1 /* $OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $ */ 2 3 /* 4 * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org> 5 * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org> 6 * 7 * Permission to use, copy, modify, and distribute this software for any 8 * purpose with or without fee is hereby granted, provided that the above 9 * copyright notice and this permission notice appear in all copies. 10 * 11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 18 */ 19 20 #include <sys/cdefs.h> 21 __FBSDID("$FreeBSD$"); 22 23 #include "opt_inet.h" 24 #include "opt_inet6.h" 25 26 #include <sys/param.h> 27 #include <sys/kernel.h> 28 #include <sys/malloc.h> 29 #include <sys/mbuf.h> 30 #include <sys/queue.h> 31 #include <sys/socket.h> 32 #include <sys/sockio.h> 33 #include <sys/sysctl.h> 34 #include <sys/module.h> 35 #include <sys/priv.h> 36 #include <sys/systm.h> 37 #include <sys/proc.h> 38 #include <sys/hash.h> 39 #include <sys/lock.h> 40 #include <sys/rwlock.h> 41 #include <sys/taskqueue.h> 42 #include <sys/eventhandler.h> 43 44 #include <net/ethernet.h> 45 #include <net/if.h> 46 #include <net/if_clone.h> 47 #include <net/if_arp.h> 48 #include <net/if_dl.h> 49 #include <net/if_llc.h> 50 #include <net/if_media.h> 51 #include <net/if_types.h> 52 #include <net/if_var.h> 53 #include <net/bpf.h> 54 55 #if defined(INET) || defined(INET6) 56 #include <netinet/in.h> 57 #endif 58 #ifdef INET 59 #include <netinet/in_systm.h> 60 #include <netinet/if_ether.h> 61 #include <netinet/ip.h> 62 #endif 63 64 #ifdef INET6 65 #include <netinet/ip6.h> 66 #endif 67 68 #include <net/if_vlan_var.h> 69 #include <net/if_lagg.h> 70 #include <net/ieee8023ad_lacp.h> 71 72 /* Special flags we should propagate to the lagg ports. */ 73 static struct { 74 int flag; 75 int (*func)(struct ifnet *, int); 76 } lagg_pflags[] = { 77 {IFF_PROMISC, ifpromisc}, 78 {IFF_ALLMULTI, if_allmulti}, 79 {0, NULL} 80 }; 81 82 SLIST_HEAD(__trhead, lagg_softc) lagg_list; /* list of laggs */ 83 static struct mtx lagg_list_mtx; 84 eventhandler_tag lagg_detach_cookie = NULL; 85 86 static int lagg_clone_create(struct if_clone *, int, caddr_t); 87 static void lagg_clone_destroy(struct ifnet *); 88 static struct if_clone *lagg_cloner; 89 static const char laggname[] = "lagg"; 90 91 static void lagg_lladdr(struct lagg_softc *, uint8_t *); 92 static void lagg_capabilities(struct lagg_softc *); 93 static void lagg_port_lladdr(struct lagg_port *, uint8_t *); 94 static void lagg_port_setlladdr(void *, int); 95 static int lagg_port_create(struct lagg_softc *, struct ifnet *); 96 static int lagg_port_destroy(struct lagg_port *, int); 97 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); 98 static void lagg_linkstate(struct lagg_softc *); 99 static void lagg_port_state(struct ifnet *, int); 100 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); 101 static int lagg_port_output(struct ifnet *, struct mbuf *, 102 const struct sockaddr *, struct route *); 103 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 104 #ifdef LAGG_PORT_STACKING 105 static int lagg_port_checkstacking(struct lagg_softc *); 106 #endif 107 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 108 static void lagg_init(void *); 109 static void lagg_stop(struct lagg_softc *); 110 static int lagg_ioctl(struct ifnet *, u_long, caddr_t); 111 static int lagg_ether_setmulti(struct lagg_softc *); 112 static int lagg_ether_cmdmulti(struct lagg_port *, int); 113 static int lagg_setflag(struct lagg_port *, int, int, 114 int (*func)(struct ifnet *, int)); 115 static int lagg_setflags(struct lagg_port *, int status); 116 static int lagg_transmit(struct ifnet *, struct mbuf *); 117 static void lagg_qflush(struct ifnet *); 118 static int lagg_media_change(struct ifnet *); 119 static void lagg_media_status(struct ifnet *, struct ifmediareq *); 120 static struct lagg_port *lagg_link_active(struct lagg_softc *, 121 struct lagg_port *); 122 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *); 123 124 /* Simple round robin */ 125 static int lagg_rr_attach(struct lagg_softc *); 126 static int lagg_rr_detach(struct lagg_softc *); 127 static int lagg_rr_start(struct lagg_softc *, struct mbuf *); 128 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 129 struct mbuf *); 130 131 /* Active failover */ 132 static int lagg_fail_attach(struct lagg_softc *); 133 static int lagg_fail_detach(struct lagg_softc *); 134 static int lagg_fail_start(struct lagg_softc *, struct mbuf *); 135 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 136 struct mbuf *); 137 138 /* Loadbalancing */ 139 static int lagg_lb_attach(struct lagg_softc *); 140 static int lagg_lb_detach(struct lagg_softc *); 141 static int lagg_lb_port_create(struct lagg_port *); 142 static void lagg_lb_port_destroy(struct lagg_port *); 143 static int lagg_lb_start(struct lagg_softc *, struct mbuf *); 144 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 145 struct mbuf *); 146 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 147 148 /* 802.3ad LACP */ 149 static int lagg_lacp_attach(struct lagg_softc *); 150 static int lagg_lacp_detach(struct lagg_softc *); 151 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); 152 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 153 struct mbuf *); 154 static void lagg_lacp_lladdr(struct lagg_softc *); 155 156 static void lagg_callout(void *); 157 158 /* lagg protocol table */ 159 static const struct { 160 int ti_proto; 161 int (*ti_attach)(struct lagg_softc *); 162 } lagg_protos[] = { 163 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach }, 164 { LAGG_PROTO_FAILOVER, lagg_fail_attach }, 165 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach }, 166 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach }, 167 { LAGG_PROTO_LACP, lagg_lacp_attach }, 168 { LAGG_PROTO_NONE, NULL } 169 }; 170 171 SYSCTL_DECL(_net_link); 172 static SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 173 "Link Aggregation"); 174 175 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */ 176 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW, 177 &lagg_failover_rx_all, 0, 178 "Accept input from any interface in a failover lagg"); 179 static int def_use_flowid = 1; /* Default value for using M_FLOWID */ 180 TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid); 181 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RW, 182 &def_use_flowid, 0, 183 "Default setting for using flow id for load sharing"); 184 185 static int 186 