if_vlan.c revision 161321
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: head/sys/net/if_vlan.c 161321 2006-08-15 16:40:27Z yar $
30 */
31
32/*
33 * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
34 * Might be extended some day to also handle IEEE 802.1p priority
35 * tagging.  This is sort of sneaky in the implementation, since
36 * we need to pretend to be enough of an Ethernet implementation
37 * to make arp work.  The way we do this is by telling everyone
38 * that we are an Ethernet, and then catch the packets that
39 * ether_output() left on our output queue when it calls
40 * if_start(), rewrite them for use by the real outgoing interface,
41 * and ask it to send them.
42 */
43
44#include "opt_inet.h"
45#include "opt_vlan.h"
46
47#include <sys/param.h>
48#include <sys/kernel.h>
49#include <sys/lock.h>
50#include <sys/malloc.h>
51#include <sys/mbuf.h>
52#include <sys/module.h>
53#include <sys/rwlock.h>
54#include <sys/queue.h>
55#include <sys/socket.h>
56#include <sys/sockio.h>
57#include <sys/sysctl.h>
58#include <sys/systm.h>
59
60#include <net/bpf.h>
61#include <net/ethernet.h>
62#include <net/if.h>
63#include <net/if_clone.h>
64#include <net/if_arp.h>
65#include <net/if_dl.h>
66#include <net/if_types.h>
67#include <net/if_vlan_var.h>
68
69#ifdef INET
70#include <netinet/in.h>
71#include <netinet/if_ether.h>
72#endif
73
74#define VLANNAME	"vlan"
75#define	VLAN_DEF_HWIDTH	4
76#define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
77
78LIST_HEAD(ifvlanhead, ifvlan);
79
80struct ifvlantrunk {
81	struct	ifnet   *parent;	/* parent interface of this trunk */
82	struct	rwlock	rw;
83#ifdef VLAN_ARRAY
84#define	VLAN_ARRAY_SIZE	(EVL_VLID_MASK + 1)
85	struct	ifvlan	*vlans[VLAN_ARRAY_SIZE]; /* static table */
86#else
87	struct	ifvlanhead *hash;	/* dynamic hash-list table */
88	uint16_t	hmask;
89	uint16_t	hwidth;
90#endif
91	int		refcnt;
92	LIST_ENTRY(ifvlantrunk) trunk_entry;
93};
94static LIST_HEAD(, ifvlantrunk) trunk_list;
95
96struct vlan_mc_entry {
97	struct ether_addr		mc_addr;
98	SLIST_ENTRY(vlan_mc_entry)	mc_entries;
99};
100
101struct	ifvlan {
102	struct	ifvlantrunk *ifv_trunk;
103	struct	ifnet *ifv_ifp;
104#define	TRUNK(ifv)	((ifv)->ifv_trunk)
105#define	PARENT(ifv)	((ifv)->ifv_trunk->parent)
106	int	ifv_pflags;	/* special flags we have set on parent */
107	struct	ifv_linkmib {
108		int	ifvm_encaplen;	/* encapsulation length */
109		int	ifvm_mtufudge;	/* MTU fudged by this much */
110		int	ifvm_mintu;	/* min transmission unit */
111		uint16_t ifvm_tag;	/* tag to apply on packets leaving if */
112	}	ifv_mib;
113	SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
114	LIST_ENTRY(ifvlan) ifv_list;
115};
116#define	ifv_tag		ifv_mib.ifvm_tag
117#define	ifv_encaplen	ifv_mib.ifvm_encaplen
118#define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
119#define	ifv_mintu	ifv_mib.ifvm_mintu
120
121/* Special flags we should propagate to parent. */
122static 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
131SYSCTL_DECL(_net_link);
132SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0, "IEEE 802.1Q VLAN");
133SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0, "for consistency");
134
135static int soft_pad = 0;
136SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW, &soft_pad, 0,
137	   "pad short frames before tagging");
138
139static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface");
140
141static 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 boots in processing,
152 * however on practice it does not. Probably this is because array
153 * is too big to fit into CPU cache.
