xref: /netbsd-src/sys/net/if_vlan.c (revision e6c7e151de239c49d2e38720a061ed9d1fa99309)
1 /*	$NetBSD: if_vlan.c,v 1.151 2020/02/01 02:58:15 riastradh 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 1.151 2020/02/01 02:58:15 riastradh Exp $");
82 
83 #ifdef _KERNEL_OPT
84 #include "opt_inet.h"
85 #include "opt_net_mpsafe.h"
86 #endif
87 
88 #include <sys/param.h>
89 #include <sys/systm.h>
90 #include <sys/kernel.h>
91 #include <sys/mbuf.h>
92 #include <sys/queue.h>
93 #include <sys/socket.h>
94 #include <sys/sockio.h>
95 #include <sys/systm.h>
96 #include <sys/proc.h>
97 #include <sys/kauth.h>
98 #include <sys/mutex.h>
99 #include <sys/kmem.h>
100 #include <sys/cpu.h>
101 #include <sys/pserialize.h>
102 #include <sys/psref.h>
103 #include <sys/pslist.h>
104 #include <sys/atomic.h>
105 #include <sys/device.h>
106 #include <sys/module.h>
107 
108 #include <net/bpf.h>
109 #include <net/if.h>
110 #include <net/if_dl.h>
111 #include <net/if_types.h>
112 #include <net/if_ether.h>
113 #include <net/if_vlanvar.h>
114 
115 #ifdef INET
116 #include <netinet/in.h>
117 #include <netinet/if_inarp.h>
118 #endif
119 #ifdef INET6
120 #include <netinet6/in6_ifattach.h>
121 #include <netinet6/in6_var.h>
122 #include <netinet6/nd6.h>
123 #endif
124 
125 #include "ioconf.h"
126 
127 struct vlan_mc_entry {
128 	LIST_ENTRY(vlan_mc_entry)	mc_entries;
129 	/*
130 	 * A key to identify this entry.  The mc_addr below can't be
131 	 * used since multiple sockaddr may mapped into the same
132 	 * ether_multi (e.g., AF_UNSPEC).
133 	 */
134 	struct ether_multi	*mc_enm;
135 	struct sockaddr_storage		mc_addr;
136 };
137 
138 struct ifvlan_linkmib {
139 	struct ifvlan *ifvm_ifvlan;
140 	const struct vlan_multisw *ifvm_msw;
141 	int	ifvm_encaplen;	/* encapsulation length */
142 	int	ifvm_mtufudge;	/* MTU fudged by this much */
143 	int	ifvm_mintu;	/* min transmission unit */
144 	uint16_t ifvm_proto;	/* encapsulation ethertype */
145 	uint16_t ifvm_tag;	/* tag to apply on packets */
146 	struct ifnet *ifvm_p;	/* parent interface of this vlan */
147 
148 	struct psref_target ifvm_psref;
149 };
150 
151 struct ifvlan {
152 	struct ethercom ifv_ec;
153 	struct ifvlan_linkmib *ifv_mib;	/*
154 					 * reader must use vlan_getref_linkmib()
155 					 * instead of direct dereference
156 					 */
157 	kmutex_t ifv_lock;		/* writer lock for ifv_mib */
158 	pserialize_t ifv_psz;
159 
160 	LIST_HEAD(__vlan_mchead, vlan_mc_entry) ifv_mc_listhead;
161 	LIST_ENTRY(ifvlan) ifv_list;
162 	struct pslist_entry ifv_hash;
163 	int ifv_flags;
164 };
165 
166 #define	IFVF_PROMISC	0x01		/* promiscuous mode enabled */
167 
168 #define	ifv_if		ifv_ec.ec_if
169 
170 #define	ifv_msw		ifv_mib.ifvm_msw
171 #define	ifv_encaplen	ifv_mib.ifvm_encaplen
172 #define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
173 #define	ifv_mintu	ifv_mib.ifvm_mintu
174 #define	ifv_tag		ifv_mib.ifvm_tag
175 
176 struct vlan_multisw {
177 	int	(*vmsw_addmulti)(struct ifvlan *, struct ifreq *);
178 	int	(*vmsw_delmulti)(struct ifvlan *, struct ifreq *);
179 	void	(*vmsw_purgemulti)(struct ifvlan *);
180 };
181 
182 static int	vlan_ether_addmulti(struct ifvlan *, struct ifreq *);
183 static int	vlan_ether_delmulti(struct ifvlan *, struct ifreq *);
184 static void	vlan_ether_purgemulti(struct ifvlan *);
185 
186 const struct vlan_multisw vlan_ether_multisw = {
187 	.vmsw_addmulti = vlan_ether_addmulti,
188 	.vmsw_delmulti = vlan_ether_delmulti,
189 	.vmsw_purgemulti = vlan_ether_purgemulti,
190 };
191 
192 static int	vlan_clone_create(struct if_clone *, int);
193 static int	vlan_clone_destroy(struct ifnet *);
194 static int	vlan_config(struct ifvlan *, struct ifnet *, uint16_t);
195 static int	vlan_ioctl(struct ifnet *, u_long, void *);
196 static void	vlan_start(struct ifnet *);
197 static int	vlan_transmit(struct ifnet *, struct mbuf *);
198 static void	vlan_unconfig(struct ifnet *);
199 static int	vlan_unconfig_locked(struct ifvlan *, struct ifvlan_linkmib *);
200 static void	vlan_hash_init(void);
201 static int	vlan_hash_fini(void);
202 static int	vlan_tag_hash(uint16_t, u_long);
203 static struct ifvlan_linkmib*	vlan_getref_linkmib(struct ifvlan *,
204     struct psref *);
205 static void	vlan_putref_linkmib(struct ifvlan_linkmib *, struct psref *);
206 static void	vlan_linkmib_update(struct ifvlan *, struct ifvlan_linkmib *);
207 static struct ifvlan_linkmib*	vlan_lookup_tag_psref(struct ifnet *,
208     uint16_t, struct psref *);
209 
210 static struct {
211 	kmutex_t lock;
212 	LIST_HEAD(vlan_ifvlist, ifvlan) list;
213 } ifv_list __cacheline_aligned;
214 
215 
216 #if !defined(VLAN_TAG_HASH_SIZE)
217 #define VLAN_TAG_HASH_SIZE 32
218 #endif
219 static struct {
220 	kmutex_t lock;
221 	struct pslist_head *lists;
222 	u_long mask;
223 } ifv_hash __cacheline_aligned = {
224 	.lists = NULL,
225 	.mask = 0,
226 };
227 
228 pserialize_t vlan_psz __read_mostly;
229 static struct psref_class *ifvm_psref_class __read_mostly;
230 
231 struct if_clone vlan_cloner =
232     IF_CLONE_INITIALIZER("vlan", vlan_clone_create, vlan_clone_destroy);
233 
234 /* Used to pad ethernet frames with < ETHER_MIN_LEN bytes */
235 static char vlan_zero_pad_buff[ETHER_MIN_LEN];
236 
237 static inline int
238 vlan_safe_ifpromisc(struct ifnet *ifp, int pswitch)
239 {
240 	int e;
241 
242 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
243 	e = ifpromisc(ifp, pswitch);
244 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
245 
246 	return e;
247 }
248 
249 static inline int
250 vlan_safe_ifpromisc_locked(struct ifnet *ifp, int pswitch)
251 {
252 	int e;
253 
254 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
255 	e = ifpromisc_locked(ifp, pswitch);
256 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
257 
258 	return e;
259 }
260 
261 void
262 vlanattach(int n)
263 {
264 
265 	/*
266 	 * Nothing to do here, initialization is handled by the
267 	 * module initialization code in vlaninit() below.
