xref: /netbsd-src/sys/net/if_vlan.c (revision 122b5006ee1bd67145794b4cde92f4fe4781a5ec)
1 /*	$NetBSD: if_vlan.c,v 1.164 2021/10/05 04:09:49 yamaguchi 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 1.164 2021/10/05 04:09:49 yamaguchi 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 	void *ifv_linkstate_hook;
160 	void *ifv_ifdetach_hook;
161 
162 	LIST_HEAD(__vlan_mchead, vlan_mc_entry) ifv_mc_listhead;
163 	struct pslist_entry ifv_hash;
164 	int ifv_flags;
165 	bool ifv_stopping;
166 };
167 
168 #define	IFVF_PROMISC	0x01		/* promiscuous mode enabled */
169 
170 #define	ifv_if		ifv_ec.ec_if
171 
172 #define	ifv_msw		ifv_mib.ifvm_msw
173 #define	ifv_encaplen	ifv_mib.ifvm_encaplen
174 #define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
175 #define	ifv_mintu	ifv_mib.ifvm_mintu
176 #define	ifv_tag		ifv_mib.ifvm_tag
177 
178 struct vlan_multisw {
179 	int	(*vmsw_addmulti)(struct ifvlan *, struct ifreq *);
180 	int	(*vmsw_delmulti)(struct ifvlan *, struct ifreq *);
181 	void	(*vmsw_purgemulti)(struct ifvlan *);
182 };
183 
184 static int	vlan_ether_addmulti(struct ifvlan *, struct ifreq *);
185 static int	vlan_ether_delmulti(struct ifvlan *, struct ifreq *);
186 static void	vlan_ether_purgemulti(struct ifvlan *);
187 
188 const struct vlan_multisw vlan_ether_multisw = {
189 	.vmsw_addmulti = vlan_ether_addmulti,
190 	.vmsw_delmulti = vlan_ether_delmulti,
191 	.vmsw_purgemulti = vlan_ether_purgemulti,
192 };
193 
194 static int	vlan_clone_create(struct if_clone *, int);
195 static int	vlan_clone_destroy(struct ifnet *);
196 static int	vlan_config(struct ifvlan *, struct ifnet *, uint16_t);
197 static int	vlan_ioctl(struct ifnet *, u_long, void *);
198 static void	vlan_start(struct ifnet *);
199 static int	vlan_transmit(struct ifnet *, struct mbuf *);
200 static void	vlan_link_state_changed(void *);
201 static void	vlan_ifdetach(void *);
202 static void	vlan_unconfig(struct ifnet *);
203 static int	vlan_unconfig_locked(struct ifvlan *, struct ifvlan_linkmib *);
204 static void	vlan_hash_init(void);
205 static int	vlan_hash_fini(void);
206 static int	vlan_tag_hash(uint16_t, u_long);
207 static struct ifvlan_linkmib*	vlan_getref_linkmib(struct ifvlan *,
208     struct psref *);
209 static void	vlan_putref_linkmib(struct ifvlan_linkmib *, struct psref *);
210 static void	vlan_linkmib_update(struct ifvlan *, struct ifvlan_linkmib *);
211 static struct ifvlan_linkmib*	vlan_lookup_tag_psref(struct ifnet *,
212     uint16_t, struct psref *);
213 
214 #if !defined(VLAN_TAG_HASH_SIZE)
215 #define VLAN_TAG_HASH_SIZE 32
216 #endif
217 static struct {
218 	kmutex_t lock;
219 	struct pslist_head *lists;
220 	u_long mask;
221 } ifv_hash __cacheline_aligned = {
222 	.lists = NULL,
223 	.mask = 0,
224 };
225 
226 pserialize_t vlan_psz __read_mostly;
227 static struct psref_class *ifvm_psref_class __read_mostly;
228 
229 struct if_clone vlan_cloner =
230     IF_CLONE_INITIALIZER("vlan", vlan_clone_create, vlan_clone_destroy);
231 
232 /* Used to pad ethernet frames with < ETHER_MIN_LEN bytes */
233 static char vlan_zero_pad_buff[ETHER_MIN_LEN];
234 
235 static uint32_t nvlanifs;
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 __unused 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 	nvlanifs = 0;
275 
276 	mutex_init(&ifv_hash.lock, MUTEX_DEFAULT, IPL_NONE);
277 	vlan_psz = pserialize_create();
278 	ifvm_psref_class = psref_class_create("vlanlinkmib", IPL_SOFTNET);
279 	if_clone_attach(&vlan_cloner);
280 
281 	vlan_hash_init();
282 	MODULE_HOOK_SET(if_vlan_vlan_input_hook, vlan_input);
283 }
284 
285 static int
286 vlandetach(void)
287 {
288 	int error;
289 
290 	if (nvlanifs > 0)
291 		return EBUSY;
292 
293 	error = vlan_hash_fini();
294 	if (error != 0)
295 		return error;
296 
297 	if_clone_detach(&vlan_cloner);
298 	psref_class_destroy(ifvm_psref_class);
299 	pserialize_destroy(vlan_psz);
300 	mutex_destroy(&ifv_hash.lock);
301 
302 	MODULE_HOOK_UNSET(if_vlan_vlan_input_hook);
303 	return 0;
304 }
305 
306 static void
307 vlan_reset_linkname(struct ifnet *ifp)
308 {
309 
310 	/*
311 	 * We start out with a "802.1Q VLAN" type and zero-length
312 	 * addresses.  When we attach to a parent interface, we
313 	 * inherit its type, address length, address, and data link
314 	 * type.
315 	 */
316 
317 	ifp->if_type = IFT_L2VLAN;
318 	ifp->if_addrlen = 0;
319 	ifp->if_dlt = DLT_NULL;
320 	if_alloc_sadl(ifp);
321 }
322 
323 static int
324 vlan_clone_create(struct if_clone *ifc, int unit)
325 {
326 	struct ifvlan *ifv;
327 	struct ifnet *ifp;
328 	struct ifvlan_linkmib *mib;
329 
330 	ifv = malloc(sizeof(struct ifvlan), M_DEVBUF, M_WAITOK | M_ZERO);
331 	mib = kmem_zalloc(sizeof(struct ifvlan_linkmib), KM_SLEEP);
332 	ifp = &ifv->ifv_if;
333 	LIST_INIT(&ifv->ifv_mc_listhead);
334 
335 	mib->ifvm_ifvlan = ifv;
336 	mib->ifvm_p = NULL;
337 	psref_target_init(&mib->ifvm_psref, ifvm_psref_class);
338 
339 	mutex_init(&ifv->ifv_lock, MUTEX_DEFAULT, IPL_NONE);
340 	ifv->ifv_psz = pserialize_create();
341 	ifv->ifv_mib = mib;
342 
343 	atomic_inc_uint(&nvlanifs);
344 
345 	if_initname(ifp, ifc->ifc_name, unit);
346 	ifp->if_softc = ifv;
347 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
348 #ifdef NET_MPSAFE
349 	ifp->if_extflags = IFEF_MPSAFE;
350 #endif
351 	ifp->if_start = vlan_start;
352 	ifp->if_transmit = vlan_transmit;
353 	ifp->if_ioctl = vlan_ioctl;
354 	IFQ_SET_READY(&ifp->if_snd);
355 	if_initialize(ifp);
356 	/*
357 	 * Set the link state to down.
