FreeBSD/Linux Kernel Cross Reference
sys/net/if_vlan.c
1 /*-
2 * Copyright 1998 Massachusetts Institute of Technology
3 * Copyright 2012 ADARA Networks, Inc.
4 *
5 * Portions of this software were developed by Robert N. M. Watson under
6 * contract to ADARA Networks, Inc.
7 *
8 * Permission to use, copy, modify, and distribute this software and
9 * its documentation for any purpose and without fee is hereby
10 * granted, provided that both the above copyright notice and this
11 * permission notice appear in all copies, that both the above
12 * copyright notice and this permission notice appear in all
13 * supporting documentation, and that the name of M.I.T. not be used
14 * in advertising or publicity pertaining to distribution of the
15 * software without specific, written prior permission. M.I.T. makes
16 * no representations about the suitability of this software for any
17 * purpose. It is provided "as is" without express or implied
18 * warranty.
19 *
20 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS
21 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
22 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
24 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
25 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
26 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
27 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
28 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 */
33
34 /*
35 * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs.
36 * This is sort of sneaky in the implementation, since
37 * we need to pretend to be enough of an Ethernet implementation
38 * to make arp work. The way we do this is by telling everyone
39 * that we are an Ethernet, and then catch the packets that
40 * ether_output() sends to us via if_transmit(), rewrite them for
41 * use by the real outgoing interface, and ask it to send them.
42 */
43
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD: releng/11.1/sys/net/if_vlan.c 319224 2017-05-30 16:34:18Z mav $");
46
47 #include "opt_inet.h"
48 #include "opt_vlan.h"
49
50 #include <sys/param.h>
51 #include <sys/eventhandler.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/module.h>
57 #include <sys/rmlock.h>
58 #include <sys/priv.h>
59 #include <sys/queue.h>
60 #include <sys/socket.h>
61 #include <sys/sockio.h>
62 #include <sys/sysctl.h>
63 #include <sys/systm.h>
64 #include <sys/sx.h>
65
66 #include <net/bpf.h>
67 #include <net/ethernet.h>
68 #include <net/if.h>
69 #include <net/if_var.h>
70 #include <net/if_clone.h>
71 #include <net/if_dl.h>
72 #include <net/if_types.h>
73 #include <net/if_vlan_var.h>
74 #include <net/vnet.h>
75
76 #ifdef INET
77 #include <netinet/in.h>
78 #include <netinet/if_ether.h>
79 #endif
80
81 #define VLAN_DEF_HWIDTH 4
82 #define VLAN_IFFLAGS (IFF_BROADCAST | IFF_MULTICAST)
83
84 #define UP_AND_RUNNING(ifp) \
85 ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING)
86
87 LIST_HEAD(ifvlanhead, ifvlan);
88
89 struct ifvlantrunk {
90 struct ifnet *parent; /* parent interface of this trunk */
91 struct rmlock lock;
92 #ifdef VLAN_ARRAY
93 #define VLAN_ARRAY_SIZE (EVL_VLID_MASK + 1)
94 struct ifvlan *vlans[VLAN_ARRAY_SIZE]; /* static table */
95 #else
96 struct ifvlanhead *hash; /* dynamic hash-list table */
97 uint16_t hmask;
98 uint16_t hwidth;
99 #endif
100 int refcnt;
101 };
102
103 struct vlan_mc_entry {
104 struct sockaddr_dl mc_addr;
105 SLIST_ENTRY(vlan_mc_entry) mc_entries;
106 };
107
108 struct ifvlan {
109 struct ifvlantrunk *ifv_trunk;
110 struct ifnet *ifv_ifp;
111 #define TRUNK(ifv) ((ifv)->ifv_trunk)
112 #define PARENT(ifv) ((ifv)->ifv_trunk->parent)
113 void *ifv_cookie;
114 int ifv_pflags; /* special flags we have set on parent */
115 int ifv_capenable;
116 struct ifv_linkmib {
117 int ifvm_encaplen; /* encapsulation length */
118 int ifvm_mtufudge; /* MTU fudged by this much */
119 int ifvm_mintu; /* min transmission unit */
120 uint16_t ifvm_proto; /* encapsulation ethertype */
121 uint16_t ifvm_tag; /* tag to apply on packets leaving if */
122 uint16_t ifvm_vid; /* VLAN ID */
123 uint8_t ifvm_pcp; /* Priority Code Point (PCP). */
124 } ifv_mib;
125 SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
126 #ifndef VLAN_ARRAY
127 LIST_ENTRY(ifvlan) ifv_list;
128 #endif
129 };
130 #define ifv_proto ifv_mib.ifvm_proto
131 #define ifv_tag ifv_mib.ifvm_tag
132 #define ifv_vid ifv_mib.ifvm_vid
133 #define ifv_pcp ifv_mib.ifvm_pcp
134 #define ifv_encaplen ifv_mib.ifvm_encaplen
135 #define ifv_mtufudge ifv_mib.ifvm_mtufudge
136 #define ifv_mintu ifv_mib.ifvm_mintu
137
138 /* Special flags we should propagate to parent. */
139 static struct {
140 int flag;
141 int (*func)(struct ifnet *, int);
142 } vlan_pflags[] = {
143 {IFF_PROMISC, ifpromisc},
144 {IFF_ALLMULTI, if_allmulti},
145 {0, NULL}
146 };
147
148 SYSCTL_DECL(_net_link);
149 static SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0,
150 "IEEE 802.1Q VLAN");
151 static SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0,
152 "for consistency");
153
154 static VNET_DEFINE(int, soft_pad);
155 #define V_soft_pad VNET(soft_pad)
156 SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW | CTLFLAG_VNET,
157 &VNET_NAME(soft_pad), 0, "pad short frames before tagging");
158
159 /*
160 * For now, make preserving PCP via an mbuf tag optional, as it increases
161 * per-packet memory allocations and frees. In the future, it would be
162 * preferable to reuse ether_vtag for this, or similar.
163 */
164 static int vlan_mtag_pcp = 0;
165 SYSCTL_INT(_net_link_vlan, OID_AUTO, mtag_pcp, CTLFLAG_RW, &vlan_mtag_pcp, 0,
166 "Retain VLAN PCP information as packets are passed up the stack");
167
168 static const char vlanname[] = "vlan";
169 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
170
171 static eventhandler_tag ifdetach_tag;
172 static eventhandler_tag iflladdr_tag;
173
174 /*
175 * We have a global mutex, that is used to serialize configuration
176 * changes and isn't used in normal packet delivery.
177 *
178 * We also have a per-trunk rmlock(9), that is locked shared on packet
179 * processing and exclusive when configuration is changed.
180 *
181 * The VLAN_ARRAY substitutes the dynamic hash with a static array
182 * with 4096 entries. In theory this can give a boost in processing,
183 * however on practice it does not. Probably this is because array
184 * is too big to fit into CPU cache.
