2 * Copyright 1998 Massachusetts Institute of Technology
4 * Permission to use, copy, modify, and distribute this software and
5 * its documentation for any purpose and without fee is hereby
6 * granted, provided that both the above copyright notice and this
7 * permission notice appear in all copies, that both the above
8 * copyright notice and this permission notice appear in all
9 * supporting documentation, and that the name of M.I.T. not be used
10 * in advertising or publicity pertaining to distribution of the
11 * software without specific, written prior permission. M.I.T. makes
12 * no representations about the suitability of this software for any
13 * purpose. It is provided "as is" without express or implied
16 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS
17 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
18 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
19 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
20 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
23 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
24 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
25 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
26 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
32 * Might be extended some day to also handle IEEE 802.1p priority
33 * tagging. This is sort of sneaky in the implementation, since
34 * we need to pretend to be enough of an Ethernet implementation
35 * to make arp work. The way we do this is by telling everyone
36 * that we are an Ethernet, and then catch the packets that
37 * ether_output() sends to us via if_transmit(), rewrite them for
38 * use by the real outgoing interface, and ask it to send them.
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
46 #include <sys/param.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
51 #include <sys/module.h>
52 #include <sys/rwlock.h>
53 #include <sys/queue.h>
54 #include <sys/socket.h>
55 #include <sys/sockio.h>
56 #include <sys/sysctl.h>
57 #include <sys/systm.h>
61 #include <net/ethernet.h>
63 #include <net/if_clone.h>
64 #include <net/if_dl.h>
65 #include <net/if_types.h>
66 #include <net/if_vlan_var.h>
69 #define VLANNAME "vlan"
70 #define VLAN_DEF_HWIDTH 4
71 #define VLAN_IFFLAGS (IFF_BROADCAST | IFF_MULTICAST)
73 #define UP_AND_RUNNING(ifp) \
74 ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
76 LIST_HEAD(ifvlanhead, ifvlan);
79 struct ifnet *parent; /* parent interface of this trunk */
82 #define VLAN_ARRAY_SIZE (EVL_VLID_MASK + 1)
83 struct ifvlan *vlans[VLAN_ARRAY_SIZE]; /* static table */
85 struct ifvlanhead *hash; /* dynamic hash-list table */
92 struct vlan_mc_entry {
93 struct sockaddr_dl mc_addr;
94 SLIST_ENTRY(vlan_mc_entry) mc_entries;
98 struct ifvlantrunk *ifv_trunk;
99 struct ifnet *ifv_ifp;
101 #define TRUNK(ifv) ((ifv)->ifv_trunk)
102 #define PARENT(ifv) ((ifv)->ifv_trunk->parent)
103 int ifv_pflags; /* special flags we have set on parent */
105 int ifvm_encaplen; /* encapsulation length */
106 int ifvm_mtufudge; /* MTU fudged by this much */
107 int ifvm_mintu; /* min transmission unit */
108 uint16_t ifvm_proto; /* encapsulation ethertype */
109 uint16_t ifvm_tag; /* tag to apply on packets leaving if */
111 SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
113 LIST_ENTRY(ifvlan) ifv_list;
116 #define ifv_proto ifv_mib.ifvm_proto
117 #define ifv_tag ifv_mib.ifvm_tag
118 #define ifv_encaplen ifv_mib.ifvm_encaplen
119 #define ifv_mtufudge ifv_mib.ifvm_mtufudge
120 #define ifv_mintu ifv_mib.ifvm_mintu
122 /* Special flags we should propagate to parent. */
125 int (*func)(struct ifnet *, int);
127 {IFF_PROMISC, ifpromisc},
128 {IFF_ALLMULTI, if_allmulti},
132 SYSCTL_DECL(_net_link);
133 static SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0,
135 static SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0,
138 static int soft_pad = 0;
139 SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW, &soft_pad, 0,
140 "pad short frames before tagging");
142 static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface");
144 static eventhandler_tag ifdetach_tag;
145 static eventhandler_tag iflladdr_tag;
148 * We have a global mutex, that is used to serialize configuration
149 * changes and isn't used in normal packet delivery.
151 * We also have a per-trunk rwlock, that is locked shared on packet
152 * processing and exclusive when configuration is changed.
154 * The VLAN_ARRAY substitutes the dynamic hash with a static array
155 * with 4096 entries. In theory this can give a boost in processing,
156 * however on practice it does not. Probably this is because array
157 * is too big to fit into CPU cache.
