xref: /freebsd/sys/net/ieee8023ad_lacp.c (revision 61b95bcb42993b24633b280791438266d78f2747)
1 /*	$NetBSD: ieee8023ad_lacp.c,v 1.3 2005/12/11 12:24:54 christos Exp $	*/
2 
3 /*-
4  * SPDX-License-Identifier: BSD-2-Clause-NetBSD
5  *
6  * Copyright (c)2005 YAMAMOTO Takashi,
7  * Copyright (c)2008 Andrew Thompson <thompsa@FreeBSD.org>
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_kern_tls.h"
36 #include "opt_ratelimit.h"
37 
38 #include <sys/param.h>
39 #include <sys/callout.h>
40 #include <sys/eventhandler.h>
41 #include <sys/mbuf.h>
42 #include <sys/systm.h>
43 #include <sys/malloc.h>
44 #include <sys/kernel.h> /* hz */
45 #include <sys/socket.h> /* for net/if.h */
46 #include <sys/sockio.h>
47 #include <sys/sysctl.h>
48 #include <machine/stdarg.h>
49 #include <sys/lock.h>
50 #include <sys/rwlock.h>
51 #include <sys/taskqueue.h>
52 #include <sys/time.h>
53 
54 #include <net/if.h>
55 #include <net/if_var.h>
56 #include <net/if_private.h>
57 #include <net/if_dl.h>
58 #include <net/ethernet.h>
59 #include <net/infiniband.h>
60 #include <net/if_media.h>
61 #include <net/if_types.h>
62 
63 #include <net/if_lagg.h>
64 #include <net/ieee8023ad_lacp.h>
65 
66 /*
67  * actor system priority and port priority.
68  * XXX should be configurable.
69  */
70 
71 #define	LACP_SYSTEM_PRIO	0x8000
72 #define	LACP_PORT_PRIO		0x8000
73 
74 const uint8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] =
75     { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 };
76 
77 static const struct tlv_template lacp_info_tlv_template[] = {
78 	{ LACP_TYPE_ACTORINFO,
79 	    sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
80 	{ LACP_TYPE_PARTNERINFO,
81 	    sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
82 	{ LACP_TYPE_COLLECTORINFO,
83 	    sizeof(struct tlvhdr) + sizeof(struct lacp_collectorinfo) },
84 	{ 0, 0 },
85 };
86 
87 static const struct tlv_template marker_info_tlv_template[] = {
88 	{ MARKER_TYPE_INFO,
89 	    sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
90 	{ 0, 0 },
91 };
92 
93 static const struct tlv_template marker_response_tlv_template[] = {
94 	{ MARKER_TYPE_RESPONSE,
95 	    sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
96 	{ 0, 0 },
97 };
98 
99 typedef void (*lacp_timer_func_t)(struct lacp_port *);
100 
101 static void	lacp_fill_actorinfo(struct lacp_port *, struct lacp_peerinfo *);
102 static void	lacp_fill_markerinfo(struct lacp_port *,
103 		    struct lacp_markerinfo *);
104 
105 static uint64_t	lacp_aggregator_bandwidth(struct lacp_aggregator *);
106 static void	lacp_suppress_distributing(struct lacp_softc *,
107 		    struct lacp_aggregator *);
108 static void	lacp_transit_expire(void *);
109 static void	lacp_update_portmap(struct lacp_softc *);
110 static void	lacp_select_active_aggregator(struct lacp_softc *);
111 static uint16_t	lacp_compose_key(struct lacp_port *);
112 static int	tlv_check(const void *, size_t, const struct tlvhdr *,
113 		    const struct tlv_template *, boolean_t);
114 static void	lacp_tick(void *);
115 
116 static void	lacp_fill_aggregator_id(struct lacp_aggregator *,
117 		    const struct lacp_port *);
118 static void	lacp_fill_aggregator_id_peer(struct lacp_peerinfo *,
119 		    const struct lacp_peerinfo *);
120 static bool	lacp_aggregator_is_compatible(const struct lacp_aggregator *,
121 		    const struct lacp_port *);
122 static bool	lacp_peerinfo_is_compatible(const struct lacp_peerinfo *,
123 		    const struct lacp_peerinfo *);
124 
125 static struct lacp_aggregator *lacp_aggregator_get(struct lacp_softc *,
126 		    struct lacp_port *);
127 static void	lacp_aggregator_addref(struct lacp_softc *,
128 		    struct lacp_aggregator *);
129 static void	lacp_aggregator_delref(struct lacp_softc *,
130 		    struct lacp_aggregator *);
131 
132 /* receive machine */
133 
134 static int	lacp_pdu_input(struct lacp_port *, struct mbuf *);
135 static int	lacp_marker_input(struct lacp_port *, struct mbuf *);
136 static void	lacp_sm_rx(struct lacp_port *, const struct lacpdu *);
137 static void	lacp_sm_rx_timer(struct lacp_port *);
138 static void	lacp_sm_rx_set_expired(struct lacp_port *);
139 static void	lacp_sm_rx_update_ntt(struct lacp_port *,
140 		    const struct lacpdu *);
141 static void	lacp_sm_rx_record_pdu(struct lacp_port *,
142 		    const struct lacpdu *);
143 static void	lacp_sm_rx_update_selected(struct lacp_port *,
144 		    const struct lacpdu *);
145 static void	lacp_sm_rx_record_default(struct lacp_port *);
146 static void	lacp_sm_rx_update_default_selected(struct lacp_port *);
147 static void	lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *,
148 		    const struct lacp_peerinfo *);
149 
150 /* mux machine */
151 
152 static void	lacp_sm_mux(struct lacp_port *);
153 static void	lacp_set_mux(struct lacp_port *, enum lacp_mux_state);
154 static void	lacp_sm_mux_timer(struct lacp_port *);
155 
156 /* periodic transmit machine */
157 
158 static void	lacp_sm_ptx_update_timeout(struct lacp_port *, uint8_t);
159 static void	lacp_sm_ptx_tx_schedule(struct lacp_port *);
160 static void	lacp_sm_ptx_timer(struct lacp_port *);
161 
162 /* transmit machine */
163 
164 static void	lacp_sm_tx(struct lacp_port *);
165 static void	lacp_sm_assert_ntt(struct lacp_port *);
166 
167 static void	lacp_run_timers(struct lacp_port *);
168 static int	lacp_compare_peerinfo(const struct lacp_peerinfo *,
169 		    const struct lacp_peerinfo *);
170 static int	lacp_compare_systemid(const struct lacp_systemid *,
171 		    const struct lacp_systemid *);
172 static void	lacp_port_enable(struct lacp_port *);
173 static void	lacp_port_disable(struct lacp_port *);
174 static void	lacp_select(struct lacp_port *);
175 static void	lacp_unselect(struct lacp_port *);
176 static void	lacp_disable_collecting(struct lacp_port *);
177 static void	lacp_enable_collecting(struct lacp_port *);
178 static void	lacp_disable_distributing(struct lacp_port *);
179 static void	lacp_enable_distributing(struct lacp_port *);
180 static int	lacp_xmit_lacpdu(struct lacp_port *);
181 static int	lacp_xmit_marker(struct lacp_port *);
182 
183 /* Debugging */
184 
185 static void	lacp_dump_lacpdu(const struct lacpdu *);
186 static const char *lacp_format_partner(const struct lacp_peerinfo *, char *,
187 		    size_t);
188 static const char *lacp_format_lagid(const struct lacp_peerinfo *,
189 		    const struct lacp_peerinfo *, char *, size_t);
190 static const char *lacp_format_lagid_aggregator(const struct lacp_aggregator *,
191 		    char *, size_t);
192 static const char *lacp_format_state(uint8_t, char *, size_t);
193 static const char *lacp_format_mac(const uint8_t *, char *, size_t);
194 static const char *lacp_format_systemid(const struct lacp_systemid *, char *,
195 		    size_t);
196 static const char *lacp_format_portid(const struct lacp_portid *, char *,
197 		    size_t);
198 static void	lacp_dprintf(const struct lacp_port *, const char *, ...)
199 		    __attribute__((__format__(__printf__, 2, 3)));
200 
201 VNET_DEFINE_STATIC(int, lacp_debug);
202 #define	V_lacp_debug	VNET(lacp_debug)
203 SYSCTL_NODE(_net_link_lagg, OID_AUTO, lacp, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
204     "ieee802.3ad");
205 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, debug, CTLFLAG_RWTUN | CTLFLAG_VNET,
206     &VNET_NAME(lacp_debug), 0, "Enable LACP debug logging (1=debug, 2=trace)");
207 
208 VNET_DEFINE_STATIC(int, lacp_default_strict_mode) = 1;
209 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, default_strict_mode,
210     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(lacp_default_strict_mode), 0,
211     "LACP strict protocol compliance default");
212 #define LACP_DPRINTF(a) if (V_lacp_debug & 0x01) { lacp_dprintf a ; }
213 #define LACP_TRACE(a) if (V_lacp_debug & 0x02) { lacp_dprintf(a,"%s\n",__func__); }
214 #define LACP_TPRINTF(a) if (V_lacp_debug & 0x04) { lacp_dprintf a ; }
215 
216 /*
217  * partner administration variables.
218  * XXX should be configurable.
