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/11.0/sys/net/if_vlan.c 302054 2016-06-21 13:48:49Z bz $");
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 struct ifv_linkmib {
116 int ifvm_encaplen; /* encapsulation length */
117 int ifvm_mtufudge; /* MTU fudged by this much */
118 int ifvm_mintu; /* min transmission unit */
119 uint16_t ifvm_proto; /* encapsulation ethertype */
120 uint16_t ifvm_tag; /* tag to apply on packets leaving if */
121 uint16_t ifvm_vid; /* VLAN ID */
122 uint8_t ifvm_pcp; /* Priority Code Point (PCP). */
123 } ifv_mib;
124 SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead;
125 #ifndef VLAN_ARRAY
126 LIST_ENTRY(ifvlan) ifv_list;
127 #endif
128 };
129 #define ifv_proto ifv_mib.ifvm_proto
130 #define ifv_tag ifv_mib.ifvm_tag
131 #define ifv_vid ifv_mib.ifvm_vid
132 #define ifv_pcp ifv_mib.ifvm_pcp
133 #define ifv_encaplen ifv_mib.ifvm_encaplen
134 #define ifv_mtufudge ifv_mib.ifvm_mtufudge
135 #define ifv_mintu ifv_mib.ifvm_mintu
136
137 /* Special flags we should propagate to parent. */
138 static struct {
139 int flag;
140 int (*func)(struct ifnet *, int);
141 } vlan_pflags[] = {
142 {IFF_PROMISC, ifpromisc},
143 {IFF_ALLMULTI, if_allmulti},
144 {0, NULL}
145 };
146
147 SYSCTL_DECL(_net_link);
148 static SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0,
149 "IEEE 802.1Q VLAN");
150 static SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0,
151 "for consistency");
152
153 static VNET_DEFINE(int, soft_pad);
154 #define V_soft_pad VNET(soft_pad)
155 SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW | CTLFLAG_VNET,
156 &VNET_NAME(soft_pad), 0, "pad short frames before tagging");
157
158 /*
159 * For now, make preserving PCP via an mbuf tag optional, as it increases
160 * per-packet memory allocations and frees. In the future, it would be
161 * preferable to reuse ether_vtag for this, or similar.
162 */
163 static int vlan_mtag_pcp = 0;
164 SYSCTL_INT(_net_link_vlan, OID_AUTO, mtag_pcp, CTLFLAG_RW, &vlan_mtag_pcp, 0,
165 "Retain VLAN PCP information as packets are passed up the stack");
166
167 static const char vlanname[] = "vlan";
168 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface");
169
170 static eventhandler_tag ifdetach_tag;
171 static eventhandler_tag iflladdr_tag;
172
173 /*
174 * We have a global mutex, that is used to serialize configuration
175 * changes and isn't used in normal packet delivery.
176 *
177 * We also have a per-trunk rmlock(9), that is locked shared on packet
178 * processing and exclusive when configuration is changed.
179 *
180 * The VLAN_ARRAY substitutes the dynamic hash with a static array
181 * with 4096 entries. In theory this can give a boost in processing,
182 * however on practice it does not. Probably this is because array
183 * is too big to fit into CPU cache.
184 */
185 static struct sx ifv_lock;
186 #define VLAN_LOCK_INIT() sx_init(&ifv_lock, "vlan_global")
187 #define VLAN_LOCK_DESTROY() sx_destroy(&ifv_lock)
188 #define VLAN_LOCK_ASSERT() sx_assert(&ifv_lock, SA_LOCKED)
189 #define VLAN_LOCK() sx_xlock(&ifv_lock)
190 #define VLAN_UNLOCK() sx_xunlock(&ifv_lock)
191 #define TRUNK_LOCK_INIT(trunk) rm_init(&(trunk)->lock, vlanname)
192 #define TRUNK_LOCK_DESTROY(trunk) rm_destroy(&(trunk)->lock)
193 #define TRUNK_LOCK(trunk) rm_wlock(&(trunk)->lock)
194 #define TRUNK_UNLOCK(trunk) rm_wunlock(&(trunk)->lock)
195 #define TRUNK_LOCK_ASSERT(trunk) rm_assert(&(trunk)->lock, RA_WLOCKED)
196 #define TRUNK_RLOCK(trunk) rm_rlock(&(trunk)->lock, &tracker)
197 #define TRUNK_RUNLOCK(trunk) rm_runlock(&(trunk)->lock, &tracker)
198 #define TRUNK_LOCK_RASSERT(trunk) rm_assert(&(trunk)->lock, RA_RLOCKED)
199 #define TRUNK_LOCK_READER struct rm_priotracker tracker
200
201 #ifndef VLAN_ARRAY
202 static void vlan_inithash(struct ifvlantrunk *trunk);
203 static void vlan_freehash(struct ifvlantrunk *trunk);
204 static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
205 static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv);
206 static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch);
207 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk,
208 uint16_t vid);
209 #endif
210 static void trunk_destroy(struct ifvlantrunk *trunk);
211
212 static void vlan_init(void *foo);
213 static void vlan_input(struct ifnet *ifp, struct mbuf *m);
214 static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr);
215 static void vlan_qflush(struct ifnet *ifp);
216 static int vlan_setflag(struct ifnet *ifp, int flag, int status,
217 int (*func)(struct ifnet *, int));
218 static int vlan_setflags(struct ifnet *ifp, int status);
219 static int vlan_setmulti(struct ifnet *ifp);
220 static int vlan_transmit(struct ifnet *ifp, struct mbuf *m);
221 static void vlan_unconfig(struct ifnet *ifp);
222 static void vlan_unconfig_locked(struct ifnet *ifp, int departing);
223 static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag);
224 static void vlan_link_state(struct ifnet *ifp);
225 static void vlan_capabilities(struct ifvlan *ifv);
226 static void vlan_trunk_capabilities(struct ifnet *ifp);
227
228 static struct ifnet *vlan_clone_match_ethervid(const char *, int *);
229 static int vlan_clone_match(struct if_clone *, const char *);
230 static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t);
231 static int vlan_clone_destroy(struct if_clone *, struct ifnet *);
232
233 static void vlan_ifdetach(void *arg, struct ifnet *ifp);
234 static void vlan_iflladdr(void *arg, struct ifnet *ifp);
235
236 static struct if_clone *vlan_cloner;
237
238 #ifdef VIMAGE
239 static VNET_DEFINE(struct if_clone *, vlan_cloner);
240 #define V_vlan_cloner VNET(vlan_cloner)
241 #endif
242
243 #ifndef VLAN_ARRAY
244 #define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m))
245
246 static void
247 vlan_inithash(struct ifvlantrunk *trunk)
248 {
249 int i, n;
250
251 /*
252 * The trunk must not be locked here since we call malloc(M_WAITOK).
253 * It is OK in case this function is called before the trunk struct
254 * gets hooked up and becomes visible from other threads.