159 static struct sx ifv_lock;
160 #define VLAN_LOCK_INIT() sx_init(&ifv_lock, "vlan_global")
161 #define VLAN_LOCK_DESTROY() sx_destroy(&ifv_lock)
162 #define VLAN_LOCK_ASSERT() sx_assert(&ifv_lock, SA_LOCKED)
163 #define VLAN_LOCK() sx_xlock(&ifv_lock)
164 #define VLAN_UNLOCK() sx_xunlock(&ifv_lock)
165 #define TRUNK_LOCK_INIT(trunk) rw_init(&(trunk)->rw, VLANNAME)
166 #define TRUNK_LOCK_DESTROY(trunk) rw_destroy(&(trunk)->rw)
167 #define TRUNK_LOCK(trunk) rw_wlock(&(trunk)->rw)
168 #define TRUNK_UNLOCK(trunk) rw_wunlock(&(trunk)->rw)
169 #define TRUNK_LOCK_ASSERT(trunk) rw_assert(&(trunk)->rw, RA_WLOCKED)
170 #define TRUNK_RLOCK(trunk) rw_rlock(&(trunk)->rw)
171 #define TRUNK_RUNLOCK(trunk) rw_runlock(&(trunk)->rw)
172 #define TRUNK_LOCK_RASSERT(trunk) rw_assert(&(trunk)->rw, RA_RLOCKED)
175 static void vlan_inithash(struct ifvlantrunk *trunk);
176 static void vlan_freehash(struct ifvlantrunk *trunk);
177 static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
178 static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
179 static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
180 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
183 static void trunk_destroy(struct ifvlantrunk *trunk);
185 static void vlan_init(void *foo);
186 static void vlan_input(struct ifnet *ifp, struct mbuf *m);
187 static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
188 static void vlan_qflush(struct ifnet *ifp);
189 static int vlan_setflag(struct ifnet *ifp, int flag, int status,
190 int (*func)(struct ifnet *, int));
191 static int vlan_setflags(struct ifnet *ifp, int status);
192 static int vlan_setmulti(struct ifnet *ifp);
193 static int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
194 static void vlan_unconfig(struct ifnet *ifp);
195 static void vlan_unconfig_locked(struct ifnet *ifp, int departing);
196 static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
197 static void vlan_link_state(struct ifnet *ifp);
198 static void vlan_capabilities(struct ifvlan *ifv);
199 static void vlan_trunk_capabilities(struct ifnet *ifp);
201 static struct ifnet *vlan_clone_match_ethertag(struct if_clone *,
202 const char *, int *);
203 static int vlan_clone_match(struct if_clone *, const char *);
204 static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
205 static int vlan_clone_destroy(struct if_clone *, struct ifnet *);
207 static void vlan_ifdetach(void *arg, struct ifnet *ifp);
208 static void vlan_iflladdr(void *arg, struct ifnet *ifp);
210 static struct if_clone vlan_cloner = IFC_CLONE_INITIALIZER(VLANNAME, NULL,
211 IF_MAXUNIT, NULL, vlan_clone_match, vlan_clone_create, vlan_clone_destroy);
214 static VNET_DEFINE(struct if_clone, vlan_cloner);
215 #define V_vlan_cloner VNET(vlan_cloner)
219 #define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
222 vlan_inithash(struct ifvlantrunk *trunk)
227 * The trunk must not be locked here since we call malloc(M_WAITOK).
228 * It is OK in case this function is called before the trunk struct
229 * gets hooked up and becomes visible from other threads.
232 KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
233 ("%s: hash already initialized", __func__));
235 trunk->hwidth = VLAN_DEF_HWIDTH;
236 n = 1 << trunk->hwidth;
237 trunk->hmask = n - 1;
238 trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
239 for (i = 0; i < n; i++)
240 LIST_INIT(&trunk->hash[i]);
244 vlan_freehash(struct ifvlantrunk *trunk)
249 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
250 for (i = 0; i < (1 << trunk->hwidth); i++)
251 KASSERT(LIST_EMPTY(&trunk->hash[i]),
252 ("%s: hash table not empty", __func__));
254 free(trunk->hash, M_VLAN);
256 trunk->hwidth = trunk->hmask = 0;
260 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
265 TRUNK_LOCK_ASSERT(trunk);
266 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
268 b = 1 << trunk->hwidth;
269 i = HASH(ifv->ifv_tag, trunk->hmask);
270 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
271 if (ifv->ifv_tag == ifv2->ifv_tag)
275 * Grow the hash when the number of vlans exceeds half of the number of
276 * hash buckets squared. This will make the average linked-list length
279 if (trunk->refcnt > (b * b) / 2) {
280 vlan_growhash(trunk, 1);
281 i = HASH(ifv->ifv_tag, trunk->hmask);
283 LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
290 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
295 TRUNK_LOCK_ASSERT(trunk);
296 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
298 b = 1 << trunk->hwidth;
299 i = HASH(ifv->ifv_tag, trunk->hmask);
300 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
303 LIST_REMOVE(ifv2, ifv_list);
304 if (trunk->refcnt < (b * b) / 2)
305 vlan_growhash(trunk, -1);
309 panic("%s: vlan not found\n", __func__);
310 return (ENOENT); /*NOTREACHED*/
314 * Grow the hash larger or smaller if memory permits.
