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 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 /* lagg protocol table */ 157 static const struct { 158 int ti_proto; 159 int (*ti_attach)(struct lagg_softc *); 160 } lagg_protos[] = { 161 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach }, 162 { LAGG_PROTO_FAILOVER, lagg_fail_attach }, 163 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach }, 164 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach }, 165 { LAGG_PROTO_LACP, lagg_lacp_attach }, 166 { LAGG_PROTO_NONE, NULL } 167 }; 168 169 SYSCTL_DECL(_net_link); 170 static SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 171 "Link Aggregation"); 172 173 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */ 174 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW, 175 &lagg_failover_rx_all, 0, 176 "Accept input from any interface in a failover lagg"); 177 static int def_use_flowid = 1; /* Default value for using M_FLOWID */ 178 TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid); 179 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RW, 180 &def_use_flowid, 0, 181 "Default setting for using flow id for load sharing"); 182 183 static int 184 lagg_modevent(module_t mod, int type, void *data) 185 { 186 187 switch (type) { 188 case MOD_LOAD: 189 mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF); 190 SLIST_INIT(&lagg_list); 191 lagg_cloner = if_clone_simple(laggname, lagg_clone_create, 192 lagg_clone_destroy, 0); 193 lagg_input_p = lagg_input; 194 lagg_linkstate_p = lagg_port_state; 195 lagg_detach_cookie = EVENTHANDLER_REGISTER( 196 ifnet_departure_event, lagg_port_ifdetach, NULL, 197 EVENTHANDLER_PRI_ANY); 198 break; 199 case MOD_UNLOAD: 200 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 201 lagg_detach_cookie); 202 if_clone_detach(lagg_cloner); 203 lagg_input_p = NULL; 204 lagg_linkstate_p = NULL; 205 mtx_destroy(&lagg_list_mtx); 206 break; 207 default: 208 return (EOPNOTSUPP); 209 } 210 return (0); 211 } 212 213 static moduledata_t lagg_mod = { 214 "if_lagg", 215 lagg_modevent, 216 0 217 }; 218 219 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 220 MODULE_VERSION(if_lagg, 1); 221 222 #if __FreeBSD_version >= 800000 223 /* 224 * This routine is run via an vlan 225 * config EVENT 226 */ 227 static void 228 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 229 { 230 struct lagg_softc *sc = ifp->if_softc; 231 struct lagg_port *lp; 232 233 if (ifp->if_softc != arg) /* Not our event */ 234 return; 235 236 LAGG_RLOCK(sc); 237 if (!SLIST_EMPTY(&sc->sc_ports)) { 238 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 239 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 240 } 241 LAGG_RUNLOCK(sc); 242 } 243 244 /* 245 * This routine is run via an vlan 246 * unconfig EVENT 247 */ 248 static void 249 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 250 { 251 struct lagg_softc *sc = ifp->if_softc; 252 struct lagg_port *lp; 253 254 if (ifp->if_softc != arg) /* Not our event */ 255 return; 256 257 LAGG_RLOCK(sc); 258 if (!SLIST_EMPTY(&sc->sc_ports)) { 259 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 260 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 261 } 262 LAGG_RUNLOCK(sc); 263 } 264 #endif 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 sysctl_ctx_init(&sc->ctx); 284 snprintf(num, sizeof(num), "%u", unit); 285 sc->use_flowid = def_use_flowid; 286 oid = SYSCTL_ADD_NODE(&sc->ctx, &SYSCTL_NODE_CHILDREN(_net_link, lagg), 287 OID_AUTO, num, CTLFLAG_RD, NULL, ""); 288 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 289 "use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid, sc->use_flowid, 290 "Use flow id for load sharing"); 291 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 292 "count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count, 293 "Total number of ports"); 294 /* Hash all layers by default */ 295 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4; 296 297 sc->sc_proto = LAGG_PROTO_NONE; 298 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) { 299 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) { 300 sc->sc_proto = lagg_protos[i].ti_proto; 301 if ((error = lagg_protos[i].ti_attach(sc)) != 0) { 302 if_free(ifp); 303 free(sc, M_DEVBUF); 304 return (error); 305 } 306 break; 307 } 308 } 309 LAGG_LOCK_INIT(sc); 310 SLIST_INIT(&sc->sc_ports); 311 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc); 312 313 /* Initialise pseudo media types */ 314 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 315 lagg_media_status); 316 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 317 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 318 319 if_initname(ifp, laggname, unit); 320 ifp->if_softc = sc; 321 ifp->if_transmit = lagg_transmit; 322 ifp->if_qflush = lagg_qflush; 323 ifp->if_init = lagg_init; 324 ifp->if_ioctl = lagg_ioctl; 325 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 326 327 /* 328 * Attach as an ordinary ethernet device, children will be attached 329 * as special device IFT_IEEE8023ADLAG. 