185 */
186 static struct sx ifv_lock;
187 #define VLAN_LOCK_INIT() sx_init(&ifv_lock, "vlan_global")
188 #define VLAN_LOCK_DESTROY() sx_destroy(&ifv_lock)
189 #define VLAN_LOCK_ASSERT() sx_assert(&ifv_lock, SA_LOCKED)
190 #define VLAN_LOCK() sx_xlock(&ifv_lock)
191 #define VLAN_UNLOCK() sx_xunlock(&ifv_lock)
192 #define TRUNK_LOCK_INIT(trunk) rm_init(&(trunk)->lock, vlanname)
193 #define TRUNK_LOCK_DESTROY(trunk) rm_destroy(&(trunk)->lock)
194 #define TRUNK_LOCK(trunk) rm_wlock(&(trunk)->lock)
195 #define TRUNK_UNLOCK(trunk) rm_wunlock(&(trunk)->lock)
196 #define TRUNK_LOCK_ASSERT(trunk) rm_assert(&(trunk)->lock, RA_WLOCKED)
197 #define TRUNK_RLOCK(trunk) rm_rlock(&(trunk)->lock, &tracker)
198 #define TRUNK_RUNLOCK(trunk) rm_runlock(&(trunk)->lock, &tracker)
199 #define TRUNK_LOCK_RASSERT(trunk) rm_assert(&(trunk)->lock, RA_RLOCKED)
200 #define TRUNK_LOCK_READER struct rm_priotracker tracker
201
202 #ifndef VLAN_ARRAY
203 static void vlan_inithash(struct ifvlantrunk *trunk);
204 static void vlan_freehash(struct ifvlantrunk *trunk);
205 static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
206 static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
207 static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
208 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
209 uint16_t vid);
210 #endif
211 static void trunk_destroy(struct ifvlantrunk *trunk);
212
213 static void vlan_init(void *foo);
214 static void vlan_input(struct ifnet *ifp, struct mbuf *m);
215 static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
216 static void vlan_qflush(struct ifnet *ifp);
217 static int vlan_setflag(struct ifnet *ifp, int flag, int status,
218 int (*func)(struct ifnet *, int));
219 static int vlan_setflags(struct ifnet *ifp, int status);
220 static int vlan_setmulti(struct ifnet *ifp);
221 static int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
222 static void vlan_unconfig(struct ifnet *ifp);
223 static void vlan_unconfig_locked(struct ifnet *ifp, int departing);
224 static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
225 static void vlan_link_state(struct ifnet *ifp);
226 static void vlan_capabilities(struct ifvlan *ifv);
227 static void vlan_trunk_capabilities(struct ifnet *ifp);
228
229 static struct ifnet *vlan_clone_match_ethervid(const char *, int *);
230 static int vlan_clone_match(struct if_clone *, const char *);
231 static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
232 static int vlan_clone_destroy(struct if_clone *, struct ifnet *);
233
234 static void vlan_ifdetach(void *arg, struct ifnet *ifp);
235 static void vlan_iflladdr(void *arg, struct ifnet *ifp);
236
237 static struct if_clone *vlan_cloner;
238
239 #ifdef VIMAGE
240 static VNET_DEFINE(struct if_clone *, vlan_cloner);
241 #define V_vlan_cloner VNET(vlan_cloner)
242 #endif
243
244 #ifndef VLAN_ARRAY
245 #define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
246
247 static void
248 vlan_inithash(struct ifvlantrunk *trunk)
249 {
250 int i, n;
251
252 /*
253 * The trunk must not be locked here since we call malloc(M_WAITOK).
254 * It is OK in case this function is called before the trunk struct
255 * gets hooked up and becomes visible from other threads.
256 */
257
258 KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
259 ("%s: hash already initialized", __func__));
260
261 trunk->hwidth = VLAN_DEF_HWIDTH;
262 n = 1 << trunk->hwidth;
263 trunk->hmask = n - 1;
264 trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
265 for (i = 0; i < n; i++)
266 LIST_INIT(&trunk->hash[i]);
267 }
268
269 static void
270 vlan_freehash(struct ifvlantrunk *trunk)
271 {
272 #ifdef INVARIANTS
273 int i;
274
275 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
276 for (i = 0; i < (1 << trunk->hwidth); i++)
277 KASSERT(LIST_EMPTY(&trunk->hash[i]),
278 ("%s: hash table not empty", __func__));
279 #endif
280 free(trunk->hash, M_VLAN);
281 trunk->hash = NULL;
282 trunk->hwidth = trunk->hmask = 0;
283 }
284
285 static int
286 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
287 {
288 int i, b;
289 struct ifvlan *ifv2;
290
291 TRUNK_LOCK_ASSERT(trunk);
292 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
293
294 b = 1 << trunk->hwidth;
295 i = HASH(ifv->ifv_vid, trunk->hmask);
296 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
297 if (ifv->ifv_vid == ifv2->ifv_vid)
298 return (EEXIST);
299
300 /*
301 * Grow the hash when the number of vlans exceeds half of the number of
302 * hash buckets squared. This will make the average linked-list length
303 * buckets/2.
304 */
305 if (trunk->refcnt > (b * b) / 2) {
306 vlan_growhash(trunk, 1);
307 i = HASH(ifv->ifv_vid, trunk->hmask);
308 }
309 LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
310 trunk->refcnt++;
311
312 return (0);
313 }
314
315 static int
316 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
317 {
318 int i, b;
319 struct ifvlan *ifv2;
320
321 TRUNK_LOCK_ASSERT(trunk);
322 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
323
324 b = 1 << trunk->hwidth;
325 i = HASH(ifv->ifv_vid, trunk->hmask);
326 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
327 if (ifv2 == ifv) {
328 trunk->refcnt--;
329 LIST_REMOVE(ifv2, ifv_list);
330 if (trunk->refcnt < (b * b) / 2)
331 vlan_growhash(trunk, -1);
332 return (0);
333 }
334
335 panic("%s: vlan not found\n", __func__);
336 return (ENOENT); /*NOTREACHED*/
337 }
338
339 /*
340 * Grow the hash larger or smaller if memory permits.
341 */
342 static void
343 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
344 {
345 struct ifvlan *ifv;
346 struct ifvlanhead *hash2;
347 int hwidth2, i, j, n, n2;
348
349 TRUNK_LOCK_ASSERT(trunk);
350 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
351
352 if (howmuch == 0) {
353 /* Harmless yet obvious coding error */
354 printf("%s: howmuch is 0\n", __func__);
355 return;
356 }
357
358 hwidth2 = trunk->hwidth + howmuch;
359 n = 1 << trunk->hwidth;
360 n2 = 1 << hwidth2;
361 /* Do not shrink the table below the default */
362 if (hwidth2 < VLAN_DEF_HWIDTH)
363 return;
364
365 /* M_NOWAIT because we're called with trunk mutex held */
366 hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
367 if (hash2 == NULL) {
368 printf("%s: out of memory -- hash size not changed\n",
369 __func__);
370 return; /* We can live with the old hash table */
371 }
372 for (j = 0; j < n2; j++)
373 LIST_INIT(&hash2[j]);
374 for (i = 0; i < n; i++)
375 while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
376 LIST_REMOVE(ifv, ifv_list);
377 j = HASH(ifv->ifv_vid, n2 - 1);
378 LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
379 }
380 free(trunk->hash, M_VLAN);
381 trunk->hash = hash2;
382 trunk->hwidth = hwidth2;
383 trunk->hmask = n2 - 1;
384
385 if (bootverbose)
386 if_printf(trunk->parent,
387 "VLAN hash table resized from %d to %d buckets\n", n, n2);
388 }
389
390 static __inline struct ifvlan *
391 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
392 {
393 struct ifvlan *ifv;
394
395 TRUNK_LOCK_RASSERT(trunk);
396
397 LIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
398 if (ifv->ifv_vid == vid)
399 return (ifv);
400 return (NULL);
401 }
402
403 #if 0
404 /* Debugging code to view the hashtables. */
405 static void
406 vlan_dumphash(struct ifvlantrunk *trunk)
407 {
408 int i;
409 struct ifvlan *ifv;
410
411 for (i = 0; i < (1 << trunk->hwidth); i++) {
412 printf("%d: ", i);
413 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
414 printf("%s ", ifv->ifv_ifp->if_xname);
415 printf("\n");
416 }
417 }
418 #endif /* 0 */
419 #else
420
421 static __inline struct ifvlan *
422 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
423 {
424
425 return trunk->vlans[vid];
426 }
427
428 static __inline int
429 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
430 {
431
432 if (trunk->vlans[ifv->ifv_vid] != NULL)
433 return EEXIST;
434 trunk->vlans[ifv->ifv_vid] = ifv;
435 trunk->refcnt++;
436
437 return (0);
438 }
439
440 static __inline int
441 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
442 {
443
444 trunk->vlans[ifv->ifv_vid] = NULL;
445 trunk->refcnt--;
446
447 return (0);
448 }
449
450 static __inline void
451 vlan_freehash(struct ifvlantrunk *trunk)
452 {
453 }
454
455 static __inline void
456 vlan_inithash(struct ifvlantrunk *trunk)
457 {
458 }
459
460 #endif /* !VLAN_ARRAY */
461
462 static void
463 trunk_destroy(struct ifvlantrunk *trunk)
464 {
465 VLAN_LOCK_ASSERT();
466
467 TRUNK_LOCK(trunk);
468 vlan_freehash(trunk);
469 trunk->parent->if_vlantrunk = NULL;
470 TRUNK_UNLOCK(trunk);
471 TRUNK_LOCK_DESTROY(trunk);
472 if_rele(trunk->parent);
473 free(trunk, M_VLAN);
474 }
475
476 /*
477 * Program our multicast filter. What we're actually doing is
478 * programming the multicast filter of the parent. This has the
479 * side effect of causing the parent interface to receive multicast
480 * traffic that it doesn't really want, which ends up being discarded
481 * later by the upper protocol layers. Unfortunately, there's no way
482 * to avoid this: there really is only one physical interface.
