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 LAGG_WLOCK(sc); 954 if (sc->sc_proto != LAGG_PROTO_NONE) { 955 /* Reset protocol first in case detach unlocks */ 956 sc->sc_proto = LAGG_PROTO_NONE; 957 error = sc->sc_detach(sc); 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 } else if (sc->sc_input != NULL) { 970 /* Still detaching */ 971 error = EBUSY; 972 } 973 if (error != 0) { 974 LAGG_WUNLOCK(sc); 975 break; 976 } 977 for (int i = 0; i < (sizeof(lagg_protos) / 978 sizeof(lagg_protos[0])); i++) { 979 if (lagg_protos[i].ti_proto == ra->ra_proto) { 980 if (sc->sc_ifflags & IFF_DEBUG) 981 printf("%s: using proto %u\n", 982 sc->sc_ifname, 983 lagg_protos[i].ti_proto); 984 sc->sc_proto = lagg_protos[i].ti_proto; 985 if (sc->sc_proto != LAGG_PROTO_NONE) 986 error = lagg_protos[i].ti_attach(sc); 987 LAGG_WUNLOCK(sc); 988 return (error); 989 } 990 } 991 LAGG_WUNLOCK(sc); 992 error = EPROTONOSUPPORT; 993 break; 994 case SIOCGLAGGFLAGS: 995 rf->rf_flags = sc->sc_flags; 996 break; 997 case SIOCSLAGGHASH: 998 error = priv_check(td, PRIV_NET_LAGG); 999 if (error) 1000 break; 1001 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1002 error = EINVAL; 1003 break; 1004 } 1005 LAGG_WLOCK(sc); 1006 sc->sc_flags &= ~LAGG_F_HASHMASK; 1007 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK; 1008 LAGG_WUNLOCK(sc); 1009 break; 1010 case SIOCGLAGGPORT: 1011 if (rp->rp_portname[0] == '\0' || 1012 (tpif = ifunit(rp->rp_portname)) == NULL) { 1013 error = EINVAL; 1014 break; 1015 } 1016 1017 LAGG_RLOCK(sc); 1018 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1019 lp->lp_softc != sc) { 1020 error = ENOENT; 1021 LAGG_RUNLOCK(sc); 1022 break; 1023 } 1024 1025 lagg_port2req(lp, rp); 1026 LAGG_RUNLOCK(sc); 1027 break; 1028 case SIOCSLAGGPORT: 1029 error = priv_check(td, PRIV_NET_LAGG); 1030 if (error) 1031 break; 1032 if (rp->rp_portname[0] == '\0' || 1033 (tpif = ifunit(rp->rp_portname)) == NULL) { 1034 error = EINVAL; 1035 break; 1036 } 1037 LAGG_WLOCK(sc); 1038 error = lagg_port_create(sc, tpif); 1039 LAGG_WUNLOCK(sc); 1040 break; 1041 case SIOCSLAGGDELPORT: 1042 error = priv_check(td, PRIV_NET_LAGG); 1043 if (error) 1044 break; 1045 if (rp->rp_portname[0] == '\0' || 1046 (tpif = ifunit(rp->rp_portname)) == NULL) { 1047 error = EINVAL; 1048 break; 1049 } 1050 1051 LAGG_WLOCK(sc); 1052 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1053 lp->lp_softc != sc) { 1054 error = ENOENT; 1055 LAGG_WUNLOCK(sc); 1056 break; 1057 } 1058 1059 error = lagg_port_destroy(lp, 1); 1060 LAGG_WUNLOCK(sc); 1061 break; 1062 case SIOCSIFFLAGS: 1063 /* Set flags on ports too */ 1064 LAGG_WLOCK(sc); 1065 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1066 lagg_setflags(lp, 1); 1067 } 1068 LAGG_WUNLOCK(sc); 1069 1070 if (!(ifp->if_flags & IFF_UP) && 1071 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1072 /* 1073 * If interface is marked down and it is running, 1074 * then stop and disable it. 1075 */ 1076 LAGG_WLOCK(sc); 1077 lagg_stop(sc); 1078 LAGG_WUNLOCK(sc); 1079 } else if ((ifp->if_flags & IFF_UP) && 1080 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1081 /* 1082 * If interface is marked up and it is stopped, then 1083 * start it. 1084 */ 1085 (*ifp->if_init)(sc); 1086 } 1087 break; 1088 case SIOCADDMULTI: 1089 case SIOCDELMULTI: 1090 LAGG_WLOCK(sc); 1091 error = lagg_ether_setmulti(sc); 1092 LAGG_WUNLOCK(sc); 1093 break; 1094 case SIOCSIFMEDIA: 1095 case SIOCGIFMEDIA: 1096 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1097 break; 1098 1099 case SIOCSIFCAP: 1100 case SIOCSIFMTU: 1101 /* Do not allow the MTU or caps to be directly changed */ 1102 error = EINVAL; 1103 break; 1104 1105 default: 1106 error = ether_ioctl(ifp, cmd, data); 1107 break; 1108 } 1109 return (error); 1110 } 1111 1112 static int 1113 lagg_ether_setmulti(struct lagg_softc *sc) 1114 { 1115 struct lagg_port *lp; 1116 1117 LAGG_WLOCK_ASSERT(sc); 1118 1119 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1120 /* First, remove any existing filter entries. */ 1121 lagg_ether_cmdmulti(lp, 0); 1122 /* copy all addresses from the lagg interface to the port */ 1123 lagg_ether_cmdmulti(lp, 1); 1124 } 1125 return (0); 1126 } 1127 1128 static int 1129 lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1130 { 1131 struct lagg_softc *sc = lp->lp_softc; 1132 struct ifnet *ifp = lp->lp_ifp; 1133 struct ifnet *scifp = sc->sc_ifp; 1134 struct lagg_mc *mc; 1135 struct ifmultiaddr *ifma, *rifma = NULL; 1136 struct sockaddr_dl sdl; 1137 int error; 1138 1139 LAGG_WLOCK_ASSERT(sc); 1140 1141 bzero((char *)&sdl, sizeof(sdl)); 1142 sdl.sdl_len = sizeof(sdl); 1143 sdl.sdl_family = AF_LINK; 1144 sdl.sdl_type = IFT_ETHER; 1145 sdl.sdl_alen = ETHER_ADDR_LEN; 1146 sdl.sdl_index = ifp->if_index; 1147 1148 if (set) { 1149 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1150 if (ifma->ifma_addr->sa_family != AF_LINK) 1151 continue; 1152 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 1153 LLADDR(&sdl), ETHER_ADDR_LEN); 1154 1155 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 1156 if (error) 1157 return (error); 1158 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1159 if (mc == NULL) 1160 return (ENOMEM); 1161 mc->mc_ifma = rifma; 1162 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1163 } 1164 } else { 1165 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1166 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1167 if_delmulti_ifma(mc->mc_ifma); 1168 free(mc, M_DEVBUF); 1169 } 1170 } 1171 return (0); 1172 } 1173 1174 /* Handle a ref counted flag that should be set on the lagg port as well */ 1175 static int 1176 lagg_setflag(struct lagg_port *lp, int flag, int status, 1177 int (*func)(struct ifnet *, int)) 1178 { 1179 struct lagg_softc *sc = lp->lp_softc; 1180 struct ifnet *scifp = sc->sc_ifp; 1181 struct ifnet *ifp = lp->lp_ifp; 1182 int error; 1183 1184 LAGG_WLOCK_ASSERT(sc); 1185 1186 status = status ? (scifp->if_flags & flag) : 0; 1187 /* Now "status" contains the flag value or 0 */ 1188 1189 /* 1190 * See if recorded ports status is different from what 1191 * we want it to be. If it is, flip it. We record ports 1192 * status in lp_ifflags so that we won't clear ports flag 1193 * we haven't set. In fact, we don't clear or set ports 1194 * flags directly, but get or release references to them. 1195 * That's why we can be sure that recorded flags still are 1196 * in accord with actual ports flags. 