1 /* $NetBSD: ieee8023ad_lacp.c,v 1.3 2005/12/11 12:24:54 christos Exp $ */
4 * Copyright (c)2005 YAMAMOTO Takashi,
5 * Copyright (c)2008 Andrew Thompson <thompsa@FreeBSD.org>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
33 #include <sys/param.h>
34 #include <sys/callout.h>
36 #include <sys/systm.h>
37 #include <sys/malloc.h>
38 #include <sys/kernel.h> /* hz */
39 #include <sys/socket.h> /* for net/if.h */
40 #include <sys/sockio.h>
41 #include <sys/sysctl.h>
42 #include <machine/stdarg.h>
44 #include <sys/rwlock.h>
47 #include <net/if_dl.h>
48 #include <net/ethernet.h>
49 #include <net/if_media.h>
50 #include <net/if_types.h>
52 #include <net/if_lagg.h>
53 #include <net/ieee8023ad_lacp.h>
56 * actor system priority and port priority.
57 * XXX should be configurable.
60 #define LACP_SYSTEM_PRIO 0x8000
61 #define LACP_PORT_PRIO 0x8000
63 const uint8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] =
64 { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 };
66 static const struct tlv_template lacp_info_tlv_template[] = {
67 { LACP_TYPE_ACTORINFO,
68 sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
69 { LACP_TYPE_PARTNERINFO,
70 sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
71 { LACP_TYPE_COLLECTORINFO,
72 sizeof(struct tlvhdr) + sizeof(struct lacp_collectorinfo) },
76 static const struct tlv_template marker_info_tlv_template[] = {
78 sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
82 static const struct tlv_template marker_response_tlv_template[] = {
83 { MARKER_TYPE_RESPONSE,
84 sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
88 typedef void (*lacp_timer_func_t)(struct lacp_port *);
90 static void lacp_fill_actorinfo(struct lacp_port *, struct lacp_peerinfo *);
91 static void lacp_fill_markerinfo(struct lacp_port *,
92 struct lacp_markerinfo *);
94 static uint64_t lacp_aggregator_bandwidth(struct lacp_aggregator *);
95 static void lacp_suppress_distributing(struct lacp_softc *,
96 struct lacp_aggregator *);
97 static void lacp_transit_expire(void *);
98 static void lacp_update_portmap(struct lacp_softc *);
99 static void lacp_select_active_aggregator(struct lacp_softc *);
100 static uint16_t lacp_compose_key(struct lacp_port *);
101 static int tlv_check(const void *, size_t, const struct tlvhdr *,
102 const struct tlv_template *, boolean_t);
103 static void lacp_tick(void *);
105 static void lacp_fill_aggregator_id(struct lacp_aggregator *,
106 const struct lacp_port *);
107 static void lacp_fill_aggregator_id_peer(struct lacp_peerinfo *,
108 const struct lacp_peerinfo *);
109 static int lacp_aggregator_is_compatible(const struct lacp_aggregator *,
110 const struct lacp_port *);
111 static int lacp_peerinfo_is_compatible(const struct lacp_peerinfo *,
112 const struct lacp_peerinfo *);
114 static struct lacp_aggregator *lacp_aggregator_get(struct lacp_softc *,
116 static void lacp_aggregator_addref(struct lacp_softc *,
117 struct lacp_aggregator *);
118 static void lacp_aggregator_delref(struct lacp_softc *,
119 struct lacp_aggregator *);
121 /* receive machine */
123 static int lacp_pdu_input(struct lacp_port *, struct mbuf *);
124 static int lacp_marker_input(struct lacp_port *, struct mbuf *);
125 static void lacp_sm_rx(struct lacp_port *, const struct lacpdu *);
126 static void lacp_sm_rx_timer(struct lacp_port *);
127 static void lacp_sm_rx_set_expired(struct lacp_port *);
128 static void lacp_sm_rx_update_ntt(struct lacp_port *,
129 const struct lacpdu *);
130 static void lacp_sm_rx_record_pdu(struct lacp_port *,
131 const struct lacpdu *);
132 static void lacp_sm_rx_update_selected(struct lacp_port *,
133 const struct lacpdu *);
134 static void lacp_sm_rx_record_default(struct lacp_port *);
135 static void lacp_sm_rx_update_default_selected(struct lacp_port *);
136 static void lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *,
137 const struct lacp_peerinfo *);
141 static void lacp_sm_mux(struct lacp_port *);
142 static void lacp_set_mux(struct lacp_port *, enum lacp_mux_state);
143 static void lacp_sm_mux_timer(struct lacp_port *);
145 /* periodic transmit machine */
147 static void lacp_sm_ptx_update_timeout(struct lacp_port *, uint8_t);
148 static void lacp_sm_ptx_tx_schedule(struct lacp_port *);
149 static void lacp_sm_ptx_timer(struct lacp_port *);
151 /* transmit machine */
153 static void lacp_sm_tx(struct lacp_port *);
154 static void lacp_sm_assert_ntt(struct lacp_port *);
156 static void lacp_run_timers(struct lacp_port *);
157 static int lacp_compare_peerinfo(const struct lacp_peerinfo *,
158 const struct lacp_peerinfo *);
159 static int lacp_compare_systemid(const struct lacp_systemid *,
160 const struct lacp_systemid *);
161 static void lacp_port_enable(struct lacp_port *);
162 static void lacp_port_disable(struct lacp_port *);
163 static void lacp_select(struct lacp_port *);
164 static void lacp_unselect(struct lacp_port *);
165 static void lacp_disable_collecting(struct lacp_port *);
166 static void lacp_enable_collecting(struct lacp_port *);
167 static void lacp_disable_distributing(struct lacp_port *);
168 static void lacp_enable_distributing(struct lacp_port *);
169 static int lacp_xmit_lacpdu(struct lacp_port *);
170 static int lacp_xmit_marker(struct lacp_port *);
174 static void lacp_dump_lacpdu(const struct lacpdu *);
175 static const char *lacp_format_partner(const struct lacp_peerinfo *, char *,
177 static const char *lacp_format_lagid(const struct lacp_peerinfo *,
178 const struct lacp_peerinfo *, char *, size_t);
179 static const char *lacp_format_lagid_aggregator(const struct lacp_aggregator *,
181 static const char *lacp_format_state(uint8_t, char *, size_t);
182 static const char *lacp_format_mac(const uint8_t *, char *, size_t);
183 static const char *lacp_format_systemid(const struct lacp_systemid *, char *,
185 static const char *lacp_format_portid(const struct lacp_portid *, char *,
187 static void lacp_dprintf(const struct lacp_port *, const char *, ...)