154 */
155static 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
171static	void vlan_inithash(struct ifvlantrunk *trunk);
172static	void vlan_freehash(struct ifvlantrunk *trunk);
173static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
174static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
175static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
176static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
177	uint16_t tag);
178#endif
179static	void trunk_destroy(struct ifvlantrunk *trunk);
180
181static	void vlan_start(struct ifnet *ifp);
182static	void vlan_init(void *foo);
183static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
184static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
185static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
186    int (*func)(struct ifnet *, int));
187static	int vlan_setflags(struct ifnet *ifp, int status);
188static	int vlan_setmulti(struct ifnet *ifp);
189static	int vlan_unconfig(struct ifnet *ifp);
190static	int vlan_unconfig_locked(struct ifnet *ifp);
191static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
192static	void vlan_link_state(struct ifnet *ifp, int link);
193static	void vlan_capabilities(struct ifvlan *ifv);
194static	void vlan_trunk_capabilities(struct ifnet *ifp);
195
196static	struct ifnet *vlan_clone_match_ethertag(struct if_clone *,
197    const char *, int *);
198static	int vlan_clone_match(struct if_clone *, const char *);
199static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
200static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
201
202static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
203
204static	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
210static void
211vlan_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
232static void
233vlan_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
248static int
249vlan_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
278static int
279vlan_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 */
305static void
306vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
307{
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 (!LIST_EMPTY(&trunk->hash[i])) {
340			ifv = LIST_FIRST(&trunk->hash[i]);
341			LIST_REMOVE(ifv, ifv_list);
342			j = HASH(ifv->ifv_tag, n2 - 1);
343			LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
344		}
345	free(trunk->hash, M_VLAN);
346	trunk->hash = hash2;
347	trunk->hwidth = hwidth2;
348	trunk->hmask = n2 - 1;
349}
350
351static __inline struct ifvlan *
352vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
353{
354	struct ifvlan *ifv;
355
356	TRUNK_LOCK_RASSERT(trunk);
357
358	LIST_FOREACH(ifv, &trunk->hash[HASH(tag, trunk->hmask)], ifv_list)
359		if (ifv->ifv_tag == tag)
360			return (ifv);
361	return (NULL);
362}
363
364#if 0
365/* Debugging code to view the hashtables. */
366static void
367vlan_dumphash(struct ifvlantrunk *trunk)
368{
369	int i;
370	struct ifvlan *ifv;
371
372	for (i = 0; i < (1 << trunk->hwidth); i++) {
373		printf("%d: ", i);
374		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
375			printf("%s ", ifv->ifv_ifp->if_xname);
376		printf("\n");
377	}
378}
379#endif /* 0 */
380#endif /* !VLAN_ARRAY */
381
382static void
383trunk_destroy(struct ifvlantrunk *trunk)
384{
385	VLAN_LOCK_ASSERT();
386
387	TRUNK_LOCK(trunk);
388#ifndef VLAN_ARRAY
389	vlan_freehash(trunk);
390#endif
391	trunk->parent->if_vlantrunk = NULL;
392	LIST_REMOVE(trunk, trunk_entry);
393	TRUNK_UNLOCK(trunk);
394	TRUNK_LOCK_DESTROY(trunk);
395	free(trunk, M_VLAN);
396}
397
398/*
399 * Program our multicast filter. What we're actually doing is
400 * programming the multicast filter of the parent. This has the
401 * side effect of causing the parent interface to receive multicast
402 * traffic that it doesn't really want, which ends up being discarded
403 * later by the upper protocol layers. Unfortunately, there's no way
404 * to avoid this: there really is only one physical interface.
405 *
406 * XXX: There is a possible race here if more than one thread is
407 *      modifying the multicast state of the vlan interface at the same time.
408 */
409static int
410vlan_setmulti(struct ifnet *ifp)
411{
412	struct ifnet		*ifp_p;
413	struct ifmultiaddr	*ifma, *rifma = NULL;
414	struct ifvlan		*sc;
415	struct vlan_mc_entry	*mc = NULL;
416	struct sockaddr_dl	sdl;
417	int			error;
418
419	/*VLAN_LOCK_ASSERT();*/
420
421	/* Find the parent. */
422	sc = ifp->if_softc;
423	ifp_p = PARENT(sc);
424
425	bzero((char *)&sdl, sizeof(sdl));
426	sdl.sdl_len = sizeof(sdl);
427	sdl.sdl_family = AF_LINK;
428	sdl.sdl_index = ifp_p->if_index;
429	sdl.sdl_type = IFT_ETHER;
430	sdl.sdl_alen = ETHER_ADDR_LEN;
431
432	/* First, remove any existing filter entries. */
433	while (SLIST_FIRST(&sc->vlan_mc_listhead) != NULL) {
434		mc = SLIST_FIRST(&sc->vlan_mc_listhead);
435		bcopy((char *)&mc->mc_addr, LLADDR(&sdl), ETHER_ADDR_LEN);
436		error = if_delmulti(ifp_p, (struct sockaddr *)&sdl);
437		if (error)
438			return (error);
439		SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
440		free(mc, M_VLAN);
441	}
442
443	/* Now program new ones. */
444	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
445		if (ifma->ifma_addr->sa_family != AF_LINK)
446			continue;
447		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
448		if (mc == NULL)
449			return (ENOMEM);
450		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
451		    (char *)&mc->mc_addr, ETHER_ADDR_LEN);
452		SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
453		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
454		    LLADDR(&sdl), ETHER_ADDR_LEN);
455		error = if_addmulti(ifp_p, (struct sockaddr *)&sdl, &rifma);
456		if (error)
457			return (error);
458	}
459
460	return (0);
461}
462
463/*
464 * A handler for network interface departure events.
465 * Track departure of trunks here so that we don't access invalid
466 * pointers or whatever if a trunk is ripped from under us, e.g.,
467 * by ejecting its hot-plug card.
468 */
469static void
470vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
471{
472	struct ifvlan *ifv;
473	int i;
474
475	/*
476	 * Check if it's a trunk interface first of all
477	 * to avoid needless locking.
478	 */
479	if (ifp->if_vlantrunk == NULL)
480		return;
481
482	VLAN_LOCK();
483	/*
484	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
485	 * Check trunk pointer after each vlan_unconfig() as it will
486	 * free it and set to NULL after the last vlan was detached.
