xref: /netbsd-src/sys/net/if_vlan.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: if_vlan.c,v 1.130 2018/06/26 06:48:02 msaitoh 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 1.130 2018/06/26 06:48:02 msaitoh 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 	.vmsw_addmulti = vlan_ether_addmulti,
195 	.vmsw_delmulti = vlan_ether_delmulti,
196 	.vmsw_purgemulti = 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 
254 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
255 	e = ifpromisc(ifp, pswitch);
256 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
257 
258 	return e;
259 }
260 
261 static inline int
262 vlan_safe_ifpromisc_locked(struct ifnet *ifp, int pswitch)
263 {
264 	int e;
265 
266 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
267 	e = ifpromisc_locked(ifp, pswitch);
268 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
269 
270 	return e;
271 }
272 
273 void
274 vlanattach(int n)
275 {
276 
277 	/*
278 	 * Nothing to do here, initialization is handled by the
279 	 * module initialization code in vlaninit() below.
280 	 */
281 }
282 
283 static void
284 vlaninit(void)
285 {
286 	mutex_init(&ifv_list.lock, MUTEX_DEFAULT, IPL_NONE);
287 	LIST_INIT(&ifv_list.list);
288 
289 	mutex_init(&ifv_hash.lock, MUTEX_DEFAULT, IPL_NONE);
290 	vlan_psz = pserialize_create();
291 	ifvm_psref_class = psref_class_create("vlanlinkmib", IPL_SOFTNET);
292 	if_clone_attach(&vlan_cloner);
293 
294 	vlan_hash_init();
295 }
296 
297 static int
298 vlandetach(void)
299 {
300 	bool is_empty;
301 	int error;
302 
303 	mutex_enter(&ifv_list.lock);
304 	is_empty = LIST_EMPTY(&ifv_list.list);
305 	mutex_exit(&ifv_list.lock);
306 
307 	if (!is_empty)
308 		return EBUSY;
309 
310 	error = vlan_hash_fini();
311 	if (error != 0)
312 		return error;
313 
314 	if_clone_detach(&vlan_cloner);
315 	psref_class_destroy(ifvm_psref_class);
316 	pserialize_destroy(vlan_psz);
317 	mutex_destroy(&ifv_hash.lock);
318 	mutex_destroy(&ifv_list.lock);
319 
320 	return 0;
321 }
322 
323 static void
324 vlan_reset_linkname(struct ifnet *ifp)
325 {
326 
327 	/*
328 	 * We start out with a "802.1Q VLAN" type and zero-length
329 	 * addresses.  When we attach to a parent interface, we
330 	 * inherit its type, address length, address, and data link
331 	 * type.
332 	 */
333 
334 	ifp->if_type = IFT_L2VLAN;
335 	ifp->if_addrlen = 0;
336 	ifp->if_dlt = DLT_NULL;
337 	if_alloc_sadl(ifp);
338 }
339 
340 static int
341 vlan_clone_create(struct if_clone *ifc, int unit)
342 {
343 	struct ifvlan *ifv;
344 	struct ifnet *ifp;
345 	struct ifvlan_linkmib *mib;
346 	int rv;
347 
348 	ifv = malloc(sizeof(struct ifvlan), M_DEVBUF, M_WAITOK|M_ZERO);
349 	mib = kmem_zalloc(sizeof(struct ifvlan_linkmib), KM_SLEEP);
350 	ifp = &ifv->ifv_if;
351 	LIST_INIT(&ifv->ifv_mc_listhead);
352 
353 	mib->ifvm_ifvlan = ifv;
354 	mib->ifvm_p = NULL;
355 	psref_target_init(&mib->ifvm_psref, ifvm_psref_class);
356 
357 	mutex_init(&ifv->ifv_lock, MUTEX_DEFAULT, IPL_NONE);
358 	ifv->ifv_mib = mib;
359 
360 	mutex_enter(&ifv_list.lock);
361 	LIST_INSERT_HEAD(&ifv_list.list, ifv, ifv_list);
362 	mutex_exit(&ifv_list.lock);
363 
364 	if_initname(ifp, ifc->ifc_name, unit);
365 	ifp->if_softc = ifv;
366 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
367 	ifp->if_extflags = IFEF_NO_LINK_STATE_CHANGE;
368 #ifdef NET_MPSAFE
369 	ifp->if_extflags |= IFEF_MPSAFE;
370 #endif
371 	ifp->if_start = vlan_start;
372 	ifp->if_transmit = vlan_transmit;
373 	ifp->if_ioctl = vlan_ioctl;
374 	IFQ_SET_READY(&ifp->if_snd);
375 
376 	rv = if_initialize(ifp);
377 	if (rv != 0) {
378 		aprint_error("%s: if_initialize failed(%d)\n", ifp->if_xname,
379 		    rv);
380 		goto fail;
381 	}
382 
383 	vlan_reset_linkname(ifp);
384 	if_register(ifp);
385 	return 0;
386 
387 fail:
388 	mutex_enter(&ifv_list.lock);
389 	LIST_REMOVE(ifv, ifv_list);
390 	mutex_exit(&ifv_list.lock);
391 
392 	mutex_destroy(&ifv->ifv_lock);
393 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
394 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
395 	free(ifv, M_DEVBUF);
396 
397 	return rv;
398 }
399 
400 static int
401 vlan_clone_destroy(struct ifnet *ifp)
402 {
403 	struct ifvlan *ifv = ifp->if_softc;
404 
405 	mutex_enter(&ifv_list.lock);
406 	LIST_REMOVE(ifv, ifv_list);
407 	mutex_exit(&ifv_list.lock);
408 
409 	IFNET_LOCK(ifp);
410 	vlan_unconfig(ifp);
411 	IFNET_UNLOCK(ifp);
412 	if_detach(ifp);
413 
414 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
415 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
416 	mutex_destroy(&ifv->ifv_lock);
417 	free(ifv, M_DEVBUF);
418 
419 	return 0;
420 }
421 
422 /*
423  * Configure a VLAN interface.
