xref: /freebsd/sys/net/ieee8023ad_lacp.c (revision b1879975794772ee51f0b4865753364c7d7626c3)
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
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 #include "opt_kern_tls.h"
34 #include "opt_ratelimit.h"
35 
36 #include <sys/param.h>
37 #include <sys/callout.h>
38 #include <sys/eventhandler.h>
39 #include <sys/mbuf.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/kernel.h> /* hz */
43 #include <sys/socket.h> /* for net/if.h */
44 #include <sys/sockio.h>
45 #include <sys/sysctl.h>
46 #include <machine/stdarg.h>
47 #include <sys/lock.h>
48 #include <sys/rwlock.h>
49 #include <sys/taskqueue.h>
50 #include <sys/time.h>
51 
52 #include <net/if.h>
53 #include <net/if_var.h>
54 #include <net/if_private.h>
55 #include <net/if_dl.h>
56 #include <net/ethernet.h>
57 #include <net/infiniband.h>
58 #include <net/if_media.h>
59 #include <net/if_types.h>
60 
61 #include <net/if_lagg.h>
62 #include <net/ieee8023ad_lacp.h>
63 
64 /*
65  * actor system priority and port priority.
66  * XXX should be configurable.
67  */
68 
69 #define	LACP_SYSTEM_PRIO	0x8000
70 #define	LACP_PORT_PRIO		0x8000
71 
72 const uint8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] =
73     { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 };
74 
75 static const struct tlv_template lacp_info_tlv_template[] = {
76 	{ LACP_TYPE_ACTORINFO,
77 	    sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
78 	{ LACP_TYPE_PARTNERINFO,
79 	    sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
80 	{ LACP_TYPE_COLLECTORINFO,
81 	    sizeof(struct tlvhdr) + sizeof(struct lacp_collectorinfo) },
82 	{ 0, 0 },
83 };
84 
85 static const struct tlv_template marker_info_tlv_template[] = {
86 	{ MARKER_TYPE_INFO,
87 	    sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
88 	{ 0, 0 },
89 };
90 
91 static const struct tlv_template marker_response_tlv_template[] = {
92 	{ MARKER_TYPE_RESPONSE,
93 	    sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
94 	{ 0, 0 },
95 };
96 
97 typedef void (*lacp_timer_func_t)(struct lacp_port *);
98 
99 static void	lacp_fill_actorinfo(struct lacp_port *, struct lacp_peerinfo *);
100 static void	lacp_fill_markerinfo(struct lacp_port *,
101 		    struct lacp_markerinfo *);
102 
103 static uint64_t	lacp_aggregator_bandwidth(struct lacp_aggregator *);
104 static void	lacp_suppress_distributing(struct lacp_softc *,
105 		    struct lacp_aggregator *);
106 static void	lacp_transit_expire(void *);
107 static void	lacp_update_portmap(struct lacp_softc *);
108 static void	lacp_select_active_aggregator(struct lacp_softc *);
109 static uint16_t	lacp_compose_key(struct lacp_port *);
110 static int	tlv_check(const void *, size_t, const struct tlvhdr *,
111 		    const struct tlv_template *, boolean_t);
112 static void	lacp_tick(void *);
113 
114 static void	lacp_fill_aggregator_id(struct lacp_aggregator *,
115 		    const struct lacp_port *);
116 static void	lacp_fill_aggregator_id_peer(struct lacp_peerinfo *,
117 		    const struct lacp_peerinfo *);
118 static bool	lacp_aggregator_is_compatible(const struct lacp_aggregator *,
119 		    const struct lacp_port *);
120 static bool	lacp_peerinfo_is_compatible(const struct lacp_peerinfo *,
121 		    const struct lacp_peerinfo *);
122 
123 static struct lacp_aggregator *lacp_aggregator_get(struct lacp_softc *,
124 		    struct lacp_port *);
125 static void	lacp_aggregator_addref(struct lacp_softc *,
126 		    struct lacp_aggregator *);
127 static void	lacp_aggregator_delref(struct lacp_softc *,
128 		    struct lacp_aggregator *);
129 
130 /* receive machine */
131 
132 static int	lacp_pdu_input(struct lacp_port *, struct mbuf *);
133 static int	lacp_marker_input(struct lacp_port *, struct mbuf *);
134 static void	lacp_sm_rx(struct lacp_port *, const struct lacpdu *);
135 static void	lacp_sm_rx_timer(struct lacp_port *);
136 static void	lacp_sm_rx_set_expired(struct lacp_port *);
137 static void	lacp_sm_rx_update_ntt(struct lacp_port *,
138 		    const struct lacpdu *);
139 static void	lacp_sm_rx_record_pdu(struct lacp_port *,
140 		    const struct lacpdu *);
141 static void	lacp_sm_rx_update_selected(struct lacp_port *,
142 		    const struct lacpdu *);
143 static void	lacp_sm_rx_record_default(struct lacp_port *);
144 static void	lacp_sm_rx_update_default_selected(struct lacp_port *);
145 static void	lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *,
146 		    const struct lacp_peerinfo *);
147 
148 /* mux machine */
149 
150 static void	lacp_sm_mux(struct lacp_port *);
151 static void	lacp_set_mux(struct lacp_port *, enum lacp_mux_state);
152 static void	lacp_sm_mux_timer(struct lacp_port *);
153 
154 /* periodic transmit machine */
155 
156 static void	lacp_sm_ptx_update_timeout(struct lacp_port *, uint8_t);
157 static void	lacp_sm_ptx_tx_schedule(struct lacp_port *);
158 static void	lacp_sm_ptx_timer(struct lacp_port *);
159 
160 /* transmit machine */
161 
162 static void	lacp_sm_tx(struct lacp_port *);
163 static void	lacp_sm_assert_ntt(struct lacp_port *);
164 
165 static void	lacp_run_timers(struct lacp_port *);
166 static int	lacp_compare_peerinfo(const struct lacp_peerinfo *,
167 		    const struct lacp_peerinfo *);
168 static int	lacp_compare_systemid(const struct lacp_systemid *,
169 		    const struct lacp_systemid *);
170 static void	lacp_port_enable(struct lacp_port *);
171 static void	lacp_port_disable(struct lacp_port *);
172 static void	lacp_select(struct lacp_port *);
173 static void	lacp_unselect(struct lacp_port *);
174 static void	lacp_disable_collecting(struct lacp_port *);
175 static void	lacp_enable_collecting(struct lacp_port *);
176 static void	lacp_disable_distributing(struct lacp_port *);
177 static void	lacp_enable_distributing(struct lacp_port *);
178 static int	lacp_xmit_lacpdu(struct lacp_port *);
179 static int	lacp_xmit_marker(struct lacp_port *);
180 
181 /* Debugging */
182 
183 static void	lacp_dump_lacpdu(const struct lacpdu *);
184 static const char *lacp_format_partner(const struct lacp_peerinfo *, char *,
185 		    size_t);
186 static const char *lacp_format_lagid(const struct lacp_peerinfo *,
187 		    const struct lacp_peerinfo *, char *, size_t);
188 static const char *lacp_format_lagid_aggregator(const struct lacp_aggregator *,
189 		    char *, size_t);
190 static const char *lacp_format_state(uint8_t, char *, size_t);
191 static const char *lacp_format_mac(const uint8_t *, char *, size_t);
192 static const char *lacp_format_systemid(const struct lacp_systemid *, char *,
193 		    size_t);
194 static const char *lacp_format_portid(const struct lacp_portid *, char *,
195 		    size_t);
196 static void	lacp_dprintf(const struct lacp_port *, const char *, ...)
