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"); 209 210 #define LACP_DPRINTF(a) if (V_lacp_debug & 0x01) { lacp_dprintf a ; } 211 #define LACP_TRACE(a) if (V_lacp_debug & 0x02) { lacp_dprintf(a,"%s\n",__func__); } 212 #define LACP_TPRINTF(a) if (V_lacp_debug & 0x04) { lacp_dprintf a ; } 213 214 /* 215 * partner administration variables. 216 * XXX should be configurable. 217 */ 218 219 static const struct lacp_peerinfo lacp_partner_admin_optimistic = { 220 .lip_systemid = { .lsi_prio = 0xffff }, 221 .lip_portid = { .lpi_prio = 0xffff }, 222 .lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION | 223 LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING, 224 }; 225 226 static const struct lacp_peerinfo lacp_partner_admin_strict = { 227 .lip_systemid = { .lsi_prio = 0xffff }, 228 .lip_portid = { .lpi_prio = 0xffff }, 229 .lip_state = 0, 230 }; 231 232 static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = { 233 [LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer, 234 [LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer, 235 [LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer, 236 }; 237 238 struct mbuf * 239 lacp_input(struct lagg_port *lgp, struct mbuf *m) 240 { 241 struct lacp_port *lp = LACP_PORT(lgp); 242 uint8_t subtype; 243 244 if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) { 245 m_freem(m); 246 return (NULL); 247 } 248 249 m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype); 250 switch (subtype) { 251 case SLOWPROTOCOLS_SUBTYPE_LACP: 252 lacp_pdu_input(lp, m); 253 return (NULL); 254 255 case SLOWPROTOCOLS_SUBTYPE_MARKER: 256 lacp_marker_input(lp, m); 257 return (NULL); 258 } 259 260 /* Not a subtype we are interested in */ 261 return (m); 262 } 263 264 /* 265 * lacp_pdu_input: process lacpdu 266 */ 267 static int 268 lacp_pdu_input(struct lacp_port *lp, struct mbuf *m) 269 { 270 struct lacp_softc *lsc = lp->lp_lsc; 271 struct lacpdu *du; 272 int error = 0; 273 274 if (m->m_pkthdr.len != sizeof(*du)) { 275 goto bad; 276 } 277 278 if ((m->m_flags & M_MCAST) == 0) { 279 goto bad; 280 } 281 282 if (m->m_len < sizeof(*du)) { 283 m = m_pullup(m, sizeof(*du)); 284 if (m == NULL) { 285 return (ENOMEM); 286 } 287 } 288 289 du = mtod(m, struct lacpdu *); 290 291 if (memcmp(&du->ldu_eh.ether_dhost, 292 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) { 293 goto bad; 294 } 295 296 /* 297 * ignore the version for compatibility with 298 * the future protocol revisions. 299 */ 300 #if 0 301 if (du->ldu_sph.sph_version != 1) { 302 goto bad; 303 } 304 #endif 305 306 /* 307 * ignore tlv types for compatibility with 308 * the future protocol revisions. 309 */ 310 if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor, 311 lacp_info_tlv_template, FALSE)) { 312 goto bad; 313 } 314 315 if (V_lacp_debug > 0) { 316 lacp_dprintf(lp, "lacpdu receive\n"); 317 lacp_dump_lacpdu(du); 318 } 319 320 if ((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_rx_test) { 321 LACP_TPRINTF((lp, "Dropping RX PDU\n")); 322 goto bad; 323 } 324 325 LACP_LOCK(lsc); 326 lacp_sm_rx(lp, du); 327 LACP_UNLOCK(lsc); 328 329 m_freem(m); 330 return (error); 331 332 bad: 333 m_freem(m); 334 return (EINVAL); 335 } 336 337 static void 338 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info) 339 { 340 struct lagg_port *lgp = lp->lp_lagg; 341 struct lagg_softc *sc = lgp->lp_softc; 342 343 info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO); 344 memcpy(&info->lip_systemid.lsi_mac, 345 IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); 346 info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO); 347 info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index); 348 info->lip_state = lp->lp_state; 349 } 350 351 static void 352 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info) 353 { 354 struct ifnet *ifp = lp->lp_ifp; 355 356 /* Fill in the port index and system id (encoded as the MAC) */ 357 info->mi_rq_port = htons(ifp->if_index); 358 memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN); 359 info->mi_rq_xid = htonl(0); 360 } 361 362 static int 363 lacp_xmit_lacpdu(struct lacp_port *lp) 364 { 365 struct lagg_port *lgp = lp->lp_lagg; 366 struct mbuf *m; 367 struct lacpdu *du; 368 int error; 369 370 LACP_LOCK_ASSERT(lp->lp_lsc); 371 372 m = m_gethdr(M_NOWAIT, MT_DATA); 373 if (m == NULL) { 374 return (ENOMEM); 375 } 376 m->m_len = m->m_pkthdr.len = sizeof(*du); 377 378 du = mtod(m, struct lacpdu *); 379 memset(du, 0, sizeof(*du)); 380 381 memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols, 382 ETHER_ADDR_LEN); 383 memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN); 384 du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW); 385 386 du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP; 387 du->ldu_sph.sph_version = 1; 388 389 TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor)); 390 du->ldu_actor = lp->lp_actor; 391 392 TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO, 393 sizeof(du->ldu_partner)); 394 du->ldu_partner = lp->lp_partner; 395 396 TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO, 397 sizeof(du->ldu_collector)); 398 du->ldu_collector.lci_maxdelay = 0; 399 400 if (V_lacp_debug > 0) { 401 lacp_dprintf(lp, "lacpdu transmit\n"); 402 lacp_dump_lacpdu(du); 403 } 404 405 m->m_flags |= M_MCAST; 406 407 /* 408 * XXX should use higher priority queue. 409 * otherwise network congestion can break aggregation. 410 */ 411 412 error = lagg_enqueue(lp->lp_ifp, m); 413 return (error); 414 } 415 416 static int 417 lacp_xmit_marker(struct lacp_port *lp) 418 { 419 struct lagg_port *lgp = lp->lp_lagg; 420 struct mbuf *m; 421 struct markerdu *mdu; 422 int error; 423 424 LACP_LOCK_ASSERT(lp->lp_lsc); 425 426 m = m_gethdr(M_NOWAIT, MT_DATA); 427 if (m == NULL) { 428 return (ENOMEM); 429 } 430 m->m_len = m->m_pkthdr.len = sizeof(*mdu); 431 432 mdu = mtod(m, struct markerdu *); 433 memset(mdu, 0, sizeof(*mdu)); 434 435 memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols, 436 ETHER_ADDR_LEN); 437 memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN); 438 mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW); 439 440 mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER; 441 mdu->mdu_sph.sph_version = 1; 442 443 /* Bump the transaction id and copy over the marker info */ 444 lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1); 445 TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info)); 446 mdu->mdu_info = lp->lp_marker; 447 448 LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n", 449 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":", 450 ntohl(mdu->mdu_info.mi_rq_xid))); 451 452 m->m_flags |= M_MCAST; 453 error = lagg_enqueue(lp->lp_ifp, m); 454 return (error); 455 } 456 457 void 458 lacp_linkstate(struct lagg_port *lgp) 459 { 460 struct lacp_port *lp = LACP_PORT(lgp); 461 struct lacp_softc *lsc = lp->lp_lsc; 462 struct ifnet *ifp = lgp->lp_ifp; 463 struct ifmediareq ifmr; 464 int error = 0; 465 u_int media; 466 uint8_t old_state; 467 uint16_t old_key; 468 469 bzero((char *)&ifmr, sizeof(ifmr)); 470 error = (*ifp->if_ioctl)(ifp, SIOCGIFXMEDIA, (caddr_t)&ifmr); 471 if (error != 0) { 472 bzero((char *)&ifmr, sizeof(ifmr)); 473 error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr); 474 } 475 if (error != 0) 476 return; 477 478 LACP_LOCK(lsc); 479 media = ifmr.ifm_active; 480 LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, " 481 "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER, 482 (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP)); 483 old_state = lp->lp_state; 484 old_key = lp->lp_key; 485 486 lp->lp_media = media; 487 /* 488 * If the port is not an active full duplex Ethernet link then it can 489 * not be aggregated. 