xref: /netbsd-src/sys/net/if_vlan.c (revision f3cfa6f6ce31685c6c4a758bc430e69eb99f50a4)
1 /*	$NetBSD: if_vlan.c,v 1.136 2019/05/15 02:59:18 ozaki-r 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 1.136 2019/05/15 02:59:18 ozaki-r 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 #endif
123 
124 #include "ioconf.h"
125 
126 struct vlan_mc_entry {
127 	LIST_ENTRY(vlan_mc_entry)	mc_entries;
128 	/*
129 	 * A key to identify this entry.  The mc_addr below can't be
130 	 * used since multiple sockaddr may mapped into the same
131 	 * ether_multi (e.g., AF_UNSPEC).
132 	 */
133 	union {
134 		struct ether_multi	*mcu_enm;
135 	} mc_u;
136 	struct sockaddr_storage		mc_addr;
137 };
138 
139 #define	mc_enm		mc_u.mcu_enm
140 
141 
142 struct ifvlan_linkmib {
143 	struct ifvlan *ifvm_ifvlan;
144 	const struct vlan_multisw *ifvm_msw;
145 	int	ifvm_encaplen;	/* encapsulation length */
146 	int	ifvm_mtufudge;	/* MTU fudged by this much */
147 	int	ifvm_mintu;	/* min transmission unit */
148 	uint16_t ifvm_proto;	/* encapsulation ethertype */
149 	uint16_t ifvm_tag;	/* tag to apply on packets */
150 	struct ifnet *ifvm_p;		/* parent interface of this vlan */
151 
152 	struct psref_target ifvm_psref;
153 };
154 
155 struct ifvlan {
156 	union {
157 		struct ethercom ifvu_ec;
158 	} ifv_u;
159 	struct ifvlan_linkmib *ifv_mib;	/*
160 					 * reader must use vlan_getref_linkmib()
161 					 * instead of direct dereference
162 					 */
163 	kmutex_t ifv_lock;		/* writer lock for ifv_mib */
164 	pserialize_t ifv_psz;
165 
166 	LIST_HEAD(__vlan_mchead, vlan_mc_entry) ifv_mc_listhead;
167 	LIST_ENTRY(ifvlan) ifv_list;
168 	struct pslist_entry ifv_hash;
169 	int ifv_flags;
170 };
171 
172 #define	IFVF_PROMISC	0x01		/* promiscuous mode enabled */
173 
174 #define	ifv_ec		ifv_u.ifvu_ec
175 
176 #define	ifv_if		ifv_ec.ec_if
177 
178 #define	ifv_msw		ifv_mib.ifvm_msw
179 #define	ifv_encaplen	ifv_mib.ifvm_encaplen
180 #define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
181 #define	ifv_mintu	ifv_mib.ifvm_mintu
182 #define	ifv_tag		ifv_mib.ifvm_tag
183 
184 struct vlan_multisw {
185 	int	(*vmsw_addmulti)(struct ifvlan *, struct ifreq *);
186 	int	(*vmsw_delmulti)(struct ifvlan *, struct ifreq *);
187 	void	(*vmsw_purgemulti)(struct ifvlan *);
188 };
189 
190 static int	vlan_ether_addmulti(struct ifvlan *, struct ifreq *);
191 static int	vlan_ether_delmulti(struct ifvlan *, struct ifreq *);
192 static void	vlan_ether_purgemulti(struct ifvlan *);
193 
194 const struct vlan_multisw vlan_ether_multisw = {
195 	.vmsw_addmulti = vlan_ether_addmulti,
196 	.vmsw_delmulti = vlan_ether_delmulti,
197 	.vmsw_purgemulti = vlan_ether_purgemulti,
198 };
199 
200 static int	vlan_clone_create(struct if_clone *, int);
201 static int	vlan_clone_destroy(struct ifnet *);
202 static int	vlan_config(struct ifvlan *, struct ifnet *,
203     uint16_t);
204 static int	vlan_ioctl(struct ifnet *, u_long, void *);
205 static void	vlan_start(struct ifnet *);
206 static int	vlan_transmit(struct ifnet *, struct mbuf *);
207 static void	vlan_unconfig(struct ifnet *);
208 static int	vlan_unconfig_locked(struct ifvlan *,
209     struct ifvlan_linkmib *);
210 static void	vlan_hash_init(void);
211 static int	vlan_hash_fini(void);
212 static int	vlan_tag_hash(uint16_t, u_long);
213 static struct ifvlan_linkmib*	vlan_getref_linkmib(struct ifvlan *,
214     struct psref *);
215 static void	vlan_putref_linkmib(struct ifvlan_linkmib *,
216     struct psref *);
217 static void	vlan_linkmib_update(struct ifvlan *,
218     struct ifvlan_linkmib *);
219 static struct ifvlan_linkmib*	vlan_lookup_tag_psref(struct ifnet *,
220     uint16_t, struct psref *);
221 
222 LIST_HEAD(vlan_ifvlist, ifvlan);
223 static struct {
224 	kmutex_t lock;
225 	struct vlan_ifvlist list;
226 } ifv_list __cacheline_aligned;
227 
228 
229 #if !defined(VLAN_TAG_HASH_SIZE)
230 #define VLAN_TAG_HASH_SIZE 32
231 #endif
232 static struct {
233 	kmutex_t lock;
234 	struct pslist_head *lists;
235 	u_long mask;
236 } ifv_hash __cacheline_aligned = {
237 	.lists = NULL,
238 	.mask = 0,
239 };
240 
241 pserialize_t vlan_psz __read_mostly;
242 static struct psref_class *ifvm_psref_class __read_mostly;
243 
244 struct if_clone vlan_cloner =
245     IF_CLONE_INITIALIZER("vlan", vlan_clone_create, vlan_clone_destroy);
246 
247 /* Used to pad ethernet frames with < ETHER_MIN_LEN bytes */
248 static char vlan_zero_pad_buff[ETHER_MIN_LEN];
249 
250 static inline int
251 vlan_safe_ifpromisc(struct ifnet *ifp, int pswitch)
252 {
253 	int e;
254 
255 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
256 	e = ifpromisc(ifp, pswitch);
257 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
258 
259 	return e;
260 }
261 
262 static inline int
263 vlan_safe_ifpromisc_locked(struct ifnet *ifp, int pswitch)
264 {
265 	int e;
266 
267 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
268 	e = ifpromisc_locked(ifp, pswitch);
269 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
270 
271 	return e;
272 }
273 
274 void
275 vlanattach(int n)
276 {
277 
278 	/*
279 	 * Nothing to do here, initialization is handled by the
280 	 * module initialization code in vlaninit() below.
