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 VNET_DEFINE(int, lacp_debug);
191 #define V_lacp_debug VNET(lacp_debug)
192 SYSCTL_NODE(_net_link_lagg, OID_AUTO, lacp, CTLFLAG_RD, 0, "ieee802.3ad");
193 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, debug, CTLFLAG_RWTUN | CTLFLAG_VNET,
194 &VNET_NAME(lacp_debug), 0, "Enable LACP debug logging (1=debug, 2=trace)");
196 static VNET_DEFINE(int, lacp_default_strict_mode) = 1;
197 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, default_strict_mode, CTLFLAG_RWTUN,
198 &VNET_NAME(lacp_default_strict_mode), 0,
199 "LACP strict protocol compliance default");
201 #define LACP_DPRINTF(a) if (V_lacp_debug & 0x01) { lacp_dprintf a ; }
202 #define LACP_TRACE(a) if (V_lacp_debug & 0x02) { lacp_dprintf(a,"%s\n",__func__); }
203 #define LACP_TPRINTF(a) if (V_lacp_debug & 0x04) { lacp_dprintf a ; }
206 * partner administration variables.
207 * XXX should be configurable.
210 static const struct lacp_peerinfo lacp_partner_admin_optimistic = {
211 .lip_systemid = { .lsi_prio = 0xffff },
212 .lip_portid = { .lpi_prio = 0xffff },
213 .lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION |
214 LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING,
217 static const struct lacp_peerinfo lacp_partner_admin_strict = {
218 .lip_systemid = { .lsi_prio = 0xffff },
219 .lip_portid = { .lpi_prio = 0xffff },
223 static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = {
224 [LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer,
225 [LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer,
226 [LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer,
230 lacp_input(struct lagg_port *lgp, struct mbuf *m)
232 struct lacp_port *lp = LACP_PORT(lgp);
235 if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) {
240 m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype);
242 case SLOWPROTOCOLS_SUBTYPE_LACP:
243 lacp_pdu_input(lp, m);
246 case SLOWPROTOCOLS_SUBTYPE_MARKER:
247 lacp_marker_input(lp, m);
251 /* Not a subtype we are interested in */
256 * lacp_pdu_input: process lacpdu
259 lacp_pdu_input(struct lacp_port *lp, struct mbuf *m)
261 struct lacp_softc *lsc = lp->lp_lsc;
265 if (m->m_pkthdr.len != sizeof(*du)) {
269 if ((m->m_flags & M_MCAST) == 0) {
273 if (m->m_len < sizeof(*du)) {
274 m = m_pullup(m, sizeof(*du));
280 du = mtod(m, struct lacpdu *);
282 if (memcmp(&du->ldu_eh.ether_dhost,
283 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
288 * ignore the version for compatibility with
289 * the future protocol revisions.
292 if (du->ldu_sph.sph_version != 1) {
298 * ignore tlv types for compatibility with
299 * the future protocol revisions.
301 if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor,
302 lacp_info_tlv_template, FALSE)) {
306 if (V_lacp_debug > 0) {
307 lacp_dprintf(lp, "lacpdu receive\n");
308 lacp_dump_lacpdu(du);
311 if ((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_rx_test) {
312 LACP_TPRINTF((lp, "Dropping RX PDU\n"));
329 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info)
331 struct lagg_port *lgp = lp->lp_lagg;
332 struct lagg_softc *sc = lgp->lp_softc;
334 info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO);
335 memcpy(&info->lip_systemid.lsi_mac,
336 IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
337 info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO);
338 info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index);
339 info->lip_state = lp->lp_state;
343 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info)
345 struct ifnet *ifp = lp->lp_ifp;
347 /* Fill in the port index and system id (encoded as the MAC) */
348 info->mi_rq_port = htons(ifp->if_index);
349 memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN);
350 info->mi_rq_xid = htonl(0);
354 lacp_xmit_lacpdu(struct lacp_port *lp)
356 struct lagg_port *lgp = lp->lp_lagg;
361 LACP_LOCK_ASSERT(lp->lp_lsc);
363 m = m_gethdr(M_NOWAIT, MT_DATA);
367 m->m_len = m->m_pkthdr.len = sizeof(*du);
369 du = mtod(m, struct lacpdu *);
370 memset(du, 0, sizeof(*du));
372 memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
374 memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
375 du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW);
377 du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP;
378 du->ldu_sph.sph_version = 1;
380 TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor));
381 du->ldu_actor = lp->lp_actor;
383 TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO,
384 sizeof(du->ldu_partner));
385 du->ldu_partner = lp->lp_partner;
387 TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO,
388 sizeof(du->ldu_collector));
389 du->ldu_collector.lci_maxdelay = 0;
391 if (V_lacp_debug > 0) {
392 lacp_dprintf(lp, "lacpdu transmit\n");
393 lacp_dump_lacpdu(du);
396 m->m_flags |= M_MCAST;
399 * XXX should use higher priority queue.
400 * otherwise network congestion can break aggregation.
