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, 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 0 /* 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 0 /* 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 lagg_input_p = NULL; 226 if_clone_detach(&lagg_cloner); 227 lockuninit(&lagg_list_lock); 228 break; 229 default: 230 return (EOPNOTSUPP); 231 } 232 return (0); 233 } 234 235 static moduledata_t lagg_mod = { 236 "if_lagg", 237 lagg_modevent, 238 0 239 }; 240 241 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 242 MODULE_VERSION(if_lagg, 1); 243 244 /* 245 * This routine is run via an vlan 246 * config EVENT 247 */ 248 static void 249 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 250 { 251 #if 0 /* XXX */ 252 struct lagg_softc *sc = ifp->if_softc; 253 struct lagg_port *lp; 254 255 if (ifp->if_softc != arg) /* Not our event */ 256 return; 257 258 LAGG_RLOCK(sc); 259 if (!SLIST_EMPTY(&sc->sc_ports)) { 260 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 261 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp); 262 } 263 LAGG_RUNLOCK(sc); 264 #endif 265 266 } 267 268 /* 269 * This routine is run via an vlan 270 * unconfig EVENT 271 */ 272 static void 273 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 274 { 275 struct lagg_softc *sc = ifp->if_softc; 276 /* 277 struct lagg_port *lp; 278 */ 279 280 if (ifp->if_softc != arg) /* Not our event */ 281 return; 282 283 LAGG_RLOCK(sc); 284 if (!SLIST_EMPTY(&sc->sc_ports)) { 285 #if 0 /* XXX */ 286 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 287 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp); 288 #endif 289 } 290 LAGG_RUNLOCK(sc); 291 } 292 293 static int 294 lagg_clone_create(struct if_clone *ifc, int unit, 295 caddr_t params __unused, caddr_t data __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_lk(&sc->sc_callout, &sc->sc_call_lock); 361 362 /* Initialise pseudo media types */ 363 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 364 lagg_media_status); 365 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 366 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 367 368 if_initname(ifp, laggname, unit); 369 ifp->if_softc = sc; 370 #if 0 /* XXX */ 371 ifp->if_transmit = lagg_transmit; 372 ifp->if_qflush = lagg_qflush; 373 #endif 374 ifp->if_mtu = ETHERMTU; 375 ifp->if_init = lagg_init; 376 ifp->if_ioctl = lagg_ioctl; 377 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 378 ifp->if_start = lagg_start; 379 ifp->if_type = IFT_ETHER; 380 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen); 381 ifq_set_ready(&ifp->if_snd); 382 ifp->if_hdrlen = ETHER_HDR_LEN; 383 384 #if 0 /* XXX */ 385 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS; 386 #endif 387 /* 388 * Attach as an ordinary ethernet device, children will be attached 389 * as special device IFT_IEEE8023ADLAG. 390 */ 391 392 ether_ifattach(ifp, eaddr, NULL); 393 394 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 395 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 396 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 397 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 398 399 /* Insert into the global list of laggs */ 400 lockmgr(&lagg_list_lock, LK_EXCLUSIVE); 401 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries); 402 lockmgr(&lagg_list_lock, LK_RELEASE); 403 404 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 405 406 return (0); 407 } 408 409 static int 410 lagg_clone_destroy(struct ifnet *ifp) 411 { 412 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 413 struct lagg_port *lp; 414 415 LAGG_WLOCK(sc); 416 417 lagg_stop(sc); 418 ifp->if_flags &= ~IFF_UP; 419 420 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 421 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 422 423 /* Shutdown and remove lagg ports */ 424 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) 425 lagg_port_destroy(lp, 1); 426 /* Unhook the