483 */
484 static int
485 vlan_setmulti(struct ifnet *ifp)
486 {
487 struct ifnet *ifp_p;
488 struct ifmultiaddr *ifma;
489 struct ifvlan *sc;
490 struct vlan_mc_entry *mc;
491 int error;
492
493 /* Find the parent. */
494 sc = ifp->if_softc;
495 TRUNK_LOCK_ASSERT(TRUNK(sc));
496 ifp_p = PARENT(sc);
497
498 CURVNET_SET_QUIET(ifp_p->if_vnet);
499
500 /* First, remove any existing filter entries. */
501 while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
502 SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
503 (void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
504 free(mc, M_VLAN);
505 }
506
507 /* Now program new ones. */
508 IF_ADDR_WLOCK(ifp);
509 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
510 if (ifma->ifma_addr->sa_family != AF_LINK)
511 continue;
512 mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
513 if (mc == NULL) {
514 IF_ADDR_WUNLOCK(ifp);
515 return (ENOMEM);
516 }
517 bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
518 mc->mc_addr.sdl_index = ifp_p->if_index;
519 SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
520 }
521 IF_ADDR_WUNLOCK(ifp);
522 SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
523 error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
524 NULL);
525 if (error)
526 return (error);
527 }
528
529 CURVNET_RESTORE();
530 return (0);
531 }
532
533 /*
534 * A handler for parent interface link layer address changes.
535 * If the parent interface link layer address is changed we
536 * should also change it on all children vlans.
537 */
538 static void
539 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
540 {
541 struct ifvlan *ifv;
542 #ifndef VLAN_ARRAY
543 struct ifvlan *next;
544 #endif
545 int i;
546
547 /*
548 * Check if it's a trunk interface first of all
549 * to avoid needless locking.
550 */
551 if (ifp->if_vlantrunk == NULL)
552 return;
553
554 VLAN_LOCK();
555 /*
556 * OK, it's a trunk. Loop over and change all vlan's lladdrs on it.
557 */
558 #ifdef VLAN_ARRAY
559 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
560 if ((ifv = ifp->if_vlantrunk->vlans[i])) {
561 #else /* VLAN_ARRAY */
562 for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
563 LIST_FOREACH_SAFE(ifv, &ifp->if_vlantrunk->hash[i], ifv_list, next) {
564 #endif /* VLAN_ARRAY */
565 VLAN_UNLOCK();
566 if_setlladdr(ifv->ifv_ifp, IF_LLADDR(ifp),
567 ifp->if_addrlen);
568 VLAN_LOCK();
569 }
570 VLAN_UNLOCK();
571
572 }
573
574 /*
575 * A handler for network interface departure events.
576 * Track departure of trunks here so that we don't access invalid
577 * pointers or whatever if a trunk is ripped from under us, e.g.,
578 * by ejecting its hot-plug card. However, if an ifnet is simply
579 * being renamed, then there's no need to tear down the state.
580 */
581 static void
582 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
583 {
584 struct ifvlan *ifv;
585 int i;
586
587 /*
588 * Check if it's a trunk interface first of all
589 * to avoid needless locking.
590 */
591 if (ifp->if_vlantrunk == NULL)
592 return;
593
594 /* If the ifnet is just being renamed, don't do anything. */
595 if (ifp->if_flags & IFF_RENAMING)
596 return;
597
598 VLAN_LOCK();
599 /*
600 * OK, it's a trunk. Loop over and detach all vlan's on it.
601 * Check trunk pointer after each vlan_unconfig() as it will
602 * free it and set to NULL after the last vlan was detached.
603 */
604 #ifdef VLAN_ARRAY
605 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
606 if ((ifv = ifp->if_vlantrunk->vlans[i])) {
607 vlan_unconfig_locked(ifv->ifv_ifp, 1);
608 if (ifp->if_vlantrunk == NULL)
609 break;
610 }
611 #else /* VLAN_ARRAY */
612 restart:
613 for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
614 if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
615 vlan_unconfig_locked(ifv->ifv_ifp, 1);
616 if (ifp->if_vlantrunk)
617 goto restart; /* trunk->hwidth can change */
618 else
619 break;
620 }
621 #endif /* VLAN_ARRAY */
622 /* Trunk should have been destroyed in vlan_unconfig(). */
623 KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
624 VLAN_UNLOCK();
625 }
626
627 /*
628 * Return the trunk device for a virtual interface.
629 */
630 static struct ifnet *
631 vlan_trunkdev(struct ifnet *ifp)
632 {
633 struct ifvlan *ifv;
634
635 if (ifp->if_type != IFT_L2VLAN)
636 return (NULL);
637 ifv = ifp->if_softc;
638 ifp = NULL;
639 VLAN_LOCK();
640 if (ifv->ifv_trunk)
641 ifp = PARENT(ifv);
642 VLAN_UNLOCK();
643 return (ifp);
644 }
645
646 /*
647 * Return the 12-bit VLAN VID for this interface, for use by external
648 * components such as Infiniband.
649 *
650 * XXXRW: Note that the function name here is historical; it should be named
651 * vlan_vid().
652 */
653 static int
654 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
655 {
656 struct ifvlan *ifv;
657
658 if (ifp->if_type != IFT_L2VLAN)
659 return (EINVAL);
660 ifv = ifp->if_softc;
661 *vidp = ifv->ifv_vid;
662 return (0);
663 }
664
665 /*
666 * Return a driver specific cookie for this interface. Synchronization
667 * with setcookie must be provided by the driver.
668 */
669 static void *
670 vlan_cookie(struct ifnet *ifp)
671 {
672 struct ifvlan *ifv;
673
674 if (ifp->if_type != IFT_L2VLAN)
675 return (NULL);
676 ifv = ifp->if_softc;
677 return (ifv->ifv_cookie);
678 }
679
680 /*
681 * Store a cookie in our softc that drivers can use to store driver
682 * private per-instance data in.
