1 /* $OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $ 2 * $FreeBSD: head/sys/net/if_lagg.c 256218 2013-10-09 19:04:40Z glebius $ 3 */ 4 /* 5 * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org> 6 * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org> 7 * Copyright (c) 2014 Marcelo Araujo <araujo@FreeBSD.org> 8 * 9 * Permission to use, copy, modify, and distribute this software for any 10 * purpose with or without fee is hereby granted, provided that the above 11 * copyright notice and this permission notice appear in all copies. 12 * 13 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 14 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 15 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 16 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 17 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 18 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 19 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 20 */ 21 22 #include <sys/cdefs.h> 23 24 #include "opt_inet.h" 25 #include "opt_inet6.h" 26 27 #include <sys/param.h> 28 #include <sys/kernel.h> 29 #include <sys/malloc.h> 30 #include <sys/mbuf.h> 31 #include <sys/queue.h> 32 #include <sys/socket.h> 33 #include <sys/sockio.h> 34 #include <sys/sysctl.h> 35 #include <sys/module.h> 36 #include <sys/priv.h> 37 #include <sys/systm.h> 38 #include <sys/proc.h> 39 #include <sys/hash.h> 40 #include <sys/lock.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/ifq_var.h> 54 #include <net/bpf.h> 55 56 #include <net/netmsg2.h> 57 #include <net/netisr2.h> 58 59 #if defined(INET) || defined(INET6) 60 #include <netinet/in.h> 61 #endif 62 #ifdef INET 63 #include <netinet/in_systm.h> 64 #include <netinet/if_ether.h> 65 #include <netinet/ip.h> 66 #endif 67 68 #ifdef INET6 69 #include <netinet/ip6.h> 70 #include <netinet6/in6_var.h> 71 #include <netinet6/in6_ifattach.h> 72 #endif 73 74 #include <net/vlan/if_vlan_var.h> 75 #include <net/lagg/if_lagg.h> 76 #include <net/lagg/ieee8023ad_lacp.h> 77 78 /* Special flags we should propagate to the lagg ports. */ 79 static struct { 80 int flag; 81 int (*func)(struct ifnet *, int); 82 } lagg_pflags[] = { 83 {IFF_PROMISC, ifpromisc}, 84 {IFF_ALLMULTI, if_allmulti}, 85 {0, NULL} 86 }; 87 88 SLIST_HEAD(, lagg_softc) lagg_list; /* list of laggs */ 89 static struct lock lagg_list_lock; 90 eventhandler_tag lagg_detach_cookie = NULL; 91 92 static int lagg_clone_create(struct if_clone *, int, caddr_t); 93 static int lagg_clone_destroy(struct ifnet *); 94 static const char * laggname = "lagg"; 95 struct if_clone lagg_cloner = IF_CLONE_INITIALIZER("lagg", 96 lagg_clone_create, 97 lagg_clone_destroy, 98 0, IF_MAXUNIT); 99 100 static void lagg_lladdr(struct lagg_softc *, uint8_t *); 101 static void lagg_capabilities(struct lagg_softc *); 102 static void lagg_port_lladdr(struct lagg_port *, uint8_t *); 103 static void lagg_port_setlladdr(void *, int); 104 static int lagg_port_create(struct lagg_softc *, struct ifnet *); 105 static int lagg_port_destroy(struct lagg_port *, int); 106 static void lagg_input(struct ifnet *, struct mbuf *); 107 static void lagg_linkstate(struct lagg_softc *); 108 #if XXX 109 static void lagg_port_state(struct ifnet *, int); 110 #endif 111 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *cr); 112 static int lagg_port_output(struct ifnet *, struct mbuf *, 113 struct sockaddr *, struct rtentry *); 114 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 115 #ifdef LAGG_PORT_STACKING 116 static int lagg_port_checkstacking(struct lagg_softc *); 117 #endif 118 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 119 static void lagg_init(void *); 120 static void lagg_stop(struct lagg_softc *); 121 static int lagg_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *cr); 122 static int lagg_ether_setmulti(struct lagg_softc *); 123 static int lagg_ether_cmdmulti(struct lagg_port *, int); 124 static int lagg_setflag(struct lagg_port *, int, int, 125 int (*func)(struct ifnet *, int)); 126 static int lagg_setflags(struct lagg_port *, int status); 127 static void lagg_start(struct ifnet *, struct ifaltq_subque *ifsq); 128 static void lagg_start_dispatch(netmsg_t msg); 129 /* Not needed? 130 static int lagg_output(struct ifnet *ifp, struct mbuf *m); 131 */ 132 #if XXX 133 static int lagg_transmit(struct ifnet *, struct mbuf *); 134 static void lagg_qflush(struct ifnet *); 135 #endif 136 static int lagg_media_change(struct ifnet *); 137 static void lagg_media_status(struct ifnet *, struct ifmediareq *); 138 static struct lagg_port *lagg_link_active(struct lagg_softc *, 139 struct lagg_port *); 140 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *); 141 static int lagg_sysctl_active(SYSCTL_HANDLER_ARGS); 142 143 /* Simple round robin */ 144 static int lagg_rr_attach(struct lagg_softc *); 145 static int lagg_rr_detach(struct lagg_softc *); 146 static struct ifnet *lagg_rr_select_tx_port(struct lagg_softc *sc, 147 struct mbuf *m); 148 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 149 struct mbuf *); 150 151 /* Active failover */ 152 static int lagg_fail_attach(struct lagg_softc *); 153 static int lagg_fail_detach(struct lagg_softc *); 154 static struct ifnet *lagg_fail_select_tx_port(struct lagg_softc *, 155 struct mbuf *); 156 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 157 struct mbuf *); 158 159 /* Loadbalancing */ 160 static int lagg_lb_attach(struct lagg_softc *); 161 static int lagg_lb_detach(struct lagg_softc *); 162 static int lagg_lb_port_create(struct lagg_port *); 163 static void lagg_lb_port_destroy(struct lagg_port *); 164 static struct ifnet *lagg_lb_select_tx_port(struct lagg_softc *, 165 struct mbuf *); 166 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 167 struct mbuf *); 168 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 169 170 /* 802.3ad LACP */ 171 static int lagg_lacp_attach(struct lagg_softc *); 172 static int lagg_lacp_detach(struct lagg_softc *); 173 static struct ifnet *lagg_lacp_select_tx_port(struct lagg_softc *, 174 struct mbuf *); 175 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 176 struct mbuf *); 177 static void lagg_lacp_lladdr(struct lagg_softc *); 178 179 static void lagg_callout(void *); 180 181 /* lagg protocol table */ 182 static const struct { 183 int ti_proto; 184 int (*ti_attach)(struct lagg_softc *); 185 } lagg_protos[] = { 186 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach }, 187 { LAGG_PROTO_FAILOVER, lagg_fail_attach }, 188 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach }, 189 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach }, 190 { LAGG_PROTO_LACP, lagg_lacp_attach }, 191 { LAGG_PROTO_NONE, NULL } 192 }; 193 194 SYSCTL_DECL(_net_link); 195 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 196 "Link Aggregation"); 197 198 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */ 199 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW, 200 &lagg_failover_rx_all, 0, 201 "Accept input from any interface in a failover lagg"); 202 static int def_use_flowid = 1; /* Default value for using M_FLOWID */ 203 TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid); 204 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RW, 205 &def_use_flowid, 0, 206 "Default setting for using flow id for load sharing"); 207 208 static int 209 lagg_modevent(module_t mod, int