lagg_modevent(module_t mod, int type, void *data) 187 { 188 189 switch (type) { 190 case MOD_LOAD: 191 mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF); 192 SLIST_INIT(&lagg_list); 193 lagg_cloner = if_clone_simple(laggname, lagg_clone_create, 194 lagg_clone_destroy, 0); 195 lagg_input_p = lagg_input; 196 lagg_linkstate_p = lagg_port_state; 197 lagg_detach_cookie = EVENTHANDLER_REGISTER( 198 ifnet_departure_event, lagg_port_ifdetach, NULL, 199 EVENTHANDLER_PRI_ANY); 200 break; 201 case MOD_UNLOAD: 202 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 203 lagg_detach_cookie); 204 if_clone_detach(lagg_cloner); 205 lagg_input_p = NULL; 206 lagg_linkstate_p = NULL; 207 mtx_destroy(&lagg_list_mtx); 208 break; 209 default: 210 return (EOPNOTSUPP); 211 } 212 return (0); 213 } 214 215 static moduledata_t lagg_mod = { 216 "if_lagg", 217 lagg_modevent, 218 0 219 }; 220 221 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 222 MODULE_VERSION(if_lagg, 1); 223 224 /* 225 * This routine is run via an vlan 226 * config EVENT 227 */ 228 static void 229 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 230 { 231 struct lagg_softc *sc = ifp->if_softc; 232 struct lagg_port *lp; 233 234 if (ifp->if_softc != arg) /* Not our event */ 235 return; 236 237 LAGG_RLOCK(sc); 238 if (!SLIST_EMPTY(&sc->sc_ports)) { 239 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 240 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 241 } 242 LAGG_RUNLOCK(sc); 243 } 244 245 /* 246 * This routine is run via an vlan 247 * unconfig EVENT 248 */ 249 static void 250 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 251 { 252 struct lagg_softc *sc = ifp->if_softc; 253 struct lagg_port *lp; 254 255 if (ifp->if_softc != arg) /* Not our event */ 256 return; 257 258 LAGG_RLOCK(sc); 259 if (!SLIST_EMPTY(&sc->sc_ports)) { 260 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 261 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 262 } 263 LAGG_RUNLOCK(sc); 264 } 265 266 static int 267 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) 268 { 269 struct lagg_softc *sc; 270 struct ifnet *ifp; 271 int i, error = 0; 272 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ 273 struct sysctl_oid *oid; 274 char num[14]; /* sufficient for 32 bits */ 275 276 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO); 277 ifp = sc->sc_ifp = if_alloc(IFT_ETHER); 278 if (ifp == NULL) { 279 free(sc, M_DEVBUF); 280 return (ENOSPC); 281 } 282 283 sc->sc_ipackets = counter_u64_alloc(M_WAITOK); 284 sc->sc_opackets = counter_u64_alloc(M_WAITOK); 285 sc->sc_ibytes = counter_u64_alloc(M_WAITOK); 286 sc->sc_obytes = counter_u64_alloc(M_WAITOK); 287 288 sysctl_ctx_init(&sc->ctx); 289 snprintf(num, sizeof(num), "%u", unit); 290 sc->use_flowid = def_use_flowid; 291 oid = SYSCTL_ADD_NODE(&sc->ctx, &SYSCTL_NODE_CHILDREN(_net_link, lagg), 292 OID_AUTO, num, CTLFLAG_RD, NULL, ""); 293 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 294 "use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid, sc->use_flowid, 295 "Use flow id for load sharing"); 296 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 297 "count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count, 298 "Total number of ports"); 299 /* Hash all layers by default */ 300 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4; 301 302 sc->sc_proto = LAGG_PROTO_NONE; 303 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) { 304 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) { 305 sc->sc_proto = lagg_protos[i].ti_proto; 306 if ((error = lagg_protos[i].ti_attach(sc)) != 0) { 307 if_free(ifp); 308 free(sc, M_DEVBUF); 309 return (error); 310 } 311 break; 312 } 313 } 314 LAGG_LOCK_INIT(sc); 315 SLIST_INIT(&sc->sc_ports); 316 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc); 317 callout_init_rw(&sc->sc_callout, &sc->sc_mtx, CALLOUT_SHAREDLOCK); 318 319 /* Initialise pseudo media types */ 320 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 321 lagg_media_status); 322 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 323 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 324 325 if_initname(ifp, laggname, unit); 326 ifp->if_softc = sc; 327 ifp->if_transmit = lagg_transmit; 328 ifp->if_qflush = lagg_qflush; 329 ifp->if_init = lagg_init; 330 ifp->if_ioctl = lagg_ioctl; 331 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 332 333 /* 334 * Attach as an ordinary ethernet device, children will be attached 335 * as special device IFT_IEEE8023ADLAG. 336 */ 337 ether_ifattach(ifp, eaddr); 338 339 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 340 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 341 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 342 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 343 344 /* Insert into the global list of laggs */ 345 mtx_lock(&lagg_list_mtx); 346 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries); 347 mtx_unlock(&lagg_list_mtx); 348 349 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 350 351 return (0); 352 } 353 354 static void 355 lagg_clone_destroy(struct ifnet *ifp) 356 { 357 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 358 struct lagg_port *lp; 359 360 LAGG_WLOCK(sc); 361 362 lagg_stop(sc); 363 ifp->if_flags &= ~IFF_UP; 364 365 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 366 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 367 368 /* Shutdown and remove lagg ports */ 369 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) 370 lagg_port_destroy(lp, 1); 371 /* Unhook the aggregation protocol */ 372 if (sc->sc_detach != NULL) 373 (*sc->sc_detach)(sc); 374 375 LAGG_WUNLOCK(sc); 376 377 sysctl_ctx_free(&sc->ctx); 378 ifmedia_removeall(&sc->sc_media); 379 ether_ifdetach(ifp); 380 if_free(ifp); 381 382 callout_drain(&sc->sc_callout); 383 counter_u64_free(sc->sc_ipackets); 384 counter_u64_free(sc->sc_opackets); 385 counter_u64_free(sc->sc_ibytes); 386 counter_u64_free(sc->sc_obytes); 387 388 mtx_lock(&lagg_list_mtx); 389 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries); 390 mtx_unlock(&lagg_list_mtx); 391 392 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task); 393 LAGG_LOCK_DESTROY(sc); 394 free(sc, M_DEVBUF); 395 } 396 397 static void 398 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr) 399 { 400 struct ifnet *ifp = sc->sc_ifp; 401 402 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 403 return; 404 405 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN); 406 /* Let the protocol know the MAC has changed */ 407 if (sc->sc_lladdr != NULL) 408 (*sc->sc_lladdr)(sc); 409 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 410 } 411 412 static void 413 lagg_capabilities(struct lagg_softc *sc) 414 { 415 struct lagg_port *lp; 416 int cap = ~0, ena = ~0; 417 u_long hwa = ~0UL; 418 419 LAGG_WLOCK_ASSERT(sc); 420 421 /* Get capabilities from the lagg ports */ 422 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 423 cap &= lp->lp_ifp->if_capabilities; 424 ena &= lp->lp_ifp->if_capenable; 425 hwa &= lp->lp_ifp->if_hwassist; 426 } 427 cap = (cap == ~0 ? 