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 * Copyright (c) 2014 Marcelo Araujo <araujo@FreeBSD.org> 7 * 8 * Permission to use, copy, modify, and distribute this software for any 9 * purpose with or without fee is hereby granted, provided that the above 10 * copyright notice and this permission notice appear in all copies. 11 * 12 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 13 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 14 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 15 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 16 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 17 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 18 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 19 */ 20 21 #include <sys/cdefs.h> 22 __FBSDID("$FreeBSD$"); 23 24 #include "opt_inet.h" 25 #include "opt_inet6.h" 26 27 #include <sys/param.h> 28 #include <sys/kernel.h> 29 #include <sys/malloc.h> 30 #include <sys/mbuf.h> 31 #include <sys/queue.h> 32 #include <sys/socket.h> 33 #include <sys/sockio.h> 34 #include <sys/sysctl.h> 35 #include <sys/module.h> 36 #include <sys/priv.h> 37 #include <sys/systm.h> 38 #include <sys/proc.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_media.h> 50 #include <net/if_types.h> 51 #include <net/if_var.h> 52 #include <net/bpf.h> 53 #include <net/vnet.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. */ 75 static 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 84 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */ 85 #define V_lagg_list VNET(lagg_list) 86 static VNET_DEFINE(struct mtx, lagg_list_mtx); 87 #define V_lagg_list_mtx VNET(lagg_list_mtx) 88 #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \ 89 "if_lagg list", NULL, MTX_DEF) 90 #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx) 91 #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx) 92 #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx) 93 eventhandler_tag lagg_detach_cookie = NULL; 94 95 static int lagg_clone_create(struct if_clone *, int, caddr_t); 96 static void lagg_clone_destroy(struct ifnet *); 97 static VNET_DEFINE(struct if_clone *, lagg_cloner); 98 #define V_lagg_cloner VNET(lagg_cloner) 99 static const char laggname[] = "lagg"; 100 101 static void lagg_lladdr(struct lagg_softc *, uint8_t *); 102 static void lagg_capabilities(struct lagg_softc *); 103 static void lagg_port_lladdr(struct lagg_port *, uint8_t *); 104 static void lagg_port_setlladdr(void *, int); 105 static int lagg_port_create(struct lagg_softc *, struct ifnet *); 106 static int lagg_port_destroy(struct lagg_port *, int); 107 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); 108 static void lagg_linkstate(struct lagg_softc *); 109 static void lagg_port_state(struct ifnet *, int); 110 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); 111 static int lagg_port_output(struct ifnet *, struct mbuf *, 112 const struct sockaddr *, struct route *); 113 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 114 #ifdef LAGG_PORT_STACKING 115 static int lagg_port_checkstacking(struct lagg_softc *); 116 #endif 117 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 118 static void lagg_init(void *); 119 static void lagg_stop(struct lagg_softc *); 120 static int lagg_ioctl(struct ifnet *, u_long, caddr_t); 121 static int lagg_ether_setmulti(struct lagg_softc *); 122 static int lagg_ether_cmdmulti(struct lagg_port *, int); 123 static int lagg_setflag(struct lagg_port *, int, int, 124 int (*func)(struct ifnet *, int)); 125 static int lagg_setflags(struct lagg_port *, int status); 126 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt); 127 static int lagg_transmit(struct ifnet *, struct mbuf *); 128 static void lagg_qflush(struct ifnet *); 129 static int lagg_media_change(struct ifnet *); 130 static void lagg_media_status(struct ifnet *, struct ifmediareq *); 131 static struct lagg_port *lagg_link_active(struct lagg_softc *, 132 struct lagg_port *); 133 134 /* Simple round robin */ 135 static void lagg_rr_attach(struct lagg_softc *); 136 static int lagg_rr_start(struct lagg_softc *, struct mbuf *); 137 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 138 struct mbuf *); 139 140 /* Active failover */ 141 static int lagg_fail_start(struct lagg_softc *, struct mbuf *); 142 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 143 struct mbuf *); 144 145 /* Loadbalancing */ 146 static void lagg_lb_attach(struct lagg_softc *); 147 static void lagg_lb_detach(struct lagg_softc *); 148 static int lagg_lb_port_create(struct lagg_port *); 149 static void lagg_lb_port_destroy(struct lagg_port *); 150 static int lagg_lb_start(struct lagg_softc *, struct mbuf *); 151 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 152 struct mbuf *); 153 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 154 155 /* Broadcast */ 156 static int lagg_bcast_start(struct lagg_softc *, struct mbuf *); 157 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *, 158 struct mbuf *); 159 160 /* 802.3ad LACP */ 161 static void lagg_lacp_attach(struct lagg_softc *); 162 static void lagg_lacp_detach(struct lagg_softc *); 163 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); 164 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 165 struct mbuf *); 166 static void lagg_lacp_lladdr(struct lagg_softc *); 167 168 /* lagg protocol table */ 169 static const struct lagg_proto { 170 lagg_proto pr_num; 171 void (*pr_attach)(struct lagg_softc *); 172 void (*pr_detach)(struct lagg_softc *); 173 int (*pr_start)(struct lagg_softc *, struct mbuf *); 174 struct mbuf * (*pr_input)(struct lagg_softc *, struct lagg_port *, 175 struct mbuf *); 176 int (*pr_addport)(struct lagg_port *); 177 void (*pr_delport)(struct lagg_port *); 178 void (*pr_linkstate)(struct lagg_port *); 179 void (*pr_init)(struct lagg_softc *); 180 void (*pr_stop)(struct lagg_softc *); 181 void (*pr_lladdr)(struct lagg_softc *); 182 void (*pr_request)(struct lagg_softc *, void *); 183 void (*pr_portreq)(struct lagg_port *, void *); 184 } lagg_protos[] = { 185 { 186 .pr_num = LAGG_PROTO_NONE 187 }, 188 { 189 .pr_num = LAGG_PROTO_ROUNDROBIN, 190 .pr_attach = lagg_rr_attach, 191 .pr_start = lagg_rr_start, 192 .pr_input = lagg_rr_input, 193 }, 194 { 195 .pr_num = LAGG_PROTO_FAILOVER, 196 .pr_start = lagg_fail_start, 197 .pr_input = lagg_fail_input, 198 }, 199 { 200 .pr_num = LAGG_PROTO_LOADBALANCE, 201 .pr_attach = lagg_lb_attach, 202 .pr_detach = lagg_lb_detach, 203 .pr_start = lagg_lb_start, 204 .pr_input = lagg_lb_input, 205 .pr_addport = lagg_lb_port_create, 206 .pr_delport = lagg_lb_port_destroy, 207 }, 208 { 209 .pr_num = LAGG_PROTO_LACP, 210 .pr_attach = lagg_lacp_attach, 211 .pr_detach = lagg_lacp_detach, 212 .pr_start = lagg_lacp_start, 213 .pr_input = lagg_lacp_input, 214 .pr_addport = lacp_port_create, 215 .pr_delport = lacp_port_destroy, 216 .pr_linkstate = lacp_linkstate, 217 .pr_init = lacp_init, 218 .pr_stop = lacp_stop, 219 .pr_lladdr = lagg_lacp_lladdr, 220 .pr_request = lacp_req, 221 .pr_portreq = lacp_portreq, 222 }, 223 { 224 .pr_num = LAGG_PROTO_ETHERCHANNEL, 225 .pr_attach = lagg_lb_attach, 226 .pr_detach = lagg_lb_detach, 227 .pr_start = lagg_lb_start, 228 .pr_input = lagg_lb_input, 229 }, 230 { 231 .pr_num = LAGG_PROTO_BROADCAST, 232 .pr_start = lagg_bcast_start, 233 .pr_input = lagg_bcast_input, 234 }, 235 }; 236 237 SYSCTL_DECL(_net_link); 238 