xref: /netbsd-src/sys/net/if_vlan.c (revision cd22f25e6f6d1cc1f197fe8c5468a80f51d1c4e1)
1 /*	$NetBSD: if_vlan.c,v 1.58 2008/04/28 20:24:09 martin 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 1.58 2008/04/28 20:24:09 martin 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);
222 	memset(ifv, 0, sizeof(struct ifvlan));
223 	ifp = &ifv->ifv_if;
224 	LIST_INIT(&ifv->ifv_mc_listhead);
225 
226 	s = splnet();
227 	LIST_INSERT_HEAD(&ifv_list, ifv, ifv_list);
228 	splx(s);
229 
230 	snprintf(ifp->if_xname, sizeof(ifp->if_xname), "%s%d", ifc->ifc_name,
231 	    unit);
232 	ifp->if_softc = ifv;
233 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
234 	ifp->if_start = vlan_start;
235 	ifp->if_ioctl = vlan_ioctl;
236 	IFQ_SET_READY(&ifp->if_snd);
237 
238 	if_attach(ifp);
239 	vlan_reset_linkname(ifp);
240 
241 	return (0);
242 }
243 
244 static int
245 vlan_clone_destroy(struct ifnet *ifp)
246 {
247 	struct ifvlan *ifv = ifp->if_softc;
248 	int s;
249 
250 	s = splnet();
251 	LIST_REMOVE(ifv, ifv_list);
252 	vlan_unconfig(ifp);
253 	splx(s);
254 
255 	if_detach(ifp);
256 	free(ifv, M_DEVBUF);
257 
258 	return (0);
259 }
260 
261 /*
262  * Configure a VLAN interface.  Must be called at splnet().
263  */
264 static int
265 vlan_config(struct ifvlan *ifv, struct ifnet *p)
266 {
267 	struct ifnet *ifp = &ifv->ifv_if;
268 	int error;
269 
270 	if (ifv->ifv_p != NULL)
271 		return (EBUSY);
272 
273 	switch (p->if_type) {
274 	case IFT_ETHER:
275 	    {
276 		struct ethercom *ec = (void *) p;
277 
278 		ifv->ifv_msw = &vlan_ether_multisw;
279 		ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
280 		ifv->ifv_mintu = ETHERMIN;
281 
282 		/*
283 		 * If the parent supports the VLAN_MTU capability,
284 		 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
285 		 * enable it.
286 		 */
287 		if (ec->ec_nvlans++ == 0 &&
288 		    (ec->ec_capabilities & ETHERCAP_VLAN_MTU) != 0) {
289 			/*
290 			 * Enable Tx/Rx of VLAN-sized frames.
291 			 */
292 			ec->ec_capenable |= ETHERCAP_VLAN_MTU;
293 			if (p->if_flags & IFF_UP) {
294 				struct ifreq ifr;
295 
296 				ifr.ifr_flags = p->if_flags;
297 				error = (*p->if_ioctl)(p, SIOCSIFFLAGS,
298 				    (void *) &ifr);
299 				if (error) {
300 					if (ec->ec_nvlans-- == 1)
301 						ec->ec_capenable &=
302 						    ~ETHERCAP_VLAN_MTU;
303 					return (error);
304 				}
305 			}
306 			ifv->ifv_mtufudge = 0;
307 		} else if ((ec->ec_capabilities & ETHERCAP_VLAN_MTU) == 0) {
308 			/*
309 			 * Fudge the MTU by the encapsulation size.  This
310 			 * makes us incompatible with strictly compliant
311 			 * 802.1Q implementations, but allows us to use
312 			 * the feature with other NetBSD implementations,
313 			 * which might still be useful.
314 			 */
315 			ifv->ifv_mtufudge = ifv->ifv_encaplen;
316 		}
317 
318 		/*
319 		 * If the parent interface can do hardware-assisted
320 		 * VLAN encapsulation, then propagate its hardware-
321 		 * assisted checksumming flags.
322 		 */
323 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING)
324 			ifp->if_capabilities = p->if_capabilities &
325 			    (IFCAP_CSUM_IPv4_Tx|IFCAP_CSUM_IPv4_Rx|
326 			     IFCAP_CSUM_TCPv4_Tx|IFCAP_CSUM_TCPv4_Rx|
327 			     IFCAP_CSUM_UDPv4_Tx|IFCAP_CSUM_UDPv4_Rx|
328 			     IFCAP_CSUM_TCPv6_Tx|IFCAP_CSUM_TCPv6_Rx|
329 			     IFCAP_CSUM_UDPv6_Tx|IFCAP_CSUM_UDPv6_Rx);
330 
331 		/*
332 		 * We inherit the parent's Ethernet address.