281 	 */
282 }
283 
284 static void
285 vlaninit(void)
286 {
287 	mutex_init(&ifv_list.lock, MUTEX_DEFAULT, IPL_NONE);
288 	LIST_INIT(&ifv_list.list);
289 
290 	mutex_init(&ifv_hash.lock, MUTEX_DEFAULT, IPL_NONE);
291 	vlan_psz = pserialize_create();
292 	ifvm_psref_class = psref_class_create("vlanlinkmib", IPL_SOFTNET);
293 	if_clone_attach(&vlan_cloner);
294 
295 	vlan_hash_init();
296 	MODULE_HOOK_SET(if_vlan_vlan_input_hook, "vlan_inp", vlan_input);
297 }
298 
299 static int
300 vlandetach(void)
301 {
302 	bool is_empty;
303 	int error;
304 
305 	mutex_enter(&ifv_list.lock);
306 	is_empty = LIST_EMPTY(&ifv_list.list);
307 	mutex_exit(&ifv_list.lock);
308 
309 	if (!is_empty)
310 		return EBUSY;
311 
312 	error = vlan_hash_fini();
313 	if (error != 0)
314 		return error;
315 
316 	if_clone_detach(&vlan_cloner);
317 	psref_class_destroy(ifvm_psref_class);
318 	pserialize_destroy(vlan_psz);
319 	mutex_destroy(&ifv_hash.lock);
320 	mutex_destroy(&ifv_list.lock);
321 
322 	MODULE_HOOK_UNSET(if_vlan_vlan_input_hook);
323 	return 0;
324 }
325 
326 static void
327 vlan_reset_linkname(struct ifnet *ifp)
328 {
329 
330 	/*
331 	 * We start out with a "802.1Q VLAN" type and zero-length
332 	 * addresses.  When we attach to a parent interface, we
333 	 * inherit its type, address length, address, and data link
334 	 * type.
335 	 */
336 
337 	ifp->if_type = IFT_L2VLAN;
338 	ifp->if_addrlen = 0;
339 	ifp->if_dlt = DLT_NULL;
340 	if_alloc_sadl(ifp);
341 }
342 
343 static int
344 vlan_clone_create(struct if_clone *ifc, int unit)
345 {
346 	struct ifvlan *ifv;
347 	struct ifnet *ifp;
348 	struct ifvlan_linkmib *mib;
349 	int rv;
350 
351 	ifv = malloc(sizeof(struct ifvlan), M_DEVBUF, M_WAITOK|M_ZERO);
352 	mib = kmem_zalloc(sizeof(struct ifvlan_linkmib), KM_SLEEP);
353 	ifp = &ifv->ifv_if;
354 	LIST_INIT(&ifv->ifv_mc_listhead);
355 
356 	mib->ifvm_ifvlan = ifv;
357 	mib->ifvm_p = NULL;
358 	psref_target_init(&mib->ifvm_psref, ifvm_psref_class);
359 
360 	mutex_init(&ifv->ifv_lock, MUTEX_DEFAULT, IPL_NONE);
361 	ifv->ifv_psz = pserialize_create();
362 	ifv->ifv_mib = mib;
363 
364 	mutex_enter(&ifv_list.lock);
365 	LIST_INSERT_HEAD(&ifv_list.list, ifv, ifv_list);
366 	mutex_exit(&ifv_list.lock);
367 
368 	if_initname(ifp, ifc->ifc_name, unit);
369 	ifp->if_softc = ifv;
370 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
371 	ifp->if_extflags = IFEF_NO_LINK_STATE_CHANGE;
372 #ifdef NET_MPSAFE
373 	ifp->if_extflags |= IFEF_MPSAFE;
374 #endif
375 	ifp->if_start = vlan_start;
376 	ifp->if_transmit = vlan_transmit;
377 	ifp->if_ioctl = vlan_ioctl;
378 	IFQ_SET_READY(&ifp->if_snd);
379 
380 	rv = if_initialize(ifp);
381 	if (rv != 0) {
382 		aprint_error("%s: if_initialize failed(%d)\n", ifp->if_xname,
383 		    rv);
384 		goto fail;
385 	}
386 
387 	vlan_reset_linkname(ifp);
388 	if_register(ifp);
389 	return 0;
390 
391 fail:
392 	mutex_enter(&ifv_list.lock);
393 	LIST_REMOVE(ifv, ifv_list);
394 	mutex_exit(&ifv_list.lock);
395 
396 	mutex_destroy(&ifv->ifv_lock);
397 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
398 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
399 	free(ifv, M_DEVBUF);
400 
401 	return rv;
402 }
403 
404 static int
405 vlan_clone_destroy(struct ifnet *ifp)
406 {
407 	struct ifvlan *ifv = ifp->if_softc;
408 
409 	mutex_enter(&ifv_list.lock);
410 	LIST_REMOVE(ifv, ifv_list);
411 	mutex_exit(&ifv_list.lock);
412 
413 	IFNET_LOCK(ifp);
414 	vlan_unconfig(ifp);
415 	IFNET_UNLOCK(ifp);
416 	if_detach(ifp);
417 
418 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
419 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
420 	pserialize_destroy(ifv->ifv_psz);
421 	mutex_destroy(&ifv->ifv_lock);
422 	free(ifv, M_DEVBUF);
423 
424 	return 0;
425 }
426 
427 /*
428  * Configure a VLAN interface.
429  */
430 static int
431 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
432 {
433 	struct ifnet *ifp = &ifv->ifv_if;
434 	struct ifvlan_linkmib *nmib = NULL;
435 	struct ifvlan_linkmib *omib = NULL;
436 	struct ifvlan_linkmib *checkmib;
437 	struct psref_target *nmib_psref = NULL;
438 	const uint16_t vid = EVL_VLANOFTAG(tag);
439 	int error = 0;
440 	int idx;
441 	bool omib_cleanup = false;
442 	struct psref psref;
443 
444 	/* VLAN ID 0 and 4095 are reserved in the spec */
445 	if ((vid == 0) || (vid == 0xfff))
446 		return EINVAL;
447 
448 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
449 	mutex_enter(&ifv->ifv_lock);
450 	omib = ifv->ifv_mib;
451 
452 	if (omib->ifvm_p != NULL) {
453 		error = EBUSY;
454 		goto done;
455 	}
456 
457 	/* Duplicate check */
458 	checkmib = vlan_lookup_tag_psref(p, vid, &psref);
459 	if (checkmib != NULL) {
460 		vlan_putref_linkmib(checkmib, &psref);
461 		error = EEXIST;
462 		goto done;
463 	}
464 
465 	*nmib = *omib;
466 	nmib_psref = &nmib->ifvm_psref;
467 
468 	psref_target_init(nmib_psref, ifvm_psref_class);
469 
470 	switch (p->if_type) {
471 	case IFT_ETHER:
472 	    {
473 		struct ethercom *ec = (void *)p;
474 		nmib->ifvm_msw = &vlan_ether_multisw;
475 		nmib->ifvm_encaplen = ETHER_VLAN_ENCAP_LEN;
476 		nmib->ifvm_mintu = ETHERMIN;
477 
478 		if (ec->ec_nvlans++ == 0) {
479 			IFNET_LOCK(p);
480 			error = ether_enable_vlan_mtu(p);
481 			IFNET_UNLOCK(p);
482 			if (error >= 0) {
483 				if (error) {
484 					ec->ec_nvlans--;
485 					goto done;
486 				}
487 				nmib->ifvm_mtufudge = 0;
488 			} else {
489 				/*
490 				 * Fudge the MTU by the encapsulation size. This
491 				 * makes us incompatible with strictly compliant
492 				 * 802.1Q implementations, but allows us to use
493 				 * the feature with other NetBSD
494 				 * implementations, which might still be useful.