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 239 "Link Aggregation"); 240 241 /* Allow input on any failover links */ 242 static VNET_DEFINE(int, lagg_failover_rx_all); 243 #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all) 244 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET, 245 &VNET_NAME(lagg_failover_rx_all), 0, 246 "Accept input from any interface in a failover lagg"); 247 248 /* Default value for using flowid */ 249 static VNET_DEFINE(int, def_use_flowid) = 1; 250 #define V_def_use_flowid VNET(def_use_flowid) 251 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN, 252 &VNET_NAME(def_use_flowid), 0, 253 "Default setting for using flow id for load sharing"); 254 255 /* Default value for flowid shift */ 256 static VNET_DEFINE(int, def_flowid_shift) = 16; 257 #define V_def_flowid_shift VNET(def_flowid_shift) 258 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN, 259 &VNET_NAME(def_flowid_shift), 0, 260 "Default setting for flowid shift for load sharing"); 261 262 static void 263 vnet_lagg_init(const void *unused __unused) 264 { 265 266 LAGG_LIST_LOCK_INIT(); 267 SLIST_INIT(&V_lagg_list); 268 V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create, 269 lagg_clone_destroy, 0); 270 } 271 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 272 vnet_lagg_init, NULL); 273 274 static void 275 vnet_lagg_uninit(const void *unused __unused) 276 { 277 278 if_clone_detach(V_lagg_cloner); 279 LAGG_LIST_LOCK_DESTROY(); 280 } 281 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 282 vnet_lagg_uninit, NULL); 283 284 static int 285 lagg_modevent(module_t mod, int type, void *data) 286 { 287 288 switch (type) { 289 case MOD_LOAD: 290 lagg_input_p = lagg_input; 291 lagg_linkstate_p = lagg_port_state; 292 lagg_detach_cookie = EVENTHANDLER_REGISTER( 293 ifnet_departure_event, lagg_port_ifdetach, NULL, 294 EVENTHANDLER_PRI_ANY); 295 break; 296 case MOD_UNLOAD: 297 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 298 lagg_detach_cookie); 299 lagg_input_p = NULL; 300 lagg_linkstate_p = NULL; 301 break; 302 default: 303 return (EOPNOTSUPP); 304 } 305 return (0); 306 } 307 308 static moduledata_t lagg_mod = { 309 "if_lagg", 310 lagg_modevent, 311 0 312 }; 313 314 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 315 MODULE_VERSION(if_lagg, 1); 316 317 static void 318 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr) 319 { 320 321 KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto", 322 __func__, sc)); 323 324 if (sc->sc_ifflags & IFF_DEBUG) 325 if_printf(sc->sc_ifp, "using proto %u\n", pr); 326 327 if (lagg_protos[pr].pr_attach != NULL) 328 lagg_protos[pr].pr_attach(sc); 329 sc->sc_proto = pr; 330 } 331 332 static void 333 lagg_proto_detach(struct lagg_softc *sc) 334 { 335 lagg_proto pr; 336 337 LAGG_WLOCK_ASSERT(sc); 338 339 pr = sc->sc_proto; 340 sc->sc_proto = LAGG_PROTO_NONE; 341 342 if (lagg_protos[pr].pr_detach != NULL) 343 lagg_protos[pr].pr_detach(sc); 344 else 345 LAGG_WUNLOCK(sc); 346 } 347 348 static int 349 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m) 350 { 351 352 return (lagg_protos[sc->sc_proto].pr_start(sc, m)); 353 } 354 355 static struct mbuf * 356 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 357 { 358 359 return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m)); 360 } 361 362 static int 363 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp) 364 { 365 366 if (lagg_protos[sc->sc_proto].pr_addport == NULL) 367 return (0); 368 else 369 return (lagg_protos[sc->sc_proto].pr_addport(lp)); 370 } 371 372 static void 373 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp) 374 { 375 376 if (lagg_protos[sc->sc_proto].pr_delport != NULL) 377 lagg_protos[sc->sc_proto].pr_delport(lp); 378 } 379 380 static void 381 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp) 382 { 383 384 if (lagg_protos[sc->sc_proto].pr_linkstate != NULL) 385 lagg_protos[sc->sc_proto].pr_linkstate(lp); 386 } 387 388 static void 389 lagg_proto_init(struct lagg_softc *sc) 390 { 391 392 if (lagg_protos[sc->sc_proto].pr_init != NULL) 393 lagg_protos[sc->sc_proto].pr_init(sc); 394 } 395 396 static void 397 lagg_proto_stop(struct lagg_softc *sc) 398 { 399 400 if (lagg_protos[sc->sc_proto].pr_stop != NULL) 401 lagg_protos[sc->sc_proto].pr_stop(sc); 402 } 403 404 static void 405 lagg_proto_lladdr(struct lagg_softc *sc) 406 { 407 408 if (lagg_protos[sc->sc_proto].pr_lladdr != NULL) 409 lagg_protos[sc->sc_proto].pr_lladdr(sc); 410 } 411 412 static void 413 lagg_proto_request(struct lagg_softc *sc, void *v) 414 { 415 416 if (lagg_protos[sc->sc_proto].pr_request != NULL) 417 lagg_protos[sc->sc_proto].pr_request(sc, v); 418 } 419 420 static void 421 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v) 422 { 423 424 if (lagg_protos[sc->sc_proto].pr_portreq != NULL) 425 lagg_protos[sc->sc_proto].pr_portreq(lp, v); 426 } 427 428 /* 429 * This routine is run via an vlan 430 * config EVENT 431 */ 432 static void 433 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 434 { 435 struct lagg_softc *sc = ifp->if_softc; 436 struct lagg_port *lp; 437 struct rm_priotracker tracker; 438 439 if (ifp->if_softc != arg) /* Not our event */ 440 return; 441 442 LAGG_RLOCK(sc, &tracker); 443 if (!SLIST_EMPTY(&sc->sc_ports)) { 444 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 445 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 446 } 447 LAGG_RUNLOCK(sc, &tracker); 448 } 449 450 /* 451 * This routine is run via an vlan 452 * unconfig EVENT 453 */ 454 static void 455 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 456 { 457 struct lagg_softc *sc = ifp->if_softc; 458 struct lagg_port *lp; 459 struct rm_priotracker tracker; 460 461 if (ifp->if_softc != arg) /* Not our event */ 462 return; 463 464 LAGG_RLOCK(sc, &tracker); 465 if (!SLIST_EMPTY(&sc->sc_ports)) { 466 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 467 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 468 } 469 LAGG_RUNLOCK(sc, &tracker); 470 } 471 472 static int 473 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) 474 { 475 struct lagg_softc *sc; 476 struct ifnet *ifp; 477 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ 478 479 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO); 480 ifp = sc->sc_ifp = if_alloc(IFT_ETHER); 481 if (ifp == NULL) { 482 free(sc, M_DEVBUF); 483 return (ENOSPC); 484 } 485 486 if (V_def_use_flowid) 487 sc->sc_opts |= LAGG_OPT_USE_FLOWID; 488 sc->flowid_shift = V_def_flowid_shift; 489 490 /* Hash all layers by default */ 491 sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4; 492 493 lagg_proto_attach(sc, LAGG_PROTO_DEFAULT); 494 495 LAGG_LOCK_INIT(sc); 496 SLIST_INIT(&sc->sc_ports); 497 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc); 498 499 /* Initialise pseudo media types */ 500 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 501 lagg_media_status); 502 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 503 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 504 505 if_initname(ifp, laggname, unit); 506 ifp->if_softc = sc; 507 ifp->if_transmit = lagg_transmit; 508 ifp->if_qflush = lagg_qflush; 509 ifp->if_init = lagg_init; 510 ifp->if_ioctl = lagg_ioctl; 511 ifp->if_get_counter = lagg_get_counter; 512 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 513 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS; 514 515 /* 516 * Attach as an ordinary ethernet device, children will be attached 517 * as special device IFT_IEEE8023ADLAG. 