aggregation protocol */ 427 if (sc->sc_detach != NULL) 428 (*sc->sc_detach)(sc); 429 430 LAGG_WUNLOCK(sc); 431 432 sysctl_ctx_free(&sc->ctx); 433 ifmedia_removeall(&sc->sc_media); 434 ether_ifdetach(ifp); 435 /* This ifp is part of lagg_softc, don't free it! */ 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 0 /* 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 < (NELEM(lagg_protos)); i++) { 1111 if (lagg_protos[i].ti_proto == ra->ra_proto) { 1112 if (sc->sc_ifflags & IFF_DEBUG) 1113 kprintf("%s: using proto %u\n", 1114 sc->sc_ifname, 1115 lagg_protos[i].ti_proto); 1116 sc->sc_proto = lagg_protos[i].ti_proto; 1117 if (sc->sc_proto != LAGG_PROTO_NONE) 1118 error = lagg_protos[i].ti_attach(sc); 1119 LAGG_WUNLOCK(sc); 1120 return (error); 1121 } 1122 } 1123 LAGG_WUNLOCK(sc); 1124 error = EPROTONOSUPPORT; 1125 break; 1126 case SIOCGLAGGFLAGS: 1127 rf->rf_flags = sc->sc_flags; 1128 break; 1129 case SIOCSLAGGHASH: 1130 error = priv_check(td, PRIV_NET_LAGG); 1131 if (error) 1132 break; 1133 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1134 error = EINVAL; 1135 break; 1136 } 1137 LAGG_WLOCK(sc); 1138 sc->sc_flags &= ~LAGG_F_HASHMASK; 1139 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK; 1140 LAGG_WUNLOCK(sc); 1141 break; 1142 case SIOCGLAGGPORT: 1143 /* 1144 * Release ifp serializers before ifnet_lock 1145 * to prevent lock order reversal. 1146 */ 1147 ifnet_deserialize_all(ifp); 1148 ifnet_lock(); 1149 if (rp->rp_portname[0] == '\0' || 1150 (tpif = ifunit(rp->rp_portname)) == NULL) { 1151 ifnet_unlock(); 1152 ifnet_serialize_all(ifp); 1153 error = EINVAL; 1154 break; 1155 } 1156 1157 LAGG_RLOCK(sc); 1158 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1159 lp->lp_softc != sc) { 1160 error = ENOENT; 1161 LAGG_RUNLOCK(sc); 1162 ifnet_unlock(); 1163 ifnet_serialize_all(ifp); 1164 break; 1165 } 1166 1167 lagg_port2req(lp, rp); 1168 LAGG_RUNLOCK(sc); 1169 ifnet_unlock(); 1170 ifnet_serialize_all(ifp); 1171 break; 1172 case SIOCSLAGGPORT: 1173 error = priv_check(td, PRIV_NET_LAGG); 1174 if (error) 1175 break; 1176 /* 1177 * Release ifp serializers before ifnet_lock 1178 * to prevent lock order reversal. 1179 */ 1180 ifnet_deserialize_all(ifp); 1181 ifnet_lock(); 1182 if (rp->rp_portname[0] == '\0' || 1183 (tpif = ifunit(rp->rp_portname)) == NULL) { 1184 ifnet_unlock(); 1185 ifnet_serialize_all(ifp); 1186 error = EINVAL; 1187 break; 1188 } 1189 LAGG_WLOCK(sc); 1190 error = lagg_port_create(sc, tpif); 1191 LAGG_WUNLOCK(sc); 1192 ifnet_unlock(); 1193 ifnet_serialize_all(ifp); 1194 break; 1195 case SIOCSLAGGDELPORT: 1196 error = priv_check(td, PRIV_NET_LAGG); 1197 if (error) 1198 break; 1199 /* 1200 * Release ifp serializers before ifnet_lock 1201 * to prevent lock order reversal. 1202 */ 1203 ifnet_deserialize_all(ifp); 1204 ifnet_lock(); 1205 if (rp->rp_portname[0] == '\0' || 1206 (tpif = ifunit(rp->rp_portname)) == NULL) { 1207 ifnet_unlock(); 1208 ifnet_serialize_all(ifp); 1209 error = EINVAL; 1210 break; 1211 } 1212 1213 LAGG_WLOCK(sc); 1214 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1215 lp->lp_softc != sc) { 1216 error = ENOENT; 1217 LAGG_WUNLOCK(sc); 1218 ifnet_unlock(); 1219 ifnet_serialize_all(ifp); 1220 break; 1221 } 1222 1223 error = lagg_port_destroy(lp, 1); 1224 LAGG_WUNLOCK(sc); 1225 ifnet_unlock(); 1226 ifnet_serialize_all(ifp); 1227 break; 1228 case SIOCSIFFLAGS: 1229 /* Set flags on ports too */ 1230 LAGG_WLOCK(sc); 1231 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1232 lagg_setflags(lp, 1); 1233 } 1234 LAGG_WUNLOCK(sc); 1235 1236 if (!(ifp->if_flags & IFF_UP) && 1237 (ifp->if_flags & IFF_RUNNING)) { 1238 /* 1239 * If interface is marked down and it is running, 1240 * then stop and disable it. 1241 */ 1242 LAGG_WLOCK(sc); 1243 lagg_stop(sc); 1244 LAGG_WUNLOCK(sc); 1245 } else if ((ifp->if_flags & IFF_UP) && 1246 !