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 = lsc->lsc_active_aggregator;
590 bzero(req, sizeof(struct lacp_opreq));
592 req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio);
593 memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac,
595 req->actor_key = ntohs(la->la_actor.lip_key);
596 req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio);
597 req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno);
598 req->actor_state = la->la_actor.lip_state;
600 req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio);
601 memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac,
603 req->partner_key = ntohs(la->la_partner.lip_key);
604 req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio);
605 req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno);
606 req->partner_state = la->la_partner.lip_state;
612 lacp_portreq(struct lagg_port *lgp, caddr_t data)
614 struct lacp_opreq *req = (struct lacp_opreq *)data;
615 struct lacp_port *lp = LACP_PORT(lgp);
616 struct lacp_softc *lsc = lp->lp_lsc;
619 req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio);
620 memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac,
622 req->actor_key = ntohs(lp->lp_actor.lip_key);
623 req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio);
624 req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno);
625 req->actor_state = lp->lp_actor.lip_state;
627 req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio);
628 memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac,
630 req->partner_key = ntohs(lp->lp_partner.lip_key);
631 req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio);
632 req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno);
633 req->partner_state = lp->lp_partner.lip_state;
638 lacp_disable_collecting(struct lacp_port *lp)
640 LACP_DPRINTF((lp, "collecting disabled\n"));
641 lp->lp_state &= ~LACP_STATE_COLLECTING;
645 lacp_enable_collecting(struct lacp_port *lp)
647 LACP_DPRINTF((lp, "collecting enabled\n"));
648 lp->lp_state |= LACP_STATE_COLLECTING;
652 lacp_disable_distributing(struct lacp_port *lp)
654 struct lacp_aggregator *la = lp->lp_aggregator;
655 struct lacp_softc *lsc = lp->lp_lsc;
656 char buf[LACP_LAGIDSTR_MAX+1];
658 LACP_LOCK_ASSERT(lsc);
660 if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) {
664 KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports"));
665 KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports));
666 KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid"));
668 LACP_DPRINTF((lp, "disable distributing on aggregator %s, "
670 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
671 la->la_nports, la->la_nports - 1));
673 TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q);
676 if (lsc->lsc_active_aggregator == la) {
677 lacp_suppress_distributing(lsc, la);
678 lacp_select_active_aggregator(lsc);
679 /* regenerate the port map, the active aggregator has changed */
680 lacp_update_portmap(lsc);
683 lp->lp_state &= ~LACP_STATE_DISTRIBUTING;
687 lacp_enable_distributing(struct lacp_port *lp)
689 struct lacp_aggregator *la = lp->lp_aggregator;
690 struct lacp_softc *lsc = lp->lp_lsc;
691 char buf[LACP_LAGIDSTR_MAX+1];
693 LACP_LOCK_ASSERT(lsc);
695 if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) {
699 LACP_DPRINTF((lp, "enable distributing on aggregator %s, "
701 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
702 la->la_nports, la->la_nports + 1));
704 KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid"));
705 TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q);
708 lp->lp_state |= LACP_STATE_DISTRIBUTING;
710 if (lsc->lsc_active_aggregator == la) {
711 lacp_suppress_distributing(lsc, la);
712 lacp_update_portmap(lsc);
714 /* try to become the active aggregator */
715 lacp_select_active_aggregator(lsc);
719 lacp_transit_expire(void *vp)
721 struct lacp_softc *lsc = vp;
723 LACP_LOCK_ASSERT(lsc);
727 lsc->lsc_suppress_distributing = FALSE;
731 lacp_attach(struct lagg_softc *sc)
733 struct lacp_softc *lsc;
735 lsc = malloc(sizeof(struct lacp_softc),
736 M_DEVBUF, M_NOWAIT|M_ZERO);
740 sc->sc_psc = (caddr_t)lsc;
743 lsc->lsc_hashkey = arc4random();
744 lsc->lsc_active_aggregator = NULL;
746 TAILQ_INIT(&lsc->lsc_aggregators);
747 LIST_INIT(&lsc->lsc_ports);
749 callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0);
750 callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0);
752 /* if the lagg is already up then do the same */
753 if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING)
760 lacp_detach(struct lagg_softc *sc)
