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 *lgs = lgp->lp_lagg;
220 struct lacp_softc *lsc = LACP_SOFTC(lgs);
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_lagg;
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_lagg);
277 if (__predict_false(lp->lp_flags & LACP_PORT_DETACHING)) {
281 if (m->m_pkthdr.len != sizeof(*du)) {
285 if ((m->m_flags & M_MCAST) == 0) {
289 if (m->m_len < sizeof(*du)) {
290 m = m_pullup(m, sizeof(*du));
296 du = mtod(m, struct lacpdu *);
298 if (memcmp(&du->ldu_eh.ether_dhost,
299 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
304 KASSERT(du->ldu_sph.sph_subtype == SLOWPROTOCOLS_SUBTYPE_LACP,
305 ("a very bad kassert!"));
309 * ignore the version for compatibility with
310 * the future protocol revisions.
314 if (du->ldu_sph.sph_version != 1) {
320 * ignore tlv types for compatibility with
321 * the future protocol revisions.
324 if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor,
325 lacp_info_tlv_template, FALSE)) {
329 #if defined(LACP_DEBUG)
330 LACP_DPRINTF((lp, "lacpdu receive\n"));
331 lacp_dump_lacpdu(du);
332 #endif /* defined(LACP_DEBUG) */
345 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info)
347 struct lagg_port *lgp = lp->lp_lagg;
348 struct lagg_softc *lgs = lgp->lp_lagg;
350 info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO);
351 memcpy(&info->lip_systemid.lsi_mac,
352 IF_LLADDR(lgs->sc_ifp), ETHER_ADDR_LEN);
353 info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO);
354 info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index);
355 info->lip_state = lp->lp_state;
359 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info)
361 struct ifnet *ifp = lp->lp_ifp;
363 /* Fill in the port index and system id (encoded as the MAC) */
364 info->mi_rq_port = htons(ifp->if_index);
365 memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN);
366 info->mi_rq_xid = htonl(0);
370 lacp_xmit_lacpdu(struct lacp_port *lp)
372 struct lagg_port *lgp = lp->lp_lagg;
377 LAGG_WLOCK_ASSERT(lgp->lp_lagg);
379 m = m_gethdr(M_DONTWAIT, MT_DATA);
383 m->m_len = m->m_pkthdr.len = sizeof(*du);
385 du = mtod(m, struct lacpdu *);
386 memset(du, 0, sizeof(*du));
388 memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
390 memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
391 du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW);
393 du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP;
394 du->ldu_sph.sph_version = 1;
396 TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor));
397 du->ldu_actor = lp->lp_actor;
399 TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO,
400 sizeof(du->ldu_partner));
401 du->ldu_partner = lp->lp_partner;
403 TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO,
404 sizeof(du->ldu_collector));
405 du->ldu_collector.lci_maxdelay = 0;
407 #if defined(LACP_DEBUG)
408 LACP_DPRINTF((lp, "lacpdu transmit\n"));
409 lacp_dump_lacpdu(du);
410 #endif /* defined(LACP_DEBUG) */
412 m->m_flags |= M_MCAST;
415 * XXX should use higher priority queue.
416 * otherwise network congestion can break aggregation.
419 error = lagg_enqueue(lp->lp_ifp, m);
424 lacp_xmit_marker(struct lacp_port *lp)
426 struct lagg_port *lgp = lp->lp_lagg;
428 struct markerdu *mdu;
431 LAGG_WLOCK_ASSERT(lgp->lp_lagg);
433 m = m_gethdr(M_DONTWAIT, MT_DATA);
437 m->m_len = m->m_pkthdr.len = sizeof(*mdu);
439 mdu = mtod(m, struct markerdu *);
440 memset(mdu, 0, sizeof(*mdu));
442 memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
444 memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
445 mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW);
447 mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER;
448 mdu->mdu_sph.sph_version = 1;
450 /* Bump the transaction id and copy over the marker info */
451 lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1);
452 TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info));
453 mdu->mdu_info = lp->lp_marker;
455 LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n",
456 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":",
457 ntohl(mdu->mdu_info.mi_rq_xid)));
459 m->m_flags |= M_MCAST;
