1 /*-
2 * Copyright 1998 Massachusetts Institute of Technology
3 *
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
14 * warranty.
15 *
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
27 * SUCH DAMAGE.
28 */
29
30 /*
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.
39 */
40
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43
44 #include "opt_inet.h"
45 #include "opt_vlan.h"
46
47 #include <sys/param.h>
48 #include <sys/eventhandler.h>
49 #include <sys/kernel.h>
50 #include <sys/lock.h>
51 #include <sys/malloc.h>
52 #include <sys/mbuf.h>
53 #include <sys/module.h>
54 #include <sys/rmlock.h>
55 #include <sys/queue.h>
56 #include <sys/socket.h>
57 #include <sys/sockio.h>
58 #include <sys/sysctl.h>
59 #include <sys/systm.h>
60 #include <sys/sx.h>
61
62 #include <net/bpf.h>
63 #include <net/ethernet.h>
64 #include <net/if.h>
65 #include <net/if_var.h>
66 #include <net/if_clone.h>
67 #include <net/if_dl.h>
68 #include <net/if_types.h>
69 #include <net/if_vlan_var.h>
70 #include <net/vnet.h>
71
72 #ifdef INET
73 #include <netinet/in.h>
74 #include <netinet/if_ether.h>
75 #endif
76
77 #define VLAN_DEF_HWIDTH 4
78 #define VLAN_IFFLAGS (IFF_BROADCAST | IFF_MULTICAST)
79
80 #define UP_AND_RUNNING(ifp) \
81 ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
82
83 LIST_HEAD(ifvlanhead, ifvlan);
84
85 struct ifvlantrunk {
86 struct ifnet *parent; /* parent interface of this trunk */
87 struct rmlock lock;
88 #ifdef VLAN_ARRAY
89 #define VLAN_ARRAY_SIZE (EVL_VLID_MASK + 1)
90 struct ifvlan *vlans[VLAN_ARRAY_SIZE]; /* static table */
91 #else
92 struct ifvlanhead *hash; /* dynamic hash-list table */
93 uint16_t hmask;
94 uint16_t hwidth;
95 #endif
96 int refcnt;
97 };
98
99 struct vlan_mc_entry {
100 struct sockaddr_dl mc_addr;
101 SLIST_ENTRY(vlan_mc_entry) mc_entries;
102 };
103
104 struct ifvlan {
105 struct ifvlantrunk *ifv_trunk;
106 struct ifnet *ifv_ifp;
107 #define TRUNK(ifv) ((ifv)->ifv_trunk)
108 #define PARENT(ifv) ((ifv)->ifv_trunk->parent)
109 void *ifv_cookie;
110 int ifv_pflags; /* special flags we have set on parent */
111 struct ifv_linkmib {
112 int ifvm_encaplen; /* encapsulation length */
113 int ifvm_mtufudge; /* MTU fudged by this much */
114 int ifvm_mintu; /* min transmission unit */
115 uint16_t ifvm_proto; /* encapsulation ethertype */
116 uint16_t ifvm_tag; /* tag to apply on packets leaving if */
117 } ifv_mib;
118 SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
119 #ifndef VLAN_ARRAY
120 LIST_ENTRY(ifvlan) ifv_list;
121 #endif
122 };
123 #define ifv_proto ifv_mib.ifvm_proto
124 #define ifv_vid ifv_mib.ifvm_tag
125 #define ifv_encaplen ifv_mib.ifvm_encaplen
126 #define ifv_mtufudge ifv_mib.ifvm_mtufudge
127 #define ifv_mintu ifv_mib.ifvm_mintu
128
129 /* Special flags we should propagate to parent. */
130 static struct {
131 int flag;
132 int (*func)(struct ifnet *, int);
133 } vlan_pflags[] = {
134 {IFF_PROMISC, ifpromisc},
135 {IFF_ALLMULTI, if_allmulti},
136 {0, NULL}
137 };
138
139 SYSCTL_DECL(_net_link);
140 static SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0,
141 "IEEE 802.1Q VLAN");
142 static SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0,
143 "for consistency");
144
145 static VNET_DEFINE(int, soft_pad);
146 #define V_soft_pad VNET(soft_pad)
147 SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW | CTLFLAG_VNET,
148 &VNET_NAME(soft_pad), 0, "pad short frames before tagging");
149
150 static const char vlanname[] = "vlan";
151 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
152
153 static eventhandler_tag ifdetach_tag;
154 static eventhandler_tag iflladdr_tag;
155
156 /*
157 * We have a global mutex, that is used to serialize configuration
158 * changes and isn't used in normal packet delivery.
159 *
160 * We also have a per-trunk rmlock(9), that is locked shared on packet
161 * processing and exclusive when configuration is changed.
162 *
163 * The VLAN_ARRAY substitutes the dynamic hash with a static array
164 * with 4096 entries. In theory this can give a boost in processing,
165 * however on practice it does not. Probably this is because array
166 * is too big to fit into CPU cache.
167 */
168 static struct sx ifv_lock;
169 #define VLAN_LOCK_INIT() sx_init(&ifv_lock, "vlan_global")
170 #define VLAN_LOCK_DESTROY() sx_destroy(&ifv_lock)
171 #define VLAN_LOCK_ASSERT() sx_assert(&ifv_lock, SA_LOCKED)
172 #define VLAN_LOCK() sx_xlock(&ifv_lock)
173 #define VLAN_UNLOCK() sx_xunlock(&ifv_lock)
174 #define TRUNK_LOCK_INIT(trunk) rm_init(&(trunk)->lock, vlanname)
175 #define TRUNK_LOCK_DESTROY(trunk) rm_destroy(&(trunk)->lock)
176 #define TRUNK_LOCK(trunk) rm_wlock(&(trunk)->lock)
177 #define TRUNK_UNLOCK(trunk) rm_wunlock(&(trunk)->lock)
178 #define TRUNK_LOCK_ASSERT(trunk) rm_assert(&(trunk)->lock, RA_WLOCKED)
179 #define TRUNK_RLOCK(trunk) rm_rlock(&(trunk)->lock, &tracker)
180 #define TRUNK_RUNLOCK(trunk) rm_runlock(&(trunk)->lock, &tracker)
181 #define TRUNK_LOCK_RASSERT(trunk) rm_assert(&(trunk)->lock, RA_RLOCKED)
182 #define TRUNK_LOCK_READER struct rm_priotracker tracker
183
184 #ifndef VLAN_ARRAY
185 static void vlan_inithash(struct ifvlantrunk *trunk);
186 static void vlan_freehash(struct ifvlantrunk *trunk);
187 static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
188 static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
189 static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
190 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
191 uint16_t vid);
192 #endif
193 static void trunk_destroy(struct ifvlantrunk *trunk);
194
195 static void vlan_init(void *foo);
196 static void vlan_input(struct ifnet *ifp, struct mbuf *m);
197 static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
198 static void vlan_qflush(struct ifnet *ifp);
199 static int vlan_setflag(struct ifnet *ifp, int flag, int status,
200 int (*func)(struct ifnet *, int));
201 static int vlan_setflags(struct ifnet *ifp, int status);
202 static int vlan_setmulti(struct ifnet *ifp);
203 static int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
204 static void vlan_unconfig(struct ifnet *ifp);
205 static void vlan_unconfig_locked(struct ifnet *ifp, int departing);
206 static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
207 static void vlan_link_state(struct ifnet *ifp);
208 static void vlan_capabilities(struct ifvlan *ifv);
209 static void vlan_trunk_capabilities(struct ifnet *ifp);
210
211 static struct ifnet *vlan_clone_match_ethervid(const char *, int *);
212 static int vlan_clone_match(struct if_clone *, const char *);
213 static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
214 static int vlan_clone_destroy(struct if_clone *, struct ifnet *);
215
216 static void vlan_ifdetach(void *arg, struct ifnet *ifp);
217 static void vlan_iflladdr(void *arg, struct ifnet *ifp);
218
219 static struct if_clone *vlan_cloner;
220
221 #ifdef VIMAGE
222 static VNET_DEFINE(struct if_clone *, vlan_cloner);
223 #define V_vlan_cloner VNET(vlan_cloner)
224 #endif
225
226 #ifndef VLAN_ARRAY
227 #define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
228
229 static void
vlan_inithash(struct ifvlantrunk * trunk)230 vlan_inithash(struct ifvlantrunk *trunk)
231 {
232 int i, n;
233
234 /*
235 * The trunk must not be locked here since we call malloc(M_WAITOK).