330 */ 331 ether_ifattach(ifp, eaddr); 332 333 #if __FreeBSD_version >= 800000 334 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 335 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 336 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 337 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 338 #endif 339 340 /* Insert into the global list of laggs */ 341 mtx_lock(&lagg_list_mtx); 342 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries); 343 mtx_unlock(&lagg_list_mtx); 344 345 return (0); 346 } 347 348 static void 349 lagg_clone_destroy(struct ifnet *ifp) 350 { 351 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 352 struct lagg_port *lp; 353 354 LAGG_WLOCK(sc); 355 356 lagg_stop(sc); 357 ifp->if_flags &= ~IFF_UP; 358 359 #if __FreeBSD_version >= 800000 360 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 361 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 362 #endif 363 364 /* Shutdown and remove lagg ports */ 365 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) 366 lagg_port_destroy(lp, 1); 367 /* Unhook the aggregation protocol */ 368 if (sc->sc_detach != NULL) 369 (*sc->sc_detach)(sc); 370 371 LAGG_WUNLOCK(sc); 372 373 sysctl_ctx_free(&sc->ctx); 374 ifmedia_removeall(&sc->sc_media); 375 ether_ifdetach(ifp); 376 if_free(ifp); 377 378 mtx_lock(&lagg_list_mtx); 379 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries); 380 mtx_unlock(&lagg_list_mtx); 381 382 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task); 383 LAGG_LOCK_DESTROY(sc); 384 free(sc, M_DEVBUF); 385 } 386 387 static void 388 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr) 389 { 390 struct ifnet *ifp = sc->sc_ifp; 391 392 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 393 return; 394 395 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN); 396 /* Let the protocol know the MAC has changed */ 397 if (sc->sc_lladdr != NULL) 398 (*sc->sc_lladdr)(sc); 399 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 400 } 401 402 static void 403 lagg_capabilities(struct lagg_softc *sc) 404 { 405 struct lagg_port *lp; 406 int cap = ~0, ena = ~0; 407 u_long hwa = ~0UL; 408 409 LAGG_WLOCK_ASSERT(sc); 410 411 /* Get capabilities from the lagg ports */ 412 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 413 cap &= lp->lp_ifp->if_capabilities; 414 ena &= lp->lp_ifp->if_capenable; 415 hwa &= lp->lp_ifp->if_hwassist; 416 } 417 cap = (cap == ~0 ? 0 : cap); 418 ena = (ena == ~0 ? 0 : ena); 419 hwa = (hwa == ~0 ? 0 : hwa); 420 421 if (sc->sc_ifp->if_capabilities != cap || 422 sc->sc_ifp->if_capenable != ena || 423 sc->sc_ifp->if_hwassist != hwa) { 424 sc->sc_ifp->if_capabilities = cap; 425 sc->sc_ifp->if_capenable = ena; 426 sc->sc_ifp->if_hwassist = hwa; 427 getmicrotime(&sc->sc_ifp->if_lastchange); 428 429 if (sc->sc_ifflags & IFF_DEBUG) 430 if_printf(sc->sc_ifp, 431 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 432 } 433 } 434 435 static void 436 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr) 437 { 438 struct lagg_softc *sc = lp->lp_softc; 439 struct ifnet *ifp = lp->lp_ifp; 440 struct lagg_llq *llq; 441 int pending = 0; 442 443 LAGG_WLOCK_ASSERT(sc); 444 445 if (lp->lp_detaching || 446 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 447 return; 448 449 /* Check to make sure its not already queued to be changed */ 450 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 451 if (llq->llq_ifp == ifp) { 452 pending = 1; 453 break; 454 } 455 } 456 457 if (!pending) { 458 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT); 459 if (llq == NULL) /* XXX what to do */ 460 return; 461 } 462 463 /* Update the lladdr even if pending, it may have changed */ 464 llq->llq_ifp = ifp; 465 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN); 466 467 if (!pending) 468 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries); 469 470 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task); 471 } 472 473 /* 474 * Set the interface MAC address from a taskqueue to avoid a LOR. 475 */ 476 static void 477 lagg_port_setlladdr(void *arg, int pending) 478 { 479 struct lagg_softc *sc = (struct lagg_softc *)arg; 480 struct lagg_llq *llq, *head; 481 struct ifnet *ifp; 482 int error; 483 484 /* Grab a local reference of the queue and remove it from the softc */ 485 LAGG_WLOCK(sc); 486 head = SLIST_FIRST(&sc->sc_llq_head); 487 SLIST_FIRST(&sc->sc_llq_head) = NULL; 488 LAGG_WUNLOCK(sc); 489 490 /* 491 * Traverse the queue and set the lladdr on each ifp. It is safe to do 492 * unlocked as we have the only reference to it. 493 */ 494 for (llq = head; llq != NULL; llq = head) { 495 ifp = llq->llq_ifp; 496 497 /* Set the link layer address */ 498 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN); 499 if (error) 500 printf("%s: setlladdr failed on %s\n", __func__, 501 ifp->if_xname); 502 503 head = SLIST_NEXT(llq, llq_entries); 504 free(llq, M_DEVBUF); 505 } 506 } 507 508 static int 509 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 510 { 511 struct lagg_softc *sc_ptr; 512 struct lagg_port *lp; 513 int error = 0; 514 515 LAGG_WLOCK_ASSERT(sc); 516 517 /* Limit the maximal number of lagg ports */ 518 if (sc->sc_count >= LAGG_MAX_PORTS) 519 return (ENOSPC); 520 521 /* Check if port has already been associated to a lagg */ 522 if (ifp->if_lagg != NULL) { 523 /* Port is already in the current lagg? */ 524 lp = (struct lagg_port *)ifp->if_lagg; 525 if (lp->lp_softc == sc) 526 return (EEXIST); 527 return (EBUSY); 528 } 529 530 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 531 if (ifp->if_type != IFT_ETHER) 532 return (EPROTONOSUPPORT); 533 534 /* Allow the first Ethernet member to define the MTU */ 535 if (SLIST_EMPTY(&sc->sc_ports)) 536 sc->sc_ifp->if_mtu = ifp->if_mtu; 537 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 538 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 539 ifp->if_xname); 540 return (EINVAL); 541 } 542 543 if ((lp = malloc(sizeof(struct lagg_port), 544 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 545 return (ENOMEM); 546 547 /* Check if port is a stacked lagg */ 548 mtx_lock(&lagg_list_mtx); 549 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) { 550 if (ifp == sc_ptr->sc_ifp) { 551 mtx_unlock(&lagg_list_mtx); 552 free(lp, M_DEVBUF); 