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 void lagg_lladdr(struct lagg_softc *, uint8_t *); 89 static void lagg_capabilities(struct lagg_softc *); 90 static void lagg_port_lladdr(struct lagg_port *, uint8_t *); 91 static void lagg_port_setlladdr(void *, int); 92 static int lagg_port_create(struct lagg_softc *, struct ifnet *); 93 static int lagg_port_destroy(struct lagg_port *, int); 94 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); 95 static void lagg_linkstate(struct lagg_softc *); 96 static void lagg_port_state(struct ifnet *, int); 97 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); 98 static int lagg_port_output(struct ifnet *, struct mbuf *, 99 struct sockaddr *, struct route *); 100 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 101 #ifdef LAGG_PORT_STACKING 102 static int lagg_port_checkstacking(struct lagg_softc *); 103 #endif 104 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 105 static void lagg_init(void *); 106 static void lagg_stop(struct lagg_softc *); 107 static int lagg_ioctl(struct ifnet *, u_long, caddr_t); 108 static int lagg_ether_setmulti(struct lagg_softc *); 109 static int lagg_ether_cmdmulti(struct lagg_port *, int); 110 static int lagg_setflag(struct lagg_port *, int, int, 111 int (*func)(struct ifnet *, int)); 112 static int lagg_setflags(struct lagg_port *, int status); 113 static void lagg_start(struct ifnet *); 114 static int lagg_media_change(struct ifnet *); 115 static void lagg_media_status(struct ifnet *, struct ifmediareq *); 116 static struct lagg_port *lagg_link_active(struct lagg_softc *, 117 struct lagg_port *); 118 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *); 119 120 IFC_SIMPLE_DECLARE(lagg, 0); 121 122 /* Simple round robin */ 123 static int lagg_rr_attach(struct lagg_softc *); 124 static int lagg_rr_detach(struct lagg_softc *); 125 static int lagg_rr_start(struct lagg_softc *, struct mbuf *); 126 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 127 struct mbuf *); 128 129 /* Active failover */ 130 static int lagg_fail_attach(struct lagg_softc *); 131 static int lagg_fail_detach(struct lagg_softc *); 132 static int lagg_fail_start(struct lagg_softc *, struct mbuf *); 133 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 134 struct mbuf *); 135 136 /* Loadbalancing */ 137 static int lagg_lb_attach(struct lagg_softc *); 138 static int lagg_lb_detach(struct lagg_softc *); 139 static int lagg_lb_port_create(struct lagg_port *); 140 static void lagg_lb_port_destroy(struct lagg_port *); 141 static int lagg_lb_start(struct lagg_softc *, struct mbuf *); 142 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 143 struct mbuf *); 144 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 145 146 /* 802.3ad LACP */ 147 static int lagg_lacp_attach(struct lagg_softc *); 148 static int lagg_lacp_detach(struct lagg_softc *); 149 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); 150 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 151 struct mbuf *); 152 static void lagg_lacp_lladdr(struct lagg_softc *); 153 154 /* lagg protocol table */ 155 static const struct { 156 int ti_proto; 157 int (*ti_attach)(struct lagg_softc *); 158 } lagg_protos[] = { 159 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach }, 160 { LAGG_PROTO_FAILOVER, lagg_fail_attach }, 161 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach }, 162 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach }, 163 { LAGG_PROTO_LACP, lagg_lacp_attach }, 164 { LAGG_PROTO_NONE, NULL } 165 }; 166 167 SYSCTL_DECL(_net_link); 168 static SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 169 "Link Aggregation"); 170 171 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */ 172 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW, 173 &lagg_failover_rx_all, 0, 174 "Accept input from any interface in a failover lagg"); 175 static int def_use_flowid = 1; /* Default value for using M_FLOWID */ 176 TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid); 177 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RW, 178 &def_use_flowid, 0, 179 "Default setting for using flow id for load sharing"); 180 181 static int 182 lagg_modevent(module_t mod, int type, void *data) 183 { 184 185 switch (type) { 186 case MOD_LOAD: 187 mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF); 188 SLIST_INIT(&lagg_list); 189 if_clone_attach(&lagg_cloner); 190 lagg_input_p = lagg_input; 191 lagg_linkstate_p = lagg_port_state; 192 lagg_detach_cookie = EVENTHANDLER_REGISTER( 193 ifnet_departure_event, lagg_port_ifdetach, NULL, 194 EVENTHANDLER_PRI_ANY); 195 break; 196 case MOD_UNLOAD: 197 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 198 lagg_detach_cookie); 199 if_clone_detach(&lagg_cloner); 200 lagg_input_p = NULL; 201 lagg_linkstate_p = NULL; 202 mtx_destroy(&lagg_list_mtx); 203 break; 204 default: 205 return (EOPNOTSUPP); 206 } 207 return (0); 208 } 209 210 static moduledata_t lagg_mod = { 211 "if_lagg", 212 lagg_modevent, 213 0 214 }; 215 216 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 217 MODULE_VERSION(if_lagg, 1); 218 219 #if __FreeBSD_version >= 800000 220 /* 221 * This routine is run via an vlan 222 * config EVENT 223 */ 224 static void 225 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 226 { 227 struct lagg_softc *sc = ifp->if_softc; 228 struct lagg_port *lp; 229 230 if (ifp->if_softc != arg) /* Not our event */ 231 return; 232 233 LAGG_RLOCK(sc); 234 if (!SLIST_EMPTY(&sc->sc_ports)) { 235 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 236 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 237 } 238 LAGG_RUNLOCK(sc); 239 } 240 241 /* 242 * This routine is run via an vlan 243 * unconfig EVENT 244 */ 245 static void 246 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 247 { 248 struct lagg_softc *sc = ifp->if_softc; 249 struct lagg_port *lp; 250 251 if (ifp->if_softc != arg) /* Not our event */ 252 return; 253 254 LAGG_RLOCK(sc); 255 if (!SLIST_EMPTY(&sc->sc_ports)) { 256 