255 */
256
257 KASSERT(trunk->hwidth == 0 && trunk->hash == NULL,
258 ("%s: hash already initialized", __func__));
259
260 trunk->hwidth = VLAN_DEF_HWIDTH;
261 n = 1 << trunk->hwidth;
262 trunk->hmask = n - 1;
263 trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK);
264 for (i = 0; i < n; i++)
265 LIST_INIT(&trunk->hash[i]);
266 }
267
268 static void
269 vlan_freehash(struct ifvlantrunk *trunk)
270 {
271 #ifdef INVARIANTS
272 int i;
273
274 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
275 for (i = 0; i < (1 << trunk->hwidth); i++)
276 KASSERT(LIST_EMPTY(&trunk->hash[i]),
277 ("%s: hash table not empty", __func__));
278 #endif
279 free(trunk->hash, M_VLAN);
280 trunk->hash = NULL;
281 trunk->hwidth = trunk->hmask = 0;
282 }
283
284 static int
285 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
286 {
287 int i, b;
288 struct ifvlan *ifv2;
289
290 TRUNK_LOCK_ASSERT(trunk);
291 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
292
293 b = 1 << trunk->hwidth;
294 i = HASH(ifv->ifv_vid, trunk->hmask);
295 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
296 if (ifv->ifv_vid == ifv2->ifv_vid)
297 return (EEXIST);
298
299 /*
300 * Grow the hash when the number of vlans exceeds half of the number of
301 * hash buckets squared. This will make the average linked-list length
302 * buckets/2.
303 */
304 if (trunk->refcnt > (b * b) / 2) {
305 vlan_growhash(trunk, 1);
306 i = HASH(ifv->ifv_vid, trunk->hmask);
307 }
308 LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list);
309 trunk->refcnt++;
310
311 return (0);
312 }
313
314 static int
315 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
316 {
317 int i, b;
318 struct ifvlan *ifv2;
319
320 TRUNK_LOCK_ASSERT(trunk);
321 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
322
323 b = 1 << trunk->hwidth;
324 i = HASH(ifv->ifv_vid, trunk->hmask);
325 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list)
326 if (ifv2 == ifv) {
327 trunk->refcnt--;
328 LIST_REMOVE(ifv2, ifv_list);
329 if (trunk->refcnt < (b * b) / 2)
330 vlan_growhash(trunk, -1);
331 return (0);
332 }
333
334 panic("%s: vlan not found\n", __func__);
335 return (ENOENT); /*NOTREACHED*/
336 }
337
338 /*
339 * Grow the hash larger or smaller if memory permits.
340 */
341 static void
342 vlan_growhash(struct ifvlantrunk *trunk, int howmuch)
343 {
344 struct ifvlan *ifv;
345 struct ifvlanhead *hash2;
346 int hwidth2, i, j, n, n2;
347
348 TRUNK_LOCK_ASSERT(trunk);
349 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__));
350
351 if (howmuch == 0) {
352 /* Harmless yet obvious coding error */
353 printf("%s: howmuch is 0\n", __func__);
354 return;
355 }
356
357 hwidth2 = trunk->hwidth + howmuch;
358 n = 1 << trunk->hwidth;
359 n2 = 1 << hwidth2;
360 /* Do not shrink the table below the default */
361 if (hwidth2 < VLAN_DEF_HWIDTH)
362 return;
363
364 /* M_NOWAIT because we're called with trunk mutex held */
365 hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT);
366 if (hash2 == NULL) {
367 printf("%s: out of memory -- hash size not changed\n",
368 __func__);
369 return; /* We can live with the old hash table */
370 }
371 for (j = 0; j < n2; j++)
372 LIST_INIT(&hash2[j]);
373 for (i = 0; i < n; i++)
374 while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) {
375 LIST_REMOVE(ifv, ifv_list);
376 j = HASH(ifv->ifv_vid, n2 - 1);
377 LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list);
378 }
379 free(trunk->hash, M_VLAN);
380 trunk->hash = hash2;
381 trunk->hwidth = hwidth2;
382 trunk->hmask = n2 - 1;
383
384 if (bootverbose)
385 if_printf(trunk->parent,
386 "VLAN hash table resized from %d to %d buckets\n", n, n2);
387 }
388
389 static __inline struct ifvlan *
390 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
391 {
392 struct ifvlan *ifv;
393
394 TRUNK_LOCK_RASSERT(trunk);
395
396 LIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list)
397 if (ifv->ifv_vid == vid)
398 return (ifv);
399 return (NULL);
400 }
401
402 #if 0
403 /* Debugging code to view the hashtables. */
404 static void
405 vlan_dumphash(struct ifvlantrunk *trunk)
406 {
407 int i;
408 struct ifvlan *ifv;
409
410 for (i = 0; i < (1 << trunk->hwidth); i++) {
411 printf("%d: ", i);
412 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
413 printf("%s ", ifv->ifv_ifp->if_xname);
414 printf("\n");
415 }
416 }
417 #endif /* 0 */
418 #else
419
420 static __inline struct ifvlan *
421 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid)
422 {
423
424 return trunk->vlans[vid];
425 }
426
427 static __inline int
428 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
429 {
430
431 if (trunk->vlans[ifv->ifv_vid] != NULL)
432 return EEXIST;
433 trunk->vlans[ifv->ifv_vid] = ifv;
434 trunk->refcnt++;
435
436 return (0);
437 }
438
439 static __inline int
440 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv)
441 {
442
443 trunk->vlans[ifv->ifv_vid] = NULL;
444 trunk->refcnt--;
445
446 return (0);
447 }
448
449 static __inline void
450 vlan_freehash(struct ifvlantrunk *trunk)
451 {
452 }
453
454 static __inline void
455 vlan_inithash(struct ifvlantrunk *trunk)
456 {
457 }
458
459 #endif /* !VLAN_ARRAY */
460
461 static void
462 trunk_destroy(struct ifvlantrunk *trunk)
463 {
464 VLAN_LOCK_ASSERT();
465
466 TRUNK_LOCK(trunk);
467 vlan_freehash(trunk);
468 trunk->parent->if_vlantrunk = NULL;
469 TRUNK_UNLOCK(trunk);
470 TRUNK_LOCK_DESTROY(trunk);
471 free(trunk, M_VLAN);
472 }
473
474 /*
475 * Program our multicast filter. What we're actually doing is
476 * programming the multicast filter of the parent. This has the
477 * side effect of causing the parent interface to receive multicast
478 * traffic that it doesn't really want, which ends up being discarded
479 * later by the upper protocol layers. Unfortunately, there's no way
480 * to avoid this: there really is only one physical interface.
481 */
482 static int
483 vlan_setmulti(struct ifnet *ifp)
484 {
485 struct ifnet *ifp_p;
486 struct ifmultiaddr *ifma;
487 struct ifvlan *sc;
488 struct vlan_mc_entry *mc;
489 int error;
490
491 /* Find the parent. */
492 sc = ifp->if_softc;
493 TRUNK_LOCK_ASSERT(TRUNK(sc));
494 ifp_p = PARENT(sc);
495
496 CURVNET_SET_QUIET(ifp_p->if_vnet);
497
498 /* First, remove any existing filter entries. */
499 while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) {
500 SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries);
501 (void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr);
502 free(mc, M_VLAN);
503 }
504
505 /* Now program new ones. */
506 IF_ADDR_WLOCK(ifp);
507 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
508 if (ifma->ifma_addr->sa_family != AF_LINK)
509 continue;
510 mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT);
511 if (mc == NULL) {
512 IF_ADDR_WUNLOCK(ifp);
513 return (ENOMEM);
514 }
515 bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
516 mc->mc_addr.sdl_index = ifp_p->if_index;
517 SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries);
518 }
519 IF_ADDR_WUNLOCK(ifp);
520 SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) {
521 error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr,
522 NULL);
523 if (error)
524 return (error);
525 }
526
527 CURVNET_RESTORE();
528 return (0);
529 }
530
531 /*
532 * A handler for parent interface link layer address changes.
