xref: /netbsd-src/sys/net/if_vlan.c (revision 87d689fb734c654d2486f87f7be32f1b53ecdbec)
1 /*	$NetBSD: if_vlan.c,v 1.122 2018/01/14 16:50:37 maxv 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 1.122 2018/01/14 16:50:37 maxv Exp $");
82 
83 #ifdef _KERNEL_OPT
84 #include "opt_inet.h"
85 #include "opt_net_mpsafe.h"
86 #endif
87 
88 #include <sys/param.h>
89 #include <sys/systm.h>
90 #include <sys/kernel.h>
91 #include <sys/mbuf.h>
92 #include <sys/queue.h>
93 #include <sys/socket.h>
94 #include <sys/sockio.h>
95 #include <sys/systm.h>
96 #include <sys/proc.h>
97 #include <sys/kauth.h>
98 #include <sys/mutex.h>
99 #include <sys/kmem.h>
100 #include <sys/cpu.h>
101 #include <sys/pserialize.h>
102 #include <sys/psref.h>
103 #include <sys/pslist.h>
104 #include <sys/atomic.h>
105 #include <sys/device.h>
106 #include <sys/module.h>
107 
108 #include <net/bpf.h>
109 #include <net/if.h>
110 #include <net/if_dl.h>
111 #include <net/if_types.h>
112 #include <net/if_ether.h>
113 #include <net/if_vlanvar.h>
114 
115 #ifdef INET
116 #include <netinet/in.h>
117 #include <netinet/if_inarp.h>
118 #endif
119 #ifdef INET6
120 #include <netinet6/in6_ifattach.h>
121 #include <netinet6/in6_var.h>
122 #endif
123 
124 #include "ioconf.h"
125 
126 struct vlan_mc_entry {
127 	LIST_ENTRY(vlan_mc_entry)	mc_entries;
128 	/*
129 	 * A key to identify this entry.  The mc_addr below can't be
130 	 * used since multiple sockaddr may mapped into the same
131 	 * ether_multi (e.g., AF_UNSPEC).
132 	 */
133 	union {
134 		struct ether_multi	*mcu_enm;
135 	} mc_u;
136 	struct sockaddr_storage		mc_addr;
137 };
138 
139 #define	mc_enm		mc_u.mcu_enm
140 
141 
142 struct ifvlan_linkmib {
143 	struct ifvlan *ifvm_ifvlan;
144 	const struct vlan_multisw *ifvm_msw;
145 	int	ifvm_encaplen;	/* encapsulation length */
146 	int	ifvm_mtufudge;	/* MTU fudged by this much */
147 	int	ifvm_mintu;	/* min transmission unit */
148 	uint16_t ifvm_proto;	/* encapsulation ethertype */
149 	uint16_t ifvm_tag;	/* tag to apply on packets */
150 	struct ifnet *ifvm_p;		/* parent interface of this vlan */
151 
152 	struct psref_target ifvm_psref;
153 };
154 
155 struct ifvlan {
156 	union {
157 		struct ethercom ifvu_ec;
158 	} ifv_u;
159 	struct ifvlan_linkmib *ifv_mib;	/*
160 					 * reader must use vlan_getref_linkmib()
161 					 * instead of direct dereference
162 					 */
163 	kmutex_t ifv_lock;		/* writer lock for ifv_mib */
164 
165 	LIST_HEAD(__vlan_mchead, vlan_mc_entry) ifv_mc_listhead;
166 	LIST_ENTRY(ifvlan) ifv_list;
167 	struct pslist_entry ifv_hash;
168 	int ifv_flags;
169 };
170 
171 #define	IFVF_PROMISC	0x01		/* promiscuous mode enabled */
172 
173 #define	ifv_ec		ifv_u.ifvu_ec
174 
175 #define	ifv_if		ifv_ec.ec_if
176 
177 #define	ifv_msw		ifv_mib.ifvm_msw
178 #define	ifv_encaplen	ifv_mib.ifvm_encaplen
179 #define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
180 #define	ifv_mintu	ifv_mib.ifvm_mintu
181 #define	ifv_tag		ifv_mib.ifvm_tag
182 
183 struct vlan_multisw {
184 	int	(*vmsw_addmulti)(struct ifvlan *, struct ifreq *);
185 	int	(*vmsw_delmulti)(struct ifvlan *, struct ifreq *);
186 	void	(*vmsw_purgemulti)(struct ifvlan *);
187 };
188 
189 static int	vlan_ether_addmulti(struct ifvlan *, struct ifreq *);
190 static int	vlan_ether_delmulti(struct ifvlan *, struct ifreq *);
191 static void	vlan_ether_purgemulti(struct ifvlan *);
192 
193 const struct vlan_multisw vlan_ether_multisw = {
194 	vlan_ether_addmulti,
195 	vlan_ether_delmulti,
196 	vlan_ether_purgemulti,
197 };
198 
199 static int	vlan_clone_create(struct if_clone *, int);
200 static int	vlan_clone_destroy(struct ifnet *);
201 static int	vlan_config(struct ifvlan *, struct ifnet *,
202     uint16_t);
203 static int	vlan_ioctl(struct ifnet *, u_long, void *);
204 static void	vlan_start(struct ifnet *);
205 static int	vlan_transmit(struct ifnet *, struct mbuf *);
206 static void	vlan_unconfig(struct ifnet *);
207 static int	vlan_unconfig_locked(struct ifvlan *,
208     struct ifvlan_linkmib *);
209 static void	vlan_hash_init(void);
210 static int	vlan_hash_fini(void);
211 static int	vlan_tag_hash(uint16_t, u_long);
212 static struct ifvlan_linkmib*	vlan_getref_linkmib(struct ifvlan *,
213     struct psref *);
214 static void	vlan_putref_linkmib(struct ifvlan_linkmib *,
215     struct psref *);
216 static void	vlan_linkmib_update(struct ifvlan *,
217     struct ifvlan_linkmib *);
218 static struct ifvlan_linkmib*	vlan_lookup_tag_psref(struct ifnet *,
219     uint16_t, struct psref *);
220 
221 LIST_HEAD(vlan_ifvlist, ifvlan);
222 static struct {
223 	kmutex_t lock;
224 	struct vlan_ifvlist list;
225 } ifv_list __cacheline_aligned;
226 
227 
228 #if !defined(VLAN_TAG_HASH_SIZE)
229 #define VLAN_TAG_HASH_SIZE 32
230 #endif
231 static struct {
232 	kmutex_t lock;
233 	struct pslist_head *lists;
234 	u_long mask;
235 } ifv_hash __cacheline_aligned = {
236 	.lists = NULL,
237 	.mask = 0,
238 };
239 
240 pserialize_t vlan_psz __read_mostly;
241 static struct psref_class *ifvm_psref_class __read_mostly;
242 
243 struct if_clone vlan_cloner =
244     IF_CLONE_INITIALIZER("vlan", vlan_clone_create, vlan_clone_destroy);
245 
246 /* Used to pad ethernet frames with < ETHER_MIN_LEN bytes */
247 static char vlan_zero_pad_buff[ETHER_MIN_LEN];
248 
249 static inline int
250 vlan_safe_ifpromisc(struct ifnet *ifp, int pswitch)
251 {
252 	int e;
253 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
254 	e = ifpromisc(ifp, pswitch);
255 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
256 	return e;
257 }
258 
259 static inline int
260 vlan_safe_ifpromisc_locked(struct ifnet *ifp, int pswitch)
261 {
262 	int e;
263 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
264 	e = ifpromisc_locked(ifp, pswitch);
265 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
266 	return e;
267 }
268 
269 void
270 vlanattach(int n)
271 {
272 
273 	/*
274 	 * Nothing to do here, initialization is handled by the
275 	 * module initialization code in vlaninit() below).