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 257 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 258 } 259 LAGG_RUNLOCK(sc); 260 } 261 #endif 262 263 static int 264 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) 265 { 266 struct lagg_softc *sc; 267 struct ifnet *ifp; 268 int i, error = 0; 269 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ 270 struct sysctl_oid *oid; 271 char num[14]; /* sufficient for 32 bits */ 272 273 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO); 274 ifp = sc->sc_ifp = if_alloc(IFT_ETHER); 275 if (ifp == NULL) { 276 free(sc, M_DEVBUF); 277 return (ENOSPC); 278 } 279 280 sysctl_ctx_init(&sc->ctx); 281 snprintf(num, sizeof(num), "%u", unit); 282 sc->use_flowid = def_use_flowid; 283 oid = SYSCTL_ADD_NODE(&sc->ctx, &SYSCTL_NODE_CHILDREN(_net_link, lagg), 284 OID_AUTO, num, CTLFLAG_RD, NULL, ""); 285 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 286 "use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid, sc->use_flowid, 287 "Use flow id for load sharing"); 288 /* Hash all layers by default */ 289 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4; 290 291 sc->sc_proto = LAGG_PROTO_NONE; 292 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) { 293 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) { 294 sc->sc_proto = lagg_protos[i].ti_proto; 295 if ((error = lagg_protos[i].ti_attach(sc)) != 0) { 296 if_free(ifp); 297 free(sc, M_DEVBUF); 298 return (error); 299 } 300 break; 301 } 302 } 303 LAGG_LOCK_INIT(sc); 304 SLIST_INIT(&sc->sc_ports); 305 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc); 306 307 /* Initialise pseudo media types */ 308 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 309 lagg_media_status); 310 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 311 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 312 313 if_initname(ifp, ifc->ifc_name, unit); 314 ifp->if_softc = sc; 315 ifp->if_start = lagg_start; 316 ifp->if_init = lagg_init; 317 ifp->if_ioctl = lagg_ioctl; 318 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 319 320 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 321 ifp->if_snd.ifq_drv_maxlen = ifqmaxlen; 322 IFQ_SET_READY(&ifp->if_snd); 323 324 /* 325 * Attach as an ordinary ethernet device, children will be attached 326 * as special device IFT_IEEE8023ADLAG. 327 */ 328 ether_ifattach(ifp, eaddr); 329 330 #if __FreeBSD_version >= 800000 331 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 332 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 333 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 334 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 335 #endif 336 337 /* Insert into the global list of laggs */ 338 mtx_lock(&lagg_list_mtx); 339 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries); 340 mtx_unlock(&lagg_list_mtx); 341 342 return (0); 343 } 344 345 static void 346 lagg_clone_destroy(struct ifnet *ifp) 347 { 348 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 349 struct lagg_port *lp; 350 351 LAGG_WLOCK(sc); 352 353 lagg_stop(sc); 354 ifp->if_flags &= ~IFF_UP; 355 356 #if __FreeBSD_version >= 800000 357 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 358 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 359 #endif 360 361 /* Shutdown and remove lagg ports */ 362 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) 363 lagg_port_destroy(lp, 1); 364 /* Unhook the aggregation protocol */ 365 if (sc->sc_detach != NULL) 366 (*sc->sc_detach)(sc); 367 368 LAGG_WUNLOCK(sc); 369 370 sysctl_ctx_free(&sc->ctx); 371 ifmedia_removeall(&sc->sc_media); 372 ether_ifdetach(ifp); 373 if_free(ifp); 374 375 mtx_lock(&lagg_list_mtx); 376 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries); 377 mtx_unlock(&lagg_list_mtx); 378 379 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task); 380 LAGG_LOCK_DESTROY(sc); 381 free(sc, M_DEVBUF); 382 } 383 384 static void 385 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr) 386 { 387 struct ifnet *ifp = sc->sc_ifp; 388 389 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 390 return; 391 392 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN); 393 /* Let the protocol know the MAC has changed */ 394 if (sc->sc_lladdr != NULL) 395 (*sc->sc_lladdr)(sc); 396 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 397 } 398 399 static void 400 lagg_capabilities(struct lagg_softc *sc) 401 { 402 struct lagg_port *lp; 403 int cap = ~0, ena = ~0; 404 u_long hwa = ~0UL; 405 406 LAGG_WLOCK_ASSERT(sc); 407 408 /* Get capabilities from the lagg ports */ 409 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 410 cap &= lp->lp_ifp->if_capabilities; 411 ena &= lp->lp_ifp->if_capenable; 412 hwa &= lp->lp_ifp->if_hwassist; 413 } 414 cap = (cap == ~0 ? 0 : cap); 415 ena = (ena == ~0 ? 0 : ena); 416 hwa = (hwa == ~0 ? 0 : hwa); 417 418 if (sc->sc_ifp->if_capabilities != cap || 419 sc->sc_ifp->if_capenable != ena || 420 sc->sc_ifp->if_hwassist != hwa) { 421 sc->sc_ifp->if_capabilities = cap; 422 sc->sc_ifp->if_capenable = ena; 423 sc->sc_ifp->if_hwassist = hwa; 424 getmicrotime(&sc->sc_ifp->if_lastchange); 425 426 if (sc->sc_ifflags & IFF_DEBUG) 427 if_printf(sc->sc_ifp, 428 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 429 } 430 } 431 432 static void 433 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr) 434 { 435 struct lagg_softc *sc = lp->lp_softc; 436 struct ifnet *ifp = lp->lp_ifp; 437 struct lagg_llq *llq; 438 int pending = 0; 439 440 LAGG_WLOCK_ASSERT(sc); 441 442 if (lp->lp_detaching || 443 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 444 return; 445 446 /* Check to make sure its not already queued to be changed */ 447 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 448 if (llq->llq_ifp == ifp) { 449 pending = 1; 450 break; 451 } 452 } 453 454 if (!pending) { 455 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT); 456 if (llq == NULL) /* XXX what to do */ 457 return; 458 } 459 460 /* Update the lladdr even if pending, it may have changed */ 461 llq->llq_ifp = ifp; 462 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN); 463 464 if (!pending) 465 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries); 466 467 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task); 468 } 469 470 /* 471 * Set the interface MAC address from a taskqueue to avoid a LOR. 472 */ 473 static void 474 lagg_port_setlladdr(void *arg, int pending) 475 { 476 struct lagg_softc *sc = (struct lagg_softc *)arg; 477 struct lagg_llq *llq, *head; 478 struct ifnet *ifp; 479 int error; 480 481 /* Grab a local reference of the queue and remove it from the softc */ 482 LAGG_WLOCK(sc); 483 head = SLIST_FIRST(&sc->sc_llq_head); 484 SLIST_FIRST(&sc->sc_llq_head) = NULL; 485 LAGG_WUNLOCK(sc); 486 487 /* 488 * Traverse the queue and set the lladdr on each ifp. It is safe to do 489 * unlocked as we have the only reference to it. 490 */ 491 for (llq = head; llq != NULL; llq = head) { 492 ifp = llq->llq_ifp; 493 494 /* Set the link layer address */ 495 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN); 496 if (error) 497 printf("%s: setlladdr failed on %s\n", __func__, 498 ifp->if_xname); 499 500 head = SLIST_NEXT(llq, llq_entries); 501 free(llq, M_DEVBUF); 502 } 503 } 504 505 static int 506 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 507 { 508 struct lagg_softc *sc_ptr; 509 struct lagg_port *lp; 510 int error = 0; 511 512 LAGG_WLOCK_ASSERT(sc); 513 514 /* Limit the maximal number of lagg ports */ 515 if (sc->sc_count >= LAGG_MAX_PORTS) 516 return (ENOSPC); 517 518 /* Check if port has already been associated to a lagg */ 519 if (ifp->if_lagg != NULL) { 520 /* Port is already in the current lagg? */ 521 lp = (struct lagg_port *)ifp->if_lagg; 522 if (lp->lp_softc == sc) 523 return (EEXIST); 524 return (EBUSY); 525 } 526 527 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 528 if (ifp->if_type != IFT_ETHER) 529 return (EPROTONOSUPPORT); 530 531 /* Allow the first Ethernet member to define the MTU */ 532 if (SLIST_EMPTY(&sc->sc_ports)) 533 sc->sc_ifp->if_mtu = ifp->if_mtu; 534 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 535 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 536 ifp->if_xname); 537 return (EINVAL); 538 } 539 540 if ((lp = malloc(sizeof(struct lagg_port), 541 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 542 return (ENOMEM); 543 544 /* Check if port is a stacked lagg */ 545 mtx_lock(&lagg_list_mtx); 546 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) { 547 if (ifp == sc_ptr->sc_ifp) { 548 mtx_unlock(&lagg_list_mtx); 549 free(lp, M_DEVBUF); 550 return (EINVAL); 551 /* XXX disable stacking for the moment, its untested */ 552 #ifdef LAGG_PORT_STACKING 553 lp->lp_flags |= LAGG_PORT_STACK; 554 if (lagg_port_checkstacking(sc_ptr) >= 555 LAGG_MAX_STACKING) { 556 mtx_unlock(&lagg_list_mtx); 557 free(lp, M_DEVBUF); 558 return (E2BIG); 559 } 560 #endif 561 } 562 } 563 mtx_unlock(&lagg_list_mtx); 564 565 /* Change the interface type */ 566 lp->lp_iftype = ifp->if_type; 567 ifp->if_type = IFT_IEEE8023ADLAG; 568 ifp->if_lagg = lp; 569 lp->lp_ioctl = ifp->if_ioctl; 570 ifp->if_ioctl = lagg_port_ioctl; 571 lp->lp_output = ifp->if_output; 572 ifp->if_output = lagg_port_output; 573 574 lp->lp_ifp = ifp; 575 lp->lp_softc = sc; 576 577 /* Save port link layer address */ 578 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 579 580 if (SLIST_EMPTY(&sc->sc_ports)) { 581 sc->sc_primary = lp; 582 lagg_lladdr(sc, IF_LLADDR(ifp)); 583 } else { 584 /* Update link layer address for this port */ 585 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp)); 586 } 587 588 /* Insert into the list of ports */ 589 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 590 sc->sc_count++; 591 592 /* Update lagg capabilities */ 593 lagg_capabilities(sc); 594 lagg_linkstate(sc); 595 596 /* Add multicast addresses and interface flags to this port */ 597 lagg_ether_cmdmulti(lp, 1); 598 lagg_setflags(lp, 1); 599 600 if (sc->sc_port_create != NULL) 601 error = (*sc->sc_port_create)(lp); 602 if (error) { 603 /* remove the port again, without calling sc_port_destroy */ 604 lagg_port_destroy(lp, 0); 605 return (error); 606 } 607 608 return (error); 609 } 610 611 #ifdef LAGG_PORT_STACKING 612 static int 613 lagg_port_checkstacking(struct lagg_softc *sc) 614 { 615 struct lagg_softc *sc_ptr; 616 struct lagg_port *lp; 617 int m = 0; 618 619 LAGG_WLOCK_ASSERT(sc); 620 621 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 622 if (lp->lp_flags & LAGG_PORT_STACK) { 623 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 624 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 625 } 626 } 627 628 return (m + 1); 629 } 630 #endif 631 632 static int 633 lagg_port_destroy(struct lagg_port *lp, int runpd) 634 { 635 struct lagg_softc *sc = lp->lp_softc; 636 struct lagg_port *lp_ptr; 637 struct lagg_llq *llq; 638 struct ifnet *ifp = lp->lp_ifp; 639 640 LAGG_WLOCK_ASSERT(sc); 641 642 if (runpd && sc->sc_port_destroy != NULL) 643 (*sc->sc_port_destroy)(lp); 644 645 /* 646 * Remove multicast addresses and interface flags from this port and 647 * reset the MAC address, skip if the interface is being detached. 