518 */ 519 ether_ifattach(ifp, eaddr); 520 521 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 522 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 523 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 524 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 525 526 /* Insert into the global list of laggs */ 527 LAGG_LIST_LOCK(); 528 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries); 529 LAGG_LIST_UNLOCK(); 530 531 return (0); 532 } 533 534 static void 535 lagg_clone_destroy(struct ifnet *ifp) 536 { 537 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 538 struct lagg_port *lp; 539 540 LAGG_WLOCK(sc); 541 542 lagg_stop(sc); 543 ifp->if_flags &= ~IFF_UP; 544 545 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 546 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 547 548 /* Shutdown and remove lagg ports */ 549 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) 550 lagg_port_destroy(lp, 1); 551 /* Unhook the aggregation protocol */ 552 lagg_proto_detach(sc); 553 554 ifmedia_removeall(&sc->sc_media); 555 ether_ifdetach(ifp); 556 if_free(ifp); 557 558 LAGG_LIST_LOCK(); 559 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries); 560 LAGG_LIST_UNLOCK(); 561 562 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task); 563 LAGG_LOCK_DESTROY(sc); 564 free(sc, M_DEVBUF); 565 } 566 567 static void 568 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr) 569 { 570 struct ifnet *ifp = sc->sc_ifp; 571 struct lagg_port lp; 572 573 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 574 return; 575 576 LAGG_WLOCK_ASSERT(sc); 577 /* 578 * Set the link layer address on the lagg interface. 579 * lagg_proto_lladdr() notifies the MAC change to 580 * the aggregation protocol. iflladdr_event handler which 581 * may trigger gratuitous ARPs for INET will be handled in 582 * a taskqueue. 583 */ 584 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN); 585 lagg_proto_lladdr(sc); 586 587 bzero(&lp, sizeof(lp)); 588 lp.lp_ifp = sc->sc_ifp; 589 lp.lp_softc = sc; 590 591 lagg_port_lladdr(&lp, lladdr); 592 } 593 594 static void 595 lagg_capabilities(struct lagg_softc *sc) 596 { 597 struct lagg_port *lp; 598 int cap = ~0, ena = ~0; 599 u_long hwa = ~0UL; 600 struct ifnet_hw_tsomax hw_tsomax; 601 602 LAGG_WLOCK_ASSERT(sc); 603 604 memset(&hw_tsomax, 0, sizeof(hw_tsomax)); 605 606 /* Get capabilities from the lagg ports */ 607 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 608 cap &= lp->lp_ifp->if_capabilities; 609 ena &= lp->lp_ifp->if_capenable; 610 hwa &= lp->lp_ifp->if_hwassist; 611 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax); 612 } 613 cap = (cap == ~0 ? 0 : cap); 614 ena = (ena == ~0 ? 0 : ena); 615 hwa = (hwa == ~0 ? 0 : hwa); 616 617 if (sc->sc_ifp->if_capabilities != cap || 618 sc->sc_ifp->if_capenable != ena || 619 sc->sc_ifp->if_hwassist != hwa || 620 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) { 621 sc->sc_ifp->if_capabilities = cap; 622 sc->sc_ifp->if_capenable = ena; 623 sc->sc_ifp->if_hwassist = hwa; 624 getmicrotime(&sc->sc_ifp->if_lastchange); 625 626 if (sc->sc_ifflags & IFF_DEBUG) 627 if_printf(sc->sc_ifp, 628 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 629 } 630 } 631 632 static void 633 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr) 634 { 635 struct lagg_softc *sc = lp->lp_softc; 636 struct ifnet *ifp = lp->lp_ifp; 637 struct lagg_llq *llq; 638 int pending = 0; 639 int primary; 640 641 LAGG_WLOCK_ASSERT(sc); 642 643 primary = (sc->sc_primary->lp_ifp == ifp) ? 1 : 0; 644 if (primary == 0 && (lp->lp_detaching || 645 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)) 646 return; 647 648 /* Check to make sure its not already queued to be changed */ 649 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 650 if (llq->llq_ifp == ifp) { 651 pending = 1; 652 break; 653 } 654 } 655 656 if (!pending) { 657 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT); 658 if (llq == NULL) /* XXX what to do */ 659 return; 660 } 661 662 /* Update the lladdr even if pending, it may have changed */ 663 llq->llq_ifp = ifp; 664 llq->llq_primary = primary; 665 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN); 666 667 if (!pending) 668 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries); 669 670 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task); 671 } 672 673 /* 674 * Set the interface MAC address from a taskqueue to avoid a LOR. 675 */ 676 static void 677 lagg_port_setlladdr(void *arg, int pending) 678 { 679 struct lagg_softc *sc = (struct lagg_softc *)arg; 680 struct lagg_llq *llq, *head; 681 struct ifnet *ifp; 682 int error; 683 684 /* Grab a local reference of the queue and remove it from the softc */ 685 LAGG_WLOCK(sc); 686 head = SLIST_FIRST(&sc->sc_llq_head); 687 SLIST_FIRST(&sc->sc_llq_head) = NULL; 688 LAGG_WUNLOCK(sc); 689 690 /* 691 * Traverse the queue and set the lladdr on each ifp. It is safe to do 692 * unlocked as we have the only reference to it. 693 */ 694 for (llq = head; llq != NULL; llq = head) { 695 ifp = llq->llq_ifp; 696 697 CURVNET_SET(ifp->if_vnet); 698 if (llq->llq_primary == 0) { 699 /* 700 * Set the link layer address on the laggport interface. 701 * if_setlladdr() triggers gratuitous ARPs for INET. 702 */ 703 error = if_setlladdr(ifp, llq->llq_lladdr, 704 ETHER_ADDR_LEN); 705 if (error) 706 printf("%s: setlladdr failed on %s\n", __func__, 707 ifp->if_xname); 708 } else 709 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 710 CURVNET_RESTORE(); 711 head = SLIST_NEXT(llq, llq_entries); 712 free(llq, M_DEVBUF); 713 } 714 } 715 716 static int 717 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 718 { 719 struct lagg_softc *sc_ptr; 720 struct lagg_port *lp, *tlp; 721 int error, i; 722 uint64_t *pval; 723 724 LAGG_WLOCK_ASSERT(sc); 725 726 /* Limit the maximal number of lagg ports */ 727 if (sc->sc_count >= LAGG_MAX_PORTS) 728 return (ENOSPC); 729 730 /* Check if port has already been associated to a lagg */ 731 if (ifp->if_lagg != NULL) { 732 /* Port is already in the current lagg? */ 733 lp = (struct lagg_port *)ifp->if_lagg; 734 if (lp->lp_softc == sc) 735 return (EEXIST); 736 return (EBUSY); 737 } 738 739 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 740 if (ifp->if_type != IFT_ETHER) 741 return (EPROTONOSUPPORT); 742 743 /* Allow the first Ethernet member to define the MTU */ 744 if (SLIST_EMPTY(&sc->sc_ports)) 745 sc->sc_ifp->if_mtu = ifp->if_mtu; 746 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 747 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 748 ifp->if_xname); 749 return (EINVAL); 750 } 751 752 if ((lp = malloc(sizeof(struct lagg_port), 753 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 754 return (ENOMEM); 755 756 /* Check if port is a stacked lagg */ 757 LAGG_LIST_LOCK(); 758 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) { 759 if (ifp == sc_ptr->sc_ifp) { 760 LAGG_LIST_UNLOCK(); 761 free(lp, M_DEVBUF); 762 return (EINVAL); 763 /* XXX disable stacking for the moment, its untested */ 764 #ifdef LAGG_PORT_STACKING 765 lp->lp_flags |= LAGG_PORT_STACK; 766 if (lagg_port_checkstacking(sc_ptr) >= 767 LAGG_MAX_STACKING) { 768 LAGG_LIST_UNLOCK(); 769 free(lp, M_DEVBUF); 770 return (E2BIG); 771 } 772 #endif 773 } 774 } 775 LAGG_LIST_UNLOCK(); 776 777 /* Change the interface type */ 778 lp->lp_iftype = ifp->if_type; 779 ifp->if_type = IFT_IEEE8023ADLAG; 780 ifp->if_lagg = lp; 781 lp->lp_ioctl = ifp->if_ioctl; 782 ifp->if_ioctl = lagg_port_ioctl; 783 lp->lp_output = ifp->if_output; 784 ifp->if_output = lagg_port_output; 785 786 lp->lp_ifp = ifp; 787 lp->lp_softc = sc; 788 789 /* Save port link layer address */ 790 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 791 792 if (SLIST_EMPTY(&sc->sc_ports)) { 793 sc->sc_primary = lp; 794 lagg_lladdr(sc, IF_LLADDR(ifp)); 795 } else { 796 /* Update link layer address for this port */ 797 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp)); 798 } 799 800 /* 801 * Insert into the list of ports. 