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>
35 #include <sys/eventhandler.h>
37 #include <sys/systm.h>
38 #include <sys/malloc.h>
39 #include <sys/kernel.h> /* hz */
40 #include <sys/socket.h> /* for net/if.h */
41 #include <sys/sockio.h>
42 #include <sys/sysctl.h>
43 #include <machine/stdarg.h>
45 #include <sys/rwlock.h>
46 #include <sys/taskqueue.h>
49 #include <net/if_var.h>
50 #include <net/if_dl.h>
51 #include <net/ethernet.h>
52 #include <net/if_media.h>
53 #include <net/if_types.h>
55 #include <net/if_lagg.h>
56 #include <net/ieee8023ad_lacp.h>
59 * actor system priority and port priority.
60 * XXX should be configurable.
63 #define LACP_SYSTEM_PRIO 0x8000
64 #define LACP_PORT_PRIO 0x8000
66 const uint8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] =
67 { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 };
69 static const struct tlv_template lacp_info_tlv_template[] = {
70 { LACP_TYPE_ACTORINFO,
71 sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
72 { LACP_TYPE_PARTNERINFO,
73 sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
74 { LACP_TYPE_COLLECTORINFO,
75 sizeof(struct tlvhdr) + sizeof(struct lacp_collectorinfo) },
79 static const struct tlv_template marker_info_tlv_template[] = {
81 sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
85 static const struct tlv_template marker_response_tlv_template[] = {
86 { MARKER_TYPE_RESPONSE,
87 sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
91 typedef void (*lacp_timer_func_t)(struct lacp_port *);
93 static void lacp_fill_actorinfo(struct lacp_port *, struct lacp_peerinfo *);
94 static void lacp_fill_markerinfo(struct lacp_port *,
95 struct lacp_markerinfo *);
97 static uint64_t lacp_aggregator_bandwidth(struct lacp_aggregator *);
98 static void lacp_suppress_distributing(struct lacp_softc *,
99 struct lacp_aggregator *);
100 static void lacp_transit_expire(void *);
101 static void lacp_update_portmap(struct lacp_softc *);
102 static void lacp_select_active_aggregator(struct lacp_softc *);
103 static uint16_t lacp_compose_key(struct lacp_port *);
104 static int tlv_check(const void *, size_t, const struct tlvhdr *,
105 const struct tlv_template *, boolean_t);
106 static void lacp_tick(void *);
108 static void lacp_fill_aggregator_id(struct lacp_aggregator *,
109 const struct lacp_port *);
110 static void lacp_fill_aggregator_id_peer(struct lacp_peerinfo *,
111 const struct lacp_peerinfo *);
112 static int lacp_aggregator_is_compatible(const struct lacp_aggregator *,
113 const struct lacp_port *);
114 static int lacp_peerinfo_is_compatible(const struct lacp_peerinfo *,
115 const struct lacp_peerinfo *);
117 static struct lacp_aggregator *lacp_aggregator_get(struct lacp_softc *,
119 static void lacp_aggregator_addref(struct lacp_softc *,
120 struct lacp_aggregator *);
121 static void lacp_aggregator_delref(struct lacp_softc *,
122 struct lacp_aggregator *);
124 /* receive machine */
126 static int lacp_pdu_input(struct lacp_port *, struct mbuf *);
127 static int lacp_marker_input(struct lacp_port *, struct mbuf *);
128 static void lacp_sm_rx(struct lacp_port *, const struct lacpdu *);
129 static void lacp_sm_rx_timer(struct lacp_port *);
130 static void lacp_sm_rx_set_expired(struct lacp_port *);
131 static void lacp_sm_rx_update_ntt(struct lacp_port *,
132 const struct lacpdu *);
133 static void lacp_sm_rx_record_pdu(struct lacp_port *,
134 const struct lacpdu *);
135 static void lacp_sm_rx_update_selected(struct lacp_port *,
136 const struct lacpdu *);
137 static void lacp_sm_rx_record_default(struct lacp_port *);
138 static void lacp_sm_rx_update_default_selected(struct lacp_port *);
139 static void lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *,
140 const struct lacp_peerinfo *);
144 static void lacp_sm_mux(struct lacp_port *);
145 static void lacp_set_mux(struct lacp_port *, enum lacp_mux_state);
146 static void lacp_sm_mux_timer(struct lacp_port *);
148 /* periodic transmit machine */
150 static void lacp_sm_ptx_update_timeout(struct lacp_port *, uint8_t);
151 static void lacp_sm_ptx_tx_schedule(struct lacp_port *);
152 static void lacp_sm_ptx_timer(struct lacp_port *);
154 /* transmit machine */
156 static void lacp_sm_tx(struct lacp_port *);
157 static void lacp_sm_assert_ntt(struct lacp_port *);
159 static void lacp_run_timers(struct lacp_port *);
160 static int lacp_compare_peerinfo(const struct lacp_peerinfo *,
161 const struct lacp_peerinfo *);
162 static int lacp_compare_systemid(const struct lacp_systemid *,
163 const struct lacp_systemid *);
164 static void lacp_port_enable(struct lacp_port *);
165 static void lacp_port_disable(struct lacp_port *);
166 static void lacp_select(struct lacp_port *);
167 static void lacp_unselect(struct lacp_port *);
168 static void lacp_disable_collecting(struct lacp_port *);
169 static void lacp_enable_collecting(struct lacp_port *);
170 static void lacp_disable_distributing(struct lacp_port *);
171 static void lacp_enable_distributing(struct lacp_port *);
172 static int lacp_xmit_lacpdu(struct lacp_port *);
173 static int lacp_xmit_marker(struct lacp_port *);
177 static void lacp_dump_lacpdu(const struct lacpdu *);
178 static const char *lacp_format_partner(const struct lacp_peerinfo *, char *,
180 static const char *lacp_format_lagid(const struct lacp_peerinfo *,
181 const struct lacp_peerinfo *, char *, size_t);
182 static const char *lacp_format_lagid_aggregator(const struct lacp_aggregator *,
184 static const char *lacp_format_state(uint8_t, char *, size_t);
185 static const char *lacp_format_mac(const uint8_t *, char *, size_t);
186 static const char *lacp_format_systemid(const struct lacp_systemid *, char *,
188 static const char *lacp_format_portid(const struct lacp_portid *, char *,
190 static void lacp_dprintf(const struct lacp_port *, const char *, ...)
