if_vlan.c revision 239519
1/*- 2 * Copyright 1998 Massachusetts Institute of Technology 3 * 4 * Permission to use, copy, modify, and distribute this software and 5 * its documentation for any purpose and without fee is hereby 6 * granted, provided that both the above copyright notice and this 7 * permission notice appear in all copies, that both the above 8 * copyright notice and this permission notice appear in all 9 * supporting documentation, and that the name of M.I.T. not be used 10 * in advertising or publicity pertaining to distribution of the 11 * software without specific, written prior permission. M.I.T. makes 12 * no representations about the suitability of this software for any 13 * purpose. It is provided "as is" without express or implied 14 * warranty. 15 * 16 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS 17 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE, 18 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 19 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT 20 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF 23 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND 24 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 25 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 26 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30/* 31 * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs. 32 * Might be extended some day to also handle IEEE 802.1p priority 33 * tagging. This is sort of sneaky in the implementation, since 34 * we need to pretend to be enough of an Ethernet implementation 35 * to make arp work. The way we do this is by telling everyone 36 * that we are an Ethernet, and then catch the packets that 37 * ether_output() sends to us via if_transmit(), rewrite them for 38 * use by the real outgoing interface, and ask it to send them. 39 */ 40 41#include <sys/cdefs.h> 42__FBSDID("$FreeBSD: head/sys/net/if_vlan.c 239519 2012-08-21 19:07:28Z jhb $"); 43 44#include "opt_vlan.h" 45 46#include <sys/param.h> 47#include <sys/kernel.h> 48#include <sys/lock.h> 49#include <sys/malloc.h> 50#include <sys/mbuf.h> 51#include <sys/module.h> 52#include <sys/rwlock.h> 53#include <sys/queue.h> 54#include <sys/socket.h> 55#include <sys/sockio.h> 56#include <sys/sysctl.h> 57#include <sys/systm.h> 58#include <sys/sx.h> 59 60#include <net/bpf.h> 61#include <net/ethernet.h> 62#include <net/if.h> 63#include <net/if_clone.h> 64#include <net/if_dl.h> 65#include <net/if_types.h> 66#include <net/if_vlan_var.h> 67#include <net/vnet.h> 68 69#define VLANNAME "vlan" 70#define VLAN_DEF_HWIDTH 4 71#define VLAN_IFFLAGS (IFF_BROADCAST | IFF_MULTICAST) 72 73#define UP_AND_RUNNING(ifp) \ 74 ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING) 75 76LIST_HEAD(ifvlanhead, ifvlan); 77 78struct ifvlantrunk { 79 struct ifnet *parent; /* parent interface of this trunk */ 80 struct rwlock rw; 81#ifdef VLAN_ARRAY 82#define VLAN_ARRAY_SIZE (EVL_VLID_MASK + 1) 83 struct ifvlan *vlans[VLAN_ARRAY_SIZE]; /* static table */ 84#else 85 struct ifvlanhead *hash; /* dynamic hash-list table */ 86 uint16_t hmask; 87 uint16_t hwidth; 88#endif 89 int refcnt; 90}; 91 92struct vlan_mc_entry { 93 struct sockaddr_dl mc_addr; 94 SLIST_ENTRY(vlan_mc_entry) mc_entries; 95}; 96 97struct ifvlan { 98 struct ifvlantrunk *ifv_trunk; 99 struct ifnet *ifv_ifp; 100 void *ifv_cookie; 101#define TRUNK(ifv) ((ifv)->ifv_trunk) 102#define PARENT(ifv) ((ifv)->ifv_trunk->parent) 103 int ifv_pflags; /* special flags we have set on parent */ 104 struct ifv_linkmib { 105 int ifvm_encaplen; /* encapsulation length */ 106 int ifvm_mtufudge; /* MTU fudged by this much */ 107 int ifvm_mintu; /* min transmission unit */ 108 uint16_t ifvm_proto; /* encapsulation ethertype */ 109 uint16_t ifvm_tag; /* tag to apply on packets leaving if */ 110 } ifv_mib; 111 SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead; 112#ifndef VLAN_ARRAY 113 LIST_ENTRY(ifvlan) ifv_list; 114#endif 115}; 116#define ifv_proto ifv_mib.ifvm_proto 117#define ifv_vid ifv_mib.ifvm_tag 118#define ifv_encaplen ifv_mib.ifvm_encaplen 119#define ifv_mtufudge ifv_mib.ifvm_mtufudge 120#define ifv_mintu ifv_mib.ifvm_mintu 121 122/* Special flags we should propagate to parent. */ 123static struct { 124 int flag; 125 int (*func)(struct ifnet *, int); 126} vlan_pflags[] = { 127 {IFF_PROMISC, ifpromisc}, 128 {IFF_ALLMULTI, if_allmulti}, 129 {0, NULL} 130}; 131 132SYSCTL_DECL(_net_link); 133static SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0, 134 "IEEE 802.1Q VLAN"); 135static SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0, 136 "for consistency"); 137 138static int soft_pad = 0; 139SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW, &soft_pad, 0, 140 "pad short frames before tagging"); 141 142static MALLOC_DEFINE(M_VLAN, VLANNAME, "802.1Q Virtual LAN Interface"); 143 144static eventhandler_tag ifdetach_tag; 145static eventhandler_tag iflladdr_tag; 146 147/* 148 * We have a global mutex, that is used to serialize configuration 149 * changes and isn't used in normal packet delivery. 150 * 151 * We also have a per-trunk rwlock, that is locked shared on packet 152 * processing and exclusive when configuration is changed. 153 * 154 * The VLAN_ARRAY substitutes the dynamic hash with a static array 155 * with 4096 entries. In theory this can give a boost in processing, 156 * however on practice it does not. Probably this is because array 157 * is too big to fit into CPU cache. 