219  */
220 
221 static const struct lacp_peerinfo lacp_partner_admin_optimistic = {
222 	.lip_systemid = { .lsi_prio = 0xffff },
223 	.lip_portid = { .lpi_prio = 0xffff },
224 	.lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION |
225 	    LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING,
226 };
227 
228 static const struct lacp_peerinfo lacp_partner_admin_strict = {
229 	.lip_systemid = { .lsi_prio = 0xffff },
230 	.lip_portid = { .lpi_prio = 0xffff },
231 	.lip_state = 0,
232 };
233 
234 static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = {
235 	[LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer,
236 	[LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer,
237 	[LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer,
238 };
239 
240 struct mbuf *
241 lacp_input(struct lagg_port *lgp, struct mbuf *m)
242 {
243 	struct lacp_port *lp = LACP_PORT(lgp);
244 	uint8_t subtype;
245 
246 	if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) {
247 		m_freem(m);
248 		return (NULL);
249 	}
250 
251 	m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype);
252 	switch (subtype) {
253 		case SLOWPROTOCOLS_SUBTYPE_LACP:
254 			lacp_pdu_input(lp, m);
255 			return (NULL);
256 
257 		case SLOWPROTOCOLS_SUBTYPE_MARKER:
258 			lacp_marker_input(lp, m);
259 			return (NULL);
260 	}
261 
262 	/* Not a subtype we are interested in */
263 	return (m);
264 }
265 
266 /*
267  * lacp_pdu_input: process lacpdu
268  */
269 static int
270 lacp_pdu_input(struct lacp_port *lp, struct mbuf *m)
271 {
272 	struct lacp_softc *lsc = lp->lp_lsc;
273 	struct lacpdu *du;
274 	int error = 0;
275 
276 	if (m->m_pkthdr.len != sizeof(*du)) {
277 		goto bad;
278 	}
279 
280 	if ((m->m_flags & M_MCAST) == 0) {
281 		goto bad;
282 	}
283 
284 	if (m->m_len < sizeof(*du)) {
285 		m = m_pullup(m, sizeof(*du));
286 		if (m == NULL) {
287 			return (ENOMEM);
288 		}
289 	}
290 
291 	du = mtod(m, struct lacpdu *);
292 
293 	if (memcmp(&du->ldu_eh.ether_dhost,
294 	    &ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
295 		goto bad;
296 	}
297 
298 	/*
299 	 * ignore the version for compatibility with
300 	 * the future protocol revisions.
301 	 */
302 #if 0
303 	if (du->ldu_sph.sph_version != 1) {
304 		goto bad;
305 	}
306 #endif
307 
308 	/*
309 	 * ignore tlv types for compatibility with
310 	 * the future protocol revisions.
311 	 */
312 	if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor,
313 	    lacp_info_tlv_template, FALSE)) {
314 		goto bad;
315 	}
316 
317         if (V_lacp_debug > 0) {
318 		lacp_dprintf(lp, "lacpdu receive\n");
319 		lacp_dump_lacpdu(du);
320 	}
321 
322 	if ((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_rx_test) {
323 		LACP_TPRINTF((lp, "Dropping RX PDU\n"));
324 		goto bad;
325 	}
326 
327 	LACP_LOCK(lsc);
328 	lacp_sm_rx(lp, du);
329 	LACP_UNLOCK(lsc);
330 
331 	m_freem(m);
332 	return (error);
333 
334 bad:
335 	m_freem(m);
336 	return (EINVAL);
337 }
338 
339 static void
340 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info)
341 {
342 	struct lagg_port *lgp = lp->lp_lagg;
343 	struct lagg_softc *sc = lgp->lp_softc;
344 
345 	info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO);
346 	memcpy(&info->lip_systemid.lsi_mac,
347 	    IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
348 	info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO);
349 	info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index);
350 	info->lip_state = lp->lp_state;
351 }
352 
353 static void
354 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info)
355 {
356 	struct ifnet *ifp = lp->lp_ifp;
357 
358 	/* Fill in the port index and system id (encoded as the MAC) */
359 	info->mi_rq_port = htons(ifp->if_index);
360 	memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN);
361 	info->mi_rq_xid = htonl(0);
362 }
363 
364 static int
365 lacp_xmit_lacpdu(struct lacp_port *lp)
366 {
367 	struct lagg_port *lgp = lp->lp_lagg;
368 	struct mbuf *m;
369 	struct lacpdu *du;
370 	int error;
371 
372 	LACP_LOCK_ASSERT(lp->lp_lsc);
373 
374 	m = m_gethdr(M_NOWAIT, MT_DATA);
375 	if (m == NULL) {
376 		return (ENOMEM);
377 	}
378 	m->m_len = m->m_pkthdr.len = sizeof(*du);
379 
380 	du = mtod(m, struct lacpdu *);
381 	memset(du, 0, sizeof(*du));
382 
383 	memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
384 	    ETHER_ADDR_LEN);
385 	memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
386 	du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW);
387 
388 	du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP;
389 	du->ldu_sph.sph_version = 1;
390 
391 	TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor));
392 	du->ldu_actor = lp->lp_actor;
393 
394 	TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO,
395 	    sizeof(du->ldu_partner));
396 	du->ldu_partner = lp->lp_partner;
397 
398 	TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO,
399 	    sizeof(du->ldu_collector));
400 	du->ldu_collector.lci_maxdelay = 0;
401 
402 	if (V_lacp_debug > 0) {
403 		lacp_dprintf(lp, "lacpdu transmit\n");
404 		lacp_dump_lacpdu(du);
405 	}
406 
407 	m->m_flags |= M_MCAST;
408 
409 	/*
410 	 * XXX should use higher priority queue.
411 	 * otherwise network congestion can break aggregation.
412 	 */
413 
414 	error = lagg_enqueue(lp->lp_ifp, m);
415 	return (error);
416 }
417 
418 static int
419 lacp_xmit_marker(struct lacp_port *lp)
420 {
421 	struct lagg_port *lgp = lp->lp_lagg;
422 	struct mbuf *m;
423 	struct markerdu *mdu;
424 	int error;
425 
426 	LACP_LOCK_ASSERT(lp->lp_lsc);
427 
428 	m = m_gethdr(M_NOWAIT, MT_DATA);
429 	if (m == NULL) {
430 		return (ENOMEM);
431 	}
432 	m->m_len = m->m_pkthdr.len = sizeof(*mdu);
433 
434 	mdu = mtod(m, struct markerdu *);
435 	memset(mdu, 0, sizeof(*mdu));
436 
437 	memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
438 	    ETHER_ADDR_LEN);
439 	memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
440 	mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW);
441 
442 	mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER;
443 	mdu->mdu_sph.sph_version = 1;
444 
445 	/* Bump the transaction id and copy over the marker info */
446 	lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1);
447 	TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info));
448 	mdu->mdu_info = lp->lp_marker;
449 
450 	LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n",
451 	    ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":",
452 	    ntohl(mdu->mdu_info.mi_rq_xid)));
453 
454 	m->m_flags |= M_MCAST;
455 	error = lagg_enqueue(lp->lp_ifp, m);
456 	return (error);
457 }
458 
459 void
460 lacp_linkstate(struct lagg_port *lgp)
461 {
462 	struct lacp_port *lp = LACP_PORT(lgp);
463 	struct lacp_softc *lsc = lp->lp_lsc;
464 	struct ifnet *ifp = lgp->lp_ifp;
465 	struct ifmediareq ifmr;
466 	int error = 0;
467 	u_int media;
468 	uint8_t old_state;
469 	uint16_t old_key;
470 
471 	bzero((char *)&ifmr, sizeof(ifmr));
472 	error = (*ifp->if_ioctl)(ifp, SIOCGIFXMEDIA, (caddr_t)&ifmr);
473 	if (error != 0) {
474 		bzero((char *)&ifmr, sizeof(ifmr));
475 		error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr);
476 	}
477 	if (error != 0)
478 		return;
479 
480 	LACP_LOCK(lsc);
481 	media = ifmr.ifm_active;
482 	LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, "
483 	    "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER,
484 	    (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP));
485 	old_state = lp->lp_state;
486 	old_key = lp->lp_key;
487 
488 	lp->lp_media = media;
489 	/*
490 	 * If the port is not an active full duplex Ethernet link then it can
491 	 * not be aggregated.
