Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/net/if_lagg.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /* $OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $ */ 2 3 /* 4 * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org> 5 * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org> 6 * Copyright (c) 2014, 2016 Marcelo Araujo <araujo@FreeBSD.org> 7 * 8 * Permission to use, copy, modify, and distribute this software for any 9 * purpose with or without fee is hereby granted, provided that the above 10 * copyright notice and this permission notice appear in all copies. 11 * 12 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 13 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 14 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 15 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 16 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 17 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 18 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 19 */ 20 21 #include <sys/cdefs.h> 22 __FBSDID("$FreeBSD$"); 23 24 #include "opt_inet.h" 25 #include "opt_inet6.h" 26 #include "opt_ratelimit.h" 27 28 #include <sys/param.h> 29 #include <sys/kernel.h> 30 #include <sys/malloc.h> 31 #include <sys/mbuf.h> 32 #include <sys/queue.h> 33 #include <sys/socket.h> 34 #include <sys/sockio.h> 35 #include <sys/sysctl.h> 36 #include <sys/module.h> 37 #include <sys/priv.h> 38 #include <sys/systm.h> 39 #include <sys/proc.h> 40 #include <sys/lock.h> 41 #include <sys/rmlock.h> 42 #include <sys/sx.h> 43 #include <sys/taskqueue.h> 44 #include <sys/eventhandler.h> 45 46 #include <net/ethernet.h> 47 #include <net/if.h> 48 #include <net/if_clone.h> 49 #include <net/if_arp.h> 50 #include <net/if_dl.h> 51 #include <net/if_media.h> 52 #include <net/if_types.h> 53 #include <net/if_var.h> 54 #include <net/bpf.h> 55 #include <net/route.h> 56 #include <net/vnet.h> 57 #include <net/infiniband.h> 58 59 #if defined(INET) || defined(INET6) 60 #include <netinet/in.h> 61 #include <netinet/ip.h> 62 #endif 63 #ifdef INET 64 #include <netinet/in_systm.h> 65 #include <netinet/if_ether.h> 66 #endif 67 68 #ifdef INET6 69 #include <netinet/ip6.h> 70 #include <netinet6/in6_var.h> 71 #include <netinet6/in6_ifattach.h> 72 #endif 73 74 #include <net/if_vlan_var.h> 75 #include <net/if_lagg.h> 76 #include <net/ieee8023ad_lacp.h> 77 78 #ifdef INET6 79 /* 80 * XXX: declare here to avoid to include many inet6 related files.. 81 * should be more generalized? 82 */ 83 extern void nd6_setmtu(struct ifnet *); 84 #endif 85 86 #define LAGG_RLOCK() struct epoch_tracker lagg_et; epoch_enter_preempt(net_epoch_preempt, &lagg_et) 87 #define LAGG_RUNLOCK() epoch_exit_preempt(net_epoch_preempt, &lagg_et) 88 #define LAGG_RLOCK_ASSERT() MPASS(in_epoch(net_epoch_preempt)) 89 #define LAGG_UNLOCK_ASSERT() MPASS(!in_epoch(net_epoch_preempt)) 90 91 #define LAGG_SX_INIT(_sc) sx_init(&(_sc)->sc_sx, "if_lagg sx") 92 #define LAGG_SX_DESTROY(_sc) sx_destroy(&(_sc)->sc_sx) 93 #define LAGG_XLOCK(_sc) sx_xlock(&(_sc)->sc_sx) 94 #define LAGG_XUNLOCK(_sc) sx_xunlock(&(_sc)->sc_sx) 95 #define LAGG_SXLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_LOCKED) 96 #define LAGG_XLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_XLOCKED) 97 98 /* Special flags we should propagate to the lagg ports. */ 99 static struct { 100 int flag; 101 int (*func)(struct ifnet *, int); 102 } lagg_pflags[] = { 103 {IFF_PROMISC, ifpromisc}, 104 {IFF_ALLMULTI, if_allmulti}, 105 {0, NULL} 106 }; 107 108 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */ 109 #define V_lagg_list VNET(lagg_list) 110 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx); 111 #define V_lagg_list_mtx VNET(lagg_list_mtx) 112 #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \ 113 "if_lagg list", NULL, MTX_DEF) 114 #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx) 115 #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx) 116 #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx) 117 eventhandler_tag lagg_detach_cookie = NULL; 118 119 static int lagg_clone_create(struct if_clone *, int, caddr_t); 120 static void lagg_clone_destroy(struct ifnet *); 121 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner); 122 #define V_lagg_cloner VNET(lagg_cloner) 123 static const char laggname[] = "lagg"; 124 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface"); 125 126 static void lagg_capabilities(struct lagg_softc *); 127 static int lagg_port_create(struct lagg_softc *, struct ifnet *); 128 static int lagg_port_destroy(struct lagg_port *, int); 129 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *); 130 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *); 131 static void lagg_linkstate(struct lagg_softc *); 132 static void lagg_port_state(struct ifnet *, int); 133 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); 134 static int lagg_port_output(struct ifnet *, struct mbuf *, 135 const struct sockaddr *, struct route *); 136 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 137 #ifdef LAGG_PORT_STACKING 138 static int lagg_port_checkstacking(struct lagg_softc *); 139 #endif 140 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 141 static void lagg_init(void *); 142 static void lagg_stop(struct lagg_softc *); 143 static int lagg_ioctl(struct ifnet *, u_long, caddr_t); 144 #ifdef RATELIMIT 145 static int lagg_snd_tag_alloc(struct ifnet *, 146 union if_snd_tag_alloc_params *, 147 struct m_snd_tag **); 148 #endif 149 static int lagg_setmulti(struct lagg_port *); 150 static int lagg_clrmulti(struct lagg_port *); 151 static int lagg_setcaps(struct lagg_port *, int cap); 152 static int lagg_setflag(struct lagg_port *, int, int, 153 int (*func)(struct ifnet *, int)); 154 static int lagg_setflags(struct lagg_port *, int status); 155 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt); 156 static int lagg_transmit_ethernet(struct ifnet *, struct mbuf *); 157 static int lagg_transmit_infiniband(struct ifnet *, struct mbuf *); 158 static void lagg_qflush(struct ifnet *); 159 static int lagg_media_change(struct ifnet *); 160 static void lagg_media_status(struct ifnet *, struct ifmediareq *); 161 static struct lagg_port *lagg_link_active(struct lagg_softc *, 162 struct lagg_port *); 163 164 /* Simple round robin */ 165 static void lagg_rr_attach(struct lagg_softc *); 166 static int lagg_rr_start(struct lagg_softc *, struct mbuf *); 167 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 168 struct mbuf *); 169 170 /* Active failover */ 171 static int lagg_fail_start(struct lagg_softc *, struct mbuf *); 172 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 173 struct mbuf *); 174 175 /* Loadbalancing */ 176 static void lagg_lb_attach(struct lagg_softc *); 177 static void lagg_lb_detach(struct lagg_softc *); 178 static int lagg_lb_port_create(struct lagg_port *); 179 static void lagg_lb_port_destroy(struct lagg_port *); 180 static int lagg_lb_start(struct lagg_softc *, struct mbuf *); 181 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 182 struct mbuf *); 183 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 184 185 /* Broadcast */ 186 static int lagg_bcast_start(struct lagg_softc *, struct mbuf *); 187 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *, 188 struct mbuf *); 189 190 /* 802.3ad LACP */ 191 static void lagg_lacp_attach(struct lagg_softc *); 192 static void lagg_lacp_detach(struct lagg_softc *); 193 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); 194 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 195 struct mbuf *); 196 static void lagg_lacp_lladdr(struct lagg_softc *); 197 198 /* lagg protocol table */ 199 static const struct lagg_proto { 200 lagg_proto pr_num; 201 void (*pr_attach)(struct lagg_softc *); 202 void (*pr_detach)(struct lagg_softc *); 203 int (*pr_start)(struct lagg_softc *, struct mbuf *); 204 struct mbuf * (*pr_input)(struct lagg_softc *, struct lagg_port *, 205 struct mbuf *); 206 int (*pr_addport)(struct lagg_port *); 207 void (*pr_delport)(struct lagg_port *); 208 void (*pr_linkstate)(struct lagg_port *); 209 void (*pr_init)(struct lagg_softc *); 210 void (*pr_stop)(struct lagg_softc *); 211 void (*pr_lladdr)(struct lagg_softc *); 212 void (*pr_request)(struct lagg_softc *, void *); 213 void (*pr_portreq)(struct lagg_port *, void *); 214 } lagg_protos[] = { 215 { 216 .pr_num = LAGG_PROTO_NONE 217 }, 218 { 219 .pr_num = LAGG_PROTO_ROUNDROBIN, 220 .pr_attach = lagg_rr_attach, 221 .pr_start = lagg_rr_start, 222 .pr_input = lagg_rr_input, 223 }, 224 { 225 .pr_num = LAGG_PROTO_FAILOVER, 226 .pr_start = lagg_fail_start, 227 .pr_input = lagg_fail_input, 228 }, 229 { 230 .pr_num = LAGG_PROTO_LOADBALANCE, 231 .pr_attach = lagg_lb_attach, 232 .pr_detach = lagg_lb_detach, 233 .pr_start = lagg_lb_start, 234 .pr_input = lagg_lb_input, 235 .pr_addport = lagg_lb_port_create, 236 .pr_delport = lagg_lb_port_destroy, 237 }, 238 { 239 .pr_num = LAGG_PROTO_LACP, 240 .pr_attach = lagg_lacp_attach, 241 .pr_detach = lagg_lacp_detach, 242 .pr_start = lagg_lacp_start, 243 .pr_input = lagg_lacp_input, 244 .pr_addport = lacp_port_create, 245 .pr_delport = lacp_port_destroy, 246 .pr_linkstate = lacp_linkstate, 247 .pr_init = lacp_init, 248 .pr_stop = lacp_stop, 249 .pr_lladdr = lagg_lacp_lladdr, 250 .pr_request = lacp_req, 251 .pr_portreq = lacp_portreq, 252 }, 253 { 254 .pr_num = LAGG_PROTO_BROADCAST, 255 .pr_start = lagg_bcast_start, 256 .pr_input = lagg_bcast_input, 257 }, 258 }; 259 260 SYSCTL_DECL(_net_link); 261 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 262 "Link Aggregation"); 263 264 /* Allow input on any failover links */ 265 VNET_DEFINE_STATIC(int, lagg_failover_rx_all); 266 #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all) 267 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET, 268 &VNET_NAME(lagg_failover_rx_all), 0, 269 "Accept input from any interface in a failover lagg"); 270 271 /* Default value for using flowid */ 272 VNET_DEFINE_STATIC(int, def_use_flowid) = 0; 273 #define V_def_use_flowid VNET(def_use_flowid) 274 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN, 275 &VNET_NAME(def_use_flowid), 0, 276 "Default setting for using flow id for load sharing"); 277 278 /* Default value for flowid shift */ 279 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16; 280 #define V_def_flowid_shift VNET(def_flowid_shift) 281 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN, 282 &VNET_NAME(def_flowid_shift), 0, 283 "Default setting for flowid shift for load sharing"); 284 285 static void 286 vnet_lagg_init(const void *unused __unused) 287 { 288 289 LAGG_LIST_LOCK_INIT(); 290 SLIST_INIT(&V_lagg_list); 291 V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create, 292 lagg_clone_destroy, 0); 293 } 294 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 295 vnet_lagg_init, NULL); 296 297 static void 298 vnet_lagg_uninit(const void *unused __unused) 299 { 300 301 if_clone_detach(V_lagg_cloner); 302 LAGG_LIST_LOCK_DESTROY(); 303 } 304 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, 305 vnet_lagg_uninit, NULL); 306 307 static int 308 lagg_modevent(module_t mod, int type, void *data) 309 { 310 311 switch (type) { 312 case MOD_LOAD: 313 lagg_input_ethernet_p = lagg_input_ethernet; 314 lagg_input_infiniband_p = lagg_input_infiniband; 315 lagg_linkstate_p = lagg_port_state; 316 lagg_detach_cookie = EVENTHANDLER_REGISTER( 317 ifnet_departure_event, lagg_port_ifdetach, NULL, 318 EVENTHANDLER_PRI_ANY); 319 break; 320 case MOD_UNLOAD: 321 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 322 lagg_detach_cookie); 323 lagg_input_ethernet_p = NULL; 324 lagg_input_infiniband_p = NULL; 325 lagg_linkstate_p = NULL; 326 break; 327 default: 328 return (EOPNOTSUPP); 329 } 330 return (0); 331 } 332 333 static moduledata_t lagg_mod = { 334 "if_lagg", 335 lagg_modevent, 336 0 337 }; 338 339 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 340 MODULE_VERSION(if_lagg, 1); 341 MODULE_DEPEND(if_lagg, if_infiniband, 1, 1, 1); 342 343 static void 344 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr) 345 { 346 347 LAGG_XLOCK_ASSERT(sc); 348 KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto", 349 __func__, sc)); 350 351 if (sc->sc_ifflags & IFF_DEBUG) 352 if_printf(sc->sc_ifp, "using proto %u\n", pr); 353 354 if (lagg_protos[pr].pr_attach != NULL) 355 lagg_protos[pr].pr_attach(sc); 356 sc->sc_proto = pr; 357 } 358 359 static void 360 lagg_proto_detach(struct lagg_softc *sc) 361 { 362 lagg_proto pr; 363 364 LAGG_XLOCK_ASSERT(sc); 365 pr = sc->sc_proto; 366 sc->sc_proto = LAGG_PROTO_NONE; 367 368 if (lagg_protos[pr].pr_detach != NULL) 369 lagg_protos[pr].pr_detach(sc); 370 } 371 372 static int 373 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m) 374 { 375 376 return (lagg_protos[sc->sc_proto].pr_start(sc, m)); 377 } 378 379 static struct mbuf * 380 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 381 { 382 383 return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m)); 384 } 385 386 static int 387 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp) 388 { 389 390 if (lagg_protos[sc->sc_proto].pr_addport == NULL) 391 return (0); 392 else 393 return (lagg_protos[sc->sc_proto].pr_addport(lp)); 394 } 395 396 static void 397 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp) 398 { 399 400 if (lagg_protos[sc->sc_proto].pr_delport != NULL) 401 lagg_protos[sc->sc_proto].pr_delport(lp); 402 } 403 404 static void 405 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp) 406 { 407 408 if (lagg_protos[sc->sc_proto].pr_linkstate != NULL) 409 lagg_protos[sc->sc_proto].pr_linkstate(lp); 410 } 411 412 static void 413 lagg_proto_init(struct lagg_softc *sc) 414 { 415 416 if (lagg_protos[sc->sc_proto].pr_init != NULL) 417 lagg_protos[sc->sc_proto].pr_init(sc); 418 } 419 420 static void 421 lagg_proto_stop(struct lagg_softc *sc) 422 { 423 424 if (lagg_protos[sc->sc_proto].pr_stop != NULL) 425 lagg_protos[sc->sc_proto].pr_stop(sc); 426 } 427 428 static void 429 lagg_proto_lladdr(struct lagg_softc *sc) 430 { 431 432 if (lagg_protos[sc->sc_proto].pr_lladdr != NULL) 433 lagg_protos[sc->sc_proto].pr_lladdr(sc); 434 } 435 436 static void 437 lagg_proto_request(struct lagg_softc *sc, void *v) 438 { 439 440 if (lagg_protos[sc->sc_proto].pr_request != NULL) 441 lagg_protos[sc->sc_proto].pr_request(sc, v); 442 } 443 444 static void 445 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v) 446 { 447 448 if (lagg_protos[sc->sc_proto].pr_portreq != NULL) 449 lagg_protos[sc->sc_proto].pr_portreq(lp, v); 450 } 451 452 /* 453 * This routine is run via an vlan 454 * config EVENT 455 */ 456 static void 457 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 458 { 459 struct lagg_softc *sc = ifp->if_softc; 460 struct lagg_port *lp; 461 462 if (ifp->if_softc != arg) /* Not our event */ 463 return; 464 465 LAGG_XLOCK(sc); 466 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 467 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 468 LAGG_XUNLOCK(sc); 469 } 470 471 /* 472 * This routine is run via an vlan 473 * unconfig EVENT 474 */ 475 static void 476 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 477 { 478 struct lagg_softc *sc = ifp->if_softc; 479 struct lagg_port *lp; 480 481 if (ifp->if_softc != arg) /* Not our event */ 482 return; 483 484 LAGG_XLOCK(sc); 485 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 486 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 487 LAGG_XUNLOCK(sc); 488 } 489 490 static int 491 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) 492 { 493 struct iflaggparam iflp; 494 struct lagg_softc *sc; 495 struct ifnet *ifp; 496 int if_type; 497 int error; 498 static const uint8_t eaddr[LAGG_ADDR_LEN]; 499 500 if (params != NULL) { 501 error = copyin(params, &iflp, sizeof(iflp)); 502 if (error) 503 return (error); 504 505 switch (iflp.lagg_type) { 506 case LAGG_TYPE_ETHERNET: 507 if_type = IFT_ETHER; 508 break; 509 case LAGG_TYPE_INFINIBAND: 510 if_type = IFT_INFINIBAND; 511 break; 512 default: 513 return (EINVAL); 514 } 515 } else { 516 if_type = IFT_ETHER; 517 } 518 519 sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO); 520 ifp = sc->sc_ifp = if_alloc(if_type); 521 if (ifp == NULL) { 522 free(sc, M_LAGG); 523 return (ENOSPC); 524 } 525 LAGG_SX_INIT(sc); 526 527 mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF); 528 callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0); 529 530 LAGG_XLOCK(sc); 531 if (V_def_use_flowid) 532 sc->sc_opts |= LAGG_OPT_USE_FLOWID; 533 sc->flowid_shift = V_def_flowid_shift; 534 535 /* Hash all layers by default */ 536 sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4; 537 538 lagg_proto_attach(sc, LAGG_PROTO_DEFAULT); 539 540 CK_SLIST_INIT(&sc->sc_ports); 541 542 switch (if_type) { 543 case IFT_ETHER: 544 /* Initialise pseudo media types */ 545 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 546 lagg_media_status); 547 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 548 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 549 