487	 */
488#ifdef VLAN_ARRAY
489	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
490		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
491			vlan_unconfig_locked(ifv->ifv_ifp);
492			if (ifp->if_vlantrunk == NULL)
493				break;
494		}
495#else /* VLAN_ARRAY */
496restart:
497	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
498		if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
499			vlan_unconfig_locked(ifv->ifv_ifp);
500			if (ifp->if_vlantrunk)
501				goto restart;	/* trunk->hwidth can change */
502			else
503				break;
504		}
505#endif /* VLAN_ARRAY */
506	/* Trunk should have been destroyed in vlan_unconfig(). */
507	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
508	VLAN_UNLOCK();
509}
510
511/*
512 * VLAN support can be loaded as a module.  The only place in the
513 * system that's intimately aware of this is ether_input.  We hook
514 * into this code through vlan_input_p which is defined there and
515 * set here.  Noone else in the system should be aware of this so
516 * we use an explicit reference here.
517 */
518extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
519
520/* For if_link_state_change() eyes only... */
521extern	void (*vlan_link_state_p)(struct ifnet *, int);
522
523static int
524vlan_modevent(module_t mod, int type, void *data)
525{
526
527	switch (type) {
528	case MOD_LOAD:
529		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
530		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
531		if (ifdetach_tag == NULL)
532			return (ENOMEM);
533		LIST_INIT(&trunk_list);
534		VLAN_LOCK_INIT();
535		vlan_input_p = vlan_input;
536		vlan_link_state_p = vlan_link_state;
537		vlan_trunk_cap_p = vlan_trunk_capabilities;
538		if_clone_attach(&vlan_cloner);
539		break;
540	case MOD_UNLOAD:
541	    {
542		struct ifvlantrunk *trunk, *trunk1;
543
544		if_clone_detach(&vlan_cloner);
545		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
546		vlan_input_p = NULL;
547		vlan_link_state_p = NULL;
548		vlan_trunk_cap_p = NULL;
549		VLAN_LOCK();
550		LIST_FOREACH_SAFE(trunk, &trunk_list, trunk_entry, trunk1)
551			trunk_destroy(trunk);
552		VLAN_UNLOCK();
553		VLAN_LOCK_DESTROY();
554		break;
555	    }
556	default:
557		return (EOPNOTSUPP);
558	}
559	return (0);
560}
561
562static moduledata_t vlan_mod = {
563	"if_vlan",
564	vlan_modevent,
565	0
566};
567
568DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
569MODULE_VERSION(if_vlan, 3);
570MODULE_DEPEND(if_vlan, miibus, 1, 1, 1);
571
572static struct ifnet *
573vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
574{
575	const char *cp;
576	struct ifnet *ifp;
577	int t = 0;
578
579	/* Check for <etherif>.<vlan> style interface names. */
580	IFNET_RLOCK();
581	TAILQ_FOREACH(ifp, &ifnet, if_link) {
582		if (ifp->if_type != IFT_ETHER)
583			continue;
584		if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
585			continue;
586		cp = name + strlen(ifp->if_xname);
587		if (*cp != '.')
588			continue;
589		for(; *cp != '\0'; cp++) {
590			if (*cp < '0' || *cp > '9')
591				continue;
592			t = (t * 10) + (*cp - '0');
593		}
594		if (tag != NULL)
595			*tag = t;
596		break;
597	}
598	IFNET_RUNLOCK();
599
600	return (ifp);
601}
602
603static int
604vlan_clone_match(struct if_clone *ifc, const char *name)
605{
606	const char *cp;
607
608	if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
609		return (1);
610
611	if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
612		return (0);
613	for (cp = name + 4; *cp != '\0'; cp++) {
614		if (*cp < '0' || *cp > '9')
615			return (0);
616	}
617
618	return (1);
619}
620
621static int
622vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
623{
624	char *dp;
625	int wildcard;
626	int unit;
627	int error;
628	int tag;
629	int ethertag;
630	struct ifvlan *ifv;
631	struct ifnet *ifp;
632	struct ifnet *p;
633	struct vlanreq vlr;
634	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
635
636	/*
637	 * There are 3 (ugh) ways to specify the cloned device:
638	 * o pass a parameter block with the clone request.
639	 * o specify parameters in the text of the clone device name
640	 * o specify no parameters and get an unattached device that
641	 *   must be configured separately.
642	 * The first technique is preferred; the latter two are
643	 * supported for backwards compatibilty.
644	 */
645	if (params) {
646		error = copyin(params, &vlr, sizeof(vlr));
647		if (error)
648			return error;
649		p = ifunit(vlr.vlr_parent);
650		if (p == NULL)
651			return ENXIO;
652		/*
653		 * Don't let the caller set up a VLAN tag with
654		 * anything except VLID bits.
655		 */
656		if (vlr.vlr_tag & ~EVL_VLID_MASK)
657			return (EINVAL);
658		error = ifc_name2unit(name, &unit);
659		if (error != 0)
660			return (error);
661
662		ethertag = 1;
663		tag = vlr.vlr_tag;
664		wildcard = (unit < 0);
665	} else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
666		ethertag = 1;
667		unit = -1;
668		wildcard = 0;
669
670		/*
671		 * Don't let the caller set up a VLAN tag with
672		 * anything except VLID bits.
