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_kern_tls.h" 27 #include "opt_ratelimit.h" 28 29 #include <sys/param.h> 30 #include <sys/kernel.h> 31 #include <sys/malloc.h> 32 #include <sys/mbuf.h> 33 #include <sys/queue.h> 34 #include <sys/socket.h> 35 #include <sys/sockio.h> 36 #include <sys/sysctl.h> 37 #include <sys/module.h> 38 #include <sys/priv.h> 39 #include <sys/systm.h> 40 #include <sys/proc.h> 41 #include <sys/lock.h> 42 #include <sys/rmlock.h> 43 #include <sys/sx.h> 44 #include <sys/taskqueue.h> 45 #include <sys/eventhandler.h> 46 47 #include <net/ethernet.h> 48 #include <net/if.h> 49 #include <net/if_clone.h> 50 #include <net/if_arp.h> 51 #include <net/if_dl.h> 52 #include <net/if_media.h> 53 #include <net/if_types.h> 54 #include <net/if_var.h> 55 #include <net/bpf.h> 56 #include <net/vnet.h> 57 58 #if defined(INET) || defined(INET6) 59 #include <netinet/in.h> 60 #include <netinet/ip.h> 61 #endif 62 #ifdef INET 63 #include <netinet/in_systm.h> 64 #include <netinet/if_ether.h> 65 #endif 66 67 #ifdef INET6 68 #include <netinet/ip6.h> 69 #include <netinet6/in6_var.h> 70 #include <netinet6/in6_ifattach.h> 71 #endif 72 73 #include <net/if_vlan_var.h> 74 #include <net/if_lagg.h> 75 #include <net/ieee8023ad_lacp.h> 76 77 #define LAGG_RLOCK() struct epoch_tracker lagg_et; epoch_enter_preempt(net_epoch_preempt, &lagg_et) 78 #define LAGG_RUNLOCK() epoch_exit_preempt(net_epoch_preempt, &lagg_et) 79 #define LAGG_RLOCK_ASSERT() MPASS(in_epoch(net_epoch_preempt)) 80 #define LAGG_UNLOCK_ASSERT() MPASS(!in_epoch(net_epoch_preempt)) 81 82 #define LAGG_SX_INIT(_sc) sx_init(&(_sc)->sc_sx, "if_lagg sx") 83 #define LAGG_SX_DESTROY(_sc) sx_destroy(&(_sc)->sc_sx) 84 #define LAGG_XLOCK(_sc) sx_xlock(&(_sc)->sc_sx) 85 #define LAGG_XUNLOCK(_sc) sx_xunlock(&(_sc)->sc_sx) 86 #define LAGG_SXLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_LOCKED) 87 #define LAGG_XLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_XLOCKED) 88 89 /* Special flags we should propagate to the lagg ports. */ 90 static struct { 91 int flag; 92 int (*func)(struct ifnet *, int); 93 } lagg_pflags[] = { 94 {IFF_PROMISC, ifpromisc}, 95 {IFF_ALLMULTI, if_allmulti}, 96 {0, NULL} 97 }; 98 99 struct lagg_snd_tag { 100 struct m_snd_tag com; 101 struct m_snd_tag *tag; 102 }; 103 104 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */ 105 #define V_lagg_list VNET(lagg_list) 106 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx); 107 #define V_lagg_list_mtx VNET(lagg_list_mtx) 108 #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \ 109 "if_lagg list", NULL, MTX_DEF) 110 #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx) 111 #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx) 112 #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx) 113 eventhandler_tag lagg_detach_cookie = NULL; 114 115 static int lagg_clone_create(struct if_clone *, int, caddr_t); 116 static void lagg_clone_destroy(struct ifnet *); 117 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner); 118 #define V_lagg_cloner VNET(lagg_cloner) 119 static const char laggname[] = "lagg"; 120 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface"); 121 122 static void lagg_capabilities(struct lagg_softc *); 123 static int lagg_port_create(struct lagg_softc *, struct ifnet *); 124 static int lagg_port_destroy(struct lagg_port *, int); 125 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); 126 static void lagg_linkstate(struct lagg_softc *); 127 static void lagg_port_state(struct ifnet *, int); 128 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); 129 static int lagg_port_output(struct ifnet *, struct mbuf *, 130 const struct sockaddr *, struct route *); 131 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 132 #ifdef LAGG_PORT_STACKING 133 static int lagg_port_checkstacking(struct lagg_softc *); 134 #endif 135 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 136 static void lagg_init(void *); 137 static void lagg_stop(struct lagg_softc *); 138 static int lagg_ioctl(struct ifnet *, u_long, caddr_t); 139 #if defined(KERN_TLS) || defined(RATELIMIT) 140 static int lagg_snd_tag_alloc(struct ifnet *, 141 union if_snd_tag_alloc_params *, 142 struct m_snd_tag **); 143 static int lagg_snd_tag_modify(struct m_snd_tag *, 144 union if_snd_tag_modify_params *); 145 static int lagg_snd_tag_query(struct m_snd_tag *, 146 union if_snd_tag_query_params *); 147 static void lagg_snd_tag_free(struct m_snd_tag *); 148 static void lagg_ratelimit_query(struct ifnet *, 149 struct if_ratelimit_query_results *); 150 #endif 151 static int lagg_setmulti(struct lagg_port *); 152 static int lagg_clrmulti(struct lagg_port *); 153 static int lagg_setcaps(struct lagg_port *, int cap); 154 static int lagg_setflag(struct lagg_port *, int, int, 155 int (*func)(struct ifnet *, int)); 156 static int lagg_setflags(struct lagg_port *, int status); 157 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt); 158 static int lagg_transmit(struct ifnet *, struct mbuf *); 159 static void lagg_qflush(struct ifnet *); 160 static int lagg_media_change(struct ifnet *); 161 static void lagg_media_status(struct ifnet *, struct ifmediareq *); 162 static struct lagg_port *lagg_link_active(struct lagg_softc *, 163 struct lagg_port *); 164 165 /* Simple round robin */ 166 static void lagg_rr_attach(struct lagg_softc *); 167 static int lagg_rr_start(struct lagg_softc *, struct mbuf *); 168 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 169 struct mbuf *); 170 171 /* Active failover */ 172 static int lagg_fail_start(struct lagg_softc *, struct mbuf *); 173 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 174 struct mbuf *); 175 176 /* Loadbalancing */ 177 static void lagg_lb_attach(struct lagg_softc *); 178 static void lagg_lb_detach(struct lagg_softc *); 179 static int lagg_lb_port_create(struct lagg_port *); 180 static void lagg_lb_port_destroy(struct lagg_port *); 181 static int lagg_lb_start(struct lagg_softc *, struct mbuf *); 182 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 183 struct mbuf *); 184 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 185 186 /* Broadcast */ 187 static int lagg_bcast_start(struct lagg_softc *, struct mbuf *); 188 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *, 189 struct mbuf *); 190 191 /* 802.3ad LACP */ 192 static void lagg_lacp_attach(struct lagg_softc *); 193 static void lagg_lacp_detach(struct lagg_softc *); 194 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); 195 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 196 struct mbuf *); 197 static void lagg_lacp_lladdr(struct lagg_softc *); 198 199 /* lagg protocol table */ 200 static const struct lagg_proto { 201 lagg_proto pr_num; 202 void (*pr_attach)(struct lagg_softc *); 203 void (*pr_detach)(struct lagg_softc *); 204 int (*pr_start)(struct lagg_softc *, struct mbuf *); 205 struct mbuf * (*pr_input)(struct lagg_softc *, struct lagg_port *, 206 struct mbuf *); 207 int (*pr_addport)(struct lagg_port *); 208 void (*pr_delport)(struct lagg_port *); 209 void (*pr_linkstate)(struct lagg_port *); 210 void (*pr_init)(struct lagg_softc *); 211 void (*pr_stop)(struct lagg_softc *); 212 void (*pr_lladdr)(struct lagg_softc *); 213 void (*pr_request)(struct lagg_softc *, void *); 214 void (*pr_portreq)(struct lagg_port *, void *); 215 } lagg_protos[] = { 216 { 217 .pr_num = LAGG_PROTO_NONE 218 }, 219 { 220 .pr_num = LAGG_PROTO_ROUNDROBIN, 221 .pr_attach = lagg_rr_attach, 222 .pr_start = lagg_rr_start, 223 .pr_input = lagg_rr_input, 224 }, 225 { 226 .pr_num = LAGG_PROTO_FAILOVER, 227 .pr_start = lagg_fail_start, 228 .pr_input = lagg_fail_input, 229 }, 230 { 231 .pr_num = LAGG_PROTO_LOADBALANCE, 232 .pr_attach = lagg_lb_attach, 233 .pr_detach = lagg_lb_detach, 234 .pr_start = lagg_lb_start, 235 .pr_input = lagg_lb_input, 236 .pr_addport = lagg_lb_port_create, 237 .pr_delport = lagg_lb_port_destroy, 238 }, 239 { 240 .pr_num = LAGG_PROTO_LACP, 241 .pr_attach = lagg_lacp_attach, 242 .pr_detach = lagg_lacp_detach, 243 .pr_start = lagg_lacp_start, 244 .pr_input = lagg_lacp_input, 245 .pr_addport = lacp_port_create, 246 .pr_delport = lacp_port_destroy, 247 .pr_linkstate = lacp_linkstate, 248 .pr_init = lacp_init, 249 .pr_stop = lacp_stop, 250 .pr_lladdr = lagg_lacp_lladdr, 251 .pr_request = lacp_req, 252 .pr_portreq = lacp_portreq, 253 }, 254 { 255 .pr_num = LAGG_PROTO_BROADCAST, 256 .pr_start = lagg_bcast_start, 257 .pr_input = lagg_bcast_input, 258 }, 259 }; 260 261 SYSCTL_DECL(_net_link); 262 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 263 "Link Aggregation"); 264 265 /* Allow input on any failover links */ 266 VNET_DEFINE_STATIC(int, lagg_failover_rx_all); 267 #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all) 268 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET, 269 &VNET_NAME(lagg_failover_rx_all), 0, 270 "Accept input from any interface in a failover lagg"); 271 272 /* Default value for using flowid */ 273 VNET_DEFINE_STATIC(int, def_use_flowid) = 0; 274 #define V_def_use_flowid VNET(def_use_flowid) 275 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN, 276 &VNET_NAME(def_use_flowid), 0, 277 "Default setting for using flow id for load sharing"); 278 279 /* Default value for using numa */ 280 VNET_DEFINE_STATIC(int, def_use_numa) = 1; 281 #define V_def_use_numa VNET(def_use_numa) 282 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN, 283 &VNET_NAME(def_use_numa), 0, 284 "Use numa to steer flows"); 285 286 /* Default value for flowid shift */ 287 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16; 288 #define V_def_flowid_shift VNET(def_flowid_shift) 289 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN, 290 &VNET_NAME(def_flowid_shift), 0, 291 "Default setting for flowid shift for load sharing"); 292 293 static void 294 vnet_lagg_init(const void *unused __unused) 295 { 296 297 LAGG_LIST_LOCK_INIT(); 298 SLIST_INIT(&V_lagg_list); 299 V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create, 300 lagg_clone_destroy, 0); 301 } 302 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 303 vnet_lagg_init, NULL); 304 305 static void 306 vnet_lagg_uninit(const void *unused __unused) 307 { 308 309 if_clone_detach(V_lagg_cloner); 310 LAGG_LIST_LOCK_DESTROY(); 311 } 312 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, 313 vnet_lagg_uninit, NULL); 314 315 static int 316 lagg_modevent(module_t mod, int type, void *data) 317 { 318 319 switch (type) { 320 case MOD_LOAD: 321 lagg_input_p = lagg_input; 322 lagg_linkstate_p = lagg_port_state; 323 lagg_detach_cookie = EVENTHANDLER_REGISTER( 324 ifnet_departure_event, lagg_port_ifdetach, NULL, 325 EVENTHANDLER_PRI_ANY); 326 break; 327 case MOD_UNLOAD: 328 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 329 lagg_detach_cookie); 330 lagg_input_p = NULL; 331 lagg_linkstate_p = NULL; 332 break; 333 default: 334 return (EOPNOTSUPP); 335 } 336 return (0); 337 } 338 339 static moduledata_t lagg_mod = { 340 "if_lagg", 341 lagg_modevent, 342 0 343 }; 344 345 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 346 MODULE_VERSION(if_lagg, 1); 347 348 static void 349 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr) 350 { 351 352 LAGG_XLOCK_ASSERT(sc); 353 KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto", 354 __func__, sc)); 355 356 if (sc->sc_ifflags & IFF_DEBUG) 357 if_printf(sc->sc_ifp, "using proto %u\n", pr); 358 359 if (lagg_protos[pr].pr_attach != NULL) 360 lagg_protos[pr].pr_attach(sc); 361 sc->sc_proto = pr; 362 } 363 364 static void 365 lagg_proto_detach(struct lagg_softc *sc) 366 { 367 lagg_proto pr; 368 369 LAGG_XLOCK_ASSERT(sc); 370 pr = sc->sc_proto; 371 sc->sc_proto = LAGG_PROTO_NONE; 372 373 if (lagg_protos[pr].pr_detach != NULL) 374 lagg_protos[pr].pr_detach(sc); 375 } 376 377 static int 378 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m) 379 { 380 381 return (lagg_protos[sc->sc_proto].pr_start(sc, m)); 382 } 383 384 static struct mbuf * 385 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 386 { 387 388 return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m)); 389 } 390 391 static int 392 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp) 393 { 394 395 if (lagg_protos[sc->sc_proto].pr_addport == NULL) 396 return (0); 397 else 398 return (lagg_protos[sc->sc_proto].pr_addport(lp)); 399 } 400 401 static void 402 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp) 403 { 404 405 if (lagg_protos[sc->sc_proto].pr_delport != NULL) 406 lagg_protos[sc->sc_proto].pr_delport(lp); 407 } 408 409 static void 410 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp) 411 { 412 413 if (lagg_protos[sc->sc_proto].pr_linkstate != NULL) 414 lagg_protos[sc->sc_proto].pr_linkstate(lp); 415 } 416 417 static void 418 lagg_proto_init(struct lagg_softc *sc) 419 { 420 421 if (lagg_protos[sc->sc_proto].pr_init != NULL) 422 lagg_protos[sc->sc_proto].pr_init(sc); 423 } 424 425 static void 426 lagg_proto_stop(struct lagg_softc *sc) 427 { 428 429 if (lagg_protos[sc->sc_proto].pr_stop != NULL) 430 lagg_protos[sc->sc_proto].pr_stop(sc); 431 } 432 433 static void 434 lagg_proto_lladdr(struct lagg_softc *sc) 435 { 436 437 if (lagg_protos[sc->sc_proto].pr_lladdr != NULL) 438 lagg_protos[sc->sc_proto].pr_lladdr(sc); 439 } 440 441 static void 442 lagg_proto_request(struct lagg_softc *sc, void *v) 443 { 444 445 if (lagg_protos[sc->sc_proto].pr_request != NULL) 446 lagg_protos[sc->sc_proto].pr_request(sc, v); 447 } 448 449 static void 450 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v) 451 { 452 453 if (lagg_protos[sc->sc_proto].pr_portreq != NULL) 454 lagg_protos[sc->sc_proto].pr_portreq(lp, v); 455 } 456 457 /* 458 * This routine is run via an vlan 459 * config EVENT 460 */ 461 static void 462 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 463 { 464 struct lagg_softc *sc = ifp->if_softc; 465 struct lagg_port *lp; 466 467 if (ifp->if_softc != arg) /* Not our event */ 468 return; 469 470 LAGG_RLOCK(); 471 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 472 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 473 LAGG_RUNLOCK(); 474 } 475 476 /* 477 * This routine is run via an vlan 478 * unconfig EVENT 479 */ 480 static void 481 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 482 { 483 struct lagg_softc *sc = ifp->if_softc; 484 struct lagg_port *lp; 485 486 if (ifp->if_softc != arg) /* Not our event */ 487 return; 488 489 LAGG_RLOCK(); 490 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 491 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 492 LAGG_RUNLOCK(); 493 } 494 495 static int 496 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) 497 { 498 struct lagg_softc *sc; 499 struct ifnet *ifp; 500 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ 501 502 sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO); 503 ifp = sc->sc_ifp = if_alloc(IFT_ETHER); 504 if (ifp == NULL) { 505 free(sc, M_LAGG); 506 return (ENOSPC); 507 } 508 LAGG_SX_INIT(sc); 509 510 LAGG_XLOCK(sc); 511 if (V_def_use_flowid) 512 sc->sc_opts |= LAGG_OPT_USE_FLOWID; 513 if (V_def_use_numa) 514 sc->sc_opts |= LAGG_OPT_USE_NUMA; 515 sc->flowid_shift = V_def_flowid_shift; 516 517 /* Hash all layers by default */ 518 sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4; 519 520 lagg_proto_attach(sc, LAGG_PROTO_DEFAULT); 521 522 CK_SLIST_INIT(&sc->sc_ports); 523 524 /* Initialise pseudo media types */ 525 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 526 lagg_media_status); 527 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 528 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 529 530 if_initname(ifp, laggname, unit); 531 ifp->if_softc = sc; 532 ifp->if_transmit = lagg_transmit; 533 ifp->if_qflush = lagg_qflush; 534 ifp->if_init = lagg_init; 535 ifp->if_ioctl = lagg_ioctl; 536 ifp->if_get_counter = lagg_get_counter; 537 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 538 #if defined(KERN_TLS) || defined(RATELIMIT) 539 ifp->if_snd_tag_alloc = lagg_snd_tag_alloc; 540 ifp->if_snd_tag_modify = lagg_snd_tag_modify; 541 ifp->if_snd_tag_query = lagg_snd_tag_query; 542 ifp->if_snd_tag_free = lagg_snd_tag_free; 543 ifp->if_ratelimit_query = lagg_ratelimit_query; 544 #endif 545 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS; 546 547 /* 548 * Attach as an ordinary ethernet device, children will be attached 549 * as special device IFT_IEEE8023ADLAG. 