333 		 */
334 		ether_ifattach(ifp, CLLADDR(p->if_sadl));
335 		ifp->if_hdrlen = sizeof(struct ether_vlan_header); /* XXX? */
336 		break;
337 	    }
338 
339 	default:
340 		return (EPROTONOSUPPORT);
341 	}
342 
343 	ifv->ifv_p = p;
344 	ifv->ifv_if.if_mtu = p->if_mtu - ifv->ifv_mtufudge;
345 	ifv->ifv_if.if_flags = p->if_flags &
346 	    (IFF_UP | IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
347 
348 	/*
349 	 * Inherit the if_type from the parent.  This allows us
350 	 * to participate in bridges of that type.
351 	 */
352 	ifv->ifv_if.if_type = p->if_type;
353 
354 	return (0);
355 }
356 
357 /*
358  * Unconfigure a VLAN interface.  Must be called at splnet().
359  */
360 static void
361 vlan_unconfig(struct ifnet *ifp)
362 {
363 	struct ifvlan *ifv = ifp->if_softc;
364 
365 	if (ifv->ifv_p == NULL)
366 		return;
367 
368 	/*
369  	 * Since the interface is being unconfigured, we need to empty the
370 	 * list of multicast groups that we may have joined while we were
371 	 * alive and remove them from the parent's list also.
372 	 */
373 	(*ifv->ifv_msw->vmsw_purgemulti)(ifv);
374 
375 	/* Disconnect from parent. */
376 	switch (ifv->ifv_p->if_type) {
377 	case IFT_ETHER:
378 	    {
379 		struct ethercom *ec = (void *) ifv->ifv_p;
380 
381 		if (ec->ec_nvlans-- == 1) {
382 			/*
383 			 * Disable Tx/Rx of VLAN-sized frames.
384 			 */
385 			ec->ec_capenable &= ~ETHERCAP_VLAN_MTU;
386 			if (ifv->ifv_p->if_flags & IFF_UP) {
387 				struct ifreq ifr;
388 
389 				ifr.ifr_flags = ifv->ifv_p->if_flags;
390 				(void) (*ifv->ifv_p->if_ioctl)(ifv->ifv_p,
391 				    SIOCSIFFLAGS, (void *) &ifr);
392 			}
393 		}
394 
395 		ether_ifdetach(ifp);
396 		vlan_reset_linkname(ifp);
397 		break;
398 	    }
399 
400 #ifdef DIAGNOSTIC
401 	default:
402 		panic("vlan_unconfig: impossible");
403 #endif
404 	}
405 
406 	ifv->ifv_p = NULL;
407 	ifv->ifv_if.if_mtu = 0;
408 	ifv->ifv_flags = 0;
409 
410 	if_down(ifp);
411 	ifp->if_flags &= ~(IFF_UP|IFF_RUNNING);
412 	ifp->if_capabilities = 0;
413 }
414 
415 /*
416  * Called when a parent interface is detaching; destroy any VLAN
417  * configuration for the parent interface.
