xref: /freebsd/sys/net/if_vlan.c (revision 732a02b4e77866604a120a275c082bb6221bd2ff)
1 /*-
2  * Copyright 1998 Massachusetts Institute of Technology
3  * Copyright 2012 ADARA Networks, Inc.
4  * Copyright 2017 Dell EMC Isilon
5  *
6  * Portions of this software were developed by Robert N. M. Watson under
7  * contract to ADARA Networks, Inc.
8  *
9  * Permission to use, copy, modify, and distribute this software and
10  * its documentation for any purpose and without fee is hereby
11  * granted, provided that both the above copyright notice and this
12  * permission notice appear in all copies, that both the above
13  * copyright notice and this permission notice appear in all
14  * supporting documentation, and that the name of M.I.T. not be used
15  * in advertising or publicity pertaining to distribution of the
16  * software without specific, written prior permission.  M.I.T. makes
17  * no representations about the suitability of this software for any
18  * purpose.  It is provided "as is" without express or implied
19  * warranty.
20  *
21  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
22  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
23  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
24  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
25  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
28  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
29  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 /*
36  * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
37  * This is sort of sneaky in the implementation, since
38  * we need to pretend to be enough of an Ethernet implementation
39  * to make arp work.  The way we do this is by telling everyone
40  * that we are an Ethernet, and then catch the packets that
41  * ether_output() sends to us via if_transmit(), rewrite them for
42  * use by the real outgoing interface, and ask it to send them.
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_inet.h"
49 #include "opt_inet6.h"
50 #include "opt_kern_tls.h"
51 #include "opt_vlan.h"
52 #include "opt_ratelimit.h"
53 
54 #include <sys/param.h>
55 #include <sys/eventhandler.h>
56 #include <sys/kernel.h>
57 #include <sys/lock.h>
58 #include <sys/malloc.h>
59 #include <sys/mbuf.h>
60 #include <sys/module.h>
61 #include <sys/rmlock.h>
62 #include <sys/priv.h>
63 #include <sys/queue.h>
64 #include <sys/socket.h>
65 #include <sys/sockio.h>
66 #include <sys/sysctl.h>
67 #include <sys/systm.h>
68 #include <sys/sx.h>
69 #include <sys/taskqueue.h>
70 
71 #include <net/bpf.h>
72 #include <net/ethernet.h>
73 #include <net/if.h>
74 #include <net/if_var.h>
75 #include <net/if_clone.h>
76 #include <net/if_dl.h>
77 #include <net/if_types.h>
78 #include <net/if_vlan_var.h>
79 #include <net/route.h>
80 #include <net/vnet.h>
81 
82 #ifdef INET
83 #include <netinet/in.h>
84 #include <netinet/if_ether.h>
85 #endif
86 
87 #ifdef INET6
88 /*
89  * XXX: declare here to avoid to include many inet6 related files..
90  * should be more generalized?
91  */
92 extern void	nd6_setmtu(struct ifnet *);
93 #endif
94 
95 #define	VLAN_DEF_HWIDTH	4
96 #define	VLAN_IFFLAGS	(IFF_BROADCAST | IFF_MULTICAST)
97 
98 #define	UP_AND_RUNNING(ifp) \
99     ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
100 
101 CK_SLIST_HEAD(ifvlanhead, ifvlan);
102 
103 struct ifvlantrunk {
104 	struct	ifnet   *parent;	/* parent interface of this trunk */
105 	struct	mtx	lock;
106 #ifdef VLAN_ARRAY
107 #define	VLAN_ARRAY_SIZE	(EVL_VLID_MASK + 1)
108 	struct	ifvlan	*vlans[VLAN_ARRAY_SIZE]; /* static table */
109 #else
110 	struct	ifvlanhead *hash;	/* dynamic hash-list table */
111 	uint16_t	hmask;
112 	uint16_t	hwidth;
113 #endif
114 	int		refcnt;
115 };
116 
117 #if defined(KERN_TLS) || defined(RATELIMIT)
118 struct vlan_snd_tag {
119 	struct m_snd_tag com;
120 	struct m_snd_tag *tag;
121 };
122 
123 static inline struct vlan_snd_tag *
124 mst_to_vst(struct m_snd_tag *mst)
125 {
126 
127 	return (__containerof(mst, struct vlan_snd_tag, com));
128 }
129 #endif
130 
131 /*
132  * This macro provides a facility to iterate over every vlan on a trunk with
133  * the assumption that none will be added/removed during iteration.
134  */
135 #ifdef VLAN_ARRAY
136 #define VLAN_FOREACH(_ifv, _trunk) \
137 	size_t _i; \
138 	for (_i = 0; _i < VLAN_ARRAY_SIZE; _i++) \
139 		if (((_ifv) = (_trunk)->vlans[_i]) != NULL)
140 #else /* VLAN_ARRAY */
141 #define VLAN_FOREACH(_ifv, _trunk) \
142 	struct ifvlan *_next; \
143 	size_t _i; \
144 	for (_i = 0; _i < (1 << (_trunk)->hwidth); _i++) \
145 		CK_SLIST_FOREACH_SAFE((_ifv), &(_trunk)->hash[_i], ifv_list, _next)
146 #endif /* VLAN_ARRAY */
147 
148 /*
149  * This macro provides a facility to iterate over every vlan on a trunk while
150  * also modifying the number of vlans on the trunk. The iteration continues
151  * until some condition is met or there are no more vlans on the trunk.
152  */
153 #ifdef VLAN_ARRAY
154 /* The VLAN_ARRAY case is simple -- just a for loop using the condition. */
155 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
156 	size_t _i; \
157 	for (_i = 0; !(_cond) && _i < VLAN_ARRAY_SIZE; _i++) \
158 		if (((_ifv) = (_trunk)->vlans[_i]))
159 #else /* VLAN_ARRAY */
160 /*
161  * The hash table case is more complicated. We allow for the hash table to be
162  * modified (i.e. vlans removed) while we are iterating over it. To allow for
163  * this we must restart the iteration every time we "touch" something during
164  * the iteration, since removal will resize the hash table and invalidate our
165  * current position. If acting on the touched element causes the trunk to be
166  * emptied, then iteration also stops.
167  */
168 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
169 	size_t _i; \
170 	bool _touch = false; \
171 	for (_i = 0; \
172 	    !(_cond) && _i < (1 << (_trunk)->hwidth); \
173 	    _i = (_touch && ((_trunk) != NULL) ? 0 : _i + 1), _touch = false) \
174 		if (((_ifv) = CK_SLIST_FIRST(&(_trunk)->hash[_i])) != NULL && \
175 		    (_touch = true))
176 #endif /* VLAN_ARRAY */
177 
178 struct vlan_mc_entry {
179 	struct sockaddr_dl		mc_addr;
180 	CK_SLIST_ENTRY(vlan_mc_entry)	mc_entries;
181 	struct epoch_context		mc_epoch_ctx;
182 };
183 
184 struct ifvlan {
185 	struct	ifvlantrunk *ifv_trunk;
186 	struct	ifnet *ifv_ifp;
187 #define	TRUNK(ifv)	((ifv)->ifv_trunk)
188 #define	PARENT(ifv)	((ifv)->ifv_trunk->parent)
189 	void	*ifv_cookie;
190 	int	ifv_pflags;	/* special flags we have set on parent */
191 	int	ifv_capenable;
192 	int	ifv_encaplen;	/* encapsulation length */
193 	int	ifv_mtufudge;	/* MTU fudged by this much */
194 	int	ifv_mintu;	/* min transmission unit */
195 	uint16_t ifv_proto;	/* encapsulation ethertype */
196 	uint16_t ifv_tag;	/* tag to apply on packets leaving if */
197 	uint16_t ifv_vid;	/* VLAN ID */
198 	uint8_t	ifv_pcp;	/* Priority Code Point (PCP). */
199 	struct task lladdr_task;
200 	CK_SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
201 #ifndef VLAN_ARRAY
202 	CK_SLIST_ENTRY(ifvlan) ifv_list;
203 #endif
204 };
205 
206 /* Special flags we should propagate to parent. */
207 static struct {
208 	int flag;
209 	int (*func)(struct ifnet *, int);
210 } vlan_pflags[] = {
211 	{IFF_PROMISC, ifpromisc},
212 	{IFF_ALLMULTI, if_allmulti},
213 	{0, NULL}
214 };
215 
216 extern int vlan_mtag_pcp;
217 
218 static const char vlanname[] = "vlan";
219 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
220 
221 static eventhandler_tag ifdetach_tag;
222 static eventhandler_tag iflladdr_tag;
223 
224 /*
225  * if_vlan uses two module-level synchronizations primitives to allow concurrent
226  * modification of vlan interfaces and (mostly) allow for vlans to be destroyed
227  * while they are being used for tx/rx. To accomplish this in a way that has
228  * acceptable performance and cooperation with other parts of the network stack
229  * there is a non-sleepable epoch(9) and an sx(9).
230  *
231  * The performance-sensitive paths that warrant using the epoch(9) are
232  * vlan_transmit and vlan_input. Both have to check for the vlan interface's
233  * existence using if_vlantrunk, and being in the network tx/rx paths the use
234  * of an epoch(9) gives a measureable improvement in performance.
