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/rmlock.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 #include <netinet6/in6_var.h> 67 #include <netinet6/in6_ifattach.h> 68 #endif 69 70 #include <net/if_vlan_var.h> 71 #include <net/if_lagg.h> 72 #include <net/ieee8023ad_lacp.h> 73 74 /* Special flags we should propagate to the lagg ports. */ 75 static struct { 76 int flag; 77 int (*func)(struct ifnet *, int); 78 } lagg_pflags[] = { 79 {IFF_PROMISC, ifpromisc}, 80 {IFF_ALLMULTI, if_allmulti}, 81 {0, NULL} 82 }; 83 84 SLIST_HEAD(__trhead, lagg_softc) lagg_list; /* list of laggs */ 85 static struct mtx lagg_list_mtx; 86 eventhandler_tag lagg_detach_cookie = NULL; 87 88 static int lagg_clone_create(struct if_clone *, int, caddr_t); 89 static void lagg_clone_destroy(struct ifnet *); 90 static struct if_clone *lagg_cloner; 91 static const char laggname[] = "lagg"; 92 93 static void lagg_lladdr(struct lagg_softc *, uint8_t *); 94 static void lagg_capabilities(struct lagg_softc *); 95 static void lagg_port_lladdr(struct lagg_port *, uint8_t *); 96 static void lagg_port_setlladdr(void *, int); 97 static int lagg_port_create(struct lagg_softc *, struct ifnet *); 98 static int lagg_port_destroy(struct lagg_port *, int); 99 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); 100 static void lagg_linkstate(struct lagg_softc *); 101 static void lagg_port_state(struct ifnet *, int); 102 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); 103 static int lagg_port_output(struct ifnet *, struct mbuf *, 104 const struct sockaddr *, struct route *); 105 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 106 #ifdef LAGG_PORT_STACKING 107 static int lagg_port_checkstacking(struct lagg_softc *); 108 #endif 109 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 110 static void lagg_init(void *); 111 static void lagg_stop(struct lagg_softc *); 112 static int lagg_ioctl(struct ifnet *, u_long, caddr_t); 113 static int lagg_ether_setmulti(struct lagg_softc *); 114 static int lagg_ether_cmdmulti(struct lagg_port *, int); 115 static int lagg_setflag(struct lagg_port *, int, int, 116 int (*func)(struct ifnet *, int)); 117 static int lagg_setflags(struct lagg_port *, int status); 118 static int lagg_transmit(struct ifnet *, struct mbuf *); 119 static void lagg_qflush(struct ifnet *); 120 static int lagg_media_change(struct ifnet *); 121 static void lagg_media_status(struct ifnet *, struct ifmediareq *); 122 static struct lagg_port *lagg_link_active(struct lagg_softc *, 123 struct lagg_port *); 124 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *); 125 static int lagg_sysctl_active(SYSCTL_HANDLER_ARGS); 126 127 /* Simple round robin */ 128 static int lagg_rr_attach(struct lagg_softc *); 129 static int lagg_rr_detach(struct lagg_softc *); 130 static int lagg_rr_start(struct lagg_softc *, struct mbuf *); 131 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 132 struct mbuf *); 133 134 /* Active failover */ 135 static int lagg_fail_attach(struct lagg_softc *); 136 static int lagg_fail_detach(struct lagg_softc *); 137 static int lagg_fail_start(struct lagg_softc *, struct mbuf *); 138 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 139 struct mbuf *); 140 141 /* Loadbalancing */ 142 static int lagg_lb_attach(struct lagg_softc *); 143 static int lagg_lb_detach(struct lagg_softc *); 144 static int lagg_lb_port_create(struct lagg_port *); 145 static void lagg_lb_port_destroy(struct lagg_port *); 146 static int lagg_lb_start(struct lagg_softc *, struct mbuf *); 147 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 148 struct mbuf *); 149 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 150 151 /* 802.3ad LACP */ 152 static int lagg_lacp_attach(struct lagg_softc *); 153 static int lagg_lacp_detach(struct lagg_softc *); 154 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); 155 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 156 struct mbuf *); 157 static void lagg_lacp_lladdr(struct lagg_softc *); 158 159 static void lagg_callout(void *); 160 161 /* lagg protocol table */ 162 static const struct { 163 int ti_proto; 164 int (*ti_attach)(struct lagg_softc *); 165 } lagg_protos[] = { 166 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach }, 167 { LAGG_PROTO_FAILOVER, lagg_fail_attach }, 168 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach }, 169 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach }, 170 { LAGG_PROTO_LACP, lagg_lacp_attach }, 171 { LAGG_PROTO_NONE, NULL } 172 }; 173 174 SYSCTL_DECL(_net_link); 175 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 176 "Link Aggregation"); 177 178 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */ 179 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW, 180 &lagg_failover_rx_all, 0, 181 "Accept input from any interface in a failover lagg"); 182 static int def_use_flowid = 1; /* Default value for using M_FLOWID */ 183 TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid); 184 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RW, 185 &def_use_flowid, 0, 186 "Default setting for using flow id for load sharing"); 187 static int def_flowid_shift = 16; /* Default value for using M_FLOWID */ 188 TUNABLE_INT("net.link.lagg.default_flowid_shift", &def_flowid_shift); 189 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RW, 190 &def_flowid_shift, 0, 191 "Default setting for flowid shift for load sharing"); 192 193 static int 194 lagg_modevent(module_t mod, int type, void *data) 195 { 196 197 switch (type) { 198 case MOD_LOAD: 199 mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF); 200 SLIST_INIT(&lagg_list); 201 lagg_cloner = if_clone_simple(laggname, lagg_clone_create, 202 lagg_clone_destroy, 0); 203 lagg_input_p = lagg_input; 204 lagg_linkstate_p = lagg_port_state; 205 lagg_detach_cookie = EVENTHANDLER_REGISTER( 206 ifnet_departure_event, lagg_port_ifdetach, NULL, 207 EVENTHANDLER_PRI_ANY); 208 break; 209 case MOD_UNLOAD: 210 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 211 lagg_detach_cookie); 212 if_clone_detach(lagg_cloner); 213 lagg_input_p = NULL; 214 lagg_linkstate_p = NULL; 215 mtx_destroy(&lagg_list_mtx); 216 break; 217 default: 218 return (EOPNOTSUPP); 219 } 220 return (0); 221 } 222 223 static moduledata_t lagg_mod = { 224 "if_lagg", 225 lagg_modevent, 226 0 227 }; 228 229 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 230 MODULE_VERSION(if_lagg, 1); 231 232 /* 233 * This routine is run via an vlan 234 * config EVENT 235 */ 236 static void 237 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 238 { 239 struct lagg_softc *sc = ifp->if_softc; 240 struct lagg_port *lp; 241 struct rm_priotracker tracker; 242 243 if (ifp->if_softc != arg) /* Not our event */ 244 