xref: /netbsd-src/sys/net/if_vlan.c (revision 8ecbf5f02b752fcb7debe1a8fab1dc82602bc760)
1 /*	$NetBSD: if_vlan.c,v 1.152 2020/06/12 11:04:45 roy 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 1.152 2020/06/12 11:04:45 roy 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 	ifv->ifv_if.if_flags = p->if_flags &
545 	    (IFF_UP | IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
546 
547 	/*
548 	 * Inherit the if_type from the parent.  This allows us
549 	 * to participate in bridges of that type.
550 	 */
551 	ifv->ifv_if.if_type = p->if_type;
552 
553 	PSLIST_ENTRY_INIT(ifv, ifv_hash);
554 	idx = vlan_tag_hash(vid, ifv_hash.mask);
555 
556 	mutex_enter(&ifv_hash.lock);
557 	PSLIST_WRITER_INSERT_HEAD(&ifv_hash.lists[idx], ifv, ifv_hash);
558 	mutex_exit(&ifv_hash.lock);
559 
560 	vlan_linkmib_update(ifv, nmib);
561 	nmib = NULL;
562 	nmib_psref = NULL;
563 	omib_cleanup = true;
564 
565 done:
566 	mutex_exit(&ifv->ifv_lock);
567 
568 	if (nmib_psref)
569 		psref_target_destroy(nmib_psref, ifvm_psref_class);
570 	if (nmib)
571 		kmem_free(nmib, sizeof(*nmib));
572 	if (omib_cleanup)
573 		kmem_free(omib, sizeof(*omib));
574 
575 	return error;
576 }
577 
578 /*
579  * Unconfigure a VLAN interface.
580  */
581 static void
582 vlan_unconfig(struct ifnet *ifp)
583 {
584 	struct ifvlan *ifv = ifp->if_softc;
585 	struct ifvlan_linkmib *nmib = NULL;
586 	int error;
587 
588 	KASSERT(IFNET_LOCKED(ifp));
589 
590 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
591 
592 	mutex_enter(&ifv->ifv_lock);
593 	error = vlan_unconfig_locked(ifv, nmib);
594 	mutex_exit(&ifv->ifv_lock);
595 
596 	if (error)
597 		kmem_free(nmib, sizeof(*nmib));
598 }
599 static int
600 vlan_unconfig_locked(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
601 {
602 	struct ifnet *p;
603 	struct ifnet *ifp = &ifv->ifv_if;
604 	struct psref_target *nmib_psref = NULL;
605 	struct ifvlan_linkmib *omib;
606 	int error = 0;
607 
608 	KASSERT(IFNET_LOCKED(ifp));
609 	KASSERT(mutex_owned(&ifv->ifv_lock));
610 
611 	ifp->if_flags &= ~(IFF_UP | IFF_RUNNING);
612 
613 	omib = ifv->ifv_mib;
614 	p = omib->ifvm_p;
615 
616 	if (p == NULL) {
617 		error = -1;
618 		goto done;
619 	}
620 
621 	*nmib = *omib;
622 	nmib_psref = &nmib->ifvm_psref;
623 	psref_target_init(nmib_psref, ifvm_psref_class);
624 
625 	/*
626 	 * Since the interface is being unconfigured, we need to empty the
627 	 * list of multicast groups that we may have joined while we were
628 	 * alive and remove them from the parent's list also.
629 	 */
630 	(*nmib->ifvm_msw->vmsw_purgemulti)(ifv);
631 
632 	/* Disconnect from parent. */
633 	switch (p->if_type) {
634 	case IFT_ETHER:
635 	    {
636 		struct ethercom *ec = (void *)p;
637 		struct vlanid_list *vlanidp;
638 		uint16_t vid = EVL_VLANOFTAG(nmib->ifvm_tag);
639 
640 		ETHER_LOCK(ec);
641 		SIMPLEQ_FOREACH(vlanidp, &ec->ec_vids, vid_list) {
642 			if (vlanidp->vid == vid) {
643 				SIMPLEQ_REMOVE(&ec->ec_vids, vlanidp,
644 				    vlanid_list, vid_list);
645 				break;
646 			}
647 		}
648 		ETHER_UNLOCK(ec);
649 		if (vlanidp != NULL)
650 			kmem_free(vlanidp, sizeof(*vlanidp));
651 
652 		if (ec->ec_vlan_cb != NULL) {
653 			/*
654 			 * Call ec_vlan_cb(). It will setup VLAN HW filter or
655 			 * HW tagging function.