490 */ 491 if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 || 492 ifp->if_link_state != LINK_STATE_UP) { 493 lacp_port_disable(lp); 494 } else { 495 lacp_port_enable(lp); 496 } 497 lp->lp_key = lacp_compose_key(lp); 498 499 if (old_state != lp->lp_state || old_key != lp->lp_key) { 500 LACP_DPRINTF((lp, "-> UNSELECTED\n")); 501 lp->lp_selected = LACP_UNSELECTED; 502 } 503 LACP_UNLOCK(lsc); 504 } 505 506 static void 507 lacp_tick(void *arg) 508 { 509 struct lacp_softc *lsc = arg; 510 struct lacp_port *lp; 511 512 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) { 513 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) 514 continue; 515 516 CURVNET_SET(lp->lp_ifp->if_vnet); 517 lacp_run_timers(lp); 518 519 lacp_select(lp); 520 lacp_sm_mux(lp); 521 lacp_sm_tx(lp); 522 lacp_sm_ptx_tx_schedule(lp); 523 CURVNET_RESTORE(); 524 } 525 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc); 526 } 527 528 int 529 lacp_port_create(struct lagg_port *lgp) 530 { 531 struct lagg_softc *sc = lgp->lp_softc; 532 struct lacp_softc *lsc = LACP_SOFTC(sc); 533 struct lacp_port *lp; 534 struct ifnet *ifp = lgp->lp_ifp; 535 struct sockaddr_dl sdl; 536 struct ifmultiaddr *rifma = NULL; 537 int error; 538 539 link_init_sdl(ifp, (struct sockaddr *)&sdl, IFT_ETHER); 540 sdl.sdl_alen = ETHER_ADDR_LEN; 541 542 bcopy(ðermulticastaddr_slowprotocols, 543 LLADDR(&sdl), ETHER_ADDR_LEN); 544 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); 545 if (error) { 546 printf("%s: ADDMULTI failed on %s\n", __func__, 547 lgp->lp_ifp->if_xname); 548 return (error); 549 } 550 551 lp = malloc(sizeof(struct lacp_port), 552 M_DEVBUF, M_NOWAIT|M_ZERO); 553 if (lp == NULL) 554 return (ENOMEM); 555 556 LACP_LOCK(lsc); 557 lgp->lp_psc = lp; 558 lp->lp_ifp = ifp; 559 lp->lp_lagg = lgp; 560 lp->lp_lsc = lsc; 561 lp->lp_ifma = rifma; 562 563 LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next); 564 565 lacp_fill_actorinfo(lp, &lp->lp_actor); 566 lacp_fill_markerinfo(lp, &lp->lp_marker); 567 lp->lp_state = LACP_STATE_ACTIVITY; 568 lp->lp_aggregator = NULL; 569 lacp_sm_rx_set_expired(lp); 570 LACP_UNLOCK(lsc); 571 lacp_linkstate(lgp); 572 573 return (0); 574 } 575 576 void 577 lacp_port_destroy(struct lagg_port *lgp) 578 { 579 struct lacp_port *lp = LACP_PORT(lgp); 580 struct lacp_softc *lsc = lp->lp_lsc; 581 int i; 582 583 LACP_LOCK(lsc); 584 for (i = 0; i < LACP_NTIMER; i++) { 585 LACP_TIMER_DISARM(lp, i); 586 } 587 588 lacp_disable_collecting(lp); 589 lacp_disable_distributing(lp); 590 lacp_unselect(lp); 591 592 LIST_REMOVE(lp, lp_next); 593 LACP_UNLOCK(lsc); 594 595 /* The address may have already been removed by if_purgemaddrs() */ 596 if (!lgp->lp_detaching) 597 if_delmulti_ifma(lp->lp_ifma); 598 599 free(lp, M_DEVBUF); 600 } 601 602 void 603 lacp_req(struct lagg_softc *sc, void *data) 604 { 605 struct lacp_opreq *req = (struct lacp_opreq *)data; 606 struct lacp_softc *lsc = LACP_SOFTC(sc); 607 struct lacp_aggregator *la; 608 609 bzero(req, sizeof(struct lacp_opreq)); 610 611 /* 612 * If the LACP softc is NULL, return with the opreq structure full of 613 * zeros. It is normal for the softc to be NULL while the lagg is 614 * being destroyed. 615 */ 616 if (NULL == lsc) 617 return; 618 619 la = lsc->lsc_active_aggregator; 620 LACP_LOCK(lsc); 621 if (la != NULL) { 622 req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio); 623 memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac, 624 ETHER_ADDR_LEN); 625 req->actor_key = ntohs(la->la_actor.lip_key); 626 req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio); 627 req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno); 628 req->actor_state = la->la_actor.lip_state; 629 630 req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio); 631 memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac, 632 ETHER_ADDR_LEN); 633 req->partner_key = ntohs(la->la_partner.lip_key); 634 req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio); 635 req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno); 636 req->partner_state = la->la_partner.lip_state; 637 } 638 LACP_UNLOCK(lsc); 639 } 640 641 void 642 lacp_portreq(struct lagg_port *lgp, void *data) 643 { 644 struct lacp_opreq *req = (struct lacp_opreq *)data; 645 struct lacp_port *lp = LACP_PORT(lgp); 646 struct lacp_softc *lsc = lp->lp_lsc; 647 648 LACP_LOCK(lsc); 649 req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio); 650 memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac, 651 ETHER_ADDR_LEN); 652 req->actor_key = ntohs(lp->lp_actor.lip_key); 653 req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio); 654 req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno); 655 req->actor_state = lp->lp_actor.lip_state; 656 657 req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio); 658 memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac, 659 ETHER_ADDR_LEN); 660 req->partner_key = ntohs(lp->lp_partner.lip_key); 661 req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio); 662 req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno); 663 req->partner_state = lp->lp_partner.lip_state; 664 LACP_UNLOCK(lsc); 665 } 666 667 static void 668 lacp_disable_collecting(struct lacp_port *lp) 669 { 670 LACP_DPRINTF((lp, "collecting disabled\n")); 671 lp->lp_state &= ~LACP_STATE_COLLECTING; 672 } 673 674 static void 675 lacp_enable_collecting(struct lacp_port *lp) 676 { 677 LACP_DPRINTF((lp, "collecting enabled\n")); 678 lp->lp_state |= LACP_STATE_COLLECTING; 679 } 680 681 static void 682 lacp_disable_distributing(struct lacp_port *lp) 683 { 684 struct lacp_aggregator *la = lp->lp_aggregator; 685 struct lacp_softc *lsc = lp->lp_lsc; 686 struct lagg_softc *sc = lsc->lsc_softc; 687 char buf[LACP_LAGIDSTR_MAX+1]; 688 689 LACP_LOCK_ASSERT(lsc); 