xref: /freebsd/sys/net/if_vlan.c (revision b3aaa0cc21c63d388230c7ef2a80abd631ff20d5)
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 #include <sys/vimage.h>
59 
60 #include <net/bpf.h>
61 #include <net/ethernet.h>
62 #include <net/if.h>
63 #include <net/if_clone.h>
64 #include <net/if_dl.h>
65 #include <net/if_types.h>
66 #include <net/if_vlan_var.h>
67 #include <net/vnet.h>
68 
69 #define VLANNAME	"vlan"
70 #define	VLAN_DEF_HWIDTH	4
71 #define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
72 
73 #define	UP_AND_RUNNING(ifp) \
74     ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
75 
76 LIST_HEAD(ifvlanhead, ifvlan);
77 
78 struct ifvlantrunk {
79 	struct	ifnet   *parent;	/* parent interface of this trunk */
80 	struct	rwlock	rw;
81 #ifdef VLAN_ARRAY
82 #define	VLAN_ARRAY_SIZE	(EVL_VLID_MASK + 1)
83 	struct	ifvlan	*vlans[VLAN_ARRAY_SIZE]; /* static table */
84 #else
85 	struct	ifvlanhead *hash;	/* dynamic hash-list table */
86 	uint16_t	hmask;
87 	uint16_t	hwidth;
88 #endif
89 	int		refcnt;
90 };
91 
92 struct vlan_mc_entry {
93 	struct ether_addr		mc_addr;
94 	SLIST_ENTRY(vlan_mc_entry)	mc_entries;
95 };
96 
97 struct	ifvlan {
98 	struct	ifvlantrunk *ifv_trunk;
99 	struct	ifnet *ifv_ifp;
100 #define	TRUNK(ifv)	((ifv)->ifv_trunk)
101 #define	PARENT(ifv)	((ifv)->ifv_trunk->parent)
102 	int	ifv_pflags;	/* special flags we have set on parent */
103 	struct	ifv_linkmib {
104 		int	ifvm_encaplen;	/* encapsulation length */
105 		int	ifvm_mtufudge;	/* MTU fudged by this much */
106 		int	ifvm_mintu;	/* min transmission unit */
107 		uint16_t ifvm_proto;	/* encapsulation ethertype */
108 		uint16_t ifvm_tag;	/* tag to apply on packets leaving if */
109 	}	ifv_mib;
110 	SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
111 #ifndef VLAN_ARRAY
112 	LIST_ENTRY(ifvlan) ifv_list;
113 #endif
114 };
115 #define	ifv_proto	ifv_mib.ifvm_proto
116 #define	ifv_tag		ifv_mib.ifvm_tag
117 #define	ifv_encaplen	ifv_mib.ifvm_encaplen
118 #define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
119 #define	ifv_mintu	ifv_mib.ifvm_mintu
120 
121 /* Special flags we should propagate to parent. */
122 static struct {
123 	int flag;
124 	int (*func)(struct ifnet *, int);
125 } vlan_pflags[] = {
126 	{IFF_PROMISC, ifpromisc},
127 	{IFF_ALLMULTI, if_allmulti},
128 	{0, NULL}
129 };
130 
131 SYSCTL_DECL(_net_link);
132 SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0, "IEEE 802.1Q VLAN");
133 SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0, "for consistency");
134 
135 static int soft_pad = 0;
136 SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW, &soft_pad, 0,
137 	   "pad short frames before tagging");
138 
139 static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface");
140 
141 static eventhandler_tag ifdetach_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, int link);
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 
204 static	struct if_clone vlan_cloner = IFC_CLONE_INITIALIZER(VLANNAME, NULL,
205     IF_MAXUNIT, NULL, vlan_clone_match, vlan_clone_create, vlan_clone_destroy);
206 
207 #ifndef VLAN_ARRAY
208 #define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
209 
210 static void
211 vlan_inithash(struct ifvlantrunk *trunk)
212 {
213 	int i, n;
214 
215 	/*
216 	 * The trunk must not be locked here since we call malloc(M_WAITOK).
217 	 * It is OK in case this function is called before the trunk struct
218 	 * gets hooked up and becomes visible from other threads.
219 	 */
220 
221 	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
222 	    ("%s: hash already initialized", __func__));
223 
224 	trunk->hwidth = VLAN_DEF_HWIDTH;
225 	n = 1 << trunk->hwidth;
226 	trunk->hmask = n - 1;
227 	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
228 	for (i = 0; i < n; i++)
229 		LIST_INIT(&trunk->hash[i]);
230 }
231 
232 static void
233 vlan_freehash(struct ifvlantrunk *trunk)
234 {
235 #ifdef INVARIANTS
236 	int i;
237 
238 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
239 	for (i = 0; i < (1 << trunk->hwidth); i++)
240 		KASSERT(LIST_EMPTY(&trunk->hash[i]),
241 		    ("%s: hash table not empty", __func__));
242 #endif
243 	free(trunk->hash, M_VLAN);
244 	trunk->hash = NULL;
245 	trunk->hwidth = trunk->hmask = 0;
246 }
247 
248 static int
249 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
250 {
251 	int i, b;
252 	struct ifvlan *ifv2;
253 
254 	TRUNK_LOCK_ASSERT(trunk);
255 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
256 
257 	b = 1 << trunk->hwidth;
258 	i = HASH(ifv->ifv_tag, trunk->hmask);
259 	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
260 		if (ifv->ifv_tag == ifv2->ifv_tag)
261 			return (EEXIST);
262 
263 	/*
264 	 * Grow the hash when the number of vlans exceeds half of the number of
265 	 * hash buckets squared. This will make the average linked-list length
266 	 * buckets/2.