656 			 */
657 			(void)(*ec->ec_vlan_cb)(ec, vid, false);
658 		}
659 		if (--ec->ec_nvlans == 0) {
660 			IFNET_LOCK(p);
661 			(void)ether_disable_vlan_mtu(p);
662 			IFNET_UNLOCK(p);
663 		}
664 
665 		/* XXX ether_ifdetach must not be called with IFNET_LOCK */
666 		mutex_exit(&ifv->ifv_lock);
667 		IFNET_UNLOCK(ifp);
668 		ether_ifdetach(ifp);
669 		IFNET_LOCK(ifp);
670 		mutex_enter(&ifv->ifv_lock);
671 
672 		/* if_free_sadl must be called with IFNET_LOCK */
673 		if_free_sadl(ifp, 1);
674 
675 		/* Restore vlan_ioctl overwritten by ether_ifdetach */
676 		ifp->if_ioctl = vlan_ioctl;
677 		vlan_reset_linkname(ifp);
678 		break;
679 	    }
680 
681 	default:
682 		panic("%s: impossible", __func__);
683 	}
684 
685 	nmib->ifvm_p = NULL;
686 	ifv->ifv_if.if_mtu = 0;
687 	ifv->ifv_flags = 0;
688 
689 	mutex_enter(&ifv_hash.lock);
690 	PSLIST_WRITER_REMOVE(ifv, ifv_hash);
691 	pserialize_perform(vlan_psz);
692 	mutex_exit(&ifv_hash.lock);
693 	PSLIST_ENTRY_DESTROY(ifv, ifv_hash);
694 
695 	vlan_linkmib_update(ifv, nmib);
696 
697 	mutex_exit(&ifv->ifv_lock);
698 
699 	nmib_psref = NULL;
700 	kmem_free(omib, sizeof(*omib));
701 
702 #ifdef INET6
703 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
704 	/* To delete v6 link local addresses */
705 	if (in6_present)
706 		in6_ifdetach(ifp);
707 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
708 #endif
709 
710 	if ((ifp->if_flags & IFF_PROMISC) != 0)
711 		vlan_safe_ifpromisc_locked(ifp, 0);
712 	if_down_locked(ifp);
713 	ifp->if_capabilities = 0;
714 	mutex_enter(&ifv->ifv_lock);
715 done:
716 
717 	if (nmib_psref)
718 		psref_target_destroy(nmib_psref, ifvm_psref_class);
719 
720 	return error;
721 }
722 
723 static void
724 vlan_hash_init(void)
725 {
726 
727 	ifv_hash.lists = hashinit(VLAN_TAG_HASH_SIZE, HASH_PSLIST, true,
728 	    &ifv_hash.mask);
729 }
730 
731 static int
732 vlan_hash_fini(void)
733 {
734 	int i;
735 
736 	mutex_enter(&ifv_hash.lock);
737 
738 	for (i = 0; i < ifv_hash.mask + 1; i++) {
739 		if (PSLIST_WRITER_FIRST(&ifv_hash.lists[i], struct ifvlan,
740 		    ifv_hash) != NULL) {
741 			mutex_exit(&ifv_hash.lock);
742 			return EBUSY;
743 		}
744 	}
745 
746 	for (i = 0; i < ifv_hash.mask + 1; i++)
747 		PSLIST_DESTROY(&ifv_hash.lists[i]);
748 
749 	mutex_exit(&ifv_hash.lock);
750 
751 	hashdone(ifv_hash.lists, HASH_PSLIST, ifv_hash.mask);
752 
753 	ifv_hash.lists = NULL;
754 	ifv_hash.mask = 0;
755 
756 	return 0;
757 }
758 
759 static int
760 vlan_tag_hash(uint16_t tag, u_long mask)
761 {
762 	uint32_t hash;
763 
764 	hash = (tag >> 8) ^ tag;
765 	hash = (hash >> 2) ^ hash;
766 
767 	return hash & mask;
768 }
769 
770 static struct ifvlan_linkmib *
771 vlan_getref_linkmib(struct ifvlan *sc, struct psref *psref)
772 {
773 	struct ifvlan_linkmib *mib;
774 	int s;
775 
776 	s = pserialize_read_enter();
777 	mib = atomic_load_consume(&sc->ifv_mib);
778 	if (mib == NULL) {
779 		pserialize_read_exit(s);
780 		return NULL;
781 	}
782 	psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
783 	pserialize_read_exit(s);
784 
785 	return mib;
786 }
787 
788 static void
789 vlan_putref_linkmib(struct ifvlan_linkmib *mib, struct psref *psref)
790 {
791 	if (mib == NULL)
792 		return;
793 	psref_release(psref, &mib->ifvm_psref, ifvm_psref_class);
794 }
795 
796 static struct ifvlan_linkmib *
797 vlan_lookup_tag_psref(struct ifnet *ifp, uint16_t tag, struct psref *psref)
798 {
799 	int idx;
800 	int s;
801 	struct ifvlan *sc;
802 
803 	idx = vlan_tag_hash(tag, ifv_hash.mask);
804 
805 	s = pserialize_read_enter();
806 	PSLIST_READER_FOREACH(sc, &ifv_hash.lists[idx], struct ifvlan,
807 	    ifv_hash) {
808 		struct ifvlan_linkmib *mib = atomic_load_consume(&sc->ifv_mib);
809 		if (mib == NULL)
810 			continue;
811 		if (mib->ifvm_tag != tag)
812 			continue;
813 		if (mib->ifvm_p != ifp)
814 			continue;
815 
816 		psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
817 		pserialize_read_exit(s);
818 		return mib;
819 	}
820 	pserialize_read_exit(s);
821 	return NULL;
822 }
823 
824 static void
825 vlan_linkmib_update(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
826 {
827 	struct ifvlan_linkmib *omib = ifv->ifv_mib;
828 
829 	KASSERT(mutex_owned(&ifv->ifv_lock));
830 
831 	atomic_store_release(&ifv->ifv_mib, nmib);
832 
833 	pserialize_perform(ifv->ifv_psz);
834 	psref_target_destroy(&omib->ifvm_psref, ifvm_psref_class);
835 }
836 
837 /*
838  * Called when a parent interface is detaching; destroy any VLAN
839  * configuration for the parent interface.