550 if_initname(ifp, laggname, unit); 551 ifp->if_transmit = lagg_transmit_ethernet; 552 break; 553 case IFT_INFINIBAND: 554 if_initname(ifp, laggname, unit); 555 ifp->if_transmit = lagg_transmit_infiniband; 556 break; 557 default: 558 break; 559 } 560 ifp->if_softc = sc; 561 ifp->if_qflush = lagg_qflush; 562 ifp->if_init = lagg_init; 563 ifp->if_ioctl = lagg_ioctl; 564 ifp->if_get_counter = lagg_get_counter; 565 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 566 #ifdef RATELIMIT 567 ifp->if_snd_tag_alloc = lagg_snd_tag_alloc; 568 #endif 569 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS; 570 571 /* 572 * Attach as an ordinary ethernet device, children will be attached 573 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG. 574 */ 575 switch (if_type) { 576 case IFT_ETHER: 577 ether_ifattach(ifp, eaddr); 578 break; 579 case IFT_INFINIBAND: 580 infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr); 581 break; 582 default: 583 break; 584 } 585 586 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 587 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 588 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 589 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 590 591 /* Insert into the global list of laggs */ 592 LAGG_LIST_LOCK(); 593 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries); 594 LAGG_LIST_UNLOCK(); 595 LAGG_XUNLOCK(sc); 596 597 return (0); 598 } 599 600 static void 601 lagg_clone_destroy(struct ifnet *ifp) 602 { 603 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 604 struct lagg_port *lp; 605 606 LAGG_XLOCK(sc); 607 sc->sc_destroying = 1; 608 lagg_stop(sc); 609 ifp->if_flags &= ~IFF_UP; 610 611 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 612 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 613 614 /* Shutdown and remove lagg ports */ 615 while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL) 616 lagg_port_destroy(lp, 1); 617 618 /* Unhook the aggregation protocol */ 619 lagg_proto_detach(sc); 620 LAGG_XUNLOCK(sc); 621 622 switch (ifp->if_type) { 623 case IFT_ETHER: 624 ifmedia_removeall(&sc->sc_media); 625 ether_ifdetach(ifp); 626 break; 627 case IFT_INFINIBAND: 628 infiniband_ifdetach(ifp); 629 break; 630 default: 631 break; 632 } 633 if_free(ifp); 634 635 LAGG_LIST_LOCK(); 636 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries); 637 LAGG_LIST_UNLOCK(); 638 639 mtx_destroy(&sc->sc_mtx); 640 LAGG_SX_DESTROY(sc); 641 free(sc, M_LAGG); 642 } 643 644 static void 645 lagg_capabilities(struct lagg_softc *sc) 646 { 647 struct lagg_port *lp; 648 int cap, ena, pena; 649 uint64_t hwa; 650 struct ifnet_hw_tsomax hw_tsomax; 651 652 LAGG_XLOCK_ASSERT(sc); 653 654 /* Get common enabled capabilities for the lagg ports */ 655 ena = ~0; 656 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 657 ena &= lp->lp_ifp->if_capenable; 658 ena = (ena == ~0 ? 0 : ena); 659 660 /* 661 * Apply common enabled capabilities back to the lagg ports. 662 * May require several iterations if they are dependent. 663 */ 664 do { 665 pena = ena; 666 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 667 lagg_setcaps(lp, ena); 668 ena &= lp->lp_ifp->if_capenable; 669 } 670 } while (pena != ena); 671 672 /* Get other capabilities from the lagg ports */ 673 cap = ~0; 674 hwa = ~(uint64_t)0; 675 memset(&hw_tsomax, 0, sizeof(hw_tsomax)); 676 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 677 cap &= lp->lp_ifp->if_capabilities; 678 hwa &= lp->lp_ifp->if_hwassist; 679 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax); 680 } 681 cap = (cap == ~0 ? 0 : cap); 682 hwa = (hwa == ~(uint64_t)0 ? 0 : hwa); 683 684 if (sc->sc_ifp->if_capabilities != cap || 685 sc->sc_ifp->if_capenable != ena || 686 sc->sc_ifp->if_hwassist != hwa || 687 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) { 688 sc->sc_ifp->if_capabilities = cap; 689 sc->sc_ifp->if_capenable = ena; 690 sc->sc_ifp->if_hwassist = hwa; 691 getmicrotime(&sc->sc_ifp->if_lastchange); 692 693 if (sc->sc_ifflags & IFF_DEBUG) 694 if_printf(sc->sc_ifp, 695 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 696 } 697 } 698 699 static int 700 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 701 { 702 struct lagg_softc *sc_ptr; 703 struct lagg_port *lp, *tlp; 704 struct ifreq ifr; 705 int error, i, oldmtu; 706 int if_type; 707 uint64_t *pval; 708 709 LAGG_XLOCK_ASSERT(sc); 710 711 if (sc->sc_ifp == ifp) { 712 if_printf(sc->sc_ifp, 713 "cannot add a lagg to itself as a port\n"); 714 return (EINVAL); 715 } 716 717 /* Limit the maximal number of lagg ports */ 718 if (sc->sc_count >= LAGG_MAX_PORTS) 719 return (ENOSPC); 720 721 /* Check if port has already been associated to a lagg */ 722 if (ifp->if_lagg != NULL) { 723 /* Port is already in the current lagg? */ 724 lp = (struct lagg_port *)ifp->if_lagg; 725 if (lp->lp_softc == sc) 726 return (EEXIST); 727 return (EBUSY); 728 } 729 730 switch (sc->sc_ifp->if_type) { 731 case IFT_ETHER: 732 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 733 if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN) 734 return (EPROTONOSUPPORT); 735 if_type = IFT_IEEE8023ADLAG; 736 break; 737 case IFT_INFINIBAND: 738 /* XXX Disallow non-infiniband interfaces */ 739 if (ifp->if_type != IFT_INFINIBAND) 740 return (EPROTONOSUPPORT); 741 if_type = IFT_INFINIBANDLAG; 742 break; 743 default: 744 break; 745 } 746 747 /* Allow the first Ethernet member to define the MTU */ 748 oldmtu = -1; 749 if (CK_SLIST_EMPTY(&sc->sc_ports)) { 750 sc->sc_ifp->if_mtu = ifp->if_mtu; 751 } else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 752 if (ifp->if_ioctl == NULL) { 753 if_printf(sc->sc_ifp, "cannot change MTU for %s\n", 754 ifp->if_xname); 755 return (EINVAL); 756 } 757 oldmtu = ifp->if_mtu; 758 strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name)); 759 ifr.ifr_mtu = sc->sc_ifp->if_mtu; 760 error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr); 761 if (error != 0) { 762 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 763 ifp->if_xname); 764 return (error); 765 } 766 ifr.ifr_mtu = oldmtu; 767 } 768 769 lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO); 770 lp->lp_softc = sc; 771 772 /* Check if port is a stacked lagg */ 773 LAGG_LIST_LOCK(); 774 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) { 775 if (ifp == sc_ptr->sc_ifp) { 776 LAGG_LIST_UNLOCK(); 777 free(lp, M_LAGG); 778 if (oldmtu != -1) 779 (*ifp->if_ioctl)(ifp, SIOCSIFMTU, 780 (caddr_t)&ifr); 781 return (EINVAL); 782 /* XXX disable stacking for the moment, its untested */ 783 #ifdef LAGG_PORT_STACKING 784 lp->lp_flags |= LAGG_PORT_STACK; 785 if (lagg_port_checkstacking(sc_ptr) >= 786 LAGG_MAX_STACKING) { 787 LAGG_LIST_UNLOCK(); 788 free(lp, M_LAGG); 789 if (oldmtu != -1) 790 (*ifp->if_ioctl)(ifp, SIOCSIFMTU, 791 (caddr_t)&ifr); 792 return (E2BIG); 793 } 794 #endif 795 } 796 } 797 LAGG_LIST_UNLOCK(); 798 799 if_ref(ifp); 800 lp->lp_ifp = ifp; 801 802 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen); 803 lp->lp_ifcapenable = ifp->if_capenable; 804 if (CK_SLIST_EMPTY(&sc->sc_ports)) { 805 bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen); 806 lagg_proto_lladdr(sc); 807 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp); 808 } else { 809 if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen); 810 } 811 lagg_setflags(lp, 1); 812 813 if (CK_SLIST_EMPTY(&sc->sc_ports)) 814 sc->sc_primary = lp; 815 816 /* Change the interface type */ 817 lp->lp_iftype = ifp->if_type; 818 ifp->if_type = if_type; 819 ifp->if_lagg = lp; 820 lp->lp_ioctl = ifp->if_ioctl; 821 ifp->if_ioctl = lagg_port_ioctl; 822 lp->lp_output = ifp->if_output; 823 ifp->if_output = lagg_port_output; 824 825 /* Read port counters */ 826 pval = lp->port_counters.val; 827 for (i = 0; i < IFCOUNTERS; i++, pval++) 828 *pval = ifp->if_get_counter(ifp, i); 829 830 /* 831 * Insert into the list of ports. 832 * Keep ports sorted by if_index. It is handy, when configuration 833 * is predictable and `ifconfig laggN create ...