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;
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).
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.
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.
242 #define _VLAN_SX_ID ifv_sx
244 static struct sx _VLAN_SX_ID;
246 #define VLAN_LOCKING_INIT() \
247 sx_init_flags(&_VLAN_SX_ID, "vlan_sx", SX_RECURSE)
249 #define VLAN_LOCKING_DESTROY() \
250 sx_destroy(&_VLAN_SX_ID)
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)
261 * We also have a per-trunk mutex that should be acquired when changing
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);
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.
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,
285 static void trunk_destroy(struct ifvlantrunk *trunk);
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 static struct m_snd_tag *vlan_next_snd_tag(struct m_snd_tag *);
299 static void vlan_ratelimit_query(struct ifnet *,
300 struct if_ratelimit_query_results *);
302 static void vlan_qflush(struct ifnet *ifp);
303 static int vlan_setflag(struct ifnet *ifp, int flag, int status,
304 int (*func)(struct ifnet *, int));
305 static int vlan_setflags(struct ifnet *ifp, int status);
306 static int vlan_setmulti(struct ifnet *ifp);
307 static int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
308 static int vlan_output(struct ifnet *ifp, struct mbuf *m,
309 const struct sockaddr *dst, struct route *ro);
310 static void vlan_unconfig(struct ifnet *ifp);
311 static void vlan_unconfig_locked(struct ifnet *ifp, int departing);
312 static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag,
314 static void vlan_link_state(struct ifnet *ifp);
315 static void vlan_capabilities(struct ifvlan *ifv);
316 static void vlan_trunk_capabilities(struct ifnet *ifp);
318 static struct ifnet *vlan_clone_match_ethervid(const char *, int *);
319 static int vlan_clone_match(struct if_clone *, const char *);
320 static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
321 static int vlan_clone_destroy(struct if_clone *, struct ifnet *);
323 static void vlan_ifdetach(void *arg, struct ifnet *ifp);
324 static void vlan_iflladdr(void *arg, struct ifnet *ifp);
326 static void vlan_lladdr_fn(void *arg, int pending);
328 static struct if_clone *vlan_cloner;
331 VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner);
332 #define V_vlan_cloner VNET(vlan_cloner)
336 vlan_mc_free(struct epoch_context *ctx)
338 struct vlan_mc_entry *mc = __containerof(ctx, struct vlan_mc_entry, mc_epoch_ctx);
343 #define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
346 vlan_inithash(struct ifvlantrunk *trunk)
351 * The trunk must not be locked here since we call malloc(M_WAITOK).
352 * It is OK in case this function is called before the trunk struct
353 * gets hooked up and becomes visible from other threads.
356 KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
357 ("%s: hash already initialized", __func__));
359 trunk->hwidth = VLAN_DEF_HWIDTH;
360 n = 1 << trunk->hwidth;
361 trunk->hmask = n - 1;
362 trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
363 for (i = 0; i < n; i++)
364 CK_SLIST_INIT(&trunk->hash[i]);
368 vlan_freehash(struct ifvlantrunk *trunk)
373 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
374 for (i = 0; i < (1 << trunk->hwidth); i++)
375 KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]),
376 ("%s: hash table not empty", __func__));
378 free(trunk->hash, M_VLAN);
380 trunk->hwidth = trunk->hmask = 0;
384 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
390 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
392 b = 1 << trunk->hwidth;
393 i = HASH(ifv->ifv_vid, trunk->hmask);
394 CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
395 if (ifv->ifv_vid == ifv2->ifv_vid)
399 * Grow the hash when the number of vlans exceeds half of the number of
400 * hash buckets squared. This will make the average linked-list length
403 if (trunk->refcnt > (b * b) / 2) {
404 vlan_growhash(trunk, 1);
405 i = HASH(ifv->ifv_vid, trunk->hmask);
407 CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
414 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
420 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
422 b = 1 << trunk->hwidth;
423 i = HASH(ifv->ifv_vid, trunk->hmask);
424 CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
427 CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list);
428 if (trunk->refcnt < (b * b) / 2)
429 vlan_growhash(trunk, -1);
433 panic("%s: vlan not found\n", __func__);
434 return (ENOENT); /*NOTREACHED*/
438 * Grow the hash larger or smaller if memory permits.