358 	 * When the parent interface attaches we will use that link state.
359 	 * When the parent interface link state changes, so will ours.
360 	 * When the parent interface detaches, set the link state to down.
361 	 */
362 	ifp->if_link_state = LINK_STATE_DOWN;
363 
364 	vlan_reset_linkname(ifp);
365 	if_register(ifp);
366 	return 0;
367 }
368 
369 static int
370 vlan_clone_destroy(struct ifnet *ifp)
371 {
372 	struct ifvlan *ifv = ifp->if_softc;
373 
374 	atomic_dec_uint(&nvlanifs);
375 
376 	IFNET_LOCK(ifp);
377 	vlan_unconfig(ifp);
378 	IFNET_UNLOCK(ifp);
379 	if_detach(ifp);
380 
381 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
382 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
383 	pserialize_destroy(ifv->ifv_psz);
384 	mutex_destroy(&ifv->ifv_lock);
385 	free(ifv, M_DEVBUF);
386 
387 	return 0;
388 }
389 
390 /*
391  * Configure a VLAN interface.
392  */
393 static int
394 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
395 {
396 	struct ifnet *ifp = &ifv->ifv_if;
397 	struct ifvlan_linkmib *nmib = NULL;
398 	struct ifvlan_linkmib *omib = NULL;
399 	struct ifvlan_linkmib *checkmib;
400 	struct psref_target *nmib_psref = NULL;
401 	const uint16_t vid = EVL_VLANOFTAG(tag);
402 	int error = 0;
403 	int idx;
404 	bool omib_cleanup = false;
405 	struct psref psref;
406 
407 	/* VLAN ID 0 and 4095 are reserved in the spec */
408 	if ((vid == 0) || (vid == 0xfff))
409 		return EINVAL;
410 
411 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
412 	mutex_enter(&ifv->ifv_lock);
413 	omib = ifv->ifv_mib;
414 
415 	if (omib->ifvm_p != NULL) {
416 		error = EBUSY;
417 		goto done;
418 	}
419 
420 	/* Duplicate check */
421 	checkmib = vlan_lookup_tag_psref(p, vid, &psref);
422 	if (checkmib != NULL) {
423 		vlan_putref_linkmib(checkmib, &psref);
424 		error = EEXIST;
425 		goto done;
426 	}
427 
428 	*nmib = *omib;
429 	nmib_psref = &nmib->ifvm_psref;
430 
431 	psref_target_init(nmib_psref, ifvm_psref_class);
432 
433 	switch (p->if_type) {
434 	case IFT_ETHER:
435 	    {
436 		struct ethercom *ec = (void *)p;
437 		struct vlanid_list *vidmem;
438 
439 		nmib->ifvm_msw = &vlan_ether_multisw;
440 		nmib->ifvm_encaplen = ETHER_VLAN_ENCAP_LEN;
441 		nmib->ifvm_mintu = ETHERMIN;
442 
443 		if (ec->ec_nvlans++ == 0) {
444 			IFNET_LOCK(p);
445 			error = ether_enable_vlan_mtu(p);
446 			IFNET_UNLOCK(p);
447 			if (error >= 0) {
448 				if (error) {
449 					ec->ec_nvlans--;
450 					goto done;
451 				}
452 				nmib->ifvm_mtufudge = 0;
453 			} else {
454 				/*
455 				 * Fudge the MTU by the encapsulation size. This
456 				 * makes us incompatible with strictly compliant
457 				 * 802.1Q implementations, but allows us to use
458 				 * the feature with other NetBSD
459 				 * implementations, which might still be useful.
460 				 */
461 				nmib->ifvm_mtufudge = nmib->ifvm_encaplen;
462 			}
463 			error = 0;
464 		}
465 		/* Add a vid to the list */
466 		vidmem = kmem_alloc(sizeof(struct vlanid_list), KM_SLEEP);
467 		vidmem->vid = vid;
468 		ETHER_LOCK(ec);
469 		SIMPLEQ_INSERT_TAIL(&ec->ec_vids, vidmem, vid_list);
470 		ETHER_UNLOCK(ec);
471 
472 		if (ec->ec_vlan_cb != NULL) {
473 			/*
474 			 * Call ec_vlan_cb(). It will setup VLAN HW filter or
475 			 * HW tagging function.
476 			 */
477 			error = (*ec->ec_vlan_cb)(ec, vid, true);
478 			if (error) {
479 				ec->ec_nvlans--;
480 				if (ec->ec_nvlans == 0) {
481 					IFNET_LOCK(p);
482 					(void)ether_disable_vlan_mtu(p);
483 					IFNET_UNLOCK(p);
484 				}
485 				goto done;
486 			}
487 		}
488 		/*
489 		 * If the parent interface can do hardware-assisted
490 		 * VLAN encapsulation, then propagate its hardware-
491 		 * assisted checksumming flags and tcp segmentation
492 		 * offload.
493 		 */
494 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
495 			ifp->if_capabilities = p->if_capabilities &
496 			    (IFCAP_TSOv4 | IFCAP_TSOv6 |
497 				IFCAP_CSUM_IPv4_Tx  | IFCAP_CSUM_IPv4_Rx |
498 				IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
499 				IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
500 				IFCAP_CSUM_TCPv6_Tx | IFCAP_CSUM_TCPv6_Rx |
501 				IFCAP_CSUM_UDPv6_Tx | IFCAP_CSUM_UDPv6_Rx);
502 		}
503 
504 		/*
505 		 * We inherit the parent's Ethernet address.
506 		 */
507 		ether_ifattach(ifp, CLLADDR(p->if_sadl));
508 		ifp->if_hdrlen = sizeof(struct ether_vlan_header); /* XXX? */
509 		break;
510 	    }
511 
512 	default:
513 		error = EPROTONOSUPPORT;
514 		goto done;
515 	}
516 
517 	nmib->ifvm_p = p;
518 	nmib->ifvm_tag = vid;
519 	ifv->ifv_if.if_mtu = p->if_mtu - nmib->ifvm_mtufudge;
520 	ifv->ifv_if.if_flags = p->if_flags &
521 	    (IFF_UP | IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
522 
523 	/*
524 	 * Inherit the if_type from the parent.  This allows us
525 	 * to participate in bridges of that type.
526 	 */
527 	ifv->ifv_if.if_type = p->if_type;
528 
529 	PSLIST_ENTRY_INIT(ifv, ifv_hash);
530 	idx = vlan_tag_hash(vid, ifv_hash.mask);
531 
532 	mutex_enter(&ifv_hash.lock);
533 	PSLIST_WRITER_INSERT_HEAD(&ifv_hash.lists[idx], ifv, ifv_hash);
534 	mutex_exit(&ifv_hash.lock);
535 
536 	vlan_linkmib_update(ifv, nmib);
537 	nmib = NULL;
538 	nmib_psref = NULL;
539 	omib_cleanup = true;
540 
541 	ifv->ifv_ifdetach_hook = ether_ifdetachhook_establish(p,
542 	    vlan_ifdetach, ifp);
543 
544 	/*
545 	 * We inherit the parents link state.
