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, 2016 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 #include "opt_ratelimit.h" 27 28 #include <sys/param.h> 29 #include <sys/kernel.h> 30 #include <sys/malloc.h> 31 #include <sys/mbuf.h> 32 #include <sys/queue.h> 33 #include <sys/socket.h> 34 #include <sys/sockio.h> 35 #include <sys/sysctl.h> 36 #include <sys/module.h> 37 #include <sys/priv.h> 38 #include <sys/systm.h> 39 #include <sys/proc.h> 40 #include <sys/lock.h> 41 #include <sys/rmlock.h> 42 #include <sys/sx.h> 43 #include <sys/taskqueue.h> 44 #include <sys/eventhandler.h> 45 46 #include <net/ethernet.h> 47 #include <net/if.h> 48 #include <net/if_clone.h> 49 #include <net/if_arp.h> 50 #include <net/if_dl.h> 51 #include <net/if_media.h> 52 #include <net/if_types.h> 53 #include <net/if_var.h> 54 #include <net/bpf.h> 55 #include <net/vnet.h> 56 57 #if defined(INET) || defined(INET6) 58 #include <netinet/in.h> 59 #include <netinet/ip.h> 60 #endif 61 #ifdef INET 62 #include <netinet/in_systm.h> 63 #include <netinet/if_ether.h> 64 #endif 65 66 #ifdef INET6 67 #include <netinet/ip6.h> 68 #include <netinet6/in6_var.h> 69 #include <netinet6/in6_ifattach.h> 70 #endif 71 72 #include <net/if_vlan_var.h> 73 #include <net/if_lagg.h> 74 #include <net/ieee8023ad_lacp.h> 75 76 /* Special flags we should propagate to the lagg ports. */ 77 static struct { 78 int flag; 79 int (*func)(struct ifnet *, int); 80 } lagg_pflags[] = { 81 {IFF_PROMISC, ifpromisc}, 82 {IFF_ALLMULTI, if_allmulti}, 83 {0, NULL} 84 }; 85 86 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */ 87 #define V_lagg_list VNET(lagg_list) 88 static VNET_DEFINE(struct mtx, lagg_list_mtx); 89 #define V_lagg_list_mtx VNET(lagg_list_mtx) 90 #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \ 91 "if_lagg list", NULL, MTX_DEF) 92 #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx) 93 #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx) 94 #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx) 95 eventhandler_tag lagg_detach_cookie = NULL; 96 97 static int lagg_clone_create(struct if_clone *, int, caddr_t); 98 static void lagg_clone_destroy(struct ifnet *); 99 static VNET_DEFINE(struct if_clone *, lagg_cloner); 100 #define V_lagg_cloner VNET(lagg_cloner) 101 static const char laggname[] = "lagg"; 102 103 static void lagg_capabilities(struct lagg_softc *); 104 static int lagg_port_create(struct lagg_softc *, struct ifnet *); 105 static int lagg_port_destroy(struct lagg_port *, int); 106 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); 107 static void lagg_linkstate(struct lagg_softc *); 108 static void lagg_port_state(struct ifnet *, int); 109 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); 110 static int lagg_port_output(struct ifnet *, struct mbuf *, 111 const struct sockaddr *, struct route *); 112 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 113 #ifdef LAGG_PORT_STACKING 114 static int lagg_port_checkstacking(struct lagg_softc *); 115 #endif 116 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 117 static void lagg_init(void *); 118 static void lagg_stop(struct lagg_softc *); 119 static int lagg_ioctl(struct ifnet *, u_long, caddr_t); 120 #ifdef RATELIMIT 121 static int lagg_snd_tag_alloc(struct ifnet *, 122 union if_snd_tag_alloc_params *, 123 struct m_snd_tag **); 124 #endif 125 static int lagg_setmulti(struct lagg_port *); 126 static int lagg_clrmulti(struct lagg_port *); 127 static int lagg_setcaps(struct lagg_port *, int cap); 128 static int lagg_setflag(struct lagg_port *, int, int, 129 int (*func)(struct ifnet *, int)); 130 static int lagg_setflags(struct lagg_port *, int status); 131 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt); 132 static int lagg_transmit(struct ifnet *, struct mbuf *); 133 static void lagg_qflush(struct ifnet *); 134 static int lagg_media_change(struct ifnet *); 135 static void lagg_media_status(struct ifnet *, struct ifmediareq *); 136 static struct lagg_port *lagg_link_active(struct lagg_softc *, 137 struct lagg_port *); 138 139 /* Simple round robin */ 140 static void lagg_rr_attach(struct lagg_softc *); 141 static int lagg_rr_start(struct lagg_softc *, struct mbuf *); 142 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 143 struct mbuf *); 144 145 /* Active failover */ 146 static int lagg_fail_start(struct lagg_softc *, struct mbuf *); 147 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 148 struct mbuf *); 149 150 /* Loadbalancing */ 151 static void lagg_lb_attach(struct lagg_softc *); 152 static void lagg_lb_detach(struct lagg_softc *); 153 static int lagg_lb_port_create(struct lagg_port *); 154 static void lagg_lb_port_destroy(struct lagg_port *); 155 static int lagg_lb_start(struct lagg_softc *, struct mbuf *); 156 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 157 struct mbuf *); 158 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 159 160 /* Broadcast */ 161 static int lagg_bcast_start(struct lagg_softc *, struct mbuf *); 162 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *, 163 struct mbuf *); 164 165 /* 802.3ad LACP */ 166 static void lagg_lacp_attach(struct lagg_softc *); 167 static void lagg_lacp_detach(struct lagg_softc *); 168 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); 169 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 170 struct mbuf *); 171 static void lagg_lacp_lladdr(struct lagg_softc *); 172 173 /* lagg protocol table */ 174 static const struct lagg_proto { 175 lagg_proto pr_num; 176 void (*pr_attach)(struct lagg_softc *); 177 void (*pr_detach)(struct lagg_softc *); 178 int (*pr_start)(struct lagg_softc *, struct mbuf *); 179 struct mbuf * (*pr_input)(struct lagg_softc *, struct lagg_port *, 180 struct mbuf *); 181 int (*pr_addport)(struct lagg_port *); 182 void (*pr_delport)(struct lagg_port *); 183 void (*pr_linkstate)(struct lagg_port *); 184 void (*pr_init)(struct lagg_softc *); 185 void (*pr_stop)(struct lagg_softc *); 186 void (*pr_lladdr)(struct lagg_softc *); 187 void (*pr_request)(struct lagg_softc *, void *); 188 void (*pr_portreq)(struct lagg_port *, void *); 189 } lagg_protos[] = { 190 { 191 .pr_num = LAGG_PROTO_NONE 192 }, 193 { 194 .pr_num = LAGG_PROTO_ROUNDROBIN, 195 .pr_attach = lagg_rr_attach, 196 .pr_start = lagg_rr_start, 197 .pr_input = lagg_rr_input, 198 }, 199 { 200 .pr_num = LAGG_PROTO_FAILOVER, 201 .pr_start = lagg_fail_start, 202 .pr_input = lagg_fail_input, 203 }, 204 { 205 .pr_num = LAGG_PROTO_LOADBALANCE, 206 .pr_attach = lagg_lb_attach, 207 .pr_detach = lagg_lb_detach, 208 .pr_start = lagg_lb_start, 209 .pr_input = lagg_lb_input, 210 .pr_addport = lagg_lb_port_create, 211 .pr_delport = lagg_lb_port_destroy, 212 }, 213 { 214 .pr_num = LAGG_PROTO_LACP, 215 .pr_attach = lagg_lacp_attach, 216 .pr_detach = lagg_lacp_detach, 217 .pr_start = lagg_lacp_start, 218 .pr_input = lagg_lacp_input, 219 .pr_addport = lacp_port_create, 220 .pr_delport = lacp_port_destroy, 221 .pr_linkstate = lacp_linkstate, 222 .pr_init = lacp_init, 223 .pr_stop = lacp_stop, 224 .pr_lladdr = lagg_lacp_lladdr, 225 .pr_request = lacp_req, 226 .pr_portreq = lacp_portreq, 227 }, 228 { 229 .pr_num = LAGG_PROTO_BROADCAST, 230 .pr_start = lagg_bcast_start, 231 .pr_input = lagg_bcast_input, 232 }, 233 }; 234 235 SYSCTL_DECL(_net_link); 236 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 237 "Link Aggregation"); 238 239 /* Allow input on any failover links */ 240 static VNET_DEFINE(int, lagg_failover_rx_all); 241 #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all) 242 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET, 243 &VNET_NAME(lagg_failover_rx_all), 0, 244 "Accept input from any interface in a failover lagg"); 245 246 /* Default value for using flowid */ 247 static VNET_DEFINE(int, def_use_flowid) = 1; 248 #define V_def_use_flowid VNET(def_use_flowid) 249 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN, 250 &VNET_NAME(def_use_flowid), 0, 251 "Default setting for using flow id for load sharing"); 252 253 /* Default value for flowid shift */ 254 static VNET_DEFINE(int, def_flowid_shift) = 16; 255 #define V_def_flowid_shift VNET(def_flowid_shift) 256 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN, 257 &VNET_NAME(def_flowid_shift), 0, 258 "Default setting for flowid shift for load sharing"); 259 260 static void 261 vnet_lagg_init(const void *unused __unused) 262 { 263 264 LAGG_LIST_LOCK_INIT(); 265 SLIST_INIT(&V_lagg_list); 266 V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create, 267 lagg_clone_destroy, 0); 268 } 269 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 270 vnet_lagg_init, NULL); 271 272 static void 273 vnet_lagg_uninit(const void *unused __unused) 274 { 275 276 if_clone_detach(V_lagg_cloner); 277 LAGG_LIST_LOCK_DESTROY(); 278 } 279 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, 280 vnet_lagg_uninit, NULL); 281 282 static int 283 lagg_modevent(module_t mod, int type, void *data) 284 { 285 286 switch (type) { 287 case MOD_LOAD: 288 lagg_input_p = lagg_input; 289 lagg_linkstate_p = lagg_port_state; 290 lagg_detach_cookie = EVENTHANDLER_REGISTER( 291 ifnet_departure_event, lagg_port_ifdetach, NULL, 292 EVENTHANDLER_PRI_ANY); 293 break; 294 case MOD_UNLOAD: 295 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 296 lagg_detach_cookie); 297 lagg_input_p = NULL; 298 lagg_linkstate_p = NULL; 299 break; 300 default: 301 return (EOPNOTSUPP); 302 } 303 return (0); 304 } 305 306 static moduledata_t lagg_mod = { 307 "if_lagg", 308 lagg_modevent, 309 0 310 }; 311 312 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 313 MODULE_VERSION(if_lagg, 1); 314 315 static void 316 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr) 317 { 318 319 LAGG_XLOCK_ASSERT(sc); 320 KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto", 321 __func__, sc)); 322 323 if (sc->sc_ifflags & IFF_DEBUG) 324 if_printf(sc->sc_ifp, "using proto %u\n", pr); 325 326 if (lagg_protos[pr].pr_attach != NULL) 327 lagg_protos[pr].pr_attach(sc); 328 sc->sc_proto = pr; 329 } 330 331 static void 332 lagg_proto_detach(struct lagg_softc *sc) 333 { 334 lagg_proto pr; 335 336 LAGG_XLOCK_ASSERT(sc); 337 LAGG_WLOCK_ASSERT(sc); 338 pr = sc->sc_proto; 339 sc->sc_proto = LAGG_PROTO_NONE; 340 341 if (lagg_protos[pr].pr_detach != NULL) 342 lagg_protos[pr].pr_detach(sc); 343 else 344 LAGG_WUNLOCK(sc); 345 } 346 347 static int 348 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m) 349 { 350 351 return (lagg_protos[sc->sc_proto].pr_start(sc, m)); 352 } 353 354 static struct mbuf * 355 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 356 { 357 358 return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m)); 359 } 360 361 static int 362 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp) 363 { 364 365 if (lagg_protos[sc->sc_proto].pr_addport == NULL) 366 return (0); 367 else 368 return (lagg_protos[sc->sc_proto].pr_addport(lp)); 369 } 370 371 static void 372 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp) 373 { 374 375 if (lagg_protos[sc->sc_proto].pr_delport != NULL) 376 lagg_protos[sc->sc_proto].pr_delport(lp); 377 } 378 379 static void 380 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp) 381 { 382 383 if (lagg_protos[sc->sc_proto].pr_linkstate != NULL) 384 lagg_protos[sc->sc_proto].pr_linkstate(lp); 385 } 386 387 static void 388 lagg_proto_init(struct lagg_softc *sc) 389 { 390 391 if (lagg_protos[sc->sc_proto].pr_init != NULL) 392 lagg_protos[sc->sc_proto].pr_init(sc); 393 } 394 395 static void 396 lagg_proto_stop(struct lagg_softc *sc) 397 { 398 399 if (lagg_protos[sc->sc_proto].pr_stop != NULL) 400 lagg_protos[sc->sc_proto].pr_stop(sc); 401 } 402 403 static void 404 lagg_proto_lladdr(struct lagg_softc *sc) 405 { 406 407 if (lagg_protos[sc->sc_proto].pr_lladdr != NULL) 408 lagg_protos[sc->sc_proto].pr_lladdr(sc); 409 } 410 411 static void 412 lagg_proto_request(struct lagg_softc *sc, void *v) 413 { 414 415 if (lagg_protos[sc->sc_proto].pr_request != NULL) 416 lagg_protos[sc->sc_proto].pr_request(sc, v); 417 } 418 419 static void 420 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v) 421 { 422 423 if (lagg_protos[sc->sc_proto].pr_portreq != NULL) 424 lagg_protos[sc->sc_proto].pr_portreq(lp, v); 425 } 426 427 /* 428 * This routine is run via an vlan 429 * config EVENT 430 */ 431 static void 432 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 433 { 434 struct lagg_softc *sc = ifp->if_softc; 435 struct lagg_port *lp; 436 437 if (ifp->if_softc != arg) /* Not our event */ 438 return; 439 440 LAGG_SLOCK(sc); 441 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 442 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 443 LAGG_SUNLOCK(sc); 444 } 445 446 /* 447 * This routine is run via an vlan 448 * unconfig EVENT 449 */ 450 static void 451 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 452 { 453 struct lagg_softc *sc = ifp->if_softc; 454 struct lagg_port *lp; 455 456 if (ifp->if_softc != arg) /* Not our event */ 457 return; 458 459 LAGG_SLOCK(sc); 460 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 461 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 462 LAGG_SUNLOCK(sc); 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 LAGG_LOCK_INIT(sc); 479 LAGG_SX_INIT(sc); 480 481 LAGG_XLOCK(sc); 482 if (V_def_use_flowid) 483 sc->sc_opts |= LAGG_OPT_USE_FLOWID; 484 sc->flowid_shift = V_def_flowid_shift; 485 486 /* Hash all layers by default */ 487 sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4; 488 489 lagg_proto_attach(sc, LAGG_PROTO_DEFAULT); 490 491 SLIST_INIT(&sc->sc_ports); 492 493 /* Initialise pseudo media types */ 494 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 495 lagg_media_status); 496 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 497 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 498 499 if_initname(ifp, laggname, unit); 500 ifp->if_softc = sc; 501 ifp->if_transmit = lagg_transmit; 502 ifp->if_qflush = lagg_qflush; 503 ifp->if_init = lagg_init; 504 ifp->if_ioctl = lagg_ioctl; 505 ifp->if_get_counter = lagg_get_counter; 506 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 507 #ifdef RATELIMIT 508 ifp->if_snd_tag_alloc = lagg_snd_tag_alloc; 509 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS | IFCAP_TXRTLMT; 510 #else 511 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS; 512 #endif 513 514 /* 515 * Attach as an ordinary ethernet device, children will be attached 516 * as special device IFT_IEEE8023ADLAG. 