1 /* $NetBSD: ieee8023ad_lacp.c,v 1.3 2005/12/11 12:24:54 christos Exp $ */
4 * Copyright (c)2005 YAMAMOTO Takashi,
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
33 #include <sys/callout.h>
35 #include <sys/systm.h>
36 #include <sys/malloc.h>
37 #include <sys/kernel.h> /* hz */
38 #include <sys/socket.h> /* for net/if.h */
39 #include <sys/sockio.h>
40 #include <machine/stdarg.h>
42 #include <sys/rwlock.h>
43 #include <sys/taskqueue.h>
46 #include <net/if_dl.h>
47 #include <net/ethernet.h>
48 #include <net/if_media.h>
49 #include <net/if_types.h>
51 #include <net/if_lagg.h>
52 #include <net/ieee8023ad_lacp.h>
55 * actor system priority and port priority.
56 * XXX should be configurable.
59 #define LACP_SYSTEM_PRIO 0x8000
60 #define LACP_PORT_PRIO 0x8000
62 const uint8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] =
63 { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 };
65 static const struct tlv_template lacp_info_tlv_template[] = {
66 { LACP_TYPE_ACTORINFO,
67 sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
68 { LACP_TYPE_PARTNERINFO,
69 sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
70 { LACP_TYPE_COLLECTORINFO,
71 sizeof(struct tlvhdr) + sizeof(struct lacp_collectorinfo) },
75 typedef void (*lacp_timer_func_t)(struct lacp_port *);
77 static const struct tlv_template marker_info_tlv_template[] = {
79 sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
83 static const struct tlv_template marker_response_tlv_template[] = {
84 { MARKER_TYPE_RESPONSE,
85 sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
89 static void lacp_fill_actorinfo(struct lacp_port *, struct lacp_peerinfo *);
90 static void lacp_fill_markerinfo(struct lacp_port *,
91 struct lacp_markerinfo *);
93 static uint64_t lacp_aggregator_bandwidth(struct lacp_aggregator *);
94 static void lacp_suppress_distributing(struct lacp_softc *,
95 struct lacp_aggregator *);
96 static void lacp_transit_expire(void *);
97 static void lacp_select_active_aggregator(struct lacp_softc *);
98 static uint16_t lacp_compose_key(struct lacp_port *);
99 static int tlv_check(const void *, size_t, const struct tlvhdr *,
100 const struct tlv_template *, boolean_t);
101 static void lacp_tick(void *);
103 static void lacp_fill_aggregator_id(struct lacp_aggregator *,
104 const struct lacp_port *);
105 static void lacp_fill_aggregator_id_peer(struct lacp_peerinfo *,
106 const struct lacp_peerinfo *);
107 static int lacp_aggregator_is_compatible(const struct lacp_aggregator *,
108 const struct lacp_port *);
109 static int lacp_peerinfo_is_compatible(const struct lacp_peerinfo *,
110 const struct lacp_peerinfo *);
112 static struct lacp_aggregator *lacp_aggregator_get(struct lacp_softc *,
114 static void lacp_aggregator_addref(struct lacp_softc *,
115 struct lacp_aggregator *);
116 static void lacp_aggregator_delref(struct lacp_softc *,
117 struct lacp_aggregator *);
119 /* receive machine */
121 static void lacp_dequeue(void *, int);
122 static int lacp_pdu_input(struct lagg_port *, struct mbuf *);
123 static int lacp_marker_input(struct lagg_port *, struct mbuf *);
124 static void lacp_sm_rx(struct lacp_port *, const struct lacpdu *);
125 static void lacp_sm_rx_timer(struct lacp_port *);
126 static void lacp_sm_rx_set_expired(struct lacp_port *);
127 static void lacp_sm_rx_update_ntt(struct lacp_port *,
128 const struct lacpdu *);
129 static void lacp_sm_rx_record_pdu(struct lacp_port *,
130 const struct lacpdu *);
131 static void lacp_sm_rx_update_selected(struct lacp_port *,
132 const struct lacpdu *);
133 static void lacp_sm_rx_record_default(struct lacp_port *);
134 static void lacp_sm_rx_update_default_selected(struct lacp_port *);
135 static void lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *,
136 const struct lacp_peerinfo *);
140 static void lacp_sm_mux(struct lacp_port *);
141 static void lacp_set_mux(struct lacp_port *, enum lacp_mux_state);
142 static void lacp_sm_mux_timer(struct lacp_port *);
144 /* periodic transmit machine */
146 static void lacp_sm_ptx_update_timeout(struct lacp_port *, uint8_t);
147 static void lacp_sm_ptx_tx_schedule(struct lacp_port *);
148 static void lacp_sm_ptx_timer(struct lacp_port *);
150 /* transmit machine */
152 static void lacp_sm_tx(struct lacp_port *);
153 static void lacp_sm_assert_ntt(struct lacp_port *);
155 static void lacp_run_timers(struct lacp_port *);
156 static int lacp_compare_peerinfo(const struct lacp_peerinfo *,
157 const struct lacp_peerinfo *);
158 static int lacp_compare_systemid(const struct lacp_systemid *,
159 const struct lacp_systemid *);
160 static void lacp_port_enable(struct lacp_port *);
161 static void lacp_port_disable(struct lacp_port *);
162 static void lacp_select(struct lacp_port *);
163 static void lacp_unselect(struct lacp_port *);
164 static void lacp_disable_collecting(struct lacp_port *);
165 static void lacp_enable_collecting(struct lacp_port *);
166 static void lacp_disable_distributing(struct lacp_port *);
167 static void lacp_enable_distributing(struct lacp_port *);
168 static int lacp_xmit_lacpdu(struct lacp_port *);
169 static int lacp_xmit_marker(struct lacp_port *);
171 #if defined(LACP_DEBUG)
172 static void lacp_dump_lacpdu(const struct lacpdu *);
173 static const char *lacp_format_partner(const struct lacp_peerinfo *, char *,
175 static const char *lacp_format_lagid(const struct lacp_peerinfo *,
176 const struct lacp_peerinfo *, char *, size_t);
177 static const char *lacp_format_lagid_aggregator(const struct lacp_aggregator *,
179 static const char *lacp_format_state(uint8_t, char *, size_t);
180 static const char *lacp_format_mac(const uint8_t *, char *, size_t);
181 static const char *lacp_format_systemid(const struct lacp_systemid *, char *,
183 static const char *lacp_format_portid(const struct lacp_portid *, char *,
185 static void lacp_dprintf(const struct lacp_port *, const char *, ...)