418  */
419 void
420 vlan_ifdetach(struct ifnet *p)
421 {
422 	struct ifvlan *ifv;
423 	int s;
424 
425 	s = splnet();
426 
427 	for (ifv = LIST_FIRST(&ifv_list); ifv != NULL;
428 	     ifv = LIST_NEXT(ifv, ifv_list)) {
429 		if (ifv->ifv_p == p)
430 			vlan_unconfig(&ifv->ifv_if);
431 	}
432 
433 	splx(s);
434 }
435 
436 static int
437 vlan_set_promisc(struct ifnet *ifp)
438 {
439 	struct ifvlan *ifv = ifp->if_softc;
440 	int error = 0;
441 
442 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
443 		if ((ifv->ifv_flags & IFVF_PROMISC) == 0) {
444 			error = ifpromisc(ifv->ifv_p, 1);
445 			if (error == 0)
446 				ifv->ifv_flags |= IFVF_PROMISC;
447 		}
448 	} else {
449 		if ((ifv->ifv_flags & IFVF_PROMISC) != 0) {
450 			error = ifpromisc(ifv->ifv_p, 0);
451 			if (error == 0)
452 				ifv->ifv_flags &= ~IFVF_PROMISC;
453 		}
454 	}
455 
456 	return (error);
457 }
458 
459 static int
460 vlan_ioctl(struct ifnet *ifp, u_long cmd, void *data)
461 {
462 	struct lwp *l = curlwp;	/* XXX */
463 	struct ifvlan *ifv = ifp->if_softc;
464 	struct ifaddr *ifa = (struct ifaddr *) data;
465 	struct ifreq *ifr = (struct ifreq *) data;
466 	struct ifnet *pr;
467 	struct vlanreq vlr;
468 	struct sockaddr *sa;
469 	int s, error = 0;
470 
471 	s = splnet();
472 
473 	switch (cmd) {
474 	case SIOCSIFADDR:
475 		if (ifv->ifv_p != NULL) {
476 			ifp->if_flags |= IFF_UP;
477 
478 			switch (ifa->ifa_addr->sa_family) {
479 #ifdef INET
480 			case AF_INET:
481 				arp_ifinit(ifp, ifa);
482 				break;
483 #endif
484 			default:
485 				break;
486 			}
487 		} else {
488 			error = EINVAL;
489 		}
490 		break;
491 
492 	case SIOCGIFADDR:
493 		sa = (struct sockaddr *)&ifr->ifr_data;
494 		memcpy(sa->sa_data, CLLADDR(ifp->if_sadl), ifp->if_addrlen);
495 		break;
496 
497 	case SIOCSIFMTU:
498 		if (ifv->ifv_p == NULL)
499 			error = EINVAL;
500 		else if (
501 		    ifr->ifr_mtu > (ifv->ifv_p->if_mtu - ifv->ifv_mtufudge) ||
502 		    ifr->ifr_mtu < (ifv->ifv_mintu - ifv->ifv_mtufudge))
503 			error = EINVAL;
504 		else if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
505 			error = 0;
506 		break;
507 
508 	case SIOCSETVLAN:
509 		if ((error = kauth_authorize_network(l->l_cred,
510 		    KAUTH_NETWORK_INTERFACE,
511 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
512 		    NULL)) != 0)
513 			break;
514 		if ((error = copyin(ifr->ifr_data, &vlr, sizeof(vlr))) != 0)
515 			break;
516 		if (vlr.vlr_parent[0] == '\0') {
517 			vlan_unconfig(ifp);
518 			break;
519 		}
520 		if (vlr.vlr_tag != EVL_VLANOFTAG(vlr.vlr_tag)) {
521 			error = EINVAL;		 /* check for valid tag */
522 			break;
523 		}
524 		if ((pr = ifunit(vlr.vlr_parent)) == 0) {
525 			error = ENOENT;
526 			break;
527 		}
528 		if ((error = vlan_config(ifv, pr)) != 0)
529 			break;
530 		ifv->ifv_tag = vlr.vlr_tag;
531 		ifp->if_flags |= IFF_RUNNING;
532 
533 		/* Update promiscuous mode, if necessary. */
534 		vlan_set_promisc(ifp);
535 		break;
536 
537 	case SIOCGETVLAN:
538 		memset(&vlr, 0, sizeof(vlr));
539 		if (ifv->ifv_p != NULL) {
540 			snprintf(vlr.vlr_parent, sizeof(vlr.vlr_parent), "%s",
541 			    ifv->ifv_p->if_xname);
542 			vlr.vlr_tag = ifv->ifv_tag;
543 		}
544 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
545 		break;
546 
547 	case SIOCSIFFLAGS:
548 		/*
549 		 * For promiscuous mode, we enable promiscuous mode on
550 		 * the parent if we need promiscuous on the VLAN interface.
551 		 */
552 		if (ifv->ifv_p != NULL)
553 			error = vlan_set_promisc(ifp);
554 		break;
555 
556 	case SIOCADDMULTI:
557 		error = (ifv->ifv_p != NULL) ?
558 		    (*ifv->ifv_msw->vmsw_addmulti)(ifv, ifr) : EINVAL;
559 		break;
560 
561 	case SIOCDELMULTI:
562 		error = (ifv->ifv_p != NULL) ?