546 	 */
547 	ifv->ifv_linkstate_hook = if_linkstate_change_establish(p,
548 	    vlan_link_state_changed, ifv);
549 	if_link_state_change(&ifv->ifv_if, p->if_link_state);
550 
551 done:
552 	mutex_exit(&ifv->ifv_lock);
553 
554 	if (nmib_psref)
555 		psref_target_destroy(nmib_psref, ifvm_psref_class);
556 	if (nmib)
557 		kmem_free(nmib, sizeof(*nmib));
558 	if (omib_cleanup)
559 		kmem_free(omib, sizeof(*omib));
560 
561 	return error;
562 }
563 
564 /*
565  * Unconfigure a VLAN interface.
566  */
567 static void
568 vlan_unconfig(struct ifnet *ifp)
569 {
570 	struct ifvlan *ifv = ifp->if_softc;
571 	struct ifvlan_linkmib *nmib = NULL;
572 	int error;
573 
574 	KASSERT(IFNET_LOCKED(ifp));
575 
576 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
577 
578 	mutex_enter(&ifv->ifv_lock);
579 	error = vlan_unconfig_locked(ifv, nmib);
580 	mutex_exit(&ifv->ifv_lock);
581 
582 	if (error)
583 		kmem_free(nmib, sizeof(*nmib));
584 }
585 static int
586 vlan_unconfig_locked(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
587 {
588 	struct ifnet *p;
589 	struct ifnet *ifp = &ifv->ifv_if;
590 	struct psref_target *nmib_psref = NULL;
591 	struct ifvlan_linkmib *omib;
592 	int error = 0;
593 
594 	KASSERT(IFNET_LOCKED(ifp));
595 	KASSERT(mutex_owned(&ifv->ifv_lock));
596 
597 	if (ifv->ifv_stopping) {
598 		error = -1;
599 		goto done;
600 	}
601 
602 	ifp->if_flags &= ~(IFF_UP | IFF_RUNNING);
603 
604 	omib = ifv->ifv_mib;
605 	p = omib->ifvm_p;
606 
607 	if (p == NULL) {
608 		error = -1;
609 		goto done;
610 	}
611 
612 	*nmib = *omib;
613 	nmib_psref = &nmib->ifvm_psref;
614 	psref_target_init(nmib_psref, ifvm_psref_class);
615 
616 	/*
617 	 * Since the interface is being unconfigured, we need to empty the
618 	 * list of multicast groups that we may have joined while we were
619 	 * alive and remove them from the parent's list also.
620 	 */
621 	(*nmib->ifvm_msw->vmsw_purgemulti)(ifv);
622 
623 	/* Disconnect from parent. */
624 	switch (p->if_type) {
625 	case IFT_ETHER:
626 	    {
627 		struct ethercom *ec = (void *)p;
628 		struct vlanid_list *vlanidp;
629 		uint16_t vid = EVL_VLANOFTAG(nmib->ifvm_tag);
630 
631 		ETHER_LOCK(ec);
632 		SIMPLEQ_FOREACH(vlanidp, &ec->ec_vids, vid_list) {
633 			if (vlanidp->vid == vid) {
634 				SIMPLEQ_REMOVE(&ec->ec_vids, vlanidp,
635 				    vlanid_list, vid_list);
636 				break;
637 			}
638 		}
639 		ETHER_UNLOCK(ec);
640 		if (vlanidp != NULL)
641 			kmem_free(vlanidp, sizeof(*vlanidp));
642 
643 		if (ec->ec_vlan_cb != NULL) {
644 			/*
645 			 * Call ec_vlan_cb(). It will setup VLAN HW filter or
646 			 * HW tagging function.
647 			 */
648 			(void)(*ec->ec_vlan_cb)(ec, vid, false);
649 		}
650 		if (--ec->ec_nvlans == 0) {
651 			IFNET_LOCK(p);
652 			(void)ether_disable_vlan_mtu(p);
653 			IFNET_UNLOCK(p);
654 		}
655 
656 		/* XXX ether_ifdetach must not be called with IFNET_LOCK */
657 		ifv->ifv_stopping = true;
658 		mutex_exit(&ifv->ifv_lock);
659 		IFNET_UNLOCK(ifp);
660 		ether_ifdetach(ifp);
661 		IFNET_LOCK(ifp);
662 		mutex_enter(&ifv->ifv_lock);
663 		ifv->ifv_stopping = false;
664 
665 		/* if_free_sadl must be called with IFNET_LOCK */
666 		if_free_sadl(ifp, 1);
667 
668 		/* Restore vlan_ioctl overwritten by ether_ifdetach */
669 		ifp->if_ioctl = vlan_ioctl;
670 		vlan_reset_linkname(ifp);
671 		break;
672 	    }
673 
674 	default:
675 		panic("%s: impossible", __func__);
676 	}
677 
678 	nmib->ifvm_p = NULL;
679 	ifv->ifv_if.if_mtu = 0;
680 	ifv->ifv_flags = 0;
681 
682 	mutex_enter(&ifv_hash.lock);
683 	PSLIST_WRITER_REMOVE(ifv, ifv_hash);
684 	pserialize_perform(vlan_psz);
685 	mutex_exit(&ifv_hash.lock);
686 	PSLIST_ENTRY_DESTROY(ifv, ifv_hash);
687 	if_linkstate_change_disestablish(p,
688 	    ifv->ifv_linkstate_hook, NULL);
689 
690 	vlan_linkmib_update(ifv, nmib);
691 	if_link_state_change(ifp, LINK_STATE_DOWN);
692 
693 	/*XXX ether_ifdetachhook_disestablish must not called with IFNET_LOCK */
694 	IFNET_UNLOCK(ifp);
695 	ether_ifdetachhook_disestablish(p, ifv->ifv_ifdetach_hook,
696 	    &ifv->ifv_lock);
697 	mutex_exit(&ifv->ifv_lock);
698 	IFNET_LOCK(ifp);
699 
700 	nmib_psref = NULL;
701 	kmem_free(omib, sizeof(*omib));
702 
703 #ifdef INET6
704 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
705 	/* To delete v6 link local addresses */
706 	if (in6_present)
707 		in6_ifdetach(ifp);
708 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
709 #endif
710 
711 	if_down_locked(ifp);
712 	ifp->if_capabilities = 0;
713 	mutex_enter(&ifv->ifv_lock);
714 done:
715 	if (nmib_psref)
716 		psref_target_destroy(nmib_psref, ifvm_psref_class);
717 
718 	return error;
719 }
720 
721 static void
722 vlan_hash_init(void)
723 {
724 
725 	ifv_hash.lists = hashinit(VLAN_TAG_HASH_SIZE, HASH_PSLIST, true,
726 	    &ifv_hash.mask);
727 }
728 
729 static int
730 vlan_hash_fini(void)
731 {
732 	int i;
733 
734 	mutex_enter(&ifv_hash.lock);
735 
736 	for (i = 0; i < ifv_hash.mask + 1; i++) {
737 		if (PSLIST_WRITER_FIRST(&ifv_hash.lists[i], struct ifvlan,
738 		    ifv_hash) != NULL) {
739 			mutex_exit(&ifv_hash.lock);
740 			return EBUSY;
741 		}
742 	}
743 
744 	for (i = 0; i < ifv_hash.mask + 1; i++)
745 		PSLIST_DESTROY(&ifv_hash.lists[i]);
746 
747 	mutex_exit(&ifv_hash.lock);
748 
749 	hashdone(ifv_hash.lists, HASH_PSLIST, ifv_hash.mask);
750 
751 	ifv_hash.lists = NULL;
752 	ifv_hash.mask = 0;
753 
754 	return 0;
755 }
756 
757 static int
758 vlan_tag_hash(uint16_t tag, u_long mask)
759 {
760 	uint32_t hash;
761 
762 	hash = (tag >> 8) ^ tag;
763 	hash = (hash >> 2) ^ hash;
764 
765 	return hash & mask;
766 }
767 
768 static struct ifvlan_linkmib *
769 vlan_getref_linkmib(struct ifvlan *sc, struct psref *psref)
770 {
771 	struct ifvlan_linkmib *mib;
772 	int s;
773 
774 	s = pserialize_read_enter();
