xref: /netbsd-src/sys/net/if_vlan.c (revision 796c32c94f6e154afc9de0f63da35c91bb739b45)
1 /*	$NetBSD: if_vlan.c,v 1.113 2017/11/27 01:34:06 jmcneill 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 1.113 2017/11/27 01:34:06 jmcneill Exp $");
82 
83 #ifdef _KERNEL_OPT
84 #include "opt_inet.h"
85 #endif
86 
87 #include <sys/param.h>
88 #include <sys/systm.h>
89 #include <sys/kernel.h>
90 #include <sys/mbuf.h>
91 #include <sys/queue.h>
92 #include <sys/socket.h>
93 #include <sys/sockio.h>
94 #include <sys/systm.h>
95 #include <sys/proc.h>
96 #include <sys/kauth.h>
97 #include <sys/mutex.h>
98 #include <sys/kmem.h>
99 #include <sys/cpu.h>
100 #include <sys/pserialize.h>
101 #include <sys/psref.h>
102 #include <sys/pslist.h>
103 #include <sys/atomic.h>
104 #include <sys/device.h>
105 #include <sys/module.h>
106 
107 #include <net/bpf.h>
108 #include <net/if.h>
109 #include <net/if_dl.h>
110 #include <net/if_types.h>
111 #include <net/if_ether.h>
112 #include <net/if_vlanvar.h>
113 
114 #ifdef INET
115 #include <netinet/in.h>
116 #include <netinet/if_inarp.h>
117 #endif
118 #ifdef INET6
119 #include <netinet6/in6_ifattach.h>
120 #include <netinet6/in6_var.h>
121 #endif
122 
123 #include "ioconf.h"
124 
125 struct vlan_mc_entry {
126 	LIST_ENTRY(vlan_mc_entry)	mc_entries;
127 	/*
128 	 * A key to identify this entry.  The mc_addr below can't be
129 	 * used since multiple sockaddr may mapped into the same
130 	 * ether_multi (e.g., AF_UNSPEC).
131 	 */
132 	union {
133 		struct ether_multi	*mcu_enm;
134 	} mc_u;
135 	struct sockaddr_storage		mc_addr;
136 };
137 
138 #define	mc_enm		mc_u.mcu_enm
139 
140 
141 struct ifvlan_linkmib {
142 	struct ifvlan *ifvm_ifvlan;
143 	const struct vlan_multisw *ifvm_msw;
144 	int	ifvm_encaplen;	/* encapsulation length */
145 	int	ifvm_mtufudge;	/* MTU fudged by this much */
146 	int	ifvm_mintu;	/* min transmission unit */
147 	uint16_t ifvm_proto;	/* encapsulation ethertype */
148 	uint16_t ifvm_tag;	/* tag to apply on packets */
149 	struct ifnet *ifvm_p;		/* parent interface of this vlan */
150 
151 	struct psref_target ifvm_psref;
152 };
153 
154 struct ifvlan {
155 	union {
156 		struct ethercom ifvu_ec;
157 	} ifv_u;
158 	struct ifvlan_linkmib *ifv_mib;	/*
159 					 * reader must use vlan_getref_linkmib()
160 					 * instead of direct dereference
161 					 */
162 	kmutex_t ifv_lock;		/* writer lock for ifv_mib */
163 
164 	LIST_HEAD(__vlan_mchead, vlan_mc_entry) ifv_mc_listhead;
165 	LIST_ENTRY(ifvlan) ifv_list;
166 	struct pslist_entry ifv_hash;
167 	int ifv_flags;
168 };
169 
170 #define	IFVF_PROMISC	0x01		/* promiscuous mode enabled */
171 
172 #define	ifv_ec		ifv_u.ifvu_ec
173 
174 #define	ifv_if		ifv_ec.ec_if
175 
176 #define	ifv_msw		ifv_mib.ifvm_msw
177 #define	ifv_encaplen	ifv_mib.ifvm_encaplen
178 #define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
179 #define	ifv_mintu	ifv_mib.ifvm_mintu
180 #define	ifv_tag		ifv_mib.ifvm_tag
181 
182 struct vlan_multisw {
183 	int	(*vmsw_addmulti)(struct ifvlan *, struct ifreq *);
184 	int	(*vmsw_delmulti)(struct ifvlan *, struct ifreq *);
185 	void	(*vmsw_purgemulti)(struct ifvlan *);
186 };
187 
188 static int	vlan_ether_addmulti(struct ifvlan *, struct ifreq *);
189 static int	vlan_ether_delmulti(struct ifvlan *, struct ifreq *);
190 static void	vlan_ether_purgemulti(struct ifvlan *);
191 
192 const struct vlan_multisw vlan_ether_multisw = {
193 	vlan_ether_addmulti,
194 	vlan_ether_delmulti,
195 	vlan_ether_purgemulti,
196 };
197 
198 static int	vlan_clone_create(struct if_clone *, int);
199 static int	vlan_clone_destroy(struct ifnet *);
200 static int	vlan_config(struct ifvlan *, struct ifnet *,
201     uint16_t);
202 static int	vlan_ioctl(struct ifnet *, u_long, void *);
203 static void	vlan_start(struct ifnet *);
204 static int	vlan_transmit(struct ifnet *, struct mbuf *);
205 static void	vlan_unconfig(struct ifnet *);
206 static int	vlan_unconfig_locked(struct ifvlan *,
207     struct ifvlan_linkmib *);
208 static void	vlan_hash_init(void);
209 static int	vlan_hash_fini(void);
210 static int	vlan_tag_hash(uint16_t, u_long);
211 static struct ifvlan_linkmib*	vlan_getref_linkmib(struct ifvlan *,
212     struct psref *);
213 static void	vlan_putref_linkmib(struct ifvlan_linkmib *,
214     struct psref *);
215 static void	vlan_linkmib_update(struct ifvlan *,
216     struct ifvlan_linkmib *);
217 static struct ifvlan_linkmib*	vlan_lookup_tag_psref(struct ifnet *,
218     uint16_t, struct psref *);
219 
220 LIST_HEAD(vlan_ifvlist, ifvlan);
221 static struct {
222 	kmutex_t lock;
223 	struct vlan_ifvlist list;
224 } ifv_list __cacheline_aligned;
225 
226 
227 #if !defined(VLAN_TAG_HASH_SIZE)
228 #define VLAN_TAG_HASH_SIZE 32
229 #endif
230 static struct {
231 	kmutex_t lock;
232 	struct pslist_head *lists;
233 	u_long mask;
234 } ifv_hash __cacheline_aligned = {
235 	.lists = NULL,
236 	.mask = 0,
237 };
238 
239 pserialize_t vlan_psz __read_mostly;
240 static struct psref_class *ifvm_psref_class __read_mostly;
241 
242 struct if_clone vlan_cloner =
243     IF_CLONE_INITIALIZER("vlan", vlan_clone_create, vlan_clone_destroy);
244 
245 /* Used to pad ethernet frames with < ETHER_MIN_LEN bytes */
246 static char vlan_zero_pad_buff[ETHER_MIN_LEN];
247 
248 static inline int
249 vlan_safe_ifpromisc(struct ifnet *ifp, int pswitch)
250 {
251 	int e;
252 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
253 	e = ifpromisc(ifp, pswitch);
254 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
255 	return e;
256 }
257 
258 void
259 vlanattach(int n)
260 {
261 
262 	/*
263 	 * Nothing to do here, initialization is handled by the
264 	 * module initialization code in vlaninit() below).
