xref: /netbsd-src/sys/net/if_vlan.c (revision cac8e449158efc7261bebc8657cbb0125a2cfdde)
1 /*	$NetBSD: if_vlan.c,v 1.59 2008/06/15 16:37:21 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2000, 2001 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran, and by Jason R. Thorpe of Zembu Labs, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright 1998 Massachusetts Institute of Technology
34  *
35  * Permission to use, copy, modify, and distribute this software and
36  * its documentation for any purpose and without fee is hereby
37  * granted, provided that both the above copyright notice and this
38  * permission notice appear in all copies, that both the above
39  * copyright notice and this permission notice appear in all
40  * supporting documentation, and that the name of M.I.T. not be used
41  * in advertising or publicity pertaining to distribution of the
42  * software without specific, written prior permission.  M.I.T. makes
43  * no representations about the suitability of this software for any
44  * purpose.  It is provided "as is" without express or implied
45  * warranty.
46  *
47  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
48  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
49  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
50  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
51  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
52  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
53  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
54  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
55  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
56  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
57  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  *
60  * from FreeBSD: if_vlan.c,v 1.16 2000/03/26 15:21:40 charnier Exp
61  * via OpenBSD: if_vlan.c,v 1.4 2000/05/15 19:15:00 chris Exp
62  */
63 
64 /*
65  * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.  Might be
66  * extended some day to also handle IEEE 802.1P priority tagging.  This is
67  * sort of sneaky in the implementation, since we need to pretend to be
68  * enough of an Ethernet implementation to make ARP work.  The way we do
69  * this is by telling everyone that we are an Ethernet interface, and then
70  * catch the packets that ether_output() left on our output queue when it
71  * calls if_start(), rewrite them for use by the real outgoing interface,
72  * and ask it to send them.
73  *
74  * TODO:
75  *
76  *	- Need some way to notify vlan interfaces when the parent
77  *	  interface changes MTU.
78  */
79 
80 #include <sys/cdefs.h>
81 __KERNEL_RCSID(0, "$NetBSD: if_vlan.c,v 1.59 2008/06/15 16:37:21 christos Exp $");
82 
83 #include "opt_inet.h"
84 #include "bpfilter.h"
85 
86 #include <sys/param.h>
87 #include <sys/kernel.h>
88 #include <sys/mbuf.h>
89 #include <sys/queue.h>
90 #include <sys/socket.h>
91 #include <sys/sockio.h>
92 #include <sys/systm.h>
93 #include <sys/proc.h>
94 #include <sys/kauth.h>
95 
96 #if NBPFILTER > 0
97 #include <net/bpf.h>
98 #endif
99 #include <net/if.h>
100 #include <net/if_dl.h>
101 #include <net/if_types.h>
102 #include <net/if_ether.h>
103 #include <net/if_vlanvar.h>
104 
105 #ifdef INET
106 #include <netinet/in.h>
107 #include <netinet/if_inarp.h>
108 #endif
109 
110 struct vlan_mc_entry {
111 	LIST_ENTRY(vlan_mc_entry)	mc_entries;
112 	/*
113 	 * A key to identify this entry.  The mc_addr below can't be
114 	 * used since multiple sockaddr may mapped into the same
115 	 * ether_multi (e.g., AF_UNSPEC).
