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_BROADCAST, 225 .pr_start = lagg_bcast_start, 226 .pr_input = lagg_bcast_input, 227 }, 228 }; 229 230 SYSCTL_DECL(_net_link); 231 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 232 "Link Aggregation"); 233 234 /* Allow input on any failover links */ 235 static VNET_DEFINE(int, lagg_failover_rx_all); 236 #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all) 237 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET, 238 &VNET_NAME(lagg_failover_rx_all), 0, 239 "Accept input from any interface in a failover lagg"); 240 241 /* Default value for using flowid */ 242 static VNET_DEFINE(int, def_use_flowid) = 1; 243 #define V_def_use_flowid VNET(def_use_flowid) 244 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN, 245 &VNET_NAME(def_use_flowid), 0, 246 "Default setting for using flow id for load sharing"); 247 248 /* Default value for flowid shift */ 249 static VNET_DEFINE(int, def_flowid_shift) = 16; 250 #define V_def_flowid_shift VNET(def_flowid_shift) 251 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN, 252 &VNET_NAME(def_flowid_shift), 0, 253 "Default setting for flowid shift for load sharing"); 254 255 static void 256 vnet_lagg_init(const void *unused __unused) 257 { 258 259 LAGG_LIST_LOCK_INIT(); 260 SLIST_INIT(&V_lagg_list); 261 V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create, 262 lagg_clone_destroy, 0); 263 } 264 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 265 vnet_lagg_init, NULL); 266 267 static void 268 vnet_lagg_uninit(const void *unused __unused) 269 { 270 271 if_clone_detach(V_lagg_cloner); 272 LAGG_LIST_LOCK_DESTROY(); 273 } 274 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 275 vnet_lagg_uninit, NULL); 276 277 static int 278 lagg_modevent(module_t mod, int type, void *data) 279 { 280 281 switch (type) { 282 case MOD_LOAD: 283 lagg_input_p = lagg_input; 284 lagg_linkstate_p = lagg_port_state; 285 lagg_detach_cookie = EVENTHANDLER_REGISTER( 286 ifnet_departure_event, lagg_port_ifdetach, NULL, 287 EVENTHANDLER_PRI_ANY); 288 break; 289 case MOD_UNLOAD: 290 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 291 lagg_detach_cookie); 292 lagg_input_p = NULL; 293 lagg_linkstate_p = NULL; 294 break; 295 default: 296 return (EOPNOTSUPP); 297 } 298 return (0); 299 } 300 301 static moduledata_t lagg_mod = { 302 "if_lagg", 303 lagg_modevent, 304 0 305 }; 306 307 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 308 MODULE_VERSION(if_lagg, 1); 309 310 static void 311 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr) 312 { 313 314 KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto", 315 __func__, sc)); 316 317 if (sc->sc_ifflags & IFF_DEBUG) 318 if_printf(sc->sc_ifp, "using proto %u\n", pr); 319 320 if (lagg_protos[pr].pr_attach != NULL) 321 lagg_protos[pr].pr_attach(sc); 322 sc->sc_proto = pr; 323 } 324 325 static void 326 lagg_proto_detach(struct lagg_softc *sc) 327 { 328 lagg_proto pr; 329 330 LAGG_WLOCK_ASSERT(sc); 331 332 pr = sc->sc_proto; 333 sc->sc_proto = LAGG_PROTO_NONE; 334 335 if (lagg_protos[pr].pr_detach != NULL) 336 lagg_protos[pr].pr_detach(sc); 337 else 338 LAGG_WUNLOCK(sc); 339 } 340 341 static int 342 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m) 343 { 344 345 return (lagg_protos[sc->sc_proto].pr_start(sc, m)); 346 } 347 348 static struct mbuf * 349 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 350 { 351 352 return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m)); 353 } 354 355 static int 356 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp) 357 { 358 359 if (lagg_protos[sc->sc_proto].pr_addport == NULL) 360 return (0); 361 else 362 return (lagg_protos[sc->sc_proto].pr_addport(lp)); 363 } 364 365 static void 366 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp) 367 { 368 369 if (lagg_protos[sc->sc_proto].pr_delport != NULL) 370 lagg_protos[sc->sc_proto].pr_delport(lp); 371 } 372 373 static void 374 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp) 375 { 376 377 if (lagg_protos[sc->sc_proto].pr_linkstate != NULL) 378 lagg_protos[sc->sc_proto].pr_linkstate(lp); 379 } 380 381 static void 382 lagg_proto_init(struct lagg_softc *sc) 383 { 384 385 if (lagg_protos[sc->sc_proto].pr_init != NULL) 386 lagg_protos[sc->sc_proto].pr_init(sc); 387 } 388 389 static void 390 lagg_proto_stop(struct lagg_softc *sc) 391 { 392 393 if (lagg_protos[sc->sc_proto].pr_stop != NULL) 394 lagg_protos[sc->sc_proto].pr_stop(sc); 395 } 396 397 static void 398 lagg_proto_lladdr(struct lagg_softc *sc) 399 { 400 401 if (lagg_protos[sc->sc_proto].pr_lladdr != NULL) 402 lagg_protos[sc->sc_proto].pr_lladdr(sc); 403 } 404 405 static void 406 lagg_proto_request(struct lagg_softc *sc, void *v) 407 { 408 409 if (lagg_protos[sc->sc_proto].pr_request != NULL) 410 lagg_protos[sc->sc_proto].pr_request(sc, v); 411 } 412 413 static void 414 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v) 415 { 416 417 if (lagg_protos[sc->sc_proto].pr_portreq != NULL) 418 lagg_protos[sc->sc_proto].pr_portreq(lp, v); 419 } 420 421 /* 422 * This routine is run via an vlan 423 * config EVENT 424 */ 425 static void 426 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 427 { 428 struct lagg_softc *sc = ifp->if_softc; 429 struct lagg_port *lp; 430 struct rm_priotracker tracker; 431 432 if (ifp->if_softc != arg) /* Not our event */ 433 return; 434 435 LAGG_RLOCK(sc, &tracker); 436 if (!SLIST_EMPTY(&sc->sc_ports)) { 437 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 438 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 439 } 440 LAGG_RUNLOCK(sc, &tracker); 441 } 442 443 /* 444 * This routine is run via an vlan 445 * unconfig EVENT 446 */ 447 static void 448 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 449 { 450 struct lagg_softc *sc = ifp->if_softc; 451 struct lagg_port *lp; 452 struct rm_priotracker tracker; 453 454 if (ifp->if_softc != arg) /* Not our event */ 455 return; 456 457 LAGG_RLOCK(sc, &tracker); 458 if (!SLIST_EMPTY(&sc->sc_ports)) { 459 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 460 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 461 } 462 LAGG_RUNLOCK(sc, &tracker); 463 } 464 465 static int 466 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) 467 { 468 struct lagg_softc *sc; 469 struct ifnet *ifp; 470 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ 471 472 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO); 473 ifp = sc->sc_ifp = if_alloc(IFT_ETHER); 474 if (ifp == NULL) { 475 free(sc, M_DEVBUF); 476 return (ENOSPC); 477 } 478 479 if (V_def_use_flowid) 480 sc->sc_opts |= LAGG_OPT_USE_FLOWID; 481 sc->flowid_shift = V_def_flowid_shift; 482 483 /* Hash all layers by default */ 484 sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4; 485 486 lagg_proto_attach(sc, LAGG_PROTO_DEFAULT); 487 488 LAGG_LOCK_INIT(sc); 489 SLIST_INIT(&sc->sc_ports); 490 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc); 491 492 /* Initialise pseudo media types */ 493 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 494 lagg_media_status); 495 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 496 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 497 498 if_initname(ifp, laggname, unit); 499 ifp->if_softc = sc; 500 ifp->if_transmit = lagg_transmit; 501 ifp->if_qflush = lagg_qflush; 502 ifp->if_init = lagg_init; 503 ifp->if_ioctl = lagg_ioctl; 504 ifp->if_get_counter = lagg_get_counter; 505 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 506 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS; 507 508 /* 509 * Attach as an ordinary ethernet device, children will be attached 510 * as special device IFT_IEEE8023ADLAG. 511 */ 512 ether_ifattach(ifp, eaddr); 513 514 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 515 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 516 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 517 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 518 519 /* Insert into the global list of laggs */ 520 LAGG_LIST_LOCK(); 521 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries); 522 LAGG_LIST_UNLOCK(); 523 524 return (0); 525 } 526 527 static void 528 lagg_clone_destroy(struct ifnet *ifp) 529 { 530 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 531 struct lagg_port *lp; 532 533 LAGG_WLOCK(sc); 534 535 lagg_stop(sc); 536 ifp->if_flags &= ~IFF_UP; 537 538 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 539 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 540 541 /* Shutdown and remove lagg ports */ 542 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) 543 lagg_port_destroy(lp, 1); 544 /* Unhook the aggregation protocol */ 545 lagg_proto_detach(sc); 546 547 ifmedia_removeall(&sc->sc_media); 548 ether_ifdetach(ifp); 549 if_free(ifp); 550 551 LAGG_LIST_LOCK(); 552 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries); 553 LAGG_LIST_UNLOCK(); 554 555 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task); 556 LAGG_LOCK_DESTROY(sc); 557 free(sc, M_DEVBUF); 558 } 559 560 static void 561 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr) 562 { 563 struct ifnet *ifp = sc->sc_ifp; 564 struct lagg_port lp; 565 566 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 567 return; 568 569 LAGG_WLOCK_ASSERT(sc); 570 /* 571 * Set the link layer address on the lagg interface. 572 * lagg_proto_lladdr() notifies the MAC change to 573 * the aggregation protocol. iflladdr_event handler which 574 * may trigger gratuitous ARPs for INET will be handled in 575 * a taskqueue. 576 */ 577 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN); 578 lagg_proto_lladdr(sc); 579 580 bzero(&lp, sizeof(lp)); 581 lp.lp_ifp = sc->sc_ifp; 582 lp.lp_softc = sc; 583 584 lagg_port_lladdr(&lp, lladdr); 585 } 586 587 static void 588 lagg_capabilities(struct lagg_softc *sc) 589 { 590 struct lagg_port *lp; 591 int cap = ~0, ena = ~0; 592 u_long hwa = ~0UL; 593 struct ifnet_hw_tsomax hw_tsomax; 594 595 LAGG_WLOCK_ASSERT(sc); 596 597 memset(&hw_tsomax, 0, sizeof(hw_tsomax)); 598 599 /* Get capabilities from the lagg ports */ 600 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 601 cap &= lp->lp_ifp->if_capabilities; 602 ena &= lp->lp_ifp->if_capenable; 603 hwa &= lp->lp_ifp->if_hwassist; 604 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax); 605 } 606 cap = (cap == ~0 ? 0 : cap); 607 ena = (ena == ~0 ? 0 : ena); 608 hwa = (hwa == ~0 ? 