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("$FreeBSD$");
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_clone.h>
74 #include <net/if_dl.h>
75 #include <net/if_types.h>
76 #include <net/if_vlan_var.h>
80 #include <netinet/in.h>
81 #include <netinet/if_ether.h>
84 #define VLAN_DEF_HWIDTH 4
85 #define VLAN_IFFLAGS (IFF_BROADCAST | IFF_MULTICAST)
87 #define UP_AND_RUNNING(ifp) \
88 ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
90 CK_SLIST_HEAD(ifvlanhead, ifvlan);
93 struct ifnet *parent; /* parent interface of this trunk */
96 #define VLAN_ARRAY_SIZE (EVL_VLID_MASK + 1)
97 struct ifvlan *vlans[VLAN_ARRAY_SIZE]; /* static table */
99 struct ifvlanhead *hash; /* dynamic hash-list table */
107 * This macro provides a facility to iterate over every vlan on a trunk with
108 * the assumption that none will be added/removed during iteration.
111 #define VLAN_FOREACH(_ifv, _trunk) \
113 for (_i = 0; _i < VLAN_ARRAY_SIZE; _i++) \
114 if (((_ifv) = (_trunk)->vlans[_i]) != NULL)
115 #else /* VLAN_ARRAY */
116 #define VLAN_FOREACH(_ifv, _trunk) \
117 struct ifvlan *_next; \
119 for (_i = 0; _i < (1 << (_trunk)->hwidth); _i++) \
120 CK_SLIST_FOREACH_SAFE((_ifv), &(_trunk)->hash[_i], ifv_list, _next)
121 #endif /* VLAN_ARRAY */
124 * This macro provides a facility to iterate over every vlan on a trunk while
125 * also modifying the number of vlans on the trunk. The iteration continues
126 * until some condition is met or there are no more vlans on the trunk.
129 /* The VLAN_ARRAY case is simple -- just a for loop using the condition. */
130 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
132 for (_i = 0; !(_cond) && _i < VLAN_ARRAY_SIZE; _i++) \
133 if (((_ifv) = (_trunk)->vlans[_i]))
134 #else /* VLAN_ARRAY */
136 * The hash table case is more complicated. We allow for the hash table to be
137 * modified (i.e. vlans removed) while we are iterating over it. To allow for
138 * this we must restart the iteration every time we "touch" something during
139 * the iteration, since removal will resize the hash table and invalidate our
140 * current position. If acting on the touched element causes the trunk to be
141 * emptied, then iteration also stops.
143 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \
145 bool _touch = false; \
147 !(_cond) && _i < (1 << (_trunk)->hwidth); \
148 _i = (_touch && ((_trunk) != NULL) ? 0 : _i + 1), _touch = false) \
149 if (((_ifv) = CK_SLIST_FIRST(&(_trunk)->hash[_i])) != NULL && \
151 #endif /* VLAN_ARRAY */
153 struct vlan_mc_entry {
154 struct sockaddr_dl mc_addr;
155 CK_SLIST_ENTRY(vlan_mc_entry) mc_entries;
156 struct epoch_context mc_epoch_ctx;
160 struct ifvlantrunk *ifv_trunk;
161 struct ifnet *ifv_ifp;
162 #define TRUNK(ifv) ((ifv)->ifv_trunk)
163 #define PARENT(ifv) ((ifv)->ifv_trunk->parent)
165 int ifv_pflags; /* special flags we have set on parent */
167 int ifv_encaplen; /* encapsulation length */
168 int ifv_mtufudge; /* MTU fudged by this much */
169 int ifv_mintu; /* min transmission unit */
170 uint16_t ifv_proto; /* encapsulation ethertype */
171 uint16_t ifv_tag; /* tag to apply on packets leaving if */
172 uint16_t ifv_vid; /* VLAN ID */
173 uint8_t ifv_pcp; /* Priority Code Point (PCP). */
174 struct task lladdr_task;
175 CK_SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
177 CK_SLIST_ENTRY(ifvlan) ifv_list;
181 /* Special flags we should propagate to parent. */
184 int (*func)(struct ifnet *, int);
186 {IFF_PROMISC, ifpromisc},
187 {IFF_ALLMULTI, if_allmulti},
191 extern int vlan_mtag_pcp;
193 static const char vlanname[] = "vlan";
194 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
196 static eventhandler_tag ifdetach_tag;
197 static eventhandler_tag iflladdr_tag;
200 * if_vlan uses two module-level synchronizations primitives to allow concurrent
201 * modification of vlan interfaces and (mostly) allow for vlans to be destroyed
202 * while they are being used for tx/rx. To accomplish this in a way that has
203 * acceptable performance and cooperation with other parts of the network stack
204 * there is a non-sleepable epoch(9) and an sx(9).
206 * The performance-sensitive paths that warrant using the epoch(9) are
207 * vlan_transmit and vlan_input. Both have to check for the vlan interface's
208 * existence using if_vlantrunk, and being in the network tx/rx paths the use
209 * of an epoch(9) gives a measureable improvement in performance.
211 * The reason for having an sx(9) is mostly because there are still areas that
212 * must be sleepable and also have safe concurrent access to a vlan interface.
213 * Since the sx(9) exists, it is used by default in most paths unless sleeping
214 * is not permitted, or if it is not clear whether sleeping is permitted.
