1 /* $NetBSD: ieee8023ad_lacp.c,v 1.3 2005/12/11 12:24:54 christos Exp $ */ 2 3 /*- 4 * SPDX-License-Identifier: BSD-2-Clause-NetBSD 5 * 6 * Copyright (c)2005 YAMAMOTO Takashi, 7 * Copyright (c)2008 Andrew Thompson <thompsa@FreeBSD.org> 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_kern_tls.h" 36 #include "opt_ratelimit.h" 37 38 #include <sys/param.h> 39 #include <sys/callout.h> 40 #include <sys/eventhandler.h> 41 #include <sys/mbuf.h> 42 #include <sys/systm.h> 43 #include <sys/malloc.h> 44 #include <sys/kernel.h> /* hz */ 45 #include <sys/socket.h> /* for net/if.h */ 46 #include <sys/sockio.h> 47 #include <sys/sysctl.h> 48 #include <machine/stdarg.h> 49 #include <sys/lock.h> 50 #include <sys/rwlock.h> 51 #include <sys/taskqueue.h> 52 53 #include <net/if.h> 54 #include <net/if_var.h> 55 #include <net/if_dl.h> 56 #include <net/ethernet.h> 57 #include <net/if_media.h> 58 #include <net/if_types.h> 59 60 #include <net/if_lagg.h> 61 #include <net/ieee8023ad_lacp.h> 62 63 /* 64 * actor system priority and port priority. 65 * XXX should be configurable. 66 */ 67 68 #define LACP_SYSTEM_PRIO 0x8000 69 #define LACP_PORT_PRIO 0x8000 70 71 const uint8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] = 72 { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 }; 73 74 static const struct tlv_template lacp_info_tlv_template[] = { 75 { LACP_TYPE_ACTORINFO, 76 sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) }, 77 { LACP_TYPE_PARTNERINFO, 78 sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) }, 79 { LACP_TYPE_COLLECTORINFO, 80 sizeof(struct tlvhdr) + sizeof(struct lacp_collectorinfo) }, 81 { 0, 0 }, 82 }; 83 84 static const struct tlv_template marker_info_tlv_template[] = { 85 { MARKER_TYPE_INFO, 86 sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) }, 87 { 0, 0 }, 88 }; 89 90 static const struct tlv_template marker_response_tlv_template[] = { 91 { MARKER_TYPE_RESPONSE, 92 sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) }, 93 { 0, 0 }, 94 }; 95 96 typedef void (*lacp_timer_func_t)(struct lacp_port *); 97 98 static void lacp_fill_actorinfo(struct lacp_port *, struct lacp_peerinfo *); 99 static void lacp_fill_markerinfo(struct lacp_port *, 100 struct lacp_markerinfo *); 101 102 static uint64_t lacp_aggregator_bandwidth(struct lacp_aggregator *); 103 static void lacp_suppress_distributing(struct lacp_softc *, 104 struct lacp_aggregator *); 105 static void lacp_transit_expire(void *); 106 static void lacp_update_portmap(struct lacp_softc *); 107 static void lacp_select_active_aggregator(struct lacp_softc *); 108 static uint16_t lacp_compose_key(struct lacp_port *); 109 static int tlv_check(const void *, size_t, const struct tlvhdr *, 110 const struct tlv_template *, boolean_t); 111 static void lacp_tick(void *); 112 113 static void lacp_fill_aggregator_id(struct lacp_aggregator *, 114 const struct lacp_port *); 115 static void lacp_fill_aggregator_id_peer(struct lacp_peerinfo *, 116 const struct lacp_peerinfo *); 117 static int lacp_aggregator_is_compatible(const struct lacp_aggregator *, 118 const struct lacp_port *); 119 static int lacp_peerinfo_is_compatible(const struct lacp_peerinfo *, 120 const struct lacp_peerinfo *); 121 122 static struct lacp_aggregator *lacp_aggregator_get(struct lacp_softc *, 123 struct lacp_port *); 124 static void lacp_aggregator_addref(struct lacp_softc *, 125 struct lacp_aggregator *); 126 static void lacp_aggregator_delref(struct lacp_softc *, 127 struct lacp_aggregator *); 128 129 /* receive machine */ 130 131 static int lacp_pdu_input(struct lacp_port *, struct mbuf *); 132 static int lacp_marker_input(struct lacp_port *, struct mbuf *); 133 static void lacp_sm_rx(struct lacp_port *, const struct lacpdu *); 134 static void lacp_sm_rx_timer(struct lacp_port *); 135 static void lacp_sm_rx_set_expired(struct lacp_port *); 136 static void lacp_sm_rx_update_ntt(struct lacp_port *, 137 const struct lacpdu *); 138 static void lacp_sm_rx_record_pdu(struct lacp_port *, 139 const struct lacpdu *); 140 static void lacp_sm_rx_update_selected(struct lacp_port *, 141 const struct lacpdu *); 142 static void lacp_sm_rx_record_default(struct lacp_port *); 143 static void lacp_sm_rx_update_default_selected(struct lacp_port *); 144 static void lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *, 145 const struct lacp_peerinfo *); 146 147 /* mux machine */ 148 149 static void lacp_sm_mux(struct lacp_port *); 150 static void lacp_set_mux(struct lacp_port *, enum lacp_mux_state); 151 static void lacp_sm_mux_timer(struct lacp_port *); 152 153 /* periodic transmit machine */ 154 155 static void lacp_sm_ptx_update_timeout(struct lacp_port *, uint8_t); 156 static void lacp_sm_ptx_tx_schedule(struct lacp_port *); 157 static void lacp_sm_ptx_timer(struct lacp_port *); 158 159 /* transmit machine */ 160 161 static void lacp_sm_tx(struct lacp_port *); 162 static void lacp_sm_assert_ntt(struct lacp_port *); 163 164 static void lacp_run_timers(struct lacp_port *); 165 static int lacp_compare_peerinfo(const struct lacp_peerinfo *, 166 const struct lacp_peerinfo *); 167 static int lacp_compare_systemid(const struct lacp_systemid *, 168 const struct lacp_systemid *); 169 static void lacp_port_enable(struct lacp_port *); 170 static void lacp_port_disable(struct lacp_port *); 171 static void lacp_select(struct lacp_port *); 172 static void lacp_unselect(struct lacp_port *); 173 static void lacp_disable_collecting(struct lacp_port *); 174 static void lacp_enable_collecting(struct lacp_port *); 175 static void lacp_disable_distributing(struct lacp_port *); 176 static void lacp_enable_distributing(struct lacp_port *); 177 static int lacp_xmit_lacpdu(struct lacp_port *); 178 static int lacp_xmit_marker(struct lacp_port *); 179 180 /* Debugging */ 181 182 static void lacp_dump_lacpdu(const struct lacpdu *); 183 static const char *lacp_format_partner(const struct lacp_peerinfo *, char *, 184 size_t); 185 static const char *lacp_format_lagid(const struct lacp_peerinfo *, 186 const struct lacp_peerinfo *, char *, size_t); 187 static const char *lacp_format_lagid_aggregator(const struct lacp_aggregator *, 188 char *, size_t); 189 static const char *lacp_format_state(uint8_t, char *, size_t); 190 static const char *lacp_format_mac(const uint8_t *, char *, size_t); 191 static const char *lacp_format_systemid(const struct lacp_systemid *, char *, 192 size_t); 193 static const char *lacp_format_portid(const struct lacp_portid *, char *, 194 size_t); 195 static void lacp_dprintf(const struct lacp_port *, const char *, ...) 196 __attribute__((__format__(__printf__, 2, 3))); 197 198 VNET_DEFINE_STATIC(int, lacp_debug); 199 #define V_lacp_debug VNET(lacp_debug) 200 SYSCTL_NODE(_net_link_lagg, OID_AUTO, lacp, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 201 "ieee802.3ad"); 202 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, debug, CTLFLAG_RWTUN | CTLFLAG_VNET, 203 &VNET_NAME(lacp_debug), 0, "Enable LACP debug logging (1=debug, 2=trace)"); 204 205 VNET_DEFINE_STATIC(int, lacp_default_strict_mode) = 1; 206 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, default_strict_mode, 207 CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(lacp_default_strict_mode), 0, 208 "LACP strict protocol compliance default"); 209 #define LACP_DPRINTF(a) if (V_lacp_debug & 0x01) { lacp_dprintf a ; } 210 #define LACP_TRACE(a) if (V_lacp_debug & 0x02) { lacp_dprintf(a,"%s\n",__func__); } 211 #define LACP_TPRINTF(a) if (V_lacp_debug & 0x04) { lacp_dprintf a ; } 212 213 /* 214 * partner administration variables. 