268 	 */
269 }
270 
271 static void
272 vlaninit(void)
273 {
274 	mutex_init(&ifv_list.lock, MUTEX_DEFAULT, IPL_NONE);
275 	LIST_INIT(&ifv_list.list);
276 
277 	mutex_init(&ifv_hash.lock, MUTEX_DEFAULT, IPL_NONE);
278 	vlan_psz = pserialize_create();
279 	ifvm_psref_class = psref_class_create("vlanlinkmib", IPL_SOFTNET);
280 	if_clone_attach(&vlan_cloner);
281 
282 	vlan_hash_init();
283 	MODULE_HOOK_SET(if_vlan_vlan_input_hook, vlan_input);
284 }
285 
286 static int
287 vlandetach(void)
288 {
289 	bool is_empty;
290 	int error;
291 
292 	mutex_enter(&ifv_list.lock);
293 	is_empty = LIST_EMPTY(&ifv_list.list);
294 	mutex_exit(&ifv_list.lock);
295 
296 	if (!is_empty)
297 		return EBUSY;
298 
299 	error = vlan_hash_fini();
300 	if (error != 0)
301 		return error;
302 
303 	if_clone_detach(&vlan_cloner);
304 	psref_class_destroy(ifvm_psref_class);
305 	pserialize_destroy(vlan_psz);
306 	mutex_destroy(&ifv_hash.lock);
307 	mutex_destroy(&ifv_list.lock);
308 
309 	MODULE_HOOK_UNSET(if_vlan_vlan_input_hook);
310 	return 0;
311 }
312 
313 static void
314 vlan_reset_linkname(struct ifnet *ifp)
315 {
316 
317 	/*
318 	 * We start out with a "802.1Q VLAN" type and zero-length
319 	 * addresses.  When we attach to a parent interface, we
320 	 * inherit its type, address length, address, and data link
321 	 * type.
322 	 */
323 
324 	ifp->if_type = IFT_L2VLAN;
325 	ifp->if_addrlen = 0;
326 	ifp->if_dlt = DLT_NULL;
327 	if_alloc_sadl(ifp);
328 }
329 
330 static int
331 vlan_clone_create(struct if_clone *ifc, int unit)
332 {
333 	struct ifvlan *ifv;
334 	struct ifnet *ifp;
335 	struct ifvlan_linkmib *mib;
336 	int rv;
337 
338 	ifv = malloc(sizeof(struct ifvlan), M_DEVBUF, M_WAITOK | M_ZERO);
339 	mib = kmem_zalloc(sizeof(struct ifvlan_linkmib), KM_SLEEP);
340 	ifp = &ifv->ifv_if;
341 	LIST_INIT(&ifv->ifv_mc_listhead);
342 
343 	mib->ifvm_ifvlan = ifv;
344 	mib->ifvm_p = NULL;
345 	psref_target_init(&mib->ifvm_psref, ifvm_psref_class);
346 
347 	mutex_init(&ifv->ifv_lock, MUTEX_DEFAULT, IPL_NONE);
348 	ifv->ifv_psz = pserialize_create();
349 	ifv->ifv_mib = mib;
350 
351 	mutex_enter(&ifv_list.lock);
352 	LIST_INSERT_HEAD(&ifv_list.list, ifv, ifv_list);
353 	mutex_exit(&ifv_list.lock);
354 
355 	if_initname(ifp, ifc->ifc_name, unit);
356 	ifp->if_softc = ifv;
357 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
358 	ifp->if_extflags = IFEF_NO_LINK_STATE_CHANGE;
359 #ifdef NET_MPSAFE
360 	ifp->if_extflags |= IFEF_MPSAFE;
361 #endif
362 	ifp->if_start = vlan_start;
363 	ifp->if_transmit = vlan_transmit;
364 	ifp->if_ioctl = vlan_ioctl;
365 	IFQ_SET_READY(&ifp->if_snd);
366 
367 	rv = if_initialize(ifp);
368 	if (rv != 0) {
369 		aprint_error("%s: if_initialize failed(%d)\n", ifp->if_xname,
370 		    rv);
371 		goto fail;
372 	}
373 
374 	vlan_reset_linkname(ifp);
375 	if_register(ifp);
376 	return 0;
377 
378 fail:
379 	mutex_enter(&ifv_list.lock);
380 	LIST_REMOVE(ifv, ifv_list);
381 	mutex_exit(&ifv_list.lock);
382 
383 	mutex_destroy(&ifv->ifv_lock);
384 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
385 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
386 	free(ifv, M_DEVBUF);
387 
388 	return rv;
389 }
390 
391 static int
392 vlan_clone_destroy(struct ifnet *ifp)
393 {
394 	struct ifvlan *ifv = ifp->if_softc;
395 
396 	mutex_enter(&ifv_list.lock);
397 	LIST_REMOVE(ifv, ifv_list);
398 	mutex_exit(&ifv_list.lock);
399 
400 	IFNET_LOCK(ifp);
401 	vlan_unconfig(ifp);
402 	IFNET_UNLOCK(ifp);
403 	if_detach(ifp);
404 
405 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
406 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
407 	pserialize_destroy(ifv->ifv_psz);
408 	mutex_destroy(&ifv->ifv_lock);
409 	free(ifv, M_DEVBUF);
410 
411 	return 0;
412 }
413 
414 /*
415  * Configure a VLAN interface.
416  */
417 static int
418 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
419 {
420 	struct ifnet *ifp = &ifv->ifv_if;
421 	struct ifvlan_linkmib *nmib = NULL;
422 	struct ifvlan_linkmib *omib = NULL;
423 	struct ifvlan_linkmib *checkmib;
424 	struct psref_target *nmib_psref = NULL;
425 	const uint16_t vid = EVL_VLANOFTAG(tag);
426 	int error = 0;
427 	int idx;
428 	bool omib_cleanup = false;
429 	struct psref psref;
430 
431 	/* VLAN ID 0 and 4095 are reserved in the spec */
432 	if ((vid == 0) || (vid == 0xfff))
433 		return EINVAL;
434 
435 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
436 	mutex_enter(&ifv->ifv_lock);
437 	omib = ifv->ifv_mib;
438 
439 	if (omib->ifvm_p != NULL) {
440 		error = EBUSY;
441 		goto done;
442 	}
443 
444 	/* Duplicate check */
445 	checkmib = vlan_lookup_tag_psref(p, vid, &psref);
446 	if (checkmib != NULL) {
447 		vlan_putref_linkmib(checkmib, &psref);
448 		error = EEXIST;
449 		goto done;
450 	}
451 
452 	*nmib = *omib;
453 	nmib_psref = &nmib->ifvm_psref;
454 
455 	psref_target_init(nmib_psref, ifvm_psref_class);
456 
457 	switch (p->if_type) {
458 	case IFT_ETHER:
459 	    {
460 		struct ethercom *ec = (void *)p;
461 		struct vlanid_list *vidmem;
462 
463 		nmib->ifvm_msw = &vlan_ether_multisw;
464 		nmib->ifvm_encaplen = ETHER_VLAN_ENCAP_LEN;
465 		nmib->ifvm_mintu = ETHERMIN;
466 
467 		if (ec->ec_nvlans++ == 0) {
468 			IFNET_LOCK(p);
469 			error = ether_enable_vlan_mtu(p);
470 			IFNET_UNLOCK(p);
471 			if (error >= 0) {
472 				if (error) {
473 					ec->ec_nvlans--;
474 					goto done;
475 				}
476 				nmib->ifvm_mtufudge = 0;
477 			} else {
478 				/*
479 				 * Fudge the MTU by the encapsulation size. This
480 				 * makes us incompatible with strictly compliant
481 				 * 802.1Q implementations, but allows us to use
482 				 * the feature with other NetBSD
483 				 * implementations, which might still be useful.