403 error = lagg_enqueue(lp->lp_ifp, m);
408 lacp_xmit_marker(struct lacp_port *lp)
410 struct lagg_port *lgp = lp->lp_lagg;
412 struct markerdu *mdu;
415 LACP_LOCK_ASSERT(lp->lp_lsc);
417 m = m_gethdr(M_NOWAIT, MT_DATA);
421 m->m_len = m->m_pkthdr.len = sizeof(*mdu);
423 mdu = mtod(m, struct markerdu *);
424 memset(mdu, 0, sizeof(*mdu));
426 memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
428 memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
429 mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW);
431 mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER;
432 mdu->mdu_sph.sph_version = 1;
434 /* Bump the transaction id and copy over the marker info */
435 lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1);
436 TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info));
437 mdu->mdu_info = lp->lp_marker;
439 LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n",
440 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":",
441 ntohl(mdu->mdu_info.mi_rq_xid)));
443 m->m_flags |= M_MCAST;
444 error = lagg_enqueue(lp->lp_ifp, m);
449 lacp_linkstate(struct lagg_port *lgp)
451 struct lacp_port *lp = LACP_PORT(lgp);
452 struct lacp_softc *lsc = lp->lp_lsc;
453 struct ifnet *ifp = lgp->lp_ifp;
454 struct ifmediareq ifmr;
460 bzero((char *)&ifmr, sizeof(ifmr));
461 error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr);
466 media = ifmr.ifm_active;
467 LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, "
468 "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER,
469 (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP));
470 old_state = lp->lp_state;
471 old_key = lp->lp_key;
473 lp->lp_media = media;
475 * If the port is not an active full duplex Ethernet link then it can
478 if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 ||
479 ifp->if_link_state != LINK_STATE_UP) {
480 lacp_port_disable(lp);
482 lacp_port_enable(lp);
484 lp->lp_key = lacp_compose_key(lp);
486 if (old_state != lp->lp_state || old_key != lp->lp_key) {
487 LACP_DPRINTF((lp, "-> UNSELECTED\n"));
488 lp->lp_selected = LACP_UNSELECTED;
496 struct lacp_softc *lsc = arg;
497 struct lacp_port *lp;
499 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
500 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0)
503 CURVNET_SET(lp->lp_ifp->if_vnet);
509 lacp_sm_ptx_tx_schedule(lp);
512 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
516 lacp_port_create(struct lagg_port *lgp)
518 struct lagg_softc *sc = lgp->lp_softc;
519 struct lacp_softc *lsc = LACP_SOFTC(sc);
520 struct lacp_port *lp;
521 struct ifnet *ifp = lgp->lp_ifp;
522 struct sockaddr_dl sdl;
523 struct ifmultiaddr *rifma = NULL;
526 bzero((char *)&sdl, sizeof(sdl));
527 sdl.sdl_len = sizeof(sdl);
528 sdl.sdl_family = AF_LINK;
529 sdl.sdl_index = ifp->if_index;
530 sdl.sdl_type = IFT_ETHER;
531 sdl.sdl_alen = ETHER_ADDR_LEN;
533 bcopy(ðermulticastaddr_slowprotocols,
534 LLADDR(&sdl), ETHER_ADDR_LEN);
535 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
537 printf("%s: ADDMULTI failed on %s\n", __func__, lgp->lp_ifname);
541 lp = malloc(sizeof(struct lacp_port),
542 M_DEVBUF, M_NOWAIT|M_ZERO);
547 lgp->lp_psc = (caddr_t)lp;
553 LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next);
555 lacp_fill_actorinfo(lp, &lp->lp_actor);
556 lacp_fill_markerinfo(lp, &lp->lp_marker);
557 lp->lp_state = LACP_STATE_ACTIVITY;
558 lp->lp_aggregator = NULL;
559 lacp_sm_rx_set_expired(lp);
567 lacp_port_destroy(struct lagg_port *lgp)
569 struct lacp_port *lp = LACP_PORT(lgp);
570 struct lacp_softc *lsc = lp->lp_lsc;
574 for (i = 0; i < LACP_NTIMER; i++) {
575 LACP_TIMER_DISARM(lp, i);
578 lacp_disable_collecting(lp);
579 lacp_disable_distributing(lp);
582 LIST_REMOVE(lp, lp_next);
585 /* The address may have already been removed by if_purgemaddrs() */
586 if (!lgp->lp_detaching)
587 if_delmulti_ifma(lp->lp_ifma);
593 lacp_req(struct lagg_softc *sc, caddr_t data)
595 struct lacp_opreq *req = (struct lacp_opreq *)data;
596 struct lacp_softc *lsc = LACP_SOFTC(sc);
597 struct lacp_aggregator *la;
599 bzero(req, sizeof(struct lacp_opreq));
602 * If the LACP softc is NULL, return with the opreq structure full of
603 * zeros. It is normal for the softc to be NULL while the lagg is
609 la = lsc->lsc_active_aggregator;
612 req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio);
613 memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac,
615 req->actor_key = ntohs(la->la_actor.lip_key);
616 req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio);
617 req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno);
618 req->actor_state = la->la_actor.lip_state;
620 req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio);
621 memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac,
623 req->partner_key = ntohs(la->la_partner.lip_key);
624 req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio);
625 req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno);
626 req->partner_state = la->la_partner.lip_state;
632 lacp_portreq(struct lagg_port *lgp, caddr_t data)
634 struct lacp_opreq *req = (struct lacp_opreq *)data;
635 struct lacp_port *lp = LACP_PORT(lgp);
636 struct lacp_softc *lsc = lp->lp_lsc;
639 req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio);
