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
33 * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
34 * Might be extended some day to also handle IEEE 802.1p priority
35 * tagging. This is sort of sneaky in the implementation, since
36 * we need to pretend to be enough of an Ethernet implementation
37 * to make arp work. The way we do this is by telling everyone
38 * that we are an Ethernet, and then catch the packets that
39 * ether_output() left on our output queue when it calls
40 * if_start(), rewrite them for use by the real outgoing interface,
41 * and ask it to send them.
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>
60 #include <net/ethernet.h>
62 #include <net/if_clone.h>
63 #include <net/if_dl.h>
64 #include <net/if_types.h>
65 #include <net/if_vlan_var.h>
67 #define VLANNAME "vlan"
68 #define VLAN_DEF_HWIDTH 4
69 #define VLAN_IFFLAGS (IFF_BROADCAST | IFF_MULTICAST)
71 #define UP_AND_RUNNING(ifp) \
72 ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
74 LIST_HEAD(ifvlanhead, ifvlan);
77 struct ifnet *parent; /* parent interface of this trunk */
80 #define VLAN_ARRAY_SIZE (EVL_VLID_MASK + 1)
81 struct ifvlan *vlans[VLAN_ARRAY_SIZE]; /* static table */
83 struct ifvlanhead *hash; /* dynamic hash-list table */
90 struct vlan_mc_entry {
91 struct ether_addr mc_addr;
92 SLIST_ENTRY(vlan_mc_entry) mc_entries;
96 struct ifvlantrunk *ifv_trunk;
97 struct ifnet *ifv_ifp;
98 #define TRUNK(ifv) ((ifv)->ifv_trunk)
99 #define PARENT(ifv) ((ifv)->ifv_trunk->parent)
100 int ifv_pflags; /* special flags we have set on parent */
102 int ifvm_encaplen; /* encapsulation length */
103 int ifvm_mtufudge; /* MTU fudged by this much */
104 int ifvm_mintu; /* min transmission unit */
105 uint16_t ifvm_proto; /* encapsulation ethertype */
106 uint16_t ifvm_tag; /* tag to apply on packets leaving if */
108 SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
110 LIST_ENTRY(ifvlan) ifv_list;
113 #define ifv_proto ifv_mib.ifvm_proto
114 #define ifv_tag ifv_mib.ifvm_tag
115 #define ifv_encaplen ifv_mib.ifvm_encaplen
116 #define ifv_mtufudge ifv_mib.ifvm_mtufudge
117 #define ifv_mintu ifv_mib.ifvm_mintu
119 /* Special flags we should propagate to parent. */
122 int (*func)(struct ifnet *, int);
124 {IFF_PROMISC, ifpromisc},
125 {IFF_ALLMULTI, if_allmulti},
129 SYSCTL_DECL(_net_link);
130 SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0, "IEEE 802.1Q VLAN");
131 SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0, "for consistency");
133 static int soft_pad = 0;
134 SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW, &soft_pad, 0,
135 "pad short frames before tagging");
137 static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface");
139 static eventhandler_tag ifdetach_tag;
142 * We have a global mutex, that is used to serialize configuration
143 * changes and isn't used in normal packet delivery.
145 * We also have a per-trunk rwlock, that is locked shared on packet
146 * processing and exclusive when configuration is changed.
148 * The VLAN_ARRAY substitutes the dynamic hash with a static array
149 * with 4096 entries. In theory this can give a boost in processing,
150 * however on practice it does not. Probably this is because array
151 * is too big to fit into CPU cache.
