2 * Copyright 1998 Massachusetts Institute of Technology
3 * Copyright 2012 ADARA Networks, Inc.
4 * Copyright 2017 Dell EMC Isilon
6 * Portions of this software were developed by Robert N. M. Watson under
7 * contract to ADARA Networks, Inc.
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
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
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.
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
49 #include "opt_inet6.h"
50 #include "opt_kern_tls.h"
52 #include "opt_ratelimit.h"
54 #include <sys/param.h>
55 #include <sys/eventhandler.h>
56 #include <sys/kernel.h>
58 #include <sys/malloc.h>
60 #include <sys/module.h>
61 #include <sys/rmlock.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>
69 #include <sys/taskqueue.h>
72 #include <net/ethernet.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>
83 #include <netinet/in.h>
84 #include <netinet/if_ether.h>
89 * XXX: declare here to avoid to include many inet6 related files..
90 * should be more generalized?
92 extern void nd6_setmtu(struct ifnet *);
95 #define VLAN_DEF_HWIDTH 4
96 #define VLAN_IFFLAGS (IFF_BROADCAST | IFF_MULTICAST)
98 #define UP_AND_RUNNING(ifp) \
99 ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
101 CK_SLIST_HEAD(ifvlanhead, ifvlan);
104 struct ifnet *parent; /* parent interface of this trunk */
107 #define VLAN_ARRAY_SIZE (EVL_VLID_MASK + 1)
108 struct ifvlan *vlans[VLAN_ARRAY_SIZE]; /* static table */
110 struct ifvlanhead *hash; /* dynamic hash-list table */
117 #if defined(KERN_TLS) || defined(RATELIMIT)
118 struct vlan_snd_tag {
119 struct m_snd_tag com;
120 struct m_snd_tag *tag;
123 static inline struct vlan_snd_tag *
124 mst_to_vst(struct m_snd_tag *mst)
127 return (__containerof(mst, struct vlan_snd_tag, com));
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.
136 #define VLAN_FOREACH(_ifv, _trunk) \
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; \
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 */
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.
154 /* The VLAN_ARRAY case is simple -- just a for loop using the condition. */
155 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
157 for (_i = 0; !(_cond) && _i < VLAN_ARRAY_SIZE; _i++) \
158 if (((_ifv) = (_trunk)->vlans[_i]))
159 #else /* VLAN_ARRAY */
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.
168 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
170 bool _touch = false; \
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 && \
176 #endif /* VLAN_ARRAY */
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;
185 struct ifvlantrunk *ifv_trunk;
186 struct ifnet *ifv_ifp;
187 #define TRUNK(ifv) ((ifv)->ifv_trunk)
188 #define PARENT(ifv) (TRUNK(ifv)->parent)
190 int ifv_pflags; /* special flags we have set on parent */
192 int ifv_encaplen; /* encapsulation length */
193 int ifv_mtufudge; /* MTU fudged by this much */
194 int ifv_mintu; /* min transmission unit */
195 struct ether_8021q_tag ifv_qtag;
196 #define ifv_proto ifv_qtag.proto
197 #define ifv_vid ifv_qtag.vid
198 #define ifv_pcp ifv_qtag.pcp
199 struct task lladdr_task;
200 CK_SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
202 CK_SLIST_ENTRY(ifvlan) ifv_list;
206 /* Special flags we should propagate to parent. */
209 int (*func)(struct ifnet *, int);
211 {IFF_PROMISC, ifpromisc},
212 {IFF_ALLMULTI, if_allmulti},
216 extern int vlan_mtag_pcp;
218 static const char vlanname[] = "vlan";
219 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
221 static eventhandler_tag ifdetach_tag;
222 static eventhandler_tag iflladdr_tag;
223 static eventhandler_tag ifevent_tag;
226 * if_vlan uses two module-level synchronizations primitives to allow concurrent
227 * modification of vlan interfaces and (mostly) allow for vlans to be destroyed
228 * while they are being used for tx/rx. To accomplish this in a way that has
229 * acceptable performance and cooperation with other parts of the network stack
230 * there is a non-sleepable epoch(9) and an sx(9).
232 * The performance-sensitive paths that warrant using the epoch(9) are
233 * vlan_transmit and vlan_input. Both have to check for the vlan interface's
234 * existence using if_vlantrunk, and being in the network tx/rx paths the use
235 * of an epoch(9) gives a measureable improvement in performance.
237 * The reason for having an sx(9) is mostly because there are still areas that
238 * must be sleepable and also have safe concurrent access to a vlan interface.
239 * Since the sx(9) exists, it is used by default in most paths unless sleeping
240 * is not permitted, or if it is not clear whether sleeping is permitted.
243 #define _VLAN_SX_ID ifv_sx
245 static struct sx _VLAN_SX_ID;
247 #define VLAN_LOCKING_INIT() \
248 sx_init_flags(&_VLAN_SX_ID, "vlan_sx", SX_RECURSE)
250 #define VLAN_LOCKING_DESTROY() \
251 sx_destroy(&_VLAN_SX_ID)
253 #define VLAN_SLOCK() sx_slock(&_VLAN_SX_ID)
254 #define VLAN_SUNLOCK() sx_sunlock(&_VLAN_SX_ID)
255 #define VLAN_XLOCK() sx_xlock(&_VLAN_SX_ID)
256 #define VLAN_XUNLOCK() sx_xunlock(&_VLAN_SX_ID)
257 #define VLAN_SLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_SLOCKED)
258 #define VLAN_XLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_XLOCKED)
259 #define VLAN_SXLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_LOCKED)
262 * We also have a per-trunk mutex that should be acquired when changing
265 #define TRUNK_LOCK_INIT(trunk) mtx_init(&(trunk)->lock, vlanname, NULL, MTX_DEF)
266 #define TRUNK_LOCK_DESTROY(trunk) mtx_destroy(&(trunk)->lock)
267 #define TRUNK_WLOCK(trunk) mtx_lock(&(trunk)->lock)
268 #define TRUNK_WUNLOCK(trunk) mtx_unlock(&(trunk)->lock)
269 #define TRUNK_WLOCK_ASSERT(trunk) mtx_assert(&(trunk)->lock, MA_OWNED);
272 * The VLAN_ARRAY substitutes the dynamic hash with a static array
273 * with 4096 entries. In theory this can give a boost in processing,
274 * however in practice it does not. Probably this is because the array
275 * is too big to fit into CPU cache.
