if_lagg.c revision 273736
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: stable/10/sys/net/if_lagg.c 273736 2014-10-27 14:38:00Z hselasky $"); 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/rmlock.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#include <netinet/ip.h> 58#endif 59#ifdef INET 60#include <netinet/in_systm.h> 61#include <netinet/if_ether.h> 62#endif 63 64#ifdef INET6 65#include <netinet/ip6.h> 66#include <netinet6/in6_var.h> 67#include <netinet6/in6_ifattach.h> 68#endif 69 70#include <net/if_vlan_var.h> 71#include <net/if_lagg.h> 72#include <net/ieee8023ad_lacp.h> 73 74/* Special flags we should propagate to the lagg ports. */ 75static struct { 76 int flag; 77 int (*func)(struct ifnet *, int); 78} lagg_pflags[] = { 79 {IFF_PROMISC, ifpromisc}, 80 {IFF_ALLMULTI, if_allmulti}, 81 {0, NULL} 82}; 83 84SLIST_HEAD(__trhead, lagg_softc) lagg_list; /* list of laggs */ 85static struct mtx lagg_list_mtx; 86eventhandler_tag lagg_detach_cookie = NULL; 87 88static int lagg_clone_create(struct if_clone *, int, caddr_t); 89static void lagg_clone_destroy(struct ifnet *); 90static struct if_clone *lagg_cloner; 91static const char laggname[] = "lagg"; 92 93static void lagg_lladdr(struct lagg_softc *, uint8_t *); 94static void lagg_capabilities(struct lagg_softc *); 95static void lagg_port_lladdr(struct lagg_port *, uint8_t *); 96static void lagg_port_setlladdr(void *, int); 97static int lagg_port_create(struct lagg_softc *, struct ifnet *); 98static int lagg_port_destroy(struct lagg_port *, int); 99static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); 100static void lagg_linkstate(struct lagg_softc *); 101static void lagg_port_state(struct ifnet *, int); 102static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); 103static int lagg_port_output(struct ifnet *, struct mbuf *, 104 const struct sockaddr *, struct route *); 105static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 106#ifdef LAGG_PORT_STACKING 107static int lagg_port_checkstacking(struct lagg_softc *); 108#endif 109static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 110static void lagg_init(void *); 111static void lagg_stop(struct lagg_softc *); 112static int lagg_ioctl(struct ifnet *, u_long, caddr_t); 113static int lagg_ether_setmulti(struct lagg_softc *); 114static int lagg_ether_cmdmulti(struct lagg_port *, int); 115static int lagg_setflag(struct lagg_port *, int, int, 116 int (*func)(struct ifnet *, int)); 117static int lagg_setflags(struct lagg_port *, int status); 118static int lagg_transmit(struct ifnet *, struct mbuf *); 119static void lagg_qflush(struct ifnet *); 120static int lagg_media_change(struct ifnet *); 121static void lagg_media_status(struct ifnet *, struct ifmediareq *); 122static struct lagg_port *lagg_link_active(struct lagg_softc *, 123 struct lagg_port *); 124static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *); 125static int lagg_sysctl_active(SYSCTL_HANDLER_ARGS); 126 127/* Simple round robin */ 128static int lagg_rr_attach(struct lagg_softc *); 129static int lagg_rr_detach(struct lagg_softc *); 130static int lagg_rr_start(struct lagg_softc *, struct mbuf *); 131static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 132 struct mbuf *); 133 134/* Active failover */ 135static int lagg_fail_attach(struct lagg_softc *); 136static int lagg_fail_detach(struct lagg_softc *); 137static int lagg_fail_start(struct lagg_softc *, struct mbuf *); 138static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 139 struct mbuf *); 140 141/* Loadbalancing */ 142static int lagg_lb_attach(struct lagg_softc *); 143static int lagg_lb_detach(struct lagg_softc *); 144static int lagg_lb_port_create(struct lagg_port *); 145static void lagg_lb_port_destroy(struct lagg_port *); 146static int lagg_lb_start(struct lagg_softc *, struct mbuf *); 147static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 148 struct mbuf *); 149static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 150 151/* 802.3ad LACP */ 152static int lagg_lacp_attach(struct lagg_softc *); 153static int lagg_lacp_detach(struct lagg_softc *); 154static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); 155static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 156 struct mbuf *); 157static void lagg_lacp_lladdr(struct lagg_softc *); 158 159static void lagg_callout(void *); 160 161/* lagg protocol table */ 162static const struct { 163 int ti_proto; 164 int (*ti_attach)(struct lagg_softc *); 165} lagg_protos[] = { 166 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach }, 167 { LAGG_PROTO_FAILOVER, lagg_fail_attach }, 168 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach }, 169 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach }, 170 { LAGG_PROTO_LACP, lagg_lacp_attach }, 171 { LAGG_PROTO_NONE, NULL } 172}; 173 174SYSCTL_DECL(_net_link); 175SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 176 "Link Aggregation"); 177 178static int lagg_failover_rx_all = 0; /* Allow input on any failover links */ 179SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW, 180 &lagg_failover_rx_all, 0, 181 "Accept input from any interface in a failover lagg"); 182static int def_use_flowid = 1; /* Default value for using M_FLOWID */ 183TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid); 184SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RW, 185 &def_use_flowid, 0, 186 "Default setting for using flow id for load sharing"); 187static int def_flowid_shift = 16; /* Default value for using M_FLOWID */ 188TUNABLE_INT("net.link.lagg.default_flowid_shift", &def_flowid_shift); 189SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RW, 190 &def_flowid_shift, 0, 191 "Default setting for flowid shift for load sharing"); 192 193static int 194lagg_modevent(module_t mod, int type, void *data) 195{ 196 197 switch (type) { 198 case MOD_LOAD: 199 mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF); 200 SLIST_INIT(&lagg_list); 201 lagg_cloner = if_clone_simple(laggname, lagg_clone_create, 202 lagg_clone_destroy, 0); 203 lagg_input_p = lagg_input; 204 lagg_linkstate_p = lagg_port_state; 205 lagg_detach_cookie = EVENTHANDLER_REGISTER( 206 ifnet_departure_event, lagg_port_ifdetach, NULL, 207 EVENTHANDLER_PRI_ANY); 208 break; 209 case MOD_UNLOAD: 210 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 211 lagg_detach_cookie); 212 if_clone_detach(lagg_cloner); 213 lagg_input_p = NULL; 214 lagg_linkstate_p = NULL; 215 mtx_destroy(&lagg_list_mtx); 216 break; 217 default: 218 return (EOPNOTSUPP); 219 } 220 return (0); 221} 222 223static moduledata_t lagg_mod = { 224 "if_lagg", 225 lagg_modevent, 226 0 227}; 228 229DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 230MODULE_VERSION(if_lagg, 1); 231 232/* 233 * This routine is run via an vlan 234 * config EVENT 235 */ 236static void 237lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 