217 #define _VLAN_SX_ID ifv_sx
219 static struct sx _VLAN_SX_ID;
221 #define VLAN_LOCKING_INIT() \
222 sx_init(&_VLAN_SX_ID, "vlan_sx")
224 #define VLAN_LOCKING_DESTROY() \
225 sx_destroy(&_VLAN_SX_ID)
227 #define VLAN_SLOCK() sx_slock(&_VLAN_SX_ID)
228 #define VLAN_SUNLOCK() sx_sunlock(&_VLAN_SX_ID)
229 #define VLAN_XLOCK() sx_xlock(&_VLAN_SX_ID)
230 #define VLAN_XUNLOCK() sx_xunlock(&_VLAN_SX_ID)
231 #define VLAN_SLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_SLOCKED)
232 #define VLAN_XLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_XLOCKED)
233 #define VLAN_SXLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_LOCKED)
237 * We also have a per-trunk mutex that should be acquired when changing
240 #define TRUNK_LOCK_INIT(trunk) mtx_init(&(trunk)->lock, vlanname, NULL, MTX_DEF)
241 #define TRUNK_LOCK_DESTROY(trunk) mtx_destroy(&(trunk)->lock)
242 #define TRUNK_WLOCK(trunk) mtx_lock(&(trunk)->lock)
243 #define TRUNK_WUNLOCK(trunk) mtx_unlock(&(trunk)->lock)
244 #define TRUNK_LOCK_ASSERT(trunk) MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(trunk)->lock))
245 #define TRUNK_WLOCK_ASSERT(trunk) mtx_assert(&(trunk)->lock, MA_OWNED);
248 * The VLAN_ARRAY substitutes the dynamic hash with a static array
249 * with 4096 entries. In theory this can give a boost in processing,
250 * however in practice it does not. Probably this is because the array
251 * is too big to fit into CPU cache.
254 static void vlan_inithash(struct ifvlantrunk *trunk);
255 static void vlan_freehash(struct ifvlantrunk *trunk);
256 static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
257 static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
258 static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
259 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
262 static void trunk_destroy(struct ifvlantrunk *trunk);
264 static void vlan_init(void *foo);
265 static void vlan_input(struct ifnet *ifp, struct mbuf *m);
266 static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
268 static int vlan_snd_tag_alloc(struct ifnet *,
269 union if_snd_tag_alloc_params *, struct m_snd_tag **);
270 static void vlan_snd_tag_free(struct m_snd_tag *);
272 static void vlan_qflush(struct ifnet *ifp);
273 static int vlan_setflag(struct ifnet *ifp, int flag, int status,
274 int (*func)(struct ifnet *, int));
275 static int vlan_setflags(struct ifnet *ifp, int status);
276 static int vlan_setmulti(struct ifnet *ifp);
277 static int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
278 static void vlan_unconfig(struct ifnet *ifp);
279 static void vlan_unconfig_locked(struct ifnet *ifp, int departing);
280 static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
281 static void vlan_link_state(struct ifnet *ifp);
282 static void vlan_capabilities(struct ifvlan *ifv);
283 static void vlan_trunk_capabilities(struct ifnet *ifp);
285 static struct ifnet *vlan_clone_match_ethervid(const char *, int *);
286 static int vlan_clone_match(struct if_clone *, const char *);
287 static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
288 static int vlan_clone_destroy(struct if_clone *, struct ifnet *);
290 static void vlan_ifdetach(void *arg, struct ifnet *ifp);
291 static void vlan_iflladdr(void *arg, struct ifnet *ifp);
293 static void vlan_lladdr_fn(void *arg, int pending);
295 static struct if_clone *vlan_cloner;
298 VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner);
299 #define V_vlan_cloner VNET(vlan_cloner)
303 vlan_mc_free(struct epoch_context *ctx)
305 struct vlan_mc_entry *mc = __containerof(ctx, struct vlan_mc_entry, mc_epoch_ctx);
310 #define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
313 vlan_inithash(struct ifvlantrunk *trunk)
318 * The trunk must not be locked here since we call malloc(M_WAITOK).
319 * It is OK in case this function is called before the trunk struct
320 * gets hooked up and becomes visible from other threads.
323 KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
324 ("%s: hash already initialized", __func__));
326 trunk->hwidth = VLAN_DEF_HWIDTH;
327 n = 1 << trunk->hwidth;
328 trunk->hmask = n - 1;
329 trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
330 for (i = 0; i < n; i++)
331 CK_SLIST_INIT(&trunk->hash[i]);
335 vlan_freehash(struct ifvlantrunk *trunk)
340 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
341 for (i = 0; i < (1 << trunk->hwidth); i++)
342 KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]),
343 ("%s: hash table not empty", __func__));
345 free(trunk->hash, M_VLAN);
347 trunk->hwidth = trunk->hmask = 0;
351 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
357 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
359 b = 1 << trunk->hwidth;
360 i = HASH(ifv->ifv_vid, trunk->hmask);
361 CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
362 if (ifv->ifv_vid == ifv2->ifv_vid)
366 * Grow the hash when the number of vlans exceeds half of the number of
367 * hash buckets squared. This will make the average linked-list length
370 if (trunk->refcnt > (b * b) / 2) {
371 vlan_growhash(trunk, 1);
372 i = HASH(ifv->ifv_vid, trunk->hmask);
374 CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
381 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
387 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
389 b = 1 << trunk->hwidth;
390 i = HASH(ifv->ifv_vid, trunk->hmask);
391 CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
394 CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list);
395 if (trunk->refcnt < (b * b) / 2)
396 vlan_growhash(trunk, -1);
400 panic("%s: vlan not found\n", __func__);
401 return (ENOENT); /*NOTREACHED*/
405 * Grow the hash larger or smaller if memory permits.
