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 struct if_clone *lagg_cloner; 89 static const char laggname[] = "lagg"; 90 91 static void lagg_lladdr(struct lagg_softc *, uint8_t *); 92 static void lagg_capabilities(struct lagg_softc *); 93 static void lagg_port_lladdr(struct lagg_port *, uint8_t *); 94 static void lagg_port_setlladdr(void *, int); 95 static int lagg_port_create(struct lagg_softc *, struct ifnet *); 96 static int lagg_port_destroy(struct lagg_port *, int); 97 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); 98 static void lagg_linkstate(struct lagg_softc *); 99 static void lagg_port_state(struct ifnet *, int); 100 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); 101 static int lagg_port_output(struct ifnet *, struct mbuf *, 102 const struct sockaddr *, struct route *); 103 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 104 #ifdef LAGG_PORT_STACKING 105 static int lagg_port_checkstacking(struct lagg_softc *); 106 #endif 107 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 108 static void lagg_init(void *); 109 static void lagg_stop(struct lagg_softc *); 110 static int lagg_ioctl(struct ifnet *, u_long, caddr_t); 111 static int lagg_ether_setmulti(struct lagg_softc *); 112 static int lagg_ether_cmdmulti(struct lagg_port *, int); 113 static int lagg_setflag(struct lagg_port *, int, int, 114 int (*func)(struct ifnet *, int)); 115 static int lagg_setflags(struct lagg_port *, int status); 116 static int lagg_transmit(struct ifnet *, struct mbuf *); 117 static void lagg_qflush(struct ifnet *); 118 static int lagg_media_change(struct ifnet *); 119 static void lagg_media_status(struct ifnet *, struct ifmediareq *); 120 static struct lagg_port *lagg_link_active(struct lagg_softc *, 121 struct lagg_port *); 122 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *); 123 124 /* Simple round robin */ 125 static int lagg_rr_attach(struct lagg_softc *); 126 static int lagg_rr_detach(struct lagg_softc *); 127 static int lagg_rr_start(struct lagg_softc *, struct mbuf *); 128 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 129 struct mbuf *); 130 131 /* Active failover */ 132 static int lagg_fail_attach(struct lagg_softc *); 133 static int lagg_fail_detach(struct lagg_softc *); 134 static int lagg_fail_start(struct lagg_softc *, struct mbuf *); 135 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 136 struct mbuf *); 137 138 /* Loadbalancing */ 139 static int lagg_lb_attach(struct lagg_softc *); 140 static int lagg_lb_detach(struct lagg_softc *); 141 static int lagg_lb_port_create(struct lagg_port *); 142 static void lagg_lb_port_destroy(struct lagg_port *); 143 static int lagg_lb_start(struct lagg_softc *, struct mbuf *); 144 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 145 struct mbuf *); 146 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 147 148 /* 802.3ad LACP */ 149 static int lagg_lacp_attach(struct lagg_softc *); 150 static int lagg_lacp_detach(struct lagg_softc *); 151 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); 152 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 153 struct mbuf *); 154 static void lagg_lacp_lladdr(struct lagg_softc *); 155 156 static void lagg_callout(void *); 157 158 /* lagg protocol table */ 159 static const struct { 160 int ti_proto; 161 int (*ti_attach)(struct lagg_softc *); 162 } lagg_protos[] = { 163 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach }, 164 { LAGG_PROTO_FAILOVER, lagg_fail_attach }, 165 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach }, 166 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach }, 167 { LAGG_PROTO_LACP, lagg_lacp_attach }, 168 { LAGG_PROTO_NONE, NULL } 169 }; 170 171 SYSCTL_DECL(_net_link); 172 static SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 173 "Link Aggregation"); 174 175 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */ 176 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW, 177 &lagg_failover_rx_all, 0, 178 "Accept input from any interface in a failover lagg"); 179 static int def_use_flowid = 1; /* Default value for using M_FLOWID */ 180 TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid); 181 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RW, 182 &def_use_flowid, 0, 183 "Default setting for using flow id for load sharing"); 184 185 static int 186 lagg_modevent(module_t mod, int type, void *data) 187 { 188 189 switch (type) { 190 case MOD_LOAD: 191 mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF); 192 SLIST_INIT(&lagg_list); 193 lagg_cloner = if_clone_simple(laggname, lagg_clone_create, 194 lagg_clone_destroy, 0); 195 lagg_input_p = lagg_input; 196 lagg_linkstate_p = lagg_port_state; 197 lagg_detach_cookie = EVENTHANDLER_REGISTER( 198 ifnet_departure_event, lagg_port_ifdetach, NULL, 199 EVENTHANDLER_PRI_ANY); 200 break; 201 case MOD_UNLOAD: 202 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 203 lagg_detach_cookie); 204 if_clone_detach(lagg_cloner); 205 lagg_input_p = NULL; 206 lagg_linkstate_p = NULL; 207 mtx_destroy(&lagg_list_mtx); 208 break; 209 default: 210 return (EOPNOTSUPP); 211 } 212 return (0); 213 } 214 215 static moduledata_t lagg_mod = { 216 "if_lagg", 217 lagg_modevent, 218 0 219 }; 220 221 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 222 MODULE_VERSION(if_lagg, 1); 223 224 /* 225 * This routine is run via an vlan 226 * config EVENT 227 */ 228 static void 229 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 230 { 231 struct lagg_softc *sc = ifp->if_softc; 232 struct lagg_port *lp; 233 234 if (ifp->if_softc != arg) /* Not our event */ 235 return; 236 237 LAGG_RLOCK(sc); 238 if (!SLIST_EMPTY(&sc->sc_ports)) { 239 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 240 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 241 } 242 LAGG_RUNLOCK(sc); 243 } 244 245 /* 246 * This routine is run via an vlan 247 * unconfig EVENT 248 */ 249 static void 250 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 251 { 252 struct lagg_softc *sc = ifp->if_softc; 253 struct lagg_port *lp; 254 255 if (ifp->if_softc != arg) /* Not our event */ 256 return; 257 258 LAGG_RLOCK(sc); 259 if (!SLIST_EMPTY(&sc->sc_ports)) { 260 