460 error = lagg_enqueue(lp->lp_ifp, m);
464 lacp_linkstate(struct lagg_port *lgp)
466 struct lacp_port *lp = LACP_PORT(lgp);
467 struct ifnet *ifp = lgp->lp_ifp;
468 struct ifmediareq ifmr;
474 LAGG_WLOCK_ASSERT(lgp->lp_lagg);
476 bzero((char *)&ifmr, sizeof(ifmr));
477 error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr);
481 media = ifmr.ifm_active;
482 LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, "
483 "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER,
484 (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP));
485 old_state = lp->lp_state;
486 old_key = lp->lp_key;
488 lp->lp_media = media;
490 * If the port is not an active full duplex Ethernet link then it can
493 if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 ||
494 ifp->if_link_state != LINK_STATE_UP) {
495 lacp_port_disable(lp);
497 lacp_port_enable(lp);
499 lp->lp_key = lacp_compose_key(lp);
501 if (old_state != lp->lp_state || old_key != lp->lp_key) {
502 LACP_DPRINTF((lp, "-> UNSELECTED\n"));
503 lp->lp_selected = LACP_UNSELECTED;
510 struct lacp_softc *lsc = arg;
511 struct lagg_softc *sc = lsc->lsc_lagg;
512 struct lacp_port *lp;
515 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
516 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0)
524 lacp_sm_ptx_tx_schedule(lp);
527 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
531 lacp_port_create(struct lagg_port *lgp)
533 struct lagg_softc *lgs = lgp->lp_lagg;
534 struct lacp_softc *lsc = LACP_SOFTC(lgs);
535 struct lacp_port *lp;
536 struct ifnet *ifp = lgp->lp_ifp;
537 struct sockaddr_dl sdl;
538 struct ifmultiaddr *rifma = NULL;
541 boolean_t active = TRUE; /* XXX should be configurable */
542 boolean_t fast = FALSE; /* XXX should be configurable */
544 LAGG_WLOCK_ASSERT(lgs);
546 bzero((char *)&sdl, sizeof(sdl));
547 sdl.sdl_len = sizeof(sdl);
548 sdl.sdl_family = AF_LINK;
549 sdl.sdl_index = ifp->if_index;
550 sdl.sdl_type = IFT_ETHER;
551 sdl.sdl_alen = ETHER_ADDR_LEN;
553 bcopy(ðermulticastaddr_slowprotocols,
554 LLADDR(&sdl), ETHER_ADDR_LEN);
555 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
557 printf("%s: ADDMULTI failed on %s\n", __func__, lgp->lp_ifname);
561 lp = malloc(sizeof(struct lacp_port),
562 M_DEVBUF, M_NOWAIT|M_ZERO);
566 lgp->lp_psc = (caddr_t)lp;
572 LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next);
574 lacp_fill_actorinfo(lp, &lp->lp_actor);
575 lacp_fill_markerinfo(lp, &lp->lp_marker);
577 (active ? LACP_STATE_ACTIVITY : 0) |
578 (fast ? LACP_STATE_TIMEOUT : 0);
579 lp->lp_aggregator = NULL;
581 lacp_sm_rx_set_expired(lp);
587 lacp_port_destroy(struct lagg_port *lgp)
589 struct lacp_port *lp = LACP_PORT(lgp);
592 LAGG_WLOCK_ASSERT(lgp->lp_lagg);
594 for (i = 0; i < LACP_NTIMER; i++) {
595 LACP_TIMER_DISARM(lp, i);
598 lacp_disable_collecting(lp);
599 lacp_disable_distributing(lp);
601 lgp->lp_flags &= ~LAGG_PORT_DISABLED;
603 /* The address may have already been removed by if_purgemaddrs() */
604 if (!lgp->lp_detaching)
605 if_delmulti_ifma(lp->lp_ifma);
607 LIST_REMOVE(lp, lp_next);
612 lacp_port_isactive(struct lagg_port *lgp)
614 struct lacp_port *lp = LACP_PORT(lgp);
615 struct lacp_softc *lsc = lp->lp_lsc;
616 struct lacp_aggregator *la = lp->lp_aggregator;
618 /* This port is joined to the active aggregator */
619 if (la != NULL && la == lsc->lsc_active_aggregator)
626 lacp_disable_collecting(struct lacp_port *lp)
628 struct lagg_port *lgp = lp->lp_lagg;
630 LACP_DPRINTF((lp, "collecting disabled\n"));
632 lp->lp_state &= ~LACP_STATE_COLLECTING;
633 lgp->lp_flags &= ~LAGG_PORT_COLLECTING;
637 lacp_enable_collecting(struct lacp_port *lp)
639 struct lagg_port *lgp = lp->lp_lagg;
641 LACP_DPRINTF((lp, "collecting enabled\n"));
643 lp->lp_state |= LACP_STATE_COLLECTING;
644 lgp->lp_flags |= LAGG_PORT_COLLECTING;
648 lacp_disable_distributing(struct lacp_port *lp)
650 struct lacp_aggregator *la = lp->lp_aggregator;
651 struct lacp_softc *lsc = lp->lp_lsc;
652 struct lagg_port *lgp = lp->lp_lagg;
653 #if defined(LACP_DEBUG)
654 char buf[LACP_LAGIDSTR_MAX+1];
655 #endif /* defined(LACP_DEBUG) */
657 LAGG_WLOCK_ASSERT(lgp->lp_lagg);