278 static void vlan_inithash(struct ifvlantrunk *trunk);
279 static void vlan_freehash(struct ifvlantrunk *trunk);
280 static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
281 static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
282 static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
283 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
286 static void trunk_destroy(struct ifvlantrunk *trunk);
288 static void vlan_init(void *foo);
289 static void vlan_input(struct ifnet *ifp, struct mbuf *m);
290 static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
291 #if defined(KERN_TLS) || defined(RATELIMIT)
292 static int vlan_snd_tag_alloc(struct ifnet *,
293 union if_snd_tag_alloc_params *, struct m_snd_tag **);
294 static int vlan_snd_tag_modify(struct m_snd_tag *,
295 union if_snd_tag_modify_params *);
296 static int vlan_snd_tag_query(struct m_snd_tag *,
297 union if_snd_tag_query_params *);
298 static void vlan_snd_tag_free(struct m_snd_tag *);
299 static struct m_snd_tag *vlan_next_snd_tag(struct m_snd_tag *);
300 static void vlan_ratelimit_query(struct ifnet *,
301 struct if_ratelimit_query_results *);
303 static void vlan_qflush(struct ifnet *ifp);
304 static int vlan_setflag(struct ifnet *ifp, int flag, int status,
305 int (*func)(struct ifnet *, int));
306 static int vlan_setflags(struct ifnet *ifp, int status);
307 static int vlan_setmulti(struct ifnet *ifp);
308 static int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
310 static void vlan_altq_start(struct ifnet *ifp);
311 static int vlan_altq_transmit(struct ifnet *ifp, struct mbuf *m);
313 static int vlan_output(struct ifnet *ifp, struct mbuf *m,
314 const struct sockaddr *dst, struct route *ro);
315 static void vlan_unconfig(struct ifnet *ifp);
316 static void vlan_unconfig_locked(struct ifnet *ifp, int departing);
317 static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag,
319 static void vlan_link_state(struct ifnet *ifp);
320 static void vlan_capabilities(struct ifvlan *ifv);
321 static void vlan_trunk_capabilities(struct ifnet *ifp);
323 static struct ifnet *vlan_clone_match_ethervid(const char *, int *);
324 static int vlan_clone_match(struct if_clone *, const char *);
325 static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
326 static int vlan_clone_destroy(struct if_clone *, struct ifnet *);
328 static void vlan_ifdetach(void *arg, struct ifnet *ifp);
329 static void vlan_iflladdr(void *arg, struct ifnet *ifp);
330 static void vlan_ifevent(void *arg, struct ifnet *ifp, int event);
332 static void vlan_lladdr_fn(void *arg, int pending);
334 static struct if_clone *vlan_cloner;
337 VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner);
338 #define V_vlan_cloner VNET(vlan_cloner)
342 vlan_mc_free(struct epoch_context *ctx)
344 struct vlan_mc_entry *mc = __containerof(ctx, struct vlan_mc_entry, mc_epoch_ctx);
349 #define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
352 vlan_inithash(struct ifvlantrunk *trunk)
357 * The trunk must not be locked here since we call malloc(M_WAITOK).
358 * It is OK in case this function is called before the trunk struct
359 * gets hooked up and becomes visible from other threads.
362 KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
363 ("%s: hash already initialized", __func__));
365 trunk->hwidth = VLAN_DEF_HWIDTH;
366 n = 1 << trunk->hwidth;
367 trunk->hmask = n - 1;
368 trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
369 for (i = 0; i < n; i++)
370 CK_SLIST_INIT(&trunk->hash[i]);
374 vlan_freehash(struct ifvlantrunk *trunk)
379 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
380 for (i = 0; i < (1 << trunk->hwidth); i++)
381 KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]),
382 ("%s: hash table not empty", __func__));
384 free(trunk->hash, M_VLAN);
386 trunk->hwidth = trunk->hmask = 0;
390 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
396 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
398 b = 1 << trunk->hwidth;
399 i = HASH(ifv->ifv_vid, trunk->hmask);
400 CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
401 if (ifv->ifv_vid == ifv2->ifv_vid)
405 * Grow the hash when the number of vlans exceeds half of the number of
406 * hash buckets squared. This will make the average linked-list length
409 if (trunk->refcnt > (b * b) / 2) {
410 vlan_growhash(trunk, 1);
411 i = HASH(ifv->ifv_vid, trunk->hmask);
413 CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
420 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
426 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
428 b = 1 << (trunk->hwidth - 1);
429 i = HASH(ifv->ifv_vid, trunk->hmask);
430 CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
433 CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list);
434 if (trunk->refcnt < (b * b) / 2)
435 vlan_growhash(trunk, -1);
439 panic("%s: vlan not found\n", __func__);
440 return (ENOENT); /*NOTREACHED*/
444 * Grow the hash larger or smaller if memory permits.
