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() NET_EPOCH_ASSERT() 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 NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx); 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 handle a non-zero value before checking 1261 * anything else to preserve compatibility. 1262 */ 1263 LAGG_XLOCK(sc); 1264 if (ro->ro_opts == 0 && 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 } 1272 if (ro->ro_opts == 0) { 1273 LAGG_XUNLOCK(sc); 1274 break; 1275 } 1276 1277 /* 1278 * Set options. LACP options are stored in sc->sc_psc, 1279 * not in sc_opts. 1280 */ 1281 int valid, lacp; 1282 1283 switch (ro->ro_opts) { 1284 case LAGG_OPT_USE_FLOWID: 1285 case -LAGG_OPT_USE_FLOWID: 1286 case LAGG_OPT_USE_NUMA: 1287 case -LAGG_OPT_USE_NUMA: 1288 case LAGG_OPT_FLOWIDSHIFT: 1289 case LAGG_OPT_RR_LIMIT: 1290 valid = 1; 1291 lacp = 0; 1292 break; 1293 case LAGG_OPT_LACP_TXTEST: 1294 case -LAGG_OPT_LACP_TXTEST: 1295 case LAGG_OPT_LACP_RXTEST: 1296 case -LAGG_OPT_LACP_RXTEST: 1297 case LAGG_OPT_LACP_STRICT: 1298 case -LAGG_OPT_LACP_STRICT: 1299 case LAGG_OPT_LACP_TIMEOUT: 1300 case -LAGG_OPT_LACP_TIMEOUT: 1301 valid = lacp = 1; 1302 break; 1303 default: 1304 valid = lacp = 0; 1305 break; 1306 } 1307 1308 if (valid == 0 || 1309 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) { 1310 /* Invalid combination of options specified. */ 1311 error = EINVAL; 1312 LAGG_XUNLOCK(sc); 1313 break; /* Return from SIOCSLAGGOPTS. */ 1314 } 1315 1316 /* 1317 * Store new options into sc->sc_opts except for 1318 * FLOWIDSHIFT, RR and LACP options. 1319 */ 1320 if (lacp == 0) { 1321 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT) 1322 sc->flowid_shift = ro->ro_flowid_shift; 1323 else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) { 1324 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN || 1325 ro->ro_bkt == 0) { 1326 error = EINVAL; 1327 LAGG_XUNLOCK(sc); 1328 break; 1329 } 1330 sc->sc_stride = ro->ro_bkt; 1331 } else if (ro->ro_opts > 0) 1332 sc->sc_opts |= ro->ro_opts; 1333 else 1334 sc->sc_opts &= ~ro->ro_opts; 1335 } else { 1336 struct lacp_softc *lsc; 1337 struct lacp_port *lp; 1338 1339 lsc = (struct lacp_softc *)sc->sc_psc; 1340 1341 switch (ro->ro_opts) { 1342 case LAGG_OPT_LACP_TXTEST: 1343 lsc->lsc_debug.lsc_tx_test = 1; 1344 break; 1345 case -LAGG_OPT_LACP_TXTEST: 1346 lsc->lsc_debug.lsc_tx_test = 0; 1347 break; 1348 case LAGG_OPT_LACP_RXTEST: 1349 lsc->lsc_debug.lsc_rx_test = 1; 1350 break; 1351 case -LAGG_OPT_LACP_RXTEST: 1352 lsc->lsc_debug.lsc_rx_test = 0; 1353 break; 1354 case LAGG_OPT_LACP_STRICT: 1355 lsc->lsc_strict_mode = 1; 1356 break; 1357 case -LAGG_OPT_LACP_STRICT: 1358 lsc->lsc_strict_mode = 0; 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 = 1; 1366 break; 1367 case -LAGG_OPT_LACP_TIMEOUT: 1368 LACP_LOCK(lsc); 1369 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1370 lp->lp_state &= ~LACP_STATE_TIMEOUT; 1371 LACP_UNLOCK(lsc); 1372 lsc->lsc_fast_timeout = 0; 1373 break; 1374 } 1375 } 1376 LAGG_XUNLOCK(sc); 1377 break; 1378 case SIOCGLAGGFLAGS: 1379 rf->rf_flags = 0; 1380 LAGG_XLOCK(sc); 1381 if (sc->sc_flags & MBUF_HASHFLAG_L2) 1382 rf->rf_flags |= LAGG_F_HASHL2; 1383 if (sc->sc_flags & MBUF_HASHFLAG_L3) 1384 rf->rf_flags |= LAGG_F_HASHL3; 1385 if (sc->sc_flags & MBUF_HASHFLAG_L4) 1386 rf->rf_flags |= LAGG_F_HASHL4; 1387 LAGG_XUNLOCK(sc); 1388 break; 1389 case SIOCSLAGGHASH: 1390 error = priv_check(td, PRIV_NET_LAGG); 1391 if (error) 1392 break; 1393 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1394 error = EINVAL; 1395 break; 1396 } 1397 LAGG_XLOCK(sc); 