Home | History | Annotate | Line # | Download | only in net
if_vlan.c revision 1.133
      1 /*	$NetBSD: if_vlan.c,v 1.133 2018/10/19 00:12:56 knakahara 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 1.133 2018/10/19 00:12:56 knakahara Exp $");
     82 
     83 #ifdef _KERNEL_OPT
     84 #include "opt_inet.h"
     85 #include "opt_net_mpsafe.h"
     86 #endif
     87 
     88 #include <sys/param.h>
     89 #include <sys/systm.h>
     90 #include <sys/kernel.h>
     91 #include <sys/mbuf.h>
     92 #include <sys/queue.h>
     93 #include <sys/socket.h>
     94 #include <sys/sockio.h>
     95 #include <sys/systm.h>
     96 #include <sys/proc.h>
     97 #include <sys/kauth.h>
     98 #include <sys/mutex.h>
     99 #include <sys/kmem.h>
    100 #include <sys/cpu.h>
    101 #include <sys/pserialize.h>
    102 #include <sys/psref.h>
    103 #include <sys/pslist.h>
    104 #include <sys/atomic.h>
    105 #include <sys/device.h>
    106 #include <sys/module.h>
    107 
    108 #include <net/bpf.h>
    109 #include <net/if.h>
    110 #include <net/if_dl.h>
    111 #include <net/if_types.h>
    112 #include <net/if_ether.h>
    113 #include <net/if_vlanvar.h>
    114 
    115 #ifdef INET
    116 #include <netinet/in.h>
    117 #include <netinet/if_inarp.h>
    118 #endif
    119 #ifdef INET6
    120 #include <netinet6/in6_ifattach.h>
    121 #include <netinet6/in6_var.h>
    122 #endif
    123 
    124 #include "ioconf.h"
    125 
    126 struct vlan_mc_entry {
    127 	LIST_ENTRY(vlan_mc_entry)	mc_entries;
    128 	/*
    129 	 * A key to identify this entry.  The mc_addr below can't be
    130 	 * used since multiple sockaddr may mapped into the same
    131 	 * ether_multi (e.g., AF_UNSPEC).
    132 	 */
    133 	union {
    134 		struct ether_multi	*mcu_enm;
    135 	} mc_u;
    136 	struct sockaddr_storage		mc_addr;
    137 };
    138 
    139 #define	mc_enm		mc_u.mcu_enm
    140 
    141 
    142 struct ifvlan_linkmib {
    143 	struct ifvlan *ifvm_ifvlan;
    144 	const struct vlan_multisw *ifvm_msw;
    145 	int	ifvm_encaplen;	/* encapsulation length */
    146 	int	ifvm_mtufudge;	/* MTU fudged by this much */
    147 	int	ifvm_mintu;	/* min transmission unit */
    148 	uint16_t ifvm_proto;	/* encapsulation ethertype */
    149 	uint16_t ifvm_tag;	/* tag to apply on packets */
    150 	struct ifnet *ifvm_p;		/* parent interface of this vlan */
    151 
    152 	struct psref_target ifvm_psref;
    153 };
    154 
    155 struct ifvlan {
    156 	union {
    157 		struct ethercom ifvu_ec;
    158 	} ifv_u;
    159 	struct ifvlan_linkmib *ifv_mib;	/*
    160 					 * reader must use vlan_getref_linkmib()
    161 					 * instead of direct dereference
    162 					 */
    163 	kmutex_t ifv_lock;		/* writer lock for ifv_mib */
    164 	pserialize_t ifv_psz;
    165 
    166 	LIST_HEAD(__vlan_mchead, vlan_mc_entry) ifv_mc_listhead;
    167 	LIST_ENTRY(ifvlan) ifv_list;
    168 	struct pslist_entry ifv_hash;
    169 	int ifv_flags;
    170 };
    171 
    172 #define	IFVF_PROMISC	0x01		/* promiscuous mode enabled */
    173 
    174 #define	ifv_ec		ifv_u.ifvu_ec
    175 
    176 #define	ifv_if		ifv_ec.ec_if
    177 
    178 #define	ifv_msw		ifv_mib.ifvm_msw
    179 #define	ifv_encaplen	ifv_mib.ifvm_encaplen
    180 #define	ifv_mtufudge	ifv_mib.ifvm_mtufudge
    181 #define	ifv_mintu	ifv_mib.ifvm_mintu
    182 #define	ifv_tag		ifv_mib.ifvm_tag
    183 
    184 struct vlan_multisw {
    185 	int	(*vmsw_addmulti)(struct ifvlan *, struct ifreq *);
    186 	int	(*vmsw_delmulti)(struct ifvlan *, struct ifreq *);
    187 	void	(*vmsw_purgemulti)(struct ifvlan *);
    188 };
    189 
    190 static int	vlan_ether_addmulti(struct ifvlan *, struct ifreq *);
    191 static int	vlan_ether_delmulti(struct ifvlan *, struct ifreq *);
    192 static void	vlan_ether_purgemulti(struct ifvlan *);
    193 
    194 const struct vlan_multisw vlan_ether_multisw = {
    195 	.vmsw_addmulti = vlan_ether_addmulti,
    196 	.vmsw_delmulti = vlan_ether_delmulti,
    197 	.vmsw_purgemulti = vlan_ether_purgemulti,
    198 };
    199 
    200 static int	vlan_clone_create(struct if_clone *, int);
    201 static int	vlan_clone_destroy(struct ifnet *);
    202 static int	vlan_config(struct ifvlan *, struct ifnet *,
    203     uint16_t);
    204 static int	vlan_ioctl(struct ifnet *, u_long, void *);
    205 static void	vlan_start(struct ifnet *);
    206 static int	vlan_transmit(struct ifnet *, struct mbuf *);
    207 static void	vlan_unconfig(struct ifnet *);
    208 static int	vlan_unconfig_locked(struct ifvlan *,
    209     struct ifvlan_linkmib *);
    210 static void	vlan_hash_init(void);
    211 static int	vlan_hash_fini(void);
    212 static int	vlan_tag_hash(uint16_t, u_long);
    213 static struct ifvlan_linkmib*	vlan_getref_linkmib(struct ifvlan *,
    214     struct psref *);
    215 static void	vlan_putref_linkmib(struct ifvlan_linkmib *,
    216     struct psref *);
    217 static void	vlan_linkmib_update(struct ifvlan *,
    218     struct ifvlan_linkmib *);
    219 static struct ifvlan_linkmib*	vlan_lookup_tag_psref(struct ifnet *,
    220     uint16_t, struct psref *);
    221 
    222 LIST_HEAD(vlan_ifvlist, ifvlan);
    223 static struct {
    224 	kmutex_t lock;
    225 	struct vlan_ifvlist list;
    226 } ifv_list __cacheline_aligned;
    227 
    228 
    229 #if !defined(VLAN_TAG_HASH_SIZE)
    230 #define VLAN_TAG_HASH_SIZE 32
    231 #endif
    232 static struct {
    233 	kmutex_t lock;
    234 	struct pslist_head *lists;
    235 	u_long mask;
    236 } ifv_hash __cacheline_aligned = {
    237 	.lists = NULL,
    238 	.mask = 0,
    239 };
    240 
    241 pserialize_t vlan_psz __read_mostly;
    242 static struct psref_class *ifvm_psref_class __read_mostly;
    243 
    244 struct if_clone vlan_cloner =
    245     IF_CLONE_INITIALIZER("vlan", vlan_clone_create, vlan_clone_destroy);
    246 
    247 /* Used to pad ethernet frames with < ETHER_MIN_LEN bytes */
    248 static char vlan_zero_pad_buff[ETHER_MIN_LEN];
    249 
    250 static inline int
    251 vlan_safe_ifpromisc(struct ifnet *ifp, int pswitch)
    252 {
    253 	int e;
    254 
    255 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
    256 	e = ifpromisc(ifp, pswitch);
    257 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    258 
    259 	return e;
    260 }
    261 
    262 static inline int
    263 vlan_safe_ifpromisc_locked(struct ifnet *ifp, int pswitch)
    264 {
    265 	int e;
    266 
    267 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
    268 	e = ifpromisc_locked(ifp, pswitch);
    269 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    270 
    271 	return e;
    272 }
    273 
    274 void
    275 vlanattach(int n)
    276 {
    277 
    278 	/*
    279 	 * Nothing to do here, initialization is handled by the
    280 	 * module initialization code in vlaninit() below.