533 * If the parent interface link layer address is changed we
534 * should also change it on all children vlans.
535 */
536 static void
537 vlan_iflladdr(void *arg __unused, struct ifnet *ifp)
538 {
539 struct ifvlan *ifv;
540 #ifndef VLAN_ARRAY
541 struct ifvlan *next;
542 #endif
543 int i;
544
545 /*
546 * Check if it's a trunk interface first of all
547 * to avoid needless locking.
548 */
549 if (ifp->if_vlantrunk == NULL)
550 return;
551
552 VLAN_LOCK();
553 /*
554 * OK, it's a trunk. Loop over and change all vlan's lladdrs on it.
555 */
556 #ifdef VLAN_ARRAY
557 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
558 if ((ifv = ifp->if_vlantrunk->vlans[i])) {
559 #else /* VLAN_ARRAY */
560 for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
561 LIST_FOREACH_SAFE(ifv, &ifp->if_vlantrunk->hash[i], ifv_list, next) {
562 #endif /* VLAN_ARRAY */
563 VLAN_UNLOCK();
564 if_setlladdr(ifv->ifv_ifp, IF_LLADDR(ifp),
565 ifp->if_addrlen);
566 VLAN_LOCK();
567 }
568 VLAN_UNLOCK();
569
570 }
571
572 /*
573 * A handler for network interface departure events.
574 * Track departure of trunks here so that we don't access invalid
575 * pointers or whatever if a trunk is ripped from under us, e.g.,
576 * by ejecting its hot-plug card. However, if an ifnet is simply
577 * being renamed, then there's no need to tear down the state.
578 */
579 static void
580 vlan_ifdetach(void *arg __unused, struct ifnet *ifp)
581 {
582 struct ifvlan *ifv;
583 int i;
584
585 /*
586 * Check if it's a trunk interface first of all
587 * to avoid needless locking.
588 */
589 if (ifp->if_vlantrunk == NULL)
590 return;
591
592 /* If the ifnet is just being renamed, don't do anything. */
593 if (ifp->if_flags & IFF_RENAMING)
594 return;
595
596 VLAN_LOCK();
597 /*
598 * OK, it's a trunk. Loop over and detach all vlan's on it.
599 * Check trunk pointer after each vlan_unconfig() as it will
600 * free it and set to NULL after the last vlan was detached.
601 */
602 #ifdef VLAN_ARRAY
603 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
604 if ((ifv = ifp->if_vlantrunk->vlans[i])) {
605 vlan_unconfig_locked(ifv->ifv_ifp, 1);
606 if (ifp->if_vlantrunk == NULL)
607 break;
608 }
609 #else /* VLAN_ARRAY */
610 restart:
611 for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++)
612 if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) {
613 vlan_unconfig_locked(ifv->ifv_ifp, 1);
614 if (ifp->if_vlantrunk)
615 goto restart; /* trunk->hwidth can change */
616 else
617 break;
618 }
619 #endif /* VLAN_ARRAY */
620 /* Trunk should have been destroyed in vlan_unconfig(). */
621 KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__));
622 VLAN_UNLOCK();
623 }
624
625 /*
626 * Return the trunk device for a virtual interface.
627 */
628 static struct ifnet *
629 vlan_trunkdev(struct ifnet *ifp)
630 {
631 struct ifvlan *ifv;
632
633 if (ifp->if_type != IFT_L2VLAN)
634 return (NULL);
635 ifv = ifp->if_softc;
636 ifp = NULL;
637 VLAN_LOCK();
638 if (ifv->ifv_trunk)
639 ifp = PARENT(ifv);
640 VLAN_UNLOCK();
641 return (ifp);
642 }
643
644 /*
645 * Return the 12-bit VLAN VID for this interface, for use by external
646 * components such as Infiniband.
647 *
648 * XXXRW: Note that the function name here is historical; it should be named
649 * vlan_vid().
650 */
651 static int
652 vlan_tag(struct ifnet *ifp, uint16_t *vidp)
653 {
654 struct ifvlan *ifv;
655
656 if (ifp->if_type != IFT_L2VLAN)
657 return (EINVAL);
658 ifv = ifp->if_softc;
659 *vidp = ifv->ifv_vid;
660 return (0);
661 }
662
663 /*
664 * Return a driver specific cookie for this interface. Synchronization
665 * with setcookie must be provided by the driver.
666 */
667 static void *
668 vlan_cookie(struct ifnet *ifp)
669 {
670 struct ifvlan *ifv;
671
672 if (ifp->if_type != IFT_L2VLAN)
673 return (NULL);
674 ifv = ifp->if_softc;
675 return (ifv->ifv_cookie);
676 }
677
678 /*
679 * Store a cookie in our softc that drivers can use to store driver
680 * private per-instance data in.
681 */
682 static int
683 vlan_setcookie(struct ifnet *ifp, void *cookie)
684 {
685 struct ifvlan *ifv;
686
687 if (ifp->if_type != IFT_L2VLAN)
688 return (EINVAL);
689 ifv = ifp->if_softc;
690 ifv->ifv_cookie = cookie;
691 return (0);
692 }
693
694 /*
695 * Return the vlan device present at the specific VID.
696 */
697 static struct ifnet *
698 vlan_devat(struct ifnet *ifp, uint16_t vid)
699 {
700 struct ifvlantrunk *trunk;
701 struct ifvlan *ifv;
702 TRUNK_LOCK_READER;
703
704 trunk = ifp->if_vlantrunk;
705 if (trunk == NULL)
706 return (NULL);
707 ifp = NULL;
708 TRUNK_RLOCK(trunk);
709 ifv = vlan_gethash(trunk, vid);
710 if (ifv)
711 ifp = ifv->ifv_ifp;
712 TRUNK_RUNLOCK(trunk);
713 return (ifp);
714 }
715
716 /*
717 * Recalculate the cached VLAN tag exposed via the MIB.
718 */
719 static void
720 vlan_tag_recalculate(struct ifvlan *ifv)
721 {
722
723 ifv->ifv_tag = EVL_MAKETAG(ifv->ifv_vid, ifv->ifv_pcp, 0);
724 }
725
726 /*
727 * VLAN support can be loaded as a module. The only place in the
728 * system that's intimately aware of this is ether_input. We hook
729 * into this code through vlan_input_p which is defined there and
730 * set here. No one else in the system should be aware of this so
731 * we use an explicit reference here.
