1 /*- 2 * Copyright 1998 Massachusetts Institute of Technology 3 * Copyright 2012 ADARA Networks, Inc. 4 * Copyright 2017 Dell EMC Isilon 5 * 6 * Portions of this software were developed by Robert N. M. Watson under 7 * contract to ADARA Networks, Inc. 8 * 9 * Permission to use, copy, modify, and distribute this software and 10 * its documentation for any purpose and without fee is hereby 11 * granted, provided that both the above copyright notice and this 12 * permission notice appear in all copies, that both the above 13 * copyright notice and this permission notice appear in all 14 * supporting documentation, and that the name of M.I.T. not be used 15 * in advertising or publicity pertaining to distribution of the 16 * software without specific, written prior permission. M.I.T. makes 17 * no representations about the suitability of this software for any 18 * purpose. It is provided "as is" without express or implied 19 * warranty. 20 * 21 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS 22 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE, 23 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 24 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT 25 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF 28 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND 29 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35 /* 36 * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs. 37 * This is sort of sneaky in the implementation, since 38 * we need to pretend to be enough of an Ethernet implementation 39 * to make arp work. The way we do this is by telling everyone 40 * that we are an Ethernet, and then catch the packets that 41 * ether_output() sends to us via if_transmit(), rewrite them for 42 * use by the real outgoing interface, and ask it to send them. 43 */ 44 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 47 48 #include "opt_inet.h" 49 #include "opt_inet6.h" 50 #include "opt_kern_tls.h" 51 #include "opt_vlan.h" 52 #include "opt_ratelimit.h" 53 54 #include <sys/param.h> 55 #include <sys/eventhandler.h> 56 #include <sys/kernel.h> 57 #include <sys/lock.h> 58 #include <sys/malloc.h> 59 #include <sys/mbuf.h> 60 #include <sys/module.h> 61 #include <sys/rmlock.h> 62 #include <sys/priv.h> 63 #include <sys/queue.h> 64 #include <sys/socket.h> 65 #include <sys/sockio.h> 66 #include <sys/sysctl.h> 67 #include <sys/systm.h> 68 #include <sys/sx.h> 69 #include <sys/taskqueue.h> 70 71 #include <net/bpf.h> 72 #include <net/ethernet.h> 73 #include <net/if.h> 74 #include <net/if_var.h> 75 #include <net/if_clone.h> 76 #include <net/if_dl.h> 77 #include <net/if_types.h> 78 #include <net/if_vlan_var.h> 79 #include <net/route.h> 80 #include <net/vnet.h> 81 82 #ifdef INET 83 #include <netinet/in.h> 84 #include <netinet/if_ether.h> 85 #endif 86 87 #ifdef INET6 88 /* 89 * XXX: declare here to avoid to include many inet6 related files.. 90 * should be more generalized? 91 */ 92 extern void nd6_setmtu(struct ifnet *); 93 #endif 94 95 #define VLAN_DEF_HWIDTH 4 96 #define VLAN_IFFLAGS (IFF_BROADCAST | IFF_MULTICAST) 97 98 #define UP_AND_RUNNING(ifp) \ 99 ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING) 100 101 CK_SLIST_HEAD(ifvlanhead, ifvlan); 102 103 struct ifvlantrunk { 104 struct ifnet *parent; /* parent interface of this trunk */ 105 struct mtx lock; 106 #ifdef VLAN_ARRAY 107 #define VLAN_ARRAY_SIZE (EVL_VLID_MASK + 1) 108 struct ifvlan *vlans[VLAN_ARRAY_SIZE]; /* static table */ 109 #else 110 struct ifvlanhead *hash; /* dynamic hash-list table */ 111 uint16_t hmask; 112 uint16_t hwidth; 113 #endif 114 int refcnt; 115 }; 116 117 #if defined(KERN_TLS) || defined(RATELIMIT) 118 struct vlan_snd_tag { 119 struct m_snd_tag com; 120 struct m_snd_tag *tag; 121 }; 122 123 static inline struct vlan_snd_tag * 124 mst_to_vst(struct m_snd_tag *mst) 125 { 126 127 return (__containerof(mst, struct vlan_snd_tag, com)); 128 } 129 #endif 130 131 /* 132 * This macro provides a facility to iterate over every vlan on a trunk with 133 * the assumption that none will be added/removed during iteration. 134 */ 135 #ifdef VLAN_ARRAY 136 #define VLAN_FOREACH(_ifv, _trunk) \ 137 size_t _i; \ 138 for (_i = 0; _i < VLAN_ARRAY_SIZE; _i++) \ 139 if (((_ifv) = (_trunk)->vlans[_i]) != NULL) 140 #else /* VLAN_ARRAY */ 141 #define VLAN_FOREACH(_ifv, _trunk) \ 142 struct ifvlan *_next; \ 143 size_t _i; \ 144 for (_i = 0; _i < (1 << (_trunk)->hwidth); _i++) \ 145 CK_SLIST_FOREACH_SAFE((_ifv), &(_trunk)->hash[_i], ifv_list, _next) 146 #endif /* VLAN_ARRAY */ 147 148 /* 149 * This macro provides a facility to iterate over every vlan on a trunk while 150 * also modifying the number of vlans on the trunk. The iteration continues 151 * until some condition is met or there are no more vlans on the trunk. 152 */ 153 #ifdef VLAN_ARRAY 154 /* The VLAN_ARRAY case is simple -- just a for loop using the condition. */ 155 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \ 156 size_t _i; \ 157 for (_i = 0; !(_cond) && _i < VLAN_ARRAY_SIZE; _i++) \ 158 if (((_ifv) = (_trunk)->vlans[_i])) 159 #else /* VLAN_ARRAY */ 160 /* 161 * The hash table case is more complicated. We allow for the hash table to be 162 * modified (i.e. vlans removed) while we are iterating over it. To allow for 163 * this we must restart the iteration every time we "touch" something during 164 * the iteration, since removal will resize the hash table and invalidate our 165 * current position. If acting on the touched element causes the trunk to be 166 * emptied, then iteration also stops. 167 */ 168 #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \ 169 size_t _i; \ 170 bool _touch = false; \ 171 for (_i = 0; \ 172 !(_cond) && _i < (1 << (_trunk)->hwidth); \ 173 _i = (_touch && ((_trunk) != NULL) ? 0 : _i + 1), _touch = false) \ 174 if (((_ifv) = CK_SLIST_FIRST(&(_trunk)->hash[_i])) != NULL && \ 175 (_touch = true)) 176 #endif /* VLAN_ARRAY */ 177 178 struct vlan_mc_entry { 179 struct sockaddr_dl mc_addr; 180 CK_SLIST_ENTRY(vlan_mc_entry) mc_entries; 181 struct epoch_context mc_epoch_ctx; 182 }; 183 184 struct ifvlan { 185 struct ifvlantrunk *ifv_trunk; 186 struct ifnet *ifv_ifp; 187 #define TRUNK(ifv) ((ifv)->ifv_trunk) 188 #define PARENT(ifv) (TRUNK(ifv)->parent) 189 void *ifv_cookie; 190 int ifv_pflags; /* special flags we have set on parent */ 191 int ifv_capenable; 192 int ifv_encaplen; /* encapsulation length */ 193 int ifv_mtufudge; /* MTU fudged by this much */ 194 int ifv_mintu; /* min transmission unit */ 195 struct ether_8021q_tag ifv_qtag; 196 #define ifv_proto ifv_qtag.proto 197 #define ifv_vid ifv_qtag.vid 198 #define ifv_pcp ifv_qtag.pcp 199 struct task lladdr_task; 200 CK_SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead; 201 #ifndef VLAN_ARRAY 202 CK_SLIST_ENTRY(ifvlan) ifv_list; 203 #endif 204 }; 205 206 /* Special flags we should propagate to parent. */ 207 static struct { 208 int flag; 209 int (*func)(struct ifnet *, int); 210 } vlan_pflags[] = { 211 {IFF_PROMISC, ifpromisc}, 212 {IFF_ALLMULTI, if_allmulti}, 213 {0, NULL} 214 }; 215 216 VNET_DECLARE(int, vlan_mtag_pcp); 217 #define V_vlan_mtag_pcp VNET(vlan_mtag_pcp) 218 219 static const char vlanname[] = "vlan"; 220 static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface"); 221 222 static eventhandler_tag ifdetach_tag; 223 static eventhandler_tag iflladdr_tag; 224 static eventhandler_tag ifevent_tag; 225 226 /* 227 * if_vlan uses two module-level synchronizations primitives to allow concurrent 228 * modification of vlan interfaces and (mostly) allow for vlans to be destroyed 229 * while they are being used for tx/rx. To accomplish this in a way that has 230 * acceptable performance and cooperation with other parts of the network stack 231 * there is a non-sleepable epoch(9) and an sx(9). 232 * 233 * The performance-sensitive paths that warrant using the epoch(9) are 234 * vlan_transmit and vlan_input. Both have to check for the vlan interface's 235 * existence using if_vlantrunk, and being in the network tx/rx paths the use 236 * of an epoch(9) gives a measureable improvement in performance. 237 * 238 * The reason for having an sx(9) is mostly because there are still areas that 239 * must be sleepable and also have safe concurrent access to a vlan interface. 