517 */ 518 ether_ifattach(ifp, eaddr); 519 520 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 521 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 522 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 523 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 524 525 /* Insert into the global list of laggs */ 526 LAGG_LIST_LOCK(); 527 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries); 528 LAGG_LIST_UNLOCK(); 529 LAGG_XUNLOCK(sc); 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_XLOCK(sc); 541 sc->sc_destroying = 1; 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); 551 552 /* Unhook the aggregation protocol */ 553 LAGG_WLOCK(sc); 554 lagg_proto_detach(sc); 555 LAGG_UNLOCK_ASSERT(sc); 556 LAGG_XUNLOCK(sc); 557 558 ifmedia_removeall(&sc->sc_media); 559 ether_ifdetach(ifp); 560 if_free(ifp); 561 562 LAGG_LIST_LOCK(); 563 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries); 564 LAGG_LIST_UNLOCK(); 565 566 LAGG_SX_DESTROY(sc); 567 LAGG_LOCK_DESTROY(sc); 568 free(sc, M_DEVBUF); 569 } 570 571 static void 572 lagg_capabilities(struct lagg_softc *sc) 573 { 574 struct lagg_port *lp; 575 int cap, ena, pena; 576 uint64_t hwa; 577 struct ifnet_hw_tsomax hw_tsomax; 578 579 LAGG_XLOCK_ASSERT(sc); 580 581 /* Get common enabled capabilities for the lagg ports */ 582 ena = ~0; 583 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 584 ena &= lp->lp_ifp->if_capenable; 585 ena = (ena == ~0 ? 0 : ena); 586 587 /* 588 * Apply common enabled capabilities back to the lagg ports. 589 * May require several iterations if they are dependent. 590 */ 591 do { 592 pena = ena; 593 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 594 lagg_setcaps(lp, ena); 595 ena &= lp->lp_ifp->if_capenable; 596 } 597 } while (pena != ena); 598 599 /* Get other capabilities from the lagg ports */ 600 cap = ~0; 601 hwa = ~(uint64_t)0; 602 memset(&hw_tsomax, 0, sizeof(hw_tsomax)); 603 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 604 cap &= lp->lp_ifp->if_capabilities; 605 hwa &= lp->lp_ifp->if_hwassist; 606 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax); 607 } 608 cap = (cap == ~0 ? 0 : cap); 609 hwa = (hwa == ~(uint64_t)0 ? 0 : hwa); 610 611 if (sc->sc_ifp->if_capabilities != cap || 612 sc->sc_ifp->if_capenable != ena || 613 sc->sc_ifp->if_hwassist != hwa || 614 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) { 615 sc->sc_ifp->if_capabilities = cap; 616 sc->sc_ifp->if_capenable = ena; 617 sc->sc_ifp->if_hwassist = hwa; 618 getmicrotime(&sc->sc_ifp->if_lastchange); 619 620 if (sc->sc_ifflags & IFF_DEBUG) 621 if_printf(sc->sc_ifp, 622 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 623 } 624 } 625 626 static int 627 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 628 { 629 struct lagg_softc *sc_ptr; 630 struct lagg_port *lp, *tlp; 631 int error, i; 632 uint64_t *pval; 633 634 LAGG_XLOCK_ASSERT(sc); 635 636 /* Limit the maximal number of lagg ports */ 637 if (sc->sc_count >= LAGG_MAX_PORTS) 638 return (ENOSPC); 639 640 /* Check if port has already been associated to a lagg */ 641 if (ifp->if_lagg != NULL) { 642 /* Port is already in the current lagg? */ 643 lp = (struct lagg_port *)ifp->if_lagg; 644 if (lp->lp_softc == sc) 645 return (EEXIST); 646 return (EBUSY); 647 } 648 649 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 650 if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN) 651 return (EPROTONOSUPPORT); 652 653 /* Allow the first Ethernet member to define the MTU */ 654 if (SLIST_EMPTY(&sc->sc_ports)) 655 sc->sc_ifp->if_mtu = ifp->if_mtu; 656 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 657 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 658 ifp->if_xname); 659 return (EINVAL); 660 } 661 662 lp = malloc(sizeof(struct lagg_port), M_DEVBUF, M_WAITOK|M_ZERO); 663 lp->lp_softc = sc; 664 665 /* Check if port is a stacked lagg */ 666 LAGG_LIST_LOCK(); 667 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) { 668 if (ifp == sc_ptr->sc_ifp) { 669 LAGG_LIST_UNLOCK(); 670 free(lp, M_DEVBUF); 671 return (EINVAL); 672 /* XXX disable stacking for the moment, its untested */ 673 #ifdef LAGG_PORT_STACKING 674 lp->lp_flags |= LAGG_PORT_STACK; 675 if (lagg_port_checkstacking(sc_ptr) >= 676 LAGG_MAX_STACKING) { 677 LAGG_LIST_UNLOCK(); 678 free(lp, M_DEVBUF); 679 return (E2BIG); 680 } 681 #endif 682 } 683 } 684 LAGG_LIST_UNLOCK(); 685 686 if_ref(ifp); 687 lp->lp_ifp = ifp; 688 689 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 690 lp->lp_ifcapenable = ifp->if_capenable; 691 if (SLIST_EMPTY(&sc->sc_ports)) { 692 LAGG_WLOCK(sc); 693 bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); 694 lagg_proto_lladdr(sc); 695 LAGG_WUNLOCK(sc); 696 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp); 697 } else { 698 if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); 699 } 700 lagg_setflags(lp, 1); 701 702 LAGG_WLOCK(sc); 703 if (SLIST_EMPTY(&sc->sc_ports)) 704 sc->sc_primary = lp; 705 706 /* Change the interface type */ 707 lp->lp_iftype = ifp->if_type; 708 ifp->if_type = IFT_IEEE8023ADLAG; 709 ifp->if_lagg = lp; 710 lp->lp_ioctl = ifp->if_ioctl; 711 ifp->if_ioctl = lagg_port_ioctl; 712 lp->lp_output = ifp->if_output; 713 ifp->if_output = lagg_port_output; 714 715 /* Read port counters */ 716 pval = lp->port_counters.val; 717 for (i = 0; i < IFCOUNTERS; i++, pval++) 718 *pval = ifp->if_get_counter(ifp, i); 719 720 /* 721 * Insert into the list of ports. 722 * Keep ports sorted by if_index. It is handy, when configuration 723 * is predictable and `ifconfig laggN create ...` command 724 * will lead to the same result each time. 725 */ 726 SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) { 727 if (tlp->lp_ifp->if_index < ifp->if_index && ( 728 SLIST_NEXT(tlp, lp_entries) == NULL || 729 SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index > 730 ifp->if_index)) 731 break; 732 } 733 if (tlp != NULL) 734 SLIST_INSERT_AFTER(tlp, lp, lp_entries); 735 else 736 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 737 sc->sc_count++; 738 739 lagg_setmulti(lp); 740 741 LAGG_WUNLOCK(sc); 742 743 if ((error = lagg_proto_addport(sc, lp)) != 0) { 744 /* Remove the port, without calling pr_delport. */ 745 LAGG_WLOCK(sc); 746 lagg_port_destroy(lp, 0); 747 LAGG_UNLOCK_ASSERT(sc); 748 return (error); 749 } 750 751 /* Update lagg capabilities */ 752 lagg_capabilities(sc); 753 lagg_linkstate(sc); 754 755 return (0); 756 } 757 758 #ifdef LAGG_PORT_STACKING 759 static int 760 lagg_port_checkstacking(struct lagg_softc *sc) 761 { 762 struct lagg_softc *sc_ptr; 763 struct lagg_port *lp; 764 int m = 0; 765 766 LAGG_SXLOCK_ASSERT(sc); 767 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 768 if (lp->lp_flags & LAGG_PORT_STACK) { 769 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 770 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 771 } 772 } 773 774 return (m + 1); 775 } 776 #endif 777 778 static int 779 lagg_port_destroy(struct lagg_port *lp, int rundelport) 780 { 781 struct