158 */ 159static struct sx ifv_lock; 160#define VLAN_LOCK_INIT() sx_init(&ifv_lock, "vlan_global") 161#define VLAN_LOCK_DESTROY() sx_destroy(&ifv_lock) 162#define VLAN_LOCK_ASSERT() sx_assert(&ifv_lock, SA_LOCKED) 163#define VLAN_LOCK() sx_xlock(&ifv_lock) 164#define VLAN_UNLOCK() sx_xunlock(&ifv_lock) 165#define TRUNK_LOCK_INIT(trunk) rw_init(&(trunk)->rw, VLANNAME) 166#define TRUNK_LOCK_DESTROY(trunk) rw_destroy(&(trunk)->rw) 167#define TRUNK_LOCK(trunk) rw_wlock(&(trunk)->rw) 168#define TRUNK_UNLOCK(trunk) rw_wunlock(&(trunk)->rw) 169#define TRUNK_LOCK_ASSERT(trunk) rw_assert(&(trunk)->rw, RA_WLOCKED) 170#define TRUNK_RLOCK(trunk) rw_rlock(&(trunk)->rw) 171#define TRUNK_RUNLOCK(trunk) rw_runlock(&(trunk)->rw) 172#define TRUNK_LOCK_RASSERT(trunk) rw_assert(&(trunk)->rw, RA_RLOCKED) 173 174#ifndef VLAN_ARRAY 175static void vlan_inithash(struct ifvlantrunk *trunk); 176static void vlan_freehash(struct ifvlantrunk *trunk); 177static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv); 178static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv); 179static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch); 180static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk, 181 uint16_t vid); 182#endif 183static void trunk_destroy(struct ifvlantrunk *trunk); 184 185static void vlan_init(void *foo); 186static void vlan_input(struct ifnet *ifp, struct mbuf *m); 187static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr); 188static void vlan_qflush(struct ifnet *ifp); 189static int vlan_setflag(struct ifnet *ifp, int flag, int status, 190 int (*func)(struct ifnet *, int)); 191static int vlan_setflags(struct ifnet *ifp, int status); 192static int vlan_setmulti(struct ifnet *ifp); 193static int vlan_transmit(struct ifnet *ifp, struct mbuf *m); 194static void vlan_unconfig(struct ifnet *ifp); 195static void vlan_unconfig_locked(struct ifnet *ifp, int departing); 196static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag); 197static void vlan_link_state(struct ifnet *ifp); 198static void vlan_capabilities(struct ifvlan *ifv); 199static void vlan_trunk_capabilities(struct ifnet *ifp); 200 201static struct ifnet *vlan_clone_match_ethervid(struct if_clone *, 202 const char *, int *); 203static int vlan_clone_match(struct if_clone *, const char *); 204static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t); 205static int vlan_clone_destroy(struct if_clone *, struct ifnet *); 206 207static void vlan_ifdetach(void *arg, struct ifnet *ifp); 208static void vlan_iflladdr(void *arg, struct ifnet *ifp); 209 210static struct if_clone vlan_cloner = IFC_CLONE_INITIALIZER(VLANNAME, NULL, 211 IF_MAXUNIT, NULL, vlan_clone_match, vlan_clone_create, vlan_clone_destroy); 212 213#ifdef VIMAGE 214static VNET_DEFINE(struct if_clone, vlan_cloner); 215#define V_vlan_cloner VNET(vlan_cloner) 216#endif 217 218#ifndef VLAN_ARRAY 219#define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m)) 220 221static void 222vlan_inithash(struct ifvlantrunk *trunk) 223{ 224 int i, n; 225 226 /* 227 * The trunk must not be locked here since we call malloc(M_WAITOK). 228 * It is OK in case this function is called before the trunk struct 229 * gets hooked up and becomes visible from other threads. 230 */ 231 232 KASSERT(trunk->hwidth == 0 && trunk->hash == NULL, 233 ("%s: hash already initialized", __func__)); 234 235 trunk->hwidth = VLAN_DEF_HWIDTH; 236 n = 1 << trunk->hwidth; 237 trunk->hmask = n - 1; 238 trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK); 239 for (i = 0; i < n; i++) 240 LIST_INIT(&trunk->hash[i]); 241} 242 243static void 244vlan_freehash(struct ifvlantrunk *trunk) 245{ 246#ifdef INVARIANTS 247 int i; 248 249 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__)); 250 for (i = 0; i < (1 << trunk->hwidth); i++) 251 KASSERT(LIST_EMPTY(&trunk->hash[i]), 252 ("%s: hash table not empty", __func__)); 253#endif 254 free(trunk->hash, M_VLAN); 255 trunk->hash = NULL; 256 trunk->hwidth = trunk->hmask = 0; 257} 258 259static int 260vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv) 261{ 262 int i, b; 263 struct ifvlan *ifv2; 264 265 TRUNK_LOCK_ASSERT(trunk); 266 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__)); 267 268 b = 1 << trunk->hwidth; 269 i = HASH(ifv->ifv_vid, trunk->hmask); 270 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list) 271 if (ifv->ifv_vid == ifv2->ifv_vid) 272 return (EEXIST); 273 274 /* 275 * Grow the hash when the number of vlans exceeds half of the number of 276 * hash buckets squared. This will make the average linked-list length 277 * buckets/2. 278 */ 279 if (trunk->refcnt > (b * b) / 2) { 280 vlan_growhash(trunk, 1); 281 i = HASH(ifv->ifv_vid, trunk->hmask); 282 } 283 LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list); 284 trunk->refcnt++; 285 286 return (0); 287} 288 289static int 290vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv) 291{ 292 int i, b; 293 struct ifvlan *ifv2; 294 295 TRUNK_LOCK_ASSERT(trunk); 296 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__)); 297 298 b = 1 << trunk->hwidth; 299 i = HASH(ifv->ifv_vid, trunk->hmask); 300 LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list) 301 if (ifv2 == ifv) { 302 trunk->refcnt--; 303 LIST_REMOVE(ifv2, ifv_list); 304 if (trunk->refcnt < (b * b) / 2) 305 vlan_growhash(trunk, -1); 306 return (0); 307 } 308 309 panic("%s: vlan not found\n", __func__); 310 return (ENOENT); /*NOTREACHED*/ 311} 312 313/* 314 * Grow the hash larger or smaller if memory permits. 315 */ 316static void 317vlan_growhash(struct ifvlantrunk *trunk, int howmuch) 318{ 319 struct ifvlan *ifv; 320 struct ifvlanhead *hash2; 321 int hwidth2, i, j, n, n2; 322 323 TRUNK_LOCK_ASSERT(trunk); 324 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__)); 325 326 if (howmuch == 0) { 327 /* Harmless yet obvious coding error */ 328 printf("%s: howmuch is 0\n", __func__); 329 return; 330 } 331 332 hwidth2 = trunk->hwidth + howmuch; 333 n = 1 << trunk->hwidth; 334 n2 = 1 << hwidth2; 335 /* Do not shrink the table below the default */ 336 if (hwidth2 < VLAN_DEF_HWIDTH) 337 return; 338 339 /* M_NOWAIT because we're called with trunk mutex held */ 340 hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT); 341 if (hash2 == NULL) { 342 printf("%s: out of memory -- hash size not changed\n", 343 __func__); 344 return; /* We can live with the old hash table */ 345 } 346 for (j = 0; j < n2; j++) 347 LIST_INIT(&hash2[j]); 348 for (i = 0; i < n; i++) 349 while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) { 350 LIST_REMOVE(ifv, ifv_list); 351 j = HASH(ifv->ifv_vid, n2 - 1); 352 LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list); 353 } 354 free(trunk->hash, M_VLAN); 355 trunk->hash = hash2; 356 