1 /* $OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $ */ 2 3 /* 4 * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org> 5 * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org> 6 * 7 * Permission to use, copy, modify, and distribute this software for any 8 * purpose with or without fee is hereby granted, provided that the above 9 * copyright notice and this permission notice appear in all copies. 10 * 11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 18 */ 19 20 #include <sys/cdefs.h> 21 __FBSDID("$FreeBSD$"); 22 23 #include "opt_inet.h" 24 #include "opt_inet6.h" 25 26 #include <sys/param.h> 27 #include <sys/kernel.h> 28 #include <sys/malloc.h> 29 #include <sys/mbuf.h> 30 #include <sys/queue.h> 31 #include <sys/socket.h> 32 #include <sys/sockio.h> 33 #include <sys/sysctl.h> 34 #include <sys/module.h> 35 #include <sys/priv.h> 36 #include <sys/systm.h> 37 #include <sys/proc.h> 38 #include <sys/hash.h> 39 #include <sys/lock.h> 40 #include <sys/rwlock.h> 41 #include <sys/taskqueue.h> 42 #include <sys/eventhandler.h> 43 44 #include <net/ethernet.h> 45 #include <net/if.h> 46 #include <net/if_clone.h> 47 #include <net/if_arp.h> 48 #include <net/if_dl.h> 49 #include <net/if_llc.h> 50 #include <net/if_media.h> 51 #include <net/if_types.h> 52 #include <net/if_var.h> 53 #include <net/bpf.h> 54 55 #if defined(INET) || defined(INET6) 56 #include <netinet/in.h> 57 #endif 58 #ifdef INET 59 #include <netinet/in_systm.h> 60 #include <netinet/if_ether.h> 61 #include <netinet/ip.h> 62 #endif 63 64 #ifdef INET6 65 #include <netinet/ip6.h> 66 #endif 67 68 #include <net/if_vlan_var.h> 69 #include <net/if_lagg.h> 70 #include <net/ieee8023ad_lacp.h> 71 72 /* Special flags we should propagate to the lagg ports. */ 73 static struct { 74 int flag; 75 int (*func)(struct ifnet *, int); 76 } lagg_pflags[] = { 77 {IFF_PROMISC, ifpromisc}, 78 {IFF_ALLMULTI, if_allmulti}, 79 {0, NULL} 80 }; 81 82 SLIST_HEAD(__trhead, lagg_softc) lagg_list; /* list of laggs */ 83 static struct mtx lagg_list_mtx; 84 eventhandler_tag lagg_detach_cookie = NULL; 85 86 static int lagg_clone_create(struct if_clone *, int, caddr_t); 87 static void lagg_clone_destroy(struct ifnet *); 88 static void lagg_lladdr(struct lagg_softc *, uint8_t *); 89 static void lagg_capabilities(struct lagg_softc *); 90 static void lagg_port_lladdr(struct lagg_port *, uint8_t *); 91 static void lagg_port_setlladdr(void *, int); 92 static int lagg_port_create(struct lagg_softc *, struct ifnet *); 93 static int lagg_port_destroy(struct lagg_port *, int); 94 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); 95 static void lagg_linkstate(struct lagg_softc *); 96 static void lagg_port_state(struct ifnet *, int); 97 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); 98 static int lagg_port_output(struct ifnet *, struct mbuf *, 99 struct sockaddr *, struct route *); 100 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 101 #ifdef LAGG_PORT_STACKING 102 static int lagg_port_checkstacking(struct lagg_softc *); 103 #endif 104 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 105 static void lagg_init(void *); 106 static void lagg_stop(struct lagg_softc *); 107 static int lagg_ioctl(struct ifnet *, u_long, caddr_t); 108 static int lagg_ether_setmulti(struct lagg_softc *); 109 static int lagg_ether_cmdmulti(struct lagg_port *, int); 110 static int lagg_setflag(struct lagg_port *, int, int, 111 int (*func)(struct ifnet *, int)); 112 static int lagg_setflags(struct lagg_port *, int status); 113 static void lagg_start(struct ifnet *); 114 static int lagg_media_change(struct ifnet *); 115 static void lagg_media_status(struct ifnet *, struct ifmediareq *); 116 static struct lagg_port *lagg_link_active(struct lagg_softc *, 117 struct lagg_port *); 118 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *); 119 120 IFC_SIMPLE_DECLARE(lagg, 0); 121 122 /* Simple round robin */ 123 static int lagg_rr_attach(struct lagg_softc *); 124 static int lagg_rr_detach(struct lagg_softc *); 125 static int lagg_rr_start(struct lagg_softc *, struct mbuf *); 126 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 127 struct mbuf *); 128 129 /* Active failover */ 130 static int lagg_fail_attach(struct lagg_softc *); 131 static int lagg_fail_detach(struct lagg_softc *); 132 static int lagg_fail_start(struct lagg_softc *, struct mbuf *); 133 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 134 struct mbuf *); 135 136 /* Loadbalancing */ 137 static int lagg_lb_attach(struct lagg_softc *); 138 static int lagg_lb_detach(struct lagg_softc *); 139 static int lagg_lb_port_create(struct lagg_port *); 140 static void lagg_lb_port_destroy(struct lagg_port *); 141 static int lagg_lb_start(struct lagg_softc *, struct mbuf *); 142 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 143 struct mbuf *); 144 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 145 146 /* 802.3ad LACP */ 147 static int lagg_lacp_attach(struct lagg_softc *); 148 static int lagg_lacp_detach(struct lagg_softc *); 149 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); 150 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 151 struct mbuf *); 152 static void lagg_lacp_lladdr(struct lagg_softc *); 153 154 /* lagg protocol table */ 155 static const struct { 156 int ti_proto; 157 int (*ti_attach)(struct lagg_softc *); 158 } lagg_protos[] = { 159 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach }, 160 { LAGG_PROTO_FAILOVER, lagg_fail_attach }, 161 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach }, 162 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach }, 163 { LAGG_PROTO_LACP, lagg_lacp_attach }, 164 { LAGG_PROTO_NONE, NULL } 165 }; 166 167 SYSCTL_DECL(_net_link); 168 static SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 169 "Link Aggregation"); 170 171 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */ 172 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW, 173 &lagg_failover_rx_all, 0, 174 "Accept input from any interface in a failover lagg"); 175 static int def_use_flowid = 1; /* Default value for using M_FLOWID */ 176 TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid); 177 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RW, 178 &def_use_flowid, 0, 179 "Default setting for using flow id for load