if_lagg.c revision 234936
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: head/sys/net/if_lagg.c 234936 2012-05-03 01:41:12Z emaste $"); 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. */ 73static 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 82SLIST_HEAD(__trhead, lagg_softc) lagg_list; /* list of laggs */ 83static struct mtx lagg_list_mtx; 84eventhandler_tag lagg_detach_cookie = NULL; 85 86static int lagg_clone_create(struct if_clone *, int, caddr_t); 87static void lagg_clone_destroy(struct ifnet *); 88static void lagg_lladdr(struct lagg_softc *, uint8_t *); 89static void lagg_capabilities(struct lagg_softc *); 90static void lagg_port_lladdr(struct lagg_port *, uint8_t *); 91static void lagg_port_setlladdr(void *, int); 92static int lagg_port_create(struct lagg_softc *, struct ifnet *); 93static int lagg_port_destroy(struct lagg_port *, int); 94static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); 95static void lagg_linkstate(struct lagg_softc *); 96static void lagg_port_state(struct ifnet *, int); 97static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); 98static int lagg_port_output(struct ifnet *, struct mbuf *, 99 struct sockaddr *, struct route *); 100static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 101#ifdef LAGG_PORT_STACKING 102static int lagg_port_checkstacking(struct lagg_softc *); 103#endif 104static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 105static void lagg_init(void *); 106static void lagg_stop(struct lagg_softc *); 107static int lagg_ioctl(struct ifnet *, u_long, caddr_t); 108static int lagg_ether_setmulti(struct lagg_softc *); 109static int lagg_ether_cmdmulti(struct lagg_port *, int); 110static int lagg_setflag(struct lagg_port *, int, int, 111 int (*func)(struct ifnet *, int)); 112static int lagg_setflags(struct lagg_port *, int status); 113static void lagg_start(struct ifnet *); 114static int lagg_media_change(struct ifnet *); 115static void lagg_media_status(struct ifnet *, struct ifmediareq *); 116static struct lagg_port *lagg_link_active(struct lagg_softc *, 117 struct lagg_port *); 118static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *); 119 120IFC_SIMPLE_DECLARE(lagg, 0); 121 122/* Simple round robin */ 123static int lagg_rr_attach(struct lagg_softc *); 124static int lagg_rr_detach(struct lagg_softc *); 125static int lagg_rr_start(struct lagg_softc *, struct mbuf *); 126static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 127 struct mbuf *); 128 129/* Active failover */ 130static int lagg_fail_attach(struct lagg_softc *); 131static int lagg_fail_detach(struct lagg_softc *); 132static int lagg_fail_start(struct lagg_softc *, struct mbuf *); 133static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 134 struct mbuf *); 135 136/* Loadbalancing */ 137static int lagg_lb_attach(struct lagg_softc *); 138static int lagg_lb_detach(struct lagg_softc *); 139static int lagg_lb_port_create(struct lagg_port *); 140static void lagg_lb_port_destroy(struct lagg_port *); 141static int lagg_lb_start(struct lagg_softc *, struct mbuf *); 142static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 143 struct mbuf *); 144static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 145 146/* 802.3ad LACP */ 147static int lagg_lacp_attach(struct lagg_softc *); 148static int lagg_lacp_detach(struct lagg_softc *); 149static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); 150static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 151 struct mbuf *); 152static void lagg_lacp_lladdr(struct lagg_softc *); 153 154/* lagg protocol table */ 155static 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 167SYSCTL_DECL(_net_link); 168static SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 169 "Link Aggregation"); 170 171static int lagg_failover_rx_all = 0; /* Allow input on any failover links */ 172SYSCTL_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"); 175static int def_use_flowid = 1; /* Default value for using M_FLOWID */ 176TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid); 177SYSCTL_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 181static int 182lagg_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 210static moduledata_t lagg_mod = { 211 "if_lagg", 212 lagg_modevent, 213 0 214}; 215 216DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 217MODULE_VERSION(if_lagg, 1); 218 219#if __FreeBSD_version >= 800000 220/* 221 * This routine is run via an vlan 222 * config EVENT 223 */ 224static void 225lagg_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 */ 245static void 246lagg_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 263static int 264lagg_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 345static void 346lagg_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 384static void 385lagg_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 399static void 