732 */
733 extern void (*vlan_input_p)(struct ifnet *, struct mbuf *);
734
735 /* For if_link_state_change() eyes only... */
736 extern void (*vlan_link_state_p)(struct ifnet *);
737
738 static int
739 vlan_modevent(module_t mod, int type, void *data)
740 {
741
742 switch (type) {
743 case MOD_LOAD:
744 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
745 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY);
746 if (ifdetach_tag == NULL)
747 return (ENOMEM);
748 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
749 vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
750 if (iflladdr_tag == NULL)
751 return (ENOMEM);
752 VLAN_LOCK_INIT();
753 vlan_input_p = vlan_input;
754 vlan_link_state_p = vlan_link_state;
755 vlan_trunk_cap_p = vlan_trunk_capabilities;
756 vlan_trunkdev_p = vlan_trunkdev;
757 vlan_cookie_p = vlan_cookie;
758 vlan_setcookie_p = vlan_setcookie;
759 vlan_tag_p = vlan_tag;
760 vlan_devat_p = vlan_devat;
761 #ifndef VIMAGE
762 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
763 vlan_clone_create, vlan_clone_destroy);
764 #endif
765 if (bootverbose)
766 printf("vlan: initialized, using "
767 #ifdef VLAN_ARRAY
768 "full-size arrays"
769 #else
770 "hash tables with chaining"
771 #endif
772
773 "\n");
774 break;
775 case MOD_UNLOAD:
776 #ifndef VIMAGE
777 if_clone_detach(vlan_cloner);
778 #endif
779 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag);
780 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag);
781 vlan_input_p = NULL;
782 vlan_link_state_p = NULL;
783 vlan_trunk_cap_p = NULL;
784 vlan_trunkdev_p = NULL;
785 vlan_tag_p = NULL;
786 vlan_cookie_p = NULL;
787 vlan_setcookie_p = NULL;
788 vlan_devat_p = NULL;
789 VLAN_LOCK_DESTROY();
790 if (bootverbose)
791 printf("vlan: unloaded\n");
792 break;
793 default:
794 return (EOPNOTSUPP);
795 }
796 return (0);
797 }
798
799 static moduledata_t vlan_mod = {
800 "if_vlan",
801 vlan_modevent,
802 0
803 };
804
805 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
806 MODULE_VERSION(if_vlan, 3);
807
808 #ifdef VIMAGE
809 static void
810 vnet_vlan_init(const void *unused __unused)
811 {
812
813 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match,
814 vlan_clone_create, vlan_clone_destroy);
815 V_vlan_cloner = vlan_cloner;
816 }
817 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
818 vnet_vlan_init, NULL);
819
820 static void
821 vnet_vlan_uninit(const void *unused __unused)
822 {
823
824 if_clone_detach(V_vlan_cloner);
825 }
826 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
827 vnet_vlan_uninit, NULL);
828 #endif
829
830 /*
831 * Check for <etherif>.<vlan> style interface names.
832 */
833 static struct ifnet *
834 vlan_clone_match_ethervid(const char *name, int *vidp)
835 {
836 char ifname[IFNAMSIZ];
837 char *cp;
838 struct ifnet *ifp;
839 int vid;
840
841 strlcpy(ifname, name, IFNAMSIZ);
842 if ((cp = strchr(ifname, '.')) == NULL)
843 return (NULL);
844 *cp = '\0';
845 if ((ifp = ifunit(ifname)) == NULL)
846 return (NULL);
847 /* Parse VID. */
848 if (*++cp == '\0')
849 return (NULL);
850 vid = 0;
851 for(; *cp >= '' && *cp <= '9'; cp++)
852 vid = (vid * 10) + (*cp - '');
853 if (*cp != '\0')
854 return (NULL);
855 if (vidp != NULL)
856 *vidp = vid;
857
858 return (ifp);
859 }
860
861 static int
862 vlan_clone_match(struct if_clone *ifc, const char *name)
863 {
864 const char *cp;
865
866 if (vlan_clone_match_ethervid(name, NULL) != NULL)
867 return (1);
868
869 if (strncmp(vlanname, name, strlen(vlanname)) != 0)
870 return (0);
871 for (cp = name + 4; *cp != '\0'; cp++) {
872 if (*cp < '' || *cp > '9')
873 return (0);
874 }
875
876 return (1);
877 }
878
879 static int
880 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
881 {
882 char *dp;
883 int wildcard;
884 int unit;
885 int error;
886 int vid;
887 int ethertag;
888 struct ifvlan *ifv;
889 struct ifnet *ifp;
890 struct ifnet *p;
891 struct ifaddr *ifa;
892 struct sockaddr_dl *sdl;
893 struct vlanreq vlr;
894 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
895
896 /*
897 * There are 3 (ugh) ways to specify the cloned device:
898 * o pass a parameter block with the clone request.
899 * o specify parameters in the text of the clone device name
900 * o specify no parameters and get an unattached device that
901 * must be configured separately.
902 * The first technique is preferred; the latter two are
903 * supported for backwards compatibility.
904 *
905 * XXXRW: Note historic use of the word "tag" here. New ioctls may be
906 * called for.
907 */
908 if (params) {
909 error = copyin(params, &vlr, sizeof(vlr));
910 if (error)
911 return error;
912 p = ifunit(vlr.vlr_parent);
913 if (p == NULL)
914 return (ENXIO);
915 error = ifc_name2unit(name, &unit);
916 if (error != 0)
917 return (error);
918
919 ethertag = 1;
920 vid = vlr.vlr_tag;
921 wildcard = (unit < 0);
922 } else if ((p = vlan_clone_match_ethervid(name, &vid)) != NULL) {
923 ethertag = 1;
924 unit = -1;
925 wildcard = 0;
926 } else {
927 ethertag = 0;
928
929 error = ifc_name2unit(name, &unit);
930 if (error != 0)
931 return (error);
932
933 wildcard = (unit < 0);
934 }
935
936 error = ifc_alloc_unit(ifc, &unit);
937 if (error != 0)
938 return (error);
939
940 /* In the wildcard case, we need to update the name. */
941 if (wildcard) {
942 for (dp = name; *dp != '\0'; dp++);
943 if (snprintf(dp, len - (dp-name), "%d", unit) >
944 len - (dp-name) - 1) {
945 panic("%s: interface name too long", __func__);
946 }
947 }
948
949 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO);
950 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER);
951 if (ifp == NULL) {
952 ifc_free_unit(ifc, unit);
953 free(ifv, M_VLAN);
954 return (ENOSPC);
955 }
956 SLIST_INIT(&ifv->vlan_mc_listhead);
957 ifp->if_softc = ifv;
958 /*
959 * Set the name manually rather than using if_initname because
960 * we don't conform to the default naming convention for interfaces.