840  */
841 void
842 vlan_ifdetach(struct ifnet *p)
843 {
844 	struct ifvlan *ifv;
845 	struct ifvlan_linkmib *mib, **nmibs;
846 	struct psref psref;
847 	int error;
848 	int bound;
849 	int i, cnt = 0;
850 
851 	bound = curlwp_bind();
852 
853 	mutex_enter(&ifv_list.lock);
854 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
855 		mib = vlan_getref_linkmib(ifv, &psref);
856 		if (mib == NULL)
857 			continue;
858 
859 		if (mib->ifvm_p == p)
860 			cnt++;
861 
862 		vlan_putref_linkmib(mib, &psref);
863 	}
864 	mutex_exit(&ifv_list.lock);
865 
866 	if (cnt == 0) {
867 		curlwp_bindx(bound);
868 		return;
869 	}
870 
871 	/*
872 	 * The value of "cnt" does not increase while ifv_list.lock
873 	 * and ifv->ifv_lock are released here, because the parent
874 	 * interface is detaching.
875 	 */
876 	nmibs = kmem_alloc(sizeof(*nmibs) * cnt, KM_SLEEP);
877 	for (i = 0; i < cnt; i++) {
878 		nmibs[i] = kmem_alloc(sizeof(*nmibs[i]), KM_SLEEP);
879 	}
880 
881 	mutex_enter(&ifv_list.lock);
882 
883 	i = 0;
884 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
885 		struct ifnet *ifp = &ifv->ifv_if;
886 
887 		/* IFNET_LOCK must be held before ifv_lock. */
888 		IFNET_LOCK(ifp);
889 		mutex_enter(&ifv->ifv_lock);
890 
891 		/* XXX ifv_mib = NULL? */
892 		if (ifv->ifv_mib->ifvm_p == p) {
893 			KASSERTMSG(i < cnt,
894 			    "no memory for unconfig, parent=%s", p->if_xname);
895 			error = vlan_unconfig_locked(ifv, nmibs[i]);
896 			if (!error) {
897 				nmibs[i] = NULL;
898 				i++;
899 			}
900 
901 		}
902 
903 		mutex_exit(&ifv->ifv_lock);
904 		IFNET_UNLOCK(ifp);
905 	}
906 
907 	mutex_exit(&ifv_list.lock);
908 
909 	curlwp_bindx(bound);
910 
911 	for (i = 0; i < cnt; i++) {
912 		if (nmibs[i])
913 			kmem_free(nmibs[i], sizeof(*nmibs[i]));
914 	}
915 
916 	kmem_free(nmibs, sizeof(*nmibs) * cnt);
917 
918 	return;
919 }
920 
921 static int
922 vlan_set_promisc(struct ifnet *ifp)
923 {
924 	struct ifvlan *ifv = ifp->if_softc;
925 	struct ifvlan_linkmib *mib;
926 	struct psref psref;
927 	int error = 0;
928 	int bound;
929 
930 	bound = curlwp_bind();
931 	mib = vlan_getref_linkmib(ifv, &psref);
932 	if (mib == NULL) {
933 		curlwp_bindx(bound);
934 		return EBUSY;
935 	}
936 
937 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
938 		if ((ifv->ifv_flags & IFVF_PROMISC) == 0) {
939 			error = vlan_safe_ifpromisc(mib->ifvm_p, 1);
940 			if (error == 0)
941 				ifv->ifv_flags |= IFVF_PROMISC;
942 		}
943 	} else {
944 		if ((ifv->ifv_flags & IFVF_PROMISC) != 0) {
945 			error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
946 			if (error == 0)
947 				ifv->ifv_flags &= ~IFVF_PROMISC;
948 		}
949 	}
950 	vlan_putref_linkmib(mib, &psref);
951 	curlwp_bindx(bound);
952 
953 	return error;
954 }
955 
956 static int
957 vlan_ioctl(struct ifnet *ifp, u_long cmd, void *data)
958 {
959 	struct lwp *l = curlwp;
960 	struct ifvlan *ifv = ifp->if_softc;
961 	struct ifaddr *ifa = (struct ifaddr *) data;
962 	struct ifreq *ifr = (struct ifreq *) data;
963 	struct ifnet *pr;
964 	struct ifcapreq *ifcr;
965 	struct vlanreq vlr;
966 	struct ifvlan_linkmib *mib;
967 	struct psref psref;
968 	int error = 0;
969 	int bound;
970 
971 	switch (cmd) {
972 	case SIOCSIFMTU:
973 		bound = curlwp_bind();
974 		mib = vlan_getref_linkmib(ifv, &psref);
975 		if (mib == NULL) {
976 			curlwp_bindx(bound);
977 			error = EBUSY;
978 			break;
979 		}
980 
981 		if (mib->ifvm_p == NULL) {
982 			vlan_putref_linkmib(mib, &psref);
983 			curlwp_bindx(bound);
984 			error = EINVAL;