116 	 */
117 	union {
118 		struct ether_multi	*mcu_enm;
119 	} mc_u;
120 	struct sockaddr_storage		mc_addr;
121 };
122 
123 #define	mc_enm		mc_u.mcu_enm
124 
125 struct ifvlan {
126 	union {
127 		struct ethercom ifvu_ec;
128 	} ifv_u;
129 	struct ifnet *ifv_p;	/* parent interface of this vlan */
130 	struct ifv_linkmib {
131 		const struct vlan_multisw *ifvm_msw;
132 		int	ifvm_encaplen;	/* encapsulation length */
133 		int	ifvm_mtufudge;	/* MTU fudged by this much */
134 		int	ifvm_mintu;	/* min transmission unit */
135 		uint16_t ifvm_proto;	/* encapsulation ethertype */
136 		uint16_t ifvm_tag;	/* tag to apply on packets */
137 	} ifv_mib;
138 	LIST_HEAD(__vlan_mchead, vlan_mc_entry) ifv_mc_listhead;
139 	LIST_ENTRY(ifvlan) ifv_list;
140 	int ifv_flags;
141 };
142 
143 #define	IFVF_PROMISC	0x01		/* promiscuous mode enabled */
144 
145 #define	ifv_ec		ifv_u.ifvu_ec
146 
147 #define	ifv_if		ifv_ec.ec_if
148 
149 #define	ifv_msw		ifv_mib.ifvm_msw
150 #define	ifv_encaplen	ifv_mib.ifvm_encaplen
151 #define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
152 #define	ifv_mintu	ifv_mib.ifvm_mintu
153 #define	ifv_tag		ifv_mib.ifvm_tag
154 
155 struct vlan_multisw {
156 	int	(*vmsw_addmulti)(struct ifvlan *, struct ifreq *);
157 	int	(*vmsw_delmulti)(struct ifvlan *, struct ifreq *);
158 	void	(*vmsw_purgemulti)(struct ifvlan *);
159 };
160 
161 static int	vlan_ether_addmulti(struct ifvlan *, struct ifreq *);
162 static int	vlan_ether_delmulti(struct ifvlan *, struct ifreq *);
163 static void	vlan_ether_purgemulti(struct ifvlan *);
164 
165 const struct vlan_multisw vlan_ether_multisw = {
166 	vlan_ether_addmulti,
167 	vlan_ether_delmulti,
168 	vlan_ether_purgemulti,
169 };
170 
171 static int	vlan_clone_create(struct if_clone *, int);
172 static int	vlan_clone_destroy(struct ifnet *);
173 static int	vlan_config(struct ifvlan *, struct ifnet *);
174 static int	vlan_ioctl(struct ifnet *, u_long, void *);
175 static void	vlan_start(struct ifnet *);
176 static void	vlan_unconfig(struct ifnet *);
177 
178 void		vlanattach(int);
179 
180 /* XXX This should be a hash table with the tag as the basis of the key. */
181 static LIST_HEAD(, ifvlan) ifv_list;
182 
183 struct if_clone vlan_cloner =
184     IF_CLONE_INITIALIZER("vlan", vlan_clone_create, vlan_clone_destroy);
185 
186 /* Used to pad ethernet frames with < ETHER_MIN_LEN bytes */
187 static char vlan_zero_pad_buff[ETHER_MIN_LEN];
188 
189 void
190 vlanattach(int n)
191 {
192 
193 	LIST_INIT(&ifv_list);
194 	if_clone_attach(&vlan_cloner);
195 }
196 
197 static void
198 vlan_reset_linkname(struct ifnet *ifp)
199 {
200 
201 	/*
202 	 * We start out with a "802.1Q VLAN" type and zero-length
203 	 * addresses.  When we attach to a parent interface, we
204 	 * inherit its type, address length, address, and data link
205 	 * type.
206 	 */
207 
208 	ifp->if_type = IFT_L2VLAN;
209 	ifp->if_addrlen = 0;
210 	ifp->if_dlt = DLT_NULL;
211 	if_alloc_sadl(ifp);
212 }
213 
214 static int
215 vlan_clone_create(struct if_clone *ifc, int unit)
216 {
217 	struct ifvlan *ifv;
218 	struct ifnet *ifp;
219 	int s;
220 
221 	ifv = malloc(sizeof(struct ifvlan), M_DEVBUF, M_WAITOK|M_ZERO);
222 	ifp = &ifv->ifv_if;
223 	LIST_INIT(&ifv->ifv_mc_listhead);
224 
225 	s = splnet();
226 	LIST_INSERT_HEAD(&ifv_list, ifv, ifv_list);
227 	splx(s);
228 
229 	if_initname(ifp, ifc->ifc_name, unit);
230 	ifp->if_softc = ifv;
231 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
232 	ifp->if_start = vlan_start;
233 	ifp->if_ioctl = vlan_ioctl;
234 	IFQ_SET_READY(&ifp->if_snd);
235 
236 	if_attach(ifp);
237 	vlan_reset_linkname(ifp);
238 
239 	return (0);
240 }
241 
242 static int
243 vlan_clone_destroy(struct ifnet *ifp)
244 {
245 	struct ifvlan *ifv = ifp->if_softc;
246 	int s;
247 
248 	s = splnet();
249 	LIST_REMOVE(ifv, ifv_list);
250 	vlan_unconfig(ifp);
251 	splx(s);
252 
253 	if_detach(ifp);
254 	free(ifv, M_DEVBUF);
255 
256 	return (0);
257 }
258 
259 /*
260  * Configure a VLAN interface.  Must be called at splnet().