return; 245 246 LAGG_RLOCK(sc, &tracker); 247 if (!SLIST_EMPTY(&sc->sc_ports)) { 248 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 249 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 250 } 251 LAGG_RUNLOCK(sc, &tracker); 252 } 253 254 /* 255 * This routine is run via an vlan 256 * unconfig EVENT 257 */ 258 static void 259 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 260 { 261 struct lagg_softc *sc = ifp->if_softc; 262 struct lagg_port *lp; 263 struct rm_priotracker tracker; 264 265 if (ifp->if_softc != arg) /* Not our event */ 266 return; 267 268 LAGG_RLOCK(sc, &tracker); 269 if (!SLIST_EMPTY(&sc->sc_ports)) { 270 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 271 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 272 } 273 LAGG_RUNLOCK(sc, &tracker); 274 } 275 276 static int 277 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) 278 { 279 struct lagg_softc *sc; 280 struct ifnet *ifp; 281 int i, error = 0; 282 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ 283 struct sysctl_oid *oid; 284 char num[14]; /* sufficient for 32 bits */ 285 286 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO); 287 ifp = sc->sc_ifp = if_alloc(IFT_ETHER); 288 if (ifp == NULL) { 289 free(sc, M_DEVBUF); 290 return (ENOSPC); 291 } 292 293 sc->sc_ipackets = counter_u64_alloc(M_WAITOK); 294 sc->sc_opackets = counter_u64_alloc(M_WAITOK); 295 sc->sc_ibytes = counter_u64_alloc(M_WAITOK); 296 sc->sc_obytes = counter_u64_alloc(M_WAITOK); 297 298 sysctl_ctx_init(&sc->ctx); 299 snprintf(num, sizeof(num), "%u", unit); 300 sc->use_flowid = def_use_flowid; 301 sc->flowid_shift = def_flowid_shift; 302 sc->sc_oid = oid = SYSCTL_ADD_NODE(&sc->ctx, 303 &SYSCTL_NODE_CHILDREN(_net_link, lagg), 304 OID_AUTO, num, CTLFLAG_RD, NULL, ""); 305 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 306 "use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid, 307 sc->use_flowid, "Use flow id for load sharing"); 308 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 309 "flowid_shift", CTLTYPE_INT|CTLFLAG_RW, &sc->flowid_shift, 310 sc->flowid_shift, 311 "Shift flowid bits to prevent multiqueue collisions"); 312 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 313 "count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count, 314 "Total number of ports"); 315 SYSCTL_ADD_PROC(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 316 "active", CTLTYPE_INT|CTLFLAG_RD, sc, 0, lagg_sysctl_active, 317 "I", "Total number of active ports"); 318 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 319 "flapping", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_flapping, 320 sc->sc_flapping, "Total number of port change events"); 321 /* Hash all layers by default */ 322 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4; 323 324 sc->sc_proto = LAGG_PROTO_NONE; 325 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) { 326 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) { 327 sc->sc_proto = lagg_protos[i].ti_proto; 328 if ((error = lagg_protos[i].ti_attach(sc)) != 0) { 329 if_free(ifp); 330 free(sc, M_DEVBUF); 331 return (error); 332 } 333 break; 334 } 335 } 336 LAGG_LOCK_INIT(sc); 337 LAGG_CALLOUT_LOCK_INIT(sc); 338 SLIST_INIT(&sc->sc_ports); 339 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc); 340 341 /* 342 * This uses the callout lock rather than the rmlock; one can't 343 * hold said rmlock during SWI. 344 */ 345 callout_init_mtx(&sc->sc_callout, &sc->sc_call_mtx, 0); 346 347 /* Initialise pseudo media types */ 348 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 349 lagg_media_status); 350 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 351 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 352 353 if_initname(ifp, laggname, unit); 354 ifp->if_softc = sc; 355 ifp->if_transmit = lagg_transmit; 356 ifp->if_qflush = lagg_qflush; 357 ifp->if_init = lagg_init; 358 ifp->if_ioctl = lagg_ioctl; 359 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 360 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS; 361 362 /* 363 * Attach as an ordinary ethernet device, children will be attached 364 * as special device IFT_IEEE8023ADLAG. 365 */ 366 ether_ifattach(ifp, eaddr); 367 368 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 369 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 370 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 371 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 372 373 /* Insert into the global list of laggs */ 374 mtx_lock(&lagg_list_mtx); 375 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries); 376 mtx_unlock(&lagg_list_mtx); 377 378 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 379 380 return (0); 381 } 382 383 static void 384 lagg_clone_destroy(struct ifnet *ifp) 385 { 386 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 387 struct lagg_port *lp; 388 389 LAGG_WLOCK(sc); 390 391 lagg_stop(sc); 392 ifp->if_flags &= ~IFF_UP; 393 394 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 395 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 396 397 /* Shutdown and remove lagg ports */ 398 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) 399 lagg_port_destroy(lp, 1); 400 /* Unhook the aggregation protocol */ 401 if (sc->sc_detach != NULL) 402 (*sc->sc_detach)(sc); 403 404 LAGG_WUNLOCK(sc); 405 406 sysctl_ctx_free(&sc->ctx); 407 ifmedia_removeall(&sc->sc_media); 408 ether_ifdetach(ifp); 409 if_free(ifp); 410 411 /* This grabs sc_callout_mtx, serialising it correctly */ 412 callout_drain(&sc->sc_callout); 413 414 /* At this point it's drained; we can free this */ 415 counter_u64_free(sc->sc_ipackets); 416 counter_u64_free(sc->sc_opackets); 417 counter_u64_free(sc->sc_ibytes); 418 counter_u64_free(sc->sc_obytes); 419 420 mtx_lock(&lagg_list_mtx); 421 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries); 422 mtx_unlock(&lagg_list_mtx); 423 424 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task); 425 LAGG_LOCK_DESTROY(sc); 426 LAGG_CALLOUT_LOCK_DESTROY(sc); 427 free(sc, M_DEVBUF); 428 } 429 430 static void 431 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr) 432 { 433 struct ifnet *ifp = sc->sc_ifp; 434 435 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 436 return; 437 438 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN); 439 /* Let the protocol know the MAC has changed */ 440 if (sc->sc_lladdr != NULL) 441 (*sc->sc_lladdr)(sc); 442 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 443 } 444 445 static void 446 lagg_capabilities(struct lagg_softc *sc) 447 { 448 struct lagg_port *lp; 449 int cap = ~0, ena = ~0; 450 u_long hwa = ~0UL; 451 452 LAGG_WLOCK_ASSERT(sc); 453 454 /* Get capabilities from the lagg ports */ 455 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 456 cap &= lp->lp_ifp->if_capabilities; 457 ena &= lp->lp_ifp->if_capenable; 458 hwa &= lp->lp_ifp->if_hwassist; 459 } 460 cap = (cap == ~0 ? 