648 */ 649 if (!lp->lp_detaching) { 650 lagg_ether_cmdmulti(lp, 0); 651 lagg_setflags(lp, 0); 652 lagg_port_lladdr(lp, lp->lp_lladdr); 653 } 654 655 /* Restore interface */ 656 ifp->if_type = lp->lp_iftype; 657 ifp->if_ioctl = lp->lp_ioctl; 658 ifp->if_output = lp->lp_output; 659 ifp->if_lagg = NULL; 660 661 /* Finally, remove the port from the lagg */ 662 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 663 sc->sc_count--; 664 665 /* Update the primary interface */ 666 if (lp == sc->sc_primary) { 667 uint8_t lladdr[ETHER_ADDR_LEN]; 668 669 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) { 670 bzero(&lladdr, ETHER_ADDR_LEN); 671 } else { 672 bcopy(lp_ptr->lp_lladdr, 673 lladdr, ETHER_ADDR_LEN); 674 } 675 lagg_lladdr(sc, lladdr); 676 sc->sc_primary = lp_ptr; 677 678 /* Update link layer address for each port */ 679 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 680 lagg_port_lladdr(lp_ptr, lladdr); 681 } 682 683 /* Remove any pending lladdr changes from the queue */ 684 if (lp->lp_detaching) { 685 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 686 if (llq->llq_ifp == ifp) { 687 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq, 688 llq_entries); 689 free(llq, M_DEVBUF); 690 break; /* Only appears once */ 691 } 692 } 693 } 694 695 if (lp->lp_ifflags) 696 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 697 698 free(lp, M_DEVBUF); 699 700 /* Update lagg capabilities */ 701 lagg_capabilities(sc); 702 lagg_linkstate(sc); 703 704 return (0); 705 } 706 707 static int 708 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 709 { 710 struct lagg_reqport *rp = (struct lagg_reqport *)data; 711 struct lagg_softc *sc; 712 struct lagg_port *lp = NULL; 713 int error = 0; 714 715 /* Should be checked by the caller */ 716 if (ifp->if_type != IFT_IEEE8023ADLAG || 717 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 718 goto fallback; 719 720 switch (cmd) { 721 case SIOCGLAGGPORT: 722 if (rp->rp_portname[0] == '\0' || 723 ifunit(rp->rp_portname) != ifp) { 724 error = EINVAL; 725 break; 726 } 727 728 LAGG_RLOCK(sc); 729 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 730 error = ENOENT; 731 LAGG_RUNLOCK(sc); 732 break; 733 } 734 735 lagg_port2req(lp, rp); 736 LAGG_RUNLOCK(sc); 737 break; 738 739 case SIOCSIFCAP: 740 if (lp->lp_ioctl == NULL) { 741 error = EINVAL; 742 break; 743 } 744 error = (*lp->lp_ioctl)(ifp, cmd, data); 745 if (error) 746 break; 747 748 /* Update lagg interface capabilities */ 749 LAGG_WLOCK(sc); 750 lagg_capabilities(sc); 751 LAGG_WUNLOCK(sc); 752 break; 753 754 case SIOCSIFMTU: 755 /* Do not allow the MTU to be changed once joined */ 756 error = EINVAL; 757 break; 758 759 default: 760 goto fallback; 761 } 762 763 return (error); 764 765 fallback: 766 if (lp->lp_ioctl != NULL) 767 return ((*lp->lp_ioctl)(ifp, cmd, data)); 768 769 return (EINVAL); 770 } 771 772 /* 773 * For direct output to child ports. 774 */ 775 static int 776 lagg_port_output(struct ifnet *ifp, struct mbuf *m, 777 struct sockaddr *dst, struct route *ro) 778 { 779 struct lagg_port *lp = ifp->if_lagg; 780 781 switch (dst->sa_family) { 782 case pseudo_AF_HDRCMPLT: 783 case AF_UNSPEC: 784 return ((*lp->lp_output)(ifp, m, dst, ro)); 785 } 786 787 /* drop any other frames */ 788 m_freem(m); 789 return (EBUSY); 790 } 791 792 static void 793 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 794 { 795 struct lagg_port *lp; 796 struct lagg_softc *sc; 797 798 if ((lp = ifp->if_lagg) == NULL) 799 return; 800 801 sc = lp->lp_softc; 802 803 LAGG_WLOCK(sc); 804 lp->lp_detaching = 1; 805 lagg_port_destroy(lp, 1); 806 LAGG_WUNLOCK(sc); 807 } 808 809 static void 810 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 811 { 812 struct lagg_softc *sc = lp->lp_softc; 813 814 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 815 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 816 rp->rp_prio = lp->lp_prio; 817 rp->rp_flags = lp->lp_flags; 818 if (sc->sc_portreq != NULL) 819 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc); 820 821 /* Add protocol specific flags */ 822 switch (sc->sc_proto) { 823 case LAGG_PROTO_FAILOVER: 824 if (lp == sc->sc_primary) 825 rp->rp_flags |= LAGG_PORT_MASTER; 826 if (lp == lagg_link_active(sc, sc->sc_primary)) 827 rp->rp_flags |= LAGG_PORT_ACTIVE; 828 break; 829 830 case LAGG_PROTO_ROUNDROBIN: 831 case LAGG_PROTO_LOADBALANCE: 832 case LAGG_PROTO_ETHERCHANNEL: 833 if (LAGG_PORTACTIVE(lp)) 834 rp->rp_flags |= LAGG_PORT_ACTIVE; 835 break; 836 837 case LAGG_PROTO_LACP: 838 /* LACP has a different definition of active */ 839 if (lacp_isactive(lp)) 840 rp->rp_flags |= LAGG_PORT_ACTIVE; 841 if (lacp_iscollecting(lp)) 842 rp->rp_flags |= LAGG_PORT_COLLECTING; 843 if (lacp_isdistributing(lp)) 844 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 845 break; 846 } 847 848 } 849 850 static void 851 lagg_init(void *xsc) 852 { 853 struct lagg_softc *sc = (struct lagg_softc *)xsc; 854 struct lagg_port *lp; 855 struct ifnet *ifp = sc->sc_ifp; 856 857 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 858 return; 859 860 LAGG_WLOCK(sc); 861 862 ifp->if_drv_flags |= IFF_DRV_RUNNING; 863 /* Update the port lladdrs */ 864 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 865 lagg_port_lladdr(lp, IF_LLADDR(ifp)); 866 867 if (sc->sc_init != NULL) 868 (*sc->sc_init)(sc); 869 870 LAGG_WUNLOCK(sc); 871 } 872 873 static void 874 lagg_stop(struct lagg_softc *sc) 875 { 876 struct ifnet *ifp = sc->sc_ifp; 877 878 LAGG_WLOCK_ASSERT(sc); 879 880 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 881 return; 882 883 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 884 885 if (sc->sc_stop != NULL) 886 (*sc->sc_stop)(sc); 887 } 888 889 static int 890 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 891 { 892 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 893 struct lagg_reqall *ra = (struct lagg_reqall *)data; 894 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 895 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 896 struct ifreq *ifr = (struct ifreq *)data; 897 struct lagg_port *lp; 898 struct ifnet *tpif; 899 struct thread *td = curthread; 900 char *buf, *outbuf; 901 int count, buflen, len, error = 0; 902 903 bzero(&rpbuf, sizeof(rpbuf)); 904 905 switch (cmd) { 906 case SIOCGLAGG: 907 LAGG_RLOCK(sc); 908 count = 0; 909 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 910 count++; 911 buflen = count * sizeof(struct lagg_reqport); 912 LAGG_RUNLOCK(sc); 913 914 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 915 916 LAGG_RLOCK(sc); 917 ra->ra_proto = sc->sc_proto; 918 if (sc->sc_req != NULL) 919 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc); 920 921 count = 0; 922 buf = outbuf; 923 len = min(ra->ra_size, buflen); 924 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 925 if (len < sizeof(rpbuf)) 926 break; 927 928 lagg_port2req(lp, &rpbuf); 929 memcpy(buf, &rpbuf, sizeof(rpbuf)); 