408 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
411 struct ifvlanhead *hash2;
412 int hwidth2, i, j, n, n2;
415 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
418 /* Harmless yet obvious coding error */
419 printf("%s: howmuch is 0\n", __func__);
423 hwidth2 = trunk->hwidth + howmuch;
424 n = 1 << trunk->hwidth;
426 /* Do not shrink the table below the default */
427 if (hwidth2 < VLAN_DEF_HWIDTH)
430 hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK);
432 printf("%s: out of memory -- hash size not changed\n",
434 return; /* We can live with the old hash table */
436 for (j = 0; j < n2; j++)
437 CK_SLIST_INIT(&hash2[j]);
438 for (i = 0; i < n; i++)
439 while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) {
440 CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list);
441 j = HASH(ifv->ifv_vid, n2 - 1);
442 CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
445 free(trunk->hash, M_VLAN);
447 trunk->hwidth = hwidth2;
448 trunk->hmask = n2 - 1;
451 if_printf(trunk->parent,
452 "VLAN hash table resized from %d to %d buckets\n", n, n2);
455 static __inline struct ifvlan *
456 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
462 CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
463 if (ifv->ifv_vid == vid)
469 /* Debugging code to view the hashtables. */
471 vlan_dumphash(struct ifvlantrunk *trunk)
476 for (i = 0; i < (1 << trunk->hwidth); i++) {
478 CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
479 printf("%s ", ifv->ifv_ifp->if_xname);
486 static __inline struct ifvlan *
487 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
490 return trunk->vlans[vid];
494 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
497 if (trunk->vlans[ifv->ifv_vid] != NULL)
499 trunk->vlans[ifv->ifv_vid] = ifv;
506 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
509 trunk->vlans[ifv->ifv_vid] = NULL;
516 vlan_freehash(struct ifvlantrunk *trunk)
521 vlan_inithash(struct ifvlantrunk *trunk)
525 #endif /* !VLAN_ARRAY */
528 trunk_destroy(struct ifvlantrunk *trunk)
532 vlan_freehash(trunk);
533 trunk->parent->if_vlantrunk = NULL;
534 TRUNK_LOCK_DESTROY(trunk);
535 if_rele(trunk->parent);
540 * Program our multicast filter. What we're actually doing is
541 * programming the multicast filter of the parent. This has the
542 * side effect of causing the parent interface to receive multicast
543 * traffic that it doesn't really want, which ends up being discarded
544 * later by the upper protocol layers. Unfortunately, there's no way
545 * to avoid this: there really is only one physical interface.
548 vlan_setmulti(struct ifnet *ifp)
551 struct ifmultiaddr *ifma;
553 struct vlan_mc_entry *mc;
558 /* Find the parent. */
562 CURVNET_SET_QUIET(ifp_p->if_vnet);
564 /* First, remove any existing filter entries. */
565 while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
566 CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
567 (void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
568 epoch_call(net_epoch_preempt, &mc->mc_epoch_ctx, vlan_mc_free);
571 /* Now program new ones. */
573 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
574 if (ifma->ifma_addr->sa_family != AF_LINK)
576 mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
578 IF_ADDR_WUNLOCK(ifp);
581 bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
582 mc->mc_addr.sdl_index = ifp_p->if_index;
583 CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
585 IF_ADDR_WUNLOCK(ifp);
586 CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
587 error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
598 * A handler for parent interface link layer address changes.
599 * If the parent interface link layer address is changed we
600 * should also change it on all children vlans.
603 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
605 struct epoch_tracker et;
607 struct ifnet *ifv_ifp;
608 struct ifvlantrunk *trunk;
609 struct sockaddr_dl *sdl;
611 /* Need the epoch since this is run on taskqueue_swi. */
613 trunk = ifp->if_vlantrunk;
620 * OK, it's a trunk. Loop over and change all vlan's lladdrs on it.
621 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR
622 * ioctl calls on the parent garbling the lladdr of the child vlan.
625 VLAN_FOREACH(ifv, trunk) {
627 * Copy new new lladdr into the ifv_ifp, enqueue a task
628 * to actually call if_setlladdr. if_setlladdr needs to
629 * be deferred to a taskqueue because it will call into
630 * the if_vlan ioctl path and try to acquire the global
633 ifv_ifp = ifv->ifv_ifp;
634 bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp),
636 sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr;
637 sdl->sdl_alen = ifp->if_addrlen;
638 taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task);
640 TRUNK_WUNLOCK(trunk);
645 * A handler for network interface departure events.
646 * Track departure of trunks here so that we don't access invalid
647 * pointers or whatever if a trunk is ripped from under us, e.g.,
648 * by ejecting its hot-plug card. However, if an ifnet is simply
649 * being renamed, then there's no need to tear down the state.
652 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
655 struct ifvlantrunk *trunk;
657 /* If the ifnet is just being renamed, don't do anything. */
658 if (ifp->if_flags & IFF_RENAMING)
661 trunk = ifp->if_vlantrunk;
668 * OK, it's a trunk. Loop over and detach all vlan's on it.