236 * It is OK in case this function is called before the trunk struct
237 * gets hooked up and becomes visible from other threads.
238 */
239
240 KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
241 ("%s: hash already initialized", __func__));
242
243 trunk->hwidth = VLAN_DEF_HWIDTH;
244 n = 1 << trunk->hwidth;
245 trunk->hmask = n - 1;
246 trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
247 for (i = 0; i < n; i++)
248 LIST_INIT(&trunk->hash[i]);
249 }
250
251 static void
vlan_freehash(struct ifvlantrunk * trunk)252 vlan_freehash(struct ifvlantrunk *trunk)
253 {
254 #ifdef INVARIANTS
255 int i;
256
257 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
258 for (i = 0; i < (1 << trunk->hwidth); i++)
259 KASSERT(LIST_EMPTY(&trunk->hash[i]),
260 ("%s: hash table not empty", __func__));
261 #endif
262 free(trunk->hash, M_VLAN);
263 trunk->hash = NULL;
264 trunk->hwidth = trunk->hmask = 0;
265 }
266
267 static int
vlan_inshash(struct ifvlantrunk * trunk,struct ifvlan * ifv)268 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
269 {
270 int i, b;
271 struct ifvlan *ifv2;
272
273 TRUNK_LOCK_ASSERT(trunk);
274 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
275
276 b = 1 << trunk->hwidth;
277 i = HASH(ifv->ifv_vid, trunk->hmask);
278 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
279 if (ifv->ifv_vid == ifv2->ifv_vid)
280 return (EEXIST);
281
282 /*
283 * Grow the hash when the number of vlans exceeds half of the number of
284 * hash buckets squared. This will make the average linked-list length
285 * buckets/2.
286 */
287 if (trunk->refcnt > (b * b) / 2) {
288 vlan_growhash(trunk, 1);
289 i = HASH(ifv->ifv_vid, trunk->hmask);
290 }
291 LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
292 trunk->refcnt++;
293
294 return (0);
295 }
296
297 static int
vlan_remhash(struct ifvlantrunk * trunk,struct ifvlan * ifv)298 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
299 {
300 int i, b;
301 struct ifvlan *ifv2;
302
303 TRUNK_LOCK_ASSERT(trunk);
304 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
305
306 b = 1 << trunk->hwidth;
307 i = HASH(ifv->ifv_vid, trunk->hmask);
308 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
309 if (ifv2 == ifv) {
310 trunk->refcnt--;
311 LIST_REMOVE(ifv2, ifv_list);
312 if (trunk->refcnt < (b * b) / 2)
313 vlan_growhash(trunk, -1);
314 return (0);
315 }
316
317 panic("%s: vlan not found\n", __func__);
318 return (ENOENT); /*NOTREACHED*/
319 }
320
321 /*
322 * Grow the hash larger or smaller if memory permits.
323 */
324 static void
vlan_growhash(struct ifvlantrunk * trunk,int howmuch)325 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
326 {
327 struct ifvlan *ifv;
328 struct ifvlanhead *hash2;
329 int hwidth2, i, j, n, n2;
330
331 TRUNK_LOCK_ASSERT(trunk);
332 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
333
334 if (howmuch == 0) {
335 /* Harmless yet obvious coding error */
336 printf("%s: howmuch is 0\n", __func__);
337 return;
338 }
339
340 hwidth2 = trunk->hwidth + howmuch;
341 n = 1 << trunk->hwidth;
342 n2 = 1 << hwidth2;
343 /* Do not shrink the table below the default */
344 if (hwidth2 < VLAN_DEF_HWIDTH)
345 return;
346
347 /* M_NOWAIT because we're called with trunk mutex held */
348 hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
349 if (hash2 == NULL) {
350 printf("%s: out of memory -- hash size not changed\n",
351 __func__);
352 return; /* We can live with the old hash table */
353 }
354 for (j = 0; j < n2; j++)
355 LIST_INIT(&hash2[j]);
356 for (i = 0; i < n; i++)
357 while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
358 LIST_REMOVE(ifv, ifv_list);
359 j = HASH(ifv->ifv_vid, n2 - 1);
360 LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
361 }
362 free(trunk->hash, M_VLAN);
363 trunk->hash = hash2;
364 trunk->hwidth = hwidth2;
365 trunk->hmask = n2 - 1;
366
367 if (bootverbose)
368 if_printf(trunk->parent,
369 "VLAN hash table resized from %d to %d buckets\n", n, n2);
370 }
371
372 static __inline struct ifvlan *
vlan_gethash(struct ifvlantrunk * trunk,uint16_t vid)373 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
374 {
375 struct ifvlan *ifv;
376
377 TRUNK_LOCK_RASSERT(trunk);
378
379 LIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
380 if (ifv->ifv_vid == vid)
381 return (ifv);
382 return (NULL);
383 }
384
385 #if 0
386 /* Debugging code to view the hashtables. */
387 static void
388 vlan_dumphash(struct ifvlantrunk *trunk)
389 {
390 int i;
391 struct ifvlan *ifv;
392
393 for (i = 0; i < (1 << trunk->hwidth); i++) {
394 printf("%d: ", i);
395 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
396 printf("%s ", ifv->ifv_ifp->if_xname);
397 printf("\n");
398 }
399 }
400 #endif /* 0 */
401 #else
402
403 static __inline struct ifvlan *
vlan_gethash(struct ifvlantrunk * trunk,uint16_t vid)404 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
405 {
406
407 return trunk->vlans[vid];
408 }
409
410 static __inline int
vlan_inshash(struct ifvlantrunk * trunk,struct ifvlan * ifv)411 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
412 {
413
414 if (trunk->vlans[ifv->ifv_vid] != NULL)
415 return EEXIST;
416 trunk->vlans[ifv->ifv_vid] = ifv;
417 trunk->refcnt++;
418
419 return (0);
420 }
421
422 static __inline int
vlan_remhash(struct ifvlantrunk * trunk,struct ifvlan * ifv)423 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
424 {
425
426 trunk->vlans[ifv->ifv_vid] = NULL;
427 trunk->refcnt--;
428
429 return (0);
430 }
431
432 static __inline void
vlan_freehash(struct ifvlantrunk * trunk)433 vlan_freehash(struct ifvlantrunk *trunk)