775 	mib = atomic_load_consume(&sc->ifv_mib);
776 	if (mib == NULL) {
777 		pserialize_read_exit(s);
778 		return NULL;
779 	}
780 	psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
781 	pserialize_read_exit(s);
782 
783 	return mib;
784 }
785 
786 static void
787 vlan_putref_linkmib(struct ifvlan_linkmib *mib, struct psref *psref)
788 {
789 	if (mib == NULL)
790 		return;
791 	psref_release(psref, &mib->ifvm_psref, ifvm_psref_class);
792 }
793 
794 static struct ifvlan_linkmib *
795 vlan_lookup_tag_psref(struct ifnet *ifp, uint16_t tag, struct psref *psref)
796 {
797 	int idx;
798 	int s;
799 	struct ifvlan *sc;
800 
801 	idx = vlan_tag_hash(tag, ifv_hash.mask);
802 
803 	s = pserialize_read_enter();
804 	PSLIST_READER_FOREACH(sc, &ifv_hash.lists[idx], struct ifvlan,
805 	    ifv_hash) {
806 		struct ifvlan_linkmib *mib = atomic_load_consume(&sc->ifv_mib);
807 		if (mib == NULL)
808 			continue;
809 		if (mib->ifvm_tag != tag)
810 			continue;
811 		if (mib->ifvm_p != ifp)
812 			continue;
813 
814 		psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
815 		pserialize_read_exit(s);
816 		return mib;
817 	}
818 	pserialize_read_exit(s);
819 	return NULL;
820 }
821 
822 static void
823 vlan_linkmib_update(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
824 {
825 	struct ifvlan_linkmib *omib = ifv->ifv_mib;
826 
827 	KASSERT(mutex_owned(&ifv->ifv_lock));
828 
829 	atomic_store_release(&ifv->ifv_mib, nmib);
830 
831 	pserialize_perform(ifv->ifv_psz);
832 	psref_target_destroy(&omib->ifvm_psref, ifvm_psref_class);
833 }
834 
835 /*
836  * Called when a parent interface is detaching; destroy any VLAN
837  * configuration for the parent interface.
838  */
839 static void
840 vlan_ifdetach(void *xifp)
841 {
842 	struct ifnet *ifp;
843 
844 	ifp = (struct ifnet *)xifp;
845 
846 	/* IFNET_LOCK must be held before ifv_lock. */
847 	IFNET_LOCK(ifp);
848 	vlan_unconfig(ifp);
849 	IFNET_UNLOCK(ifp);
850 }
851 
852 static int
853 vlan_set_promisc(struct ifnet *ifp)
854 {
855 	struct ifvlan *ifv = ifp->if_softc;
856 	struct ifvlan_linkmib *mib;
857 	struct psref psref;
858 	int error = 0;
859 	int bound;
860 
861 	bound = curlwp_bind();
862 	mib = vlan_getref_linkmib(ifv, &psref);
863 	if (mib == NULL) {
864 		curlwp_bindx(bound);
865 		return EBUSY;
866 	}
867 
868 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
869 		if ((ifv->ifv_flags & IFVF_PROMISC) == 0) {
870 			error = vlan_safe_ifpromisc(mib->ifvm_p, 1);
871 			if (error == 0)
872 				ifv->ifv_flags |= IFVF_PROMISC;
873 		}
874 	} else {
875 		if ((ifv->ifv_flags & IFVF_PROMISC) != 0) {
876 			error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
877 			if (error == 0)
878 				ifv->ifv_flags &= ~IFVF_PROMISC;
879 		}
880 	}
881 	vlan_putref_linkmib(mib, &psref);
882 	curlwp_bindx(bound);
883 
884 	return error;
885 }
886 
887 static int
888 vlan_ioctl(struct ifnet *ifp, u_long cmd, void *data)
889 {
890 	struct lwp *l = curlwp;
891 	struct ifvlan *ifv = ifp->if_softc;
892 	struct ifaddr *ifa = (struct ifaddr *) data;
893 	struct ifreq *ifr = (struct ifreq *) data;
894 	struct ifnet *pr;
895 	struct ifcapreq *ifcr;
896 	struct vlanreq vlr;
897 	struct ifvlan_linkmib *mib;
898 	struct psref psref;
899 	int error = 0;
900 	int bound;
901 
902 	switch (cmd) {
903 	case SIOCSIFMTU:
904 		bound = curlwp_bind();
905 		mib = vlan_getref_linkmib(ifv, &psref);
906 		if (mib == NULL) {
907 			curlwp_bindx(bound);
908 			error = EBUSY;
909 			break;
910 		}
911 
912 		if (mib->ifvm_p == NULL) {
913 			vlan_putref_linkmib(mib, &psref);
914 			curlwp_bindx(bound);
915 			error = EINVAL;
916 		} else if (
917 		    ifr->ifr_mtu > (mib->ifvm_p->if_mtu - mib->ifvm_mtufudge) ||
918 		    ifr->ifr_mtu < (mib->ifvm_mintu - mib->ifvm_mtufudge)) {
919 			vlan_putref_linkmib(mib, &psref);
920 			curlwp_bindx(bound);
921 			error = EINVAL;
922 		} else {
923 			vlan_putref_linkmib(mib, &psref);
924 			curlwp_bindx(bound);
925 
926 			error = ifioctl_common(ifp, cmd, data);
927 			if (error == ENETRESET)
928 				error = 0;
929 		}
930 
931 		break;
932 
933 	case SIOCSETVLAN:
934 		if ((error = kauth_authorize_network(l->l_cred,
935 		    KAUTH_NETWORK_INTERFACE,
936 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
937 		    NULL)) != 0)
938 			break;
939 		if ((error = copyin(ifr->ifr_data, &vlr, sizeof(vlr))) != 0)
940 			break;
941 
942 		if (vlr.vlr_parent[0] == '\0') {
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 			    (ifp->if_flags & IFF_PROMISC) != 0)
953 				error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
954 
955 			vlan_putref_linkmib(mib, &psref);
956 			curlwp_bindx(bound);
957 
958 			vlan_unconfig(ifp);
959 			break;
960 		}
961 		if (vlr.vlr_tag != EVL_VLANOFTAG(vlr.vlr_tag)) {
962 			error = EINVAL;		 /* check for valid tag */
963 			break;
964 		}
965 		if ((pr = ifunit(vlr.vlr_parent)) == NULL) {
966 			error = ENOENT;
967 			break;
968 		}
969 
970 		error = vlan_config(ifv, pr, vlr.vlr_tag);
971 		if (error != 0)
972 			break;
973 
974 		/* Update promiscuous mode, if necessary. */
975 		vlan_set_promisc(ifp);
976 
977 		ifp->if_flags |= IFF_RUNNING;
978 		break;
979 
980 	case SIOCGETVLAN:
981 		memset(&vlr, 0, sizeof(vlr));
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 		if (mib->ifvm_p != NULL) {
990 			snprintf(vlr.vlr_parent, sizeof(vlr.vlr_parent), "%s",
991 			    mib->ifvm_p->if_xname);
992 			vlr.vlr_tag = mib->ifvm_tag;
993 		}
994 		vlan_putref_linkmib(mib, &psref);
995 		curlwp_bindx(bound);
996 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
997 		break;
998 
999 	case SIOCSIFFLAGS:
1000 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
1001 			break;
1002 		/*
1003 		 * For promiscuous mode, we enable promiscuous mode on
1004 		 * the parent if we need promiscuous on the VLAN interface.