424  */
425 static int
426 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
427 {
428 	struct ifnet *ifp = &ifv->ifv_if;
429 	struct ifvlan_linkmib *nmib = NULL;
430 	struct ifvlan_linkmib *omib = NULL;
431 	struct ifvlan_linkmib *checkmib;
432 	struct psref_target *nmib_psref = NULL;
433 	const uint16_t vid = EVL_VLANOFTAG(tag);
434 	int error = 0;
435 	int idx;
436 	bool omib_cleanup = false;
437 	struct psref psref;
438 
439 	/* VLAN ID 0 and 4095 are reserved in the spec */
440 	if ((vid == 0) || (vid == 0xfff))
441 		return EINVAL;
442 
443 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
444 	mutex_enter(&ifv->ifv_lock);
445 	omib = ifv->ifv_mib;
446 
447 	if (omib->ifvm_p != NULL) {
448 		error = EBUSY;
449 		goto done;
450 	}
451 
452 	/* Duplicate check */
453 	checkmib = vlan_lookup_tag_psref(p, vid, &psref);
454 	if (checkmib != NULL) {
455 		vlan_putref_linkmib(checkmib, &psref);
456 		error = EEXIST;
457 		goto done;
458 	}
459 
460 	*nmib = *omib;
461 	nmib_psref = &nmib->ifvm_psref;
462 
463 	psref_target_init(nmib_psref, ifvm_psref_class);
464 
465 	switch (p->if_type) {
466 	case IFT_ETHER:
467 	    {
468 		struct ethercom *ec = (void *)p;
469 		nmib->ifvm_msw = &vlan_ether_multisw;
470 		nmib->ifvm_encaplen = ETHER_VLAN_ENCAP_LEN;
471 		nmib->ifvm_mintu = ETHERMIN;
472 
473 		if (ec->ec_nvlans++ == 0) {
474 			IFNET_LOCK(p);
475 			error = ether_enable_vlan_mtu(p);
476 			IFNET_UNLOCK(p);
477 			if (error >= 0) {
478 				if (error) {
479 					ec->ec_nvlans--;
480 					goto done;
481 				}
482 				nmib->ifvm_mtufudge = 0;
483 			} else {
484 				/*
485 				 * Fudge the MTU by the encapsulation size. This
486 				 * makes us incompatible with strictly compliant
487 				 * 802.1Q implementations, but allows us to use
488 				 * the feature with other NetBSD
489 				 * implementations, which might still be useful.
490 				 */
491 				nmib->ifvm_mtufudge = nmib->ifvm_encaplen;
492 			}
493 			error = 0;
494 		}
495 
496 		/*
497 		 * If the parent interface can do hardware-assisted
498 		 * VLAN encapsulation, then propagate its hardware-
499 		 * assisted checksumming flags and tcp segmentation
500 		 * offload.
501 		 */
502 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
503 			ec->ec_capenable |= ETHERCAP_VLAN_HWTAGGING;
504 			ifp->if_capabilities = p->if_capabilities &
505 			    (IFCAP_TSOv4 | IFCAP_TSOv6 |
506 			     IFCAP_CSUM_IPv4_Tx|IFCAP_CSUM_IPv4_Rx|
507 			     IFCAP_CSUM_TCPv4_Tx|IFCAP_CSUM_TCPv4_Rx|
508 			     IFCAP_CSUM_UDPv4_Tx|IFCAP_CSUM_UDPv4_Rx|
509 			     IFCAP_CSUM_TCPv6_Tx|IFCAP_CSUM_TCPv6_Rx|
510 			     IFCAP_CSUM_UDPv6_Tx|IFCAP_CSUM_UDPv6_Rx);
511 		}
512 
513 		/*
514 		 * We inherit the parent's Ethernet address.
515 		 */
516 		ether_ifattach(ifp, CLLADDR(p->if_sadl));
517 		ifp->if_hdrlen = sizeof(struct ether_vlan_header); /* XXX? */
518 		break;
519 	    }
520 
521 	default:
522 		error = EPROTONOSUPPORT;
523 		goto done;
524 	}
525 
526 	nmib->ifvm_p = p;
527 	nmib->ifvm_tag = vid;
528 	ifv->ifv_if.if_mtu = p->if_mtu - nmib->ifvm_mtufudge;
529 	ifv->ifv_if.if_flags = p->if_flags &
530 	    (IFF_UP | IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
531 
532 	/*
533 	 * Inherit the if_type from the parent.  This allows us
534 	 * to participate in bridges of that type.
535 	 */
536 	ifv->ifv_if.if_type = p->if_type;
537 
538 	PSLIST_ENTRY_INIT(ifv, ifv_hash);
539 	idx = vlan_tag_hash(vid, ifv_hash.mask);
540 
541 	mutex_enter(&ifv_hash.lock);
542 	PSLIST_WRITER_INSERT_HEAD(&ifv_hash.lists[idx], ifv, ifv_hash);
543 	mutex_exit(&ifv_hash.lock);
544 
545 	vlan_linkmib_update(ifv, nmib);
546 	nmib = NULL;
547 	nmib_psref = NULL;
548 	omib_cleanup = true;
549 
550 done:
551 	mutex_exit(&ifv->ifv_lock);
552 
553 	if (nmib_psref)
554 		psref_target_destroy(nmib_psref, ifvm_psref_class);
555 	if (nmib)
556 		kmem_free(nmib, sizeof(*nmib));
557 	if (omib_cleanup)
558 		kmem_free(omib, sizeof(*omib));
559 
560 	return error;
561 }
562 
563 /*
564  * Unconfigure a VLAN interface.
565  */
566 static void
567 vlan_unconfig(struct ifnet *ifp)
568 {
569 	struct ifvlan *ifv = ifp->if_softc;
570 	struct ifvlan_linkmib *nmib = NULL;
571 	int error;
572 
573 	KASSERT(IFNET_LOCKED(ifp));
574 
575 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
576 
577 	mutex_enter(&ifv->ifv_lock);
578 	error = vlan_unconfig_locked(ifv, nmib);
579 	mutex_exit(&ifv->ifv_lock);
580 
581 	if (error)
582 		kmem_free(nmib, sizeof(*nmib));
583 }
584 static int
585 vlan_unconfig_locked(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
586 {
587 	struct ifnet *p;
588 	struct ifnet *ifp = &ifv->ifv_if;
589 	struct psref_target *nmib_psref = NULL;
590 	struct ifvlan_linkmib *omib;
591 	int error = 0;
592 
593 	KASSERT(IFNET_LOCKED(ifp));
594 	KASSERT(mutex_owned(&ifv->ifv_lock));
595 
596 	ifp->if_flags &= ~(IFF_UP|IFF_RUNNING);
597 
598 	omib = ifv->ifv_mib;
599 	p = omib->ifvm_p;
600 
601 	if (p == NULL) {
602 		error = -1;
603 		goto done;
604 	}
605 
606 	*nmib = *omib;
607 	nmib_psref = &nmib->ifvm_psref;
608 	psref_target_init(nmib_psref, ifvm_psref_class);
609 
610 	/*
611  	 * Since the interface is being unconfigured, we need to empty the
612 	 * list of multicast groups that we may have joined while we were
613 	 * alive and remove them from the parent's list also.