267 	 */
268 	if (trunk->refcnt > (b * b) / 2) {
269 		vlan_growhash(trunk, 1);
270 		i = HASH(ifv->ifv_tag, trunk->hmask);
271 	}
272 	LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
273 	trunk->refcnt++;
274 
275 	return (0);
276 }
277 
278 static int
279 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
280 {
281 	int i, b;
282 	struct ifvlan *ifv2;
283 
284 	TRUNK_LOCK_ASSERT(trunk);
285 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
286 
287 	b = 1 << trunk->hwidth;
288 	i = HASH(ifv->ifv_tag, trunk->hmask);
289 	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
290 		if (ifv2 == ifv) {
291 			trunk->refcnt--;
292 			LIST_REMOVE(ifv2, ifv_list);
293 			if (trunk->refcnt < (b * b) / 2)
294 				vlan_growhash(trunk, -1);
295 			return (0);
296 		}
297 
298 	panic("%s: vlan not found\n", __func__);
299 	return (ENOENT); /*NOTREACHED*/
300 }
301 
302 /*
303  * Grow the hash larger or smaller if memory permits.
304  */
305 static void
306 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
307 {
308 	struct ifvlan *ifv;
309 	struct ifvlanhead *hash2;
310 	int hwidth2, i, j, n, n2;
311 
312 	TRUNK_LOCK_ASSERT(trunk);
313 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
314 
315 	if (howmuch == 0) {
316 		/* Harmless yet obvious coding error */
317 		printf("%s: howmuch is 0\n", __func__);
318 		return;
319 	}
320 
321 	hwidth2 = trunk->hwidth + howmuch;
322 	n = 1 << trunk->hwidth;
323 	n2 = 1 << hwidth2;
324 	/* Do not shrink the table below the default */
325 	if (hwidth2 < VLAN_DEF_HWIDTH)
326 		return;
327 
328 	/* M_NOWAIT because we're called with trunk mutex held */
329 	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
330 	if (hash2 == NULL) {
331 		printf("%s: out of memory -- hash size not changed\n",
332 		    __func__);
333 		return;		/* We can live with the old hash table */
334 	}
335 	for (j = 0; j < n2; j++)
336 		LIST_INIT(&hash2[j]);
337 	for (i = 0; i < n; i++)
338 		while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
339 			LIST_REMOVE(ifv, ifv_list);
340 			j = HASH(ifv->ifv_tag, n2 - 1);
341 			LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
342 		}
343 	free(trunk->hash, M_VLAN);
344 	trunk->hash = hash2;
345 	trunk->hwidth = hwidth2;
346 	trunk->hmask = n2 - 1;
347 
348 	if (bootverbose)
349 		if_printf(trunk->parent,
350 		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
351 }
352 
353 static __inline struct ifvlan *
354 vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
355 {
356 	struct ifvlan *ifv;
357 
358 	TRUNK_LOCK_RASSERT(trunk);
359 
360 	LIST_FOREACH(ifv, &trunk->hash[HASH(tag, trunk->hmask)], ifv_list)
361 		if (ifv->ifv_tag == tag)
362 			return (ifv);
363 	return (NULL);
364 }
365 
366 #if 0
367 /* Debugging code to view the hashtables. */
368 static void
369 vlan_dumphash(struct ifvlantrunk *trunk)
370 {
371 	int i;
372 	struct ifvlan *ifv;
373 
374 	for (i = 0; i < (1 << trunk->hwidth); i++) {
375 		printf("%d: ", i);
376 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
377 			printf("%s ", ifv->ifv_ifp->if_xname);
378 		printf("\n");
379 	}
380 }
381 #endif /* 0 */
382 #endif /* !VLAN_ARRAY */
383 
384 static void
385 trunk_destroy(struct ifvlantrunk *trunk)
386 {
387 	VLAN_LOCK_ASSERT();
388 
389 	TRUNK_LOCK(trunk);
390 #ifndef VLAN_ARRAY
391 	vlan_freehash(trunk);
392 #endif
393 	trunk->parent->if_vlantrunk = NULL;
394 	TRUNK_UNLOCK(trunk);
395 	TRUNK_LOCK_DESTROY(trunk);
396 	free(trunk, M_VLAN);
397 }
398 
399 /*
400  * Program our multicast filter. What we're actually doing is
401  * programming the multicast filter of the parent. This has the
402  * side effect of causing the parent interface to receive multicast
403  * traffic that it doesn't really want, which ends up being discarded
404  * later by the upper protocol layers. Unfortunately, there's no way
405  * to avoid this: there really is only one physical interface.
406  *
407  * XXX: There is a possible race here if more than one thread is
408  *      modifying the multicast state of the vlan interface at the same time.