434 {
435 }
436
437 static __inline void
vlan_inithash(struct ifvlantrunk * trunk)438 vlan_inithash(struct ifvlantrunk *trunk)
439 {
440 }
441
442 #endif /* !VLAN_ARRAY */
443
444 static void
trunk_destroy(struct ifvlantrunk * trunk)445 trunk_destroy(struct ifvlantrunk *trunk)
446 {
447 VLAN_LOCK_ASSERT();
448
449 TRUNK_LOCK(trunk);
450 vlan_freehash(trunk);
451 trunk->parent->if_vlantrunk = NULL;
452 TRUNK_UNLOCK(trunk);
453 TRUNK_LOCK_DESTROY(trunk);
454 free(trunk, M_VLAN);
455 }
456
457 /*
458 * Program our multicast filter. What we're actually doing is
459 * programming the multicast filter of the parent. This has the
460 * side effect of causing the parent interface to receive multicast
461 * traffic that it doesn't really want, which ends up being discarded
462 * later by the upper protocol layers. Unfortunately, there's no way
463 * to avoid this: there really is only one physical interface.
464 */
465 static int
vlan_setmulti(struct ifnet * ifp)466 vlan_setmulti(struct ifnet *ifp)
467 {
468 struct ifnet *ifp_p;
469 struct ifmultiaddr *ifma;
470 struct ifvlan *sc;
471 struct vlan_mc_entry *mc;
472 int error;
473
474 /* Find the parent. */
475 sc = ifp->if_softc;
476 TRUNK_LOCK_ASSERT(TRUNK(sc));
477 ifp_p = PARENT(sc);
478
479 CURVNET_SET_QUIET(ifp_p->if_vnet);
480
481 /* First, remove any existing filter entries. */
482 while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
483 SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
484 (void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
485 free(mc, M_VLAN);
486 }
487
488 /* Now program new ones. */
489 IF_ADDR_WLOCK(ifp);
490 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
491 if (ifma->ifma_addr->sa_family != AF_LINK)
492 continue;
493 mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
494 if (mc == NULL) {
495 IF_ADDR_WUNLOCK(ifp);
496 return (ENOMEM);
497 }
498 bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
499 mc->mc_addr.sdl_index = ifp_p->if_index;
500 SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
501 }
502 IF_ADDR_WUNLOCK(ifp);
503 SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
504 error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
505 NULL);
506 if (error)
507 return (error);
508 }
509
510 CURVNET_RESTORE();
511 return (0);
512 }
513
514 /*
515 * A handler for parent interface link layer address changes.
516 * If the parent interface link layer address is changed we
517 * should also change it on all children vlans.
518 */
519 static void
vlan_iflladdr(void * arg __unused,struct ifnet * ifp)520 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
521 {
522 struct ifvlan *ifv;
523 #ifndef VLAN_ARRAY
524 struct ifvlan *next;
525 #endif
526 int i;
527
528 /*
529 * Check if it's a trunk interface first of all
530 * to avoid needless locking.
531 */
532 if (ifp->if_vlantrunk == NULL)
533 return;
534
535 VLAN_LOCK();
536 /*
537 * OK, it's a trunk. Loop over and change all vlan's lladdrs on it.
538 */
539 #ifdef VLAN_ARRAY
540 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
541 if ((ifv = ifp->if_vlantrunk->vlans[i])) {
542 #else /* VLAN_ARRAY */
543 for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
544 LIST_FOREACH_SAFE(ifv, &ifp->if_vlantrunk->hash[i], ifv_list, next) {
545 #endif /* VLAN_ARRAY */
546 VLAN_UNLOCK();
547 if_setlladdr(ifv->ifv_ifp, IF_LLADDR(ifp),
548 ifp->if_addrlen);
549 VLAN_LOCK();
550 }
551 VLAN_UNLOCK();
552
553 }
554
555 /*
556 * A handler for network interface departure events.
557 * Track departure of trunks here so that we don't access invalid
558 * pointers or whatever if a trunk is ripped from under us, e.g.,
559 * by ejecting its hot-plug card. However, if an ifnet is simply
560 * being renamed, then there's no need to tear down the state.
561 */
562 static void
563 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
564 {
565 struct ifvlan *ifv;
566 int i;
567
568 /*
569 * Check if it's a trunk interface first of all
570 * to avoid needless locking.
571 */
572 if (ifp->if_vlantrunk == NULL)
573 return;
574
575 /* If the ifnet is just being renamed, don't do anything. */
576 if (ifp->if_flags & IFF_RENAMING)
577 return;
578
579 VLAN_LOCK();
580 /*
581 * OK, it's a trunk. Loop over and detach all vlan's on it.
582 * Check trunk pointer after each vlan_unconfig() as it will
583 * free it and set to NULL after the last vlan was detached.
584 */
585 #ifdef VLAN_ARRAY
586 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
587 if ((ifv = ifp->if_vlantrunk->vlans[i])) {
588 vlan_unconfig_locked(ifv->ifv_ifp, 1);
589 if (ifp->if_vlantrunk == NULL)
590 break;
591 }
592 #else /* VLAN_ARRAY */
593 restart:
594 for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
595 if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
596 vlan_unconfig_locked(ifv->ifv_ifp, 1);
597 if (ifp->if_vlantrunk)
598 goto restart; /* trunk->hwidth can change */
599 else
600 break;
601 }
602 #endif /* VLAN_ARRAY */
603 /* Trunk should have been destroyed in vlan_unconfig(). */
604 KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
605 VLAN_UNLOCK();
606 }
607
608 /*
609 * Return the trunk device for a virtual interface.
610 */
611 static struct ifnet *
612 vlan_trunkdev(struct ifnet *ifp)
613 {
614 struct ifvlan *ifv;
615
616 if (ifp->if_type != IFT_L2VLAN)
617 return (NULL);
618 ifv = ifp->if_softc;
619 ifp = NULL;
620 VLAN_LOCK();
621 if (ifv->ifv_trunk)
622 ifp = PARENT(ifv);
623 VLAN_UNLOCK();
624 return (ifp);
625 }
626
627 /*
628 * Return the 12-bit VLAN VID for this interface, for use by external
629 * components such as Infiniband.
