if_vlan.c revision 191217
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 191217 2009-04-17 17:40:47Z jhb $
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_route.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#include <sys/vimage.h>
60
61#include <net/bpf.h>
62#include <net/ethernet.h>
63#include <net/if.h>
64#include <net/if_clone.h>
65#include <net/if_dl.h>
66#include <net/if_types.h>
67#include <net/if_vlan_var.h>
68#include <net/route.h>
69#include <net/vnet.h>
70
71#define VLANNAME	"vlan"
72#define	VLAN_DEF_HWIDTH	4
73#define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
74
75#define	UP_AND_RUNNING(ifp) \
76    ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
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};
93
94struct vlan_mc_entry {
95	struct ether_addr		mc_addr;
96	SLIST_ENTRY(vlan_mc_entry)	mc_entries;
97};
98
99struct	ifvlan {
100	struct	ifvlantrunk *ifv_trunk;
101	struct	ifnet *ifv_ifp;
102#define	TRUNK(ifv)	((ifv)->ifv_trunk)
103#define	PARENT(ifv)	((ifv)->ifv_trunk->parent)
104	int	ifv_pflags;	/* special flags we have set on parent */
105	struct	ifv_linkmib {
106		int	ifvm_encaplen;	/* encapsulation length */
107		int	ifvm_mtufudge;	/* MTU fudged by this much */
108		int	ifvm_mintu;	/* min transmission unit */
109		uint16_t ifvm_proto;	/* encapsulation ethertype */
110		uint16_t ifvm_tag;	/* tag to apply on packets leaving if */
111	}	ifv_mib;
112	SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
113#ifndef VLAN_ARRAY
114	LIST_ENTRY(ifvlan) ifv_list;
115#endif
116};
117#define	ifv_proto	ifv_mib.ifvm_proto
118#define	ifv_tag		ifv_mib.ifvm_tag
119#define	ifv_encaplen	ifv_mib.ifvm_encaplen
120#define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
121#define	ifv_mintu	ifv_mib.ifvm_mintu
122
123/* Special flags we should propagate to parent. */
124static struct {
125	int flag;
126	int (*func)(struct ifnet *, int);
127} vlan_pflags[] = {
128	{IFF_PROMISC, ifpromisc},
129	{IFF_ALLMULTI, if_allmulti},
130	{0, NULL}
131};
132
133SYSCTL_DECL(_net_link);
134SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0, "IEEE 802.1Q VLAN");
135SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0, "for consistency");
136
137static int soft_pad = 0;
138SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW, &soft_pad, 0,
139	   "pad short frames before tagging");
140
141static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface");
142
143static eventhandler_tag ifdetach_tag;
144
145/*
146 * We have a global mutex, that is used to serialize configuration
147 * changes and isn't used in normal packet delivery.
148 *
149 * We also have a per-trunk rwlock, that is locked shared on packet
150 * processing and exclusive when configuration is changed.
151 *
152 * The VLAN_ARRAY substitutes the dynamic hash with a static array
153 * with 4096 entries. In theory this can give a boost in processing,
154 * however on practice it does not. Probably this is because array
155 * is too big to fit into CPU cache.
156 */
157static struct mtx ifv_mtx;
158#define	VLAN_LOCK_INIT()	mtx_init(&ifv_mtx, "vlan_global", NULL, MTX_DEF)
159#define	VLAN_LOCK_DESTROY()	mtx_destroy(&ifv_mtx)
160#define	VLAN_LOCK_ASSERT()	mtx_assert(&ifv_mtx, MA_OWNED)
161#define	VLAN_LOCK()		mtx_lock(&ifv_mtx)
162#define	VLAN_UNLOCK()		mtx_unlock(&ifv_mtx)
163#define	TRUNK_LOCK_INIT(trunk)	rw_init(&(trunk)->rw, VLANNAME)
164#define	TRUNK_LOCK_DESTROY(trunk) rw_destroy(&(trunk)->rw)
165#define	TRUNK_LOCK(trunk)	rw_wlock(&(trunk)->rw)
166#define	TRUNK_UNLOCK(trunk)	rw_wunlock(&(trunk)->rw)
167#define	TRUNK_LOCK_ASSERT(trunk) rw_assert(&(trunk)->rw, RA_WLOCKED)
168#define	TRUNK_RLOCK(trunk)	rw_rlock(&(trunk)->rw)
169#define	TRUNK_RUNLOCK(trunk)	rw_runlock(&(trunk)->rw)
170#define	TRUNK_LOCK_RASSERT(trunk) rw_assert(&(trunk)->rw, RA_RLOCKED)
171
172#ifndef VLAN_ARRAY
173static	void vlan_inithash(struct ifvlantrunk *trunk);
174static	void vlan_freehash(struct ifvlantrunk *trunk);
175static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
176static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
177static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
178static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
179	uint16_t tag);
180#endif
181static	void trunk_destroy(struct ifvlantrunk *trunk);
182
183static	void vlan_start(struct ifnet *ifp);
184static	void vlan_init(void *foo);
185static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
186static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
187static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
188    int (*func)(struct ifnet *, int));
189static	int vlan_setflags(struct ifnet *ifp, int status);
190static	int vlan_setmulti(struct ifnet *ifp);
191static	int vlan_unconfig(struct ifnet *ifp);
192static	int vlan_unconfig_locked(struct ifnet *ifp);
193static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
194static	void vlan_link_state(struct ifnet *ifp, int link);
195static	void vlan_capabilities(struct ifvlan *ifv);
196static	void vlan_trunk_capabilities(struct ifnet *ifp);
197
198static	struct ifnet *vlan_clone_match_ethertag(struct if_clone *,
199    const char *, int *);
200static	int vlan_clone_match(struct if_clone *, const char *);
201static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
202static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
203
204static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
205
206static	struct if_clone vlan_cloner = IFC_CLONE_INITIALIZER(VLANNAME, NULL,
207    IF_MAXUNIT, NULL, vlan_clone_match, vlan_clone_create, vlan_clone_destroy);
208
209#ifndef VLAN_ARRAY
210#define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
211
212static void
213vlan_inithash(struct ifvlantrunk *trunk)
214{
215	int i, n;
216
217	/*
218	 * The trunk must not be locked here since we call malloc(M_WAITOK).