240 * Since the sx(9) exists, it is used by default in most paths unless sleeping 241 * is not permitted, or if it is not clear whether sleeping is permitted. 242 * 243 */ 244 #define _VLAN_SX_ID ifv_sx 245 246 static struct sx _VLAN_SX_ID; 247 248 #define VLAN_LOCKING_INIT() \ 249 sx_init_flags(&_VLAN_SX_ID, "vlan_sx", SX_RECURSE) 250 251 #define VLAN_LOCKING_DESTROY() \ 252 sx_destroy(&_VLAN_SX_ID) 253 254 #define VLAN_SLOCK() sx_slock(&_VLAN_SX_ID) 255 #define VLAN_SUNLOCK() sx_sunlock(&_VLAN_SX_ID) 256 #define VLAN_XLOCK() sx_xlock(&_VLAN_SX_ID) 257 #define VLAN_XUNLOCK() sx_xunlock(&_VLAN_SX_ID) 258 #define VLAN_SLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_SLOCKED) 259 #define VLAN_XLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_XLOCKED) 260 #define VLAN_SXLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_LOCKED) 261 262 /* 263 * We also have a per-trunk mutex that should be acquired when changing 264 * its state. 265 */ 266 #define TRUNK_LOCK_INIT(trunk) mtx_init(&(trunk)->lock, vlanname, NULL, MTX_DEF) 267 #define TRUNK_LOCK_DESTROY(trunk) mtx_destroy(&(trunk)->lock) 268 #define TRUNK_WLOCK(trunk) mtx_lock(&(trunk)->lock) 269 #define TRUNK_WUNLOCK(trunk) mtx_unlock(&(trunk)->lock) 270 #define TRUNK_WLOCK_ASSERT(trunk) mtx_assert(&(trunk)->lock, MA_OWNED); 271 272 /* 273 * The VLAN_ARRAY substitutes the dynamic hash with a static array 274 * with 4096 entries. In theory this can give a boost in processing, 275 * however in practice it does not. Probably this is because the array 276 * is too big to fit into CPU cache. 277 */ 278 #ifndef VLAN_ARRAY 279 static void vlan_inithash(struct ifvlantrunk *trunk); 280 static void vlan_freehash(struct ifvlantrunk *trunk); 281 static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv); 282 static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv); 283 static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch); 284 static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk, 285 uint16_t vid); 286 #endif 287 static void trunk_destroy(struct ifvlantrunk *trunk); 288 289 static void vlan_init(void *foo); 290 static void vlan_input(struct ifnet *ifp, struct mbuf *m); 291 static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr); 292 #if defined(KERN_TLS) || defined(RATELIMIT) 293 static int vlan_snd_tag_alloc(struct ifnet *, 294 union if_snd_tag_alloc_params *, struct m_snd_tag **); 295 static int vlan_snd_tag_modify(struct m_snd_tag *, 296 union if_snd_tag_modify_params *); 297 static int vlan_snd_tag_query(struct m_snd_tag *, 298 union if_snd_tag_query_params *); 299 static void vlan_snd_tag_free(struct m_snd_tag *); 300 static struct m_snd_tag *vlan_next_snd_tag(struct m_snd_tag *); 301 static void vlan_ratelimit_query(struct ifnet *, 302 struct if_ratelimit_query_results *); 303 #endif 304 static void vlan_qflush(struct ifnet *ifp); 305 static int vlan_setflag(struct ifnet *ifp, int flag, int status, 306 int (*func)(struct ifnet *, int)); 307 static int vlan_setflags(struct ifnet *ifp, int status); 308 static int vlan_setmulti(struct ifnet *ifp); 309 static int vlan_transmit(struct ifnet *ifp, struct mbuf *m); 310 #ifdef ALTQ 311 static void vlan_altq_start(struct ifnet *ifp); 312 static int vlan_altq_transmit(struct ifnet *ifp, struct mbuf *m); 313 #endif 314 static int vlan_output(struct ifnet *ifp, struct mbuf *m, 315 const struct sockaddr *dst, struct route *ro); 316 static void vlan_unconfig(struct ifnet *ifp); 317 static void vlan_unconfig_locked(struct ifnet *ifp, int departing); 318 static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag, 319 uint16_t proto); 320 static void vlan_link_state(struct ifnet *ifp); 321 static void vlan_capabilities(struct ifvlan *ifv); 322 static void vlan_trunk_capabilities(struct ifnet *ifp); 323 324 static struct ifnet *vlan_clone_match_ethervid(const char *, int *); 325 static int vlan_clone_match(struct if_clone *, const char *); 326 static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t); 327 static int vlan_clone_destroy(struct if_clone *, struct ifnet *); 328 329 static void vlan_ifdetach(void *arg, struct ifnet *ifp); 330 static void vlan_iflladdr(void *arg, struct ifnet *ifp); 331 static void vlan_ifevent(void *arg, struct ifnet *ifp, int event); 332 333 static void vlan_lladdr_fn(void *arg, int pending); 334 335 static struct if_clone *vlan_cloner; 336 337 #ifdef VIMAGE 338 VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner); 339 #define V_vlan_cloner VNET(vlan_cloner) 340 #endif 341 342 #ifdef RATELIMIT 343 static const struct if_snd_tag_sw vlan_snd_tag_ul_sw = { 344 .snd_tag_modify = vlan_snd_tag_modify, 345 .snd_tag_query = vlan_snd_tag_query, 346 .snd_tag_free = vlan_snd_tag_free, 347 .next_snd_tag = vlan_next_snd_tag, 348 .type = IF_SND_TAG_TYPE_UNLIMITED 349 }; 350 351 static const struct if_snd_tag_sw vlan_snd_tag_rl_sw = { 352 .snd_tag_modify = vlan_snd_tag_modify, 353 .snd_tag_query = vlan_snd_tag_query, 354 .snd_tag_free = vlan_snd_tag_free, 355 .next_snd_tag = vlan_next_snd_tag, 356 .type = IF_SND_TAG_TYPE_RATE_LIMIT 357 }; 358 #endif 359 360 #ifdef KERN_TLS 361 static const struct if_snd_tag_sw vlan_snd_tag_tls_sw = { 362 .snd_tag_modify = vlan_snd_tag_modify, 363 .snd_tag_query = vlan_snd_tag_query, 364 .snd_tag_free = vlan_snd_tag_free, 365 .next_snd_tag = vlan_next_snd_tag, 366 .type = IF_SND_TAG_TYPE_TLS 367 }; 368 369 #ifdef RATELIMIT 370 static const struct if_snd_tag_sw vlan_snd_tag_tls_rl_sw = { 371 .snd_tag_modify = vlan_snd_tag_modify, 372 .snd_tag_query = vlan_snd_tag_query, 373 .snd_tag_free = vlan_snd_tag_free, 374 .next_snd_tag = vlan_next_snd_tag, 375 .type = IF_SND_TAG_TYPE_TLS_RATE_LIMIT 376 }; 377 #endif 378 #endif 379 380 static void 381 vlan_mc_free(struct epoch_context *ctx) 382 { 383 struct vlan_mc_entry *mc = __containerof(ctx, struct vlan_mc_entry, mc_epoch_ctx); 384 free(mc, M_VLAN); 385 } 386 387 #ifndef VLAN_ARRAY 388 #define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m)) 389 390 static void 391 vlan_inithash(struct ifvlantrunk *trunk) 392 { 393 int i, n; 394 395 /* 396 * The trunk must not be locked here since we call malloc(M_WAITOK). 397 * It is OK in case this function is called before the trunk struct 398 * gets hooked up and becomes visible from other threads. 399 */ 400 401 KASSERT(trunk->hwidth == 0 && trunk->hash == NULL, 402 ("%s: hash already initialized", __func__)); 403 404 trunk->hwidth = VLAN_DEF_HWIDTH; 405 n = 1 << trunk->hwidth; 406 trunk->hmask = n - 1; 407 trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK); 408 for (i = 0; i < n; i++) 409 CK_SLIST_INIT(&trunk->hash[i]); 410 } 411 412 static void 413 vlan_freehash(struct ifvlantrunk *trunk) 414 { 415 #ifdef INVARIANTS 416 int i; 417 418 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__)); 419 for (i = 0; i < (1 << trunk->hwidth); i++) 420 KASSERT(CK_SLIST_EMPTY(&trunk->hash[i]), 421 ("%s: hash table not empty", __func__)); 422 #endif 423 free(trunk->hash, M_VLAN); 424 trunk->hash = NULL; 425 trunk->hwidth = trunk->hmask = 0; 426 } 427 428 static int 429 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv) 430 { 431 int i, b; 432 struct ifvlan *ifv2; 433 434 VLAN_XLOCK_ASSERT(); 435 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__)); 436 437 b = 1 << trunk->hwidth; 438 i = HASH(ifv->ifv_vid, trunk->hmask); 439 CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list) 440 if (ifv->ifv_vid == ifv2->ifv_vid) 441 return (EEXIST); 442 443 /* 444 * Grow the hash when the number of vlans exceeds half of the number of 445 * hash buckets squared. This will make the average linked-list length 446 * buckets/2. 447 */ 448 if (trunk->refcnt > (b * b) / 2) { 449 vlan_growhash(trunk, 1); 450 i = HASH(ifv->ifv_vid, trunk->hmask); 451 } 452 CK_SLIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list); 453 trunk->refcnt++; 454 455 return (0); 456 } 457 458 static int 459 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv) 460 { 461 int i, b; 462 struct ifvlan *ifv2; 463 464 VLAN_XLOCK_ASSERT(); 465 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__)); 466 467 b = 1 << (trunk->hwidth - 1); 468 i = HASH(ifv->ifv_vid, trunk->hmask); 469 CK_SLIST_FOREACH(ifv2, &trunk->hash[i], ifv_list) 470 if (ifv2 == ifv) { 471 trunk->refcnt--; 472 CK_SLIST_REMOVE(&trunk->hash[i], ifv2, ifvlan, ifv_list); 473 if (trunk->refcnt < (b * b) / 2) 474 vlan_growhash(trunk, -1); 475 return (0); 476 } 477 478 panic("%s: vlan not found\n", __func__); 479 return (ENOENT); /*NOTREACHED*/ 480 } 481 482 /* 483 * Grow the hash larger or smaller if memory permits. 