0 : cap); 428 ena = (ena == ~0 ? 0 : ena); 429 hwa = (hwa == ~0 ? 0 : hwa); 430 431 if (sc->sc_ifp->if_capabilities != cap || 432 sc->sc_ifp->if_capenable != ena || 433 sc->sc_ifp->if_hwassist != hwa) { 434 sc->sc_ifp->if_capabilities = cap; 435 sc->sc_ifp->if_capenable = ena; 436 sc->sc_ifp->if_hwassist = hwa; 437 getmicrotime(&sc->sc_ifp->if_lastchange); 438 439 if (sc->sc_ifflags & IFF_DEBUG) 440 if_printf(sc->sc_ifp, 441 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 442 } 443 } 444 445 static void 446 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr) 447 { 448 struct lagg_softc *sc = lp->lp_softc; 449 struct ifnet *ifp = lp->lp_ifp; 450 struct lagg_llq *llq; 451 int pending = 0; 452 453 LAGG_WLOCK_ASSERT(sc); 454 455 if (lp->lp_detaching || 456 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 457 return; 458 459 /* Check to make sure its not already queued to be changed */ 460 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 461 if (llq->llq_ifp == ifp) { 462 pending = 1; 463 break; 464 } 465 } 466 467 if (!pending) { 468 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT); 469 if (llq == NULL) /* XXX what to do */ 470 return; 471 } 472 473 /* Update the lladdr even if pending, it may have changed */ 474 llq->llq_ifp = ifp; 475 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN); 476 477 if (!pending) 478 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries); 479 480 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task); 481 } 482 483 /* 484 * Set the interface MAC address from a taskqueue to avoid a LOR. 485 */ 486 static void 487 lagg_port_setlladdr(void *arg, int pending) 488 { 489 struct lagg_softc *sc = (struct lagg_softc *)arg; 490 struct lagg_llq *llq, *head; 491 struct ifnet *ifp; 492 int error; 493 494 /* Grab a local reference of the queue and remove it from the softc */ 495 LAGG_WLOCK(sc); 496 head = SLIST_FIRST(&sc->sc_llq_head); 497 SLIST_FIRST(&sc->sc_llq_head) = NULL; 498 LAGG_WUNLOCK(sc); 499 500 /* 501 * Traverse the queue and set the lladdr on each ifp. It is safe to do 502 * unlocked as we have the only reference to it. 503 */ 504 for (llq = head; llq != NULL; llq = head) { 505 ifp = llq->llq_ifp; 506 507 /* Set the link layer address */ 508 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN); 509 if (error) 510 printf("%s: setlladdr failed on %s\n", __func__, 511 ifp->if_xname); 512 513 head = SLIST_NEXT(llq, llq_entries); 514 free(llq, M_DEVBUF); 515 } 516 } 517 518 static int 519 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 520 { 521 struct lagg_softc *sc_ptr; 522 struct lagg_port *lp; 523 int error = 0; 524 525 LAGG_WLOCK_ASSERT(sc); 526 527 /* Limit the maximal number of lagg ports */ 528 if (sc->sc_count >= LAGG_MAX_PORTS) 529 return (ENOSPC); 530 531 /* Check if port has already been associated to a lagg */ 532 if (ifp->if_lagg != NULL) { 533 /* Port is already in the current lagg? */ 534 lp = (struct lagg_port *)ifp->if_lagg; 535 if (lp->lp_softc == sc) 536 return (EEXIST); 537 return (EBUSY); 538 } 539 540 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 541 if (ifp->if_type != IFT_ETHER) 542 return (EPROTONOSUPPORT); 543 544 /* Allow the first Ethernet member to define the MTU */ 545 if (SLIST_EMPTY(&sc->sc_ports)) 546 sc->sc_ifp->if_mtu = ifp->if_mtu; 547 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 548 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 549 ifp->if_xname); 550 return (EINVAL); 551 } 552 553 if ((lp = malloc(sizeof(struct lagg_port), 554 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 555 return (ENOMEM); 556 557 /* Check if port is a stacked lagg */ 558 mtx_lock(&lagg_list_mtx); 559 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) { 560 if (ifp == sc_ptr->sc_ifp) { 561 mtx_unlock(&lagg_list_mtx); 562 free(lp, M_DEVBUF); 563 return (EINVAL); 564 /* XXX disable stacking for the moment, its untested */ 565 #ifdef LAGG_PORT_STACKING 566 lp->lp_flags |= LAGG_PORT_STACK; 567 if (lagg_port_checkstacking(sc_ptr) >= 568 LAGG_MAX_STACKING) { 569 mtx_unlock(&lagg_list_mtx); 570 free(lp, M_DEVBUF); 571 return (E2BIG); 572 } 573 #endif 574 } 575 } 576 mtx_unlock(&lagg_list_mtx); 577 578 /* Change the interface type */ 579 lp->lp_iftype = ifp->if_type; 580 ifp->if_type = IFT_IEEE8023ADLAG; 581 ifp->if_lagg = lp; 582 lp->lp_ioctl = ifp->if_ioctl; 583 ifp->if_ioctl = lagg_port_ioctl; 584 lp->lp_output = ifp->if_output; 585 ifp->if_output = lagg_port_output; 586 587 lp->lp_ifp = ifp; 588 lp->lp_softc = sc; 589 590 /* Save port link layer address */ 591 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 592 593 if (SLIST_EMPTY(&sc->sc_ports)) { 594 sc->sc_primary = lp; 595 lagg_lladdr(sc, IF_LLADDR(ifp)); 596 } else { 597 /* Update link layer address for this port */ 598 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp)); 599 } 600 601 /* Insert into the list of ports */ 602 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 603 sc->sc_count++; 604 605 /* Update lagg capabilities */ 606 lagg_capabilities(sc); 607 lagg_linkstate(sc); 608 609 /* Add multicast addresses and interface flags to this port */ 610 lagg_ether_cmdmulti(lp, 1); 611 lagg_setflags(lp, 1); 612 613 if (sc->sc_port_create != NULL) 614 error = (*sc->sc_port_create)(lp); 615 if (error) { 616 /* remove the port again, without calling sc_port_destroy */ 617 lagg_port_destroy(lp, 0); 618 return (error); 619 } 620 621 return (error); 622 } 623 624 #ifdef LAGG_PORT_STACKING 625 static int 626 lagg_port_checkstacking(struct lagg_softc *sc) 627 { 628 struct lagg_softc *sc_ptr; 629 struct lagg_port *lp; 630 int m = 0; 631 632 LAGG_WLOCK_ASSERT(sc); 633 634 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 635 if (lp->lp_flags & LAGG_PORT_STACK) { 636 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 637 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 638 } 639 } 640 641 return (m + 1); 642 } 643 #endif 644 645 static int 646 lagg_port_destroy(struct lagg_port *lp, int runpd) 647 { 648 struct lagg_softc *sc = lp->lp_softc; 649 struct lagg_port *lp_ptr; 650 struct lagg_llq *llq; 651 struct ifnet *ifp = lp->lp_ifp; 652 653 LAGG_WLOCK_ASSERT(sc); 654 655 if (runpd && sc->sc_port_destroy != NULL) 656 (*sc->sc_port_destroy)(lp); 657 658 /* 659 * Remove multicast addresses and interface flags from this port and 660 * reset the MAC address, skip if the interface is being detached. 