614 	 */
615 	(*nmib->ifvm_msw->vmsw_purgemulti)(ifv);
616 
617 	/* Disconnect from parent. */
618 	switch (p->if_type) {
619 	case IFT_ETHER:
620 	    {
621 		struct ethercom *ec = (void *)p;
622 		if (--ec->ec_nvlans == 0) {
623 			IFNET_LOCK(p);
624 			(void) ether_disable_vlan_mtu(p);
625 			IFNET_UNLOCK(p);
626 		}
627 
628 		/* XXX ether_ifdetach must not be called with IFNET_LOCK */
629 		mutex_exit(&ifv->ifv_lock);
630 		IFNET_UNLOCK(ifp);
631 		ether_ifdetach(ifp);
632 		IFNET_LOCK(ifp);
633 		mutex_enter(&ifv->ifv_lock);
634 
635 		/* Restore vlan_ioctl overwritten by ether_ifdetach */
636 		ifp->if_ioctl = vlan_ioctl;
637 		vlan_reset_linkname(ifp);
638 		break;
639 	    }
640 
641 	default:
642 		panic("%s: impossible", __func__);
643 	}
644 
645 	nmib->ifvm_p = NULL;
646 	ifv->ifv_if.if_mtu = 0;
647 	ifv->ifv_flags = 0;
648 
649 	mutex_enter(&ifv_hash.lock);
650 	PSLIST_WRITER_REMOVE(ifv, ifv_hash);
651 	pserialize_perform(vlan_psz);
652 	mutex_exit(&ifv_hash.lock);
653 	PSLIST_ENTRY_DESTROY(ifv, ifv_hash);
654 
655 	vlan_linkmib_update(ifv, nmib);
656 
657 	mutex_exit(&ifv->ifv_lock);
658 
659 	nmib_psref = NULL;
660 	kmem_free(omib, sizeof(*omib));
661 
662 #ifdef INET6
663 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
664 	/* To delete v6 link local addresses */
665 	if (in6_present)
666 		in6_ifdetach(ifp);
667 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
668 #endif
669 
670 	if ((ifp->if_flags & IFF_PROMISC) != 0)
671 		vlan_safe_ifpromisc_locked(ifp, 0);
672 	if_down_locked(ifp);
673 	ifp->if_capabilities = 0;
674 	mutex_enter(&ifv->ifv_lock);
675 done:
676 
677 	if (nmib_psref)
678 		psref_target_destroy(nmib_psref, ifvm_psref_class);
679 
680 	return error;
681 }
682 
683 static void
684 vlan_hash_init(void)
685 {
686 
687 	ifv_hash.lists = hashinit(VLAN_TAG_HASH_SIZE, HASH_PSLIST, true,
688 	    &ifv_hash.mask);
689 }
690 
691 static int
692 vlan_hash_fini(void)
693 {
694 	int i;
695 
696 	mutex_enter(&ifv_hash.lock);
697 
698 	for (i = 0; i < ifv_hash.mask + 1; i++) {
699 		if (PSLIST_WRITER_FIRST(&ifv_hash.lists[i], struct ifvlan,
700 		    ifv_hash) != NULL) {
701 			mutex_exit(&ifv_hash.lock);
702 			return EBUSY;
703 		}
704 	}
705 
706 	for (i = 0; i < ifv_hash.mask + 1; i++)
707 		PSLIST_DESTROY(&ifv_hash.lists[i]);
708 
709 	mutex_exit(&ifv_hash.lock);
710 
711 	hashdone(ifv_hash.lists, HASH_PSLIST, ifv_hash.mask);
712 
713 	ifv_hash.lists = NULL;
714 	ifv_hash.mask = 0;
715 
716 	return 0;
717 }
718 
719 static int
720 vlan_tag_hash(uint16_t tag, u_long mask)
721 {
722 	uint32_t hash;
723 
724 	hash = (tag >> 8) ^ tag;
725 	hash = (hash >> 2) ^ hash;
726 
727 	return hash & mask;
728 }
729 
730 static struct ifvlan_linkmib *
731 vlan_getref_linkmib(struct ifvlan *sc, struct psref *psref)
732 {
733 	struct ifvlan_linkmib *mib;
734 	int s;
735 
736 	s = pserialize_read_enter();
737 	mib = sc->ifv_mib;
738 	if (mib == NULL) {
739 		pserialize_read_exit(s);
740 		return NULL;
741 	}
742 	membar_datadep_consumer();
743 	psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
744 	pserialize_read_exit(s);
745 
746 	return mib;
747 }
748 
749 static void
750 vlan_putref_linkmib(struct ifvlan_linkmib *mib, struct psref *psref)
751 {
752 	if (mib == NULL)
753 		return;
754 	psref_release(psref, &mib->ifvm_psref, ifvm_psref_class);
755 }
756 
757 static struct ifvlan_linkmib *
758 vlan_lookup_tag_psref(struct ifnet *ifp, uint16_t tag, struct psref *psref)
759 {
760 	int idx;
761 	int s;
762 	struct ifvlan *sc;
763 
764 	idx = vlan_tag_hash(tag, ifv_hash.mask);
765 
766 	s = pserialize_read_enter();
767 	PSLIST_READER_FOREACH(sc, &ifv_hash.lists[idx], struct ifvlan,
768 	    ifv_hash) {
769 		struct ifvlan_linkmib *mib = sc->ifv_mib;
770 		if (mib == NULL)
771 			continue;
772 		if (mib->ifvm_tag != tag)
773 			continue;
774 		if (mib->ifvm_p != ifp)
775 			continue;
776 
777 		psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
778 		pserialize_read_exit(s);
779 		return mib;
780 	}
781 	pserialize_read_exit(s);
782 	return NULL;
783 }
784 
785 static void
786 vlan_linkmib_update(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
787 {
788 	struct ifvlan_linkmib *omib = ifv->ifv_mib;
789 
790 	KASSERT(mutex_owned(&ifv->ifv_lock));
791 
792 	membar_producer();
793 	ifv->ifv_mib = nmib;
794 
795 	pserialize_perform(vlan_psz);
796 	psref_target_destroy(&omib->ifvm_psref, ifvm_psref_class);
797 }
798 
799 /*
800  * Called when a parent interface is detaching; destroy any VLAN
801  * configuration for the parent interface.
