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>
47 #include "opt_inet6.h"
48 #include "opt_kern_tls.h"
50 #include "opt_ratelimit.h"
52 #include <sys/param.h>
53 #include <sys/eventhandler.h>
54 #include <sys/kernel.h>
56 #include <sys/malloc.h>
58 #include <sys/module.h>
59 #include <sys/rmlock.h>
61 #include <sys/queue.h>
62 #include <sys/socket.h>
63 #include <sys/sockio.h>
64 #include <sys/sysctl.h>
65 #include <sys/systm.h>
67 #include <sys/taskqueue.h>
70 #include <net/ethernet.h>
72 #include <net/if_var.h>
73 #include <net/if_private.h>
74 #include <net/if_clone.h>
75 #include <net/if_dl.h>
76 #include <net/if_types.h>
77 #include <net/if_vlan_var.h>
78 #include <net/route.h>
82 #include <netinet/in.h>
83 #include <netinet/if_ether.h>
86 #include <netlink/netlink.h>
87 #include <netlink/netlink_ctl.h>
88 #include <netlink/netlink_route.h>
89 #include <netlink/route/route_var.h>
91 #define VLAN_DEF_HWIDTH 4
92 #define VLAN_IFFLAGS (IFF_BROADCAST | IFF_MULTICAST)
94 #define UP_AND_RUNNING(ifp) \
95 ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
97 CK_SLIST_HEAD(ifvlanhead, ifvlan);
100 struct ifnet *parent; /* parent interface of this trunk */
103 #define VLAN_ARRAY_SIZE (EVL_VLID_MASK + 1)
104 struct ifvlan *vlans[VLAN_ARRAY_SIZE]; /* static table */
106 struct ifvlanhead *hash; /* dynamic hash-list table */
113 #if defined(KERN_TLS) || defined(RATELIMIT)
114 struct vlan_snd_tag {
115 struct m_snd_tag com;
116 struct m_snd_tag *tag;
119 static inline struct vlan_snd_tag *
120 mst_to_vst(struct m_snd_tag *mst)
123 return (__containerof(mst, struct vlan_snd_tag, com));
128 * This macro provides a facility to iterate over every vlan on a trunk with
129 * the assumption that none will be added/removed during iteration.
132 #define VLAN_FOREACH(_ifv, _trunk) \
134 for (_i = 0; _i < VLAN_ARRAY_SIZE; _i++) \
135 if (((_ifv) = (_trunk)->vlans[_i]) != NULL)
136 #else /* VLAN_ARRAY */
137 #define VLAN_FOREACH(_ifv, _trunk) \
138 struct ifvlan *_next; \
140 for (_i = 0; _i < (1 << (_trunk)->hwidth); _i++) \
141 CK_SLIST_FOREACH_SAFE((_ifv), &(_trunk)->hash[_i], ifv_list, _next)
142 #endif /* VLAN_ARRAY */
145 * This macro provides a facility to iterate over every vlan on a trunk while
146 * also modifying the number of vlans on the trunk. The iteration continues
147 * until some condition is met or there are no more vlans on the trunk.
150 /* The VLAN_ARRAY case is simple -- just a for loop using the condition. */
151 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
153 for (_i = 0; !(_cond) && _i < VLAN_ARRAY_SIZE; _i++) \
154 if (((_ifv) = (_trunk)->vlans[_i]))
155 #else /* VLAN_ARRAY */
157 * The hash table case is more complicated. We allow for the hash table to be
158 * modified (i.e. vlans removed) while we are iterating over it. To allow for
159 * this we must restart the iteration every time we "touch" something during
160 * the iteration, since removal will resize the hash table and invalidate our
161 * current position. If acting on the touched element causes the trunk to be
162 * emptied, then iteration also stops.
164 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
166 bool _touch = false; \
168 !(_cond) && _i < (1 << (_trunk)->hwidth); \
169 _i = (_touch && ((_trunk) != NULL) ? 0 : _i + 1), _touch = false) \
170 if (((_ifv) = CK_SLIST_FIRST(&(_trunk)->hash[_i])) != NULL && \
172 #endif /* VLAN_ARRAY */
174 struct vlan_mc_entry {
175 struct sockaddr_dl mc_addr;
176 CK_SLIST_ENTRY(vlan_mc_entry) mc_entries;
177 struct epoch_context mc_epoch_ctx;
181 struct ifvlantrunk *ifv_trunk;
182 struct ifnet *ifv_ifp;
183 #define TRUNK(ifv) ((ifv)->ifv_trunk)
184 #define PARENT(ifv) (TRUNK(ifv)->parent)
186 int ifv_pflags; /* special flags we have set on parent */
188 int ifv_encaplen; /* encapsulation length */
189 int ifv_mtufudge; /* MTU fudged by this much */
190 int ifv_mintu; /* min transmission unit */
191 struct ether_8021q_tag ifv_qtag;
192 #define ifv_proto ifv_qtag.proto
193 #define ifv_vid ifv_qtag.vid
194 #define ifv_pcp ifv_qtag.pcp
195 struct task lladdr_task;
196 CK_SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
198 CK_SLIST_ENTRY(ifvlan) ifv_list;
202 /* Special flags we should propagate to parent. */
205 int (*func)(struct ifnet *, int);
207 {IFF_PROMISC, ifpromisc},
208 {IFF_ALLMULTI, if_allmulti},
212 VNET_DECLARE(int, vlan_mtag_pcp);
213 #define V_vlan_mtag_pcp VNET(vlan_mtag_pcp)
215 static const char vlanname[] = "vlan";
216 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
218 static eventhandler_tag ifdetach_tag;
219 static eventhandler_tag iflladdr_tag;
220 static eventhandler_tag ifevent_tag;
223 * if_vlan uses two module-level synchronizations primitives to allow concurrent
224 * modification of vlan interfaces and (mostly) allow for vlans to be destroyed
225 * while they are being used for tx/rx. To accomplish this in a way that has
226 * acceptable performance and cooperation with other parts of the network stack
227 * there is a non-sleepable epoch(9) and an sx(9).
229 * The performance-sensitive paths that warrant using the epoch(9) are
230 * vlan_transmit and vlan_input. Both have to check for the vlan interface's
231 * existence using if_vlantrunk, and being in the network tx/rx paths the use
232 * of an epoch(9) gives a measureable improvement in performance.
234 * The reason for having an sx(9) is mostly because there are still areas that
235 * must be sleepable and also have safe concurrent access to a vlan interface.
236 * Since the sx(9) exists, it is used by default in most paths unless sleeping
237 * is not permitted, or if it is not clear whether sleeping is permitted.
240 #define _VLAN_SX_ID ifv_sx
242 static struct sx _VLAN_SX_ID;
244 #define VLAN_LOCKING_INIT() \
245 sx_init_flags(&_VLAN_SX_ID, "vlan_sx", SX_RECURSE)
247 #define VLAN_LOCKING_DESTROY() \
248 sx_destroy(&_VLAN_SX_ID)
250 #define VLAN_SLOCK() sx_slock(&_VLAN_SX_ID)
251 #define VLAN_SUNLOCK() sx_sunlock(&_VLAN_SX_ID)
252 #define VLAN_XLOCK() sx_xlock(&_VLAN_SX_ID)
253 #define VLAN_XUNLOCK() sx_xunlock(&_VLAN_SX_ID)
254 #define VLAN_SLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_SLOCKED)
255 #define VLAN_XLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_XLOCKED)
256 #define VLAN_SXLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_LOCKED)
259 * We also have a per-trunk mutex that should be acquired when changing
262 #define TRUNK_LOCK_INIT(trunk) mtx_init(&(trunk)->lock, vlanname, NULL, MTX_DEF)
263 #define TRUNK_LOCK_DESTROY(trunk) mtx_destroy(&(trunk)->lock)
264 #define TRUNK_WLOCK(trunk) mtx_lock(&(trunk)->lock)
265 #define TRUNK_WUNLOCK(trunk) mtx_unlock(&(trunk)->lock)
266 #define TRUNK_WLOCK_ASSERT(trunk) mtx_assert(&(trunk)->lock, MA_OWNED);
269 * The VLAN_ARRAY substitutes the dynamic hash with a static array
270 * with 4096 entries. In theory this can give a boost in processing,
271 * however in practice it does not. Probably this is because the array
272 * is too big to fit into CPU cache.
