if_vlan.c revision 165662
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 165662 2006-12-30 21:10:25Z yar $
30 */
31
32/*
33 * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
34 * Might be extended some day to also handle IEEE 802.1p priority
35 * tagging.  This is sort of sneaky in the implementation, since
36 * we need to pretend to be enough of an Ethernet implementation
37 * to make arp work.  The way we do this is by telling everyone
38 * that we are an Ethernet, and then catch the packets that
39 * ether_output() left on our output queue when it calls
40 * if_start(), rewrite them for use by the real outgoing interface,
41 * and ask it to send them.
42 */
43
44#include "opt_inet.h"
45#include "opt_vlan.h"
46
47#include <sys/param.h>
48#include <sys/kernel.h>
49#include <sys/lock.h>
50#include <sys/malloc.h>
51#include <sys/mbuf.h>
52#include <sys/module.h>
53#include <sys/rwlock.h>
54#include <sys/queue.h>
55#include <sys/socket.h>
56#include <sys/sockio.h>
57#include <sys/sysctl.h>
58#include <sys/systm.h>
59
60#include <net/bpf.h>
61#include <net/ethernet.h>
62#include <net/if.h>
63#include <net/if_clone.h>
64#include <net/if_arp.h>
65#include <net/if_dl.h>
66#include <net/if_types.h>
67#include <net/if_vlan_var.h>
68
69#ifdef INET
70#include <netinet/in.h>
71#include <netinet/if_ether.h>
72#endif
73
74#define VLANNAME	"vlan"
75#define	VLAN_DEF_HWIDTH	4
76#define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
77
78#define	UP_AND_RUNNING(ifp) \
79    ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
80
81LIST_HEAD(ifvlanhead, ifvlan);
82
83struct ifvlantrunk {
84	struct	ifnet   *parent;	/* parent interface of this trunk */
85	struct	rwlock	rw;
86#ifdef VLAN_ARRAY
87#define	VLAN_ARRAY_SIZE	(EVL_VLID_MASK + 1)
88	struct	ifvlan	*vlans[VLAN_ARRAY_SIZE]; /* static table */
89#else
90	struct	ifvlanhead *hash;	/* dynamic hash-list table */
91	uint16_t	hmask;
92	uint16_t	hwidth;
93#endif
94	int		refcnt;
95	LIST_ENTRY(ifvlantrunk) trunk_entry;
96};
97static LIST_HEAD(, ifvlantrunk) trunk_list;
98
99struct vlan_mc_entry {
100	struct ether_addr		mc_addr;
101	SLIST_ENTRY(vlan_mc_entry)	mc_entries;
102};
103
104struct	ifvlan {
105	struct	ifvlantrunk *ifv_trunk;
106	struct	ifnet *ifv_ifp;
107#define	TRUNK(ifv)	((ifv)->ifv_trunk)
108#define	PARENT(ifv)	((ifv)->ifv_trunk->parent)
109	int	ifv_pflags;	/* special flags we have set on parent */
110	struct	ifv_linkmib {
111		int	ifvm_encaplen;	/* encapsulation length */
112		int	ifvm_mtufudge;	/* MTU fudged by this much */
113		int	ifvm_mintu;	/* min transmission unit */
114		uint16_t ifvm_proto;	/* encapsulation ethertype */
115		uint16_t ifvm_tag;	/* tag to apply on packets leaving if */
116	}	ifv_mib;
117	SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
118	LIST_ENTRY(ifvlan) ifv_list;
119};
120#define	ifv_proto	ifv_mib.ifvm_proto
121#define	ifv_tag		ifv_mib.ifvm_tag
122#define	ifv_encaplen	ifv_mib.ifvm_encaplen
123#define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
124#define	ifv_mintu	ifv_mib.ifvm_mintu
125
126/* Special flags we should propagate to parent. */
127static struct {
128	int flag;
129	int (*func)(struct ifnet *, int);
130} vlan_pflags[] = {
131	{IFF_PROMISC, ifpromisc},
132	{IFF_ALLMULTI, if_allmulti},
133	{0, NULL}
134};
135
136SYSCTL_DECL(_net_link);
137SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0, "IEEE 802.1Q VLAN");
138SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0, "for consistency");
139
140static int soft_pad = 0;
141SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW, &soft_pad, 0,
142	   "pad short frames before tagging");
143
144static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface");
145
146static eventhandler_tag ifdetach_tag;
147
148/*
149 * We have a global mutex, that is used to serialize configuration
150 * changes and isn't used in normal packet delivery.
151 *
152 * We also have a per-trunk rwlock, that is locked shared on packet
153 * processing and exclusive when configuration is changed.
154 *
155 * The VLAN_ARRAY substitutes the dynamic hash with a static array
156 * with 4096 entries. In theory this can give a boost in processing,
157 * however on practice it does not. Probably this is because array
158 * is too big to fit into CPU cache.
