)
1022 		return (EBUSY);
1023 
1024 	if (p->if_vlantrunk == NULL) {
1025 		trunk = malloc(sizeof(struct ifvlantrunk),
1026 		    M_VLAN, M_WAITOK | M_ZERO);
1027 #ifndef VLAN_ARRAY
1028 		vlan_inithash(trunk);
1029 #endif
1030 		VLAN_LOCK();
1031 		if (p->if_vlantrunk != NULL) {
1032 			/* A race that that is very unlikely to be hit. */
1033 #ifndef VLAN_ARRAY
1034 			vlan_freehash(trunk);
1035 #endif
1036 			free(trunk, M_VLAN);
1037 			goto exists;
1038 		}
1039 		TRUNK_LOCK_INIT(trunk);
1040 		TRUNK_LOCK(trunk);
1041 		p->if_vlantrunk = trunk;
1042 		trunk->parent = p;
1043 	} else {
1044 		VLAN_LOCK();
1045 exists:
1046 		trunk = p->if_vlantrunk;
1047 		TRUNK_LOCK(trunk);
1048 	}
1049 
1050 	ifv->ifv_tag = tag;	/* must set this before vlan_inshash() */
1051 #ifdef VLAN_ARRAY
1052 	if (trunk->vlans[tag] != NULL) {
1053 		error = EEXIST;
1054 		goto done;
1055 	}
1056 	trunk->vlans[tag] = ifv;
1057 	trunk->refcnt++;
1058 #else
1059 	error = vlan_inshash(trunk, ifv);
1060 	if (error)
1061 		goto done;
1062 #endif
1063 	ifv->ifv_proto = ETHERTYPE_VLAN;
1064 	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1065 	ifv->ifv_mintu = ETHERMIN;
1066 	ifv->ifv_pflags = 0;
1067 
1068 	/*
1069 	 * If the parent supports the VLAN_MTU capability,
1070 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1071 	 * use it.
1072 	 */
1073 	if (p->if_capenable & IFCAP_VLAN_MTU) {
1074 		/*
1075 		 * No need to fudge the MTU since the parent can
1076 		 * handle extended frames.
1077 		 */
1078 		ifv->ifv_mtufudge = 0;
1079 	} else {
1080 		/*
1081 		 * Fudge the MTU by the encapsulation size.  This
1082 		 * makes us incompatible with strictly compliant
1083 		 * 802.1Q implementations, but allows us to use
1084 		 * the feature with other NetBSD implementations,
1085 		 * which might still be useful.
1086 		 */
1087 		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1088 	}
1089 
1090 	ifv->ifv_trunk = trunk;
1091 	ifp = ifv->ifv_ifp;
1092 	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1093 	ifp->if_baudrate = p->if_baudrate;
1094 	/*
1095 	 * Copy only a selected subset of flags from the parent.
1096 	 * Other flags are none of our business.
1097 	 */
1098 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1099 	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1100 	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1101 #undef VLAN_COPY_FLAGS
1102 
1103 	ifp->if_link_state = p->if_link_state;
1104 
1105 	vlan_capabilities(ifv);
1106 
1107 	/*
1108 	 * Set up our ``Ethernet address'' to reflect the underlying
1109 	 * physical interface's.
1110 	 */
1111 	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), ETHER_ADDR_LEN);
1112 
1113 	/*
1114 	 * Configure multicast addresses that may already be
1115 	 * joined on the vlan device.