1398 sc->sc_flags = 0; 1399 if (rf->rf_flags & LAGG_F_HASHL2) 1400 sc->sc_flags |= MBUF_HASHFLAG_L2; 1401 if (rf->rf_flags & LAGG_F_HASHL3) 1402 sc->sc_flags |= MBUF_HASHFLAG_L3; 1403 if (rf->rf_flags & LAGG_F_HASHL4) 1404 sc->sc_flags |= MBUF_HASHFLAG_L4; 1405 LAGG_XUNLOCK(sc); 1406 break; 1407 case SIOCGLAGGPORT: 1408 if (rp->rp_portname[0] == '\0' || 1409 (tpif = ifunit_ref(rp->rp_portname)) == NULL) { 1410 error = EINVAL; 1411 break; 1412 } 1413 1414 LAGG_RLOCK(); 1415 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1416 lp->lp_softc != sc) { 1417 error = ENOENT; 1418 LAGG_RUNLOCK(); 1419 if_rele(tpif); 1420 break; 1421 } 1422 1423 lagg_port2req(lp, rp); 1424 LAGG_RUNLOCK(); 1425 if_rele(tpif); 1426 break; 1427 case SIOCSLAGGPORT: 1428 error = priv_check(td, PRIV_NET_LAGG); 1429 if (error) 1430 break; 1431 if (rp->rp_portname[0] == '\0' || 1432 (tpif = ifunit_ref(rp->rp_portname)) == NULL) { 1433 error = EINVAL; 1434 break; 1435 } 1436 #ifdef INET6 1437 /* 1438 * A laggport interface should not have inet6 address 1439 * because two interfaces with a valid link-local 1440 * scope zone must not be merged in any form. This 1441 * restriction is needed to prevent violation of 1442 * link-local scope zone. Attempts to add a laggport 1443 * interface which has inet6 addresses triggers 1444 * removal of all inet6 addresses on the member 1445 * interface. 1446 */ 1447 if (in6ifa_llaonifp(tpif)) { 1448 in6_ifdetach(tpif); 1449 if_printf(sc->sc_ifp, 1450 "IPv6 addresses on %s have been removed " 1451 "before adding it as a member to prevent " 1452 "IPv6 address scope violation.\n", 1453 tpif->if_xname); 1454 } 1455 #endif 1456 LAGG_XLOCK(sc); 1457 error = lagg_port_create(sc, tpif); 1458 LAGG_XUNLOCK(sc); 1459 if_rele(tpif); 1460 VLAN_CAPABILITIES(ifp); 1461 break; 1462 case SIOCSLAGGDELPORT: 1463 error = priv_check(td, PRIV_NET_LAGG); 1464 if (error) 1465 break; 1466 if (rp->rp_portname[0] == '\0' || 1467 (tpif = ifunit_ref(rp->rp_portname)) == NULL) { 1468 error = EINVAL; 1469 break; 1470 } 1471 1472 LAGG_XLOCK(sc); 1473 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1474 lp->lp_softc != sc) { 1475 error = ENOENT; 1476 LAGG_XUNLOCK(sc); 1477 if_rele(tpif); 1478 break; 1479 } 1480 1481 error = lagg_port_destroy(lp, 1); 1482 LAGG_XUNLOCK(sc); 1483 if_rele(tpif); 1484 VLAN_CAPABILITIES(ifp); 1485 break; 1486 case SIOCSIFFLAGS: 1487 /* Set flags on ports too */ 1488 LAGG_XLOCK(sc); 1489 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1490 lagg_setflags(lp, 1); 1491 } 1492 1493 if (!(ifp->if_flags & IFF_UP) && 1494 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1495 /* 1496 * If interface is marked down and it is running, 1497 * then stop and disable it. 1498 */ 1499 lagg_stop(sc); 1500 LAGG_XUNLOCK(sc); 1501 } else if ((ifp->if_flags & IFF_UP) && 1502 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1503 /* 1504 * If interface is marked up and it is stopped, then 1505 * start it. 1506 */ 1507 LAGG_XUNLOCK(sc); 1508 (*ifp->if_init)(sc); 1509 } else 1510 LAGG_XUNLOCK(sc); 1511 break; 1512 case SIOCADDMULTI: 1513 case SIOCDELMULTI: 1514 LAGG_XLOCK(sc); 1515 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1516 lagg_clrmulti(lp); 1517 lagg_setmulti(lp); 1518 } 1519 LAGG_XUNLOCK(sc); 1520 error = 0; 1521 break; 1522 case SIOCSIFMEDIA: 1523 case SIOCGIFMEDIA: 1524 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1525 break; 1526 1527 case SIOCSIFCAP: 1528 LAGG_XLOCK(sc); 1529 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1530 if (lp->lp_ioctl != NULL) 1531 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); 