409  */
410 static int
411 vlan_setmulti(struct ifnet *ifp)
412 {
413 	struct ifnet		*ifp_p;
414 	struct ifmultiaddr	*ifma, *rifma = NULL;
415 	struct ifvlan		*sc;
416 	struct vlan_mc_entry	*mc;
417 	struct sockaddr_dl	sdl;
418 	int			error;
419 
420 	/*VLAN_LOCK_ASSERT();*/
421 
422 	/* Find the parent. */
423 	sc = ifp->if_softc;
424 	ifp_p = PARENT(sc);
425 
426 	CURVNET_SET_QUIET(ifp_p->if_vnet);
427 
428 	bzero((char *)&sdl, sizeof(sdl));
429 	sdl.sdl_len = sizeof(sdl);
430 	sdl.sdl_family = AF_LINK;
431 	sdl.sdl_index = ifp_p->if_index;
432 	sdl.sdl_type = IFT_ETHER;
433 	sdl.sdl_alen = ETHER_ADDR_LEN;
434 
435 	/* First, remove any existing filter entries. */
436 	while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
437 		bcopy((char *)&mc->mc_addr, LLADDR(&sdl), ETHER_ADDR_LEN);
438 		error = if_delmulti(ifp_p, (struct sockaddr *)&sdl);
439 		if (error)
440 			return (error);
441 		SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
442 		free(mc, M_VLAN);
443 	}
444 
445 	/* Now program new ones. */
446 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
447 		if (ifma->ifma_addr->sa_family != AF_LINK)
448 			continue;
449 		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
450 		if (mc == NULL)
451 			return (ENOMEM);
452 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
453 		    (char *)&mc->mc_addr, ETHER_ADDR_LEN);
454 		SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
455 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
456 		    LLADDR(&sdl), ETHER_ADDR_LEN);
457 		error = if_addmulti(ifp_p, (struct sockaddr *)&sdl, &rifma);
458 		if (error)
459 			return (error);
460 	}
461 
462 	CURVNET_RESTORE();
463 	return (0);
464 }
465 
466 /*
467  * A handler for network interface departure events.
468  * Track departure of trunks here so that we don't access invalid
469  * pointers or whatever if a trunk is ripped from under us, e.g.,
470  * by ejecting its hot-plug card.
471  */
472 static void
473 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
474 {
475 	struct ifvlan *ifv;
476 	int i;
477 
478 	/*
479 	 * Check if it's a trunk interface first of all
480 	 * to avoid needless locking.
481 	 */
482 	if (ifp->if_vlantrunk == NULL)
483 		return;
484 
485 	VLAN_LOCK();
486 	/*
487 	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
488 	 * Check trunk pointer after each vlan_unconfig() as it will
489 	 * free it and set to NULL after the last vlan was detached.
490 	 */
491 #ifdef VLAN_ARRAY
492 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
493 		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
494 			vlan_unconfig_locked(ifv->ifv_ifp);
495 			if (ifp->if_vlantrunk == NULL)
496 				break;
497 		}
498 #else /* VLAN_ARRAY */
499 restart:
500 	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
501 		if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
502 			vlan_unconfig_locked(ifv->ifv_ifp);
503 			if (ifp->if_vlantrunk)
504 				goto restart;	/* trunk->hwidth can change */
505 			else
506 				break;
507 		}
508 #endif /* VLAN_ARRAY */
509 	/* Trunk should have been destroyed in vlan_unconfig(). */
510 	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
511 	VLAN_UNLOCK();
512 }
513 
514 /*
515  * VLAN support can be loaded as a module.  The only place in the
516  * system that's intimately aware of this is ether_input.  We hook
517  * into this code through vlan_input_p which is defined there and
518  * set here.  Noone else in the system should be aware of this so
519  * we use an explicit reference here.
520  */
521 extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
522 
523 /* For if_link_state_change() eyes only... */
524 extern	void (*vlan_link_state_p)(struct ifnet *, int);
525 
526 static int
527 vlan_modevent(module_t mod, int type, void *data)
528 {
529 
530 	switch (type) {
531 	case MOD_LOAD:
532 		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
533 		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
534 		if (ifdetach_tag == NULL)
535 			return (ENOMEM);
536 		VLAN_LOCK_INIT();
537 		vlan_input_p = vlan_input;
538 		vlan_link_state_p = vlan_link_state;
539 		vlan_trunk_cap_p = vlan_trunk_capabilities;
540 		if_clone_attach(&vlan_cloner);
541 		if (bootverbose)
542 			printf("vlan: initialized, using "
543 #ifdef VLAN_ARRAY
544 			       "full-size arrays"
545 #else
546 			       "hash tables with chaining"
547 #endif
548 
549 			       "\n");
550 		break;
551 	case MOD_UNLOAD:
552 		if_clone_detach(&vlan_cloner);
553 		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
554 		vlan_input_p = NULL;
555 		vlan_link_state_p = NULL;
556 		vlan_trunk_cap_p = NULL;
557 		VLAN_LOCK_DESTROY();
558 		if (bootverbose)
559 			printf("vlan: unloaded\n");
560 		break;
561 	default:
562 		return (EOPNOTSUPP);
563 	}
564 	return (0);
565 }
566 
567 static moduledata_t vlan_mod = {
568 	"if_vlan",
569 	vlan_modevent,
570 	0
571 };
572 
573 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
574 MODULE_VERSION(if_vlan, 3);
575 MODULE_DEPEND(if_vlan, miibus, 1, 1, 1);
576 
577 static struct ifnet *
578 vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
579 {
580 	INIT_VNET_NET(curvnet);
581 	const char *cp;
582 	struct ifnet *ifp;
583 	int t = 0;
584 
585 	/* Check for <etherif>.<vlan> style interface names. */
586 	IFNET_RLOCK();
587 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
588 		if (ifp->if_type != IFT_ETHER)
589 			continue;
590 		if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
591 			continue;
592 		cp = name + strlen(ifp->if_xname);
593 		if (*cp != '.')