276 	 */
277 }
278 
279 static void
280 vlaninit(void)
281 {
282 	mutex_init(&ifv_list.lock, MUTEX_DEFAULT, IPL_NONE);
283 	LIST_INIT(&ifv_list.list);
284 
285 	mutex_init(&ifv_hash.lock, MUTEX_DEFAULT, IPL_NONE);
286 	vlan_psz = pserialize_create();
287 	ifvm_psref_class = psref_class_create("vlanlinkmib", IPL_SOFTNET);
288 	if_clone_attach(&vlan_cloner);
289 
290 	vlan_hash_init();
291 }
292 
293 static int
294 vlandetach(void)
295 {
296 	int error = 0;
297 
298 	mutex_enter(&ifv_list.lock);
299 	if (!LIST_EMPTY(&ifv_list.list)) {
300 		mutex_exit(&ifv_list.lock);
301 		return EBUSY;
302 	}
303 	mutex_exit(&ifv_list.lock);
304 
305 	error = vlan_hash_fini();
306 	if (error != 0)
307 		return error;
308 
309 	if_clone_detach(&vlan_cloner);
310 	psref_class_destroy(ifvm_psref_class);
311 	pserialize_destroy(vlan_psz);
312 	mutex_destroy(&ifv_hash.lock);
313 	mutex_destroy(&ifv_list.lock);
314 
315 	return 0;
316 }
317 
318 static void
319 vlan_reset_linkname(struct ifnet *ifp)
320 {
321 
322 	/*
323 	 * We start out with a "802.1Q VLAN" type and zero-length
324 	 * addresses.  When we attach to a parent interface, we
325 	 * inherit its type, address length, address, and data link
326 	 * type.
327 	 */
328 
329 	ifp->if_type = IFT_L2VLAN;
330 	ifp->if_addrlen = 0;
331 	ifp->if_dlt = DLT_NULL;
332 	if_alloc_sadl(ifp);
333 }
334 
335 static int
336 vlan_clone_create(struct if_clone *ifc, int unit)
337 {
338 	struct ifvlan *ifv;
339 	struct ifnet *ifp;
340 	struct ifvlan_linkmib *mib;
341 	int rv;
342 
343 	ifv = malloc(sizeof(struct ifvlan), M_DEVBUF, M_WAITOK|M_ZERO);
344 	mib = kmem_zalloc(sizeof(struct ifvlan_linkmib), KM_SLEEP);
345 	ifp = &ifv->ifv_if;
346 	LIST_INIT(&ifv->ifv_mc_listhead);
347 
348 	mib->ifvm_ifvlan = ifv;
349 	mib->ifvm_p = NULL;
350 	psref_target_init(&mib->ifvm_psref, ifvm_psref_class);
351 
352 	mutex_init(&ifv->ifv_lock, MUTEX_DEFAULT, IPL_NONE);
353 	ifv->ifv_mib = mib;
354 
355 	mutex_enter(&ifv_list.lock);
356 	LIST_INSERT_HEAD(&ifv_list.list, ifv, ifv_list);
357 	mutex_exit(&ifv_list.lock);
358 
359 	if_initname(ifp, ifc->ifc_name, unit);
360 	ifp->if_softc = ifv;
361 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
362 	ifp->if_extflags = IFEF_NO_LINK_STATE_CHANGE;
363 #ifdef NET_MPSAFE
364 	ifp->if_extflags |= IFEF_MPSAFE;
365 #endif
366 	ifp->if_start = vlan_start;
367 	ifp->if_transmit = vlan_transmit;
368 	ifp->if_ioctl = vlan_ioctl;
369 	IFQ_SET_READY(&ifp->if_snd);
370 
371 	rv = if_initialize(ifp);
372 	if (rv != 0) {
373 		aprint_error("%s: if_initialize failed(%d)\n", ifp->if_xname,
374 		    rv);
375 		goto fail;
376 	}
377 
378 	vlan_reset_linkname(ifp);
379 	if_register(ifp);
380 	return 0;
381 
382 fail:
383 	mutex_enter(&ifv_list.lock);
384 	LIST_REMOVE(ifv, ifv_list);
385 	mutex_exit(&ifv_list.lock);
386 
387 	mutex_destroy(&ifv->ifv_lock);
388 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
389 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
390 	free(ifv, M_DEVBUF);
391 
392 	return rv;
393 }
394 
395 static int
396 vlan_clone_destroy(struct ifnet *ifp)
397 {
398 	struct ifvlan *ifv = ifp->if_softc;
399 
400 	mutex_enter(&ifv_list.lock);
401 	LIST_REMOVE(ifv, ifv_list);
402 	mutex_exit(&ifv_list.lock);
403 
404 	IFNET_LOCK(ifp);
405 	vlan_unconfig(ifp);
406 	IFNET_UNLOCK(ifp);
407 	if_detach(ifp);
408 
409 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
410 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
411 	mutex_destroy(&ifv->ifv_lock);
412 	free(ifv, M_DEVBUF);
413 
414 	return (0);
415 }
416 
417 /*
418  * Configure a VLAN interface.
419  */
420 static int
421 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
422 {
423 	struct ifnet *ifp = &ifv->ifv_if;
424 	struct ifvlan_linkmib *nmib = NULL;
425 	struct ifvlan_linkmib *omib = NULL;
426 	struct ifvlan_linkmib *checkmib = NULL;
427 	struct psref_target *nmib_psref = NULL;
428 	uint16_t vid = EVL_VLANOFTAG(tag);
429 	int error = 0;
430 	int idx;
431 	bool omib_cleanup = false;
432 	struct psref psref;
433 
434 	/* VLAN ID 0 and 4095 are reserved in the spec */
435 	if ((vid == 0) || (vid == 0xfff))
436 		return EINVAL;
437 
438 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
439 
440 	mutex_enter(&ifv->ifv_lock);
441 	omib = ifv->ifv_mib;
442 
443 	if (omib->ifvm_p != NULL) {
444 		error = EBUSY;
445 		goto done;
446 	}
447 
448 	/* Duplicate check */
449 	checkmib = vlan_lookup_tag_psref(p, vid, &psref);
450 	if (checkmib != NULL) {
451 		vlan_putref_linkmib(checkmib, &psref);
452 		error = EEXIST;
453 		goto done;
454 	}
455 
456 	*nmib = *omib;
457 	nmib_psref = &nmib->ifvm_psref;
458 
459 	psref_target_init(nmib_psref, ifvm_psref_class);
460 
461 	switch (p->if_type) {
462 	case IFT_ETHER:
463 	    {
464 		struct ethercom *ec = (void *) p;
465 		nmib->ifvm_msw = &vlan_ether_multisw;
466 		nmib->ifvm_encaplen = ETHER_VLAN_ENCAP_LEN;
467 		nmib->ifvm_mintu = ETHERMIN;
468 
469 		if (ec->ec_nvlans++ == 0) {
470 			IFNET_LOCK(p);
471 			error = ether_enable_vlan_mtu(p);
472 			IFNET_UNLOCK(p);
473 			if (error >= 0) {
474 				if (error) {
475 					ec->ec_nvlans--;
476 					goto done;
477 				}
478 				nmib->ifvm_mtufudge = 0;
479 			} else {
480 				/*
481 				 * Fudge the MTU by the encapsulation size. This
482 				 * makes us incompatible with strictly compliant
483 				 * 802.1Q implementations, but allows us to use
484 				 * the feature with other NetBSD
485 				 * implementations, which might still be useful.