1197 */ 1198 if (status != (lp->lp_ifflags & flag)) { 1199 error = (*func)(ifp, status); 1200 if (error) 1201 return (error); 1202 lp->lp_ifflags &= ~flag; 1203 lp->lp_ifflags |= status; 1204 } 1205 return (0); 1206 } 1207 1208 /* 1209 * Handle IFF_* flags that require certain changes on the lagg port 1210 * if "status" is true, update ports flags respective to the lagg 1211 * if "status" is false, forcedly clear the flags set on port. 1212 */ 1213 static int 1214 lagg_setflags(struct lagg_port *lp, int status) 1215 { 1216 int error, i; 1217 1218 for (i = 0; lagg_pflags[i].flag; i++) { 1219 error = lagg_setflag(lp, lagg_pflags[i].flag, 1220 status, lagg_pflags[i].func); 1221 if (error) 1222 return (error); 1223 } 1224 return (0); 1225 } 1226 1227 static void 1228 lagg_start(struct ifnet *ifp) 1229 { 1230 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1231 struct mbuf *m; 1232 int error = 0; 1233 1234 LAGG_RLOCK(sc); 1235 /* We need a Tx algorithm and at least one port */ 1236 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1237 IF_DRAIN(&ifp->if_snd); 1238 LAGG_RUNLOCK(sc); 1239 return; 1240 } 1241 1242 for (;; error = 0) { 1243 IFQ_DEQUEUE(&ifp->if_snd, m); 1244 if (m == NULL) 1245 break; 1246 1247 ETHER_BPF_MTAP(ifp, m); 1248 1249 error = (*sc->sc_start)(sc, m); 1250 if (error == 0) 1251 ifp->if_opackets++; 1252 else 1253 ifp->if_oerrors++; 1254 } 1255 LAGG_RUNLOCK(sc); 1256 } 1257 1258 static struct mbuf * 1259 lagg_input(struct ifnet *ifp, struct mbuf *m) 1260 { 1261 struct lagg_port *lp = ifp->if_lagg; 1262 struct lagg_softc *sc = lp->lp_softc; 1263 struct ifnet *scifp = sc->sc_ifp; 1264 1265 LAGG_RLOCK(sc); 1266 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1267 (lp->lp_flags & LAGG_PORT_DISABLED) || 1268 sc->sc_proto == LAGG_PROTO_NONE) { 1269 LAGG_RUNLOCK(sc); 1270 m_freem(m); 1271 return (NULL); 1272 } 1273 1274 ETHER_BPF_MTAP(scifp, m); 1275 1276 m = (*sc->sc_input)(sc, lp, m); 1277 1278 if (m != NULL) { 1279 scifp->if_ipackets++; 1280 scifp->if_ibytes += m->m_pkthdr.len; 1281 1282 if (scifp->if_flags & IFF_MONITOR) { 1283 m_freem(m); 1284 m = NULL; 1285 } 1286 } 1287 1288 LAGG_RUNLOCK(sc); 1289 return (m); 1290 } 1291 1292 static int 1293 lagg_media_change(struct ifnet *ifp) 1294 { 1295 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1296 1297 if (sc->sc_ifflags & IFF_DEBUG) 1298 printf("%s\n", __func__); 1299 1300 /* Ignore */ 1301 return (0); 1302 } 1303 1304 static void 1305 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1306 { 1307 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1308 struct lagg_port *lp; 1309 1310 imr->ifm_status = IFM_AVALID; 1311 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1312 1313 LAGG_RLOCK(sc); 1314 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1315 if (LAGG_PORTACTIVE(lp)) 1316 imr->ifm_status |= IFM_ACTIVE; 1317 } 1318 LAGG_RUNLOCK(sc); 1319 } 1320 1321 static void 1322 lagg_linkstate(struct lagg_softc *sc) 1323 { 1324 struct lagg_port *lp; 1325 int new_link = LINK_STATE_DOWN; 1326 uint64_t speed; 1327 1328 /* Our link is considered up if at least one of our ports is active */ 1329 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1330 if (lp->lp_link_state == LINK_STATE_UP) { 1331 new_link = LINK_STATE_UP; 1332 break; 1333 } 1334 } 1335 if_link_state_change(sc->sc_ifp, new_link); 1336 1337 /* Update if_baudrate to reflect the max possible speed */ 1338 switch (sc->sc_proto) { 1339 case LAGG_PROTO_FAILOVER: 1340 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1341 sc->sc_primary->lp_ifp->if_baudrate : 0; 1342 break; 1343 case LAGG_PROTO_ROUNDROBIN: 