495 				 */
496 				nmib->ifvm_mtufudge = nmib->ifvm_encaplen;
497 			}
498 			error = 0;
499 		}
500 
501 		/*
502 		 * If the parent interface can do hardware-assisted
503 		 * VLAN encapsulation, then propagate its hardware-
504 		 * assisted checksumming flags and tcp segmentation
505 		 * offload.
506 		 */
507 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
508 			ec->ec_capenable |= ETHERCAP_VLAN_HWTAGGING;
509 			ifp->if_capabilities = p->if_capabilities &
510 			    (IFCAP_TSOv4 | IFCAP_TSOv6 |
511 			     IFCAP_CSUM_IPv4_Tx|IFCAP_CSUM_IPv4_Rx|
512 			     IFCAP_CSUM_TCPv4_Tx|IFCAP_CSUM_TCPv4_Rx|
513 			     IFCAP_CSUM_UDPv4_Tx|IFCAP_CSUM_UDPv4_Rx|
514 			     IFCAP_CSUM_TCPv6_Tx|IFCAP_CSUM_TCPv6_Rx|
515 			     IFCAP_CSUM_UDPv6_Tx|IFCAP_CSUM_UDPv6_Rx);
516 		}
517 
518 		/*
519 		 * We inherit the parent's Ethernet address.
520 		 */
521 		ether_ifattach(ifp, CLLADDR(p->if_sadl));
522 		ifp->if_hdrlen = sizeof(struct ether_vlan_header); /* XXX? */
523 		break;
524 	    }
525 
526 	default:
527 		error = EPROTONOSUPPORT;
528 		goto done;
529 	}
530 
531 	nmib->ifvm_p = p;
532 	nmib->ifvm_tag = vid;
533 	ifv->ifv_if.if_mtu = p->if_mtu - nmib->ifvm_mtufudge;
534 	ifv->ifv_if.if_flags = p->if_flags &
535 	    (IFF_UP | IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
536 
537 	/*
538 	 * Inherit the if_type from the parent.  This allows us
539 	 * to participate in bridges of that type.
540 	 */
541 	ifv->ifv_if.if_type = p->if_type;
542 
543 	PSLIST_ENTRY_INIT(ifv, ifv_hash);
544 	idx = vlan_tag_hash(vid, ifv_hash.mask);
545 
546 	mutex_enter(&ifv_hash.lock);
547 	PSLIST_WRITER_INSERT_HEAD(&ifv_hash.lists[idx], ifv, ifv_hash);
548 	mutex_exit(&ifv_hash.lock);
549 
550 	vlan_linkmib_update(ifv, nmib);
551 	nmib = NULL;
552 	nmib_psref = NULL;
553 	omib_cleanup = true;
554 
555 done:
556 	mutex_exit(&ifv->ifv_lock);
557 
558 	if (nmib_psref)
559 		psref_target_destroy(nmib_psref, ifvm_psref_class);
560 	if (nmib)
561 		kmem_free(nmib, sizeof(*nmib));
562 	if (omib_cleanup)
563 		kmem_free(omib, sizeof(*omib));
564 
565 	return error;
566 }
567 
568 /*
569  * Unconfigure a VLAN interface.
570  */
571 static void
572 vlan_unconfig(struct ifnet *ifp)
573 {
574 	struct ifvlan *ifv = ifp->if_softc;
575 	struct ifvlan_linkmib *nmib = NULL;
576 	int error;
577 
578 	KASSERT(IFNET_LOCKED(ifp));
579 
580 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
581 
582 	mutex_enter(&ifv->ifv_lock);
583 	error = vlan_unconfig_locked(ifv, nmib);
584 	mutex_exit(&ifv->ifv_lock);
585 
586 	if (error)
587 		kmem_free(nmib, sizeof(*nmib));
588 }
589 static int
590 vlan_unconfig_locked(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
591 {
592 	struct ifnet *p;
593 	struct ifnet *ifp = &ifv->ifv_if;
594 	struct psref_target *nmib_psref = NULL;
595 	struct ifvlan_linkmib *omib;
596 	int error = 0;
597 
598 	KASSERT(IFNET_LOCKED(ifp));
599 	KASSERT(mutex_owned(&ifv->ifv_lock));
600 
601 	ifp->if_flags &= ~(IFF_UP|IFF_RUNNING);
602 
603 	omib = ifv->ifv_mib;
604 	p = omib->ifvm_p;
605 
606 	if (p == NULL) {
607 		error = -1;
608 		goto done;
609 	}
610 
611 	*nmib = *omib;
612 	nmib_psref = &nmib->ifvm_psref;
613 	psref_target_init(nmib_psref, ifvm_psref_class);
614 
615 	/*
616  	 * Since the interface is being unconfigured, we need to empty the
617 	 * list of multicast groups that we may have joined while we were
618 	 * alive and remove them from the parent's list also.
