if_vlan.c revision 167704
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 167704 2007-03-19 16:31:28Z 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_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_arp.h>
64#include <net/if_dl.h>
65#include <net/if_types.h>
66#include <net/if_vlan_var.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, int link);
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	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 ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
434		bcopy((char *)&mc->mc_addr, LLADDR(&sdl), ETHER_ADDR_LEN);
435		error = if_delmulti(ifp_p, (struct sockaddr *)&sdl);
436		if (error)
437			return (error);
438		SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
439		free(mc, M_VLAN);
440	}
441
442	/* Now program new ones. */
443	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
444		if (ifma->ifma_addr->sa_family != AF_LINK)
445			continue;
446		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
447		if (mc == NULL)
448			return (ENOMEM);
449		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
450		    (char *)&mc->mc_addr, ETHER_ADDR_LEN);
451		SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
452		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
453		    LLADDR(&sdl), ETHER_ADDR_LEN);
454		error = if_addmulti(ifp_p, (struct sockaddr *)&sdl, &rifma);
455		if (error)
456			return (error);
457	}
458
459	return (0);
460}
461
462/*
463 * A handler for network interface departure events.
464 * Track departure of trunks here so that we don't access invalid
465 * pointers or whatever if a trunk is ripped from under us, e.g.,
466 * by ejecting its hot-plug card.
467 */
468static void
469vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
470{
471	struct ifvlan *ifv;
472	int i;
473
474	/*
475	 * Check if it's a trunk interface first of all
476	 * to avoid needless locking.
477	 */
478	if (ifp->if_vlantrunk == NULL)
479		return;
480
481	VLAN_LOCK();
482	/*
483	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
484	 * Check trunk pointer after each vlan_unconfig() as it will
485	 * free it and set to NULL after the last vlan was detached.
486	 */
487#ifdef VLAN_ARRAY
488	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
489		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
490			vlan_unconfig_locked(ifv->ifv_ifp);
491			if (ifp->if_vlantrunk == NULL)
492				break;
493		}
494#else /* VLAN_ARRAY */
495restart:
496	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
497		if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
498			vlan_unconfig_locked(ifv->ifv_ifp);
499			if (ifp->if_vlantrunk)
500				goto restart;	/* trunk->hwidth can change */
501			else
502				break;
503		}
504#endif /* VLAN_ARRAY */
505	/* Trunk should have been destroyed in vlan_unconfig(). */
506	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
507	VLAN_UNLOCK();
508}
509
510/*
511 * VLAN support can be loaded as a module.  The only place in the
512 * system that's intimately aware of this is ether_input.  We hook
513 * into this code through vlan_input_p which is defined there and
514 * set here.  Noone else in the system should be aware of this so
515 * we use an explicit reference here.
516 */
517extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
518
519/* For if_link_state_change() eyes only... */
520extern	void (*vlan_link_state_p)(struct ifnet *, int);
521
522static int
523vlan_modevent(module_t mod, int type, void *data)
524{
525
526	switch (type) {
527	case MOD_LOAD:
528		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
529		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
530		if (ifdetach_tag == NULL)
531			return (ENOMEM);
532		VLAN_LOCK_INIT();
533		vlan_input_p = vlan_input;
534		vlan_link_state_p = vlan_link_state;
535		vlan_trunk_cap_p = vlan_trunk_capabilities;
536		if_clone_attach(&vlan_cloner);
537		if (bootverbose)
538			printf("vlan: initialized, using "
539#ifdef VLAN_ARRAY
540			       "full-size arrays"
541#else
542			       "hash tables with chaining"
543#endif
544
545			       "\n");
546		break;
547	case MOD_UNLOAD:
548		if_clone_detach(&vlan_cloner);
549		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
550		vlan_input_p = NULL;
551		vlan_link_state_p = NULL;
552		vlan_trunk_cap_p = NULL;
553		VLAN_LOCK_DESTROY();
554		if (bootverbose)
555			printf("vlan: unloaded\n");
556		break;
557	default:
558		return (EOPNOTSUPP);
559	}
560	return (0);
561}
562
563static moduledata_t vlan_mod = {
564	"if_vlan",
565	vlan_modevent,
566	0
567};
568
569DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
570MODULE_VERSION(if_vlan, 3);
571MODULE_DEPEND(if_vlan, miibus, 1, 1, 1);
572
573static struct ifnet *
574vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
575{
576	const char *cp;
577	struct ifnet *ifp;
578	int t = 0;
579
580	/* Check for <etherif>.<vlan> style interface names. */
581	IFNET_RLOCK();
582	TAILQ_FOREACH(ifp, &ifnet, if_link) {
583		if (ifp->if_type != IFT_ETHER)
584			continue;
585		if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
586			continue;
587		cp = name + strlen(ifp->if_xname);
588		if (*cp != '.')