188 __attribute__((__format__(__printf__, 2, 3)));
190 static int lacp_debug = 0;
191 SYSCTL_INT(_net, OID_AUTO, lacp_debug, CTLFLAG_RW | CTLFLAG_TUN,
192 &lacp_debug, 0, "Enable LACP debug logging (1=debug, 2=trace)");
193 TUNABLE_INT("net.lacp_debug", &lacp_debug);
195 #define LACP_DPRINTF(a) if (lacp_debug > 0) { lacp_dprintf a ; }
196 #define LACP_TRACE(a) if (lacp_debug > 1) { lacp_dprintf(a,"%s\n",__func__); }
199 * partner administration variables.
200 * XXX should be configurable.
203 static const struct lacp_peerinfo lacp_partner_admin = {
204 .lip_systemid = { .lsi_prio = 0xffff },
205 .lip_portid = { .lpi_prio = 0xffff },
208 .lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION |
209 LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING,
216 static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = {
217 [LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer,
218 [LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer,
219 [LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer,
223 lacp_input(struct lagg_port *lgp, struct mbuf *m)
225 struct lacp_port *lp = LACP_PORT(lgp);
228 if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) {
233 m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype);
235 case SLOWPROTOCOLS_SUBTYPE_LACP:
236 lacp_pdu_input(lp, m);
239 case SLOWPROTOCOLS_SUBTYPE_MARKER:
240 lacp_marker_input(lp, m);
244 /* Not a subtype we are interested in */
249 * lacp_pdu_input: process lacpdu
252 lacp_pdu_input(struct lacp_port *lp, struct mbuf *m)
254 struct lacp_softc *lsc = lp->lp_lsc;
258 if (m->m_pkthdr.len != sizeof(*du)) {
262 if ((m->m_flags & M_MCAST) == 0) {
266 if (m->m_len < sizeof(*du)) {
267 m = m_pullup(m, sizeof(*du));
273 du = mtod(m, struct lacpdu *);
275 if (memcmp(&du->ldu_eh.ether_dhost,
276 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
281 * ignore the version for compatibility with
282 * the future protocol revisions.
285 if (du->ldu_sph.sph_version != 1) {
291 * ignore tlv types for compatibility with
292 * the future protocol revisions.
294 if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor,
295 lacp_info_tlv_template, FALSE)) {
299 if (lacp_debug > 0) {
300 lacp_dprintf(lp, "lacpdu receive\n");
301 lacp_dump_lacpdu(du);
317 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info)
319 struct lagg_port *lgp = lp->lp_lagg;
320 struct lagg_softc *sc = lgp->lp_softc;
322 info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO);
323 memcpy(&info->lip_systemid.lsi_mac,
324 IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
325 info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO);
326 info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index);
327 info->lip_state = lp->lp_state;
331 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info)
333 struct ifnet *ifp = lp->lp_ifp;
335 /* Fill in the port index and system id (encoded as the MAC) */
336 info->mi_rq_port = htons(ifp->if_index);
337 memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN);
338 info->mi_rq_xid = htonl(0);
342 lacp_xmit_lacpdu(struct lacp_port *lp)
344 struct lagg_port *lgp = lp->lp_lagg;
349 LACP_LOCK_ASSERT(lp->lp_lsc);
351 m = m_gethdr(M_DONTWAIT, MT_DATA);
355 m->m_len = m->m_pkthdr.len = sizeof(*du);
357 du = mtod(m, struct lacpdu *);
358 memset(du, 0, sizeof(*du));
360 memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
362 memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
363 du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW);
365 du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP;
366 du->ldu_sph.sph_version = 1;
368 TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor));
369 du->ldu_actor = lp->lp_actor;
371 TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO,
372 sizeof(du->ldu_partner));
373 du->ldu_partner = lp->lp_partner;
375 TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO,
376 sizeof(du->ldu_collector));
377 du->ldu_collector.lci_maxdelay = 0;
379 if (lacp_debug > 0) {
380 lacp_dprintf(lp, "lacpdu transmit\n");
381 lacp_dump_lacpdu(du);
384 m->m_flags |= M_MCAST;
387 * XXX should use higher priority queue.
388 * otherwise network congestion can break aggregation.