215 * XXX should be configurable. 216 */ 217 218 static const struct lacp_peerinfo lacp_partner_admin_optimistic = { 219 .lip_systemid = { .lsi_prio = 0xffff }, 220 .lip_portid = { .lpi_prio = 0xffff }, 221 .lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION | 222 LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING, 223 }; 224 225 static const struct lacp_peerinfo lacp_partner_admin_strict = { 226 .lip_systemid = { .lsi_prio = 0xffff }, 227 .lip_portid = { .lpi_prio = 0xffff }, 228 .lip_state = 0, 229 }; 230 231 static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = { 232 [LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer, 233 [LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer, 234 [LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer, 235 }; 236 237 struct mbuf * 238 lacp_input(struct lagg_port *lgp, struct mbuf *m) 239 { 240 struct lacp_port *lp = LACP_PORT(lgp); 241 uint8_t subtype; 242 243 if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) { 244 m_freem(m); 245 return (NULL); 246 } 247 248 m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype); 249 switch (subtype) { 250 case SLOWPROTOCOLS_SUBTYPE_LACP: 251 lacp_pdu_input(lp, m); 252 return (NULL); 253 254 case SLOWPROTOCOLS_SUBTYPE_MARKER: 255 lacp_marker_input(lp, m); 256 return (NULL); 257 } 258 259 /* Not a subtype we are interested in */ 260 return (m); 261 } 262 263 /* 264 * lacp_pdu_input: process lacpdu 265 */ 266 static int 267 lacp_pdu_input(struct lacp_port *lp, struct mbuf *m) 268 { 269 struct lacp_softc *lsc = lp->lp_lsc; 270 struct lacpdu *du; 271 int error = 0; 272 273 if (m->m_pkthdr.len != sizeof(*du)) { 274 goto bad; 275 } 276 277 if ((m->m_flags & M_MCAST) == 0) { 278 goto bad; 279 } 280 281 if (m->m_len < sizeof(*du)) { 282 m = m_pullup(m, sizeof(*du)); 283 if (m == NULL) { 284 return (ENOMEM); 285 } 286 } 287 288 du = mtod(m, struct lacpdu *); 289 290 if (memcmp(&du->ldu_eh.ether_dhost, 291 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) { 292 goto bad; 293 } 294 295 /* 296 * ignore the version for compatibility with 297 * the future protocol revisions. 298 */ 299 #if 0 300 if (du->ldu_sph.sph_version != 1) { 301 goto bad; 302 } 303 #endif 304 305 /* 306 * ignore tlv types for compatibility with 307 * the future protocol revisions. 308 */ 309 if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor, 310 lacp_info_tlv_template, FALSE)) { 311 goto bad; 312 } 313 314 if (V_lacp_debug > 0) { 315 lacp_dprintf(lp, "lacpdu receive\n"); 316 lacp_dump_lacpdu(du); 317 } 318 319 if ((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_rx_test) { 320 LACP_TPRINTF((lp, "Dropping RX PDU\n")); 321 goto bad; 322 } 323 324 LACP_LOCK(lsc); 325 lacp_sm_rx(lp, du); 326 LACP_UNLOCK(lsc); 327 328 m_freem(m); 329 return (error); 330 331 bad: 332 m_freem(m); 333 return (EINVAL); 334 } 335 336 static void 337 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info) 338 { 339 struct lagg_port *lgp = lp->lp_lagg; 340 struct lagg_softc *sc = lgp->lp_softc; 341 342 info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO); 343 memcpy(&info->lip_systemid.lsi_mac, 344 IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); 345 info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO); 346 info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index); 347 info->lip_state = lp->lp_state; 348 } 349 350 static void 351 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info) 352 { 353 struct ifnet *ifp = lp->lp_ifp; 354 355 /* Fill in the port index and system id (encoded as the MAC) */ 356 info->mi_rq_port = htons(ifp->if_index); 357 memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN); 358 info->mi_rq_xid = htonl(0); 359 } 360 361 static int 362 lacp_xmit_lacpdu(struct lacp_port *lp) 363 { 364 struct lagg_port *lgp = lp->lp_lagg; 365 struct mbuf *m; 366 struct lacpdu *du; 367 int error; 368 369 LACP_LOCK_ASSERT(lp->lp_lsc); 370 371 m = m_gethdr(M_NOWAIT, MT_DATA); 372 if (m == NULL) { 373 return (ENOMEM); 374 } 375 m->m_len = m->m_pkthdr.len = sizeof(*du); 376 377 du = mtod(m, struct lacpdu *); 378 memset(du, 0, sizeof(*du)); 379 380 memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols, 381 ETHER_ADDR_LEN); 382 memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN); 383 du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW); 384 385 du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP; 386 du->ldu_sph.sph_version = 1; 387 388 TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor)); 389 du->ldu_actor = lp->lp_actor; 390 391 TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO, 392 sizeof(du->ldu_partner)); 393 du->ldu_partner = lp->lp_partner; 394 395 TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO, 396 sizeof(du->ldu_collector)); 397 du->ldu_collector.lci_maxdelay = 0; 398 399 if (V_lacp_debug > 0) { 400 lacp_dprintf(lp, "lacpdu transmit\n"); 401 lacp_dump_lacpdu(du); 402 } 403 404 m->m_flags |= M_MCAST; 405 406 /* 407 * XXX should use higher priority queue. 408 * otherwise network congestion can break aggregation. 409 */ 410 411 error = lagg_enqueue(lp->lp_ifp, m); 412 return (error); 413 } 414 415 static int 416 lacp_xmit_marker(struct lacp_port *lp) 417 { 418 struct lagg_port *lgp = lp->lp_lagg; 419 struct mbuf *m; 420 struct markerdu *mdu; 421 int error; 422 423 LACP_LOCK_ASSERT(lp->lp_lsc); 424 425 m = m_gethdr(M_NOWAIT, MT_DATA); 426 if (m == NULL) { 427 return (ENOMEM); 428 } 429 m->m_len = m->m_pkthdr.len = sizeof(*mdu); 430 431 mdu = mtod(m, struct markerdu *); 432 memset(mdu, 0, sizeof(*mdu)); 433 434 memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols, 435 ETHER_ADDR_LEN); 436 memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN); 437 mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW); 438 439 mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER; 440 mdu->mdu_sph.sph_version = 1; 441 442 /* Bump the transaction id and copy over the marker info */ 443 lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1); 444 TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info)); 445 mdu->mdu_info = lp->lp_marker; 446 447 LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n", 448 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":", 449 ntohl(mdu->mdu_info.mi_rq_xid))); 450 451 m->m_flags |= M_MCAST; 452 error = lagg_enqueue(lp->lp_ifp, m); 453 return (error); 454 } 455 456 void 457 lacp_linkstate(struct lagg_port *lgp) 458 { 459 struct lacp_port *lp = LACP_PORT(lgp); 460 struct lacp_softc *lsc = lp->lp_lsc; 461 struct ifnet *ifp = lgp->lp_ifp; 462 struct ifmediareq ifmr; 463 int error = 0; 464 u_int media; 465 uint8_t old_state; 466 uint16_t old_key; 467 468 bzero((char *)&ifmr, sizeof(ifmr)); 469 error = (*ifp->if_ioctl)(ifp, SIOCGIFXMEDIA, (caddr_t)&ifmr); 470 if (error != 0) { 471 bzero((char *)&ifmr, sizeof(ifmr)); 472 error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr); 473 } 474 if (error != 0) 475 return; 476 477 LACP_LOCK(lsc); 478 media = ifmr.ifm_active; 479 LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, " 480 "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER, 481 (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP)); 482 old_state = lp->lp_state; 483 old_key = lp->lp_key; 484 485 lp->lp_media = media; 486 /* 487 * If the port is not an active full duplex Ethernet link then it can 488 * not be aggregated. 