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_NODE(_net_link_lagg, OID_AUTO, lacp, CTLFLAG_RD, 0, "ieee802.3ad");
192 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, debug, CTLFLAG_RW | CTLFLAG_TUN,
193 &lacp_debug, 0, "Enable LACP debug logging (1=debug, 2=trace)");
194 TUNABLE_INT("net.link.lagg.lacp.debug", &lacp_debug);
196 /* bitmap of ports */
197 static int lacp_rx_test = 0;
198 static int lacp_tx_test = 0;
199 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, rxtest, CTLFLAG_RW, &lacp_rx_test, 0,
201 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, txtest, CTLFLAG_RW, &lacp_tx_test, 0,
204 static int lacp_strict = 1;
205 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, strict, CTLFLAG_RW, &lacp_strict,
206 0, "Strict spec compliance");
208 #define LACP_DPRINTF(a) if (lacp_debug & 0x01) { lacp_dprintf a ; }
209 #define LACP_TRACE(a) if (lacp_debug & 0x02) { lacp_dprintf(a,"%s\n",__func__); }
210 #define LACP_TPRINTF(a) if (lacp_debug & 0x04) { lacp_dprintf a ; }
213 * partner administration variables.
214 * XXX should be configurable.
217 static const struct lacp_peerinfo lacp_partner_admin_optimistic = {
218 .lip_systemid = { .lsi_prio = 0xffff },
219 .lip_portid = { .lpi_prio = 0xffff },
220 .lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION |
221 LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING,
224 static const struct lacp_peerinfo lacp_partner_admin_strict = {
225 .lip_systemid = { .lsi_prio = 0xffff },
226 .lip_portid = { .lpi_prio = 0xffff },
230 static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = {
231 [LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer,
232 [LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer,
233 [LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer,
237 lacp_input(struct lagg_port *lgp, struct mbuf *m)
239 struct lacp_port *lp = LACP_PORT(lgp);
242 if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) {
247 m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype);
249 case SLOWPROTOCOLS_SUBTYPE_LACP:
250 lacp_pdu_input(lp, m);
253 case SLOWPROTOCOLS_SUBTYPE_MARKER:
254 lacp_marker_input(lp, m);
258 /* Not a subtype we are interested in */
263 * lacp_pdu_input: process lacpdu
266 lacp_pdu_input(struct lacp_port *lp, struct mbuf *m)
268 struct lacp_softc *lsc = lp->lp_lsc;
272 if (m->m_pkthdr.len != sizeof(*du)) {
276 if ((m->m_flags & M_MCAST) == 0) {
280 if (m->m_len < sizeof(*du)) {
281 m = m_pullup(m, sizeof(*du));
287 du = mtod(m, struct lacpdu *);
289 if (memcmp(&du->ldu_eh.ether_dhost,
290 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
295 * ignore the version for compatibility with
296 * the future protocol revisions.
299 if (du->ldu_sph.sph_version != 1) {
305 * ignore tlv types for compatibility with
306 * the future protocol revisions.
308 if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor,
309 lacp_info_tlv_template, FALSE)) {
313 if (lacp_debug > 0) {
314 lacp_dprintf(lp, "lacpdu receive\n");
315 lacp_dump_lacpdu(du);
318 if ((1 << lp->lp_ifp->if_dunit) & lacp_rx_test) {
319 LACP_TPRINTF((lp, "Dropping RX PDU\n"));
336 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info)
338 struct lagg_port *lgp = lp->lp_lagg;
339 struct lagg_softc *sc = lgp->lp_softc;
341 info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO);
342 memcpy(&info->lip_systemid.lsi_mac,
343 IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
344 info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO);
345 info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index);
346 info->lip_state = lp->lp_state;
350 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info)
352 struct ifnet *ifp = lp->lp_ifp;
354 /* Fill in the port index and system id (encoded as the MAC) */
355 info->mi_rq_port = htons(ifp->if_index);
356 memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN);
357 info->mi_rq_xid = htonl(0);
361 lacp_xmit_lacpdu(struct lacp_port *lp)
363 struct lagg_port *lgp = lp->lp_lagg;
368 LACP_LOCK_ASSERT(lp->lp_lsc);
370 m = m_gethdr(M_NOWAIT, MT_DATA);
374 m->m_len = m->m_pkthdr.len = sizeof(*du);
376 du = mtod(m, struct lacpdu *);
377 memset(du, 0, sizeof(*du));
379 memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
381 memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
382 du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW);
384 du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP;
385 du->ldu_sph.sph_version = 1;
387 TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor));
388 du->ldu_actor = lp->lp_actor;
390 TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO,
391 sizeof(du->ldu_partner));
392 du->ldu_partner = lp->lp_partner;
394 TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO,
395 sizeof(du->ldu_collector));
396 du->ldu_collector.lci_maxdelay = 0;
398 if (lacp_debug > 0) {
399 lacp_dprintf(lp, "lacpdu transmit\n");
400 lacp_dump_lacpdu(du);
403 m->m_flags |= M_MCAST;
406 * XXX should use higher priority queue.