441 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
444 struct ifvlanhead *hash2;
445 int hwidth2, i, j, n, n2;
448 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
451 /* Harmless yet obvious coding error */
452 printf("%s: howmuch is 0\n", __func__);
456 hwidth2 = trunk->hwidth + howmuch;
457 n = 1 << trunk->hwidth;
459 /* Do not shrink the table below the default */
460 if (hwidth2 < VLAN_DEF_HWIDTH)
463 hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK);
465 printf("%s: out of memory -- hash size not changed\n",
467 return; /* We can live with the old hash table */
469 for (j = 0; j < n2; j++)
470 CK_SLIST_INIT(&hash2[j]);
471 for (i = 0; i < n; i++)
472 while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) {
473 CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list);
474 j = HASH(ifv->ifv_vid, n2 - 1);
475 CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
478 free(trunk->hash, M_VLAN);
480 trunk->hwidth = hwidth2;
481 trunk->hmask = n2 - 1;
484 if_printf(trunk->parent,
485 "VLAN hash table resized from %d to %d buckets\n", n, n2);
488 static __inline struct ifvlan *
489 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
495 CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
496 if (ifv->ifv_vid == vid)
502 /* Debugging code to view the hashtables. */
504 vlan_dumphash(struct ifvlantrunk *trunk)
509 for (i = 0; i < (1 << trunk->hwidth); i++) {
511 CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
512 printf("%s ", ifv->ifv_ifp->if_xname);
519 static __inline struct ifvlan *
520 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
523 return trunk->vlans[vid];
527 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
530 if (trunk->vlans[ifv->ifv_vid] != NULL)
532 trunk->vlans[ifv->ifv_vid] = ifv;
539 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
542 trunk->vlans[ifv->ifv_vid] = NULL;
549 vlan_freehash(struct ifvlantrunk *trunk)
554 vlan_inithash(struct ifvlantrunk *trunk)
558 #endif /* !VLAN_ARRAY */
561 trunk_destroy(struct ifvlantrunk *trunk)
565 vlan_freehash(trunk);
566 trunk->parent->if_vlantrunk = NULL;
567 TRUNK_LOCK_DESTROY(trunk);
568 if_rele(trunk->parent);
573 * Program our multicast filter. What we're actually doing is
574 * programming the multicast filter of the parent. This has the
575 * side effect of causing the parent interface to receive multicast
576 * traffic that it doesn't really want, which ends up being discarded
577 * later by the upper protocol layers. Unfortunately, there's no way
578 * to avoid this: there really is only one physical interface.
581 vlan_setmulti(struct ifnet *ifp)
584 struct ifmultiaddr *ifma;
586 struct vlan_mc_entry *mc;
591 /* Find the parent. */
595 CURVNET_SET_QUIET(ifp_p->if_vnet);
597 /* First, remove any existing filter entries. */
598 while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
599 CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
600 (void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
601 NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
604 /* Now program new ones. */
606 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
607 if (ifma->ifma_addr->sa_family != AF_LINK)
609 mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
611 IF_ADDR_WUNLOCK(ifp);
614 bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
615 mc->mc_addr.sdl_index = ifp_p->if_index;
616 CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
618 IF_ADDR_WUNLOCK(ifp);
619 CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
620 error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
631 * A handler for parent interface link layer address changes.
632 * If the parent interface link layer address is changed we
633 * should also change it on all children vlans.
636 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
638 struct epoch_tracker et;
640 struct ifnet *ifv_ifp;
641 struct ifvlantrunk *trunk;
642 struct sockaddr_dl *sdl;
644 /* Need the epoch since this is run on taskqueue_swi. */
646 trunk = ifp->if_vlantrunk;
653 * OK, it's a trunk. Loop over and change all vlan's lladdrs on it.
654 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
655 * ioctl calls on the parent garbling the lladdr of the child vlan.
658 VLAN_FOREACH(ifv, trunk) {
660 * Copy new new lladdr into the ifv_ifp, enqueue a task
661 * to actually call if_setlladdr. if_setlladdr needs to
662 * be deferred to a taskqueue because it will call into
663 * the if_vlan ioctl path and try to acquire the global
666 ifv_ifp = ifv->ifv_ifp;
667 bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
669 sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
670 sdl->sdl_alen = ifp->if_addrlen;
671 taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
673 TRUNK_WUNLOCK(trunk);
678 * A handler for network interface departure events.
679 * Track departure of trunks here so that we don't access invalid
680 * pointers or whatever if a trunk is ripped from under us, e.g.,
681 * by ejecting its hot-plug card. However, if an ifnet is simply
682 * being renamed, then there's no need to tear down the state.
685 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
688 struct ifvlantrunk *trunk;
690 /* If the ifnet is just being renamed, don't do anything. */
691 if (ifp->if_flags & IFF_RENAMING)
694 trunk = ifp->if_vlantrunk;
701 * OK, it's a trunk. Loop over and detach all vlan's on it.
702 * Check trunk pointer after each vlan_unconfig() as it will
703 * free it and set to NULL after the last vlan was detached.
705 VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
706 ifp->if_vlantrunk == NULL)
707 vlan_unconfig_locked(ifv->ifv_ifp, 1);
709 /* Trunk should have been destroyed in vlan_unconfig(). */
710 KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
715 * Return the trunk device for a virtual interface.
717 static struct ifnet *
718 vlan_trunkdev(struct ifnet *ifp)
724 if (ifp->if_type != IFT_L2VLAN)
735 * Return the 12-bit VLAN VID for this interface, for use by external
736 * components such as Infiniband.