lagg_softc *sc = lp->lp_softc; 782 struct lagg_port *lp_ptr, *lp0; 783 struct ifnet *ifp = lp->lp_ifp; 784 uint64_t *pval, vdiff; 785 int i; 786 787 LAGG_XLOCK_ASSERT(sc); 788 789 if (rundelport) { 790 LAGG_WLOCK(sc); 791 lagg_proto_delport(sc, lp); 792 } else 793 LAGG_WLOCK_ASSERT(sc); 794 795 if (lp->lp_detaching == 0) 796 lagg_clrmulti(lp); 797 798 /* Restore interface */ 799 ifp->if_type = lp->lp_iftype; 800 ifp->if_ioctl = lp->lp_ioctl; 801 ifp->if_output = lp->lp_output; 802 ifp->if_lagg = NULL; 803 804 /* Update detached port counters */ 805 pval = lp->port_counters.val; 806 for (i = 0; i < IFCOUNTERS; i++, pval++) { 807 vdiff = ifp->if_get_counter(ifp, i) - *pval; 808 sc->detached_counters.val[i] += vdiff; 809 } 810 811 /* Finally, remove the port from the lagg */ 812 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 813 sc->sc_count--; 814 815 /* Update the primary interface */ 816 if (lp == sc->sc_primary) { 817 uint8_t lladdr[ETHER_ADDR_LEN]; 818 819 if ((lp0 = SLIST_FIRST(&sc->sc_ports)) == NULL) 820 bzero(&lladdr, ETHER_ADDR_LEN); 821 else 822 bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN); 823 sc->sc_primary = lp0; 824 if (sc->sc_destroying == 0) { 825 bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); 826 lagg_proto_lladdr(sc); 827 LAGG_WUNLOCK(sc); 828 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp); 829 } else 830 LAGG_WUNLOCK(sc); 831 832 /* 833 * Update lladdr for each port (new primary needs update 834 * as well, to switch from old lladdr to its 'real' one) 835 */ 836 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 837 if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN); 838 } else 839 LAGG_WUNLOCK(sc); 840 841 if (lp->lp_ifflags) 842 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 843 844 if (lp->lp_detaching == 0) { 845 lagg_setflags(lp, 0); 846 lagg_setcaps(lp, lp->lp_ifcapenable); 847 if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN); 848 } 849 850 if_rele(ifp); 851 free(lp, M_DEVBUF); 852 853 /* Update lagg capabilities */ 854 lagg_capabilities(sc); 855 lagg_linkstate(sc); 856 857 return (0); 858 } 859 860 static int 861 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 862 { 863 struct lagg_reqport *rp = (struct lagg_reqport *)data; 864 struct lagg_softc *sc; 865 struct lagg_port *lp = NULL; 866 int error = 0; 867 868 /* Should be checked by the caller */ 869 if (ifp->if_type != IFT_IEEE8023ADLAG || 870 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 871 goto fallback; 872 873 switch (cmd) { 874 case SIOCGLAGGPORT: 875 if (rp->rp_portname[0] == '\0' || 876 ifunit(rp->rp_portname) != ifp) { 877 error = EINVAL; 878 break; 879 } 880 881 LAGG_SLOCK(sc); 882 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 883 error = ENOENT; 884 LAGG_SUNLOCK(sc); 885 break; 886 } 887 888 lagg_port2req(lp, rp); 889 LAGG_SUNLOCK(sc); 890 break; 891 892 case SIOCSIFCAP: 893 if (lp->lp_ioctl == NULL) { 894 error = EINVAL; 895 break; 896 } 897 error = (*lp->lp_ioctl)(ifp, cmd, data); 898 if (error) 899 break; 900 901 /* Update lagg interface capabilities */ 902 LAGG_XLOCK(sc); 903 lagg_capabilities(sc); 904 LAGG_XUNLOCK(sc); 905 VLAN_CAPABILITIES(sc->sc_ifp); 906 break; 907 908 case SIOCSIFMTU: 909 /* Do not allow the MTU to be changed once joined */ 910 error = EINVAL; 911 break; 912 913 default: 914 goto fallback; 915 } 916 917 return (error); 918 919 fallback: 920 if (lp != NULL && lp->lp_ioctl != NULL) 921 return ((*lp->lp_ioctl)(ifp, cmd, data)); 922 923 return (EINVAL); 924 } 925 926 /* 927 * Requests counter @cnt data. 928 * 929 * Counter value is calculated the following way: 930 * 1) for each port, sum difference between current and "initial" measurements. 931 * 2) add lagg logical interface counters. 932 * 3) add data from detached_counters array. 933 * 934 * We also do the following things on ports attach/detach: 935 * 1) On port attach we store all counters it has into port_counter array. 936 * 2) On port detach we add the different between "initial" and 937 * current counters data to detached_counters array. 938 */ 939 static uint64_t 940 lagg_get_counter(struct ifnet *ifp, ift_counter cnt) 941 { 942 struct lagg_softc *sc; 943 struct lagg_port *lp; 944 struct ifnet *lpifp; 945 struct rm_priotracker tracker; 946 uint64_t newval, oldval, vsum; 947 948 /* Revise this when we've got non-generic counters. */ 949 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt)); 950 951 sc = (struct lagg_softc *)ifp->if_softc; 952 LAGG_RLOCK(sc, &tracker); 953 954 vsum = 0; 955 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 956 /* Saved attached value */ 957 oldval = lp->port_counters.val[cnt]; 958 /* current value */ 959 lpifp = lp->lp_ifp; 960 newval = lpifp->if_get_counter(lpifp, cnt); 961 /* Calculate diff and save new */ 962 vsum += newval - oldval; 963 } 964 965 /* 966 * Add counter data which might be added by upper 967 * layer protocols operating on logical interface. 968 */ 969 vsum += if_get_counter_default(ifp, cnt); 970 971 /* 972 * Add counter data from detached ports counters 973 */ 974 vsum += sc->detached_counters.val[cnt]; 975 976 LAGG_RUNLOCK(sc, &tracker); 977 978 return (vsum); 979 } 980 981 /* 982 * For direct output to child ports. 983 */ 984 static int 985 lagg_port_output(struct ifnet *ifp, struct mbuf *m, 986 const struct sockaddr *dst, struct route *ro) 987 { 988 struct lagg_port *lp = ifp->if_lagg; 989 990 switch (dst->sa_family) { 991 case pseudo_AF_HDRCMPLT: 992 case AF_UNSPEC: 993 return ((*lp->lp_output)(ifp, m, dst, ro)); 994 } 995 996 /* drop any other frames */ 997 m_freem(m); 998 return (ENETDOWN); 999 } 1000 1001 static void 1002 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 1003 { 1004 struct lagg_port *lp; 1005 struct lagg_softc *sc; 1006 1007 if ((lp = ifp->if_lagg) == NULL) 1008 return; 1009 /* If the ifnet is just being renamed, don't do anything. */ 1010 if (ifp->if_flags & IFF_RENAMING) 1011 return; 1012 1013 sc = lp->lp_softc; 1014 1015 LAGG_XLOCK(sc); 1016 lp->lp_detaching = 1; 1017 lagg_port_destroy(lp, 1); 1018 LAGG_XUNLOCK(sc); 1019 VLAN_CAPABILITIES(sc->sc_ifp); 1020 } 1021 1022 static void 1023 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 1024 { 1025 struct lagg_softc *sc = lp->lp_softc; 1026 1027 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 1028 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 1029 rp->rp_prio = lp->lp_prio; 1030 rp->rp_flags = lp->lp_flags; 1031 lagg_proto_portreq(sc, lp, &rp->rp_psc); 1032 1033 /* Add protocol specific flags */ 1034 switch (sc->sc_proto) { 1035 case LAGG_PROTO_FAILOVER: 1036 if (lp == sc->sc_primary) 1037 rp->rp_flags |= LAGG_PORT_MASTER; 1038 if (lp == lagg_link_active(sc, sc->sc_primary)) 1039 rp->rp_flags |= LAGG_PORT_ACTIVE; 1040 break; 1041 1042 case LAGG_PROTO_ROUNDROBIN: 1043 case LAGG_PROTO_LOADBALANCE: 1044 case LAGG_PROTO_BROADCAST: 1045 if (LAGG_PORTACTIVE(lp)) 1046 rp->rp_flags |= LAGG_PORT_ACTIVE; 1047 break; 1048 1049 case LAGG_PROTO_LACP: 1050 /* LACP has a different definition of active */ 1051 if (lacp_isactive(lp)) 1052 rp->rp_flags |= LAGG_PORT_ACTIVE; 1053 if (lacp_iscollecting(lp)) 1054 rp->rp_flags |= LAGG_PORT_COLLECTING; 1055 if (lacp_isdistributing(lp)) 1056 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 1057 break; 1058 } 1059 1060 } 1061 1062 static void 1063 lagg_init(void *xsc) 1064 { 1065 struct lagg_softc *sc = (struct lagg_softc *)xsc; 1066 struct ifnet *ifp = sc->sc_ifp; 1067 struct lagg_port *lp; 1068 1069 LAGG_XLOCK(sc); 1070 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1071 LAGG_XUNLOCK(sc); 1072 return; 1073 } 1074 1075 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1076 1077 /* 1078 * Update the port lladdrs if needed. 