235  *
236  * The reason for having an sx(9) is mostly because there are still areas that
237  * must be sleepable and also have safe concurrent access to a vlan interface.
238  * Since the sx(9) exists, it is used by default in most paths unless sleeping
239  * is not permitted, or if it is not clear whether sleeping is permitted.
240  *
241  */
242 #define _VLAN_SX_ID ifv_sx
243 
244 static struct sx _VLAN_SX_ID;
245 
246 #define VLAN_LOCKING_INIT() \
247 	sx_init(&_VLAN_SX_ID, "vlan_sx")
248 
249 #define VLAN_LOCKING_DESTROY() \
250 	sx_destroy(&_VLAN_SX_ID)
251 
252 #define	VLAN_SLOCK()			sx_slock(&_VLAN_SX_ID)
253 #define	VLAN_SUNLOCK()			sx_sunlock(&_VLAN_SX_ID)
254 #define	VLAN_XLOCK()			sx_xlock(&_VLAN_SX_ID)
255 #define	VLAN_XUNLOCK()			sx_xunlock(&_VLAN_SX_ID)
256 #define	VLAN_SLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_SLOCKED)
257 #define	VLAN_XLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_XLOCKED)
258 #define	VLAN_SXLOCK_ASSERT()		sx_assert(&_VLAN_SX_ID, SA_LOCKED)
259 
260 /*
261  * We also have a per-trunk mutex that should be acquired when changing
262  * its state.
263  */
264 #define	TRUNK_LOCK_INIT(trunk)		mtx_init(&(trunk)->lock, vlanname, NULL, MTX_DEF)
265 #define	TRUNK_LOCK_DESTROY(trunk)	mtx_destroy(&(trunk)->lock)
266 #define	TRUNK_WLOCK(trunk)		mtx_lock(&(trunk)->lock)
267 #define	TRUNK_WUNLOCK(trunk)		mtx_unlock(&(trunk)->lock)
268 #define	TRUNK_WLOCK_ASSERT(trunk)	mtx_assert(&(trunk)->lock, MA_OWNED);
269 
270 /*
271  * The VLAN_ARRAY substitutes the dynamic hash with a static array
272  * with 4096 entries. In theory this can give a boost in processing,
273  * however in practice it does not. Probably this is because the array
274  * is too big to fit into CPU cache.
275  */
276 #ifndef VLAN_ARRAY
277 static	void vlan_inithash(struct ifvlantrunk *trunk);
278 static	void vlan_freehash(struct ifvlantrunk *trunk);
279 static	int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
280 static	int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
281 static	void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
282 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
283 	uint16_t vid);
284 #endif
285 static	void trunk_destroy(struct ifvlantrunk *trunk);
286 
287 static	void vlan_init(void *foo);
288 static	void vlan_input(struct ifnet *ifp, struct mbuf *m);
289 static	int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
290 #if defined(KERN_TLS) || defined(RATELIMIT)
291 static	int vlan_snd_tag_alloc(struct ifnet *,
292     union if_snd_tag_alloc_params *, struct m_snd_tag **);
293 static	int vlan_snd_tag_modify(struct m_snd_tag *,
294     union if_snd_tag_modify_params *);
295 static	int vlan_snd_tag_query(struct m_snd_tag *,
296     union if_snd_tag_query_params *);
297 static	void vlan_snd_tag_free(struct m_snd_tag *);
298 #endif
299 static	void vlan_qflush(struct ifnet *ifp);
300 static	int vlan_setflag(struct ifnet *ifp, int flag, int status,
301     int (*func)(struct ifnet *, int));
302 static	int vlan_setflags(struct ifnet *ifp, int status);
303 static	int vlan_setmulti(struct ifnet *ifp);
304 static	int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
305 static	int vlan_output(struct ifnet *ifp, struct mbuf *m,
306     const struct sockaddr *dst, struct route *ro);
307 static	void vlan_unconfig(struct ifnet *ifp);
308 static	void vlan_unconfig_locked(struct ifnet *ifp, int departing);
309 static	int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
310 static	void vlan_link_state(struct ifnet *ifp);
311 static	void vlan_capabilities(struct ifvlan *ifv);
312 static	void vlan_trunk_capabilities(struct ifnet *ifp);
313 
314 static	struct ifnet *vlan_clone_match_ethervid(const char *, int *);
315 static	int vlan_clone_match(struct if_clone *, const char *);
316 static	int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
317 static	int vlan_clone_destroy(struct if_clone *, struct ifnet *);
318 
319 static	void vlan_ifdetach(void *arg, struct ifnet *ifp);
320 static  void vlan_iflladdr(void *arg, struct ifnet *ifp);
321 
322 static  void vlan_lladdr_fn(void *arg, int pending);
323 
324 static struct if_clone *vlan_cloner;
325 
326 #ifdef VIMAGE
327 VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner);
328 #define	V_vlan_cloner	VNET(vlan_cloner)
329 #endif
330 
331 static void
332 vlan_mc_free(struct epoch_context *ctx)
333 {
334 	struct vlan_mc_entry *mc = __containerof(ctx, struct vlan_mc_entry, mc_epoch_ctx);
335 	free(mc, M_VLAN);
336 }
337 
338 #ifndef VLAN_ARRAY
339 #define HASH(n, m)	((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
340 
341 static void
342 vlan_inithash(struct ifvlantrunk *trunk)
343 {
344 	int i, n;
345 
346 	/*
347 	 * The trunk must not be locked here since we call malloc(M_WAITOK).
348 	 * It is OK in case this function is called before the trunk struct
349 	 * gets hooked up and becomes visible from other threads.
350 	 */
351 
352 	KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
353 	    ("%s: hash already initialized", __func__));
354 
355 	trunk->hwidth = VLAN_DEF_HWIDTH;
356 	n = 1 << trunk->hwidth;
357 	trunk->hmask = n - 1;
358 	trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
359 	for (i = 0; i < n; i++)
360 		CK_SLIST_INIT(&trunk->hash[i]);
361 }
362 
363 static void
364 vlan_freehash(struct ifvlantrunk *trunk)
365 {
366 #ifdef INVARIANTS
367 	int i;
368 
369 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
370 	for (i = 0; i < (1 << trunk->hwidth); i++)
371 		KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]),
372 		    ("%s: hash table not empty", __func__));
373 #endif
374 	free(trunk->hash, M_VLAN);
375 	trunk->hash = NULL;
376 	trunk->hwidth = trunk->hmask = 0;
377 }
378 
379 static int
380 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
381 {
382 	int i, b;
383 	struct ifvlan *ifv2;
384 
385 	VLAN_XLOCK_ASSERT();
386 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
387 
388 	b = 1 << trunk->hwidth;
389 	i = HASH(ifv->ifv_vid, trunk->hmask);
390 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
391 		if (ifv->ifv_vid == ifv2->ifv_vid)
392 			return (EEXIST);
393 
394 	/*
395 	 * Grow the hash when the number of vlans exceeds half of the number of
396 	 * hash buckets squared. This will make the average linked-list length
397 	 * buckets/2.
398 	 */
399 	if (trunk->refcnt > (b * b) / 2) {
400 		vlan_growhash(trunk, 1);
401 		i = HASH(ifv->ifv_vid, trunk->hmask);
402 	}
403 	CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
404 	trunk->refcnt++;
405 
406 	return (0);
407 }
408 
409 static int
410 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
411 {
412 	int i, b;
413 	struct ifvlan *ifv2;
414 
415 	VLAN_XLOCK_ASSERT();
416 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
417 
418 	b = 1 << trunk->hwidth;
419 	i = HASH(ifv->ifv_vid, trunk->hmask);
420 	CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
421 		if (ifv2 == ifv) {
422 			trunk->refcnt--;
423 			CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list);
424 			if (trunk->refcnt < (b * b) / 2)
425 				vlan_growhash(trunk, -1);
426 			return (0);
427 		}
428 
429 	panic("%s: vlan not found\n", __func__);
430 	return (ENOENT); /*NOTREACHED*/
431 }
432 
433 /*
434  * Grow the hash larger or smaller if memory permits.
