1 /* $NetBSD: ieee8023ad_lacp.c,v 1.3 2005/12/11 12:24:54 christos Exp $ */ 2 3 /*- 4 * Copyright (c)2005 YAMAMOTO Takashi, 5 * Copyright (c)2008 Andrew Thompson <thompsa@FreeBSD.org> 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 __FBSDID("$FreeBSD$"); 32 33 #include <sys/param.h> 34 #include <sys/callout.h> 35 #include <sys/eventhandler.h> 36 #include <sys/mbuf.h> 37 #include <sys/systm.h> 38 #include <sys/malloc.h> 39 #include <sys/kernel.h> /* hz */ 40 #include <sys/socket.h> /* for net/if.h */ 41 #include <sys/sockio.h> 42 #include <sys/sysctl.h> 43 #include <machine/stdarg.h> 44 #include <sys/lock.h> 45 #include <sys/rwlock.h> 46 #include <sys/taskqueue.h> 47 48 #include <net/if.h> 49 #include <net/if_var.h> 50 #include <net/if_dl.h> 51 #include <net/ethernet.h> 52 #include <net/if_media.h> 53 #include <net/if_types.h> 54 55 #include <net/if_lagg.h> 56 #include <net/ieee8023ad_lacp.h> 57 58 /* 59 * actor system priority and port priority. 60 * XXX should be configurable. 61 */ 62 63 #define LACP_SYSTEM_PRIO 0x8000 64 #define LACP_PORT_PRIO 0x8000 65 66 const uint8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] = 67 { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 }; 68 69 static const struct tlv_template lacp_info_tlv_template[] = { 70 { LACP_TYPE_ACTORINFO, 71 sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) }, 72 { LACP_TYPE_PARTNERINFO, 73 sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) }, 74 { LACP_TYPE_COLLECTORINFO, 75 sizeof(struct tlvhdr) + sizeof(struct lacp_collectorinfo) }, 76 { 0, 0 }, 77 }; 78 79 static const struct tlv_template marker_info_tlv_template[] = { 80 { MARKER_TYPE_INFO, 81 sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) }, 82 { 0, 0 }, 83 }; 84 85 static const struct tlv_template marker_response_tlv_template[] = { 86 { MARKER_TYPE_RESPONSE, 87 sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) }, 88 { 0, 0 }, 89 }; 90 91 typedef void (*lacp_timer_func_t)(struct lacp_port *); 92 93 static void lacp_fill_actorinfo(struct lacp_port *, struct lacp_peerinfo *); 94 static void lacp_fill_markerinfo(struct lacp_port *, 95 struct lacp_markerinfo *); 96 97 static uint64_t lacp_aggregator_bandwidth(struct lacp_aggregator *); 98 static void lacp_suppress_distributing(struct lacp_softc *, 99 struct lacp_aggregator *); 100 static void lacp_transit_expire(void *); 101 static void lacp_update_portmap(struct lacp_softc *); 102 static void lacp_select_active_aggregator(struct lacp_softc *); 103 static uint16_t lacp_compose_key(struct lacp_port *); 104 static int tlv_check(const void *, size_t, const struct tlvhdr *, 105 const struct tlv_template *, boolean_t); 106 static void lacp_tick(void *); 107 108 static void lacp_fill_aggregator_id(struct lacp_aggregator *, 109 const struct lacp_port *); 110 static void lacp_fill_aggregator_id_peer(struct lacp_peerinfo *, 111 const struct lacp_peerinfo *); 112 static int lacp_aggregator_is_compatible(const struct lacp_aggregator *, 113 const struct lacp_port *); 114 static int lacp_peerinfo_is_compatible(const struct lacp_peerinfo *, 115 const struct lacp_peerinfo *); 116 117 static struct lacp_aggregator *lacp_aggregator_get(struct lacp_softc *, 118 struct lacp_port *); 119 static void lacp_aggregator_addref(struct lacp_softc *, 120 struct lacp_aggregator *); 121 static void lacp_aggregator_delref(struct lacp_softc *, 122 struct lacp_aggregator *); 123 124 /* receive machine */ 125 126 static int lacp_pdu_input(struct lacp_port *, struct mbuf *); 127 static int lacp_marker_input(struct lacp_port *, struct mbuf *); 128 static void lacp_sm_rx(struct lacp_port *, const struct lacpdu *); 129 static void lacp_sm_rx_timer(struct lacp_port *); 130 static void lacp_sm_rx_set_expired(struct lacp_port *); 131 static void lacp_sm_rx_update_ntt(struct lacp_port *, 132 const struct lacpdu *); 133 static void lacp_sm_rx_record_pdu(struct lacp_port *, 134 const struct lacpdu *); 135 static void lacp_sm_rx_update_selected(struct lacp_port *, 136 const struct lacpdu *); 137 static void lacp_sm_rx_record_default(struct lacp_port *); 138 static void lacp_sm_rx_update_default_selected(struct lacp_port *); 139 static void lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *, 140 const struct lacp_peerinfo *); 141 142 /* mux machine */ 143 144 static void lacp_sm_mux(struct lacp_port *); 145 static void lacp_set_mux(struct lacp_port *, enum lacp_mux_state); 146 static void lacp_sm_mux_timer(struct lacp_port *); 147 148 /* periodic transmit machine */ 149 150 static void lacp_sm_ptx_update_timeout(struct lacp_port *, uint8_t); 151 static void lacp_sm_ptx_tx_schedule(struct lacp_port *); 152 static void lacp_sm_ptx_timer(struct lacp_port *); 153 154 /* transmit machine */ 155 156 static void lacp_sm_tx(struct lacp_port *); 157 static void lacp_sm_assert_ntt(struct lacp_port *); 158 159 static void lacp_run_timers(struct lacp_port *); 160 static int lacp_compare_peerinfo(const struct lacp_peerinfo *, 161 const struct lacp_peerinfo *); 162 static int lacp_compare_systemid(const struct lacp_systemid *, 163 const struct lacp_systemid *); 164 static void lacp_port_enable(struct lacp_port *); 165 static void lacp_port_disable(struct lacp_port *); 166 static void lacp_select(struct lacp_port *); 167 static void lacp_unselect(struct lacp_port *); 168 static void lacp_disable_collecting(struct lacp_port *); 169 static void lacp_enable_collecting(struct lacp_port *); 170 static void lacp_disable_distributing(struct lacp_port *); 171 static void lacp_enable_distributing(struct lacp_port *); 172 static int lacp_xmit_lacpdu(struct lacp_port *); 173 static int lacp_xmit_marker(struct lacp_port *); 174 175 /* Debugging */ 176 177 static void lacp_dump_lacpdu(const struct lacpdu *); 178 static const char *lacp_format_partner(const struct lacp_peerinfo *, char *, 179 size_t); 180 static const char *lacp_format_lagid(const struct lacp_peerinfo *, 181 const struct lacp_peerinfo *, char *, size_t); 182 static const char *lacp_format_lagid_aggregator(const struct lacp_aggregator *, 183 char *, size_t); 184 static const char *lacp_format_state(uint8_t, char *, size_t); 185 static const char *lacp_format_mac(const uint8_t *, char *, size_t); 186 static const char *lacp_format_systemid(const struct lacp_systemid *, char *, 187 size_t); 188 static const char *lacp_format_portid(const struct lacp_portid *, char *, 189 size_t); 190 static void lacp_dprintf(const struct lacp_port *, const char *, ...) 191 __attribute__((__format__(__printf__, 2, 3))); 192 193 static int lacp_debug = 0; 194 SYSCTL_NODE(_net_link_lagg, OID_AUTO, lacp, CTLFLAG_RD, 0, "ieee802.3ad"); 195 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, debug, CTLFLAG_RW | CTLFLAG_TUN, 196 &lacp_debug, 0, "Enable LACP debug logging (1=debug, 2=trace)"); 197 TUNABLE_INT("net.link.lagg.lacp.debug", &lacp_debug); 198 199 #define LACP_DPRINTF(a) if (lacp_debug & 0x01) { lacp_dprintf a ; } 200 #define LACP_TRACE(a) if (lacp_debug & 0x02) { lacp_dprintf(a,"%s\n",__func__); } 201 #define LACP_TPRINTF(a) if (lacp_debug & 0x04) { lacp_dprintf a ; } 202 203 /* 204 * partner administration variables. 