317 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
320 struct ifvlanhead *hash2;
321 int hwidth2, i, j, n, n2;
323 TRUNK_LOCK_ASSERT(trunk);
324 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
327 /* Harmless yet obvious coding error */
328 printf("%s: howmuch is 0\n", __func__);
332 hwidth2 = trunk->hwidth + howmuch;
333 n = 1 << trunk->hwidth;
335 /* Do not shrink the table below the default */
336 if (hwidth2 < VLAN_DEF_HWIDTH)
339 /* M_NOWAIT because we're called with trunk mutex held */
340 hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
342 printf("%s: out of memory -- hash size not changed\n",
344 return; /* We can live with the old hash table */
346 for (j = 0; j < n2; j++)
347 LIST_INIT(&hash2[j]);
348 for (i = 0; i < n; i++)
349 while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
350 LIST_REMOVE(ifv, ifv_list);
351 j = HASH(ifv->ifv_tag, n2 - 1);
352 LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
354 free(trunk->hash, M_VLAN);
356 trunk->hwidth = hwidth2;
357 trunk->hmask = n2 - 1;
360 if_printf(trunk->parent,
361 "VLAN hash table resized from %d to %d buckets\n", n, n2);
364 static __inline struct ifvlan *
365 vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
369 TRUNK_LOCK_RASSERT(trunk);
371 LIST_FOREACH(ifv, &trunk->hash[HASH(tag, trunk->hmask)], ifv_list)
372 if (ifv->ifv_tag == tag)
378 /* Debugging code to view the hashtables. */
380 vlan_dumphash(struct ifvlantrunk *trunk)
385 for (i = 0; i < (1 << trunk->hwidth); i++) {
387 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
388 printf("%s ", ifv->ifv_ifp->if_xname);
395 static __inline struct ifvlan *
396 vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
399 return trunk->vlans[tag];
403 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
406 if (trunk->vlans[ifv->ifv_tag] != NULL)
408 trunk->vlans[ifv->ifv_tag] = ifv;
415 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
418 trunk->vlans[ifv->ifv_tag] = NULL;
425 vlan_freehash(struct ifvlantrunk *trunk)
430 vlan_inithash(struct ifvlantrunk *trunk)
434 #endif /* !VLAN_ARRAY */
437 trunk_destroy(struct ifvlantrunk *trunk)
442 vlan_freehash(trunk);
443 trunk->parent->if_vlantrunk = NULL;
445 TRUNK_LOCK_DESTROY(trunk);
450 * Program our multicast filter. What we're actually doing is
451 * programming the multicast filter of the parent. This has the
452 * side effect of causing the parent interface to receive multicast
453 * traffic that it doesn't really want, which ends up being discarded
454 * later by the upper protocol layers. Unfortunately, there's no way
455 * to avoid this: there really is only one physical interface.
457 * XXX: There is a possible race here if more than one thread is
458 * modifying the multicast state of the vlan interface at the same time.
461 vlan_setmulti(struct ifnet *ifp)
464 struct ifmultiaddr *ifma, *rifma = NULL;
466 struct vlan_mc_entry *mc;
469 /*VLAN_LOCK_ASSERT();*/
471 /* Find the parent. */
475 CURVNET_SET_QUIET(ifp_p->if_vnet);
477 /* First, remove any existing filter entries. */
478 while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
479 error = if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
482 SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
486 /* Now program new ones. */
487 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
488 if (ifma->ifma_addr->sa_family != AF_LINK)
490 mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
493 bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
494 mc->mc_addr.sdl_index = ifp_p->if_index;
495 SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
496 error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
507 * A handler for parent interface link layer address changes.
508 * If the parent interface link layer address is changed we
509 * should also change it on all children vlans.
512 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
521 * Check if it's a trunk interface first of all
522 * to avoid needless locking.
524 if (ifp->if_vlantrunk == NULL)
529 * OK, it's a trunk. Loop over and change all vlan's lladdrs on it.
532 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
533 if ((ifv = ifp->if_vlantrunk->vlans[i])) {
534 #else /* VLAN_ARRAY */
535 for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
536 LIST_FOREACH_SAFE(ifv, &ifp->if_vlantrunk->hash[i], ifv_list, next) {
537 #endif /* VLAN_ARRAY */
539 if_setlladdr(ifv->ifv_ifp, IF_LLADDR(ifp),
548 * A handler for network interface departure events.