391 error = lagg_enqueue(lp->lp_ifp, m);
396 lacp_xmit_marker(struct lacp_port *lp)
398 struct lagg_port *lgp = lp->lp_lagg;
400 struct markerdu *mdu;
403 LACP_LOCK_ASSERT(lp->lp_lsc);
405 m = m_gethdr(M_DONTWAIT, MT_DATA);
409 m->m_len = m->m_pkthdr.len = sizeof(*mdu);
411 mdu = mtod(m, struct markerdu *);
412 memset(mdu, 0, sizeof(*mdu));
414 memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
416 memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
417 mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW);
419 mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER;
420 mdu->mdu_sph.sph_version = 1;
422 /* Bump the transaction id and copy over the marker info */
423 lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1);
424 TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info));
425 mdu->mdu_info = lp->lp_marker;
427 LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n",
428 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":",
429 ntohl(mdu->mdu_info.mi_rq_xid)));
431 m->m_flags |= M_MCAST;
432 error = lagg_enqueue(lp->lp_ifp, m);
437 lacp_linkstate(struct lagg_port *lgp)
439 struct lacp_port *lp = LACP_PORT(lgp);
440 struct lacp_softc *lsc = lp->lp_lsc;
441 struct ifnet *ifp = lgp->lp_ifp;
442 struct ifmediareq ifmr;
448 bzero((char *)&ifmr, sizeof(ifmr));
449 error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr);
454 media = ifmr.ifm_active;
455 LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, "
456 "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER,
457 (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP));
458 old_state = lp->lp_state;
459 old_key = lp->lp_key;
461 lp->lp_media = media;
463 * If the port is not an active full duplex Ethernet link then it can
466 if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 ||
467 ifp->if_link_state != LINK_STATE_UP) {
468 lacp_port_disable(lp);
470 lacp_port_enable(lp);
472 lp->lp_key = lacp_compose_key(lp);
474 if (old_state != lp->lp_state || old_key != lp->lp_key) {
475 LACP_DPRINTF((lp, "-> UNSELECTED\n"));
476 lp->lp_selected = LACP_UNSELECTED;
484 struct lacp_softc *lsc = arg;
485 struct lacp_port *lp;
487 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
488 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0)
496 lacp_sm_ptx_tx_schedule(lp);
498 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
502 lacp_port_create(struct lagg_port *lgp)
504 struct lagg_softc *sc = lgp->lp_softc;
505 struct lacp_softc *lsc = LACP_SOFTC(sc);
506 struct lacp_port *lp;
507 struct ifnet *ifp = lgp->lp_ifp;
508 struct sockaddr_dl sdl;
509 struct ifmultiaddr *rifma = NULL;
512 boolean_t active = TRUE; /* XXX should be configurable */
513 boolean_t fast = FALSE; /* XXX should be configurable */
515 bzero((char *)&sdl, sizeof(sdl));
516 sdl.sdl_len = sizeof(sdl);
517 sdl.sdl_family = AF_LINK;
518 sdl.sdl_index = ifp->if_index;
519 sdl.sdl_type = IFT_ETHER;
520 sdl.sdl_alen = ETHER_ADDR_LEN;
522 bcopy(ðermulticastaddr_slowprotocols,
523 LLADDR(&sdl), ETHER_ADDR_LEN);
524 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
526 printf("%s: ADDMULTI failed on %s\n", __func__, lgp->lp_ifname);
530 lp = malloc(sizeof(struct lacp_port),
531 M_DEVBUF, M_NOWAIT|M_ZERO);
536 lgp->lp_psc = (caddr_t)lp;
542 LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next);
544 lacp_fill_actorinfo(lp, &lp->lp_actor);
545 lacp_fill_markerinfo(lp, &lp->lp_marker);
547 (active ? LACP_STATE_ACTIVITY : 0) |
548 (fast ? LACP_STATE_TIMEOUT : 0);
549 lp->lp_aggregator = NULL;
550 lacp_sm_rx_set_expired(lp);
558 lacp_port_destroy(struct lagg_port *lgp)
560 struct lacp_port *lp = LACP_PORT(lgp);
561 struct lacp_softc *lsc = lp->lp_lsc;
565 for (i = 0; i < LACP_NTIMER; i++) {
566 LACP_TIMER_DISARM(lp, i);
569 lacp_disable_collecting(lp);
570 lacp_disable_distributing(lp);
573 /* The address may have already been removed by if_purgemaddrs() */
574 if (!lgp->lp_detaching)
575 if_delmulti_ifma(lp->lp_ifma);
577 LIST_REMOVE(lp, lp_next);
583 lacp_req(struct lagg_softc *sc, caddr_t data)
585 struct lacp_opreq *req = (struct lacp_opreq *)data;
586 struct lacp_softc *lsc = LACP_SOFTC(sc);
587 struct lacp_aggregator *la;
589 bzero(req, sizeof(struct lacp_opreq));
592 * If the LACP softc is NULL, return with the opreq structure full of
593 * zeros. It is normal for the softc to be NULL while the lagg is
599 la = lsc->lsc_active_aggregator;
602 req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio);
603 memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac,
605 req->actor_key = ntohs(la->la_actor.lip_key);
606 req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio);
607 req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno);
608 req->actor_state = la->la_actor.lip_state;
610 req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio);
611 memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac,
613 req->partner_key = ntohs(la->la_partner.lip_key);
614 req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio);
615 req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno);
616 req->partner_state = la->la_partner.lip_state;
622 lacp_portreq(struct lagg_port *lgp, caddr_t data)
624 struct lacp_opreq *req = (struct lacp_opreq *)data;
625 struct lacp_port *lp = LACP_PORT(lgp);
626 struct lacp_softc *lsc = lp->lp_lsc;
629 req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio);
630 memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac,
632 req->actor_key = ntohs(lp->lp_actor.lip_key);
633 req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio);
634 req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno);
635 req->actor_state = lp->lp_actor.lip_state;
637 req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio);
638 memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac,
640 req->partner_key = ntohs(lp->lp_partner.lip_key);