191 __attribute__((__format__(__printf__, 2, 3)));
193 static int lacp_debug = 0;
194 SYSCTL_NODE(_net_link_lagg, OID_AUTO, lacp, CTLFLAG_RD, 0, "ieee802.3ad");
195 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, debug, CTLFLAG_RW | CTLFLAG_TUN,
196 &lacp_debug, 0, "Enable LACP debug logging (1=debug, 2=trace)");
197 TUNABLE_INT("net.link.lagg.lacp.debug", &lacp_debug);
199 #define LACP_DPRINTF(a) if (lacp_debug & 0x01) { lacp_dprintf a ; }
200 #define LACP_TRACE(a) if (lacp_debug & 0x02) { lacp_dprintf(a,"%s\n",__func__); }
201 #define LACP_TPRINTF(a) if (lacp_debug & 0x04) { lacp_dprintf a ; }
204 * partner administration variables.
205 * XXX should be configurable.
208 static const struct lacp_peerinfo lacp_partner_admin_optimistic = {
209 .lip_systemid = { .lsi_prio = 0xffff },
210 .lip_portid = { .lpi_prio = 0xffff },
211 .lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION |
212 LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING,
215 static const struct lacp_peerinfo lacp_partner_admin_strict = {
216 .lip_systemid = { .lsi_prio = 0xffff },
217 .lip_portid = { .lpi_prio = 0xffff },
221 static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = {
222 [LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer,
223 [LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer,
224 [LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer,
228 lacp_input(struct lagg_port *lgp, struct mbuf *m)
230 struct lacp_port *lp = LACP_PORT(lgp);
233 if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) {
238 m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype);
240 case SLOWPROTOCOLS_SUBTYPE_LACP:
241 lacp_pdu_input(lp, m);
244 case SLOWPROTOCOLS_SUBTYPE_MARKER:
245 lacp_marker_input(lp, m);
249 /* Not a subtype we are interested in */
254 * lacp_pdu_input: process lacpdu
257 lacp_pdu_input(struct lacp_port *lp, struct mbuf *m)
259 struct lacp_softc *lsc = lp->lp_lsc;
263 if (m->m_pkthdr.len != sizeof(*du)) {
267 if ((m->m_flags & M_MCAST) == 0) {
271 if (m->m_len < sizeof(*du)) {
272 m = m_pullup(m, sizeof(*du));
278 du = mtod(m, struct lacpdu *);
280 if (memcmp(&du->ldu_eh.ether_dhost,
281 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
286 * ignore the version for compatibility with
287 * the future protocol revisions.
290 if (du->ldu_sph.sph_version != 1) {
296 * ignore tlv types for compatibility with
297 * the future protocol revisions.
299 if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor,
300 lacp_info_tlv_template, FALSE)) {
304 if (lacp_debug > 0) {
305 lacp_dprintf(lp, "lacpdu receive\n");
306 lacp_dump_lacpdu(du);
309 if ((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_rx_test) {
310 LACP_TPRINTF((lp, "Dropping RX PDU\n"));
327 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info)
329 struct lagg_port *lgp = lp->lp_lagg;
330 struct lagg_softc *sc = lgp->lp_softc;
332 info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO);
333 memcpy(&info->lip_systemid.lsi_mac,
334 IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
335 info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO);
336 info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index);
337 info->lip_state = lp->lp_state;
341 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info)
343 struct ifnet *ifp = lp->lp_ifp;
345 /* Fill in the port index and system id (encoded as the MAC) */
346 info->mi_rq_port = htons(ifp->if_index);
347 memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN);
348 info->mi_rq_xid = htonl(0);
352 lacp_xmit_lacpdu(struct lacp_port *lp)
354 struct lagg_port *lgp = lp->lp_lagg;
359 LACP_LOCK_ASSERT(lp->lp_lsc);
361 m = m_gethdr(M_NOWAIT, MT_DATA);
365 m->m_len = m->m_pkthdr.len = sizeof(*du);
367 du = mtod(m, struct lacpdu *);
368 memset(du, 0, sizeof(*du));
370 memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
372 memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
373 du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW);
375 du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP;
376 du->ldu_sph.sph_version = 1;
378 TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor));
379 du->ldu_actor = lp->lp_actor;
381 TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO,
382 sizeof(du->ldu_partner));
383 du->ldu_partner = lp->lp_partner;
385 TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO,
386 sizeof(du->ldu_collector));
387 du->ldu_collector.lci_maxdelay = 0;
389 if (lacp_debug > 0) {
390 lacp_dprintf(lp, "lacpdu transmit\n");
391 lacp_dump_lacpdu(du);
394 m->m_flags |= M_MCAST;
397 * XXX should use higher priority queue.
398 * otherwise network congestion can break aggregation.