690 691 if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) { 692 return; 693 } 694 695 KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports")); 696 KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports)); 697 KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid")); 698 699 LACP_DPRINTF((lp, "disable distributing on aggregator %s, " 700 "nports %d -> %d\n", 701 lacp_format_lagid_aggregator(la, buf, sizeof(buf)), 702 la->la_nports, la->la_nports - 1)); 703 704 TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q); 705 la->la_nports--; 706 sc->sc_active = la->la_nports; 707 708 if (lsc->lsc_active_aggregator == la) { 709 lacp_suppress_distributing(lsc, la); 710 lacp_select_active_aggregator(lsc); 711 /* regenerate the port map, the active aggregator has changed */ 712 lacp_update_portmap(lsc); 713 } 714 715 lp->lp_state &= ~LACP_STATE_DISTRIBUTING; 716 if_link_state_change(sc->sc_ifp, 717 sc->sc_active ? LINK_STATE_UP : LINK_STATE_DOWN); 718 } 719 720 static void 721 lacp_enable_distributing(struct lacp_port *lp) 722 { 723 struct lacp_aggregator *la = lp->lp_aggregator; 724 struct lacp_softc *lsc = lp->lp_lsc; 725 struct lagg_softc *sc = lsc->lsc_softc; 726 char buf[LACP_LAGIDSTR_MAX+1]; 727 728 LACP_LOCK_ASSERT(lsc); 729 730 if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) { 731 return; 732 } 733 734 LACP_DPRINTF((lp, "enable distributing on aggregator %s, " 735 "nports %d -> %d\n", 736 lacp_format_lagid_aggregator(la, buf, sizeof(buf)), 737 la->la_nports, la->la_nports + 1)); 738 739 KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid")); 740 TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q); 741 la->la_nports++; 742 sc->sc_active = la->la_nports; 743 744 lp->lp_state |= LACP_STATE_DISTRIBUTING; 745 746 if (lsc->lsc_active_aggregator == la) { 747 lacp_suppress_distributing(lsc, la); 748 lacp_update_portmap(lsc); 749 } else 750 /* try to become the active aggregator */ 751 lacp_select_active_aggregator(lsc); 752 753 if_link_state_change(sc->sc_ifp, 754 sc->sc_active ? LINK_STATE_UP : LINK_STATE_DOWN); 755 } 756 757 static void 758 lacp_transit_expire(void *vp) 759 { 760 struct lacp_softc *lsc = vp; 761 762 LACP_LOCK_ASSERT(lsc); 763 764 CURVNET_SET(lsc->lsc_softc->sc_ifp->if_vnet); 765 LACP_TRACE(NULL); 766 CURVNET_RESTORE(); 767 768 lsc->lsc_suppress_distributing = FALSE; 769 } 770 771 void 772 lacp_attach(struct lagg_softc *sc) 773 { 774 struct lacp_softc *lsc; 775 776 lsc = malloc(sizeof(struct lacp_softc), M_DEVBUF, M_WAITOK | M_ZERO); 777 778 sc->sc_psc = lsc; 779 lsc->lsc_softc = sc; 780 781 lsc->lsc_hashkey = m_ether_tcpip_hash_init(); 782 lsc->lsc_active_aggregator = NULL; 783 lsc->lsc_strict_mode = VNET(lacp_default_strict_mode); 784 LACP_LOCK_INIT(lsc); 785 TAILQ_INIT(&lsc->lsc_aggregators); 786 LIST_INIT(&lsc->lsc_ports); 787 788 callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0); 789 callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0); 790 791 /* if the lagg is already up then do the same */ 792 if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) 793 lacp_init(sc); 794 } 795 796 void 797 lacp_detach(void *psc) 798 { 799 struct lacp_softc *lsc = (struct lacp_softc *)psc; 800 801 KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators), 802 ("aggregators still active")); 803 KASSERT(lsc->lsc_active_aggregator == NULL, 804 ("aggregator still attached")); 805 806 callout_drain(&lsc->lsc_transit_callout); 807 callout_drain(&lsc->lsc_callout); 808 809 LACP_LOCK_DESTROY(lsc); 810 free(lsc, M_DEVBUF); 811 } 812 813 void 814 lacp_init(struct lagg_softc *sc) 815 { 816 struct lacp_softc *lsc = LACP_SOFTC(sc); 817 818 LACP_LOCK(lsc); 819 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc); 820 LACP_UNLOCK(lsc); 821 } 822 823 void 824 lacp_stop(struct lagg_softc *sc) 825 { 826 struct lacp_softc *lsc = LACP_SOFTC(sc); 827 828 LACP_LOCK(lsc); 829 callout_stop(&lsc->lsc_transit_callout); 830 callout_stop(&lsc->lsc_callout); 831 LACP_UNLOCK(lsc); 832 } 833 834 struct lagg_port * 835 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m) 836 { 837 struct lacp_softc *lsc = LACP_SOFTC(sc); 838 struct lacp_portmap *pm; 839 struct lacp_port *lp; 840 struct lacp_port **map; 841 uint32_t hash; 842 int count; 843 844 if (__predict_false(lsc->lsc_suppress_distributing)) { 845 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__)); 846 return (NULL); 847 } 848 849 pm = &lsc->lsc_pmap[lsc->lsc_activemap]; 850 if (pm->pm_count == 0) { 851 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__)); 852 return (NULL); 853 } 854 855 #ifdef NUMA 856 if ((sc->sc_opts & LAGG_OPT_USE_NUMA) && 857 pm->pm_num_dom > 1 && m->m_pkthdr.numa_domain < MAXMEMDOM) { 858 count = pm->pm_numa[m->m_pkthdr.numa_domain].count; 859 if (count > 0) { 860 map = pm->pm_numa[m->m_pkthdr.numa_domain].map; 861 } else { 862 /* No ports on this domain; use global hash. */ 863 map = pm->pm_map; 864 count = pm->pm_count; 865 } 866 } else 867 #endif 868 { 869 map = pm->pm_map; 870 count = pm->pm_count; 871 } 872 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && 873 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 874 hash = m->m_pkthdr.flowid >> sc->flowid_shift; 875 else 876 hash = m_ether_tcpip_hash(sc->sc_flags, m, lsc->lsc_hashkey); 877 878 hash %= count; 879 lp = map[hash]; 880 881 KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0, 882 ("aggregated port is not distributing")); 883 884 return (lp->lp_lagg); 885 } 886 887 #if defined(RATELIMIT) || defined(KERN_TLS) 888 struct lagg_port * 889 lacp_select_tx_port_by_hash(struct lagg_softc *sc, uint32_t flowid) 890 { 891 struct lacp_softc *lsc = LACP_SOFTC(sc); 892 struct lacp_portmap *pm; 893 struct lacp_port *lp; 894 uint32_t hash; 895 896 if (__predict_false(lsc->lsc_suppress_distributing)) { 897 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__)); 898 return (NULL); 899 } 900 901 pm = &lsc->lsc_pmap[lsc->lsc_activemap]; 902 if (pm->pm_count == 0) { 903 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__)); 904 return (NULL); 905 } 906 907 hash = flowid >> sc->flowid_shift; 908 hash %= pm->pm_count; 909 lp = pm->pm_map[hash]; 910 911 return (lp->lp_lagg); 912 } 913 #endif 914 915 /* 916 * lacp_suppress_distributing: drop transmit packets for a while 917 * to preserve packet ordering. 