1116 	 */
1117 	(void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1118 
1119 	/* We are ready for operation now. */
1120 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1121 done:
1122 	TRUNK_UNLOCK(trunk);
1123 	if (error == 0)
1124 		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_tag);
1125 	VLAN_UNLOCK();
1126 
1127 	return (error);
1128 }
1129 
1130 static int
1131 vlan_unconfig(struct ifnet *ifp)
1132 {
1133 	int ret;
1134 
1135 	VLAN_LOCK();
1136 	ret = vlan_unconfig_locked(ifp);
1137 	VLAN_UNLOCK();
1138 	return (ret);
1139 }
1140 
1141 static int
1142 vlan_unconfig_locked(struct ifnet *ifp)
1143 {
1144 	struct ifvlantrunk *trunk;
1145 	struct vlan_mc_entry *mc;
1146 	struct ifvlan *ifv;
1147 	struct ifnet  *parent;
1148 	int error;
1149 
1150 	VLAN_LOCK_ASSERT();
1151 
1152 	ifv = ifp->if_softc;
1153 	trunk = ifv->ifv_trunk;
1154 	parent = NULL;
1155 
1156 	if (trunk != NULL) {
1157 		struct sockaddr_dl sdl;
1158 
1159 		TRUNK_LOCK(trunk);
1160 		parent = trunk->parent;
1161 
1162 		/*
1163 		 * Since the interface is being unconfigured, we need to
1164 		 * empty the list of multicast groups that we may have joined
1165 		 * while we were alive from the parent's list.
1166 		 */
1167 		bzero((char *)&sdl, sizeof(sdl));
1168 		sdl.sdl_len = sizeof(sdl);
1169 		sdl.sdl_family = AF_LINK;
1170 		sdl.sdl_index = parent->if_index;
1171 		sdl.sdl_type = IFT_ETHER;
1172 		sdl.sdl_alen = ETHER_ADDR_LEN;
1173 
1174 		while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1175 			bcopy((char *)&mc->mc_addr, LLADDR(&sdl),
1176 			    ETHER_ADDR_LEN);
1177 			error = if_delmulti(parent, (struct sockaddr *)&sdl);
1178 			if (error)
1179 				return (error);
1180 			SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1181 			free(mc, M_VLAN);
1182 		}
1183 
1184 		vlan_setflags(ifp, 0); /* clear special flags on parent */
1185 #ifdef VLAN_ARRAY
1186 		trunk->vlans[ifv->ifv_tag] = NULL;
1187 		trunk->refcnt--;
1188 #else
1189 		vlan_remhash(trunk, ifv);
1190 #endif
1191 		ifv->ifv_trunk = NULL;
1192 
1193 		/*
1194 		 * Check if we were the last.
1195 		 */
1196 		if (trunk->refcnt == 0) {
1197 			trunk->parent->if_vlantrunk = NULL;
1198 			/*
1199 			 * XXXGL: If some ithread has already entered
1200 			 * vlan_input() and is now blocked on the trunk
1201 			 * lock, then it should preempt us right after
1202 			 * unlock and finish its work. Then we will acquire
1203 			 * lock again in trunk_destroy().
1204 			 */
1205 			TRUNK_UNLOCK(trunk);
1206 			trunk_destroy(trunk);
1207 		} else
1208 			TRUNK_UNLOCK(trunk);
1209 	}
1210 
1211 	/* Disconnect from parent. */
1212 	if (ifv->ifv_pflags)
1213 		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1214 	ifp->if_mtu = ETHERMTU;
1215 	ifp->if_link_state = LINK_STATE_UNKNOWN;
1216 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1217 
1218 	/*
1219 	 * Only dispatch an event if vlan was
1220 	 * attached, otherwise there is nothing
1221 	 * to cleanup anyway.
1222 	 */
1223 	if (parent != NULL)
1224 		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_tag);
1225 
1226 	return (0);
1227 }
1228 
1229 /* Handle a reference counted flag that should be set on the parent as well */
1230 static int
1231 vlan_setflag(struct ifnet *ifp, int flag, int status,
1232 	     int (*func)(struct ifnet *, int))
1233 {
1234 	struct ifvlan *ifv;
1235 	int error;
1236 
1237 	/* XXX VLAN_LOCK_ASSERT(); */
1238 
1239 	ifv = ifp->if_softc;
1240 	status = status ? (ifp->if_flags & flag) : 0;
1241 	/* Now "status" contains the flag value or 0 */
1242 
1243 	/*
1244 	 * See if recorded parent's status is different from what
1245 	 * we want it to be.  If it is, flip it.  We record parent's
1246 	 * status in ifv_pflags so that we won't clear parent's flag
1247 	 * we haven't set.  In fact, we don't clear or set parent's
1248 	 * flags directly, but get or release references to them.