669 * Check trunk pointer after each vlan_unconfig() as it will
670 * free it and set to NULL after the last vlan was detached.
672 VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk,
673 ifp->if_vlantrunk == NULL)
674 vlan_unconfig_locked(ifv->ifv_ifp, 1);
676 /* Trunk should have been destroyed in vlan_unconfig(). */
677 KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
682 * Return the trunk device for a virtual interface.
684 static struct ifnet *
685 vlan_trunkdev(struct ifnet *ifp)
687 struct epoch_tracker et;
690 if (ifp->if_type != IFT_L2VLAN)
703 * Return the 12-bit VLAN VID for this interface, for use by external
704 * components such as Infiniband.
706 * XXXRW: Note that the function name here is historical; it should be named
710 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
714 if (ifp->if_type != IFT_L2VLAN)
717 *vidp = ifv->ifv_vid;
722 vlan_pcp(struct ifnet *ifp, uint16_t *pcpp)
726 if (ifp->if_type != IFT_L2VLAN)
729 *pcpp = ifv->ifv_pcp;
734 * Return a driver specific cookie for this interface. Synchronization
735 * with setcookie must be provided by the driver.
738 vlan_cookie(struct ifnet *ifp)
742 if (ifp->if_type != IFT_L2VLAN)
745 return (ifv->ifv_cookie);
749 * Store a cookie in our softc that drivers can use to store driver
750 * private per-instance data in.
753 vlan_setcookie(struct ifnet *ifp, void *cookie)
757 if (ifp->if_type != IFT_L2VLAN)
760 ifv->ifv_cookie = cookie;
765 * Return the vlan device present at the specific VID.
767 static struct ifnet *
768 vlan_devat(struct ifnet *ifp, uint16_t vid)
770 struct epoch_tracker et;
771 struct ifvlantrunk *trunk;
775 trunk = ifp->if_vlantrunk;
781 ifv = vlan_gethash(trunk, vid);
789 * Recalculate the cached VLAN tag exposed via the MIB.
792 vlan_tag_recalculate(struct ifvlan *ifv)
795 ifv->ifv_tag = EVL_MAKETAG(ifv->ifv_vid, ifv->ifv_pcp, 0);
799 * VLAN support can be loaded as a module. The only place in the
800 * system that's intimately aware of this is ether_input. We hook
801 * into this code through vlan_input_p which is defined there and
802 * set here. No one else in the system should be aware of this so
803 * we use an explicit reference here.
805 extern void (*vlan_input_p)(struct ifnet *, struct mbuf *);
807 /* For if_link_state_change() eyes only... */
808 extern void (*vlan_link_state_p)(struct ifnet *);
811 vlan_modevent(module_t mod, int type, void *data)
816 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
817 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
818 if (ifdetach_tag == NULL)
820 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
821 vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
822 if (iflladdr_tag == NULL)
825 vlan_input_p = vlan_input;
826 vlan_link_state_p = vlan_link_state;
827 vlan_trunk_cap_p = vlan_trunk_capabilities;
828 vlan_trunkdev_p = vlan_trunkdev;
829 vlan_cookie_p = vlan_cookie;
830 vlan_setcookie_p = vlan_setcookie;
831 vlan_tag_p = vlan_tag;
832 vlan_pcp_p = vlan_pcp;
833 vlan_devat_p = vlan_devat;
835 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
836 vlan_clone_create, vlan_clone_destroy);
839 printf("vlan: initialized, using "
843 "hash tables with chaining"
850 if_clone_detach(vlan_cloner);
852 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
853 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
855 vlan_link_state_p = NULL;
856 vlan_trunk_cap_p = NULL;
857 vlan_trunkdev_p = NULL;
859 vlan_cookie_p = NULL;
860 vlan_setcookie_p = NULL;
862 VLAN_LOCKING_DESTROY();
864 printf("vlan: unloaded\n");
872 static moduledata_t vlan_mod = {
878 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
879 MODULE_VERSION(if_vlan, 3);
883 vnet_vlan_init(const void *unused __unused)
886 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
887 vlan_clone_create, vlan_clone_destroy);
888 V_vlan_cloner = vlan_cloner;
890 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
891 vnet_vlan_init, NULL);
894 vnet_vlan_uninit(const void *unused __unused)
897 if_clone_detach(V_vlan_cloner);
899 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
900 vnet_vlan_uninit, NULL);
904 * Check for <etherif>.<vlan> style interface names.
906 static struct ifnet *
907 vlan_clone_match_ethervid(const char *name, int *vidp)
909 char ifname[IFNAMSIZ];
914 strlcpy(ifname, name, IFNAMSIZ);
915 if ((cp = strchr(ifname, '.')) == NULL)
918 if ((ifp = ifunit_ref(ifname)) == NULL)
926 for(; *cp >= '0' && *cp <= '9'; cp++)
927 vid = (vid * 10) + (*cp - '0');
939 vlan_clone_match(struct if_clone *ifc, const char *name)
943 if (vlan_clone_match_ethervid(name, NULL) != NULL)
946 if (strncmp(vlanname, name, strlen(vlanname)) != 0)
948 for (cp = name + 4; *cp != '\0'; cp++) {
949 if (*cp < '0' || *cp > '9')
957 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
968 struct sockaddr_dl *sdl;
970 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
973 * There are 3 (ugh) ways to specify the cloned device:
974 * o pass a parameter block with the clone request.