549 * Track departure of trunks here so that we don't access invalid
550 * pointers or whatever if a trunk is ripped from under us, e.g.,
551 * by ejecting its hot-plug card. However, if an ifnet is simply
552 * being renamed, then there's no need to tear down the state.
555 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
561 * Check if it's a trunk interface first of all
562 * to avoid needless locking.
564 if (ifp->if_vlantrunk == NULL)
567 /* If the ifnet is just being renamed, don't do anything. */
568 if (ifp->if_flags & IFF_RENAMING)
573 * OK, it's a trunk. Loop over and detach all vlan's on it.
574 * Check trunk pointer after each vlan_unconfig() as it will
575 * free it and set to NULL after the last vlan was detached.
578 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
579 if ((ifv = ifp->if_vlantrunk->vlans[i])) {
580 vlan_unconfig_locked(ifv->ifv_ifp, 1);
581 if (ifp->if_vlantrunk == NULL)
584 #else /* VLAN_ARRAY */
586 for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
587 if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
588 vlan_unconfig_locked(ifv->ifv_ifp, 1);
589 if (ifp->if_vlantrunk)
590 goto restart; /* trunk->hwidth can change */
594 #endif /* VLAN_ARRAY */
595 /* Trunk should have been destroyed in vlan_unconfig(). */
596 KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
601 * Return the trunk device for a virtual interface.
603 static struct ifnet *
604 vlan_trunkdev(struct ifnet *ifp)
608 if (ifp->if_type != IFT_L2VLAN)
620 * Return the 16bit vlan tag for this interface.
623 vlan_tag(struct ifnet *ifp, uint16_t *tagp)
627 if (ifp->if_type != IFT_L2VLAN)
630 *tagp = ifv->ifv_tag;
635 * Return a driver specific cookie for this interface. Synchronization
636 * with setcookie must be provided by the driver.
639 vlan_cookie(struct ifnet *ifp)
643 if (ifp->if_type != IFT_L2VLAN)
646 return (ifv->ifv_cookie);
650 * Store a cookie in our softc that drivers can use to store driver
651 * private per-instance data in.
654 vlan_setcookie(struct ifnet *ifp, void *cookie)
658 if (ifp->if_type != IFT_L2VLAN)
661 ifv->ifv_cookie = cookie;
666 * Return the vlan device present at the specific tag.
668 static struct ifnet *
669 vlan_devat(struct ifnet *ifp, uint16_t tag)
671 struct ifvlantrunk *trunk;
674 trunk = ifp->if_vlantrunk;
679 ifv = vlan_gethash(trunk, tag);
682 TRUNK_RUNLOCK(trunk);
687 * VLAN support can be loaded as a module. The only place in the
688 * system that's intimately aware of this is ether_input. We hook
689 * into this code through vlan_input_p which is defined there and
690 * set here. Noone else in the system should be aware of this so
691 * we use an explicit reference here.
693 extern void (*vlan_input_p)(struct ifnet *, struct mbuf *);
695 /* For if_link_state_change() eyes only... */
696 extern void (*vlan_link_state_p)(struct ifnet *);
699 vlan_modevent(module_t mod, int type, void *data)
704 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
705 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
706 if (ifdetach_tag == NULL)
708 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
709 vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
710 if (iflladdr_tag == NULL)
713 vlan_input_p = vlan_input;
714 vlan_link_state_p = vlan_link_state;
715 vlan_trunk_cap_p = vlan_trunk_capabilities;
716 vlan_trunkdev_p = vlan_trunkdev;
717 vlan_cookie_p = vlan_cookie;
718 vlan_setcookie_p = vlan_setcookie;
719 vlan_tag_p = vlan_tag;
720 vlan_devat_p = vlan_devat;
722 if_clone_attach(&vlan_cloner);
725 printf("vlan: initialized, using "
729 "hash tables with chaining"
736 if_clone_detach(&vlan_cloner);
738 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
739 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
741 vlan_link_state_p = NULL;
742 vlan_trunk_cap_p = NULL;
743 vlan_trunkdev_p = NULL;
745 vlan_cookie_p = vlan_cookie;
746 vlan_setcookie_p = vlan_setcookie;
750 printf("vlan: unloaded\n");
758 static moduledata_t vlan_mod = {
764 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
765 MODULE_VERSION(if_vlan, 3);
769 vnet_vlan_init(const void *unused __unused)
772 V_vlan_cloner = vlan_cloner;
773 if_clone_attach(&V_vlan_cloner);
775 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
776 vnet_vlan_init, NULL);
779 vnet_vlan_uninit(const void *unused __unused)
782 if_clone_detach(&V_vlan_cloner);
784 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
785 vnet_vlan_uninit, NULL);
788 static struct ifnet *
789 vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
795 /* Check for <etherif>.<vlan> style interface names. */
796 IFNET_RLOCK_NOSLEEP();
797 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
799 * We can handle non-ethernet hardware types as long as
800 * they handle the tagging and headers themselves.