153 static struct mtx ifv_mtx;
154 #define VLAN_LOCK_INIT() mtx_init(&ifv_mtx, "vlan_global", NULL, MTX_DEF)
155 #define VLAN_LOCK_DESTROY() mtx_destroy(&ifv_mtx)
156 #define VLAN_LOCK_ASSERT() mtx_assert(&ifv_mtx, MA_OWNED)
157 #define VLAN_LOCK() mtx_lock(&ifv_mtx)
158 #define VLAN_UNLOCK() mtx_unlock(&ifv_mtx)
159 #define TRUNK_LOCK_INIT(trunk) rw_init(&(trunk)->rw, VLANNAME)
160 #define TRUNK_LOCK_DESTROY(trunk) rw_destroy(&(trunk)->rw)
161 #define TRUNK_LOCK(trunk) rw_wlock(&(trunk)->rw)
162 #define TRUNK_UNLOCK(trunk) rw_wunlock(&(trunk)->rw)
163 #define TRUNK_LOCK_ASSERT(trunk) rw_assert(&(trunk)->rw, RA_WLOCKED)
164 #define TRUNK_RLOCK(trunk) rw_rlock(&(trunk)->rw)
165 #define TRUNK_RUNLOCK(trunk) rw_runlock(&(trunk)->rw)
166 #define TRUNK_LOCK_RASSERT(trunk) rw_assert(&(trunk)->rw, RA_RLOCKED)
169 static void vlan_inithash(struct ifvlantrunk *trunk);
170 static void vlan_freehash(struct ifvlantrunk *trunk);
171 static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
172 static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
173 static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
174 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
177 static void trunk_destroy(struct ifvlantrunk *trunk);
179 static void vlan_start(struct ifnet *ifp);
180 static void vlan_init(void *foo);
181 static void vlan_input(struct ifnet *ifp, struct mbuf *m);
182 static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
183 static int vlan_setflag(struct ifnet *ifp, int flag, int status,
184 int (*func)(struct ifnet *, int));
185 static int vlan_setflags(struct ifnet *ifp, int status);
186 static int vlan_setmulti(struct ifnet *ifp);
187 static int vlan_unconfig(struct ifnet *ifp);
188 static int vlan_unconfig_locked(struct ifnet *ifp);
189 static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
190 static void vlan_link_state(struct ifnet *ifp, int link);
191 static void vlan_capabilities(struct ifvlan *ifv);
192 static void vlan_trunk_capabilities(struct ifnet *ifp);
194 static struct ifnet *vlan_clone_match_ethertag(struct if_clone *,
195 const char *, int *);
196 static int vlan_clone_match(struct if_clone *, const char *);
197 static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
198 static int vlan_clone_destroy(struct if_clone *, struct ifnet *);
200 static void vlan_ifdetach(void *arg, struct ifnet *ifp);
202 static struct if_clone vlan_cloner = IFC_CLONE_INITIALIZER(VLANNAME, NULL,
203 IF_MAXUNIT, NULL, vlan_clone_match, vlan_clone_create, vlan_clone_destroy);
206 #define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
209 vlan_inithash(struct ifvlantrunk *trunk)
214 * The trunk must not be locked here since we call malloc(M_WAITOK).
215 * It is OK in case this function is called before the trunk struct
216 * gets hooked up and becomes visible from other threads.
219 KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
220 ("%s: hash already initialized", __func__));
222 trunk->hwidth = VLAN_DEF_HWIDTH;
223 n = 1 << trunk->hwidth;
224 trunk->hmask = n - 1;
225 trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
226 for (i = 0; i < n; i++)
227 LIST_INIT(&trunk->hash[i]);
231 vlan_freehash(struct ifvlantrunk *trunk)
236 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
237 for (i = 0; i < (1 << trunk->hwidth); i++)
238 KASSERT(LIST_EMPTY(&trunk->hash[i]),
239 ("%s: hash table not empty", __func__));
241 free(trunk->hash, M_VLAN);
243 trunk->hwidth = trunk->hmask = 0;
247 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
252 TRUNK_LOCK_ASSERT(trunk);
253 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
255 b = 1 << trunk->hwidth;
256 i = HASH(ifv->ifv_tag, trunk->hmask);
257 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
258 if (ifv->ifv_tag == ifv2->ifv_tag)
262 * Grow the hash when the number of vlans exceeds half of the number of
263 * hash buckets squared. This will make the average linked-list length
266 if (trunk->refcnt > (b * b) / 2) {
267 vlan_growhash(trunk, 1);
268 i = HASH(ifv->ifv_tag, trunk->hmask);
270 LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
277 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
282 TRUNK_LOCK_ASSERT(trunk);
283 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
285 b = 1 << trunk->hwidth;
286 i = HASH(ifv->ifv_tag, trunk->hmask);
287 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
290 LIST_REMOVE(ifv2, ifv_list);
291 if (trunk->refcnt < (b * b) / 2)
292 vlan_growhash(trunk, -1);
296 panic("%s: vlan not found\n", __func__);
297 return (ENOENT); /*NOTREACHED*/
301 * Grow the hash larger or smaller if memory permits.