975 * o specify parameters in the text of the clone device name
976 * o specify no parameters and get an unattached device that
977 * must be configured separately.
978 * The first technique is preferred; the latter two are
979 * supported for backwards compatibility.
981 * XXXRW: Note historic use of the word "tag" here. New ioctls may be
985 error = copyin(params, &vlr, sizeof(vlr));
988 p = ifunit_ref(vlr.vlr_parent);
991 error = ifc_name2unit(name, &unit);
997 wildcard = (unit < 0);
998 } else if ((p = vlan_clone_match_ethervid(name, &vid)) != NULL) {
1003 error = ifc_name2unit(name, &unit);
1007 wildcard = (unit < 0);
1010 error = ifc_alloc_unit(ifc, &unit);
1017 /* In the wildcard case, we need to update the name. */
1019 for (dp = name; *dp != '\0'; dp++);
1020 if (snprintf(dp, len - (dp-name), "%d", unit) >
1021 len - (dp-name) - 1) {
1022 panic("%s: interface name too long", __func__);
1026 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
1027 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
1029 ifc_free_unit(ifc, unit);
1035 CK_SLIST_INIT(&ifv->vlan_mc_listhead);
1036 ifp->if_softc = ifv;
1038 * Set the name manually rather than using if_initname because
1039 * we don't conform to the default naming convention for interfaces.
1041 strlcpy(ifp->if_xname, name, IFNAMSIZ);
1042 ifp->if_dname = vlanname;
1043 ifp->if_dunit = unit;
1045 ifp->if_init = vlan_init;
1046 ifp->if_transmit = vlan_transmit;
1047 ifp->if_qflush = vlan_qflush;
1048 ifp->if_ioctl = vlan_ioctl;
1050 ifp->if_snd_tag_alloc = vlan_snd_tag_alloc;
1051 ifp->if_snd_tag_free = vlan_snd_tag_free;
1053 ifp->if_flags = VLAN_IFFLAGS;
1054 ether_ifattach(ifp, eaddr);
1055 /* Now undo some of the damage... */
1056 ifp->if_baudrate = 0;
1057 ifp->if_type = IFT_L2VLAN;
1058 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
1060 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1061 sdl->sdl_type = IFT_L2VLAN;
1064 error = vlan_config(ifv, p, vid);
1068 * Since we've partially failed, we need to back
1069 * out all the way, otherwise userland could get
1070 * confused. Thus, we destroy the interface.
1072 ether_ifdetach(ifp);
1075 ifc_free_unit(ifc, unit);
1086 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1088 struct ifvlan *ifv = ifp->if_softc;
1089 int unit = ifp->if_dunit;
1091 ether_ifdetach(ifp); /* first, remove it from system-wide lists */
1092 vlan_unconfig(ifp); /* now it can be unconfigured and freed */
1094 * We should have the only reference to the ifv now, so we can now
1095 * drain any remaining lladdr task before freeing the ifnet and the
1098 taskqueue_drain(taskqueue_thread, &ifv->lladdr_task);
1102 ifc_free_unit(ifc, unit);
1108 * The ifp->if_init entry point for vlan(4) is a no-op.
1111 vlan_init(void *foo __unused)
1116 * The if_transmit method for vlan(4) interface.
1119 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1121 struct epoch_tracker et;
1124 int error, len, mcast;
1126 NET_EPOCH_ENTER(et);
1127 ifv = ifp->if_softc;
1128 if (TRUNK(ifv) == NULL) {
1129 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1135 len = m->m_pkthdr.len;
1136 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1141 * Do not run parent's if_transmit() if the parent is not up,
1142 * or parent's driver will cause a system crash.
1144 if (!UP_AND_RUNNING(p)) {
1145 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1151 if (!ether_8021q_frame(&m, ifp, p, ifv->ifv_vid, ifv->ifv_pcp)) {
1152 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1158 * Send it, precisely as ether_output() would have.
1160 error = (p->if_transmit)(p, m);
1162 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1163 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1164 if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1166 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1172 * The ifp->if_qflush entry point for vlan(4) is a no-op.
1175 vlan_qflush(struct ifnet *ifp __unused)
1180 vlan_input(struct ifnet *ifp, struct mbuf *m)
1182 struct epoch_tracker et;
1183 struct ifvlantrunk *trunk;
1188 NET_EPOCH_ENTER(et);
1189 trunk = ifp->if_vlantrunk;
1190 if (trunk == NULL) {
1196 if (m->m_flags & M_VLANTAG) {
1198 * Packet is tagged, but m contains a normal
1199 * Ethernet frame; the tag is stored out-of-band.
1201 tag = m->m_pkthdr.ether_vtag;
1202 m->m_flags &= ~M_VLANTAG;
1204 struct ether_vlan_header *evl;
1207 * Packet is tagged in-band as specified by 802.1q.
1209 switch (ifp->if_type) {
1211 if (m->m_len < sizeof(*evl) &&
1212 (m = m_pullup(m, sizeof(*evl))) == NULL) {
1213 if_printf(ifp, "cannot pullup VLAN header\n");
1217 evl = mtod(m, struct ether_vlan_header *);
1218 tag = ntohs(evl->evl_tag);
1221 * Remove the 802.1q header by copying the Ethernet
1222 * addresses over it and adjusting the beginning of
1223 * the data in the mbuf. The encapsulated Ethernet
1224 * type field is already in place.