186 __attribute__((__format__(__printf__, 2, 3)));
187 #define LACP_DPRINTF(a) lacp_dprintf a
189 #define LACP_DPRINTF(a) /* nothing */
193 * partner administration variables.
194 * XXX should be configurable.
197 static const struct lacp_peerinfo lacp_partner_admin = {
198 .lip_systemid = { .lsi_prio = 0xffff },
199 .lip_portid = { .lpi_prio = 0xffff },
202 .lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION |
203 LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING,
210 static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = {
211 [LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer,
212 [LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer,
213 [LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer,
217 lacp_input(struct lagg_port *lgp, struct mbuf *m)
219 struct lagg_softc *sc = lgp->lp_softc;
220 struct lacp_softc *lsc = LACP_SOFTC(sc);
223 if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) {
228 m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype);
230 case SLOWPROTOCOLS_SUBTYPE_LACP:
231 IF_HANDOFF(&lsc->lsc_queue, m, NULL);
232 taskqueue_enqueue(taskqueue_swi, &lsc->lsc_qtask);
235 case SLOWPROTOCOLS_SUBTYPE_MARKER:
236 lacp_marker_input(lgp, m);
240 /* Unknown LACP packet type */
247 lacp_dequeue(void *arg, int pending)
249 struct lacp_softc *lsc = (struct lacp_softc *)arg;
250 struct lagg_softc *sc = lsc->lsc_softc;
251 struct lagg_port *lgp;
256 IF_DEQUEUE(&lsc->lsc_queue, m);
259 lgp = m->m_pkthdr.rcvif->if_lagg;
260 lacp_pdu_input(lgp, m);
266 * lacp_pdu_input: process lacpdu
269 lacp_pdu_input(struct lagg_port *lgp, struct mbuf *m)
271 struct lacp_port *lp = LACP_PORT(lgp);
275 LAGG_WLOCK_ASSERT(lgp->lp_softc);
277 if (m->m_pkthdr.len != sizeof(*du)) {
281 if ((m->m_flags & M_MCAST) == 0) {
285 if (m->m_len < sizeof(*du)) {
286 m = m_pullup(m, sizeof(*du));
292 du = mtod(m, struct lacpdu *);
294 if (memcmp(&du->ldu_eh.ether_dhost,
295 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
300 * ignore the version for compatibility with
301 * the future protocol revisions.
304 if (du->ldu_sph.sph_version != 1) {
310 * ignore tlv types for compatibility with
311 * the future protocol revisions.
313 if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor,
314 lacp_info_tlv_template, FALSE)) {
318 #if defined(LACP_DEBUG)
319 LACP_DPRINTF((lp, "lacpdu receive\n"));
320 lacp_dump_lacpdu(du);
321 #endif /* defined(LACP_DEBUG) */
334 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info)
336 struct lagg_port *lgp = lp->lp_lagg;
337 struct lagg_softc *sc = lgp->lp_softc;
339 info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO);
340 memcpy(&info->lip_systemid.lsi_mac,
341 IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
342 info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO);
343 info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index);
344 info->lip_state = lp->lp_state;
348 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info)
350 struct ifnet *ifp = lp->lp_ifp;
352 /* Fill in the port index and system id (encoded as the MAC) */
353 info->mi_rq_port = htons(ifp->if_index);
354 memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN);
355 info->mi_rq_xid = htonl(0);
359 lacp_xmit_lacpdu(struct lacp_port *lp)
361 struct lagg_port *lgp = lp->lp_lagg;
366 LAGG_WLOCK_ASSERT(lgp->lp_softc);
368 m = m_gethdr(M_DONTWAIT, MT_DATA);
372 m->m_len = m->m_pkthdr.len = sizeof(*du);
374 du = mtod(m, struct lacpdu *);
375 memset(du, 0, sizeof(*du));
377 memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
379 memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
380 du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW);
382 du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP;
383 du->ldu_sph.sph_version = 1;
385 TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor));
386 du->ldu_actor = lp->lp_actor;
388 TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO,
389 sizeof(du->ldu_partner));
390 du->ldu_partner = lp->lp_partner;
392 TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO,
393 sizeof(du->ldu_collector));
394 du->ldu_collector.lci_maxdelay = 0;
396 #if defined(LACP_DEBUG)
397 LACP_DPRINTF((lp, "lacpdu transmit\n"));
398 lacp_dump_lacpdu(du);
399 #endif /* defined(LACP_DEBUG) */
401 m->m_flags |= M_MCAST;
404 * XXX should use higher priority queue.
405 * otherwise network congestion can break aggregation.