484 				 */
485 				nmib->ifvm_mtufudge = nmib->ifvm_encaplen;
486 			}
487 			error = 0;
488 		}
489 		/* Add a vid to the list */
490 		vidmem = kmem_alloc(sizeof(struct vlanid_list), KM_SLEEP);
491 		vidmem->vid = vid;
492 		ETHER_LOCK(ec);
493 		SIMPLEQ_INSERT_TAIL(&ec->ec_vids, vidmem, vid_list);
494 		ETHER_UNLOCK(ec);
495 
496 		if (ec->ec_vlan_cb != NULL) {
497 			/*
498 			 * Call ec_vlan_cb(). It will setup VLAN HW filter or
499 			 * HW tagging function.
500 			 */
501 			error = (*ec->ec_vlan_cb)(ec, vid, true);
502 			if (error) {
503 				ec->ec_nvlans--;
504 				if (ec->ec_nvlans == 0) {
505 					IFNET_LOCK(p);
506 					(void)ether_disable_vlan_mtu(p);
507 					IFNET_UNLOCK(p);
508 				}
509 				goto done;
510 			}
511 		}
512 		/*
513 		 * If the parent interface can do hardware-assisted
514 		 * VLAN encapsulation, then propagate its hardware-
515 		 * assisted checksumming flags and tcp segmentation
516 		 * offload.
517 		 */
518 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
519 			ifp->if_capabilities = p->if_capabilities &
520 			    (IFCAP_TSOv4 | IFCAP_TSOv6 |
521 				IFCAP_CSUM_IPv4_Tx  | IFCAP_CSUM_IPv4_Rx |
522 				IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
523 				IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
524 				IFCAP_CSUM_TCPv6_Tx | IFCAP_CSUM_TCPv6_Rx |
525 				IFCAP_CSUM_UDPv6_Tx | IFCAP_CSUM_UDPv6_Rx);
526 		}
527 
528 		/*
529 		 * We inherit the parent's Ethernet address.
530 		 */
531 		ether_ifattach(ifp, CLLADDR(p->if_sadl));
532 		ifp->if_hdrlen = sizeof(struct ether_vlan_header); /* XXX? */
533 		break;
534 	    }
535 
536 	default:
537 		error = EPROTONOSUPPORT;
538 		goto done;
539 	}
540 
541 	nmib->ifvm_p = p;
542 	nmib->ifvm_tag = vid;
543 	ifv->ifv_if.if_mtu = p->if_mtu - nmib->ifvm_mtufudge;
544 #ifdef INET6
545 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
546 	if (in6_present)
547 		nd6_setmtu(ifp);
548 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
549 #endif
550 	ifv->ifv_if.if_flags = p->if_flags &
551 	    (IFF_UP | IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
552 
553 	/*
554 	 * Inherit the if_type from the parent.  This allows us
555 	 * to participate in bridges of that type.
556 	 */
557 	ifv->ifv_if.if_type = p->if_type;
558 
559 	PSLIST_ENTRY_INIT(ifv, ifv_hash);
560 	idx = vlan_tag_hash(vid, ifv_hash.mask);
561 
562 	mutex_enter(&ifv_hash.lock);
563 	PSLIST_WRITER_INSERT_HEAD(&ifv_hash.lists[idx], ifv, ifv_hash);
564 	mutex_exit(&ifv_hash.lock);
565 
566 	vlan_linkmib_update(ifv, nmib);
567 	nmib = NULL;
568 	nmib_psref = NULL;
569 	omib_cleanup = true;
570 
571 done:
572 	mutex_exit(&ifv->ifv_lock);
573 
574 	if (nmib_psref)
575 		psref_target_destroy(nmib_psref, ifvm_psref_class);
576 	if (nmib)
577 		kmem_free(nmib, sizeof(*nmib));
578 	if (omib_cleanup)
579 		kmem_free(omib, sizeof(*omib));
580 
581 	return error;
582 }
583 
584 /*
585  * Unconfigure a VLAN interface.
586  */
587 static void
588 vlan_unconfig(struct ifnet *ifp)
589 {
590 	struct ifvlan *ifv = ifp->if_softc;
591 	struct ifvlan_linkmib *nmib = NULL;
592 	int error;
593 
594 	KASSERT(IFNET_LOCKED(ifp));
595 
596 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
597 
598 	mutex_enter(&ifv->ifv_lock);
599 	error = vlan_unconfig_locked(ifv, nmib);
600 	mutex_exit(&ifv->ifv_lock);
601 
602 	if (error)
603 		kmem_free(nmib, sizeof(*nmib));
604 }
605 static int
606 vlan_unconfig_locked(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
607 {
608 	struct ifnet *p;
609 	struct ifnet *ifp = &ifv->ifv_if;
610 	struct psref_target *nmib_psref = NULL;
611 	struct ifvlan_linkmib *omib;
612 	int error = 0;
613 
614 	KASSERT(IFNET_LOCKED(ifp));
615 	KASSERT(mutex_owned(&ifv->ifv_lock));
616 
617 	ifp->if_flags &= ~(IFF_UP | IFF_RUNNING);
618 
619 	omib = ifv->ifv_mib;
620 	p = omib->ifvm_p;
621 
622 	if (p == NULL) {
623 		error = -1;
624 		goto done;
625 	}
626 
627 	*nmib = *omib;
628 	nmib_psref = &nmib->ifvm_psref;
629 	psref_target_init(nmib_psref, ifvm_psref_class);
630 
631 	/*
632 	 * Since the interface is being unconfigured, we need to empty the
633 	 * list of multicast groups that we may have joined while we were
634 	 * alive and remove them from the parent's list also.
635 	 */
636 	(*nmib->ifvm_msw->vmsw_purgemulti)(ifv);
637 
638 	/* Disconnect from parent. */
639 	switch (p->if_type) {
640 	case IFT_ETHER:
641 	    {
642 		struct ethercom *ec = (void *)p;
643 		struct vlanid_list *vlanidp;
644 		uint16_t vid = EVL_VLANOFTAG(nmib->ifvm_tag);
645 
646 		ETHER_LOCK(ec);
647 		SIMPLEQ_FOREACH(vlanidp, &ec->ec_vids, vid_list) {
648 			if (vlanidp->vid == vid) {
649 				SIMPLEQ_REMOVE(&ec->ec_vids, vlanidp,
650 				    vlanid_list, vid_list);
651 				break;
652 			}
653 		}
654 		ETHER_UNLOCK(ec);
655 		if (vlanidp != NULL)
656 			kmem_free(vlanidp, sizeof(*vlanidp));
657 
658 		if (ec->ec_vlan_cb != NULL) {
659 			/*
660 			 * Call ec_vlan_cb(). It will setup VLAN HW filter or
661 			 * HW tagging function.