261  */
262 static int
263 vlan_config(struct ifvlan *ifv, struct ifnet *p)
264 {
265 	struct ifnet *ifp = &ifv->ifv_if;
266 	int error;
267 
268 	if (ifv->ifv_p != NULL)
269 		return (EBUSY);
270 
271 	switch (p->if_type) {
272 	case IFT_ETHER:
273 	    {
274 		struct ethercom *ec = (void *) p;
275 
276 		ifv->ifv_msw = &vlan_ether_multisw;
277 		ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
278 		ifv->ifv_mintu = ETHERMIN;
279 
280 		/*
281 		 * If the parent supports the VLAN_MTU capability,
282 		 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
283 		 * enable it.
284 		 */
285 		if (ec->ec_nvlans++ == 0 &&
286 		    (ec->ec_capabilities & ETHERCAP_VLAN_MTU) != 0) {
287 			/*
288 			 * Enable Tx/Rx of VLAN-sized frames.
289 			 */
290 			ec->ec_capenable |= ETHERCAP_VLAN_MTU;
291 			if (p->if_flags & IFF_UP) {
292 				struct ifreq ifr;
293 
294 				ifr.ifr_flags = p->if_flags;
295 				error = (*p->if_ioctl)(p, SIOCSIFFLAGS,
296 				    (void *) &ifr);
297 				if (error) {
298 					if (ec->ec_nvlans-- == 1)
299 						ec->ec_capenable &=
300 						    ~ETHERCAP_VLAN_MTU;
301 					return (error);
302 				}
303 			}
304 			ifv->ifv_mtufudge = 0;
305 		} else if ((ec->ec_capabilities & ETHERCAP_VLAN_MTU) == 0) {
306 			/*
307 			 * Fudge the MTU by the encapsulation size.  This
308 			 * makes us incompatible with strictly compliant
309 			 * 802.1Q implementations, but allows us to use
310 			 * the feature with other NetBSD implementations,
311 			 * which might still be useful.
312 			 */
313 			ifv->ifv_mtufudge = ifv->ifv_encaplen;
314 		}
315 
316 		/*
317 		 * If the parent interface can do hardware-assisted
318 		 * VLAN encapsulation, then propagate its hardware-
319 		 * assisted checksumming flags.
320 		 */
321 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING)
322 			ifp->if_capabilities = p->if_capabilities &
323 			    (IFCAP_CSUM_IPv4_Tx|IFCAP_CSUM_IPv4_Rx|
324 			     IFCAP_CSUM_TCPv4_Tx|IFCAP_CSUM_TCPv4_Rx|
325 			     IFCAP_CSUM_UDPv4_Tx|IFCAP_CSUM_UDPv4_Rx|
326 			     IFCAP_CSUM_TCPv6_Tx|IFCAP_CSUM_TCPv6_Rx|
327 			     IFCAP_CSUM_UDPv6_Tx|IFCAP_CSUM_UDPv6_Rx);
328 
329 		/*
330 		 * We inherit the parent's Ethernet address.
331 		 */
332 		ether_ifattach(ifp, CLLADDR(p->if_sadl));
333 		ifp->if_hdrlen = sizeof(struct ether_vlan_header); /* XXX? */
334 		break;
335 	    }
336 
337 	default:
338 		return (EPROTONOSUPPORT);
339 	}
340 
341 	ifv->ifv_p = p;
342 	ifv->ifv_if.if_mtu = p->if_mtu - ifv->ifv_mtufudge;
343 	ifv->ifv_if.if_flags = p->if_flags &
344 	    (IFF_UP | IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
345 
346 	/*
347 	 * Inherit the if_type from the parent.  This allows us
348 	 * to participate in bridges of that type.