304 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
307 struct ifvlanhead *hash2;
308 int hwidth2, i, j, n, n2;
310 TRUNK_LOCK_ASSERT(trunk);
311 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
314 /* Harmless yet obvious coding error */
315 printf("%s: howmuch is 0\n", __func__);
319 hwidth2 = trunk->hwidth + howmuch;
320 n = 1 << trunk->hwidth;
322 /* Do not shrink the table below the default */
323 if (hwidth2 < VLAN_DEF_HWIDTH)
326 /* M_NOWAIT because we're called with trunk mutex held */
327 hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
329 printf("%s: out of memory -- hash size not changed\n",
331 return; /* We can live with the old hash table */
333 for (j = 0; j < n2; j++)
334 LIST_INIT(&hash2[j]);
335 for (i = 0; i < n; i++)
336 while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
337 LIST_REMOVE(ifv, ifv_list);
338 j = HASH(ifv->ifv_tag, n2 - 1);
339 LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
341 free(trunk->hash, M_VLAN);
343 trunk->hwidth = hwidth2;
344 trunk->hmask = n2 - 1;
347 if_printf(trunk->parent,
348 "VLAN hash table resized from %d to %d buckets\n", n, n2);
351 static __inline struct ifvlan *
352 vlan_gethash(struct ifvlantrunk *trunk, uint16_t tag)
356 TRUNK_LOCK_RASSERT(trunk);
358 LIST_FOREACH(ifv, &trunk->hash[HASH(tag, trunk->hmask)], ifv_list)
359 if (ifv->ifv_tag == tag)
365 /* Debugging code to view the hashtables. */
367 vlan_dumphash(struct ifvlantrunk *trunk)
372 for (i = 0; i < (1 << trunk->hwidth); i++) {
374 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
375 printf("%s ", ifv->ifv_ifp->if_xname);
380 #endif /* !VLAN_ARRAY */
383 trunk_destroy(struct ifvlantrunk *trunk)
389 vlan_freehash(trunk);
391 trunk->parent->if_vlantrunk = NULL;
393 TRUNK_LOCK_DESTROY(trunk);
398 * Program our multicast filter. What we're actually doing is
399 * programming the multicast filter of the parent. This has the
400 * side effect of causing the parent interface to receive multicast
401 * traffic that it doesn't really want, which ends up being discarded
402 * later by the upper protocol layers. Unfortunately, there's no way
403 * to avoid this: there really is only one physical interface.
405 * XXX: There is a possible race here if more than one thread is
406 * modifying the multicast state of the vlan interface at the same time.
409 vlan_setmulti(struct ifnet *ifp)
412 struct ifmultiaddr *ifma, *rifma = NULL;
414 struct vlan_mc_entry *mc;
415 struct sockaddr_dl sdl;
418 /*VLAN_LOCK_ASSERT();*/
420 /* Find the parent. */
424 bzero((char *)&sdl, sizeof(sdl));
425 sdl.sdl_len = sizeof(sdl);
426 sdl.sdl_family = AF_LINK;
427 sdl.sdl_index = ifp_p->if_index;
428 sdl.sdl_type = IFT_ETHER;
429 sdl.sdl_alen = ETHER_ADDR_LEN;
431 /* First, remove any existing filter entries. */
432 while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
433 bcopy((char *)&mc->mc_addr, LLADDR(&sdl), ETHER_ADDR_LEN);
434 error = if_delmulti(ifp_p, (struct sockaddr *)&sdl);
437 SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
441 /* Now program new ones. */
442 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
443 if (ifma->ifma_addr->sa_family != AF_LINK)
445 mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
448 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
449 (char *)&mc->mc_addr, ETHER_ADDR_LEN);
450 SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
451 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
452 LLADDR(&sdl), ETHER_ADDR_LEN);
453 error = if_addmulti(ifp_p, (struct sockaddr *)&sdl, &rifma);
462 * A handler for network interface departure events.