550 */ 551 ether_ifattach(ifp, eaddr); 552 553 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 554 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 555 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 556 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 557 558 /* Insert into the global list of laggs */ 559 LAGG_LIST_LOCK(); 560 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries); 561 LAGG_LIST_UNLOCK(); 562 LAGG_XUNLOCK(sc); 563 564 return (0); 565 } 566 567 static void 568 lagg_clone_destroy(struct ifnet *ifp) 569 { 570 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 571 struct lagg_port *lp; 572 573 LAGG_XLOCK(sc); 574 sc->sc_destroying = 1; 575 lagg_stop(sc); 576 ifp->if_flags &= ~IFF_UP; 577 578 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 579 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 580 581 /* Shutdown and remove lagg ports */ 582 while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL) 583 lagg_port_destroy(lp, 1); 584 585 /* Unhook the aggregation protocol */ 586 lagg_proto_detach(sc); 587 LAGG_XUNLOCK(sc); 588 589 ifmedia_removeall(&sc->sc_media); 590 ether_ifdetach(ifp); 591 if_free(ifp); 592 593 LAGG_LIST_LOCK(); 594 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries); 595 LAGG_LIST_UNLOCK(); 596 597 LAGG_SX_DESTROY(sc); 598 free(sc, M_LAGG); 599 } 600 601 static void 602 lagg_capabilities(struct lagg_softc *sc) 603 { 604 struct lagg_port *lp; 605 int cap, ena, pena; 606 uint64_t hwa; 607 struct ifnet_hw_tsomax hw_tsomax; 608 609 LAGG_XLOCK_ASSERT(sc); 610 611 /* Get common enabled capabilities for the lagg ports */ 612 ena = ~0; 613 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 614 ena &= lp->lp_ifp->if_capenable; 615 ena = (ena == ~0 ? 0 : ena); 616 617 /* 618 * Apply common enabled capabilities back to the lagg ports. 619 * May require several iterations if they are dependent. 620 */ 621 do { 622 pena = ena; 623 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 624 lagg_setcaps(lp, ena); 625 ena &= lp->lp_ifp->if_capenable; 626 } 627 } while (pena != ena); 628 629 /* Get other capabilities from the lagg ports */ 630 cap = ~0; 631 hwa = ~(uint64_t)0; 632 memset(&hw_tsomax, 0, sizeof(hw_tsomax)); 633 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 634 cap &= lp->lp_ifp->if_capabilities; 635 hwa &= lp->lp_ifp->if_hwassist; 636 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax); 637 } 638 cap = (cap == ~0 ? 0 : cap); 639 hwa = (hwa == ~(uint64_t)0 ? 0 : hwa); 640 641 if (sc->sc_ifp->if_capabilities != cap || 642 sc->sc_ifp->if_capenable != ena || 643 sc->sc_ifp->if_hwassist != hwa || 644 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) { 645 sc->sc_ifp->if_capabilities = cap; 646 sc->sc_ifp->if_capenable = ena; 647 sc->sc_ifp->if_hwassist = hwa; 648 getmicrotime(&sc->sc_ifp->if_lastchange); 649 650 if (sc->sc_ifflags & IFF_DEBUG) 651 if_printf(sc->sc_ifp, 652 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 653 } 654 } 655 656 static int 657 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 658 { 659 struct lagg_softc *sc_ptr; 660 struct lagg_port *lp, *tlp; 661 struct ifreq ifr; 662 int error, i, oldmtu; 663 uint64_t *pval; 664 665 LAGG_XLOCK_ASSERT(sc); 666 667 if (sc->sc_ifp == ifp) { 668 if_printf(sc->sc_ifp, 669 "cannot add a lagg to itself as a port\n"); 670 return (EINVAL); 671 } 672 673 /* Limit the maximal number of lagg ports */ 674 if (sc->sc_count >= LAGG_MAX_PORTS) 675 return (ENOSPC); 676 677 /* Check if port has already been associated to a lagg */ 678 if (ifp->if_lagg != NULL) { 679 /* Port is already in the current lagg? */ 680 lp = (struct lagg_port *)ifp->if_lagg; 681 if (lp->lp_softc == sc) 682 return (EEXIST); 683 return (EBUSY); 684 } 685 686 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 687 if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN) 688 return (EPROTONOSUPPORT); 689 690 /* Allow the first Ethernet member to define the MTU */ 691 oldmtu = -1; 692 if (CK_SLIST_EMPTY(&sc->sc_ports)) { 693 sc->sc_ifp->if_mtu = ifp->if_mtu; 694 } else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 695 if (ifp->if_ioctl == NULL) { 696 if_printf(sc->sc_ifp, "cannot change MTU for %s\n", 697 ifp->if_xname); 698 return (EINVAL); 699 } 700 oldmtu = ifp->if_mtu; 701 strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name)); 702 ifr.ifr_mtu = sc->sc_ifp->if_mtu; 703 error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr); 704 if (error != 0) { 705 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 706 ifp->if_xname); 707 return (error); 708 } 709 ifr.ifr_mtu = oldmtu; 710 } 711 712 lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO); 713 lp->lp_softc = sc; 714 715 /* Check if port is a stacked lagg */ 716 LAGG_LIST_LOCK(); 717 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) { 718 if (ifp == sc_ptr->sc_ifp) { 719 LAGG_LIST_UNLOCK(); 720 free(lp, M_LAGG); 721 if (oldmtu != -1) 722 (*ifp->if_ioctl)(ifp, SIOCSIFMTU, 723 (caddr_t)&ifr); 724 return (EINVAL); 725 /* XXX disable stacking for the moment, its untested */ 726 #ifdef LAGG_PORT_STACKING 727 lp->lp_flags |= LAGG_PORT_STACK; 728 if (lagg_port_checkstacking(sc_ptr) >= 729 LAGG_MAX_STACKING) { 730 LAGG_LIST_UNLOCK(); 731 free(lp, M_LAGG); 732 if (oldmtu != -1) 733 (*ifp->if_ioctl)(ifp, SIOCSIFMTU, 734 (caddr_t)&ifr); 735 return (E2BIG); 736 } 737 #endif 738 } 739 } 740 LAGG_LIST_UNLOCK(); 741 742 if_ref(ifp); 743 lp->lp_ifp = ifp; 744 745 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 746 lp->lp_ifcapenable = ifp->if_capenable; 747 if (CK_SLIST_EMPTY(&sc->sc_ports)) { 748 bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); 749 lagg_proto_lladdr(sc); 750 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp); 751 } else { 752 if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); 753 } 754 lagg_setflags(lp, 1); 755 756 if (CK_SLIST_EMPTY(&sc->sc_ports)) 757 sc->sc_primary = lp; 758 759 /* Change the interface type */ 760 lp->lp_iftype = ifp->if_type; 761 ifp->if_type = IFT_IEEE8023ADLAG; 762 ifp->if_lagg = lp; 763 lp->lp_ioctl = ifp->if_ioctl; 764 ifp->if_ioctl = lagg_port_ioctl; 765 lp->lp_output = ifp->if_output; 766 ifp->if_output = lagg_port_output; 767 768 /* Read port counters */ 769 pval = lp->port_counters.val; 770 for (i = 0; i < IFCOUNTERS; i++, pval++) 771 *pval = ifp->if_get_counter(ifp, i); 772 773 /* 774 * Insert into the list of ports. 775 * Keep ports sorted by if_index. It is handy, when configuration 776 * is predictable and `ifconfig laggN create ...` command 777 * will lead to the same result each time. 778 */ 779 CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) { 780 if (tlp->lp_ifp->if_index < ifp->if_index && ( 781 CK_SLIST_NEXT(tlp, lp_entries) == NULL || 782 ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index > 783 ifp->if_index)) 784 break; 785 } 786 if (tlp != NULL) 787 CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries); 788 else 789 CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 790 sc->sc_count++; 791 792 lagg_setmulti(lp); 793 794 795 if ((error = lagg_proto_addport(sc, lp)) != 0) { 796 /* Remove the port, without calling pr_delport. */ 797 lagg_port_destroy(lp, 0); 798 if (oldmtu != -1) 799 (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr); 800 return (error); 801 } 802 803 /* Update lagg capabilities */ 804 lagg_capabilities(sc); 805 lagg_linkstate(sc); 806 807 return (0); 808 } 809 810 #ifdef LAGG_PORT_STACKING 811 static int 812 lagg_port_checkstacking(struct lagg_softc *sc) 813 { 814 struct lagg_softc *sc_ptr; 815 struct lagg_port *lp; 816 int m = 0; 817 818 LAGG_SXLOCK_ASSERT(sc); 819 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 820 if (lp->lp_flags & LAGG_PORT_STACK) { 821 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 822 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 