0 : hwa); 609 610 if (sc->sc_ifp->if_capabilities != cap || 611 sc->sc_ifp->if_capenable != ena || 612 sc->sc_ifp->if_hwassist != hwa || 613 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) { 614 sc->sc_ifp->if_capabilities = cap; 615 sc->sc_ifp->if_capenable = ena; 616 sc->sc_ifp->if_hwassist = hwa; 617 getmicrotime(&sc->sc_ifp->if_lastchange); 618 619 if (sc->sc_ifflags & IFF_DEBUG) 620 if_printf(sc->sc_ifp, 621 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 622 } 623 } 624 625 static void 626 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr) 627 { 628 struct lagg_softc *sc = lp->lp_softc; 629 struct ifnet *ifp = lp->lp_ifp; 630 struct lagg_llq *llq; 631 int pending = 0; 632 int primary; 633 634 LAGG_WLOCK_ASSERT(sc); 635 636 primary = (sc->sc_primary->lp_ifp == ifp) ? 1 : 0; 637 if (primary == 0 && (lp->lp_detaching || 638 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)) 639 return; 640 641 /* Check to make sure its not already queued to be changed */ 642 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 643 if (llq->llq_ifp == ifp && llq->llq_primary == primary) { 644 pending = 1; 645 break; 646 } 647 } 648 649 if (!pending) { 650 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT); 651 if (llq == NULL) /* XXX what to do */ 652 return; 653 } 654 655 /* Update the lladdr even if pending, it may have changed */ 656 llq->llq_ifp = ifp; 657 llq->llq_primary = primary; 658 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN); 659 660 if (!pending) 661 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries); 662 663 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task); 664 } 665 666 /* 667 * Set the interface MAC address from a taskqueue to avoid a LOR. 668 */ 669 static void 670 lagg_port_setlladdr(void *arg, int pending) 671 { 672 struct lagg_softc *sc = (struct lagg_softc *)arg; 673 struct lagg_llq *llq, *head; 674 struct ifnet *ifp; 675 int error; 676 677 /* Grab a local reference of the queue and remove it from the softc */ 678 LAGG_WLOCK(sc); 679 head = SLIST_FIRST(&sc->sc_llq_head); 680 SLIST_FIRST(&sc->sc_llq_head) = NULL; 681 LAGG_WUNLOCK(sc); 682 683 /* 684 * Traverse the queue and set the lladdr on each ifp. It is safe to do 685 * unlocked as we have the only reference to it. 686 */ 687 for (llq = head; llq != NULL; llq = head) { 688 ifp = llq->llq_ifp; 689 690 CURVNET_SET(ifp->if_vnet); 691 if (llq->llq_primary == 0) { 692 /* 693 * Set the link layer address on the laggport interface. 694 * if_setlladdr() triggers gratuitous ARPs for INET. 695 */ 696 error = if_setlladdr(ifp, llq->llq_lladdr, 697 ETHER_ADDR_LEN); 698 if (error) 699 printf("%s: setlladdr failed on %s\n", __func__, 700 ifp->if_xname); 701 } else 702 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 703 CURVNET_RESTORE(); 704 head = SLIST_NEXT(llq, llq_entries); 705 free(llq, M_DEVBUF); 706 } 707 } 708 709 static int 710 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 711 { 712 struct lagg_softc *sc_ptr; 713 struct lagg_port *lp, *tlp; 714 int error, i; 715 uint64_t *pval; 716 717 LAGG_WLOCK_ASSERT(sc); 718 719 /* Limit the maximal number of lagg ports */ 720 if (sc->sc_count >= LAGG_MAX_PORTS) 721 return (ENOSPC); 722 723 /* Check if port has already been associated to a lagg */ 724 if (ifp->if_lagg != NULL) { 725 /* Port is already in the current lagg? */ 726 lp = (struct lagg_port *)ifp->if_lagg; 727 if (lp->lp_softc == sc) 728 return (EEXIST); 729 return (EBUSY); 730 } 731 732 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 733 if (ifp->if_type != IFT_ETHER) 734 return (EPROTONOSUPPORT); 735 736 /* Allow the first Ethernet member to define the MTU */ 737 if (SLIST_EMPTY(&sc->sc_ports)) 738 sc->sc_ifp->if_mtu = ifp->if_mtu; 739 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 740 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 741 ifp->if_xname); 742 return (EINVAL); 743 } 744 745 if ((lp = malloc(sizeof(struct lagg_port), 746 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 747 return (ENOMEM); 748 749 /* Check if port is a stacked lagg */ 750 LAGG_LIST_LOCK(); 751 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) { 752 if (ifp == sc_ptr->sc_ifp) { 753 LAGG_LIST_UNLOCK(); 754 free(lp, M_DEVBUF); 755 return (EINVAL); 756 /* XXX disable stacking for the moment, its untested */ 757 #ifdef LAGG_PORT_STACKING 758 lp->lp_flags |= LAGG_PORT_STACK; 759 if (lagg_port_checkstacking(sc_ptr) >= 760 LAGG_MAX_STACKING) { 761 LAGG_LIST_UNLOCK(); 762 free(lp, M_DEVBUF); 763 return (E2BIG); 764 } 765 #endif 766 } 767 } 768 LAGG_LIST_UNLOCK(); 769 770 /* Change the interface type */ 771 lp->lp_iftype = ifp->if_type; 772 ifp->if_type = IFT_IEEE8023ADLAG; 773 ifp->if_lagg = lp; 774 lp->lp_ioctl = ifp->if_ioctl; 775 ifp->if_ioctl = lagg_port_ioctl; 776 lp->lp_output = ifp->if_output; 777 ifp->if_output = lagg_port_output; 778 779 lp->lp_ifp = ifp; 780 lp->lp_softc = sc; 781 782 /* Save port link layer address */ 783 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 784 785 if (SLIST_EMPTY(&sc->sc_ports)) { 786 sc->sc_primary = lp; 787 lagg_lladdr(sc, IF_LLADDR(ifp)); 788 } else { 789 /* Update link layer address for this port */ 790 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp)); 791 } 792 793 /* 794 * Insert into the list of ports. 