662 			 */
663 			(void)(*ec->ec_vlan_cb)(ec, vid, false);
664 		}
665 		if (--ec->ec_nvlans == 0) {
666 			IFNET_LOCK(p);
667 			(void)ether_disable_vlan_mtu(p);
668 			IFNET_UNLOCK(p);
669 		}
670 
671 		/* XXX ether_ifdetach must not be called with IFNET_LOCK */
672 		mutex_exit(&ifv->ifv_lock);
673 		IFNET_UNLOCK(ifp);
674 		ether_ifdetach(ifp);
675 		IFNET_LOCK(ifp);
676 		mutex_enter(&ifv->ifv_lock);
677 
678 		/* if_free_sadl must be called with IFNET_LOCK */
679 		if_free_sadl(ifp, 1);
680 
681 		/* Restore vlan_ioctl overwritten by ether_ifdetach */
682 		ifp->if_ioctl = vlan_ioctl;
683 		vlan_reset_linkname(ifp);
684 		break;
685 	    }
686 
687 	default:
688 		panic("%s: impossible", __func__);
689 	}
690 
691 	nmib->ifvm_p = NULL;
692 	ifv->ifv_if.if_mtu = 0;
693 	ifv->ifv_flags = 0;
694 
695 	mutex_enter(&ifv_hash.lock);
696 	PSLIST_WRITER_REMOVE(ifv, ifv_hash);
697 	pserialize_perform(vlan_psz);
698 	mutex_exit(&ifv_hash.lock);
699 	PSLIST_ENTRY_DESTROY(ifv, ifv_hash);
700 
701 	vlan_linkmib_update(ifv, nmib);
702 
703 	mutex_exit(&ifv->ifv_lock);
704 
705 	nmib_psref = NULL;
706 	kmem_free(omib, sizeof(*omib));
707 
708 #ifdef INET6
709 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
710 	/* To delete v6 link local addresses */
711 	if (in6_present)
712 		in6_ifdetach(ifp);
713 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
714 #endif
715 
716 	if ((ifp->if_flags & IFF_PROMISC) != 0)
717 		vlan_safe_ifpromisc_locked(ifp, 0);
718 	if_down_locked(ifp);
719 	ifp->if_capabilities = 0;
720 	mutex_enter(&ifv->ifv_lock);
721 done:
722 
723 	if (nmib_psref)
724 		psref_target_destroy(nmib_psref, ifvm_psref_class);
725 
726 	return error;
727 }
728 
729 static void
730 vlan_hash_init(void)
731 {
732 
733 	ifv_hash.lists = hashinit(VLAN_TAG_HASH_SIZE, HASH_PSLIST, true,
734 	    &ifv_hash.mask);
735 }
736 
737 static int
738 vlan_hash_fini(void)
739 {
740 	int i;
741 
742 	mutex_enter(&ifv_hash.lock);
743 
744 	for (i = 0; i < ifv_hash.mask + 1; i++) {
745 		if (PSLIST_WRITER_FIRST(&ifv_hash.lists[i], struct ifvlan,
746 		    ifv_hash) != NULL) {
747 			mutex_exit(&ifv_hash.lock);
748 			return EBUSY;
749 		}
750 	}
751 
752 	for (i = 0; i < ifv_hash.mask + 1; i++)
753 		PSLIST_DESTROY(&ifv_hash.lists[i]);
754 
755 	mutex_exit(&ifv_hash.lock);
756 
757 	hashdone(ifv_hash.lists, HASH_PSLIST, ifv_hash.mask);
758 
759 	ifv_hash.lists = NULL;
760 	ifv_hash.mask = 0;
761 
762 	return 0;
763 }
764 
765 static int
766 vlan_tag_hash(uint16_t tag, u_long mask)
767 {
768 	uint32_t hash;
769 
770 	hash = (tag >> 8) ^ tag;
771 	hash = (hash >> 2) ^ hash;
772 
773 	return hash & mask;
774 }
775 
776 static struct ifvlan_linkmib *
777 vlan_getref_linkmib(struct ifvlan *sc, struct psref *psref)
778 {
779 	struct ifvlan_linkmib *mib;
780 	int s;
781 
782 	s = pserialize_read_enter();
783 	mib = atomic_load_consume(&sc->ifv_mib);
784 	if (mib == NULL) {
785 		pserialize_read_exit(s);
786 		return NULL;
787 	}
788 	psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
789 	pserialize_read_exit(s);
790 
791 	return mib;
792 }
793 
794 static void
795 vlan_putref_linkmib(struct ifvlan_linkmib *mib, struct psref *psref)
796 {
797 	if (mib == NULL)
798 		return;
799 	psref_release(psref, &mib->ifvm_psref, ifvm_psref_class);
800 }
801 
802 static struct ifvlan_linkmib *
803 vlan_lookup_tag_psref(struct ifnet *ifp, uint16_t tag, struct psref *psref)
804 {
805 	int idx;
806 	int s;
807 	struct ifvlan *sc;
808 
809 	idx = vlan_tag_hash(tag, ifv_hash.mask);
810 
811 	s = pserialize_read_enter();
812 	PSLIST_READER_FOREACH(sc, &ifv_hash.lists[idx], struct ifvlan,
813 	    ifv_hash) {
814 		struct ifvlan_linkmib *mib = atomic_load_consume(&sc->ifv_mib);
815 		if (mib == NULL)
816 			continue;
817 		if (mib->ifvm_tag != tag)
818 			continue;
819 		if (mib->ifvm_p != ifp)
820 			continue;
821 
822 		psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
823 		pserialize_read_exit(s);
824 		return mib;
825 	}
826 	pserialize_read_exit(s);
827 	return NULL;
828 }
829 
830 static void
831 vlan_linkmib_update(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
832 {
833 	struct ifvlan_linkmib *omib = ifv->ifv_mib;
834 
835 	KASSERT(mutex_owned(&ifv->ifv_lock));
836 
837 	atomic_store_release(&ifv->ifv_mib, nmib);
838 
839 	pserialize_perform(ifv->ifv_psz);
840 	psref_target_destroy(&omib->ifvm_psref, ifvm_psref_class);
841 }
842 
843 /*
844  * Called when a parent interface is detaching; destroy any VLAN
845  * configuration for the parent interface.
846  */
847 void
848 vlan_ifdetach(struct ifnet *p)
849 {
850 	struct ifvlan *ifv;
851 	struct ifvlan_linkmib *mib, **nmibs;
852 	struct psref psref;
853 	int error;
854 	int bound;
855 	int i, cnt = 0;
856 
857 	bound = curlwp_bind();
858 
859 	mutex_enter(&ifv_list.lock);
860 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
861 		mib = vlan_getref_linkmib(ifv, &psref);
862 		if (mib == NULL)
863 			continue;
864 
865 		if (mib->ifvm_p == p)
866 			cnt++;
867 
868 		vlan_putref_linkmib(mib, &psref);
869 	}
870 	mutex_exit(&ifv_list.lock);
871 
872 	if (cnt == 0) {
873 		curlwp_bindx(bound);
874 		return;
875 	}
876 
877 	/*
878 	 * The value of "cnt" does not increase while ifv_list.lock
879 	 * and ifv->ifv_lock are released here, because the parent
880 	 * interface is detaching.