407 * otherwise network congestion can break aggregation.
410 error = lagg_enqueue(lp->lp_ifp, m);
415 lacp_xmit_marker(struct lacp_port *lp)
417 struct lagg_port *lgp = lp->lp_lagg;
419 struct markerdu *mdu;
422 LACP_LOCK_ASSERT(lp->lp_lsc);
424 m = m_gethdr(M_NOWAIT, MT_DATA);
428 m->m_len = m->m_pkthdr.len = sizeof(*mdu);
430 mdu = mtod(m, struct markerdu *);
431 memset(mdu, 0, sizeof(*mdu));
433 memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
435 memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
436 mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW);
438 mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER;
439 mdu->mdu_sph.sph_version = 1;
441 /* Bump the transaction id and copy over the marker info */
442 lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1);
443 TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info));
444 mdu->mdu_info = lp->lp_marker;
446 LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n",
447 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":",
448 ntohl(mdu->mdu_info.mi_rq_xid)));
450 m->m_flags |= M_MCAST;
451 error = lagg_enqueue(lp->lp_ifp, m);
456 lacp_linkstate(struct lagg_port *lgp)
458 struct lacp_port *lp = LACP_PORT(lgp);
459 struct lacp_softc *lsc = lp->lp_lsc;
460 struct ifnet *ifp = lgp->lp_ifp;
461 struct ifmediareq ifmr;
467 bzero((char *)&ifmr, sizeof(ifmr));
468 error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr);
473 media = ifmr.ifm_active;
474 LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, "
475 "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER,
476 (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP));
477 old_state = lp->lp_state;
478 old_key = lp->lp_key;
480 lp->lp_media = media;
482 * If the port is not an active full duplex Ethernet link then it can
485 if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 ||
486 ifp->if_link_state != LINK_STATE_UP) {
487 lacp_port_disable(lp);
489 lacp_port_enable(lp);
491 lp->lp_key = lacp_compose_key(lp);
493 if (old_state != lp->lp_state || old_key != lp->lp_key) {
494 LACP_DPRINTF((lp, "-> UNSELECTED\n"));
495 lp->lp_selected = LACP_UNSELECTED;
503 struct lacp_softc *lsc = arg;
504 struct lacp_port *lp;
506 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
507 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0)
515 lacp_sm_ptx_tx_schedule(lp);
517 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
521 lacp_port_create(struct lagg_port *lgp)
523 struct lagg_softc *sc = lgp->lp_softc;
524 struct lacp_softc *lsc = LACP_SOFTC(sc);
525 struct lacp_port *lp;
526 struct ifnet *ifp = lgp->lp_ifp;
527 struct sockaddr_dl sdl;
528 struct ifmultiaddr *rifma = NULL;
531 boolean_t active = TRUE; /* XXX should be configurable */
532 boolean_t fast = FALSE; /* XXX should be configurable */
534 bzero((char *)&sdl, sizeof(sdl));
535 sdl.sdl_len = sizeof(sdl);
536 sdl.sdl_family = AF_LINK;
537 sdl.sdl_index = ifp->if_index;
538 sdl.sdl_type = IFT_ETHER;
539 sdl.sdl_alen = ETHER_ADDR_LEN;
541 bcopy(ðermulticastaddr_slowprotocols,
542 LLADDR(&sdl), ETHER_ADDR_LEN);
543 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
545 printf("%s: ADDMULTI failed on %s\n", __func__, lgp->lp_ifname);
549 lp = malloc(sizeof(struct lacp_port),
550 M_DEVBUF, M_NOWAIT|M_ZERO);
555 lgp->lp_psc = (caddr_t)lp;
561 LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next);
563 lacp_fill_actorinfo(lp, &lp->lp_actor);
564 lacp_fill_markerinfo(lp, &lp->lp_marker);
566 (active ? LACP_STATE_ACTIVITY : 0) |
567 (fast ? LACP_STATE_TIMEOUT : 0);
568 lp->lp_aggregator = NULL;
569 lacp_sm_rx_set_expired(lp);
577 lacp_port_destroy(struct lagg_port *lgp)
579 struct lacp_port *lp = LACP_PORT(lgp);
580 struct lacp_softc *lsc = lp->lp_lsc;
584 for (i = 0; i < LACP_NTIMER; i++) {
585 LACP_TIMER_DISARM(lp, i);
588 lacp_disable_collecting(lp);
589 lacp_disable_distributing(lp);