802  */
803 void
804 vlan_ifdetach(struct ifnet *p)
805 {
806 	struct ifvlan *ifv;
807 	struct ifvlan_linkmib *mib, **nmibs;
808 	struct psref psref;
809 	int error;
810 	int bound;
811 	int i, cnt = 0;
812 
813 	bound = curlwp_bind();
814 
815 	mutex_enter(&ifv_list.lock);
816 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
817 		mib = vlan_getref_linkmib(ifv, &psref);
818 		if (mib == NULL)
819 			continue;
820 
821 		if (mib->ifvm_p == p)
822 			cnt++;
823 
824 		vlan_putref_linkmib(mib, &psref);
825 	}
826 	mutex_exit(&ifv_list.lock);
827 
828 	if (cnt == 0) {
829 		curlwp_bindx(bound);
830 		return;
831 	}
832 
833 	/*
834 	 * The value of "cnt" does not increase while ifv_list.lock
835 	 * and ifv->ifv_lock are released here, because the parent
836 	 * interface is detaching.
837 	 */
838 	nmibs = kmem_alloc(sizeof(*nmibs) * cnt, KM_SLEEP);
839 	for (i = 0; i < cnt; i++) {
840 		nmibs[i] = kmem_alloc(sizeof(*nmibs[i]), KM_SLEEP);
841 	}
842 
843 	mutex_enter(&ifv_list.lock);
844 
845 	i = 0;
846 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
847 		struct ifnet *ifp = &ifv->ifv_if;
848 
849 		/* IFNET_LOCK must be held before ifv_lock. */
850 		IFNET_LOCK(ifp);
851 		mutex_enter(&ifv->ifv_lock);
852 
853 		/* XXX ifv_mib = NULL? */
854 		if (ifv->ifv_mib->ifvm_p == p) {
855 			KASSERTMSG(i < cnt, "no memory for unconfig, parent=%s",
856 			    p->if_xname);
857 			error = vlan_unconfig_locked(ifv, nmibs[i]);
858 			if (!error) {
859 				nmibs[i] = NULL;
860 				i++;
861 			}
862 
863 		}
864 
865 		mutex_exit(&ifv->ifv_lock);
866 		IFNET_UNLOCK(ifp);
867 	}
868 
869 	mutex_exit(&ifv_list.lock);
870 
871 	curlwp_bindx(bound);
872 
873 	for (i = 0; i < cnt; i++) {
874 		if (nmibs[i])
875 			kmem_free(nmibs[i], sizeof(*nmibs[i]));
876 	}
877 
878 	kmem_free(nmibs, sizeof(*nmibs) * cnt);
879 
880 	return;
881 }
882 
883 static int
884 vlan_set_promisc(struct ifnet *ifp)
885 {
886 	struct ifvlan *ifv = ifp->if_softc;
887 	struct ifvlan_linkmib *mib;
888 	struct psref psref;
889 	int error = 0;
890 	int bound;
891 
892 	bound = curlwp_bind();
893 	mib = vlan_getref_linkmib(ifv, &psref);
894 	if (mib == NULL) {
895 		curlwp_bindx(bound);
896 		return EBUSY;
897 	}
898 
899 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
900 		if ((ifv->ifv_flags & IFVF_PROMISC) == 0) {
901 			error = vlan_safe_ifpromisc(mib->ifvm_p, 1);
902 			if (error == 0)
903 				ifv->ifv_flags |= IFVF_PROMISC;
904 		}
905 	} else {
906 		if ((ifv->ifv_flags & IFVF_PROMISC) != 0) {
907 			error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
908 			if (error == 0)
909 				ifv->ifv_flags &= ~IFVF_PROMISC;
910 		}
911 	}
912 	vlan_putref_linkmib(mib, &psref);
913 	curlwp_bindx(bound);
914 
915 	return error;
916 }
917 
918 static int
919 vlan_ioctl(struct ifnet *ifp, u_long cmd, void *data)
920 {
921 	struct lwp *l = curlwp;
922 	struct ifvlan *ifv = ifp->if_softc;
923 	struct ifaddr *ifa = (struct ifaddr *) data;
924 	struct ifreq *ifr = (struct ifreq *) data;
925 	struct ifnet *pr;
926 	struct ifcapreq *ifcr;
927 	struct vlanreq vlr;
928 	struct ifvlan_linkmib *mib;
929 	struct psref psref;
930 	int error = 0;
931 	int bound;
932 
933 	switch (cmd) {
934 	case SIOCSIFMTU:
935 		bound = curlwp_bind();
936 		mib = vlan_getref_linkmib(ifv, &psref);
937 		if (mib == NULL) {
938 			curlwp_bindx(bound);
939 			error = EBUSY;
940 			break;
941 		}
942 
943 		if (mib->ifvm_p == NULL) {
944 			vlan_putref_linkmib(mib, &psref);
945 			curlwp_bindx(bound);
946 			error = EINVAL;
947 		} else if (
948 		    ifr->ifr_mtu > (mib->ifvm_p->if_mtu - mib->ifvm_mtufudge) ||
949 		    ifr->ifr_mtu < (mib->ifvm_mintu - mib->ifvm_mtufudge)) {
950 			vlan_putref_linkmib(mib, &psref);
951 			curlwp_bindx(bound);
952 			error = EINVAL;
953 		} else {
954 			vlan_putref_linkmib(mib, &psref);
955 			curlwp_bindx(bound);
956 
957 			error = ifioctl_common(ifp, cmd, data);
958 			if (error == ENETRESET)
959 				error = 0;
960 		}
961 
962 		break;
963 
964 	case SIOCSETVLAN:
965 		if ((error = kauth_authorize_network(l->l_cred,
966 		    KAUTH_NETWORK_INTERFACE,
967 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
968 		    NULL)) != 0)
969 			break;
970 		if ((error = copyin(ifr->ifr_data, &vlr, sizeof(vlr))) != 0)
971 			break;
972 
973 		if (vlr.vlr_parent[0] == '\0') {
974 			bound = curlwp_bind();
975 			mib = vlan_getref_linkmib(ifv, &psref);
976 			if (mib == NULL) {
977 				curlwp_bindx(bound);
978 				error = EBUSY;
979 				break;
980 			}
981 
982 			if (mib->ifvm_p != NULL &&
983 			    (ifp->if_flags & IFF_PROMISC) != 0)
984 				error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
985 
986 			vlan_putref_linkmib(mib, &psref);