641 req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio);
642 req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno);
643 req->partner_state = lp->lp_partner.lip_state;
648 lacp_disable_collecting(struct lacp_port *lp)
650 LACP_DPRINTF((lp, "collecting disabled\n"));
651 lp->lp_state &= ~LACP_STATE_COLLECTING;
655 lacp_enable_collecting(struct lacp_port *lp)
657 LACP_DPRINTF((lp, "collecting enabled\n"));
658 lp->lp_state |= LACP_STATE_COLLECTING;
662 lacp_disable_distributing(struct lacp_port *lp)
664 struct lacp_aggregator *la = lp->lp_aggregator;
665 struct lacp_softc *lsc = lp->lp_lsc;
666 char buf[LACP_LAGIDSTR_MAX+1];
668 LACP_LOCK_ASSERT(lsc);
670 if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) {
674 KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports"));
675 KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports));
676 KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid"));
678 LACP_DPRINTF((lp, "disable distributing on aggregator %s, "
680 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
681 la->la_nports, la->la_nports - 1));
683 TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q);
686 if (lsc->lsc_active_aggregator == la) {
687 lacp_suppress_distributing(lsc, la);
688 lacp_select_active_aggregator(lsc);
689 /* regenerate the port map, the active aggregator has changed */
690 lacp_update_portmap(lsc);
693 lp->lp_state &= ~LACP_STATE_DISTRIBUTING;
697 lacp_enable_distributing(struct lacp_port *lp)
699 struct lacp_aggregator *la = lp->lp_aggregator;
700 struct lacp_softc *lsc = lp->lp_lsc;
701 char buf[LACP_LAGIDSTR_MAX+1];
703 LACP_LOCK_ASSERT(lsc);
705 if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) {
709 LACP_DPRINTF((lp, "enable distributing on aggregator %s, "
711 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
712 la->la_nports, la->la_nports + 1));
714 KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid"));
715 TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q);
718 lp->lp_state |= LACP_STATE_DISTRIBUTING;
720 if (lsc->lsc_active_aggregator == la) {
721 lacp_suppress_distributing(lsc, la);
722 lacp_update_portmap(lsc);
724 /* try to become the active aggregator */
725 lacp_select_active_aggregator(lsc);
729 lacp_transit_expire(void *vp)
731 struct lacp_softc *lsc = vp;
733 LACP_LOCK_ASSERT(lsc);
737 lsc->lsc_suppress_distributing = FALSE;
741 lacp_attach(struct lagg_softc *sc)
743 struct lacp_softc *lsc;
745 lsc = malloc(sizeof(struct lacp_softc),
746 M_DEVBUF, M_NOWAIT|M_ZERO);
750 sc->sc_psc = (caddr_t)lsc;
753 lsc->lsc_hashkey = arc4random();
754 lsc->lsc_active_aggregator = NULL;
756 TAILQ_INIT(&lsc->lsc_aggregators);
757 LIST_INIT(&lsc->lsc_ports);
759 callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0);
760 callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0);
762 /* if the lagg is already up then do the same */
763 if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING)
770 lacp_detach(struct lagg_softc *sc)
772 struct lacp_softc *lsc = LACP_SOFTC(sc);
774 KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators),
775 ("aggregators still active"));
776 KASSERT(lsc->lsc_active_aggregator == NULL,
777 ("aggregator still attached"));
780 callout_drain(&lsc->lsc_transit_callout);
781 callout_drain(&lsc->lsc_callout);
783 LACP_LOCK_DESTROY(lsc);
789 lacp_init(struct lagg_softc *sc)
791 struct lacp_softc *lsc = LACP_SOFTC(sc);
794 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
799 lacp_stop(struct lagg_softc *sc)
801 struct lacp_softc *lsc = LACP_SOFTC(sc);
804 callout_stop(&lsc->lsc_transit_callout);
805 callout_stop(&lsc->lsc_callout);
810 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
812 struct lacp_softc *lsc = LACP_SOFTC(sc);
813 struct lacp_portmap *pm;
814 struct lacp_port *lp;
817 if (__predict_false(lsc->lsc_suppress_distributing)) {
818 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__));
822 pm = &lsc->lsc_pmap[lsc->lsc_activemap];
823 if (pm->pm_count == 0) {
824 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__));
828 if (sc->use_flowid && (m->m_flags & M_FLOWID))
829 hash = m->m_pkthdr.flowid;
831 hash = lagg_hashmbuf(sc, m, lsc->lsc_hashkey);
832 hash %= pm->pm_count;
833 lp = pm->pm_map[hash];
835 KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0,
836 ("aggregated port is not distributing"));
838 return (lp->lp_lagg);
841 * lacp_suppress_distributing: drop transmit packets for a while
842 * to preserve packet ordering.
846 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la)
848 struct lacp_port *lp;
850 if (lsc->lsc_active_aggregator != la) {
856 lsc->lsc_suppress_distributing = TRUE;
858 /* send a marker frame down each port to verify the queues are empty */
859 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
860 lp->lp_flags |= LACP_PORT_MARK;
861 lacp_xmit_marker(lp);
864 /* set a timeout for the marker frames */
865 callout_reset(&lsc->lsc_transit_callout,
866 LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc);
870 lacp_compare_peerinfo(const struct lacp_peerinfo *a,
871 const struct lacp_peerinfo *b)
873 return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state)));
877 lacp_compare_systemid(const struct lacp_systemid *a,
878 const struct lacp_systemid *b)
880 return (memcmp(a, b, sizeof(*a)));
885 lacp_compare_portid(const struct lacp_portid *a,
886 const struct lacp_portid *b)
888 return (memcmp(a, b, sizeof(*a)));
893 lacp_aggregator_bandwidth(struct lacp_aggregator *la)
895 struct lacp_port *lp;
898 lp = TAILQ_FIRST(&la->la_ports);
903 speed = ifmedia_baudrate(lp->lp_media);
904 speed *= la->la_nports;
906 LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n",
907 lp->lp_media, la->la_nports));
914 * lacp_select_active_aggregator: select an aggregator to be used to transmit
915 * packets from lagg(4) interface.