sharing"); 180 181 static int 182 lagg_modevent(module_t mod, int type, void *data) 183 { 184 185 switch (type) { 186 case MOD_LOAD: 187 mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF); 188 SLIST_INIT(&lagg_list); 189 if_clone_attach(&lagg_cloner); 190 lagg_input_p = lagg_input; 191 lagg_linkstate_p = lagg_port_state; 192 lagg_detach_cookie = EVENTHANDLER_REGISTER( 193 ifnet_departure_event, lagg_port_ifdetach, NULL, 194 EVENTHANDLER_PRI_ANY); 195 break; 196 case MOD_UNLOAD: 197 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 198 lagg_detach_cookie); 199 if_clone_detach(&lagg_cloner); 200 lagg_input_p = NULL; 201 lagg_linkstate_p = NULL; 202 mtx_destroy(&lagg_list_mtx); 203 break; 204 default: 205 return (EOPNOTSUPP); 206 } 207 return (0); 208 } 209 210 static moduledata_t lagg_mod = { 211 "if_lagg", 212 lagg_modevent, 213 0 214 }; 215 216 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 217 MODULE_VERSION(if_lagg, 1); 218 219 #if __FreeBSD_version >= 800000 220 /* 221 * This routine is run via an vlan 222 * config EVENT 223 */ 224 static void 225 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 226 { 227 struct lagg_softc *sc = ifp->if_softc; 228 struct lagg_port *lp; 229 230 if (ifp->if_softc != arg) /* Not our event */ 231 return; 232 233 LAGG_RLOCK(sc); 234 if (!SLIST_EMPTY(&sc->sc_ports)) { 235 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 236 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 237 } 238 LAGG_RUNLOCK(sc); 239 } 240 241 /* 242 * This routine is run via an vlan 243 * unconfig EVENT 244 */ 245 static void 246 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 247 { 248 struct lagg_softc *sc = ifp->if_softc; 249 struct lagg_port *lp; 250 251 if (ifp->if_softc != arg) /* Not our event */ 252 return; 253 254 LAGG_RLOCK(sc); 255 if (!SLIST_EMPTY(&sc->sc_ports)) { 256 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 257 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 258 } 259 LAGG_RUNLOCK(sc); 260 } 261 #endif 262 263 static int 264 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) 265 { 266 struct lagg_softc *sc; 267 struct ifnet *ifp; 268 int i, error = 0; 269 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ 270 struct sysctl_oid *oid; 271 char num[14]; /* sufficient for 32 bits */ 272 273 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO); 274 ifp = sc->sc_ifp = if_alloc(IFT_ETHER); 275 if (ifp == NULL) { 276 free(sc, M_DEVBUF); 277 return (ENOSPC); 278 } 279 280 sysctl_ctx_init(&sc->ctx); 281 snprintf(num, sizeof(num), "%u", unit); 282 sc->use_flowid = def_use_flowid; 283 oid = SYSCTL_ADD_NODE(&sc->ctx, &SYSCTL_NODE_CHILDREN(_net_link, lagg), 284 OID_AUTO, num, CTLFLAG_RD, NULL, ""); 285 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 286 "use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid, sc->use_flowid, 287 "Use flow id for load sharing"); 288 /* Hash all layers by default */ 289 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4; 290 291 sc->sc_proto = LAGG_PROTO_NONE; 292 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) { 293 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) { 294 sc->sc_proto = lagg_protos[i].ti_proto; 295 if ((error = lagg_protos[i].ti_attach(sc)) != 0) { 296 if_free(ifp); 297 free(sc, M_DEVBUF); 298 return (error); 299 } 300 break; 301 } 302 } 303 LAGG_LOCK_INIT(sc); 304 SLIST_INIT(&sc->sc_ports); 305 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc); 306 307 /* Initialise pseudo media types */ 308 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 309 lagg_media_status); 310 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 311 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 312 313 if_initname(ifp, ifc->ifc_name, unit); 314 ifp->if_softc = sc; 315 ifp->if_start = lagg_start; 316 ifp->if_init = lagg_init; 317 ifp->if_ioctl = lagg_ioctl; 318 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 319 320 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 321 ifp->if_snd.ifq_drv_maxlen = ifqmaxlen; 322 IFQ_SET_READY(&ifp->if_snd); 323 324 /* 325 * Attach as an ordinary ethernet device, children will be attached 326 * as special device IFT_IEEE8023ADLAG. 327 */ 328 ether_ifattach(ifp, eaddr); 329 330 #if __FreeBSD_version >= 800000 331 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 332 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 333 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 334 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 335 #endif 336 337 /* Insert into the global list of laggs */ 338 mtx_lock(&lagg_list_mtx); 339 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries); 340 mtx_unlock(&lagg_list_mtx); 341 342 return (0); 343 } 344 345 static void 346 lagg_clone_destroy(struct ifnet *ifp) 347 { 348 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 349 struct lagg_port *lp; 350 351 LAGG_WLOCK(sc); 352 353 lagg_stop(sc); 354 ifp->if_flags &= ~IFF_UP; 355 356 #if __FreeBSD_version >= 800000 357 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 358 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 359 #endif 360 361 /* Shutdown and remove lagg ports */ 362 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) 363 lagg_port_destroy(lp, 1); 364 /* Unhook the aggregation protocol */ 365 if (sc->sc_detach != NULL) 366 (*sc->sc_detach)(sc); 367 368 LAGG_WUNLOCK(sc); 369 370 sysctl_ctx_free(&sc->ctx); 371 ifmedia_removeall(&sc->sc_media); 372 ether_ifdetach(ifp); 373 if_free(ifp); 374 375 mtx_lock(&lagg_list_mtx); 376 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries); 377 mtx_unlock(&lagg_list_mtx); 378 379 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task); 380 LAGG_LOCK_DESTROY(sc); 381 free(sc, M_DEVBUF); 382 } 383 384 static void 385 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr) 386 { 387 struct ifnet *ifp = sc->sc_ifp; 388 389 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 390 return; 391 392 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN); 393 /* Let the protocol know the MAC has changed */ 394 if (sc->sc_lladdr != NULL) 395 (*sc->sc_lladdr)(sc); 396 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 397 } 398 399 static void 400 lagg_capabilities(struct lagg_softc *sc) 401 { 402 struct lagg_port *lp; 403 int cap = ~0, ena = ~0; 404 u_long hwa = ~0UL; 405 406 LAGG_WLOCK_ASSERT(sc); 407 408 /* Get capabilities from the lagg ports */ 409 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 410 cap &= lp->lp_ifp->if_capabilities; 411 ena &= lp->lp_ifp->if_capenable; 412 hwa &= lp->lp_ifp->if_hwassist; 413 } 414 cap = (cap == ~0 ? 