987 			curlwp_bindx(bound);
988 
989 			vlan_unconfig(ifp);
990 			break;
991 		}
992 		if (vlr.vlr_tag != EVL_VLANOFTAG(vlr.vlr_tag)) {
993 			error = EINVAL;		 /* check for valid tag */
994 			break;
995 		}
996 		if ((pr = ifunit(vlr.vlr_parent)) == NULL) {
997 			error = ENOENT;
998 			break;
999 		}
1000 		error = vlan_config(ifv, pr, vlr.vlr_tag);
1001 		if (error != 0) {
1002 			break;
1003 		}
1004 
1005 		/* Update promiscuous mode, if necessary. */
1006 		vlan_set_promisc(ifp);
1007 
1008 		ifp->if_flags |= IFF_RUNNING;
1009 		break;
1010 
1011 	case SIOCGETVLAN:
1012 		memset(&vlr, 0, sizeof(vlr));
1013 		bound = curlwp_bind();
1014 		mib = vlan_getref_linkmib(ifv, &psref);
1015 		if (mib == NULL) {
1016 			curlwp_bindx(bound);
1017 			error = EBUSY;
1018 			break;
1019 		}
1020 		if (mib->ifvm_p != NULL) {
1021 			snprintf(vlr.vlr_parent, sizeof(vlr.vlr_parent), "%s",
1022 			    mib->ifvm_p->if_xname);
1023 			vlr.vlr_tag = mib->ifvm_tag;
1024 		}
1025 		vlan_putref_linkmib(mib, &psref);
1026 		curlwp_bindx(bound);
1027 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1028 		break;
1029 
1030 	case SIOCSIFFLAGS:
1031 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
1032 			break;
1033 		/*
1034 		 * For promiscuous mode, we enable promiscuous mode on
1035 		 * the parent if we need promiscuous on the VLAN interface.
1036 		 */
1037 		bound = curlwp_bind();
1038 		mib = vlan_getref_linkmib(ifv, &psref);
1039 		if (mib == NULL) {
1040 			curlwp_bindx(bound);
1041 			error = EBUSY;
1042 			break;
1043 		}
1044 
1045 		if (mib->ifvm_p != NULL)
1046 			error = vlan_set_promisc(ifp);
1047 		vlan_putref_linkmib(mib, &psref);
1048 		curlwp_bindx(bound);
1049 		break;
1050 
1051 	case SIOCADDMULTI:
1052 		mutex_enter(&ifv->ifv_lock);
1053 		mib = ifv->ifv_mib;
1054 		if (mib == NULL) {
1055 			error = EBUSY;
1056 			mutex_exit(&ifv->ifv_lock);
1057 			break;
1058 		}
1059 
1060 		error = (mib->ifvm_p != NULL) ?
1061 		    (*mib->ifvm_msw->vmsw_addmulti)(ifv, ifr) : EINVAL;
1062 		mib = NULL;
1063 		mutex_exit(&ifv->ifv_lock);
1064 		break;
1065 
1066 	case SIOCDELMULTI:
1067 		mutex_enter(&ifv->ifv_lock);
1068 		mib = ifv->ifv_mib;
1069 		if (mib == NULL) {
1070 			error = EBUSY;
1071 			mutex_exit(&ifv->ifv_lock);
1072 			break;
1073 		}
1074 		error = (mib->ifvm_p != NULL) ?
1075 		    (*mib->ifvm_msw->vmsw_delmulti)(ifv, ifr) : EINVAL;
1076 		mib = NULL;
1077 		mutex_exit(&ifv->ifv_lock);
1078 		break;
1079 
1080 	case SIOCSIFCAP:
1081 		ifcr = data;
1082 		/* make sure caps are enabled on parent */
1083 		bound = curlwp_bind();
1084 		mib = vlan_getref_linkmib(ifv, &psref);
1085 		if (mib == NULL) {
1086 			curlwp_bindx(bound);
1087 			error = EBUSY;
1088 			break;
1089 		}
1090 
1091 		if (mib->ifvm_p == NULL) {
1092 			vlan_putref_linkmib(mib, &psref);
1093 			curlwp_bindx(bound);
1094 			error = EINVAL;
1095 			break;
1096 		}
1097 		if ((mib->ifvm_p->if_capenable & ifcr->ifcr_capenable) !=
1098 		    ifcr->ifcr_capenable) {
1099 			vlan_putref_linkmib(mib, &psref);
1100 			curlwp_bindx(bound);
1101 			error = EINVAL;
1102 			break;
1103 		}
1104 
1105 		vlan_putref_linkmib(mib, &psref);
1106 		curlwp_bindx(bound);
1107 
1108 		if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
1109 			error = 0;
1110 		break;
1111 	case SIOCINITIFADDR:
1112 		bound = curlwp_bind();
1113 		mib = vlan_getref_linkmib(ifv, &psref);
1114 		if (mib == NULL) {
1115 			curlwp_bindx(bound);
1116 			error = EBUSY;
1117 			break;
1118 		}
1119 
1120 		if (mib->ifvm_p == NULL) {
1121 			error = EINVAL;
1122 			vlan_putref_linkmib(mib, &psref);
1123 			curlwp_bindx(bound);
1124 			break;
1125 		}
1126 		vlan_putref_linkmib(mib, &psref);
1127 		curlwp_bindx(bound);
1128 
1129 		ifp->if_flags |= IFF_UP;
1130 #ifdef INET
1131 		if (ifa->ifa_addr->sa_family == AF_INET)
1132 			arp_ifinit(ifp, ifa);
1133 #endif
1134 		break;
1135 
1136 	default:
1137 		error = ether_ioctl(ifp, cmd, data);
1138 	}
1139 
1140 	return error;
1141 }
1142 
1143 static int
1144 vlan_ether_addmulti(struct ifvlan *ifv, struct ifreq *ifr)
1145 {
1146 	const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr);
1147 	struct vlan_mc_entry *mc;
1148 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1149 	struct ifvlan_linkmib *mib;
1150 	int error;
1151 
1152 	KASSERT(mutex_owned(&ifv->ifv_lock));
1153 
1154 	if (sa->sa_len > sizeof(struct sockaddr_storage))
1155 		return EINVAL;
1156 
1157 	error = ether_addmulti(sa, &ifv->ifv_ec);
1158 	if (error != ENETRESET)
1159 		return error;
1160 
1161 	/*
1162 	 * This is a new multicast address.  We have to tell parent
1163 	 * about it.  Also, remember this multicast address so that
1164 	 * we can delete it on unconfigure.