640 memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac,
642 req->actor_key = ntohs(lp->lp_actor.lip_key);
643 req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio);
644 req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno);
645 req->actor_state = lp->lp_actor.lip_state;
647 req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio);
648 memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac,
650 req->partner_key = ntohs(lp->lp_partner.lip_key);
651 req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio);
652 req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno);
653 req->partner_state = lp->lp_partner.lip_state;
658 lacp_disable_collecting(struct lacp_port *lp)
660 LACP_DPRINTF((lp, "collecting disabled\n"));
661 lp->lp_state &= ~LACP_STATE_COLLECTING;
665 lacp_enable_collecting(struct lacp_port *lp)
667 LACP_DPRINTF((lp, "collecting enabled\n"));
668 lp->lp_state |= LACP_STATE_COLLECTING;
672 lacp_disable_distributing(struct lacp_port *lp)
674 struct lacp_aggregator *la = lp->lp_aggregator;
675 struct lacp_softc *lsc = lp->lp_lsc;
676 struct lagg_softc *sc = lsc->lsc_softc;
677 char buf[LACP_LAGIDSTR_MAX+1];
679 LACP_LOCK_ASSERT(lsc);
681 if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) {
685 KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports"));
686 KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports));
687 KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid"));
689 LACP_DPRINTF((lp, "disable distributing on aggregator %s, "
691 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
692 la->la_nports, la->la_nports - 1));
694 TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q);
696 sc->sc_active = la->la_nports;
698 if (lsc->lsc_active_aggregator == la) {
699 lacp_suppress_distributing(lsc, la);
700 lacp_select_active_aggregator(lsc);
701 /* regenerate the port map, the active aggregator has changed */
702 lacp_update_portmap(lsc);
705 lp->lp_state &= ~LACP_STATE_DISTRIBUTING;
709 lacp_enable_distributing(struct lacp_port *lp)
711 struct lacp_aggregator *la = lp->lp_aggregator;
712 struct lacp_softc *lsc = lp->lp_lsc;
713 struct lagg_softc *sc = lsc->lsc_softc;
714 char buf[LACP_LAGIDSTR_MAX+1];
716 LACP_LOCK_ASSERT(lsc);
718 if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) {
722 LACP_DPRINTF((lp, "enable distributing on aggregator %s, "
724 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
725 la->la_nports, la->la_nports + 1));
727 KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid"));
728 TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q);
730 sc->sc_active = la->la_nports;
732 lp->lp_state |= LACP_STATE_DISTRIBUTING;
734 if (lsc->lsc_active_aggregator == la) {
735 lacp_suppress_distributing(lsc, la);
736 lacp_update_portmap(lsc);
738 /* try to become the active aggregator */
739 lacp_select_active_aggregator(lsc);
743 lacp_transit_expire(void *vp)
745 struct lacp_softc *lsc = vp;
747 LACP_LOCK_ASSERT(lsc);
749 CURVNET_SET(lsc->lsc_softc->sc_ifp->if_vnet);
753 lsc->lsc_suppress_distributing = FALSE;
757 lacp_attach(struct lagg_softc *sc)
759 struct lacp_softc *lsc;
761 lsc = malloc(sizeof(struct lacp_softc), M_DEVBUF, M_WAITOK | M_ZERO);
763 sc->sc_psc = (caddr_t)lsc;
766 lsc->lsc_hashkey = arc4random();
767 lsc->lsc_active_aggregator = NULL;
768 lsc->lsc_strict_mode = VNET(lacp_default_strict_mode);
770 TAILQ_INIT(&lsc->lsc_aggregators);
771 LIST_INIT(&lsc->lsc_ports);
773 callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0);
774 callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0);
776 /* if the lagg is already up then do the same */
777 if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING)
782 lacp_detach(void *psc)
784 struct lacp_softc *lsc = (struct lacp_softc *)psc;
786 KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators),
787 ("aggregators still active"));
788 KASSERT(lsc->lsc_active_aggregator == NULL,
789 ("aggregator still attached"));
791 callout_drain(&lsc->lsc_transit_callout);
792 callout_drain(&lsc->lsc_callout);
794 LACP_LOCK_DESTROY(lsc);
800 lacp_init(struct lagg_softc *sc)
802 struct lacp_softc *lsc = LACP_SOFTC(sc);
805 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
810 lacp_stop(struct lagg_softc *sc)
812 struct lacp_softc *lsc = LACP_SOFTC(sc);
815 callout_stop(&lsc->lsc_transit_callout);
816 callout_stop(&lsc->lsc_callout);
821 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
823 struct lacp_softc *lsc = LACP_SOFTC(sc);
824 struct lacp_portmap *pm;
825 struct lacp_port *lp;
828 if (__predict_false(lsc->lsc_suppress_distributing)) {
829 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__));
833 pm = &lsc->lsc_pmap[lsc->lsc_activemap];
834 if (pm->pm_count == 0) {
835 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__));
839 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
840 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
841 hash = m->m_pkthdr.flowid >> sc->flowid_shift;
843 hash = lagg_hashmbuf(sc, m, lsc->lsc_hashkey);
844 hash %= pm->pm_count;
845 lp = pm->pm_map[hash];
847 KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0,
848 ("aggregated port is not distributing"));
850 return (lp->lp_lagg);
853 * lacp_suppress_distributing: drop transmit packets for a while
854 * to preserve packet ordering.