930 count++; 931 buf += sizeof(rpbuf); 932 len -= sizeof(rpbuf); 933 } 934 LAGG_RUNLOCK(sc); 935 ra->ra_ports = count; 936 ra->ra_size = count * sizeof(rpbuf); 937 error = copyout(outbuf, ra->ra_port, ra->ra_size); 938 free(outbuf, M_TEMP); 939 break; 940 case SIOCSLAGG: 941 error = priv_check(td, PRIV_NET_LAGG); 942 if (error) 943 break; 944 if (ra->ra_proto >= LAGG_PROTO_MAX) { 945 error = EPROTONOSUPPORT; 946 break; 947 } 948 LAGG_WLOCK(sc); 949 if (sc->sc_proto != LAGG_PROTO_NONE) { 950 /* Reset protocol first in case detach unlocks */ 951 sc->sc_proto = LAGG_PROTO_NONE; 952 error = sc->sc_detach(sc); 953 sc->sc_detach = NULL; 954 sc->sc_start = NULL; 955 sc->sc_input = NULL; 956 sc->sc_port_create = NULL; 957 sc->sc_port_destroy = NULL; 958 sc->sc_linkstate = NULL; 959 sc->sc_init = NULL; 960 sc->sc_stop = NULL; 961 sc->sc_lladdr = NULL; 962 sc->sc_req = NULL; 963 sc->sc_portreq = NULL; 964 } else if (sc->sc_input != NULL) { 965 /* Still detaching */ 966 error = EBUSY; 967 } 968 if (error != 0) { 969 LAGG_WUNLOCK(sc); 970 break; 971 } 972 for (int i = 0; i < (sizeof(lagg_protos) / 973 sizeof(lagg_protos[0])); i++) { 974 if (lagg_protos[i].ti_proto == ra->ra_proto) { 975 if (sc->sc_ifflags & IFF_DEBUG) 976 printf("%s: using proto %u\n", 977 sc->sc_ifname, 978 lagg_protos[i].ti_proto); 979 sc->sc_proto = lagg_protos[i].ti_proto; 980 if (sc->sc_proto != LAGG_PROTO_NONE) 981 error = lagg_protos[i].ti_attach(sc); 982 LAGG_WUNLOCK(sc); 983 return (error); 984 } 985 } 986 LAGG_WUNLOCK(sc); 987 error = EPROTONOSUPPORT; 988 break; 989 case SIOCGLAGGFLAGS: 990 rf->rf_flags = sc->sc_flags; 991 break; 992 case SIOCSLAGGHASH: 993 error = priv_check(td, PRIV_NET_LAGG); 994 if (error) 995 break; 996 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 997 error = EINVAL; 998 break; 999 } 1000 LAGG_WLOCK(sc); 1001 sc->sc_flags &= ~LAGG_F_HASHMASK; 1002 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK; 1003 LAGG_WUNLOCK(sc); 1004 break; 1005 case SIOCGLAGGPORT: 1006 if (rp->rp_portname[0] == '\0' || 1007 (tpif = ifunit(rp->rp_portname)) == NULL) { 1008 error = EINVAL; 1009 break; 1010 } 1011 1012 LAGG_RLOCK(sc); 1013 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1014 lp->lp_softc != sc) { 1015 error = ENOENT; 1016 LAGG_RUNLOCK(sc); 1017 break; 1018 } 1019 1020 lagg_port2req(lp, rp); 1021 LAGG_RUNLOCK(sc); 1022 break; 1023 case SIOCSLAGGPORT: 1024 error = priv_check(td, PRIV_NET_LAGG); 1025 if (error) 1026 break; 1027 if (rp->rp_portname[0] == '\0' || 1028 (tpif = ifunit(rp->rp_portname)) == NULL) { 1029 error = EINVAL; 1030 break; 1031 } 1032 LAGG_WLOCK(sc); 1033 error = lagg_port_create(sc, tpif); 1034 LAGG_WUNLOCK(sc); 1035 break; 1036 case SIOCSLAGGDELPORT: 1037 error = priv_check(td, PRIV_NET_LAGG); 1038 if (error) 1039 break; 1040 if (rp->rp_portname[0] == '\0' || 1041 (tpif = ifunit(rp->rp_portname)) == NULL) { 1042 error = EINVAL; 1043 break; 1044 } 1045 1046 LAGG_WLOCK(sc); 1047 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1048 lp->lp_softc != sc) { 1049 error = ENOENT; 1050 LAGG_WUNLOCK(sc); 1051 break; 1052 } 1053 1054 error = lagg_port_destroy(lp, 1); 1055 LAGG_WUNLOCK(sc); 1056 break; 1057 case SIOCSIFFLAGS: 1058 /* Set flags on ports too */ 1059 LAGG_WLOCK(sc); 1060 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1061 lagg_setflags(lp, 1); 1062 } 1063 LAGG_WUNLOCK(sc); 1064 1065 if (!(ifp->if_flags & IFF_UP) && 1066 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1067 /* 1068 * If interface is marked down and it is running, 1069 * then stop and disable it. 1070 */ 1071 LAGG_WLOCK(sc); 1072 lagg_stop(sc); 1073 LAGG_WUNLOCK(sc); 1074 } else if ((ifp->if_flags & IFF_UP) && 1075 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1076 /* 1077 * If interface is marked up and it is stopped, then 1078 * start it. 1079 */ 1080 (*ifp->if_init)(sc); 1081 } 1082 break; 1083 case SIOCADDMULTI: 1084 case SIOCDELMULTI: 1085 LAGG_WLOCK(sc); 1086 error = lagg_ether_setmulti(sc); 1087 LAGG_WUNLOCK(sc); 1088 break; 1089 case SIOCSIFMEDIA: 1090 case SIOCGIFMEDIA: 1091 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1092 break; 1093 1094 case SIOCSIFCAP: 1095 case SIOCSIFMTU: 1096 /* Do not allow the MTU or caps to be directly changed */ 1097 error = EINVAL; 1098 break; 1099 1100 default: 1101 error = ether_ioctl(ifp, cmd, data); 1102 break; 1103 } 1104 return (error); 1105 } 1106 1107 static int 1108 lagg_ether_setmulti(struct lagg_softc *sc) 1109 { 1110 struct lagg_port *lp; 1111 1112 LAGG_WLOCK_ASSERT(sc); 1113 1114 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1115 /* First, remove any existing filter entries. */ 1116 lagg_ether_cmdmulti(lp, 0); 1117 /* copy all addresses from the lagg interface to the port */ 1118 lagg_ether_cmdmulti(lp, 1); 1119 } 1120 return (0); 1121 } 1122 1123 static int 1124 lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1125 { 1126 struct lagg_softc *sc = lp->lp_softc; 1127 struct ifnet *ifp = lp->lp_ifp; 1128 struct ifnet *scifp = sc->sc_ifp; 1129 struct lagg_mc *mc; 1130 struct ifmultiaddr *ifma, *rifma = NULL; 1131 struct sockaddr_dl sdl; 1132 int error; 1133 1134 LAGG_WLOCK_ASSERT(sc); 1135 1136 bzero((char *)&sdl, sizeof(sdl)); 1137 sdl.sdl_len = sizeof(sdl); 1138 sdl.sdl_family = AF_LINK; 1139 sdl.sdl_type = IFT_ETHER; 1140 sdl.sdl_alen = ETHER_ADDR_LEN; 1141 sdl.sdl_index = ifp->if_index; 1142 1143 if (set) { 1144 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1145 if (ifma->ifma_addr->sa_family != AF_LINK) 1146 continue; 1147 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 1148 LLADDR(&sdl), ETHER_ADDR_LEN); 1149 1150 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 1151 if (error) 1152 return (error); 1153 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1154 if (mc == NULL) 1155 return (ENOMEM); 1156 mc->mc_ifma = rifma; 1157 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1158 } 1159 } else { 1160 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1161 