683 */
684 static int
685 vlan_setcookie(struct ifnet *ifp, void *cookie)
686 {
687 struct ifvlan *ifv;
688
689 if (ifp->if_type != IFT_L2VLAN)
690 return (EINVAL);
691 ifv = ifp->if_softc;
692 ifv->ifv_cookie = cookie;
693 return (0);
694 }
695
696 /*
697 * Return the vlan device present at the specific VID.
698 */
699 static struct ifnet *
700 vlan_devat(struct ifnet *ifp, uint16_t vid)
701 {
702 struct ifvlantrunk *trunk;
703 struct ifvlan *ifv;
704 TRUNK_LOCK_READER;
705
706 trunk = ifp->if_vlantrunk;
707 if (trunk == NULL)
708 return (NULL);
709 ifp = NULL;
710 TRUNK_RLOCK(trunk);
711 ifv = vlan_gethash(trunk, vid);
712 if (ifv)
713 ifp = ifv->ifv_ifp;
714 TRUNK_RUNLOCK(trunk);
715 return (ifp);
716 }
717
718 /*
719 * Recalculate the cached VLAN tag exposed via the MIB.
720 */
721 static void
722 vlan_tag_recalculate(struct ifvlan *ifv)
723 {
724
725 ifv->ifv_tag = EVL_MAKETAG(ifv->ifv_vid, ifv->ifv_pcp, 0);
726 }
727
728 /*
729 * VLAN support can be loaded as a module. The only place in the
730 * system that's intimately aware of this is ether_input. We hook
731 * into this code through vlan_input_p which is defined there and
732 * set here. No one else in the system should be aware of this so
733 * we use an explicit reference here.
734 */
735 extern void (*vlan_input_p)(struct ifnet *, struct mbuf *);
736
737 /* For if_link_state_change() eyes only... */
738 extern void (*vlan_link_state_p)(struct ifnet *);
739
740 static int
741 vlan_modevent(module_t mod, int type, void *data)
742 {
743
744 switch (type) {
745 case MOD_LOAD:
746 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
747 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
748 if (ifdetach_tag == NULL)
749 return (ENOMEM);
750 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
751 vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
752 if (iflladdr_tag == NULL)
753 return (ENOMEM);
754 VLAN_LOCK_INIT();
755 vlan_input_p = vlan_input;
756 vlan_link_state_p = vlan_link_state;
757 vlan_trunk_cap_p = vlan_trunk_capabilities;
758 vlan_trunkdev_p = vlan_trunkdev;
759 vlan_cookie_p = vlan_cookie;
760 vlan_setcookie_p = vlan_setcookie;
761 vlan_tag_p = vlan_tag;
762 vlan_devat_p = vlan_devat;
763 #ifndef VIMAGE
764 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
765 vlan_clone_create, vlan_clone_destroy);
766 #endif
767 if (bootverbose)
768 printf("vlan: initialized, using "
769 #ifdef VLAN_ARRAY
770 "full-size arrays"
771 #else
772 "hash tables with chaining"
773 #endif
774
775 "\n");
776 break;
777 case MOD_UNLOAD:
778 #ifndef VIMAGE
779 if_clone_detach(vlan_cloner);
780 #endif
781 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
782 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
783 vlan_input_p = NULL;
784 vlan_link_state_p = NULL;
785 vlan_trunk_cap_p = NULL;
786 vlan_trunkdev_p = NULL;
787 vlan_tag_p = NULL;
788 vlan_cookie_p = NULL;
789 vlan_setcookie_p = NULL;
790 vlan_devat_p = NULL;
791 VLAN_LOCK_DESTROY();
792 if (bootverbose)
793 printf("vlan: unloaded\n");
794 break;
795 default:
796 return (EOPNOTSUPP);
797 }
798 return (0);
799 }
800
801 static moduledata_t vlan_mod = {
802 "if_vlan",
803 vlan_modevent,
804 0
805 };
806
807 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
808 MODULE_VERSION(if_vlan, 3);
809
810 #ifdef VIMAGE
811 static void
812 vnet_vlan_init(const void *unused __unused)
813 {
814
815 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
816 vlan_clone_create, vlan_clone_destroy);
817 V_vlan_cloner = vlan_cloner;
818 }
819 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
820 vnet_vlan_init, NULL);
821
822 static void
823 vnet_vlan_uninit(const void *unused __unused)
824 {
825
826 if_clone_detach(V_vlan_cloner);
827 }
828 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
829 vnet_vlan_uninit, NULL);
830 #endif
831
832 /*
833 * Check for <etherif>.<vlan> style interface names.
834 */
835 static struct ifnet *
836 vlan_clone_match_ethervid(const char *name, int *vidp)
837 {
838 char ifname[IFNAMSIZ];
839 char *cp;
840 struct ifnet *ifp;
841 int vid;
842
843 strlcpy(ifname, name, IFNAMSIZ);
844 if ((cp = strchr(ifname, '.')) == NULL)
845 return (NULL);
846 *cp = '\0';
847 if ((ifp = ifunit_ref(ifname)) == NULL)
848 return (NULL);
849 /* Parse VID. */
850 if (*++cp == '\0') {
851 if_rele(ifp);
852 return (NULL);
853 }
854 vid = 0;
855 for(; *cp >= '' && *cp <= '9'; cp++)
856 vid = (vid * 10) + (*cp - '');
857 if (*cp != '\0') {
858 if_rele(ifp);
859 return (NULL);
860 }
861 if (vidp != NULL)
862 *vidp = vid;
863
864 return (ifp);
865 }
866
867 static int
868 vlan_clone_match(struct if_clone *ifc, const char *name)
869 {
870 const char *cp;
871
872 if (vlan_clone_match_ethervid(name, NULL) != NULL)
873 return (1);
874
875 if (strncmp(vlanname, name, strlen(vlanname)) != 0)
876 return (0);
877 for (cp = name + 4; *cp != '\0'; cp++) {
878 if (*cp < '' || *cp > '9')
879 return (0);
880 }
881
882 return (1);
883 }
884
885 static int
886 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
887 {
888 char *dp;
889 int wildcard;
890 int unit;
891 int error;
892 int vid;
893 struct ifvlan *ifv;
894 struct ifnet *ifp;
895 struct ifnet *p;
896 struct ifaddr *ifa;
897 struct sockaddr_dl *sdl;
898 struct vlanreq vlr;
899 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
900
901 /*
902 * There are 3 (ugh) ways to specify the cloned device:
903 * o pass a parameter block with the clone request.
904 * o specify parameters in the text of the clone device name
905 * o specify no parameters and get an unattached device that
906 * must be configured separately.
907 * The first technique is preferred; the latter two are
908 * supported for backwards compatibility.
909 *
910 * XXXRW: Note historic use of the word "tag" here. New ioctls may be
911 * called for.