447 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
450 struct ifvlanhead *hash2;
451 int hwidth2, i, j, n, n2;
454 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
457 /* Harmless yet obvious coding error */
458 printf("%s: howmuch is 0\n", __func__);
462 hwidth2 = trunk->hwidth + howmuch;
463 n = 1 << trunk->hwidth;
465 /* Do not shrink the table below the default */
466 if (hwidth2 < VLAN_DEF_HWIDTH)
469 hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK);
471 printf("%s: out of memory -- hash size not changed\n",
473 return; /* We can live with the old hash table */
475 for (j = 0; j < n2; j++)
476 CK_SLIST_INIT(&hash2[j]);
477 for (i = 0; i < n; i++)
478 while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) {
479 CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list);
480 j = HASH(ifv->ifv_vid, n2 - 1);
481 CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
484 free(trunk->hash, M_VLAN);
486 trunk->hwidth = hwidth2;
487 trunk->hmask = n2 - 1;
490 if_printf(trunk->parent,
491 "VLAN hash table resized from %d to %d buckets\n", n, n2);
494 static __inline struct ifvlan *
495 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
501 CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
502 if (ifv->ifv_vid == vid)
508 /* Debugging code to view the hashtables. */
510 vlan_dumphash(struct ifvlantrunk *trunk)
515 for (i = 0; i < (1 << trunk->hwidth); i++) {
517 CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
518 printf("%s ", ifv->ifv_ifp->if_xname);
525 static __inline struct ifvlan *
526 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
529 return trunk->vlans[vid];
533 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
536 if (trunk->vlans[ifv->ifv_vid] != NULL)
538 trunk->vlans[ifv->ifv_vid] = ifv;
545 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
548 trunk->vlans[ifv->ifv_vid] = NULL;
555 vlan_freehash(struct ifvlantrunk *trunk)
560 vlan_inithash(struct ifvlantrunk *trunk)
564 #endif /* !VLAN_ARRAY */
567 trunk_destroy(struct ifvlantrunk *trunk)
571 vlan_freehash(trunk);
572 trunk->parent->if_vlantrunk = NULL;
573 TRUNK_LOCK_DESTROY(trunk);
574 if_rele(trunk->parent);
579 * Program our multicast filter. What we're actually doing is
580 * programming the multicast filter of the parent. This has the
581 * side effect of causing the parent interface to receive multicast
582 * traffic that it doesn't really want, which ends up being discarded
583 * later by the upper protocol layers. Unfortunately, there's no way
584 * to avoid this: there really is only one physical interface.
587 vlan_setmulti(struct ifnet *ifp)
590 struct ifmultiaddr *ifma;
592 struct vlan_mc_entry *mc;
597 /* Find the parent. */
601 CURVNET_SET_QUIET(ifp_p->if_vnet);
603 /* First, remove any existing filter entries. */
604 while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
605 CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
606 (void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
607 NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
610 /* Now program new ones. */
612 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
613 if (ifma->ifma_addr->sa_family != AF_LINK)
615 mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
617 IF_ADDR_WUNLOCK(ifp);
621 bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
622 mc->mc_addr.sdl_index = ifp_p->if_index;
623 CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
625 IF_ADDR_WUNLOCK(ifp);
626 CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
627 error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
640 * A handler for interface ifnet events.
643 vlan_ifevent(void *arg __unused, struct ifnet *ifp, int event)
645 struct epoch_tracker et;
647 struct ifvlantrunk *trunk;
649 if (event != IFNET_EVENT_UPDATE_BAUDRATE)
653 trunk = ifp->if_vlantrunk;
660 VLAN_FOREACH(ifv, trunk) {
661 ifv->ifv_ifp->if_baudrate = ifp->if_baudrate;
663 TRUNK_WUNLOCK(trunk);
668 * A handler for parent interface link layer address changes.
669 * If the parent interface link layer address is changed we
670 * should also change it on all children vlans.
673 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
675 struct epoch_tracker et;
677 struct ifnet *ifv_ifp;
678 struct ifvlantrunk *trunk;
679 struct sockaddr_dl *sdl;
681 /* Need the epoch since this is run on taskqueue_swi. */
683 trunk = ifp->if_vlantrunk;
690 * OK, it's a trunk. Loop over and change all vlan's lladdrs on it.
691 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
692 * ioctl calls on the parent garbling the lladdr of the child vlan.
695 VLAN_FOREACH(ifv, trunk) {
697 * Copy new new lladdr into the ifv_ifp, enqueue a task
698 * to actually call if_setlladdr. if_setlladdr needs to
699 * be deferred to a taskqueue because it will call into
700 * the if_vlan ioctl path and try to acquire the global
703 ifv_ifp = ifv->ifv_ifp;
704 bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
706 sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
707 sdl->sdl_alen = ifp->if_addrlen;
708 taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
710 TRUNK_WUNLOCK(trunk);
715 * A handler for network interface departure events.
716 * Track departure of trunks here so that we don't access invalid
717 * pointers or whatever if a trunk is ripped from under us, e.g.,
718 * by ejecting its hot-plug card. However, if an ifnet is simply
719 * being renamed, then there's no need to tear down the state.
722 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
725 struct ifvlantrunk *trunk;
727 /* If the ifnet is just being renamed, don't do anything. */
728 if (ifp->if_flags & IFF_RENAMING)
731 trunk = ifp->if_vlantrunk;
738 * OK, it's a trunk. Loop over and detach all vlan's on it.
739 * Check trunk pointer after each vlan_unconfig() as it will
740 * free it and set to NULL after the last vlan was detached.
742 VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
743 ifp->if_vlantrunk == NULL)
744 vlan_unconfig_locked(ifv->ifv_ifp, 1);
746 /* Trunk should have been destroyed in vlan_unconfig(). */
747 KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
752 * Return the trunk device for a virtual interface.
754 static struct ifnet *
755 vlan_trunkdev(struct ifnet *ifp)
761 if (ifp->if_type != IFT_L2VLAN)
772 * Return the 12-bit VLAN VID for this interface, for use by external
773 * components such as Infiniband.
775 * XXXRW: Note that the function name here is historical; it should be named
779 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
783 if (ifp->if_type != IFT_L2VLAN)
786 *vidp = ifv->ifv_vid;
791 vlan_pcp(struct ifnet *ifp, uint16_t *pcpp)
795 if (ifp->if_type != IFT_L2VLAN)
798 *pcpp = ifv->ifv_pcp;
803 * Return a driver specific cookie for this interface. Synchronization
804 * with setcookie must be provided by the driver.
807 vlan_cookie(struct ifnet *ifp)
811 if (ifp->if_type != IFT_L2VLAN)
814 return (ifv->ifv_cookie);
818 * Store a cookie in our softc that drivers can use to store driver
819 * private per-instance data in.
822 vlan_setcookie(struct ifnet *ifp, void *cookie)
826 if (ifp->if_type != IFT_L2VLAN)
829 ifv->ifv_cookie = cookie;
834 * Return the vlan device present at the specific VID.