408 error = lagg_enqueue(lp->lp_ifp, m);
413 lacp_xmit_marker(struct lacp_port *lp)
415 struct lagg_port *lgp = lp->lp_lagg;
417 struct markerdu *mdu;
420 LAGG_WLOCK_ASSERT(lgp->lp_softc);
422 m = m_gethdr(M_DONTWAIT, MT_DATA);
426 m->m_len = m->m_pkthdr.len = sizeof(*mdu);
428 mdu = mtod(m, struct markerdu *);
429 memset(mdu, 0, sizeof(*mdu));
431 memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
433 memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
434 mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW);
436 mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER;
437 mdu->mdu_sph.sph_version = 1;
439 /* Bump the transaction id and copy over the marker info */
440 lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1);
441 TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info));
442 mdu->mdu_info = lp->lp_marker;
444 LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n",
445 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":",
446 ntohl(mdu->mdu_info.mi_rq_xid)));
448 m->m_flags |= M_MCAST;
449 error = lagg_enqueue(lp->lp_ifp, m);
453 lacp_linkstate(struct lagg_port *lgp)
455 struct lacp_port *lp = LACP_PORT(lgp);
456 struct ifnet *ifp = lgp->lp_ifp;
457 struct ifmediareq ifmr;
463 LAGG_WLOCK_ASSERT(lgp->lp_softc);
465 bzero((char *)&ifmr, sizeof(ifmr));
466 error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr);
470 media = ifmr.ifm_active;
471 LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, "
472 "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER,
473 (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP));
474 old_state = lp->lp_state;
475 old_key = lp->lp_key;
477 lp->lp_media = media;
479 * If the port is not an active full duplex Ethernet link then it can
482 if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 ||
483 ifp->if_link_state != LINK_STATE_UP) {
484 lacp_port_disable(lp);
486 lacp_port_enable(lp);
488 lp->lp_key = lacp_compose_key(lp);
490 if (old_state != lp->lp_state || old_key != lp->lp_key) {
491 LACP_DPRINTF((lp, "-> UNSELECTED\n"));
492 lp->lp_selected = LACP_UNSELECTED;
499 struct lacp_softc *lsc = arg;
500 struct lagg_softc *sc = lsc->lsc_softc;
501 struct lacp_port *lp;
504 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
505 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0)
513 lacp_sm_ptx_tx_schedule(lp);
516 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
520 lacp_port_create(struct lagg_port *lgp)
522 struct lagg_softc *sc = lgp->lp_softc;
523 struct lacp_softc *lsc = LACP_SOFTC(sc);
524 struct lacp_port *lp;
525 struct ifnet *ifp = lgp->lp_ifp;
526 struct sockaddr_dl sdl;
527 struct ifmultiaddr *rifma = NULL;
530 boolean_t active = TRUE; /* XXX should be configurable */
531 boolean_t fast = FALSE; /* XXX should be configurable */
533 LAGG_WLOCK_ASSERT(sc);
535 bzero((char *)&sdl, sizeof(sdl));
536 sdl.sdl_len = sizeof(sdl);
537 sdl.sdl_family = AF_LINK;
538 sdl.sdl_index = ifp->if_index;
539 sdl.sdl_type = IFT_ETHER;
540 sdl.sdl_alen = ETHER_ADDR_LEN;
542 bcopy(ðermulticastaddr_slowprotocols,
543 LLADDR(&sdl), ETHER_ADDR_LEN);
544 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
546 printf("%s: ADDMULTI failed on %s\n", __func__, lgp->lp_ifname);
550 lp = malloc(sizeof(struct lacp_port),
551 M_DEVBUF, M_NOWAIT|M_ZERO);
555 lgp->lp_psc = (caddr_t)lp;
561 LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next);
563 lacp_fill_actorinfo(lp, &lp->lp_actor);
564 lacp_fill_markerinfo(lp, &lp->lp_marker);
566 (active ? LACP_STATE_ACTIVITY : 0) |
567 (fast ? LACP_STATE_TIMEOUT : 0);
568 lp->lp_aggregator = NULL;
570 lacp_sm_rx_set_expired(lp);
576 lacp_port_destroy(struct lagg_port *lgp)
578 struct lacp_port *lp = LACP_PORT(lgp);
581 LAGG_WLOCK_ASSERT(lgp->lp_softc);
583 for (i = 0; i < LACP_NTIMER; i++) {
584 LACP_TIMER_DISARM(lp, i);
587 lacp_disable_collecting(lp);
588 lacp_disable_distributing(lp);
590 lgp->lp_flags &= ~LAGG_PORT_DISABLED;
592 /* The address may have already been removed by if_purgemaddrs() */
593 if (!lgp->lp_detaching)
594 if_delmulti_ifma(lp->lp_ifma);
596 LIST_REMOVE(lp, lp_next);
601 lacp_port_isactive(struct lagg_port *lgp)
603 struct lacp_port *lp = LACP_PORT(lgp);
604 struct lacp_softc *lsc = lp->lp_lsc;
605 struct lacp_aggregator *la = lp->lp_aggregator;
607 /* This port is joined to the active aggregator */
608 if (la != NULL && la == lsc->lsc_active_aggregator)
615 lacp_disable_collecting(struct lacp_port *lp)
617 struct lagg_port *lgp = lp->lp_lagg;
619 LACP_DPRINTF((lp, "collecting disabled\n"));
621 lp->lp_state &= ~LACP_STATE_COLLECTING;
622 lgp->lp_flags &= ~LAGG_PORT_COLLECTING;
626 lacp_enable_collecting(struct lacp_port *lp)
628 struct lagg_port *lgp = lp->lp_lagg;
630 LACP_DPRINTF((lp, "collecting enabled\n"));
632 lp->lp_state |= LACP_STATE_COLLECTING;
633 lgp->lp_flags |= LAGG_PORT_COLLECTING;
637 lacp_disable_distributing(struct lacp_port *lp)
639 struct lacp_aggregator *la = lp->lp_aggregator;
640 struct lacp_softc *lsc = lp->lp_lsc;
641 struct lagg_port *lgp = lp->lp_lagg;
642 #if defined(LACP_DEBUG)
643 char buf[LACP_LAGIDSTR_MAX+1];
644 #endif /* defined(LACP_DEBUG) */
646 LAGG_WLOCK_ASSERT(lgp->lp_softc);
648 if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) {
652 KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports"));
653 KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports));