197 		    __attribute__((__format__(__printf__, 2, 3)));
198 
199 VNET_DEFINE_STATIC(int, lacp_debug);
200 #define	V_lacp_debug	VNET(lacp_debug)
201 SYSCTL_NODE(_net_link_lagg, OID_AUTO, lacp, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
202     "ieee802.3ad");
203 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, debug, CTLFLAG_RWTUN | CTLFLAG_VNET,
204     &VNET_NAME(lacp_debug), 0, "Enable LACP debug logging (1=debug, 2=trace)");
205 
206 VNET_DEFINE_STATIC(int, lacp_default_strict_mode) = 1;
207 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, default_strict_mode,
208     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(lacp_default_strict_mode), 0,
209     "LACP strict protocol compliance default");
210 #define LACP_DPRINTF(a) if (V_lacp_debug & 0x01) { lacp_dprintf a ; }
211 #define LACP_TRACE(a) if (V_lacp_debug & 0x02) { lacp_dprintf(a,"%s\n",__func__); }
212 #define LACP_TPRINTF(a) if (V_lacp_debug & 0x04) { lacp_dprintf a ; }
213 
214 /*
215  * partner administration variables.
216  * XXX should be configurable.
217  */
218 
219 static const struct lacp_peerinfo lacp_partner_admin_optimistic = {
220 	.lip_systemid = { .lsi_prio = 0xffff },
221 	.lip_portid = { .lpi_prio = 0xffff },
222 	.lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION |
223 	    LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING,
224 };
225 
226 static const struct lacp_peerinfo lacp_partner_admin_strict = {
227 	.lip_systemid = { .lsi_prio = 0xffff },
228 	.lip_portid = { .lpi_prio = 0xffff },
229 	.lip_state = 0,
230 };
231 
232 static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = {
233 	[LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer,
234 	[LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer,
235 	[LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer,
236 };
237 
238 struct mbuf *
239 lacp_input(struct lagg_port *lgp, struct mbuf *m)
240 {
241 	struct lacp_port *lp = LACP_PORT(lgp);
242 	uint8_t subtype;
243 
244 	if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) {
245 		m_freem(m);
246 		return (NULL);
247 	}
248 
249 	m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype);
250 	switch (subtype) {
251 		case SLOWPROTOCOLS_SUBTYPE_LACP:
252 			lacp_pdu_input(lp, m);
253 			return (NULL);
254 
255 		case SLOWPROTOCOLS_SUBTYPE_MARKER:
256 			lacp_marker_input(lp, m);
257 			return (NULL);
258 	}
259 
260 	/* Not a subtype we are interested in */
261 	return (m);
262 }
263 
264 /*
265  * lacp_pdu_input: process lacpdu
266  */
267 static int
268 lacp_pdu_input(struct lacp_port *lp, struct mbuf *m)
269 {
270 	struct lacp_softc *lsc = lp->lp_lsc;
271 	struct lacpdu *du;
272 	int error = 0;
273 
274 	if (m->m_pkthdr.len != sizeof(*du)) {
275 		goto bad;
276 	}
277 
278 	if ((m->m_flags & M_MCAST) == 0) {
279 		goto bad;
280 	}
281 
282 	if (m->m_len < sizeof(*du)) {
283 		m = m_pullup(m, sizeof(*du));
284 		if (m == NULL) {
285 			return (ENOMEM);
286 		}
287 	}
288 
289 	du = mtod(m, struct lacpdu *);
290 
291 	if (memcmp(&du->ldu_eh.ether_dhost,
292 	    &ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
293 		goto bad;
294 	}
295 
296 	/*
297 	 * ignore the version for compatibility with
298 	 * the future protocol revisions.
299 	 */
300 #if 0
301 	if (du->ldu_sph.sph_version != 1) {
302 		goto bad;
303 	}
304 #endif
305 
306 	/*
307 	 * ignore tlv types for compatibility with
308 	 * the future protocol revisions.
309 	 */
310 	if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor,
311 	    lacp_info_tlv_template, FALSE)) {
312 		goto bad;
313 	}
314 
315         if (V_lacp_debug > 0) {
316 		lacp_dprintf(lp, "lacpdu receive\n");
317 		lacp_dump_lacpdu(du);
318 	}
319 
320 	if ((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_rx_test) {
321 		LACP_TPRINTF((lp, "Dropping RX PDU\n"));
322 		goto bad;
323 	}
324 
325 	LACP_LOCK(lsc);
326 	lacp_sm_rx(lp, du);
327 	LACP_UNLOCK(lsc);
328 
329 	m_freem(m);
330 	return (error);
331 
332 bad:
333 	m_freem(m);
334 	return (EINVAL);
335 }
336 
337 static void
338 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info)
339 {
340 	struct lagg_port *lgp = lp->lp_lagg;
341 	struct lagg_softc *sc = lgp->lp_softc;
342 
343 	info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO);
344 	memcpy(&info->lip_systemid.lsi_mac,
345 	    IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
346 	info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO);
347 	info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index);
348 	info->lip_state = lp->lp_state;
349 }
350 
351 static void
352 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info)
353 {
354 	struct ifnet *ifp = lp->lp_ifp;
355 
356 	/* Fill in the port index and system id (encoded as the MAC) */
357 	info->mi_rq_port = htons(ifp->if_index);
358 	memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN);
359 	info->mi_rq_xid = htonl(0);
360 }
361 
362 static int
363 lacp_xmit_lacpdu(struct lacp_port *lp)
364 {
365 	struct lagg_port *lgp = lp->lp_lagg;
366 	struct mbuf *m;
367 	struct lacpdu *du;
368 	int error;
369 
370 	LACP_LOCK_ASSERT(lp->lp_lsc);
371 
372 	m = m_gethdr(M_NOWAIT, MT_DATA);
373 	if (m == NULL) {
374 		return (ENOMEM);
375 	}
376 	m->m_len = m->m_pkthdr.len = sizeof(*du);
377 
378 	du = mtod(m, struct lacpdu *);
379 	memset(du, 0, sizeof(*du));
380 
381 	memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
382 	    ETHER_ADDR_LEN);
383 	memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
384 	du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW);
385 
386 	du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP;
387 	du->ldu_sph.sph_version = 1;
388 
389 	TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor));
390 	du->ldu_actor = lp->lp_actor;
391 
392 	TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO,
393 	    sizeof(du->ldu_partner));
394 	du->ldu_partner = lp->lp_partner;
395 
396 	TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO,
397 	    sizeof(du->ldu_collector));
398 	du->ldu_collector.lci_maxdelay = 0;
399 
400 	if (V_lacp_debug > 0) {
401 		lacp_dprintf(lp, "lacpdu transmit\n");
402 		lacp_dump_lacpdu(du);
403 	}
404 
405 	m->m_flags |= M_MCAST;
406 
407 	/*
408 	 * XXX should use higher priority queue.
409 	 * otherwise network congestion can break aggregation.