486 				 */
487 				nmib->ifvm_mtufudge = nmib->ifvm_encaplen;
488 			}
489 			error = 0;
490 		}
491 
492 		/*
493 		 * If the parent interface can do hardware-assisted
494 		 * VLAN encapsulation, then propagate its hardware-
495 		 * assisted checksumming flags and tcp segmentation
496 		 * offload.
497 		 */
498 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
499 		        ec->ec_capenable |= ETHERCAP_VLAN_HWTAGGING;
500 			ifp->if_capabilities = p->if_capabilities &
501 			    (IFCAP_TSOv4 | IFCAP_TSOv6 |
502 			     IFCAP_CSUM_IPv4_Tx|IFCAP_CSUM_IPv4_Rx|
503 			     IFCAP_CSUM_TCPv4_Tx|IFCAP_CSUM_TCPv4_Rx|
504 			     IFCAP_CSUM_UDPv4_Tx|IFCAP_CSUM_UDPv4_Rx|
505 			     IFCAP_CSUM_TCPv6_Tx|IFCAP_CSUM_TCPv6_Rx|
506 			     IFCAP_CSUM_UDPv6_Tx|IFCAP_CSUM_UDPv6_Rx);
507                 }
508 		/*
509 		 * We inherit the parent's Ethernet address.
510 		 */
511 		ether_ifattach(ifp, CLLADDR(p->if_sadl));
512 		ifp->if_hdrlen = sizeof(struct ether_vlan_header); /* XXX? */
513 		break;
514 	    }
515 
516 	default:
517 		error = EPROTONOSUPPORT;
518 		goto done;
519 	}
520 
521 	nmib->ifvm_p = p;
522 	nmib->ifvm_tag = vid;
523 	ifv->ifv_if.if_mtu = p->if_mtu - nmib->ifvm_mtufudge;
524 	ifv->ifv_if.if_flags = p->if_flags &
525 	    (IFF_UP | IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
526 
527 	/*
528 	 * Inherit the if_type from the parent.  This allows us
529 	 * to participate in bridges of that type.
530 	 */
531 	ifv->ifv_if.if_type = p->if_type;
532 
533 	PSLIST_ENTRY_INIT(ifv, ifv_hash);
534 	idx = vlan_tag_hash(vid, ifv_hash.mask);
535 
536 	mutex_enter(&ifv_hash.lock);
537 	PSLIST_WRITER_INSERT_HEAD(&ifv_hash.lists[idx], ifv, ifv_hash);
538 	mutex_exit(&ifv_hash.lock);
539 
540 	vlan_linkmib_update(ifv, nmib);
541 	nmib = NULL;
542 	nmib_psref = NULL;
543 	omib_cleanup = true;
544 
545 done:
546 	mutex_exit(&ifv->ifv_lock);
547 
548 	if (nmib_psref)
549 		psref_target_destroy(nmib_psref, ifvm_psref_class);
550 
551 	if (nmib)
552 		kmem_free(nmib, sizeof(*nmib));
553 
554 	if (omib_cleanup)
555 		kmem_free(omib, sizeof(*omib));
556 
557 	return error;
558 }
559 
560 /*
561  * Unconfigure a VLAN interface.
562  */
563 static void
564 vlan_unconfig(struct ifnet *ifp)
565 {
566 	struct ifvlan *ifv = ifp->if_softc;
567 	struct ifvlan_linkmib *nmib = NULL;
568 	int error;
569 
570 	KASSERT(IFNET_LOCKED(ifp));
571 
572 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
573 
574 	mutex_enter(&ifv->ifv_lock);
575 	error = vlan_unconfig_locked(ifv, nmib);
576 	mutex_exit(&ifv->ifv_lock);
577 
578 	if (error)
579 		kmem_free(nmib, sizeof(*nmib));
580 }
581 static int
582 vlan_unconfig_locked(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
583 {
584 	struct ifnet *p;
585 	struct ifnet *ifp = &ifv->ifv_if;
586 	struct psref_target *nmib_psref = NULL;
587 	struct ifvlan_linkmib *omib;
588 	int error = 0;
589 
590 	KASSERT(IFNET_LOCKED(ifp));
591 	KASSERT(mutex_owned(&ifv->ifv_lock));
592 
593 	ifp->if_flags &= ~(IFF_UP|IFF_RUNNING);
594 
595 	omib = ifv->ifv_mib;
596 	p = omib->ifvm_p;
597 
598 	if (p == NULL) {
599 		error = -1;
600 		goto done;
601 	}
602 
603 	*nmib = *omib;
604 	nmib_psref = &nmib->ifvm_psref;
605 	psref_target_init(nmib_psref, ifvm_psref_class);
606 
607 	/*
608  	 * Since the interface is being unconfigured, we need to empty the
609 	 * list of multicast groups that we may have joined while we were
610 	 * alive and remove them from the parent's list also.