1249 	 * That's why we can be sure that recorded flags still are
1250 	 * in accord with actual parent's flags.
1251 	 */
1252 	if (status != (ifv->ifv_pflags & flag)) {
1253 		error = (*func)(PARENT(ifv), status);
1254 		if (error)
1255 			return (error);
1256 		ifv->ifv_pflags &= ~flag;
1257 		ifv->ifv_pflags |= status;
1258 	}
1259 	return (0);
1260 }
1261 
1262 /*
1263  * Handle IFF_* flags that require certain changes on the parent:
1264  * if "status" is true, update parent's flags respective to our if_flags;
1265  * if "status" is false, forcedly clear the flags set on parent.
1266  */
1267 static int
1268 vlan_setflags(struct ifnet *ifp, int status)
1269 {
1270 	int error, i;
1271 
1272 	for (i = 0; vlan_pflags[i].flag; i++) {
1273 		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1274 				     status, vlan_pflags[i].func);
1275 		if (error)
1276 			return (error);
1277 	}
1278 	return (0);
1279 }
1280 
1281 /* Inform all vlans that their parent has changed link state */
1282 static void
1283 vlan_link_state(struct ifnet *ifp)
1284 {
1285 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1286 	struct ifvlan *ifv;
1287 	int i;
1288 
1289 	TRUNK_LOCK(trunk);
1290 #ifdef VLAN_ARRAY
1291 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1292 		if (trunk->vlans[i] != NULL) {
1293 			ifv = trunk->vlans[i];
1294 #else
1295 	for (i = 0; i < (1 << trunk->hwidth); i++)
1296 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1297 #endif
1298 			ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1299 			if_link_state_change(ifv->ifv_ifp,
1300 			    trunk->parent->if_link_state);
1301 		}
1302 	TRUNK_UNLOCK(trunk);
1303 }
1304 
1305 static void
1306 vlan_capabilities(struct ifvlan *ifv)
1307 {
1308 	struct ifnet *p = PARENT(ifv);
1309 	struct ifnet *ifp = ifv->ifv_ifp;
1310 
1311 	TRUNK_LOCK_ASSERT(TRUNK(ifv));
1312 
1313 	/*
1314 	 * If the parent interface can do checksum offloading
1315 	 * on VLANs, then propagate its hardware-assisted
1316 	 * checksumming flags. Also assert that checksum
1317 	 * offloading requires hardware VLAN tagging.