275 static void vlan_inithash(struct ifvlantrunk *trunk);
276 static void vlan_freehash(struct ifvlantrunk *trunk);
277 static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
278 static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
279 static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
280 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
283 static void trunk_destroy(struct ifvlantrunk *trunk);
285 static void vlan_init(void *foo);
286 static void vlan_input(struct ifnet *ifp, struct mbuf *m);
287 static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
288 #if defined(KERN_TLS) || defined(RATELIMIT)
289 static int vlan_snd_tag_alloc(struct ifnet *,
290 union if_snd_tag_alloc_params *, struct m_snd_tag **);
291 static int vlan_snd_tag_modify(struct m_snd_tag *,
292 union if_snd_tag_modify_params *);
293 static int vlan_snd_tag_query(struct m_snd_tag *,
294 union if_snd_tag_query_params *);
295 static void vlan_snd_tag_free(struct m_snd_tag *);
296 static struct m_snd_tag *vlan_next_snd_tag(struct m_snd_tag *);
297 static void vlan_ratelimit_query(struct ifnet *,
298 struct if_ratelimit_query_results *);
300 static void vlan_qflush(struct ifnet *ifp);
301 static int vlan_setflag(struct ifnet *ifp, int flag, int status,
302 int (*func)(struct ifnet *, int));
303 static int vlan_setflags(struct ifnet *ifp, int status);
304 static int vlan_setmulti(struct ifnet *ifp);
305 static int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
307 static void vlan_altq_start(struct ifnet *ifp);
308 static int vlan_altq_transmit(struct ifnet *ifp, struct mbuf *m);
310 static int vlan_output(struct ifnet *ifp, struct mbuf *m,
311 const struct sockaddr *dst, struct route *ro);
312 static void vlan_unconfig(struct ifnet *ifp);
313 static void vlan_unconfig_locked(struct ifnet *ifp, int departing);
314 static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag,
316 static void vlan_link_state(struct ifnet *ifp);
317 static void vlan_capabilities(struct ifvlan *ifv);
318 static void vlan_trunk_capabilities(struct ifnet *ifp);
320 static struct ifnet *vlan_clone_match_ethervid(const char *, int *);
321 static int vlan_clone_match(struct if_clone *, const char *);
322 static int vlan_clone_create(struct if_clone *, char *, size_t,
323 struct ifc_data *, struct ifnet **);
324 static int vlan_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
326 static int vlan_clone_create_nl(struct if_clone *ifc, char *name, size_t len,
327 struct ifc_data_nl *ifd);
328 static int vlan_clone_modify_nl(struct ifnet *ifp, struct ifc_data_nl *ifd);
329 static void vlan_clone_dump_nl(struct ifnet *ifp, struct nl_writer *nw);
331 static void vlan_ifdetach(void *arg, struct ifnet *ifp);
332 static void vlan_iflladdr(void *arg, struct ifnet *ifp);
333 static void vlan_ifevent(void *arg, struct ifnet *ifp, int event);
335 static void vlan_lladdr_fn(void *arg, int pending);
337 static struct if_clone *vlan_cloner;
340 VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner);
341 #define V_vlan_cloner VNET(vlan_cloner)
345 static const struct if_snd_tag_sw vlan_snd_tag_ul_sw = {
346 .snd_tag_modify = vlan_snd_tag_modify,
347 .snd_tag_query = vlan_snd_tag_query,
348 .snd_tag_free = vlan_snd_tag_free,
349 .next_snd_tag = vlan_next_snd_tag,
350 .type = IF_SND_TAG_TYPE_UNLIMITED
353 static const struct if_snd_tag_sw vlan_snd_tag_rl_sw = {
354 .snd_tag_modify = vlan_snd_tag_modify,
355 .snd_tag_query = vlan_snd_tag_query,
356 .snd_tag_free = vlan_snd_tag_free,
357 .next_snd_tag = vlan_next_snd_tag,
358 .type = IF_SND_TAG_TYPE_RATE_LIMIT
363 static const struct if_snd_tag_sw vlan_snd_tag_tls_sw = {
364 .snd_tag_modify = vlan_snd_tag_modify,
365 .snd_tag_query = vlan_snd_tag_query,
366 .snd_tag_free = vlan_snd_tag_free,
367 .next_snd_tag = vlan_next_snd_tag,
368 .type = IF_SND_TAG_TYPE_TLS
372 static const struct if_snd_tag_sw vlan_snd_tag_tls_rl_sw = {
373 .snd_tag_modify = vlan_snd_tag_modify,
374 .snd_tag_query = vlan_snd_tag_query,
375 .snd_tag_free = vlan_snd_tag_free,
376 .next_snd_tag = vlan_next_snd_tag,
377 .type = IF_SND_TAG_TYPE_TLS_RATE_LIMIT
383 vlan_mc_free(struct epoch_context *ctx)
385 struct vlan_mc_entry *mc = __containerof(ctx, struct vlan_mc_entry, mc_epoch_ctx);
390 #define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
393 vlan_inithash(struct ifvlantrunk *trunk)
398 * The trunk must not be locked here since we call malloc(M_WAITOK).
399 * It is OK in case this function is called before the trunk struct
400 * gets hooked up and becomes visible from other threads.
403 KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
404 ("%s: hash already initialized", __func__));
406 trunk->hwidth = VLAN_DEF_HWIDTH;
407 n = 1 << trunk->hwidth;
408 trunk->hmask = n - 1;
409 trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
410 for (i = 0; i < n; i++)
411 CK_SLIST_INIT(&trunk->hash[i]);
415 vlan_freehash(struct ifvlantrunk *trunk)
420 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
421 for (i = 0; i < (1 << trunk->hwidth); i++)
422 KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]),
423 ("%s: hash table not empty", __func__));
425 free(trunk->hash, M_VLAN);
427 trunk->hwidth = trunk->hmask = 0;
431 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
437 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
439 b = 1 << trunk->hwidth;
440 i = HASH(ifv->ifv_vid, trunk->hmask);
441 CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
442 if (ifv->ifv_vid == ifv2->ifv_vid)
446 * Grow the hash when the number of vlans exceeds half of the number of
447 * hash buckets squared. This will make the average linked-list length
450 if (trunk->refcnt > (b * b) / 2) {
451 vlan_growhash(trunk, 1);
452 i = HASH(ifv->ifv_vid, trunk->hmask);
454 CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
461 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
467 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
469 b = 1 << (trunk->hwidth - 1);
470 i = HASH(ifv->ifv_vid, trunk->hmask);
471 CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
474 CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list);
475 if (trunk->refcnt < (b * b) / 2)
476 vlan_growhash(trunk, -1);
480 panic("%s: vlan not found\n", __func__);
481 return (ENOENT); /*NOTREACHED*/
485 * Grow the hash larger or smaller if memory permits.