265 	 */
266 }
267 
268 static void
269 vlaninit(void)
270 {
271 	mutex_init(&ifv_list.lock, MUTEX_DEFAULT, IPL_NONE);
272 	LIST_INIT(&ifv_list.list);
273 
274 	mutex_init(&ifv_hash.lock, MUTEX_DEFAULT, IPL_NONE);
275 	vlan_psz = pserialize_create();
276 	ifvm_psref_class = psref_class_create("vlanlinkmib", IPL_SOFTNET);
277 	if_clone_attach(&vlan_cloner);
278 
279 	vlan_hash_init();
280 }
281 
282 static int
283 vlandetach(void)
284 {
285 	int error = 0;
286 
287 	mutex_enter(&ifv_list.lock);
288 	if (!LIST_EMPTY(&ifv_list.list)) {
289 		mutex_exit(&ifv_list.lock);
290 		return EBUSY;
291 	}
292 	mutex_exit(&ifv_list.lock);
293 
294 	error = vlan_hash_fini();
295 	if (error != 0)
296 		return error;
297 
298 	if_clone_detach(&vlan_cloner);
299 	psref_class_destroy(ifvm_psref_class);
300 	pserialize_destroy(vlan_psz);
301 	mutex_destroy(&ifv_hash.lock);
302 	mutex_destroy(&ifv_list.lock);
303 
304 	return 0;
305 }
306 
307 static void
308 vlan_reset_linkname(struct ifnet *ifp)
309 {
310 
311 	/*
312 	 * We start out with a "802.1Q VLAN" type and zero-length
313 	 * addresses.  When we attach to a parent interface, we
314 	 * inherit its type, address length, address, and data link
315 	 * type.
316 	 */
317 
318 	ifp->if_type = IFT_L2VLAN;
319 	ifp->if_addrlen = 0;
320 	ifp->if_dlt = DLT_NULL;
321 	if_alloc_sadl(ifp);
322 }
323 
324 static int
325 vlan_clone_create(struct if_clone *ifc, int unit)
326 {
327 	struct ifvlan *ifv;
328 	struct ifnet *ifp;
329 	struct ifvlan_linkmib *mib;
330 	int rv;
331 
332 	ifv = malloc(sizeof(struct ifvlan), M_DEVBUF, M_WAITOK|M_ZERO);
333 	mib = kmem_zalloc(sizeof(struct ifvlan_linkmib), KM_SLEEP);
334 	ifp = &ifv->ifv_if;
335 	LIST_INIT(&ifv->ifv_mc_listhead);
336 
337 	mib->ifvm_ifvlan = ifv;
338 	mib->ifvm_p = NULL;
339 	psref_target_init(&mib->ifvm_psref, ifvm_psref_class);
340 
341 	mutex_init(&ifv->ifv_lock, MUTEX_DEFAULT, IPL_NONE);
342 	ifv->ifv_mib = mib;
343 
344 	mutex_enter(&ifv_list.lock);
345 	LIST_INSERT_HEAD(&ifv_list.list, ifv, ifv_list);
346 	mutex_exit(&ifv_list.lock);
347 
348 	if_initname(ifp, ifc->ifc_name, unit);
349 	ifp->if_softc = ifv;
350 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
351 	ifp->if_extflags = IFEF_MPSAFE | IFEF_NO_LINK_STATE_CHANGE;
352 	ifp->if_start = vlan_start;
353 	ifp->if_transmit = vlan_transmit;
354 	ifp->if_ioctl = vlan_ioctl;
355 	IFQ_SET_READY(&ifp->if_snd);
356 
357 	rv = if_initialize(ifp);
358 	if (rv != 0) {
359 		aprint_error("%s: if_initialize failed(%d)\n", ifp->if_xname,
360 		    rv);
361 		goto fail;
362 	}
363 
364 	vlan_reset_linkname(ifp);
365 	if_register(ifp);
366 	return 0;
367 
368 fail:
369 	mutex_enter(&ifv_list.lock);
370 	LIST_REMOVE(ifv, ifv_list);
371 	mutex_exit(&ifv_list.lock);
372 
373 	mutex_destroy(&ifv->ifv_lock);
374 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
375 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
376 	free(ifv, M_DEVBUF);
377 
378 	return rv;
379 }
380 
381 static int
382 vlan_clone_destroy(struct ifnet *ifp)
383 {
384 	struct ifvlan *ifv = ifp->if_softc;
385 
386 	mutex_enter(&ifv_list.lock);
387 	LIST_REMOVE(ifv, ifv_list);
388 	mutex_exit(&ifv_list.lock);
389 
390 	vlan_unconfig(ifp);
391 	if_detach(ifp);
392 
393 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
394 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
395 	mutex_destroy(&ifv->ifv_lock);
396 	free(ifv, M_DEVBUF);
397 
398 	return (0);
399 }
400 
401 /*
402  * Configure a VLAN interface.
403  */
404 static int
405 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
406 {
407 	struct ifnet *ifp = &ifv->ifv_if;
408 	struct ifvlan_linkmib *nmib = NULL;
409 	struct ifvlan_linkmib *omib = NULL;
410 	struct ifvlan_linkmib *checkmib = NULL;
411 	struct psref_target *nmib_psref = NULL;
412 	uint16_t vid = EVL_VLANOFTAG(tag);
413 	int error = 0;
414 	int idx;
415 	bool omib_cleanup = false;
416 	struct psref psref;
417 
418 	/* VLAN ID 0 and 4095 are reserved in the spec */
419 	if ((vid == 0) || (vid == 0xfff))
420 		return EINVAL;
421 
422 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
423 
424 	mutex_enter(&ifv->ifv_lock);
425 	omib = ifv->ifv_mib;
426 
427 	if (omib->ifvm_p != NULL) {
428 		error = EBUSY;
429 		goto done;
430 	}
431 
432 	/* Duplicate check */
433 	checkmib = vlan_lookup_tag_psref(p, vid, &psref);
434 	if (checkmib != NULL) {
435 		vlan_putref_linkmib(checkmib, &psref);
436 		error = EEXIST;
437 		goto done;
438 	}
439 
440 	*nmib = *omib;
441 	nmib_psref = &nmib->ifvm_psref;
442 
443 	psref_target_init(nmib_psref, ifvm_psref_class);
444 
445 	switch (p->if_type) {
446 	case IFT_ETHER:
447 	    {
448 		struct ethercom *ec = (void *) p;
449 		nmib->ifvm_msw = &vlan_ether_multisw;
450 		nmib->ifvm_encaplen = ETHER_VLAN_ENCAP_LEN;
451 		nmib->ifvm_mintu = ETHERMIN;
452 
453 		if (ec->ec_nvlans++ == 0) {
454 			if ((error = ether_enable_vlan_mtu(p)) >= 0) {
455 				if (error) {
456 					ec->ec_nvlans--;
457 					goto done;
458 				}
459 				nmib->ifvm_mtufudge = 0;
460 			} else {
461 				/*
462 				 * Fudge the MTU by the encapsulation size. This
463 				 * makes us incompatible with strictly compliant
464 				 * 802.1Q implementations, but allows us to use
465 				 * the feature with other NetBSD
466 				 * implementations, which might still be useful.