463 * Track departure of trunks here so that we don't access invalid
464 * pointers or whatever if a trunk is ripped from under us, e.g.,
465 * by ejecting its hot-plug card.
468 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
474 * Check if it's a trunk interface first of all
475 * to avoid needless locking.
477 if (ifp->if_vlantrunk == NULL)
482 * OK, it's a trunk. Loop over and detach all vlan's on it.
483 * Check trunk pointer after each vlan_unconfig() as it will
484 * free it and set to NULL after the last vlan was detached.
487 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
488 if ((ifv = ifp->if_vlantrunk->vlans[i])) {
489 vlan_unconfig_locked(ifv->ifv_ifp);
490 if (ifp->if_vlantrunk == NULL)
493 #else /* VLAN_ARRAY */
495 for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
496 if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
497 vlan_unconfig_locked(ifv->ifv_ifp);
498 if (ifp->if_vlantrunk)
499 goto restart; /* trunk->hwidth can change */
503 #endif /* VLAN_ARRAY */
504 /* Trunk should have been destroyed in vlan_unconfig(). */
505 KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
510 * VLAN support can be loaded as a module. The only place in the
511 * system that's intimately aware of this is ether_input. We hook
512 * into this code through vlan_input_p which is defined there and
513 * set here. Noone else in the system should be aware of this so
514 * we use an explicit reference here.
516 extern void (*vlan_input_p)(struct ifnet *, struct mbuf *);
518 /* For if_link_state_change() eyes only... */
519 extern void (*vlan_link_state_p)(struct ifnet *, int);
522 vlan_modevent(module_t mod, int type, void *data)
527 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
528 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
529 if (ifdetach_tag == NULL)
532 vlan_input_p = vlan_input;
533 vlan_link_state_p = vlan_link_state;
534 vlan_trunk_cap_p = vlan_trunk_capabilities;
535 if_clone_attach(&vlan_cloner);
537 printf("vlan: initialized, using "
541 "hash tables with chaining"
547 if_clone_detach(&vlan_cloner);
548 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
550 vlan_link_state_p = NULL;
551 vlan_trunk_cap_p = NULL;
554 printf("vlan: unloaded\n");
562 static moduledata_t vlan_mod = {
568 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
569 MODULE_VERSION(if_vlan, 3);
570 MODULE_DEPEND(if_vlan, miibus, 1, 1, 1);
572 static struct ifnet *
573 vlan_clone_match_ethertag(struct if_clone *ifc, const char *name, int *tag)
579 /* Check for <etherif>.<vlan> style interface names. */
581 TAILQ_FOREACH(ifp, &ifnet, if_link) {
582 if (ifp->if_type != IFT_ETHER)
584 if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0)
586 cp = name + strlen(ifp->if_xname);
589 for(; *cp != '\0'; cp++) {
590 if (*cp < '0' || *cp > '9')
592 t = (t * 10) + (*cp - '0');
604 vlan_clone_match(struct if_clone *ifc, const char *name)
608 if (vlan_clone_match_ethertag(ifc, name, NULL) != NULL)
611 if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0)
613 for (cp = name + 4; *cp != '\0'; cp++) {
614 if (*cp < '0' || *cp > '9')
622 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
634 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
637 * There are 3 (ugh) ways to specify the cloned device:
638 * o pass a parameter block with the clone request.
639 * o specify parameters in the text of the clone device name
640 * o specify no parameters and get an unattached device that
641 * must be configured separately.
642 * The first technique is preferred; the latter two are
643 * supported for backwards compatibilty.
646 error = copyin(params, &vlr, sizeof(vlr));
649 p = ifunit(vlr.vlr_parent);
653 * Don't let the caller set up a VLAN tag with
654 * anything except VLID bits.