238{ 239 struct lagg_softc *sc = ifp->if_softc; 240 struct lagg_port *lp; 241 struct rm_priotracker tracker; 242 243 if (ifp->if_softc != arg) /* Not our event */ 244 return; 245 246 LAGG_RLOCK(sc, &tracker); 247 if (!SLIST_EMPTY(&sc->sc_ports)) { 248 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 249 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 250 } 251 LAGG_RUNLOCK(sc, &tracker); 252} 253 254/* 255 * This routine is run via an vlan 256 * unconfig EVENT 257 */ 258static void 259lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 260{ 261 struct lagg_softc *sc = ifp->if_softc; 262 struct lagg_port *lp; 263 struct rm_priotracker tracker; 264 265 if (ifp->if_softc != arg) /* Not our event */ 266 return; 267 268 LAGG_RLOCK(sc, &tracker); 269 if (!SLIST_EMPTY(&sc->sc_ports)) { 270 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 271 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 272 } 273 LAGG_RUNLOCK(sc, &tracker); 274} 275 276static int 277lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) 278{ 279 struct lagg_softc *sc; 280 struct ifnet *ifp; 281 int i, error = 0; 282 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ 283 struct sysctl_oid *oid; 284 char num[14]; /* sufficient for 32 bits */ 285 286 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO); 287 ifp = sc->sc_ifp = if_alloc(IFT_ETHER); 288 if (ifp == NULL) { 289 free(sc, M_DEVBUF); 290 return (ENOSPC); 291 } 292 293 sc->sc_ipackets = counter_u64_alloc(M_WAITOK); 294 sc->sc_opackets = counter_u64_alloc(M_WAITOK); 295 sc->sc_ibytes = counter_u64_alloc(M_WAITOK); 296 sc->sc_obytes = counter_u64_alloc(M_WAITOK); 297 298 sysctl_ctx_init(&sc->ctx); 299 snprintf(num, sizeof(num), "%u", unit); 300 sc->use_flowid = def_use_flowid; 301 sc->flowid_shift = def_flowid_shift; 302 sc->sc_oid = oid = SYSCTL_ADD_NODE(&sc->ctx, 303 &SYSCTL_NODE_CHILDREN(_net_link, lagg), 304 OID_AUTO, num, CTLFLAG_RD, NULL, ""); 305 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 306 "use_flowid", CTLFLAG_RW, &sc->use_flowid, 307 sc->use_flowid, "Use flow id for load sharing"); 308 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 309 "flowid_shift", CTLFLAG_RW, &sc->flowid_shift, 310 sc->flowid_shift, 311 "Shift flowid bits to prevent multiqueue collisions"); 312 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 313 "count", CTLFLAG_RD, &sc->sc_count, sc->sc_count, 314 "Total number of ports"); 315 SYSCTL_ADD_PROC(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 316 "active", CTLTYPE_INT|CTLFLAG_RD, sc, 0, lagg_sysctl_active, 317 "I", "Total number of active ports"); 318 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 319 "flapping", CTLFLAG_RD, &sc->sc_flapping, 320 sc->sc_flapping, "Total number of port change events"); 321 /* Hash all layers by default */ 322 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4; 323 324 sc->sc_proto = LAGG_PROTO_NONE; 325 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) { 326 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) { 327 sc->sc_proto = lagg_protos[i].ti_proto; 328 if ((error = lagg_protos[i].ti_attach(sc)) != 0) { 329 if_free(ifp); 330 free(sc, M_DEVBUF); 331 return (error); 332 } 333 break; 334 } 335 } 336 LAGG_LOCK_INIT(sc); 337 LAGG_CALLOUT_LOCK_INIT(sc); 338 SLIST_INIT(&sc->sc_ports); 339 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc); 340 341 /* 342 * This uses the callout lock rather than the rmlock; one can't 343 * hold said rmlock during SWI. 344 */ 345 callout_init_mtx(&sc->sc_callout, &sc->sc_call_mtx, 0); 346 347 /* Initialise pseudo media types */ 348 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 349 lagg_media_status); 350 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 351 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 352 353 if_initname(ifp, laggname, unit); 354 ifp->if_softc = sc; 355 ifp->if_transmit = lagg_transmit; 356 ifp->if_qflush = lagg_qflush; 357 ifp->if_init = lagg_init; 358 ifp->if_ioctl = lagg_ioctl; 359 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 360 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS; 361 362 /* 363 * Attach as an ordinary ethernet device, children will be attached 364 * as special device IFT_IEEE8023ADLAG. 365 */ 366 ether_ifattach(ifp, eaddr); 367 368 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 369 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 370 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 371 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 372 373 /* Insert into the global list of laggs */ 374 mtx_lock(&lagg_list_mtx); 375 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries); 376 mtx_unlock(&lagg_list_mtx); 377 378 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 379 380 return (0); 381} 382 383static void 384lagg_clone_destroy(struct ifnet *ifp) 385{ 386 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 387 struct lagg_port *lp; 388 389 LAGG_WLOCK(sc); 390 391 lagg_stop(sc); 392 ifp->if_flags &= ~IFF_UP; 393 394 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 395 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 396 397 /* Shutdown and remove lagg ports */ 398 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) 399 lagg_port_destroy(lp, 1); 400 /* Unhook the aggregation protocol */ 401 if (sc->sc_detach != NULL) 402 (*sc->sc_detach)(sc); 403 404 LAGG_WUNLOCK(sc); 405 406 sysctl_ctx_free(&sc->ctx); 407 ifmedia_removeall(&sc->sc_media); 408 ether_ifdetach(ifp); 409 if_free(ifp); 410 411 /* This grabs sc_callout_mtx, serialising it correctly */ 412 callout_drain(&sc->sc_callout); 413 414 /* At this point it's drained; we can free this */ 415 counter_u64_free(sc->sc_ipackets); 416 counter_u64_free(sc->sc_opackets); 417 counter_u64_free(sc->sc_ibytes); 418 counter_u64_free(sc->sc_obytes); 419 420 mtx_lock(&lagg_list_mtx); 421 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries); 422 mtx_unlock(&lagg_list_mtx); 423 424 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task); 425 LAGG_LOCK_DESTROY(sc); 426 LAGG_CALLOUT_LOCK_DESTROY(sc); 427 free(sc, M_DEVBUF); 428} 429 430static void 431lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr) 432{ 433 struct ifnet *ifp = sc->sc_ifp; 434 435 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 436 return; 437 438 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN); 439 /* Let the protocol know the MAC has changed */ 440 if (sc->sc_lladdr != NULL) 441 (*sc->sc_lladdr)(sc); 442 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 443} 444 445static void 446lagg_capabilities(struct lagg_softc *sc) 447{ 448 struct lagg_port *lp; 449 int cap = ~0, ena = ~0; 450 u_long hwa = ~0UL; 451#if defined(INET) || defined(INET6) 452 u_int hw_tsomax = IP_MAXPACKET; /* Initialize to the maximum value. */ 453#else 454 u_int hw_tsomax = ~0; /* if_hw_tsomax is only for INET/INET6, but.. */ 455#endif 456 457 LAGG_WLOCK_ASSERT(sc); 458 459 /* Get capabilities from the lagg ports */ 460 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 461 cap &= lp->lp_ifp->if_capabilities; 462 ena &= lp->lp_ifp->if_capenable; 463 hwa &= lp->lp_ifp->if_hwassist; 464 /* Set to the minimum value of the lagg ports. */ 465 if (lp->lp_ifp->if_hw_tsomax < hw_tsomax && 466 lp->lp_ifp->if_hw_tsomax > 0) 467 hw_tsomax = lp->lp_ifp->if_hw_tsomax; 468 } 469 cap = (cap == ~0 ? 