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 261 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 262 } 263 LAGG_RUNLOCK(sc); 264 } 265 266 static int 267 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) 268 { 269 struct lagg_softc *sc; 270 struct ifnet *ifp; 271 int i, error = 0; 272 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ 273 struct sysctl_oid *oid; 274 char num[14]; /* sufficient for 32 bits */ 275 276 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO); 277 ifp = sc->sc_ifp = if_alloc(IFT_ETHER); 278 if (ifp == NULL) { 279 free(sc, M_DEVBUF); 280 return (ENOSPC); 281 } 282 283 sc->sc_ipackets = counter_u64_alloc(M_WAITOK); 284 sc->sc_opackets = counter_u64_alloc(M_WAITOK); 285 sc->sc_ibytes = counter_u64_alloc(M_WAITOK); 286 sc->sc_obytes = counter_u64_alloc(M_WAITOK); 287 288 sysctl_ctx_init(&sc->ctx); 289 snprintf(num, sizeof(num), "%u", unit); 290 sc->use_flowid = def_use_flowid; 291 oid = SYSCTL_ADD_NODE(&sc->ctx, &SYSCTL_NODE_CHILDREN(_net_link, lagg), 292 OID_AUTO, num, CTLFLAG_RD, NULL, ""); 293 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 294 "use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid, sc->use_flowid, 295 "Use flow id for load sharing"); 296 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 297 "count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count, 298 "Total number of ports"); 299 /* Hash all layers by default */ 300 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4; 301 302 sc->sc_proto = LAGG_PROTO_NONE; 303 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) { 304 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) { 305 sc->sc_proto = lagg_protos[i].ti_proto; 306 if ((error = lagg_protos[i].ti_attach(sc)) != 0) { 307 if_free(ifp); 308 free(sc, M_DEVBUF); 309 return (error); 310 } 311 break; 312 } 313 } 314 LAGG_LOCK_INIT(sc); 315 SLIST_INIT(&sc->sc_ports); 316 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc); 317 callout_init_rw(&sc->sc_callout, &sc->sc_mtx, CALLOUT_SHAREDLOCK); 318 319 /* Initialise pseudo media types */ 320 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 321 lagg_media_status); 322 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 323 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 324 325 if_initname(ifp, laggname, unit); 326 ifp->if_softc = sc; 327 ifp->if_transmit = lagg_transmit; 328 ifp->if_qflush = lagg_qflush; 329 ifp->if_init = lagg_init; 330 ifp->if_ioctl = lagg_ioctl; 331 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 332 333 /* 334 * Attach as an ordinary ethernet device, children will be attached 335 * as special device IFT_IEEE8023ADLAG. 336 */ 337 ether_ifattach(ifp, eaddr); 338 339 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 340 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 341 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 342 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 343 344 /* Insert into the global list of laggs */ 345 mtx_lock(&lagg_list_mtx); 346 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries); 347 mtx_unlock(&lagg_list_mtx); 348 349 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 350 351 return (0); 352 } 353 354 static void 355 lagg_clone_destroy(struct ifnet *ifp) 356 { 357 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 358 struct lagg_port *lp; 359 360 LAGG_WLOCK(sc); 361 362 lagg_stop(sc); 363 ifp->if_flags &= ~IFF_UP; 364 365 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 366 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 367 368 /* Shutdown and remove lagg ports */ 369 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) 370 lagg_port_destroy(lp, 1); 371 /* Unhook the aggregation protocol */ 372 if (sc->sc_detach != NULL) 373 (*sc->sc_detach)(sc); 374 375 LAGG_WUNLOCK(sc); 376 377 sysctl_ctx_free(&sc->ctx); 378 ifmedia_removeall(&sc->sc_media); 379 ether_ifdetach(ifp); 380 if_free(ifp); 381 382 callout_drain(&sc->sc_callout); 383 counter_u64_free(sc->sc_ipackets); 384 counter_u64_free(sc->sc_opackets); 385 counter_u64_free(sc->sc_ibytes); 386 counter_u64_free(sc->sc_obytes); 387 388 mtx_lock(&lagg_list_mtx); 389 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries); 390 mtx_unlock(&lagg_list_mtx); 391 392 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task); 393 LAGG_LOCK_DESTROY(sc); 394 free(sc, M_DEVBUF); 395 } 396 397 static void 398 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr) 399 { 400 struct ifnet *ifp = sc->sc_ifp; 401 402 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 403 return; 404 405 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN); 406 /* Let the protocol know the MAC has changed */ 407 if (sc->sc_lladdr != NULL) 408 (*sc->sc_lladdr)(sc); 409 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 410 } 411 412 static void 413 lagg_capabilities(struct lagg_softc *sc) 414 { 415 struct lagg_port *lp; 416 int cap = ~0, ena = ~0; 417 u_long hwa = ~0UL; 418 419 LAGG_WLOCK_ASSERT(sc); 420 421 /* Get capabilities from the lagg ports */ 422 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 423 cap &= lp->lp_ifp->if_capabilities; 424 ena &= lp->lp_ifp->if_capenable; 425 hwa &= lp->lp_ifp->if_hwassist; 426 } 427 cap = (cap == ~0 ? 0 : cap); 428 ena = (ena == ~0 ? 0 : ena); 429 hwa = (hwa == ~0 ? 