762 struct lacp_softc *lsc = LACP_SOFTC(sc);
764 KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators),
765 ("aggregators still active"));
766 KASSERT(lsc->lsc_active_aggregator == NULL,
767 ("aggregator still attached"));
770 callout_drain(&lsc->lsc_transit_callout);
771 callout_drain(&lsc->lsc_callout);
773 LACP_LOCK_DESTROY(lsc);
779 lacp_init(struct lagg_softc *sc)
781 struct lacp_softc *lsc = LACP_SOFTC(sc);
784 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
789 lacp_stop(struct lagg_softc *sc)
791 struct lacp_softc *lsc = LACP_SOFTC(sc);
794 callout_stop(&lsc->lsc_transit_callout);
795 callout_stop(&lsc->lsc_callout);
800 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
802 struct lacp_softc *lsc = LACP_SOFTC(sc);
803 struct lacp_portmap *pm;
804 struct lacp_port *lp;
807 if (__predict_false(lsc->lsc_suppress_distributing)) {
808 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__));
812 pm = &lsc->lsc_pmap[lsc->lsc_activemap];
813 if (pm->pm_count == 0) {
814 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__));
818 if (sc->use_flowid && (m->m_flags & M_FLOWID))
819 hash = m->m_pkthdr.flowid;
821 hash = lagg_hashmbuf(sc, m, lsc->lsc_hashkey);
822 hash %= pm->pm_count;
823 lp = pm->pm_map[hash];
825 KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0,
826 ("aggregated port is not distributing"));
828 return (lp->lp_lagg);
831 * lacp_suppress_distributing: drop transmit packets for a while
832 * to preserve packet ordering.
836 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la)
838 struct lacp_port *lp;
840 if (lsc->lsc_active_aggregator != la) {
846 lsc->lsc_suppress_distributing = TRUE;
848 /* send a marker frame down each port to verify the queues are empty */
849 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
850 lp->lp_flags |= LACP_PORT_MARK;
851 lacp_xmit_marker(lp);
854 /* set a timeout for the marker frames */
855 callout_reset(&lsc->lsc_transit_callout,
856 LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc);
860 lacp_compare_peerinfo(const struct lacp_peerinfo *a,
861 const struct lacp_peerinfo *b)
863 return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state)));
867 lacp_compare_systemid(const struct lacp_systemid *a,
868 const struct lacp_systemid *b)
870 return (memcmp(a, b, sizeof(*a)));
875 lacp_compare_portid(const struct lacp_portid *a,
876 const struct lacp_portid *b)
878 return (memcmp(a, b, sizeof(*a)));
883 lacp_aggregator_bandwidth(struct lacp_aggregator *la)
885 struct lacp_port *lp;
888 lp = TAILQ_FIRST(&la->la_ports);
893 speed = ifmedia_baudrate(lp->lp_media);
894 speed *= la->la_nports;
896 LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n",
897 lp->lp_media, la->la_nports));
904 * lacp_select_active_aggregator: select an aggregator to be used to transmit
905 * packets from lagg(4) interface.
909 lacp_select_active_aggregator(struct lacp_softc *lsc)
911 struct lagg_softc *sc = lsc->lsc_softc;
912 struct lacp_aggregator *la;
913 struct lacp_aggregator *best_la = NULL;
914 uint64_t best_speed = 0;
915 char buf[LACP_LAGIDSTR_MAX+1];
919 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
922 if (la->la_nports == 0) {
926 speed = lacp_aggregator_bandwidth(la);
927 LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n",
928 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
929 speed, la->la_nports));
931 /* This aggregator is chosen if
932 * the partner has a better system priority
933 * or, the total aggregated speed is higher
934 * or, it is already the chosen aggregator
936 if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) <
937 LACP_SYS_PRI(best_la->la_partner)) ||
938 speed > best_speed ||
939 (speed == best_speed &&
940 la == lsc->lsc_active_aggregator)) {
946 KASSERT(best_la == NULL || best_la->la_nports > 0,
947 ("invalid aggregator refcnt"));
948 KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports),
949 ("invalid aggregator list"));
951 if (lsc->lsc_active_aggregator != best_la) {
952 LACP_DPRINTF((NULL, "active aggregator changed\n"));
953 LACP_DPRINTF((NULL, "old %s\n",
954 lacp_format_lagid_aggregator(lsc->lsc_active_aggregator,
957 LACP_DPRINTF((NULL, "active aggregator not changed\n"));
959 LACP_DPRINTF((NULL, "new %s\n",
960 lacp_format_lagid_aggregator(best_la, buf, sizeof(buf))));
962 if (lsc->lsc_active_aggregator != best_la) {
963 sc->sc_ifp->if_baudrate = best_speed;
964 lsc->lsc_active_aggregator = best_la;
965 lacp_update_portmap(lsc);
967 lacp_suppress_distributing(lsc, best_la);