467 				 */
468 				nmib->ifvm_mtufudge = nmib->ifvm_encaplen;
469 			}
470 			error = 0;
471 		}
472 
473 		/*
474 		 * If the parent interface can do hardware-assisted
475 		 * VLAN encapsulation, then propagate its hardware-
476 		 * assisted checksumming flags and tcp segmentation
477 		 * offload.
478 		 */
479 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
480 		        ec->ec_capenable |= ETHERCAP_VLAN_HWTAGGING;
481 			ifp->if_capabilities = p->if_capabilities &
482 			    (IFCAP_TSOv4 | IFCAP_TSOv6 |
483 			     IFCAP_CSUM_IPv4_Tx|IFCAP_CSUM_IPv4_Rx|
484 			     IFCAP_CSUM_TCPv4_Tx|IFCAP_CSUM_TCPv4_Rx|
485 			     IFCAP_CSUM_UDPv4_Tx|IFCAP_CSUM_UDPv4_Rx|
486 			     IFCAP_CSUM_TCPv6_Tx|IFCAP_CSUM_TCPv6_Rx|
487 			     IFCAP_CSUM_UDPv6_Tx|IFCAP_CSUM_UDPv6_Rx);
488                 }
489 		/*
490 		 * We inherit the parent's Ethernet address.
491 		 */
492 		ether_ifattach(ifp, CLLADDR(p->if_sadl));
493 		ifp->if_hdrlen = sizeof(struct ether_vlan_header); /* XXX? */
494 		break;
495 	    }
496 
497 	default:
498 		error = EPROTONOSUPPORT;
499 		goto done;
500 	}
501 
502 	nmib->ifvm_p = p;
503 	nmib->ifvm_tag = vid;
504 	ifv->ifv_if.if_mtu = p->if_mtu - nmib->ifvm_mtufudge;
505 	ifv->ifv_if.if_flags = p->if_flags &
506 	    (IFF_UP | IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
507 
508 	/*
509 	 * Inherit the if_type from the parent.  This allows us
510 	 * to participate in bridges of that type.
511 	 */
512 	ifv->ifv_if.if_type = p->if_type;
513 
514 	PSLIST_ENTRY_INIT(ifv, ifv_hash);
515 	idx = vlan_tag_hash(vid, ifv_hash.mask);
516 
517 	mutex_enter(&ifv_hash.lock);
518 	PSLIST_WRITER_INSERT_HEAD(&ifv_hash.lists[idx], ifv, ifv_hash);
519 	mutex_exit(&ifv_hash.lock);
520 
521 	vlan_linkmib_update(ifv, nmib);
522 	nmib = NULL;
523 	nmib_psref = NULL;
524 	omib_cleanup = true;
525 
526 done:
527 	mutex_exit(&ifv->ifv_lock);
528 
529 	if (nmib_psref)
530 		psref_target_destroy(nmib_psref, ifvm_psref_class);
531 
532 	if (nmib)
533 		kmem_free(nmib, sizeof(*nmib));
534 
535 	if (omib_cleanup)
536 		kmem_free(omib, sizeof(*omib));
537 
538 	return error;
539 }
540 
541 /*
542  * Unconfigure a VLAN interface.
543  */
544 static void
545 vlan_unconfig(struct ifnet *ifp)
546 {
547 	struct ifvlan *ifv = ifp->if_softc;
548 	struct ifvlan_linkmib *nmib = NULL;
549 	int error;
550 
551 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
552 
553 	mutex_enter(&ifv->ifv_lock);
554 	error = vlan_unconfig_locked(ifv, nmib);
555 	mutex_exit(&ifv->ifv_lock);
556 
557 	if (error)
558 		kmem_free(nmib, sizeof(*nmib));
559 }
560 static int
561 vlan_unconfig_locked(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
562 {
563 	struct ifnet *p;
564 	struct ifnet *ifp = &ifv->ifv_if;
565 	struct psref_target *nmib_psref = NULL;
566 	struct ifvlan_linkmib *omib;
567 	int error = 0;
568 
569 	KASSERT(mutex_owned(&ifv->ifv_lock));
570 
571 	omib = ifv->ifv_mib;
572 	p = omib->ifvm_p;
573 
574 	if (p == NULL) {
575 		error = -1;
576 		goto done;
577 	}
578 
579 	*nmib = *omib;
580 	nmib_psref = &nmib->ifvm_psref;
581 	psref_target_init(nmib_psref, ifvm_psref_class);
582 
583 	/*
584  	 * Since the interface is being unconfigured, we need to empty the
585 	 * list of multicast groups that we may have joined while we were
586 	 * alive and remove them from the parent's list also.