673		 */
674		if (tag & ~EVL_VLID_MASK)
675			return (EINVAL);
676	} else {
677		ethertag = 0;
678
679		error = ifc_name2unit(name, &unit);
680		if (error != 0)
681			return (error);
682
683		wildcard = (unit < 0);
684	}
685
686	error = ifc_alloc_unit(ifc, &unit);
687	if (error != 0)
688		return (error);
689
690	/* In the wildcard case, we need to update the name. */
691	if (wildcard) {
692		for (dp = name; *dp != '\0'; dp++);
693		if (snprintf(dp, len - (dp-name), "%d", unit) >
694		    len - (dp-name) - 1) {
695			panic("%s: interface name too long", __func__);
696		}
697	}
698
699	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
700	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
701	if (ifp == NULL) {
702		ifc_free_unit(ifc, unit);
703		free(ifv, M_VLAN);
704		return (ENOSPC);
705	}
706	SLIST_INIT(&ifv->vlan_mc_listhead);
707
708	ifp->if_softc = ifv;
709	/*
710	 * Set the name manually rather than using if_initname because
711	 * we don't conform to the default naming convention for interfaces.
712	 */
713	strlcpy(ifp->if_xname, name, IFNAMSIZ);
714	ifp->if_dname = ifc->ifc_name;
715	ifp->if_dunit = unit;
716	/* NB: flags are not set here */
717	ifp->if_linkmib = &ifv->ifv_mib;
718	ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
719	/* NB: mtu is not set here */
720
721	ifp->if_init = vlan_init;
722	ifp->if_start = vlan_start;
723	ifp->if_ioctl = vlan_ioctl;
724	ifp->if_snd.ifq_maxlen = ifqmaxlen;
725	ifp->if_flags = VLAN_IFFLAGS;
726	ether_ifattach(ifp, eaddr);
727	/* Now undo some of the damage... */
728	ifp->if_baudrate = 0;
729	ifp->if_type = IFT_L2VLAN;
730	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
731
732	if (ethertag) {
733		error = vlan_config(ifv, p, tag);
734		if (error != 0) {
735			/*
736			 * Since we've partialy failed, we need to back
737			 * out all the way, otherwise userland could get
738			 * confused.  Thus, we destroy the interface.
739			 */
740			ether_ifdetach(ifp);
741			vlan_unconfig(ifp);
742			if_free_type(ifp, IFT_ETHER);
743			free(ifv, M_VLAN);
744
745			return (error);
746		}
747
748		/* Update flags on the parent, if necessary. */
749		vlan_setflags(ifp, 1);
750	}
751
752	return (0);
753}
754
755static int
756vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
757{
758	struct ifvlan *ifv = ifp->if_softc;
759	int unit = ifp->if_dunit;
760
761	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
762	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
763	if_free_type(ifp, IFT_ETHER);
764	free(ifv, M_VLAN);
765	ifc_free_unit(ifc, unit);
766
767	return (0);
768}
769
770/*
771 * The ifp->if_init entry point for vlan(4) is a no-op.
772 */
773static void
774vlan_init(void *foo __unused)
775{
776}
777
778/*
779 * The if_start method for vlan(4) interface. It doesn't
780 * raises the IFF_DRV_OACTIVE flag, since it is called
781 * only from IFQ_HANDOFF() macro in ether_output_frame().
782 * If the interface queue is full, and vlan_start() is
783 * not called, the queue would never get emptied and
784 * interface would stall forever.
785 */
786static void
787vlan_start(struct ifnet *ifp)
788{
789	struct ifvlan *ifv;
790	struct ifnet *p;
791	struct mbuf *m;
792	int error;
793
794	ifv = ifp->if_softc;
795	p = PARENT(ifv);
796
797	for (;;) {
798		IF_DEQUEUE(&ifp->if_snd, m);
799		if (m == 0)
800			break;
801		BPF_MTAP(ifp, m);
802
803		/*
804		 * Do not run parent's if_start() if the parent is not up,
805		 * or parent's driver will cause a system crash.
806		 */
807		if (!((p->if_flags & IFF_UP) &&
808		    (p->if_drv_flags & IFF_DRV_RUNNING))) {
809			m_freem(m);
810			ifp->if_collisions++;
811			continue;
812		}
813
814		/*
815		 * Pad the frame to the minimum size allowed if told to.
816		 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
817		 * paragraph C.4.4.3.b.  It can help to work around buggy
818		 * bridges that violate paragraph C.4.4.3.a from the same
819		 * document, i.e., fail to pad short frames after untagging.
820		 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
821		 * untagging it will produce a 62-byte frame, which is a runt
822		 * and requires padding.  There are VLAN-enabled network
823		 * devices that just discard such runts instead or mishandle
824		 * them somehow.
825		 */
826		if (soft_pad) {
827			static char pad[8];	/* just zeros */
828			int n;
829
830			for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
831			     n > 0; n -= sizeof(pad))
832				if (!m_append(m, min(n, sizeof(pad)), pad))
833					break;
834
835			if (n > 0) {
836				if_printf(ifp, "cannot pad short frame\n");
837				ifp->if_oerrors++;
838				m_freem(m);
839				continue;
840			}
841		}
842
843		/*
844		 * If underlying interface can do VLAN tag insertion itself,
845		 * just pass the packet along. However, we need some way to
846		 * tell the interface where the packet came from so that it
847		 * knows how to find the VLAN tag to use, so we attach a
848		 * packet tag that holds it.