654 KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid"));
656 LACP_DPRINTF((lp, "disable distributing on aggregator %s, "
658 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
659 la->la_nports, la->la_nports - 1));
661 TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q);
664 lacp_suppress_distributing(lsc, la);
666 lp->lp_state &= ~LACP_STATE_DISTRIBUTING;
667 lgp->lp_flags &= ~LAGG_PORT_DISTRIBUTING;
669 if (lsc->lsc_active_aggregator == la) {
670 lacp_select_active_aggregator(lsc);
675 lacp_enable_distributing(struct lacp_port *lp)
677 struct lacp_aggregator *la = lp->lp_aggregator;
678 struct lacp_softc *lsc = lp->lp_lsc;
679 struct lagg_port *lgp = lp->lp_lagg;
680 #if defined(LACP_DEBUG)
681 char buf[LACP_LAGIDSTR_MAX+1];
682 #endif /* defined(LACP_DEBUG) */
684 LAGG_WLOCK_ASSERT(lgp->lp_softc);
686 if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) {
690 LACP_DPRINTF((lp, "enable distributing on aggregator %s, "
692 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
693 la->la_nports, la->la_nports + 1));
695 KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid"));
696 TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q);
699 lacp_suppress_distributing(lsc, la);
701 lp->lp_state |= LACP_STATE_DISTRIBUTING;
702 lgp->lp_flags |= LAGG_PORT_DISTRIBUTING;
704 if (lsc->lsc_active_aggregator != la) {
705 lacp_select_active_aggregator(lsc);
710 lacp_transit_expire(void *vp)
712 struct lacp_softc *lsc = vp;
714 LACP_DPRINTF((NULL, "%s\n", __func__));
715 lsc->lsc_suppress_distributing = FALSE;
719 lacp_attach(struct lagg_softc *sc)
721 struct lacp_softc *lsc;
723 LAGG_WLOCK_ASSERT(sc);
725 lsc = malloc(sizeof(struct lacp_softc),
726 M_DEVBUF, M_NOWAIT|M_ZERO);
730 sc->sc_psc = (caddr_t)lsc;
733 lsc->lsc_hashkey = arc4random();
734 lsc->lsc_active_aggregator = NULL;
735 TAILQ_INIT(&lsc->lsc_aggregators);
736 LIST_INIT(&lsc->lsc_ports);
738 TASK_INIT(&lsc->lsc_qtask, 0, lacp_dequeue, lsc);
739 mtx_init(&lsc->lsc_queue.ifq_mtx, "lacp queue", NULL, MTX_DEF);
740 lsc->lsc_queue.ifq_maxlen = ifqmaxlen;
742 callout_init(&lsc->lsc_transit_callout, CALLOUT_MPSAFE);
743 callout_init(&lsc->lsc_callout, CALLOUT_MPSAFE);
745 /* if the lagg is already up then do the same */
746 if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING)
753 lacp_detach(struct lagg_softc *sc)
755 struct lacp_softc *lsc = LACP_SOFTC(sc);
757 KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators),
758 ("aggregators still active"));
759 KASSERT(lsc->lsc_active_aggregator == NULL,
760 ("aggregator still attached"));
763 callout_drain(&lsc->lsc_transit_callout);
764 callout_drain(&lsc->lsc_callout);
765 taskqueue_drain(taskqueue_swi, &lsc->lsc_qtask);
766 IF_DRAIN(&lsc->lsc_queue);
767 mtx_destroy(&lsc->lsc_queue.ifq_mtx);
774 lacp_init(struct lagg_softc *sc)
776 struct lacp_softc *lsc = LACP_SOFTC(sc);
778 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
782 lacp_stop(struct lagg_softc *sc)
784 struct lacp_softc *lsc = LACP_SOFTC(sc);
786 callout_stop(&lsc->lsc_transit_callout);
787 callout_stop(&lsc->lsc_callout);
791 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
793 struct lacp_softc *lsc = LACP_SOFTC(sc);
794 struct lacp_aggregator *la;
795 struct lacp_port *lp;
799 LAGG_RLOCK_ASSERT(sc);
801 if (__predict_false(lsc->lsc_suppress_distributing)) {
802 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__));
806 la = lsc->lsc_active_aggregator;
807 if (__predict_false(la == NULL)) {
808 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__));
812 nports = la->la_nports;
813 KASSERT(nports > 0, ("no ports available"));
815 hash = lagg_hashmbuf(m, lsc->lsc_hashkey);
817 lp = TAILQ_FIRST(&la->la_ports);
819 lp = TAILQ_NEXT(lp, lp_dist_q);
822 KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0,
823 ("aggregated port is not distributing"));
825 return (lp->lp_lagg);
828 * lacp_suppress_distributing: drop transmit packets for a while
829 * to preserve packet ordering.
833 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la)
835 struct lacp_port *lp;
837 if (lsc->lsc_active_aggregator != la) {
841 LACP_DPRINTF((NULL, "%s\n", __func__));
842 lsc->lsc_suppress_distributing = TRUE;
844 /* send a marker frame down each port to verify the queues are empty */
845 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
846 lp->lp_flags |= LACP_PORT_MARK;
847 lacp_xmit_marker(lp);
850 /* set a timeout for the marker frames */
851 callout_reset(&lsc->lsc_transit_callout,
852 LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc);
856 lacp_compare_peerinfo(const struct lacp_peerinfo *a,
857 const struct lacp_peerinfo *b)
859 return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state)));
863 lacp_compare_systemid(const struct lacp_systemid *a,
864 const struct lacp_systemid *b)
866 return (memcmp(a, b, sizeof(*a)));
871 lacp_compare_portid(const struct lacp_portid *a,
872 const struct lacp_portid *b)
874 return (memcmp(a, b, sizeof(*a)));
879 lacp_aggregator_bandwidth(struct lacp_aggregator *la)
881 struct lacp_port *lp;
884 lp = TAILQ_FIRST(&la->la_ports);
889 speed = ifmedia_baudrate(lp->lp_media);
890 speed *= la->la_nports;
892 LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n",
893 lp->lp_media, la->la_nports));
900 * lacp_select_active_aggregator: select an aggregator to be used to transmit
901 * packets from lagg(4) interface.