619 	 */
620 	(*nmib->ifvm_msw->vmsw_purgemulti)(ifv);
621 
622 	/* Disconnect from parent. */
623 	switch (p->if_type) {
624 	case IFT_ETHER:
625 	    {
626 		struct ethercom *ec = (void *)p;
627 		if (--ec->ec_nvlans == 0) {
628 			IFNET_LOCK(p);
629 			(void) ether_disable_vlan_mtu(p);
630 			IFNET_UNLOCK(p);
631 		}
632 
633 		/* XXX ether_ifdetach must not be called with IFNET_LOCK */
634 		mutex_exit(&ifv->ifv_lock);
635 		IFNET_UNLOCK(ifp);
636 		ether_ifdetach(ifp);
637 		IFNET_LOCK(ifp);
638 		mutex_enter(&ifv->ifv_lock);
639 
640 		/* if_free_sadl must be called with IFNET_LOCK */
641 		if_free_sadl(ifp, 1);
642 
643 		/* Restore vlan_ioctl overwritten by ether_ifdetach */
644 		ifp->if_ioctl = vlan_ioctl;
645 		vlan_reset_linkname(ifp);
646 		break;
647 	    }
648 
649 	default:
650 		panic("%s: impossible", __func__);
651 	}
652 
653 	nmib->ifvm_p = NULL;
654 	ifv->ifv_if.if_mtu = 0;
655 	ifv->ifv_flags = 0;
656 
657 	mutex_enter(&ifv_hash.lock);
658 	PSLIST_WRITER_REMOVE(ifv, ifv_hash);
659 	pserialize_perform(vlan_psz);
660 	mutex_exit(&ifv_hash.lock);
661 	PSLIST_ENTRY_DESTROY(ifv, ifv_hash);
662 
663 	vlan_linkmib_update(ifv, nmib);
664 
665 	mutex_exit(&ifv->ifv_lock);
666 
667 	nmib_psref = NULL;
668 	kmem_free(omib, sizeof(*omib));
669 
670 #ifdef INET6
671 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
672 	/* To delete v6 link local addresses */
673 	if (in6_present)
674 		in6_ifdetach(ifp);
675 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
676 #endif
677 
678 	if ((ifp->if_flags & IFF_PROMISC) != 0)
679 		vlan_safe_ifpromisc_locked(ifp, 0);
680 	if_down_locked(ifp);
681 	ifp->if_capabilities = 0;
682 	mutex_enter(&ifv->ifv_lock);
683 done:
684 
685 	if (nmib_psref)
686 		psref_target_destroy(nmib_psref, ifvm_psref_class);
687 
688 	return error;
689 }
690 
691 static void
692 vlan_hash_init(void)
693 {
694 
695 	ifv_hash.lists = hashinit(VLAN_TAG_HASH_SIZE, HASH_PSLIST, true,
696 	    &ifv_hash.mask);
697 }
698 
699 static int
700 vlan_hash_fini(void)
701 {
702 	int i;
703 
704 	mutex_enter(&ifv_hash.lock);
705 
706 	for (i = 0; i < ifv_hash.mask + 1; i++) {
707 		if (PSLIST_WRITER_FIRST(&ifv_hash.lists[i], struct ifvlan,
708 		    ifv_hash) != NULL) {
709 			mutex_exit(&ifv_hash.lock);
710 			return EBUSY;
711 		}
712 	}
713 
714 	for (i = 0; i < ifv_hash.mask + 1; i++)
715 		PSLIST_DESTROY(&ifv_hash.lists[i]);
716 
717 	mutex_exit(&ifv_hash.lock);
718 
719 	hashdone(ifv_hash.lists, HASH_PSLIST, ifv_hash.mask);
720 
721 	ifv_hash.lists = NULL;
722 	ifv_hash.mask = 0;
723 
724 	return 0;
725 }
726 
727 static int
728 vlan_tag_hash(uint16_t tag, u_long mask)
729 {
730 	uint32_t hash;
731 
732 	hash = (tag >> 8) ^ tag;
733 	hash = (hash >> 2) ^ hash;
734 
735 	return hash & mask;
736 }
737 
738 static struct ifvlan_linkmib *
739 vlan_getref_linkmib(struct ifvlan *sc, struct psref *psref)
740 {
741 	struct ifvlan_linkmib *mib;
742 	int s;
743 
744 	s = pserialize_read_enter();
745 	mib = sc->ifv_mib;
746 	if (mib == NULL) {
747 		pserialize_read_exit(s);
748 		return NULL;
749 	}
750 	membar_datadep_consumer();
751 	psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
752 	pserialize_read_exit(s);
753 
754 	return mib;
755 }
756 
757 static void
758 vlan_putref_linkmib(struct ifvlan_linkmib *mib, struct psref *psref)
759 {
760 	if (mib == NULL)
761 		return;
762 	psref_release(psref, &mib->ifvm_psref, ifvm_psref_class);
763 }
764 
765 static struct ifvlan_linkmib *
766 vlan_lookup_tag_psref(struct ifnet *ifp, uint16_t tag, struct psref *psref)
767 {
768 	int idx;
769 	int s;
770 	struct ifvlan *sc;
771 
772 	idx = vlan_tag_hash(tag, ifv_hash.mask);
773 
774 	s = pserialize_read_enter();
775 	PSLIST_READER_FOREACH(sc, &ifv_hash.lists[idx], struct ifvlan,
776 	    ifv_hash) {
777 		struct ifvlan_linkmib *mib = sc->ifv_mib;
778 		if (mib == NULL)
779 			continue;
780 		if (mib->ifvm_tag != tag)
781 			continue;
782 		if (mib->ifvm_p != ifp)
783 			continue;
784 
785 		psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
786 		pserialize_read_exit(s);
787 		return mib;
788 	}
789 	pserialize_read_exit(s);
790 	return NULL;
791 }
792 
793 static void
794 vlan_linkmib_update(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
795 {
796 	struct ifvlan_linkmib *omib = ifv->ifv_mib;
797 
798 	KASSERT(mutex_owned(&ifv->ifv_lock));
799 
800 	membar_producer();
801 	ifv->ifv_mib = nmib;
802 
803 	pserialize_perform(ifv->ifv_psz);
804 	psref_target_destroy(&omib->ifvm_psref, ifvm_psref_class);
805 }
806 
807 /*
808  * Called when a parent interface is detaching; destroy any VLAN
809  * configuration for the parent interface.
810  */
811 void
812 vlan_ifdetach(struct ifnet *p)
813 {
814 	struct ifvlan *ifv;
815 	struct ifvlan_linkmib *mib, **nmibs;
816 	struct psref psref;
817 	int error;
818 	int bound;
819 	int i, cnt = 0;
820 
821 	bound = curlwp_bind();
822 
823 	mutex_enter(&ifv_list.lock);
824 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
825 		mib = vlan_getref_linkmib(ifv, &psref);
826 		if (mib == NULL)
827 			continue;
828 
829 		if (mib->ifvm_p == p)
830 			cnt++;
831 
832 		vlan_putref_linkmib(mib, &psref);
833 	}
834 	mutex_exit(&ifv_list.lock);
835 
836 	if (cnt == 0) {
837 		curlwp_bindx(bound);
838 		return;
839 	}
840 
841 	/*
842 	 * The value of "cnt" does not increase while ifv_list.lock
843 	 * and ifv->ifv_lock are released here, because the parent
844 	 * interface is detaching.