400lagg_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 432static void 433lagg_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 */ 473static void 474lagg_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 505static int 506lagg_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 return (EBUSY); 521 522 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 523 if (ifp->if_type != IFT_ETHER) 524 return (EPROTONOSUPPORT); 525 526 /* Allow the first Ethernet member to define the MTU */ 527 if (SLIST_EMPTY(&sc->sc_ports)) 528 sc->sc_ifp->if_mtu = ifp->if_mtu; 529 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 530 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 531 ifp->if_xname); 532 return (EINVAL); 533 } 534 535 if ((lp = malloc(sizeof(struct lagg_port), 536 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 537 return (ENOMEM); 538 539 /* Check if port is a stacked lagg */ 540 mtx_lock(&lagg_list_mtx); 541 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) { 542 if (ifp == sc_ptr->sc_ifp) { 543 mtx_unlock(&lagg_list_mtx); 544 free(lp, M_DEVBUF); 545 return (EINVAL); 546 /* XXX disable stacking for the moment, its untested */ 547#ifdef LAGG_PORT_STACKING 548 lp->lp_flags |= LAGG_PORT_STACK; 549 if (lagg_port_checkstacking(sc_ptr) >= 550 LAGG_MAX_STACKING) { 551 mtx_unlock(&lagg_list_mtx); 552 free(lp, M_DEVBUF); 553 return (E2BIG); 554 } 555#endif 556 } 557 } 558 mtx_unlock(&lagg_list_mtx); 559 560 /* Change the interface type */ 561 lp->lp_iftype = ifp->if_type; 562 ifp->if_type = IFT_IEEE8023ADLAG; 563 ifp->if_lagg = lp; 564 lp->lp_ioctl = ifp->if_ioctl; 565 ifp->if_ioctl = lagg_port_ioctl; 566 lp->lp_output = ifp->if_output; 567 ifp->if_output = lagg_port_output; 568 569 lp->lp_ifp = ifp; 570 lp->lp_softc = sc; 571 572 /* Save port link layer address */ 573 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 574 575 if (SLIST_EMPTY(&sc->sc_ports)) { 576 sc->sc_primary = lp; 577 lagg_lladdr(sc, IF_LLADDR(ifp)); 578 } else { 579 /* Update link layer address for this port */ 580 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp)); 581 } 582 583 /* Insert into the list of ports */ 584 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 585 sc->sc_count++; 586 587 /* Update lagg capabilities */ 588 lagg_capabilities(sc); 589 lagg_linkstate(sc); 590 591 /* Add multicast addresses and interface flags to this port */ 592 lagg_ether_cmdmulti(lp, 1); 593 lagg_setflags(lp, 1); 594 595 if (sc->sc_port_create != NULL) 596 error = (*sc->sc_port_create)(lp); 597 if (error) { 598 /* remove the port again, without calling sc_port_destroy */ 599 lagg_port_destroy(lp, 0); 600 return (error); 601 } 602 603 return (error); 604} 605 606#ifdef LAGG_PORT_STACKING 607static int 608lagg_port_checkstacking(struct lagg_softc *sc) 609{ 610 struct lagg_softc *sc_ptr; 611 struct lagg_port *lp; 612 int m = 0; 613 614 LAGG_WLOCK_ASSERT(sc); 615 616 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 617 if (lp->lp_flags & LAGG_PORT_STACK) { 618 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 619 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 620 } 621 } 622 623 return (m + 1); 624} 625#endif 626 627static int 628lagg_port_destroy(struct lagg_port *lp, int runpd) 629{ 630 struct lagg_softc *sc = lp->lp_softc; 631 struct lagg_port *lp_ptr; 632 struct lagg_llq *llq; 633 struct ifnet *ifp = lp->lp_ifp; 634 635 LAGG_WLOCK_ASSERT(sc); 636 637 if (runpd && sc->sc_port_destroy != NULL) 638 (*sc->sc_port_destroy)(lp); 639 640 /* 641 * Remove multicast addresses and interface flags from this port and 642 * reset the MAC address, skip if the interface is being detached. 643 */ 644 if (!lp->lp_detaching) { 645 lagg_ether_cmdmulti(lp, 0); 646 lagg_setflags(lp, 0); 647 lagg_port_lladdr(lp, lp->lp_lladdr); 648 } 649 650 /* Restore interface */ 651 ifp->if_type = lp->lp_iftype; 652 ifp->if_ioctl = lp->lp_ioctl; 653 ifp->if_output = lp->lp_output; 654 ifp->if_lagg = NULL; 655 656 /* Finally, remove the port from the lagg */ 657 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 658 sc->sc_count--; 659 660 /* Update the primary interface */ 661 if (lp == sc->sc_primary) { 662 uint8_t lladdr[ETHER_ADDR_LEN]; 663 664 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) { 665 bzero(&lladdr, ETHER_ADDR_LEN); 666 } else { 667 bcopy(lp_ptr->lp_lladdr, 668 lladdr, ETHER_ADDR_LEN); 669 } 670 lagg_lladdr(sc, lladdr); 671 sc->sc_primary = lp_ptr; 672 673 /* Update link layer address for each port */ 674 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 675 lagg_port_lladdr(lp_ptr, lladdr); 676 } 677 678 /* Remove any pending lladdr changes from the queue */ 679 if (lp->lp_detaching) { 680 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 681 if (llq->llq_ifp == ifp) { 682 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq, 683 llq_entries); 684 free(llq, M_DEVBUF); 685 break; /* Only appears once */ 686 } 687 } 688 } 689 690 if (lp->lp_ifflags) 691 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 692 693 free(lp, M_DEVBUF); 694 695 /* Update lagg capabilities */ 696 lagg_capabilities(sc); 697 lagg_linkstate(sc); 698 699 return (0); 700} 701 702static int 703lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 704{ 705 struct lagg_reqport *rp = (struct lagg_reqport *)data; 706 struct lagg_softc *sc; 707 struct lagg_port *lp = NULL; 708 int error = 0; 709 710 /* Should be checked by the caller */ 711 if (ifp->if_type != IFT_IEEE8023ADLAG || 712 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 713 goto fallback; 714 715 switch (cmd) { 716 case SIOCGLAGGPORT: 717 if (rp->rp_portname[0] == '\0' || 718 ifunit(rp->rp_portname) != ifp) { 719 error = EINVAL; 720 break; 721 } 722 723 LAGG_RLOCK(sc); 724 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 725 error = ENOENT; 726 LAGG_RUNLOCK(sc); 727 break; 728 } 729 730 lagg_port2req(lp, rp); 731 LAGG_RUNLOCK(sc); 732 break; 733 734 case SIOCSIFCAP: 735 if (lp->lp_ioctl == NULL) { 736 error = EINVAL; 737 break; 738 } 739 error = (*lp->lp_ioctl)(ifp, cmd, data); 740 if (error) 741 break; 742 743 /* Update lagg interface capabilities */ 744 LAGG_WLOCK(sc); 745 lagg_capabilities(sc); 746 LAGG_WUNLOCK(sc); 747 break; 748 749 case SIOCSIFMTU: 750 /* Do not allow the MTU to be changed once joined */ 751 error = EINVAL; 752 break; 753 754 default: 755 goto fallback; 756 } 757 758 return (error); 759 760fallback: 761 if (lp->lp_ioctl != NULL) 762 return ((*lp->lp_ioctl)(ifp, cmd, data)); 763 764 return (EINVAL); 765} 766 767/* 768 * For direct output to child ports. 769 */ 770static int 771lagg_port_output(struct ifnet *ifp, struct mbuf *m, 772 struct sockaddr *dst, struct route *ro) 773{ 774 struct lagg_port *lp = ifp->if_lagg; 775 776 switch (dst->sa_family) { 777 case pseudo_AF_HDRCMPLT: 778 case AF_UNSPEC: 779 return ((*lp->lp_output)(ifp, m, dst, ro)); 780 } 781 782 /* drop any other frames */ 783 m_freem(m); 784 return (EBUSY); 785} 786 787static void 788lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 789{ 790 struct lagg_port *lp; 791 struct lagg_softc *sc; 792 793 if ((lp = ifp->if_lagg) == NULL) 794 return; 795 796 sc = lp->lp_softc; 797 798 LAGG_WLOCK(sc); 799 lp->lp_detaching = 1; 800 lagg_port_destroy(lp, 1); 801 LAGG_WUNLOCK(sc); 802} 803 804static void 805lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 806{ 807 struct lagg_softc *sc = lp->lp_softc; 808 809 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 810 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 811 rp->rp_prio = lp->lp_prio; 812 rp->rp_flags = lp->lp_flags; 813 if (sc->sc_portreq != NULL) 814 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc); 815 816 /* Add protocol specific flags */ 817 switch (sc->sc_proto) { 818 case LAGG_PROTO_FAILOVER: 819 if (lp == sc->sc_primary) 820 rp->rp_flags |= LAGG_PORT_MASTER; 821 if (lp == lagg_link_active(sc, sc->sc_primary)) 822 rp->rp_flags |= LAGG_PORT_ACTIVE; 823 break; 824 825 case LAGG_PROTO_ROUNDROBIN: 826 case LAGG_PROTO_LOADBALANCE: 827 case LAGG_PROTO_ETHERCHANNEL: 828 if (LAGG_PORTACTIVE(lp)) 829 rp->rp_flags |= LAGG_PORT_ACTIVE; 830 break; 831 832 case LAGG_PROTO_LACP: 833 /* LACP has a different definition of active */ 834 if (lacp_isactive(lp)) 835 rp->rp_flags |= LAGG_PORT_ACTIVE; 836 if (lacp_iscollecting(lp)) 837 rp->rp_flags |= LAGG_PORT_COLLECTING; 838 if (lacp_isdistributing(lp)) 839 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 840 break; 841 } 842 843} 844 845static void 846lagg_init(void *xsc) 847{ 848 struct lagg_softc *sc = (struct lagg_softc *)xsc; 849 struct lagg_port *lp; 850 struct ifnet *ifp = sc->sc_ifp; 851 852 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 853 return; 854 855 LAGG_WLOCK(sc); 856 857 ifp->if_drv_flags |= IFF_DRV_RUNNING; 858 /* Update the port lladdrs */ 859 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 860 lagg_port_lladdr(lp, IF_LLADDR(ifp)); 861 862 if (sc->sc_init != NULL) 863 (*sc->sc_init)(sc); 864 865 LAGG_WUNLOCK(sc); 866} 867 868static void 869lagg_stop(struct lagg_softc *sc) 870{ 871 struct ifnet *ifp = sc->sc_ifp; 872 873 LAGG_WLOCK_ASSERT(sc); 874 875 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 876 return; 877 878 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 879 880 if (sc->sc_stop != NULL) 881 (*sc->sc_stop)(sc); 882} 883 884static int 885lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 886{ 887 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 888 struct lagg_reqall *ra = (struct lagg_reqall *)data; 889 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 890 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 891 struct ifreq *ifr = (struct ifreq *)data; 892 struct lagg_port *lp; 893 struct ifnet *tpif; 894 struct thread *td = curthread; 895 char *buf, *outbuf; 896 int count, buflen, len, error = 0; 897 898 bzero(&rpbuf, sizeof(rpbuf)); 