435  */
436 static void
437 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
438 {
439 	struct ifvlan *ifv;
440 	struct ifvlanhead *hash2;
441 	int hwidth2, i, j, n, n2;
442 
443 	VLAN_XLOCK_ASSERT();
444 	KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
445 
446 	if (howmuch == 0) {
447 		/* Harmless yet obvious coding error */
448 		printf("%s: howmuch is 0\n", __func__);
449 		return;
450 	}
451 
452 	hwidth2 = trunk->hwidth + howmuch;
453 	n = 1 << trunk->hwidth;
454 	n2 = 1 << hwidth2;
455 	/* Do not shrink the table below the default */
456 	if (hwidth2 < VLAN_DEF_HWIDTH)
457 		return;
458 
459 	hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK);
460 	if (hash2 == NULL) {
461 		printf("%s: out of memory -- hash size not changed\n",
462 		    __func__);
463 		return;		/* We can live with the old hash table */
464 	}
465 	for (j = 0; j < n2; j++)
466 		CK_SLIST_INIT(&hash2[j]);
467 	for (i = 0; i < n; i++)
468 		while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) {
469 			CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list);
470 			j = HASH(ifv->ifv_vid, n2 - 1);
471 			CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
472 		}
473 	NET_EPOCH_WAIT();
474 	free(trunk->hash, M_VLAN);
475 	trunk->hash = hash2;
476 	trunk->hwidth = hwidth2;
477 	trunk->hmask = n2 - 1;
478 
479 	if (bootverbose)
480 		if_printf(trunk->parent,
481 		    "VLAN hash table resized from %d to %d buckets\n", n, n2);
482 }
483 
484 static __inline struct ifvlan *
485 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
486 {
487 	struct ifvlan *ifv;
488 
489 	NET_EPOCH_ASSERT();
490 
491 	CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
492 		if (ifv->ifv_vid == vid)
493 			return (ifv);
494 	return (NULL);
495 }
496 
497 #if 0
498 /* Debugging code to view the hashtables. */
499 static void
500 vlan_dumphash(struct ifvlantrunk *trunk)
501 {
502 	int i;
503 	struct ifvlan *ifv;
504 
505 	for (i = 0; i < (1 << trunk->hwidth); i++) {
506 		printf("%d: ", i);
507 		CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
508 			printf("%s ", ifv->ifv_ifp->if_xname);
509 		printf("\n");
510 	}
511 }
512 #endif /* 0 */
513 #else
514 
515 static __inline struct ifvlan *
516 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
517 {
518 
519 	return trunk->vlans[vid];
520 }
521 
522 static __inline int
523 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
524 {
525 
526 	if (trunk->vlans[ifv->ifv_vid] != NULL)
527 		return EEXIST;
528 	trunk->vlans[ifv->ifv_vid] = ifv;
529 	trunk->refcnt++;
530 
531 	return (0);
532 }
533 
534 static __inline int
535 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
536 {
537 
538 	trunk->vlans[ifv->ifv_vid] = NULL;
539 	trunk->refcnt--;
540 
541 	return (0);
542 }
543 
544 static __inline void
545 vlan_freehash(struct ifvlantrunk *trunk)
546 {
547 }
548 
549 static __inline void
550 vlan_inithash(struct ifvlantrunk *trunk)
551 {
552 }
553 
554 #endif /* !VLAN_ARRAY */
555 
556 static void
557 trunk_destroy(struct ifvlantrunk *trunk)
558 {
559 	VLAN_XLOCK_ASSERT();
560 
561 	vlan_freehash(trunk);
562 	trunk->parent->if_vlantrunk = NULL;
563 	TRUNK_LOCK_DESTROY(trunk);
564 	if_rele(trunk->parent);
565 	free(trunk, M_VLAN);
566 }
567 
568 /*
569  * Program our multicast filter. What we're actually doing is
570  * programming the multicast filter of the parent. This has the
571  * side effect of causing the parent interface to receive multicast
572  * traffic that it doesn't really want, which ends up being discarded
573  * later by the upper protocol layers. Unfortunately, there's no way
574  * to avoid this: there really is only one physical interface.
575  */
576 static int
577 vlan_setmulti(struct ifnet *ifp)
578 {
579 	struct ifnet		*ifp_p;
580 	struct ifmultiaddr	*ifma;
581 	struct ifvlan		*sc;
582 	struct vlan_mc_entry	*mc;
583 	int			error;
584 
585 	VLAN_XLOCK_ASSERT();
586 
587 	/* Find the parent. */
588 	sc = ifp->if_softc;
589 	ifp_p = PARENT(sc);
590 
591 	CURVNET_SET_QUIET(ifp_p->if_vnet);
592 
593 	/* First, remove any existing filter entries. */
594 	while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
595 		CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
596 		(void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
597 		NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
598 	}
599 
600 	/* Now program new ones. */
601 	IF_ADDR_WLOCK(ifp);
602 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
603 		if (ifma->ifma_addr->sa_family != AF_LINK)
604 			continue;
605 		mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
606 		if (mc == NULL) {
607 			IF_ADDR_WUNLOCK(ifp);
608 			return (ENOMEM);
609 		}
610 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
611 		mc->mc_addr.sdl_index = ifp_p->if_index;
612 		CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
613 	}
614 	IF_ADDR_WUNLOCK(ifp);
615 	CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
616 		error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
617 		    NULL);
618 		if (error)
619 			return (error);
620 	}
621 
622 	CURVNET_RESTORE();
623 	return (0);
624 }
625 
626 /*
627  * A handler for parent interface link layer address changes.
628  * If the parent interface link layer address is changed we
629  * should also change it on all children vlans.
630  */
631 static void
632 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
633 {
634 	struct epoch_tracker et;
635 	struct ifvlan *ifv;
636 	struct ifnet *ifv_ifp;
637 	struct ifvlantrunk *trunk;
638 	struct sockaddr_dl *sdl;
639 
640 	/* Need the epoch since this is run on taskqueue_swi. */
641 	NET_EPOCH_ENTER(et);
642 	trunk = ifp->if_vlantrunk;
643 	if (trunk == NULL) {
644 		NET_EPOCH_EXIT(et);
645 		return;
646 	}
647 
648 	/*
649 	 * OK, it's a trunk.  Loop over and change all vlan's lladdrs on it.
650 	 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
651 	 * ioctl calls on the parent garbling the lladdr of the child vlan.
652 	 */
653 	TRUNK_WLOCK(trunk);
654 	VLAN_FOREACH(ifv, trunk) {
655 		/*
656 		 * Copy new new lladdr into the ifv_ifp, enqueue a task
657 		 * to actually call if_setlladdr. if_setlladdr needs to
658 		 * be deferred to a taskqueue because it will call into
659 		 * the if_vlan ioctl path and try to acquire the global
660 		 * lock.
661 		 */
662 		ifv_ifp = ifv->ifv_ifp;
663 		bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
664 		    ifp->if_addrlen);
665 		sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
666 		sdl->sdl_alen = ifp->if_addrlen;
667 		taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
668 	}
669 	TRUNK_WUNLOCK(trunk);
670 	NET_EPOCH_EXIT(et);
671 }
672 
673 /*
674  * A handler for network interface departure events.
675  * Track departure of trunks here so that we don't access invalid
676  * pointers or whatever if a trunk is ripped from under us, e.g.,
677  * by ejecting its hot-plug card.  However, if an ifnet is simply
678  * being renamed, then there's no need to tear down the state.
679  */
680 static void
681 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
682 {
683 	struct ifvlan *ifv;
684 	struct ifvlantrunk *trunk;
685 
686 	/* If the ifnet is just being renamed, don't do anything. */
687 	if (ifp->if_flags & IFF_RENAMING)
688 		return;
689 	VLAN_XLOCK();
690 	trunk = ifp->if_vlantrunk;
691 	if (trunk == NULL) {
692 		VLAN_XUNLOCK();
693 		return;
694 	}
695 
696 	/*
697 	 * OK, it's a trunk.  Loop over and detach all vlan's on it.
698 	 * Check trunk pointer after each vlan_unconfig() as it will
699 	 * free it and set to NULL after the last vlan was detached.
700 	 */
701 	VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
702 	    ifp->if_vlantrunk == NULL)
703 		vlan_unconfig_locked(ifv->ifv_ifp, 1);
704 
705 	/* Trunk should have been destroyed in vlan_unconfig(). */
706 	KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
707 	VLAN_XUNLOCK();
708 }
709 
710 /*
711  * Return the trunk device for a virtual interface.
712  */
713 static struct ifnet  *
714 vlan_trunkdev(struct ifnet *ifp)
715 {
716 	struct ifvlan *ifv;
717 
718 	NET_EPOCH_ASSERT();
719 
720 	if (ifp->if_type != IFT_L2VLAN)
721 		return (NULL);
722 
723 	ifv = ifp->if_softc;
724 	ifp = NULL;
725 	if (ifv->ifv_trunk)
726 		ifp = PARENT(ifv);
727 	return (ifp);
728 }
729 
730 /*
731  * Return the 12-bit VLAN VID for this interface, for use by external
732  * components such as Infiniband.
733  *
734  * XXXRW: Note that the function name here is historical; it should be named
735  * vlan_vid().
736  */
737 static int
738 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
739 {
740 	struct ifvlan *ifv;
741 
742 	if (ifp->if_type != IFT_L2VLAN)
743 		return (EINVAL);
744 	ifv = ifp->if_softc;
745 	*vidp = ifv->ifv_vid;
746 	return (0);
747 }
748 
749 static int
750 vlan_pcp(struct ifnet *ifp, uint16_t *pcpp)
751 {
752 	struct ifvlan *ifv;
753 
754 	if (ifp->if_type != IFT_L2VLAN)
755 		return (EINVAL);
756 	ifv = ifp->if_softc;
757 	*pcpp = ifv->ifv_pcp;
758 	return (0);
759 }
760 
761 /*
762  * Return a driver specific cookie for this interface.  Synchronization
763  * with setcookie must be provided by the driver.
764  */
765 static void *
766 vlan_cookie(struct ifnet *ifp)
767 {
768 	struct ifvlan *ifv;
769 
770 	if (ifp->if_type != IFT_L2VLAN)
771 		return (NULL);
772 	ifv = ifp->if_softc;
773 	return (ifv->ifv_cookie);
774 }
775 
776 /*
777  * Store a cookie in our softc that drivers can use to store driver
778  * private per-instance data in.