587 	 */
588 	(*nmib->ifvm_msw->vmsw_purgemulti)(ifv);
589 
590 	/* Disconnect from parent. */
591 	switch (p->if_type) {
592 	case IFT_ETHER:
593 	    {
594 		struct ethercom *ec = (void *)p;
595 		if (--ec->ec_nvlans == 0)
596 			(void)ether_disable_vlan_mtu(p);
597 
598 		ether_ifdetach(ifp);
599 		/* Restore vlan_ioctl overwritten by ether_ifdetach */
600 		ifp->if_ioctl = vlan_ioctl;
601 		vlan_reset_linkname(ifp);
602 		break;
603 	    }
604 
605 #ifdef DIAGNOSTIC
606 	default:
607 		panic("vlan_unconfig: impossible");
608 #endif
609 	}
610 
611 	nmib->ifvm_p = NULL;
612 	ifv->ifv_if.if_mtu = 0;
613 	ifv->ifv_flags = 0;
614 
615 	mutex_enter(&ifv_hash.lock);
616 	PSLIST_WRITER_REMOVE(ifv, ifv_hash);
617 	pserialize_perform(vlan_psz);
618 	mutex_exit(&ifv_hash.lock);
619 	PSLIST_ENTRY_DESTROY(ifv, ifv_hash);
620 
621 	vlan_linkmib_update(ifv, nmib);
622 
623 	mutex_exit(&ifv->ifv_lock);
624 
625 	nmib_psref = NULL;
626 	kmem_free(omib, sizeof(*omib));
627 
628 #ifdef INET6
629 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
630 	/* To delete v6 link local addresses */
631 	if (in6_present)
632 		in6_ifdetach(ifp);
633 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
634 #endif
635 
636 	if ((ifp->if_flags & IFF_PROMISC) != 0)
637 		vlan_safe_ifpromisc(ifp, 0);
638 	if_down(ifp);
639 	ifp->if_flags &= ~(IFF_UP|IFF_RUNNING);
640 	ifp->if_capabilities = 0;
641 	mutex_enter(&ifv->ifv_lock);
642 done:
643 
644 	if (nmib_psref)
645 		psref_target_destroy(nmib_psref, ifvm_psref_class);
646 
647 	return error;
648 }
649 
650 static void
651 vlan_hash_init(void)
652 {
653 
654 	ifv_hash.lists = hashinit(VLAN_TAG_HASH_SIZE, HASH_PSLIST, true,
655 	    &ifv_hash.mask);
656 }
657 
658 static int
659 vlan_hash_fini(void)
660 {
661 	int i;
662 
663 	mutex_enter(&ifv_hash.lock);
664 
665 	for (i = 0; i < ifv_hash.mask + 1; i++) {
666 		if (PSLIST_WRITER_FIRST(&ifv_hash.lists[i], struct ifvlan,
667 		    ifv_hash) != NULL) {
668 			mutex_exit(&ifv_hash.lock);
669 			return EBUSY;
670 		}
671 	}
672 
673 	for (i = 0; i < ifv_hash.mask + 1; i++)
674 		PSLIST_DESTROY(&ifv_hash.lists[i]);
675 
676 	mutex_exit(&ifv_hash.lock);
677 
678 	hashdone(ifv_hash.lists, HASH_PSLIST, ifv_hash.mask);
679 
680 	ifv_hash.lists = NULL;
681 	ifv_hash.mask = 0;
682 
683 	return 0;
684 }
685 
686 static int
687 vlan_tag_hash(uint16_t tag, u_long mask)
688 {
689 	uint32_t hash;
690 
691 	hash = (tag >> 8) ^ tag;
692 	hash = (hash >> 2) ^ hash;
693 
694 	return hash & mask;
695 }
696 
697 static struct ifvlan_linkmib *
698 vlan_getref_linkmib(struct ifvlan *sc, struct psref *psref)
699 {
700 	struct ifvlan_linkmib *mib;
701 	int s;
702 
703 	s = pserialize_read_enter();
704 	mib = sc->ifv_mib;
705 	if (mib == NULL) {
706 		pserialize_read_exit(s);
707 		return NULL;
708 	}
709 	membar_datadep_consumer();
710 	psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
711 	pserialize_read_exit(s);
712 
713 	return mib;
714 }
715 
716 static void
717 vlan_putref_linkmib(struct ifvlan_linkmib *mib, struct psref *psref)
718 {
719 	if (mib == NULL)
720 		return;
721 	psref_release(psref, &mib->ifvm_psref, ifvm_psref_class);
722 }
723 
724 static struct ifvlan_linkmib *
725 vlan_lookup_tag_psref(struct ifnet *ifp, uint16_t tag, struct psref *psref)
726 {
727 	int idx;
728 	int s;
729 	struct ifvlan *sc;
730 
731 	idx = vlan_tag_hash(tag, ifv_hash.mask);
732 
733 	s = pserialize_read_enter();
734 	PSLIST_READER_FOREACH(sc, &ifv_hash.lists[idx], struct ifvlan,
735 	    ifv_hash) {
736 		struct ifvlan_linkmib *mib = sc->ifv_mib;
737 		if (mib == NULL)
738 			continue;
739 		if (mib->ifvm_tag != tag)
740 			continue;
741 		if (mib->ifvm_p != ifp)
742 			continue;
743 
744 		psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
745 		pserialize_read_exit(s);
746 		return mib;
747 	}
748 	pserialize_read_exit(s);
749 	return NULL;
750 }
751 
752 static void
753 vlan_linkmib_update(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
754 {
755 	struct ifvlan_linkmib *omib = ifv->ifv_mib;
756 
757 	KASSERT(mutex_owned(&ifv->ifv_lock));
758 
759 	membar_producer();
760 	ifv->ifv_mib = nmib;
761 
762 	pserialize_perform(vlan_psz);
763 	psref_target_destroy(&omib->ifvm_psref, ifvm_psref_class);
764 }
765 
766 /*
767  * Called when a parent interface is detaching; destroy any VLAN
768  * configuration for the parent interface.
769  */
770 void
771 vlan_ifdetach(struct ifnet *p)
772 {
773 	struct ifvlan *ifv;
774 	struct ifvlan_linkmib *mib, **nmibs;
775 	struct psref psref;
776 	int error;
777 	int bound;
778 	int i, cnt = 0;
779 
780 	bound = curlwp_bind();
781 	mutex_enter(&ifv_list.lock);
782 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
783 		mib = vlan_getref_linkmib(ifv, &psref);
784 		if (mib == NULL)
785 			continue;
786 
787 		if (mib->ifvm_p == p)
788 			cnt++;
789 
790 		vlan_putref_linkmib(mib, &psref);
791 	}
792 	mutex_exit(&ifv_list.lock);
793 
794 	/*
795 	 * The value of "cnt" does not increase while ifv_list.lock
796 	 * and ifv->ifv_lock are released here, because the parent
797 	 * interface is detaching.