985 		} else if (
986 		    ifr->ifr_mtu > (mib->ifvm_p->if_mtu - mib->ifvm_mtufudge) ||
987 		    ifr->ifr_mtu < (mib->ifvm_mintu - mib->ifvm_mtufudge)) {
988 			vlan_putref_linkmib(mib, &psref);
989 			curlwp_bindx(bound);
990 			error = EINVAL;
991 		} else {
992 			vlan_putref_linkmib(mib, &psref);
993 			curlwp_bindx(bound);
994 
995 			error = ifioctl_common(ifp, cmd, data);
996 			if (error == ENETRESET)
997 				error = 0;
998 		}
999 
1000 		break;
1001 
1002 	case SIOCSETVLAN:
1003 		if ((error = kauth_authorize_network(l->l_cred,
1004 		    KAUTH_NETWORK_INTERFACE,
1005 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
1006 		    NULL)) != 0)
1007 			break;
1008 		if ((error = copyin(ifr->ifr_data, &vlr, sizeof(vlr))) != 0)
1009 			break;
1010 
1011 		if (vlr.vlr_parent[0] == '\0') {
1012 			bound = curlwp_bind();
1013 			mib = vlan_getref_linkmib(ifv, &psref);
1014 			if (mib == NULL) {
1015 				curlwp_bindx(bound);
1016 				error = EBUSY;
1017 				break;
1018 			}
1019 
1020 			if (mib->ifvm_p != NULL &&
1021 			    (ifp->if_flags & IFF_PROMISC) != 0)
1022 				error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
1023 
1024 			vlan_putref_linkmib(mib, &psref);
1025 			curlwp_bindx(bound);
1026 
1027 			vlan_unconfig(ifp);
1028 			break;
1029 		}
1030 		if (vlr.vlr_tag != EVL_VLANOFTAG(vlr.vlr_tag)) {
1031 			error = EINVAL;		 /* check for valid tag */
1032 			break;
1033 		}
1034 		if ((pr = ifunit(vlr.vlr_parent)) == NULL) {
1035 			error = ENOENT;
1036 			break;
1037 		}
1038 
1039 		error = vlan_config(ifv, pr, vlr.vlr_tag);
1040 		if (error != 0)
1041 			break;
1042 
1043 		/* Update promiscuous mode, if necessary. */
1044 		vlan_set_promisc(ifp);
1045 
1046 		ifp->if_flags |= IFF_RUNNING;
1047 		break;
1048 
1049 	case SIOCGETVLAN:
1050 		memset(&vlr, 0, sizeof(vlr));
1051 		bound = curlwp_bind();
1052 		mib = vlan_getref_linkmib(ifv, &psref);
1053 		if (mib == NULL) {
1054 			curlwp_bindx(bound);
1055 			error = EBUSY;
1056 			break;
1057 		}
1058 		if (mib->ifvm_p != NULL) {
1059 			snprintf(vlr.vlr_parent, sizeof(vlr.vlr_parent), "%s",
1060 			    mib->ifvm_p->if_xname);
1061 			vlr.vlr_tag = mib->ifvm_tag;
1062 		}
1063 		vlan_putref_linkmib(mib, &psref);
1064 		curlwp_bindx(bound);
1065 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1066 		break;
1067 
1068 	case SIOCSIFFLAGS:
1069 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
1070 			break;
1071 		/*
1072 		 * For promiscuous mode, we enable promiscuous mode on
1073 		 * the parent if we need promiscuous on the VLAN interface.
1074 		 */
1075 		bound = curlwp_bind();
1076 		mib = vlan_getref_linkmib(ifv, &psref);
1077 		if (mib == NULL) {
1078 			curlwp_bindx(bound);
1079 			error = EBUSY;
1080 			break;
1081 		}
1082 
1083 		if (mib->ifvm_p != NULL)
1084 			error = vlan_set_promisc(ifp);
1085 		vlan_putref_linkmib(mib, &psref);
1086 		curlwp_bindx(bound);
1087 		break;
1088 
1089 	case SIOCADDMULTI:
1090 		mutex_enter(&ifv->ifv_lock);
1091 		mib = ifv->ifv_mib;
1092 		if (mib == NULL) {
1093 			error = EBUSY;
1094 			mutex_exit(&ifv->ifv_lock);
1095 			break;
1096 		}
1097 
1098 		error = (mib->ifvm_p != NULL) ?
1099 		    (*mib->ifvm_msw->vmsw_addmulti)(ifv, ifr) : EINVAL;
1100 		mib = NULL;
1101 		mutex_exit(&ifv->ifv_lock);
1102 		break;
1103 
1104 	case SIOCDELMULTI:
1105 		mutex_enter(&ifv->ifv_lock);
1106 		mib = ifv->ifv_mib;
1107 		if (mib == NULL) {
1108 			error = EBUSY;
1109 			mutex_exit(&ifv->ifv_lock);
1110 			break;
1111 		}
1112 		error = (mib->ifvm_p != NULL) ?