trunk->hwidth = hwidth2; 357 trunk->hmask = n2 - 1; 358 359 if (bootverbose) 360 if_printf(trunk->parent, 361 "VLAN hash table resized from %d to %d buckets\n", n, n2); 362} 363 364static __inline struct ifvlan * 365vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid) 366{ 367 struct ifvlan *ifv; 368 369 TRUNK_LOCK_RASSERT(trunk); 370 371 LIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list) 372 if (ifv->ifv_vid == vid) 373 return (ifv); 374 return (NULL); 375} 376 377#if 0 378/* Debugging code to view the hashtables. */ 379static void 380vlan_dumphash(struct ifvlantrunk *trunk) 381{ 382 int i; 383 struct ifvlan *ifv; 384 385 for (i = 0; i < (1 << trunk->hwidth); i++) { 386 printf("%d: ", i); 387 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) 388 printf("%s ", ifv->ifv_ifp->if_xname); 389 printf("\n"); 390 } 391} 392#endif /* 0 */ 393#else 394 395static __inline struct ifvlan * 396vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid) 397{ 398 399 return trunk->vlans[vid]; 400} 401 402static __inline int 403vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv) 404{ 405 406 if (trunk->vlans[ifv->ifv_vid] != NULL) 407 return EEXIST; 408 trunk->vlans[ifv->ifv_vid] = ifv; 409 trunk->refcnt++; 410 411 return (0); 412} 413 414static __inline int 415vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv) 416{ 417 418 trunk->vlans[ifv->ifv_vid] = NULL; 419 trunk->refcnt--; 420 421 return (0); 422} 423 424static __inline void 425vlan_freehash(struct ifvlantrunk *trunk) 426{ 427} 428 429static __inline void 430vlan_inithash(struct ifvlantrunk *trunk) 431{ 432} 433 434#endif /* !VLAN_ARRAY */ 435 436static void 437trunk_destroy(struct ifvlantrunk *trunk) 438{ 439 VLAN_LOCK_ASSERT(); 440 441 TRUNK_LOCK(trunk); 442 vlan_freehash(trunk); 443 trunk->parent->if_vlantrunk = NULL; 444 TRUNK_UNLOCK(trunk); 445 TRUNK_LOCK_DESTROY(trunk); 446 free(trunk, M_VLAN); 447} 448 449/* 450 * Program our multicast filter. What we're actually doing is 451 * programming the multicast filter of the parent. This has the 452 * side effect of causing the parent interface to receive multicast 453 * traffic that it doesn't really want, which ends up being discarded 454 * later by the upper protocol layers. Unfortunately, there's no way 455 * to avoid this: there really is only one physical interface. 456 * 457 * XXX: There is a possible race here if more than one thread is 458 * modifying the multicast state of the vlan interface at the same time. 459 */ 460static int 461vlan_setmulti(struct ifnet *ifp) 462{ 463 struct ifnet *ifp_p; 464 struct ifmultiaddr *ifma, *rifma = NULL; 465 struct ifvlan *sc; 466 struct vlan_mc_entry *mc; 467 int error; 468 469 /*VLAN_LOCK_ASSERT();*/ 470 471 /* Find the parent. */ 472 sc = ifp->if_softc; 473 ifp_p = PARENT(sc); 474 475 CURVNET_SET_QUIET(ifp_p->if_vnet); 476 477 /* First, remove any existing filter entries. */ 478 while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) { 479 error = if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr); 480 if (error) 481 return (error); 482 SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries); 483 free(mc, M_VLAN); 484 } 485 486 /* Now program new ones. */ 487 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 488 if (ifma->ifma_addr->sa_family != AF_LINK) 489 continue; 490 mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT); 491 if (mc == NULL) 492 return (ENOMEM); 493 bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len); 494 mc->mc_addr.sdl_index = ifp_p->if_index; 495 SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries); 496 error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr, 497 &rifma); 498 if (error) 499 return (error); 500 } 501 502 CURVNET_RESTORE(); 503 return (0); 504} 505 506/* 507 * A handler for parent interface link layer address changes. 508 * If the parent interface link layer address is changed we 509 * should also change it on all children vlans. 510 */ 511static void 512vlan_iflladdr(void *arg __unused, struct ifnet *ifp) 513{ 514 struct ifvlan *ifv; 515#ifndef VLAN_ARRAY 516 struct ifvlan *next; 517#endif 518 int i; 519 520 /* 521 * Check if it's a trunk interface first of all 522 * to avoid needless locking. 523 */ 524 if (ifp->if_vlantrunk == NULL) 525 return; 526 527 VLAN_LOCK(); 528 /* 529 * OK, it's a trunk. Loop over and change all vlan's lladdrs on it. 530 */ 531#ifdef VLAN_ARRAY 532 for (i = 0; i < VLAN_ARRAY_SIZE; i++) 533 if ((ifv = ifp->if_vlantrunk->vlans[i])) { 534#else /* VLAN_ARRAY */ 535 for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++) 536 LIST_FOREACH_SAFE(ifv, &ifp->if_vlantrunk->hash[i], ifv_list, next) { 537#endif /* VLAN_ARRAY */ 538 VLAN_UNLOCK(); 539 if_setlladdr(ifv->ifv_ifp, IF_LLADDR(ifp), 540 ifp->if_addrlen); 541 VLAN_LOCK(); 542 } 543 VLAN_UNLOCK(); 544 545} 546 547/* 548 * A handler for network interface departure events. 549 * Track departure of trunks here so that we don't access invalid 550 * pointers or whatever if a trunk is ripped from under us, e.g., 551 * by ejecting its hot-plug card. However, if an ifnet is simply 552 * being renamed, then there's no need to tear down the state. 553 */ 554static void 555vlan_ifdetach(void *arg __unused, struct ifnet *ifp) 556{ 557 struct ifvlan *ifv; 558 int i; 559 560 /* 561 * Check if it's a trunk interface first of all 562 * to avoid needless locking. 563 */ 564 if (ifp->if_vlantrunk == NULL) 565 return; 566 567 /* If the ifnet is just being renamed, don't do anything. */ 568 if (ifp->if_flags & IFF_RENAMING) 569 return; 570 571 VLAN_LOCK(); 572 /* 573 * OK, it's a trunk. Loop over and detach all vlan's on it. 574 * Check trunk pointer after each vlan_unconfig() as it will 575 * free it and set to NULL after the last vlan was detached. 