410 	 */
411 
412 	error = lagg_enqueue(lp->lp_ifp, m);
413 	return (error);
414 }
415 
416 static int
417 lacp_xmit_marker(struct lacp_port *lp)
418 {
419 	struct lagg_port *lgp = lp->lp_lagg;
420 	struct mbuf *m;
421 	struct markerdu *mdu;
422 	int error;
423 
424 	LACP_LOCK_ASSERT(lp->lp_lsc);
425 
426 	m = m_gethdr(M_NOWAIT, MT_DATA);
427 	if (m == NULL) {
428 		return (ENOMEM);
429 	}
430 	m->m_len = m->m_pkthdr.len = sizeof(*mdu);
431 
432 	mdu = mtod(m, struct markerdu *);
433 	memset(mdu, 0, sizeof(*mdu));
434 
435 	memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
436 	    ETHER_ADDR_LEN);
437 	memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
438 	mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW);
439 
440 	mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER;
441 	mdu->mdu_sph.sph_version = 1;
442 
443 	/* Bump the transaction id and copy over the marker info */
444 	lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1);
445 	TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info));
446 	mdu->mdu_info = lp->lp_marker;
447 
448 	LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n",
449 	    ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":",
450 	    ntohl(mdu->mdu_info.mi_rq_xid)));
451 
452 	m->m_flags |= M_MCAST;
453 	error = lagg_enqueue(lp->lp_ifp, m);
454 	return (error);
455 }
456 
457 void
458 lacp_linkstate(struct lagg_port *lgp)
459 {
460 	struct lacp_port *lp = LACP_PORT(lgp);
461 	struct lacp_softc *lsc = lp->lp_lsc;
462 	struct ifnet *ifp = lgp->lp_ifp;
463 	struct ifmediareq ifmr;
464 	int error = 0;
465 	u_int media;
466 	uint8_t old_state;
467 	uint16_t old_key;
468 
469 	bzero((char *)&ifmr, sizeof(ifmr));
470 	error = (*ifp->if_ioctl)(ifp, SIOCGIFXMEDIA, (caddr_t)&ifmr);
471 	if (error != 0) {
472 		bzero((char *)&ifmr, sizeof(ifmr));
473 		error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr);
474 	}
475 	if (error != 0)
476 		return;
477 
478 	LACP_LOCK(lsc);
479 	media = ifmr.ifm_active;
480 	LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, "
481 	    "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER,
482 	    (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP));
483 	old_state = lp->lp_state;
484 	old_key = lp->lp_key;
485 
486 	lp->lp_media = media;
487 	/*
488 	 * If the port is not an active full duplex Ethernet link then it can
489 	 * not be aggregated.
490 	 */
491 	if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 ||
492 	    ifp->if_link_state != LINK_STATE_UP) {
493 		lacp_port_disable(lp);
494 	} else {
495 		lacp_port_enable(lp);
496 	}
497 	lp->lp_key = lacp_compose_key(lp);
498 
499 	if (old_state != lp->lp_state || old_key != lp->lp_key) {
500 		LACP_DPRINTF((lp, "-> UNSELECTED\n"));
501 		lp->lp_selected = LACP_UNSELECTED;
502 	}
503 	LACP_UNLOCK(lsc);
504 }
505 
506 static void
507 lacp_tick(void *arg)
508 {
509 	struct lacp_softc *lsc = arg;
510 	struct lacp_port *lp;
511 
512 	LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
513 		if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0)
514 			continue;
515 
516 		CURVNET_SET(lp->lp_ifp->if_vnet);
517 		lacp_run_timers(lp);
518 
519 		lacp_select(lp);
520 		lacp_sm_mux(lp);
521 		lacp_sm_tx(lp);
522 		lacp_sm_ptx_tx_schedule(lp);
523 		CURVNET_RESTORE();
524 	}
525 	callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
526 }
527 
528 int
529 lacp_port_create(struct lagg_port *lgp)
530 {
531 	struct lagg_softc *sc = lgp->lp_softc;
532 	struct lacp_softc *lsc = LACP_SOFTC(sc);
533 	struct lacp_port *lp;
534 	struct ifnet *ifp = lgp->lp_ifp;
535 	struct sockaddr_dl sdl;
536 	struct ifmultiaddr *rifma = NULL;
537 	int error;
538 
539 	link_init_sdl(ifp, (struct sockaddr *)&sdl, IFT_ETHER);
540 	sdl.sdl_alen = ETHER_ADDR_LEN;
541 
542 	bcopy(&ethermulticastaddr_slowprotocols,
543 	    LLADDR(&sdl), ETHER_ADDR_LEN);
544 	error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
545 	if (error) {
546 		printf("%s: ADDMULTI failed on %s\n", __func__,
547 		    lgp->lp_ifp->if_xname);
548 		return (error);
549 	}
550 
551 	lp = malloc(sizeof(struct lacp_port),
552 	    M_DEVBUF, M_NOWAIT|M_ZERO);
553 	if (lp == NULL)
554 		return (ENOMEM);
555 
556 	LACP_LOCK(lsc);
557 	lgp->lp_psc = lp;
558 	lp->lp_ifp = ifp;
559 	lp->lp_lagg = lgp;
560 	lp->lp_lsc = lsc;
561 	lp->lp_ifma = rifma;
562 
563 	LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next);
564 
565 	lacp_fill_actorinfo(lp, &lp->lp_actor);
566 	lacp_fill_markerinfo(lp, &lp->lp_marker);
567 	lp->lp_state = LACP_STATE_ACTIVITY;
568 	lp->lp_aggregator = NULL;
569 	lacp_sm_rx_set_expired(lp);
570 	LACP_UNLOCK(lsc);
571 	lacp_linkstate(lgp);
572 
573 	return (0);
574 }
575 
576 void
577 lacp_port_destroy(struct lagg_port *lgp)
578 {
579 	struct lacp_port *lp = LACP_PORT(lgp);
580 	struct lacp_softc *lsc = lp->lp_lsc;
581 	int i;
582 
583 	LACP_LOCK(lsc);
584 	for (i = 0; i < LACP_NTIMER; i++) {
585 		LACP_TIMER_DISARM(lp, i);
586 	}
587 
588 	lacp_disable_collecting(lp);
589 	lacp_disable_distributing(lp);
590 	lacp_unselect(lp);
591 
592 	LIST_REMOVE(lp, lp_next);
593 	LACP_UNLOCK(lsc);
594 
595 	/* The address may have already been removed by if_purgemaddrs() */
596 	if (!lgp->lp_detaching)
597 		if_delmulti_ifma(lp->lp_ifma);
598 
599 	free(lp, M_DEVBUF);
600 }
601 
602 void
603 lacp_req(struct lagg_softc *sc, void *data)
604 {
605 	struct lacp_opreq *req = (struct lacp_opreq *)data;
606 	struct lacp_softc *lsc = LACP_SOFTC(sc);
607 	struct lacp_aggregator *la;
608 
609 	bzero(req, sizeof(struct lacp_opreq));
610 
611 	/*
612 	 * If the LACP softc is NULL, return with the opreq structure full of
613 	 * zeros.  It is normal for the softc to be NULL while the lagg is
614 	 * being destroyed.