1532 } 1533 lagg_capabilities(sc); 1534 LAGG_XUNLOCK(sc); 1535 VLAN_CAPABILITIES(ifp); 1536 error = 0; 1537 break; 1538 1539 case SIOCSIFMTU: 1540 LAGG_XLOCK(sc); 1541 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1542 if (lp->lp_ioctl != NULL) 1543 error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); 1544 else 1545 error = EINVAL; 1546 if (error != 0) { 1547 if_printf(ifp, 1548 "failed to change MTU to %d on port %s, " 1549 "reverting all ports to original MTU (%d)\n", 1550 ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu); 1551 break; 1552 } 1553 } 1554 if (error == 0) { 1555 ifp->if_mtu = ifr->ifr_mtu; 1556 } else { 1557 /* set every port back to the original MTU */ 1558 ifr->ifr_mtu = ifp->if_mtu; 1559 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1560 if (lp->lp_ioctl != NULL) 1561 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); 1562 } 1563 } 1564 LAGG_XUNLOCK(sc); 1565 break; 1566 1567 default: 1568 error = ether_ioctl(ifp, cmd, data); 1569 break; 1570 } 1571 return (error); 1572 } 1573 1574 #if defined(KERN_TLS) || defined(RATELIMIT) 1575 static inline struct lagg_snd_tag * 1576 mst_to_lst(struct m_snd_tag *mst) 1577 { 1578 1579 return (__containerof(mst, struct lagg_snd_tag, com)); 1580 } 1581 1582 /* 1583 * Look up the port used by a specific flow. This only works for lagg 1584 * protocols with deterministic port mappings (e.g. not roundrobin). 1585 * In addition protocols which use a hash to map flows to ports must 1586 * be configured to use the mbuf flowid rather than hashing packet 1587 * contents. 1588 */ 1589 static struct lagg_port * 1590 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype) 1591 { 1592 struct lagg_softc *sc; 1593 struct lagg_port *lp; 1594 struct lagg_lb *lb; 1595 uint32_t p; 1596 1597 sc = ifp->if_softc; 1598 1599 switch (sc->sc_proto) { 1600 case LAGG_PROTO_FAILOVER: 1601 return (lagg_link_active(sc, sc->sc_primary)); 1602 case LAGG_PROTO_LOADBALANCE: 1603 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 || 1604 flowtype == M_HASHTYPE_NONE) 1605 return (NULL); 1606 p = flowid >> sc->flowid_shift; 1607 p %= sc->sc_count; 1608 lb = (struct lagg_lb *)sc->sc_psc; 1609 lp = lb->lb_ports[p]; 1610 return (lagg_link_active(sc, lp)); 1611 case LAGG_PROTO_LACP: 1612 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 || 1613 flowtype == M_HASHTYPE_NONE) 1614 return (NULL); 1615 return (lacp_select_tx_port_by_hash(sc, flowid)); 1616 default: 1617 return (NULL); 1618 } 1619 } 1620 1621 static int 1622 lagg_snd_tag_alloc(struct ifnet *ifp, 1623 union if_snd_tag_alloc_params *params, 1624 struct m_snd_tag **ppmt) 1625 { 1626 struct lagg_snd_tag *lst; 1627 struct lagg_softc *sc; 1628 struct lagg_port *lp; 1629 struct ifnet *lp_ifp; 1630 int error; 1631 1632 sc = ifp->if_softc; 1633 1634 LAGG_RLOCK(); 1635 lp = lookup_snd_tag_port(ifp, params->hdr.flowid, params->hdr.flowtype); 1636 if (lp == NULL) { 1637 LAGG_RUNLOCK(); 1638 return (EOPNOTSUPP); 1639 } 1640 if (lp->lp_ifp == NULL || lp->lp_ifp->if_snd_tag_alloc == NULL) { 1641 LAGG_RUNLOCK(); 1642 return (EOPNOTSUPP); 1643 } 1644 lp_ifp = lp->lp_ifp; 1645 if_ref(lp_ifp); 1646 LAGG_RUNLOCK(); 1647 1648 lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT); 1649 if (lst == NULL) { 1650 if_rele(lp_ifp); 1651 return (ENOMEM); 1652 } 1653 1654 error = lp_ifp->if_snd_tag_alloc(lp_ifp, params, &lst->tag); 1655 if_rele(lp_ifp); 1656 if (error) { 1657 free(lst, M_LAGG); 1658 return (error); 1659 } 1660 1661 m_snd_tag_init(&lst->com, ifp); 1662 1663 *ppmt = &lst->com; 1664 return (0); 1665 } 1666 1667 static int 1668 