630 *
631 * XXXRW: Note that the function name here is historical; it should be named
632 * vlan_vid().
633 */
634 static int
635 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
636 {
637 struct ifvlan *ifv;
638
639 if (ifp->if_type != IFT_L2VLAN)
640 return (EINVAL);
641 ifv = ifp->if_softc;
642 *vidp = ifv->ifv_vid;
643 return (0);
644 }
645
646 /*
647 * Return a driver specific cookie for this interface. Synchronization
648 * with setcookie must be provided by the driver.
649 */
650 static void *
651 vlan_cookie(struct ifnet *ifp)
652 {
653 struct ifvlan *ifv;
654
655 if (ifp->if_type != IFT_L2VLAN)
656 return (NULL);
657 ifv = ifp->if_softc;
658 return (ifv->ifv_cookie);
659 }
660
661 /*
662 * Store a cookie in our softc that drivers can use to store driver
663 * private per-instance data in.
664 */
665 static int
666 vlan_setcookie(struct ifnet *ifp, void *cookie)
667 {
668 struct ifvlan *ifv;
669
670 if (ifp->if_type != IFT_L2VLAN)
671 return (EINVAL);
672 ifv = ifp->if_softc;
673 ifv->ifv_cookie = cookie;
674 return (0);
675 }
676
677 /*
678 * Return the vlan device present at the specific VID.
679 */
680 static struct ifnet *
681 vlan_devat(struct ifnet *ifp, uint16_t vid)
682 {
683 struct ifvlantrunk *trunk;
684 struct ifvlan *ifv;
685 TRUNK_LOCK_READER;
686
687 trunk = ifp->if_vlantrunk;
688 if (trunk == NULL)
689 return (NULL);
690 ifp = NULL;
691 TRUNK_RLOCK(trunk);
692 ifv = vlan_gethash(trunk, vid);
693 if (ifv)
694 ifp = ifv->ifv_ifp;
695 TRUNK_RUNLOCK(trunk);
696 return (ifp);
697 }
698
699 /*
700 * VLAN support can be loaded as a module. The only place in the
701 * system that's intimately aware of this is ether_input. We hook
702 * into this code through vlan_input_p which is defined there and
703 * set here. Noone else in the system should be aware of this so
704 * we use an explicit reference here.
705 */
706 extern void (*vlan_input_p)(struct ifnet *, struct mbuf *);
707
708 /* For if_link_state_change() eyes only... */
709 extern void (*vlan_link_state_p)(struct ifnet *);
710
711 static int
712 vlan_modevent(module_t mod, int type, void *data)
713 {
714
715 switch (type) {
716 case MOD_LOAD:
717 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
718 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
719 if (ifdetach_tag == NULL)
720 return (ENOMEM);
721 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
722 vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
723 if (iflladdr_tag == NULL)
724 return (ENOMEM);
725 VLAN_LOCK_INIT();
726 vlan_input_p = vlan_input;
727 vlan_link_state_p = vlan_link_state;
728 vlan_trunk_cap_p = vlan_trunk_capabilities;
729 vlan_trunkdev_p = vlan_trunkdev;
730 vlan_cookie_p = vlan_cookie;
731 vlan_setcookie_p = vlan_setcookie;
732 vlan_tag_p = vlan_tag;
733 vlan_devat_p = vlan_devat;
734 #ifndef VIMAGE
735 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
736 vlan_clone_create, vlan_clone_destroy);
737 #endif
738 if (bootverbose)
739 printf("vlan: initialized, using "
740 #ifdef VLAN_ARRAY
741 "full-size arrays"
742 #else
743 "hash tables with chaining"
744 #endif
745
746 "\n");
747 break;
748 case MOD_UNLOAD:
749 #ifndef VIMAGE
750 if_clone_detach(vlan_cloner);
751 #endif
752 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
753 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
754 vlan_input_p = NULL;
755 vlan_link_state_p = NULL;
756 vlan_trunk_cap_p = NULL;
757 vlan_trunkdev_p = NULL;
758 vlan_tag_p = NULL;
759 vlan_cookie_p = NULL;
760 vlan_setcookie_p = NULL;
761 vlan_devat_p = NULL;
762 VLAN_LOCK_DESTROY();
763 if (bootverbose)
764 printf("vlan: unloaded\n");
765 break;
766 default:
767 return (EOPNOTSUPP);
768 }
769 return (0);
770 }
771
772 static moduledata_t vlan_mod = {
773 "if_vlan",
774 vlan_modevent,
775 0
776 };
777
778 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
779 MODULE_VERSION(if_vlan, 3);
780
781 #ifdef VIMAGE
782 static void
783 vnet_vlan_init(const void *unused __unused)
784 {
785
786 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
787 vlan_clone_create, vlan_clone_destroy);
788 V_vlan_cloner = vlan_cloner;
789 }
790 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
791 vnet_vlan_init, NULL);
792
793 static void
794 vnet_vlan_uninit(const void *unused __unused)
795 {
796
797 if_clone_detach(V_vlan_cloner);
798 }
799 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
800 vnet_vlan_uninit, NULL);
801 #endif
802
803 /*
804 * Check for <etherif>.<vlan> style interface names.
805 */
806 static struct ifnet *
807 vlan_clone_match_ethervid(const char *name, int *vidp)
808 {
809 char ifname[IFNAMSIZ];
810 char *cp;
811 struct ifnet *ifp;
812 int vid;
813
814 strlcpy(ifname, name, IFNAMSIZ);
815 if ((cp = strchr(ifname, '.')) == NULL)
816 return (NULL);
817 *cp = '\0';
818 if ((ifp = ifunit(ifname)) == NULL)
819 return (NULL);
820 /* Parse VID. */
821 if (*++cp == '\0')
822 return (NULL);
823 vid = 0;
824 for(; *cp >= '0' && *cp <= '9'; cp++)
825 vid = (vid * 10) + (*cp - '0');
826 if (*cp != '\0')
827 return (NULL);
828 if (vidp != NULL)
829 *vidp = vid;
830
831 return (ifp);
832 }
833
834 static int
835 vlan_clone_match(struct if_clone *ifc, const char *name)
836 {
837 const char *cp;
838
839 if (vlan_clone_match_ethervid(name, NULL) != NULL)
840 return (1);
841
842 if (strncmp(vlanname, name, strlen(vlanname)) != 0)
843 return (0);
844 for (cp = name + 4; *cp != '\0'; cp++) {
845 if (*cp < '0' || *cp > '9')
846 return (0);
847 }
848
849 return (1);
850 }
851
852 static int
853 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
854 {
855 char *dp;
856 int wildcard;
857 int unit;
858 int error;
859 int vid;
860 int ethertag;
861 struct ifvlan *ifv;
862 struct ifnet *ifp;
863 struct ifnet *p;
864 struct ifaddr *ifa;
865 struct sockaddr_dl *sdl;
866 struct vlanreq vlr;
867 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
868
869 /*
870 * There are 3 (ugh) ways to specify the cloned device:
871 * o pass a parameter block with the clone request.