661 */ 662 if (!lp->lp_detaching) { 663 lagg_ether_cmdmulti(lp, 0); 664 lagg_setflags(lp, 0); 665 lagg_port_lladdr(lp, lp->lp_lladdr); 666 } 667 668 /* Restore interface */ 669 ifp->if_type = lp->lp_iftype; 670 ifp->if_ioctl = lp->lp_ioctl; 671 ifp->if_output = lp->lp_output; 672 ifp->if_lagg = NULL; 673 674 /* Finally, remove the port from the lagg */ 675 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 676 sc->sc_count--; 677 678 /* Update the primary interface */ 679 if (lp == sc->sc_primary) { 680 uint8_t lladdr[ETHER_ADDR_LEN]; 681 682 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) { 683 bzero(&lladdr, ETHER_ADDR_LEN); 684 } else { 685 bcopy(lp_ptr->lp_lladdr, 686 lladdr, ETHER_ADDR_LEN); 687 } 688 lagg_lladdr(sc, lladdr); 689 sc->sc_primary = lp_ptr; 690 691 /* Update link layer address for each port */ 692 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 693 lagg_port_lladdr(lp_ptr, lladdr); 694 } 695 696 /* Remove any pending lladdr changes from the queue */ 697 if (lp->lp_detaching) { 698 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 699 if (llq->llq_ifp == ifp) { 700 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq, 701 llq_entries); 702 free(llq, M_DEVBUF); 703 break; /* Only appears once */ 704 } 705 } 706 } 707 708 if (lp->lp_ifflags) 709 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 710 711 free(lp, M_DEVBUF); 712 713 /* Update lagg capabilities */ 714 lagg_capabilities(sc); 715 lagg_linkstate(sc); 716 717 return (0); 718 } 719 720 static int 721 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 722 { 723 struct lagg_reqport *rp = (struct lagg_reqport *)data; 724 struct lagg_softc *sc; 725 struct lagg_port *lp = NULL; 726 int error = 0; 727 728 /* Should be checked by the caller */ 729 if (ifp->if_type != IFT_IEEE8023ADLAG || 730 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 731 goto fallback; 732 733 switch (cmd) { 734 case SIOCGLAGGPORT: 735 if (rp->rp_portname[0] == '\0' || 736 ifunit(rp->rp_portname) != ifp) { 737 error = EINVAL; 738 break; 739 } 740 741 LAGG_RLOCK(sc); 742 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 743 error = ENOENT; 744 LAGG_RUNLOCK(sc); 745 break; 746 } 747 748 lagg_port2req(lp, rp); 749 LAGG_RUNLOCK(sc); 750 break; 751 752 case SIOCSIFCAP: 753 if (lp->lp_ioctl == NULL) { 754 error = EINVAL; 755 break; 756 } 757 error = (*lp->lp_ioctl)(ifp, cmd, data); 758 if (error) 759 break; 760 761 /* Update lagg interface capabilities */ 762 LAGG_WLOCK(sc); 763 lagg_capabilities(sc); 764 LAGG_WUNLOCK(sc); 765 break; 766 767 case SIOCSIFMTU: 768 /* Do not allow the MTU to be changed once joined */ 769 error = EINVAL; 770 break; 771 772 default: 773 goto fallback; 774 } 775 776 return (error); 777 778 fallback: 779 if (lp->lp_ioctl != NULL) 780 return ((*lp->lp_ioctl)(ifp, cmd, data)); 781 782 return (EINVAL); 783 } 784 785 /* 786 * For direct output to child ports. 787 */ 788 static int 789 lagg_port_output(struct ifnet *ifp, struct mbuf *m, 790 const struct sockaddr *dst, struct route *ro) 791 { 792 struct lagg_port *lp = ifp->if_lagg; 793 794 switch (dst->sa_family) { 795 case pseudo_AF_HDRCMPLT: 796 case AF_UNSPEC: 797 return ((*lp->lp_output)(ifp, m, dst, ro)); 798 } 799 800 /* drop any other frames */ 801 m_freem(m); 802 return (ENETDOWN); 803 } 804 805 static void 806 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 807 { 808 struct lagg_port *lp; 809 struct lagg_softc *sc; 810 811 if ((lp = ifp->if_lagg) == NULL) 812 return; 813 /* If the ifnet is just being renamed, don't do anything. */ 814 if (ifp->if_flags & IFF_RENAMING) 815 return; 816 817 sc = lp->lp_softc; 818 819 LAGG_WLOCK(sc); 820 lp->lp_detaching = 1; 821 lagg_port_destroy(lp, 1); 822 LAGG_WUNLOCK(sc); 823 } 824 825 static void 826 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 827 { 828 struct lagg_softc *sc = lp->lp_softc; 829 830 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 831 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 832 rp->rp_prio = lp->lp_prio; 833 rp->rp_flags = lp->lp_flags; 834 if (sc->sc_portreq != NULL) 835 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc); 836 837 /* Add protocol specific flags */ 838 switch (sc->sc_proto) { 839 case LAGG_PROTO_FAILOVER: 840 if (lp == sc->sc_primary) 841 rp->rp_flags |= LAGG_PORT_MASTER; 842 if (lp == lagg_link_active(sc, sc->sc_primary)) 843 rp->rp_flags |= LAGG_PORT_ACTIVE; 844 break; 845 846 case LAGG_PROTO_ROUNDROBIN: 847 case LAGG_PROTO_LOADBALANCE: 848 case LAGG_PROTO_ETHERCHANNEL: 849 if (LAGG_PORTACTIVE(lp)) 850 rp->rp_flags |= LAGG_PORT_ACTIVE; 851 break; 852 853 case LAGG_PROTO_LACP: 854 /* LACP has a different definition of active */ 855 if (lacp_isactive(lp)) 856 rp->rp_flags |= LAGG_PORT_ACTIVE; 857 if (lacp_iscollecting(lp)) 858 rp->rp_flags |= LAGG_PORT_COLLECTING; 859 if (lacp_isdistributing(lp)) 860 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 861 break; 862 } 863 864 } 865 866 static void 867 lagg_init(void *xsc) 868 { 869 struct lagg_softc *sc = (struct lagg_softc *)xsc; 870 struct lagg_port *lp; 871 struct ifnet *ifp = sc->sc_ifp; 872 873 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 874 return; 875 876 LAGG_WLOCK(sc); 877 878 ifp->if_drv_flags |= IFF_DRV_RUNNING; 879 /* Update the port lladdrs */ 880 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 881 lagg_port_lladdr(lp, IF_LLADDR(ifp)); 882 883 if (sc->sc_init != NULL) 884 (*sc->sc_init)(sc); 885 886 LAGG_WUNLOCK(sc); 887 } 888 889 static void 890 lagg_stop(struct lagg_softc *sc) 891 { 892 struct ifnet *ifp = sc->sc_ifp; 893 894 LAGG_WLOCK_ASSERT(sc); 895 896 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 897 return; 898 899 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 900 901 if (sc->sc_stop != NULL) 902 (*sc->sc_stop)(sc); 903 } 904 905 static int 906 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 907 { 908 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 909 