484 */ 485 static void 486 vlan_growhash(struct ifvlantrunk *trunk, int howmuch) 487 { 488 struct ifvlan *ifv; 489 struct ifvlanhead *hash2; 490 int hwidth2, i, j, n, n2; 491 492 VLAN_XLOCK_ASSERT(); 493 KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__)); 494 495 if (howmuch == 0) { 496 /* Harmless yet obvious coding error */ 497 printf("%s: howmuch is 0\n", __func__); 498 return; 499 } 500 501 hwidth2 = trunk->hwidth + howmuch; 502 n = 1 << trunk->hwidth; 503 n2 = 1 << hwidth2; 504 /* Do not shrink the table below the default */ 505 if (hwidth2 < VLAN_DEF_HWIDTH) 506 return; 507 508 hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_WAITOK); 509 if (hash2 == NULL) { 510 printf("%s: out of memory -- hash size not changed\n", 511 __func__); 512 return; /* We can live with the old hash table */ 513 } 514 for (j = 0; j < n2; j++) 515 CK_SLIST_INIT(&hash2[j]); 516 for (i = 0; i < n; i++) 517 while ((ifv = CK_SLIST_FIRST(&trunk->hash[i])) != NULL) { 518 CK_SLIST_REMOVE(&trunk->hash[i], ifv, ifvlan, ifv_list); 519 j = HASH(ifv->ifv_vid, n2 - 1); 520 CK_SLIST_INSERT_HEAD(&hash2[j], ifv, ifv_list); 521 } 522 NET_EPOCH_WAIT(); 523 free(trunk->hash, M_VLAN); 524 trunk->hash = hash2; 525 trunk->hwidth = hwidth2; 526 trunk->hmask = n2 - 1; 527 528 if (bootverbose) 529 if_printf(trunk->parent, 530 "VLAN hash table resized from %d to %d buckets\n", n, n2); 531 } 532 533 static __inline struct ifvlan * 534 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid) 535 { 536 struct ifvlan *ifv; 537 538 NET_EPOCH_ASSERT(); 539 540 CK_SLIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list) 541 if (ifv->ifv_vid == vid) 542 return (ifv); 543 return (NULL); 544 } 545 546 #if 0 547 /* Debugging code to view the hashtables. */ 548 static void 549 vlan_dumphash(struct ifvlantrunk *trunk) 550 { 551 int i; 552 struct ifvlan *ifv; 553 554 for (i = 0; i < (1 << trunk->hwidth); i++) { 555 printf("%d: ", i); 556 CK_SLIST_FOREACH(ifv, &trunk->hash[i], ifv_list) 557 printf("%s ", ifv->ifv_ifp->if_xname); 558 printf("\n"); 559 } 560 } 561 #endif /* 0 */ 562 #else 563 564 static __inline struct ifvlan * 565 vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid) 566 { 567 568 return trunk->vlans[vid]; 569 } 570 571 static __inline int 572 vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv) 573 { 574 575 if (trunk->vlans[ifv->ifv_vid] != NULL) 576 return EEXIST; 577 trunk->vlans[ifv->ifv_vid] = ifv; 578 trunk->refcnt++; 579 580 return (0); 581 } 582 583 static __inline int 584 vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv) 585 { 586 587 trunk->vlans[ifv->ifv_vid] = NULL; 588 trunk->refcnt--; 589 590 return (0); 591 } 592 593 static __inline void 594 vlan_freehash(struct ifvlantrunk *trunk) 595 { 596 } 597 598 static __inline void 599 vlan_inithash(struct ifvlantrunk *trunk) 600 { 601 } 602 603 #endif /* !VLAN_ARRAY */ 604 605 static void 606 trunk_destroy(struct ifvlantrunk *trunk) 607 { 608 VLAN_XLOCK_ASSERT(); 609 610 vlan_freehash(trunk); 611 trunk->parent->if_vlantrunk = NULL; 612 TRUNK_LOCK_DESTROY(trunk); 613 if_rele(trunk->parent); 614 free(trunk, M_VLAN); 615 } 616 617 /* 618 * Program our multicast filter. What we're actually doing is 619 * programming the multicast filter of the parent. This has the 620 * side effect of causing the parent interface to receive multicast 621 * traffic that it doesn't really want, which ends up being discarded 622 * later by the upper protocol layers. Unfortunately, there's no way 623 * to avoid this: there really is only one physical interface. 624 */ 625 static int 626 vlan_setmulti(struct ifnet *ifp) 627 { 628 struct ifnet *ifp_p; 629 struct ifmultiaddr *ifma; 630 struct ifvlan *sc; 631 struct vlan_mc_entry *mc; 632 int error; 633 634 VLAN_XLOCK_ASSERT(); 635 636 /* Find the parent. */ 637 sc = ifp->if_softc; 638 ifp_p = PARENT(sc); 639 640 CURVNET_SET_QUIET(ifp_p->if_vnet); 641 642 /* First, remove any existing filter entries. */ 643 while ((mc = CK_SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) { 644 CK_SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries); 645 (void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr); 646 NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx); 647 } 648 649 /* Now program new ones. */ 650 IF_ADDR_WLOCK(ifp); 651 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 652 if (ifma->ifma_addr->sa_family != AF_LINK) 653 continue; 654 mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT); 655 if (mc == NULL) { 656 IF_ADDR_WUNLOCK(ifp); 657 CURVNET_RESTORE(); 658 return (ENOMEM); 659 } 660 bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len); 661 mc->mc_addr.sdl_index = ifp_p->if_index; 662 CK_SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries); 663 } 664 IF_ADDR_WUNLOCK(ifp); 665 CK_SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) { 666 error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr, 667 NULL); 668 if (error) { 669 CURVNET_RESTORE(); 670 return (error); 671 } 672 } 673 674 CURVNET_RESTORE(); 675 return (0); 676 } 677 678 /* 679 * A handler for interface ifnet events. 680 */ 681 static void 682 vlan_ifevent(void *arg __unused, struct ifnet *ifp, int event) 683 { 684 struct epoch_tracker et; 685 struct ifvlan *ifv; 686 struct ifvlantrunk *trunk; 687 688 if (event != IFNET_EVENT_UPDATE_BAUDRATE) 689 return; 690 691 NET_EPOCH_ENTER(et); 692 trunk = ifp->if_vlantrunk; 693 if (trunk == NULL) { 694 NET_EPOCH_EXIT(et); 695 return; 696 } 697 698 TRUNK_WLOCK(trunk); 699 VLAN_FOREACH(ifv, trunk) { 700 ifv->ifv_ifp->if_baudrate = ifp->if_baudrate; 701 } 702 TRUNK_WUNLOCK(trunk); 703 NET_EPOCH_EXIT(et); 704 } 705 706 /* 707 * A handler for parent interface link layer address changes. 708 * If the parent interface link layer address is changed we 709 * should also change it on all children vlans. 710 */ 711 static void 712 vlan_iflladdr(void *arg __unused, struct ifnet *ifp) 713 { 714 struct epoch_tracker et; 715 struct ifvlan *ifv; 716 struct ifnet *ifv_ifp; 717 struct ifvlantrunk *trunk; 718 struct sockaddr_dl *sdl; 719 720 /* Need the epoch since this is run on taskqueue_swi. */ 721 NET_EPOCH_ENTER(et); 722 trunk = ifp->if_vlantrunk; 723 if (trunk == NULL) { 724 NET_EPOCH_EXIT(et); 725 return; 726 } 727 728 /* 729 * OK, it's a trunk. Loop over and change all vlan's lladdrs on it. 730 * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR 731 * ioctl calls on the parent garbling the lladdr of the child vlan. 732 */ 733 TRUNK_WLOCK(trunk); 734 VLAN_FOREACH(ifv, trunk) { 735 /* 736 * Copy new new lladdr into the ifv_ifp, enqueue a task 737 * to actually call if_setlladdr. if_setlladdr needs to 738 * be deferred to a taskqueue because it will call into 739 * the if_vlan ioctl path and try to acquire the global 740 * lock. 741 */ 742 ifv_ifp = ifv->ifv_ifp; 743 bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp), 744 ifp->if_addrlen); 745 sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr; 746 sdl->sdl_alen = ifp->if_addrlen; 747 taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task); 748 } 749 TRUNK_WUNLOCK(trunk); 750 NET_EPOCH_EXIT(et); 751 } 752 753 /* 754 * A handler for network interface departure events. 755 * Track departure of trunks here so that we don't access invalid 756 * pointers or whatever if a trunk is ripped from under us, e.g., 757 * by ejecting its hot-plug card. However, if an ifnet is simply 758 * being renamed, then there's no need to tear down the state. 759 */ 760 static void 761 vlan_ifdetach(void *arg __unused, struct ifnet *ifp) 762 { 763 struct ifvlan *ifv; 764 struct ifvlantrunk *trunk; 765 766 /* If the ifnet is just being renamed, don't do anything. */ 767 if (ifp->if_flags & IFF_RENAMING) 768 return; 769 VLAN_XLOCK(); 770 trunk = ifp->if_vlantrunk; 771 if (trunk == NULL) { 772 VLAN_XUNLOCK(); 773 return; 774 } 775 776 /* 777 * OK, it's a trunk. Loop over and detach all vlan's on it. 778 * Check trunk pointer after each vlan_unconfig() as it will 779 * free it and set to NULL after the last vlan was detached. 