1344 case LAGG_PROTO_LOADBALANCE: 1345 case LAGG_PROTO_ETHERCHANNEL: 1346 speed = 0; 1347 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1348 speed += lp->lp_ifp->if_baudrate; 1349 sc->sc_ifp->if_baudrate = speed; 1350 break; 1351 case LAGG_PROTO_LACP: 1352 /* LACP updates if_baudrate itself */ 1353 break; 1354 } 1355 } 1356 1357 static void 1358 lagg_port_state(struct ifnet *ifp, int state) 1359 { 1360 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1361 struct lagg_softc *sc = NULL; 1362 1363 if (lp != NULL) 1364 sc = lp->lp_softc; 1365 if (sc == NULL) 1366 return; 1367 1368 LAGG_WLOCK(sc); 1369 lagg_linkstate(sc); 1370 if (sc->sc_linkstate != NULL) 1371 (*sc->sc_linkstate)(lp); 1372 LAGG_WUNLOCK(sc); 1373 } 1374 1375 struct lagg_port * 1376 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1377 { 1378 struct lagg_port *lp_next, *rval = NULL; 1379 // int new_link = LINK_STATE_DOWN; 1380 1381 LAGG_RLOCK_ASSERT(sc); 1382 /* 1383 * Search a port which reports an active link state. 1384 */ 1385 1386 if (lp == NULL) 1387 goto search; 1388 if (LAGG_PORTACTIVE(lp)) { 1389 rval = lp; 1390 goto found; 1391 } 1392 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1393 LAGG_PORTACTIVE(lp_next)) { 1394 rval = lp_next; 1395 goto found; 1396 } 1397 1398 search: 1399 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1400 if (LAGG_PORTACTIVE(lp_next)) { 1401 rval = lp_next; 1402 goto found; 1403 } 1404 } 1405 1406 found: 1407 if (rval != NULL) { 1408 /* 1409 * The IEEE 802.1D standard assumes that a lagg with 1410 * multiple ports is always full duplex. This is valid 1411 * for load sharing laggs and if at least two links 1412 * are active. Unfortunately, checking the latter would 1413 * be too expensive at this point. 1414 XXX 1415 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1416 (sc->sc_count > 1)) 1417 new_link = LINK_STATE_FULL_DUPLEX; 1418 else 1419 new_link = rval->lp_link_state; 1420 */ 1421 } 1422 1423 return (rval); 1424 } 1425 1426 static const void * 1427 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf) 1428 { 1429 if (m->m_pkthdr.len < (off + len)) { 1430 return (NULL); 1431 } else if (m->m_len < (off + len)) { 1432 m_copydata(m, off, len, buf); 1433 return (buf); 1434 } 1435 return (mtod(m, char *) + off); 1436 } 1437 1438 uint32_t 1439 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key) 1440 { 1441 uint16_t etype; 1442 uint32_t p = key; 1443 int off; 1444 struct ether_header *eh; 1445 const struct ether_vlan_header *vlan; 1446 #ifdef INET 1447 const struct ip *ip; 1448 const uint32_t *ports; 1449 int iphlen; 1450 #endif 1451 #ifdef INET6 1452 const struct ip6_hdr *ip6; 1453 uint32_t flow; 1454 #endif 1455 union { 1456 #ifdef INET 1457 struct ip ip; 1458 #endif 1459 #ifdef INET6 1460 struct ip6_hdr ip6; 1461 #endif 1462 struct ether_vlan_header vlan; 1463 uint32_t port; 1464 } buf; 1465 1466 1467 off = sizeof(*eh); 1468 if (m->m_len < off) 1469 goto out; 1470 eh = mtod(m, struct ether_header *); 1471 etype = ntohs(eh->ether_type); 1472 if (sc->sc_flags & LAGG_F_HASHL2) { 1473 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p); 1474 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p); 1475 } 1476 1477 /* Special handling for encapsulating VLAN frames */ 1478 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) { 1479 