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_vid 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);
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_ethervid(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_vid, trunk->hmask);
270 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
271 if (ifv->ifv_vid == ifv2->ifv_vid)
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_vid, 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_vid, 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_vid, 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 vid)
369 TRUNK_LOCK_RASSERT(trunk);
371 LIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
372 if (ifv->ifv_vid == vid)
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 vid)
399 return trunk->vlans[vid];
403 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
406 if (trunk->vlans[ifv->ifv_vid] != NULL)
408 trunk->vlans[ifv->ifv_vid] = ifv;
415 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
418 trunk->vlans[ifv->ifv_vid] = 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);
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);
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 12-bit VLAN VID for this interface, for use by external
621 * components such as Infiniband.
623 * XXXRW: Note that the function name here is historical; it should be named
627 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
631 if (ifp->if_type != IFT_L2VLAN)
634 *vidp = ifv->ifv_vid;
639 * Return a driver specific cookie for this interface. Synchronization
640 * with setcookie must be provided by the driver.
643 vlan_cookie(struct ifnet *ifp)
647 if (ifp->if_type != IFT_L2VLAN)
650 return (ifv->ifv_cookie);
654 * Store a cookie in our softc that drivers can use to store driver
655 * private per-instance data in.
658 vlan_setcookie(struct ifnet *ifp, void *cookie)
662 if (ifp->if_type != IFT_L2VLAN)
665 ifv->ifv_cookie = cookie;
670 * Return the vlan device present at the specific VID.
672 static struct ifnet *
673 vlan_devat(struct ifnet *ifp, uint16_t vid)
675 struct ifvlantrunk *trunk;
678 trunk = ifp->if_vlantrunk;
683 ifv = vlan_gethash(trunk, vid);
686 TRUNK_RUNLOCK(trunk);
691 * VLAN support can be loaded as a module. The only place in the
692 * system that's intimately aware of this is ether_input. We hook
693 * into this code through vlan_input_p which is defined there and
694 * set here. Noone else in the system should be aware of this so
695 * we use an explicit reference here.
697 extern void (*vlan_input_p)(struct ifnet *, struct mbuf *);
699 /* For if_link_state_change() eyes only... */
700 extern void (*vlan_link_state_p)(struct ifnet *);
703 vlan_modevent(module_t mod, int type, void *data)
708 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
709 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
710 if (ifdetach_tag == NULL)
712 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
713 vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
714 if (iflladdr_tag == NULL)
717 vlan_input_p = vlan_input;
718 vlan_link_state_p = vlan_link_state;
719 vlan_trunk_cap_p = vlan_trunk_capabilities;
720 vlan_trunkdev_p = vlan_trunkdev;
721 vlan_cookie_p = vlan_cookie;
722 vlan_setcookie_p = vlan_setcookie;
723 vlan_tag_p = vlan_tag;
724 vlan_devat_p = vlan_devat;
726 if_clone_attach(&vlan_cloner);
729 printf("vlan: initialized, using "
733 "hash tables with chaining"
740 if_clone_detach(&vlan_cloner);
742 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
743 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
745 vlan_link_state_p = NULL;
746 vlan_trunk_cap_p = NULL;
747 vlan_trunkdev_p = NULL;
749 vlan_cookie_p = vlan_cookie;
750 vlan_setcookie_p = vlan_setcookie;
754 printf("vlan: unloaded\n");
762 static moduledata_t vlan_mod = {
768 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
769 MODULE_VERSION(if_vlan, 3);
773 vnet_vlan_init(const void *unused __unused)
776 V_vlan_cloner = vlan_cloner;
777 if_clone_attach(&V_vlan_cloner);
779 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
780 vnet_vlan_init, NULL);
783 vnet_vlan_uninit(const void *unused __unused)
786 if_clone_detach(&V_vlan_cloner);
788 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
789 vnet_vlan_uninit, NULL);
792 static struct ifnet *
793 vlan_clone_match_ethervid(struct if_clone *ifc, const char *name, int *vidp)
799 /* Check for <etherif>.<vlan> style interface names. */
800 IFNET_RLOCK_NOSLEEP();
801 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
803 * We can handle non-ethernet hardware types as long as
804 * they handle the tagging and headers themselves.