738 * XXXRW: Note that the function name here is historical; it should be named
742 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
746 if (ifp->if_type != IFT_L2VLAN)
749 *vidp = ifv->ifv_vid;
754 vlan_pcp(struct ifnet *ifp, uint16_t *pcpp)
758 if (ifp->if_type != IFT_L2VLAN)
761 *pcpp = ifv->ifv_pcp;
766 * Return a driver specific cookie for this interface. Synchronization
767 * with setcookie must be provided by the driver.
770 vlan_cookie(struct ifnet *ifp)
774 if (ifp->if_type != IFT_L2VLAN)
777 return (ifv->ifv_cookie);
781 * Store a cookie in our softc that drivers can use to store driver
782 * private per-instance data in.
785 vlan_setcookie(struct ifnet *ifp, void *cookie)
789 if (ifp->if_type != IFT_L2VLAN)
792 ifv->ifv_cookie = cookie;
797 * Return the vlan device present at the specific VID.
799 static struct ifnet *
800 vlan_devat(struct ifnet *ifp, uint16_t vid)
802 struct ifvlantrunk *trunk;
807 trunk = ifp->if_vlantrunk;
811 ifv = vlan_gethash(trunk, vid);
818 * VLAN support can be loaded as a module. The only place in the
819 * system that's intimately aware of this is ether_input. We hook
820 * into this code through vlan_input_p which is defined there and
821 * set here. No one else in the system should be aware of this so
822 * we use an explicit reference here.
824 extern void (*vlan_input_p)(struct ifnet *, struct mbuf *);
826 /* For if_link_state_change() eyes only... */
827 extern void (*vlan_link_state_p)(struct ifnet *);
830 vlan_modevent(module_t mod, int type, void *data)
835 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
836 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
837 if (ifdetach_tag == NULL)
839 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
840 vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
841 if (iflladdr_tag == NULL)
844 vlan_input_p = vlan_input;
845 vlan_link_state_p = vlan_link_state;
846 vlan_trunk_cap_p = vlan_trunk_capabilities;
847 vlan_trunkdev_p = vlan_trunkdev;
848 vlan_cookie_p = vlan_cookie;
849 vlan_setcookie_p = vlan_setcookie;
850 vlan_tag_p = vlan_tag;
851 vlan_pcp_p = vlan_pcp;
852 vlan_devat_p = vlan_devat;
854 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
855 vlan_clone_create, vlan_clone_destroy);
858 printf("vlan: initialized, using "
862 "hash tables with chaining"
869 if_clone_detach(vlan_cloner);
871 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
872 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
874 vlan_link_state_p = NULL;
875 vlan_trunk_cap_p = NULL;
876 vlan_trunkdev_p = NULL;
878 vlan_cookie_p = NULL;
879 vlan_setcookie_p = NULL;
881 VLAN_LOCKING_DESTROY();
883 printf("vlan: unloaded\n");
891 static moduledata_t vlan_mod = {
897 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
898 MODULE_VERSION(if_vlan, 3);
902 vnet_vlan_init(const void *unused __unused)
905 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
906 vlan_clone_create, vlan_clone_destroy);
907 V_vlan_cloner = vlan_cloner;
909 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
910 vnet_vlan_init, NULL);
913 vnet_vlan_uninit(const void *unused __unused)
916 if_clone_detach(V_vlan_cloner);
918 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
919 vnet_vlan_uninit, NULL);
923 * Check for <etherif>.<vlan>[.<vlan> ...] style interface names.
925 static struct ifnet *
926 vlan_clone_match_ethervid(const char *name, int *vidp)
928 char ifname[IFNAMSIZ];
933 strlcpy(ifname, name, IFNAMSIZ);
934 if ((cp = strrchr(ifname, '.')) == NULL)
937 if ((ifp = ifunit_ref(ifname)) == NULL)
945 for(; *cp >= '0' && *cp <= '9'; cp++)
946 vid = (vid * 10) + (*cp - '0');
958 vlan_clone_match(struct if_clone *ifc, const char *name)
963 ifp = vlan_clone_match_ethervid(name, NULL);
969 if (strncmp(vlanname, name, strlen(vlanname)) != 0)
971 for (cp = name + 4; *cp != '\0'; cp++) {
972 if (*cp < '0' || *cp > '9')
980 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
983 bool wildcard = false;
984 bool subinterface = false;
988 uint16_t proto = ETHERTYPE_VLAN;
991 struct ifnet *p = NULL;
993 struct sockaddr_dl *sdl;
995 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
999 * There are three ways to specify the cloned device:
1000 * o pass a parameter block with the clone request.
1001 * o specify parameters in the text of the clone device name
1002 * o specify no parameters and get an unattached device that
1003 * must be configured separately.
1004 * The first technique is preferred; the latter two are supported
1005 * for backwards compatibility.