973 * Updated the inactive portmap array with the new list of ports and
977 lacp_update_portmap(struct lacp_softc *lsc)
979 struct lacp_aggregator *la;
980 struct lacp_portmap *p;
981 struct lacp_port *lp;
985 newmap = lsc->lsc_activemap == 0 ? 1 : 0;
986 p = &lsc->lsc_pmap[newmap];
987 la = lsc->lsc_active_aggregator;
988 bzero(p, sizeof(struct lacp_portmap));
990 if (la != NULL && la->la_nports > 0) {
991 p->pm_count = la->la_nports;
993 TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q)
995 KASSERT(i == p->pm_count, ("Invalid port count"));
998 /* switch the active portmap over */
999 atomic_store_rel_int(&lsc->lsc_activemap, newmap);
1000 LACP_DPRINTF((NULL, "Set table %d with %d ports\n",
1002 lsc->lsc_pmap[lsc->lsc_activemap].pm_count));
1006 lacp_compose_key(struct lacp_port *lp)
1008 struct lagg_port *lgp = lp->lp_lagg;
1009 struct lagg_softc *sc = lgp->lp_softc;
1010 u_int media = lp->lp_media;
1013 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) {
1016 * non-aggregatable links should have unique keys.
1018 * XXX this isn't really unique as if_index is 16 bit.
1021 /* bit 0..14: (some bits of) if_index of this port */
1022 key = lp->lp_ifp->if_index;
1026 u_int subtype = IFM_SUBTYPE(media);
1028 KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type"));
1029 KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface"));
1031 /* bit 0..4: IFM_SUBTYPE */
1033 /* bit 5..14: (some bits of) if_index of lagg device */
1034 key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5);
1037 return (htons(key));
1041 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1043 char buf[LACP_LAGIDSTR_MAX+1];
1045 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1047 lacp_format_lagid(&la->la_actor, &la->la_partner,
1049 la->la_refcnt, la->la_refcnt + 1));
1051 KASSERT(la->la_refcnt > 0, ("refcount <= 0"));
1053 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount"));
1057 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1059 char buf[LACP_LAGIDSTR_MAX+1];
1061 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1063 lacp_format_lagid(&la->la_actor, &la->la_partner,
1065 la->la_refcnt, la->la_refcnt - 1));
1067 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt"));
1069 if (la->la_refcnt > 0) {
1073 KASSERT(la->la_refcnt == 0, ("refcount not zero"));
1074 KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active"));
1076 TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q);
1082 * lacp_aggregator_get: allocate an aggregator.
1085 static struct lacp_aggregator *
1086 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp)
1088 struct lacp_aggregator *la;
1090 la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT);
1094 TAILQ_INIT(&la->la_ports);
1096 TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q);
1103 * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port.
1107 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp)
1109 lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner);
1110 lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor);
1112 la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION;
1116 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr,
1117 const struct lacp_peerinfo *lpi_port)
1119 memset(lpi_aggr, 0, sizeof(*lpi_aggr));
1120 lpi_aggr->lip_systemid = lpi_port->lip_systemid;
1121 lpi_aggr->lip_key = lpi_port->lip_key;
1125 * lacp_aggregator_is_compatible: check if a port can join to an aggregator.
1129 lacp_aggregator_is_compatible(const struct lacp_aggregator *la,
1130 const struct lacp_port *lp)
1132 if (!(lp->lp_state & LACP_STATE_AGGREGATION) ||
1133 !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) {
1137 if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) {
1141 if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) {
1145 if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) {
1153 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a,
1154 const struct lacp_peerinfo *b)
1156 if (memcmp(&a->lip_systemid, &b->lip_systemid,
1157 sizeof(a->lip_systemid))) {
1161 if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) {
1169 lacp_port_enable(struct lacp_port *lp)
1171 lp->lp_state |= LACP_STATE_AGGREGATION;
1175 lacp_port_disable(struct lacp_port *lp)
1177 lacp_set_mux(lp, LACP_MUX_DETACHED);
1179 lp->lp_state &= ~LACP_STATE_AGGREGATION;
1180 lp->lp_selected = LACP_UNSELECTED;
1181 lacp_sm_rx_record_default(lp);
1182 lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION;
1183 lp->lp_state &= ~LACP_STATE_EXPIRED;
1187 * lacp_select: select an aggregator. create one if necessary.