159 */
160static struct mtx ifv_mtx;
161#define	VLAN_LOCK_INIT()	mtx_init(&ifv_mtx, "vlan_global", NULL, MTX_DEF)
162#define	VLAN_LOCK_DESTROY()	mtx_destroy(&ifv_mtx)
163#define	VLAN_LOCK_ASSERT()	mtx_assert(&ifv_mtx, MA_OWNED)
164#define	VLAN_LOCK()		mtx_lock(&ifv_mtx)
165#define	VLAN_UNLOCK()		mtx_unlock(&ifv_mtx)
166#define	TRUNK_LOCK_INIT(trunk)	rw_init(&(trunk)->rw, VLANNAME)
167#define	TRUNK_LOCK_DESTROY(trunk) rw_destroy(&(trunk)->rw)
168#define	TRUNK_LOCK(trunk)	rw_wlock(&(trunk)->rw)
169#define	TRUNK_UNLOCK(trunk)	rw_wunlock(&(trunk)->rw)
170#define	TRUNK_LOCK_ASSERT(trunk) rw_assert(&(trunk)->rw, RA_WLOCKED)
171#define	TRUNK_RLOCK(trunk)	rw_rlock(&(trunk)->rw)
172#define	TRUNK_RUNLOCK(trunk)	rw_runlock(&(trunk)->rw)
173#define	TRUNK_LOCK_RASSERT(trunk) rw_assert(&(trunk)->rw, RA_RLOCKED)
174
175#ifndef VLAN_ARRAY
176static	void vlan_inithash(struct ifvlantrunk *trunk);
177static	void vlan_freehash(struct ifvlantrunk *trunk);
178static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
179static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
180static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
181static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
182	uint16_t tag);
183#endif
184static	void trunk_destroy(struct ifvlantrunk *trunk);
185
186static	void vlan_start(struct ifnet *ifp);
187static	void vlan_init(void *foo);
188static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
189static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
190static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
191    int (*func)(struct ifnet *, int));
192static	int vlan_setflags(struct ifnet *ifp, int status);
193static	int vlan_setmulti(struct ifnet *ifp);
194static	int vlan_unconfig(struct ifnet *ifp);
195static	int vlan_unconfig_locked(struct ifnet *ifp);
196static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
197static	void vlan_link_state(struct ifnet *ifp, int link);
198static	void vlan_capabilities(struct ifvlan *ifv);
199static	void vlan_trunk_capabilities(struct ifnet *ifp);
200
201static	struct ifnet *vlan_clone_match_ethertag(struct if_clone *,
202    const char *, int *);
203static	int vlan_clone_match(struct if_clone *, const char *);
204static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
205static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
206
207static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
208
209static	struct if_clone vlan_cloner = IFC_CLONE_INITIALIZER(VLANNAME, NULL,
210    IF_MAXUNIT, NULL, vlan_clone_match, vlan_clone_create, vlan_clone_destroy);
211
212#ifndef VLAN_ARRAY
213#define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
214
215static void
216vlan_inithash(struct ifvlantrunk *trunk)
217{
218	int i, n;
219
220	/*
221	 * The trunk must not be locked here since we call malloc(M_WAITOK).
222	 * It is OK in case this function is called before the trunk struct
223	 * gets hooked up and becomes visible from other threads.
224	 */
225
226	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
227	    ("%s: hash already initialized", __func__));
228
229	trunk->hwidth = VLAN_DEF_HWIDTH;
230	n = 1 << trunk->hwidth;
231	trunk->hmask = n - 1;
232	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
233	for (i = 0; i < n; i++)
234		LIST_INIT(&trunk->hash[i]);
235}
236
237static void
238vlan_freehash(struct ifvlantrunk *trunk)
239{
240#ifdef INVARIANTS
241	int i;
242
243	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
244	for (i = 0; i < (1 << trunk->hwidth); i++)
245		KASSERT(LIST_EMPTY(&trunk->hash[i]),
246		    ("%s: hash table not empty", __func__));
247#endif
248	free(trunk->hash, M_VLAN);
249	trunk->hash = NULL;
250	trunk->hwidth = trunk->hmask = 0;
251}
252
253static int
254vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
255{
256	int i, b;
257	struct ifvlan *ifv2;
258
259	TRUNK_LOCK_ASSERT(trunk);
260	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
261
262	b = 1 << trunk->hwidth;
263	i = HASH(ifv->ifv_tag, trunk->hmask);
264	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
265		if (ifv->ifv_tag == ifv2->ifv_tag)
266			return (EEXIST);
267
268	/*
269	 * Grow the hash when the number of vlans exceeds half of the number of
270	 * hash buckets squared. This will make the average linked-list length
271	 * buckets/2.
272	 */
273	if (trunk->refcnt > (b * b) / 2) {
274		vlan_growhash(trunk, 1);
275		i = HASH(ifv->ifv_tag, trunk->hmask);
276	}
277	LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
278	trunk->refcnt++;
279
280	return (0);
281}
282
283static int
284vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
285{
286	int i, b;
287	struct ifvlan *ifv2;
288
289	TRUNK_LOCK_ASSERT(trunk);
290	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
291
292	b = 1 << trunk->hwidth;
293	i = HASH(ifv->ifv_tag, trunk->hmask);
294	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
295		if (ifv2 == ifv) {
296			trunk->refcnt--;
297			LIST_REMOVE(ifv2, ifv_list);
298			if (trunk->refcnt < (b * b) / 2)
299				vlan_growhash(trunk, -1);
300			return (0);
301		}
302
303	panic("%s: vlan not found\n", __func__);
304	return (ENOENT); /*NOTREACHED*/
305}
306
307/*
308 * Grow the hash larger or smaller if memory permits.