1079 * This might be if_setlladdr() notification 1080 * that lladdr has been changed. 1081 */ 1082 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1083 if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp), 1084 ETHER_ADDR_LEN) != 0) 1085 if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN); 1086 } 1087 1088 lagg_proto_init(sc); 1089 1090 LAGG_XUNLOCK(sc); 1091 } 1092 1093 static void 1094 lagg_stop(struct lagg_softc *sc) 1095 { 1096 struct ifnet *ifp = sc->sc_ifp; 1097 1098 LAGG_XLOCK_ASSERT(sc); 1099 1100 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 1101 return; 1102 1103 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1104 1105 lagg_proto_stop(sc); 1106 } 1107 1108 static int 1109 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1110 { 1111 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1112 struct lagg_reqall *ra = (struct lagg_reqall *)data; 1113 struct lagg_reqopts *ro = (struct lagg_reqopts *)data; 1114 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 1115 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 1116 struct ifreq *ifr = (struct ifreq *)data; 1117 struct lagg_port *lp; 1118 struct ifnet *tpif; 1119 struct thread *td = curthread; 1120 char *buf, *outbuf; 1121 int count, buflen, len, error = 0; 1122 1123 bzero(&rpbuf, sizeof(rpbuf)); 1124 1125 switch (cmd) { 1126 case SIOCGLAGG: 1127 LAGG_SLOCK(sc); 1128 buflen = sc->sc_count * sizeof(struct lagg_reqport); 1129 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 1130 ra->ra_proto = sc->sc_proto; 1131 lagg_proto_request(sc, &ra->ra_psc); 1132 count = 0; 1133 buf = outbuf; 1134 len = min(ra->ra_size, buflen); 1135 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1136 if (len < sizeof(rpbuf)) 1137 break; 1138 1139 lagg_port2req(lp, &rpbuf); 1140 memcpy(buf, &rpbuf, sizeof(rpbuf)); 1141 count++; 1142 buf += sizeof(rpbuf); 1143 len -= sizeof(rpbuf); 1144 } 1145 LAGG_SUNLOCK(sc); 1146 ra->ra_ports = count; 1147 ra->ra_size = count * sizeof(rpbuf); 1148 error = copyout(outbuf, ra->ra_port, ra->ra_size); 1149 free(outbuf, M_TEMP); 1150 break; 1151 case SIOCSLAGG: 1152 error = priv_check(td, PRIV_NET_LAGG); 1153 if (error) 1154 break; 1155 if (ra->ra_proto >= LAGG_PROTO_MAX) { 1156 error = EPROTONOSUPPORT; 1157 break; 1158 } 1159 1160 LAGG_XLOCK(sc); 1161 LAGG_WLOCK(sc); 1162 lagg_proto_detach(sc); 1163 LAGG_UNLOCK_ASSERT(sc); 1164 lagg_proto_attach(sc, ra->ra_proto); 1165 LAGG_XUNLOCK(sc); 1166 break; 1167 case SIOCGLAGGOPTS: 1168 LAGG_SLOCK(sc); 1169 ro->ro_opts = sc->sc_opts; 1170 if (sc->sc_proto == LAGG_PROTO_LACP) { 1171 struct lacp_softc *lsc; 1172 1173 lsc = (struct lacp_softc *)sc->sc_psc; 1174 if (lsc->lsc_debug.lsc_tx_test != 0) 1175 ro->ro_opts |= LAGG_OPT_LACP_TXTEST; 1176 if (lsc->lsc_debug.lsc_rx_test != 0) 1177 ro->ro_opts |= LAGG_OPT_LACP_RXTEST; 1178 if (lsc->lsc_strict_mode != 0) 1179 ro->ro_opts |= LAGG_OPT_LACP_STRICT; 1180 if (lsc->lsc_fast_timeout != 0) 1181 ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT; 1182 1183 ro->ro_active = sc->sc_active; 1184 } else { 1185 ro->ro_active = 0; 1186 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1187 ro->ro_active += LAGG_PORTACTIVE(lp); 1188 } 1189 ro->ro_bkt = sc->sc_bkt; 1190 ro->ro_flapping = sc->sc_flapping; 1191 ro->ro_flowid_shift = sc->flowid_shift; 1192 LAGG_SUNLOCK(sc); 1193 break; 1194 case SIOCSLAGGOPTS: 1195 if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) { 1196 if (ro->ro_bkt == 0) 1197 sc->sc_bkt = 1; // Minimum 1 packet per iface. 1198 else 1199 sc->sc_bkt = ro->ro_bkt; 1200 } 1201 error = priv_check(td, PRIV_NET_LAGG); 1202 if (error) 1203 break; 1204 if (ro->ro_opts == 0) 1205 break; 1206 /* 1207 * Set options. LACP options are stored in sc->sc_psc, 1208 * not in sc_opts. 1209 */ 1210 int valid, lacp; 1211 1212 switch (ro->ro_opts) { 1213 case LAGG_OPT_USE_FLOWID: 1214 case -LAGG_OPT_USE_FLOWID: 1215 case LAGG_OPT_FLOWIDSHIFT: 1216 valid = 1; 1217 lacp = 0; 1218 break; 1219 case LAGG_OPT_LACP_TXTEST: 1220 case -LAGG_OPT_LACP_TXTEST: 1221 case LAGG_OPT_LACP_RXTEST: 1222 case -LAGG_OPT_LACP_RXTEST: 1223 case LAGG_OPT_LACP_STRICT: 1224 case -LAGG_OPT_LACP_STRICT: 1225 case LAGG_OPT_LACP_TIMEOUT: 1226 case -LAGG_OPT_LACP_TIMEOUT: 1227 valid = lacp = 1; 1228 break; 1229 default: 1230 valid = lacp = 0; 1231 break; 1232 } 1233 1234 LAGG_XLOCK(sc); 1235 1236 if (valid == 0 || 1237 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) { 1238 /* Invalid combination of options specified. */ 1239 error = EINVAL; 1240 LAGG_XUNLOCK(sc); 1241 break; /* Return from SIOCSLAGGOPTS. */ 1242 } 1243 /* 1244 * Store new options into sc->sc_opts except for 1245 * FLOWIDSHIFT and LACP options. 1246 */ 1247 if (lacp == 0) { 1248 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT) 1249 sc->flowid_shift = ro->ro_flowid_shift; 1250 else if (ro->ro_opts > 0) 1251 sc->sc_opts |= ro->ro_opts; 1252 else 1253 sc->sc_opts &= ~ro->ro_opts; 1254 } else { 1255 struct lacp_softc *lsc; 1256 struct lacp_port *lp; 1257 1258 lsc = (struct lacp_softc *)sc->sc_psc; 1259 1260 switch (ro->ro_opts) { 1261 case LAGG_OPT_LACP_TXTEST: 1262 lsc->lsc_debug.lsc_tx_test = 1; 1263 break; 1264 case -LAGG_OPT_LACP_TXTEST: 1265 lsc->lsc_debug.lsc_tx_test = 0; 1266 break; 1267 case LAGG_OPT_LACP_RXTEST: 1268 lsc->lsc_debug.lsc_rx_test = 1; 1269 break; 1270 case -LAGG_OPT_LACP_RXTEST: 1271 lsc->lsc_debug.lsc_rx_test = 0; 1272 break; 1273 case LAGG_OPT_LACP_STRICT: 1274 lsc->lsc_strict_mode = 1; 1275 break; 1276 case -LAGG_OPT_LACP_STRICT: 1277 lsc->lsc_strict_mode = 0; 1278 break; 1279 case LAGG_OPT_LACP_TIMEOUT: 1280 LACP_LOCK(lsc); 1281 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1282 lp->lp_state |= LACP_STATE_TIMEOUT; 1283 LACP_UNLOCK(lsc); 1284 lsc->lsc_fast_timeout = 1; 1285 break; 1286 case -LAGG_OPT_LACP_TIMEOUT: 1287 LACP_LOCK(lsc); 1288 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1289 lp->lp_state &= ~LACP_STATE_TIMEOUT; 1290 LACP_UNLOCK(lsc); 1291 lsc->lsc_fast_timeout = 0; 1292 break; 1293 } 1294 } 1295 LAGG_XUNLOCK(sc); 1296 break; 1297 case SIOCGLAGGFLAGS: 1298 rf->rf_flags = 0; 1299 LAGG_SLOCK(sc); 1300 if (sc->sc_flags & MBUF_HASHFLAG_L2) 1301 rf->rf_flags |= LAGG_F_HASHL2; 1302 if (sc->sc_flags & MBUF_HASHFLAG_L3) 1303 rf->rf_flags |= LAGG_F_HASHL3; 1304 if (sc->sc_flags & MBUF_HASHFLAG_L4) 1305 rf->rf_flags |= LAGG_F_HASHL4; 1306 LAGG_SUNLOCK(sc); 1307 break; 1308 case SIOCSLAGGHASH: 1309 error = priv_check(td, PRIV_NET_LAGG); 1310 if (error) 1311 break; 1312 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1313 