553 return (EINVAL); 554 /* XXX disable stacking for the moment, its untested */ 555 #ifdef LAGG_PORT_STACKING 556 lp->lp_flags |= LAGG_PORT_STACK; 557 if (lagg_port_checkstacking(sc_ptr) >= 558 LAGG_MAX_STACKING) { 559 mtx_unlock(&lagg_list_mtx); 560 free(lp, M_DEVBUF); 561 return (E2BIG); 562 } 563 #endif 564 } 565 } 566 mtx_unlock(&lagg_list_mtx); 567 568 /* Change the interface type */ 569 lp->lp_iftype = ifp->if_type; 570 ifp->if_type = IFT_IEEE8023ADLAG; 571 ifp->if_lagg = lp; 572 lp->lp_ioctl = ifp->if_ioctl; 573 ifp->if_ioctl = lagg_port_ioctl; 574 lp->lp_output = ifp->if_output; 575 ifp->if_output = lagg_port_output; 576 577 lp->lp_ifp = ifp; 578 lp->lp_softc = sc; 579 580 /* Save port link layer address */ 581 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 582 583 if (SLIST_EMPTY(&sc->sc_ports)) { 584 sc->sc_primary = lp; 585 lagg_lladdr(sc, IF_LLADDR(ifp)); 586 } else { 587 /* Update link layer address for this port */ 588 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp)); 589 } 590 591 /* Insert into the list of ports */ 592 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 593 sc->sc_count++; 594 595 /* Update lagg capabilities */ 596 lagg_capabilities(sc); 597 lagg_linkstate(sc); 598 599 /* Add multicast addresses and interface flags to this port */ 600 lagg_ether_cmdmulti(lp, 1); 601 lagg_setflags(lp, 1); 602 603 if (sc->sc_port_create != NULL) 604 error = (*sc->sc_port_create)(lp); 605 if (error) { 606 /* remove the port again, without calling sc_port_destroy */ 607 lagg_port_destroy(lp, 0); 608 return (error); 609 } 610 611 return (error); 612 } 613 614 #ifdef LAGG_PORT_STACKING 615 static int 616 lagg_port_checkstacking(struct lagg_softc *sc) 617 { 618 struct lagg_softc *sc_ptr; 619 struct lagg_port *lp; 620 int m = 0; 621 622 LAGG_WLOCK_ASSERT(sc); 623 624 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 625 if (lp->lp_flags & LAGG_PORT_STACK) { 626 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 627 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 628 } 629 } 630 631 return (m + 1); 632 } 633 #endif 634 635 static int 636 lagg_port_destroy(struct lagg_port *lp, int runpd) 637 { 638 struct lagg_softc *sc = lp->lp_softc; 639 struct lagg_port *lp_ptr; 640 struct lagg_llq *llq; 641 struct ifnet *ifp = lp->lp_ifp; 642 643 LAGG_WLOCK_ASSERT(sc); 644 645 if (runpd && sc->sc_port_destroy != NULL) 646 (*sc->sc_port_destroy)(lp); 647 648 /* 649 * Remove multicast addresses and interface flags from this port and 650 * reset the MAC address, skip if the interface is being detached. 651 */ 652 if (!lp->lp_detaching) { 653 lagg_ether_cmdmulti(lp, 0); 654 lagg_setflags(lp, 0); 655 lagg_port_lladdr(lp, lp->lp_lladdr); 656 } 657 658 /* Restore interface */ 659 ifp->if_type = lp->lp_iftype; 660 ifp->if_ioctl = lp->lp_ioctl; 661 ifp->if_output = lp->lp_output; 662 ifp->if_lagg = NULL; 663 664 /* Finally, remove the port from the lagg */ 665 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 666 sc->sc_count--; 667 668 /* Update the primary interface */ 669 if (lp == sc->sc_primary) { 670 uint8_t lladdr[ETHER_ADDR_LEN]; 671 672 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) { 673 bzero(&lladdr, ETHER_ADDR_LEN); 674 } else { 675 bcopy(lp_ptr->lp_lladdr, 676 lladdr, ETHER_ADDR_LEN); 677 } 678 lagg_lladdr(sc, lladdr); 679 sc->sc_primary = lp_ptr; 680 681 /* Update link layer address for each port */ 682 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 683 lagg_port_lladdr(lp_ptr, lladdr); 684 } 685 686 /* Remove any pending lladdr changes from the queue */ 687 if (lp->lp_detaching) { 688 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 689 if (llq->llq_ifp == ifp) { 690 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq, 691 llq_entries); 692 free(llq, M_DEVBUF); 693 break; /* Only appears once */ 694 } 695 } 696 } 697 698 if (lp->lp_ifflags) 699 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 700 701 free(lp, M_DEVBUF); 702 703 /* Update lagg capabilities */ 704 lagg_capabilities(sc); 705 lagg_linkstate(sc); 706 707 return (0); 708 } 709 710 static int 711 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 712 { 713 struct lagg_reqport *rp = (struct lagg_reqport *)data; 714 struct lagg_softc *sc; 715 struct lagg_port *lp = NULL; 716 int error = 0; 717 718 /* Should be checked by the caller */ 719 if (ifp->if_type != IFT_IEEE8023ADLAG || 720 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 721 goto fallback; 722 723 switch (cmd) { 724 case SIOCGLAGGPORT: 725 if (rp->rp_portname[0] == '\0' || 726 ifunit(rp->rp_portname) != ifp) { 727 error = EINVAL; 728 break; 729 } 730 731 LAGG_RLOCK(sc); 732 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 733 error = ENOENT; 734 LAGG_RUNLOCK(sc); 735 break; 736 } 737 738 lagg_port2req(lp, rp); 739 LAGG_RUNLOCK(sc); 740 break; 741 742 case SIOCSIFCAP: 743 if (lp->lp_ioctl == NULL) { 744 error = EINVAL; 745 break; 746 } 747 error = (*lp->lp_ioctl)(ifp, cmd, data); 748 if (error) 749 break; 750 751 /* Update lagg interface capabilities */ 752 LAGG_WLOCK(sc); 753 lagg_capabilities(sc); 754 LAGG_WUNLOCK(sc); 755 break; 756 757 case SIOCSIFMTU: 758 /* Do not allow the MTU to be changed once joined */ 759 error = EINVAL; 760 break; 761 762 default: 763 goto fallback; 764 } 765 766 return (error); 767 768 fallback: 769 if (lp->lp_ioctl != NULL) 770 return ((*lp->lp_ioctl)(ifp, cmd, data)); 771 772 return (EINVAL); 773 } 774 775 /* 776 * For direct output to child ports. 