823 } 824 } 825 826 return (m + 1); 827 } 828 #endif 829 830 static void 831 lagg_port_destroy_cb(epoch_context_t ec) 832 { 833 struct lagg_port *lp; 834 struct ifnet *ifp; 835 836 lp = __containerof(ec, struct lagg_port, lp_epoch_ctx); 837 ifp = lp->lp_ifp; 838 839 if_rele(ifp); 840 free(lp, M_LAGG); 841 } 842 843 static int 844 lagg_port_destroy(struct lagg_port *lp, int rundelport) 845 { 846 struct lagg_softc *sc = lp->lp_softc; 847 struct lagg_port *lp_ptr, *lp0; 848 struct ifnet *ifp = lp->lp_ifp; 849 uint64_t *pval, vdiff; 850 int i; 851 852 LAGG_XLOCK_ASSERT(sc); 853 854 if (rundelport) 855 lagg_proto_delport(sc, lp); 856 857 if (lp->lp_detaching == 0) 858 lagg_clrmulti(lp); 859 860 /* Restore interface */ 861 ifp->if_type = lp->lp_iftype; 862 ifp->if_ioctl = lp->lp_ioctl; 863 ifp->if_output = lp->lp_output; 864 ifp->if_lagg = NULL; 865 866 /* Update detached port counters */ 867 pval = lp->port_counters.val; 868 for (i = 0; i < IFCOUNTERS; i++, pval++) { 869 vdiff = ifp->if_get_counter(ifp, i) - *pval; 870 sc->detached_counters.val[i] += vdiff; 871 } 872 873 /* Finally, remove the port from the lagg */ 874 CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 875 sc->sc_count--; 876 877 /* Update the primary interface */ 878 if (lp == sc->sc_primary) { 879 uint8_t lladdr[ETHER_ADDR_LEN]; 880 881 if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL) 882 bzero(&lladdr, ETHER_ADDR_LEN); 883 else 884 bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN); 885 sc->sc_primary = lp0; 886 if (sc->sc_destroying == 0) { 887 bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); 888 lagg_proto_lladdr(sc); 889 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp); 890 } 891 892 /* 893 * Update lladdr for each port (new primary needs update 894 * as well, to switch from old lladdr to its 'real' one) 895 */ 896 CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 897 if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN); 898 } 899 900 if (lp->lp_ifflags) 901 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 902 903 if (lp->lp_detaching == 0) { 904 lagg_setflags(lp, 0); 905 lagg_setcaps(lp, lp->lp_ifcapenable); 906 if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN); 907 } 908 909 /* 910 * free port and release it's ifnet reference after a grace period has 911 * elapsed. 912 */ 913 epoch_call(net_epoch_preempt, &lp->lp_epoch_ctx, lagg_port_destroy_cb); 914 /* Update lagg capabilities */ 915 lagg_capabilities(sc); 916 lagg_linkstate(sc); 917 918 return (0); 919 } 920 921 static int 922 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 923 { 924 struct lagg_reqport *rp = (struct lagg_reqport *)data; 925 struct lagg_softc *sc; 926 struct lagg_port *lp = NULL; 927 int error = 0; 928 929 /* Should be checked by the caller */ 930 if (ifp->if_type != IFT_IEEE8023ADLAG || 931 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 932 goto fallback; 933 934 switch (cmd) { 935 case SIOCGLAGGPORT: 936 if (rp->rp_portname[0] == '\0' || 937 ifunit(rp->rp_portname) != ifp) { 938 error = EINVAL; 939 break; 940 } 941 942 LAGG_RLOCK(); 943 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 944 error = ENOENT; 945 LAGG_RUNLOCK(); 946 break; 947 } 948 949 lagg_port2req(lp, rp); 950 LAGG_RUNLOCK(); 951 break; 952 953 case SIOCSIFCAP: 954 if (lp->lp_ioctl == NULL) { 955 error = EINVAL; 956 break; 957 } 958 error = (*lp->lp_ioctl)(ifp, cmd, data); 959 if (error) 960 break; 961 962 /* Update lagg interface capabilities */ 963 LAGG_XLOCK(sc); 964 lagg_capabilities(sc); 965 LAGG_XUNLOCK(sc); 966 VLAN_CAPABILITIES(sc->sc_ifp); 967 break; 968 969 case SIOCSIFMTU: 970 /* Do not allow the MTU to be changed once joined */ 971 error = EINVAL; 972 break; 973 974 default: 975 goto fallback; 976 } 977 978 return (error); 979 980 fallback: 981 if (lp != NULL && lp->lp_ioctl != NULL) 982 return ((*lp->lp_ioctl)(ifp, cmd, data)); 983 984 return (EINVAL); 985 } 986 987 /* 988 * Requests counter @cnt data. 989 * 990 * Counter value is calculated the following way: 991 * 1) for each port, sum difference between current and "initial" measurements. 992 * 2) add lagg logical interface counters. 993 * 3) add data from detached_counters array. 994 * 995 * We also do the following things on ports attach/detach: 996 * 1) On port attach we store all counters it has into port_counter array. 997 * 2) On port detach we add the different between "initial" and 998 * current counters data to detached_counters array. 999 */ 1000 static uint64_t 1001 lagg_get_counter(struct ifnet *ifp, ift_counter cnt) 1002 { 1003 struct lagg_softc *sc; 1004 struct lagg_port *lp; 1005 struct ifnet *lpifp; 1006 uint64_t newval, oldval, vsum; 1007 1008 /* Revise this when we've got non-generic counters. */ 1009 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt)); 1010 1011 sc = (struct lagg_softc *)ifp->if_softc; 1012 1013 vsum = 0; 1014 LAGG_RLOCK(); 1015 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1016 /* Saved attached value */ 1017 oldval = lp->port_counters.val[cnt]; 1018 /* current value */ 1019 lpifp = lp->lp_ifp; 1020 newval = lpifp->if_get_counter(lpifp, cnt); 1021 /* Calculate diff and save new */ 1022 vsum += newval - oldval; 1023 } 1024 LAGG_RUNLOCK(); 1025 1026 /* 1027 * Add counter data which might be added by upper 1028 * layer protocols operating on logical interface. 1029 */ 1030 vsum += if_get_counter_default(ifp, cnt); 1031 1032 /* 1033 * Add counter data from detached ports counters 1034 */ 1035 vsum += sc->detached_counters.val[cnt]; 1036 1037 1038 return (vsum); 1039 } 1040 1041 /* 1042 * For direct output to child ports. 1043 */ 1044 static int 1045 lagg_port_output(struct ifnet *ifp, struct mbuf *m, 1046 const struct sockaddr *dst, struct route *ro) 1047 { 1048 struct lagg_port *lp = ifp->if_lagg; 1049 1050 switch (dst->sa_family) { 1051 case pseudo_AF_HDRCMPLT: 1052 case AF_UNSPEC: 1053 return ((*lp->lp_output)(ifp, m, dst, ro)); 1054 } 1055 1056 /* drop any other frames */ 1057 m_freem(m); 1058 return (ENETDOWN); 1059 } 1060 1061 static void 1062 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 1063 { 1064 struct lagg_port *lp; 1065 struct lagg_softc *sc; 1066 1067 if ((lp = ifp->if_lagg) == NULL) 1068 return; 1069 /* If the ifnet is just being renamed, don't do anything. */ 1070 if (ifp->if_flags & IFF_RENAMING) 1071 return; 1072 1073 sc = lp->lp_softc; 1074 1075 LAGG_XLOCK(sc); 1076 lp->lp_detaching = 1; 1077 lagg_port_destroy(lp, 1); 1078 LAGG_XUNLOCK(sc); 1079 VLAN_CAPABILITIES(sc->sc_ifp); 1080 } 1081 1082 static void 1083 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 1084 { 1085 struct lagg_softc *sc = lp->lp_softc; 1086 1087 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 1088 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 1089 rp->rp_prio = lp->lp_prio; 1090 rp->rp_flags = lp->lp_flags; 1091 lagg_proto_portreq(sc, lp, &rp->rp_psc); 1092 1093 /* Add protocol specific flags */ 1094 switch (sc->sc_proto) { 1095 case LAGG_PROTO_FAILOVER: 1096 if (lp == sc->sc_primary) 1097 rp->rp_flags |= LAGG_PORT_MASTER; 1098 if (lp == lagg_link_active(sc, sc->sc_primary)) 1099 rp->rp_flags |= LAGG_PORT_ACTIVE; 1100 break; 1101 1102 case LAGG_PROTO_ROUNDROBIN: 1103 case LAGG_PROTO_LOADBALANCE: 1104 case LAGG_PROTO_BROADCAST: 1105 if (LAGG_PORTACTIVE(lp)) 1106 rp->rp_flags |= LAGG_PORT_ACTIVE; 1107 break; 1108 1109 case LAGG_PROTO_LACP: 1110 /* LACP has a different definition of active */ 1111 if (lacp_isactive(lp)) 1112 rp->rp_flags |= LAGG_PORT_ACTIVE; 1113 if (lacp_iscollecting(lp)) 1114 rp->rp_flags |= LAGG_PORT_COLLECTING; 1115 if (lacp_isdistributing(lp)) 1116 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 1117 break; 1118 } 1119 1120 } 1121 1122 static void 1123 lagg_init(void *xsc) 1124 { 1125 struct lagg_softc *sc = (struct lagg_softc *)xsc; 1126 struct ifnet *ifp = sc->sc_ifp; 1127 struct lagg_port *lp; 1128 1129 LAGG_XLOCK(sc); 1130 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1131 LAGG_XUNLOCK(sc); 1132 return; 1133 } 1134 1135 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1136 1137 /* 1138 * Update the port lladdrs if needed. 