592 /* The address may have already been removed by if_purgemaddrs() */
593 if (!lgp->lp_detaching)
594 if_delmulti_ifma(lp->lp_ifma);
596 LIST_REMOVE(lp, lp_next);
602 lacp_req(struct lagg_softc *sc, caddr_t data)
604 struct lacp_opreq *req = (struct lacp_opreq *)data;
605 struct lacp_softc *lsc = LACP_SOFTC(sc);
606 struct lacp_aggregator *la = lsc->lsc_active_aggregator;
609 bzero(req, sizeof(struct lacp_opreq));
611 req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio);
612 memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac,
614 req->actor_key = ntohs(la->la_actor.lip_key);
615 req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio);
616 req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno);
617 req->actor_state = la->la_actor.lip_state;
619 req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio);
620 memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac,
622 req->partner_key = ntohs(la->la_partner.lip_key);
623 req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio);
624 req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno);
625 req->partner_state = la->la_partner.lip_state;
631 lacp_portreq(struct lagg_port *lgp, caddr_t data)
633 struct lacp_opreq *req = (struct lacp_opreq *)data;
634 struct lacp_port *lp = LACP_PORT(lgp);
635 struct lacp_softc *lsc = lp->lp_lsc;
638 req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio);
639 memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac,
641 req->actor_key = ntohs(lp->lp_actor.lip_key);
642 req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio);
643 req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno);
644 req->actor_state = lp->lp_actor.lip_state;
646 req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio);
647 memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac,
649 req->partner_key = ntohs(lp->lp_partner.lip_key);
650 req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio);
651 req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno);
652 req->partner_state = lp->lp_partner.lip_state;
657 lacp_disable_collecting(struct lacp_port *lp)
659 LACP_DPRINTF((lp, "collecting disabled\n"));
660 lp->lp_state &= ~LACP_STATE_COLLECTING;
664 lacp_enable_collecting(struct lacp_port *lp)
666 LACP_DPRINTF((lp, "collecting enabled\n"));
667 lp->lp_state |= LACP_STATE_COLLECTING;
671 lacp_disable_distributing(struct lacp_port *lp)
673 struct lacp_aggregator *la = lp->lp_aggregator;
674 struct lacp_softc *lsc = lp->lp_lsc;
675 struct lagg_softc *sc = lsc->lsc_softc;
676 char buf[LACP_LAGIDSTR_MAX+1];
678 LACP_LOCK_ASSERT(lsc);
680 if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) {
684 KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports"));
685 KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports));
686 KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid"));
688 LACP_DPRINTF((lp, "disable distributing on aggregator %s, "
690 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
691 la->la_nports, la->la_nports - 1));
693 TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q);
695 sc->sc_active = la->la_nports;
697 if (lsc->lsc_active_aggregator == la) {
698 lacp_suppress_distributing(lsc, la);
699 lacp_select_active_aggregator(lsc);
700 /* regenerate the port map, the active aggregator has changed */
701 lacp_update_portmap(lsc);
704 lp->lp_state &= ~LACP_STATE_DISTRIBUTING;
708 lacp_enable_distributing(struct lacp_port *lp)
710 struct lacp_aggregator *la = lp->lp_aggregator;
711 struct lacp_softc *lsc = lp->lp_lsc;
712 struct lagg_softc *sc = lsc->lsc_softc;
713 char buf[LACP_LAGIDSTR_MAX+1];
715 LACP_LOCK_ASSERT(lsc);
717 if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) {
721 LACP_DPRINTF((lp, "enable distributing on aggregator %s, "
723 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
724 la->la_nports, la->la_nports + 1));
726 KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid"));