881 	 */
882 	nmibs = kmem_alloc(sizeof(*nmibs) * cnt, KM_SLEEP);
883 	for (i = 0; i < cnt; i++) {
884 		nmibs[i] = kmem_alloc(sizeof(*nmibs[i]), KM_SLEEP);
885 	}
886 
887 	mutex_enter(&ifv_list.lock);
888 
889 	i = 0;
890 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
891 		struct ifnet *ifp = &ifv->ifv_if;
892 
893 		/* IFNET_LOCK must be held before ifv_lock. */
894 		IFNET_LOCK(ifp);
895 		mutex_enter(&ifv->ifv_lock);
896 
897 		/* XXX ifv_mib = NULL? */
898 		if (ifv->ifv_mib->ifvm_p == p) {
899 			KASSERTMSG(i < cnt,
900 			    "no memory for unconfig, parent=%s", p->if_xname);
901 			error = vlan_unconfig_locked(ifv, nmibs[i]);
902 			if (!error) {
903 				nmibs[i] = NULL;
904 				i++;
905 			}
906 
907 		}
908 
909 		mutex_exit(&ifv->ifv_lock);
910 		IFNET_UNLOCK(ifp);
911 	}
912 
913 	mutex_exit(&ifv_list.lock);
914 
915 	curlwp_bindx(bound);
916 
917 	for (i = 0; i < cnt; i++) {
918 		if (nmibs[i])
919 			kmem_free(nmibs[i], sizeof(*nmibs[i]));
920 	}
921 
922 	kmem_free(nmibs, sizeof(*nmibs) * cnt);
923 
924 	return;
925 }
926 
927 static int
928 vlan_set_promisc(struct ifnet *ifp)
929 {
930 	struct ifvlan *ifv = ifp->if_softc;
931 	struct ifvlan_linkmib *mib;
932 	struct psref psref;
933 	int error = 0;
934 	int bound;
935 
936 	bound = curlwp_bind();
937 	mib = vlan_getref_linkmib(ifv, &psref);
938 	if (mib == NULL) {
939 		curlwp_bindx(bound);
940 		return EBUSY;
941 	}
942 
943 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
944 		if ((ifv->ifv_flags & IFVF_PROMISC) == 0) {
945 			error = vlan_safe_ifpromisc(mib->ifvm_p, 1);
946 			if (error == 0)
947 				ifv->ifv_flags |= IFVF_PROMISC;
948 		}
949 	} else {
950 		if ((ifv->ifv_flags & IFVF_PROMISC) != 0) {
951 			error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
952 			if (error == 0)
953 				ifv->ifv_flags &= ~IFVF_PROMISC;
954 		}
955 	}
956 	vlan_putref_linkmib(mib, &psref);
957 	curlwp_bindx(bound);
958 
959 	return error;
960 }
961 
962 static int
963 vlan_ioctl(struct ifnet *ifp, u_long cmd, void *data)
964 {
965 	struct lwp *l = curlwp;
966 	struct ifvlan *ifv = ifp->if_softc;
967 	struct ifaddr *ifa = (struct ifaddr *) data;
968 	struct ifreq *ifr = (struct ifreq *) data;
969 	struct ifnet *pr;
970 	struct ifcapreq *ifcr;
971 	struct vlanreq vlr;
972 	struct ifvlan_linkmib *mib;
973 	struct psref psref;
974 	int error = 0;
975 	int bound;
976 
977 	switch (cmd) {
978 	case SIOCSIFMTU:
979 		bound = curlwp_bind();
980 		mib = vlan_getref_linkmib(ifv, &psref);
981 		if (mib == NULL) {
982 			curlwp_bindx(bound);
983 			error = EBUSY;
984 			break;
985 		}
986 
987 		if (mib->ifvm_p == NULL) {
988 			vlan_putref_linkmib(mib, &psref);
989 			curlwp_bindx(bound);
990 			error = EINVAL;
991 		} else if (
992 		    ifr->ifr_mtu > (mib->ifvm_p->if_mtu - mib->ifvm_mtufudge) ||
993 		    ifr->ifr_mtu < (mib->ifvm_mintu - mib->ifvm_mtufudge)) {
994 			vlan_putref_linkmib(mib, &psref);
995 			curlwp_bindx(bound);
996 			error = EINVAL;
997 		} else {
998 			vlan_putref_linkmib(mib, &psref);
999 			curlwp_bindx(bound);
1000 
1001 			error = ifioctl_common(ifp, cmd, data);
1002 			if (error == ENETRESET)
1003 				error = 0;
1004 		}
1005 
1006 		break;
1007 
1008 	case SIOCSETVLAN:
1009 		if ((error = kauth_authorize_network(l->l_cred,
1010 		    KAUTH_NETWORK_INTERFACE,
1011 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
1012 		    NULL)) != 0)
1013 			break;
1014 		if ((error = copyin(ifr->ifr_data, &vlr, sizeof(vlr))) != 0)
1015 			break;
1016 
1017 		if (vlr.vlr_parent[0] == '\0') {
1018 			bound = curlwp_bind();
1019 			mib = vlan_getref_linkmib(ifv, &psref);
1020 			if (mib == NULL) {
1021 				curlwp_bindx(bound);
1022 				error = EBUSY;
1023 				break;
1024 			}
1025 
1026 			if (mib->ifvm_p != NULL &&
1027 			    (ifp->if_flags & IFF_PROMISC) != 0)
1028 				error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
1029 
1030 			vlan_putref_linkmib(mib, &psref);
1031 			curlwp_bindx(bound);
1032 
1033 			vlan_unconfig(ifp);
1034 			break;
1035 		}
1036 		if (vlr.vlr_tag != EVL_VLANOFTAG(vlr.vlr_tag)) {
1037 			error = EINVAL;		 /* check for valid tag */
1038 			break;
1039 		}
1040 		if ((pr = ifunit(vlr.vlr_parent)) == NULL) {
1041 			error = ENOENT;
1042 			break;
1043 		}
1044 
1045 		error = vlan_config(ifv, pr, vlr.vlr_tag);
1046 		if (error != 0)
1047 			break;
1048 
1049 		/* Update promiscuous mode, if necessary. */
1050 		vlan_set_promisc(ifp);
1051 
1052 		ifp->if_flags |= IFF_RUNNING;
1053 		break;
1054 
1055 	case SIOCGETVLAN:
1056 		memset(&vlr, 0, sizeof(vlr));
1057 		bound = curlwp_bind();
1058 		mib = vlan_getref_linkmib(ifv, &psref);
1059 		if (mib == NULL) {
1060 			curlwp_bindx(bound);
1061 			error = EBUSY;
1062 			break;
1063 		}
1064 		if (mib->ifvm_p != NULL) {
1065 			snprintf(vlr.vlr_parent, sizeof(vlr.vlr_parent), "%s",
1066 			    mib->ifvm_p->if_xname);
1067 			vlr.vlr_tag = mib->ifvm_tag;
1068 		}
1069 		vlan_putref_linkmib(mib, &psref);
1070 		curlwp_bindx(bound);
1071 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1072 		break;
1073 
1074 	case SIOCSIFFLAGS:
1075 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
1076 			break;
1077 		/*
1078 		 * For promiscuous mode, we enable promiscuous mode on
1079 		 * the parent if we need promiscuous on the VLAN interface.