779  */
780 static int
781 vlan_setcookie(struct ifnet *ifp, void *cookie)
782 {
783 	struct ifvlan *ifv;
784 
785 	if (ifp->if_type != IFT_L2VLAN)
786 		return (EINVAL);
787 	ifv = ifp->if_softc;
788 	ifv->ifv_cookie = cookie;
789 	return (0);
790 }
791 
792 /*
793  * Return the vlan device present at the specific VID.
794  */
795 static struct ifnet *
796 vlan_devat(struct ifnet *ifp, uint16_t vid)
797 {
798 	struct ifvlantrunk *trunk;
799 	struct ifvlan *ifv;
800 
801 	NET_EPOCH_ASSERT();
802 
803 	trunk = ifp->if_vlantrunk;
804 	if (trunk == NULL)
805 		return (NULL);
806 	ifp = NULL;
807 	ifv = vlan_gethash(trunk, vid);
808 	if (ifv)
809 		ifp = ifv->ifv_ifp;
810 	return (ifp);
811 }
812 
813 /*
814  * Recalculate the cached VLAN tag exposed via the MIB.
815  */
816 static void
817 vlan_tag_recalculate(struct ifvlan *ifv)
818 {
819 
820        ifv->ifv_tag = EVL_MAKETAG(ifv->ifv_vid, ifv->ifv_pcp, 0);
821 }
822 
823 /*
824  * VLAN support can be loaded as a module.  The only place in the
825  * system that's intimately aware of this is ether_input.  We hook
826  * into this code through vlan_input_p which is defined there and
827  * set here.  No one else in the system should be aware of this so
828  * we use an explicit reference here.
829  */
830 extern	void (*vlan_input_p)(struct ifnet *, struct mbuf *);
831 
832 /* For if_link_state_change() eyes only... */
833 extern	void (*vlan_link_state_p)(struct ifnet *);
834 
835 static int
836 vlan_modevent(module_t mod, int type, void *data)
837 {
838 
839 	switch (type) {
840 	case MOD_LOAD:
841 		ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
842 		    vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
843 		if (ifdetach_tag == NULL)
844 			return (ENOMEM);
845 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
846 		    vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
847 		if (iflladdr_tag == NULL)
848 			return (ENOMEM);
849 		VLAN_LOCKING_INIT();
850 		vlan_input_p = vlan_input;
851 		vlan_link_state_p = vlan_link_state;
852 		vlan_trunk_cap_p = vlan_trunk_capabilities;
853 		vlan_trunkdev_p = vlan_trunkdev;
854 		vlan_cookie_p = vlan_cookie;
855 		vlan_setcookie_p = vlan_setcookie;
856 		vlan_tag_p = vlan_tag;
857 		vlan_pcp_p = vlan_pcp;
858 		vlan_devat_p = vlan_devat;
859 #ifndef VIMAGE
860 		vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
861 		    vlan_clone_create, vlan_clone_destroy);
862 #endif
863 		if (bootverbose)
864 			printf("vlan: initialized, using "
865 #ifdef VLAN_ARRAY
866 			       "full-size arrays"
867 #else
868 			       "hash tables with chaining"
869 #endif
870 
871 			       "\n");
872 		break;
873 	case MOD_UNLOAD:
874 #ifndef VIMAGE
875 		if_clone_detach(vlan_cloner);
876 #endif
877 		EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
878 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
879 		vlan_input_p = NULL;
880 		vlan_link_state_p = NULL;
881 		vlan_trunk_cap_p = NULL;
882 		vlan_trunkdev_p = NULL;
883 		vlan_tag_p = NULL;
884 		vlan_cookie_p = NULL;
885 		vlan_setcookie_p = NULL;
886 		vlan_devat_p = NULL;
887 		VLAN_LOCKING_DESTROY();
888 		if (bootverbose)
889 			printf("vlan: unloaded\n");
890 		break;
891 	default:
892 		return (EOPNOTSUPP);
893 	}
894 	return (0);
895 }
896 
897 static moduledata_t vlan_mod = {
898 	"if_vlan",
899 	vlan_modevent,
900 	0
901 };
902 
903 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
904 MODULE_VERSION(if_vlan, 3);
905 
906 #ifdef VIMAGE
907 static void
908 vnet_vlan_init(const void *unused __unused)
909 {
910 
911 	vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
912 		    vlan_clone_create, vlan_clone_destroy);
913 	V_vlan_cloner = vlan_cloner;
914 }
915 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
916     vnet_vlan_init, NULL);
917 
918 static void
919 vnet_vlan_uninit(const void *unused __unused)
920 {
921 
922 	if_clone_detach(V_vlan_cloner);
923 }
924 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
925     vnet_vlan_uninit, NULL);
926 #endif
927 
928 /*
929  * Check for <etherif>.<vlan> style interface names.
930  */
931 static struct ifnet *
932 vlan_clone_match_ethervid(const char *name, int *vidp)
933 {
934 	char ifname[IFNAMSIZ];
935 	char *cp;
936 	struct ifnet *ifp;
937 	int vid;
938 
939 	strlcpy(ifname, name, IFNAMSIZ);
940 	if ((cp = strchr(ifname, '.')) == NULL)
941 		return (NULL);
942 	*cp = '\0';
943 	if ((ifp = ifunit_ref(ifname)) == NULL)
944 		return (NULL);
945 	/* Parse VID. */
946 	if (*++cp == '\0') {
947 		if_rele(ifp);
948 		return (NULL);
949 	}
950 	vid = 0;
951 	for(; *cp >= '0' && *cp <= '9'; cp++)
952 		vid = (vid * 10) + (*cp - '0');
953 	if (*cp != '\0') {
954 		if_rele(ifp);
955 		return (NULL);
956 	}
957 	if (vidp != NULL)
958 		*vidp = vid;
959 
960 	return (ifp);
961 }
962 
963 static int
964 vlan_clone_match(struct if_clone *ifc, const char *name)
965 {
966 	struct ifnet *ifp;
967 	const char *cp;
968 
969 	ifp = vlan_clone_match_ethervid(name, NULL);
970 	if (ifp != NULL) {
971 		if_rele(ifp);
972 		return (1);
973 	}
974 
975 	if (strncmp(vlanname, name, strlen(vlanname)) != 0)
976 		return (0);
977 	for (cp = name + 4; *cp != '\0'; cp++) {
978 		if (*cp < '0' || *cp > '9')
979 			return (0);
980 	}
981 
982 	return (1);
983 }
984 
985 static int
986 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
987 {
988 	char *dp;
989 	int wildcard;
990 	int unit;
991 	int error;
992 	int vid;
993 	struct ifvlan *ifv;
994 	struct ifnet *ifp;
995 	struct ifnet *p;
996 	struct ifaddr *ifa;
997 	struct sockaddr_dl *sdl;
998 	struct vlanreq vlr;
999 	static const u_char eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
1000 
1001 	/*
1002 	 * There are 3 (ugh) ways to specify the cloned device:
1003 	 * o pass a parameter block with the clone request.
1004 	 * o specify parameters in the text of the clone device name
1005 	 * o specify no parameters and get an unattached device that
1006 	 *   must be configured separately.
1007 	 * The first technique is preferred; the latter two are
1008 	 * supported for backwards compatibility.
1009 	 *
1010 	 * XXXRW: Note historic use of the word "tag" here.  New ioctls may be
1011 	 * called for.
1012 	 */
1013 	if (params) {
1014 		error = copyin(params, &vlr, sizeof(vlr));
1015 		if (error)
1016 			return error;
1017 		p = ifunit_ref(vlr.vlr_parent);
1018 		if (p == NULL)
1019 			return (ENXIO);
1020 		error = ifc_name2unit(name, &unit);
1021 		if (error != 0) {
1022 			if_rele(p);
1023 			return (error);
1024 		}
1025 		vid = vlr.vlr_tag;
1026 		wildcard = (unit < 0);
1027 	} else if ((p = vlan_clone_match_ethervid(name, &vid)) != NULL) {
1028 		unit = -1;
1029 		wildcard = 0;
1030 	} else {
1031 		p = NULL;
1032 		error = ifc_name2unit(name, &unit);
1033 		if (error != 0)
1034 			return (error);
1035 
1036 		wildcard = (unit < 0);
1037 	}
1038 
1039 	error = ifc_alloc_unit(ifc, &unit);
1040 	if (error != 0) {
1041 		if (p != NULL)
1042 			if_rele(p);
1043 		return (error);
1044 	}
1045 
1046 	/* In the wildcard case, we need to update the name. */
1047 	if (wildcard) {
1048 		for (dp = name; *dp != '\0'; dp++);
1049 		if (snprintf(dp, len - (dp-name), "%d", unit) >
1050 		    len - (dp-name) - 1) {
1051 			panic("%s: interface name too long", __func__);
1052 		}
1053 	}
1054 
1055 	ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
1056 	ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
1057 	if (ifp == NULL) {
1058 		ifc_free_unit(ifc, unit);
1059 		free(ifv, M_VLAN);
1060 		if (p != NULL)
1061 			if_rele(p);
1062 		return (ENOSPC);
1063 	}
1064 	CK_SLIST_INIT(&ifv->vlan_mc_listhead);
1065 	ifp->if_softc = ifv;
1066 	/*
1067 	 * Set the name manually rather than using if_initname because
1068 	 * we don't conform to the default naming convention for interfaces.