780 */ 781 VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk, 782 ifp->if_vlantrunk == NULL) 783 vlan_unconfig_locked(ifv->ifv_ifp, 1); 784 785 /* Trunk should have been destroyed in vlan_unconfig(). */ 786 KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__)); 787 VLAN_XUNLOCK(); 788 } 789 790 /* 791 * Return the trunk device for a virtual interface. 792 */ 793 static struct ifnet * 794 vlan_trunkdev(struct ifnet *ifp) 795 { 796 struct ifvlan *ifv; 797 798 NET_EPOCH_ASSERT(); 799 800 if (ifp->if_type != IFT_L2VLAN) 801 return (NULL); 802 803 ifv = ifp->if_softc; 804 ifp = NULL; 805 if (ifv->ifv_trunk) 806 ifp = PARENT(ifv); 807 return (ifp); 808 } 809 810 /* 811 * Return the 12-bit VLAN VID for this interface, for use by external 812 * components such as Infiniband. 813 * 814 * XXXRW: Note that the function name here is historical; it should be named 815 * vlan_vid(). 816 */ 817 static int 818 vlan_tag(struct ifnet *ifp, uint16_t *vidp) 819 { 820 struct ifvlan *ifv; 821 822 if (ifp->if_type != IFT_L2VLAN) 823 return (EINVAL); 824 ifv = ifp->if_softc; 825 *vidp = ifv->ifv_vid; 826 return (0); 827 } 828 829 static int 830 vlan_pcp(struct ifnet *ifp, uint16_t *pcpp) 831 { 832 struct ifvlan *ifv; 833 834 if (ifp->if_type != IFT_L2VLAN) 835 return (EINVAL); 836 ifv = ifp->if_softc; 837 *pcpp = ifv->ifv_pcp; 838 return (0); 839 } 840 841 /* 842 * Return a driver specific cookie for this interface. Synchronization 843 * with setcookie must be provided by the driver. 844 */ 845 static void * 846 vlan_cookie(struct ifnet *ifp) 847 { 848 struct ifvlan *ifv; 849 850 if (ifp->if_type != IFT_L2VLAN) 851 return (NULL); 852 ifv = ifp->if_softc; 853 return (ifv->ifv_cookie); 854 } 855 856 /* 857 * Store a cookie in our softc that drivers can use to store driver 858 * private per-instance data in. 859 */ 860 static int 861 vlan_setcookie(struct ifnet *ifp, void *cookie) 862 { 863 struct ifvlan *ifv; 864 865 if (ifp->if_type != IFT_L2VLAN) 866 return (EINVAL); 867 ifv = ifp->if_softc; 868 ifv->ifv_cookie = cookie; 869 return (0); 870 } 871 872 /* 873 * Return the vlan device present at the specific VID. 874 */ 875 static struct ifnet * 876 vlan_devat(struct ifnet *ifp, uint16_t vid) 877 { 878 struct ifvlantrunk *trunk; 879 struct ifvlan *ifv; 880 881 NET_EPOCH_ASSERT(); 882 883 trunk = ifp->if_vlantrunk; 884 if (trunk == NULL) 885 return (NULL); 886 ifp = NULL; 887 ifv = vlan_gethash(trunk, vid); 888 if (ifv) 889 ifp = ifv->ifv_ifp; 890 return (ifp); 891 } 892 893 /* 894 * VLAN support can be loaded as a module. The only place in the 895 * system that's intimately aware of this is ether_input. We hook 896 * into this code through vlan_input_p which is defined there and 897 * set here. No one else in the system should be aware of this so 898 * we use an explicit reference here. 899 */ 900 extern void (*vlan_input_p)(struct ifnet *, struct mbuf *); 901 902 /* For if_link_state_change() eyes only... */ 903 extern void (*vlan_link_state_p)(struct ifnet *); 904 905 static int 906 vlan_modevent(module_t mod, int type, void *data) 907 { 908 909 switch (type) { 910 case MOD_LOAD: 911 ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event, 912 vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY); 913 if (ifdetach_tag == NULL) 914 return (ENOMEM); 915 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event, 916 vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY); 917 if (iflladdr_tag == NULL) 918 return (ENOMEM); 919 ifevent_tag = EVENTHANDLER_REGISTER(ifnet_event, 920 vlan_ifevent, NULL, EVENTHANDLER_PRI_ANY); 921 if (ifevent_tag == NULL) 922 return (ENOMEM); 923 VLAN_LOCKING_INIT(); 924 vlan_input_p = vlan_input; 925 vlan_link_state_p = vlan_link_state; 926 vlan_trunk_cap_p = vlan_trunk_capabilities; 927 vlan_trunkdev_p = vlan_trunkdev; 928 vlan_cookie_p = vlan_cookie; 929 vlan_setcookie_p = vlan_setcookie; 930 vlan_tag_p = vlan_tag; 931 vlan_pcp_p = vlan_pcp; 932 vlan_devat_p = vlan_devat; 933 #ifndef VIMAGE 934 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match, 935 vlan_clone_create, vlan_clone_destroy); 936 #endif 937 if (bootverbose) 938 printf("vlan: initialized, using " 939 #ifdef VLAN_ARRAY 940 "full-size arrays" 941 #else 942 "hash tables with chaining" 943 #endif 944 945 "\n"); 946 break; 947 case MOD_UNLOAD: 948 #ifndef VIMAGE 949 if_clone_detach(vlan_cloner); 950 #endif 951 EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag); 952 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag); 953 EVENTHANDLER_DEREGISTER(ifnet_event, ifevent_tag); 954 vlan_input_p = NULL; 955 vlan_link_state_p = NULL; 956 vlan_trunk_cap_p = NULL; 957 vlan_trunkdev_p = NULL; 958 vlan_tag_p = NULL; 959 vlan_cookie_p = NULL; 960 vlan_setcookie_p = NULL; 961 vlan_devat_p = NULL; 962 VLAN_LOCKING_DESTROY(); 963 if (bootverbose) 964 printf("vlan: unloaded\n"); 965 break; 966 default: 967 return (EOPNOTSUPP); 968 } 969 return (0); 970 } 971 972 static moduledata_t vlan_mod = { 973 "if_vlan", 974 vlan_modevent, 975 0 976 }; 977 978 DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 979 MODULE_VERSION(if_vlan, 3); 980 981 #ifdef VIMAGE 982 static void 983 vnet_vlan_init(const void *unused __unused) 984 { 985 986 vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match, 987 vlan_clone_create, vlan_clone_destroy); 988 V_vlan_cloner = vlan_cloner; 989 } 990 VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 991 vnet_vlan_init, NULL); 992 993 static void 994 vnet_vlan_uninit(const void *unused __unused) 995 { 996 997 if_clone_detach(V_vlan_cloner); 998 } 999 VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, 1000 vnet_vlan_uninit, NULL); 1001 #endif 1002 1003 /* 1004 * Check for <etherif>.<vlan>[.<vlan> ...] style interface names. 1005 */ 1006 static struct ifnet * 1007 vlan_clone_match_ethervid(const char *name, int *vidp) 1008 { 1009 char ifname[IFNAMSIZ]; 1010 char *cp; 1011 struct ifnet *ifp; 1012 int vid; 1013 1014 strlcpy(ifname, name, IFNAMSIZ); 1015 if ((cp = strrchr(ifname, '.')) == NULL) 1016 return (NULL); 1017 *cp = '\0'; 1018 if ((ifp = ifunit_ref(ifname)) == NULL) 1019 return (NULL); 1020 /* Parse VID. */ 1021 if (*++cp == '\0') { 1022 if_rele(ifp); 1023 return (NULL); 1024 } 1025 vid = 0; 1026 for(; *cp >= '0' && *cp <= '9'; cp++) 1027 vid = (vid * 10) + (*cp - '0'); 1028 if (*cp != '\0') { 1029 if_rele(ifp); 1030 return (NULL); 1031 } 1032 if (vidp != NULL) 1033 *vidp = vid; 1034 1035 return (ifp); 1036 } 1037 1038 static int 1039 vlan_clone_match(struct if_clone *ifc, const char *name) 1040 { 1041 struct ifnet *ifp; 1042 const char *cp; 1043 1044 ifp = vlan_clone_match_ethervid(name, NULL); 1045 if (ifp != NULL) { 1046 if_rele(ifp); 1047 return (1); 1048 } 1049 1050 if (strncmp(vlanname, name, strlen(vlanname)) != 0) 1051 return (0); 1052 for (cp = name + 4; *cp != '\0'; cp++) { 1053 if (*cp < '0' || *cp > '9') 1054 return (0); 1055 } 1056 1057 return (1); 1058 } 1059 1060 static int 1061 vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params) 1062 { 1063 char *dp; 1064 bool wildcard = false; 1065 bool subinterface = false; 1066 int unit; 1067 int error; 1068 int vid = 0; 1069 uint16_t proto = ETHERTYPE_VLAN; 1070 struct ifvlan *ifv; 1071 struct ifnet *ifp; 1072 struct ifnet *p = NULL; 1073 struct ifaddr *ifa; 1074 struct sockaddr_dl *sdl; 1075 struct vlanreq vlr; 1076 static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */ 1077 1078 1079 /* 1080 * There are three ways to specify the cloned device: 1081 * o pass a parameter block with the clone request. 1082 * o specify parameters in the text of the clone device name 1083 * o specify no parameters and get an unattached device that 1084 * must be configured separately. 1085 * The first technique is preferred; the latter two are supported 1086 * for backwards compatibility. 