611 	 */
612 	(*nmib->ifvm_msw->vmsw_purgemulti)(ifv);
613 
614 	/* Disconnect from parent. */
615 	switch (p->if_type) {
616 	case IFT_ETHER:
617 	    {
618 		struct ethercom *ec = (void *)p;
619 		if (--ec->ec_nvlans == 0) {
620 			IFNET_LOCK(p);
621 			(void) ether_disable_vlan_mtu(p);
622 			IFNET_UNLOCK(p);
623 		}
624 
625 		ether_ifdetach(ifp);
626 		/* Restore vlan_ioctl overwritten by ether_ifdetach */
627 		ifp->if_ioctl = vlan_ioctl;
628 		vlan_reset_linkname(ifp);
629 		break;
630 	    }
631 
632 #ifdef DIAGNOSTIC
633 	default:
634 		panic("vlan_unconfig: impossible");
635 #endif
636 	}
637 
638 	nmib->ifvm_p = NULL;
639 	ifv->ifv_if.if_mtu = 0;
640 	ifv->ifv_flags = 0;
641 
642 	mutex_enter(&ifv_hash.lock);
643 	PSLIST_WRITER_REMOVE(ifv, ifv_hash);
644 	pserialize_perform(vlan_psz);
645 	mutex_exit(&ifv_hash.lock);
646 	PSLIST_ENTRY_DESTROY(ifv, ifv_hash);
647 
648 	vlan_linkmib_update(ifv, nmib);
649 
650 	mutex_exit(&ifv->ifv_lock);
651 
652 	nmib_psref = NULL;
653 	kmem_free(omib, sizeof(*omib));
654 
655 #ifdef INET6
656 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
657 	/* To delete v6 link local addresses */
658 	if (in6_present)
659 		in6_ifdetach(ifp);
660 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
661 #endif
662 
663 	if ((ifp->if_flags & IFF_PROMISC) != 0)
664 		vlan_safe_ifpromisc_locked(ifp, 0);
665 	if_down_locked(ifp);
666 	ifp->if_capabilities = 0;
667 	mutex_enter(&ifv->ifv_lock);
668 done:
669 
670 	if (nmib_psref)
671 		psref_target_destroy(nmib_psref, ifvm_psref_class);
672 
673 	return error;
674 }
675 
676 static void
677 vlan_hash_init(void)
678 {
679 
680 	ifv_hash.lists = hashinit(VLAN_TAG_HASH_SIZE, HASH_PSLIST, true,
681 	    &ifv_hash.mask);
682 }
683 
684 static int
685 vlan_hash_fini(void)
686 {
687 	int i;
688 
689 	mutex_enter(&ifv_hash.lock);
690 
691 	for (i = 0; i < ifv_hash.mask + 1; i++) {
692 		if (PSLIST_WRITER_FIRST(&ifv_hash.lists[i], struct ifvlan,
693 		    ifv_hash) != NULL) {
694 			mutex_exit(&ifv_hash.lock);
695 			return EBUSY;
696 		}
697 	}
698 
699 	for (i = 0; i < ifv_hash.mask + 1; i++)
700 		PSLIST_DESTROY(&ifv_hash.lists[i]);
701 
702 	mutex_exit(&ifv_hash.lock);
703 
704 	hashdone(ifv_hash.lists, HASH_PSLIST, ifv_hash.mask);
705 
706 	ifv_hash.lists = NULL;
707 	ifv_hash.mask = 0;
708 
709 	return 0;
710 }
711 
712 static int
713 vlan_tag_hash(uint16_t tag, u_long mask)
714 {
715 	uint32_t hash;
716 
717 	hash = (tag >> 8) ^ tag;
718 	hash = (hash >> 2) ^ hash;
719 
720 	return hash & mask;
721 }
722 
723 static struct ifvlan_linkmib *
724 vlan_getref_linkmib(struct ifvlan *sc, struct psref *psref)
725 {
726 	struct ifvlan_linkmib *mib;
727 	int s;
728 
729 	s = pserialize_read_enter();
730 	mib = sc->ifv_mib;
731 	if (mib == NULL) {
732 		pserialize_read_exit(s);
733 		return NULL;
734 	}
735 	membar_datadep_consumer();
736 	psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
737 	pserialize_read_exit(s);
738 
739 	return mib;
740 }
741 
742 static void
743 vlan_putref_linkmib(struct ifvlan_linkmib *mib, struct psref *psref)
744 {
745 	if (mib == NULL)
746 		return;
747 	psref_release(psref, &mib->ifvm_psref, ifvm_psref_class);
748 }
749 
750 static struct ifvlan_linkmib *
751 vlan_lookup_tag_psref(struct ifnet *ifp, uint16_t tag, struct psref *psref)
752 {
753 	int idx;
754 	int s;
755 	struct ifvlan *sc;
756 
757 	idx = vlan_tag_hash(tag, ifv_hash.mask);
758 
759 	s = pserialize_read_enter();
760 	PSLIST_READER_FOREACH(sc, &ifv_hash.lists[idx], struct ifvlan,
761 	    ifv_hash) {
762 		struct ifvlan_linkmib *mib = sc->ifv_mib;
763 		if (mib == NULL)
764 			continue;
765 		if (mib->ifvm_tag != tag)
766 			continue;
767 		if (mib->ifvm_p != ifp)
768 			continue;
769 
770 		psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
771 		pserialize_read_exit(s);
772 		return mib;
773 	}
774 	pserialize_read_exit(s);
775 	return NULL;
776 }
777 
778 static void
779 vlan_linkmib_update(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
780 {
781 	struct ifvlan_linkmib *omib = ifv->ifv_mib;
782 
783 	KASSERT(mutex_owned(&ifv->ifv_lock));
784 
785 	membar_producer();
786 	ifv->ifv_mib = nmib;
787 
788 	pserialize_perform(vlan_psz);
789 	psref_target_destroy(&omib->ifvm_psref, ifvm_psref_class);
790 }
791 
792 /*
793  * Called when a parent interface is detaching; destroy any VLAN
794  * configuration for the parent interface.
795  */
796 void
797 vlan_ifdetach(struct ifnet *p)
798 {
799 	struct ifvlan *ifv;
800 	struct ifvlan_linkmib *mib, **nmibs;
801 	struct psref psref;
802 	int error;
803 	int bound;
804 	int i, cnt = 0;
805 
806 	bound = curlwp_bind();
807 
808 	mutex_enter(&ifv_list.lock);
809 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
810 		mib = vlan_getref_linkmib(ifv, &psref);
811 		if (mib == NULL)
812 			continue;
813 
814 		if (mib->ifvm_p == p)
815 			cnt++;
816 
817 		vlan_putref_linkmib(mib, &psref);
818 	}
819 	mutex_exit(&ifv_list.lock);
820 
821 	if (cnt == 0) {
822 		curlwp_bindx(bound);
823 		return;
824 	}
825 
826 	/*
827 	 * The value of "cnt" does not increase while ifv_list.lock
828 	 * and ifv->ifv_lock are released here, because the parent
829 	 * interface is detaching.