0 : hwa); 430 431 if (sc->sc_ifp->if_capabilities != cap || 432 sc->sc_ifp->if_capenable != ena || 433 sc->sc_ifp->if_hwassist != hwa) { 434 sc->sc_ifp->if_capabilities = cap; 435 sc->sc_ifp->if_capenable = ena; 436 sc->sc_ifp->if_hwassist = hwa; 437 getmicrotime(&sc->sc_ifp->if_lastchange); 438 439 if (sc->sc_ifflags & IFF_DEBUG) 440 if_printf(sc->sc_ifp, 441 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 442 } 443 } 444 445 static void 446 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr) 447 { 448 struct lagg_softc *sc = lp->lp_softc; 449 struct ifnet *ifp = lp->lp_ifp; 450 struct lagg_llq *llq; 451 int pending = 0; 452 453 LAGG_WLOCK_ASSERT(sc); 454 455 if (lp->lp_detaching || 456 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 457 return; 458 459 /* Check to make sure its not already queued to be changed */ 460 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 461 if (llq->llq_ifp == ifp) { 462 pending = 1; 463 break; 464 } 465 } 466 467 if (!pending) { 468 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT); 469 if (llq == NULL) /* XXX what to do */ 470 return; 471 } 472 473 /* Update the lladdr even if pending, it may have changed */ 474 llq->llq_ifp = ifp; 475 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN); 476 477 if (!pending) 478 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries); 479 480 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task); 481 } 482 483 /* 484 * Set the interface MAC address from a taskqueue to avoid a LOR. 485 */ 486 static void 487 lagg_port_setlladdr(void *arg, int pending) 488 { 489 struct lagg_softc *sc = (struct lagg_softc *)arg; 490 struct lagg_llq *llq, *head; 491 struct ifnet *ifp; 492 int error; 493 494 /* Grab a local reference of the queue and remove it from the softc */ 495 LAGG_WLOCK(sc); 496 head = SLIST_FIRST(&sc->sc_llq_head); 497 SLIST_FIRST(&sc->sc_llq_head) = NULL; 498 LAGG_WUNLOCK(sc); 499 500 /* 501 * Traverse the queue and set the lladdr on each ifp. It is safe to do 502 * unlocked as we have the only reference to it. 503 */ 504 for (llq = head; llq != NULL; llq = head) { 505 ifp = llq->llq_ifp; 506 507 /* Set the link layer address */ 508 CURVNET_SET(ifp->if_vnet); 509 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN); 510 CURVNET_RESTORE(); 511 if (error) 512 printf("%s: setlladdr failed on %s\n", __func__, 513 ifp->if_xname); 514 515 head = SLIST_NEXT(llq, llq_entries); 516 free(llq, M_DEVBUF); 517 } 518 } 519 520 static int 521 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 522 { 523 struct lagg_softc *sc_ptr; 524 struct lagg_port *lp; 525 int error = 0; 526 527 LAGG_WLOCK_ASSERT(sc); 528 529 /* Limit the maximal number of lagg ports */ 530 if (sc->sc_count >= LAGG_MAX_PORTS) 531 return (ENOSPC); 532 533 /* Check if port has already been associated to a lagg */ 534 if (ifp->if_lagg != NULL) { 535 /* Port is already in the current lagg? */ 536 lp = (struct lagg_port *)ifp->if_lagg; 537 if (lp->lp_softc == sc) 538 return (EEXIST); 539 return (EBUSY); 540 } 541 542 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 543 if (ifp->if_type != IFT_ETHER) 544 return (EPROTONOSUPPORT); 545 546 /* Allow the first Ethernet member to define the MTU */ 547 if (SLIST_EMPTY(&sc->sc_ports)) 548 sc->sc_ifp->if_mtu = ifp->if_mtu; 549 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 550 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 551 ifp->if_xname); 552 return (EINVAL); 553 } 554 555 if ((lp = malloc(sizeof(struct lagg_port), 556 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 557 return (ENOMEM); 558 559 /* Check if port is a stacked lagg */ 560 mtx_lock(&lagg_list_mtx); 561 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) { 562 if (ifp == sc_ptr->sc_ifp) { 563 mtx_unlock(&lagg_list_mtx); 564 free(lp, M_DEVBUF); 565 return (EINVAL); 566 /* XXX disable stacking for the moment, its untested */ 567 #ifdef LAGG_PORT_STACKING 568 lp->lp_flags |= LAGG_PORT_STACK; 569 if (lagg_port_checkstacking(sc_ptr) >= 570 LAGG_MAX_STACKING) { 571 mtx_unlock(&lagg_list_mtx); 572 free(lp, M_DEVBUF); 573 return (E2BIG); 574 } 575 #endif 576 } 577 } 578 mtx_unlock(&lagg_list_mtx); 579 580 /* Change the interface type */ 581 lp->lp_iftype = ifp->if_type; 582 ifp->if_type = IFT_IEEE8023ADLAG; 583 ifp->if_lagg = lp; 584 lp->lp_ioctl = ifp->if_ioctl; 585 ifp->if_ioctl = lagg_port_ioctl; 586 lp->lp_output = ifp->if_output; 587 ifp->if_output = lagg_port_output; 588 589 lp->lp_ifp = ifp; 590 lp->lp_softc = sc; 591 592 /* Save port link layer address */ 593 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 594 595 if (SLIST_EMPTY(&sc->sc_ports)) { 596 sc->sc_primary = lp; 597 lagg_lladdr(sc, IF_LLADDR(ifp)); 598 } else { 599 /* Update link layer address for this port */ 600 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp)); 601 } 602 603 /* Insert into the list of ports */ 604 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 605 sc->sc_count++; 606 607 /* Update lagg capabilities */ 608 lagg_capabilities(sc); 609 lagg_linkstate(sc); 610 611 /* Add multicast addresses and interface flags to this port */ 612 lagg_ether_cmdmulti(lp, 1); 613 lagg_setflags(lp, 1); 614 615 if (sc->sc_port_create != NULL) 616 error = (*sc->sc_port_create)(lp); 617 if (error) { 618 /* remove the port again, without calling sc_port_destroy */ 619 lagg_port_destroy(lp, 0); 620 return (error); 621 } 622 623 return (error); 624 } 625 626 #ifdef LAGG_PORT_STACKING 627 static int 628 lagg_port_checkstacking(struct lagg_softc *sc) 629 { 630 struct lagg_softc *sc_ptr; 631 struct lagg_port *lp; 632 int m = 0; 633 634 LAGG_WLOCK_ASSERT(sc); 635 636 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 637 if (lp->lp_flags & LAGG_PORT_STACK) { 638 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 639 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 640 } 641 } 642 643 return (m + 1); 644 } 645 #endif 646 647 static int 648 lagg_port_destroy(struct lagg_port *lp, int runpd) 649 { 650 struct lagg_softc *sc = lp->lp_softc; 651 struct lagg_port *lp_ptr; 652 struct lagg_llq *llq; 653 struct ifnet *ifp = lp->lp_ifp; 654 655 LAGG_WLOCK_ASSERT(sc); 656 657 if (runpd && sc->sc_port_destroy != NULL) 658 (*sc->sc_port_destroy)(lp); 659 660 /* 661 * Remove multicast addresses and interface flags from this port and 662 * reset the MAC address, skip if the interface is being detached. 