905 lacp_select_active_aggregator(struct lacp_softc *lsc)
907 struct lacp_aggregator *la;
908 struct lacp_aggregator *best_la = NULL;
909 uint64_t best_speed = 0;
910 #if defined(LACP_DEBUG)
911 char buf[LACP_LAGIDSTR_MAX+1];
912 #endif /* defined(LACP_DEBUG) */
914 LACP_DPRINTF((NULL, "%s:\n", __func__));
916 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
919 if (la->la_nports == 0) {
923 speed = lacp_aggregator_bandwidth(la);
924 LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n",
925 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
926 speed, la->la_nports));
928 /* This aggregator is chosen if
929 * the partner has a better system priority
930 * or, the total aggregated speed is higher
931 * or, it is already the chosen aggregator
933 if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) <
934 LACP_SYS_PRI(best_la->la_partner)) ||
935 speed > best_speed ||
936 (speed == best_speed &&
937 la == lsc->lsc_active_aggregator)) {
943 KASSERT(best_la == NULL || best_la->la_nports > 0,
944 ("invalid aggregator refcnt"));
945 KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports),
946 ("invalid aggregator list"));
948 #if defined(LACP_DEBUG)
949 if (lsc->lsc_active_aggregator != best_la) {
950 LACP_DPRINTF((NULL, "active aggregator changed\n"));
951 LACP_DPRINTF((NULL, "old %s\n",
952 lacp_format_lagid_aggregator(lsc->lsc_active_aggregator,
955 LACP_DPRINTF((NULL, "active aggregator not changed\n"));
957 LACP_DPRINTF((NULL, "new %s\n",
958 lacp_format_lagid_aggregator(best_la, buf, sizeof(buf))));
959 #endif /* defined(LACP_DEBUG) */
961 if (lsc->lsc_active_aggregator != best_la) {
962 lsc->lsc_active_aggregator = best_la;
964 lacp_suppress_distributing(lsc, best_la);
970 lacp_compose_key(struct lacp_port *lp)
972 struct lagg_port *lgp = lp->lp_lagg;
973 struct lagg_softc *sc = lgp->lp_softc;
974 u_int media = lp->lp_media;
977 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) {
980 * non-aggregatable links should have unique keys.
982 * XXX this isn't really unique as if_index is 16 bit.
985 /* bit 0..14: (some bits of) if_index of this port */
986 key = lp->lp_ifp->if_index;
990 u_int subtype = IFM_SUBTYPE(media);
992 KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type"));
993 KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface"));
995 /* bit 0..4: IFM_SUBTYPE */
997 /* bit 5..14: (some bits of) if_index of lagg device */
998 key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5);
1001 return (htons(key));
1005 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1007 #if defined(LACP_DEBUG)
1008 char buf[LACP_LAGIDSTR_MAX+1];
1011 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1013 lacp_format_lagid(&la->la_actor, &la->la_partner,
1015 la->la_refcnt, la->la_refcnt + 1));
1017 KASSERT(la->la_refcnt > 0, ("refcount <= 0"));
1019 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount"));
1023 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1025 #if defined(LACP_DEBUG)
1026 char buf[LACP_LAGIDSTR_MAX+1];
1029 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1031 lacp_format_lagid(&la->la_actor, &la->la_partner,
1033 la->la_refcnt, la->la_refcnt - 1));
1035 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt"));
1037 if (la->la_refcnt > 0) {
1041 KASSERT(la->la_refcnt == 0, ("refcount not zero"));
1042 KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active"));
1044 TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q);
1050 * lacp_aggregator_get: allocate an aggregator.
1053 static struct lacp_aggregator *
1054 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp)
1056 struct lacp_aggregator *la;
1058 la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT);
1062 TAILQ_INIT(&la->la_ports);
1064 TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q);
1071 * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port.
1075 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp)
1077 lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner);
1078 lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor);
1080 la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION;
1084 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr,
1085 const struct lacp_peerinfo *lpi_port)
1087 memset(lpi_aggr, 0, sizeof(*lpi_aggr));
1088 lpi_aggr->lip_systemid = lpi_port->lip_systemid;
1089 lpi_aggr->lip_key = lpi_port->lip_key;
1093 * lacp_aggregator_is_compatible: check if a port can join to an aggregator.
1097 lacp_aggregator_is_compatible(const struct lacp_aggregator *la,
1098 const struct lacp_port *lp)
1100 if (!(lp->lp_state & LACP_STATE_AGGREGATION) ||
1101 !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) {
1105 if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) {
1109 if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) {
1113 if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) {
1121 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a,
1122 const struct lacp_peerinfo *b)
1124 if (memcmp(&a->lip_systemid, &b->lip_systemid,
1125 sizeof(a->lip_systemid))) {
1129 if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) {
1137 lacp_port_enable(struct lacp_port *lp)
1139 struct lagg_port *lgp = lp->lp_lagg;
1141 lp->lp_state |= LACP_STATE_AGGREGATION;
1142 lgp->lp_flags &= ~LAGG_PORT_DISABLED;
1146 lacp_port_disable(struct lacp_port *lp)
1148 struct lagg_port *lgp = lp->lp_lagg;
1150 lacp_set_mux(lp, LACP_MUX_DETACHED);
1152 lp->lp_state &= ~LACP_STATE_AGGREGATION;
1153 lp->lp_selected = LACP_UNSELECTED;
1154 lacp_sm_rx_record_default(lp);
1155 lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION;
1156 lp->lp_state &= ~LACP_STATE_EXPIRED;
1157 lgp->lp_flags |= LAGG_PORT_DISABLED;
1161 * lacp_select: select an aggregator. create one if necessary.