576 */ 577#ifdef VLAN_ARRAY 578 for (i = 0; i < VLAN_ARRAY_SIZE; i++) 579 if ((ifv = ifp->if_vlantrunk->vlans[i])) { 580 vlan_unconfig_locked(ifv->ifv_ifp, 1); 581 if (ifp->if_vlantrunk == NULL) 582 break; 583 } 584#else /* VLAN_ARRAY */ 585restart: 586 for (i = 0; i < (1 << ifp->if_vlantrunk->hwidth); i++) 587 if ((ifv = LIST_FIRST(&ifp->if_vlantrunk->hash[i]))) { 588 vlan_unconfig_locked(ifv->ifv_ifp, 1); 589 if (ifp->if_vlantrunk) 590 goto restart; /* trunk->hwidth can change */ 591 else 592 break; 593 } 594#endif /* VLAN_ARRAY */ 595 /* Trunk should have been destroyed in vlan_unconfig(). */ 596 KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__)); 597 VLAN_UNLOCK(); 598} 599 600/* 601 * Return the trunk device for a virtual interface. 602 */ 603static struct ifnet * 604vlan_trunkdev(struct ifnet *ifp) 605{ 606 struct ifvlan *ifv; 607 608 if (ifp->if_type != IFT_L2VLAN) 609 return (NULL); 610 ifv = ifp->if_softc; 611 ifp = NULL; 612 VLAN_LOCK(); 613 if (ifv->ifv_trunk) 614 ifp = PARENT(ifv); 615 VLAN_UNLOCK(); 616 return (ifp); 617} 618 619/* 620 * Return the 12-bit VLAN VID for this interface, for use by external 621 * components such as Infiniband. 622 * 623 * XXXRW: Note that the function name here is historical; it should be named 624 * vlan_vid(). 625 */ 626static int 627vlan_tag(struct ifnet *ifp, uint16_t *vidp) 628{ 629 struct ifvlan *ifv; 630 631 if (ifp->if_type != IFT_L2VLAN) 632 return (EINVAL); 633 ifv = ifp->if_softc; 634 *vidp = ifv->ifv_vid; 635 return (0); 636} 637 638/* 639 * Return a driver specific cookie for this interface. Synchronization 640 * with setcookie must be provided by the driver. 641 */ 642static void * 643vlan_cookie(struct ifnet *ifp) 644{ 645 struct ifvlan *ifv; 646 647 if (ifp->if_type != IFT_L2VLAN) 648 return (NULL); 649 ifv = ifp->if_softc; 650 return (ifv->ifv_cookie); 651} 652 653/* 654 * Store a cookie in our softc that drivers can use to store driver 655 * private per-instance data in. 656 */ 657static int 658vlan_setcookie(struct ifnet *ifp, void *cookie) 659{ 660 struct ifvlan *ifv; 661 662 if (ifp->if_type != IFT_L2VLAN) 663 return (EINVAL); 664 ifv = ifp->if_softc; 665 ifv->ifv_cookie = cookie; 666 return (0); 667} 668 669/* 670 * Return the vlan device present at the specific VID. 671 */ 672static struct ifnet * 673vlan_devat(struct ifnet *ifp, uint16_t vid) 674{ 675 struct ifvlantrunk *trunk; 676 struct ifvlan *ifv; 677 678 trunk = ifp->if_vlantrunk; 679 if (trunk == NULL) 680 return (NULL); 681 ifp = NULL; 682 TRUNK_RLOCK(trunk); 683 ifv = vlan_gethash(trunk, vid); 684 if (ifv) 685 ifp = ifv->ifv_ifp; 686 TRUNK_RUNLOCK(trunk); 687 return (ifp); 688} 689 690/* 691 * VLAN support can be loaded as a module. The only place in the 692 * system that's intimately aware of this is ether_input. We hook 693 * into this code through vlan_input_p which is defined there and 694 * set here. Noone else in the system should be aware of this so 695 * we use an explicit reference here. 696 */ 697extern void (*vlan_input_p)(struct ifnet *, struct mbuf *); 698 699/* For if_link_state_change() eyes only... */ 700extern void (*vlan_link_state_p)(struct ifnet *); 701 702static int 703vlan_modevent(module_t mod, int type, void *data) 704{ 705 706 switch (type) { 707 case MOD_LOAD: 708 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event, 709 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY); 710 if (ifdetach_tag == NULL) 711 return (ENOMEM); 712 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event, 713 vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY); 714 if (iflladdr_tag == NULL) 715 return (ENOMEM); 716 VLAN_LOCK_INIT(); 717 vlan_input_p = vlan_input; 718 vlan_link_state_p = vlan_link_state; 719 vlan_trunk_cap_p = vlan_trunk_capabilities; 720 vlan_trunkdev_p = vlan_trunkdev; 721 vlan_cookie_p = vlan_cookie; 722 vlan_setcookie_p = vlan_setcookie; 723 vlan_tag_p = vlan_tag; 724 vlan_devat_p = vlan_devat; 725#ifndef VIMAGE 726 if_clone_attach(&vlan_cloner); 727#endif 728 if (bootverbose) 729 printf("vlan: initialized, using " 730#ifdef VLAN_ARRAY 731 "full-size arrays" 732#else 733 "hash tables with chaining" 734#endif 735 736 "\n"); 737 break; 738 case MOD_UNLOAD: 739#ifndef VIMAGE 740 if_clone_detach(&vlan_cloner); 741#endif 742 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag); 743 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag); 744 vlan_input_p = NULL; 745 vlan_link_state_p = NULL; 746 vlan_trunk_cap_p = NULL; 747 vlan_trunkdev_p = NULL; 748 vlan_tag_p = NULL; 749 vlan_cookie_p = NULL; 750 vlan_setcookie_p = NULL; 751 vlan_devat_p = NULL; 752 VLAN_LOCK_DESTROY(); 753 if (bootverbose) 754 printf("vlan: unloaded\n"); 755 break; 756 default: 757 return (EOPNOTSUPP); 758 } 759 return (0); 760} 761 762static moduledata_t vlan_mod = { 763 "if_vlan", 764 vlan_modevent, 765 0 766}; 767 768DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 769MODULE_VERSION(if_vlan, 3); 770 771#ifdef VIMAGE 772static void 773vnet_vlan_init(const void *unused __unused) 774{ 775 776 V_vlan_cloner = vlan_cloner; 777 if_clone_attach(&V_vlan_cloner); 778} 779VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 780 vnet_vlan_init, NULL); 781 782static void 783vnet_vlan_uninit(const void *unused __unused) 784{ 785 786 if_clone_detach(&V_vlan_cloner); 787} 788VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST, 789 vnet_vlan_uninit, NULL); 790#endif 791 792static struct ifnet * 793vlan_clone_match_ethervid(struct if_clone *ifc, const char *name, int *vidp) 794{ 795 const char *cp; 796 struct ifnet *ifp; 797 int vid; 798 799 /* Check for <etherif>.<vlan> style interface names. */ 800 IFNET_RLOCK_NOSLEEP(); 801 TAILQ_FOREACH(ifp, &V_ifnet, if_link) { 802 /* 803 * We can handle non-ethernet hardware types as long as 804 * they handle the tagging and headers themselves. 805 */ 806 if (ifp->if_type != IFT_ETHER && 807 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0) 808 continue; 809 if (strncmp(ifp->if_xname, name, strlen(ifp->if_xname)) != 0) 810 continue; 811 cp = name + strlen(ifp->if_xname); 812 if (*cp++ != '.') 