802 * Keep ports sorted by if_index. It is handy, when configuration 803 * is predictable and `ifconfig laggN create ...` command 804 * will lead to the same result each time. 805 */ 806 SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) { 807 if (tlp->lp_ifp->if_index < ifp->if_index && ( 808 SLIST_NEXT(tlp, lp_entries) == NULL || 809 SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index > 810 ifp->if_index)) 811 break; 812 } 813 if (tlp != NULL) 814 SLIST_INSERT_AFTER(tlp, lp, lp_entries); 815 else 816 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 817 sc->sc_count++; 818 819 /* Update lagg capabilities */ 820 lagg_capabilities(sc); 821 lagg_linkstate(sc); 822 823 /* Read port counters */ 824 pval = lp->port_counters.val; 825 for (i = 0; i < IFCOUNTERS; i++, pval++) 826 *pval = ifp->if_get_counter(ifp, i); 827 /* Add multicast addresses and interface flags to this port */ 828 lagg_ether_cmdmulti(lp, 1); 829 lagg_setflags(lp, 1); 830 831 if ((error = lagg_proto_addport(sc, lp)) != 0) { 832 /* Remove the port, without calling pr_delport. */ 833 lagg_port_destroy(lp, 0); 834 return (error); 835 } 836 837 return (0); 838 } 839 840 #ifdef LAGG_PORT_STACKING 841 static int 842 lagg_port_checkstacking(struct lagg_softc *sc) 843 { 844 struct lagg_softc *sc_ptr; 845 struct lagg_port *lp; 846 int m = 0; 847 848 LAGG_WLOCK_ASSERT(sc); 849 850 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 851 if (lp->lp_flags & LAGG_PORT_STACK) { 852 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 853 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 854 } 855 } 856 857 return (m + 1); 858 } 859 #endif 860 861 static int 862 lagg_port_destroy(struct lagg_port *lp, int rundelport) 863 { 864 struct lagg_softc *sc = lp->lp_softc; 865 struct lagg_port *lp_ptr; 866 struct lagg_llq *llq; 867 struct ifnet *ifp = lp->lp_ifp; 868 uint64_t *pval, vdiff; 869 int i; 870 871 LAGG_WLOCK_ASSERT(sc); 872 873 if (rundelport) 874 lagg_proto_delport(sc, lp); 875 876 /* 877 * Remove multicast addresses and interface flags from this port and 878 * reset the MAC address, skip if the interface is being detached. 879 */ 880 if (!lp->lp_detaching) { 881 lagg_ether_cmdmulti(lp, 0); 882 lagg_setflags(lp, 0); 883 lagg_port_lladdr(lp, lp->lp_lladdr); 884 } 885 886 /* Restore interface */ 887 ifp->if_type = lp->lp_iftype; 888 ifp->if_ioctl = lp->lp_ioctl; 889 ifp->if_output = lp->lp_output; 890 ifp->if_lagg = NULL; 891 892 /* Update detached port counters */ 893 pval = lp->port_counters.val; 894 for (i = 0; i < IFCOUNTERS; i++, pval++) { 895 vdiff = ifp->if_get_counter(ifp, i) - *pval; 896 sc->detached_counters.val[i] += vdiff; 897 } 898 899 /* Finally, remove the port from the lagg */ 900 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 901 sc->sc_count--; 902 903 /* Update the primary interface */ 904 if (lp == sc->sc_primary) { 905 uint8_t lladdr[ETHER_ADDR_LEN]; 906 907 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) { 908 bzero(&lladdr, ETHER_ADDR_LEN); 909 } else { 910 bcopy(lp_ptr->lp_lladdr, 911 lladdr, ETHER_ADDR_LEN); 912 } 913 lagg_lladdr(sc, lladdr); 914 sc->sc_primary = lp_ptr; 915 916 /* Update link layer address for each port */ 917 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 918 lagg_port_lladdr(lp_ptr, lladdr); 919 } 920 921 /* Remove any pending lladdr changes from the queue */ 922 if (lp->lp_detaching) { 923 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 924 if (llq->llq_ifp == ifp) { 925 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq, 926 llq_entries); 927 free(llq, M_DEVBUF); 928 break; /* Only appears once */ 929 } 930 } 931 } 932 933 if (lp->lp_ifflags) 934 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 935 936 free(lp, M_DEVBUF); 937 938 /* Update lagg capabilities */ 939 lagg_capabilities(sc); 940 lagg_linkstate(sc); 941 942 return (0); 943 } 944 945 static int 946 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 947 { 948 struct lagg_reqport *rp = (struct lagg_reqport *)data; 949 struct lagg_softc *sc; 950 struct lagg_port *lp = NULL; 951 int error = 0; 952 struct rm_priotracker tracker; 953 954 /* Should be checked by the caller */ 955 if (ifp->if_type != IFT_IEEE8023ADLAG || 956 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 957 goto fallback; 958 959 switch (cmd) { 960 case SIOCGLAGGPORT: 961 if (rp->rp_portname[0] == '\0' || 962 ifunit(rp->rp_portname) != ifp) { 963 error = EINVAL; 964 break; 965 } 966 967 LAGG_RLOCK(sc, &tracker); 968 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 969 error = ENOENT; 970 LAGG_RUNLOCK(sc, &tracker); 971 break; 972 } 973 974 lagg_port2req(lp, rp); 975 LAGG_RUNLOCK(sc, &tracker); 976 break; 977 978 case SIOCSIFCAP: 979 if (lp->lp_ioctl == NULL) { 980 error = EINVAL; 981 break; 982 } 983 error = (*lp->lp_ioctl)(ifp, cmd, data); 984 if (error) 985 break; 986 987 /* Update lagg interface capabilities */ 988 LAGG_WLOCK(sc); 989 lagg_capabilities(sc); 990 LAGG_WUNLOCK(sc); 991 break; 992 993 case SIOCSIFMTU: 994 /* Do not allow the MTU to be changed once joined */ 995 error = EINVAL; 996 break; 997 998 default: 999 goto fallback; 1000 } 1001 1002 return (error); 1003 1004 fallback: 1005 if (lp->lp_ioctl != NULL) 1006 return ((*lp->lp_ioctl)(ifp, cmd, data)); 1007 1008 return (EINVAL); 1009 } 1010 1011 /* 1012 * Requests counter @cnt data. 1013 * 1014 * Counter value is calculated the following way: 1015 * 1) for each port, sum difference between current and "initial" measurements. 1016 * 2) add lagg logical interface counters. 1017 * 3) add data from detached_counters array. 1018 * 1019 * We also do the following things on ports attach/detach: 1020 * 1) On port attach we store all counters it has into port_counter array. 1021 * 2) On port detach we add the different between "initial" and 1022 * current counters data to detached_counters array. 1023 */ 1024 static uint64_t 1025 lagg_get_counter(struct ifnet *ifp, ift_counter cnt) 1026 { 1027 struct lagg_softc *sc; 1028 struct lagg_port *lp; 1029 struct ifnet *lpifp; 1030 struct rm_priotracker tracker; 1031 uint64_t newval, oldval, vsum; 1032 1033 /* Revise this when we've got non-generic counters. */ 1034 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt)); 1035 1036 sc = (struct lagg_softc *)ifp->if_softc; 1037 LAGG_RLOCK(sc, &tracker); 1038 1039 vsum = 0; 1040 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1041 /* Saved attached value */ 1042 oldval = lp->port_counters.val[cnt]; 1043 /* current value */ 1044 lpifp = lp->lp_ifp; 1045 newval = lpifp->if_get_counter(lpifp, cnt); 1046 /* Calculate diff and save new */ 1047 vsum += newval - oldval; 1048 } 1049 1050 /* 1051 * Add counter data which might be added by upper 1052 * layer protocols operating on logical interface. 