type, void *data) 210 { 211 212 switch (type) { 213 case MOD_LOAD: 214 lockinit(&lagg_list_lock, "if_lagg list", 0, 0); 215 SLIST_INIT(&lagg_list); 216 if_clone_attach(&lagg_cloner); 217 lagg_input_p = lagg_input; 218 lagg_detach_cookie = EVENTHANDLER_REGISTER( 219 ifnet_departure_event, lagg_port_ifdetach, NULL, 220 EVENTHANDLER_PRI_ANY); 221 break; 222 case MOD_UNLOAD: 223 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 224 lagg_detach_cookie); 225 if_clone_detach(&lagg_cloner); 226 lagg_input_p = NULL; 227 if_clone_detach(&lagg_cloner); 228 lockuninit(&lagg_list_lock); 229 break; 230 default: 231 return (EOPNOTSUPP); 232 } 233 return (0); 234 } 235 236 static moduledata_t lagg_mod = { 237 "if_lagg", 238 lagg_modevent, 239 0 240 }; 241 242 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 243 MODULE_VERSION(if_lagg, 1); 244 245 /* 246 * This routine is run via an vlan 247 * config EVENT 248 */ 249 static void 250 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 251 { 252 #if XXX 253 struct lagg_softc *sc = ifp->if_softc; 254 struct lagg_port *lp; 255 256 if (ifp->if_softc != arg) /* Not our event */ 257 return; 258 259 LAGG_RLOCK(sc); 260 if (!SLIST_EMPTY(&sc->sc_ports)) { 261 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 262 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp); 263 } 264 LAGG_RUNLOCK(sc); 265 #endif 266 267 } 268 269 /* 270 * This routine is run via an vlan 271 * unconfig EVENT 272 */ 273 static void 274 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 275 { 276 struct lagg_softc *sc = ifp->if_softc; 277 /* 278 struct lagg_port *lp; 279 */ 280 281 if (ifp->if_softc != arg) /* Not our event */ 282 return; 283 284 LAGG_RLOCK(sc); 285 if (!SLIST_EMPTY(&sc->sc_ports)) { 286 #if XXX 287 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 288 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp); 289 #endif 290 } 291 LAGG_RUNLOCK(sc); 292 } 293 294 static int 295 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params __unused) 296 { 297 struct lagg_softc *sc; 298 struct ifnet *ifp; 299 int i, error = 0; 300 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ 301 struct sysctl_oid *oid; 302 char num[14]; /* sufficient for 32 bits */ 303 304 sc = kmalloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO); 305 ifp = sc->sc_ifp = &sc->sc_if; // XXX if_alloc(IFT_ETHER); 306 /* 307 if (ifp == NULL) { 308 kfree(sc, M_DEVBUF); 309 return (ENOSPC); 310 } 311 */ 312 /* 313 sc->sc_ipackets = counter_u64_alloc(M_WAITOK); 314 sc->sc_opackets = counter_u64_alloc(M_WAITOK); 315 sc->sc_ibytes = counter_u64_alloc(M_WAITOK); 316 sc->sc_obytes = counter_u64_alloc(M_WAITOK); 317 */ 318 sysctl_ctx_init(&sc->ctx); 319 ksnprintf(num, sizeof(num), "%u", unit); 320 sc->use_flowid = def_use_flowid; 321 sc->sc_oid = oid = SYSCTL_ADD_NODE(&sc->ctx, 322 &SYSCTL_NODE_CHILDREN(_net_link, lagg), 323 OID_AUTO, num, CTLFLAG_RD, NULL, ""); 324 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 325 "use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid, sc->use_flowid, 326 "Use flow id for load sharing"); 327 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 328 "count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count, 329 "Total number of ports"); 330 SYSCTL_ADD_PROC(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 331 "active", CTLTYPE_INT|CTLFLAG_RD, sc, 0, lagg_sysctl_active, 332 "I", "Total number of active ports"); 333 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 334 "flapping", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_flapping, 335 sc->sc_flapping, "Total number of port change events"); 336 /* Hash all layers by default */ 337 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4; 338 339 sc->sc_proto = LAGG_PROTO_NONE; 340 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) { 341 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) { 342 sc->sc_proto = lagg_protos[i].ti_proto; 343 if ((error = lagg_protos[i].ti_attach(sc)) != 0) { 344 if_free(ifp); 345 kfree(sc, M_DEVBUF); 346 return (error); 347 } 348 break; 349 } 350 } 351 LAGG_LOCK_INIT(sc); 352 LAGG_CALLOUT_LOCK_INIT(sc); 353 SLIST_INIT(&sc->sc_ports); 354 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc); 355 356 /* 357 * This uses the callout lock rather than the rmlock; one can't 358 * hold said rmlock during SWI. 359 */ 360 callout_init(&sc->sc_callout); 361 /*, &sc->sc_call_lock, 0); */ 362 363 /* Initialise pseudo media types */ 364 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 365 lagg_media_status); 366 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 367 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 368 369 if_initname(ifp, laggname, unit); 370 ifp->if_softc = sc; 371 #if XXX 372 ifp->if_transmit = lagg_transmit; 373 ifp->if_qflush = lagg_qflush; 374 #endif 375 ifp->if_mtu = ETHERMTU; 376 ifp->if_init = lagg_init; 377 ifp->if_ioctl = lagg_ioctl; 378 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 379 ifp->if_start = lagg_start; 380 ifp->if_type = IFT_ETHER; 381 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen); 382 ifq_set_ready(&ifp->if_snd); 383 ifp->if_hdrlen = ETHER_HDR_LEN; 384 385 #if XXX 386 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS; 387 #endif 388 /* 389 * Attach as an ordinary ethernet device, children will be attached 390 * as special device IFT_IEEE8023ADLAG. 391 */ 392 393 ether_ifattach(ifp, eaddr, NULL); 394 395 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 396 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 397 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 398 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 399 400 /* Insert into the global list of laggs */ 401 lockmgr(&lagg_list_lock, LK_EXCLUSIVE); 402 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries); 403 lockmgr(&lagg_list_lock, LK_RELEASE); 404 405 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 406 407 return (0); 408 } 409 410 static int 411 lagg_clone_destroy(struct ifnet *ifp) 412 { 413 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 414 struct lagg_port *lp; 415 416 LAGG_WLOCK(sc); 417 418 lagg_stop(sc); 419 ifp->if_flags &= ~IFF_UP; 420 421 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 422 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 423 424 /* Shutdown and remove lagg ports */ 425 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) 426 lagg_port_destroy(lp, 1); 427 /* Unhook the aggregation protocol */ 428 if (sc->sc_detach != NULL) 429 (*sc->sc_detach)(sc); 430 431 LAGG_WUNLOCK(sc); 432 433 sysctl_ctx_free(&sc->ctx); 434 ifmedia_removeall(&sc->sc_media); 435 ether_ifdetach(ifp); 436 if_free(ifp); 437 438 /* This grabs sc_callout_mtx, serialising it correctly */ 439 callout_drain(&sc->sc_callout); 440 441 #if 0 442 /* At this point it's drained; we can free this */ 443 counter_u64_free(sc->sc_ipackets); 444 counter_u64_free(sc->sc_opackets); 445 counter_u64_free(sc->sc_ibytes); 446 counter_u64_free(sc->sc_obytes); 447 #endif 448 449 lockmgr(&lagg_list_lock, LK_EXCLUSIVE); 450 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries); 451 lockmgr(&lagg_list_lock, LK_RELEASE); 452 453 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task); 454 LAGG_LOCK_DESTROY(sc); 455 LAGG_CALLOUT_LOCK_DESTROY(sc); 456 kfree(sc, M_DEVBUF); 457 458 return 0; 459 } 460 461 static void 462 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr) 463 { 464 struct ifnet *ifp = sc->sc_ifp; 465 466 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 467 return; 468 469 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN); 470 /* Let the protocol know the MAC has changed */ 471 if (sc->sc_lladdr != NULL) 472 (*sc->sc_lladdr)(sc); 473 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 474 } 475 476 static void 477 lagg_capabilities(struct lagg_softc *sc) 478 { 479 struct lagg_port *lp; 480 int cap = ~0, ena = ~0; 481 u_long hwa = ~0UL; 482 483 LAGG_WLOCK_ASSERT(sc); 484 485 /* Get capabilities from the lagg ports */ 486 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 487 cap &= lp->lp_ifp->if_capabilities; 488 ena &= lp->lp_ifp->if_capenable; 489 hwa &= lp->lp_ifp->if_hwassist; 490 } 491 cap = (cap == ~0 ? 