488 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
491 struct ifvlanhead *hash2;
492 int hwidth2, i, j, n, n2;
495 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
498 /* Harmless yet obvious coding error */
499 printf("%s: howmuch is 0\n", __func__);
503 hwidth2 = trunk->hwidth + howmuch;
504 n = 1 << trunk->hwidth;
506 /* Do not shrink the table below the default */
507 if (hwidth2 < VLAN_DEF_HWIDTH)
510 hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK);
512 printf("%s: out of memory -- hash size not changed\n",
514 return; /* We can live with the old hash table */
516 for (j = 0; j < n2; j++)
517 CK_SLIST_INIT(&hash2[j]);
518 for (i = 0; i < n; i++)
519 while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) {
520 CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list);
521 j = HASH(ifv->ifv_vid, n2 - 1);
522 CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
525 free(trunk->hash, M_VLAN);
527 trunk->hwidth = hwidth2;
528 trunk->hmask = n2 - 1;
531 if_printf(trunk->parent,
532 "VLAN hash table resized from %d to %d buckets\n", n, n2);
535 static __inline struct ifvlan *
536 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
542 CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
543 if (ifv->ifv_vid == vid)
549 /* Debugging code to view the hashtables. */
551 vlan_dumphash(struct ifvlantrunk *trunk)
556 for (i = 0; i < (1 << trunk->hwidth); i++) {
558 CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
559 printf("%s ", ifv->ifv_ifp->if_xname);
566 static __inline struct ifvlan *
567 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
570 return trunk->vlans[vid];
574 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
577 if (trunk->vlans[ifv->ifv_vid] != NULL)
579 trunk->vlans[ifv->ifv_vid] = ifv;
586 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
589 trunk->vlans[ifv->ifv_vid] = NULL;
596 vlan_freehash(struct ifvlantrunk *trunk)
601 vlan_inithash(struct ifvlantrunk *trunk)
605 #endif /* !VLAN_ARRAY */
608 trunk_destroy(struct ifvlantrunk *trunk)
612 vlan_freehash(trunk);
613 trunk->parent->if_vlantrunk = NULL;
614 TRUNK_LOCK_DESTROY(trunk);
615 if_rele(trunk->parent);
620 * Program our multicast filter. What we're actually doing is
621 * programming the multicast filter of the parent. This has the
622 * side effect of causing the parent interface to receive multicast
623 * traffic that it doesn't really want, which ends up being discarded
624 * later by the upper protocol layers. Unfortunately, there's no way
625 * to avoid this: there really is only one physical interface.
628 vlan_setmulti(struct ifnet *ifp)
631 struct ifmultiaddr *ifma;
633 struct vlan_mc_entry *mc;
638 /* Find the parent. */
642 CURVNET_SET_QUIET(ifp_p->if_vnet);
644 /* First, remove any existing filter entries. */
645 while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
646 CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
647 (void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
648 NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
651 /* Now program new ones. */
653 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
654 if (ifma->ifma_addr->sa_family != AF_LINK)
656 mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
658 IF_ADDR_WUNLOCK(ifp);
662 bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
663 mc->mc_addr.sdl_index = ifp_p->if_index;
664 CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
666 IF_ADDR_WUNLOCK(ifp);
667 CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
668 error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
681 * A handler for interface ifnet events.
684 vlan_ifevent(void *arg __unused, struct ifnet *ifp, int event)
686 struct epoch_tracker et;
688 struct ifvlantrunk *trunk;
690 if (event != IFNET_EVENT_UPDATE_BAUDRATE)
694 trunk = ifp->if_vlantrunk;
701 VLAN_FOREACH(ifv, trunk) {
702 ifv->ifv_ifp->if_baudrate = ifp->if_baudrate;
704 TRUNK_WUNLOCK(trunk);
709 * A handler for parent interface link layer address changes.
710 * If the parent interface link layer address is changed we
711 * should also change it on all children vlans.
714 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
716 struct epoch_tracker et;
718 struct ifnet *ifv_ifp;
719 struct ifvlantrunk *trunk;
720 struct sockaddr_dl *sdl;
722 /* Need the epoch since this is run on taskqueue_swi. */
724 trunk = ifp->if_vlantrunk;
731 * OK, it's a trunk. Loop over and change all vlan's lladdrs on it.
732 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
733 * ioctl calls on the parent garbling the lladdr of the child vlan.
736 VLAN_FOREACH(ifv, trunk) {
738 * Copy new new lladdr into the ifv_ifp, enqueue a task
739 * to actually call if_setlladdr. if_setlladdr needs to
740 * be deferred to a taskqueue because it will call into
741 * the if_vlan ioctl path and try to acquire the global
744 ifv_ifp = ifv->ifv_ifp;
745 bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
747 sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
748 sdl->sdl_alen = ifp->if_addrlen;
749 taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
751 TRUNK_WUNLOCK(trunk);
756 * A handler for network interface departure events.
757 * Track departure of trunks here so that we don't access invalid
758 * pointers or whatever if a trunk is ripped from under us, e.g.,
759 * by ejecting its hot-plug card. However, if an ifnet is simply
760 * being renamed, then there's no need to tear down the state.
763 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
766 struct ifvlantrunk *trunk;
768 /* If the ifnet is just being renamed, don't do anything. */
769 if (ifp->if_flags & IFF_RENAMING)
772 trunk = ifp->if_vlantrunk;
779 * OK, it's a trunk. Loop over and detach all vlan's on it.
780 * Check trunk pointer after each vlan_unconfig() as it will
781 * free it and set to NULL after the last vlan was detached.
783 VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
784 ifp->if_vlantrunk == NULL)
785 vlan_unconfig_locked(ifv->ifv_ifp, 1);
787 /* Trunk should have been destroyed in vlan_unconfig(). */
788 KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
793 * Return the trunk device for a virtual interface.
795 static struct ifnet *
796 vlan_trunkdev(struct ifnet *ifp)
802 if (ifp->if_type != IFT_L2VLAN)
813 * Return the 12-bit VLAN VID for this interface, for use by external
814 * components such as Infiniband.
816 * XXXRW: Note that the function name here is historical; it should be named
820 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
824 if (ifp->if_type != IFT_L2VLAN)
827 *vidp = ifv->ifv_vid;
832 vlan_pcp(struct ifnet *ifp, uint16_t *pcpp)
836 if (ifp->if_type != IFT_L2VLAN)
839 *pcpp = ifv->ifv_pcp;
844 * Return a driver specific cookie for this interface. Synchronization
845 * with setcookie must be provided by the driver.
848 vlan_cookie(struct ifnet *ifp)
852 if (ifp->if_type != IFT_L2VLAN)
855 return (ifv->ifv_cookie);
859 * Store a cookie in our softc that drivers can use to store driver
860 * private per-instance data in.
863 vlan_setcookie(struct ifnet *ifp, void *cookie)
867 if (ifp->if_type != IFT_L2VLAN)
870 ifv->ifv_cookie = cookie;
875 * Return the vlan device present at the specific VID.
877 static struct ifnet *
878 vlan_devat(struct ifnet *ifp, uint16_t vid)
880 struct ifvlantrunk *trunk;
885 trunk = ifp->if_vlantrunk;
889 ifv = vlan_gethash(trunk, vid);
896 * VLAN support can be loaded as a module. The only place in the
897 * system that's intimately aware of this is ether_input. We hook
898 * into this code through vlan_input_p which is defined there and
899 * set here. No one else in the system should be aware of this so
900 * we use an explicit reference here.