1190 lacp_select(struct lacp_port *lp)
1192 struct lacp_softc *lsc = lp->lp_lsc;
1193 struct lacp_aggregator *la;
1194 char buf[LACP_LAGIDSTR_MAX+1];
1196 if (lp->lp_aggregator) {
1200 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1201 ("timer_wait_while still active"));
1203 LACP_DPRINTF((lp, "port lagid=%s\n",
1204 lacp_format_lagid(&lp->lp_actor, &lp->lp_partner,
1205 buf, sizeof(buf))));
1207 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
1208 if (lacp_aggregator_is_compatible(la, lp)) {
1214 la = lacp_aggregator_get(lsc, lp);
1216 LACP_DPRINTF((lp, "aggregator creation failed\n"));
1219 * will retry on the next tick.
1224 lacp_fill_aggregator_id(la, lp);
1225 LACP_DPRINTF((lp, "aggregator created\n"));
1227 LACP_DPRINTF((lp, "compatible aggregator found\n"));
1228 if (la->la_refcnt == LACP_MAX_PORTS)
1230 lacp_aggregator_addref(lsc, la);
1233 LACP_DPRINTF((lp, "aggregator lagid=%s\n",
1234 lacp_format_lagid(&la->la_actor, &la->la_partner,
1235 buf, sizeof(buf))));
1237 lp->lp_aggregator = la;
1238 lp->lp_selected = LACP_SELECTED;
1242 * lacp_unselect: finish unselect/detach process.
1246 lacp_unselect(struct lacp_port *lp)
1248 struct lacp_softc *lsc = lp->lp_lsc;
1249 struct lacp_aggregator *la = lp->lp_aggregator;
1251 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1252 ("timer_wait_while still active"));
1258 lp->lp_aggregator = NULL;
1259 lacp_aggregator_delref(lsc, la);
1265 lacp_sm_mux(struct lacp_port *lp)
1267 enum lacp_mux_state new_state;
1269 (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0;
1270 boolean_t p_collecting =
1271 (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0;
1272 enum lacp_selected selected = lp->lp_selected;
1273 struct lacp_aggregator *la;
1276 lacp_dprintf(lp, "%s: state %d\n", __func__, lp->lp_mux_state);
1279 la = lp->lp_aggregator;
1280 KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL,
1281 ("MUX not detached"));
1282 new_state = lp->lp_mux_state;
1283 switch (lp->lp_mux_state) {
1284 case LACP_MUX_DETACHED:
1285 if (selected != LACP_UNSELECTED) {
1286 new_state = LACP_MUX_WAITING;
1289 case LACP_MUX_WAITING:
1290 KASSERT(la->la_pending > 0 ||
1291 !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1292 ("timer_wait_while still active"));
1293 if (selected == LACP_SELECTED && la->la_pending == 0) {
1294 new_state = LACP_MUX_ATTACHED;
1295 } else if (selected == LACP_UNSELECTED) {
1296 new_state = LACP_MUX_DETACHED;
1299 case LACP_MUX_ATTACHED:
1300 if (selected == LACP_SELECTED && p_sync) {
1301 new_state = LACP_MUX_COLLECTING;
1302 } else if (selected != LACP_SELECTED) {
1303 new_state = LACP_MUX_DETACHED;
1306 case LACP_MUX_COLLECTING:
1307 if (selected == LACP_SELECTED && p_sync && p_collecting) {
1308 new_state = LACP_MUX_DISTRIBUTING;
1309 } else if (selected != LACP_SELECTED || !p_sync) {
1310 new_state = LACP_MUX_ATTACHED;
1313 case LACP_MUX_DISTRIBUTING:
1314 if (selected != LACP_SELECTED || !p_sync || !p_collecting) {
1315 new_state = LACP_MUX_COLLECTING;
1319 panic("%s: unknown state", __func__);
1322 if (lp->lp_mux_state == new_state) {
1326 lacp_set_mux(lp, new_state);
1331 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state)
1333 struct lacp_aggregator *la = lp->lp_aggregator;