836 static struct ifnet *
837 vlan_devat(struct ifnet *ifp, uint16_t vid)
839 struct ifvlantrunk *trunk;
844 trunk = ifp->if_vlantrunk;
848 ifv = vlan_gethash(trunk, vid);
855 * VLAN support can be loaded as a module. The only place in the
856 * system that's intimately aware of this is ether_input. We hook
857 * into this code through vlan_input_p which is defined there and
858 * set here. No one else in the system should be aware of this so
859 * we use an explicit reference here.
861 extern void (*vlan_input_p)(struct ifnet *, struct mbuf *);
863 /* For if_link_state_change() eyes only... */
864 extern void (*vlan_link_state_p)(struct ifnet *);
867 vlan_modevent(module_t mod, int type, void *data)
872 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
873 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
874 if (ifdetach_tag == NULL)
876 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
877 vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
878 if (iflladdr_tag == NULL)
880 ifevent_tag = EVENTHANDLER_REGISTER(ifnet_event,
881 vlan_ifevent, NULL, EVENTHANDLER_PRI_ANY);
882 if (ifevent_tag == NULL)
885 vlan_input_p = vlan_input;
886 vlan_link_state_p = vlan_link_state;
887 vlan_trunk_cap_p = vlan_trunk_capabilities;
888 vlan_trunkdev_p = vlan_trunkdev;
889 vlan_cookie_p = vlan_cookie;
890 vlan_setcookie_p = vlan_setcookie;
891 vlan_tag_p = vlan_tag;
892 vlan_pcp_p = vlan_pcp;
893 vlan_devat_p = vlan_devat;
895 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
896 vlan_clone_create, vlan_clone_destroy);
899 printf("vlan: initialized, using "
903 "hash tables with chaining"
910 if_clone_detach(vlan_cloner);
912 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
913 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
914 EVENTHANDLER_DEREGISTER(ifnet_event, ifevent_tag);
916 vlan_link_state_p = NULL;
917 vlan_trunk_cap_p = NULL;
918 vlan_trunkdev_p = NULL;
920 vlan_cookie_p = NULL;
921 vlan_setcookie_p = NULL;
923 VLAN_LOCKING_DESTROY();
925 printf("vlan: unloaded\n");
933 static moduledata_t vlan_mod = {
939 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
940 MODULE_VERSION(if_vlan, 3);
944 vnet_vlan_init(const void *unused __unused)
947 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
948 vlan_clone_create, vlan_clone_destroy);
949 V_vlan_cloner = vlan_cloner;
951 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
952 vnet_vlan_init, NULL);
955 vnet_vlan_uninit(const void *unused __unused)
958 if_clone_detach(V_vlan_cloner);
960 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
961 vnet_vlan_uninit, NULL);
965 * Check for <etherif>.<vlan>[.<vlan> ...] style interface names.
967 static struct ifnet *
968 vlan_clone_match_ethervid(const char *name, int *vidp)
970 char ifname[IFNAMSIZ];
975 strlcpy(ifname, name, IFNAMSIZ);
976 if ((cp = strrchr(ifname, '.')) == NULL)
979 if ((ifp = ifunit_ref(ifname)) == NULL)
987 for(; *cp >= '0' && *cp <= '9'; cp++)
988 vid = (vid * 10) + (*cp - '0');
1000 vlan_clone_match(struct if_clone *ifc, const char *name)
1005 ifp = vlan_clone_match_ethervid(name, NULL);
1011 if (strncmp(vlanname, name, strlen(vlanname)) != 0)
1013 for (cp = name + 4; *cp != '\0'; cp++) {
1014 if (*cp < '0' || *cp > '9')
1022 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
1025 bool wildcard = false;
1026 bool subinterface = false;
1030 uint16_t proto = ETHERTYPE_VLAN;
1033 struct ifnet *p = NULL;
1035 struct sockaddr_dl *sdl;
1037 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
1041 * There are three ways to specify the cloned device:
1042 * o pass a parameter block with the clone request.
1043 * o specify parameters in the text of the clone device name
1044 * o specify no parameters and get an unattached device that
1045 * must be configured separately.
1046 * The first technique is preferred; the latter two are supported
1047 * for backwards compatibility.
1049 * XXXRW: Note historic use of the word "tag" here. New ioctls may be
1054 error = copyin(params, &vlr, sizeof(vlr));
1058 proto = vlr.vlr_proto;
1060 #ifdef COMPAT_FREEBSD12
1062 proto = ETHERTYPE_VLAN;
1064 p = ifunit_ref(vlr.vlr_parent);
1069 if ((error = ifc_name2unit(name, &unit)) == 0) {
1072 * vlanX interface. Set wildcard to true if the unit number
1075 wildcard = (unit < 0);
1077 struct ifnet *p_tmp = vlan_clone_match_ethervid(name, &vid);
1078 if (p_tmp != NULL) {
1080 subinterface = true;
1081 unit = IF_DUNIT_NONE;
1099 if (!subinterface) {
1100 /* vlanX interface, mark X as busy or allocate new unit # */
1101 error = ifc_alloc_unit(ifc, &unit);
1109 /* In the wildcard case, we need to update the name. */
1111 for (dp = name; *dp != '\0'; dp++);
1112 if (snprintf(dp, len - (dp-name), "%d", unit) >
1113 len - (dp-name) - 1) {
1114 panic("%s: interface name too long", __func__);
1118 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
1119 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
1122 ifc_free_unit(ifc, unit);
1128 CK_SLIST_INIT(&ifv->vlan_mc_listhead);
1129 ifp->if_softc = ifv;
1131 * Set the name manually rather than using if_initname because
1132 * we don't conform to the default naming convention for interfaces.