912 */
913 if (params) {
914 error = copyin(params, &vlr, sizeof(vlr));
915 if (error)
916 return error;
917 p = ifunit_ref(vlr.vlr_parent);
918 if (p == NULL)
919 return (ENXIO);
920 error = ifc_name2unit(name, &unit);
921 if (error != 0) {
922 if_rele(p);
923 return (error);
924 }
925 vid = vlr.vlr_tag;
926 wildcard = (unit < 0);
927 } else if ((p = vlan_clone_match_ethervid(name, &vid)) != NULL) {
928 unit = -1;
929 wildcard = 0;
930 } else {
931 p = NULL;
932 error = ifc_name2unit(name, &unit);
933 if (error != 0)
934 return (error);
935
936 wildcard = (unit < 0);
937 }
938
939 error = ifc_alloc_unit(ifc, &unit);
940 if (error != 0) {
941 if (p != NULL)
942 if_rele(p);
943 return (error);
944 }
945
946 /* In the wildcard case, we need to update the name. */
947 if (wildcard) {
948 for (dp = name; *dp != '\0'; dp++);
949 if (snprintf(dp, len - (dp-name), "%d", unit) >
950 len - (dp-name) - 1) {
951 panic("%s: interface name too long", __func__);
952 }
953 }
954
955 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
956 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
957 if (ifp == NULL) {
958 ifc_free_unit(ifc, unit);
959 free(ifv, M_VLAN);
960 if (p != NULL)
961 if_rele(p);
962 return (ENOSPC);
963 }
964 SLIST_INIT(&ifv->vlan_mc_listhead);
965 ifp->if_softc = ifv;
966 /*
967 * Set the name manually rather than using if_initname because
968 * we don't conform to the default naming convention for interfaces.
969 */
970 strlcpy(ifp->if_xname, name, IFNAMSIZ);
971 ifp->if_dname = vlanname;
972 ifp->if_dunit = unit;
973 /* NB: flags are not set here */
974 ifp->if_linkmib = &ifv->ifv_mib;
975 ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
976 /* NB: mtu is not set here */
977
978 ifp->if_init = vlan_init;
979 ifp->if_transmit = vlan_transmit;
980 ifp->if_qflush = vlan_qflush;
981 ifp->if_ioctl = vlan_ioctl;
982 ifp->if_flags = VLAN_IFFLAGS;
983 ether_ifattach(ifp, eaddr);
984 /* Now undo some of the damage... */
985 ifp->if_baudrate = 0;
986 ifp->if_type = IFT_L2VLAN;
987 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
988 ifa = ifp->if_addr;
989 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
990 sdl->sdl_type = IFT_L2VLAN;
991
992 if (p != NULL) {
993 error = vlan_config(ifv, p, vid);
994 if_rele(p);
995 if (error != 0) {
996 /*
997 * Since we've partially failed, we need to back
998 * out all the way, otherwise userland could get
999 * confused. Thus, we destroy the interface.
1000 */
1001 ether_ifdetach(ifp);
1002 vlan_unconfig(ifp);
1003 if_free(ifp);
1004 ifc_free_unit(ifc, unit);
1005 free(ifv, M_VLAN);
1006
1007 return (error);
1008 }
1009
1010 /* Update flags on the parent, if necessary. */
1011 vlan_setflags(ifp, 1);
1012 }
1013
1014 return (0);
1015 }
1016
1017 static int
1018 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1019 {
1020 struct ifvlan *ifv = ifp->if_softc;
1021 int unit = ifp->if_dunit;
1022
1023 ether_ifdetach(ifp); /* first, remove it from system-wide lists */
1024 vlan_unconfig(ifp); /* now it can be unconfigured and freed */
1025 if_free(ifp);
1026 free(ifv, M_VLAN);
1027 ifc_free_unit(ifc, unit);
1028
1029 return (0);
1030 }
1031
1032 /*
1033 * The ifp->if_init entry point for vlan(4) is a no-op.
1034 */
1035 static void
1036 vlan_init(void *foo __unused)
1037 {
1038 }
1039
1040 /*
1041 * The if_transmit method for vlan(4) interface.
1042 */
1043 static int
1044 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1045 {
1046 struct ifvlan *ifv;
1047 struct ifnet *p;
1048 struct m_tag *mtag;
1049 uint16_t tag;
1050 int error, len, mcast;
1051
1052 ifv = ifp->if_softc;
1053 p = PARENT(ifv);
1054 len = m->m_pkthdr.len;
1055 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1056
1057 BPF_MTAP(ifp, m);
1058
1059 /*
1060 * Do not run parent's if_transmit() if the parent is not up,
1061 * or parent's driver will cause a system crash.
1062 */
1063 if (!UP_AND_RUNNING(p)) {
1064 m_freem(m);
1065 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1066 return (ENETDOWN);
1067 }
1068
1069 /*
1070 * Pad the frame to the minimum size allowed if told to.
1071 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
1072 * paragraph C.4.4.3.b. It can help to work around buggy
1073 * bridges that violate paragraph C.4.4.3.a from the same
1074 * document, i.e., fail to pad short frames after untagging.
1075 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
1076 * untagging it will produce a 62-byte frame, which is a runt
1077 * and requires padding. There are VLAN-enabled network
1078 * devices that just discard such runts instead or mishandle
1079 * them somehow.
1080 */
1081 if (V_soft_pad && p->if_type == IFT_ETHER) {
1082 static char pad[8]; /* just zeros */
1083 int n;
1084
1085 for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
1086 n > 0; n -= sizeof(pad))
1087 if (!m_append(m, min(n, sizeof(pad)), pad))
1088 break;
1089
1090 if (n > 0) {
1091 if_printf(ifp, "cannot pad short frame\n");
1092 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1093 m_freem(m);
1094 return (0);
1095 }
1096 }
1097
1098 /*
1099 * If underlying interface can do VLAN tag insertion itself,
1100 * just pass the packet along. However, we need some way to
1101 * tell the interface where the packet came from so that it
1102 * knows how to find the VLAN tag to use, so we attach a
1103 * packet tag that holds it.
1104 */
1105 if (vlan_mtag_pcp && (mtag = m_tag_locate(m, MTAG_8021Q,
1106 MTAG_8021Q_PCP_OUT, NULL)) != NULL)
1107 tag = EVL_MAKETAG(ifv->ifv_vid, *(uint8_t *)(mtag + 1), 0);
1108 else
1109 tag = ifv->ifv_tag;
1110 if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1111 m->m_pkthdr.ether_vtag = tag;
1112 m->m_flags |= M_VLANTAG;
1113 } else {
1114 m = ether_vlanencap(m, tag);
1115 if (m == NULL) {
1116 if_printf(ifp, "unable to prepend VLAN header\n");
1117 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1118 return (0);
1119 }
1120 }
1121
1122 /*
1123 * Send it, precisely as ether_output() would have.
1124 */
1125 error = (p->if_transmit)(p, m);
1126 if (error == 0) {
1127 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1128 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1129 if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1130 } else
1131 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1132 return (error);
1133 }
1134
1135 /*
1136 * The ifp->if_qflush entry point for vlan(4) is a no-op.
1137 */
1138 static void
1139 vlan_qflush(struct ifnet *ifp __unused)
1140 {
1141 }
1142
1143 static void
1144 vlan_input(struct ifnet *ifp, struct mbuf *m)
1145 {
1146 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1147 struct ifvlan *ifv;
1148 TRUNK_LOCK_READER;
1149 struct m_tag *mtag;
1150 uint16_t vid, tag;
1151
1152 KASSERT(trunk != NULL, ("%s: no trunk", __func__));
1153
1154 if (m->m_flags & M_VLANTAG) {
1155 /*
1156 * Packet is tagged, but m contains a normal
1157 * Ethernet frame; the tag is stored out-of-band.
1158 */
1159 tag = m->m_pkthdr.ether_vtag;
1160 m->m_flags &= ~M_VLANTAG;
1161 } else {
1162 struct ether_vlan_header *evl;
1163
1164 /*
1165 * Packet is tagged in-band as specified by 802.1q.