error = EINVAL; 1314 break; 1315 } 1316 LAGG_XLOCK(sc); 1317 sc->sc_flags = 0; 1318 if (rf->rf_flags & LAGG_F_HASHL2) 1319 sc->sc_flags |= MBUF_HASHFLAG_L2; 1320 if (rf->rf_flags & LAGG_F_HASHL3) 1321 sc->sc_flags |= MBUF_HASHFLAG_L3; 1322 if (rf->rf_flags & LAGG_F_HASHL4) 1323 sc->sc_flags |= MBUF_HASHFLAG_L4; 1324 LAGG_XUNLOCK(sc); 1325 break; 1326 case SIOCGLAGGPORT: 1327 if (rp->rp_portname[0] == '\0' || 1328 (tpif = ifunit_ref(rp->rp_portname)) == NULL) { 1329 error = EINVAL; 1330 break; 1331 } 1332 1333 LAGG_SLOCK(sc); 1334 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1335 lp->lp_softc != sc) { 1336 error = ENOENT; 1337 LAGG_SUNLOCK(sc); 1338 if_rele(tpif); 1339 break; 1340 } 1341 1342 lagg_port2req(lp, rp); 1343 LAGG_SUNLOCK(sc); 1344 if_rele(tpif); 1345 break; 1346 case SIOCSLAGGPORT: 1347 error = priv_check(td, PRIV_NET_LAGG); 1348 if (error) 1349 break; 1350 if (rp->rp_portname[0] == '\0' || 1351 (tpif = ifunit_ref(rp->rp_portname)) == NULL) { 1352 error = EINVAL; 1353 break; 1354 } 1355 #ifdef INET6 1356 /* 1357 * A laggport interface should not have inet6 address 1358 * because two interfaces with a valid link-local 1359 * scope zone must not be merged in any form. This 1360 * restriction is needed to prevent violation of 1361 * link-local scope zone. Attempts to add a laggport 1362 * interface which has inet6 addresses triggers 1363 * removal of all inet6 addresses on the member 1364 * interface. 1365 */ 1366 if (in6ifa_llaonifp(tpif)) { 1367 in6_ifdetach(tpif); 1368 if_printf(sc->sc_ifp, 1369 "IPv6 addresses on %s have been removed " 1370 "before adding it as a member to prevent " 1371 "IPv6 address scope violation.\n", 1372 tpif->if_xname); 1373 } 1374 #endif 1375 LAGG_XLOCK(sc); 1376 error = lagg_port_create(sc, tpif); 1377 LAGG_XUNLOCK(sc); 1378 if_rele(tpif); 1379 VLAN_CAPABILITIES(ifp); 1380 break; 1381 case SIOCSLAGGDELPORT: 1382 error = priv_check(td, PRIV_NET_LAGG); 1383 if (error) 1384 break; 1385 if (rp->rp_portname[0] == '\0' || 1386 (tpif = ifunit_ref(rp->rp_portname)) == NULL) { 1387 error = EINVAL; 1388 break; 1389 } 1390 1391 LAGG_XLOCK(sc); 1392 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1393 lp->lp_softc != sc) { 1394 error = ENOENT; 1395 LAGG_XUNLOCK(sc); 1396 if_rele(tpif); 1397 break; 1398 } 1399 1400 error = lagg_port_destroy(lp, 1); 1401 LAGG_XUNLOCK(sc); 1402 if_rele(tpif); 1403 VLAN_CAPABILITIES(ifp); 1404 break; 1405 case SIOCSIFFLAGS: 1406 /* Set flags on ports too */ 1407 LAGG_XLOCK(sc); 1408 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1409 lagg_setflags(lp, 1); 1410 } 1411 1412 if (!(ifp->if_flags & IFF_UP) && 1413 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1414 /* 1415 * If interface is marked down and it is running, 1416 * then stop and disable it. 1417 */ 1418 lagg_stop(sc); 1419 LAGG_XUNLOCK(sc); 1420 } else if ((ifp->if_flags & IFF_UP) && 1421 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1422 /* 1423 * If interface is marked up and it is stopped, then 1424 * start it. 1425 */ 1426 LAGG_XUNLOCK(sc); 1427 (*ifp->if_init)(sc); 1428 } else 1429 LAGG_XUNLOCK(sc); 1430 break; 1431 case SIOCADDMULTI: 1432 case SIOCDELMULTI: 1433 LAGG_WLOCK(sc); 1434 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1435 lagg_clrmulti(lp); 1436 lagg_setmulti(lp); 1437 } 1438 LAGG_WUNLOCK(sc); 1439 error = 0; 1440 break; 1441 case SIOCSIFMEDIA: 1442 case SIOCGIFMEDIA: 1443 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1444 break; 1445 1446 case SIOCSIFCAP: 1447 LAGG_XLOCK(sc); 1448 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1449 if (lp->lp_ioctl != NULL) 1450 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); 1451 } 1452 lagg_capabilities(sc); 1453 LAGG_XUNLOCK(sc); 1454 VLAN_CAPABILITIES(ifp); 1455 error = 0; 1456 break; 1457 1458 case SIOCSIFMTU: 1459 /* Do not allow the MTU to be directly changed */ 1460 error = EINVAL; 1461 break; 1462 1463 default: 1464 error = ether_ioctl(ifp, cmd, data); 1465 break; 1466 } 1467 return (error); 1468 } 1469 1470 #ifdef RATELIMIT 1471 static int 1472 lagg_snd_tag_alloc(struct ifnet *ifp, 1473 union if_snd_tag_alloc_params *params, 1474 struct m_snd_tag **ppmt) 1475 { 1476 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1477 struct lagg_port *lp; 1478 struct lagg_lb *lb; 1479 uint32_t p; 1480 1481 switch (sc->sc_proto) { 1482 case LAGG_PROTO_FAILOVER: 1483 lp = lagg_link_active(sc, sc->sc_primary); 1484 break; 1485 case LAGG_PROTO_LOADBALANCE: 1486 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 || 1487 params->hdr.flowtype == M_HASHTYPE_NONE) 1488 return (EOPNOTSUPP); 1489 p = params->hdr.flowid >> sc->flowid_shift; 1490 p %= sc->sc_count; 1491 lb = (struct lagg_lb *)sc->sc_psc; 1492 lp = lb->lb_ports[p]; 1493 lp = lagg_link_active(sc, lp); 1494 break; 1495 case LAGG_PROTO_LACP: 1496 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 || 1497 params->hdr.flowtype == M_HASHTYPE_NONE) 1498 return (EOPNOTSUPP); 1499 lp = lacp_select_tx_port_by_hash(sc, params->hdr.flowid); 1500 break; 1501 default: 1502 return (EOPNOTSUPP); 1503 } 1504 if (lp == NULL) 1505 return (EOPNOTSUPP); 1506 ifp = lp->lp_ifp; 1507 if (ifp == NULL || ifp->if_snd_tag_alloc == NULL || 1508 (ifp->if_capenable & IFCAP_TXRTLMT) == 0) 1509 return (EOPNOTSUPP); 1510 1511 /* forward allocation request */ 1512 return (ifp->if_snd_tag_alloc(ifp, params, ppmt)); 1513 } 1514 #endif 1515 1516 static int 1517 lagg_setmulti(struct lagg_port *lp) 1518 { 1519 struct lagg_softc *sc = lp->lp_softc; 1520 struct ifnet *ifp = lp->lp_ifp; 1521 struct ifnet *scifp = sc->sc_ifp; 1522 struct lagg_mc *mc; 1523 struct ifmultiaddr *ifma; 1524 int error; 1525 1526 LAGG_WLOCK_ASSERT(sc); 1527 IF_ADDR_WLOCK(scifp); 1528 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1529 if (ifma->ifma_addr->sa_family != AF_LINK) 1530 continue; 1531 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1532 if (mc == NULL) { 1533 IF_ADDR_WUNLOCK(scifp); 1534 return (ENOMEM); 1535 } 1536 bcopy(ifma->ifma_addr, &mc->mc_addr, 1537 ifma->ifma_addr->sa_len); 1538 mc->mc_addr.sdl_index = ifp->if_index; 1539 mc->mc_ifma = NULL; 1540 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1541 } 1542 IF_ADDR_WUNLOCK(scifp); 1543 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) { 1544 error = if_addmulti(ifp, 1545 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma); 1546 if (error) 1547 return (error); 1548 } 1549 return (0); 1550 } 1551 1552 static int 1553 lagg_clrmulti(struct lagg_port *lp) 1554 { 1555 struct lagg_mc *mc; 1556 1557 LAGG_WLOCK_ASSERT(lp->lp_softc); 1558 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1559 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1560 if (mc->mc_ifma && lp->lp_detaching == 0) 1561 if_delmulti_ifma(mc->mc_ifma); 1562 free(mc, M_DEVBUF); 1563 } 1564 return (0); 1565 } 1566 1567 static int 1568 lagg_setcaps(struct lagg_port *lp, int cap) 1569 { 1570 struct ifreq ifr; 1571 1572 if (lp->lp_ifp->if_capenable == cap) 1573 return (0); 1574 if (lp->lp_ioctl == NULL) 1575 return (ENXIO); 1576 ifr.ifr_reqcap = cap; 1577 return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr)); 1578 } 1579 1580 /* Handle a ref counted flag that should be set on the lagg port as well */ 1581 static int 1582 lagg_setflag(struct lagg_port *lp, int flag, int status, 1583 int (*func)(struct ifnet *, int)) 1584 { 1585 struct lagg_softc *sc = lp->lp_softc; 1586 struct ifnet *scifp = sc->sc_ifp; 1587 struct ifnet *ifp = lp->lp_ifp; 1588 int error; 1589 1590 LAGG_XLOCK_ASSERT(sc); 1591 1592 status = status ? (scifp->if_flags & flag) : 0; 1593 /* Now "status" contains the flag value or 0 */ 1594 1595 /* 1596 * See if recorded ports status is different from what 1597 * we want it to be. If it is, flip it. We record ports 1598 * status in lp_ifflags so that we won't clear ports flag 1599 * we haven't set. In fact, we don't clear or set ports 1600 * flags directly, but get or release references to them. 1601 * That's why we can be sure that recorded flags still are 1602 * in accord with actual ports flags. 