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
30/*
31 * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
32 * Might be extended some day to also handle IEEE 802.1p priority
33 * tagging.  This is sort of sneaky in the implementation, since
34 * we need to pretend to be enough of an Ethernet implementation
35 * to make arp work.  The way we do this is by telling everyone
36 * that we are an Ethernet, and then catch the packets that
37 * ether_output() sends to us via if_transmit(), rewrite them for
38 * use by the real outgoing interface, and ask it to send them.
39 */
40
41#include <sys/cdefs.h>
42__FBSDID("$FreeBSD$");
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#include <sys/sx.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/vnet.h>
69
70#ifdef INET
71#include <netinet/in.h>
72#include <netinet/if_ether.h>
73#endif
74
75#define VLANNAME	"vlan"
76#define	VLAN_DEF_HWIDTH	4
77#define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
78
79#define	UP_AND_RUNNING(ifp) \
80    ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
81
82LIST_HEAD(ifvlanhead, ifvlan);
83
84struct ifvlantrunk {
85	struct	ifnet   *parent;	/* parent interface of this trunk */
86	struct	rwlock	rw;
87#ifdef VLAN_ARRAY
88#define	VLAN_ARRAY_SIZE	(EVL_VLID_MASK + 1)
89	struct	ifvlan	*vlans[VLAN_ARRAY_SIZE]; /* static table */
90#else
91	struct	ifvlanhead *hash;	/* dynamic hash-list table */
92	uint16_t	hmask;
93	uint16_t	hwidth;
94#endif
95	int		refcnt;
96};
97
98struct vlan_mc_entry {
99	struct sockaddr_dl		mc_addr;
100	SLIST_ENTRY(vlan_mc_entry)	mc_entries;
101};
102
103struct	ifvlan {
104	struct	ifvlantrunk *ifv_trunk;
105	struct	ifnet *ifv_ifp;
106	void	*ifv_cookie;
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#ifndef VLAN_ARRAY
119	LIST_ENTRY(ifvlan) ifv_list;
120#endif
121};
122#define	ifv_proto	ifv_mib.ifvm_proto
123#define	ifv_tag		ifv_mib.ifvm_tag
124#define	ifv_encaplen	ifv_mib.ifvm_encaplen
125#define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
126#define	ifv_mintu	ifv_mib.ifvm_mintu
127
128/* Special flags we should propagate to parent. */
129static struct {
130	int flag;
131	int (*func)(struct ifnet *, int);
132} vlan_pflags[] = {
133	{IFF_PROMISC, ifpromisc},
134	{IFF_ALLMULTI, if_allmulti},
135	{0, NULL}
136};
137
138SYSCTL_DECL(_net_link);
139static SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0,
140    "IEEE 802.1Q VLAN");
141static SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0,
142    "for consistency");
143
144static int soft_pad = 0;
145SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW, &soft_pad, 0,
146	   "pad short frames before tagging");
147
148static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface");
149
150static eventhandler_tag ifdetach_tag;
151static eventhandler_tag iflladdr_tag;
152
153/*
154 * We have a global mutex, that is used to serialize configuration
155 * changes and isn't used in normal packet delivery.
156 *
157 * We also have a per-trunk rwlock, that is locked shared on packet
158 * processing and exclusive when configuration is changed.
159 *
160 * The VLAN_ARRAY substitutes the dynamic hash with a static array
161 * with 4096 entries. In theory this can give a boost in processing,
162 * however on practice it does not. Probably this is because array
163 * is too big to fit into CPU cache.
164 */
165static struct sx ifv_lock;
166#define	VLAN_LOCK_INIT()	sx_init(&ifv_lock, "vlan_global")
167#define	VLAN_LOCK_DESTROY()	sx_destroy(&ifv_lock)
168#define	VLAN_LOCK_ASSERT()	sx_assert(&ifv_lock, SA_LOCKED)
169#define	VLAN_LOCK()		sx_xlock(&ifv_lock)
170#define	VLAN_UNLOCK()		sx_xunlock(&ifv_lock)
171#define	TRUNK_LOCK_INIT(trunk)	rw_init(&(trunk)->rw, VLANNAME)
172#define	TRUNK_LOCK_DESTROY(trunk) rw_destroy(&(trunk)->rw)
173#define	TRUNK_LOCK(trunk)	rw_wlock(&(trunk)->rw)
174#define	TRUNK_UNLOCK(trunk)	rw_wunlock(&(trunk)->rw)
175#define	TRUNK_LOCK_ASSERT(trunk) rw_assert(&(trunk)->rw, RA_WLOCKED)
176#define	TRUNK_RLOCK(trunk)	rw_rlock(&(trunk)->rw)
177#define	TRUNK_RUNLOCK(trunk)	rw_runlock(&(trunk)->rw)
178#define	TRUNK_LOCK_RASSERT(trunk) rw_assert(&(trunk)->rw, RA_RLOCKED)
179
180#ifndef VLAN_ARRAY
181static	void vlan_inithash(struct ifvlantrunk *trunk);
182static	void vlan_freehash(struct ifvlantrunk *trunk);
183static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
184static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
185static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
186static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
187	uint16_t tag);
188#endif
189static	void trunk_destroy(struct ifvlantrunk *trunk);
190
191static	void vlan_init(void *foo);
192static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
193static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
194static	void vlan_qflush(struct ifnet *ifp);
195static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
196    int (*func)(struct ifnet *, int));
197static	int vlan_setflags(struct ifnet *ifp, int status);
198static	int vlan_setmulti(struct ifnet *ifp);
199static	int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
200static	void vlan_unconfig(struct ifnet *ifp);
201static	void vlan_unconfig_locked(struct ifnet *ifp, int departing);
202static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
203static	void vlan_link_state(struct ifnet *ifp);
204static	void vlan_capabilities(struct ifvlan *ifv);
205static	void vlan_trunk_capabilities(struct ifnet *ifp);
206
207static	struct ifnet *vlan_clone_match_ethertag(struct if_clone *,
208    const char *, int *);
209static	int vlan_clone_match(struct if_clone *, const char *);
210static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
211static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
212
213static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
214static  void vlan_iflladdr(void *arg, struct ifnet *ifp);
215
216static	struct if_clone vlan_cloner = IFC_CLONE_INITIALIZER(VLANNAME, NULL,
217    IF_MAXUNIT, NULL, vlan_clone_match, vlan_clone_create, vlan_clone_destroy);
218
219#ifdef VIMAGE
220static VNET_DEFINE(struct if_clone, vlan_cloner);
221#define	V_vlan_cloner	VNET(vlan_cloner)
222#endif
223
224#ifndef VLAN_ARRAY
225#define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
226
227static void
228vlan_inithash(struct ifvlantrunk *trunk)
229{
230	int i, n;
231
232	/*
233	 * The trunk must not be locked here since we call malloc(M_WAITOK).