    281 	 */
    282 }
    283 
    284 static void
    285 vlaninit(void)
    286 {
    287 	mutex_init(&ifv_list.lock, MUTEX_DEFAULT, IPL_NONE);
    288 	LIST_INIT(&ifv_list.list);
    289 
    290 	mutex_init(&ifv_hash.lock, MUTEX_DEFAULT, IPL_NONE);
    291 	vlan_psz = pserialize_create();
    292 	ifvm_psref_class = psref_class_create("vlanlinkmib", IPL_SOFTNET);
    293 	if_clone_attach(&vlan_cloner);
    294 
    295 	vlan_hash_init();
    296 }
    297 
    298 static int
    299 vlandetach(void)
    300 {
    301 	bool is_empty;
    302 	int error;
    303 
    304 	mutex_enter(&ifv_list.lock);
    305 	is_empty = LIST_EMPTY(&ifv_list.list);
    306 	mutex_exit(&ifv_list.lock);
    307 
    308 	if (!is_empty)
    309 		return EBUSY;
    310 
    311 	error = vlan_hash_fini();
    312 	if (error != 0)
    313 		return error;
    314 
    315 	if_clone_detach(&vlan_cloner);
    316 	psref_class_destroy(ifvm_psref_class);
    317 	pserialize_destroy(vlan_psz);
    318 	mutex_destroy(&ifv_hash.lock);
    319 	mutex_destroy(&ifv_list.lock);
    320 
    321 	return 0;
    322 }
    323 
    324 static void
    325 vlan_reset_linkname(struct ifnet *ifp)
    326 {
    327 
    328 	/*
    329 	 * We start out with a "802.1Q VLAN" type and zero-length
    330 	 * addresses.  When we attach to a parent interface, we
    331 	 * inherit its type, address length, address, and data link
    332 	 * type.
    333 	 */
    334 
    335 	ifp->if_type = IFT_L2VLAN;
    336 	ifp->if_addrlen = 0;
    337 	ifp->if_dlt = DLT_NULL;
    338 	if_alloc_sadl(ifp);
    339 }
    340 
    341 static int
    342 vlan_clone_create(struct if_clone *ifc, int unit)
    343 {
    344 	struct ifvlan *ifv;
    345 	struct ifnet *ifp;
    346 	struct ifvlan_linkmib *mib;
    347 	int rv;
    348 
    349 	ifv = malloc(sizeof(struct ifvlan), M_DEVBUF, M_WAITOK|M_ZERO);
    350 	mib = kmem_zalloc(sizeof(struct ifvlan_linkmib), KM_SLEEP);
    351 	ifp = &ifv->ifv_if;
    352 	LIST_INIT(&ifv->ifv_mc_listhead);
    353 
    354 	mib->ifvm_ifvlan = ifv;
    355 	mib->ifvm_p = NULL;
    356 	psref_target_init(&mib->ifvm_psref, ifvm_psref_class);
    357 
    358 	mutex_init(&ifv->ifv_lock, MUTEX_DEFAULT, IPL_NONE);
    359 	ifv->ifv_psz = pserialize_create();
    360 	ifv->ifv_mib = mib;
    361 
    362 	mutex_enter(&ifv_list.lock);
    363 	LIST_INSERT_HEAD(&ifv_list.list, ifv, ifv_list);
    364 	mutex_exit(&ifv_list.lock);
    365 
    366 	if_initname(ifp, ifc->ifc_name, unit);
    367 	ifp->if_softc = ifv;
    368 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
    369 	ifp->if_extflags = IFEF_NO_LINK_STATE_CHANGE;
    370 #ifdef NET_MPSAFE
    371 	ifp->if_extflags |= IFEF_MPSAFE;
    372 #endif
    373 	ifp->if_start = vlan_start;
    374 	ifp->if_transmit = vlan_transmit;
    375 	ifp->if_ioctl = vlan_ioctl;
    376 	IFQ_SET_READY(&ifp->if_snd);
    377 
    378 	rv = if_initialize(ifp);
    379 	if (rv != 0) {
    380 		aprint_error("%s: if_initialize failed(%d)\n", ifp->if_xname,
    381 		    rv);
    382 		goto fail;
    383 	}
    384 
    385 	vlan_reset_linkname(ifp);
    386 	if_register(ifp);
    387 	return 0;
    388 
    389 fail:
    390 	mutex_enter(&ifv_list.lock);
    391 	LIST_REMOVE(ifv, ifv_list);
    392 	mutex_exit(&ifv_list.lock);
    393 
    394 	mutex_destroy(&ifv->ifv_lock);
    395 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
    396 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
    397 	free(ifv, M_DEVBUF);
    398 
    399 	return rv;
    400 }
    401 
    402 static int
    403 vlan_clone_destroy(struct ifnet *ifp)
    404 {
    405 	struct ifvlan *ifv = ifp->if_softc;
    406 
    407 	mutex_enter(&ifv_list.lock);
    408 	LIST_REMOVE(ifv, ifv_list);
    409 	mutex_exit(&ifv_list.lock);
    410 
    411 	IFNET_LOCK(ifp);
    412 	vlan_unconfig(ifp);
    413 	IFNET_UNLOCK(ifp);
    414 	if_detach(ifp);
    415 
    416 	psref_target_destroy(&ifv->ifv_mib->ifvm_psref, ifvm_psref_class);
    417 	kmem_free(ifv->ifv_mib, sizeof(struct ifvlan_linkmib));
    418 	pserialize_destroy(ifv->ifv_psz);
    419 	mutex_destroy(&ifv->ifv_lock);
    420 	free(ifv, M_DEVBUF);
    421 
    422 	return 0;
    423 }
    424 
    425 /*
    426  * Configure a VLAN interface.
    427  */
    428 static int
    429 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
    430 {
    431 	struct ifnet *ifp = &ifv->ifv_if;
    432 	struct ifvlan_linkmib *nmib = NULL;
    433 	struct ifvlan_linkmib *omib = NULL;
    434 	struct ifvlan_linkmib *checkmib;
    435 	struct psref_target *nmib_psref = NULL;
    436 	const uint16_t vid = EVL_VLANOFTAG(tag);
    437 	int error = 0;
    438 	int idx;
    439 	bool omib_cleanup = false;
    440 	struct psref psref;
    441 
    442 	/* VLAN ID 0 and 4095 are reserved in the spec */
    443 	if ((vid == 0) || (vid == 0xfff))
    444 		return EINVAL;
    445 
    446 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
    447 	mutex_enter(&ifv->ifv_lock);
    448 	omib = ifv->ifv_mib;
    449 
    450 	if (omib->ifvm_p != NULL) {
    451 		error = EBUSY;
    452 		goto done;
    453 	}
    454 
    455 	/* Duplicate check */
    456 	checkmib = vlan_lookup_tag_psref(p, vid, &psref);
    457 	if (checkmib != NULL) {
    458 		vlan_putref_linkmib(checkmib, &psref);
    459 		error = EEXIST;
    460 		goto done;
    461 	}
    462 
    463 	*nmib = *omib;
    464 	nmib_psref = &nmib->ifvm_psref;
    465 
    466 	psref_target_init(nmib_psref, ifvm_psref_class);
    467 
    468 	switch (p->if_type) {
    469 	case IFT_ETHER:
    470 	    {
    471 		struct ethercom *ec = (void *)p;
    472 		nmib->ifvm_msw = &vlan_ether_multisw;
    473 		nmib->ifvm_encaplen = ETHER_VLAN_ENCAP_LEN;
    474 		nmib->ifvm_mintu = ETHERMIN;
    475 
    476 		if (ec->ec_nvlans++ == 0) {
    477 			IFNET_LOCK(p);
    478 			error = ether_enable_vlan_mtu(p);
    479 			IFNET_UNLOCK(p);
    480 			if (error >= 0) {
    481 				if (error) {
    482 					ec->ec_nvlans--;
    483 					goto done;
    484 				}
    485 				nmib->ifvm_mtufudge = 0;
    486 			} else {
    487 				/*
    488 				 * Fudge the MTU by the encapsulation size. This
    489 				 * makes us incompatible with strictly compliant
    490 				 * 802.1Q implementations, but allows us to use
    491 				 * the feature with other NetBSD
    492 				 * implementations, which might still be useful.