p = hash32_buf(&m->m_pkthdr.ether_vtag, 1480 sizeof(m->m_pkthdr.ether_vtag), p); 1481 } else if (etype == ETHERTYPE_VLAN) { 1482 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf); 1483 if (vlan == NULL) 1484 goto out; 1485 1486 if (sc->sc_flags & LAGG_F_HASHL2) 1487 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p); 1488 etype = ntohs(vlan->evl_proto); 1489 off += sizeof(*vlan) - sizeof(*eh); 1490 } 1491 1492 switch (etype) { 1493 #ifdef INET 1494 case ETHERTYPE_IP: 1495 ip = lagg_gethdr(m, off, sizeof(*ip), &buf); 1496 if (ip == NULL) 1497 goto out; 1498 1499 if (sc->sc_flags & LAGG_F_HASHL3) { 1500 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p); 1501 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p); 1502 } 1503 if (!(sc->sc_flags & LAGG_F_HASHL4)) 1504 break; 1505 switch (ip->ip_p) { 1506 case IPPROTO_TCP: 1507 case IPPROTO_UDP: 1508 case IPPROTO_SCTP: 1509 iphlen = ip->ip_hl << 2; 1510 if (iphlen < sizeof(*ip)) 1511 break; 1512 off += iphlen; 1513 ports = lagg_gethdr(m, off, sizeof(*ports), &buf); 1514 if (ports == NULL) 1515 break; 1516 p = hash32_buf(ports, sizeof(*ports), p); 1517 break; 1518 } 1519 break; 1520 #endif 1521 #ifdef INET6 1522 case ETHERTYPE_IPV6: 1523 if (!(sc->sc_flags & LAGG_F_HASHL3)) 1524 break; 1525 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf); 1526 if (ip6 == NULL) 1527 goto out; 1528 1529 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p); 1530 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p); 1531 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK; 1532 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */ 1533 break; 1534 #endif 1535 } 1536 out: 1537 return (p); 1538 } 1539 1540 int 1541 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1542 { 1543 1544 return (ifp->if_transmit)(ifp, m); 1545 } 1546 1547 /* 1548 * Simple round robin aggregation 1549 */ 1550 1551 static int 1552 lagg_rr_attach(struct lagg_softc *sc) 1553 { 1554 sc->sc_detach = lagg_rr_detach; 1555 sc->sc_start = lagg_rr_start; 1556 sc->sc_input = lagg_rr_input; 1557 sc->sc_port_create = NULL; 1558 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1559 sc->sc_seq = 0; 1560 1561 return (0); 1562 } 1563 1564 static int 1565 lagg_rr_detach(struct lagg_softc *sc) 1566 { 1567 return (0); 1568 } 1569 1570 static int 1571 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 1572 { 1573 struct lagg_port *lp; 1574 uint32_t p; 1575 1576 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1577 p %= sc->sc_count; 1578 lp = SLIST_FIRST(&sc->sc_ports); 1579 while (p--) 1580 lp = SLIST_NEXT(lp, lp_entries); 1581 1582 /* 1583 * Check the port's link state. This will return the next active 1584 * port if the link is down or the port is NULL. 1585 */ 1586 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1587 m_freem(m); 1588 return (ENOENT); 1589 } 1590 1591 /* Send mbuf */ 1592 return (lagg_enqueue(lp->lp_ifp, m)); 1593 } 1594 1595 static struct mbuf * 1596 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1597 { 1598 struct ifnet *ifp = sc->sc_ifp; 1599 1600 /* Just pass in the packet to our lagg device */ 1601 m->m_pkthdr.rcvif = ifp; 1602 1603 return (m); 1604 } 1605 1606 /* 1607 * Active failover 1608 */ 1609 1610 static int 1611 lagg_fail_attach(struct lagg_softc *sc) 1612 { 1613 sc->sc_detach = lagg_fail_detach; 1614 