234	 * It is OK in case this function is called before the trunk struct
235	 * gets hooked up and becomes visible from other threads.
236	 */
237
238	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
239	    ("%s: hash already initialized", __func__));
240
241	trunk->hwidth = VLAN_DEF_HWIDTH;
242	n = 1 << trunk->hwidth;
243	trunk->hmask = n - 1;
244	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
245	for (i = 0; i < n; i++)
246		LIST_INIT(&trunk->hash[i]);
247}
248
249static void
250vlan_freehash(struct ifvlantrunk *trunk)
251{
252#ifdef INVARIANTS
253	int i;
254
255	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
256	for (i = 0; i < (1 << trunk->hwidth); i++)
257		KASSERT(LIST_EMPTY(&trunk->hash[i]),
258		    ("%s: hash table not empty", __func__));
259#endif
260	free(trunk->hash, M_VLAN);
261	trunk->hash = NULL;
262	trunk->hwidth = trunk->hmask = 0;
263}
264
265static int
266vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
267{
268	int i, b;
269	struct ifvlan *ifv2;
270
271	TRUNK_LOCK_ASSERT(trunk);
272	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
273
274	b = 1 << trunk->hwidth;
275	i = HASH(ifv->ifv_tag, trunk->hmask);
276	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
277		if (ifv->ifv_tag == ifv2->ifv_tag)
278			return (EEXIST);
279
280	/*
281	 * Grow the hash when the number of vlans exceeds half of the number of
282	 * hash buckets squared. This will make the average linked-list length
283	 * buckets/2.
284	 */
285	if (trunk->refcnt > (b * b) / 2) {
286		vlan_growhash(trunk, 1);
287		i = HASH(ifv->ifv_tag, trunk->hmask);
288	}
289	LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
290	trunk->refcnt++;
291
292	return (0);
293}
294
295static int
296vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
297{
298	int i, b;
299	struct ifvlan *ifv2;
300
301	TRUNK_LOCK_ASSERT(trunk);
302	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
303
304	b = 1 << trunk->hwidth;
305	i = HASH(ifv->ifv_tag, trunk->hmask);
306	LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
307		if (ifv2 == ifv) {
308			trunk->refcnt--;
309			LIST_REMOVE(ifv2, ifv_list);
310			if (trunk->refcnt < (b * b) / 2)
311				vlan_growhash(trunk, -1);
312			return (0);
313		}
314
315	panic("%s: vlan not found\n", __func__);
316	return (ENOENT); /*NOTREACHED*/
317}
318
319/*
320 * Grow the hash larger or smaller if memory permits.
321 */
322static void
323vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
324{
325	struct ifvlan *ifv;
326	struct ifvlanhead *hash2;
327	int hwidth2, i, j, n, n2;
328
329	TRUNK_LOCK_ASSERT(trunk);
330	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
331
332	if (howmuch == 0) {
333		/* Harmless yet obvious coding error */
334		printf("%s: howmuch is 0\n", __func__);
335		return;
336	}
337
338	hwidth2 = trunk->hwidth + howmuch;
339	n = 1 << trunk->hwidth;
340	n2 = 1 << hwidth2;
341	/* Do not shrink the table below the default */
342	if (hwidth2 < VLAN_DEF_HWIDTH)
343		return;
344
345	/* M_NOWAIT because we're called with trunk mutex held */
346	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
347	if (hash2 == NULL) {
348		printf("%s: out of memory -- hash size not changed\n",
349		    __func__);
350		return;		/* We can live with the old hash table */
351	}
352	for (j = 0; j < n2; j++)
353		LIST_INIT(&hash2[j]);
354	for (i = 0; i < n; i++)
355		while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
356			LIST_REMOVE(ifv, ifv_list);
357			j = HASH(ifv->ifv_tag, n2 - 1);
358			LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
359		}
360	free(trunk->hash, M_VLAN);
361	trunk->hash = hash2;
362	trunk->hwidth = hwidth2;
363	trunk->hmask = n2 - 1;
364
365	if (bootverbose)
366		if_printf(trunk->parent,
367		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
368}
369
370static __inline struct ifvlan *
371vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
372{
373	struct ifvlan *ifv;
374
375	TRUNK_LOCK_RASSERT(trunk);
376
377	LIST_FOREACH(ifv, &trunk->hash[HASH(tag, trunk->hmask)], ifv_list)
378		if (ifv->ifv_tag == tag)
379			return (ifv);
380	return (NULL);
381}
382
383#if 0
384/* Debugging code to view the hashtables. */
385static void
386vlan_dumphash(struct ifvlantrunk *trunk)
387{
388	int i;
389	struct ifvlan *ifv;
390
391	for (i = 0; i < (1 << trunk->hwidth); i++) {
392		printf("%d: ", i);
393		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
394			printf("%s ", ifv->ifv_ifp->if_xname);
395		printf("\n");
396	}
397}
398#endif /* 0 */
399#else
400
401static __inline struct ifvlan *
402vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
403{
404
405	return trunk->vlans[tag];
406}
407
408static __inline int
409vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
410{
411
412	if (trunk->vlans[ifv->ifv_tag] != NULL)
413		return EEXIST;
414	trunk->vlans[ifv->ifv_tag] = ifv;
415	trunk->refcnt++;
416
417	return (0);
418}
419
420static __inline int
421vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
422{
423
424	trunk->vlans[ifv->ifv_tag] = NULL;
425	trunk->refcnt--;
426
427	return (0);
428}
429
430static __inline void
431vlan_freehash(struct ifvlantrunk *trunk)
432{
433}
434
435static __inline void
436vlan_inithash(struct ifvlantrunk *trunk)
437{
438}
439
440#endif /* !VLAN_ARRAY */
441
442static void
443trunk_destroy(struct ifvlantrunk *trunk)
444{
445	VLAN_LOCK_ASSERT();
446
447	TRUNK_LOCK(trunk);
448	vlan_freehash(trunk);
449	trunk->parent->if_vlantrunk = NULL;
450	TRUNK_UNLOCK(trunk);
451	TRUNK_LOCK_DESTROY(trunk);
452	free(trunk, M_VLAN);
453}
454
455/*
456 * Program our multicast filter. What we're actually doing is
457 * programming the multicast filter of the parent. This has the
458 * side effect of causing the parent interface to receive multicast
459 * traffic that it doesn't really want, which ends up being discarded
460 * later by the upper protocol layers. Unfortunately, there's no way
461 * to avoid this: there really is only one physical interface.