589			continue;
590		for(; *cp != '\0'; cp++) {
591			if (*cp < '0' || *cp > '9')
592				continue;
593			t = (t * 10) + (*cp - '0');
594		}
595		if (tag != NULL)
596			*tag = t;
597		break;
598	}
599	IFNET_RUNLOCK();
600
601	return (ifp);
602}
603
604static int
605vlan_clone_match(struct if_clone *ifc, const char *name)
606{
607	const char *cp;
608
609	if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
610		return (1);
611
612	if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
613		return (0);
614	for (cp = name + 4; *cp != '\0'; cp++) {
615		if (*cp < '0' || *cp > '9')
616			return (0);
617	}
618
619	return (1);
620}
621
622static int
623vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
624{
625	char *dp;
626	int wildcard;
627	int unit;
628	int error;
629	int tag;
630	int ethertag;
631	struct ifvlan *ifv;
632	struct ifnet *ifp;
633	struct ifnet *p;
634	struct vlanreq vlr;
635	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
636
637	/*
638	 * There are 3 (ugh) ways to specify the cloned device:
639	 * o pass a parameter block with the clone request.
640	 * o specify parameters in the text of the clone device name
641	 * o specify no parameters and get an unattached device that
642	 *   must be configured separately.
643	 * The first technique is preferred; the latter two are
644	 * supported for backwards compatibilty.
645	 */
646	if (params) {
647		error = copyin(params, &vlr, sizeof(vlr));
648		if (error)
649			return error;
650		p = ifunit(vlr.vlr_parent);
651		if (p == NULL)
652			return ENXIO;
653		/*
654		 * Don't let the caller set up a VLAN tag with
655		 * anything except VLID bits.
656		 */
657		if (vlr.vlr_tag & ~EVL_VLID_MASK)
658			return (EINVAL);
659		error = ifc_name2unit(name, &unit);
660		if (error != 0)
661			return (error);
662
663		ethertag = 1;
664		tag = vlr.vlr_tag;
665		wildcard = (unit < 0);
666	} else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
667		ethertag = 1;
668		unit = -1;
669		wildcard = 0;
670
671		/*
672		 * Don't let the caller set up a VLAN tag with
673		 * anything except VLID bits.
674		 */
675		if (tag & ~EVL_VLID_MASK)
676			return (EINVAL);
677	} else {
678		ethertag = 0;
679
680		error = ifc_name2unit(name, &unit);
681		if (error != 0)
682			return (error);
683
684		wildcard = (unit < 0);
685	}
686
687	error = ifc_alloc_unit(ifc, &unit);
688	if (error != 0)
689		return (error);
690
691	/* In the wildcard case, we need to update the name. */
692	if (wildcard) {
693		for (dp = name; *dp != '\0'; dp++);
694		if (snprintf(dp, len - (dp-name), "%d", unit) >
695		    len - (dp-name) - 1) {
696			panic("%s: interface name too long", __func__);
697		}
698	}
699
700	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
701	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
702	if (ifp == NULL) {
703		ifc_free_unit(ifc, unit);
704		free(ifv, M_VLAN);
705		return (ENOSPC);
706	}
707	SLIST_INIT(&ifv->vlan_mc_listhead);
708
709	ifp->if_softc = ifv;
710	/*
711	 * Set the name manually rather than using if_initname because
712	 * we don't conform to the default naming convention for interfaces.