1113 		    (*mib->ifvm_msw->vmsw_delmulti)(ifv, ifr) : EINVAL;
1114 		mib = NULL;
1115 		mutex_exit(&ifv->ifv_lock);
1116 		break;
1117 
1118 	case SIOCSIFCAP:
1119 		ifcr = data;
1120 		/* make sure caps are enabled on parent */
1121 		bound = curlwp_bind();
1122 		mib = vlan_getref_linkmib(ifv, &psref);
1123 		if (mib == NULL) {
1124 			curlwp_bindx(bound);
1125 			error = EBUSY;
1126 			break;
1127 		}
1128 
1129 		if (mib->ifvm_p == NULL) {
1130 			vlan_putref_linkmib(mib, &psref);
1131 			curlwp_bindx(bound);
1132 			error = EINVAL;
1133 			break;
1134 		}
1135 		if ((mib->ifvm_p->if_capenable & ifcr->ifcr_capenable) !=
1136 		    ifcr->ifcr_capenable) {
1137 			vlan_putref_linkmib(mib, &psref);
1138 			curlwp_bindx(bound);
1139 			error = EINVAL;
1140 			break;
1141 		}
1142 
1143 		vlan_putref_linkmib(mib, &psref);
1144 		curlwp_bindx(bound);
1145 
1146 		if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
1147 			error = 0;
1148 		break;
1149 	case SIOCINITIFADDR:
1150 		bound = curlwp_bind();
1151 		mib = vlan_getref_linkmib(ifv, &psref);
1152 		if (mib == NULL) {
1153 			curlwp_bindx(bound);
1154 			error = EBUSY;
1155 			break;
1156 		}
1157 
1158 		if (mib->ifvm_p == NULL) {
1159 			error = EINVAL;
1160 			vlan_putref_linkmib(mib, &psref);
1161 			curlwp_bindx(bound);
1162 			break;
1163 		}
1164 		vlan_putref_linkmib(mib, &psref);
1165 		curlwp_bindx(bound);
1166 
1167 		ifp->if_flags |= IFF_UP;
1168 #ifdef INET
1169 		if (ifa->ifa_addr->sa_family == AF_INET)
1170 			arp_ifinit(ifp, ifa);
1171 #endif
1172 		break;
1173 
1174 	default:
1175 		error = ether_ioctl(ifp, cmd, data);
1176 	}
1177 
1178 	return error;
1179 }
1180 
1181 static int
1182 vlan_ether_addmulti(struct ifvlan *ifv, struct ifreq *ifr)
1183 {
1184 	const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr);
1185 	struct vlan_mc_entry *mc;
1186 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1187 	struct ifvlan_linkmib *mib;
1188 	int error;
1189 
1190 	KASSERT(mutex_owned(&ifv->ifv_lock));
1191 
1192 	if (sa->sa_len > sizeof(struct sockaddr_storage))
1193 		return EINVAL;
1194 
1195 	error = ether_addmulti(sa, &ifv->ifv_ec);
1196 	if (error != ENETRESET)
1197 		return error;
1198 
1199 	/*
1200 	 * This is a new multicast address.  We have to tell parent
1201 	 * about it.  Also, remember this multicast address so that
1202 	 * we can delete it on unconfigure.
1203 	 */
1204 	mc = malloc(sizeof(struct vlan_mc_entry), M_DEVBUF, M_NOWAIT);
1205 	if (mc == NULL) {
1206 		error = ENOMEM;
1207 		goto alloc_failed;
1208 	}
1209 
1210 	/*
1211 	 * Since ether_addmulti() returned ENETRESET, the following two
1212 	 * statements shouldn't fail. Here ifv_ec is implicitly protected
1213 	 * by the ifv_lock lock.
1214 	 */
1215 	error = ether_multiaddr(sa, addrlo, addrhi);
1216 	KASSERT(error == 0);
1217 
1218 	ETHER_LOCK(&ifv->ifv_ec);
1219 	mc->mc_enm = ether_lookup_multi(addrlo, addrhi, &ifv->ifv_ec);
1220 	ETHER_UNLOCK(&ifv->ifv_ec);
1221 
1222 	KASSERT(mc->mc_enm != NULL);
1223 
1224 	memcpy(&mc->mc_addr, sa, sa->sa_len);
1225 	LIST_INSERT_HEAD(&ifv->ifv_mc_listhead, mc, mc_entries);
1226 
1227 	mib = ifv->ifv_mib;
1228 
1229 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1230 	error = if_mcast_op(mib->ifvm_p, SIOCADDMULTI, sa);
1231 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1232 
1233 	if (error != 0)
1234 		goto ioctl_failed;
1235 	return error;
1236 
1237 ioctl_failed:
1238 	LIST_REMOVE(mc, mc_entries);
1239 	free(mc, M_DEVBUF);
1240 
1241 alloc_failed:
1242 	(void)ether_delmulti(sa, &ifv->ifv_ec);
1243 	return error;
1244 }
1245 
1246 static int
1247 vlan_ether_delmulti(struct ifvlan *ifv, struct ifreq *ifr)
1248 {
1249 	const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr);
1250 	struct ether_multi *enm;
1251 	struct vlan_mc_entry *mc;
1252 	struct ifvlan_linkmib *mib;
1253 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1254 	int error;
1255 
1256 	KASSERT(mutex_owned(&ifv->ifv_lock));
1257 
1258 	/*
1259 	 * Find a key to lookup vlan_mc_entry.  We have to do this
1260 	 * before calling ether_delmulti for obvious reasons.