219	 * It is OK in case this function is called before the trunk struct
220	 * gets hooked up and becomes visible from other threads.
221	 */
222
223	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
224	    ("%s: hash already initialized", __func__));
225
226	trunk->hwidth = VLAN_DEF_HWIDTH;
227	n = 1 << trunk->hwidth;
228	trunk->hmask = n - 1;
229	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
230	for (i = 0; i < n; i++)
231		LIST_INIT(&trunk->hash[i]);
232}
233
234static void
235vlan_freehash(struct ifvlantrunk *trunk)
236{
237#ifdef INVARIANTS
238	int i;
239
240	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
241	for (i = 0; i < (1 << trunk->hwidth); i++)
242		KASSERT(LIST_EMPTY(&trunk->hash[i]),
243		    ("%s: hash table not empty", __func__));
244#endif
245	free(trunk->hash, M_VLAN);
246	trunk->hash = NULL;
247	trunk->hwidth = trunk->hmask = 0;
248}
249
250static int
251vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
252{
253	int i, b;
254	struct ifvlan *ifv2;
255
256	TRUNK_LOCK_ASSERT(trunk);
257	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
258
259	b = 1 << trunk->hwidth;
260	i = HASH(ifv->ifv_tag, trunk->hmask);
261	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
262		if (ifv->ifv_tag == ifv2->ifv_tag)
263			return (EEXIST);
264
265	/*
266	 * Grow the hash when the number of vlans exceeds half of the number of
267	 * hash buckets squared. This will make the average linked-list length
268	 * buckets/2.
269	 */
270	if (trunk->refcnt > (b * b) / 2) {
271		vlan_growhash(trunk, 1);
272		i = HASH(ifv->ifv_tag, trunk->hmask);
273	}
274	LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
275	trunk->refcnt++;
276
277	return (0);
278}
279
280static int
281vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
282{
283	int i, b;
284	struct ifvlan *ifv2;
285
286	TRUNK_LOCK_ASSERT(trunk);
287	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
288
289	b = 1 << trunk->hwidth;
290	i = HASH(ifv->ifv_tag, trunk->hmask);
291	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
292		if (ifv2 == ifv) {
293			trunk->refcnt--;
294			LIST_REMOVE(ifv2, ifv_list);
295			if (trunk->refcnt < (b * b) / 2)
296				vlan_growhash(trunk, -1);
297			return (0);
298		}
299
300	panic("%s: vlan not found\n", __func__);
301	return (ENOENT); /*NOTREACHED*/
302}
303
304/*
305 * Grow the hash larger or smaller if memory permits.
306 */
307static void
308vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
309{
310	struct ifvlan *ifv;
311	struct ifvlanhead *hash2;
312	int hwidth2, i, j, n, n2;
313
314	TRUNK_LOCK_ASSERT(trunk);
315	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
316
317	if (howmuch == 0) {
318		/* Harmless yet obvious coding error */
319		printf("%s: howmuch is 0\n", __func__);
320		return;
321	}
322
323	hwidth2 = trunk->hwidth + howmuch;
324	n = 1 << trunk->hwidth;
325	n2 = 1 << hwidth2;
326	/* Do not shrink the table below the default */
327	if (hwidth2 < VLAN_DEF_HWIDTH)
328		return;
329
330	/* M_NOWAIT because we're called with trunk mutex held */
331	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
332	if (hash2 == NULL) {
333		printf("%s: out of memory -- hash size not changed\n",
334		    __func__);
335		return;		/* We can live with the old hash table */
336	}
337	for (j = 0; j < n2; j++)
338		LIST_INIT(&hash2[j]);
339	for (i = 0; i < n; i++)
340		while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
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	if (bootverbose)
351		if_printf(trunk->parent,
352		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
353}
354
355static __inline struct ifvlan *
356vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
357{
358	struct ifvlan *ifv;
359
360	TRUNK_LOCK_RASSERT(trunk);
361
362	LIST_FOREACH(ifv, &trunk->hash[HASH(tag, trunk->hmask)], ifv_list)
363		if (ifv->ifv_tag == tag)
364			return (ifv);
365	return (NULL);
366}
367
368#if 0
369/* Debugging code to view the hashtables. */
370static void
371vlan_dumphash(struct ifvlantrunk *trunk)
372{
373	int i;
374	struct ifvlan *ifv;
375
376	for (i = 0; i < (1 << trunk->hwidth); i++) {
377		printf("%d: ", i);
378		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
379			printf("%s ", ifv->ifv_ifp->if_xname);
380		printf("\n");
381	}
382}
383#endif /* 0 */
384#endif /* !VLAN_ARRAY */
385
386static void
387trunk_destroy(struct ifvlantrunk *trunk)
388{
389	VLAN_LOCK_ASSERT();
390
391	TRUNK_LOCK(trunk);
392#ifndef VLAN_ARRAY
393	vlan_freehash(trunk);
394#endif
395	trunk->parent->if_vlantrunk = NULL;
396	TRUNK_UNLOCK(trunk);
397	TRUNK_LOCK_DESTROY(trunk);
398	free(trunk, M_VLAN);
399}
400
401/*
402 * Program our multicast filter. What we're actually doing is
403 * programming the multicast filter of the parent. This has the
404 * side effect of causing the parent interface to receive multicast
405 * traffic that it doesn't really want, which ends up being discarded
406 * later by the upper protocol layers. Unfortunately, there's no way
407 * to avoid this: there really is only one physical interface.