918 */ 919 920 static void 921 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la) 922 { 923 struct lacp_port *lp; 924 925 if (lsc->lsc_active_aggregator != la) { 926 return; 927 } 928 929 LACP_TRACE(NULL); 930 931 lsc->lsc_suppress_distributing = TRUE; 932 933 /* send a marker frame down each port to verify the queues are empty */ 934 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) { 935 lp->lp_flags |= LACP_PORT_MARK; 936 lacp_xmit_marker(lp); 937 } 938 939 /* set a timeout for the marker frames */ 940 callout_reset(&lsc->lsc_transit_callout, 941 LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc); 942 } 943 944 static int 945 lacp_compare_peerinfo(const struct lacp_peerinfo *a, 946 const struct lacp_peerinfo *b) 947 { 948 return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state))); 949 } 950 951 static int 952 lacp_compare_systemid(const struct lacp_systemid *a, 953 const struct lacp_systemid *b) 954 { 955 return (memcmp(a, b, sizeof(*a))); 956 } 957 958 #if 0 /* unused */ 959 static int 960 lacp_compare_portid(const struct lacp_portid *a, 961 const struct lacp_portid *b) 962 { 963 return (memcmp(a, b, sizeof(*a))); 964 } 965 #endif 966 967 static uint64_t 968 lacp_aggregator_bandwidth(struct lacp_aggregator *la) 969 { 970 struct lacp_port *lp; 971 uint64_t speed; 972 973 lp = TAILQ_FIRST(&la->la_ports); 974 if (lp == NULL) { 975 return (0); 976 } 977 978 speed = ifmedia_baudrate(lp->lp_media); 979 speed *= la->la_nports; 980 if (speed == 0) { 981 LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n", 982 lp->lp_media, la->la_nports)); 983 } 984 985 return (speed); 986 } 987 988 /* 989 * lacp_select_active_aggregator: select an aggregator to be used to transmit 990 * packets from lagg(4) interface. 991 */ 992 993 static void 994 lacp_select_active_aggregator(struct lacp_softc *lsc) 995 { 996 struct lacp_aggregator *la; 997 struct lacp_aggregator *best_la = NULL; 998 uint64_t best_speed = 0; 999 char buf[LACP_LAGIDSTR_MAX+1]; 1000 1001 LACP_TRACE(NULL); 1002 1003 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) { 1004 uint64_t speed; 1005 1006 if (la->la_nports == 0) { 1007 continue; 1008 } 1009 1010 speed = lacp_aggregator_bandwidth(la); 1011 LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n", 1012 lacp_format_lagid_aggregator(la, buf, sizeof(buf)), 1013 speed, la->la_nports)); 1014 1015 /* 1016 * This aggregator is chosen if the partner has a better 1017 * system priority or, the total aggregated speed is higher 1018 * or, it is already the chosen aggregator 1019 */ 1020 if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) < 1021 LACP_SYS_PRI(best_la->la_partner)) || 1022 speed > best_speed || 1023 (speed == best_speed && 1024 la == lsc->lsc_active_aggregator)) { 1025 best_la = la; 1026 best_speed = speed; 1027 } 1028 } 1029 1030 KASSERT(best_la == NULL || best_la->la_nports > 0, 1031 ("invalid aggregator refcnt")); 1032 KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports), 1033 ("invalid aggregator list")); 1034 1035 if (lsc->lsc_active_aggregator != best_la) { 1036 LACP_DPRINTF((NULL, "active aggregator changed\n")); 1037 LACP_DPRINTF((NULL, "old %s\n", 1038 lacp_format_lagid_aggregator(lsc->lsc_active_aggregator, 1039 buf, sizeof(buf)))); 1040 } else { 1041 LACP_DPRINTF((NULL, "active aggregator not changed\n")); 1042 } 1043 LACP_DPRINTF((NULL, "new %s\n", 1044 lacp_format_lagid_aggregator(best_la, buf, sizeof(buf)))); 1045 1046 if (lsc->lsc_active_aggregator != best_la) { 1047 lsc->lsc_active_aggregator = best_la; 1048 lacp_update_portmap(lsc); 1049 if (best_la) { 1050 lacp_suppress_distributing(lsc, best_la); 1051 } 1052 } 1053 } 1054 1055 /* 1056 * Updated the inactive portmap array with the new list of ports and 1057 * make it live. 1058 */ 1059 static void 1060 lacp_update_portmap(struct lacp_softc *lsc) 1061 { 1062 struct lagg_softc *sc = lsc->lsc_softc; 1063 struct lacp_aggregator *la; 1064 struct lacp_portmap *p; 1065 struct lacp_port *lp; 1066 uint64_t speed; 1067 u_int newmap; 1068 int i; 1069 #ifdef NUMA 1070 int count; 1071 uint8_t domain; 1072 #endif 1073 1074 newmap = lsc->lsc_activemap == 0 ? 1 : 0; 1075 p = &lsc->lsc_pmap[newmap]; 1076 la = lsc->lsc_active_aggregator; 1077 speed = 0; 1078 bzero(p, sizeof(struct lacp_portmap)); 1079 1080 if (la != NULL && la->la_nports > 0) { 1081 p->pm_count = la->la_nports; 1082 i = 0; 1083 TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q) { 1084 p->pm_map[i++] = lp; 1085 #ifdef NUMA 1086 domain = lp->lp_ifp->if_numa_domain; 1087 if (domain >= MAXMEMDOM) 1088 continue; 1089 count = p->pm_numa[domain].count; 1090 p->pm_numa[domain].map[count] = lp; 1091 p->pm_numa[domain].count++; 1092 #endif 1093 } 1094 KASSERT(i == p->pm_count, ("Invalid port count")); 1095 1096 #ifdef NUMA 1097 for (i = 0; i < MAXMEMDOM; i++) { 1098 if (p->pm_numa[i].count != 0) 1099 p->pm_num_dom++; 1100 } 1101 #endif 1102 speed = lacp_aggregator_bandwidth(la); 1103 } 1104 sc->sc_ifp->if_baudrate = speed; 1105 1106 /* switch the active portmap over */ 1107 atomic_store_rel_int(&lsc->lsc_activemap, newmap); 1108 LACP_DPRINTF((NULL, "Set table %d with %d ports\n", 1109 lsc->lsc_activemap, 1110 lsc->lsc_pmap[lsc->lsc_activemap].pm_count)); 1111 } 1112 1113 static uint16_t 1114 lacp_compose_key(struct lacp_port *lp) 1115 { 1116 struct lagg_port *lgp = lp->lp_lagg; 1117 struct lagg_softc *sc = lgp->lp_softc; 1118 u_int media = lp->lp_media; 1119 uint16_t key; 1120 1121 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) { 1122 1123 /* 1124 * non-aggregatable links should have unique keys. 1125 * 1126 * XXX this isn't really unique as if_index is 16 bit. 1127 */ 1128 1129 /* bit 0..14: (some bits of) if_index of this port */ 1130 key = lp->lp_ifp->if_index; 1131 /* bit 15: 1 */ 1132 key |= 0x8000; 1133 } else { 1134 u_int subtype = IFM_SUBTYPE(media); 1135 1136 KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type")); 1137 KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface")); 1138 1139 /* bit 0..4: IFM_SUBTYPE modulo speed */ 1140 switch (subtype) { 1141 case IFM_10_T: 1142 case IFM_10_2: 1143 case IFM_10_5: 1144 case IFM_10_STP: 1145 case IFM_10_FL: 1146 key = IFM_10_T; 1147 break; 1148 case IFM_100_TX: 1149 case IFM_100_FX: 1150 case IFM_100_T4: 1151 case IFM_100_VG: 1152 case IFM_100_T2: 1153 case IFM_100_T: 1154 case IFM_100_SGMII: 1155 key = IFM_100_TX; 1156 break; 1157 case IFM_1000_SX: 1158 case IFM_1000_LX: 1159 case IFM_1000_CX: 1160 case IFM_1000_T: 1161 case IFM_1000_KX: 1162 case IFM_1000_SGMII: 1163 case IFM_1000_CX_SGMII: 1164 key = IFM_1000_SX; 1165 break; 1166 case IFM_10G_LR: 1167 case IFM_10G_SR: 1168 case IFM_10G_CX4: 1169 case IFM_10G_TWINAX: 1170 case IFM_10G_TWINAX_LONG: 1171 case IFM_10G_LRM: 1172 case IFM_10G_T: 1173 case IFM_10G_KX4: 1174 case IFM_10G_KR: 1175 case IFM_10G_CR1: 1176 case IFM_10G_ER: 1177 case IFM_10G_SFI: 1178 case IFM_10G_AOC: 1179 key = IFM_10G_LR; 1180 break; 1181 case IFM_20G_KR2: 1182 key = IFM_20G_KR2; 1183 break; 1184 case IFM_2500_KX: 1185 case IFM_2500_T: 1186 case IFM_2500_X: 1187 key = IFM_2500_KX; 1188 break; 1189 case IFM_5000_T: 1190 case IFM_5000_KR: 1191 case