663 */ 664 if (!lp->lp_detaching) { 665 lagg_ether_cmdmulti(lp, 0); 666 lagg_setflags(lp, 0); 667 lagg_port_lladdr(lp, lp->lp_lladdr); 668 } 669 670 /* Restore interface */ 671 ifp->if_type = lp->lp_iftype; 672 ifp->if_ioctl = lp->lp_ioctl; 673 ifp->if_output = lp->lp_output; 674 ifp->if_lagg = NULL; 675 676 /* Finally, remove the port from the lagg */ 677 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 678 sc->sc_count--; 679 680 /* Update the primary interface */ 681 if (lp == sc->sc_primary) { 682 uint8_t lladdr[ETHER_ADDR_LEN]; 683 684 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) { 685 bzero(&lladdr, ETHER_ADDR_LEN); 686 } else { 687 bcopy(lp_ptr->lp_lladdr, 688 lladdr, ETHER_ADDR_LEN); 689 } 690 lagg_lladdr(sc, lladdr); 691 sc->sc_primary = lp_ptr; 692 693 /* Update link layer address for each port */ 694 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 695 lagg_port_lladdr(lp_ptr, lladdr); 696 } 697 698 /* Remove any pending lladdr changes from the queue */ 699 if (lp->lp_detaching) { 700 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 701 if (llq->llq_ifp == ifp) { 702 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq, 703 llq_entries); 704 free(llq, M_DEVBUF); 705 break; /* Only appears once */ 706 } 707 } 708 } 709 710 if (lp->lp_ifflags) 711 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 712 713 free(lp, M_DEVBUF); 714 715 /* Update lagg capabilities */ 716 lagg_capabilities(sc); 717 lagg_linkstate(sc); 718 719 return (0); 720 } 721 722 static int 723 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 724 { 725 struct lagg_reqport *rp = (struct lagg_reqport *)data; 726 struct lagg_softc *sc; 727 struct lagg_port *lp = NULL; 728 int error = 0; 729 730 /* Should be checked by the caller */ 731 if (ifp->if_type != IFT_IEEE8023ADLAG || 732 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 733 goto fallback; 734 735 switch (cmd) { 736 case SIOCGLAGGPORT: 737 if (rp->rp_portname[0] == '\0' || 738 ifunit(rp->rp_portname) != ifp) { 739 error = EINVAL; 740 break; 741 } 742 743 LAGG_RLOCK(sc); 744 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 745 error = ENOENT; 746 LAGG_RUNLOCK(sc); 747 break; 748 } 749 750 lagg_port2req(lp, rp); 751 LAGG_RUNLOCK(sc); 752 break; 753 754 case SIOCSIFCAP: 755 if (lp->lp_ioctl == NULL) { 756 error = EINVAL; 757 break; 758 } 759 error = (*lp->lp_ioctl)(ifp, cmd, data); 760 if (error) 761 break; 762 763 /* Update lagg interface capabilities */ 764 LAGG_WLOCK(sc); 765 lagg_capabilities(sc); 766 LAGG_WUNLOCK(sc); 767 break; 768 769 case SIOCSIFMTU: 770 /* Do not allow the MTU to be changed once joined */ 771 error = EINVAL; 772 break; 773 774 default: 775 goto fallback; 776 } 777 778 return (error); 779 780 fallback: 781 if (lp->lp_ioctl != NULL) 782 return ((*lp->lp_ioctl)(ifp, cmd, data)); 783 784 return (EINVAL); 785 } 786 787 /* 788 * For direct output to child ports. 789 */ 790 static int 791 lagg_port_output(struct ifnet *ifp, struct mbuf *m, 792 const struct sockaddr *dst, struct route *ro) 793 { 794 struct lagg_port *lp = ifp->if_lagg; 795 796 switch (dst->sa_family) { 797 case pseudo_AF_HDRCMPLT: 798 case AF_UNSPEC: 799 return ((*lp->lp_output)(ifp, m, dst, ro)); 800 } 801 802 /* drop any other frames */ 803 m_freem(m); 804 return (ENETDOWN); 805 } 806 807 static void 808 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 809 { 810 struct lagg_port *lp; 811 struct lagg_softc *sc; 812 813 if ((lp = ifp->if_lagg) == NULL) 814 return; 815 /* If the ifnet is just being renamed, don't do anything. */ 816 if (ifp->if_flags & IFF_RENAMING) 817 return; 818 819 sc = lp->lp_softc; 820 821 LAGG_WLOCK(sc); 822 lp->lp_detaching = 1; 823 lagg_port_destroy(lp, 1); 824 LAGG_WUNLOCK(sc); 825 } 826 827 static void 828 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 829 { 830 struct lagg_softc *sc = lp->lp_softc; 831 832 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 833 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 834 rp->rp_prio = lp->lp_prio; 835 rp->rp_flags = lp->lp_flags; 836 if (sc->sc_portreq != NULL) 837 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc); 838 839 /* Add protocol specific flags */ 840 switch (sc->sc_proto) { 841 case LAGG_PROTO_FAILOVER: 842 if (lp == sc->sc_primary) 843 rp->rp_flags |= LAGG_PORT_MASTER; 844 if (lp == lagg_link_active(sc, sc->sc_primary)) 845 rp->rp_flags |= LAGG_PORT_ACTIVE; 846 break; 847 848 case LAGG_PROTO_ROUNDROBIN: 849 case LAGG_PROTO_LOADBALANCE: 850 case LAGG_PROTO_ETHERCHANNEL: 851 if (LAGG_PORTACTIVE(lp)) 852 rp->rp_flags |= LAGG_PORT_ACTIVE; 853 break; 854 855 case LAGG_PROTO_LACP: 856 /* LACP has a different definition of active */ 857 if (lacp_isactive(lp)) 858 rp->rp_flags |= LAGG_PORT_ACTIVE; 859 if (lacp_iscollecting(lp)) 860 rp->rp_flags |= LAGG_PORT_COLLECTING; 861 if (lacp_isdistributing(lp)) 862 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 863 break; 864 } 865 866 } 867 868 static void 869 lagg_init(void *xsc) 870 { 871 struct lagg_softc *sc = (struct lagg_softc *)xsc; 872 struct lagg_port *lp; 873 struct ifnet *ifp = sc->sc_ifp; 874 875 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 876 return; 877 878 LAGG_WLOCK(sc); 879 880 ifp->if_drv_flags |= IFF_DRV_RUNNING; 881 /* Update the port lladdrs */ 882 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 883 lagg_port_lladdr(lp, IF_LLADDR(ifp)); 884 885 if (sc->sc_init != NULL) 886 (*sc->sc_init)(sc); 887 888 LAGG_WUNLOCK(sc); 889 } 890 891 static void 892 lagg_stop(struct lagg_softc *sc) 893 { 894 struct ifnet *ifp = sc->sc_ifp; 895 896 LAGG_WLOCK_ASSERT(sc); 897 898 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 899 return; 900 901 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 902 903 if (sc->sc_stop != NULL) 904 (*sc->sc_stop)(sc); 905 } 906 907 static int 908 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 909 { 910 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 911 struct lagg_reqall *ra = (struct lagg_reqall *)data; 912 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 913 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 914 struct ifreq *ifr = (struct ifreq *)data; 915 struct lagg_port *lp; 916 struct ifnet *tpif; 917 struct thread *td = curthread; 918 char *buf, *outbuf; 919 int count, buflen, len, error = 0; 920 921 bzero(&rpbuf, sizeof(rpbuf)); 922 923 switch (cmd) { 924 case SIOCGLAGG: 925 LAGG_RLOCK(sc); 926 count = 0; 927 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 928 count++; 929 buflen = count * sizeof(struct lagg_reqport); 930 LAGG_RUNLOCK(sc); 931 932 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 933 934 LAGG_RLOCK(sc); 935 ra->ra_proto = sc->sc_proto; 936 if (sc->sc_req != NULL) 937 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc); 938 939 count = 0; 940 buf = outbuf; 941 len = min(ra->ra_size, buflen); 942 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 943 if (len < sizeof(rpbuf)) 944 break; 945 946 lagg_port2req(lp, &rpbuf); 947 memcpy(buf, &rpbuf, sizeof(rpbuf)); 948 count++; 949 buf += sizeof(rpbuf); 950 len -= sizeof(rpbuf); 951 } 952 LAGG_RUNLOCK(sc); 953 ra->ra_ports = count; 954 ra->ra_size = count * sizeof(rpbuf); 955 error = copyout(outbuf, ra->ra_port, ra->ra_size); 956 free(outbuf, M_TEMP); 957 break; 958 case SIOCSLAGG: 959 error = priv_check(td, PRIV_NET_LAGG); 960 if (error) 961 break; 962 if (ra->ra_proto >= LAGG_PROTO_MAX) { 963 error = EPROTONOSUPPORT; 964 break; 965 } 966 LAGG_WLOCK(sc); 967 if (sc->sc_proto != LAGG_PROTO_NONE) { 968 /* Reset protocol first in case detach unlocks */ 969 sc->sc_proto = LAGG_PROTO_NONE; 970 error = sc->sc_detach(sc); 971 sc->sc_detach = NULL; 972 sc->sc_start = NULL; 973 sc->sc_input = NULL; 974 sc->sc_port_create = NULL; 975 sc->sc_port_destroy = NULL; 976 sc->sc_linkstate = NULL; 977 sc->sc_init = NULL; 978 sc->sc_stop = NULL; 979 sc->sc_lladdr = NULL; 980 sc->sc_req = NULL; 981 sc->sc_portreq = NULL; 982 } else if (sc->sc_input != NULL) { 983 /* Still detaching */ 984 error = EBUSY; 985 } 986 if (error != 0) { 987 LAGG_WUNLOCK(sc); 988 break; 989 } 990 for (int i = 0; i < (sizeof(lagg_protos) / 991 sizeof(lagg_protos[0])); i++) { 992 if (lagg_protos[i].ti_proto == ra->ra_proto) { 993 if (sc->sc_ifflags & IFF_DEBUG) 994 printf("%s: using proto %u\n", 995 sc->sc_ifname, 996 lagg_protos[i].ti_proto); 997 sc->sc_proto = lagg_protos[i].ti_proto; 998 if (sc->sc_proto != LAGG_PROTO_NONE) 999 error = lagg_protos[i].ti_attach(sc); 1000 LAGG_WUNLOCK(sc); 1001 return (error); 1002 } 1003 } 1004 LAGG_WUNLOCK(sc); 1005 error = EPROTONOSUPPORT; 1006 break; 1007 case SIOCGLAGGFLAGS: 1008 rf->rf_flags = sc->sc_flags; 1009 break; 1010 case SIOCSLAGGHASH: 1011 error = priv_check(td, PRIV_NET_LAGG); 1012 if (error) 1013 break; 1014 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1015 error = EINVAL; 1016 break; 1017 } 1018 LAGG_WLOCK(sc); 1019 sc->sc_flags &= ~LAGG_F_HASHMASK; 1020 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK; 1021 LAGG_WUNLOCK(sc); 1022 break; 1023 case SIOCGLAGGPORT: 1024 if (rp->rp_portname[0] == '\0' || 1025 (tpif = ifunit(rp->rp_portname)) == NULL) { 1026 error = EINVAL; 1027 break; 1028 } 1029 1030 LAGG_RLOCK(sc); 1031 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1032 lp->lp_softc != sc) { 1033 error = ENOENT; 1034 LAGG_RUNLOCK(sc); 1035 break; 1036 } 1037 1038 lagg_port2req(lp, rp); 1039 LAGG_RUNLOCK(sc); 1040 break; 1041 case SIOCSLAGGPORT: 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 LAGG_WLOCK(sc); 1051 error = lagg_port_create(sc, tpif); 1052 LAGG_WUNLOCK(sc); 1053 break; 1054 case SIOCSLAGGDELPORT: 1055 error = priv_check(td, PRIV_NET_LAGG); 1056 if (error) 1057 break; 1058 if (rp->rp_portname[0] == '\0' || 1059 (tpif = ifunit(rp->rp_portname)) == NULL) { 1060 error = EINVAL; 1061 break; 1062 } 1063 1064 LAGG_WLOCK(sc); 1065 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1066 lp->lp_softc != sc) { 1067 error = ENOENT; 1068 LAGG_WUNLOCK(sc); 1069 break; 1070 } 1071 1072 error = lagg_port_destroy(lp, 1); 1073 LAGG_WUNLOCK(sc); 1074 break; 1075 case SIOCSIFFLAGS: 1076 /* Set flags on ports too */ 1077 LAGG_WLOCK(sc); 1078 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1079 lagg_setflags(lp, 1); 1080 } 1081 LAGG_WUNLOCK(sc); 1082 1083 if (!(ifp->if_flags & IFF_UP) && 1084 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1085 /* 1086 * If interface is marked down and it is running, 1087 * then stop and disable it. 1088 */ 1089 LAGG_WLOCK(sc); 1090 lagg_stop(sc); 1091 LAGG_WUNLOCK(sc); 1092 } else if ((ifp->if_flags & IFF_UP) && 1093 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1094 /* 1095 * If interface is marked up and it is stopped, then 1096 * start it. 1097 */ 1098 (*ifp->if_init)(sc); 1099 } 1100 break; 1101 case SIOCADDMULTI: 1102 case SIOCDELMULTI: 1103 LAGG_WLOCK(sc); 1104 error = lagg_ether_setmulti(sc); 1105 LAGG_WUNLOCK(sc); 1106 break; 1107 case SIOCSIFMEDIA: 1108 case SIOCGIFMEDIA: 1109 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1110 break; 1111 1112 case SIOCSIFCAP: 1113 case SIOCSIFMTU: 1114 /* Do not allow the MTU or caps to be directly changed */ 1115 error = EINVAL; 1116 break; 1117 1118 default: 1119 error = ether_ioctl(ifp, cmd, data); 1120 break; 1121 } 1122 return (error); 1123 } 1124 1125 static int 1126 lagg_ether_setmulti(struct lagg_softc *sc) 1127 { 1128 struct lagg_port *lp; 1129 1130 LAGG_WLOCK_ASSERT(sc); 1131 1132 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1133 /* First, remove any existing filter entries. */ 1134 lagg_ether_cmdmulti(lp, 0); 1135 /* copy all addresses from the lagg interface to the port */ 1136 lagg_ether_cmdmulti(lp, 1); 1137 } 1138 return (0); 1139 } 1140 1141 static int 1142 lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1143 { 1144 struct lagg_softc *sc = lp->lp_softc; 1145 struct ifnet *ifp = lp->lp_ifp; 1146 struct ifnet *scifp = sc->sc_ifp; 1147 struct lagg_mc *mc; 1148 struct ifmultiaddr *ifma, *rifma = NULL; 1149 struct sockaddr_dl sdl; 1150 int error; 1151 1152 LAGG_WLOCK_ASSERT(sc); 1153 1154 bzero((char *)&sdl, sizeof(sdl)); 1155 sdl.sdl_len = sizeof(sdl); 1156 sdl.sdl_family = AF_LINK; 1157 sdl.sdl_type = IFT_ETHER; 1158 sdl.sdl_alen = ETHER_ADDR_LEN; 1159 sdl.sdl_index = ifp->if_index; 1160 1161 if (set) { 1162 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1163 if (ifma->ifma_addr->sa_family != AF_LINK) 1164 continue; 1165 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 1166 LLADDR(&sdl), ETHER_ADDR_LEN); 1167 1168 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 1169 if (error) 1170 return (error); 1171 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1172 if (mc == NULL) 1173 return (ENOMEM); 1174 mc->mc_ifma = rifma; 1175 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1176 } 1177 } else { 1178 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1179 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1180 if_delmulti_ifma(mc->mc_ifma); 1181 free(mc, M_DEVBUF); 1182 } 1183 } 1184 return (0); 1185 } 1186 1187 /* Handle a ref counted flag that should be set on the lagg port as well */ 1188 static int 1189 lagg_setflag(struct lagg_port *lp, int flag, int status, 1190 int (*func)(struct ifnet *, int)) 1191 { 1192 struct lagg_softc *sc = lp->lp_softc; 1193 struct ifnet *scifp = sc->sc_ifp; 1194 struct ifnet *ifp = lp->lp_ifp; 1195 int error; 1196 1197 LAGG_WLOCK_ASSERT(sc); 1198 1199 status = status ? (scifp->if_flags & flag) : 0; 1200 /* Now "status" contains the flag value or 0 */ 1201 1202 /* 1203 * See if recorded ports status is different from what 1204 * we want it to be. If it is, flip it. We record ports 1205 * status in lp_ifflags so that we won't clear ports flag 1206 * we haven't set. In fact, we don't clear or set ports 1207 * flags directly, but get or release references to them. 1208 * That's why we can be sure that recorded flags still are 1209 * in accord with actual ports flags. 