lagg_snd_tag_modify(struct m_snd_tag *mst, 1669 union if_snd_tag_modify_params *params) 1670 { 1671 struct lagg_snd_tag *lst; 1672 1673 lst = mst_to_lst(mst); 1674 return (lst->tag->ifp->if_snd_tag_modify(lst->tag, params)); 1675 } 1676 1677 static int 1678 lagg_snd_tag_query(struct m_snd_tag *mst, 1679 union if_snd_tag_query_params *params) 1680 { 1681 struct lagg_snd_tag *lst; 1682 1683 lst = mst_to_lst(mst); 1684 return (lst->tag->ifp->if_snd_tag_query(lst->tag, params)); 1685 } 1686 1687 static void 1688 lagg_snd_tag_free(struct m_snd_tag *mst) 1689 { 1690 struct lagg_snd_tag *lst; 1691 1692 lst = mst_to_lst(mst); 1693 m_snd_tag_rele(lst->tag); 1694 free(lst, M_LAGG); 1695 } 1696 1697 static void 1698 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q) 1699 { 1700 /* 1701 * For lagg, we have an indirect 1702 * interface. The caller needs to 1703 * get a ratelimit tag on the actual 1704 * interface the flow will go on. 1705 */ 1706 q->rate_table = NULL; 1707 q->flags = RT_IS_INDIRECT; 1708 q->max_flows = 0; 1709 q->number_of_rates = 0; 1710 } 1711 #endif 1712 1713 static int 1714 lagg_setmulti(struct lagg_port *lp) 1715 { 1716 struct lagg_softc *sc = lp->lp_softc; 1717 struct ifnet *ifp = lp->lp_ifp; 1718 struct ifnet *scifp = sc->sc_ifp; 1719 struct lagg_mc *mc; 1720 struct ifmultiaddr *ifma; 1721 int error; 1722 1723 IF_ADDR_WLOCK(scifp); 1724 CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1725 if (ifma->ifma_addr->sa_family != AF_LINK) 1726 continue; 1727 mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT); 1728 if (mc == NULL) { 1729 IF_ADDR_WUNLOCK(scifp); 1730 return (ENOMEM); 1731 } 1732 bcopy(ifma->ifma_addr, &mc->mc_addr, 1733 ifma->ifma_addr->sa_len); 1734 mc->mc_addr.sdl_index = ifp->if_index; 1735 mc->mc_ifma = NULL; 1736 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1737 } 1738 IF_ADDR_WUNLOCK(scifp); 1739 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) { 1740 error = if_addmulti(ifp, 1741 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma); 1742 if (error) 1743 return (error); 1744 } 1745 return (0); 1746 } 1747 1748 static int 1749 lagg_clrmulti(struct lagg_port *lp) 1750 { 1751 struct lagg_mc *mc; 1752 1753 LAGG_XLOCK_ASSERT(lp->lp_softc); 1754 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1755 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1756 if (mc->mc_ifma && lp->lp_detaching == 0) 1757 if_delmulti_ifma(mc->mc_ifma); 1758 free(mc, M_LAGG); 1759 } 1760 return (0); 1761 } 1762 1763 static int 1764 lagg_setcaps(struct lagg_port *lp, int cap) 1765 { 1766 struct ifreq ifr; 1767 1768 if (lp->lp_ifp->if_capenable == cap) 1769 return (0); 1770 if (lp->lp_ioctl == NULL) 1771 return (ENXIO); 1772 ifr.ifr_reqcap = cap; 1773 return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr)); 1774 } 1775 1776 /* Handle a ref counted flag that should be set on the lagg port as well */ 1777 static int 1778 lagg_setflag(struct lagg_port *lp, int flag, int status, 1779 int (*func)(struct ifnet *, int)) 1780 { 1781 struct lagg_softc *sc = lp->lp_softc; 1782 struct ifnet *scifp = sc->sc_ifp; 1783 struct ifnet *ifp = lp->lp_ifp; 1784 int error; 1785 1786 LAGG_XLOCK_ASSERT(sc); 1787 1788 status = status ? (scifp->if_flags & flag) : 0; 1789 /* Now "status" contains the flag value or 0 */ 1790 1791 /* 1792 * See if recorded ports status is different from what 1793 * we want it to be. If it is, flip it. We record ports 1794 * status in lp_ifflags so that we won't clear ports flag 1795 * we haven't set. In fact, we don't clear or set ports 1796 * flags directly, but get or release references to them. 1797 * That's why we can be sure that recorded flags still are 1798 * in accord with actual ports flags. 