845 	 */
846 	nmibs = kmem_alloc(sizeof(*nmibs) * cnt, KM_SLEEP);
847 	for (i = 0; i < cnt; i++) {
848 		nmibs[i] = kmem_alloc(sizeof(*nmibs[i]), KM_SLEEP);
849 	}
850 
851 	mutex_enter(&ifv_list.lock);
852 
853 	i = 0;
854 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
855 		struct ifnet *ifp = &ifv->ifv_if;
856 
857 		/* IFNET_LOCK must be held before ifv_lock. */
858 		IFNET_LOCK(ifp);
859 		mutex_enter(&ifv->ifv_lock);
860 
861 		/* XXX ifv_mib = NULL? */
862 		if (ifv->ifv_mib->ifvm_p == p) {
863 			KASSERTMSG(i < cnt, "no memory for unconfig, parent=%s",
864 			    p->if_xname);
865 			error = vlan_unconfig_locked(ifv, nmibs[i]);
866 			if (!error) {
867 				nmibs[i] = NULL;
868 				i++;
869 			}
870 
871 		}
872 
873 		mutex_exit(&ifv->ifv_lock);
874 		IFNET_UNLOCK(ifp);
875 	}
876 
877 	mutex_exit(&ifv_list.lock);
878 
879 	curlwp_bindx(bound);
880 
881 	for (i = 0; i < cnt; i++) {
882 		if (nmibs[i])
883 			kmem_free(nmibs[i], sizeof(*nmibs[i]));
884 	}
885 
886 	kmem_free(nmibs, sizeof(*nmibs) * cnt);
887 
888 	return;
889 }
890 
891 static int
892 vlan_set_promisc(struct ifnet *ifp)
893 {
894 	struct ifvlan *ifv = ifp->if_softc;
895 	struct ifvlan_linkmib *mib;
896 	struct psref psref;
897 	int error = 0;
898 	int bound;
899 
900 	bound = curlwp_bind();
901 	mib = vlan_getref_linkmib(ifv, &psref);
902 	if (mib == NULL) {
903 		curlwp_bindx(bound);
904 		return EBUSY;
905 	}
906 
907 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
908 		if ((ifv->ifv_flags & IFVF_PROMISC) == 0) {
909 			error = vlan_safe_ifpromisc(mib->ifvm_p, 1);
910 			if (error == 0)
911 				ifv->ifv_flags |= IFVF_PROMISC;
912 		}
913 	} else {
914 		if ((ifv->ifv_flags & IFVF_PROMISC) != 0) {
915 			error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
916 			if (error == 0)
917 				ifv->ifv_flags &= ~IFVF_PROMISC;
918 		}
919 	}
920 	vlan_putref_linkmib(mib, &psref);
921 	curlwp_bindx(bound);
922 
923 	return error;
924 }
925 
926 static int
927 vlan_ioctl(struct ifnet *ifp, u_long cmd, void *data)
928 {
929 	struct lwp *l = curlwp;
930 	struct ifvlan *ifv = ifp->if_softc;
931 	struct ifaddr *ifa = (struct ifaddr *) data;
932 	struct ifreq *ifr = (struct ifreq *) data;
933 	struct ifnet *pr;
934 	struct ifcapreq *ifcr;
935 	struct vlanreq vlr;
936 	struct ifvlan_linkmib *mib;
937 	struct psref psref;
938 	int error = 0;
939 	int bound;
940 
941 	switch (cmd) {
942 	case SIOCSIFMTU:
943 		bound = curlwp_bind();
944 		mib = vlan_getref_linkmib(ifv, &psref);
945 		if (mib == NULL) {
946 			curlwp_bindx(bound);
947 			error = EBUSY;
948 			break;
949 		}
950 
951 		if (mib->ifvm_p == NULL) {
952 			vlan_putref_linkmib(mib, &psref);
953 			curlwp_bindx(bound);
954 			error = EINVAL;
955 		} else if (
956 		    ifr->ifr_mtu > (mib->ifvm_p->if_mtu - mib->ifvm_mtufudge) ||
957 		    ifr->ifr_mtu < (mib->ifvm_mintu - mib->ifvm_mtufudge)) {
958 			vlan_putref_linkmib(mib, &psref);
959 			curlwp_bindx(bound);
960 			error = EINVAL;
961 		} else {
962 			vlan_putref_linkmib(mib, &psref);
963 			curlwp_bindx(bound);
964 
965 			error = ifioctl_common(ifp, cmd, data);
966 			if (error == ENETRESET)
967 				error = 0;
968 		}
969 
970 		break;
971 
972 	case SIOCSETVLAN:
973 		if ((error = kauth_authorize_network(l->l_cred,
974 		    KAUTH_NETWORK_INTERFACE,
975 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
976 		    NULL)) != 0)
977 			break;
978 		if ((error = copyin(ifr->ifr_data, &vlr, sizeof(vlr))) != 0)
979 			break;
980 
981 		if (vlr.vlr_parent[0] == '\0') {
982 			bound = curlwp_bind();
983 			mib = vlan_getref_linkmib(ifv, &psref);
984 			if (mib == NULL) {
985 				curlwp_bindx(bound);
986 				error = EBUSY;
987 				break;
988 			}
989 
990 			if (mib->ifvm_p != NULL &&
991 			    (ifp->if_flags & IFF_PROMISC) != 0)
992 				error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
993 
994 			vlan_putref_linkmib(mib, &psref);
995 			curlwp_bindx(bound);
996 
997 			vlan_unconfig(ifp);
998 			break;
999 		}
1000 		if (vlr.vlr_tag != EVL_VLANOFTAG(vlr.vlr_tag)) {
1001 			error = EINVAL;		 /* check for valid tag */
1002 			break;
1003 		}
1004 		if ((pr = ifunit(vlr.vlr_parent)) == NULL) {
1005 			error = ENOENT;
1006 			break;
1007 		}
1008 		error = vlan_config(ifv, pr, vlr.vlr_tag);
1009 		if (error != 0) {
1010 			break;
1011 		}
1012 
1013 		/* Update promiscuous mode, if necessary. */
1014 		vlan_set_promisc(ifp);
1015 
1016 		ifp->if_flags |= IFF_RUNNING;
1017 		break;
1018 
1019 	case SIOCGETVLAN:
1020 		memset(&vlr, 0, sizeof(vlr));
1021 		bound = curlwp_bind();
1022 		mib = vlan_getref_linkmib(ifv, &psref);
1023 		if (mib == NULL) {
1024 			curlwp_bindx(bound);
1025 			error = EBUSY;
1026 			break;
1027 		}
1028 		if (mib->ifvm_p != NULL) {
1029 			snprintf(vlr.vlr_parent, sizeof(vlr.vlr_parent), "%s",
1030 			    mib->ifvm_p->if_xname);
1031 			vlr.vlr_tag = mib->ifvm_tag;
1032 		}
1033 		vlan_putref_linkmib(mib, &psref);
1034 		curlwp_bindx(bound);
1035 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1036 		break;
1037 
1038 	case SIOCSIFFLAGS:
1039 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
1040 			break;
1041 		/*
1042 		 * For promiscuous mode, we enable promiscuous mode on
1043 		 * the parent if we need promiscuous on the VLAN interface.
1044 		 */
1045 		bound = curlwp_bind();
1046 		mib = vlan_getref_linkmib(ifv, &psref);
1047 		if (mib == NULL) {
1048 			curlwp_bindx(bound);
1049 			error = EBUSY;
1050 			break;
1051 		}
1052 
1053 		if (mib->ifvm_p != NULL)
1054 			error = vlan_set_promisc(ifp);
1055 		vlan_putref_linkmib(mib, &psref);
1056 		curlwp_bindx(bound);
1057 		break;
1058 
1059 	case SIOCADDMULTI:
1060 		mutex_enter(&ifv->ifv_lock);
1061 		mib = ifv->ifv_mib;
1062 		if (mib == NULL) {
1063 			error = EBUSY;
1064 			mutex_exit(&ifv->ifv_lock);
1065 			break;
1066 		}
1067 
1068 		error = (mib->ifvm_p != NULL) ?