1261 	 */
1262 	if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0)
1263 		return error;
1264 
1265 	ETHER_LOCK(&ifv->ifv_ec);
1266 	enm = ether_lookup_multi(addrlo, addrhi, &ifv->ifv_ec);
1267 	ETHER_UNLOCK(&ifv->ifv_ec);
1268 	if (enm == NULL)
1269 		return EINVAL;
1270 
1271 	LIST_FOREACH(mc, &ifv->ifv_mc_listhead, mc_entries) {
1272 		if (mc->mc_enm == enm)
1273 			break;
1274 	}
1275 
1276 	/* We woun't delete entries we didn't add */
1277 	if (mc == NULL)
1278 		return EINVAL;
1279 
1280 	error = ether_delmulti(sa, &ifv->ifv_ec);
1281 	if (error != ENETRESET)
1282 		return error;
1283 
1284 	/* We no longer use this multicast address.  Tell parent so. */
1285 	mib = ifv->ifv_mib;
1286 	error = if_mcast_op(mib->ifvm_p, SIOCDELMULTI, sa);
1287 
1288 	if (error == 0) {
1289 		/* And forget about this address. */
1290 		LIST_REMOVE(mc, mc_entries);
1291 		free(mc, M_DEVBUF);
1292 	} else {
1293 		(void)ether_addmulti(sa, &ifv->ifv_ec);
1294 	}
1295 
1296 	return error;
1297 }
1298 
1299 /*
1300  * Delete any multicast address we have asked to add from parent
1301  * interface.  Called when the vlan is being unconfigured.
1302  */
1303 static void
1304 vlan_ether_purgemulti(struct ifvlan *ifv)
1305 {
1306 	struct vlan_mc_entry *mc;
1307 	struct ifvlan_linkmib *mib;
1308 
1309 	KASSERT(mutex_owned(&ifv->ifv_lock));
1310 	mib = ifv->ifv_mib;
1311 	if (mib == NULL) {
1312 		return;
1313 	}
1314 
1315 	while ((mc = LIST_FIRST(&ifv->ifv_mc_listhead)) != NULL) {
1316 		(void)if_mcast_op(mib->ifvm_p, SIOCDELMULTI,
1317 		    sstocsa(&mc->mc_addr));
1318 		LIST_REMOVE(mc, mc_entries);
1319 		free(mc, M_DEVBUF);
1320 	}
1321 }
1322 
1323 static void
1324 vlan_start(struct ifnet *ifp)
1325 {
1326 	struct ifvlan *ifv = ifp->if_softc;
1327 	struct ifnet *p;
1328 	struct ethercom *ec;
1329 	struct mbuf *m;
1330 	struct ifvlan_linkmib *mib;
1331 	struct psref psref;
1332 	int error;
1333 
1334 	mib = vlan_getref_linkmib(ifv, &psref);
1335 	if (mib == NULL)
1336 		return;
1337 	p = mib->ifvm_p;
1338 	ec = (void *)mib->ifvm_p;
1339 
1340 	ifp->if_flags |= IFF_OACTIVE;
1341 
1342 	for (;;) {
1343 		IFQ_DEQUEUE(&ifp->if_snd, m);
1344 		if (m == NULL)
1345 			break;
1346 
1347 #ifdef ALTQ
1348 		/*
1349 		 * KERNEL_LOCK is required for ALTQ even if NET_MPSAFE is
1350 		 * defined.
1351 		 */
1352 		KERNEL_LOCK(1, NULL);
1353 		/*
1354 		 * If ALTQ is enabled on the parent interface, do
1355 		 * classification; the queueing discipline might
1356 		 * not require classification, but might require
1357 		 * the address family/header pointer in the pktattr.
1358 		 */
1359 		if (ALTQ_IS_ENABLED(&p->if_snd)) {
1360 			switch (p->if_type) {
1361 			case IFT_ETHER:
1362 				altq_etherclassify(&p->if_snd, m);
1363 				break;
1364 			default:
1365 				panic("%s: impossible (altq)", __func__);
1366 			}
1367 		}
1368 		KERNEL_UNLOCK_ONE(NULL);
1369 #endif /* ALTQ */
1370 
1371 		bpf_mtap(ifp, m, BPF_D_OUT);
1372 		/*
1373 		 * If the parent can insert the tag itself, just mark
1374 		 * the tag in the mbuf header.
1375 		 */
1376 		if (ec->ec_capenable & ETHERCAP_VLAN_HWTAGGING) {
1377 			vlan_set_tag(m, mib->ifvm_tag);
1378 		} else {
1379 			/*
1380 			 * insert the tag ourselves
1381 			 */
1382 			M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1383 			if (m == NULL) {
1384 				printf("%s: unable to prepend encap header",
1385 				    p->if_xname);
1386 				if_statinc(ifp, if_oerrors);
1387 				continue;
1388 			}
1389 
1390 			switch (p->if_type) {
1391 			case IFT_ETHER:
1392 			    {
1393 				struct ether_vlan_header *evl;
1394 
1395 				if (m->m_len < sizeof(struct ether_vlan_header))
1396 					m = m_pullup(m,
1397 					    sizeof(struct ether_vlan_header));
1398 				if (m == NULL) {
1399 					printf("%s: unable to pullup encap "
1400 					    "header", p->if_xname);
1401 					if_statinc(ifp, if_oerrors);
1402 					continue;
1403 				}
1404 
1405 				/*
1406 				 * Transform the Ethernet header into an
1407 				 * Ethernet header with 802.1Q encapsulation.