798 	 */
799 	nmibs = kmem_alloc(sizeof(*nmibs) * cnt, KM_SLEEP);
800 	for (i=0; i < cnt; i++) {
801 		nmibs[i] = kmem_alloc(sizeof(*nmibs[i]), KM_SLEEP);
802 	}
803 
804 	mutex_enter(&ifv_list.lock);
805 
806 	i = 0;
807 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
808 		mutex_enter(&ifv->ifv_lock);
809 		if (ifv->ifv_mib->ifvm_p == p) {
810 			KASSERTMSG(i < cnt, "no memory for unconfig, parent=%s",
811 			    p->if_xname);
812 			error = vlan_unconfig_locked(ifv, nmibs[i]);
813 			if (!error) {
814 				nmibs[i] = NULL;
815 				i++;
816 			}
817 
818 		}
819 		mutex_exit(&ifv->ifv_lock);
820 	}
821 
822 	mutex_exit(&ifv_list.lock);
823 	curlwp_bindx(bound);
824 
825 	for (i=0; i < cnt; i++) {
826 		if (nmibs[i])
827 			kmem_free(nmibs[i], sizeof(*nmibs[i]));
828 	}
829 
830 	kmem_free(nmibs, sizeof(*nmibs) * cnt);
831 
832 	return;
833 }
834 
835 static int
836 vlan_set_promisc(struct ifnet *ifp)
837 {
838 	struct ifvlan *ifv = ifp->if_softc;
839 	struct ifvlan_linkmib *mib;
840 	struct psref psref;
841 	int error = 0;
842 	int bound;
843 
844 	bound = curlwp_bind();
845 	mib = vlan_getref_linkmib(ifv, &psref);
846 	if (mib == NULL) {
847 		curlwp_bindx(bound);
848 		return EBUSY;
849 	}
850 
851 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
852 		if ((ifv->ifv_flags & IFVF_PROMISC) == 0) {
853 			error = vlan_safe_ifpromisc(mib->ifvm_p, 1);
854 			if (error == 0)
855 				ifv->ifv_flags |= IFVF_PROMISC;
856 		}
857 	} else {
858 		if ((ifv->ifv_flags & IFVF_PROMISC) != 0) {
859 			error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
860 			if (error == 0)
861 				ifv->ifv_flags &= ~IFVF_PROMISC;
862 		}
863 	}
864 	vlan_putref_linkmib(mib, &psref);
865 	curlwp_bindx(bound);
866 
867 	return (error);
868 }
869 
870 static int
871 vlan_ioctl(struct ifnet *ifp, u_long cmd, void *data)
872 {
873 	struct lwp *l = curlwp;	/* XXX */
874 	struct ifvlan *ifv = ifp->if_softc;
875 	struct ifaddr *ifa = (struct ifaddr *) data;
876 	struct ifreq *ifr = (struct ifreq *) data;
877 	struct ifnet *pr;
878 	struct ifcapreq *ifcr;
879 	struct vlanreq vlr;
880 	struct ifvlan_linkmib *mib;
881 	struct psref psref;
882 	int error = 0;
883 	int bound;
884 
885 	switch (cmd) {
886 	case SIOCSIFMTU:
887 		bound = curlwp_bind();
888 		mib = vlan_getref_linkmib(ifv, &psref);
889 		if (mib == NULL) {
890 			curlwp_bindx(bound);
891 			error = EBUSY;
892 			break;
893 		}
894 
895 		if (mib->ifvm_p == NULL) {
896 			vlan_putref_linkmib(mib, &psref);
897 			curlwp_bindx(bound);
898 			error = EINVAL;
899 		} else if (
900 		    ifr->ifr_mtu > (mib->ifvm_p->if_mtu - mib->ifvm_mtufudge) ||
901 		    ifr->ifr_mtu < (mib->ifvm_mintu - mib->ifvm_mtufudge)) {
902 			vlan_putref_linkmib(mib, &psref);
903 			curlwp_bindx(bound);
904 			error = EINVAL;
905 		} else {
906 			vlan_putref_linkmib(mib, &psref);
907 			curlwp_bindx(bound);
908 
909 			error = ifioctl_common(ifp, cmd, data);
910 			if (error == ENETRESET)
911 					error = 0;
912 		}
913 
914 		break;
915 
916 	case SIOCSETVLAN:
917 		if ((error = kauth_authorize_network(l->l_cred,
918 		    KAUTH_NETWORK_INTERFACE,
919 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
920 		    NULL)) != 0)
921 			break;
922 		if ((error = copyin(ifr->ifr_data, &vlr, sizeof(vlr))) != 0)
923 			break;
924 
925 		if (vlr.vlr_parent[0] == '\0') {
926 			bound = curlwp_bind();
927 			mib = vlan_getref_linkmib(ifv, &psref);
928 			if (mib == NULL) {
929 				curlwp_bindx(bound);
930 				error = EBUSY;
931 				break;
932 			}
933 
934 			if (mib->ifvm_p != NULL &&
935 			    (ifp->if_flags & IFF_PROMISC) != 0)
936 				error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
937 
938 			vlan_putref_linkmib(mib, &psref);
939 			curlwp_bindx(bound);
940 
941 			vlan_unconfig(ifp);
942 			break;
943 		}
944 		if (vlr.vlr_tag != EVL_VLANOFTAG(vlr.vlr_tag)) {
945 			error = EINVAL;		 /* check for valid tag */
946 			break;
947 		}
948 		if ((pr = ifunit(vlr.vlr_parent)) == NULL) {
949 			error = ENOENT;
950 			break;
951 		}
952 		error = vlan_config(ifv, pr, vlr.vlr_tag);
953 		if (error != 0) {
954 			break;
955 		}
956 		ifp->if_flags |= IFF_RUNNING;
957 
958 		/* Update promiscuous mode, if necessary. */
959 		vlan_set_promisc(ifp);
960 		break;
961 
962 	case SIOCGETVLAN:
963 		memset(&vlr, 0, sizeof(vlr));
964 		bound = curlwp_bind();
965 		mib = vlan_getref_linkmib(ifv, &psref);
966 		if (mib == NULL) {
967 			curlwp_bindx(bound);
968 			error = EBUSY;
969 			break;
970 		}
971 		if (mib->ifvm_p != NULL) {
972 			snprintf(vlr.vlr_parent, sizeof(vlr.vlr_parent), "%s",
973 			    mib->ifvm_p->if_xname);
974 			vlr.vlr_tag = mib->ifvm_tag;
975 		}
976 		vlan_putref_linkmib(mib, &psref);
977 		curlwp_bindx(bound);
978 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
979 		break;
980 
981 	case SIOCSIFFLAGS:
982 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
983 			break;
984 		/*
985 		 * For promiscuous mode, we enable promiscuous mode on
986 		 * the parent if we need promiscuous on the VLAN interface.
987 		 */
988 		bound = curlwp_bind();
989 		mib = vlan_getref_linkmib(ifv, &psref);
990 		if (mib == NULL) {
991 			curlwp_bindx(bound);
992 			error = EBUSY;
993 			break;
994 		}
995 
996 		if (mib->ifvm_p != NULL)
997 			error = vlan_set_promisc(ifp);
998 		vlan_putref_linkmib(mib, &psref);
999 		curlwp_bindx(bound);
1000 		break;
1001 
1002 	case SIOCADDMULTI:
1003 		mutex_enter(&ifv->ifv_lock);
1004 		mib = ifv->ifv_mib;
1005 		if (mib == NULL) {
1006 			error = EBUSY;
1007 			mutex_exit(&ifv->ifv_lock);
1008 			break;
1009 		}
1010 
1011 		error = (mib->ifvm_p != NULL) ?
1012 		    (*mib->ifvm_msw->vmsw_addmulti)(ifv, ifr) : EINVAL;
1013 		mib = NULL;
1014 		mutex_exit(&ifv->ifv_lock);
1015 		break;
1016 
1017 	case SIOCDELMULTI:
1018 		mutex_enter(&ifv->ifv_lock);
1019 		mib = ifv->ifv_mib;
1020 		if (mib == NULL) {
1021 			error = EBUSY;
1022 			mutex_exit(&ifv->ifv_lock);
1023 			break;
1024 		}
1025 		error = (mib->ifvm_p != NULL) ?