659 if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) {
663 KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports"));
664 KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports));
665 KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid"));
667 LACP_DPRINTF((lp, "disable distributing on aggregator %s, "
669 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
670 la->la_nports, la->la_nports - 1));
672 TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q);
675 lacp_suppress_distributing(lsc, la);
677 lp->lp_state &= ~LACP_STATE_DISTRIBUTING;
678 lgp->lp_flags &= ~LAGG_PORT_DISTRIBUTING;
680 if (lsc->lsc_active_aggregator == la) {
681 lacp_select_active_aggregator(lsc);
686 lacp_enable_distributing(struct lacp_port *lp)
688 struct lacp_aggregator *la = lp->lp_aggregator;
689 struct lacp_softc *lsc = lp->lp_lsc;
690 struct lagg_port *lgp = lp->lp_lagg;
691 #if defined(LACP_DEBUG)
692 char buf[LACP_LAGIDSTR_MAX+1];
693 #endif /* defined(LACP_DEBUG) */
695 LAGG_WLOCK_ASSERT(lgp->lp_lagg);
697 if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) {
701 LACP_DPRINTF((lp, "enable distributing on aggregator %s, "
703 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
704 la->la_nports, la->la_nports + 1));
706 KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid"));
707 TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q);
710 lacp_suppress_distributing(lsc, la);
712 lp->lp_state |= LACP_STATE_DISTRIBUTING;
713 lgp->lp_flags |= LAGG_PORT_DISTRIBUTING;
715 if (lsc->lsc_active_aggregator != la) {
716 lacp_select_active_aggregator(lsc);
721 lacp_transit_expire(void *vp)
723 struct lacp_softc *lsc = vp;
725 LACP_DPRINTF((NULL, "%s\n", __func__));
726 lsc->lsc_suppress_distributing = FALSE;
730 lacp_attach(struct lagg_softc *lgs)
732 struct lacp_softc *lsc;
734 LAGG_WLOCK_ASSERT(lgs);
736 lsc = malloc(sizeof(struct lacp_softc),
737 M_DEVBUF, M_NOWAIT|M_ZERO);
741 lgs->sc_psc = (caddr_t)lsc;
744 lsc->lsc_hashkey = arc4random();
745 lsc->lsc_active_aggregator = NULL;
746 TAILQ_INIT(&lsc->lsc_aggregators);
747 LIST_INIT(&lsc->lsc_ports);
749 TASK_INIT(&lsc->lsc_qtask, 0, lacp_dequeue, lsc);
750 mtx_init(&lsc->lsc_queue.ifq_mtx, "lacp queue", NULL, MTX_DEF);
751 lsc->lsc_queue.ifq_maxlen = ifqmaxlen;
753 callout_init(&lsc->lsc_transit_callout, CALLOUT_MPSAFE);
754 callout_init(&lsc->lsc_callout, CALLOUT_MPSAFE);
756 /* if the lagg is already up then do the same */
757 if (lgs->sc_ifp->if_drv_flags & IFF_DRV_RUNNING)
764 lacp_detach(struct lagg_softc *lgs)
766 struct lacp_softc *lsc = LACP_SOFTC(lgs);
768 KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators),
769 ("aggregators still active"));
770 KASSERT(lsc->lsc_active_aggregator == NULL,
771 ("aggregator still attached"));
774 callout_drain(&lsc->lsc_transit_callout);
775 callout_drain(&lsc->lsc_callout);
776 taskqueue_drain(taskqueue_swi, &lsc->lsc_qtask);
777 IF_DRAIN(&lsc->lsc_queue);
778 mtx_destroy(&lsc->lsc_queue.ifq_mtx);
785 lacp_init(struct lagg_softc *lgs)
787 struct lacp_softc *lsc = LACP_SOFTC(lgs);
789 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
793 lacp_stop(struct lagg_softc *lgs)
795 struct lacp_softc *lsc = LACP_SOFTC(lgs);
797 callout_stop(&lsc->lsc_transit_callout);
798 callout_stop(&lsc->lsc_callout);
802 lacp_select_tx_port(struct lagg_softc *lgs, struct mbuf *m)
804 struct lacp_softc *lsc = LACP_SOFTC(lgs);
805 struct lacp_aggregator *la;
806 struct lacp_port *lp;
810 LAGG_RLOCK_ASSERT(lgs);
812 if (__predict_false(lsc->lsc_suppress_distributing)) {
813 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__));
817 la = lsc->lsc_active_aggregator;
818 if (__predict_false(la == NULL)) {
819 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__));
823 nports = la->la_nports;
824 KASSERT(nports > 0, ("no ports available"));
826 hash = lagg_hashmbuf(m, lsc->lsc_hashkey);
828 lp = TAILQ_FIRST(&la->la_ports);
830 lp = TAILQ_NEXT(lp, lp_dist_q);
833 KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0,
834 ("aggregated port is not distributing"));
836 return (lp->lp_lagg);
839 * lacp_suppress_distributing: drop transmit packets for a while
840 * to preserve packet ordering.