1080 		 */
1081 		bound = curlwp_bind();
1082 		mib = vlan_getref_linkmib(ifv, &psref);
1083 		if (mib == NULL) {
1084 			curlwp_bindx(bound);
1085 			error = EBUSY;
1086 			break;
1087 		}
1088 
1089 		if (mib->ifvm_p != NULL)
1090 			error = vlan_set_promisc(ifp);
1091 		vlan_putref_linkmib(mib, &psref);
1092 		curlwp_bindx(bound);
1093 		break;
1094 
1095 	case SIOCADDMULTI:
1096 		mutex_enter(&ifv->ifv_lock);
1097 		mib = ifv->ifv_mib;
1098 		if (mib == NULL) {
1099 			error = EBUSY;
1100 			mutex_exit(&ifv->ifv_lock);
1101 			break;
1102 		}
1103 
1104 		error = (mib->ifvm_p != NULL) ?
1105 		    (*mib->ifvm_msw->vmsw_addmulti)(ifv, ifr) : EINVAL;
1106 		mib = NULL;
1107 		mutex_exit(&ifv->ifv_lock);
1108 		break;
1109 
1110 	case SIOCDELMULTI:
1111 		mutex_enter(&ifv->ifv_lock);
1112 		mib = ifv->ifv_mib;
1113 		if (mib == NULL) {
1114 			error = EBUSY;
1115 			mutex_exit(&ifv->ifv_lock);
1116 			break;
1117 		}
1118 		error = (mib->ifvm_p != NULL) ?
1119 		    (*mib->ifvm_msw->vmsw_delmulti)(ifv, ifr) : EINVAL;
1120 		mib = NULL;
1121 		mutex_exit(&ifv->ifv_lock);
1122 		break;
1123 
1124 	case SIOCSIFCAP:
1125 		ifcr = data;
1126 		/* make sure caps are enabled on parent */
1127 		bound = curlwp_bind();
1128 		mib = vlan_getref_linkmib(ifv, &psref);
1129 		if (mib == NULL) {
1130 			curlwp_bindx(bound);
1131 			error = EBUSY;
1132 			break;
1133 		}
1134 
1135 		if (mib->ifvm_p == NULL) {
1136 			vlan_putref_linkmib(mib, &psref);
1137 			curlwp_bindx(bound);
1138 			error = EINVAL;
1139 			break;
1140 		}
1141 		if ((mib->ifvm_p->if_capenable & ifcr->ifcr_capenable) !=
1142 		    ifcr->ifcr_capenable) {
1143 			vlan_putref_linkmib(mib, &psref);
1144 			curlwp_bindx(bound);
1145 			error = EINVAL;
1146 			break;
1147 		}
1148 
1149 		vlan_putref_linkmib(mib, &psref);
1150 		curlwp_bindx(bound);
1151 
1152 		if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
1153 			error = 0;
1154 		break;
1155 	case SIOCINITIFADDR:
1156 		bound = curlwp_bind();
1157 		mib = vlan_getref_linkmib(ifv, &psref);
1158 		if (mib == NULL) {
1159 			curlwp_bindx(bound);
1160 			error = EBUSY;
1161 			break;
1162 		}
1163 
1164 		if (mib->ifvm_p == NULL) {
1165 			error = EINVAL;
1166 			vlan_putref_linkmib(mib, &psref);
1167 			curlwp_bindx(bound);
1168 			break;
1169 		}
1170 		vlan_putref_linkmib(mib, &psref);
1171 		curlwp_bindx(bound);
1172 
1173 		ifp->if_flags |= IFF_UP;
1174 #ifdef INET
1175 		if (ifa->ifa_addr->sa_family == AF_INET)
1176 			arp_ifinit(ifp, ifa);
1177 #endif
1178 		break;
1179 
1180 	default:
1181 		error = ether_ioctl(ifp, cmd, data);
1182 	}
1183 
1184 	return error;
1185 }
1186 
1187 static int
1188 vlan_ether_addmulti(struct ifvlan *ifv, struct ifreq *ifr)
1189 {
1190 	const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr);
1191 	struct vlan_mc_entry *mc;
1192 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1193 	struct ifvlan_linkmib *mib;
1194 	int error;
1195 
1196 	KASSERT(mutex_owned(&ifv->ifv_lock));
1197 
1198 	if (sa->sa_len > sizeof(struct sockaddr_storage))
1199 		return EINVAL;
1200 
1201 	error = ether_addmulti(sa, &ifv->ifv_ec);
1202 	if (error != ENETRESET)
1203 		return error;
1204 
1205 	/*
1206 	 * This is a new multicast address.  We have to tell parent
1207 	 * about it.  Also, remember this multicast address so that
1208 	 * we can delete it on unconfigure.
1209 	 */
1210 	mc = malloc(sizeof(struct vlan_mc_entry), M_DEVBUF, M_NOWAIT);
1211 	if (mc == NULL) {
1212 		error = ENOMEM;
1213 		goto alloc_failed;
1214 	}
1215 
1216 	/*
1217 	 * Since ether_addmulti() returned ENETRESET, the following two
1218 	 * statements shouldn't fail. Here ifv_ec is implicitly protected
1219 	 * by the ifv_lock lock.
1220 	 */
1221 	error = ether_multiaddr(sa, addrlo, addrhi);
1222 	KASSERT(error == 0);
1223 
1224 	ETHER_LOCK(&ifv->ifv_ec);
1225 	mc->mc_enm = ether_lookup_multi(addrlo, addrhi, &ifv->ifv_ec);
1226 	ETHER_UNLOCK(&ifv->ifv_ec);
1227 
1228 	KASSERT(mc->mc_enm != NULL);
1229 
1230 	memcpy(&mc->mc_addr, sa, sa->sa_len);
1231 	LIST_INSERT_HEAD(&ifv->ifv_mc_listhead, mc, mc_entries);
1232 
1233 	mib = ifv->ifv_mib;
1234 
1235 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1236 	error = if_mcast_op(mib->ifvm_p, SIOCADDMULTI, sa);
1237 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1238 
1239 	if (error != 0)
1240 		goto ioctl_failed;
1241 	return error;
1242 
1243 ioctl_failed:
1244 	LIST_REMOVE(mc, mc_entries);
1245 	free(mc, M_DEVBUF);
1246 
1247 alloc_failed:
1248 	(void)ether_delmulti(sa, &ifv->ifv_ec);
1249 	return error;
1250 }
1251 
1252 static int
1253 vlan_ether_delmulti(struct ifvlan *ifv, struct ifreq *ifr)
1254 {
1255 	const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr);
1256 	struct ether_multi *enm;
1257 	struct vlan_mc_entry *mc;
1258 	struct ifvlan_linkmib *mib;
1259 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1260 	int error;
1261 
1262 	KASSERT(mutex_owned(&ifv->ifv_lock));
1263 
1264 	/*
1265 	 * Find a key to lookup vlan_mc_entry.  We have to do this
1266 	 * before calling ether_delmulti for obvious reasons.