1087 * 1088 * XXXRW: Note historic use of the word "tag" here. New ioctls may be 1089 * called for. 1090 */ 1091 1092 if (params) { 1093 error = copyin(params, &vlr, sizeof(vlr)); 1094 if (error) 1095 return error; 1096 vid = vlr.vlr_tag; 1097 proto = vlr.vlr_proto; 1098 1099 #ifdef COMPAT_FREEBSD12 1100 if (proto == 0) 1101 proto = ETHERTYPE_VLAN; 1102 #endif 1103 p = ifunit_ref(vlr.vlr_parent); 1104 if (p == NULL) 1105 return (ENXIO); 1106 } 1107 1108 if ((error = ifc_name2unit(name, &unit)) == 0) { 1109 1110 /* 1111 * vlanX interface. Set wildcard to true if the unit number 1112 * is not fixed (-1) 1113 */ 1114 wildcard = (unit < 0); 1115 } else { 1116 struct ifnet *p_tmp = vlan_clone_match_ethervid(name, &vid); 1117 if (p_tmp != NULL) { 1118 error = 0; 1119 subinterface = true; 1120 unit = IF_DUNIT_NONE; 1121 wildcard = false; 1122 if (p != NULL) { 1123 if_rele(p_tmp); 1124 if (p != p_tmp) 1125 error = EINVAL; 1126 } else 1127 p = p_tmp; 1128 } else 1129 error = ENXIO; 1130 } 1131 1132 if (error != 0) { 1133 if (p != NULL) 1134 if_rele(p); 1135 return (error); 1136 } 1137 1138 if (!subinterface) { 1139 /* vlanX interface, mark X as busy or allocate new unit # */ 1140 error = ifc_alloc_unit(ifc, &unit); 1141 if (error != 0) { 1142 if (p != NULL) 1143 if_rele(p); 1144 return (error); 1145 } 1146 } 1147 1148 /* In the wildcard case, we need to update the name. */ 1149 if (wildcard) { 1150 for (dp = name; *dp != '\0'; dp++); 1151 if (snprintf(dp, len - (dp-name), "%d", unit) > 1152 len - (dp-name) - 1) { 1153 panic("%s: interface name too long", __func__); 1154 } 1155 } 1156 1157 ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO); 1158 ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER); 1159 if (ifp == NULL) { 1160 if (!subinterface) 1161 ifc_free_unit(ifc, unit); 1162 free(ifv, M_VLAN); 1163 if (p != NULL) 1164 if_rele(p); 1165 return (ENOSPC); 1166 } 1167 CK_SLIST_INIT(&ifv->vlan_mc_listhead); 1168 ifp->if_softc = ifv; 1169 /* 1170 * Set the name manually rather than using if_initname because 1171 * we don't conform to the default naming convention for interfaces. 1172 */ 1173 strlcpy(ifp->if_xname, name, IFNAMSIZ); 1174 ifp->if_dname = vlanname; 1175 ifp->if_dunit = unit; 1176 1177 ifp->if_init = vlan_init; 1178 #ifdef ALTQ 1179 ifp->if_start = vlan_altq_start; 1180 ifp->if_transmit = vlan_altq_transmit; 1181 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 1182 ifp->if_snd.ifq_drv_maxlen = 0; 1183 IFQ_SET_READY(&ifp->if_snd); 1184 #else 1185 ifp->if_transmit = vlan_transmit; 1186 #endif 1187 ifp->if_qflush = vlan_qflush; 1188 ifp->if_ioctl = vlan_ioctl; 1189 #if defined(KERN_TLS) || defined(RATELIMIT) 1190 ifp->if_snd_tag_alloc = vlan_snd_tag_alloc; 1191 ifp->if_ratelimit_query = vlan_ratelimit_query; 1192 #endif 1193 ifp->if_flags = VLAN_IFFLAGS; 1194 ether_ifattach(ifp, eaddr); 1195 /* Now undo some of the damage... */ 1196 ifp->if_baudrate = 0; 1197 ifp->if_type = IFT_L2VLAN; 1198 ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN; 1199 ifa = ifp->if_addr; 1200 sdl = (struct sockaddr_dl *)ifa->ifa_addr; 1201 sdl->sdl_type = IFT_L2VLAN; 1202 1203 if (p != NULL) { 1204 error = vlan_config(ifv, p, vid, proto); 1205 if_rele(p); 1206 if (error != 0) { 1207 /* 1208 * Since we've partially failed, we need to back 1209 * out all the way, otherwise userland could get 1210 * confused. Thus, we destroy the interface. 1211 */ 1212 ether_ifdetach(ifp); 1213 vlan_unconfig(ifp); 1214 if_free(ifp); 1215 if (!subinterface) 1216 ifc_free_unit(ifc, unit); 1217 free(ifv, M_VLAN); 1218 1219 return (error); 1220 } 1221 } 1222 1223 return (0); 1224 } 1225 1226 static int 1227 vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp) 1228 { 1229 struct ifvlan *ifv = ifp->if_softc; 1230 int unit = ifp->if_dunit; 1231 1232 if (ifp->if_vlantrunk) 1233 return (EBUSY); 1234 1235 #ifdef ALTQ 1236 IFQ_PURGE(&ifp->if_snd); 1237 #endif 1238 ether_ifdetach(ifp); /* first, remove it from system-wide lists */ 1239 vlan_unconfig(ifp); /* now it can be unconfigured and freed */ 1240 /* 1241 * We should have the only reference to the ifv now, so we can now 1242 * drain any remaining lladdr task before freeing the ifnet and the 1243 * ifvlan. 1244 */ 1245 taskqueue_drain(taskqueue_thread, &ifv->lladdr_task); 1246 NET_EPOCH_WAIT(); 1247 if_free(ifp); 1248 free(ifv, M_VLAN); 1249 if (unit != IF_DUNIT_NONE) 1250 ifc_free_unit(ifc, unit); 1251 1252 return (0); 1253 } 1254 1255 /* 1256 * The ifp->if_init entry point for vlan(4) is a no-op. 1257 */ 1258 static void 1259 vlan_init(void *foo __unused) 1260 { 1261 } 1262 1263 /* 1264 * The if_transmit method for vlan(4) interface. 1265 */ 1266 static int 1267 vlan_transmit(struct ifnet *ifp, struct mbuf *m) 1268 { 1269 struct ifvlan *ifv; 1270 struct ifnet *p; 1271 int error, len, mcast; 1272 1273 NET_EPOCH_ASSERT(); 1274 1275 ifv = ifp->if_softc; 1276 if (TRUNK(ifv) == NULL) { 1277 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1278 m_freem(m); 1279 return (ENETDOWN); 1280 } 1281 p = PARENT(ifv); 1282 len = m->m_pkthdr.len; 1283 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1284 1285 BPF_MTAP(ifp, m); 1286 1287 #if defined(KERN_TLS) || defined(RATELIMIT) 1288 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) { 1289 struct vlan_snd_tag *vst; 1290 struct m_snd_tag *mst; 1291 1292 MPASS(m->m_pkthdr.snd_tag->ifp == ifp); 1293 mst = m->m_pkthdr.snd_tag; 1294 vst = mst_to_vst(mst); 1295 if (vst->tag->ifp != p) { 1296 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1297 m_freem(m); 1298 return (EAGAIN); 1299 } 1300 1301 m->m_pkthdr.snd_tag = m_snd_tag_ref(vst->tag); 1302 m_snd_tag_rele(mst); 1303 } 1304 #endif 1305 1306 /* 1307 * Do not run parent's if_transmit() if the parent is not up, 1308 * or parent's driver will cause a system crash. 1309 */ 1310 if (!UP_AND_RUNNING(p)) { 1311 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1312 m_freem(m); 1313 return (ENETDOWN); 1314 } 1315 1316 if (!ether_8021q_frame(&m, ifp, p, &ifv->ifv_qtag)) { 1317 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1318 return (0); 1319 } 1320 1321 /* 1322 * Send it, precisely as ether_output() would have. 1323 */ 1324 error = (p->if_transmit)(p, m); 1325 if (error == 0) { 1326 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); 1327 if_inc_counter(ifp, IFCOUNTER_OBYTES, len); 1328 if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast); 1329 } else 1330 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1331 return (error); 1332 } 1333 1334 static int 1335 vlan_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, 1336 struct route *ro) 1337 { 1338 struct ifvlan *ifv; 1339 struct ifnet *p; 1340 1341 NET_EPOCH_ASSERT(); 1342 1343 /* 1344 * Find the first non-VLAN parent interface. 1345 */ 1346 ifv = ifp->if_softc; 1347 do { 1348 if (TRUNK(ifv) == NULL) { 1349 m_freem(m); 1350 return (ENETDOWN); 1351 } 1352 p = PARENT(ifv); 1353 ifv = p->if_softc; 1354 } while (p->if_type == IFT_L2VLAN); 1355 1356 return p->if_output(ifp, m, dst, ro); 1357 } 1358 1359 #ifdef ALTQ 1360 static void 1361 vlan_altq_start(if_t ifp) 1362 { 1363 struct ifaltq *ifq = &ifp->if_snd; 1364 struct mbuf *m; 1365 1366 IFQ_LOCK(ifq); 1367 IFQ_DEQUEUE_NOLOCK(ifq, m); 1368 while (m != NULL) { 1369 vlan_transmit(ifp, m); 1370 IFQ_DEQUEUE_NOLOCK(ifq, m); 1371 } 1372 IFQ_UNLOCK(ifq); 1373 } 1374 1375 static int 1376 vlan_altq_transmit(if_t ifp, struct mbuf *m) 1377 { 1378 int err; 1379 1380 if (ALTQ_IS_ENABLED(&ifp->if_snd)) { 1381 IFQ_ENQUEUE(&ifp->if_snd, m, err); 1382 if (err == 0) 1383 vlan_altq_start(ifp); 1384 } else 1385 err = vlan_transmit(ifp, m); 1386 1387 return (err); 1388 } 1389 #endif /* ALTQ */ 1390 1391 /* 1392 * The ifp->if_qflush entry point for vlan(4) is a no-op. 1393 */ 1394 static void 1395 vlan_qflush(struct ifnet *ifp __unused) 1396 { 1397 } 1398 1399 static void 1400 vlan_input(struct ifnet *ifp, struct mbuf *m) 1401 { 1402 struct ifvlantrunk *trunk; 1403 struct ifvlan *ifv; 1404 struct m_tag *mtag; 1405 uint16_t vid, tag; 1406 1407 NET_EPOCH_ASSERT(); 1408 1409 trunk = ifp->if_vlantrunk; 1410 if (trunk == NULL) { 1411 m_freem(m); 1412 return; 1413 } 1414 1415 if (m->m_flags & M_VLANTAG) { 1416 /* 1417 * Packet is tagged, but m contains a normal 1418 * Ethernet frame; the tag is stored out-of-band. 1419 */ 1420 tag = m->m_pkthdr.ether_vtag; 1421 m->m_flags &= ~M_VLANTAG; 1422 } else { 1423 struct ether_vlan_header *evl; 1424 1425 /* 1426 * Packet is tagged in-band as specified by 802.1q. 1427 */ 1428 switch (ifp->if_type) { 1429 case IFT_ETHER: 1430 if (m->m_len < sizeof(*evl) && 1431 (m = m_pullup(m, sizeof(*evl))) == NULL) { 1432 if_printf(ifp, "cannot pullup VLAN header\n"); 1433 return; 1434 } 1435 evl = mtod(m, struct ether_vlan_header *); 1436 tag = ntohs(evl->evl_tag); 1437 1438 /* 1439 * Remove the 802.1q header by copying the Ethernet 1440 * addresses over it and adjusting the beginning of 1441 * the data in the mbuf. The encapsulated Ethernet 1442 * type field is already in place. 