492 	 */
493 	if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 ||
494 	    ifp->if_link_state != LINK_STATE_UP) {
495 		lacp_port_disable(lp);
496 	} else {
497 		lacp_port_enable(lp);
498 	}
499 	lp->lp_key = lacp_compose_key(lp);
500 
501 	if (old_state != lp->lp_state || old_key != lp->lp_key) {
502 		LACP_DPRINTF((lp, "-> UNSELECTED\n"));
503 		lp->lp_selected = LACP_UNSELECTED;
504 	}
505 	LACP_UNLOCK(lsc);
506 }
507 
508 static void
509 lacp_tick(void *arg)
510 {
511 	struct lacp_softc *lsc = arg;
512 	struct lacp_port *lp;
513 
514 	LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
515 		if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0)
516 			continue;
517 
518 		CURVNET_SET(lp->lp_ifp->if_vnet);
519 		lacp_run_timers(lp);
520 
521 		lacp_select(lp);
522 		lacp_sm_mux(lp);
523 		lacp_sm_tx(lp);
524 		lacp_sm_ptx_tx_schedule(lp);
525 		CURVNET_RESTORE();
526 	}
527 	callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
528 }
529 
530 int
531 lacp_port_create(struct lagg_port *lgp)
532 {
533 	struct lagg_softc *sc = lgp->lp_softc;
534 	struct lacp_softc *lsc = LACP_SOFTC(sc);
535 	struct lacp_port *lp;
536 	struct ifnet *ifp = lgp->lp_ifp;
537 	struct sockaddr_dl sdl;
538 	struct ifmultiaddr *rifma = NULL;
539 	int error;
540 
541 	link_init_sdl(ifp, (struct sockaddr *)&sdl, IFT_ETHER);
542 	sdl.sdl_alen = ETHER_ADDR_LEN;
543 
544 	bcopy(&ethermulticastaddr_slowprotocols,
545 	    LLADDR(&sdl), ETHER_ADDR_LEN);
546 	error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
547 	if (error) {
548 		printf("%s: ADDMULTI failed on %s\n", __func__,
549 		    lgp->lp_ifp->if_xname);
550 		return (error);
551 	}
552 
553 	lp = malloc(sizeof(struct lacp_port),
554 	    M_DEVBUF, M_NOWAIT|M_ZERO);
555 	if (lp == NULL)
556 		return (ENOMEM);
557 
558 	LACP_LOCK(lsc);
559 	lgp->lp_psc = lp;
560 	lp->lp_ifp = ifp;
561 	lp->lp_lagg = lgp;
562 	lp->lp_lsc = lsc;
563 	lp->lp_ifma = rifma;
564 
565 	LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next);
566 
567 	lacp_fill_actorinfo(lp, &lp->lp_actor);
568 	lacp_fill_markerinfo(lp, &lp->lp_marker);
569 	lp->lp_state = LACP_STATE_ACTIVITY;
570 	lp->lp_aggregator = NULL;
571 	lacp_sm_rx_set_expired(lp);
572 	LACP_UNLOCK(lsc);
573 	lacp_linkstate(lgp);
574 
575 	return (0);
576 }
577 
578 void
579 lacp_port_destroy(struct lagg_port *lgp)
580 {
581 	struct lacp_port *lp = LACP_PORT(lgp);
582 	struct lacp_softc *lsc = lp->lp_lsc;
583 	int i;
584 
585 	LACP_LOCK(lsc);
586 	for (i = 0; i < LACP_NTIMER; i++) {
587 		LACP_TIMER_DISARM(lp, i);
588 	}
589 
590 	lacp_disable_collecting(lp);
591 	lacp_disable_distributing(lp);
592 	lacp_unselect(lp);
593 
594 	LIST_REMOVE(lp, lp_next);
595 	LACP_UNLOCK(lsc);
596 
597 	/* The address may have already been removed by if_purgemaddrs() */
598 	if (!lgp->lp_detaching)
599 		if_delmulti_ifma(lp->lp_ifma);
600 
601 	free(lp, M_DEVBUF);
602 }
603 
604 void
605 lacp_req(struct lagg_softc *sc, void *data)
606 {
607 	struct lacp_opreq *req = (struct lacp_opreq *)data;
608 	struct lacp_softc *lsc = LACP_SOFTC(sc);
609 	struct lacp_aggregator *la;
610 
611 	bzero(req, sizeof(struct lacp_opreq));
612 
613 	/*
614 	 * If the LACP softc is NULL, return with the opreq structure full of
615 	 * zeros.  It is normal for the softc to be NULL while the lagg is
616 	 * being destroyed.
617 	 */
618 	if (NULL == lsc)
619 		return;
620 
621 	la = lsc->lsc_active_aggregator;
622 	LACP_LOCK(lsc);
623 	if (la != NULL) {
624 		req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio);
625 		memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac,
626 		    ETHER_ADDR_LEN);
627 		req->actor_key = ntohs(la->la_actor.lip_key);
628 		req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio);
629 		req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno);
630 		req->actor_state = la->la_actor.lip_state;
631 
632 		req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio);
633 		memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac,
634 		    ETHER_ADDR_LEN);
635 		req->partner_key = ntohs(la->la_partner.lip_key);
636 		req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio);
637 		req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno);
638 		req->partner_state = la->la_partner.lip_state;
639 	}
640 	LACP_UNLOCK(lsc);
641 }
642 
643 void
644 lacp_portreq(struct lagg_port *lgp, void *data)
645 {
646 	struct lacp_opreq *req = (struct lacp_opreq *)data;
647 	struct lacp_port *lp = LACP_PORT(lgp);
648 	struct lacp_softc *lsc = lp->lp_lsc;
649 
650 	LACP_LOCK(lsc);
651 	req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio);
652 	memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac,
653 	    ETHER_ADDR_LEN);
654 	req->actor_key = ntohs(lp->lp_actor.lip_key);
655 	req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio);
656 	req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno);
657 	req->actor_state = lp->lp_actor.lip_state;
658 
659 	req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio);
660 	memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac,
661 	    ETHER_ADDR_LEN);
662 	req->partner_key = ntohs(lp->lp_partner.lip_key);
663 	req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio);
664 	req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno);
665 	req->partner_state = lp->lp_partner.lip_state;
666 	LACP_UNLOCK(lsc);
667 }
668 
669 static void
670 lacp_disable_collecting(struct lacp_port *lp)
671 {
672 	LACP_DPRINTF((lp, "collecting disabled\n"));
673 	lp->lp_state &= ~LACP_STATE_COLLECTING;
674 }
675 
676 static void
677 lacp_enable_collecting(struct lacp_port *lp)
678 {
679 	LACP_DPRINTF((lp, "collecting enabled\n"));
680 	lp->lp_state |= LACP_STATE_COLLECTING;
681 }
682 
683 static void
684 lacp_disable_distributing(struct lacp_port *lp)
685 {
686 	struct lacp_aggregator *la = lp->lp_aggregator;
687 	struct lacp_softc *lsc = lp->lp_lsc;
688 	struct lagg_softc *sc = lsc->lsc_softc;
689 	char buf[LACP_LAGIDSTR_MAX+1];
690 
691 	LACP_LOCK_ASSERT(lsc);
692 
693 	if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) {
694 		return;
695 	}
696 
697 	KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports"));
698 	KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports));
699 	KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid"));
700 
701 	LACP_DPRINTF((lp, "disable distributing on aggregator %s, "
702 	    "nports %d -> %d\n",
703 	    lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
704 	    la->la_nports, la->la_nports - 1));
705 
706 	TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q);
707 	la->la_nports--;
708 	sc->sc_active = la->la_nports;
709 
710 	if (lsc->lsc_active_aggregator == la) {
711 		lacp_suppress_distributing(lsc, la);
712 		lacp_select_active_aggregator(lsc);
713 		/* regenerate the port map, the active aggregator has changed */
714 		lacp_update_portmap(lsc);
715 	}
716 
717 	lp->lp_state &= ~LACP_STATE_DISTRIBUTING;
718 	if_link_state_change(sc->sc_ifp,
719 	    sc->sc_active ? LINK_STATE_UP : LINK_STATE_DOWN);
720 }
721 
722 static void
723 lacp_enable_distributing(struct lacp_port *lp)
724 {
725 	struct lacp_aggregator *la = lp->lp_aggregator;
726 	struct lacp_softc *lsc = lp->lp_lsc;
727 	struct lagg_softc *sc = lsc->lsc_softc;
728 	char buf[LACP_LAGIDSTR_MAX+1];
729 
730 	LACP_LOCK_ASSERT(lsc);
731 
732 	if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) {
733 		return;
734 	}
735 
736 	LACP_DPRINTF((lp, "enable distributing on aggregator %s, "
737 	    "nports %d -> %d\n",
738 	    lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
739 	    la->la_nports, la->la_nports + 1));
740 
741 	KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid"));
742 	TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q);
743 	la->la_nports++;
744 	sc->sc_active = la->la_nports;
745 
746 	lp->lp_state |= LACP_STATE_DISTRIBUTING;
747 
748 	if (lsc->lsc_active_aggregator == la) {
749 		lacp_suppress_distributing(lsc, la);
750 		lacp_update_portmap(lsc);
751 	} else
752 		/* try to become the active aggregator */
753 		lacp_select_active_aggregator(lsc);
754 
755 	if_link_state_change(sc->sc_ifp,
756 	    sc->sc_active ? LINK_STATE_UP : LINK_STATE_DOWN);
757 }
758 
759 static void
760 lacp_transit_expire(void *vp)
761 {
762 	struct lacp_softc *lsc = vp;
763 
764 	LACP_LOCK_ASSERT(lsc);
765 
766 	CURVNET_SET(lsc->lsc_softc->sc_ifp->if_vnet);
767 	LACP_TRACE(NULL);
768 	CURVNET_RESTORE();
769 
770 	lsc->lsc_suppress_distributing = FALSE;
771 }
772 
773 void
774 lacp_attach(struct lagg_softc *sc)
775 {
776 	struct lacp_softc *lsc;
777 
778 	lsc = malloc(sizeof(struct lacp_softc), M_DEVBUF, M_WAITOK | M_ZERO);
779 
780 	sc->sc_psc = lsc;
781 	lsc->lsc_softc = sc;
782 
783 	lsc->lsc_hashkey = m_ether_tcpip_hash_init();
784 	lsc->lsc_active_aggregator = NULL;
785 	lsc->lsc_strict_mode = VNET(lacp_default_strict_mode);
786 	LACP_LOCK_INIT(lsc);
787 	TAILQ_INIT(&lsc->lsc_aggregators);
788 	LIST_INIT(&lsc->lsc_ports);
789 
790 	callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0);