1318 	 */
1319 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1320 		ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1321 
1322 	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1323 	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1324 		ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1325 		ifp->if_hwassist = p->if_hwassist;
1326 	} else {
1327 		ifp->if_capenable = 0;
1328 		ifp->if_hwassist = 0;
1329 	}
1330 }
1331 
1332 static void
1333 vlan_trunk_capabilities(struct ifnet *ifp)
1334 {
1335 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1336 	struct ifvlan *ifv;
1337 	int i;
1338 
1339 	TRUNK_LOCK(trunk);
1340 #ifdef VLAN_ARRAY
1341 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1342 		if (trunk->vlans[i] != NULL) {
1343 			ifv = trunk->vlans[i];
1344 #else
1345 	for (i = 0; i < (1 << trunk->hwidth); i++) {
1346 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1347 #endif
1348 			vlan_capabilities(ifv);
1349 	}
1350 	TRUNK_UNLOCK(trunk);
1351 }
1352 
1353 static int
1354 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1355 {
1356 	struct ifnet *p;
1357 	struct ifreq *ifr;
1358 	struct ifvlan *ifv;
1359 	struct vlanreq vlr;
1360 	int error = 0;
1361 
1362 	ifr = (struct ifreq *)data;
1363 	ifv = ifp->if_softc;
1364 
1365 	switch (cmd) {
1366 	case SIOCGIFMEDIA:
1367 		VLAN_LOCK();
1368 		if (TRUNK(ifv) != NULL) {
1369 			error = (*PARENT(ifv)->if_ioctl)(PARENT(ifv),
1370 					SIOCGIFMEDIA, data);
1371 			VLAN_UNLOCK();
1372 			/* Limit the result to the parent's current config. */
1373 			if (error == 0) {
1374 				struct ifmediareq *ifmr;
1375 
1376 				ifmr = (struct ifmediareq *)data;
1377 				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1378 					ifmr->ifm_count = 1;
1379 					error = copyout(&ifmr->ifm_current,
1380 						ifmr->ifm_ulist,
1381 						sizeof(int));
1382 				}
1383 			}
1384 		} else {
1385 			VLAN_UNLOCK();
1386 			error = EINVAL;
1387 		}
1388 		break;
1389 
1390 	case SIOCSIFMEDIA:
1391 		error = EINVAL;
1392 		break;
1393 
1394 	case SIOCSIFMTU:
1395 		/*
1396 		 * Set the interface MTU.
1397 		 */
1398 		VLAN_LOCK();
1399 		if (TRUNK(ifv) != NULL) {
1400 			if (ifr->ifr_mtu >
1401 			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1402 			    ifr->ifr_mtu <
1403 			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1404 				error = EINVAL;
1405 			else
1406 				ifp->if_mtu = ifr->ifr_mtu;
1407 		} else
1408 			error = EINVAL;
1409 		VLAN_UNLOCK();
1410 		break;
1411 
1412 	case SIOCSETVLAN:
1413 		error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1414 		if (error)
1415 			break;
1416 		if (vlr.vlr_parent[0] == '\0') {
1417 			vlan_unconfig(ifp);
1418 			break;
1419 		}
1420 		p = ifunit(vlr.vlr_parent);
1421 		if (p == NULL) {
1422 			error = ENOENT;
1423 			break;
1424 		}
1425 		/*
1426 		 * Don't let the caller set up a VLAN tag with
1427 		 * anything except VLID bits.
1428 		 */
1429 		if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1430 			error = EINVAL;
1431 			break;
1432 		}
1433 		error = vlan_config(ifv, p, vlr.vlr_tag);
1434 		if (error)
1435 			break;
1436 
1437 		/* Update flags on the parent, if necessary. */
1438 		vlan_setflags(ifp, 1);
1439 		break;
1440 
1441 	case SIOCGETVLAN:
1442 		bzero(&vlr, sizeof(vlr));
1443 		VLAN_LOCK();
1444 		if (TRUNK(ifv) != NULL) {
1445 			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1446 			    sizeof(vlr.vlr_parent));
1447 			vlr.vlr_tag = ifv->ifv_tag;
1448 		}
1449 		VLAN_UNLOCK();
1450 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1451 		break;
1452 
1453 	case SIOCSIFFLAGS:
1454 		/*
1455 		 * We should propagate selected flags to the parent,
1456 		 * e.g., promiscuous mode.
1457 		 */
1458 		if (TRUNK(ifv) != NULL)
1459 			error = vlan_setflags(ifp, 1);
1460 		break;
1461 
1462 	case SIOCADDMULTI:
1463 	case SIOCDELMULTI:
1464 		/*
1465 		 * If we don't have a parent, just remember the membership for
1466 		 * when we do.
1467 		 */
1468 		if (TRUNK(ifv) != NULL)
1469 			error = vlan_setmulti(ifp);
1470 		break;
1471 
1472 	default:
1473 		error = ether_ioctl(ifp, cmd, data);
1474 	}
1475 
1476 	return (error);
1477 }
1478