1053 */ 1054 vsum += if_get_counter_default(ifp, cnt); 1055 1056 /* 1057 * Add counter data from detached ports counters 1058 */ 1059 vsum += sc->detached_counters.val[cnt]; 1060 1061 LAGG_RUNLOCK(sc, &tracker); 1062 1063 return (vsum); 1064 } 1065 1066 /* 1067 * For direct output to child ports. 1068 */ 1069 static int 1070 lagg_port_output(struct ifnet *ifp, struct mbuf *m, 1071 const struct sockaddr *dst, struct route *ro) 1072 { 1073 struct lagg_port *lp = ifp->if_lagg; 1074 1075 switch (dst->sa_family) { 1076 case pseudo_AF_HDRCMPLT: 1077 case AF_UNSPEC: 1078 return ((*lp->lp_output)(ifp, m, dst, ro)); 1079 } 1080 1081 /* drop any other frames */ 1082 m_freem(m); 1083 return (ENETDOWN); 1084 } 1085 1086 static void 1087 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 1088 { 1089 struct lagg_port *lp; 1090 struct lagg_softc *sc; 1091 1092 if ((lp = ifp->if_lagg) == NULL) 1093 return; 1094 /* If the ifnet is just being renamed, don't do anything. */ 1095 if (ifp->if_flags & IFF_RENAMING) 1096 return; 1097 1098 sc = lp->lp_softc; 1099 1100 LAGG_WLOCK(sc); 1101 lp->lp_detaching = 1; 1102 lagg_port_destroy(lp, 1); 1103 LAGG_WUNLOCK(sc); 1104 } 1105 1106 static void 1107 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 1108 { 1109 struct lagg_softc *sc = lp->lp_softc; 1110 1111 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 1112 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 1113 rp->rp_prio = lp->lp_prio; 1114 rp->rp_flags = lp->lp_flags; 1115 lagg_proto_portreq(sc, lp, &rp->rp_psc); 1116 1117 /* Add protocol specific flags */ 1118 switch (sc->sc_proto) { 1119 case LAGG_PROTO_FAILOVER: 1120 if (lp == sc->sc_primary) 1121 rp->rp_flags |= LAGG_PORT_MASTER; 1122 if (lp == lagg_link_active(sc, sc->sc_primary)) 1123 rp->rp_flags |= LAGG_PORT_ACTIVE; 1124 break; 1125 1126 case LAGG_PROTO_ROUNDROBIN: 1127 case LAGG_PROTO_LOADBALANCE: 1128 case LAGG_PROTO_ETHERCHANNEL: 1129 case LAGG_PROTO_BROADCAST: 1130 if (LAGG_PORTACTIVE(lp)) 1131 rp->rp_flags |= LAGG_PORT_ACTIVE; 1132 break; 1133 1134 case LAGG_PROTO_LACP: 1135 /* LACP has a different definition of active */ 1136 if (lacp_isactive(lp)) 1137 rp->rp_flags |= LAGG_PORT_ACTIVE; 1138 if (lacp_iscollecting(lp)) 1139 rp->rp_flags |= LAGG_PORT_COLLECTING; 1140 if (lacp_isdistributing(lp)) 1141 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 1142 break; 1143 } 1144 1145 } 1146 1147 static void 1148 lagg_init(void *xsc) 1149 { 1150 struct lagg_softc *sc = (struct lagg_softc *)xsc; 1151 struct lagg_port *lp; 1152 struct ifnet *ifp = sc->sc_ifp; 1153 1154 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1155 return; 1156 1157 LAGG_WLOCK(sc); 1158 1159 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1160 /* Update the port lladdrs */ 1161 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1162 lagg_port_lladdr(lp, IF_LLADDR(ifp)); 1163 1164 lagg_proto_init(sc); 1165 1166 LAGG_WUNLOCK(sc); 1167 } 1168 1169 static void 1170 lagg_stop(struct lagg_softc *sc) 1171 { 1172 struct ifnet *ifp = sc->sc_ifp; 1173 1174 LAGG_WLOCK_ASSERT(sc); 1175 1176 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 1177 return; 1178 1179 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1180 1181 lagg_proto_stop(sc); 1182 } 1183 1184 static int 1185 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1186 { 1187 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1188 struct lagg_reqall *ra = (struct lagg_reqall *)data; 1189 struct lagg_reqopts *ro = (struct lagg_reqopts *)data; 1190 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 1191 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 1192 struct ifreq *ifr = (struct ifreq *)data; 1193 struct lagg_port *lp; 1194 struct ifnet *tpif; 1195 struct thread *td = curthread; 1196 char *buf, *outbuf; 1197 int count, buflen, len, error = 0; 1198 struct rm_priotracker tracker; 1199 1200 bzero(&rpbuf, sizeof(rpbuf)); 1201 1202 switch (cmd) { 1203 case SIOCGLAGG: 1204 LAGG_RLOCK(sc, &tracker); 1205 count = 0; 1206 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1207 count++; 1208 buflen = count * sizeof(struct lagg_reqport); 1209 LAGG_RUNLOCK(sc, &tracker); 1210 1211 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 1212 1213 LAGG_RLOCK(sc, &tracker); 1214 ra->ra_proto = sc->sc_proto; 1215 lagg_proto_request(sc, &ra->ra_psc); 1216 count = 0; 1217 buf = outbuf; 1218 len = min(ra->ra_size, buflen); 1219 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1220 if (len < sizeof(rpbuf)) 1221 break; 1222 1223 lagg_port2req(lp, &rpbuf); 1224 memcpy(buf, &rpbuf, sizeof(rpbuf)); 1225 count++; 1226 buf += sizeof(rpbuf); 1227 len -= sizeof(rpbuf); 1228 } 1229 LAGG_RUNLOCK(sc, &tracker); 1230 ra->ra_ports = count; 1231 ra->ra_size = count * sizeof(rpbuf); 1232 error = copyout(outbuf, ra->ra_port, ra->ra_size); 1233 free(outbuf, M_TEMP); 1234 break; 1235 case SIOCSLAGG: 1236 error = priv_check(td, PRIV_NET_LAGG); 1237 if (error) 1238 break; 1239 if (ra->ra_proto < 1 || ra->ra_proto >= LAGG_PROTO_MAX) { 1240 error = EPROTONOSUPPORT; 1241 break; 1242 } 1243 1244 LAGG_WLOCK(sc); 1245 lagg_proto_detach(sc); 1246 lagg_proto_attach(sc, ra->ra_proto); 1247 break; 1248 case SIOCGLAGGOPTS: 1249 ro->ro_opts = sc->sc_opts; 1250 if (sc->sc_proto == LAGG_PROTO_LACP) { 1251 struct lacp_softc *lsc; 1252 1253 lsc = (struct lacp_softc *)sc->sc_psc; 1254 if (lsc->lsc_debug.lsc_tx_test != 0) 1255 ro->ro_opts |= LAGG_OPT_LACP_TXTEST; 1256 if (lsc->lsc_debug.lsc_rx_test != 0) 1257 ro->ro_opts |= LAGG_OPT_LACP_RXTEST; 1258 if (lsc->lsc_strict_mode != 0) 1259 ro->ro_opts |= LAGG_OPT_LACP_STRICT; 1260 if (lsc->lsc_fast_timeout != 0) 1261 ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT; 1262 1263 ro->ro_active = sc->sc_active; 1264 } else { 1265 ro->ro_active = 0; 1266 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1267 ro->ro_active += LAGG_PORTACTIVE(lp); 1268 } 1269 ro->ro_flapping = sc->sc_flapping; 1270 ro->ro_flowid_shift = sc->flowid_shift; 1271 break; 1272 case SIOCSLAGGOPTS: 1273 error = priv_check(td, PRIV_NET_LAGG); 1274 if (error) 1275 break; 1276 if (ro->ro_opts == 0) 1277 break; 1278 /* 1279 * Set options. LACP options are stored in sc->sc_psc, 1280 * not in sc_opts. 1281 */ 1282 int valid, lacp; 1283 1284 switch (ro->ro_opts) { 1285 case LAGG_OPT_USE_FLOWID: 1286 case -LAGG_OPT_USE_FLOWID: 1287 case LAGG_OPT_FLOWIDSHIFT: 1288 valid = 1; 1289 lacp = 0; 1290 break; 1291 case LAGG_OPT_LACP_TXTEST: 1292 case -LAGG_OPT_LACP_TXTEST: 1293 case LAGG_OPT_LACP_RXTEST: 1294 case -LAGG_OPT_LACP_RXTEST: 1295 case LAGG_OPT_LACP_STRICT: 1296 case -LAGG_OPT_LACP_STRICT: 1297 case LAGG_OPT_LACP_TIMEOUT: 1298 case -LAGG_OPT_LACP_TIMEOUT: 1299 valid = lacp = 1; 1300 break; 1301 default: 1302 valid = lacp = 0; 1303 break; 1304 } 1305 1306 LAGG_WLOCK(sc); 1307 if (valid == 0 || 1308 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) { 1309 /* Invalid combination of options specified. */ 1310 error = EINVAL; 1311 LAGG_WUNLOCK(sc); 1312 break; /* Return from SIOCSLAGGOPTS. */ 1313 } 1314 /* 1315 * Store new options into sc->sc_opts except for 1316 * FLOWIDSHIFT and LACP options. 