791 	callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0);
792 
793 	/* if the lagg is already up then do the same */
794 	if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING)
795 		lacp_init(sc);
796 }
797 
798 void
799 lacp_detach(void *psc)
800 {
801 	struct lacp_softc *lsc = (struct lacp_softc *)psc;
802 
803 	KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators),
804 	    ("aggregators still active"));
805 	KASSERT(lsc->lsc_active_aggregator == NULL,
806 	    ("aggregator still attached"));
807 
808 	callout_drain(&lsc->lsc_transit_callout);
809 	callout_drain(&lsc->lsc_callout);
810 
811 	LACP_LOCK_DESTROY(lsc);
812 	free(lsc, M_DEVBUF);
813 }
814 
815 void
816 lacp_init(struct lagg_softc *sc)
817 {
818 	struct lacp_softc *lsc = LACP_SOFTC(sc);
819 
820 	LACP_LOCK(lsc);
821 	callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
822 	LACP_UNLOCK(lsc);
823 }
824 
825 void
826 lacp_stop(struct lagg_softc *sc)
827 {
828 	struct lacp_softc *lsc = LACP_SOFTC(sc);
829 
830 	LACP_LOCK(lsc);
831 	callout_stop(&lsc->lsc_transit_callout);
832 	callout_stop(&lsc->lsc_callout);
833 	LACP_UNLOCK(lsc);
834 }
835 
836 struct lagg_port *
837 lacp_select_tx_port_by_hash(struct lagg_softc *sc, uint32_t hash,
838     uint8_t numa_domain, int *err)
839 {
840 	struct lacp_softc *lsc = LACP_SOFTC(sc);
841 	struct lacp_portmap *pm;
842 	struct lacp_port *lp;
843 	struct lacp_port **map;
844 	int count;
845 
846 	if (__predict_false(lsc->lsc_suppress_distributing)) {
847 		LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__));
848 		*err = ENOBUFS;
849 		return (NULL);
850 	}
851 
852 	pm = &lsc->lsc_pmap[lsc->lsc_activemap];
853 	if (pm->pm_count == 0) {
854 		LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__));
855 		*err = ENETDOWN;
856 		return (NULL);
857 	}
858 
859 #ifdef NUMA
860 	if ((sc->sc_opts & LAGG_OPT_USE_NUMA) &&
861 	    pm->pm_num_dom > 1 && numa_domain < MAXMEMDOM) {
862 		count = pm->pm_numa[numa_domain].count;
863 		if (count > 0) {
864 			map = pm->pm_numa[numa_domain].map;
865 		} else {
866 			/* No ports on this domain; use global hash. */
867 			map = pm->pm_map;
868 			count = pm->pm_count;
869 		}
870 	} else
871 #endif
872 	{
873 		map = pm->pm_map;
874 		count = pm->pm_count;
875 	}
876 
877 	hash %= count;
878 	lp = map[hash];
879 
880 	return (lp->lp_lagg);
881 }
882 
883 struct lagg_port *
884 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m, int *err)
885 {
886 	struct lacp_softc *lsc = LACP_SOFTC(sc);
887 	uint32_t hash;
888 	uint8_t numa_domain;
889 
890 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
891 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
892 		hash = m->m_pkthdr.flowid >> sc->flowid_shift;
893 	else
894 		hash = m_ether_tcpip_hash(sc->sc_flags, m, lsc->lsc_hashkey);
895 
896 	numa_domain = m->m_pkthdr.numa_domain;
897 	return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, err));
898 }
899 
900 /*
901  * lacp_suppress_distributing: drop transmit packets for a while
902  * to preserve packet ordering.
903  */
904 
905 static void
906 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la)
907 {
908 	struct lacp_port *lp;
909 
910 	if (lsc->lsc_active_aggregator != la) {
911 		return;
912 	}
913 
914 	LACP_TRACE(NULL);
915 
916 	lsc->lsc_suppress_distributing = TRUE;
917 
918 	/* send a marker frame down each port to verify the queues are empty */
919 	LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
920 		lp->lp_flags |= LACP_PORT_MARK;
921 		if (lacp_xmit_marker(lp) != 0)
922 			lp->lp_flags &= ~LACP_PORT_MARK;
923 	}
924 
925 	/* set a timeout for the marker frames */
926 	callout_reset(&lsc->lsc_transit_callout,
927 	    LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc);
928 }
929 
930 static int
931 lacp_compare_peerinfo(const struct lacp_peerinfo *a,
932     const struct lacp_peerinfo *b)
933 {
934 	return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state)));
935 }
936 
937 static int
938 lacp_compare_systemid(const struct lacp_systemid *a,
939     const struct lacp_systemid *b)
940 {
941 	return (memcmp(a, b, sizeof(*a)));
942 }
943 
944 #if 0	/* unused */
945 static int
946 lacp_compare_portid(const struct lacp_portid *a,
947     const struct lacp_portid *b)
948 {
949 	return (memcmp(a, b, sizeof(*a)));
950 }
951 #endif
952 
953 static uint64_t
954 lacp_aggregator_bandwidth(struct lacp_aggregator *la)
955 {
956 	struct lacp_port *lp;
957 	uint64_t speed;
958 
959 	lp = TAILQ_FIRST(&la->la_ports);
960 	if (lp == NULL) {
961 		return (0);
962 	}
963 
964 	speed = ifmedia_baudrate(lp->lp_media);
965 	speed *= la->la_nports;
966 	if (speed == 0) {
967 		LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n",
968 		    lp->lp_media, la->la_nports));
969 	}
970 
971 	return (speed);
972 }
973 
974 /*
975  * lacp_select_active_aggregator: select an aggregator to be used to transmit
976  * packets from lagg(4) interface.
977  */
978 
979 static void
980 lacp_select_active_aggregator(struct lacp_softc *lsc)
981 {
982 	struct lacp_aggregator *la;
983 	struct lacp_aggregator *best_la = NULL;
984 	uint64_t best_speed = 0;
985 	char buf[LACP_LAGIDSTR_MAX+1];
986 
987 	LACP_TRACE(NULL);
988 
989 	TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
990 		uint64_t speed;
991 
992 		if (la->la_nports == 0) {
993 			continue;
994 		}
995 
996 		speed = lacp_aggregator_bandwidth(la);
997 		LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n",
998 		    lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
999 		    speed, la->la_nports));
1000 
1001 		/*
1002 		 * This aggregator is chosen if the partner has a better
1003 		 * system priority or, the total aggregated speed is higher
1004 		 * or, it is already the chosen aggregator
1005 		 */
1006 		if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) <
1007 		    LACP_SYS_PRI(best_la->la_partner)) ||
1008 		    speed > best_speed ||
1009 		    (speed == best_speed &&
1010 		    la == lsc->lsc_active_aggregator)) {
1011 			best_la = la;
1012 			best_speed = speed;
1013 		}
1014 	}
1015 
1016 	KASSERT(best_la == NULL || best_la->la_nports > 0,
1017 	    ("invalid aggregator refcnt"));
1018 	KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports),
1019 	    ("invalid aggregator list"));
1020 
1021 	if (lsc->lsc_active_aggregator != best_la) {
1022 		LACP_DPRINTF((NULL, "active aggregator changed\n"));
1023 		LACP_DPRINTF((NULL, "old %s\n",
1024 		    lacp_format_lagid_aggregator(lsc->lsc_active_aggregator,
1025 		    buf, sizeof(buf))));
1026 	} else {
1027 		LACP_DPRINTF((NULL, "active aggregator not changed\n"));
1028 	}
1029 	LACP_DPRINTF((NULL, "new %s\n",
1030 	    lacp_format_lagid_aggregator(best_la, buf, sizeof(buf))));
1031 
1032 	if (lsc->lsc_active_aggregator != best_la) {
1033 		lsc->lsc_active_aggregator = best_la;
1034 		lacp_update_portmap(lsc);
1035 		if (best_la) {
1036 			lacp_suppress_distributing(lsc, best_la);
1037 		}
1038 	}
1039 }
1040 
1041 /*
1042  * Updated the inactive portmap array with the new list of ports and
1043  * make it live.
1044  */
1045 static void
1046 lacp_update_portmap(struct lacp_softc *lsc)
1047 {
1048 	struct lagg_softc *sc = lsc->lsc_softc;
1049 	struct lacp_aggregator *la;
1050 	struct lacp_portmap *p;
1051 	struct lacp_port *lp;
1052 	uint64_t speed;
1053 	u_int newmap;
1054 	int i;
1055 #ifdef NUMA
1056 	int count;
1057 	uint8_t domain;
1058 #endif
1059 
1060 	newmap = lsc->lsc_activemap == 0 ? 1 : 0;
1061 	p = &lsc->lsc_pmap[newmap];
1062 	la = lsc->lsc_active_aggregator;
1063 	speed = 0;
1064 	bzero(p, sizeof(struct lacp_portmap));
1065 
1066 	if (la != NULL && la->la_nports > 0) {
1067 		p->pm_count = la->la_nports;
1068 		i = 0;
1069 		TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q) {
1070 			p->pm_map[i++] = lp;
1071 #ifdef NUMA
1072 			domain = lp->lp_ifp->if_numa_domain;
1073 			if (domain >= MAXMEMDOM)
1074 				continue;
1075 			count = p->pm_numa[domain].count;
1076 			p->pm_numa[domain].map[count] = lp;
1077 			p->pm_numa[domain].count++;
1078 #endif
1079 		}
1080 		KASSERT(i == p->pm_count, ("Invalid port count"));
1081 
1082 #ifdef NUMA
1083 		for (i = 0; i < MAXMEMDOM; i++) {
1084 			if (p->pm_numa[i].count != 0)
1085 				p->pm_num_dom++;
1086 		}
1087 #endif
1088 		speed = lacp_aggregator_bandwidth(la);
1089 	}
1090 	sc->sc_ifp->if_baudrate = speed;
1091 	EVENTHANDLER_INVOKE(ifnet_event, sc->sc_ifp,
1092 	    IFNET_EVENT_UPDATE_BAUDRATE);
1093 
1094 	/* switch the active portmap over */
1095 	atomic_store_rel_int(&lsc->lsc_activemap, newmap);
1096 	LACP_DPRINTF((NULL, "Set table %d with %d ports\n",
1097 		    lsc->lsc_activemap,
1098 		    lsc->lsc_pmap[lsc->lsc_activemap].pm_count));
1099 }
1100 
1101 static uint16_t
1102 lacp_compose_key(struct lacp_port *lp)
1103 {
1104 	struct lagg_port *lgp = lp->lp_lagg;
1105 	struct lagg_softc *sc = lgp->lp_softc;
1106 	u_int media = lp->lp_media;
1107 	uint16_t key;
1108 
1109 	if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) {
1110 		/*
1111 		 * non-aggregatable links should have unique keys.