1139 * This might be if_setlladdr() notification 1140 * that lladdr has been changed. 1141 */ 1142 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1143 if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp), 1144 ETHER_ADDR_LEN) != 0) 1145 if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN); 1146 } 1147 1148 lagg_proto_init(sc); 1149 1150 LAGG_XUNLOCK(sc); 1151 } 1152 1153 static void 1154 lagg_stop(struct lagg_softc *sc) 1155 { 1156 struct ifnet *ifp = sc->sc_ifp; 1157 1158 LAGG_XLOCK_ASSERT(sc); 1159 1160 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 1161 return; 1162 1163 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1164 1165 lagg_proto_stop(sc); 1166 } 1167 1168 static int 1169 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1170 { 1171 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1172 struct lagg_reqall *ra = (struct lagg_reqall *)data; 1173 struct lagg_reqopts *ro = (struct lagg_reqopts *)data; 1174 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 1175 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 1176 struct ifreq *ifr = (struct ifreq *)data; 1177 struct lagg_port *lp; 1178 struct ifnet *tpif; 1179 struct thread *td = curthread; 1180 char *buf, *outbuf; 1181 int count, buflen, len, error = 0; 1182 1183 bzero(&rpbuf, sizeof(rpbuf)); 1184 1185 switch (cmd) { 1186 case SIOCGLAGG: 1187 LAGG_XLOCK(sc); 1188 buflen = sc->sc_count * sizeof(struct lagg_reqport); 1189 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 1190 ra->ra_proto = sc->sc_proto; 1191 lagg_proto_request(sc, &ra->ra_psc); 1192 count = 0; 1193 buf = outbuf; 1194 len = min(ra->ra_size, buflen); 1195 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1196 if (len < sizeof(rpbuf)) 1197 break; 1198 1199 lagg_port2req(lp, &rpbuf); 1200 memcpy(buf, &rpbuf, sizeof(rpbuf)); 1201 count++; 1202 buf += sizeof(rpbuf); 1203 len -= sizeof(rpbuf); 1204 } 1205 LAGG_XUNLOCK(sc); 1206 ra->ra_ports = count; 1207 ra->ra_size = count * sizeof(rpbuf); 1208 error = copyout(outbuf, ra->ra_port, ra->ra_size); 1209 free(outbuf, M_TEMP); 1210 break; 1211 case SIOCSLAGG: 1212 error = priv_check(td, PRIV_NET_LAGG); 1213 if (error) 1214 break; 1215 if (ra->ra_proto >= LAGG_PROTO_MAX) { 1216 error = EPROTONOSUPPORT; 1217 break; 1218 } 1219 1220 LAGG_XLOCK(sc); 1221 lagg_proto_detach(sc); 1222 LAGG_UNLOCK_ASSERT(); 1223 lagg_proto_attach(sc, ra->ra_proto); 1224 LAGG_XUNLOCK(sc); 1225 break; 1226 case SIOCGLAGGOPTS: 1227 LAGG_XLOCK(sc); 1228 ro->ro_opts = sc->sc_opts; 1229 if (sc->sc_proto == LAGG_PROTO_LACP) { 1230 struct lacp_softc *lsc; 1231 1232 lsc = (struct lacp_softc *)sc->sc_psc; 1233 if (lsc->lsc_debug.lsc_tx_test != 0) 1234 ro->ro_opts |= LAGG_OPT_LACP_TXTEST; 1235 if (lsc->lsc_debug.lsc_rx_test != 0) 1236 ro->ro_opts |= LAGG_OPT_LACP_RXTEST; 1237 if (lsc->lsc_strict_mode != 0) 1238 ro->ro_opts |= LAGG_OPT_LACP_STRICT; 1239 if (lsc->lsc_fast_timeout != 0) 1240 ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT; 1241 1242 ro->ro_active = sc->sc_active; 1243 } else { 1244 ro->ro_active = 0; 1245 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1246 ro->ro_active += LAGG_PORTACTIVE(lp); 1247 } 1248 ro->ro_bkt = sc->sc_stride; 1249 ro->ro_flapping = sc->sc_flapping; 1250 ro->ro_flowid_shift = sc->flowid_shift; 1251 LAGG_XUNLOCK(sc); 1252 break; 1253 case SIOCSLAGGOPTS: 1254 error = priv_check(td, PRIV_NET_LAGG); 1255 if (error) 1256 break; 1257 1258 /* 1259 * The stride option was added without defining a corresponding 1260 * LAGG_OPT flag, so we must handle it before processing any 1261 * remaining options. 1262 */ 1263 LAGG_XLOCK(sc); 1264 if (ro->ro_bkt != 0) { 1265 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) { 1266 LAGG_XUNLOCK(sc); 1267 error = EINVAL; 1268 break; 1269 } 1270 sc->sc_stride = ro->ro_bkt; 1271 } else { 1272 sc->sc_stride = 0; 1273 } 1274 1275 if (ro->ro_opts == 0) { 1276 LAGG_XUNLOCK(sc); 1277 break; 1278 } 1279 1280 /* 1281 * Set options. LACP options are stored in sc->sc_psc, 1282 * not in sc_opts. 1283 */ 1284 int valid, lacp; 1285 1286 switch (ro->ro_opts) { 1287 case LAGG_OPT_USE_FLOWID: 1288 case -LAGG_OPT_USE_FLOWID: 1289 case LAGG_OPT_USE_NUMA: 1290 case -LAGG_OPT_USE_NUMA: 1291 case LAGG_OPT_FLOWIDSHIFT: 1292 valid = 1; 1293 lacp = 0; 1294 break; 1295 case LAGG_OPT_LACP_TXTEST: 1296 case -LAGG_OPT_LACP_TXTEST: 1297 case LAGG_OPT_LACP_RXTEST: 1298 case -LAGG_OPT_LACP_RXTEST: 1299 case LAGG_OPT_LACP_STRICT: 1300 case -LAGG_OPT_LACP_STRICT: 1301 case LAGG_OPT_LACP_TIMEOUT: 1302 case -LAGG_OPT_LACP_TIMEOUT: 1303 valid = lacp = 1; 1304 break; 1305 default: 1306 valid = lacp = 0; 1307 break; 1308 } 1309 1310 if (valid == 0 || 1311 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) { 1312 /* Invalid combination of options specified. */ 1313 error = EINVAL; 1314 LAGG_XUNLOCK(sc); 1315 break; /* Return from SIOCSLAGGOPTS. */ 1316 } 1317 /* 1318 * Store new options into sc->sc_opts except for 1319 * FLOWIDSHIFT and LACP options. 1320 */ 1321 if (lacp == 0) { 1322 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT) 1323 sc->flowid_shift = ro->ro_flowid_shift; 1324 else if (ro->ro_opts > 0) 1325 sc->sc_opts |= ro->ro_opts; 1326 else 1327 sc->sc_opts &= ~ro->ro_opts; 1328 } else { 1329 struct lacp_softc *lsc; 1330 struct lacp_port *lp; 1331 1332 lsc = (struct lacp_softc *)sc->sc_psc; 1333 1334 switch (ro->ro_opts) { 1335 case LAGG_OPT_LACP_TXTEST: 1336 lsc->lsc_debug.lsc_tx_test = 1; 1337 break; 1338 case -LAGG_OPT_LACP_TXTEST: 1339 lsc->lsc_debug.lsc_tx_test = 0; 1340 break; 1341 case LAGG_OPT_LACP_RXTEST: 1342 lsc->lsc_debug.lsc_rx_test = 1; 1343 break; 1344 case -LAGG_OPT_LACP_RXTEST: 1345 lsc->lsc_debug.lsc_rx_test = 0; 1346 break; 1347 case LAGG_OPT_LACP_STRICT: 1348 lsc->lsc_strict_mode = 1; 1349 break; 1350 case -LAGG_OPT_LACP_STRICT: 1351 lsc->lsc_strict_mode = 0; 1352 break; 1353 case LAGG_OPT_LACP_TIMEOUT: 1354 LACP_LOCK(lsc); 1355 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1356 lp->lp_state |= LACP_STATE_TIMEOUT; 1357 LACP_UNLOCK(lsc); 1358 lsc->lsc_fast_timeout = 1; 1359 break; 1360 case -LAGG_OPT_LACP_TIMEOUT: 1361 LACP_LOCK(lsc); 1362 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1363 lp->lp_state &= ~LACP_STATE_TIMEOUT; 1364 LACP_UNLOCK(lsc); 1365 lsc->lsc_fast_timeout = 0; 1366 break; 1367 } 1368 } 1369 LAGG_XUNLOCK(sc); 1370 break; 1371 case SIOCGLAGGFLAGS: 1372 rf->rf_flags = 0; 1373 LAGG_XLOCK(sc); 1374 if (sc->sc_flags & MBUF_HASHFLAG_L2) 1375 rf->rf_flags |= LAGG_F_HASHL2; 1376 if (sc->sc_flags & MBUF_HASHFLAG_L3) 1377 rf->rf_flags |= LAGG_F_HASHL3; 1378 if (sc->sc_flags & MBUF_HASHFLAG_L4) 1379 rf->rf_flags |= LAGG_F_HASHL4; 1380 LAGG_XUNLOCK(sc); 1381 break; 1382 case SIOCSLAGGHASH: 1383 error = priv_check(td, PRIV_NET_LAGG); 1384 if (error) 1385 break; 1386 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1387 error = EINVAL; 1388 break; 1389 } 1390 LAGG_XLOCK(sc); 1391 sc->sc_flags = 0; 1392 if (rf->rf_flags & LAGG_F_HASHL2) 1393 sc->sc_flags |= MBUF_HASHFLAG_L2; 1394 if (rf->rf_flags & LAGG_F_HASHL3) 1395 sc->sc_flags |= MBUF_HASHFLAG_L3; 1396 if (rf->rf_flags & LAGG_F_HASHL4) 1397 sc->sc_flags |= MBUF_HASHFLAG_L4; 1398 LAGG_XUNLOCK(sc); 1399 break; 1400 case SIOCGLAGGPORT: 1401 if (rp->rp_portname[0] == '\0' || 1402 (tpif = ifunit_ref(rp->rp_portname)) == NULL) { 1403 error = EINVAL; 1404 break; 1405 } 1406 1407 LAGG_RLOCK(); 1408 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1409 lp->lp_softc != sc) { 1410 error = ENOENT; 1411 LAGG_RUNLOCK(); 1412 if_rele(tpif); 1413 break; 1414 } 1415 1416 lagg_port2req(lp, rp); 1417 LAGG_RUNLOCK(); 1418 if_rele(tpif); 1419 break; 1420 case SIOCSLAGGPORT: 1421 error = priv_check(td, PRIV_NET_LAGG); 1422 if (error) 1423 break; 1424 if (rp->rp_portname[0] == '\0' || 1425 (tpif = ifunit_ref(rp->rp_portname)) == NULL) { 1426 error = EINVAL; 1427 break; 1428 } 1429 #ifdef INET6 1430 /* 1431 * A laggport interface should not have inet6 address 1432 * because two interfaces with a valid link-local 1433 * scope zone must not be merged in any form. This 1434 * restriction is needed to prevent violation of 1435 * link-local scope zone. Attempts to add a laggport 1436 * interface which has inet6 addresses triggers 1437 * removal of all inet6 addresses on the member 1438 * interface. 