872 * o specify parameters in the text of the clone device name
873 * o specify no parameters and get an unattached device that
874 * must be configured separately.
875 * The first technique is preferred; the latter two are
876 * supported for backwards compatibilty.
877 *
878 * XXXRW: Note historic use of the word "tag" here. New ioctls may be
879 * called for.
880 */
881 if (params) {
882 error = copyin(params, &vlr, sizeof(vlr));
883 if (error)
884 return error;
885 p = ifunit(vlr.vlr_parent);
886 if (p == NULL)
887 return (ENXIO);
888 error = ifc_name2unit(name, &unit);
889 if (error != 0)
890 return (error);
891
892 ethertag = 1;
893 vid = vlr.vlr_tag;
894 wildcard = (unit < 0);
895 } else if ((p = vlan_clone_match_ethervid(name, &vid)) != NULL) {
896 ethertag = 1;
897 unit = -1;
898 wildcard = 0;
899 } else {
900 ethertag = 0;
901
902 error = ifc_name2unit(name, &unit);
903 if (error != 0)
904 return (error);
905
906 wildcard = (unit < 0);
907 }
908
909 error = ifc_alloc_unit(ifc, &unit);
910 if (error != 0)
911 return (error);
912
913 /* In the wildcard case, we need to update the name. */
914 if (wildcard) {
915 for (dp = name; *dp != '\0'; dp++);
916 if (snprintf(dp, len - (dp-name), "%d", unit) >
917 len - (dp-name) - 1) {
918 panic("%s: interface name too long", __func__);
919 }
920 }
921
922 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
923 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
924 if (ifp == NULL) {
925 ifc_free_unit(ifc, unit);
926 free(ifv, M_VLAN);
927 return (ENOSPC);
928 }
929 SLIST_INIT(&ifv->vlan_mc_listhead);
930 ifp->if_softc = ifv;
931 /*
932 * Set the name manually rather than using if_initname because
933 * we don't conform to the default naming convention for interfaces.
934 */
935 strlcpy(ifp->if_xname, name, IFNAMSIZ);
936 ifp->if_dname = vlanname;
937 ifp->if_dunit = unit;
938 /* NB: flags are not set here */
939 ifp->if_linkmib = &ifv->ifv_mib;
940 ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
941 /* NB: mtu is not set here */
942
943 ifp->if_init = vlan_init;
944 ifp->if_transmit = vlan_transmit;
945 ifp->if_qflush = vlan_qflush;
946 ifp->if_ioctl = vlan_ioctl;
947 ifp->if_flags = VLAN_IFFLAGS;
948 ether_ifattach(ifp, eaddr);
949 /* Now undo some of the damage... */
950 ifp->if_baudrate = 0;
951 ifp->if_type = IFT_L2VLAN;
952 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
953 ifa = ifp->if_addr;
954 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
955 sdl->sdl_type = IFT_L2VLAN;
956
957 if (ethertag) {
958 error = vlan_config(ifv, p, vid);
959 if (error != 0) {
960 /*
961 * Since we've partially failed, we need to back
962 * out all the way, otherwise userland could get
963 * confused. Thus, we destroy the interface.
964 */
965 ether_ifdetach(ifp);
966 vlan_unconfig(ifp);
967 if_free(ifp);
968 ifc_free_unit(ifc, unit);
969 free(ifv, M_VLAN);
970
971 return (error);
972 }
973
974 /* Update flags on the parent, if necessary. */
975 vlan_setflags(ifp, 1);
976 }
977
978 return (0);
979 }
980
981 static int
982 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
983 {
984 struct ifvlan *ifv = ifp->if_softc;
985 int unit = ifp->if_dunit;
986
987 ether_ifdetach(ifp); /* first, remove it from system-wide lists */
988 vlan_unconfig(ifp); /* now it can be unconfigured and freed */
989 if_free(ifp);
990 free(ifv, M_VLAN);
991 ifc_free_unit(ifc, unit);
992
993 return (0);
994 }
995
996 /*
997 * The ifp->if_init entry point for vlan(4) is a no-op.
998 */
999 static void
1000 vlan_init(void *foo __unused)
1001 {
1002 }
1003
1004 /*
1005 * The if_transmit method for vlan(4) interface.
1006 */
1007 static int
1008 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1009 {
1010 struct ifvlan *ifv;
1011 struct ifnet *p;
1012 int error, len, mcast;
1013
1014 ifv = ifp->if_softc;
1015 p = PARENT(ifv);
1016 len = m->m_pkthdr.len;
1017 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1018
1019 BPF_MTAP(ifp, m);
1020
1021 /*
1022 * Do not run parent's if_transmit() if the parent is not up,
1023 * or parent's driver will cause a system crash.
1024 */
1025 if (!UP_AND_RUNNING(p)) {
1026 m_freem(m);
1027 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1028 return (ENETDOWN);
1029 }
1030
1031 /*
1032 * Pad the frame to the minimum size allowed if told to.
1033 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
1034 * paragraph C.4.4.3.b. It can help to work around buggy
1035 * bridges that violate paragraph C.4.4.3.a from the same
1036 * document, i.e., fail to pad short frames after untagging.
1037 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
1038 * untagging it will produce a 62-byte frame, which is a runt
1039 * and requires padding. There are VLAN-enabled network
1040 * devices that just discard such runts instead or mishandle
1041 * them somehow.
1042 */
1043 if (V_soft_pad && p->if_type == IFT_ETHER) {
1044 static char pad[8]; /* just zeros */
1045 int n;
1046
1047 for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
1048 n > 0; n -= sizeof(pad))
1049 if (!m_append(m, min(n, sizeof(pad)), pad))
1050 break;
1051
1052 if (n > 0) {
1053 if_printf(ifp, "cannot pad short frame\n");
1054 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1055 m_freem(m);
1056 return (0);
1057 }
1058 }
1059
1060 /*
1061 * If underlying interface can do VLAN tag insertion itself,
1062 * just pass the packet along. However, we need some way to
1063 * tell the interface where the packet came from so that it
1064 * knows how to find the VLAN tag to use, so we attach a
1065 * packet tag that holds it.
1066 */
1067 if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1068 m->m_pkthdr.ether_vtag = ifv->ifv_vid;
1069 m->m_flags |= M_VLANTAG;
1070 } else {
1071 m = ether_vlanencap(m, ifv->ifv_vid);
1072 if (m == NULL) {
1073 if_printf(ifp, "unable to prepend VLAN header\n");
1074 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1075 return (0);
1076 }
1077 }
1078
1079 /*
1080 * Send it, precisely as ether_output() would have.