sc->sc_start = lagg_fail_start; 1615 sc->sc_input = lagg_fail_input; 1616 sc->sc_port_create = NULL; 1617 sc->sc_port_destroy = NULL; 1618 1619 return (0); 1620 } 1621 1622 static int 1623 lagg_fail_detach(struct lagg_softc *sc) 1624 { 1625 return (0); 1626 } 1627 1628 static int 1629 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 1630 { 1631 struct lagg_port *lp; 1632 1633 /* Use the master port if active or the next available port */ 1634 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 1635 m_freem(m); 1636 return (ENOENT); 1637 } 1638 1639 /* Send mbuf */ 1640 return (lagg_enqueue(lp->lp_ifp, m)); 1641 } 1642 1643 static struct mbuf * 1644 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1645 { 1646 struct ifnet *ifp = sc->sc_ifp; 1647 struct lagg_port *tmp_tp; 1648 1649 if (lp == sc->sc_primary || lagg_failover_rx_all) { 1650 m->m_pkthdr.rcvif = ifp; 1651 return (m); 1652 } 1653 1654 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1655 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1656 /* 1657 * If tmp_tp is null, we've recieved a packet when all 1658 * our links are down. Weird, but process it anyways. 1659 */ 1660 if ((tmp_tp == NULL || tmp_tp == lp)) { 1661 m->m_pkthdr.rcvif = ifp; 1662 return (m); 1663 } 1664 } 1665 1666 m_freem(m); 1667 return (NULL); 1668 } 1669 1670 /* 1671 * Loadbalancing 1672 */ 1673 1674 static int 1675 lagg_lb_attach(struct lagg_softc *sc) 1676 { 1677 struct lagg_port *lp; 1678 struct lagg_lb *lb; 1679 1680 if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb), 1681 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 1682 return (ENOMEM); 1683 1684 sc->sc_detach = lagg_lb_detach; 1685 sc->sc_start = lagg_lb_start; 1686 sc->sc_input = lagg_lb_input; 1687 sc->sc_port_create = lagg_lb_port_create; 1688 sc->sc_port_destroy = lagg_lb_port_destroy; 1689 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1690 1691 lb->lb_key = arc4random(); 1692 sc->sc_psc = (caddr_t)lb; 1693 1694 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1695 lagg_lb_port_create(lp); 1696 1697 return (0); 1698 } 1699 1700 static int 1701 lagg_lb_detach(struct lagg_softc *sc) 1702 { 1703 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1704 if (lb != NULL) 1705 free(lb, M_DEVBUF); 1706 return (0); 1707 } 1708 1709 static int 1710 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 1711 { 1712 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1713 struct lagg_port *lp_next; 1714 int i = 0; 1715 1716 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 1717 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1718 if (lp_next == lp) 1719 continue; 1720 if (i >= LAGG_MAX_PORTS) 1721 return (EINVAL); 1722 if (sc->sc_ifflags & IFF_DEBUG) 1723 printf("%s: port %s at index %d\n", 1724 sc->sc_ifname, lp_next->lp_ifname, i); 1725 lb->lb_ports[i++] = lp_next; 1726 } 1727 1728 return (0); 1729 } 1730 1731 static int 1732 lagg_lb_port_create(struct lagg_port *lp) 1733 { 1734 struct lagg_softc *sc = lp->lp_softc; 1735 return (lagg_lb_porttable(sc, NULL)); 1736 } 1737 1738 static void 1739 lagg_lb_port_destroy(struct lagg_port *lp) 1740 { 1741 struct lagg_softc *sc = lp->lp_softc; 1742 lagg_lb_porttable(sc, lp); 1743 } 1744 1745 static int 1746 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 1747 { 1748 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1749 