849		 */
850		if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
851			struct m_tag *mtag = (struct m_tag *)
852			    uma_zalloc(zone_mtag_vlan, M_NOWAIT);
853			if (mtag == NULL) {
854				ifp->if_oerrors++;
855				m_freem(m);
856				continue;
857			}
858			VLAN_TAG_VALUE(mtag) = ifv->ifv_tag;
859			m_tag_prepend(m, mtag);
860			m->m_flags |= M_VLANTAG;
861		} else {
862			struct ether_vlan_header *evl;
863
864			M_PREPEND(m, ifv->ifv_encaplen, M_DONTWAIT);
865			if (m == NULL) {
866				if_printf(ifp,
867				    "unable to prepend VLAN header\n");
868				ifp->if_oerrors++;
869				continue;
870			}
871			/* M_PREPEND takes care of m_len, m_pkthdr.len for us */
872
873			if (m->m_len < sizeof(*evl)) {
874				m = m_pullup(m, sizeof(*evl));
875				if (m == NULL) {
876					if_printf(ifp,
877					    "cannot pullup VLAN header\n");
878					ifp->if_oerrors++;
879					continue;
880				}
881			}
882
883			/*
884			 * Transform the Ethernet header into an Ethernet header
885			 * with 802.1Q encapsulation.
886			 */
887			bcopy(mtod(m, char *) + ifv->ifv_encaplen,
888			      mtod(m, char *), ETHER_HDR_LEN);
889			evl = mtod(m, struct ether_vlan_header *);
890			evl->evl_proto = evl->evl_encap_proto;
891			evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
892			evl->evl_tag = htons(ifv->ifv_tag);
893#ifdef DEBUG
894			printf("%s: %*D\n", __func__, (int)sizeof(*evl),
895			    (unsigned char *)evl, ":");
896#endif
897		}
898
899		/*
900		 * Send it, precisely as ether_output() would have.
901		 * We are already running at splimp.
902		 */
903		IFQ_HANDOFF(p, m, error);
904		if (!error)
905			ifp->if_opackets++;
906		else
907			ifp->if_oerrors++;
908	}
909}
910
911static void
912vlan_input(struct ifnet *ifp, struct mbuf *m)
913{
914	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
915	struct ifvlan *ifv;
916	struct m_tag *mtag;
917	uint16_t tag;
918
919	KASSERT(trunk != NULL, ("%s: no trunk", __func__));
920
921	if (m->m_flags & M_VLANTAG) {
922		/*
923		 * Packet is tagged, but m contains a normal
924		 * Ethernet frame; the tag is stored out-of-band.
925		 */
926		mtag = m_tag_locate(m, MTAG_VLAN, MTAG_VLAN_TAG, NULL);
927		KASSERT(mtag != NULL,
928			("%s: M_VLANTAG without m_tag", __func__));
929		tag = EVL_VLANOFTAG(VLAN_TAG_VALUE(mtag));
930		m_tag_delete(m, mtag);
931		m->m_flags &= ~M_VLANTAG;
932	} else {
933		struct ether_vlan_header *evl;
934
935		/*
936		 * Packet is tagged in-band as specified by 802.1q.
937		 */
938		mtag = NULL;
939		switch (ifp->if_type) {
940		case IFT_ETHER:
941			if (m->m_len < sizeof(*evl) &&
942			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
943				if_printf(ifp, "cannot pullup VLAN header\n");
944				return;
945			}
946			evl = mtod(m, struct ether_vlan_header *);
947			KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN,
948				("%s: bad encapsulation protocol (%u)",
949				 __func__, ntohs(evl->evl_encap_proto)));
950
951			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
952
953			/*
954			 * Restore the original ethertype.  We'll remove
955			 * the encapsulation after we've found the vlan
956			 * interface corresponding to the tag.
957			 */
958			evl->evl_encap_proto = evl->evl_proto;
959			break;
960		default:
961#ifdef INVARIANTS
962			panic("%s: %s has unsupported if_type %u",
963			      __func__, ifp->if_xname, ifp->if_type);
964#endif
965			m_freem(m);
966			ifp->if_noproto++;
967			return;
968		}
969	}
970
971	TRUNK_RLOCK(trunk);
972#ifdef VLAN_ARRAY
973	ifv = trunk->vlans[tag];
974#else
975	ifv = vlan_gethash(trunk, tag);
976#endif
977	if (ifv == NULL || (ifv->ifv_ifp->if_flags & IFF_UP) == 0) {
978		TRUNK_RUNLOCK(trunk);
979		m_freem(m);
980		ifp->if_noproto++;
981		return;
982	}
983	TRUNK_RUNLOCK(trunk);
984
985	if (mtag == NULL) {
986		/*
987		 * Packet had an in-line encapsulation header;
988		 * remove it.  The original header has already
989		 * been fixed up above.