0 : cap); 492 ena = (ena == ~0 ? 0 : ena); 493 hwa = (hwa == ~0 ? 0 : hwa); 494 495 if (sc->sc_ifp->if_capabilities != cap || 496 sc->sc_ifp->if_capenable != ena || 497 sc->sc_ifp->if_hwassist != hwa) { 498 sc->sc_ifp->if_capabilities = cap; 499 sc->sc_ifp->if_capenable = ena; 500 sc->sc_ifp->if_hwassist = hwa; 501 getmicrotime(&sc->sc_ifp->if_lastchange); 502 503 if (sc->sc_ifflags & IFF_DEBUG) 504 if_printf(sc->sc_ifp, 505 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 506 } 507 } 508 509 static void 510 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr) 511 { 512 struct lagg_softc *sc = lp->lp_softc; 513 struct ifnet *ifp = lp->lp_ifp; 514 struct lagg_llq *llq; 515 int pending = 0; 516 517 LAGG_WLOCK_ASSERT(sc); 518 519 if (lp->lp_detaching || 520 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 521 return; 522 523 /* Check to make sure its not already queued to be changed */ 524 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 525 if (llq->llq_ifp == ifp) { 526 pending = 1; 527 break; 528 } 529 } 530 531 if (!pending) { 532 llq = kmalloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT); 533 if (llq == NULL) /* XXX what to do */ 534 return; 535 } 536 537 /* Update the lladdr even if pending, it may have changed */ 538 llq->llq_ifp = ifp; 539 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN); 540 541 if (!pending) 542 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries); 543 544 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task); 545 } 546 547 /* 548 * Set the interface MAC address from a taskqueue to avoid a LOR. 549 */ 550 static void 551 lagg_port_setlladdr(void *arg, int pending) 552 { 553 struct lagg_softc *sc = (struct lagg_softc *)arg; 554 struct lagg_llq *llq, *head; 555 struct ifnet *ifp; 556 int error; 557 558 /* Grab a local reference of the queue and remove it from the softc */ 559 LAGG_WLOCK(sc); 560 head = SLIST_FIRST(&sc->sc_llq_head); 561 SLIST_FIRST(&sc->sc_llq_head) = NULL; 562 LAGG_WUNLOCK(sc); 563 564 /* 565 * Traverse the queue and set the lladdr on each ifp. It is safe to do 566 * unlocked as we have the only reference to it. 567 */ 568 for (llq = head; llq != NULL; llq = head) { 569 ifp = llq->llq_ifp; 570 571 /* Set the link layer address */ 572 /* CURVNET_SET(ifp->if_vnet); */ 573 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN); 574 /* CURVNET_RESTORE(); */ 575 if (error) 576 kprintf("%s: setlladdr failed on %s\n", __func__, 577 ifp->if_xname); 578 579 head = SLIST_NEXT(llq, llq_entries); 580 kfree(llq, M_DEVBUF); 581 } 582 } 583 584 static int 585 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 586 { 587 struct lagg_softc *sc_ptr; 588 struct lagg_port *lp; 589 int error = 0; 590 591 LAGG_WLOCK_ASSERT(sc); 592 593 /* Limit the maximal number of lagg ports */ 594 if (sc->sc_count >= LAGG_MAX_PORTS) 595 return (ENOSPC); 596 597 /* Check if port has already been associated to a lagg */ 598 if (ifp->if_lagg != NULL) { 599 /* Port is already in the current lagg? */ 600 lp = (struct lagg_port *)ifp->if_lagg; 601 if (lp->lp_softc == sc) 602 return (EEXIST); 603 return (EBUSY); 604 } 605 606 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 607 if (ifp->if_type != IFT_ETHER) 608 return (EPROTONOSUPPORT); 609 610 #ifdef INET6 611 /* 612 * The member interface should not have inet6 address because 613 * two interfaces with a valid link-local scope zone must not be 614 * merged in any form. This restriction is needed to 615 * prevent violation of link-local scope zone. Attempts to 616 * add a member interface which has inet6 addresses triggers 617 * removal of all inet6 addresses on the member interface. 618 */ 619 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 620 if (in6ifa_llaonifp(lp->lp_ifp)) { 621 in6_ifdetach(lp->lp_ifp); 622 if_printf(sc->sc_ifp, 623 "IPv6 addresses on %s have been removed " 624 "before adding it as a member to prevent " 625 "IPv6 address scope violation.\n", 626 lp->lp_ifp->if_xname); 627 } 628 } 629 if (in6ifa_llaonifp(ifp)) { 630 in6_ifdetach(ifp); 631 if_printf(sc->sc_ifp, 632 "IPv6 addresses on %s have been removed " 633 "before adding it as a member to prevent " 634 "IPv6 address scope violation.\n", 635 ifp->if_xname); 636 } 637 #endif 638 /* Allow the first Ethernet member to define the MTU */ 639 if (SLIST_EMPTY(&sc->sc_ports)) 640 sc->sc_ifp->if_mtu = ifp->if_mtu; 641 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 642 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 643 ifp->if_xname); 644 return (EINVAL); 645 } 646 647 if ((lp = kmalloc(sizeof(struct lagg_port), 648 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 649 return (ENOMEM); 650 651 /* Check if port is a stacked lagg */ 652 lockmgr(&lagg_list_lock, LK_EXCLUSIVE); 653 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) { 654 if (ifp == sc_ptr->sc_ifp) { 655 lockmgr(&lagg_list_lock, LK_RELEASE); 656 kfree(lp, M_DEVBUF); 657 return (EINVAL); 658 /* XXX disable stacking for the moment, its untested */ 659 #ifdef LAGG_PORT_STACKING 660 lp->lp_flags |= LAGG_PORT_STACK; 661 if (lagg_port_checkstacking(sc_ptr) >= 662 LAGG_MAX_STACKING) { 663 lockmgr(&lagg_list_lock, LK_RELEASE); 664 kfree(lp, M_DEVBUF); 665 return (E2BIG); 666 } 667 #endif 668 } 669 } 670 lockmgr(&lagg_list_lock, LK_RELEASE); 671 672 /* Change the interface type */ 673 lp->lp_iftype = ifp->if_type; 674 ifp->if_type = IFT_IEEE8023ADLAG; 675 ifp->if_lagg = lp; 676 lp->lp_ioctl = ifp->if_ioctl; 677 ifp->if_ioctl = lagg_port_ioctl; 678 lp->lp_output = ifp->if_output; 679 ifp->if_output = lagg_port_output; 680 681 lp->lp_ifp = ifp; 682 lp->lp_softc = sc; 683 684 /* Save port link layer address */ 685 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 686 687 if (SLIST_EMPTY(&sc->sc_ports)) { 688 sc->sc_primary = lp; 689 lagg_lladdr(sc, IF_LLADDR(ifp)); 690 } else { 691 /* Update link layer address for this port */ 692 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp)); 693 } 694 695 /* Insert into the list of ports */ 696 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 697 sc->sc_count++; 698 699 /* Update lagg capabilities */ 700 lagg_capabilities(sc); 701 lagg_linkstate(sc); 702 703 /* Add multicast addresses and interface flags to this port */ 704 lagg_ether_cmdmulti(lp, 1); 705 lagg_setflags(lp, 1); 706 707 if (sc->sc_port_create != NULL) 708 error = (*sc->sc_port_create)(lp); 709 if (error) { 710 /* remove the port again, without calling sc_port_destroy */ 711 lagg_port_destroy(lp, 0); 712 return (error); 713 } 714 715 return (error); 716 } 717 718 #ifdef LAGG_PORT_STACKING 719 static int 720 lagg_port_checkstacking(struct lagg_softc *sc) 721 { 722 struct lagg_softc *sc_ptr; 723 struct lagg_port *lp; 724 int m = 0; 725 726 LAGG_WLOCK_ASSERT(sc); 727 728 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 729 if (lp->lp_flags & LAGG_PORT_STACK) { 730 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 731 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 732 } 733 } 734 735 return (m + 1); 736 } 737 #endif 738 739 static int 740 lagg_port_destroy(struct lagg_port *lp, int runpd) 741 { 742 struct lagg_softc *sc = lp->lp_softc; 743 struct lagg_port *lp_ptr; 744 struct lagg_llq *llq; 745 struct ifnet *ifp = lp->lp_ifp; 746 747 LAGG_WLOCK_ASSERT(sc); 748 749 if (runpd && sc->sc_port_destroy != NULL) 750 (*sc->sc_port_destroy)(lp); 751 752 /* 753 * Remove multicast addresses and interface flags from this port and 754 * reset the MAC address, skip if the interface is being detached. 