401 error = lagg_enqueue(lp->lp_ifp, m);
406 lacp_xmit_marker(struct lacp_port *lp)
408 struct lagg_port *lgp = lp->lp_lagg;
410 struct markerdu *mdu;
413 LACP_LOCK_ASSERT(lp->lp_lsc);
415 m = m_gethdr(M_NOWAIT, MT_DATA);
419 m->m_len = m->m_pkthdr.len = sizeof(*mdu);
421 mdu = mtod(m, struct markerdu *);
422 memset(mdu, 0, sizeof(*mdu));
424 memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
426 memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
427 mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW);
429 mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER;
430 mdu->mdu_sph.sph_version = 1;
432 /* Bump the transaction id and copy over the marker info */
433 lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1);
434 TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info));
435 mdu->mdu_info = lp->lp_marker;
437 LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n",
438 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":",
439 ntohl(mdu->mdu_info.mi_rq_xid)));
441 m->m_flags |= M_MCAST;
442 error = lagg_enqueue(lp->lp_ifp, m);
447 lacp_linkstate(struct lagg_port *lgp)
449 struct lacp_port *lp = LACP_PORT(lgp);
450 struct lacp_softc *lsc = lp->lp_lsc;
451 struct ifnet *ifp = lgp->lp_ifp;
452 struct ifmediareq ifmr;
458 bzero((char *)&ifmr, sizeof(ifmr));
459 error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr);
464 media = ifmr.ifm_active;
465 LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, "
466 "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER,
467 (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP));
468 old_state = lp->lp_state;
469 old_key = lp->lp_key;
471 lp->lp_media = media;
473 * If the port is not an active full duplex Ethernet link then it can
476 if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 ||
477 ifp->if_link_state != LINK_STATE_UP) {
478 lacp_port_disable(lp);
480 lacp_port_enable(lp);
482 lp->lp_key = lacp_compose_key(lp);
484 if (old_state != lp->lp_state || old_key != lp->lp_key) {
485 LACP_DPRINTF((lp, "-> UNSELECTED\n"));
486 lp->lp_selected = LACP_UNSELECTED;
494 struct lacp_softc *lsc = arg;
495 struct lacp_port *lp;
497 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
498 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0)
506 lacp_sm_ptx_tx_schedule(lp);
508 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
512 lacp_port_create(struct lagg_port *lgp)
514 struct lagg_softc *sc = lgp->lp_softc;
515 struct lacp_softc *lsc = LACP_SOFTC(sc);
516 struct lacp_port *lp;
517 struct ifnet *ifp = lgp->lp_ifp;
518 struct sockaddr_dl sdl;
519 struct ifmultiaddr *rifma = NULL;
522 boolean_t active = TRUE; /* XXX should be configurable */
523 boolean_t fast = FALSE; /* XXX should be configurable */
525 link_init_sdl(ifp, (struct sockaddr *)&sdl, IFT_ETHER);
526 sdl.sdl_alen = ETHER_ADDR_LEN;
528 bcopy(ðermulticastaddr_slowprotocols,
529 LLADDR(&sdl), ETHER_ADDR_LEN);
530 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
532 printf("%s: ADDMULTI failed on %s\n", __func__, lgp->lp_ifname);
536 lp = malloc(sizeof(struct lacp_port),
537 M_DEVBUF, M_NOWAIT|M_ZERO);
542 lgp->lp_psc = (caddr_t)lp;
548 LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next);
550 lacp_fill_actorinfo(lp, &lp->lp_actor);
551 lacp_fill_markerinfo(lp, &lp->lp_marker);
553 (active ? LACP_STATE_ACTIVITY : 0) |
554 (fast ? LACP_STATE_TIMEOUT : 0);
555 lp->lp_aggregator = NULL;
556 lacp_sm_rx_set_expired(lp);
564 lacp_port_destroy(struct lagg_port *lgp)
566 struct lacp_port *lp = LACP_PORT(lgp);
567 struct lacp_softc *lsc = lp->lp_lsc;
571 for (i = 0; i < LACP_NTIMER; i++) {
572 LACP_TIMER_DISARM(lp, i);
575 lacp_disable_collecting(lp);
576 lacp_disable_distributing(lp);
579 /* The address may have already been removed by if_purgemaddrs() */
580 if (!lgp->lp_detaching)
581 if_delmulti_ifma(lp->lp_ifma);
583 LIST_REMOVE(lp, lp_next);
589 lacp_req(struct lagg_softc *sc, caddr_t data)
591 struct lacp_opreq *req = (struct lacp_opreq *)data;
592 struct lacp_softc *lsc = LACP_SOFTC(sc);
593 struct lacp_aggregator *la;
595 bzero(req, sizeof(struct lacp_opreq));
598 * If the LACP softc is NULL, return with the opreq structure full of
599 * zeros. It is normal for the softc to be NULL while the lagg is
605 la = lsc->lsc_active_aggregator;
608 req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio);
609 memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac,
611 req->actor_key = ntohs(la->la_actor.lip_key);
612 req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio);
613 req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno);
614 req->actor_state = la->la_actor.lip_state;
616 req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio);
617 memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac,
619 req->partner_key = ntohs(la->la_partner.lip_key);
620 req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio);
621 req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno);
622 req->partner_state = la->la_partner.lip_state;
628 lacp_portreq(struct lagg_port *lgp, caddr_t data)
630 struct lacp_opreq *req = (struct lacp_opreq *)data;
631 struct lacp_port *lp = LACP_PORT(lgp);
632 struct lacp_softc *lsc = lp->lp_lsc;
635 req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio);
636 memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac,
638 req->actor_key = ntohs(lp->lp_actor.lip_key);
639 req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio);
640 req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno);
641 req->actor_state = lp->lp_actor.lip_state;
643 req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio);
644 memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac,
646 req->partner_key = ntohs(lp->lp_partner.lip_key);
647 req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio);
648 req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno);
649 req->partner_state = lp->lp_partner.lip_state;
654 lacp_disable_collecting(struct lacp_port *lp)
656 LACP_DPRINTF((lp, "collecting disabled\n"));
657 lp->lp_state &= ~LACP_STATE_COLLECTING;
661 lacp_enable_collecting(struct lacp_port *lp)
663 LACP_DPRINTF((lp, "collecting enabled\n"));
664 lp->lp_state |= LACP_STATE_COLLECTING;
668 lacp_disable_distributing(struct lacp_port *lp)
670 struct lacp_aggregator *la = lp->lp_aggregator;
671 struct lacp_softc *lsc = lp->lp_lsc;
672 struct lagg_softc *sc = lsc->lsc_softc;
673 char buf[LACP_LAGIDSTR_MAX+1];
675 LACP_LOCK_ASSERT(lsc);
677 if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) {