349 	 */
350 	ifv->ifv_if.if_type = p->if_type;
351 
352 	return (0);
353 }
354 
355 /*
356  * Unconfigure a VLAN interface.  Must be called at splnet().
357  */
358 static void
359 vlan_unconfig(struct ifnet *ifp)
360 {
361 	struct ifvlan *ifv = ifp->if_softc;
362 
363 	if (ifv->ifv_p == NULL)
364 		return;
365 
366 	/*
367  	 * Since the interface is being unconfigured, we need to empty the
368 	 * list of multicast groups that we may have joined while we were
369 	 * alive and remove them from the parent's list also.
370 	 */
371 	(*ifv->ifv_msw->vmsw_purgemulti)(ifv);
372 
373 	/* Disconnect from parent. */
374 	switch (ifv->ifv_p->if_type) {
375 	case IFT_ETHER:
376 	    {
377 		struct ethercom *ec = (void *) ifv->ifv_p;
378 
379 		if (ec->ec_nvlans-- == 1) {
380 			/*
381 			 * Disable Tx/Rx of VLAN-sized frames.
382 			 */
383 			ec->ec_capenable &= ~ETHERCAP_VLAN_MTU;
384 			if (ifv->ifv_p->if_flags & IFF_UP) {
385 				struct ifreq ifr;
386 
387 				ifr.ifr_flags = ifv->ifv_p->if_flags;
388 				(void) (*ifv->ifv_p->if_ioctl)(ifv->ifv_p,
389 				    SIOCSIFFLAGS, (void *) &ifr);
390 			}
391 		}
392 
393 		ether_ifdetach(ifp);
394 		vlan_reset_linkname(ifp);
395 		break;
396 	    }
397 
398 #ifdef DIAGNOSTIC
399 	default:
400 		panic("vlan_unconfig: impossible");
401 #endif
402 	}
403 
404 	ifv->ifv_p = NULL;
405 	ifv->ifv_if.if_mtu = 0;
406 	ifv->ifv_flags = 0;
407 
408 	if_down(ifp);
409 	ifp->if_flags &= ~(IFF_UP|IFF_RUNNING);
410 	ifp->if_capabilities = 0;
411 }
412 
413 /*
414  * Called when a parent interface is detaching; destroy any VLAN
415  * configuration for the parent interface.
416  */
417 void
418 vlan_ifdetach(struct ifnet *p)
419 {
420 	struct ifvlan *ifv;
421 	int s;
422 
423 	s = splnet();
424 
425 	for (ifv = LIST_FIRST(&ifv_list); ifv != NULL;
426 	     ifv = LIST_NEXT(ifv, ifv_list)) {
427 		if (ifv->ifv_p == p)
428 			vlan_unconfig(&ifv->ifv_if);
429 	}
430 
431 	splx(s);
432 }
433 
434 static int
435 vlan_set_promisc(struct ifnet *ifp)
436 {
437 	struct ifvlan *ifv = ifp->if_softc;
438 	int error = 0;
439 
440 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
441 		if ((ifv->ifv_flags & IFVF_PROMISC) == 0) {
442 			error = ifpromisc(ifv->ifv_p, 1);
443 			if (error == 0)
444 				ifv->ifv_flags |= IFVF_PROMISC;
445 		}
446 	} else {
447 		if ((ifv->ifv_flags & IFVF_PROMISC) != 0) {
448 			error = ifpromisc(ifv->ifv_p, 0);
449 			if (error == 0)
450 				ifv->ifv_flags &= ~IFVF_PROMISC;
451 		}
452 	}
453 
454 	return (error);
455 }
456 
457 static int
458 vlan_ioctl(struct ifnet *ifp, u_long cmd, void *data)
459 {
460 	struct lwp *l = curlwp;	/* XXX */
461 	struct ifvlan *ifv = ifp->if_softc;
462 	struct ifaddr *ifa = (struct ifaddr *) data;
463 	struct ifreq *ifr = (struct ifreq *) data;
464 	struct ifnet *pr;
465 	struct vlanreq vlr;
466 	struct sockaddr *sa;
467 	int s, error = 0;
468 
469 	s = splnet();
470 
471 	switch (cmd) {
472 	case SIOCSIFADDR:
473 		if (ifv->ifv_p != NULL) {
474 			ifp->if_flags |= IFF_UP;
475 
476 			switch (ifa->ifa_addr->sa_family) {
477 #ifdef INET