830 	 */
831 	nmibs = kmem_alloc(sizeof(*nmibs) * cnt, KM_SLEEP);
832 	for (i = 0; i < cnt; i++) {
833 		nmibs[i] = kmem_alloc(sizeof(*nmibs[i]), KM_SLEEP);
834 	}
835 
836 	mutex_enter(&ifv_list.lock);
837 
838 	i = 0;
839 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
840 		struct ifnet *ifp = &ifv->ifv_if;
841 
842 		/* Need IFNET_LOCK that must be held before ifv_lock. */
843 		IFNET_LOCK(ifp);
844 		mutex_enter(&ifv->ifv_lock);
845 		if (ifv->ifv_mib->ifvm_p == p) {
846 			KASSERTMSG(i < cnt, "no memory for unconfig, parent=%s",
847 			    p->if_xname);
848 			error = vlan_unconfig_locked(ifv, nmibs[i]);
849 			if (!error) {
850 				nmibs[i] = NULL;
851 				i++;
852 			}
853 
854 		}
855 		mutex_exit(&ifv->ifv_lock);
856 		IFNET_UNLOCK(ifp);
857 	}
858 
859 	mutex_exit(&ifv_list.lock);
860 
861 	curlwp_bindx(bound);
862 
863 	for (i = 0; i < cnt; i++) {
864 		if (nmibs[i])
865 			kmem_free(nmibs[i], sizeof(*nmibs[i]));
866 	}
867 
868 	kmem_free(nmibs, sizeof(*nmibs) * cnt);
869 
870 	return;
871 }
872 
873 static int
874 vlan_set_promisc(struct ifnet *ifp)
875 {
876 	struct ifvlan *ifv = ifp->if_softc;
877 	struct ifvlan_linkmib *mib;
878 	struct psref psref;
879 	int error = 0;
880 	int bound;
881 
882 	bound = curlwp_bind();
883 	mib = vlan_getref_linkmib(ifv, &psref);
884 	if (mib == NULL) {
885 		curlwp_bindx(bound);
886 		return EBUSY;
887 	}
888 
889 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
890 		if ((ifv->ifv_flags & IFVF_PROMISC) == 0) {
891 			error = vlan_safe_ifpromisc(mib->ifvm_p, 1);
892 			if (error == 0)
893 				ifv->ifv_flags |= IFVF_PROMISC;
894 		}
895 	} else {
896 		if ((ifv->ifv_flags & IFVF_PROMISC) != 0) {
897 			error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
898 			if (error == 0)
899 				ifv->ifv_flags &= ~IFVF_PROMISC;
900 		}
901 	}
902 	vlan_putref_linkmib(mib, &psref);
903 	curlwp_bindx(bound);
904 
905 	return (error);
906 }
907 
908 static int
909 vlan_ioctl(struct ifnet *ifp, u_long cmd, void *data)
910 {
911 	struct lwp *l = curlwp;	/* XXX */
912 	struct ifvlan *ifv = ifp->if_softc;
913 	struct ifaddr *ifa = (struct ifaddr *) data;
914 	struct ifreq *ifr = (struct ifreq *) data;
915 	struct ifnet *pr;
916 	struct ifcapreq *ifcr;
917 	struct vlanreq vlr;
918 	struct ifvlan_linkmib *mib;
919 	struct psref psref;
920 	int error = 0;
921 	int bound;
922 
923 	switch (cmd) {
924 	case SIOCSIFMTU:
925 		bound = curlwp_bind();
926 		mib = vlan_getref_linkmib(ifv, &psref);
927 		if (mib == NULL) {
928 			curlwp_bindx(bound);
929 			error = EBUSY;
930 			break;
931 		}
932 
933 		if (mib->ifvm_p == NULL) {
934 			vlan_putref_linkmib(mib, &psref);
935 			curlwp_bindx(bound);
936 			error = EINVAL;
937 		} else if (
938 		    ifr->ifr_mtu > (mib->ifvm_p->if_mtu - mib->ifvm_mtufudge) ||
939 		    ifr->ifr_mtu < (mib->ifvm_mintu - mib->ifvm_mtufudge)) {
940 			vlan_putref_linkmib(mib, &psref);
941 			curlwp_bindx(bound);
942 			error = EINVAL;
943 		} else {
944 			vlan_putref_linkmib(mib, &psref);
945 			curlwp_bindx(bound);
946 
947 			error = ifioctl_common(ifp, cmd, data);
948 			if (error == ENETRESET)
949 					error = 0;
950 		}
951 
952 		break;
953 
954 	case SIOCSETVLAN:
955 		if ((error = kauth_authorize_network(l->l_cred,
956 		    KAUTH_NETWORK_INTERFACE,
957 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
958 		    NULL)) != 0)
959 			break;
960 		if ((error = copyin(ifr->ifr_data, &vlr, sizeof(vlr))) != 0)
961 			break;
962 
963 		if (vlr.vlr_parent[0] == '\0') {
964 			bound = curlwp_bind();
965 			mib = vlan_getref_linkmib(ifv, &psref);
966 			if (mib == NULL) {
967 				curlwp_bindx(bound);
968 				error = EBUSY;
969 				break;
970 			}
971 
972 			if (mib->ifvm_p != NULL &&
973 			    (ifp->if_flags & IFF_PROMISC) != 0)
974 				error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
975 
976 			vlan_putref_linkmib(mib, &psref);
977 			curlwp_bindx(bound);
978 
979 			vlan_unconfig(ifp);
980 			break;
981 		}
982 		if (vlr.vlr_tag != EVL_VLANOFTAG(vlr.vlr_tag)) {
983 			error = EINVAL;		 /* check for valid tag */
984 			break;
985 		}
986 		if ((pr = ifunit(vlr.vlr_parent)) == NULL) {
987 			error = ENOENT;
988 			break;
989 		}
990 		error = vlan_config(ifv, pr, vlr.vlr_tag);
991 		if (error != 0) {
992 			break;
993 		}
994 
995 		/* Update promiscuous mode, if necessary. */
996 		vlan_set_promisc(ifp);
997 
998 		ifp->if_flags |= IFF_RUNNING;
999 		break;
1000 
1001 	case SIOCGETVLAN:
1002 		memset(&vlr, 0, sizeof(vlr));
1003 		bound = curlwp_bind();
1004 		mib = vlan_getref_linkmib(ifv, &psref);
1005 		if (mib == NULL) {
1006 			curlwp_bindx(bound);
1007 			error = EBUSY;
1008 			break;
1009 		}
1010 		if (mib->ifvm_p != NULL) {
1011 			snprintf(vlr.vlr_parent, sizeof(vlr.vlr_parent), "%s",
1012 			    mib->ifvm_p->if_xname);
1013 			vlr.vlr_tag = mib->ifvm_tag;
1014 		}
1015 		vlan_putref_linkmib(mib, &psref);
1016 		curlwp_bindx(bound);
1017 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1018 		break;
1019 
1020 	case SIOCSIFFLAGS:
1021 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
1022 			break;
1023 		/*
1024 		 * For promiscuous mode, we enable promiscuous mode on
1025 		 * the parent if we need promiscuous on the VLAN interface.
1026 		 */
1027 		bound = curlwp_bind();
1028 		mib = vlan_getref_linkmib(ifv, &psref);
1029 		if (mib == NULL) {
1030 			curlwp_bindx(bound);
1031 			error = EBUSY;
1032 			break;
1033 		}
1034 
1035 		if (mib->ifvm_p != NULL)
1036 			error = vlan_set_promisc(ifp);
1037 		vlan_putref_linkmib(mib, &psref);
1038 		curlwp_bindx(bound);
1039 		break;
1040 
1041 	case SIOCADDMULTI:
1042 		mutex_enter(&ifv->ifv_lock);
1043 		mib = ifv->ifv_mib;
1044 		if (mib == NULL) {
1045 			error = EBUSY;
1046 			mutex_exit(&ifv->ifv_lock);
1047 			break;
1048 		}
1049 
1050 		error = (mib->ifvm_p != NULL) ?