806 if (ifp->if_type != IFT_ETHER &&
807 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
809 if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
811 cp = name + strlen(ifp->if_xname);
817 for(; *cp >= '0' && *cp <= '9'; cp++)
818 vid = (vid * 10) + (*cp - '0');
825 IFNET_RUNLOCK_NOSLEEP();
831 vlan_clone_match(struct if_clone *ifc, const char *name)
835 if (vlan_clone_match_ethervid(ifc, name, NULL) != NULL)
838 if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
840 for (cp = name + 4; *cp != '\0'; cp++) {
841 if (*cp < '0' || *cp > '9')
849 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
861 struct sockaddr_dl *sdl;
863 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
866 * There are 3 (ugh) ways to specify the cloned device:
867 * o pass a parameter block with the clone request.
868 * o specify parameters in the text of the clone device name
869 * o specify no parameters and get an unattached device that
870 * must be configured separately.
871 * The first technique is preferred; the latter two are
872 * supported for backwards compatibilty.
874 * XXXRW: Note historic use of the word "tag" here. New ioctls may be
878 error = copyin(params, &vlr, sizeof(vlr));
881 p = ifunit(vlr.vlr_parent);
885 * Don't let the caller set up a VLAN VID with
886 * anything except VLID bits.
888 if (vlr.vlr_tag & ~EVL_VLID_MASK)
890 error = ifc_name2unit(name, &unit);
896 wildcard = (unit < 0);
897 } else if ((p = vlan_clone_match_ethervid(ifc, name, &vid)) != NULL) {
903 * Don't let the caller set up a VLAN VID with
904 * anything except VLID bits.
906 if (vid & ~EVL_VLID_MASK)
911 error = ifc_name2unit(name, &unit);
915 wildcard = (unit < 0);
918 error = ifc_alloc_unit(ifc, &unit);
922 /* In the wildcard case, we need to update the name. */
924 for (dp = name; *dp != '\0'; dp++);
925 if (snprintf(dp, len - (dp-name), "%d", unit) >
926 len - (dp-name) - 1) {
927 panic("%s: interface name too long", __func__);
931 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
932 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
934 ifc_free_unit(ifc, unit);
938 SLIST_INIT(&ifv->vlan_mc_listhead);
942 * Set the name manually rather than using if_initname because
943 * we don't conform to the default naming convention for interfaces.
945 strlcpy(ifp->if_xname, name, IFNAMSIZ);
946 ifp->if_dname = ifc->ifc_name;
947 ifp->if_dunit = unit;
948 /* NB: flags are not set here */
949 ifp->if_linkmib = &ifv->ifv_mib;
950 ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
951 /* NB: mtu is not set here */
953 ifp->if_init = vlan_init;
954 ifp->if_transmit = vlan_transmit;
955 ifp->if_qflush = vlan_qflush;
956 ifp->if_ioctl = vlan_ioctl;
957 ifp->if_flags = VLAN_IFFLAGS;
958 ether_ifattach(ifp, eaddr);
959 /* Now undo some of the damage... */
960 ifp->if_baudrate = 0;
961 ifp->if_type = IFT_L2VLAN;
962 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
964 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
965 sdl->sdl_type = IFT_L2VLAN;
968 error = vlan_config(ifv, p, vid);
971 * Since we've partialy failed, we need to back
972 * out all the way, otherwise userland could get
973 * confused. Thus, we destroy the interface.