795 * Keep ports sorted by if_index. It is handy, when configuration 796 * is predictable and `ifconfig laggN create ...` command 797 * will lead to the same result each time. 798 */ 799 SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) { 800 if (tlp->lp_ifp->if_index < ifp->if_index && ( 801 SLIST_NEXT(tlp, lp_entries) == NULL || 802 SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index > 803 ifp->if_index)) 804 break; 805 } 806 if (tlp != NULL) 807 SLIST_INSERT_AFTER(tlp, lp, lp_entries); 808 else 809 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 810 sc->sc_count++; 811 812 /* Update lagg capabilities */ 813 lagg_capabilities(sc); 814 lagg_linkstate(sc); 815 816 /* Read port counters */ 817 pval = lp->port_counters.val; 818 for (i = 0; i < IFCOUNTERS; i++, pval++) 819 *pval = ifp->if_get_counter(ifp, i); 820 /* Add multicast addresses and interface flags to this port */ 821 lagg_ether_cmdmulti(lp, 1); 822 lagg_setflags(lp, 1); 823 824 if ((error = lagg_proto_addport(sc, lp)) != 0) { 825 /* Remove the port, without calling pr_delport. */ 826 lagg_port_destroy(lp, 0); 827 return (error); 828 } 829 830 return (0); 831 } 832 833 #ifdef LAGG_PORT_STACKING 834 static int 835 lagg_port_checkstacking(struct lagg_softc *sc) 836 { 837 struct lagg_softc *sc_ptr; 838 struct lagg_port *lp; 839 int m = 0; 840 841 LAGG_WLOCK_ASSERT(sc); 842 843 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 844 if (lp->lp_flags & LAGG_PORT_STACK) { 845 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 846 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 847 } 848 } 849 850 return (m + 1); 851 } 852 #endif 853 854 static int 855 lagg_port_destroy(struct lagg_port *lp, int rundelport) 856 { 857 struct lagg_softc *sc = lp->lp_softc; 858 struct lagg_port *lp_ptr, *lp0; 859 struct lagg_llq *llq; 860 struct ifnet *ifp = lp->lp_ifp; 861 uint64_t *pval, vdiff; 862 int i; 863 864 LAGG_WLOCK_ASSERT(sc); 865 866 if (rundelport) 867 lagg_proto_delport(sc, lp); 868 869 /* 870 * Remove multicast addresses and interface flags from this port and 871 * reset the MAC address, skip if the interface is being detached. 872 */ 873 if (!lp->lp_detaching) { 874 lagg_ether_cmdmulti(lp, 0); 875 lagg_setflags(lp, 0); 876 lagg_port_lladdr(lp, lp->lp_lladdr); 877 } 878 879 /* Restore interface */ 880 ifp->if_type = lp->lp_iftype; 881 ifp->if_ioctl = lp->lp_ioctl; 882 ifp->if_output = lp->lp_output; 883 ifp->if_lagg = NULL; 884 885 /* Update detached port counters */ 886 pval = lp->port_counters.val; 887 for (i = 0; i < IFCOUNTERS; i++, pval++) { 888 vdiff = ifp->if_get_counter(ifp, i) - *pval; 889 sc->detached_counters.val[i] += vdiff; 890 } 891 892 /* Finally, remove the port from the lagg */ 893 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 894 sc->sc_count--; 895 896 /* Update the primary interface */ 897 if (lp == sc->sc_primary) { 898 uint8_t lladdr[ETHER_ADDR_LEN]; 899 900 if ((lp0 = SLIST_FIRST(&sc->sc_ports)) == NULL) { 901 bzero(&lladdr, ETHER_ADDR_LEN); 902 } else { 903 bcopy(lp0->lp_lladdr, 904 lladdr, ETHER_ADDR_LEN); 905 } 906 lagg_lladdr(sc, lladdr); 907 908 /* 909 * Update link layer address for each port. No port is 910 * marked as primary at this moment. 911 */ 912 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 913 lagg_port_lladdr(lp_ptr, lladdr); 914 /* 915 * Mark lp0 as the new primary. This invokes an 916 * iflladdr_event. 917 */ 918 sc->sc_primary = lp0; 919 if (lp0 != NULL) 920 lagg_port_lladdr(lp0, lladdr); 921 } 922 923 /* Remove any pending lladdr changes from the queue */ 924 if (lp->lp_detaching) { 925 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 926 if (llq->llq_ifp == ifp) { 927 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq, 928 llq_entries); 929 free(llq, M_DEVBUF); 930 break; /* Only appears once */ 931 } 932 } 933 } 934 935 if (lp->lp_ifflags) 936 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 937 938 free(lp, M_DEVBUF); 939 940 /* Update lagg capabilities */ 941 lagg_capabilities(sc); 942 lagg_linkstate(sc); 943 944 return (0); 945 } 946 947 static int 948 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 949 { 950 struct lagg_reqport *rp = (struct lagg_reqport *)data; 951 struct lagg_softc *sc; 952 struct lagg_port *lp = NULL; 953 int error = 0; 954 struct rm_priotracker tracker; 955 956 /* Should be checked by the caller */ 957 if (ifp->if_type != IFT_IEEE8023ADLAG || 958 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 959 goto fallback; 960 961 switch (cmd) { 962 case SIOCGLAGGPORT: 963 if (rp->rp_portname[0] == '\0' || 964 ifunit(rp->rp_portname) != ifp) { 965 error = EINVAL; 966 break; 967 } 968 969 LAGG_RLOCK(sc, &tracker); 970 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 971 error = ENOENT; 972 LAGG_RUNLOCK(sc, &tracker); 973 break; 974 } 975 976 lagg_port2req(lp, rp); 977 LAGG_RUNLOCK(sc, &tracker); 978 break; 979 980 case SIOCSIFCAP: 981 if (lp->lp_ioctl == NULL) { 982 error = EINVAL; 983 break; 984 } 985 error = (*lp->lp_ioctl)(ifp, cmd, data); 986 if (error) 987 break; 988 989 /* Update lagg interface capabilities */ 990 LAGG_WLOCK(sc); 991 lagg_capabilities(sc); 992 LAGG_WUNLOCK(sc); 993 break; 994 995 case SIOCSIFMTU: 996 /* Do not allow the MTU to be changed once joined */ 997 error = EINVAL; 998 break; 999 1000 default: 1001 goto fallback; 1002 } 1003 1004 return (error); 1005 1006 fallback: 1007 if (lp->lp_ioctl != NULL) 1008 return ((*lp->lp_ioctl)(ifp, cmd, data)); 1009 1010 return (EINVAL); 1011 } 1012 1013 /* 1014 * Requests counter @cnt data. 1015 * 1016 * Counter value is calculated the following way: 1017 * 1) for each port, sum difference between current and "initial" measurements. 