1051 		    (*mib->ifvm_msw->vmsw_addmulti)(ifv, ifr) : EINVAL;
1052 		mib = NULL;
1053 		mutex_exit(&ifv->ifv_lock);
1054 		break;
1055 
1056 	case SIOCDELMULTI:
1057 		mutex_enter(&ifv->ifv_lock);
1058 		mib = ifv->ifv_mib;
1059 		if (mib == NULL) {
1060 			error = EBUSY;
1061 			mutex_exit(&ifv->ifv_lock);
1062 			break;
1063 		}
1064 		error = (mib->ifvm_p != NULL) ?
1065 		    (*mib->ifvm_msw->vmsw_delmulti)(ifv, ifr) : EINVAL;
1066 		mib = NULL;
1067 		mutex_exit(&ifv->ifv_lock);
1068 		break;
1069 
1070 	case SIOCSIFCAP:
1071 		ifcr = data;
1072 		/* make sure caps are enabled on parent */
1073 		bound = curlwp_bind();
1074 		mib = vlan_getref_linkmib(ifv, &psref);
1075 		if (mib == NULL) {
1076 			curlwp_bindx(bound);
1077 			error = EBUSY;
1078 			break;
1079 		}
1080 
1081 		if (mib->ifvm_p == NULL) {
1082 			vlan_putref_linkmib(mib, &psref);
1083 			curlwp_bindx(bound);
1084 			error = EINVAL;
1085 			break;
1086 		}
1087 		if ((mib->ifvm_p->if_capenable & ifcr->ifcr_capenable) !=
1088 		    ifcr->ifcr_capenable) {
1089 			vlan_putref_linkmib(mib, &psref);
1090 			curlwp_bindx(bound);
1091 			error = EINVAL;
1092 			break;
1093 		}
1094 
1095 		vlan_putref_linkmib(mib, &psref);
1096 		curlwp_bindx(bound);
1097 
1098 		if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
1099 			error = 0;
1100 		break;
1101 	case SIOCINITIFADDR:
1102 		bound = curlwp_bind();
1103 		mib = vlan_getref_linkmib(ifv, &psref);
1104 		if (mib == NULL) {
1105 			curlwp_bindx(bound);
1106 			error = EBUSY;
1107 			break;
1108 		}
1109 
1110 		if (mib->ifvm_p == NULL) {
1111 			error = EINVAL;
1112 			vlan_putref_linkmib(mib, &psref);
1113 			curlwp_bindx(bound);
1114 			break;
1115 		}
1116 		vlan_putref_linkmib(mib, &psref);
1117 		curlwp_bindx(bound);
1118 
1119 		ifp->if_flags |= IFF_UP;
1120 #ifdef INET
1121 		if (ifa->ifa_addr->sa_family == AF_INET)
1122 			arp_ifinit(ifp, ifa);
1123 #endif
1124 		break;
1125 
1126 	default:
1127 		error = ether_ioctl(ifp, cmd, data);
1128 	}
1129 
1130 	return (error);
1131 }
1132 
1133 static int
1134 vlan_ether_addmulti(struct ifvlan *ifv, struct ifreq *ifr)
1135 {
1136 	const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr);
1137 	struct vlan_mc_entry *mc;
1138 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1139 	struct ifvlan_linkmib *mib;
1140 	int error;
1141 
1142 	KASSERT(mutex_owned(&ifv->ifv_lock));
1143 
1144 	if (sa->sa_len > sizeof(struct sockaddr_storage))
1145 		return (EINVAL);
1146 
1147 	error = ether_addmulti(sa, &ifv->ifv_ec);
1148 	if (error != ENETRESET)
1149 		return (error);
1150 
1151 	/*
1152 	 * This is new multicast address.  We have to tell parent
1153 	 * about it.  Also, remember this multicast address so that
1154 	 * we can delete them on unconfigure.
1155 	 */
1156 	mc = malloc(sizeof(struct vlan_mc_entry), M_DEVBUF, M_NOWAIT);
1157 	if (mc == NULL) {
1158 		error = ENOMEM;
1159 		goto alloc_failed;
1160 	}
1161 
1162 	/*
1163 	 * As ether_addmulti() returns ENETRESET, following two
1164 	 * statement shouldn't fail.
1165 	 */
1166 	(void)ether_multiaddr(sa, addrlo, addrhi);
1167 	ETHER_LOOKUP_MULTI(addrlo, addrhi, &ifv->ifv_ec, mc->mc_enm);
1168 	memcpy(&mc->mc_addr, sa, sa->sa_len);
1169 	LIST_INSERT_HEAD(&ifv->ifv_mc_listhead, mc, mc_entries);
1170 
1171 	mib = ifv->ifv_mib;
1172 
1173 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1174 	IFNET_LOCK(mib->ifvm_p);
1175 	error = if_mcast_op(mib->ifvm_p, SIOCADDMULTI, sa);
1176 	IFNET_UNLOCK(mib->ifvm_p);
1177 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1178 
1179 	if (error != 0)
1180 		goto ioctl_failed;
1181 	return (error);
1182 
1183  ioctl_failed:
1184 	LIST_REMOVE(mc, mc_entries);
1185 	free(mc, M_DEVBUF);
1186  alloc_failed:
1187 	(void)ether_delmulti(sa, &ifv->ifv_ec);
1188 	return (error);
1189 }
1190 
1191 static int
1192 vlan_ether_delmulti(struct ifvlan *ifv, struct ifreq *ifr)
1193 {
1194 	const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr);
1195 	struct ether_multi *enm;
1196 	struct vlan_mc_entry *mc;
1197 	struct ifvlan_linkmib *mib;
1198 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1199 	int error;
1200 
1201 	KASSERT(mutex_owned(&ifv->ifv_lock));
1202 
1203 	/*
1204 	 * Find a key to lookup vlan_mc_entry.  We have to do this
1205 	 * before calling ether_delmulti for obvious reason.
1206 	 */
1207 	if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0)
1208 		return (error);
1209 	ETHER_LOOKUP_MULTI(addrlo, addrhi, &ifv->ifv_ec, enm);
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 	IFNET_LOCK(mib->ifvm_p);
1218 	error = if_mcast_op(mib->ifvm_p, SIOCDELMULTI, sa);
1219 	IFNET_UNLOCK(mib->ifvm_p);
1220 
1221 	if (error == 0) {
1222 		/* And forget about this address. */
1223 		for (mc = LIST_FIRST(&ifv->ifv_mc_listhead); mc != NULL;
1224 		    mc = LIST_NEXT(mc, mc_entries)) {
1225 			if (mc->mc_enm == enm) {
1226 				LIST_REMOVE(mc, mc_entries);
1227 				free(mc, M_DEVBUF);
1228 				break;
1229 			}
1230 		}
1231 		KASSERT(mc != NULL);
1232 	} else
1233 		(void)ether_addmulti(sa, &ifv->ifv_ec);
1234 	return (error);
1235 }
1236 
1237 /*
1238  * Delete any multicast address we have asked to add from parent
1239  * interface.  Called when the vlan is being unconfigured.