0 : cap); 461 ena = (ena == ~0 ? 0 : ena); 462 hwa = (hwa == ~0 ? 0 : hwa); 463 464 if (sc->sc_ifp->if_capabilities != cap || 465 sc->sc_ifp->if_capenable != ena || 466 sc->sc_ifp->if_hwassist != hwa) { 467 sc->sc_ifp->if_capabilities = cap; 468 sc->sc_ifp->if_capenable = ena; 469 sc->sc_ifp->if_hwassist = hwa; 470 getmicrotime(&sc->sc_ifp->if_lastchange); 471 472 if (sc->sc_ifflags & IFF_DEBUG) 473 if_printf(sc->sc_ifp, 474 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 475 } 476 } 477 478 static void 479 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr) 480 { 481 struct lagg_softc *sc = lp->lp_softc; 482 struct ifnet *ifp = lp->lp_ifp; 483 struct lagg_llq *llq; 484 int pending = 0; 485 486 LAGG_WLOCK_ASSERT(sc); 487 488 if (lp->lp_detaching || 489 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 490 return; 491 492 /* Check to make sure its not already queued to be changed */ 493 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 494 if (llq->llq_ifp == ifp) { 495 pending = 1; 496 break; 497 } 498 } 499 500 if (!pending) { 501 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT); 502 if (llq == NULL) /* XXX what to do */ 503 return; 504 } 505 506 /* Update the lladdr even if pending, it may have changed */ 507 llq->llq_ifp = ifp; 508 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN); 509 510 if (!pending) 511 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries); 512 513 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task); 514 } 515 516 /* 517 * Set the interface MAC address from a taskqueue to avoid a LOR. 518 */ 519 static void 520 lagg_port_setlladdr(void *arg, int pending) 521 { 522 struct lagg_softc *sc = (struct lagg_softc *)arg; 523 struct lagg_llq *llq, *head; 524 struct ifnet *ifp; 525 int error; 526 527 /* Grab a local reference of the queue and remove it from the softc */ 528 LAGG_WLOCK(sc); 529 head = SLIST_FIRST(&sc->sc_llq_head); 530 SLIST_FIRST(&sc->sc_llq_head) = NULL; 531 LAGG_WUNLOCK(sc); 532 533 /* 534 * Traverse the queue and set the lladdr on each ifp. It is safe to do 535 * unlocked as we have the only reference to it. 536 */ 537 for (llq = head; llq != NULL; llq = head) { 538 ifp = llq->llq_ifp; 539 540 /* Set the link layer address */ 541 CURVNET_SET(ifp->if_vnet); 542 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN); 543 CURVNET_RESTORE(); 544 if (error) 545 printf("%s: setlladdr failed on %s\n", __func__, 546 ifp->if_xname); 547 548 head = SLIST_NEXT(llq, llq_entries); 549 free(llq, M_DEVBUF); 550 } 551 } 552 553 static int 554 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 555 { 556 struct lagg_softc *sc_ptr; 557 struct lagg_port *lp; 558 int error = 0; 559 560 LAGG_WLOCK_ASSERT(sc); 561 562 /* Limit the maximal number of lagg ports */ 563 if (sc->sc_count >= LAGG_MAX_PORTS) 564 return (ENOSPC); 565 566 /* Check if port has already been associated to a lagg */ 567 if (ifp->if_lagg != NULL) { 568 /* Port is already in the current lagg? */ 569 lp = (struct lagg_port *)ifp->if_lagg; 570 if (lp->lp_softc == sc) 571 return (EEXIST); 572 return (EBUSY); 573 } 574 575 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 576 if (ifp->if_type != IFT_ETHER) 577 return (EPROTONOSUPPORT); 578 579 #ifdef INET6 580 /* 581 * The member interface should not have inet6 address because 582 * two interfaces with a valid link-local scope zone must not be 583 * merged in any form. This restriction is needed to 584 * prevent violation of link-local scope zone. Attempts to 585 * add a member interface which has inet6 addresses triggers 586 * removal of all inet6 addresses on the member interface. 587 */ 588 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 589 if (in6ifa_llaonifp(lp->lp_ifp)) { 590 in6_ifdetach(lp->lp_ifp); 591 if_printf(sc->sc_ifp, 592 "IPv6 addresses on %s have been removed " 593 "before adding it as a member to prevent " 594 "IPv6 address scope violation.\n", 595 lp->lp_ifp->if_xname); 596 } 597 } 598 if (in6ifa_llaonifp(ifp)) { 599 in6_ifdetach(ifp); 600 if_printf(sc->sc_ifp, 601 "IPv6 addresses on %s have been removed " 602 "before adding it as a member to prevent " 603 "IPv6 address scope violation.\n", 604 ifp->if_xname); 605 } 606 #endif 607 /* Allow the first Ethernet member to define the MTU */ 608 if (SLIST_EMPTY(&sc->sc_ports)) 609 sc->sc_ifp->if_mtu = ifp->if_mtu; 610 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 611 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 612 ifp->if_xname); 613 return (EINVAL); 614 } 615 616 if ((lp = malloc(sizeof(struct lagg_port), 617 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 618 return (ENOMEM); 619 620 /* Check if port is a stacked lagg */ 621 mtx_lock(&lagg_list_mtx); 622 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) { 623 if (ifp == sc_ptr->sc_ifp) { 624 mtx_unlock(&lagg_list_mtx); 625 free(lp, M_DEVBUF); 626 return (EINVAL); 627 /* XXX disable stacking for the moment, its untested */ 628 #ifdef LAGG_PORT_STACKING 629 lp->lp_flags |= LAGG_PORT_STACK; 630 if (lagg_port_checkstacking(sc_ptr) >= 631 LAGG_MAX_STACKING) { 632 mtx_unlock(&lagg_list_mtx); 633 free(lp, M_DEVBUF); 634 return (E2BIG); 635 } 636 #endif 637 } 638 } 639 mtx_unlock(&lagg_list_mtx); 640 641 /* Change the interface type */ 642 lp->lp_iftype = ifp->if_type; 643 ifp->if_type = IFT_IEEE8023ADLAG; 644 ifp->if_lagg = lp; 645 lp->lp_ioctl = ifp->if_ioctl; 646 ifp->if_ioctl = lagg_port_ioctl; 647 lp->lp_output = ifp->if_output; 648 ifp->if_output = lagg_port_output; 649 650 lp->lp_ifp = ifp; 651 lp->lp_softc = sc; 652 653 /* Save port link layer address */ 654 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 655 656 if (SLIST_EMPTY(&sc->sc_ports)) { 657 sc->sc_primary = lp; 658 lagg_lladdr(sc, IF_LLADDR(ifp)); 659 } else { 660 /* Update link layer address for this port */ 661 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp)); 662 } 663 664 /* Insert into the list of ports */ 665 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 666 sc->sc_count++; 667 668 /* Update lagg capabilities */ 669 lagg_capabilities(sc); 670 lagg_linkstate(sc); 671 672 /* Add multicast addresses and interface flags to this port */ 673 lagg_ether_cmdmulti(lp, 1); 674 lagg_setflags(lp, 1); 675 676 if (sc->sc_port_create != NULL) 677 error = (*sc->sc_port_create)(lp); 678 if (error) { 679 /* remove the port again, without calling sc_port_destroy */ 680 lagg_port_destroy(lp, 0); 681 return (error); 682 } 683 684 return (error); 685 } 686 687 #ifdef LAGG_PORT_STACKING 688 static int 689 lagg_port_checkstacking(struct lagg_softc *sc) 690 { 691 struct lagg_softc *sc_ptr; 692 struct lagg_port *lp; 693 int m = 0; 694 695 LAGG_WLOCK_ASSERT(sc); 696 697 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 698 if (lp->lp_flags & LAGG_PORT_STACK) { 699 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 700 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 701 } 702 } 703 704 return (m + 1); 705 } 706 #endif 707 708 static int 709 lagg_port_destroy(struct lagg_port *lp, int runpd) 710 { 711 struct lagg_softc *sc = lp->lp_softc; 712 struct lagg_port *lp_ptr; 713 struct lagg_llq *llq; 714 struct ifnet *ifp = lp->lp_ifp; 715 716 LAGG_WLOCK_ASSERT(sc); 717 718 if (runpd && sc->sc_port_destroy != NULL) 719 (*sc->sc_port_destroy)(lp); 720 721 /* 722 * Remove multicast addresses and interface flags from this port and 723 * reset the MAC address, skip if the interface is being detached. 