1134 strlcpy(ifp->if_xname, name, IFNAMSIZ);
1135 ifp->if_dname = vlanname;
1136 ifp->if_dunit = unit;
1138 ifp->if_init = vlan_init;
1140 ifp->if_start = vlan_altq_start;
1141 ifp->if_transmit = vlan_altq_transmit;
1142 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
1143 ifp->if_snd.ifq_drv_maxlen = 0;
1144 IFQ_SET_READY(&ifp->if_snd);
1146 ifp->if_transmit = vlan_transmit;
1148 ifp->if_qflush = vlan_qflush;
1149 ifp->if_ioctl = vlan_ioctl;
1150 #if defined(KERN_TLS) || defined(RATELIMIT)
1151 ifp->if_snd_tag_alloc = vlan_snd_tag_alloc;
1152 ifp->if_snd_tag_modify = vlan_snd_tag_modify;
1153 ifp->if_snd_tag_query = vlan_snd_tag_query;
1154 ifp->if_snd_tag_free = vlan_snd_tag_free;
1155 ifp->if_next_snd_tag = vlan_next_snd_tag;
1156 ifp->if_ratelimit_query = vlan_ratelimit_query;
1158 ifp->if_flags = VLAN_IFFLAGS;
1159 ether_ifattach(ifp, eaddr);
1160 /* Now undo some of the damage... */
1161 ifp->if_baudrate = 0;
1162 ifp->if_type = IFT_L2VLAN;
1163 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
1165 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1166 sdl->sdl_type = IFT_L2VLAN;
1169 error = vlan_config(ifv, p, vid, proto);
1173 * Since we've partially failed, we need to back
1174 * out all the way, otherwise userland could get
1175 * confused. Thus, we destroy the interface.
1177 ether_ifdetach(ifp);
1181 ifc_free_unit(ifc, unit);
1192 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1194 struct ifvlan *ifv = ifp->if_softc;
1195 int unit = ifp->if_dunit;
1197 if (ifp->if_vlantrunk)
1201 IFQ_PURGE(&ifp->if_snd);
1203 ether_ifdetach(ifp); /* first, remove it from system-wide lists */
1204 vlan_unconfig(ifp); /* now it can be unconfigured and freed */
1206 * We should have the only reference to the ifv now, so we can now
1207 * drain any remaining lladdr task before freeing the ifnet and the
1210 taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
1214 if (unit != IF_DUNIT_NONE)
1215 ifc_free_unit(ifc, unit);
1221 * The ifp->if_init entry point for vlan(4) is a no-op.
1224 vlan_init(void *foo __unused)
1229 * The if_transmit method for vlan(4) interface.
1232 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1236 int error, len, mcast;
1240 ifv = ifp->if_softc;
1241 if (TRUNK(ifv) == NULL) {
1242 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1247 len = m->m_pkthdr.len;
1248 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1252 #if defined(KERN_TLS) || defined(RATELIMIT)
1253 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
1254 struct vlan_snd_tag *vst;
1255 struct m_snd_tag *mst;
1257 MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
1258 mst = m->m_pkthdr.snd_tag;
1259 vst = mst_to_vst(mst);
1260 if (vst->tag->ifp != p) {
1261 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1266 m->m_pkthdr.snd_tag = m_snd_tag_ref(vst->tag);
1267 m_snd_tag_rele(mst);
1272 * Do not run parent's if_transmit() if the parent is not up,
1273 * or parent's driver will cause a system crash.
1275 if (!UP_AND_RUNNING(p)) {
1276 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1281 if (!ether_8021q_frame(&m, ifp, p, &ifv->ifv_qtag)) {
1282 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1287 * Send it, precisely as ether_output() would have.
1289 error = (p->if_transmit)(p, m);
1291 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1292 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1293 if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1295 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1300 vlan_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1309 * Find the first non-VLAN parent interface.
1311 ifv = ifp->if_softc;
1313 if (TRUNK(ifv) == NULL) {
1319 } while (p->if_type == IFT_L2VLAN);
1321 return p->if_output(ifp, m, dst, ro);
1326 vlan_altq_start(if_t ifp)
1328 struct ifaltq *ifq = &ifp->if_snd;
1332 IFQ_DEQUEUE_NOLOCK(ifq, m);
1334 vlan_transmit(ifp, m);
1335 IFQ_DEQUEUE_NOLOCK(ifq, m);
1341 vlan_altq_transmit(if_t ifp, struct mbuf *m)
1345 if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
1346 IFQ_ENQUEUE(&ifp->if_snd, m, err);
1348 vlan_altq_start(ifp);
1350 err = vlan_transmit(ifp, m);
1357 * The ifp->if_qflush entry point for vlan(4) is a no-op.
1360 vlan_qflush(struct ifnet *ifp __unused)
1365 vlan_input(struct ifnet *ifp, struct mbuf *m)
1367 struct ifvlantrunk *trunk;
1374 trunk = ifp->if_vlantrunk;
1375 if (trunk == NULL) {
1380 if (m->m_flags & M_VLANTAG) {
1382 * Packet is tagged, but m contains a normal
1383 * Ethernet frame; the tag is stored out-of-band.
1385 tag = m->m_pkthdr.ether_vtag;
1386 m->m_flags &= ~M_VLANTAG;
1388 struct ether_vlan_header *evl;
1391 * Packet is tagged in-band as specified by 802.1q.
1393 switch (ifp->if_type) {
1395 if (m->m_len < sizeof(*evl) &&
1396 (m = m_pullup(m, sizeof(*evl))) == NULL) {
1397 if_printf(ifp, "cannot pullup VLAN header\n");
1400 evl = mtod(m, struct ether_vlan_header *);
1401 tag = ntohs(evl->evl_tag);
1404 * Remove the 802.1q header by copying the Ethernet
1405 * addresses over it and adjusting the beginning of
1406 * the data in the mbuf. The encapsulated Ethernet
1407 * type field is already in place.
1409 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1410 ETHER_HDR_LEN - ETHER_TYPE_LEN);
1411 m_adj(m, ETHER_VLAN_ENCAP_LEN);
1416 panic("%s: %s has unsupported if_type %u",
1417 __func__, ifp->if_xname, ifp->if_type);
1419 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1425 vid = EVL_VLANOFTAG(tag);
1427 ifv = vlan_gethash(trunk, vid);
1428 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1429 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1434 if (vlan_mtag_pcp) {
1436 * While uncommon, it is possible that we will find a 802.1q
1437 * packet encapsulated inside another packet that also had an
1438 * 802.1q header. For example, ethernet tunneled over IPSEC
1439 * arriving over ethernet. In that case, we replace the
1440 * existing 802.1q PCP m_tag value.