IFM_5000_KR_S: 1192 case IFM_5000_KR1: 1193 key = IFM_5000_T; 1194 break; 1195 case IFM_50G_PCIE: 1196 case IFM_50G_CR2: 1197 case IFM_50G_KR2: 1198 case IFM_50G_KR4: 1199 case IFM_50G_SR2: 1200 case IFM_50G_LR2: 1201 case IFM_50G_LAUI2_AC: 1202 case IFM_50G_LAUI2: 1203 case IFM_50G_AUI2_AC: 1204 case IFM_50G_AUI2: 1205 case IFM_50G_CP: 1206 case IFM_50G_SR: 1207 case IFM_50G_LR: 1208 case IFM_50G_FR: 1209 case IFM_50G_KR_PAM4: 1210 case IFM_50G_AUI1_AC: 1211 case IFM_50G_AUI1: 1212 key = IFM_50G_PCIE; 1213 break; 1214 case IFM_56G_R4: 1215 key = IFM_56G_R4; 1216 break; 1217 case IFM_25G_PCIE: 1218 case IFM_25G_CR: 1219 case IFM_25G_KR: 1220 case IFM_25G_SR: 1221 case IFM_25G_LR: 1222 case IFM_25G_ACC: 1223 case IFM_25G_AOC: 1224 case IFM_25G_T: 1225 case IFM_25G_CR_S: 1226 case IFM_25G_CR1: 1227 case IFM_25G_KR_S: 1228 case IFM_25G_AUI: 1229 case IFM_25G_KR1: 1230 key = IFM_25G_PCIE; 1231 break; 1232 case IFM_40G_CR4: 1233 case IFM_40G_SR4: 1234 case IFM_40G_LR4: 1235 case IFM_40G_XLPPI: 1236 case IFM_40G_KR4: 1237 case IFM_40G_XLAUI: 1238 case IFM_40G_XLAUI_AC: 1239 case IFM_40G_ER4: 1240 key = IFM_40G_CR4; 1241 break; 1242 case IFM_100G_CR4: 1243 case IFM_100G_SR4: 1244 case IFM_100G_KR4: 1245 case IFM_100G_LR4: 1246 case IFM_100G_CAUI4_AC: 1247 case IFM_100G_CAUI4: 1248 case IFM_100G_AUI4_AC: 1249 case IFM_100G_AUI4: 1250 case IFM_100G_CR_PAM4: 1251 case IFM_100G_KR_PAM4: 1252 case IFM_100G_CP2: 1253 case IFM_100G_SR2: 1254 case IFM_100G_DR: 1255 case IFM_100G_KR2_PAM4: 1256 case IFM_100G_CAUI2_AC: 1257 case IFM_100G_CAUI2: 1258 case IFM_100G_AUI2_AC: 1259 case IFM_100G_AUI2: 1260 key = IFM_100G_CR4; 1261 break; 1262 case IFM_200G_CR4_PAM4: 1263 case IFM_200G_SR4: 1264 case IFM_200G_FR4: 1265 case IFM_200G_LR4: 1266 case IFM_200G_DR4: 1267 case IFM_200G_KR4_PAM4: 1268 case IFM_200G_AUI4_AC: 1269 case IFM_200G_AUI4: 1270 case IFM_200G_AUI8_AC: 1271 case IFM_200G_AUI8: 1272 key = IFM_200G_CR4_PAM4; 1273 break; 1274 case IFM_400G_FR8: 1275 case IFM_400G_LR8: 1276 case IFM_400G_DR4: 1277 case IFM_400G_AUI8_AC: 1278 case IFM_400G_AUI8: 1279 key = IFM_400G_FR8; 1280 break; 1281 default: 1282 key = subtype; 1283 break; 1284 } 1285 /* bit 5..14: (some bits of) if_index of lagg device */ 1286 key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5); 1287 /* bit 15: 0 */ 1288 } 1289 return (htons(key)); 1290 } 1291 1292 static void 1293 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la) 1294 { 1295 char buf[LACP_LAGIDSTR_MAX+1]; 1296 1297 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n", 1298 __func__, 1299 lacp_format_lagid(&la->la_actor, &la->la_partner, 1300 buf, sizeof(buf)), 1301 la->la_refcnt, la->la_refcnt + 1)); 1302 1303 KASSERT(la->la_refcnt > 0, ("refcount <= 0")); 1304 la->la_refcnt++; 1305 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount")); 1306 } 1307 1308 static void 1309 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la) 1310 { 1311 char buf[LACP_LAGIDSTR_MAX+1]; 1312 1313 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n", 1314 __func__, 1315 lacp_format_lagid(&la->la_actor, &la->la_partner, 1316 buf, sizeof(buf)), 1317 la->la_refcnt, la->la_refcnt - 1)); 1318 1319 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt")); 1320 la->la_refcnt--; 1321 if (la->la_refcnt > 0) { 1322 return; 1323 } 1324 1325 KASSERT(la->la_refcnt == 0, ("refcount not zero")); 1326 KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active")); 1327 1328 TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q); 1329 1330 free(la, M_DEVBUF); 1331 } 1332 1333 /* 1334 * lacp_aggregator_get: allocate an aggregator. 1335 */ 1336 1337 static struct lacp_aggregator * 1338 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp) 1339 { 1340 struct lacp_aggregator *la; 1341 1342 la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT); 1343 if (la) { 1344 la->la_refcnt = 1; 1345 la->la_nports = 0; 1346 TAILQ_INIT(&la->la_ports); 1347 la->la_pending = 0; 1348 TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q); 1349 } 1350 1351 return (la); 1352 } 1353 1354 /* 1355 * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port. 1356 */ 1357 1358 static void 1359 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp) 1360 { 1361 lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner); 1362 lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor); 1363 1364 la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION; 1365 } 1366 1367 static void 1368 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr, 1369 const struct lacp_peerinfo *lpi_port) 1370 { 1371 memset(lpi_aggr, 0, sizeof(*lpi_aggr)); 1372 lpi_aggr->lip_systemid = lpi_port->lip_systemid; 1373 lpi_aggr->lip_key = lpi_port->lip_key; 1374 } 1375 1376 /* 1377 * lacp_aggregator_is_compatible: check if a port can join to an aggregator. 1378 */ 1379 1380 static int 1381 lacp_aggregator_is_compatible(const struct lacp_aggregator *la, 1382 const struct lacp_port *lp) 1383 { 1384 if (!(lp->lp_state & LACP_STATE_AGGREGATION) || 1385 !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) { 1386 return (0); 1387 } 1388 1389 if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) { 1390 return (0); 1391 } 1392 1393 if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) { 1394 return (0); 1395 } 1396 1397 if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) { 1398 return (0); 1399 } 1400 1401 return (1); 1402 } 1403 1404 static int 1405 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a, 1406 const struct lacp_peerinfo *b) 1407 { 1408 if (memcmp(&a->lip_systemid, &b->lip_systemid, 1409 sizeof(a->lip_systemid))) { 1410 return (0); 1411 } 1412 1413 if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) { 1414 return (0); 1415 } 1416 1417 return (1); 1418 } 1419 1420 static void 1421 lacp_port_enable(struct lacp_port *lp) 1422 { 1423 lp->lp_state |= LACP_STATE_AGGREGATION; 1424 } 1425 1426 static void 1427 lacp_port_disable(struct lacp_port *lp) 1428 { 1429 lacp_set_mux(lp, LACP_MUX_DETACHED); 1430 1431 lp->lp_state &= ~LACP_STATE_AGGREGATION; 1432 lp->lp_selected = LACP_UNSELECTED; 1433 lacp_sm_rx_record_default(lp); 1434 lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION; 1435 lp->lp_state &= ~LACP_STATE_EXPIRED; 1436 } 1437 1438 /* 1439 * lacp_select: select an aggregator. create one if necessary. 