if_vlan.c revision 201350
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 201350 2009-12-31 20:29:58Z brooks $
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
75LIST_HEAD(ifvlanhead, ifvlan);
76
77struct 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
91struct vlan_mc_entry {
92	struct ether_addr		mc_addr;
93	SLIST_ENTRY(vlan_mc_entry)	mc_entries;
94};
95
96struct	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. */
121static 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
130SYSCTL_DECL(_net_link);
131SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0, "IEEE 802.1Q VLAN");
132SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0, "for consistency");
133
134static int soft_pad = 0;
135SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW, &soft_pad, 0,
136	   "pad short frames before tagging");
137
138static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface");
139
140static eventhandler_tag ifdetach_tag;
141
142/*
143 * We have a global mutex, that is used to serialize configuration
144 * changes and isn't used in normal packet delivery.
145 *
146 * We also have a per-trunk rwlock, that is locked shared on packet
147 * processing and exclusive when configuration is changed.
148 *
149 * The VLAN_ARRAY substitutes the dynamic hash with a static array
150 * with 4096 entries. In theory this can give a boost in processing,
151 * however on practice it does not. Probably this is because array
152 * is too big to fit into CPU cache.
153 */
154static struct mtx ifv_mtx;
155#define	VLAN_LOCK_INIT()	mtx_init(&ifv_mtx, "vlan_global", NULL, MTX_DEF)
156#define	VLAN_LOCK_DESTROY()	mtx_destroy(&ifv_mtx)
157#define	VLAN_LOCK_ASSERT()	mtx_assert(&ifv_mtx, MA_OWNED)
158#define	VLAN_LOCK()		mtx_lock(&ifv_mtx)
159#define	VLAN_UNLOCK()		mtx_unlock(&ifv_mtx)
160#define	TRUNK_LOCK_INIT(trunk)	rw_init(&(trunk)->rw, VLANNAME)
161#define	TRUNK_LOCK_DESTROY(trunk) rw_destroy(&(trunk)->rw)
162#define	TRUNK_LOCK(trunk)	rw_wlock(&(trunk)->rw)
163#define	TRUNK_UNLOCK(trunk)	rw_wunlock(&(trunk)->rw)
164#define	TRUNK_LOCK_ASSERT(trunk) rw_assert(&(trunk)->rw, RA_WLOCKED)
165#define	TRUNK_RLOCK(trunk)	rw_rlock(&(trunk)->rw)
166#define	TRUNK_RUNLOCK(trunk)	rw_runlock(&(trunk)->rw)
167#define	TRUNK_LOCK_RASSERT(trunk) rw_assert(&(trunk)->rw, RA_RLOCKED)
168
169#ifndef VLAN_ARRAY
170static	void vlan_inithash(struct ifvlantrunk *trunk);
171static	void vlan_freehash(struct ifvlantrunk *trunk);
172static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
173static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
174static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
175static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
176	uint16_t tag);
177#endif
178static	void trunk_destroy(struct ifvlantrunk *trunk);
179
180static	void vlan_start(struct ifnet *ifp);
181static	void vlan_init(void *foo);
182static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
183static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
184static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
185    int (*func)(struct ifnet *, int));
186static	int vlan_setflags(struct ifnet *ifp, int status);
187static	int vlan_setmulti(struct ifnet *ifp);
188static	int vlan_unconfig(struct ifnet *ifp);
189static	int vlan_unconfig_locked(struct ifnet *ifp);
190static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
191static	void vlan_link_state(struct ifnet *ifp);
192static	void vlan_capabilities(struct ifvlan *ifv);
193static	void vlan_trunk_capabilities(struct ifnet *ifp);
194
195static	struct ifnet *vlan_clone_match_ethertag(struct if_clone *,
196    const char *, int *);
197static	int vlan_clone_match(struct if_clone *, const char *);
198static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
199static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
200
201static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
202
203static	struct if_clone vlan_cloner = IFC_CLONE_INITIALIZER(VLANNAME, NULL,
204    IF_MAXUNIT, NULL, vlan_clone_match, vlan_clone_create, vlan_clone_destroy);
205
206#ifndef VLAN_ARRAY
207#define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
208
209static void
210vlan_inithash(struct ifvlantrunk *trunk)
211{
212	int i, n;
213
214	/*
215	 * The trunk must not be locked here since we call malloc(M_WAITOK).
216	 * It is OK in case this function is called before the trunk struct
217	 * gets hooked up and becomes visible from other threads.