724 */ 725 if (!lp->lp_detaching) { 726 lagg_ether_cmdmulti(lp, 0); 727 lagg_setflags(lp, 0); 728 lagg_port_lladdr(lp, lp->lp_lladdr); 729 } 730 731 /* Restore interface */ 732 ifp->if_type = lp->lp_iftype; 733 ifp->if_ioctl = lp->lp_ioctl; 734 ifp->if_output = lp->lp_output; 735 ifp->if_lagg = NULL; 736 737 /* Finally, remove the port from the lagg */ 738 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 739 sc->sc_count--; 740 741 /* Update the primary interface */ 742 if (lp == sc->sc_primary) { 743 uint8_t lladdr[ETHER_ADDR_LEN]; 744 745 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) { 746 bzero(&lladdr, ETHER_ADDR_LEN); 747 } else { 748 bcopy(lp_ptr->lp_lladdr, 749 lladdr, ETHER_ADDR_LEN); 750 } 751 lagg_lladdr(sc, lladdr); 752 sc->sc_primary = lp_ptr; 753 754 /* Update link layer address for each port */ 755 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 756 lagg_port_lladdr(lp_ptr, lladdr); 757 } 758 759 /* Remove any pending lladdr changes from the queue */ 760 if (lp->lp_detaching) { 761 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 762 if (llq->llq_ifp == ifp) { 763 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq, 764 llq_entries); 765 free(llq, M_DEVBUF); 766 break; /* Only appears once */ 767 } 768 } 769 } 770 771 if (lp->lp_ifflags) 772 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 773 774 free(lp, M_DEVBUF); 775 776 /* Update lagg capabilities */ 777 lagg_capabilities(sc); 778 lagg_linkstate(sc); 779 780 return (0); 781 } 782 783 static int 784 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 785 { 786 struct lagg_reqport *rp = (struct lagg_reqport *)data; 787 struct lagg_softc *sc; 788 struct lagg_port *lp = NULL; 789 int error = 0; 790 struct rm_priotracker tracker; 791 792 /* Should be checked by the caller */ 793 if (ifp->if_type != IFT_IEEE8023ADLAG || 794 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 795 goto fallback; 796 797 switch (cmd) { 798 case SIOCGLAGGPORT: 799 if (rp->rp_portname[0] == '\0' || 800 ifunit(rp->rp_portname) != ifp) { 801 error = EINVAL; 802 break; 803 } 804 805 LAGG_RLOCK(sc, &tracker); 806 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 807 error = ENOENT; 808 LAGG_RUNLOCK(sc, &tracker); 809 break; 810 } 811 812 lagg_port2req(lp, rp); 813 LAGG_RUNLOCK(sc, &tracker); 814 break; 815 816 case SIOCSIFCAP: 817 if (lp->lp_ioctl == NULL) { 818 error = EINVAL; 819 break; 820 } 821 error = (*lp->lp_ioctl)(ifp, cmd, data); 822 if (error) 823 break; 824 825 /* Update lagg interface capabilities */ 826 LAGG_WLOCK(sc); 827 lagg_capabilities(sc); 828 LAGG_WUNLOCK(sc); 829 break; 830 831 case SIOCSIFMTU: 832 /* Do not allow the MTU to be changed once joined */ 833 error = EINVAL; 834 break; 835 836 default: 837 goto fallback; 838 } 839 840 return (error); 841 842 fallback: 843 if (lp->lp_ioctl != NULL) 844 return ((*lp->lp_ioctl)(ifp, cmd, data)); 845 846 return (EINVAL); 847 } 848 849 /* 850 * For direct output to child ports. 851 */ 852 static int 853 lagg_port_output(struct ifnet *ifp, struct mbuf *m, 854 const struct sockaddr *dst, struct route *ro) 855 { 856 struct lagg_port *lp = ifp->if_lagg; 857 858 switch (dst->sa_family) { 859 case pseudo_AF_HDRCMPLT: 860 case AF_UNSPEC: 861 return ((*lp->lp_output)(ifp, m, dst, ro)); 862 } 863 864 /* drop any other frames */ 865 m_freem(m); 866 return (ENETDOWN); 867 } 868 869 static void 870 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 871 { 872 struct lagg_port *lp; 873 struct lagg_softc *sc; 874 875 if ((lp = ifp->if_lagg) == NULL) 876 return; 877 /* If the ifnet is just being renamed, don't do anything. */ 878 if (ifp->if_flags & IFF_RENAMING) 879 return; 880 881 sc = lp->lp_softc; 882 883 LAGG_WLOCK(sc); 884 lp->lp_detaching = 1; 885 lagg_port_destroy(lp, 1); 886 LAGG_WUNLOCK(sc); 887 } 888 889 static void 890 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 891 { 892 struct lagg_softc *sc = lp->lp_softc; 893 894 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 895 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 896 rp->rp_prio = lp->lp_prio; 897 rp->rp_flags = lp->lp_flags; 898 if (sc->sc_portreq != NULL) 899 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc); 900 901 /* Add protocol specific flags */ 902 switch (sc->sc_proto) { 903 case LAGG_PROTO_FAILOVER: 904 if (lp == sc->sc_primary) 905 rp->rp_flags |= LAGG_PORT_MASTER; 906 if (lp == lagg_link_active(sc, sc->sc_primary)) 907 rp->rp_flags |= LAGG_PORT_ACTIVE; 908 break; 909 910 case LAGG_PROTO_ROUNDROBIN: 911 case LAGG_PROTO_LOADBALANCE: 912 case LAGG_PROTO_ETHERCHANNEL: 913 if (LAGG_PORTACTIVE(lp)) 914 rp->rp_flags |= LAGG_PORT_ACTIVE; 915 break; 916 917 case LAGG_PROTO_LACP: 918 /* LACP has a different definition of active */ 919 if (lacp_isactive(lp)) 920 rp->rp_flags |= LAGG_PORT_ACTIVE; 921 if (lacp_iscollecting(lp)) 922 rp->rp_flags |= LAGG_PORT_COLLECTING; 923 if (lacp_isdistributing(lp)) 924 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 925 break; 926 } 927 928 } 929 930 static void 931 lagg_init(void *xsc) 932 { 933 struct lagg_softc *sc = (struct lagg_softc *)xsc; 934 struct lagg_port *lp; 935 struct ifnet *ifp = sc->sc_ifp; 936 937 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 938 return; 939 940 LAGG_WLOCK(sc); 941 942 ifp->if_drv_flags |= IFF_DRV_RUNNING; 943 /* Update the port lladdrs */ 944 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 945 lagg_port_lladdr(lp, IF_LLADDR(ifp)); 946 947 if (sc->sc_init != NULL) 948 (*sc->sc_init)(sc); 949 950 LAGG_WUNLOCK(sc); 951 } 952 953 static void 954 lagg_stop(struct lagg_softc *sc) 955 { 956 struct ifnet *ifp = sc->sc_ifp; 957 958 LAGG_WLOCK_ASSERT(sc); 959 960 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 961 return; 962 963 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 964 965 if (sc->sc_stop != NULL) 966 (*sc->sc_stop)(sc); 