1443 */ 1444 bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN, 1445 ETHER_HDR_LEN - ETHER_TYPE_LEN); 1446 m_adj(m, ETHER_VLAN_ENCAP_LEN); 1447 break; 1448 1449 default: 1450 #ifdef INVARIANTS 1451 panic("%s: %s has unsupported if_type %u", 1452 __func__, ifp->if_xname, ifp->if_type); 1453 #endif 1454 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1); 1455 m_freem(m); 1456 return; 1457 } 1458 } 1459 1460 vid = EVL_VLANOFTAG(tag); 1461 1462 ifv = vlan_gethash(trunk, vid); 1463 if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) { 1464 if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1); 1465 m_freem(m); 1466 return; 1467 } 1468 1469 if (V_vlan_mtag_pcp) { 1470 /* 1471 * While uncommon, it is possible that we will find a 802.1q 1472 * packet encapsulated inside another packet that also had an 1473 * 802.1q header. For example, ethernet tunneled over IPSEC 1474 * arriving over ethernet. In that case, we replace the 1475 * existing 802.1q PCP m_tag value. 1476 */ 1477 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 1478 if (mtag == NULL) { 1479 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN, 1480 sizeof(uint8_t), M_NOWAIT); 1481 if (mtag == NULL) { 1482 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1); 1483 m_freem(m); 1484 return; 1485 } 1486 m_tag_prepend(m, mtag); 1487 } 1488 *(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag); 1489 } 1490 1491 m->m_pkthdr.rcvif = ifv->ifv_ifp; 1492 if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1); 1493 1494 /* Pass it back through the parent's input routine. */ 1495 (*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m); 1496 } 1497 1498 static void 1499 vlan_lladdr_fn(void *arg, int pending __unused) 1500 { 1501 struct ifvlan *ifv; 1502 struct ifnet *ifp; 1503 1504 ifv = (struct ifvlan *)arg; 1505 ifp = ifv->ifv_ifp; 1506 1507 CURVNET_SET(ifp->if_vnet); 1508 1509 /* The ifv_ifp already has the lladdr copied in. */ 1510 if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen); 1511 1512 CURVNET_RESTORE(); 1513 } 1514 1515 static int 1516 vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid, 1517 uint16_t proto) 1518 { 1519 struct epoch_tracker et; 1520 struct ifvlantrunk *trunk; 1521 struct ifnet *ifp; 1522 int error = 0; 1523 1524 /* 1525 * We can handle non-ethernet hardware types as long as 1526 * they handle the tagging and headers themselves. 1527 */ 1528 if (p->if_type != IFT_ETHER && 1529 p->if_type != IFT_L2VLAN && 1530 (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0) 1531 return (EPROTONOSUPPORT); 1532 if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS) 1533 return (EPROTONOSUPPORT); 1534 /* 1535 * Don't let the caller set up a VLAN VID with 1536 * anything except VLID bits. 1537 * VID numbers 0x0 and 0xFFF are reserved. 1538 */ 1539 if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK)) 1540 return (EINVAL); 1541 if (ifv->ifv_trunk) { 1542 trunk = ifv->ifv_trunk; 1543 if (trunk->parent != p) 1544 return (EBUSY); 1545 1546 VLAN_XLOCK(); 1547 1548 ifv->ifv_proto = proto; 1549 1550 if (ifv->ifv_vid != vid) { 1551 /* Re-hash */ 1552 vlan_remhash(trunk, ifv); 1553 ifv->ifv_vid = vid; 1554 error = vlan_inshash(trunk, ifv); 1555 } 1556 /* Will unlock */ 1557 goto done; 1558 } 1559 1560 VLAN_XLOCK(); 1561 if (p->if_vlantrunk == NULL) { 1562 trunk = malloc(sizeof(struct ifvlantrunk), 1563 M_VLAN, M_WAITOK | M_ZERO); 1564 vlan_inithash(trunk); 1565 TRUNK_LOCK_INIT(trunk); 1566 TRUNK_WLOCK(trunk); 1567 p->if_vlantrunk = trunk; 1568 trunk->parent = p; 1569 if_ref(trunk->parent); 1570 TRUNK_WUNLOCK(trunk); 1571 } else { 1572 trunk = p->if_vlantrunk; 1573 } 1574 1575 ifv->ifv_vid = vid; /* must set this before vlan_inshash() */ 1576 ifv->ifv_pcp = 0; /* Default: best effort delivery. */ 1577 error = vlan_inshash(trunk, ifv); 1578 if (error) 1579 goto done; 1580 ifv->ifv_proto = proto; 1581 ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN; 1582 ifv->ifv_mintu = ETHERMIN; 1583 ifv->ifv_pflags = 0; 1584 ifv->ifv_capenable = -1; 1585 1586 /* 1587 * If the parent supports the VLAN_MTU capability, 1588 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames, 1589 * use it. 1590 */ 1591 if (p->if_capenable & IFCAP_VLAN_MTU) { 1592 /* 1593 * No need to fudge the MTU since the parent can 1594 * handle extended frames. 1595 */ 1596 ifv->ifv_mtufudge = 0; 1597 } else { 1598 /* 1599 * Fudge the MTU by the encapsulation size. This 1600 * makes us incompatible with strictly compliant 1601 * 802.1Q implementations, but allows us to use 1602 * the feature with other NetBSD implementations, 1603 * which might still be useful. 1604 */ 1605 ifv->ifv_mtufudge = ifv->ifv_encaplen; 1606 } 1607 1608 ifv->ifv_trunk = trunk; 1609 ifp = ifv->ifv_ifp; 1610 /* 1611 * Initialize fields from our parent. This duplicates some 1612 * work with ether_ifattach() but allows for non-ethernet 1613 * interfaces to also work. 1614 */ 1615 ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge; 1616 ifp->if_baudrate = p->if_baudrate; 1617 ifp->if_input = p->if_input; 1618 ifp->if_resolvemulti = p->if_resolvemulti; 1619 ifp->if_addrlen = p->if_addrlen; 1620 ifp->if_broadcastaddr = p->if_broadcastaddr; 1621 ifp->if_pcp = ifv->ifv_pcp; 1622 1623 /* 1624 * We wrap the parent's if_output using vlan_output to ensure that it 1625 * can't become stale. 1626 */ 1627 ifp->if_output = vlan_output; 1628 1629 /* 1630 * Copy only a selected subset of flags from the parent. 1631 * Other flags are none of our business. 1632 */ 1633 #define VLAN_COPY_FLAGS (IFF_SIMPLEX) 1634 ifp->if_flags &= ~VLAN_COPY_FLAGS; 1635 ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS; 1636 #undef VLAN_COPY_FLAGS 1637 1638 ifp->if_link_state = p->if_link_state; 1639 1640 NET_EPOCH_ENTER(et); 1641 vlan_capabilities(ifv); 1642 NET_EPOCH_EXIT(et); 1643 1644 /* 1645 * Set up our interface address to reflect the underlying 1646 * physical interface's. 1647 */ 1648 TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv); 1649 ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen = 1650 p->if_addrlen; 1651 1652 /* 1653 * Do not schedule link address update if it was the same 1654 * as previous parent's. This helps avoid updating for each 1655 * associated llentry. 1656 */ 1657 if (memcmp(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen) != 0) { 1658 bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen); 1659 taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task); 1660 } 1661 1662 /* We are ready for operation now. */ 1663 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1664 1665 /* Update flags on the parent, if necessary. */ 1666 vlan_setflags(ifp, 1); 1667 1668 /* 1669 * Configure multicast addresses that may already be 1670 * joined on the vlan device. 1671 */ 1672 (void)vlan_setmulti(ifp); 1673 1674 done: 1675 if (error == 0) 1676 EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid); 1677 VLAN_XUNLOCK(); 1678 1679 return (error); 1680 } 1681 1682 static void 1683 vlan_unconfig(struct ifnet *ifp) 1684 { 1685 1686 VLAN_XLOCK(); 1687 vlan_unconfig_locked(ifp, 0); 1688 VLAN_XUNLOCK(); 1689 } 1690 1691 static void 1692 vlan_unconfig_locked(struct ifnet *ifp, int departing) 1693 { 1694 struct ifvlantrunk *trunk; 1695 struct vlan_mc_entry *mc; 1696 struct ifvlan *ifv; 1697 struct ifnet *parent; 1698 int error; 1699 1700 VLAN_XLOCK_ASSERT(); 1701 1702 ifv = ifp->if_softc; 1703 trunk = ifv->ifv_trunk; 1704 parent = NULL; 1705 1706 if (trunk != NULL) { 1707 parent = trunk->parent; 1708 1709 /* 1710 * Since the interface is being unconfigured, we need to 1711 * empty the list of multicast groups that we may have joined 1712 * while we were alive from the parent's list. 1713 */ 1714 while ((mc = CK_SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) { 1715 /* 1716 * If the parent interface is being detached, 1717 * all its multicast addresses have already 1718 * been removed. Warn about errors if 1719 * if_delmulti() does fail, but don't abort as 1720 * all callers expect vlan destruction to 1721 * succeed. 