1164 lacp_select(struct lacp_port *lp)
1166 struct lacp_softc *lsc = lp->lp_lsc;
1167 struct lacp_aggregator *la;
1168 #if defined(LACP_DEBUG)
1169 char buf[LACP_LAGIDSTR_MAX+1];
1172 if (lp->lp_aggregator) {
1176 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1177 ("timer_wait_while still active"));
1179 LACP_DPRINTF((lp, "port lagid=%s\n",
1180 lacp_format_lagid(&lp->lp_actor, &lp->lp_partner,
1181 buf, sizeof(buf))));
1183 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
1184 if (lacp_aggregator_is_compatible(la, lp)) {
1190 la = lacp_aggregator_get(lsc, lp);
1192 LACP_DPRINTF((lp, "aggregator creation failed\n"));
1195 * will retry on the next tick.
1200 lacp_fill_aggregator_id(la, lp);
1201 LACP_DPRINTF((lp, "aggregator created\n"));
1203 LACP_DPRINTF((lp, "compatible aggregator found\n"));
1204 lacp_aggregator_addref(lsc, la);
1207 LACP_DPRINTF((lp, "aggregator lagid=%s\n",
1208 lacp_format_lagid(&la->la_actor, &la->la_partner,
1209 buf, sizeof(buf))));
1211 lp->lp_aggregator = la;
1212 lp->lp_selected = LACP_SELECTED;
1216 * lacp_unselect: finish unselect/detach process.
1220 lacp_unselect(struct lacp_port *lp)
1222 struct lacp_softc *lsc = lp->lp_lsc;
1223 struct lacp_aggregator *la = lp->lp_aggregator;
1225 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1226 ("timer_wait_while still active"));
1232 lp->lp_aggregator = NULL;
1233 lacp_aggregator_delref(lsc, la);
1239 lacp_sm_mux(struct lacp_port *lp)
1241 enum lacp_mux_state new_state;
1243 (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0;
1244 boolean_t p_collecting =
1245 (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0;
1246 enum lacp_selected selected = lp->lp_selected;
1247 struct lacp_aggregator *la;
1249 /* LACP_DPRINTF((lp, "%s: state %d\n", __func__, lp->lp_mux_state)); */
1252 la = lp->lp_aggregator;
1253 KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL,
1254 ("MUX not detached"));
1255 new_state = lp->lp_mux_state;
1256 switch (lp->lp_mux_state) {
1257 case LACP_MUX_DETACHED:
1258 if (selected != LACP_UNSELECTED) {
1259 new_state = LACP_MUX_WAITING;
1262 case LACP_MUX_WAITING:
1263 KASSERT(la->la_pending > 0 ||
1264 !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1265 ("timer_wait_while still active"));
1266 if (selected == LACP_SELECTED && la->la_pending == 0) {
1267 new_state = LACP_MUX_ATTACHED;
1268 } else if (selected == LACP_UNSELECTED) {
1269 new_state = LACP_MUX_DETACHED;
1272 case LACP_MUX_ATTACHED:
1273 if (selected == LACP_SELECTED && p_sync) {
1274 new_state = LACP_MUX_COLLECTING;
1275 } else if (selected != LACP_SELECTED) {
1276 new_state = LACP_MUX_DETACHED;
1279 case LACP_MUX_COLLECTING:
1280 if (selected == LACP_SELECTED && p_sync && p_collecting) {
1281 new_state = LACP_MUX_DISTRIBUTING;
1282 } else if (selected != LACP_SELECTED || !p_sync) {
1283 new_state = LACP_MUX_ATTACHED;
1286 case LACP_MUX_DISTRIBUTING:
1287 if (selected != LACP_SELECTED || !p_sync || !p_collecting) {
1288 new_state = LACP_MUX_COLLECTING;
1292 panic("%s: unknown state", __func__);
1295 if (lp->lp_mux_state == new_state) {
1299 lacp_set_mux(lp, new_state);
1304 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state)
1306 struct lacp_aggregator *la = lp->lp_aggregator;
1308 if (lp->lp_mux_state == new_state) {
1312 switch (new_state) {
1313 case LACP_MUX_DETACHED:
1314 lp->lp_state &= ~LACP_STATE_SYNC;
1315 lacp_disable_distributing(lp);
1316 lacp_disable_collecting(lp);
1317 lacp_sm_assert_ntt(lp);
1319 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) {
1320 KASSERT(la->la_pending > 0,
1321 ("timer_wait_while not active"));
1324 LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE);
1327 case LACP_MUX_WAITING:
1328 LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE,
1329 LACP_AGGREGATE_WAIT_TIME);
1332 case LACP_MUX_ATTACHED:
1333 lp->lp_state |= LACP_STATE_SYNC;
1334 lacp_disable_collecting(lp);
1335 lacp_sm_assert_ntt(lp);
1337 case LACP_MUX_COLLECTING:
1338 lacp_enable_collecting(lp);
1339 lacp_disable_distributing(lp);
1340 lacp_sm_assert_ntt(lp);
1342 case LACP_MUX_DISTRIBUTING:
1343 lacp_enable_distributing(lp);
1346 panic("%s: unknown state", __func__);
1349 LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state));
1351 lp->lp_mux_state = new_state;
1355 lacp_sm_mux_timer(struct lacp_port *lp)
1357 struct lacp_aggregator *la = lp->lp_aggregator;
1358 #if defined(LACP_DEBUG)
1359 char buf[LACP_LAGIDSTR_MAX+1];
1362 KASSERT(la->la_pending > 0, ("no pending event"));
1364 LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__,
1365 lacp_format_lagid(&la->la_actor, &la->la_partner,
1367 la->la_pending, la->la_pending - 1));
1372 /* periodic transmit machine */
1375 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate)
1377 if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state,
1378 LACP_STATE_TIMEOUT)) {
1382 LACP_DPRINTF((lp, "partner timeout changed\n"));
1385 * FAST_PERIODIC -> SLOW_PERIODIC
1387 * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC
1389 * let lacp_sm_ptx_tx_schedule to update timeout.