1026 		    (*mib->ifvm_msw->vmsw_delmulti)(ifv, ifr) : EINVAL;
1027 		mib = NULL;
1028 		mutex_exit(&ifv->ifv_lock);
1029 		break;
1030 
1031 	case SIOCSIFCAP:
1032 		ifcr = data;
1033 		/* make sure caps are enabled on parent */
1034 		bound = curlwp_bind();
1035 		mib = vlan_getref_linkmib(ifv, &psref);
1036 		if (mib == NULL) {
1037 			curlwp_bindx(bound);
1038 			error = EBUSY;
1039 			break;
1040 		}
1041 
1042 		if (mib->ifvm_p == NULL) {
1043 			vlan_putref_linkmib(mib, &psref);
1044 			curlwp_bindx(bound);
1045 			error = EINVAL;
1046 			break;
1047 		}
1048 		if ((mib->ifvm_p->if_capenable & ifcr->ifcr_capenable) !=
1049 		    ifcr->ifcr_capenable) {
1050 			vlan_putref_linkmib(mib, &psref);
1051 			curlwp_bindx(bound);
1052 			error = EINVAL;
1053 			break;
1054 		}
1055 
1056 		vlan_putref_linkmib(mib, &psref);
1057 		curlwp_bindx(bound);
1058 
1059 		if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
1060 			error = 0;
1061 		break;
1062 	case SIOCINITIFADDR:
1063 		bound = curlwp_bind();
1064 		mib = vlan_getref_linkmib(ifv, &psref);
1065 		if (mib == NULL) {
1066 			curlwp_bindx(bound);
1067 			error = EBUSY;
1068 			break;
1069 		}
1070 
1071 		if (mib->ifvm_p == NULL) {
1072 			error = EINVAL;
1073 			vlan_putref_linkmib(mib, &psref);
1074 			curlwp_bindx(bound);
1075 			break;
1076 		}
1077 		vlan_putref_linkmib(mib, &psref);
1078 		curlwp_bindx(bound);
1079 
1080 		ifp->if_flags |= IFF_UP;
1081 #ifdef INET
1082 		if (ifa->ifa_addr->sa_family == AF_INET)
1083 			arp_ifinit(ifp, ifa);
1084 #endif
1085 		break;
1086 
1087 	default:
1088 		error = ether_ioctl(ifp, cmd, data);
1089 	}
1090 
1091 	return (error);
1092 }
1093 
1094 static int
1095 vlan_ether_addmulti(struct ifvlan *ifv, struct ifreq *ifr)
1096 {
1097 	const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr);
1098 	struct vlan_mc_entry *mc;
1099 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1100 	struct ifvlan_linkmib *mib;
1101 	int error;
1102 
1103 	KASSERT(mutex_owned(&ifv->ifv_lock));
1104 
1105 	if (sa->sa_len > sizeof(struct sockaddr_storage))
1106 		return (EINVAL);
1107 
1108 	error = ether_addmulti(sa, &ifv->ifv_ec);
1109 	if (error != ENETRESET)
1110 		return (error);
1111 
1112 	/*
1113 	 * This is new multicast address.  We have to tell parent
1114 	 * about it.  Also, remember this multicast address so that
1115 	 * we can delete them on unconfigure.
1116 	 */
1117 	mc = malloc(sizeof(struct vlan_mc_entry), M_DEVBUF, M_NOWAIT);
1118 	if (mc == NULL) {
1119 		error = ENOMEM;
1120 		goto alloc_failed;
1121 	}
1122 
1123 	/*
1124 	 * As ether_addmulti() returns ENETRESET, following two
1125 	 * statement shouldn't fail.
1126 	 */
1127 	(void)ether_multiaddr(sa, addrlo, addrhi);
1128 	ETHER_LOOKUP_MULTI(addrlo, addrhi, &ifv->ifv_ec, mc->mc_enm);
1129 	memcpy(&mc->mc_addr, sa, sa->sa_len);
1130 	LIST_INSERT_HEAD(&ifv->ifv_mc_listhead, mc, mc_entries);
1131 
1132 	mib = ifv->ifv_mib;
1133 
1134 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1135 	error = if_mcast_op(mib->ifvm_p, SIOCADDMULTI, sa);
1136 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1137 
1138 	if (error != 0)
1139 		goto ioctl_failed;
1140 	return (error);
1141 
1142  ioctl_failed:
1143 	LIST_REMOVE(mc, mc_entries);
1144 	free(mc, M_DEVBUF);
1145  alloc_failed:
1146 	(void)ether_delmulti(sa, &ifv->ifv_ec);
1147 	return (error);
1148 }
1149 
1150 static int
1151 vlan_ether_delmulti(struct ifvlan *ifv, struct ifreq *ifr)
1152 {
1153 	const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr);
1154 	struct ether_multi *enm;
1155 	struct vlan_mc_entry *mc;
1156 	struct ifvlan_linkmib *mib;
1157 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1158 	int error;
1159 
1160 	KASSERT(mutex_owned(&ifv->ifv_lock));
1161 
1162 	/*
1163 	 * Find a key to lookup vlan_mc_entry.  We have to do this
1164 	 * before calling ether_delmulti for obvious reason.
1165 	 */
1166 	if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0)
1167 		return (error);
1168 	ETHER_LOOKUP_MULTI(addrlo, addrhi, &ifv->ifv_ec, enm);
1169 
1170 	error = ether_delmulti(sa, &ifv->ifv_ec);
1171 	if (error != ENETRESET)
1172 		return (error);
1173 
1174 	/* We no longer use this multicast address.  Tell parent so. */
1175 	mib = ifv->ifv_mib;
1176 	error = if_mcast_op(mib->ifvm_p, SIOCDELMULTI, sa);
1177 
1178 	if (error == 0) {
1179 		/* And forget about this address. */
1180 		for (mc = LIST_FIRST(&ifv->ifv_mc_listhead); mc != NULL;
1181 		    mc = LIST_NEXT(mc, mc_entries)) {
1182 			if (mc->mc_enm == enm) {
1183 				LIST_REMOVE(mc, mc_entries);
1184 				free(mc, M_DEVBUF);
1185 				break;
1186 			}
1187 		}
1188 		KASSERT(mc != NULL);
1189 	} else
1190 		(void)ether_addmulti(sa, &ifv->ifv_ec);
1191 	return (error);
1192 }
1193 
1194 /*
1195  * Delete any multicast address we have asked to add from parent
1196  * interface.  Called when the vlan is being unconfigured.