0 : cap); 415 ena = (ena == ~0 ? 0 : ena); 416 hwa = (hwa == ~0 ? 0 : hwa); 417 418 if (sc->sc_ifp->if_capabilities != cap || 419 sc->sc_ifp->if_capenable != ena || 420 sc->sc_ifp->if_hwassist != hwa) { 421 sc->sc_ifp->if_capabilities = cap; 422 sc->sc_ifp->if_capenable = ena; 423 sc->sc_ifp->if_hwassist = hwa; 424 getmicrotime(&sc->sc_ifp->if_lastchange); 425 426 if (sc->sc_ifflags & IFF_DEBUG) 427 if_printf(sc->sc_ifp, 428 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 429 } 430 } 431 432 static void 433 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr) 434 { 435 struct lagg_softc *sc = lp->lp_softc; 436 struct ifnet *ifp = lp->lp_ifp; 437 struct lagg_llq *llq; 438 int pending = 0; 439 440 LAGG_WLOCK_ASSERT(sc); 441 442 if (lp->lp_detaching || 443 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 444 return; 445 446 /* Check to make sure its not already queued to be changed */ 447 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 448 if (llq->llq_ifp == ifp) { 449 pending = 1; 450 break; 451 } 452 } 453 454 if (!pending) { 455 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT); 456 if (llq == NULL) /* XXX what to do */ 457 return; 458 } 459 460 /* Update the lladdr even if pending, it may have changed */ 461 llq->llq_ifp = ifp; 462 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN); 463 464 if (!pending) 465 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries); 466 467 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task); 468 } 469 470 /* 471 * Set the interface MAC address from a taskqueue to avoid a LOR. 472 */ 473 static void 474 lagg_port_setlladdr(void *arg, int pending) 475 { 476 struct lagg_softc *sc = (struct lagg_softc *)arg; 477 struct lagg_llq *llq, *head; 478 struct ifnet *ifp; 479 int error; 480 481 /* Grab a local reference of the queue and remove it from the softc */ 482 LAGG_WLOCK(sc); 483 head = SLIST_FIRST(&sc->sc_llq_head); 484 SLIST_FIRST(&sc->sc_llq_head) = NULL; 485 LAGG_WUNLOCK(sc); 486 487 /* 488 * Traverse the queue and set the lladdr on each ifp. It is safe to do 489 * unlocked as we have the only reference to it. 490 */ 491 for (llq = head; llq != NULL; llq = head) { 492 ifp = llq->llq_ifp; 493 494 /* Set the link layer address */ 495 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN); 496 if (error) 497 printf("%s: setlladdr failed on %s\n", __func__, 498 ifp->if_xname); 499 500 head = SLIST_NEXT(llq, llq_entries); 501 free(llq, M_DEVBUF); 502 } 503 } 504 505 static int 506 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 507 { 508 struct lagg_softc *sc_ptr; 509 struct lagg_port *lp; 510 int error = 0; 511 512 LAGG_WLOCK_ASSERT(sc); 513 514 /* Limit the maximal number of lagg ports */ 515 if (sc->sc_count >= LAGG_MAX_PORTS) 516 return (ENOSPC); 517 518 /* Check if port has already been associated to a lagg */ 519 if (ifp->if_lagg != NULL) 520 return (EBUSY); 521 522 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 523 if (ifp->if_type != IFT_ETHER) 524 return (EPROTONOSUPPORT); 525 526 /* Allow the first Ethernet member to define the MTU */ 527 if (SLIST_EMPTY(&sc->sc_ports)) 528 sc->sc_ifp->if_mtu = ifp->if_mtu; 529 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 530 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 531 ifp->if_xname); 532 return (EINVAL); 533 } 534 535 if ((lp = malloc(sizeof(struct lagg_port), 536 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 537 return (ENOMEM); 538 539 /* Check if port is a stacked lagg */ 540 mtx_lock(&lagg_list_mtx); 541 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) { 542 if (ifp == sc_ptr->sc_ifp) { 543 mtx_unlock(&lagg_list_mtx); 544 free(lp, M_DEVBUF); 545 return (EINVAL); 546 /* XXX disable stacking for the moment, its untested */ 547 #ifdef LAGG_PORT_STACKING 548 lp->lp_flags |= LAGG_PORT_STACK; 549 if (lagg_port_checkstacking(sc_ptr) >= 550 LAGG_MAX_STACKING) { 551 mtx_unlock(&lagg_list_mtx); 552 free(lp, M_DEVBUF); 553 return (E2BIG); 554 } 555 #endif 556 } 557 } 558 mtx_unlock(&lagg_list_mtx); 559 560 /* Change the interface type */ 561 lp->lp_iftype = ifp->if_type; 562 ifp->if_type = IFT_IEEE8023ADLAG; 563 ifp->if_lagg = lp; 564 lp->lp_ioctl = ifp->if_ioctl; 565 ifp->if_ioctl = lagg_port_ioctl; 566 lp->lp_output = ifp->if_output; 567 ifp->if_output = lagg_port_output; 568 569 lp->lp_ifp = ifp; 570 lp->lp_softc = sc; 571 572 /* Save port link layer address */ 573 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 574 575 if (SLIST_EMPTY(&sc->sc_ports)) { 576 sc->sc_primary = lp; 577 lagg_lladdr(sc, IF_LLADDR(ifp)); 578 } else { 579 /* Update link layer address for this port */ 580 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp)); 581 } 582 583 /* Insert into the list of ports */ 584 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 585 sc->sc_count++; 586 587 /* Update lagg capabilities */ 588 lagg_capabilities(sc); 589 lagg_linkstate(sc); 590 591 /* Add multicast addresses and interface flags to this port */ 592 lagg_ether_cmdmulti(lp, 1); 593 lagg_setflags(lp, 1); 594 595 if (sc->sc_port_create != NULL) 596 error = (*sc->sc_port_create)(lp); 597 if (error) { 598 /* remove the port again, without calling sc_port_destroy */ 599 lagg_port_destroy(lp, 0); 600 return (error); 601 } 602 603 return (error); 604 } 605 606 #ifdef LAGG_PORT_STACKING 607 static int 608 lagg_port_checkstacking(struct lagg_softc *sc) 609 { 610 struct lagg_softc *sc_ptr; 611 struct lagg_port *lp; 612 int m = 0; 613 614 LAGG_WLOCK_ASSERT(sc); 615 616 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 617 if (lp->lp_flags & LAGG_PORT_STACK) { 618 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 619 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 620 } 621 } 622 623 return (m + 1); 624 } 625 #endif 626 627 static int 628 lagg_port_destroy(struct lagg_port *lp, int runpd) 629 { 630 struct lagg_softc *sc = lp->lp_softc; 631 struct lagg_port *lp_ptr; 632 struct lagg_llq *llq; 633 struct ifnet *ifp = lp->lp_ifp; 634 635 LAGG_WLOCK_ASSERT(sc); 636 637 if (runpd && sc->sc_port_destroy != NULL) 638 (*sc->sc_port_destroy)(lp); 639 640 /* 641 * Remove multicast addresses and interface flags from this port and 642 * reset the MAC address, skip if the interface is being detached. 