218	 */
219
220	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
221	    ("%s: hash already initialized", __func__));
222
223	trunk->hwidth = VLAN_DEF_HWIDTH;
224	n = 1 << trunk->hwidth;
225	trunk->hmask = n - 1;
226	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
227	for (i = 0; i < n; i++)
228		LIST_INIT(&trunk->hash[i]);
229}
230
231static void
232vlan_freehash(struct ifvlantrunk *trunk)
233{
234#ifdef INVARIANTS
235	int i;
236
237	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
238	for (i = 0; i < (1 << trunk->hwidth); i++)
239		KASSERT(LIST_EMPTY(&trunk->hash[i]),
240		    ("%s: hash table not empty", __func__));
241#endif
242	free(trunk->hash, M_VLAN);
243	trunk->hash = NULL;
244	trunk->hwidth = trunk->hmask = 0;
245}
246
247static int
248vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
249{
250	int i, b;
251	struct ifvlan *ifv2;
252
253	TRUNK_LOCK_ASSERT(trunk);
254	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
255
256	b = 1 << trunk->hwidth;
257	i = HASH(ifv->ifv_tag, trunk->hmask);
258	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
259		if (ifv->ifv_tag == ifv2->ifv_tag)
260			return (EEXIST);
261
262	/*
263	 * Grow the hash when the number of vlans exceeds half of the number of
264	 * hash buckets squared. This will make the average linked-list length
265	 * buckets/2.
266	 */
267	if (trunk->refcnt > (b * b) / 2) {
268		vlan_growhash(trunk, 1);
269		i = HASH(ifv->ifv_tag, trunk->hmask);
270	}
271	LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
272	trunk->refcnt++;
273
274	return (0);
275}
276
277static int
278vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
279{
280	int i, b;
281	struct ifvlan *ifv2;
282
283	TRUNK_LOCK_ASSERT(trunk);
284	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
285
286	b = 1 << trunk->hwidth;
287	i = HASH(ifv->ifv_tag, trunk->hmask);
288	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
289		if (ifv2 == ifv) {
290			trunk->refcnt--;
291			LIST_REMOVE(ifv2, ifv_list);
292			if (trunk->refcnt < (b * b) / 2)
293				vlan_growhash(trunk, -1);
294			return (0);
295		}
296
297	panic("%s: vlan not found\n", __func__);
298	return (ENOENT); /*NOTREACHED*/
299}
300
301/*
302 * Grow the hash larger or smaller if memory permits.
303 */
304static void
305vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
306{
307	struct ifvlan *ifv;
308	struct ifvlanhead *hash2;
309	int hwidth2, i, j, n, n2;
310
311	TRUNK_LOCK_ASSERT(trunk);
312	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
313
314	if (howmuch == 0) {
315		/* Harmless yet obvious coding error */
316		printf("%s: howmuch is 0\n", __func__);
317		return;
318	}
319
320	hwidth2 = trunk->hwidth + howmuch;
321	n = 1 << trunk->hwidth;
322	n2 = 1 << hwidth2;
323	/* Do not shrink the table below the default */
324	if (hwidth2 < VLAN_DEF_HWIDTH)
325		return;
326
327	/* M_NOWAIT because we're called with trunk mutex held */
328	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
329	if (hash2 == NULL) {
330		printf("%s: out of memory -- hash size not changed\n",
331		    __func__);
332		return;		/* We can live with the old hash table */
333	}
334	for (j = 0; j < n2; j++)
335		LIST_INIT(&hash2[j]);
336	for (i = 0; i < n; i++)
337		while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
338			LIST_REMOVE(ifv, ifv_list);
339			j = HASH(ifv->ifv_tag, n2 - 1);
340			LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
341		}
342	free(trunk->hash, M_VLAN);
343	trunk->hash = hash2;
344	trunk->hwidth = hwidth2;
345	trunk->hmask = n2 - 1;
346
347	if (bootverbose)
348		if_printf(trunk->parent,
349		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
350}
351
352static __inline struct ifvlan *
353vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
354{
355	struct ifvlan *ifv;
356
357	TRUNK_LOCK_RASSERT(trunk);
358
359	LIST_FOREACH(ifv, &trunk->hash[HASH(tag, trunk->hmask)], ifv_list)
360		if (ifv->ifv_tag == tag)
361			return (ifv);
362	return (NULL);
363}
364
365#if 0
366/* Debugging code to view the hashtables. */
367static void
368vlan_dumphash(struct ifvlantrunk *trunk)
369{
370	int i;
371	struct ifvlan *ifv;
372
373	for (i = 0; i < (1 << trunk->hwidth); i++) {
374		printf("%d: ", i);
375		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
376			printf("%s ", ifv->ifv_ifp->if_xname);
377		printf("\n");
378	}
379}
380#endif /* 0 */
381#endif /* !VLAN_ARRAY */
382
383static void
384trunk_destroy(struct ifvlantrunk *trunk)
385{
386	VLAN_LOCK_ASSERT();
387
388	TRUNK_LOCK(trunk);
389#ifndef VLAN_ARRAY
390	vlan_freehash(trunk);
391#endif
392	trunk->parent->if_vlantrunk = NULL;
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;
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	CURVNET_SET_QUIET(ifp_p->if_vnet);
426
427	bzero((char *)&sdl, sizeof(sdl));
428	sdl.sdl_len = sizeof(sdl);
429	sdl.sdl_family = AF_LINK;
430	sdl.sdl_index = ifp_p->if_index;
431	sdl.sdl_type = IFT_ETHER;
432	sdl.sdl_alen = ETHER_ADDR_LEN;
433
434	/* First, remove any existing filter entries. */
435	while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
436		bcopy((char *)&mc->mc_addr, LLADDR(&sdl), ETHER_ADDR_LEN);
437		error = if_delmulti(ifp_p, (struct sockaddr *)&sdl);
438		if (error)
439			return (error);
440		SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
441		free(mc, M_VLAN);
442	}
443
444	/* Now program new ones. */
445	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
446		if (ifma->ifma_addr->sa_family != AF_LINK)
447			continue;
448		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
449		if (mc == NULL)
450			return (ENOMEM);
451		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
452		    (char *)&mc->mc_addr, ETHER_ADDR_LEN);
453		SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
454		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
455		    LLADDR(&sdl), ETHER_ADDR_LEN);
456		error = if_addmulti(ifp_p, (struct sockaddr *)&sdl, &rifma);
457		if (error)
458			return (error);
459	}
460
461	CURVNET_RESTORE();
462	return (0);
463}
464
465/*
466 * A handler for network interface departure events.