1112 		 *
1113 		 * XXX this isn't really unique as if_index is 16 bit.
1114 		 */
1115 
1116 		/* bit 0..14:	(some bits of) if_index of this port */
1117 		key = lp->lp_ifp->if_index;
1118 		/* bit 15:	1 */
1119 		key |= 0x8000;
1120 	} else {
1121 		u_int subtype = IFM_SUBTYPE(media);
1122 
1123 		KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type"));
1124 		KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface"));
1125 
1126 		/* bit 0..4:	IFM_SUBTYPE modulo speed */
1127 		switch (subtype) {
1128 		case IFM_10_T:
1129 		case IFM_10_2:
1130 		case IFM_10_5:
1131 		case IFM_10_STP:
1132 		case IFM_10_FL:
1133 			key = IFM_10_T;
1134 			break;
1135 		case IFM_100_TX:
1136 		case IFM_100_FX:
1137 		case IFM_100_T4:
1138 		case IFM_100_VG:
1139 		case IFM_100_T2:
1140 		case IFM_100_T:
1141 		case IFM_100_SGMII:
1142 			key = IFM_100_TX;
1143 			break;
1144 		case IFM_1000_SX:
1145 		case IFM_1000_LX:
1146 		case IFM_1000_CX:
1147 		case IFM_1000_T:
1148 		case IFM_1000_KX:
1149 		case IFM_1000_SGMII:
1150 		case IFM_1000_CX_SGMII:
1151 			key = IFM_1000_SX;
1152 			break;
1153 		case IFM_10G_LR:
1154 		case IFM_10G_SR:
1155 		case IFM_10G_CX4:
1156 		case IFM_10G_TWINAX:
1157 		case IFM_10G_TWINAX_LONG:
1158 		case IFM_10G_LRM:
1159 		case IFM_10G_T:
1160 		case IFM_10G_KX4:
1161 		case IFM_10G_KR:
1162 		case IFM_10G_CR1:
1163 		case IFM_10G_ER:
1164 		case IFM_10G_SFI:
1165 		case IFM_10G_AOC:
1166 			key = IFM_10G_LR;
1167 			break;
1168 		case IFM_20G_KR2:
1169 			key = IFM_20G_KR2;
1170 			break;
1171 		case IFM_2500_KX:
1172 		case IFM_2500_T:
1173 		case IFM_2500_X:
1174 			key = IFM_2500_KX;
1175 			break;
1176 		case IFM_5000_T:
1177 		case IFM_5000_KR:
1178 		case IFM_5000_KR_S:
1179 		case IFM_5000_KR1:
1180 			key = IFM_5000_T;
1181 			break;
1182 		case IFM_50G_PCIE:
1183 		case IFM_50G_CR2:
1184 		case IFM_50G_KR2:
1185 		case IFM_50G_KR4:
1186 		case IFM_50G_SR2:
1187 		case IFM_50G_LR2:
1188 		case IFM_50G_LAUI2_AC:
1189 		case IFM_50G_LAUI2:
1190 		case IFM_50G_AUI2_AC:
1191 		case IFM_50G_AUI2:
1192 		case IFM_50G_CP:
1193 		case IFM_50G_SR:
1194 		case IFM_50G_LR:
1195 		case IFM_50G_FR:
1196 		case IFM_50G_KR_PAM4:
1197 		case IFM_50G_AUI1_AC:
1198 		case IFM_50G_AUI1:
1199 			key = IFM_50G_PCIE;
1200 			break;
1201 		case IFM_56G_R4:
1202 			key = IFM_56G_R4;
1203 			break;
1204 		case IFM_25G_PCIE:
1205 		case IFM_25G_CR:
1206 		case IFM_25G_KR:
1207 		case IFM_25G_SR:
1208 		case IFM_25G_LR:
1209 		case IFM_25G_ACC:
1210 		case IFM_25G_AOC:
1211 		case IFM_25G_T:
1212 		case IFM_25G_CR_S:
1213 		case IFM_25G_CR1:
1214 		case IFM_25G_KR_S:
1215 		case IFM_25G_AUI:
1216 		case IFM_25G_KR1:
1217 			key = IFM_25G_PCIE;
1218 			break;
1219 		case IFM_40G_CR4:
1220 		case IFM_40G_SR4:
1221 		case IFM_40G_LR4:
1222 		case IFM_40G_LM4:
1223 		case IFM_40G_XLPPI:
1224 		case IFM_40G_KR4:
1225 		case IFM_40G_XLAUI:
1226 		case IFM_40G_XLAUI_AC:
1227 		case IFM_40G_ER4:
1228 			key = IFM_40G_CR4;
1229 			break;
1230 		case IFM_100G_CR4:
1231 		case IFM_100G_SR4:
1232 		case IFM_100G_KR4:
1233 		case IFM_100G_LR4:
1234 		case IFM_100G_CAUI4_AC:
1235 		case IFM_100G_CAUI4:
1236 		case IFM_100G_AUI4_AC:
1237 		case IFM_100G_AUI4:
1238 		case IFM_100G_CR_PAM4:
1239 		case IFM_100G_KR_PAM4:
1240 		case IFM_100G_CP2:
1241 		case IFM_100G_SR2:
1242 		case IFM_100G_DR:
1243 		case IFM_100G_KR2_PAM4:
1244 		case IFM_100G_CAUI2_AC:
1245 		case IFM_100G_CAUI2:
1246 		case IFM_100G_AUI2_AC:
1247 		case IFM_100G_AUI2:
1248 			key = IFM_100G_CR4;
1249 			break;
1250 		case IFM_200G_CR4_PAM4:
1251 		case IFM_200G_SR4:
1252 		case IFM_200G_FR4:
1253 		case IFM_200G_LR4:
1254 		case IFM_200G_DR4:
1255 		case IFM_200G_KR4_PAM4:
1256 		case IFM_200G_AUI4_AC:
1257 		case IFM_200G_AUI4:
1258 		case IFM_200G_AUI8_AC:
1259 		case IFM_200G_AUI8:
1260 			key = IFM_200G_CR4_PAM4;
1261 			break;
1262 		case IFM_400G_FR8:
1263 		case IFM_400G_LR8:
1264 		case IFM_400G_DR4:
1265 		case IFM_400G_AUI8_AC:
1266 		case IFM_400G_AUI8:
1267 			key = IFM_400G_FR8;
1268 			break;
1269 		default:
1270 			key = subtype;
1271 			break;
1272 		}
1273 		/* bit 5..14:	(some bits of) if_index of lagg device */
1274 		key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5);
1275 		/* bit 15:	0 */
1276 	}
1277 	return (htons(key));
1278 }
1279 
1280 static void
1281 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1282 {
1283 	char buf[LACP_LAGIDSTR_MAX+1];
1284 
1285 	LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1286 	    __func__,
1287 	    lacp_format_lagid(&la->la_actor, &la->la_partner,
1288 	    buf, sizeof(buf)),
1289 	    la->la_refcnt, la->la_refcnt + 1));
1290 
1291 	KASSERT(la->la_refcnt > 0, ("refcount <= 0"));
1292 	la->la_refcnt++;
1293 	KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount"));
1294 }
1295 
1296 static void
1297 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1298 {
1299 	char buf[LACP_LAGIDSTR_MAX+1];
1300 
1301 	LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1302 	    __func__,
1303 	    lacp_format_lagid(&la->la_actor, &la->la_partner,
1304 	    buf, sizeof(buf)),
1305 	    la->la_refcnt, la->la_refcnt - 1));
1306 
1307 	KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt"));
1308 	la->la_refcnt--;
1309 	if (la->la_refcnt > 0) {
1310 		return;
1311 	}
1312 
1313 	KASSERT(la->la_refcnt == 0, ("refcount not zero"));
1314 	KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active"));
1315 
1316 	TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q);
1317 
1318 	free(la, M_DEVBUF);
1319 }
1320 
1321 /*
1322  * lacp_aggregator_get: allocate an aggregator.
1323  */
1324 
1325 static struct lacp_aggregator *
1326 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp)
1327 {
1328 	struct lacp_aggregator *la;
1329 
1330 	la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT);
1331 	if (la) {
1332 		la->la_refcnt = 1;
1333 		la->la_nports = 0;
1334 		TAILQ_INIT(&la->la_ports);
1335 		la->la_pending = 0;
1336 		TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q);
1337 	}
1338 
1339 	return (la);
1340 }
1341 
1342 /*
1343  * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port.
1344  */
1345 
1346 static void
1347 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp)
1348 {
1349 	lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner);
1350 	lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor);
1351 
1352 	la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION;
1353 }
1354 
1355 static void
1356 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr,
1357     const struct lacp_peerinfo *lpi_port)
1358 {
1359 	memset(lpi_aggr, 0, sizeof(*lpi_aggr));
1360 	lpi_aggr->lip_systemid = lpi_port->lip_systemid;
1361 	lpi_aggr->lip_key = lpi_port->lip_key;
1362 }
1363 
1364 /*
1365  * lacp_aggregator_is_compatible: check if a port can join to an aggregator.