1069 	 */
1070 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
1071 	ifp->if_dname = vlanname;
1072 	ifp->if_dunit = unit;
1073 
1074 	ifp->if_init = vlan_init;
1075 	ifp->if_transmit = vlan_transmit;
1076 	ifp->if_qflush = vlan_qflush;
1077 	ifp->if_ioctl = vlan_ioctl;
1078 #if defined(KERN_TLS) || defined(RATELIMIT)
1079 	ifp->if_snd_tag_alloc = vlan_snd_tag_alloc;
1080 	ifp->if_snd_tag_modify = vlan_snd_tag_modify;
1081 	ifp->if_snd_tag_query = vlan_snd_tag_query;
1082 	ifp->if_snd_tag_free = vlan_snd_tag_free;
1083 #endif
1084 	ifp->if_flags = VLAN_IFFLAGS;
1085 	ether_ifattach(ifp, eaddr);
1086 	/* Now undo some of the damage... */
1087 	ifp->if_baudrate = 0;
1088 	ifp->if_type = IFT_L2VLAN;
1089 	ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
1090 	ifa = ifp->if_addr;
1091 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1092 	sdl->sdl_type = IFT_L2VLAN;
1093 
1094 	if (p != NULL) {
1095 		error = vlan_config(ifv, p, vid);
1096 		if_rele(p);
1097 		if (error != 0) {
1098 			/*
1099 			 * Since we've partially failed, we need to back
1100 			 * out all the way, otherwise userland could get
1101 			 * confused.  Thus, we destroy the interface.
1102 			 */
1103 			ether_ifdetach(ifp);
1104 			vlan_unconfig(ifp);
1105 			if_free(ifp);
1106 			ifc_free_unit(ifc, unit);
1107 			free(ifv, M_VLAN);
1108 
1109 			return (error);
1110 		}
1111 	}
1112 
1113 	return (0);
1114 }
1115 
1116 static int
1117 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1118 {
1119 	struct ifvlan *ifv = ifp->if_softc;
1120 	int unit = ifp->if_dunit;
1121 
1122 	ether_ifdetach(ifp);	/* first, remove it from system-wide lists */
1123 	vlan_unconfig(ifp);	/* now it can be unconfigured and freed */
1124 	/*
1125 	 * We should have the only reference to the ifv now, so we can now
1126 	 * drain any remaining lladdr task before freeing the ifnet and the
1127 	 * ifvlan.
1128 	 */
1129 	taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
1130 	NET_EPOCH_WAIT();
1131 	if_free(ifp);
1132 	free(ifv, M_VLAN);
1133 	ifc_free_unit(ifc, unit);
1134 
1135 	return (0);
1136 }
1137 
1138 /*
1139  * The ifp->if_init entry point for vlan(4) is a no-op.
1140  */
1141 static void
1142 vlan_init(void *foo __unused)
1143 {
1144 }
1145 
1146 /*
1147  * The if_transmit method for vlan(4) interface.
1148  */
1149 static int
1150 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1151 {
1152 	struct ifvlan *ifv;
1153 	struct ifnet *p;
1154 	int error, len, mcast;
1155 
1156 	NET_EPOCH_ASSERT();
1157 
1158 	ifv = ifp->if_softc;
1159 	if (TRUNK(ifv) == NULL) {
1160 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1161 		m_freem(m);
1162 		return (ENETDOWN);
1163 	}
1164 	p = PARENT(ifv);
1165 	len = m->m_pkthdr.len;
1166 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1167 
1168 	BPF_MTAP(ifp, m);
1169 
1170 #if defined(KERN_TLS) || defined(RATELIMIT)
1171 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
1172 		struct vlan_snd_tag *vst;
1173 		struct m_snd_tag *mst;
1174 
1175 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
1176 		mst = m->m_pkthdr.snd_tag;
1177 		vst = mst_to_vst(mst);
1178 		if (vst->tag->ifp != p) {
1179 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1180 			m_freem(m);
1181 			return (EAGAIN);
1182 		}
1183 
1184 		m->m_pkthdr.snd_tag = m_snd_tag_ref(vst->tag);
1185 		m_snd_tag_rele(mst);
1186 	}
1187 #endif
1188 
1189 	/*
1190 	 * Do not run parent's if_transmit() if the parent is not up,
1191 	 * or parent's driver will cause a system crash.
1192 	 */
1193 	if (!UP_AND_RUNNING(p)) {
1194 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1195 		m_freem(m);
1196 		return (ENETDOWN);
1197 	}
1198 
1199 	if (!ether_8021q_frame(&m, ifp, p, ifv->ifv_vid, ifv->ifv_pcp)) {
1200 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1201 		return (0);
1202 	}
1203 
1204 	/*
1205 	 * Send it, precisely as ether_output() would have.
1206 	 */
1207 	error = (p->if_transmit)(p, m);
1208 	if (error == 0) {
1209 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1210 		if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1211 		if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1212 	} else
1213 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1214 	return (error);
1215 }
1216 
1217 static int
1218 vlan_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1219     struct route *ro)
1220 {
1221 	struct ifvlan *ifv;
1222 	struct ifnet *p;
1223 
1224 	NET_EPOCH_ASSERT();
1225 
1226 	ifv = ifp->if_softc;
1227 	if (TRUNK(ifv) == NULL) {
1228 		m_freem(m);
1229 		return (ENETDOWN);
1230 	}
1231 	p = PARENT(ifv);
1232 	return p->if_output(ifp, m, dst, ro);
1233 }
1234 
1235 
1236 /*
1237  * The ifp->if_qflush entry point for vlan(4) is a no-op.
1238  */
1239 static void
1240 vlan_qflush(struct ifnet *ifp __unused)
1241 {
1242 }
1243 
1244 static void
1245 vlan_input(struct ifnet *ifp, struct mbuf *m)
1246 {
1247 	struct ifvlantrunk *trunk;
1248 	struct ifvlan *ifv;
1249 	struct m_tag *mtag;
1250 	uint16_t vid, tag;
1251 
1252 	NET_EPOCH_ASSERT();
1253 
1254 	trunk = ifp->if_vlantrunk;
1255 	if (trunk == NULL) {
1256 		m_freem(m);
1257 		return;
1258 	}
1259 
1260 	if (m->m_flags & M_VLANTAG) {
1261 		/*
1262 		 * Packet is tagged, but m contains a normal
1263 		 * Ethernet frame; the tag is stored out-of-band.
1264 		 */
1265 		tag = m->m_pkthdr.ether_vtag;
1266 		m->m_flags &= ~M_VLANTAG;
1267 	} else {
1268 		struct ether_vlan_header *evl;
1269 
1270 		/*
1271 		 * Packet is tagged in-band as specified by 802.1q.
1272 		 */
1273 		switch (ifp->if_type) {
1274 		case IFT_ETHER:
1275 			if (m->m_len < sizeof(*evl) &&
1276 			    (m = m_pullup(m, sizeof(*evl))) == NULL) {
1277 				if_printf(ifp, "cannot pullup VLAN header\n");
1278 				return;
1279 			}
1280 			evl = mtod(m, struct ether_vlan_header *);
1281 			tag = ntohs(evl->evl_tag);
1282 
1283 			/*
1284 			 * Remove the 802.1q header by copying the Ethernet
1285 			 * addresses over it and adjusting the beginning of
1286 			 * the data in the mbuf.  The encapsulated Ethernet
1287 			 * type field is already in place.
1288 			 */
1289 			bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1290 			      ETHER_HDR_LEN - ETHER_TYPE_LEN);
1291 			m_adj(m, ETHER_VLAN_ENCAP_LEN);
1292 			break;
1293 
1294 		default:
1295 #ifdef INVARIANTS
1296 			panic("%s: %s has unsupported if_type %u",
1297 			      __func__, ifp->if_xname, ifp->if_type);
1298 #endif
1299 			if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1300 			m_freem(m);
1301 			return;
1302 		}
1303 	}
1304 
1305 	vid = EVL_VLANOFTAG(tag);
1306 
1307 	ifv = vlan_gethash(trunk, vid);
1308 	if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1309 		if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1310 		m_freem(m);
1311 		return;
1312 	}
1313 
1314 	if (vlan_mtag_pcp) {
1315 		/*
1316 		 * While uncommon, it is possible that we will find a 802.1q
1317 		 * packet encapsulated inside another packet that also had an
1318 		 * 802.1q header.  For example, ethernet tunneled over IPSEC
1319 		 * arriving over ethernet.  In that case, we replace the
1320 		 * existing 802.1q PCP m_tag value.