467 * Track departure of trunks here so that we don't access invalid
468 * pointers or whatever if a trunk is ripped from under us, e.g.,
469 * by ejecting its hot-plug card.  However, if an ifnet is simply
470 * being renamed, then there's no need to tear down the state.
471 */
472static void
473vlan_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	/* If the ifnet is just being renamed, don't do anything. */
486	if (ifp->if_flags & IFF_RENAMING)
487		return;
488
489	VLAN_LOCK();
490	/*
491	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
492	 * Check trunk pointer after each vlan_unconfig() as it will
493	 * free it and set to NULL after the last vlan was detached.
494	 */
495#ifdef VLAN_ARRAY
496	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
497		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
498			vlan_unconfig_locked(ifv->ifv_ifp);
499			if (ifp->if_vlantrunk == NULL)
500				break;
501		}
502#else /* VLAN_ARRAY */
503restart:
504	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
505		if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
506			vlan_unconfig_locked(ifv->ifv_ifp);
507			if (ifp->if_vlantrunk)
508				goto restart;	/* trunk->hwidth can change */
509			else
510				break;
511		}
512#endif /* VLAN_ARRAY */
513	/* Trunk should have been destroyed in vlan_unconfig(). */
514	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
515	VLAN_UNLOCK();
516}
517
518/*
519 * VLAN support can be loaded as a module.  The only place in the
520 * system that's intimately aware of this is ether_input.  We hook
521 * into this code through vlan_input_p which is defined there and
522 * set here.  Noone else in the system should be aware of this so
523 * we use an explicit reference here.
524 */
525extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
526
527/* For if_link_state_change() eyes only... */
528extern	void (*vlan_link_state_p)(struct ifnet *);
529
530static int
531vlan_modevent(module_t mod, int type, void *data)
532{
533
534	switch (type) {
535	case MOD_LOAD:
536		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
537		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
538		if (ifdetach_tag == NULL)
539			return (ENOMEM);
540		VLAN_LOCK_INIT();
541		vlan_input_p = vlan_input;
542		vlan_link_state_p = vlan_link_state;
543		vlan_trunk_cap_p = vlan_trunk_capabilities;
544		if_clone_attach(&vlan_cloner);
545		if (bootverbose)
546			printf("vlan: initialized, using "
547#ifdef VLAN_ARRAY
548			       "full-size arrays"
549#else
550			       "hash tables with chaining"
551#endif
552
553			       "\n");
554		break;
555	case MOD_UNLOAD:
556		if_clone_detach(&vlan_cloner);
557		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
558		vlan_input_p = NULL;
559		vlan_link_state_p = NULL;
560		vlan_trunk_cap_p = NULL;
561		VLAN_LOCK_DESTROY();
562		if (bootverbose)
563			printf("vlan: unloaded\n");
564		break;
565	default:
566		return (EOPNOTSUPP);
567	}
568	return (0);
569}
570
571static moduledata_t vlan_mod = {
572	"if_vlan",
573	vlan_modevent,
574	0
575};
576
577DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
578MODULE_VERSION(if_vlan, 3);
579
580static struct ifnet *
581vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
582{
583	const char *cp;
584	struct ifnet *ifp;
585	int t = 0;
586
587	/* Check for <etherif>.<vlan> style interface names. */
588	IFNET_RLOCK_NOSLEEP();
589	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
590		if (ifp->if_type != IFT_ETHER)
591			continue;
592		if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
593			continue;
594		cp = name + strlen(ifp->if_xname);
595		if (*cp != '.')