713	 */
714	strlcpy(ifp->if_xname, name, IFNAMSIZ);
715	ifp->if_dname = ifc->ifc_name;
716	ifp->if_dunit = unit;
717	/* NB: flags are not set here */
718	ifp->if_linkmib = &ifv->ifv_mib;
719	ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
720	/* NB: mtu is not set here */
721
722	ifp->if_init = vlan_init;
723	ifp->if_start = vlan_start;
724	ifp->if_ioctl = vlan_ioctl;
725	ifp->if_snd.ifq_maxlen = ifqmaxlen;
726	ifp->if_flags = VLAN_IFFLAGS;
727	ether_ifattach(ifp, eaddr);
728	/* Now undo some of the damage... */
729	ifp->if_baudrate = 0;
730	ifp->if_type = IFT_L2VLAN;
731	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
732
733	if (ethertag) {
734		error = vlan_config(ifv, p, tag);
735		if (error != 0) {
736			/*
737			 * Since we've partialy failed, we need to back
738			 * out all the way, otherwise userland could get
739			 * confused.  Thus, we destroy the interface.
740			 */
741			ether_ifdetach(ifp);
742			vlan_unconfig(ifp);
743			if_free_type(ifp, IFT_ETHER);
744			free(ifv, M_VLAN);
745
746			return (error);
747		}
748
749		/* Update flags on the parent, if necessary. */
750		vlan_setflags(ifp, 1);
751	}
752
753	return (0);
754}
755
756static int
757vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
758{
759	struct ifvlan *ifv = ifp->if_softc;
760	int unit = ifp->if_dunit;
761
762	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
763	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
764	if_free_type(ifp, IFT_ETHER);
765	free(ifv, M_VLAN);
766	ifc_free_unit(ifc, unit);
767
768	return (0);
769}
770
771/*
772 * The ifp->if_init entry point for vlan(4) is a no-op.
773 */
774static void
775vlan_init(void *foo __unused)
776{
777}
778
779/*
780 * The if_start method for vlan(4) interface. It doesn't
781 * raises the IFF_DRV_OACTIVE flag, since it is called
782 * only from IFQ_HANDOFF() macro in ether_output_frame().
783 * If the interface queue is full, and vlan_start() is
784 * not called, the queue would never get emptied and
785 * interface would stall forever.
786 */
787static void
788vlan_start(struct ifnet *ifp)
789{
790	struct ifvlan *ifv;
791	struct ifnet *p;
792	struct mbuf *m;
793	int error;
794
795	ifv = ifp->if_softc;
796	p = PARENT(ifv);
797
798	for (;;) {
799		IF_DEQUEUE(&ifp->if_snd, m);
800		if (m == NULL)
801			break;
802		BPF_MTAP(ifp, m);
803
804		/*
805		 * Do not run parent's if_start() if the parent is not up,
806		 * or parent's driver will cause a system crash.
807		 */
808		if (!UP_AND_RUNNING(p)) {
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			m->m_pkthdr.ether_vtag = ifv->ifv_tag;
852			m->m_flags |= M_VLANTAG;
853		} else {
854			struct ether_vlan_header *evl;
855
856			M_PREPEND(m, ifv->ifv_encaplen, M_DONTWAIT);
857			if (m == NULL) {
858				if_printf(ifp,
859				    "unable to prepend VLAN header\n");
860				ifp->if_oerrors++;
861				continue;
862			}
863			/* M_PREPEND takes care of m_len, m_pkthdr.len for us */
864
865			if (m->m_len < sizeof(*evl)) {
866				m = m_pullup(m, sizeof(*evl));
867				if (m == NULL) {
868					if_printf(ifp,
869					    "cannot pullup VLAN header\n");
870					ifp->if_oerrors++;
871					continue;
872				}
873			}
874
875			/*
876			 * Transform the Ethernet header into an Ethernet header
877			 * with 802.1Q encapsulation.
878			 */
879			evl = mtod(m, struct ether_vlan_header *);
880			bcopy((char *)evl + ifv->ifv_encaplen,
881			      (char *)evl, ETHER_HDR_LEN - ETHER_TYPE_LEN);
882			evl->evl_encap_proto = htons(ifv->ifv_proto);
883			evl->evl_tag = htons(ifv->ifv_tag);
884#ifdef DEBUG
885			printf("%s: %*D\n", __func__, (int)sizeof(*evl),
886			    (unsigned char *)evl, ":");
887#endif
888		}
889
890		/*
891		 * Send it, precisely as ether_output() would have.
892		 * We are already running at splimp.