0 : cap); 470 ena = (ena == ~0 ? 0 : ena); 471 hwa = (hwa == ~0 ? 0 : hwa); 472 473 if (sc->sc_ifp->if_capabilities != cap || 474 sc->sc_ifp->if_capenable != ena || 475 sc->sc_ifp->if_hwassist != hwa || 476 sc->sc_ifp->if_hw_tsomax != hw_tsomax) { 477 sc->sc_ifp->if_capabilities = cap; 478 sc->sc_ifp->if_capenable = ena; 479 sc->sc_ifp->if_hwassist = hwa; 480 sc->sc_ifp->if_hw_tsomax = hw_tsomax; 481 getmicrotime(&sc->sc_ifp->if_lastchange); 482 483 if (sc->sc_ifflags & IFF_DEBUG) 484 if_printf(sc->sc_ifp, 485 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 486 } 487} 488 489static void 490lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr) 491{ 492 struct lagg_softc *sc = lp->lp_softc; 493 struct ifnet *ifp = lp->lp_ifp; 494 struct lagg_llq *llq; 495 int pending = 0; 496 497 LAGG_WLOCK_ASSERT(sc); 498 499 if (lp->lp_detaching || 500 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 501 return; 502 503 /* Check to make sure its not already queued to be changed */ 504 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 505 if (llq->llq_ifp == ifp) { 506 pending = 1; 507 break; 508 } 509 } 510 511 if (!pending) { 512 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT); 513 if (llq == NULL) /* XXX what to do */ 514 return; 515 } 516 517 /* Update the lladdr even if pending, it may have changed */ 518 llq->llq_ifp = ifp; 519 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN); 520 521 if (!pending) 522 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries); 523 524 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task); 525} 526 527/* 528 * Set the interface MAC address from a taskqueue to avoid a LOR. 529 */ 530static void 531lagg_port_setlladdr(void *arg, int pending) 532{ 533 struct lagg_softc *sc = (struct lagg_softc *)arg; 534 struct lagg_llq *llq, *head; 535 struct ifnet *ifp; 536 int error; 537 538 /* Grab a local reference of the queue and remove it from the softc */ 539 LAGG_WLOCK(sc); 540 head = SLIST_FIRST(&sc->sc_llq_head); 541 SLIST_FIRST(&sc->sc_llq_head) = NULL; 542 LAGG_WUNLOCK(sc); 543 544 /* 545 * Traverse the queue and set the lladdr on each ifp. It is safe to do 546 * unlocked as we have the only reference to it. 547 */ 548 for (llq = head; llq != NULL; llq = head) { 549 ifp = llq->llq_ifp; 550 551 /* Set the link layer address */ 552 CURVNET_SET(ifp->if_vnet); 553 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN); 554 CURVNET_RESTORE(); 555 if (error) 556 printf("%s: setlladdr failed on %s\n", __func__, 557 ifp->if_xname); 558 559 head = SLIST_NEXT(llq, llq_entries); 560 free(llq, M_DEVBUF); 561 } 562} 563 564static int 565lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 566{ 567 struct lagg_softc *sc_ptr; 568 struct lagg_port *lp, *tlp; 569 int error = 0; 570 571 LAGG_WLOCK_ASSERT(sc); 572 573 /* Limit the maximal number of lagg ports */ 574 if (sc->sc_count >= LAGG_MAX_PORTS) 575 return (ENOSPC); 576 577 /* Check if port has already been associated to a lagg */ 578 if (ifp->if_lagg != NULL) { 579 /* Port is already in the current lagg? */ 580 lp = (struct lagg_port *)ifp->if_lagg; 581 if (lp->lp_softc == sc) 582 return (EEXIST); 583 return (EBUSY); 584 } 585 586 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 587 if (ifp->if_type != IFT_ETHER) 588 return (EPROTONOSUPPORT); 589 590#ifdef INET6 591 /* 592 * The member interface should not have inet6 address because 593 * two interfaces with a valid link-local scope zone must not be 594 * merged in any form. This restriction is needed to 595 * prevent violation of link-local scope zone. Attempts to 596 * add a member interface which has inet6 addresses triggers 597 * removal of all inet6 addresses on the member interface. 598 */ 599 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 600 if (in6ifa_llaonifp(lp->lp_ifp)) { 601 in6_ifdetach(lp->lp_ifp); 602 if_printf(sc->sc_ifp, 603 "IPv6 addresses on %s have been removed " 604 "before adding it as a member to prevent " 605 "IPv6 address scope violation.\n", 606 lp->lp_ifp->if_xname); 607 } 608 } 609 if (in6ifa_llaonifp(ifp)) { 610 in6_ifdetach(ifp); 611 if_printf(sc->sc_ifp, 612 "IPv6 addresses on %s have been removed " 613 "before adding it as a member to prevent " 614 "IPv6 address scope violation.\n", 615 ifp->if_xname); 616 } 617#endif 618 /* Allow the first Ethernet member to define the MTU */ 619 if (SLIST_EMPTY(&sc->sc_ports)) 620 sc->sc_ifp->if_mtu = ifp->if_mtu; 621 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 622 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 623 ifp->if_xname); 624 return (EINVAL); 625 } 626 627 if ((lp = malloc(sizeof(struct lagg_port), 628 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 629 return (ENOMEM); 630 631 /* Check if port is a stacked lagg */ 632 mtx_lock(&lagg_list_mtx); 633 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) { 634 if (ifp == sc_ptr->sc_ifp) { 635 mtx_unlock(&lagg_list_mtx); 636 free(lp, M_DEVBUF); 637 return (EINVAL); 638 /* XXX disable stacking for the moment, its untested */ 639#ifdef LAGG_PORT_STACKING 640 lp->lp_flags |= LAGG_PORT_STACK; 641 if (lagg_port_checkstacking(sc_ptr) >= 642 LAGG_MAX_STACKING) { 643 mtx_unlock(&lagg_list_mtx); 644 free(lp, M_DEVBUF); 645 return (E2BIG); 646 } 647#endif 648 } 649 } 650 mtx_unlock(&lagg_list_mtx); 651 652 /* Change the interface type */ 653 lp->lp_iftype = ifp->if_type; 654 ifp->if_type = IFT_IEEE8023ADLAG; 655 ifp->if_lagg = lp; 656 lp->lp_ioctl = ifp->if_ioctl; 657 ifp->if_ioctl = lagg_port_ioctl; 658 lp->lp_output = ifp->if_output; 659 ifp->if_output = lagg_port_output; 660 661 lp->lp_ifp = ifp; 662 lp->lp_softc = sc; 663 664 /* Save port link layer address */ 665 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 666 667 if (SLIST_EMPTY(&sc->sc_ports)) { 668 sc->sc_primary = lp; 669 lagg_lladdr(sc, IF_LLADDR(ifp)); 670 } else { 671 /* Update link layer address for this port */ 672 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp)); 673 } 674 675 /* Insert into the list of ports. Keep ports sorted by if_index. */ 676 SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) { 677 if (tlp->lp_ifp->if_index < ifp->if_index && ( 678 SLIST_NEXT(tlp, lp_entries) == NULL || 679 SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index < 680 ifp->if_index)) 681 break; 682 } 683 if (tlp != NULL) 684 