1007 * XXXRW: Note historic use of the word "tag" here. New ioctls may be
1012 error = copyin(params, &vlr, sizeof(vlr));
1016 proto = vlr.vlr_proto;
1018 p = ifunit_ref(vlr.vlr_parent);
1023 if ((error = ifc_name2unit(name, &unit)) == 0) {
1026 * vlanX interface. Set wildcard to true if the unit number
1029 wildcard = (unit < 0);
1031 struct ifnet *p_tmp = vlan_clone_match_ethervid(name, &vid);
1032 if (p_tmp != NULL) {
1034 subinterface = true;
1035 unit = IF_DUNIT_NONE;
1053 if (!subinterface) {
1054 /* vlanX interface, mark X as busy or allocate new unit # */
1055 error = ifc_alloc_unit(ifc, &unit);
1063 /* In the wildcard case, we need to update the name. */
1065 for (dp = name; *dp != '\0'; dp++);
1066 if (snprintf(dp, len - (dp-name), "%d", unit) >
1067 len - (dp-name) - 1) {
1068 panic("%s: interface name too long", __func__);
1072 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
1073 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
1076 ifc_free_unit(ifc, unit);
1082 CK_SLIST_INIT(&ifv->vlan_mc_listhead);
1083 ifp->if_softc = ifv;
1085 * Set the name manually rather than using if_initname because
1086 * we don't conform to the default naming convention for interfaces.
1088 strlcpy(ifp->if_xname, name, IFNAMSIZ);
1089 ifp->if_dname = vlanname;
1090 ifp->if_dunit = unit;
1092 ifp->if_init = vlan_init;
1093 ifp->if_transmit = vlan_transmit;
1094 ifp->if_qflush = vlan_qflush;
1095 ifp->if_ioctl = vlan_ioctl;
1096 #if defined(KERN_TLS) || defined(RATELIMIT)
1097 ifp->if_snd_tag_alloc = vlan_snd_tag_alloc;
1098 ifp->if_snd_tag_modify = vlan_snd_tag_modify;
1099 ifp->if_snd_tag_query = vlan_snd_tag_query;
1100 ifp->if_snd_tag_free = vlan_snd_tag_free;
1101 ifp->if_next_snd_tag = vlan_next_snd_tag;
1102 ifp->if_ratelimit_query = vlan_ratelimit_query;
1104 ifp->if_flags = VLAN_IFFLAGS;
1105 ether_ifattach(ifp, eaddr);
1106 /* Now undo some of the damage... */
1107 ifp->if_baudrate = 0;
1108 ifp->if_type = IFT_L2VLAN;
1109 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
1111 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1112 sdl->sdl_type = IFT_L2VLAN;
1115 error = vlan_config(ifv, p, vid, proto);
1119 * Since we've partially failed, we need to back
1120 * out all the way, otherwise userland could get
1121 * confused. Thus, we destroy the interface.
1123 ether_ifdetach(ifp);
1127 ifc_free_unit(ifc, unit);
1138 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1140 struct ifvlan *ifv = ifp->if_softc;
1141 int unit = ifp->if_dunit;
1143 if (ifp->if_vlantrunk)
1146 ether_ifdetach(ifp); /* first, remove it from system-wide lists */
1147 vlan_unconfig(ifp); /* now it can be unconfigured and freed */
1149 * We should have the only reference to the ifv now, so we can now
1150 * drain any remaining lladdr task before freeing the ifnet and the
1153 taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
1157 if (unit != IF_DUNIT_NONE)
1158 ifc_free_unit(ifc, unit);
1164 * The ifp->if_init entry point for vlan(4) is a no-op.
1167 vlan_init(void *foo __unused)
1172 * The if_transmit method for vlan(4) interface.
1175 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1179 int error, len, mcast;
1183 ifv = ifp->if_softc;
1184 if (TRUNK(ifv) == NULL) {
1185 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1190 len = m->m_pkthdr.len;
1191 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1195 #if defined(KERN_TLS) || defined(RATELIMIT)
1196 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
1197 struct vlan_snd_tag *vst;
1198 struct m_snd_tag *mst;
1200 MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
1201 mst = m->m_pkthdr.snd_tag;
1202 vst = mst_to_vst(mst);
1203 if (vst->tag->ifp != p) {
1204 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1209 m->m_pkthdr.snd_tag = m_snd_tag_ref(vst->tag);
1210 m_snd_tag_rele(mst);
1215 * Do not run parent's if_transmit() if the parent is not up,
1216 * or parent's driver will cause a system crash.
1218 if (!UP_AND_RUNNING(p)) {
1219 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1224 if (!ether_8021q_frame(&m, ifp, p, &ifv->ifv_qtag)) {
1225 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1230 * Send it, precisely as ether_output() would have.
1232 error = (p->if_transmit)(p, m);
1234 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1235 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1236 if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1238 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1243 vlan_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1252 * Find the first non-VLAN parent interface.
1254 ifv = ifp->if_softc;
1256 if (TRUNK(ifv) == NULL) {
1262 } while (p->if_type == IFT_L2VLAN);
1264 return p->if_output(ifp, m, dst, ro);
1268 * The ifp->if_qflush entry point for vlan(4) is a no-op.