961 */
962 strlcpy(ifp->if_xname, name, IFNAMSIZ);
963 ifp->if_dname = vlanname;
964 ifp->if_dunit = unit;
965 /* NB: flags are not set here */
966 ifp->if_linkmib = &ifv->ifv_mib;
967 ifp->if_linkmiblen = sizeof(ifv->ifv_mib);
968 /* NB: mtu is not set here */
969
970 ifp->if_init = vlan_init;
971 ifp->if_transmit = vlan_transmit;
972 ifp->if_qflush = vlan_qflush;
973 ifp->if_ioctl = vlan_ioctl;
974 ifp->if_flags = VLAN_IFFLAGS;
975 ether_ifattach(ifp, eaddr);
976 /* Now undo some of the damage... */
977 ifp->if_baudrate = 0;
978 ifp->if_type = IFT_L2VLAN;
979 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN;
980 ifa = ifp->if_addr;
981 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
982 sdl->sdl_type = IFT_L2VLAN;
983
984 if (ethertag) {
985 error = vlan_config(ifv, p, vid);
986 if (error != 0) {
987 /*
988 * Since we've partially failed, we need to back
989 * out all the way, otherwise userland could get
990 * confused. Thus, we destroy the interface.
991 */
992 ether_ifdetach(ifp);
993 vlan_unconfig(ifp);
994 if_free(ifp);
995 ifc_free_unit(ifc, unit);
996 free(ifv, M_VLAN);
997
998 return (error);
999 }
1000
1001 /* Update flags on the parent, if necessary. */
1002 vlan_setflags(ifp, 1);
1003 }
1004
1005 return (0);
1006 }
1007
1008 static int
1009 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
1010 {
1011 struct ifvlan *ifv = ifp->if_softc;
1012 int unit = ifp->if_dunit;
1013
1014 ether_ifdetach(ifp); /* first, remove it from system-wide lists */
1015 vlan_unconfig(ifp); /* now it can be unconfigured and freed */
1016 if_free(ifp);
1017 free(ifv, M_VLAN);
1018 ifc_free_unit(ifc, unit);
1019
1020 return (0);
1021 }
1022
1023 /*
1024 * The ifp->if_init entry point for vlan(4) is a no-op.
1025 */
1026 static void
1027 vlan_init(void *foo __unused)
1028 {
1029 }
1030
1031 /*
1032 * The if_transmit method for vlan(4) interface.
1033 */
1034 static int
1035 vlan_transmit(struct ifnet *ifp, struct mbuf *m)
1036 {
1037 struct ifvlan *ifv;
1038 struct ifnet *p;
1039 struct m_tag *mtag;
1040 uint16_t tag;
1041 int error, len, mcast;
1042
1043 ifv = ifp->if_softc;
1044 p = PARENT(ifv);
1045 len = m->m_pkthdr.len;
1046 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1047
1048 BPF_MTAP(ifp, m);
1049
1050 /*
1051 * Do not run parent's if_transmit() if the parent is not up,
1052 * or parent's driver will cause a system crash.
1053 */
1054 if (!UP_AND_RUNNING(p)) {
1055 m_freem(m);
1056 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1057 return (ENETDOWN);
1058 }
1059
1060 /*
1061 * Pad the frame to the minimum size allowed if told to.
1062 * This option is in accord with IEEE Std 802.1Q, 2003 Ed.,
1063 * paragraph C.4.4.3.b. It can help to work around buggy
1064 * bridges that violate paragraph C.4.4.3.a from the same
1065 * document, i.e., fail to pad short frames after untagging.
1066 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but
1067 * untagging it will produce a 62-byte frame, which is a runt
1068 * and requires padding. There are VLAN-enabled network
1069 * devices that just discard such runts instead or mishandle
1070 * them somehow.
1071 */
1072 if (V_soft_pad && p->if_type == IFT_ETHER) {
1073 static char pad[8]; /* just zeros */
1074 int n;
1075
1076 for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len;
1077 n > 0; n -= sizeof(pad))
1078 if (!m_append(m, min(n, sizeof(pad)), pad))
1079 break;
1080
1081 if (n > 0) {
1082 if_printf(ifp, "cannot pad short frame\n");
1083 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1084 m_freem(m);
1085 return (0);
1086 }
1087 }
1088
1089 /*
1090 * If underlying interface can do VLAN tag insertion itself,
1091 * just pass the packet along. However, we need some way to
1092 * tell the interface where the packet came from so that it
1093 * knows how to find the VLAN tag to use, so we attach a
1094 * packet tag that holds it.
1095 */
1096 if (vlan_mtag_pcp && (mtag = m_tag_locate(m, MTAG_8021Q,
1097 MTAG_8021Q_PCP_OUT, NULL)) != NULL)
1098 tag = EVL_MAKETAG(ifv->ifv_vid, *(uint8_t *)(mtag + 1), 0);
1099 else
1100 tag = ifv->ifv_tag;
1101 if (p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1102 m->m_pkthdr.ether_vtag = tag;
1103 m->m_flags |= M_VLANTAG;
1104 } else {
1105 m = ether_vlanencap(m, tag);
1106 if (m == NULL) {
1107 if_printf(ifp, "unable to prepend VLAN header\n");
1108 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1109 return (0);
1110 }
1111 }
1112
1113 /*
1114 * Send it, precisely as ether_output() would have.
1115 */
1116 error = (p->if_transmit)(p, m);
1117 if (error == 0) {
1118 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1119 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
1120 if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast);
1121 } else
1122 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1123 return (error);
1124 }
1125
1126 /*
1127 * The ifp->if_qflush entry point for vlan(4) is a no-op.
1128 */
1129 static void
1130 vlan_qflush(struct ifnet *ifp __unused)
1131 {
1132 }
1133
1134 static void
1135 vlan_input(struct ifnet *ifp, struct mbuf *m)
1136 {
1137 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1138 struct ifvlan *ifv;
1139 TRUNK_LOCK_READER;
1140 struct m_tag *mtag;
1141 uint16_t vid, tag;
1142
1143 KASSERT(trunk != NULL, ("%s: no trunk", __func__));
1144
1145 if (m->m_flags & M_VLANTAG) {
1146 /*
1147 * Packet is tagged, but m contains a normal
1148 * Ethernet frame; the tag is stored out-of-band.
1149 */
1150 tag = m->m_pkthdr.ether_vtag;
1151 m->m_flags &= ~M_VLANTAG;
1152 } else {
1153 struct ether_vlan_header *evl;
1154
1155 /*
1156 * Packet is tagged in-band as specified by 802.1q.
1157 */
1158 switch (ifp->if_type) {
1159 case IFT_ETHER:
1160 if (m->m_len < sizeof(*evl) &&
1161 (m = m_pullup(m, sizeof(*evl))) == NULL) {
1162 if_printf(ifp, "cannot pullup VLAN header\n");
1163 return;
1164 }
1165 evl = mtod(m, struct ether_vlan_header *);
1166 tag = ntohs(evl->evl_tag);
1167
1168 /*
1169 * Remove the 802.1q header by copying the Ethernet
1170 * addresses over it and adjusting the beginning of
1171 * the data in the mbuf. The encapsulated Ethernet
1172 * type field is already in place.