(ifp->if_flags & IFF_RUNNING)) { 1247 /* 1248 * If interface is marked up and it is stopped, then 1249 * start it. 1250 */ 1251 (*ifp->if_init)(sc); 1252 } 1253 break; 1254 case SIOCADDMULTI: 1255 case SIOCDELMULTI: 1256 LAGG_WLOCK(sc); 1257 error = lagg_ether_setmulti(sc); 1258 LAGG_WUNLOCK(sc); 1259 break; 1260 case SIOCSIFMEDIA: 1261 case SIOCGIFMEDIA: 1262 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1263 break; 1264 1265 case SIOCSIFCAP: 1266 case SIOCSIFMTU: 1267 /* Do not allow the MTU or caps to be directly changed */ 1268 error = EINVAL; 1269 break; 1270 1271 default: 1272 error = ether_ioctl(ifp, cmd, data); 1273 break; 1274 } 1275 return (error); 1276 } 1277 1278 static int 1279 lagg_ether_setmulti(struct lagg_softc *sc) 1280 { 1281 struct lagg_port *lp; 1282 1283 LAGG_WLOCK_ASSERT(sc); 1284 1285 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1286 /* First, remove any existing filter entries. */ 1287 lagg_ether_cmdmulti(lp, 0); 1288 /* copy all addresses from the lagg interface to the port */ 1289 lagg_ether_cmdmulti(lp, 1); 1290 } 1291 return (0); 1292 } 1293 1294 static int 1295 lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1296 { 1297 struct lagg_softc *sc = lp->lp_softc; 1298 struct ifnet *ifp = lp->lp_ifp; 1299 struct ifnet *scifp = sc->sc_ifp; 1300 struct lagg_mc *mc; 1301 struct ifmultiaddr *ifma, *rifma = NULL; 1302 struct sockaddr_dl sdl; 1303 int error; 1304 1305 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp); 1306 LAGG_WLOCK_ASSERT(sc); 1307 1308 bzero((char *)&sdl, sizeof(sdl)); 1309 sdl.sdl_len = sizeof(sdl); 1310 sdl.sdl_family = AF_LINK; 1311 sdl.sdl_type = IFT_ETHER; 1312 sdl.sdl_alen = ETHER_ADDR_LEN; 1313 sdl.sdl_index = ifp->if_index; 1314 1315 if (set) { 1316 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1317 if (ifma->ifma_addr->sa_family != AF_LINK) 1318 continue; 1319 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 1320 LLADDR(&sdl), ETHER_ADDR_LEN); 1321 1322 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 1323 if (error) 1324 return (error); 1325 mc = kmalloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1326 if (mc == NULL) 1327 return (ENOMEM); 1328 mc->mc_ifma = rifma; 1329 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1330 } 1331 } else { 1332 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1333 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1334 if_delmulti(ifp, (struct sockaddr *)mc->mc_ifma); 1335 kfree(mc, M_DEVBUF); 1336 } 1337 } 1338 return (0); 1339 } 1340 1341 /* Handle a ref counted flag that should be set on the lagg port as well */ 1342 static int 1343 lagg_setflag(struct lagg_port *lp, int flag, int status, 1344 int (*func)(struct ifnet *, int)) 1345 { 1346 struct lagg_softc *sc = lp->lp_softc; 1347 struct ifnet *scifp = sc->sc_ifp; 1348 struct ifnet *ifp = lp->lp_ifp; 1349 int error; 1350 1351 LAGG_WLOCK_ASSERT(sc); 1352 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp); 1353 1354 status = status ? (scifp->if_flags & flag) : 0; 1355 /* Now "status" contains the flag value or 0 */ 1356 1357 /* 1358 * See if recorded ports status is different from what 1359 * we want it to be. If it is, flip it. We record ports 1360 * status in lp_ifflags so that we won't clear ports flag 1361 * we haven't set. In fact, we don't clear or set ports 1362 * flags directly, but get or release references to them. 1363 * That's why we can be sure that recorded flags still are 1364 * in accord with actual ports flags. 