899 900 switch (cmd) { 901 case SIOCGLAGG: 902 LAGG_RLOCK(sc); 903 count = 0; 904 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 905 count++; 906 buflen = count * sizeof(struct lagg_reqport); 907 LAGG_RUNLOCK(sc); 908 909 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 910 911 LAGG_RLOCK(sc); 912 ra->ra_proto = sc->sc_proto; 913 if (sc->sc_req != NULL) 914 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc); 915 916 count = 0; 917 buf = outbuf; 918 len = min(ra->ra_size, buflen); 919 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 920 if (len < sizeof(rpbuf)) 921 break; 922 923 lagg_port2req(lp, &rpbuf); 924 memcpy(buf, &rpbuf, sizeof(rpbuf)); 925 count++; 926 buf += sizeof(rpbuf); 927 len -= sizeof(rpbuf); 928 } 929 LAGG_RUNLOCK(sc); 930 ra->ra_ports = count; 931 ra->ra_size = count * sizeof(rpbuf); 932 error = copyout(outbuf, ra->ra_port, ra->ra_size); 933 free(outbuf, M_TEMP); 934 break; 935 case SIOCSLAGG: 936 error = priv_check(td, PRIV_NET_LAGG); 937 if (error) 938 break; 939 if (ra->ra_proto >= LAGG_PROTO_MAX) { 940 error = EPROTONOSUPPORT; 941 break; 942 } 943 LAGG_WLOCK(sc); 944 if (sc->sc_proto != LAGG_PROTO_NONE) { 945 /* Reset protocol first in case detach unlocks */ 946 sc->sc_proto = LAGG_PROTO_NONE; 947 error = sc->sc_detach(sc); 948 sc->sc_detach = NULL; 949 sc->sc_start = NULL; 950 sc->sc_input = NULL; 951 sc->sc_port_create = NULL; 952 sc->sc_port_destroy = NULL; 953 sc->sc_linkstate = NULL; 954 sc->sc_init = NULL; 955 sc->sc_stop = NULL; 956 sc->sc_lladdr = NULL; 957 sc->sc_req = NULL; 958 sc->sc_portreq = NULL; 959 } else if (sc->sc_input != NULL) { 960 /* Still detaching */ 961 error = EBUSY; 962 } 963 if (error != 0) { 964 LAGG_WUNLOCK(sc); 965 break; 966 } 967 for (int i = 0; i < (sizeof(lagg_protos) / 968 sizeof(lagg_protos[0])); i++) { 969 if (lagg_protos[i].ti_proto == ra->ra_proto) { 970 if (sc->sc_ifflags & IFF_DEBUG) 971 printf("%s: using proto %u\n", 972 sc->sc_ifname, 973 lagg_protos[i].ti_proto); 974 sc->sc_proto = lagg_protos[i].ti_proto; 975 if (sc->sc_proto != LAGG_PROTO_NONE) 976 error = lagg_protos[i].ti_attach(sc); 977 LAGG_WUNLOCK(sc); 978 return (error); 979 } 980 } 981 LAGG_WUNLOCK(sc); 982 error = EPROTONOSUPPORT; 983 break; 984 case SIOCGLAGGFLAGS: 985 rf->rf_flags = sc->sc_flags; 986 break; 987 case SIOCSLAGGHASH: 988 error = priv_check(td, PRIV_NET_LAGG); 989 if (error) 990 break; 991 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 992 error = EINVAL; 993 break; 994 } 995 LAGG_WLOCK(sc); 996 sc->sc_flags &= ~LAGG_F_HASHMASK; 997 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK; 998 LAGG_WUNLOCK(sc); 999 break; 1000 case SIOCGLAGGPORT: 1001 if (rp->rp_portname[0] == '\0' || 1002 (tpif = ifunit(rp->rp_portname)) == NULL) { 1003 error = EINVAL; 1004 break; 1005 } 1006 1007 LAGG_RLOCK(sc); 1008 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1009 lp->lp_softc != sc) { 1010 error = ENOENT; 1011 LAGG_RUNLOCK(sc); 1012 break; 1013 } 1014 1015 lagg_port2req(lp, rp); 1016 LAGG_RUNLOCK(sc); 1017 break; 1018 case SIOCSLAGGPORT: 1019 error = priv_check(td, PRIV_NET_LAGG); 1020 if (error) 1021 break; 1022 if (rp->rp_portname[0] == '\0' || 1023 (tpif = ifunit(rp->rp_portname)) == NULL) { 1024 error = EINVAL; 1025 break; 1026 } 1027 LAGG_WLOCK(sc); 1028 error = lagg_port_create(sc, tpif); 1029 LAGG_WUNLOCK(sc); 1030 break; 1031 case SIOCSLAGGDELPORT: 1032 error = priv_check(td, PRIV_NET_LAGG); 1033 if (error) 1034 break; 1035 if (rp->rp_portname[0] == '\0' || 1036 (tpif = ifunit(rp->rp_portname)) == NULL) { 1037 error = EINVAL; 1038 break; 1039 } 1040 1041 LAGG_WLOCK(sc); 1042 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1043 lp->lp_softc != sc) { 1044 error = ENOENT; 1045 LAGG_WUNLOCK(sc); 1046 break; 1047 } 1048 1049 error = lagg_port_destroy(lp, 1); 1050 LAGG_WUNLOCK(sc); 1051 break; 1052 case SIOCSIFFLAGS: 1053 /* Set flags on ports too */ 1054 LAGG_WLOCK(sc); 1055 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1056 lagg_setflags(lp, 1); 1057 } 1058 LAGG_WUNLOCK(sc); 1059 1060 if (!(ifp->if_flags & IFF_UP) && 1061 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1062 /* 1063 * If interface is marked down and it is running, 1064 * then stop and disable it. 1065 */ 1066 LAGG_WLOCK(sc); 1067 lagg_stop(sc); 1068 LAGG_WUNLOCK(sc); 1069 } else if ((ifp->if_flags & IFF_UP) && 1070 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1071 /* 1072 * If interface is marked up and it is stopped, then 1073 * start it. 1074 */ 1075 (*ifp->if_init)(sc); 1076 } 1077 break; 1078 case SIOCADDMULTI: 1079 case SIOCDELMULTI: 1080 LAGG_WLOCK(sc); 1081 error = lagg_ether_setmulti(sc); 1082 LAGG_WUNLOCK(sc); 1083 break; 1084 case SIOCSIFMEDIA: 1085 case SIOCGIFMEDIA: 1086 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1087 break; 1088 1089 case SIOCSIFCAP: 1090 case SIOCSIFMTU: 1091 /* Do not allow the MTU or caps to be