1005 		 */
1006 		bound = curlwp_bind();
1007 		mib = vlan_getref_linkmib(ifv, &psref);
1008 		if (mib == NULL) {
1009 			curlwp_bindx(bound);
1010 			error = EBUSY;
1011 			break;
1012 		}
1013 
1014 		if (mib->ifvm_p != NULL)
1015 			error = vlan_set_promisc(ifp);
1016 		vlan_putref_linkmib(mib, &psref);
1017 		curlwp_bindx(bound);
1018 		break;
1019 
1020 	case SIOCADDMULTI:
1021 		mutex_enter(&ifv->ifv_lock);
1022 		mib = ifv->ifv_mib;
1023 		if (mib == NULL) {
1024 			error = EBUSY;
1025 			mutex_exit(&ifv->ifv_lock);
1026 			break;
1027 		}
1028 
1029 		error = (mib->ifvm_p != NULL) ?
1030 		    (*mib->ifvm_msw->vmsw_addmulti)(ifv, ifr) : EINVAL;
1031 		mib = NULL;
1032 		mutex_exit(&ifv->ifv_lock);
1033 		break;
1034 
1035 	case SIOCDELMULTI:
1036 		mutex_enter(&ifv->ifv_lock);
1037 		mib = ifv->ifv_mib;
1038 		if (mib == NULL) {
1039 			error = EBUSY;
1040 			mutex_exit(&ifv->ifv_lock);
1041 			break;
1042 		}
1043 		error = (mib->ifvm_p != NULL) ?
1044 		    (*mib->ifvm_msw->vmsw_delmulti)(ifv, ifr) : EINVAL;
1045 		mib = NULL;
1046 		mutex_exit(&ifv->ifv_lock);
1047 		break;
1048 
1049 	case SIOCSIFCAP:
1050 		ifcr = data;
1051 		/* make sure caps are enabled on parent */
1052 		bound = curlwp_bind();
1053 		mib = vlan_getref_linkmib(ifv, &psref);
1054 		if (mib == NULL) {
1055 			curlwp_bindx(bound);
1056 			error = EBUSY;
1057 			break;
1058 		}
1059 
1060 		if (mib->ifvm_p == NULL) {
1061 			vlan_putref_linkmib(mib, &psref);
1062 			curlwp_bindx(bound);
1063 			error = EINVAL;
1064 			break;
1065 		}
1066 		if ((mib->ifvm_p->if_capenable & ifcr->ifcr_capenable) !=
1067 		    ifcr->ifcr_capenable) {
1068 			vlan_putref_linkmib(mib, &psref);
1069 			curlwp_bindx(bound);
1070 			error = EINVAL;
1071 			break;
1072 		}
1073 
1074 		vlan_putref_linkmib(mib, &psref);
1075 		curlwp_bindx(bound);
1076 
1077 		if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
1078 			error = 0;
1079 		break;
1080 	case SIOCINITIFADDR:
1081 		bound = curlwp_bind();
1082 		mib = vlan_getref_linkmib(ifv, &psref);
1083 		if (mib == NULL) {
1084 			curlwp_bindx(bound);
1085 			error = EBUSY;
1086 			break;
1087 		}
1088 
1089 		if (mib->ifvm_p == NULL) {
1090 			error = EINVAL;
1091 			vlan_putref_linkmib(mib, &psref);
1092 			curlwp_bindx(bound);
1093 			break;
1094 		}
1095 		vlan_putref_linkmib(mib, &psref);
1096 		curlwp_bindx(bound);
1097 
1098 		ifp->if_flags |= IFF_UP;
1099 #ifdef INET
1100 		if (ifa->ifa_addr->sa_family == AF_INET)
1101 			arp_ifinit(ifp, ifa);
1102 #endif
1103 		break;
1104 
1105 	default:
1106 		error = ether_ioctl(ifp, cmd, data);
1107 	}
1108 
1109 	return error;
1110 }
1111 
1112 static int
1113 vlan_ether_addmulti(struct ifvlan *ifv, struct ifreq *ifr)
1114 {
1115 	const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr);
1116 	struct vlan_mc_entry *mc;
1117 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1118 	struct ifvlan_linkmib *mib;
1119 	int error;
1120 
1121 	KASSERT(mutex_owned(&ifv->ifv_lock));
1122 
1123 	if (sa->sa_len > sizeof(struct sockaddr_storage))
1124 		return EINVAL;
1125 
1126 	error = ether_addmulti(sa, &ifv->ifv_ec);
1127 	if (error != ENETRESET)
1128 		return error;
1129 
1130 	/*
1131 	 * This is a new multicast address.  We have to tell parent
1132 	 * about it.  Also, remember this multicast address so that
1133 	 * we can delete it on unconfigure.
1134 	 */
1135 	mc = malloc(sizeof(struct vlan_mc_entry), M_DEVBUF, M_NOWAIT);
1136 	if (mc == NULL) {
1137 		error = ENOMEM;
1138 		goto alloc_failed;
1139 	}
1140 
1141 	/*
1142 	 * Since ether_addmulti() returned ENETRESET, the following two
1143 	 * statements shouldn't fail. Here ifv_ec is implicitly protected
1144 	 * by the ifv_lock lock.