990		 */
991		bcopy(mtod(m, caddr_t),
992		      mtod(m, caddr_t) + ETHER_VLAN_ENCAP_LEN,
993		      ETHER_HDR_LEN);
994		m_adj(m, ETHER_VLAN_ENCAP_LEN);
995	}
996
997	m->m_pkthdr.rcvif = ifv->ifv_ifp;
998	ifv->ifv_ifp->if_ipackets++;
999
1000	/* Pass it back through the parent's input routine. */
1001	(*ifp->if_input)(ifv->ifv_ifp, m);
1002}
1003
1004static int
1005vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
1006{
1007	struct ifvlantrunk *trunk;
1008	struct ifnet *ifp;
1009	int error = 0;
1010
1011	/* VID numbers 0x0 and 0xFFF are reserved */
1012	if (tag == 0 || tag == 0xFFF)
1013		return (EINVAL);
1014	if (p->if_type != IFT_ETHER)
1015		return (EPROTONOSUPPORT);
1016	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1017		return (EPROTONOSUPPORT);
1018	if (ifv->ifv_trunk)
1019		return (EBUSY);
1020
1021	if (p->if_vlantrunk == NULL) {
1022		trunk = malloc(sizeof(struct ifvlantrunk),
1023		    M_VLAN, M_WAITOK | M_ZERO);
1024#ifndef VLAN_ARRAY
1025		vlan_inithash(trunk);
1026#endif
1027		VLAN_LOCK();
1028		if (p->if_vlantrunk != NULL) {
1029			/* A race that that is very unlikely to be hit. */
1030#ifndef VLAN_ARRAY
1031			vlan_freehash(trunk);
1032#endif
1033			free(trunk, M_VLAN);
1034			goto exists;
1035		}
1036		TRUNK_LOCK_INIT(trunk);
1037		LIST_INSERT_HEAD(&trunk_list, trunk, trunk_entry);
1038		TRUNK_LOCK(trunk);
1039		p->if_vlantrunk = trunk;
1040		trunk->parent = p;
1041	} else {
1042		VLAN_LOCK();
1043exists:
1044		trunk = p->if_vlantrunk;
1045		TRUNK_LOCK(trunk);
1046	}
1047
1048	ifv->ifv_tag = tag;	/* must set this before vlan_inshash() */
1049#ifdef VLAN_ARRAY
1050	if (trunk->vlans[tag] != NULL) {
1051		error = EEXIST;
1052		goto done;
1053	}
1054	trunk->vlans[tag] = ifv;
1055	trunk->refcnt++;
1056#else
1057	error = vlan_inshash(trunk, ifv);
1058	if (error)
1059		goto done;
1060#endif
1061	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1062	ifv->ifv_mintu = ETHERMIN;
1063	ifv->ifv_pflags = 0;
1064
1065	/*
1066	 * If the parent supports the VLAN_MTU capability,
1067	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1068	 * use it.
1069	 */
1070	if (p->if_capenable & IFCAP_VLAN_MTU) {
1071		/*
1072		 * No need to fudge the MTU since the parent can
1073		 * handle extended frames.
1074		 */
1075		ifv->ifv_mtufudge = 0;
1076	} else {
1077		/*
1078		 * Fudge the MTU by the encapsulation size.  This
1079		 * makes us incompatible with strictly compliant
1080		 * 802.1Q implementations, but allows us to use
1081		 * the feature with other NetBSD implementations,
1082		 * which might still be useful.
1083		 */
1084		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1085	}
1086
1087	ifv->ifv_trunk = trunk;
1088	ifp = ifv->ifv_ifp;
1089	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1090	ifp->if_baudrate = p->if_baudrate;
1091	/*
1092	 * Copy only a selected subset of flags from the parent.
1093	 * Other flags are none of our business.
1094	 */
1095#define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1096	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1097	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1098#undef VLAN_COPY_FLAGS
1099
1100	ifp->if_link_state = p->if_link_state;
1101
1102	vlan_capabilities(ifv);
1103
1104	/*
1105	 * Set up our ``Ethernet address'' to reflect the underlying
1106	 * physical interface's.
1107	 */
1108	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), ETHER_ADDR_LEN);
1109
1110	/*
1111	 * Configure multicast addresses that may already be
1112	 * joined on the vlan device.
1113	 */
1114	(void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1115
1116	/* We are ready for operation now. */
1117	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1118done:
1119	TRUNK_UNLOCK(trunk);
1120	VLAN_UNLOCK();
1121
1122	return (error);
1123}
1124
1125static int
1126vlan_unconfig(struct ifnet *ifp)
1127{
1128	int ret;
1129
1130	VLAN_LOCK();
1131	ret = vlan_unconfig_locked(ifp);
1132	VLAN_UNLOCK();
1133	return (ret);
1134}
1135
1136static int
1137vlan_unconfig_locked(struct ifnet *ifp)
1138{
1139	struct ifvlantrunk *trunk;
1140	struct vlan_mc_entry *mc;
1141	struct ifvlan *ifv;
1142	int error;
1143
1144	VLAN_LOCK_ASSERT();
1145
1146	ifv = ifp->if_softc;
1147	trunk = ifv->ifv_trunk;
1148
1149	if (trunk) {
1150		struct sockaddr_dl sdl;
1151		struct ifnet *p = trunk->parent;
1152
1153		TRUNK_LOCK(trunk);
1154
1155		/*
1156		 * Since the interface is being unconfigured, we need to
1157		 * empty the list of multicast groups that we may have joined
1158		 * while we were alive from the parent's list.