727 TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q);
729 sc->sc_active = la->la_nports;
731 lp->lp_state |= LACP_STATE_DISTRIBUTING;
733 if (lsc->lsc_active_aggregator == la) {
734 lacp_suppress_distributing(lsc, la);
735 lacp_update_portmap(lsc);
737 /* try to become the active aggregator */
738 lacp_select_active_aggregator(lsc);
742 lacp_transit_expire(void *vp)
744 struct lacp_softc *lsc = vp;
746 LACP_LOCK_ASSERT(lsc);
750 lsc->lsc_suppress_distributing = FALSE;
754 lacp_attach(struct lagg_softc *sc)
756 struct lacp_softc *lsc;
758 lsc = malloc(sizeof(struct lacp_softc),
759 M_DEVBUF, M_NOWAIT|M_ZERO);
763 sc->sc_psc = (caddr_t)lsc;
766 lsc->lsc_hashkey = arc4random();
767 lsc->lsc_active_aggregator = NULL;
769 TAILQ_INIT(&lsc->lsc_aggregators);
770 LIST_INIT(&lsc->lsc_ports);
772 callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0);
773 callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0);
775 /* if the lagg is already up then do the same */
776 if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING)
783 lacp_detach(struct lagg_softc *sc)
785 struct lacp_softc *lsc = LACP_SOFTC(sc);
787 KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators),
788 ("aggregators still active"));
789 KASSERT(lsc->lsc_active_aggregator == NULL,
790 ("aggregator still attached"));
793 callout_drain(&lsc->lsc_transit_callout);
794 callout_drain(&lsc->lsc_callout);
796 LACP_LOCK_DESTROY(lsc);
802 lacp_init(struct lagg_softc *sc)
804 struct lacp_softc *lsc = LACP_SOFTC(sc);
807 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
812 lacp_stop(struct lagg_softc *sc)
814 struct lacp_softc *lsc = LACP_SOFTC(sc);
817 callout_stop(&lsc->lsc_transit_callout);
818 callout_stop(&lsc->lsc_callout);
823 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
825 struct lacp_softc *lsc = LACP_SOFTC(sc);
826 struct lacp_portmap *pm;
827 struct lacp_port *lp;
830 if (__predict_false(lsc->lsc_suppress_distributing)) {
831 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__));
835 pm = &lsc->lsc_pmap[lsc->lsc_activemap];
836 if (pm->pm_count == 0) {
837 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__));
841 if (sc->use_flowid && (m->m_flags & M_FLOWID))
842 hash = m->m_pkthdr.flowid;
844 hash = lagg_hashmbuf(sc, m, lsc->lsc_hashkey);
845 hash %= pm->pm_count;
846 lp = pm->pm_map[hash];
848 KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0,
849 ("aggregated port is not distributing"));
851 return (lp->lp_lagg);
854 * lacp_suppress_distributing: drop transmit packets for a while
855 * to preserve packet ordering.
859 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la)
861 struct lacp_port *lp;
863 if (lsc->lsc_active_aggregator != la) {
869 lsc->lsc_suppress_distributing = TRUE;
871 /* send a marker frame down each port to verify the queues are empty */
872 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
873 lp->lp_flags |= LACP_PORT_MARK;
874 lacp_xmit_marker(lp);
877 /* set a timeout for the marker frames */
878 callout_reset(&lsc->lsc_transit_callout,
879 LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc);
883 lacp_compare_peerinfo(const struct lacp_peerinfo *a,
884 const struct lacp_peerinfo *b)
886 return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state)));
890 lacp_compare_systemid(const struct lacp_systemid *a,
891 const struct lacp_systemid *b)
893 return (memcmp(a, b, sizeof(*a)));
898 lacp_compare_portid(const struct lacp_portid *a,
899 const struct lacp_portid *b)
901 return (memcmp(a, b, sizeof(*a)));
906 lacp_aggregator_bandwidth(struct lacp_aggregator *la)
908 struct lacp_port *lp;
911 lp = TAILQ_FIRST(&la->la_ports);
916 speed = ifmedia_baudrate(lp->lp_media);
917 speed *= la->la_nports;
919 LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n",
920 lp->lp_media, la->la_nports));
927 * lacp_select_active_aggregator: select an aggregator to be used to transmit
928 * packets from lagg(4) interface.