1210 */ 1211 if (status != (lp->lp_ifflags & flag)) { 1212 error = (*func)(ifp, status); 1213 if (error) 1214 return (error); 1215 lp->lp_ifflags &= ~flag; 1216 lp->lp_ifflags |= status; 1217 } 1218 return (0); 1219 } 1220 1221 /* 1222 * Handle IFF_* flags that require certain changes on the lagg port 1223 * if "status" is true, update ports flags respective to the lagg 1224 * if "status" is false, forcedly clear the flags set on port. 1225 */ 1226 static int 1227 lagg_setflags(struct lagg_port *lp, int status) 1228 { 1229 int error, i; 1230 1231 for (i = 0; lagg_pflags[i].flag; i++) { 1232 error = lagg_setflag(lp, lagg_pflags[i].flag, 1233 status, lagg_pflags[i].func); 1234 if (error) 1235 return (error); 1236 } 1237 return (0); 1238 } 1239 1240 static int 1241 lagg_transmit(struct ifnet *ifp, struct mbuf *m) 1242 { 1243 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1244 int error, len, mcast; 1245 1246 len = m->m_pkthdr.len; 1247 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1248 1249 LAGG_RLOCK(sc); 1250 /* We need a Tx algorithm and at least one port */ 1251 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1252 LAGG_RUNLOCK(sc); 1253 m_freem(m); 1254 ifp->if_oerrors++; 1255 return (ENXIO); 1256 } 1257 1258 ETHER_BPF_MTAP(ifp, m); 1259 1260 error = (*sc->sc_start)(sc, m); 1261 LAGG_RUNLOCK(sc); 1262 1263 if (error == 0) { 1264 counter_u64_add(sc->sc_opackets, 1); 1265 counter_u64_add(sc->sc_obytes, len); 1266 ifp->if_omcasts += mcast; 1267 } else 1268 ifp->if_oerrors++; 1269 1270 return (error); 1271 } 1272 1273 /* 1274 * The ifp->if_qflush entry point for lagg(4) is no-op. 1275 */ 1276 static void 1277 lagg_qflush(struct ifnet *ifp __unused) 1278 { 1279 } 1280 1281 static struct mbuf * 1282 lagg_input(struct ifnet *ifp, struct mbuf *m) 1283 { 1284 struct lagg_port *lp = ifp->if_lagg; 1285 struct lagg_softc *sc = lp->lp_softc; 1286 struct ifnet *scifp = sc->sc_ifp; 1287 1288 LAGG_RLOCK(sc); 1289 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1290 (lp->lp_flags & LAGG_PORT_DISABLED) || 1291 sc->sc_proto == LAGG_PROTO_NONE) { 1292 LAGG_RUNLOCK(sc); 1293 m_freem(m); 1294 return (NULL); 1295 } 1296 1297 ETHER_BPF_MTAP(scifp, m); 1298 1299 m = (*sc->sc_input)(sc, lp, m); 1300 1301 if (m != NULL) { 1302 counter_u64_add(sc->sc_ipackets, 1); 1303 counter_u64_add(sc->sc_ibytes, m->m_pkthdr.len); 1304 1305 if (scifp->if_flags & IFF_MONITOR) { 1306 m_freem(m); 1307 m = NULL; 1308 } 1309 } 1310 1311 LAGG_RUNLOCK(sc); 1312 return (m); 1313 } 1314 1315 static int 1316 lagg_media_change(struct ifnet *ifp) 1317 { 1318 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1319 1320 if (sc->sc_ifflags & IFF_DEBUG) 1321 printf("%s\n", __func__); 1322 1323 /* Ignore */ 1324 return (0); 1325 } 1326 1327 static void 1328 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1329 { 1330 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1331 struct lagg_port *lp; 1332 1333 imr->ifm_status = IFM_AVALID; 1334 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1335 1336 LAGG_RLOCK(sc); 1337 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1338 if (LAGG_PORTACTIVE(lp)) 1339 imr->ifm_status |= IFM_ACTIVE; 1340 } 1341 LAGG_RUNLOCK(sc); 1342 } 1343 1344 static void 1345 lagg_linkstate(struct lagg_softc *sc) 1346 { 1347 struct lagg_port *lp; 1348 int new_link = LINK_STATE_DOWN; 1349 uint64_t speed; 1350 1351 /* Our link is considered up if at least one of our ports is active */ 1352 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1353 if (lp->lp_link_state == LINK_STATE_UP) { 1354 new_link = LINK_STATE_UP; 1355 break; 1356 } 1357 } 1358 if_link_state_change(sc->sc_ifp, new_link); 1359 1360 /* Update if_baudrate to reflect the max possible speed */ 1361 switch (sc->sc_proto) { 1362 case LAGG_PROTO_FAILOVER: 1363 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1364 sc->sc_primary->lp_ifp->if_baudrate : 0; 1365 break; 1366 case LAGG_PROTO_ROUNDROBIN: 1367 case LAGG_PROTO_LOADBALANCE: 1368 case LAGG_PROTO_ETHERCHANNEL: 1369 speed = 0; 1370 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1371 speed += lp->lp_ifp->if_baudrate; 1372 sc->sc_ifp->if_baudrate = speed; 1373 break; 1374 case LAGG_PROTO_LACP: 1375 /* LACP updates if_baudrate itself */ 1376 break; 1377 } 1378 } 1379 1380 static void 1381 lagg_port_state(struct ifnet *ifp, int state) 1382 { 1383 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1384 struct lagg_softc *sc = NULL; 1385 1386 if (lp != NULL) 1387 sc = lp->lp_softc; 1388 if (sc == NULL) 1389 return; 1390 1391 LAGG_WLOCK(sc); 1392 lagg_linkstate(sc); 1393 if (sc->sc_linkstate != NULL) 1394 (*sc->sc_linkstate)(lp); 1395 LAGG_WUNLOCK(sc); 1396 } 1397 1398 struct lagg_port * 1399 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1400 { 1401 struct lagg_port *lp_next, *rval = NULL; 1402 // int new_link = LINK_STATE_DOWN; 1403 1404 LAGG_RLOCK_ASSERT(sc); 1405 /* 1406 * Search a port which reports an active link state. 1407 */ 1408 1409 if (lp == NULL) 1410 goto search; 1411 if (LAGG_PORTACTIVE(lp)) { 1412 rval = lp; 1413 goto found; 1414 } 1415 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1416 LAGG_PORTACTIVE(lp_next)) { 1417 rval = lp_next; 1418 goto found; 1419 } 1420 1421 search: 1422 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1423 if (LAGG_PORTACTIVE(lp_next)) { 1424 rval = lp_next; 1425 goto found; 1426 } 1427 } 1428 1429 found: 1430 if (rval != NULL) { 1431 /* 1432 * The IEEE 802.1D standard assumes that a lagg with 1433 * multiple ports is always full duplex. This is valid 1434 * for load sharing laggs and if at least two links 1435 * are active. Unfortunately, checking the latter would 1436 * be too expensive at this point. 