594 			continue;
595 		for(; *cp != '\0'; cp++) {
596 			if (*cp < '0' || *cp > '9')
597 				continue;
598 			t = (t * 10) + (*cp - '0');
599 		}
600 		if (tag != NULL)
601 			*tag = t;
602 		break;
603 	}
604 	IFNET_RUNLOCK();
605 
606 	return (ifp);
607 }
608 
609 static int
610 vlan_clone_match(struct if_clone *ifc, const char *name)
611 {
612 	const char *cp;
613 
614 	if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
615 		return (1);
616 
617 	if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
618 		return (0);
619 	for (cp = name + 4; *cp != '\0'; cp++) {
620 		if (*cp < '0' || *cp > '9')
621 			return (0);
622 	}
623 
624 	return (1);
625 }
626 
627 static int
628 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
629 {
630 	char *dp;
631 	int wildcard;
632 	int unit;
633 	int error;
634 	int tag;
635 	int ethertag;
636 	struct ifvlan *ifv;
637 	struct ifnet *ifp;
638 	struct ifnet *p;
639 	struct vlanreq vlr;
640 	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
641 
642 	/*
643 	 * There are 3 (ugh) ways to specify the cloned device:
644 	 * o pass a parameter block with the clone request.
645 	 * o specify parameters in the text of the clone device name
646 	 * o specify no parameters and get an unattached device that
647 	 *   must be configured separately.
648 	 * The first technique is preferred; the latter two are
649 	 * supported for backwards compatibilty.
650 	 */
651 	if (params) {
652 		error = copyin(params, &vlr, sizeof(vlr));
653 		if (error)
654 			return error;
655 		p = ifunit(vlr.vlr_parent);
656 		if (p == NULL)
657 			return ENXIO;
658 		/*
659 		 * Don't let the caller set up a VLAN tag with
660 		 * anything except VLID bits.
661 		 */
662 		if (vlr.vlr_tag & ~EVL_VLID_MASK)
663 			return (EINVAL);
664 		error = ifc_name2unit(name, &unit);
665 		if (error != 0)
666 			return (error);
667 
668 		ethertag = 1;
669 		tag = vlr.vlr_tag;
670 		wildcard = (unit < 0);
671 	} else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
672 		ethertag = 1;
673 		unit = -1;
674 		wildcard = 0;
675 
676 		/*
677 		 * Don't let the caller set up a VLAN tag with
678 		 * anything except VLID bits.
679 		 */
680 		if (tag & ~EVL_VLID_MASK)
681 			return (EINVAL);
682 	} else {
683 		ethertag = 0;
684 
685 		error = ifc_name2unit(name, &unit);
686 		if (error != 0)
687 			return (error);
688 
689 		wildcard = (unit < 0);
690 	}
691 
692 	error = ifc_alloc_unit(ifc, &unit);
693 	if (error != 0)
694 		return (error);
695 
696 	/* In the wildcard case, we need to update the name. */
697 	if (wildcard) {
698 		for (dp = name; *dp != '\0'; dp++);
699 		if (snprintf(dp, len - (dp-name), "%d", unit) >
700 		    len - (dp-name) - 1) {
701 			panic("%s: interface name too long", __func__);
702 		}
703 	}
704 
705 	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
706 	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
707 	if (ifp == NULL) {
708 		ifc_free_unit(ifc, unit);
709 		free(ifv, M_VLAN);
710 		return (ENOSPC);
711 	}
712 	SLIST_INIT(&ifv->vlan_mc_listhead);
713 
714 	ifp->if_softc = ifv;
715 	/*
716 	 * Set the name manually rather than using if_initname because
717 	 * we don't conform to the default naming convention for interfaces.
718 	 */
719 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
720 	ifp->if_dname = ifc->ifc_name;
721 	ifp->if_dunit = unit;
722 	/* NB: flags are not set here */
723 	ifp->if_linkmib = &ifv->ifv_mib;
724 	ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
725 	/* NB: mtu is not set here */
726 
727 	ifp->if_init = vlan_init;
728 	ifp->if_start = vlan_start;
729 	ifp->if_ioctl = vlan_ioctl;
730 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
731 	ifp->if_flags = VLAN_IFFLAGS;
732 	ether_ifattach(ifp, eaddr);
733 	/* Now undo some of the damage... */
734 	ifp->if_baudrate = 0;
735 	ifp->if_type = IFT_L2VLAN;
736 	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
737 
738 	if (ethertag) {
739 		error = vlan_config(ifv, p, tag);
740 		if (error != 0) {
741 			/*
742 			 * Since we've partialy failed, we need to back
743 			 * out all the way, otherwise userland could get
744 			 * confused.  Thus, we destroy the interface.
745 			 */
746 			ether_ifdetach(ifp);
747 			vlan_unconfig(ifp);
748 			if_free_type(ifp, IFT_ETHER);
749 			ifc_free_unit(ifc, unit);
750 			free(ifv, M_VLAN);
751 
752 			return (error);
753 		}
754 
755 		/* Update flags on the parent, if necessary. */
756 		vlan_setflags(ifp, 1);
757 	}
758 
759 	return (0);
760 }
761 
762 static int
763 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
764 {
765 	struct ifvlan *ifv = ifp->if_softc;
766 	int unit = ifp->if_dunit;
767 
768 	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
769 	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
770 	if_free_type(ifp, IFT_ETHER);
771 	free(ifv, M_VLAN);
772 	ifc_free_unit(ifc, unit);
773 
774 	return (0);
775 }
776 
777 /*
778  * The ifp->if_init entry point for vlan(4) is a no-op.