1018 * 2) add lagg logical interface counters. 1019 * 3) add data from detached_counters array. 1020 * 1021 * We also do the following things on ports attach/detach: 1022 * 1) On port attach we store all counters it has into port_counter array. 1023 * 2) On port detach we add the different between "initial" and 1024 * current counters data to detached_counters array. 1025 */ 1026 static uint64_t 1027 lagg_get_counter(struct ifnet *ifp, ift_counter cnt) 1028 { 1029 struct lagg_softc *sc; 1030 struct lagg_port *lp; 1031 struct ifnet *lpifp; 1032 struct rm_priotracker tracker; 1033 uint64_t newval, oldval, vsum; 1034 1035 /* Revise this when we've got non-generic counters. */ 1036 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt)); 1037 1038 sc = (struct lagg_softc *)ifp->if_softc; 1039 LAGG_RLOCK(sc, &tracker); 1040 1041 vsum = 0; 1042 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1043 /* Saved attached value */ 1044 oldval = lp->port_counters.val[cnt]; 1045 /* current value */ 1046 lpifp = lp->lp_ifp; 1047 newval = lpifp->if_get_counter(lpifp, cnt); 1048 /* Calculate diff and save new */ 1049 vsum += newval - oldval; 1050 } 1051 1052 /* 1053 * Add counter data which might be added by upper 1054 * layer protocols operating on logical interface. 1055 */ 1056 vsum += if_get_counter_default(ifp, cnt); 1057 1058 /* 1059 * Add counter data from detached ports counters 1060 */ 1061 vsum += sc->detached_counters.val[cnt]; 1062 1063 LAGG_RUNLOCK(sc, &tracker); 1064 1065 return (vsum); 1066 } 1067 1068 /* 1069 * For direct output to child ports. 1070 */ 1071 static int 1072 lagg_port_output(struct ifnet *ifp, struct mbuf *m, 1073 const struct sockaddr *dst, struct route *ro) 1074 { 1075 struct lagg_port *lp = ifp->if_lagg; 1076 1077 switch (dst->sa_family) { 1078 case pseudo_AF_HDRCMPLT: 1079 case AF_UNSPEC: 1080 return ((*lp->lp_output)(ifp, m, dst, ro)); 1081 } 1082 1083 /* drop any other frames */ 1084 m_freem(m); 1085 return (ENETDOWN); 1086 } 1087 1088 static void 1089 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 1090 { 1091 struct lagg_port *lp; 1092 struct lagg_softc *sc; 1093 1094 if ((lp = ifp->if_lagg) == NULL) 1095 return; 1096 /* If the ifnet is just being renamed, don't do anything. */ 1097 if (ifp->if_flags & IFF_RENAMING) 1098 return; 1099 1100 sc = lp->lp_softc; 1101 1102 LAGG_WLOCK(sc); 1103 lp->lp_detaching = 1; 1104 lagg_port_destroy(lp, 1); 1105 LAGG_WUNLOCK(sc); 1106 } 1107 1108 static void 1109 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 1110 { 1111 struct lagg_softc *sc = lp->lp_softc; 1112 1113 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 1114 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 1115 rp->rp_prio = lp->lp_prio; 1116 rp->rp_flags = lp->lp_flags; 1117 lagg_proto_portreq(sc, lp, &rp->rp_psc); 1118 1119 /* Add protocol specific flags */ 1120 switch (sc->sc_proto) { 1121 case LAGG_PROTO_FAILOVER: 1122 if (lp == sc->sc_primary) 1123 rp->rp_flags |= LAGG_PORT_MASTER; 1124 if (lp == lagg_link_active(sc, sc->sc_primary)) 1125 rp->rp_flags |= LAGG_PORT_ACTIVE; 1126 break; 1127 1128 case LAGG_PROTO_ROUNDROBIN: 1129 case LAGG_PROTO_LOADBALANCE: 1130 case LAGG_PROTO_BROADCAST: 1131 if (LAGG_PORTACTIVE(lp)) 1132 rp->rp_flags |= LAGG_PORT_ACTIVE; 1133 break; 1134 1135 case LAGG_PROTO_LACP: 1136 /* LACP has a different definition of active */ 1137 if (lacp_isactive(lp)) 1138 rp->rp_flags |= LAGG_PORT_ACTIVE; 1139 if (lacp_iscollecting(lp)) 1140 rp->rp_flags |= LAGG_PORT_COLLECTING; 1141 if (lacp_isdistributing(lp)) 1142 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 1143 break; 1144 } 1145 1146 } 1147 1148 static void 1149 lagg_init(void *xsc) 1150 { 1151 struct lagg_softc *sc = (struct lagg_softc *)xsc; 1152 struct lagg_port *lp; 1153 struct ifnet *ifp = sc->sc_ifp; 1154 1155 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1156 return; 1157 1158 LAGG_WLOCK(sc); 1159 1160 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1161 /* Update the port lladdrs */ 1162 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1163 lagg_port_lladdr(lp, IF_LLADDR(ifp)); 1164 1165 lagg_proto_init(sc); 1166 1167 LAGG_WUNLOCK(sc); 1168 } 1169 1170 static void 1171 lagg_stop(struct lagg_softc *sc) 1172 { 1173 struct ifnet *ifp = sc->sc_ifp; 1174 1175 LAGG_WLOCK_ASSERT(sc); 1176 1177 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 1178 return; 1179 1180 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1181 1182 lagg_proto_stop(sc); 1183 } 1184 1185 static int 1186 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1187 { 1188 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1189 struct lagg_reqall *ra = (struct lagg_reqall *)data; 1190 struct lagg_reqopts *ro = (struct lagg_reqopts *)data; 1191 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 1192 