1069 		    (*mib->ifvm_msw->vmsw_addmulti)(ifv, ifr) : EINVAL;
1070 		mib = NULL;
1071 		mutex_exit(&ifv->ifv_lock);
1072 		break;
1073 
1074 	case SIOCDELMULTI:
1075 		mutex_enter(&ifv->ifv_lock);
1076 		mib = ifv->ifv_mib;
1077 		if (mib == NULL) {
1078 			error = EBUSY;
1079 			mutex_exit(&ifv->ifv_lock);
1080 			break;
1081 		}
1082 		error = (mib->ifvm_p != NULL) ?
1083 		    (*mib->ifvm_msw->vmsw_delmulti)(ifv, ifr) : EINVAL;
1084 		mib = NULL;
1085 		mutex_exit(&ifv->ifv_lock);
1086 		break;
1087 
1088 	case SIOCSIFCAP:
1089 		ifcr = data;
1090 		/* make sure caps are enabled on parent */
1091 		bound = curlwp_bind();
1092 		mib = vlan_getref_linkmib(ifv, &psref);
1093 		if (mib == NULL) {
1094 			curlwp_bindx(bound);
1095 			error = EBUSY;
1096 			break;
1097 		}
1098 
1099 		if (mib->ifvm_p == NULL) {
1100 			vlan_putref_linkmib(mib, &psref);
1101 			curlwp_bindx(bound);
1102 			error = EINVAL;
1103 			break;
1104 		}
1105 		if ((mib->ifvm_p->if_capenable & ifcr->ifcr_capenable) !=
1106 		    ifcr->ifcr_capenable) {
1107 			vlan_putref_linkmib(mib, &psref);
1108 			curlwp_bindx(bound);
1109 			error = EINVAL;
1110 			break;
1111 		}
1112 
1113 		vlan_putref_linkmib(mib, &psref);
1114 		curlwp_bindx(bound);
1115 
1116 		if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
1117 			error = 0;
1118 		break;
1119 	case SIOCINITIFADDR:
1120 		bound = curlwp_bind();
1121 		mib = vlan_getref_linkmib(ifv, &psref);
1122 		if (mib == NULL) {
1123 			curlwp_bindx(bound);
1124 			error = EBUSY;
1125 			break;
1126 		}
1127 
1128 		if (mib->ifvm_p == NULL) {
1129 			error = EINVAL;
1130 			vlan_putref_linkmib(mib, &psref);
1131 			curlwp_bindx(bound);
1132 			break;
1133 		}
1134 		vlan_putref_linkmib(mib, &psref);
1135 		curlwp_bindx(bound);
1136 
1137 		ifp->if_flags |= IFF_UP;
1138 #ifdef INET
1139 		if (ifa->ifa_addr->sa_family == AF_INET)
1140 			arp_ifinit(ifp, ifa);
1141 #endif
1142 		break;
1143 
1144 	default:
1145 		error = ether_ioctl(ifp, cmd, data);
1146 	}
1147 
1148 	return error;
1149 }
1150 
1151 static int
1152 vlan_ether_addmulti(struct ifvlan *ifv, struct ifreq *ifr)
1153 {
1154 	const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr);
1155 	struct vlan_mc_entry *mc;
1156 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1157 	struct ifvlan_linkmib *mib;
1158 	int error;
1159 
1160 	KASSERT(mutex_owned(&ifv->ifv_lock));
1161 
1162 	if (sa->sa_len > sizeof(struct sockaddr_storage))
1163 		return EINVAL;
1164 
1165 	error = ether_addmulti(sa, &ifv->ifv_ec);
1166 	if (error != ENETRESET)
1167 		return error;
1168 
1169 	/*
1170 	 * This is a new multicast address.  We have to tell parent
1171 	 * about it.  Also, remember this multicast address so that
1172 	 * we can delete it on unconfigure.
1173 	 */
1174 	mc = malloc(sizeof(struct vlan_mc_entry), M_DEVBUF, M_NOWAIT);
1175 	if (mc == NULL) {
1176 		error = ENOMEM;
1177 		goto alloc_failed;
1178 	}
1179 
1180 	/*
1181 	 * Since ether_addmulti() returned ENETRESET, the following two
1182 	 * statements shouldn't fail. Here ifv_ec is implicitly protected
1183 	 * by the ifv_lock lock.
1184 	 */
1185 	error = ether_multiaddr(sa, addrlo, addrhi);
1186 	KASSERT(error == 0);
1187 
1188 	ETHER_LOCK(&ifv->ifv_ec);
1189 	mc->mc_enm = ether_lookup_multi(addrlo, addrhi, &ifv->ifv_ec);
1190 	ETHER_UNLOCK(&ifv->ifv_ec);
1191 
1192 	KASSERT(mc->mc_enm != NULL);
1193 
1194 	memcpy(&mc->mc_addr, sa, sa->sa_len);
1195 	LIST_INSERT_HEAD(&ifv->ifv_mc_listhead, mc, mc_entries);
1196 
1197 	mib = ifv->ifv_mib;
1198 
1199 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1200 	error = if_mcast_op(mib->ifvm_p, SIOCADDMULTI, sa);
1201 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1202 
1203 	if (error != 0)
1204 		goto ioctl_failed;
1205 	return error;
1206 
1207 ioctl_failed:
1208 	LIST_REMOVE(mc, mc_entries);
1209 	free(mc, M_DEVBUF);
1210 
1211 alloc_failed:
1212 	(void)ether_delmulti(sa, &ifv->ifv_ec);
1213 	return error;
1214 }
1215 
1216 static int
1217 vlan_ether_delmulti(struct ifvlan *ifv, struct ifreq *ifr)
1218 {
1219 	const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr);
1220 	struct ether_multi *enm;
1221 	struct vlan_mc_entry *mc;
1222 	struct ifvlan_linkmib *mib;
1223 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1224 	int error;
1225 
1226 	KASSERT(mutex_owned(&ifv->ifv_lock));
1227 
1228 	/*
1229 	 * Find a key to lookup vlan_mc_entry.  We have to do this
1230 	 * before calling ether_delmulti for obvious reasons.
1231 	 */
1232 	if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0)
1233 		return error;
1234 
1235 	ETHER_LOCK(&ifv->ifv_ec);
1236 	enm = ether_lookup_multi(addrlo, addrhi, &ifv->ifv_ec);
1237 	ETHER_UNLOCK(&ifv->ifv_ec);
1238 	if (enm == NULL)
1239 		return EINVAL;
1240 
1241 	LIST_FOREACH(mc, &ifv->ifv_mc_listhead, mc_entries) {
1242 		if (mc->mc_enm == enm)
1243 			break;
1244 	}
1245 
1246 	/* We woun't delete entries we didn't add */
1247 	if (mc == NULL)
1248 		return EINVAL;
1249 
1250 	error = ether_delmulti(sa, &ifv->ifv_ec);
1251 	if (error != ENETRESET)
1252 		return error;
1253 
1254 	/* We no longer use this multicast address.  Tell parent so. */
1255 	mib = ifv->ifv_mib;
1256 	error = if_mcast_op(mib->ifvm_p, SIOCDELMULTI, sa);
1257 
1258 	if (error == 0) {
1259 		/* And forget about this address. */
1260 		LIST_REMOVE(mc, mc_entries);
1261 		free(mc, M_DEVBUF);
1262 	} else {
1263 		(void)ether_addmulti(sa, &ifv->ifv_ec);
1264 	}
1265 
1266 	return error;
1267 }
1268 
1269 /*
1270  * Delete any multicast address we have asked to add from parent
1271  * interface.  Called when the vlan is being unconfigured.