1197  */
1198 static void
1199 vlan_ether_purgemulti(struct ifvlan *ifv)
1200 {
1201 	struct vlan_mc_entry *mc;
1202 	struct ifvlan_linkmib *mib;
1203 
1204 	KASSERT(mutex_owned(&ifv->ifv_lock));
1205 	mib = ifv->ifv_mib;
1206 	if (mib == NULL) {
1207 		return;
1208 	}
1209 
1210 	while ((mc = LIST_FIRST(&ifv->ifv_mc_listhead)) != NULL) {
1211 		(void)if_mcast_op(mib->ifvm_p, SIOCDELMULTI,
1212 		    (const struct sockaddr *)&mc->mc_addr);
1213 		LIST_REMOVE(mc, mc_entries);
1214 		free(mc, M_DEVBUF);
1215 	}
1216 }
1217 
1218 static void
1219 vlan_start(struct ifnet *ifp)
1220 {
1221 	struct ifvlan *ifv = ifp->if_softc;
1222 	struct ifnet *p;
1223 	struct ethercom *ec;
1224 	struct mbuf *m;
1225 	struct ifvlan_linkmib *mib;
1226 	struct psref psref;
1227 	int error;
1228 
1229 	mib = vlan_getref_linkmib(ifv, &psref);
1230 	if (mib == NULL)
1231 		return;
1232 	p = mib->ifvm_p;
1233 	ec = (void *)mib->ifvm_p;
1234 
1235 	ifp->if_flags |= IFF_OACTIVE;
1236 
1237 	for (;;) {
1238 		IFQ_DEQUEUE(&ifp->if_snd, m);
1239 		if (m == NULL)
1240 			break;
1241 
1242 #ifdef ALTQ
1243 		/*
1244 		 * KERNEL_LOCK is required for ALTQ even if NET_MPSAFE is defined.
1245 		 */
1246 		KERNEL_LOCK(1, NULL);
1247 		/*
1248 		 * If ALTQ is enabled on the parent interface, do
1249 		 * classification; the queueing discipline might
1250 		 * not require classification, but might require
1251 		 * the address family/header pointer in the pktattr.
1252 		 */
1253 		if (ALTQ_IS_ENABLED(&p->if_snd)) {
1254 			switch (p->if_type) {
1255 			case IFT_ETHER:
1256 				altq_etherclassify(&p->if_snd, m);
1257 				break;
1258 #ifdef DIAGNOSTIC
1259 			default:
1260 				panic("vlan_start: impossible (altq)");
1261 #endif
1262 			}
1263 		}
1264 		KERNEL_UNLOCK_ONE(NULL);
1265 #endif /* ALTQ */
1266 
1267 		bpf_mtap(ifp, m);
1268 		/*
1269 		 * If the parent can insert the tag itself, just mark
1270 		 * the tag in the mbuf header.
1271 		 */
1272 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
1273 			vlan_set_tag(m, mib->ifvm_tag);
1274 		} else {
1275 			/*
1276 			 * insert the tag ourselves
1277 			 */
1278 			M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1279 			if (m == NULL) {
1280 				printf("%s: unable to prepend encap header",
1281 				    p->if_xname);
1282 				ifp->if_oerrors++;
1283 				continue;
1284 			}
1285 
1286 			switch (p->if_type) {
1287 			case IFT_ETHER:
1288 			    {
1289 				struct ether_vlan_header *evl;
1290 
1291 				if (m->m_len < sizeof(struct ether_vlan_header))
1292 					m = m_pullup(m,
1293 					    sizeof(struct ether_vlan_header));
1294 				if (m == NULL) {
1295 					printf("%s: unable to pullup encap "
1296 					    "header", p->if_xname);
1297 					ifp->if_oerrors++;
1298 					continue;
1299 				}
1300 
1301 				/*
1302 				 * Transform the Ethernet header into an
1303 				 * Ethernet header with 802.1Q encapsulation.
1304 				 */
1305 				memmove(mtod(m, void *),
1306 				    mtod(m, char *) + mib->ifvm_encaplen,
1307 				    sizeof(struct ether_header));
1308 				evl = mtod(m, struct ether_vlan_header *);
1309 				evl->evl_proto = evl->evl_encap_proto;
1310 				evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1311 				evl->evl_tag = htons(mib->ifvm_tag);
1312 
1313 				/*
1314 				 * To cater for VLAN-aware layer 2 ethernet
1315 				 * switches which may need to strip the tag
1316 				 * before forwarding the packet, make sure
1317 				 * the packet+tag is at least 68 bytes long.
1318 				 * This is necessary because our parent will
1319 				 * only pad to 64 bytes (ETHER_MIN_LEN) and
1320 				 * some switches will not pad by themselves
1321 				 * after deleting a tag.
1322 				 */
1323 				if (m->m_pkthdr.len <
1324 				    (ETHER_MIN_LEN - ETHER_CRC_LEN +
1325 				     ETHER_VLAN_ENCAP_LEN)) {
1326 					m_copyback(m, m->m_pkthdr.len,
1327 					    (ETHER_MIN_LEN - ETHER_CRC_LEN +
1328 					     ETHER_VLAN_ENCAP_LEN) -
1329 					     m->m_pkthdr.len,
1330 					    vlan_zero_pad_buff);
1331 				}
1332 				break;
1333 			    }
1334 
1335 #ifdef DIAGNOSTIC
1336 			default:
1337 				panic("vlan_start: impossible");
1338 #endif
1339 			}
1340 		}
1341 
1342 		if ((p->if_flags & IFF_RUNNING) == 0) {
1343 			m_freem(m);
1344 			continue;
1345 		}
1346 
1347 		error = if_transmit_lock(p, m);
1348 		if (error) {
1349 			/* mbuf is already freed */
1350 			ifp->if_oerrors++;
1351 			continue;
1352 		}
1353 		ifp->if_opackets++;
1354 	}
1355 
1356 	ifp->if_flags &= ~IFF_OACTIVE;
1357 
1358 	/* Remove reference to mib before release */
1359 	p = NULL;
1360 	ec = NULL;
1361 
1362 	vlan_putref_linkmib(mib, &psref);
1363 }
1364 
1365 static int
1366 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1367 {
1368 	struct ifvlan *ifv = ifp->if_softc;
1369 	struct ifnet *p;
1370 	struct ethercom *ec;
1371 	struct ifvlan_linkmib *mib;
1372 	struct psref psref;
1373 	int error;
1374 	size_t pktlen = m->m_pkthdr.len;
1375 	bool mcast = (m->m_flags & M_MCAST) != 0;
1376 
1377 	mib = vlan_getref_linkmib(ifv, &psref);
1378 	if (mib == NULL) {
1379 		m_freem(m);
1380 		return ENETDOWN;
1381 	}
1382 
1383 	p = mib->ifvm_p;
1384 	ec = (void *)mib->ifvm_p;
1385 
1386 	bpf_mtap(ifp, m);
1387 
1388 	if (pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_OUT) != 0) {
1389 		if (m != NULL)
1390 			m_freem(m);
1391 		error = 0;
1392 		goto out;
1393 	}
1394 
1395 	/*
1396 	 * If the parent can insert the tag itself, just mark
1397 	 * the tag in the mbuf header.