SLIST_INSERT_AFTER(tlp, lp, lp_entries); 685 else 686 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 687 sc->sc_count++; 688 689 /* Update lagg capabilities */ 690 lagg_capabilities(sc); 691 lagg_linkstate(sc); 692 693 /* Add multicast addresses and interface flags to this port */ 694 lagg_ether_cmdmulti(lp, 1); 695 lagg_setflags(lp, 1); 696 697 if (sc->sc_port_create != NULL) 698 error = (*sc->sc_port_create)(lp); 699 if (error) { 700 /* remove the port again, without calling sc_port_destroy */ 701 lagg_port_destroy(lp, 0); 702 return (error); 703 } 704 705 return (error); 706} 707 708#ifdef LAGG_PORT_STACKING 709static int 710lagg_port_checkstacking(struct lagg_softc *sc) 711{ 712 struct lagg_softc *sc_ptr; 713 struct lagg_port *lp; 714 int m = 0; 715 716 LAGG_WLOCK_ASSERT(sc); 717 718 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 719 if (lp->lp_flags & LAGG_PORT_STACK) { 720 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 721 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 722 } 723 } 724 725 return (m + 1); 726} 727#endif 728 729static int 730lagg_port_destroy(struct lagg_port *lp, int runpd) 731{ 732 struct lagg_softc *sc = lp->lp_softc; 733 struct lagg_port *lp_ptr; 734 struct lagg_llq *llq; 735 struct ifnet *ifp = lp->lp_ifp; 736 737 LAGG_WLOCK_ASSERT(sc); 738 739 if (runpd && sc->sc_port_destroy != NULL) 740 (*sc->sc_port_destroy)(lp); 741 742 /* 743 * Remove multicast addresses and interface flags from this port and 744 * reset the MAC address, skip if the interface is being detached. 745 */ 746 if (!lp->lp_detaching) { 747 lagg_ether_cmdmulti(lp, 0); 748 lagg_setflags(lp, 0); 749 lagg_port_lladdr(lp, lp->lp_lladdr); 750 } 751 752 /* Restore interface */ 753 ifp->if_type = lp->lp_iftype; 754 ifp->if_ioctl = lp->lp_ioctl; 755 ifp->if_output = lp->lp_output; 756 ifp->if_lagg = NULL; 757 758 /* Finally, remove the port from the lagg */ 759 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 760 sc->sc_count--; 761 762 /* Update the primary interface */ 763 if (lp == sc->sc_primary) { 764 uint8_t lladdr[ETHER_ADDR_LEN]; 765 766 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) { 767 bzero(&lladdr, ETHER_ADDR_LEN); 768 } else { 769 bcopy(lp_ptr->lp_lladdr, 770 lladdr, ETHER_ADDR_LEN); 771 } 772 lagg_lladdr(sc, lladdr); 773 sc->sc_primary = lp_ptr; 774 775 /* Update link layer address for each port */ 776 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 777 lagg_port_lladdr(lp_ptr, lladdr); 778 } 779 780 /* Remove any pending lladdr changes from the queue */ 781 if (lp->lp_detaching) { 782 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 783 if (llq->llq_ifp == ifp) { 784 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq, 785 llq_entries); 786 free(llq, M_DEVBUF); 787 break; /* Only appears once */ 788 } 789 } 790 } 791 792 if (lp->lp_ifflags) 793 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 794 795 free(lp, M_DEVBUF); 796 797 /* Update lagg capabilities */ 798 lagg_capabilities(sc); 799 lagg_linkstate(sc); 800 801 return (0); 802} 803 804static int 805lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 806{ 807 struct lagg_reqport *rp = (struct lagg_reqport *)data; 808 struct lagg_softc *sc; 809 struct lagg_port *lp = NULL; 810 int error = 0; 811 struct rm_priotracker tracker; 812 813 /* Should be checked by the caller */ 814 if (ifp->if_type != IFT_IEEE8023ADLAG || 815 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 816 goto fallback; 817 818 switch (cmd) { 819 case SIOCGLAGGPORT: 820 if (rp->rp_portname[0] == '\0' || 821 ifunit(rp->rp_portname) != ifp) { 822 error = EINVAL; 823 break; 824 } 825 826 LAGG_RLOCK(sc, &tracker); 827 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 828 error = ENOENT; 829 LAGG_RUNLOCK(sc, &tracker); 830 break; 831 } 832 833 lagg_port2req(lp, rp); 834 LAGG_RUNLOCK(sc, &tracker); 835 break; 836 837 case SIOCSIFCAP: 838 if (lp->lp_ioctl == NULL) { 839 error = EINVAL; 840 break; 841 } 842 error = (*lp->lp_ioctl)(ifp, cmd, data); 843 if (error) 844 break; 845 846 /* Update lagg interface capabilities */ 847 LAGG_WLOCK(sc); 848 lagg_capabilities(sc); 849 LAGG_WUNLOCK(sc); 850 break; 851 852 case SIOCSIFMTU: 853 /* Do not allow the MTU to be changed once joined */ 854 error = EINVAL; 855 break; 856 857 default: 858 goto fallback; 859 } 860 861 return (error); 862 863fallback: 864 if (lp->lp_ioctl != NULL) 865 return ((*lp->lp_ioctl)(ifp, cmd, data)); 866 867 return (EINVAL); 868} 869 870/* 871 * For direct output to child ports. 872 */ 873static int 874lagg_port_output(struct ifnet *ifp, struct mbuf *m, 875 const struct sockaddr *dst, struct route *ro) 876{ 877 struct lagg_port *lp = ifp->if_lagg; 878 879 switch (dst->sa_family) { 880 case pseudo_AF_HDRCMPLT: 881 case AF_UNSPEC: 882 return ((*lp->lp_output)(ifp, m, dst, ro)); 883 } 884 885 /* drop any other frames */ 886 m_freem(m); 887 return (ENETDOWN); 888} 889 890static void 891lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 892{ 893 struct lagg_port *lp; 894 struct lagg_softc *sc; 895 896 if ((lp = ifp->if_lagg) == NULL) 897 return; 898 /* If the ifnet is just being renamed, don't do anything. */ 899 if (ifp->if_flags & IFF_RENAMING) 900 return; 901 902 sc = lp->lp_softc; 903 904 LAGG_WLOCK(sc); 905 lp->lp_detaching = 1; 906 lagg_port_destroy(lp, 1); 907 LAGG_WUNLOCK(sc); 908} 909 910static void 911lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 912{ 913 struct lagg_softc *sc = lp->lp_softc; 914 915 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 916 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 917 rp->rp_prio = lp->lp_prio; 918 rp->rp_flags = lp->lp_flags; 919 if (sc->sc_portreq != NULL) 920 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc); 921 922 /* Add protocol specific flags */ 923 switch (sc->sc_proto) { 924 case LAGG_PROTO_FAILOVER: 925 if (lp == sc->sc_primary) 926 rp->rp_flags |= LAGG_PORT_MASTER; 927 if (lp == lagg_link_active(sc, sc->sc_primary)) 928 rp->rp_flags |= LAGG_PORT_ACTIVE; 929 break; 930 931 case LAGG_PROTO_ROUNDROBIN: 932 case LAGG_PROTO_LOADBALANCE: 933 case LAGG_PROTO_ETHERCHANNEL: 934 if (LAGG_PORTACTIVE(lp)) 935 rp->rp_flags |= LAGG_PORT_ACTIVE; 936 break; 937 938 case LAGG_PROTO_LACP: 939 /* LACP has a different definition of active */ 940 if (lacp_isactive(lp)) 941 rp->rp_flags |= LAGG_PORT_ACTIVE; 942 if (lacp_iscollecting(lp)) 943 rp->rp_flags |= LAGG_PORT_COLLECTING; 944 if (lacp_isdistributing(lp)) 945 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 946 break; 947 } 948 949} 950 951static void 952lagg_init(void *xsc) 953{ 954 struct lagg_softc *sc = (struct lagg_softc *)xsc; 955 struct lagg_port *lp; 956 struct ifnet *ifp = sc->sc_ifp; 957 958 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 959 return; 960 961 LAGG_WLOCK(sc); 962 963 ifp->if_drv_flags |= IFF_DRV_RUNNING; 964 /* Update the