932 lacp_select_active_aggregator(struct lacp_softc *lsc)
934 struct lacp_aggregator *la;
935 struct lacp_aggregator *best_la = NULL;
936 uint64_t best_speed = 0;
937 char buf[LACP_LAGIDSTR_MAX+1];
941 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
944 if (la->la_nports == 0) {
948 speed = lacp_aggregator_bandwidth(la);
949 LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n",
950 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
951 speed, la->la_nports));
953 /* This aggregator is chosen if
954 * the partner has a better system priority
955 * or, the total aggregated speed is higher
956 * or, it is already the chosen aggregator
958 if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) <
959 LACP_SYS_PRI(best_la->la_partner)) ||
960 speed > best_speed ||
961 (speed == best_speed &&
962 la == lsc->lsc_active_aggregator)) {
968 KASSERT(best_la == NULL || best_la->la_nports > 0,
969 ("invalid aggregator refcnt"));
970 KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports),
971 ("invalid aggregator list"));
973 if (lsc->lsc_active_aggregator != best_la) {
974 LACP_DPRINTF((NULL, "active aggregator changed\n"));
975 LACP_DPRINTF((NULL, "old %s\n",
976 lacp_format_lagid_aggregator(lsc->lsc_active_aggregator,
979 LACP_DPRINTF((NULL, "active aggregator not changed\n"));
981 LACP_DPRINTF((NULL, "new %s\n",
982 lacp_format_lagid_aggregator(best_la, buf, sizeof(buf))));
984 if (lsc->lsc_active_aggregator != best_la) {
985 lsc->lsc_active_aggregator = best_la;
986 lacp_update_portmap(lsc);
988 lacp_suppress_distributing(lsc, best_la);
994 * Updated the inactive portmap array with the new list of ports and
998 lacp_update_portmap(struct lacp_softc *lsc)
1000 struct lagg_softc *sc = lsc->lsc_softc;
1001 struct lacp_aggregator *la;
1002 struct lacp_portmap *p;
1003 struct lacp_port *lp;
1008 newmap = lsc->lsc_activemap == 0 ? 1 : 0;
1009 p = &lsc->lsc_pmap[newmap];
1010 la = lsc->lsc_active_aggregator;
1012 bzero(p, sizeof(struct lacp_portmap));
1014 if (la != NULL && la->la_nports > 0) {
1015 p->pm_count = la->la_nports;
1017 TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q)
1018 p->pm_map[i++] = lp;
1019 KASSERT(i == p->pm_count, ("Invalid port count"));
1020 speed = lacp_aggregator_bandwidth(la);
1022 sc->sc_ifp->if_baudrate = speed;
1024 /* switch the active portmap over */
1025 atomic_store_rel_int(&lsc->lsc_activemap, newmap);
1026 LACP_DPRINTF((NULL, "Set table %d with %d ports\n",
1028 lsc->lsc_pmap[lsc->lsc_activemap].pm_count));
1032 lacp_compose_key(struct lacp_port *lp)
1034 struct lagg_port *lgp = lp->lp_lagg;
1035 struct lagg_softc *sc = lgp->lp_softc;
1036 u_int media = lp->lp_media;
1039 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) {
1042 * non-aggregatable links should have unique keys.
1044 * XXX this isn't really unique as if_index is 16 bit.
1047 /* bit 0..14: (some bits of) if_index of this port */
1048 key = lp->lp_ifp->if_index;
1052 u_int subtype = IFM_SUBTYPE(media);
1054 KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type"));
1055 KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface"));
1057 /* bit 0..4: IFM_SUBTYPE */
1059 /* bit 5..14: (some bits of) if_index of lagg device */
1060 key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5);
1063 return (htons(key));
1067 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1069 char buf[LACP_LAGIDSTR_MAX+1];
1071 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1073 lacp_format_lagid(&la->la_actor, &la->la_partner,
1075 la->la_refcnt, la->la_refcnt + 1));
1077 KASSERT(la->la_refcnt > 0, ("refcount <= 0"));
1079 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount"));
1083 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1085 char buf[LACP_LAGIDSTR_MAX+1];
1087 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1089 lacp_format_lagid(&la->la_actor, &la->la_partner,
1091 la->la_refcnt, la->la_refcnt - 1));
1093 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt"));
1095 if (la->la_refcnt > 0) {
1099 KASSERT(la->la_refcnt == 0, ("refcount not zero"));
1100 KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active"));
1102 TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q);
1108 * lacp_aggregator_get: allocate an aggregator.
1111 static struct lacp_aggregator *
1112 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp)
1114 struct lacp_aggregator *la;
1116 la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT);
1120 TAILQ_INIT(&la->la_ports);
1122 TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q);
1129 * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port.