    493 				 */
    494 				nmib->ifvm_mtufudge = nmib->ifvm_encaplen;
    495 			}
    496 			error = 0;
    497 		}
    498 
    499 		/*
    500 		 * If the parent interface can do hardware-assisted
    501 		 * VLAN encapsulation, then propagate its hardware-
    502 		 * assisted checksumming flags and tcp segmentation
    503 		 * offload.
    504 		 */
    505 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
    506 			ec->ec_capenable |= ETHERCAP_VLAN_HWTAGGING;
    507 			ifp->if_capabilities = p->if_capabilities &
    508 			    (IFCAP_TSOv4 | IFCAP_TSOv6 |
    509 			     IFCAP_CSUM_IPv4_Tx|IFCAP_CSUM_IPv4_Rx|
    510 			     IFCAP_CSUM_TCPv4_Tx|IFCAP_CSUM_TCPv4_Rx|
    511 			     IFCAP_CSUM_UDPv4_Tx|IFCAP_CSUM_UDPv4_Rx|
    512 			     IFCAP_CSUM_TCPv6_Tx|IFCAP_CSUM_TCPv6_Rx|
    513 			     IFCAP_CSUM_UDPv6_Tx|IFCAP_CSUM_UDPv6_Rx);
    514 		}
    515 
    516 		/*
    517 		 * We inherit the parent's Ethernet address.
    518 		 */
    519 		ether_ifattach(ifp, CLLADDR(p->if_sadl));
    520 		ifp->if_hdrlen = sizeof(struct ether_vlan_header); /* XXX? */
    521 		break;
    522 	    }
    523 
    524 	default:
    525 		error = EPROTONOSUPPORT;
    526 		goto done;
    527 	}
    528 
    529 	nmib->ifvm_p = p;
    530 	nmib->ifvm_tag = vid;
    531 	ifv->ifv_if.if_mtu = p->if_mtu - nmib->ifvm_mtufudge;
    532 	ifv->ifv_if.if_flags = p->if_flags &
    533 	    (IFF_UP | IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
    534 
    535 	/*
    536 	 * Inherit the if_type from the parent.  This allows us
    537 	 * to participate in bridges of that type.
    538 	 */
    539 	ifv->ifv_if.if_type = p->if_type;
    540 
    541 	PSLIST_ENTRY_INIT(ifv, ifv_hash);
    542 	idx = vlan_tag_hash(vid, ifv_hash.mask);
    543 
    544 	mutex_enter(&ifv_hash.lock);
    545 	PSLIST_WRITER_INSERT_HEAD(&ifv_hash.lists[idx], ifv, ifv_hash);
    546 	mutex_exit(&ifv_hash.lock);
    547 
    548 	vlan_linkmib_update(ifv, nmib);
    549 	nmib = NULL;
    550 	nmib_psref = NULL;
    551 	omib_cleanup = true;
    552 
    553 done:
    554 	mutex_exit(&ifv->ifv_lock);
    555 
    556 	if (nmib_psref)
    557 		psref_target_destroy(nmib_psref, ifvm_psref_class);
    558 	if (nmib)
    559 		kmem_free(nmib, sizeof(*nmib));
    560 	if (omib_cleanup)
    561 		kmem_free(omib, sizeof(*omib));
    562 
    563 	return error;
    564 }
    565 
    566 /*
    567  * Unconfigure a VLAN interface.
    568  */
    569 static void
    570 vlan_unconfig(struct ifnet *ifp)
    571 {
    572 	struct ifvlan *ifv = ifp->if_softc;
    573 	struct ifvlan_linkmib *nmib = NULL;
    574 	int error;
    575 
    576 	KASSERT(IFNET_LOCKED(ifp));
    577 
    578 	nmib = kmem_alloc(sizeof(*nmib), KM_SLEEP);
    579 
    580 	mutex_enter(&ifv->ifv_lock);
    581 	error = vlan_unconfig_locked(ifv, nmib);
    582 	mutex_exit(&ifv->ifv_lock);
    583 
    584 	if (error)
    585 		kmem_free(nmib, sizeof(*nmib));
    586 }
    587 static int
    588 vlan_unconfig_locked(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
    589 {
    590 	struct ifnet *p;
    591 	struct ifnet *ifp = &ifv->ifv_if;
    592 	struct psref_target *nmib_psref = NULL;
    593 	struct ifvlan_linkmib *omib;
    594 	int error = 0;
    595 
    596 	KASSERT(IFNET_LOCKED(ifp));
    597 	KASSERT(mutex_owned(&ifv->ifv_lock));
    598 
    599 	ifp->if_flags &= ~(IFF_UP|IFF_RUNNING);
    600 
    601 	omib = ifv->ifv_mib;
    602 	p = omib->ifvm_p;
    603 
    604 	if (p == NULL) {
    605 		error = -1;
    606 		goto done;
    607 	}
    608 
    609 	*nmib = *omib;
    610 	nmib_psref = &nmib->ifvm_psref;
    611 	psref_target_init(nmib_psref, ifvm_psref_class);
    612 
    613 	/*
    614  	 * Since the interface is being unconfigured, we need to empty the
    615 	 * list of multicast groups that we may have joined while we were
    616 	 * alive and remove them from the parent's list also.