755 */ 756 if (!lp->lp_detaching) { 757 lagg_ether_cmdmulti(lp, 0); 758 lagg_setflags(lp, 0); 759 lagg_port_lladdr(lp, lp->lp_lladdr); 760 } 761 762 /* Restore interface */ 763 ifp->if_type = lp->lp_iftype; 764 ifp->if_ioctl = lp->lp_ioctl; 765 ifp->if_output = lp->lp_output; 766 ifp->if_lagg = NULL; 767 768 /* Finally, remove the port from the lagg */ 769 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 770 sc->sc_count--; 771 772 /* Update the primary interface */ 773 if (lp == sc->sc_primary) { 774 uint8_t lladdr[ETHER_ADDR_LEN]; 775 776 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) { 777 bzero(&lladdr, ETHER_ADDR_LEN); 778 } else { 779 bcopy(lp_ptr->lp_lladdr, 780 lladdr, ETHER_ADDR_LEN); 781 } 782 lagg_lladdr(sc, lladdr); 783 sc->sc_primary = lp_ptr; 784 785 /* Update link layer address for each port */ 786 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 787 lagg_port_lladdr(lp_ptr, lladdr); 788 } 789 790 /* Remove any pending lladdr changes from the queue */ 791 if (lp->lp_detaching) { 792 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 793 if (llq->llq_ifp == ifp) { 794 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq, 795 llq_entries); 796 kfree(llq, M_DEVBUF); 797 break; /* Only appears once */ 798 } 799 } 800 } 801 802 if (lp->lp_ifflags) 803 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 804 805 kfree(lp, M_DEVBUF); 806 807 /* Update lagg capabilities */ 808 lagg_capabilities(sc); 809 lagg_linkstate(sc); 810 811 return (0); 812 } 813 814 static int 815 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr) 816 { 817 struct lagg_reqport *rp = (struct lagg_reqport *)data; 818 struct lagg_softc *sc; 819 struct lagg_port *lp = NULL; 820 int error = 0; 821 822 ASSERT_IFNET_SERIALIZED_ALL(ifp); 823 824 /* Should be checked by the caller */ 825 if (ifp->if_type != IFT_IEEE8023ADLAG || 826 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 827 goto fallback; 828 829 switch (cmd) { 830 case SIOCGLAGGPORT: 831 if (rp->rp_portname[0] == '\0') { 832 error = EINVAL; 833 break; 834 } 835 836 /* 837 * Release ifp serializers before ifnet_lock 838 * to prevent lock order reversal. 839 */ 840 ifnet_deserialize_all(ifp); 841 ifnet_lock(); 842 if (ifunit(rp->rp_portname) != ifp) { 843 ifnet_unlock(); 844 ifnet_serialize_all(ifp); 845 error = EINVAL; 846 break; 847 } 848 ifnet_unlock(); 849 ifnet_serialize_all(ifp); 850 851 LAGG_RLOCK(sc); 852 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 853 error = ENOENT; 854 LAGG_RUNLOCK(sc); 855 break; 856 } 857 858 lagg_port2req(lp, rp); 859 LAGG_RUNLOCK(sc); 860 break; 861 862 case SIOCSIFCAP: 863 if (lp->lp_ioctl == NULL) { 864 error = EINVAL; 865 break; 866 } 867 868 error = (*lp->lp_ioctl)(ifp, cmd, data, cr); 869 if (error) 870 break; 871 872 /* Update lagg interface capabilities */ 873 LAGG_WLOCK(sc); 874 lagg_capabilities(sc); 875 LAGG_WUNLOCK(sc); 876 break; 877 case SIOCGIFMEDIA: 878 if (lp->lp_ioctl == NULL) { 879 error = EINVAL; 880 break; 881 } 882 883 error = (*lp->lp_ioctl)(ifp, cmd, data, cr); 884 break; 885 case SIOCSIFMTU: 886 /* Do not allow the MTU to be changed once joined */ 887 error = EINVAL; 888 break; 889 890 default: 891 goto fallback; 892 } 893 894 return (error); 895 896 fallback: 897 if (lp->lp_ioctl != NULL) { 898 int result; 899 result = ((*lp->lp_ioctl)(ifp, cmd, data, cr)); 900 } 901 return (EINVAL); 902 } 903 904 /* 905 * For direct output to child ports. 906 */ 907 static int 908 lagg_port_output(struct ifnet *ifp, struct mbuf *m, 909 struct sockaddr *dst, struct rtentry *ro) 910 { 911 struct lagg_port *lp = ifp->if_lagg; 912 913 switch (dst->sa_family) { 914 case pseudo_AF_HDRCMPLT: 915 case AF_UNSPEC: 916 return ((*lp->lp_output)(ifp, m, dst, ro)); 917 } 918 919 /* drop any other frames */ 920 m_freem(m); 921 return (ENETDOWN); 922 } 923 924 static void 925 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 926 { 927 struct lagg_port *lp; 928 struct lagg_softc *sc; 929 930 if ((lp = ifp->if_lagg) == NULL) 931 return; 932 #if XXX 933 /* If the ifnet is just being renamed, don't do anything. */ 934 if (ifp->if_flags & IFF_RENAMING) 935 return; 936 #endif 937 sc = lp->lp_softc; 938 939 LAGG_WLOCK(sc); 940 lp->lp_detaching = 1; 941 lagg_port_destroy(lp, 1); 942 LAGG_WUNLOCK(sc); 943 } 944 945 static void 946 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 947 { 948 struct lagg_softc *sc = lp->lp_softc; 949 950 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 951 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 952 rp->rp_prio = lp->lp_prio; 953 rp->rp_flags = lp->lp_flags; 954 if (sc->sc_portreq != NULL) 955 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc); 956 957 /* Add protocol specific flags */ 958 switch (sc->sc_proto) { 959 case LAGG_PROTO_FAILOVER: 960 if (lp == sc->sc_primary) 961 rp->rp_flags |= LAGG_PORT_MASTER; 962 if (lp == lagg_link_active(sc, sc->sc_primary)) 963 rp->rp_flags |= LAGG_PORT_ACTIVE; 964 break; 965 966 case LAGG_PROTO_ROUNDROBIN: 967 case LAGG_PROTO_LOADBALANCE: 968 case LAGG_PROTO_ETHERCHANNEL: 969 if (LAGG_PORTACTIVE(lp)) 970 rp->rp_flags |= LAGG_PORT_ACTIVE; 971 break; 972 973 case LAGG_PROTO_LACP: 974 /* LACP has a different definition of active */ 975 if (lacp_isactive(lp)) 976 rp->rp_flags |= LAGG_PORT_ACTIVE; 977 if (lacp_iscollecting(lp)) 978 rp->rp_flags |= LAGG_PORT_COLLECTING; 979 if (lacp_isdistributing(lp)) 980 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 981 break; 982 } 983 984 } 985 986 static void 987 lagg_init(void *xsc) 988 { 989 struct lagg_softc *sc = (struct lagg_softc *)xsc; 990 struct lagg_port *lp; 991 struct ifnet *ifp = sc->sc_ifp; 992 993 if (ifp->if_flags & IFF_RUNNING) 994 return; 995 996 LAGG_WLOCK(sc); 997 998 ifp->if_flags |= IFF_RUNNING; 999 /* Update the port lladdrs */ 1000 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1001 lagg_port_lladdr(lp, IF_LLADDR(ifp)); 1002 1003 if (sc->sc_init != NULL) 1004 (*sc->sc_init)(sc); 1005 1006 LAGG_WUNLOCK(sc); 1007 } 1008 1009 static void 1010 lagg_stop(struct lagg_softc *sc) 1011 { 1012 struct ifnet *ifp = sc->sc_ifp; 1013 1014 LAGG_WLOCK_ASSERT(sc); 1015 1016 if ((ifp->if_flags & IFF_RUNNING) == 0) 1017 return; 1018 1019 ifp->if_flags &= ~IFF_RUNNING; 1020 1021 if (sc->sc_stop != NULL) 1022 (*sc->sc_stop)(sc); 1023 } 1024 1025 static int 1026 