1165 	 */
1166 	mc = malloc(sizeof(struct vlan_mc_entry), M_DEVBUF, M_NOWAIT);
1167 	if (mc == NULL) {
1168 		error = ENOMEM;
1169 		goto alloc_failed;
1170 	}
1171 
1172 	/*
1173 	 * Since ether_addmulti() returned ENETRESET, the following two
1174 	 * statements shouldn't fail. Here ifv_ec is implicitly protected
1175 	 * by the ifv_lock lock.
1176 	 */
1177 	error = ether_multiaddr(sa, addrlo, addrhi);
1178 	KASSERT(error == 0);
1179 
1180 	ETHER_LOCK(&ifv->ifv_ec);
1181 	mc->mc_enm = ether_lookup_multi(addrlo, addrhi, &ifv->ifv_ec);
1182 	ETHER_UNLOCK(&ifv->ifv_ec);
1183 
1184 	KASSERT(mc->mc_enm != NULL);
1185 
1186 	memcpy(&mc->mc_addr, sa, sa->sa_len);
1187 	LIST_INSERT_HEAD(&ifv->ifv_mc_listhead, mc, mc_entries);
1188 
1189 	mib = ifv->ifv_mib;
1190 
1191 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1192 	IFNET_LOCK(mib->ifvm_p);
1193 	error = if_mcast_op(mib->ifvm_p, SIOCADDMULTI, sa);
1194 	IFNET_UNLOCK(mib->ifvm_p);
1195 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
1196 
1197 	if (error != 0)
1198 		goto ioctl_failed;
1199 	return error;
1200 
1201 ioctl_failed:
1202 	LIST_REMOVE(mc, mc_entries);
1203 	free(mc, M_DEVBUF);
1204 
1205 alloc_failed:
1206 	(void)ether_delmulti(sa, &ifv->ifv_ec);
1207 	return error;
1208 }
1209 
1210 static int
1211 vlan_ether_delmulti(struct ifvlan *ifv, struct ifreq *ifr)
1212 {
1213 	const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr);
1214 	struct ether_multi *enm;
1215 	struct vlan_mc_entry *mc;
1216 	struct ifvlan_linkmib *mib;
1217 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
1218 	int error;
1219 
1220 	KASSERT(mutex_owned(&ifv->ifv_lock));
1221 
1222 	/*
1223 	 * Find a key to lookup vlan_mc_entry.  We have to do this
1224 	 * before calling ether_delmulti for obvious reasons.
1225 	 */
1226 	if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0)
1227 		return error;
1228 
1229 	ETHER_LOCK(&ifv->ifv_ec);
1230 	enm = ether_lookup_multi(addrlo, addrhi, &ifv->ifv_ec);
1231 	ETHER_UNLOCK(&ifv->ifv_ec);
1232 	if (enm == NULL)
1233 		return EINVAL;
1234 
1235 	LIST_FOREACH(mc, &ifv->ifv_mc_listhead, mc_entries) {
1236 		if (mc->mc_enm == enm)
1237 			break;
1238 	}
1239 
1240 	/* We woun't delete entries we didn't add */
1241 	if (mc == NULL)
1242 		return EINVAL;
1243 
1244 	error = ether_delmulti(sa, &ifv->ifv_ec);
1245 	if (error != ENETRESET)
1246 		return error;
1247 
1248 	/* We no longer use this multicast address.  Tell parent so. */
1249 	mib = ifv->ifv_mib;
1250 	IFNET_LOCK(mib->ifvm_p);
1251 	error = if_mcast_op(mib->ifvm_p, SIOCDELMULTI, sa);
1252 	IFNET_UNLOCK(mib->ifvm_p);
1253 
1254 	if (error == 0) {
1255 		/* And forget about this address. */
1256 		LIST_REMOVE(mc, mc_entries);
1257 		free(mc, M_DEVBUF);
1258 	} else {
1259 		(void)ether_addmulti(sa, &ifv->ifv_ec);
1260 	}
1261 
1262 	return error;
1263 }
1264 
1265 /*
1266  * Delete any multicast address we have asked to add from parent
1267  * interface.  Called when the vlan is being unconfigured.
1268  */
1269 static void
1270 vlan_ether_purgemulti(struct ifvlan *ifv)
1271 {
1272 	struct vlan_mc_entry *mc;
1273 	struct ifvlan_linkmib *mib;
1274 
1275 	KASSERT(mutex_owned(&ifv->ifv_lock));
1276 	mib = ifv->ifv_mib;
1277 	if (mib == NULL) {
1278 		return;
1279 	}
1280 
1281 	while ((mc = LIST_FIRST(&ifv->ifv_mc_listhead)) != NULL) {
1282 		IFNET_LOCK(mib->ifvm_p);
1283 		(void)if_mcast_op(mib->ifvm_p, SIOCDELMULTI,
1284 		    sstocsa(&mc->mc_addr));
1285 		IFNET_UNLOCK(mib->ifvm_p);
1286 		LIST_REMOVE(mc, mc_entries);
1287 		free(mc, M_DEVBUF);
1288 	}
1289 }
1290 
1291 static void
1292 vlan_start(struct ifnet *ifp)
1293 {
1294 	struct ifvlan *ifv = ifp->if_softc;
1295 	struct ifnet *p;
1296 	struct ethercom *ec;
1297 	struct mbuf *m;
1298 	struct ifvlan_linkmib *mib;
1299 	struct psref psref;
1300 	int error;
1301 
1302 	mib = vlan_getref_linkmib(ifv, &psref);
1303 	if (mib == NULL)
1304 		return;
1305 	p = mib->ifvm_p;
1306 	ec = (void *)mib->ifvm_p;
1307 
1308 	ifp->if_flags |= IFF_OACTIVE;
1309 
1310 	for (;;) {
1311 		IFQ_DEQUEUE(&ifp->if_snd, m);
1312 		if (m == NULL)
1313 			break;
1314 
1315 #ifdef ALTQ
1316 		/*
1317 		 * KERNEL_LOCK is required for ALTQ even if NET_MPSAFE is
1318 		 * defined.