919 lacp_select_active_aggregator(struct lacp_softc *lsc)
921 struct lagg_softc *sc = lsc->lsc_softc;
922 struct lacp_aggregator *la;
923 struct lacp_aggregator *best_la = NULL;
924 uint64_t best_speed = 0;
925 char buf[LACP_LAGIDSTR_MAX+1];
929 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
932 if (la->la_nports == 0) {
936 speed = lacp_aggregator_bandwidth(la);
937 LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n",
938 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
939 speed, la->la_nports));
941 /* This aggregator is chosen if
942 * the partner has a better system priority
943 * or, the total aggregated speed is higher
944 * or, it is already the chosen aggregator
946 if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) <
947 LACP_SYS_PRI(best_la->la_partner)) ||
948 speed > best_speed ||
949 (speed == best_speed &&
950 la == lsc->lsc_active_aggregator)) {
956 KASSERT(best_la == NULL || best_la->la_nports > 0,
957 ("invalid aggregator refcnt"));
958 KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports),
959 ("invalid aggregator list"));
961 if (lsc->lsc_active_aggregator != best_la) {
962 LACP_DPRINTF((NULL, "active aggregator changed\n"));
963 LACP_DPRINTF((NULL, "old %s\n",
964 lacp_format_lagid_aggregator(lsc->lsc_active_aggregator,
967 LACP_DPRINTF((NULL, "active aggregator not changed\n"));
969 LACP_DPRINTF((NULL, "new %s\n",
970 lacp_format_lagid_aggregator(best_la, buf, sizeof(buf))));
972 if (lsc->lsc_active_aggregator != best_la) {
973 sc->sc_ifp->if_baudrate = best_speed;
974 lsc->lsc_active_aggregator = best_la;
975 lacp_update_portmap(lsc);
977 lacp_suppress_distributing(lsc, best_la);
983 * Updated the inactive portmap array with the new list of ports and
987 lacp_update_portmap(struct lacp_softc *lsc)
989 struct lacp_aggregator *la;
990 struct lacp_portmap *p;
991 struct lacp_port *lp;
995 newmap = lsc->lsc_activemap == 0 ? 1 : 0;
996 p = &lsc->lsc_pmap[newmap];
997 la = lsc->lsc_active_aggregator;
998 bzero(p, sizeof(struct lacp_portmap));
1000 if (la != NULL && la->la_nports > 0) {
1001 p->pm_count = la->la_nports;
1003 TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q)
1004 p->pm_map[i++] = lp;
1005 KASSERT(i == p->pm_count, ("Invalid port count"));
1008 /* switch the active portmap over */
1009 atomic_store_rel_int(&lsc->lsc_activemap, newmap);
1010 LACP_DPRINTF((NULL, "Set table %d with %d ports\n",
1012 lsc->lsc_pmap[lsc->lsc_activemap].pm_count));
1016 lacp_compose_key(struct lacp_port *lp)
1018 struct lagg_port *lgp = lp->lp_lagg;
1019 struct lagg_softc *sc = lgp->lp_softc;
1020 u_int media = lp->lp_media;
1023 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) {
1026 * non-aggregatable links should have unique keys.
1028 * XXX this isn't really unique as if_index is 16 bit.
1031 /* bit 0..14: (some bits of) if_index of this port */
1032 key = lp->lp_ifp->if_index;
1036 u_int subtype = IFM_SUBTYPE(media);
1038 KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type"));
1039 KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface"));
1041 /* bit 0..4: IFM_SUBTYPE modulo speed */
1066 case IFM_10G_TWINAX:
1067 case IFM_10G_TWINAX_LONG:
1080 /* bit 5..14: (some bits of) if_index of lagg device */
1081 key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5);
1084 return (htons(key));
1088 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1090 char buf[LACP_LAGIDSTR_MAX+1];
1092 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1094 lacp_format_lagid(&la->la_actor, &la->la_partner,
1096 la->la_refcnt, la->la_refcnt + 1));
1098 KASSERT(la->la_refcnt > 0, ("refcount <= 0"));
1100 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount"));
1104 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1106 char buf[LACP_LAGIDSTR_MAX+1];
1108 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1110 lacp_format_lagid(&la->la_actor, &la->la_partner,
1112 la->la_refcnt, la->la_refcnt - 1));
1114 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt"));
1116 if (la->la_refcnt > 0) {
1120 KASSERT(la->la_refcnt == 0, ("refcount not zero"));
1121 KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active"));
1123 TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q);
1129 * lacp_aggregator_get: allocate an aggregator.
1132 static struct lacp_aggregator *
1133 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp)
1135 struct lacp_aggregator *la;
1137 la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT);
1141 TAILQ_INIT(&la->la_ports);
1143 TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q);
1150 * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port.
1154 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp)
1156 lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner);
1157 lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor);
1159 la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION;
1163 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr,
1164 const struct lacp_peerinfo *lpi_port)
1166 memset(lpi_aggr, 0, sizeof(*lpi_aggr));
1167 lpi_aggr->lip_systemid = lpi_port->lip_systemid;
1168 lpi_aggr->lip_key = lpi_port->lip_key;
1172 * lacp_aggregator_is_compatible: check if a port can join to an aggregator.
1176 lacp_aggregator_is_compatible(const struct lacp_aggregator *la,
1177 const struct lacp_port *lp)
1179 if (!(lp->lp_state & LACP_STATE_AGGREGATION) ||
1180 !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) {
1184 if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) {
1188 if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) {
1192 if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) {
1200 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a,
1201 const struct lacp_peerinfo *b)
1203 if (memcmp(&a->lip_systemid, &b->lip_systemid,
1204 sizeof(a->lip_systemid))) {
1208 if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) {
1216 lacp_port_enable(struct lacp_port *lp)
1218 lp->lp_state |= LACP_STATE_AGGREGATION;
1222 lacp_port_disable(struct lacp_port *lp)
1224 lacp_set_mux(lp, LACP_MUX_DETACHED);
1226 lp->lp_state &= ~LACP_STATE_AGGREGATION;
1227 lp->lp_selected = LACP_UNSELECTED;
1228 lacp_sm_rx_record_default(lp);
1229 lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION;
1230 lp->lp_state &= ~LACP_STATE_EXPIRED;
1234 * lacp_select: select an aggregator. create one if necessary.