1133 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp)
1135 lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner);
1136 lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor);
1138 la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION;
1142 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr,
1143 const struct lacp_peerinfo *lpi_port)
1145 memset(lpi_aggr, 0, sizeof(*lpi_aggr));
1146 lpi_aggr->lip_systemid = lpi_port->lip_systemid;
1147 lpi_aggr->lip_key = lpi_port->lip_key;
1151 * lacp_aggregator_is_compatible: check if a port can join to an aggregator.
1155 lacp_aggregator_is_compatible(const struct lacp_aggregator *la,
1156 const struct lacp_port *lp)
1158 if (!(lp->lp_state & LACP_STATE_AGGREGATION) ||
1159 !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) {
1163 if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) {
1167 if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) {
1171 if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) {
1179 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a,
1180 const struct lacp_peerinfo *b)
1182 if (memcmp(&a->lip_systemid, &b->lip_systemid,
1183 sizeof(a->lip_systemid))) {
1187 if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) {
1195 lacp_port_enable(struct lacp_port *lp)
1197 lp->lp_state |= LACP_STATE_AGGREGATION;
1201 lacp_port_disable(struct lacp_port *lp)
1203 lacp_set_mux(lp, LACP_MUX_DETACHED);
1205 lp->lp_state &= ~LACP_STATE_AGGREGATION;
1206 lp->lp_selected = LACP_UNSELECTED;
1207 lacp_sm_rx_record_default(lp);
1208 lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION;
1209 lp->lp_state &= ~LACP_STATE_EXPIRED;
1213 * lacp_select: select an aggregator. create one if necessary.
1216 lacp_select(struct lacp_port *lp)
1218 struct lacp_softc *lsc = lp->lp_lsc;
1219 struct lacp_aggregator *la;
1220 char buf[LACP_LAGIDSTR_MAX+1];
1222 if (lp->lp_aggregator) {
1226 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1227 ("timer_wait_while still active"));
1229 LACP_DPRINTF((lp, "port lagid=%s\n",
1230 lacp_format_lagid(&lp->lp_actor, &lp->lp_partner,
1231 buf, sizeof(buf))));
1233 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
1234 if (lacp_aggregator_is_compatible(la, lp)) {
1240 la = lacp_aggregator_get(lsc, lp);
1242 LACP_DPRINTF((lp, "aggregator creation failed\n"));
1245 * will retry on the next tick.
1250 lacp_fill_aggregator_id(la, lp);
1251 LACP_DPRINTF((lp, "aggregator created\n"));
1253 LACP_DPRINTF((lp, "compatible aggregator found\n"));
1254 if (la->la_refcnt == LACP_MAX_PORTS)
1256 lacp_aggregator_addref(lsc, la);
1259 LACP_DPRINTF((lp, "aggregator lagid=%s\n",
1260 lacp_format_lagid(&la->la_actor, &la->la_partner,
1261 buf, sizeof(buf))));
1263 lp->lp_aggregator = la;
1264 lp->lp_selected = LACP_SELECTED;
1268 * lacp_unselect: finish unselect/detach process.
1272 lacp_unselect(struct lacp_port *lp)
1274 struct lacp_softc *lsc = lp->lp_lsc;
1275 struct lacp_aggregator *la = lp->lp_aggregator;
1277 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1278 ("timer_wait_while still active"));
1284 lp->lp_aggregator = NULL;
1285 lacp_aggregator_delref(lsc, la);
1291 lacp_sm_mux(struct lacp_port *lp)
1293 struct lagg_port *lgp = lp->lp_lagg;
1294 struct lagg_softc *sc = lgp->lp_softc;
1295 enum lacp_mux_state new_state;
1297 (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0;
1298 boolean_t p_collecting =
1299 (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0;
1300 enum lacp_selected selected = lp->lp_selected;
1301 struct lacp_aggregator *la;
1304 lacp_dprintf(lp, "%s: state= 0x%x, selected= 0x%x, "
1305 "p_sync= 0x%x, p_collecting= 0x%x\n", __func__,
1306 lp->lp_mux_state, selected, p_sync, p_collecting);
1309 la = lp->lp_aggregator;
1310 KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL,
1311 ("MUX not detached"));
1312 new_state = lp->lp_mux_state;
1313 switch (lp->lp_mux_state) {
1314 case LACP_MUX_DETACHED:
1315 if (selected != LACP_UNSELECTED) {