205 * XXX should be configurable. 206 */ 207 208 static const struct lacp_peerinfo lacp_partner_admin_optimistic = { 209 .lip_systemid = { .lsi_prio = 0xffff }, 210 .lip_portid = { .lpi_prio = 0xffff }, 211 .lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION | 212 LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING, 213 }; 214 215 static const struct lacp_peerinfo lacp_partner_admin_strict = { 216 .lip_systemid = { .lsi_prio = 0xffff }, 217 .lip_portid = { .lpi_prio = 0xffff }, 218 .lip_state = 0, 219 }; 220 221 static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = { 222 [LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer, 223 [LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer, 224 [LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer, 225 }; 226 227 struct mbuf * 228 lacp_input(struct lagg_port *lgp, struct mbuf *m) 229 { 230 struct lacp_port *lp = LACP_PORT(lgp); 231 uint8_t subtype; 232 233 if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) { 234 m_freem(m); 235 return (NULL); 236 } 237 238 m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype); 239 switch (subtype) { 240 case SLOWPROTOCOLS_SUBTYPE_LACP: 241 lacp_pdu_input(lp, m); 242 return (NULL); 243 244 case SLOWPROTOCOLS_SUBTYPE_MARKER: 245 lacp_marker_input(lp, m); 246 return (NULL); 247 } 248 249 /* Not a subtype we are interested in */ 250 return (m); 251 } 252 253 /* 254 * lacp_pdu_input: process lacpdu 255 */ 256 static int 257 lacp_pdu_input(struct lacp_port *lp, struct mbuf *m) 258 { 259 struct lacp_softc *lsc = lp->lp_lsc; 260 struct lacpdu *du; 261 int error = 0; 262 263 if (m->m_pkthdr.len != sizeof(*du)) { 264 goto bad; 265 } 266 267 if ((m->m_flags & M_MCAST) == 0) { 268 goto bad; 269 } 270 271 if (m->m_len < sizeof(*du)) { 272 m = m_pullup(m, sizeof(*du)); 273 if (m == NULL) { 274 return (ENOMEM); 275 } 276 } 277 278 du = mtod(m, struct lacpdu *); 279 280 if (memcmp(&du->ldu_eh.ether_dhost, 281 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) { 282 goto bad; 283 } 284 285 /* 286 * ignore the version for compatibility with 287 * the future protocol revisions. 288 */ 289 #if 0 290 if (du->ldu_sph.sph_version != 1) { 291 goto bad; 292 } 293 #endif 294 295 /* 296 * ignore tlv types for compatibility with 297 * the future protocol revisions. 298 */ 299 if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor, 300 lacp_info_tlv_template, FALSE)) { 301 goto bad; 302 } 303 304 if (lacp_debug > 0) { 305 lacp_dprintf(lp, "lacpdu receive\n"); 306 lacp_dump_lacpdu(du); 307 } 308 309 if ((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_rx_test) { 310 LACP_TPRINTF((lp, "Dropping RX PDU\n")); 311 goto bad; 312 } 313 314 LACP_LOCK(lsc); 315 lacp_sm_rx(lp, du); 316 LACP_UNLOCK(lsc); 317 318 m_freem(m); 319 return (error); 320 321 bad: 322 m_freem(m); 323 return (EINVAL); 324 } 325 326 static void 327 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info) 328 { 329 struct lagg_port *lgp = lp->lp_lagg; 330 struct lagg_softc *sc = lgp->lp_softc; 331 332 info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO); 333 memcpy(&info->lip_systemid.lsi_mac, 334 IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); 335 info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO); 336 info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index); 337 info->lip_state = lp->lp_state; 338 } 339 340 static void 341 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info) 342 { 343 struct ifnet *ifp = lp->lp_ifp; 344 345 /* Fill in the port index and system id (encoded as the MAC) */ 346 info->mi_rq_port = htons(ifp->if_index); 347 memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN); 348 info->mi_rq_xid = htonl(0); 349 } 350 351 static int 352 lacp_xmit_lacpdu(struct lacp_port *lp) 353 { 354 struct lagg_port *lgp = lp->lp_lagg; 355 struct mbuf *m; 356 struct lacpdu *du; 357 int error; 358 359 LACP_LOCK_ASSERT(lp->lp_lsc); 360 361 m = m_gethdr(M_NOWAIT, MT_DATA); 362 if (m == NULL) { 363 return (ENOMEM); 364 } 365 m->m_len = m->m_pkthdr.len = sizeof(*du); 366 367 du = mtod(m, struct lacpdu *); 368 memset(du, 0, sizeof(*du)); 369 370 memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols, 371 ETHER_ADDR_LEN); 372 memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN); 373 du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW); 374 375 du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP; 376 du->ldu_sph.sph_version = 1; 377 378 TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor)); 379 du->ldu_actor = lp->lp_actor; 380 381 TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO, 382 sizeof(du->ldu_partner)); 383 du->ldu_partner = lp->lp_partner; 384 385 TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO, 386 sizeof(du->ldu_collector)); 387 du->ldu_collector.lci_maxdelay = 0; 388 389 if (lacp_debug > 0) { 390 lacp_dprintf(lp, "lacpdu transmit\n"); 391 lacp_dump_lacpdu(du); 392 } 393 394 m->m_flags |= M_MCAST; 395 396 /* 397 * XXX should use higher priority queue. 398 * otherwise network congestion can break aggregation. 399 */ 400 401 error = lagg_enqueue(lp->lp_ifp, m); 402 return (error); 403 } 404 405 static int 406 lacp_xmit_marker(struct lacp_port *lp) 407 { 408 struct lagg_port *lgp = lp->lp_lagg; 409 struct mbuf *m; 410 struct markerdu *mdu; 411 int error; 412 413 LACP_LOCK_ASSERT(lp->lp_lsc); 414 415 m = m_gethdr(M_NOWAIT, MT_DATA); 416 if (m == NULL) { 417 return (ENOMEM); 418 } 419 m->m_len = m->m_pkthdr.len = sizeof(*mdu); 420 421 mdu = mtod(m, struct markerdu *); 422 memset(mdu, 0, sizeof(*mdu)); 423 424 memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols, 425 ETHER_ADDR_LEN); 426 memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN); 427 mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW); 428 429 mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER; 430 mdu->mdu_sph.sph_version = 1; 431 432 /* Bump the transaction id and copy over the marker info */ 433 lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1); 434 TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info)); 435 mdu->mdu_info = lp->lp_marker; 436 437 LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n", 438 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":", 439 ntohl(mdu->mdu_info.mi_rq_xid))); 440 441 m->m_flags |= M_MCAST; 442 error = lagg_enqueue(lp->lp_ifp, m); 443 return (error); 444 } 445 446 void 447 lacp_linkstate(struct lagg_port *lgp) 448 { 449 struct lacp_port *lp = LACP_PORT(lgp); 450 struct lacp_softc *lsc = lp->lp_lsc; 451 struct ifnet *ifp = lgp->lp_ifp; 452 struct ifmediareq ifmr; 453 int error = 0; 454 u_int media; 455 uint8_t old_state; 456 uint16_t old_key; 457 458 bzero((char *)&ifmr, sizeof(ifmr)); 459 error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr); 460 if (error != 0) 461 return; 462 463 LACP_LOCK(lsc); 464 media = ifmr.ifm_active; 465 LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, " 466 "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER, 467 (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP)); 468 old_state = lp->lp_state; 469 old_key = lp->lp_key; 470 471 lp->lp_media = media; 472 /* 473 * If the port is not an active full duplex Ethernet link then it can 474 * not be aggregated. 