967 } 968 969 static int 970 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 971 { 972 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 973 struct lagg_reqall *ra = (struct lagg_reqall *)data; 974 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 975 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 976 struct ifreq *ifr = (struct ifreq *)data; 977 struct lagg_port *lp; 978 struct ifnet *tpif; 979 struct thread *td = curthread; 980 char *buf, *outbuf; 981 int count, buflen, len, error = 0; 982 struct rm_priotracker tracker; 983 984 bzero(&rpbuf, sizeof(rpbuf)); 985 986 switch (cmd) { 987 case SIOCGLAGG: 988 LAGG_RLOCK(sc, &tracker); 989 count = 0; 990 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 991 count++; 992 buflen = count * sizeof(struct lagg_reqport); 993 LAGG_RUNLOCK(sc, &tracker); 994 995 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 996 997 LAGG_RLOCK(sc, &tracker); 998 ra->ra_proto = sc->sc_proto; 999 if (sc->sc_req != NULL) 1000 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc); 1001 1002 count = 0; 1003 buf = outbuf; 1004 len = min(ra->ra_size, buflen); 1005 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1006 if (len < sizeof(rpbuf)) 1007 break; 1008 1009 lagg_port2req(lp, &rpbuf); 1010 memcpy(buf, &rpbuf, sizeof(rpbuf)); 1011 count++; 1012 buf += sizeof(rpbuf); 1013 len -= sizeof(rpbuf); 1014 } 1015 LAGG_RUNLOCK(sc, &tracker); 1016 ra->ra_ports = count; 1017 ra->ra_size = count * sizeof(rpbuf); 1018 error = copyout(outbuf, ra->ra_port, ra->ra_size); 1019 free(outbuf, M_TEMP); 1020 break; 1021 case SIOCSLAGG: 1022 error = priv_check(td, PRIV_NET_LAGG); 1023 if (error) 1024 break; 1025 if (ra->ra_proto >= LAGG_PROTO_MAX) { 1026 error = EPROTONOSUPPORT; 1027 break; 1028 } 1029 LAGG_WLOCK(sc); 1030 if (sc->sc_proto != LAGG_PROTO_NONE) { 1031 /* Reset protocol first in case detach unlocks */ 1032 sc->sc_proto = LAGG_PROTO_NONE; 1033 error = sc->sc_detach(sc); 1034 sc->sc_detach = NULL; 1035 sc->sc_start = NULL; 1036 sc->sc_input = NULL; 1037 sc->sc_port_create = NULL; 1038 sc->sc_port_destroy = NULL; 1039 sc->sc_linkstate = NULL; 1040 sc->sc_init = NULL; 1041 sc->sc_stop = NULL; 1042 sc->sc_lladdr = NULL; 1043 sc->sc_req = NULL; 1044 sc->sc_portreq = NULL; 1045 } else if (sc->sc_input != NULL) { 1046 /* Still detaching */ 1047 error = EBUSY; 1048 } 1049 if (error != 0) { 1050 LAGG_WUNLOCK(sc); 1051 break; 1052 } 1053 for (int i = 0; i < (sizeof(lagg_protos) / 1054 sizeof(lagg_protos[0])); i++) { 1055 if (lagg_protos[i].ti_proto == ra->ra_proto) { 1056 if (sc->sc_ifflags & IFF_DEBUG) 1057 printf("%s: using proto %u\n", 1058 sc->sc_ifname, 1059 lagg_protos[i].ti_proto); 1060 sc->sc_proto = lagg_protos[i].ti_proto; 1061 if (sc->sc_proto != LAGG_PROTO_NONE) 1062 error = lagg_protos[i].ti_attach(sc); 1063 LAGG_WUNLOCK(sc); 1064 return (error); 1065 } 1066 } 1067 LAGG_WUNLOCK(sc); 1068 error = EPROTONOSUPPORT; 1069 break; 1070 case SIOCGLAGGFLAGS: 1071 rf->rf_flags = sc->sc_flags; 1072 break; 1073 case SIOCSLAGGHASH: 1074 error = priv_check(td, PRIV_NET_LAGG); 1075 if (error) 1076 break; 1077 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1078 error = EINVAL; 1079 break; 1080 } 1081 LAGG_WLOCK(sc); 1082 sc->sc_flags &= ~LAGG_F_HASHMASK; 1083 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK; 1084 LAGG_WUNLOCK(sc); 1085 break; 1086 case SIOCGLAGGPORT: 1087 if (rp->rp_portname[0] == '\0' || 1088 (tpif = ifunit(rp->rp_portname)) == NULL) { 1089 error = EINVAL; 1090 break; 1091 } 1092 1093 LAGG_RLOCK(sc, &tracker); 1094 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1095 lp->lp_softc != sc) { 1096 error = ENOENT; 1097 LAGG_RUNLOCK(sc, &tracker); 1098 break; 1099 } 1100 1101 lagg_port2req(lp, rp); 1102 LAGG_RUNLOCK(sc, &tracker); 1103 break; 1104 case SIOCSLAGGPORT: 1105 error = priv_check(td, PRIV_NET_LAGG); 1106 if (error) 1107 break; 1108 if (rp->rp_portname[0] == '\0' || 1109 (tpif = ifunit(rp->rp_portname)) == NULL) { 1110 error = EINVAL; 1111 break; 1112 } 1113 LAGG_WLOCK(sc); 1114 error = lagg_port_create(sc, tpif); 1115 LAGG_WUNLOCK(sc); 1116 break; 1117 case SIOCSLAGGDELPORT: 1118 error = priv_check(td, PRIV_NET_LAGG); 1119 if (error) 1120 break; 1121 if (rp->rp_portname[0] == '\0' || 1122 (tpif = ifunit(rp->rp_portname)) == NULL) { 1123 error = EINVAL; 1124 break; 1125 } 1126 1127 LAGG_WLOCK(sc); 1128 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1129 lp->lp_softc != sc) { 1130 error = ENOENT; 1131 LAGG_WUNLOCK(sc); 1132 break; 1133 } 1134 1135 error = lagg_port_destroy(lp, 1); 1136 LAGG_WUNLOCK(sc); 1137 break; 1138 case SIOCSIFFLAGS: 1139 /* Set flags on ports too */ 1140 LAGG_WLOCK(sc); 1141 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1142 lagg_setflags(lp, 1); 1143 } 1144 LAGG_WUNLOCK(sc); 1145 1146 if (!(ifp->if_flags & IFF_UP) && 1147 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1148 /* 1149 * If interface is marked down and it is running, 1150 * then stop and disable it. 1151 */ 1152 LAGG_WLOCK(sc); 1153 lagg_stop(sc); 1154 LAGG_WUNLOCK(sc); 1155 } else if ((ifp->if_flags & IFF_UP) && 1156 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1157 /* 1158 * If interface is marked up and it is stopped, then 1159 * start it. 1160 */ 1161 (*ifp->if_init)(sc); 1162 } 1163 break; 1164 case SIOCADDMULTI: 1165 case SIOCDELMULTI: 1166 LAGG_WLOCK(sc); 1167 error = lagg_ether_setmulti(sc); 1168 LAGG_WUNLOCK(sc); 1169 break; 1170 case SIOCSIFMEDIA: 1171 case SIOCGIFMEDIA: 1172 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1173 break; 1174 1175 case SIOCSIFCAP: 1176 case SIOCSIFMTU: 1177 /* Do not allow the MTU or caps to be directly changed */ 1178 error = EINVAL; 1179 break; 1180 1181 default: 1182 error = ether_ioctl(ifp, cmd, data); 1183 break; 1184 } 1185 return (error); 1186 } 1187 1188 static int 1189 lagg_ether_setmulti(struct lagg_softc *sc) 1190 { 1191 struct lagg_port *lp; 1192 1193 LAGG_WLOCK_ASSERT(sc); 1194 1195 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1196 /* First, remove any existing filter entries. */ 1197 lagg_ether_cmdmulti(lp, 0); 1198 /* copy all addresses from the lagg interface to the port */ 1199 lagg_ether_cmdmulti(lp, 1); 1200 } 1201 return (0); 1202 } 1203 1204 static int 1205 lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1206 { 1207 struct lagg_softc *sc = lp->lp_softc; 1208 struct ifnet *ifp = lp->lp_ifp; 1209 struct ifnet *scifp = sc->sc_ifp; 1210 struct lagg_mc *mc; 1211 struct ifmultiaddr *ifma, *rifma = NULL; 1212 struct sockaddr_dl sdl; 1213 int error; 1214 1215 LAGG_WLOCK_ASSERT(sc); 1216 1217 link_init_sdl(ifp, (struct sockaddr *)&sdl, IFT_ETHER); 1218 sdl.sdl_alen = ETHER_ADDR_LEN; 1219 1220 if (set) { 1221 