1439 */ 1440 if (in6ifa_llaonifp(tpif)) { 1441 in6_ifdetach(tpif); 1442 if_printf(sc->sc_ifp, 1443 "IPv6 addresses on %s have been removed " 1444 "before adding it as a member to prevent " 1445 "IPv6 address scope violation.\n", 1446 tpif->if_xname); 1447 } 1448 #endif 1449 LAGG_XLOCK(sc); 1450 error = lagg_port_create(sc, tpif); 1451 LAGG_XUNLOCK(sc); 1452 if_rele(tpif); 1453 VLAN_CAPABILITIES(ifp); 1454 break; 1455 case SIOCSLAGGDELPORT: 1456 error = priv_check(td, PRIV_NET_LAGG); 1457 if (error) 1458 break; 1459 if (rp->rp_portname[0] == '\0' || 1460 (tpif = ifunit_ref(rp->rp_portname)) == NULL) { 1461 error = EINVAL; 1462 break; 1463 } 1464 1465 LAGG_XLOCK(sc); 1466 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1467 lp->lp_softc != sc) { 1468 error = ENOENT; 1469 LAGG_XUNLOCK(sc); 1470 if_rele(tpif); 1471 break; 1472 } 1473 1474 error = lagg_port_destroy(lp, 1); 1475 LAGG_XUNLOCK(sc); 1476 if_rele(tpif); 1477 VLAN_CAPABILITIES(ifp); 1478 break; 1479 case SIOCSIFFLAGS: 1480 /* Set flags on ports too */ 1481 LAGG_XLOCK(sc); 1482 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1483 lagg_setflags(lp, 1); 1484 } 1485 1486 if (!(ifp->if_flags & IFF_UP) && 1487 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1488 /* 1489 * If interface is marked down and it is running, 1490 * then stop and disable it. 1491 */ 1492 lagg_stop(sc); 1493 LAGG_XUNLOCK(sc); 1494 } else if ((ifp->if_flags & IFF_UP) && 1495 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1496 /* 1497 * If interface is marked up and it is stopped, then 1498 * start it. 1499 */ 1500 LAGG_XUNLOCK(sc); 1501 (*ifp->if_init)(sc); 1502 } else 1503 LAGG_XUNLOCK(sc); 1504 break; 1505 case SIOCADDMULTI: 1506 case SIOCDELMULTI: 1507 LAGG_XLOCK(sc); 1508 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1509 lagg_clrmulti(lp); 1510 lagg_setmulti(lp); 1511 } 1512 LAGG_XUNLOCK(sc); 1513 error = 0; 1514 break; 1515 case SIOCSIFMEDIA: 1516 case SIOCGIFMEDIA: 1517 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1518 break; 1519 1520 case SIOCSIFCAP: 1521 LAGG_XLOCK(sc); 1522 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1523 if (lp->lp_ioctl != NULL) 1524 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); 1525 } 1526 lagg_capabilities(sc); 1527 LAGG_XUNLOCK(sc); 1528 VLAN_CAPABILITIES(ifp); 1529 error = 0; 1530 break; 1531 1532 case SIOCSIFMTU: 1533 LAGG_XLOCK(sc); 1534 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1535 if (lp->lp_ioctl != NULL) 1536 error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); 1537 else 1538 error = EINVAL; 1539 if (error != 0) { 1540 if_printf(ifp, 1541 "failed to change MTU to %d on port %s, " 1542 "reverting all ports to original MTU (%d)\n", 1543 ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu); 1544 break; 1545 } 1546 } 1547 if (error == 0) { 1548 ifp->if_mtu = ifr->ifr_mtu; 1549 } else { 1550 /* set every port back to the original MTU */ 1551 ifr->ifr_mtu = ifp->if_mtu; 1552 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1553 if (lp->lp_ioctl != NULL) 1554 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); 1555 } 1556 } 1557 LAGG_XUNLOCK(sc); 1558 break; 1559 1560 default: 1561 error = ether_ioctl(ifp, cmd, data); 1562 break; 1563 } 1564 return (error); 1565 } 1566 1567 #if defined(KERN_TLS) || defined(RATELIMIT) 1568 static inline struct lagg_snd_tag * 1569 mst_to_lst(struct m_snd_tag *mst) 1570 { 1571 1572 return (__containerof(mst, struct lagg_snd_tag, com)); 1573 } 1574 1575 /* 1576 * Look up the port used by a specific flow. This only works for lagg 1577 * protocols with deterministic port mappings (e.g. not roundrobin). 1578 * In addition protocols which use a hash to map flows to ports must 1579 * be configured to use the mbuf flowid rather than hashing packet 1580 * contents. 1581 */ 1582 static struct lagg_port * 1583 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype) 1584 { 1585 struct lagg_softc *sc; 1586 struct lagg_port *lp; 1587 struct lagg_lb *lb; 1588 uint32_t p; 1589 1590 sc = ifp->if_softc; 1591 1592 switch (sc->sc_proto) { 1593 case LAGG_PROTO_FAILOVER: 1594 return (lagg_link_active(sc, sc->sc_primary)); 1595 case LAGG_PROTO_LOADBALANCE: 1596 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 || 1597 flowtype == M_HASHTYPE_NONE) 1598 return (NULL); 1599 p = flowid >> sc->flowid_shift; 1600 p %= sc->sc_count; 1601 lb = (struct lagg_lb *)sc->sc_psc; 1602 lp = lb->lb_ports[p]; 1603 return (lagg_link_active(sc, lp)); 1604 case LAGG_PROTO_LACP: 1605 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 || 1606 flowtype == M_HASHTYPE_NONE) 1607 return (NULL); 1608 return (lacp_select_tx_port_by_hash(sc, flowid)); 1609 default: 1610 return (NULL); 1611 } 1612 } 1613 1614 static int 1615 lagg_snd_tag_alloc(struct ifnet *ifp, 1616 union if_snd_tag_alloc_params *params, 1617 struct m_snd_tag **ppmt) 1618 { 1619 struct lagg_snd_tag *lst; 1620 struct lagg_softc *sc; 1621 struct lagg_port *lp; 1622 struct ifnet *lp_ifp; 1623 int error; 1624 1625 sc = ifp->if_softc; 1626 1627 LAGG_RLOCK(); 1628 lp = lookup_snd_tag_port(ifp, params->hdr.flowid, params->hdr.flowtype); 1629 if (lp == NULL) { 1630 LAGG_RUNLOCK(); 1631 return (EOPNOTSUPP); 1632 } 1633 if (lp->lp_ifp == NULL || lp->lp_ifp->if_snd_tag_alloc == NULL) { 1634 LAGG_RUNLOCK(); 1635 return (EOPNOTSUPP); 1636 } 1637 lp_ifp = lp->lp_ifp; 1638 if_ref(lp_ifp); 1639 LAGG_RUNLOCK(); 1640 1641 lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT); 1642 if (lst == NULL) { 1643 if_rele(lp_ifp); 1644 return (ENOMEM); 1645 } 1646 1647 error = lp_ifp->if_snd_tag_alloc(lp_ifp, params, &lst->tag); 1648 if_rele(lp_ifp); 1649 if (error) { 1650 free(lst, M_LAGG); 1651 return (error); 1652 } 1653 1654 m_snd_tag_init(&lst->com, ifp); 1655 1656 *ppmt = &lst->com; 1657 return (0); 1658 } 1659 1660 static int 1661 lagg_snd_tag_modify(struct m_snd_tag *mst, 1662 union if_snd_tag_modify_params *params) 1663 { 1664 struct lagg_snd_tag *lst; 1665 1666 lst = mst_to_lst(mst); 1667 return (lst->tag->ifp->if_snd_tag_modify(lst->tag, params)); 1668 } 1669 1670 static int 1671 lagg_snd_tag_query(struct m_snd_tag *mst, 1672 union if_snd_tag_query_params *params) 1673 { 1674 struct lagg_snd_tag *lst; 1675 1676 lst = mst_to_lst(mst); 1677 return (lst->tag->ifp->if_snd_tag_query(lst->tag, params)); 1678 } 1679 1680 static void 1681 lagg_snd_tag_free(struct m_snd_tag *mst) 1682 { 1683 struct lagg_snd_tag *lst; 1684 1685 lst = mst_to_lst(mst); 1686 m_snd_tag_rele(lst->tag); 1687 free(lst, M_LAGG); 1688 } 1689 1690 static void 1691 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q) 1692 { 1693 /* 1694 * For lagg, we have an indirect 1695 * interface. The caller needs to 1696 * get a ratelimit tag on the actual 1697 * interface the flow will go on. 1698 */ 1699 q->rate_table = NULL; 1700 q->flags = RT_IS_INDIRECT; 1701 q->max_flows = 0; 1702 q->number_of_rates = 0; 1703 } 1704 #endif 1705 1706 static int 1707 lagg_setmulti(struct lagg_port *lp) 1708 { 1709 struct lagg_softc *sc = lp->lp_softc; 1710 struct ifnet *ifp = lp->lp_ifp; 1711 struct ifnet *scifp = sc->sc_ifp; 1712 struct lagg_mc *mc; 1713 struct ifmultiaddr *ifma; 1714 int error; 1715 1716 IF_ADDR_WLOCK(scifp); 1717 CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1718 if (ifma->ifma_addr->sa_family != AF_LINK) 1719 continue; 1720 mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT); 1721 if (mc == NULL) { 1722 IF_ADDR_WUNLOCK(scifp); 1723 return (ENOMEM); 1724 } 1725 bcopy(ifma->ifma_addr, &mc->mc_addr, 1726 ifma->ifma_addr->sa_len); 1727 mc->mc_addr.sdl_index = ifp->if_index; 1728 mc->mc_ifma = NULL; 1729 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1730 } 1731 IF_ADDR_WUNLOCK(scifp); 1732 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) { 