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr) 1027 { 1028 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1029 struct lagg_reqall *ra = (struct lagg_reqall *)data; 1030 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 1031 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 1032 struct ifreq *ifr = (struct ifreq *)data; 1033 struct lagg_port *lp; 1034 struct ifnet *tpif; 1035 struct thread *td = curthread; 1036 char *buf, *outbuf; 1037 int count, buflen, len, error = 0; 1038 1039 ASSERT_IFNET_SERIALIZED_ALL(ifp); 1040 1041 bzero(&rpbuf, sizeof(rpbuf)); 1042 1043 switch (cmd) { 1044 case SIOCGLAGG: 1045 LAGG_RLOCK(sc); 1046 count = 0; 1047 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1048 count++; 1049 buflen = count * sizeof(struct lagg_reqport); 1050 LAGG_RUNLOCK(sc); 1051 1052 outbuf = kmalloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 1053 1054 LAGG_RLOCK(sc); 1055 ra->ra_proto = sc->sc_proto; 1056 if (sc->sc_req != NULL) 1057 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc); 1058 1059 count = 0; 1060 buf = outbuf; 1061 len = min(ra->ra_size, buflen); 1062 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1063 if (len < sizeof(rpbuf)) 1064 break; 1065 1066 lagg_port2req(lp, &rpbuf); 1067 memcpy(buf, &rpbuf, sizeof(rpbuf)); 1068 count++; 1069 buf += sizeof(rpbuf); 1070 len -= sizeof(rpbuf); 1071 } 1072 LAGG_RUNLOCK(sc); 1073 ra->ra_ports = count; 1074 ra->ra_size = count * sizeof(rpbuf); 1075 error = copyout(outbuf, ra->ra_port, ra->ra_size); 1076 kfree(outbuf, M_TEMP); 1077 break; 1078 case SIOCSLAGG: 1079 error = priv_check(td, PRIV_NET_LAGG); 1080 if (error) 1081 break; 1082 if (ra->ra_proto >= LAGG_PROTO_MAX) { 1083 error = EPROTONOSUPPORT; 1084 break; 1085 } 1086 LAGG_WLOCK(sc); 1087 if (sc->sc_proto != LAGG_PROTO_NONE) { 1088 /* Reset protocol first in case detach unlocks */ 1089 sc->sc_proto = LAGG_PROTO_NONE; 1090 error = sc->sc_detach(sc); 1091 sc->sc_detach = NULL; 1092 sc->sc_start = NULL; 1093 sc->sc_input = NULL; 1094 sc->sc_port_create = NULL; 1095 sc->sc_port_destroy = NULL; 1096 sc->sc_linkstate = NULL; 1097 sc->sc_init = NULL; 1098 sc->sc_stop = NULL; 1099 sc->sc_lladdr = NULL; 1100 sc->sc_req = NULL; 1101 sc->sc_portreq = NULL; 1102 } else if (sc->sc_input != NULL) { 1103 /* Still detaching */ 1104 error = EBUSY; 1105 } 1106 if (error != 0) { 1107 LAGG_WUNLOCK(sc); 1108 break; 1109 } 1110 for (int i = 0; i < (sizeof(lagg_protos) / 1111 sizeof(lagg_protos[0])); i++) { 1112 if (lagg_protos[i].ti_proto == ra->ra_proto) { 1113 if (sc->sc_ifflags & IFF_DEBUG) 1114 kprintf("%s: using proto %u\n", 1115 sc->sc_ifname, 1116 lagg_protos[i].ti_proto); 1117 sc->sc_proto = lagg_protos[i].ti_proto; 1118 if (sc->sc_proto != LAGG_PROTO_NONE) 1119 error = lagg_protos[i].ti_attach(sc); 1120 LAGG_WUNLOCK(sc); 1121 return (error); 1122 } 1123 } 1124 LAGG_WUNLOCK(sc); 1125 error = EPROTONOSUPPORT; 1126 break; 1127 case SIOCGLAGGFLAGS: 1128 rf->rf_flags = sc->sc_flags; 1129 break; 1130 case SIOCSLAGGHASH: 1131 error = priv_check(td, PRIV_NET_LAGG); 1132 if (error) 1133 break; 1134 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1135 error = EINVAL; 1136 break; 1137 } 1138 LAGG_WLOCK(sc); 1139 sc->sc_flags &= ~LAGG_F_HASHMASK; 1140 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK; 1141 LAGG_WUNLOCK(sc); 1142 break; 1143 case SIOCGLAGGPORT: 1144 /* 1145 * Release ifp serializers before ifnet_lock 1146 * to prevent lock order reversal. 1147 */ 1148 ifnet_deserialize_all(ifp); 1149 ifnet_lock(); 1150 if (rp->rp_portname[0] == '\0' || 1151 (tpif = ifunit(rp->rp_portname)) == NULL) { 1152 ifnet_unlock(); 1153 ifnet_serialize_all(ifp); 1154 error = EINVAL; 1155 break; 1156 } 1157 1158 LAGG_RLOCK(sc); 1159 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1160 lp->lp_softc != sc) { 1161 error = ENOENT; 1162 LAGG_RUNLOCK(sc); 1163 ifnet_unlock(); 1164 ifnet_serialize_all(ifp); 1165 break; 1166 } 1167 1168 lagg_port2req(lp, rp); 1169 LAGG_RUNLOCK(sc); 1170 ifnet_unlock(); 1171 ifnet_serialize_all(ifp); 1172 break; 1173 case SIOCSLAGGPORT: 1174 error = priv_check(td, PRIV_NET_LAGG); 1175 if (error) 1176 break; 1177 /* 1178 * Release ifp serializers before ifnet_lock 1179 * to prevent lock order reversal. 1180 */ 1181 ifnet_deserialize_all(ifp); 1182 ifnet_lock(); 1183 if (rp->rp_portname[0] == '\0' || 1184 (tpif = ifunit(rp->rp_portname)) == NULL) { 1185 ifnet_unlock(); 1186 ifnet_serialize_all(ifp); 1187 error = EINVAL; 1188 break; 1189 } 1190 LAGG_WLOCK(sc); 1191 error = lagg_port_create(sc, tpif); 1192 LAGG_WUNLOCK(sc); 1193 ifnet_unlock(); 1194 ifnet_serialize_all(ifp); 1195 break; 1196 case SIOCSLAGGDELPORT: 1197 error = priv_check(td, PRIV_NET_LAGG); 1198 if (error) 1199 break; 1200 /* 1201 * Release ifp serializers before ifnet_lock 1202 * to prevent lock order reversal. 1203 */ 1204 ifnet_deserialize_all(ifp); 1205 ifnet_lock(); 1206 if (rp->rp_portname[0] == '\0' || 1207 (tpif = ifunit(rp->rp_portname)) == NULL) { 1208 ifnet_unlock(); 1209 ifnet_serialize_all(ifp); 1210 error = EINVAL; 1211 break; 1212 } 1213 1214 LAGG_WLOCK(sc); 1215 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1216 lp->lp_softc != sc) { 1217 error = ENOENT; 1218 LAGG_WUNLOCK(sc); 1219 ifnet_unlock(); 1220 ifnet_serialize_all(ifp); 1221 break; 1222 } 1223 1224 error = lagg_port_destroy(lp, 1); 1225 LAGG_WUNLOCK(sc); 1226 ifnet_unlock(); 1227 ifnet_serialize_all(ifp); 1228 break; 1229 case SIOCSIFFLAGS: 1230 /* Set flags on ports too */ 1231 LAGG_WLOCK(sc); 1232 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1233 lagg_setflags(lp, 1); 1234 } 1235 LAGG_WUNLOCK(sc); 1236 1237 if (!(ifp->if_flags & IFF_UP) && 1238 (ifp->if_flags & IFF_RUNNING)) { 1239 /* 1240 * If interface is marked down and it is running, 1241 * then stop and disable it. 1242 */ 1243 LAGG_WLOCK(sc); 1244 lagg_stop(sc); 1245 LAGG_WUNLOCK(sc); 1246 } else if ((ifp->if_flags & IFF_UP) && 1247 !(ifp->if_flags & IFF_RUNNING)) { 1248 /* 1249 * If interface is marked up and it is stopped, then 1250 * start it. 1251 */ 1252 (*ifp->if_init)(sc); 1253 } 1254 break; 1255 case SIOCADDMULTI: 1256 case SIOCDELMULTI: 1257 LAGG_WLOCK(sc); 1258 error = lagg_ether_setmulti(sc); 1259 LAGG_WUNLOCK(sc); 1260 break; 1261 case SIOCSIFMEDIA: 1262 case SIOCGIFMEDIA: 1263 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1264 break; 1265 1266 case SIOCSIFCAP: 1267 case SIOCSIFMTU: 1268 /* Do not allow the MTU or caps to be directly changed */ 1269 error = EINVAL; 1270 break; 1271 1272 default: 1273 error = ether_ioctl(ifp, cmd, data); 1274 break; 1275 } 1276 return (error); 1277 } 1278 1279 static int 1280 lagg_ether_setmulti(struct lagg_softc *sc) 1281 { 1282 struct lagg_port *lp; 1283 1284 LAGG_WLOCK_ASSERT(sc); 1285 1286 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1287 /* First, remove any existing filter entries. */ 1288 lagg_ether_cmdmulti(lp, 0); 1289 /* copy all addresses from the lagg interface to the port */ 1290 lagg_ether_cmdmulti(lp, 1); 1291 } 1292 return (0); 1293 } 1294 1295 static int 1296 lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1297 { 1298 struct lagg_softc *sc = lp->lp_softc; 1299 struct ifnet *ifp = lp->lp_ifp; 1300 struct ifnet *scifp = sc->sc_ifp; 1301 struct lagg_mc *mc; 1302 struct ifmultiaddr *ifma, *rifma = NULL; 1303 struct sockaddr_dl sdl; 1304 int error; 1305 1306 