1437 XXX 1438 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1439 (sc->sc_count > 1)) 1440 new_link = LINK_STATE_FULL_DUPLEX; 1441 else 1442 new_link = rval->lp_link_state; 1443 */ 1444 } 1445 1446 return (rval); 1447 } 1448 1449 static const void * 1450 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf) 1451 { 1452 if (m->m_pkthdr.len < (off + len)) { 1453 return (NULL); 1454 } else if (m->m_len < (off + len)) { 1455 m_copydata(m, off, len, buf); 1456 return (buf); 1457 } 1458 return (mtod(m, char *) + off); 1459 } 1460 1461 uint32_t 1462 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key) 1463 { 1464 uint16_t etype; 1465 uint32_t p = key; 1466 int off; 1467 struct ether_header *eh; 1468 const struct ether_vlan_header *vlan; 1469 #ifdef INET 1470 const struct ip *ip; 1471 const uint32_t *ports; 1472 int iphlen; 1473 #endif 1474 #ifdef INET6 1475 const struct ip6_hdr *ip6; 1476 uint32_t flow; 1477 #endif 1478 union { 1479 #ifdef INET 1480 struct ip ip; 1481 #endif 1482 #ifdef INET6 1483 struct ip6_hdr ip6; 1484 #endif 1485 struct ether_vlan_header vlan; 1486 uint32_t port; 1487 } buf; 1488 1489 1490 off = sizeof(*eh); 1491 if (m->m_len < off) 1492 goto out; 1493 eh = mtod(m, struct ether_header *); 1494 etype = ntohs(eh->ether_type); 1495 if (sc->sc_flags & LAGG_F_HASHL2) { 1496 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p); 1497 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p); 1498 } 1499 1500 /* Special handling for encapsulating VLAN frames */ 1501 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) { 1502 p = hash32_buf(&m->m_pkthdr.ether_vtag, 1503 sizeof(m->m_pkthdr.ether_vtag), p); 1504 } else if (etype == ETHERTYPE_VLAN) { 1505 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf); 1506 if (vlan == NULL) 1507 goto out; 1508 1509 if (sc->sc_flags & LAGG_F_HASHL2) 1510 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p); 1511 etype = ntohs(vlan->evl_proto); 1512 off += sizeof(*vlan) - sizeof(*eh); 1513 } 1514 1515 switch (etype) { 1516 #ifdef INET 1517 case ETHERTYPE_IP: 1518 ip = lagg_gethdr(m, off, sizeof(*ip), &buf); 1519 if (ip == NULL) 1520 goto out; 1521 1522 if (sc->sc_flags & LAGG_F_HASHL3) { 1523 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p); 1524 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p); 1525 } 1526 if (!(sc->sc_flags & LAGG_F_HASHL4)) 1527 break; 1528 switch (ip->ip_p) { 1529 case IPPROTO_TCP: 1530 case IPPROTO_UDP: 1531 case IPPROTO_SCTP: 1532 iphlen = ip->ip_hl << 2; 1533 if (iphlen < sizeof(*ip)) 1534 break; 1535 off += iphlen; 1536 ports = lagg_gethdr(m, off, sizeof(*ports), &buf); 1537 if (ports == NULL) 1538 break; 1539 p = hash32_buf(ports, sizeof(*ports), p); 1540 break; 1541 } 1542 break; 1543 #endif 1544 #ifdef INET6 1545 case ETHERTYPE_IPV6: 1546 if (!(sc->sc_flags & LAGG_F_HASHL3)) 1547 break; 1548 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf); 1549 if (ip6 == NULL) 1550 goto out; 1551 1552 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p); 1553 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p); 1554 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK; 1555 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */ 1556 break; 1557 #endif 1558 } 1559 out: 1560 return (p); 1561 } 1562 1563 int 1564 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1565 { 1566 1567 return (ifp->if_transmit)(ifp, m); 1568 } 1569 1570 /* 1571 * Simple round robin aggregation 1572 */ 1573 1574 static int 1575 lagg_rr_attach(struct lagg_softc *sc) 1576 { 1577 sc->sc_detach = lagg_rr_detach; 1578 sc->sc_start = lagg_rr_start; 1579 sc->sc_input = lagg_rr_input; 1580 sc->sc_port_create = NULL; 1581 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1582 sc->sc_seq = 0; 1583 1584 return (0); 1585 } 1586 1587 static int 1588 lagg_rr_detach(struct lagg_softc *sc) 1589 { 1590 return (0); 1591 } 1592 1593 static int 1594 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 1595 { 1596 struct lagg_port *lp; 1597 uint32_t p; 1598 1599 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1600 p %= sc->sc_count; 1601 lp = SLIST_FIRST(&sc->sc_ports); 1602 while (p--) 1603 lp = SLIST_NEXT(lp, lp_entries); 1604 1605 /* 1606 * Check the port's link state. This will return the next active 1607 * port if the link is down or the port is NULL. 1608 */ 1609 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1610 m_freem(m); 1611 return (ENETDOWN); 1612 } 1613 1614 /* Send mbuf */ 1615 return (lagg_enqueue(lp->lp_ifp, m)); 1616 } 1617 1618 static struct mbuf * 1619 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1620 { 1621 struct ifnet *ifp = sc->sc_ifp; 1622 1623 /* Just pass in the packet to our lagg device */ 1624 m->m_pkthdr.rcvif = ifp; 1625 1626 return (m); 1627 } 1628 1629 /* 1630 * Active failover 1631 */ 1632 1633 static int 1634 lagg_fail_attach(struct lagg_softc *sc) 1635 { 1636 sc->sc_detach = lagg_fail_detach; 1637 sc->sc_start = lagg_fail_start; 1638 sc->sc_input = lagg_fail_input; 1639 sc->sc_port_create = NULL; 1640 sc->sc_port_destroy = NULL; 1641 1642 return (0); 1643 } 1644 1645 static int 1646 lagg_fail_detach(struct lagg_softc *sc) 1647 { 1648 return (0); 1649 } 1650 1651 static int 1652 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 1653 { 1654 struct lagg_port *lp; 1655 1656 /* Use the master port if active or the next available port */ 1657 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 1658 m_freem(m); 1659 return (ENETDOWN); 1660 } 1661 1662 /* Send mbuf */ 1663 return (lagg_enqueue(lp->lp_ifp, m)); 1664 } 1665 1666 static struct mbuf * 1667 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1668 { 1669 struct ifnet *ifp = sc->sc_ifp; 1670 struct lagg_port *tmp_tp; 1671 1672 if (lp == sc->sc_primary || lagg_failover_rx_all) { 1673 m->m_pkthdr.rcvif = ifp; 1674 return (m); 1675 } 1676 1677 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1678 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1679 /* 1680 * If tmp_tp is null, we've recieved a packet when all 1681 * our links are down. Weird, but process it anyways. 