` command 834 * will lead to the same result each time. 835 */ 836 CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) { 837 if (tlp->lp_ifp->if_index < ifp->if_index && ( 838 CK_SLIST_NEXT(tlp, lp_entries) == NULL || 839 ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index > 840 ifp->if_index)) 841 break; 842 } 843 if (tlp != NULL) 844 CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries); 845 else 846 CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 847 sc->sc_count++; 848 849 lagg_setmulti(lp); 850 851 852 if ((error = lagg_proto_addport(sc, lp)) != 0) { 853 /* Remove the port, without calling pr_delport. */ 854 lagg_port_destroy(lp, 0); 855 if (oldmtu != -1) 856 (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr); 857 return (error); 858 } 859 860 /* Update lagg capabilities */ 861 lagg_capabilities(sc); 862 lagg_linkstate(sc); 863 864 return (0); 865 } 866 867 #ifdef LAGG_PORT_STACKING 868 static int 869 lagg_port_checkstacking(struct lagg_softc *sc) 870 { 871 struct lagg_softc *sc_ptr; 872 struct lagg_port *lp; 873 int m = 0; 874 875 LAGG_SXLOCK_ASSERT(sc); 876 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 877 if (lp->lp_flags & LAGG_PORT_STACK) { 878 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 879 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 880 } 881 } 882 883 return (m + 1); 884 } 885 #endif 886 887 static void 888 lagg_port_destroy_cb(epoch_context_t ec) 889 { 890 struct lagg_port *lp; 891 struct ifnet *ifp; 892 893 lp = __containerof(ec, struct lagg_port, lp_epoch_ctx); 894 ifp = lp->lp_ifp; 895 896 if_rele(ifp); 897 free(lp, M_LAGG); 898 } 899 900 static int 901 lagg_port_destroy(struct lagg_port *lp, int rundelport) 902 { 903 struct lagg_softc *sc = lp->lp_softc; 904 struct lagg_port *lp_ptr, *lp0; 905 struct ifnet *ifp = lp->lp_ifp; 906 uint64_t *pval, vdiff; 907 int i; 908 909 LAGG_XLOCK_ASSERT(sc); 910 911 if (rundelport) 912 lagg_proto_delport(sc, lp); 913 914 if (lp->lp_detaching == 0) 915 lagg_clrmulti(lp); 916 917 /* Restore interface */ 918 ifp->if_type = lp->lp_iftype; 919 ifp->if_ioctl = lp->lp_ioctl; 920 ifp->if_output = lp->lp_output; 921 ifp->if_lagg = NULL; 922 923 /* Update detached port counters */ 924 pval = lp->port_counters.val; 925 for (i = 0; i < IFCOUNTERS; i++, pval++) { 926 vdiff = ifp->if_get_counter(ifp, i) - *pval; 927 sc->detached_counters.val[i] += vdiff; 928 } 929 930 /* Finally, remove the port from the lagg */ 931 CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 932 sc->sc_count--; 933 934 /* Update the primary interface */ 935 if (lp == sc->sc_primary) { 936 uint8_t lladdr[LAGG_ADDR_LEN]; 937 938 if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL) 939 bzero(&lladdr, LAGG_ADDR_LEN); 940 else 941 bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN); 942 sc->sc_primary = lp0; 943 if (sc->sc_destroying == 0) { 944 bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen); 945 lagg_proto_lladdr(sc); 946 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp); 947 948 /* 949 * Update lladdr for each port (new primary needs update 950 * as well, to switch from old lladdr to its 'real' one). 951 * We can skip this if the lagg is being destroyed. 952 */ 953 CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 954 if_setlladdr(lp_ptr->lp_ifp, lladdr, 955 lp_ptr->lp_ifp->if_addrlen); 956 } 957 } 958 959 if (lp->lp_ifflags) 960 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 961 962 if (lp->lp_detaching == 0) { 963 lagg_setflags(lp, 0); 964 lagg_setcaps(lp, lp->lp_ifcapenable); 965 if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen); 966 } 967 968 /* 969 * free port and release it's ifnet reference after a grace period has 970 * elapsed. 971 */ 972 epoch_call(net_epoch_preempt, &lp->lp_epoch_ctx, lagg_port_destroy_cb); 973 /* Update lagg capabilities */ 974 lagg_capabilities(sc); 975 lagg_linkstate(sc); 976 977 return (0); 978 } 979 980 static int 981 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 982 { 983 struct lagg_reqport *rp = (struct lagg_reqport *)data; 984 struct lagg_softc *sc; 985 struct lagg_port *lp = NULL; 986 int error = 0; 987 988 /* Should be checked by the caller */ 989 switch (ifp->if_type) { 990 case IFT_IEEE8023ADLAG: 991 case IFT_INFINIBANDLAG: 992 if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 993 goto fallback; 994 break; 995 default: 996 goto fallback; 997 } 998 999 switch (cmd) { 1000 case SIOCGLAGGPORT: 1001 if (rp->rp_portname[0] == '\0' || 1002 ifunit(rp->rp_portname) != ifp) { 1003 error = EINVAL; 1004 break; 1005 } 1006 1007 LAGG_RLOCK(); 1008 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 1009 error = ENOENT; 1010 LAGG_RUNLOCK(); 1011 break; 1012 } 1013 1014 lagg_port2req(lp, rp); 1015 LAGG_RUNLOCK(); 1016 break; 1017 1018 case SIOCSIFCAP: 1019 if (lp->lp_ioctl == NULL) { 1020 error = EINVAL; 1021 break; 1022 } 1023 error = (*lp->lp_ioctl)(ifp, cmd, data); 1024 if (error) 1025 break; 1026 1027 /* Update lagg interface capabilities */ 1028 LAGG_XLOCK(sc); 1029 lagg_capabilities(sc); 1030 LAGG_XUNLOCK(sc); 1031 VLAN_CAPABILITIES(sc->sc_ifp); 1032 break; 1033 1034 case SIOCSIFMTU: 1035 /* Do not allow the MTU to be changed once joined */ 1036 error = EINVAL; 1037 break; 1038 1039 default: 1040 goto fallback; 1041 } 1042 1043 return (error); 1044 1045 fallback: 1046 if (lp != NULL && lp->lp_ioctl != NULL) 1047 return ((*lp->lp_ioctl)(ifp, cmd, data)); 1048 1049 return (EINVAL); 1050 } 1051 1052 /* 1053 * Requests counter @cnt data. 1054 * 1055 * Counter value is calculated the following way: 1056 * 1) for each port, sum difference between current and "initial" measurements. 1057 * 2) add lagg logical interface counters. 1058 * 3) add data from detached_counters array. 1059 * 1060 * We also do the following things on ports attach/detach: 1061 * 1) On port attach we store all counters it has into port_counter array. 1062 * 2) On port detach we add the different between "initial" and 1063 * current counters data to detached_counters array. 1064 */ 1065 static uint64_t 1066 lagg_get_counter(struct ifnet *ifp, ift_counter cnt) 1067 { 1068 struct lagg_softc *sc; 1069 struct lagg_port *lp; 1070 struct ifnet *lpifp; 1071 uint64_t newval, oldval, vsum; 1072 1073 /* Revise this when we've got non-generic counters. */ 1074 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt)); 1075 1076 sc = (struct lagg_softc *)ifp->if_softc; 1077 1078 vsum = 0; 1079 LAGG_RLOCK(); 1080 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1081 /* Saved attached value */ 1082 oldval = lp->port_counters.val[cnt]; 1083 /* current value */ 1084 lpifp = lp->lp_ifp; 1085 newval = lpifp->if_get_counter(lpifp, cnt); 1086 /* Calculate diff and save new */ 1087 vsum += newval - oldval; 1088 } 1089 LAGG_RUNLOCK(); 1090 1091 /* 1092 * Add counter data which might be added by upper 1093 * layer protocols operating on logical interface. 1094 */ 1095 vsum += if_get_counter_default(ifp, cnt); 1096 1097 /* 1098 * Add counter data from detached ports counters 1099 */ 1100 vsum += sc->detached_counters.val[cnt]; 1101 1102 1103 return (vsum); 1104 } 1105 1106 /* 1107 * For direct output to child ports. 1108 */ 1109 static int 1110 lagg_port_output(struct ifnet *ifp, struct mbuf *m, 1111 const struct sockaddr *dst, struct route *ro) 1112 { 1113 struct lagg_port *lp = ifp->if_lagg; 1114 1115 switch (dst->sa_family) { 1116 case pseudo_AF_HDRCMPLT: 1117 case AF_UNSPEC: 1118 if (lp != NULL) 1119 return ((*lp->lp_output)(ifp, m, dst, ro)); 1120 } 1121 1122 /* drop any other frames */ 1123 m_freem(m); 1124 return (ENETDOWN); 1125 } 1126 1127 static void 1128 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 1129 { 1130 struct lagg_port *lp; 1131 struct lagg_softc *sc; 1132 1133 if ((lp = ifp->if_lagg) == NULL) 1134 return; 1135 /* If the ifnet is just being renamed, don't do anything. */ 1136 if (ifp->if_flags & IFF_RENAMING) 1137 return; 1138 1139 sc = lp->lp_softc; 1140 1141 LAGG_XLOCK(sc); 1142 lp->lp_detaching = 1; 1143 lagg_port_destroy(lp, 1); 1144 LAGG_XUNLOCK(sc); 1145 VLAN_CAPABILITIES(sc->sc_ifp); 1146 } 1147 1148 static void 1149 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 1150 { 1151 struct lagg_softc *sc = lp->lp_softc; 1152 1153 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 1154 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 1155 rp->rp_prio = lp->lp_prio; 1156 rp->rp_flags = lp->lp_flags; 1157 lagg_proto_portreq(sc, lp, &rp->rp_psc); 1158 1159 /* Add protocol specific flags */ 1160 switch (sc->sc_proto) { 1161 case LAGG_PROTO_FAILOVER: 1162 if (lp == sc->sc_primary) 1163 rp->rp_flags |= LAGG_PORT_MASTER; 1164 if (lp == lagg_link_active(sc, sc->sc_primary)) 1165 rp->rp_flags |= LAGG_PORT_ACTIVE; 1166 break; 1167 1168 case LAGG_PROTO_ROUNDROBIN: 1169 case LAGG_PROTO_LOADBALANCE: 1170 case LAGG_PROTO_BROADCAST: 1171 if (LAGG_PORTACTIVE(lp)) 1172 rp->rp_flags |= LAGG_PORT_ACTIVE; 1173 break; 1174 1175 case LAGG_PROTO_LACP: 1176 /* LACP has a different definition of active */ 1177 if (lacp_isactive(lp)) 1178 rp->rp_flags |= LAGG_PORT_ACTIVE; 1179 if (lacp_iscollecting(lp)) 1180 rp->rp_flags |= LAGG_PORT_COLLECTING; 1181 if (lacp_isdistributing(lp)) 1182 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 1183 break; 1184 } 1185 1186 } 1187 1188 static void 1189 lagg_watchdog_infiniband(void *arg) 1190 { 1191 struct lagg_softc *sc; 1192 struct lagg_port *lp; 1193 struct ifnet *ifp; 1194 struct ifnet *lp_ifp; 1195 1196 sc = arg; 1197 1198 /* 1199 * Because infiniband nodes have a fixed MAC address, which is 1200 * generated by the so-called GID, we need to regularly update 1201 * the link level address of the parent lagg<N> device when 1202 * the active port changes. Possibly we could piggy-back on 1203 * link up/down events aswell, but using a timer also provides 1204 * a guarantee against too frequent events. This operation 1205 * does not have to be atomic. 