596			continue;
597		for(; *cp != '\0'; cp++) {
598			if (*cp < '0' || *cp > '9')
599				continue;
600			t = (t * 10) + (*cp - '0');
601		}
602		if (tag != NULL)
603			*tag = t;
604		break;
605	}
606	IFNET_RUNLOCK_NOSLEEP();
607
608	return (ifp);
609}
610
611static int
612vlan_clone_match(struct if_clone *ifc, const char *name)
613{
614	const char *cp;
615
616	if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
617		return (1);
618
619	if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
620		return (0);
621	for (cp = name + 4; *cp != '\0'; cp++) {
622		if (*cp < '0' || *cp > '9')
623			return (0);
624	}
625
626	return (1);
627}
628
629static int
630vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
631{
632	char *dp;
633	int wildcard;
634	int unit;
635	int error;
636	int tag;
637	int ethertag;
638	struct ifvlan *ifv;
639	struct ifnet *ifp;
640	struct ifnet *p;
641	struct vlanreq vlr;
642	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
643
644	/*
645	 * There are 3 (ugh) ways to specify the cloned device:
646	 * o pass a parameter block with the clone request.
647	 * o specify parameters in the text of the clone device name
648	 * o specify no parameters and get an unattached device that
649	 *   must be configured separately.
650	 * The first technique is preferred; the latter two are
651	 * supported for backwards compatibilty.
652	 */
653	if (params) {
654		error = copyin(params, &vlr, sizeof(vlr));
655		if (error)
656			return error;
657		p = ifunit(vlr.vlr_parent);
658		if (p == NULL)
659			return ENXIO;
660		/*
661		 * Don't let the caller set up a VLAN tag with
662		 * anything except VLID bits.
663		 */
664		if (vlr.vlr_tag & ~EVL_VLID_MASK)
665			return (EINVAL);
666		error = ifc_name2unit(name, &unit);
667		if (error != 0)
668			return (error);
669
670		ethertag = 1;
671		tag = vlr.vlr_tag;
672		wildcard = (unit < 0);
673	} else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
674		ethertag = 1;
675		unit = -1;
676		wildcard = 0;
677
678		/*
679		 * Don't let the caller set up a VLAN tag with
680		 * anything except VLID bits.
681		 */
682		if (tag & ~EVL_VLID_MASK)
683			return (EINVAL);
684	} else {
685		ethertag = 0;
686
687		error = ifc_name2unit(name, &unit);
688		if (error != 0)
689			return (error);
690
691		wildcard = (unit < 0);
692	}
693
694	error = ifc_alloc_unit(ifc, &unit);
695	if (error != 0)
696		return (error);
697
698	/* In the wildcard case, we need to update the name. */
699	if (wildcard) {
700		for (dp = name; *dp != '\0'; dp++);
701		if (snprintf(dp, len - (dp-name), "%d", unit) >
702		    len - (dp-name) - 1) {
703			panic("%s: interface name too long", __func__);
704		}
705	}
706
707	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
708	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
709	if (ifp == NULL) {
710		ifc_free_unit(ifc, unit);
711		free(ifv, M_VLAN);
712		return (ENOSPC);
713	}
714	SLIST_INIT(&ifv->vlan_mc_listhead);
715
716	ifp->if_softc = ifv;
717	/*
718	 * Set the name manually rather than using if_initname because
719	 * we don't conform to the default naming convention for interfaces.
720	 */
721	strlcpy(ifp->if_xname, name, IFNAMSIZ);
722	ifp->if_dname = ifc->ifc_name;
723	ifp->if_dunit = unit;
724	/* NB: flags are not set here */
725	ifp->if_linkmib = &ifv->ifv_mib;
726	ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
727	/* NB: mtu is not set here */
728
729	ifp->if_init = vlan_init;
730	ifp->if_start = vlan_start;
731	ifp->if_ioctl = vlan_ioctl;
732	ifp->if_snd.ifq_maxlen = ifqmaxlen;
733	ifp->if_flags = VLAN_IFFLAGS;
734	ether_ifattach(ifp, eaddr);
735	/* Now undo some of the damage... */
736	ifp->if_baudrate = 0;
737	ifp->if_type = IFT_L2VLAN;
738	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
739
740	if (ethertag) {
741		error = vlan_config(ifv, p, tag);
742		if (error != 0) {
743			/*
744			 * Since we've partialy failed, we need to back
745			 * out all the way, otherwise userland could get
746			 * confused.  Thus, we destroy the interface.
747			 */
748			ether_ifdetach(ifp);
749			vlan_unconfig(ifp);
750			if_free_type(ifp, IFT_ETHER);
751			ifc_free_unit(ifc, unit);
752			free(ifv, M_VLAN);
753
754			return (error);
755		}
756
757		/* Update flags on the parent, if necessary. */
758		vlan_setflags(ifp, 1);
759	}
760
761	return (0);
762}
763
764static int
765vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
766{
767	struct ifvlan *ifv = ifp->if_softc;
768	int unit = ifp->if_dunit;
769
770	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
771	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
772	if_free_type(ifp, IFT_ETHER);
773	free(ifv, M_VLAN);
774	ifc_free_unit(ifc, unit);
775
776	return (0);
777}
778
779/*
780 * The ifp->if_init entry point for vlan(4) is a no-op.
781 */
782static void
783vlan_init(void *foo __unused)
784{
785}
786
787/*
788 * The if_start method for vlan(4) interface. It doesn't
789 * raises the IFF_DRV_OACTIVE flag, since it is called
790 * only from IFQ_HANDOFF() macro in ether_output_frame().