    617 	 */
    618 	(*nmib->ifvm_msw->vmsw_purgemulti)(ifv);
    619 
    620 	/* Disconnect from parent. */
    621 	switch (p->if_type) {
    622 	case IFT_ETHER:
    623 	    {
    624 		struct ethercom *ec = (void *)p;
    625 		if (--ec->ec_nvlans == 0) {
    626 			IFNET_LOCK(p);
    627 			(void) ether_disable_vlan_mtu(p);
    628 			IFNET_UNLOCK(p);
    629 		}
    630 
    631 		/* XXX ether_ifdetach must not be called with IFNET_LOCK */
    632 		mutex_exit(&ifv->ifv_lock);
    633 		IFNET_UNLOCK(ifp);
    634 		ether_ifdetach(ifp);
    635 		IFNET_LOCK(ifp);
    636 		mutex_enter(&ifv->ifv_lock);
    637 
    638 		/* if_free_sadl must be called with IFNET_LOCK */
    639 		if_free_sadl(ifp, 1);
    640 
    641 		/* Restore vlan_ioctl overwritten by ether_ifdetach */
    642 		ifp->if_ioctl = vlan_ioctl;
    643 		vlan_reset_linkname(ifp);
    644 		break;
    645 	    }
    646 
    647 	default:
    648 		panic("%s: impossible", __func__);
    649 	}
    650 
    651 	nmib->ifvm_p = NULL;
    652 	ifv->ifv_if.if_mtu = 0;
    653 	ifv->ifv_flags = 0;
    654 
    655 	mutex_enter(&ifv_hash.lock);
    656 	PSLIST_WRITER_REMOVE(ifv, ifv_hash);
    657 	pserialize_perform(vlan_psz);
    658 	mutex_exit(&ifv_hash.lock);
    659 	PSLIST_ENTRY_DESTROY(ifv, ifv_hash);
    660 
    661 	vlan_linkmib_update(ifv, nmib);
    662 
    663 	mutex_exit(&ifv->ifv_lock);
    664 
    665 	nmib_psref = NULL;
    666 	kmem_free(omib, sizeof(*omib));
    667 
    668 #ifdef INET6
    669 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
    670 	/* To delete v6 link local addresses */
    671 	if (in6_present)
    672 		in6_ifdetach(ifp);
    673 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    674 #endif
    675 
    676 	if ((ifp->if_flags & IFF_PROMISC) != 0)
    677 		vlan_safe_ifpromisc_locked(ifp, 0);
    678 	if_down_locked(ifp);
    679 	ifp->if_capabilities = 0;
    680 	mutex_enter(&ifv->ifv_lock);
    681 done:
    682 
    683 	if (nmib_psref)
    684 		psref_target_destroy(nmib_psref, ifvm_psref_class);
    685 
    686 	return error;
    687 }
    688 
    689 static void
    690 vlan_hash_init(void)
    691 {
    692 
    693 	ifv_hash.lists = hashinit(VLAN_TAG_HASH_SIZE, HASH_PSLIST, true,
    694 	    &ifv_hash.mask);
    695 }
    696 
    697 static int
    698 vlan_hash_fini(void)
    699 {
    700 	int i;
    701 
    702 	mutex_enter(&ifv_hash.lock);
    703 
    704 	for (i = 0; i < ifv_hash.mask + 1; i++) {
    705 		if (PSLIST_WRITER_FIRST(&ifv_hash.lists[i], struct ifvlan,
    706 		    ifv_hash) != NULL) {
    707 			mutex_exit(&ifv_hash.lock);
    708 			return EBUSY;
    709 		}
    710 	}
    711 
    712 	for (i = 0; i < ifv_hash.mask + 1; i++)
    713 		PSLIST_DESTROY(&ifv_hash.lists[i]);
    714 
    715 	mutex_exit(&ifv_hash.lock);
    716 
    717 	hashdone(ifv_hash.lists, HASH_PSLIST, ifv_hash.mask);
    718 
    719 	ifv_hash.lists = NULL;
    720 	ifv_hash.mask = 0;
    721 
    722 	return 0;
    723 }
    724 
    725 static int
    726 vlan_tag_hash(uint16_t tag, u_long mask)
    727 {
    728 	uint32_t hash;
    729 
    730 	hash = (tag >> 8) ^ tag;
    731 	hash = (hash >> 2) ^ hash;
    732 
    733 	return hash & mask;
    734 }
    735 
    736 static struct ifvlan_linkmib *
    737 vlan_getref_linkmib(struct ifvlan *sc, struct psref *psref)
    738 {
    739 	struct ifvlan_linkmib *mib;
    740 	int s;
    741 
    742 	s = pserialize_read_enter();
    743 	mib = sc->ifv_mib;
    744 	if (mib == NULL) {
    745 		pserialize_read_exit(s);
    746 		return NULL;
    747 	}
    748 	membar_datadep_consumer();
    749 	psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
    750 	pserialize_read_exit(s);
    751 
    752 	return mib;
    753 }
    754 
    755 static void
    756 vlan_putref_linkmib(struct ifvlan_linkmib *mib, struct psref *psref)
    757 {
    758 	if (mib == NULL)
    759 		return;
    760 	psref_release(psref, &mib->ifvm_psref, ifvm_psref_class);
    761 }
    762 
    763 static struct ifvlan_linkmib *
    764 vlan_lookup_tag_psref(struct ifnet *ifp, uint16_t tag, struct psref *psref)
    765 {
    766 	int idx;
    767 	int s;
    768 	struct ifvlan *sc;
    769 
    770 	idx = vlan_tag_hash(tag, ifv_hash.mask);
    771 
    772 	s = pserialize_read_enter();
    773 	PSLIST_READER_FOREACH(sc, &ifv_hash.lists[idx], struct ifvlan,
    774 	    ifv_hash) {
    775 		struct ifvlan_linkmib *mib = sc->ifv_mib;
    776 		if (mib == NULL)
    777 			continue;
    778 		if (mib->ifvm_tag != tag)
    779 			continue;
    780 		if (mib->ifvm_p != ifp)
    781 			continue;
    782 
    783 		psref_acquire(psref, &mib->ifvm_psref, ifvm_psref_class);
    784 		pserialize_read_exit(s);
    785 		return mib;
    786 	}
    787 	pserialize_read_exit(s);
    788 	return NULL;
    789 }
    790 
    791 static void
    792 vlan_linkmib_update(struct ifvlan *ifv, struct ifvlan_linkmib *nmib)
    793 {
    794 	struct ifvlan_linkmib *omib = ifv->ifv_mib;
    795 
    796 	KASSERT(mutex_owned(&ifv->ifv_lock));
    797 
    798 	membar_producer();
    799 	ifv->ifv_mib = nmib;
    800 
    801 	pserialize_perform(ifv->ifv_psz);
    802 	psref_target_destroy(&omib->ifvm_psref, ifvm_psref_class);
    803 }
    804 
    805 /*
    806  * Called when a parent interface is detaching; destroy any VLAN
    807  * configuration for the parent interface.
    808  */
    809 void
    810 vlan_ifdetach(struct ifnet *p)
    811 {
    812 	struct ifvlan *ifv;
    813 	struct ifvlan_linkmib *mib, **nmibs;
    814 	struct psref psref;
    815 	int error;
    816 	int bound;
    817 	int i, cnt = 0;
    818 
    819 	bound = curlwp_bind();
    820 
    821 	mutex_enter(&ifv_list.lock);
    822 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
    823 		mib = vlan_getref_linkmib(ifv, &psref);
    824 		if (mib == NULL)
    825 			continue;
    826 
    827 		if (mib->ifvm_p == p)
    828 			cnt++;
    829 
    830 		vlan_putref_linkmib(mib, &psref);
    831 	}
    832 	mutex_exit(&ifv_list.lock);
    833 
    834 	if (cnt == 0) {
    835 		curlwp_bindx(bound);
    836 		return;
    837 	}
    838 
    839 	/*
    840 	 * The value of "cnt" does not increase while ifv_list.lock
    841 	 * and ifv->ifv_lock are released here, because the parent
    842 	 * interface is detaching.