1682 */ 1683 if ((tmp_tp == NULL || tmp_tp == lp)) { 1684 m->m_pkthdr.rcvif = ifp; 1685 return (m); 1686 } 1687 } 1688 1689 m_freem(m); 1690 return (NULL); 1691 } 1692 1693 /* 1694 * Loadbalancing 1695 */ 1696 1697 static int 1698 lagg_lb_attach(struct lagg_softc *sc) 1699 { 1700 struct lagg_port *lp; 1701 struct lagg_lb *lb; 1702 1703 if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb), 1704 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 1705 return (ENOMEM); 1706 1707 sc->sc_detach = lagg_lb_detach; 1708 sc->sc_start = lagg_lb_start; 1709 sc->sc_input = lagg_lb_input; 1710 sc->sc_port_create = lagg_lb_port_create; 1711 sc->sc_port_destroy = lagg_lb_port_destroy; 1712 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1713 1714 lb->lb_key = arc4random(); 1715 sc->sc_psc = (caddr_t)lb; 1716 1717 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1718 lagg_lb_port_create(lp); 1719 1720 return (0); 1721 } 1722 1723 static int 1724 lagg_lb_detach(struct lagg_softc *sc) 1725 { 1726 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1727 if (lb != NULL) 1728 free(lb, M_DEVBUF); 1729 return (0); 1730 } 1731 1732 static int 1733 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 1734 { 1735 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1736 struct lagg_port *lp_next; 1737 int i = 0; 1738 1739 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 1740 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1741 if (lp_next == lp) 1742 continue; 1743 if (i >= LAGG_MAX_PORTS) 1744 return (EINVAL); 1745 if (sc->sc_ifflags & IFF_DEBUG) 1746 printf("%s: port %s at index %d\n", 1747 sc->sc_ifname, lp_next->lp_ifname, i); 1748 lb->lb_ports[i++] = lp_next; 1749 } 1750 1751 return (0); 1752 } 1753 1754 static int 1755 lagg_lb_port_create(struct lagg_port *lp) 1756 { 1757 struct lagg_softc *sc = lp->lp_softc; 1758 return (lagg_lb_porttable(sc, NULL)); 1759 } 1760 1761 static void 1762 lagg_lb_port_destroy(struct lagg_port *lp) 1763 { 1764 struct lagg_softc *sc = lp->lp_softc; 1765 lagg_lb_porttable(sc, lp); 1766 } 1767 1768 static int 1769 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 1770 { 1771 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1772 struct lagg_port *lp = NULL; 1773 uint32_t p = 0; 1774 1775 if (sc->use_flowid && (m->m_flags & M_FLOWID)) 1776 p = m->m_pkthdr.flowid; 1777 else 1778 p = lagg_hashmbuf(sc, m, lb->lb_key); 1779 p %= sc->sc_count; 1780 lp = lb->lb_ports[p]; 1781 1782 /* 1783 * Check the port's link state. This will return the next active 1784 * port if the link is down or the port is NULL. 1785 */ 1786 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1787 m_freem(m); 1788 return (ENETDOWN); 1789 } 1790 1791 /* Send mbuf */ 1792 return (lagg_enqueue(lp->lp_ifp, m)); 1793 } 1794 1795 static struct mbuf * 1796 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1797 { 1798 struct ifnet *ifp = sc->sc_ifp; 1799 1800 /* Just pass in the packet to our lagg device */ 1801 m->m_pkthdr.rcvif = ifp; 1802 1803 return (m); 1804 } 1805 1806 /* 1807 * 802.3ad LACP 1808 */ 1809 1810 static int 1811 lagg_lacp_attach(struct lagg_softc *sc) 1812 { 1813 struct lagg_port *lp; 1814 int error; 1815 1816 sc->sc_detach = lagg_lacp_detach; 1817 sc->sc_port_create = lacp_port_create; 1818 sc->sc_port_destroy = lacp_port_destroy; 1819 sc->sc_linkstate = lacp_linkstate; 1820 sc->sc_start = lagg_lacp_start; 1821 sc->sc_input = lagg_lacp_input; 1822 sc->sc_init = lacp_init; 1823 sc->sc_stop = lacp_stop; 1824 sc->sc_lladdr = lagg_lacp_lladdr; 1825 sc->sc_req = lacp_req; 1826 sc->sc_portreq = lacp_portreq; 1827 1828 error = lacp_attach(sc); 1829 if (error) 1830 return (error); 1831 1832 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1833 lacp_port_create(lp); 1834 1835 return (error); 1836 } 1837 1838 static int 1839 lagg_lacp_detach(struct lagg_softc *sc) 1840 { 1841 struct lagg_port *lp; 1842 int error; 1843 1844 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1845 lacp_port_destroy(lp); 1846 1847 /* unlocking is safe here */ 1848 LAGG_WUNLOCK(sc); 1849 error = lacp_detach(sc); 1850 LAGG_WLOCK(sc); 1851 1852 return (error); 1853 } 1854 1855 static void 1856 lagg_lacp_lladdr(struct lagg_softc *sc) 1857 { 1858 struct lagg_port *lp; 1859 1860 /* purge all the lacp ports */ 1861 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1862 lacp_port_destroy(lp); 1863 1864 /* add them back in */ 1865 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1866 lacp_port_create(lp); 1867 } 1868 1869 static int 1870 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 1871 { 1872 struct lagg_port *lp; 1873 1874 lp = lacp_select_tx_port(sc, m); 1875 if (lp == NULL) { 1876 m_freem(m); 1877 return (ENETDOWN); 1878 } 1879 1880 /* Send mbuf */ 1881 return (lagg_enqueue(lp->lp_ifp, m)); 1882 } 1883 1884 static struct mbuf * 1885 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1886 { 1887 struct ifnet *ifp = sc->sc_ifp; 1888 struct ether_header *eh; 1889 u_short etype; 1890 1891 eh = mtod(m, struct ether_header *); 1892 etype = ntohs(eh->ether_type); 1893 1894 /* Tap off LACP control messages */ 1895 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 1896 m = lacp_input(lp, m); 1897 if (m == NULL) 1898 return (NULL); 1899 } 1900 1901 /* 1902 * If the port is not collecting or not in the active aggregator then 1903 * free and return. 1904 */ 1905 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 1906 m_freem(m); 1907 return (NULL); 1908 } 1909 1910 m->m_pkthdr.rcvif = ifp; 1911 return (m); 1912 } 1913 1914 static void 1915 lagg_callout(void *arg) 1916 { 1917 struct lagg_softc *sc = (struct lagg_softc *)arg; 1918 struct ifnet *ifp = sc->sc_ifp; 1919 1920 ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets); 1921 ifp->if_opackets = counter_u64_fetch(sc->sc_opackets); 1922 ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes); 1923 ifp->if_obytes = counter_u64_fetch(sc->sc_obytes); 1924 1925 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 1926 } 1927