309 */
310static void
311vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
312{
313
314	struct ifvlan *ifv;
315	struct ifvlanhead *hash2;
316	int hwidth2, i, j, n, n2;
317
318	TRUNK_LOCK_ASSERT(trunk);
319	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
320
321	if (howmuch == 0) {
322		/* Harmless yet obvious coding error */
323		printf("%s: howmuch is 0\n", __func__);
324		return;
325	}
326
327	hwidth2 = trunk->hwidth + howmuch;
328	n = 1 << trunk->hwidth;
329	n2 = 1 << hwidth2;
330	/* Do not shrink the table below the default */
331	if (hwidth2 < VLAN_DEF_HWIDTH)
332		return;
333
334	/* M_NOWAIT because we're called with trunk mutex held */
335	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
336	if (hash2 == NULL) {
337		printf("%s: out of memory -- hash size not changed\n",
338		    __func__);
339		return;		/* We can live with the old hash table */
340	}
341	for (j = 0; j < n2; j++)
342		LIST_INIT(&hash2[j]);
343	for (i = 0; i < n; i++)
344		while (!LIST_EMPTY(&trunk->hash[i])) {
345			ifv = LIST_FIRST(&trunk->hash[i]);
346			LIST_REMOVE(ifv, ifv_list);
347			j = HASH(ifv->ifv_tag, n2 - 1);
348			LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
349		}
350	free(trunk->hash, M_VLAN);
351	trunk->hash = hash2;
352	trunk->hwidth = hwidth2;
353	trunk->hmask = n2 - 1;
354}
355
356static __inline struct ifvlan *
357vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
358{
359	struct ifvlan *ifv;
360
361	TRUNK_LOCK_RASSERT(trunk);
362
363	LIST_FOREACH(ifv, &trunk->hash[HASH(tag, trunk->hmask)], ifv_list)
364		if (ifv->ifv_tag == tag)
365			return (ifv);
366	return (NULL);
367}
368
369#if 0
370/* Debugging code to view the hashtables. */
371static void
372vlan_dumphash(struct ifvlantrunk *trunk)
373{
374	int i;
375	struct ifvlan *ifv;
376
377	for (i = 0; i < (1 << trunk->hwidth); i++) {
378		printf("%d: ", i);
379		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
380			printf("%s ", ifv->ifv_ifp->if_xname);
381		printf("\n");
382	}
383}
384#endif /* 0 */
385#endif /* !VLAN_ARRAY */
386
387static void
388trunk_destroy(struct ifvlantrunk *trunk)
389{
390	VLAN_LOCK_ASSERT();
391
392	TRUNK_LOCK(trunk);
393#ifndef VLAN_ARRAY
394	vlan_freehash(trunk);
395#endif
396	trunk->parent->if_vlantrunk = NULL;
397	LIST_REMOVE(trunk, trunk_entry);
398	TRUNK_UNLOCK(trunk);
399	TRUNK_LOCK_DESTROY(trunk);
400	free(trunk, M_VLAN);
401}
402
403/*
404 * Program our multicast filter. What we're actually doing is
405 * programming the multicast filter of the parent. This has the
406 * side effect of causing the parent interface to receive multicast
407 * traffic that it doesn't really want, which ends up being discarded
408 * later by the upper protocol layers. Unfortunately, there's no way
409 * to avoid this: there really is only one physical interface.
410 *
411 * XXX: There is a possible race here if more than one thread is
412 *      modifying the multicast state of the vlan interface at the same time.
413 */
414static int
415vlan_setmulti(struct ifnet *ifp)
416{
417	struct ifnet		*ifp_p;
418	struct ifmultiaddr	*ifma, *rifma = NULL;
419	struct ifvlan		*sc;
420	struct vlan_mc_entry	*mc = NULL;
421	struct sockaddr_dl	sdl;
422	int			error;
423
424	/*VLAN_LOCK_ASSERT();*/
425
426	/* Find the parent. */
427	sc = ifp->if_softc;
428	ifp_p = PARENT(sc);
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 (SLIST_FIRST(&sc->vlan_mc_listhead) != NULL) {
439		mc = SLIST_FIRST(&sc->vlan_mc_listhead);
440		bcopy((char *)&mc->mc_addr, LLADDR(&sdl), ETHER_ADDR_LEN);
441		error = if_delmulti(ifp_p, (struct sockaddr *)&sdl);
442		if (error)
443			return (error);
444		SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
445		free(mc, M_VLAN);
446	}
447
448	/* Now program new ones. */
449	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
450		if (ifma->ifma_addr->sa_family != AF_LINK)
451			continue;
452		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
453		if (mc == NULL)
454			return (ENOMEM);
455		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
456		    (char *)&mc->mc_addr, ETHER_ADDR_LEN);
457		SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
458		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
459		    LLADDR(&sdl), ETHER_ADDR_LEN);
460		error = if_addmulti(ifp_p, (struct sockaddr *)&sdl, &rifma);
461		if (error)
462			return (error);
463	}
464
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		LIST_INIT(&trunk_list);
539		VLAN_LOCK_INIT();
540		vlan_input_p = vlan_input;
541		vlan_link_state_p = vlan_link_state;
542		vlan_trunk_cap_p = vlan_trunk_capabilities;
543		if_clone_attach(&vlan_cloner);
544		break;
545	case MOD_UNLOAD:
546	    {
547		struct ifvlantrunk *trunk, *trunk1;
548
549		if_clone_detach(&vlan_cloner);
550		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
551		vlan_input_p = NULL;
552		vlan_link_state_p = NULL;
553		vlan_trunk_cap_p = NULL;
554		VLAN_LOCK();
555		LIST_FOREACH_SAFE(trunk, &trunk_list, trunk_entry, trunk1)
556			trunk_destroy(trunk);
557		VLAN_UNLOCK();
558		VLAN_LOCK_DESTROY();
559		break;
560	    }
561	default:
562		return (EOPNOTSUPP);
563	}
564	return (0);
565}
566
567static moduledata_t vlan_mod = {
568	"if_vlan",
569	vlan_modevent,
570	0
571};
572
573DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
574MODULE_VERSION(if_vlan, 3);
575MODULE_DEPEND(if_vlan, miibus, 1, 1, 1);
576
577static struct ifnet *
578vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
579{
580	const char *cp;
581	struct ifnet *ifp;
582	int t = 0;
583
584	/* Check for <etherif>.<vlan> style interface names. */
585	IFNET_RLOCK();
586	TAILQ_FOREACH(ifp, &ifnet, if_link) {
587		if (ifp->if_type != IFT_ETHER)
588			continue;
589		if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
590			continue;
591		cp = name + strlen(ifp->if_xname);
592		if (*cp != '.')