1722 */ 1723 if (!departing) { 1724 error = if_delmulti(parent, 1725 (struct sockaddr *)&mc->mc_addr); 1726 if (error) 1727 if_printf(ifp, 1728 "Failed to delete multicast address from parent: %d\n", 1729 error); 1730 } 1731 CK_SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries); 1732 NET_EPOCH_CALL(vlan_mc_free, &mc->mc_epoch_ctx); 1733 } 1734 1735 vlan_setflags(ifp, 0); /* clear special flags on parent */ 1736 1737 vlan_remhash(trunk, ifv); 1738 ifv->ifv_trunk = NULL; 1739 1740 /* 1741 * Check if we were the last. 1742 */ 1743 if (trunk->refcnt == 0) { 1744 parent->if_vlantrunk = NULL; 1745 NET_EPOCH_WAIT(); 1746 trunk_destroy(trunk); 1747 } 1748 } 1749 1750 /* Disconnect from parent. */ 1751 if (ifv->ifv_pflags) 1752 if_printf(ifp, "%s: ifv_pflags unclean\n", __func__); 1753 ifp->if_mtu = ETHERMTU; 1754 ifp->if_link_state = LINK_STATE_UNKNOWN; 1755 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1756 1757 /* 1758 * Only dispatch an event if vlan was 1759 * attached, otherwise there is nothing 1760 * to cleanup anyway. 1761 */ 1762 if (parent != NULL) 1763 EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid); 1764 } 1765 1766 /* Handle a reference counted flag that should be set on the parent as well */ 1767 static int 1768 vlan_setflag(struct ifnet *ifp, int flag, int status, 1769 int (*func)(struct ifnet *, int)) 1770 { 1771 struct ifvlan *ifv; 1772 int error; 1773 1774 VLAN_SXLOCK_ASSERT(); 1775 1776 ifv = ifp->if_softc; 1777 status = status ? (ifp->if_flags & flag) : 0; 1778 /* Now "status" contains the flag value or 0 */ 1779 1780 /* 1781 * See if recorded parent's status is different from what 1782 * we want it to be. If it is, flip it. We record parent's 1783 * status in ifv_pflags so that we won't clear parent's flag 1784 * we haven't set. In fact, we don't clear or set parent's 1785 * flags directly, but get or release references to them. 1786 * That's why we can be sure that recorded flags still are 1787 * in accord with actual parent's flags. 1788 */ 1789 if (status != (ifv->ifv_pflags & flag)) { 1790 error = (*func)(PARENT(ifv), status); 1791 if (error) 1792 return (error); 1793 ifv->ifv_pflags &= ~flag; 1794 ifv->ifv_pflags |= status; 1795 } 1796 return (0); 1797 } 1798 1799 /* 1800 * Handle IFF_* flags that require certain changes on the parent: 1801 * if "status" is true, update parent's flags respective to our if_flags; 1802 * if "status" is false, forcedly clear the flags set on parent. 1803 */ 1804 static int 1805 vlan_setflags(struct ifnet *ifp, int status) 1806 { 1807 int error, i; 1808 1809 for (i = 0; vlan_pflags[i].flag; i++) { 1810 error = vlan_setflag(ifp, vlan_pflags[i].flag, 1811 status, vlan_pflags[i].func); 1812 if (error) 1813 return (error); 1814 } 1815 return (0); 1816 } 1817 1818 /* Inform all vlans that their parent has changed link state */ 1819 static void 1820 vlan_link_state(struct ifnet *ifp) 1821 { 1822 struct epoch_tracker et; 1823 struct ifvlantrunk *trunk; 1824 struct ifvlan *ifv; 1825 1826 NET_EPOCH_ENTER(et); 1827 trunk = ifp->if_vlantrunk; 1828 if (trunk == NULL) { 1829 NET_EPOCH_EXIT(et); 1830 return; 1831 } 1832 1833 TRUNK_WLOCK(trunk); 1834 VLAN_FOREACH(ifv, trunk) { 1835 ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate; 1836 if_link_state_change(ifv->ifv_ifp, 1837 trunk->parent->if_link_state); 1838 } 1839 TRUNK_WUNLOCK(trunk); 1840 NET_EPOCH_EXIT(et); 1841 } 1842 1843 static void 1844 vlan_capabilities(struct ifvlan *ifv) 1845 { 1846 struct ifnet *p; 1847 struct ifnet *ifp; 1848 struct ifnet_hw_tsomax hw_tsomax; 1849 int cap = 0, ena = 0, mena; 1850 u_long hwa = 0; 1851 1852 NET_EPOCH_ASSERT(); 1853 VLAN_SXLOCK_ASSERT(); 1854 1855 p = PARENT(ifv); 1856 ifp = ifv->ifv_ifp; 1857 1858 /* Mask parent interface enabled capabilities disabled by user. */ 1859 mena = p->if_capenable & ifv->ifv_capenable; 1860 1861 /* 1862 * If the parent interface can do checksum offloading 1863 * on VLANs, then propagate its hardware-assisted 1864 * checksumming flags. Also assert that checksum 1865 * offloading requires hardware VLAN tagging. 1866 */ 1867 if (p->if_capabilities & IFCAP_VLAN_HWCSUM) 1868 cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6); 1869 if (p->if_capenable & IFCAP_VLAN_HWCSUM && 1870 p->if_capenable & IFCAP_VLAN_HWTAGGING) { 1871 ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6); 1872 if (ena & IFCAP_TXCSUM) 1873 hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP | 1874 CSUM_UDP | CSUM_SCTP); 1875 if (ena & IFCAP_TXCSUM_IPV6) 1876 hwa |= p->if_hwassist & (CSUM_TCP_IPV6 | 1877 CSUM_UDP_IPV6 | CSUM_SCTP_IPV6); 1878 } 1879 1880 /* 1881 * If the parent interface can do TSO on VLANs then 1882 * propagate the hardware-assisted flag. TSO on VLANs 1883 * does not necessarily require hardware VLAN tagging. 1884 */ 1885 memset(&hw_tsomax, 0, sizeof(hw_tsomax)); 1886 if_hw_tsomax_common(p, &hw_tsomax); 1887 if_hw_tsomax_update(ifp, &hw_tsomax); 1888 if (p->if_capabilities & IFCAP_VLAN_HWTSO) 1889 cap |= p->if_capabilities & IFCAP_TSO; 1890 if (p->if_capenable & IFCAP_VLAN_HWTSO) { 1891 ena |= mena & IFCAP_TSO; 1892 if (ena & IFCAP_TSO) 1893 hwa |= p->if_hwassist & CSUM_TSO; 1894 } 1895 1896 /* 1897 * If the parent interface can do LRO and checksum offloading on 1898 * VLANs, then guess it may do LRO on VLANs. False positive here 1899 * cost nothing, while false negative may lead to some confusions. 1900 */ 1901 if (p->if_capabilities & IFCAP_VLAN_HWCSUM) 1902 cap |= p->if_capabilities & IFCAP_LRO; 1903 if (p->if_capenable & IFCAP_VLAN_HWCSUM) 1904 ena |= p->if_capenable & IFCAP_LRO; 1905 1906 /* 1907 * If the parent interface can offload TCP connections over VLANs then 1908 * propagate its TOE capability to the VLAN interface. 1909 * 1910 * All TOE drivers in the tree today can deal with VLANs. If this 1911 * changes then IFCAP_VLAN_TOE should be promoted to a full capability 1912 * with its own bit. 1913 */ 1914 #define IFCAP_VLAN_TOE IFCAP_TOE 1915 if (p->if_capabilities & IFCAP_VLAN_TOE) 1916 cap |= p->if_capabilities & IFCAP_TOE; 1917 if (p->if_capenable & IFCAP_VLAN_TOE) { 1918 TOEDEV(ifp) = TOEDEV(p); 1919 ena |= mena & IFCAP_TOE; 1920 } 1921 1922 /* 1923 * If the parent interface supports dynamic link state, so does the 1924 * VLAN interface. 1925 */ 1926 cap |= (p->if_capabilities & IFCAP_LINKSTATE); 1927 ena |= (mena & IFCAP_LINKSTATE); 1928 1929 #ifdef RATELIMIT 1930 /* 1931 * If the parent interface supports ratelimiting, so does the 1932 * VLAN interface. 1933 */ 1934 cap |= (p->if_capabilities & IFCAP_TXRTLMT); 1935 ena |= (mena & IFCAP_TXRTLMT); 1936 #endif 1937 1938 /* 1939 * If the parent interface supports unmapped mbufs, so does 1940 * the VLAN interface. Note that this should be fine even for 1941 * interfaces that don't support hardware tagging as headers 1942 * are prepended in normal mbufs to unmapped mbufs holding 1943 * payload data. 1944 */ 1945 cap |= (p->if_capabilities & IFCAP_MEXTPG); 1946 ena |= (mena & IFCAP_MEXTPG); 1947 1948 /* 1949 * If the parent interface can offload encryption and segmentation 1950 * of TLS records over TCP, propagate it's capability to the VLAN 1951 * interface. 1952 * 1953 * All TLS drivers in the tree today can deal with VLANs. If 1954 * this ever changes, then a new IFCAP_VLAN_TXTLS can be 1955 * defined. 