615 	 */
616 	if (NULL == lsc)
617 		return;
618 
619 	la = lsc->lsc_active_aggregator;
620 	LACP_LOCK(lsc);
621 	if (la != NULL) {
622 		req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio);
623 		memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac,
624 		    ETHER_ADDR_LEN);
625 		req->actor_key = ntohs(la->la_actor.lip_key);
626 		req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio);
627 		req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno);
628 		req->actor_state = la->la_actor.lip_state;
629 
630 		req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio);
631 		memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac,
632 		    ETHER_ADDR_LEN);
633 		req->partner_key = ntohs(la->la_partner.lip_key);
634 		req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio);
635 		req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno);
636 		req->partner_state = la->la_partner.lip_state;
637 	}
638 	LACP_UNLOCK(lsc);
639 }
640 
641 void
642 lacp_portreq(struct lagg_port *lgp, void *data)
643 {
644 	struct lacp_opreq *req = (struct lacp_opreq *)data;
645 	struct lacp_port *lp = LACP_PORT(lgp);
646 	struct lacp_softc *lsc = lp->lp_lsc;
647 
648 	LACP_LOCK(lsc);
649 	req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio);
650 	memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac,
651 	    ETHER_ADDR_LEN);
652 	req->actor_key = ntohs(lp->lp_actor.lip_key);
653 	req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio);
654 	req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno);
655 	req->actor_state = lp->lp_actor.lip_state;
656 
657 	req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio);
658 	memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac,
659 	    ETHER_ADDR_LEN);
660 	req->partner_key = ntohs(lp->lp_partner.lip_key);
661 	req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio);
662 	req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno);
663 	req->partner_state = lp->lp_partner.lip_state;
664 	LACP_UNLOCK(lsc);
665 }
666 
667 static void
668 lacp_disable_collecting(struct lacp_port *lp)
669 {
670 	LACP_DPRINTF((lp, "collecting disabled\n"));
671 	lp->lp_state &= ~LACP_STATE_COLLECTING;
672 }
673 
674 static void
675 lacp_enable_collecting(struct lacp_port *lp)
676 {
677 	LACP_DPRINTF((lp, "collecting enabled\n"));
678 	lp->lp_state |= LACP_STATE_COLLECTING;
679 }
680 
681 static void
682 lacp_disable_distributing(struct lacp_port *lp)
683 {
684 	struct lacp_aggregator *la = lp->lp_aggregator;
685 	struct lacp_softc *lsc = lp->lp_lsc;
686 	struct lagg_softc *sc = lsc->lsc_softc;
687 	char buf[LACP_LAGIDSTR_MAX+1];
688 
689 	LACP_LOCK_ASSERT(lsc);
690 
691 	if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) {
692 		return;
693 	}
694 
695 	KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports"));
696 	KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports));
697 	KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid"));
698 
699 	LACP_DPRINTF((lp, "disable distributing on aggregator %s, "
700 	    "nports %d -> %d\n",
701 	    lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
702 	    la->la_nports, la->la_nports - 1));
703 
704 	TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q);
705 	la->la_nports--;
706 	sc->sc_active = la->la_nports;
707 
708 	if (lsc->lsc_active_aggregator == la) {
709 		lacp_suppress_distributing(lsc, la);
710 		lacp_select_active_aggregator(lsc);
711 		/* regenerate the port map, the active aggregator has changed */
712 		lacp_update_portmap(lsc);
713 	}
714 
715 	lp->lp_state &= ~LACP_STATE_DISTRIBUTING;
716 	if_link_state_change(sc->sc_ifp,
717 	    sc->sc_active ? LINK_STATE_UP : LINK_STATE_DOWN);
718 }
719 
720 static void
721 lacp_enable_distributing(struct lacp_port *lp)
722 {
723 	struct lacp_aggregator *la = lp->lp_aggregator;
724 	struct lacp_softc *lsc = lp->lp_lsc;
725 	struct lagg_softc *sc = lsc->lsc_softc;
726 	char buf[LACP_LAGIDSTR_MAX+1];
727 
728 	LACP_LOCK_ASSERT(lsc);
729 
730 	if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) {
731 		return;
732 	}
733 
734 	LACP_DPRINTF((lp, "enable distributing on aggregator %s, "
735 	    "nports %d -> %d\n",
736 	    lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
737 	    la->la_nports, la->la_nports + 1));
738 
739 	KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid"));
740 	TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q);
741 	la->la_nports++;
742 	sc->sc_active = la->la_nports;
743 
744 	lp->lp_state |= LACP_STATE_DISTRIBUTING;
745 
746 	if (lsc->lsc_active_aggregator == la) {
747 		lacp_suppress_distributing(lsc, la);
748 		lacp_update_portmap(lsc);
749 	} else
750 		/* try to become the active aggregator */
751 		lacp_select_active_aggregator(lsc);
752 
753 	if_link_state_change(sc->sc_ifp,
754 	    sc->sc_active ? LINK_STATE_UP : LINK_STATE_DOWN);
755 }
756 
757 static void
758 lacp_transit_expire(void *vp)
759 {
760 	struct lacp_softc *lsc = vp;
761 
762 	LACP_LOCK_ASSERT(lsc);
763 
764 	CURVNET_SET(lsc->lsc_softc->sc_ifp->if_vnet);
765 	LACP_TRACE(NULL);
766 	CURVNET_RESTORE();
767 
768 	lsc->lsc_suppress_distributing = FALSE;
769 }
770 
771 void
772 lacp_attach(struct lagg_softc *sc)
773 {
774 	struct lacp_softc *lsc;
775 
776 	lsc = malloc(sizeof(struct lacp_softc), M_DEVBUF, M_WAITOK | M_ZERO);
777 
778 	sc->sc_psc = lsc;
779 	lsc->lsc_softc = sc;
780 
781 	lsc->lsc_hashkey = m_ether_tcpip_hash_init();
782 	lsc->lsc_active_aggregator = NULL;
783 	lsc->lsc_strict_mode = VNET(lacp_default_strict_mode);
784 	LACP_LOCK_INIT(lsc);
785 	TAILQ_INIT(&lsc->lsc_aggregators);
786 	LIST_INIT(&lsc->lsc_ports);
787 
788 	callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0);
789 	callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0);
790 
791 	/* if the lagg is already up then do the same */
792 	if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING)
793 		lacp_init(sc);
794 }
795 
796 void
797 lacp_detach(void *psc)
798 {
799 	struct lacp_softc *lsc = (struct lacp_softc *)psc;
800 
801 	KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators),
802 	    ("aggregators still active"));
803 	KASSERT(lsc->lsc_active_aggregator == NULL,
804 	    ("aggregator still attached"));
805 
806 	callout_drain(&lsc->lsc_transit_callout);
807 	callout_drain(&lsc->lsc_callout);
808 
809 	LACP_LOCK_DESTROY(lsc);
810 	free(lsc, M_DEVBUF);
811 }
812 
813 void
814 lacp_init(struct lagg_softc *sc)
815 {
816 	struct lacp_softc *lsc = LACP_SOFTC(sc);
817 
818 	LACP_LOCK(lsc);
819 	callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
820 	LACP_UNLOCK(lsc);
821 }
822 
823 void
824 lacp_stop(struct lagg_softc *sc)
825 {
826 	struct lacp_softc *lsc = LACP_SOFTC(sc);
827 
828 	LACP_LOCK(lsc);
829 	callout_stop(&lsc->lsc_transit_callout);
830 	callout_stop(&lsc->lsc_callout);
831 	LACP_UNLOCK(lsc);
832 }
833 
834 struct lagg_port *
835 lacp_select_tx_port_by_hash(struct lagg_softc *sc, uint32_t hash,
836     uint8_t numa_domain, int *err)
837 {
838 	struct lacp_softc *lsc = LACP_SOFTC(sc);
839 	struct lacp_portmap *pm;
840 	struct lacp_port *lp;
841 	struct lacp_port **map;
842 	int count;
843 
844 	if (__predict_false(lsc->lsc_suppress_distributing)) {
845 		LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__));
846 		*err = ENOBUFS;
847 		return (NULL);
848 	}
849 
850 	pm = &lsc->lsc_pmap[lsc->lsc_activemap];
851 	if (pm->pm_count == 0) {
852 		LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__));
853 		*err = ENETDOWN;
854 		return (NULL);
855 	}
856 
857 #ifdef NUMA
858 	if ((sc->sc_opts & LAGG_OPT_USE_NUMA) &&
859 	    pm->pm_num_dom > 1 && numa_domain < MAXMEMDOM) {
860 		count = pm->pm_numa[numa_domain].count;
861 		if (count > 0) {
862 			map = pm->pm_numa[numa_domain].map;
863 		} else {
864 			/* No ports on this domain; use global hash. */
865 			map = pm->pm_map;
866 			count = pm->pm_count;
867 		}
868 	} else
869 #endif
870 	{
871 		map = pm->pm_map;
872 		count = pm->pm_count;
873 	}
874 
875 	hash %= count;
876 	lp = map[hash];
877 
878 	return (lp->lp_lagg);
879 }
880 
881 struct lagg_port *
882 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m, int *err)
883 {
884 	struct lacp_softc *lsc = LACP_SOFTC(sc);
885 	uint32_t hash;
886 	uint8_t numa_domain;
887 
888 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
889 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
890 		hash = m->m_pkthdr.flowid >> sc->flowid_shift;
891 	else
892 		hash = m_ether_tcpip_hash(sc->sc_flags, m, lsc->lsc_hashkey);
893 
894 	numa_domain = m->m_pkthdr.numa_domain;
895 	return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, err));
896 }
897 
898 /*
899  * lacp_suppress_distributing: drop transmit packets for a while
900  * to preserve packet ordering.