844 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la)
846 struct lacp_port *lp;
848 if (lsc->lsc_active_aggregator != la) {
852 LACP_DPRINTF((NULL, "%s\n", __func__));
853 lsc->lsc_suppress_distributing = TRUE;
855 /* send a marker frame down each port to verify the queues are empty */
856 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
857 lp->lp_flags |= LACP_PORT_MARK;
858 lacp_xmit_marker(lp);
861 /* set a timeout for the marker frames */
862 callout_reset(&lsc->lsc_transit_callout,
863 LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc);
867 lacp_compare_peerinfo(const struct lacp_peerinfo *a,
868 const struct lacp_peerinfo *b)
870 return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state)));
874 lacp_compare_systemid(const struct lacp_systemid *a,
875 const struct lacp_systemid *b)
877 return (memcmp(a, b, sizeof(*a)));
882 lacp_compare_portid(const struct lacp_portid *a,
883 const struct lacp_portid *b)
885 return (memcmp(a, b, sizeof(*a)));
890 lacp_aggregator_bandwidth(struct lacp_aggregator *la)
892 struct lacp_port *lp;
895 lp = TAILQ_FIRST(&la->la_ports);
900 speed = ifmedia_baudrate(lp->lp_media);
901 speed *= la->la_nports;
903 LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n",
904 lp->lp_media, la->la_nports));
911 * lacp_select_active_aggregator: select an aggregator to be used to transmit
912 * packets from lagg(4) interface.
916 lacp_select_active_aggregator(struct lacp_softc *lsc)
918 struct lacp_aggregator *la;
919 struct lacp_aggregator *best_la = NULL;
920 uint64_t best_speed = 0;
921 #if defined(LACP_DEBUG)
922 char buf[LACP_LAGIDSTR_MAX+1];
923 #endif /* defined(LACP_DEBUG) */
925 LACP_DPRINTF((NULL, "%s:\n", __func__));
927 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
930 if (la->la_nports == 0) {
934 speed = lacp_aggregator_bandwidth(la);
935 LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n",
936 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
937 speed, la->la_nports));
939 /* This aggregator is chosen if
940 * the partner has a better system priority
941 * or, the total aggregated speed is higher
942 * or, it is already the chosen aggregator
944 if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) <
945 LACP_SYS_PRI(best_la->la_partner)) ||
946 speed > best_speed ||
947 (speed == best_speed &&
948 la == lsc->lsc_active_aggregator)) {
954 KASSERT(best_la == NULL || best_la->la_nports > 0,
955 ("invalid aggregator refcnt"));
956 KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports),
957 ("invalid aggregator list"));
959 #if defined(LACP_DEBUG)
960 if (lsc->lsc_active_aggregator != best_la) {
961 LACP_DPRINTF((NULL, "active aggregator changed\n"));
962 LACP_DPRINTF((NULL, "old %s\n",
963 lacp_format_lagid_aggregator(lsc->lsc_active_aggregator,
966 LACP_DPRINTF((NULL, "active aggregator not changed\n"));
968 LACP_DPRINTF((NULL, "new %s\n",
969 lacp_format_lagid_aggregator(best_la, buf, sizeof(buf))));
970 #endif /* defined(LACP_DEBUG) */
972 if (lsc->lsc_active_aggregator != best_la) {
973 lsc->lsc_active_aggregator = best_la;
975 lacp_suppress_distributing(lsc, best_la);
981 lacp_compose_key(struct lacp_port *lp)
983 struct lagg_port *lgp = lp->lp_lagg;
984 struct lagg_softc *lgs = lgp->lp_lagg;
985 u_int media = lp->lp_media;
988 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) {
991 * non-aggregatable links should have unique keys.
993 * XXX this isn't really unique as if_index is 16 bit.
996 /* bit 0..14: (some bits of) if_index of this port */
997 key = lp->lp_ifp->if_index;
1001 u_int subtype = IFM_SUBTYPE(media);
1003 KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type"));
1004 KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface"));
1006 /* bit 0..4: IFM_SUBTYPE */
1008 /* bit 5..14: (some bits of) if_index of lagg device */
1009 key |= 0x7fe0 & ((lgs->sc_ifp->if_index) << 5);
1012 return (htons(key));
1016 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1018 #if defined(LACP_DEBUG)
1019 char buf[LACP_LAGIDSTR_MAX+1];
1022 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1024 lacp_format_lagid(&la->la_actor, &la->la_partner,
1026 la->la_refcnt, la->la_refcnt + 1));
1028 KASSERT(la->la_refcnt > 0, ("refcount <= 0"));
1030 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount"));
1034 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1036 #if defined(LACP_DEBUG)
1037 char buf[LACP_LAGIDSTR_MAX+1];
1040 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1042 lacp_format_lagid(&la->la_actor, &la->la_partner,
1044 la->la_refcnt, la->la_refcnt - 1));
1046 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt"));
1048 if (la->la_refcnt > 0) {
1052 KASSERT(la->la_refcnt == 0, ("refcount not zero"));
1053 KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active"));
1055 TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q);
1061 * lacp_aggregator_get: allocate an aggregator.