656 if (vlr.vlr_tag & ~EVL_VLID_MASK)
658 error = ifc_name2unit(name, &unit);
664 wildcard = (unit < 0);
665 } else if ((p = vlan_clone_match_ethertag(ifc, name, &tag)) != NULL) {
671 * Don't let the caller set up a VLAN tag with
672 * anything except VLID bits.
674 if (tag & ~EVL_VLID_MASK)
679 error = ifc_name2unit(name, &unit);
683 wildcard = (unit < 0);
686 error = ifc_alloc_unit(ifc, &unit);
690 /* In the wildcard case, we need to update the name. */
692 for (dp = name; *dp != '\0'; dp++);
693 if (snprintf(dp, len - (dp-name), "%d", unit) >
694 len - (dp-name) - 1) {
695 panic("%s: interface name too long", __func__);
699 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
700 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
702 ifc_free_unit(ifc, unit);
706 SLIST_INIT(&ifv->vlan_mc_listhead);
710 * Set the name manually rather than using if_initname because
711 * we don't conform to the default naming convention for interfaces.
713 strlcpy(ifp->if_xname, name, IFNAMSIZ);
714 ifp->if_dname = ifc->ifc_name;
715 ifp->if_dunit = unit;
716 /* NB: flags are not set here */
717 ifp->if_linkmib = &ifv->ifv_mib;
718 ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
719 /* NB: mtu is not set here */
721 ifp->if_init = vlan_init;
722 ifp->if_start = vlan_start;
723 ifp->if_ioctl = vlan_ioctl;
724 ifp->if_snd.ifq_maxlen = ifqmaxlen;
725 ifp->if_flags = VLAN_IFFLAGS;
726 ether_ifattach(ifp, eaddr);
727 /* Now undo some of the damage... */
728 ifp->if_baudrate = 0;
729 ifp->if_type = IFT_L2VLAN;
730 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
733 error = vlan_config(ifv, p, tag);
736 * Since we've partialy failed, we need to back
737 * out all the way, otherwise userland could get
738 * confused. Thus, we destroy the interface.
742 if_free_type(ifp, IFT_ETHER);
748 /* Update flags on the parent, if necessary. */
749 vlan_setflags(ifp, 1);
756 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
758 struct ifvlan *ifv = ifp->if_softc;
759 int unit = ifp->if_dunit;
761 ether_ifdetach(ifp); /* first, remove it from system-wide lists */
762 vlan_unconfig(ifp); /* now it can be unconfigured and freed */
763 if_free_type(ifp, IFT_ETHER);
765 ifc_free_unit(ifc, unit);
771 * The ifp->if_init entry point for vlan(4) is a no-op.
774 vlan_init(void *foo __unused)
779 * The if_start method for vlan(4) interface. It doesn't
780 * raises the IFF_DRV_OACTIVE flag, since it is called
781 * only from IFQ_HANDOFF() macro in ether_output_frame().
782 * If the interface queue is full, and vlan_start() is
783 * not called, the queue would never get emptied and
784 * interface would stall forever.
787 vlan_start(struct ifnet *ifp)
798 IF_DEQUEUE(&ifp->if_snd, m);
804 * Do not run parent's if_start() if the parent is not up,
805 * or parent's driver will cause a system crash.
807 if (!UP_AND_RUNNING(p)) {
809 ifp->if_collisions++;
814 * Pad the frame to the minimum size allowed if told to.
815 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
816 * paragraph C.4.4.3.b. It can help to work around buggy
817 * bridges that violate paragraph C.4.4.3.a from the same
818 * document, i.e., fail to pad short frames after untagging.
819 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
820 * untagging it will produce a 62-byte frame, which is a runt
821 * and requires padding. There are VLAN-enabled network
822 * devices that just discard such runts instead or mishandle
826 static char pad[8]; /* just zeros */
829 for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
830 n > 0; n -= sizeof(pad))
831 if (!m_append(m, min(n, sizeof(pad)), pad))
835 if_printf(ifp, "cannot pad short frame\n");
843 * If underlying interface can do VLAN tag insertion itself,
844 * just pass the packet along. However, we need some way to
845 * tell the interface where the packet came from so that it
846 * knows how to find the VLAN tag to use, so we attach a
847 * packet tag that holds it.