902 extern void (*vlan_input_p)(struct ifnet *, struct mbuf *);
904 /* For if_link_state_change() eyes only... */
905 extern void (*vlan_link_state_p)(struct ifnet *);
907 static struct if_clone_addreq_v2 vlan_addreq = {
909 .match_f = vlan_clone_match,
910 .create_f = vlan_clone_create,
911 .destroy_f = vlan_clone_destroy,
912 .create_nl_f = vlan_clone_create_nl,
913 .modify_nl_f = vlan_clone_modify_nl,
914 .dump_nl_f = vlan_clone_dump_nl,
918 vlan_modevent(module_t mod, int type, void *data)
923 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
924 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
925 if (ifdetach_tag == NULL)
927 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
928 vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
929 if (iflladdr_tag == NULL)
931 ifevent_tag = EVENTHANDLER_REGISTER(ifnet_event,
932 vlan_ifevent, NULL, EVENTHANDLER_PRI_ANY);
933 if (ifevent_tag == NULL)
936 vlan_input_p = vlan_input;
937 vlan_link_state_p = vlan_link_state;
938 vlan_trunk_cap_p = vlan_trunk_capabilities;
939 vlan_trunkdev_p = vlan_trunkdev;
940 vlan_cookie_p = vlan_cookie;
941 vlan_setcookie_p = vlan_setcookie;
942 vlan_tag_p = vlan_tag;
943 vlan_pcp_p = vlan_pcp;
944 vlan_devat_p = vlan_devat;
946 vlan_cloner = ifc_attach_cloner(vlanname, (struct if_clone_addreq *)&vlan_addreq);
949 printf("vlan: initialized, using "
953 "hash tables with chaining"
960 ifc_detach_cloner(vlan_cloner);
962 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
963 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
964 EVENTHANDLER_DEREGISTER(ifnet_event, ifevent_tag);
966 vlan_link_state_p = NULL;
967 vlan_trunk_cap_p = NULL;
968 vlan_trunkdev_p = NULL;
970 vlan_cookie_p = NULL;
971 vlan_setcookie_p = NULL;
973 VLAN_LOCKING_DESTROY();
975 printf("vlan: unloaded\n");
983 static moduledata_t vlan_mod = {
989 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
990 MODULE_VERSION(if_vlan, 3);
994 vnet_vlan_init(const void *unused __unused)
996 vlan_cloner = ifc_attach_cloner(vlanname, (struct if_clone_addreq *)&vlan_addreq);
997 V_vlan_cloner = vlan_cloner;
999 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
1000 vnet_vlan_init, NULL);
1003 vnet_vlan_uninit(const void *unused __unused)
1006 ifc_detach_cloner(V_vlan_cloner);
1008 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
1009 vnet_vlan_uninit, NULL);
1013 * Check for <etherif>.<vlan>[.<vlan> ...] style interface names.
1015 static struct ifnet *
1016 vlan_clone_match_ethervid(const char *name, int *vidp)
1018 char ifname[IFNAMSIZ];
1023 strlcpy(ifname, name, IFNAMSIZ);
1024 if ((cp = strrchr(ifname, '.')) == NULL)
1027 if ((ifp = ifunit_ref(ifname)) == NULL)
1030 if (*++cp == '\0') {
1035 for(; *cp >= '0' && *cp <= '9'; cp++)
1036 vid = (vid * 10) + (*cp - '0');
1048 vlan_clone_match(struct if_clone *ifc, const char *name)
1053 ifp = vlan_clone_match_ethervid(name, NULL);
1059 if (strncmp(vlanname, name, strlen(vlanname)) != 0)
1061 for (cp = name + 4; *cp != '\0'; cp++) {
1062 if (*cp < '0' || *cp > '9')
1070 vlan_clone_create(struct if_clone *ifc, char *name, size_t len,
1071 struct ifc_data *ifd, struct ifnet **ifpp)
1074 bool wildcard = false;
1075 bool subinterface = false;
1079 uint16_t proto = ETHERTYPE_VLAN;
1082 struct ifnet *p = NULL;
1084 struct sockaddr_dl *sdl;
1086 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
1090 * There are three ways to specify the cloned device:
1091 * o pass a parameter block with the clone request.
1092 * o specify parameters in the text of the clone device name
1093 * o specify no parameters and get an unattached device that
1094 * must be configured separately.
1095 * The first technique is preferred; the latter two are supported
1096 * for backwards compatibility.
1098 * XXXRW: Note historic use of the word "tag" here. New ioctls may be
1102 if (ifd->params != NULL) {
1103 error = ifc_copyin(ifd, &vlr, sizeof(vlr));
1107 proto = vlr.vlr_proto;
1109 proto = ETHERTYPE_VLAN;
1110 p = ifunit_ref(vlr.vlr_parent);
1115 if ((error = ifc_name2unit(name, &unit)) == 0) {
1118 * vlanX interface. Set wildcard to true if the unit number
1121 wildcard = (unit < 0);
1123 struct ifnet *p_tmp = vlan_clone_match_ethervid(name, &vid);
1124 if (p_tmp != NULL) {
1126 subinterface = true;
1127 unit = IF_DUNIT_NONE;
1145 if (!subinterface) {
1146 /* vlanX interface, mark X as busy or allocate new unit # */
1147 error = ifc_alloc_unit(ifc, &unit);
1155 /* In the wildcard case, we need to update the name. */
1157 for (dp = name; *dp != '\0'; dp++);
1158 if (snprintf(dp, len - (dp-name), "%d", unit) >
1159 len - (dp-name) - 1) {
1160 panic("%s: interface name too long", __func__);
1164 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
1165 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
1168 ifc_free_unit(ifc, unit);
1174 CK_SLIST_INIT(&ifv->vlan_mc_listhead);
1175 ifp->if_softc = ifv;
1177 * Set the name manually rather than using if_initname because
1178 * we don't conform to the default naming convention for interfaces.
1180 strlcpy(ifp->if_xname, name, IFNAMSIZ);
1181 ifp->if_dname = vlanname;
1182 ifp->if_dunit = unit;
1184 ifp->if_init = vlan_init;
1186 ifp->if_start = vlan_altq_start;
1187 ifp->if_transmit = vlan_altq_transmit;
1188 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
1189 ifp->if_snd.ifq_drv_maxlen = 0;
1190 IFQ_SET_READY(&ifp->if_snd);
1192 ifp->if_transmit = vlan_transmit;
1194 ifp->if_qflush = vlan_qflush;
1195 ifp->if_ioctl = vlan_ioctl;
1196 #if defined(KERN_TLS) || defined(RATELIMIT)
1197 ifp->if_snd_tag_alloc = vlan_snd_tag_alloc;
1198 ifp->if_ratelimit_query = vlan_ratelimit_query;
1200 ifp->if_flags = VLAN_IFFLAGS;
1201 ether_ifattach(ifp, eaddr);
1202 /* Now undo some of the damage... */
1203 ifp->if_baudrate = 0;
1204 ifp->if_type = IFT_L2VLAN;
1205 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
1207 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1208 sdl->sdl_type = IFT_L2VLAN;
1211 error = vlan_config(ifv, p, vid, proto);
1215 * Since we've partially failed, we need to back
1216 * out all the way, otherwise userland could get
1217 * confused. Thus, we destroy the interface.