681 KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports"));
682 KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports));
683 KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid"));
685 LACP_DPRINTF((lp, "disable distributing on aggregator %s, "
687 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
688 la->la_nports, la->la_nports - 1));
690 TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q);
692 sc->sc_active = la->la_nports;
694 if (lsc->lsc_active_aggregator == la) {
695 lacp_suppress_distributing(lsc, la);
696 lacp_select_active_aggregator(lsc);
697 /* regenerate the port map, the active aggregator has changed */
698 lacp_update_portmap(lsc);
701 lp->lp_state &= ~LACP_STATE_DISTRIBUTING;
705 lacp_enable_distributing(struct lacp_port *lp)
707 struct lacp_aggregator *la = lp->lp_aggregator;
708 struct lacp_softc *lsc = lp->lp_lsc;
709 struct lagg_softc *sc = lsc->lsc_softc;
710 char buf[LACP_LAGIDSTR_MAX+1];
712 LACP_LOCK_ASSERT(lsc);
714 if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) {
718 LACP_DPRINTF((lp, "enable distributing on aggregator %s, "
720 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
721 la->la_nports, la->la_nports + 1));
723 KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid"));
724 TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q);
726 sc->sc_active = la->la_nports;
728 lp->lp_state |= LACP_STATE_DISTRIBUTING;
730 if (lsc->lsc_active_aggregator == la) {
731 lacp_suppress_distributing(lsc, la);
732 lacp_update_portmap(lsc);
734 /* try to become the active aggregator */
735 lacp_select_active_aggregator(lsc);
739 lacp_transit_expire(void *vp)
741 struct lacp_softc *lsc = vp;
743 LACP_LOCK_ASSERT(lsc);
747 lsc->lsc_suppress_distributing = FALSE;
751 lacp_attach_sysctl(struct lacp_softc *lsc, struct sysctl_oid *p_oid)
753 struct lagg_softc *sc = lsc->lsc_softc;
755 SYSCTL_ADD_UINT(&sc->ctx, SYSCTL_CHILDREN(p_oid), OID_AUTO,
758 &lsc->lsc_strict_mode,
759 lsc->lsc_strict_mode,
760 "Enable LACP strict mode");
764 lacp_attach_sysctl_debug(struct lacp_softc *lsc, struct sysctl_oid *p_oid)
766 struct lagg_softc *sc = lsc->lsc_softc;
767 struct sysctl_oid *oid;
769 /* Create a child of the parent lagg interface */
770 oid = SYSCTL_ADD_NODE(&sc->ctx, SYSCTL_CHILDREN(p_oid),
771 OID_AUTO, "debug", CTLFLAG_RD, NULL, "DEBUG");
773 SYSCTL_ADD_UINT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
776 &lsc->lsc_debug.lsc_rx_test,
777 lsc->lsc_debug.lsc_rx_test,
778 "Bitmap of if_dunit entries to drop RX frames for");
779 SYSCTL_ADD_UINT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
782 &lsc->lsc_debug.lsc_tx_test,
783 lsc->lsc_debug.lsc_tx_test,
784 "Bitmap of if_dunit entries to drop TX frames for");
788 lacp_attach(struct lagg_softc *sc)
790 struct lacp_softc *lsc;
791 struct sysctl_oid *oid;
793 lsc = malloc(sizeof(struct lacp_softc),
794 M_DEVBUF, M_NOWAIT|M_ZERO);
798 sc->sc_psc = (caddr_t)lsc;
801 lsc->lsc_hashkey = arc4random();
802 lsc->lsc_active_aggregator = NULL;
803 lsc->lsc_strict_mode = 1;
805 TAILQ_INIT(&lsc->lsc_aggregators);
806 LIST_INIT(&lsc->lsc_ports);
808 /* Create a child of the parent lagg interface */
809 oid = SYSCTL_ADD_NODE(&sc->ctx, SYSCTL_CHILDREN(sc->sc_oid),
810 OID_AUTO, "lacp", CTLFLAG_RD, NULL, "LACP");
812 /* Attach sysctl nodes */
813 lacp_attach_sysctl(lsc, oid);
814 lacp_attach_sysctl_debug(lsc, oid);
816 callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0);
817 callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0);
819 /* if the lagg is already up then do the same */
820 if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING)
827 lacp_detach(struct lagg_softc *sc)
829 struct lacp_softc *lsc = LACP_SOFTC(sc);
831 KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators),
832 ("aggregators still active"));
833 KASSERT(lsc->lsc_active_aggregator == NULL,
834 ("aggregator still attached"));
837 callout_drain(&lsc->lsc_transit_callout);
838 callout_drain(&lsc->lsc_callout);
840 LACP_LOCK_DESTROY(lsc);
846 lacp_init(struct lagg_softc *sc)
848 struct lacp_softc *lsc = LACP_SOFTC(sc);
851 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
856 lacp_stop(struct lagg_softc *sc)
858 struct lacp_softc *lsc = LACP_SOFTC(sc);
861 callout_stop(&lsc->lsc_transit_callout);
862 callout_stop(&lsc->lsc_callout);
867 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
869 struct lacp_softc *lsc = LACP_SOFTC(sc);
870 struct lacp_portmap *pm;
871 struct lacp_port *lp;
874 if (__predict_false(lsc->lsc_suppress_distributing)) {
875 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__));
879 pm = &lsc->lsc_pmap[lsc->lsc_activemap];
880 if (pm->pm_count == 0) {
881 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__));
885 if (sc->use_flowid && (m->m_flags & M_FLOWID))
886 hash = m->m_pkthdr.flowid >> sc->flowid_shift;
888 hash = lagg_hashmbuf(sc, m, lsc->lsc_hashkey);
889 hash %= pm->pm_count;
890 lp = pm->pm_map[hash];
892 KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0,
893 ("aggregated port is not distributing"));
895 return (lp->lp_lagg);
898 * lacp_suppress_distributing: drop transmit packets for a while
899 * to preserve packet ordering.
903 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la)
905 struct lacp_port *lp;
907 if (lsc->lsc_active_aggregator != la) {
913 lsc->lsc_suppress_distributing = TRUE;
915 /* send a marker frame down each port to verify the queues are empty */
916 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
917 lp->lp_flags |= LACP_PORT_MARK;
918 lacp_xmit_marker(lp);
921 /* set a timeout for the marker frames */
922 callout_reset(&lsc->lsc_transit_callout,
923 LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc);
927 lacp_compare_peerinfo(const struct lacp_peerinfo *a,
928 const struct lacp_peerinfo *b)
930 return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state)));
934 lacp_compare_systemid(const struct lacp_systemid *a,
935 const struct lacp_systemid *b)
937 return (memcmp(a, b, sizeof(*a)));
942 lacp_compare_portid(const struct lacp_portid *a,
943 const struct lacp_portid *b)
945 return (memcmp(a, b, sizeof(*a)));
950 lacp_aggregator_bandwidth(struct lacp_aggregator *la)
952 struct lacp_port *lp;
955 lp = TAILQ_FIRST(&la->la_ports);
960 speed = ifmedia_baudrate(lp->lp_media);
961 speed *= la->la_nports;
963 LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n",
964 lp->lp_media, la->la_nports));
971 * lacp_select_active_aggregator: select an aggregator to be used to transmit
972 * packets from lagg(4) interface.