978 ifc_free_unit(ifc, unit);
984 /* Update flags on the parent, if necessary. */
985 vlan_setflags(ifp, 1);
992 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
994 struct ifvlan *ifv = ifp->if_softc;
995 int unit = ifp->if_dunit;
997 ether_ifdetach(ifp); /* first, remove it from system-wide lists */
998 vlan_unconfig(ifp); /* now it can be unconfigured and freed */
1001 ifc_free_unit(ifc, unit);
1007 * The ifp->if_init entry point for vlan(4) is a no-op.
1010 vlan_init(void *foo __unused)
1015 * The if_transmit method for vlan(4) interface.
1018 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1022 int error, len, mcast;
1024 ifv = ifp->if_softc;
1026 len = m->m_pkthdr.len;
1027 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1032 * Do not run parent's if_transmit() if the parent is not up,
1033 * or parent's driver will cause a system crash.
1035 if (!UP_AND_RUNNING(p)) {
1042 * Pad the frame to the minimum size allowed if told to.
1043 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
1044 * paragraph C.4.4.3.b. It can help to work around buggy
1045 * bridges that violate paragraph C.4.4.3.a from the same
1046 * document, i.e., fail to pad short frames after untagging.
1047 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
1048 * untagging it will produce a 62-byte frame, which is a runt
1049 * and requires padding. There are VLAN-enabled network
1050 * devices that just discard such runts instead or mishandle
1053 if (soft_pad && p->if_type == IFT_ETHER) {
1054 static char pad[8]; /* just zeros */
1057 for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
1058 n > 0; n -= sizeof(pad))
1059 if (!m_append(m, min(n, sizeof(pad)), pad))
1063 if_printf(ifp, "cannot pad short frame\n");
1071 * If underlying interface can do VLAN tag insertion itself,
1072 * just pass the packet along. However, we need some way to
1073 * tell the interface where the packet came from so that it
1074 * knows how to find the VLAN tag to use, so we attach a
1075 * packet tag that holds it.
1077 if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1078 m->m_pkthdr.ether_vtag = ifv->ifv_vid;
1079 m->m_flags |= M_VLANTAG;
1081 m = ether_vlanencap(m, ifv->ifv_vid);
1083 if_printf(ifp, "unable to prepend VLAN header\n");
1090 * Send it, precisely as ether_output() would have.
1092 error = (p->if_transmit)(p, m);
1095 ifp->if_omcasts += mcast;
1096 ifp->if_obytes += len;
1103 * The ifp->if_qflush entry point for vlan(4) is a no-op.
1106 vlan_qflush(struct ifnet *ifp __unused)
1111 vlan_input(struct ifnet *ifp, struct mbuf *m)
1113 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1117 KASSERT(trunk != NULL, ("%s: no trunk", __func__));
1119 if (m->m_flags & M_VLANTAG) {
1121 * Packet is tagged, but m contains a normal
1122 * Ethernet frame; the tag is stored out-of-band.
1124 vid = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
1125 m->m_flags &= ~M_VLANTAG;
1127 struct ether_vlan_header *evl;
1130 * Packet is tagged in-band as specified by 802.1q.
1132 switch (ifp->if_type) {
1134 if (m->m_len < sizeof(*evl) &&
1135 (m = m_pullup(m, sizeof(*evl))) == NULL) {
1136 if_printf(ifp, "cannot pullup VLAN header\n");
1139 evl = mtod(m, struct ether_vlan_header *);
1140 vid = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1143 * Remove the 802.1q header by copying the Ethernet
1144 * addresses over it and adjusting the beginning of
1145 * the data in the mbuf. The encapsulated Ethernet
1146 * type field is already in place.