408 *
409 * XXX: There is a possible race here if more than one thread is
410 *      modifying the multicast state of the vlan interface at the same time.
411 */
412static int
413vlan_setmulti(struct ifnet *ifp)
414{
415	struct ifnet		*ifp_p;
416	struct ifmultiaddr	*ifma, *rifma = NULL;
417	struct ifvlan		*sc;
418	struct vlan_mc_entry	*mc;
419	struct sockaddr_dl	sdl;
420	int			error;
421
422	/*VLAN_LOCK_ASSERT();*/
423
424	/* Find the parent. */
425	sc = ifp->if_softc;
426	ifp_p = PARENT(sc);
427
428	CURVNET_SET_QUIET(ifp_p->if_vnet);
429
430	bzero((char *)&sdl, sizeof(sdl));
431	sdl.sdl_len = sizeof(sdl);
432	sdl.sdl_family = AF_LINK;
433	sdl.sdl_index = ifp_p->if_index;
434	sdl.sdl_type = IFT_ETHER;
435	sdl.sdl_alen = ETHER_ADDR_LEN;
436
437	/* First, remove any existing filter entries. */
438	while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
439		bcopy((char *)&mc->mc_addr, LLADDR(&sdl), ETHER_ADDR_LEN);
440		error = if_delmulti(ifp_p, (struct sockaddr *)&sdl);
441		if (error)
442			return (error);
443		SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
444		free(mc, M_VLAN);
445	}
446
447	/* Now program new ones. */
448	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
449		if (ifma->ifma_addr->sa_family != AF_LINK)
450			continue;
451		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
452		if (mc == NULL)
453			return (ENOMEM);
454		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
455		    (char *)&mc->mc_addr, ETHER_ADDR_LEN);
456		SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
457		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
458		    LLADDR(&sdl), ETHER_ADDR_LEN);
459		error = if_addmulti(ifp_p, (struct sockaddr *)&sdl, &rifma);
460		if (error)
461			return (error);
462	}
463
464	CURVNET_RESTORE();
465	return (0);
466}
467
468/*
469 * A handler for network interface departure events.
470 * Track departure of trunks here so that we don't access invalid
471 * pointers or whatever if a trunk is ripped from under us, e.g.,
472 * by ejecting its hot-plug card.
473 */
474static void
475vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
476{
477	struct ifvlan *ifv;
478	int i;
479
480	/*
481	 * Check if it's a trunk interface first of all
482	 * to avoid needless locking.
483	 */
484	if (ifp->if_vlantrunk == NULL)
485		return;
486
487	VLAN_LOCK();
488	/*
489	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
490	 * Check trunk pointer after each vlan_unconfig() as it will
491	 * free it and set to NULL after the last vlan was detached.
492	 */
493#ifdef VLAN_ARRAY
494	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
495		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
496			vlan_unconfig_locked(ifv->ifv_ifp);
497			if (ifp->if_vlantrunk == NULL)
498				break;
499		}
500#else /* VLAN_ARRAY */
501restart:
502	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
503		if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
504			vlan_unconfig_locked(ifv->ifv_ifp);
505			if (ifp->if_vlantrunk)
506				goto restart;	/* trunk->hwidth can change */
507			else
508				break;
509		}
510#endif /* VLAN_ARRAY */
511	/* Trunk should have been destroyed in vlan_unconfig(). */
512	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
513	VLAN_UNLOCK();
514}
515
516/*
517 * VLAN support can be loaded as a module.  The only place in the
518 * system that's intimately aware of this is ether_input.  We hook
519 * into this code through vlan_input_p which is defined there and
520 * set here.  Noone else in the system should be aware of this so
521 * we use an explicit reference here.