489 */ 490 if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 || 491 ifp->if_link_state != LINK_STATE_UP) { 492 lacp_port_disable(lp); 493 } else { 494 lacp_port_enable(lp); 495 } 496 lp->lp_key = lacp_compose_key(lp); 497 498 if (old_state != lp->lp_state || old_key != lp->lp_key) { 499 LACP_DPRINTF((lp, "-> UNSELECTED\n")); 500 lp->lp_selected = LACP_UNSELECTED; 501 } 502 LACP_UNLOCK(lsc); 503 } 504 505 static void 506 lacp_tick(void *arg) 507 { 508 struct lacp_softc *lsc = arg; 509 struct lacp_port *lp; 510 511 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) { 512 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) 513 continue; 514 515 CURVNET_SET(lp->lp_ifp->if_vnet); 516 lacp_run_timers(lp); 517 518 lacp_select(lp); 519 lacp_sm_mux(lp); 520 lacp_sm_tx(lp); 521 lacp_sm_ptx_tx_schedule(lp); 522 CURVNET_RESTORE(); 523 } 524 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc); 525 } 526 527 int 528 lacp_port_create(struct lagg_port *lgp) 529 { 530 struct lagg_softc *sc = lgp->lp_softc; 531 struct lacp_softc *lsc = LACP_SOFTC(sc); 532 struct lacp_port *lp; 533 struct ifnet *ifp = lgp->lp_ifp; 534 struct sockaddr_dl sdl; 535 struct ifmultiaddr *rifma = NULL; 536 int error; 537 538 link_init_sdl(ifp, (struct sockaddr *)&sdl, IFT_ETHER); 539 sdl.sdl_alen = ETHER_ADDR_LEN; 540 541 bcopy(ðermulticastaddr_slowprotocols, 542 LLADDR(&sdl), ETHER_ADDR_LEN); 543 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 544 if (error) { 545 printf("%s: ADDMULTI failed on %s\n", __func__, 546 lgp->lp_ifp->if_xname); 547 return (error); 548 } 549 550 lp = malloc(sizeof(struct lacp_port), 551 M_DEVBUF, M_NOWAIT|M_ZERO); 552 if (lp == NULL) 553 return (ENOMEM); 554 555 LACP_LOCK(lsc); 556 lgp->lp_psc = lp; 557 lp->lp_ifp = ifp; 558 lp->lp_lagg = lgp; 559 lp->lp_lsc = lsc; 560 lp->lp_ifma = rifma; 561 562 LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next); 563 564 lacp_fill_actorinfo(lp, &lp->lp_actor); 565 lacp_fill_markerinfo(lp, &lp->lp_marker); 566 lp->lp_state = LACP_STATE_ACTIVITY; 567 lp->lp_aggregator = NULL; 568 lacp_sm_rx_set_expired(lp); 569 LACP_UNLOCK(lsc); 570 lacp_linkstate(lgp); 571 572 return (0); 573 } 574 575 void 576 lacp_port_destroy(struct lagg_port *lgp) 577 { 578 struct lacp_port *lp = LACP_PORT(lgp); 579 struct lacp_softc *lsc = lp->lp_lsc; 580 int i; 581 582 LACP_LOCK(lsc); 583 for (i = 0; i < LACP_NTIMER; i++) { 584 LACP_TIMER_DISARM(lp, i); 585 } 586 587 lacp_disable_collecting(lp); 588 lacp_disable_distributing(lp); 589 lacp_unselect(lp); 590 591 LIST_REMOVE(lp, lp_next); 592 LACP_UNLOCK(lsc); 593 594 /* The address may have already been removed by if_purgemaddrs() */ 595 if (!lgp->lp_detaching) 596 if_delmulti_ifma(lp->lp_ifma); 597 598 free(lp, M_DEVBUF); 599 } 600 601 void 602 lacp_req(struct lagg_softc *sc, void *data) 603 { 604 struct lacp_opreq *req = (struct lacp_opreq *)data; 605 struct lacp_softc *lsc = LACP_SOFTC(sc); 606 struct lacp_aggregator *la; 607 608 bzero(req, sizeof(struct lacp_opreq)); 609 610 /* 611 * If the LACP softc is NULL, return with the opreq structure full of 612 * zeros. It is normal for the softc to be NULL while the lagg is 613 * being destroyed. 614 */ 615 if (NULL == lsc) 616 return; 617 618 la = lsc->lsc_active_aggregator; 619 LACP_LOCK(lsc); 620 if (la != NULL) { 621 req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio); 622 memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac, 623 ETHER_ADDR_LEN); 624 req->actor_key = ntohs(la->la_actor.lip_key); 625 req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio); 626 req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno); 627 req->actor_state = la->la_actor.lip_state; 628 629 req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio); 630 memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac, 631 ETHER_ADDR_LEN); 632 req->partner_key = ntohs(la->la_partner.lip_key); 633 req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio); 634 req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno); 635 req->partner_state = la->la_partner.lip_state; 636 } 637 LACP_UNLOCK(lsc); 638 } 639 640 void 641 lacp_portreq(struct lagg_port *lgp, void *data) 642 { 643 struct lacp_opreq *req = (struct lacp_opreq *)data; 644 struct lacp_port *lp = LACP_PORT(lgp); 645 struct lacp_softc *lsc = lp->lp_lsc; 646 647 LACP_LOCK(lsc); 648 req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio); 649 memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac, 650 ETHER_ADDR_LEN); 651 req->actor_key = ntohs(lp->lp_actor.lip_key); 652 req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio); 653 req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno); 654 req->actor_state = lp->lp_actor.lip_state; 655 656 req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio); 657 memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac, 658 ETHER_ADDR_LEN); 659 req->partner_key = ntohs(lp->lp_partner.lip_key); 660 req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio); 661 req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno); 662 req->partner_state = lp->lp_partner.lip_state; 663 LACP_UNLOCK(lsc); 664 } 665 666 static void 667 lacp_disable_collecting(struct lacp_port *lp) 668 { 669 LACP_DPRINTF((lp, "collecting disabled\n")); 670 lp->lp_state &= ~LACP_STATE_COLLECTING; 671 } 672 673 static void 674 lacp_enable_collecting(struct lacp_port *lp) 675 { 676 LACP_DPRINTF((lp, "collecting enabled\n")); 677 lp->lp_state |= LACP_STATE_COLLECTING; 678 } 679 680 static void 681 lacp_disable_distributing(struct lacp_port *lp) 682 { 683 struct lacp_aggregator *la = lp->lp_aggregator; 684 struct lacp_softc *lsc = lp->lp_lsc; 685 struct lagg_softc *sc = lsc->lsc_softc; 686 char buf[LACP_LAGIDSTR_MAX+1]; 687 688 LACP_LOCK_ASSERT(lsc); 689 690 if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) { 691 return; 692 } 693 694 KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports")); 695 KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports)); 696 KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid")); 697 698 LACP_DPRINTF((lp, "disable distributing on aggregator %s, " 699 "nports %d -> %d\n", 700 lacp_format_lagid_aggregator(la, buf, sizeof(buf)), 701 la->la_nports, la->la_nports - 1)); 702 703 TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q); 704 la->la_nports--; 