1321 		 */
1322 		mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1323 		if (mtag == NULL) {
1324 			mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1325 			    sizeof(uint8_t), M_NOWAIT);
1326 			if (mtag == NULL) {
1327 				if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1328 				m_freem(m);
1329 				return;
1330 			}
1331 			m_tag_prepend(m, mtag);
1332 		}
1333 		*(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1334 	}
1335 
1336 	m->m_pkthdr.rcvif = ifv->ifv_ifp;
1337 	if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1338 
1339 	/* Pass it back through the parent's input routine. */
1340 	(*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
1341 }
1342 
1343 static void
1344 vlan_lladdr_fn(void *arg, int pending __unused)
1345 {
1346 	struct ifvlan *ifv;
1347 	struct ifnet *ifp;
1348 
1349 	ifv = (struct ifvlan *)arg;
1350 	ifp = ifv->ifv_ifp;
1351 
1352 	CURVNET_SET(ifp->if_vnet);
1353 
1354 	/* The ifv_ifp already has the lladdr copied in. */
1355 	if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1356 
1357 	CURVNET_RESTORE();
1358 }
1359 
1360 static int
1361 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid)
1362 {
1363 	struct epoch_tracker et;
1364 	struct ifvlantrunk *trunk;
1365 	struct ifnet *ifp;
1366 	int error = 0;
1367 
1368 	/*
1369 	 * We can handle non-ethernet hardware types as long as
1370 	 * they handle the tagging and headers themselves.
1371 	 */
1372 	if (p->if_type != IFT_ETHER &&
1373 	    (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1374 		return (EPROTONOSUPPORT);
1375 	if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1376 		return (EPROTONOSUPPORT);
1377 	/*
1378 	 * Don't let the caller set up a VLAN VID with
1379 	 * anything except VLID bits.
1380 	 * VID numbers 0x0 and 0xFFF are reserved.
1381 	 */
1382 	if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1383 		return (EINVAL);
1384 	if (ifv->ifv_trunk)
1385 		return (EBUSY);
1386 
1387 	VLAN_XLOCK();
1388 	if (p->if_vlantrunk == NULL) {
1389 		trunk = malloc(sizeof(struct ifvlantrunk),
1390 		    M_VLAN, M_WAITOK | M_ZERO);
1391 		vlan_inithash(trunk);
1392 		TRUNK_LOCK_INIT(trunk);
1393 		TRUNK_WLOCK(trunk);
1394 		p->if_vlantrunk = trunk;
1395 		trunk->parent = p;
1396 		if_ref(trunk->parent);
1397 		TRUNK_WUNLOCK(trunk);
1398 	} else {
1399 		trunk = p->if_vlantrunk;
1400 	}
1401 
1402 	ifv->ifv_vid = vid;	/* must set this before vlan_inshash() */
1403 	ifv->ifv_pcp = 0;       /* Default: best effort delivery. */
1404 	vlan_tag_recalculate(ifv);
1405 	error = vlan_inshash(trunk, ifv);
1406 	if (error)
1407 		goto done;
1408 	ifv->ifv_proto = ETHERTYPE_VLAN;
1409 	ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1410 	ifv->ifv_mintu = ETHERMIN;
1411 	ifv->ifv_pflags = 0;
1412 	ifv->ifv_capenable = -1;
1413 
1414 	/*
1415 	 * If the parent supports the VLAN_MTU capability,
1416 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1417 	 * use it.
1418 	 */
1419 	if (p->if_capenable & IFCAP_VLAN_MTU) {
1420 		/*
1421 		 * No need to fudge the MTU since the parent can
1422 		 * handle extended frames.
1423 		 */
1424 		ifv->ifv_mtufudge = 0;
1425 	} else {
1426 		/*
1427 		 * Fudge the MTU by the encapsulation size.  This
1428 		 * makes us incompatible with strictly compliant
1429 		 * 802.1Q implementations, but allows us to use
1430 		 * the feature with other NetBSD implementations,
1431 		 * which might still be useful.
1432 		 */
1433 		ifv->ifv_mtufudge = ifv->ifv_encaplen;
1434 	}
1435 
1436 	ifv->ifv_trunk = trunk;
1437 	ifp = ifv->ifv_ifp;
1438 	/*
1439 	 * Initialize fields from our parent.  This duplicates some
1440 	 * work with ether_ifattach() but allows for non-ethernet
1441 	 * interfaces to also work.
1442 	 */
1443 	ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1444 	ifp->if_baudrate = p->if_baudrate;
1445 	ifp->if_input = p->if_input;
1446 	ifp->if_resolvemulti = p->if_resolvemulti;
1447 	ifp->if_addrlen = p->if_addrlen;
1448 	ifp->if_broadcastaddr = p->if_broadcastaddr;
1449 	ifp->if_pcp = ifv->ifv_pcp;
1450 
1451 	/*
1452 	 * We wrap the parent's if_output using vlan_output to ensure that it
1453 	 * can't become stale.
1454 	 */
1455 	ifp->if_output = vlan_output;
1456 
1457 	/*
1458 	 * Copy only a selected subset of flags from the parent.
1459 	 * Other flags are none of our business.
1460 	 */
1461 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1462 	ifp->if_flags &= ~VLAN_COPY_FLAGS;
1463 	ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1464 #undef VLAN_COPY_FLAGS
1465 
1466 	ifp->if_link_state = p->if_link_state;
1467 
1468 	NET_EPOCH_ENTER(et);
1469 	vlan_capabilities(ifv);
1470 	NET_EPOCH_EXIT(et);
1471 
1472 	/*
1473 	 * Set up our interface address to reflect the underlying
1474 	 * physical interface's.
1475 	 */
1476 	TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
1477 	((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1478 	    p->if_addrlen;
1479 
1480 	/*
1481 	 * Do not schedule link address update if it was the same
1482 	 * as previous parent's. This helps avoid updating for each
1483 	 * associated llentry.
1484 	 */
1485 	if (memcmp(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen) != 0) {
1486 		bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1487 		taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
1488 	}
1489 
1490 	/* We are ready for operation now. */
1491 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1492 
1493 	/* Update flags on the parent, if necessary. */
1494 	vlan_setflags(ifp, 1);
1495 
1496 	/*
1497 	 * Configure multicast addresses that may already be
1498 	 * joined on the vlan device.
1499 	 */
1500 	(void)vlan_setmulti(ifp);
1501 
1502 done:
1503 	if (error == 0)
1504 		EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1505 	VLAN_XUNLOCK();
1506 
1507 	return (error);
1508 }
1509 
1510 static void
1511 vlan_unconfig(struct ifnet *ifp)
1512 {
1513 
1514 	VLAN_XLOCK();
1515 	vlan_unconfig_locked(ifp, 0);
1516 	VLAN_XUNLOCK();
1517 }
1518 
1519 static void
1520 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1521 {
1522 	struct ifvlantrunk *trunk;
1523 	struct vlan_mc_entry *mc;
1524 	struct ifvlan *ifv;
1525 	struct ifnet  *parent;
1526 	int error;
1527 
1528 	VLAN_XLOCK_ASSERT();
1529 
1530 	ifv = ifp->if_softc;
1531 	trunk = ifv->ifv_trunk;
1532 	parent = NULL;
1533 
1534 	if (trunk != NULL) {
1535 		parent = trunk->parent;
1536 
1537 		/*
1538 		 * Since the interface is being unconfigured, we need to
1539 		 * empty the list of multicast groups that we may have joined
1540 		 * while we were alive from the parent's list.
1541 		 */
1542 		while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1543 			/*
1544 			 * If the parent interface is being detached,
1545 			 * all its multicast addresses have already
1546 			 * been removed.  Warn about errors if
1547 			 * if_delmulti() does fail, but don't abort as
1548 			 * all callers expect vlan destruction to
1549 			 * succeed.
1550 			 */
1551 			if (!departing) {
1552 				error = if_delmulti(parent,
1553 				    (struct sockaddr *)&mc->mc_addr);
1554 				if (error)
1555 					if_printf(ifp,
1556 		    "Failed to delete multicast address from parent: %d\n",
1557 					    error);
1558 			}
1559 			CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1560 			NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
1561 		}
1562 
1563 		vlan_setflags(ifp, 0); /* clear special flags on parent */
1564 
1565 		vlan_remhash(trunk, ifv);
1566 		ifv->ifv_trunk = NULL;
1567 
1568 		/*
1569 		 * Check if we were the last.
1570 		 */
1571 		if (trunk->refcnt == 0) {
1572 			parent->if_vlantrunk = NULL;
1573 			NET_EPOCH_WAIT();
1574 			trunk_destroy(trunk);
1575 		}
1576 	}
1577 
1578 	/* Disconnect from parent. */
1579 	if (ifv->ifv_pflags)
1580 		if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1581 	ifp->if_mtu = ETHERMTU;
1582 	ifp->if_link_state = LINK_STATE_UNKNOWN;
1583 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1584 
1585 	/*
1586 	 * Only dispatch an event if vlan was
1587 	 * attached, otherwise there is nothing
1588 	 * to cleanup anyway.
1589 	 */
1590 	if (parent != NULL)
1591 		EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1592 }
1593 
1594 /* Handle a reference counted flag that should be set on the parent as well */
1595 static int
1596 vlan_setflag(struct ifnet *ifp, int flag, int status,
1597 	     int (*func)(struct ifnet *, int))
1598 {
1599 	struct ifvlan *ifv;
1600 	int error;
1601 
1602 	VLAN_SXLOCK_ASSERT();
1603 
1604 	ifv = ifp->if_softc;
1605 	status = status ? (ifp->if_flags & flag) : 0;
1606 	/* Now "status" contains the flag value or 0 */
1607 
1608 	/*
1609 	 * See if recorded parent's status is different from what
1610 	 * we want it to be.  If it is, flip it.  We record parent's
1611 	 * status in ifv_pflags so that we won't clear parent's flag
1612 	 * we haven't set.  In fact, we don't clear or set parent's
1613 	 * flags directly, but get or release references to them.