976 lacp_select_active_aggregator(struct lacp_softc *lsc)
978 struct lacp_aggregator *la;
979 struct lacp_aggregator *best_la = NULL;
980 uint64_t best_speed = 0;
981 char buf[LACP_LAGIDSTR_MAX+1];
985 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
988 if (la->la_nports == 0) {
992 speed = lacp_aggregator_bandwidth(la);
993 LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n",
994 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
995 speed, la->la_nports));
997 /* This aggregator is chosen if
998 * the partner has a better system priority
999 * or, the total aggregated speed is higher
1000 * or, it is already the chosen aggregator
1002 if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) <
1003 LACP_SYS_PRI(best_la->la_partner)) ||
1004 speed > best_speed ||
1005 (speed == best_speed &&
1006 la == lsc->lsc_active_aggregator)) {
1012 KASSERT(best_la == NULL || best_la->la_nports > 0,
1013 ("invalid aggregator refcnt"));
1014 KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports),
1015 ("invalid aggregator list"));
1017 if (lsc->lsc_active_aggregator != best_la) {
1018 LACP_DPRINTF((NULL, "active aggregator changed\n"));
1019 LACP_DPRINTF((NULL, "old %s\n",
1020 lacp_format_lagid_aggregator(lsc->lsc_active_aggregator,
1021 buf, sizeof(buf))));
1023 LACP_DPRINTF((NULL, "active aggregator not changed\n"));
1025 LACP_DPRINTF((NULL, "new %s\n",
1026 lacp_format_lagid_aggregator(best_la, buf, sizeof(buf))));
1028 if (lsc->lsc_active_aggregator != best_la) {
1029 lsc->lsc_active_aggregator = best_la;
1030 lacp_update_portmap(lsc);
1032 lacp_suppress_distributing(lsc, best_la);
1038 * Updated the inactive portmap array with the new list of ports and
1042 lacp_update_portmap(struct lacp_softc *lsc)
1044 struct lagg_softc *sc = lsc->lsc_softc;
1045 struct lacp_aggregator *la;
1046 struct lacp_portmap *p;
1047 struct lacp_port *lp;
1052 newmap = lsc->lsc_activemap == 0 ? 1 : 0;
1053 p = &lsc->lsc_pmap[newmap];
1054 la = lsc->lsc_active_aggregator;
1056 bzero(p, sizeof(struct lacp_portmap));
1058 if (la != NULL && la->la_nports > 0) {
1059 p->pm_count = la->la_nports;
1061 TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q)
1062 p->pm_map[i++] = lp;
1063 KASSERT(i == p->pm_count, ("Invalid port count"));
1064 speed = lacp_aggregator_bandwidth(la);
1066 sc->sc_ifp->if_baudrate = speed;
1068 /* switch the active portmap over */
1069 atomic_store_rel_int(&lsc->lsc_activemap, newmap);
1070 LACP_DPRINTF((NULL, "Set table %d with %d ports\n",
1072 lsc->lsc_pmap[lsc->lsc_activemap].pm_count));
1076 lacp_compose_key(struct lacp_port *lp)
1078 struct lagg_port *lgp = lp->lp_lagg;
1079 struct lagg_softc *sc = lgp->lp_softc;
1080 u_int media = lp->lp_media;
1083 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) {
1086 * non-aggregatable links should have unique keys.
1088 * XXX this isn't really unique as if_index is 16 bit.
1091 /* bit 0..14: (some bits of) if_index of this port */
1092 key = lp->lp_ifp->if_index;
1096 u_int subtype = IFM_SUBTYPE(media);
1098 KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type"));
1099 KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface"));
1101 /* bit 0..4: IFM_SUBTYPE modulo speed */
1126 case IFM_10G_TWINAX:
1127 case IFM_10G_TWINAX_LONG:
1140 /* bit 5..14: (some bits of) if_index of lagg device */
1141 key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5);
1144 return (htons(key));
1148 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1150 char buf[LACP_LAGIDSTR_MAX+1];
1152 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1154 lacp_format_lagid(&la->la_actor, &la->la_partner,
1156 la->la_refcnt, la->la_refcnt + 1));
1158 KASSERT(la->la_refcnt > 0, ("refcount <= 0"));
1160 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount"));
1164 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1166 char buf[LACP_LAGIDSTR_MAX+1];
1168 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1170 lacp_format_lagid(&la->la_actor, &la->la_partner,
1172 la->la_refcnt, la->la_refcnt - 1));
1174 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt"));
1176 if (la->la_refcnt > 0) {
1180 KASSERT(la->la_refcnt == 0, ("refcount not zero"));
1181 KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active"));
1183 TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q);
1189 * lacp_aggregator_get: allocate an aggregator.
1192 static struct lacp_aggregator *
1193 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp)
1195 struct lacp_aggregator *la;
1197 la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT);
1201 TAILQ_INIT(&la->la_ports);
1203 TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q);
1210 * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port.
1214 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp)
1216 lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner);
1217 lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor);
1219 la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION;
1223 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr,
1224 const struct lacp_peerinfo *lpi_port)
1226 memset(lpi_aggr, 0, sizeof(*lpi_aggr));
1227 lpi_aggr->lip_systemid = lpi_port->lip_systemid;
1228 lpi_aggr->lip_key = lpi_port->lip_key;
1232 * lacp_aggregator_is_compatible: check if a port can join to an aggregator.