1316 new_state = LACP_MUX_WAITING;
1319 case LACP_MUX_WAITING:
1320 KASSERT(la->la_pending > 0 ||
1321 !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1322 ("timer_wait_while still active"));
1323 if (selected == LACP_SELECTED && la->la_pending == 0) {
1324 new_state = LACP_MUX_ATTACHED;
1325 } else if (selected == LACP_UNSELECTED) {
1326 new_state = LACP_MUX_DETACHED;
1329 case LACP_MUX_ATTACHED:
1330 if (selected == LACP_SELECTED && p_sync) {
1331 new_state = LACP_MUX_COLLECTING;
1332 } else if (selected != LACP_SELECTED) {
1333 new_state = LACP_MUX_DETACHED;
1336 case LACP_MUX_COLLECTING:
1337 if (selected == LACP_SELECTED && p_sync && p_collecting) {
1338 new_state = LACP_MUX_DISTRIBUTING;
1339 } else if (selected != LACP_SELECTED || !p_sync) {
1340 new_state = LACP_MUX_ATTACHED;
1343 case LACP_MUX_DISTRIBUTING:
1344 if (selected != LACP_SELECTED || !p_sync || !p_collecting) {
1345 new_state = LACP_MUX_COLLECTING;
1346 lacp_dprintf(lp, "Interface stopped DISTRIBUTING, possible flaping\n");
1351 panic("%s: unknown state", __func__);
1354 if (lp->lp_mux_state == new_state) {
1358 lacp_set_mux(lp, new_state);
1363 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state)
1365 struct lacp_aggregator *la = lp->lp_aggregator;
1367 if (lp->lp_mux_state == new_state) {
1371 switch (new_state) {
1372 case LACP_MUX_DETACHED:
1373 lp->lp_state &= ~LACP_STATE_SYNC;
1374 lacp_disable_distributing(lp);
1375 lacp_disable_collecting(lp);
1376 lacp_sm_assert_ntt(lp);
1378 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) {
1379 KASSERT(la->la_pending > 0,
1380 ("timer_wait_while not active"));
1383 LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE);
1386 case LACP_MUX_WAITING:
1387 LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE,
1388 LACP_AGGREGATE_WAIT_TIME);
1391 case LACP_MUX_ATTACHED:
1392 lp->lp_state |= LACP_STATE_SYNC;
1393 lacp_disable_collecting(lp);
1394 lacp_sm_assert_ntt(lp);
1396 case LACP_MUX_COLLECTING:
1397 lacp_enable_collecting(lp);
1398 lacp_disable_distributing(lp);
1399 lacp_sm_assert_ntt(lp);
1401 case LACP_MUX_DISTRIBUTING:
1402 lacp_enable_distributing(lp);
1405 panic("%s: unknown state", __func__);
1408 LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state));
1410 lp->lp_mux_state = new_state;
1414 lacp_sm_mux_timer(struct lacp_port *lp)
1416 struct lacp_aggregator *la = lp->lp_aggregator;
1417 char buf[LACP_LAGIDSTR_MAX+1];
1419 KASSERT(la->la_pending > 0, ("no pending event"));
1421 LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__,
1422 lacp_format_lagid(&la->la_actor, &la->la_partner,
1424 la->la_pending, la->la_pending - 1));
1429 /* periodic transmit machine */
1432 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate)
1434 if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state,
1435 LACP_STATE_TIMEOUT)) {
1439 LACP_DPRINTF((lp, "partner timeout changed\n"));
1442 * FAST_PERIODIC -> SLOW_PERIODIC
1444 * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC
1446 * let lacp_sm_ptx_tx_schedule to update timeout.
1449 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1452 * if timeout has been shortened, assert NTT.
1455 if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) {
1456 lacp_sm_assert_ntt(lp);
1461 lacp_sm_ptx_tx_schedule(struct lacp_port *lp)
1465 if (!(lp->lp_state & LACP_STATE_ACTIVITY) &&
1466 !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) {
1472 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1476 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) {
1480 timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ?
1481 LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME;
1483 LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout);
1487 lacp_sm_ptx_timer(struct lacp_port *lp)
1489 lacp_sm_assert_ntt(lp);
1493 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du)
1498 * check LACP_DISABLED first
1501 if (!(lp->lp_state & LACP_STATE_AGGREGATION)) {
1506 * check loopback condition.