1366  */
1367 
1368 static bool
1369 lacp_aggregator_is_compatible(const struct lacp_aggregator *la,
1370     const struct lacp_port *lp)
1371 {
1372 	if (!(lp->lp_state & LACP_STATE_AGGREGATION) ||
1373 	    !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) {
1374 		return (false);
1375 	}
1376 
1377 	if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION))
1378 		return (false);
1379 
1380 	if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner))
1381 		return (false);
1382 
1383 	if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor))
1384 		return (false);
1385 
1386 	return (true);
1387 }
1388 
1389 static bool
1390 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a,
1391     const struct lacp_peerinfo *b)
1392 {
1393 	if (memcmp(&a->lip_systemid, &b->lip_systemid,
1394 	    sizeof(a->lip_systemid)) != 0) {
1395 		return (false);
1396 	}
1397 
1398 	if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key)) != 0)
1399 		return (false);
1400 
1401 	return (true);
1402 }
1403 
1404 static void
1405 lacp_port_enable(struct lacp_port *lp)
1406 {
1407 	lp->lp_state |= LACP_STATE_AGGREGATION;
1408 }
1409 
1410 static void
1411 lacp_port_disable(struct lacp_port *lp)
1412 {
1413 	lacp_set_mux(lp, LACP_MUX_DETACHED);
1414 
1415 	lp->lp_state &= ~LACP_STATE_AGGREGATION;
1416 	lp->lp_selected = LACP_UNSELECTED;
1417 	lacp_sm_rx_record_default(lp);
1418 	lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION;
1419 	lp->lp_state &= ~LACP_STATE_EXPIRED;
1420 }
1421 
1422 /*
1423  * lacp_select: select an aggregator.  create one if necessary.
1424  */
1425 static void
1426 lacp_select(struct lacp_port *lp)
1427 {
1428 	struct lacp_softc *lsc = lp->lp_lsc;
1429 	struct lacp_aggregator *la;
1430 	char buf[LACP_LAGIDSTR_MAX+1];
1431 
1432 	if (lp->lp_aggregator) {
1433 		return;
1434 	}
1435 
1436 	/* If we haven't heard from our peer, skip this step. */
1437 	if (lp->lp_state & LACP_STATE_DEFAULTED)
1438 		return;
1439 
1440 	KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1441 	    ("timer_wait_while still active"));
1442 
1443 	LACP_DPRINTF((lp, "port lagid=%s\n",
1444 	    lacp_format_lagid(&lp->lp_actor, &lp->lp_partner,
1445 	    buf, sizeof(buf))));
1446 
1447 	TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
1448 		if (lacp_aggregator_is_compatible(la, lp)) {
1449 			break;
1450 		}
1451 	}
1452 
1453 	if (la == NULL) {
1454 		la = lacp_aggregator_get(lsc, lp);
1455 		if (la == NULL) {
1456 			LACP_DPRINTF((lp, "aggregator creation failed\n"));
1457 
1458 			/*
1459 			 * will retry on the next tick.
1460 			 */
1461 
1462 			return;
1463 		}
1464 		lacp_fill_aggregator_id(la, lp);
1465 		LACP_DPRINTF((lp, "aggregator created\n"));
1466 	} else {
1467 		LACP_DPRINTF((lp, "compatible aggregator found\n"));
1468 		if (la->la_refcnt == LACP_MAX_PORTS)
1469 			return;
1470 		lacp_aggregator_addref(lsc, la);
1471 	}
1472 
1473 	LACP_DPRINTF((lp, "aggregator lagid=%s\n",
1474 	    lacp_format_lagid(&la->la_actor, &la->la_partner,
1475 	    buf, sizeof(buf))));
1476 
1477 	lp->lp_aggregator = la;
1478 	lp->lp_selected = LACP_SELECTED;
1479 }
1480 
1481 /*
1482  * lacp_unselect: finish unselect/detach process.
1483  */
1484 
1485 static void
1486 lacp_unselect(struct lacp_port *lp)
1487 {
1488 	struct lacp_softc *lsc = lp->lp_lsc;
1489 	struct lacp_aggregator *la = lp->lp_aggregator;
1490 
1491 	KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1492 	    ("timer_wait_while still active"));
1493 
1494 	if (la == NULL) {
1495 		return;
1496 	}
1497 
1498 	lp->lp_aggregator = NULL;
1499 	lacp_aggregator_delref(lsc, la);
1500 }
1501 
1502 /* mux machine */
1503 
1504 static void
1505 lacp_sm_mux(struct lacp_port *lp)
1506 {
1507 	struct lagg_port *lgp = lp->lp_lagg;
1508 	struct lagg_softc *sc = lgp->lp_softc;
1509 	enum lacp_mux_state new_state;
1510 	boolean_t p_sync =
1511 		    (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0;
1512 	boolean_t p_collecting =
1513 	    (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0;
1514 	enum lacp_selected selected = lp->lp_selected;
1515 	struct lacp_aggregator *la;
1516 
1517 	if (V_lacp_debug > 1)
1518 		lacp_dprintf(lp, "%s: state= 0x%x, selected= 0x%x, "
1519 		    "p_sync= 0x%x, p_collecting= 0x%x\n", __func__,
1520 		    lp->lp_mux_state, selected, p_sync, p_collecting);
1521 
1522 re_eval:
1523 	la = lp->lp_aggregator;
1524 	KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL,
1525 	    ("MUX not detached"));
1526 	new_state = lp->lp_mux_state;
1527 	switch (lp->lp_mux_state) {
1528 	case LACP_MUX_DETACHED:
1529 		if (selected != LACP_UNSELECTED) {
1530 			new_state = LACP_MUX_WAITING;
1531 		}
1532 		break;
1533 	case LACP_MUX_WAITING:
1534 		KASSERT(la->la_pending > 0 ||
1535 		    !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1536 		    ("timer_wait_while still active"));
1537 		if (selected == LACP_SELECTED && la->la_pending == 0) {
1538 			new_state = LACP_MUX_ATTACHED;
1539 		} else if (selected == LACP_UNSELECTED) {
1540 			new_state = LACP_MUX_DETACHED;
1541 		}
1542 		break;
1543 	case LACP_MUX_ATTACHED:
1544 		if (selected == LACP_SELECTED && p_sync) {
1545 			new_state = LACP_MUX_COLLECTING;
1546 		} else if (selected != LACP_SELECTED) {
1547 			new_state = LACP_MUX_DETACHED;
1548 		}
1549 		break;
1550 	case LACP_MUX_COLLECTING:
1551 		if (selected == LACP_SELECTED && p_sync && p_collecting) {
1552 			new_state = LACP_MUX_DISTRIBUTING;
1553 		} else if (selected != LACP_SELECTED || !p_sync) {
1554 			new_state = LACP_MUX_ATTACHED;
1555 		}
1556 		break;
1557 	case LACP_MUX_DISTRIBUTING:
1558 		if (selected != LACP_SELECTED || !p_sync || !p_collecting) {
1559 			new_state = LACP_MUX_COLLECTING;
1560 			lacp_dprintf(lp, "Interface stopped DISTRIBUTING, possible flapping\n");
1561 			sc->sc_flapping++;
1562 		}
1563 		break;
1564 	default:
1565 		panic("%s: unknown state", __func__);
1566 	}
1567 
1568 	if (lp->lp_mux_state == new_state) {
1569 		return;
1570 	}
1571 
1572 	lacp_set_mux(lp, new_state);
1573 	goto re_eval;
1574 }
1575 
1576 static void
1577 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state)
1578 {
1579 	struct lacp_aggregator *la = lp->lp_aggregator;
1580 
1581 	if (lp->lp_mux_state == new_state) {
1582 		return;
1583 	}
1584 
1585 	switch (new_state) {
1586 	case LACP_MUX_DETACHED:
1587 		lp->lp_state &= ~LACP_STATE_SYNC;
1588 		lacp_disable_distributing(lp);
1589 		lacp_disable_collecting(lp);
1590 		lacp_sm_assert_ntt(lp);
1591 		/* cancel timer */
1592 		if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) {
1593 			KASSERT(la->la_pending > 0,
1594 			    ("timer_wait_while not active"));
1595 			la->la_pending--;
1596 		}
1597 		LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE);
1598 		lacp_unselect(lp);
1599 		break;
1600 	case LACP_MUX_WAITING:
1601 		LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE,
1602 		    LACP_AGGREGATE_WAIT_TIME);
1603 		la->la_pending++;
1604 		break;
1605 	case LACP_MUX_ATTACHED:
1606 		lp->lp_state |= LACP_STATE_SYNC;
1607 		lacp_disable_collecting(lp);
1608 		lacp_sm_assert_ntt(lp);
1609 		break;
1610 	case LACP_MUX_COLLECTING:
1611 		lacp_enable_collecting(lp);
1612 		lacp_disable_distributing(lp);
1613 		lacp_sm_assert_ntt(lp);
1614 		break;
1615 	case LACP_MUX_DISTRIBUTING:
1616 		lacp_enable_distributing(lp);
1617 		break;
1618 	default:
1619 		panic("%s: unknown state", __func__);
1620 	}
1621 
1622 	LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state));
1623 
1624 	lp->lp_mux_state = new_state;
1625 }
1626 
1627 static void
1628 lacp_sm_mux_timer(struct lacp_port *lp)
1629 {
1630 	struct lacp_aggregator *la = lp->lp_aggregator;
1631 	char buf[LACP_LAGIDSTR_MAX+1];
1632 
1633 	KASSERT(la->la_pending > 0, ("no pending event"));
1634 
1635 	LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__,
1636 	    lacp_format_lagid(&la->la_actor, &la->la_partner,
1637 	    buf, sizeof(buf)),
1638 	    la->la_pending, la->la_pending - 1));
1639 
1640 	la->la_pending--;
1641 }
1642 
1643 /* periodic transmit machine */
1644 
1645 static void
1646 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate)
1647 {
1648 	if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state,
1649 	    LACP_STATE_TIMEOUT)) {
1650 		return;
1651 	}
1652 
1653 	LACP_DPRINTF((lp, "partner timeout changed\n"));
1654 
1655 	/*
1656 	 * FAST_PERIODIC -> SLOW_PERIODIC
1657 	 * or
1658 	 * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC
1659 	 *
1660 	 * let lacp_sm_ptx_tx_schedule to update timeout.