791 * If the interface queue is full, and vlan_start() is
792 * not called, the queue would never get emptied and
793 * interface would stall forever.
794 */
795static void
796vlan_start(struct ifnet *ifp)
797{
798	struct ifvlan *ifv;
799	struct ifnet *p;
800	struct mbuf *m;
801	int error;
802
803	ifv = ifp->if_softc;
804	p = PARENT(ifv);
805
806	for (;;) {
807		IF_DEQUEUE(&ifp->if_snd, m);
808		if (m == NULL)
809			break;
810		BPF_MTAP(ifp, m);
811
812		/*
813		 * Do not run parent's if_start() if the parent is not up,
814		 * or parent's driver will cause a system crash.
815		 */
816		if (!UP_AND_RUNNING(p)) {
817			m_freem(m);
818			ifp->if_collisions++;
819			continue;
820		}
821
822		/*
823		 * Pad the frame to the minimum size allowed if told to.
824		 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
825		 * paragraph C.4.4.3.b.  It can help to work around buggy
826		 * bridges that violate paragraph C.4.4.3.a from the same
827		 * document, i.e., fail to pad short frames after untagging.
828		 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
829		 * untagging it will produce a 62-byte frame, which is a runt
830		 * and requires padding.  There are VLAN-enabled network
831		 * devices that just discard such runts instead or mishandle
832		 * them somehow.
833		 */
834		if (soft_pad) {
835			static char pad[8];	/* just zeros */
836			int n;
837
838			for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
839			     n > 0; n -= sizeof(pad))
840				if (!m_append(m, min(n, sizeof(pad)), pad))
841					break;
842
843			if (n > 0) {
844				if_printf(ifp, "cannot pad short frame\n");
845				ifp->if_oerrors++;
846				m_freem(m);
847				continue;
848			}
849		}
850
851		/*
852		 * If underlying interface can do VLAN tag insertion itself,
853		 * just pass the packet along. However, we need some way to
854		 * tell the interface where the packet came from so that it
855		 * knows how to find the VLAN tag to use, so we attach a
856		 * packet tag that holds it.
857		 */
858		if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
859			m->m_pkthdr.ether_vtag = ifv->ifv_tag;
860			m->m_flags |= M_VLANTAG;
861		} else {
862			m = ether_vlanencap(m, ifv->ifv_tag);
863			if (m == NULL) {
864				if_printf(ifp,
865				    "unable to prepend VLAN header\n");
866				ifp->if_oerrors++;
867				continue;
868			}
869		}
870
871		/*
872		 * Send it, precisely as ether_output() would have.
873		 * We are already running at splimp.
874		 */
875		error = (p->if_transmit)(p, m);
876		if (!error)
877			ifp->if_opackets++;
878		else
879			ifp->if_oerrors++;
880	}
881}
882
883static void
884vlan_input(struct ifnet *ifp, struct mbuf *m)
885{
886	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
887	struct ifvlan *ifv;
888	uint16_t tag;
889
890	KASSERT(trunk != NULL, ("%s: no trunk", __func__));
891
892	if (m->m_flags & M_VLANTAG) {
893		/*
894		 * Packet is tagged, but m contains a normal
895		 * Ethernet frame; the tag is stored out-of-band.
896		 */
897		tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
898		m->m_flags &= ~M_VLANTAG;
899	} else {
900		struct ether_vlan_header *evl;
901
902		/*
903		 * Packet is tagged in-band as specified by 802.1q.
904		 */
905		switch (ifp->if_type) {
906		case IFT_ETHER:
907			if (m->m_len < sizeof(*evl) &&
908			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
909				if_printf(ifp, "cannot pullup VLAN header\n");
910				return;
911			}
912			evl = mtod(m, struct ether_vlan_header *);
913			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
914
915			/*
916			 * Remove the 802.1q header by copying the Ethernet
917			 * addresses over it and adjusting the beginning of
918			 * the data in the mbuf.  The encapsulated Ethernet
919			 * type field is already in place.