    843 	 */
    844 	nmibs = kmem_alloc(sizeof(*nmibs) * cnt, KM_SLEEP);
    845 	for (i = 0; i < cnt; i++) {
    846 		nmibs[i] = kmem_alloc(sizeof(*nmibs[i]), KM_SLEEP);
    847 	}
    848 
    849 	mutex_enter(&ifv_list.lock);
    850 
    851 	i = 0;
    852 	LIST_FOREACH(ifv, &ifv_list.list, ifv_list) {
    853 		struct ifnet *ifp = &ifv->ifv_if;
    854 
    855 		/* IFNET_LOCK must be held before ifv_lock. */
    856 		IFNET_LOCK(ifp);
    857 		mutex_enter(&ifv->ifv_lock);
    858 
    859 		/* XXX ifv_mib = NULL? */
    860 		if (ifv->ifv_mib->ifvm_p == p) {
    861 			KASSERTMSG(i < cnt, "no memory for unconfig, parent=%s",
    862 			    p->if_xname);
    863 			error = vlan_unconfig_locked(ifv, nmibs[i]);
    864 			if (!error) {
    865 				nmibs[i] = NULL;
    866 				i++;
    867 			}
    868 
    869 		}
    870 
    871 		mutex_exit(&ifv->ifv_lock);
    872 		IFNET_UNLOCK(ifp);
    873 	}
    874 
    875 	mutex_exit(&ifv_list.lock);
    876 
    877 	curlwp_bindx(bound);
    878 
    879 	for (i = 0; i < cnt; i++) {
    880 		if (nmibs[i])
    881 			kmem_free(nmibs[i], sizeof(*nmibs[i]));
    882 	}
    883 
    884 	kmem_free(nmibs, sizeof(*nmibs) * cnt);
    885 
    886 	return;
    887 }
    888 
    889 static int
    890 vlan_set_promisc(struct ifnet *ifp)
    891 {
    892 	struct ifvlan *ifv = ifp->if_softc;
    893 	struct ifvlan_linkmib *mib;
    894 	struct psref psref;
    895 	int error = 0;
    896 	int bound;
    897 
    898 	bound = curlwp_bind();
    899 	mib = vlan_getref_linkmib(ifv, &psref);
    900 	if (mib == NULL) {
    901 		curlwp_bindx(bound);
    902 		return EBUSY;
    903 	}
    904 
    905 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
    906 		if ((ifv->ifv_flags & IFVF_PROMISC) == 0) {
    907 			error = vlan_safe_ifpromisc(mib->ifvm_p, 1);
    908 			if (error == 0)
    909 				ifv->ifv_flags |= IFVF_PROMISC;
    910 		}
    911 	} else {
    912 		if ((ifv->ifv_flags & IFVF_PROMISC) != 0) {
    913 			error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
    914 			if (error == 0)
    915 				ifv->ifv_flags &= ~IFVF_PROMISC;
    916 		}
    917 	}
    918 	vlan_putref_linkmib(mib, &psref);
    919 	curlwp_bindx(bound);
    920 
    921 	return error;
    922 }
    923 
    924 static int
    925 vlan_ioctl(struct ifnet *ifp, u_long cmd, void *data)
    926 {
    927 	struct lwp *l = curlwp;
    928 	struct ifvlan *ifv = ifp->if_softc;
    929 	struct ifaddr *ifa = (struct ifaddr *) data;
    930 	struct ifreq *ifr = (struct ifreq *) data;
    931 	struct ifnet *pr;
    932 	struct ifcapreq *ifcr;
    933 	struct vlanreq vlr;
    934 	struct ifvlan_linkmib *mib;
    935 	struct psref psref;
    936 	int error = 0;
    937 	int bound;
    938 
    939 	switch (cmd) {
    940 	case SIOCSIFMTU:
    941 		bound = curlwp_bind();
    942 		mib = vlan_getref_linkmib(ifv, &psref);
    943 		if (mib == NULL) {
    944 			curlwp_bindx(bound);
    945 			error = EBUSY;
    946 			break;
    947 		}
    948 
    949 		if (mib->ifvm_p == NULL) {
    950 			vlan_putref_linkmib(mib, &psref);
    951 			curlwp_bindx(bound);
    952 			error = EINVAL;
    953 		} else if (
    954 		    ifr->ifr_mtu > (mib->ifvm_p->if_mtu - mib->ifvm_mtufudge) ||
    955 		    ifr->ifr_mtu < (mib->ifvm_mintu - mib->ifvm_mtufudge)) {
    956 			vlan_putref_linkmib(mib, &psref);
    957 			curlwp_bindx(bound);
    958 			error = EINVAL;
    959 		} else {
    960 			vlan_putref_linkmib(mib, &psref);
    961 			curlwp_bindx(bound);
    962 
    963 			error = ifioctl_common(ifp, cmd, data);
    964 			if (error == ENETRESET)
    965 				error = 0;
    966 		}
    967 
    968 		break;
    969 
    970 	case SIOCSETVLAN:
    971 		if ((error = kauth_authorize_network(l->l_cred,
    972 		    KAUTH_NETWORK_INTERFACE,
    973 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
    974 		    NULL)) != 0)
    975 			break;
    976 		if ((error = copyin(ifr->ifr_data, &vlr, sizeof(vlr))) != 0)
    977 			break;
    978 
    979 		if (vlr.vlr_parent[0] == '\0') {
    980 			bound = curlwp_bind();
    981 			mib = vlan_getref_linkmib(ifv, &psref);
    982 			if (mib == NULL) {
    983 				curlwp_bindx(bound);
    984 				error = EBUSY;
    985 				break;
    986 			}
    987 
    988 			if (mib->ifvm_p != NULL &&
    989 			    (ifp->if_flags & IFF_PROMISC) != 0)
    990 				error = vlan_safe_ifpromisc(mib->ifvm_p, 0);
    991 
    992 			vlan_putref_linkmib(mib, &psref);
    993 			curlwp_bindx(bound);
    994 
    995 			vlan_unconfig(ifp);
    996 			break;
    997 		}
    998 		if (vlr.vlr_tag != EVL_VLANOFTAG(vlr.vlr_tag)) {
    999 			error = EINVAL;		 /* check for valid tag */
   1000 			break;
   1001 		}
   1002 		if ((pr = ifunit(vlr.vlr_parent)) == NULL) {
   1003 			error = ENOENT;
   1004 			break;
   1005 		}
   1006 		error = vlan_config(ifv, pr, vlr.vlr_tag);
   1007 		if (error != 0) {
   1008 			break;
   1009 		}
   1010 
   1011 		/* Update promiscuous mode, if necessary. */
   1012 		vlan_set_promisc(ifp);
   1013 
   1014 		ifp->if_flags |= IFF_RUNNING;
   1015 		break;
   1016 
   1017 	case SIOCGETVLAN:
   1018 		memset(&vlr, 0, sizeof(vlr));
   1019 		bound = curlwp_bind();
   1020 		mib = vlan_getref_linkmib(ifv, &psref);
   1021 		if (mib == NULL) {
   1022 			curlwp_bindx(bound);
   1023 			error = EBUSY;
   1024 			break;
   1025 		}
   1026 		if (mib->ifvm_p != NULL) {
   1027 			snprintf(vlr.vlr_parent, sizeof(vlr.vlr_parent), "%s",
   1028 			    mib->ifvm_p->if_xname);
   1029 			vlr.vlr_tag = mib->ifvm_tag;
   1030 		}
   1031 		vlan_putref_linkmib(mib, &psref);
   1032 		curlwp_bindx(bound);
   1033 		error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
   1034 		break;
   1035 
   1036 	case SIOCSIFFLAGS:
   1037 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
   1038 			break;
   1039 		/*
   1040 		 * For promiscuous mode, we enable promiscuous mode on
   1041 		 * the parent if we need promiscuous on the VLAN interface.
   1042 		 */
   1043 		bound = curlwp_bind();
   1044 		mib = vlan_getref_linkmib(ifv, &psref);
   1045 		if (mib == NULL) {
   1046 			curlwp_bindx(bound);
   1047 			error = EBUSY;
   1048 			break;
   1049 		}
   1050 
   1051 		if (mib->ifvm_p != NULL)
   1052 			error = vlan_set_promisc(ifp);
   1053 		vlan_putref_linkmib(mib, &psref);
   1054 		curlwp_bindx(bound);
   1055 		break;
   1056 
   1057 	case SIOCADDMULTI:
   1058 		mutex_enter(&ifv->ifv_lock);
   1059 		mib = ifv->ifv_mib;
   1060 		if (mib == NULL) {
   1061 			error = EBUSY;
   1062 			mutex_exit(&ifv->ifv_lock);
   1063 			break;
   1064 		}
   1065 
   1066 		error = (mib->ifvm_p != NULL) ?