1317 */ 1318 if (lacp == 0) { 1319 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT) 1320 sc->flowid_shift = ro->ro_flowid_shift; 1321 else if (ro->ro_opts > 0) 1322 sc->sc_opts |= ro->ro_opts; 1323 else 1324 sc->sc_opts &= ~ro->ro_opts; 1325 } else { 1326 struct lacp_softc *lsc; 1327 struct lacp_port *lp; 1328 1329 lsc = (struct lacp_softc *)sc->sc_psc; 1330 1331 switch (ro->ro_opts) { 1332 case LAGG_OPT_LACP_TXTEST: 1333 lsc->lsc_debug.lsc_tx_test = 1; 1334 break; 1335 case -LAGG_OPT_LACP_TXTEST: 1336 lsc->lsc_debug.lsc_tx_test = 0; 1337 break; 1338 case LAGG_OPT_LACP_RXTEST: 1339 lsc->lsc_debug.lsc_rx_test = 1; 1340 break; 1341 case -LAGG_OPT_LACP_RXTEST: 1342 lsc->lsc_debug.lsc_rx_test = 0; 1343 break; 1344 case LAGG_OPT_LACP_STRICT: 1345 lsc->lsc_strict_mode = 1; 1346 break; 1347 case -LAGG_OPT_LACP_STRICT: 1348 lsc->lsc_strict_mode = 0; 1349 break; 1350 case LAGG_OPT_LACP_TIMEOUT: 1351 LACP_LOCK(lsc); 1352 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1353 lp->lp_state |= LACP_STATE_TIMEOUT; 1354 LACP_UNLOCK(lsc); 1355 lsc->lsc_fast_timeout = 1; 1356 break; 1357 case -LAGG_OPT_LACP_TIMEOUT: 1358 LACP_LOCK(lsc); 1359 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1360 lp->lp_state &= ~LACP_STATE_TIMEOUT; 1361 LACP_UNLOCK(lsc); 1362 lsc->lsc_fast_timeout = 0; 1363 break; 1364 } 1365 } 1366 LAGG_WUNLOCK(sc); 1367 break; 1368 case SIOCGLAGGFLAGS: 1369 rf->rf_flags = 0; 1370 LAGG_RLOCK(sc, &tracker); 1371 if (sc->sc_flags & MBUF_HASHFLAG_L2) 1372 rf->rf_flags |= LAGG_F_HASHL2; 1373 if (sc->sc_flags & MBUF_HASHFLAG_L3) 1374 rf->rf_flags |= LAGG_F_HASHL3; 1375 if (sc->sc_flags & MBUF_HASHFLAG_L4) 1376 rf->rf_flags |= LAGG_F_HASHL4; 1377 LAGG_RUNLOCK(sc, &tracker); 1378 break; 1379 case SIOCSLAGGHASH: 1380 error = priv_check(td, PRIV_NET_LAGG); 1381 if (error) 1382 break; 1383 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1384 error = EINVAL; 1385 break; 1386 } 1387 LAGG_WLOCK(sc); 1388 sc->sc_flags = 0; 1389 if (rf->rf_flags & LAGG_F_HASHL2) 1390 sc->sc_flags |= MBUF_HASHFLAG_L2; 1391 if (rf->rf_flags & LAGG_F_HASHL3) 1392 sc->sc_flags |= MBUF_HASHFLAG_L3; 1393 if (rf->rf_flags & LAGG_F_HASHL4) 1394 sc->sc_flags |= MBUF_HASHFLAG_L4; 1395 LAGG_WUNLOCK(sc); 1396 break; 1397 case SIOCGLAGGPORT: 1398 if (rp->rp_portname[0] == '\0' || 1399 (tpif = ifunit(rp->rp_portname)) == NULL) { 1400 error = EINVAL; 1401 break; 1402 } 1403 1404 LAGG_RLOCK(sc, &tracker); 1405 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1406 lp->lp_softc != sc) { 1407 error = ENOENT; 1408 LAGG_RUNLOCK(sc, &tracker); 1409 break; 1410 } 1411 1412 lagg_port2req(lp, rp); 1413 LAGG_RUNLOCK(sc, &tracker); 1414 break; 1415 case SIOCSLAGGPORT: 1416 error = priv_check(td, PRIV_NET_LAGG); 1417 if (error) 1418 break; 1419 if (rp->rp_portname[0] == '\0' || 1420 (tpif = ifunit(rp->rp_portname)) == NULL) { 1421 error = EINVAL; 1422 break; 1423 } 1424 #ifdef INET6 1425 /* 1426 * A laggport interface should not have inet6 address 1427 * because two interfaces with a valid link-local 1428 * scope zone must not be merged in any form. This 1429 * restriction is needed to prevent violation of 1430 * link-local scope zone. Attempts to add a laggport 1431 * interface which has inet6 addresses triggers 1432 * removal of all inet6 addresses on the member 1433 * interface. 1434 */ 1435 if (in6ifa_llaonifp(tpif)) { 1436 in6_ifdetach(tpif); 1437 if_printf(sc->sc_ifp, 1438 "IPv6 addresses on %s have been removed " 1439 "before adding it as a member to prevent " 1440 "IPv6 address scope violation.\n", 1441 tpif->if_xname); 1442 } 1443 #endif 1444 LAGG_WLOCK(sc); 1445 error = lagg_port_create(sc, tpif); 1446 LAGG_WUNLOCK(sc); 1447 break; 1448 case SIOCSLAGGDELPORT: 1449 error = priv_check(td, PRIV_NET_LAGG); 1450 if (error) 1451 break; 1452 if (rp->rp_portname[0] == '\0' || 1453 (tpif = ifunit(rp->rp_portname)) == NULL) { 1454 error = EINVAL; 1455 break; 1456 } 1457 1458 LAGG_WLOCK(sc); 1459 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1460 lp->lp_softc != sc) { 1461 error = ENOENT; 1462 LAGG_WUNLOCK(sc); 1463 break; 1464 } 1465 1466 error = lagg_port_destroy(lp, 1); 1467 LAGG_WUNLOCK(sc); 1468 break; 1469 case SIOCSIFFLAGS: 1470 /* Set flags on ports too */ 1471 LAGG_WLOCK(sc); 1472 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1473 lagg_setflags(lp, 1); 1474 } 1475 LAGG_WUNLOCK(sc); 1476 1477 if (!(ifp->if_flags & IFF_UP) && 1478 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1479 /* 1480 * If interface is marked down and it is running, 1481 * then stop and disable it. 1482 */ 1483 LAGG_WLOCK(sc); 1484 lagg_stop(sc); 1485 LAGG_WUNLOCK(sc); 1486 } else if ((ifp->if_flags & IFF_UP) && 1487 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1488 /* 1489 * If interface is marked up and it is stopped, then 1490 * start it. 1491 */ 1492 (*ifp->if_init)(sc); 1493 } 1494 break; 1495 case SIOCADDMULTI: 1496 case SIOCDELMULTI: 1497 LAGG_WLOCK(sc); 1498 error = lagg_ether_setmulti(sc); 1499 LAGG_WUNLOCK(sc); 1500 break; 1501 case SIOCSIFMEDIA: 1502 case SIOCGIFMEDIA: 1503 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1504 break; 1505 1506 case SIOCSIFCAP: 1507 case SIOCSIFMTU: 1508 /* Do not allow the MTU or caps to be directly changed */ 1509 error = EINVAL; 1510 break; 1511 1512 default: 1513 error = ether_ioctl(ifp, cmd, data); 1514 break; 1515 } 1516 return (error); 1517 } 1518 1519 static int 1520 lagg_ether_setmulti(struct lagg_softc *sc) 1521 { 1522 struct lagg_port *lp; 1523 1524 LAGG_WLOCK_ASSERT(sc); 1525 1526 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1527 /* First, remove any existing filter entries. */ 1528 lagg_ether_cmdmulti(lp, 0); 1529 /* copy all addresses from the lagg interface to the port */ 1530 lagg_ether_cmdmulti(lp, 1); 1531 } 1532 return (0); 1533 } 1534 1535 static int 1536 lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1537 { 1538 struct lagg_softc *sc = lp->lp_softc; 1539 struct ifnet *ifp = lp->lp_ifp; 1540 struct ifnet *scifp = sc->sc_ifp; 1541 struct lagg_mc *mc; 1542 struct ifmultiaddr *ifma; 1543 int error; 1544 1545 LAGG_WLOCK_ASSERT(sc); 1546 1547 if (set) { 1548 IF_ADDR_WLOCK(scifp); 1549 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1550 if (ifma->ifma_addr->sa_family != AF_LINK) 1551 continue; 1552 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1553 if (mc == NULL) { 1554 IF_ADDR_WUNLOCK(scifp); 1555 return (ENOMEM); 1556 } 1557 bcopy(ifma->ifma_addr, &mc->mc_addr, 1558 ifma->ifma_addr->sa_len); 1559 mc->mc_addr.sdl_index = ifp->if_index; 1560 mc->mc_ifma = NULL; 1561 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1562 } 1563 IF_ADDR_WUNLOCK(scifp); 1564 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) { 1565 error = if_addmulti(ifp, 1566 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma); 1567 if (error) 1568 return (error); 1569 } 1570 } else { 1571 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1572 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1573 if (mc->mc_ifma && !lp->lp_detaching) 1574 if_delmulti_ifma(mc->mc_ifma); 1575 free(mc, M_DEVBUF); 1576 } 1577 } 1578 return (0); 1579 } 1580 1581 /* Handle a ref counted flag that should be set on the lagg port as well */ 1582 static int 1583 lagg_setflag(struct lagg_port *lp, int flag, int status, 1584 int (*func)(struct ifnet *, int)) 1585 { 1586 struct lagg_softc *sc = lp->lp_softc; 1587 struct ifnet *scifp = sc->sc_ifp; 1588 struct ifnet *ifp = lp->lp_ifp; 1589 int error; 1590 1591 LAGG_WLOCK_ASSERT(sc); 1592 1593 status = status ? (scifp->if_flags & flag) : 0; 1594 /* Now "status" contains the flag value or 0 */ 1595 1596 /* 1597 * See if recorded ports status is different from what 1598 * we want it to be. If it is, flip it. We record ports 1599 * status in lp_ifflags so that we won't clear ports flag 1600 * we haven't set. In fact, we don't clear or set ports 1601 * flags directly, but get or release references to them. 1602 * That's why we can be sure that recorded flags still are 1603 * in accord with actual ports flags. 