1236 lacp_aggregator_is_compatible(const struct lacp_aggregator *la,
1237 const struct lacp_port *lp)
1239 if (!(lp->lp_state & LACP_STATE_AGGREGATION) ||
1240 !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) {
1244 if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) {
1248 if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) {
1252 if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) {
1260 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a,
1261 const struct lacp_peerinfo *b)
1263 if (memcmp(&a->lip_systemid, &b->lip_systemid,
1264 sizeof(a->lip_systemid))) {
1268 if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) {
1276 lacp_port_enable(struct lacp_port *lp)
1278 lp->lp_state |= LACP_STATE_AGGREGATION;
1282 lacp_port_disable(struct lacp_port *lp)
1284 lacp_set_mux(lp, LACP_MUX_DETACHED);
1286 lp->lp_state &= ~LACP_STATE_AGGREGATION;
1287 lp->lp_selected = LACP_UNSELECTED;
1288 lacp_sm_rx_record_default(lp);
1289 lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION;
1290 lp->lp_state &= ~LACP_STATE_EXPIRED;
1294 * lacp_select: select an aggregator. create one if necessary.
1297 lacp_select(struct lacp_port *lp)
1299 struct lacp_softc *lsc = lp->lp_lsc;
1300 struct lacp_aggregator *la;
1301 char buf[LACP_LAGIDSTR_MAX+1];
1303 if (lp->lp_aggregator) {
1307 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1308 ("timer_wait_while still active"));
1310 LACP_DPRINTF((lp, "port lagid=%s\n",
1311 lacp_format_lagid(&lp->lp_actor, &lp->lp_partner,
1312 buf, sizeof(buf))));
1314 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
1315 if (lacp_aggregator_is_compatible(la, lp)) {
1321 la = lacp_aggregator_get(lsc, lp);
1323 LACP_DPRINTF((lp, "aggregator creation failed\n"));
1326 * will retry on the next tick.
1331 lacp_fill_aggregator_id(la, lp);
1332 LACP_DPRINTF((lp, "aggregator created\n"));
1334 LACP_DPRINTF((lp, "compatible aggregator found\n"));
1335 if (la->la_refcnt == LACP_MAX_PORTS)
1337 lacp_aggregator_addref(lsc, la);
1340 LACP_DPRINTF((lp, "aggregator lagid=%s\n",
1341 lacp_format_lagid(&la->la_actor, &la->la_partner,
1342 buf, sizeof(buf))));
1344 lp->lp_aggregator = la;
1345 lp->lp_selected = LACP_SELECTED;
1349 * lacp_unselect: finish unselect/detach process.
1353 lacp_unselect(struct lacp_port *lp)
1355 struct lacp_softc *lsc = lp->lp_lsc;
1356 struct lacp_aggregator *la = lp->lp_aggregator;
1358 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1359 ("timer_wait_while still active"));
1365 lp->lp_aggregator = NULL;
1366 lacp_aggregator_delref(lsc, la);
1372 lacp_sm_mux(struct lacp_port *lp)
1374 struct lagg_port *lgp = lp->lp_lagg;
1375 struct lagg_softc *sc = lgp->lp_softc;
1376 enum lacp_mux_state new_state;
1378 (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0;
1379 boolean_t p_collecting =
1380 (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0;
1381 enum lacp_selected selected = lp->lp_selected;
1382 struct lacp_aggregator *la;
1385 lacp_dprintf(lp, "%s: state= 0x%x, selected= 0x%x, "
1386 "p_sync= 0x%x, p_collecting= 0x%x\n", __func__,
1387 lp->lp_mux_state, selected, p_sync, p_collecting);
1390 la = lp->lp_aggregator;
1391 KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL,
1392 ("MUX not detached"));
1393 new_state = lp->lp_mux_state;
1394 switch (lp->lp_mux_state) {
1395 case LACP_MUX_DETACHED:
1396 if (selected != LACP_UNSELECTED) {
1397 new_state = LACP_MUX_WAITING;
1400 case LACP_MUX_WAITING:
1401 KASSERT(la->la_pending > 0 ||
1402 !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1403 ("timer_wait_while still active"));
1404 if (selected == LACP_SELECTED && la->la_pending == 0) {
1405 new_state = LACP_MUX_ATTACHED;
1406 } else if (selected == LACP_UNSELECTED) {
1407 new_state = LACP_MUX_DETACHED;
1410 case LACP_MUX_ATTACHED:
1411 if (selected == LACP_SELECTED && p_sync) {
1412 new_state = LACP_MUX_COLLECTING;
1413 } else if (selected != LACP_SELECTED) {
1414 new_state = LACP_MUX_DETACHED;
1417 case LACP_MUX_COLLECTING:
1418 if (selected == LACP_SELECTED && p_sync && p_collecting) {
1419 new_state = LACP_MUX_DISTRIBUTING;
1420 } else if (selected != LACP_SELECTED || !p_sync) {
1421 new_state = LACP_MUX_ATTACHED;
1424 case LACP_MUX_DISTRIBUTING:
1425 if (selected != LACP_SELECTED || !p_sync || !p_collecting) {
1426 new_state = LACP_MUX_COLLECTING;
1427 lacp_dprintf(lp, "Interface stopped DISTRIBUTING, possible flapping\n");
1432 panic("%s: unknown state", __func__);
1435 if (lp->lp_mux_state == new_state) {
1439 lacp_set_mux(lp, new_state);
1444 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state)
1446 struct lacp_aggregator *la = lp->lp_aggregator;
1448 if (lp->lp_mux_state == new_state) {
1452 switch (new_state) {
1453 case LACP_MUX_DETACHED:
1454 lp->lp_state &= ~LACP_STATE_SYNC;
1455 lacp_disable_distributing(lp);
1456 lacp_disable_collecting(lp);
1457 lacp_sm_assert_ntt(lp);
1459 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) {
1460 KASSERT(la->la_pending > 0,
1461 ("timer_wait_while not active"));
1464 LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE);
1467 case LACP_MUX_WAITING:
1468 LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE,
1469 LACP_AGGREGATE_WAIT_TIME);
1472 case LACP_MUX_ATTACHED:
1473 lp->lp_state |= LACP_STATE_SYNC;
1474 lacp_disable_collecting(lp);
1475 lacp_sm_assert_ntt(lp);
1477 case LACP_MUX_COLLECTING:
1478 lacp_enable_collecting(lp);
1479 lacp_disable_distributing(lp);
1480 lacp_sm_assert_ntt(lp);
1482 case LACP_MUX_DISTRIBUTING:
1483 lacp_enable_distributing(lp);
1486 panic("%s: unknown state", __func__);
1489 LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state));
1491 lp->lp_mux_state = new_state;
1495 lacp_sm_mux_timer(struct lacp_port *lp)
1497 struct lacp_aggregator *la = lp->lp_aggregator;
1498 char buf[LACP_LAGIDSTR_MAX+1];
1500 KASSERT(la->la_pending > 0, ("no pending event"));
1502 LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__,
1503 lacp_format_lagid(&la->la_actor, &la->la_partner,
1505 la->la_pending, la->la_pending - 1));
1510 /* periodic transmit machine */
1513 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate)
1515 if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state,
1516 LACP_STATE_TIMEOUT)) {
1520 LACP_DPRINTF((lp, "partner timeout changed\n"));
1523 * FAST_PERIODIC -> SLOW_PERIODIC
1525 * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC
1527 * let lacp_sm_ptx_tx_schedule to update timeout.