1267 	 */
1268 	if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0)
1269 		return error;
1270 
1271 	ETHER_LOCK(&ifv->ifv_ec);
1272 	enm = ether_lookup_multi(addrlo, addrhi, &ifv->ifv_ec);
1273 	ETHER_UNLOCK(&ifv->ifv_ec);
1274 	if (enm == NULL)
1275 		return EINVAL;
1276 
1277 	LIST_FOREACH(mc, &ifv->ifv_mc_listhead, mc_entries) {
1278 		if (mc->mc_enm == enm)
1279 			break;
1280 	}
1281 
1282 	/* We woun't delete entries we didn't add */
1283 	if (mc == NULL)
1284 		return EINVAL;
1285 
1286 	error = ether_delmulti(sa, &ifv->ifv_ec);
1287 	if (error != ENETRESET)
1288 		return error;
1289 
1290 	/* We no longer use this multicast address.  Tell parent so. */
1291 	mib = ifv->ifv_mib;
1292 	error = if_mcast_op(mib->ifvm_p, SIOCDELMULTI, sa);
1293 
1294 	if (error == 0) {
1295 		/* And forget about this address. */
1296 		LIST_REMOVE(mc, mc_entries);
1297 		free(mc, M_DEVBUF);
1298 	} else {
1299 		(void)ether_addmulti(sa, &ifv->ifv_ec);
1300 	}
1301 
1302 	return error;
1303 }
1304 
1305 /*
1306  * Delete any multicast address we have asked to add from parent
1307  * interface.  Called when the vlan is being unconfigured.
1308  */
1309 static void
1310 vlan_ether_purgemulti(struct ifvlan *ifv)
1311 {
1312 	struct vlan_mc_entry *mc;
1313 	struct ifvlan_linkmib *mib;
1314 
1315 	KASSERT(mutex_owned(&ifv->ifv_lock));
1316 	mib = ifv->ifv_mib;
1317 	if (mib == NULL) {
1318 		return;
1319 	}
1320 
1321 	while ((mc = LIST_FIRST(&ifv->ifv_mc_listhead)) != NULL) {
1322 		(void)if_mcast_op(mib->ifvm_p, SIOCDELMULTI,
1323 		    sstocsa(&mc->mc_addr));
1324 		LIST_REMOVE(mc, mc_entries);
1325 		free(mc, M_DEVBUF);
1326 	}
1327 }
1328 
1329 static void
1330 vlan_start(struct ifnet *ifp)
1331 {
1332 	struct ifvlan *ifv = ifp->if_softc;
1333 	struct ifnet *p;
1334 	struct ethercom *ec;
1335 	struct mbuf *m;
1336 	struct ifvlan_linkmib *mib;
1337 	struct psref psref;
1338 	int error;
1339 
1340 	mib = vlan_getref_linkmib(ifv, &psref);
1341 	if (mib == NULL)
1342 		return;
1343 	p = mib->ifvm_p;
1344 	ec = (void *)mib->ifvm_p;
1345 
1346 	ifp->if_flags |= IFF_OACTIVE;
1347 
1348 	for (;;) {
1349 		IFQ_DEQUEUE(&ifp->if_snd, m);
1350 		if (m == NULL)
1351 			break;
1352 
1353 #ifdef ALTQ
1354 		/*
1355 		 * KERNEL_LOCK is required for ALTQ even if NET_MPSAFE is
1356 		 * defined.
1357 		 */
1358 		KERNEL_LOCK(1, NULL);
1359 		/*
1360 		 * If ALTQ is enabled on the parent interface, do
1361 		 * classification; the queueing discipline might
1362 		 * not require classification, but might require
1363 		 * the address family/header pointer in the pktattr.
1364 		 */
1365 		if (ALTQ_IS_ENABLED(&p->if_snd)) {
1366 			switch (p->if_type) {
1367 			case IFT_ETHER:
1368 				altq_etherclassify(&p->if_snd, m);
1369 				break;
1370 			default:
1371 				panic("%s: impossible (altq)", __func__);
1372 			}
1373 		}
1374 		KERNEL_UNLOCK_ONE(NULL);
1375 #endif /* ALTQ */
1376 
1377 		bpf_mtap(ifp, m, BPF_D_OUT);
1378 		/*
1379 		 * If the parent can insert the tag itself, just mark
1380 		 * the tag in the mbuf header.
1381 		 */
1382 		if (ec->ec_capenable & ETHERCAP_VLAN_HWTAGGING) {
1383 			vlan_set_tag(m, mib->ifvm_tag);
1384 		} else {
1385 			/*
1386 			 * insert the tag ourselves
1387 			 */
1388 			M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1389 			if (m == NULL) {
1390 				printf("%s: unable to prepend encap header",
1391 				    p->if_xname);
1392 				if_statinc(ifp, if_oerrors);
1393 				continue;
1394 			}
1395 
1396 			switch (p->if_type) {
1397 			case IFT_ETHER:
1398 			    {
1399 				struct ether_vlan_header *evl;
1400 
1401 				if (m->m_len < sizeof(struct ether_vlan_header))
1402 					m = m_pullup(m,
1403 					    sizeof(struct ether_vlan_header));
1404 				if (m == NULL) {
1405 					printf("%s: unable to pullup encap "
1406 					    "header", p->if_xname);
1407 					if_statinc(ifp, if_oerrors);
1408 					continue;
1409 				}
1410 
1411 				/*
1412 				 * Transform the Ethernet header into an
1413 				 * Ethernet header with 802.1Q encapsulation.
1414 				 */
1415 				memmove(mtod(m, void *),
1416 				    mtod(m, char *) + mib->ifvm_encaplen,
1417 				    sizeof(struct ether_header));
1418 				evl = mtod(m, struct ether_vlan_header *);
1419 				evl->evl_proto = evl->evl_encap_proto;
1420 				evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1421 				evl->evl_tag = htons(mib->ifvm_tag);
1422 
1423 				/*
1424 				 * To cater for VLAN-aware layer 2 ethernet
1425 				 * switches which may need to strip the tag
1426 				 * before forwarding the packet, make sure
1427 				 * the packet+tag is at least 68 bytes long.
1428 				 * This is necessary because our parent will
1429 				 * only pad to 64 bytes (ETHER_MIN_LEN) and
1430 				 * some switches will not pad by themselves
1431 				 * after deleting a tag.
1432 				 */
1433 				const size_t min_data_len = ETHER_MIN_LEN -
1434 				    ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
1435 				if (m->m_pkthdr.len < min_data_len) {
1436 					m_copyback(m, m->m_pkthdr.len,
1437 					    min_data_len - m->m_pkthdr.len,
1438 					    vlan_zero_pad_buff);
1439 				}
1440 				break;
1441 			    }
1442 
1443 			default:
1444 				panic("%s: impossible", __func__);
1445 			}
1446 		}
1447 
1448 		if ((p->if_flags & IFF_RUNNING) == 0) {
1449 			m_freem(m);
1450 			continue;
1451 		}
1452 
1453 		error = if_transmit_lock(p, m);
1454 		if (error) {
1455 			/* mbuf is already freed */
1456 			if_statinc(ifp, if_oerrors);
1457 			continue;
1458 		}
1459 		if_statinc(ifp, if_opackets);
1460 	}
1461 
1462 	ifp->if_flags &= ~IFF_OACTIVE;
1463 
1464 	/* Remove reference to mib before release */
1465 	vlan_putref_linkmib(mib, &psref);
1466 }
1467 
1468 static int
1469 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1470 {
1471 	struct ifvlan *ifv = ifp->if_softc;
1472 	struct ifnet *p;
1473 	struct ethercom *ec;
1474 	struct ifvlan_linkmib *mib;
1475 	struct psref psref;
1476 	int error;
1477 	size_t pktlen = m->m_pkthdr.len;
1478 	bool mcast = (m->m_flags & M_MCAST) != 0;
1479 
1480 	mib = vlan_getref_linkmib(ifv, &psref);
1481 	if (mib == NULL) {
1482 		m_freem(m);
1483 		return ENETDOWN;
1484 	}
1485 
1486 	p = mib->ifvm_p;
1487 	ec = (void *)mib->ifvm_p;
1488 
1489 	bpf_mtap(ifp, m, BPF_D_OUT);
1490 
1491 	if ((error = pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_OUT)) != 0)
1492 		goto out;
1493 	if (m == NULL)
1494 		goto out;
1495 
1496 	/*
1497 	 * If the parent can insert the tag itself, just mark
1498 	 * the tag in the mbuf header.