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 1193 struct ifreq *ifr = (struct ifreq *)data; 1194 struct lagg_port *lp; 1195 struct ifnet *tpif; 1196 struct thread *td = curthread; 1197 char *buf, *outbuf; 1198 int count, buflen, len, error = 0; 1199 struct rm_priotracker tracker; 1200 1201 bzero(&rpbuf, sizeof(rpbuf)); 1202 1203 switch (cmd) { 1204 case SIOCGLAGG: 1205 LAGG_RLOCK(sc, &tracker); 1206 count = 0; 1207 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1208 count++; 1209 buflen = count * sizeof(struct lagg_reqport); 1210 LAGG_RUNLOCK(sc, &tracker); 1211 1212 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 1213 1214 LAGG_RLOCK(sc, &tracker); 1215 ra->ra_proto = sc->sc_proto; 1216 lagg_proto_request(sc, &ra->ra_psc); 1217 count = 0; 1218 buf = outbuf; 1219 len = min(ra->ra_size, buflen); 1220 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1221 if (len < sizeof(rpbuf)) 1222 break; 1223 1224 lagg_port2req(lp, &rpbuf); 1225 memcpy(buf, &rpbuf, sizeof(rpbuf)); 1226 count++; 1227 buf += sizeof(rpbuf); 1228 len -= sizeof(rpbuf); 1229 } 1230 LAGG_RUNLOCK(sc, &tracker); 1231 ra->ra_ports = count; 1232 ra->ra_size = count * sizeof(rpbuf); 1233 error = copyout(outbuf, ra->ra_port, ra->ra_size); 1234 free(outbuf, M_TEMP); 1235 break; 1236 case SIOCSLAGG: 1237 error = priv_check(td, PRIV_NET_LAGG); 1238 if (error) 1239 break; 1240 if (ra->ra_proto < 1 || ra->ra_proto >= LAGG_PROTO_MAX) { 1241 error = EPROTONOSUPPORT; 1242 break; 1243 } 1244 1245 LAGG_WLOCK(sc); 1246 lagg_proto_detach(sc); 1247 lagg_proto_attach(sc, ra->ra_proto); 1248 break; 1249 case SIOCGLAGGOPTS: 1250 ro->ro_opts = sc->sc_opts; 1251 if (sc->sc_proto == LAGG_PROTO_LACP) { 1252 struct lacp_softc *lsc; 1253 1254 lsc = (struct lacp_softc *)sc->sc_psc; 1255 if (lsc->lsc_debug.lsc_tx_test != 0) 1256 ro->ro_opts |= LAGG_OPT_LACP_TXTEST; 1257 if (lsc->lsc_debug.lsc_rx_test != 0) 1258 ro->ro_opts |= LAGG_OPT_LACP_RXTEST; 1259 if (lsc->lsc_strict_mode != 0) 1260 ro->ro_opts |= LAGG_OPT_LACP_STRICT; 1261 if (lsc->lsc_fast_timeout != 0) 1262 ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT; 1263 1264 ro->ro_active = sc->sc_active; 1265 } else { 1266 ro->ro_active = 0; 1267 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1268 ro->ro_active += LAGG_PORTACTIVE(lp); 1269 } 1270 ro->ro_flapping = sc->sc_flapping; 1271 ro->ro_flowid_shift = sc->flowid_shift; 1272 break; 1273 case SIOCSLAGGOPTS: 1274 error = priv_check(td, PRIV_NET_LAGG); 1275 if (error) 1276 break; 1277 if (ro->ro_opts == 0) 1278 break; 1279 /* 1280 * Set options. LACP options are stored in sc->sc_psc, 1281 * not in sc_opts. 1282 */ 1283 int valid, lacp; 1284 1285 switch (ro->ro_opts) { 1286 case LAGG_OPT_USE_FLOWID: 1287 case -LAGG_OPT_USE_FLOWID: 1288 case LAGG_OPT_FLOWIDSHIFT: 1289 valid = 1; 1290 lacp = 0; 1291 break; 1292 case LAGG_OPT_LACP_TXTEST: 1293 case -LAGG_OPT_LACP_TXTEST: 1294 case LAGG_OPT_LACP_RXTEST: 1295 case -LAGG_OPT_LACP_RXTEST: 1296 case LAGG_OPT_LACP_STRICT: 1297 case -LAGG_OPT_LACP_STRICT: 1298 case LAGG_OPT_LACP_TIMEOUT: 1299 case -LAGG_OPT_LACP_TIMEOUT: 1300 valid = lacp = 1; 1301 break; 1302 default: 1303 valid = lacp = 0; 1304 break; 1305 } 1306 1307 LAGG_WLOCK(sc); 1308 if (valid == 0 || 1309 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) { 1310 /* Invalid combination of options specified. */ 1311 error = EINVAL; 1312 LAGG_WUNLOCK(sc); 1313 break; /* Return from SIOCSLAGGOPTS. */ 1314 } 1315 /* 1316 * Store new options into sc->sc_opts except for 1317 * FLOWIDSHIFT and LACP options. 1318 */ 1319 if (lacp == 0) { 1320 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT) 1321 sc->flowid_shift = ro->ro_flowid_shift; 1322 else if (ro->ro_opts > 0) 1323 sc->sc_opts |= ro->ro_opts; 1324 else 1325 sc->sc_opts &= ~ro->ro_opts; 1326 } else { 1327 struct lacp_softc *lsc; 1328 struct lacp_port *lp; 1329 1330 lsc = (struct lacp_softc *)sc->sc_psc; 1331 1332 switch (ro->ro_opts) { 1333 case LAGG_OPT_LACP_TXTEST: 1334 lsc->lsc_debug.lsc_tx_test = 1; 1335 break; 1336 case -LAGG_OPT_LACP_TXTEST: 1337 lsc->lsc_debug.lsc_tx_test = 0; 1338 break; 1339 case LAGG_OPT_LACP_RXTEST: 1340 lsc->lsc_debug.lsc_rx_test = 1; 1341 break; 1342 case -LAGG_OPT_LACP_RXTEST: 1343 lsc->lsc_debug.lsc_rx_test = 0; 1344 break; 1345 case LAGG_OPT_LACP_STRICT: 1346 lsc->lsc_strict_mode = 1; 1347 break; 1348 case -LAGG_OPT_LACP_STRICT: 1349 lsc->lsc_strict_mode = 0; 1350 break; 1351 case LAGG_OPT_LACP_TIMEOUT: 1352 LACP_LOCK(lsc); 1353 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1354 lp->lp_state |= LACP_STATE_TIMEOUT; 1355 LACP_UNLOCK(lsc); 1356 lsc->lsc_fast_timeout = 1; 1357 break; 1358 case -LAGG_OPT_LACP_TIMEOUT: 1359 LACP_LOCK(lsc); 1360 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1361 lp->lp_state &= ~LACP_STATE_TIMEOUT; 1362 LACP_UNLOCK(lsc); 1363 lsc->lsc_fast_timeout = 0; 1364 break; 1365 } 1366 } 1367 LAGG_WUNLOCK(sc); 1368 break; 1369 case SIOCGLAGGFLAGS: 1370 rf->rf_flags = 0; 1371 LAGG_RLOCK(sc, &tracker); 1372 if (sc->sc_flags & MBUF_HASHFLAG_L2) 1373 rf->rf_flags |= LAGG_F_HASHL2; 1374 if (sc->sc_flags & MBUF_HASHFLAG_L3) 1375 rf->rf_flags |= LAGG_F_HASHL3; 1376 if (sc->sc_flags & MBUF_HASHFLAG_L4) 1377 rf->rf_flags |= LAGG_F_HASHL4; 1378 