directly changed */ 1092 error = EINVAL; 1093 break; 1094 1095 default: 1096 error = ether_ioctl(ifp, cmd, data); 1097 break; 1098 } 1099 return (error); 1100} 1101 1102static int 1103lagg_ether_setmulti(struct lagg_softc *sc) 1104{ 1105 struct lagg_port *lp; 1106 1107 LAGG_WLOCK_ASSERT(sc); 1108 1109 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1110 /* First, remove any existing filter entries. */ 1111 lagg_ether_cmdmulti(lp, 0); 1112 /* copy all addresses from the lagg interface to the port */ 1113 lagg_ether_cmdmulti(lp, 1); 1114 } 1115 return (0); 1116} 1117 1118static int 1119lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1120{ 1121 struct lagg_softc *sc = lp->lp_softc; 1122 struct ifnet *ifp = lp->lp_ifp; 1123 struct ifnet *scifp = sc->sc_ifp; 1124 struct lagg_mc *mc; 1125 struct ifmultiaddr *ifma, *rifma = NULL; 1126 struct sockaddr_dl sdl; 1127 int error; 1128 1129 LAGG_WLOCK_ASSERT(sc); 1130 1131 bzero((char *)&sdl, sizeof(sdl)); 1132 sdl.sdl_len = sizeof(sdl); 1133 sdl.sdl_family = AF_LINK; 1134 sdl.sdl_type = IFT_ETHER; 1135 sdl.sdl_alen = ETHER_ADDR_LEN; 1136 sdl.sdl_index = ifp->if_index; 1137 1138 if (set) { 1139 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1140 if (ifma->ifma_addr->sa_family != AF_LINK) 1141 continue; 1142 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 1143 LLADDR(&sdl), ETHER_ADDR_LEN); 1144 1145 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 1146 if (error) 1147 return (error); 1148 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1149 if (mc == NULL) 1150 return (ENOMEM); 1151 mc->mc_ifma = rifma; 1152 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1153 } 1154 } else { 1155 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1156 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1157 if_delmulti_ifma(mc->mc_ifma); 1158 free(mc, M_DEVBUF); 1159 } 1160 } 1161 return (0); 1162} 1163 1164/* Handle a ref counted flag that should be set on the lagg port as well */ 1165static int 1166lagg_setflag(struct lagg_port *lp, int flag, int status, 1167 int (*func)(struct ifnet *, int)) 1168{ 1169 struct lagg_softc *sc = lp->lp_softc; 1170 struct ifnet *scifp = sc->sc_ifp; 1171 struct ifnet *ifp = lp->lp_ifp; 1172 int error; 1173 1174 LAGG_WLOCK_ASSERT(sc); 1175 1176 status = status ? (scifp->if_flags & flag) : 0; 1177 /* Now "status" contains the flag value or 0 */ 1178 1179 /* 1180 * See if recorded ports status is different from what 1181 * we want it to be. If it is, flip it. We record ports 1182 * status in lp_ifflags so that we won't clear ports flag 1183 * we haven't set. In fact, we don't clear or set ports 1184 * flags directly, but get or release references to them. 1185 * That's why we can be sure that recorded flags still are 1186 * in accord with actual ports flags. 1187 */ 1188 if (status != (lp->lp_ifflags & flag)) { 1189 error = (*func)(ifp, status); 1190 if (error) 1191 return (error); 1192 lp->lp_ifflags &= ~flag; 1193 lp->lp_ifflags |= status; 1194 } 1195 return (0); 1196} 1197 1198/* 1199 * Handle IFF_* flags that require certain changes on the lagg port 1200 * if "status" is true, update ports flags respective to the lagg 1201 * if "status" is false, forcedly clear the flags set on port. 1202 */ 1203static int 1204lagg_setflags(struct lagg_port *lp, int status) 1205{ 1206 int error, i; 1207 1208 for (i = 0; lagg_pflags[i].flag; i++) { 1209 error = lagg_setflag(lp, lagg_pflags[i].flag, 1210 status, lagg_pflags[i].func); 1211 if (error) 1212 return (error); 1213 } 1214 return (0); 1215} 1216 1217static void 1218lagg_start(struct ifnet *ifp) 1219{ 1220 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1221 struct mbuf *m; 1222 int error = 0; 1223 1224 LAGG_RLOCK(sc); 1225 /* We need a Tx algorithm and at least one port */ 1226 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1227 IF_DRAIN(&ifp->if_snd); 1228 LAGG_RUNLOCK(sc); 1229 return; 1230 } 1231 1232 for (;; error = 0) { 1233 IFQ_DEQUEUE(&ifp->if_snd, m); 1234 if (m == NULL) 1235 break; 1236 1237 ETHER_BPF_MTAP(ifp, m); 1238 1239 error = (*sc->sc_start)(sc, m); 1240 if (error == 0) 1241 ifp->if_opackets++; 1242 else 1243 ifp->if_oerrors++; 1244 } 1245 LAGG_RUNLOCK(sc); 1246} 1247 1248static struct mbuf * 1249lagg_input(struct ifnet *ifp, struct mbuf *m) 1250{ 1251 struct lagg_port *lp = ifp->if_lagg; 1252 struct lagg_softc *sc = lp->lp_softc; 1253 struct ifnet *scifp = sc->sc_ifp; 1254 1255 LAGG_RLOCK(sc); 1256 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1257 (lp->lp_flags & LAGG_PORT_DISABLED) || 1258 sc->sc_proto == LAGG_PROTO_NONE) { 1259 LAGG_RUNLOCK(sc); 1260 m_freem(m); 1261 return (NULL); 1262 } 1263 1264 ETHER_BPF_MTAP(scifp, m); 1265 1266 m = (*sc->sc_input)(sc, lp, m); 1267 1268 if (m != NULL) { 1269 scifp->if_ipackets++; 1270 scifp->if_ibytes += m->m_pkthdr.len; 1271 1272 if (scifp->if_flags & IFF_MONITOR) { 1273 m_freem(m); 1274 m = NULL; 1275 } 