1166 */
1167 switch (ifp->if_type) {
1168 case IFT_ETHER:
1169 if (m->m_len < sizeof(*evl) &&
1170 (m = m_pullup(m, sizeof(*evl))) == NULL) {
1171 if_printf(ifp, "cannot pullup VLAN header\n");
1172 return;
1173 }
1174 evl = mtod(m, struct ether_vlan_header *);
1175 tag = ntohs(evl->evl_tag);
1176
1177 /*
1178 * Remove the 802.1q header by copying the Ethernet
1179 * addresses over it and adjusting the beginning of
1180 * the data in the mbuf. The encapsulated Ethernet
1181 * type field is already in place.
1182 */
1183 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1184 ETHER_HDR_LEN - ETHER_TYPE_LEN);
1185 m_adj(m, ETHER_VLAN_ENCAP_LEN);
1186 break;
1187
1188 default:
1189 #ifdef INVARIANTS
1190 panic("%s: %s has unsupported if_type %u",
1191 __func__, ifp->if_xname, ifp->if_type);
1192 #endif
1193 m_freem(m);
1194 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1195 return;
1196 }
1197 }
1198
1199 vid = EVL_VLANOFTAG(tag);
1200
1201 TRUNK_RLOCK(trunk);
1202 ifv = vlan_gethash(trunk, vid);
1203 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1204 TRUNK_RUNLOCK(trunk);
1205 m_freem(m);
1206 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1207 return;
1208 }
1209 TRUNK_RUNLOCK(trunk);
1210
1211 if (vlan_mtag_pcp) {
1212 /*
1213 * While uncommon, it is possible that we will find a 802.1q
1214 * packet encapsulated inside another packet that also had an
1215 * 802.1q header. For example, ethernet tunneled over IPSEC
1216 * arriving over ethernet. In that case, we replace the
1217 * existing 802.1q PCP m_tag value.
1218 */
1219 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1220 if (mtag == NULL) {
1221 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1222 sizeof(uint8_t), M_NOWAIT);
1223 if (mtag == NULL) {
1224 m_freem(m);
1225 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1226 return;
1227 }
1228 m_tag_prepend(m, mtag);
1229 }
1230 *(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1231 }
1232
1233 m->m_pkthdr.rcvif = ifv->ifv_ifp;
1234 if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1235
1236 /* Pass it back through the parent's input routine. */
1237 (*ifp->if_input)(ifv->ifv_ifp, m);
1238 }
1239
1240 static int
1241 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid)
1242 {
1243 struct ifvlantrunk *trunk;
1244 struct ifnet *ifp;
1245 int error = 0;
1246
1247 /*
1248 * We can handle non-ethernet hardware types as long as
1249 * they handle the tagging and headers themselves.
1250 */
1251 if (p->if_type != IFT_ETHER &&
1252 (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1253 return (EPROTONOSUPPORT);
1254 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1255 return (EPROTONOSUPPORT);
1256 /*
1257 * Don't let the caller set up a VLAN VID with
1258 * anything except VLID bits.
1259 * VID numbers 0x0 and 0xFFF are reserved.
1260 */
1261 if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1262 return (EINVAL);
1263 if (ifv->ifv_trunk)
1264 return (EBUSY);
1265
1266 if (p->if_vlantrunk == NULL) {
1267 trunk = malloc(sizeof(struct ifvlantrunk),
1268 M_VLAN, M_WAITOK | M_ZERO);
1269 vlan_inithash(trunk);
1270 VLAN_LOCK();
1271 if (p->if_vlantrunk != NULL) {
1272 /* A race that is very unlikely to be hit. */
1273 vlan_freehash(trunk);
1274 free(trunk, M_VLAN);
1275 goto exists;
1276 }
1277 TRUNK_LOCK_INIT(trunk);
1278 TRUNK_LOCK(trunk);
1279 p->if_vlantrunk = trunk;
1280 trunk->parent = p;
1281 if_ref(trunk->parent);
1282 } else {
1283 VLAN_LOCK();
1284 exists:
1285 trunk = p->if_vlantrunk;
1286 TRUNK_LOCK(trunk);
1287 }
1288
1289 ifv->ifv_vid = vid; /* must set this before vlan_inshash() */
1290 ifv->ifv_pcp = 0; /* Default: best effort delivery. */
1291 vlan_tag_recalculate(ifv);
1292 error = vlan_inshash(trunk, ifv);
1293 if (error)
1294 goto done;
1295 ifv->ifv_proto = ETHERTYPE_VLAN;
1296 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1297 ifv->ifv_mintu = ETHERMIN;
1298 ifv->ifv_pflags = 0;
1299 ifv->ifv_capenable = -1;
1300
1301 /*
1302 * If the parent supports the VLAN_MTU capability,
1303 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1304 * use it.
1305 */
1306 if (p->if_capenable & IFCAP_VLAN_MTU) {
1307 /*
1308 * No need to fudge the MTU since the parent can
1309 * handle extended frames.
1310 */
1311 ifv->ifv_mtufudge = 0;
1312 } else {
1313 /*
1314 * Fudge the MTU by the encapsulation size. This
1315 * makes us incompatible with strictly compliant
1316 * 802.1Q implementations, but allows us to use
1317 * the feature with other NetBSD implementations,
1318 * which might still be useful.
1319 */
1320 ifv->ifv_mtufudge = ifv->ifv_encaplen;
1321 }
1322
1323 ifv->ifv_trunk = trunk;
1324 ifp = ifv->ifv_ifp;
1325 /*
1326 * Initialize fields from our parent. This duplicates some
1327 * work with ether_ifattach() but allows for non-ethernet
1328 * interfaces to also work.
1329 */
1330 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1331 ifp->if_baudrate = p->if_baudrate;
1332 ifp->if_output = p->if_output;
1333 ifp->if_input = p->if_input;
1334 ifp->if_resolvemulti = p->if_resolvemulti;
1335 ifp->if_addrlen = p->if_addrlen;
1336 ifp->if_broadcastaddr = p->if_broadcastaddr;
1337
1338 /*
1339 * Copy only a selected subset of flags from the parent.
1340 * Other flags are none of our business.
1341 */
1342 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1343 ifp->if_flags &= ~VLAN_COPY_FLAGS;
1344 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1345 #undef VLAN_COPY_FLAGS
1346
1347 ifp->if_link_state = p->if_link_state;
1348
1349 vlan_capabilities(ifv);
1350
1351 /*
1352 * Set up our interface address to reflect the underlying
1353 * physical interface's.
1354 */
1355 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1356 ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1357 p->if_addrlen;
1358
1359 /*
1360 * Configure multicast addresses that may already be
1361 * joined on the vlan device.
1362 */
1363 (void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1364
1365 /* We are ready for operation now. */
1366 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1367 done:
1368 TRUNK_UNLOCK(trunk);
1369 if (error == 0)
1370 EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1371 VLAN_UNLOCK();
1372
1373 return (error);
1374 }
1375
1376 static void
1377 vlan_unconfig(struct ifnet *ifp)
1378 {
1379
1380 VLAN_LOCK();
1381 vlan_unconfig_locked(ifp, 0);
1382 VLAN_UNLOCK();
1383 }
1384
1385 static void
1386 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1387 {
1388 struct ifvlantrunk *trunk;
1389 struct vlan_mc_entry *mc;
1390 struct ifvlan *ifv;
1391 struct ifnet *parent;
1392 int error;
1393
1394 VLAN_LOCK_ASSERT();
1395
1396 ifv = ifp->if_softc;
1397 trunk = ifv->ifv_trunk;
1398 parent = NULL;
1399
1400 if (trunk != NULL) {
1401
1402 TRUNK_LOCK(trunk);
1403 parent = trunk->parent;
1404
1405 /*
1406 * Since the interface is being unconfigured, we need to
1407 * empty the list of multicast groups that we may have joined
1408 * while we were alive from the parent's list.