462 *
463 * XXX: There is a possible race here if more than one thread is
464 *      modifying the multicast state of the vlan interface at the same time.
465 */
466static int
467vlan_setmulti(struct ifnet *ifp)
468{
469	struct ifnet		*ifp_p;
470	struct ifmultiaddr	*ifma, *rifma = NULL;
471	struct ifvlan		*sc;
472	struct vlan_mc_entry	*mc;
473	int			error;
474
475	/*VLAN_LOCK_ASSERT();*/
476
477	/* Find the parent. */
478	sc = ifp->if_softc;
479	ifp_p = PARENT(sc);
480
481	CURVNET_SET_QUIET(ifp_p->if_vnet);
482
483	/* First, remove any existing filter entries. */
484	while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
485		error = if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
486		if (error)
487			return (error);
488		SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
489		free(mc, M_VLAN);
490	}
491
492	/* Now program new ones. */
493	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
494		if (ifma->ifma_addr->sa_family != AF_LINK)
495			continue;
496		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
497		if (mc == NULL)
498			return (ENOMEM);
499		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
500		mc->mc_addr.sdl_index = ifp_p->if_index;
501		SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
502		error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
503		    &rifma);
504		if (error)
505			return (error);
506	}
507
508	CURVNET_RESTORE();
509	return (0);
510}
511
512/*
513 * A handler for parent interface link layer address changes.
514 * If the parent interface link layer address is changed we
515 * should also change it on all children vlans.
516 */
517static void
518vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
519{
520	struct ifvlan *ifv;
521#ifndef VLAN_ARRAY
522	struct ifvlan *next;
523#endif
524	int i;
525
526	/*
527	 * Check if it's a trunk interface first of all
528	 * to avoid needless locking.
529	 */
530	if (ifp->if_vlantrunk == NULL)
531		return;
532
533	VLAN_LOCK();
534	/*
535	 * OK, it's a trunk.  Loop over and change all vlan's lladdrs on it.
536	 */
537#ifdef VLAN_ARRAY
538	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
539		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
540#else /* VLAN_ARRAY */
541	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
542		LIST_FOREACH_SAFE(ifv, &ifp->if_vlantrunk->hash[i], ifv_list, next) {
543#endif /* VLAN_ARRAY */
544			VLAN_UNLOCK();
545			if_setlladdr(ifv->ifv_ifp, IF_LLADDR(ifp),
546			    ifp->if_addrlen);
547			VLAN_LOCK();
548		}
549	VLAN_UNLOCK();
550
551}
552
553/*
554 * A handler for network interface departure events.
555 * Track departure of trunks here so that we don't access invalid
556 * pointers or whatever if a trunk is ripped from under us, e.g.,
557 * by ejecting its hot-plug card.  However, if an ifnet is simply
558 * being renamed, then there's no need to tear down the state.
559 */
560static void
561vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
562{
563	struct ifvlan *ifv;
564	int i;
565
566	/*
567	 * Check if it's a trunk interface first of all
568	 * to avoid needless locking.
569	 */
570	if (ifp->if_vlantrunk == NULL)
571		return;
572
573	/* If the ifnet is just being renamed, don't do anything. */
574	if (ifp->if_flags & IFF_RENAMING)
575		return;
576
577	VLAN_LOCK();
578	/*
579	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
580	 * Check trunk pointer after each vlan_unconfig() as it will
581	 * free it and set to NULL after the last vlan was detached.
582	 */
583#ifdef VLAN_ARRAY
584	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
585		if ((ifv = ifp->if_vlantrunk->vlans[i])) {
586			vlan_unconfig_locked(ifv->ifv_ifp, 1);
587			if (ifp->if_vlantrunk == NULL)
588				break;
589		}
590#else /* VLAN_ARRAY */
591restart:
592	for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
593		if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
594			vlan_unconfig_locked(ifv->ifv_ifp, 1);
595			if (ifp->if_vlantrunk)
596				goto restart;	/* trunk->hwidth can change */
597			else
598				break;
599		}
600#endif /* VLAN_ARRAY */
601	/* Trunk should have been destroyed in vlan_unconfig(). */
602	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
603	VLAN_UNLOCK();
604}
605
606/*
607 * Return the trunk device for a virtual interface.
608 */
609static struct ifnet  *
610vlan_trunkdev(struct ifnet *ifp)
611{
612	struct ifvlan *ifv;
613
614	if (ifp->if_type != IFT_L2VLAN)
615		return (NULL);
616	ifv = ifp->if_softc;
617	ifp = NULL;
618	VLAN_LOCK();
619	if (ifv->ifv_trunk)
620		ifp = PARENT(ifv);
621	VLAN_UNLOCK();
622	return (ifp);
623}
624
625/*
626 * Return the 16bit vlan tag for this interface.
627 */
628static int
629vlan_tag(struct ifnet *ifp, uint16_t *tagp)
630{
631	struct ifvlan *ifv;
632
633	if (ifp->if_type != IFT_L2VLAN)
634		return (EINVAL);
635	ifv = ifp->if_softc;
636	*tagp = ifv->ifv_tag;
637	return (0);
638}
639
640/*
641 * Return a driver specific cookie for this interface.  Synchronization
642 * with setcookie must be provided by the driver.
643 */
644static void *
645vlan_cookie(struct ifnet *ifp)
646{
647	struct ifvlan *ifv;
648
649	if (ifp->if_type != IFT_L2VLAN)
650		return (NULL);
651	ifv = ifp->if_softc;
652	return (ifv->ifv_cookie);
653}
654
655/*
656 * Store a cookie in our softc that drivers can use to store driver
657 * private per-instance data in.
658 */
659static int
660vlan_setcookie(struct ifnet *ifp, void *cookie)
661{
662	struct ifvlan *ifv;
663
664	if (ifp->if_type != IFT_L2VLAN)
665		return (EINVAL);
666	ifv = ifp->if_softc;
667	ifv->ifv_cookie = cookie;
668	return (0);
669}
670
671/*
672 * Return the vlan device present at the specific tag.