779  */
780 static void
781 vlan_init(void *foo __unused)
782 {
783 }
784 
785 /*
786  * The if_start method for vlan(4) interface. It doesn't
787  * raises the IFF_DRV_OACTIVE flag, since it is called
788  * only from IFQ_HANDOFF() macro in ether_output_frame().
789  * If the interface queue is full, and vlan_start() is
790  * not called, the queue would never get emptied and
791  * interface would stall forever.
792  */
793 static void
794 vlan_start(struct ifnet *ifp)
795 {
796 	struct ifvlan *ifv;
797 	struct ifnet *p;
798 	struct mbuf *m;
799 	int error;
800 
801 	ifv = ifp->if_softc;
802 	p = PARENT(ifv);
803 
804 	for (;;) {
805 		IF_DEQUEUE(&ifp->if_snd, m);
806 		if (m == NULL)
807 			break;
808 		BPF_MTAP(ifp, m);
809 
810 		/*
811 		 * Do not run parent's if_start() if the parent is not up,
812 		 * or parent's driver will cause a system crash.
813 		 */
814 		if (!UP_AND_RUNNING(p)) {
815 			m_freem(m);
816 			ifp->if_collisions++;
817 			continue;
818 		}
819 
820 		/*
821 		 * Pad the frame to the minimum size allowed if told to.
822 		 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
823 		 * paragraph C.4.4.3.b.  It can help to work around buggy
824 		 * bridges that violate paragraph C.4.4.3.a from the same
825 		 * document, i.e., fail to pad short frames after untagging.
826 		 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
827 		 * untagging it will produce a 62-byte frame, which is a runt
828 		 * and requires padding.  There are VLAN-enabled network
829 		 * devices that just discard such runts instead or mishandle
830 		 * them somehow.
831 		 */
832 		if (soft_pad) {
833 			static char pad[8];	/* just zeros */
834 			int n;
835 
836 			for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
837 			     n > 0; n -= sizeof(pad))
838 				if (!m_append(m, min(n, sizeof(pad)), pad))
839 					break;
840 
841 			if (n > 0) {
842 				if_printf(ifp, "cannot pad short frame\n");
843 				ifp->if_oerrors++;
844 				m_freem(m);
845 				continue;
846 			}
847 		}
848 
849 		/*
850 		 * If underlying interface can do VLAN tag insertion itself,
851 		 * just pass the packet along. However, we need some way to
852 		 * tell the interface where the packet came from so that it
853 		 * knows how to find the VLAN tag to use, so we attach a
854 		 * packet tag that holds it.
855 		 */
856 		if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
857 			m->m_pkthdr.ether_vtag = ifv->ifv_tag;
858 			m->m_flags |= M_VLANTAG;
859 		} else {
860 			m = ether_vlanencap(m, ifv->ifv_tag);
861 			if (m == NULL) {
862 				if_printf(ifp,
863 				    "unable to prepend VLAN header\n");
864 				ifp->if_oerrors++;
865 				continue;
866 			}
867 		}
868 
869 		/*
870 		 * Send it, precisely as ether_output() would have.
871 		 * We are already running at splimp.
872 		 */
873 		error = (p->if_transmit)(p, m);
874 		if (!error)
875 			ifp->if_opackets++;
876 		else
877 			ifp->if_oerrors++;
878 	}
879 }
880 
881 static void
882 vlan_input(struct ifnet *ifp, struct mbuf *m)
883 {
884 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
885 	struct ifvlan *ifv;
886 	uint16_t tag;
887 
888 	KASSERT(trunk != NULL, ("%s: no trunk", __func__));
889 
890 	if (m->m_flags & M_VLANTAG) {
891 		/*
892 		 * Packet is tagged, but m contains a normal
893 		 * Ethernet frame; the tag is stored out-of-band.
894 		 */
895 		tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
896 		m->m_flags &= ~M_VLANTAG;
897 	} else {
898 		struct ether_vlan_header *evl;
899 
900 		/*
901 		 * Packet is tagged in-band as specified by 802.1q.
902 		 */
903 		switch (ifp->if_type) {
904 		case IFT_ETHER:
905 			if (m->m_len < sizeof(*evl) &&
906 			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
907 				if_printf(ifp, "cannot pullup VLAN header\n");
908 				return;
909 			}
910 			evl = mtod(m, struct ether_vlan_header *);
911 			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
912 
913 			/*
914 			 * Remove the 802.1q header by copying the Ethernet
915 			 * addresses over it and adjusting the beginning of
916 			 * the data in the mbuf.  The encapsulated Ethernet
917 			 * type field is already in place.