522 */
523extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
524
525/* For if_link_state_change() eyes only... */
526extern	void (*vlan_link_state_p)(struct ifnet *, int);
527
528static int
529vlan_modevent(module_t mod, int type, void *data)
530{
531
532	switch (type) {
533	case MOD_LOAD:
534		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
535		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
536		if (ifdetach_tag == NULL)
537			return (ENOMEM);
538		VLAN_LOCK_INIT();
539		vlan_input_p = vlan_input;
540		vlan_link_state_p = vlan_link_state;
541		vlan_trunk_cap_p = vlan_trunk_capabilities;
542		if_clone_attach(&vlan_cloner);
543		if (bootverbose)
544			printf("vlan: initialized, using "
545#ifdef VLAN_ARRAY
546			       "full-size arrays"
547#else
548			       "hash tables with chaining"
549#endif
550
551			       "\n");
552		break;
553	case MOD_UNLOAD:
554		if_clone_detach(&vlan_cloner);
555		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
556		vlan_input_p = NULL;
557		vlan_link_state_p = NULL;
558		vlan_trunk_cap_p = NULL;
559		VLAN_LOCK_DESTROY();
560		if (bootverbose)
561			printf("vlan: unloaded\n");
562		break;
563	default:
564		return (EOPNOTSUPP);
565	}
566	return (0);
567}
568
569static moduledata_t vlan_mod = {
570	"if_vlan",
571	vlan_modevent,
572	0
573};
574
575DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
576MODULE_VERSION(if_vlan, 3);
577
578static struct ifnet *
579vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
580{
581	INIT_VNET_NET(curvnet);
582	const char *cp;
583	struct ifnet *ifp;
584	int t = 0;
585
586	/* Check for <etherif>.<vlan> style interface names. */
587	IFNET_RLOCK();
588	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
589		if (ifp->if_type != IFT_ETHER)
590			continue;
591		if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
592			continue;
593		cp = name + strlen(ifp->if_xname);
594		if (*cp != '.')
595			continue;
596		for(; *cp != '\0'; cp++) {
597			if (*cp < '0' || *cp > '9')
598				continue;
599			t = (t * 10) + (*cp - '0');
600		}
601		if (tag != NULL)
602			*tag = t;
603		break;
604	}
605	IFNET_RUNLOCK();
606
607	return (ifp);
608}
609
610static int
611vlan_clone_match(struct if_clone *ifc, const char *name)
612{
613	const char *cp;
614
615	if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
616		return (1);
617
618	if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
619		return (0);
620	for (cp = name + 4; *cp != '\0'; cp++) {
621		if (*cp < '0' || *cp > '9')
622			return (0);
623	}
624
625	return (1);
626}
627
628static int
629vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
630{
631	char *dp;
632	int wildcard;
633	int unit;
634	int error;
635	int tag;
636	int ethertag;
637	struct ifvlan *ifv;
638	struct ifnet *ifp;
639	struct ifnet *p;
640	struct vlanreq vlr;
641	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
642
643	/*
644	 * There are 3 (ugh) ways to specify the cloned device:
645	 * o pass a parameter block with the clone request.
646	 * o specify parameters in the text of the clone device name
647	 * o specify no parameters and get an unattached device that
648	 *   must be configured separately.
649	 * The first technique is preferred; the latter two are
650	 * supported for backwards compatibilty.
651	 */
652	if (params) {
653		error = copyin(params, &vlr, sizeof(vlr));
654		if (error)
655			return error;
656		p = ifunit(vlr.vlr_parent);
657		if (p == NULL)
658			return ENXIO;
659		/*
660		 * Don't let the caller set up a VLAN tag with
661		 * anything except VLID bits.
662		 */
663		if (vlr.vlr_tag & ~EVL_VLID_MASK)
664			return (EINVAL);
665		error = ifc_name2unit(name, &unit);
666		if (error != 0)
667			return (error);
668
669		ethertag = 1;
670		tag = vlr.vlr_tag;
671		wildcard = (unit < 0);
672	} else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
673		ethertag = 1;
674		unit = -1;
675		wildcard = 0;
676
677		/*
678		 * Don't let the caller set up a VLAN tag with
679		 * anything except VLID bits.
680		 */
681		if (tag & ~EVL_VLID_MASK)
682			return (EINVAL);
683	} else {
684		ethertag = 0;
685
686		error = ifc_name2unit(name, &unit);
687		if (error != 0)
688			return (error);
689
690		wildcard = (unit < 0);
691	}
692
693	error = ifc_alloc_unit(ifc, &unit);
694	if (error != 0)
695		return (error);
696
697	/* In the wildcard case, we need to update the name. */
698	if (wildcard) {
699		for (dp = name; *dp != '\0'; dp++);
700		if (snprintf(dp, len - (dp-name), "%d", unit) >
701		    len - (dp-name) - 1) {
702			panic("%s: interface name too long", __func__);
703		}
704	}
705
706	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
707	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
708	if (ifp == NULL) {
709		ifc_free_unit(ifc, unit);
710		free(ifv, M_VLAN);
711		return (ENOSPC);
712	}
713	SLIST_INIT(&ifv->vlan_mc_listhead);
714
715	ifp->if_softc = ifv;
716	/*
717	 * Set the name manually rather than using if_initname because
718	 * we don't conform to the default naming convention for interfaces.
719	 */
720	strlcpy(ifp->if_xname, name, IFNAMSIZ);
721	ifp->if_dname = ifc->ifc_name;
722	ifp->if_dunit = unit;
723	/* NB: flags are not set here */
724	ifp->if_linkmib = &ifv->ifv_mib;
725	ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
726	/* NB: mtu is not set here */
727
728	ifp->if_init = vlan_init;
729	ifp->if_start = vlan_start;
730	ifp->if_ioctl = vlan_ioctl;
731	ifp->if_snd.ifq_maxlen = ifqmaxlen;
732	ifp->if_flags = VLAN_IFFLAGS;
733	ether_ifattach(ifp, eaddr);
734	/* Now undo some of the damage... */
735	ifp->if_baudrate = 0;
736	ifp->if_type = IFT_L2VLAN;
737	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
738
739	if (ethertag) {
740		error = vlan_config(ifv, p, tag);
741		if (error != 0) {
742			/*
743			 * Since we've partialy failed, we need to back
744			 * out all the way, otherwise userland could get
745			 * confused.  Thus, we destroy the interface.