1226 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1227 ETHER_HDR_LEN - ETHER_TYPE_LEN);
1228 m_adj(m, ETHER_VLAN_ENCAP_LEN);
1233 panic("%s: %s has unsupported if_type %u",
1234 __func__, ifp->if_xname, ifp->if_type);
1236 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1243 vid = EVL_VLANOFTAG(tag);
1245 ifv = vlan_gethash(trunk, vid);
1246 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1248 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1253 if (vlan_mtag_pcp) {
1255 * While uncommon, it is possible that we will find a 802.1q
1256 * packet encapsulated inside another packet that also had an
1257 * 802.1q header. For example, ethernet tunneled over IPSEC
1258 * arriving over ethernet. In that case, we replace the
1259 * existing 802.1q PCP m_tag value.
1261 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1263 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1264 sizeof(uint8_t), M_NOWAIT);
1266 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1271 m_tag_prepend(m, mtag);
1273 *(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1276 m->m_pkthdr.rcvif = ifv->ifv_ifp;
1277 if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1280 /* Pass it back through the parent's input routine. */
1281 (*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m);
1285 vlan_lladdr_fn(void *arg, int pending __unused)
1290 ifv = (struct ifvlan *)arg;
1293 CURVNET_SET(ifp->if_vnet);
1295 /* The ifv_ifp already has the lladdr copied in. */
1296 if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1302 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid)
1304 struct epoch_tracker et;
1305 struct ifvlantrunk *trunk;
1310 * We can handle non-ethernet hardware types as long as
1311 * they handle the tagging and headers themselves.
1313 if (p->if_type != IFT_ETHER &&
1314 (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1315 return (EPROTONOSUPPORT);
1316 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1317 return (EPROTONOSUPPORT);
1319 * Don't let the caller set up a VLAN VID with
1320 * anything except VLID bits.
1321 * VID numbers 0x0 and 0xFFF are reserved.
1323 if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1329 if (p->if_vlantrunk == NULL) {
1330 trunk = malloc(sizeof(struct ifvlantrunk),
1331 M_VLAN, M_WAITOK | M_ZERO);
1332 vlan_inithash(trunk);
1333 TRUNK_LOCK_INIT(trunk);
1335 p->if_vlantrunk = trunk;
1337 if_ref(trunk->parent);
1338 TRUNK_WUNLOCK(trunk);
1340 trunk = p->if_vlantrunk;
1343 ifv->ifv_vid = vid; /* must set this before vlan_inshash() */
1344 ifv->ifv_pcp = 0; /* Default: best effort delivery. */
1345 vlan_tag_recalculate(ifv);
1346 error = vlan_inshash(trunk, ifv);
1349 ifv->ifv_proto = ETHERTYPE_VLAN;
1350 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1351 ifv->ifv_mintu = ETHERMIN;
1352 ifv->ifv_pflags = 0;
1353 ifv->ifv_capenable = -1;
1356 * If the parent supports the VLAN_MTU capability,
1357 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1360 if (p->if_capenable & IFCAP_VLAN_MTU) {
1362 * No need to fudge the MTU since the parent can
1363 * handle extended frames.
1365 ifv->ifv_mtufudge = 0;
1368 * Fudge the MTU by the encapsulation size. This
1369 * makes us incompatible with strictly compliant
1370 * 802.1Q implementations, but allows us to use
1371 * the feature with other NetBSD implementations,
1372 * which might still be useful.
1374 ifv->ifv_mtufudge = ifv->ifv_encaplen;
1377 ifv->ifv_trunk = trunk;
1380 * Initialize fields from our parent. This duplicates some
1381 * work with ether_ifattach() but allows for non-ethernet
1382 * interfaces to also work.
1384 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1385 ifp->if_baudrate = p->if_baudrate;
1386 ifp->if_output = p->if_output;
1387 ifp->if_input = p->if_input;
1388 ifp->if_resolvemulti = p->if_resolvemulti;
1389 ifp->if_addrlen = p->if_addrlen;
1390 ifp->if_broadcastaddr = p->if_broadcastaddr;
1391 ifp->if_pcp = ifv->ifv_pcp;
1394 * Copy only a selected subset of flags from the parent.
1395 * Other flags are none of our business.
1397 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1398 ifp->if_flags &= ~VLAN_COPY_FLAGS;
1399 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1400 #undef VLAN_COPY_FLAGS
1402 ifp->if_link_state = p->if_link_state;
1404 NET_EPOCH_ENTER(et);
1405 vlan_capabilities(ifv);
1409 * Set up our interface address to reflect the underlying
1410 * physical interface's.
1412 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1413 ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1416 TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv);
1418 /* We are ready for operation now. */
1419 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1421 /* Update flags on the parent, if necessary. */
1422 vlan_setflags(ifp, 1);
1425 * Configure multicast addresses that may already be
1426 * joined on the vlan device.
1428 (void)vlan_setmulti(ifp);
1432 EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1439 vlan_unconfig(struct ifnet *ifp)
1443 vlan_unconfig_locked(ifp, 0);
1448 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1450 struct ifvlantrunk *trunk;
1451 struct vlan_mc_entry *mc;
1453 struct ifnet *parent;
1456 VLAN_XLOCK_ASSERT();
1458 ifv = ifp->if_softc;
1459 trunk = ifv->ifv_trunk;
1462 if (trunk != NULL) {
1463 parent = trunk->parent;
1466 * Since the interface is being unconfigured, we need to
1467 * empty the list of multicast groups that we may have joined
1468 * while we were alive from the parent's list.