705 sc->sc_active = la->la_nports; 706 707 if (lsc->lsc_active_aggregator == la) { 708 lacp_suppress_distributing(lsc, la); 709 lacp_select_active_aggregator(lsc); 710 /* regenerate the port map, the active aggregator has changed */ 711 lacp_update_portmap(lsc); 712 } 713 714 lp->lp_state &= ~LACP_STATE_DISTRIBUTING; 715 if_link_state_change(sc->sc_ifp, 716 sc->sc_active ? LINK_STATE_UP : LINK_STATE_DOWN); 717 } 718 719 static void 720 lacp_enable_distributing(struct lacp_port *lp) 721 { 722 struct lacp_aggregator *la = lp->lp_aggregator; 723 struct lacp_softc *lsc = lp->lp_lsc; 724 struct lagg_softc *sc = lsc->lsc_softc; 725 char buf[LACP_LAGIDSTR_MAX+1]; 726 727 LACP_LOCK_ASSERT(lsc); 728 729 if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) { 730 return; 731 } 732 733 LACP_DPRINTF((lp, "enable distributing on aggregator %s, " 734 "nports %d -> %d\n", 735 lacp_format_lagid_aggregator(la, buf, sizeof(buf)), 736 la->la_nports, la->la_nports + 1)); 737 738 KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid")); 739 TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q); 740 la->la_nports++; 741 sc->sc_active = la->la_nports; 742 743 lp->lp_state |= LACP_STATE_DISTRIBUTING; 744 745 if (lsc->lsc_active_aggregator == la) { 746 lacp_suppress_distributing(lsc, la); 747 lacp_update_portmap(lsc); 748 } else 749 /* try to become the active aggregator */ 750 lacp_select_active_aggregator(lsc); 751 752 if_link_state_change(sc->sc_ifp, 753 sc->sc_active ? LINK_STATE_UP : LINK_STATE_DOWN); 754 } 755 756 static void 757 lacp_transit_expire(void *vp) 758 { 759 struct lacp_softc *lsc = vp; 760 761 LACP_LOCK_ASSERT(lsc); 762 763 CURVNET_SET(lsc->lsc_softc->sc_ifp->if_vnet); 764 LACP_TRACE(NULL); 765 CURVNET_RESTORE(); 766 767 lsc->lsc_suppress_distributing = FALSE; 768 } 769 770 void 771 lacp_attach(struct lagg_softc *sc) 772 { 773 struct lacp_softc *lsc; 774 775 lsc = malloc(sizeof(struct lacp_softc), M_DEVBUF, M_WAITOK | M_ZERO); 776 777 sc->sc_psc = lsc; 778 lsc->lsc_softc = sc; 779 780 lsc->lsc_hashkey = m_ether_tcpip_hash_init(); 781 lsc->lsc_active_aggregator = NULL; 782 lsc->lsc_strict_mode = VNET(lacp_default_strict_mode); 783 LACP_LOCK_INIT(lsc); 784 TAILQ_INIT(&lsc->lsc_aggregators); 785 LIST_INIT(&lsc->lsc_ports); 786 787 callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0); 788 callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0); 789 790 /* if the lagg is already up then do the same */ 791 if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) 792 lacp_init(sc); 793 } 794 795 void 796 lacp_detach(void *psc) 797 { 798 struct lacp_softc *lsc = (struct lacp_softc *)psc; 799 800 KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators), 801 ("aggregators still active")); 802 KASSERT(lsc->lsc_active_aggregator == NULL, 803 ("aggregator still attached")); 804 805 callout_drain(&lsc->lsc_transit_callout); 806 callout_drain(&lsc->lsc_callout); 807 808 LACP_LOCK_DESTROY(lsc); 809 free(lsc, M_DEVBUF); 810 } 811 812 void 813 lacp_init(struct lagg_softc *sc) 814 { 815 struct lacp_softc *lsc = LACP_SOFTC(sc); 816 817 LACP_LOCK(lsc); 818 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc); 819 LACP_UNLOCK(lsc); 820 } 821 822 void 823 lacp_stop(struct lagg_softc *sc) 824 { 825 struct lacp_softc *lsc = LACP_SOFTC(sc); 826 827 LACP_LOCK(lsc); 828 callout_stop(&lsc->lsc_transit_callout); 829 callout_stop(&lsc->lsc_callout); 830 LACP_UNLOCK(lsc); 831 } 832 833 struct lagg_port * 834 lacp_select_tx_port_by_hash(struct lagg_softc *sc, uint32_t hash, uint8_t numa_domain) 835 { 836 struct lacp_softc *lsc = LACP_SOFTC(sc); 837 struct lacp_portmap *pm; 838 struct lacp_port *lp; 839 struct lacp_port **map; 840 int count; 841 842 if (__predict_false(lsc->lsc_suppress_distributing)) { 843 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__)); 844 return (NULL); 845 } 846 847 pm = &lsc->lsc_pmap[lsc->lsc_activemap]; 848 if (pm->pm_count == 0) { 849 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__)); 850 return (NULL); 851 } 852 853 #ifdef NUMA 854 if ((sc->sc_opts & LAGG_OPT_USE_NUMA) && 855 pm->pm_num_dom > 1 && numa_domain < MAXMEMDOM) { 856 count = pm->pm_numa[numa_domain].count; 857 if (count > 0) { 858 map = pm->pm_numa[numa_domain].map; 859 } else { 860 /* No ports on this domain; use global hash. */ 861 map = pm->pm_map; 862 count = pm->pm_count; 863 } 864 } else 865 #endif 866 { 867 map = pm->pm_map; 868 count = pm->pm_count; 869 } 870 871 hash %= count; 872 lp = map[hash]; 873 874 KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0, 875 ("aggregated port is not distributing")); 876 877 return (lp->lp_lagg); 878 } 879 880 struct lagg_port * 881 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m) 882 { 883 struct lacp_softc *lsc = LACP_SOFTC(sc); 884 uint32_t hash; 885 uint8_t numa_domain; 886 887 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && 888 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 889 hash = m->m_pkthdr.flowid >> sc->flowid_shift; 890 else 891 hash = m_ether_tcpip_hash(sc->sc_flags, m, lsc->lsc_hashkey); 892 893 numa_domain = m->m_pkthdr.numa_domain; 894 return (lacp_select_tx_port_by_hash(sc, hash, numa_domain)); 895 } 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 /* 998 * This aggregator is chosen if the partner has a better 999 * system priority 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 #ifdef NUMA 1052 int count; 1053 uint8_t domain; 1054 #endif 1055 1056 newmap = lsc->lsc_activemap == 0 ? 1 : 0; 1057 p = &lsc->lsc_pmap[newmap]; 1058 la = lsc->lsc_active_aggregator; 1059 speed = 0; 1060 bzero(p, sizeof(struct lacp_portmap)); 1061 1062 if (la != NULL && la->la_nports > 0) { 1063 p->pm_count = la->la_nports; 1064 i = 0; 1065 TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q) { 1066 p->pm_map[i++] = lp; 1067 #ifdef NUMA 1068 domain = lp->lp_ifp->if_numa_domain; 1069 if (domain >= MAXMEMDOM) 1070 continue; 1071 count = p->pm_numa[domain].count; 1072 p->pm_numa[domain].map[count] = lp; 1073 p->pm_numa[domain].count++; 1074 #endif 1075 } 1076 KASSERT(i == p->pm_count, ("Invalid port count")); 1077 1078 #ifdef NUMA 1079 for (i = 0; i < MAXMEMDOM; i++) { 1080 if (p->pm_numa[i].count != 0) 1081 p->pm_num_dom++; 1082 } 1083 #endif 1084 speed = lacp_aggregator_bandwidth(la); 1085 } 1086 sc->sc_ifp->if_baudrate = speed; 1087 1088 /* switch the active portmap over */ 1089 atomic_store_rel_int(&lsc->lsc_activemap, newmap); 1090 LACP_DPRINTF((NULL, "Set table %d with %d ports\n", 1091 lsc->lsc_activemap, 1092 lsc->lsc_pmap[lsc->lsc_activemap].pm_count)); 1093 } 1094 1095 static uint16_t 1096 lacp_compose_key(struct lacp_port *lp) 1097 { 1098 struct lagg_port *lgp = lp->lp_lagg; 1099 struct lagg_softc *sc = lgp->lp_softc; 1100 u_int media = lp->lp_media; 1101 uint16_t key; 1102 1103 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) { 1104 /* 1105 * non-aggregatable links should have unique keys. 1106 * 1107 * XXX this isn't really unique as if_index is 16 bit. 1108 */ 1109 1110 /* bit 0..14: (some bits of) if_index of this port */ 1111 key = lp->lp_ifp->if_index; 1112 /* bit 15: 1 */ 1113 key |= 0x8000; 1114 } else { 1115 u_int subtype = IFM_SUBTYPE(media); 1116 1117 KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type")); 1118 KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface")); 1119 1120 /* bit 0..4: IFM_SUBTYPE modulo speed */ 1121 switch (subtype) { 1122 case IFM_10_T: 1123 case IFM_10_2: 1124 case IFM_10_5: 1125 case IFM_10_STP: 1126 case IFM_10_FL: 1127 key = IFM_10_T; 1128 break; 1129 case IFM_100_TX: 1130 case IFM_100_FX: 1131 case IFM_100_T4: 1132 case IFM_100_VG: 1133 case IFM_100_T2: 1134 case IFM_100_T: 1135 case IFM_100_SGMII: 1136 key = IFM_100_TX; 1137 break; 1138 case IFM_1000_SX: 1139 case IFM_1000_LX: 1140 case IFM_1000_CX: 1141 case IFM_1000_T: 1142 case IFM_1000_KX: 1143 case IFM_1000_SGMII: 1144 case IFM_1000_CX_SGMII: 1145 key = IFM_1000_SX; 1146 break; 1147 case IFM_10G_LR: 1148 case IFM_10G_SR: 1149 case IFM_10G_CX4: 1150 case IFM_10G_TWINAX: 1151 case IFM_10G_TWINAX_LONG: 1152 case IFM_10G_LRM: 1153 case IFM_10G_T: 1154 case IFM_10G_KX4: 1155 case IFM_10G_KR: 1156 case IFM_10G_CR1: 1157 case IFM_10G_ER: 1158 case IFM_10G_SFI: 1159 case IFM_10G_AOC: 1160 key = IFM_10G_LR; 1161 break; 1162 case IFM_20G_KR2: 1163 key = IFM_20G_KR2; 1164 break; 1165 case IFM_2500_KX: 1166 case IFM_2500_T: 1167 case IFM_2500_X: 1168 key = IFM_2500_KX; 1169 break; 1170 case IFM_5000_T: 1171 case IFM_5000_KR: 1172 case IFM_5000_KR_S: 1173 case IFM_5000_KR1: 1174 key = IFM_5000_T; 1175 break; 1176 case IFM_50G_PCIE: 1177 case IFM_50G_CR2: 1178 case IFM_50G_KR2: 1179 case IFM_50G_KR4: 1180 case IFM_50G_SR2: 1181 case IFM_50G_LR2: 1182 case IFM_50G_LAUI2_AC: 1183 case IFM_50G_LAUI2: 1184 case IFM_50G_AUI2_AC: 1185 case IFM_50G_AUI2: 1186 case IFM_50G_CP: 1187 case IFM_50G_SR: 1188 case IFM_50G_LR: 1189 case IFM_50G_FR: 1190 case IFM_50G_KR_PAM4: 1191 case IFM_50G_AUI1_AC: 1192 case IFM_50G_AUI1: 1193 key = IFM_50G_PCIE; 1194 break; 1195 case IFM_56G_R4: 1196 key = IFM_56G_R4; 1197 break; 1198 case IFM_25G_PCIE: 1199 case IFM_25G_CR: 1200 case IFM_25G_KR: 1201 case IFM_25G_SR: 1202 case IFM_25G_LR: 1203 case IFM_25G_ACC: 1204 case IFM_25G_AOC: 1205 case IFM_25G_T: 1206 case IFM_25G_CR_S: 1207 case IFM_25G_CR1: 1208 case IFM_25G_KR_S: 1209 case IFM_25G_AUI: 1210 case IFM_25G_KR1: 1211 key = IFM_25G_PCIE; 1212 break; 1213 case IFM_40G_CR4: 1214 case IFM_40G_SR4: 1215 case IFM_40G_LR4: 1216 case IFM_40G_XLPPI: 1217 case IFM_40G_KR4: 1218 case IFM_40G_XLAUI: 1219 case IFM_40G_XLAUI_AC: 1220 case IFM_40G_ER4: 1221 key = IFM_40G_CR4; 1222 break; 1223 case IFM_100G_CR4: 1224 case IFM_100G_SR4: 1225 case IFM_100G_KR4: 1226 case IFM_100G_LR4: 1227 case IFM_100G_CAUI4_AC: 1228 case IFM_100G_CAUI4: 1229 case IFM_100G_AUI4_AC: 1230 case IFM_100G_AUI4: 1231 case IFM_100G_CR_PAM4: 1232 case IFM_100G_KR_PAM4: 1233 case IFM_100G_CP2: 1234 case IFM_100G_SR2: 1235 case IFM_100G_DR: 1236 case IFM_100G_KR2_PAM4: 1237 case IFM_100G_CAUI2_AC: 1238 case IFM_100G_CAUI2: 1239 case IFM_100G_AUI2_AC: 1240 case IFM_100G_AUI2: 1241 key = IFM_100G_CR4; 1242 break; 1243 case IFM_200G_CR4_PAM4: 1244 case IFM_200G_SR4: 1245 case IFM_200G_FR4: 1246 case IFM_200G_LR4: 1247 case IFM_200G_DR4: 1248 case IFM_200G_KR4_PAM4: 1249 case IFM_200G_AUI4_AC: 1250 case IFM_200G_AUI4: 1251 case IFM_200G_AUI8_AC: 1252 case IFM_200G_AUI8: 1253 key = IFM_200G_CR4_PAM4; 1254 break; 1255 case IFM_400G_FR8: 1256 case IFM_400G_LR8: 1257 case IFM_400G_DR4: 1258 case IFM_400G_AUI8_AC: 1259 case IFM_400G_AUI8: 1260 key = IFM_400G_FR8; 1261 break; 1262 default: 1263 key = subtype; 1264 break; 1265 } 1266 /* bit 5..14: (some bits of) if_index of lagg device */ 1267 key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5); 1268 /* bit 15: 0 */ 1269 } 1270 return (htons(key)); 1271 } 1272 1273 static void 1274 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la) 1275 { 1276 char buf[LACP_LAGIDSTR_MAX+1]; 1277 1278 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n", 1279 __func__, 1280 lacp_format_lagid(&la->la_actor, &la->la_partner, 1281 buf, sizeof(buf)), 1282 la->la_refcnt, la->la_refcnt + 1)); 1283 1284 KASSERT(la->la_refcnt > 0, ("refcount <= 0")); 1285 la->la_refcnt++; 1286 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount")); 1287 } 1288 1289 static void 1290 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la) 1291 { 1292 char buf[LACP_LAGIDSTR_MAX+1]; 1293 1294 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n", 1295 __func__, 1296 lacp_format_lagid(&la->la_actor, &la->la_partner, 1297 buf, sizeof(buf)), 1298 la->la_refcnt, la->la_refcnt - 1)); 1299 1300 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt")); 1301 la->la_refcnt--; 1302 if (la->la_refcnt > 0) { 1303 return; 1304 } 1305 1306 KASSERT(la->la_refcnt == 0, ("refcount not zero")); 1307 KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active")); 1308 1309 TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q); 1310 1311 free(la, M_DEVBUF); 1312 } 1313 1314 /* 1315 * lacp_aggregator_get: allocate an aggregator. 1316 */ 1317 1318 static struct lacp_aggregator * 1319 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp) 1320 { 1321 struct lacp_aggregator *la; 1322 1323 la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT); 1324 if (la) { 1325 la->la_refcnt = 1; 1326 la->la_nports = 0; 1327 TAILQ_INIT(&la->la_ports); 1328 la->la_pending = 0; 1329 TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q); 1330 } 1331 1332 return (la); 1333 } 1334 1335 /* 1336 * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port. 1337 */ 1338 1339 static void 1340 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp) 1341 { 1342 lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner); 1343 lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor); 1344 1345 la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION; 1346 } 1347 1348 static void 1349 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr, 1350 const struct lacp_peerinfo *lpi_port) 1351 { 1352 memset(lpi_aggr, 0, sizeof(*lpi_aggr)); 1353 lpi_aggr->lip_systemid = lpi_port->lip_systemid; 1354 lpi_aggr->lip_key = lpi_port->lip_key; 1355 } 1356 1357 /* 1358 * lacp_aggregator_is_compatible: check if a port can join to an aggregator. 1359 */ 1360 1361 static int 1362 lacp_aggregator_is_compatible(const struct lacp_aggregator *la, 1363 const struct lacp_port *lp) 1364 { 1365 if (!(lp->lp_state & LACP_STATE_AGGREGATION) || 1366 !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) { 1367 return (0); 1368 } 1369 1370 if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) { 1371 return (0); 1372 } 1373 1374 if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) { 1375 return (0); 1376 } 1377 1378 if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) { 1379 return (0); 1380 } 1381 1382 return (1); 1383 } 1384 1385 static int 1386 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a, 1387 const struct lacp_peerinfo *b) 1388 { 1389 if (memcmp(&a->lip_systemid, &b->lip_systemid, 1390 sizeof(a->lip_systemid))) { 1391 return (0); 1392 } 1393 1394 if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) { 1395 return (0); 1396 } 1397 1398 return (1); 1399 } 1400 1401 static void 1402 lacp_port_enable(struct lacp_port *lp) 1403 { 1404 lp->lp_state |= LACP_STATE_AGGREGATION; 1405 } 1406 1407 static void 1408 lacp_port_disable(struct lacp_port *lp) 1409 { 1410 lacp_set_mux(lp, LACP_MUX_DETACHED); 1411 1412 lp->lp_state &= ~LACP_STATE_AGGREGATION; 1413 lp->lp_selected = LACP_UNSELECTED; 1414 lacp_sm_rx_record_default(lp); 1415 lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION; 1416 lp->lp_state &= ~LACP_STATE_EXPIRED; 1417 } 1418 1419 /* 1420 * lacp_select: select an aggregator. create one if necessary. 