1392 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1395 * if timeout has been shortened, assert NTT.
1398 if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) {
1399 lacp_sm_assert_ntt(lp);
1404 lacp_sm_ptx_tx_schedule(struct lacp_port *lp)
1408 if (!(lp->lp_state & LACP_STATE_ACTIVITY) &&
1409 !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) {
1415 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1419 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) {
1423 timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ?
1424 LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME;
1426 LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout);
1430 lacp_sm_ptx_timer(struct lacp_port *lp)
1432 lacp_sm_assert_ntt(lp);
1436 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du)
1441 * check LACP_DISABLED first
1444 if (!(lp->lp_state & LACP_STATE_AGGREGATION)) {
1449 * check loopback condition.
1452 if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid,
1453 &lp->lp_actor.lip_systemid)) {
1458 * EXPIRED, DEFAULTED, CURRENT -> CURRENT
1461 lacp_sm_rx_update_selected(lp, du);
1462 lacp_sm_rx_update_ntt(lp, du);
1463 lacp_sm_rx_record_pdu(lp, du);
1465 timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ?
1466 LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME;
1467 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout);
1469 lp->lp_state &= ~LACP_STATE_EXPIRED;
1472 * kick transmit machine without waiting the next tick.
1479 lacp_sm_rx_set_expired(struct lacp_port *lp)
1481 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1482 lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT;
1483 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME);
1484 lp->lp_state |= LACP_STATE_EXPIRED;
1488 lacp_sm_rx_timer(struct lacp_port *lp)
1490 if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) {
1491 /* CURRENT -> EXPIRED */
1492 LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__));
1493 lacp_sm_rx_set_expired(lp);
1495 /* EXPIRED -> DEFAULTED */
1496 LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__));
1497 lacp_sm_rx_update_default_selected(lp);
1498 lacp_sm_rx_record_default(lp);
1499 lp->lp_state &= ~LACP_STATE_EXPIRED;
1504 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du)
1508 #if defined(LACP_DEBUG)
1509 char buf[LACP_STATESTR_MAX+1];
1512 /* LACP_DPRINTF((lp, "%s\n", __func__)); */
1514 oldpstate = lp->lp_partner.lip_state;
1516 active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY)
1517 || ((lp->lp_state & LACP_STATE_ACTIVITY) &&
1518 (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY));
1520 lp->lp_partner = du->ldu_actor;
1522 ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1523 LACP_STATE_AGGREGATION) &&
1524 !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner))
1525 || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) {
1528 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1531 lp->lp_state &= ~LACP_STATE_DEFAULTED;
1533 if (oldpstate != lp->lp_partner.lip_state) {
1534 LACP_DPRINTF((lp, "old pstate %s\n",
1535 lacp_format_state(oldpstate, buf, sizeof(buf))));
1536 LACP_DPRINTF((lp, "new pstate %s\n",
1537 lacp_format_state(lp->lp_partner.lip_state, buf,
1541 lacp_sm_ptx_update_timeout(lp, oldpstate);
1545 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du)
1547 /* LACP_DPRINTF((lp, "%s\n", __func__)); */
1549 if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) ||
1550 !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1551 LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) {
1552 LACP_DPRINTF((lp, "%s: assert ntt\n", __func__));
1553 lacp_sm_assert_ntt(lp);
1558 lacp_sm_rx_record_default(struct lacp_port *lp)
1562 /* LACP_DPRINTF((lp, "%s\n", __func__)); */
1564 oldpstate = lp->lp_partner.lip_state;
1565 lp->lp_partner = lacp_partner_admin;
1566 lp->lp_state |= LACP_STATE_DEFAULTED;
1567 lacp_sm_ptx_update_timeout(lp, oldpstate);
1571 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp,
1572 const struct lacp_peerinfo *info)
1574 /* LACP_DPRINTF((lp, "%s\n", __func__)); */
1576 if (lacp_compare_peerinfo(&lp->lp_partner, info) ||
1577 !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state,
1578 LACP_STATE_AGGREGATION)) {
1579 lp->lp_selected = LACP_UNSELECTED;
1580 /* mux machine will clean up lp->lp_aggregator */
1585 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du)
1587 /* LACP_DPRINTF((lp, "%s\n", __func__)); */
1589 lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor);
1593 lacp_sm_rx_update_default_selected(struct lacp_port *lp)
1595 /* LACP_DPRINTF((lp, "%s\n", __func__)); */
1597 lacp_sm_rx_update_selected_from_peerinfo(lp, &lacp_partner_admin);
1600 /* transmit machine */
1603 lacp_sm_tx(struct lacp_port *lp)
1607 if (!(lp->lp_state & LACP_STATE_AGGREGATION)
1609 || (!(lp->lp_state & LACP_STATE_ACTIVITY)
1610 && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY))
1613 lp->lp_flags &= ~LACP_PORT_NTT;
1616 if (!(lp->lp_flags & LACP_PORT_NTT)) {
1620 /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */
1621 if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent,
1622 (3 / LACP_FAST_PERIODIC_TIME)) == 0) {
1623 LACP_DPRINTF((lp, "rate limited pdu\n"));
1627 error = lacp_xmit_lacpdu(lp);