478 			case AF_INET:
479 				arp_ifinit(ifp, ifa);
480 				break;
481 #endif
482 			default:
483 				break;
484 			}
485 		} else {
486 			error = EINVAL;
487 		}
488 		break;
489 
490 	case SIOCGIFADDR:
491 		sa = (struct sockaddr *)&ifr->ifr_data;
492 		memcpy(sa->sa_data, CLLADDR(ifp->if_sadl), ifp->if_addrlen);
493 		break;
494 
495 	case SIOCSIFMTU:
496 		if (ifv->ifv_p == NULL)
497 			error = EINVAL;
498 		else if (
499 		    ifr->ifr_mtu > (ifv->ifv_p->if_mtu - ifv->ifv_mtufudge) ||
500 		    ifr->ifr_mtu < (ifv->ifv_mintu - ifv->ifv_mtufudge))
501 			error = EINVAL;
502 		else if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
503 			error = 0;
504 		break;
505 
506 	case SIOCSETVLAN:
507 		if ((error = kauth_authorize_network(l->l_cred,
508 		    KAUTH_NETWORK_INTERFACE,
509 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
510 		    NULL)) != 0)
511 			break;
512 		if ((error = copyin(ifr->ifr_data, &vlr, sizeof(vlr))) != 0)
513 			break;
514 		if (vlr.vlr_parent[0] == '\0') {
515 			vlan_unconfig(ifp);
516 			break;
517 		}
518 		if (vlr.vlr_tag != EVL_VLANOFTAG(vlr.vlr_tag)) {
519 			error = EINVAL;		 /* check for valid tag */
520 			break;
521 		}
522 		if ((pr = ifunit(vlr.vlr_parent)) == 0) {
523 			error = ENOENT;
524 			break;
525 		}
526 		if ((error = vlan_config(ifv, pr)) != 0)
527 			break;
528 		ifv->ifv_tag = vlr.vlr_tag;
529 		ifp->if_flags |= IFF_RUNNING;
530 
531 		/* Update promiscuous mode, if necessary. */
532 		vlan_set_promisc(ifp);
533 		break;
534 
535 	case SIOCGETVLAN:
536 		memset(&vlr, 0, sizeof(vlr));
537 		if (ifv->ifv_p != NULL) {
538 			snprintf(vlr.vlr_parent, sizeof(vlr.vlr_parent), "%s",
539 			    ifv->ifv_p->if_xname);
540 			vlr.vlr_tag = ifv->ifv_tag;
541 		}
542 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
543 		break;
544 
545 	case SIOCSIFFLAGS:
546 		/*
547 		 * For promiscuous mode, we enable promiscuous mode on
548 		 * the parent if we need promiscuous on the VLAN interface.
549 		 */
550 		if (ifv->ifv_p != NULL)
551 			error = vlan_set_promisc(ifp);
552 		break;
553 
554 	case SIOCADDMULTI:
555 		error = (ifv->ifv_p != NULL) ?
556 		    (*ifv->ifv_msw->vmsw_addmulti)(ifv, ifr) : EINVAL;
557 		break;
558 
559 	case SIOCDELMULTI:
560 		error = (ifv->ifv_p != NULL) ?
561 		    (*ifv->ifv_msw->vmsw_delmulti)(ifv, ifr) : EINVAL;
562 		break;
563 
564 	default:
565 		error = EINVAL;
566 	}
567 
568 	splx(s);
569 
570 	return (error);
571 }
572 
573 static int
574 vlan_ether_addmulti(struct ifvlan *ifv, struct ifreq *ifr)
575 {
576 	const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr);
577 	struct vlan_mc_entry *mc;
578 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
579 	int error;
580 
581 	if (sa->sa_len > sizeof(struct sockaddr_storage))
582 		return (EINVAL);
583 
584 	error = ether_addmulti(sa, &ifv->ifv_ec);
585 	if (error != ENETRESET)
586 		return (error);
587 
588 	/*
589 	 * This is new multicast address.  We have to tell parent
590 	 * about it.  Also, remember this multicast address so that
591 	 * we can delete them on unconfigure.