1421 */ 1422 static void 1423 lacp_select(struct lacp_port *lp) 1424 { 1425 struct lacp_softc *lsc = lp->lp_lsc; 1426 struct lacp_aggregator *la; 1427 char buf[LACP_LAGIDSTR_MAX+1]; 1428 1429 if (lp->lp_aggregator) { 1430 return; 1431 } 1432 1433 /* If we haven't heard from our peer, skip this step. */ 1434 if (lp->lp_state & LACP_STATE_DEFAULTED) 1435 return; 1436 1437 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), 1438 ("timer_wait_while still active")); 1439 1440 LACP_DPRINTF((lp, "port lagid=%s\n", 1441 lacp_format_lagid(&lp->lp_actor, &lp->lp_partner, 1442 buf, sizeof(buf)))); 1443 1444 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) { 1445 if (lacp_aggregator_is_compatible(la, lp)) { 1446 break; 1447 } 1448 } 1449 1450 if (la == NULL) { 1451 la = lacp_aggregator_get(lsc, lp); 1452 if (la == NULL) { 1453 LACP_DPRINTF((lp, "aggregator creation failed\n")); 1454 1455 /* 1456 * will retry on the next tick. 1457 */ 1458 1459 return; 1460 } 1461 lacp_fill_aggregator_id(la, lp); 1462 LACP_DPRINTF((lp, "aggregator created\n")); 1463 } else { 1464 LACP_DPRINTF((lp, "compatible aggregator found\n")); 1465 if (la->la_refcnt == LACP_MAX_PORTS) 1466 return; 1467 lacp_aggregator_addref(lsc, la); 1468 } 1469 1470 LACP_DPRINTF((lp, "aggregator lagid=%s\n", 1471 lacp_format_lagid(&la->la_actor, &la->la_partner, 1472 buf, sizeof(buf)))); 1473 1474 lp->lp_aggregator = la; 1475 lp->lp_selected = LACP_SELECTED; 1476 } 1477 1478 /* 1479 * lacp_unselect: finish unselect/detach process. 1480 */ 1481 1482 static void 1483 lacp_unselect(struct lacp_port *lp) 1484 { 1485 struct lacp_softc *lsc = lp->lp_lsc; 1486 struct lacp_aggregator *la = lp->lp_aggregator; 1487 1488 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), 1489 ("timer_wait_while still active")); 1490 1491 if (la == NULL) { 1492 return; 1493 } 1494 1495 lp->lp_aggregator = NULL; 1496 lacp_aggregator_delref(lsc, la); 1497 } 1498 1499 /* mux machine */ 1500 1501 static void 1502 lacp_sm_mux(struct lacp_port *lp) 1503 { 1504 struct lagg_port *lgp = lp->lp_lagg; 1505 struct lagg_softc *sc = lgp->lp_softc; 1506 enum lacp_mux_state new_state; 1507 boolean_t p_sync = 1508 (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0; 1509 boolean_t p_collecting = 1510 (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0; 1511 enum lacp_selected selected = lp->lp_selected; 1512 struct lacp_aggregator *la; 1513 1514 if (V_lacp_debug > 1) 1515 lacp_dprintf(lp, "%s: state= 0x%x, selected= 0x%x, " 1516 "p_sync= 0x%x, p_collecting= 0x%x\n", __func__, 1517 lp->lp_mux_state, selected, p_sync, p_collecting); 1518 1519 re_eval: 1520 la = lp->lp_aggregator; 1521 KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL, 1522 ("MUX not detached")); 1523 new_state = lp->lp_mux_state; 1524 switch (lp->lp_mux_state) { 1525 case LACP_MUX_DETACHED: 1526 if (selected != LACP_UNSELECTED) { 1527 new_state = LACP_MUX_WAITING; 1528 } 1529 break; 1530 case LACP_MUX_WAITING: 1531 KASSERT(la->la_pending > 0 || 1532 !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), 1533 ("timer_wait_while still active")); 1534 if (selected == LACP_SELECTED && la->la_pending == 0) { 1535 new_state = LACP_MUX_ATTACHED; 1536 } else if (selected == LACP_UNSELECTED) { 1537 new_state = LACP_MUX_DETACHED; 1538 } 1539 break; 1540 case LACP_MUX_ATTACHED: 1541 if (selected == LACP_SELECTED && p_sync) { 1542 new_state = LACP_MUX_COLLECTING; 1543 } else if (selected != LACP_SELECTED) { 1544 new_state = LACP_MUX_DETACHED; 1545 } 1546 break; 1547 case LACP_MUX_COLLECTING: 1548 if (selected == LACP_SELECTED && p_sync && p_collecting) { 1549 new_state = LACP_MUX_DISTRIBUTING; 1550 } else if (selected != LACP_SELECTED || !p_sync) { 1551 new_state = LACP_MUX_ATTACHED; 1552 } 1553 break; 1554 case LACP_MUX_DISTRIBUTING: 1555 if (selected != LACP_SELECTED || !p_sync || !p_collecting) { 1556 new_state = LACP_MUX_COLLECTING; 1557 lacp_dprintf(lp, "Interface stopped DISTRIBUTING, possible flapping\n"); 1558 sc->sc_flapping++; 1559 } 1560 break; 1561 default: 1562 panic("%s: unknown state", __func__); 1563 } 1564 1565 if (lp->lp_mux_state == new_state) { 1566 return; 1567 } 1568 1569 lacp_set_mux(lp, new_state); 1570 goto re_eval; 1571 } 1572 1573 static void 1574 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state) 1575 { 1576 struct lacp_aggregator *la = lp->lp_aggregator; 1577 1578 if (lp->lp_mux_state == new_state) { 1579 return; 1580 } 1581 1582 switch (new_state) { 1583 case LACP_MUX_DETACHED: 1584 lp->lp_state &= ~LACP_STATE_SYNC; 1585 lacp_disable_distributing(lp); 1586 lacp_disable_collecting(lp); 1587 lacp_sm_assert_ntt(lp); 1588 /* cancel timer */ 1589 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) { 1590 KASSERT(la->la_pending > 0, 1591 ("timer_wait_while not active")); 1592 la->la_pending--; 1593 } 1594 LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE); 1595 lacp_unselect(lp); 1596 break; 1597 case LACP_MUX_WAITING: 1598 LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE, 1599 LACP_AGGREGATE_WAIT_TIME); 1600 la->la_pending++; 1601 break; 1602 case LACP_MUX_ATTACHED: 1603 lp->lp_state |= LACP_STATE_SYNC; 1604 lacp_disable_collecting(lp); 1605 lacp_sm_assert_ntt(lp); 1606 break; 1607 case LACP_MUX_COLLECTING: 1608 lacp_enable_collecting(lp); 1609 lacp_disable_distributing(lp); 1610 lacp_sm_assert_ntt(lp); 1611 break; 1612 case LACP_MUX_DISTRIBUTING: 1613 lacp_enable_distributing(lp); 1614 break; 1615 default: 1616 panic("%s: unknown state", __func__); 1617 } 1618 1619 LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state)); 1620 1621 lp->lp_mux_state = new_state; 1622 } 1623 1624 static void 1625 lacp_sm_mux_timer(struct lacp_port *lp) 1626 { 1627 struct lacp_aggregator *la = lp->lp_aggregator; 1628 char buf[LACP_LAGIDSTR_MAX+1]; 1629 1630 KASSERT(la->la_pending > 0, ("no pending event")); 1631 1632 LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__, 1633 lacp_format_lagid(&la->la_actor, &la->la_partner, 1634 buf, sizeof(buf)), 1635 la->la_pending, la->la_pending - 1)); 1636 1637 la->la_pending--; 1638 } 1639 1640 /* periodic transmit machine */ 1641 1642 static void 1643 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate) 1644 { 1645 if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state, 1646 LACP_STATE_TIMEOUT)) { 1647 return; 1648 } 1649 1650 LACP_DPRINTF((lp, "partner timeout changed\n")); 1651 1652 /* 1653 * FAST_PERIODIC -> SLOW_PERIODIC 1654 * or 1655 * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC 1656 * 1657 * let lacp_sm_ptx_tx_schedule to update timeout. 