1440 */ 1441 static void 1442 lacp_select(struct lacp_port *lp) 1443 { 1444 struct lacp_softc *lsc = lp->lp_lsc; 1445 struct lacp_aggregator *la; 1446 char buf[LACP_LAGIDSTR_MAX+1]; 1447 1448 if (lp->lp_aggregator) { 1449 return; 1450 } 1451 1452 /* If we haven't heard from our peer, skip this step. */ 1453 if (lp->lp_state & LACP_STATE_DEFAULTED) 1454 return; 1455 1456 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), 1457 ("timer_wait_while still active")); 1458 1459 LACP_DPRINTF((lp, "port lagid=%s\n", 1460 lacp_format_lagid(&lp->lp_actor, &lp->lp_partner, 1461 buf, sizeof(buf)))); 1462 1463 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) { 1464 if (lacp_aggregator_is_compatible(la, lp)) { 1465 break; 1466 } 1467 } 1468 1469 if (la == NULL) { 1470 la = lacp_aggregator_get(lsc, lp); 1471 if (la == NULL) { 1472 LACP_DPRINTF((lp, "aggregator creation failed\n")); 1473 1474 /* 1475 * will retry on the next tick. 1476 */ 1477 1478 return; 1479 } 1480 lacp_fill_aggregator_id(la, lp); 1481 LACP_DPRINTF((lp, "aggregator created\n")); 1482 } else { 1483 LACP_DPRINTF((lp, "compatible aggregator found\n")); 1484 if (la->la_refcnt == LACP_MAX_PORTS) 1485 return; 1486 lacp_aggregator_addref(lsc, la); 1487 } 1488 1489 LACP_DPRINTF((lp, "aggregator lagid=%s\n", 1490 lacp_format_lagid(&la->la_actor, &la->la_partner, 1491 buf, sizeof(buf)))); 1492 1493 lp->lp_aggregator = la; 1494 lp->lp_selected = LACP_SELECTED; 1495 } 1496 1497 /* 1498 * lacp_unselect: finish unselect/detach process. 1499 */ 1500 1501 static void 1502 lacp_unselect(struct lacp_port *lp) 1503 { 1504 struct lacp_softc *lsc = lp->lp_lsc; 1505 struct lacp_aggregator *la = lp->lp_aggregator; 1506 1507 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), 1508 ("timer_wait_while still active")); 1509 1510 if (la == NULL) { 1511 return; 1512 } 1513 1514 lp->lp_aggregator = NULL; 1515 lacp_aggregator_delref(lsc, la); 1516 } 1517 1518 /* mux machine */ 1519 1520 static void 1521 lacp_sm_mux(struct lacp_port *lp) 1522 { 1523 struct lagg_port *lgp = lp->lp_lagg; 1524 struct lagg_softc *sc = lgp->lp_softc; 1525 enum lacp_mux_state new_state; 1526 boolean_t p_sync = 1527 (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0; 1528 boolean_t p_collecting = 1529 (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0; 1530 enum lacp_selected selected = lp->lp_selected; 1531 struct lacp_aggregator *la; 1532 1533 if (V_lacp_debug > 1) 1534 lacp_dprintf(lp, "%s: state= 0x%x, selected= 0x%x, " 1535 "p_sync= 0x%x, p_collecting= 0x%x\n", __func__, 1536 lp->lp_mux_state, selected, p_sync, p_collecting); 1537 1538 re_eval: 1539 la = lp->lp_aggregator; 1540 KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL, 1541 ("MUX not detached")); 1542 new_state = lp->lp_mux_state; 1543 switch (lp->lp_mux_state) { 1544 case LACP_MUX_DETACHED: 1545 if (selected != LACP_UNSELECTED) { 1546 new_state = LACP_MUX_WAITING; 1547 } 1548 break; 1549 case LACP_MUX_WAITING: 1550 KASSERT(la->la_pending > 0 || 1551 !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), 1552 ("timer_wait_while still active")); 1553 if (selected == LACP_SELECTED && la->la_pending == 0) { 1554 new_state = LACP_MUX_ATTACHED; 1555 } else if (selected == LACP_UNSELECTED) { 1556 new_state = LACP_MUX_DETACHED; 1557 } 1558 break; 1559 case LACP_MUX_ATTACHED: 1560 if (selected == LACP_SELECTED && p_sync) { 1561 new_state = LACP_MUX_COLLECTING; 1562 } else if (selected != LACP_SELECTED) { 1563 new_state = LACP_MUX_DETACHED; 1564 } 1565 break; 1566 case LACP_MUX_COLLECTING: 1567 if (selected == LACP_SELECTED && p_sync && p_collecting) { 1568 new_state = LACP_MUX_DISTRIBUTING; 1569 } else if (selected != LACP_SELECTED || !p_sync) { 1570 new_state = LACP_MUX_ATTACHED; 1571 } 1572 break; 1573 case LACP_MUX_DISTRIBUTING: 1574 if (selected != LACP_SELECTED || !p_sync || !p_collecting) { 1575 new_state = LACP_MUX_COLLECTING; 1576 lacp_dprintf(lp, "Interface stopped DISTRIBUTING, possible flapping\n"); 1577 sc->sc_flapping++; 1578 } 1579 break; 1580 default: 1581 panic("%s: unknown state", __func__); 1582 } 1583 1584 if (lp->lp_mux_state == new_state) { 1585 return; 1586 } 1587 1588 lacp_set_mux(lp, new_state); 1589 goto re_eval; 1590 } 1591 1592 static void 1593 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state) 1594 { 1595 struct lacp_aggregator *la = lp->lp_aggregator; 1596 1597 if (lp->lp_mux_state == new_state) { 1598 return; 1599 } 1600 1601 switch (new_state) { 1602 case LACP_MUX_DETACHED: 1603 lp->lp_state &= ~LACP_STATE_SYNC; 1604 lacp_disable_distributing(lp); 1605 lacp_disable_collecting(lp); 1606 lacp_sm_assert_ntt(lp); 1607 /* cancel timer */ 1608 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) { 1609 KASSERT(la->la_pending > 0, 1610 ("timer_wait_while not active")); 1611 la->la_pending--; 1612 } 1613 LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE); 1614 lacp_unselect(lp); 1615 break; 1616 case LACP_MUX_WAITING: 1617 LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE, 1618 LACP_AGGREGATE_WAIT_TIME); 1619 la->la_pending++; 1620 break; 1621 case LACP_MUX_ATTACHED: 1622 lp->lp_state |= LACP_STATE_SYNC; 1623 lacp_disable_collecting(lp); 1624 lacp_sm_assert_ntt(lp); 1625 break; 1626 case LACP_MUX_COLLECTING: 1627 lacp_enable_collecting(lp); 1628 lacp_disable_distributing(lp); 1629 lacp_sm_assert_ntt(lp); 1630 break; 1631 case LACP_MUX_DISTRIBUTING: 1632 lacp_enable_distributing(lp); 1633 break; 1634 default: 1635 panic("%s: unknown state", __func__); 1636 } 1637 1638 LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state)); 1639 1640 lp->lp_mux_state = new_state; 1641 } 1642 1643 static void 1644 lacp_sm_mux_timer(struct lacp_port *lp) 1645 { 1646 struct lacp_aggregator *la = lp->lp_aggregator; 1647 char buf[LACP_LAGIDSTR_MAX+1]; 1648 1649 KASSERT(la->la_pending > 0, ("no pending event")); 1650 1651 LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__, 1652 lacp_format_lagid(&la->la_actor, &la->la_partner, 1653 buf, sizeof(buf)), 1654 la->la_pending, la->la_pending - 1)); 1655 1656 la->la_pending--; 1657 } 1658 1659 /* periodic transmit machine */ 1660 1661 static void 1662 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate) 1663 { 1664 if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state, 1665 LACP_STATE_TIMEOUT)) { 1666 return; 1667 } 1668 1669 LACP_DPRINTF((lp, "partner timeout changed\n")); 1670 1671 /* 1672 * FAST_PERIODIC -> SLOW_PERIODIC 1673 * or 1674 * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC 1675 * 1676 * let lacp_sm_ptx_tx_schedule to update timeout. 1677 */ 1678 1679 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC); 1680 1681 /* 1682 * if timeout has been shortened, assert NTT. 1683 */ 1684 1685 if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) { 1686 lacp_sm_assert_ntt(lp); 1687 } 1688 } 1689 1690 static void 1691 lacp_sm_ptx_tx_schedule(struct lacp_port *lp) 1692 { 1693 int timeout; 1694 1695 if (!(lp->lp_state & LACP_STATE_ACTIVITY) && 1696 !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) { 1697 1698 /* 1699 * NO_PERIODIC 1700 */ 1701 1702 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC); 1703 return; 1704 } 1705 1706 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) { 1707 return; 1708 } 1709 1710 timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ? 