643 */ 644 if (!lp->lp_detaching) { 645 lagg_ether_cmdmulti(lp, 0); 646 lagg_setflags(lp, 0); 647 lagg_port_lladdr(lp, lp->lp_lladdr); 648 } 649 650 /* Restore interface */ 651 ifp->if_type = lp->lp_iftype; 652 ifp->if_ioctl = lp->lp_ioctl; 653 ifp->if_output = lp->lp_output; 654 ifp->if_lagg = NULL; 655 656 /* Finally, remove the port from the lagg */ 657 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 658 sc->sc_count--; 659 660 /* Update the primary interface */ 661 if (lp == sc->sc_primary) { 662 uint8_t lladdr[ETHER_ADDR_LEN]; 663 664 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) { 665 bzero(&lladdr, ETHER_ADDR_LEN); 666 } else { 667 bcopy(lp_ptr->lp_lladdr, 668 lladdr, ETHER_ADDR_LEN); 669 } 670 lagg_lladdr(sc, lladdr); 671 sc->sc_primary = lp_ptr; 672 673 /* Update link layer address for each port */ 674 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 675 lagg_port_lladdr(lp_ptr, lladdr); 676 } 677 678 /* Remove any pending lladdr changes from the queue */ 679 if (lp->lp_detaching) { 680 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 681 if (llq->llq_ifp == ifp) { 682 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq, 683 llq_entries); 684 free(llq, M_DEVBUF); 685 break; /* Only appears once */ 686 } 687 } 688 } 689 690 if (lp->lp_ifflags) 691 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 692 693 free(lp, M_DEVBUF); 694 695 /* Update lagg capabilities */ 696 lagg_capabilities(sc); 697 lagg_linkstate(sc); 698 699 return (0); 700 } 701 702 static int 703 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 704 { 705 struct lagg_reqport *rp = (struct lagg_reqport *)data; 706 struct lagg_softc *sc; 707 struct lagg_port *lp = NULL; 708 int error = 0; 709 710 /* Should be checked by the caller */ 711 if (ifp->if_type != IFT_IEEE8023ADLAG || 712 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 713 goto fallback; 714 715 switch (cmd) { 716 case SIOCGLAGGPORT: 717 if (rp->rp_portname[0] == '\0' || 718 ifunit(rp->rp_portname) != ifp) { 719 error = EINVAL; 720 break; 721 } 722 723 LAGG_RLOCK(sc); 724 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 725 error = ENOENT; 726 LAGG_RUNLOCK(sc); 727 break; 728 } 729 730 lagg_port2req(lp, rp); 731 LAGG_RUNLOCK(sc); 732 break; 733 734 case SIOCSIFCAP: 735 if (lp->lp_ioctl == NULL) { 736 error = EINVAL; 737 break; 738 } 739 error = (*lp->lp_ioctl)(ifp, cmd, data); 740 if (error) 741 break; 742 743 /* Update lagg interface capabilities */ 744 LAGG_WLOCK(sc); 745 lagg_capabilities(sc); 746 LAGG_WUNLOCK(sc); 747 break; 748 749 case SIOCSIFMTU: 750 /* Do not allow the MTU to be changed once joined */ 751 error = EINVAL; 752 break; 753 754 default: 755 goto fallback; 756 } 757 758 return (error); 759 760 fallback: 761 if (lp->lp_ioctl != NULL) 762 return ((*lp->lp_ioctl)(ifp, cmd, data)); 763 764 return (EINVAL); 765 } 766 767 static int 768 lagg_port_output(struct ifnet *ifp, struct mbuf *m, 769 struct sockaddr *dst, struct route *ro) 770 { 771 struct lagg_port *lp = ifp->if_lagg; 772 struct ether_header *eh; 773 short type = 0; 774 775 switch (dst->sa_family) { 776 case pseudo_AF_HDRCMPLT: 777 case AF_UNSPEC: 778 eh = (struct ether_header *)dst->sa_data; 779 type = eh->ether_type; 780 break; 781 } 782 783 /* 784 * Only allow ethernet types required to initiate or maintain the link, 785 * aggregated frames take a different path. 786 */ 787 switch (ntohs(type)) { 788 case ETHERTYPE_PAE: /* EAPOL PAE/802.1x */ 789 return ((*lp->lp_output)(ifp, m, dst, ro)); 790 } 791 792 /* drop any other frames */ 793 m_freem(m); 794 return (EBUSY); 795 } 796 797 static void 798 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 799 { 800 struct lagg_port *lp; 801 struct lagg_softc *sc; 802 803 if ((lp = ifp->if_lagg) == NULL) 804 return; 805 806 sc = lp->lp_softc; 807 808 LAGG_WLOCK(sc); 809 lp->lp_detaching = 1; 810 lagg_port_destroy(lp, 1); 811 LAGG_WUNLOCK(sc); 812 } 813 814 static void 815 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 816 { 817 struct lagg_softc *sc = lp->lp_softc; 818 819 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 820 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 821 rp->rp_prio = lp->lp_prio; 822 rp->rp_flags = lp->lp_flags; 823 if (sc->sc_portreq != NULL) 824 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc); 825 826 /* Add protocol specific flags */ 827 switch (sc->sc_proto) { 828 case LAGG_PROTO_FAILOVER: 829 if (lp == sc->sc_primary) 830 rp->rp_flags |= LAGG_PORT_MASTER; 831 if (lp == lagg_link_active(sc, sc->sc_primary)) 832 rp->rp_flags |= LAGG_PORT_ACTIVE; 833 break; 834 835 case LAGG_PROTO_ROUNDROBIN: 836 case LAGG_PROTO_LOADBALANCE: 837 case LAGG_PROTO_ETHERCHANNEL: 838 if (LAGG_PORTACTIVE(lp)) 839 rp->rp_flags |= LAGG_PORT_ACTIVE; 840 break; 841 842 case LAGG_PROTO_LACP: 843 /* LACP has a different definition of active */ 844 if (lacp_isactive(lp)) 845 rp->rp_flags |= LAGG_PORT_ACTIVE; 846 if (lacp_iscollecting(lp)) 847 rp->rp_flags |= LAGG_PORT_COLLECTING; 848 if (lacp_isdistributing(lp)) 849 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 850 break; 851 } 852 853 } 854 855 static void 856 lagg_init(void *xsc) 857 { 858 struct lagg_softc *sc = (struct lagg_softc *)xsc; 859 struct lagg_port *lp; 860 struct ifnet *ifp = sc->sc_ifp; 861 862 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 863 return; 864 865 LAGG_WLOCK(sc); 866 867 ifp->if_drv_flags |= IFF_DRV_RUNNING; 868 /* Update the port lladdrs */ 869 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 870 lagg_port_lladdr(lp, IF_LLADDR(ifp)); 871 872 if (sc->sc_init != NULL) 873 (*sc->sc_init)(sc); 874 875 LAGG_WUNLOCK(sc); 876 } 877 878 static void 879 lagg_stop(struct lagg_softc *sc) 880 { 881 struct ifnet *ifp = sc->sc_ifp; 882 883 LAGG_WLOCK_ASSERT(sc); 884 885 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 886 return; 887 888 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 889 890 if (sc->sc_stop != NULL) 891 (*sc->sc_stop)(sc); 892 } 893 