802 if (ifp->if_type != IFT_ETHER &&
803 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
805 if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
807 cp = name + strlen(ifp->if_xname);
813 for(; *cp >= '0' && *cp <= '9'; cp++)
814 t = (t * 10) + (*cp - '0');
821 IFNET_RUNLOCK_NOSLEEP();
827 vlan_clone_match(struct if_clone *ifc, const char *name)
831 if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
834 if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
836 for (cp = name + 4; *cp != '\0'; cp++) {
837 if (*cp < '0' || *cp > '9')
845 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
857 struct sockaddr_dl *sdl;
859 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
862 * There are 3 (ugh) ways to specify the cloned device:
863 * o pass a parameter block with the clone request.
864 * o specify parameters in the text of the clone device name
865 * o specify no parameters and get an unattached device that
866 * must be configured separately.
867 * The first technique is preferred; the latter two are
868 * supported for backwards compatibilty.
871 error = copyin(params, &vlr, sizeof(vlr));
874 p = ifunit(vlr.vlr_parent);
878 * Don't let the caller set up a VLAN tag with
879 * anything except VLID bits.
881 if (vlr.vlr_tag & ~EVL_VLID_MASK)
883 error = ifc_name2unit(name, &unit);
889 wildcard = (unit < 0);
890 } else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
896 * Don't let the caller set up a VLAN tag with
897 * anything except VLID bits.
899 if (tag & ~EVL_VLID_MASK)
904 error = ifc_name2unit(name, &unit);
908 wildcard = (unit < 0);
911 error = ifc_alloc_unit(ifc, &unit);
915 /* In the wildcard case, we need to update the name. */
917 for (dp = name; *dp != '\0'; dp++);
918 if (snprintf(dp, len - (dp-name), "%d", unit) >
919 len - (dp-name) - 1) {
920 panic("%s: interface name too long", __func__);
924 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
925 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
927 ifc_free_unit(ifc, unit);
931 SLIST_INIT(&ifv->vlan_mc_listhead);
935 * Set the name manually rather than using if_initname because
936 * we don't conform to the default naming convention for interfaces.
938 strlcpy(ifp->if_xname, name, IFNAMSIZ);
939 ifp->if_dname = ifc->ifc_name;
940 ifp->if_dunit = unit;
941 /* NB: flags are not set here */
942 ifp->if_linkmib = &ifv->ifv_mib;
943 ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
944 /* NB: mtu is not set here */
946 ifp->if_init = vlan_init;
947 ifp->if_transmit = vlan_transmit;
948 ifp->if_qflush = vlan_qflush;
949 ifp->if_ioctl = vlan_ioctl;
950 ifp->if_flags = VLAN_IFFLAGS;
951 ether_ifattach(ifp, eaddr);
952 /* Now undo some of the damage... */
953 ifp->if_baudrate = 0;
954 ifp->if_type = IFT_L2VLAN;
955 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
957 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
958 sdl->sdl_type = IFT_L2VLAN;
961 error = vlan_config(ifv, p, tag);
964 * Since we've partially failed, we need to back
965 * out all the way, otherwise userland could get
966 * confused. Thus, we destroy the interface.
970 if_free_type(ifp, IFT_ETHER);
971 ifc_free_unit(ifc, unit);
977 /* Update flags on the parent, if necessary. */
978 vlan_setflags(ifp, 1);
985 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
987 struct ifvlan *ifv = ifp->if_softc;
988 int unit = ifp->if_dunit;
990 ether_ifdetach(ifp); /* first, remove it from system-wide lists */
991 vlan_unconfig(ifp); /* now it can be unconfigured and freed */
992 if_free_type(ifp, IFT_ETHER);
994 ifc_free_unit(ifc, unit);
1000 * The ifp->if_init entry point for vlan(4) is a no-op.
1003 vlan_init(void *foo __unused)
1008 * The if_transmit method for vlan(4) interface.
1011 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1015 int error, len, mcast;
1017 ifv = ifp->if_softc;
1019 len = m->m_pkthdr.len;
1020 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1025 * Do not run parent's if_transmit() if the parent is not up,
1026 * or parent's driver will cause a system crash.
1028 if (!UP_AND_RUNNING(p)) {
1035 * Pad the frame to the minimum size allowed if told to.
1036 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
1037 * paragraph C.4.4.3.b. It can help to work around buggy
1038 * bridges that violate paragraph C.4.4.3.a from the same
1039 * document, i.e., fail to pad short frames after untagging.