1499 	 */
1500 	if (ec->ec_capenable & ETHERCAP_VLAN_HWTAGGING) {
1501 		vlan_set_tag(m, mib->ifvm_tag);
1502 	} else {
1503 		/*
1504 		 * insert the tag ourselves
1505 		 */
1506 		M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1507 		if (m == NULL) {
1508 			printf("%s: unable to prepend encap header",
1509 			    p->if_xname);
1510 			if_statinc(ifp, if_oerrors);
1511 			error = ENOBUFS;
1512 			goto out;
1513 		}
1514 
1515 		switch (p->if_type) {
1516 		case IFT_ETHER:
1517 		    {
1518 			struct ether_vlan_header *evl;
1519 
1520 			if (m->m_len < sizeof(struct ether_vlan_header))
1521 				m = m_pullup(m,
1522 				    sizeof(struct ether_vlan_header));
1523 			if (m == NULL) {
1524 				printf("%s: unable to pullup encap "
1525 				    "header", p->if_xname);
1526 				if_statinc(ifp, if_oerrors);
1527 				error = ENOBUFS;
1528 				goto out;
1529 			}
1530 
1531 			/*
1532 			 * Transform the Ethernet header into an
1533 			 * Ethernet header with 802.1Q encapsulation.
1534 			 */
1535 			memmove(mtod(m, void *),
1536 			    mtod(m, char *) + mib->ifvm_encaplen,
1537 			    sizeof(struct ether_header));
1538 			evl = mtod(m, struct ether_vlan_header *);
1539 			evl->evl_proto = evl->evl_encap_proto;
1540 			evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1541 			evl->evl_tag = htons(mib->ifvm_tag);
1542 
1543 			/*
1544 			 * To cater for VLAN-aware layer 2 ethernet
1545 			 * switches which may need to strip the tag
1546 			 * before forwarding the packet, make sure
1547 			 * the packet+tag is at least 68 bytes long.
1548 			 * This is necessary because our parent will
1549 			 * only pad to 64 bytes (ETHER_MIN_LEN) and
1550 			 * some switches will not pad by themselves
1551 			 * after deleting a tag.
1552 			 */
1553 			const size_t min_data_len = ETHER_MIN_LEN -
1554 			    ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
1555 			if (m->m_pkthdr.len < min_data_len) {
1556 				m_copyback(m, m->m_pkthdr.len,
1557 				    min_data_len - m->m_pkthdr.len,
1558 				    vlan_zero_pad_buff);
1559 			}
1560 			break;
1561 		    }
1562 
1563 		default:
1564 			panic("%s: impossible", __func__);
1565 		}
1566 	}
1567 
1568 	if ((p->if_flags & IFF_RUNNING) == 0) {
1569 		m_freem(m);
1570 		error = ENETDOWN;
1571 		goto out;
1572 	}
1573 
1574 	error = if_transmit_lock(p, m);
1575 	net_stat_ref_t nsr = IF_STAT_GETREF(ifp);
1576 	if (error) {
1577 		/* mbuf is already freed */
1578 		if_statinc_ref(nsr, if_oerrors);
1579 	} else {
1580 		if_statinc_ref(nsr, if_opackets);
1581 		if_statadd_ref(nsr, if_obytes, pktlen);
1582 		if (mcast)
1583 			if_statinc_ref(nsr, if_omcasts);
1584 	}
1585 	IF_STAT_PUTREF(ifp);
1586 
1587 out:
1588 	/* Remove reference to mib before release */
1589 	vlan_putref_linkmib(mib, &psref);
1590 	return error;
1591 }
1592 
1593 /*
1594  * Given an Ethernet frame, find a valid vlan interface corresponding to the
1595  * given source interface and tag, then run the real packet through the
1596  * parent's input routine.
1597  */
1598 void
1599 vlan_input(struct ifnet *ifp, struct mbuf *m)
1600 {
1601 	struct ifvlan *ifv;
1602 	uint16_t vid;
1603 	struct ifvlan_linkmib *mib;
1604 	struct psref psref;
1605 	bool have_vtag;
1606 
1607 	have_vtag = vlan_has_tag(m);
1608 	if (have_vtag) {
1609 		vid = EVL_VLANOFTAG(vlan_get_tag(m));
1610 		m->m_flags &= ~M_VLANTAG;
1611 	} else {
1612 		struct ether_vlan_header *evl;
1613 
1614 		if (ifp->if_type != IFT_ETHER) {
1615 			panic("%s: impossible", __func__);
1616 		}
1617 
1618 		if (m->m_len < sizeof(struct ether_vlan_header) &&
1619 		    (m = m_pullup(m,
1620 		     sizeof(struct ether_vlan_header))) == NULL) {
1621 			printf("%s: no memory for VLAN header, "
1622 			    "dropping packet.\n", ifp->if_xname);
1623 			return;
1624 		}
1625 		evl = mtod(m, struct ether_vlan_header *);
1626 		KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN);
1627 
1628 		vid = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1629 
1630 		/*
1631 		 * Restore the original ethertype.  We'll remove
1632 		 * the encapsulation after we've found the vlan
1633 		 * interface corresponding to the tag.
1634 		 */
1635 		evl->evl_encap_proto = evl->evl_proto;
1636 	}
1637 
1638 	mib = vlan_lookup_tag_psref(ifp, vid, &psref);
1639 	if (mib == NULL) {
1640 		m_freem(m);
1641 		if_statinc(ifp, if_noproto);
1642 		return;
1643 	}
1644 	KASSERT(mib->ifvm_encaplen == ETHER_VLAN_ENCAP_LEN);
1645 
1646 	ifv = mib->ifvm_ifvlan;
1647 	if ((ifv->ifv_if.if_flags & (IFF_UP | IFF_RUNNING)) !=
1648 	    (IFF_UP | IFF_RUNNING)) {
1649 		m_freem(m);
1650 		if_statinc(ifp, if_noproto);
1651 		goto out;
1652 	}
1653 
1654 	/*
1655 	 * Now, remove the encapsulation header.  The original
1656 	 * header has already been fixed up above.
1657 	 */
1658 	if (!have_vtag) {
1659 		memmove(mtod(m, char *) + mib->ifvm_encaplen,
1660 		    mtod(m, void *), sizeof(struct ether_header));
1661 		m_adj(m, mib->ifvm_encaplen);
1662 	}
1663 
1664 	m_set_rcvif(m, &ifv->ifv_if);
1665 
1666 	if (pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_IN) != 0)
1667 		goto out;
1668 	if (m == NULL)
1669 		goto out;
1670 
1671 	m->m_flags &= ~M_PROMISC;
1672 	if_input(&ifv->ifv_if, m);
1673 out:
1674 	vlan_putref_linkmib(mib, &psref);
1675 }
1676 
1677 /*
1678  * Module infrastructure
1679  */
1680 #include "if_module.h"
1681 
1682 IF_MODULE(MODULE_CLASS_DRIVER, vlan, NULL)
1683