1064 static struct lacp_aggregator *
1065 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp)
1067 struct lacp_aggregator *la;
1069 la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT);
1073 TAILQ_INIT(&la->la_ports);
1075 TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q);
1082 * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port.
1086 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp)
1088 lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner);
1089 lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor);
1091 la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION;
1095 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr,
1096 const struct lacp_peerinfo *lpi_port)
1098 memset(lpi_aggr, 0, sizeof(*lpi_aggr));
1099 lpi_aggr->lip_systemid = lpi_port->lip_systemid;
1100 lpi_aggr->lip_key = lpi_port->lip_key;
1104 * lacp_aggregator_is_compatible: check if a port can join to an aggregator.
1108 lacp_aggregator_is_compatible(const struct lacp_aggregator *la,
1109 const struct lacp_port *lp)
1111 if (!(lp->lp_state & LACP_STATE_AGGREGATION) ||
1112 !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) {
1116 if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) {
1120 if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) {
1124 if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) {
1132 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a,
1133 const struct lacp_peerinfo *b)
1135 if (memcmp(&a->lip_systemid, &b->lip_systemid,
1136 sizeof(a->lip_systemid))) {
1140 if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) {
1148 lacp_port_enable(struct lacp_port *lp)
1150 struct lagg_port *lgp = lp->lp_lagg;
1152 lp->lp_state |= LACP_STATE_AGGREGATION;
1153 lgp->lp_flags &= ~LAGG_PORT_DISABLED;
1157 lacp_port_disable(struct lacp_port *lp)
1159 struct lagg_port *lgp = lp->lp_lagg;
1161 lacp_set_mux(lp, LACP_MUX_DETACHED);
1163 lp->lp_state &= ~LACP_STATE_AGGREGATION;
1164 lp->lp_selected = LACP_UNSELECTED;
1165 lacp_sm_rx_record_default(lp);
1166 lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION;
1167 lp->lp_state &= ~LACP_STATE_EXPIRED;
1168 lgp->lp_flags |= LAGG_PORT_DISABLED;
1172 * lacp_select: select an aggregator. create one if necessary.
1175 lacp_select(struct lacp_port *lp)
1177 struct lacp_softc *lsc = lp->lp_lsc;
1178 struct lacp_aggregator *la;
1179 #if defined(LACP_DEBUG)
1180 char buf[LACP_LAGIDSTR_MAX+1];
1183 if (lp->lp_aggregator) {
1187 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1188 ("timer_wait_while still active"));
1190 LACP_DPRINTF((lp, "port lagid=%s\n",
1191 lacp_format_lagid(&lp->lp_actor, &lp->lp_partner,
1192 buf, sizeof(buf))));
1194 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
1195 if (lacp_aggregator_is_compatible(la, lp)) {
1201 la = lacp_aggregator_get(lsc, lp);
1203 LACP_DPRINTF((lp, "aggregator creation failed\n"));
1206 * will retry on the next tick.
1211 lacp_fill_aggregator_id(la, lp);
1212 LACP_DPRINTF((lp, "aggregator created\n"));
1214 LACP_DPRINTF((lp, "compatible aggregator found\n"));
1215 lacp_aggregator_addref(lsc, la);
1218 LACP_DPRINTF((lp, "aggregator lagid=%s\n",
1219 lacp_format_lagid(&la->la_actor, &la->la_partner,
1220 buf, sizeof(buf))));
1222 lp->lp_aggregator = la;
1223 lp->lp_selected = LACP_SELECTED;
1227 * lacp_unselect: finish unselect/detach process.