1040 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
1041 * untagging it will produce a 62-byte frame, which is a runt
1042 * and requires padding. There are VLAN-enabled network
1043 * devices that just discard such runts instead or mishandle
1046 if (soft_pad && p->if_type == IFT_ETHER) {
1047 static char pad[8]; /* just zeros */
1050 for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
1051 n > 0; n -= sizeof(pad))
1052 if (!m_append(m, min(n, sizeof(pad)), pad))
1056 if_printf(ifp, "cannot pad short frame\n");
1064 * If underlying interface can do VLAN tag insertion itself,
1065 * just pass the packet along. However, we need some way to
1066 * tell the interface where the packet came from so that it
1067 * knows how to find the VLAN tag to use, so we attach a
1068 * packet tag that holds it.
1070 if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1071 m->m_pkthdr.ether_vtag = ifv->ifv_tag;
1072 m->m_flags |= M_VLANTAG;
1074 m = ether_vlanencap(m, ifv->ifv_tag);
1076 if_printf(ifp, "unable to prepend VLAN header\n");
1083 * Send it, precisely as ether_output() would have.
1085 error = (p->if_transmit)(p, m);
1088 ifp->if_omcasts += mcast;
1089 ifp->if_obytes += len;
1096 * The ifp->if_qflush entry point for vlan(4) is a no-op.
1099 vlan_qflush(struct ifnet *ifp __unused)
1104 vlan_input(struct ifnet *ifp, struct mbuf *m)
1106 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1110 KASSERT(trunk != NULL, ("%s: no trunk", __func__));
1112 if (m->m_flags & M_VLANTAG) {
1114 * Packet is tagged, but m contains a normal
1115 * Ethernet frame; the tag is stored out-of-band.
1117 tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
1118 m->m_flags &= ~M_VLANTAG;
1120 struct ether_vlan_header *evl;
1123 * Packet is tagged in-band as specified by 802.1q.
1125 switch (ifp->if_type) {
1127 if (m->m_len < sizeof(*evl) &&
1128 (m = m_pullup(m, sizeof(*evl))) == NULL) {
1129 if_printf(ifp, "cannot pullup VLAN header\n");
1132 evl = mtod(m, struct ether_vlan_header *);
1133 tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1136 * Remove the 802.1q header by copying the Ethernet
1137 * addresses over it and adjusting the beginning of
1138 * the data in the mbuf. The encapsulated Ethernet
1139 * type field is already in place.
1141 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1142 ETHER_HDR_LEN - ETHER_TYPE_LEN);
1143 m_adj(m, ETHER_VLAN_ENCAP_LEN);
1148 panic("%s: %s has unsupported if_type %u",
1149 __func__, ifp->if_xname, ifp->if_type);
1158 ifv = vlan_gethash(trunk, tag);
1159 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1160 TRUNK_RUNLOCK(trunk);
1165 TRUNK_RUNLOCK(trunk);
1167 m->m_pkthdr.rcvif = ifv->ifv_ifp;
1168 ifv->ifv_ifp->if_ipackets++;
1170 /* Pass it back through the parent's input routine. */
1171 (*ifp->if_input)(ifv->ifv_ifp, m);
1175 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
1177 struct ifvlantrunk *trunk;
1181 /* VID numbers 0x0 and 0xFFF are reserved */
1182 if (tag == 0 || tag == 0xFFF)
1184 if (p->if_type != IFT_ETHER &&
1185 (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1186 return (EPROTONOSUPPORT);
1187 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1188 return (EPROTONOSUPPORT);
1192 if (p->if_vlantrunk == NULL) {
1193 trunk = malloc(sizeof(struct ifvlantrunk),
1194 M_VLAN, M_WAITOK | M_ZERO);
1195 vlan_inithash(trunk);
1197 if (p->if_vlantrunk != NULL) {
1198 /* A race that that is very unlikely to be hit. */
1199 vlan_freehash(trunk);
1200 free(trunk, M_VLAN);
1203 TRUNK_LOCK_INIT(trunk);
1205 p->if_vlantrunk = trunk;
1210 trunk = p->if_vlantrunk;
1214 ifv->ifv_tag = tag; /* must set this before vlan_inshash() */
1215 error = vlan_inshash(trunk, ifv);
1218 ifv->ifv_proto = ETHERTYPE_VLAN;
1219 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1220 ifv->ifv_mintu = ETHERMIN;
1221 ifv->ifv_pflags = 0;
1224 * If the parent supports the VLAN_MTU capability,
1225 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1228 if (p->if_capenable & IFCAP_VLAN_MTU) {
1230 * No need to fudge the MTU since the parent can
1231 * handle extended frames.
1233 ifv->ifv_mtufudge = 0;
1236 * Fudge the MTU by the encapsulation size. This
1237 * makes us incompatible with strictly compliant
1238 * 802.1Q implementations, but allows us to use
1239 * the feature with other NetBSD implementations,
1240 * which might still be useful.