1206 */ 1207 LAGG_RLOCK(); 1208 lp = lagg_link_active(sc, sc->sc_primary); 1209 if (lp != NULL) { 1210 ifp = sc->sc_ifp; 1211 lp_ifp = lp->lp_ifp; 1212 1213 if (ifp != NULL && lp_ifp != NULL && 1214 (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0 || 1215 memcmp(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen) != 0)) { 1216 memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen); 1217 memcpy(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen); 1218 1219 CURVNET_SET(ifp->if_vnet); 1220 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 1221 CURVNET_RESTORE(); 1222 } 1223 } 1224 LAGG_RUNLOCK(); 1225 1226 callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg); 1227 } 1228 1229 static void 1230 lagg_init(void *xsc) 1231 { 1232 struct lagg_softc *sc = (struct lagg_softc *)xsc; 1233 struct ifnet *ifp = sc->sc_ifp; 1234 struct lagg_port *lp; 1235 1236 LAGG_XLOCK(sc); 1237 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1238 LAGG_XUNLOCK(sc); 1239 return; 1240 } 1241 1242 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1243 1244 /* 1245 * Update the port lladdrs if needed. 1246 * This might be if_setlladdr() notification 1247 * that lladdr has been changed. 1248 */ 1249 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1250 if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp), 1251 ifp->if_addrlen) != 0) 1252 if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen); 1253 } 1254 1255 lagg_proto_init(sc); 1256 1257 if (ifp->if_type == IFT_INFINIBAND) { 1258 mtx_lock(&sc->sc_mtx); 1259 lagg_watchdog_infiniband(sc); 1260 mtx_unlock(&sc->sc_mtx); 1261 } 1262 1263 LAGG_XUNLOCK(sc); 1264 } 1265 1266 static void 1267 lagg_stop(struct lagg_softc *sc) 1268 { 1269 struct ifnet *ifp = sc->sc_ifp; 1270 1271 LAGG_XLOCK_ASSERT(sc); 1272 1273 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 1274 return; 1275 1276 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1277 1278 lagg_proto_stop(sc); 1279 1280 mtx_lock(&sc->sc_mtx); 1281 callout_stop(&sc->sc_watchdog); 1282 mtx_unlock(&sc->sc_mtx); 1283 1284 callout_drain(&sc->sc_watchdog); 1285 } 1286 1287 static int 1288 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1289 { 1290 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1291 struct lagg_reqall *ra = (struct lagg_reqall *)data; 1292 struct lagg_reqopts *ro = (struct lagg_reqopts *)data; 1293 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 1294 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 1295 struct ifreq *ifr = (struct ifreq *)data; 1296 struct lagg_port *lp; 1297 struct ifnet *tpif; 1298 struct thread *td = curthread; 1299 char *buf, *outbuf; 1300 int count, buflen, len, error = 0, oldmtu; 1301 1302 bzero(&rpbuf, sizeof(rpbuf)); 1303 1304 switch (cmd) { 1305 case SIOCGLAGG: 1306 LAGG_XLOCK(sc); 1307 buflen = sc->sc_count * sizeof(struct lagg_reqport); 1308 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 1309 ra->ra_proto = sc->sc_proto; 1310 lagg_proto_request(sc, &ra->ra_psc); 1311 count = 0; 1312 buf = outbuf; 1313 len = min(ra->ra_size, buflen); 1314 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1315 if (len < sizeof(rpbuf)) 1316 break; 1317 1318 lagg_port2req(lp, &rpbuf); 1319 memcpy(buf, &rpbuf, sizeof(rpbuf)); 1320 count++; 1321 buf += sizeof(rpbuf); 1322 len -= sizeof(rpbuf); 1323 } 1324 LAGG_XUNLOCK(sc); 1325 ra->ra_ports = count; 1326 ra->ra_size = count * sizeof(rpbuf); 1327 error = copyout(outbuf, ra->ra_port, ra->ra_size); 1328 free(outbuf, M_TEMP); 1329 break; 1330 case SIOCSLAGG: 1331 error = priv_check(td, PRIV_NET_LAGG); 1332 if (error) 1333 break; 1334 if (ra->ra_proto >= LAGG_PROTO_MAX) { 1335 error = EPROTONOSUPPORT; 1336 break; 1337 } 1338 /* Infiniband only supports the failover protocol. */ 1339 if (ra->ra_proto != LAGG_PROTO_FAILOVER && 1340 ifp->if_type == IFT_INFINIBAND) { 1341 error = EPROTONOSUPPORT; 1342 break; 1343 } 1344 LAGG_XLOCK(sc); 1345 lagg_proto_detach(sc); 1346 LAGG_UNLOCK_ASSERT(); 1347 lagg_proto_attach(sc, ra->ra_proto); 1348 LAGG_XUNLOCK(sc); 1349 break; 1350 case SIOCGLAGGOPTS: 1351 LAGG_XLOCK(sc); 1352 ro->ro_opts = sc->sc_opts; 1353 if (sc->sc_proto == LAGG_PROTO_LACP) { 1354 struct lacp_softc *lsc; 1355 1356 lsc = (struct lacp_softc *)sc->sc_psc; 1357 if (lsc->lsc_debug.lsc_tx_test != 0) 1358 ro->ro_opts |= LAGG_OPT_LACP_TXTEST; 1359 if (lsc->lsc_debug.lsc_rx_test != 0) 1360 ro->ro_opts |= LAGG_OPT_LACP_RXTEST; 1361 if (lsc->lsc_strict_mode != 0) 1362 ro->ro_opts |= LAGG_OPT_LACP_STRICT; 1363 if (lsc->lsc_fast_timeout != 0) 1364 ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO; 1365 1366 ro->ro_active = sc->sc_active; 1367 } else { 1368 ro->ro_active = 0; 1369 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1370 ro->ro_active += LAGG_PORTACTIVE(lp); 1371 } 1372 ro->ro_bkt = sc->sc_stride; 1373 ro->ro_flapping = sc->sc_flapping; 1374 ro->ro_flowid_shift = sc->flowid_shift; 1375 LAGG_XUNLOCK(sc); 1376 break; 1377 case SIOCSLAGGOPTS: 1378 error = priv_check(td, PRIV_NET_LAGG); 1379 if (error) 1380 break; 1381 1382 /* 1383 * The stride option was added without defining a corresponding 1384 * LAGG_OPT flag, so handle a non-zero value before checking 1385 * anything else to preserve compatibility. 1386 */ 1387 LAGG_XLOCK(sc); 1388 if (ro->ro_opts == 0 && ro->ro_bkt != 0) { 1389 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) { 1390 LAGG_XUNLOCK(sc); 1391 error = EINVAL; 1392 break; 1393 } 1394 sc->sc_stride = ro->ro_bkt; 1395 } 1396 if (ro->ro_opts == 0) { 1397 LAGG_XUNLOCK(sc); 1398 break; 1399 } 1400 1401 /* 1402 * Set options. LACP options are stored in sc->sc_psc, 1403 * not in sc_opts. 1404 */ 1405 int valid, lacp; 1406 1407 switch (ro->ro_opts) { 1408 case LAGG_OPT_USE_FLOWID: 1409 case -LAGG_OPT_USE_FLOWID: 1410 case LAGG_OPT_FLOWIDSHIFT: 1411 case LAGG_OPT_RR_LIMIT: 1412 valid = 1; 1413 lacp = 0; 1414 break; 1415 case LAGG_OPT_LACP_TXTEST: 1416 case -LAGG_OPT_LACP_TXTEST: 1417 case LAGG_OPT_LACP_RXTEST: 1418 case -LAGG_OPT_LACP_RXTEST: 1419 case LAGG_OPT_LACP_STRICT: 1420 case -LAGG_OPT_LACP_STRICT: 1421 case LAGG_OPT_LACP_FAST_TIMO: 1422 case -LAGG_OPT_LACP_FAST_TIMO: 1423 valid = lacp = 1; 1424 break; 1425 default: 1426 valid = lacp = 0; 1427 break; 1428 } 1429 1430 if (valid == 0 || 1431 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) { 1432 /* Invalid combination of options specified. */ 1433 error = EINVAL; 1434 LAGG_XUNLOCK(sc); 1435 break; /* Return from SIOCSLAGGOPTS. */ 1436 } 1437 1438 /* 1439 * Store new options into sc->sc_opts except for 1440 * FLOWIDSHIFT, RR and LACP options. 