1231 lacp_unselect(struct lacp_port *lp)
1233 struct lacp_softc *lsc = lp->lp_lsc;
1234 struct lacp_aggregator *la = lp->lp_aggregator;
1236 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1237 ("timer_wait_while still active"));
1243 lp->lp_aggregator = NULL;
1244 lacp_aggregator_delref(lsc, la);
1250 lacp_sm_mux(struct lacp_port *lp)
1252 enum lacp_mux_state new_state;
1254 (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0;
1255 boolean_t p_collecting =
1256 (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0;
1257 enum lacp_selected selected = lp->lp_selected;
1258 struct lacp_aggregator *la;
1260 /* LACP_DPRINTF((lp, "%s: state %d\n", __func__, lp->lp_mux_state)); */
1263 la = lp->lp_aggregator;
1264 KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL,
1265 ("MUX not detached"));
1266 new_state = lp->lp_mux_state;
1267 switch (lp->lp_mux_state) {
1268 case LACP_MUX_DETACHED:
1269 if (selected != LACP_UNSELECTED) {
1270 new_state = LACP_MUX_WAITING;
1273 case LACP_MUX_WAITING:
1274 KASSERT(la->la_pending > 0 ||
1275 !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1276 ("timer_wait_while still active"));
1277 if (selected == LACP_SELECTED && la->la_pending == 0) {
1278 new_state = LACP_MUX_ATTACHED;
1279 } else if (selected == LACP_UNSELECTED) {
1280 new_state = LACP_MUX_DETACHED;
1283 case LACP_MUX_ATTACHED:
1284 if (selected == LACP_SELECTED && p_sync) {
1285 new_state = LACP_MUX_COLLECTING;
1286 } else if (selected != LACP_SELECTED) {
1287 new_state = LACP_MUX_DETACHED;
1290 case LACP_MUX_COLLECTING:
1291 if (selected == LACP_SELECTED && p_sync && p_collecting) {
1292 new_state = LACP_MUX_DISTRIBUTING;
1293 } else if (selected != LACP_SELECTED || !p_sync) {
1294 new_state = LACP_MUX_ATTACHED;
1297 case LACP_MUX_DISTRIBUTING:
1298 if (selected != LACP_SELECTED || !p_sync || !p_collecting) {
1299 new_state = LACP_MUX_COLLECTING;
1303 panic("%s: unknown state", __func__);
1306 if (lp->lp_mux_state == new_state) {
1310 lacp_set_mux(lp, new_state);
1315 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state)
1317 struct lacp_aggregator *la = lp->lp_aggregator;
1319 if (lp->lp_mux_state == new_state) {
1323 switch (new_state) {
1324 case LACP_MUX_DETACHED:
1325 lp->lp_state &= ~LACP_STATE_SYNC;
1326 lacp_disable_distributing(lp);
1327 lacp_disable_collecting(lp);
1328 lacp_sm_assert_ntt(lp);
1330 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) {
1331 KASSERT(la->la_pending > 0,
1332 ("timer_wait_while not active"));
1335 LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE);
1338 case LACP_MUX_WAITING:
1339 LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE,
1340 LACP_AGGREGATE_WAIT_TIME);
1343 case LACP_MUX_ATTACHED:
1344 lp->lp_state |= LACP_STATE_SYNC;
1345 lacp_disable_collecting(lp);
1346 lacp_sm_assert_ntt(lp);
1348 case LACP_MUX_COLLECTING:
1349 lacp_enable_collecting(lp);
1350 lacp_disable_distributing(lp);
1351 lacp_sm_assert_ntt(lp);
1353 case LACP_MUX_DISTRIBUTING:
1354 lacp_enable_distributing(lp);
1357 panic("%s: unknown state", __func__);
1360 LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state));
1362 lp->lp_mux_state = new_state;
1366 lacp_sm_mux_timer(struct lacp_port *lp)
1368 struct lacp_aggregator *la = lp->lp_aggregator;
1369 #if defined(LACP_DEBUG)
1370 char buf[LACP_LAGIDSTR_MAX+1];
1373 KASSERT(la->la_pending > 0, ("no pending event"));
1375 LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__,
1376 lacp_format_lagid(&la->la_actor, &la->la_partner,
1378 la->la_pending, la->la_pending - 1));
1383 /* periodic transmit machine */
1386 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate)
1388 if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state,
1389 LACP_STATE_TIMEOUT)) {
1393 LACP_DPRINTF((lp, "partner timeout changed\n"));
1396 * FAST_PERIODIC -> SLOW_PERIODIC
1398 * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC
1400 * let lacp_sm_ptx_tx_schedule to update timeout.
1403 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1406 * if timeout has been shortened, assert NTT.
1409 if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) {
1410 lacp_sm_assert_ntt(lp);
1415 lacp_sm_ptx_tx_schedule(struct lacp_port *lp)
1419 if (!(lp->lp_state & LACP_STATE_ACTIVITY) &&
1420 !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) {
1426 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1430 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) {
1434 timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ?
1435 LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME;
1437 LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout);
1441 lacp_sm_ptx_timer(struct lacp_port *lp)
1443 lacp_sm_assert_ntt(lp);
1447 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du)
1452 * check LACP_DISABLED first
1455 if (!(lp->lp_state & LACP_STATE_AGGREGATION)) {
1460 * check loopback condition.
1463 if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid,
1464 &lp->lp_actor.lip_systemid)) {
1469 * EXPIRED, DEFAULTED, CURRENT -> CURRENT
1472 lacp_sm_rx_update_selected(lp, du);
1473 lacp_sm_rx_update_ntt(lp, du);
1474 lacp_sm_rx_record_pdu(lp, du);
1476 timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ?
1477 LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME;
1478 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout);
1480 lp->lp_state &= ~LACP_STATE_EXPIRED;
1483 * kick transmit machine without waiting the next tick.