1604 */ 1605 if (status != (lp->lp_ifflags & flag)) { 1606 error = (*func)(ifp, status); 1607 if (error) 1608 return (error); 1609 lp->lp_ifflags &= ~flag; 1610 lp->lp_ifflags |= status; 1611 } 1612 return (0); 1613 } 1614 1615 /* 1616 * Handle IFF_* flags that require certain changes on the lagg port 1617 * if "status" is true, update ports flags respective to the lagg 1618 * if "status" is false, forcedly clear the flags set on port. 1619 */ 1620 static int 1621 lagg_setflags(struct lagg_port *lp, int status) 1622 { 1623 int error, i; 1624 1625 for (i = 0; lagg_pflags[i].flag; i++) { 1626 error = lagg_setflag(lp, lagg_pflags[i].flag, 1627 status, lagg_pflags[i].func); 1628 if (error) 1629 return (error); 1630 } 1631 return (0); 1632 } 1633 1634 static int 1635 lagg_transmit(struct ifnet *ifp, struct mbuf *m) 1636 { 1637 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1638 int error, len, mcast; 1639 struct rm_priotracker tracker; 1640 1641 len = m->m_pkthdr.len; 1642 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1643 1644 LAGG_RLOCK(sc, &tracker); 1645 /* We need a Tx algorithm and at least one port */ 1646 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1647 LAGG_RUNLOCK(sc, &tracker); 1648 m_freem(m); 1649 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1650 return (ENXIO); 1651 } 1652 1653 ETHER_BPF_MTAP(ifp, m); 1654 1655 error = lagg_proto_start(sc, m); 1656 LAGG_RUNLOCK(sc, &tracker); 1657 1658 if (error != 0) 1659 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1660 1661 return (error); 1662 } 1663 1664 /* 1665 * The ifp->if_qflush entry point for lagg(4) is no-op. 1666 */ 1667 static void 1668 lagg_qflush(struct ifnet *ifp __unused) 1669 { 1670 } 1671 1672 static struct mbuf * 1673 lagg_input(struct ifnet *ifp, struct mbuf *m) 1674 { 1675 struct lagg_port *lp = ifp->if_lagg; 1676 struct lagg_softc *sc = lp->lp_softc; 1677 struct ifnet *scifp = sc->sc_ifp; 1678 struct rm_priotracker tracker; 1679 1680 LAGG_RLOCK(sc, &tracker); 1681 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1682 (lp->lp_flags & LAGG_PORT_DISABLED) || 1683 sc->sc_proto == LAGG_PROTO_NONE) { 1684 LAGG_RUNLOCK(sc, &tracker); 1685 m_freem(m); 1686 return (NULL); 1687 } 1688 1689 ETHER_BPF_MTAP(scifp, m); 1690 1691 if (lp->lp_detaching != 0) { 1692 m_freem(m); 1693 m = NULL; 1694 } else 1695 m = lagg_proto_input(sc, lp, m); 1696 1697 if (m != NULL) { 1698 if (scifp->if_flags & IFF_MONITOR) { 1699 m_freem(m); 1700 m = NULL; 1701 } 1702 } 1703 1704 LAGG_RUNLOCK(sc, &tracker); 1705 return (m); 1706 } 1707 1708 static int 1709 lagg_media_change(struct ifnet *ifp) 1710 { 1711 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1712 1713 if (sc->sc_ifflags & IFF_DEBUG) 1714 printf("%s\n", __func__); 1715 1716 /* Ignore */ 1717 return (0); 1718 } 1719 1720 static void 1721 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1722 { 1723 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1724 struct lagg_port *lp; 1725 struct rm_priotracker tracker; 1726 1727 imr->ifm_status = IFM_AVALID; 1728 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1729 1730 LAGG_RLOCK(sc, &tracker); 1731 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1732 if (LAGG_PORTACTIVE(lp)) 1733 imr->ifm_status |= IFM_ACTIVE; 1734 } 1735 LAGG_RUNLOCK(sc, &tracker); 1736 } 1737 1738 static void 1739 lagg_linkstate(struct lagg_softc *sc) 1740 { 1741 struct lagg_port *lp; 1742 int new_link = LINK_STATE_DOWN; 1743 uint64_t speed; 1744 1745 /* Our link is considered up if at least one of our ports is active */ 1746 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1747 if (lp->lp_ifp->if_link_state == LINK_STATE_UP) { 1748 new_link = LINK_STATE_UP; 1749 break; 1750 } 1751 } 1752 if_link_state_change(sc->sc_ifp, new_link); 1753 1754 /* Update if_baudrate to reflect the max possible speed */ 1755 switch (sc->sc_proto) { 1756 case LAGG_PROTO_FAILOVER: 1757 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1758 sc->sc_primary->lp_ifp->if_baudrate : 0; 1759 break; 1760 case LAGG_PROTO_ROUNDROBIN: 1761 case LAGG_PROTO_LOADBALANCE: 1762 case LAGG_PROTO_ETHERCHANNEL: 1763 case LAGG_PROTO_BROADCAST: 1764 speed = 0; 1765 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1766 speed += lp->lp_ifp->if_baudrate; 1767 sc->sc_ifp->if_baudrate = speed; 1768 break; 1769 case LAGG_PROTO_LACP: 1770 /* LACP updates if_baudrate itself */ 1771 break; 1772 } 1773 } 1774 1775 static void 1776 lagg_port_state(struct ifnet *ifp, int state) 1777 { 1778 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1779 struct lagg_softc *sc = NULL; 1780 1781 if (lp != NULL) 1782 sc = lp->lp_softc; 1783 if (sc == NULL) 1784 return; 1785 1786 LAGG_WLOCK(sc); 1787 lagg_linkstate(sc); 1788 lagg_proto_linkstate(sc, lp); 1789 LAGG_WUNLOCK(sc); 1790 } 1791 1792 struct lagg_port * 1793 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1794 { 1795 struct lagg_port *lp_next, *rval = NULL; 1796 // int new_link = LINK_STATE_DOWN; 1797 1798 LAGG_RLOCK_ASSERT(sc); 1799 /* 1800 * Search a port which reports an active link state. 1801 */ 1802 1803 if (lp == NULL) 1804 goto search; 1805 if (LAGG_PORTACTIVE(lp)) { 1806 rval = lp; 1807 goto found; 1808 } 1809 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1810 LAGG_PORTACTIVE(lp_next)) { 1811 rval = lp_next; 1812 goto found; 1813 } 1814 1815 search: 1816 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1817 if (LAGG_PORTACTIVE(lp_next)) { 1818 rval = lp_next; 1819 goto found; 1820 } 1821 } 1822 1823 found: 1824 if (rval != NULL) { 1825 /* 1826 * The IEEE 802.1D standard assumes that a lagg with 1827 * multiple ports is always full duplex. This is valid 1828 * for load sharing laggs and if at least two links 1829 * are active. Unfortunately, checking the latter would 1830 * be too expensive at this point. 