1733 error = if_addmulti(ifp, 1734 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma); 1735 if (error) 1736 return (error); 1737 } 1738 return (0); 1739 } 1740 1741 static int 1742 lagg_clrmulti(struct lagg_port *lp) 1743 { 1744 struct lagg_mc *mc; 1745 1746 LAGG_XLOCK_ASSERT(lp->lp_softc); 1747 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1748 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1749 if (mc->mc_ifma && lp->lp_detaching == 0) 1750 if_delmulti_ifma(mc->mc_ifma); 1751 free(mc, M_LAGG); 1752 } 1753 return (0); 1754 } 1755 1756 static int 1757 lagg_setcaps(struct lagg_port *lp, int cap) 1758 { 1759 struct ifreq ifr; 1760 1761 if (lp->lp_ifp->if_capenable == cap) 1762 return (0); 1763 if (lp->lp_ioctl == NULL) 1764 return (ENXIO); 1765 ifr.ifr_reqcap = cap; 1766 return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr)); 1767 } 1768 1769 /* Handle a ref counted flag that should be set on the lagg port as well */ 1770 static int 1771 lagg_setflag(struct lagg_port *lp, int flag, int status, 1772 int (*func)(struct ifnet *, int)) 1773 { 1774 struct lagg_softc *sc = lp->lp_softc; 1775 struct ifnet *scifp = sc->sc_ifp; 1776 struct ifnet *ifp = lp->lp_ifp; 1777 int error; 1778 1779 LAGG_XLOCK_ASSERT(sc); 1780 1781 status = status ? (scifp->if_flags & flag) : 0; 1782 /* Now "status" contains the flag value or 0 */ 1783 1784 /* 1785 * See if recorded ports status is different from what 1786 * we want it to be. If it is, flip it. We record ports 1787 * status in lp_ifflags so that we won't clear ports flag 1788 * we haven't set. In fact, we don't clear or set ports 1789 * flags directly, but get or release references to them. 1790 * That's why we can be sure that recorded flags still are 1791 * in accord with actual ports flags. 1792 */ 1793 if (status != (lp->lp_ifflags & flag)) { 1794 error = (*func)(ifp, status); 1795 if (error) 1796 return (error); 1797 lp->lp_ifflags &= ~flag; 1798 lp->lp_ifflags |= status; 1799 } 1800 return (0); 1801 } 1802 1803 /* 1804 * Handle IFF_* flags that require certain changes on the lagg port 1805 * if "status" is true, update ports flags respective to the lagg 1806 * if "status" is false, forcedly clear the flags set on port. 1807 */ 1808 static int 1809 lagg_setflags(struct lagg_port *lp, int status) 1810 { 1811 int error, i; 1812 1813 for (i = 0; lagg_pflags[i].flag; i++) { 1814 error = lagg_setflag(lp, lagg_pflags[i].flag, 1815 status, lagg_pflags[i].func); 1816 if (error) 1817 return (error); 1818 } 1819 return (0); 1820 } 1821 1822 static int 1823 lagg_transmit(struct ifnet *ifp, struct mbuf *m) 1824 { 1825 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1826 int error; 1827 1828 #if defined(KERN_TLS) || defined(RATELIMIT) 1829 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) 1830 MPASS(m->m_pkthdr.snd_tag->ifp == ifp); 1831 #endif 1832 LAGG_RLOCK(); 1833 /* We need a Tx algorithm and at least one port */ 1834 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1835 LAGG_RUNLOCK(); 1836 m_freem(m); 1837 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1838 return (ENXIO); 1839 } 1840 1841 ETHER_BPF_MTAP(ifp, m); 1842 1843 error = lagg_proto_start(sc, m); 1844 LAGG_RUNLOCK(); 1845 1846 if (error != 0) 1847 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1848 1849 return (error); 1850 } 1851 1852 /* 1853 * The ifp->if_qflush entry point for lagg(4) is no-op. 1854 */ 1855 static void 1856 lagg_qflush(struct ifnet *ifp __unused) 1857 { 1858 } 1859 1860 static struct mbuf * 1861 lagg_input(struct ifnet *ifp, struct mbuf *m) 1862 { 1863 struct lagg_port *lp = ifp->if_lagg; 1864 struct lagg_softc *sc = lp->lp_softc; 1865 struct ifnet *scifp = sc->sc_ifp; 1866 1867 LAGG_RLOCK(); 1868 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1869 lp->lp_detaching != 0 || 1870 sc->sc_proto == LAGG_PROTO_NONE) { 1871 LAGG_RUNLOCK(); 1872 m_freem(m); 1873 return (NULL); 1874 } 1875 1876 ETHER_BPF_MTAP(scifp, m); 1877 1878 m = lagg_proto_input(sc, lp, m); 1879 if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) { 1880 m_freem(m); 1881 m = NULL; 1882 } 1883 1884 LAGG_RUNLOCK(); 1885 return (m); 1886 } 1887 1888 static int 1889 lagg_media_change(struct ifnet *ifp) 1890 { 1891 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1892 1893 if (sc->sc_ifflags & IFF_DEBUG) 1894 printf("%s\n", __func__); 1895 1896 /* Ignore */ 1897 return (0); 1898 } 1899 1900 static void 1901 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1902 { 1903 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1904 struct lagg_port *lp; 1905 1906 imr->ifm_status = IFM_AVALID; 1907 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1908 1909 LAGG_RLOCK(); 1910 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1911 if (LAGG_PORTACTIVE(lp)) 1912 imr->ifm_status |= IFM_ACTIVE; 1913 } 1914 LAGG_RUNLOCK(); 1915 } 1916 1917 static void 1918 lagg_linkstate(struct lagg_softc *sc) 1919 { 1920 struct lagg_port *lp; 1921 int new_link = LINK_STATE_DOWN; 1922 uint64_t speed; 1923 1924 LAGG_XLOCK_ASSERT(sc); 1925 1926 /* LACP handles link state itself */ 1927 if (sc->sc_proto == LAGG_PROTO_LACP) 1928 return; 1929 1930 /* Our link is considered up if at least one of our ports is active */ 1931 LAGG_RLOCK(); 1932 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1933 if (lp->lp_ifp->if_link_state == LINK_STATE_UP) { 1934 new_link = LINK_STATE_UP; 1935 break; 1936 } 1937 } 1938 LAGG_RUNLOCK(); 1939 if_link_state_change(sc->sc_ifp, new_link); 1940 1941 /* Update if_baudrate to reflect the max possible speed */ 1942 switch (sc->sc_proto) { 1943 case LAGG_PROTO_FAILOVER: 1944 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1945 sc->sc_primary->lp_ifp->if_baudrate : 0; 1946 break; 1947 case LAGG_PROTO_ROUNDROBIN: 1948 case LAGG_PROTO_LOADBALANCE: 1949 case LAGG_PROTO_BROADCAST: 1950 speed = 0; 1951 LAGG_RLOCK(); 1952 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1953 speed += lp->lp_ifp->if_baudrate; 1954 LAGG_RUNLOCK(); 1955 sc->sc_ifp->if_baudrate = speed; 1956 break; 1957 case LAGG_PROTO_LACP: 1958 /* LACP updates if_baudrate itself */ 1959 break; 1960 } 1961 } 1962 1963 static void 1964 lagg_port_state(struct ifnet *ifp, int state) 1965 { 1966 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1967 struct lagg_softc *sc = NULL; 1968 1969 if (lp != NULL) 1970 sc = lp->lp_softc; 1971 if (sc == NULL) 1972 return; 1973 1974 LAGG_XLOCK(sc); 1975 lagg_linkstate(sc); 1976 lagg_proto_linkstate(sc, lp); 1977 LAGG_XUNLOCK(sc); 1978 } 1979 1980 struct lagg_port * 1981 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1982 { 1983 struct lagg_port *lp_next, *rval = NULL; 1984 1985 /* 1986 * Search a port which reports an active link state. 1987 */ 1988 1989 #ifdef INVARIANTS 1990 /* 1991 * This is called with either LAGG_RLOCK() held or 1992 * LAGG_XLOCK(sc) held. 1993 */ 1994 if (!in_epoch(net_epoch_preempt)) 1995 LAGG_XLOCK_ASSERT(sc); 1996 #endif 1997 1998 if (lp == NULL) 1999 goto search; 2000 if (LAGG_PORTACTIVE(lp)) { 2001 rval = lp; 2002 goto found; 2003 } 2004 if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL && 2005 LAGG_PORTACTIVE(lp_next)) { 2006 rval = lp_next; 2007 goto found; 2008 } 2009 2010 search: 2011 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 2012 if (LAGG_PORTACTIVE(lp_next)) { 2013 return (lp_next); 2014 } 2015 } 2016 found: 2017 return (rval); 2018 } 2019 2020 int 2021 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 2022 { 2023 2024 #if defined(KERN_TLS) || defined(RATELIMIT) 2025 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) { 2026 struct lagg_snd_tag *lst; 2027 struct m_snd_tag *mst; 2028 2029 mst = m->m_pkthdr.snd_tag; 2030 lst = mst_to_lst(mst); 2031 if (lst->tag->ifp != ifp) { 2032 m_freem(m); 2033 return (EAGAIN); 2034 } 2035 m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag); 2036 m_snd_tag_rele(mst); 2037 } 2038 #endif 2039 return (ifp->if_transmit)(ifp, m); 2040 } 2041 2042 /* 2043 * Simple round robin aggregation 2044 */ 2045 static void 2046 lagg_rr_attach(struct lagg_softc *sc) 2047 { 