struct lagg_port *lp = NULL; 1750 uint32_t p = 0; 1751 1752 if (sc->use_flowid && (m->m_flags & M_FLOWID)) 1753 p = m->m_pkthdr.flowid; 1754 else 1755 p = lagg_hashmbuf(sc, m, lb->lb_key); 1756 p %= sc->sc_count; 1757 lp = lb->lb_ports[p]; 1758 1759 /* 1760 * Check the port's link state. This will return the next active 1761 * port if the link is down or the port is NULL. 1762 */ 1763 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1764 m_freem(m); 1765 return (ENOENT); 1766 } 1767 1768 /* Send mbuf */ 1769 return (lagg_enqueue(lp->lp_ifp, m)); 1770 } 1771 1772 static struct mbuf * 1773 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1774 { 1775 struct ifnet *ifp = sc->sc_ifp; 1776 1777 /* Just pass in the packet to our lagg device */ 1778 m->m_pkthdr.rcvif = ifp; 1779 1780 return (m); 1781 } 1782 1783 /* 1784 * 802.3ad LACP 1785 */ 1786 1787 static int 1788 lagg_lacp_attach(struct lagg_softc *sc) 1789 { 1790 struct lagg_port *lp; 1791 int error; 1792 1793 sc->sc_detach = lagg_lacp_detach; 1794 sc->sc_port_create = lacp_port_create; 1795 sc->sc_port_destroy = lacp_port_destroy; 1796 sc->sc_linkstate = lacp_linkstate; 1797 sc->sc_start = lagg_lacp_start; 1798 sc->sc_input = lagg_lacp_input; 1799 sc->sc_init = lacp_init; 1800 sc->sc_stop = lacp_stop; 1801 sc->sc_lladdr = lagg_lacp_lladdr; 1802 sc->sc_req = lacp_req; 1803 sc->sc_portreq = lacp_portreq; 1804 1805 error = lacp_attach(sc); 1806 if (error) 1807 return (error); 1808 1809 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1810 lacp_port_create(lp); 1811 1812 return (error); 1813 } 1814 1815 static int 1816 lagg_lacp_detach(struct lagg_softc *sc) 1817 { 1818 struct lagg_port *lp; 1819 int error; 1820 1821 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1822 lacp_port_destroy(lp); 1823 1824 /* unlocking is safe here */ 1825 LAGG_WUNLOCK(sc); 1826 error = lacp_detach(sc); 1827 LAGG_WLOCK(sc); 1828 1829 return (error); 1830 } 1831 1832 static void 1833 lagg_lacp_lladdr(struct lagg_softc *sc) 1834 { 1835 struct lagg_port *lp; 1836 1837 /* purge all the lacp ports */ 1838 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1839 lacp_port_destroy(lp); 1840 1841 /* add them back in */ 1842 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1843 lacp_port_create(lp); 1844 } 1845 1846 static int 1847 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 1848 { 1849 struct lagg_port *lp; 1850 1851 lp = lacp_select_tx_port(sc, m); 1852 if (lp == NULL) { 1853 m_freem(m); 1854 return (EBUSY); 1855 } 1856 1857 /* Send mbuf */ 1858 return (lagg_enqueue(lp->lp_ifp, m)); 1859 } 1860 1861 static struct mbuf * 1862 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1863 { 1864 struct ifnet *ifp = sc->sc_ifp; 1865 struct ether_header *eh; 1866 u_short etype; 1867 1868 eh = mtod(m, struct ether_header *); 1869 etype = ntohs(eh->ether_type); 1870 1871 /* Tap off LACP control messages */ 1872 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 1873 m = lacp_input(lp, m); 1874 if (m == NULL) 1875 return (NULL); 1876 } 1877 1878 /* 1879 * If the port is not collecting or not in the active aggregator then 1880 * free and return. 1881 */ 1882 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 1883 m_freem(m); 1884 return (NULL); 1885 } 1886 1887 m->m_pkthdr.rcvif = ifp; 1888 return (m); 1889 } 1890