593			continue;
594		for(; *cp != '\0'; cp++) {
595			if (*cp < '0' || *cp > '9')
596				continue;
597			t = (t * 10) + (*cp - '0');
598		}
599		if (tag != NULL)
600			*tag = t;
601		break;
602	}
603	IFNET_RUNLOCK();
604
605	return (ifp);
606}
607
608static int
609vlan_clone_match(struct if_clone *ifc, const char *name)
610{
611	const char *cp;
612
613	if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
614		return (1);
615
616	if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
617		return (0);
618	for (cp = name + 4; *cp != '\0'; cp++) {
619		if (*cp < '0' || *cp > '9')
620			return (0);
621	}
622
623	return (1);
624}
625
626static int
627vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
628{
629	char *dp;
630	int wildcard;
631	int unit;
632	int error;
633	int tag;
634	int ethertag;
635	struct ifvlan *ifv;
636	struct ifnet *ifp;
637	struct ifnet *p;
638	struct vlanreq vlr;
639	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
640
641	/*
642	 * There are 3 (ugh) ways to specify the cloned device:
643	 * o pass a parameter block with the clone request.
644	 * o specify parameters in the text of the clone device name
645	 * o specify no parameters and get an unattached device that
646	 *   must be configured separately.
647	 * The first technique is preferred; the latter two are
648	 * supported for backwards compatibilty.
649	 */
650	if (params) {
651		error = copyin(params, &vlr, sizeof(vlr));
652		if (error)
653			return error;
654		p = ifunit(vlr.vlr_parent);
655		if (p == NULL)
656			return ENXIO;
657		/*
658		 * Don't let the caller set up a VLAN tag with
659		 * anything except VLID bits.
660		 */
661		if (vlr.vlr_tag & ~EVL_VLID_MASK)
662			return (EINVAL);
663		error = ifc_name2unit(name, &unit);
664		if (error != 0)
665			return (error);
666
667		ethertag = 1;
668		tag = vlr.vlr_tag;
669		wildcard = (unit < 0);
670	} else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
671		ethertag = 1;
672		unit = -1;
673		wildcard = 0;
674
675		/*
676		 * Don't let the caller set up a VLAN tag with
677		 * anything except VLID bits.
678		 */
679		if (tag & ~EVL_VLID_MASK)
680			return (EINVAL);
681	} else {
682		ethertag = 0;
683
684		error = ifc_name2unit(name, &unit);
685		if (error != 0)
686			return (error);
687
688		wildcard = (unit < 0);
689	}
690
691	error = ifc_alloc_unit(ifc, &unit);
692	if (error != 0)
693		return (error);
694
695	/* In the wildcard case, we need to update the name. */
696	if (wildcard) {
697		for (dp = name; *dp != '\0'; dp++);
698		if (snprintf(dp, len - (dp-name), "%d", unit) >
699		    len - (dp-name) - 1) {
700			panic("%s: interface name too long", __func__);
701		}
702	}
703
704	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
705	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
706	if (ifp == NULL) {
707		ifc_free_unit(ifc, unit);
708		free(ifv, M_VLAN);
709		return (ENOSPC);
710	}
711	SLIST_INIT(&ifv->vlan_mc_listhead);
712
713	ifp->if_softc = ifv;
714	/*
715	 * Set the name manually rather than using if_initname because
716	 * we don't conform to the default naming convention for interfaces.
717	 */
718	strlcpy(ifp->if_xname, name, IFNAMSIZ);
719	ifp->if_dname = ifc->ifc_name;
720	ifp->if_dunit = unit;
721	/* NB: flags are not set here */
722	ifp->if_linkmib = &ifv->ifv_mib;
723	ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
724	/* NB: mtu is not set here */
725
726	ifp->if_init = vlan_init;
727	ifp->if_start = vlan_start;
728	ifp->if_ioctl = vlan_ioctl;
729	ifp->if_snd.ifq_maxlen = ifqmaxlen;
730	ifp->if_flags = VLAN_IFFLAGS;
731	ether_ifattach(ifp, eaddr);
732	/* Now undo some of the damage... */
733	ifp->if_baudrate = 0;
734	ifp->if_type = IFT_L2VLAN;
735	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
736
737	if (ethertag) {
738		error = vlan_config(ifv, p, tag);
739		if (error != 0) {
740			/*
741			 * Since we've partialy failed, we need to back
742			 * out all the way, otherwise userland could get
743			 * confused.  Thus, we destroy the interface.
744			 */
745			ether_ifdetach(ifp);
746			vlan_unconfig(ifp);
747			if_free_type(ifp, IFT_ETHER);
748			free(ifv, M_VLAN);
749
750			return (error);
751		}
752
753		/* Update flags on the parent, if necessary. */
754		vlan_setflags(ifp, 1);
755	}
756
757	return (0);
758}
759
760static int
761vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
762{
763	struct ifvlan *ifv = ifp->if_softc;
764	int unit = ifp->if_dunit;
765
766	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
767	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
768	if_free_type(ifp, IFT_ETHER);
769	free(ifv, M_VLAN);
770	ifc_free_unit(ifc, unit);
771
772	return (0);
773}
774
775/*
776 * The ifp->if_init entry point for vlan(4) is a no-op.