1148 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1149 ETHER_HDR_LEN - ETHER_TYPE_LEN);
1150 m_adj(m, ETHER_VLAN_ENCAP_LEN);
1155 panic("%s: %s has unsupported if_type %u",
1156 __func__, ifp->if_xname, ifp->if_type);
1165 ifv = vlan_gethash(trunk, vid);
1166 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1167 TRUNK_RUNLOCK(trunk);
1172 TRUNK_RUNLOCK(trunk);
1174 m->m_pkthdr.rcvif = ifv->ifv_ifp;
1175 ifv->ifv_ifp->if_ipackets++;
1177 /* Pass it back through the parent's input routine. */
1178 (*ifp->if_input)(ifv->ifv_ifp, m);
1182 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid)
1184 struct ifvlantrunk *trunk;
1188 /* VID numbers 0x0 and 0xFFF are reserved */
1189 if (vid == 0 || vid == 0xFFF)
1191 if (p->if_type != IFT_ETHER &&
1192 (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1193 return (EPROTONOSUPPORT);
1194 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1195 return (EPROTONOSUPPORT);
1199 if (p->if_vlantrunk == NULL) {
1200 trunk = malloc(sizeof(struct ifvlantrunk),
1201 M_VLAN, M_WAITOK | M_ZERO);
1202 vlan_inithash(trunk);
1204 if (p->if_vlantrunk != NULL) {
1205 /* A race that that is very unlikely to be hit. */
1206 vlan_freehash(trunk);
1207 free(trunk, M_VLAN);
1210 TRUNK_LOCK_INIT(trunk);
1212 p->if_vlantrunk = trunk;
1217 trunk = p->if_vlantrunk;
1221 ifv->ifv_vid = vid; /* must set this before vlan_inshash() */
1222 error = vlan_inshash(trunk, ifv);
1225 ifv->ifv_proto = ETHERTYPE_VLAN;
1226 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1227 ifv->ifv_mintu = ETHERMIN;
1228 ifv->ifv_pflags = 0;
1231 * If the parent supports the VLAN_MTU capability,
1232 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1235 if (p->if_capenable & IFCAP_VLAN_MTU) {
1237 * No need to fudge the MTU since the parent can
1238 * handle extended frames.
1240 ifv->ifv_mtufudge = 0;
1243 * Fudge the MTU by the encapsulation size. This
1244 * makes us incompatible with strictly compliant
1245 * 802.1Q implementations, but allows us to use
1246 * the feature with other NetBSD implementations,
1247 * which might still be useful.
1249 ifv->ifv_mtufudge = ifv->ifv_encaplen;
1252 ifv->ifv_trunk = trunk;
1255 * Initialize fields from our parent. This duplicates some
1256 * work with ether_ifattach() but allows for non-ethernet
1257 * interfaces to also work.
1259 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1260 ifp->if_baudrate = p->if_baudrate;
1261 ifp->if_output = p->if_output;
1262 ifp->if_input = p->if_input;
1263 ifp->if_resolvemulti = p->if_resolvemulti;
1264 ifp->if_addrlen = p->if_addrlen;
1265 ifp->if_broadcastaddr = p->if_broadcastaddr;
1268 * Copy only a selected subset of flags from the parent.
1269 * Other flags are none of our business.
1271 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1272 ifp->if_flags &= ~VLAN_COPY_FLAGS;
1273 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1274 #undef VLAN_COPY_FLAGS
1276 ifp->if_link_state = p->if_link_state;
1278 vlan_capabilities(ifv);
1281 * Set up our interface address to reflect the underlying
1282 * physical interface's.
1284 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1285 ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1289 * Configure multicast addresses that may already be
1290 * joined on the vlan device.
1292 (void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1294 /* We are ready for operation now. */
1295 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1297 TRUNK_UNLOCK(trunk);
1299 EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1306 vlan_unconfig(struct ifnet *ifp)
1310 vlan_unconfig_locked(ifp);
1315 vlan_unconfig_locked(struct ifnet *ifp)
1317 struct ifvlantrunk *trunk;
1318 struct vlan_mc_entry *mc;
1320 struct ifnet *parent;
1324 ifv = ifp->if_softc;
1325 trunk = ifv->ifv_trunk;
1328 if (trunk != NULL) {
1331 parent = trunk->parent;
1334 * Since the interface is being unconfigured, we need to
1335 * empty the list of multicast groups that we may have joined
1336 * while we were alive from the parent's list.