1956 */ 1957 if (p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT)) 1958 cap |= p->if_capabilities & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT); 1959 if (p->if_capenable & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT)) 1960 ena |= mena & (IFCAP_TXTLS | IFCAP_TXTLS_RTLMT); 1961 1962 ifp->if_capabilities = cap; 1963 ifp->if_capenable = ena; 1964 ifp->if_hwassist = hwa; 1965 } 1966 1967 static void 1968 vlan_trunk_capabilities(struct ifnet *ifp) 1969 { 1970 struct epoch_tracker et; 1971 struct ifvlantrunk *trunk; 1972 struct ifvlan *ifv; 1973 1974 VLAN_SLOCK(); 1975 trunk = ifp->if_vlantrunk; 1976 if (trunk == NULL) { 1977 VLAN_SUNLOCK(); 1978 return; 1979 } 1980 NET_EPOCH_ENTER(et); 1981 VLAN_FOREACH(ifv, trunk) 1982 vlan_capabilities(ifv); 1983 NET_EPOCH_EXIT(et); 1984 VLAN_SUNLOCK(); 1985 } 1986 1987 static int 1988 vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1989 { 1990 struct ifnet *p; 1991 struct ifreq *ifr; 1992 #ifdef INET 1993 struct ifaddr *ifa; 1994 #endif 1995 struct ifvlan *ifv; 1996 struct ifvlantrunk *trunk; 1997 struct vlanreq vlr; 1998 int error = 0, oldmtu; 1999 2000 ifr = (struct ifreq *)data; 2001 #ifdef INET 2002 ifa = (struct ifaddr *) data; 2003 #endif 2004 ifv = ifp->if_softc; 2005 2006 switch (cmd) { 2007 case SIOCSIFADDR: 2008 ifp->if_flags |= IFF_UP; 2009 #ifdef INET 2010 if (ifa->ifa_addr->sa_family == AF_INET) 2011 arp_ifinit(ifp, ifa); 2012 #endif 2013 break; 2014 case SIOCGIFADDR: 2015 bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0], 2016 ifp->if_addrlen); 2017 break; 2018 case SIOCGIFMEDIA: 2019 VLAN_SLOCK(); 2020 if (TRUNK(ifv) != NULL) { 2021 p = PARENT(ifv); 2022 if_ref(p); 2023 error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data); 2024 if_rele(p); 2025 /* Limit the result to the parent's current config. */ 2026 if (error == 0) { 2027 struct ifmediareq *ifmr; 2028 2029 ifmr = (struct ifmediareq *)data; 2030 if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) { 2031 ifmr->ifm_count = 1; 2032 error = copyout(&ifmr->ifm_current, 2033 ifmr->ifm_ulist, 2034 sizeof(int)); 2035 } 2036 } 2037 } else { 2038 error = EINVAL; 2039 } 2040 VLAN_SUNLOCK(); 2041 break; 2042 2043 case SIOCSIFMEDIA: 2044 error = EINVAL; 2045 break; 2046 2047 case SIOCSIFMTU: 2048 /* 2049 * Set the interface MTU. 2050 */ 2051 VLAN_SLOCK(); 2052 trunk = TRUNK(ifv); 2053 if (trunk != NULL) { 2054 TRUNK_WLOCK(trunk); 2055 if (ifr->ifr_mtu > 2056 (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) || 2057 ifr->ifr_mtu < 2058 (ifv->ifv_mintu - ifv->ifv_mtufudge)) 2059 error = EINVAL; 2060 else 2061 ifp->if_mtu = ifr->ifr_mtu; 2062 TRUNK_WUNLOCK(trunk); 2063 } else 2064 error = EINVAL; 2065 VLAN_SUNLOCK(); 2066 break; 2067 2068 case SIOCSETVLAN: 2069 #ifdef VIMAGE 2070 /* 2071 * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN 2072 * interface to be delegated to a jail without allowing the 2073 * jail to change what underlying interface/VID it is 2074 * associated with. We are not entirely convinced that this 2075 * is the right way to accomplish that policy goal. 2076 */ 2077 if (ifp->if_vnet != ifp->if_home_vnet) { 2078 error = EPERM; 2079 break; 2080 } 2081 #endif 2082 error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr)); 2083 if (error) 2084 break; 2085 if (vlr.vlr_parent[0] == '\0') { 2086 vlan_unconfig(ifp); 2087 break; 2088 } 2089 p = ifunit_ref(vlr.vlr_parent); 2090 if (p == NULL) { 2091 error = ENOENT; 2092 break; 2093 } 2094 #ifdef COMPAT_FREEBSD12 2095 if (vlr.vlr_proto == 0) 2096 vlr.vlr_proto = ETHERTYPE_VLAN; 2097 #endif 2098 oldmtu = ifp->if_mtu; 2099 error = vlan_config(ifv, p, vlr.vlr_tag, vlr.vlr_proto); 2100 if_rele(p); 2101 2102 /* 2103 * VLAN MTU may change during addition of the vlandev. 2104 * If it did, do network layer specific procedure. 2105 */ 2106 if (ifp->if_mtu != oldmtu) { 2107 #ifdef INET6 2108 nd6_setmtu(ifp); 2109 #endif 2110 rt_updatemtu(ifp); 2111 } 2112 break; 2113 2114 case SIOCGETVLAN: 2115 #ifdef VIMAGE 2116 if (ifp->if_vnet != ifp->if_home_vnet) { 2117 error = EPERM; 2118 break; 2119 } 2120 #endif 2121 bzero(&vlr, sizeof(vlr)); 2122 VLAN_SLOCK(); 2123 if (TRUNK(ifv) != NULL) { 2124 strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname, 2125 sizeof(vlr.vlr_parent)); 2126 vlr.vlr_tag = ifv->ifv_vid; 2127 vlr.vlr_proto = ifv->ifv_proto; 2128 } 2129 VLAN_SUNLOCK(); 2130 error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr)); 2131 break; 2132 2133 case SIOCSIFFLAGS: 2134 /* 2135 * We should propagate selected flags to the parent, 2136 * e.g., promiscuous mode. 2137 */ 2138 VLAN_XLOCK(); 2139 if (TRUNK(ifv) != NULL) 2140 error = vlan_setflags(ifp, 1); 2141 VLAN_XUNLOCK(); 2142 break; 2143 2144 case SIOCADDMULTI: 2145 case SIOCDELMULTI: 2146 /* 2147 * If we don't have a parent, just remember the membership for 2148 * when we do. 2149 * 2150 * XXX We need the rmlock here to avoid sleeping while 2151 * holding in6_multi_mtx. 2152 */ 2153 VLAN_XLOCK(); 2154 trunk = TRUNK(ifv); 2155 if (trunk != NULL) 2156 error = vlan_setmulti(ifp); 2157 VLAN_XUNLOCK(); 2158 2159 break; 2160 case SIOCGVLANPCP: 2161 #ifdef VIMAGE 2162 if (ifp->if_vnet != ifp->if_home_vnet) { 2163 error = EPERM; 2164 break; 2165 } 2166 #endif 2167 ifr->ifr_vlan_pcp = ifv->ifv_pcp; 2168 break; 2169 2170 case SIOCSVLANPCP: 2171 #ifdef VIMAGE 2172 if (ifp->if_vnet != ifp->if_home_vnet) { 2173 error = EPERM; 2174 break; 2175 } 2176 #endif 2177 error = priv_check(curthread, PRIV_NET_SETVLANPCP); 2178 if (error) 2179 break; 2180 if (ifr->ifr_vlan_pcp > VLAN_PCP_MAX) { 2181 error = EINVAL; 2182 break; 2183 } 2184 ifv->ifv_pcp = ifr->ifr_vlan_pcp; 2185 ifp->if_pcp = ifv->ifv_pcp; 2186 /* broadcast event about PCP change */ 2187 EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP); 2188 break; 2189 2190 case SIOCSIFCAP: 2191 VLAN_SLOCK(); 2192 ifv->ifv_capenable = ifr->ifr_reqcap; 2193 trunk = TRUNK(ifv); 2194 if (trunk != NULL) { 2195 struct epoch_tracker et; 2196 2197 NET_EPOCH_ENTER(et); 2198 vlan_capabilities(ifv); 2199 NET_EPOCH_EXIT(et); 2200 } 2201 VLAN_SUNLOCK(); 2202 break; 2203 2204 default: 2205 error = EINVAL; 2206 break; 2207 } 2208 2209 return (error); 2210 } 2211 2212 #if defined(KERN_TLS) || defined(RATELIMIT) 2213 static int 2214 vlan_snd_tag_alloc(struct ifnet *ifp, 2215 union if_snd_tag_alloc_params *params, 2216 struct m_snd_tag **ppmt) 2217 { 2218 struct epoch_tracker et; 2219 const struct if_snd_tag_sw *sw; 2220 struct vlan_snd_tag *vst; 2221 struct ifvlan *ifv; 2222 struct ifnet *parent; 2223 struct m_snd_tag *mst; 2224 int error; 2225 2226 NET_EPOCH_ENTER(et); 2227 ifv = ifp->if_softc; 2228 2229 switch (params->hdr.type) { 2230 #ifdef RATELIMIT 2231 case IF_SND_TAG_TYPE_UNLIMITED: 2232 sw = &vlan_snd_tag_ul_sw; 2233 break; 2234 case IF_SND_TAG_TYPE_RATE_LIMIT: 2235 sw = &vlan_snd_tag_rl_sw; 2236 break; 2237 #endif 2238 #ifdef KERN_TLS 2239 case IF_SND_TAG_TYPE_TLS: 2240 sw = &vlan_snd_tag_tls_sw; 2241 break; 2242 case IF_SND_TAG_TYPE_TLS_RX: 2243 sw = NULL; 2244 if (params->tls_rx.vlan_id != 0) 2245 goto failure; 2246 params->tls_rx.vlan_id = ifv->ifv_vid; 2247 break; 2248 #ifdef RATELIMIT 2249 case IF_SND_TAG_TYPE_TLS_RATE_LIMIT: 2250 sw = &vlan_snd_tag_tls_rl_sw; 2251 break; 2252 #endif 2253 #endif 2254 default: 2255 goto failure; 2256 } 2257 2258 if (ifv->ifv_trunk != NULL) 2259 parent = PARENT(ifv); 2260 else 2261 parent = NULL; 2262 if (parent == NULL) 2263 goto failure; 2264 if_ref(parent); 2265 NET_EPOCH_EXIT(et); 2266 2267 if (sw != NULL) { 2268 vst = malloc(sizeof(*vst), M_VLAN, M_NOWAIT); 2269 if (vst == NULL) { 2270 if_rele(parent); 2271 return (ENOMEM); 2272 } 2273 } else 2274 vst = NULL; 2275 2276 error = m_snd_tag_alloc(parent, params, &mst); 2277 if_rele(parent); 2278 if (error) { 2279 free(vst, M_VLAN); 2280 return (error); 2281 } 2282 2283 if (sw != NULL) { 2284 m_snd_tag_init(&vst->com, ifp, sw); 2285 vst->tag = mst; 2286 2287 *ppmt = &vst->com; 2288 } else 2289 *ppmt = mst; 2290 2291 return (0); 2292 failure: 2293 NET_EPOCH_EXIT(et); 2294 return (EOPNOTSUPP); 2295 } 2296 2297 static struct m_snd_tag * 2298 vlan_next_snd_tag(struct m_snd_tag *mst) 2299 { 2300 struct vlan_snd_tag *vst; 2301 2302 vst = mst_to_vst(mst); 2303 return (vst->tag); 2304 } 2305 2306 static int 2307 vlan_snd_tag_modify(struct m_snd_tag *mst, 2308 union if_snd_tag_modify_params *params) 2309 { 2310 struct vlan_snd_tag *vst; 2311 2312 vst = mst_to_vst(mst); 2313 return (vst->tag->sw->snd_tag_modify(vst->tag, params)); 2314 } 2315 2316 static int 2317 vlan_snd_tag_query(struct m_snd_tag *mst, 2318 union if_snd_tag_query_params *params) 2319 { 2320 struct vlan_snd_tag *vst; 2321 2322 vst = mst_to_vst(mst); 2323 return (vst->tag->sw->snd_tag_query(vst->tag, params)); 2324 } 2325 2326 static void 2327 vlan_snd_tag_free(struct m_snd_tag *mst) 2328 { 2329 struct vlan_snd_tag *vst; 2330 2331 vst = mst_to_vst(mst); 2332 m_snd_tag_rele(vst->tag); 2333 free(vst, M_VLAN); 2334 } 2335 2336 static void 2337 vlan_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q) 2338 { 2339 /* 2340 * For vlan, we have an indirect 2341 * interface. The caller needs to 2342 * get a ratelimit tag on the actual 2343 * interface the flow will go on. 2344 */ 2345 q->rate_table = NULL; 2346 q->flags = RT_IS_INDIRECT; 2347 q->max_flows = 0; 2348 q->number_of_rates = 0; 2349 } 2350 2351 #endif 2352