1490 lacp_sm_rx_set_expired(struct lacp_port *lp)
1492 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1493 lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT;
1494 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME);
1495 lp->lp_state |= LACP_STATE_EXPIRED;
1499 lacp_sm_rx_timer(struct lacp_port *lp)
1501 if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) {
1502 /* CURRENT -> EXPIRED */
1503 LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__));
1504 lacp_sm_rx_set_expired(lp);
1506 /* EXPIRED -> DEFAULTED */
1507 LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__));
1508 lacp_sm_rx_update_default_selected(lp);
1509 lacp_sm_rx_record_default(lp);
1510 lp->lp_state &= ~LACP_STATE_EXPIRED;
1515 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du)
1519 #if defined(LACP_DEBUG)
1520 char buf[LACP_STATESTR_MAX+1];
1523 /* LACP_DPRINTF((lp, "%s\n", __func__)); */
1525 oldpstate = lp->lp_partner.lip_state;
1527 active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY)
1528 || ((lp->lp_state & LACP_STATE_ACTIVITY) &&
1529 (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY));
1531 lp->lp_partner = du->ldu_actor;
1533 ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1534 LACP_STATE_AGGREGATION) &&
1535 !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner))
1536 || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) {
1539 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1542 lp->lp_state &= ~LACP_STATE_DEFAULTED;
1544 if (oldpstate != lp->lp_partner.lip_state) {
1545 LACP_DPRINTF((lp, "old pstate %s\n",
1546 lacp_format_state(oldpstate, buf, sizeof(buf))));
1547 LACP_DPRINTF((lp, "new pstate %s\n",
1548 lacp_format_state(lp->lp_partner.lip_state, buf,
1552 lacp_sm_ptx_update_timeout(lp, oldpstate);
1556 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du)
1558 /* LACP_DPRINTF((lp, "%s\n", __func__)); */
1560 if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) ||
1561 !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1562 LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) {
1563 LACP_DPRINTF((lp, "%s: assert ntt\n", __func__));
1564 lacp_sm_assert_ntt(lp);
1569 lacp_sm_rx_record_default(struct lacp_port *lp)
1573 /* LACP_DPRINTF((lp, "%s\n", __func__)); */
1575 oldpstate = lp->lp_partner.lip_state;
1576 lp->lp_partner = lacp_partner_admin;
1577 lp->lp_state |= LACP_STATE_DEFAULTED;
1578 lacp_sm_ptx_update_timeout(lp, oldpstate);
1582 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp,
1583 const struct lacp_peerinfo *info)
1585 /* LACP_DPRINTF((lp, "%s\n", __func__)); */
1587 if (lacp_compare_peerinfo(&lp->lp_partner, info) ||
1588 !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state,
1589 LACP_STATE_AGGREGATION)) {
1590 lp->lp_selected = LACP_UNSELECTED;
1591 /* mux machine will clean up lp->lp_aggregator */
1596 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du)
1598 /* LACP_DPRINTF((lp, "%s\n", __func__)); */
1600 lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor);
1604 lacp_sm_rx_update_default_selected(struct lacp_port *lp)
1606 /* LACP_DPRINTF((lp, "%s\n", __func__)); */
1608 lacp_sm_rx_update_selected_from_peerinfo(lp, &lacp_partner_admin);
1611 /* transmit machine */
1614 lacp_sm_tx(struct lacp_port *lp)
1618 if (!(lp->lp_state & LACP_STATE_AGGREGATION)
1620 || (!(lp->lp_state & LACP_STATE_ACTIVITY)
1621 && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY))
1624 lp->lp_flags &= ~LACP_PORT_NTT;
1627 if (!(lp->lp_flags & LACP_PORT_NTT)) {
1631 /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */
1632 if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent,
1633 (3 / LACP_FAST_PERIODIC_TIME)) == 0) {
1634 LACP_DPRINTF((lp, "rate limited pdu\n"));
1638 error = lacp_xmit_lacpdu(lp);
1641 lp->lp_flags &= ~LACP_PORT_NTT;
1643 LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n",
1649 lacp_sm_assert_ntt(struct lacp_port *lp)
1652 lp->lp_flags |= LACP_PORT_NTT;
1656 lacp_run_timers(struct lacp_port *lp)
1660 for (i = 0; i < LACP_NTIMER; i++) {
1661 KASSERT(lp->lp_timer[i] >= 0,
1662 ("invalid timer value %d", lp->lp_timer[i]));
1663 if (lp->lp_timer[i] == 0) {
1665 } else if (--lp->lp_timer[i] <= 0) {
1666 if (lacp_timer_funcs[i]) {
1667 (*lacp_timer_funcs[i])(lp);
1674 lacp_marker_input(struct lagg_port *lgp, struct mbuf *m)
1676 struct lacp_port *lp = LACP_PORT(lgp);
1677 struct lacp_port *lp2;
1678 struct lacp_softc *lsc = lp->lp_lsc;
1679 struct markerdu *mdu;
1683 LAGG_RLOCK_ASSERT(lgp->lp_lagg);
1685 if (__predict_false(lp->lp_flags & LACP_PORT_DETACHING)) {