475 */ 476 if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 || 477 ifp->if_link_state != LINK_STATE_UP) { 478 lacp_port_disable(lp); 479 } else { 480 lacp_port_enable(lp); 481 } 482 lp->lp_key = lacp_compose_key(lp); 483 484 if (old_state != lp->lp_state || old_key != lp->lp_key) { 485 LACP_DPRINTF((lp, "-> UNSELECTED\n")); 486 lp->lp_selected = LACP_UNSELECTED; 487 } 488 LACP_UNLOCK(lsc); 489 } 490 491 static void 492 lacp_tick(void *arg) 493 { 494 struct lacp_softc *lsc = arg; 495 struct lacp_port *lp; 496 497 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) { 498 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) 499 continue; 500 501 lacp_run_timers(lp); 502 503 lacp_select(lp); 504 lacp_sm_mux(lp); 505 lacp_sm_tx(lp); 506 lacp_sm_ptx_tx_schedule(lp); 507 } 508 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc); 509 } 510 511 int 512 lacp_port_create(struct lagg_port *lgp) 513 { 514 struct lagg_softc *sc = lgp->lp_softc; 515 struct lacp_softc *lsc = LACP_SOFTC(sc); 516 struct lacp_port *lp; 517 struct ifnet *ifp = lgp->lp_ifp; 518 struct sockaddr_dl sdl; 519 struct ifmultiaddr *rifma = NULL; 520 int error; 521 522 boolean_t active = TRUE; /* XXX should be configurable */ 523 boolean_t fast = FALSE; /* XXX should be configurable */ 524 525 link_init_sdl(ifp, (struct sockaddr *)&sdl, IFT_ETHER); 526 sdl.sdl_alen = ETHER_ADDR_LEN; 527 528 bcopy(ðermulticastaddr_slowprotocols, 529 LLADDR(&sdl), ETHER_ADDR_LEN); 530 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 531 if (error) { 532 printf("%s: ADDMULTI failed on %s\n", __func__, lgp->lp_ifname); 533 return (error); 534 } 535 536 lp = malloc(sizeof(struct lacp_port), 537 M_DEVBUF, M_NOWAIT|M_ZERO); 538 if (lp == NULL) 539 return (ENOMEM); 540 541 LACP_LOCK(lsc); 542 lgp->lp_psc = (caddr_t)lp; 543 lp->lp_ifp = ifp; 544 lp->lp_lagg = lgp; 545 lp->lp_lsc = lsc; 546 lp->lp_ifma = rifma; 547 548 LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next); 549 550 lacp_fill_actorinfo(lp, &lp->lp_actor); 551 lacp_fill_markerinfo(lp, &lp->lp_marker); 552 lp->lp_state = 553 (active ? LACP_STATE_ACTIVITY : 0) | 554 (fast ? LACP_STATE_TIMEOUT : 0); 555 lp->lp_aggregator = NULL; 556 lacp_sm_rx_set_expired(lp); 557 LACP_UNLOCK(lsc); 558 lacp_linkstate(lgp); 559 560 return (0); 561 } 562 563 void 564 lacp_port_destroy(struct lagg_port *lgp) 565 { 566 struct lacp_port *lp = LACP_PORT(lgp); 567 struct lacp_softc *lsc = lp->lp_lsc; 568 int i; 569 570 LACP_LOCK(lsc); 571 for (i = 0; i < LACP_NTIMER; i++) { 572 LACP_TIMER_DISARM(lp, i); 573 } 574 575 lacp_disable_collecting(lp); 576 lacp_disable_distributing(lp); 577 lacp_unselect(lp); 578 579 /* The address may have already been removed by if_purgemaddrs() */ 580 if (!lgp->lp_detaching) 581 if_delmulti_ifma(lp->lp_ifma); 582 583 LIST_REMOVE(lp, lp_next); 584 LACP_UNLOCK(lsc); 585 free(lp, M_DEVBUF); 586 } 587 588 void 589 lacp_req(struct lagg_softc *sc, caddr_t data) 590 { 591 struct lacp_opreq *req = (struct lacp_opreq *)data; 592 struct lacp_softc *lsc = LACP_SOFTC(sc); 593 struct lacp_aggregator *la; 594 595 bzero(req, sizeof(struct lacp_opreq)); 596 597 /* 598 * If the LACP softc is NULL, return with the opreq structure full of 599 * zeros. It is normal for the softc to be NULL while the lagg is 600 * being destroyed. 601 */ 602 if (NULL == lsc) 603 return; 604 605 la = lsc->lsc_active_aggregator; 606 LACP_LOCK(lsc); 607 if (la != NULL) { 608 req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio); 609 memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac, 610 ETHER_ADDR_LEN); 611 req->actor_key = ntohs(la->la_actor.lip_key); 612 req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio); 613 req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno); 614 req->actor_state = la->la_actor.lip_state; 615 616 req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio); 617 memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac, 618 ETHER_ADDR_LEN); 619 req->partner_key = ntohs(la->la_partner.lip_key); 620 req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio); 621 req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno); 622 req->partner_state = la->la_partner.lip_state; 623 } 624 LACP_UNLOCK(lsc); 625 } 626 627 void 628 lacp_portreq(struct lagg_port *lgp, caddr_t data) 629 { 630 struct lacp_opreq *req = (struct lacp_opreq *)data; 631 struct lacp_port *lp = LACP_PORT(lgp); 632 struct lacp_softc *lsc = lp->lp_lsc; 633 634 LACP_LOCK(lsc); 635 req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio); 636 memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac, 637 ETHER_ADDR_LEN); 638 req->actor_key = ntohs(lp->lp_actor.lip_key); 639 req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio); 640 req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno); 641 req->actor_state = lp->lp_actor.lip_state; 642 643 req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio); 644 memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac, 645 ETHER_ADDR_LEN); 646 req->partner_key = ntohs(lp->lp_partner.lip_key); 647 req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio); 648 req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno); 649 req->partner_state = lp->lp_partner.lip_state; 650 LACP_UNLOCK(lsc); 651 } 652 653 static void 654 lacp_disable_collecting(struct lacp_port *lp) 655 { 656 LACP_DPRINTF((lp, "collecting disabled\n")); 657 lp->lp_state &= ~LACP_STATE_COLLECTING; 658 } 659 660 static void 661 lacp_enable_collecting(struct lacp_port *lp) 662 { 663 LACP_DPRINTF((lp, "collecting enabled\n")); 664 lp->lp_state |= LACP_STATE_COLLECTING; 665 } 666 667 static void 668 lacp_disable_distributing(struct lacp_port *lp) 669 { 670 struct lacp_aggregator *la = lp->lp_aggregator; 671 struct lacp_softc *lsc = lp->lp_lsc; 672 struct lagg_softc *sc = lsc->lsc_softc; 673 char buf[LACP_LAGIDSTR_MAX+1]; 674 675 LACP_LOCK_ASSERT(lsc); 676 677 if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) { 678 return; 679 } 680 681 KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports")); 682 KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports)); 683 KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid")); 684 685 LACP_DPRINTF((lp, "disable distributing on aggregator %s, " 686 "nports %d -> %d\n", 687 lacp_format_lagid_aggregator(la, buf, sizeof(buf)), 688 la->la_nports, la->la_nports - 1)); 689 690 TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q); 691 la->la_nports--; 692 sc->sc_active = la->la_nports; 693 694 if (lsc->lsc_active_aggregator == la) { 695 lacp_suppress_distributing(lsc, la); 696 lacp_select_active_aggregator(lsc); 697 /* regenerate the port map, the active aggregator has changed */ 698 lacp_update_portmap(lsc); 699 } 700 701 lp->lp_state &= ~LACP_STATE_DISTRIBUTING; 702 } 703 704 static void 705 lacp_enable_distributing(struct lacp_port *lp) 706 { 707 struct lacp_aggregator *la = lp->lp_aggregator; 708 struct lacp_softc *lsc = lp->lp_lsc; 709 struct lagg_softc *sc = lsc->lsc_softc; 710 char buf[LACP_LAGIDSTR_MAX+1]; 711 712 LACP_LOCK_ASSERT(lsc); 713 714 if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) { 715 