1145 	 */
1146 	error = ether_multiaddr(sa, addrlo, addrhi);
1147 	KASSERT(error == 0);
1148 
1149 	ETHER_LOCK(&ifv->ifv_ec);
1150 	mc->mc_enm = ether_lookup_multi(addrlo, addrhi, &ifv->ifv_ec);
1151 	ETHER_UNLOCK(&ifv->ifv_ec);
1152 
1153 	KASSERT(mc->mc_enm != NULL);
1154 
1155 	memcpy(&mc->mc_addr, sa, sa->sa_len);
1156 	LIST_INSERT_HEAD(&ifv->ifv_mc_listhead, mc, mc_entries);
1157 
1158 	mib = ifv->ifv_mib;
1159 
1160 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1161 	error = if_mcast_op(mib->ifvm_p, SIOCADDMULTI, sa);
1162 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1163 
1164 	if (error != 0)
1165 		goto ioctl_failed;
1166 	return error;
1167 
1168 ioctl_failed:
1169 	LIST_REMOVE(mc, mc_entries);
1170 	free(mc, M_DEVBUF);
1171 
1172 alloc_failed:
1173 	(void)ether_delmulti(sa, &ifv->ifv_ec);
1174 	return error;
1175 }
1176 
1177 static int
1178 vlan_ether_delmulti(struct ifvlan *ifv, struct ifreq *ifr)
1179 {
1180 	const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr);
1181 	struct ether_multi *enm;
1182 	struct vlan_mc_entry *mc;
1183 	struct ifvlan_linkmib *mib;
1184 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1185 	int error;
1186 
1187 	KASSERT(mutex_owned(&ifv->ifv_lock));
1188 
1189 	/*
1190 	 * Find a key to lookup vlan_mc_entry.  We have to do this
1191 	 * before calling ether_delmulti for obvious reasons.
1192 	 */
1193 	if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0)
1194 		return error;
1195 
1196 	ETHER_LOCK(&ifv->ifv_ec);
1197 	enm = ether_lookup_multi(addrlo, addrhi, &ifv->ifv_ec);
1198 	ETHER_UNLOCK(&ifv->ifv_ec);
1199 	if (enm == NULL)
1200 		return EINVAL;
1201 
1202 	LIST_FOREACH(mc, &ifv->ifv_mc_listhead, mc_entries) {
1203 		if (mc->mc_enm == enm)
1204 			break;
1205 	}
1206 
1207 	/* We woun't delete entries we didn't add */
1208 	if (mc == NULL)
1209 		return EINVAL;
1210 
1211 	error = ether_delmulti(sa, &ifv->ifv_ec);
1212 	if (error != ENETRESET)
1213 		return error;
1214 
1215 	/* We no longer use this multicast address.  Tell parent so. */
1216 	mib = ifv->ifv_mib;
1217 	error = if_mcast_op(mib->ifvm_p, SIOCDELMULTI, sa);
1218 
1219 	if (error == 0) {
1220 		/* And forget about this address. */
1221 		LIST_REMOVE(mc, mc_entries);
1222 		free(mc, M_DEVBUF);
1223 	} else {
1224 		(void)ether_addmulti(sa, &ifv->ifv_ec);
1225 	}
1226 
1227 	return error;
1228 }
1229 
1230 /*
1231  * Delete any multicast address we have asked to add from parent
1232  * interface.  Called when the vlan is being unconfigured.
1233  */
1234 static void
1235 vlan_ether_purgemulti(struct ifvlan *ifv)
1236 {
1237 	struct vlan_mc_entry *mc;
1238 	struct ifvlan_linkmib *mib;
1239 
1240 	KASSERT(mutex_owned(&ifv->ifv_lock));
1241 	mib = ifv->ifv_mib;
1242 	if (mib == NULL) {
1243 		return;
1244 	}
1245 
1246 	while ((mc = LIST_FIRST(&ifv->ifv_mc_listhead)) != NULL) {
1247 		(void)if_mcast_op(mib->ifvm_p, SIOCDELMULTI,
1248 		    sstocsa(&mc->mc_addr));
1249 		LIST_REMOVE(mc, mc_entries);
1250 		free(mc, M_DEVBUF);
1251 	}
1252 }
1253 
1254 static void
1255 vlan_start(struct ifnet *ifp)
1256 {
1257 	struct ifvlan *ifv = ifp->if_softc;
1258 	struct ifnet *p;
1259 	struct ethercom *ec;
1260 	struct mbuf *m;
1261 	struct ifvlan_linkmib *mib;
1262 	struct psref psref;
1263 	struct ether_header *eh;
1264 	int error;
1265 
1266 	mib = vlan_getref_linkmib(ifv, &psref);
1267 	if (mib == NULL)
1268 		return;
1269 
1270 	if (__predict_false(mib->ifvm_p == NULL)) {
1271 		vlan_putref_linkmib(mib, &psref);
1272 		return;
1273 	}
1274 
1275 	p = mib->ifvm_p;
1276 	ec = (void *)mib->ifvm_p;
1277 
1278 	ifp->if_flags |= IFF_OACTIVE;
1279 
1280 	for (;;) {
1281 		IFQ_DEQUEUE(&ifp->if_snd, m);
1282 		if (m == NULL)
1283 			break;
1284 
1285 		if (m->m_len < sizeof(*eh)) {
1286 			m = m_pullup(m, sizeof(*eh));
1287 			if (m == NULL) {
1288 				if_statinc(ifp, if_oerrors);
1289 				continue;
1290 			}
1291 		}
1292 
1293 		eh = mtod(m, struct ether_header *);
1294 		if (ntohs(eh->ether_type) == ETHERTYPE_VLAN) {
1295 			m_freem(m);
1296 			if_statinc(ifp, if_noproto);
1297 			continue;
1298 		}
1299 
1300 #ifdef ALTQ
1301 		/*
1302 		 * KERNEL_LOCK is required for ALTQ even if NET_MPSAFE is
1303 		 * defined.
1304 		 */
1305 		KERNEL_LOCK(1, NULL);
1306 		/*
1307 		 * If ALTQ is enabled on the parent interface, do
1308 		 * classification; the queueing discipline might
1309 		 * not require classification, but might require
1310 		 * the address family/header pointer in the pktattr.
1311 		 */
1312 		if (ALTQ_IS_ENABLED(&p->if_snd)) {
1313 			switch (p->if_type) {
1314 			case IFT_ETHER:
1315 				altq_etherclassify(&p->if_snd, m);
1316 				break;
1317 			default:
1318 				panic("%s: impossible (altq)", __func__);
1319 			}
1320 		}
1321 		KERNEL_UNLOCK_ONE(NULL);
1322 #endif /* ALTQ */
1323 
1324 		bpf_mtap(ifp, m, BPF_D_OUT);
1325 		/*
1326 		 * If the parent can insert the tag itself, just mark
1327 		 * the tag in the mbuf header.