918 			 */
919 			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
920 			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
921 			m_adj(m, ETHER_VLAN_ENCAP_LEN);
922 			break;
923 
924 		default:
925 #ifdef INVARIANTS
926 			panic("%s: %s has unsupported if_type %u",
927 			      __func__, ifp->if_xname, ifp->if_type);
928 #endif
929 			m_freem(m);
930 			ifp->if_noproto++;
931 			return;
932 		}
933 	}
934 
935 	TRUNK_RLOCK(trunk);
936 #ifdef VLAN_ARRAY
937 	ifv = trunk->vlans[tag];
938 #else
939 	ifv = vlan_gethash(trunk, tag);
940 #endif
941 	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
942 		TRUNK_RUNLOCK(trunk);
943 		m_freem(m);
944 		ifp->if_noproto++;
945 		return;
946 	}
947 	TRUNK_RUNLOCK(trunk);
948 
949 	m->m_pkthdr.rcvif = ifv->ifv_ifp;
950 	ifv->ifv_ifp->if_ipackets++;
951 
952 	/* Pass it back through the parent's input routine. */
953 	(*ifp->if_input)(ifv->ifv_ifp, m);
954 }
955 
956 static int
957 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
958 {
959 	struct ifvlantrunk *trunk;
960 	struct ifnet *ifp;
961 	int error = 0;
962 
963 	/* VID numbers 0x0 and 0xFFF are reserved */
964 	if (tag == 0 || tag == 0xFFF)
965 		return (EINVAL);
966 	if (p->if_type != IFT_ETHER)
967 		return (EPROTONOSUPPORT);
968 	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
969 		return (EPROTONOSUPPORT);
970 	if (ifv->ifv_trunk)
971 		return (EBUSY);
972 
973 	if (p->if_vlantrunk == NULL) {
974 		trunk = malloc(sizeof(struct ifvlantrunk),
975 		    M_VLAN, M_WAITOK | M_ZERO);
976 #ifndef VLAN_ARRAY
977 		vlan_inithash(trunk);
978 #endif
979 		VLAN_LOCK();
980 		if (p->if_vlantrunk != NULL) {
981 			/* A race that that is very unlikely to be hit. */
982 #ifndef VLAN_ARRAY
983 			vlan_freehash(trunk);
984 #endif
985 			free(trunk, M_VLAN);
986 			goto exists;
987 		}
988 		TRUNK_LOCK_INIT(trunk);
989 		TRUNK_LOCK(trunk);
990 		p->if_vlantrunk = trunk;
991 		trunk->parent = p;
992 	} else {
993 		VLAN_LOCK();
994 exists:
995 		trunk = p->if_vlantrunk;
996 		TRUNK_LOCK(trunk);
997 	}
998 
999 	ifv->ifv_tag = tag;	/* must set this before vlan_inshash() */
1000 #ifdef VLAN_ARRAY
1001 	if (trunk->vlans[tag] != NULL) {
1002 		error = EEXIST;
1003 		goto done;
1004 	}
1005 	trunk->vlans[tag] = ifv;
1006 	trunk->refcnt++;
1007 #else
1008 	error = vlan_inshash(trunk, ifv);
1009 	if (error)
1010 		goto done;
1011 #endif
1012 	ifv->ifv_proto = ETHERTYPE_VLAN;
1013 	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1014 	ifv->ifv_mintu = ETHERMIN;
1015 	ifv->ifv_pflags = 0;
1016 
1017 	/*
1018 	 * If the parent supports the VLAN_MTU capability,
1019 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1020 	 * use it.
1021 	 */
1022 	if (p->if_capenable & IFCAP_VLAN_MTU) {
1023 		/*
1024 		 * No need to fudge the MTU since the parent can
1025 		 * handle extended frames.
1026 		 */
1027 		ifv->ifv_mtufudge = 0;
1028 	} else {
1029 		/*
1030 		 * Fudge the MTU by the encapsulation size.  This
1031 		 * makes us incompatible with strictly compliant
1032 		 * 802.1Q implementations, but allows us to use
1033 		 * the feature with other NetBSD implementations,
1034 		 * which might still be useful.
1035 		 */
1036 		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1037 	}
1038 
1039 	ifv->ifv_trunk = trunk;
1040 	ifp = ifv->ifv_ifp;
1041 	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1042 	ifp->if_baudrate = p->if_baudrate;
1043 	/*
1044 	 * Copy only a selected subset of flags from the parent.
1045 	 * Other flags are none of our business.
1046 	 */
1047 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1048 	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1049 	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1050 #undef VLAN_COPY_FLAGS
1051 
1052 	ifp->if_link_state = p->if_link_state;
1053 
1054 	vlan_capabilities(ifv);
1055 
1056 	/*
1057 	 * Set up our ``Ethernet address'' to reflect the underlying
1058 	 * physical interface's.
1059 	 */
1060 	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), ETHER_ADDR_LEN);
1061 
1062 	/*
1063 	 * Configure multicast addresses that may already be
1064 	 * joined on the vlan device.
1065 	 */
1066 	(void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1067 
1068 	/* We are ready for operation now. */
1069 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1070 done:
1071 	TRUNK_UNLOCK(trunk);
1072 	if (error == 0)
1073 		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_tag);
1074 	VLAN_UNLOCK();
1075 
1076 	return (error);
1077 }
1078 
1079 static int
1080 vlan_unconfig(struct ifnet *ifp)
1081 {
1082 	int ret;
1083 
1084 	VLAN_LOCK();
1085 	ret = vlan_unconfig_locked(ifp);
1086 	VLAN_UNLOCK();
1087 	return (ret);
1088 }
1089 
1090 static int
1091 vlan_unconfig_locked(struct ifnet *ifp)
1092 {
1093 	struct ifvlantrunk *trunk;
1094 	struct vlan_mc_entry *mc;
1095 	struct ifvlan *ifv;
1096 	struct ifnet  *parent;
1097 	int error;
1098 
1099 	VLAN_LOCK_ASSERT();
1100 
1101 	ifv = ifp->if_softc;
1102 	trunk = ifv->ifv_trunk;
1103 	parent = NULL;
1104 
1105 	if (trunk != NULL) {
1106 		struct sockaddr_dl sdl;
1107 
1108 		TRUNK_LOCK(trunk);
1109 		parent = trunk->parent;
1110 
1111 		/*
1112 		 * Since the interface is being unconfigured, we need to
1113 		 * empty the list of multicast groups that we may have joined
1114 		 * while we were alive from the parent's list.