1159		 */
1160		bzero((char *)&sdl, sizeof(sdl));
1161		sdl.sdl_len = sizeof(sdl);
1162		sdl.sdl_family = AF_LINK;
1163		sdl.sdl_index = p->if_index;
1164		sdl.sdl_type = IFT_ETHER;
1165		sdl.sdl_alen = ETHER_ADDR_LEN;
1166
1167		while(SLIST_FIRST(&ifv->vlan_mc_listhead) != NULL) {
1168			mc = SLIST_FIRST(&ifv->vlan_mc_listhead);
1169			bcopy((char *)&mc->mc_addr, LLADDR(&sdl),
1170			    ETHER_ADDR_LEN);
1171			error = if_delmulti(p, (struct sockaddr *)&sdl);
1172			if (error)
1173				return (error);
1174			SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1175			free(mc, M_VLAN);
1176		}
1177
1178		vlan_setflags(ifp, 0); /* clear special flags on parent */
1179#ifdef VLAN_ARRAY
1180		trunk->vlans[ifv->ifv_tag] = NULL;
1181		trunk->refcnt--;
1182#else
1183		vlan_remhash(trunk, ifv);
1184#endif
1185		ifv->ifv_trunk = NULL;
1186
1187		/*
1188		 * Check if we were the last.
1189		 */
1190		if (trunk->refcnt == 0) {
1191			trunk->parent->if_vlantrunk = NULL;
1192			/*
1193			 * XXXGL: If some ithread has already entered
1194			 * vlan_input() and is now blocked on the trunk
1195			 * lock, then it should preempt us right after
1196			 * unlock and finish its work. Then we will acquire
1197			 * lock again in trunk_destroy().
1198			 * XXX: not true in case of VLAN_ARRAY
1199			 */
1200			TRUNK_UNLOCK(trunk);
1201			trunk_destroy(trunk);
1202		} else
1203			TRUNK_UNLOCK(trunk);
1204	}
1205
1206	/* Disconnect from parent. */
1207	if (ifv->ifv_pflags)
1208		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1209	ifp->if_mtu = ETHERMTU;
1210	ifp->if_link_state = LINK_STATE_UNKNOWN;
1211	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1212
1213	return (0);
1214}
1215
1216/* Handle a reference counted flag that should be set on the parent as well */
1217static int
1218vlan_setflag(struct ifnet *ifp, int flag, int status,
1219	     int (*func)(struct ifnet *, int))
1220{
1221	struct ifvlan *ifv;
1222	int error;
1223
1224	/* XXX VLAN_LOCK_ASSERT(); */
1225
1226	ifv = ifp->if_softc;
1227	status = status ? (ifp->if_flags & flag) : 0;
1228	/* Now "status" contains the flag value or 0 */
1229
1230	/*
1231	 * See if recorded parent's status is different from what
1232	 * we want it to be.  If it is, flip it.  We record parent's
1233	 * status in ifv_pflags so that we won't clear parent's flag
1234	 * we haven't set.  In fact, we don't clear or set parent's
1235	 * flags directly, but get or release references to them.
1236	 * That's why we can be sure that recorded flags still are
1237	 * in accord with actual parent's flags.
1238	 */
1239	if (status != (ifv->ifv_pflags & flag)) {
1240		error = (*func)(PARENT(ifv), status);
1241		if (error)
1242			return (error);
1243		ifv->ifv_pflags &= ~flag;
1244		ifv->ifv_pflags |= status;
1245	}
1246	return (0);
1247}
1248
1249/*
1250 * Handle IFF_* flags that require certain changes on the parent:
1251 * if "status" is true, update parent's flags respective to our if_flags;
1252 * if "status" is false, forcedly clear the flags set on parent.
1253 */
1254static int
1255vlan_setflags(struct ifnet *ifp, int status)
1256{
1257	int error, i;
1258
1259	for (i = 0; vlan_pflags[i].flag; i++) {
1260		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1261				     status, vlan_pflags[i].func);
1262		if (error)
1263			return (error);
1264	}
1265	return (0);
1266}
1267
1268/* Inform all vlans that their parent has changed link state */
1269static void
1270vlan_link_state(struct ifnet *ifp, int link)
1271{
1272	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1273	struct ifvlan *ifv;
1274	int i;
1275
1276	TRUNK_LOCK(trunk);
1277#ifdef VLAN_ARRAY
1278	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1279		if (trunk->vlans[i] != NULL) {
1280			ifv = trunk->vlans[i];
1281#else
1282	for (i = 0; i < (1 << trunk->hwidth); i++) {
1283		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1284#endif
1285			if_link_state_change(ifv->ifv_ifp,
1286			    trunk->parent->if_link_state);
1287	}
1288	TRUNK_UNLOCK(trunk);
1289}
1290
1291static void
1292vlan_capabilities(struct ifvlan *ifv)
1293{
1294	struct ifnet *p = PARENT(ifv);
1295	struct ifnet *ifp = ifv->ifv_ifp;
1296
1297	TRUNK_LOCK_ASSERT(TRUNK(ifv));
1298
1299	/*
1300	 * If the parent interface can do checksum offloading
1301	 * on VLANs, then propagate its hardware-assisted
1302	 * checksumming flags. Also assert that checksum
1303	 * offloading requires hardware VLAN tagging.