2048 sc->sc_seq = 0; 2049 } 2050 2051 static int 2052 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 2053 { 2054 struct lagg_port *lp; 2055 uint32_t p; 2056 2057 p = atomic_fetchadd_32(&sc->sc_seq, 1); 2058 if (sc->sc_stride > 0) 2059 p /= sc->sc_stride; 2060 2061 p %= sc->sc_count; 2062 lp = CK_SLIST_FIRST(&sc->sc_ports); 2063 2064 while (p--) 2065 lp = CK_SLIST_NEXT(lp, lp_entries); 2066 2067 /* 2068 * Check the port's link state. This will return the next active 2069 * port if the link is down or the port is NULL. 2070 */ 2071 if ((lp = lagg_link_active(sc, lp)) == NULL) { 2072 m_freem(m); 2073 return (ENETDOWN); 2074 } 2075 2076 /* Send mbuf */ 2077 return (lagg_enqueue(lp->lp_ifp, m)); 2078 } 2079 2080 static struct mbuf * 2081 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2082 { 2083 struct ifnet *ifp = sc->sc_ifp; 2084 2085 /* Just pass in the packet to our lagg device */ 2086 m->m_pkthdr.rcvif = ifp; 2087 2088 return (m); 2089 } 2090 2091 /* 2092 * Broadcast mode 2093 */ 2094 static int 2095 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m) 2096 { 2097 int active_ports = 0; 2098 int errors = 0; 2099 int ret; 2100 struct lagg_port *lp, *last = NULL; 2101 struct mbuf *m0; 2102 2103 LAGG_RLOCK_ASSERT(); 2104 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 2105 if (!LAGG_PORTACTIVE(lp)) 2106 continue; 2107 2108 active_ports++; 2109 2110 if (last != NULL) { 2111 m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT); 2112 if (m0 == NULL) { 2113 ret = ENOBUFS; 2114 errors++; 2115 break; 2116 } 2117 2118 ret = lagg_enqueue(last->lp_ifp, m0); 2119 if (ret != 0) 2120 errors++; 2121 } 2122 last = lp; 2123 } 2124 2125 if (last == NULL) { 2126 m_freem(m); 2127 return (ENOENT); 2128 } 2129 if ((last = lagg_link_active(sc, last)) == NULL) { 2130 m_freem(m); 2131 return (ENETDOWN); 2132 } 2133 2134 ret = lagg_enqueue(last->lp_ifp, m); 2135 if (ret != 0) 2136 errors++; 2137 2138 if (errors == 0) 2139 return (ret); 2140 2141 return (0); 2142 } 2143 2144 static struct mbuf* 2145 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2146 { 2147 struct ifnet *ifp = sc->sc_ifp; 2148 2149 /* Just pass in the packet to our lagg device */ 2150 m->m_pkthdr.rcvif = ifp; 2151 return (m); 2152 } 2153 2154 /* 2155 * Active failover 2156 */ 2157 static int 2158 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 2159 { 2160 struct lagg_port *lp; 2161 2162 /* Use the master port if active or the next available port */ 2163 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 2164 m_freem(m); 2165 return (ENETDOWN); 2166 } 2167 2168 /* Send mbuf */ 2169 return (lagg_enqueue(lp->lp_ifp, m)); 2170 } 2171 2172 static struct mbuf * 2173 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2174 { 2175 struct ifnet *ifp = sc->sc_ifp; 2176 struct lagg_port *tmp_tp; 2177 2178 if (lp == sc->sc_primary || V_lagg_failover_rx_all) { 2179 m->m_pkthdr.rcvif = ifp; 2180 return (m); 2181 } 2182 2183 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 2184 tmp_tp = lagg_link_active(sc, sc->sc_primary); 2185 /* 2186 * If tmp_tp is null, we've received a packet when all 2187 * our links are down. Weird, but process it anyways. 2188 */ 2189 if ((tmp_tp == NULL || tmp_tp == lp)) { 2190 m->m_pkthdr.rcvif = ifp; 2191 return (m); 2192 } 2193 } 2194 2195 m_freem(m); 2196 return (NULL); 2197 } 2198 2199 /* 2200 * Loadbalancing 2201 */ 2202 static void 2203 lagg_lb_attach(struct lagg_softc *sc) 2204 { 2205 struct lagg_port *lp; 2206 struct lagg_lb *lb; 2207 2208 LAGG_XLOCK_ASSERT(sc); 2209 lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO); 2210 lb->lb_key = m_ether_tcpip_hash_init(); 2211 sc->sc_psc = lb; 2212 2213 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2214 lagg_lb_port_create(lp); 2215 } 2216 2217 static void 2218 lagg_lb_detach(struct lagg_softc *sc) 2219 { 2220 struct lagg_lb *lb; 2221 2222 lb = (struct lagg_lb *)sc->sc_psc; 2223 if (lb != NULL) 2224 free(lb, M_LAGG); 2225 } 2226 2227 static int 2228 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 2229 { 2230 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 2231 struct lagg_port *lp_next; 2232 int i = 0, rv; 2233 2234 rv = 0; 2235 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 2236 LAGG_XLOCK_ASSERT(sc); 2237 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 2238 if (lp_next == lp) 2239 continue; 2240 if (i >= LAGG_MAX_PORTS) { 2241 rv = EINVAL; 2242 break; 2243 } 2244 if (sc->sc_ifflags & IFF_DEBUG) 2245 printf("%s: port %s at index %d\n", 2246 sc->sc_ifname, lp_next->lp_ifp->if_xname, i); 2247 lb->lb_ports[i++] = lp_next; 2248 } 2249 2250 return (rv); 2251 } 2252 2253 static int 2254 lagg_lb_port_create(struct lagg_port *lp) 2255 { 2256 struct lagg_softc *sc = lp->lp_softc; 2257 return (lagg_lb_porttable(sc, NULL)); 2258 } 2259 2260 static void 2261 lagg_lb_port_destroy(struct lagg_port *lp) 2262 { 2263 struct lagg_softc *sc = lp->lp_softc; 2264 lagg_lb_porttable(sc, lp); 2265 } 2266 2267 static int 2268 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 2269 { 2270 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 2271 struct lagg_port *lp = NULL; 2272 uint32_t p = 0; 2273 2274 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && 2275 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 2276 p = m->m_pkthdr.flowid >> sc->flowid_shift; 2277 else 2278 p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key); 2279 p %= sc->sc_count; 2280 lp = lb->lb_ports[p]; 2281 2282 /* 2283 * Check the port's link state. This will return the next active 2284 * port if the link is down or the port is NULL. 2285 */ 2286 if ((lp = lagg_link_active(sc, lp)) == NULL) { 2287 m_freem(m); 2288 return (ENETDOWN); 2289 } 2290 2291 /* Send mbuf */ 2292 return (lagg_enqueue(lp->lp_ifp, m)); 2293 } 2294 2295 static struct mbuf * 2296 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2297 { 2298 struct ifnet *ifp = sc->sc_ifp; 2299 2300 /* Just pass in the packet to our lagg device */ 2301 m->m_pkthdr.rcvif = ifp; 2302 2303 return (m); 2304 } 2305 2306 /* 2307 * 802.3ad LACP 2308 */ 2309 static void 2310 lagg_lacp_attach(struct lagg_softc *sc) 2311 { 2312 struct lagg_port *lp; 2313 2314 lacp_attach(sc); 2315 LAGG_XLOCK_ASSERT(sc); 2316 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2317 lacp_port_create(lp); 2318 } 2319 2320 static void 2321 lagg_lacp_detach(struct lagg_softc *sc) 2322 { 2323 struct lagg_port *lp; 2324 void *psc; 2325 2326 LAGG_XLOCK_ASSERT(sc); 2327 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2328 lacp_port_destroy(lp); 2329 2330 psc = sc->sc_psc; 2331 sc->sc_psc = NULL; 2332 lacp_detach(psc); 2333 } 2334 2335 static void 2336 lagg_lacp_lladdr(struct lagg_softc *sc) 2337 { 2338 struct lagg_port *lp; 2339 2340 LAGG_SXLOCK_ASSERT(sc); 2341 2342 /* purge all the lacp ports */ 2343 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2344 lacp_port_destroy(lp); 2345 2346 /* add them back in */ 2347 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2348 lacp_port_create(lp); 2349 } 2350 2351 static int 2352 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 2353 { 2354 struct lagg_port *lp; 2355 2356 lp = lacp_select_tx_port(sc, m); 2357 if (lp == NULL) { 2358 m_freem(m); 2359 return (ENETDOWN); 2360 } 2361 2362 /* Send mbuf */ 2363 return (lagg_enqueue(lp->lp_ifp, m)); 2364 } 2365 2366 static struct mbuf * 2367 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2368 { 2369 struct ifnet *ifp = sc->sc_ifp; 2370 struct ether_header *eh; 2371 u_short etype; 2372 2373 eh = mtod(m, struct ether_header *); 2374 etype = ntohs(eh->ether_type); 2375 2376 /* Tap off LACP control messages */ 2377 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 2378 m = lacp_input(lp, m); 2379 if (m == NULL) 2380 return (NULL); 2381 } 2382 2383 /* 2384 * If the port is not collecting or not in the active aggregator then 2385 * free and return. 2386 */ 2387 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 2388 m_freem(m); 2389 return (NULL); 2390 } 2391 2392 m->m_pkthdr.rcvif = ifp; 2393 return (m); 2394 } 2395 2396