1272  */
1273 static void
1274 vlan_ether_purgemulti(struct ifvlan *ifv)
1275 {
1276 	struct vlan_mc_entry *mc;
1277 	struct ifvlan_linkmib *mib;
1278 
1279 	KASSERT(mutex_owned(&ifv->ifv_lock));
1280 	mib = ifv->ifv_mib;
1281 	if (mib == NULL) {
1282 		return;
1283 	}
1284 
1285 	while ((mc = LIST_FIRST(&ifv->ifv_mc_listhead)) != NULL) {
1286 		(void)if_mcast_op(mib->ifvm_p, SIOCDELMULTI,
1287 		    sstocsa(&mc->mc_addr));
1288 		LIST_REMOVE(mc, mc_entries);
1289 		free(mc, M_DEVBUF);
1290 	}
1291 }
1292 
1293 static void
1294 vlan_start(struct ifnet *ifp)
1295 {
1296 	struct ifvlan *ifv = ifp->if_softc;
1297 	struct ifnet *p;
1298 	struct ethercom *ec;
1299 	struct mbuf *m;
1300 	struct ifvlan_linkmib *mib;
1301 	struct psref psref;
1302 	int error;
1303 
1304 	mib = vlan_getref_linkmib(ifv, &psref);
1305 	if (mib == NULL)
1306 		return;
1307 	p = mib->ifvm_p;
1308 	ec = (void *)mib->ifvm_p;
1309 
1310 	ifp->if_flags |= IFF_OACTIVE;
1311 
1312 	for (;;) {
1313 		IFQ_DEQUEUE(&ifp->if_snd, m);
1314 		if (m == NULL)
1315 			break;
1316 
1317 #ifdef ALTQ
1318 		/*
1319 		 * KERNEL_LOCK is required for ALTQ even if NET_MPSAFE is
1320 		 * defined.
1321 		 */
1322 		KERNEL_LOCK(1, NULL);
1323 		/*
1324 		 * If ALTQ is enabled on the parent interface, do
1325 		 * classification; the queueing discipline might
1326 		 * not require classification, but might require
1327 		 * the address family/header pointer in the pktattr.
1328 		 */
1329 		if (ALTQ_IS_ENABLED(&p->if_snd)) {
1330 			switch (p->if_type) {
1331 			case IFT_ETHER:
1332 				altq_etherclassify(&p->if_snd, m);
1333 				break;
1334 			default:
1335 				panic("%s: impossible (altq)", __func__);
1336 			}
1337 		}
1338 		KERNEL_UNLOCK_ONE(NULL);
1339 #endif /* ALTQ */
1340 
1341 		bpf_mtap(ifp, m, BPF_D_OUT);
1342 		/*
1343 		 * If the parent can insert the tag itself, just mark
1344 		 * the tag in the mbuf header.
1345 		 */
1346 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
1347 			vlan_set_tag(m, mib->ifvm_tag);
1348 		} else {
1349 			/*
1350 			 * insert the tag ourselves
1351 			 */
1352 			M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1353 			if (m == NULL) {
1354 				printf("%s: unable to prepend encap header",
1355 				    p->if_xname);
1356 				ifp->if_oerrors++;
1357 				continue;
1358 			}
1359 
1360 			switch (p->if_type) {
1361 			case IFT_ETHER:
1362 			    {
1363 				struct ether_vlan_header *evl;
1364 
1365 				if (m->m_len < sizeof(struct ether_vlan_header))
1366 					m = m_pullup(m,
1367 					    sizeof(struct ether_vlan_header));
1368 				if (m == NULL) {
1369 					printf("%s: unable to pullup encap "
1370 					    "header", p->if_xname);
1371 					ifp->if_oerrors++;
1372 					continue;
1373 				}
1374 
1375 				/*
1376 				 * Transform the Ethernet header into an
1377 				 * Ethernet header with 802.1Q encapsulation.
1378 				 */
1379 				memmove(mtod(m, void *),
1380 				    mtod(m, char *) + mib->ifvm_encaplen,
1381 				    sizeof(struct ether_header));
1382 				evl = mtod(m, struct ether_vlan_header *);
1383 				evl->evl_proto = evl->evl_encap_proto;
1384 				evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1385 				evl->evl_tag = htons(mib->ifvm_tag);
1386 
1387 				/*
1388 				 * To cater for VLAN-aware layer 2 ethernet
1389 				 * switches which may need to strip the tag
1390 				 * before forwarding the packet, make sure
1391 				 * the packet+tag is at least 68 bytes long.
1392 				 * This is necessary because our parent will
1393 				 * only pad to 64 bytes (ETHER_MIN_LEN) and
1394 				 * some switches will not pad by themselves
1395 				 * after deleting a tag.
1396 				 */
1397 				const size_t min_data_len = ETHER_MIN_LEN -
1398 				    ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
1399 				if (m->m_pkthdr.len < min_data_len) {
1400 					m_copyback(m, m->m_pkthdr.len,
1401 					    min_data_len - m->m_pkthdr.len,
1402 					    vlan_zero_pad_buff);
1403 				}
1404 				break;
1405 			    }
1406 
1407 			default:
1408 				panic("%s: impossible", __func__);
1409 			}
1410 		}
1411 
1412 		if ((p->if_flags & IFF_RUNNING) == 0) {
1413 			m_freem(m);
1414 			continue;
1415 		}
1416 
1417 		error = if_transmit_lock(p, m);
1418 		if (error) {
1419 			/* mbuf is already freed */
1420 			ifp->if_oerrors++;
1421 			continue;
1422 		}
1423 		ifp->if_opackets++;
1424 	}
1425 
1426 	ifp->if_flags &= ~IFF_OACTIVE;
1427 
1428 	/* Remove reference to mib before release */
1429 	vlan_putref_linkmib(mib, &psref);
1430 }
1431 
1432 static int
1433 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1434 {
1435 	struct ifvlan *ifv = ifp->if_softc;
1436 	struct ifnet *p;
1437 	struct ethercom *ec;
1438 	struct ifvlan_linkmib *mib;
1439 	struct psref psref;
1440 	int error;
1441 	size_t pktlen = m->m_pkthdr.len;
1442 	bool mcast = (m->m_flags & M_MCAST) != 0;
1443 
1444 	mib = vlan_getref_linkmib(ifv, &psref);
1445 	if (mib == NULL) {
1446 		m_freem(m);
1447 		return ENETDOWN;
1448 	}
1449 
1450 	p = mib->ifvm_p;
1451 	ec = (void *)mib->ifvm_p;
1452 
1453 	bpf_mtap(ifp, m, BPF_D_OUT);
1454 
1455 	if ((error = pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_OUT)) != 0)
1456 		goto out;
1457 	if (m == NULL)
1458 		goto out;
1459 
1460 	/*
1461 	 * If the parent can insert the tag itself, just mark
1462 	 * the tag in the mbuf header.