1219 ether_ifdetach(ifp);
1223 ifc_free_unit(ifc, unit);
1236 * Parsers of IFLA_INFO_DATA inside IFLA_LINKINFO of RTM_NEWLINK
1237 * {{nla_len=8, nla_type=IFLA_LINK}, 2},
1238 * {{nla_len=12, nla_type=IFLA_IFNAME}, "xvlan22"},
1239 * {{nla_len=24, nla_type=IFLA_LINKINFO},
1241 * {{nla_len=8, nla_type=IFLA_INFO_KIND}, "vlan"...},
1242 * {{nla_len=12, nla_type=IFLA_INFO_DATA}, "\x06\x00\x01\x00\x16\x00\x00\x00"}]}
1245 struct nl_parsed_vlan {
1247 uint16_t vlan_proto;
1248 struct ifla_vlan_flags vlan_flags;
1251 #define _OUT(_field) offsetof(struct nl_parsed_vlan, _field)
1252 static const struct nlattr_parser nla_p_vlan[] = {
1253 { .type = IFLA_VLAN_ID, .off = _OUT(vlan_id), .cb = nlattr_get_uint16 },
1254 { .type = IFLA_VLAN_FLAGS, .off = _OUT(vlan_flags), .cb = nlattr_get_nla },
1255 { .type = IFLA_VLAN_PROTOCOL, .off = _OUT(vlan_proto), .cb = nlattr_get_uint16 },
1258 NL_DECLARE_ATTR_PARSER(vlan_parser, nla_p_vlan);
1261 vlan_clone_create_nl(struct if_clone *ifc, char *name, size_t len,
1262 struct ifc_data_nl *ifd)
1264 struct epoch_tracker et;
1265 struct ifnet *ifp_parent;
1266 struct nl_pstate *npt = ifd->npt;
1267 struct nl_parsed_link *lattrs = ifd->lattrs;
1271 * lattrs.ifla_ifname is the new interface name
1272 * lattrs.ifi_index contains parent interface index
1273 * lattrs.ifla_idata contains un-parsed vlan data
1275 struct nl_parsed_vlan attrs = {
1277 .vlan_proto = ETHERTYPE_VLAN
1280 if (lattrs->ifla_idata == NULL) {
1281 nlmsg_report_err_msg(npt, "vlan id is required, guessing not supported");
1285 error = nl_parse_nested(lattrs->ifla_idata, &vlan_parser, npt, &attrs);
1288 if (attrs.vlan_id > 4095) {
1289 nlmsg_report_err_msg(npt, "Invalid VID: %d", attrs.vlan_id);
1292 if (attrs.vlan_proto != ETHERTYPE_VLAN && attrs.vlan_proto != ETHERTYPE_QINQ) {
1293 nlmsg_report_err_msg(npt, "Unsupported ethertype: 0x%04X", attrs.vlan_proto);
1297 struct vlanreq params = {
1298 .vlr_tag = attrs.vlan_id,
1299 .vlr_proto = attrs.vlan_proto,
1301 struct ifc_data ifd_new = { .flags = IFC_F_SYSSPACE, .unit = ifd->unit, .params = ¶ms };
1303 NET_EPOCH_ENTER(et);
1304 ifp_parent = ifnet_byindex(lattrs->ifi_index);
1305 if (ifp_parent != NULL)
1306 strlcpy(params.vlr_parent, if_name(ifp_parent), sizeof(params.vlr_parent));
1309 if (ifp_parent == NULL) {
1310 nlmsg_report_err_msg(npt, "unable to find parent interface %u", lattrs->ifi_index);
1314 error = vlan_clone_create(ifc, name, len, &ifd_new, &ifd->ifp);
1320 vlan_clone_modify_nl(struct ifnet *ifp, struct ifc_data_nl *ifd)
1322 struct nl_parsed_link *lattrs = ifd->lattrs;
1324 if ((lattrs->ifla_idata != NULL) && ((ifd->flags & IFC_F_CREATE) == 0)) {
1325 struct epoch_tracker et;
1326 struct nl_parsed_vlan attrs = {
1327 .vlan_proto = ETHERTYPE_VLAN,
1331 error = nl_parse_nested(lattrs->ifla_idata, &vlan_parser, ifd->npt, &attrs);
1335 NET_EPOCH_ENTER(et);
1336 struct ifnet *ifp_parent = ifnet_byindex_ref(lattrs->ifla_link);
1339 if (ifp_parent == NULL) {
1340 nlmsg_report_err_msg(ifd->npt, "unable to find parent interface %u",
1345 struct ifvlan *ifv = ifp->if_softc;
1346 error = vlan_config(ifv, ifp_parent, attrs.vlan_id, attrs.vlan_proto);
1348 if_rele(ifp_parent);
1353 return (nl_modify_ifp_generic(ifp, ifd->lattrs, ifd->bm, ifd->npt));
1357 * {{nla_len=24, nla_type=IFLA_LINKINFO},
1359 * {{nla_len=8, nla_type=IFLA_INFO_KIND}, "vlan"...},
1360 * {{nla_len=12, nla_type=IFLA_INFO_DATA}, "\x06\x00\x01\x00\x16\x00\x00\x00"}]}
1363 vlan_clone_dump_nl(struct ifnet *ifp, struct nl_writer *nw)
1365 uint32_t parent_index = 0;
1366 uint16_t vlan_id = 0;
1367 uint16_t vlan_proto = 0;
1370 struct ifvlan *ifv = ifp->if_softc;
1371 if (TRUNK(ifv) != NULL)
1372 parent_index = PARENT(ifv)->if_index;
1373 vlan_id = ifv->ifv_vid;
1374 vlan_proto = ifv->ifv_proto;
1377 if (parent_index != 0)
1378 nlattr_add_u32(nw, IFLA_LINK, parent_index);
1380 int off = nlattr_add_nested(nw, IFLA_LINKINFO);
1382 nlattr_add_string(nw, IFLA_INFO_KIND, "vlan");
1383 int off2 = nlattr_add_nested(nw, IFLA_INFO_DATA);
1385 nlattr_add_u16(nw, IFLA_VLAN_ID, vlan_id);
1386 nlattr_add_u16(nw, IFLA_VLAN_PROTOCOL, vlan_proto);
1387 nlattr_set_len(nw, off2);
1389 nlattr_set_len(nw, off);
1394 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp, uint32_t flags)
1396 struct ifvlan *ifv = ifp->if_softc;
1397 int unit = ifp->if_dunit;
1399 if (ifp->if_vlantrunk)
1403 IFQ_PURGE(&ifp->if_snd);
1405 ether_ifdetach(ifp); /* first, remove it from system-wide lists */
1406 vlan_unconfig(ifp); /* now it can be unconfigured and freed */
1408 * We should have the only reference to the ifv now, so we can now
1409 * drain any remaining lladdr task before freeing the ifnet and the
1412 taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
1416 if (unit != IF_DUNIT_NONE)
1417 ifc_free_unit(ifc, unit);
1423 * The ifp->if_init entry point for vlan(4) is a no-op.
1426 vlan_init(void *foo __unused)
1431 * The if_transmit method for vlan(4) interface.
1434 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1438 int error, len, mcast;
1442 ifv = ifp->if_softc;
1443 if (TRUNK(ifv) == NULL) {
1444 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1449 len = m->m_pkthdr.len;
1450 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1454 #if defined(KERN_TLS) || defined(RATELIMIT)
1455 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
1456 struct vlan_snd_tag *vst;
1457 struct m_snd_tag *mst;
1459 MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
1460 mst = m->m_pkthdr.snd_tag;
1461 vst = mst_to_vst(mst);
1462 if (vst->tag->ifp != p) {
1463 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1468 m->m_pkthdr.snd_tag = m_snd_tag_ref(vst->tag);
1469 m_snd_tag_rele(mst);
1474 * Do not run parent's if_transmit() if the parent is not up,
1475 * or parent's driver will cause a system crash.