1442 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1444 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1445 sizeof(uint8_t), M_NOWAIT);
1447 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1451 m_tag_prepend(m, mtag);
1453 *(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1456 m->m_pkthdr.rcvif = ifv->ifv_ifp;
1457 if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1459 /* Pass it back through the parent's input routine. */
1460 (*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
1464 vlan_lladdr_fn(void *arg, int pending __unused)
1469 ifv = (struct ifvlan *)arg;
1472 CURVNET_SET(ifp->if_vnet);
1474 /* The ifv_ifp already has the lladdr copied in. */
1475 if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1481 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid,
1484 struct epoch_tracker et;
1485 struct ifvlantrunk *trunk;
1490 * We can handle non-ethernet hardware types as long as
1491 * they handle the tagging and headers themselves.
1493 if (p->if_type != IFT_ETHER &&
1494 p->if_type != IFT_L2VLAN &&
1495 (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1496 return (EPROTONOSUPPORT);
1497 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1498 return (EPROTONOSUPPORT);
1500 * Don't let the caller set up a VLAN VID with
1501 * anything except VLID bits.
1502 * VID numbers 0x0 and 0xFFF are reserved.
1504 if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1510 if (p->if_vlantrunk == NULL) {
1511 trunk = malloc(sizeof(struct ifvlantrunk),
1512 M_VLAN, M_WAITOK | M_ZERO);
1513 vlan_inithash(trunk);
1514 TRUNK_LOCK_INIT(trunk);
1516 p->if_vlantrunk = trunk;
1518 if_ref(trunk->parent);
1519 TRUNK_WUNLOCK(trunk);
1521 trunk = p->if_vlantrunk;
1524 ifv->ifv_vid = vid; /* must set this before vlan_inshash() */
1525 ifv->ifv_pcp = 0; /* Default: best effort delivery. */
1526 error = vlan_inshash(trunk, ifv);
1529 ifv->ifv_proto = proto;
1530 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1531 ifv->ifv_mintu = ETHERMIN;
1532 ifv->ifv_pflags = 0;
1533 ifv->ifv_capenable = -1;
1536 * If the parent supports the VLAN_MTU capability,
1537 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1540 if (p->if_capenable & IFCAP_VLAN_MTU) {
1542 * No need to fudge the MTU since the parent can
1543 * handle extended frames.
1545 ifv->ifv_mtufudge = 0;
1548 * Fudge the MTU by the encapsulation size. This
1549 * makes us incompatible with strictly compliant
1550 * 802.1Q implementations, but allows us to use
1551 * the feature with other NetBSD implementations,
1552 * which might still be useful.
1554 ifv->ifv_mtufudge = ifv->ifv_encaplen;
1557 ifv->ifv_trunk = trunk;
1560 * Initialize fields from our parent. This duplicates some
1561 * work with ether_ifattach() but allows for non-ethernet
1562 * interfaces to also work.
1564 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1565 ifp->if_baudrate = p->if_baudrate;
1566 ifp->if_input = p->if_input;
1567 ifp->if_resolvemulti = p->if_resolvemulti;
1568 ifp->if_addrlen = p->if_addrlen;
1569 ifp->if_broadcastaddr = p->if_broadcastaddr;
1570 ifp->if_pcp = ifv->ifv_pcp;
1573 * We wrap the parent's if_output using vlan_output to ensure that it
1574 * can't become stale.
1576 ifp->if_output = vlan_output;
1579 * Copy only a selected subset of flags from the parent.
1580 * Other flags are none of our business.
1582 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1583 ifp->if_flags &= ~VLAN_COPY_FLAGS;
1584 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1585 #undef VLAN_COPY_FLAGS
1587 ifp->if_link_state = p->if_link_state;
1589 NET_EPOCH_ENTER(et);
1590 vlan_capabilities(ifv);
1594 * Set up our interface address to reflect the underlying
1595 * physical interface's.
1597 TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
1598 ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1602 * Do not schedule link address update if it was the same
1603 * as previous parent's. This helps avoid updating for each
1604 * associated llentry.
1606 if (memcmp(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen) != 0) {
1607 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1608 taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
1611 /* We are ready for operation now. */
1612 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1614 /* Update flags on the parent, if necessary. */
1615 vlan_setflags(ifp, 1);
1618 * Configure multicast addresses that may already be
1619 * joined on the vlan device.
1621 (void)vlan_setmulti(ifp);
1625 EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1632 vlan_unconfig(struct ifnet *ifp)
1636 vlan_unconfig_locked(ifp, 0);
1641 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1643 struct ifvlantrunk *trunk;
1644 struct vlan_mc_entry *mc;
1646 struct ifnet *parent;
1649 VLAN_XLOCK_ASSERT();
1651 ifv = ifp->if_softc;
1652 trunk = ifv->ifv_trunk;
1655 if (trunk != NULL) {
1656 parent = trunk->parent;
1659 * Since the interface is being unconfigured, we need to
1660 * empty the list of multicast groups that we may have joined
1661 * while we were alive from the parent's list.
1663 while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1665 * If the parent interface is being detached,
1666 * all its multicast addresses have already
1667 * been removed. Warn about errors if
1668 * if_delmulti() does fail, but don't abort as
1669 * all callers expect vlan destruction to
1673 error = if_delmulti(parent,
1674 (struct sockaddr *)&mc->mc_addr);
1677 "Failed to delete multicast address from parent: %d\n",
1680 CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1681 NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
1684 vlan_setflags(ifp, 0); /* clear special flags on parent */
1686 vlan_remhash(trunk, ifv);
1687 ifv->ifv_trunk = NULL;
1690 * Check if we were the last.
1692 if (trunk->refcnt == 0) {
1693 parent->if_vlantrunk = NULL;
1695 trunk_destroy(trunk);
1699 /* Disconnect from parent. */
1700 if (ifv->ifv_pflags)
1701 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1702 ifp->if_mtu = ETHERMTU;
1703 ifp->if_link_state = LINK_STATE_UNKNOWN;
1704 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1707 * Only dispatch an event if vlan was
1708 * attached, otherwise there is nothing
1709 * to cleanup anyway.