893		 */
894		IFQ_HANDOFF(p, m, error);
895		if (!error)
896			ifp->if_opackets++;
897		else
898			ifp->if_oerrors++;
899	}
900}
901
902static void
903vlan_input(struct ifnet *ifp, struct mbuf *m)
904{
905	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
906	struct ifvlan *ifv;
907	uint16_t tag;
908
909	KASSERT(trunk != NULL, ("%s: no trunk", __func__));
910
911	if (m->m_flags & M_VLANTAG) {
912		/*
913		 * Packet is tagged, but m contains a normal
914		 * Ethernet frame; the tag is stored out-of-band.
915		 */
916		tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
917		m->m_flags &= ~M_VLANTAG;
918	} else {
919		struct ether_vlan_header *evl;
920
921		/*
922		 * Packet is tagged in-band as specified by 802.1q.
923		 */
924		switch (ifp->if_type) {
925		case IFT_ETHER:
926			if (m->m_len < sizeof(*evl) &&
927			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
928				if_printf(ifp, "cannot pullup VLAN header\n");
929				return;
930			}
931			evl = mtod(m, struct ether_vlan_header *);
932			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
933
934			/*
935			 * Remove the 802.1q header by copying the Ethernet
936			 * addresses over it and adjusting the beginning of
937			 * the data in the mbuf.  The encapsulated Ethernet
938			 * type field is already in place.
939			 */
940			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
941			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
942			m_adj(m, ETHER_VLAN_ENCAP_LEN);
943			break;
944
945		default:
946#ifdef INVARIANTS
947			panic("%s: %s has unsupported if_type %u",
948			      __func__, ifp->if_xname, ifp->if_type);
949#endif
950			m_freem(m);
951			ifp->if_noproto++;
952			return;
953		}
954	}
955
956	TRUNK_RLOCK(trunk);
957#ifdef VLAN_ARRAY
958	ifv = trunk->vlans[tag];
959#else
960	ifv = vlan_gethash(trunk, tag);
961#endif
962	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
963		TRUNK_RUNLOCK(trunk);
964		m_freem(m);
965		ifp->if_noproto++;
966		return;
967	}
968	TRUNK_RUNLOCK(trunk);
969
970	m->m_pkthdr.rcvif = ifv->ifv_ifp;
971	ifv->ifv_ifp->if_ipackets++;
972
973	/* Pass it back through the parent's input routine. */
974	(*ifp->if_input)(ifv->ifv_ifp, m);
975}
976
977static int
978vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
979{
980	struct ifvlantrunk *trunk;
981	struct ifnet *ifp;
982	int error = 0;
983
984	/* VID numbers 0x0 and 0xFFF are reserved */
985	if (tag == 0 || tag == 0xFFF)
986		return (EINVAL);
987	if (p->if_type != IFT_ETHER)
988		return (EPROTONOSUPPORT);
989	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
990		return (EPROTONOSUPPORT);
991	if (ifv->ifv_trunk)
992		return (EBUSY);
993
994	if (p->if_vlantrunk == NULL) {
995		trunk = malloc(sizeof(struct ifvlantrunk),
996		    M_VLAN, M_WAITOK | M_ZERO);
997#ifndef VLAN_ARRAY
998		vlan_inithash(trunk);
999#endif
1000		VLAN_LOCK();
1001		if (p->if_vlantrunk != NULL) {
1002			/* A race that that is very unlikely to be hit. */
1003#ifndef VLAN_ARRAY
1004			vlan_freehash(trunk);
1005#endif
1006			free(trunk, M_VLAN);
1007			goto exists;
1008		}
1009		TRUNK_LOCK_INIT(trunk);
1010		TRUNK_LOCK(trunk);
1011		p->if_vlantrunk = trunk;
1012		trunk->parent = p;
1013	} else {
1014		VLAN_LOCK();
1015exists:
1016		trunk = p->if_vlantrunk;
1017		TRUNK_LOCK(trunk);
1018	}
1019
1020	ifv->ifv_tag = tag;	/* must set this before vlan_inshash() */
1021#ifdef VLAN_ARRAY
1022	if (trunk->vlans[tag] != NULL) {
1023		error = EEXIST;
1024		goto done;
1025	}
1026	trunk->vlans[tag] = ifv;
1027	trunk->refcnt++;
1028#else
1029	error = vlan_inshash(trunk, ifv);
1030	if (error)
1031		goto done;
1032#endif
1033	ifv->ifv_proto = ETHERTYPE_VLAN;
1034	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1035	ifv->ifv_mintu = ETHERMIN;
1036	ifv->ifv_pflags = 0;
1037
1038	/*
1039	 * If the parent supports the VLAN_MTU capability,
1040	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1041	 * use it.