673 */
674static struct ifnet *
675vlan_devat(struct ifnet *ifp, uint16_t tag)
676{
677	struct ifvlantrunk *trunk;
678	struct ifvlan *ifv;
679
680	trunk = ifp->if_vlantrunk;
681	if (trunk == NULL)
682		return (NULL);
683	ifp = NULL;
684	TRUNK_RLOCK(trunk);
685	ifv = vlan_gethash(trunk, tag);
686	if (ifv)
687		ifp = ifv->ifv_ifp;
688	TRUNK_RUNLOCK(trunk);
689	return (ifp);
690}
691
692/*
693 * VLAN support can be loaded as a module.  The only place in the
694 * system that's intimately aware of this is ether_input.  We hook
695 * into this code through vlan_input_p which is defined there and
696 * set here.  Noone else in the system should be aware of this so
697 * we use an explicit reference here.
698 */
699extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
700
701/* For if_link_state_change() eyes only... */
702extern	void (*vlan_link_state_p)(struct ifnet *);
703
704static int
705vlan_modevent(module_t mod, int type, void *data)
706{
707
708	switch (type) {
709	case MOD_LOAD:
710		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
711		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
712		if (ifdetach_tag == NULL)
713			return (ENOMEM);
714		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
715		    vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
716		if (iflladdr_tag == NULL)
717			return (ENOMEM);
718		VLAN_LOCK_INIT();
719		vlan_input_p = vlan_input;
720		vlan_link_state_p = vlan_link_state;
721		vlan_trunk_cap_p = vlan_trunk_capabilities;
722		vlan_trunkdev_p = vlan_trunkdev;
723		vlan_cookie_p = vlan_cookie;
724		vlan_setcookie_p = vlan_setcookie;
725		vlan_tag_p = vlan_tag;
726		vlan_devat_p = vlan_devat;
727#ifndef VIMAGE
728		if_clone_attach(&vlan_cloner);
729#endif
730		if (bootverbose)
731			printf("vlan: initialized, using "
732#ifdef VLAN_ARRAY
733			       "full-size arrays"
734#else
735			       "hash tables with chaining"
736#endif
737
738			       "\n");
739		break;
740	case MOD_UNLOAD:
741#ifndef VIMAGE
742		if_clone_detach(&vlan_cloner);
743#endif
744		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
745		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
746		vlan_input_p = NULL;
747		vlan_link_state_p = NULL;
748		vlan_trunk_cap_p = NULL;
749		vlan_trunkdev_p = NULL;
750		vlan_tag_p = NULL;
751		vlan_cookie_p = NULL;
752		vlan_setcookie_p = NULL;
753		vlan_devat_p = NULL;
754		VLAN_LOCK_DESTROY();
755		if (bootverbose)
756			printf("vlan: unloaded\n");
757		break;
758	default:
759		return (EOPNOTSUPP);
760	}
761	return (0);
762}
763
764static moduledata_t vlan_mod = {
765	"if_vlan",
766	vlan_modevent,
767	0
768};
769
770DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
771MODULE_VERSION(if_vlan, 3);
772
773#ifdef VIMAGE
774static void
775vnet_vlan_init(const void *unused __unused)
776{
777
778	V_vlan_cloner = vlan_cloner;
779	if_clone_attach(&V_vlan_cloner);
780}
781VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
782    vnet_vlan_init, NULL);
783
784static void
785vnet_vlan_uninit(const void *unused __unused)
786{
787
788	if_clone_detach(&V_vlan_cloner);
789}
790VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
791    vnet_vlan_uninit, NULL);
792#endif
793
794static struct ifnet *
795vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
796{
797	const char *cp;
798	struct ifnet *ifp;
799	int t;
800
801	/* Check for <etherif>.<vlan> style interface names. */
802	IFNET_RLOCK_NOSLEEP();
803	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
804		/*
805		 * We can handle non-ethernet hardware types as long as
806		 * they handle the tagging and headers themselves.
807		 */
808		if (ifp->if_type != IFT_ETHER &&
809		    (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
810			continue;
811		if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
812			continue;
813		cp = name + strlen(ifp->if_xname);
814		if (*cp++ != '.')
815			continue;
816		if (*cp == '\0')
817			continue;
818		t = 0;
819		for(; *cp >= '0' && *cp <= '9'; cp++)
820			t = (t * 10) + (*cp - '0');
821		if (*cp != '\0')
822			continue;
823		if (tag != NULL)
824			*tag = t;
825		break;
826	}
827	IFNET_RUNLOCK_NOSLEEP();
828
829	return (ifp);
830}
831
832static int
833vlan_clone_match(struct if_clone *ifc, const char *name)
834{
835	const char *cp;
836
837	if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
838		return (1);
839
840	if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
841		return (0);
842	for (cp = name + 4; *cp != '\0'; cp++) {
843		if (*cp < '0' || *cp > '9')
844			return (0);
845	}
846
847	return (1);
848}
849
850static int
851vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
852{
853	char *dp;
854	int wildcard;
855	int unit;
856	int error;
857	int tag;
858	int ethertag;
859	struct ifvlan *ifv;
860	struct ifnet *ifp;
861	struct ifnet *p;
862	struct ifaddr *ifa;
863	struct sockaddr_dl *sdl;
864	struct vlanreq vlr;
865	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
866
867	/*
868	 * There are 3 (ugh) ways to specify the cloned device:
869	 * o pass a parameter block with the clone request.
870	 * o specify parameters in the text of the clone device name
871	 * o specify no parameters and get an unattached device that
872	 *   must be configured separately.
873	 * The first technique is preferred; the latter two are
874	 * supported for backwards compatibilty.
875	 */
876	if (params) {
877		error = copyin(params, &vlr, sizeof(vlr));
878		if (error)
879			return error;
880		p = ifunit(vlr.vlr_parent);
881		if (p == NULL)
882			return ENXIO;
883		/*
884		 * Don't let the caller set up a VLAN tag with
885		 * anything except VLID bits.
886		 */
887		if (vlr.vlr_tag & ~EVL_VLID_MASK)
888			return (EINVAL);
889		error = ifc_name2unit(name, &unit);
890		if (error != 0)
891			return (error);
892
893		ethertag = 1;
894		tag = vlr.vlr_tag;
895		wildcard = (unit < 0);
896	} else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
897		ethertag = 1;
898		unit = -1;
899		wildcard = 0;
900
901		/*
902		 * Don't let the caller set up a VLAN tag with
903		 * anything except VLID bits.