1276 } 1277 1278 LAGG_RUNLOCK(sc); 1279 return (m); 1280} 1281 1282static int 1283lagg_media_change(struct ifnet *ifp) 1284{ 1285 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1286 1287 if (sc->sc_ifflags & IFF_DEBUG) 1288 printf("%s\n", __func__); 1289 1290 /* Ignore */ 1291 return (0); 1292} 1293 1294static void 1295lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1296{ 1297 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1298 struct lagg_port *lp; 1299 1300 imr->ifm_status = IFM_AVALID; 1301 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1302 1303 LAGG_RLOCK(sc); 1304 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1305 if (LAGG_PORTACTIVE(lp)) 1306 imr->ifm_status |= IFM_ACTIVE; 1307 } 1308 LAGG_RUNLOCK(sc); 1309} 1310 1311static void 1312lagg_linkstate(struct lagg_softc *sc) 1313{ 1314 struct lagg_port *lp; 1315 int new_link = LINK_STATE_DOWN; 1316 uint64_t speed; 1317 1318 /* Our link is considered up if at least one of our ports is active */ 1319 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1320 if (lp->lp_link_state == LINK_STATE_UP) { 1321 new_link = LINK_STATE_UP; 1322 break; 1323 } 1324 } 1325 if_link_state_change(sc->sc_ifp, new_link); 1326 1327 /* Update if_baudrate to reflect the max possible speed */ 1328 switch (sc->sc_proto) { 1329 case LAGG_PROTO_FAILOVER: 1330 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1331 sc->sc_primary->lp_ifp->if_baudrate : 0; 1332 break; 1333 case LAGG_PROTO_ROUNDROBIN: 1334 case LAGG_PROTO_LOADBALANCE: 1335 case LAGG_PROTO_ETHERCHANNEL: 1336 speed = 0; 1337 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1338 speed += lp->lp_ifp->if_baudrate; 1339 sc->sc_ifp->if_baudrate = speed; 1340 break; 1341 case LAGG_PROTO_LACP: 1342 /* LACP updates if_baudrate itself */ 1343 break; 1344 } 1345} 1346 1347static void 1348lagg_port_state(struct ifnet *ifp, int state) 1349{ 1350 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1351 struct lagg_softc *sc = NULL; 1352 1353 if (lp != NULL) 1354 sc = lp->lp_softc; 1355 if (sc == NULL) 1356 return; 1357 1358 LAGG_WLOCK(sc); 1359 lagg_linkstate(sc); 1360 if (sc->sc_linkstate != NULL) 1361 (*sc->sc_linkstate)(lp); 1362 LAGG_WUNLOCK(sc); 1363} 1364 1365struct lagg_port * 1366lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1367{ 1368 struct lagg_port *lp_next, *rval = NULL; 1369 // int new_link = LINK_STATE_DOWN; 1370 1371 LAGG_RLOCK_ASSERT(sc); 1372 /* 1373 * Search a port which reports an active link state. 1374 */ 1375 1376 if (lp == NULL) 1377 goto search; 1378 if (LAGG_PORTACTIVE(lp)) { 1379 rval = lp; 1380 goto found; 1381 } 1382 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1383 LAGG_PORTACTIVE(lp_next)) { 1384 rval = lp_next; 1385 goto found; 1386 } 1387 1388search: 1389 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1390 if (LAGG_PORTACTIVE(lp_next)) { 1391 rval = lp_next; 1392 goto found; 1393 } 1394 } 1395 1396found: 1397 if (rval != NULL) { 1398 /* 1399 * The IEEE 802.1D standard assumes that a lagg with 1400 * multiple ports is always full duplex. This is valid 1401 * for load sharing laggs and if at least two links 1402 * are active. Unfortunately, checking the latter would 1403 * be too expensive at this point. 1404 XXX 1405 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1406 (sc->sc_count > 1)) 1407 new_link = LINK_STATE_FULL_DUPLEX; 1408 else 1409 new_link = rval->lp_link_state; 1410 */ 1411 } 1412 1413 return (rval); 1414} 1415 1416static const void * 1417lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf) 1418{ 1419 if (m->m_pkthdr.len < (off + len)) { 1420 return (NULL); 1421 } else if (m->m_len < (off + len)) { 1422 m_copydata(m, off, len, buf); 1423 return (buf); 1424 } 1425 return (mtod(m, char *) + off); 1426} 1427 1428uint32_t 1429lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key) 1430{ 1431 uint16_t etype; 1432 uint32_t p = key; 1433 int off; 1434 struct ether_header *eh; 1435 const struct ether_vlan_header *vlan; 1436#ifdef INET 1437 const struct ip *ip; 1438 const uint32_t *ports; 1439 int iphlen; 1440#endif 1441#ifdef INET6 1442 const struct ip6_hdr *ip6; 1443 uint32_t flow; 1444#endif 1445 union { 1446#ifdef INET 1447 struct ip ip; 1448#endif 1449#ifdef INET6 1450 struct ip6_hdr ip6; 1451#endif 1452 struct ether_vlan_header vlan; 1453 uint32_t port; 1454 } buf; 1455 1456 1457 off = sizeof(*eh); 1458 if (m->m_len < off) 1459 goto out; 1460 eh = mtod(m, struct ether_header *); 1461 etype = ntohs(eh->ether_type); 1462 if (sc->sc_flags & LAGG_F_HASHL2) { 1463 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p); 1464 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p); 1465 } 1466 1467 /* Special handling for encapsulating VLAN frames */ 1468 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) { 1469 p = hash32_buf(&m->m_pkthdr.ether_vtag, 1470 sizeof(m->m_pkthdr.ether_vtag), p); 1471 } else if (etype == ETHERTYPE_VLAN) { 1472 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf); 1473 if (vlan == NULL) 1474 goto out; 1475 1476 if (sc->sc_flags & LAGG_F_HASHL2) 1477 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p); 1478 etype = ntohs(vlan->evl_proto); 1479 off += sizeof(*vlan) - sizeof(*eh); 1480 } 1481 1482 switch (etype) { 1483#ifdef INET 1484 case ETHERTYPE_IP: 1485 ip = lagg_gethdr(m, off, sizeof(*ip), &buf); 1486 if (ip == NULL) 1487 goto out; 1488 1489 if (sc->sc_flags & LAGG_F_HASHL3) { 1490 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p); 1491 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p); 1492 } 1493 if (!