894 static int 895 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 896 { 897 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 898 struct lagg_reqall *ra = (struct lagg_reqall *)data; 899 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 900 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 901 struct ifreq *ifr = (struct ifreq *)data; 902 struct lagg_port *lp; 903 struct ifnet *tpif; 904 struct thread *td = curthread; 905 char *buf, *outbuf; 906 int count, buflen, len, error = 0; 907 908 bzero(&rpbuf, sizeof(rpbuf)); 909 910 switch (cmd) { 911 case SIOCGLAGG: 912 LAGG_RLOCK(sc); 913 count = 0; 914 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 915 count++; 916 buflen = count * sizeof(struct lagg_reqport); 917 LAGG_RUNLOCK(sc); 918 919 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 920 921 LAGG_RLOCK(sc); 922 ra->ra_proto = sc->sc_proto; 923 if (sc->sc_req != NULL) 924 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc); 925 926 count = 0; 927 buf = outbuf; 928 len = min(ra->ra_size, buflen); 929 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 930 if (len < sizeof(rpbuf)) 931 break; 932 933 lagg_port2req(lp, &rpbuf); 934 memcpy(buf, &rpbuf, sizeof(rpbuf)); 935 count++; 936 buf += sizeof(rpbuf); 937 len -= sizeof(rpbuf); 938 } 939 LAGG_RUNLOCK(sc); 940 ra->ra_ports = count; 941 ra->ra_size = count * sizeof(rpbuf); 942 error = copyout(outbuf, ra->ra_port, ra->ra_size); 943 free(outbuf, M_TEMP); 944 break; 945 case SIOCSLAGG: 946 error = priv_check(td, PRIV_NET_LAGG); 947 if (error) 948 break; 949 if (ra->ra_proto >= LAGG_PROTO_MAX) { 950 error = EPROTONOSUPPORT; 951 break; 952 } 953 if (sc->sc_proto != LAGG_PROTO_NONE) { 954 LAGG_WLOCK(sc); 955 error = sc->sc_detach(sc); 956 /* Reset protocol and pointers */ 957 sc->sc_proto = LAGG_PROTO_NONE; 958 sc->sc_detach = NULL; 959 sc->sc_start = NULL; 960 sc->sc_input = NULL; 961 sc->sc_port_create = NULL; 962 sc->sc_port_destroy = NULL; 963 sc->sc_linkstate = NULL; 964 sc->sc_init = NULL; 965 sc->sc_stop = NULL; 966 sc->sc_lladdr = NULL; 967 sc->sc_req = NULL; 968 sc->sc_portreq = NULL; 969 LAGG_WUNLOCK(sc); 970 } 971 if (error != 0) 972 break; 973 for (int i = 0; i < (sizeof(lagg_protos) / 974 sizeof(lagg_protos[0])); i++) { 975 if (lagg_protos[i].ti_proto == ra->ra_proto) { 976 if (sc->sc_ifflags & IFF_DEBUG) 977 printf("%s: using proto %u\n", 978 sc->sc_ifname, 979 lagg_protos[i].ti_proto); 980 LAGG_WLOCK(sc); 981 sc->sc_proto = lagg_protos[i].ti_proto; 982 if (sc->sc_proto != LAGG_PROTO_NONE) 983 error = lagg_protos[i].ti_attach(sc); 984 LAGG_WUNLOCK(sc); 985 return (error); 986 } 987 } 988 error = EPROTONOSUPPORT; 989 break; 990 case SIOCGLAGGFLAGS: 991 rf->rf_flags = sc->sc_flags; 992 break; 993 case SIOCSLAGGHASH: 994 error = priv_check(td, PRIV_NET_LAGG); 995 if (error) 996 break; 997 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 998 error = EINVAL; 999 break; 1000 } 1001 LAGG_WLOCK(sc); 1002 sc->sc_flags &= ~LAGG_F_HASHMASK; 1003 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK; 1004 LAGG_WUNLOCK(sc); 1005 break; 1006 case SIOCGLAGGPORT: 1007 if (rp->rp_portname[0] == '\0' || 1008 (tpif = ifunit(rp->rp_portname)) == NULL) { 1009 error = EINVAL; 1010 break; 1011 } 1012 1013 LAGG_RLOCK(sc); 1014 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1015 lp->lp_softc != sc) { 1016 error = ENOENT; 1017 LAGG_RUNLOCK(sc); 1018 break; 1019 } 1020 1021 lagg_port2req(lp, rp); 1022 LAGG_RUNLOCK(sc); 1023 break; 1024 case SIOCSLAGGPORT: 1025 error = priv_check(td, PRIV_NET_LAGG); 1026 if (error) 1027 break; 1028 if (rp->rp_portname[0] == '\0' || 1029 (tpif = ifunit(rp->rp_portname)) == NULL) { 1030 error = EINVAL; 1031 break; 1032 } 1033 LAGG_WLOCK(sc); 1034 error = lagg_port_create(sc, tpif); 1035 LAGG_WUNLOCK(sc); 1036 break; 1037 case SIOCSLAGGDELPORT: 1038 error = priv_check(td, PRIV_NET_LAGG); 1039 if (error) 1040 break; 1041 if (rp->rp_portname[0] == '\0' || 1042 (tpif = ifunit(rp->rp_portname)) == NULL) { 1043 error = EINVAL; 1044 break; 1045 } 1046 1047 LAGG_WLOCK(sc); 1048 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1049 lp->lp_softc != sc) { 1050 error = ENOENT; 1051 LAGG_WUNLOCK(sc); 1052 break; 1053 } 1054 1055 error = lagg_port_destroy(lp, 1); 1056 LAGG_WUNLOCK(sc); 1057 break; 1058 case SIOCSIFFLAGS: 1059 /* Set flags on ports too */ 1060 LAGG_WLOCK(sc); 1061 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1062 lagg_setflags(lp, 1); 1063 } 1064 LAGG_WUNLOCK(sc); 1065 1066 if (!(ifp->if_flags & IFF_UP) && 1067 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1068 /* 1069 * If interface is marked down and it is running, 1070 * then stop and disable it. 1071 */ 1072 LAGG_WLOCK(sc); 1073 lagg_stop(sc); 1074 LAGG_WUNLOCK(sc); 1075 } else if ((ifp->if_flags & IFF_UP) && 1076 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1077 /* 1078 * If interface is marked up and it is stopped, then 1079 * start it. 1080 */ 1081 (*ifp->if_init)(sc); 1082 } 1083 break; 1084 case SIOCADDMULTI: 1085 case SIOCDELMULTI: 1086 LAGG_WLOCK(sc); 1087 error = lagg_ether_setmulti(sc); 1088 LAGG_WUNLOCK(sc); 1089 break; 1090 case SIOCSIFMEDIA: 1091 case SIOCGIFMEDIA: 1092 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1093 break; 1094 1095 case SIOCSIFCAP: 1096 case SIOCSIFMTU: 1097 /* Do not allow the MTU or caps to be directly changed */ 1098 error = EINVAL; 1099 break; 1100 1101 default: 1102 error = ether_ioctl(ifp, cmd, data); 1103 break; 1104 } 1105 return (error); 1106 } 1107 1108 static int 1109 lagg_ether_setmulti(struct lagg_softc *sc) 1110 { 1111 struct lagg_port *lp; 1112 1113 LAGG_WLOCK_ASSERT(sc); 1114 1115 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1116 /* First, remove any existing filter entries. */ 1117 lagg_ether_cmdmulti(lp, 0); 1118 /* copy all addresses from the lagg interface to the port */ 1119 lagg_ether_cmdmulti(lp, 1); 1120 } 1121 return (0); 1122 } 1123 1124 static int 1125 lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1126 { 1127 struct lagg_softc *sc = lp->lp_softc; 1128 struct ifnet *ifp = lp->lp_ifp; 1129 struct ifnet *scifp = sc->sc_ifp; 1130 struct lagg_mc *mc; 1131 struct ifmultiaddr *ifma, *rifma = NULL; 1132 struct sockaddr_dl sdl; 1133 int error; 1134 1135 LAGG_WLOCK_ASSERT(sc); 1136 1137 