port lladdrs */ 965 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 966 lagg_port_lladdr(lp, IF_LLADDR(ifp)); 967 968 if (sc->sc_init != NULL) 969 (*sc->sc_init)(sc); 970 971 LAGG_WUNLOCK(sc); 972} 973 974static void 975lagg_stop(struct lagg_softc *sc) 976{ 977 struct ifnet *ifp = sc->sc_ifp; 978 979 LAGG_WLOCK_ASSERT(sc); 980 981 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 982 return; 983 984 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 985 986 if (sc->sc_stop != NULL) 987 (*sc->sc_stop)(sc); 988} 989 990static int 991lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 992{ 993 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 994 struct lagg_reqall *ra = (struct lagg_reqall *)data; 995 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 996 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 997 struct ifreq *ifr = (struct ifreq *)data; 998 struct lagg_port *lp; 999 struct ifnet *tpif; 1000 struct thread *td = curthread; 1001 char *buf, *outbuf; 1002 int count, buflen, len, error = 0; 1003 struct rm_priotracker tracker; 1004 1005 bzero(&rpbuf, sizeof(rpbuf)); 1006 1007 switch (cmd) { 1008 case SIOCGLAGG: 1009 LAGG_RLOCK(sc, &tracker); 1010 count = 0; 1011 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1012 count++; 1013 buflen = count * sizeof(struct lagg_reqport); 1014 LAGG_RUNLOCK(sc, &tracker); 1015 1016 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 1017 1018 LAGG_RLOCK(sc, &tracker); 1019 ra->ra_proto = sc->sc_proto; 1020 if (sc->sc_req != NULL) 1021 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc); 1022 1023 count = 0; 1024 buf = outbuf; 1025 len = min(ra->ra_size, buflen); 1026 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1027 if (len < sizeof(rpbuf)) 1028 break; 1029 1030 lagg_port2req(lp, &rpbuf); 1031 memcpy(buf, &rpbuf, sizeof(rpbuf)); 1032 count++; 1033 buf += sizeof(rpbuf); 1034 len -= sizeof(rpbuf); 1035 } 1036 LAGG_RUNLOCK(sc, &tracker); 1037 ra->ra_ports = count; 1038 ra->ra_size = count * sizeof(rpbuf); 1039 error = copyout(outbuf, ra->ra_port, ra->ra_size); 1040 free(outbuf, M_TEMP); 1041 break; 1042 case SIOCSLAGG: 1043 error = priv_check(td, PRIV_NET_LAGG); 1044 if (error) 1045 break; 1046 if (ra->ra_proto >= LAGG_PROTO_MAX) { 1047 error = EPROTONOSUPPORT; 1048 break; 1049 } 1050 LAGG_WLOCK(sc); 1051 if (sc->sc_proto != LAGG_PROTO_NONE) { 1052 /* Reset protocol first in case detach unlocks */ 1053 sc->sc_proto = LAGG_PROTO_NONE; 1054 error = sc->sc_detach(sc); 1055 sc->sc_detach = NULL; 1056 sc->sc_start = NULL; 1057 sc->sc_input = NULL; 1058 sc->sc_port_create = NULL; 1059 sc->sc_port_destroy = NULL; 1060 sc->sc_linkstate = NULL; 1061 sc->sc_init = NULL; 1062 sc->sc_stop = NULL; 1063 sc->sc_lladdr = NULL; 1064 sc->sc_req = NULL; 1065 sc->sc_portreq = NULL; 1066 } else if (sc->sc_input != NULL) { 1067 /* Still detaching */ 1068 error = EBUSY; 1069 } 1070 if (error != 0) { 1071 LAGG_WUNLOCK(sc); 1072 break; 1073 } 1074 for (int i = 0; i < (sizeof(lagg_protos) / 1075 sizeof(lagg_protos[0])); i++) { 1076 if (lagg_protos[i].ti_proto == ra->ra_proto) { 1077 if (sc->sc_ifflags & IFF_DEBUG) 1078 printf("%s: using proto %u\n", 1079 sc->sc_ifname, 1080 lagg_protos[i].ti_proto); 1081 sc->sc_proto = lagg_protos[i].ti_proto; 1082 if (sc->sc_proto != LAGG_PROTO_NONE) 1083 error = lagg_protos[i].ti_attach(sc); 1084 LAGG_WUNLOCK(sc); 1085 return (error); 1086 } 1087 } 1088 LAGG_WUNLOCK(sc); 1089 error = EPROTONOSUPPORT; 1090 break; 1091 case SIOCGLAGGFLAGS: 1092 rf->rf_flags = sc->sc_flags; 1093 break; 1094 case SIOCSLAGGHASH: 1095 error = priv_check(td, PRIV_NET_LAGG); 1096 if (error) 1097 break; 1098 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1099 error = EINVAL; 1100 break; 1101 } 1102 LAGG_WLOCK(sc); 1103 sc->sc_flags &= ~LAGG_F_HASHMASK; 1104 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK; 1105 LAGG_WUNLOCK(sc); 1106 break; 1107 case SIOCGLAGGPORT: 1108 if (rp->rp_portname[0] == '\0' || 1109 (tpif = ifunit(rp->rp_portname)) == NULL) { 1110 error = EINVAL; 1111 break; 1112 } 1113 1114 LAGG_RLOCK(sc, &tracker); 1115 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1116 lp->lp_softc != sc) { 1117 error = ENOENT; 1118 LAGG_RUNLOCK(sc, &tracker); 1119 break; 1120 } 1121 1122 lagg_port2req(lp, rp); 1123 LAGG_RUNLOCK(sc, &tracker); 1124 break; 1125 case SIOCSLAGGPORT: 1126 error = priv_check(td, PRIV_NET_LAGG); 1127 if (error) 1128 break; 1129 if (rp->rp_portname[0] == '\0' || 1130 (tpif = ifunit(rp->rp_portname)) == NULL) { 1131 error = EINVAL; 1132 break; 1133 } 1134 LAGG_WLOCK(sc); 1135 error = lagg_port_create(sc, tpif); 1136 LAGG_WUNLOCK(sc); 1137 break; 1138 case SIOCSLAGGDELPORT: 1139 error = priv_check(td, PRIV_NET_LAGG); 1140 if (error) 1141 break; 1142 if (rp->rp_portname[0] == '\0' || 1143 (tpif = ifunit(rp->rp_portname)) == NULL) { 1144 error = EINVAL; 1145 break; 1146 } 1147 1148 LAGG_WLOCK(sc); 1149 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1150 lp->lp_softc != sc) { 1151 error = ENOENT; 1152 LAGG_WUNLOCK(sc); 1153 break; 1154 } 1155 1156 error = lagg_port_destroy(lp, 1); 1157 LAGG_WUNLOCK(sc); 1158 break; 1159 case SIOCSIFFLAGS: 1160 /* Set flags on ports too */ 1161 LAGG_WLOCK(sc); 1162 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1163 lagg_setflags(lp, 1); 1164 } 1165 LAGG_WUNLOCK(sc); 1166 1167 if (!(ifp->if_flags & IFF_UP) && 1168 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1169 /* 1170 * If interface is marked down and it is running, 1171 * then stop and disable it. 1172 */ 1173 LAGG_WLOCK(sc); 1174 lagg_stop(sc); 1175 LAGG_WUNLOCK(sc); 1176 } else if ((ifp->if_flags & IFF_UP) && 1177 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1178 /* 1179 * If interface is marked up and it is stopped, then 1180 * start it. 1181 */ 1182 (*ifp->if_init)(sc); 1183 } 1184 break; 1185 case SIOCADDMULTI: 1186 case SIOCDELMULTI: 1187 LAGG_WLOCK(sc); 1188 error = lagg_ether_setmulti(sc); 1189 LAGG_WUNLOCK(sc); 1190 break; 1191 case SIOCSIFMEDIA: 1192 case SIOCGIFMEDIA: 1193 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1194 break; 1195 1196 case SIOCSIFCAP: 1197 case SIOCSIFMTU: 1198 /* Do not allow the MTU or caps to be directly changed */ 1199 error = EINVAL; 1200 break; 1201 1202 default: 1203 error = ether_ioctl(ifp, cmd, data); 1204 break; 1205 } 1206 return (error); 1207} 1208 1209static int 1210lagg_ether_setmulti(struct lagg_softc *sc) 1211{ 1212 struct lagg_port *lp; 1213 1214 LAGG_WLOCK_ASSERT(sc); 1215 1216 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1217 /* First, remove any existing filter entries. */ 1218 lagg_ether_cmdmulti(lp, 0); 1219 /* copy all addresses from the lagg interface to the port */ 1220 lagg_ether_cmdmulti(lp, 1); 1221 } 1222 return (0); 1223} 1224 1225static int 1226lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1227{ 1228 struct