1630 lp->lp_flags &= ~LACP_PORT_NTT;
1632 LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n",
1638 lacp_sm_assert_ntt(struct lacp_port *lp)
1641 lp->lp_flags |= LACP_PORT_NTT;
1645 lacp_run_timers(struct lacp_port *lp)
1649 for (i = 0; i < LACP_NTIMER; i++) {
1650 KASSERT(lp->lp_timer[i] >= 0,
1651 ("invalid timer value %d", lp->lp_timer[i]));
1652 if (lp->lp_timer[i] == 0) {
1654 } else if (--lp->lp_timer[i] <= 0) {
1655 if (lacp_timer_funcs[i]) {
1656 (*lacp_timer_funcs[i])(lp);
1663 lacp_marker_input(struct lagg_port *lgp, struct mbuf *m)
1665 struct lacp_port *lp = LACP_PORT(lgp);
1666 struct lacp_port *lp2;
1667 struct lacp_softc *lsc = lp->lp_lsc;
1668 struct markerdu *mdu;
1672 LAGG_RLOCK_ASSERT(lgp->lp_softc);
1674 if (m->m_pkthdr.len != sizeof(*mdu)) {
1678 if ((m->m_flags & M_MCAST) == 0) {
1682 if (m->m_len < sizeof(*mdu)) {
1683 m = m_pullup(m, sizeof(*mdu));
1689 mdu = mtod(m, struct markerdu *);
1691 if (memcmp(&mdu->mdu_eh.ether_dhost,
1692 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
1696 if (mdu->mdu_sph.sph_version != 1) {
1700 switch (mdu->mdu_tlv.tlv_type) {
1701 case MARKER_TYPE_INFO:
1702 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
1703 marker_info_tlv_template, TRUE)) {
1706 mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE;
1707 memcpy(&mdu->mdu_eh.ether_dhost,
1708 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN);
1709 memcpy(&mdu->mdu_eh.ether_shost,
1710 lgp->lp_lladdr, ETHER_ADDR_LEN);
1711 error = lagg_enqueue(lp->lp_ifp, m);
1714 case MARKER_TYPE_RESPONSE:
1715 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
1716 marker_response_tlv_template, TRUE)) {
1719 LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n",
1720 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system,
1721 ":", ntohl(mdu->mdu_info.mi_rq_xid)));
1723 /* Verify that it is the last marker we sent out */
1724 if (memcmp(&mdu->mdu_info, &lp->lp_marker,
1725 sizeof(struct lacp_markerinfo)))
1728 lp->lp_flags &= ~LACP_PORT_MARK;
1730 if (lsc->lsc_suppress_distributing) {
1731 /* Check if any ports are waiting for a response */
1732 LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) {
1733 if (lp2->lp_flags & LACP_PORT_MARK) {
1740 /* All interface queues are clear */
1741 LACP_DPRINTF((NULL, "queue flush complete\n"));
1742 lsc->lsc_suppress_distributing = FALSE;
1756 LACP_DPRINTF((lp, "bad marker frame\n"));
1762 tlv_check(const void *p, size_t size, const struct tlvhdr *tlv,
1763 const struct tlv_template *tmpl, boolean_t check_type)
1765 while (/* CONSTCOND */ 1) {
1766 if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) {
1769 if ((check_type && tlv->tlv_type != tmpl->tmpl_type) ||
1770 tlv->tlv_length != tmpl->tmpl_length) {
1773 if (tmpl->tmpl_type == 0) {
1776 tlv = (const struct tlvhdr *)
1777 ((const char *)tlv + tlv->tlv_length);
1784 #if defined(LACP_DEBUG)
1786 lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen)
1788 snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X",
1800 lacp_format_systemid(const struct lacp_systemid *sysid,
1801 char *buf, size_t buflen)
1803 char macbuf[LACP_MACSTR_MAX+1];
1805 snprintf(buf, buflen, "%04X,%s",
1806 ntohs(sysid->lsi_prio),
1807 lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf)));
1813 lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen)
1815 snprintf(buf, buflen, "%04X,%04X",
1816 ntohs(portid->lpi_prio),
1817 ntohs(portid->lpi_portno));
1823 lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen)
1825 char sysid[LACP_SYSTEMIDSTR_MAX+1];
1826 char portid[LACP_PORTIDSTR_MAX+1];
1828 snprintf(buf, buflen, "(%s,%04X,%s)",
1829 lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)),
1830 ntohs(peer->lip_key),
1831 lacp_format_portid(&peer->lip_portid, portid, sizeof(portid)));
1837 lacp_format_lagid(const struct lacp_peerinfo *a,
1838 const struct lacp_peerinfo *b, char *buf, size_t buflen)
1840 char astr[LACP_PARTNERSTR_MAX+1];
1841 char bstr[LACP_PARTNERSTR_MAX+1];
1845 * there's a convention to display small numbered peer
1849 if (lacp_compare_peerinfo(a, b) > 0) {
1850 const struct lacp_peerinfo *t;
1858 snprintf(buf, buflen, "[%s,%s]",
1859 lacp_format_partner(a, astr, sizeof(astr)),
1860 lacp_format_partner(b, bstr, sizeof(bstr)));
1866 lacp_format_lagid_aggregator(const struct lacp_aggregator *la,
1867 char *buf, size_t buflen)
1873 return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen));
1877 lacp_format_state(uint8_t state, char *buf, size_t buflen)
1879 snprintf(buf, buflen, "%b", state, LACP_STATE_BITS);
1884 lacp_dump_lacpdu(const struct lacpdu *du)
1886 char buf[LACP_PARTNERSTR_MAX+1];
1887 char buf2[LACP_STATESTR_MAX+1];
1889 printf("actor=%s\n",
1890 lacp_format_partner(&du->ldu_actor, buf, sizeof(buf)));
1891 printf("actor.state=%s\n",
1892 lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2)));
1893 printf("partner=%s\n",
1894 lacp_format_partner(&du->ldu_partner, buf, sizeof(buf)));
1895 printf("partner.state=%s\n",
1896 lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2)));
1898 printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay));
1902 lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...)
1907 printf("%s: ", lp->lp_ifp->if_xname);