1304	 */
1305	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1306		ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1307
1308	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1309	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1310		ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1311		ifp->if_hwassist = p->if_hwassist;
1312	} else {
1313		ifp->if_capenable = 0;
1314		ifp->if_hwassist = 0;
1315	}
1316}
1317
1318static void
1319vlan_trunk_capabilities(struct ifnet *ifp)
1320{
1321	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1322	struct ifvlan *ifv;
1323	int i;
1324
1325	TRUNK_LOCK(trunk);
1326#ifdef VLAN_ARRAY
1327	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1328		if (trunk->vlans[i] != NULL) {
1329			ifv = trunk->vlans[i];
1330#else
1331	for (i = 0; i < (1 << trunk->hwidth); i++) {
1332		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1333#endif
1334			vlan_capabilities(ifv);
1335	}
1336	TRUNK_UNLOCK(trunk);
1337}
1338
1339static int
1340vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1341{
1342	struct ifaddr *ifa;
1343	struct ifnet *p;
1344	struct ifreq *ifr;
1345	struct ifvlan *ifv;
1346	struct vlanreq vlr;
1347	int error = 0;
1348
1349	ifr = (struct ifreq *)data;
1350	ifa = (struct ifaddr *)data;
1351	ifv = ifp->if_softc;
1352
1353	switch (cmd) {
1354	case SIOCSIFADDR:
1355		ifp->if_flags |= IFF_UP;
1356
1357		switch (ifa->ifa_addr->sa_family) {
1358#ifdef INET
1359		case AF_INET:
1360			arp_ifinit(ifv->ifv_ifp, ifa);
1361			break;
1362#endif
1363		default:
1364			break;
1365		}
1366		break;
1367
1368	case SIOCGIFADDR:
1369		{
1370			struct sockaddr *sa;
1371
1372			sa = (struct sockaddr *) &ifr->ifr_data;
1373			bcopy(IF_LLADDR(ifp), (caddr_t)sa->sa_data,
1374			    ETHER_ADDR_LEN);
1375		}
1376		break;
1377
1378	case SIOCGIFMEDIA:
1379		VLAN_LOCK();
1380		if (TRUNK(ifv) != NULL) {
1381			error = (*PARENT(ifv)->if_ioctl)(PARENT(ifv),
1382					SIOCGIFMEDIA, data);
1383			VLAN_UNLOCK();
1384			/* Limit the result to the parent's current config. */
1385			if (error == 0) {
1386				struct ifmediareq *ifmr;
1387
1388				ifmr = (struct ifmediareq *)data;
1389				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1390					ifmr->ifm_count = 1;
1391					error = copyout(&ifmr->ifm_current,
1392						ifmr->ifm_ulist,
1393						sizeof(int));
1394				}
1395			}
1396		} else {
1397			VLAN_UNLOCK();
1398			error = EINVAL;
1399		}
1400		break;
1401
1402	case SIOCSIFMEDIA:
1403		error = EINVAL;
1404		break;
1405
1406	case SIOCSIFMTU:
1407		/*
1408		 * Set the interface MTU.
1409		 */
1410		VLAN_LOCK();
1411		if (TRUNK(ifv) != NULL) {
1412			if (ifr->ifr_mtu >
1413			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1414			    ifr->ifr_mtu <
1415			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1416				error = EINVAL;
1417			else
1418				ifp->if_mtu = ifr->ifr_mtu;
1419		} else
1420			error = EINVAL;
1421		VLAN_UNLOCK();
1422		break;
1423
1424	case SIOCSETVLAN:
1425		error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1426		if (error)
1427			break;
1428		if (vlr.vlr_parent[0] == '\0') {
1429			vlan_unconfig(ifp);
1430			break;
1431		}
1432		p = ifunit(vlr.vlr_parent);
1433		if (p == 0) {
1434			error = ENOENT;
1435			break;
1436		}
1437		/*
1438		 * Don't let the caller set up a VLAN tag with
1439		 * anything except VLID bits.
1440		 */
1441		if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1442			error = EINVAL;
1443			break;
1444		}
1445		error = vlan_config(ifv, p, vlr.vlr_tag);
1446		if (error)
1447			break;
1448
1449		/* Update flags on the parent, if necessary. */
1450		vlan_setflags(ifp, 1);
1451		break;
1452
1453	case SIOCGETVLAN:
1454		bzero(&vlr, sizeof(vlr));
1455		VLAN_LOCK();
1456		if (TRUNK(ifv) != NULL) {
1457			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1458			    sizeof(vlr.vlr_parent));
1459			vlr.vlr_tag = ifv->ifv_tag;
1460		}
1461		VLAN_UNLOCK();
1462		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1463		break;
1464
1465	case SIOCSIFFLAGS:
1466		/*
1467		 * We should propagate selected flags to the parent,
1468		 * e.g., promiscuous mode.
1469		 */
1470		if (TRUNK(ifv) != NULL)
1471			error = vlan_setflags(ifp, 1);
1472		break;
1473
1474	case SIOCADDMULTI:
1475	case SIOCDELMULTI:
1476		/*
1477		 * If we don't have a parent, just remember the membership for
1478		 * when we do.
1479		 */
1480		if (TRUNK(ifv) != NULL)
1481			error = vlan_setmulti(ifp);
1482		break;
1483
1484	default:
1485		error = EINVAL;
1486	}
1487
1488	return (error);
1489}
1490