1408 				 */
1409 				memmove(mtod(m, void *),
1410 				    mtod(m, char *) + mib->ifvm_encaplen,
1411 				    sizeof(struct ether_header));
1412 				evl = mtod(m, struct ether_vlan_header *);
1413 				evl->evl_proto = evl->evl_encap_proto;
1414 				evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1415 				evl->evl_tag = htons(mib->ifvm_tag);
1416 
1417 				/*
1418 				 * To cater for VLAN-aware layer 2 ethernet
1419 				 * switches which may need to strip the tag
1420 				 * before forwarding the packet, make sure
1421 				 * the packet+tag is at least 68 bytes long.
1422 				 * This is necessary because our parent will
1423 				 * only pad to 64 bytes (ETHER_MIN_LEN) and
1424 				 * some switches will not pad by themselves
1425 				 * after deleting a tag.
1426 				 */
1427 				const size_t min_data_len = ETHER_MIN_LEN -
1428 				    ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
1429 				if (m->m_pkthdr.len < min_data_len) {
1430 					m_copyback(m, m->m_pkthdr.len,
1431 					    min_data_len - m->m_pkthdr.len,
1432 					    vlan_zero_pad_buff);
1433 				}
1434 				break;
1435 			    }
1436 
1437 			default:
1438 				panic("%s: impossible", __func__);
1439 			}
1440 		}
1441 
1442 		if ((p->if_flags & IFF_RUNNING) == 0) {
1443 			m_freem(m);
1444 			continue;
1445 		}
1446 
1447 		error = if_transmit_lock(p, m);
1448 		if (error) {
1449 			/* mbuf is already freed */
1450 			if_statinc(ifp, if_oerrors);
1451 			continue;
1452 		}
1453 		if_statinc(ifp, if_opackets);
1454 	}
1455 
1456 	ifp->if_flags &= ~IFF_OACTIVE;
1457 
1458 	/* Remove reference to mib before release */
1459 	vlan_putref_linkmib(mib, &psref);
1460 }
1461 
1462 static int
1463 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1464 {
1465 	struct ifvlan *ifv = ifp->if_softc;
1466 	struct ifnet *p;
1467 	struct ethercom *ec;
1468 	struct ifvlan_linkmib *mib;
1469 	struct psref psref;
1470 	int error;
1471 	size_t pktlen = m->m_pkthdr.len;
1472 	bool mcast = (m->m_flags & M_MCAST) != 0;
1473 
1474 	mib = vlan_getref_linkmib(ifv, &psref);
1475 	if (mib == NULL) {
1476 		m_freem(m);
1477 		return ENETDOWN;
1478 	}
1479 
1480 	p = mib->ifvm_p;
1481 	ec = (void *)mib->ifvm_p;
1482 
1483 	bpf_mtap(ifp, m, BPF_D_OUT);
1484 
1485 	if ((error = pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_OUT)) != 0)
1486 		goto out;
1487 	if (m == NULL)
1488 		goto out;
1489 
1490 	/*
1491 	 * If the parent can insert the tag itself, just mark
1492 	 * the tag in the mbuf header.
1493 	 */
1494 	if (ec->ec_capenable & ETHERCAP_VLAN_HWTAGGING) {
1495 		vlan_set_tag(m, mib->ifvm_tag);
1496 	} else {
1497 		/*
1498 		 * insert the tag ourselves
1499 		 */
1500 		M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1501 		if (m == NULL) {
1502 			printf("%s: unable to prepend encap header",
1503 			    p->if_xname);
1504 			if_statinc(ifp, if_oerrors);
1505 			error = ENOBUFS;
1506 			goto out;
1507 		}
1508 
1509 		switch (p->if_type) {
1510 		case IFT_ETHER:
1511 		    {
1512 			struct ether_vlan_header *evl;
1513 
1514 			if (m->m_len < sizeof(struct ether_vlan_header))
1515 				m = m_pullup(m,
1516 				    sizeof(struct ether_vlan_header));
1517 			if (m == NULL) {
1518 				printf("%s: unable to pullup encap "
1519 				    "header", p->if_xname);
1520 				if_statinc(ifp, if_oerrors);
1521 				error = ENOBUFS;
1522 				goto out;
1523 			}
1524 
1525 			/*
1526 			 * Transform the Ethernet header into an
1527 			 * Ethernet header with 802.1Q encapsulation.
1528 			 */
1529 			memmove(mtod(m, void *),
1530 			    mtod(m, char *) + mib->ifvm_encaplen,
1531 			    sizeof(struct ether_header));
1532 			evl = mtod(m, struct ether_vlan_header *);
1533 			evl->evl_proto = evl->evl_encap_proto;
1534 			evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1535 			evl->evl_tag = htons(mib->ifvm_tag);
1536 
1537 			/*
1538 			 * To cater for VLAN-aware layer 2 ethernet
1539 			 * switches which may need to strip the tag
1540 			 * before forwarding the packet, make sure
1541 			 * the packet+tag is at least 68 bytes long.