904		 */
905		if (tag & ~EVL_VLID_MASK)
906			return (EINVAL);
907	} else {
908		ethertag = 0;
909
910		error = ifc_name2unit(name, &unit);
911		if (error != 0)
912			return (error);
913
914		wildcard = (unit < 0);
915	}
916
917	error = ifc_alloc_unit(ifc, &unit);
918	if (error != 0)
919		return (error);
920
921	/* In the wildcard case, we need to update the name. */
922	if (wildcard) {
923		for (dp = name; *dp != '\0'; dp++);
924		if (snprintf(dp, len - (dp-name), "%d", unit) >
925		    len - (dp-name) - 1) {
926			panic("%s: interface name too long", __func__);
927		}
928	}
929
930	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
931	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
932	if (ifp == NULL) {
933		ifc_free_unit(ifc, unit);
934		free(ifv, M_VLAN);
935		return (ENOSPC);
936	}
937	SLIST_INIT(&ifv->vlan_mc_listhead);
938
939	ifp->if_softc = ifv;
940	/*
941	 * Set the name manually rather than using if_initname because
942	 * we don't conform to the default naming convention for interfaces.
943	 */
944	strlcpy(ifp->if_xname, name, IFNAMSIZ);
945	ifp->if_dname = ifc->ifc_name;
946	ifp->if_dunit = unit;
947	/* NB: flags are not set here */
948	ifp->if_linkmib = &ifv->ifv_mib;
949	ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
950	/* NB: mtu is not set here */
951
952	ifp->if_init = vlan_init;
953	ifp->if_transmit = vlan_transmit;
954	ifp->if_qflush = vlan_qflush;
955	ifp->if_ioctl = vlan_ioctl;
956	ifp->if_flags = VLAN_IFFLAGS;
957	ether_ifattach(ifp, eaddr);
958	/* Now undo some of the damage... */
959	ifp->if_baudrate = 0;
960	ifp->if_type = IFT_L2VLAN;
961	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
962	ifa = ifp->if_addr;
963	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
964	sdl->sdl_type = IFT_L2VLAN;
965
966	if (ethertag) {
967		error = vlan_config(ifv, p, tag);
968		if (error != 0) {
969			/*
970			 * Since we've partially failed, we need to back
971			 * out all the way, otherwise userland could get
972			 * confused.  Thus, we destroy the interface.
973			 */
974			ether_ifdetach(ifp);
975			vlan_unconfig(ifp);
976			if_free_type(ifp, IFT_ETHER);
977			ifc_free_unit(ifc, unit);
978			free(ifv, M_VLAN);
979
980			return (error);
981		}
982
983		/* Update flags on the parent, if necessary. */
984		vlan_setflags(ifp, 1);
985	}
986
987	return (0);
988}
989
990static int
991vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
992{
993	struct ifvlan *ifv = ifp->if_softc;
994	int unit = ifp->if_dunit;
995
996	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
997	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
998	if_free_type(ifp, IFT_ETHER);
999	free(ifv, M_VLAN);
1000	ifc_free_unit(ifc, unit);
1001
1002	return (0);
1003}
1004
1005/*
1006 * The ifp->if_init entry point for vlan(4) is a no-op.
1007 */
1008static void
1009vlan_init(void *foo __unused)
1010{
1011}
1012
1013/*
1014 * The if_transmit method for vlan(4) interface.
1015 */
1016static int
1017vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1018{
1019	struct ifvlan *ifv;
1020	struct ifnet *p;
1021	int error, len, mcast;
1022
1023	ifv = ifp->if_softc;
1024	p = PARENT(ifv);
1025	len = m->m_pkthdr.len;
1026	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1027
1028	BPF_MTAP(ifp, m);
1029
1030	/*
1031	 * Do not run parent's if_transmit() if the parent is not up,
1032	 * or parent's driver will cause a system crash.
1033	 */
1034	if (!UP_AND_RUNNING(p)) {
1035		m_freem(m);
1036		ifp->if_oerrors++;
1037		return (ENETDOWN);
1038	}
1039
1040	/*
1041	 * Pad the frame to the minimum size allowed if told to.
1042	 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
1043	 * paragraph C.4.4.3.b.  It can help to work around buggy
1044	 * bridges that violate paragraph C.4.4.3.a from the same
1045	 * document, i.e., fail to pad short frames after untagging.
1046	 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
1047	 * untagging it will produce a 62-byte frame, which is a runt
1048	 * and requires padding.  There are VLAN-enabled network
1049	 * devices that just discard such runts instead or mishandle
1050	 * them somehow.
1051	 */
1052	if (soft_pad && p->if_type == IFT_ETHER) {
1053		static char pad[8];	/* just zeros */
1054		int n;
1055
1056		for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
1057		     n > 0; n -= sizeof(pad))
1058			if (!m_append(m, min(n, sizeof(pad)), pad))
1059				break;
1060
1061		if (n > 0) {
1062			if_printf(ifp, "cannot pad short frame\n");
1063			ifp->if_oerrors++;
1064			m_freem(m);
1065			return (0);
1066		}
1067	}
1068
1069	/*
1070	 * If underlying interface can do VLAN tag insertion itself,
1071	 * just pass the packet along. However, we need some way to
1072	 * tell the interface where the packet came from so that it
1073	 * knows how to find the VLAN tag to use, so we attach a
1074	 * packet tag that holds it.
1075	 */
1076	if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1077		m->m_pkthdr.ether_vtag = ifv->ifv_tag;
1078		m->m_flags |= M_VLANTAG;
1079	} else {
1080		m = ether_vlanencap(m, ifv->ifv_tag);
1081		if (m == NULL) {
1082			if_printf(ifp, "unable to prepend VLAN header\n");
1083			ifp->if_oerrors++;
1084			return (0);
1085		}
1086	}
1087
1088	/*
1089	 * Send it, precisely as ether_output() would have.
1090	 */
1091	error = (p->if_transmit)(p, m);
1092	if (!error) {
1093		ifp->if_opackets++;
1094		ifp->if_omcasts += mcast;
1095		ifp->if_obytes += len;
1096	} else
1097		ifp->if_oerrors++;
1098	return (error);
1099}
1100
1101/*
1102 * The ifp->if_qflush entry point for vlan(4) is a no-op.
1103 */
1104static void
1105vlan_qflush(struct ifnet *ifp __unused)
1106{
1107}
1108
1109static void
1110vlan_input(struct ifnet *ifp, struct mbuf *m)
1111{
1112	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1113	struct ifvlan *ifv;
1114	uint16_t tag;
1115
1116	KASSERT(trunk != NULL, ("%s: no trunk", __func__));
1117
1118	if (m->m_flags & M_VLANTAG) {
1119		/*
1120		 * Packet is tagged, but m contains a normal
1121		 * Ethernet frame; the tag is stored out-of-band.