   1067 		    (*mib->ifvm_msw->vmsw_addmulti)(ifv, ifr) : EINVAL;
   1068 		mib = NULL;
   1069 		mutex_exit(&ifv->ifv_lock);
   1070 		break;
   1071 
   1072 	case SIOCDELMULTI:
   1073 		mutex_enter(&ifv->ifv_lock);
   1074 		mib = ifv->ifv_mib;
   1075 		if (mib == NULL) {
   1076 			error = EBUSY;
   1077 			mutex_exit(&ifv->ifv_lock);
   1078 			break;
   1079 		}
   1080 		error = (mib->ifvm_p != NULL) ?
   1081 		    (*mib->ifvm_msw->vmsw_delmulti)(ifv, ifr) : EINVAL;
   1082 		mib = NULL;
   1083 		mutex_exit(&ifv->ifv_lock);
   1084 		break;
   1085 
   1086 	case SIOCSIFCAP:
   1087 		ifcr = data;
   1088 		/* make sure caps are enabled on parent */
   1089 		bound = curlwp_bind();
   1090 		mib = vlan_getref_linkmib(ifv, &psref);
   1091 		if (mib == NULL) {
   1092 			curlwp_bindx(bound);
   1093 			error = EBUSY;
   1094 			break;
   1095 		}
   1096 
   1097 		if (mib->ifvm_p == NULL) {
   1098 			vlan_putref_linkmib(mib, &psref);
   1099 			curlwp_bindx(bound);
   1100 			error = EINVAL;
   1101 			break;
   1102 		}
   1103 		if ((mib->ifvm_p->if_capenable & ifcr->ifcr_capenable) !=
   1104 		    ifcr->ifcr_capenable) {
   1105 			vlan_putref_linkmib(mib, &psref);
   1106 			curlwp_bindx(bound);
   1107 			error = EINVAL;
   1108 			break;
   1109 		}
   1110 
   1111 		vlan_putref_linkmib(mib, &psref);
   1112 		curlwp_bindx(bound);
   1113 
   1114 		if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
   1115 			error = 0;
   1116 		break;
   1117 	case SIOCINITIFADDR:
   1118 		bound = curlwp_bind();
   1119 		mib = vlan_getref_linkmib(ifv, &psref);
   1120 		if (mib == NULL) {
   1121 			curlwp_bindx(bound);
   1122 			error = EBUSY;
   1123 			break;
   1124 		}
   1125 
   1126 		if (mib->ifvm_p == NULL) {
   1127 			error = EINVAL;
   1128 			vlan_putref_linkmib(mib, &psref);
   1129 			curlwp_bindx(bound);
   1130 			break;
   1131 		}
   1132 		vlan_putref_linkmib(mib, &psref);
   1133 		curlwp_bindx(bound);
   1134 
   1135 		ifp->if_flags |= IFF_UP;
   1136 #ifdef INET
   1137 		if (ifa->ifa_addr->sa_family == AF_INET)
   1138 			arp_ifinit(ifp, ifa);
   1139 #endif
   1140 		break;
   1141 
   1142 	default:
   1143 		error = ether_ioctl(ifp, cmd, data);
   1144 	}
   1145 
   1146 	return error;
   1147 }
   1148 
   1149 static int
   1150 vlan_ether_addmulti(struct ifvlan *ifv, struct ifreq *ifr)
   1151 {
   1152 	const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr);
   1153 	struct vlan_mc_entry *mc;
   1154 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
   1155 	struct ifvlan_linkmib *mib;
   1156 	int error;
   1157 
   1158 	KASSERT(mutex_owned(&ifv->ifv_lock));
   1159 
   1160 	if (sa->sa_len > sizeof(struct sockaddr_storage))
   1161 		return EINVAL;
   1162 
   1163 	error = ether_addmulti(sa, &ifv->ifv_ec);
   1164 	if (error != ENETRESET)
   1165 		return error;
   1166 
   1167 	/*
   1168 	 * This is a new multicast address.  We have to tell parent
   1169 	 * about it.  Also, remember this multicast address so that
   1170 	 * we can delete it on unconfigure.
   1171 	 */
   1172 	mc = malloc(sizeof(struct vlan_mc_entry), M_DEVBUF, M_NOWAIT);
   1173 	if (mc == NULL) {
   1174 		error = ENOMEM;
   1175 		goto alloc_failed;
   1176 	}
   1177 
   1178 	/*
   1179 	 * Since ether_addmulti() returned ENETRESET, the following two
   1180 	 * statements shouldn't fail. Here ifv_ec is implicitly protected
   1181 	 * by the ifv_lock lock.
   1182 	 */
   1183 	error = ether_multiaddr(sa, addrlo, addrhi);
   1184 	KASSERT(error == 0);
   1185 
   1186 	ETHER_LOCK(&ifv->ifv_ec);
   1187 	mc->mc_enm = ether_lookup_multi(addrlo, addrhi, &ifv->ifv_ec);
   1188 	ETHER_UNLOCK(&ifv->ifv_ec);
   1189 
   1190 	KASSERT(mc->mc_enm != NULL);
   1191 
   1192 	memcpy(&mc->mc_addr, sa, sa->sa_len);
   1193 	LIST_INSERT_HEAD(&ifv->ifv_mc_listhead, mc, mc_entries);
   1194 
   1195 	mib = ifv->ifv_mib;
   1196 
   1197 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
   1198 	IFNET_LOCK(mib->ifvm_p);
   1199 	error = if_mcast_op(mib->ifvm_p, SIOCADDMULTI, sa);
   1200 	IFNET_UNLOCK(mib->ifvm_p);
   1201 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(mib->ifvm_p);
   1202 
   1203 	if (error != 0)
   1204 		goto ioctl_failed;
   1205 	return error;
   1206 
   1207 ioctl_failed:
   1208 	LIST_REMOVE(mc, mc_entries);
   1209 	free(mc, M_DEVBUF);
   1210 
   1211 alloc_failed:
   1212 	(void)ether_delmulti(sa, &ifv->ifv_ec);
   1213 	return error;
   1214 }
   1215 
   1216 static int
   1217 vlan_ether_delmulti(struct ifvlan *ifv, struct ifreq *ifr)
   1218 {
   1219 	const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr);
   1220 	struct ether_multi *enm;
   1221 	struct vlan_mc_entry *mc;
   1222 	struct ifvlan_linkmib *mib;
   1223 	uint8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN];
   1224 	int error;
   1225 
   1226 	KASSERT(mutex_owned(&ifv->ifv_lock));
   1227 
   1228 	/*
   1229 	 * Find a key to lookup vlan_mc_entry.  We have to do this
   1230 	 * before calling ether_delmulti for obvious reasons.
   1231 	 */
   1232 	if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0)
   1233 		return error;
   1234 
   1235 	ETHER_LOCK(&ifv->ifv_ec);
   1236 	enm = ether_lookup_multi(addrlo, addrhi, &ifv->ifv_ec);
   1237 	ETHER_UNLOCK(&ifv->ifv_ec);
   1238 	if (enm == NULL)
   1239 		return EINVAL;
   1240 
   1241 	LIST_FOREACH(mc, &ifv->ifv_mc_listhead, mc_entries) {
   1242 		if (mc->mc_enm == enm)
   1243 			break;
   1244 	}
   1245 
   1246 	/* We woun't delete entries we didn't add */
   1247 	if (mc == NULL)
   1248 		return EINVAL;
   1249 
   1250 	error = ether_delmulti(sa, &ifv->ifv_ec);
   1251 	if (error != ENETRESET)
   1252 		return error;
   1253 
   1254 	/* We no longer use this multicast address.  Tell parent so. */
   1255 	mib = ifv->ifv_mib;
   1256 	IFNET_LOCK(mib->ifvm_p);
   1257 	error = if_mcast_op(mib->ifvm_p, SIOCDELMULTI, sa);
   1258 	IFNET_UNLOCK(mib->ifvm_p);
   1259 
   1260 	if (error == 0) {
   1261 		/* And forget about this address. */
   1262 		LIST_REMOVE(mc, mc_entries);
   1263 		free(mc, M_DEVBUF);
   1264 	} else {
   1265 		(void)ether_addmulti(sa, &ifv->ifv_ec);
   1266 	}
   1267 
   1268 	return error;
   1269 }
   1270 
   1271 /*
   1272  * Delete any multicast address we have asked to add from parent
   1273  * interface.  Called when the vlan is being unconfigured.