1477 if (!UP_AND_RUNNING(p)) {
1478 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1483 if (!ether_8021q_frame(&m, ifp, p, &ifv->ifv_qtag)) {
1484 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1489 * Send it, precisely as ether_output() would have.
1491 error = (p->if_transmit)(p, m);
1493 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1494 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1495 if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1497 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1502 vlan_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1511 * Find the first non-VLAN parent interface.
1513 ifv = ifp->if_softc;
1515 if (TRUNK(ifv) == NULL) {
1521 } while (p->if_type == IFT_L2VLAN);
1523 return p->if_output(ifp, m, dst, ro);
1528 vlan_altq_start(if_t ifp)
1530 struct ifaltq *ifq = &ifp->if_snd;
1534 IFQ_DEQUEUE_NOLOCK(ifq, m);
1536 vlan_transmit(ifp, m);
1537 IFQ_DEQUEUE_NOLOCK(ifq, m);
1543 vlan_altq_transmit(if_t ifp, struct mbuf *m)
1547 if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
1548 IFQ_ENQUEUE(&ifp->if_snd, m, err);
1550 vlan_altq_start(ifp);
1552 err = vlan_transmit(ifp, m);
1559 * The ifp->if_qflush entry point for vlan(4) is a no-op.
1562 vlan_qflush(struct ifnet *ifp __unused)
1567 vlan_input(struct ifnet *ifp, struct mbuf *m)
1569 struct ifvlantrunk *trunk;
1576 trunk = ifp->if_vlantrunk;
1577 if (trunk == NULL) {
1582 if (m->m_flags & M_VLANTAG) {
1584 * Packet is tagged, but m contains a normal
1585 * Ethernet frame; the tag is stored out-of-band.
1587 tag = m->m_pkthdr.ether_vtag;
1588 m->m_flags &= ~M_VLANTAG;
1590 struct ether_vlan_header *evl;
1593 * Packet is tagged in-band as specified by 802.1q.
1595 switch (ifp->if_type) {
1597 if (m->m_len < sizeof(*evl) &&
1598 (m = m_pullup(m, sizeof(*evl))) == NULL) {
1599 if_printf(ifp, "cannot pullup VLAN header\n");
1602 evl = mtod(m, struct ether_vlan_header *);
1603 tag = ntohs(evl->evl_tag);
1606 * Remove the 802.1q header by copying the Ethernet
1607 * addresses over it and adjusting the beginning of
1608 * the data in the mbuf. The encapsulated Ethernet
1609 * type field is already in place.
1611 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1612 ETHER_HDR_LEN - ETHER_TYPE_LEN);
1613 m_adj(m, ETHER_VLAN_ENCAP_LEN);
1618 panic("%s: %s has unsupported if_type %u",
1619 __func__, ifp->if_xname, ifp->if_type);
1621 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1627 vid = EVL_VLANOFTAG(tag);
1629 ifv = vlan_gethash(trunk, vid);
1630 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1631 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1636 if (V_vlan_mtag_pcp) {
1638 * While uncommon, it is possible that we will find a 802.1q
1639 * packet encapsulated inside another packet that also had an
1640 * 802.1q header. For example, ethernet tunneled over IPSEC
1641 * arriving over ethernet. In that case, we replace the
1642 * existing 802.1q PCP m_tag value.
1644 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1646 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1647 sizeof(uint8_t), M_NOWAIT);
1649 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1653 m_tag_prepend(m, mtag);
1655 *(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1658 m->m_pkthdr.rcvif = ifv->ifv_ifp;
1659 if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1661 /* Pass it back through the parent's input routine. */
1662 (*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
1666 vlan_lladdr_fn(void *arg, int pending __unused)
1671 ifv = (struct ifvlan *)arg;
1674 CURVNET_SET(ifp->if_vnet);
1676 /* The ifv_ifp already has the lladdr copied in. */
1677 if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1683 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid,
1686 struct epoch_tracker et;
1687 struct ifvlantrunk *trunk;
1692 * We can handle non-ethernet hardware types as long as
1693 * they handle the tagging and headers themselves.
1695 if (p->if_type != IFT_ETHER &&
1696 p->if_type != IFT_L2VLAN &&
1697 (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1698 return (EPROTONOSUPPORT);
1699 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1700 return (EPROTONOSUPPORT);
1702 * Don't let the caller set up a VLAN VID with
1703 * anything except VLID bits.
1704 * VID numbers 0x0 and 0xFFF are reserved.
1706 if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1708 if (ifv->ifv_trunk) {
1709 trunk = ifv->ifv_trunk;
1710 if (trunk->parent != p)
1715 ifv->ifv_proto = proto;
1717 if (ifv->ifv_vid != vid) {
1719 vlan_remhash(trunk, ifv);
1721 error = vlan_inshash(trunk, ifv);
1728 if (p->if_vlantrunk == NULL) {
1729 trunk = malloc(sizeof(struct ifvlantrunk),
1730 M_VLAN, M_WAITOK | M_ZERO);
1731 vlan_inithash(trunk);
1732 TRUNK_LOCK_INIT(trunk);
1734 p->if_vlantrunk = trunk;
1736 if_ref(trunk->parent);
1737 TRUNK_WUNLOCK(trunk);
1739 trunk = p->if_vlantrunk;
1742 ifv->ifv_vid = vid; /* must set this before vlan_inshash() */
1743 ifv->ifv_pcp = 0; /* Default: best effort delivery. */
1744 error = vlan_inshash(trunk, ifv);
1747 ifv->ifv_proto = proto;
1748 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1749 ifv->ifv_mintu = ETHERMIN;
1750 ifv->ifv_pflags = 0;
1751 ifv->ifv_capenable = -1;
1754 * If the parent supports the VLAN_MTU capability,
1755 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1758 if (p->if_capenable & IFCAP_VLAN_MTU) {
1760 * No need to fudge the MTU since the parent can
1761 * handle extended frames.
1763 ifv->ifv_mtufudge = 0;
1766 * Fudge the MTU by the encapsulation size. This
1767 * makes us incompatible with strictly compliant
1768 * 802.1Q implementations, but allows us to use
1769 * the feature with other NetBSD implementations,
1770 * which might still be useful.
1772 ifv->ifv_mtufudge = ifv->ifv_encaplen;
1775 ifv->ifv_trunk = trunk;
1778 * Initialize fields from our parent. This duplicates some
1779 * work with ether_ifattach() but allows for non-ethernet
1780 * interfaces to also work.
1782 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1783 ifp->if_baudrate = p->if_baudrate;
1784 ifp->if_input = p->if_input;
1785 ifp->if_resolvemulti = p->if_resolvemulti;
1786 ifp->if_addrlen = p->if_addrlen;
1787 ifp->if_broadcastaddr = p->if_broadcastaddr;
1788 ifp->if_pcp = ifv->ifv_pcp;
1791 * We wrap the parent's if_output using vlan_output to ensure that it
1792 * can't become stale.
1794 ifp->if_output = vlan_output;
1797 * Copy only a selected subset of flags from the parent.
1798 * Other flags are none of our business.
1800 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1801 ifp->if_flags &= ~VLAN_COPY_FLAGS;
1802 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1803 #undef VLAN_COPY_FLAGS
1805 ifp->if_link_state = p->if_link_state;
1807 NET_EPOCH_ENTER(et);
1808 vlan_capabilities(ifv);
1812 * Set up our interface address to reflect the underlying
1813 * physical interface's.