813 continue; 814 if (*cp == '\0') 815 continue; 816 vid = 0; 817 for(; *cp >= '0' && *cp <= '9'; cp++) 818 vid = (vid * 10) + (*cp - '0'); 819 if (*cp != '\0') 820 continue; 821 if (vidp != NULL) 822 *vidp = vid; 823 break; 824 } 825 IFNET_RUNLOCK_NOSLEEP(); 826 827 return (ifp); 828} 829 830static int 831vlan_clone_match(struct if_clone *ifc, const char *name) 832{ 833 const char *cp; 834 835 if (vlan_clone_match_ethervid(ifc, name, NULL) != NULL) 836 return (1); 837 838 if (strncmp(VLANNAME, name, strlen(VLANNAME)) != 0) 839 return (0); 840 for (cp = name + 4; *cp != '\0'; cp++) { 841 if (*cp < '0' || *cp > '9') 842 return (0); 843 } 844 845 return (1); 846} 847 848static int 849vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params) 850{ 851 char *dp; 852 int wildcard; 853 int unit; 854 int error; 855 int vid; 856 int ethertag; 857 struct ifvlan *ifv; 858 struct ifnet *ifp; 859 struct ifnet *p; 860 struct ifaddr *ifa; 861 struct sockaddr_dl *sdl; 862 struct vlanreq vlr; 863 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */ 864 865 /* 866 * There are 3 (ugh) ways to specify the cloned device: 867 * o pass a parameter block with the clone request. 868 * o specify parameters in the text of the clone device name 869 * o specify no parameters and get an unattached device that 870 * must be configured separately. 871 * The first technique is preferred; the latter two are 872 * supported for backwards compatibilty. 873 * 874 * XXXRW: Note historic use of the word "tag" here. New ioctls may be 875 * called for. 876 */ 877 if (params) { 878 error = copyin(params, &vlr, sizeof(vlr)); 879 if (error) 880 return error; 881 p = ifunit(vlr.vlr_parent); 882 if (p == NULL) 883 return ENXIO; 884 /* 885 * Don't let the caller set up a VLAN VID with 886 * anything except VLID bits. 887 */ 888 if (vlr.vlr_tag & ~EVL_VLID_MASK) 889 return (EINVAL); 890 error = ifc_name2unit(name, &unit); 891 if (error != 0) 892 return (error); 893 894 ethertag = 1; 895 vid = vlr.vlr_tag; 896 wildcard = (unit < 0); 897 } else if ((p = vlan_clone_match_ethervid(ifc, name, &vid)) != NULL) { 898 ethertag = 1; 899 unit = -1; 900 wildcard = 0; 901 902 /* 903 * Don't let the caller set up a VLAN VID with 904 * anything except VLID bits. 905 */ 906 if (vid & ~EVL_VLID_MASK) 907 return (EINVAL); 908 } else { 909 ethertag = 0; 910 911 error = ifc_name2unit(name, &unit); 912 if (error != 0) 913 return (error); 914 915 wildcard = (unit < 0); 916 } 917 918 error = ifc_alloc_unit(ifc, &unit); 919 if (error != 0) 920 return (error); 921 922 /* In the wildcard case, we need to update the name. */ 923 if (wildcard) { 924 for (dp = name; *dp != '\0'; dp++); 925 if (snprintf(dp, len - (dp-name), "%d", unit) > 926 len - (dp-name) - 1) { 927 panic("%s: interface name too long", __func__); 928 } 929 } 930 931 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO); 932 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER); 933 if (ifp == NULL) { 934 ifc_free_unit(ifc, unit); 935 free(ifv, M_VLAN); 936 return (ENOSPC); 937 } 938 SLIST_INIT(&ifv->vlan_mc_listhead); 939 940 ifp->if_softc = ifv; 941 /* 942 * Set the name manually rather than using if_initname because 943 * we don't conform to the default naming convention for interfaces. 944 */ 945 strlcpy(ifp->if_xname, name, IFNAMSIZ); 946 ifp->if_dname = ifc->ifc_name; 947 ifp->if_dunit = unit; 948 /* NB: flags are not set here */ 949 ifp->if_linkmib = &ifv->ifv_mib; 950 ifp->if_linkmiblen = sizeof(ifv->ifv_mib); 951 /* NB: mtu is not set here */ 952 953 ifp->if_init = vlan_init; 954 ifp->if_transmit = vlan_transmit; 955 ifp->if_qflush = vlan_qflush; 956 ifp->if_ioctl = vlan_ioctl; 957 ifp->if_flags = VLAN_IFFLAGS; 958 ether_ifattach(ifp, eaddr); 959 /* Now undo some of the damage... */ 960 ifp->if_baudrate = 0; 961 ifp->if_type = IFT_L2VLAN; 962 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN; 963 ifa = ifp->if_addr; 964 sdl = (struct sockaddr_dl *)ifa->ifa_addr; 965 sdl->sdl_type = IFT_L2VLAN; 966 967 if (ethertag) { 968 error = vlan_config(ifv, p, vid); 969 if (error != 0) { 970 /* 971 * Since we've partially failed, we need to back 972 * out all the way, otherwise userland could get 973 * confused. Thus, we destroy the interface. 974 */ 975 ether_ifdetach(ifp); 976 vlan_unconfig(ifp); 977 if_free(ifp); 978 ifc_free_unit(ifc, unit); 979 free(ifv, M_VLAN); 980 981 return (error); 982 } 983 984 /* Update flags on the parent, if necessary. */ 985 vlan_setflags(ifp, 1); 986 } 987 988 return (0); 989} 990 991static int 992vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp) 993{ 994 struct ifvlan *ifv = ifp->if_softc; 995 int unit = ifp->if_dunit; 996 997 ether_ifdetach(ifp); /* first, remove it from system-wide lists */ 998 vlan_unconfig(ifp); /* now it can be unconfigured and freed */ 999 if_free(ifp); 1000 free(ifv, M_VLAN); 1001 ifc_free_unit(ifc, unit); 1002 1003 return (0); 1004} 1005 1006/* 1007 * The ifp->if_init entry point for vlan(4) is a no-op. 1008 */ 1009static void 1010vlan_init(void *foo __unused) 1011{ 1012} 1013 1014/* 1015 * The if_transmit method for vlan(4) interface. 1016 */ 1017static int 1018vlan_transmit(struct ifnet *ifp, struct mbuf *m) 1019{ 1020 struct ifvlan *ifv; 1021 struct ifnet *p; 1022 int error, len, mcast; 1023 1024 ifv = ifp->if_softc; 1025 p = PARENT(ifv); 1026 len = m->m_pkthdr.len; 1027 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1028 1029 BPF_MTAP(ifp, m); 1030 1031 /* 1032 * Do not run parent's if_transmit() if the parent is not up, 1033 * or parent's driver will cause a system crash. 1034 */ 1035 if (!UP_AND_RUNNING(p)) { 1036 m_freem(m); 1037 ifp->if_oerrors++; 1038 return (0); 1039 } 1040 1041 /* 1042 * Pad the frame to the minimum size allowed if told to. 1043 * This option is in accord with IEEE Std 802.1Q, 2003 Ed., 1044 * paragraph C.4.4.3.b. It can help to work around buggy 1045 * bridges that violate paragraph C.4.4.3.a from the same 1046 * document, i.e., fail to pad short frames after untagging. 1047 * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but 1048 * untagging it will produce a 62-byte frame, which is a runt 1049 * and requires padding. There are VLAN-enabled network 1050 * devices that just discard such runts instead or mishandle 1051 * them somehow. 