   1274  */
   1275 static void
   1276 vlan_ether_purgemulti(struct ifvlan *ifv)
   1277 {
   1278 	struct vlan_mc_entry *mc;
   1279 	struct ifvlan_linkmib *mib;
   1280 
   1281 	KASSERT(mutex_owned(&ifv->ifv_lock));
   1282 	mib = ifv->ifv_mib;
   1283 	if (mib == NULL) {
   1284 		return;
   1285 	}
   1286 
   1287 	while ((mc = LIST_FIRST(&ifv->ifv_mc_listhead)) != NULL) {
   1288 		IFNET_LOCK(mib->ifvm_p);
   1289 		(void)if_mcast_op(mib->ifvm_p, SIOCDELMULTI,
   1290 		    sstocsa(&mc->mc_addr));
   1291 		IFNET_UNLOCK(mib->ifvm_p);
   1292 		LIST_REMOVE(mc, mc_entries);
   1293 		free(mc, M_DEVBUF);
   1294 	}
   1295 }
   1296 
   1297 static void
   1298 vlan_start(struct ifnet *ifp)
   1299 {
   1300 	struct ifvlan *ifv = ifp->if_softc;
   1301 	struct ifnet *p;
   1302 	struct ethercom *ec;
   1303 	struct mbuf *m;
   1304 	struct ifvlan_linkmib *mib;
   1305 	struct psref psref;
   1306 	int error;
   1307 
   1308 	mib = vlan_getref_linkmib(ifv, &psref);
   1309 	if (mib == NULL)
   1310 		return;
   1311 	p = mib->ifvm_p;
   1312 	ec = (void *)mib->ifvm_p;
   1313 
   1314 	ifp->if_flags |= IFF_OACTIVE;
   1315 
   1316 	for (;;) {
   1317 		IFQ_DEQUEUE(&ifp->if_snd, m);
   1318 		if (m == NULL)
   1319 			break;
   1320 
   1321 #ifdef ALTQ
   1322 		/*
   1323 		 * KERNEL_LOCK is required for ALTQ even if NET_MPSAFE is
   1324 		 * defined.
   1325 		 */
   1326 		KERNEL_LOCK(1, NULL);
   1327 		/*
   1328 		 * If ALTQ is enabled on the parent interface, do
   1329 		 * classification; the queueing discipline might
   1330 		 * not require classification, but might require
   1331 		 * the address family/header pointer in the pktattr.
   1332 		 */
   1333 		if (ALTQ_IS_ENABLED(&p->if_snd)) {
   1334 			switch (p->if_type) {
   1335 			case IFT_ETHER:
   1336 				altq_etherclassify(&p->if_snd, m);
   1337 				break;
   1338 			default:
   1339 				panic("%s: impossible (altq)", __func__);
   1340 			}
   1341 		}
   1342 		KERNEL_UNLOCK_ONE(NULL);
   1343 #endif /* ALTQ */
   1344 
   1345 		bpf_mtap(ifp, m, BPF_D_OUT);
   1346 		/*
   1347 		 * If the parent can insert the tag itself, just mark
   1348 		 * the tag in the mbuf header.
   1349 		 */
   1350 		if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
   1351 			vlan_set_tag(m, mib->ifvm_tag);
   1352 		} else {
   1353 			/*
   1354 			 * insert the tag ourselves
   1355 			 */
   1356 			M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
   1357 			if (m == NULL) {
   1358 				printf("%s: unable to prepend encap header",
   1359 				    p->if_xname);
   1360 				ifp->if_oerrors++;
   1361 				continue;
   1362 			}
   1363 
   1364 			switch (p->if_type) {
   1365 			case IFT_ETHER:
   1366 			    {
   1367 				struct ether_vlan_header *evl;
   1368 
   1369 				if (m->m_len < sizeof(struct ether_vlan_header))
   1370 					m = m_pullup(m,
   1371 					    sizeof(struct ether_vlan_header));
   1372 				if (m == NULL) {
   1373 					printf("%s: unable to pullup encap "
   1374 					    "header", p->if_xname);
   1375 					ifp->if_oerrors++;
   1376 					continue;
   1377 				}
   1378 
   1379 				/*
   1380 				 * Transform the Ethernet header into an
   1381 				 * Ethernet header with 802.1Q encapsulation.
   1382 				 */
   1383 				memmove(mtod(m, void *),
   1384 				    mtod(m, char *) + mib->ifvm_encaplen,
   1385 				    sizeof(struct ether_header));
   1386 				evl = mtod(m, struct ether_vlan_header *);
   1387 				evl->evl_proto = evl->evl_encap_proto;
   1388 				evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
   1389 				evl->evl_tag = htons(mib->ifvm_tag);
   1390 
   1391 				/*
   1392 				 * To cater for VLAN-aware layer 2 ethernet
   1393 				 * switches which may need to strip the tag
   1394 				 * before forwarding the packet, make sure
   1395 				 * the packet+tag is at least 68 bytes long.
   1396 				 * This is necessary because our parent will
   1397 				 * only pad to 64 bytes (ETHER_MIN_LEN) and
   1398 				 * some switches will not pad by themselves
   1399 				 * after deleting a tag.
   1400 				 */
   1401 				const size_t min_data_len = ETHER_MIN_LEN -
   1402 				    ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
   1403 				if (m->m_pkthdr.len < min_data_len) {
   1404 					m_copyback(m, m->m_pkthdr.len,
   1405 					    min_data_len - m->m_pkthdr.len,
   1406 					    vlan_zero_pad_buff);
   1407 				}
   1408 				break;
   1409 			    }
   1410 
   1411 			default:
   1412 				panic("%s: impossible", __func__);
   1413 			}
   1414 		}
   1415 
   1416 		if ((p->if_flags & IFF_RUNNING) == 0) {
   1417 			m_freem(m);
   1418 			continue;
   1419 		}
   1420 
   1421 		error = if_transmit_lock(p, m);
   1422 		if (error) {
   1423 			/* mbuf is already freed */
   1424 			ifp->if_oerrors++;
   1425 			continue;
   1426 		}
   1427 		ifp->if_opackets++;
   1428 	}
   1429 
   1430 	ifp->if_flags &= ~IFF_OACTIVE;
   1431 
   1432 	/* Remove reference to mib before release */
   1433 	vlan_putref_linkmib(mib, &psref);
   1434 }
   1435 
   1436 static int
   1437 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
   1438 {
   1439 	struct ifvlan *ifv = ifp->if_softc;
   1440 	struct ifnet *p;
   1441 	struct ethercom *ec;
   1442 	struct ifvlan_linkmib *mib;
   1443 	struct psref psref;
   1444 	int error;
   1445 	size_t pktlen = m->m_pkthdr.len;
   1446 	bool mcast = (m->m_flags & M_MCAST) != 0;
   1447 
   1448 	mib = vlan_getref_linkmib(ifv, &psref);
   1449 	if (mib == NULL) {
   1450 		m_freem(m);
   1451 		return ENETDOWN;
   1452 	}
   1453 
   1454 	p = mib->ifvm_p;
   1455 	ec = (void *)mib->ifvm_p;
   1456 
   1457 	bpf_mtap(ifp, m, BPF_D_OUT);
   1458 
   1459 	if ((error = pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_OUT)) != 0)
   1460 		goto out;
   1461 	if (m == NULL)
   1462 		goto out;
   1463 
   1464 	/*
   1465 	 * If the parent can insert the tag itself, just mark
   1466 	 * the tag in the mbuf header.