lagg_softc *sc = lp->lp_softc; 1229 struct ifnet *ifp = lp->lp_ifp; 1230 struct ifnet *scifp = sc->sc_ifp; 1231 struct lagg_mc *mc; 1232 struct ifmultiaddr *ifma; 1233 int error; 1234 1235 LAGG_WLOCK_ASSERT(sc); 1236 1237 if (set) { 1238 IF_ADDR_WLOCK(scifp); 1239 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1240 if (ifma->ifma_addr->sa_family != AF_LINK) 1241 continue; 1242 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1243 if (mc == NULL) { 1244 IF_ADDR_WUNLOCK(scifp); 1245 return (ENOMEM); 1246 } 1247 bcopy(ifma->ifma_addr, &mc->mc_addr, 1248 ifma->ifma_addr->sa_len); 1249 mc->mc_addr.sdl_index = ifp->if_index; 1250 mc->mc_ifma = NULL; 1251 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1252 } 1253 IF_ADDR_WUNLOCK(scifp); 1254 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) { 1255 error = if_addmulti(ifp, 1256 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma); 1257 if (error) 1258 return (error); 1259 } 1260 } else { 1261 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1262 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1263 if (mc->mc_ifma && !lp->lp_detaching) 1264 if_delmulti_ifma(mc->mc_ifma); 1265 free(mc, M_DEVBUF); 1266 } 1267 } 1268 return (0); 1269} 1270 1271/* Handle a ref counted flag that should be set on the lagg port as well */ 1272static int 1273lagg_setflag(struct lagg_port *lp, int flag, int status, 1274 int (*func)(struct ifnet *, int)) 1275{ 1276 struct lagg_softc *sc = lp->lp_softc; 1277 struct ifnet *scifp = sc->sc_ifp; 1278 struct ifnet *ifp = lp->lp_ifp; 1279 int error; 1280 1281 LAGG_WLOCK_ASSERT(sc); 1282 1283 status = status ? (scifp->if_flags & flag) : 0; 1284 /* Now "status" contains the flag value or 0 */ 1285 1286 /* 1287 * See if recorded ports status is different from what 1288 * we want it to be. If it is, flip it. We record ports 1289 * status in lp_ifflags so that we won't clear ports flag 1290 * we haven't set. In fact, we don't clear or set ports 1291 * flags directly, but get or release references to them. 1292 * That's why we can be sure that recorded flags still are 1293 * in accord with actual ports flags. 1294 */ 1295 if (status != (lp->lp_ifflags & flag)) { 1296 error = (*func)(ifp, status); 1297 if (error) 1298 return (error); 1299 lp->lp_ifflags &= ~flag; 1300 lp->lp_ifflags |= status; 1301 } 1302 return (0); 1303} 1304 1305/* 1306 * Handle IFF_* flags that require certain changes on the lagg port 1307 * if "status" is true, update ports flags respective to the lagg 1308 * if "status" is false, forcedly clear the flags set on port. 1309 */ 1310static int 1311lagg_setflags(struct lagg_port *lp, int status) 1312{ 1313 int error, i; 1314 1315 for (i = 0; lagg_pflags[i].flag; i++) { 1316 error = lagg_setflag(lp, lagg_pflags[i].flag, 1317 status, lagg_pflags[i].func); 1318 if (error) 1319 return (error); 1320 } 1321 return (0); 1322} 1323 1324static int 1325lagg_transmit(struct ifnet *ifp, struct mbuf *m) 1326{ 1327 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1328 int error, len, mcast; 1329 struct rm_priotracker tracker; 1330 1331 len = m->m_pkthdr.len; 1332 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1333 1334 LAGG_RLOCK(sc, &tracker); 1335 /* We need a Tx algorithm and at least one port */ 1336 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1337 LAGG_RUNLOCK(sc, &tracker); 1338 m_freem(m); 1339 ifp->if_oerrors++; 1340 return (ENXIO); 1341 } 1342 1343 ETHER_BPF_MTAP(ifp, m); 1344 1345 error = (*sc->sc_start)(sc, m); 1346 LAGG_RUNLOCK(sc, &tracker); 1347 1348 if (error == 0) { 1349 counter_u64_add(sc->sc_opackets, 1); 1350 counter_u64_add(sc->sc_obytes, len); 1351 ifp->if_omcasts += mcast; 1352 } else 1353 ifp->if_oerrors++; 1354 1355 return (error); 1356} 1357 1358/* 1359 * The ifp->if_qflush entry point for lagg(4) is no-op. 1360 */ 1361static void 1362lagg_qflush(struct ifnet *ifp __unused) 1363{ 1364} 1365 1366static struct mbuf * 1367lagg_input(struct ifnet *ifp, struct mbuf *m) 1368{ 1369 struct lagg_port *lp = ifp->if_lagg; 1370 struct lagg_softc *sc = lp->lp_softc; 1371 struct ifnet *scifp = sc->sc_ifp; 1372 struct rm_priotracker tracker; 1373 1374 LAGG_RLOCK(sc, &tracker); 1375 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1376 (lp->lp_flags & LAGG_PORT_DISABLED) || 1377 sc->sc_proto == LAGG_PROTO_NONE) { 1378 LAGG_RUNLOCK(sc, &tracker); 1379 m_freem(m); 1380 return (NULL); 1381 } 1382 1383 ETHER_BPF_MTAP(scifp, m); 1384 1385 m = (*sc->sc_input)(sc, lp, m); 1386 1387 if (m != NULL) { 1388 counter_u64_add(sc->sc_ipackets, 1); 1389 counter_u64_add(sc->sc_ibytes, m->m_pkthdr.len); 1390 1391 if (scifp->if_flags & IFF_MONITOR) { 1392 m_freem(m); 1393 m = NULL; 1394 } 1395 } 1396 1397 LAGG_RUNLOCK(sc, &tracker); 1398 return (m); 1399} 1400 1401static int 1402lagg_media_change(struct ifnet *ifp) 1403{ 1404 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1405 1406 if (sc->sc_ifflags & IFF_DEBUG) 1407 printf("%s\n", __func__); 1408 1409 /* Ignore */ 1410 return (0); 1411} 1412 1413static void 1414lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1415{ 1416 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1417 struct lagg_port *lp; 1418 struct rm_priotracker tracker; 1419 1420 imr->ifm_status = IFM_AVALID; 1421 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1422 1423 LAGG_RLOCK(sc, &tracker); 1424 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1425 if (LAGG_PORTACTIVE(lp)) 1426 imr->ifm_status |= IFM_ACTIVE; 1427 } 1428 LAGG_RUNLOCK(sc, &tracker); 1429} 1430 1431static void 1432lagg_linkstate(struct lagg_softc *sc) 1433{ 1434 struct lagg_port *lp; 1435 int new_link = LINK_STATE_DOWN; 1436 uint64_t speed; 1437 1438 /* Our link is considered up if at least one of our ports is active */ 1439 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1440 if (lp->lp_link_state == LINK_STATE_UP) { 1441 new_link = LINK_STATE_UP; 1442 break; 1443 } 1444 } 1445 if_link_state_change(sc->sc_ifp, new_link); 1446 1447 /* Update if_baudrate to reflect the max possible speed */ 1448 switch (sc->sc_proto) { 1449 case LAGG_PROTO_FAILOVER: 1450 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1451 sc->sc_primary->lp_ifp->if_baudrate : 0; 1452 break; 1453 case LAGG_PROTO_ROUNDROBIN: 1454 case LAGG_PROTO_LOADBALANCE: 1455 case LAGG_PROTO_ETHERCHANNEL: 1456 speed = 0; 1457 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1458 speed += lp->lp_ifp->if_baudrate; 1459 sc->sc_ifp->if_baudrate = speed; 1460 break; 1461 case LAGG_PROTO_LACP: 1462 /* LACP updates if_baudrate itself */ 1463 break; 1464 } 1465} 1466 1467static void 1468lagg_port_state(struct ifnet *ifp, int state) 1469{ 1470 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1471 struct lagg_softc *sc = NULL; 1472 1473 if (lp != NULL) 1474 sc = lp->lp_softc; 1475 if (sc == NULL) 1476 return; 1477 1478 LAGG_WLOCK(sc); 1479 lagg_linkstate(sc); 1480 if (sc->sc_linkstate != NULL) 1481 (*sc->sc_linkstate)(lp); 1482 LAGG_WUNLOCK(sc); 1483} 1484 1485struct lagg_port * 1486lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1487{ 1488 struct lagg_port *lp_next, *rval = NULL; 1489 // int new_link = LINK_STATE_DOWN; 1490 1491 LAGG_RLOCK_ASSERT(sc); 1492 /* 1493 * Search a port which reports an active link state. 