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1162 if_delmulti_ifma(mc->mc_ifma); 1163 free(mc, M_DEVBUF); 1164 } 1165 } 1166 return (0); 1167 } 1168 1169 /* Handle a ref counted flag that should be set on the lagg port as well */ 1170 static int 1171 lagg_setflag(struct lagg_port *lp, int flag, int status, 1172 int (*func)(struct ifnet *, int)) 1173 { 1174 struct lagg_softc *sc = lp->lp_softc; 1175 struct ifnet *scifp = sc->sc_ifp; 1176 struct ifnet *ifp = lp->lp_ifp; 1177 int error; 1178 1179 LAGG_WLOCK_ASSERT(sc); 1180 1181 status = status ? (scifp->if_flags & flag) : 0; 1182 /* Now "status" contains the flag value or 0 */ 1183 1184 /* 1185 * See if recorded ports status is different from what 1186 * we want it to be. If it is, flip it. We record ports 1187 * status in lp_ifflags so that we won't clear ports flag 1188 * we haven't set. In fact, we don't clear or set ports 1189 * flags directly, but get or release references to them. 1190 * That's why we can be sure that recorded flags still are 1191 * in accord with actual ports flags. 1192 */ 1193 if (status != (lp->lp_ifflags & flag)) { 1194 error = (*func)(ifp, status); 1195 if (error) 1196 return (error); 1197 lp->lp_ifflags &= ~flag; 1198 lp->lp_ifflags |= status; 1199 } 1200 return (0); 1201 } 1202 1203 /* 1204 * Handle IFF_* flags that require certain changes on the lagg port 1205 * if "status" is true, update ports flags respective to the lagg 1206 * if "status" is false, forcedly clear the flags set on port. 1207 */ 1208 static int 1209 lagg_setflags(struct lagg_port *lp, int status) 1210 { 1211 int error, i; 1212 1213 for (i = 0; lagg_pflags[i].flag; i++) { 1214 error = lagg_setflag(lp, lagg_pflags[i].flag, 1215 status, lagg_pflags[i].func); 1216 if (error) 1217 return (error); 1218 } 1219 return (0); 1220 } 1221 1222 static void 1223 lagg_start(struct ifnet *ifp) 1224 { 1225 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1226 struct mbuf *m; 1227 int error = 0; 1228 1229 LAGG_RLOCK(sc); 1230 /* We need a Tx algorithm and at least one port */ 1231 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1232 IF_DRAIN(&ifp->if_snd); 1233 LAGG_RUNLOCK(sc); 1234 return; 1235 } 1236 1237 for (;; error = 0) { 1238 IFQ_DEQUEUE(&ifp->if_snd, m); 1239 if (m == NULL) 1240 break; 1241 1242 ETHER_BPF_MTAP(ifp, m); 1243 1244 error = (*sc->sc_start)(sc, m); 1245 if (error == 0) 1246 ifp->if_opackets++; 1247 else 1248 ifp->if_oerrors++; 1249 } 1250 LAGG_RUNLOCK(sc); 1251 } 1252 1253 static struct mbuf * 1254 lagg_input(struct ifnet *ifp, struct mbuf *m) 1255 { 1256 struct lagg_port *lp = ifp->if_lagg; 1257 struct lagg_softc *sc = lp->lp_softc; 1258 struct ifnet *scifp = sc->sc_ifp; 1259 1260 LAGG_RLOCK(sc); 1261 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1262 (lp->lp_flags & LAGG_PORT_DISABLED) || 1263 sc->sc_proto == LAGG_PROTO_NONE) { 1264 LAGG_RUNLOCK(sc); 1265 m_freem(m); 1266 return (NULL); 1267 } 1268 1269 ETHER_BPF_MTAP(scifp, m); 1270 1271 m = (*sc->sc_input)(sc, lp, m); 1272 1273 if (m != NULL) { 1274 scifp->if_ipackets++; 1275 scifp->if_ibytes += m->m_pkthdr.len; 1276 1277 if (scifp->if_flags & IFF_MONITOR) { 1278 m_freem(m); 1279 m = NULL; 1280 } 1281 } 1282 1283 LAGG_RUNLOCK(sc); 1284 return (m); 1285 } 1286 1287 static int 1288 lagg_media_change(struct ifnet *ifp) 1289 { 1290 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1291 1292 if (sc->sc_ifflags & IFF_DEBUG) 1293 printf("%s\n", __func__); 1294 1295 /* Ignore */ 1296 return (0); 1297 } 1298 1299 static void 1300 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1301 { 1302 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1303 struct lagg_port *lp; 1304 1305 imr->ifm_status = IFM_AVALID; 1306 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1307 1308 LAGG_RLOCK(sc); 1309 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1310 if (LAGG_PORTACTIVE(lp)) 1311 imr->ifm_status |= IFM_ACTIVE; 1312 } 1313 LAGG_RUNLOCK(sc); 1314 } 1315 1316 static void 1317 lagg_linkstate(struct lagg_softc *sc) 1318 { 1319 struct lagg_port *lp; 1320 int new_link = LINK_STATE_DOWN; 1321 uint64_t speed; 1322 1323 /* Our link is considered up if at least one of our ports is active */ 1324 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1325 if (lp->lp_link_state == LINK_STATE_UP) { 1326 new_link = LINK_STATE_UP; 1327 break; 1328 } 1329 } 1330 if_link_state_change(sc->sc_ifp, new_link); 1331 1332 /* Update if_baudrate to reflect the max possible speed */ 1333 switch (sc->sc_proto) { 1334 case LAGG_PROTO_FAILOVER: 1335 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1336 sc->sc_primary->lp_ifp->if_baudrate : 0; 1337 break; 1338 case LAGG_PROTO_ROUNDROBIN: 1339 case LAGG_PROTO_LOADBALANCE: 1340 case LAGG_PROTO_ETHERCHANNEL: 1341 speed = 0; 1342 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1343 speed += lp->lp_ifp->if_baudrate; 1344 sc->sc_ifp->if_baudrate = speed; 1345 break; 1346 case LAGG_PROTO_LACP: 1347 /* LACP updates if_baudrate itself */ 1348 break; 1349 } 1350 } 1351 1352 static void 1353 lagg_port_state(struct ifnet *ifp, int state) 1354 { 1355 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1356 struct lagg_softc *sc = NULL; 1357 1358 if (lp != NULL) 1359 sc = lp->lp_softc; 1360 if (sc == NULL) 1361 return; 1362 1363 LAGG_WLOCK(sc); 1364 lagg_linkstate(sc); 1365 if (sc->sc_linkstate != NULL) 1366 (*sc->sc_linkstate)(lp); 1367 LAGG_WUNLOCK(sc); 1368 } 1369 1370 struct lagg_port * 1371 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1372 { 1373 struct lagg_port *lp_next, *rval = NULL; 1374 // int new_link = LINK_STATE_DOWN; 1375 1376 LAGG_RLOCK_ASSERT(sc); 1377 /* 1378 * Search a port which reports an active link state. 