1603 */ 1604 if (status != (lp->lp_ifflags & flag)) { 1605 error = (*func)(ifp, status); 1606 if (error) 1607 return (error); 1608 lp->lp_ifflags &= ~flag; 1609 lp->lp_ifflags |= status; 1610 } 1611 return (0); 1612 } 1613 1614 /* 1615 * Handle IFF_* flags that require certain changes on the lagg port 1616 * if "status" is true, update ports flags respective to the lagg 1617 * if "status" is false, forcedly clear the flags set on port. 1618 */ 1619 static int 1620 lagg_setflags(struct lagg_port *lp, int status) 1621 { 1622 int error, i; 1623 1624 for (i = 0; lagg_pflags[i].flag; i++) { 1625 error = lagg_setflag(lp, lagg_pflags[i].flag, 1626 status, lagg_pflags[i].func); 1627 if (error) 1628 return (error); 1629 } 1630 return (0); 1631 } 1632 1633 static int 1634 lagg_transmit(struct ifnet *ifp, struct mbuf *m) 1635 { 1636 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1637 int error, len, mcast; 1638 struct rm_priotracker tracker; 1639 1640 len = m->m_pkthdr.len; 1641 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1642 1643 LAGG_RLOCK(sc, &tracker); 1644 /* We need a Tx algorithm and at least one port */ 1645 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1646 LAGG_RUNLOCK(sc, &tracker); 1647 m_freem(m); 1648 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1649 return (ENXIO); 1650 } 1651 1652 ETHER_BPF_MTAP(ifp, m); 1653 1654 error = lagg_proto_start(sc, m); 1655 LAGG_RUNLOCK(sc, &tracker); 1656 1657 if (error != 0) 1658 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1659 1660 return (error); 1661 } 1662 1663 /* 1664 * The ifp->if_qflush entry point for lagg(4) is no-op. 1665 */ 1666 static void 1667 lagg_qflush(struct ifnet *ifp __unused) 1668 { 1669 } 1670 1671 static struct mbuf * 1672 lagg_input(struct ifnet *ifp, struct mbuf *m) 1673 { 1674 struct lagg_port *lp = ifp->if_lagg; 1675 struct lagg_softc *sc = lp->lp_softc; 1676 struct ifnet *scifp = sc->sc_ifp; 1677 struct rm_priotracker tracker; 1678 1679 LAGG_RLOCK(sc, &tracker); 1680 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1681 (lp->lp_flags & LAGG_PORT_DISABLED) || 1682 sc->sc_proto == LAGG_PROTO_NONE) { 1683 LAGG_RUNLOCK(sc, &tracker); 1684 m_freem(m); 1685 return (NULL); 1686 } 1687 1688 ETHER_BPF_MTAP(scifp, m); 1689 1690 if (lp->lp_detaching != 0) { 1691 m_freem(m); 1692 m = NULL; 1693 } else 1694 m = lagg_proto_input(sc, lp, m); 1695 1696 if (m != NULL) { 1697 if (scifp->if_flags & IFF_MONITOR) { 1698 m_freem(m); 1699 m = NULL; 1700 } 1701 } 1702 1703 LAGG_RUNLOCK(sc, &tracker); 1704 return (m); 1705 } 1706 1707 static int 1708 lagg_media_change(struct ifnet *ifp) 1709 { 1710 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1711 1712 if (sc->sc_ifflags & IFF_DEBUG) 1713 printf("%s\n", __func__); 1714 1715 /* Ignore */ 1716 return (0); 1717 } 1718 1719 static void 1720 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1721 { 1722 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1723 struct lagg_port *lp; 1724 1725 imr->ifm_status = IFM_AVALID; 1726 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1727 1728 LAGG_SLOCK(sc); 1729 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1730 if (LAGG_PORTACTIVE(lp)) 1731 imr->ifm_status |= IFM_ACTIVE; 1732 } 1733 LAGG_SUNLOCK(sc); 1734 } 1735 1736 static void 1737 lagg_linkstate(struct lagg_softc *sc) 1738 { 1739 struct lagg_port *lp; 1740 int new_link = LINK_STATE_DOWN; 1741 uint64_t speed; 1742 1743 LAGG_XLOCK_ASSERT(sc); 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_BROADCAST: 1763 speed = 0; 1764 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1765 speed += lp->lp_ifp->if_baudrate; 1766 sc->sc_ifp->if_baudrate = speed; 1767 break; 1768 case LAGG_PROTO_LACP: 1769 /* LACP updates if_baudrate itself */ 1770 break; 1771 } 1772 } 1773 1774 static void 1775 lagg_port_state(struct ifnet *ifp, int state) 1776 { 1777 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1778 struct lagg_softc *sc = NULL; 1779 1780 if (lp != NULL) 1781 sc = lp->lp_softc; 1782 if (sc == NULL) 1783 return; 1784 1785 LAGG_XLOCK(sc); 1786 lagg_linkstate(sc); 1787 lagg_proto_linkstate(sc, lp); 1788 LAGG_XUNLOCK(sc); 1789 } 1790 1791 struct lagg_port * 1792 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1793 { 1794 struct lagg_port *lp_next, *rval = NULL; 1795 1796 /* 1797 * Search a port which reports an active link state. 1798 */ 1799 1800 if (lp == NULL) 1801 goto search; 1802 if (LAGG_PORTACTIVE(lp)) { 1803 rval = lp; 1804 goto found; 1805 } 1806 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1807 LAGG_PORTACTIVE(lp_next)) { 1808 rval = lp_next; 1809 goto found; 1810 } 1811 1812 search: 1813 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1814 if (LAGG_PORTACTIVE(lp_next)) { 1815 rval = lp_next; 1816 goto found; 1817 } 1818 } 1819 1820 found: 1821 return (rval); 1822 } 1823 1824 int 1825 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1826 { 1827 1828 return (ifp->if_transmit)(ifp, m); 1829 } 1830 1831 /* 1832 * Simple round robin aggregation 1833 */ 1834 static void 1835 lagg_rr_attach(struct lagg_softc *sc) 1836 { 1837 sc->sc_seq = 0; 1838 sc->sc_bkt_count = sc->sc_bkt; 1839 } 1840 1841 static int 1842 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 1843 { 1844 struct lagg_port *lp; 1845 uint32_t p; 1846 1847 if (sc->sc_bkt_count == 0 && sc->sc_bkt > 0) 1848 sc->sc_bkt_count = sc->sc_bkt; 1849 1850 if (sc->sc_bkt > 0) { 1851 atomic_subtract_int(&sc->sc_bkt_count, 1); 1852 if (atomic_cmpset_int(&sc->sc_bkt_count, 0, sc->sc_bkt)) 1853 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1854 else 1855 p = sc->sc_seq; 1856 } else 1857 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1858 1859 p %= sc->sc_count; 1860 lp = SLIST_FIRST(&sc->sc_ports); 1861 1862 while (p--) 1863 lp = SLIST_NEXT(lp, lp_entries); 1864 1865 /* 1866 * Check the port's link state. This will return the next active 1867 * port if the link is down or the port is NULL. 1868 */ 1869 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1870 m_freem(m); 1871 return (ENETDOWN); 1872 } 1873 1874 /* Send mbuf */ 1875 return (lagg_enqueue(lp->lp_ifp, m)); 1876 } 1877 1878 static struct mbuf * 1879 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1880 { 1881 struct ifnet *ifp = sc->sc_ifp; 1882 1883 /* Just pass in the packet to our lagg device */ 1884 m->m_pkthdr.rcvif = ifp; 1885 1886 return (m); 1887 } 1888 1889 /* 1890 * Broadcast mode 1891 */ 1892 static int 1893 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m) 1894 { 1895 int active_ports = 0; 1896 int errors = 0; 1897 int ret; 1898 struct lagg_port *lp, *last = NULL; 1899 struct mbuf *m0; 1900 1901 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1902 if (!LAGG_PORTACTIVE(lp)) 1903 continue; 1904 1905 active_ports++; 1906 1907 if (last != NULL) { 1908 m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT); 1909 if (m0 == NULL) { 1910 ret = ENOBUFS; 1911 errors++; 1912 break; 1913 } 1914 1915 ret = lagg_enqueue(last->lp_ifp, m0); 1916 if (ret != 0) 1917 errors++; 1918 } 1919 last = lp; 1920 } 1921 if (last == NULL) { 1922 m_freem(m); 1923 return (ENOENT); 1924 } 1925 if ((last = lagg_link_active(sc, last)) == NULL) { 1926 m_freem(m); 1927 return (ENETDOWN); 1928 } 1929 1930 ret = lagg_enqueue(last->lp_ifp, m); 1931 if (ret != 0) 1932 errors++; 1933 1934 if (errors == 0) 1935 return (ret); 1936 1937 return (0); 1938 } 1939 1940 static struct mbuf* 1941 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1942 { 1943 struct ifnet *ifp = sc->sc_ifp; 1944 1945 /* Just pass in the packet to our lagg device */ 1946 m->m_pkthdr.rcvif = ifp; 1947 return (m); 1948 } 1949 1950 /* 1951 * Active failover 1952 */ 1953 static int 1954 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 1955 { 1956 struct lagg_port *lp; 1957 1958 /* Use the master port if active or the next available port */ 1959 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 1960 m_freem(m); 1961 return (ENETDOWN); 1962 } 1963 1964 /* Send mbuf */ 1965 return (lagg_enqueue(lp->lp_ifp, m)); 1966 } 1967 1968 static struct mbuf * 1969 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1970 { 1971 struct ifnet *ifp = sc->sc_ifp; 1972 struct lagg_port *tmp_tp; 1973 1974 if (lp == sc->sc_primary || V_lagg_failover_rx_all) { 1975 m->m_pkthdr.rcvif = ifp; 1976 return (m); 1977 } 1978 1979 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1980 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1981 /* 1982 * If tmp_tp is null, we've received a packet when all 1983 * our links are down. Weird, but process it anyways. 1984 */ 1985 if ((tmp_tp == NULL || tmp_tp == lp)) { 1986 m->m_pkthdr.rcvif = ifp; 1987 return (m); 1988 } 1989 } 1990 1991 m_freem(m); 1992 return (NULL); 1993 } 1994 1995 /* 1996 * Loadbalancing 1997 */ 1998 static void 1999 lagg_lb_attach(struct lagg_softc *sc) 2000 { 2001 struct lagg_port *lp; 2002 struct lagg_lb *lb; 2003 2004 lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO); 2005 lb->lb_key = m_ether_tcpip_hash_init(); 2006 sc->sc_psc = lb; 2007 2008 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2009 lagg_lb_port_create(lp); 2010 } 2011 2012 static void 2013 lagg_lb_detach(struct lagg_softc *sc) 2014 { 2015 struct lagg_lb *lb; 2016 2017 lb = (struct lagg_lb *)sc->sc_psc; 2018 LAGG_WUNLOCK(sc); 2019 if (lb != NULL) 2020 free(lb, M_DEVBUF); 2021 } 2022 2023 static int 2024 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 2025 { 2026 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 2027 struct lagg_port *lp_next; 2028 int i = 0; 2029 2030 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 2031 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 2032 if (lp_next == lp) 2033 continue; 2034 if (i >= LAGG_MAX_PORTS) 2035 return (EINVAL); 2036 if (sc->sc_ifflags & IFF_DEBUG) 2037 printf("%s: port %s at index %d\n", 2038 sc->sc_ifname, lp_next->lp_ifp->if_xname, i); 2039 lb->lb_ports[i++] = lp_next; 2040 } 2041 2042 return (0); 2043 } 2044 2045 static int 2046 lagg_lb_port_create(struct lagg_port *lp) 2047 { 2048 struct lagg_softc *sc = lp->lp_softc; 2049 return (lagg_lb_porttable(sc, NULL)); 2050 } 2051 2052 static void 2053 lagg_lb_port_destroy(struct lagg_port *lp) 2054 { 2055 struct lagg_softc *sc = lp->lp_softc; 2056 lagg_lb_porttable(sc, lp); 2057 } 2058 2059 static int 2060 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 2061 { 2062 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 2063 struct lagg_port *lp = NULL; 2064 uint32_t p = 0; 2065 2066 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && 2067 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 2068 p = m->m_pkthdr.flowid >> sc->flowid_shift; 2069 else 2070 p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key); 2071 p %= sc->sc_count; 2072 lp = lb->lb_ports[p]; 2073 2074 /* 2075 * Check the port's link state. This will return the next active 2076 * port if the link is down or the port is NULL. 2077 */ 2078 if ((lp = lagg_link_active(sc, lp)) == NULL) { 2079 m_freem(m); 2080 return (ENETDOWN); 2081 } 2082 2083 /* Send mbuf */ 2084 return (lagg_enqueue(lp->lp_ifp, m)); 2085 } 2086 2087 static struct mbuf * 2088 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2089 { 2090 struct ifnet *ifp = sc->sc_ifp; 2091 2092 /* Just pass in the packet to our lagg device */ 2093 m->m_pkthdr.rcvif = ifp; 2094 2095 return (m); 2096 } 2097 2098 /* 2099 * 802.3ad LACP 2100 */ 2101 static void 2102 lagg_lacp_attach(struct lagg_softc *sc) 2103 { 2104 struct lagg_port *lp; 2105 2106 lacp_attach(sc); 2107 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2108 lacp_port_create(lp); 2109 } 2110 2111 static void 2112 lagg_lacp_detach(struct lagg_softc *sc) 2113 { 2114 struct lagg_port *lp; 2115 void *psc; 2116 2117 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2118 lacp_port_destroy(lp); 2119 2120 psc = sc->sc_psc; 2121 sc->sc_psc = NULL; 2122 LAGG_WUNLOCK(sc); 2123 2124 lacp_detach(psc); 2125 } 2126 2127 static void 2128 lagg_lacp_lladdr(struct lagg_softc *sc) 2129 { 2130 struct lagg_port *lp; 2131 2132 LAGG_SXLOCK_ASSERT(sc); 2133 2134 /* purge all the lacp ports */ 2135 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2136 lacp_port_destroy(lp); 2137 2138 /* add them back in */ 2139 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2140 lacp_port_create(lp); 2141 } 2142 2143 static int 2144 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 2145 { 2146 struct lagg_port *lp; 2147 2148 lp = lacp_select_tx_port(sc, m); 2149 if (lp == NULL) { 2150 m_freem(m); 2151 return (ENETDOWN); 2152 } 2153 2154 /* Send mbuf */ 2155 return (lagg_enqueue(lp->lp_ifp, m)); 2156 } 2157 2158 static struct mbuf * 2159 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2160 { 2161 struct ifnet *ifp = sc->sc_ifp; 2162 struct ether_header *eh; 2163 u_short etype; 2164 2165 eh = mtod(m, struct ether_header *); 2166 etype = ntohs(eh->ether_type); 2167 2168 /* Tap off LACP control messages */ 2169 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 2170 m = lacp_input(lp, m); 2171 if (m == NULL) 2172 return (NULL); 2173 } 2174 2175 /* 2176 * If the port is not collecting or not in the active aggregator then 2177 * free and return. 2178 */ 2179 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 2180 m_freem(m); 2181 return (NULL); 2182 } 2183 2184 m->m_pkthdr.rcvif = ifp; 2185 return (m); 2186 } 2187 2188