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1222 if (ifma->ifma_addr->sa_family != AF_LINK) 1223 continue; 1224 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 1225 LLADDR(&sdl), ETHER_ADDR_LEN); 1226 1227 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 1228 if (error) 1229 return (error); 1230 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1231 if (mc == NULL) 1232 return (ENOMEM); 1233 mc->mc_ifma = rifma; 1234 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1235 } 1236 } else { 1237 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1238 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1239 if_delmulti_ifma(mc->mc_ifma); 1240 free(mc, M_DEVBUF); 1241 } 1242 } 1243 return (0); 1244 } 1245 1246 /* Handle a ref counted flag that should be set on the lagg port as well */ 1247 static int 1248 lagg_setflag(struct lagg_port *lp, int flag, int status, 1249 int (*func)(struct ifnet *, int)) 1250 { 1251 struct lagg_softc *sc = lp->lp_softc; 1252 struct ifnet *scifp = sc->sc_ifp; 1253 struct ifnet *ifp = lp->lp_ifp; 1254 int error; 1255 1256 LAGG_WLOCK_ASSERT(sc); 1257 1258 status = status ? (scifp->if_flags & flag) : 0; 1259 /* Now "status" contains the flag value or 0 */ 1260 1261 /* 1262 * See if recorded ports status is different from what 1263 * we want it to be. If it is, flip it. We record ports 1264 * status in lp_ifflags so that we won't clear ports flag 1265 * we haven't set. In fact, we don't clear or set ports 1266 * flags directly, but get or release references to them. 1267 * That's why we can be sure that recorded flags still are 1268 * in accord with actual ports flags. 1269 */ 1270 if (status != (lp->lp_ifflags & flag)) { 1271 error = (*func)(ifp, status); 1272 if (error) 1273 return (error); 1274 lp->lp_ifflags &= ~flag; 1275 lp->lp_ifflags |= status; 1276 } 1277 return (0); 1278 } 1279 1280 /* 1281 * Handle IFF_* flags that require certain changes on the lagg port 1282 * if "status" is true, update ports flags respective to the lagg 1283 * if "status" is false, forcedly clear the flags set on port. 1284 */ 1285 static int 1286 lagg_setflags(struct lagg_port *lp, int status) 1287 { 1288 int error, i; 1289 1290 for (i = 0; lagg_pflags[i].flag; i++) { 1291 error = lagg_setflag(lp, lagg_pflags[i].flag, 1292 status, lagg_pflags[i].func); 1293 if (error) 1294 return (error); 1295 } 1296 return (0); 1297 } 1298 1299 static int 1300 lagg_transmit(struct ifnet *ifp, struct mbuf *m) 1301 { 1302 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1303 int error, len, mcast; 1304 struct rm_priotracker tracker; 1305 1306 len = m->m_pkthdr.len; 1307 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1308 1309 LAGG_RLOCK(sc, &tracker); 1310 /* We need a Tx algorithm and at least one port */ 1311 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1312 LAGG_RUNLOCK(sc, &tracker); 1313 m_freem(m); 1314 ifp->if_oerrors++; 1315 return (ENXIO); 1316 } 1317 1318 ETHER_BPF_MTAP(ifp, m); 1319 1320 error = (*sc->sc_start)(sc, m); 1321 LAGG_RUNLOCK(sc, &tracker); 1322 1323 if (error == 0) { 1324 counter_u64_add(sc->sc_opackets, 1); 1325 counter_u64_add(sc->sc_obytes, len); 1326 ifp->if_omcasts += mcast; 1327 } else 1328 ifp->if_oerrors++; 1329 1330 return (error); 1331 } 1332 1333 /* 1334 * The ifp->if_qflush entry point for lagg(4) is no-op. 1335 */ 1336 static void 1337 lagg_qflush(struct ifnet *ifp __unused) 1338 { 1339 } 1340 1341 static struct mbuf * 1342 lagg_input(struct ifnet *ifp, struct mbuf *m) 1343 { 1344 struct lagg_port *lp = ifp->if_lagg; 1345 struct lagg_softc *sc = lp->lp_softc; 1346 struct ifnet *scifp = sc->sc_ifp; 1347 struct rm_priotracker tracker; 1348 1349 LAGG_RLOCK(sc, &tracker); 1350 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1351 (lp->lp_flags & LAGG_PORT_DISABLED) || 1352 sc->sc_proto == LAGG_PROTO_NONE) { 1353 LAGG_RUNLOCK(sc, &tracker); 1354 m_freem(m); 1355 return (NULL); 1356 } 1357 1358 ETHER_BPF_MTAP(scifp, m); 1359 1360 m = (*sc->sc_input)(sc, lp, m); 1361 1362 if (m != NULL) { 1363 counter_u64_add(sc->sc_ipackets, 1); 1364 counter_u64_add(sc->sc_ibytes, m->m_pkthdr.len); 1365 1366 if (scifp->if_flags & IFF_MONITOR) { 1367 m_freem(m); 1368 m = NULL; 1369 } 1370 } 1371 1372 LAGG_RUNLOCK(sc, &tracker); 1373 return (m); 1374 } 1375 1376 static int 1377 lagg_media_change(struct ifnet *ifp) 1378 { 1379 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1380 1381 if (sc->sc_ifflags & IFF_DEBUG) 1382 printf("%s\n", __func__); 1383 1384 /* Ignore */ 1385 return (0); 1386 } 1387 1388 static void 1389 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1390 { 1391 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1392 struct lagg_port *lp; 1393 struct rm_priotracker tracker; 1394 1395 imr->ifm_status = IFM_AVALID; 1396 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1397 1398 LAGG_RLOCK(sc, &tracker); 1399 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1400 if (LAGG_PORTACTIVE(lp)) 1401 imr->ifm_status |= IFM_ACTIVE; 1402 } 1403 LAGG_RUNLOCK(sc, &tracker); 1404 } 1405 1406 static void 1407 lagg_linkstate(struct lagg_softc *sc) 1408 { 1409 struct lagg_port *lp; 1410 int new_link = LINK_STATE_DOWN; 1411 uint64_t speed; 1412 1413 /* Our link is considered up if at least one of our ports is active */ 1414 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1415 if (lp->lp_link_state == LINK_STATE_UP) { 1416 new_link = LINK_STATE_UP; 1417 break; 1418 } 1419 } 1420 if_link_state_change(sc->sc_ifp, new_link); 1421 1422 /* Update if_baudrate to reflect the max possible speed */ 1423 switch (sc->sc_proto) { 1424 case LAGG_PROTO_FAILOVER: 1425 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1426 sc->sc_primary->lp_ifp->if_baudrate : 0; 1427 break; 1428 case LAGG_PROTO_ROUNDROBIN: 1429 case LAGG_PROTO_LOADBALANCE: 1430 case LAGG_PROTO_ETHERCHANNEL: 1431 speed = 0; 1432 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1433 speed += lp->lp_ifp->if_baudrate; 1434 sc->sc_ifp->if_baudrate = speed; 1435 break; 1436 case LAGG_PROTO_LACP: 1437 /* LACP updates if_baudrate itself */ 1438 break; 1439 } 1440 } 1441 1442 static void 1443 lagg_port_state(struct ifnet *ifp, int state) 1444 { 1445 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1446 struct lagg_softc *sc = NULL; 1447 1448 if (lp != NULL) 1449 sc = lp->lp_softc; 1450 if (sc == NULL) 1451 return; 1452 1453 LAGG_WLOCK(sc); 1454 lagg_linkstate(sc); 1455 if (sc->sc_linkstate != NULL) 1456 (*sc->sc_linkstate)(lp); 1457 LAGG_WUNLOCK(sc); 1458 } 1459 1460 struct lagg_port * 1461 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1462 { 1463 struct lagg_port *lp_next, *rval = NULL; 1464 // int new_link = LINK_STATE_DOWN; 1465 1466 LAGG_RLOCK_ASSERT(sc); 1467 /* 1468 * Search a port which reports an active link state. 