1365 */ 1366 if (status != (lp->lp_ifflags & flag)) { 1367 error = (*func)(ifp, status); 1368 if (error) 1369 return (error); 1370 lp->lp_ifflags &= ~flag; 1371 lp->lp_ifflags |= status; 1372 } 1373 return (0); 1374 } 1375 1376 /* 1377 * Handle IFF_* flags that require certain changes on the lagg port 1378 * if "status" is true, update ports flags respective to the lagg 1379 * if "status" is false, forcedly clear the flags set on port. 1380 */ 1381 static int 1382 lagg_setflags(struct lagg_port *lp, int status) 1383 { 1384 int error, i; 1385 1386 ASSERT_IFNET_NOT_SERIALIZED_ALL(lp->lp_ifp); 1387 1388 for (i = 0; lagg_pflags[i].flag; i++) { 1389 error = lagg_setflag(lp, lagg_pflags[i].flag, 1390 status, lagg_pflags[i].func); 1391 if (error) 1392 return (error); 1393 } 1394 return (0); 1395 } 1396 1397 1398 #if 0 /* XXX not needed? */ 1399 static int 1400 lagg_output(struct ifnet *ifp, struct mbuf *m) 1401 { 1402 struct lagg_softc *sc = ifp->if_softc; 1403 int error, len, mcast; 1404 1405 len = m->m_pkthdr.len; 1406 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1407 1408 LAGG_RLOCK(sc); 1409 /* We need a Tx algorithm and at least one port */ 1410 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1411 LAGG_RUNLOCK(sc); 1412 m_freem(m); 1413 ifp->if_oerrors++; 1414 return (ENXIO); 1415 } 1416 1417 BPF_MTAP(ifp, m); 1418 1419 error = (*sc->sc_start)(sc, m); 1420 LAGG_RUNLOCK(sc); 1421 1422 if (error == 0) { 1423 IFNET_STAT_INC(ifp, opackets, 1); 1424 IFNET_STAT_INC(ifp, obytes, len); 1425 ifp->if_omcasts += mcast; 1426 } else 1427 ifp->if_oerrors++; 1428 1429 return error; 1430 } 1431 #endif 1432 1433 #if 0 /* XXX */ 1434 /* 1435 * The ifp->if_qflush entry point for lagg(4) is no-op. 1436 */ 1437 static void 1438 lagg_qflush(struct ifnet *ifp __unused) 1439 { 1440 } 1441 #endif 1442 static void 1443 lagg_input(struct ifnet *ifp, struct mbuf *m) 1444 { 1445 struct lagg_port *lp = ifp->if_lagg; 1446 struct lagg_softc *sc = lp->lp_softc; 1447 struct ifnet *scifp = sc->sc_ifp; 1448 1449 LAGG_RLOCK(sc); 1450 if ((scifp->if_flags & IFF_RUNNING) == 0 || 1451 (lp->lp_flags & LAGG_PORT_DISABLED) || 1452 sc->sc_proto == LAGG_PROTO_NONE) { 1453 LAGG_RUNLOCK(sc); 1454 m_freem(m); 1455 return; 1456 } 1457 1458 BPF_MTAP(scifp, m); 1459 m = (*sc->sc_input)(sc, lp, m); 1460 1461 LAGG_RUNLOCK(sc); 1462 1463 if (m != NULL) { 1464 IFNET_STAT_INC(ifp, ipackets, 1); 1465 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len); 1466 1467 if (scifp->if_flags & IFF_MONITOR) { 1468 m_freem(m); 1469 m = NULL; 1470 } 1471 ether_reinput_oncpu(scifp, m, REINPUT_RUNBPF); 1472 } 1473 } 1474 1475 static int 1476 lagg_media_change(struct ifnet *ifp) 1477 { 1478 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1479 1480 if (sc->sc_ifflags & IFF_DEBUG) 1481 kprintf("%s\n", __func__); 1482 /* Ignore */ 1483 return (0); 1484 } 1485 1486 static void 1487 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1488 { 1489 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1490 struct lagg_port *lp; 1491 1492 imr->ifm_status = IFM_AVALID; 1493 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1494 1495 LAGG_RLOCK(sc); 1496 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1497 if (LAGG_PORTACTIVE(lp)) 1498 imr->ifm_status |= IFM_ACTIVE; 1499 } 1500 LAGG_RUNLOCK(sc); 1501 } 1502 1503 static void 1504 lagg_linkstate(struct lagg_softc *sc) 1505 { 1506 struct lagg_port *lp; 1507 int new_link = LINK_STATE_DOWN; 1508 uint64_t speed; 1509 1510 /* Our link is considered up if at least one of our ports is active */ 1511 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1512 if (lp->lp_link_state == LINK_STATE_UP) { 1513 new_link = LINK_STATE_UP; 1514 break; 1515 } 1516 } 1517 if_link_state_change(sc->sc_ifp); 1518 1519 /* Update if_baudrate to reflect the max possible speed */ 1520 switch (sc->sc_proto) { 1521 case LAGG_PROTO_FAILOVER: 1522 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1523 sc->sc_primary->lp_ifp->if_baudrate : 0; 1524 break; 1525 case LAGG_PROTO_ROUNDROBIN: 1526 case LAGG_PROTO_LOADBALANCE: 1527 case LAGG_PROTO_ETHERCHANNEL: 1528 speed = 0; 1529 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1530 speed += lp->lp_ifp->if_baudrate; 1531 sc->sc_ifp->if_baudrate = speed; 1532 break; 1533 case LAGG_PROTO_LACP: 1534 /* LACP updates if_baudrate itself */ 1535 break; 1536 } 1537 } 1538 1539 #if 0 /* XXX */ 1540 static void 1541 lagg_port_state(struct ifnet *ifp, int state) 1542 { 1543 