849 if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
850 m->m_pkthdr.ether_vtag = ifv->ifv_tag;
851 m->m_flags |= M_VLANTAG;
853 m = ether_vlanencap(m, ifv->ifv_tag);
856 "unable to prepend VLAN header\n");
863 * Send it, precisely as ether_output() would have.
864 * We are already running at splimp.
866 IFQ_HANDOFF(p, m, error);
875 vlan_input(struct ifnet *ifp, struct mbuf *m)
877 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
881 KASSERT(trunk != NULL, ("%s: no trunk", __func__));
883 if (m->m_flags & M_VLANTAG) {
885 * Packet is tagged, but m contains a normal
886 * Ethernet frame; the tag is stored out-of-band.
888 tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
889 m->m_flags &= ~M_VLANTAG;
891 struct ether_vlan_header *evl;
894 * Packet is tagged in-band as specified by 802.1q.
896 switch (ifp->if_type) {
898 if (m->m_len < sizeof(*evl) &&
899 (m = m_pullup(m, sizeof(*evl))) == NULL) {
900 if_printf(ifp, "cannot pullup VLAN header\n");
903 evl = mtod(m, struct ether_vlan_header *);
904 tag = EVL_VLANOFTAG(ntohs(evl->evl_tag));
907 * Remove the 802.1q header by copying the Ethernet
908 * addresses over it and adjusting the beginning of
909 * the data in the mbuf. The encapsulated Ethernet
910 * type field is already in place.
912 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
913 ETHER_HDR_LEN - ETHER_TYPE_LEN);
914 m_adj(m, ETHER_VLAN_ENCAP_LEN);
919 panic("%s: %s has unsupported if_type %u",
920 __func__, ifp->if_xname, ifp->if_type);
930 ifv = trunk->vlans[tag];
932 ifv = vlan_gethash(trunk, tag);
934 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
935 TRUNK_RUNLOCK(trunk);
940 TRUNK_RUNLOCK(trunk);
942 m->m_pkthdr.rcvif = ifv->ifv_ifp;
943 ifv->ifv_ifp->if_ipackets++;
945 /* Pass it back through the parent's input routine. */
946 (*ifp->if_input)(ifv->ifv_ifp, m);
950 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag)
952 struct ifvlantrunk *trunk;
956 /* VID numbers 0x0 and 0xFFF are reserved */
957 if (tag == 0 || tag == 0xFFF)
959 if (p->if_type != IFT_ETHER)
960 return (EPROTONOSUPPORT);
961 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
962 return (EPROTONOSUPPORT);
966 if (p->if_vlantrunk == NULL) {
967 trunk = malloc(sizeof(struct ifvlantrunk),
968 M_VLAN, M_WAITOK | M_ZERO);
970 vlan_inithash(trunk);
973 if (p->if_vlantrunk != NULL) {
974 /* A race that that is very unlikely to be hit. */
976 vlan_freehash(trunk);
981 TRUNK_LOCK_INIT(trunk);
983 p->if_vlantrunk = trunk;
988 trunk = p->if_vlantrunk;
992 ifv->ifv_tag = tag; /* must set this before vlan_inshash() */
994 if (trunk->vlans[tag] != NULL) {
998 trunk->vlans[tag] = ifv;
1001 error = vlan_inshash(trunk, ifv);
1005 ifv->ifv_proto = ETHERTYPE_VLAN;
1006 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1007 ifv->ifv_mintu = ETHERMIN;
1008 ifv->ifv_pflags = 0;
1011 * If the parent supports the VLAN_MTU capability,
1012 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1015 if (p->if_capenable & IFCAP_VLAN_MTU) {
1017 * No need to fudge the MTU since the parent can
1018 * handle extended frames.
1020 ifv->ifv_mtufudge = 0;
1023 * Fudge the MTU by the encapsulation size. This
1024 * makes us incompatible with strictly compliant
1025 * 802.1Q implementations, but allows us to use
1026 * the feature with other NetBSD implementations,
1027 * which might still be useful.