1237 lacp_select(struct lacp_port *lp)
1239 struct lacp_softc *lsc = lp->lp_lsc;
1240 struct lacp_aggregator *la;
1241 char buf[LACP_LAGIDSTR_MAX+1];
1243 if (lp->lp_aggregator) {
1247 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1248 ("timer_wait_while still active"));
1250 LACP_DPRINTF((lp, "port lagid=%s\n",
1251 lacp_format_lagid(&lp->lp_actor, &lp->lp_partner,
1252 buf, sizeof(buf))));
1254 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
1255 if (lacp_aggregator_is_compatible(la, lp)) {
1261 la = lacp_aggregator_get(lsc, lp);
1263 LACP_DPRINTF((lp, "aggregator creation failed\n"));
1266 * will retry on the next tick.
1271 lacp_fill_aggregator_id(la, lp);
1272 LACP_DPRINTF((lp, "aggregator created\n"));
1274 LACP_DPRINTF((lp, "compatible aggregator found\n"));
1275 if (la->la_refcnt == LACP_MAX_PORTS)
1277 lacp_aggregator_addref(lsc, la);
1280 LACP_DPRINTF((lp, "aggregator lagid=%s\n",
1281 lacp_format_lagid(&la->la_actor, &la->la_partner,
1282 buf, sizeof(buf))));
1284 lp->lp_aggregator = la;
1285 lp->lp_selected = LACP_SELECTED;
1289 * lacp_unselect: finish unselect/detach process.
1293 lacp_unselect(struct lacp_port *lp)
1295 struct lacp_softc *lsc = lp->lp_lsc;
1296 struct lacp_aggregator *la = lp->lp_aggregator;
1298 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1299 ("timer_wait_while still active"));
1305 lp->lp_aggregator = NULL;
1306 lacp_aggregator_delref(lsc, la);
1312 lacp_sm_mux(struct lacp_port *lp)
1314 enum lacp_mux_state new_state;
1316 (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0;
1317 boolean_t p_collecting =
1318 (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0;
1319 enum lacp_selected selected = lp->lp_selected;
1320 struct lacp_aggregator *la;
1323 lacp_dprintf(lp, "%s: state %d\n", __func__, lp->lp_mux_state);
1326 la = lp->lp_aggregator;
1327 KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL,
1328 ("MUX not detached"));
1329 new_state = lp->lp_mux_state;
1330 switch (lp->lp_mux_state) {
1331 case LACP_MUX_DETACHED:
1332 if (selected != LACP_UNSELECTED) {
1333 new_state = LACP_MUX_WAITING;
1336 case LACP_MUX_WAITING:
1337 KASSERT(la->la_pending > 0 ||
1338 !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1339 ("timer_wait_while still active"));
1340 if (selected == LACP_SELECTED && la->la_pending == 0) {
1341 new_state = LACP_MUX_ATTACHED;
1342 } else if (selected == LACP_UNSELECTED) {
1343 new_state = LACP_MUX_DETACHED;
1346 case LACP_MUX_ATTACHED:
1347 if (selected == LACP_SELECTED && p_sync) {
1348 new_state = LACP_MUX_COLLECTING;
1349 } else if (selected != LACP_SELECTED) {
1350 new_state = LACP_MUX_DETACHED;
1353 case LACP_MUX_COLLECTING:
1354 if (selected == LACP_SELECTED && p_sync && p_collecting) {
1355 new_state = LACP_MUX_DISTRIBUTING;
1356 } else if (selected != LACP_SELECTED || !p_sync) {
1357 new_state = LACP_MUX_ATTACHED;
1360 case LACP_MUX_DISTRIBUTING:
1361 if (selected != LACP_SELECTED || !p_sync || !p_collecting) {
1362 new_state = LACP_MUX_COLLECTING;
1366 panic("%s: unknown state", __func__);
1369 if (lp->lp_mux_state == new_state) {
1373 lacp_set_mux(lp, new_state);
1378 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state)
1380 struct lacp_aggregator *la = lp->lp_aggregator;
1382 if (lp->lp_mux_state == new_state) {
1386 switch (new_state) {
1387 case LACP_MUX_DETACHED:
1388 lp->lp_state &= ~LACP_STATE_SYNC;
1389 lacp_disable_distributing(lp);
1390 lacp_disable_collecting(lp);
1391 lacp_sm_assert_ntt(lp);
1393 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) {
1394 KASSERT(la->la_pending > 0,
1395 ("timer_wait_while not active"));
1398 LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE);
1401 case LACP_MUX_WAITING:
1402 LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE,
1403 LACP_AGGREGATE_WAIT_TIME);
1406 case LACP_MUX_ATTACHED:
1407 lp->lp_state |= LACP_STATE_SYNC;
1408 lacp_disable_collecting(lp);
1409 lacp_sm_assert_ntt(lp);
1411 case LACP_MUX_COLLECTING:
1412 lacp_enable_collecting(lp);
1413 lacp_disable_distributing(lp);
1414 lacp_sm_assert_ntt(lp);
1416 case LACP_MUX_DISTRIBUTING:
1417 lacp_enable_distributing(lp);
1420 panic("%s: unknown state", __func__);
1423 LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state));
1425 lp->lp_mux_state = new_state;
1429 lacp_sm_mux_timer(struct lacp_port *lp)
1431 struct lacp_aggregator *la = lp->lp_aggregator;
1432 char buf[LACP_LAGIDSTR_MAX+1];
1434 KASSERT(la->la_pending > 0, ("no pending event"));
1436 LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__,
1437 lacp_format_lagid(&la->la_actor, &la->la_partner,
1439 la->la_pending, la->la_pending - 1));
1444 /* periodic transmit machine */
1447 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate)
1449 if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state,
1450 LACP_STATE_TIMEOUT)) {
1454 LACP_DPRINTF((lp, "partner timeout changed\n"));
1457 * FAST_PERIODIC -> SLOW_PERIODIC
1459 * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC
1461 * let lacp_sm_ptx_tx_schedule to update timeout.