1319 		 */
1320 		KERNEL_LOCK(1, NULL);
1321 		/*
1322 		 * If ALTQ is enabled on the parent interface, do
1323 		 * classification; the queueing discipline might
1324 		 * not require classification, but might require
1325 		 * the address family/header pointer in the pktattr.
1326 		 */
1327 		if (ALTQ_IS_ENABLED(&p->if_snd)) {
1328 			switch (p->if_type) {
1329 			case IFT_ETHER:
1330 				altq_etherclassify(&p->if_snd, m);
1331 				break;
1332 			default:
1333 				panic("%s: impossible (altq)", __func__);
1334 			}
1335 		}
1336 		KERNEL_UNLOCK_ONE(NULL);
1337 #endif /* ALTQ */
1338 
1339 		bpf_mtap(ifp, m, BPF_D_OUT);
1340 		/*
1341 		 * If the parent can insert the tag itself, just mark
1342 		 * the tag in the mbuf header.
1343 		 */
1344 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
1345 			vlan_set_tag(m, mib->ifvm_tag);
1346 		} else {
1347 			/*
1348 			 * insert the tag ourselves
1349 			 */
1350 			M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1351 			if (m == NULL) {
1352 				printf("%s: unable to prepend encap header",
1353 				    p->if_xname);
1354 				ifp->if_oerrors++;
1355 				continue;
1356 			}
1357 
1358 			switch (p->if_type) {
1359 			case IFT_ETHER:
1360 			    {
1361 				struct ether_vlan_header *evl;
1362 
1363 				if (m->m_len < sizeof(struct ether_vlan_header))
1364 					m = m_pullup(m,
1365 					    sizeof(struct ether_vlan_header));
1366 				if (m == NULL) {
1367 					printf("%s: unable to pullup encap "
1368 					    "header", p->if_xname);
1369 					ifp->if_oerrors++;
1370 					continue;
1371 				}
1372 
1373 				/*
1374 				 * Transform the Ethernet header into an
1375 				 * Ethernet header with 802.1Q encapsulation.
1376 				 */
1377 				memmove(mtod(m, void *),
1378 				    mtod(m, char *) + mib->ifvm_encaplen,
1379 				    sizeof(struct ether_header));
1380 				evl = mtod(m, struct ether_vlan_header *);
1381 				evl->evl_proto = evl->evl_encap_proto;
1382 				evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1383 				evl->evl_tag = htons(mib->ifvm_tag);
1384 
1385 				/*
1386 				 * To cater for VLAN-aware layer 2 ethernet
1387 				 * switches which may need to strip the tag
1388 				 * before forwarding the packet, make sure
1389 				 * the packet+tag is at least 68 bytes long.
1390 				 * This is necessary because our parent will
1391 				 * only pad to 64 bytes (ETHER_MIN_LEN) and
1392 				 * some switches will not pad by themselves
1393 				 * after deleting a tag.
1394 				 */
1395 				const size_t min_data_len = ETHER_MIN_LEN -
1396 				    ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
1397 				if (m->m_pkthdr.len < min_data_len) {
1398 					m_copyback(m, m->m_pkthdr.len,
1399 					    min_data_len - m->m_pkthdr.len,
1400 					    vlan_zero_pad_buff);
1401 				}
1402 				break;
1403 			    }
1404 
1405 			default:
1406 				panic("%s: impossible", __func__);
1407 			}
1408 		}
1409 
1410 		if ((p->if_flags & IFF_RUNNING) == 0) {
1411 			m_freem(m);
1412 			continue;
1413 		}
1414 
1415 		error = if_transmit_lock(p, m);
1416 		if (error) {
1417 			/* mbuf is already freed */
1418 			ifp->if_oerrors++;
1419 			continue;
1420 		}
1421 		ifp->if_opackets++;
1422 	}
1423 
1424 	ifp->if_flags &= ~IFF_OACTIVE;
1425 
1426 	/* Remove reference to mib before release */
1427 	vlan_putref_linkmib(mib, &psref);
1428 }
1429 
1430 static int
1431 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1432 {
1433 	struct ifvlan *ifv = ifp->if_softc;
1434 	struct ifnet *p;
1435 	struct ethercom *ec;
1436 	struct ifvlan_linkmib *mib;
1437 	struct psref psref;
1438 	int error;
1439 	size_t pktlen = m->m_pkthdr.len;
1440 	bool mcast = (m->m_flags & M_MCAST) != 0;
1441 
1442 	mib = vlan_getref_linkmib(ifv, &psref);
1443 	if (mib == NULL) {
1444 		m_freem(m);
1445 		return ENETDOWN;
1446 	}
1447 
1448 	p = mib->ifvm_p;
1449 	ec = (void *)mib->ifvm_p;
1450 
1451 	bpf_mtap(ifp, m, BPF_D_OUT);
1452 
1453 	if ((error = pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_OUT)) != 0)
1454 		goto out;
1455 	if (m == NULL)
1456 		goto out;
1457 
1458 	/*
1459 	 * If the parent can insert the tag itself, just mark
1460 	 * the tag in the mbuf header.