1335 if (lp->lp_mux_state == new_state) {
1339 switch (new_state) {
1340 case LACP_MUX_DETACHED:
1341 lp->lp_state &= ~LACP_STATE_SYNC;
1342 lacp_disable_distributing(lp);
1343 lacp_disable_collecting(lp);
1344 lacp_sm_assert_ntt(lp);
1346 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) {
1347 KASSERT(la->la_pending > 0,
1348 ("timer_wait_while not active"));
1351 LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE);
1354 case LACP_MUX_WAITING:
1355 LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE,
1356 LACP_AGGREGATE_WAIT_TIME);
1359 case LACP_MUX_ATTACHED:
1360 lp->lp_state |= LACP_STATE_SYNC;
1361 lacp_disable_collecting(lp);
1362 lacp_sm_assert_ntt(lp);
1364 case LACP_MUX_COLLECTING:
1365 lacp_enable_collecting(lp);
1366 lacp_disable_distributing(lp);
1367 lacp_sm_assert_ntt(lp);
1369 case LACP_MUX_DISTRIBUTING:
1370 lacp_enable_distributing(lp);
1373 panic("%s: unknown state", __func__);
1376 LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state));
1378 lp->lp_mux_state = new_state;
1382 lacp_sm_mux_timer(struct lacp_port *lp)
1384 struct lacp_aggregator *la = lp->lp_aggregator;
1385 char buf[LACP_LAGIDSTR_MAX+1];
1387 KASSERT(la->la_pending > 0, ("no pending event"));
1389 LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__,
1390 lacp_format_lagid(&la->la_actor, &la->la_partner,
1392 la->la_pending, la->la_pending - 1));
1397 /* periodic transmit machine */
1400 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate)
1402 if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state,
1403 LACP_STATE_TIMEOUT)) {
1407 LACP_DPRINTF((lp, "partner timeout changed\n"));
1410 * FAST_PERIODIC -> SLOW_PERIODIC
1412 * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC
1414 * let lacp_sm_ptx_tx_schedule to update timeout.
1417 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1420 * if timeout has been shortened, assert NTT.
1423 if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) {
1424 lacp_sm_assert_ntt(lp);
1429 lacp_sm_ptx_tx_schedule(struct lacp_port *lp)
1433 if (!(lp->lp_state & LACP_STATE_ACTIVITY) &&
1434 !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) {
1440 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1444 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) {
1448 timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ?
1449 LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME;
1451 LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout);
1455 lacp_sm_ptx_timer(struct lacp_port *lp)
1457 lacp_sm_assert_ntt(lp);
1461 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du)
1466 * check LACP_DISABLED first
1469 if (!(lp->lp_state & LACP_STATE_AGGREGATION)) {
1474 * check loopback condition.
1477 if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid,
1478 &lp->lp_actor.lip_systemid)) {
1483 * EXPIRED, DEFAULTED, CURRENT -> CURRENT
1486 lacp_sm_rx_update_selected(lp, du);
1487 lacp_sm_rx_update_ntt(lp, du);
1488 lacp_sm_rx_record_pdu(lp, du);
1490 timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ?
1491 LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME;
1492 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout);
1494 lp->lp_state &= ~LACP_STATE_EXPIRED;
1497 * kick transmit machine without waiting the next tick.