1271 vlan_qflush(struct ifnet *ifp __unused)
1276 vlan_input(struct ifnet *ifp, struct mbuf *m)
1278 struct ifvlantrunk *trunk;
1285 trunk = ifp->if_vlantrunk;
1286 if (trunk == NULL) {
1291 if (m->m_flags & M_VLANTAG) {
1293 * Packet is tagged, but m contains a normal
1294 * Ethernet frame; the tag is stored out-of-band.
1296 tag = m->m_pkthdr.ether_vtag;
1297 m->m_flags &= ~M_VLANTAG;
1299 struct ether_vlan_header *evl;
1302 * Packet is tagged in-band as specified by 802.1q.
1304 switch (ifp->if_type) {
1306 if (m->m_len < sizeof(*evl) &&
1307 (m = m_pullup(m, sizeof(*evl))) == NULL) {
1308 if_printf(ifp, "cannot pullup VLAN header\n");
1311 evl = mtod(m, struct ether_vlan_header *);
1312 tag = ntohs(evl->evl_tag);
1315 * Remove the 802.1q header by copying the Ethernet
1316 * addresses over it and adjusting the beginning of
1317 * the data in the mbuf. The encapsulated Ethernet
1318 * type field is already in place.
1320 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1321 ETHER_HDR_LEN - ETHER_TYPE_LEN);
1322 m_adj(m, ETHER_VLAN_ENCAP_LEN);
1327 panic("%s: %s has unsupported if_type %u",
1328 __func__, ifp->if_xname, ifp->if_type);
1330 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1336 vid = EVL_VLANOFTAG(tag);
1338 ifv = vlan_gethash(trunk, vid);
1339 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1340 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1345 if (vlan_mtag_pcp) {
1347 * While uncommon, it is possible that we will find a 802.1q
1348 * packet encapsulated inside another packet that also had an
1349 * 802.1q header. For example, ethernet tunneled over IPSEC
1350 * arriving over ethernet. In that case, we replace the
1351 * existing 802.1q PCP m_tag value.
1353 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1355 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1356 sizeof(uint8_t), M_NOWAIT);
1358 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1362 m_tag_prepend(m, mtag);
1364 *(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1367 m->m_pkthdr.rcvif = ifv->ifv_ifp;
1368 if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1370 /* Pass it back through the parent's input routine. */
1371 (*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
1375 vlan_lladdr_fn(void *arg, int pending __unused)
1380 ifv = (struct ifvlan *)arg;
1383 CURVNET_SET(ifp->if_vnet);
1385 /* The ifv_ifp already has the lladdr copied in. */
1386 if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1392 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid,
1395 struct epoch_tracker et;
1396 struct ifvlantrunk *trunk;
1401 * We can handle non-ethernet hardware types as long as
1402 * they handle the tagging and headers themselves.
1404 if (p->if_type != IFT_ETHER &&
1405 p->if_type != IFT_L2VLAN &&
1406 (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1407 return (EPROTONOSUPPORT);
1408 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1409 return (EPROTONOSUPPORT);
1411 * Don't let the caller set up a VLAN VID with
1412 * anything except VLID bits.
1413 * VID numbers 0x0 and 0xFFF are reserved.
1415 if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1421 if (p->if_vlantrunk == NULL) {
1422 trunk = malloc(sizeof(struct ifvlantrunk),
1423 M_VLAN, M_WAITOK | M_ZERO);
1424 vlan_inithash(trunk);
1425 TRUNK_LOCK_INIT(trunk);
1427 p->if_vlantrunk = trunk;
1429 if_ref(trunk->parent);
1430 TRUNK_WUNLOCK(trunk);
1432 trunk = p->if_vlantrunk;
1435 ifv->ifv_vid = vid; /* must set this before vlan_inshash() */
1436 ifv->ifv_pcp = 0; /* Default: best effort delivery. */
1437 error = vlan_inshash(trunk, ifv);
1440 ifv->ifv_proto = proto;
1441 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1442 ifv->ifv_mintu = ETHERMIN;
1443 ifv->ifv_pflags = 0;
1444 ifv->ifv_capenable = -1;
1447 * If the parent supports the VLAN_MTU capability,
1448 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1451 if (p->if_capenable & IFCAP_VLAN_MTU) {
1453 * No need to fudge the MTU since the parent can
1454 * handle extended frames.
1456 ifv->ifv_mtufudge = 0;
1459 * Fudge the MTU by the encapsulation size. This
1460 * makes us incompatible with strictly compliant
1461 * 802.1Q implementations, but allows us to use
1462 * the feature with other NetBSD implementations,
1463 * which might still be useful.
1465 ifv->ifv_mtufudge = ifv->ifv_encaplen;
1468 ifv->ifv_trunk = trunk;
1471 * Initialize fields from our parent. This duplicates some
1472 * work with ether_ifattach() but allows for non-ethernet
1473 * interfaces to also work.