bzero((char *)&sdl, sizeof(sdl)); 1138 sdl.sdl_len = sizeof(sdl); 1139 sdl.sdl_family = AF_LINK; 1140 sdl.sdl_type = IFT_ETHER; 1141 sdl.sdl_alen = ETHER_ADDR_LEN; 1142 sdl.sdl_index = ifp->if_index; 1143 1144 if (set) { 1145 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1146 if (ifma->ifma_addr->sa_family != AF_LINK) 1147 continue; 1148 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 1149 LLADDR(&sdl), ETHER_ADDR_LEN); 1150 1151 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 1152 if (error) 1153 return (error); 1154 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1155 if (mc == NULL) 1156 return (ENOMEM); 1157 mc->mc_ifma = rifma; 1158 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1159 } 1160 } else { 1161 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1162 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1163 if_delmulti_ifma(mc->mc_ifma); 1164 free(mc, M_DEVBUF); 1165 } 1166 } 1167 return (0); 1168 } 1169 1170 /* Handle a ref counted flag that should be set on the lagg port as well */ 1171 static int 1172 lagg_setflag(struct lagg_port *lp, int flag, int status, 1173 int (*func)(struct ifnet *, int)) 1174 { 1175 struct lagg_softc *sc = lp->lp_softc; 1176 struct ifnet *scifp = sc->sc_ifp; 1177 struct ifnet *ifp = lp->lp_ifp; 1178 int error; 1179 1180 LAGG_WLOCK_ASSERT(sc); 1181 1182 status = status ? (scifp->if_flags & flag) : 0; 1183 /* Now "status" contains the flag value or 0 */ 1184 1185 /* 1186 * See if recorded ports status is different from what 1187 * we want it to be. If it is, flip it. We record ports 1188 * status in lp_ifflags so that we won't clear ports flag 1189 * we haven't set. In fact, we don't clear or set ports 1190 * flags directly, but get or release references to them. 1191 * That's why we can be sure that recorded flags still are 1192 * in accord with actual ports flags. 1193 */ 1194 if (status != (lp->lp_ifflags & flag)) { 1195 error = (*func)(ifp, status); 1196 if (error) 1197 return (error); 1198 lp->lp_ifflags &= ~flag; 1199 lp->lp_ifflags |= status; 1200 } 1201 return (0); 1202 } 1203 1204 /* 1205 * Handle IFF_* flags that require certain changes on the lagg port 1206 * if "status" is true, update ports flags respective to the lagg 1207 * if "status" is false, forcedly clear the flags set on port. 1208 */ 1209 static int 1210 lagg_setflags(struct lagg_port *lp, int status) 1211 { 1212 int error, i; 1213 1214 for (i = 0; lagg_pflags[i].flag; i++) { 1215 error = lagg_setflag(lp, lagg_pflags[i].flag, 1216 status, lagg_pflags[i].func); 1217 if (error) 1218 return (error); 1219 } 1220 return (0); 1221 } 1222 1223 static void 1224 lagg_start(struct ifnet *ifp) 1225 { 1226 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1227 struct mbuf *m; 1228 int error = 0; 1229 1230 LAGG_RLOCK(sc); 1231 /* We need a Tx algorithm and at least one port */ 1232 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1233 IF_DRAIN(&ifp->if_snd); 1234 LAGG_RUNLOCK(sc); 1235 return; 1236 } 1237 1238 for (;; error = 0) { 1239 IFQ_DEQUEUE(&ifp->if_snd, m); 1240 if (m == NULL) 1241 break; 1242 1243 ETHER_BPF_MTAP(ifp, m); 1244 1245 error = (*sc->sc_start)(sc, m); 1246 if (error == 0) 1247 ifp->if_opackets++; 1248 else 1249 ifp->if_oerrors++; 1250 } 1251 LAGG_RUNLOCK(sc); 1252 } 1253 1254 static struct mbuf * 1255 lagg_input(struct ifnet *ifp, struct mbuf *m) 1256 { 1257 struct lagg_port *lp = ifp->if_lagg; 1258 struct lagg_softc *sc = lp->lp_softc; 1259 struct ifnet *scifp = sc->sc_ifp; 1260 1261 LAGG_RLOCK(sc); 1262 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1263 (lp->lp_flags & LAGG_PORT_DISABLED) || 1264 sc->sc_proto == LAGG_PROTO_NONE) { 1265 LAGG_RUNLOCK(sc); 1266 m_freem(m); 1267 return (NULL); 1268 } 1269 1270 ETHER_BPF_MTAP(scifp, m); 1271 1272 m = (*sc->sc_input)(sc, lp, m); 1273 1274 if (m != NULL) { 1275 scifp->if_ipackets++; 1276 scifp->if_ibytes += m->m_pkthdr.len; 1277 1278 if (scifp->if_flags & IFF_MONITOR) { 1279 m_freem(m); 1280 m = NULL; 1281 } 1282 } 1283 1284 LAGG_RUNLOCK(sc); 1285 return (m); 1286 } 1287 1288 static int 1289 lagg_media_change(struct ifnet *ifp) 1290 { 1291 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1292 1293 if (sc->sc_ifflags & IFF_DEBUG) 1294 printf("%s\n", __func__); 1295 1296 /* Ignore */ 1297 return (0); 1298 } 1299 1300 static void 1301 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1302 { 1303 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1304 struct lagg_port *lp; 1305 1306 imr->ifm_status = IFM_AVALID; 1307 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1308 1309 LAGG_RLOCK(sc); 1310 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1311 if (LAGG_PORTACTIVE(lp)) 1312 imr->ifm_status |= IFM_ACTIVE; 1313 } 1314 LAGG_RUNLOCK(sc); 1315 } 1316 1317 static void 1318 lagg_linkstate(struct lagg_softc *sc) 1319 { 1320 struct lagg_port *lp; 1321 int new_link = LINK_STATE_DOWN; 1322 uint64_t speed; 1323 1324 /* Our link is considered up if at least one of our ports is active */ 1325 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1326 if (lp->lp_link_state == LINK_STATE_UP) { 1327 new_link = LINK_STATE_UP; 1328 break; 1329 } 1330 } 1331 if_link_state_change(sc->sc_ifp, new_link); 1332 1333 /* Update if_baudrate to reflect the max possible speed */ 1334 switch (sc->sc_proto) { 1335 case LAGG_PROTO_FAILOVER: 1336 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1337 sc->sc_primary->lp_ifp->if_baudrate : 0; 1338 break; 1339 case LAGG_PROTO_ROUNDROBIN: 1340 case LAGG_PROTO_LOADBALANCE: 1341 case LAGG_PROTO_ETHERCHANNEL: 1342 speed = 0; 1343 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1344 speed += lp->lp_ifp->if_baudrate; 1345 sc->sc_ifp->if_baudrate = speed; 1346 break; 1347 case LAGG_PROTO_LACP: 1348 /* LACP updates if_baudrate itself */ 1349 break; 1350 } 1351 } 1352 1353 static void 1354 lagg_port_state(struct ifnet *ifp, int state) 1355 { 1356 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1357 struct lagg_softc *sc = NULL; 1358 1359 if (lp != NULL) 1360 sc = lp->lp_softc; 1361 if (sc == NULL) 1362 return; 1363 1364 LAGG_WLOCK(sc); 1365 lagg_linkstate(sc); 1366 if (sc->sc_linkstate != NULL) 1367 (*sc->sc_linkstate)(lp); 1368 LAGG_WUNLOCK(sc); 1369 } 1370 1371 struct lagg_port * 1372 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1373 { 1374 struct lagg_port *lp_next, *rval = NULL; 1375 // int new_link = LINK_STATE_DOWN; 1376 1377 LAGG_RLOCK_ASSERT(sc); 1378 /* 1379 * Search a port which reports an active link state. 