   1467 	 */
   1468 	if (ec->ec_capabilities & ETHERCAP_VLAN_HWTAGGING) {
   1469 		vlan_set_tag(m, mib->ifvm_tag);
   1470 	} else {
   1471 		/*
   1472 		 * insert the tag ourselves
   1473 		 */
   1474 		M_PREPEND(m, mib->ifvm_encaplen, M_DONTWAIT);
   1475 		if (m == NULL) {
   1476 			printf("%s: unable to prepend encap header",
   1477 			    p->if_xname);
   1478 			ifp->if_oerrors++;
   1479 			error = ENOBUFS;
   1480 			goto out;
   1481 		}
   1482 
   1483 		switch (p->if_type) {
   1484 		case IFT_ETHER:
   1485 		    {
   1486 			struct ether_vlan_header *evl;
   1487 
   1488 			if (m->m_len < sizeof(struct ether_vlan_header))
   1489 				m = m_pullup(m,
   1490 				    sizeof(struct ether_vlan_header));
   1491 			if (m == NULL) {
   1492 				printf("%s: unable to pullup encap "
   1493 				    "header", p->if_xname);
   1494 				ifp->if_oerrors++;
   1495 				error = ENOBUFS;
   1496 				goto out;
   1497 			}
   1498 
   1499 			/*
   1500 			 * Transform the Ethernet header into an
   1501 			 * Ethernet header with 802.1Q encapsulation.
   1502 			 */
   1503 			memmove(mtod(m, void *),
   1504 			    mtod(m, char *) + mib->ifvm_encaplen,
   1505 			    sizeof(struct ether_header));
   1506 			evl = mtod(m, struct ether_vlan_header *);
   1507 			evl->evl_proto = evl->evl_encap_proto;
   1508 			evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
   1509 			evl->evl_tag = htons(mib->ifvm_tag);
   1510 
   1511 			/*
   1512 			 * To cater for VLAN-aware layer 2 ethernet
   1513 			 * switches which may need to strip the tag
   1514 			 * before forwarding the packet, make sure
   1515 			 * the packet+tag is at least 68 bytes long.
   1516 			 * This is necessary because our parent will
   1517 			 * only pad to 64 bytes (ETHER_MIN_LEN) and
   1518 			 * some switches will not pad by themselves
   1519 			 * after deleting a tag.
   1520 			 */
   1521 			const size_t min_data_len = ETHER_MIN_LEN -
   1522 			    ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
   1523 			if (m->m_pkthdr.len < min_data_len) {
   1524 				m_copyback(m, m->m_pkthdr.len,
   1525 				    min_data_len - m->m_pkthdr.len,
   1526 				    vlan_zero_pad_buff);
   1527 			}
   1528 			break;
   1529 		    }
   1530 
   1531 		default:
   1532 			panic("%s: impossible", __func__);
   1533 		}
   1534 	}
   1535 
   1536 	if ((p->if_flags & IFF_RUNNING) == 0) {
   1537 		m_freem(m);
   1538 		error = ENETDOWN;
   1539 		goto out;
   1540 	}
   1541 
   1542 	error = if_transmit_lock(p, m);
   1543 	if (error) {
   1544 		/* mbuf is already freed */
   1545 		ifp->if_oerrors++;
   1546 	} else {
   1547 
   1548 		ifp->if_opackets++;
   1549 		ifp->if_obytes += pktlen;
   1550 		if (mcast)
   1551 			ifp->if_omcasts++;
   1552 	}
   1553 
   1554 out:
   1555 	/* Remove reference to mib before release */
   1556 	vlan_putref_linkmib(mib, &psref);
   1557 	return error;
   1558 }
   1559 
   1560 /*
   1561  * Given an Ethernet frame, find a valid vlan interface corresponding to the
   1562  * given source interface and tag, then run the real packet through the
   1563  * parent's input routine.
   1564  */
   1565 void
   1566 vlan_input(struct ifnet *ifp, struct mbuf *m)
   1567 {
   1568 	struct ifvlan *ifv;
   1569 	uint16_t vid;
   1570 	struct ifvlan_linkmib *mib;
   1571 	struct psref psref;
   1572 	bool have_vtag;
   1573 
   1574 	have_vtag = vlan_has_tag(m);
   1575 	if (have_vtag) {
   1576 		vid = EVL_VLANOFTAG(vlan_get_tag(m));
   1577 		m->m_flags &= ~M_VLANTAG;
   1578 	} else {
   1579 		struct ether_vlan_header *evl;
   1580 
   1581 		if (ifp->if_type != IFT_ETHER) {
   1582 			panic("%s: impossible", __func__);
   1583 		}
   1584 
   1585 		if (m->m_len < sizeof(struct ether_vlan_header) &&
   1586 		    (m = m_pullup(m,
   1587 		     sizeof(struct ether_vlan_header))) == NULL) {
   1588 			printf("%s: no memory for VLAN header, "
   1589 			    "dropping packet.\n", ifp->if_xname);
   1590 			return;
   1591 		}
   1592 		evl = mtod(m, struct ether_vlan_header *);
   1593 		KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN);
   1594 
   1595 		vid = EVL_VLANOFTAG(ntohs(evl->evl_tag));
   1596 
   1597 		/*
   1598 		 * Restore the original ethertype.  We'll remove
   1599 		 * the encapsulation after we've found the vlan
   1600 		 * interface corresponding to the tag.
   1601 		 */
   1602 		evl->evl_encap_proto = evl->evl_proto;
   1603 	}
   1604 
   1605 	mib = vlan_lookup_tag_psref(ifp, vid, &psref);
   1606 	if (mib == NULL) {
   1607 		m_freem(m);
   1608 		ifp->if_noproto++;
   1609 		return;
   1610 	}
   1611 	KASSERT(mib->ifvm_encaplen == ETHER_VLAN_ENCAP_LEN);
   1612 
   1613 	ifv = mib->ifvm_ifvlan;
   1614 	if ((ifv->ifv_if.if_flags & (IFF_UP|IFF_RUNNING)) !=
   1615 	    (IFF_UP|IFF_RUNNING)) {
   1616 		m_freem(m);
   1617 		ifp->if_noproto++;
   1618 		goto out;
   1619 	}
   1620 
   1621 	/*
   1622 	 * Now, remove the encapsulation header.  The original
   1623 	 * header has already been fixed up above.
   1624 	 */
   1625 	if (!have_vtag) {
   1626 		memmove(mtod(m, char *) + mib->ifvm_encaplen,
   1627 		    mtod(m, void *), sizeof(struct ether_header));
   1628 		m_adj(m, mib->ifvm_encaplen);
   1629 	}
   1630 
   1631 	m_set_rcvif(m, &ifv->ifv_if);
   1632 	ifv->ifv_if.if_ipackets++;
   1633 
   1634 	if (pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_IN) != 0)
   1635 		goto out;
   1636 	if (m == NULL)
   1637 		goto out;
   1638 
   1639 	m->m_flags &= ~M_PROMISC;
   1640 	if_input(&ifv->ifv_if, m);
   1641 out:
   1642 	vlan_putref_linkmib(mib, &psref);
   1643 }
   1644 
   1645 /*
   1646  * Module infrastructure
   1647  */
   1648 #include "if_module.h"
   1649 
   1650 IF_MODULE(MODULE_CLASS_DRIVER, vlan, "")
   1651