1464 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1467 * if timeout has been shortened, assert NTT.
1470 if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) {
1471 lacp_sm_assert_ntt(lp);
1476 lacp_sm_ptx_tx_schedule(struct lacp_port *lp)
1480 if (!(lp->lp_state & LACP_STATE_ACTIVITY) &&
1481 !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) {
1487 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1491 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) {
1495 timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ?
1496 LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME;
1498 LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout);
1502 lacp_sm_ptx_timer(struct lacp_port *lp)
1504 lacp_sm_assert_ntt(lp);
1508 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du)
1513 * check LACP_DISABLED first
1516 if (!(lp->lp_state & LACP_STATE_AGGREGATION)) {
1521 * check loopback condition.
1524 if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid,
1525 &lp->lp_actor.lip_systemid)) {
1530 * EXPIRED, DEFAULTED, CURRENT -> CURRENT
1533 lacp_sm_rx_update_selected(lp, du);
1534 lacp_sm_rx_update_ntt(lp, du);
1535 lacp_sm_rx_record_pdu(lp, du);
1537 timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ?
1538 LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME;
1539 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout);
1541 lp->lp_state &= ~LACP_STATE_EXPIRED;
1544 * kick transmit machine without waiting the next tick.
1551 lacp_sm_rx_set_expired(struct lacp_port *lp)
1553 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1554 lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT;
1555 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME);
1556 lp->lp_state |= LACP_STATE_EXPIRED;
1560 lacp_sm_rx_timer(struct lacp_port *lp)
1562 if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) {
1563 /* CURRENT -> EXPIRED */
1564 LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__));
1565 lacp_sm_rx_set_expired(lp);
1567 /* EXPIRED -> DEFAULTED */
1568 LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__));
1569 lacp_sm_rx_update_default_selected(lp);
1570 lacp_sm_rx_record_default(lp);
1571 lp->lp_state &= ~LACP_STATE_EXPIRED;
1576 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du)
1580 char buf[LACP_STATESTR_MAX+1];
1584 oldpstate = lp->lp_partner.lip_state;
1586 active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY)
1587 || ((lp->lp_state & LACP_STATE_ACTIVITY) &&
1588 (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY));
1590 lp->lp_partner = du->ldu_actor;
1592 ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1593 LACP_STATE_AGGREGATION) &&
1594 !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner))
1595 || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) {
1598 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1601 lp->lp_state &= ~LACP_STATE_DEFAULTED;
1603 if (oldpstate != lp->lp_partner.lip_state) {
1604 LACP_DPRINTF((lp, "old pstate %s\n",
1605 lacp_format_state(oldpstate, buf, sizeof(buf))));
1606 LACP_DPRINTF((lp, "new pstate %s\n",
1607 lacp_format_state(lp->lp_partner.lip_state, buf,
1611 lacp_sm_ptx_update_timeout(lp, oldpstate);
1615 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du)
1620 if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) ||
1621 !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1622 LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) {
1623 LACP_DPRINTF((lp, "%s: assert ntt\n", __func__));
1624 lacp_sm_assert_ntt(lp);
1629 lacp_sm_rx_record_default(struct lacp_port *lp)
1635 oldpstate = lp->lp_partner.lip_state;
1636 lp->lp_partner = lacp_partner_admin;
1637 lp->lp_state |= LACP_STATE_DEFAULTED;
1638 lacp_sm_ptx_update_timeout(lp, oldpstate);
1642 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp,
1643 const struct lacp_peerinfo *info)
1648 if (lacp_compare_peerinfo(&lp->lp_partner, info) ||
1649 !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state,
1650 LACP_STATE_AGGREGATION)) {
1651 lp->lp_selected = LACP_UNSELECTED;
1652 /* mux machine will clean up lp->lp_aggregator */
1657 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du)
1662 lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor);
1666 lacp_sm_rx_update_default_selected(struct lacp_port *lp)
1671 lacp_sm_rx_update_selected_from_peerinfo(lp, &lacp_partner_admin);
1674 /* transmit machine */
1677 lacp_sm_tx(struct lacp_port *lp)
1681 if (!(lp->lp_state & LACP_STATE_AGGREGATION)
1683 || (!(lp->lp_state & LACP_STATE_ACTIVITY)
1684 && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY))
1687 lp->lp_flags &= ~LACP_PORT_NTT;
1690 if (!(lp->lp_flags & LACP_PORT_NTT)) {
1694 /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */
1695 if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent,
1696 (3 / LACP_FAST_PERIODIC_TIME)) == 0) {
1697 LACP_DPRINTF((lp, "rate limited pdu\n"));
1701 error = lacp_xmit_lacpdu(lp);
1704 lp->lp_flags &= ~LACP_PORT_NTT;
1706 LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n",