1409 */
1410 while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1411 /*
1412 * If the parent interface is being detached,
1413 * all its multicast addresses have already
1414 * been removed. Warn about errors if
1415 * if_delmulti() does fail, but don't abort as
1416 * all callers expect vlan destruction to
1417 * succeed.
1418 */
1419 if (!departing) {
1420 error = if_delmulti(parent,
1421 (struct sockaddr *)&mc->mc_addr);
1422 if (error)
1423 if_printf(ifp,
1424 "Failed to delete multicast address from parent: %d\n",
1425 error);
1426 }
1427 SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1428 free(mc, M_VLAN);
1429 }
1430
1431 vlan_setflags(ifp, 0); /* clear special flags on parent */
1432 vlan_remhash(trunk, ifv);
1433 ifv->ifv_trunk = NULL;
1434
1435 /*
1436 * Check if we were the last.
1437 */
1438 if (trunk->refcnt == 0) {
1439 parent->if_vlantrunk = NULL;
1440 /*
1441 * XXXGL: If some ithread has already entered
1442 * vlan_input() and is now blocked on the trunk
1443 * lock, then it should preempt us right after
1444 * unlock and finish its work. Then we will acquire
1445 * lock again in trunk_destroy().
1446 */
1447 TRUNK_UNLOCK(trunk);
1448 trunk_destroy(trunk);
1449 } else
1450 TRUNK_UNLOCK(trunk);
1451 }
1452
1453 /* Disconnect from parent. */
1454 if (ifv->ifv_pflags)
1455 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1456 ifp->if_mtu = ETHERMTU;
1457 ifp->if_link_state = LINK_STATE_UNKNOWN;
1458 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1459
1460 /*
1461 * Only dispatch an event if vlan was
1462 * attached, otherwise there is nothing
1463 * to cleanup anyway.
1464 */
1465 if (parent != NULL)
1466 EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1467 }
1468
1469 /* Handle a reference counted flag that should be set on the parent as well */
1470 static int
1471 vlan_setflag(struct ifnet *ifp, int flag, int status,
1472 int (*func)(struct ifnet *, int))
1473 {
1474 struct ifvlan *ifv;
1475 int error;
1476
1477 /* XXX VLAN_LOCK_ASSERT(); */
1478
1479 ifv = ifp->if_softc;
1480 status = status ? (ifp->if_flags & flag) : 0;
1481 /* Now "status" contains the flag value or 0 */
1482
1483 /*
1484 * See if recorded parent's status is different from what
1485 * we want it to be. If it is, flip it. We record parent's
1486 * status in ifv_pflags so that we won't clear parent's flag
1487 * we haven't set. In fact, we don't clear or set parent's
1488 * flags directly, but get or release references to them.
1489 * That's why we can be sure that recorded flags still are
1490 * in accord with actual parent's flags.
1491 */
1492 if (status != (ifv->ifv_pflags & flag)) {
1493 error = (*func)(PARENT(ifv), status);
1494 if (error)
1495 return (error);
1496 ifv->ifv_pflags &= ~flag;
1497 ifv->ifv_pflags |= status;
1498 }
1499 return (0);
1500 }
1501
1502 /*
1503 * Handle IFF_* flags that require certain changes on the parent:
1504 * if "status" is true, update parent's flags respective to our if_flags;
1505 * if "status" is false, forcedly clear the flags set on parent.
1506 */
1507 static int
1508 vlan_setflags(struct ifnet *ifp, int status)
1509 {
1510 int error, i;
1511
1512 for (i = 0; vlan_pflags[i].flag; i++) {
1513 error = vlan_setflag(ifp, vlan_pflags[i].flag,
1514 status, vlan_pflags[i].func);
1515 if (error)
1516 return (error);
1517 }
1518 return (0);
1519 }
1520
1521 /* Inform all vlans that their parent has changed link state */
1522 static void
1523 vlan_link_state(struct ifnet *ifp)
1524 {
1525 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1526 struct ifvlan *ifv;
1527 int i;
1528
1529 TRUNK_LOCK(trunk);
1530 #ifdef VLAN_ARRAY
1531 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1532 if (trunk->vlans[i] != NULL) {
1533 ifv = trunk->vlans[i];
1534 #else
1535 for (i = 0; i < (1 << trunk->hwidth); i++)
1536 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1537 #endif
1538 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1539 if_link_state_change(ifv->ifv_ifp,
1540 trunk->parent->if_link_state);
1541 }
1542 TRUNK_UNLOCK(trunk);
1543 }
1544
1545 static void
1546 vlan_capabilities(struct ifvlan *ifv)
1547 {
1548 struct ifnet *p = PARENT(ifv);
1549 struct ifnet *ifp = ifv->ifv_ifp;
1550 struct ifnet_hw_tsomax hw_tsomax;
1551 int cap = 0, ena = 0, mena;
1552 u_long hwa = 0;
1553
1554 TRUNK_LOCK_ASSERT(TRUNK(ifv));
1555
1556 /* Mask parent interface enabled capabilities disabled by user. */
1557 mena = p->if_capenable & ifv->ifv_capenable;
1558
1559 /*
1560 * If the parent interface can do checksum offloading
1561 * on VLANs, then propagate its hardware-assisted
1562 * checksumming flags. Also assert that checksum
1563 * offloading requires hardware VLAN tagging.
1564 */
1565 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1566 cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1567 if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1568 p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1569 ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6);
1570 if (ena & IFCAP_TXCSUM)
1571 hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP |
1572 CSUM_UDP | CSUM_SCTP);
1573 if (ena & IFCAP_TXCSUM_IPV6)
1574 hwa |= p->if_hwassist & (CSUM_TCP_IPV6 |
1575 CSUM_UDP_IPV6 | CSUM_SCTP_IPV6);
1576 }
1577
1578 /*
1579 * If the parent interface can do TSO on VLANs then
1580 * propagate the hardware-assisted flag. TSO on VLANs
1581 * does not necessarily require hardware VLAN tagging.
1582 */
1583 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1584 if_hw_tsomax_common(p, &hw_tsomax);
1585 if_hw_tsomax_update(ifp, &hw_tsomax);
1586 if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1587 cap |= p->if_capabilities & IFCAP_TSO;
1588 if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1589 ena |= mena & IFCAP_TSO;
1590 if (ena & IFCAP_TSO)
1591 hwa |= p->if_hwassist & CSUM_TSO;
1592 }
1593
1594 /*
1595 * If the parent interface can do LRO and checksum offloading on
1596 * VLANs, then guess it may do LRO on VLANs. False positive here
1597 * cost nothing, while false negative may lead to some confusions.
1598 */
1599 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1600 cap |= p->if_capabilities & IFCAP_LRO;
1601 if (p->if_capenable & IFCAP_VLAN_HWCSUM)
1602 ena |= p->if_capenable & IFCAP_LRO;
1603
1604 /*
1605 * If the parent interface can offload TCP connections over VLANs then
1606 * propagate its TOE capability to the VLAN interface.
1607 *
1608 * All TOE drivers in the tree today can deal with VLANs. If this
1609 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1610 * with its own bit.
1611 */
1612 #define IFCAP_VLAN_TOE IFCAP_TOE
1613 if (p->if_capabilities & IFCAP_VLAN_TOE)
1614 cap |= p->if_capabilities & IFCAP_TOE;
1615 if (p->if_capenable & IFCAP_VLAN_TOE) {
1616 TOEDEV(ifp) = TOEDEV(p);
1617 ena |= mena & IFCAP_TOE;
1618 }
1619
1620 /*
1621 * If the parent interface supports dynamic link state, so does the
1622 * VLAN interface.