592 	 */
593 	MALLOC(mc, struct vlan_mc_entry *, sizeof(struct vlan_mc_entry),
594 	    M_DEVBUF, M_NOWAIT);
595 	if (mc == NULL) {
596 		error = ENOMEM;
597 		goto alloc_failed;
598 	}
599 
600 	/*
601 	 * As ether_addmulti() returns ENETRESET, following two
602 	 * statement shouldn't fail.
603 	 */
604 	(void)ether_multiaddr(sa, addrlo, addrhi);
605 	ETHER_LOOKUP_MULTI(addrlo, addrhi, &ifv->ifv_ec, mc->mc_enm);
606 	memcpy(&mc->mc_addr, sa, sa->sa_len);
607 	LIST_INSERT_HEAD(&ifv->ifv_mc_listhead, mc, mc_entries);
608 
609 	error = (*ifv->ifv_p->if_ioctl)(ifv->ifv_p, SIOCADDMULTI,
610 	    (void *)ifr);
611 	if (error != 0)
612 		goto ioctl_failed;
613 	return (error);
614 
615  ioctl_failed:
616 	LIST_REMOVE(mc, mc_entries);
617 	FREE(mc, M_DEVBUF);
618  alloc_failed:
619 	(void)ether_delmulti(sa, &ifv->ifv_ec);
620 	return (error);
621 }
622 
623 static int
624 vlan_ether_delmulti(struct ifvlan *ifv, struct ifreq *ifr)
625 {
626 	const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr);
627 	struct ether_multi *enm;
628 	struct vlan_mc_entry *mc;
629 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
630 	int error;
631 
632 	/*
633 	 * Find a key to lookup vlan_mc_entry.  We have to do this
634 	 * before calling ether_delmulti for obvious reason.
635 	 */
636 	if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0)
637 		return (error);
638 	ETHER_LOOKUP_MULTI(addrlo, addrhi, &ifv->ifv_ec, enm);
639 
640 	error = ether_delmulti(sa, &ifv->ifv_ec);
641 	if (error != ENETRESET)
642 		return (error);
643 
644 	/* We no longer use this multicast address.  Tell parent so. */
645 	error = (*ifv->ifv_p->if_ioctl)(ifv->ifv_p, SIOCDELMULTI,
646 	    (void *)ifr);
647 	if (error == 0) {
648 		/* And forget about this address. */
649 		for (mc = LIST_FIRST(&ifv->ifv_mc_listhead); mc != NULL;
650 		    mc = LIST_NEXT(mc, mc_entries)) {
651 			if (mc->mc_enm == enm) {
652 				LIST_REMOVE(mc, mc_entries);
653 				FREE(mc, M_DEVBUF);
654 				break;
655 			}
656 		}
657 		KASSERT(mc != NULL);
658 	} else
659 		(void)ether_addmulti(sa, &ifv->ifv_ec);
660 	return (error);
661 }
662 
663 /*
664  * Delete any multicast address we have asked to add from parent
665  * interface.  Called when the vlan is being unconfigured.
666  */
667 static void
668 vlan_ether_purgemulti(struct ifvlan *ifv)
669 {
670 	struct ifnet *ifp = ifv->ifv_p;		/* Parent. */
671 	struct vlan_mc_entry *mc;
672 	union {
673 		struct ifreq ifreq;
674 		struct {
675 			char ifr_name[IFNAMSIZ];
676 			struct sockaddr_storage ifr_ss;
677 		} ifreq_storage;
678 	} ifreq;
679 	struct ifreq *ifr = &ifreq.ifreq;
680 
681 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
682 	while ((mc = LIST_FIRST(&ifv->ifv_mc_listhead)) != NULL) {
683 		ifreq_setaddr(SIOCDELMULTI, ifr,
684 		    (const struct sockaddr *)&mc->mc_addr);
685 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, (void *)ifr);
686 		LIST_REMOVE(mc, mc_entries);
687 		FREE(mc, M_DEVBUF);
688 	}
689 }
690 
691 static void
692 vlan_start(struct ifnet *ifp)
693 {
694 	struct ifvlan *ifv = ifp->if_softc;
695 	struct ifnet *p = ifv->ifv_p;
696 	struct ethercom *ec = (void *) ifv->ifv_p;
697 	struct mbuf *m;
698 	int error;
699 	ALTQ_DECL(struct altq_pktattr pktattr;)
700 
701 	ifp->if_flags |= IFF_OACTIVE;
702 
703 	for (;;) {
704 		IFQ_DEQUEUE(&ifp->if_snd, m);
705 		if (m == NULL)
706 			break;
707 
708 #ifdef ALTQ
709 		/*
710 		 * If ALTQ is enabled on the parent interface, do
711 		 * classification; the queueing discipline might
712 		 * not require classification, but might require
713 		 * the address family/header pointer in the pktattr.