1475 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1476 ifp->if_baudrate = p->if_baudrate;
1477 ifp->if_input = p->if_input;
1478 ifp->if_resolvemulti = p->if_resolvemulti;
1479 ifp->if_addrlen = p->if_addrlen;
1480 ifp->if_broadcastaddr = p->if_broadcastaddr;
1481 ifp->if_pcp = ifv->ifv_pcp;
1484 * We wrap the parent's if_output using vlan_output to ensure that it
1485 * can't become stale.
1487 ifp->if_output = vlan_output;
1490 * Copy only a selected subset of flags from the parent.
1491 * Other flags are none of our business.
1493 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1494 ifp->if_flags &= ~VLAN_COPY_FLAGS;
1495 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1496 #undef VLAN_COPY_FLAGS
1498 ifp->if_link_state = p->if_link_state;
1500 NET_EPOCH_ENTER(et);
1501 vlan_capabilities(ifv);
1505 * Set up our interface address to reflect the underlying
1506 * physical interface's.
1508 TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
1509 ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1513 * Do not schedule link address update if it was the same
1514 * as previous parent's. This helps avoid updating for each
1515 * associated llentry.
1517 if (memcmp(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen) != 0) {
1518 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1519 taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
1522 /* We are ready for operation now. */
1523 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1525 /* Update flags on the parent, if necessary. */
1526 vlan_setflags(ifp, 1);
1529 * Configure multicast addresses that may already be
1530 * joined on the vlan device.
1532 (void)vlan_setmulti(ifp);
1536 EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1543 vlan_unconfig(struct ifnet *ifp)
1547 vlan_unconfig_locked(ifp, 0);
1552 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1554 struct ifvlantrunk *trunk;
1555 struct vlan_mc_entry *mc;
1557 struct ifnet *parent;
1560 VLAN_XLOCK_ASSERT();
1562 ifv = ifp->if_softc;
1563 trunk = ifv->ifv_trunk;
1566 if (trunk != NULL) {
1567 parent = trunk->parent;
1570 * Since the interface is being unconfigured, we need to
1571 * empty the list of multicast groups that we may have joined
1572 * while we were alive from the parent's list.
1574 while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1576 * If the parent interface is being detached,
1577 * all its multicast addresses have already
1578 * been removed. Warn about errors if
1579 * if_delmulti() does fail, but don't abort as
1580 * all callers expect vlan destruction to
1584 error = if_delmulti(parent,
1585 (struct sockaddr *)&mc->mc_addr);
1588 "Failed to delete multicast address from parent: %d\n",
1591 CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1592 NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
1595 vlan_setflags(ifp, 0); /* clear special flags on parent */
1597 vlan_remhash(trunk, ifv);
1598 ifv->ifv_trunk = NULL;
1601 * Check if we were the last.
1603 if (trunk->refcnt == 0) {
1604 parent->if_vlantrunk = NULL;
1606 trunk_destroy(trunk);
1610 /* Disconnect from parent. */
1611 if (ifv->ifv_pflags)
1612 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1613 ifp->if_mtu = ETHERMTU;
1614 ifp->if_link_state = LINK_STATE_UNKNOWN;
1615 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1618 * Only dispatch an event if vlan was
1619 * attached, otherwise there is nothing
1620 * to cleanup anyway.
1623 EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1626 /* Handle a reference counted flag that should be set on the parent as well */
1628 vlan_setflag(struct ifnet *ifp, int flag, int status,
1629 int (*func)(struct ifnet *, int))
1634 VLAN_SXLOCK_ASSERT();
1636 ifv = ifp->if_softc;
1637 status = status ? (ifp->if_flags & flag) : 0;
1638 /* Now "status" contains the flag value or 0 */
1641 * See if recorded parent's status is different from what
1642 * we want it to be. If it is, flip it. We record parent's
1643 * status in ifv_pflags so that we won't clear parent's flag
1644 * we haven't set. In fact, we don't clear or set parent's
1645 * flags directly, but get or release references to them.
1646 * That's why we can be sure that recorded flags still are
1647 * in accord with actual parent's flags.
1649 if (status != (ifv->ifv_pflags & flag)) {
1650 error = (*func)(PARENT(ifv), status);
1653 ifv->ifv_pflags &= ~flag;
1654 ifv->ifv_pflags |= status;
1660 * Handle IFF_* flags that require certain changes on the parent:
1661 * if "status" is true, update parent's flags respective to our if_flags;
1662 * if "status" is false, forcedly clear the flags set on parent.