777 */ 778 static int 779 lagg_port_output(struct ifnet *ifp, struct mbuf *m, 780 struct sockaddr *dst, struct route *ro) 781 { 782 struct lagg_port *lp = ifp->if_lagg; 783 784 switch (dst->sa_family) { 785 case pseudo_AF_HDRCMPLT: 786 case AF_UNSPEC: 787 return ((*lp->lp_output)(ifp, m, dst, ro)); 788 } 789 790 /* drop any other frames */ 791 m_freem(m); 792 return (EBUSY); 793 } 794 795 static void 796 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 797 { 798 struct lagg_port *lp; 799 struct lagg_softc *sc; 800 801 if ((lp = ifp->if_lagg) == NULL) 802 return; 803 /* If the ifnet is just being renamed, don't do anything. */ 804 if (ifp->if_flags & IFF_RENAMING) 805 return; 806 807 sc = lp->lp_softc; 808 809 LAGG_WLOCK(sc); 810 lp->lp_detaching = 1; 811 lagg_port_destroy(lp, 1); 812 LAGG_WUNLOCK(sc); 813 } 814 815 static void 816 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 817 { 818 struct lagg_softc *sc = lp->lp_softc; 819 820 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 821 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 822 rp->rp_prio = lp->lp_prio; 823 rp->rp_flags = lp->lp_flags; 824 if (sc->sc_portreq != NULL) 825 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc); 826 827 /* Add protocol specific flags */ 828 switch (sc->sc_proto) { 829 case LAGG_PROTO_FAILOVER: 830 if (lp == sc->sc_primary) 831 rp->rp_flags |= LAGG_PORT_MASTER; 832 if (lp == lagg_link_active(sc, sc->sc_primary)) 833 rp->rp_flags |= LAGG_PORT_ACTIVE; 834 break; 835 836 case LAGG_PROTO_ROUNDROBIN: 837 case LAGG_PROTO_LOADBALANCE: 838 case LAGG_PROTO_ETHERCHANNEL: 839 if (LAGG_PORTACTIVE(lp)) 840 rp->rp_flags |= LAGG_PORT_ACTIVE; 841 break; 842 843 case LAGG_PROTO_LACP: 844 /* LACP has a different definition of active */ 845 if (lacp_isactive(lp)) 846 rp->rp_flags |= LAGG_PORT_ACTIVE; 847 if (lacp_iscollecting(lp)) 848 rp->rp_flags |= LAGG_PORT_COLLECTING; 849 if (lacp_isdistributing(lp)) 850 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 851 break; 852 } 853 854 } 855 856 static void 857 lagg_init(void *xsc) 858 { 859 struct lagg_softc *sc = (struct lagg_softc *)xsc; 860 struct lagg_port *lp; 861 struct ifnet *ifp = sc->sc_ifp; 862 863 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 864 return; 865 866 LAGG_WLOCK(sc); 867 868 ifp->if_drv_flags |= IFF_DRV_RUNNING; 869 /* Update the port lladdrs */ 870 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 871 lagg_port_lladdr(lp, IF_LLADDR(ifp)); 872 873 if (sc->sc_init != NULL) 874 (*sc->sc_init)(sc); 875 876 LAGG_WUNLOCK(sc); 877 } 878 879 static void 880 lagg_stop(struct lagg_softc *sc) 881 { 882 struct ifnet *ifp = sc->sc_ifp; 883 884 LAGG_WLOCK_ASSERT(sc); 885 886 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 887 return; 888 889 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 890 891 if (sc->sc_stop != NULL) 892 (*sc->sc_stop)(sc); 893 } 894 895 static int 896 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 897 { 898 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 899 struct lagg_reqall *ra = (struct lagg_reqall *)data; 900 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 901 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 902 struct ifreq *ifr = (struct ifreq *)data; 903 struct lagg_port *lp; 904 struct ifnet *tpif; 905 struct thread *td = curthread; 906 char *buf, *outbuf; 907 int count, buflen, len, error = 0; 908 909 bzero(&rpbuf, sizeof(rpbuf)); 910 911 switch (cmd) { 912 case SIOCGLAGG: 913 LAGG_RLOCK(sc); 914 count = 0; 915 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 916 count++; 917 buflen = count * sizeof(struct lagg_reqport); 918 LAGG_RUNLOCK(sc); 919 920 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 921 922 LAGG_RLOCK(sc); 923 ra->ra_proto = sc->sc_proto; 924 if (sc->sc_req != NULL) 925 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc); 926 927 count = 0; 928 buf = outbuf; 929 len = min(ra->ra_size, buflen); 930 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 931 if (len < sizeof(rpbuf)) 932 break; 933 934 lagg_port2req(lp, &rpbuf); 935 memcpy(buf, &rpbuf, sizeof(rpbuf)); 936 count++; 937 buf += sizeof(rpbuf); 938 len -= sizeof(rpbuf); 939 } 940 LAGG_RUNLOCK(sc); 941 ra->ra_ports = count; 942 ra->ra_size = count * sizeof(rpbuf); 943 error = copyout(outbuf, ra->ra_port, ra->ra_size); 944 free(outbuf, M_TEMP); 945 break; 946 case SIOCSLAGG: 947 error = priv_check(td, PRIV_NET_LAGG); 948 if (error) 949 break; 950 if (ra->ra_proto >= LAGG_PROTO_MAX) { 951 error = EPROTONOSUPPORT; 952 break; 953 } 954 LAGG_WLOCK(sc); 955 if (sc->sc_proto != LAGG_PROTO_NONE) { 956 /* Reset protocol first in case detach unlocks */ 957 sc->sc_proto = LAGG_PROTO_NONE; 958 error = sc->sc_detach(sc); 959 sc->sc_detach = NULL; 960 sc->sc_start = NULL; 961 sc->sc_input = NULL; 962 sc->sc_port_create = NULL; 963 sc->sc_port_destroy = NULL; 964 sc->sc_linkstate = NULL; 965 sc->sc_init = NULL; 966 sc->sc_stop = NULL; 967 sc->sc_lladdr = NULL; 968 sc->sc_req = NULL; 969 sc->sc_portreq = NULL; 970 } else if (sc->sc_input != NULL) { 971 /* Still detaching */ 972 error = EBUSY; 973 } 974 if (error != 0) { 975 LAGG_WUNLOCK(sc); 976 break; 977 } 978 for (int i = 0; i < (sizeof(lagg_protos) / 979 sizeof(lagg_protos[0])); i++) { 980 if (lagg_protos[i].ti_proto == ra->ra_proto) { 981 if (sc->sc_ifflags & IFF_DEBUG) 982 printf("%s: using proto %u\n", 983 sc->sc_ifname, 984 lagg_protos[i].ti_proto); 985 sc->sc_proto = lagg_protos[i].ti_proto; 986 if (sc->sc_proto != LAGG_PROTO_NONE) 987 error = lagg_protos[i].ti_attach(sc); 988 LAGG_WUNLOCK(sc); 989 return (error); 990 } 991 } 992 LAGG_WUNLOCK(sc); 993 error = EPROTONOSUPPORT; 994 break; 995 case SIOCGLAGGFLAGS: 996 rf->rf_flags = sc->sc_flags; 997 break; 998 case SIOCSLAGGHASH: 999 error = priv_check(td, PRIV_NET_LAGG); 1000 if (error) 1001 break; 1002 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1003 error = EINVAL; 1004 break; 1005 } 1006 LAGG_WLOCK(sc); 1007 sc->sc_flags &= ~LAGG_F_HASHMASK; 1008 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK; 1009 LAGG_WUNLOCK(sc); 1010 break; 1011 case SIOCGLAGGPORT: 1012 if (rp->rp_portname[0] == '\0' || 1013 (tpif = ifunit(rp->rp_portname)) == NULL) { 1014 error = EINVAL; 1015 break; 1016 } 1017 1018 LAGG_RLOCK(sc); 1019 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1020 lp->lp_softc != sc) { 1021 error = ENOENT; 1022 LAGG_RUNLOCK(sc); 1023 break; 1024 } 1025 1026 lagg_port2req(lp, rp); 1027 LAGG_RUNLOCK(sc); 1028 break; 1029 case SIOCSLAGGPORT: 1030 error = priv_check(td, PRIV_NET_LAGG); 1031 if (error) 1032 break; 1033 if (rp->rp_portname[0] == '\0' || 1034 (tpif = ifunit(rp->rp_portname)) == NULL) { 1035 error = EINVAL; 1036 break; 1037 } 1038 LAGG_WLOCK(sc); 1039 error = lagg_port_create(sc, tpif); 1040 LAGG_WUNLOCK(sc); 1041 break; 1042 case SIOCSLAGGDELPORT: 1043 error = priv_check(td, PRIV_NET_LAGG); 1044 if (error) 1045 break; 1046 if (rp->rp_portname[0] == '\0' || 1047 (tpif = ifunit(rp->rp_portname)) == NULL) { 1048 error = EINVAL; 1049 break; 1050 } 1051 1052 LAGG_WLOCK(sc); 1053 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1054 lp->lp_softc != sc) { 1055 error = ENOENT; 1056 LAGG_WUNLOCK(sc); 1057 break; 1058 } 1059 1060 error = lagg_port_destroy(lp, 1); 1061 LAGG_WUNLOCK(sc); 1062 break; 1063 case SIOCSIFFLAGS: 1064 /* Set flags on ports too */ 1065 LAGG_WLOCK(sc); 1066 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1067 lagg_setflags(lp, 1); 1068 } 1069 LAGG_WUNLOCK(sc); 1070 1071 if (!(ifp->if_flags & IFF_UP) && 1072 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1073 /* 1074 * If interface is marked down and it is running, 1075 * then stop and disable it. 1076 */ 1077 LAGG_WLOCK(sc); 1078 lagg_stop(sc); 1079 LAGG_WUNLOCK(sc); 1080 } else if ((ifp->if_flags & IFF_UP) && 1081 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1082 /* 1083 * If interface is marked up and it is stopped, then 1084 * start it. 1085 */ 1086 (*ifp->if_init)(sc); 1087 } 1088 break; 1089 case SIOCADDMULTI: 1090 case SIOCDELMULTI: 1091 LAGG_WLOCK(sc); 1092 error = lagg_ether_setmulti(sc); 1093 LAGG_WUNLOCK(sc); 1094 break; 1095 case SIOCSIFMEDIA: 1096 case SIOCGIFMEDIA: 1097 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1098 break; 1099 1100 case SIOCSIFCAP: 1101 case SIOCSIFMTU: 1102 /* Do not allow the MTU or caps to be directly changed */ 1103 error = EINVAL; 1104 break; 1105 1106 default: 1107 error = ether_ioctl(ifp, cmd, data); 1108 break; 1109 } 1110 return (error); 1111 } 1112 1113 static int 1114 lagg_ether_setmulti(struct lagg_softc *sc) 1115 { 1116 struct lagg_port *lp; 1117 1118 LAGG_WLOCK_ASSERT(sc); 1119 1120 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1121 /* First, remove any existing filter entries. */ 1122 lagg_ether_cmdmulti(lp, 0); 1123 /* copy all addresses from the lagg interface to the port */ 1124 lagg_ether_cmdmulti(lp, 1); 1125 } 1126 return (0); 1127 } 1128 1129 static int 1130 lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1131 { 1132 struct lagg_softc *sc = lp->lp_softc; 1133 struct ifnet *ifp = lp->lp_ifp; 1134 struct ifnet *scifp = sc->sc_ifp; 1135 struct lagg_mc *mc; 1136 struct ifmultiaddr *ifma, *rifma = NULL; 1137 struct sockaddr_dl sdl; 1138 int error; 1139 1140 LAGG_WLOCK_ASSERT(sc); 1141 1142 bzero((char *)&sdl, sizeof(sdl)); 1143 sdl.sdl_len = sizeof(sdl); 1144 sdl.sdl_family = AF_LINK; 1145 sdl.sdl_type = IFT_ETHER; 1146 sdl.sdl_alen = ETHER_ADDR_LEN; 1147 sdl.sdl_index = ifp->if_index; 1148 1149 if (set) { 1150 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1151 if (ifma->ifma_addr->sa_family != AF_LINK) 1152 continue; 1153 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 1154 LLADDR(&sdl), ETHER_ADDR_LEN); 1155 1156 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 1157 if (error) 1158 return (error); 1159 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1160 if (mc == NULL) 1161 return (ENOMEM); 1162 mc->mc_ifma = rifma; 1163 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1164 } 1165 } else { 1166 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1167 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1168 if_delmulti_ifma(mc->mc_ifma); 1169 free(mc, M_DEVBUF); 1170 } 1171 } 1172 return (0); 1173 } 1174 1175 /* Handle a ref counted flag that should be set on the lagg port as well */ 1176 static int 1177 lagg_setflag(struct lagg_port *lp, int flag, int status, 1178 int (*func)(struct ifnet *, int)) 1179 { 1180 struct lagg_softc *sc = lp->lp_softc; 1181 struct ifnet *scifp = sc->sc_ifp; 1182 struct ifnet *ifp = lp->lp_ifp; 1183 int error; 1184 1185 LAGG_WLOCK_ASSERT(sc); 1186 1187 status = status ? (scifp->if_flags & flag) : 0; 1188 /* Now "status" contains the flag value or 0 */ 1189 1190 /* 1191 * See if recorded ports status is different from what 1192 * we want it to be. If it is, flip it. We record ports 1193 * status in lp_ifflags so that we won't clear ports flag 1194 * we haven't set. In fact, we don't clear or set ports 1195 * flags directly, but get or release references to them. 1196 * That's why we can be sure that recorded flags still are 1197 * in accord with actual ports flags. 