1042	 */
1043	if (p->if_capenable & IFCAP_VLAN_MTU) {
1044		/*
1045		 * No need to fudge the MTU since the parent can
1046		 * handle extended frames.
1047		 */
1048		ifv->ifv_mtufudge = 0;
1049	} else {
1050		/*
1051		 * Fudge the MTU by the encapsulation size.  This
1052		 * makes us incompatible with strictly compliant
1053		 * 802.1Q implementations, but allows us to use
1054		 * the feature with other NetBSD implementations,
1055		 * which might still be useful.
1056		 */
1057		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1058	}
1059
1060	ifv->ifv_trunk = trunk;
1061	ifp = ifv->ifv_ifp;
1062	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1063	ifp->if_baudrate = p->if_baudrate;
1064	/*
1065	 * Copy only a selected subset of flags from the parent.
1066	 * Other flags are none of our business.
1067	 */
1068#define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1069	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1070	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1071#undef VLAN_COPY_FLAGS
1072
1073	ifp->if_link_state = p->if_link_state;
1074
1075	vlan_capabilities(ifv);
1076
1077	/*
1078	 * Set up our ``Ethernet address'' to reflect the underlying
1079	 * physical interface's.
1080	 */
1081	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), ETHER_ADDR_LEN);
1082
1083	/*
1084	 * Configure multicast addresses that may already be
1085	 * joined on the vlan device.
1086	 */
1087	(void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1088
1089	/* We are ready for operation now. */
1090	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1091done:
1092	TRUNK_UNLOCK(trunk);
1093	VLAN_UNLOCK();
1094
1095	return (error);
1096}
1097
1098static int
1099vlan_unconfig(struct ifnet *ifp)
1100{
1101	int ret;
1102
1103	VLAN_LOCK();
1104	ret = vlan_unconfig_locked(ifp);
1105	VLAN_UNLOCK();
1106	return (ret);
1107}
1108
1109static int
1110vlan_unconfig_locked(struct ifnet *ifp)
1111{
1112	struct ifvlantrunk *trunk;
1113	struct vlan_mc_entry *mc;
1114	struct ifvlan *ifv;
1115	int error;
1116
1117	VLAN_LOCK_ASSERT();
1118
1119	ifv = ifp->if_softc;
1120	trunk = ifv->ifv_trunk;
1121
1122	if (trunk) {
1123		struct sockaddr_dl sdl;
1124		struct ifnet *p = trunk->parent;
1125
1126		TRUNK_LOCK(trunk);
1127
1128		/*
1129		 * Since the interface is being unconfigured, we need to
1130		 * empty the list of multicast groups that we may have joined
1131		 * while we were alive from the parent's list.
1132		 */
1133		bzero((char *)&sdl, sizeof(sdl));
1134		sdl.sdl_len = sizeof(sdl);
1135		sdl.sdl_family = AF_LINK;
1136		sdl.sdl_index = p->if_index;
1137		sdl.sdl_type = IFT_ETHER;
1138		sdl.sdl_alen = ETHER_ADDR_LEN;
1139
1140		while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1141			bcopy((char *)&mc->mc_addr, LLADDR(&sdl),
1142			    ETHER_ADDR_LEN);
1143			error = if_delmulti(p, (struct sockaddr *)&sdl);
1144			if (error)
1145				return (error);
1146			SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1147			free(mc, M_VLAN);
1148		}
1149
1150		vlan_setflags(ifp, 0); /* clear special flags on parent */
1151#ifdef VLAN_ARRAY
1152		trunk->vlans[ifv->ifv_tag] = NULL;
1153		trunk->refcnt--;
1154#else
1155		vlan_remhash(trunk, ifv);
1156#endif
1157		ifv->ifv_trunk = NULL;
1158
1159		/*
1160		 * Check if we were the last.