LAGG_RUNLOCK(sc, &tracker); 1379 break; 1380 case SIOCSLAGGHASH: 1381 error = priv_check(td, PRIV_NET_LAGG); 1382 if (error) 1383 break; 1384 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1385 error = EINVAL; 1386 break; 1387 } 1388 LAGG_WLOCK(sc); 1389 sc->sc_flags = 0; 1390 if (rf->rf_flags & LAGG_F_HASHL2) 1391 sc->sc_flags |= MBUF_HASHFLAG_L2; 1392 if (rf->rf_flags & LAGG_F_HASHL3) 1393 sc->sc_flags |= MBUF_HASHFLAG_L3; 1394 if (rf->rf_flags & LAGG_F_HASHL4) 1395 sc->sc_flags |= MBUF_HASHFLAG_L4; 1396 LAGG_WUNLOCK(sc); 1397 break; 1398 case SIOCGLAGGPORT: 1399 if (rp->rp_portname[0] == '\0' || 1400 (tpif = ifunit(rp->rp_portname)) == NULL) { 1401 error = EINVAL; 1402 break; 1403 } 1404 1405 LAGG_RLOCK(sc, &tracker); 1406 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1407 lp->lp_softc != sc) { 1408 error = ENOENT; 1409 LAGG_RUNLOCK(sc, &tracker); 1410 break; 1411 } 1412 1413 lagg_port2req(lp, rp); 1414 LAGG_RUNLOCK(sc, &tracker); 1415 break; 1416 case SIOCSLAGGPORT: 1417 error = priv_check(td, PRIV_NET_LAGG); 1418 if (error) 1419 break; 1420 if (rp->rp_portname[0] == '\0' || 1421 (tpif = ifunit(rp->rp_portname)) == NULL) { 1422 error = EINVAL; 1423 break; 1424 } 1425 #ifdef INET6 1426 /* 1427 * A laggport interface should not have inet6 address 1428 * because two interfaces with a valid link-local 1429 * scope zone must not be merged in any form. This 1430 * restriction is needed to prevent violation of 1431 * link-local scope zone. Attempts to add a laggport 1432 * interface which has inet6 addresses triggers 1433 * removal of all inet6 addresses on the member 1434 * interface. 1435 */ 1436 if (in6ifa_llaonifp(tpif)) { 1437 in6_ifdetach(tpif); 1438 if_printf(sc->sc_ifp, 1439 "IPv6 addresses on %s have been removed " 1440 "before adding it as a member to prevent " 1441 "IPv6 address scope violation.\n", 1442 tpif->if_xname); 1443 } 1444 #endif 1445 LAGG_WLOCK(sc); 1446 error = lagg_port_create(sc, tpif); 1447 LAGG_WUNLOCK(sc); 1448 break; 1449 case SIOCSLAGGDELPORT: 1450 error = priv_check(td, PRIV_NET_LAGG); 1451 if (error) 1452 break; 1453 if (rp->rp_portname[0] == '\0' || 1454 (tpif = ifunit(rp->rp_portname)) == NULL) { 1455 error = EINVAL; 1456 break; 1457 } 1458 1459 LAGG_WLOCK(sc); 1460 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1461 lp->lp_softc != sc) { 1462 error = ENOENT; 1463 LAGG_WUNLOCK(sc); 1464 break; 1465 } 1466 1467 error = lagg_port_destroy(lp, 1); 1468 LAGG_WUNLOCK(sc); 1469 break; 1470 case SIOCSIFFLAGS: 1471 /* Set flags on ports too */ 1472 LAGG_WLOCK(sc); 1473 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1474 lagg_setflags(lp, 1); 1475 } 1476 LAGG_WUNLOCK(sc); 1477 1478 if (!(ifp->if_flags & IFF_UP) && 1479 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1480 /* 1481 * If interface is marked down and it is running, 1482 * then stop and disable it. 1483 */ 1484 LAGG_WLOCK(sc); 1485 lagg_stop(sc); 1486 LAGG_WUNLOCK(sc); 1487 } else if ((ifp->if_flags & IFF_UP) && 1488 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1489 /* 1490 * If interface is marked up and it is stopped, then 1491 * start it. 1492 */ 1493 (*ifp->if_init)(sc); 1494 } 1495 break; 1496 case SIOCADDMULTI: 1497 case SIOCDELMULTI: 1498 LAGG_WLOCK(sc); 1499 error = lagg_ether_setmulti(sc); 1500 LAGG_WUNLOCK(sc); 1501 break; 1502 case SIOCSIFMEDIA: 1503 case SIOCGIFMEDIA: 1504 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1505 break; 1506 1507 case SIOCSIFCAP: 1508 case SIOCSIFMTU: 1509 /* Do not allow the MTU or caps to be directly changed */ 1510 error = EINVAL; 1511 break; 1512 1513 default: 1514 error = ether_ioctl(ifp, cmd, data); 1515 break; 1516 } 1517 return (error); 1518 } 1519 1520 static int 1521 lagg_ether_setmulti(struct lagg_softc *sc) 1522 { 1523 struct lagg_port *lp; 1524 1525 LAGG_WLOCK_ASSERT(sc); 1526 1527 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1528 /* First, remove any existing filter entries. */ 1529 lagg_ether_cmdmulti(lp, 0); 1530 /* copy all addresses from the lagg interface to the port */ 1531 lagg_ether_cmdmulti(lp, 1); 1532 } 1533 return (0); 1534 } 1535 1536 static int 1537 lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1538 { 1539 struct lagg_softc *sc = lp->lp_softc; 1540 struct ifnet *ifp = lp->lp_ifp; 1541 struct ifnet *scifp = sc->sc_ifp; 1542 struct lagg_mc *mc; 1543 struct ifmultiaddr *ifma; 1544 int error; 1545 1546 LAGG_WLOCK_ASSERT(sc); 1547 1548 if (set) { 1549 IF_ADDR_WLOCK(scifp); 1550 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1551 if (ifma->ifma_addr->sa_family != AF_LINK) 1552 continue; 1553 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1554 if (mc == NULL) { 1555 IF_ADDR_WUNLOCK(scifp); 1556 return (ENOMEM); 1557 } 1558 bcopy(ifma->ifma_addr, &mc->mc_addr, 1559 ifma->ifma_addr->sa_len); 1560 mc->mc_addr.sdl_index = ifp->if_index; 1561 mc->mc_ifma = NULL; 1562 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1563 } 1564 IF_ADDR_WUNLOCK(scifp); 1565 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) { 1566 error = if_addmulti(ifp, 1567 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma); 1568 if (error) 1569 return (error); 1570 } 1571 } else { 1572 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1573 