1614 	 * That's why we can be sure that recorded flags still are
1615 	 * in accord with actual parent's flags.
1616 	 */
1617 	if (status != (ifv->ifv_pflags & flag)) {
1618 		error = (*func)(PARENT(ifv), status);
1619 		if (error)
1620 			return (error);
1621 		ifv->ifv_pflags &= ~flag;
1622 		ifv->ifv_pflags |= status;
1623 	}
1624 	return (0);
1625 }
1626 
1627 /*
1628  * Handle IFF_* flags that require certain changes on the parent:
1629  * if "status" is true, update parent's flags respective to our if_flags;
1630  * if "status" is false, forcedly clear the flags set on parent.
1631  */
1632 static int
1633 vlan_setflags(struct ifnet *ifp, int status)
1634 {
1635 	int error, i;
1636 
1637 	for (i = 0; vlan_pflags[i].flag; i++) {
1638 		error = vlan_setflag(ifp, vlan_pflags[i].flag,
1639 				     status, vlan_pflags[i].func);
1640 		if (error)
1641 			return (error);
1642 	}
1643 	return (0);
1644 }
1645 
1646 /* Inform all vlans that their parent has changed link state */
1647 static void
1648 vlan_link_state(struct ifnet *ifp)
1649 {
1650 	struct epoch_tracker et;
1651 	struct ifvlantrunk *trunk;
1652 	struct ifvlan *ifv;
1653 
1654 	NET_EPOCH_ENTER(et);
1655 	trunk = ifp->if_vlantrunk;
1656 	if (trunk == NULL) {
1657 		NET_EPOCH_EXIT(et);
1658 		return;
1659 	}
1660 
1661 	TRUNK_WLOCK(trunk);
1662 	VLAN_FOREACH(ifv, trunk) {
1663 		ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1664 		if_link_state_change(ifv->ifv_ifp,
1665 		    trunk->parent->if_link_state);
1666 	}
1667 	TRUNK_WUNLOCK(trunk);
1668 	NET_EPOCH_EXIT(et);
1669 }
1670 
1671 static void
1672 vlan_capabilities(struct ifvlan *ifv)
1673 {
1674 	struct ifnet *p;
1675 	struct ifnet *ifp;
1676 	struct ifnet_hw_tsomax hw_tsomax;
1677 	int cap = 0, ena = 0, mena;
1678 	u_long hwa = 0;
1679 
1680 	NET_EPOCH_ASSERT();
1681 	VLAN_SXLOCK_ASSERT();
1682 
1683 	p = PARENT(ifv);
1684 	ifp = ifv->ifv_ifp;
1685 
1686 	/* Mask parent interface enabled capabilities disabled by user. */
1687 	mena = p->if_capenable & ifv->ifv_capenable;
1688 
1689 	/*
1690 	 * If the parent interface can do checksum offloading
1691 	 * on VLANs, then propagate its hardware-assisted
1692 	 * checksumming flags. Also assert that checksum
1693 	 * offloading requires hardware VLAN tagging.
1694 	 */
1695 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1696 		cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1697 	if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1698 	    p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1699 		ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1700 		if (ena & IFCAP_TXCSUM)
1701 			hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
1702 			    CSUM_UDP | CSUM_SCTP);
1703 		if (ena & IFCAP_TXCSUM_IPV6)
1704 			hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
1705 			    CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
1706 	}
1707 
1708 	/*
1709 	 * If the parent interface can do TSO on VLANs then
1710 	 * propagate the hardware-assisted flag. TSO on VLANs
1711 	 * does not necessarily require hardware VLAN tagging.
1712 	 */
1713 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1714 	if_hw_tsomax_common(p, &hw_tsomax);
1715 	if_hw_tsomax_update(ifp, &hw_tsomax);
1716 	if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1717 		cap |= p->if_capabilities & IFCAP_TSO;
1718 	if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1719 		ena |= mena & IFCAP_TSO;
1720 		if (ena & IFCAP_TSO)
1721 			hwa |= p->if_hwassist & CSUM_TSO;
1722 	}
1723 
1724 	/*
1725 	 * If the parent interface can do LRO and checksum offloading on
1726 	 * VLANs, then guess it may do LRO on VLANs.  False positive here
1727 	 * cost nothing, while false negative may lead to some confusions.
1728 	 */
1729 	if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1730 		cap |= p->if_capabilities & IFCAP_LRO;
1731 	if (p->if_capenable & IFCAP_VLAN_HWCSUM)
1732 		ena |= p->if_capenable & IFCAP_LRO;
1733 
1734 	/*
1735 	 * If the parent interface can offload TCP connections over VLANs then
1736 	 * propagate its TOE capability to the VLAN interface.
1737 	 *
1738 	 * All TOE drivers in the tree today can deal with VLANs.  If this
1739 	 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1740 	 * with its own bit.
1741 	 */
1742 #define	IFCAP_VLAN_TOE IFCAP_TOE
1743 	if (p->if_capabilities & IFCAP_VLAN_TOE)
1744 		cap |= p->if_capabilities & IFCAP_TOE;
1745 	if (p->if_capenable & IFCAP_VLAN_TOE) {
1746 		TOEDEV(ifp) = TOEDEV(p);
1747 		ena |= mena & IFCAP_TOE;
1748 	}
1749 
1750 	/*
1751 	 * If the parent interface supports dynamic link state, so does the
1752 	 * VLAN interface.
1753 	 */
1754 	cap |= (p->if_capabilities & IFCAP_LINKSTATE);
1755 	ena |= (mena & IFCAP_LINKSTATE);
1756 
1757 #ifdef RATELIMIT
1758 	/*
1759 	 * If the parent interface supports ratelimiting, so does the
1760 	 * VLAN interface.
1761 	 */
1762 	cap |= (p->if_capabilities & IFCAP_TXRTLMT);
1763 	ena |= (mena & IFCAP_TXRTLMT);
1764 #endif
1765 
1766 	/*
1767 	 * If the parent interface supports unmapped mbufs, so does
1768 	 * the VLAN interface.  Note that this should be fine even for
1769 	 * interfaces that don't support hardware tagging as headers
1770 	 * are prepended in normal mbufs to unmapped mbufs holding
1771 	 * payload data.
1772 	 */
1773 	cap |= (p->if_capabilities & IFCAP_NOMAP);
1774 	ena |= (mena & IFCAP_NOMAP);
1775 
1776 	/*
1777 	 * If the parent interface can offload encryption and segmentation
1778 	 * of TLS records over TCP, propagate it's capability to the VLAN
1779 	 * interface.
1780 	 *
1781 	 * All TLS drivers in the tree today can deal with VLANs.  If
1782 	 * this ever changes, then a new IFCAP_VLAN_TXTLS can be
1783 	 * defined.
1784 	 */
1785 	if (p->if_capabilities & IFCAP_TXTLS)
1786 		cap |= p->if_capabilities & IFCAP_TXTLS;
1787 	if (p->if_capenable & IFCAP_TXTLS)
1788 		ena |= mena & IFCAP_TXTLS;
1789 
1790 	ifp->if_capabilities = cap;
1791 	ifp->if_capenable = ena;
1792 	ifp->if_hwassist = hwa;
1793 }
1794 
1795 static void
1796 vlan_trunk_capabilities(struct ifnet *ifp)
1797 {
1798 	struct epoch_tracker et;
1799 	struct ifvlantrunk *trunk;
1800 	struct ifvlan *ifv;
1801 
1802 	VLAN_SLOCK();
1803 	trunk = ifp->if_vlantrunk;
1804 	if (trunk == NULL) {
1805 		VLAN_SUNLOCK();
1806 		return;
1807 	}
1808 	NET_EPOCH_ENTER(et);
1809 	VLAN_FOREACH(ifv, trunk)
1810 		vlan_capabilities(ifv);
1811 	NET_EPOCH_EXIT(et);
1812 	VLAN_SUNLOCK();
1813 }
1814 
1815 static int
1816 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1817 {
1818 	struct ifnet *p;
1819 	struct ifreq *ifr;
1820 	struct ifaddr *ifa;
1821 	struct ifvlan *ifv;
1822 	struct ifvlantrunk *trunk;
1823 	struct vlanreq vlr;
1824 	int error = 0, oldmtu;
1825 
1826 	ifr = (struct ifreq *)data;
1827 	ifa = (struct ifaddr *) data;
1828 	ifv = ifp->if_softc;
1829 
1830 	switch (cmd) {
1831 	case SIOCSIFADDR:
1832 		ifp->if_flags |= IFF_UP;
1833 #ifdef INET
1834 		if (ifa->ifa_addr->sa_family == AF_INET)
1835 			arp_ifinit(ifp, ifa);
1836 #endif
1837 		break;
1838 	case SIOCGIFADDR:
1839 		bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
1840 		    ifp->if_addrlen);
1841 		break;
1842 	case SIOCGIFMEDIA:
1843 		VLAN_SLOCK();
1844 		if (TRUNK(ifv) != NULL) {
1845 			p = PARENT(ifv);
1846 			if_ref(p);
1847 			error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1848 			if_rele(p);
1849 			/* Limit the result to the parent's current config. */
1850 			if (error == 0) {
1851 				struct ifmediareq *ifmr;
1852 
1853 				ifmr = (struct ifmediareq *)data;
1854 				if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1855 					ifmr->ifm_count = 1;
1856 					error = copyout(&ifmr->ifm_current,
1857 						ifmr->ifm_ulist,
1858 						sizeof(int));
1859 				}
1860 			}
1861 		} else {
1862 			error = EINVAL;
1863 		}
1864 		VLAN_SUNLOCK();
1865 		break;
1866 
1867 	case SIOCSIFMEDIA:
1868 		error = EINVAL;
1869 		break;
1870 
1871 	case SIOCSIFMTU:
1872 		/*
1873 		 * Set the interface MTU.