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp); 1307 LAGG_WLOCK_ASSERT(sc); 1308 1309 bzero((char *)&sdl, sizeof(sdl)); 1310 sdl.sdl_len = sizeof(sdl); 1311 sdl.sdl_family = AF_LINK; 1312 sdl.sdl_type = IFT_ETHER; 1313 sdl.sdl_alen = ETHER_ADDR_LEN; 1314 sdl.sdl_index = ifp->if_index; 1315 1316 if (set) { 1317 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1318 if (ifma->ifma_addr->sa_family != AF_LINK) 1319 continue; 1320 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 1321 LLADDR(&sdl), ETHER_ADDR_LEN); 1322 1323 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 1324 if (error) 1325 return (error); 1326 mc = kmalloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1327 if (mc == NULL) 1328 return (ENOMEM); 1329 mc->mc_ifma = rifma; 1330 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1331 } 1332 } else { 1333 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1334 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1335 if_delmulti(ifp, (struct sockaddr *)mc->mc_ifma); 1336 kfree(mc, M_DEVBUF); 1337 } 1338 } 1339 return (0); 1340 } 1341 1342 /* Handle a ref counted flag that should be set on the lagg port as well */ 1343 static int 1344 lagg_setflag(struct lagg_port *lp, int flag, int status, 1345 int (*func)(struct ifnet *, int)) 1346 { 1347 struct lagg_softc *sc = lp->lp_softc; 1348 struct ifnet *scifp = sc->sc_ifp; 1349 struct ifnet *ifp = lp->lp_ifp; 1350 int error; 1351 1352 LAGG_WLOCK_ASSERT(sc); 1353 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp); 1354 1355 status = status ? (scifp->if_flags & flag) : 0; 1356 /* Now "status" contains the flag value or 0 */ 1357 1358 /* 1359 * See if recorded ports status is different from what 1360 * we want it to be. If it is, flip it. We record ports 1361 * status in lp_ifflags so that we won't clear ports flag 1362 * we haven't set. In fact, we don't clear or set ports 1363 * flags directly, but get or release references to them. 1364 * That's why we can be sure that recorded flags still are 1365 * in accord with actual ports flags. 1366 */ 1367 if (status != (lp->lp_ifflags & flag)) { 1368 error = (*func)(ifp, status); 1369 if (error) 1370 return (error); 1371 lp->lp_ifflags &= ~flag; 1372 lp->lp_ifflags |= status; 1373 } 1374 return (0); 1375 } 1376 1377 /* 1378 * Handle IFF_* flags that require certain changes on the lagg port 1379 * if "status" is true, update ports flags respective to the lagg 1380 * if "status" is false, forcedly clear the flags set on port. 1381 */ 1382 static int 1383 lagg_setflags(struct lagg_port *lp, int status) 1384 { 1385 int error, i; 1386 1387 ASSERT_IFNET_NOT_SERIALIZED_ALL(lp->lp_ifp); 1388 1389 for (i = 0; lagg_pflags[i].flag; i++) { 1390 error = lagg_setflag(lp, lagg_pflags[i].flag, 1391 status, lagg_pflags[i].func); 1392 if (error) 1393 return (error); 1394 } 1395 return (0); 1396 } 1397 1398 1399 #ifdef XXX /* not needed? */ 1400 static int 1401 lagg_output(struct ifnet *ifp, struct mbuf *m) 1402 { 1403 struct lagg_softc *sc = ifp->if_softc; 1404 int error, len, mcast; 1405 1406 len = m->m_pkthdr.len; 1407 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1408 1409 LAGG_RLOCK(sc); 1410 /* We need a Tx algorithm and at least one port */ 1411 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1412 LAGG_RUNLOCK(sc); 1413 m_freem(m); 1414 ifp->if_oerrors++; 1415 return (ENXIO); 1416 } 1417 1418 BPF_MTAP(ifp, m); 1419 1420 error = (*sc->sc_start)(sc, m); 1421 LAGG_RUNLOCK(sc); 1422 1423 if (error == 0) { 1424 IFNET_STAT_INC(ifp, opackets, 1); 1425 IFNET_STAT_INC(ifp, obytes, len); 1426 ifp->if_omcasts += mcast; 1427 } else 1428 ifp->if_oerrors++; 1429 1430 return error; 1431 } 1432 #endif 1433 1434 #ifdef XXX 1435 /* 1436 * The ifp->if_qflush entry point for lagg(4) is no-op. 1437 */ 1438 static void 1439 lagg_qflush(struct ifnet *ifp __unused) 1440 { 1441 } 1442 #endif 1443 static void 1444 lagg_input(struct ifnet *ifp, struct mbuf *m) 1445 { 1446 struct lagg_port *lp = ifp->if_lagg; 1447 struct lagg_softc *sc = lp->lp_softc; 1448 struct ifnet *scifp = sc->sc_ifp; 1449 1450 LAGG_RLOCK(sc); 1451 if ((scifp->if_flags & IFF_RUNNING) == 0 || 1452 (lp->lp_flags & LAGG_PORT_DISABLED) || 1453 sc->sc_proto == LAGG_PROTO_NONE) { 1454 LAGG_RUNLOCK(sc); 1455 m_freem(m); 1456 return; 1457 } 1458 1459 BPF_MTAP(scifp, m); 1460 m = (*sc->sc_input)(sc, lp, m); 1461 1462 LAGG_RUNLOCK(sc); 1463 1464 if (m != NULL) { 1465 IFNET_STAT_INC(ifp, ipackets, 1); 1466 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len); 1467 1468 if (scifp->if_flags & IFF_MONITOR) { 1469 m_freem(m); 1470 m = NULL; 1471 } 1472 ether_reinput_oncpu(scifp, m, REINPUT_RUNBPF); 1473 } 1474 } 1475 1476 static int 1477 lagg_media_change(struct ifnet *ifp) 1478 { 1479 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1480 1481 if (sc->sc_ifflags & IFF_DEBUG) 1482 kprintf("%s\n", __func__); 1483 /* Ignore */ 1484 return (0); 1485 } 1486 1487 static void 1488 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1489 { 1490 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1491 struct lagg_port *lp; 1492 1493 imr->ifm_status = IFM_AVALID; 1494 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1495 1496 LAGG_RLOCK(sc); 1497 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1498 if (LAGG_PORTACTIVE(lp)) 1499 imr->ifm_status |= IFM_ACTIVE; 1500 } 1501 LAGG_RUNLOCK(sc); 1502 } 1503 1504 static void 1505 lagg_linkstate(struct lagg_softc *sc) 1506 { 1507 struct lagg_port *lp; 1508 int new_link = LINK_STATE_DOWN; 1509 uint64_t speed; 1510 1511 /* Our link is considered up if at least one of our ports is active */ 1512 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1513 if (lp->lp_link_state == LINK_STATE_UP) { 1514 new_link = LINK_STATE_UP; 1515 break; 1516 } 1517 } 1518 if_link_state_change(sc->sc_ifp); 1519 1520 /* Update if_baudrate to reflect the max possible speed */ 1521 switch (sc->sc_proto) { 1522 case LAGG_PROTO_FAILOVER: 1523 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1524 sc->sc_primary->lp_ifp->if_baudrate : 0; 1525 break; 1526 case LAGG_PROTO_ROUNDROBIN: 1527 case LAGG_PROTO_LOADBALANCE: 1528 case LAGG_PROTO_ETHERCHANNEL: 1529 speed = 0; 1530 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1531 speed += lp->lp_ifp->if_baudrate; 1532 sc->sc_ifp->if_baudrate = speed; 1533 break; 1534 case LAGG_PROTO_LACP: 1535 /* LACP updates if_baudrate itself */ 1536 break; 1537 } 1538 } 1539 1540 #if XXX 1541 static void 1542 lagg_port_state(struct ifnet *ifp, int state) 1543 { 1544 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1545 struct lagg_softc *sc = NULL; 1546 1547 if (lp != NULL) 1548 sc = lp->lp_softc; 1549 if (sc == NULL) 1550 return; 1551 1552 LAGG_WLOCK(sc); 1553 lagg_linkstate(sc); 1554 if (sc->sc_linkstate != NULL) 1555 (*sc->sc_linkstate)(lp); 1556 LAGG_WUNLOCK(sc); 1557 } 1558 #endif 1559 1560 struct lagg_port * 1561 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1562 { 1563 struct lagg_port *lp_next, *rval = NULL; 1564 // int new_link = LINK_STATE_DOWN; 1565 1566 LAGG_RLOCK_ASSERT(sc); 1567 /* 1568 * Search a port which reports an active link state. 