1469 */ 1470 1471 if (lp == NULL) 1472 goto search; 1473 if (LAGG_PORTACTIVE(lp)) { 1474 rval = lp; 1475 goto found; 1476 } 1477 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1478 LAGG_PORTACTIVE(lp_next)) { 1479 rval = lp_next; 1480 goto found; 1481 } 1482 1483 search: 1484 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1485 if (LAGG_PORTACTIVE(lp_next)) { 1486 rval = lp_next; 1487 goto found; 1488 } 1489 } 1490 1491 found: 1492 if (rval != NULL) { 1493 /* 1494 * The IEEE 802.1D standard assumes that a lagg with 1495 * multiple ports is always full duplex. This is valid 1496 * for load sharing laggs and if at least two links 1497 * are active. Unfortunately, checking the latter would 1498 * be too expensive at this point. 1499 XXX 1500 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1501 (sc->sc_count > 1)) 1502 new_link = LINK_STATE_FULL_DUPLEX; 1503 else 1504 new_link = rval->lp_link_state; 1505 */ 1506 } 1507 1508 return (rval); 1509 } 1510 1511 static const void * 1512 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf) 1513 { 1514 if (m->m_pkthdr.len < (off + len)) { 1515 return (NULL); 1516 } else if (m->m_len < (off + len)) { 1517 m_copydata(m, off, len, buf); 1518 return (buf); 1519 } 1520 return (mtod(m, char *) + off); 1521 } 1522 1523 static int 1524 lagg_sysctl_active(SYSCTL_HANDLER_ARGS) 1525 { 1526 struct lagg_softc *sc = (struct lagg_softc *)arg1; 1527 struct lagg_port *lp; 1528 int error; 1529 1530 /* LACP tracks active links automatically, the others do not */ 1531 if (sc->sc_proto != LAGG_PROTO_LACP) { 1532 sc->sc_active = 0; 1533 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1534 sc->sc_active += LAGG_PORTACTIVE(lp); 1535 } 1536 1537 error = sysctl_handle_int(oidp, &sc->sc_active, 0, req); 1538 if ((error) || (req->newptr == NULL)) 1539 return (error); 1540 1541 return (0); 1542 } 1543 1544 uint32_t 1545 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key) 1546 { 1547 uint16_t etype; 1548 uint32_t p = key; 1549 int off; 1550 struct ether_header *eh; 1551 const struct ether_vlan_header *vlan; 1552 #ifdef INET 1553 const struct ip *ip; 1554 const uint32_t *ports; 1555 int iphlen; 1556 #endif 1557 #ifdef INET6 1558 const struct ip6_hdr *ip6; 1559 uint32_t flow; 1560 #endif 1561 union { 1562 #ifdef INET 1563 struct ip ip; 1564 #endif 1565 #ifdef INET6 1566 struct ip6_hdr ip6; 1567 #endif 1568 struct ether_vlan_header vlan; 1569 uint32_t port; 1570 } buf; 1571 1572 1573 off = sizeof(*eh); 1574 if (m->m_len < off) 1575 goto out; 1576 eh = mtod(m, struct ether_header *); 1577 etype = ntohs(eh->ether_type); 1578 if (sc->sc_flags & LAGG_F_HASHL2) { 1579 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p); 1580 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p); 1581 } 1582 1583 /* Special handling for encapsulating VLAN frames */ 1584 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) { 1585 p = hash32_buf(&m->m_pkthdr.ether_vtag, 1586 sizeof(m->m_pkthdr.ether_vtag), p); 1587 } else if (etype == ETHERTYPE_VLAN) { 1588 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf); 1589 if (vlan == NULL) 1590 goto out; 1591 1592 if (sc->sc_flags & LAGG_F_HASHL2) 1593 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p); 1594 etype = ntohs(vlan->evl_proto); 1595 off += sizeof(*vlan) - sizeof(*eh); 1596 } 1597 1598 switch (etype) { 1599 #ifdef INET 1600 case ETHERTYPE_IP: 1601 ip = lagg_gethdr(m, off, sizeof(*ip), &buf); 1602 if (ip == NULL) 1603 goto out; 1604 1605 if (sc->sc_flags & LAGG_F_HASHL3) { 1606 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p); 1607 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p); 1608 } 1609 if (!(sc->sc_flags & LAGG_F_HASHL4)) 1610 break; 1611 switch (ip->ip_p) { 1612 case IPPROTO_TCP: 1613 case IPPROTO_UDP: 1614 case IPPROTO_SCTP: 1615 iphlen = ip->ip_hl << 2; 1616 if (iphlen < sizeof(*ip)) 1617 break; 1618 off += iphlen; 1619 ports = lagg_gethdr(m, off, sizeof(*ports), &buf); 1620 if (ports == NULL) 1621 break; 1622 p = hash32_buf(ports, sizeof(*ports), p); 1623 break; 1624 } 1625 break; 1626 #endif 1627 #ifdef INET6 1628 case ETHERTYPE_IPV6: 1629 if (!(sc->sc_flags & LAGG_F_HASHL3)) 1630 break; 1631 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf); 1632 if (ip6 == NULL) 1633 goto out; 1634 1635 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p); 1636 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p); 1637 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK; 1638 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */ 1639 break; 1640 #endif 1641 } 1642 out: 1643 return (p); 1644 } 1645 1646 int 1647 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1648 { 1649 1650 return (ifp->if_transmit)(ifp, m); 1651 } 1652 1653 /* 1654 * Simple round robin aggregation 1655 */ 1656 1657 static int 1658 lagg_rr_attach(struct lagg_softc *sc) 1659 { 1660 sc->sc_detach = lagg_rr_detach; 1661 sc->sc_start = lagg_rr_start; 1662 sc->sc_input = lagg_rr_input; 1663 sc->sc_port_create = NULL; 1664 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1665 sc->sc_seq = 0; 1666 1667 return (0); 1668 } 1669 1670 static int 1671 lagg_rr_detach(struct lagg_softc *sc) 1672 { 1673 return (0); 1674 } 1675 1676 static int 1677 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 1678 { 1679 struct lagg_port *lp; 1680 uint32_t p; 1681 1682 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1683 p %= sc->sc_count; 1684 lp = SLIST_FIRST(&sc->sc_ports); 1685 while (p--) 1686 lp = SLIST_NEXT(lp, lp_entries); 1687 1688 /* 1689 * Check the port's link state. This will return the next active 1690 * port if the link is down or the port is NULL. 1691 */ 1692 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1693 m_freem(m); 1694 return (ENETDOWN); 1695 } 1696 1697 /* Send mbuf */ 1698 return (lagg_enqueue(lp->lp_ifp, m)); 1699 } 1700 1701 static struct mbuf * 1702 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1703 { 1704 struct ifnet *ifp = sc->sc_ifp; 1705 1706 /* Just pass in the packet to our lagg device */ 1707 m->m_pkthdr.rcvif = ifp; 1708 1709 return (m); 1710 } 1711 1712 /* 1713 * Active failover 1714 */ 1715 1716 static int 1717 lagg_fail_attach(struct lagg_softc *sc) 1718 { 1719 sc->sc_detach = lagg_fail_detach; 1720 sc->sc_start = lagg_fail_start; 1721 sc->sc_input = lagg_fail_input; 1722 sc->sc_port_create = NULL; 1723 sc->sc_port_destroy = NULL; 1724 1725 return (0); 1726 } 1727 1728 static int 1729 lagg_fail_detach(struct lagg_softc *sc) 1730 { 1731 return (0); 1732 } 1733 1734 static int 1735 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 1736 { 1737 struct lagg_port *lp; 1738 1739 /* Use the master port if active or the next available port */ 1740 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 1741 m_freem(m); 1742 return (ENETDOWN); 1743 } 1744 1745 /* Send mbuf */ 1746 return (lagg_enqueue(lp->lp_ifp, m)); 1747 } 1748 1749 static struct mbuf * 1750 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1751 { 1752 struct ifnet *ifp = sc->sc_ifp; 1753 struct lagg_port *tmp_tp; 1754 1755 if (lp == sc->sc_primary || lagg_failover_rx_all) { 1756 m->m_pkthdr.rcvif = ifp; 1757 return (m); 1758 } 1759 1760 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1761 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1762 /* 1763 * If tmp_tp is null, we've recieved a packet when all 1764 * our links are down. Weird, but process it anyways. 