1115 		 */
1116 		bzero((char *)&sdl, sizeof(sdl));
1117 		sdl.sdl_len = sizeof(sdl);
1118 		sdl.sdl_family = AF_LINK;
1119 		sdl.sdl_index = parent->if_index;
1120 		sdl.sdl_type = IFT_ETHER;
1121 		sdl.sdl_alen = ETHER_ADDR_LEN;
1122 
1123 		while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1124 			bcopy((char *)&mc->mc_addr, LLADDR(&sdl),
1125 			    ETHER_ADDR_LEN);
1126 			error = if_delmulti(parent, (struct sockaddr *)&sdl);
1127 			if (error)
1128 				return (error);
1129 			SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1130 			free(mc, M_VLAN);
1131 		}
1132 
1133 		vlan_setflags(ifp, 0); /* clear special flags on parent */
1134 #ifdef VLAN_ARRAY
1135 		trunk->vlans[ifv->ifv_tag] = NULL;
1136 		trunk->refcnt--;
1137 #else
1138 		vlan_remhash(trunk, ifv);
1139 #endif
1140 		ifv->ifv_trunk = NULL;
1141 
1142 		/*
1143 		 * Check if we were the last.
1144 		 */
1145 		if (trunk->refcnt == 0) {
1146 			trunk->parent->if_vlantrunk = NULL;
1147 			/*
1148 			 * XXXGL: If some ithread has already entered
1149 			 * vlan_input() and is now blocked on the trunk
1150 			 * lock, then it should preempt us right after
1151 			 * unlock and finish its work. Then we will acquire
1152 			 * lock again in trunk_destroy().
1153 			 */
1154 			TRUNK_UNLOCK(trunk);
1155 			trunk_destroy(trunk);
1156 		} else
1157 			TRUNK_UNLOCK(trunk);
1158 	}
1159 
1160 	/* Disconnect from parent. */
1161 	if (ifv->ifv_pflags)
1162 		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1163 	ifp->if_mtu = ETHERMTU;
1164 	ifp->if_link_state = LINK_STATE_UNKNOWN;
1165 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1166 
1167 	/*
1168 	 * Only dispatch an event if vlan was
1169 	 * attached, otherwise there is nothing
1170 	 * to cleanup anyway.
1171 	 */
1172 	if (parent != NULL)
1173 		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_tag);
1174 
1175 	return (0);
1176 }
1177 
1178 /* Handle a reference counted flag that should be set on the parent as well */
1179 static int
1180 vlan_setflag(struct ifnet *ifp, int flag, int status,
1181 	     int (*func)(struct ifnet *, int))
1182 {
1183 	struct ifvlan *ifv;
1184 	int error;
1185 
1186 	/* XXX VLAN_LOCK_ASSERT(); */
1187 
1188 	ifv = ifp->if_softc;
1189 	status = status ? (ifp->if_flags & flag) : 0;
1190 	/* Now "status" contains the flag value or 0 */
1191 
1192 	/*
1193 	 * See if recorded parent's status is different from what
1194 	 * we want it to be.  If it is, flip it.  We record parent's
1195 	 * status in ifv_pflags so that we won't clear parent's flag
1196 	 * we haven't set.  In fact, we don't clear or set parent's
1197 	 * flags directly, but get or release references to them.
1198 	 * That's why we can be sure that recorded flags still are
1199 	 * in accord with actual parent's flags.
1200 	 */
1201 	if (status != (ifv->ifv_pflags & flag)) {
1202 		error = (*func)(PARENT(ifv), status);
1203 		if (error)
1204 			return (error);
1205 		ifv->ifv_pflags &= ~flag;
1206 		ifv->ifv_pflags |= status;
1207 	}
1208 	return (0);
1209 }
1210 
1211 /*
1212  * Handle IFF_* flags that require certain changes on the parent:
1213  * if "status" is true, update parent's flags respective to our if_flags;
1214  * if "status" is false, forcedly clear the flags set on parent.
1215  */
1216 static int
1217 vlan_setflags(struct ifnet *ifp, int status)
1218 {
1219 	int error, i;
1220 
1221 	for (i = 0; vlan_pflags[i].flag; i++) {
1222 		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1223 				     status, vlan_pflags[i].func);
1224 		if (error)
1225 			return (error);
1226 	}
1227 	return (0);
1228 }
1229 
1230 /* Inform all vlans that their parent has changed link state */
1231 static void
1232 vlan_link_state(struct ifnet *ifp, int link)
1233 {
1234 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1235 	struct ifvlan *ifv;
1236 	int i;
1237 
1238 	TRUNK_LOCK(trunk);
1239 #ifdef VLAN_ARRAY
1240 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1241 		if (trunk->vlans[i] != NULL) {
1242 			ifv = trunk->vlans[i];
1243 #else
1244 	for (i = 0; i < (1 << trunk->hwidth); i++)
1245 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1246 #endif
1247 			ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1248 			if_link_state_change(ifv->ifv_ifp,
1249 			    trunk->parent->if_link_state);
1250 		}
1251 	TRUNK_UNLOCK(trunk);
1252 }
1253 
1254 static void
1255 vlan_capabilities(struct ifvlan *ifv)
1256 {
1257 	struct ifnet *p = PARENT(ifv);
1258 	struct ifnet *ifp = ifv->ifv_ifp;
1259 
1260 	TRUNK_LOCK_ASSERT(TRUNK(ifv));
1261 
1262 	/*
1263 	 * If the parent interface can do checksum offloading
1264 	 * on VLANs, then propagate its hardware-assisted
1265 	 * checksumming flags. Also assert that checksum
1266 	 * offloading requires hardware VLAN tagging.