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1544 struct lagg_softc *sc = NULL; 1545 1546 if (lp != NULL) 1547 sc = lp->lp_softc; 1548 if (sc == NULL) 1549 return; 1550 1551 LAGG_WLOCK(sc); 1552 lagg_linkstate(sc); 1553 if (sc->sc_linkstate != NULL) 1554 (*sc->sc_linkstate)(lp); 1555 LAGG_WUNLOCK(sc); 1556 } 1557 #endif 1558 1559 struct lagg_port * 1560 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1561 { 1562 struct lagg_port *lp_next, *rval = NULL; 1563 // int new_link = LINK_STATE_DOWN; 1564 1565 LAGG_RLOCK_ASSERT(sc); 1566 /* 1567 * Search a port which reports an active link state. 1568 */ 1569 1570 if (lp == NULL) 1571 goto search; 1572 if (LAGG_PORTACTIVE(lp)) { 1573 rval = lp; 1574 goto found; 1575 } 1576 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1577 LAGG_PORTACTIVE(lp_next)) { 1578 rval = lp_next; 1579 goto found; 1580 } 1581 1582 search: 1583 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1584 if (LAGG_PORTACTIVE(lp_next)) { 1585 rval = lp_next; 1586 goto found; 1587 } 1588 } 1589 1590 found: 1591 if (rval != NULL) { 1592 /* 1593 * The IEEE 802.1D standard assumes that a lagg with 1594 * multiple ports is always full duplex. This is valid 1595 * for load sharing laggs and if at least two links 1596 * are active. Unfortunately, checking the latter would 1597 * be too expensive at this point. 1598 XXX 1599 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1600 (sc->sc_count > 1)) 1601 new_link = LINK_STATE_FULL_DUPLEX; 1602 else 1603 new_link = rval->lp_link_state; 1604 */ 1605 } 1606 1607 return (rval); 1608 } 1609 1610 static const void * 1611 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf) 1612 { 1613 if (m->m_pkthdr.len < (off + len)) { 1614 return (NULL); 1615 } else if (m->m_len < (off + len)) { 1616 m_copydata(m, off, len, buf); 1617 return (buf); 1618 } 1619 return (mtod(m, char *) + off); 1620 } 1621 1622 static int 1623 lagg_sysctl_active(SYSCTL_HANDLER_ARGS) 1624 { 1625 struct lagg_softc *sc = (struct lagg_softc *)arg1; 1626 struct lagg_port *lp; 1627 int error; 1628 1629 /* LACP tracks active links automatically, the others do not */ 1630 if (sc->sc_proto != LAGG_PROTO_LACP) { 1631 sc->sc_active = 0; 1632 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1633 sc->sc_active += LAGG_PORTACTIVE(lp); 1634 } 1635 1636 error = sysctl_handle_int(oidp, &sc->sc_active, 0, req); 1637 if ((error) || (req->newptr == NULL)) 1638 return (error); 1639 1640 return (0); 1641 } 1642 1643 uint32_t 1644 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key) 1645 { 1646 uint16_t etype; 1647 uint32_t p = key; 1648 int off; 1649 struct ether_header *eh; 1650 const struct ether_vlan_header *vlan; 1651 #ifdef INET 1652 const struct ip *ip; 1653 const uint32_t *ports; 1654 int iphlen; 1655 #endif 1656 #ifdef INET6 1657 const struct ip6_hdr *ip6; 1658 uint32_t flow; 1659 #endif 1660 union { 1661 #ifdef INET 1662 struct ip ip; 1663 #endif 1664 #ifdef INET6 1665 struct ip6_hdr ip6; 1666 #endif 1667 struct ether_vlan_header vlan; 1668 uint32_t port; 1669 } buf; 1670 1671 1672 off = sizeof(*eh); 1673 if (m->m_len < off) 1674 goto out; 1675 eh = mtod(m, struct ether_header *); 1676 etype = ntohs(eh->ether_type); 1677 if (sc->sc_flags & LAGG_F_HASHL2) { 1678 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p); 1679 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p); 1680 } 1681 1682 /* Special handling for encapsulating VLAN frames */ 1683 #if 0 /* XXX */ 1684 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) { 1685 p = hash32_buf(&m->m_pkthdr.ether_vtag, 1686 sizeof(m->m_pkthdr.ether_vtag), p); 1687 } else 1688 #endif 1689 if (etype == ETHERTYPE_VLAN) { 1690 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf); 1691 if (vlan == NULL) 1692 goto out; 1693 1694 if (sc->sc_flags & LAGG_F_HASHL2) 1695 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p); 1696 etype = ntohs(vlan->evl_proto); 1697 off += sizeof(*vlan) - sizeof(*eh); 1698 } 1699 1700 switch (etype) { 1701 #ifdef INET 1702 case ETHERTYPE_IP: 1703 ip = lagg_gethdr(m, off, sizeof(*ip), &buf); 1704 if (ip == NULL) 1705 goto out; 1706 1707 if (sc->sc_flags & LAGG_F_HASHL3) { 1708 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p); 1709 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p); 1710 } 1711 if (!