(sc->sc_flags & LAGG_F_HASHL4)) 1494 break; 1495 switch (ip->ip_p) { 1496 case IPPROTO_TCP: 1497 case IPPROTO_UDP: 1498 case IPPROTO_SCTP: 1499 iphlen = ip->ip_hl << 2; 1500 if (iphlen < sizeof(*ip)) 1501 break; 1502 off += iphlen; 1503 ports = lagg_gethdr(m, off, sizeof(*ports), &buf); 1504 if (ports == NULL) 1505 break; 1506 p = hash32_buf(ports, sizeof(*ports), p); 1507 break; 1508 } 1509 break; 1510#endif 1511#ifdef INET6 1512 case ETHERTYPE_IPV6: 1513 if (!(sc->sc_flags & LAGG_F_HASHL3)) 1514 break; 1515 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf); 1516 if (ip6 == NULL) 1517 goto out; 1518 1519 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p); 1520 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p); 1521 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK; 1522 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */ 1523 break; 1524#endif 1525 } 1526out: 1527 return (p); 1528} 1529 1530int 1531lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1532{ 1533 1534 return (ifp->if_transmit)(ifp, m); 1535} 1536 1537/* 1538 * Simple round robin aggregation 1539 */ 1540 1541static int 1542lagg_rr_attach(struct lagg_softc *sc) 1543{ 1544 sc->sc_detach = lagg_rr_detach; 1545 sc->sc_start = lagg_rr_start; 1546 sc->sc_input = lagg_rr_input; 1547 sc->sc_port_create = NULL; 1548 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1549 sc->sc_seq = 0; 1550 1551 return (0); 1552} 1553 1554static int 1555lagg_rr_detach(struct lagg_softc *sc) 1556{ 1557 return (0); 1558} 1559 1560static int 1561lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 1562{ 1563 struct lagg_port *lp; 1564 uint32_t p; 1565 1566 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1567 p %= sc->sc_count; 1568 lp = SLIST_FIRST(&sc->sc_ports); 1569 while (p--) 1570 lp = SLIST_NEXT(lp, lp_entries); 1571 1572 /* 1573 * Check the port's link state. This will return the next active 1574 * port if the link is down or the port is NULL. 1575 */ 1576 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1577 m_freem(m); 1578 return (ENOENT); 1579 } 1580 1581 /* Send mbuf */ 1582 return (lagg_enqueue(lp->lp_ifp, m)); 1583} 1584 1585static struct mbuf * 1586lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1587{ 1588 struct ifnet *ifp = sc->sc_ifp; 1589 1590 /* Just pass in the packet to our lagg device */ 1591 m->m_pkthdr.rcvif = ifp; 1592 1593 return (m); 1594} 1595 1596/* 1597 * Active failover 1598 */ 1599 1600static int 1601lagg_fail_attach(struct lagg_softc *sc) 1602{ 1603 sc->sc_detach = lagg_fail_detach; 1604 sc->sc_start = lagg_fail_start; 1605 sc->sc_input = lagg_fail_input; 1606 sc->sc_port_create = NULL; 1607 sc->sc_port_destroy = NULL; 1608 1609 return (0); 1610} 1611 1612static int 1613lagg_fail_detach(struct lagg_softc *sc) 1614{ 1615 return (0); 1616} 1617 1618static int 1619lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 1620{ 1621 struct lagg_port *lp; 1622 1623 /* Use the master port if active or the next available port */ 1624 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 1625 m_freem(m); 1626 return (ENOENT); 1627 } 1628 1629 /* Send mbuf */ 1630 return (lagg_enqueue(lp->lp_ifp, m)); 1631} 1632 1633static struct mbuf * 1634lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1635{ 1636 struct ifnet *ifp = sc->sc_ifp; 1637 struct lagg_port *tmp_tp; 1638 1639 if (lp == sc->sc_primary || lagg_failover_rx_all) { 1640 m->m_pkthdr.rcvif = ifp; 1641 return (m); 1642 } 1643 1644 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1645 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1646 /* 1647 * If tmp_tp is null, we've recieved a packet when all 1648 * our links are down. Weird, but process it anyways. 