1173 */
1174 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
1175 ETHER_HDR_LEN - ETHER_TYPE_LEN);
1176 m_adj(m, ETHER_VLAN_ENCAP_LEN);
1177 break;
1178
1179 default:
1180 #ifdef INVARIANTS
1181 panic("%s: %s has unsupported if_type %u",
1182 __func__, ifp->if_xname, ifp->if_type);
1183 #endif
1184 m_freem(m);
1185 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1186 return;
1187 }
1188 }
1189
1190 vid = EVL_VLANOFTAG(tag);
1191
1192 TRUNK_RLOCK(trunk);
1193 ifv = vlan_gethash(trunk, vid);
1194 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) {
1195 TRUNK_RUNLOCK(trunk);
1196 m_freem(m);
1197 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
1198 return;
1199 }
1200 TRUNK_RUNLOCK(trunk);
1201
1202 if (vlan_mtag_pcp) {
1203 /*
1204 * While uncommon, it is possible that we will find a 802.1q
1205 * packet encapsulated inside another packet that also had an
1206 * 802.1q header. For example, ethernet tunneled over IPSEC
1207 * arriving over ethernet. In that case, we replace the
1208 * existing 802.1q PCP m_tag value.
1209 */
1210 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
1211 if (mtag == NULL) {
1212 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN,
1213 sizeof(uint8_t), M_NOWAIT);
1214 if (mtag == NULL) {
1215 m_freem(m);
1216 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1217 return;
1218 }
1219 m_tag_prepend(m, mtag);
1220 }
1221 *(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag);
1222 }
1223
1224 m->m_pkthdr.rcvif = ifv->ifv_ifp;
1225 if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1);
1226
1227 /* Pass it back through the parent's input routine. */
1228 (*ifp->if_input)(ifv->ifv_ifp, m);
1229 }
1230
1231 static int
1232 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid)
1233 {
1234 struct ifvlantrunk *trunk;
1235 struct ifnet *ifp;
1236 int error = 0;
1237
1238 /*
1239 * We can handle non-ethernet hardware types as long as
1240 * they handle the tagging and headers themselves.
1241 */
1242 if (p->if_type != IFT_ETHER &&
1243 (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
1244 return (EPROTONOSUPPORT);
1245 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS)
1246 return (EPROTONOSUPPORT);
1247 /*
1248 * Don't let the caller set up a VLAN VID with
1249 * anything except VLID bits.
1250 * VID numbers 0x0 and 0xFFF are reserved.
1251 */
1252 if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK))
1253 return (EINVAL);
1254 if (ifv->ifv_trunk)
1255 return (EBUSY);
1256
1257 if (p->if_vlantrunk == NULL) {
1258 trunk = malloc(sizeof(struct ifvlantrunk),
1259 M_VLAN, M_WAITOK | M_ZERO);
1260 vlan_inithash(trunk);
1261 VLAN_LOCK();
1262 if (p->if_vlantrunk != NULL) {
1263 /* A race that is very unlikely to be hit. */
1264 vlan_freehash(trunk);
1265 free(trunk, M_VLAN);
1266 goto exists;
1267 }
1268 TRUNK_LOCK_INIT(trunk);
1269 TRUNK_LOCK(trunk);
1270 p->if_vlantrunk = trunk;
1271 trunk->parent = p;
1272 } else {
1273 VLAN_LOCK();
1274 exists:
1275 trunk = p->if_vlantrunk;
1276 TRUNK_LOCK(trunk);
1277 }
1278
1279 ifv->ifv_vid = vid; /* must set this before vlan_inshash() */
1280 ifv->ifv_pcp = 0; /* Default: best effort delivery. */
1281 vlan_tag_recalculate(ifv);
1282 error = vlan_inshash(trunk, ifv);
1283 if (error)
1284 goto done;
1285 ifv->ifv_proto = ETHERTYPE_VLAN;
1286 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN;
1287 ifv->ifv_mintu = ETHERMIN;
1288 ifv->ifv_pflags = 0;
1289
1290 /*
1291 * If the parent supports the VLAN_MTU capability,
1292 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames,
1293 * use it.
1294 */
1295 if (p->if_capenable & IFCAP_VLAN_MTU) {
1296 /*
1297 * No need to fudge the MTU since the parent can
1298 * handle extended frames.
1299 */
1300 ifv->ifv_mtufudge = 0;
1301 } else {
1302 /*
1303 * Fudge the MTU by the encapsulation size. This
1304 * makes us incompatible with strictly compliant
1305 * 802.1Q implementations, but allows us to use
1306 * the feature with other NetBSD implementations,
1307 * which might still be useful.
1308 */
1309 ifv->ifv_mtufudge = ifv->ifv_encaplen;
1310 }
1311
1312 ifv->ifv_trunk = trunk;
1313 ifp = ifv->ifv_ifp;
1314 /*
1315 * Initialize fields from our parent. This duplicates some
1316 * work with ether_ifattach() but allows for non-ethernet
1317 * interfaces to also work.
1318 */
1319 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge;
1320 ifp->if_baudrate = p->if_baudrate;
1321 ifp->if_output = p->if_output;
1322 ifp->if_input = p->if_input;
1323 ifp->if_resolvemulti = p->if_resolvemulti;
1324 ifp->if_addrlen = p->if_addrlen;
1325 ifp->if_broadcastaddr = p->if_broadcastaddr;
1326
1327 /*
1328 * Copy only a selected subset of flags from the parent.
1329 * Other flags are none of our business.
1330 */
1331 #define VLAN_COPY_FLAGS (IFF_SIMPLEX)
1332 ifp->if_flags &= ~VLAN_COPY_FLAGS;
1333 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS;
1334 #undef VLAN_COPY_FLAGS
1335
1336 ifp->if_link_state = p->if_link_state;
1337
1338 vlan_capabilities(ifv);
1339
1340 /*
1341 * Set up our interface address to reflect the underlying
1342 * physical interface's.
1343 */
1344 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen);
1345 ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen =
1346 p->if_addrlen;
1347
1348 /*
1349 * Configure multicast addresses that may already be
1350 * joined on the vlan device.
1351 */
1352 (void)vlan_setmulti(ifp); /* XXX: VLAN lock held */
1353
1354 /* We are ready for operation now. */
1355 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1356 done:
1357 TRUNK_UNLOCK(trunk);
1358 if (error == 0)
1359 EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid);
1360 VLAN_UNLOCK();
1361
1362 return (error);
1363 }
1364
1365 static void
1366 vlan_unconfig(struct ifnet *ifp)
1367 {
1368
1369 VLAN_LOCK();
1370 vlan_unconfig_locked(ifp, 0);
1371 VLAN_UNLOCK();
1372 }
1373
1374 static void
1375 vlan_unconfig_locked(struct ifnet *ifp, int departing)
1376 {
1377 struct ifvlantrunk *trunk;
1378 struct vlan_mc_entry *mc;
1379 struct ifvlan *ifv;
1380 struct ifnet *parent;
1381 int error;
1382
1383 VLAN_LOCK_ASSERT();
1384
1385 ifv = ifp->if_softc;
1386 trunk = ifv->ifv_trunk;
1387 parent = NULL;
1388
1389 if (trunk != NULL) {
1390
1391 TRUNK_LOCK(trunk);
1392 parent = trunk->parent;
1393
1394 /*
1395 * Since the interface is being unconfigured, we need to
1396 * empty the list of multicast groups that we may have joined
1397 * while we were alive from the parent's list.