1665 vlan_setflags(struct ifnet *ifp, int status)
1669 for (i = 0; vlan_pflags[i].flag; i++) {
1670 error = vlan_setflag(ifp, vlan_pflags[i].flag,
1671 status, vlan_pflags[i].func);
1678 /* Inform all vlans that their parent has changed link state */
1680 vlan_link_state(struct ifnet *ifp)
1682 struct epoch_tracker et;
1683 struct ifvlantrunk *trunk;
1686 NET_EPOCH_ENTER(et);
1687 trunk = ifp->if_vlantrunk;
1688 if (trunk == NULL) {
1694 VLAN_FOREACH(ifv, trunk) {
1695 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1696 if_link_state_change(ifv->ifv_ifp,
1697 trunk->parent->if_link_state);
1699 TRUNK_WUNLOCK(trunk);
1704 vlan_capabilities(struct ifvlan *ifv)
1708 struct ifnet_hw_tsomax hw_tsomax;
1709 int cap = 0, ena = 0, mena;
1713 VLAN_SXLOCK_ASSERT();
1718 /* Mask parent interface enabled capabilities disabled by user. */
1719 mena = p->if_capenable & ifv->ifv_capenable;
1722 * If the parent interface can do checksum offloading
1723 * on VLANs, then propagate its hardware-assisted
1724 * checksumming flags. Also assert that checksum
1725 * offloading requires hardware VLAN tagging.
1727 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1728 cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1729 if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1730 p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1731 ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1732 if (ena & IFCAP_TXCSUM)
1733 hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
1734 CSUM_UDP | CSUM_SCTP);
1735 if (ena & IFCAP_TXCSUM_IPV6)
1736 hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
1737 CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
1741 * If the parent interface can do TSO on VLANs then
1742 * propagate the hardware-assisted flag. TSO on VLANs
1743 * does not necessarily require hardware VLAN tagging.
1745 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1746 if_hw_tsomax_common(p, &hw_tsomax);
1747 if_hw_tsomax_update(ifp, &hw_tsomax);
1748 if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1749 cap |= p->if_capabilities & IFCAP_TSO;
1750 if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1751 ena |= mena & IFCAP_TSO;
1752 if (ena & IFCAP_TSO)
1753 hwa |= p->if_hwassist & CSUM_TSO;
1757 * If the parent interface can do LRO and checksum offloading on
1758 * VLANs, then guess it may do LRO on VLANs. False positive here
1759 * cost nothing, while false negative may lead to some confusions.
1761 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1762 cap |= p->if_capabilities & IFCAP_LRO;
1763 if (p->if_capenable & IFCAP_VLAN_HWCSUM)
1764 ena |= p->if_capenable & IFCAP_LRO;
1767 * If the parent interface can offload TCP connections over VLANs then
1768 * propagate its TOE capability to the VLAN interface.
1770 * All TOE drivers in the tree today can deal with VLANs. If this
1771 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1774 #define IFCAP_VLAN_TOE IFCAP_TOE
1775 if (p->if_capabilities & IFCAP_VLAN_TOE)
1776 cap |= p->if_capabilities & IFCAP_TOE;
1777 if (p->if_capenable & IFCAP_VLAN_TOE) {
1778 TOEDEV(ifp) = TOEDEV(p);
1779 ena |= mena & IFCAP_TOE;
1783 * If the parent interface supports dynamic link state, so does the
1786 cap |= (p->if_capabilities & IFCAP_LINKSTATE);
1787 ena |= (mena & IFCAP_LINKSTATE);
1791 * If the parent interface supports ratelimiting, so does the
1794 cap |= (p->if_capabilities & IFCAP_TXRTLMT);
1795 ena |= (mena & IFCAP_TXRTLMT);
1799 * If the parent interface supports unmapped mbufs, so does
1800 * the VLAN interface. Note that this should be fine even for
1801 * interfaces that don't support hardware tagging as headers
1802 * are prepended in normal mbufs to unmapped mbufs holding
1805 cap |= (p->if_capabilities & IFCAP_MEXTPG);
1806 ena |= (mena & IFCAP_MEXTPG);
1809 * If the parent interface can offload encryption and segmentation
1810 * of TLS records over TCP, propagate it's capability to the VLAN
1813 * All TLS drivers in the tree today can deal with VLANs. If
1814 * this ever changes, then a new IFCAP_VLAN_TXTLS can be
1817 if (p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
1818 cap |= p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
1819 if (p->if_capenable & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
1820 ena |= mena & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
1822 ifp->if_capabilities = cap;
1823 ifp->if_capenable = ena;
1824 ifp->if_hwassist = hwa;
1828 vlan_trunk_capabilities(struct ifnet *ifp)
1830 struct epoch_tracker et;
1831 struct ifvlantrunk *trunk;
1835 trunk = ifp->if_vlantrunk;
1836 if (trunk == NULL) {
1840 NET_EPOCH_ENTER(et);
1841 VLAN_FOREACH(ifv, trunk)
1842 vlan_capabilities(ifv);
1848 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1854 struct ifvlantrunk *trunk;
1856 int error = 0, oldmtu;
1858 ifr = (struct ifreq *)data;
1859 ifa = (struct ifaddr *) data;
1860 ifv = ifp->if_softc;
1864 ifp->if_flags |= IFF_UP;
1866 if (ifa->ifa_addr->sa_family == AF_INET)
1867 arp_ifinit(ifp, ifa);
1871 bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
1876 if (TRUNK(ifv) != NULL) {
1879 error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1881 /* Limit the result to the parent's current config. */
1883 struct ifmediareq *ifmr;
1885 ifmr = (struct ifmediareq *)data;
1886 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1887 ifmr->ifm_count = 1;
1888 error = copyout(&ifmr->ifm_current,
1905 * Set the interface MTU.