1441 */ 1442 if (lacp == 0) { 1443 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT) 1444 sc->flowid_shift = ro->ro_flowid_shift; 1445 else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) { 1446 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN || 1447 ro->ro_bkt == 0) { 1448 error = EINVAL; 1449 LAGG_XUNLOCK(sc); 1450 break; 1451 } 1452 sc->sc_stride = ro->ro_bkt; 1453 } else if (ro->ro_opts > 0) 1454 sc->sc_opts |= ro->ro_opts; 1455 else 1456 sc->sc_opts &= ~ro->ro_opts; 1457 } else { 1458 struct lacp_softc *lsc; 1459 struct lacp_port *lp; 1460 1461 lsc = (struct lacp_softc *)sc->sc_psc; 1462 1463 switch (ro->ro_opts) { 1464 case LAGG_OPT_LACP_TXTEST: 1465 lsc->lsc_debug.lsc_tx_test = 1; 1466 break; 1467 case -LAGG_OPT_LACP_TXTEST: 1468 lsc->lsc_debug.lsc_tx_test = 0; 1469 break; 1470 case LAGG_OPT_LACP_RXTEST: 1471 lsc->lsc_debug.lsc_rx_test = 1; 1472 break; 1473 case -LAGG_OPT_LACP_RXTEST: 1474 lsc->lsc_debug.lsc_rx_test = 0; 1475 break; 1476 case LAGG_OPT_LACP_STRICT: 1477 lsc->lsc_strict_mode = 1; 1478 break; 1479 case -LAGG_OPT_LACP_STRICT: 1480 lsc->lsc_strict_mode = 0; 1481 break; 1482 case LAGG_OPT_LACP_FAST_TIMO: 1483 LACP_LOCK(lsc); 1484 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1485 lp->lp_state |= LACP_STATE_TIMEOUT; 1486 LACP_UNLOCK(lsc); 1487 lsc->lsc_fast_timeout = 1; 1488 break; 1489 case -LAGG_OPT_LACP_FAST_TIMO: 1490 LACP_LOCK(lsc); 1491 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1492 lp->lp_state &= ~LACP_STATE_TIMEOUT; 1493 LACP_UNLOCK(lsc); 1494 lsc->lsc_fast_timeout = 0; 1495 break; 1496 } 1497 } 1498 LAGG_XUNLOCK(sc); 1499 break; 1500 case SIOCGLAGGFLAGS: 1501 rf->rf_flags = 0; 1502 LAGG_XLOCK(sc); 1503 if (sc->sc_flags & MBUF_HASHFLAG_L2) 1504 rf->rf_flags |= LAGG_F_HASHL2; 1505 if (sc->sc_flags & MBUF_HASHFLAG_L3) 1506 rf->rf_flags |= LAGG_F_HASHL3; 1507 if (sc->sc_flags & MBUF_HASHFLAG_L4) 1508 rf->rf_flags |= LAGG_F_HASHL4; 1509 LAGG_XUNLOCK(sc); 1510 break; 1511 case SIOCSLAGGHASH: 1512 error = priv_check(td, PRIV_NET_LAGG); 1513 if (error) 1514 break; 1515 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1516 error = EINVAL; 1517 break; 1518 } 1519 LAGG_XLOCK(sc); 1520 sc->sc_flags = 0; 1521 if (rf->rf_flags & LAGG_F_HASHL2) 1522 sc->sc_flags |= MBUF_HASHFLAG_L2; 1523 if (rf->rf_flags & LAGG_F_HASHL3) 1524 sc->sc_flags |= MBUF_HASHFLAG_L3; 1525 if (rf->rf_flags & LAGG_F_HASHL4) 1526 sc->sc_flags |= MBUF_HASHFLAG_L4; 1527 LAGG_XUNLOCK(sc); 1528 break; 1529 case SIOCGLAGGPORT: 1530 if (rp->rp_portname[0] == '\0' || 1531 (tpif = ifunit_ref(rp->rp_portname)) == NULL) { 1532 error = EINVAL; 1533 break; 1534 } 1535 1536 LAGG_RLOCK(); 1537 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1538 lp->lp_softc != sc) { 1539 error = ENOENT; 1540 LAGG_RUNLOCK(); 1541 if_rele(tpif); 1542 break; 1543 } 1544 1545 lagg_port2req(lp, rp); 1546 LAGG_RUNLOCK(); 1547 if_rele(tpif); 1548 break; 1549 case SIOCSLAGGPORT: 1550 error = priv_check(td, PRIV_NET_LAGG); 1551 if (error) 1552 break; 1553 if (rp->rp_portname[0] == '\0' || 1554 (tpif = ifunit_ref(rp->rp_portname)) == NULL) { 1555 error = EINVAL; 1556 break; 1557 } 1558 #ifdef INET6 1559 /* 1560 * A laggport interface should not have inet6 address 1561 * because two interfaces with a valid link-local 1562 * scope zone must not be merged in any form. This 1563 * restriction is needed to prevent violation of 1564 * link-local scope zone. Attempts to add a laggport 1565 * interface which has inet6 addresses triggers 1566 * removal of all inet6 addresses on the member 1567 * interface. 1568 */ 1569 if (in6ifa_llaonifp(tpif)) { 1570 in6_ifdetach(tpif); 1571 if_printf(sc->sc_ifp, 1572 "IPv6 addresses on %s have been removed " 1573 "before adding it as a member to prevent " 1574 "IPv6 address scope violation.\n", 1575 tpif->if_xname); 1576 } 1577 #endif 1578 oldmtu = ifp->if_mtu; 1579 LAGG_XLOCK(sc); 1580 error = lagg_port_create(sc, tpif); 1581 LAGG_XUNLOCK(sc); 1582 if_rele(tpif); 1583 1584 /* 1585 * LAGG MTU may change during addition of the first port. 1586 * If it did, do network layer specific procedure. 1587 */ 1588 if (ifp->if_mtu != oldmtu) { 1589 #ifdef INET6 1590 nd6_setmtu(ifp); 1591 #endif 1592 rt_updatemtu(ifp); 1593 } 1594 1595 VLAN_CAPABILITIES(ifp); 1596 break; 1597 case SIOCSLAGGDELPORT: 1598 error = priv_check(td, PRIV_NET_LAGG); 1599 if (error) 1600 break; 1601 if (rp->rp_portname[0] == '\0' || 1602 (tpif = ifunit_ref(rp->rp_portname)) == NULL) { 1603 error = EINVAL; 1604 break; 1605 } 1606 1607 LAGG_XLOCK(sc); 1608 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1609 lp->lp_softc != sc) { 1610 error = ENOENT; 1611 LAGG_XUNLOCK(sc); 1612 if_rele(tpif); 1613 break; 1614 } 1615 1616 error = lagg_port_destroy(lp, 1); 1617 LAGG_XUNLOCK(sc); 1618 if_rele(tpif); 1619 VLAN_CAPABILITIES(ifp); 1620 break; 1621 case SIOCSIFFLAGS: 1622 /* Set flags on ports too */ 1623 LAGG_XLOCK(sc); 1624 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1625 lagg_setflags(lp, 1); 1626 } 1627 1628 if (!(ifp->if_flags & IFF_UP) && 1629 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1630 /* 1631 * If interface is marked down and it is running, 1632 * then stop and disable it. 1633 */ 1634 lagg_stop(sc); 1635 LAGG_XUNLOCK(sc); 1636 } else if ((ifp->if_flags & IFF_UP) && 1637 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1638 /* 1639 * If interface is marked up and it is stopped, then 1640 * start it. 1641 */ 1642 LAGG_XUNLOCK(sc); 1643 (*ifp->if_init)(sc); 1644 } else 1645 LAGG_XUNLOCK(sc); 1646 break; 1647 case SIOCADDMULTI: 1648 case SIOCDELMULTI: 1649 LAGG_XLOCK(sc); 1650 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1651 lagg_clrmulti(lp); 1652 lagg_setmulti(lp); 1653 } 1654 LAGG_XUNLOCK(sc); 1655 error = 0; 1656 break; 1657 case SIOCSIFMEDIA: 1658 case SIOCGIFMEDIA: 1659 if (ifp->if_type == IFT_INFINIBAND) 1660 error = EINVAL; 1661 else 1662 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1663 break; 1664 1665 case SIOCSIFCAP: 1666 LAGG_XLOCK(sc); 1667 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1668 if (lp->lp_ioctl != NULL) 1669 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); 1670 } 1671 lagg_capabilities(sc); 1672 LAGG_XUNLOCK(sc); 1673 VLAN_CAPABILITIES(ifp); 1674 error = 0; 1675 break; 1676 1677 case SIOCSIFMTU: 1678 LAGG_XLOCK(sc); 1679 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1680 if (lp->lp_ioctl != NULL) 1681 error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); 1682 else 1683 error = EINVAL; 1684 if (error != 0) { 1685 if_printf(ifp, 1686 "failed to change MTU to %d on port %s, " 1687 "reverting all ports to original MTU (%d)\n", 1688 ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu); 1689 break; 1690 } 1691 } 1692 if (error == 0) { 1693 ifp->if_mtu = ifr->ifr_mtu; 1694 } else { 1695 /* set every port back to the original MTU */ 1696 ifr->ifr_mtu = ifp->if_mtu; 1697 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1698 if (lp->lp_ioctl != NULL) 1699 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); 1700 } 1701 } 1702 LAGG_XUNLOCK(sc); 1703 break; 1704 1705 default: 1706 error = ether_ioctl(ifp, cmd, data); 1707 break; 1708 } 1709 return (error); 1710 } 1711 1712 #ifdef RATELIMIT 1713 static int 1714 lagg_snd_tag_alloc(struct ifnet *ifp, 1715 union if_snd_tag_alloc_params *params, 1716 struct m_snd_tag **ppmt) 1717 { 1718 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1719 struct lagg_port *lp; 1720 struct lagg_lb *lb; 1721 uint32_t p; 1722 1723 LAGG_RLOCK(); 1724 switch (sc->sc_proto) { 1725 case LAGG_PROTO_FAILOVER: 1726 lp = lagg_link_active(sc, sc->sc_primary); 1727 break; 1728 case LAGG_PROTO_LOADBALANCE: 1729 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 || 1730 params->hdr.flowtype == M_HASHTYPE_NONE) { 1731 LAGG_RUNLOCK(); 1732 return (EOPNOTSUPP); 1733 } 1734 p = params->hdr.flowid >> sc->flowid_shift; 1735 p %= sc->sc_count; 1736 lb = (struct lagg_lb *)sc->sc_psc; 1737 lp = lb->lb_ports[p]; 1738 lp = lagg_link_active(sc, lp); 1739 break; 1740 case LAGG_PROTO_LACP: 1741 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 || 1742 params->hdr.flowtype == M_HASHTYPE_NONE) { 1743 LAGG_RUNLOCK(); 1744 return (EOPNOTSUPP); 1745 } 1746 lp = lacp_select_tx_port_by_hash(sc, params->hdr.flowid); 1747 break; 1748 default: 1749 LAGG_RUNLOCK(); 1750 return (EOPNOTSUPP); 1751 } 1752 if (lp == NULL) { 1753 LAGG_RUNLOCK(); 1754 return (EOPNOTSUPP); 1755 } 1756 ifp = lp->lp_ifp; 1757 LAGG_RUNLOCK(); 1758 if (ifp == NULL || ifp->if_snd_tag_alloc == NULL || 1759 (ifp->if_capenable & IFCAP_TXRTLMT) == 0) 1760 return (EOPNOTSUPP); 1761 1762 /* forward allocation request */ 1763 return (ifp->if_snd_tag_alloc(ifp, params, ppmt)); 1764 } 1765 #endif 1766 1767 static int 1768 lagg_setmulti(struct lagg_port *lp) 1769 { 1770 struct lagg_softc *sc = lp->lp_softc; 1771 struct ifnet *ifp = lp->lp_ifp; 1772 struct ifnet *scifp = sc->sc_ifp; 1773 struct lagg_mc *mc; 1774 struct ifmultiaddr *ifma; 1775 int error; 1776 1777 IF_ADDR_WLOCK(scifp); 1778 CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1779 if (ifma->ifma_addr->sa_family != AF_LINK) 1780 continue; 1781 mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT); 1782 if (mc == NULL) { 1783 IF_ADDR_WUNLOCK(scifp); 1784 return (ENOMEM); 1785 } 1786 bcopy(ifma->ifma_addr, &mc->mc_addr, 1787 ifma->ifma_addr->sa_len); 1788 mc->mc_addr.sdl_index = ifp->if_index; 1789 mc->mc_ifma = NULL; 1790 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1791 } 1792 IF_ADDR_WUNLOCK(scifp); 1793 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) { 