1052 */ 1053 if (soft_pad && p->if_type == IFT_ETHER) { 1054 static char pad[8]; /* just zeros */ 1055 int n; 1056 1057 for (n = ETHERMIN + ETHER_HDR_LEN - m->m_pkthdr.len; 1058 n > 0; n -= sizeof(pad)) 1059 if (!m_append(m, min(n, sizeof(pad)), pad)) 1060 break; 1061 1062 if (n > 0) { 1063 if_printf(ifp, "cannot pad short frame\n"); 1064 ifp->if_oerrors++; 1065 m_freem(m); 1066 return (0); 1067 } 1068 } 1069 1070 /* 1071 * If underlying interface can do VLAN tag insertion itself, 1072 * just pass the packet along. However, we need some way to 1073 * tell the interface where the packet came from so that it 1074 * knows how to find the VLAN tag to use, so we attach a 1075 * packet tag that holds it. 1076 */ 1077 if (p->if_capenable & IFCAP_VLAN_HWTAGGING) { 1078 m->m_pkthdr.ether_vtag = ifv->ifv_vid; 1079 m->m_flags |= M_VLANTAG; 1080 } else { 1081 m = ether_vlanencap(m, ifv->ifv_vid); 1082 if (m == NULL) { 1083 if_printf(ifp, "unable to prepend VLAN header\n"); 1084 ifp->if_oerrors++; 1085 return (0); 1086 } 1087 } 1088 1089 /* 1090 * Send it, precisely as ether_output() would have. 1091 */ 1092 error = (p->if_transmit)(p, m); 1093 if (!error) { 1094 ifp->if_opackets++; 1095 ifp->if_omcasts += mcast; 1096 ifp->if_obytes += len; 1097 } else 1098 ifp->if_oerrors++; 1099 return (error); 1100} 1101 1102/* 1103 * The ifp->if_qflush entry point for vlan(4) is a no-op. 1104 */ 1105static void 1106vlan_qflush(struct ifnet *ifp __unused) 1107{ 1108} 1109 1110static void 1111vlan_input(struct ifnet *ifp, struct mbuf *m) 1112{ 1113 struct ifvlantrunk *trunk = ifp->if_vlantrunk; 1114 struct ifvlan *ifv; 1115 uint16_t vid; 1116 1117 KASSERT(trunk != NULL, ("%s: no trunk", __func__)); 1118 1119 if (m->m_flags & M_VLANTAG) { 1120 /* 1121 * Packet is tagged, but m contains a normal 1122 * Ethernet frame; the tag is stored out-of-band. 1123 */ 1124 vid = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag); 1125 m->m_flags &= ~M_VLANTAG; 1126 } else { 1127 struct ether_vlan_header *evl; 1128 1129 /* 1130 * Packet is tagged in-band as specified by 802.1q. 1131 */ 1132 switch (ifp->if_type) { 1133 case IFT_ETHER: 1134 if (m->m_len < sizeof(*evl) && 1135 (m = m_pullup(m, sizeof(*evl))) == NULL) { 1136 if_printf(ifp, "cannot pullup VLAN header\n"); 1137 return; 1138 } 1139 evl = mtod(m, struct ether_vlan_header *); 1140 vid = EVL_VLANOFTAG(ntohs(evl->evl_tag)); 1141 1142 /* 1143 * Remove the 802.1q header by copying the Ethernet 1144 * addresses over it and adjusting the beginning of 1145 * the data in the mbuf. The encapsulated Ethernet 1146 * type field is already in place. 1147 */ 1148 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN, 1149 ETHER_HDR_LEN - ETHER_TYPE_LEN); 1150 m_adj(m, ETHER_VLAN_ENCAP_LEN); 1151 break; 1152 1153 default: 1154#ifdef INVARIANTS 1155 panic("%s: %s has unsupported if_type %u", 1156 __func__, ifp->if_xname, ifp->if_type); 1157#endif 1158 m_freem(m); 1159 ifp->if_noproto++; 1160 return; 1161 } 1162 } 1163 1164 TRUNK_RLOCK(trunk); 1165 ifv = vlan_gethash(trunk, vid); 1166 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) { 1167 TRUNK_RUNLOCK(trunk); 1168 m_freem(m); 1169 ifp->if_noproto++; 1170 return; 1171 } 1172 TRUNK_RUNLOCK(trunk); 1173 1174 m->m_pkthdr.rcvif = ifv->ifv_ifp; 1175 ifv->ifv_ifp->if_ipackets++; 1176 1177 /* Pass it back through the parent's input routine. */ 1178 (*ifp->if_input)(ifv->ifv_ifp, m); 1179} 1180 1181static int 1182vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid) 1183{ 1184 struct ifvlantrunk *trunk; 1185 struct ifnet *ifp; 1186 int error = 0; 1187 1188 /* VID numbers 0x0 and 0xFFF are reserved */ 1189 if (vid == 0 || vid == 0xFFF) 1190 return (EINVAL); 1191 if (p->if_type != IFT_ETHER && 1192 (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0) 1193 return (EPROTONOSUPPORT); 1194 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS) 1195 return (EPROTONOSUPPORT); 1196 if (ifv->ifv_trunk) 1197 return (EBUSY); 1198 1199 if (p->if_vlantrunk == NULL) { 1200 trunk = malloc(sizeof(struct ifvlantrunk), 1201 M_VLAN, M_WAITOK | M_ZERO); 1202 vlan_inithash(trunk); 1203 VLAN_LOCK(); 1204 if (p->if_vlantrunk != NULL) { 1205 /* A race that that is very unlikely to be hit. */ 1206 vlan_freehash(trunk); 1207 free(trunk, M_VLAN); 1208 goto exists; 1209 } 1210 TRUNK_LOCK_INIT(trunk); 1211 TRUNK_LOCK(trunk); 1212 p->if_vlantrunk = trunk; 1213 trunk->parent = p; 1214 } else { 1215 VLAN_LOCK(); 1216exists: 1217 trunk = p->if_vlantrunk; 1218 TRUNK_LOCK(trunk); 1219 } 1220 1221 ifv->ifv_vid = vid; /* must set this before vlan_inshash() */ 1222 error = vlan_inshash(trunk, ifv); 1223 if (error) 1224 goto done; 1225 ifv->ifv_proto = ETHERTYPE_VLAN; 1226 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN; 1227 ifv->ifv_mintu = ETHERMIN; 1228 ifv->ifv_pflags = 0; 1229 1230 /* 1231 * If the parent supports the VLAN_MTU capability, 1232 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames, 1233 * use it. 1234 */ 1235 if (p->if_capenable & IFCAP_VLAN_MTU) { 1236 /* 1237 * No need to fudge the MTU since the parent can 1238 * handle extended frames. 1239 */ 1240 ifv->ifv_mtufudge = 0; 1241 } else { 1242 /* 1243 * Fudge the MTU by the encapsulation size. This 1244 * makes us incompatible with strictly compliant 1245 * 802.1Q implementations, but allows us to use 1246 * the feature with other NetBSD implementations, 1247 * which might still be useful. 1248 */ 1249 ifv->ifv_mtufudge = ifv->ifv_encaplen; 1250 } 1251 1252 ifv->ifv_trunk = trunk; 1253 ifp = ifv->ifv_ifp; 1254 /* 1255 * Initialize fields from our parent. This duplicates some 1256 * work with ether_ifattach() but allows for non-ethernet 1257 * interfaces to also work. 1258 */ 1259 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge; 1260 ifp->if_baudrate = p->if_baudrate; 1261 ifp->if_output = p->if_output; 1262 ifp->if_input = p->if_input; 1263 ifp->if_resolvemulti = p->if_resolvemulti; 1264 ifp->if_addrlen = p->if_addrlen; 1265 ifp->if_broadcastaddr = p->if_broadcastaddr; 1266 1267 /* 1268 * Copy only a selected subset of flags from the parent. 1269 * Other flags are none of our business. 