struct lagg_reqall *ra = (struct lagg_reqall *)data; 910 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 911 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 912 struct ifreq *ifr = (struct ifreq *)data; 913 struct lagg_port *lp; 914 struct ifnet *tpif; 915 struct thread *td = curthread; 916 char *buf, *outbuf; 917 int count, buflen, len, error = 0; 918 919 bzero(&rpbuf, sizeof(rpbuf)); 920 921 switch (cmd) { 922 case SIOCGLAGG: 923 LAGG_RLOCK(sc); 924 count = 0; 925 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 926 count++; 927 buflen = count * sizeof(struct lagg_reqport); 928 LAGG_RUNLOCK(sc); 929 930 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 931 932 LAGG_RLOCK(sc); 933 ra->ra_proto = sc->sc_proto; 934 if (sc->sc_req != NULL) 935 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc); 936 937 count = 0; 938 buf = outbuf; 939 len = min(ra->ra_size, buflen); 940 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 941 if (len < sizeof(rpbuf)) 942 break; 943 944 lagg_port2req(lp, &rpbuf); 945 memcpy(buf, &rpbuf, sizeof(rpbuf)); 946 count++; 947 buf += sizeof(rpbuf); 948 len -= sizeof(rpbuf); 949 } 950 LAGG_RUNLOCK(sc); 951 ra->ra_ports = count; 952 ra->ra_size = count * sizeof(rpbuf); 953 error = copyout(outbuf, ra->ra_port, ra->ra_size); 954 free(outbuf, M_TEMP); 955 break; 956 case SIOCSLAGG: 957 error = priv_check(td, PRIV_NET_LAGG); 958 if (error) 959 break; 960 if (ra->ra_proto >= LAGG_PROTO_MAX) { 961 error = EPROTONOSUPPORT; 962 break; 963 } 964 LAGG_WLOCK(sc); 965 if (sc->sc_proto != LAGG_PROTO_NONE) { 966 /* Reset protocol first in case detach unlocks */ 967 sc->sc_proto = LAGG_PROTO_NONE; 968 error = sc->sc_detach(sc); 969 sc->sc_detach = NULL; 970 sc->sc_start = NULL; 971 sc->sc_input = NULL; 972 sc->sc_port_create = NULL; 973 sc->sc_port_destroy = NULL; 974 sc->sc_linkstate = NULL; 975 sc->sc_init = NULL; 976 sc->sc_stop = NULL; 977 sc->sc_lladdr = NULL; 978 sc->sc_req = NULL; 979 sc->sc_portreq = NULL; 980 } else if (sc->sc_input != NULL) { 981 /* Still detaching */ 982 error = EBUSY; 983 } 984 if (error != 0) { 985 LAGG_WUNLOCK(sc); 986 break; 987 } 988 for (int i = 0; i < (sizeof(lagg_protos) / 989 sizeof(lagg_protos[0])); i++) { 990 if (lagg_protos[i].ti_proto == ra->ra_proto) { 991 if (sc->sc_ifflags & IFF_DEBUG) 992 printf("%s: using proto %u\n", 993 sc->sc_ifname, 994 lagg_protos[i].ti_proto); 995 sc->sc_proto = lagg_protos[i].ti_proto; 996 if (sc->sc_proto != LAGG_PROTO_NONE) 997 error = lagg_protos[i].ti_attach(sc); 998 LAGG_WUNLOCK(sc); 999 return (error); 1000 } 1001 } 1002 LAGG_WUNLOCK(sc); 1003 error = EPROTONOSUPPORT; 1004 break; 1005 case SIOCGLAGGFLAGS: 1006 rf->rf_flags = sc->sc_flags; 1007 break; 1008 case SIOCSLAGGHASH: 1009 error = priv_check(td, PRIV_NET_LAGG); 1010 if (error) 1011 break; 1012 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1013 error = EINVAL; 1014 break; 1015 } 1016 LAGG_WLOCK(sc); 1017 sc->sc_flags &= ~LAGG_F_HASHMASK; 1018 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK; 1019 LAGG_WUNLOCK(sc); 1020 break; 1021 case SIOCGLAGGPORT: 1022 if (rp->rp_portname[0] == '\0' || 1023 (tpif = ifunit(rp->rp_portname)) == NULL) { 1024 error = EINVAL; 1025 break; 1026 } 1027 1028 LAGG_RLOCK(sc); 1029 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1030 lp->lp_softc != sc) { 1031 error = ENOENT; 1032 LAGG_RUNLOCK(sc); 1033 break; 1034 } 1035 1036 lagg_port2req(lp, rp); 1037 LAGG_RUNLOCK(sc); 1038 break; 1039 case SIOCSLAGGPORT: 1040 error = priv_check(td, PRIV_NET_LAGG); 1041 if (error) 1042 break; 1043 if (rp->rp_portname[0] == '\0' || 1044 (tpif = ifunit(rp->rp_portname)) == NULL) { 1045 error = EINVAL; 1046 break; 1047 } 1048 LAGG_WLOCK(sc); 1049 error = lagg_port_create(sc, tpif); 1050 LAGG_WUNLOCK(sc); 1051 break; 1052 case SIOCSLAGGDELPORT: 1053 error = priv_check(td, PRIV_NET_LAGG); 1054 if (error) 1055 break; 1056 if (rp->rp_portname[0] == '\0' || 1057 (tpif = ifunit(rp->rp_portname)) == NULL) { 1058 error = EINVAL; 1059 break; 1060 } 1061 1062 LAGG_WLOCK(sc); 1063 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1064 lp->lp_softc != sc) { 1065 error = ENOENT; 1066 LAGG_WUNLOCK(sc); 1067 break; 1068 } 1069 1070 error = lagg_port_destroy(lp, 1); 1071 LAGG_WUNLOCK(sc); 1072 break; 1073 case SIOCSIFFLAGS: 1074 /* Set flags on ports too */ 1075 LAGG_WLOCK(sc); 1076 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1077 lagg_setflags(lp, 1); 1078 } 1079 LAGG_WUNLOCK(sc); 1080 1081 if (!(ifp->if_flags & IFF_UP) && 1082 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1083 /* 1084 * If interface is marked down and it is running, 1085 * then stop and disable it. 1086 */ 1087 LAGG_WLOCK(sc); 1088 lagg_stop(sc); 1089 LAGG_WUNLOCK(sc); 1090 } else if ((ifp->if_flags & IFF_UP) && 1091 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1092 /* 1093 * If interface is marked up and it is stopped, then 1094 * start it. 1095 */ 1096 (*ifp->if_init)(sc); 1097 } 1098 break; 1099 case SIOCADDMULTI: 1100 case SIOCDELMULTI: 1101 LAGG_WLOCK(sc); 1102 error = lagg_ether_setmulti(sc); 1103 LAGG_WUNLOCK(sc); 1104 break; 1105 case SIOCSIFMEDIA: 1106 case SIOCGIFMEDIA: 1107 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1108 break; 1109 1110 case SIOCSIFCAP: 1111 case SIOCSIFMTU: 1112 /* Do not allow the MTU or caps to be directly changed */ 1113 error = EINVAL; 1114 break; 1115 1116 default: 1117 error = ether_ioctl(ifp, cmd, data); 1118 break; 1119 } 1120 return (error); 1121 } 1122 1123 static int 1124 lagg_ether_setmulti(struct lagg_softc *sc) 1125 { 1126 struct lagg_port *lp; 1127 1128 LAGG_WLOCK_ASSERT(sc); 1129 1130 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1131 /* First, remove any existing filter entries. */ 1132 lagg_ether_cmdmulti(lp, 0); 1133 /* copy all addresses from the lagg interface to the port */ 1134 lagg_ether_cmdmulti(lp, 1); 1135 } 1136 return (0); 1137 } 1138 1139 static int 1140 lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1141 { 1142 struct lagg_softc *sc = lp->lp_softc; 1143 struct ifnet *ifp = lp->lp_ifp; 1144 struct ifnet *scifp = sc->sc_ifp; 1145 struct lagg_mc *mc; 1146 struct ifmultiaddr *ifma, *rifma = NULL; 1147 struct sockaddr_dl sdl; 1148 int error; 1149 1150 LAGG_WLOCK_ASSERT(sc); 1151 1152 bzero((char *)&sdl, sizeof(sdl)); 1153 sdl.sdl_len = sizeof(sdl); 1154 sdl.sdl_family = AF_LINK; 1155 sdl.sdl_type = IFT_ETHER; 1156 sdl.sdl_alen = ETHER_ADDR_LEN; 1157 sdl.sdl_index = ifp->if_index; 1158 1159 if (set) { 1160 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1161 if (ifma->ifma_addr->sa_family != AF_LINK) 1162 continue; 1163 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 1164 LLADDR(&sdl), ETHER_ADDR_LEN); 1165 1166 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 1167 if (error) 1168 return (error); 1169 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1170 if (mc == NULL) 1171 return (ENOMEM); 1172 mc->mc_ifma = rifma; 1173 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1174 } 1175 } else { 1176 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1177 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1178 if_delmulti_ifma(mc->mc_ifma); 1179 free(mc, M_DEVBUF); 1180 } 1181 } 1182 return (0); 1183 } 1184 1185 /* Handle a ref counted flag that should be set on the lagg port as well */ 1186 static int 1187 lagg_setflag(struct lagg_port *lp, int flag, int status, 1188 int (*func)(struct ifnet *, int)) 1189 { 1190 struct lagg_softc *sc = lp->lp_softc; 1191 struct ifnet *scifp = sc->sc_ifp; 1192 struct ifnet *ifp = lp->lp_ifp; 1193 int error; 1194 1195 LAGG_WLOCK_ASSERT(sc); 1196 1197 status = status ? (scifp->if_flags & flag) : 0; 1198 /* Now "status" contains the flag value or 0 */ 1199 1200 /* 1201 * See if recorded ports status is different from what 1202 * we want it to be. If it is, flip it. We record ports 1203 * status in lp_ifflags so that we won't clear ports flag 1204 * we haven't set. In fact, we don't clear or set ports 1205 * flags directly, but get or release references to them. 1206 * That's why we can be sure that recorded flags still are 1207 * in accord with actual ports flags. 1208 */ 1209 if (status != (lp->lp_ifflags & flag)) { 1210 error = (*func)(ifp, status); 1211 if (error) 1212 return (error); 1213 lp->lp_ifflags &= ~flag; 1214 lp->lp_ifflags |= status; 1215 } 1216 return (0); 1217 } 1218 1219 /* 1220 * Handle IFF_* flags that require certain changes on the lagg port 1221 * if "status" is true, update ports flags respective to the lagg 1222 * if "status" is false, forcedly clear the flags set on port. 1223 */ 1224 static int 1225 lagg_setflags(struct lagg_port *lp, int status) 1226 { 1227 int error, i; 1228 1229 for (i = 0; lagg_pflags[i].flag; i++) { 1230 error = lagg_setflag(lp, lagg_pflags[i].flag, 1231 status, lagg_pflags[i].func); 1232 if (error) 1233 return (error); 1234 } 1235 return (0); 1236 } 1237 1238 static int 1239 lagg_transmit(struct ifnet *ifp, struct mbuf *m) 1240 { 1241 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1242 int error, len, mcast; 1243 1244 len = m->m_pkthdr.len; 1245 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1246 1247 LAGG_RLOCK(sc); 1248 /* We need a Tx algorithm and at least one port */ 1249 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1250 LAGG_RUNLOCK(sc); 1251 m_freem(m); 1252 ifp->if_oerrors++; 1253 return (ENXIO); 1254 } 1255 1256 ETHER_BPF_MTAP(ifp, m); 1257 1258 error = (*sc->sc_start)(sc, m); 1259 LAGG_RUNLOCK(sc); 1260 1261 if (error == 0) { 1262 counter_u64_add(sc->sc_opackets, 1); 1263 counter_u64_add(sc->sc_obytes, len); 1264 ifp->if_omcasts += mcast; 1265 } else 1266 ifp->if_oerrors++; 1267 1268 return (error); 1269 } 1270 1271 /* 1272 * The ifp->if_qflush entry point for lagg(4) is no-op. 1273 */ 1274 static void 1275 lagg_qflush(struct ifnet *ifp __unused) 1276 { 1277 } 1278 1279 static struct mbuf * 1280 lagg_input(struct ifnet *ifp, struct mbuf *m) 1281 { 1282 struct lagg_port *lp = ifp->if_lagg; 1283 struct lagg_softc *sc = lp->lp_softc; 1284 struct ifnet *scifp = sc->sc_ifp; 1285 1286 LAGG_RLOCK(sc); 1287 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1288 (lp->lp_flags & LAGG_PORT_DISABLED) || 1289 sc->sc_proto == LAGG_PROTO_NONE) { 1290 LAGG_RUNLOCK(sc); 1291 m_freem(m); 1292 return (NULL); 1293 } 1294 1295 ETHER_BPF_MTAP(scifp, m); 1296 1297 m = (*sc->sc_input)(sc, lp, m); 1298 1299 if (m != NULL) { 1300 counter_u64_add(sc->sc_ipackets, 1); 1301 counter_u64_add(sc->sc_ibytes, m->m_pkthdr.len); 1302 1303 if (scifp->if_flags & IFF_MONITOR) { 1304 m_freem(m); 1305 m = NULL; 1306 } 1307 } 1308 1309 LAGG_RUNLOCK(sc); 1310 return (m); 1311 } 1312 1313 static int 1314 lagg_media_change(struct ifnet *ifp) 1315 { 1316 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1317 1318 if (sc->sc_ifflags & IFF_DEBUG) 1319 printf("%s\n", __func__); 1320 1321 /* Ignore */ 1322 return (0); 1323 } 1324 1325 static void 1326 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1327 { 1328 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1329 struct lagg_port *lp; 1330 1331 imr->ifm_status = IFM_AVALID; 1332 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1333 1334 LAGG_RLOCK(sc); 1335 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1336 if (LAGG_PORTACTIVE(lp)) 1337 imr->ifm_status |= IFM_ACTIVE; 1338 } 1339 LAGG_RUNLOCK(sc); 1340 } 1341 1342 static void 1343 lagg_linkstate(struct lagg_softc *sc) 1344 { 1345 struct lagg_port *lp; 1346 int new_link = LINK_STATE_DOWN; 1347 uint64_t speed; 1348 1349 /* Our link is considered up if at least one of our ports is active */ 1350 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1351 if (lp->lp_link_state == LINK_STATE_UP) { 1352 new_link = LINK_STATE_UP; 1353 break; 1354 } 1355 } 1356 if_link_state_change(sc->sc_ifp, new_link); 1357 1358 /* Update if_baudrate to reflect the max possible speed */ 1359 switch (sc->sc_proto) { 1360 case LAGG_PROTO_FAILOVER: 1361 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1362 sc->sc_primary->lp_ifp->if_baudrate : 0; 1363 break; 1364 case LAGG_PROTO_ROUNDROBIN: 1365 case LAGG_PROTO_LOADBALANCE: 1366 case LAGG_PROTO_ETHERCHANNEL: 1367 speed = 0; 1368 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1369 speed += lp->lp_ifp->if_baudrate; 1370 sc->sc_ifp->if_baudrate = speed; 1371 break; 1372 case LAGG_PROTO_LACP: 1373 /* LACP updates if_baudrate itself */ 1374 break; 1375 } 1376 } 1377 1378 static void 1379 lagg_port_state(struct ifnet *ifp, int state) 1380 { 1381 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1382 struct lagg_softc *sc = NULL; 1383 1384 if (lp != NULL) 1385 sc = lp->lp_softc; 1386 if (sc == NULL) 1387 return; 1388 1389 LAGG_WLOCK(sc); 1390 lagg_linkstate(sc); 1391 if (sc->sc_linkstate != NULL) 1392 (*sc->sc_linkstate)(lp); 1393 LAGG_WUNLOCK(sc); 1394 } 1395 1396 struct lagg_port * 1397 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1398 { 1399 struct lagg_port *lp_next, *rval = NULL; 1400 // int new_link = LINK_STATE_DOWN; 1401 1402 LAGG_RLOCK_ASSERT(sc); 1403 /* 1404 * Search a port which reports an active link state. 