1799 */ 1800 if (status != (lp->lp_ifflags & flag)) { 1801 error = (*func)(ifp, status); 1802 if (error) 1803 return (error); 1804 lp->lp_ifflags &= ~flag; 1805 lp->lp_ifflags |= status; 1806 } 1807 return (0); 1808 } 1809 1810 /* 1811 * Handle IFF_* flags that require certain changes on the lagg port 1812 * if "status" is true, update ports flags respective to the lagg 1813 * if "status" is false, forcedly clear the flags set on port. 1814 */ 1815 static int 1816 lagg_setflags(struct lagg_port *lp, int status) 1817 { 1818 int error, i; 1819 1820 for (i = 0; lagg_pflags[i].flag; i++) { 1821 error = lagg_setflag(lp, lagg_pflags[i].flag, 1822 status, lagg_pflags[i].func); 1823 if (error) 1824 return (error); 1825 } 1826 return (0); 1827 } 1828 1829 static int 1830 lagg_transmit(struct ifnet *ifp, struct mbuf *m) 1831 { 1832 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1833 int error; 1834 1835 #if defined(KERN_TLS) || defined(RATELIMIT) 1836 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) 1837 MPASS(m->m_pkthdr.snd_tag->ifp == ifp); 1838 #endif 1839 LAGG_RLOCK(); 1840 /* We need a Tx algorithm and at least one port */ 1841 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1842 LAGG_RUNLOCK(); 1843 m_freem(m); 1844 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1845 return (ENXIO); 1846 } 1847 1848 ETHER_BPF_MTAP(ifp, m); 1849 1850 error = lagg_proto_start(sc, m); 1851 LAGG_RUNLOCK(); 1852 1853 if (error != 0) 1854 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1855 1856 return (error); 1857 } 1858 1859 /* 1860 * The ifp->if_qflush entry point for lagg(4) is no-op. 1861 */ 1862 static void 1863 lagg_qflush(struct ifnet *ifp __unused) 1864 { 1865 } 1866 1867 static struct mbuf * 1868 lagg_input(struct ifnet *ifp, struct mbuf *m) 1869 { 1870 struct lagg_port *lp = ifp->if_lagg; 1871 struct lagg_softc *sc = lp->lp_softc; 1872 struct ifnet *scifp = sc->sc_ifp; 1873 1874 LAGG_RLOCK(); 1875 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1876 lp->lp_detaching != 0 || 1877 sc->sc_proto == LAGG_PROTO_NONE) { 1878 LAGG_RUNLOCK(); 1879 m_freem(m); 1880 return (NULL); 1881 } 1882 1883 ETHER_BPF_MTAP(scifp, m); 1884 1885 m = lagg_proto_input(sc, lp, m); 1886 if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) { 1887 m_freem(m); 1888 m = NULL; 1889 } 1890 1891 LAGG_RUNLOCK(); 1892 return (m); 1893 } 1894 1895 static int 1896 lagg_media_change(struct ifnet *ifp) 1897 { 1898 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1899 1900 if (sc->sc_ifflags & IFF_DEBUG) 1901 printf("%s\n", __func__); 1902 1903 /* Ignore */ 1904 return (0); 1905 } 1906 1907 static void 1908 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1909 { 1910 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1911 struct lagg_port *lp; 1912 1913 imr->ifm_status = IFM_AVALID; 1914 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1915 1916 LAGG_RLOCK(); 1917 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1918 if (LAGG_PORTACTIVE(lp)) 1919 imr->ifm_status |= IFM_ACTIVE; 1920 } 1921 LAGG_RUNLOCK(); 1922 } 1923 1924 static void 1925 lagg_linkstate(struct lagg_softc *sc) 1926 { 1927 struct lagg_port *lp; 1928 int new_link = LINK_STATE_DOWN; 1929 uint64_t speed; 1930 1931 LAGG_XLOCK_ASSERT(sc); 1932 1933 /* LACP handles link state itself */ 1934 if (sc->sc_proto == LAGG_PROTO_LACP) 1935 return; 1936 1937 /* Our link is considered up if at least one of our ports is active */ 1938 LAGG_RLOCK(); 1939 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1940 