1494 */ 1495 1496 if (lp == NULL) 1497 goto search; 1498 if (LAGG_PORTACTIVE(lp)) { 1499 rval = lp; 1500 goto found; 1501 } 1502 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1503 LAGG_PORTACTIVE(lp_next)) { 1504 rval = lp_next; 1505 goto found; 1506 } 1507 1508search: 1509 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1510 if (LAGG_PORTACTIVE(lp_next)) { 1511 rval = lp_next; 1512 goto found; 1513 } 1514 } 1515 1516found: 1517 if (rval != NULL) { 1518 /* 1519 * The IEEE 802.1D standard assumes that a lagg with 1520 * multiple ports is always full duplex. This is valid 1521 * for load sharing laggs and if at least two links 1522 * are active. Unfortunately, checking the latter would 1523 * be too expensive at this point. 1524 XXX 1525 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1526 (sc->sc_count > 1)) 1527 new_link = LINK_STATE_FULL_DUPLEX; 1528 else 1529 new_link = rval->lp_link_state; 1530 */ 1531 } 1532 1533 return (rval); 1534} 1535 1536static const void * 1537lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf) 1538{ 1539 if (m->m_pkthdr.len < (off + len)) { 1540 return (NULL); 1541 } else if (m->m_len < (off + len)) { 1542 m_copydata(m, off, len, buf); 1543 return (buf); 1544 } 1545 return (mtod(m, char *) + off); 1546} 1547 1548static int 1549lagg_sysctl_active(SYSCTL_HANDLER_ARGS) 1550{ 1551 struct lagg_softc *sc = (struct lagg_softc *)arg1; 1552 struct lagg_port *lp; 1553 int error; 1554 1555 /* LACP tracks active links automatically, the others do not */ 1556 if (sc->sc_proto != LAGG_PROTO_LACP) { 1557 sc->sc_active = 0; 1558 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1559 sc->sc_active += LAGG_PORTACTIVE(lp); 1560 } 1561 1562 error = sysctl_handle_int(oidp, &sc->sc_active, 0, req); 1563 if ((error) || (req->newptr == NULL)) 1564 return (error); 1565 1566 return (0); 1567} 1568 1569uint32_t 1570lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key) 1571{ 1572 uint16_t etype; 1573 uint32_t p = key; 1574 int off; 1575 struct ether_header *eh; 1576 const struct ether_vlan_header *vlan; 1577#ifdef INET 1578 const struct ip *ip; 1579 const uint32_t *ports; 1580 int iphlen; 1581#endif 1582#ifdef INET6 1583 const struct ip6_hdr *ip6; 1584 uint32_t flow; 1585#endif 1586 union { 1587#ifdef INET 1588 struct ip ip; 1589#endif 1590#ifdef INET6 1591 struct ip6_hdr ip6; 1592#endif 1593 struct ether_vlan_header vlan; 1594 uint32_t port; 1595 } buf; 1596 1597 1598 off = sizeof(*eh); 1599 if (m->m_len < off) 1600 goto out; 1601 eh = mtod(m, struct ether_header *); 1602 etype = ntohs(eh->ether_type); 1603 if (sc->sc_flags & LAGG_F_HASHL2) { 1604 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p); 1605 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p); 1606 } 1607 1608 /* Special handling for encapsulating VLAN frames */ 1609 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) { 1610 p = hash32_buf(&m->m_pkthdr.ether_vtag, 1611 sizeof(m->m_pkthdr.ether_vtag), p); 1612 } else if (etype == ETHERTYPE_VLAN) { 1613 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf); 1614 if (vlan == NULL) 1615 goto out; 1616 1617 if (sc->sc_flags & LAGG_F_HASHL2) 1618 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p); 1619 etype = ntohs(vlan->evl_proto); 1620 off += sizeof(*vlan) - sizeof(*eh); 1621 } 1622 1623 switch (etype) { 1624#ifdef INET 1625 case ETHERTYPE_IP: 1626 ip = lagg_gethdr(m, off, sizeof(*ip), &buf); 1627 if (ip == NULL) 1628 goto out; 1629 1630 if (sc->sc_flags & LAGG_F_HASHL3) { 1631 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p); 1632 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p); 1633 } 1634 if (!(sc->sc_flags & LAGG_F_HASHL4)) 1635 break; 1636 switch (ip->ip_p) { 1637 case IPPROTO_TCP: 1638 case IPPROTO_UDP: 1639 case IPPROTO_SCTP: 1640 iphlen = ip->ip_hl << 2; 1641 if (iphlen < sizeof(*ip)) 1642 break; 1643 off += iphlen; 1644 ports = lagg_gethdr(m, off, sizeof(*ports), &buf); 1645 if (ports == NULL) 1646 break; 1647 p = hash32_buf(ports, sizeof(*ports), p); 1648 break; 1649 } 1650 break; 1651#endif 1652#ifdef INET6 1653 case ETHERTYPE_IPV6: 1654 if (!(sc->sc_flags & LAGG_F_HASHL3)) 1655 break; 1656 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf); 1657 if (ip6 == NULL) 1658 goto out; 1659 1660 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p); 1661 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p); 1662 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK; 1663 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */ 1664 break; 1665#endif 1666 } 1667out: 1668 return (p); 1669} 1670 1671int 1672lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1673{ 1674 1675 return (ifp->if_transmit)(ifp, m); 1676} 1677 1678/* 1679 * Simple round robin aggregation 1680 */ 1681 1682static int 1683lagg_rr_attach(struct lagg_softc *sc) 1684{ 1685 sc->sc_detach = lagg_rr_detach; 1686 sc->sc_start = lagg_rr_start; 1687 sc->sc_input = lagg_rr_input; 1688 sc->sc_port_create = NULL; 1689 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1690 sc->sc_seq = 0; 1691 1692 return (0); 1693} 1694 1695static int 1696lagg_rr_detach(struct lagg_softc *sc) 1697{ 1698 return (0); 1699} 1700 1701static int 1702lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 1703{ 1704 struct lagg_port *lp; 1705 uint32_t p; 1706 1707 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1708 p %= sc->sc_count; 1709 lp = SLIST_FIRST(&sc->sc_ports); 1710 while (p--) 1711 lp = SLIST_NEXT(lp, lp_entries); 1712 1713 /* 1714 * Check the port's link state. This will return the next active 1715 * port if the link is down or the port is NULL. 1716 */ 1717 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1718 m_freem(m); 1719 return (ENETDOWN); 1720 } 1721 1722 /* Send mbuf */ 1723 return (lagg_enqueue(lp->lp_ifp, m)); 1724} 1725 1726static struct mbuf * 1727lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1728{ 1729 struct ifnet *ifp = sc->sc_ifp; 1730 1731 /* Just pass in the packet to our lagg device */ 1732 m->m_pkthdr.rcvif = ifp; 1733 1734 return (m); 1735} 1736 1737/* 1738 * Active failover 1739 */ 1740 1741static int 1742lagg_fail_attach(struct lagg_softc *sc) 1743{ 1744 sc->sc_detach = lagg_fail_detach; 1745 sc->sc_start = lagg_fail_start; 1746 sc->sc_input = lagg_fail_input; 1747 sc->sc_port_create = NULL; 1748 sc->sc_port_destroy = NULL; 1749 1750 return (0); 1751} 1752 1753static int 1754lagg_fail_detach(struct lagg_softc *sc) 1755{ 1756 return (0); 1757} 1758 1759static int 1760lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 1761{ 1762 struct lagg_port *lp; 1763 1764 /* Use the master port if active or the next available port */ 1765 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 1766 m_freem(m); 1767 return (ENETDOWN); 1768 } 1769 1770 /* Send mbuf */ 1771 return (lagg_enqueue(lp->lp_ifp, m)); 1772} 1773 1774static struct mbuf * 1775lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1776{ 1777 struct ifnet *ifp = sc->sc_ifp; 1778 struct lagg_port *tmp_tp; 1779 1780 if (lp == sc->sc_primary || lagg_failover_rx_all) { 1781 m->m_pkthdr.rcvif = ifp; 1782 return (m); 1783 } 1784 1785 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1786 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1787 /* 1788 * If tmp_tp is null, we've recieved a packet when all 1789 * our links are down. Weird, but process it anyways. 