(sc->sc_flags & LAGG_F_HASHL4)) 1712 break; 1713 switch (ip->ip_p) { 1714 case IPPROTO_TCP: 1715 case IPPROTO_UDP: 1716 iphlen = ip->ip_hl << 2; 1717 if (iphlen < sizeof(*ip)) 1718 break; 1719 off += iphlen; 1720 ports = lagg_gethdr(m, off, sizeof(*ports), &buf); 1721 if (ports == NULL) 1722 break; 1723 p = hash32_buf(ports, sizeof(*ports), p); 1724 break; 1725 } 1726 break; 1727 #endif 1728 #ifdef INET6 1729 case ETHERTYPE_IPV6: 1730 if (!(sc->sc_flags & LAGG_F_HASHL3)) 1731 break; 1732 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf); 1733 if (ip6 == NULL) 1734 goto out; 1735 1736 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p); 1737 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p); 1738 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK; 1739 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */ 1740 break; 1741 #endif 1742 } 1743 out: 1744 return (p); 1745 } 1746 1747 static void 1748 lagg_start(struct ifnet *ifp, struct ifaltq_subque *ifsq) 1749 { 1750 struct lagg_softc *sc = ifp->if_softc; 1751 struct mbuf *m; 1752 struct ifnet *ifp_p; 1753 struct netmsg_packet *nmp; 1754 lwkt_port_t p_port; 1755 1756 ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq); 1757 ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq); 1758 1759 if (((ifp->if_flags & IFF_RUNNING) == 0) 1760 || (sc->sc_proto == LAGG_PROTO_NONE) 1761 || (sc->sc_count == 0)) { 1762 ifsq_purge(ifsq); 1763 return; 1764 } 1765 1766 1767 LAGG_RLOCK(sc); 1768 for (;;) { 1769 m = ifsq_dequeue(ifsq); 1770 if (m == NULL){ 1771 break; 1772 } 1773 // Choose output port 1774 ifp_p = (*sc->sc_select_tx_port)(sc, m); 1775 1776 if (ifp_p == NULL) { 1777 ifsq_purge(ifsq); 1778 break; 1779 } 1780 p_port = netisr_cpuport( 1781 ifsq_get_cpuid(ifq_get_subq_default(&ifp_p->if_snd))); 1782 1783 BPF_MTAP(ifp, m); 1784 1785 nmp = &m->m_hdr.mh_netmsg; 1786 1787 netmsg_init(&nmp->base, NULL, &netisr_apanic_rport, 1788 0, lagg_start_dispatch); 1789 nmp->nm_packet = m; 1790 nmp->base.lmsg.u.ms_resultp = ifp_p; 1791 1792 lwkt_sendmsg(p_port, &nmp->base.lmsg); 1793 IFNET_STAT_INC(ifp, opackets, 1); 1794 } 1795 LAGG_RUNLOCK(sc); 1796 } 1797 1798 1799 static void 1800 lagg_start_dispatch(netmsg_t msg) 1801 { 1802 struct netmsg_packet *nmp = &msg->packet; 1803 struct mbuf *m; 1804 struct ifnet *ifp; 1805 struct altq_pktattr pktattr; 1806 1807 m = nmp->nm_packet; 1808 ifp = msg->lmsg.u.ms_resultp; 1809 1810 M_ASSERTPKTHDR(m); 1811 1812 /* Does altq mix with lacp? */ 1813 if (ifq_is_enabled(&ifp->if_snd)) 1814 altq_etherclassify(&ifp->if_snd, m, &pktattr); 1815 1816 ifq_dispatch(ifp, m, &pktattr); 1817 } 1818 1819 1820 int 1821 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1822 { 1823 struct altq_pktattr pktattr; 1824 1825 if (ifq_is_enabled(&ifp->if_snd)) 1826 altq_etherclassify(&ifp->if_snd, m, &pktattr); 1827 1828 ifq_dispatch(ifp, m, &pktattr); 1829 return 0; 1830 } 1831 1832 /* 1833 * Simple round robin aggregation 1834 */ 1835 static int 1836 lagg_rr_attach(struct lagg_softc *sc) 1837 { 1838 sc->sc_detach = lagg_rr_detach; 1839 sc->sc_input = lagg_rr_input; 1840 sc->sc_select_tx_port = lagg_rr_select_tx_port; 1841 sc->sc_port_create = NULL; 1842 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1843 sc->sc_seq = 0; 1844 1845 return (0); 1846 } 1847 1848 static int 1849 lagg_rr_detach(struct lagg_softc *sc) 1850 { 1851 return (0); 1852 } 1853 1854 static struct ifnet * 1855 lagg_rr_select_tx_port(struct lagg_softc *sc, struct mbuf *m) 1856 { 1857 struct lagg_port *lp; 1858 uint32_t p; 1859 1860 