if (lp->lp_ifp->if_link_state == LINK_STATE_UP) { 1941 new_link = LINK_STATE_UP; 1942 break; 1943 } 1944 } 1945 LAGG_RUNLOCK(); 1946 if_link_state_change(sc->sc_ifp, new_link); 1947 1948 /* Update if_baudrate to reflect the max possible speed */ 1949 switch (sc->sc_proto) { 1950 case LAGG_PROTO_FAILOVER: 1951 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1952 sc->sc_primary->lp_ifp->if_baudrate : 0; 1953 break; 1954 case LAGG_PROTO_ROUNDROBIN: 1955 case LAGG_PROTO_LOADBALANCE: 1956 case LAGG_PROTO_BROADCAST: 1957 speed = 0; 1958 LAGG_RLOCK(); 1959 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1960 speed += lp->lp_ifp->if_baudrate; 1961 LAGG_RUNLOCK(); 1962 sc->sc_ifp->if_baudrate = speed; 1963 break; 1964 case LAGG_PROTO_LACP: 1965 /* LACP updates if_baudrate itself */ 1966 break; 1967 } 1968 } 1969 1970 static void 1971 lagg_port_state(struct ifnet *ifp, int state) 1972 { 1973 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1974 struct lagg_softc *sc = NULL; 1975 1976 if (lp != NULL) 1977 sc = lp->lp_softc; 1978 if (sc == NULL) 1979 return; 1980 1981 LAGG_XLOCK(sc); 1982 lagg_linkstate(sc); 1983 lagg_proto_linkstate(sc, lp); 1984 LAGG_XUNLOCK(sc); 1985 } 1986 1987 struct lagg_port * 1988 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1989 { 1990 struct lagg_port *lp_next, *rval = NULL; 1991 1992 /* 1993 * Search a port which reports an active link state. 1994 */ 1995 1996 #ifdef INVARIANTS 1997 /* 1998 * This is called with either LAGG_RLOCK() held or 1999 * LAGG_XLOCK(sc) held. 2000 */ 2001 if (!in_epoch(net_epoch_preempt)) 2002 LAGG_XLOCK_ASSERT(sc); 2003 #endif 2004 2005 if (lp == NULL) 2006 goto search; 2007 if (LAGG_PORTACTIVE(lp)) { 2008 rval = lp; 2009 goto found; 2010 } 2011 if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL && 2012 LAGG_PORTACTIVE(lp_next)) { 2013 rval = lp_next; 2014 goto found; 2015 } 2016 2017 search: 2018 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 2019 if (LAGG_PORTACTIVE(lp_next)) { 2020 return (lp_next); 2021 } 2022 } 2023 found: 2024 return (rval); 2025 } 2026 2027 int 2028 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 2029 { 2030 2031 #if defined(KERN_TLS) || defined(RATELIMIT) 2032 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) { 2033 struct lagg_snd_tag *lst; 2034 struct m_snd_tag *mst; 2035 2036 mst = m->m_pkthdr.snd_tag; 2037 lst = mst_to_lst(mst); 2038 if (lst->tag->ifp != ifp) { 2039 m_freem(m); 2040 return (EAGAIN); 2041 } 2042 m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag); 2043 m_snd_tag_rele(mst); 2044 } 2045 #endif 2046 return (ifp->if_transmit)(ifp, m); 2047 } 2048 2049 /* 2050 * Simple round robin aggregation 2051 */ 2052 static void 2053 lagg_rr_attach(struct lagg_softc *sc) 2054 { 2055 sc->sc_seq = 0; 2056 sc->sc_stride = 1; 2057 } 2058 2059 static int 2060 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 2061 { 2062 struct lagg_port *lp; 2063 uint32_t p; 2064 2065 p = atomic_fetchadd_32(&sc->sc_seq, 1); 2066 p /= sc->sc_stride; 2067 p %= sc->sc_count; 2068 lp = CK_SLIST_FIRST(&sc->sc_ports); 2069 2070 while (p--) 2071 lp = CK_SLIST_NEXT(lp, lp_entries); 2072 2073 /* 2074 * Check the port's link state. This will return the next active 2075 * port if the link is down or the port is NULL. 2076 */ 2077 if ((lp = lagg_link_active(sc, lp)) == NULL) { 2078 m_freem(m); 2079 return (ENETDOWN); 2080 } 2081 2082 /* Send mbuf */ 2083 return (lagg_enqueue(lp->lp_ifp, m)); 2084 } 2085 2086 static struct mbuf * 2087 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2088 { 2089 struct ifnet *ifp = sc->sc_ifp; 