1623 */
1624 cap |= (p->if_capabilities & IFCAP_LINKSTATE);
1625 ena |= (mena & IFCAP_LINKSTATE);
1626
1627 ifp->if_capabilities = cap;
1628 ifp->if_capenable = ena;
1629 ifp->if_hwassist = hwa;
1630 }
1631
1632 static void
1633 vlan_trunk_capabilities(struct ifnet *ifp)
1634 {
1635 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1636 struct ifvlan *ifv;
1637 int i;
1638
1639 TRUNK_LOCK(trunk);
1640 #ifdef VLAN_ARRAY
1641 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1642 if (trunk->vlans[i] != NULL) {
1643 ifv = trunk->vlans[i];
1644 #else
1645 for (i = 0; i < (1 << trunk->hwidth); i++) {
1646 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1647 #endif
1648 vlan_capabilities(ifv);
1649 }
1650 TRUNK_UNLOCK(trunk);
1651 }
1652
1653 static int
1654 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1655 {
1656 struct ifnet *p;
1657 struct ifreq *ifr;
1658 struct ifaddr *ifa;
1659 struct ifvlan *ifv;
1660 struct ifvlantrunk *trunk;
1661 struct vlanreq vlr;
1662 int error = 0;
1663
1664 ifr = (struct ifreq *)data;
1665 ifa = (struct ifaddr *) data;
1666 ifv = ifp->if_softc;
1667
1668 switch (cmd) {
1669 case SIOCSIFADDR:
1670 ifp->if_flags |= IFF_UP;
1671 #ifdef INET
1672 if (ifa->ifa_addr->sa_family == AF_INET)
1673 arp_ifinit(ifp, ifa);
1674 #endif
1675 break;
1676 case SIOCGIFADDR:
1677 {
1678 struct sockaddr *sa;
1679
1680 sa = (struct sockaddr *)&ifr->ifr_data;
1681 bcopy(IF_LLADDR(ifp), sa->sa_data, ifp->if_addrlen);
1682 }
1683 break;
1684 case SIOCGIFMEDIA:
1685 VLAN_LOCK();
1686 if (TRUNK(ifv) != NULL) {
1687 p = PARENT(ifv);
1688 if_ref(p);
1689 VLAN_UNLOCK();
1690 error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1691 if_rele(p);
1692 /* Limit the result to the parent's current config. */
1693 if (error == 0) {
1694 struct ifmediareq *ifmr;
1695
1696 ifmr = (struct ifmediareq *)data;
1697 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1698 ifmr->ifm_count = 1;
1699 error = copyout(&ifmr->ifm_current,
1700 ifmr->ifm_ulist,
1701 sizeof(int));
1702 }
1703 }
1704 } else {
1705 VLAN_UNLOCK();
1706 error = EINVAL;
1707 }
1708 break;
1709
1710 case SIOCSIFMEDIA:
1711 error = EINVAL;
1712 break;
1713
1714 case SIOCSIFMTU:
1715 /*
1716 * Set the interface MTU.
1717 */
1718 VLAN_LOCK();
1719 if (TRUNK(ifv) != NULL) {
1720 if (ifr->ifr_mtu >
1721 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1722 ifr->ifr_mtu <
1723 (ifv->ifv_mintu - ifv->ifv_mtufudge))
1724 error = EINVAL;
1725 else
1726 ifp->if_mtu = ifr->ifr_mtu;
1727 } else
1728 error = EINVAL;
1729 VLAN_UNLOCK();
1730 break;
1731
1732 case SIOCSETVLAN:
1733 #ifdef VIMAGE
1734 /*
1735 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1736 * interface to be delegated to a jail without allowing the
1737 * jail to change what underlying interface/VID it is
1738 * associated with. We are not entirely convinced that this
1739 * is the right way to accomplish that policy goal.
1740 */
1741 if (ifp->if_vnet != ifp->if_home_vnet) {
1742 error = EPERM;
1743 break;
1744 }
1745 #endif
1746 error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1747 if (error)
1748 break;
1749 if (vlr.vlr_parent[0] == '\0') {
1750 vlan_unconfig(ifp);
1751 break;
1752 }
1753 p = ifunit_ref(vlr.vlr_parent);
1754 if (p == NULL) {
1755 error = ENOENT;
1756 break;
1757 }
1758 error = vlan_config(ifv, p, vlr.vlr_tag);
1759 if_rele(p);
1760 if (error)
1761 break;
1762
1763 /* Update flags on the parent, if necessary. */
1764 vlan_setflags(ifp, 1);
1765 break;
1766
1767 case SIOCGETVLAN:
1768 #ifdef VIMAGE
1769 if (ifp->if_vnet != ifp->if_home_vnet) {
1770 error = EPERM;
1771 break;
1772 }
1773 #endif
1774 bzero(&vlr, sizeof(vlr));
1775 VLAN_LOCK();
1776 if (TRUNK(ifv) != NULL) {
1777 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1778 sizeof(vlr.vlr_parent));
1779 vlr.vlr_tag = ifv->ifv_vid;
1780 }
1781 VLAN_UNLOCK();
1782 error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1783 break;
1784
1785 case SIOCSIFFLAGS:
1786 /*
1787 * We should propagate selected flags to the parent,
1788 * e.g., promiscuous mode.
1789 */
1790 if (TRUNK(ifv) != NULL)
1791 error = vlan_setflags(ifp, 1);
1792 break;
1793
1794 case SIOCADDMULTI:
1795 case SIOCDELMULTI:
1796 /*
1797 * If we don't have a parent, just remember the membership for
1798 * when we do.
1799 */
1800 trunk = TRUNK(ifv);
1801 if (trunk != NULL) {
1802 TRUNK_LOCK(trunk);
1803 error = vlan_setmulti(ifp);
1804 TRUNK_UNLOCK(trunk);
1805 }
1806 break;
1807
1808 case SIOCGVLANPCP:
1809 #ifdef VIMAGE
1810 if (ifp->if_vnet != ifp->if_home_vnet) {
1811 error = EPERM;
1812 break;
1813 }
1814 #endif
1815 ifr->ifr_vlan_pcp = ifv->ifv_pcp;
1816 break;
1817
1818 case SIOCSVLANPCP:
1819 #ifdef VIMAGE
1820 if (ifp->if_vnet != ifp->if_home_vnet) {
1821 error = EPERM;
1822 break;
1823 }
1824 #endif
1825 error = priv_check(curthread, PRIV_NET_SETVLANPCP);
1826 if (error)
1827 break;
1828 if (ifr->ifr_vlan_pcp > 7) {
1829 error = EINVAL;
1830 break;
1831 }
1832 ifv->ifv_pcp = ifr->ifr_vlan_pcp;
1833 vlan_tag_recalculate(ifv);
1834 break;
1835
1836 case SIOCSIFCAP:
1837 VLAN_LOCK();
1838 ifv->ifv_capenable = ifr->ifr_reqcap;
1839 trunk = TRUNK(ifv);
1840 if (trunk != NULL) {
1841 TRUNK_LOCK(trunk);
1842 vlan_capabilities(ifv);
1843 TRUNK_UNLOCK(trunk);
1844 }
1845 VLAN_UNLOCK();
1846 break;
1847
1848 default:
1849 error = EINVAL;
1850 break;
1851 }
1852
1853 return (error);
1854 }
Cache object: 7429dde31137396669b870821c07c8d8
|