1504 lacp_sm_rx_set_expired(struct lacp_port *lp)
1506 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1507 lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT;
1508 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME);
1509 lp->lp_state |= LACP_STATE_EXPIRED;
1513 lacp_sm_rx_timer(struct lacp_port *lp)
1515 if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) {
1516 /* CURRENT -> EXPIRED */
1517 LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__));
1518 lacp_sm_rx_set_expired(lp);
1520 /* EXPIRED -> DEFAULTED */
1521 LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__));
1522 lacp_sm_rx_update_default_selected(lp);
1523 lacp_sm_rx_record_default(lp);
1524 lp->lp_state &= ~LACP_STATE_EXPIRED;
1529 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du)
1533 char buf[LACP_STATESTR_MAX+1];
1537 oldpstate = lp->lp_partner.lip_state;
1539 active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY)
1540 || ((lp->lp_state & LACP_STATE_ACTIVITY) &&
1541 (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY));
1543 lp->lp_partner = du->ldu_actor;
1545 ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1546 LACP_STATE_AGGREGATION) &&
1547 !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner))
1548 || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) {
1551 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1554 lp->lp_state &= ~LACP_STATE_DEFAULTED;
1556 if (oldpstate != lp->lp_partner.lip_state) {
1557 LACP_DPRINTF((lp, "old pstate %s\n",
1558 lacp_format_state(oldpstate, buf, sizeof(buf))));
1559 LACP_DPRINTF((lp, "new pstate %s\n",
1560 lacp_format_state(lp->lp_partner.lip_state, buf,
1564 lacp_sm_ptx_update_timeout(lp, oldpstate);
1568 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du)
1573 if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) ||
1574 !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1575 LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) {
1576 LACP_DPRINTF((lp, "%s: assert ntt\n", __func__));
1577 lacp_sm_assert_ntt(lp);
1582 lacp_sm_rx_record_default(struct lacp_port *lp)
1588 oldpstate = lp->lp_partner.lip_state;
1589 lp->lp_partner = lacp_partner_admin;
1590 lp->lp_state |= LACP_STATE_DEFAULTED;
1591 lacp_sm_ptx_update_timeout(lp, oldpstate);
1595 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp,
1596 const struct lacp_peerinfo *info)
1601 if (lacp_compare_peerinfo(&lp->lp_partner, info) ||
1602 !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state,
1603 LACP_STATE_AGGREGATION)) {
1604 lp->lp_selected = LACP_UNSELECTED;
1605 /* mux machine will clean up lp->lp_aggregator */
1610 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du)
1615 lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor);
1619 lacp_sm_rx_update_default_selected(struct lacp_port *lp)
1624 lacp_sm_rx_update_selected_from_peerinfo(lp, &lacp_partner_admin);
1627 /* transmit machine */
1630 lacp_sm_tx(struct lacp_port *lp)
1634 if (!(lp->lp_state & LACP_STATE_AGGREGATION)
1636 || (!(lp->lp_state & LACP_STATE_ACTIVITY)
1637 && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY))
1640 lp->lp_flags &= ~LACP_PORT_NTT;
1643 if (!(lp->lp_flags & LACP_PORT_NTT)) {
1647 /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */
1648 if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent,
1649 (3 / LACP_FAST_PERIODIC_TIME)) == 0) {
1650 LACP_DPRINTF((lp, "rate limited pdu\n"));
1654 error = lacp_xmit_lacpdu(lp);
1657 lp->lp_flags &= ~LACP_PORT_NTT;
1659 LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n",
1665 lacp_sm_assert_ntt(struct lacp_port *lp)
1668 lp->lp_flags |= LACP_PORT_NTT;
1672 lacp_run_timers(struct lacp_port *lp)
1676 for (i = 0; i < LACP_NTIMER; i++) {
1677 KASSERT(lp->lp_timer[i] >= 0,
1678 ("invalid timer value %d", lp->lp_timer[i]));
1679 if (lp->lp_timer[i] == 0) {
1681 } else if (--lp->lp_timer[i] <= 0) {
1682 if (lacp_timer_funcs[i]) {
1683 (*lacp_timer_funcs[i])(lp);
1690 lacp_marker_input(struct lacp_port *lp, struct mbuf *m)
1692 struct lacp_softc *lsc = lp->lp_lsc;
1693 struct lagg_port *lgp = lp->lp_lagg;
1694 struct lacp_port *lp2;
1695 struct markerdu *mdu;
1699 if (m->m_pkthdr.len != sizeof(*mdu)) {
1703 if ((m->m_flags & M_MCAST) == 0) {
1707 if (m->m_len < sizeof(*mdu)) {
1708 m = m_pullup(m, sizeof(*mdu));