1658 */ 1659 1660 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC); 1661 1662 /* 1663 * if timeout has been shortened, assert NTT. 1664 */ 1665 1666 if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) { 1667 lacp_sm_assert_ntt(lp); 1668 } 1669 } 1670 1671 static void 1672 lacp_sm_ptx_tx_schedule(struct lacp_port *lp) 1673 { 1674 int timeout; 1675 1676 if (!(lp->lp_state & LACP_STATE_ACTIVITY) && 1677 !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) { 1678 /* 1679 * NO_PERIODIC 1680 */ 1681 1682 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC); 1683 return; 1684 } 1685 1686 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) { 1687 return; 1688 } 1689 1690 timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ? 1691 LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME; 1692 1693 LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout); 1694 } 1695 1696 static void 1697 lacp_sm_ptx_timer(struct lacp_port *lp) 1698 { 1699 lacp_sm_assert_ntt(lp); 1700 } 1701 1702 static void 1703 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du) 1704 { 1705 int timeout; 1706 1707 /* 1708 * check LACP_DISABLED first 1709 */ 1710 1711 if (!(lp->lp_state & LACP_STATE_AGGREGATION)) { 1712 return; 1713 } 1714 1715 /* 1716 * check loopback condition. 1717 */ 1718 1719 if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid, 1720 &lp->lp_actor.lip_systemid)) { 1721 return; 1722 } 1723 1724 /* 1725 * EXPIRED, DEFAULTED, CURRENT -> CURRENT 1726 */ 1727 1728 lacp_sm_rx_update_selected(lp, du); 1729 lacp_sm_rx_update_ntt(lp, du); 1730 lacp_sm_rx_record_pdu(lp, du); 1731 1732 timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ? 1733 LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME; 1734 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout); 1735 1736 lp->lp_state &= ~LACP_STATE_EXPIRED; 1737 1738 /* 1739 * kick transmit machine without waiting the next tick. 1740 */ 1741 1742 lacp_sm_tx(lp); 1743 } 1744 1745 static void 1746 lacp_sm_rx_set_expired(struct lacp_port *lp) 1747 { 1748 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC; 1749 lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT; 1750 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME); 1751 lp->lp_state |= LACP_STATE_EXPIRED; 1752 } 1753 1754 static void 1755 lacp_sm_rx_timer(struct lacp_port *lp) 1756 { 1757 if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) { 1758 /* CURRENT -> EXPIRED */ 1759 LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__)); 1760 lacp_sm_rx_set_expired(lp); 1761 } else { 1762 /* EXPIRED -> DEFAULTED */ 1763 LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__)); 1764 lacp_sm_rx_update_default_selected(lp); 1765 lacp_sm_rx_record_default(lp); 1766 lp->lp_state &= ~LACP_STATE_EXPIRED; 1767 } 1768 } 1769 1770 static void 1771 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du) 1772 { 1773 boolean_t active; 1774 uint8_t oldpstate; 1775 char buf[LACP_STATESTR_MAX+1]; 1776 1777 LACP_TRACE(lp); 1778 1779 oldpstate = lp->lp_partner.lip_state; 1780 1781 active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY) 1782 || ((lp->lp_state & LACP_STATE_ACTIVITY) && 1783 (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY)); 1784 1785 lp->lp_partner = du->ldu_actor; 1786 if (active && 1787 ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state, 1788 LACP_STATE_AGGREGATION) && 1789 !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner)) 1790 || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) { 1791 /* 1792 * XXX Maintain legacy behavior of leaving the 1793 * LACP_STATE_SYNC bit unchanged from the partner's 1794 * advertisement if lsc_strict_mode is false. 1795 * TODO: We should re-examine the concept of the "strict mode" 1796 * to ensure it makes sense to maintain a non-strict mode. 1797 */ 1798 if (lp->lp_lsc->lsc_strict_mode) 1799 lp->lp_partner.lip_state |= LACP_STATE_SYNC; 1800 } else { 1801 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC; 1802 } 1803 1804 lp->lp_state &= ~LACP_STATE_DEFAULTED; 1805 1806 if (oldpstate != lp->lp_partner.lip_state) { 1807 LACP_DPRINTF((lp, "old pstate %s\n", 1808 lacp_format_state(oldpstate, buf, sizeof(buf)))); 1809 LACP_DPRINTF((lp, "new pstate %s\n", 1810 lacp_format_state(lp->lp_partner.lip_state, buf, 1811 sizeof(buf)))); 1812 } 1813 1814 lacp_sm_ptx_update_timeout(lp, oldpstate); 1815 } 1816 1817 static void 1818 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du) 1819 { 1820 1821 LACP_TRACE(lp); 1822 1823 if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) || 1824 !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state, 1825 LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) { 1826 LACP_DPRINTF((lp, "%s: assert ntt\n", __func__)); 1827 lacp_sm_assert_ntt(lp); 1828 } 1829 } 1830 1831 static void 1832 lacp_sm_rx_record_default(struct lacp_port *lp) 1833 { 1834 uint8_t oldpstate; 1835 1836 LACP_TRACE(lp); 1837 1838 oldpstate = lp->lp_partner.lip_state; 1839 if (lp->lp_lsc->lsc_strict_mode) 1840 lp->lp_partner = lacp_partner_admin_strict; 1841 else 1842 lp->lp_partner = lacp_partner_admin_optimistic; 1843 lp->lp_state |= LACP_STATE_DEFAULTED; 1844 lacp_sm_ptx_update_timeout(lp, oldpstate); 1845 } 1846 1847 static void 1848 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp, 1849 const struct lacp_peerinfo *info) 1850 { 1851 1852 LACP_TRACE(lp); 1853 1854 if (lacp_compare_peerinfo(&lp->lp_partner, info) || 1855 !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state, 1856 LACP_STATE_AGGREGATION)) { 1857 lp->lp_selected = LACP_UNSELECTED; 1858 /* mux machine will clean up lp->lp_aggregator */ 1859 } 1860 } 1861 1862 static void 1863 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du) 1864 { 1865 1866 LACP_TRACE(lp); 1867 1868 lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor); 1869 } 1870 1871 static void 1872 lacp_sm_rx_update_default_selected(struct lacp_port *lp) 1873 { 1874 1875 LACP_TRACE(lp); 1876 1877 if (lp->lp_lsc->lsc_strict_mode) 1878 lacp_sm_rx_update_selected_from_peerinfo(lp, 1879 &lacp_partner_admin_strict); 1880 else 1881 lacp_sm_rx_update_selected_from_peerinfo(lp, 1882 &lacp_partner_admin_optimistic); 1883 } 1884 1885 /* transmit machine */ 1886 1887 static void 1888 lacp_sm_tx(struct lacp_port *lp) 1889 { 1890 int error = 0; 1891 1892 if (!(lp->lp_state & LACP_STATE_AGGREGATION) 1893 #if 1 1894 || (!(lp->lp_state & LACP_STATE_ACTIVITY) 1895 && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) 1896 #endif 1897 ) { 1898 lp->lp_flags &= ~LACP_PORT_NTT; 1899 } 1900 1901 if (!