901  */
902 
903 static void
904 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la)
905 {
906 	struct lacp_port *lp;
907 
908 	if (lsc->lsc_active_aggregator != la) {
909 		return;
910 	}
911 
912 	LACP_TRACE(NULL);
913 
914 	lsc->lsc_suppress_distributing = TRUE;
915 
916 	/* send a marker frame down each port to verify the queues are empty */
917 	LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
918 		lp->lp_flags |= LACP_PORT_MARK;
919 		if (lacp_xmit_marker(lp) != 0)
920 			lp->lp_flags &= ~LACP_PORT_MARK;
921 	}
922 
923 	/* set a timeout for the marker frames */
924 	callout_reset(&lsc->lsc_transit_callout,
925 	    LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc);
926 }
927 
928 static int
929 lacp_compare_peerinfo(const struct lacp_peerinfo *a,
930     const struct lacp_peerinfo *b)
931 {
932 	return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state)));
933 }
934 
935 static int
936 lacp_compare_systemid(const struct lacp_systemid *a,
937     const struct lacp_systemid *b)
938 {
939 	return (memcmp(a, b, sizeof(*a)));
940 }
941 
942 #if 0	/* unused */
943 static int
944 lacp_compare_portid(const struct lacp_portid *a,
945     const struct lacp_portid *b)
946 {
947 	return (memcmp(a, b, sizeof(*a)));
948 }
949 #endif
950 
951 static uint64_t
952 lacp_aggregator_bandwidth(struct lacp_aggregator *la)
953 {
954 	struct lacp_port *lp;
955 	uint64_t speed;
956 
957 	lp = TAILQ_FIRST(&la->la_ports);
958 	if (lp == NULL) {
959 		return (0);
960 	}
961 
962 	speed = ifmedia_baudrate(lp->lp_media);
963 	speed *= la->la_nports;
964 	if (speed == 0) {
965 		LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n",
966 		    lp->lp_media, la->la_nports));
967 	}
968 
969 	return (speed);
970 }
971 
972 /*
973  * lacp_select_active_aggregator: select an aggregator to be used to transmit
974  * packets from lagg(4) interface.
975  */
976 
977 static void
978 lacp_select_active_aggregator(struct lacp_softc *lsc)
979 {
980 	struct lacp_aggregator *la;
981 	struct lacp_aggregator *best_la = NULL;
982 	uint64_t best_speed = 0;
983 	char buf[LACP_LAGIDSTR_MAX+1];
984 
985 	LACP_TRACE(NULL);
986 
987 	TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
988 		uint64_t speed;
989 
990 		if (la->la_nports == 0) {
991 			continue;
992 		}
993 
994 		speed = lacp_aggregator_bandwidth(la);
995 		LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n",
996 		    lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
997 		    speed, la->la_nports));
998 
999 		/*
1000 		 * This aggregator is chosen if the partner has a better
1001 		 * system priority or, the total aggregated speed is higher
1002 		 * or, it is already the chosen aggregator
1003 		 */
1004 		if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) <
1005 		    LACP_SYS_PRI(best_la->la_partner)) ||
1006 		    speed > best_speed ||
1007 		    (speed == best_speed &&
1008 		    la == lsc->lsc_active_aggregator)) {
1009 			best_la = la;
1010 			best_speed = speed;
1011 		}
1012 	}
1013 
1014 	KASSERT(best_la == NULL || best_la->la_nports > 0,
1015 	    ("invalid aggregator refcnt"));
1016 	KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports),
1017 	    ("invalid aggregator list"));
1018 
1019 	if (lsc->lsc_active_aggregator != best_la) {
1020 		LACP_DPRINTF((NULL, "active aggregator changed\n"));
1021 		LACP_DPRINTF((NULL, "old %s\n",
1022 		    lacp_format_lagid_aggregator(lsc->lsc_active_aggregator,
1023 		    buf, sizeof(buf))));
1024 	} else {
1025 		LACP_DPRINTF((NULL, "active aggregator not changed\n"));
1026 	}
1027 	LACP_DPRINTF((NULL, "new %s\n",
1028 	    lacp_format_lagid_aggregator(best_la, buf, sizeof(buf))));
1029 
1030 	if (lsc->lsc_active_aggregator != best_la) {
1031 		lsc->lsc_active_aggregator = best_la;
1032 		lacp_update_portmap(lsc);
1033 		if (best_la) {
1034 			lacp_suppress_distributing(lsc, best_la);
1035 		}
1036 	}
1037 }
1038 
1039 /*
1040  * Updated the inactive portmap array with the new list of ports and
1041  * make it live.