2090 2091 /* Just pass in the packet to our lagg device */ 2092 m->m_pkthdr.rcvif = ifp; 2093 2094 return (m); 2095 } 2096 2097 /* 2098 * Broadcast mode 2099 */ 2100 static int 2101 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m) 2102 { 2103 int active_ports = 0; 2104 int errors = 0; 2105 int ret; 2106 struct lagg_port *lp, *last = NULL; 2107 struct mbuf *m0; 2108 2109 LAGG_RLOCK_ASSERT(); 2110 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 2111 if (!LAGG_PORTACTIVE(lp)) 2112 continue; 2113 2114 active_ports++; 2115 2116 if (last != NULL) { 2117 m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT); 2118 if (m0 == NULL) { 2119 ret = ENOBUFS; 2120 errors++; 2121 break; 2122 } 2123 2124 ret = lagg_enqueue(last->lp_ifp, m0); 2125 if (ret != 0) 2126 errors++; 2127 } 2128 last = lp; 2129 } 2130 2131 if (last == NULL) { 2132 m_freem(m); 2133 return (ENOENT); 2134 } 2135 if ((last = lagg_link_active(sc, last)) == NULL) { 2136 m_freem(m); 2137 return (ENETDOWN); 2138 } 2139 2140 ret = lagg_enqueue(last->lp_ifp, m); 2141 if (ret != 0) 2142 errors++; 2143 2144 if (errors == 0) 2145 return (ret); 2146 2147 return (0); 2148 } 2149 2150 static struct mbuf* 2151 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2152 { 2153 struct ifnet *ifp = sc->sc_ifp; 2154 2155 /* Just pass in the packet to our lagg device */ 2156 m->m_pkthdr.rcvif = ifp; 2157 return (m); 2158 } 2159 2160 /* 2161 * Active failover 2162 */ 2163 static int 2164 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 2165 { 2166 struct lagg_port *lp; 2167 2168 /* Use the master port if active or the next available port */ 2169 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 2170 m_freem(m); 2171 return (ENETDOWN); 2172 } 2173 2174 /* Send mbuf */ 2175 return (lagg_enqueue(lp->lp_ifp, m)); 2176 } 2177 2178 static struct mbuf * 2179 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2180 { 2181 struct ifnet *ifp = sc->sc_ifp; 2182 struct lagg_port *tmp_tp; 2183 2184 if (lp == sc->sc_primary || V_lagg_failover_rx_all) { 2185 m->m_pkthdr.rcvif = ifp; 2186 return (m); 2187 } 2188 2189 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 2190 tmp_tp = lagg_link_active(sc, sc->sc_primary); 2191 /* 2192 * If tmp_tp is null, we've received a packet when all 2193 * our links are down. Weird, but process it anyways. 2194 */ 2195 if ((tmp_tp == NULL || tmp_tp == lp)) { 2196 m->m_pkthdr.rcvif = ifp; 2197 return (m); 2198 } 2199 } 2200 2201 m_freem(m); 2202 return (NULL); 2203 } 2204 2205 /* 2206 * Loadbalancing 2207 */ 2208 static void 2209 lagg_lb_attach(struct lagg_softc *sc) 2210 { 2211 struct lagg_port *lp; 2212 struct lagg_lb *lb; 2213 2214 LAGG_XLOCK_ASSERT(sc); 2215 lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO); 2216 lb->lb_key = m_ether_tcpip_hash_init(); 2217 sc->sc_psc = lb; 2218 2219 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2220 lagg_lb_port_create(lp); 2221 } 2222 2223 static void 2224 lagg_lb_detach(struct lagg_softc *sc) 2225 { 2226 struct lagg_lb *lb; 2227 2228 lb = (struct lagg_lb *)sc->sc_psc; 2229 if (lb != NULL) 2230 free(lb, M_LAGG); 2231 } 2232 2233 static int 2234 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 2235 { 2236 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 2237 struct lagg_port *lp_next; 2238 int i = 0, rv; 2239 2240 rv = 0; 2241 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 2242 LAGG_XLOCK_ASSERT(sc); 2243 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 2244 if (lp_next == lp) 2245 continue; 2246 if (i >= LAGG_MAX_PORTS) { 2247 rv = EINVAL; 2248 break; 2249 } 2250 if (sc->sc_ifflags & IFF_DEBUG) 2251 printf("%s: port %s at index %d\n", 2252 sc->sc_ifname, lp_next->lp_ifp->if_xname, i); 2253 lb->lb_ports[i++] = lp_next; 2254 } 2255 2256 return (rv); 2257 } 2258 2259 static int 2260 lagg_lb_port_create(struct lagg_port *lp) 2261 { 2262 struct lagg_softc *sc = lp->lp_softc; 2263 return (lagg_lb_porttable(sc, NULL)); 2264 } 2265 2266 static void 2267 lagg_lb_port_destroy(struct lagg_port *lp) 2268 { 2269 struct lagg_softc *sc = lp->lp_softc; 2270 lagg_lb_porttable(sc, lp); 2271 } 2272 2273 static int 2274 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 2275 { 2276 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 2277 struct lagg_port *lp = NULL; 2278 uint32_t p = 0; 2279 2280 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && 2281 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 2282 p = m->m_pkthdr.flowid >> sc->flowid_shift; 2283 else 2284 p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key); 2285 p %= sc->sc_count; 2286 lp = lb->lb_ports[p]; 2287 2288 /* 2289 * Check the port's link state. This will return the next active 2290 * port if the link is down or the port is NULL. 2291 */ 2292 if ((lp = lagg_link_active(sc, lp)) == NULL) { 2293 m_freem(m); 2294 return (ENETDOWN); 2295 } 2296 2297 /* Send mbuf */ 2298 return (lagg_enqueue(lp->lp_ifp, m)); 2299 } 2300 2301 static struct mbuf * 2302 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2303 { 2304 struct ifnet *ifp = sc->sc_ifp; 2305 2306 /* Just pass in the packet to our lagg device */ 2307 m->m_pkthdr.rcvif = ifp; 2308 2309 return (m); 2310 } 2311 2312 /* 2313 * 802.3ad LACP 2314 */ 2315 static void 2316 lagg_lacp_attach(struct lagg_softc *sc) 2317 { 2318 struct lagg_port *lp; 2319 2320 lacp_attach(sc); 2321 LAGG_XLOCK_ASSERT(sc); 2322 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2323 lacp_port_create(lp); 2324 } 2325 2326 static void 2327 lagg_lacp_detach(struct lagg_softc *sc) 2328 { 2329 struct lagg_port *lp; 2330 void *psc; 2331 2332 LAGG_XLOCK_ASSERT(sc); 2333 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2334 lacp_port_destroy(lp); 2335 2336 psc = sc->sc_psc; 2337 sc->sc_psc = NULL; 2338 lacp_detach(psc); 2339 } 2340 2341 static void 2342 lagg_lacp_lladdr(struct lagg_softc *sc) 2343 { 2344 struct lagg_port *lp; 2345 2346 LAGG_SXLOCK_ASSERT(sc); 2347 2348 /* purge all the lacp ports */ 2349 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2350 lacp_port_destroy(lp); 2351 2352 /* add them back in */ 2353 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2354 lacp_port_create(lp); 2355 } 2356 2357 static int 2358 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 2359 { 2360 struct lagg_port *lp; 2361 2362 lp = lacp_select_tx_port(sc, m); 2363 if (lp == NULL) { 2364 m_freem(m); 2365 return (ENETDOWN); 2366 } 2367 2368 /* Send mbuf */ 2369 return (lagg_enqueue(lp->lp_ifp, m)); 2370 } 2371 2372 static struct mbuf * 2373 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2374 { 2375 struct ifnet *ifp = sc->sc_ifp; 2376 struct ether_header *eh; 2377 u_short etype; 2378 2379 eh = mtod(m, struct ether_header *); 2380 etype = ntohs(eh->ether_type); 2381 2382 /* Tap off LACP control messages */ 2383 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 2384 m = lacp_input(lp, m); 2385 if (m == NULL) 2386 return (NULL); 2387 } 2388 2389 /* 2390 * If the port is not collecting or not in the active aggregator then 2391 * free and return. 2392 */ 2393 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 2394 m_freem(m); 2395 return (NULL); 2396 } 2397 2398 m->m_pkthdr.rcvif = ifp; 2399 return (m); 2400 } 2401 2402