1815 TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
1816 ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1820 * Do not schedule link address update if it was the same
1821 * as previous parent's. This helps avoid updating for each
1822 * associated llentry.
1824 if (memcmp(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen) != 0) {
1825 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1826 taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
1829 /* We are ready for operation now. */
1830 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1832 /* Update flags on the parent, if necessary. */
1833 vlan_setflags(ifp, 1);
1836 * Configure multicast addresses that may already be
1837 * joined on the vlan device.
1839 (void)vlan_setmulti(ifp);
1843 EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1850 vlan_unconfig(struct ifnet *ifp)
1854 vlan_unconfig_locked(ifp, 0);
1859 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1861 struct ifvlantrunk *trunk;
1862 struct vlan_mc_entry *mc;
1864 struct ifnet *parent;
1867 VLAN_XLOCK_ASSERT();
1869 ifv = ifp->if_softc;
1870 trunk = ifv->ifv_trunk;
1873 if (trunk != NULL) {
1874 parent = trunk->parent;
1877 * Since the interface is being unconfigured, we need to
1878 * empty the list of multicast groups that we may have joined
1879 * while we were alive from the parent's list.
1881 while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1883 * If the parent interface is being detached,
1884 * all its multicast addresses have already
1885 * been removed. Warn about errors if
1886 * if_delmulti() does fail, but don't abort as
1887 * all callers expect vlan destruction to
1891 error = if_delmulti(parent,
1892 (struct sockaddr *)&mc->mc_addr);
1895 "Failed to delete multicast address from parent: %d\n",
1898 CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1899 NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx);
1902 vlan_setflags(ifp, 0); /* clear special flags on parent */
1904 vlan_remhash(trunk, ifv);
1905 ifv->ifv_trunk = NULL;
1908 * Check if we were the last.
1910 if (trunk->refcnt == 0) {
1911 parent->if_vlantrunk = NULL;
1913 trunk_destroy(trunk);
1917 /* Disconnect from parent. */
1918 if (ifv->ifv_pflags)
1919 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1920 ifp->if_mtu = ETHERMTU;
1921 ifp->if_link_state = LINK_STATE_UNKNOWN;
1922 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1925 * Only dispatch an event if vlan was
1926 * attached, otherwise there is nothing
1927 * to cleanup anyway.
1930 EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1933 /* Handle a reference counted flag that should be set on the parent as well */
1935 vlan_setflag(struct ifnet *ifp, int flag, int status,
1936 int (*func)(struct ifnet *, int))
1941 VLAN_SXLOCK_ASSERT();
1943 ifv = ifp->if_softc;
1944 status = status ? (ifp->if_flags & flag) : 0;
1945 /* Now "status" contains the flag value or 0 */
1948 * See if recorded parent's status is different from what
1949 * we want it to be. If it is, flip it. We record parent's
1950 * status in ifv_pflags so that we won't clear parent's flag
1951 * we haven't set. In fact, we don't clear or set parent's
1952 * flags directly, but get or release references to them.
1953 * That's why we can be sure that recorded flags still are
1954 * in accord with actual parent's flags.
1956 if (status != (ifv->ifv_pflags & flag)) {
1957 error = (*func)(PARENT(ifv), status);
1960 ifv->ifv_pflags &= ~flag;
1961 ifv->ifv_pflags |= status;
1967 * Handle IFF_* flags that require certain changes on the parent:
1968 * if "status" is true, update parent's flags respective to our if_flags;
1969 * if "status" is false, forcedly clear the flags set on parent.
1972 vlan_setflags(struct ifnet *ifp, int status)
1976 for (i = 0; vlan_pflags[i].flag; i++) {
1977 error = vlan_setflag(ifp, vlan_pflags[i].flag,
1978 status, vlan_pflags[i].func);
1985 /* Inform all vlans that their parent has changed link state */
1987 vlan_link_state(struct ifnet *ifp)
1989 struct epoch_tracker et;
1990 struct ifvlantrunk *trunk;
1993 NET_EPOCH_ENTER(et);
1994 trunk = ifp->if_vlantrunk;
1995 if (trunk == NULL) {
2001 VLAN_FOREACH(ifv, trunk) {
2002 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
2003 if_link_state_change(ifv->ifv_ifp,
2004 trunk->parent->if_link_state);
2006 TRUNK_WUNLOCK(trunk);
2011 vlan_capabilities(struct ifvlan *ifv)
2015 struct ifnet_hw_tsomax hw_tsomax;
2016 int cap = 0, ena = 0, mena;
2020 VLAN_SXLOCK_ASSERT();
2025 /* Mask parent interface enabled capabilities disabled by user. */
2026 mena = p->if_capenable & ifv->ifv_capenable;
2029 * If the parent interface can do checksum offloading
2030 * on VLANs, then propagate its hardware-assisted
2031 * checksumming flags. Also assert that checksum
2032 * offloading requires hardware VLAN tagging.
2034 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
2035 cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
2036 if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
2037 p->if_capenable & IFCAP_VLAN_HWTAGGING) {
2038 ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
2039 if (ena & IFCAP_TXCSUM)
2040 hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
2041 CSUM_UDP | CSUM_SCTP);
2042 if (ena & IFCAP_TXCSUM_IPV6)
2043 hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
2044 CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
2048 * If the parent interface can do TSO on VLANs then
2049 * propagate the hardware-assisted flag. TSO on VLANs
2050 * does not necessarily require hardware VLAN tagging.
2052 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
2053 if_hw_tsomax_common(p, &hw_tsomax);
2054 if_hw_tsomax_update(ifp, &hw_tsomax);
2055 if (p->if_capabilities & IFCAP_VLAN_HWTSO)
2056 cap |= p->if_capabilities & IFCAP_TSO;
2057 if (p->if_capenable & IFCAP_VLAN_HWTSO) {
2058 ena |= mena & IFCAP_TSO;
2059 if (ena & IFCAP_TSO)
2060 hwa |= p->if_hwassist & CSUM_TSO;
2064 * If the parent interface can do LRO and checksum offloading on
2065 * VLANs, then guess it may do LRO on VLANs. False positive here
2066 * cost nothing, while false negative may lead to some confusions.
2068 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
2069 cap |= p->if_capabilities & IFCAP_LRO;
2070 if (p->if_capenable & IFCAP_VLAN_HWCSUM)
2071 ena |= mena & IFCAP_LRO;
2074 * If the parent interface can offload TCP connections over VLANs then
2075 * propagate its TOE capability to the VLAN interface.