1242 ifv->ifv_mtufudge = ifv->ifv_encaplen;
1245 ifv->ifv_trunk = trunk;
1248 * Initialize fields from our parent. This duplicates some
1249 * work with ether_ifattach() but allows for non-ethernet
1250 * interfaces to also work.
1252 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1253 ifp->if_baudrate = p->if_baudrate;
1254 ifp->if_output = p->if_output;
1255 ifp->if_input = p->if_input;
1256 ifp->if_resolvemulti = p->if_resolvemulti;
1257 ifp->if_addrlen = p->if_addrlen;
1258 ifp->if_broadcastaddr = p->if_broadcastaddr;
1261 * Copy only a selected subset of flags from the parent.
1262 * Other flags are none of our business.
1264 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1265 ifp->if_flags &= ~VLAN_COPY_FLAGS;
1266 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1267 #undef VLAN_COPY_FLAGS
1269 ifp->if_link_state = p->if_link_state;
1271 vlan_capabilities(ifv);
1274 * Set up our interface address to reflect the underlying
1275 * physical interface's.
1277 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1278 ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1282 * Configure multicast addresses that may already be
1283 * joined on the vlan device.
1285 (void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1287 /* We are ready for operation now. */
1288 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1290 TRUNK_UNLOCK(trunk);
1292 EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_tag);
1299 vlan_unconfig(struct ifnet *ifp)
1303 vlan_unconfig_locked(ifp, 0);
1308 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1310 struct ifvlantrunk *trunk;
1311 struct vlan_mc_entry *mc;
1313 struct ifnet *parent;
1318 ifv = ifp->if_softc;
1319 trunk = ifv->ifv_trunk;
1322 if (trunk != NULL) {
1325 parent = trunk->parent;
1328 * Since the interface is being unconfigured, we need to
1329 * empty the list of multicast groups that we may have joined
1330 * while we were alive from the parent's list.
1332 while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1334 * If the parent interface is being detached,
1335 * all its multicast addresses have already
1336 * been removed. Warn about errors if
1337 * if_delmulti() does fail, but don't abort as
1338 * all callers expect vlan destruction to
1342 error = if_delmulti(parent,
1343 (struct sockaddr *)&mc->mc_addr);
1346 "Failed to delete multicast address from parent: %d\n",
1349 SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1353 vlan_setflags(ifp, 0); /* clear special flags on parent */
1354 vlan_remhash(trunk, ifv);
1355 ifv->ifv_trunk = NULL;
1358 * Check if we were the last.
1360 if (trunk->refcnt == 0) {
1361 trunk->parent->if_vlantrunk = NULL;
1363 * XXXGL: If some ithread has already entered
1364 * vlan_input() and is now blocked on the trunk
1365 * lock, then it should preempt us right after
1366 * unlock and finish its work. Then we will acquire
1367 * lock again in trunk_destroy().
1369 TRUNK_UNLOCK(trunk);
1370 trunk_destroy(trunk);
1372 TRUNK_UNLOCK(trunk);
1375 /* Disconnect from parent. */
1376 if (ifv->ifv_pflags)
1377 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1378 ifp->if_mtu = ETHERMTU;
1379 ifp->if_link_state = LINK_STATE_UNKNOWN;
1380 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1383 * Only dispatch an event if vlan was
1384 * attached, otherwise there is nothing
1385 * to cleanup anyway.
1388 EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_tag);
1391 /* Handle a reference counted flag that should be set on the parent as well */
1393 vlan_setflag(struct ifnet *ifp, int flag, int status,
1394 int (*func)(struct ifnet *, int))
1399 /* XXX VLAN_LOCK_ASSERT(); */
1401 ifv = ifp->if_softc;
1402 status = status ? (ifp->if_flags & flag) : 0;
1403 /* Now "status" contains the flag value or 0 */
1406 * See if recorded parent's status is different from what
1407 * we want it to be. If it is, flip it. We record parent's
1408 * status in ifv_pflags so that we won't clear parent's flag
1409 * we haven't set. In fact, we don't clear or set parent's
1410 * flags directly, but get or release references to them.
1411 * That's why we can be sure that recorded flags still are
1412 * in accord with actual parent's flags.
1414 if (status != (ifv->ifv_pflags & flag)) {
1415 error = (*func)(PARENT(ifv), status);
1418 ifv->ifv_pflags &= ~flag;
1419 ifv->ifv_pflags |= status;
1425 * Handle IFF_* flags that require certain changes on the parent:
1426 * if "status" is true, update parent's flags respective to our if_flags;
1427 * if "status" is false, forcedly clear the flags set on parent.