858 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la)
860 struct lacp_port *lp;
862 if (lsc->lsc_active_aggregator != la) {
868 lsc->lsc_suppress_distributing = TRUE;
870 /* send a marker frame down each port to verify the queues are empty */
871 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
872 lp->lp_flags |= LACP_PORT_MARK;
873 lacp_xmit_marker(lp);
876 /* set a timeout for the marker frames */
877 callout_reset(&lsc->lsc_transit_callout,
878 LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc);
882 lacp_compare_peerinfo(const struct lacp_peerinfo *a,
883 const struct lacp_peerinfo *b)
885 return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state)));
889 lacp_compare_systemid(const struct lacp_systemid *a,
890 const struct lacp_systemid *b)
892 return (memcmp(a, b, sizeof(*a)));
897 lacp_compare_portid(const struct lacp_portid *a,
898 const struct lacp_portid *b)
900 return (memcmp(a, b, sizeof(*a)));
905 lacp_aggregator_bandwidth(struct lacp_aggregator *la)
907 struct lacp_port *lp;
910 lp = TAILQ_FIRST(&la->la_ports);
915 speed = ifmedia_baudrate(lp->lp_media);
916 speed *= la->la_nports;
918 LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n",
919 lp->lp_media, la->la_nports));
926 * lacp_select_active_aggregator: select an aggregator to be used to transmit
927 * packets from lagg(4) interface.
931 lacp_select_active_aggregator(struct lacp_softc *lsc)
933 struct lacp_aggregator *la;
934 struct lacp_aggregator *best_la = NULL;
935 uint64_t best_speed = 0;
936 char buf[LACP_LAGIDSTR_MAX+1];
940 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
943 if (la->la_nports == 0) {
947 speed = lacp_aggregator_bandwidth(la);
948 LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n",
949 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
950 speed, la->la_nports));
952 /* This aggregator is chosen if
953 * the partner has a better system priority
954 * or, the total aggregated speed is higher
955 * or, it is already the chosen aggregator
957 if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) <
958 LACP_SYS_PRI(best_la->la_partner)) ||
959 speed > best_speed ||
960 (speed == best_speed &&
961 la == lsc->lsc_active_aggregator)) {
967 KASSERT(best_la == NULL || best_la->la_nports > 0,
968 ("invalid aggregator refcnt"));
969 KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports),
970 ("invalid aggregator list"));
972 if (lsc->lsc_active_aggregator != best_la) {
973 LACP_DPRINTF((NULL, "active aggregator changed\n"));
974 LACP_DPRINTF((NULL, "old %s\n",
975 lacp_format_lagid_aggregator(lsc->lsc_active_aggregator,
978 LACP_DPRINTF((NULL, "active aggregator not changed\n"));
980 LACP_DPRINTF((NULL, "new %s\n",
981 lacp_format_lagid_aggregator(best_la, buf, sizeof(buf))));
983 if (lsc->lsc_active_aggregator != best_la) {
984 lsc->lsc_active_aggregator = best_la;
985 lacp_update_portmap(lsc);
987 lacp_suppress_distributing(lsc, best_la);
993 * Updated the inactive portmap array with the new list of ports and
997 lacp_update_portmap(struct lacp_softc *lsc)
999 struct lagg_softc *sc = lsc->lsc_softc;
1000 struct lacp_aggregator *la;
1001 struct lacp_portmap *p;
1002 struct lacp_port *lp;
1007 newmap = lsc->lsc_activemap == 0 ? 1 : 0;
1008 p = &lsc->lsc_pmap[newmap];
1009 la = lsc->lsc_active_aggregator;
1011 bzero(p, sizeof(struct lacp_portmap));
1013 if (la != NULL && la->la_nports > 0) {
1014 p->pm_count = la->la_nports;
1016 TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q)
1017 p->pm_map[i++] = lp;
1018 KASSERT(i == p->pm_count, ("Invalid port count"));
1019 speed = lacp_aggregator_bandwidth(la);
1021 sc->sc_ifp->if_baudrate = speed;
1023 /* switch the active portmap over */
1024 atomic_store_rel_int(&lsc->lsc_activemap, newmap);
1025 LACP_DPRINTF((NULL, "Set table %d with %d ports\n",
1027 lsc->lsc_pmap[lsc->lsc_activemap].pm_count));
1031 lacp_compose_key(struct lacp_port *lp)
1033 struct lagg_port *lgp = lp->lp_lagg;
1034 struct lagg_softc *sc = lgp->lp_softc;
1035 u_int media = lp->lp_media;
1038 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) {
1041 * non-aggregatable links should have unique keys.
1043 * XXX this isn't really unique as if_index is 16 bit.