1470 while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1472 * If the parent interface is being detached,
1473 * all its multicast addresses have already
1474 * been removed. Warn about errors if
1475 * if_delmulti() does fail, but don't abort as
1476 * all callers expect vlan destruction to
1480 error = if_delmulti(parent,
1481 (struct sockaddr *)&mc->mc_addr);
1484 "Failed to delete multicast address from parent: %d\n",
1487 CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1488 epoch_call(net_epoch_preempt, &mc->mc_epoch_ctx, vlan_mc_free);
1491 vlan_setflags(ifp, 0); /* clear special flags on parent */
1493 vlan_remhash(trunk, ifv);
1494 ifv->ifv_trunk = NULL;
1497 * Check if we were the last.
1499 if (trunk->refcnt == 0) {
1500 parent->if_vlantrunk = NULL;
1502 trunk_destroy(trunk);
1506 /* Disconnect from parent. */
1507 if (ifv->ifv_pflags)
1508 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1509 ifp->if_mtu = ETHERMTU;
1510 ifp->if_link_state = LINK_STATE_UNKNOWN;
1511 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1514 * Only dispatch an event if vlan was
1515 * attached, otherwise there is nothing
1516 * to cleanup anyway.
1519 EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1522 /* Handle a reference counted flag that should be set on the parent as well */
1524 vlan_setflag(struct ifnet *ifp, int flag, int status,
1525 int (*func)(struct ifnet *, int))
1530 VLAN_SXLOCK_ASSERT();
1532 ifv = ifp->if_softc;
1533 status = status ? (ifp->if_flags & flag) : 0;
1534 /* Now "status" contains the flag value or 0 */
1537 * See if recorded parent's status is different from what
1538 * we want it to be. If it is, flip it. We record parent's
1539 * status in ifv_pflags so that we won't clear parent's flag
1540 * we haven't set. In fact, we don't clear or set parent's
1541 * flags directly, but get or release references to them.
1542 * That's why we can be sure that recorded flags still are
1543 * in accord with actual parent's flags.
1545 if (status != (ifv->ifv_pflags & flag)) {
1546 error = (*func)(PARENT(ifv), status);
1549 ifv->ifv_pflags &= ~flag;
1550 ifv->ifv_pflags |= status;
1556 * Handle IFF_* flags that require certain changes on the parent:
1557 * if "status" is true, update parent's flags respective to our if_flags;
1558 * if "status" is false, forcedly clear the flags set on parent.
1561 vlan_setflags(struct ifnet *ifp, int status)
1565 for (i = 0; vlan_pflags[i].flag; i++) {
1566 error = vlan_setflag(ifp, vlan_pflags[i].flag,
1567 status, vlan_pflags[i].func);
1574 /* Inform all vlans that their parent has changed link state */
1576 vlan_link_state(struct ifnet *ifp)
1578 struct epoch_tracker et;
1579 struct ifvlantrunk *trunk;
1582 /* Called from a taskqueue_swi task, so we cannot sleep. */
1583 NET_EPOCH_ENTER(et);
1584 trunk = ifp->if_vlantrunk;
1585 if (trunk == NULL) {
1591 VLAN_FOREACH(ifv, trunk) {
1592 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1593 if_link_state_change(ifv->ifv_ifp,
1594 trunk->parent->if_link_state);
1596 TRUNK_WUNLOCK(trunk);
1601 vlan_capabilities(struct ifvlan *ifv)
1605 struct ifnet_hw_tsomax hw_tsomax;
1606 int cap = 0, ena = 0, mena;
1609 VLAN_SXLOCK_ASSERT();
1614 /* Mask parent interface enabled capabilities disabled by user. */
1615 mena = p->if_capenable & ifv->ifv_capenable;
1618 * If the parent interface can do checksum offloading
1619 * on VLANs, then propagate its hardware-assisted
1620 * checksumming flags. Also assert that checksum
1621 * offloading requires hardware VLAN tagging.
1623 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1624 cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1625 if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1626 p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1627 ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1628 if (ena & IFCAP_TXCSUM)
1629 hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
1630 CSUM_UDP | CSUM_SCTP);
1631 if (ena & IFCAP_TXCSUM_IPV6)
1632 hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
1633 CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
1637 * If the parent interface can do TSO on VLANs then
1638 * propagate the hardware-assisted flag. TSO on VLANs
1639 * does not necessarily require hardware VLAN tagging.
1641 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1642 if_hw_tsomax_common(p, &hw_tsomax);
1643 if_hw_tsomax_update(ifp, &hw_tsomax);
1644 if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1645 cap |= p->if_capabilities & IFCAP_TSO;
1646 if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1647 ena |= mena & IFCAP_TSO;
1648 if (ena & IFCAP_TSO)
1649 hwa |= p->if_hwassist & CSUM_TSO;
1653 * If the parent interface can do LRO and checksum offloading on
1654 * VLANs, then guess it may do LRO on VLANs. False positive here
1655 * cost nothing, while false negative may lead to some confusions.