1122		 */
1123		tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
1124		m->m_flags &= ~M_VLANTAG;
1125	} else {
1126		struct ether_vlan_header *evl;
1127
1128		/*
1129		 * Packet is tagged in-band as specified by 802.1q.
1130		 */
1131		switch (ifp->if_type) {
1132		case IFT_ETHER:
1133			if (m->m_len < sizeof(*evl) &&
1134			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
1135				if_printf(ifp, "cannot pullup VLAN header\n");
1136				return;
1137			}
1138			evl = mtod(m, struct ether_vlan_header *);
1139			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1140
1141			/*
1142			 * Remove the 802.1q header by copying the Ethernet
1143			 * addresses over it and adjusting the beginning of
1144			 * the data in the mbuf.  The encapsulated Ethernet
1145			 * type field is already in place.
1146			 */
1147			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1148			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
1149			m_adj(m, ETHER_VLAN_ENCAP_LEN);
1150			break;
1151
1152		default:
1153#ifdef INVARIANTS
1154			panic("%s: %s has unsupported if_type %u",
1155			      __func__, ifp->if_xname, ifp->if_type);
1156#endif
1157			m_freem(m);
1158			ifp->if_noproto++;
1159			return;
1160		}
1161	}
1162
1163	TRUNK_RLOCK(trunk);
1164	ifv = vlan_gethash(trunk, tag);
1165	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1166		TRUNK_RUNLOCK(trunk);
1167		m_freem(m);
1168		ifp->if_noproto++;
1169		return;
1170	}
1171	TRUNK_RUNLOCK(trunk);
1172
1173	m->m_pkthdr.rcvif = ifv->ifv_ifp;
1174	ifv->ifv_ifp->if_ipackets++;
1175
1176	/* Pass it back through the parent's input routine. */
1177	(*ifp->if_input)(ifv->ifv_ifp, m);
1178}
1179
1180static int
1181vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
1182{
1183	struct ifvlantrunk *trunk;
1184	struct ifnet *ifp;
1185	int error = 0;
1186
1187	/* VID numbers 0x0 and 0xFFF are reserved */
1188	if (tag == 0 || tag == 0xFFF)
1189		return (EINVAL);
1190	if (p->if_type != IFT_ETHER &&
1191	    (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1192		return (EPROTONOSUPPORT);
1193	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1194		return (EPROTONOSUPPORT);
1195	if (ifv->ifv_trunk)
1196		return (EBUSY);
1197
1198	if (p->if_vlantrunk == NULL) {
1199		trunk = malloc(sizeof(struct ifvlantrunk),
1200		    M_VLAN, M_WAITOK | M_ZERO);
1201		vlan_inithash(trunk);
1202		VLAN_LOCK();
1203		if (p->if_vlantrunk != NULL) {
1204			/* A race that that is very unlikely to be hit. */
1205			vlan_freehash(trunk);
1206			free(trunk, M_VLAN);
1207			goto exists;
1208		}
1209		TRUNK_LOCK_INIT(trunk);
1210		TRUNK_LOCK(trunk);
1211		p->if_vlantrunk = trunk;
1212		trunk->parent = p;
1213	} else {
1214		VLAN_LOCK();
1215exists:
1216		trunk = p->if_vlantrunk;
1217		TRUNK_LOCK(trunk);
1218	}
1219
1220	ifv->ifv_tag = tag;	/* must set this before vlan_inshash() */
1221	error = vlan_inshash(trunk, ifv);
1222	if (error)
1223		goto done;
1224	ifv->ifv_proto = ETHERTYPE_VLAN;
1225	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1226	ifv->ifv_mintu = ETHERMIN;
1227	ifv->ifv_pflags = 0;
1228
1229	/*
1230	 * If the parent supports the VLAN_MTU capability,
1231	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1232	 * use it.
1233	 */
1234	if (p->if_capenable & IFCAP_VLAN_MTU) {
1235		/*
1236		 * No need to fudge the MTU since the parent can
1237		 * handle extended frames.
1238		 */
1239		ifv->ifv_mtufudge = 0;
1240	} else {
1241		/*
1242		 * Fudge the MTU by the encapsulation size.  This
1243		 * makes us incompatible with strictly compliant
1244		 * 802.1Q implementations, but allows us to use
1245		 * the feature with other NetBSD implementations,
1246		 * which might still be useful.
1247		 */
1248		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1249	}
1250
1251	ifv->ifv_trunk = trunk;
1252	ifp = ifv->ifv_ifp;
1253	/*
1254	 * Initialize fields from our parent.  This duplicates some
1255	 * work with ether_ifattach() but allows for non-ethernet
1256	 * interfaces to also work.
1257	 */
1258	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1259	ifp->if_baudrate = p->if_baudrate;
1260	ifp->if_output = p->if_output;
1261	ifp->if_input = p->if_input;
1262	ifp->if_resolvemulti = p->if_resolvemulti;
1263	ifp->if_addrlen = p->if_addrlen;
1264	ifp->if_broadcastaddr = p->if_broadcastaddr;
1265
1266	/*
1267	 * Copy only a selected subset of flags from the parent.
1268	 * Other flags are none of our business.
1269	 */
1270#define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1271	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1272	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1273#undef VLAN_COPY_FLAGS
1274
1275	ifp->if_link_state = p->if_link_state;
1276
1277	vlan_capabilities(ifv);
1278
1279	/*
1280	 * Set up our interface address to reflect the underlying
1281	 * physical interface's.
1282	 */
1283	bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1284	((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1285	    p->if_addrlen;
1286
1287	/*
1288	 * Configure multicast addresses that may already be
1289	 * joined on the vlan device.