1909 if (trunk != NULL) {
1912 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1914 (ifv->ifv_mintu - ifv->ifv_mtufudge))
1917 ifp->if_mtu = ifr->ifr_mtu;
1918 TRUNK_WUNLOCK(trunk);
1927 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1928 * interface to be delegated to a jail without allowing the
1929 * jail to change what underlying interface/VID it is
1930 * associated with. We are not entirely convinced that this
1931 * is the right way to accomplish that policy goal.
1933 if (ifp->if_vnet != ifp->if_home_vnet) {
1938 error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
1941 if (vlr.vlr_parent[0] == '\0') {
1945 p = ifunit_ref(vlr.vlr_parent);
1950 oldmtu = ifp->if_mtu;
1951 error = vlan_config(ifv, p, vlr.vlr_tag, vlr.vlr_proto);
1955 * VLAN MTU may change during addition of the vlandev.
1956 * If it did, do network layer specific procedure.
1958 if (ifp->if_mtu != oldmtu) {
1968 if (ifp->if_vnet != ifp->if_home_vnet) {
1973 bzero(&vlr, sizeof(vlr));
1975 if (TRUNK(ifv) != NULL) {
1976 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1977 sizeof(vlr.vlr_parent));
1978 vlr.vlr_tag = ifv->ifv_vid;
1979 vlr.vlr_proto = ifv->ifv_proto;
1982 error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
1987 * We should propagate selected flags to the parent,
1988 * e.g., promiscuous mode.
1991 if (TRUNK(ifv) != NULL)
1992 error = vlan_setflags(ifp, 1);
1999 * If we don't have a parent, just remember the membership for
2002 * XXX We need the rmlock here to avoid sleeping while
2003 * holding in6_multi_mtx.
2008 error = vlan_setmulti(ifp);
2014 if (ifp->if_vnet != ifp->if_home_vnet) {
2019 ifr->ifr_vlan_pcp = ifv->ifv_pcp;
2024 if (ifp->if_vnet != ifp->if_home_vnet) {
2029 error = priv_check(curthread, PRIV_NET_SETVLANPCP);
2032 if (ifr->ifr_vlan_pcp > 7) {
2036 ifv->ifv_pcp = ifr->ifr_vlan_pcp;
2037 ifp->if_pcp = ifv->ifv_pcp;
2038 /* broadcast event about PCP change */
2039 EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
2044 ifv->ifv_capenable = ifr->ifr_reqcap;
2046 if (trunk != NULL) {
2047 struct epoch_tracker et;
2049 NET_EPOCH_ENTER(et);
2050 vlan_capabilities(ifv);
2064 #if defined(KERN_TLS) || defined(RATELIMIT)
2066 vlan_snd_tag_alloc(struct ifnet *ifp,
2067 union if_snd_tag_alloc_params *params,
2068 struct m_snd_tag **ppmt)
2070 struct epoch_tracker et;
2071 struct vlan_snd_tag *vst;
2073 struct ifnet *parent;
2076 NET_EPOCH_ENTER(et);
2077 ifv = ifp->if_softc;
2078 if (ifv->ifv_trunk != NULL)
2079 parent = PARENT(ifv);
2082 if (parent == NULL) {
2084 return (EOPNOTSUPP);
2089 vst = malloc(sizeof(*vst), M_VLAN, M_NOWAIT);
2095 error = m_snd_tag_alloc(parent, params, &vst->tag);
2102 m_snd_tag_init(&vst->com, ifp, vst->tag->type);
2108 static struct m_snd_tag *
2109 vlan_next_snd_tag(struct m_snd_tag *mst)
2111 struct vlan_snd_tag *vst;
2113 vst = mst_to_vst(mst);
2118 vlan_snd_tag_modify(struct m_snd_tag *mst,
2119 union if_snd_tag_modify_params *params)
2121 struct vlan_snd_tag *vst;
2123 vst = mst_to_vst(mst);
2124 return (vst->tag->ifp->if_snd_tag_modify(vst->tag, params));
2128 vlan_snd_tag_query(struct m_snd_tag *mst,
2129 union if_snd_tag_query_params *params)
2131 struct vlan_snd_tag *vst;
2133 vst = mst_to_vst(mst);
2134 return (vst->tag->ifp->if_snd_tag_query(vst->tag, params));
2138 vlan_snd_tag_free(struct m_snd_tag *mst)
2140 struct vlan_snd_tag *vst;
2142 vst = mst_to_vst(mst);
2143 m_snd_tag_rele(vst->tag);
2148 vlan_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
2151 * For vlan, we have an indirect
2152 * interface. The caller needs to
2153 * get a ratelimit tag on the actual
2154 * interface the flow will go on.
2156 q->rate_table = NULL;
2157 q->flags = RT_IS_INDIRECT;
2159 q->number_of_rates = 0;