1463 	 */
1464 	if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
1465 		vlan_set_tag(m, mib->ifvm_tag);
1466 	} else {
1467 		/*
1468 		 * insert the tag ourselves
1469 		 */
1470 		M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1471 		if (m == NULL) {
1472 			printf("%s: unable to prepend encap header",
1473 			    p->if_xname);
1474 			ifp->if_oerrors++;
1475 			error = ENOBUFS;
1476 			goto out;
1477 		}
1478 
1479 		switch (p->if_type) {
1480 		case IFT_ETHER:
1481 		    {
1482 			struct ether_vlan_header *evl;
1483 
1484 			if (m->m_len < sizeof(struct ether_vlan_header))
1485 				m = m_pullup(m,
1486 				    sizeof(struct ether_vlan_header));
1487 			if (m == NULL) {
1488 				printf("%s: unable to pullup encap "
1489 				    "header", p->if_xname);
1490 				ifp->if_oerrors++;
1491 				error = ENOBUFS;
1492 				goto out;
1493 			}
1494 
1495 			/*
1496 			 * Transform the Ethernet header into an
1497 			 * Ethernet header with 802.1Q encapsulation.
1498 			 */
1499 			memmove(mtod(m, void *),
1500 			    mtod(m, char *) + mib->ifvm_encaplen,
1501 			    sizeof(struct ether_header));
1502 			evl = mtod(m, struct ether_vlan_header *);
1503 			evl->evl_proto = evl->evl_encap_proto;
1504 			evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1505 			evl->evl_tag = htons(mib->ifvm_tag);
1506 
1507 			/*
1508 			 * To cater for VLAN-aware layer 2 ethernet
1509 			 * switches which may need to strip the tag
1510 			 * before forwarding the packet, make sure
1511 			 * the packet+tag is at least 68 bytes long.
1512 			 * This is necessary because our parent will
1513 			 * only pad to 64 bytes (ETHER_MIN_LEN) and
1514 			 * some switches will not pad by themselves
1515 			 * after deleting a tag.
1516 			 */
1517 			const size_t min_data_len = ETHER_MIN_LEN -
1518 			    ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
1519 			if (m->m_pkthdr.len < min_data_len) {
1520 				m_copyback(m, m->m_pkthdr.len,
1521 				    min_data_len - m->m_pkthdr.len,
1522 				    vlan_zero_pad_buff);
1523 			}
1524 			break;
1525 		    }
1526 
1527 		default:
1528 			panic("%s: impossible", __func__);
1529 		}
1530 	}
1531 
1532 	if ((p->if_flags & IFF_RUNNING) == 0) {
1533 		m_freem(m);
1534 		error = ENETDOWN;
1535 		goto out;
1536 	}
1537 
1538 	error = if_transmit_lock(p, m);
1539 	if (error) {
1540 		/* mbuf is already freed */
1541 		ifp->if_oerrors++;
1542 	} else {
1543 
1544 		ifp->if_opackets++;
1545 		ifp->if_obytes += pktlen;
1546 		if (mcast)
1547 			ifp->if_omcasts++;
1548 	}
1549 
1550 out:
1551 	/* Remove reference to mib before release */
1552 	vlan_putref_linkmib(mib, &psref);
1553 	return error;
1554 }
1555 
1556 /*
1557  * Given an Ethernet frame, find a valid vlan interface corresponding to the
1558  * given source interface and tag, then run the real packet through the
1559  * parent's input routine.
1560  */
1561 void
1562 vlan_input(struct ifnet *ifp, struct mbuf *m)
1563 {
1564 	struct ifvlan *ifv;
1565 	uint16_t vid;
1566 	struct ifvlan_linkmib *mib;
1567 	struct psref psref;
1568 	bool have_vtag;
1569 
1570 	have_vtag = vlan_has_tag(m);
1571 	if (have_vtag) {
1572 		vid = EVL_VLANOFTAG(vlan_get_tag(m));
1573 		m->m_flags &= ~M_VLANTAG;
1574 	} else {
1575 		struct ether_vlan_header *evl;
1576 
1577 		if (ifp->if_type != IFT_ETHER) {
1578 			panic("%s: impossible", __func__);
1579 		}
1580 
1581 		if (m->m_len < sizeof(struct ether_vlan_header) &&
1582 		    (m = m_pullup(m,
1583 		     sizeof(struct ether_vlan_header))) == NULL) {
1584 			printf("%s: no memory for VLAN header, "
1585 			    "dropping packet.\n", ifp->if_xname);
1586 			return;
1587 		}
1588 		evl = mtod(m, struct ether_vlan_header *);
1589 		KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN);
1590 
1591 		vid = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1592 
1593 		/*
1594 		 * Restore the original ethertype.  We'll remove
1595 		 * the encapsulation after we've found the vlan
1596 		 * interface corresponding to the tag.
1597 		 */
1598 		evl->evl_encap_proto = evl->evl_proto;
1599 	}
1600 
1601 	mib = vlan_lookup_tag_psref(ifp, vid, &psref);
1602 	if (mib == NULL) {
1603 		m_freem(m);
1604 		ifp->if_noproto++;
1605 		return;
1606 	}
1607 	KASSERT(mib->ifvm_encaplen == ETHER_VLAN_ENCAP_LEN);
1608 
1609 	ifv = mib->ifvm_ifvlan;
1610 	if ((ifv->ifv_if.if_flags & (IFF_UP|IFF_RUNNING)) !=
1611 	    (IFF_UP|IFF_RUNNING)) {
1612 		m_freem(m);
1613 		ifp->if_noproto++;
1614 		goto out;
1615 	}
1616 
1617 	/*
1618 	 * Now, remove the encapsulation header.  The original
1619 	 * header has already been fixed up above.
1620 	 */
1621 	if (!have_vtag) {
1622 		memmove(mtod(m, char *) + mib->ifvm_encaplen,
1623 		    mtod(m, void *), sizeof(struct ether_header));
1624 		m_adj(m, mib->ifvm_encaplen);
1625 	}
1626 
1627 	m_set_rcvif(m, &ifv->ifv_if);
1628 	ifv->ifv_if.if_ipackets++;
1629 
1630 	if (pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_IN) != 0)
1631 		goto out;
1632 	if (m == NULL)
1633 		goto out;
1634 
1635 	m->m_flags &= ~M_PROMISC;
1636 	if_input(&ifv->ifv_if, m);
1637 out:
1638 	vlan_putref_linkmib(mib, &psref);
1639 }
1640 
1641 /*
1642  * Module infrastructure
1643  */
1644 #include "if_module.h"
1645 
1646 IF_MODULE(MODULE_CLASS_DRIVER, vlan, NULL)
1647