return; 716 } 717 718 LACP_DPRINTF((lp, "enable distributing on aggregator %s, " 719 "nports %d -> %d\n", 720 lacp_format_lagid_aggregator(la, buf, sizeof(buf)), 721 la->la_nports, la->la_nports + 1)); 722 723 KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid")); 724 TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q); 725 la->la_nports++; 726 sc->sc_active = la->la_nports; 727 728 lp->lp_state |= LACP_STATE_DISTRIBUTING; 729 730 if (lsc->lsc_active_aggregator == la) { 731 lacp_suppress_distributing(lsc, la); 732 lacp_update_portmap(lsc); 733 } else 734 /* try to become the active aggregator */ 735 lacp_select_active_aggregator(lsc); 736 } 737 738 static void 739 lacp_transit_expire(void *vp) 740 { 741 struct lacp_softc *lsc = vp; 742 743 LACP_LOCK_ASSERT(lsc); 744 745 LACP_TRACE(NULL); 746 747 lsc->lsc_suppress_distributing = FALSE; 748 } 749 750 static void 751 lacp_attach_sysctl(struct lacp_softc *lsc, struct sysctl_oid *p_oid) 752 { 753 struct lagg_softc *sc = lsc->lsc_softc; 754 755 SYSCTL_ADD_UINT(&sc->ctx, SYSCTL_CHILDREN(p_oid), OID_AUTO, 756 "lacp_strict_mode", 757 CTLFLAG_RW, 758 &lsc->lsc_strict_mode, 759 lsc->lsc_strict_mode, 760 "Enable LACP strict mode"); 761 } 762 763 static void 764 lacp_attach_sysctl_debug(struct lacp_softc *lsc, struct sysctl_oid *p_oid) 765 { 766 struct lagg_softc *sc = lsc->lsc_softc; 767 struct sysctl_oid *oid; 768 769 /* Create a child of the parent lagg interface */ 770 oid = SYSCTL_ADD_NODE(&sc->ctx, SYSCTL_CHILDREN(p_oid), 771 OID_AUTO, "debug", CTLFLAG_RD, NULL, "DEBUG"); 772 773 SYSCTL_ADD_UINT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 774 "rx_test", 775 CTLFLAG_RW, 776 &lsc->lsc_debug.lsc_rx_test, 777 lsc->lsc_debug.lsc_rx_test, 778 "Bitmap of if_dunit entries to drop RX frames for"); 779 SYSCTL_ADD_UINT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 780 "tx_test", 781 CTLFLAG_RW, 782 &lsc->lsc_debug.lsc_tx_test, 783 lsc->lsc_debug.lsc_tx_test, 784 "Bitmap of if_dunit entries to drop TX frames for"); 785 } 786 787 int 788 lacp_attach(struct lagg_softc *sc) 789 { 790 struct lacp_softc *lsc; 791 struct sysctl_oid *oid; 792 793 lsc = malloc(sizeof(struct lacp_softc), 794 M_DEVBUF, M_NOWAIT|M_ZERO); 795 if (lsc == NULL) 796 return (ENOMEM); 797 798 sc->sc_psc = (caddr_t)lsc; 799 lsc->lsc_softc = sc; 800 801 lsc->lsc_hashkey = arc4random(); 802 lsc->lsc_active_aggregator = NULL; 803 lsc->lsc_strict_mode = 1; 804 LACP_LOCK_INIT(lsc); 805 TAILQ_INIT(&lsc->lsc_aggregators); 806 LIST_INIT(&lsc->lsc_ports); 807 808 /* Create a child of the parent lagg interface */ 809 oid = SYSCTL_ADD_NODE(&sc->ctx, SYSCTL_CHILDREN(sc->sc_oid), 810 OID_AUTO, "lacp", CTLFLAG_RD, NULL, "LACP"); 811 812 /* Attach sysctl nodes */ 813 lacp_attach_sysctl(lsc, oid); 814 lacp_attach_sysctl_debug(lsc, oid); 815 816 callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0); 817 callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0); 818 819 /* if the lagg is already up then do the same */ 820 if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) 821 lacp_init(sc); 822 823 return (0); 824 } 825 826 int 827 lacp_detach(struct lagg_softc *sc) 828 { 829 struct lacp_softc *lsc = LACP_SOFTC(sc); 830 831 KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators), 832 ("aggregators still active")); 833 KASSERT(lsc->lsc_active_aggregator == NULL, 834 ("aggregator still attached")); 835 836 sc->sc_psc = NULL; 837 callout_drain(&lsc->lsc_transit_callout); 838 callout_drain(&lsc->lsc_callout); 839 840 LACP_LOCK_DESTROY(lsc); 841 free(lsc, M_DEVBUF); 842 return (0); 843 } 844 845 void 846 lacp_init(struct lagg_softc *sc) 847 { 848 struct lacp_softc *lsc = LACP_SOFTC(sc); 849 850 LACP_LOCK(lsc); 851 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc); 852 LACP_UNLOCK(lsc); 853 } 854 855 void 856 lacp_stop(struct lagg_softc *sc) 857 { 858 struct lacp_softc *lsc = LACP_SOFTC(sc); 859 860 LACP_LOCK(lsc); 861 callout_stop(&lsc->lsc_transit_callout); 862 callout_stop(&lsc->lsc_callout); 863 LACP_UNLOCK(lsc); 864 } 865 866 struct lagg_port * 867 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m) 868 { 869 struct lacp_softc *lsc = LACP_SOFTC(sc); 870 struct lacp_portmap *pm; 871 struct lacp_port *lp; 872 uint32_t hash; 873 874 if (__predict_false(lsc->lsc_suppress_distributing)) { 875 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__)); 876 return (NULL); 877 } 878 879 pm = &lsc->lsc_pmap[lsc->lsc_activemap]; 880 if (pm->pm_count == 0) { 881 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__)); 882 return (NULL); 883 } 884 885 if (sc->use_flowid && (m->m_flags & M_FLOWID)) 886 hash = m->m_pkthdr.flowid >> sc->flowid_shift; 887 else 888 hash = lagg_hashmbuf(sc, m, lsc->lsc_hashkey); 889 hash %= pm->pm_count; 890 lp = pm->pm_map[hash]; 891 892 KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0, 893 ("aggregated port is not distributing")); 894 895 return (lp->lp_lagg); 896 } 897 /* 898 * lacp_suppress_distributing: drop transmit packets for a while 899 * to preserve packet ordering. 900 */ 901 902 static void 903 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la) 904 { 905 struct lacp_port *lp; 906 907 if (lsc->lsc_active_aggregator != la) { 908 return; 909 } 910 911 LACP_TRACE(NULL); 912 913 lsc->lsc_suppress_distributing = TRUE; 914 915 /* send a marker frame down each port to verify the queues are empty */ 916 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) { 917 lp->lp_flags |= LACP_PORT_MARK; 918 lacp_xmit_marker(lp); 919 } 920 921 /* set a timeout for the marker frames */ 922 callout_reset(&lsc->lsc_transit_callout, 923 LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc); 924 } 925 926 static int 927 lacp_compare_peerinfo(const struct lacp_peerinfo *a, 928 const struct lacp_peerinfo *b) 929 { 930 return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state))); 931 } 932 933 static int 934 lacp_compare_systemid(const struct lacp_systemid *a, 935 const struct lacp_systemid *b) 936 { 937 return (memcmp(a, b, sizeof(*a))); 938 } 939 940 #if 0 /* unused */ 941 static int 942 lacp_compare_portid(const struct lacp_portid *a, 943 const struct lacp_portid *b) 944 { 945 return (memcmp(a, b, sizeof(*a))); 946 } 947 #endif 948 949 static uint64_t 950 lacp_aggregator_bandwidth(struct lacp_aggregator *la) 951 { 952 struct lacp_port *lp; 953 uint64_t speed; 954 955 lp = TAILQ_FIRST(&la->la_ports); 956 if (lp == NULL) { 957 return (0); 958 } 959 960 speed = ifmedia_baudrate(lp->lp_media); 961 speed *= la->la_nports; 962 if (speed == 0) { 963 LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n", 964 lp->lp_media, la->la_nports)); 965 } 966 967 return (speed); 968 } 969 970 /* 971 * lacp_select_active_aggregator: select an aggregator to be used to transmit 972 * packets from lagg(4) interface. 