746			 */
747			ether_ifdetach(ifp);
748			vlan_unconfig(ifp);
749			if_free_type(ifp, IFT_ETHER);
750			ifc_free_unit(ifc, unit);
751			free(ifv, M_VLAN);
752
753			return (error);
754		}
755
756		/* Update flags on the parent, if necessary. */
757		vlan_setflags(ifp, 1);
758	}
759
760	return (0);
761}
762
763static int
764vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
765{
766	struct ifvlan *ifv = ifp->if_softc;
767	int unit = ifp->if_dunit;
768
769	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
770	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
771	if_free_type(ifp, IFT_ETHER);
772	free(ifv, M_VLAN);
773	ifc_free_unit(ifc, unit);
774
775	return (0);
776}
777
778/*
779 * The ifp->if_init entry point for vlan(4) is a no-op.
780 */
781static void
782vlan_init(void *foo __unused)
783{
784}
785
786/*
787 * The if_start method for vlan(4) interface. It doesn't
788 * raises the IFF_DRV_OACTIVE flag, since it is called
789 * only from IFQ_HANDOFF() macro in ether_output_frame().
790 * If the interface queue is full, and vlan_start() is
791 * not called, the queue would never get emptied and
792 * interface would stall forever.
793 */
794static void
795vlan_start(struct ifnet *ifp)
796{
797	struct ifvlan *ifv;
798	struct ifnet *p;
799	struct mbuf *m;
800	int error;
801
802	ifv = ifp->if_softc;
803	p = PARENT(ifv);
804
805	for (;;) {
806		IF_DEQUEUE(&ifp->if_snd, m);
807		if (m == NULL)
808			break;
809		BPF_MTAP(ifp, m);
810
811		/*
812		 * Do not run parent's if_start() if the parent is not up,
813		 * or parent's driver will cause a system crash.
814		 */
815		if (!UP_AND_RUNNING(p)) {
816			m_freem(m);
817			ifp->if_collisions++;
818			continue;
819		}
820
821		/*
822		 * Pad the frame to the minimum size allowed if told to.
823		 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
824		 * paragraph C.4.4.3.b.  It can help to work around buggy
825		 * bridges that violate paragraph C.4.4.3.a from the same
826		 * document, i.e., fail to pad short frames after untagging.
827		 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
828		 * untagging it will produce a 62-byte frame, which is a runt
829		 * and requires padding.  There are VLAN-enabled network
830		 * devices that just discard such runts instead or mishandle
831		 * them somehow.
832		 */
833		if (soft_pad) {
834			static char pad[8];	/* just zeros */
835			int n;
836
837			for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
838			     n > 0; n -= sizeof(pad))
839				if (!m_append(m, min(n, sizeof(pad)), pad))
840					break;
841
842			if (n > 0) {
843				if_printf(ifp, "cannot pad short frame\n");
844				ifp->if_oerrors++;
845				m_freem(m);
846				continue;
847			}
848		}
849
850		/*
851		 * If underlying interface can do VLAN tag insertion itself,
852		 * just pass the packet along. However, we need some way to
853		 * tell the interface where the packet came from so that it
854		 * knows how to find the VLAN tag to use, so we attach a
855		 * packet tag that holds it.
856		 */
857		if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
858			m->m_pkthdr.ether_vtag = ifv->ifv_tag;
859			m->m_flags |= M_VLANTAG;
860		} else {
861			m = ether_vlanencap(m, ifv->ifv_tag);
862			if (m == NULL) {
863				if_printf(ifp,
864				    "unable to prepend VLAN header\n");
865				ifp->if_oerrors++;
866				continue;
867			}
868		}
869
870		/*
871		 * Send it, precisely as ether_output() would have.
872		 * We are already running at splimp.
873		 */
874		error = (p->if_transmit)(p, m);
875		if (!error)
876			ifp->if_opackets++;
877		else
878			ifp->if_oerrors++;
879	}
880}
881
882static void
883vlan_input(struct ifnet *ifp, struct mbuf *m)
884{
885	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
886	struct ifvlan *ifv;
887	uint16_t tag;
888
889	KASSERT(trunk != NULL, ("%s: no trunk", __func__));
890
891	if (m->m_flags & M_VLANTAG) {
892		/*
893		 * Packet is tagged, but m contains a normal
894		 * Ethernet frame; the tag is stored out-of-band.
895		 */
896		tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
897		m->m_flags &= ~M_VLANTAG;
898	} else {
899		struct ether_vlan_header *evl;
900
901		/*
902		 * Packet is tagged in-band as specified by 802.1q.
903		 */
904		switch (ifp->if_type) {
905		case IFT_ETHER:
906			if (m->m_len < sizeof(*evl) &&
907			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
908				if_printf(ifp, "cannot pullup VLAN header\n");
909				return;
910			}
911			evl = mtod(m, struct ether_vlan_header *);
912			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
913
914			/*
915			 * Remove the 802.1q header by copying the Ethernet
916			 * addresses over it and adjusting the beginning of
917			 * the data in the mbuf.  The encapsulated Ethernet
918			 * type field is already in place.