1198 */ 1199 if (status != (lp->lp_ifflags & flag)) { 1200 error = (*func)(ifp, status); 1201 if (error) 1202 return (error); 1203 lp->lp_ifflags &= ~flag; 1204 lp->lp_ifflags |= status; 1205 } 1206 return (0); 1207 } 1208 1209 /* 1210 * Handle IFF_* flags that require certain changes on the lagg port 1211 * if "status" is true, update ports flags respective to the lagg 1212 * if "status" is false, forcedly clear the flags set on port. 1213 */ 1214 static int 1215 lagg_setflags(struct lagg_port *lp, int status) 1216 { 1217 int error, i; 1218 1219 for (i = 0; lagg_pflags[i].flag; i++) { 1220 error = lagg_setflag(lp, lagg_pflags[i].flag, 1221 status, lagg_pflags[i].func); 1222 if (error) 1223 return (error); 1224 } 1225 return (0); 1226 } 1227 1228 static int 1229 lagg_transmit(struct ifnet *ifp, struct mbuf *m) 1230 { 1231 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1232 int error, len, mcast; 1233 1234 len = m->m_pkthdr.len; 1235 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1236 1237 LAGG_RLOCK(sc); 1238 /* We need a Tx algorithm and at least one port */ 1239 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1240 LAGG_RUNLOCK(sc); 1241 m_freem(m); 1242 ifp->if_oerrors++; 1243 return (ENXIO); 1244 } 1245 1246 ETHER_BPF_MTAP(ifp, m); 1247 1248 error = (*sc->sc_start)(sc, m); 1249 LAGG_RUNLOCK(sc); 1250 1251 if (error == 0) { 1252 ifp->if_opackets++; 1253 ifp->if_omcasts += mcast; 1254 ifp->if_obytes += len; 1255 } else 1256 ifp->if_oerrors++; 1257 1258 return (error); 1259 } 1260 1261 /* 1262 * The ifp->if_qflush entry point for lagg(4) is no-op. 1263 */ 1264 static void 1265 lagg_qflush(struct ifnet *ifp __unused) 1266 { 1267 } 1268 1269 static struct mbuf * 1270 lagg_input(struct ifnet *ifp, struct mbuf *m) 1271 { 1272 struct lagg_port *lp = ifp->if_lagg; 1273 struct lagg_softc *sc = lp->lp_softc; 1274 struct ifnet *scifp = sc->sc_ifp; 1275 1276 LAGG_RLOCK(sc); 1277 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1278 (lp->lp_flags & LAGG_PORT_DISABLED) || 1279 sc->sc_proto == LAGG_PROTO_NONE) { 1280 LAGG_RUNLOCK(sc); 1281 m_freem(m); 1282 return (NULL); 1283 } 1284 1285 ETHER_BPF_MTAP(scifp, m); 1286 1287 m = (*sc->sc_input)(sc, lp, m); 1288 1289 if (m != NULL) { 1290 scifp->if_ipackets++; 1291 scifp->if_ibytes += m->m_pkthdr.len; 1292 1293 if (scifp->if_flags & IFF_MONITOR) { 1294 m_freem(m); 1295 m = NULL; 1296 } 1297 } 1298 1299 LAGG_RUNLOCK(sc); 1300 return (m); 1301 } 1302 1303 static int 1304 lagg_media_change(struct ifnet *ifp) 1305 { 1306 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1307 1308 if (sc->sc_ifflags & IFF_DEBUG) 1309 printf("%s\n", __func__); 1310 1311 /* Ignore */ 1312 return (0); 1313 } 1314 1315 static void 1316 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1317 { 1318 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1319 struct lagg_port *lp; 1320 1321 imr->ifm_status = IFM_AVALID; 1322 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1323 1324 LAGG_RLOCK(sc); 1325 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1326 if (LAGG_PORTACTIVE(lp)) 1327 imr->ifm_status |= IFM_ACTIVE; 1328 } 1329 LAGG_RUNLOCK(sc); 1330 } 1331 1332 static void 1333 lagg_linkstate(struct lagg_softc *sc) 1334 { 1335 struct lagg_port *lp; 1336 int new_link = LINK_STATE_DOWN; 1337 uint64_t speed; 1338 1339 /* Our link is considered up if at least one of our ports is active */ 1340 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1341 if (lp->lp_link_state == LINK_STATE_UP) { 1342 new_link = LINK_STATE_UP; 1343 break; 1344 } 1345 } 1346 if_link_state_change(sc->sc_ifp, new_link); 1347 1348 /* Update if_baudrate to reflect the max possible speed */ 1349 switch (sc->sc_proto) { 1350 case LAGG_PROTO_FAILOVER: 1351 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1352 sc->sc_primary->lp_ifp->if_baudrate : 0; 1353 break; 1354 case LAGG_PROTO_ROUNDROBIN: 1355 case LAGG_PROTO_LOADBALANCE: 1356 case LAGG_PROTO_ETHERCHANNEL: 1357 speed = 0; 1358 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1359 speed += lp->lp_ifp->if_baudrate; 1360 sc->sc_ifp->if_baudrate = speed; 1361 break; 1362 case LAGG_PROTO_LACP: 1363 /* LACP updates if_baudrate itself */ 1364 break; 1365 } 1366 } 1367 1368 static void 1369 lagg_port_state(struct ifnet *ifp, int state) 1370 { 1371 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1372 struct lagg_softc *sc = NULL; 1373 1374 if (lp != NULL) 1375 sc = lp->lp_softc; 1376 if (sc == NULL) 1377 return; 1378 1379 LAGG_WLOCK(sc); 1380 lagg_linkstate(sc); 1381 if (sc->sc_linkstate != NULL) 1382 (*sc->sc_linkstate)(lp); 1383 LAGG_WUNLOCK(sc); 1384 } 1385 1386 struct lagg_port * 1387 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1388 { 1389 struct lagg_port *lp_next, *rval = NULL; 1390 // int new_link = LINK_STATE_DOWN; 1391 1392 LAGG_RLOCK_ASSERT(sc); 1393 /* 1394 * Search a port which reports an active link state. 1395 */ 1396 1397 if (lp == NULL) 1398 goto search; 1399 if (LAGG_PORTACTIVE(lp)) { 1400 rval = lp; 1401 goto found; 1402 } 1403 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1404 LAGG_PORTACTIVE(lp_next)) { 1405 rval = lp_next; 1406 goto found; 1407 } 1408 1409 search: 1410 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1411 if (LAGG_PORTACTIVE(lp_next)) { 1412 rval = lp_next; 1413 goto found; 1414 } 1415 } 1416 1417 found: 1418 if (rval != NULL) { 1419 /* 1420 * The IEEE 802.1D standard assumes that a lagg with 1421 * multiple ports is always full duplex. This is valid 1422 * for load sharing laggs and if at least two links 1423 * are active. Unfortunately, checking the latter would 1424 * be too expensive at this point. 