973 */ 974 975 static void 976 lacp_select_active_aggregator(struct lacp_softc *lsc) 977 { 978 struct lacp_aggregator *la; 979 struct lacp_aggregator *best_la = NULL; 980 uint64_t best_speed = 0; 981 char buf[LACP_LAGIDSTR_MAX+1]; 982 983 LACP_TRACE(NULL); 984 985 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) { 986 uint64_t speed; 987 988 if (la->la_nports == 0) { 989 continue; 990 } 991 992 speed = lacp_aggregator_bandwidth(la); 993 LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n", 994 lacp_format_lagid_aggregator(la, buf, sizeof(buf)), 995 speed, la->la_nports)); 996 997 /* This aggregator is chosen if 998 * the partner has a better system priority 999 * or, the total aggregated speed is higher 1000 * or, it is already the chosen aggregator 1001 */ 1002 if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) < 1003 LACP_SYS_PRI(best_la->la_partner)) || 1004 speed > best_speed || 1005 (speed == best_speed && 1006 la == lsc->lsc_active_aggregator)) { 1007 best_la = la; 1008 best_speed = speed; 1009 } 1010 } 1011 1012 KASSERT(best_la == NULL || best_la->la_nports > 0, 1013 ("invalid aggregator refcnt")); 1014 KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports), 1015 ("invalid aggregator list")); 1016 1017 if (lsc->lsc_active_aggregator != best_la) { 1018 LACP_DPRINTF((NULL, "active aggregator changed\n")); 1019 LACP_DPRINTF((NULL, "old %s\n", 1020 lacp_format_lagid_aggregator(lsc->lsc_active_aggregator, 1021 buf, sizeof(buf)))); 1022 } else { 1023 LACP_DPRINTF((NULL, "active aggregator not changed\n")); 1024 } 1025 LACP_DPRINTF((NULL, "new %s\n", 1026 lacp_format_lagid_aggregator(best_la, buf, sizeof(buf)))); 1027 1028 if (lsc->lsc_active_aggregator != best_la) { 1029 lsc->lsc_active_aggregator = best_la; 1030 lacp_update_portmap(lsc); 1031 if (best_la) { 1032 lacp_suppress_distributing(lsc, best_la); 1033 } 1034 } 1035 } 1036 1037 /* 1038 * Updated the inactive portmap array with the new list of ports and 1039 * make it live. 1040 */ 1041 static void 1042 lacp_update_portmap(struct lacp_softc *lsc) 1043 { 1044 struct lagg_softc *sc = lsc->lsc_softc; 1045 struct lacp_aggregator *la; 1046 struct lacp_portmap *p; 1047 struct lacp_port *lp; 1048 uint64_t speed; 1049 u_int newmap; 1050 int i; 1051 1052 newmap = lsc->lsc_activemap == 0 ? 1 : 0; 1053 p = &lsc->lsc_pmap[newmap]; 1054 la = lsc->lsc_active_aggregator; 1055 speed = 0; 1056 bzero(p, sizeof(struct lacp_portmap)); 1057 1058 if (la != NULL && la->la_nports > 0) { 1059 p->pm_count = la->la_nports; 1060 i = 0; 1061 TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q) 1062 p->pm_map[i++] = lp; 1063 KASSERT(i == p->pm_count, ("Invalid port count")); 1064 speed = lacp_aggregator_bandwidth(la); 1065 } 1066 sc->sc_ifp->if_baudrate = speed; 1067 1068 /* switch the active portmap over */ 1069 atomic_store_rel_int(&lsc->lsc_activemap, newmap); 1070 LACP_DPRINTF((NULL, "Set table %d with %d ports\n", 1071 lsc->lsc_activemap, 1072 lsc->lsc_pmap[lsc->lsc_activemap].pm_count)); 1073 } 1074 1075 static uint16_t 1076 lacp_compose_key(struct lacp_port *lp) 1077 { 1078 struct lagg_port *lgp = lp->lp_lagg; 1079 struct lagg_softc *sc = lgp->lp_softc; 1080 u_int media = lp->lp_media; 1081 uint16_t key; 1082 1083 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) { 1084 1085 /* 1086 * non-aggregatable links should have unique keys. 1087 * 1088 * XXX this isn't really unique as if_index is 16 bit. 1089 */ 1090 1091 /* bit 0..14: (some bits of) if_index of this port */ 1092 key = lp->lp_ifp->if_index; 1093 /* bit 15: 1 */ 1094 key |= 0x8000; 1095 } else { 1096 u_int subtype = IFM_SUBTYPE(media); 1097 1098 KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type")); 1099 KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface")); 1100 1101 /* bit 0..4: IFM_SUBTYPE modulo speed */ 1102 switch (subtype) { 1103 case IFM_10_T: 1104 case IFM_10_2: 1105 case IFM_10_5: 1106 case IFM_10_STP: 1107 case IFM_10_FL: 1108 key = IFM_10_T; 1109 break; 1110 case IFM_100_TX: 1111 case IFM_100_FX: 1112 case IFM_100_T4: 1113 case IFM_100_VG: 1114 case IFM_100_T2: 1115 key = IFM_100_TX; 1116 break; 1117 case IFM_1000_SX: 1118 case IFM_1000_LX: 1119 case IFM_1000_CX: 1120 case IFM_1000_T: 1121 key = IFM_1000_SX; 1122 break; 1123 case IFM_10G_LR: 1124 case IFM_10G_SR: 1125 case IFM_10G_CX4: 1126 case IFM_10G_TWINAX: 1127 case IFM_10G_TWINAX_LONG: 1128 case IFM_10G_LRM: 1129 case IFM_10G_T: 1130 key = IFM_10G_LR; 1131 break; 1132 case IFM_40G_CR4: 1133 case IFM_40G_SR4: 1134 case IFM_40G_LR4: 1135 key = IFM_40G_CR4; 1136 break; 1137 default: 1138 key = subtype; 1139 } 1140 /* bit 5..14: (some bits of) if_index of lagg device */ 1141 key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5); 1142 /* bit 15: 0 */ 1143 } 1144 return (htons(key)); 1145 } 1146 1147 static void 1148 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la) 1149 { 1150 char buf[LACP_LAGIDSTR_MAX+1]; 1151 1152 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n", 1153 __func__, 1154 lacp_format_lagid(&la->la_actor, &la->la_partner, 1155 buf, sizeof(buf)), 1156 la->la_refcnt, la->la_refcnt + 1)); 1157 1158 KASSERT(la->la_refcnt > 0, ("refcount <= 0")); 1159 la->la_refcnt++; 1160 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount")); 1161 } 1162 1163 static void 1164 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la) 1165 { 1166 char buf[LACP_LAGIDSTR_MAX+1]; 1167 1168 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n", 1169 __func__, 1170 lacp_format_lagid(&la->la_actor, &la->la_partner, 1171 buf, sizeof(buf)), 1172 la->la_refcnt, la->la_refcnt - 1)); 1173 1174 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt")); 1175 la->la_refcnt--; 1176 if (la->la_refcnt > 0) { 1177 return; 1178 } 1179 1180 KASSERT(la->la_refcnt == 0, ("refcount not zero")); 1181 KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active")); 1182 1183 TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q); 1184 1185 free(la, M_DEVBUF); 1186 } 1187 1188 /* 1189 * lacp_aggregator_get: allocate an aggregator. 1190 */ 1191 1192 static struct lacp_aggregator * 1193 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp) 1194 { 1195 struct lacp_aggregator *la; 1196 1197 la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT); 1198 if (la) { 1199 la->la_refcnt = 1; 1200 la->la_nports = 0; 1201 TAILQ_INIT(&la->la_ports); 1202 la->la_pending = 0; 1203 TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q); 1204 } 1205 1206 return (la); 1207 } 1208 1209 /* 1210 * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port. 1211 */ 1212 1213 static void 1214 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp) 1215 { 1216 lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner); 1217 lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor); 1218 1219 la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION; 1220 } 1221 1222 static void 1223 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr, 1224 const struct lacp_peerinfo *lpi_port) 1225 { 1226 memset(lpi_aggr, 0, sizeof(*lpi_aggr)); 1227 lpi_aggr->lip_systemid = lpi_port->lip_systemid; 1228 lpi_aggr->lip_key = lpi_port->lip_key; 1229 } 1230 1231 /* 1232 * lacp_aggregator_is_compatible: check if a port can join to an aggregator. 1233 */ 1234 1235 static int 1236 lacp_aggregator_is_compatible(const struct lacp_aggregator *la, 1237 const struct lacp_port *lp) 1238 { 1239 if (!(lp->lp_state & LACP_STATE_AGGREGATION) || 1240 !