1461 	 */
1462 	if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
1463 		vlan_set_tag(m, mib->ifvm_tag);
1464 	} else {
1465 		/*
1466 		 * insert the tag ourselves
1467 		 */
1468 		M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
1469 		if (m == NULL) {
1470 			printf("%s: unable to prepend encap header",
1471 			    p->if_xname);
1472 			ifp->if_oerrors++;
1473 			error = ENOBUFS;
1474 			goto out;
1475 		}
1476 
1477 		switch (p->if_type) {
1478 		case IFT_ETHER:
1479 		    {
1480 			struct ether_vlan_header *evl;
1481 
1482 			if (m->m_len < sizeof(struct ether_vlan_header))
1483 				m = m_pullup(m,
1484 				    sizeof(struct ether_vlan_header));
1485 			if (m == NULL) {
1486 				printf("%s: unable to pullup encap "
1487 				    "header", p->if_xname);
1488 				ifp->if_oerrors++;
1489 				error = ENOBUFS;
1490 				goto out;
1491 			}
1492 
1493 			/*
1494 			 * Transform the Ethernet header into an
1495 			 * Ethernet header with 802.1Q encapsulation.
1496 			 */
1497 			memmove(mtod(m, void *),
1498 			    mtod(m, char *) + mib->ifvm_encaplen,
1499 			    sizeof(struct ether_header));
1500 			evl = mtod(m, struct ether_vlan_header *);
1501 			evl->evl_proto = evl->evl_encap_proto;
1502 			evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1503 			evl->evl_tag = htons(mib->ifvm_tag);
1504 
1505 			/*
1506 			 * To cater for VLAN-aware layer 2 ethernet
1507 			 * switches which may need to strip the tag
1508 			 * before forwarding the packet, make sure
1509 			 * the packet+tag is at least 68 bytes long.
1510 			 * This is necessary because our parent will
1511 			 * only pad to 64 bytes (ETHER_MIN_LEN) and
1512 			 * some switches will not pad by themselves
1513 			 * after deleting a tag.
1514 			 */
1515 			const size_t min_data_len = ETHER_MIN_LEN -
1516 			    ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
1517 			if (m->m_pkthdr.len < min_data_len) {
1518 				m_copyback(m, m->m_pkthdr.len,
1519 				    min_data_len - m->m_pkthdr.len,
1520 				    vlan_zero_pad_buff);
1521 			}
1522 			break;
1523 		    }
1524 
1525 		default:
1526 			panic("%s: impossible", __func__);
1527 		}
1528 	}
1529 
1530 	if ((p->if_flags & IFF_RUNNING) == 0) {
1531 		m_freem(m);
1532 		error = ENETDOWN;
1533 		goto out;
1534 	}
1535 
1536 	error = if_transmit_lock(p, m);
1537 	if (error) {
1538 		/* mbuf is already freed */
1539 		ifp->if_oerrors++;
1540 	} else {
1541 
1542 		ifp->if_opackets++;
1543 		ifp->if_obytes += pktlen;
1544 		if (mcast)
1545 			ifp->if_omcasts++;
1546 	}
1547 
1548 out:
1549 	/* Remove reference to mib before release */
1550 	vlan_putref_linkmib(mib, &psref);
1551 	return error;
1552 }
1553 
1554 /*
1555  * Given an Ethernet frame, find a valid vlan interface corresponding to the
1556  * given source interface and tag, then run the real packet through the
1557  * parent's input routine.
1558  */
1559 void
1560 vlan_input(struct ifnet *ifp, struct mbuf *m)
1561 {
1562 	struct ifvlan *ifv;
1563 	uint16_t vid;
1564 	struct ifvlan_linkmib *mib;
1565 	struct psref psref;
1566 	bool have_vtag;
1567 
1568 	have_vtag = vlan_has_tag(m);
1569 	if (have_vtag) {
1570 		vid = EVL_VLANOFTAG(vlan_get_tag(m));
1571 		m->m_flags &= ~M_VLANTAG;
1572 	} else {
1573 		struct ether_vlan_header *evl;
1574 
1575 		if (ifp->if_type != IFT_ETHER) {
1576 			panic("%s: impossible", __func__);
1577 		}
1578 
1579 		if (m->m_len < sizeof(struct ether_vlan_header) &&
1580 		    (m = m_pullup(m,
1581 		     sizeof(struct ether_vlan_header))) == NULL) {
1582 			printf("%s: no memory for VLAN header, "
1583 			    "dropping packet.\n", ifp->if_xname);
1584 			return;
1585 		}
1586 		evl = mtod(m, struct ether_vlan_header *);
1587 		KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN);
1588 
1589 		vid = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1590 
1591 		/*
1592 		 * Restore the original ethertype.  We'll remove
1593 		 * the encapsulation after we've found the vlan
1594 		 * interface corresponding to the tag.
1595 		 */
1596 		evl->evl_encap_proto = evl->evl_proto;
1597 	}
1598 
1599 	mib = vlan_lookup_tag_psref(ifp, vid, &psref);
1600 	if (mib == NULL) {
1601 		m_freem(m);
1602 		ifp->if_noproto++;
1603 		return;
1604 	}
1605 	KASSERT(mib->ifvm_encaplen == ETHER_VLAN_ENCAP_LEN);
1606 
1607 	ifv = mib->ifvm_ifvlan;
1608 	if ((ifv->ifv_if.if_flags & (IFF_UP|IFF_RUNNING)) !=
1609 	    (IFF_UP|IFF_RUNNING)) {
1610 		m_freem(m);
1611 		ifp->if_noproto++;
1612 		goto out;
1613 	}
1614 
1615 	/*
1616 	 * Now, remove the encapsulation header.  The original
1617 	 * header has already been fixed up above.
1618 	 */
1619 	if (!have_vtag) {
1620 		memmove(mtod(m, char *) + mib->ifvm_encaplen,
1621 		    mtod(m, void *), sizeof(struct ether_header));
1622 		m_adj(m, mib->ifvm_encaplen);
1623 	}
1624 
1625 	m_set_rcvif(m, &ifv->ifv_if);
1626 	ifv->ifv_if.if_ipackets++;
1627 
1628 	if (pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_IN) != 0)
1629 		goto out;
1630 	if (m == NULL)
1631 		goto out;
1632 
1633 	m->m_flags &= ~M_PROMISC;
1634 	if_input(&ifv->ifv_if, m);
1635 out:
1636 	vlan_putref_linkmib(mib, &psref);
1637 }
1638 
1639 /*
1640  * Module infrastructure
1641  */
1642 #include "if_module.h"
1643 
1644 IF_MODULE(MODULE_CLASS_DRIVER, vlan, "")
1645