1046 /* bit 0..14: (some bits of) if_index of this port */
1047 key = lp->lp_ifp->if_index;
1051 u_int subtype = IFM_SUBTYPE(media);
1053 KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type"));
1054 KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface"));
1056 /* bit 0..4: IFM_SUBTYPE modulo speed */
1078 case IFM_1000_SGMII:
1079 case IFM_1000_CX_SGMII:
1085 case IFM_10G_TWINAX:
1086 case IFM_10G_TWINAX_LONG:
1137 /* bit 5..14: (some bits of) if_index of lagg device */
1138 key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5);
1141 return (htons(key));
1145 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1147 char buf[LACP_LAGIDSTR_MAX+1];
1149 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1151 lacp_format_lagid(&la->la_actor, &la->la_partner,
1153 la->la_refcnt, la->la_refcnt + 1));
1155 KASSERT(la->la_refcnt > 0, ("refcount <= 0"));
1157 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount"));
1161 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1163 char buf[LACP_LAGIDSTR_MAX+1];
1165 LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1167 lacp_format_lagid(&la->la_actor, &la->la_partner,
1169 la->la_refcnt, la->la_refcnt - 1));
1171 KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt"));
1173 if (la->la_refcnt > 0) {
1177 KASSERT(la->la_refcnt == 0, ("refcount not zero"));
1178 KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active"));
1180 TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q);
1186 * lacp_aggregator_get: allocate an aggregator.
1189 static struct lacp_aggregator *
1190 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp)
1192 struct lacp_aggregator *la;
1194 la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT);
1198 TAILQ_INIT(&la->la_ports);
1200 TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q);
1207 * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port.
1211 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp)
1213 lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner);
1214 lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor);
1216 la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION;
1220 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr,
1221 const struct lacp_peerinfo *lpi_port)
1223 memset(lpi_aggr, 0, sizeof(*lpi_aggr));
1224 lpi_aggr->lip_systemid = lpi_port->lip_systemid;
1225 lpi_aggr->lip_key = lpi_port->lip_key;
1229 * lacp_aggregator_is_compatible: check if a port can join to an aggregator.
1233 lacp_aggregator_is_compatible(const struct lacp_aggregator *la,
1234 const struct lacp_port *lp)
1236 if (!(lp->lp_state & LACP_STATE_AGGREGATION) ||
1237 !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) {
1241 if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) {
1245 if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) {
1249 if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) {
1257 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a,
1258 const struct lacp_peerinfo *b)
1260 if (memcmp(&a->lip_systemid, &b->lip_systemid,
1261 sizeof(a->lip_systemid))) {
1265 if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) {
1273 lacp_port_enable(struct lacp_port *lp)
1275 lp->lp_state |= LACP_STATE_AGGREGATION;
1279 lacp_port_disable(struct lacp_port *lp)
1281 lacp_set_mux(lp, LACP_MUX_DETACHED);
1283 lp->lp_state &= ~LACP_STATE_AGGREGATION;
1284 lp->lp_selected = LACP_UNSELECTED;
1285 lacp_sm_rx_record_default(lp);
1286 lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION;
1287 lp->lp_state &= ~LACP_STATE_EXPIRED;
1291 * lacp_select: select an aggregator. create one if necessary.
1294 lacp_select(struct lacp_port *lp)
1296 struct lacp_softc *lsc = lp->lp_lsc;
1297 struct lacp_aggregator *la;
1298 char buf[LACP_LAGIDSTR_MAX+1];
1300 if (lp->lp_aggregator) {
1304 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1305 ("timer_wait_while still active"));
1307 LACP_DPRINTF((lp, "port lagid=%s\n",
1308 lacp_format_lagid(&lp->lp_actor, &lp->lp_partner,
1309 buf, sizeof(buf))));
1311 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
1312 if (lacp_aggregator_is_compatible(la, lp)) {
1318 la = lacp_aggregator_get(lsc, lp);
1320 LACP_DPRINTF((lp, "aggregator creation failed\n"));
1323 * will retry on the next tick.
1328 lacp_fill_aggregator_id(la, lp);
1329 LACP_DPRINTF((lp, "aggregator created\n"));
1331 LACP_DPRINTF((lp, "compatible aggregator found\n"));
1332 if (la->la_refcnt == LACP_MAX_PORTS)
1334 lacp_aggregator_addref(lsc, la);
1337 LACP_DPRINTF((lp, "aggregator lagid=%s\n",
1338 lacp_format_lagid(&la->la_actor, &la->la_partner,
1339 buf, sizeof(buf))));
1341 lp->lp_aggregator = la;
1342 lp->lp_selected = LACP_SELECTED;
1346 * lacp_unselect: finish unselect/detach process.