1380 */ 1381 1382 if (lp == NULL) 1383 goto search; 1384 if (LAGG_PORTACTIVE(lp)) { 1385 rval = lp; 1386 goto found; 1387 } 1388 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1389 LAGG_PORTACTIVE(lp_next)) { 1390 rval = lp_next; 1391 goto found; 1392 } 1393 1394 search: 1395 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1396 if (LAGG_PORTACTIVE(lp_next)) { 1397 rval = lp_next; 1398 goto found; 1399 } 1400 } 1401 1402 found: 1403 if (rval != NULL) { 1404 /* 1405 * The IEEE 802.1D standard assumes that a lagg with 1406 * multiple ports is always full duplex. This is valid 1407 * for load sharing laggs and if at least two links 1408 * are active. Unfortunately, checking the latter would 1409 * be too expensive at this point. 1410 XXX 1411 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1412 (sc->sc_count > 1)) 1413 new_link = LINK_STATE_FULL_DUPLEX; 1414 else 1415 new_link = rval->lp_link_state; 1416 */ 1417 } 1418 1419 return (rval); 1420 } 1421 1422 static const void * 1423 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf) 1424 { 1425 if (m->m_pkthdr.len < (off + len)) { 1426 return (NULL); 1427 } else if (m->m_len < (off + len)) { 1428 m_copydata(m, off, len, buf); 1429 return (buf); 1430 } 1431 return (mtod(m, char *) + off); 1432 } 1433 1434 uint32_t 1435 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key) 1436 { 1437 uint16_t etype; 1438 uint32_t p = key; 1439 int off; 1440 struct ether_header *eh; 1441 const struct ether_vlan_header *vlan; 1442 #ifdef INET 1443 const struct ip *ip; 1444 const uint32_t *ports; 1445 int iphlen; 1446 #endif 1447 #ifdef INET6 1448 const struct ip6_hdr *ip6; 1449 uint32_t flow; 1450 #endif 1451 union { 1452 #ifdef INET 1453 struct ip ip; 1454 #endif 1455 #ifdef INET6 1456 struct ip6_hdr ip6; 1457 #endif 1458 struct ether_vlan_header vlan; 1459 uint32_t port; 1460 } buf; 1461 1462 1463 off = sizeof(*eh); 1464 if (m->m_len < off) 1465 goto out; 1466 eh = mtod(m, struct ether_header *); 1467 etype = ntohs(eh->ether_type); 1468 if (sc->sc_flags & LAGG_F_HASHL2) { 1469 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p); 1470 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p); 1471 } 1472 1473 /* Special handling for encapsulating VLAN frames */ 1474 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) { 1475 p = hash32_buf(&m->m_pkthdr.ether_vtag, 1476 sizeof(m->m_pkthdr.ether_vtag), p); 1477 } else if (etype == ETHERTYPE_VLAN) { 1478 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf); 1479 if (vlan == NULL) 1480 goto out; 1481 1482 if (sc->sc_flags & LAGG_F_HASHL2) 1483 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p); 1484 etype = ntohs(vlan->evl_proto); 1485 off += sizeof(*vlan) - sizeof(*eh); 1486 } 1487 1488 switch (etype) { 1489 #ifdef INET 1490 case ETHERTYPE_IP: 1491 ip = lagg_gethdr(m, off, sizeof(*ip), &buf); 1492 if (ip == NULL) 1493 goto out; 1494 1495 if (sc->sc_flags & LAGG_F_HASHL3) { 1496 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p); 1497 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p); 1498 } 1499 if (!(sc->sc_flags & LAGG_F_HASHL4)) 1500 break; 1501 switch (ip->ip_p) { 1502 case IPPROTO_TCP: 1503 case IPPROTO_UDP: 1504 case IPPROTO_SCTP: 1505 iphlen = ip->ip_hl << 2; 1506 if (iphlen < sizeof(*ip)) 1507 break; 1508 off += iphlen; 1509 ports = lagg_gethdr(m, off, sizeof(*ports), &buf); 1510 if (ports == NULL) 1511 break; 1512 p = hash32_buf(ports, sizeof(*ports), p); 1513 break; 1514 } 1515 break; 1516 #endif 1517 #ifdef INET6 1518 case ETHERTYPE_IPV6: 1519 if (!(sc->sc_flags & LAGG_F_HASHL3)) 1520 break; 1521 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf); 1522 if (ip6 == NULL) 1523 goto out; 1524 1525 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p); 1526 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p); 1527 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK; 1528 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */ 1529 break; 1530 #endif 1531 } 1532 out: 1533 return (p); 1534 } 1535 1536 int 1537 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1538 { 1539 1540 return (ifp->if_transmit)(ifp, m); 1541 } 1542 1543 /* 1544 * Simple round robin aggregation 1545 */ 1546 1547 static int 1548 lagg_rr_attach(struct lagg_softc *sc) 1549 { 1550 sc->sc_detach = lagg_rr_detach; 1551 sc->sc_start = lagg_rr_start; 1552 sc->sc_input = lagg_rr_input; 1553 sc->sc_port_create = NULL; 1554 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1555 sc->sc_seq = 0; 1556 1557 return (0); 1558 } 1559 1560 static int 1561 lagg_rr_detach(struct lagg_softc *sc) 1562 { 1563 return (0); 1564 } 1565 1566 static int 1567 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 1568 { 1569 struct lagg_port *lp; 1570 uint32_t p; 1571 1572 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1573 p %= sc->sc_count; 1574 lp = SLIST_FIRST(&sc->sc_ports); 1575 while (p--) 1576 lp = SLIST_NEXT(lp, lp_entries); 1577 1578 /* 1579 * Check the port's link state. This will return the next active 1580 * port if the link is down or the port is NULL. 1581 */ 1582 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1583 m_freem(m); 1584 return (ENOENT); 1585 } 1586 1587 /* Send mbuf */ 1588 return (lagg_enqueue(lp->lp_ifp, m)); 1589 } 1590 1591 static struct mbuf * 1592 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1593 { 1594 struct ifnet *ifp = sc->sc_ifp; 1595 1596 /* Just pass in the packet to our lagg device */ 1597 m->m_pkthdr.rcvif = ifp; 1598 1599 return (m); 1600 } 1601 1602 /* 1603 * Active failover 1604 */ 1605 1606 static int 1607 lagg_fail_attach(struct lagg_softc *sc) 1608 { 1609 sc->sc_detach = lagg_fail_detach; 1610 sc->sc_start = lagg_fail_start; 1611 sc->sc_input = lagg_fail_input; 1612 sc->sc_port_create = NULL; 1613 sc->sc_port_destroy = NULL; 1614 1615 return (0); 1616 } 1617 1618 static int 1619 lagg_fail_detach(struct lagg_softc *sc) 1620 { 1621 return (0); 1622 } 1623 1624 static int 1625 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 1626 { 1627 struct lagg_port *lp; 1628 1629 /* Use the master port if active or the next available port */ 1630 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 1631 m_freem(m); 1632 return (ENOENT); 1633 } 1634 1635 /* Send mbuf */ 1636 return (lagg_enqueue(lp->lp_ifp, m)); 1637 } 1638 1639 static struct mbuf * 1640 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1641 { 1642 struct ifnet *ifp = sc->sc_ifp; 1643 struct lagg_port *tmp_tp; 1644 1645 if (lp == sc->sc_primary || lagg_failover_rx_all) { 1646 m->m_pkthdr.rcvif = ifp; 1647 return (m); 1648 } 1649 1650 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1651 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1652 /* 1653 * If tmp_tp is null, we've recieved a packet when all 1654 * our links are down. Weird, but process it anyways. 