1338 while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1340 * This may fail if the parent interface is
1341 * being detached. Regardless, we should do a
1342 * best effort to free this interface as much
1343 * as possible as all callers expect vlan
1344 * destruction to succeed.
1346 (void)if_delmulti(parent,
1347 (struct sockaddr *)&mc->mc_addr);
1348 SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1352 vlan_setflags(ifp, 0); /* clear special flags on parent */
1353 vlan_remhash(trunk, ifv);
1354 ifv->ifv_trunk = NULL;
1357 * Check if we were the last.
1359 if (trunk->refcnt == 0) {
1360 trunk->parent->if_vlantrunk = NULL;
1362 * XXXGL: If some ithread has already entered
1363 * vlan_input() and is now blocked on the trunk
1364 * lock, then it should preempt us right after
1365 * unlock and finish its work. Then we will acquire
1366 * lock again in trunk_destroy().
1368 TRUNK_UNLOCK(trunk);
1369 trunk_destroy(trunk);
1371 TRUNK_UNLOCK(trunk);
1374 /* Disconnect from parent. */
1375 if (ifv->ifv_pflags)
1376 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1377 ifp->if_mtu = ETHERMTU;
1378 ifp->if_link_state = LINK_STATE_UNKNOWN;
1379 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1382 * Only dispatch an event if vlan was
1383 * attached, otherwise there is nothing
1384 * to cleanup anyway.
1387 EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1390 /* Handle a reference counted flag that should be set on the parent as well */
1392 vlan_setflag(struct ifnet *ifp, int flag, int status,
1393 int (*func)(struct ifnet *, int))
1398 /* XXX VLAN_LOCK_ASSERT(); */
1400 ifv = ifp->if_softc;
1401 status = status ? (ifp->if_flags & flag) : 0;
1402 /* Now "status" contains the flag value or 0 */
1405 * See if recorded parent's status is different from what
1406 * we want it to be. If it is, flip it. We record parent's
1407 * status in ifv_pflags so that we won't clear parent's flag
1408 * we haven't set. In fact, we don't clear or set parent's
1409 * flags directly, but get or release references to them.
1410 * That's why we can be sure that recorded flags still are
1411 * in accord with actual parent's flags.
1413 if (status != (ifv->ifv_pflags & flag)) {
1414 error = (*func)(PARENT(ifv), status);
1417 ifv->ifv_pflags &= ~flag;
1418 ifv->ifv_pflags |= status;
1424 * Handle IFF_* flags that require certain changes on the parent:
1425 * if "status" is true, update parent's flags respective to our if_flags;
1426 * if "status" is false, forcedly clear the flags set on parent.
1429 vlan_setflags(struct ifnet *ifp, int status)
1433 for (i = 0; vlan_pflags[i].flag; i++) {
1434 error = vlan_setflag(ifp, vlan_pflags[i].flag,
1435 status, vlan_pflags[i].func);
1442 /* Inform all vlans that their parent has changed link state */
1444 vlan_link_state(struct ifnet *ifp)
1446 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1452 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1453 if (trunk->vlans[i] != NULL) {
1454 ifv = trunk->vlans[i];
1456 for (i = 0; i < (1 << trunk->hwidth); i++)
1457 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1459 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1460 if_link_state_change(ifv->ifv_ifp,
1461 trunk->parent->if_link_state);
1463 TRUNK_UNLOCK(trunk);
1467 vlan_capabilities(struct ifvlan *ifv)
1469 struct ifnet *p = PARENT(ifv);
1470 struct ifnet *ifp = ifv->ifv_ifp;
1472 TRUNK_LOCK_ASSERT(TRUNK(ifv));
1475 * If the parent interface can do checksum offloading
1476 * on VLANs, then propagate its hardware-assisted
1477 * checksumming flags. Also assert that checksum
1478 * offloading requires hardware VLAN tagging.