1398 */
1399 while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) {
1400 /*
1401 * If the parent interface is being detached,
1402 * all its multicast addresses have already
1403 * been removed. Warn about errors if
1404 * if_delmulti() does fail, but don't abort as
1405 * all callers expect vlan destruction to
1406 * succeed.
1407 */
1408 if (!departing) {
1409 error = if_delmulti(parent,
1410 (struct sockaddr *)&mc->mc_addr);
1411 if (error)
1412 if_printf(ifp,
1413 "Failed to delete multicast address from parent: %d\n",
1414 error);
1415 }
1416 SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries);
1417 free(mc, M_VLAN);
1418 }
1419
1420 vlan_setflags(ifp, 0); /* clear special flags on parent */
1421 vlan_remhash(trunk, ifv);
1422 ifv->ifv_trunk = NULL;
1423
1424 /*
1425 * Check if we were the last.
1426 */
1427 if (trunk->refcnt == 0) {
1428 parent->if_vlantrunk = NULL;
1429 /*
1430 * XXXGL: If some ithread has already entered
1431 * vlan_input() and is now blocked on the trunk
1432 * lock, then it should preempt us right after
1433 * unlock and finish its work. Then we will acquire
1434 * lock again in trunk_destroy().
1435 */
1436 TRUNK_UNLOCK(trunk);
1437 trunk_destroy(trunk);
1438 } else
1439 TRUNK_UNLOCK(trunk);
1440 }
1441
1442 /* Disconnect from parent. */
1443 if (ifv->ifv_pflags)
1444 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__);
1445 ifp->if_mtu = ETHERMTU;
1446 ifp->if_link_state = LINK_STATE_UNKNOWN;
1447 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1448
1449 /*
1450 * Only dispatch an event if vlan was
1451 * attached, otherwise there is nothing
1452 * to cleanup anyway.
1453 */
1454 if (parent != NULL)
1455 EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid);
1456 }
1457
1458 /* Handle a reference counted flag that should be set on the parent as well */
1459 static int
1460 vlan_setflag(struct ifnet *ifp, int flag, int status,
1461 int (*func)(struct ifnet *, int))
1462 {
1463 struct ifvlan *ifv;
1464 int error;
1465
1466 /* XXX VLAN_LOCK_ASSERT(); */
1467
1468 ifv = ifp->if_softc;
1469 status = status ? (ifp->if_flags & flag) : 0;
1470 /* Now "status" contains the flag value or 0 */
1471
1472 /*
1473 * See if recorded parent's status is different from what
1474 * we want it to be. If it is, flip it. We record parent's
1475 * status in ifv_pflags so that we won't clear parent's flag
1476 * we haven't set. In fact, we don't clear or set parent's
1477 * flags directly, but get or release references to them.
1478 * That's why we can be sure that recorded flags still are
1479 * in accord with actual parent's flags.
1480 */
1481 if (status != (ifv->ifv_pflags & flag)) {
1482 error = (*func)(PARENT(ifv), status);
1483 if (error)
1484 return (error);
1485 ifv->ifv_pflags &= ~flag;
1486 ifv->ifv_pflags |= status;
1487 }
1488 return (0);
1489 }
1490
1491 /*
1492 * Handle IFF_* flags that require certain changes on the parent:
1493 * if "status" is true, update parent's flags respective to our if_flags;
1494 * if "status" is false, forcedly clear the flags set on parent.
1495 */
1496 static int
1497 vlan_setflags(struct ifnet *ifp, int status)
1498 {
1499 int error, i;
1500
1501 for (i = 0; vlan_pflags[i].flag; i++) {
1502 error = vlan_setflag(ifp, vlan_pflags[i].flag,
1503 status, vlan_pflags[i].func);
1504 if (error)
1505 return (error);
1506 }
1507 return (0);
1508 }
1509
1510 /* Inform all vlans that their parent has changed link state */
1511 static void
1512 vlan_link_state(struct ifnet *ifp)
1513 {
1514 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1515 struct ifvlan *ifv;
1516 int i;
1517
1518 TRUNK_LOCK(trunk);
1519 #ifdef VLAN_ARRAY
1520 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1521 if (trunk->vlans[i] != NULL) {
1522 ifv = trunk->vlans[i];
1523 #else
1524 for (i = 0; i < (1 << trunk->hwidth); i++)
1525 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) {
1526 #endif
1527 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate;
1528 if_link_state_change(ifv->ifv_ifp,
1529 trunk->parent->if_link_state);
1530 }
1531 TRUNK_UNLOCK(trunk);
1532 }
1533
1534 static void
1535 vlan_capabilities(struct ifvlan *ifv)
1536 {
1537 struct ifnet *p = PARENT(ifv);
1538 struct ifnet *ifp = ifv->ifv_ifp;
1539 struct ifnet_hw_tsomax hw_tsomax;
1540
1541 TRUNK_LOCK_ASSERT(TRUNK(ifv));
1542
1543 /*
1544 * If the parent interface can do checksum offloading
1545 * on VLANs, then propagate its hardware-assisted
1546 * checksumming flags. Also assert that checksum
1547 * offloading requires hardware VLAN tagging.
1548 */
1549 if (p->if_capabilities & IFCAP_VLAN_HWCSUM)
1550 ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM;
1551
1552 if (p->if_capenable & IFCAP_VLAN_HWCSUM &&
1553 p->if_capenable & IFCAP_VLAN_HWTAGGING) {
1554 ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM;
1555 ifp->if_hwassist = p->if_hwassist & (CSUM_IP | CSUM_TCP |
1556 CSUM_UDP | CSUM_SCTP);
1557 } else {
1558 ifp->if_capenable = 0;
1559 ifp->if_hwassist = 0;
1560 }
1561 /*
1562 * If the parent interface can do TSO on VLANs then
1563 * propagate the hardware-assisted flag. TSO on VLANs
1564 * does not necessarily require hardware VLAN tagging.
1565 */
1566 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
1567 if_hw_tsomax_common(p, &hw_tsomax);
1568 if_hw_tsomax_update(ifp, &hw_tsomax);
1569 if (p->if_capabilities & IFCAP_VLAN_HWTSO)
1570 ifp->if_capabilities |= p->if_capabilities & IFCAP_TSO;
1571 if (p->if_capenable & IFCAP_VLAN_HWTSO) {
1572 ifp->if_capenable |= p->if_capenable & IFCAP_TSO;
1573 ifp->if_hwassist |= p->if_hwassist & CSUM_TSO;
1574 } else {
1575 ifp->if_capenable &= ~(p->if_capenable & IFCAP_TSO);
1576 ifp->if_hwassist &= ~(p->if_hwassist & CSUM_TSO);
1577 }
1578
1579 /*
1580 * If the parent interface can offload TCP connections over VLANs then
1581 * propagate its TOE capability to the VLAN interface.
1582 *
1583 * All TOE drivers in the tree today can deal with VLANs. If this
1584 * changes then IFCAP_VLAN_TOE should be promoted to a full capability
1585 * with its own bit.