1240  */
1241 static void
1242 vlan_ether_purgemulti(struct ifvlan *ifv)
1243 {
1244 	struct vlan_mc_entry *mc;
1245 	struct ifvlan_linkmib *mib;
1246 
1247 	KASSERT(mutex_owned(&ifv->ifv_lock));
1248 	mib = ifv->ifv_mib;
1249 	if (mib == NULL) {
1250 		return;
1251 	}
1252 
1253 	while ((mc = LIST_FIRST(&ifv->ifv_mc_listhead)) != NULL) {
1254 		IFNET_LOCK(mib->ifvm_p);
1255 		(void)if_mcast_op(mib->ifvm_p, SIOCDELMULTI,
1256 		    (const struct sockaddr *)&mc->mc_addr);
1257 		IFNET_UNLOCK(mib->ifvm_p);
1258 		LIST_REMOVE(mc, mc_entries);
1259 		free(mc, M_DEVBUF);
1260 	}
1261 }
1262 
1263 static void
1264 vlan_start(struct ifnet *ifp)
1265 {
1266 	struct ifvlan *ifv = ifp->if_softc;
1267 	struct ifnet *p;
1268 	struct ethercom *ec;
1269 	struct mbuf *m;
1270 	struct ifvlan_linkmib *mib;
1271 	struct psref psref;
1272 	int error;
1273 
1274 	mib = vlan_getref_linkmib(ifv, &psref);
1275 	if (mib == NULL)
1276 		return;
1277 	p = mib->ifvm_p;
1278 	ec = (void *)mib->ifvm_p;
1279 
1280 	ifp->if_flags |= IFF_OACTIVE;
1281 
1282 	for (;;) {
1283 		IFQ_DEQUEUE(&ifp->if_snd, m);
1284 		if (m == NULL)
1285 			break;
1286 
1287 #ifdef ALTQ
1288 		/*
1289 		 * KERNEL_LOCK is required for ALTQ even if NET_MPSAFE is defined.
1290 		 */
1291 		KERNEL_LOCK(1, NULL);
1292 		/*
1293 		 * If ALTQ is enabled on the parent interface, do
1294 		 * classification; the queueing discipline might
1295 		 * not require classification, but might require
1296 		 * the address family/header pointer in the pktattr.
1297 		 */
1298 		if (ALTQ_IS_ENABLED(&p->if_snd)) {
1299 			switch (p->if_type) {
1300 			case IFT_ETHER:
1301 				altq_etherclassify(&p->if_snd, m);
1302 				break;
1303 #ifdef DIAGNOSTIC
1304 			default:
1305 				panic("vlan_start: impossible (altq)");
1306 #endif
1307 			}
1308 		}
1309 		KERNEL_UNLOCK_ONE(NULL);
1310 #endif /* ALTQ */
1311 
1312 		bpf_mtap(ifp, m);
1313 		/*
1314 		 * If the parent can insert the tag itself, just mark
1315 		 * the tag in the mbuf header.
1316 		 */
1317 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
1318 			vlan_set_tag(m, mib->ifvm_tag);
1319 		} else {
1320 			/*
1321 			 * insert the tag ourselves
1322 			 */
1323 			M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1324 			if (m == NULL) {
1325 				printf("%s: unable to prepend encap header",
1326 				    p->if_xname);
1327 				ifp->if_oerrors++;
1328 				continue;
1329 			}
1330 
1331 			switch (p->if_type) {
1332 			case IFT_ETHER:
1333 			    {
1334 				struct ether_vlan_header *evl;
1335 
1336 				if (m->m_len < sizeof(struct ether_vlan_header))
1337 					m = m_pullup(m,
1338 					    sizeof(struct ether_vlan_header));
1339 				if (m == NULL) {
1340 					printf("%s: unable to pullup encap "
1341 					    "header", p->if_xname);
1342 					ifp->if_oerrors++;
1343 					continue;
1344 				}
1345 
1346 				/*
1347 				 * Transform the Ethernet header into an
1348 				 * Ethernet header with 802.1Q encapsulation.
1349 				 */
1350 				memmove(mtod(m, void *),
1351 				    mtod(m, char *) + mib->ifvm_encaplen,
1352 				    sizeof(struct ether_header));
1353 				evl = mtod(m, struct ether_vlan_header *);
1354 				evl->evl_proto = evl->evl_encap_proto;
1355 				evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1356 				evl->evl_tag = htons(mib->ifvm_tag);
1357 
1358 				/*
1359 				 * To cater for VLAN-aware layer 2 ethernet
1360 				 * switches which may need to strip the tag
1361 				 * before forwarding the packet, make sure
1362 				 * the packet+tag is at least 68 bytes long.
1363 				 * This is necessary because our parent will
1364 				 * only pad to 64 bytes (ETHER_MIN_LEN) and
1365 				 * some switches will not pad by themselves
1366 				 * after deleting a tag.
1367 				 */
1368 				if (m->m_pkthdr.len <
1369 				    (ETHER_MIN_LEN - ETHER_CRC_LEN +
1370 				     ETHER_VLAN_ENCAP_LEN)) {
1371 					m_copyback(m, m->m_pkthdr.len,
1372 					    (ETHER_MIN_LEN - ETHER_CRC_LEN +
1373 					     ETHER_VLAN_ENCAP_LEN) -
1374 					     m->m_pkthdr.len,
1375 					    vlan_zero_pad_buff);
1376 				}
1377 				break;
1378 			    }
1379 
1380 #ifdef DIAGNOSTIC
1381 			default:
1382 				panic("vlan_start: impossible");
1383 #endif
1384 			}
1385 		}
1386 
1387 		if ((p->if_flags & IFF_RUNNING) == 0) {
1388 			m_freem(m);
1389 			continue;
1390 		}
1391 
1392 		error = if_transmit_lock(p, m);
1393 		if (error) {
1394 			/* mbuf is already freed */
1395 			ifp->if_oerrors++;
1396 			continue;
1397 		}
1398 		ifp->if_opackets++;
1399 	}
1400 
1401 	ifp->if_flags &= ~IFF_OACTIVE;
1402 
1403 	/* Remove reference to mib before release */
1404 	p = NULL;
1405 	ec = NULL;
1406 
1407 	vlan_putref_linkmib(mib, &psref);
1408 }
1409 
1410 static int
1411 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1412 {
1413 	struct ifvlan *ifv = ifp->if_softc;
1414 	struct ifnet *p;
1415 	struct ethercom *ec;
1416 	struct ifvlan_linkmib *mib;
1417 	struct psref psref;
1418 	int error;
1419 	size_t pktlen = m->m_pkthdr.len;
1420 	bool mcast = (m->m_flags & M_MCAST) != 0;
1421 
1422 	mib = vlan_getref_linkmib(ifv, &psref);
1423 	if (mib == NULL) {
1424 		m_freem(m);
1425 		return ENETDOWN;
1426 	}
1427 
1428 	p = mib->ifvm_p;
1429 	ec = (void *)mib->ifvm_p;
1430 
1431 	bpf_mtap(ifp, m);
1432 
1433 	if (pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_OUT) != 0) {
1434 		if (m != NULL)
1435 			m_freem(m);
1436 		error = 0;
1437 		goto out;
1438 	}
1439 
1440 	/*
1441 	 * If the parent can insert the tag itself, just mark
1442 	 * the tag in the mbuf header.