1689 if (m->m_pkthdr.len != sizeof(*mdu)) {
1693 if ((m->m_flags & M_MCAST) == 0) {
1697 if (m->m_len < sizeof(*mdu)) {
1698 m = m_pullup(m, sizeof(*mdu));
1704 mdu = mtod(m, struct markerdu *);
1706 if (memcmp(&mdu->mdu_eh.ether_dhost,
1707 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
1712 KASSERT(mdu->mdu_sph.sph_subtype == SLOWPROTOCOLS_SUBTYPE_MARKER,
1713 ("a very bad kassert!"));
1716 if (mdu->mdu_sph.sph_version != 1) {
1720 switch (mdu->mdu_tlv.tlv_type) {
1721 case MARKER_TYPE_INFO:
1722 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
1723 marker_info_tlv_template, TRUE)) {
1726 mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE;
1727 memcpy(&mdu->mdu_eh.ether_dhost,
1728 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN);
1729 memcpy(&mdu->mdu_eh.ether_shost,
1730 lgp->lp_lladdr, ETHER_ADDR_LEN);
1731 error = lagg_enqueue(lp->lp_ifp, m);
1734 case MARKER_TYPE_RESPONSE:
1735 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
1736 marker_response_tlv_template, TRUE)) {
1739 LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n",
1740 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system,
1741 ":", ntohl(mdu->mdu_info.mi_rq_xid)));
1743 /* Verify that it is the last marker we sent out */
1744 if (memcmp(&mdu->mdu_info, &lp->lp_marker,
1745 sizeof(struct lacp_markerinfo)))
1748 lp->lp_flags &= ~LACP_PORT_MARK;
1750 if (lsc->lsc_suppress_distributing) {
1751 /* Check if any ports are waiting for a response */
1752 LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) {
1753 if (lp2->lp_flags & LACP_PORT_MARK) {
1760 /* All interface queues are clear */
1761 LACP_DPRINTF((NULL, "queue flush complete\n"));
1762 lsc->lsc_suppress_distributing = FALSE;
1776 LACP_DPRINTF((lp, "bad marker frame\n"));
1782 tlv_check(const void *p, size_t size, const struct tlvhdr *tlv,
1783 const struct tlv_template *tmpl, boolean_t check_type)
1785 while (/* CONSTCOND */ 1) {
1786 if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) {
1789 if ((check_type && tlv->tlv_type != tmpl->tmpl_type) ||
1790 tlv->tlv_length != tmpl->tmpl_length) {
1793 if (tmpl->tmpl_type == 0) {
1796 tlv = (const struct tlvhdr *)
1797 ((const char *)tlv + tlv->tlv_length);
1804 #if defined(LACP_DEBUG)
1806 lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen)
1808 snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X",
1820 lacp_format_systemid(const struct lacp_systemid *sysid,
1821 char *buf, size_t buflen)
1823 char macbuf[LACP_MACSTR_MAX+1];
1825 snprintf(buf, buflen, "%04X,%s",
1826 ntohs(sysid->lsi_prio),
1827 lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf)));
1833 lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen)
1835 snprintf(buf, buflen, "%04X,%04X",
1836 ntohs(portid->lpi_prio),
1837 ntohs(portid->lpi_portno));
1843 lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen)
1845 char sysid[LACP_SYSTEMIDSTR_MAX+1];
1846 char portid[LACP_PORTIDSTR_MAX+1];
1848 snprintf(buf, buflen, "(%s,%04X,%s)",
1849 lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)),
1850 ntohs(peer->lip_key),
1851 lacp_format_portid(&peer->lip_portid, portid, sizeof(portid)));
1857 lacp_format_lagid(const struct lacp_peerinfo *a,
1858 const struct lacp_peerinfo *b, char *buf, size_t buflen)
1860 char astr[LACP_PARTNERSTR_MAX+1];
1861 char bstr[LACP_PARTNERSTR_MAX+1];
1865 * there's a convention to display small numbered peer
1869 if (lacp_compare_peerinfo(a, b) > 0) {
1870 const struct lacp_peerinfo *t;
1878 snprintf(buf, buflen, "[%s,%s]",
1879 lacp_format_partner(a, astr, sizeof(astr)),
1880 lacp_format_partner(b, bstr, sizeof(bstr)));
1886 lacp_format_lagid_aggregator(const struct lacp_aggregator *la,
1887 char *buf, size_t buflen)
1893 return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen));
1897 lacp_format_state(uint8_t state, char *buf, size_t buflen)
1899 snprintf(buf, buflen, "%b", state, LACP_STATE_BITS);
1904 lacp_dump_lacpdu(const struct lacpdu *du)
1906 char buf[LACP_PARTNERSTR_MAX+1];
1907 char buf2[LACP_STATESTR_MAX+1];
1909 printf("actor=%s\n",
1910 lacp_format_partner(&du->ldu_actor, buf, sizeof(buf)));
1911 printf("actor.state=%s\n",
1912 lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2)));
1913 printf("partner=%s\n",
1914 lacp_format_partner(&du->ldu_partner, buf, sizeof(buf)));
1915 printf("partner.state=%s\n",
1916 lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2)));
1918 printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay));
1922 lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...)
1927 printf("%s: ", lp->lp_ifp->if_xname);