1081 */
1082 error = (p->if_transmit)(p, m);
1083 if (error == 0) {
1084 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1085 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1086 if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1087 } else
1088 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1089 return (error);
1090 }
1091
1092 /*
1093 * The ifp->if_qflush entry point for vlan(4) is a no-op.
1094 */
1095 static void
1096 vlan_qflush(struct ifnet *ifp __unused)
1097 {
1098 }
1099
1100 static void
1101 vlan_input(struct ifnet *ifp, struct mbuf *m)
1102 {
1103 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1104 struct ifvlan *ifv;
1105 TRUNK_LOCK_READER;
1106 uint16_t vid;
1107
1108 KASSERT(trunk != NULL, ("%s: no trunk", __func__));
1109
1110 if (m->m_flags & M_VLANTAG) {
1111 /*
1112 * Packet is tagged, but m contains a normal
1113 * Ethernet frame; the tag is stored out-of-band.
1114 */
1115 vid = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
1116 m->m_flags &= ~M_VLANTAG;
1117 } else {
1118 struct ether_vlan_header *evl;
1119
1120 /*
1121 * Packet is tagged in-band as specified by 802.1q.
1122 */
1123 switch (ifp->if_type) {
1124 case IFT_ETHER:
1125 if (m->m_len < sizeof(*evl) &&
1126 (m = m_pullup(m, sizeof(*evl))) == NULL) {
1127 if_printf(ifp, "cannot pullup VLAN header\n");
1128 return;
1129 }
1130 evl = mtod(m, struct ether_vlan_header *);
1131 vid = EVL_VLANOFTAG(ntohs(evl->evl_tag));
1132
1133 /*
1134 * Remove the 802.1q header by copying the Ethernet
1135 * addresses over it and adjusting the beginning of
1136 * the data in the mbuf. The encapsulated Ethernet
1137 * type field is already in place.
1138 */
1139 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1140 ETHER_HDR_LEN - ETHER_TYPE_LEN);
1141 m_adj(m, ETHER_VLAN_ENCAP_LEN);
1142 break;
1143
1144 default:
1145 #ifdef INVARIANTS
1146 panic("%s: %s has unsupported if_type %u",
1147 __func__, ifp->if_xname, ifp->if_type);
1148 #endif
1149 m_freem(m);
1150 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1151 return;
1152 }
1153 }
1154
1155 TRUNK_RLOCK(trunk);
1156 ifv = vlan_gethash(trunk, vid);
1157 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1158 TRUNK_RUNLOCK(trunk);
1159 m_freem(m);
1160 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1161 return;
1162 }
1163 TRUNK_RUNLOCK(trunk);
1164
1165 m->m_pkthdr.rcvif = ifv->ifv_ifp;
1166 if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1167
1168 /* Pass it back through the parent's input routine. */
1169 (*ifp->if_input)(ifv->ifv_ifp, m);
1170 }
1171
1172 static int
1173 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid)
1174 {
1175 struct ifvlantrunk *trunk;
1176 struct ifnet *ifp;
1177 int error = 0;
1178
1179 /*
1180 * We can handle non-ethernet hardware types as long as
1181 * they handle the tagging and headers themselves.
1182 */
1183 if (p->if_type != IFT_ETHER &&
1184 (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1185 return (EPROTONOSUPPORT);
1186 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1187 return (EPROTONOSUPPORT);
1188 /*
1189 * Don't let the caller set up a VLAN VID with
1190 * anything except VLID bits.
1191 * VID numbers 0x0 and 0xFFF are reserved.
1192 */
1193 if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1194 return (EINVAL);
1195 if (ifv->ifv_trunk)
1196 return (EBUSY);
1197
1198 if (p->if_vlantrunk == NULL) {
1199 trunk = malloc(sizeof(struct ifvlantrunk),
1200 M_VLAN, M_WAITOK | M_ZERO);
1201 vlan_inithash(trunk);
1202 VLAN_LOCK();
1203 if (p->if_vlantrunk != NULL) {
1204 /* A race that that is very unlikely to be hit. */
1205 vlan_freehash(trunk);
1206 free(trunk, M_VLAN);
1207 goto exists;
1208 }
1209 TRUNK_LOCK_INIT(trunk);
1210 TRUNK_LOCK(trunk);
1211 p->if_vlantrunk = trunk;
1212 trunk->parent = p;
1213 } else {
1214 VLAN_LOCK();
1215 exists:
1216 trunk = p->if_vlantrunk;
1217 TRUNK_LOCK(trunk);
1218 }
1219
1220 ifv->ifv_vid = vid; /* must set this before vlan_inshash() */
1221 error = vlan_inshash(trunk, ifv);
1222 if (error)
1223 goto done;
1224 ifv->ifv_proto = ETHERTYPE_VLAN;
1225 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1226 ifv->ifv_mintu = ETHERMIN;
1227 ifv->ifv_pflags = 0;
1228
1229 /*
1230 * If the parent supports the VLAN_MTU capability,
1231 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1232 * use it.
1233 */
1234 if (p->if_capenable & IFCAP_VLAN_MTU) {
1235 /*
1236 * No need to fudge the MTU since the parent can
1237 * handle extended frames.
1238 */
1239 ifv->ifv_mtufudge = 0;
1240 } else {
1241 /*
1242 * Fudge the MTU by the encapsulation size. This
1243 * makes us incompatible with strictly compliant
1244 * 802.1Q implementations, but allows us to use
1245 * the feature with other NetBSD implementations,
1246 * which might still be useful.
1247 */
1248 ifv->ifv_mtufudge = ifv->ifv_encaplen;
1249 }
1250
1251 ifv->ifv_trunk = trunk;
1252 ifp = ifv->ifv_ifp;
1253 /*
1254 * Initialize fields from our parent. This duplicates some
1255 * work with ether_ifattach() but allows for non-ethernet
1256 * interfaces to also work.
1257 */
1258 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1259 ifp->if_baudrate = p->if_baudrate;
1260 ifp->if_output = p->if_output;
1261 ifp->if_input = p->if_input;
1262 ifp->if_resolvemulti = p->if_resolvemulti;
1263 ifp->if_addrlen = p->if_addrlen;
1264 ifp->if_broadcastaddr = p->if_broadcastaddr;
1265
1266 /*
1267 * Copy only a selected subset of flags from the parent.
1268 * Other flags are none of our business.
1269 */
1270 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1271 ifp->if_flags &= ~VLAN_COPY_FLAGS;
1272 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1273 #undef VLAN_COPY_FLAGS
1274
1275 ifp->if_link_state = p->if_link_state;
1276
1277 vlan_capabilities(ifv);
1278
1279 /*
1280 * Set up our interface address to reflect the underlying
1281 * physical interface's.