(lp->lp_flags & LACP_PORT_NTT)) { 1902 return; 1903 } 1904 1905 /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */ 1906 if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent, 1907 (3 / LACP_FAST_PERIODIC_TIME)) == 0) { 1908 LACP_DPRINTF((lp, "rate limited pdu\n")); 1909 return; 1910 } 1911 1912 if (((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_tx_test) == 0) { 1913 error = lacp_xmit_lacpdu(lp); 1914 } else { 1915 LACP_TPRINTF((lp, "Dropping TX PDU\n")); 1916 } 1917 1918 if (error == 0) { 1919 lp->lp_flags &= ~LACP_PORT_NTT; 1920 } else { 1921 LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n", 1922 error)); 1923 } 1924 } 1925 1926 static void 1927 lacp_sm_assert_ntt(struct lacp_port *lp) 1928 { 1929 1930 lp->lp_flags |= LACP_PORT_NTT; 1931 } 1932 1933 static void 1934 lacp_run_timers(struct lacp_port *lp) 1935 { 1936 int i; 1937 1938 for (i = 0; i < LACP_NTIMER; i++) { 1939 KASSERT(lp->lp_timer[i] >= 0, 1940 ("invalid timer value %d", lp->lp_timer[i])); 1941 if (lp->lp_timer[i] == 0) { 1942 continue; 1943 } else if (--lp->lp_timer[i] <= 0) { 1944 if (lacp_timer_funcs[i]) { 1945 (*lacp_timer_funcs[i])(lp); 1946 } 1947 } 1948 } 1949 } 1950 1951 int 1952 lacp_marker_input(struct lacp_port *lp, struct mbuf *m) 1953 { 1954 struct lacp_softc *lsc = lp->lp_lsc; 1955 struct lagg_port *lgp = lp->lp_lagg; 1956 struct lacp_port *lp2; 1957 struct markerdu *mdu; 1958 int error = 0; 1959 int pending = 0; 1960 1961 if (m->m_pkthdr.len != sizeof(*mdu)) { 1962 goto bad; 1963 } 1964 1965 if ((m->m_flags & M_MCAST) == 0) { 1966 goto bad; 1967 } 1968 1969 if (m->m_len < sizeof(*mdu)) { 1970 m = m_pullup(m, sizeof(*mdu)); 1971 if (m == NULL) { 1972 return (ENOMEM); 1973 } 1974 } 1975 1976 mdu = mtod(m, struct markerdu *); 1977 1978 if (memcmp(&mdu->mdu_eh.ether_dhost, 1979 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) { 1980 goto bad; 1981 } 1982 1983 if (mdu->mdu_sph.sph_version != 1) { 1984 goto bad; 1985 } 1986 1987 switch (mdu->mdu_tlv.tlv_type) { 1988 case MARKER_TYPE_INFO: 1989 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv, 1990 marker_info_tlv_template, TRUE)) { 1991 goto bad; 1992 } 1993 mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE; 1994 memcpy(&mdu->mdu_eh.ether_dhost, 1995 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN); 1996 memcpy(&mdu->mdu_eh.ether_shost, 1997 lgp->lp_lladdr, ETHER_ADDR_LEN); 1998 error = lagg_enqueue(lp->lp_ifp, m); 1999 break; 2000 2001 case MARKER_TYPE_RESPONSE: 2002 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv, 2003 marker_response_tlv_template, TRUE)) { 2004 goto bad; 2005 } 2006 LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n", 2007 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, 2008 ":", ntohl(mdu->mdu_info.mi_rq_xid))); 2009 2010 /* Verify that it is the last marker we sent out */ 2011 if (memcmp(&mdu->mdu_info, &lp->lp_marker, 2012 sizeof(struct lacp_markerinfo))) 2013 goto bad; 2014 2015 LACP_LOCK(lsc); 2016 lp->lp_flags &= ~LACP_PORT_MARK; 2017 2018 if (lsc->lsc_suppress_distributing) { 2019 /* Check if any ports are waiting for a response */ 2020 LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) { 2021 if (lp2->lp_flags & LACP_PORT_MARK) { 2022 pending = 1; 2023 break; 2024 } 2025 } 2026 2027 if (pending == 0) { 2028 /* All interface queues are clear */ 2029 LACP_DPRINTF((NULL, "queue flush complete\n")); 2030 lsc->lsc_suppress_distributing = FALSE; 2031 } 2032 } 2033 LACP_UNLOCK(lsc); 2034 m_freem(m); 2035 break; 2036 2037 default: 2038 goto bad; 2039 } 2040 2041 return (error); 2042 2043 bad: 2044 LACP_DPRINTF((lp, "bad marker frame\n")); 2045 m_freem(m); 2046 return (EINVAL); 2047 } 2048 2049 static int 2050 tlv_check(const void *p, size_t size, const struct tlvhdr *tlv, 2051 const struct tlv_template *tmpl, boolean_t check_type) 2052 { 2053 while (/* CONSTCOND */ 1) { 2054 if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) { 2055 return (EINVAL); 2056 } 2057 if ((check_type && tlv->tlv_type != tmpl->tmpl_type) || 2058 tlv->tlv_length != tmpl->tmpl_length) { 2059 return (EINVAL); 2060 } 2061 if (tmpl->tmpl_type == 0) { 2062 break; 2063 } 2064 tlv = (const struct tlvhdr *) 2065 ((const char *)tlv + tlv->tlv_length); 2066 tmpl++; 2067 } 2068 2069 return (0); 2070 } 2071 2072 /* Debugging */ 2073 const char * 2074 lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen) 2075 { 2076 snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X", 2077 (int)mac[0], 2078 (int)mac[1], 2079 (int)mac[2], 2080 (int)mac[3], 2081 (int)mac[4], 2082 (int)mac[5]); 2083 2084 return (buf); 2085 } 2086 2087 const char * 2088 lacp_format_systemid(const struct lacp_systemid *sysid, 2089 char *buf, size_t buflen) 2090 { 2091 char macbuf[LACP_MACSTR_MAX+1]; 2092 2093 snprintf(buf, buflen, "%04X,%s", 2094 ntohs(sysid->lsi_prio), 2095 lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf))); 2096 2097 return (buf); 2098 } 2099 2100 const char * 2101 lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen) 2102 { 2103 snprintf(buf, buflen, "%04X,%04X", 2104 ntohs(portid->lpi_prio), 2105 ntohs(portid->lpi_portno)); 2106 2107 return (buf); 2108 } 2109 2110 const char * 2111 lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen) 2112 { 2113 char sysid[LACP_SYSTEMIDSTR_MAX+1]; 2114 char portid[LACP_PORTIDSTR_MAX+1]; 2115 2116 snprintf(buf, buflen, "(%s,%04X,%s)", 2117 lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)), 2118 ntohs(peer->lip_key), 2119 lacp_format_portid(&peer->lip_portid, portid, sizeof(portid))); 2120 2121 return (buf); 2122 } 2123 2124 const char * 2125 lacp_format_lagid(const struct lacp_peerinfo *a, 2126 const struct lacp_peerinfo *b, char *buf, size_t buflen) 2127 { 2128 char astr[LACP_PARTNERSTR_MAX+1]; 2129 char bstr[LACP_PARTNERSTR_MAX+1]; 2130 2131 #if 0 2132 /* 2133 * there's a convention to display small numbered peer 2134 * in the left. 2135 */ 2136 2137 if (lacp_compare_peerinfo(a, b) > 0) { 2138 const struct lacp_peerinfo *t; 2139 2140 t = a; 2141 a = b; 2142 b = t; 2143 } 2144 #endif 2145 2146 snprintf(buf, buflen, "[%s,%s]", 2147 lacp_format_partner(a, astr, sizeof(astr)), 2148 lacp_format_partner(b, bstr, sizeof(bstr))); 2149 2150 return (buf); 2151 } 2152 2153 const char * 2154 lacp_format_lagid_aggregator(const struct lacp_aggregator *la, 2155 char *buf, size_t buflen) 2156 { 2157 if (la == NULL) { 2158 return ("(none)"); 2159 } 2160 2161 return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen)); 2162 } 2163 2164 const char * 2165 lacp_format_state(uint8_t state, char *buf, size_t buflen) 2166 { 2167 snprintf(buf, buflen, "%b", state, LACP_STATE_BITS); 2168 return (buf); 2169 } 2170 2171 static void 2172 lacp_dump_lacpdu(const struct lacpdu *du) 2173 { 2174 char buf[LACP_PARTNERSTR_MAX+1]; 2175 char buf2[LACP_STATESTR_MAX+1]; 2176 2177 printf("actor=%s\n", 2178 lacp_format_partner(&du->ldu_actor, buf, sizeof(buf))); 2179 printf("actor.state=%s\n", 2180 lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2))); 2181 printf("partner=%s\n", 2182 lacp_format_partner(&du->ldu_partner, buf, sizeof(buf))); 2183 printf("partner.state=%s\n", 2184 lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2))); 2185 2186 printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay)); 2187 } 2188 2189 static void 2190 lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...) 2191 { 2192 va_list va; 2193 2194 if (lp) { 2195 printf("%s: ", lp->lp_ifp->if_xname); 2196 } 2197 2198 va_start(va, fmt); 2199 vprintf(fmt, va); 2200 va_end(va); 2201 } 2202