1794 error = if_addmulti(ifp, 1795 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma); 1796 if (error) 1797 return (error); 1798 } 1799 return (0); 1800 } 1801 1802 static int 1803 lagg_clrmulti(struct lagg_port *lp) 1804 { 1805 struct lagg_mc *mc; 1806 1807 LAGG_XLOCK_ASSERT(lp->lp_softc); 1808 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1809 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1810 if (mc->mc_ifma && lp->lp_detaching == 0) 1811 if_delmulti_ifma(mc->mc_ifma); 1812 free(mc, M_LAGG); 1813 } 1814 return (0); 1815 } 1816 1817 static int 1818 lagg_setcaps(struct lagg_port *lp, int cap) 1819 { 1820 struct ifreq ifr; 1821 1822 if (lp->lp_ifp->if_capenable == cap) 1823 return (0); 1824 if (lp->lp_ioctl == NULL) 1825 return (ENXIO); 1826 ifr.ifr_reqcap = cap; 1827 return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr)); 1828 } 1829 1830 /* Handle a ref counted flag that should be set on the lagg port as well */ 1831 static int 1832 lagg_setflag(struct lagg_port *lp, int flag, int status, 1833 int (*func)(struct ifnet *, int)) 1834 { 1835 struct lagg_softc *sc = lp->lp_softc; 1836 struct ifnet *scifp = sc->sc_ifp; 1837 struct ifnet *ifp = lp->lp_ifp; 1838 int error; 1839 1840 LAGG_XLOCK_ASSERT(sc); 1841 1842 status = status ? (scifp->if_flags & flag) : 0; 1843 /* Now "status" contains the flag value or 0 */ 1844 1845 /* 1846 * See if recorded ports status is different from what 1847 * we want it to be. If it is, flip it. We record ports 1848 * status in lp_ifflags so that we won't clear ports flag 1849 * we haven't set. In fact, we don't clear or set ports 1850 * flags directly, but get or release references to them. 1851 * That's why we can be sure that recorded flags still are 1852 * in accord with actual ports flags. 1853 */ 1854 if (status != (lp->lp_ifflags & flag)) { 1855 error = (*func)(ifp, status); 1856 if (error) 1857 return (error); 1858 lp->lp_ifflags &= ~flag; 1859 lp->lp_ifflags |= status; 1860 } 1861 return (0); 1862 } 1863 1864 /* 1865 * Handle IFF_* flags that require certain changes on the lagg port 1866 * if "status" is true, update ports flags respective to the lagg 1867 * if "status" is false, forcedly clear the flags set on port. 1868 */ 1869 static int 1870 lagg_setflags(struct lagg_port *lp, int status) 1871 { 1872 int error, i; 1873 1874 for (i = 0; lagg_pflags[i].flag; i++) { 1875 error = lagg_setflag(lp, lagg_pflags[i].flag, 1876 status, lagg_pflags[i].func); 1877 if (error) 1878 return (error); 1879 } 1880 return (0); 1881 } 1882 1883 static int 1884 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m) 1885 { 1886 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1887 int error; 1888 1889 LAGG_RLOCK(); 1890 /* We need a Tx algorithm and at least one port */ 1891 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1892 LAGG_RUNLOCK(); 1893 m_freem(m); 1894 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1895 return (ENXIO); 1896 } 1897 1898 ETHER_BPF_MTAP(ifp, m); 1899 1900 error = lagg_proto_start(sc, m); 1901 LAGG_RUNLOCK(); 1902 1903 if (error != 0) 1904 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1905 1906 return (error); 1907 } 1908 1909 static int 1910 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m) 1911 { 1912 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1913 int error; 1914 1915 #if defined(KERN_TLS) || defined(RATELIMIT) 1916 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) 1917 MPASS(m->m_pkthdr.snd_tag->ifp == ifp); 1918 #endif 1919 LAGG_RLOCK(); 1920 /* We need a Tx algorithm and at least one port */ 1921 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1922 LAGG_RUNLOCK(); 1923 m_freem(m); 1924 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1925 return (ENXIO); 1926 } 1927 1928 INFINIBAND_BPF_MTAP(ifp, m); 1929 1930 error = lagg_proto_start(sc, m); 1931 LAGG_RUNLOCK(); 1932 return (error); 1933 } 1934 1935 /* 1936 * The ifp->if_qflush entry point for lagg(4) is no-op. 1937 */ 1938 static void 1939 lagg_qflush(struct ifnet *ifp __unused) 1940 { 1941 } 1942 1943 static struct mbuf * 1944 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m) 1945 { 1946 struct lagg_port *lp = ifp->if_lagg; 1947 struct lagg_softc *sc = lp->lp_softc; 1948 struct ifnet *scifp = sc->sc_ifp; 1949 1950 LAGG_RLOCK(); 1951 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1952 lp->lp_detaching != 0 || 1953 sc->sc_proto == LAGG_PROTO_NONE) { 1954 LAGG_RUNLOCK(); 1955 m_freem(m); 1956 return (NULL); 1957 } 1958 1959 ETHER_BPF_MTAP(scifp, m); 1960 1961 m = lagg_proto_input(sc, lp, m); 1962 if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) { 1963 m_freem(m); 1964 m = NULL; 1965 } 1966 1967 LAGG_RUNLOCK(); 1968 return (m); 1969 } 1970 1971 static struct mbuf * 1972 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m) 1973 { 1974 struct lagg_port *lp = ifp->if_lagg; 1975 struct lagg_softc *sc = lp->lp_softc; 1976 struct ifnet *scifp = sc->sc_ifp; 1977 1978 LAGG_RLOCK(); 1979 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1980 lp->lp_detaching != 0 || 1981 sc->sc_proto == LAGG_PROTO_NONE) { 1982 LAGG_RUNLOCK(); 1983 m_freem(m); 1984 return (NULL); 1985 } 1986 1987 INFINIBAND_BPF_MTAP(scifp, m); 1988 1989 m = lagg_proto_input(sc, lp, m); 1990 if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) { 1991 m_freem(m); 1992 m = NULL; 1993 } 1994 1995 LAGG_RUNLOCK(); 1996 return (m); 1997 } 1998 1999 static int 2000 lagg_media_change(struct ifnet *ifp) 2001 { 2002 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 2003 2004 if (sc->sc_ifflags & IFF_DEBUG) 2005 printf("%s\n", __func__); 2006 2007 /* Ignore */ 2008 return (0); 2009 } 2010 2011 static void 2012 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 2013 { 2014 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 2015 struct lagg_port *lp; 2016 2017 imr->ifm_status = IFM_AVALID; 2018 imr->ifm_active = IFM_ETHER | IFM_AUTO; 2019 2020 LAGG_RLOCK(); 2021 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 2022 if (LAGG_PORTACTIVE(lp)) 2023 imr->ifm_status |= IFM_ACTIVE; 2024 } 2025 LAGG_RUNLOCK(); 2026 } 2027 2028 static void 2029 lagg_linkstate(struct lagg_softc *sc) 2030 { 2031 struct lagg_port *lp; 2032 int new_link = LINK_STATE_DOWN; 2033 uint64_t speed; 2034 2035 LAGG_XLOCK_ASSERT(sc); 2036 2037 /* LACP handles link state itself */ 2038 if (sc->sc_proto == LAGG_PROTO_LACP) 2039 return; 2040 2041 /* Our link is considered up if at least one of our ports is active */ 2042 LAGG_RLOCK(); 2043 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 2044 if (lp->lp_ifp->if_link_state == LINK_STATE_UP) { 2045 new_link = LINK_STATE_UP; 2046 break; 2047 } 2048 } 2049 LAGG_RUNLOCK(); 2050 if_link_state_change(sc->sc_ifp, new_link); 2051 2052 /* Update if_baudrate to reflect the max possible speed */ 2053 switch (sc->sc_proto) { 2054 case LAGG_PROTO_FAILOVER: 2055 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 2056 sc->sc_primary->lp_ifp->if_baudrate : 0; 2057 break; 2058 case LAGG_PROTO_ROUNDROBIN: 2059 case LAGG_PROTO_LOADBALANCE: 2060 case LAGG_PROTO_BROADCAST: 2061 speed = 0; 2062 LAGG_RLOCK(); 2063 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2064 speed += lp->lp_ifp->if_baudrate; 2065 LAGG_RUNLOCK(); 2066 sc->sc_ifp->if_baudrate = speed; 2067 break; 2068 case LAGG_PROTO_LACP: 2069 /* LACP updates if_baudrate itself */ 2070 break; 2071 } 2072 } 2073 2074 static void 2075 lagg_port_state(struct ifnet *ifp, int state) 2076 { 2077 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 2078 struct lagg_softc *sc = NULL; 2079 2080 if (lp != NULL) 2081 sc = lp->lp_softc; 2082 if (sc == NULL) 2083 return; 2084 2085 LAGG_XLOCK(sc); 2086 lagg_linkstate(sc); 2087 lagg_proto_linkstate(sc, lp); 2088 LAGG_XUNLOCK(sc); 2089 } 2090 2091 struct lagg_port * 2092 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 2093 { 2094 struct lagg_port *lp_next, *rval = NULL; 2095 2096 /* 2097 * Search a port which reports an active link state. 