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1861 p %= sc->sc_count; 1862 lp = SLIST_FIRST(&sc->sc_ports); 1863 while (p--) 1864 lp = SLIST_NEXT(lp, lp_entries); 1865 1866 /* 1867 * Check the port's link state. This will return the next active 1868 * port if the link is down or the port is NULL. 1869 */ 1870 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1871 return (NULL); 1872 } 1873 1874 return (lp->lp_ifp); 1875 } 1876 1877 static struct mbuf * 1878 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1879 { 1880 struct ifnet *ifp = sc->sc_ifp; 1881 1882 /* Just pass in the packet to our lagg device */ 1883 m->m_pkthdr.rcvif = ifp; 1884 1885 return (m); 1886 } 1887 1888 /* 1889 * Active failover 1890 */ 1891 1892 static int 1893 lagg_fail_attach(struct lagg_softc *sc) 1894 { 1895 sc->sc_detach = lagg_fail_detach; 1896 sc->sc_select_tx_port = lagg_fail_select_tx_port; 1897 sc->sc_input = lagg_fail_input; 1898 sc->sc_port_create = NULL; 1899 sc->sc_port_destroy = NULL; 1900 1901 return (0); 1902 } 1903 1904 static int 1905 lagg_fail_detach(struct lagg_softc *sc) 1906 { 1907 return (0); 1908 } 1909 1910 struct ifnet * 1911 lagg_fail_select_tx_port(struct lagg_softc *sc, struct mbuf *m) 1912 { 1913 struct lagg_port *lp; 1914 1915 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) 1916 return NULL; 1917 1918 return lp->lp_ifp; 1919 } 1920 1921 static struct mbuf * 1922 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1923 { 1924 struct ifnet *ifp = sc->sc_ifp; 1925 struct lagg_port *tmp_tp; 1926 1927 if (lp == sc->sc_primary || lagg_failover_rx_all) { 1928 m->m_pkthdr.rcvif = ifp; 1929 return (m); 1930 } 1931 1932 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1933 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1934 /* 1935 * If tmp_tp is null, we've recieved a packet when all 1936 * our links are down. Weird, but process it anyways. 1937 */ 1938 if ((tmp_tp == NULL || tmp_tp == lp)) { 1939 m->m_pkthdr.rcvif = ifp; 1940 return (m); 1941 } 1942 } 1943 1944 m_freem(m); 1945 return (NULL); 1946 } 1947 1948 /* 1949 * Loadbalancing 1950 */ 1951 1952 static int 1953 lagg_lb_attach(struct lagg_softc *sc) 1954 { 1955 struct lagg_port *lp; 1956 struct lagg_lb *lb; 1957 1958 if ((lb = (struct lagg_lb *)kmalloc(sizeof(struct lagg_lb), 1959 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 1960 return (ENOMEM); 1961 1962 sc->sc_detach = lagg_lb_detach; 1963 sc->sc_select_tx_port = lagg_lb_select_tx_port; 1964 sc->sc_input = lagg_lb_input; 1965 sc->sc_port_create = lagg_lb_port_create; 1966 sc->sc_port_destroy = lagg_lb_port_destroy; 1967 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1968 1969 lb->lb_key = karc4random(); 1970 sc->sc_psc = (caddr_t)lb; 1971 1972 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1973 lagg_lb_port_create(lp); 1974 1975 return (0); 1976 } 1977 1978 static int 1979 lagg_lb_detach(struct lagg_softc *sc) 1980 { 1981 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1982 if (lb != NULL) 1983 kfree(lb, M_DEVBUF); 1984 return (0); 1985 } 1986 1987 static int 1988 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 1989 { 1990 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1991 struct lagg_port *lp_next; 1992 int i = 0; 1993 1994 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 1995 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1996 if (lp_next == lp) 1997 continue; 1998 if (i >= LAGG_MAX_PORTS) 1999 return (EINVAL); 2000 if (sc->sc_ifflags & IFF_DEBUG) 2001 kprintf("%s: port %s at index %d\n", 2002 sc->sc_ifname, lp_next->lp_ifname, i); 2003 lb->lb_ports[i++] = lp_next; 2004 } 2005 2006 return (0); 2007 } 2008 2009 static int 2010 lagg_lb_port_create(struct lagg_port *lp) 2011 { 2012 struct lagg_softc *sc = lp->lp_softc; 2013 return (lagg_lb_porttable(sc, NULL)); 2014 } 2015 2016 static void 2017 lagg_lb_port_destroy(struct lagg_port *lp) 2018 { 2019 struct lagg_softc *sc = lp->lp_softc; 