1712 lacp_sm_assert_ntt(struct lacp_port *lp)
1715 lp->lp_flags |= LACP_PORT_NTT;
1719 lacp_run_timers(struct lacp_port *lp)
1723 for (i = 0; i < LACP_NTIMER; i++) {
1724 KASSERT(lp->lp_timer[i] >= 0,
1725 ("invalid timer value %d", lp->lp_timer[i]));
1726 if (lp->lp_timer[i] == 0) {
1728 } else if (--lp->lp_timer[i] <= 0) {
1729 if (lacp_timer_funcs[i]) {
1730 (*lacp_timer_funcs[i])(lp);
1737 lacp_marker_input(struct lacp_port *lp, struct mbuf *m)
1739 struct lacp_softc *lsc = lp->lp_lsc;
1740 struct lagg_port *lgp = lp->lp_lagg;
1741 struct lacp_port *lp2;
1742 struct markerdu *mdu;
1746 if (m->m_pkthdr.len != sizeof(*mdu)) {
1750 if ((m->m_flags & M_MCAST) == 0) {
1754 if (m->m_len < sizeof(*mdu)) {
1755 m = m_pullup(m, sizeof(*mdu));
1761 mdu = mtod(m, struct markerdu *);
1763 if (memcmp(&mdu->mdu_eh.ether_dhost,
1764 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
1768 if (mdu->mdu_sph.sph_version != 1) {
1772 switch (mdu->mdu_tlv.tlv_type) {
1773 case MARKER_TYPE_INFO:
1774 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
1775 marker_info_tlv_template, TRUE)) {
1778 mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE;
1779 memcpy(&mdu->mdu_eh.ether_dhost,
1780 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN);
1781 memcpy(&mdu->mdu_eh.ether_shost,
1782 lgp->lp_lladdr, ETHER_ADDR_LEN);
1783 error = lagg_enqueue(lp->lp_ifp, m);
1786 case MARKER_TYPE_RESPONSE:
1787 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
1788 marker_response_tlv_template, TRUE)) {
1791 LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n",
1792 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system,
1793 ":", ntohl(mdu->mdu_info.mi_rq_xid)));
1795 /* Verify that it is the last marker we sent out */
1796 if (memcmp(&mdu->mdu_info, &lp->lp_marker,
1797 sizeof(struct lacp_markerinfo)))
1801 lp->lp_flags &= ~LACP_PORT_MARK;
1803 if (lsc->lsc_suppress_distributing) {
1804 /* Check if any ports are waiting for a response */
1805 LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) {
1806 if (lp2->lp_flags & LACP_PORT_MARK) {
1813 /* All interface queues are clear */
1814 LACP_DPRINTF((NULL, "queue flush complete\n"));
1815 lsc->lsc_suppress_distributing = FALSE;
1829 LACP_DPRINTF((lp, "bad marker frame\n"));
1835 tlv_check(const void *p, size_t size, const struct tlvhdr *tlv,
1836 const struct tlv_template *tmpl, boolean_t check_type)
1838 while (/* CONSTCOND */ 1) {
1839 if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) {
1842 if ((check_type && tlv->tlv_type != tmpl->tmpl_type) ||
1843 tlv->tlv_length != tmpl->tmpl_length) {
1846 if (tmpl->tmpl_type == 0) {
1849 tlv = (const struct tlvhdr *)
1850 ((const char *)tlv + tlv->tlv_length);
1859 lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen)
1861 snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X",
1873 lacp_format_systemid(const struct lacp_systemid *sysid,
1874 char *buf, size_t buflen)
1876 char macbuf[LACP_MACSTR_MAX+1];
1878 snprintf(buf, buflen, "%04X,%s",
1879 ntohs(sysid->lsi_prio),
1880 lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf)));
1886 lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen)
1888 snprintf(buf, buflen, "%04X,%04X",
1889 ntohs(portid->lpi_prio),
1890 ntohs(portid->lpi_portno));
1896 lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen)
1898 char sysid[LACP_SYSTEMIDSTR_MAX+1];
1899 char portid[LACP_PORTIDSTR_MAX+1];
1901 snprintf(buf, buflen, "(%s,%04X,%s)",
1902 lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)),
1903 ntohs(peer->lip_key),
1904 lacp_format_portid(&peer->lip_portid, portid, sizeof(portid)));
1910 lacp_format_lagid(const struct lacp_peerinfo *a,
1911 const struct lacp_peerinfo *b, char *buf, size_t buflen)
1913 char astr[LACP_PARTNERSTR_MAX+1];
1914 char bstr[LACP_PARTNERSTR_MAX+1];
1918 * there's a convention to display small numbered peer
1922 if (lacp_compare_peerinfo(a, b) > 0) {
1923 const struct lacp_peerinfo *t;
1931 snprintf(buf, buflen, "[%s,%s]",
1932 lacp_format_partner(a, astr, sizeof(astr)),
1933 lacp_format_partner(b, bstr, sizeof(bstr)));
1939 lacp_format_lagid_aggregator(const struct lacp_aggregator *la,
1940 char *buf, size_t buflen)
1946 return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen));
1950 lacp_format_state(uint8_t state, char *buf, size_t buflen)
1952 snprintf(buf, buflen, "%b", state, LACP_STATE_BITS);
1957 lacp_dump_lacpdu(const struct lacpdu *du)
1959 char buf[LACP_PARTNERSTR_MAX+1];
1960 char buf2[LACP_STATESTR_MAX+1];
1962 printf("actor=%s\n",
1963 lacp_format_partner(&du->ldu_actor, buf, sizeof(buf)));
1964 printf("actor.state=%s\n",
1965 lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2)));
1966 printf("partner=%s\n",
1967 lacp_format_partner(&du->ldu_partner, buf, sizeof(buf)));
1968 printf("partner.state=%s\n",
1969 lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2)));
1971 printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay));
1975 lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...)
1980 printf("%s: ", lp->lp_ifp->if_xname);