1712 EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1715 /* Handle a reference counted flag that should be set on the parent as well */
1717 vlan_setflag(struct ifnet *ifp, int flag, int status,
1718 int (*func)(struct ifnet *, int))
1723 VLAN_SXLOCK_ASSERT();
1725 ifv = ifp->if_softc;
1726 status = status ? (ifp->if_flags & flag) : 0;
1727 /* Now "status" contains the flag value or 0 */
1730 * See if recorded parent's status is different from what
1731 * we want it to be. If it is, flip it. We record parent's
1732 * status in ifv_pflags so that we won't clear parent's flag
1733 * we haven't set. In fact, we don't clear or set parent's
1734 * flags directly, but get or release references to them.
1735 * That's why we can be sure that recorded flags still are
1736 * in accord with actual parent's flags.
1738 if (status != (ifv->ifv_pflags & flag)) {
1739 error = (*func)(PARENT(ifv), status);
1742 ifv->ifv_pflags &= ~flag;
1743 ifv->ifv_pflags |= status;
1749 * Handle IFF_* flags that require certain changes on the parent:
1750 * if "status" is true, update parent's flags respective to our if_flags;
1751 * if "status" is false, forcedly clear the flags set on parent.
1754 vlan_setflags(struct ifnet *ifp, int status)
1758 for (i = 0; vlan_pflags[i].flag; i++) {
1759 error = vlan_setflag(ifp, vlan_pflags[i].flag,
1760 status, vlan_pflags[i].func);
1767 /* Inform all vlans that their parent has changed link state */
1769 vlan_link_state(struct ifnet *ifp)
1771 struct epoch_tracker et;
1772 struct ifvlantrunk *trunk;
1775 NET_EPOCH_ENTER(et);
1776 trunk = ifp->if_vlantrunk;
1777 if (trunk == NULL) {
1783 VLAN_FOREACH(ifv, trunk) {
1784 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1785 if_link_state_change(ifv->ifv_ifp,
1786 trunk->parent->if_link_state);
1788 TRUNK_WUNLOCK(trunk);
1793 vlan_capabilities(struct ifvlan *ifv)
1797 struct ifnet_hw_tsomax hw_tsomax;
1798 int cap = 0, ena = 0, mena;
1802 VLAN_SXLOCK_ASSERT();
1807 /* Mask parent interface enabled capabilities disabled by user. */
1808 mena = p->if_capenable & ifv->ifv_capenable;
1811 * If the parent interface can do checksum offloading
1812 * on VLANs, then propagate its hardware-assisted
1813 * checksumming flags. Also assert that checksum
1814 * offloading requires hardware VLAN tagging.
1816 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1817 cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1818 if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1819 p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1820 ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1821 if (ena & IFCAP_TXCSUM)
1822 hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
1823 CSUM_UDP | CSUM_SCTP);
1824 if (ena & IFCAP_TXCSUM_IPV6)
1825 hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
1826 CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
1830 * If the parent interface can do TSO on VLANs then
1831 * propagate the hardware-assisted flag. TSO on VLANs
1832 * does not necessarily require hardware VLAN tagging.
1834 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1835 if_hw_tsomax_common(p, &hw_tsomax);
1836 if_hw_tsomax_update(ifp, &hw_tsomax);
1837 if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1838 cap |= p->if_capabilities & IFCAP_TSO;
1839 if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1840 ena |= mena & IFCAP_TSO;
1841 if (ena & IFCAP_TSO)
1842 hwa |= p->if_hwassist & CSUM_TSO;
1846 * If the parent interface can do LRO and checksum offloading on
1847 * VLANs, then guess it may do LRO on VLANs. False positive here
1848 * cost nothing, while false negative may lead to some confusions.
1850 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1851 cap |= p->if_capabilities & IFCAP_LRO;
1852 if (p->if_capenable & IFCAP_VLAN_HWCSUM)
1853 ena |= p->if_capenable & IFCAP_LRO;
1856 * If the parent interface can offload TCP connections over VLANs then
1857 * propagate its TOE capability to the VLAN interface.
1859 * All TOE drivers in the tree today can deal with VLANs. If this
1860 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1863 #define IFCAP_VLAN_TOE IFCAP_TOE
1864 if (p->if_capabilities & IFCAP_VLAN_TOE)
1865 cap |= p->if_capabilities & IFCAP_TOE;
1866 if (p->if_capenable & IFCAP_VLAN_TOE) {
1867 TOEDEV(ifp) = TOEDEV(p);
1868 ena |= mena & IFCAP_TOE;
1872 * If the parent interface supports dynamic link state, so does the
1875 cap |= (p->if_capabilities & IFCAP_LINKSTATE);
1876 ena |= (mena & IFCAP_LINKSTATE);
1880 * If the parent interface supports ratelimiting, so does the
1883 cap |= (p->if_capabilities & IFCAP_TXRTLMT);
1884 ena |= (mena & IFCAP_TXRTLMT);
1888 * If the parent interface supports unmapped mbufs, so does
1889 * the VLAN interface. Note that this should be fine even for
1890 * interfaces that don't support hardware tagging as headers
1891 * are prepended in normal mbufs to unmapped mbufs holding
1894 cap |= (p->if_capabilities & IFCAP_MEXTPG);
1895 ena |= (mena & IFCAP_MEXTPG);
1898 * If the parent interface can offload encryption and segmentation
1899 * of TLS records over TCP, propagate it's capability to the VLAN
1902 * All TLS drivers in the tree today can deal with VLANs. If
1903 * this ever changes, then a new IFCAP_VLAN_TXTLS can be
1906 if (p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
1907 cap |= p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
1908 if (p->if_capenable & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
1909 ena |= mena & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
1911 ifp->if_capabilities = cap;
1912 ifp->if_capenable = ena;
1913 ifp->if_hwassist = hwa;
1917 vlan_trunk_capabilities(struct ifnet *ifp)
1919 struct epoch_tracker et;
1920 struct ifvlantrunk *trunk;
1924 trunk = ifp->if_vlantrunk;
1925 if (trunk == NULL) {
1929 NET_EPOCH_ENTER(et);
1930 VLAN_FOREACH(ifv, trunk)
1931 vlan_capabilities(ifv);
1937 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1943 struct ifvlantrunk *trunk;
1945 int error = 0, oldmtu;
1947 ifr = (struct ifreq *)data;
1948 ifa = (struct ifaddr *) data;
1949 ifv = ifp->if_softc;
1953 ifp->if_flags |= IFF_UP;
1955 if (ifa->ifa_addr->sa_family == AF_INET)
1956 arp_ifinit(ifp, ifa);
1960 bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
1965 if (TRUNK(ifv) != NULL) {
1968 error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1970 /* Limit the result to the parent's current config. */
1972 struct ifmediareq *ifmr;
1974 ifmr = (struct ifmediareq *)data;
1975 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1976 ifmr->ifm_count = 1;
1977 error = copyout(&ifmr->ifm_current,
1994 * Set the interface MTU.