2020 lagg_lb_porttable(sc, lp); 2021 } 2022 2023 2024 struct ifnet * 2025 lagg_lb_select_tx_port(struct lagg_softc *sc, struct mbuf *m) 2026 { 2027 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 2028 struct lagg_port *lp = NULL; 2029 uint32_t p = 0; 2030 2031 /* XXX 2032 if (sc->use_flowid && (m->m_flags & M_FLOWID)) 2033 p = m->m_pkthdr.flowid; 2034 else 2035 */ 2036 p = lagg_hashmbuf(sc, m, lb->lb_key); 2037 p %= sc->sc_count; 2038 lp = lb->lb_ports[p]; 2039 2040 /* 2041 * Check the port's link state. This will return the next active 2042 * port if the link is down or the port is NULL. 2043 */ 2044 if ((lp = lagg_link_active(sc, lp)) == NULL) 2045 return NULL; 2046 2047 return lp->lp_ifp; 2048 } 2049 2050 static struct mbuf * 2051 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2052 { 2053 struct ifnet *ifp = sc->sc_ifp; 2054 2055 /* Just pass in the packet to our lagg device */ 2056 m->m_pkthdr.rcvif = ifp; 2057 2058 return (m); 2059 } 2060 2061 /* 2062 * 802.3ad LACP 2063 */ 2064 static int 2065 lagg_lacp_attach(struct lagg_softc *sc) 2066 { 2067 struct lagg_port *lp; 2068 int error; 2069 2070 sc->sc_detach = lagg_lacp_detach; 2071 sc->sc_port_create = lacp_port_create; 2072 sc->sc_port_destroy = lacp_port_destroy; 2073 sc->sc_linkstate = lacp_linkstate; 2074 sc->sc_select_tx_port = lagg_lacp_select_tx_port; 2075 sc->sc_input = lagg_lacp_input; 2076 sc->sc_init = lacp_init; 2077 sc->sc_stop = lacp_stop; 2078 sc->sc_lladdr = lagg_lacp_lladdr; 2079 sc->sc_req = lacp_req; 2080 sc->sc_portreq = lacp_portreq; 2081 2082 error = lacp_attach(sc); 2083 if (error) 2084 return (error); 2085 2086 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2087 lacp_port_create(lp); 2088 2089 return (error); 2090 } 2091 2092 static int 2093 lagg_lacp_detach(struct lagg_softc *sc) 2094 { 2095 struct lagg_port *lp; 2096 int error; 2097 2098 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2099 lacp_port_destroy(lp); 2100 2101 /* unlocking is safe here */ 2102 LAGG_WUNLOCK(sc); 2103 error = lacp_detach(sc); 2104 LAGG_WLOCK(sc); 2105 2106 return (error); 2107 } 2108 2109 static void 2110 lagg_lacp_lladdr(struct lagg_softc *sc) 2111 { 2112 struct lagg_port *lp; 2113 2114 /* purge all the lacp ports */ 2115 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2116 lacp_port_destroy(lp); 2117 2118 /* add them back in */ 2119 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2120 lacp_port_create(lp); 2121 } 2122 2123 struct ifnet * 2124 lagg_lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m) 2125 { 2126 struct lagg_port *lp; 2127 lp = lacp_select_tx_port(sc, m); 2128 if (lp == NULL) 2129 return NULL; 2130 2131 return lp->lp_ifp; 2132 } 2133 2134 static struct mbuf * 2135 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2136 { 2137 struct ifnet *ifp = sc->sc_ifp; 2138 struct ether_header *eh; 2139 u_short etype; 2140 2141 eh = mtod(m, struct ether_header *); 2142 etype = ntohs(eh->ether_type); 2143 2144 /* Tap off LACP control messages */ 2145 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 2146 m = lacp_input(lp, m); 2147 if (m == NULL) 2148 return (NULL); 2149 } 2150 2151 /* 2152 * If the port is not collecting or not in the active aggregator then 2153 * free and return. 2154 */ 2155 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 2156 m_freem(m); 2157 return (NULL); 2158 } 2159 2160 m->m_pkthdr.rcvif = ifp; 2161 return (m); 2162 } 2163 2164 static void 2165 lagg_callout(void *arg) 2166 { 2167 struct lagg_softc *sc = (struct lagg_softc *)arg; 2168 #if 0 /* XXX */ 2169 struct ifnet *ifp = sc->sc_ifp; 2170 2171 ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets); 2172 ifp->if_opackets = counter_u64_fetch(sc->sc_opackets); 2173 ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes); 2174 ifp->if_obytes = counter_u64_fetch(sc->sc_obytes); 2175 #endif 2176 2177 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 2178 } 2179