1480 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1481 ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1483 if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1484 p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1485 ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1486 ifp->if_hwassist = p->if_hwassist & (CSUM_IP | CSUM_TCP |
1487 CSUM_UDP | CSUM_SCTP | CSUM_IP_FRAGS | CSUM_FRAGMENT);
1489 ifp->if_capenable = 0;
1490 ifp->if_hwassist = 0;
1493 * If the parent interface can do TSO on VLANs then
1494 * propagate the hardware-assisted flag. TSO on VLANs
1495 * does not necessarily require hardware VLAN tagging.
1497 if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1498 ifp->if_capabilities |= p->if_capabilities & IFCAP_TSO;
1499 if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1500 ifp->if_capenable |= p->if_capenable & IFCAP_TSO;
1501 ifp->if_hwassist |= p->if_hwassist & CSUM_TSO;
1503 ifp->if_capenable &= ~(p->if_capenable & IFCAP_TSO);
1504 ifp->if_hwassist &= ~(p->if_hwassist & CSUM_TSO);
1509 vlan_trunk_capabilities(struct ifnet *ifp)
1511 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1517 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1518 if (trunk->vlans[i] != NULL) {
1519 ifv = trunk->vlans[i];
1521 for (i = 0; i < (1 << trunk->hwidth); i++) {
1522 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1524 vlan_capabilities(ifv);
1526 TRUNK_UNLOCK(trunk);
1530 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1539 ifr = (struct ifreq *)data;
1540 ifa = (struct ifaddr *) data;
1541 ifv = ifp->if_softc;
1545 ifp->if_flags |= IFF_UP;
1547 if (ifa->ifa_addr->sa_family == AF_INET)
1548 arp_ifinit(ifp, ifa);
1553 struct sockaddr *sa;
1555 sa = (struct sockaddr *)&ifr->ifr_data;
1556 bcopy(IF_LLADDR(ifp), sa->sa_data, ifp->if_addrlen);
1561 if (TRUNK(ifv) != NULL) {
1564 error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1565 /* Limit the result to the parent's current config. */
1567 struct ifmediareq *ifmr;
1569 ifmr = (struct ifmediareq *)data;
1570 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1571 ifmr->ifm_count = 1;
1572 error = copyout(&ifmr->ifm_current,
1589 * Set the interface MTU.
1592 if (TRUNK(ifv) != NULL) {
1594 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1596 (ifv->ifv_mintu - ifv->ifv_mtufudge))
1599 ifp->if_mtu = ifr->ifr_mtu;
1608 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1609 * interface to be delegated to a jail without allowing the
1610 * jail to change what underlying interface/VID it is
1611 * associated with. We are not entirely convinced that this
1612 * is the right way to accomplish that policy goal.
1614 if (ifp->if_vnet != ifp->if_home_vnet) {
1619 error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1622 if (vlr.vlr_parent[0] == '\0') {
1626 p = ifunit(vlr.vlr_parent);
1632 * Don't let the caller set up a VLAN VID with
1633 * anything except VLID bits.
1635 if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1639 error = vlan_config(ifv, p, vlr.vlr_tag);
1643 /* Update flags on the parent, if necessary. */
1644 vlan_setflags(ifp, 1);
1649 if (ifp->if_vnet != ifp->if_home_vnet) {
1654 bzero(&vlr, sizeof(vlr));
1656 if (TRUNK(ifv) != NULL) {
1657 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1658 sizeof(vlr.vlr_parent));
1659 vlr.vlr_tag = ifv->ifv_vid;
1662 error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1667 * We should propagate selected flags to the parent,
1668 * e.g., promiscuous mode.
1670 if (TRUNK(ifv) != NULL)
1671 error = vlan_setflags(ifp, 1);
1677 * If we don't have a parent, just remember the membership for
1680 if (TRUNK(ifv) != NULL)
1681 error = vlan_setmulti(ifp);