920			 */
921			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
922			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
923			m_adj(m, ETHER_VLAN_ENCAP_LEN);
924			break;
925
926		default:
927#ifdef INVARIANTS
928			panic("%s: %s has unsupported if_type %u",
929			      __func__, ifp->if_xname, ifp->if_type);
930#endif
931			m_freem(m);
932			ifp->if_noproto++;
933			return;
934		}
935	}
936
937	TRUNK_RLOCK(trunk);
938#ifdef VLAN_ARRAY
939	ifv = trunk->vlans[tag];
940#else
941	ifv = vlan_gethash(trunk, tag);
942#endif
943	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
944		TRUNK_RUNLOCK(trunk);
945		m_freem(m);
946		ifp->if_noproto++;
947		return;
948	}
949	TRUNK_RUNLOCK(trunk);
950
951	m->m_pkthdr.rcvif = ifv->ifv_ifp;
952	ifv->ifv_ifp->if_ipackets++;
953
954	/* Pass it back through the parent's input routine. */
955	(*ifp->if_input)(ifv->ifv_ifp, m);
956}
957
958static int
959vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
960{
961	struct ifvlantrunk *trunk;
962	struct ifnet *ifp;
963	int error = 0;
964
965	/* VID numbers 0x0 and 0xFFF are reserved */
966	if (tag == 0 || tag == 0xFFF)
967		return (EINVAL);
968	if (p->if_type != IFT_ETHER)
969		return (EPROTONOSUPPORT);
970	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
971		return (EPROTONOSUPPORT);
972	if (ifv->ifv_trunk)
973		return (EBUSY);
974
975	if (p->if_vlantrunk == NULL) {
976		trunk = malloc(sizeof(struct ifvlantrunk),
977		    M_VLAN, M_WAITOK | M_ZERO);
978#ifndef VLAN_ARRAY
979		vlan_inithash(trunk);
980#endif
981		VLAN_LOCK();
982		if (p->if_vlantrunk != NULL) {
983			/* A race that that is very unlikely to be hit. */
984#ifndef VLAN_ARRAY
985			vlan_freehash(trunk);
986#endif
987			free(trunk, M_VLAN);
988			goto exists;
989		}
990		TRUNK_LOCK_INIT(trunk);
991		TRUNK_LOCK(trunk);
992		p->if_vlantrunk = trunk;
993		trunk->parent = p;
994	} else {
995		VLAN_LOCK();
996exists:
997		trunk = p->if_vlantrunk;
998		TRUNK_LOCK(trunk);
999	}
1000
1001	ifv->ifv_tag = tag;	/* must set this before vlan_inshash() */
1002#ifdef VLAN_ARRAY
1003	if (trunk->vlans[tag] != NULL) {
1004		error = EEXIST;
1005		goto done;
1006	}
1007	trunk->vlans[tag] = ifv;
1008	trunk->refcnt++;
1009#else
1010	error = vlan_inshash(trunk, ifv);
1011	if (error)
1012		goto done;
1013#endif
1014	ifv->ifv_proto = ETHERTYPE_VLAN;
1015	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1016	ifv->ifv_mintu = ETHERMIN;
1017	ifv->ifv_pflags = 0;
1018
1019	/*
1020	 * If the parent supports the VLAN_MTU capability,
1021	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1022	 * use it.
1023	 */
1024	if (p->if_capenable & IFCAP_VLAN_MTU) {
1025		/*
1026		 * No need to fudge the MTU since the parent can
1027		 * handle extended frames.
1028		 */
1029		ifv->ifv_mtufudge = 0;
1030	} else {
1031		/*
1032		 * Fudge the MTU by the encapsulation size.  This
1033		 * makes us incompatible with strictly compliant
1034		 * 802.1Q implementations, but allows us to use
1035		 * the feature with other NetBSD implementations,
1036		 * which might still be useful.
1037		 */
1038		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1039	}
1040
1041	ifv->ifv_trunk = trunk;
1042	ifp = ifv->ifv_ifp;
1043	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1044	ifp->if_baudrate = p->if_baudrate;
1045	/*
1046	 * Copy only a selected subset of flags from the parent.
1047	 * Other flags are none of our business.
1048	 */
1049#define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1050	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1051	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1052#undef VLAN_COPY_FLAGS
1053
1054	ifp->if_link_state = p->if_link_state;
1055
1056	vlan_capabilities(ifv);
1057
1058	/*
1059	 * Set up our ``Ethernet address'' to reflect the underlying
1060	 * physical interface's.
1061	 */
1062	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), ETHER_ADDR_LEN);
1063
1064	/*
1065	 * Configure multicast addresses that may already be
1066	 * joined on the vlan device.
1067	 */
1068	(void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1069
1070	/* We are ready for operation now. */
1071	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1072done:
1073	TRUNK_UNLOCK(trunk);
1074	if (error == 0)
1075		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_tag);
1076	VLAN_UNLOCK();
1077
1078	return (error);
1079}
1080
1081static int
1082vlan_unconfig(struct ifnet *ifp)
1083{
1084	int ret;
1085
1086	VLAN_LOCK();
1087	ret = vlan_unconfig_locked(ifp);
1088	VLAN_UNLOCK();
1089	return (ret);
1090}
1091
1092static int
1093vlan_unconfig_locked(struct ifnet *ifp)
1094{
1095	struct ifvlantrunk *trunk;
1096	struct vlan_mc_entry *mc;
1097	struct ifvlan *ifv;
1098	struct ifnet  *parent;
1099	int error;
1100
1101	VLAN_LOCK_ASSERT();
1102
1103	ifv = ifp->if_softc;
1104	trunk = ifv->ifv_trunk;
1105	parent = NULL;
1106
1107	if (trunk != NULL) {
1108		struct sockaddr_dl sdl;
1109
1110		TRUNK_LOCK(trunk);
1111		parent = trunk->parent;
1112
1113		/*
1114		 * Since the interface is being unconfigured, we need to
1115		 * empty the list of multicast groups that we may have joined
1116		 * while we were alive from the parent's list.