919			 */
920			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
921			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
922			m_adj(m, ETHER_VLAN_ENCAP_LEN);
923			break;
924
925		default:
926#ifdef INVARIANTS
927			panic("%s: %s has unsupported if_type %u",
928			      __func__, ifp->if_xname, ifp->if_type);
929#endif
930			m_freem(m);
931			ifp->if_noproto++;
932			return;
933		}
934	}
935
936	TRUNK_RLOCK(trunk);
937#ifdef VLAN_ARRAY
938	ifv = trunk->vlans[tag];
939#else
940	ifv = vlan_gethash(trunk, tag);
941#endif
942	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
943		TRUNK_RUNLOCK(trunk);
944		m_freem(m);
945		ifp->if_noproto++;
946		return;
947	}
948	TRUNK_RUNLOCK(trunk);
949
950	m->m_pkthdr.rcvif = ifv->ifv_ifp;
951	ifv->ifv_ifp->if_ipackets++;
952
953	/* Pass it back through the parent's input routine. */
954	(*ifp->if_input)(ifv->ifv_ifp, m);
955}
956
957static int
958vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
959{
960	struct ifvlantrunk *trunk;
961	struct ifnet *ifp;
962	int error = 0;
963
964	/* VID numbers 0x0 and 0xFFF are reserved */
965	if (tag == 0 || tag == 0xFFF)
966		return (EINVAL);
967	if (p->if_type != IFT_ETHER)
968		return (EPROTONOSUPPORT);
969	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
970		return (EPROTONOSUPPORT);
971	if (ifv->ifv_trunk)
972		return (EBUSY);
973
974	if (p->if_vlantrunk == NULL) {
975		trunk = malloc(sizeof(struct ifvlantrunk),
976		    M_VLAN, M_WAITOK | M_ZERO);
977#ifndef VLAN_ARRAY
978		vlan_inithash(trunk);
979#endif
980		VLAN_LOCK();
981		if (p->if_vlantrunk != NULL) {
982			/* A race that that is very unlikely to be hit. */
983#ifndef VLAN_ARRAY
984			vlan_freehash(trunk);
985#endif
986			free(trunk, M_VLAN);
987			goto exists;
988		}
989		TRUNK_LOCK_INIT(trunk);
990		TRUNK_LOCK(trunk);
991		p->if_vlantrunk = trunk;
992		trunk->parent = p;
993	} else {
994		VLAN_LOCK();
995exists:
996		trunk = p->if_vlantrunk;
997		TRUNK_LOCK(trunk);
998	}
999
1000	ifv->ifv_tag = tag;	/* must set this before vlan_inshash() */
1001#ifdef VLAN_ARRAY
1002	if (trunk->vlans[tag] != NULL) {
1003		error = EEXIST;
1004		goto done;
1005	}
1006	trunk->vlans[tag] = ifv;
1007	trunk->refcnt++;
1008#else
1009	error = vlan_inshash(trunk, ifv);
1010	if (error)
1011		goto done;
1012#endif
1013	ifv->ifv_proto = ETHERTYPE_VLAN;
1014	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1015	ifv->ifv_mintu = ETHERMIN;
1016	ifv->ifv_pflags = 0;
1017
1018	/*
1019	 * If the parent supports the VLAN_MTU capability,
1020	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1021	 * use it.
1022	 */
1023	if (p->if_capenable & IFCAP_VLAN_MTU) {
1024		/*
1025		 * No need to fudge the MTU since the parent can
1026		 * handle extended frames.
1027		 */
1028		ifv->ifv_mtufudge = 0;
1029	} else {
1030		/*
1031		 * Fudge the MTU by the encapsulation size.  This
1032		 * makes us incompatible with strictly compliant
1033		 * 802.1Q implementations, but allows us to use
1034		 * the feature with other NetBSD implementations,
1035		 * which might still be useful.
1036		 */
1037		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1038	}
1039
1040	ifv->ifv_trunk = trunk;
1041	ifp = ifv->ifv_ifp;
1042	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1043	ifp->if_baudrate = p->if_baudrate;
1044	/*
1045	 * Copy only a selected subset of flags from the parent.
1046	 * Other flags are none of our business.
1047	 */
1048#define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1049	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1050	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1051#undef VLAN_COPY_FLAGS
1052
1053	ifp->if_link_state = p->if_link_state;
1054
1055	vlan_capabilities(ifv);
1056
1057	/*
1058	 * Set up our ``Ethernet address'' to reflect the underlying
1059	 * physical interface's.
1060	 */
1061	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), ETHER_ADDR_LEN);
1062
1063	/*
1064	 * Configure multicast addresses that may already be
1065	 * joined on the vlan device.