1530 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1533 * if timeout has been shortened, assert NTT.
1536 if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) {
1537 lacp_sm_assert_ntt(lp);
1542 lacp_sm_ptx_tx_schedule(struct lacp_port *lp)
1546 if (!(lp->lp_state & LACP_STATE_ACTIVITY) &&
1547 !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) {
1553 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1557 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) {
1561 timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ?
1562 LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME;
1564 LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout);
1568 lacp_sm_ptx_timer(struct lacp_port *lp)
1570 lacp_sm_assert_ntt(lp);
1574 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du)
1579 * check LACP_DISABLED first
1582 if (!(lp->lp_state & LACP_STATE_AGGREGATION)) {
1587 * check loopback condition.
1590 if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid,
1591 &lp->lp_actor.lip_systemid)) {
1596 * EXPIRED, DEFAULTED, CURRENT -> CURRENT
1599 lacp_sm_rx_update_selected(lp, du);
1600 lacp_sm_rx_update_ntt(lp, du);
1601 lacp_sm_rx_record_pdu(lp, du);
1603 timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ?
1604 LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME;
1605 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout);
1607 lp->lp_state &= ~LACP_STATE_EXPIRED;
1610 * kick transmit machine without waiting the next tick.
1617 lacp_sm_rx_set_expired(struct lacp_port *lp)
1619 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1620 lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT;
1621 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME);
1622 lp->lp_state |= LACP_STATE_EXPIRED;
1626 lacp_sm_rx_timer(struct lacp_port *lp)
1628 if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) {
1629 /* CURRENT -> EXPIRED */
1630 LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__));
1631 lacp_sm_rx_set_expired(lp);
1633 /* EXPIRED -> DEFAULTED */
1634 LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__));
1635 lacp_sm_rx_update_default_selected(lp);
1636 lacp_sm_rx_record_default(lp);
1637 lp->lp_state &= ~LACP_STATE_EXPIRED;
1642 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du)
1646 char buf[LACP_STATESTR_MAX+1];
1650 oldpstate = lp->lp_partner.lip_state;
1652 active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY)
1653 || ((lp->lp_state & LACP_STATE_ACTIVITY) &&
1654 (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY));
1656 lp->lp_partner = du->ldu_actor;
1658 ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1659 LACP_STATE_AGGREGATION) &&
1660 !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner))
1661 || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) {
1664 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1667 lp->lp_state &= ~LACP_STATE_DEFAULTED;
1669 if (oldpstate != lp->lp_partner.lip_state) {
1670 LACP_DPRINTF((lp, "old pstate %s\n",
1671 lacp_format_state(oldpstate, buf, sizeof(buf))));
1672 LACP_DPRINTF((lp, "new pstate %s\n",
1673 lacp_format_state(lp->lp_partner.lip_state, buf,
1677 /* XXX Hack, still need to implement 5.4.9 para 2,3,4 */
1678 if (lp->lp_lsc->lsc_strict_mode)
1679 lp->lp_partner.lip_state |= LACP_STATE_SYNC;
1681 lacp_sm_ptx_update_timeout(lp, oldpstate);
1685 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du)
1690 if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) ||
1691 !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1692 LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) {
1693 LACP_DPRINTF((lp, "%s: assert ntt\n", __func__));
1694 lacp_sm_assert_ntt(lp);
1699 lacp_sm_rx_record_default(struct lacp_port *lp)
1705 oldpstate = lp->lp_partner.lip_state;
1706 if (lp->lp_lsc->lsc_strict_mode)
1707 lp->lp_partner = lacp_partner_admin_strict;
1709 lp->lp_partner = lacp_partner_admin_optimistic;;
1710 lp->lp_state |= LACP_STATE_DEFAULTED;
1711 lacp_sm_ptx_update_timeout(lp, oldpstate);
1715 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp,
1716 const struct lacp_peerinfo *info)
1721 if (lacp_compare_peerinfo(&lp->lp_partner, info) ||
1722 !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state,
1723 LACP_STATE_AGGREGATION)) {
1724 lp->lp_selected = LACP_UNSELECTED;
1725 /* mux machine will clean up lp->lp_aggregator */
1730 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du)
1735 lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor);
1739 lacp_sm_rx_update_default_selected(struct lacp_port *lp)
1744 if (lp->lp_lsc->lsc_strict_mode)
1745 lacp_sm_rx_update_selected_from_peerinfo(lp,
1746 &lacp_partner_admin_strict);
1748 lacp_sm_rx_update_selected_from_peerinfo(lp,
1749 &lacp_partner_admin_optimistic);
1752 /* transmit machine */
1755 lacp_sm_tx(struct lacp_port *lp)
1759 if (!(lp->lp_state & LACP_STATE_AGGREGATION)
1761 || (!(lp->lp_state & LACP_STATE_ACTIVITY)
1762 && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY))
1765 lp->lp_flags &= ~LACP_PORT_NTT;
1768 if (!(lp->lp_flags & LACP_PORT_NTT)) {
1772 /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */
1773 if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent,
1774 (3 / LACP_FAST_PERIODIC_TIME)) == 0) {
1775 LACP_DPRINTF((lp, "rate limited pdu\n"));