1542 			 * This is necessary because our parent will
1543 			 * only pad to 64 bytes (ETHER_MIN_LEN) and
1544 			 * some switches will not pad by themselves
1545 			 * after deleting a tag.
1546 			 */
1547 			const size_t min_data_len = ETHER_MIN_LEN -
1548 			    ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
1549 			if (m->m_pkthdr.len < min_data_len) {
1550 				m_copyback(m, m->m_pkthdr.len,
1551 				    min_data_len - m->m_pkthdr.len,
1552 				    vlan_zero_pad_buff);
1553 			}
1554 			break;
1555 		    }
1556 
1557 		default:
1558 			panic("%s: impossible", __func__);
1559 		}
1560 	}
1561 
1562 	if ((p->if_flags & IFF_RUNNING) == 0) {
1563 		m_freem(m);
1564 		error = ENETDOWN;
1565 		goto out;
1566 	}
1567 
1568 	error = if_transmit_lock(p, m);
1569 	net_stat_ref_t nsr = IF_STAT_GETREF(ifp);
1570 	if (error) {
1571 		/* mbuf is already freed */
1572 		if_statinc_ref(nsr, if_oerrors);
1573 	} else {
1574 		if_statinc_ref(nsr, if_opackets);
1575 		if_statadd_ref(nsr, if_obytes, pktlen);
1576 		if (mcast)
1577 			if_statinc_ref(nsr, if_omcasts);
1578 	}
1579 	IF_STAT_PUTREF(ifp);
1580 
1581 out:
1582 	/* Remove reference to mib before release */
1583 	vlan_putref_linkmib(mib, &psref);
1584 	return error;
1585 }
1586 
1587 /*
1588  * Given an Ethernet frame, find a valid vlan interface corresponding to the
1589  * given source interface and tag, then run the real packet through the
1590  * parent's input routine.
1591  */
1592 void
1593 vlan_input(struct ifnet *ifp, struct mbuf *m)
1594 {
1595 	struct ifvlan *ifv;
1596 	uint16_t vid;
1597 	struct ifvlan_linkmib *mib;
1598 	struct psref psref;
1599 	bool have_vtag;
1600 
1601 	have_vtag = vlan_has_tag(m);
1602 	if (have_vtag) {
1603 		vid = EVL_VLANOFTAG(vlan_get_tag(m));
1604 		m->m_flags &= ~M_VLANTAG;
1605 	} else {
1606 		struct ether_vlan_header *evl;
1607 
1608 		if (ifp->if_type != IFT_ETHER) {
1609 			panic("%s: impossible", __func__);
1610 		}
1611 
1612 		if (m->m_len < sizeof(struct ether_vlan_header) &&
1613 		    (m = m_pullup(m,
1614 		     sizeof(struct ether_vlan_header))) == NULL) {
1615 			printf("%s: no memory for VLAN header, "
1616 			    "dropping packet.\n", ifp->if_xname);
1617 			return;
1618 		}
1619 		evl = mtod(m, struct ether_vlan_header *);
1620 		KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN);
1621 
1622 		vid = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1623 
1624 		/*
1625 		 * Restore the original ethertype.  We'll remove
1626 		 * the encapsulation after we've found the vlan
1627 		 * interface corresponding to the tag.
1628 		 */
1629 		evl->evl_encap_proto = evl->evl_proto;
1630 	}
1631 
1632 	mib = vlan_lookup_tag_psref(ifp, vid, &psref);
1633 	if (mib == NULL) {
1634 		m_freem(m);
1635 		if_statinc(ifp, if_noproto);
1636 		return;
1637 	}
1638 	KASSERT(mib->ifvm_encaplen == ETHER_VLAN_ENCAP_LEN);
1639 
1640 	ifv = mib->ifvm_ifvlan;
1641 	if ((ifv->ifv_if.if_flags & (IFF_UP | IFF_RUNNING)) !=
1642 	    (IFF_UP | IFF_RUNNING)) {
1643 		m_freem(m);
1644 		if_statinc(ifp, if_noproto);
1645 		goto out;
1646 	}
1647 
1648 	/*
1649 	 * Now, remove the encapsulation header.  The original
1650 	 * header has already been fixed up above.
1651 	 */
1652 	if (!have_vtag) {
1653 		memmove(mtod(m, char *) + mib->ifvm_encaplen,
1654 		    mtod(m, void *), sizeof(struct ether_header));
1655 		m_adj(m, mib->ifvm_encaplen);
1656 	}
1657 
1658 	m_set_rcvif(m, &ifv->ifv_if);
1659 
1660 	if (pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_IN) != 0)
1661 		goto out;
1662 	if (m == NULL)
1663 		goto out;
1664 
1665 	m->m_flags &= ~M_PROMISC;
1666 	if_input(&ifv->ifv_if, m);
1667 out:
1668 	vlan_putref_linkmib(mib, &psref);
1669 }
1670 
1671 /*
1672  * Module infrastructure
1673  */
1674 #include "if_module.h"
1675 
1676 IF_MODULE(MODULE_CLASS_DRIVER, vlan, NULL)
1677