1042  */
1043 static void
1044 lacp_update_portmap(struct lacp_softc *lsc)
1045 {
1046 	struct lagg_softc *sc = lsc->lsc_softc;
1047 	struct lacp_aggregator *la;
1048 	struct lacp_portmap *p;
1049 	struct lacp_port *lp;
1050 	uint64_t speed;
1051 	u_int newmap;
1052 	int i;
1053 #ifdef NUMA
1054 	int count;
1055 	uint8_t domain;
1056 #endif
1057 
1058 	newmap = lsc->lsc_activemap == 0 ? 1 : 0;
1059 	p = &lsc->lsc_pmap[newmap];
1060 	la = lsc->lsc_active_aggregator;
1061 	speed = 0;
1062 	bzero(p, sizeof(struct lacp_portmap));
1063 
1064 	if (la != NULL && la->la_nports > 0) {
1065 		p->pm_count = la->la_nports;
1066 		i = 0;
1067 		TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q) {
1068 			p->pm_map[i++] = lp;
1069 #ifdef NUMA
1070 			domain = lp->lp_ifp->if_numa_domain;
1071 			if (domain >= MAXMEMDOM)
1072 				continue;
1073 			count = p->pm_numa[domain].count;
1074 			p->pm_numa[domain].map[count] = lp;
1075 			p->pm_numa[domain].count++;
1076 #endif
1077 		}
1078 		KASSERT(i == p->pm_count, ("Invalid port count"));
1079 
1080 #ifdef NUMA
1081 		for (i = 0; i < MAXMEMDOM; i++) {
1082 			if (p->pm_numa[i].count != 0)
1083 				p->pm_num_dom++;
1084 		}
1085 #endif
1086 		speed = lacp_aggregator_bandwidth(la);
1087 	}
1088 	sc->sc_ifp->if_baudrate = speed;
1089 	EVENTHANDLER_INVOKE(ifnet_event, sc->sc_ifp,
1090 	    IFNET_EVENT_UPDATE_BAUDRATE);
1091 
1092 	/* switch the active portmap over */
1093 	atomic_store_rel_int(&lsc->lsc_activemap, newmap);
1094 	LACP_DPRINTF((NULL, "Set table %d with %d ports\n",
1095 		    lsc->lsc_activemap,
1096 		    lsc->lsc_pmap[lsc->lsc_activemap].pm_count));
1097 }
1098 
1099 static uint16_t
1100 lacp_compose_key(struct lacp_port *lp)
1101 {
1102 	struct lagg_port *lgp = lp->lp_lagg;
1103 	struct lagg_softc *sc = lgp->lp_softc;
1104 	u_int media = lp->lp_media;
1105 	uint16_t key;
1106 
1107 	if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) {
1108 		/*
1109 		 * non-aggregatable links should have unique keys.
1110 		 *
1111 		 * XXX this isn't really unique as if_index is 16 bit.
1112 		 */
1113 
1114 		/* bit 0..14:	(some bits of) if_index of this port */
1115 		key = lp->lp_ifp->if_index;
1116 		/* bit 15:	1 */
1117 		key |= 0x8000;
1118 	} else {
1119 		u_int subtype = IFM_SUBTYPE(media);
1120 
1121 		KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type"));
1122 		KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface"));
1123 
1124 		/* bit 0..4:	IFM_SUBTYPE modulo speed */
1125 		switch (subtype) {
1126 		case IFM_10_T:
1127 		case IFM_10_2:
1128 		case IFM_10_5:
1129 		case IFM_10_STP:
1130 		case IFM_10_FL:
1131 			key = IFM_10_T;
1132 			break;
1133 		case IFM_100_TX:
1134 		case IFM_100_FX:
1135 		case IFM_100_T4:
1136 		case IFM_100_VG:
1137 		case IFM_100_T2:
1138 		case IFM_100_T:
1139 		case IFM_100_SGMII:
1140 			key = IFM_100_TX;
1141 			break;
1142 		case IFM_1000_SX:
1143 		case IFM_1000_LX:
1144 		case IFM_1000_CX:
1145 		case IFM_1000_T:
1146 		case IFM_1000_KX:
1147 		case IFM_1000_SGMII:
1148 		case IFM_1000_CX_SGMII:
1149 		case IFM_1000_BX:
1150 			key = IFM_1000_SX;
1151 			break;
1152 		case IFM_10G_LR:
1153 		case IFM_10G_SR:
1154 		case IFM_10G_CX4:
1155 		case IFM_10G_TWINAX:
1156 		case IFM_10G_TWINAX_LONG:
1157 		case IFM_10G_LRM:
1158 		case IFM_10G_T:
1159 		case IFM_10G_KX4:
1160 		case IFM_10G_KR:
1161 		case IFM_10G_CR1:
1162 		case IFM_10G_ER:
1163 		case IFM_10G_SFI:
1164 		case IFM_10G_AOC:
1165 			key = IFM_10G_LR;
1166 			break;
1167 		case IFM_20G_KR2:
1168 			key = IFM_20G_KR2;
1169 			break;
1170 		case IFM_2500_KX:
1171 		case IFM_2500_T:
1172 		case IFM_2500_X:
1173 			key = IFM_2500_KX;
1174 			break;
1175 		case IFM_5000_T:
1176 		case IFM_5000_KR:
1177 		case IFM_5000_KR_S:
1178 		case IFM_5000_KR1:
1179 			key = IFM_5000_T;
1180 			break;
1181 		case IFM_50G_PCIE:
1182 		case IFM_50G_CR2:
1183 		case IFM_50G_KR2:
1184 		case IFM_50G_KR4:
1185 		case IFM_50G_SR2:
1186 		case IFM_50G_LR2:
1187 		case IFM_50G_LAUI2_AC:
1188 		case IFM_50G_LAUI2:
1189 		case IFM_50G_AUI2_AC:
1190 		case IFM_50G_AUI2:
1191 		case IFM_50G_CP:
1192 		case IFM_50G_SR:
1193 		case IFM_50G_LR:
1194 		case IFM_50G_FR:
1195 		case IFM_50G_KR_PAM4:
1196 		case IFM_50G_AUI1_AC:
1197 		case IFM_50G_AUI1:
1198 			key = IFM_50G_PCIE;
1199 			break;
1200 		case IFM_56G_R4:
1201 			key = IFM_56G_R4;
1202 			break;
1203 		case IFM_25G_PCIE:
1204 		case IFM_25G_CR:
1205 		case IFM_25G_KR:
1206 		case IFM_25G_SR:
1207 		case IFM_25G_LR:
1208 		case IFM_25G_ACC:
1209 		case IFM_25G_AOC:
1210 		case IFM_25G_T:
1211 		case IFM_25G_CR_S:
1212 		case IFM_25G_CR1:
1213 		case IFM_25G_KR_S:
1214 		case IFM_25G_AUI:
1215 		case IFM_25G_KR1:
1216 			key = IFM_25G_PCIE;
1217 			break;
1218 		case IFM_40G_CR4:
1219 		case IFM_40G_SR4:
1220 		case IFM_40G_LR4:
1221 		case IFM_40G_LM4:
1222 		case IFM_40G_XLPPI:
1223 		case IFM_40G_KR4:
1224 		case IFM_40G_XLAUI:
1225 		case IFM_40G_XLAUI_AC:
1226 		case IFM_40G_ER4:
1227 			key = IFM_40G_CR4;
1228 			break;
1229 		case IFM_100G_CR4:
1230 		case IFM_100G_SR4:
1231 		case IFM_100G_KR4:
1232 		case IFM_100G_LR4:
1233 		case IFM_100G_CAUI4_AC:
1234 		case IFM_100G_CAUI4:
1235 		case IFM_100G_AUI4_AC:
1236 		case IFM_100G_AUI4:
1237 		case IFM_100G_CR_PAM4:
1238 		case IFM_100G_KR_PAM4:
1239 		case IFM_100G_CP2:
1240 		case IFM_100G_SR2:
1241 		case IFM_100G_DR:
1242 		case IFM_100G_KR2_PAM4:
1243 		case IFM_100G_CAUI2_AC:
1244 		case IFM_100G_CAUI2:
1245 		case IFM_100G_AUI2_AC:
1246 		case IFM_100G_AUI2:
1247 			key = IFM_100G_CR4;
1248 			break;
1249 		case IFM_200G_CR4_PAM4:
1250 		case IFM_200G_SR4:
1251 		case IFM_200G_FR4:
1252 		case IFM_200G_LR4:
1253 		case IFM_200G_DR4:
1254 		case IFM_200G_KR4_PAM4:
1255 		case IFM_200G_AUI4_AC:
1256 		case IFM_200G_AUI4:
1257 		case IFM_200G_AUI8_AC:
1258 		case IFM_200G_AUI8:
1259 			key = IFM_200G_CR4_PAM4;
1260 			break;
1261 		case IFM_400G_FR8:
1262 		case IFM_400G_LR8:
1263 		case IFM_400G_DR4:
1264 		case IFM_400G_AUI8_AC:
1265 		case IFM_400G_AUI8:
1266 			key = IFM_400G_FR8;
1267 			break;
1268 		default:
1269 			key = subtype;
1270 			break;
1271 		}
1272 		/* bit 5..14:	(some bits of) if_index of lagg device */
1273 		key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5);
1274 		/* bit 15:	0 */
1275 	}
1276 	return (htons(key));
1277 }
1278 
1279 static void
1280 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1281 {
1282 	char buf[LACP_LAGIDSTR_MAX+1];
1283 
1284 	LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1285 	    __func__,
1286 	    lacp_format_lagid(&la->la_actor, &la->la_partner,
1287 	    buf, sizeof(buf)),
1288 	    la->la_refcnt, la->la_refcnt + 1));
1289 
1290 	KASSERT(la->la_refcnt > 0, ("refcount <= 0"));
1291 	la->la_refcnt++;
1292 	KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount"));
1293 }
1294 
1295 static void
1296 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la)
1297 {
1298 	char buf[LACP_LAGIDSTR_MAX+1];
1299 
1300 	LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
1301 	    __func__,
1302 	    lacp_format_lagid(&la->la_actor, &la->la_partner,
1303 	    buf, sizeof(buf)),
1304 	    la->la_refcnt, la->la_refcnt - 1));
1305 
1306 	KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt"));
1307 	la->la_refcnt--;
1308 	if (la->la_refcnt > 0) {
1309 		return;
1310 	}
1311 
1312 	KASSERT(la->la_refcnt == 0, ("refcount not zero"));
1313 	KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active"));
1314 
1315 	TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q);
1316 
1317 	free(la, M_DEVBUF);
1318 }
1319 
1320 /*
1321  * lacp_aggregator_get: allocate an aggregator.
1322  */
1323 
1324 static struct lacp_aggregator *
1325 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp)
1326 {
1327 	struct lacp_aggregator *la;
1328 
1329 	la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT);
1330 	if (la) {
1331 		la->la_refcnt = 1;
1332 		la->la_nports = 0;
1333 		TAILQ_INIT(&la->la_ports);
1334 		la->la_pending = 0;
1335 		TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q);
1336 	}
1337 
1338 	return (la);
1339 }
1340 
1341 /*
1342  * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port.
1343  */
1344 
1345 static void
1346 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp)
1347 {
1348 	lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner);
1349 	lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor);
1350 
1351 	la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION;
1352 }
1353 
1354 static void
1355 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr,
1356     const struct lacp_peerinfo *lpi_port)
1357 {
1358 	memset(lpi_aggr, 0, sizeof(*lpi_aggr));
1359 	lpi_aggr->lip_systemid = lpi_port->lip_systemid;
1360 	lpi_aggr->lip_key = lpi_port->lip_key;
1361 }
1362 
1363 /*
1364  * lacp_aggregator_is_compatible: check if a port can join to an aggregator.
1365  */
1366 
1367 static bool
1368 lacp_aggregator_is_compatible(const struct lacp_aggregator *la,
1369     const struct lacp_port *lp)
1370 {
1371 	if (!(lp->lp_state & LACP_STATE_AGGREGATION) ||
1372 	    !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) {
1373 		return (false);
1374 	}
1375 
1376 	if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION))
1377 		return (false);
1378 
1379 	if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner))
1380 		return (false);
1381 
1382 	if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor))
1383 		return (false);
1384 
1385 	return (true);
1386 }
1387 
1388 static bool
1389 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a,
1390     const struct lacp_peerinfo *b)
1391 {
1392 	if (memcmp(&a->lip_systemid, &b->lip_systemid,
1393 	    sizeof(a->lip_systemid)) != 0) {
1394 		return (false);
1395 	}
1396 
1397 	if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key)) != 0)
1398 		return (false);
1399 
1400 	return (true);
1401 }
1402 
1403 static void
1404 lacp_port_enable(struct lacp_port *lp)
1405 {
1406 	lp->lp_state |= LACP_STATE_AGGREGATION;
1407 }
1408 
1409 static void
1410 lacp_port_disable(struct lacp_port *lp)
1411 {
1412 	lacp_set_mux(lp, LACP_MUX_DETACHED);
1413 
1414 	lp->lp_state &= ~LACP_STATE_AGGREGATION;
1415 	lp->lp_selected = LACP_UNSELECTED;
1416 	lacp_sm_rx_record_default(lp);
1417 	lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION;
1418 	lp->lp_state &= ~LACP_STATE_EXPIRED;
1419 }
1420 
1421 /*
1422  * lacp_select: select an aggregator.  create one if necessary.
1423  */
1424 static void
1425 lacp_select(struct lacp_port *lp)
1426 {
1427 	struct lacp_softc *lsc = lp->lp_lsc;
1428 	struct lacp_aggregator *la;
1429 	char buf[LACP_LAGIDSTR_MAX+1];
1430 
1431 	if (lp->lp_aggregator) {
1432 		return;
1433 	}
1434 
1435 	/* If we haven't heard from our peer, skip this step. */
1436 	if (lp->lp_state & LACP_STATE_DEFAULTED)
1437 		return;
1438 
1439 	KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1440 	    ("timer_wait_while still active"));
1441 
1442 	LACP_DPRINTF((lp, "port lagid=%s\n",
1443 	    lacp_format_lagid(&lp->lp_actor, &lp->lp_partner,
1444 	    buf, sizeof(buf))));
1445 
1446 	TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
1447 		if (lacp_aggregator_is_compatible(la, lp)) {
1448 			break;
1449 		}
1450 	}
1451 
1452 	if (la == NULL) {
1453 		la = lacp_aggregator_get(lsc, lp);
1454 		if (la == NULL) {
1455 			LACP_DPRINTF((lp, "aggregator creation failed\n"));
1456 
1457 			/*
1458 			 * will retry on the next tick.
1459 			 */
1460 
1461 			return;
1462 		}
1463 		lacp_fill_aggregator_id(la, lp);
1464 		LACP_DPRINTF((lp, "aggregator created\n"));
1465 	} else {
1466 		LACP_DPRINTF((lp, "compatible aggregator found\n"));
1467 		if (la->la_refcnt == LACP_MAX_PORTS)
1468 			return;
1469 		lacp_aggregator_addref(lsc, la);
1470 	}
1471 
1472 	LACP_DPRINTF((lp, "aggregator lagid=%s\n",
1473 	    lacp_format_lagid(&la->la_actor, &la->la_partner,
1474 	    buf, sizeof(buf))));
1475 
1476 	lp->lp_aggregator = la;
1477 	lp->lp_selected = LACP_SELECTED;
1478 }
1479 
1480 /*
1481  * lacp_unselect: finish unselect/detach process.