777 */
778static void
779vlan_init(void *foo __unused)
780{
781}
782
783/*
784 * The if_start method for vlan(4) interface. It doesn't
785 * raises the IFF_DRV_OACTIVE flag, since it is called
786 * only from IFQ_HANDOFF() macro in ether_output_frame().
787 * If the interface queue is full, and vlan_start() is
788 * not called, the queue would never get emptied and
789 * interface would stall forever.
790 */
791static void
792vlan_start(struct ifnet *ifp)
793{
794	struct ifvlan *ifv;
795	struct ifnet *p;
796	struct mbuf *m;
797	int error;
798
799	ifv = ifp->if_softc;
800	p = PARENT(ifv);
801
802	for (;;) {
803		IF_DEQUEUE(&ifp->if_snd, m);
804		if (m == NULL)
805			break;
806		BPF_MTAP(ifp, m);
807
808		/*
809		 * Do not run parent's if_start() if the parent is not up,
810		 * or parent's driver will cause a system crash.
811		 */
812		if (!UP_AND_RUNNING(p)) {
813			m_freem(m);
814			ifp->if_collisions++;
815			continue;
816		}
817
818		/*
819		 * Pad the frame to the minimum size allowed if told to.
820		 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
821		 * paragraph C.4.4.3.b.  It can help to work around buggy
822		 * bridges that violate paragraph C.4.4.3.a from the same
823		 * document, i.e., fail to pad short frames after untagging.
824		 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
825		 * untagging it will produce a 62-byte frame, which is a runt
826		 * and requires padding.  There are VLAN-enabled network
827		 * devices that just discard such runts instead or mishandle
828		 * them somehow.
829		 */
830		if (soft_pad) {
831			static char pad[8];	/* just zeros */
832			int n;
833
834			for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
835			     n > 0; n -= sizeof(pad))
836				if (!m_append(m, min(n, sizeof(pad)), pad))
837					break;
838
839			if (n > 0) {
840				if_printf(ifp, "cannot pad short frame\n");
841				ifp->if_oerrors++;
842				m_freem(m);
843				continue;
844			}
845		}
846
847		/*
848		 * If underlying interface can do VLAN tag insertion itself,
849		 * just pass the packet along. However, we need some way to
850		 * tell the interface where the packet came from so that it
851		 * knows how to find the VLAN tag to use, so we attach a
852		 * packet tag that holds it.
853		 */
854		if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
855			m->m_pkthdr.ether_vtag = ifv->ifv_tag;
856			m->m_flags |= M_VLANTAG;
857		} else {
858			struct ether_vlan_header *evl;
859
860			M_PREPEND(m, ifv->ifv_encaplen, M_DONTWAIT);
861			if (m == NULL) {
862				if_printf(ifp,
863				    "unable to prepend VLAN header\n");
864				ifp->if_oerrors++;
865				continue;
866			}
867			/* M_PREPEND takes care of m_len, m_pkthdr.len for us */
868
869			if (m->m_len < sizeof(*evl)) {
870				m = m_pullup(m, sizeof(*evl));
871				if (m == NULL) {
872					if_printf(ifp,
873					    "cannot pullup VLAN header\n");
874					ifp->if_oerrors++;
875					continue;
876				}
877			}
878
879			/*
880			 * Transform the Ethernet header into an Ethernet header
881			 * with 802.1Q encapsulation.
882			 */
883			evl = mtod(m, struct ether_vlan_header *);
884			bcopy((char *)evl + ifv->ifv_encaplen,
885			      (char *)evl, ETHER_HDR_LEN - ETHER_TYPE_LEN);
886			evl->evl_encap_proto = htons(ifv->ifv_proto);
887			evl->evl_tag = htons(ifv->ifv_tag);
888#ifdef DEBUG
889			printf("%s: %*D\n", __func__, (int)sizeof(*evl),
890			    (unsigned char *)evl, ":");
891#endif
892		}
893
894		/*
895		 * Send it, precisely as ether_output() would have.
896		 * We are already running at splimp.
897		 */
898		IFQ_HANDOFF(p, m, error);
899		if (!error)
900			ifp->if_opackets++;
901		else
902			ifp->if_oerrors++;
903	}
904}
905
906static void
907vlan_input(struct ifnet *ifp, struct mbuf *m)
908{
909	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
910	struct ifvlan *ifv;
911	uint16_t tag;
912
913	KASSERT(trunk != NULL, ("%s: no trunk", __func__));
914
915	if (m->m_flags & M_VLANTAG) {
916		/*
917		 * Packet is tagged, but m contains a normal
918		 * Ethernet frame; the tag is stored out-of-band.
919		 */
920		tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
921		m->m_flags &= ~M_VLANTAG;
922	} else {
923		struct ether_vlan_header *evl;
924
925		/*
926		 * Packet is tagged in-band as specified by 802.1q.
927		 */
928		switch (ifp->if_type) {
929		case IFT_ETHER:
930			if (m->m_len < sizeof(*evl) &&
931			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
932				if_printf(ifp, "cannot pullup VLAN header\n");
933				return;
934			}
935			evl = mtod(m, struct ether_vlan_header *);
936			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
937
938			/*
939			 * Remove the 802.1q header by copying the Ethernet
940			 * addresses over it and adjusting the beginning of
941			 * the data in the mbuf.  The encapsulated Ethernet
942			 * type field is already in place.