1282 */
1283 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1284 ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1285 p->if_addrlen;
1286
1287 /*
1288 * Configure multicast addresses that may already be
1289 * joined on the vlan device.
1290 */
1291 (void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1292
1293 /* We are ready for operation now. */
1294 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1295 done:
1296 TRUNK_UNLOCK(trunk);
1297 if (error == 0)
1298 EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1299 VLAN_UNLOCK();
1300
1301 return (error);
1302 }
1303
1304 static void
1305 vlan_unconfig(struct ifnet *ifp)
1306 {
1307
1308 VLAN_LOCK();
1309 vlan_unconfig_locked(ifp, 0);
1310 VLAN_UNLOCK();
1311 }
1312
1313 static void
1314 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1315 {
1316 struct ifvlantrunk *trunk;
1317 struct vlan_mc_entry *mc;
1318 struct ifvlan *ifv;
1319 struct ifnet *parent;
1320 int error;
1321
1322 VLAN_LOCK_ASSERT();
1323
1324 ifv = ifp->if_softc;
1325 trunk = ifv->ifv_trunk;
1326 parent = NULL;
1327
1328 if (trunk != NULL) {
1329
1330 TRUNK_LOCK(trunk);
1331 parent = trunk->parent;
1332
1333 /*
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.
1337 */
1338 while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1339 /*
1340 * If the parent interface is being detached,
1341 * all its multicast addresses have already
1342 * been removed. Warn about errors if
1343 * if_delmulti() does fail, but don't abort as
1344 * all callers expect vlan destruction to
1345 * succeed.
1346 */
1347 if (!departing) {
1348 error = if_delmulti(parent,
1349 (struct sockaddr *)&mc->mc_addr);
1350 if (error)
1351 if_printf(ifp,
1352 "Failed to delete multicast address from parent: %d\n",
1353 error);
1354 }
1355 SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1356 free(mc, M_VLAN);
1357 }
1358
1359 vlan_setflags(ifp, 0); /* clear special flags on parent */
1360 vlan_remhash(trunk, ifv);
1361 ifv->ifv_trunk = NULL;
1362
1363 /*
1364 * Check if we were the last.
1365 */
1366 if (trunk->refcnt == 0) {
1367 parent->if_vlantrunk = NULL;
1368 /*
1369 * XXXGL: If some ithread has already entered
1370 * vlan_input() and is now blocked on the trunk
1371 * lock, then it should preempt us right after
1372 * unlock and finish its work. Then we will acquire
1373 * lock again in trunk_destroy().
1374 */
1375 TRUNK_UNLOCK(trunk);
1376 trunk_destroy(trunk);
1377 } else
1378 TRUNK_UNLOCK(trunk);
1379 }
1380
1381 /* Disconnect from parent. */
1382 if (ifv->ifv_pflags)
1383 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1384 ifp->if_mtu = ETHERMTU;
1385 ifp->if_link_state = LINK_STATE_UNKNOWN;
1386 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1387
1388 /*
1389 * Only dispatch an event if vlan was
1390 * attached, otherwise there is nothing
1391 * to cleanup anyway.
1392 */
1393 if (parent != NULL)
1394 EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1395 }
1396
1397 /* Handle a reference counted flag that should be set on the parent as well */
1398 static int
1399 vlan_setflag(struct ifnet *ifp, int flag, int status,
1400 int (*func)(struct ifnet *, int))
1401 {
1402 struct ifvlan *ifv;
1403 int error;
1404
1405 /* XXX VLAN_LOCK_ASSERT(); */
1406
1407 ifv = ifp->if_softc;
1408 status = status ? (ifp->if_flags & flag) : 0;
1409 /* Now "status" contains the flag value or 0 */
1410
1411 /*
1412 * See if recorded parent's status is different from what
1413 * we want it to be. If it is, flip it. We record parent's
1414 * status in ifv_pflags so that we won't clear parent's flag
1415 * we haven't set. In fact, we don't clear or set parent's
1416 * flags directly, but get or release references to them.
1417 * That's why we can be sure that recorded flags still are
1418 * in accord with actual parent's flags.
1419 */
1420 if (status != (ifv->ifv_pflags & flag)) {
1421 error = (*func)(PARENT(ifv), status);
1422 if (error)
1423 return (error);
1424 ifv->ifv_pflags &= ~flag;
1425 ifv->ifv_pflags |= status;
1426 }
1427 return (0);
1428 }
1429
1430 /*
1431 * Handle IFF_* flags that require certain changes on the parent:
1432 * if "status" is true, update parent's flags respective to our if_flags;
1433 * if "status" is false, forcedly clear the flags set on parent.
1434 */
1435 static int
1436 vlan_setflags(struct ifnet *ifp, int status)
1437 {
1438 int error, i;
1439
1440 for (i = 0; vlan_pflags[i].flag; i++) {
1441 error = vlan_setflag(ifp, vlan_pflags[i].flag,
1442 status, vlan_pflags[i].func);
1443 if (error)
1444 return (error);
1445 }
1446 return (0);
1447 }
1448
1449 /* Inform all vlans that their parent has changed link state */
1450 static void
1451 vlan_link_state(struct ifnet *ifp)
1452 {
1453 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1454 struct ifvlan *ifv;
1455 int i;
1456
1457 TRUNK_LOCK(trunk);
1458 #ifdef VLAN_ARRAY
1459 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1460 if (trunk->vlans[i] != NULL) {
1461 ifv = trunk->vlans[i];
1462 #else
1463 for (i = 0; i < (1 << trunk->hwidth); i++)
1464 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1465 #endif
1466 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1467 if_link_state_change(ifv->ifv_ifp,
1468 trunk->parent->if_link_state);
1469 }
1470 TRUNK_UNLOCK(trunk);
1471 }
1472
1473 static void
1474 vlan_capabilities(struct ifvlan *ifv)
1475 {
1476 struct ifnet *p = PARENT(ifv);
1477 struct ifnet *ifp = ifv->ifv_ifp;
1478 struct ifnet_hw_tsomax hw_tsomax;
1479
1480 TRUNK_LOCK_ASSERT(TRUNK(ifv));
1481
1482 /*
1483 * If the parent interface can do checksum offloading
1484 * on VLANs, then propagate its hardware-assisted
1485 * checksumming flags. Also assert that checksum
1486 * offloading requires hardware VLAN tagging.
1487 */
1488 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1489 ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1490
1491 if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1492 p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1493 ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1494 ifp->if_hwassist = p->if_hwassist & (CSUM_IP | CSUM_TCP |
1495 CSUM_UDP | CSUM_SCTP);
1496 } else {
1497 ifp->if_capenable = 0;
1498 ifp->if_hwassist = 0;
1499 }
1500 /*
1501 * If the parent interface can do TSO on VLANs then
1502 * propagate the hardware-assisted flag. TSO on VLANs
1503 * does not necessarily require hardware VLAN tagging.