1405 */ 1406 1407 if (lp == NULL) 1408 goto search; 1409 if (LAGG_PORTACTIVE(lp)) { 1410 rval = lp; 1411 goto found; 1412 } 1413 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1414 LAGG_PORTACTIVE(lp_next)) { 1415 rval = lp_next; 1416 goto found; 1417 } 1418 1419 search: 1420 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1421 if (LAGG_PORTACTIVE(lp_next)) { 1422 rval = lp_next; 1423 goto found; 1424 } 1425 } 1426 1427 found: 1428 if (rval != NULL) { 1429 /* 1430 * The IEEE 802.1D standard assumes that a lagg with 1431 * multiple ports is always full duplex. This is valid 1432 * for load sharing laggs and if at least two links 1433 * are active. Unfortunately, checking the latter would 1434 * be too expensive at this point. 1435 XXX 1436 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1437 (sc->sc_count > 1)) 1438 new_link = LINK_STATE_FULL_DUPLEX; 1439 else 1440 new_link = rval->lp_link_state; 1441 */ 1442 } 1443 1444 return (rval); 1445 } 1446 1447 static const void * 1448 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf) 1449 { 1450 if (m->m_pkthdr.len < (off + len)) { 1451 return (NULL); 1452 } else if (m->m_len < (off + len)) { 1453 m_copydata(m, off, len, buf); 1454 return (buf); 1455 } 1456 return (mtod(m, char *) + off); 1457 } 1458 1459 uint32_t 1460 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key) 1461 { 1462 uint16_t etype; 1463 uint32_t p = key; 1464 int off; 1465 struct ether_header *eh; 1466 const struct ether_vlan_header *vlan; 1467 #ifdef INET 1468 const struct ip *ip; 1469 const uint32_t *ports; 1470 int iphlen; 1471 #endif 1472 #ifdef INET6 1473 const struct ip6_hdr *ip6; 1474 uint32_t flow; 1475 #endif 1476 union { 1477 #ifdef INET 1478 struct ip ip; 1479 #endif 1480 #ifdef INET6 1481 struct ip6_hdr ip6; 1482 #endif 1483 struct ether_vlan_header vlan; 1484 uint32_t port; 1485 } buf; 1486 1487 1488 off = sizeof(*eh); 1489 if (m->m_len < off) 1490 goto out; 1491 eh = mtod(m, struct ether_header *); 1492 etype = ntohs(eh->ether_type); 1493 if (sc->sc_flags & LAGG_F_HASHL2) { 1494 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p); 1495 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p); 1496 } 1497 1498 /* Special handling for encapsulating VLAN frames */ 1499 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) { 1500 p = hash32_buf(&m->m_pkthdr.ether_vtag, 1501 sizeof(m->m_pkthdr.ether_vtag), p); 1502 } else if (etype == ETHERTYPE_VLAN) { 1503 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf); 1504 if (vlan == NULL) 1505 goto out; 1506 1507 if (sc->sc_flags & LAGG_F_HASHL2) 1508 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p); 1509 etype = ntohs(vlan->evl_proto); 1510 off += sizeof(*vlan) - sizeof(*eh); 1511 } 1512 1513 switch (etype) { 1514 #ifdef INET 1515 case ETHERTYPE_IP: 1516 ip = lagg_gethdr(m, off, sizeof(*ip), &buf); 1517 if (ip == NULL) 1518 goto out; 1519 1520 if (sc->sc_flags & LAGG_F_HASHL3) { 1521 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p); 1522 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p); 1523 } 1524 if (!(sc->sc_flags & LAGG_F_HASHL4)) 1525 break; 1526 switch (ip->ip_p) { 1527 case IPPROTO_TCP: 1528 case IPPROTO_UDP: 1529 case IPPROTO_SCTP: 1530 iphlen = ip->ip_hl << 2; 1531 if (iphlen < sizeof(*ip)) 1532 break; 1533 off += iphlen; 1534 ports = lagg_gethdr(m, off, sizeof(*ports), &buf); 1535 if (ports == NULL) 1536 break; 1537 p = hash32_buf(ports, sizeof(*ports), p); 1538 break; 1539 } 1540 break; 1541 #endif 1542 #ifdef INET6 1543 case ETHERTYPE_IPV6: 1544 if (!(sc->sc_flags & LAGG_F_HASHL3)) 1545 break; 1546 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf); 1547 if (ip6 == NULL) 1548 goto out; 1549 1550 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p); 1551 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p); 1552 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK; 1553 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */ 1554 break; 1555 #endif 1556 } 1557 out: 1558 return (p); 1559 } 1560 1561 int 1562 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1563 { 1564 1565 return (ifp->if_transmit)(ifp, m); 1566 } 1567 1568 /* 1569 * Simple round robin aggregation 1570 */ 1571 1572 static int 1573 lagg_rr_attach(struct lagg_softc *sc) 1574 { 1575 sc->sc_detach = lagg_rr_detach; 1576 sc->sc_start = lagg_rr_start; 1577 sc->sc_input = lagg_rr_input; 1578 sc->sc_port_create = NULL; 1579 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1580 sc->sc_seq = 0; 1581 1582 return (0); 1583 } 1584 1585 static int 1586 lagg_rr_detach(struct lagg_softc *sc) 1587 { 1588 return (0); 1589 } 1590 1591 static int 1592 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 1593 { 1594 struct lagg_port *lp; 1595 uint32_t p; 1596 1597 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1598 p %= sc->sc_count; 1599 lp = SLIST_FIRST(&sc->sc_ports); 1600 while (p--) 1601 lp = SLIST_NEXT(lp, lp_entries); 1602 1603 /* 1604 * Check the port's link state. This will return the next active 1605 * port if the link is down or the port is NULL. 1606 */ 1607 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1608 m_freem(m); 1609 return (ENOENT); 1610 } 1611 1612 /* Send mbuf */ 1613 return (lagg_enqueue(lp->lp_ifp, m)); 1614 } 1615 1616 static struct mbuf * 1617 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1618 { 1619 struct ifnet *ifp = sc->sc_ifp; 1620 1621 /* Just pass in the packet to our lagg device */ 1622 m->m_pkthdr.rcvif = ifp; 1623 1624 return (m); 1625 } 1626 1627 /* 1628 * Active failover 1629 */ 1630 1631 static int 1632 lagg_fail_attach(struct lagg_softc *sc) 1633 { 1634 sc->sc_detach = lagg_fail_detach; 1635 sc->sc_start = lagg_fail_start; 1636 sc->sc_input = lagg_fail_input; 1637 sc->sc_port_create = NULL; 1638 sc->sc_port_destroy = NULL; 1639 1640 return (0); 1641 } 1642 1643 static int 1644 lagg_fail_detach(struct lagg_softc *sc) 1645 { 1646 return (0); 1647 } 1648 1649 static int 1650 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 1651 { 1652 struct lagg_port *lp; 1653 1654 /* Use the master port if active or the next available port */ 1655 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 1656 m_freem(m); 1657 return (ENOENT); 1658 } 1659 1660 /* Send mbuf */ 1661 return (lagg_enqueue(lp->lp_ifp, m)); 1662 } 1663 1664 static struct mbuf * 1665 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1666 { 1667 struct ifnet *ifp = sc->sc_ifp; 1668 struct lagg_port *tmp_tp; 1669 1670 if (lp == sc->sc_primary || lagg_failover_rx_all) { 1671 m->m_pkthdr.rcvif = ifp; 1672 return (m); 1673 } 1674 1675 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1676 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1677 /* 1678 * If tmp_tp is null, we've recieved a packet when all 1679 * our links are down. Weird, but process it anyways. 