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) { 1241 return (0); 1242 } 1243 1244 if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) { 1245 return (0); 1246 } 1247 1248 if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) { 1249 return (0); 1250 } 1251 1252 if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) { 1253 return (0); 1254 } 1255 1256 return (1); 1257 } 1258 1259 static int 1260 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a, 1261 const struct lacp_peerinfo *b) 1262 { 1263 if (memcmp(&a->lip_systemid, &b->lip_systemid, 1264 sizeof(a->lip_systemid))) { 1265 return (0); 1266 } 1267 1268 if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) { 1269 return (0); 1270 } 1271 1272 return (1); 1273 } 1274 1275 static void 1276 lacp_port_enable(struct lacp_port *lp) 1277 { 1278 lp->lp_state |= LACP_STATE_AGGREGATION; 1279 } 1280 1281 static void 1282 lacp_port_disable(struct lacp_port *lp) 1283 { 1284 lacp_set_mux(lp, LACP_MUX_DETACHED); 1285 1286 lp->lp_state &= ~LACP_STATE_AGGREGATION; 1287 lp->lp_selected = LACP_UNSELECTED; 1288 lacp_sm_rx_record_default(lp); 1289 lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION; 1290 lp->lp_state &= ~LACP_STATE_EXPIRED; 1291 } 1292 1293 /* 1294 * lacp_select: select an aggregator. create one if necessary. 1295 */ 1296 static void 1297 lacp_select(struct lacp_port *lp) 1298 { 1299 struct lacp_softc *lsc = lp->lp_lsc; 1300 struct lacp_aggregator *la; 1301 char buf[LACP_LAGIDSTR_MAX+1]; 1302 1303 if (lp->lp_aggregator) { 1304 return; 1305 } 1306 1307 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), 1308 ("timer_wait_while still active")); 1309 1310 LACP_DPRINTF((lp, "port lagid=%s\n", 1311 lacp_format_lagid(&lp->lp_actor, &lp->lp_partner, 1312 buf, sizeof(buf)))); 1313 1314 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) { 1315 if (lacp_aggregator_is_compatible(la, lp)) { 1316 break; 1317 } 1318 } 1319 1320 if (la == NULL) { 1321 la = lacp_aggregator_get(lsc, lp); 1322 if (la == NULL) { 1323 LACP_DPRINTF((lp, "aggregator creation failed\n")); 1324 1325 /* 1326 * will retry on the next tick. 1327 */ 1328 1329 return; 1330 } 1331 lacp_fill_aggregator_id(la, lp); 1332 LACP_DPRINTF((lp, "aggregator created\n")); 1333 } else { 1334 LACP_DPRINTF((lp, "compatible aggregator found\n")); 1335 if (la->la_refcnt == LACP_MAX_PORTS) 1336 return; 1337 lacp_aggregator_addref(lsc, la); 1338 } 1339 1340 LACP_DPRINTF((lp, "aggregator lagid=%s\n", 1341 lacp_format_lagid(&la->la_actor, &la->la_partner, 1342 buf, sizeof(buf)))); 1343 1344 lp->lp_aggregator = la; 1345 lp->lp_selected = LACP_SELECTED; 1346 } 1347 1348 /* 1349 * lacp_unselect: finish unselect/detach process. 1350 */ 1351 1352 static void 1353 lacp_unselect(struct lacp_port *lp) 1354 { 1355 struct lacp_softc *lsc = lp->lp_lsc; 1356 struct lacp_aggregator *la = lp->lp_aggregator; 1357 1358 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), 1359 ("timer_wait_while still active")); 1360 1361 if (la == NULL) { 1362 return; 1363 } 1364 1365 lp->lp_aggregator = NULL; 1366 lacp_aggregator_delref(lsc, la); 1367 } 1368 1369 /* mux machine */ 1370 1371 static void 1372 lacp_sm_mux(struct lacp_port *lp) 1373 { 1374 struct lagg_port *lgp = lp->lp_lagg; 1375 struct lagg_softc *sc = lgp->lp_softc; 1376 enum lacp_mux_state new_state; 1377 boolean_t p_sync = 1378 (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0; 1379 boolean_t p_collecting = 1380 (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0; 1381 enum lacp_selected selected = lp->lp_selected; 1382 struct lacp_aggregator *la; 1383 1384 if (lacp_debug > 1) 1385 lacp_dprintf(lp, "%s: state= 0x%x, selected= 0x%x, " 1386 "p_sync= 0x%x, p_collecting= 0x%x\n", __func__, 1387 lp->lp_mux_state, selected, p_sync, p_collecting); 1388 1389 re_eval: 1390 la = lp->lp_aggregator; 1391 KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL, 1392 ("MUX not detached")); 1393 new_state = lp->lp_mux_state; 1394 switch (lp->lp_mux_state) { 1395 case LACP_MUX_DETACHED: 1396 if (selected != LACP_UNSELECTED) { 1397 new_state = LACP_MUX_WAITING; 1398 } 1399 break; 1400 case LACP_MUX_WAITING: 1401 KASSERT(la->la_pending > 0 || 1402 !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), 1403 ("timer_wait_while still active")); 1404 if (selected == LACP_SELECTED && la->la_pending == 0) { 1405 new_state = LACP_MUX_ATTACHED; 1406 } else if (selected == LACP_UNSELECTED) { 1407 new_state = LACP_MUX_DETACHED; 1408 } 1409 break; 1410 case LACP_MUX_ATTACHED: 1411 if (selected == LACP_SELECTED && p_sync) { 1412 new_state = LACP_MUX_COLLECTING; 1413 } else if (selected != LACP_SELECTED) { 1414 new_state = LACP_MUX_DETACHED; 1415 } 1416 break; 1417 case LACP_MUX_COLLECTING: 1418 if (selected == LACP_SELECTED && p_sync && p_collecting) { 1419 new_state = LACP_MUX_DISTRIBUTING; 1420 } else if (selected != LACP_SELECTED || !p_sync) { 1421 new_state = LACP_MUX_ATTACHED; 1422 } 1423 break; 1424 case LACP_MUX_DISTRIBUTING: 1425 if (selected != LACP_SELECTED || !p_sync || !p_collecting) { 1426 new_state = LACP_MUX_COLLECTING; 1427 lacp_dprintf(lp, "Interface stopped DISTRIBUTING, possible flapping\n"); 1428 sc->sc_flapping++; 1429 } 1430 break; 1431 default: 1432 panic("%s: unknown state", __func__); 1433 } 1434 1435 if (lp->lp_mux_state == new_state) { 1436 return; 1437 } 1438 1439 lacp_set_mux(lp, new_state); 1440 goto re_eval; 1441 } 1442 1443 static void 1444 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state) 1445 { 1446 struct lacp_aggregator *la = lp->lp_aggregator; 1447 1448 if (lp->lp_mux_state == new_state) { 1449 return; 1450 } 1451 1452 switch (new_state) { 1453 case LACP_MUX_DETACHED: 1454 lp->lp_state &= ~LACP_STATE_SYNC; 1455 lacp_disable_distributing(lp); 1456 lacp_disable_collecting(lp); 1457 lacp_sm_assert_ntt(lp); 1458 /* cancel timer */ 1459 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) { 1460 KASSERT(la->la_pending > 0, 1461 ("timer_wait_while not active")); 1462 la->la_pending--; 1463 } 1464 LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE); 1465 lacp_unselect(lp); 1466 break; 1467 case LACP_MUX_WAITING: 1468 LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE, 1469 LACP_AGGREGATE_WAIT_TIME); 1470 la->la_pending++; 1471 break; 1472 case LACP_MUX_ATTACHED: 1473 lp->lp_state |= LACP_STATE_SYNC; 1474 lacp_disable_collecting(lp); 1475 lacp_sm_assert_ntt(lp); 1476 break; 1477 case LACP_MUX_COLLECTING: 1478 lacp_enable_collecting(lp); 1479 lacp_disable_distributing(lp); 1480 lacp_sm_assert_ntt(lp); 1481 break; 1482 case LACP_MUX_DISTRIBUTING: 1483 lacp_enable_distributing(lp); 1484 break; 1485 default: 1486 panic("%s: unknown state", __func__); 1487 } 1488 1489 LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state)); 1490 1491 lp->lp_mux_state = new_state; 1492 } 1493 1494 static void 1495 lacp_sm_mux_timer(struct lacp_port *lp) 1496 { 1497 struct lacp_aggregator *la = lp->lp_aggregator; 1498 char buf[LACP_LAGIDSTR_MAX+1]; 1499 1500 KASSERT(la->la_pending > 0, ("no pending event")); 1501 1502 LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__, 1503 lacp_format_lagid(&la->la_actor, &la->la_partner, 1504 buf, sizeof(buf)), 1505 la->la_pending, la->la_pending - 1)); 1506 1507 la->la_pending--; 1508 } 1509 1510 /* periodic transmit machine */ 1511 1512 static void 1513 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate) 1514 { 1515 if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state, 1516 LACP_STATE_TIMEOUT)) { 1517 return; 1518 } 1519 1520 LACP_DPRINTF((lp, "partner timeout changed\n")); 1521 1522 /* 1523 * FAST_PERIODIC -> SLOW_PERIODIC 1524 * or 1525 * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC 1526 * 1527 * let lacp_sm_ptx_tx_schedule to update timeout. 