1586 */
1587 #define IFCAP_VLAN_TOE IFCAP_TOE
1588 if (p->if_capabilities & IFCAP_VLAN_TOE)
1589 ifp->if_capabilities |= p->if_capabilities & IFCAP_TOE;
1590 if (p->if_capenable & IFCAP_VLAN_TOE) {
1591 TOEDEV(ifp) = TOEDEV(p);
1592 ifp->if_capenable |= p->if_capenable & IFCAP_TOE;
1593 }
1594 }
1595
1596 static void
1597 vlan_trunk_capabilities(struct ifnet *ifp)
1598 {
1599 struct ifvlantrunk *trunk = ifp->if_vlantrunk;
1600 struct ifvlan *ifv;
1601 int i;
1602
1603 TRUNK_LOCK(trunk);
1604 #ifdef VLAN_ARRAY
1605 for (i = 0; i < VLAN_ARRAY_SIZE; i++)
1606 if (trunk->vlans[i] != NULL) {
1607 ifv = trunk->vlans[i];
1608 #else
1609 for (i = 0; i < (1 << trunk->hwidth); i++) {
1610 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list)
1611 #endif
1612 vlan_capabilities(ifv);
1613 }
1614 TRUNK_UNLOCK(trunk);
1615 }
1616
1617 static int
1618 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1619 {
1620 struct ifnet *p;
1621 struct ifreq *ifr;
1622 struct ifaddr *ifa;
1623 struct ifvlan *ifv;
1624 struct ifvlantrunk *trunk;
1625 struct vlanreq vlr;
1626 int error = 0;
1627
1628 ifr = (struct ifreq *)data;
1629 ifa = (struct ifaddr *) data;
1630 ifv = ifp->if_softc;
1631
1632 switch (cmd) {
1633 case SIOCSIFADDR:
1634 ifp->if_flags |= IFF_UP;
1635 #ifdef INET
1636 if (ifa->ifa_addr->sa_family == AF_INET)
1637 arp_ifinit(ifp, ifa);
1638 #endif
1639 break;
1640 case SIOCGIFADDR:
1641 {
1642 struct sockaddr *sa;
1643
1644 sa = (struct sockaddr *)&ifr->ifr_data;
1645 bcopy(IF_LLADDR(ifp), sa->sa_data, ifp->if_addrlen);
1646 }
1647 break;
1648 case SIOCGIFMEDIA:
1649 VLAN_LOCK();
1650 if (TRUNK(ifv) != NULL) {
1651 p = PARENT(ifv);
1652 VLAN_UNLOCK();
1653 error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data);
1654 /* Limit the result to the parent's current config. */
1655 if (error == 0) {
1656 struct ifmediareq *ifmr;
1657
1658 ifmr = (struct ifmediareq *)data;
1659 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) {
1660 ifmr->ifm_count = 1;
1661 error = copyout(&ifmr->ifm_current,
1662 ifmr->ifm_ulist,
1663 sizeof(int));
1664 }
1665 }
1666 } else {
1667 VLAN_UNLOCK();
1668 error = EINVAL;
1669 }
1670 break;
1671
1672 case SIOCSIFMEDIA:
1673 error = EINVAL;
1674 break;
1675
1676 case SIOCSIFMTU:
1677 /*
1678 * Set the interface MTU.
1679 */
1680 VLAN_LOCK();
1681 if (TRUNK(ifv) != NULL) {
1682 if (ifr->ifr_mtu >
1683 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) ||
1684 ifr->ifr_mtu <
1685 (ifv->ifv_mintu - ifv->ifv_mtufudge))
1686 error = EINVAL;
1687 else
1688 ifp->if_mtu = ifr->ifr_mtu;
1689 } else
1690 error = EINVAL;
1691 VLAN_UNLOCK();
1692 break;
1693
1694 case SIOCSETVLAN:
1695 #ifdef VIMAGE
1696 /*
1697 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN
1698 * interface to be delegated to a jail without allowing the
1699 * jail to change what underlying interface/VID it is
1700 * associated with. We are not entirely convinced that this
1701 * is the right way to accomplish that policy goal.
1702 */
1703 if (ifp->if_vnet != ifp->if_home_vnet) {
1704 error = EPERM;
1705 break;
1706 }
1707 #endif
1708 error = copyin(ifr->ifr_data, &vlr, sizeof(vlr));
1709 if (error)
1710 break;
1711 if (vlr.vlr_parent[0] == '\0') {
1712 vlan_unconfig(ifp);
1713 break;
1714 }
1715 p = ifunit(vlr.vlr_parent);
1716 if (p == NULL) {
1717 error = ENOENT;
1718 break;
1719 }
1720 error = vlan_config(ifv, p, vlr.vlr_tag);
1721 if (error)
1722 break;
1723
1724 /* Update flags on the parent, if necessary. */
1725 vlan_setflags(ifp, 1);
1726 break;
1727
1728 case SIOCGETVLAN:
1729 #ifdef VIMAGE
1730 if (ifp->if_vnet != ifp->if_home_vnet) {
1731 error = EPERM;
1732 break;
1733 }
1734 #endif
1735 bzero(&vlr, sizeof(vlr));
1736 VLAN_LOCK();
1737 if (TRUNK(ifv) != NULL) {
1738 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname,
1739 sizeof(vlr.vlr_parent));
1740 vlr.vlr_tag = ifv->ifv_vid;
1741 }
1742 VLAN_UNLOCK();
1743 error = copyout(&vlr, ifr->ifr_data, sizeof(vlr));
1744 break;
1745
1746 case SIOCSIFFLAGS:
1747 /*
1748 * We should propagate selected flags to the parent,
1749 * e.g., promiscuous mode.
1750 */
1751 if (TRUNK(ifv) != NULL)
1752 error = vlan_setflags(ifp, 1);
1753 break;
1754
1755 case SIOCADDMULTI:
1756 case SIOCDELMULTI:
1757 /*
1758 * If we don't have a parent, just remember the membership for
1759 * when we do.
1760 */
1761 trunk = TRUNK(ifv);
1762 if (trunk != NULL) {
1763 TRUNK_LOCK(trunk);
1764 error = vlan_setmulti(ifp);
1765 TRUNK_UNLOCK(trunk);
1766 }
1767 break;
1768
1769 case SIOCGVLANPCP:
1770 #ifdef VIMAGE
1771 if (ifp->if_vnet != ifp->if_home_vnet) {
1772 error = EPERM;
1773 break;
1774 }
1775 #endif
1776 ifr->ifr_vlan_pcp = ifv->ifv_pcp;
1777 break;
1778
1779 case SIOCSVLANPCP:
1780 #ifdef VIMAGE
1781 if (ifp->if_vnet != ifp->if_home_vnet) {
1782 error = EPERM;
1783 break;
1784 }
1785 #endif
1786 error = priv_check(curthread, PRIV_NET_SETVLANPCP);
1787 if (error)
1788 break;
1789 if (ifr->ifr_vlan_pcp > 7) {
1790 error = EINVAL;
1791 break;
1792 }
1793 ifv->ifv_pcp = ifr->ifr_vlan_pcp;
1794 vlan_tag_recalculate(ifv);
1795 break;
1796
1797 default:
1798 error = EINVAL;
1799 break;
1800 }
1801
1802 return (error);
1803 }
Cache object: 9e01dc5dd621d55026c18957b75d2a44
|