563 		    (*ifv->ifv_msw->vmsw_delmulti)(ifv, ifr) : EINVAL;
564 		break;
565 
566 	default:
567 		error = EINVAL;
568 	}
569 
570 	splx(s);
571 
572 	return (error);
573 }
574 
575 static int
576 vlan_ether_addmulti(struct ifvlan *ifv, struct ifreq *ifr)
577 {
578 	const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr);
579 	struct vlan_mc_entry *mc;
580 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
581 	int error;
582 
583 	if (sa->sa_len > sizeof(struct sockaddr_storage))
584 		return (EINVAL);
585 
586 	error = ether_addmulti(sa, &ifv->ifv_ec);
587 	if (error != ENETRESET)
588 		return (error);
589 
590 	/*
591 	 * This is new multicast address.  We have to tell parent
592 	 * about it.  Also, remember this multicast address so that
593 	 * we can delete them on unconfigure.
594 	 */
595 	MALLOC(mc, struct vlan_mc_entry *, sizeof(struct vlan_mc_entry),
596 	    M_DEVBUF, M_NOWAIT);
597 	if (mc == NULL) {
598 		error = ENOMEM;
599 		goto alloc_failed;
600 	}
601 
602 	/*
603 	 * As ether_addmulti() returns ENETRESET, following two
604 	 * statement shouldn't fail.
605 	 */
606 	(void)ether_multiaddr(sa, addrlo, addrhi);
607 	ETHER_LOOKUP_MULTI(addrlo, addrhi, &ifv->ifv_ec, mc->mc_enm);
608 	memcpy(&mc->mc_addr, sa, sa->sa_len);
609 	LIST_INSERT_HEAD(&ifv->ifv_mc_listhead, mc, mc_entries);
610 
611 	error = (*ifv->ifv_p->if_ioctl)(ifv->ifv_p, SIOCADDMULTI,
612 	    (void *)ifr);
613 	if (error != 0)
614 		goto ioctl_failed;
615 	return (error);
616 
617  ioctl_failed:
618 	LIST_REMOVE(mc, mc_entries);
619 	FREE(mc, M_DEVBUF);
620  alloc_failed:
621 	(void)ether_delmulti(sa, &ifv->ifv_ec);
622 	return (error);
623 }
624 
625 static int
626 vlan_ether_delmulti(struct ifvlan *ifv, struct ifreq *ifr)
627 {
628 	const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr);
629 	struct ether_multi *enm;
630 	struct vlan_mc_entry *mc;
631 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
632 	int error;
633 
634 	/*
635 	 * Find a key to lookup vlan_mc_entry.  We have to do this
636 	 * before calling ether_delmulti for obvious reason.
637 	 */
638 	if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0)
639 		return (error);
640 	ETHER_LOOKUP_MULTI(addrlo, addrhi, &ifv->ifv_ec, enm);
641 
642 	error = ether_delmulti(sa, &ifv->ifv_ec);
643 	if (error != ENETRESET)
644 		return (error);
645 
646 	/* We no longer use this multicast address.  Tell parent so. */
647 	error = (*ifv->ifv_p->if_ioctl)(ifv->ifv_p, SIOCDELMULTI,
648 	    (void *)ifr);
649 	if (error == 0) {
650 		/* And forget about this address. */
651 		for (mc = LIST_FIRST(&ifv->ifv_mc_listhead); mc != NULL;
652 		    mc = LIST_NEXT(mc, mc_entries)) {
653 			if (mc->mc_enm == enm) {
654 				LIST_REMOVE(mc, mc_entries);
655 				FREE(mc, M_DEVBUF);
656 				break;
657 			}
658 		}
659 		KASSERT(mc != NULL);
660 	} else
661 		(void)ether_addmulti(sa, &ifv->ifv_ec);
662 	return (error);
663 }
664 
665 /*
666  * Delete any multicast address we have asked to add from parent
667  * interface.  Called when the vlan is being unconfigured.