714 		 */
715 		if (ALTQ_IS_ENABLED(&p->if_snd)) {
716 			switch (p->if_type) {
717 			case IFT_ETHER:
718 				altq_etherclassify(&p->if_snd, m, &pktattr);
719 				break;
720 #ifdef DIAGNOSTIC
721 			default:
722 				panic("vlan_start: impossible (altq)");
723 #endif
724 			}
725 		}
726 #endif /* ALTQ */
727 
728 #if NBPFILTER > 0
729 		if (ifp->if_bpf)
730 			bpf_mtap(ifp->if_bpf, m);
731 #endif
732 		/*
733 		 * If the parent can insert the tag itself, just mark
734 		 * the tag in the mbuf header.
735 		 */
736 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
737 			struct m_tag *mtag;
738 
739 			mtag = m_tag_get(PACKET_TAG_VLAN, sizeof(u_int),
740 			    M_NOWAIT);
741 			if (mtag == NULL) {
742 				ifp->if_oerrors++;
743 				m_freem(m);
744 				continue;
745 			}
746 			*(u_int *)(mtag + 1) = ifv->ifv_tag;
747 			m_tag_prepend(m, mtag);
748 		} else {
749 			/*
750 			 * insert the tag ourselves
751 			 */
752 			M_PREPEND(m, ifv->ifv_encaplen, M_DONTWAIT);
753 			if (m == NULL) {
754 				printf("%s: unable to prepend encap header",
755 				    ifv->ifv_p->if_xname);
756 				ifp->if_oerrors++;
757 				continue;
758 			}
759 
760 			switch (p->if_type) {
761 			case IFT_ETHER:
762 			    {
763 				struct ether_vlan_header *evl;
764 
765 				if (m->m_len < sizeof(struct ether_vlan_header))
766 					m = m_pullup(m,
767 					    sizeof(struct ether_vlan_header));
768 				if (m == NULL) {
769 					printf("%s: unable to pullup encap "
770 					    "header", ifv->ifv_p->if_xname);
771 					ifp->if_oerrors++;
772 					continue;
773 				}
774 
775 				/*
776 				 * Transform the Ethernet header into an
777 				 * Ethernet header with 802.1Q encapsulation.
778 				 */
779 				memmove(mtod(m, void *),
780 				    mtod(m, char *) + ifv->ifv_encaplen,
781 				    sizeof(struct ether_header));
782 				evl = mtod(m, struct ether_vlan_header *);
783 				evl->evl_proto = evl->evl_encap_proto;
784 				evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
785 				evl->evl_tag = htons(ifv->ifv_tag);
786 
787 				/*
788 				 * To cater for VLAN-aware layer 2 ethernet
789 				 * switches which may need to strip the tag
790 				 * before forwarding the packet, make sure
791 				 * the packet+tag is at least 68 bytes long.
792 				 * This is necessary because our parent will
793 				 * only pad to 64 bytes (ETHER_MIN_LEN) and
794 				 * some switches will not pad by themselves
795 				 * after deleting a tag.