1430 vlan_setflags(struct ifnet *ifp, int status)
1434 for (i = 0; vlan_pflags[i].flag; i++) {
1435 error = vlan_setflag(ifp, vlan_pflags[i].flag,
1436 status, vlan_pflags[i].func);
1443 /* Inform all vlans that their parent has changed link state */
1445 vlan_link_state(struct ifnet *ifp)
1447 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1453 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1454 if (trunk->vlans[i] != NULL) {
1455 ifv = trunk->vlans[i];
1457 for (i = 0; i < (1 << trunk->hwidth); i++)
1458 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1460 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1461 if_link_state_change(ifv->ifv_ifp,
1462 trunk->parent->if_link_state);
1464 TRUNK_UNLOCK(trunk);
1468 vlan_capabilities(struct ifvlan *ifv)
1470 struct ifnet *p = PARENT(ifv);
1471 struct ifnet *ifp = ifv->ifv_ifp;
1473 TRUNK_LOCK_ASSERT(TRUNK(ifv));
1476 * If the parent interface can do checksum offloading
1477 * on VLANs, then propagate its hardware-assisted
1478 * checksumming flags. Also assert that checksum
1479 * offloading requires hardware VLAN tagging.
1481 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1482 ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1484 if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1485 p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1486 ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1487 ifp->if_hwassist = p->if_hwassist & (CSUM_IP | CSUM_TCP |
1488 CSUM_UDP | CSUM_SCTP | CSUM_IP_FRAGS | CSUM_FRAGMENT);
1490 ifp->if_capenable = 0;
1491 ifp->if_hwassist = 0;
1494 * If the parent interface can do TSO on VLANs then
1495 * propagate the hardware-assisted flag. TSO on VLANs
1496 * does not necessarily require hardware VLAN tagging.
1498 if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1499 ifp->if_capabilities |= p->if_capabilities & IFCAP_TSO;
1500 if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1501 ifp->if_capenable |= p->if_capenable & IFCAP_TSO;
1502 ifp->if_hwassist |= p->if_hwassist & CSUM_TSO;
1504 ifp->if_capenable &= ~(p->if_capenable & IFCAP_TSO);
1505 ifp->if_hwassist &= ~(p->if_hwassist & CSUM_TSO);
1510 vlan_trunk_capabilities(struct ifnet *ifp)
1512 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1518 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1519 if (trunk->vlans[i] != NULL) {
1520 ifv = trunk->vlans[i];
1522 for (i = 0; i < (1 << trunk->hwidth); i++) {
1523 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1525 vlan_capabilities(ifv);
1527 TRUNK_UNLOCK(trunk);
1531 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1540 ifr = (struct ifreq *)data;
1541 ifa = (struct ifaddr *) data;
1542 ifv = ifp->if_softc;
1546 ifp->if_flags |= IFF_UP;
1548 if (ifa->ifa_addr->sa_family == AF_INET)
1549 arp_ifinit(ifp, ifa);
1554 struct sockaddr *sa;
1556 sa = (struct sockaddr *)&ifr->ifr_data;
1557 bcopy(IF_LLADDR(ifp), sa->sa_data, ifp->if_addrlen);
1562 if (TRUNK(ifv) != NULL) {
1565 error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1566 /* Limit the result to the parent's current config. */
1568 struct ifmediareq *ifmr;
1570 ifmr = (struct ifmediareq *)data;
1571 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1572 ifmr->ifm_count = 1;
1573 error = copyout(&ifmr->ifm_current,
1590 * Set the interface MTU.
1593 if (TRUNK(ifv) != NULL) {
1595 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1597 (ifv->ifv_mintu - ifv->ifv_mtufudge))
1600 ifp->if_mtu = ifr->ifr_mtu;
1608 if (ifp->if_vnet != ifp->if_home_vnet) {
1613 error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1616 if (vlr.vlr_parent[0] == '\0') {
1620 p = ifunit(vlr.vlr_parent);
1626 * Don't let the caller set up a VLAN tag with
1627 * anything except VLID bits.
1629 if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1633 error = vlan_config(ifv, p, vlr.vlr_tag);
1637 /* Update flags on the parent, if necessary. */
1638 vlan_setflags(ifp, 1);
1643 if (ifp->if_vnet != ifp->if_home_vnet) {
1648 bzero(&vlr, sizeof(vlr));
1650 if (TRUNK(ifv) != NULL) {
1651 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1652 sizeof(vlr.vlr_parent));
1653 vlr.vlr_tag = ifv->ifv_tag;
1656 error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1661 * We should propagate selected flags to the parent,
1662 * e.g., promiscuous mode.
1664 if (TRUNK(ifv) != NULL)
1665 error = vlan_setflags(ifp, 1);
1671 * If we don't have a parent, just remember the membership for
1674 if (TRUNK(ifv) != NULL)
1675 error = vlan_setmulti(ifp);