1379 */ 1380 1381 if (lp == NULL) 1382 goto search; 1383 if (LAGG_PORTACTIVE(lp)) { 1384 rval = lp; 1385 goto found; 1386 } 1387 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1388 LAGG_PORTACTIVE(lp_next)) { 1389 rval = lp_next; 1390 goto found; 1391 } 1392 1393 search: 1394 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1395 if (LAGG_PORTACTIVE(lp_next)) { 1396 rval = lp_next; 1397 goto found; 1398 } 1399 } 1400 1401 found: 1402 if (rval != NULL) { 1403 /* 1404 * The IEEE 802.1D standard assumes that a lagg with 1405 * multiple ports is always full duplex. This is valid 1406 * for load sharing laggs and if at least two links 1407 * are active. Unfortunately, checking the latter would 1408 * be too expensive at this point. 1409 XXX 1410 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1411 (sc->sc_count > 1)) 1412 new_link = LINK_STATE_FULL_DUPLEX; 1413 else 1414 new_link = rval->lp_link_state; 1415 */ 1416 } 1417 1418 return (rval); 1419 } 1420 1421 static const void * 1422 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf) 1423 { 1424 if (m->m_pkthdr.len < (off + len)) { 1425 return (NULL); 1426 } else if (m->m_len < (off + len)) { 1427 m_copydata(m, off, len, buf); 1428 return (buf); 1429 } 1430 return (mtod(m, char *) + off); 1431 } 1432 1433 uint32_t 1434 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key) 1435 { 1436 uint16_t etype; 1437 uint32_t p = key; 1438 int off; 1439 struct ether_header *eh; 1440 const struct ether_vlan_header *vlan; 1441 #ifdef INET 1442 const struct ip *ip; 1443 const uint32_t *ports; 1444 int iphlen; 1445 #endif 1446 #ifdef INET6 1447 const struct ip6_hdr *ip6; 1448 uint32_t flow; 1449 #endif 1450 union { 1451 #ifdef INET 1452 struct ip ip; 1453 #endif 1454 #ifdef INET6 1455 struct ip6_hdr ip6; 1456 #endif 1457 struct ether_vlan_header vlan; 1458 uint32_t port; 1459 } buf; 1460 1461 1462 off = sizeof(*eh); 1463 if (m->m_len < off) 1464 goto out; 1465 eh = mtod(m, struct ether_header *); 1466 etype = ntohs(eh->ether_type); 1467 if (sc->sc_flags & LAGG_F_HASHL2) { 1468 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p); 1469 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p); 1470 } 1471 1472 /* Special handling for encapsulating VLAN frames */ 1473 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) { 1474 p = hash32_buf(&m->m_pkthdr.ether_vtag, 1475 sizeof(m->m_pkthdr.ether_vtag), p); 1476 } else if (etype == ETHERTYPE_VLAN) { 1477 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf); 1478 if (vlan == NULL) 1479 goto out; 1480 1481 if (sc->sc_flags & LAGG_F_HASHL2) 1482 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p); 1483 etype = ntohs(vlan->evl_proto); 1484 off += sizeof(*vlan) - sizeof(*eh); 1485 } 1486 1487 switch (etype) { 1488 #ifdef INET 1489 case ETHERTYPE_IP: 1490 ip = lagg_gethdr(m, off, sizeof(*ip), &buf); 1491 if (ip == NULL) 1492 goto out; 1493 1494 if (sc->sc_flags & LAGG_F_HASHL3) { 1495 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p); 1496 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p); 1497 } 1498 if (!(sc->sc_flags & LAGG_F_HASHL4)) 1499 break; 1500 switch (ip->ip_p) { 1501 case IPPROTO_TCP: 1502 case IPPROTO_UDP: 1503 case IPPROTO_SCTP: 1504 iphlen = ip->ip_hl << 2; 1505 if (iphlen < sizeof(*ip)) 1506 break; 1507 off += iphlen; 1508 ports = lagg_gethdr(m, off, sizeof(*ports), &buf); 1509 if (ports == NULL) 1510 break; 1511 p = hash32_buf(ports, sizeof(*ports), p); 1512 break; 1513 } 1514 break; 1515 #endif 1516 #ifdef INET6 1517 case ETHERTYPE_IPV6: 1518 if (!(sc->sc_flags & LAGG_F_HASHL3)) 1519 break; 1520 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf); 1521 if (ip6 == NULL) 1522 goto out; 1523 1524 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p); 1525 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p); 1526 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK; 1527 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */ 1528 break; 1529 #endif 1530 } 1531 out: 1532 return (p); 1533 } 1534 1535 int 1536 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1537 { 1538 1539 return (ifp->if_transmit)(ifp, m); 1540 } 1541 1542 /* 1543 * Simple round robin aggregation 1544 */ 1545 1546 static int 1547 lagg_rr_attach(struct lagg_softc *sc) 1548 { 1549 sc->sc_detach = lagg_rr_detach; 1550 sc->sc_start = lagg_rr_start; 1551 sc->sc_input = lagg_rr_input; 1552 sc->sc_port_create = NULL; 1553 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1554 sc->sc_seq = 0; 1555 1556 return (0); 1557 } 1558 1559 static int 1560 lagg_rr_detach(struct lagg_softc *sc) 1561 { 1562 return (0); 1563 } 1564 1565 static int 1566 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 1567 { 1568 struct lagg_port *lp; 1569 uint32_t p; 1570 1571 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1572 p %= sc->sc_count; 1573 lp = SLIST_FIRST(&sc->sc_ports); 1574 while (p--) 1575 lp = SLIST_NEXT(lp, lp_entries); 1576 1577 /* 1578 * Check the port's link state. This will return the next active 1579 * port if the link is down or the port is NULL. 1580 */ 1581 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1582 m_freem(m); 1583 return (ENOENT); 1584 } 1585 1586 /* Send mbuf */ 1587 return (lagg_enqueue(lp->lp_ifp, m)); 1588 } 1589 1590 static struct mbuf * 1591 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1592 { 1593 struct ifnet *ifp = sc->sc_ifp; 1594 1595 /* Just pass in the packet to our lagg device */ 1596 m->m_pkthdr.rcvif = ifp; 1597 1598 return (m); 1599 } 1600 1601 /* 1602 * Active failover 1603 */ 1604 1605 static int 1606 lagg_fail_attach(struct lagg_softc *sc) 1607 { 1608 sc->sc_detach = lagg_fail_detach; 1609 sc->sc_start = lagg_fail_start; 1610 sc->sc_input = lagg_fail_input; 1611 sc->sc_port_create = NULL; 1612 sc->sc_port_destroy = NULL; 1613 1614 return (0); 1615 } 1616 1617 static int 1618 lagg_fail_detach(struct lagg_softc *sc) 1619 { 1620 return (0); 1621 } 1622 1623 static int 1624 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 1625 { 1626 struct lagg_port *lp; 1627 1628 /* Use the master port if active or the next available port */ 1629 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 1630 m_freem(m); 1631 return (ENOENT); 1632 } 1633 1634 /* Send mbuf */ 1635 return (lagg_enqueue(lp->lp_ifp, m)); 1636 } 1637 1638 static struct mbuf * 1639 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1640 { 1641 struct ifnet *ifp = sc->sc_ifp; 1642 struct lagg_port *tmp_tp; 1643 1644 if (lp == sc->sc_primary || lagg_failover_rx_all) { 1645 m->m_pkthdr.rcvif = ifp; 1646 return (m); 1647 } 1648 1649 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1650 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1651 /* 1652 * If tmp_tp is null, we've recieved a packet when all 1653 * our links are down. Weird, but process it anyways. 