1569 */ 1570 1571 if (lp == NULL) 1572 goto search; 1573 if (LAGG_PORTACTIVE(lp)) { 1574 rval = lp; 1575 goto found; 1576 } 1577 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1578 LAGG_PORTACTIVE(lp_next)) { 1579 rval = lp_next; 1580 goto found; 1581 } 1582 1583 search: 1584 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1585 if (LAGG_PORTACTIVE(lp_next)) { 1586 rval = lp_next; 1587 goto found; 1588 } 1589 } 1590 1591 found: 1592 if (rval != NULL) { 1593 /* 1594 * The IEEE 802.1D standard assumes that a lagg with 1595 * multiple ports is always full duplex. This is valid 1596 * for load sharing laggs and if at least two links 1597 * are active. Unfortunately, checking the latter would 1598 * be too expensive at this point. 1599 XXX 1600 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1601 (sc->sc_count > 1)) 1602 new_link = LINK_STATE_FULL_DUPLEX; 1603 else 1604 new_link = rval->lp_link_state; 1605 */ 1606 } 1607 1608 return (rval); 1609 } 1610 1611 static const void * 1612 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf) 1613 { 1614 if (m->m_pkthdr.len < (off + len)) { 1615 return (NULL); 1616 } else if (m->m_len < (off + len)) { 1617 m_copydata(m, off, len, buf); 1618 return (buf); 1619 } 1620 return (mtod(m, char *) + off); 1621 } 1622 1623 static int 1624 lagg_sysctl_active(SYSCTL_HANDLER_ARGS) 1625 { 1626 struct lagg_softc *sc = (struct lagg_softc *)arg1; 1627 struct lagg_port *lp; 1628 int error; 1629 1630 /* LACP tracks active links automatically, the others do not */ 1631 if (sc->sc_proto != LAGG_PROTO_LACP) { 1632 sc->sc_active = 0; 1633 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1634 sc->sc_active += LAGG_PORTACTIVE(lp); 1635 } 1636 1637 error = sysctl_handle_int(oidp, &sc->sc_active, 0, req); 1638 if ((error) || (req->newptr == NULL)) 1639 return (error); 1640 1641 return (0); 1642 } 1643 1644 uint32_t 1645 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key) 1646 { 1647 uint16_t etype; 1648 uint32_t p = key; 1649 int off; 1650 struct ether_header *eh; 1651 const struct ether_vlan_header *vlan; 1652 #ifdef INET 1653 const struct ip *ip; 1654 const uint32_t *ports; 1655 int iphlen; 1656 #endif 1657 #ifdef INET6 1658 const struct ip6_hdr *ip6; 1659 uint32_t flow; 1660 #endif 1661 union { 1662 #ifdef INET 1663 struct ip ip; 1664 #endif 1665 #ifdef INET6 1666 struct ip6_hdr ip6; 1667 #endif 1668 struct ether_vlan_header vlan; 1669 uint32_t port; 1670 } buf; 1671 1672 1673 off = sizeof(*eh); 1674 if (m->m_len < off) 1675 goto out; 1676 eh = mtod(m, struct ether_header *); 1677 etype = ntohs(eh->ether_type); 1678 if (sc->sc_flags & LAGG_F_HASHL2) { 1679 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p); 1680 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p); 1681 } 1682 1683 /* Special handling for encapsulating VLAN frames */ 1684 #if XXX 1685 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) { 1686 p = hash32_buf(&m->m_pkthdr.ether_vtag, 1687 sizeof(m->m_pkthdr.ether_vtag), p); 1688 } else 1689 #endif 1690 if (etype == ETHERTYPE_VLAN) { 1691 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf); 1692 if (vlan == NULL) 1693 goto out; 1694 1695 if (sc->sc_flags & LAGG_F_HASHL2) 1696 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p); 1697 etype = ntohs(vlan->evl_proto); 1698 off += sizeof(*vlan) - sizeof(*eh); 1699 } 1700 1701 switch (etype) { 1702 #ifdef INET 1703 case ETHERTYPE_IP: 1704 ip = lagg_gethdr(m, off, sizeof(*ip), &buf); 1705 if (ip == NULL) 1706 goto out; 1707 1708 if (sc->sc_flags & LAGG_F_HASHL3) { 1709 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p); 1710 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p); 1711 } 1712 if (!(sc->sc_flags & LAGG_F_HASHL4)) 1713 break; 1714 switch (ip->ip_p) { 1715 case IPPROTO_TCP: 1716 case IPPROTO_UDP: 1717 iphlen = ip->ip_hl << 2; 1718 if (iphlen < sizeof(*ip)) 1719 break; 1720 off += iphlen; 1721 ports = lagg_gethdr(m, off, sizeof(*ports), &buf); 1722 if (ports == NULL) 1723 break; 1724 p = hash32_buf(ports, sizeof(*ports), p); 1725 break; 1726 } 1727 break; 1728 #endif 1729 #ifdef INET6 1730 case ETHERTYPE_IPV6: 1731 if (!(sc->sc_flags & LAGG_F_HASHL3)) 1732 break; 1733 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf); 1734 if (ip6 == NULL) 1735 goto out; 1736 1737 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p); 1738 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p); 1739 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK; 1740 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */ 1741 break; 1742 #endif 1743 } 1744 out: 1745 return (p); 1746 } 1747 1748 static void 1749 lagg_start(struct ifnet *ifp, struct ifaltq_subque *ifsq) 1750 { 1751 struct lagg_softc *sc = ifp->if_softc; 1752 struct mbuf *m; 1753 struct ifnet *ifp_p; 1754 struct netmsg_packet *nmp; 1755 lwkt_port_t p_port; 1756 1757 ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq); 1758 ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq); 1759 1760 if (((ifp->if_flags & IFF_RUNNING) == 0) 1761 || (sc->sc_proto == LAGG_PROTO_NONE) 1762 || (sc->sc_count == 0)) { 1763 ifsq_purge(ifsq); 1764 return; 1765 } 1766 1767 1768 LAGG_RLOCK(sc); 1769 for (;;) { 1770 m = ifsq_dequeue(ifsq); 1771 if (m == NULL){ 1772 break; 1773 } 1774 // Choose output port 1775 ifp_p = (*sc->sc_select_tx_port)(sc, m); 1776 1777 if (ifp_p == NULL) { 1778 ifsq_purge(ifsq); 1779 break; 1780 } 1781 p_port = netisr_cpuport( 1782 ifsq_get_cpuid(ifq_get_subq_default(&ifp_p->if_snd))); 1783 1784 BPF_MTAP(ifp, m); 1785 1786 nmp = &m->m_hdr.mh_netmsg; 1787 1788 netmsg_init(&nmp->base, NULL, &netisr_apanic_rport, 1789 0, lagg_start_dispatch); 1790 nmp->nm_packet = m; 1791 nmp->base.lmsg.u.ms_resultp = ifp_p; 1792 1793 lwkt_sendmsg(p_port, &nmp->base.lmsg); 1794 IFNET_STAT_INC(ifp, opackets, 1); 1795 } 1796 LAGG_RUNLOCK(sc); 1797 } 1798 1799 1800 static void 1801 lagg_start_dispatch(netmsg_t msg) 1802 { 1803 struct netmsg_packet *nmp = &msg->packet; 1804 struct mbuf *m; 1805 struct ifnet *ifp; 1806 struct altq_pktattr pktattr; 1807 1808 m = nmp->nm_packet; 1809 ifp = msg->lmsg.u.ms_resultp; 1810 1811 M_ASSERTPKTHDR(m); 1812 1813 /* Does altq mix with lacp? */ 1814 if (ifq_is_enabled(&ifp->if_snd)) 1815 altq_etherclassify(&ifp->if_snd, m, &pktattr); 1816 1817 ifq_dispatch(ifp, m, &pktattr); 1818 } 1819 1820 1821 int 1822 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1823 { 1824 struct altq_pktattr pktattr; 1825 1826 if (ifq_is_enabled(&ifp->if_snd)) 1827 altq_etherclassify(&ifp->if_snd, m, &pktattr); 1828 1829 ifq_dispatch(ifp, m, &pktattr); 1830 return 0; 1831 } 1832 1833 /* 1834 * Simple round robin aggregation 1835 */ 1836 static int 1837 lagg_rr_attach(struct lagg_softc *sc) 1838 { 1839 sc->sc_detach = lagg_rr_detach; 1840 sc->sc_input = lagg_rr_input; 1841 sc->sc_select_tx_port = lagg_rr_select_tx_port; 1842 sc->sc_port_create = NULL; 1843 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1844 sc->sc_seq = 0; 1845 1846 return (0); 1847 } 1848 1849 static int 1850 lagg_rr_detach(struct lagg_softc *sc) 1851 { 1852 return (0); 1853 } 1854 1855 static struct ifnet * 1856 lagg_rr_select_tx_port(struct lagg_softc *sc, struct mbuf *m) 1857 { 1858 struct lagg_port *lp; 1859 uint32_t p; 1860 1861 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1862 p %= sc->sc_count; 1863 lp = SLIST_FIRST(&sc->sc_ports); 1864 while (p--) 1865 lp = SLIST_NEXT(lp, lp_entries); 1866 1867 /* 1868 * Check the port's link state. This will return the next active 1869 * port if the link is down or the port is NULL. 1870 */ 1871 