1765 */ 1766 if ((tmp_tp == NULL || tmp_tp == lp)) { 1767 m->m_pkthdr.rcvif = ifp; 1768 return (m); 1769 } 1770 } 1771 1772 m_freem(m); 1773 return (NULL); 1774 } 1775 1776 /* 1777 * Loadbalancing 1778 */ 1779 1780 static int 1781 lagg_lb_attach(struct lagg_softc *sc) 1782 { 1783 struct lagg_port *lp; 1784 struct lagg_lb *lb; 1785 1786 if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb), 1787 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 1788 return (ENOMEM); 1789 1790 sc->sc_detach = lagg_lb_detach; 1791 sc->sc_start = lagg_lb_start; 1792 sc->sc_input = lagg_lb_input; 1793 sc->sc_port_create = lagg_lb_port_create; 1794 sc->sc_port_destroy = lagg_lb_port_destroy; 1795 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1796 1797 lb->lb_key = arc4random(); 1798 sc->sc_psc = (caddr_t)lb; 1799 1800 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1801 lagg_lb_port_create(lp); 1802 1803 return (0); 1804 } 1805 1806 static int 1807 lagg_lb_detach(struct lagg_softc *sc) 1808 { 1809 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1810 if (lb != NULL) 1811 free(lb, M_DEVBUF); 1812 return (0); 1813 } 1814 1815 static int 1816 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 1817 { 1818 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1819 struct lagg_port *lp_next; 1820 int i = 0; 1821 1822 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 1823 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1824 if (lp_next == lp) 1825 continue; 1826 if (i >= LAGG_MAX_PORTS) 1827 return (EINVAL); 1828 if (sc->sc_ifflags & IFF_DEBUG) 1829 printf("%s: port %s at index %d\n", 1830 sc->sc_ifname, lp_next->lp_ifname, i); 1831 lb->lb_ports[i++] = lp_next; 1832 } 1833 1834 return (0); 1835 } 1836 1837 static int 1838 lagg_lb_port_create(struct lagg_port *lp) 1839 { 1840 struct lagg_softc *sc = lp->lp_softc; 1841 return (lagg_lb_porttable(sc, NULL)); 1842 } 1843 1844 static void 1845 lagg_lb_port_destroy(struct lagg_port *lp) 1846 { 1847 struct lagg_softc *sc = lp->lp_softc; 1848 lagg_lb_porttable(sc, lp); 1849 } 1850 1851 static int 1852 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 1853 { 1854 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1855 struct lagg_port *lp = NULL; 1856 uint32_t p = 0; 1857 1858 if (sc->use_flowid && (m->m_flags & M_FLOWID)) 1859 p = m->m_pkthdr.flowid >> sc->flowid_shift; 1860 else 1861 p = lagg_hashmbuf(sc, m, lb->lb_key); 1862 p %= sc->sc_count; 1863 lp = lb->lb_ports[p]; 1864 1865 /* 1866 * Check the port's link state. This will return the next active 1867 * port if the link is down or the port is NULL. 1868 */ 1869 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1870 m_freem(m); 1871 return (ENETDOWN); 1872 } 1873 1874 /* Send mbuf */ 1875 return (lagg_enqueue(lp->lp_ifp, m)); 1876 } 1877 1878 static struct mbuf * 1879 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1880 { 1881 struct ifnet *ifp = sc->sc_ifp; 1882 1883 /* Just pass in the packet to our lagg device */ 1884 m->m_pkthdr.rcvif = ifp; 1885 1886 return (m); 1887 } 1888 1889 /* 1890 * 802.3ad LACP 1891 */ 1892 1893 static int 1894 lagg_lacp_attach(struct lagg_softc *sc) 1895 { 1896 struct lagg_port *lp; 1897 int error; 1898 1899 sc->sc_detach = lagg_lacp_detach; 1900 sc->sc_port_create = lacp_port_create; 1901 sc->sc_port_destroy = lacp_port_destroy; 1902 sc->sc_linkstate = lacp_linkstate; 1903 sc->sc_start = lagg_lacp_start; 1904 sc->sc_input = lagg_lacp_input; 1905 sc->sc_init = lacp_init; 1906 sc->sc_stop = lacp_stop; 1907 sc->sc_lladdr = lagg_lacp_lladdr; 1908 sc->sc_req = lacp_req; 1909 sc->sc_portreq = lacp_portreq; 1910 1911 error = lacp_attach(sc); 1912 if (error) 1913 return (error); 1914 1915 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1916 lacp_port_create(lp); 1917 1918 return (error); 1919 } 1920 1921 static int 1922 lagg_lacp_detach(struct lagg_softc *sc) 1923 { 1924 struct lagg_port *lp; 1925 int error; 1926 1927 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1928 lacp_port_destroy(lp); 1929 1930 /* unlocking is safe here */ 1931 LAGG_WUNLOCK(sc); 1932 error = lacp_detach(sc); 1933 LAGG_WLOCK(sc); 1934 1935 return (error); 1936 } 1937 1938 static void 1939 lagg_lacp_lladdr(struct lagg_softc *sc) 1940 { 1941 struct lagg_port *lp; 1942 1943 /* purge all the lacp ports */ 1944 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1945 lacp_port_destroy(lp); 1946 1947 /* add them back in */ 1948 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1949 lacp_port_create(lp); 1950 } 1951 1952 static int 1953 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 1954 { 1955 struct lagg_port *lp; 1956 1957 lp = lacp_select_tx_port(sc, m); 1958 if (lp == NULL) { 1959 m_freem(m); 1960 return (ENETDOWN); 1961 } 1962 1963 /* Send mbuf */ 1964 return (lagg_enqueue(lp->lp_ifp, m)); 1965 } 1966 1967 static struct mbuf * 1968 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1969 { 1970 struct ifnet *ifp = sc->sc_ifp; 1971 struct ether_header *eh; 1972 u_short etype; 1973 1974 eh = mtod(m, struct ether_header *); 1975 etype = ntohs(eh->ether_type); 1976 1977 /* Tap off LACP control messages */ 1978 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 1979 m = lacp_input(lp, m); 1980 if (m == NULL) 1981 return (NULL); 1982 } 1983 1984 /* 1985 * If the port is not collecting or not in the active aggregator then 1986 * free and return. 1987 */ 1988 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 1989 m_freem(m); 1990 return (NULL); 1991 } 1992 1993 m->m_pkthdr.rcvif = ifp; 1994 return (m); 1995 } 1996 1997 static void 1998 lagg_callout(void *arg) 1999 { 2000 struct lagg_softc *sc = (struct lagg_softc *)arg; 2001 struct ifnet *ifp = sc->sc_ifp; 2002 2003 ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets); 2004 ifp->if_opackets = counter_u64_fetch(sc->sc_opackets); 2005 ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes); 2006 ifp->if_obytes = counter_u64_fetch(sc->sc_obytes); 2007 2008 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 2009 } 2010