943			 */
944			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
945			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
946			m_adj(m, ETHER_VLAN_ENCAP_LEN);
947			break;
948
949		default:
950#ifdef INVARIANTS
951			panic("%s: %s has unsupported if_type %u",
952			      __func__, ifp->if_xname, ifp->if_type);
953#endif
954			m_freem(m);
955			ifp->if_noproto++;
956			return;
957		}
958	}
959
960	TRUNK_RLOCK(trunk);
961#ifdef VLAN_ARRAY
962	ifv = trunk->vlans[tag];
963#else
964	ifv = vlan_gethash(trunk, tag);
965#endif
966	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
967		TRUNK_RUNLOCK(trunk);
968		m_freem(m);
969		ifp->if_noproto++;
970		return;
971	}
972	TRUNK_RUNLOCK(trunk);
973
974	m->m_pkthdr.rcvif = ifv->ifv_ifp;
975	ifv->ifv_ifp->if_ipackets++;
976
977	/* Pass it back through the parent's input routine. */
978	(*ifp->if_input)(ifv->ifv_ifp, m);
979}
980
981static int
982vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
983{
984	struct ifvlantrunk *trunk;
985	struct ifnet *ifp;
986	int error = 0;
987
988	/* VID numbers 0x0 and 0xFFF are reserved */
989	if (tag == 0 || tag == 0xFFF)
990		return (EINVAL);
991	if (p->if_type != IFT_ETHER)
992		return (EPROTONOSUPPORT);
993	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
994		return (EPROTONOSUPPORT);
995	if (ifv->ifv_trunk)
996		return (EBUSY);
997
998	if (p->if_vlantrunk == NULL) {
999		trunk = malloc(sizeof(struct ifvlantrunk),
1000		    M_VLAN, M_WAITOK | M_ZERO);
1001#ifndef VLAN_ARRAY
1002		vlan_inithash(trunk);
1003#endif
1004		VLAN_LOCK();
1005		if (p->if_vlantrunk != NULL) {
1006			/* A race that that is very unlikely to be hit. */
1007#ifndef VLAN_ARRAY
1008			vlan_freehash(trunk);
1009#endif
1010			free(trunk, M_VLAN);
1011			goto exists;
1012		}
1013		TRUNK_LOCK_INIT(trunk);
1014		LIST_INSERT_HEAD(&trunk_list, trunk, trunk_entry);
1015		TRUNK_LOCK(trunk);
1016		p->if_vlantrunk = trunk;
1017		trunk->parent = p;
1018	} else {
1019		VLAN_LOCK();
1020exists:
1021		trunk = p->if_vlantrunk;
1022		TRUNK_LOCK(trunk);
1023	}
1024
1025	ifv->ifv_tag = tag;	/* must set this before vlan_inshash() */
1026#ifdef VLAN_ARRAY
1027	if (trunk->vlans[tag] != NULL) {
1028		error = EEXIST;
1029		goto done;
1030	}
1031	trunk->vlans[tag] = ifv;
1032	trunk->refcnt++;
1033#else
1034	error = vlan_inshash(trunk, ifv);
1035	if (error)
1036		goto done;
1037#endif
1038	ifv->ifv_proto = ETHERTYPE_VLAN;
1039	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1040	ifv->ifv_mintu = ETHERMIN;
1041	ifv->ifv_pflags = 0;
1042
1043	/*
1044	 * If the parent supports the VLAN_MTU capability,
1045	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1046	 * use it.
1047	 */
1048	if (p->if_capenable & IFCAP_VLAN_MTU) {
1049		/*
1050		 * No need to fudge the MTU since the parent can
1051		 * handle extended frames.
1052		 */
1053		ifv->ifv_mtufudge = 0;
1054	} else {
1055		/*
1056		 * Fudge the MTU by the encapsulation size.  This
1057		 * makes us incompatible with strictly compliant
1058		 * 802.1Q implementations, but allows us to use
1059		 * the feature with other NetBSD implementations,
1060		 * which might still be useful.
1061		 */
1062		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1063	}
1064
1065	ifv->ifv_trunk = trunk;
1066	ifp = ifv->ifv_ifp;
1067	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1068	ifp->if_baudrate = p->if_baudrate;
1069	/*
1070	 * Copy only a selected subset of flags from the parent.
1071	 * Other flags are none of our business.
1072	 */
1073#define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1074	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1075	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1076#undef VLAN_COPY_FLAGS
1077
1078	ifp->if_link_state = p->if_link_state;
1079
1080	vlan_capabilities(ifv);
1081
1082	/*
1083	 * Set up our ``Ethernet address'' to reflect the underlying
1084	 * physical interface's.
1085	 */
1086	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), ETHER_ADDR_LEN);
1087
1088	/*
1089	 * Configure multicast addresses that may already be
1090	 * joined on the vlan device.
1091	 */
1092	(void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1093
1094	/* We are ready for operation now. */
1095	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1096done:
1097	TRUNK_UNLOCK(trunk);
1098	VLAN_UNLOCK();
1099
1100	return (error);
1101}
1102
1103static int
1104vlan_unconfig(struct ifnet *ifp)
1105{
1106	int ret;
1107
1108	VLAN_LOCK();
1109	ret = vlan_unconfig_locked(ifp);
1110	VLAN_UNLOCK();
1111	return (ret);
1112}
1113
1114static int
1115vlan_unconfig_locked(struct ifnet *ifp)
1116{
1117	struct ifvlantrunk *trunk;
1118	struct vlan_mc_entry *mc;
1119	struct ifvlan *ifv;
1120	int error;
1121
1122	VLAN_LOCK_ASSERT();
1123
1124	ifv = ifp->if_softc;
1125	trunk = ifv->ifv_trunk;
1126
1127	if (trunk) {
1128		struct sockaddr_dl sdl;
1129		struct ifnet *p = trunk->parent;
1130
1131		TRUNK_LOCK(trunk);
1132
1133		/*
1134		 * Since the interface is being unconfigured, we need to
1135		 * empty the list of multicast groups that we may have joined
1136		 * while we were alive from the parent's list.