1029 ifv->ifv_mtufudge = ifv->ifv_encaplen;
1032 ifv->ifv_trunk = trunk;
1034 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1035 ifp->if_baudrate = p->if_baudrate;
1037 * Copy only a selected subset of flags from the parent.
1038 * Other flags are none of our business.
1040 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1041 ifp->if_flags &= ~VLAN_COPY_FLAGS;
1042 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1043 #undef VLAN_COPY_FLAGS
1045 ifp->if_link_state = p->if_link_state;
1047 vlan_capabilities(ifv);
1050 * Set up our ``Ethernet address'' to reflect the underlying
1051 * physical interface's.
1053 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), ETHER_ADDR_LEN);
1056 * Configure multicast addresses that may already be
1057 * joined on the vlan device.
1059 (void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1061 /* We are ready for operation now. */
1062 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1064 TRUNK_UNLOCK(trunk);
1071 vlan_unconfig(struct ifnet *ifp)
1076 ret = vlan_unconfig_locked(ifp);
1082 vlan_unconfig_locked(struct ifnet *ifp)
1084 struct ifvlantrunk *trunk;
1085 struct vlan_mc_entry *mc;
1091 ifv = ifp->if_softc;
1092 trunk = ifv->ifv_trunk;
1095 struct sockaddr_dl sdl;
1096 struct ifnet *p = trunk->parent;
1101 * Since the interface is being unconfigured, we need to
1102 * empty the list of multicast groups that we may have joined
1103 * while we were alive from the parent's list.
1105 bzero((char *)&sdl, sizeof(sdl));
1106 sdl.sdl_len = sizeof(sdl);
1107 sdl.sdl_family = AF_LINK;
1108 sdl.sdl_index = p->if_index;
1109 sdl.sdl_type = IFT_ETHER;
1110 sdl.sdl_alen = ETHER_ADDR_LEN;
1112 while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1113 bcopy((char *)&mc->mc_addr, LLADDR(&sdl),
1115 error = if_delmulti(p, (struct sockaddr *)&sdl);
1118 SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1122 vlan_setflags(ifp, 0); /* clear special flags on parent */
1124 trunk->vlans[ifv->ifv_tag] = NULL;
1127 vlan_remhash(trunk, ifv);
1129 ifv->ifv_trunk = NULL;
1132 * Check if we were the last.
1134 if (trunk->refcnt == 0) {
1135 trunk->parent->if_vlantrunk = NULL;
1137 * XXXGL: If some ithread has already entered
1138 * vlan_input() and is now blocked on the trunk
1139 * lock, then it should preempt us right after
1140 * unlock and finish its work. Then we will acquire
1141 * lock again in trunk_destroy().
1143 TRUNK_UNLOCK(trunk);
1144 trunk_destroy(trunk);
1146 TRUNK_UNLOCK(trunk);
1149 /* Disconnect from parent. */
1150 if (ifv->ifv_pflags)
1151 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1152 ifp->if_mtu = ETHERMTU;
1153 ifp->if_link_state = LINK_STATE_UNKNOWN;
1154 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1159 /* Handle a reference counted flag that should be set on the parent as well */
1161 vlan_setflag(struct ifnet *ifp, int flag, int status,
1162 int (*func)(struct ifnet *, int))
1167 /* XXX VLAN_LOCK_ASSERT(); */
1169 ifv = ifp->if_softc;
1170 status = status ? (ifp->if_flags & flag) : 0;
1171 /* Now "status" contains the flag value or 0 */
1174 * See if recorded parent's status is different from what
1175 * we want it to be. If it is, flip it. We record parent's
1176 * status in ifv_pflags so that we won't clear parent's flag
1177 * we haven't set. In fact, we don't clear or set parent's
1178 * flags directly, but get or release references to them.
1179 * That's why we can be sure that recorded flags still are
1180 * in accord with actual parent's flags.