1270 */ 1271#define VLAN_COPY_FLAGS (IFF_SIMPLEX) 1272 ifp->if_flags &= ~VLAN_COPY_FLAGS; 1273 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS; 1274#undef VLAN_COPY_FLAGS 1275 1276 ifp->if_link_state = p->if_link_state; 1277 1278 vlan_capabilities(ifv); 1279 1280 /* 1281 * Set up our interface address to reflect the underlying 1282 * physical interface's. 1283 */ 1284 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen); 1285 ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen = 1286 p->if_addrlen; 1287 1288 /* 1289 * Configure multicast addresses that may already be 1290 * joined on the vlan device. 1291 */ 1292 (void)vlan_setmulti(ifp); /* XXX: VLAN lock held */ 1293 1294 /* We are ready for operation now. */ 1295 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1296done: 1297 TRUNK_UNLOCK(trunk); 1298 if (error == 0) 1299 EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid); 1300 VLAN_UNLOCK(); 1301 1302 return (error); 1303} 1304 1305static void 1306vlan_unconfig(struct ifnet *ifp) 1307{ 1308 1309 VLAN_LOCK(); 1310 vlan_unconfig_locked(ifp, 0); 1311 VLAN_UNLOCK(); 1312} 1313 1314static void 1315vlan_unconfig_locked(struct ifnet *ifp, int departing) 1316{ 1317 struct ifvlantrunk *trunk; 1318 struct vlan_mc_entry *mc; 1319 struct ifvlan *ifv; 1320 struct ifnet *parent; 1321 int error; 1322 1323 VLAN_LOCK_ASSERT(); 1324 1325 ifv = ifp->if_softc; 1326 trunk = ifv->ifv_trunk; 1327 parent = NULL; 1328 1329 if (trunk != NULL) { 1330 1331 TRUNK_LOCK(trunk); 1332 parent = trunk->parent; 1333 1334 /* 1335 * Since the interface is being unconfigured, we need to 1336 * empty the list of multicast groups that we may have joined 1337 * while we were alive from the parent's list. 1338 */ 1339 while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) { 1340 /* 1341 * If the parent interface is being detached, 1342 * all its multicast addresses have already 1343 * been removed. Warn about errors if 1344 * if_delmulti() does fail, but don't abort as 1345 * all callers expect vlan destruction to 1346 * succeed. 1347 */ 1348 if (!departing) { 1349 error = if_delmulti(parent, 1350 (struct sockaddr *)&mc->mc_addr); 1351 if (error) 1352 if_printf(ifp, 1353 "Failed to delete multicast address from parent: %d\n", 1354 error); 1355 } 1356 SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries); 1357 free(mc, M_VLAN); 1358 } 1359 1360 vlan_setflags(ifp, 0); /* clear special flags on parent */ 1361 vlan_remhash(trunk, ifv); 1362 ifv->ifv_trunk = NULL; 1363 1364 /* 1365 * Check if we were the last. 1366 */ 1367 if (trunk->refcnt == 0) { 1368 trunk->parent->if_vlantrunk = NULL; 1369 /* 1370 * XXXGL: If some ithread has already entered 1371 * vlan_input() and is now blocked on the trunk 1372 * lock, then it should preempt us right after 1373 * unlock and finish its work. Then we will acquire 1374 * lock again in trunk_destroy(). 1375 */ 1376 TRUNK_UNLOCK(trunk); 1377 trunk_destroy(trunk); 1378 } else 1379 TRUNK_UNLOCK(trunk); 1380 } 1381 1382 /* Disconnect from parent. */ 1383 if (ifv->ifv_pflags) 1384 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__); 1385 ifp->if_mtu = ETHERMTU; 1386 ifp->if_link_state = LINK_STATE_UNKNOWN; 1387 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1388 1389 /* 1390 * Only dispatch an event if vlan was 1391 * attached, otherwise there is nothing 1392 * to cleanup anyway. 1393 */ 1394 if (parent != NULL) 1395 EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid); 1396} 1397 1398/* Handle a reference counted flag that should be set on the parent as well */ 1399static int 1400vlan_setflag(struct ifnet *ifp, int flag, int status, 1401 int (*func)(struct ifnet *, int)) 1402{ 1403 struct ifvlan *ifv; 1404 int error; 1405 1406 /* XXX VLAN_LOCK_ASSERT(); */ 1407 1408 ifv = ifp->if_softc; 1409 status = status ? (ifp->if_flags & flag) : 0; 1410 /* Now "status" contains the flag value or 0 */ 1411 1412 /* 1413 * See if recorded parent's status is different from what 1414 * we want it to be. If it is, flip it. We record parent's 1415 * status in ifv_pflags so that we won't clear parent's flag 1416 * we haven't set. In fact, we don't clear or set parent's 1417 * flags directly, but get or release references to them. 1418 * That's why we can be sure that recorded flags still are 1419 * in accord with actual parent's flags. 1420 */ 1421 if (status != (ifv->ifv_pflags & flag)) { 1422 error = (*func)(PARENT(ifv), status); 1423 if (error) 1424 return (error); 1425 ifv->ifv_pflags &= ~flag; 1426 ifv->ifv_pflags |= status; 1427 } 1428 return (0); 1429} 1430 1431/* 1432 * Handle IFF_* flags that require certain changes on the parent: 1433 * if "status" is true, update parent's flags respective to our if_flags; 1434 * if "status" is false, forcedly clear the flags set on parent. 1435 */ 1436static int 1437vlan_setflags(struct ifnet *ifp, int status) 1438{ 1439 int error, i; 1440 1441 for (i = 0; vlan_pflags[i].flag; i++) { 1442 error = vlan_setflag(ifp, vlan_pflags[i].flag, 1443 status, vlan_pflags[i].func); 1444 if (error) 1445 return (error); 1446 } 1447 return (0); 1448} 1449 1450/* Inform all vlans that their parent has changed link state */ 1451static void 1452vlan_link_state(struct ifnet *ifp) 1453{ 1454 struct ifvlantrunk *trunk = ifp->if_vlantrunk; 1455 struct ifvlan *ifv; 1456 int i; 1457 1458 TRUNK_LOCK(trunk); 1459#ifdef VLAN_ARRAY 1460 for (i = 0; i < VLAN_ARRAY_SIZE; i++) 1461 if (trunk->vlans[i] != NULL) { 1462 ifv = trunk->vlans[i]; 1463#else 1464 for (i = 0; i < (1 << trunk->hwidth); i++) 1465 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) { 1466#endif 1467 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate; 1468 if_link_state_change(ifv->ifv_ifp, 1469 trunk->parent->if_link_state); 1470 } 1471 TRUNK_UNLOCK(trunk); 1472} 1473 1474static void 1475vlan_capabilities(struct ifvlan *ifv) 1476{ 1477 struct ifnet *p = PARENT(ifv); 1478 struct ifnet *ifp = ifv->ifv_ifp; 1479 1480 TRUNK_LOCK_ASSERT(TRUNK(ifv)); 1481 1482 /* 1483 * If the parent interface can do checksum offloading 1484 * on VLANs, then propagate its hardware-assisted 1485 * checksumming flags. Also assert that checksum 1486 * offloading requires hardware VLAN tagging. 