1654 */ 1655 if ((tmp_tp == NULL || tmp_tp == lp)) { 1656 m->m_pkthdr.rcvif = ifp; 1657 return (m); 1658 } 1659 } 1660 1661 m_freem(m); 1662 return (NULL); 1663 } 1664 1665 /* 1666 * Loadbalancing 1667 */ 1668 1669 static int 1670 lagg_lb_attach(struct lagg_softc *sc) 1671 { 1672 struct lagg_port *lp; 1673 struct lagg_lb *lb; 1674 1675 if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb), 1676 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 1677 return (ENOMEM); 1678 1679 sc->sc_detach = lagg_lb_detach; 1680 sc->sc_start = lagg_lb_start; 1681 sc->sc_input = lagg_lb_input; 1682 sc->sc_port_create = lagg_lb_port_create; 1683 sc->sc_port_destroy = lagg_lb_port_destroy; 1684 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1685 1686 lb->lb_key = arc4random(); 1687 sc->sc_psc = (caddr_t)lb; 1688 1689 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1690 lagg_lb_port_create(lp); 1691 1692 return (0); 1693 } 1694 1695 static int 1696 lagg_lb_detach(struct lagg_softc *sc) 1697 { 1698 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1699 if (lb != NULL) 1700 free(lb, M_DEVBUF); 1701 return (0); 1702 } 1703 1704 static int 1705 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 1706 { 1707 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1708 struct lagg_port *lp_next; 1709 int i = 0; 1710 1711 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 1712 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1713 if (lp_next == lp) 1714 continue; 1715 if (i >= LAGG_MAX_PORTS) 1716 return (EINVAL); 1717 if (sc->sc_ifflags & IFF_DEBUG) 1718 printf("%s: port %s at index %d\n", 1719 sc->sc_ifname, lp_next->lp_ifname, i); 1720 lb->lb_ports[i++] = lp_next; 1721 } 1722 1723 return (0); 1724 } 1725 1726 static int 1727 lagg_lb_port_create(struct lagg_port *lp) 1728 { 1729 struct lagg_softc *sc = lp->lp_softc; 1730 return (lagg_lb_porttable(sc, NULL)); 1731 } 1732 1733 static void 1734 lagg_lb_port_destroy(struct lagg_port *lp) 1735 { 1736 struct lagg_softc *sc = lp->lp_softc; 1737 lagg_lb_porttable(sc, lp); 1738 } 1739 1740 static int 1741 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 1742 { 1743 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1744 struct lagg_port *lp = NULL; 1745 uint32_t p = 0; 1746 1747 if (sc->use_flowid && (m->m_flags & M_FLOWID)) 1748 p = m->m_pkthdr.flowid; 1749 else 1750 p = lagg_hashmbuf(sc, m, lb->lb_key); 1751 p %= sc->sc_count; 1752 lp = lb->lb_ports[p]; 1753 1754 /* 1755 * Check the port's link state. This will return the next active 1756 * port if the link is down or the port is NULL. 1757 */ 1758 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1759 m_freem(m); 1760 return (ENOENT); 1761 } 1762 1763 /* Send mbuf */ 1764 return (lagg_enqueue(lp->lp_ifp, m)); 1765 } 1766 1767 static struct mbuf * 1768 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1769 { 1770 struct ifnet *ifp = sc->sc_ifp; 1771 1772 /* Just pass in the packet to our lagg device */ 1773 m->m_pkthdr.rcvif = ifp; 1774 1775 return (m); 1776 } 1777 1778 /* 1779 * 802.3ad LACP 1780 */ 1781 1782 static int 1783 lagg_lacp_attach(struct lagg_softc *sc) 1784 { 1785 struct lagg_port *lp; 1786 int error; 1787 1788 sc->sc_detach = lagg_lacp_detach; 1789 sc->sc_port_create = lacp_port_create; 1790 sc->sc_port_destroy = lacp_port_destroy; 1791 sc->sc_linkstate = lacp_linkstate; 1792 sc->sc_start = lagg_lacp_start; 1793 sc->sc_input = lagg_lacp_input; 1794 sc->sc_init = lacp_init; 1795 sc->sc_stop = lacp_stop; 1796 sc->sc_lladdr = lagg_lacp_lladdr; 1797 sc->sc_req = lacp_req; 1798 sc->sc_portreq = lacp_portreq; 1799 1800 error = lacp_attach(sc); 1801 if (error) 1802 return (error); 1803 1804 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1805 lacp_port_create(lp); 1806 1807 return (error); 1808 } 1809 1810 static int 1811 lagg_lacp_detach(struct lagg_softc *sc) 1812 { 1813 struct lagg_port *lp; 1814 int error; 1815 1816 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1817 lacp_port_destroy(lp); 1818 1819 /* unlocking is safe here */ 1820 LAGG_WUNLOCK(sc); 1821 error = lacp_detach(sc); 1822 LAGG_WLOCK(sc); 1823 1824 return (error); 1825 } 1826 1827 static void 1828 lagg_lacp_lladdr(struct lagg_softc *sc) 1829 { 1830 struct lagg_port *lp; 1831 1832 /* purge all the lacp ports */ 1833 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1834 lacp_port_destroy(lp); 1835 1836 /* add them back in */ 1837 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1838 lacp_port_create(lp); 1839 } 1840 1841 static int 1842 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 1843 { 1844 struct lagg_port *lp; 1845 1846 lp = lacp_select_tx_port(sc, m); 1847 if (lp == NULL) { 1848 m_freem(m); 1849 return (EBUSY); 1850 } 1851 1852 /* Send mbuf */ 1853 return (lagg_enqueue(lp->lp_ifp, m)); 1854 } 1855 1856 static struct mbuf * 1857 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1858 { 1859 struct ifnet *ifp = sc->sc_ifp; 1860 struct ether_header *eh; 1861 u_short etype; 1862 1863 eh = mtod(m, struct ether_header *); 1864 etype = ntohs(eh->ether_type); 1865 1866 /* Tap off LACP control messages */ 1867 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 1868 m = lacp_input(lp, m); 1869 if (m == NULL) 1870 return (NULL); 1871 } 1872 1873 /* 1874 * If the port is not collecting or not in the active aggregator then 1875 * free and return. 1876 */ 1877 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 1878 m_freem(m); 1879 return (NULL); 1880 } 1881 1882 m->m_pkthdr.rcvif = ifp; 1883 return (m); 1884 } 1885