668  */
669 static void
670 vlan_ether_purgemulti(struct ifvlan *ifv)
671 {
672 	struct ifnet *ifp = ifv->ifv_p;		/* Parent. */
673 	struct vlan_mc_entry *mc;
674 	union {
675 		struct ifreq ifreq;
676 		struct {
677 			char ifr_name[IFNAMSIZ];
678 			struct sockaddr_storage ifr_ss;
679 		} ifreq_storage;
680 	} ifreq;
681 	struct ifreq *ifr = &ifreq.ifreq;
682 
683 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
684 	while ((mc = LIST_FIRST(&ifv->ifv_mc_listhead)) != NULL) {
685 		ifreq_setaddr(SIOCDELMULTI, ifr,
686 		    (const struct sockaddr *)&mc->mc_addr);
687 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, (void *)ifr);
688 		LIST_REMOVE(mc, mc_entries);
689 		FREE(mc, M_DEVBUF);
690 	}
691 }
692 
693 static void
694 vlan_start(struct ifnet *ifp)
695 {
696 	struct ifvlan *ifv = ifp->if_softc;
697 	struct ifnet *p = ifv->ifv_p;
698 	struct ethercom *ec = (void *) ifv->ifv_p;
699 	struct mbuf *m;
700 	int error;
701 	ALTQ_DECL(struct altq_pktattr pktattr;)
702 
703 	ifp->if_flags |= IFF_OACTIVE;
704 
705 	for (;;) {
706 		IFQ_DEQUEUE(&ifp->if_snd, m);
707 		if (m == NULL)
708 			break;
709 
710 #ifdef ALTQ
711 		/*
712 		 * If ALTQ is enabled on the parent interface, do
713 		 * classification; the queueing discipline might
714 		 * not require classification, but might require
715 		 * the address family/header pointer in the pktattr.
716 		 */
717 		if (ALTQ_IS_ENABLED(&p->if_snd)) {
718 			switch (p->if_type) {
719 			case IFT_ETHER:
720 				altq_etherclassify(&p->if_snd, m, &pktattr);
721 				break;
722 #ifdef DIAGNOSTIC
723 			default:
724 				panic("vlan_start: impossible (altq)");
725 #endif
726 			}
727 		}
728 #endif /* ALTQ */
729 
730 #if NBPFILTER > 0
731 		if (ifp->if_bpf)
732 			bpf_mtap(ifp->if_bpf, m);
733 #endif
734 		/*
735 		 * If the parent can insert the tag itself, just mark
736 		 * the tag in the mbuf header.
737 		 */
738 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
739 			struct m_tag *mtag;
740 
741 			mtag = m_tag_get(PACKET_TAG_VLAN, sizeof(u_int),
742 			    M_NOWAIT);
743 			if (mtag == NULL) {
744 				ifp->if_oerrors++;
745 				m_freem(m);
746 				continue;
747 			}
748 			*(u_int *)(mtag + 1) = ifv->ifv_tag;
749 			m_tag_prepend(m, mtag);
750 		} else {
751 			/*
752 			 * insert the tag ourselves
753 			 */
754 			M_PREPEND(m, ifv->ifv_encaplen, M_DONTWAIT);
755 			if (m == NULL) {
756 				printf("%s: unable to prepend encap header",
757 				    ifv->ifv_p->if_xname);
758 				ifp->if_oerrors++;
759 				continue;
760 			}
761 
762 			switch (p->if_type) {
763 			case IFT_ETHER:
764 			    {
765 				struct ether_vlan_header *evl;
766 
767 				if (m->m_len < sizeof(struct ether_vlan_header))
768 					m = m_pullup(m,
769 					    sizeof(struct ether_vlan_header));
770 				if (m == NULL) {
771 					printf("%s: unable to pullup encap "
772 					    "header", ifv->ifv_p->if_xname);
773 					ifp->if_oerrors++;
774 					continue;
775 				}
776 
777 				/*
778 				 * Transform the Ethernet header into an
779 				 * Ethernet header with 802.1Q encapsulation.
780 				 */
781 				memmove(mtod(m, void *),
782 				    mtod(m, char *) + ifv->ifv_encaplen,
783 				    sizeof(struct ether_header));
784 				evl = mtod(m, struct ether_vlan_header *);
785 				evl->evl_proto = evl->evl_encap_proto;
786 				evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
787 				evl->evl_tag = htons(ifv->ifv_tag);
788 
789 				/*
790 				 * To cater for VLAN-aware layer 2 ethernet
791 				 * switches which may need to strip the tag
792 				 * before forwarding the packet, make sure
793 				 * the packet+tag is at least 68 bytes long.
794 				 * This is necessary because our parent will
795 				 * only pad to 64 bytes (ETHER_MIN_LEN) and
796 				 * some switches will not pad by themselves
797 				 * after deleting a tag.