1998 if (trunk != NULL) {
2001 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
2003 (ifv->ifv_mintu - ifv->ifv_mtufudge))
2006 ifp->if_mtu = ifr->ifr_mtu;
2007 TRUNK_WUNLOCK(trunk);
2016 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
2017 * interface to be delegated to a jail without allowing the
2018 * jail to change what underlying interface/VID it is
2019 * associated with. We are not entirely convinced that this
2020 * is the right way to accomplish that policy goal.
2022 if (ifp->if_vnet != ifp->if_home_vnet) {
2027 error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
2030 if (vlr.vlr_parent[0] == '\0') {
2034 p = ifunit_ref(vlr.vlr_parent);
2039 #ifdef COMPAT_FREEBSD12
2040 if (vlr.vlr_proto == 0)
2041 vlr.vlr_proto = ETHERTYPE_VLAN;
2043 oldmtu = ifp->if_mtu;
2044 error = vlan_config(ifv, p, vlr.vlr_tag, vlr.vlr_proto);
2048 * VLAN MTU may change during addition of the vlandev.
2049 * If it did, do network layer specific procedure.
2051 if (ifp->if_mtu != oldmtu) {
2061 if (ifp->if_vnet != ifp->if_home_vnet) {
2066 bzero(&vlr, sizeof(vlr));
2068 if (TRUNK(ifv) != NULL) {
2069 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
2070 sizeof(vlr.vlr_parent));
2071 vlr.vlr_tag = ifv->ifv_vid;
2072 vlr.vlr_proto = ifv->ifv_proto;
2075 error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
2080 * We should propagate selected flags to the parent,
2081 * e.g., promiscuous mode.
2084 if (TRUNK(ifv) != NULL)
2085 error = vlan_setflags(ifp, 1);
2092 * If we don't have a parent, just remember the membership for
2095 * XXX We need the rmlock here to avoid sleeping while
2096 * holding in6_multi_mtx.
2101 error = vlan_setmulti(ifp);
2107 if (ifp->if_vnet != ifp->if_home_vnet) {
2112 ifr->ifr_vlan_pcp = ifv->ifv_pcp;
2117 if (ifp->if_vnet != ifp->if_home_vnet) {
2122 error = priv_check(curthread, PRIV_NET_SETVLANPCP);
2125 if (ifr->ifr_vlan_pcp > VLAN_PCP_MAX) {
2129 ifv->ifv_pcp = ifr->ifr_vlan_pcp;
2130 ifp->if_pcp = ifv->ifv_pcp;
2131 /* broadcast event about PCP change */
2132 EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
2137 ifv->ifv_capenable = ifr->ifr_reqcap;
2139 if (trunk != NULL) {
2140 struct epoch_tracker et;
2142 NET_EPOCH_ENTER(et);
2143 vlan_capabilities(ifv);
2157 #if defined(KERN_TLS) || defined(RATELIMIT)
2159 vlan_snd_tag_alloc(struct ifnet *ifp,
2160 union if_snd_tag_alloc_params *params,
2161 struct m_snd_tag **ppmt)
2163 struct epoch_tracker et;
2164 struct vlan_snd_tag *vst;
2166 struct ifnet *parent;
2169 NET_EPOCH_ENTER(et);
2170 ifv = ifp->if_softc;
2171 if (ifv->ifv_trunk != NULL)
2172 parent = PARENT(ifv);
2175 if (parent == NULL) {
2177 return (EOPNOTSUPP);
2182 vst = malloc(sizeof(*vst), M_VLAN, M_NOWAIT);
2188 error = m_snd_tag_alloc(parent, params, &vst->tag);
2195 m_snd_tag_init(&vst->com, ifp, vst->tag->type);
2201 static struct m_snd_tag *
2202 vlan_next_snd_tag(struct m_snd_tag *mst)
2204 struct vlan_snd_tag *vst;
2206 vst = mst_to_vst(mst);
2211 vlan_snd_tag_modify(struct m_snd_tag *mst,
2212 union if_snd_tag_modify_params *params)
2214 struct vlan_snd_tag *vst;
2216 vst = mst_to_vst(mst);
2217 return (vst->tag->ifp->if_snd_tag_modify(vst->tag, params));
2221 vlan_snd_tag_query(struct m_snd_tag *mst,
2222 union if_snd_tag_query_params *params)
2224 struct vlan_snd_tag *vst;
2226 vst = mst_to_vst(mst);
2227 return (vst->tag->ifp->if_snd_tag_query(vst->tag, params));
2231 vlan_snd_tag_free(struct m_snd_tag *mst)
2233 struct vlan_snd_tag *vst;
2235 vst = mst_to_vst(mst);
2236 m_snd_tag_rele(vst->tag);
2241 vlan_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
2244 * For vlan, we have an indirect
2245 * interface. The caller needs to
2246 * get a ratelimit tag on the actual
2247 * interface the flow will go on.
2249 q->rate_table = NULL;
2250 q->flags = RT_IS_INDIRECT;
2252 q->number_of_rates = 0;