1161		 */
1162		if (trunk->refcnt == 0) {
1163			trunk->parent->if_vlantrunk = NULL;
1164			/*
1165			 * XXXGL: If some ithread has already entered
1166			 * vlan_input() and is now blocked on the trunk
1167			 * lock, then it should preempt us right after
1168			 * unlock and finish its work. Then we will acquire
1169			 * lock again in trunk_destroy().
1170			 */
1171			TRUNK_UNLOCK(trunk);
1172			trunk_destroy(trunk);
1173		} else
1174			TRUNK_UNLOCK(trunk);
1175	}
1176
1177	/* Disconnect from parent. */
1178	if (ifv->ifv_pflags)
1179		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1180	ifp->if_mtu = ETHERMTU;
1181	ifp->if_link_state = LINK_STATE_UNKNOWN;
1182	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1183
1184	return (0);
1185}
1186
1187/* Handle a reference counted flag that should be set on the parent as well */
1188static int
1189vlan_setflag(struct ifnet *ifp, int flag, int status,
1190	     int (*func)(struct ifnet *, int))
1191{
1192	struct ifvlan *ifv;
1193	int error;
1194
1195	/* XXX VLAN_LOCK_ASSERT(); */
1196
1197	ifv = ifp->if_softc;
1198	status = status ? (ifp->if_flags & flag) : 0;
1199	/* Now "status" contains the flag value or 0 */
1200
1201	/*
1202	 * See if recorded parent's status is different from what
1203	 * we want it to be.  If it is, flip it.  We record parent's
1204	 * status in ifv_pflags so that we won't clear parent's flag
1205	 * we haven't set.  In fact, we don't clear or set parent's
1206	 * flags directly, but get or release references to them.
1207	 * That's why we can be sure that recorded flags still are
1208	 * in accord with actual parent's flags.
1209	 */
1210	if (status != (ifv->ifv_pflags & flag)) {
1211		error = (*func)(PARENT(ifv), status);
1212		if (error)
1213			return (error);
1214		ifv->ifv_pflags &= ~flag;
1215		ifv->ifv_pflags |= status;
1216	}
1217	return (0);
1218}
1219
1220/*
1221 * Handle IFF_* flags that require certain changes on the parent:
1222 * if "status" is true, update parent's flags respective to our if_flags;
1223 * if "status" is false, forcedly clear the flags set on parent.
1224 */
1225static int
1226vlan_setflags(struct ifnet *ifp, int status)
1227{
1228	int error, i;
1229
1230	for (i = 0; vlan_pflags[i].flag; i++) {
1231		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1232				     status, vlan_pflags[i].func);
1233		if (error)
1234			return (error);
1235	}
1236	return (0);
1237}
1238
1239/* Inform all vlans that their parent has changed link state */
1240static void
1241vlan_link_state(struct ifnet *ifp, int link)
1242{
1243	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1244	struct ifvlan *ifv;
1245	int i;
1246
1247	TRUNK_LOCK(trunk);
1248#ifdef VLAN_ARRAY
1249	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1250		if (trunk->vlans[i] != NULL) {
1251			ifv = trunk->vlans[i];
1252#else
1253	for (i = 0; i < (1 << trunk->hwidth); i++)
1254		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1255#endif
1256			ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1257			if_link_state_change(ifv->ifv_ifp,
1258			    trunk->parent->if_link_state);
1259		}
1260	TRUNK_UNLOCK(trunk);
1261}
1262
1263static void
1264vlan_capabilities(struct ifvlan *ifv)
1265{
1266	struct ifnet *p = PARENT(ifv);
1267	struct ifnet *ifp = ifv->ifv_ifp;
1268
1269	TRUNK_LOCK_ASSERT(TRUNK(ifv));
1270
1271	/*
1272	 * If the parent interface can do checksum offloading
1273	 * on VLANs, then propagate its hardware-assisted
1274	 * checksumming flags. Also assert that checksum
1275	 * offloading requires hardware VLAN tagging.