1657 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1658 cap |= p->if_capabilities & IFCAP_LRO;
1659 if (p->if_capenable & IFCAP_VLAN_HWCSUM)
1660 ena |= p->if_capenable & IFCAP_LRO;
1663 * If the parent interface can offload TCP connections over VLANs then
1664 * propagate its TOE capability to the VLAN interface.
1666 * All TOE drivers in the tree today can deal with VLANs. If this
1667 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1670 #define IFCAP_VLAN_TOE IFCAP_TOE
1671 if (p->if_capabilities & IFCAP_VLAN_TOE)
1672 cap |= p->if_capabilities & IFCAP_TOE;
1673 if (p->if_capenable & IFCAP_VLAN_TOE) {
1674 TOEDEV(ifp) = TOEDEV(p);
1675 ena |= mena & IFCAP_TOE;
1679 * If the parent interface supports dynamic link state, so does the
1682 cap |= (p->if_capabilities & IFCAP_LINKSTATE);
1683 ena |= (mena & IFCAP_LINKSTATE);
1687 * If the parent interface supports ratelimiting, so does the
1690 cap |= (p->if_capabilities & IFCAP_TXRTLMT);
1691 ena |= (mena & IFCAP_TXRTLMT);
1694 ifp->if_capabilities = cap;
1695 ifp->if_capenable = ena;
1696 ifp->if_hwassist = hwa;
1700 vlan_trunk_capabilities(struct ifnet *ifp)
1702 struct epoch_tracker et;
1703 struct ifvlantrunk *trunk;
1707 trunk = ifp->if_vlantrunk;
1708 if (trunk == NULL) {
1712 NET_EPOCH_ENTER(et);
1713 VLAN_FOREACH(ifv, trunk) {
1714 vlan_capabilities(ifv);
1721 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1727 struct ifvlantrunk *trunk;
1731 ifr = (struct ifreq *)data;
1732 ifa = (struct ifaddr *) data;
1733 ifv = ifp->if_softc;
1737 ifp->if_flags |= IFF_UP;
1739 if (ifa->ifa_addr->sa_family == AF_INET)
1740 arp_ifinit(ifp, ifa);
1744 bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0],
1749 if (TRUNK(ifv) != NULL) {
1752 error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1754 /* Limit the result to the parent's current config. */
1756 struct ifmediareq *ifmr;
1758 ifmr = (struct ifmediareq *)data;
1759 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1760 ifmr->ifm_count = 1;
1761 error = copyout(&ifmr->ifm_current,
1778 * Set the interface MTU.
1782 if (trunk != NULL) {
1785 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1787 (ifv->ifv_mintu - ifv->ifv_mtufudge))
1790 ifp->if_mtu = ifr->ifr_mtu;
1791 TRUNK_WUNLOCK(trunk);
1800 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1801 * interface to be delegated to a jail without allowing the
1802 * jail to change what underlying interface/VID it is
1803 * associated with. We are not entirely convinced that this
1804 * is the right way to accomplish that policy goal.
1806 if (ifp->if_vnet != ifp->if_home_vnet) {
1811 error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr));
1814 if (vlr.vlr_parent[0] == '\0') {
1818 p = ifunit_ref(vlr.vlr_parent);
1823 error = vlan_config(ifv, p, vlr.vlr_tag);
1829 if (ifp->if_vnet != ifp->if_home_vnet) {
1834 bzero(&vlr, sizeof(vlr));
1836 if (TRUNK(ifv) != NULL) {
1837 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1838 sizeof(vlr.vlr_parent));
1839 vlr.vlr_tag = ifv->ifv_vid;
1842 error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr));
1847 * We should propagate selected flags to the parent,
1848 * e.g., promiscuous mode.
1851 if (TRUNK(ifv) != NULL)
1852 error = vlan_setflags(ifp, 1);
1859 * If we don't have a parent, just remember the membership for
1862 * XXX We need the rmlock here to avoid sleeping while
1863 * holding in6_multi_mtx.
1868 error = vlan_setmulti(ifp);
1874 if (ifp->if_vnet != ifp->if_home_vnet) {
1879 ifr->ifr_vlan_pcp = ifv->ifv_pcp;
1884 if (ifp->if_vnet != ifp->if_home_vnet) {
1889 error = priv_check(curthread, PRIV_NET_SETVLANPCP);
1892 if (ifr->ifr_vlan_pcp > 7) {
1896 ifv->ifv_pcp = ifr->ifr_vlan_pcp;
1897 ifp->if_pcp = ifv->ifv_pcp;
1898 vlan_tag_recalculate(ifv);
1899 /* broadcast event about PCP change */
1900 EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP);
1905 ifv->ifv_capenable = ifr->ifr_reqcap;
1907 if (trunk != NULL) {
1908 struct epoch_tracker et;
1910 NET_EPOCH_ENTER(et);
1911 vlan_capabilities(ifv);
1927 vlan_snd_tag_alloc(struct ifnet *ifp,
1928 union if_snd_tag_alloc_params *params,
1929 struct m_snd_tag **ppmt)
1932 /* get trunk device */
1933 ifp = vlan_trunkdev(ifp);
1934 if (ifp == NULL || (ifp->if_capenable & IFCAP_TXRTLMT) == 0)
1935 return (EOPNOTSUPP);
1936 /* forward allocation request */
1937 return (ifp->if_snd_tag_alloc(ifp, params, ppmt));
1941 vlan_snd_tag_free(struct m_snd_tag *tag)
1943 tag->ifp->if_snd_tag_free(tag);