1661 	 */
1662 
1663 	LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1664 
1665 	/*
1666 	 * if timeout has been shortened, assert NTT.
1667 	 */
1668 
1669 	if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) {
1670 		lacp_sm_assert_ntt(lp);
1671 	}
1672 }
1673 
1674 static void
1675 lacp_sm_ptx_tx_schedule(struct lacp_port *lp)
1676 {
1677 	int timeout;
1678 
1679 	if (!(lp->lp_state & LACP_STATE_ACTIVITY) &&
1680 	    !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) {
1681 		/*
1682 		 * NO_PERIODIC
1683 		 */
1684 
1685 		LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1686 		return;
1687 	}
1688 
1689 	if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) {
1690 		return;
1691 	}
1692 
1693 	timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ?
1694 	    LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME;
1695 
1696 	LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout);
1697 }
1698 
1699 static void
1700 lacp_sm_ptx_timer(struct lacp_port *lp)
1701 {
1702 	lacp_sm_assert_ntt(lp);
1703 }
1704 
1705 static void
1706 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du)
1707 {
1708 	int timeout;
1709 
1710 	/*
1711 	 * check LACP_DISABLED first
1712 	 */
1713 
1714 	if (!(lp->lp_state & LACP_STATE_AGGREGATION)) {
1715 		return;
1716 	}
1717 
1718 	/*
1719 	 * check loopback condition.
1720 	 */
1721 
1722 	if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid,
1723 	    &lp->lp_actor.lip_systemid)) {
1724 		return;
1725 	}
1726 
1727 	/*
1728 	 * EXPIRED, DEFAULTED, CURRENT -> CURRENT
1729 	 */
1730 
1731 	microuptime(&lp->lp_last_lacpdu_rx);
1732 	lacp_sm_rx_update_selected(lp, du);
1733 	lacp_sm_rx_update_ntt(lp, du);
1734 	lacp_sm_rx_record_pdu(lp, du);
1735 
1736 	timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ?
1737 	    LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME;
1738 	LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout);
1739 
1740 	lp->lp_state &= ~LACP_STATE_EXPIRED;
1741 
1742 	/*
1743 	 * kick transmit machine without waiting the next tick.
1744 	 */
1745 
1746 	lacp_sm_tx(lp);
1747 }
1748 
1749 static void
1750 lacp_sm_rx_set_expired(struct lacp_port *lp)
1751 {
1752 	lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1753 	lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT;
1754 	LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME);
1755 	lp->lp_state |= LACP_STATE_EXPIRED;
1756 }
1757 
1758 static void
1759 lacp_sm_rx_timer(struct lacp_port *lp)
1760 {
1761 	if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) {
1762 		/* CURRENT -> EXPIRED */
1763 		LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__));
1764 		lacp_sm_rx_set_expired(lp);
1765 	} else {
1766 		/* EXPIRED -> DEFAULTED */
1767 		LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__));
1768 		lacp_sm_rx_update_default_selected(lp);
1769 		lacp_sm_rx_record_default(lp);
1770 		lp->lp_state &= ~LACP_STATE_EXPIRED;
1771 	}
1772 }
1773 
1774 static void
1775 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du)
1776 {
1777 	boolean_t active;
1778 	uint8_t oldpstate;
1779 	char buf[LACP_STATESTR_MAX+1];
1780 
1781 	LACP_TRACE(lp);
1782 
1783 	oldpstate = lp->lp_partner.lip_state;
1784 
1785 	active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY)
1786 	    || ((lp->lp_state & LACP_STATE_ACTIVITY) &&
1787 	    (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY));
1788 
1789 	lp->lp_partner = du->ldu_actor;
1790 	if (active &&
1791 	    ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1792 	    LACP_STATE_AGGREGATION) &&
1793 	    !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner))
1794 	    || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) {
1795 		/*
1796 		 * XXX Maintain legacy behavior of leaving the
1797 		 * LACP_STATE_SYNC bit unchanged from the partner's
1798 		 * advertisement if lsc_strict_mode is false.
1799 		 * TODO: We should re-examine the concept of the "strict mode"
1800 		 * to ensure it makes sense to maintain a non-strict mode.
1801 		 */
1802 		if (lp->lp_lsc->lsc_strict_mode)
1803 			lp->lp_partner.lip_state |= LACP_STATE_SYNC;
1804 	} else {
1805 		lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1806 	}
1807 
1808 	lp->lp_state &= ~LACP_STATE_DEFAULTED;
1809 
1810 	if (oldpstate != lp->lp_partner.lip_state) {
1811 		LACP_DPRINTF((lp, "old pstate %s\n",
1812 		    lacp_format_state(oldpstate, buf, sizeof(buf))));
1813 		LACP_DPRINTF((lp, "new pstate %s\n",
1814 		    lacp_format_state(lp->lp_partner.lip_state, buf,
1815 		    sizeof(buf))));
1816 	}
1817 
1818 	lacp_sm_ptx_update_timeout(lp, oldpstate);
1819 }
1820 
1821 static void
1822 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du)
1823 {
1824 
1825 	LACP_TRACE(lp);
1826 
1827 	if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) ||
1828 	    !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1829 	    LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) {
1830 		LACP_DPRINTF((lp, "%s: assert ntt\n", __func__));
1831 		lacp_sm_assert_ntt(lp);
1832 	}
1833 }
1834 
1835 static void
1836 lacp_sm_rx_record_default(struct lacp_port *lp)
1837 {
1838 	uint8_t oldpstate;
1839 
1840 	LACP_TRACE(lp);
1841 
1842 	oldpstate = lp->lp_partner.lip_state;
1843 	if (lp->lp_lsc->lsc_strict_mode)
1844 		lp->lp_partner = lacp_partner_admin_strict;
1845 	else
1846 		lp->lp_partner = lacp_partner_admin_optimistic;
1847 	lp->lp_state |= LACP_STATE_DEFAULTED;
1848 	lacp_sm_ptx_update_timeout(lp, oldpstate);
1849 }
1850 
1851 static void
1852 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp,
1853     const struct lacp_peerinfo *info)
1854 {
1855 
1856 	LACP_TRACE(lp);
1857 
1858 	if (lacp_compare_peerinfo(&lp->lp_partner, info) ||
1859 	    !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state,
1860 	    LACP_STATE_AGGREGATION)) {
1861 		lp->lp_selected = LACP_UNSELECTED;
1862 		/* mux machine will clean up lp->lp_aggregator */
1863 	}
1864 }
1865 
1866 static void
1867 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du)
1868 {
1869 
1870 	LACP_TRACE(lp);
1871 
1872 	lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor);
1873 }
1874 
1875 static void
1876 lacp_sm_rx_update_default_selected(struct lacp_port *lp)
1877 {
1878 
1879 	LACP_TRACE(lp);
1880 
1881 	if (lp->lp_lsc->lsc_strict_mode)
1882 		lacp_sm_rx_update_selected_from_peerinfo(lp,
1883 		    &lacp_partner_admin_strict);
1884 	else
1885 		lacp_sm_rx_update_selected_from_peerinfo(lp,
1886 		    &lacp_partner_admin_optimistic);
1887 }
1888 
1889 /* transmit machine */
1890 
1891 static void
1892 lacp_sm_tx(struct lacp_port *lp)
1893 {
1894 	int error = 0;
1895 
1896 	if (!(lp->lp_state & LACP_STATE_AGGREGATION)
1897 #if 1
1898 	    || (!(lp->lp_state & LACP_STATE_ACTIVITY)
1899 	    && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY))
1900 #endif
1901 	    ) {
1902 		lp->lp_flags &= ~LACP_PORT_NTT;
1903 	}
1904 
1905 	if (!(lp->lp_flags & LACP_PORT_NTT)) {
1906 		return;
1907 	}
1908 
1909 	/* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */
1910 	if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent,
1911 		    (3 / LACP_FAST_PERIODIC_TIME)) == 0) {
1912 		LACP_DPRINTF((lp, "rate limited pdu\n"));
1913 		return;
1914 	}
1915 
1916 	if (((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_tx_test) == 0) {
1917 		error = lacp_xmit_lacpdu(lp);
1918 	} else {
1919 		LACP_TPRINTF((lp, "Dropping TX PDU\n"));
1920 	}
1921 
1922 	if (error == 0) {
1923 		lp->lp_flags &= ~LACP_PORT_NTT;
1924 	} else {
1925 		LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n",
1926 		    error));
1927 	}
1928 }
1929 
1930 static void