1711 LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME; 1712 1713 LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout); 1714 } 1715 1716 static void 1717 lacp_sm_ptx_timer(struct lacp_port *lp) 1718 { 1719 lacp_sm_assert_ntt(lp); 1720 } 1721 1722 static void 1723 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du) 1724 { 1725 int timeout; 1726 1727 /* 1728 * check LACP_DISABLED first 1729 */ 1730 1731 if (!(lp->lp_state & LACP_STATE_AGGREGATION)) { 1732 return; 1733 } 1734 1735 /* 1736 * check loopback condition. 1737 */ 1738 1739 if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid, 1740 &lp->lp_actor.lip_systemid)) { 1741 return; 1742 } 1743 1744 /* 1745 * EXPIRED, DEFAULTED, CURRENT -> CURRENT 1746 */ 1747 1748 lacp_sm_rx_update_selected(lp, du); 1749 lacp_sm_rx_update_ntt(lp, du); 1750 lacp_sm_rx_record_pdu(lp, du); 1751 1752 timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ? 1753 LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME; 1754 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout); 1755 1756 lp->lp_state &= ~LACP_STATE_EXPIRED; 1757 1758 /* 1759 * kick transmit machine without waiting the next tick. 1760 */ 1761 1762 lacp_sm_tx(lp); 1763 } 1764 1765 static void 1766 lacp_sm_rx_set_expired(struct lacp_port *lp) 1767 { 1768 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC; 1769 lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT; 1770 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME); 1771 lp->lp_state |= LACP_STATE_EXPIRED; 1772 } 1773 1774 static void 1775 lacp_sm_rx_timer(struct lacp_port *lp) 1776 { 1777 if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) { 1778 /* CURRENT -> EXPIRED */ 1779 LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__)); 1780 lacp_sm_rx_set_expired(lp); 1781 } else { 1782 /* EXPIRED -> DEFAULTED */ 1783 LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__)); 1784 lacp_sm_rx_update_default_selected(lp); 1785 lacp_sm_rx_record_default(lp); 1786 lp->lp_state &= ~LACP_STATE_EXPIRED; 1787 } 1788 } 1789 1790 static void 1791 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du) 1792 { 1793 boolean_t active; 1794 uint8_t oldpstate; 1795 char buf[LACP_STATESTR_MAX+1]; 1796 1797 LACP_TRACE(lp); 1798 1799 oldpstate = lp->lp_partner.lip_state; 1800 1801 active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY) 1802 || ((lp->lp_state & LACP_STATE_ACTIVITY) && 1803 (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY)); 1804 1805 lp->lp_partner = du->ldu_actor; 1806 if (active && 1807 ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state, 1808 LACP_STATE_AGGREGATION) && 1809 !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner)) 1810 || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) { 1811 /* 1812 * XXX Maintain legacy behavior of leaving the 1813 * LACP_STATE_SYNC bit unchanged from the partner's 1814 * advertisement if lsc_strict_mode is false. 1815 * TODO: We should re-examine the concept of the "strict mode" 1816 * to ensure it makes sense to maintain a non-strict mode. 1817 */ 1818 if (lp->lp_lsc->lsc_strict_mode) 1819 lp->lp_partner.lip_state |= LACP_STATE_SYNC; 1820 } else { 1821 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC; 1822 } 1823 1824 lp->lp_state &= ~LACP_STATE_DEFAULTED; 1825 1826 if (oldpstate != lp->lp_partner.lip_state) { 1827 LACP_DPRINTF((lp, "old pstate %s\n", 1828 lacp_format_state(oldpstate, buf, sizeof(buf)))); 1829 LACP_DPRINTF((lp, "new pstate %s\n", 1830 lacp_format_state(lp->lp_partner.lip_state, buf, 1831 sizeof(buf)))); 1832 } 1833 1834 lacp_sm_ptx_update_timeout(lp, oldpstate); 1835 } 1836 1837 static void 1838 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du) 1839 { 1840 1841 LACP_TRACE(lp); 1842 1843 if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) || 1844 !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state, 1845 LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) { 1846 LACP_DPRINTF((lp, "%s: assert ntt\n", __func__)); 1847 lacp_sm_assert_ntt(lp); 1848 } 1849 } 1850 1851 static void 1852 lacp_sm_rx_record_default(struct lacp_port *lp) 1853 { 1854 uint8_t oldpstate; 1855 1856 LACP_TRACE(lp); 1857 1858 oldpstate = lp->lp_partner.lip_state; 1859 if (lp->lp_lsc->lsc_strict_mode) 1860 lp->lp_partner = lacp_partner_admin_strict; 1861 else 1862 lp->lp_partner = lacp_partner_admin_optimistic; 1863 lp->lp_state |= LACP_STATE_DEFAULTED; 1864 lacp_sm_ptx_update_timeout(lp, oldpstate); 1865 } 1866 1867 static void 1868 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp, 1869 const struct lacp_peerinfo *info) 1870 { 1871 1872 LACP_TRACE(lp); 1873 1874 if (lacp_compare_peerinfo(&lp->lp_partner, info) || 1875 !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state, 1876 LACP_STATE_AGGREGATION)) { 1877 lp->lp_selected = LACP_UNSELECTED; 1878 /* mux machine will clean up lp->lp_aggregator */ 1879 } 1880 } 1881 1882 static void 1883 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du) 1884 { 1885 1886 LACP_TRACE(lp); 1887 1888 lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor); 1889 } 1890 1891 static void 1892 lacp_sm_rx_update_default_selected(struct lacp_port *lp) 1893 { 1894 1895 LACP_TRACE(lp); 1896 1897 if (lp->lp_lsc->lsc_strict_mode) 1898 lacp_sm_rx_update_selected_from_peerinfo(lp, 1899 &lacp_partner_admin_strict); 1900 else 1901 lacp_sm_rx_update_selected_from_peerinfo(lp, 1902 &lacp_partner_admin_optimistic); 1903 } 1904 1905 /* transmit machine */ 1906 1907 static void 1908 lacp_sm_tx(struct lacp_port *lp) 1909 { 1910 int error = 0; 1911 1912 if (!(lp->lp_state & LACP_STATE_AGGREGATION) 1913 #if 1 1914 || (!(lp->lp_state & LACP_STATE_ACTIVITY) 1915 && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) 1916 #endif 1917 ) { 1918 lp->lp_flags &= ~LACP_PORT_NTT; 1919 } 1920 1921 if (!