1182 if (status != (ifv->ifv_pflags & flag)) {
1183 error = (*func)(PARENT(ifv), status);
1186 ifv->ifv_pflags &= ~flag;
1187 ifv->ifv_pflags |= status;
1193 * Handle IFF_* flags that require certain changes on the parent:
1194 * if "status" is true, update parent's flags respective to our if_flags;
1195 * if "status" is false, forcedly clear the flags set on parent.
1198 vlan_setflags(struct ifnet *ifp, int status)
1202 for (i = 0; vlan_pflags[i].flag; i++) {
1203 error = vlan_setflag(ifp, vlan_pflags[i].flag,
1204 status, vlan_pflags[i].func);
1211 /* Inform all vlans that their parent has changed link state */
1213 vlan_link_state(struct ifnet *ifp, int link)
1215 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1221 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1222 if (trunk->vlans[i] != NULL) {
1223 ifv = trunk->vlans[i];
1225 for (i = 0; i < (1 << trunk->hwidth); i++)
1226 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1228 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1229 if_link_state_change(ifv->ifv_ifp,
1230 trunk->parent->if_link_state);
1232 TRUNK_UNLOCK(trunk);
1236 vlan_capabilities(struct ifvlan *ifv)
1238 struct ifnet *p = PARENT(ifv);
1239 struct ifnet *ifp = ifv->ifv_ifp;
1241 TRUNK_LOCK_ASSERT(TRUNK(ifv));
1244 * If the parent interface can do checksum offloading
1245 * on VLANs, then propagate its hardware-assisted
1246 * checksumming flags. Also assert that checksum
1247 * offloading requires hardware VLAN tagging.
1249 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1250 ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1252 if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1253 p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1254 ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1255 ifp->if_hwassist = p->if_hwassist;
1257 ifp->if_capenable = 0;
1258 ifp->if_hwassist = 0;
1263 vlan_trunk_capabilities(struct ifnet *ifp)
1265 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1271 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1272 if (trunk->vlans[i] != NULL) {
1273 ifv = trunk->vlans[i];
1275 for (i = 0; i < (1 << trunk->hwidth); i++) {
1276 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1278 vlan_capabilities(ifv);
1280 TRUNK_UNLOCK(trunk);
1284 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1293 ifr = (struct ifreq *)data;
1294 ifa = (struct ifaddr *)data;
1295 ifv = ifp->if_softc;
1300 if (TRUNK(ifv) != NULL) {
1301 error = (*PARENT(ifv)->if_ioctl)(PARENT(ifv),
1302 SIOCGIFMEDIA, data);
1304 /* Limit the result to the parent's current config. */
1306 struct ifmediareq *ifmr;
1308 ifmr = (struct ifmediareq *)data;
1309 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1310 ifmr->ifm_count = 1;
1311 error = copyout(&ifmr->ifm_current,
1328 * Set the interface MTU.
1331 if (TRUNK(ifv) != NULL) {
1333 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1335 (ifv->ifv_mintu - ifv->ifv_mtufudge))
1338 ifp->if_mtu = ifr->ifr_mtu;
1345 error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1348 if (vlr.vlr_parent[0] == '\0') {
1352 p = ifunit(vlr.vlr_parent);
1358 * Don't let the caller set up a VLAN tag with
1359 * anything except VLID bits.
1361 if (vlr.vlr_tag & ~EVL_VLID_MASK) {
1365 error = vlan_config(ifv, p, vlr.vlr_tag);
1369 /* Update flags on the parent, if necessary. */
1370 vlan_setflags(ifp, 1);
1374 bzero(&vlr, sizeof(vlr));
1376 if (TRUNK(ifv) != NULL) {
1377 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1378 sizeof(vlr.vlr_parent));
1379 vlr.vlr_tag = ifv->ifv_tag;
1382 error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1387 * We should propagate selected flags to the parent,
1388 * e.g., promiscuous mode.
1390 if (TRUNK(ifv) != NULL)
1391 error = vlan_setflags(ifp, 1);
1397 * If we don't have a parent, just remember the membership for
1400 if (TRUNK(ifv) != NULL)
1401 error = vlan_setmulti(ifp);
1405 error = ether_ioctl(ifp, cmd, data);