1931 lacp_sm_assert_ntt(struct lacp_port *lp)
1932 {
1933 
1934 	lp->lp_flags |= LACP_PORT_NTT;
1935 }
1936 
1937 static void
1938 lacp_run_timers(struct lacp_port *lp)
1939 {
1940 	int i;
1941 	struct timeval time_diff;
1942 
1943 	for (i = 0; i < LACP_NTIMER; i++) {
1944 		KASSERT(lp->lp_timer[i] >= 0,
1945 		    ("invalid timer value %d", lp->lp_timer[i]));
1946 		if (lp->lp_timer[i] == 0) {
1947 			continue;
1948 		} else {
1949 			if (i == LACP_TIMER_CURRENT_WHILE) {
1950 				microuptime(&time_diff);
1951 				timevalsub(&time_diff, &lp->lp_last_lacpdu_rx);
1952 				if (time_diff.tv_sec) {
1953 					/* At least one sec has elapsed since last LACP packet. */
1954 					--lp->lp_timer[i];
1955 				}
1956 			} else {
1957 				--lp->lp_timer[i];
1958 			}
1959 
1960 			if ((lp->lp_timer[i] <= 0) && (lacp_timer_funcs[i])) {
1961 				(*lacp_timer_funcs[i])(lp);
1962 			}
1963 		}
1964 	}
1965 }
1966 
1967 int
1968 lacp_marker_input(struct lacp_port *lp, struct mbuf *m)
1969 {
1970 	struct lacp_softc *lsc = lp->lp_lsc;
1971 	struct lagg_port *lgp = lp->lp_lagg;
1972 	struct lacp_port *lp2;
1973 	struct markerdu *mdu;
1974 	int error = 0;
1975 	int pending = 0;
1976 
1977 	if (m->m_pkthdr.len != sizeof(*mdu)) {
1978 		goto bad;
1979 	}
1980 
1981 	if ((m->m_flags & M_MCAST) == 0) {
1982 		goto bad;
1983 	}
1984 
1985 	if (m->m_len < sizeof(*mdu)) {
1986 		m = m_pullup(m, sizeof(*mdu));
1987 		if (m == NULL) {
1988 			return (ENOMEM);
1989 		}
1990 	}
1991 
1992 	mdu = mtod(m, struct markerdu *);
1993 
1994 	if (memcmp(&mdu->mdu_eh.ether_dhost,
1995 	    &ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
1996 		goto bad;
1997 	}
1998 
1999 	if (mdu->mdu_sph.sph_version != 1) {
2000 		goto bad;
2001 	}
2002 
2003 	switch (mdu->mdu_tlv.tlv_type) {
2004 	case MARKER_TYPE_INFO:
2005 		if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
2006 		    marker_info_tlv_template, TRUE)) {
2007 			goto bad;
2008 		}
2009 		mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE;
2010 		memcpy(&mdu->mdu_eh.ether_dhost,
2011 		    &ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN);
2012 		memcpy(&mdu->mdu_eh.ether_shost,
2013 		    lgp->lp_lladdr, ETHER_ADDR_LEN);
2014 		error = lagg_enqueue(lp->lp_ifp, m);
2015 		break;
2016 
2017 	case MARKER_TYPE_RESPONSE:
2018 		if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
2019 		    marker_response_tlv_template, TRUE)) {
2020 			goto bad;
2021 		}
2022 		LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n",
2023 		    ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system,
2024 		    ":", ntohl(mdu->mdu_info.mi_rq_xid)));
2025 
2026 		/* Verify that it is the last marker we sent out */
2027 		if (memcmp(&mdu->mdu_info, &lp->lp_marker,
2028 		    sizeof(struct lacp_markerinfo)))
2029 			goto bad;
2030 
2031 		LACP_LOCK(lsc);
2032 		lp->lp_flags &= ~LACP_PORT_MARK;
2033 
2034 		if (lsc->lsc_suppress_distributing) {
2035 			/* Check if any ports are waiting for a response */
2036 			LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) {
2037 				if (lp2->lp_flags & LACP_PORT_MARK) {
2038 					pending = 1;
2039 					break;
2040 				}
2041 			}
2042 
2043 			if (pending == 0) {
2044 				/* All interface queues are clear */
2045 				LACP_DPRINTF((NULL, "queue flush complete\n"));
2046 				lsc->lsc_suppress_distributing = FALSE;
2047 			}
2048 		}
2049 		LACP_UNLOCK(lsc);
2050 		m_freem(m);
2051 		break;
2052 
2053 	default:
2054 		goto bad;
2055 	}
2056 
2057 	return (error);
2058 
2059 bad:
2060 	LACP_DPRINTF((lp, "bad marker frame\n"));
2061 	m_freem(m);
2062 	return (EINVAL);
2063 }
2064 
2065 static int
2066 tlv_check(const void *p, size_t size, const struct tlvhdr *tlv,
2067     const struct tlv_template *tmpl, boolean_t check_type)
2068 {
2069 	while (/* CONSTCOND */ 1) {
2070 		if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) {
2071 			return (EINVAL);
2072 		}
2073 		if ((check_type && tlv->tlv_type != tmpl->tmpl_type) ||
2074 		    tlv->tlv_length != tmpl->tmpl_length) {
2075 			return (EINVAL);
2076 		}
2077 		if (tmpl->tmpl_type == 0) {
2078 			break;
2079 		}
2080 		tlv = (const struct tlvhdr *)
2081 		    ((const char *)tlv + tlv->tlv_length);
2082 		tmpl++;
2083 	}
2084 
2085 	return (0);
2086 }
2087 
2088 /* Debugging */
2089 const char *
2090 lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen)
2091 {
2092 	snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X",
2093 	    (int)mac[0],
2094 	    (int)mac[1],
2095 	    (int)mac[2],
2096 	    (int)mac[3],
2097 	    (int)mac[4],
2098 	    (int)mac[5]);
2099 
2100 	return (buf);
2101 }
2102 
2103 const char *
2104 lacp_format_systemid(const struct lacp_systemid *sysid,
2105     char *buf, size_t buflen)
2106 {
2107 	char macbuf[LACP_MACSTR_MAX+1];
2108 
2109 	snprintf(buf, buflen, "%04X,%s",
2110 	    ntohs(sysid->lsi_prio),
2111 	    lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf)));
2112 
2113 	return (buf);
2114 }
2115 
2116 const char *
2117 lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen)
2118 {
2119 	snprintf(buf, buflen, "%04X,%04X",
2120 	    ntohs(portid->lpi_prio),
2121 	    ntohs(portid->lpi_portno));
2122 
2123 	return (buf);
2124 }
2125 
2126 const char *
2127 lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen)
2128 {
2129 	char sysid[LACP_SYSTEMIDSTR_MAX+1];
2130 	char portid[LACP_PORTIDSTR_MAX+1];
2131 
2132 	snprintf(buf, buflen, "(%s,%04X,%s)",
2133 	    lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)),
2134 	    ntohs(peer->lip_key),
2135 	    lacp_format_portid(&peer->lip_portid, portid, sizeof(portid)));
2136 
2137 	return (buf);
2138 }
2139 
2140 const char *
2141 lacp_format_lagid(const struct lacp_peerinfo *a,
2142     const struct lacp_peerinfo *b, char *buf, size_t buflen)
2143 {
2144 	char astr[LACP_PARTNERSTR_MAX+1];
2145 	char bstr[LACP_PARTNERSTR_MAX+1];
2146 
2147 #if 0
2148 	/*
2149 	 * there's a convention to display small numbered peer
2150 	 * in the left.
2151 	 */
2152 
2153 	if (lacp_compare_peerinfo(a, b) > 0) {
2154 		const struct lacp_peerinfo *t;
2155 
2156 		t = a;
2157 		a = b;
2158 		b = t;
2159 	}
2160 #endif
2161 
2162 	snprintf(buf, buflen, "[%s,%s]",
2163 	    lacp_format_partner(a, astr, sizeof(astr)),
2164 	    lacp_format_partner(b, bstr, sizeof(bstr)));
2165 
2166 	return (buf);
2167 }
2168 
2169 const char *
2170 lacp_format_lagid_aggregator(const struct lacp_aggregator *la,
2171     char *buf, size_t buflen)
2172 {
2173 	if (la == NULL) {
2174 		return ("(none)");
2175 	}
2176 
2177 	return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen));
2178 }
2179 
2180 const char *
2181 lacp_format_state(uint8_t state, char *buf, size_t buflen)
2182 {
2183 	snprintf(buf, buflen, "%b", state, LACP_STATE_BITS);
2184 	return (buf);
2185 }
2186 
2187 static void
2188 lacp_dump_lacpdu(const struct lacpdu *du)
2189 {
2190 	char buf[LACP_PARTNERSTR_MAX+1];
2191 	char buf2[LACP_STATESTR_MAX+1];
2192 
2193 	printf("actor=%s\n",
2194 	    lacp_format_partner(&du->ldu_actor, buf, sizeof(buf)));
2195 	printf("actor.state=%s\n",
2196 	    lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2)));
2197 	printf("partner=%s\n",
2198 	    lacp_format_partner(&du->ldu_partner, buf, sizeof(buf)));
2199 	printf("partner.state=%s\n",
2200 	    lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2)));
2201 
2202 	printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay));
2203 }
2204 
2205 static void
2206 lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...)
2207 {
2208 	va_list va;
2209 
2210 	if (lp) {
2211 		printf("%s: ", lp->lp_ifp->if_xname);
2212 	}
2213 
2214 	va_start(va, fmt);
2215 	vprintf(fmt, va);
2216 	va_end(va);
2217 }
2218