1790 */ 1791 if ((tmp_tp == NULL || tmp_tp == lp)) { 1792 m->m_pkthdr.rcvif = ifp; 1793 return (m); 1794 } 1795 } 1796 1797 m_freem(m); 1798 return (NULL); 1799} 1800 1801/* 1802 * Loadbalancing 1803 */ 1804 1805static int 1806lagg_lb_attach(struct lagg_softc *sc) 1807{ 1808 struct lagg_port *lp; 1809 struct lagg_lb *lb; 1810 1811 if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb), 1812 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 1813 return (ENOMEM); 1814 1815 sc->sc_detach = lagg_lb_detach; 1816 sc->sc_start = lagg_lb_start; 1817 sc->sc_input = lagg_lb_input; 1818 sc->sc_port_create = lagg_lb_port_create; 1819 sc->sc_port_destroy = lagg_lb_port_destroy; 1820 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1821 1822 lb->lb_key = arc4random(); 1823 sc->sc_psc = (caddr_t)lb; 1824 1825 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1826 lagg_lb_port_create(lp); 1827 1828 return (0); 1829} 1830 1831static int 1832lagg_lb_detach(struct lagg_softc *sc) 1833{ 1834 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1835 if (lb != NULL) 1836 free(lb, M_DEVBUF); 1837 return (0); 1838} 1839 1840static int 1841lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 1842{ 1843 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1844 struct lagg_port *lp_next; 1845 int i = 0; 1846 1847 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 1848 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1849 if (lp_next == lp) 1850 continue; 1851 if (i >= LAGG_MAX_PORTS) 1852 return (EINVAL); 1853 if (sc->sc_ifflags & IFF_DEBUG) 1854 printf("%s: port %s at index %d\n", 1855 sc->sc_ifname, lp_next->lp_ifname, i); 1856 lb->lb_ports[i++] = lp_next; 1857 } 1858 1859 return (0); 1860} 1861 1862static int 1863lagg_lb_port_create(struct lagg_port *lp) 1864{ 1865 struct lagg_softc *sc = lp->lp_softc; 1866 return (lagg_lb_porttable(sc, NULL)); 1867} 1868 1869static void 1870lagg_lb_port_destroy(struct lagg_port *lp) 1871{ 1872 struct lagg_softc *sc = lp->lp_softc; 1873 lagg_lb_porttable(sc, lp); 1874} 1875 1876static int 1877lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 1878{ 1879 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1880 struct lagg_port *lp = NULL; 1881 uint32_t p = 0; 1882 1883 if (sc->use_flowid && (m->m_flags & M_FLOWID)) 1884 p = m->m_pkthdr.flowid >> sc->flowid_shift; 1885 else 1886 p = lagg_hashmbuf(sc, m, lb->lb_key); 1887 p %= sc->sc_count; 1888 lp = lb->lb_ports[p]; 1889 1890 /* 1891 * Check the port's link state. This will return the next active 1892 * port if the link is down or the port is NULL. 1893 */ 1894 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1895 m_freem(m); 1896 return (ENETDOWN); 1897 } 1898 1899 /* Send mbuf */ 1900 return (lagg_enqueue(lp->lp_ifp, m)); 1901} 1902 1903static struct mbuf * 1904lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1905{ 1906 struct ifnet *ifp = sc->sc_ifp; 1907 1908 /* Just pass in the packet to our lagg device */ 1909 m->m_pkthdr.rcvif = ifp; 1910 1911 return (m); 1912} 1913 1914/* 1915 * 802.3ad LACP 1916 */ 1917 1918static int 1919lagg_lacp_attach(struct lagg_softc *sc) 1920{ 1921 struct lagg_port *lp; 1922 int error; 1923 1924 sc->sc_detach = lagg_lacp_detach; 1925 sc->sc_port_create = lacp_port_create; 1926 sc->sc_port_destroy = lacp_port_destroy; 1927 sc->sc_linkstate = lacp_linkstate; 1928 sc->sc_start = lagg_lacp_start; 1929 sc->sc_input = lagg_lacp_input; 1930 sc->sc_init = lacp_init; 1931 sc->sc_stop = lacp_stop; 1932 sc->sc_lladdr = lagg_lacp_lladdr; 1933 sc->sc_req = lacp_req; 1934 sc->sc_portreq = lacp_portreq; 1935 1936 error = lacp_attach(sc); 1937 if (error) 1938 return (error); 1939 1940 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1941 lacp_port_create(lp); 1942 1943 return (error); 1944} 1945 1946static int 1947lagg_lacp_detach(struct lagg_softc *sc) 1948{ 1949 struct lagg_port *lp; 1950 int error; 1951 1952 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1953 lacp_port_destroy(lp); 1954 1955 /* unlocking is safe here */ 1956 LAGG_WUNLOCK(sc); 1957 error = lacp_detach(sc); 1958 LAGG_WLOCK(sc); 1959 1960 return (error); 1961} 1962 1963static void 1964lagg_lacp_lladdr(struct lagg_softc *sc) 1965{ 1966 struct lagg_port *lp; 1967 1968 /* purge all the lacp ports */ 1969 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1970 lacp_port_destroy(lp); 1971 1972 /* add them back in */ 1973 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1974 lacp_port_create(lp); 1975} 1976 1977static int 1978lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 1979{ 1980 struct lagg_port *lp; 1981 1982 lp = lacp_select_tx_port(sc, m); 1983 if (lp == NULL) { 1984 m_freem(m); 1985 return (ENETDOWN); 1986 } 1987 1988 /* Send mbuf */ 1989 return (lagg_enqueue(lp->lp_ifp, m)); 1990} 1991 1992static struct mbuf * 1993lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1994{ 1995 struct ifnet *ifp = sc->sc_ifp; 1996 struct ether_header *eh; 1997 u_short etype; 1998 1999 eh = mtod(m, struct ether_header *); 2000 etype = ntohs(eh->ether_type); 2001 2002 /* Tap off LACP control messages */ 2003 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 2004 m = lacp_input(lp, m); 2005 if (m == NULL) 2006 return (NULL); 2007 } 2008 2009 /* 2010 * If the port is not collecting or not in the active aggregator then 2011 * free and return. 2012 */ 2013 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 2014 m_freem(m); 2015 return (NULL); 2016 } 2017 2018 m->m_pkthdr.rcvif = ifp; 2019 return (m); 2020} 2021 2022static void 2023lagg_callout(void *arg) 2024{ 2025 struct lagg_softc *sc = (struct lagg_softc *)arg; 2026 struct ifnet *ifp = sc->sc_ifp; 2027 2028 ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets); 2029 ifp->if_opackets = counter_u64_fetch(sc->sc_opackets); 2030 ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes); 2031 ifp->if_obytes = counter_u64_fetch(sc->sc_obytes); 2032 2033 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 2034} 2035