1528 */ 1529 1530 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC); 1531 1532 /* 1533 * if timeout has been shortened, assert NTT. 1534 */ 1535 1536 if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) { 1537 lacp_sm_assert_ntt(lp); 1538 } 1539 } 1540 1541 static void 1542 lacp_sm_ptx_tx_schedule(struct lacp_port *lp) 1543 { 1544 int timeout; 1545 1546 if (!(lp->lp_state & LACP_STATE_ACTIVITY) && 1547 !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) { 1548 1549 /* 1550 * NO_PERIODIC 1551 */ 1552 1553 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC); 1554 return; 1555 } 1556 1557 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) { 1558 return; 1559 } 1560 1561 timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ? 1562 LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME; 1563 1564 LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout); 1565 } 1566 1567 static void 1568 lacp_sm_ptx_timer(struct lacp_port *lp) 1569 { 1570 lacp_sm_assert_ntt(lp); 1571 } 1572 1573 static void 1574 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du) 1575 { 1576 int timeout; 1577 1578 /* 1579 * check LACP_DISABLED first 1580 */ 1581 1582 if (!(lp->lp_state & LACP_STATE_AGGREGATION)) { 1583 return; 1584 } 1585 1586 /* 1587 * check loopback condition. 1588 */ 1589 1590 if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid, 1591 &lp->lp_actor.lip_systemid)) { 1592 return; 1593 } 1594 1595 /* 1596 * EXPIRED, DEFAULTED, CURRENT -> CURRENT 1597 */ 1598 1599 lacp_sm_rx_update_selected(lp, du); 1600 lacp_sm_rx_update_ntt(lp, du); 1601 lacp_sm_rx_record_pdu(lp, du); 1602 1603 timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ? 1604 LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME; 1605 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout); 1606 1607 lp->lp_state &= ~LACP_STATE_EXPIRED; 1608 1609 /* 1610 * kick transmit machine without waiting the next tick. 1611 */ 1612 1613 lacp_sm_tx(lp); 1614 } 1615 1616 static void 1617 lacp_sm_rx_set_expired(struct lacp_port *lp) 1618 { 1619 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC; 1620 lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT; 1621 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME); 1622 lp->lp_state |= LACP_STATE_EXPIRED; 1623 } 1624 1625 static void 1626 lacp_sm_rx_timer(struct lacp_port *lp) 1627 { 1628 if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) { 1629 /* CURRENT -> EXPIRED */ 1630 LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__)); 1631 lacp_sm_rx_set_expired(lp); 1632 } else { 1633 /* EXPIRED -> DEFAULTED */ 1634 LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__)); 1635 lacp_sm_rx_update_default_selected(lp); 1636 lacp_sm_rx_record_default(lp); 1637 lp->lp_state &= ~LACP_STATE_EXPIRED; 1638 } 1639 } 1640 1641 static void 1642 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du) 1643 { 1644 boolean_t active; 1645 uint8_t oldpstate; 1646 char buf[LACP_STATESTR_MAX+1]; 1647 1648 LACP_TRACE(lp); 1649 1650 oldpstate = lp->lp_partner.lip_state; 1651 1652 active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY) 1653 || ((lp->lp_state & LACP_STATE_ACTIVITY) && 1654 (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY)); 1655 1656 lp->lp_partner = du->ldu_actor; 1657 if (active && 1658 ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state, 1659 LACP_STATE_AGGREGATION) && 1660 !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner)) 1661 || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) { 1662 /* XXX nothing? */ 1663 } else { 1664 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC; 1665 } 1666 1667 lp->lp_state &= ~LACP_STATE_DEFAULTED; 1668 1669 if (oldpstate != lp->lp_partner.lip_state) { 1670 LACP_DPRINTF((lp, "old pstate %s\n", 1671 lacp_format_state(oldpstate, buf, sizeof(buf)))); 1672 LACP_DPRINTF((lp, "new pstate %s\n", 1673 lacp_format_state(lp->lp_partner.lip_state, buf, 1674 sizeof(buf)))); 1675 } 1676 1677 /* XXX Hack, still need to implement 5.4.9 para 2,3,4 */ 1678 if (lp->lp_lsc->lsc_strict_mode) 1679 lp->lp_partner.lip_state |= LACP_STATE_SYNC; 1680 1681 lacp_sm_ptx_update_timeout(lp, oldpstate); 1682 } 1683 1684 static void 1685 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du) 1686 { 1687 1688 LACP_TRACE(lp); 1689 1690 if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) || 1691 !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state, 1692 LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) { 1693 LACP_DPRINTF((lp, "%s: assert ntt\n", __func__)); 1694 lacp_sm_assert_ntt(lp); 1695 } 1696 } 1697 1698 static void 1699 lacp_sm_rx_record_default(struct lacp_port *lp) 1700 { 1701 uint8_t oldpstate; 1702 1703 LACP_TRACE(lp); 1704 1705 oldpstate = lp->lp_partner.lip_state; 1706 if (lp->lp_lsc->lsc_strict_mode) 1707 lp->lp_partner = lacp_partner_admin_strict; 1708 else 1709 lp->lp_partner = lacp_partner_admin_optimistic;; 1710 lp->lp_state |= LACP_STATE_DEFAULTED; 1711 lacp_sm_ptx_update_timeout(lp, oldpstate); 1712 } 1713 1714 static void 1715 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp, 1716 const struct lacp_peerinfo *info) 1717 { 1718 1719 LACP_TRACE(lp); 1720 1721 if (lacp_compare_peerinfo(&lp->lp_partner, info) || 1722 !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state, 1723 LACP_STATE_AGGREGATION)) { 1724 lp->lp_selected = LACP_UNSELECTED; 1725 /* mux machine will clean up lp->lp_aggregator */ 1726 } 1727 } 1728 1729 static void 1730 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du) 1731 { 1732 1733 LACP_TRACE(lp); 1734 1735 lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor); 1736 } 1737 1738 static void 1739 lacp_sm_rx_update_default_selected(struct lacp_port *lp) 1740 { 1741 1742 LACP_TRACE(lp); 1743 1744 if (lp->lp_lsc->lsc_strict_mode) 1745 lacp_sm_rx_update_selected_from_peerinfo(lp, 1746 &lacp_partner_admin_strict); 1747 else 1748 lacp_sm_rx_update_selected_from_peerinfo(lp, 1749 &lacp_partner_admin_optimistic); 1750 } 1751 1752 /* transmit machine */ 1753 1754 static void 1755 lacp_sm_tx(struct lacp_port *lp) 1756 { 1757 int error = 0; 1758 1759 if (!(lp->lp_state & LACP_STATE_AGGREGATION) 1760 #if 1 1761 || (!(lp->lp_state & LACP_STATE_ACTIVITY) 1762 && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) 1763 #endif 1764 ) { 1765 lp->lp_flags &= ~LACP_PORT_NTT; 1766 } 1767 1768 if (!