1509 if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid,
1510 &lp->lp_actor.lip_systemid)) {
1515 * EXPIRED, DEFAULTED, CURRENT -> CURRENT
1518 lacp_sm_rx_update_selected(lp, du);
1519 lacp_sm_rx_update_ntt(lp, du);
1520 lacp_sm_rx_record_pdu(lp, du);
1522 timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ?
1523 LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME;
1524 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout);
1526 lp->lp_state &= ~LACP_STATE_EXPIRED;
1529 * kick transmit machine without waiting the next tick.
1536 lacp_sm_rx_set_expired(struct lacp_port *lp)
1538 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1539 lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT;
1540 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME);
1541 lp->lp_state |= LACP_STATE_EXPIRED;
1545 lacp_sm_rx_timer(struct lacp_port *lp)
1547 if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) {
1548 /* CURRENT -> EXPIRED */
1549 LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__));
1550 lacp_sm_rx_set_expired(lp);
1552 /* EXPIRED -> DEFAULTED */
1553 LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__));
1554 lacp_sm_rx_update_default_selected(lp);
1555 lacp_sm_rx_record_default(lp);
1556 lp->lp_state &= ~LACP_STATE_EXPIRED;
1561 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du)
1565 char buf[LACP_STATESTR_MAX+1];
1569 oldpstate = lp->lp_partner.lip_state;
1571 active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY)
1572 || ((lp->lp_state & LACP_STATE_ACTIVITY) &&
1573 (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY));
1575 lp->lp_partner = du->ldu_actor;
1577 ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1578 LACP_STATE_AGGREGATION) &&
1579 !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner))
1580 || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) {
1583 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1586 lp->lp_state &= ~LACP_STATE_DEFAULTED;
1588 if (oldpstate != lp->lp_partner.lip_state) {
1589 LACP_DPRINTF((lp, "old pstate %s\n",
1590 lacp_format_state(oldpstate, buf, sizeof(buf))));
1591 LACP_DPRINTF((lp, "new pstate %s\n",
1592 lacp_format_state(lp->lp_partner.lip_state, buf,
1596 /* XXX Hack, still need to implement 5.4.9 para 2,3,4 */
1598 lp->lp_partner.lip_state |= LACP_STATE_SYNC;
1600 lacp_sm_ptx_update_timeout(lp, oldpstate);
1604 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du)
1609 if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) ||
1610 !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1611 LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) {
1612 LACP_DPRINTF((lp, "%s: assert ntt\n", __func__));
1613 lacp_sm_assert_ntt(lp);
1618 lacp_sm_rx_record_default(struct lacp_port *lp)
1624 oldpstate = lp->lp_partner.lip_state;
1626 lp->lp_partner = lacp_partner_admin_strict;
1628 lp->lp_partner = lacp_partner_admin_optimistic;;
1629 lp->lp_state |= LACP_STATE_DEFAULTED;
1630 lacp_sm_ptx_update_timeout(lp, oldpstate);
1634 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp,
1635 const struct lacp_peerinfo *info)
1640 if (lacp_compare_peerinfo(&lp->lp_partner, info) ||
1641 !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state,
1642 LACP_STATE_AGGREGATION)) {
1643 lp->lp_selected = LACP_UNSELECTED;
1644 /* mux machine will clean up lp->lp_aggregator */
1649 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du)
1654 lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor);
1658 lacp_sm_rx_update_default_selected(struct lacp_port *lp)
1664 lacp_sm_rx_update_selected_from_peerinfo(lp,
1665 &lacp_partner_admin_strict);
1667 lacp_sm_rx_update_selected_from_peerinfo(lp,
1668 &lacp_partner_admin_optimistic);
1671 /* transmit machine */
1674 lacp_sm_tx(struct lacp_port *lp)
1678 if (!(lp->lp_state & LACP_STATE_AGGREGATION)
1680 || (!(lp->lp_state & LACP_STATE_ACTIVITY)
1681 && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY))
1684 lp->lp_flags &= ~LACP_PORT_NTT;
1687 if (!(lp->lp_flags & LACP_PORT_NTT)) {
1691 /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */
1692 if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent,
1693 (3 / LACP_FAST_PERIODIC_TIME)) == 0) {
1694 LACP_DPRINTF((lp, "rate limited pdu\n"));
1698 if (((1 << lp->lp_ifp->if_dunit) & lacp_tx_test) == 0)
1699 error = lacp_xmit_lacpdu(lp);
1701 LACP_TPRINTF((lp, "Dropping TX PDU\n"));
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);