1290	 */
1291	(void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1292
1293	/* We are ready for operation now. */
1294	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1295done:
1296	TRUNK_UNLOCK(trunk);
1297	if (error == 0)
1298		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_tag);
1299	VLAN_UNLOCK();
1300
1301	return (error);
1302}
1303
1304static void
1305vlan_unconfig(struct ifnet *ifp)
1306{
1307
1308	VLAN_LOCK();
1309	vlan_unconfig_locked(ifp, 0);
1310	VLAN_UNLOCK();
1311}
1312
1313static void
1314vlan_unconfig_locked(struct ifnet *ifp, int departing)
1315{
1316	struct ifvlantrunk *trunk;
1317	struct vlan_mc_entry *mc;
1318	struct ifvlan *ifv;
1319	struct ifnet  *parent;
1320	int error;
1321
1322	VLAN_LOCK_ASSERT();
1323
1324	ifv = ifp->if_softc;
1325	trunk = ifv->ifv_trunk;
1326	parent = NULL;
1327
1328	if (trunk != NULL) {
1329
1330		TRUNK_LOCK(trunk);
1331		parent = trunk->parent;
1332
1333		/*
1334		 * Since the interface is being unconfigured, we need to
1335		 * empty the list of multicast groups that we may have joined
1336		 * while we were alive from the parent's list.
1337		 */
1338		while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1339			/*
1340			 * If the parent interface is being detached,
1341			 * all its multicast addresses have already
1342			 * been removed.  Warn about errors if
1343			 * if_delmulti() does fail, but don't abort as
1344			 * all callers expect vlan destruction to
1345			 * succeed.
1346			 */
1347			if (!departing) {
1348				error = if_delmulti(parent,
1349				    (struct sockaddr *)&mc->mc_addr);
1350				if (error)
1351					if_printf(ifp,
1352		    "Failed to delete multicast address from parent: %d\n",
1353					    error);
1354			}
1355			SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1356			free(mc, M_VLAN);
1357		}
1358
1359		vlan_setflags(ifp, 0); /* clear special flags on parent */
1360		vlan_remhash(trunk, ifv);
1361		ifv->ifv_trunk = NULL;
1362
1363		/*
1364		 * Check if we were the last.
1365		 */
1366		if (trunk->refcnt == 0) {
1367			trunk->parent->if_vlantrunk = NULL;
1368			/*
1369			 * XXXGL: If some ithread has already entered
1370			 * vlan_input() and is now blocked on the trunk
1371			 * lock, then it should preempt us right after
1372			 * unlock and finish its work. Then we will acquire
1373			 * lock again in trunk_destroy().
1374			 */
1375			TRUNK_UNLOCK(trunk);
1376			trunk_destroy(trunk);
1377		} else
1378			TRUNK_UNLOCK(trunk);
1379	}
1380
1381	/* Disconnect from parent. */
1382	if (ifv->ifv_pflags)
1383		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1384	ifp->if_mtu = ETHERMTU;
1385	ifp->if_link_state = LINK_STATE_UNKNOWN;
1386	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1387
1388	/*
1389	 * Only dispatch an event if vlan was
1390	 * attached, otherwise there is nothing
1391	 * to cleanup anyway.
1392	 */
1393	if (parent != NULL)
1394		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_tag);
1395}
1396
1397/* Handle a reference counted flag that should be set on the parent as well */
1398static int
1399vlan_setflag(struct ifnet *ifp, int flag, int status,
1400	     int (*func)(struct ifnet *, int))
1401{
1402	struct ifvlan *ifv;
1403	int error;
1404
1405	/* XXX VLAN_LOCK_ASSERT(); */
1406
1407	ifv = ifp->if_softc;
1408	status = status ? (ifp->if_flags & flag) : 0;
1409	/* Now "status" contains the flag value or 0 */
1410
1411	/*
1412	 * See if recorded parent's status is different from what
1413	 * we want it to be.  If it is, flip it.  We record parent's
1414	 * status in ifv_pflags so that we won't clear parent's flag
1415	 * we haven't set.  In fact, we don't clear or set parent's
1416	 * flags directly, but get or release references to them.
1417	 * That's why we can be sure that recorded flags still are
1418	 * in accord with actual parent's flags.
1419	 */
1420	if (status != (ifv->ifv_pflags & flag)) {
1421		error = (*func)(PARENT(ifv), status);
1422		if (error)
1423			return (error);
1424		ifv->ifv_pflags &= ~flag;
1425		ifv->ifv_pflags |= status;
1426	}
1427	return (0);
1428}
1429
1430/*
1431 * Handle IFF_* flags that require certain changes on the parent:
1432 * if "status" is true, update parent's flags respective to our if_flags;
1433 * if "status" is false, forcedly clear the flags set on parent.
1434 */
1435static int
1436vlan_setflags(struct ifnet *ifp, int status)
1437{
1438	int error, i;
1439
1440	for (i = 0; vlan_pflags[i].flag; i++) {
1441		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1442				     status, vlan_pflags[i].func);
1443		if (error)
1444			return (error);
1445	}
1446	return (0);
1447}
1448
1449/* Inform all vlans that their parent has changed link state */
1450static void
1451vlan_link_state(struct ifnet *ifp)
1452{
1453	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1454	struct ifvlan *ifv;
1455	int i;
1456
1457	TRUNK_LOCK(trunk);
1458#ifdef VLAN_ARRAY
1459	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1460		if (trunk->vlans[i] != NULL) {
1461			ifv = trunk->vlans[i];
1462#else
1463	for (i = 0; i < (1 << trunk->hwidth); i++)
1464		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1465#endif
1466			ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1467			if_link_state_change(ifv->ifv_ifp,
1468			    trunk->parent->if_link_state);
1469		}
1470	TRUNK_UNLOCK(trunk);
1471}
1472
1473static void
1474vlan_capabilities(struct ifvlan *ifv)
1475{
1476	struct ifnet *p = PARENT(ifv);
1477	struct ifnet *ifp = ifv->ifv_ifp;
1478
1479	TRUNK_LOCK_ASSERT(TRUNK(ifv));
1480
1481	/*
1482	 * If the parent interface can do checksum offloading
1483	 * on VLANs, then propagate its hardware-assisted
1484	 * checksumming flags. Also assert that checksum
1485	 * offloading requires hardware VLAN tagging.
1486	 */
1487	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1488		ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1489
1490	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1491	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1492		ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1493		ifp->if_hwassist = p->if_hwassist & (CSUM_IP | CSUM_TCP |
1494		    CSUM_UDP | CSUM_SCTP | CSUM_FRAGMENT);
1495	} else {
1496		ifp->if_capenable = 0;
1497		ifp->if_hwassist = 0;
1498	}
1499	/*
1500	 * If the parent interface can do TSO on VLANs then
1501	 * propagate the hardware-assisted flag. TSO on VLANs
1502	 * does not necessarily require hardware VLAN tagging.