(lp->lp_flags & LACP_PORT_NTT)) { 1922 return; 1923 } 1924 1925 /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */ 1926 if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent, 1927 (3 / LACP_FAST_PERIODIC_TIME)) == 0) { 1928 LACP_DPRINTF((lp, "rate limited pdu\n")); 1929 return; 1930 } 1931 1932 if (((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_tx_test) == 0) { 1933 error = lacp_xmit_lacpdu(lp); 1934 } else { 1935 LACP_TPRINTF((lp, "Dropping TX PDU\n")); 1936 } 1937 1938 if (error == 0) { 1939 lp->lp_flags &= ~LACP_PORT_NTT; 1940 } else { 1941 LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n", 1942 error)); 1943 } 1944 } 1945 1946 static void 1947 lacp_sm_assert_ntt(struct lacp_port *lp) 1948 { 1949 1950 lp->lp_flags |= LACP_PORT_NTT; 1951 } 1952 1953 static void 1954 lacp_run_timers(struct lacp_port *lp) 1955 { 1956 int i; 1957 1958 for (i = 0; i < LACP_NTIMER; i++) { 1959 KASSERT(lp->lp_timer[i] >= 0, 1960 ("invalid timer value %d", lp->lp_timer[i])); 1961 if (lp->lp_timer[i] == 0) { 1962 continue; 1963 } else if (--lp->lp_timer[i] <= 0) { 1964 if (lacp_timer_funcs[i]) { 1965 (*lacp_timer_funcs[i])(lp); 1966 } 1967 } 1968 } 1969 } 1970 1971 int 1972 lacp_marker_input(struct lacp_port *lp, struct mbuf *m) 1973 { 1974 struct lacp_softc *lsc = lp->lp_lsc; 1975 struct lagg_port *lgp = lp->lp_lagg; 1976 struct lacp_port *lp2; 1977 struct markerdu *mdu; 1978 int error = 0; 1979 int pending = 0; 1980 1981 if (m->m_pkthdr.len != sizeof(*mdu)) { 1982 goto bad; 1983 } 1984 1985 if ((m->m_flags & M_MCAST) == 0) { 1986 goto bad; 1987 } 1988 1989 if (m->m_len < sizeof(*mdu)) { 1990 m = m_pullup(m, sizeof(*mdu)); 1991 if (m == NULL) { 1992 return (ENOMEM); 1993 } 1994 } 1995 1996 mdu = mtod(m, struct markerdu *); 1997 1998 if (memcmp(&mdu->mdu_eh.ether_dhost, 1999 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) { 2000 goto bad; 2001 } 2002 2003 if (mdu->mdu_sph.sph_version != 1) { 2004 goto bad; 2005 } 2006 2007 switch (mdu->mdu_tlv.tlv_type) { 2008 case MARKER_TYPE_INFO: 2009 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv, 2010 marker_info_tlv_template, TRUE)) { 2011 goto bad; 2012 } 2013 mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE; 2014 memcpy(&mdu->mdu_eh.ether_dhost, 2015 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN); 2016 memcpy(&mdu->mdu_eh.ether_shost, 2017 lgp->lp_lladdr, ETHER_ADDR_LEN); 2018 error = lagg_enqueue(lp->lp_ifp, m); 2019 break; 2020 2021 case MARKER_TYPE_RESPONSE: 2022 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv, 2023 marker_response_tlv_template, TRUE)) { 2024 goto bad; 2025 } 2026 LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n", 2027 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, 2028 ":", ntohl(mdu->mdu_info.mi_rq_xid))); 2029 2030 /* Verify that it is the last marker we sent out */ 2031 if (memcmp(&mdu->mdu_info, &lp->lp_marker, 2032 sizeof(struct lacp_markerinfo))) 2033 goto bad; 2034 2035 LACP_LOCK(lsc); 2036 lp->lp_flags &= ~LACP_PORT_MARK; 2037 2038 if (lsc->lsc_suppress_distributing) { 2039 /* Check if any ports are waiting for a response */ 2040 LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) { 2041 if (lp2->lp_flags & LACP_PORT_MARK) { 2042 pending = 1; 2043 break; 2044 } 2045 } 2046 2047 if (pending == 0) { 2048 /* All interface queues are clear */ 2049 LACP_DPRINTF((NULL, "queue flush complete\n")); 2050 lsc->lsc_suppress_distributing = FALSE; 2051 } 2052 } 2053 LACP_UNLOCK(lsc); 2054 m_freem(m); 2055 break; 2056 2057 default: 2058 goto bad; 2059 } 2060 2061 return (error); 2062 2063 bad: 2064 LACP_DPRINTF((lp, "bad marker frame\n")); 2065 m_freem(m); 2066 return (EINVAL); 2067 } 2068 2069 static int 2070 tlv_check(const void *p, size_t size, const struct tlvhdr *tlv, 2071 const struct tlv_template *tmpl, boolean_t check_type) 2072 { 2073 while (/* CONSTCOND */ 1) { 2074 if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) { 2075 return (EINVAL); 2076 } 2077 if ((check_type && tlv->tlv_type != tmpl->tmpl_type) || 2078 tlv->tlv_length != tmpl->tmpl_length) { 2079 return (EINVAL); 2080 } 2081 if (tmpl->tmpl_type == 0) { 2082 break; 2083 } 2084 tlv = (const struct tlvhdr *) 2085 ((const char *)tlv + tlv->tlv_length); 2086 tmpl++; 2087 } 2088 2089 return (0); 2090 } 2091 2092 /* Debugging */ 2093 const char * 2094 lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen) 2095 { 2096 snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X", 2097 (int)mac[0], 2098 (int)mac[1], 2099 (int)mac[2], 2100 (int)mac[3], 2101 (int)mac[4], 2102 (int)mac[5]); 2103 2104 return (buf); 2105 } 2106 2107 const char * 2108 lacp_format_systemid(const struct lacp_systemid *sysid, 2109 char *buf, size_t buflen) 2110 { 2111 char macbuf[LACP_MACSTR_MAX+1]; 2112 2113 snprintf(buf, buflen, "%04X,%s", 2114 ntohs(sysid->lsi_prio), 2115 lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf))); 2116 2117 return (buf); 2118 } 2119 2120 const char * 2121 lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen) 2122 { 2123 snprintf(buf, buflen, "%04X,%04X", 2124 ntohs(portid->lpi_prio), 2125 ntohs(portid->lpi_portno)); 2126 2127 return (buf); 2128 } 2129 2130 const char * 2131 lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen) 2132 { 2133 char sysid[LACP_SYSTEMIDSTR_MAX+1]; 2134 char portid[LACP_PORTIDSTR_MAX+1]; 2135 2136 snprintf(buf, buflen, "(%s,%04X,%s)", 2137 lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)), 2138 ntohs(peer->lip_key), 2139 lacp_format_portid(&peer->lip_portid, portid, sizeof(portid))); 2140 2141 return (buf); 2142 } 2143 2144 const char * 2145 lacp_format_lagid(const struct lacp_peerinfo *a, 2146 const struct lacp_peerinfo *b, char *buf, size_t buflen) 2147 { 2148 char astr[LACP_PARTNERSTR_MAX+1]; 2149 char bstr[LACP_PARTNERSTR_MAX+1]; 2150 2151 #if 0 2152 /* 2153 * there's a convention to display small numbered peer 2154 * in the left. 2155 */ 2156 2157 if (lacp_compare_peerinfo(a, b) > 0) { 2158 const struct lacp_peerinfo *t; 2159 2160 t = a; 2161 a = b; 2162 b = t; 2163 } 2164 #endif 2165 2166 snprintf(buf, buflen, "[%s,%s]", 2167 lacp_format_partner(a, astr, sizeof(astr)), 2168 lacp_format_partner(b, bstr, sizeof(bstr))); 2169 2170 return (buf); 2171 } 2172 2173 const char * 2174 lacp_format_lagid_aggregator(const struct lacp_aggregator *la, 2175 char *buf, size_t buflen) 2176 { 2177 if (la == NULL) { 2178 return ("(none)"); 2179 } 2180 2181 return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen)); 2182 } 2183 2184 const char * 2185 lacp_format_state(uint8_t state, char *buf, size_t buflen) 2186 { 2187 snprintf(buf, buflen, "%b", state, LACP_STATE_BITS); 2188 return (buf); 2189 } 2190 2191 static void 2192 lacp_dump_lacpdu(const struct lacpdu *du) 2193 { 2194 char buf[LACP_PARTNERSTR_MAX+1]; 2195 char buf2[LACP_STATESTR_MAX+1]; 2196 2197 printf("actor=%s\n", 2198 lacp_format_partner(&du->ldu_actor, buf, sizeof(buf))); 2199 printf("actor.state=%s\n", 2200 lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2))); 2201 printf("partner=%s\n", 2202 lacp_format_partner(&du->ldu_partner, buf, sizeof(buf))); 2203 printf("partner.state=%s\n", 2204 lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2))); 2205 2206 printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay)); 2207 } 2208 2209 static void 2210 lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...) 2211 { 2212 va_list va; 2213 2214 if (lp) { 2215 printf("%s: ", lp->lp_ifp->if_xname); 2216 } 2217 2218 va_start(va, fmt); 2219 vprintf(fmt, va); 2220 va_end(va); 2221 } 2222