1504 */
1505 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1506 if_hw_tsomax_common(p, &hw_tsomax);
1507 if_hw_tsomax_update(ifp, &hw_tsomax);
1508 if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1509 ifp->if_capabilities |= p->if_capabilities & IFCAP_TSO;
1510 if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1511 ifp->if_capenable |= p->if_capenable & IFCAP_TSO;
1512 ifp->if_hwassist |= p->if_hwassist & CSUM_TSO;
1513 } else {
1514 ifp->if_capenable &= ~(p->if_capenable & IFCAP_TSO);
1515 ifp->if_hwassist &= ~(p->if_hwassist & CSUM_TSO);
1516 }
1517
1518 /*
1519 * If the parent interface can offload TCP connections over VLANs then
1520 * propagate its TOE capability to the VLAN interface.
1521 *
1522 * All TOE drivers in the tree today can deal with VLANs. If this
1523 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1524 * with its own bit.
1525 */
1526 #define IFCAP_VLAN_TOE IFCAP_TOE
1527 if (p->if_capabilities & IFCAP_VLAN_TOE)
1528 ifp->if_capabilities |= p->if_capabilities & IFCAP_TOE;
1529 if (p->if_capenable & IFCAP_VLAN_TOE) {
1530 TOEDEV(ifp) = TOEDEV(p);
1531 ifp->if_capenable |= p->if_capenable & IFCAP_TOE;
1532 }
1533 }
1534
1535 static void
1536 vlan_trunk_capabilities(struct ifnet *ifp)
1537 {
1538 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1539 struct ifvlan *ifv;
1540 int i;
1541
1542 TRUNK_LOCK(trunk);
1543 #ifdef VLAN_ARRAY
1544 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1545 if (trunk->vlans[i] != NULL) {
1546 ifv = trunk->vlans[i];
1547 #else
1548 for (i = 0; i < (1 << trunk->hwidth); i++) {
1549 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1550 #endif
1551 vlan_capabilities(ifv);
1552 }
1553 TRUNK_UNLOCK(trunk);
1554 }
1555
1556 static int
1557 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1558 {
1559 struct ifnet *p;
1560 struct ifreq *ifr;
1561 struct ifaddr *ifa;
1562 struct ifvlan *ifv;
1563 struct ifvlantrunk *trunk;
1564 struct vlanreq vlr;
1565 int error = 0;
1566
1567 ifr = (struct ifreq *)data;
1568 ifa = (struct ifaddr *) data;
1569 ifv = ifp->if_softc;
1570
1571 switch (cmd) {
1572 case SIOCSIFADDR:
1573 ifp->if_flags |= IFF_UP;
1574 #ifdef INET
1575 if (ifa->ifa_addr->sa_family == AF_INET)
1576 arp_ifinit(ifp, ifa);
1577 #endif
1578 break;
1579 case SIOCGIFADDR:
1580 {
1581 struct sockaddr *sa;
1582
1583 sa = (struct sockaddr *)&ifr->ifr_data;
1584 bcopy(IF_LLADDR(ifp), sa->sa_data, ifp->if_addrlen);
1585 }
1586 break;
1587 case SIOCGIFMEDIA:
1588 VLAN_LOCK();
1589 if (TRUNK(ifv) != NULL) {
1590 p = PARENT(ifv);
1591 VLAN_UNLOCK();
1592 error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1593 /* Limit the result to the parent's current config. */
1594 if (error == 0) {
1595 struct ifmediareq *ifmr;
1596
1597 ifmr = (struct ifmediareq *)data;
1598 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1599 ifmr->ifm_count = 1;
1600 error = copyout(&ifmr->ifm_current,
1601 ifmr->ifm_ulist,
1602 sizeof(int));
1603 }
1604 }
1605 } else {
1606 VLAN_UNLOCK();
1607 error = EINVAL;
1608 }
1609 break;
1610
1611 case SIOCSIFMEDIA:
1612 error = EINVAL;
1613 break;
1614
1615 case SIOCSIFMTU:
1616 /*
1617 * Set the interface MTU.
1618 */
1619 VLAN_LOCK();
1620 if (TRUNK(ifv) != NULL) {
1621 if (ifr->ifr_mtu >
1622 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1623 ifr->ifr_mtu <
1624 (ifv->ifv_mintu - ifv->ifv_mtufudge))
1625 error = EINVAL;
1626 else
1627 ifp->if_mtu = ifr->ifr_mtu;
1628 } else
1629 error = EINVAL;
1630 VLAN_UNLOCK();
1631 break;
1632
1633 case SIOCSETVLAN:
1634 #ifdef VIMAGE
1635 /*
1636 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1637 * interface to be delegated to a jail without allowing the
1638 * jail to change what underlying interface/VID it is
1639 * associated with. We are not entirely convinced that this
1640 * is the right way to accomplish that policy goal.
1641 */
1642 if (ifp->if_vnet != ifp->if_home_vnet) {
1643 error = EPERM;
1644 break;
1645 }
1646 #endif
1647 error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1648 if (error)
1649 break;
1650 if (vlr.vlr_parent[0] == '\0') {
1651 vlan_unconfig(ifp);
1652 break;
1653 }
1654 p = ifunit(vlr.vlr_parent);
1655 if (p == NULL) {
1656 error = ENOENT;
1657 break;
1658 }
1659 error = vlan_config(ifv, p, vlr.vlr_tag);
1660 if (error)
1661 break;
1662
1663 /* Update flags on the parent, if necessary. */
1664 vlan_setflags(ifp, 1);
1665 break;
1666
1667 case SIOCGETVLAN:
1668 #ifdef VIMAGE
1669 if (ifp->if_vnet != ifp->if_home_vnet) {
1670 error = EPERM;
1671 break;
1672 }
1673 #endif
1674 bzero(&vlr, sizeof(vlr));
1675 VLAN_LOCK();
1676 if (TRUNK(ifv) != NULL) {
1677 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1678 sizeof(vlr.vlr_parent));
1679 vlr.vlr_tag = ifv->ifv_vid;
1680 }
1681 VLAN_UNLOCK();
1682 error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1683 break;
1684
1685 case SIOCSIFFLAGS:
1686 /*
1687 * We should propagate selected flags to the parent,
1688 * e.g., promiscuous mode.
1689 */
1690 if (TRUNK(ifv) != NULL)
1691 error = vlan_setflags(ifp, 1);
1692 break;
1693
1694 case SIOCADDMULTI:
1695 case SIOCDELMULTI:
1696 /*
1697 * If we don't have a parent, just remember the membership for
1698 * when we do.
1699 */
1700 trunk = TRUNK(ifv);
1701 if (trunk != NULL) {
1702 TRUNK_LOCK(trunk);
1703 error = vlan_setmulti(ifp);
1704 TRUNK_UNLOCK(trunk);
1705 }
1706 break;
1707
1708 default:
1709 error = EINVAL;
1710 break;
1711 }
1712
1713 return (error);
1714 }
1715