1137		 */
1138		bzero((char *)&sdl, sizeof(sdl));
1139		sdl.sdl_len = sizeof(sdl);
1140		sdl.sdl_family = AF_LINK;
1141		sdl.sdl_index = p->if_index;
1142		sdl.sdl_type = IFT_ETHER;
1143		sdl.sdl_alen = ETHER_ADDR_LEN;
1144
1145		while(SLIST_FIRST(&ifv->vlan_mc_listhead) != NULL) {
1146			mc = SLIST_FIRST(&ifv->vlan_mc_listhead);
1147			bcopy((char *)&mc->mc_addr, LLADDR(&sdl),
1148			    ETHER_ADDR_LEN);
1149			error = if_delmulti(p, (struct sockaddr *)&sdl);
1150			if (error)
1151				return (error);
1152			SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1153			free(mc, M_VLAN);
1154		}
1155
1156		vlan_setflags(ifp, 0); /* clear special flags on parent */
1157#ifdef VLAN_ARRAY
1158		trunk->vlans[ifv->ifv_tag] = NULL;
1159		trunk->refcnt--;
1160#else
1161		vlan_remhash(trunk, ifv);
1162#endif
1163		ifv->ifv_trunk = NULL;
1164
1165		/*
1166		 * Check if we were the last.
1167		 */
1168		if (trunk->refcnt == 0) {
1169			trunk->parent->if_vlantrunk = NULL;
1170			/*
1171			 * XXXGL: If some ithread has already entered
1172			 * vlan_input() and is now blocked on the trunk
1173			 * lock, then it should preempt us right after
1174			 * unlock and finish its work. Then we will acquire
1175			 * lock again in trunk_destroy().
1176			 */
1177			TRUNK_UNLOCK(trunk);
1178			trunk_destroy(trunk);
1179		} else
1180			TRUNK_UNLOCK(trunk);
1181	}
1182
1183	/* Disconnect from parent. */
1184	if (ifv->ifv_pflags)
1185		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1186	ifp->if_mtu = ETHERMTU;
1187	ifp->if_link_state = LINK_STATE_UNKNOWN;
1188	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1189
1190	return (0);
1191}
1192
1193/* Handle a reference counted flag that should be set on the parent as well */
1194static int
1195vlan_setflag(struct ifnet *ifp, int flag, int status,
1196	     int (*func)(struct ifnet *, int))
1197{
1198	struct ifvlan *ifv;
1199	int error;
1200
1201	/* XXX VLAN_LOCK_ASSERT(); */
1202
1203	ifv = ifp->if_softc;
1204	status = status ? (ifp->if_flags & flag) : 0;
1205	/* Now "status" contains the flag value or 0 */
1206
1207	/*
1208	 * See if recorded parent's status is different from what
1209	 * we want it to be.  If it is, flip it.  We record parent's
1210	 * status in ifv_pflags so that we won't clear parent's flag
1211	 * we haven't set.  In fact, we don't clear or set parent's
1212	 * flags directly, but get or release references to them.
1213	 * That's why we can be sure that recorded flags still are
1214	 * in accord with actual parent's flags.
1215	 */
1216	if (status != (ifv->ifv_pflags & flag)) {
1217		error = (*func)(PARENT(ifv), status);
1218		if (error)
1219			return (error);
1220		ifv->ifv_pflags &= ~flag;
1221		ifv->ifv_pflags |= status;
1222	}
1223	return (0);
1224}
1225
1226/*
1227 * Handle IFF_* flags that require certain changes on the parent:
1228 * if "status" is true, update parent's flags respective to our if_flags;
1229 * if "status" is false, forcedly clear the flags set on parent.
1230 */
1231static int
1232vlan_setflags(struct ifnet *ifp, int status)
1233{
1234	int error, i;
1235
1236	for (i = 0; vlan_pflags[i].flag; i++) {
1237		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1238				     status, vlan_pflags[i].func);
1239		if (error)
1240			return (error);
1241	}
1242	return (0);
1243}
1244
1245/* Inform all vlans that their parent has changed link state */
1246static void
1247vlan_link_state(struct ifnet *ifp, int link)
1248{
1249	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1250	struct ifvlan *ifv;
1251	int i;
1252
1253	TRUNK_LOCK(trunk);
1254#ifdef VLAN_ARRAY
1255	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1256		if (trunk->vlans[i] != NULL) {
1257			ifv = trunk->vlans[i];
1258#else
1259	for (i = 0; i < (1 << trunk->hwidth); i++)
1260		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1261#endif
1262			ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1263			if_link_state_change(ifv->ifv_ifp,
1264			    trunk->parent->if_link_state);
1265		}
1266	TRUNK_UNLOCK(trunk);
1267}
1268
1269static void
1270vlan_capabilities(struct ifvlan *ifv)
1271{
1272	struct ifnet *p = PARENT(ifv);
1273	struct ifnet *ifp = ifv->ifv_ifp;
1274
1275	TRUNK_LOCK_ASSERT(TRUNK(ifv));
1276
1277	/*
1278	 * If the parent interface can do checksum offloading
1279	 * on VLANs, then propagate its hardware-assisted
1280	 * checksumming flags. Also assert that checksum
1281	 * offloading requires hardware VLAN tagging.