1655 */ 1656 if ((tmp_tp == NULL || tmp_tp == lp)) { 1657 m->m_pkthdr.rcvif = ifp; 1658 return (m); 1659 } 1660 } 1661 1662 m_freem(m); 1663 return (NULL); 1664 } 1665 1666 /* 1667 * Loadbalancing 1668 */ 1669 1670 static int 1671 lagg_lb_attach(struct lagg_softc *sc) 1672 { 1673 struct lagg_port *lp; 1674 struct lagg_lb *lb; 1675 1676 if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb), 1677 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 1678 return (ENOMEM); 1679 1680 sc->sc_detach = lagg_lb_detach; 1681 sc->sc_start = lagg_lb_start; 1682 sc->sc_input = lagg_lb_input; 1683 sc->sc_port_create = lagg_lb_port_create; 1684 sc->sc_port_destroy = lagg_lb_port_destroy; 1685 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1686 1687 lb->lb_key = arc4random(); 1688 sc->sc_psc = (caddr_t)lb; 1689 1690 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1691 lagg_lb_port_create(lp); 1692 1693 return (0); 1694 } 1695 1696 static int 1697 lagg_lb_detach(struct lagg_softc *sc) 1698 { 1699 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1700 if (lb != NULL) 1701 free(lb, M_DEVBUF); 1702 return (0); 1703 } 1704 1705 static int 1706 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 1707 { 1708 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1709 struct lagg_port *lp_next; 1710 int i = 0; 1711 1712 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 1713 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1714 if (lp_next == lp) 1715 continue; 1716 if (i >= LAGG_MAX_PORTS) 1717 return (EINVAL); 1718 if (sc->sc_ifflags & IFF_DEBUG) 1719 printf("%s: port %s at index %d\n", 1720 sc->sc_ifname, lp_next->lp_ifname, i); 1721 lb->lb_ports[i++] = lp_next; 1722 } 1723 1724 return (0); 1725 } 1726 1727 static int 1728 lagg_lb_port_create(struct lagg_port *lp) 1729 { 1730 struct lagg_softc *sc = lp->lp_softc; 1731 return (lagg_lb_porttable(sc, NULL)); 1732 } 1733 1734 static void 1735 lagg_lb_port_destroy(struct lagg_port *lp) 1736 { 1737 struct lagg_softc *sc = lp->lp_softc; 1738 lagg_lb_porttable(sc, lp); 1739 } 1740 1741 static int 1742 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 1743 { 1744 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1745 struct lagg_port *lp = NULL; 1746 uint32_t p = 0; 1747 1748 if (sc->use_flowid && (m->m_flags & M_FLOWID)) 1749 p = m->m_pkthdr.flowid; 1750 else 1751 p = lagg_hashmbuf(sc, m, lb->lb_key); 1752 p %= sc->sc_count; 1753 lp = lb->lb_ports[p]; 1754 1755 /* 1756 * Check the port's link state. This will return the next active 1757 * port if the link is down or the port is NULL. 1758 */ 1759 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1760 m_freem(m); 1761 return (ENOENT); 1762 } 1763 1764 /* Send mbuf */ 1765 return (lagg_enqueue(lp->lp_ifp, m)); 1766 } 1767 1768 static struct mbuf * 1769 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1770 { 1771 struct ifnet *ifp = sc->sc_ifp; 1772 1773 /* Just pass in the packet to our lagg device */ 1774 m->m_pkthdr.rcvif = ifp; 1775 1776 return (m); 1777 } 1778 1779 /* 1780 * 802.3ad LACP 1781 */ 1782 1783 static int 1784 lagg_lacp_attach(struct lagg_softc *sc) 1785 { 1786 struct lagg_port *lp; 1787 int error; 1788 1789 sc->sc_detach = lagg_lacp_detach; 1790 sc->sc_port_create = lacp_port_create; 1791 sc->sc_port_destroy = lacp_port_destroy; 1792 sc->sc_linkstate = lacp_linkstate; 1793 sc->sc_start = lagg_lacp_start; 1794 sc->sc_input = lagg_lacp_input; 1795 sc->sc_init = lacp_init; 1796 sc->sc_stop = lacp_stop; 1797 sc->sc_lladdr = lagg_lacp_lladdr; 1798 sc->sc_req = lacp_req; 1799 sc->sc_portreq = lacp_portreq; 1800 1801 error = lacp_attach(sc); 1802 if (error) 1803 return (error); 1804 1805 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1806 lacp_port_create(lp); 1807 1808 return (error); 1809 } 1810 1811 static int 1812 lagg_lacp_detach(struct lagg_softc *sc) 1813 { 1814 struct lagg_port *lp; 1815 int error; 1816 1817 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1818 lacp_port_destroy(lp); 1819 1820 /* unlocking is safe here */ 1821 LAGG_WUNLOCK(sc); 1822 error = lacp_detach(sc); 1823 LAGG_WLOCK(sc); 1824 1825 return (error); 1826 } 1827 1828 static void 1829 lagg_lacp_lladdr(struct lagg_softc *sc) 1830 { 1831 struct lagg_port *lp; 1832 1833 /* purge all the lacp ports */ 1834 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1835 lacp_port_destroy(lp); 1836 1837 /* add them back in */ 1838 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1839 lacp_port_create(lp); 1840 } 1841 1842 static int 1843 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 1844 { 1845 struct lagg_port *lp; 1846 1847 lp = lacp_select_tx_port(sc, m); 1848 if (lp == NULL) { 1849 m_freem(m); 1850 return (EBUSY); 1851 } 1852 1853 /* Send mbuf */ 1854 return (lagg_enqueue(lp->lp_ifp, m)); 1855 } 1856 1857 static struct mbuf * 1858 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1859 { 1860 struct ifnet *ifp = sc->sc_ifp; 1861 struct ether_header *eh; 1862 u_short etype; 1863 1864 eh = mtod(m, struct ether_header *); 1865 etype = ntohs(eh->ether_type); 1866 1867 /* Tap off LACP control messages */ 1868 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 1869 m = lacp_input(lp, m); 1870 if (m == NULL) 1871 return (NULL); 1872 } 1873 1874 /* 1875 * If the port is not collecting or not in the active aggregator then 1876 * free and return. 1877 */ 1878 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 1879 m_freem(m); 1880 return (NULL); 1881 } 1882 1883 m->m_pkthdr.rcvif = ifp; 1884 return (m); 1885 } 1886