1398 	 */
1399 	if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
1400 		vlan_set_tag(m, mib->ifvm_tag);
1401 	} else {
1402 		/*
1403 		 * insert the tag ourselves
1404 		 */
1405 		M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1406 		if (m == NULL) {
1407 			printf("%s: unable to prepend encap header",
1408 			    p->if_xname);
1409 			ifp->if_oerrors++;
1410 			error = ENOBUFS;
1411 			goto out;
1412 		}
1413 
1414 		switch (p->if_type) {
1415 		case IFT_ETHER:
1416 		    {
1417 			struct ether_vlan_header *evl;
1418 
1419 			if (m->m_len < sizeof(struct ether_vlan_header))
1420 				m = m_pullup(m,
1421 				    sizeof(struct ether_vlan_header));
1422 			if (m == NULL) {
1423 				printf("%s: unable to pullup encap "
1424 				    "header", p->if_xname);
1425 				ifp->if_oerrors++;
1426 				error = ENOBUFS;
1427 				goto out;
1428 			}
1429 
1430 			/*
1431 			 * Transform the Ethernet header into an
1432 			 * Ethernet header with 802.1Q encapsulation.
1433 			 */
1434 			memmove(mtod(m, void *),
1435 			    mtod(m, char *) + mib->ifvm_encaplen,
1436 			    sizeof(struct ether_header));
1437 			evl = mtod(m, struct ether_vlan_header *);
1438 			evl->evl_proto = evl->evl_encap_proto;
1439 			evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1440 			evl->evl_tag = htons(mib->ifvm_tag);
1441 
1442 			/*
1443 			 * To cater for VLAN-aware layer 2 ethernet
1444 			 * switches which may need to strip the tag
1445 			 * before forwarding the packet, make sure
1446 			 * the packet+tag is at least 68 bytes long.
1447 			 * This is necessary because our parent will
1448 			 * only pad to 64 bytes (ETHER_MIN_LEN) and
1449 			 * some switches will not pad by themselves
1450 			 * after deleting a tag.
1451 			 */
1452 			if (m->m_pkthdr.len <
1453 			    (ETHER_MIN_LEN - ETHER_CRC_LEN +
1454 			     ETHER_VLAN_ENCAP_LEN)) {
1455 				m_copyback(m, m->m_pkthdr.len,
1456 				    (ETHER_MIN_LEN - ETHER_CRC_LEN +
1457 				     ETHER_VLAN_ENCAP_LEN) -
1458 				     m->m_pkthdr.len,
1459 				    vlan_zero_pad_buff);
1460 			}
1461 			break;
1462 		    }
1463 
1464 #ifdef DIAGNOSTIC
1465 		default:
1466 			panic("vlan_transmit: impossible");
1467 #endif
1468 		}
1469 	}
1470 
1471 	if ((p->if_flags & IFF_RUNNING) == 0) {
1472 		m_freem(m);
1473 		error = ENETDOWN;
1474 		goto out;
1475 	}
1476 
1477 	error = if_transmit_lock(p, m);
1478 	if (error) {
1479 		/* mbuf is already freed */
1480 		ifp->if_oerrors++;
1481 	} else {
1482 
1483 		ifp->if_opackets++;
1484 		ifp->if_obytes += pktlen;
1485 		if (mcast)
1486 			ifp->if_omcasts++;
1487 	}
1488 
1489 out:
1490 	/* Remove reference to mib before release */
1491 	p = NULL;
1492 	ec = NULL;
1493 
1494 	vlan_putref_linkmib(mib, &psref);
1495 	return error;
1496 }
1497 
1498 /*
1499  * Given an Ethernet frame, find a valid vlan interface corresponding to the
1500  * given source interface and tag, then run the real packet through the
1501  * parent's input routine.
1502  */
1503 void
1504 vlan_input(struct ifnet *ifp, struct mbuf *m)
1505 {
1506 	struct ifvlan *ifv;
1507 	uint16_t vid;
1508 	struct ifvlan_linkmib *mib;
1509 	struct psref psref;
1510 	bool have_vtag;
1511 
1512 	have_vtag = vlan_has_tag(m);
1513 	if (have_vtag) {
1514 		vid = EVL_VLANOFTAG(vlan_get_tag(m));
1515 		m->m_flags &= ~M_VLANTAG;
1516 	} else {
1517 		switch (ifp->if_type) {
1518 		case IFT_ETHER:
1519 		    {
1520 			struct ether_vlan_header *evl;
1521 
1522 			if (m->m_len < sizeof(struct ether_vlan_header) &&
1523 			    (m = m_pullup(m,
1524 			     sizeof(struct ether_vlan_header))) == NULL) {
1525 				printf("%s: no memory for VLAN header, "
1526 				    "dropping packet.\n", ifp->if_xname);
1527 				return;
1528 			}
1529 			evl = mtod(m, struct ether_vlan_header *);
1530 			KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN);
1531 
1532 			vid = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1533 
1534 			/*
1535 			 * Restore the original ethertype.  We'll remove
1536 			 * the encapsulation after we've found the vlan
1537 			 * interface corresponding to the tag.
1538 			 */
1539 			evl->evl_encap_proto = evl->evl_proto;
1540 			break;
1541 		    }
1542 
1543 		default:
1544 			vid = (uint16_t) -1;	/* XXX GCC */
1545 #ifdef DIAGNOSTIC
1546 			panic("vlan_input: impossible");
1547 #endif
1548 		}
1549 	}
1550 
1551 	mib = vlan_lookup_tag_psref(ifp, vid, &psref);
1552 	if (mib == NULL) {
1553 		m_freem(m);
1554 		ifp->if_noproto++;
1555 		return;
1556 	}
1557 
1558 	ifv = mib->ifvm_ifvlan;
1559 	if ((ifv->ifv_if.if_flags & (IFF_UP|IFF_RUNNING)) !=
1560 	    (IFF_UP|IFF_RUNNING)) {
1561 		m_freem(m);
1562 		ifp->if_noproto++;
1563 		goto out;
1564 	}
1565 
1566 	/*
1567 	 * Now, remove the encapsulation header.  The original
1568 	 * header has already been fixed up above.
1569 	 */
1570 	if (!have_vtag) {
1571 		memmove(mtod(m, char *) + mib->ifvm_encaplen,
1572 		    mtod(m, void *), sizeof(struct ether_header));
1573 		m_adj(m, mib->ifvm_encaplen);
1574 	}
1575 
1576 	m_set_rcvif(m, &ifv->ifv_if);
1577 	ifv->ifv_if.if_ipackets++;
1578 
1579 	if (pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_IN) != 0) {
1580 		if (m != NULL)
1581 			m_freem(m);
1582 		goto out;
1583 	}
1584 
1585 	m->m_flags &= ~M_PROMISC;
1586 	if_input(&ifv->ifv_if, m);
1587 out:
1588 	vlan_putref_linkmib(mib, &psref);
1589 }
1590 
1591 /*
1592  * Module infrastructure
1593  */
1594 #include "if_module.h"
1595 
1596 IF_MODULE(MODULE_CLASS_DRIVER, vlan, "")
1597