1779 if (((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_tx_test) == 0) {
1780 error = lacp_xmit_lacpdu(lp);
1782 LACP_TPRINTF((lp, "Dropping TX PDU\n"));
1786 lp->lp_flags &= ~LACP_PORT_NTT;
1788 LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n",
1794 lacp_sm_assert_ntt(struct lacp_port *lp)
1797 lp->lp_flags |= LACP_PORT_NTT;
1801 lacp_run_timers(struct lacp_port *lp)
1805 for (i = 0; i < LACP_NTIMER; i++) {
1806 KASSERT(lp->lp_timer[i] >= 0,
1807 ("invalid timer value %d", lp->lp_timer[i]));
1808 if (lp->lp_timer[i] == 0) {
1810 } else if (--lp->lp_timer[i] <= 0) {
1811 if (lacp_timer_funcs[i]) {
1812 (*lacp_timer_funcs[i])(lp);
1819 lacp_marker_input(struct lacp_port *lp, struct mbuf *m)
1821 struct lacp_softc *lsc = lp->lp_lsc;
1822 struct lagg_port *lgp = lp->lp_lagg;
1823 struct lacp_port *lp2;
1824 struct markerdu *mdu;
1828 if (m->m_pkthdr.len != sizeof(*mdu)) {
1832 if ((m->m_flags & M_MCAST) == 0) {
1836 if (m->m_len < sizeof(*mdu)) {
1837 m = m_pullup(m, sizeof(*mdu));
1843 mdu = mtod(m, struct markerdu *);
1845 if (memcmp(&mdu->mdu_eh.ether_dhost,
1846 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
1850 if (mdu->mdu_sph.sph_version != 1) {
1854 switch (mdu->mdu_tlv.tlv_type) {
1855 case MARKER_TYPE_INFO:
1856 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
1857 marker_info_tlv_template, TRUE)) {
1860 mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE;
1861 memcpy(&mdu->mdu_eh.ether_dhost,
1862 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN);
1863 memcpy(&mdu->mdu_eh.ether_shost,
1864 lgp->lp_lladdr, ETHER_ADDR_LEN);
1865 error = lagg_enqueue(lp->lp_ifp, m);
1868 case MARKER_TYPE_RESPONSE:
1869 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
1870 marker_response_tlv_template, TRUE)) {
1873 LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n",
1874 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system,
1875 ":", ntohl(mdu->mdu_info.mi_rq_xid)));
1877 /* Verify that it is the last marker we sent out */
1878 if (memcmp(&mdu->mdu_info, &lp->lp_marker,
1879 sizeof(struct lacp_markerinfo)))
1883 lp->lp_flags &= ~LACP_PORT_MARK;
1885 if (lsc->lsc_suppress_distributing) {
1886 /* Check if any ports are waiting for a response */
1887 LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) {
1888 if (lp2->lp_flags & LACP_PORT_MARK) {
1895 /* All interface queues are clear */
1896 LACP_DPRINTF((NULL, "queue flush complete\n"));
1897 lsc->lsc_suppress_distributing = FALSE;
1911 LACP_DPRINTF((lp, "bad marker frame\n"));
1917 tlv_check(const void *p, size_t size, const struct tlvhdr *tlv,
1918 const struct tlv_template *tmpl, boolean_t check_type)
1920 while (/* CONSTCOND */ 1) {
1921 if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) {
1924 if ((check_type && tlv->tlv_type != tmpl->tmpl_type) ||
1925 tlv->tlv_length != tmpl->tmpl_length) {
1928 if (tmpl->tmpl_type == 0) {
1931 tlv = (const struct tlvhdr *)
1932 ((const char *)tlv + tlv->tlv_length);
1941 lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen)
1943 snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X",
1955 lacp_format_systemid(const struct lacp_systemid *sysid,
1956 char *buf, size_t buflen)
1958 char macbuf[LACP_MACSTR_MAX+1];
1960 snprintf(buf, buflen, "%04X,%s",
1961 ntohs(sysid->lsi_prio),
1962 lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf)));
1968 lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen)
1970 snprintf(buf, buflen, "%04X,%04X",
1971 ntohs(portid->lpi_prio),
1972 ntohs(portid->lpi_portno));
1978 lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen)
1980 char sysid[LACP_SYSTEMIDSTR_MAX+1];
1981 char portid[LACP_PORTIDSTR_MAX+1];
1983 snprintf(buf, buflen, "(%s,%04X,%s)",
1984 lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)),
1985 ntohs(peer->lip_key),
1986 lacp_format_portid(&peer->lip_portid, portid, sizeof(portid)));
1992 lacp_format_lagid(const struct lacp_peerinfo *a,
1993 const struct lacp_peerinfo *b, char *buf, size_t buflen)
1995 char astr[LACP_PARTNERSTR_MAX+1];
1996 char bstr[LACP_PARTNERSTR_MAX+1];
2000 * there's a convention to display small numbered peer
2004 if (lacp_compare_peerinfo(a, b) > 0) {
2005 const struct lacp_peerinfo *t;
2013 snprintf(buf, buflen, "[%s,%s]",
2014 lacp_format_partner(a, astr, sizeof(astr)),
2015 lacp_format_partner(b, bstr, sizeof(bstr)));
2021 lacp_format_lagid_aggregator(const struct lacp_aggregator *la,
2022 char *buf, size_t buflen)
2028 return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen));
2032 lacp_format_state(uint8_t state, char *buf, size_t buflen)
2034 snprintf(buf, buflen, "%b", state, LACP_STATE_BITS);
2039 lacp_dump_lacpdu(const struct lacpdu *du)
2041 char buf[LACP_PARTNERSTR_MAX+1];
2042 char buf2[LACP_STATESTR_MAX+1];
2044 printf("actor=%s\n",
2045 lacp_format_partner(&du->ldu_actor, buf, sizeof(buf)));
2046 printf("actor.state=%s\n",
2047 lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2)));
2048 printf("partner=%s\n",
2049 lacp_format_partner(&du->ldu_partner, buf, sizeof(buf)));
2050 printf("partner.state=%s\n",
2051 lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2)));
2053 printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay));
2057 lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...)
2062 printf("%s: ", lp->lp_ifp->if_xname);