1350 lacp_unselect(struct lacp_port *lp)
1352 struct lacp_softc *lsc = lp->lp_lsc;
1353 struct lacp_aggregator *la = lp->lp_aggregator;
1355 KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1356 ("timer_wait_while still active"));
1362 lp->lp_aggregator = NULL;
1363 lacp_aggregator_delref(lsc, la);
1369 lacp_sm_mux(struct lacp_port *lp)
1371 struct lagg_port *lgp = lp->lp_lagg;
1372 struct lagg_softc *sc = lgp->lp_softc;
1373 enum lacp_mux_state new_state;
1375 (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0;
1376 boolean_t p_collecting =
1377 (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0;
1378 enum lacp_selected selected = lp->lp_selected;
1379 struct lacp_aggregator *la;
1381 if (V_lacp_debug > 1)
1382 lacp_dprintf(lp, "%s: state= 0x%x, selected= 0x%x, "
1383 "p_sync= 0x%x, p_collecting= 0x%x\n", __func__,
1384 lp->lp_mux_state, selected, p_sync, p_collecting);
1387 la = lp->lp_aggregator;
1388 KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL,
1389 ("MUX not detached"));
1390 new_state = lp->lp_mux_state;
1391 switch (lp->lp_mux_state) {
1392 case LACP_MUX_DETACHED:
1393 if (selected != LACP_UNSELECTED) {
1394 new_state = LACP_MUX_WAITING;
1397 case LACP_MUX_WAITING:
1398 KASSERT(la->la_pending > 0 ||
1399 !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1400 ("timer_wait_while still active"));
1401 if (selected == LACP_SELECTED && la->la_pending == 0) {
1402 new_state = LACP_MUX_ATTACHED;
1403 } else if (selected == LACP_UNSELECTED) {
1404 new_state = LACP_MUX_DETACHED;
1407 case LACP_MUX_ATTACHED:
1408 if (selected == LACP_SELECTED && p_sync) {
1409 new_state = LACP_MUX_COLLECTING;
1410 } else if (selected != LACP_SELECTED) {
1411 new_state = LACP_MUX_DETACHED;
1414 case LACP_MUX_COLLECTING:
1415 if (selected == LACP_SELECTED && p_sync && p_collecting) {
1416 new_state = LACP_MUX_DISTRIBUTING;
1417 } else if (selected != LACP_SELECTED || !p_sync) {
1418 new_state = LACP_MUX_ATTACHED;
1421 case LACP_MUX_DISTRIBUTING:
1422 if (selected != LACP_SELECTED || !p_sync || !p_collecting) {
1423 new_state = LACP_MUX_COLLECTING;
1424 lacp_dprintf(lp, "Interface stopped DISTRIBUTING, possible flapping\n");
1429 panic("%s: unknown state", __func__);
1432 if (lp->lp_mux_state == new_state) {
1436 lacp_set_mux(lp, new_state);
1441 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state)
1443 struct lacp_aggregator *la = lp->lp_aggregator;
1445 if (lp->lp_mux_state == new_state) {
1449 switch (new_state) {
1450 case LACP_MUX_DETACHED:
1451 lp->lp_state &= ~LACP_STATE_SYNC;
1452 lacp_disable_distributing(lp);
1453 lacp_disable_collecting(lp);
1454 lacp_sm_assert_ntt(lp);
1456 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) {
1457 KASSERT(la->la_pending > 0,
1458 ("timer_wait_while not active"));
1461 LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE);
1464 case LACP_MUX_WAITING:
1465 LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE,
1466 LACP_AGGREGATE_WAIT_TIME);
1469 case LACP_MUX_ATTACHED:
1470 lp->lp_state |= LACP_STATE_SYNC;
1471 lacp_disable_collecting(lp);
1472 lacp_sm_assert_ntt(lp);
1474 case LACP_MUX_COLLECTING:
1475 lacp_enable_collecting(lp);
1476 lacp_disable_distributing(lp);
1477 lacp_sm_assert_ntt(lp);
1479 case LACP_MUX_DISTRIBUTING:
1480 lacp_enable_distributing(lp);
1483 panic("%s: unknown state", __func__);
1486 LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state));
1488 lp->lp_mux_state = new_state;
1492 lacp_sm_mux_timer(struct lacp_port *lp)
1494 struct lacp_aggregator *la = lp->lp_aggregator;
1495 char buf[LACP_LAGIDSTR_MAX+1];
1497 KASSERT(la->la_pending > 0, ("no pending event"));
1499 LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__,
1500 lacp_format_lagid(&la->la_actor, &la->la_partner,
1502 la->la_pending, la->la_pending - 1));
1507 /* periodic transmit machine */
1510 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate)
1512 if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state,
1513 LACP_STATE_TIMEOUT)) {
1517 LACP_DPRINTF((lp, "partner timeout changed\n"));
1520 * FAST_PERIODIC -> SLOW_PERIODIC
1522 * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC
1524 * let lacp_sm_ptx_tx_schedule to update timeout.
1527 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1530 * if timeout has been shortened, assert NTT.
1533 if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) {
1534 lacp_sm_assert_ntt(lp);
1539 lacp_sm_ptx_tx_schedule(struct lacp_port *lp)
1543 if (!(lp->lp_state & LACP_STATE_ACTIVITY) &&
1544 !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) {
1550 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1554 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) {
1558 timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ?
1559 LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME;
1561 LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout);
1565 lacp_sm_ptx_timer(struct lacp_port *lp)
1567 lacp_sm_assert_ntt(lp);
1571 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du)
1576 * check LACP_DISABLED first
1579 if (!(lp->lp_state & LACP_STATE_AGGREGATION)) {
1584 * check loopback condition.
1587 if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid,
1588 &lp->lp_actor.lip_systemid)) {
1593 * EXPIRED, DEFAULTED, CURRENT -> CURRENT
1596 lacp_sm_rx_update_selected(lp, du);
1597 lacp_sm_rx_update_ntt(lp, du);
1598 lacp_sm_rx_record_pdu(lp, du);
1600 timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ?