1282	 */
1283	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1284		ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1285
1286	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1287	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1288		ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1289		ifp->if_hwassist = p->if_hwassist;
1290	} else {
1291		ifp->if_capenable = 0;
1292		ifp->if_hwassist = 0;
1293	}
1294}
1295
1296static void
1297vlan_trunk_capabilities(struct ifnet *ifp)
1298{
1299	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1300	struct ifvlan *ifv;
1301	int i;
1302
1303	TRUNK_LOCK(trunk);
1304#ifdef VLAN_ARRAY
1305	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1306		if (trunk->vlans[i] != NULL) {
1307			ifv = trunk->vlans[i];
1308#else
1309	for (i = 0; i < (1 << trunk->hwidth); i++) {
1310		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1311#endif
1312			vlan_capabilities(ifv);
1313	}
1314	TRUNK_UNLOCK(trunk);
1315}
1316
1317static int
1318vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1319{
1320	struct ifaddr *ifa;
1321	struct ifnet *p;
1322	struct ifreq *ifr;
1323	struct ifvlan *ifv;
1324	struct vlanreq vlr;
1325	int error = 0;
1326
1327	ifr = (struct ifreq *)data;
1328	ifa = (struct ifaddr *)data;
1329	ifv = ifp->if_softc;
1330
1331	switch (cmd) {
1332	case SIOCSIFADDR:
1333		ifp->if_flags |= IFF_UP;
1334
1335		switch (ifa->ifa_addr->sa_family) {
1336#ifdef INET
1337		case AF_INET:
1338			arp_ifinit(ifv->ifv_ifp, ifa);
1339			break;
1340#endif
1341		default:
1342			break;
1343		}
1344		break;
1345
1346	case SIOCGIFADDR:
1347		{
1348			struct sockaddr *sa;
1349
1350			sa = (struct sockaddr *) &ifr->ifr_data;
1351			bcopy(IF_LLADDR(ifp), (caddr_t)sa->sa_data,
1352			    ETHER_ADDR_LEN);
1353		}
1354		break;
1355
1356	case SIOCGIFMEDIA:
1357		VLAN_LOCK();
1358		if (TRUNK(ifv) != NULL) {
1359			error = (*PARENT(ifv)->if_ioctl)(PARENT(ifv),
1360					SIOCGIFMEDIA, data);
1361			VLAN_UNLOCK();
1362			/* Limit the result to the parent's current config. */
1363			if (error == 0) {
1364				struct ifmediareq *ifmr;
1365
1366				ifmr = (struct ifmediareq *)data;
1367				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1368					ifmr->ifm_count = 1;
1369					error = copyout(&ifmr->ifm_current,
1370						ifmr->ifm_ulist,
1371						sizeof(int));
1372				}
1373			}
1374		} else {
1375			VLAN_UNLOCK();
1376			error = EINVAL;
1377		}
1378		break;
1379
1380	case SIOCSIFMEDIA:
1381		error = EINVAL;
1382		break;
1383
1384	case SIOCSIFMTU:
1385		/*
1386		 * Set the interface MTU.
1387		 */
1388		VLAN_LOCK();
1389		if (TRUNK(ifv) != NULL) {
1390			if (ifr->ifr_mtu >
1391			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1392			    ifr->ifr_mtu <
1393			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1394				error = EINVAL;
1395			else
1396				ifp->if_mtu = ifr->ifr_mtu;
1397		} else
1398			error = EINVAL;
1399		VLAN_UNLOCK();
1400		break;
1401
1402	case SIOCSETVLAN:
1403		error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1404		if (error)
1405			break;
1406		if (vlr.vlr_parent[0] == '\0') {
1407			vlan_unconfig(ifp);
1408			break;
1409		}
1410		p = ifunit(vlr.vlr_parent);
1411		if (p == 0) {
1412			error = ENOENT;
1413			break;
1414		}
1415		/*
1416		 * Don't let the caller set up a VLAN tag with
1417		 * anything except VLID bits.
1418		 */
1419		if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1420			error = EINVAL;
1421			break;
1422		}
1423		error = vlan_config(ifv, p, vlr.vlr_tag);
1424		if (error)
1425			break;
1426
1427		/* Update flags on the parent, if necessary. */
1428		vlan_setflags(ifp, 1);
1429		break;
1430
1431	case SIOCGETVLAN:
1432		bzero(&vlr, sizeof(vlr));
1433		VLAN_LOCK();
1434		if (TRUNK(ifv) != NULL) {
1435			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1436			    sizeof(vlr.vlr_parent));
1437			vlr.vlr_tag = ifv->ifv_tag;
1438		}
1439		VLAN_UNLOCK();
1440		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1441		break;
1442
1443	case SIOCSIFFLAGS:
1444		/*
1445		 * We should propagate selected flags to the parent,
1446		 * e.g., promiscuous mode.
1447		 */
1448		if (TRUNK(ifv) != NULL)
1449			error = vlan_setflags(ifp, 1);
1450		break;
1451
1452	case SIOCADDMULTI:
1453	case SIOCDELMULTI:
1454		/*
1455		 * If we don't have a parent, just remember the membership for
1456		 * when we do.
1457		 */
1458		if (TRUNK(ifv) != NULL)
1459			error = vlan_setmulti(ifp);
1460		break;
1461
1462	default:
1463		error = EINVAL;
1464	}
1465
1466	return (error);
1467}
1468