1117		 */
1118		bzero((char *)&sdl, sizeof(sdl));
1119		sdl.sdl_len = sizeof(sdl);
1120		sdl.sdl_family = AF_LINK;
1121		sdl.sdl_index = parent->if_index;
1122		sdl.sdl_type = IFT_ETHER;
1123		sdl.sdl_alen = ETHER_ADDR_LEN;
1124
1125		while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1126			bcopy((char *)&mc->mc_addr, LLADDR(&sdl),
1127			    ETHER_ADDR_LEN);
1128			error = if_delmulti(parent, (struct sockaddr *)&sdl);
1129			if (error)
1130				return (error);
1131			SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1132			free(mc, M_VLAN);
1133		}
1134
1135		vlan_setflags(ifp, 0); /* clear special flags on parent */
1136#ifdef VLAN_ARRAY
1137		trunk->vlans[ifv->ifv_tag] = NULL;
1138		trunk->refcnt--;
1139#else
1140		vlan_remhash(trunk, ifv);
1141#endif
1142		ifv->ifv_trunk = NULL;
1143
1144		/*
1145		 * Check if we were the last.
1146		 */
1147		if (trunk->refcnt == 0) {
1148			trunk->parent->if_vlantrunk = NULL;
1149			/*
1150			 * XXXGL: If some ithread has already entered
1151			 * vlan_input() and is now blocked on the trunk
1152			 * lock, then it should preempt us right after
1153			 * unlock and finish its work. Then we will acquire
1154			 * lock again in trunk_destroy().
1155			 */
1156			TRUNK_UNLOCK(trunk);
1157			trunk_destroy(trunk);
1158		} else
1159			TRUNK_UNLOCK(trunk);
1160	}
1161
1162	/* Disconnect from parent. */
1163	if (ifv->ifv_pflags)
1164		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1165	ifp->if_mtu = ETHERMTU;
1166	ifp->if_link_state = LINK_STATE_UNKNOWN;
1167	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1168
1169	/*
1170	 * Only dispatch an event if vlan was
1171	 * attached, otherwise there is nothing
1172	 * to cleanup anyway.
1173	 */
1174	if (parent != NULL)
1175		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_tag);
1176
1177	return (0);
1178}
1179
1180/* Handle a reference counted flag that should be set on the parent as well */
1181static int
1182vlan_setflag(struct ifnet *ifp, int flag, int status,
1183	     int (*func)(struct ifnet *, int))
1184{
1185	struct ifvlan *ifv;
1186	int error;
1187
1188	/* XXX VLAN_LOCK_ASSERT(); */
1189
1190	ifv = ifp->if_softc;
1191	status = status ? (ifp->if_flags & flag) : 0;
1192	/* Now "status" contains the flag value or 0 */
1193
1194	/*
1195	 * See if recorded parent's status is different from what
1196	 * we want it to be.  If it is, flip it.  We record parent's
1197	 * status in ifv_pflags so that we won't clear parent's flag
1198	 * we haven't set.  In fact, we don't clear or set parent's
1199	 * flags directly, but get or release references to them.
1200	 * That's why we can be sure that recorded flags still are
1201	 * in accord with actual parent's flags.
1202	 */
1203	if (status != (ifv->ifv_pflags & flag)) {
1204		error = (*func)(PARENT(ifv), status);
1205		if (error)
1206			return (error);
1207		ifv->ifv_pflags &= ~flag;
1208		ifv->ifv_pflags |= status;
1209	}
1210	return (0);
1211}
1212
1213/*
1214 * Handle IFF_* flags that require certain changes on the parent:
1215 * if "status" is true, update parent's flags respective to our if_flags;
1216 * if "status" is false, forcedly clear the flags set on parent.
1217 */
1218static int
1219vlan_setflags(struct ifnet *ifp, int status)
1220{
1221	int error, i;
1222
1223	for (i = 0; vlan_pflags[i].flag; i++) {
1224		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1225				     status, vlan_pflags[i].func);
1226		if (error)
1227			return (error);
1228	}
1229	return (0);
1230}
1231
1232/* Inform all vlans that their parent has changed link state */
1233static void
1234vlan_link_state(struct ifnet *ifp)
1235{
1236	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1237	struct ifvlan *ifv;
1238	int i;
1239
1240	TRUNK_LOCK(trunk);
1241#ifdef VLAN_ARRAY
1242	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1243		if (trunk->vlans[i] != NULL) {
1244			ifv = trunk->vlans[i];
1245#else
1246	for (i = 0; i < (1 << trunk->hwidth); i++)
1247		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1248#endif
1249			ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1250			if_link_state_change(ifv->ifv_ifp,
1251			    trunk->parent->if_link_state);
1252		}
1253	TRUNK_UNLOCK(trunk);
1254}
1255
1256static void
1257vlan_capabilities(struct ifvlan *ifv)
1258{
1259	struct ifnet *p = PARENT(ifv);
1260	struct ifnet *ifp = ifv->ifv_ifp;
1261
1262	TRUNK_LOCK_ASSERT(TRUNK(ifv));
1263
1264	/*
1265	 * If the parent interface can do checksum offloading
1266	 * on VLANs, then propagate its hardware-assisted
1267	 * checksumming flags. Also assert that checksum
1268	 * offloading requires hardware VLAN tagging.