1328 		 */
1329 		if (ec->ec_capenable & ETHERCAP_VLAN_HWTAGGING) {
1330 			vlan_set_tag(m, mib->ifvm_tag);
1331 		} else {
1332 			/*
1333 			 * insert the tag ourselves
1334 			 */
1335 			M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1336 			if (m == NULL) {
1337 				printf("%s: unable to prepend encap header",
1338 				    p->if_xname);
1339 				if_statinc(ifp, if_oerrors);
1340 				continue;
1341 			}
1342 
1343 			switch (p->if_type) {
1344 			case IFT_ETHER:
1345 			    {
1346 				struct ether_vlan_header *evl;
1347 
1348 				if (m->m_len < sizeof(struct ether_vlan_header))
1349 					m = m_pullup(m,
1350 					    sizeof(struct ether_vlan_header));
1351 				if (m == NULL) {
1352 					printf("%s: unable to pullup encap "
1353 					    "header", p->if_xname);
1354 					if_statinc(ifp, if_oerrors);
1355 					continue;
1356 				}
1357 
1358 				/*
1359 				 * Transform the Ethernet header into an
1360 				 * Ethernet header with 802.1Q encapsulation.
1361 				 */
1362 				memmove(mtod(m, void *),
1363 				    mtod(m, char *) + mib->ifvm_encaplen,
1364 				    sizeof(struct ether_header));
1365 				evl = mtod(m, struct ether_vlan_header *);
1366 				evl->evl_proto = evl->evl_encap_proto;
1367 				evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1368 				evl->evl_tag = htons(mib->ifvm_tag);
1369 
1370 				/*
1371 				 * To cater for VLAN-aware layer 2 ethernet
1372 				 * switches which may need to strip the tag
1373 				 * before forwarding the packet, make sure
1374 				 * the packet+tag is at least 68 bytes long.
1375 				 * This is necessary because our parent will
1376 				 * only pad to 64 bytes (ETHER_MIN_LEN) and
1377 				 * some switches will not pad by themselves
1378 				 * after deleting a tag.
1379 				 */
1380 				const size_t min_data_len = ETHER_MIN_LEN -
1381 				    ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
1382 				if (m->m_pkthdr.len < min_data_len) {
1383 					m_copyback(m, m->m_pkthdr.len,
1384 					    min_data_len - m->m_pkthdr.len,
1385 					    vlan_zero_pad_buff);
1386 				}
1387 				break;
1388 			    }
1389 
1390 			default:
1391 				panic("%s: impossible", __func__);
1392 			}
1393 		}
1394 
1395 		if ((p->if_flags & IFF_RUNNING) == 0) {
1396 			m_freem(m);
1397 			continue;
1398 		}
1399 
1400 		error = if_transmit_lock(p, m);
1401 		if (error) {
1402 			/* mbuf is already freed */
1403 			if_statinc(ifp, if_oerrors);
1404 			continue;
1405 		}
1406 		if_statinc(ifp, if_opackets);
1407 	}
1408 
1409 	ifp->if_flags &= ~IFF_OACTIVE;
1410 
1411 	/* Remove reference to mib before release */
1412 	vlan_putref_linkmib(mib, &psref);
1413 }
1414 
1415 static int
1416 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1417 {
1418 	struct ifvlan *ifv = ifp->if_softc;
1419 	struct ifnet *p;
1420 	struct ethercom *ec;
1421 	struct ifvlan_linkmib *mib;
1422 	struct psref psref;
1423 	struct ether_header *eh;
1424 	int error;
1425 	size_t pktlen = m->m_pkthdr.len;
1426 	bool mcast = (m->m_flags & M_MCAST) != 0;
1427 
1428 	if (m->m_len < sizeof(*eh)) {
1429 		m = m_pullup(m, sizeof(*eh));
1430 		if (m == NULL) {
1431 			if_statinc(ifp, if_oerrors);
1432 			return ENOBUFS;
1433 		}
1434 	}
1435 
1436 	eh = mtod(m, struct ether_header *);
1437 	if (ntohs(eh->ether_type) == ETHERTYPE_VLAN) {
1438 		m_freem(m);
1439 		if_statinc(ifp, if_noproto);
1440 		return EPROTONOSUPPORT;
1441 	}
1442 
1443 	mib = vlan_getref_linkmib(ifv, &psref);
1444 	if (mib == NULL) {
1445 		m_freem(m);
1446 		return ENETDOWN;
1447 	}
1448 
1449 	if (__predict_false(mib->ifvm_p == NULL)) {
1450 		vlan_putref_linkmib(mib, &psref);
1451 		m_freem(m);
1452 		return ENETDOWN;
1453 	}
1454 
1455 	p = mib->ifvm_p;
1456 	ec = (void *)mib->ifvm_p;
1457 
1458 	bpf_mtap(ifp, m, BPF_D_OUT);
1459 
1460 	if ((error = pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_OUT)) != 0)
1461 		goto out;
1462 	if (m == NULL)
1463 		goto out;
1464 
1465 	/*
1466 	 * If the parent can insert the tag itself, just mark
1467 	 * the tag in the mbuf header.
1468 	 */
1469 	if (ec->ec_capenable & ETHERCAP_VLAN_HWTAGGING) {
1470 		vlan_set_tag(m, mib->ifvm_tag);
1471 	} else {
1472 		/*
1473 		 * insert the tag ourselves
1474 		 */
1475 		M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1476 		if (m == NULL) {
1477 			printf("%s: unable to prepend encap header",
1478 			    p->if_xname);
1479 			if_statinc(ifp, if_oerrors);
1480 			error = ENOBUFS;
1481 			goto out;
1482 		}
1483 
1484 		switch (p->if_type) {
1485 		case IFT_ETHER:
1486 		    {
1487 			struct ether_vlan_header *evl;
1488 
1489 			if (m->m_len < sizeof(struct ether_vlan_header))
1490 				m = m_pullup(m,
1491 				    sizeof(struct ether_vlan_header));
1492 			if (m == NULL) {
1493 				printf("%s: unable to pullup encap "
1494 				    "header", p->if_xname);
1495 				if_statinc(ifp, if_oerrors);
1496 				error = ENOBUFS;
1497 				goto out;
1498 			}
1499 
1500 			/*
1501 			 * Transform the Ethernet header into an
1502 			 * Ethernet header with 802.1Q encapsulation.
1503 			 */
1504 			memmove(mtod(m, void *),
1505 			    mtod(m, char *) + mib->ifvm_encaplen,
1506 			    sizeof(struct ether_header));
1507 			evl = mtod(m, struct ether_vlan_header *);
1508 			evl->evl_proto = evl->evl_encap_proto;
1509 			evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1510 			evl->evl_tag = htons(mib->ifvm_tag);
1511 
1512 			/*
1513 			 * To cater for VLAN-aware layer 2 ethernet
1514 			 * switches which may need to strip the tag
1515 			 * before forwarding the packet, make sure
1516 			 * the packet+tag is at least 68 bytes long.