798 				 */
799 				if (m->m_pkthdr.len <
800 				    (ETHER_MIN_LEN + ETHER_VLAN_ENCAP_LEN)) {
801 					m_copyback(m, m->m_pkthdr.len,
802 					    (ETHER_MIN_LEN +
803 					     ETHER_VLAN_ENCAP_LEN) -
804 					     m->m_pkthdr.len,
805 					    vlan_zero_pad_buff);
806 				}
807 				break;
808 			    }
809 
810 #ifdef DIAGNOSTIC
811 			default:
812 				panic("vlan_start: impossible");
813 #endif
814 			}
815 		}
816 
817 		/*
818 		 * Send it, precisely as the parent's output routine
819 		 * would have.  We are already running at splnet.
820 		 */
821 		IFQ_ENQUEUE(&p->if_snd, m, &pktattr, error);
822 		if (error) {
823 			/* mbuf is already freed */
824 			ifp->if_oerrors++;
825 			continue;
826 		}
827 
828 		ifp->if_opackets++;
829 		if ((p->if_flags & (IFF_RUNNING|IFF_OACTIVE)) == IFF_RUNNING)
830 			(*p->if_start)(p);
831 	}
832 
833 	ifp->if_flags &= ~IFF_OACTIVE;
834 }
835 
836 /*
837  * Given an Ethernet frame, find a valid vlan interface corresponding to the
838  * given source interface and tag, then run the real packet through the
839  * parent's input routine.
840  */
841 void
842 vlan_input(struct ifnet *ifp, struct mbuf *m)
843 {
844 	struct ifvlan *ifv;
845 	u_int tag;
846 	struct m_tag *mtag;
847 
848 	mtag = m_tag_find(m, PACKET_TAG_VLAN, NULL);
849 	if (mtag != NULL) {
850 		/* m contains a normal ethernet frame, the tag is in mtag */
851 		tag = EVL_VLANOFTAG(*(u_int *)(mtag + 1));
852 		m_tag_delete(m, mtag);
853 	} else {
854 		switch (ifp->if_type) {
855 		case IFT_ETHER:
856 		    {
857 			struct ether_vlan_header *evl;
858 
859 			if (m->m_len < sizeof(struct ether_vlan_header) &&
860 			    (m = m_pullup(m,
861 			     sizeof(struct ether_vlan_header))) == NULL) {
862 				printf("%s: no memory for VLAN header, "
863 				    "dropping packet.\n", ifp->if_xname);
864 				return;
865 			}
866 			evl = mtod(m, struct ether_vlan_header *);
867 			KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN);
868 
869 			tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
870 
871 			/*
872 			 * Restore the original ethertype.  We'll remove
873 			 * the encapsulation after we've found the vlan
874 			 * interface corresponding to the tag.
875 			 */
876 			evl->evl_encap_proto = evl->evl_proto;
877 			break;
878 		    }
879 
880 		default:
881 			tag = (u_int) -1;	/* XXX GCC */
882 #ifdef DIAGNOSTIC
883 			panic("vlan_input: impossible");
884 #endif
885 		}
886 	}
887 
888 	for (ifv = LIST_FIRST(&ifv_list); ifv != NULL;
889 	    ifv = LIST_NEXT(ifv, ifv_list))
890 		if (ifp == ifv->ifv_p && tag == ifv->ifv_tag)
891 			break;
892 
893 	if (ifv == NULL ||
894 	    (ifv->ifv_if.if_flags & (IFF_UP|IFF_RUNNING)) !=
895 	     (IFF_UP|IFF_RUNNING)) {
896 		m_freem(m);
897 		ifp->if_noproto++;
898 		return;
899 	}
900 
901 	/*
902 	 * Now, remove the encapsulation header.  The original
903 	 * header has already been fixed up above.
904 	 */
905 	if (mtag == NULL) {
906 		memmove(mtod(m, char *) + ifv->ifv_encaplen,
907 		    mtod(m, void *), sizeof(struct ether_header));
908 		m_adj(m, ifv->ifv_encaplen);
909 	}
910 
911 	m->m_pkthdr.rcvif = &ifv->ifv_if;
912 	ifv->ifv_if.if_ipackets++;
913 
914 #if NBPFILTER > 0
915 	if (ifv->ifv_if.if_bpf)
916 		bpf_mtap(ifv->ifv_if.if_bpf, m);
917 #endif
918 
919 	/* Pass it back through the parent's input routine. */
920 	(*ifp->if_input)(&ifv->ifv_if, m);
921 }
922