1649 */ 1650 if ((tmp_tp == NULL || tmp_tp == lp)) { 1651 m->m_pkthdr.rcvif = ifp; 1652 return (m); 1653 } 1654 } 1655 1656 m_freem(m); 1657 return (NULL); 1658} 1659 1660/* 1661 * Loadbalancing 1662 */ 1663 1664static int 1665lagg_lb_attach(struct lagg_softc *sc) 1666{ 1667 struct lagg_port *lp; 1668 struct lagg_lb *lb; 1669 1670 if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb), 1671 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 1672 return (ENOMEM); 1673 1674 sc->sc_detach = lagg_lb_detach; 1675 sc->sc_start = lagg_lb_start; 1676 sc->sc_input = lagg_lb_input; 1677 sc->sc_port_create = lagg_lb_port_create; 1678 sc->sc_port_destroy = lagg_lb_port_destroy; 1679 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1680 1681 lb->lb_key = arc4random(); 1682 sc->sc_psc = (caddr_t)lb; 1683 1684 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1685 lagg_lb_port_create(lp); 1686 1687 return (0); 1688} 1689 1690static int 1691lagg_lb_detach(struct lagg_softc *sc) 1692{ 1693 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1694 if (lb != NULL) 1695 free(lb, M_DEVBUF); 1696 return (0); 1697} 1698 1699static int 1700lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 1701{ 1702 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1703 struct lagg_port *lp_next; 1704 int i = 0; 1705 1706 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 1707 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1708 if (lp_next == lp) 1709 continue; 1710 if (i >= LAGG_MAX_PORTS) 1711 return (EINVAL); 1712 if (sc->sc_ifflags & IFF_DEBUG) 1713 printf("%s: port %s at index %d\n", 1714 sc->sc_ifname, lp_next->lp_ifname, i); 1715 lb->lb_ports[i++] = lp_next; 1716 } 1717 1718 return (0); 1719} 1720 1721static int 1722lagg_lb_port_create(struct lagg_port *lp) 1723{ 1724 struct lagg_softc *sc = lp->lp_softc; 1725 return (lagg_lb_porttable(sc, NULL)); 1726} 1727 1728static void 1729lagg_lb_port_destroy(struct lagg_port *lp) 1730{ 1731 struct lagg_softc *sc = lp->lp_softc; 1732 lagg_lb_porttable(sc, lp); 1733} 1734 1735static int 1736lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 1737{ 1738 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1739 struct lagg_port *lp = NULL; 1740 uint32_t p = 0; 1741 1742 if (sc->use_flowid && (m->m_flags & M_FLOWID)) 1743 p = m->m_pkthdr.flowid; 1744 else 1745 p = lagg_hashmbuf(sc, m, lb->lb_key); 1746 p %= sc->sc_count; 1747 lp = lb->lb_ports[p]; 1748 1749 /* 1750 * Check the port's link state. This will return the next active 1751 * port if the link is down or the port is NULL. 1752 */ 1753 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1754 m_freem(m); 1755 return (ENOENT); 1756 } 1757 1758 /* Send mbuf */ 1759 return (lagg_enqueue(lp->lp_ifp, m)); 1760} 1761 1762static struct mbuf * 1763lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1764{ 1765 struct ifnet *ifp = sc->sc_ifp; 1766 1767 /* Just pass in the packet to our lagg device */ 1768 m->m_pkthdr.rcvif = ifp; 1769 1770 return (m); 1771} 1772 1773/* 1774 * 802.3ad LACP 1775 */ 1776 1777static int 1778lagg_lacp_attach(struct lagg_softc *sc) 1779{ 1780 struct lagg_port *lp; 1781 int error; 1782 1783 sc->sc_detach = lagg_lacp_detach; 1784 sc->sc_port_create = lacp_port_create; 1785 sc->sc_port_destroy = lacp_port_destroy; 1786 sc->sc_linkstate = lacp_linkstate; 1787 sc->sc_start = lagg_lacp_start; 1788 sc->sc_input = lagg_lacp_input; 1789 sc->sc_init = lacp_init; 1790 sc->sc_stop = lacp_stop; 1791 sc->sc_lladdr = lagg_lacp_lladdr; 1792 sc->sc_req = lacp_req; 1793 sc->sc_portreq = lacp_portreq; 1794 1795 error = lacp_attach(sc); 1796 if (error) 1797 return (error); 1798 1799 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1800 lacp_port_create(lp); 1801 1802 return (error); 1803} 1804 1805static int 1806lagg_lacp_detach(struct lagg_softc *sc) 1807{ 1808 struct lagg_port *lp; 1809 int error; 1810 1811 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1812 lacp_port_destroy(lp); 1813 1814 /* unlocking is safe here */ 1815 LAGG_WUNLOCK(sc); 1816 error = lacp_detach(sc); 1817 LAGG_WLOCK(sc); 1818 1819 return (error); 1820} 1821 1822static void 1823lagg_lacp_lladdr(struct lagg_softc *sc) 1824{ 1825 struct lagg_port *lp; 1826 1827 /* purge all the lacp ports */ 1828 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1829 lacp_port_destroy(lp); 1830 1831 /* add them back in */ 1832 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1833 lacp_port_create(lp); 1834} 1835 1836static int 1837lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 1838{ 1839 struct lagg_port *lp; 1840 1841 lp = lacp_select_tx_port(sc, m); 1842 if (lp == NULL) { 1843 m_freem(m); 1844 return (EBUSY); 1845 } 1846 1847 /* Send mbuf */ 1848 return (lagg_enqueue(lp->lp_ifp, m)); 1849} 1850 1851static struct mbuf * 1852lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1853{ 1854 struct ifnet *ifp = sc->sc_ifp; 1855 struct ether_header *eh; 1856 u_short etype; 1857 1858 eh = mtod(m, struct ether_header *); 1859 etype = ntohs(eh->ether_type); 1860 1861 /* Tap off LACP control messages */ 1862 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 1863 m = lacp_input(lp, m); 1864 if (m == NULL) 1865 return (NULL); 1866 } 1867 1868 /* 1869 * If the port is not collecting or not in the active aggregator then 1870 * free and return. 1871 */ 1872 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 1873 m_freem(m); 1874 return (NULL); 1875 } 1876 1877 m->m_pkthdr.rcvif = ifp; 1878 return (m); 1879} 1880