1874 		 */
1875 		VLAN_SLOCK();
1876 		trunk = TRUNK(ifv);
1877 		if (trunk != NULL) {
1878 			TRUNK_WLOCK(trunk);
1879 			if (ifr->ifr_mtu >
1880 			     (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1881 			    ifr->ifr_mtu <
1882 			     (ifv->ifv_mintu - ifv->ifv_mtufudge))
1883 				error = EINVAL;
1884 			else
1885 				ifp->if_mtu = ifr->ifr_mtu;
1886 			TRUNK_WUNLOCK(trunk);
1887 		} else
1888 			error = EINVAL;
1889 		VLAN_SUNLOCK();
1890 		break;
1891 
1892 	case SIOCSETVLAN:
1893 #ifdef VIMAGE
1894 		/*
1895 		 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1896 		 * interface to be delegated to a jail without allowing the
1897 		 * jail to change what underlying interface/VID it is
1898 		 * associated with.  We are not entirely convinced that this
1899 		 * is the right way to accomplish that policy goal.
1900 		 */
1901 		if (ifp->if_vnet != ifp->if_home_vnet) {
1902 			error = EPERM;
1903 			break;
1904 		}
1905 #endif
1906 		error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
1907 		if (error)
1908 			break;
1909 		if (vlr.vlr_parent[0] == '\0') {
1910 			vlan_unconfig(ifp);
1911 			break;
1912 		}
1913 		p = ifunit_ref(vlr.vlr_parent);
1914 		if (p == NULL) {
1915 			error = ENOENT;
1916 			break;
1917 		}
1918 		oldmtu = ifp->if_mtu;
1919 		error = vlan_config(ifv, p, vlr.vlr_tag);
1920 		if_rele(p);
1921 
1922 		/*
1923 		 * VLAN MTU may change during addition of the vlandev.
1924 		 * If it did, do network layer specific procedure.
1925 		 */
1926 		if (ifp->if_mtu != oldmtu) {
1927 #ifdef INET6
1928 			nd6_setmtu(ifp);
1929 #endif
1930 			rt_updatemtu(ifp);
1931 		}
1932 		break;
1933 
1934 	case SIOCGETVLAN:
1935 #ifdef VIMAGE
1936 		if (ifp->if_vnet != ifp->if_home_vnet) {
1937 			error = EPERM;
1938 			break;
1939 		}
1940 #endif
1941 		bzero(&vlr, sizeof(vlr));
1942 		VLAN_SLOCK();
1943 		if (TRUNK(ifv) != NULL) {
1944 			strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1945 			    sizeof(vlr.vlr_parent));
1946 			vlr.vlr_tag = ifv->ifv_vid;
1947 		}
1948 		VLAN_SUNLOCK();
1949 		error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
1950 		break;
1951 
1952 	case SIOCSIFFLAGS:
1953 		/*
1954 		 * We should propagate selected flags to the parent,
1955 		 * e.g., promiscuous mode.
1956 		 */
1957 		VLAN_XLOCK();
1958 		if (TRUNK(ifv) != NULL)
1959 			error = vlan_setflags(ifp, 1);
1960 		VLAN_XUNLOCK();
1961 		break;
1962 
1963 	case SIOCADDMULTI:
1964 	case SIOCDELMULTI:
1965 		/*
1966 		 * If we don't have a parent, just remember the membership for
1967 		 * when we do.
1968 		 *
1969 		 * XXX We need the rmlock here to avoid sleeping while
1970 		 * holding in6_multi_mtx.
1971 		 */
1972 		VLAN_XLOCK();
1973 		trunk = TRUNK(ifv);
1974 		if (trunk != NULL)
1975 			error = vlan_setmulti(ifp);
1976 		VLAN_XUNLOCK();
1977 
1978 		break;
1979 	case SIOCGVLANPCP:
1980 #ifdef VIMAGE
1981 		if (ifp->if_vnet != ifp->if_home_vnet) {
1982 			error = EPERM;
1983 			break;
1984 		}
1985 #endif
1986 		ifr->ifr_vlan_pcp = ifv->ifv_pcp;
1987 		break;
1988 
1989 	case SIOCSVLANPCP:
1990 #ifdef VIMAGE
1991 		if (ifp->if_vnet != ifp->if_home_vnet) {
1992 			error = EPERM;
1993 			break;
1994 		}
1995 #endif
1996 		error = priv_check(curthread, PRIV_NET_SETVLANPCP);
1997 		if (error)
1998 			break;
1999 		if (ifr->ifr_vlan_pcp > 7) {
2000 			error = EINVAL;
2001 			break;
2002 		}
2003 		ifv->ifv_pcp = ifr->ifr_vlan_pcp;
2004 		ifp->if_pcp = ifv->ifv_pcp;
2005 		vlan_tag_recalculate(ifv);
2006 		/* broadcast event about PCP change */
2007 		EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
2008 		break;
2009 
2010 	case SIOCSIFCAP:
2011 		VLAN_SLOCK();
2012 		ifv->ifv_capenable = ifr->ifr_reqcap;
2013 		trunk = TRUNK(ifv);
2014 		if (trunk != NULL) {
2015 			struct epoch_tracker et;
2016 
2017 			NET_EPOCH_ENTER(et);
2018 			vlan_capabilities(ifv);
2019 			NET_EPOCH_EXIT(et);
2020 		}
2021 		VLAN_SUNLOCK();
2022 		break;
2023 
2024 	default:
2025 		error = EINVAL;
2026 		break;
2027 	}
2028 
2029 	return (error);
2030 }
2031 
2032 #if defined(KERN_TLS) || defined(RATELIMIT)
2033 static int
2034 vlan_snd_tag_alloc(struct ifnet *ifp,
2035     union if_snd_tag_alloc_params *params,
2036     struct m_snd_tag **ppmt)
2037 {
2038 	struct epoch_tracker et;
2039 	struct vlan_snd_tag *vst;
2040 	struct ifvlan *ifv;
2041 	struct ifnet *parent;
2042 	int error;
2043 
2044 	NET_EPOCH_ENTER(et);
2045 	ifv = ifp->if_softc;
2046 	if (ifv->ifv_trunk != NULL)
2047 		parent = PARENT(ifv);
2048 	else
2049 		parent = NULL;
2050 	if (parent == NULL || parent->if_snd_tag_alloc == NULL) {
2051 		NET_EPOCH_EXIT(et);
2052 		return (EOPNOTSUPP);
2053 	}
2054 	if_ref(parent);
2055 	NET_EPOCH_EXIT(et);
2056 
2057 	vst = malloc(sizeof(*vst), M_VLAN, M_NOWAIT);
2058 	if (vst == NULL) {
2059 		if_rele(parent);
2060 		return (ENOMEM);
2061 	}
2062 
2063 	error = parent->if_snd_tag_alloc(parent, params, &vst->tag);
2064 	if_rele(parent);
2065 	if (error) {
2066 		free(vst, M_VLAN);
2067 		return (error);
2068 	}
2069 
2070 	m_snd_tag_init(&vst->com, ifp);
2071 
2072 	*ppmt = &vst->com;
2073 	return (0);
2074 }
2075 
2076 static int
2077 vlan_snd_tag_modify(struct m_snd_tag *mst,
2078     union if_snd_tag_modify_params *params)
2079 {
2080 	struct vlan_snd_tag *vst;
2081 
2082 	vst = mst_to_vst(mst);
2083 	return (vst->tag->ifp->if_snd_tag_modify(vst->tag, params));
2084 }
2085 
2086 static int
2087 vlan_snd_tag_query(struct m_snd_tag *mst,
2088     union if_snd_tag_query_params *params)
2089 {
2090 	struct vlan_snd_tag *vst;
2091 
2092 	vst = mst_to_vst(mst);
2093 	return (vst->tag->ifp->if_snd_tag_query(vst->tag, params));
2094 }
2095 
2096 static void
2097 vlan_snd_tag_free(struct m_snd_tag *mst)
2098 {
2099 	struct vlan_snd_tag *vst;
2100 
2101 	vst = mst_to_vst(mst);
2102 	m_snd_tag_rele(vst->tag);
2103 	free(vst, M_VLAN);
2104 }
2105 #endif
2106