1443 	 */
1444 	if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
1445 		vlan_set_tag(m, mib->ifvm_tag);
1446 	} else {
1447 		/*
1448 		 * insert the tag ourselves
1449 		 */
1450 		M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1451 		if (m == NULL) {
1452 			printf("%s: unable to prepend encap header",
1453 			    p->if_xname);
1454 			ifp->if_oerrors++;
1455 			error = ENOBUFS;
1456 			goto out;
1457 		}
1458 
1459 		switch (p->if_type) {
1460 		case IFT_ETHER:
1461 		    {
1462 			struct ether_vlan_header *evl;
1463 
1464 			if (m->m_len < sizeof(struct ether_vlan_header))
1465 				m = m_pullup(m,
1466 				    sizeof(struct ether_vlan_header));
1467 			if (m == NULL) {
1468 				printf("%s: unable to pullup encap "
1469 				    "header", p->if_xname);
1470 				ifp->if_oerrors++;
1471 				error = ENOBUFS;
1472 				goto out;
1473 			}
1474 
1475 			/*
1476 			 * Transform the Ethernet header into an
1477 			 * Ethernet header with 802.1Q encapsulation.
1478 			 */
1479 			memmove(mtod(m, void *),
1480 			    mtod(m, char *) + mib->ifvm_encaplen,
1481 			    sizeof(struct ether_header));
1482 			evl = mtod(m, struct ether_vlan_header *);
1483 			evl->evl_proto = evl->evl_encap_proto;
1484 			evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1485 			evl->evl_tag = htons(mib->ifvm_tag);
1486 
1487 			/*
1488 			 * To cater for VLAN-aware layer 2 ethernet
1489 			 * switches which may need to strip the tag
1490 			 * before forwarding the packet, make sure
1491 			 * the packet+tag is at least 68 bytes long.
1492 			 * This is necessary because our parent will
1493 			 * only pad to 64 bytes (ETHER_MIN_LEN) and
1494 			 * some switches will not pad by themselves
1495 			 * after deleting a tag.
1496 			 */
1497 			if (m->m_pkthdr.len <
1498 			    (ETHER_MIN_LEN - ETHER_CRC_LEN +
1499 			     ETHER_VLAN_ENCAP_LEN)) {
1500 				m_copyback(m, m->m_pkthdr.len,
1501 				    (ETHER_MIN_LEN - ETHER_CRC_LEN +
1502 				     ETHER_VLAN_ENCAP_LEN) -
1503 				     m->m_pkthdr.len,
1504 				    vlan_zero_pad_buff);
1505 			}
1506 			break;
1507 		    }
1508 
1509 #ifdef DIAGNOSTIC
1510 		default:
1511 			panic("vlan_transmit: impossible");
1512 #endif
1513 		}
1514 	}
1515 
1516 	if ((p->if_flags & IFF_RUNNING) == 0) {
1517 		m_freem(m);
1518 		error = ENETDOWN;
1519 		goto out;
1520 	}
1521 
1522 	error = if_transmit_lock(p, m);
1523 	if (error) {
1524 		/* mbuf is already freed */
1525 		ifp->if_oerrors++;
1526 	} else {
1527 
1528 		ifp->if_opackets++;
1529 		ifp->if_obytes += pktlen;
1530 		if (mcast)
1531 			ifp->if_omcasts++;
1532 	}
1533 
1534 out:
1535 	/* Remove reference to mib before release */
1536 	p = NULL;
1537 	ec = NULL;
1538 
1539 	vlan_putref_linkmib(mib, &psref);
1540 	return error;
1541 }
1542 
1543 /*
1544  * Given an Ethernet frame, find a valid vlan interface corresponding to the
1545  * given source interface and tag, then run the real packet through the
1546  * parent's input routine.
1547  */
1548 void
1549 vlan_input(struct ifnet *ifp, struct mbuf *m)
1550 {
1551 	struct ifvlan *ifv;
1552 	uint16_t vid;
1553 	struct ifvlan_linkmib *mib;
1554 	struct psref psref;
1555 	bool have_vtag;
1556 
1557 	have_vtag = vlan_has_tag(m);
1558 	if (have_vtag) {
1559 		vid = EVL_VLANOFTAG(vlan_get_tag(m));
1560 		m->m_flags &= ~M_VLANTAG;
1561 	} else {
1562 		switch (ifp->if_type) {
1563 		case IFT_ETHER:
1564 		    {
1565 			struct ether_vlan_header *evl;
1566 
1567 			if (m->m_len < sizeof(struct ether_vlan_header) &&
1568 			    (m = m_pullup(m,
1569 			     sizeof(struct ether_vlan_header))) == NULL) {
1570 				printf("%s: no memory for VLAN header, "
1571 				    "dropping packet.\n", ifp->if_xname);
1572 				return;
1573 			}
1574 			evl = mtod(m, struct ether_vlan_header *);
1575 			KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN);
1576 
1577 			vid = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1578 
1579 			/*
1580 			 * Restore the original ethertype.  We'll remove
1581 			 * the encapsulation after we've found the vlan
1582 			 * interface corresponding to the tag.
1583 			 */
1584 			evl->evl_encap_proto = evl->evl_proto;
1585 			break;
1586 		    }
1587 
1588 		default:
1589 			vid = (uint16_t) -1;	/* XXX GCC */
1590 #ifdef DIAGNOSTIC
1591 			panic("vlan_input: impossible");
1592 #endif
1593 		}
1594 	}
1595 
1596 	mib = vlan_lookup_tag_psref(ifp, vid, &psref);
1597 	if (mib == NULL) {
1598 		m_freem(m);
1599 		ifp->if_noproto++;
1600 		return;
1601 	}
1602 
1603 	ifv = mib->ifvm_ifvlan;
1604 	if ((ifv->ifv_if.if_flags & (IFF_UP|IFF_RUNNING)) !=
1605 	    (IFF_UP|IFF_RUNNING)) {
1606 		m_freem(m);
1607 		ifp->if_noproto++;
1608 		goto out;
1609 	}
1610 
1611 	/*
1612 	 * Now, remove the encapsulation header.  The original
1613 	 * header has already been fixed up above.
1614 	 */
1615 	if (!have_vtag) {
1616 		memmove(mtod(m, char *) + mib->ifvm_encaplen,
1617 		    mtod(m, void *), sizeof(struct ether_header));
1618 		m_adj(m, mib->ifvm_encaplen);
1619 	}
1620 
1621 	m_set_rcvif(m, &ifv->ifv_if);
1622 	ifv->ifv_if.if_ipackets++;
1623 
1624 	if (pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_IN) != 0) {
1625 		if (m != NULL)
1626 			m_freem(m);
1627 		goto out;
1628 	}
1629 
1630 	m->m_flags &= ~M_PROMISC;
1631 	if_input(&ifv->ifv_if, m);
1632 out:
1633 	vlan_putref_linkmib(mib, &psref);
1634 }
1635 
1636 /*
1637  * Module infrastructure
1638  */
1639 #include "if_module.h"
1640 
1641 IF_MODULE(MODULE_CLASS_DRIVER, vlan, "")
1642