1276	 */
1277	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1278		ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1279
1280	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1281	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1282		ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1283		ifp->if_hwassist = p->if_hwassist;
1284	} else {
1285		ifp->if_capenable = 0;
1286		ifp->if_hwassist = 0;
1287	}
1288}
1289
1290static void
1291vlan_trunk_capabilities(struct ifnet *ifp)
1292{
1293	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1294	struct ifvlan *ifv;
1295	int i;
1296
1297	TRUNK_LOCK(trunk);
1298#ifdef VLAN_ARRAY
1299	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1300		if (trunk->vlans[i] != NULL) {
1301			ifv = trunk->vlans[i];
1302#else
1303	for (i = 0; i < (1 << trunk->hwidth); i++) {
1304		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1305#endif
1306			vlan_capabilities(ifv);
1307	}
1308	TRUNK_UNLOCK(trunk);
1309}
1310
1311static int
1312vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1313{
1314	struct ifaddr *ifa;
1315	struct ifnet *p;
1316	struct ifreq *ifr;
1317	struct ifvlan *ifv;
1318	struct vlanreq vlr;
1319	int error = 0;
1320
1321	ifr = (struct ifreq *)data;
1322	ifa = (struct ifaddr *)data;
1323	ifv = ifp->if_softc;
1324
1325	switch (cmd) {
1326	case SIOCGIFMEDIA:
1327		VLAN_LOCK();
1328		if (TRUNK(ifv) != NULL) {
1329			error = (*PARENT(ifv)->if_ioctl)(PARENT(ifv),
1330					SIOCGIFMEDIA, data);
1331			VLAN_UNLOCK();
1332			/* Limit the result to the parent's current config. */
1333			if (error == 0) {
1334				struct ifmediareq *ifmr;
1335
1336				ifmr = (struct ifmediareq *)data;
1337				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1338					ifmr->ifm_count = 1;
1339					error = copyout(&ifmr->ifm_current,
1340						ifmr->ifm_ulist,
1341						sizeof(int));
1342				}
1343			}
1344		} else {
1345			VLAN_UNLOCK();
1346			error = EINVAL;
1347		}
1348		break;
1349
1350	case SIOCSIFMEDIA:
1351		error = EINVAL;
1352		break;
1353
1354	case SIOCSIFMTU:
1355		/*
1356		 * Set the interface MTU.
1357		 */
1358		VLAN_LOCK();
1359		if (TRUNK(ifv) != NULL) {
1360			if (ifr->ifr_mtu >
1361			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1362			    ifr->ifr_mtu <
1363			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1364				error = EINVAL;
1365			else
1366				ifp->if_mtu = ifr->ifr_mtu;
1367		} else
1368			error = EINVAL;
1369		VLAN_UNLOCK();
1370		break;
1371
1372	case SIOCSETVLAN:
1373		error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1374		if (error)
1375			break;
1376		if (vlr.vlr_parent[0] == '\0') {
1377			vlan_unconfig(ifp);
1378			break;
1379		}
1380		p = ifunit(vlr.vlr_parent);
1381		if (p == 0) {
1382			error = ENOENT;
1383			break;
1384		}
1385		/*
1386		 * Don't let the caller set up a VLAN tag with
1387		 * anything except VLID bits.
1388		 */
1389		if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1390			error = EINVAL;
1391			break;
1392		}
1393		error = vlan_config(ifv, p, vlr.vlr_tag);
1394		if (error)
1395			break;
1396
1397		/* Update flags on the parent, if necessary. */
1398		vlan_setflags(ifp, 1);
1399		break;
1400
1401	case SIOCGETVLAN:
1402		bzero(&vlr, sizeof(vlr));
1403		VLAN_LOCK();
1404		if (TRUNK(ifv) != NULL) {
1405			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1406			    sizeof(vlr.vlr_parent));
1407			vlr.vlr_tag = ifv->ifv_tag;
1408		}
1409		VLAN_UNLOCK();
1410		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1411		break;
1412
1413	case SIOCSIFFLAGS:
1414		/*
1415		 * We should propagate selected flags to the parent,
1416		 * e.g., promiscuous mode.
1417		 */
1418		if (TRUNK(ifv) != NULL)
1419			error = vlan_setflags(ifp, 1);
1420		break;
1421
1422	case SIOCADDMULTI:
1423	case SIOCDELMULTI:
1424		/*
1425		 * If we don't have a parent, just remember the membership for
1426		 * when we do.
1427		 */
1428		if (TRUNK(ifv) != NULL)
1429			error = vlan_setmulti(ifp);
1430		break;
1431
1432	default:
1433		error = ether_ioctl(ifp, cmd, data);
1434	}
1435
1436	return (error);
1437}
1438