1269	 */
1270	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1271		ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1272
1273	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1274	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1275		ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1276		ifp->if_hwassist = p->if_hwassist;
1277	} else {
1278		ifp->if_capenable = 0;
1279		ifp->if_hwassist = 0;
1280	}
1281}
1282
1283static void
1284vlan_trunk_capabilities(struct ifnet *ifp)
1285{
1286	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1287	struct ifvlan *ifv;
1288	int i;
1289
1290	TRUNK_LOCK(trunk);
1291#ifdef VLAN_ARRAY
1292	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1293		if (trunk->vlans[i] != NULL) {
1294			ifv = trunk->vlans[i];
1295#else
1296	for (i = 0; i < (1 << trunk->hwidth); i++) {
1297		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1298#endif
1299			vlan_capabilities(ifv);
1300	}
1301	TRUNK_UNLOCK(trunk);
1302}
1303
1304static int
1305vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1306{
1307	struct ifnet *p;
1308	struct ifreq *ifr;
1309	struct ifvlan *ifv;
1310	struct vlanreq vlr;
1311	int error = 0;
1312
1313	ifr = (struct ifreq *)data;
1314	ifv = ifp->if_softc;
1315
1316	switch (cmd) {
1317	case SIOCGIFMEDIA:
1318		VLAN_LOCK();
1319		if (TRUNK(ifv) != NULL) {
1320			error = (*PARENT(ifv)->if_ioctl)(PARENT(ifv),
1321					SIOCGIFMEDIA, data);
1322			VLAN_UNLOCK();
1323			/* Limit the result to the parent's current config. */
1324			if (error == 0) {
1325				struct ifmediareq *ifmr;
1326
1327				ifmr = (struct ifmediareq *)data;
1328				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1329					ifmr->ifm_count = 1;
1330					error = copyout(&ifmr->ifm_current,
1331						ifmr->ifm_ulist,
1332						sizeof(int));
1333				}
1334			}
1335		} else {
1336			VLAN_UNLOCK();
1337			error = EINVAL;
1338		}
1339		break;
1340
1341	case SIOCSIFMEDIA:
1342		error = EINVAL;
1343		break;
1344
1345	case SIOCSIFMTU:
1346		/*
1347		 * Set the interface MTU.
1348		 */
1349		VLAN_LOCK();
1350		if (TRUNK(ifv) != NULL) {
1351			if (ifr->ifr_mtu >
1352			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1353			    ifr->ifr_mtu <
1354			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1355				error = EINVAL;
1356			else
1357				ifp->if_mtu = ifr->ifr_mtu;
1358		} else
1359			error = EINVAL;
1360		VLAN_UNLOCK();
1361		break;
1362
1363	case SIOCSETVLAN:
1364		error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1365		if (error)
1366			break;
1367		if (vlr.vlr_parent[0] == '\0') {
1368			vlan_unconfig(ifp);
1369			break;
1370		}
1371		p = ifunit(vlr.vlr_parent);
1372		if (p == NULL) {
1373			error = ENOENT;
1374			break;
1375		}
1376		/*
1377		 * Don't let the caller set up a VLAN tag with
1378		 * anything except VLID bits.
1379		 */
1380		if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1381			error = EINVAL;
1382			break;
1383		}
1384		error = vlan_config(ifv, p, vlr.vlr_tag);
1385		if (error)
1386			break;
1387
1388		/* Update flags on the parent, if necessary. */
1389		vlan_setflags(ifp, 1);
1390		break;
1391
1392	case SIOCGETVLAN:
1393		bzero(&vlr, sizeof(vlr));
1394		VLAN_LOCK();
1395		if (TRUNK(ifv) != NULL) {
1396			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1397			    sizeof(vlr.vlr_parent));
1398			vlr.vlr_tag = ifv->ifv_tag;
1399		}
1400		VLAN_UNLOCK();
1401		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1402		break;
1403
1404	case SIOCSIFFLAGS:
1405		/*
1406		 * We should propagate selected flags to the parent,
1407		 * e.g., promiscuous mode.
1408		 */
1409		if (TRUNK(ifv) != NULL)
1410			error = vlan_setflags(ifp, 1);
1411		break;
1412
1413	case SIOCADDMULTI:
1414	case SIOCDELMULTI:
1415		/*
1416		 * If we don't have a parent, just remember the membership for
1417		 * when we do.
1418		 */
1419		if (TRUNK(ifv) != NULL)
1420			error = vlan_setmulti(ifp);
1421		break;
1422
1423	default:
1424		error = ether_ioctl(ifp, cmd, data);
1425	}
1426
1427	return (error);
1428}
1429