1066	 */
1067	(void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1068
1069	/* We are ready for operation now. */
1070	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1071done:
1072	TRUNK_UNLOCK(trunk);
1073	if (error == 0)
1074		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_tag);
1075	VLAN_UNLOCK();
1076
1077	return (error);
1078}
1079
1080static int
1081vlan_unconfig(struct ifnet *ifp)
1082{
1083	int ret;
1084
1085	VLAN_LOCK();
1086	ret = vlan_unconfig_locked(ifp);
1087	VLAN_UNLOCK();
1088	return (ret);
1089}
1090
1091static int
1092vlan_unconfig_locked(struct ifnet *ifp)
1093{
1094	struct ifvlantrunk *trunk;
1095	struct vlan_mc_entry *mc;
1096	struct ifvlan *ifv;
1097	struct ifnet  *parent;
1098	int error;
1099
1100	VLAN_LOCK_ASSERT();
1101
1102	ifv = ifp->if_softc;
1103	trunk = ifv->ifv_trunk;
1104	parent = NULL;
1105
1106	if (trunk != NULL) {
1107		struct sockaddr_dl sdl;
1108
1109		TRUNK_LOCK(trunk);
1110		parent = trunk->parent;
1111
1112		/*
1113		 * Since the interface is being unconfigured, we need to
1114		 * empty the list of multicast groups that we may have joined
1115		 * while we were alive from the parent's list.
1116		 */
1117		bzero((char *)&sdl, sizeof(sdl));
1118		sdl.sdl_len = sizeof(sdl);
1119		sdl.sdl_family = AF_LINK;
1120		sdl.sdl_index = parent->if_index;
1121		sdl.sdl_type = IFT_ETHER;
1122		sdl.sdl_alen = ETHER_ADDR_LEN;
1123
1124		while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1125			bcopy((char *)&mc->mc_addr, LLADDR(&sdl),
1126			    ETHER_ADDR_LEN);
1127			error = if_delmulti(parent, (struct sockaddr *)&sdl);
1128			if (error)
1129				return (error);
1130			SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1131			free(mc, M_VLAN);
1132		}
1133
1134		vlan_setflags(ifp, 0); /* clear special flags on parent */
1135#ifdef VLAN_ARRAY
1136		trunk->vlans[ifv->ifv_tag] = NULL;
1137		trunk->refcnt--;
1138#else
1139		vlan_remhash(trunk, ifv);
1140#endif
1141		ifv->ifv_trunk = NULL;
1142
1143		/*
1144		 * Check if we were the last.
1145		 */
1146		if (trunk->refcnt == 0) {
1147			trunk->parent->if_vlantrunk = NULL;
1148			/*
1149			 * XXXGL: If some ithread has already entered
1150			 * vlan_input() and is now blocked on the trunk
1151			 * lock, then it should preempt us right after
1152			 * unlock and finish its work. Then we will acquire
1153			 * lock again in trunk_destroy().
1154			 */
1155			TRUNK_UNLOCK(trunk);
1156			trunk_destroy(trunk);
1157		} else
1158			TRUNK_UNLOCK(trunk);
1159	}
1160
1161	/* Disconnect from parent. */
1162	if (ifv->ifv_pflags)
1163		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1164	ifp->if_mtu = ETHERMTU;
1165	ifp->if_link_state = LINK_STATE_UNKNOWN;
1166	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1167
1168	/*
1169	 * Only dispatch an event if vlan was
1170	 * attached, otherwise there is nothing
1171	 * to cleanup anyway.
1172	 */
1173	if (parent != NULL)
1174		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_tag);
1175
1176	return (0);
1177}
1178
1179/* Handle a reference counted flag that should be set on the parent as well */
1180static int
1181vlan_setflag(struct ifnet *ifp, int flag, int status,
1182	     int (*func)(struct ifnet *, int))
1183{
1184	struct ifvlan *ifv;
1185	int error;
1186
1187	/* XXX VLAN_LOCK_ASSERT(); */
1188
1189	ifv = ifp->if_softc;
1190	status = status ? (ifp->if_flags & flag) : 0;
1191	/* Now "status" contains the flag value or 0 */
1192
1193	/*
1194	 * See if recorded parent's status is different from what
1195	 * we want it to be.  If it is, flip it.  We record parent's
1196	 * status in ifv_pflags so that we won't clear parent's flag
1197	 * we haven't set.  In fact, we don't clear or set parent's
1198	 * flags directly, but get or release references to them.
1199	 * That's why we can be sure that recorded flags still are
1200	 * in accord with actual parent's flags.
1201	 */
1202	if (status != (ifv->ifv_pflags & flag)) {
1203		error = (*func)(PARENT(ifv), status);
1204		if (error)
1205			return (error);
1206		ifv->ifv_pflags &= ~flag;
1207		ifv->ifv_pflags |= status;
1208	}
1209	return (0);
1210}
1211
1212/*
1213 * Handle IFF_* flags that require certain changes on the parent:
1214 * if "status" is true, update parent's flags respective to our if_flags;
1215 * if "status" is false, forcedly clear the flags set on parent.
1216 */
1217static int
1218vlan_setflags(struct ifnet *ifp, int status)
1219{
1220	int error, i;
1221
1222	for (i = 0; vlan_pflags[i].flag; i++) {
1223		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1224				     status, vlan_pflags[i].func);
1225		if (error)
1226			return (error);
1227	}
1228	return (0);
1229}
1230
1231/* Inform all vlans that their parent has changed link state */
1232static void
1233vlan_link_state(struct ifnet *ifp, int link)
1234{
1235	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1236	struct ifvlan *ifv;
1237	int i;
1238
1239	TRUNK_LOCK(trunk);
1240#ifdef VLAN_ARRAY
1241	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1242		if (trunk->vlans[i] != NULL) {
1243			ifv = trunk->vlans[i];
1244#else
1245	for (i = 0; i < (1 << trunk->hwidth); i++)
1246		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1247#endif
1248			ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1249			if_link_state_change(ifv->ifv_ifp,
1250			    trunk->parent->if_link_state);
1251		}
1252	TRUNK_UNLOCK(trunk);
1253}
1254
1255static void
1256vlan_capabilities(struct ifvlan *ifv)
1257{
1258	struct ifnet *p = PARENT(ifv);
1259	struct ifnet *ifp = ifv->ifv_ifp;
1260
1261	TRUNK_LOCK_ASSERT(TRUNK(ifv));
1262
1263	/*
1264	 * If the parent interface can do checksum offloading
1265	 * on VLANs, then propagate its hardware-assisted
1266	 * checksumming flags. Also assert that checksum
1267	 * offloading requires hardware VLAN tagging.