1714 mdu = mtod(m, struct markerdu *);
1716 if (memcmp(&mdu->mdu_eh.ether_dhost,
1717 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
1721 if (mdu->mdu_sph.sph_version != 1) {
1725 switch (mdu->mdu_tlv.tlv_type) {
1726 case MARKER_TYPE_INFO:
1727 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
1728 marker_info_tlv_template, TRUE)) {
1731 mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE;
1732 memcpy(&mdu->mdu_eh.ether_dhost,
1733 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN);
1734 memcpy(&mdu->mdu_eh.ether_shost,
1735 lgp->lp_lladdr, ETHER_ADDR_LEN);
1736 error = lagg_enqueue(lp->lp_ifp, m);
1739 case MARKER_TYPE_RESPONSE:
1740 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
1741 marker_response_tlv_template, TRUE)) {
1744 LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n",
1745 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system,
1746 ":", ntohl(mdu->mdu_info.mi_rq_xid)));
1748 /* Verify that it is the last marker we sent out */
1749 if (memcmp(&mdu->mdu_info, &lp->lp_marker,
1750 sizeof(struct lacp_markerinfo)))
1754 lp->lp_flags &= ~LACP_PORT_MARK;
1756 if (lsc->lsc_suppress_distributing) {
1757 /* Check if any ports are waiting for a response */
1758 LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) {
1759 if (lp2->lp_flags & LACP_PORT_MARK) {
1766 /* All interface queues are clear */
1767 LACP_DPRINTF((NULL, "queue flush complete\n"));
1768 lsc->lsc_suppress_distributing = FALSE;
1782 LACP_DPRINTF((lp, "bad marker frame\n"));
1788 tlv_check(const void *p, size_t size, const struct tlvhdr *tlv,
1789 const struct tlv_template *tmpl, boolean_t check_type)
1791 while (/* CONSTCOND */ 1) {
1792 if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) {
1795 if ((check_type && tlv->tlv_type != tmpl->tmpl_type) ||
1796 tlv->tlv_length != tmpl->tmpl_length) {
1799 if (tmpl->tmpl_type == 0) {
1802 tlv = (const struct tlvhdr *)
1803 ((const char *)tlv + tlv->tlv_length);
1812 lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen)
1814 snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X",
1826 lacp_format_systemid(const struct lacp_systemid *sysid,
1827 char *buf, size_t buflen)
1829 char macbuf[LACP_MACSTR_MAX+1];
1831 snprintf(buf, buflen, "%04X,%s",
1832 ntohs(sysid->lsi_prio),
1833 lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf)));
1839 lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen)
1841 snprintf(buf, buflen, "%04X,%04X",
1842 ntohs(portid->lpi_prio),
1843 ntohs(portid->lpi_portno));
1849 lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen)
1851 char sysid[LACP_SYSTEMIDSTR_MAX+1];
1852 char portid[LACP_PORTIDSTR_MAX+1];
1854 snprintf(buf, buflen, "(%s,%04X,%s)",
1855 lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)),
1856 ntohs(peer->lip_key),
1857 lacp_format_portid(&peer->lip_portid, portid, sizeof(portid)));
1863 lacp_format_lagid(const struct lacp_peerinfo *a,
1864 const struct lacp_peerinfo *b, char *buf, size_t buflen)
1866 char astr[LACP_PARTNERSTR_MAX+1];
1867 char bstr[LACP_PARTNERSTR_MAX+1];
1871 * there's a convention to display small numbered peer
1875 if (lacp_compare_peerinfo(a, b) > 0) {
1876 const struct lacp_peerinfo *t;
1884 snprintf(buf, buflen, "[%s,%s]",
1885 lacp_format_partner(a, astr, sizeof(astr)),
1886 lacp_format_partner(b, bstr, sizeof(bstr)));
1892 lacp_format_lagid_aggregator(const struct lacp_aggregator *la,
1893 char *buf, size_t buflen)
1899 return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen));
1903 lacp_format_state(uint8_t state, char *buf, size_t buflen)
1905 snprintf(buf, buflen, "%b", state, LACP_STATE_BITS);
1910 lacp_dump_lacpdu(const struct lacpdu *du)
1912 char buf[LACP_PARTNERSTR_MAX+1];
1913 char buf2[LACP_STATESTR_MAX+1];
1915 printf("actor=%s\n",
1916 lacp_format_partner(&du->ldu_actor, buf, sizeof(buf)));
1917 printf("actor.state=%s\n",
1918 lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2)));
1919 printf("partner=%s\n",
1920 lacp_format_partner(&du->ldu_partner, buf, sizeof(buf)));
1921 printf("partner.state=%s\n",
1922 lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2)));
1924 printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay));
1928 lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...)
1933 printf("%s: ", lp->lp_ifp->if_xname);