1425 XXX 1426 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1427 (sc->sc_count > 1)) 1428 new_link = LINK_STATE_FULL_DUPLEX; 1429 else 1430 new_link = rval->lp_link_state; 1431 */ 1432 } 1433 1434 return (rval); 1435 } 1436 1437 static const void * 1438 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf) 1439 { 1440 if (m->m_pkthdr.len < (off + len)) { 1441 return (NULL); 1442 } else if (m->m_len < (off + len)) { 1443 m_copydata(m, off, len, buf); 1444 return (buf); 1445 } 1446 return (mtod(m, char *) + off); 1447 } 1448 1449 uint32_t 1450 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key) 1451 { 1452 uint16_t etype; 1453 uint32_t p = key; 1454 int off; 1455 struct ether_header *eh; 1456 const struct ether_vlan_header *vlan; 1457 #ifdef INET 1458 const struct ip *ip; 1459 const uint32_t *ports; 1460 int iphlen; 1461 #endif 1462 #ifdef INET6 1463 const struct ip6_hdr *ip6; 1464 uint32_t flow; 1465 #endif 1466 union { 1467 #ifdef INET 1468 struct ip ip; 1469 #endif 1470 #ifdef INET6 1471 struct ip6_hdr ip6; 1472 #endif 1473 struct ether_vlan_header vlan; 1474 uint32_t port; 1475 } buf; 1476 1477 1478 off = sizeof(*eh); 1479 if (m->m_len < off) 1480 goto out; 1481 eh = mtod(m, struct ether_header *); 1482 etype = ntohs(eh->ether_type); 1483 if (sc->sc_flags & LAGG_F_HASHL2) { 1484 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p); 1485 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p); 1486 } 1487 1488 /* Special handling for encapsulating VLAN frames */ 1489 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) { 1490 p = hash32_buf(&m->m_pkthdr.ether_vtag, 1491 sizeof(m->m_pkthdr.ether_vtag), p); 1492 } else if (etype == ETHERTYPE_VLAN) { 1493 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf); 1494 if (vlan == NULL) 1495 goto out; 1496 1497 if (sc->sc_flags & LAGG_F_HASHL2) 1498 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p); 1499 etype = ntohs(vlan->evl_proto); 1500 off += sizeof(*vlan) - sizeof(*eh); 1501 } 1502 1503 switch (etype) { 1504 #ifdef INET 1505 case ETHERTYPE_IP: 1506 ip = lagg_gethdr(m, off, sizeof(*ip), &buf); 1507 if (ip == NULL) 1508 goto out; 1509 1510 if (sc->sc_flags & LAGG_F_HASHL3) { 1511 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p); 1512 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p); 1513 } 1514 if (!(sc->sc_flags & LAGG_F_HASHL4)) 1515 break; 1516 switch (ip->ip_p) { 1517 case IPPROTO_TCP: 1518 case IPPROTO_UDP: 1519 case IPPROTO_SCTP: 1520 iphlen = ip->ip_hl << 2; 1521 if (iphlen < sizeof(*ip)) 1522 break; 1523 off += iphlen; 1524 ports = lagg_gethdr(m, off, sizeof(*ports), &buf); 1525 if (ports == NULL) 1526 break; 1527 p = hash32_buf(ports, sizeof(*ports), p); 1528 break; 1529 } 1530 break; 1531 #endif 1532 #ifdef INET6 1533 case ETHERTYPE_IPV6: 1534 if (!(sc->sc_flags & LAGG_F_HASHL3)) 1535 break; 1536 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf); 1537 if (ip6 == NULL) 1538 goto out; 1539 1540 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p); 1541 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p); 1542 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK; 1543 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */ 1544 break; 1545 #endif 1546 } 1547 out: 1548 return (p); 1549 } 1550 1551 int 1552 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1553 { 1554 1555 return (ifp->if_transmit)(ifp, m); 1556 } 1557 1558 /* 1559 * Simple round robin aggregation 1560 */ 1561 1562 static int 1563 lagg_rr_attach(struct lagg_softc *sc) 1564 { 1565 sc->sc_detach = lagg_rr_detach; 1566 sc->sc_start = lagg_rr_start; 1567 sc->sc_input = lagg_rr_input; 1568 sc->sc_port_create = NULL; 1569 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1570 sc->sc_seq = 0; 1571 1572 return (0); 1573 } 1574 1575 static int 1576 lagg_rr_detach(struct lagg_softc *sc) 1577 { 1578 return (0); 1579 } 1580 1581 static int 1582 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 1583 { 1584 struct lagg_port *lp; 1585 uint32_t p; 1586 1587 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1588 p %= sc->sc_count; 1589 lp = SLIST_FIRST(&sc->sc_ports); 1590 while (p--) 1591 lp = SLIST_NEXT(lp, lp_entries); 1592 1593 /* 1594 * Check the port's link state. This will return the next active 1595 * port if the link is down or the port is NULL. 1596 */ 1597 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1598 m_freem(m); 1599 return (ENOENT); 1600 } 1601 1602 /* Send mbuf */ 1603 return (lagg_enqueue(lp->lp_ifp, m)); 1604 } 1605 1606 static struct mbuf * 1607 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1608 { 1609 struct ifnet *ifp = sc->sc_ifp; 1610 1611 /* Just pass in the packet to our lagg device */ 1612 m->m_pkthdr.rcvif = ifp; 1613 1614 return (m); 1615 } 1616 1617 /* 1618 * Active failover 1619 */ 1620 1621 static int 1622 lagg_fail_attach(struct lagg_softc *sc) 1623 { 1624 sc->sc_detach = lagg_fail_detach; 1625 sc->sc_start = lagg_fail_start; 1626 sc->sc_input = lagg_fail_input; 1627 sc->sc_port_create = NULL; 1628 sc->sc_port_destroy = NULL; 1629 1630 return (0); 1631 } 1632 1633 static int 1634 lagg_fail_detach(struct lagg_softc *sc) 1635 { 1636 return (0); 1637 } 1638 1639 static int 1640 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 1641 { 1642 struct lagg_port *lp; 1643 1644 /* Use the master port if active or the next available port */ 1645 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 1646 m_freem(m); 1647 return (ENOENT); 1648 } 1649 1650 /* Send mbuf */ 1651 return (lagg_enqueue(lp->lp_ifp, m)); 1652 } 1653 1654 static struct mbuf * 1655 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1656 { 1657 struct ifnet *ifp = sc->sc_ifp; 1658 struct lagg_port *tmp_tp; 1659 1660 if (lp == sc->sc_primary || lagg_failover_rx_all) { 1661 m->m_pkthdr.rcvif = ifp; 1662 return (m); 1663 } 1664 1665 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1666 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1667 /* 1668 * If tmp_tp is null, we've recieved a packet when all 1669 * our links are down. Weird, but process it anyways. 