2098 */ 2099 2100 #ifdef INVARIANTS 2101 /* 2102 * This is called with either LAGG_RLOCK() held or 2103 * LAGG_XLOCK(sc) held. 2104 */ 2105 if (!in_epoch(net_epoch_preempt)) 2106 LAGG_XLOCK_ASSERT(sc); 2107 #endif 2108 2109 if (lp == NULL) 2110 goto search; 2111 if (LAGG_PORTACTIVE(lp)) { 2112 rval = lp; 2113 goto found; 2114 } 2115 if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL && 2116 LAGG_PORTACTIVE(lp_next)) { 2117 rval = lp_next; 2118 goto found; 2119 } 2120 2121 search: 2122 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 2123 if (LAGG_PORTACTIVE(lp_next)) { 2124 return (lp_next); 2125 } 2126 } 2127 found: 2128 return (rval); 2129 } 2130 2131 int 2132 lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 2133 { 2134 2135 return (ifp->if_transmit)(ifp, m); 2136 } 2137 2138 /* 2139 * Simple round robin aggregation 2140 */ 2141 static void 2142 lagg_rr_attach(struct lagg_softc *sc) 2143 { 2144 sc->sc_seq = 0; 2145 sc->sc_stride = 1; 2146 } 2147 2148 static int 2149 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 2150 { 2151 struct lagg_port *lp; 2152 uint32_t p; 2153 2154 p = atomic_fetchadd_32(&sc->sc_seq, 1); 2155 p /= sc->sc_stride; 2156 p %= sc->sc_count; 2157 lp = CK_SLIST_FIRST(&sc->sc_ports); 2158 2159 while (p--) 2160 lp = CK_SLIST_NEXT(lp, lp_entries); 2161 2162 /* 2163 * Check the port's link state. This will return the next active 2164 * port if the link is down or the port is NULL. 2165 */ 2166 if ((lp = lagg_link_active(sc, lp)) == NULL) { 2167 m_freem(m); 2168 return (ENETDOWN); 2169 } 2170 2171 /* Send mbuf */ 2172 return (lagg_enqueue(lp->lp_ifp, m)); 2173 } 2174 2175 static struct mbuf * 2176 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2177 { 2178 struct ifnet *ifp = sc->sc_ifp; 2179 2180 /* Just pass in the packet to our lagg device */ 2181 m->m_pkthdr.rcvif = ifp; 2182 2183 return (m); 2184 } 2185 2186 /* 2187 * Broadcast mode 2188 */ 2189 static int 2190 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m) 2191 { 2192 int errors = 0; 2193 int ret; 2194 struct lagg_port *lp, *last = NULL; 2195 struct mbuf *m0; 2196 2197 LAGG_RLOCK_ASSERT(); 2198 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 2199 if (!LAGG_PORTACTIVE(lp)) 2200 continue; 2201 2202 if (last != NULL) { 2203 m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT); 2204 if (m0 == NULL) { 2205 ret = ENOBUFS; 2206 errors++; 2207 break; 2208 } 2209 2210 ret = lagg_enqueue(last->lp_ifp, m0); 2211 if (ret != 0) 2212 errors++; 2213 } 2214 last = lp; 2215 } 2216 2217 if (last == NULL) { 2218 m_freem(m); 2219 return (ENOENT); 2220 } 2221 if ((last = lagg_link_active(sc, last)) == NULL) { 2222 m_freem(m); 2223 return (ENETDOWN); 2224 } 2225 2226 ret = lagg_enqueue(last->lp_ifp, m); 2227 if (ret != 0) 2228 errors++; 2229 2230 if (errors == 0) 2231 return (ret); 2232 2233 return (0); 2234 } 2235 2236 static struct mbuf* 2237 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2238 { 2239 struct ifnet *ifp = sc->sc_ifp; 2240 2241 /* Just pass in the packet to our lagg device */ 2242 m->m_pkthdr.rcvif = ifp; 2243 return (m); 2244 } 2245 2246 /* 2247 * Active failover 2248 */ 2249 static int 2250 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 2251 { 2252 struct lagg_port *lp; 2253 2254 /* Use the master port if active or the next available port */ 2255 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 2256 m_freem(m); 2257 return (ENETDOWN); 2258 } 2259 2260 /* Send mbuf */ 2261 return (lagg_enqueue(lp->lp_ifp, m)); 2262 } 2263 2264 static struct mbuf * 2265 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2266 { 2267 struct ifnet *ifp = sc->sc_ifp; 2268 struct lagg_port *tmp_tp; 2269 2270 if (lp == sc->sc_primary || V_lagg_failover_rx_all) { 2271 m->m_pkthdr.rcvif = ifp; 2272 return (m); 2273 } 2274 2275 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 2276 tmp_tp = lagg_link_active(sc, sc->sc_primary); 2277 /* 2278 * If tmp_tp is null, we've received a packet when all 2279 * our links are down. Weird, but process it anyways. 2280 */ 2281 if ((tmp_tp == NULL || tmp_tp == lp)) { 2282 m->m_pkthdr.rcvif = ifp; 2283 return (m); 2284 } 2285 } 2286 2287 m_freem(m); 2288 return (NULL); 2289 } 2290 2291 /* 2292 * Loadbalancing 2293 */ 2294 static void 2295 lagg_lb_attach(struct lagg_softc *sc) 2296 { 2297 struct lagg_port *lp; 2298 struct lagg_lb *lb; 2299 2300 LAGG_XLOCK_ASSERT(sc); 2301 lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO); 2302 lb->lb_key = m_ether_tcpip_hash_init(); 2303 sc->sc_psc = lb; 2304 2305 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2306 lagg_lb_port_create(lp); 2307 } 2308 2309 static void 2310 lagg_lb_detach(struct lagg_softc *sc) 2311 { 2312 struct lagg_lb *lb; 2313 2314 lb = (struct lagg_lb *)sc->sc_psc; 2315 if (lb != NULL) 2316 free(lb, M_LAGG); 2317 } 2318 2319 static int 2320 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 2321 { 2322 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 2323 struct lagg_port *lp_next; 2324 int i = 0, rv; 2325 2326 rv = 0; 2327 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 2328 LAGG_XLOCK_ASSERT(sc); 2329 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 2330 if (lp_next == lp) 2331 continue; 2332 if (i >= LAGG_MAX_PORTS) { 2333 rv = EINVAL; 2334 break; 2335 } 2336 if (sc->sc_ifflags & IFF_DEBUG) 2337 printf("%s: port %s at index %d\n", 2338 sc->sc_ifname, lp_next->lp_ifp->if_xname, i); 2339 lb->lb_ports[i++] = lp_next; 2340 } 2341 2342 return (rv); 2343 } 2344 2345 static int 2346 lagg_lb_port_create(struct lagg_port *lp) 2347 { 2348 struct lagg_softc *sc = lp->lp_softc; 2349 return (lagg_lb_porttable(sc, NULL)); 2350 } 2351 2352 static void 2353 lagg_lb_port_destroy(struct lagg_port *lp) 2354 { 2355 struct lagg_softc *sc = lp->lp_softc; 2356 lagg_lb_porttable(sc, lp); 2357 } 2358 2359 static int 2360 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 2361 { 2362 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 2363 struct lagg_port *lp = NULL; 2364 uint32_t p = 0; 2365 2366 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && 2367 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 2368 p = m->m_pkthdr.flowid >> sc->flowid_shift; 2369 else 2370 p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key); 2371 p %= sc->sc_count; 2372 lp = lb->lb_ports[p]; 2373 2374 /* 2375 * Check the port's link state. This will return the next active 2376 * port if the link is down or the port is NULL. 2377 */ 2378 if ((lp = lagg_link_active(sc, lp)) == NULL) { 2379 m_freem(m); 2380 return (ENETDOWN); 2381 } 2382 2383 /* Send mbuf */ 2384 return (lagg_enqueue(lp->lp_ifp, m)); 2385 } 2386 2387 static struct mbuf * 2388 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2389 { 2390 struct ifnet *ifp = sc->sc_ifp; 2391 2392 /* Just pass in the packet to our lagg device */ 2393 m->m_pkthdr.rcvif = ifp; 2394 2395 return (m); 2396 } 2397 2398 /* 2399 * 802.3ad LACP 2400 */ 2401 static void 2402 lagg_lacp_attach(struct lagg_softc *sc) 2403 { 2404 struct lagg_port *lp; 2405 2406 lacp_attach(sc); 2407 LAGG_XLOCK_ASSERT(sc); 2408 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2409 lacp_port_create(lp); 2410 } 2411 2412 static void 2413 lagg_lacp_detach(struct lagg_softc *sc) 2414 { 2415 struct lagg_port *lp; 2416 void *psc; 2417 2418 LAGG_XLOCK_ASSERT(sc); 2419 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2420 lacp_port_destroy(lp); 2421 2422 psc = sc->sc_psc; 2423 sc->sc_psc = NULL; 2424 lacp_detach(psc); 2425 } 2426 2427 static void 2428 lagg_lacp_lladdr(struct lagg_softc *sc) 2429 { 2430 struct lagg_port *lp; 2431 2432 LAGG_SXLOCK_ASSERT(sc); 2433 2434 /* purge all the lacp ports */ 2435 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2436 lacp_port_destroy(lp); 2437 2438 /* add them back in */ 2439 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2440 lacp_port_create(lp); 2441 } 2442 2443 static int 2444 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 2445 { 2446 struct lagg_port *lp; 2447 2448 lp = lacp_select_tx_port(sc, m); 2449 if (lp == NULL) { 2450 m_freem(m); 2451 return (ENETDOWN); 2452 } 2453 2454 /* Send mbuf */ 2455 return (lagg_enqueue(lp->lp_ifp, m)); 2456 } 2457 2458 static struct mbuf * 2459 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2460 { 2461 struct ifnet *ifp = sc->sc_ifp; 2462 struct ether_header *eh; 2463 u_short etype; 2464 2465 eh = mtod(m, struct ether_header *); 2466 etype = ntohs(eh->ether_type); 2467 2468 /* Tap off LACP control messages */ 2469 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 2470 m = lacp_input(lp, m); 2471 if (m == NULL) 2472 return (NULL); 2473 } 2474 2475 /* 2476 * If the port is not collecting or not in the active aggregator then 2477 * free and return. 2478 */ 2479 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 2480 m_freem(m); 2481 return (NULL); 2482 } 2483 2484 m->m_pkthdr.rcvif = ifp; 2485 return (m); 2486 } 2487
Cache object: b2f7474d720ab58cc966bd28f0ec44a1
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]