1487 */ 1488 if (p->if_capabilities & IFCAP_VLAN_HWCSUM) 1489 ifp->if_capabilities = p->if_capabilities & IFCAP_HWCSUM; 1490 1491 if (p->if_capenable & IFCAP_VLAN_HWCSUM && 1492 p->if_capenable & IFCAP_VLAN_HWTAGGING) { 1493 ifp->if_capenable = p->if_capenable & IFCAP_HWCSUM; 1494 ifp->if_hwassist = p->if_hwassist & (CSUM_IP | CSUM_TCP | 1495 CSUM_UDP | CSUM_SCTP | CSUM_IP_FRAGS | CSUM_FRAGMENT); 1496 } else { 1497 ifp->if_capenable = 0; 1498 ifp->if_hwassist = 0; 1499 } 1500 /* 1501 * If the parent interface can do TSO on VLANs then 1502 * propagate the hardware-assisted flag. TSO on VLANs 1503 * does not necessarily require hardware VLAN tagging. 1504 */ 1505 if (p->if_capabilities & IFCAP_VLAN_HWTSO) 1506 ifp->if_capabilities |= p->if_capabilities & IFCAP_TSO; 1507 if (p->if_capenable & IFCAP_VLAN_HWTSO) { 1508 ifp->if_capenable |= p->if_capenable & IFCAP_TSO; 1509 ifp->if_hwassist |= p->if_hwassist & CSUM_TSO; 1510 } else { 1511 ifp->if_capenable &= ~(p->if_capenable & IFCAP_TSO); 1512 ifp->if_hwassist &= ~(p->if_hwassist & CSUM_TSO); 1513 } 1514 1515 /* 1516 * If the parent interface can offload TCP connections over VLANs then 1517 * propagate its TOE capability to the VLAN interface. 1518 * 1519 * All TOE drivers in the tree today can deal with VLANs. If this 1520 * changes then IFCAP_VLAN_TOE should be promoted to a full capability 1521 * with its own bit. 1522 */ 1523#define IFCAP_VLAN_TOE IFCAP_TOE 1524 if (p->if_capabilities & IFCAP_VLAN_TOE) 1525 ifp->if_capabilities |= p->if_capabilities & IFCAP_TOE; 1526 if (p->if_capenable & IFCAP_VLAN_TOE) { 1527 TOEDEV(ifp) = TOEDEV(p); 1528 ifp->if_capenable |= p->if_capenable & IFCAP_TOE; 1529 } 1530} 1531 1532static void 1533vlan_trunk_capabilities(struct ifnet *ifp) 1534{ 1535 struct ifvlantrunk *trunk = ifp->if_vlantrunk; 1536 struct ifvlan *ifv; 1537 int i; 1538 1539 TRUNK_LOCK(trunk); 1540#ifdef VLAN_ARRAY 1541 for (i = 0; i < VLAN_ARRAY_SIZE; i++) 1542 if (trunk->vlans[i] != NULL) { 1543 ifv = trunk->vlans[i]; 1544#else 1545 for (i = 0; i < (1 << trunk->hwidth); i++) { 1546 LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) 1547#endif 1548 vlan_capabilities(ifv); 1549 } 1550 TRUNK_UNLOCK(trunk); 1551} 1552 1553static int 1554vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1555{ 1556 struct ifnet *p; 1557 struct ifreq *ifr; 1558 struct ifaddr *ifa; 1559 struct ifvlan *ifv; 1560 struct vlanreq vlr; 1561 int error = 0; 1562 1563 ifr = (struct ifreq *)data; 1564 ifa = (struct ifaddr *) data; 1565 ifv = ifp->if_softc; 1566 1567 switch (cmd) { 1568 case SIOCSIFADDR: 1569 ifp->if_flags |= IFF_UP; 1570#ifdef INET 1571 if (ifa->ifa_addr->sa_family == AF_INET) 1572 arp_ifinit(ifp, ifa); 1573#endif 1574 break; 1575 case SIOCGIFADDR: 1576 { 1577 struct sockaddr *sa; 1578 1579 sa = (struct sockaddr *)&ifr->ifr_data; 1580 bcopy(IF_LLADDR(ifp), sa->sa_data, ifp->if_addrlen); 1581 } 1582 break; 1583 case SIOCGIFMEDIA: 1584 VLAN_LOCK(); 1585 if (TRUNK(ifv) != NULL) { 1586 p = PARENT(ifv); 1587 VLAN_UNLOCK(); 1588 error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data); 1589 /* Limit the result to the parent's current config. */ 1590 if (error == 0) { 1591 struct ifmediareq *ifmr; 1592 1593 ifmr = (struct ifmediareq *)data; 1594 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) { 1595 ifmr->ifm_count = 1; 1596 error = copyout(&ifmr->ifm_current, 1597 ifmr->ifm_ulist, 1598 sizeof(int)); 1599 } 1600 } 1601 } else { 1602 VLAN_UNLOCK(); 1603 error = EINVAL; 1604 } 1605 break; 1606 1607 case SIOCSIFMEDIA: 1608 error = EINVAL; 1609 break; 1610 1611 case SIOCSIFMTU: 1612 /* 1613 * Set the interface MTU. 1614 */ 1615 VLAN_LOCK(); 1616 if (TRUNK(ifv) != NULL) { 1617 if (ifr->ifr_mtu > 1618 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) || 1619 ifr->ifr_mtu < 1620 (ifv->ifv_mintu - ifv->ifv_mtufudge)) 1621 error = EINVAL; 1622 else 1623 ifp->if_mtu = ifr->ifr_mtu; 1624 } else 1625 error = EINVAL; 1626 VLAN_UNLOCK(); 1627 break; 1628 1629 case SIOCSETVLAN: 1630#ifdef VIMAGE 1631 /* 1632 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN 1633 * interface to be delegated to a jail without allowing the 1634 * jail to change what underlying interface/VID it is 1635 * associated with. We are not entirely convinced that this 1636 * is the right way to accomplish that policy goal. 1637 */ 1638 if (ifp->if_vnet != ifp->if_home_vnet) { 1639 error = EPERM; 1640 break; 1641 } 1642#endif 1643 error = copyin(ifr->ifr_data, &vlr, sizeof(vlr)); 1644 if (error) 1645 break; 1646 if (vlr.vlr_parent[0] == '\0') { 1647 vlan_unconfig(ifp); 1648 break; 1649 } 1650 p = ifunit(vlr.vlr_parent); 1651 if (p == NULL) { 1652 error = ENOENT; 1653 break; 1654 } 1655 /* 1656 * Don't let the caller set up a VLAN VID with 1657 * anything except VLID bits. 1658 */ 1659 if (vlr.vlr_tag & ~EVL_VLID_MASK) { 1660 error = EINVAL; 1661 break; 1662 } 1663 error = vlan_config(ifv, p, vlr.vlr_tag); 1664 if (error) 1665 break; 1666 1667 /* Update flags on the parent, if necessary. */ 1668 vlan_setflags(ifp, 1); 1669 break; 1670 1671 case SIOCGETVLAN: 1672#ifdef VIMAGE 1673 if (ifp->if_vnet != ifp->if_home_vnet) { 1674 error = EPERM; 1675 break; 1676 } 1677#endif 1678 bzero(&vlr, sizeof(vlr)); 1679 VLAN_LOCK(); 1680 if (TRUNK(ifv) != NULL) { 1681 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname, 1682 sizeof(vlr.vlr_parent)); 1683 vlr.vlr_tag = ifv->ifv_vid; 1684 } 1685 VLAN_UNLOCK(); 1686 error = copyout(&vlr, ifr->ifr_data, sizeof(vlr)); 1687 break; 1688 1689 case SIOCSIFFLAGS: 1690 /* 1691 * We should propagate selected flags to the parent, 1692 * e.g., promiscuous mode. 1693 */ 1694 if (TRUNK(ifv) != NULL) 1695 error = vlan_setflags(ifp, 1); 1696 break; 1697 1698 case SIOCADDMULTI: 1699 case SIOCDELMULTI: 1700 /* 1701 * If we don't have a parent, just remember the membership for 1702 * when we do. 1703 */ 1704 if (TRUNK(ifv) != NULL) 1705 error = vlan_setmulti(ifp); 1706 break; 1707 1708 default: 1709 error = EINVAL; 1710 break; 1711 } 1712 1713 return (error); 1714} 1715