1268	 */
1269	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1270		ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1271
1272	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1273	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1274		ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1275		ifp->if_hwassist = p->if_hwassist;
1276	} else {
1277		ifp->if_capenable = 0;
1278		ifp->if_hwassist = 0;
1279	}
1280}
1281
1282static void
1283vlan_trunk_capabilities(struct ifnet *ifp)
1284{
1285	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1286	struct ifvlan *ifv;
1287	int i;
1288
1289	TRUNK_LOCK(trunk);
1290#ifdef VLAN_ARRAY
1291	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1292		if (trunk->vlans[i] != NULL) {
1293			ifv = trunk->vlans[i];
1294#else
1295	for (i = 0; i < (1 << trunk->hwidth); i++) {
1296		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1297#endif
1298			vlan_capabilities(ifv);
1299	}
1300	TRUNK_UNLOCK(trunk);
1301}
1302
1303static int
1304vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1305{
1306	struct ifnet *p;
1307	struct ifreq *ifr;
1308	struct ifvlan *ifv;
1309	struct vlanreq vlr;
1310	int error = 0;
1311
1312	ifr = (struct ifreq *)data;
1313	ifv = ifp->if_softc;
1314
1315	switch (cmd) {
1316	case SIOCGIFMEDIA:
1317		VLAN_LOCK();
1318		if (TRUNK(ifv) != NULL) {
1319			error = (*PARENT(ifv)->if_ioctl)(PARENT(ifv),
1320					SIOCGIFMEDIA, data);
1321			VLAN_UNLOCK();
1322			/* Limit the result to the parent's current config. */
1323			if (error == 0) {
1324				struct ifmediareq *ifmr;
1325
1326				ifmr = (struct ifmediareq *)data;
1327				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1328					ifmr->ifm_count = 1;
1329					error = copyout(&ifmr->ifm_current,
1330						ifmr->ifm_ulist,
1331						sizeof(int));
1332				}
1333			}
1334		} else {
1335			VLAN_UNLOCK();
1336			error = EINVAL;
1337		}
1338		break;
1339
1340	case SIOCSIFMEDIA:
1341		error = EINVAL;
1342		break;
1343
1344	case SIOCSIFMTU:
1345		/*
1346		 * Set the interface MTU.
1347		 */
1348		VLAN_LOCK();
1349		if (TRUNK(ifv) != NULL) {
1350			if (ifr->ifr_mtu >
1351			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1352			    ifr->ifr_mtu <
1353			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1354				error = EINVAL;
1355			else
1356				ifp->if_mtu = ifr->ifr_mtu;
1357		} else
1358			error = EINVAL;
1359		VLAN_UNLOCK();
1360		break;
1361
1362	case SIOCSETVLAN:
1363		error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1364		if (error)
1365			break;
1366		if (vlr.vlr_parent[0] == '\0') {
1367			vlan_unconfig(ifp);
1368			break;
1369		}
1370		p = ifunit(vlr.vlr_parent);
1371		if (p == 0) {
1372			error = ENOENT;
1373			break;
1374		}
1375		/*
1376		 * Don't let the caller set up a VLAN tag with
1377		 * anything except VLID bits.
1378		 */
1379		if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1380			error = EINVAL;
1381			break;
1382		}
1383		error = vlan_config(ifv, p, vlr.vlr_tag);
1384		if (error)
1385			break;
1386
1387		/* Update flags on the parent, if necessary. */
1388		vlan_setflags(ifp, 1);
1389		break;
1390
1391	case SIOCGETVLAN:
1392		bzero(&vlr, sizeof(vlr));
1393		VLAN_LOCK();
1394		if (TRUNK(ifv) != NULL) {
1395			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1396			    sizeof(vlr.vlr_parent));
1397			vlr.vlr_tag = ifv->ifv_tag;
1398		}
1399		VLAN_UNLOCK();
1400		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1401		break;
1402
1403	case SIOCSIFFLAGS:
1404		/*
1405		 * We should propagate selected flags to the parent,
1406		 * e.g., promiscuous mode.
1407		 */
1408		if (TRUNK(ifv) != NULL)
1409			error = vlan_setflags(ifp, 1);
1410		break;
1411
1412	case SIOCADDMULTI:
1413	case SIOCDELMULTI:
1414		/*
1415		 * If we don't have a parent, just remember the membership for
1416		 * when we do.
1417		 */
1418		if (TRUNK(ifv) != NULL)
1419			error = vlan_setmulti(ifp);
1420		break;
1421
1422	default:
1423		error = ether_ioctl(ifp, cmd, data);
1424	}
1425
1426	return (error);
1427}
1428