1680 */ 1681 if ((tmp_tp == NULL || tmp_tp == lp)) { 1682 m->m_pkthdr.rcvif = ifp; 1683 return (m); 1684 } 1685 } 1686 1687 m_freem(m); 1688 return (NULL); 1689 } 1690 1691 /* 1692 * Loadbalancing 1693 */ 1694 1695 static int 1696 lagg_lb_attach(struct lagg_softc *sc) 1697 { 1698 struct lagg_port *lp; 1699 struct lagg_lb *lb; 1700 1701 if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb), 1702 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 1703 return (ENOMEM); 1704 1705 sc->sc_detach = lagg_lb_detach; 1706 sc->sc_start = lagg_lb_start; 1707 sc->sc_input = lagg_lb_input; 1708 sc->sc_port_create = lagg_lb_port_create; 1709 sc->sc_port_destroy = lagg_lb_port_destroy; 1710 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1711 1712 lb->lb_key = arc4random(); 1713 sc->sc_psc = (caddr_t)lb; 1714 1715 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1716 lagg_lb_port_create(lp); 1717 1718 return (0); 1719 } 1720 1721 static int 1722 lagg_lb_detach(struct lagg_softc *sc) 1723 { 1724 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1725 if (lb != NULL) 1726 free(lb, M_DEVBUF); 1727 return (0); 1728 } 1729 1730 static int 1731 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 1732 { 1733 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1734 struct lagg_port *lp_next; 1735 int i = 0; 1736 1737 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 1738 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1739 if (lp_next == lp) 1740 continue; 1741 if (i >= LAGG_MAX_PORTS) 1742 return (EINVAL); 1743 if (sc->sc_ifflags & IFF_DEBUG) 1744 printf("%s: port %s at index %d\n", 1745 sc->sc_ifname, lp_next->lp_ifname, i); 1746 lb->lb_ports[i++] = lp_next; 1747 } 1748 1749 return (0); 1750 } 1751 1752 static int 1753 lagg_lb_port_create(struct lagg_port *lp) 1754 { 1755 struct lagg_softc *sc = lp->lp_softc; 1756 return (lagg_lb_porttable(sc, NULL)); 1757 } 1758 1759 static void 1760 lagg_lb_port_destroy(struct lagg_port *lp) 1761 { 1762 struct lagg_softc *sc = lp->lp_softc; 1763 lagg_lb_porttable(sc, lp); 1764 } 1765 1766 static int 1767 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 1768 { 1769 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1770 struct lagg_port *lp = NULL; 1771 uint32_t p = 0; 1772 1773 if (sc->use_flowid && (m->m_flags & M_FLOWID)) 1774 p = m->m_pkthdr.flowid; 1775 else 1776 p = lagg_hashmbuf(sc, m, lb->lb_key); 1777 p %= sc->sc_count; 1778 lp = lb->lb_ports[p]; 1779 1780 /* 1781 * Check the port's link state. This will return the next active 1782 * port if the link is down or the port is NULL. 1783 */ 1784 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1785 m_freem(m); 1786 return (ENOENT); 1787 } 1788 1789 /* Send mbuf */ 1790 return (lagg_enqueue(lp->lp_ifp, m)); 1791 } 1792 1793 static struct mbuf * 1794 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1795 { 1796 struct ifnet *ifp = sc->sc_ifp; 1797 1798 /* Just pass in the packet to our lagg device */ 1799 m->m_pkthdr.rcvif = ifp; 1800 1801 return (m); 1802 } 1803 1804 /* 1805 * 802.3ad LACP 1806 */ 1807 1808 static int 1809 lagg_lacp_attach(struct lagg_softc *sc) 1810 { 1811 struct lagg_port *lp; 1812 int error; 1813 1814 sc->sc_detach = lagg_lacp_detach; 1815 sc->sc_port_create = lacp_port_create; 1816 sc->sc_port_destroy = lacp_port_destroy; 1817 sc->sc_linkstate = lacp_linkstate; 1818 sc->sc_start = lagg_lacp_start; 1819 sc->sc_input = lagg_lacp_input; 1820 sc->sc_init = lacp_init; 1821 sc->sc_stop = lacp_stop; 1822 sc->sc_lladdr = lagg_lacp_lladdr; 1823 sc->sc_req = lacp_req; 1824 sc->sc_portreq = lacp_portreq; 1825 1826 error = lacp_attach(sc); 1827 if (error) 1828 return (error); 1829 1830 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1831 lacp_port_create(lp); 1832 1833 return (error); 1834 } 1835 1836 static int 1837 lagg_lacp_detach(struct lagg_softc *sc) 1838 { 1839 struct lagg_port *lp; 1840 int error; 1841 1842 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1843 lacp_port_destroy(lp); 1844 1845 /* unlocking is safe here */ 1846 LAGG_WUNLOCK(sc); 1847 error = lacp_detach(sc); 1848 LAGG_WLOCK(sc); 1849 1850 return (error); 1851 } 1852 1853 static void 1854 lagg_lacp_lladdr(struct lagg_softc *sc) 1855 { 1856 struct lagg_port *lp; 1857 1858 /* purge all the lacp ports */ 1859 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1860 lacp_port_destroy(lp); 1861 1862 /* add them back in */ 1863 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1864 lacp_port_create(lp); 1865 } 1866 1867 static int 1868 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 1869 { 1870 struct lagg_port *lp; 1871 1872 lp = lacp_select_tx_port(sc, m); 1873 if (lp == NULL) { 1874 m_freem(m); 1875 return (ENETDOWN); 1876 } 1877 1878 /* Send mbuf */ 1879 return (lagg_enqueue(lp->lp_ifp, m)); 1880 } 1881 1882 static struct mbuf * 1883 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1884 { 1885 struct ifnet *ifp = sc->sc_ifp; 1886 struct ether_header *eh; 1887 u_short etype; 1888 1889 eh = mtod(m, struct ether_header *); 1890 etype = ntohs(eh->ether_type); 1891 1892 /* Tap off LACP control messages */ 1893 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 1894 m = lacp_input(lp, m); 1895 if (m == NULL) 1896 return (NULL); 1897 } 1898 1899 /* 1900 * If the port is not collecting or not in the active aggregator then 1901 * free and return. 1902 */ 1903 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 1904 m_freem(m); 1905 return (NULL); 1906 } 1907 1908 m->m_pkthdr.rcvif = ifp; 1909 return (m); 1910 } 1911 1912 static void 1913 lagg_callout(void *arg) 1914 { 1915 struct lagg_softc *sc = (struct lagg_softc *)arg; 1916 struct ifnet *ifp = sc->sc_ifp; 1917 1918 ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets); 1919 ifp->if_opackets = counter_u64_fetch(sc->sc_opackets); 1920 ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes); 1921 ifp->if_obytes = counter_u64_fetch(sc->sc_obytes); 1922 1923 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 1924 } 1925