(lp->lp_flags & LACP_PORT_NTT)) { 1769 return; 1770 } 1771 1772 /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */ 1773 if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent, 1774 (3 / LACP_FAST_PERIODIC_TIME)) == 0) { 1775 LACP_DPRINTF((lp, "rate limited pdu\n")); 1776 return; 1777 } 1778 1779 if (((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_tx_test) == 0) { 1780 error = lacp_xmit_lacpdu(lp); 1781 } else { 1782 LACP_TPRINTF((lp, "Dropping TX PDU\n")); 1783 } 1784 1785 if (error == 0) { 1786 lp->lp_flags &= ~LACP_PORT_NTT; 1787 } else { 1788 LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n", 1789 error)); 1790 } 1791 } 1792 1793 static void 1794 lacp_sm_assert_ntt(struct lacp_port *lp) 1795 { 1796 1797 lp->lp_flags |= LACP_PORT_NTT; 1798 } 1799 1800 static void 1801 lacp_run_timers(struct lacp_port *lp) 1802 { 1803 int i; 1804 1805 for (i = 0; i < LACP_NTIMER; i++) { 1806 KASSERT(lp->lp_timer[i] >= 0, 1807 ("invalid timer value %d", lp->lp_timer[i])); 1808 if (lp->lp_timer[i] == 0) { 1809 continue; 1810 } else if (--lp->lp_timer[i] <= 0) { 1811 if (lacp_timer_funcs[i]) { 1812 (*lacp_timer_funcs[i])(lp); 1813 } 1814 } 1815 } 1816 } 1817 1818 int 1819 lacp_marker_input(struct lacp_port *lp, struct mbuf *m) 1820 { 1821 struct lacp_softc *lsc = lp->lp_lsc; 1822 struct lagg_port *lgp = lp->lp_lagg; 1823 struct lacp_port *lp2; 1824 struct markerdu *mdu; 1825 int error = 0; 1826 int pending = 0; 1827 1828 if (m->m_pkthdr.len != sizeof(*mdu)) { 1829 goto bad; 1830 } 1831 1832 if ((m->m_flags & M_MCAST) == 0) { 1833 goto bad; 1834 } 1835 1836 if (m->m_len < sizeof(*mdu)) { 1837 m = m_pullup(m, sizeof(*mdu)); 1838 if (m == NULL) { 1839 return (ENOMEM); 1840 } 1841 } 1842 1843 mdu = mtod(m, struct markerdu *); 1844 1845 if (memcmp(&mdu->mdu_eh.ether_dhost, 1846 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) { 1847 goto bad; 1848 } 1849 1850 if (mdu->mdu_sph.sph_version != 1) { 1851 goto bad; 1852 } 1853 1854 switch (mdu->mdu_tlv.tlv_type) { 1855 case MARKER_TYPE_INFO: 1856 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv, 1857 marker_info_tlv_template, TRUE)) { 1858 goto bad; 1859 } 1860 mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE; 1861 memcpy(&mdu->mdu_eh.ether_dhost, 1862 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN); 1863 memcpy(&mdu->mdu_eh.ether_shost, 1864 lgp->lp_lladdr, ETHER_ADDR_LEN); 1865 error = lagg_enqueue(lp->lp_ifp, m); 1866 break; 1867 1868 case MARKER_TYPE_RESPONSE: 1869 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv, 1870 marker_response_tlv_template, TRUE)) { 1871 goto bad; 1872 } 1873 LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n", 1874 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, 1875 ":", ntohl(mdu->mdu_info.mi_rq_xid))); 1876 1877 /* Verify that it is the last marker we sent out */ 1878 if (memcmp(&mdu->mdu_info, &lp->lp_marker, 1879 sizeof(struct lacp_markerinfo))) 1880 goto bad; 1881 1882 LACP_LOCK(lsc); 1883 lp->lp_flags &= ~LACP_PORT_MARK; 1884 1885 if (lsc->lsc_suppress_distributing) { 1886 /* Check if any ports are waiting for a response */ 1887 LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) { 1888 if (lp2->lp_flags & LACP_PORT_MARK) { 1889 pending = 1; 1890 break; 1891 } 1892 } 1893 1894 if (pending == 0) { 1895 /* All interface queues are clear */ 1896 LACP_DPRINTF((NULL, "queue flush complete\n")); 1897 lsc->lsc_suppress_distributing = FALSE; 1898 } 1899 } 1900 LACP_UNLOCK(lsc); 1901 m_freem(m); 1902 break; 1903 1904 default: 1905 goto bad; 1906 } 1907 1908 return (error); 1909 1910 bad: 1911 LACP_DPRINTF((lp, "bad marker frame\n")); 1912 m_freem(m); 1913 return (EINVAL); 1914 } 1915 1916 static int 1917 tlv_check(const void *p, size_t size, const struct tlvhdr *tlv, 1918 const struct tlv_template *tmpl, boolean_t check_type) 1919 { 1920 while (/* CONSTCOND */ 1) { 1921 if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) { 1922 return (EINVAL); 1923 } 1924 if ((check_type && tlv->tlv_type != tmpl->tmpl_type) || 1925 tlv->tlv_length != tmpl->tmpl_length) { 1926 return (EINVAL); 1927 } 1928 if (tmpl->tmpl_type == 0) { 1929 break; 1930 } 1931 tlv = (const struct tlvhdr *) 1932 ((const char *)tlv + tlv->tlv_length); 1933 tmpl++; 1934 } 1935 1936 return (0); 1937 } 1938 1939 /* Debugging */ 1940 const char * 1941 lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen) 1942 { 1943 snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X", 1944 (int)mac[0], 1945 (int)mac[1], 1946 (int)mac[2], 1947 (int)mac[3], 1948 (int)mac[4], 1949 (int)mac[5]); 1950 1951 return (buf); 1952 } 1953 1954 const char * 1955 lacp_format_systemid(const struct lacp_systemid *sysid, 1956 char *buf, size_t buflen) 1957 { 1958 char macbuf[LACP_MACSTR_MAX+1]; 1959 1960 snprintf(buf, buflen, "%04X,%s", 1961 ntohs(sysid->lsi_prio), 1962 lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf))); 1963 1964 return (buf); 1965 } 1966 1967 const char * 1968 lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen) 1969 { 1970 snprintf(buf, buflen, "%04X,%04X", 1971 ntohs(portid->lpi_prio), 1972 ntohs(portid->lpi_portno)); 1973 1974 return (buf); 1975 } 1976 1977 const char * 1978 lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen) 1979 { 1980 char sysid[LACP_SYSTEMIDSTR_MAX+1]; 1981 char portid[LACP_PORTIDSTR_MAX+1]; 1982 1983 snprintf(buf, buflen, "(%s,%04X,%s)", 1984 lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)), 1985 ntohs(peer->lip_key), 1986 lacp_format_portid(&peer->lip_portid, portid, sizeof(portid))); 1987 1988 return (buf); 1989 } 1990 1991 const char * 1992 lacp_format_lagid(const struct lacp_peerinfo *a, 1993 const struct lacp_peerinfo *b, char *buf, size_t buflen) 1994 { 1995 char astr[LACP_PARTNERSTR_MAX+1]; 1996 char bstr[LACP_PARTNERSTR_MAX+1]; 1997 1998 #if 0 1999 /* 2000 * there's a convention to display small numbered peer 2001 * in the left. 2002 */ 2003 2004 if (lacp_compare_peerinfo(a, b) > 0) { 2005 const struct lacp_peerinfo *t; 2006 2007 t = a; 2008 a = b; 2009 b = t; 2010 } 2011 #endif 2012 2013 snprintf(buf, buflen, "[%s,%s]", 2014 lacp_format_partner(a, astr, sizeof(astr)), 2015 lacp_format_partner(b, bstr, sizeof(bstr))); 2016 2017 return (buf); 2018 } 2019 2020 const char * 2021 lacp_format_lagid_aggregator(const struct lacp_aggregator *la, 2022 char *buf, size_t buflen) 2023 { 2024 if (la == NULL) { 2025 return ("(none)"); 2026 } 2027 2028 return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen)); 2029 } 2030 2031 const char * 2032 lacp_format_state(uint8_t state, char *buf, size_t buflen) 2033 { 2034 snprintf(buf, buflen, "%b", state, LACP_STATE_BITS); 2035 return (buf); 2036 } 2037 2038 static void 2039 lacp_dump_lacpdu(const struct lacpdu *du) 2040 { 2041 char buf[LACP_PARTNERSTR_MAX+1]; 2042 char buf2[LACP_STATESTR_MAX+1]; 2043 2044 printf("actor=%s\n", 2045 lacp_format_partner(&du->ldu_actor, buf, sizeof(buf))); 2046 printf("actor.state=%s\n", 2047 lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2))); 2048 printf("partner=%s\n", 2049 lacp_format_partner(&du->ldu_partner, buf, sizeof(buf))); 2050 printf("partner.state=%s\n", 2051 lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2))); 2052 2053 printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay)); 2054 } 2055 2056 static void 2057 lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...) 2058 { 2059 va_list va; 2060 2061 if (lp) { 2062 printf("%s: ", lp->lp_ifp->if_xname); 2063 } 2064 2065 va_start(va, fmt); 2066 vprintf(fmt, va); 2067 va_end(va); 2068 } 2069