1517 			 * This is necessary because our parent will
1518 			 * only pad to 64 bytes (ETHER_MIN_LEN) and
1519 			 * some switches will not pad by themselves
1520 			 * after deleting a tag.
1521 			 */
1522 			const size_t min_data_len = ETHER_MIN_LEN -
1523 			    ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
1524 			if (m->m_pkthdr.len < min_data_len) {
1525 				m_copyback(m, m->m_pkthdr.len,
1526 				    min_data_len - m->m_pkthdr.len,
1527 				    vlan_zero_pad_buff);
1528 			}
1529 			break;
1530 		    }
1531 
1532 		default:
1533 			panic("%s: impossible", __func__);
1534 		}
1535 	}
1536 
1537 	if ((p->if_flags & IFF_RUNNING) == 0) {
1538 		m_freem(m);
1539 		error = ENETDOWN;
1540 		goto out;
1541 	}
1542 
1543 	error = if_transmit_lock(p, m);
1544 	net_stat_ref_t nsr = IF_STAT_GETREF(ifp);
1545 	if (error) {
1546 		/* mbuf is already freed */
1547 		if_statinc_ref(nsr, if_oerrors);
1548 	} else {
1549 		if_statinc_ref(nsr, if_opackets);
1550 		if_statadd_ref(nsr, if_obytes, pktlen);
1551 		if (mcast)
1552 			if_statinc_ref(nsr, if_omcasts);
1553 	}
1554 	IF_STAT_PUTREF(ifp);
1555 
1556 out:
1557 	/* Remove reference to mib before release */
1558 	vlan_putref_linkmib(mib, &psref);
1559 	return error;
1560 }
1561 
1562 /*
1563  * Given an Ethernet frame, find a valid vlan interface corresponding to the
1564  * given source interface and tag, then run the real packet through the
1565  * parent's input routine.
1566  */
1567 void
1568 vlan_input(struct ifnet *ifp, struct mbuf *m)
1569 {
1570 	struct ifvlan *ifv;
1571 	uint16_t vid;
1572 	struct ifvlan_linkmib *mib;
1573 	struct psref psref;
1574 	bool have_vtag;
1575 
1576 	have_vtag = vlan_has_tag(m);
1577 	if (have_vtag) {
1578 		vid = EVL_VLANOFTAG(vlan_get_tag(m));
1579 		m->m_flags &= ~M_VLANTAG;
1580 	} else {
1581 		struct ether_vlan_header *evl;
1582 
1583 		if (ifp->if_type != IFT_ETHER) {
1584 			panic("%s: impossible", __func__);
1585 		}
1586 
1587 		if (m->m_len < sizeof(struct ether_vlan_header) &&
1588 		    (m = m_pullup(m,
1589 		     sizeof(struct ether_vlan_header))) == NULL) {
1590 			printf("%s: no memory for VLAN header, "
1591 			    "dropping packet.\n", ifp->if_xname);
1592 			return;
1593 		}
1594 
1595 		if (m_makewritable(&m, 0,
1596 		    sizeof(struct ether_vlan_header), M_DONTWAIT)) {
1597 			m_freem(m);
1598 			if_statinc(ifp, if_ierrors);
1599 			return;
1600 		}
1601 
1602 		evl = mtod(m, struct ether_vlan_header *);
1603 		KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN);
1604 
1605 		vid = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1606 
1607 		/*
1608 		 * Restore the original ethertype.  We'll remove
1609 		 * the encapsulation after we've found the vlan
1610 		 * interface corresponding to the tag.
1611 		 */
1612 		evl->evl_encap_proto = evl->evl_proto;
1613 	}
1614 
1615 	mib = vlan_lookup_tag_psref(ifp, vid, &psref);
1616 	if (mib == NULL) {
1617 		m_freem(m);
1618 		if_statinc(ifp, if_noproto);
1619 		return;
1620 	}
1621 	KASSERT(mib->ifvm_encaplen == ETHER_VLAN_ENCAP_LEN);
1622 
1623 	ifv = mib->ifvm_ifvlan;
1624 	if ((ifv->ifv_if.if_flags & (IFF_UP | IFF_RUNNING)) !=
1625 	    (IFF_UP | IFF_RUNNING)) {
1626 		m_freem(m);
1627 		if_statinc(ifp, if_noproto);
1628 		goto out;
1629 	}
1630 
1631 	/*
1632 	 * Now, remove the encapsulation header.  The original
1633 	 * header has already been fixed up above.
1634 	 */
1635 	if (!have_vtag) {
1636 		memmove(mtod(m, char *) + mib->ifvm_encaplen,
1637 		    mtod(m, void *), sizeof(struct ether_header));
1638 		m_adj(m, mib->ifvm_encaplen);
1639 	}
1640 
1641 	/*
1642 	 * Drop promiscuously received packets if we are not in
1643 	 * promiscuous mode
1644 	 */
1645 	if ((m->m_flags & (M_BCAST | M_MCAST)) == 0 &&
1646 	    (ifp->if_flags & IFF_PROMISC) &&
1647 	    (ifv->ifv_if.if_flags & IFF_PROMISC) == 0) {
1648 		struct ether_header *eh;
1649 
1650 		eh = mtod(m, struct ether_header *);
1651 		if (memcmp(CLLADDR(ifv->ifv_if.if_sadl),
1652 		    eh->ether_dhost, ETHER_ADDR_LEN) != 0) {
1653 			m_freem(m);
1654 			if_statinc(&ifv->ifv_if, if_ierrors);
1655 			goto out;
1656 		}
1657 	}
1658 
1659 	m_set_rcvif(m, &ifv->ifv_if);
1660 
1661 	if (pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_IN) != 0)
1662 		goto out;
1663 	if (m == NULL)
1664 		goto out;
1665 
1666 	m->m_flags &= ~M_PROMISC;
1667 	if_input(&ifv->ifv_if, m);
1668 out:
1669 	vlan_putref_linkmib(mib, &psref);
1670 }
1671 
1672 /*
1673  * If the parent link state changed, the vlan link state should change also.
1674  */
1675 static void
1676 vlan_link_state_changed(void *xifv)
1677 {
1678 	struct ifvlan *ifv = xifv;
1679 	struct ifnet *ifp, *p;
1680 	struct ifvlan_linkmib *mib;
1681 	struct psref psref;
1682 
1683 	mib = vlan_getref_linkmib(ifv, &psref);
1684 	if (mib == NULL)
1685 		return;
1686 
1687 	if (mib->ifvm_p == NULL) {
1688 		vlan_putref_linkmib(mib, &psref);
1689 		return;
1690 	}
1691 
1692 	ifp = &ifv->ifv_if;
1693 	p = mib->ifvm_p;
1694 	if_link_state_change(ifp, p->if_link_state);
1695 
1696 	vlan_putref_linkmib(mib, &psref);
1697 }
1698 
1699 /*
1700  * Module infrastructure
1701  */
1702 #include "if_module.h"
1703 
1704 IF_MODULE(MODULE_CLASS_DRIVER, vlan, NULL)
1705