1670 */ 1671 if ((tmp_tp == NULL || tmp_tp == lp)) { 1672 m->m_pkthdr.rcvif = ifp; 1673 return (m); 1674 } 1675 } 1676 1677 m_freem(m); 1678 return (NULL); 1679 } 1680 1681 /* 1682 * Loadbalancing 1683 */ 1684 1685 static int 1686 lagg_lb_attach(struct lagg_softc *sc) 1687 { 1688 struct lagg_port *lp; 1689 struct lagg_lb *lb; 1690 1691 if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb), 1692 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 1693 return (ENOMEM); 1694 1695 sc->sc_detach = lagg_lb_detach; 1696 sc->sc_start = lagg_lb_start; 1697 sc->sc_input = lagg_lb_input; 1698 sc->sc_port_create = lagg_lb_port_create; 1699 sc->sc_port_destroy = lagg_lb_port_destroy; 1700 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1701 1702 lb->lb_key = arc4random(); 1703 sc->sc_psc = (caddr_t)lb; 1704 1705 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1706 lagg_lb_port_create(lp); 1707 1708 return (0); 1709 } 1710 1711 static int 1712 lagg_lb_detach(struct lagg_softc *sc) 1713 { 1714 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1715 if (lb != NULL) 1716 free(lb, M_DEVBUF); 1717 return (0); 1718 } 1719 1720 static int 1721 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 1722 { 1723 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1724 struct lagg_port *lp_next; 1725 int i = 0; 1726 1727 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 1728 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1729 if (lp_next == lp) 1730 continue; 1731 if (i >= LAGG_MAX_PORTS) 1732 return (EINVAL); 1733 if (sc->sc_ifflags & IFF_DEBUG) 1734 printf("%s: port %s at index %d\n", 1735 sc->sc_ifname, lp_next->lp_ifname, i); 1736 lb->lb_ports[i++] = lp_next; 1737 } 1738 1739 return (0); 1740 } 1741 1742 static int 1743 lagg_lb_port_create(struct lagg_port *lp) 1744 { 1745 struct lagg_softc *sc = lp->lp_softc; 1746 return (lagg_lb_porttable(sc, NULL)); 1747 } 1748 1749 static void 1750 lagg_lb_port_destroy(struct lagg_port *lp) 1751 { 1752 struct lagg_softc *sc = lp->lp_softc; 1753 lagg_lb_porttable(sc, lp); 1754 } 1755 1756 static int 1757 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 1758 { 1759 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1760 struct lagg_port *lp = NULL; 1761 uint32_t p = 0; 1762 1763 if (sc->use_flowid && (m->m_flags & M_FLOWID)) 1764 p = m->m_pkthdr.flowid; 1765 else 1766 p = lagg_hashmbuf(sc, m, lb->lb_key); 1767 p %= sc->sc_count; 1768 lp = lb->lb_ports[p]; 1769 1770 /* 1771 * Check the port's link state. This will return the next active 1772 * port if the link is down or the port is NULL. 1773 */ 1774 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1775 m_freem(m); 1776 return (ENOENT); 1777 } 1778 1779 /* Send mbuf */ 1780 return (lagg_enqueue(lp->lp_ifp, m)); 1781 } 1782 1783 static struct mbuf * 1784 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1785 { 1786 struct ifnet *ifp = sc->sc_ifp; 1787 1788 /* Just pass in the packet to our lagg device */ 1789 m->m_pkthdr.rcvif = ifp; 1790 1791 return (m); 1792 } 1793 1794 /* 1795 * 802.3ad LACP 1796 */ 1797 1798 static int 1799 lagg_lacp_attach(struct lagg_softc *sc) 1800 { 1801 struct lagg_port *lp; 1802 int error; 1803 1804 sc->sc_detach = lagg_lacp_detach; 1805 sc->sc_port_create = lacp_port_create; 1806 sc->sc_port_destroy = lacp_port_destroy; 1807 sc->sc_linkstate = lacp_linkstate; 1808 sc->sc_start = lagg_lacp_start; 1809 sc->sc_input = lagg_lacp_input; 1810 sc->sc_init = lacp_init; 1811 sc->sc_stop = lacp_stop; 1812 sc->sc_lladdr = lagg_lacp_lladdr; 1813 sc->sc_req = lacp_req; 1814 sc->sc_portreq = lacp_portreq; 1815 1816 error = lacp_attach(sc); 1817 if (error) 1818 return (error); 1819 1820 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1821 lacp_port_create(lp); 1822 1823 return (error); 1824 } 1825 1826 static int 1827 lagg_lacp_detach(struct lagg_softc *sc) 1828 { 1829 struct lagg_port *lp; 1830 int error; 1831 1832 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1833 lacp_port_destroy(lp); 1834 1835 /* unlocking is safe here */ 1836 LAGG_WUNLOCK(sc); 1837 error = lacp_detach(sc); 1838 LAGG_WLOCK(sc); 1839 1840 return (error); 1841 } 1842 1843 static void 1844 lagg_lacp_lladdr(struct lagg_softc *sc) 1845 { 1846 struct lagg_port *lp; 1847 1848 /* purge all the lacp ports */ 1849 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1850 lacp_port_destroy(lp); 1851 1852 /* add them back in */ 1853 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1854 lacp_port_create(lp); 1855 } 1856 1857 static int 1858 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 1859 { 1860 struct lagg_port *lp; 1861 1862 lp = lacp_select_tx_port(sc, m); 1863 if (lp == NULL) { 1864 m_freem(m); 1865 return (EBUSY); 1866 } 1867 1868 /* Send mbuf */ 1869 return (lagg_enqueue(lp->lp_ifp, m)); 1870 } 1871 1872 static struct mbuf * 1873 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1874 { 1875 struct ifnet *ifp = sc->sc_ifp; 1876 struct ether_header *eh; 1877 u_short etype; 1878 1879 eh = mtod(m, struct ether_header *); 1880 etype = ntohs(eh->ether_type); 1881 1882 /* Tap off LACP control messages */ 1883 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 1884 m = lacp_input(lp, m); 1885 if (m == NULL) 1886 return (NULL); 1887 } 1888 1889 /* 1890 * If the port is not collecting or not in the active aggregator then 1891 * free and return. 1892 */ 1893 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 1894 m_freem(m); 1895 return (NULL); 1896 } 1897 1898 m->m_pkthdr.rcvif = ifp; 1899 return (m); 1900 } 1901