1831 XXX 1832 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1833 (sc->sc_count > 1)) 1834 new_link = LINK_STATE_FULL_DUPLEX; 1835 else 1836 new_link = rval->lp_link_state; 1837 */ 1838 } 1839 1840 return (rval); 1841 } 1842 1843 int 1844 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1845 { 1846 1847 return (ifp->if_transmit)(ifp, m); 1848 } 1849 1850 /* 1851 * Simple round robin aggregation 1852 */ 1853 static void 1854 lagg_rr_attach(struct lagg_softc *sc) 1855 { 1856 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1857 sc->sc_seq = 0; 1858 } 1859 1860 static int 1861 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 1862 { 1863 struct lagg_port *lp; 1864 uint32_t p; 1865 1866 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1867 p %= sc->sc_count; 1868 lp = SLIST_FIRST(&sc->sc_ports); 1869 while (p--) 1870 lp = SLIST_NEXT(lp, lp_entries); 1871 1872 /* 1873 * Check the port's link state. This will return the next active 1874 * port if the link is down or the port is NULL. 1875 */ 1876 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1877 m_freem(m); 1878 return (ENETDOWN); 1879 } 1880 1881 /* Send mbuf */ 1882 return (lagg_enqueue(lp->lp_ifp, m)); 1883 } 1884 1885 static struct mbuf * 1886 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1887 { 1888 struct ifnet *ifp = sc->sc_ifp; 1889 1890 /* Just pass in the packet to our lagg device */ 1891 m->m_pkthdr.rcvif = ifp; 1892 1893 return (m); 1894 } 1895 1896 /* 1897 * Broadcast mode 1898 */ 1899 static int 1900 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m) 1901 { 1902 int active_ports = 0; 1903 int errors = 0; 1904 int ret; 1905 struct lagg_port *lp, *last = NULL; 1906 struct mbuf *m0; 1907 1908 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1909 if (!LAGG_PORTACTIVE(lp)) 1910 continue; 1911 1912 active_ports++; 1913 1914 if (last != NULL) { 1915 m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT); 1916 if (m0 == NULL) { 1917 ret = ENOBUFS; 1918 errors++; 1919 break; 1920 } 1921 1922 ret = lagg_enqueue(last->lp_ifp, m0); 1923 if (ret != 0) 1924 errors++; 1925 } 1926 last = lp; 1927 } 1928 if (last == NULL) { 1929 m_freem(m); 1930 return (ENOENT); 1931 } 1932 if ((last = lagg_link_active(sc, last)) == NULL) { 1933 m_freem(m); 1934 return (ENETDOWN); 1935 } 1936 1937 ret = lagg_enqueue(last->lp_ifp, m); 1938 if (ret != 0) 1939 errors++; 1940 1941 if (errors == 0) 1942 return (ret); 1943 1944 return (0); 1945 } 1946 1947 static struct mbuf* 1948 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1949 { 1950 struct ifnet *ifp = sc->sc_ifp; 1951 1952 /* Just pass in the packet to our lagg device */ 1953 m->m_pkthdr.rcvif = ifp; 1954 return (m); 1955 } 1956 1957 /* 1958 * Active failover 1959 */ 1960 static int 1961 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 1962 { 1963 struct lagg_port *lp; 1964 1965 /* Use the master port if active or the next available port */ 1966 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 1967 m_freem(m); 1968 return (ENETDOWN); 1969 } 1970 1971 /* Send mbuf */ 1972 return (lagg_enqueue(lp->lp_ifp, m)); 1973 } 1974 1975 static struct mbuf * 1976 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1977 { 1978 struct ifnet *ifp = sc->sc_ifp; 1979 struct lagg_port *tmp_tp; 1980 1981 if (lp == sc->sc_primary || V_lagg_failover_rx_all) { 1982 m->m_pkthdr.rcvif = ifp; 1983 return (m); 1984 } 1985 1986 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1987 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1988 /* 1989 * If tmp_tp is null, we've recieved a packet when all 1990 * our links are down. Weird, but process it anyways. 1991 */ 1992 if ((tmp_tp == NULL || tmp_tp == lp)) { 1993 m->m_pkthdr.rcvif = ifp; 1994 return (m); 1995 } 1996 } 1997 1998 m_freem(m); 1999 return (NULL); 2000 } 2001 2002 /* 2003 * Loadbalancing 2004 */ 2005 static void 2006 lagg_lb_attach(struct lagg_softc *sc) 2007 { 2008 struct lagg_port *lp; 2009 struct lagg_lb *lb; 2010 2011 lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO); 2012 2013 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 2014 2015 lb->lb_key = m_ether_tcpip_hash_init(); 2016 sc->sc_psc = lb; 2017 2018 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2019 lagg_lb_port_create(lp); 2020 } 2021 2022 static void 2023 lagg_lb_detach(struct lagg_softc *sc) 2024 { 2025 struct lagg_lb *lb; 2026 2027 lb = (struct lagg_lb *)sc->sc_psc; 2028 LAGG_WUNLOCK(sc); 2029 if (lb != NULL) 2030 free(lb, M_DEVBUF); 2031 } 2032 2033 static int 2034 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 2035 { 2036 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 2037 struct lagg_port *lp_next; 2038 int i = 0; 2039 2040 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 2041 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 2042 if (lp_next == lp) 2043 continue; 2044 if (i >= LAGG_MAX_PORTS) 2045 return (EINVAL); 2046 if (sc->sc_ifflags & IFF_DEBUG) 2047 printf("%s: port %s at index %d\n", 2048 sc->sc_ifname, lp_next->lp_ifp->if_xname, i); 2049 lb->lb_ports[i++] = lp_next; 2050 } 2051 2052 return (0); 2053 } 2054 2055 static int 2056 lagg_lb_port_create(struct lagg_port *lp) 2057 { 2058 struct lagg_softc *sc = lp->lp_softc; 2059 return (lagg_lb_porttable(sc, NULL)); 2060 } 2061 2062 static void 2063 lagg_lb_port_destroy(struct lagg_port *lp) 2064 { 2065 struct lagg_softc *sc = lp->lp_softc; 2066 lagg_lb_porttable(sc, lp); 2067 } 2068 2069 static int 2070 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 2071 { 2072 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 2073 struct lagg_port *lp = NULL; 2074 uint32_t p = 0; 2075 2076 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && 2077 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 2078 p = m->m_pkthdr.flowid >> sc->flowid_shift; 2079 else 2080 p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key); 2081 p %= sc->sc_count; 2082 lp = lb->lb_ports[p]; 2083 2084 /* 2085 * Check the port's link state. This will return the next active 2086 * port if the link is down or the port is NULL. 2087 */ 2088 if ((lp = lagg_link_active(sc, lp)) == NULL) { 2089 m_freem(m); 2090 return (ENETDOWN); 2091 } 2092 2093 /* Send mbuf */ 2094 return (lagg_enqueue(lp->lp_ifp, m)); 2095 } 2096 2097 static struct mbuf * 2098 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2099 { 2100 struct ifnet *ifp = sc->sc_ifp; 2101 2102 /* Just pass in the packet to our lagg device */ 2103 m->m_pkthdr.rcvif = ifp; 2104 2105 return (m); 2106 } 2107 2108 /* 2109 * 802.3ad LACP 2110 */ 2111 static void 2112 lagg_lacp_attach(struct lagg_softc *sc) 2113 { 2114 struct lagg_port *lp; 2115 2116 lacp_attach(sc); 2117 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2118 lacp_port_create(lp); 2119 } 2120 2121 static void 2122 lagg_lacp_detach(struct lagg_softc *sc) 2123 { 2124 struct lagg_port *lp; 2125 void *psc; 2126 2127 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2128 lacp_port_destroy(lp); 2129 2130 psc = sc->sc_psc; 2131 sc->sc_psc = NULL; 2132 LAGG_WUNLOCK(sc); 2133 2134 lacp_detach(psc); 2135 } 2136 2137 static void 2138 lagg_lacp_lladdr(struct lagg_softc *sc) 2139 { 2140 struct lagg_port *lp; 2141 2142 /* purge all the lacp ports */ 2143 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2144 lacp_port_destroy(lp); 2145 2146 /* add them back in */ 2147 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2148 lacp_port_create(lp); 2149 } 2150 2151 static int 2152 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 2153 { 2154 struct lagg_port *lp; 2155 2156 lp = lacp_select_tx_port(sc, m); 2157 if (lp == NULL) { 2158 m_freem(m); 2159 return (ENETDOWN); 2160 } 2161 2162 /* Send mbuf */ 2163 return (lagg_enqueue(lp->lp_ifp, m)); 2164 } 2165 2166 static struct mbuf * 2167 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2168 { 2169 struct ifnet *ifp = sc->sc_ifp; 2170 struct ether_header *eh; 2171 u_short etype; 2172 2173 eh = mtod(m, struct ether_header *); 2174 etype = ntohs(eh->ether_type); 2175 2176 /* Tap off LACP control messages */ 2177 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 2178 m = lacp_input(lp, m); 2179 if (m == NULL) 2180 return (NULL); 2181 } 2182 2183 /* 2184 * If the port is not collecting or not in the active aggregator then 2185 * free and return. 2186 */ 2187 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 2188 m_freem(m); 2189 return (NULL); 2190 } 2191 2192 m->m_pkthdr.rcvif = ifp; 2193 return (m); 2194 } 2195 2196