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1574 if (mc->mc_ifma && !lp->lp_detaching) 1575 if_delmulti_ifma(mc->mc_ifma); 1576 free(mc, M_DEVBUF); 1577 } 1578 } 1579 return (0); 1580 } 1581 1582 /* Handle a ref counted flag that should be set on the lagg port as well */ 1583 static int 1584 lagg_setflag(struct lagg_port *lp, int flag, int status, 1585 int (*func)(struct ifnet *, int)) 1586 { 1587 struct lagg_softc *sc = lp->lp_softc; 1588 struct ifnet *scifp = sc->sc_ifp; 1589 struct ifnet *ifp = lp->lp_ifp; 1590 int error; 1591 1592 LAGG_WLOCK_ASSERT(sc); 1593 1594 status = status ? (scifp->if_flags & flag) : 0; 1595 /* Now "status" contains the flag value or 0 */ 1596 1597 /* 1598 * See if recorded ports status is different from what 1599 * we want it to be. If it is, flip it. We record ports 1600 * status in lp_ifflags so that we won't clear ports flag 1601 * we haven't set. In fact, we don't clear or set ports 1602 * flags directly, but get or release references to them. 1603 * That's why we can be sure that recorded flags still are 1604 * in accord with actual ports flags. 1605 */ 1606 if (status != (lp->lp_ifflags & flag)) { 1607 error = (*func)(ifp, status); 1608 if (error) 1609 return (error); 1610 lp->lp_ifflags &= ~flag; 1611 lp->lp_ifflags |= status; 1612 } 1613 return (0); 1614 } 1615 1616 /* 1617 * Handle IFF_* flags that require certain changes on the lagg port 1618 * if "status" is true, update ports flags respective to the lagg 1619 * if "status" is false, forcedly clear the flags set on port. 1620 */ 1621 static int 1622 lagg_setflags(struct lagg_port *lp, int status) 1623 { 1624 int error, i; 1625 1626 for (i = 0; lagg_pflags[i].flag; i++) { 1627 error = lagg_setflag(lp, lagg_pflags[i].flag, 1628 status, lagg_pflags[i].func); 1629 if (error) 1630 return (error); 1631 } 1632 return (0); 1633 } 1634 1635 static int 1636 lagg_transmit(struct ifnet *ifp, struct mbuf *m) 1637 { 1638 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1639 int error, len, mcast; 1640 struct rm_priotracker tracker; 1641 1642 len = m->m_pkthdr.len; 1643 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1644 1645 LAGG_RLOCK(sc, &tracker); 1646 /* We need a Tx algorithm and at least one port */ 1647 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1648 LAGG_RUNLOCK(sc, &tracker); 1649 m_freem(m); 1650 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1651 return (ENXIO); 1652 } 1653 1654 ETHER_BPF_MTAP(ifp, m); 1655 1656 error = lagg_proto_start(sc, m); 1657 LAGG_RUNLOCK(sc, &tracker); 1658 1659 if (error != 0) 1660 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1661 1662 return (error); 1663 } 1664 1665 /* 1666 * The ifp->if_qflush entry point for lagg(4) is no-op. 1667 */ 1668 static void 1669 lagg_qflush(struct ifnet *ifp __unused) 1670 { 1671 } 1672 1673 static struct mbuf * 1674 lagg_input(struct ifnet *ifp, struct mbuf *m) 1675 { 1676 struct lagg_port *lp = ifp->if_lagg; 1677 struct lagg_softc *sc = lp->lp_softc; 1678 struct ifnet *scifp = sc->sc_ifp; 1679 struct rm_priotracker tracker; 1680 1681 LAGG_RLOCK(sc, &tracker); 1682 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1683 (lp->lp_flags & LAGG_PORT_DISABLED) || 1684 sc->sc_proto == LAGG_PROTO_NONE) { 1685 LAGG_RUNLOCK(sc, &tracker); 1686 m_freem(m); 1687 return (NULL); 1688 } 1689 1690 ETHER_BPF_MTAP(scifp, m); 1691 1692 if (lp->lp_detaching != 0) { 1693 m_freem(m); 1694 m = NULL; 1695 } else 1696 m = lagg_proto_input(sc, lp, m); 1697 1698 if (m != NULL) { 1699 if (scifp->if_flags & IFF_MONITOR) { 1700 m_freem(m); 1701 m = NULL; 1702 } 1703 } 1704 1705 LAGG_RUNLOCK(sc, &tracker); 1706 return (m); 1707 } 1708 1709 static int 1710 lagg_media_change(struct ifnet *ifp) 1711 { 1712 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1713 1714 if (sc->sc_ifflags & IFF_DEBUG) 1715 printf("%s\n", __func__); 1716 1717 /* Ignore */ 1718 return (0); 1719 } 1720 1721 static void 1722 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1723 { 1724 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1725 struct lagg_port *lp; 1726 struct rm_priotracker tracker; 1727 1728 imr->ifm_status = IFM_AVALID; 1729 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1730 1731 LAGG_RLOCK(sc, &tracker); 1732 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1733 if (LAGG_PORTACTIVE(lp)) 1734 imr->ifm_status |= IFM_ACTIVE; 1735 } 1736 LAGG_RUNLOCK(sc, &tracker); 1737 } 1738 1739 static void 1740 lagg_linkstate(struct lagg_softc *sc) 1741 { 1742 struct lagg_port *lp; 1743 int new_link = LINK_STATE_DOWN; 1744 uint64_t speed; 1745 1746 /* Our link is considered up if at least one of our ports is active */ 1747 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1748 if (lp->lp_ifp->if_link_state == LINK_STATE_UP) { 1749 new_link = LINK_STATE_UP; 1750 break; 1751 } 1752 } 1753 if_link_state_change(sc->sc_ifp, new_link); 1754 1755 /* Update if_baudrate to reflect the max possible speed */ 1756 switch (sc->sc_proto) { 1757 case LAGG_PROTO_FAILOVER: 1758 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1759 sc->sc_primary->lp_ifp->if_baudrate : 0; 1760 break; 1761 case LAGG_PROTO_ROUNDROBIN: 1762 case LAGG_PROTO_LOADBALANCE: 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