1601 LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME;
1602 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout);
1604 lp->lp_state &= ~LACP_STATE_EXPIRED;
1607 * kick transmit machine without waiting the next tick.
1614 lacp_sm_rx_set_expired(struct lacp_port *lp)
1616 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1617 lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT;
1618 LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME);
1619 lp->lp_state |= LACP_STATE_EXPIRED;
1623 lacp_sm_rx_timer(struct lacp_port *lp)
1625 if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) {
1626 /* CURRENT -> EXPIRED */
1627 LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__));
1628 lacp_sm_rx_set_expired(lp);
1630 /* EXPIRED -> DEFAULTED */
1631 LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__));
1632 lacp_sm_rx_update_default_selected(lp);
1633 lacp_sm_rx_record_default(lp);
1634 lp->lp_state &= ~LACP_STATE_EXPIRED;
1639 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du)
1643 char buf[LACP_STATESTR_MAX+1];
1647 oldpstate = lp->lp_partner.lip_state;
1649 active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY)
1650 || ((lp->lp_state & LACP_STATE_ACTIVITY) &&
1651 (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY));
1653 lp->lp_partner = du->ldu_actor;
1655 ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1656 LACP_STATE_AGGREGATION) &&
1657 !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner))
1658 || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) {
1661 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1664 lp->lp_state &= ~LACP_STATE_DEFAULTED;
1666 if (oldpstate != lp->lp_partner.lip_state) {
1667 LACP_DPRINTF((lp, "old pstate %s\n",
1668 lacp_format_state(oldpstate, buf, sizeof(buf))));
1669 LACP_DPRINTF((lp, "new pstate %s\n",
1670 lacp_format_state(lp->lp_partner.lip_state, buf,
1674 /* XXX Hack, still need to implement 5.4.9 para 2,3,4 */
1675 if (lp->lp_lsc->lsc_strict_mode)
1676 lp->lp_partner.lip_state |= LACP_STATE_SYNC;
1678 lacp_sm_ptx_update_timeout(lp, oldpstate);
1682 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du)
1687 if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) ||
1688 !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1689 LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) {
1690 LACP_DPRINTF((lp, "%s: assert ntt\n", __func__));
1691 lacp_sm_assert_ntt(lp);
1696 lacp_sm_rx_record_default(struct lacp_port *lp)
1702 oldpstate = lp->lp_partner.lip_state;
1703 if (lp->lp_lsc->lsc_strict_mode)
1704 lp->lp_partner = lacp_partner_admin_strict;
1706 lp->lp_partner = lacp_partner_admin_optimistic;
1707 lp->lp_state |= LACP_STATE_DEFAULTED;
1708 lacp_sm_ptx_update_timeout(lp, oldpstate);
1712 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp,
1713 const struct lacp_peerinfo *info)
1718 if (lacp_compare_peerinfo(&lp->lp_partner, info) ||
1719 !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state,
1720 LACP_STATE_AGGREGATION)) {
1721 lp->lp_selected = LACP_UNSELECTED;
1722 /* mux machine will clean up lp->lp_aggregator */
1727 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du)
1732 lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor);
1736 lacp_sm_rx_update_default_selected(struct lacp_port *lp)
1741 if (lp->lp_lsc->lsc_strict_mode)
1742 lacp_sm_rx_update_selected_from_peerinfo(lp,
1743 &lacp_partner_admin_strict);
1745 lacp_sm_rx_update_selected_from_peerinfo(lp,
1746 &lacp_partner_admin_optimistic);
1749 /* transmit machine */
1752 lacp_sm_tx(struct lacp_port *lp)
1756 if (!(lp->lp_state & LACP_STATE_AGGREGATION)
1758 || (!(lp->lp_state & LACP_STATE_ACTIVITY)
1759 && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY))
1762 lp->lp_flags &= ~LACP_PORT_NTT;
1765 if (!(lp->lp_flags & LACP_PORT_NTT)) {
1769 /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */
1770 if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent,
1771 (3 / LACP_FAST_PERIODIC_TIME)) == 0) {
1772 LACP_DPRINTF((lp, "rate limited pdu\n"));
1776 if (((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_tx_test) == 0) {
1777 error = lacp_xmit_lacpdu(lp);
1779 LACP_TPRINTF((lp, "Dropping TX PDU\n"));
1783 lp->lp_flags &= ~LACP_PORT_NTT;
1785 LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n",
1791 lacp_sm_assert_ntt(struct lacp_port *lp)
1794 lp->lp_flags |= LACP_PORT_NTT;
1798 lacp_run_timers(struct lacp_port *lp)
1802 for (i = 0; i < LACP_NTIMER; i++) {
1803 KASSERT(lp->lp_timer[i] >= 0,
1804 ("invalid timer value %d", lp->lp_timer[i]));
1805 if (lp->lp_timer[i] == 0) {
1807 } else if (--lp->lp_timer[i] <= 0) {
1808 if (lacp_timer_funcs[i]) {
1809 (*lacp_timer_funcs[i])(lp);