796 				 */
797 				if (m->m_pkthdr.len <
798 				    (ETHER_MIN_LEN + ETHER_VLAN_ENCAP_LEN)) {
799 					m_copyback(m, m->m_pkthdr.len,
800 					    (ETHER_MIN_LEN +
801 					     ETHER_VLAN_ENCAP_LEN) -
802 					     m->m_pkthdr.len,
803 					    vlan_zero_pad_buff);
804 				}
805 				break;
806 			    }
807 
808 #ifdef DIAGNOSTIC
809 			default:
810 				panic("vlan_start: impossible");
811 #endif
812 			}
813 		}
814 
815 		/*
816 		 * Send it, precisely as the parent's output routine
817 		 * would have.  We are already running at splnet.
818 		 */
819 		IFQ_ENQUEUE(&p->if_snd, m, &pktattr, error);
820 		if (error) {
821 			/* mbuf is already freed */
822 			ifp->if_oerrors++;
823 			continue;
824 		}
825 
826 		ifp->if_opackets++;
827 		if ((p->if_flags & (IFF_RUNNING|IFF_OACTIVE)) == IFF_RUNNING)
828 			(*p->if_start)(p);
829 	}
830 
831 	ifp->if_flags &= ~IFF_OACTIVE;
832 }
833 
834 /*
835  * Given an Ethernet frame, find a valid vlan interface corresponding to the
836  * given source interface and tag, then run the real packet through the
837  * parent's input routine.
838  */
839 void
840 vlan_input(struct ifnet *ifp, struct mbuf *m)
841 {
842 	struct ifvlan *ifv;
843 	u_int tag;
844 	struct m_tag *mtag;
845 
846 	mtag = m_tag_find(m, PACKET_TAG_VLAN, NULL);
847 	if (mtag != NULL) {
848 		/* m contains a normal ethernet frame, the tag is in mtag */
849 		tag = EVL_VLANOFTAG(*(u_int *)(mtag + 1));
850 		m_tag_delete(m, mtag);
851 	} else {
852 		switch (ifp->if_type) {
853 		case IFT_ETHER:
854 		    {
855 			struct ether_vlan_header *evl;
856 
857 			if (m->m_len < sizeof(struct ether_vlan_header) &&
858 			    (m = m_pullup(m,
859 			     sizeof(struct ether_vlan_header))) == NULL) {
860 				printf("%s: no memory for VLAN header, "
861 				    "dropping packet.\n", ifp->if_xname);
862 				return;
863 			}
864 			evl = mtod(m, struct ether_vlan_header *);
865 			KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN);
866 
867 			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
868 
869 			/*
870 			 * Restore the original ethertype.  We'll remove
871 			 * the encapsulation after we've found the vlan
872 			 * interface corresponding to the tag.
873 			 */
874 			evl->evl_encap_proto = evl->evl_proto;
875 			break;
876 		    }
877 
878 		default:
879 			tag = (u_int) -1;	/* XXX GCC */
880 #ifdef DIAGNOSTIC
881 			panic("vlan_input: impossible");
882 #endif
883 		}
884 	}
885 
886 	for (ifv = LIST_FIRST(&ifv_list); ifv != NULL;
887 	    ifv = LIST_NEXT(ifv, ifv_list))
888 		if (ifp == ifv->ifv_p && tag == ifv->ifv_tag)
889 			break;
890 
891 	if (ifv == NULL ||
892 	    (ifv->ifv_if.if_flags & (IFF_UP|IFF_RUNNING)) !=
893 	     (IFF_UP|IFF_RUNNING)) {
894 		m_freem(m);
895 		ifp->if_noproto++;
896 		return;
897 	}
898 
899 	/*
900 	 * Now, remove the encapsulation header.  The original
901 	 * header has already been fixed up above.
902 	 */
903 	if (mtag == NULL) {
904 		memmove(mtod(m, char *) + ifv->ifv_encaplen,
905 		    mtod(m, void *), sizeof(struct ether_header));
906 		m_adj(m, ifv->ifv_encaplen);
907 	}
908 
909 	m->m_pkthdr.rcvif = &ifv->ifv_if;
910 	ifv->ifv_if.if_ipackets++;
911 
912 #if NBPFILTER > 0
913 	if (ifv->ifv_if.if_bpf)
914 		bpf_mtap(ifv->ifv_if.if_bpf, m);
915 #endif
916 
917 	/* Pass it back through the parent's input routine. */
918 	(*ifp->if_input)(&ifv->ifv_if, m);
919 }
920