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1872 return (NULL); 1873 } 1874 1875 return (lp->lp_ifp); 1876 } 1877 1878 static struct mbuf * 1879 lagg_rr_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 * Active failover 1891 */ 1892 1893 static int 1894 lagg_fail_attach(struct lagg_softc *sc) 1895 { 1896 sc->sc_detach = lagg_fail_detach; 1897 sc->sc_select_tx_port = lagg_fail_select_tx_port; 1898 sc->sc_input = lagg_fail_input; 1899 sc->sc_port_create = NULL; 1900 sc->sc_port_destroy = NULL; 1901 1902 return (0); 1903 } 1904 1905 static int 1906 lagg_fail_detach(struct lagg_softc *sc) 1907 { 1908 return (0); 1909 } 1910 1911 struct ifnet * 1912 lagg_fail_select_tx_port(struct lagg_softc *sc, struct mbuf *m) 1913 { 1914 struct lagg_port *lp; 1915 1916 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) 1917 return NULL; 1918 1919 return lp->lp_ifp; 1920 } 1921 1922 static struct mbuf * 1923 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1924 { 1925 struct ifnet *ifp = sc->sc_ifp; 1926 struct lagg_port *tmp_tp; 1927 1928 if (lp == sc->sc_primary || lagg_failover_rx_all) { 1929 m->m_pkthdr.rcvif = ifp; 1930 return (m); 1931 } 1932 1933 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1934 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1935 /* 1936 * If tmp_tp is null, we've recieved a packet when all 1937 * our links are down. Weird, but process it anyways. 1938 */ 1939 if ((tmp_tp == NULL || tmp_tp == lp)) { 1940 m->m_pkthdr.rcvif = ifp; 1941 return (m); 1942 } 1943 } 1944 1945 m_freem(m); 1946 return (NULL); 1947 } 1948 1949 /* 1950 * Loadbalancing 1951 */ 1952 1953 static int 1954 lagg_lb_attach(struct lagg_softc *sc) 1955 { 1956 struct lagg_port *lp; 1957 struct lagg_lb *lb; 1958 1959 if ((lb = (struct lagg_lb *)kmalloc(sizeof(struct lagg_lb), 1960 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 1961 return (ENOMEM); 1962 1963 sc->sc_detach = lagg_lb_detach; 1964 sc->sc_select_tx_port = lagg_lb_select_tx_port; 1965 sc->sc_input = lagg_lb_input; 1966 sc->sc_port_create = lagg_lb_port_create; 1967 sc->sc_port_destroy = lagg_lb_port_destroy; 1968 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1969 1970 lb->lb_key = karc4random(); 1971 sc->sc_psc = (caddr_t)lb; 1972 1973 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1974 lagg_lb_port_create(lp); 1975 1976 return (0); 1977 } 1978 1979 static int 1980 lagg_lb_detach(struct lagg_softc *sc) 1981 { 1982 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1983 if (lb != NULL) 1984 kfree(lb, M_DEVBUF); 1985 return (0); 1986 } 1987 1988 static int 1989 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 1990 { 1991 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1992 struct lagg_port *lp_next; 1993 int i = 0; 1994 1995 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 1996 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1997 if (lp_next == lp) 1998 continue; 1999 if (i >= LAGG_MAX_PORTS) 2000 return (EINVAL); 2001 if (sc->sc_ifflags & IFF_DEBUG) 2002 kprintf("%s: port %s at index %d\n", 2003 sc->sc_ifname, lp_next->lp_ifname, i); 2004 lb->lb_ports[i++] = lp_next; 2005 } 2006 2007 return (0); 2008 } 2009 2010 static int 2011 lagg_lb_port_create(struct lagg_port *lp) 2012 { 2013 struct lagg_softc *sc = lp->lp_softc; 2014 return (lagg_lb_porttable(sc, NULL)); 2015 } 2016 2017 static void 2018 lagg_lb_port_destroy(struct lagg_port *lp) 2019 { 2020 struct lagg_softc *sc = lp->lp_softc; 2021 lagg_lb_porttable(sc, lp); 2022 } 2023 2024 2025 struct ifnet * 2026 lagg_lb_select_tx_port(struct lagg_softc *sc, struct mbuf *m) 2027 { 2028 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 2029 struct lagg_port *lp = NULL; 2030 uint32_t p = 0; 2031 2032 /* XXX 2033 if (sc->use_flowid && (m->m_flags & M_FLOWID)) 2034 p = m->m_pkthdr.flowid; 2035 else 2036 */ 2037 p = lagg_hashmbuf(sc, m, lb->lb_key); 2038 p %= sc->sc_count; 2039 lp = lb->lb_ports[p]; 2040 2041 /* 2042 * Check the port's link state. This will return the next active 2043 * port if the link is down or the port is NULL. 2044 */ 2045 if ((lp = lagg_link_active(sc, lp)) == NULL) 2046 return NULL; 2047 2048 return lp->lp_ifp; 2049 } 2050 2051 static struct mbuf * 2052 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2053 { 2054 struct ifnet *ifp = sc->sc_ifp; 2055 2056 /* Just pass in the packet to our lagg device */ 2057 m->m_pkthdr.rcvif = ifp; 2058 2059 return (m); 2060 } 2061 2062 /* 2063 * 802.3ad LACP 2064 */ 2065 static int 2066 lagg_lacp_attach(struct lagg_softc *sc) 2067 { 2068 struct lagg_port *lp; 2069 int error; 2070 2071 sc->sc_detach = lagg_lacp_detach; 2072 sc->sc_port_create = lacp_port_create; 2073 sc->sc_port_destroy = lacp_port_destroy; 2074 sc->sc_linkstate = lacp_linkstate; 2075 sc->sc_select_tx_port = lagg_lacp_select_tx_port; 2076 sc->sc_input = lagg_lacp_input; 2077 sc->sc_init = lacp_init; 2078 sc->sc_stop = lacp_stop; 2079 sc->sc_lladdr = lagg_lacp_lladdr; 2080 sc->sc_req = lacp_req; 2081 sc->sc_portreq = lacp_portreq; 2082 2083 error = lacp_attach(sc); 2084 if (error) 2085 return (error); 2086 2087 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2088 lacp_port_create(lp); 2089 2090 return (error); 2091 } 2092 2093 static int 2094 lagg_lacp_detach(struct lagg_softc *sc) 2095 { 2096 struct lagg_port *lp; 2097 int error; 2098 2099 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2100 lacp_port_destroy(lp); 2101 2102 /* unlocking is safe here */ 2103 LAGG_WUNLOCK(sc); 2104 error = lacp_detach(sc); 2105 LAGG_WLOCK(sc); 2106 2107 return (error); 2108 } 2109 2110 static void 2111 lagg_lacp_lladdr(struct lagg_softc *sc) 2112 { 2113 struct lagg_port *lp; 2114 2115 /* purge all the lacp ports */ 2116 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2117 lacp_port_destroy(lp); 2118 2119 /* add them back in */ 2120 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2121 lacp_port_create(lp); 2122 } 2123 2124 struct ifnet * 2125 lagg_lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m) 2126 { 2127 struct lagg_port *lp; 2128 lp = lacp_select_tx_port(sc, m); 2129 if (lp == NULL) 2130 return NULL; 2131 2132 return lp->lp_ifp; 2133 } 2134 2135 static struct mbuf * 2136 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2137 { 2138 struct ifnet *ifp = sc->sc_ifp; 2139 struct ether_header *eh; 2140 u_short etype; 2141 2142 eh = mtod(m, struct ether_header *); 2143 etype = ntohs(eh->ether_type); 2144 2145 /* Tap off LACP control messages */ 2146 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 2147 m = lacp_input(lp, m); 2148 if (m == NULL) 2149 return (NULL); 2150 } 2151 2152 /* 2153 * If the port is not collecting or not in the active aggregator then 2154 * free and return. 2155 */ 2156 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 2157 m_freem(m); 2158 return (NULL); 2159 } 2160 2161 m->m_pkthdr.rcvif = ifp; 2162 return (m); 2163 } 2164 2165 static void 2166 lagg_callout(void *arg) 2167 { 2168 struct lagg_softc *sc = (struct lagg_softc *)arg; 2169 #if XXX 2170 struct ifnet *ifp = sc->sc_ifp; 2171 2172 ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets); 2173 ifp->if_opackets = counter_u64_fetch(sc->sc_opackets); 2174 ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes); 2175 ifp->if_obytes = counter_u64_fetch(sc->sc_obytes); 2176 #endif 2177 2178 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 2179 } 2180