1503	 */
1504	if (p->if_hw_tsomax > 0)
1505		ifp->if_hw_tsomax = p->if_hw_tsomax;
1506	if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1507		ifp->if_capabilities |= p->if_capabilities & IFCAP_TSO;
1508	if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1509		ifp->if_capenable |= p->if_capenable & IFCAP_TSO;
1510		ifp->if_hwassist |= p->if_hwassist & CSUM_TSO;
1511	} else {
1512		ifp->if_capenable &= ~(p->if_capenable & IFCAP_TSO);
1513		ifp->if_hwassist &= ~(p->if_hwassist & CSUM_TSO);
1514	}
1515
1516	/*
1517	 * If the parent interface can offload TCP connections over VLANs then
1518	 * propagate its TOE capability to the VLAN interface.
1519	 *
1520	 * All TOE drivers in the tree today can deal with VLANs.  If this
1521	 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1522	 * with its own bit.
1523	 */
1524#define	IFCAP_VLAN_TOE IFCAP_TOE
1525	if (p->if_capabilities & IFCAP_VLAN_TOE)
1526		ifp->if_capabilities |= p->if_capabilities & IFCAP_TOE;
1527	if (p->if_capenable & IFCAP_VLAN_TOE) {
1528		TOEDEV(ifp) = TOEDEV(p);
1529		ifp->if_capenable |= p->if_capenable & IFCAP_TOE;
1530	}
1531}
1532
1533static void
1534vlan_trunk_capabilities(struct ifnet *ifp)
1535{
1536	struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1537	struct ifvlan *ifv;
1538	int i;
1539
1540	TRUNK_LOCK(trunk);
1541#ifdef VLAN_ARRAY
1542	for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1543		if (trunk->vlans[i] != NULL) {
1544			ifv = trunk->vlans[i];
1545#else
1546	for (i = 0; i < (1 << trunk->hwidth); i++) {
1547		LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1548#endif
1549			vlan_capabilities(ifv);
1550	}
1551	TRUNK_UNLOCK(trunk);
1552}
1553
1554static int
1555vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1556{
1557	struct ifnet *p;
1558	struct ifreq *ifr;
1559	struct ifaddr *ifa;
1560	struct ifvlan *ifv;
1561	struct vlanreq vlr;
1562	int error = 0;
1563
1564	ifr = (struct ifreq *)data;
1565	ifa = (struct ifaddr *) data;
1566	ifv = ifp->if_softc;
1567
1568	switch (cmd) {
1569	case SIOCSIFADDR:
1570		ifp->if_flags |= IFF_UP;
1571#ifdef INET
1572		if (ifa->ifa_addr->sa_family == AF_INET)
1573			arp_ifinit(ifp, ifa);
1574#endif
1575		break;
1576	case SIOCGIFADDR:
1577                {
1578			struct sockaddr *sa;
1579
1580			sa = (struct sockaddr *)&ifr->ifr_data;
1581			bcopy(IF_LLADDR(ifp), sa->sa_data, ifp->if_addrlen);
1582                }
1583		break;
1584	case SIOCGIFMEDIA:
1585		VLAN_LOCK();
1586		if (TRUNK(ifv) != NULL) {
1587			p = PARENT(ifv);
1588			VLAN_UNLOCK();
1589			error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1590			/* Limit the result to the parent's current config. */
1591			if (error == 0) {
1592				struct ifmediareq *ifmr;
1593
1594				ifmr = (struct ifmediareq *)data;
1595				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1596					ifmr->ifm_count = 1;
1597					error = copyout(&ifmr->ifm_current,
1598						ifmr->ifm_ulist,
1599						sizeof(int));
1600				}
1601			}
1602		} else {
1603			VLAN_UNLOCK();
1604			error = EINVAL;
1605		}
1606		break;
1607
1608	case SIOCSIFMEDIA:
1609		error = EINVAL;
1610		break;
1611
1612	case SIOCSIFMTU:
1613		/*
1614		 * Set the interface MTU.
1615		 */
1616		VLAN_LOCK();
1617		if (TRUNK(ifv) != NULL) {
1618			if (ifr->ifr_mtu >
1619			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1620			    ifr->ifr_mtu <
1621			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1622				error = EINVAL;
1623			else
1624				ifp->if_mtu = ifr->ifr_mtu;
1625		} else
1626			error = EINVAL;
1627		VLAN_UNLOCK();
1628		break;
1629
1630	case SIOCSETVLAN:
1631#ifdef VIMAGE
1632		if (ifp->if_vnet != ifp->if_home_vnet) {
1633			error = EPERM;
1634			break;
1635		}
1636#endif
1637		error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1638		if (error)
1639			break;
1640		if (vlr.vlr_parent[0] == '\0') {
1641			vlan_unconfig(ifp);
1642			break;
1643		}
1644		p = ifunit(vlr.vlr_parent);
1645		if (p == NULL) {
1646			error = ENOENT;
1647			break;
1648		}
1649		/*
1650		 * Don't let the caller set up a VLAN tag with
1651		 * anything except VLID bits.
1652		 */
1653		if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1654			error = EINVAL;
1655			break;
1656		}
1657		error = vlan_config(ifv, p, vlr.vlr_tag);
1658		if (error)
1659			break;
1660
1661		/* Update flags on the parent, if necessary. */
1662		vlan_setflags(ifp, 1);
1663		break;
1664
1665	case SIOCGETVLAN:
1666#ifdef VIMAGE
1667		if (ifp->if_vnet != ifp->if_home_vnet) {
1668			error = EPERM;
1669			break;
1670		}
1671#endif
1672		bzero(&vlr, sizeof(vlr));
1673		VLAN_LOCK();
1674		if (TRUNK(ifv) != NULL) {
1675			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1676			    sizeof(vlr.vlr_parent));
1677			vlr.vlr_tag = ifv->ifv_tag;
1678		}
1679		VLAN_UNLOCK();
1680		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1681		break;
1682
1683	case SIOCSIFFLAGS:
1684		/*
1685		 * We should propagate selected flags to the parent,
1686		 * e.g., promiscuous mode.
1687		 */
1688		if (TRUNK(ifv) != NULL)
1689			error = vlan_setflags(ifp, 1);
1690		break;
1691
1692	case SIOCADDMULTI:
1693	case SIOCDELMULTI:
1694		/*
1695		 * If we don't have a parent, just remember the membership for
1696		 * when we do.
1697		 */
1698		if (TRUNK(ifv) != NULL)
1699			error = vlan_setmulti(ifp);
1700		break;
1701
1702	default:
1703		error = EINVAL;
1704		break;
1705	}
1706
1707	return (error);
1708}
1709