1267 	 */
1268 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1269 		ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1270 
1271 	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1272 	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1273 		ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1274 		ifp->if_hwassist = p->if_hwassist;
1275 	} else {
1276 		ifp->if_capenable = 0;
1277 		ifp->if_hwassist = 0;
1278 	}
1279 }
1280 
1281 static void
1282 vlan_trunk_capabilities(struct ifnet *ifp)
1283 {
1284 	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1285 	struct ifvlan *ifv;
1286 	int i;
1287 
1288 	TRUNK_LOCK(trunk);
1289 #ifdef VLAN_ARRAY
1290 	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1291 		if (trunk->vlans[i] != NULL) {
1292 			ifv = trunk->vlans[i];
1293 #else
1294 	for (i = 0; i < (1 << trunk->hwidth); i++) {
1295 		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1296 #endif
1297 			vlan_capabilities(ifv);
1298 	}
1299 	TRUNK_UNLOCK(trunk);
1300 }
1301 
1302 static int
1303 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1304 {
1305 	struct ifnet *p;
1306 	struct ifreq *ifr;
1307 	struct ifvlan *ifv;
1308 	struct vlanreq vlr;
1309 	int error = 0;
1310 
1311 	ifr = (struct ifreq *)data;
1312 	ifv = ifp->if_softc;
1313 
1314 	switch (cmd) {
1315 	case SIOCGIFMEDIA:
1316 		VLAN_LOCK();
1317 		if (TRUNK(ifv) != NULL) {
1318 			error = (*PARENT(ifv)->if_ioctl)(PARENT(ifv),
1319 					SIOCGIFMEDIA, data);
1320 			VLAN_UNLOCK();
1321 			/* Limit the result to the parent's current config. */
1322 			if (error == 0) {
1323 				struct ifmediareq *ifmr;
1324 
1325 				ifmr = (struct ifmediareq *)data;
1326 				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1327 					ifmr->ifm_count = 1;
1328 					error = copyout(&ifmr->ifm_current,
1329 						ifmr->ifm_ulist,
1330 						sizeof(int));
1331 				}
1332 			}
1333 		} else {
1334 			VLAN_UNLOCK();
1335 			error = EINVAL;
1336 		}
1337 		break;
1338 
1339 	case SIOCSIFMEDIA:
1340 		error = EINVAL;
1341 		break;
1342 
1343 	case SIOCSIFMTU:
1344 		/*
1345 		 * Set the interface MTU.
1346 		 */
1347 		VLAN_LOCK();
1348 		if (TRUNK(ifv) != NULL) {
1349 			if (ifr->ifr_mtu >
1350 			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1351 			    ifr->ifr_mtu <
1352 			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1353 				error = EINVAL;
1354 			else
1355 				ifp->if_mtu = ifr->ifr_mtu;
1356 		} else
1357 			error = EINVAL;
1358 		VLAN_UNLOCK();
1359 		break;
1360 
1361 	case SIOCSETVLAN:
1362 		error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1363 		if (error)
1364 			break;
1365 		if (vlr.vlr_parent[0] == '\0') {
1366 			vlan_unconfig(ifp);
1367 			break;
1368 		}
1369 		p = ifunit(vlr.vlr_parent);
1370 		if (p == 0) {
1371 			error = ENOENT;
1372 			break;
1373 		}
1374 		/*
1375 		 * Don't let the caller set up a VLAN tag with
1376 		 * anything except VLID bits.
1377 		 */
1378 		if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1379 			error = EINVAL;
1380 			break;
1381 		}
1382 		error = vlan_config(ifv, p, vlr.vlr_tag);
1383 		if (error)
1384 			break;
1385 
1386 		/* Update flags on the parent, if necessary. */
1387 		vlan_setflags(ifp, 1);
1388 		break;
1389 
1390 	case SIOCGETVLAN:
1391 		bzero(&vlr, sizeof(vlr));
1392 		VLAN_LOCK();
1393 		if (TRUNK(ifv) != NULL) {
1394 			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1395 			    sizeof(vlr.vlr_parent));
1396 			vlr.vlr_tag = ifv->ifv_tag;
1397 		}
1398 		VLAN_UNLOCK();
1399 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1400 		break;
1401 
1402 	case SIOCSIFFLAGS:
1403 		/*
1404 		 * We should propagate selected flags to the parent,
1405 		 * e.g., promiscuous mode.
1406 		 */
1407 		if (TRUNK(ifv) != NULL)
1408 			error = vlan_setflags(ifp, 1);
1409 		break;
1410 
1411 	case SIOCADDMULTI:
1412 	case SIOCDELMULTI:
1413 		/*
1414 		 * If we don't have a parent, just remember the membership for
1415 		 * when we do.
1416 		 */
1417 		if (TRUNK(ifv) != NULL)
1418 			error = vlan_setmulti(ifp);
1419 		break;
1420 
1421 	default:
1422 		error = ether_ioctl(ifp, cmd, data);
1423 	}
1424 
1425 	return (error);
1426 }
1427