1816 lacp_marker_input(struct lacp_port *lp, struct mbuf *m)
1818 struct lacp_softc *lsc = lp->lp_lsc;
1819 struct lagg_port *lgp = lp->lp_lagg;
1820 struct lacp_port *lp2;
1821 struct markerdu *mdu;
1825 if (m->m_pkthdr.len != sizeof(*mdu)) {
1829 if ((m->m_flags & M_MCAST) == 0) {
1833 if (m->m_len < sizeof(*mdu)) {
1834 m = m_pullup(m, sizeof(*mdu));
1840 mdu = mtod(m, struct markerdu *);
1842 if (memcmp(&mdu->mdu_eh.ether_dhost,
1843 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
1847 if (mdu->mdu_sph.sph_version != 1) {
1851 switch (mdu->mdu_tlv.tlv_type) {
1852 case MARKER_TYPE_INFO:
1853 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
1854 marker_info_tlv_template, TRUE)) {
1857 mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE;
1858 memcpy(&mdu->mdu_eh.ether_dhost,
1859 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN);
1860 memcpy(&mdu->mdu_eh.ether_shost,
1861 lgp->lp_lladdr, ETHER_ADDR_LEN);
1862 error = lagg_enqueue(lp->lp_ifp, m);
1865 case MARKER_TYPE_RESPONSE:
1866 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
1867 marker_response_tlv_template, TRUE)) {
1870 LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n",
1871 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system,
1872 ":", ntohl(mdu->mdu_info.mi_rq_xid)));
1874 /* Verify that it is the last marker we sent out */
1875 if (memcmp(&mdu->mdu_info, &lp->lp_marker,
1876 sizeof(struct lacp_markerinfo)))
1880 lp->lp_flags &= ~LACP_PORT_MARK;
1882 if (lsc->lsc_suppress_distributing) {
1883 /* Check if any ports are waiting for a response */
1884 LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) {
1885 if (lp2->lp_flags & LACP_PORT_MARK) {
1892 /* All interface queues are clear */
1893 LACP_DPRINTF((NULL, "queue flush complete\n"));
1894 lsc->lsc_suppress_distributing = FALSE;
1908 LACP_DPRINTF((lp, "bad marker frame\n"));
1914 tlv_check(const void *p, size_t size, const struct tlvhdr *tlv,
1915 const struct tlv_template *tmpl, boolean_t check_type)
1917 while (/* CONSTCOND */ 1) {
1918 if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) {
1921 if ((check_type && tlv->tlv_type != tmpl->tmpl_type) ||
1922 tlv->tlv_length != tmpl->tmpl_length) {
1925 if (tmpl->tmpl_type == 0) {
1928 tlv = (const struct tlvhdr *)
1929 ((const char *)tlv + tlv->tlv_length);
1938 lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen)
1940 snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X",
1952 lacp_format_systemid(const struct lacp_systemid *sysid,
1953 char *buf, size_t buflen)
1955 char macbuf[LACP_MACSTR_MAX+1];
1957 snprintf(buf, buflen, "%04X,%s",
1958 ntohs(sysid->lsi_prio),
1959 lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf)));
1965 lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen)
1967 snprintf(buf, buflen, "%04X,%04X",
1968 ntohs(portid->lpi_prio),
1969 ntohs(portid->lpi_portno));
1975 lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen)
1977 char sysid[LACP_SYSTEMIDSTR_MAX+1];
1978 char portid[LACP_PORTIDSTR_MAX+1];
1980 snprintf(buf, buflen, "(%s,%04X,%s)",
1981 lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)),
1982 ntohs(peer->lip_key),
1983 lacp_format_portid(&peer->lip_portid, portid, sizeof(portid)));
1989 lacp_format_lagid(const struct lacp_peerinfo *a,
1990 const struct lacp_peerinfo *b, char *buf, size_t buflen)
1992 char astr[LACP_PARTNERSTR_MAX+1];
1993 char bstr[LACP_PARTNERSTR_MAX+1];
1997 * there's a convention to display small numbered peer
2001 if (lacp_compare_peerinfo(a, b) > 0) {
2002 const struct lacp_peerinfo *t;
2010 snprintf(buf, buflen, "[%s,%s]",
2011 lacp_format_partner(a, astr, sizeof(astr)),
2012 lacp_format_partner(b, bstr, sizeof(bstr)));
2018 lacp_format_lagid_aggregator(const struct lacp_aggregator *la,
2019 char *buf, size_t buflen)
2025 return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen));
2029 lacp_format_state(uint8_t state, char *buf, size_t buflen)
2031 snprintf(buf, buflen, "%b", state, LACP_STATE_BITS);
2036 lacp_dump_lacpdu(const struct lacpdu *du)
2038 char buf[LACP_PARTNERSTR_MAX+1];
2039 char buf2[LACP_STATESTR_MAX+1];
2041 printf("actor=%s\n",
2042 lacp_format_partner(&du->ldu_actor, buf, sizeof(buf)));
2043 printf("actor.state=%s\n",
2044 lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2)));
2045 printf("partner=%s\n",
2046 lacp_format_partner(&du->ldu_partner, buf, sizeof(buf)));
2047 printf("partner.state=%s\n",
2048 lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2)));
2050 printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay));
2054 lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...)
2059 printf("%s: ", lp->lp_ifp->if_xname);