2077 * All TOE drivers in the tree today can deal with VLANs. If this
2078 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
2081 #define IFCAP_VLAN_TOE IFCAP_TOE
2082 if (p->if_capabilities & IFCAP_VLAN_TOE)
2083 cap |= p->if_capabilities & IFCAP_TOE;
2084 if (p->if_capenable & IFCAP_VLAN_TOE) {
2085 SETTOEDEV(ifp, TOEDEV(p));
2086 ena |= mena & IFCAP_TOE;
2090 * If the parent interface supports dynamic link state, so does the
2093 cap |= (p->if_capabilities & IFCAP_LINKSTATE);
2094 ena |= (mena & IFCAP_LINKSTATE);
2098 * If the parent interface supports ratelimiting, so does the
2101 cap |= (p->if_capabilities & IFCAP_TXRTLMT);
2102 ena |= (mena & IFCAP_TXRTLMT);
2106 * If the parent interface supports unmapped mbufs, so does
2107 * the VLAN interface. Note that this should be fine even for
2108 * interfaces that don't support hardware tagging as headers
2109 * are prepended in normal mbufs to unmapped mbufs holding
2112 cap |= (p->if_capabilities & IFCAP_MEXTPG);
2113 ena |= (mena & IFCAP_MEXTPG);
2116 * If the parent interface can offload encryption and segmentation
2117 * of TLS records over TCP, propagate it's capability to the VLAN
2120 * All TLS drivers in the tree today can deal with VLANs. If
2121 * this ever changes, then a new IFCAP_VLAN_TXTLS can be
2124 if (p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
2125 cap |= p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
2126 if (p->if_capenable & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT))
2127 ena |= mena & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT);
2129 ifp->if_capabilities = cap;
2130 ifp->if_capenable = ena;
2131 ifp->if_hwassist = hwa;
2135 vlan_trunk_capabilities(struct ifnet *ifp)
2137 struct epoch_tracker et;
2138 struct ifvlantrunk *trunk;
2142 trunk = ifp->if_vlantrunk;
2143 if (trunk == NULL) {
2147 NET_EPOCH_ENTER(et);
2148 VLAN_FOREACH(ifv, trunk)
2149 vlan_capabilities(ifv);
2155 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
2163 struct ifvlantrunk *trunk;
2165 int error = 0, oldmtu;
2167 ifr = (struct ifreq *)data;
2169 ifa = (struct ifaddr *) data;
2171 ifv = ifp->if_softc;
2175 ifp->if_flags |= IFF_UP;
2177 if (ifa->ifa_addr->sa_family == AF_INET)
2178 arp_ifinit(ifp, ifa);
2182 bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
2187 if (TRUNK(ifv) != NULL) {
2190 error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
2192 /* Limit the result to the parent's current config. */
2194 struct ifmediareq *ifmr;
2196 ifmr = (struct ifmediareq *)data;
2197 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
2198 ifmr->ifm_count = 1;
2199 error = copyout(&ifmr->ifm_current,
2216 * Set the interface MTU.
2220 if (trunk != NULL) {
2223 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
2225 (ifv->ifv_mintu - ifv->ifv_mtufudge))
2228 ifp->if_mtu = ifr->ifr_mtu;
2229 TRUNK_WUNLOCK(trunk);
2238 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
2239 * interface to be delegated to a jail without allowing the
2240 * jail to change what underlying interface/VID it is
2241 * associated with. We are not entirely convinced that this
2242 * is the right way to accomplish that policy goal.
2244 if (ifp->if_vnet != ifp->if_home_vnet) {
2249 error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
2252 if (vlr.vlr_parent[0] == '\0') {
2256 p = ifunit_ref(vlr.vlr_parent);
2261 if (vlr.vlr_proto == 0)
2262 vlr.vlr_proto = ETHERTYPE_VLAN;
2263 oldmtu = ifp->if_mtu;
2264 error = vlan_config(ifv, p, vlr.vlr_tag, vlr.vlr_proto);
2268 * VLAN MTU may change during addition of the vlandev.
2269 * If it did, do network layer specific procedure.
2271 if (ifp->if_mtu != oldmtu)
2277 if (ifp->if_vnet != ifp->if_home_vnet) {
2282 bzero(&vlr, sizeof(vlr));
2284 if (TRUNK(ifv) != NULL) {
2285 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
2286 sizeof(vlr.vlr_parent));
2287 vlr.vlr_tag = ifv->ifv_vid;
2288 vlr.vlr_proto = ifv->ifv_proto;
2291 error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
2296 * We should propagate selected flags to the parent,
2297 * e.g., promiscuous mode.
2300 if (TRUNK(ifv) != NULL)
2301 error = vlan_setflags(ifp, 1);
2308 * If we don't have a parent, just remember the membership for
2311 * XXX We need the rmlock here to avoid sleeping while
2312 * holding in6_multi_mtx.
2317 error = vlan_setmulti(ifp);
2323 if (ifp->if_vnet != ifp->if_home_vnet) {
2328 ifr->ifr_vlan_pcp = ifv->ifv_pcp;
2333 if (ifp->if_vnet != ifp->if_home_vnet) {
2338 error = priv_check(curthread, PRIV_NET_SETVLANPCP);
2341 if (ifr->ifr_vlan_pcp > VLAN_PCP_MAX) {
2345 ifv->ifv_pcp = ifr->ifr_vlan_pcp;
2346 ifp->if_pcp = ifv->ifv_pcp;
2347 /* broadcast event about PCP change */
2348 EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
2353 ifv->ifv_capenable = ifr->ifr_reqcap;
2355 if (trunk != NULL) {
2356 struct epoch_tracker et;
2358 NET_EPOCH_ENTER(et);
2359 vlan_capabilities(ifv);
2373 #if defined(KERN_TLS) || defined(RATELIMIT)
2375 vlan_snd_tag_alloc(struct ifnet *ifp,
2376 union if_snd_tag_alloc_params *params,
2377 struct m_snd_tag **ppmt)
2379 struct epoch_tracker et;
2380 const struct if_snd_tag_sw *sw;
2381 struct vlan_snd_tag *vst;
2383 struct ifnet *parent;
2384 struct m_snd_tag *mst;
2387 NET_EPOCH_ENTER(et);
2388 ifv = ifp->if_softc;
2390 switch (params->hdr.type) {
2392 case IF_SND_TAG_TYPE_UNLIMITED:
2393 sw = &vlan_snd_tag_ul_sw;
2395 case IF_SND_TAG_TYPE_RATE_LIMIT:
2396 sw = &vlan_snd_tag_rl_sw;
2400 case IF_SND_TAG_TYPE_TLS:
2401 sw = &vlan_snd_tag_tls_sw;
2403 case IF_SND_TAG_TYPE_TLS_RX:
2405 if (params->tls_rx.vlan_id != 0)
2407 params->tls_rx.vlan_id = ifv->ifv_vid;
2410 case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
2411 sw = &vlan_snd_tag_tls_rl_sw;
2419 if (ifv->ifv_trunk != NULL)
2420 parent = PARENT(ifv);
2429 vst = malloc(sizeof(*vst), M_VLAN, M_NOWAIT);
2437 error = m_snd_tag_alloc(parent, params, &mst);
2445 m_snd_tag_init(&vst->com, ifp, sw);
2455 return (EOPNOTSUPP);
2458 static struct m_snd_tag *
2459 vlan_next_snd_tag(struct m_snd_tag *mst)
2461 struct vlan_snd_tag *vst;
2463 vst = mst_to_vst(mst);
2468 vlan_snd_tag_modify(struct m_snd_tag *mst,
2469 union if_snd_tag_modify_params *params)
2471 struct vlan_snd_tag *vst;
2473 vst = mst_to_vst(mst);
2474 return (vst->tag->sw->snd_tag_modify(vst->tag, params));
2478 vlan_snd_tag_query(struct m_snd_tag *mst,
2479 union if_snd_tag_query_params *params)
2481 struct vlan_snd_tag *vst;
2483 vst = mst_to_vst(mst);
2484 return (vst->tag->sw->snd_tag_query(vst->tag, params));
2488 vlan_snd_tag_free(struct m_snd_tag *mst)
2490 struct vlan_snd_tag *vst;
2492 vst = mst_to_vst(mst);
2493 m_snd_tag_rele(vst->tag);
2498 vlan_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
2501 * For vlan, we have an indirect
2502 * interface. The caller needs to
2503 * get a ratelimit tag on the actual
2504 * interface the flow will go on.
2506 q->rate_table = NULL;
2507 q->flags = RT_IS_INDIRECT;
2509 q->number_of_rates = 0;