1482  */
1483 
1484 static void
1485 lacp_unselect(struct lacp_port *lp)
1486 {
1487 	struct lacp_softc *lsc = lp->lp_lsc;
1488 	struct lacp_aggregator *la = lp->lp_aggregator;
1489 
1490 	KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1491 	    ("timer_wait_while still active"));
1492 
1493 	if (la == NULL) {
1494 		return;
1495 	}
1496 
1497 	lp->lp_aggregator = NULL;
1498 	lacp_aggregator_delref(lsc, la);
1499 }
1500 
1501 /* mux machine */
1502 
1503 static void
1504 lacp_sm_mux(struct lacp_port *lp)
1505 {
1506 	struct lagg_port *lgp = lp->lp_lagg;
1507 	struct lagg_softc *sc = lgp->lp_softc;
1508 	enum lacp_mux_state new_state;
1509 	boolean_t p_sync =
1510 		    (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0;
1511 	boolean_t p_collecting =
1512 	    (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0;
1513 	enum lacp_selected selected = lp->lp_selected;
1514 	struct lacp_aggregator *la;
1515 
1516 	if (V_lacp_debug > 1)
1517 		lacp_dprintf(lp, "%s: state= 0x%x, selected= 0x%x, "
1518 		    "p_sync= 0x%x, p_collecting= 0x%x\n", __func__,
1519 		    lp->lp_mux_state, selected, p_sync, p_collecting);
1520 
1521 re_eval:
1522 	la = lp->lp_aggregator;
1523 	KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL,
1524 	    ("MUX not detached"));
1525 	new_state = lp->lp_mux_state;
1526 	switch (lp->lp_mux_state) {
1527 	case LACP_MUX_DETACHED:
1528 		if (selected != LACP_UNSELECTED) {
1529 			new_state = LACP_MUX_WAITING;
1530 		}
1531 		break;
1532 	case LACP_MUX_WAITING:
1533 		KASSERT(la->la_pending > 0 ||
1534 		    !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
1535 		    ("timer_wait_while still active"));
1536 		if (selected == LACP_SELECTED && la->la_pending == 0) {
1537 			new_state = LACP_MUX_ATTACHED;
1538 		} else if (selected == LACP_UNSELECTED) {
1539 			new_state = LACP_MUX_DETACHED;
1540 		}
1541 		break;
1542 	case LACP_MUX_ATTACHED:
1543 		if (selected == LACP_SELECTED && p_sync) {
1544 			new_state = LACP_MUX_COLLECTING;
1545 		} else if (selected != LACP_SELECTED) {
1546 			new_state = LACP_MUX_DETACHED;
1547 		}
1548 		break;
1549 	case LACP_MUX_COLLECTING:
1550 		if (selected == LACP_SELECTED && p_sync && p_collecting) {
1551 			new_state = LACP_MUX_DISTRIBUTING;
1552 		} else if (selected != LACP_SELECTED || !p_sync) {
1553 			new_state = LACP_MUX_ATTACHED;
1554 		}
1555 		break;
1556 	case LACP_MUX_DISTRIBUTING:
1557 		if (selected != LACP_SELECTED || !p_sync || !p_collecting) {
1558 			new_state = LACP_MUX_COLLECTING;
1559 			lacp_dprintf(lp, "Interface stopped DISTRIBUTING, possible flapping\n");
1560 			sc->sc_flapping++;
1561 		}
1562 		break;
1563 	default:
1564 		panic("%s: unknown state", __func__);
1565 	}
1566 
1567 	if (lp->lp_mux_state == new_state) {
1568 		return;
1569 	}
1570 
1571 	lacp_set_mux(lp, new_state);
1572 	goto re_eval;
1573 }
1574 
1575 static void
1576 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state)
1577 {
1578 	struct lacp_aggregator *la = lp->lp_aggregator;
1579 
1580 	if (lp->lp_mux_state == new_state) {
1581 		return;
1582 	}
1583 
1584 	switch (new_state) {
1585 	case LACP_MUX_DETACHED:
1586 		lp->lp_state &= ~LACP_STATE_SYNC;
1587 		lacp_disable_distributing(lp);
1588 		lacp_disable_collecting(lp);
1589 		lacp_sm_assert_ntt(lp);
1590 		/* cancel timer */
1591 		if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) {
1592 			KASSERT(la->la_pending > 0,
1593 			    ("timer_wait_while not active"));
1594 			la->la_pending--;
1595 		}
1596 		LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE);
1597 		lacp_unselect(lp);
1598 		break;
1599 	case LACP_MUX_WAITING:
1600 		LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE,
1601 		    LACP_AGGREGATE_WAIT_TIME);
1602 		la->la_pending++;
1603 		break;
1604 	case LACP_MUX_ATTACHED:
1605 		lp->lp_state |= LACP_STATE_SYNC;
1606 		lacp_disable_collecting(lp);
1607 		lacp_sm_assert_ntt(lp);
1608 		break;
1609 	case LACP_MUX_COLLECTING:
1610 		lacp_enable_collecting(lp);
1611 		lacp_disable_distributing(lp);
1612 		lacp_sm_assert_ntt(lp);
1613 		break;
1614 	case LACP_MUX_DISTRIBUTING:
1615 		lacp_enable_distributing(lp);
1616 		break;
1617 	default:
1618 		panic("%s: unknown state", __func__);
1619 	}
1620 
1621 	LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state));
1622 
1623 	lp->lp_mux_state = new_state;
1624 }
1625 
1626 static void
1627 lacp_sm_mux_timer(struct lacp_port *lp)
1628 {
1629 	struct lacp_aggregator *la = lp->lp_aggregator;
1630 	char buf[LACP_LAGIDSTR_MAX+1];
1631 
1632 	KASSERT(la->la_pending > 0, ("no pending event"));
1633 
1634 	LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__,
1635 	    lacp_format_lagid(&la->la_actor, &la->la_partner,
1636 	    buf, sizeof(buf)),
1637 	    la->la_pending, la->la_pending - 1));
1638 
1639 	la->la_pending--;
1640 }
1641 
1642 /* periodic transmit machine */
1643 
1644 static void
1645 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate)
1646 {
1647 	if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state,
1648 	    LACP_STATE_TIMEOUT)) {
1649 		return;
1650 	}
1651 
1652 	LACP_DPRINTF((lp, "partner timeout changed\n"));
1653 
1654 	/*
1655 	 * FAST_PERIODIC -> SLOW_PERIODIC
1656 	 * or
1657 	 * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC
1658 	 *
1659 	 * let lacp_sm_ptx_tx_schedule to update timeout.
1660 	 */
1661 
1662 	LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1663 
1664 	/*
1665 	 * if timeout has been shortened, assert NTT.
1666 	 */
1667 
1668 	if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) {
1669 		lacp_sm_assert_ntt(lp);
1670 	}
1671 }
1672 
1673 static void
1674 lacp_sm_ptx_tx_schedule(struct lacp_port *lp)
1675 {
1676 	int timeout;
1677 
1678 	if (!(lp->lp_state & LACP_STATE_ACTIVITY) &&
1679 	    !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) {
1680 		/*
1681 		 * NO_PERIODIC
1682 		 */
1683 
1684 		LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
1685 		return;
1686 	}
1687 
1688 	if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) {
1689 		return;
1690 	}
1691 
1692 	timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ?
1693 	    LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME;
1694 
1695 	LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout);
1696 }
1697 
1698 static void
1699 lacp_sm_ptx_timer(struct lacp_port *lp)
1700 {
1701 	lacp_sm_assert_ntt(lp);
1702 }
1703 
1704 static void
1705 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du)
1706 {
1707 	int timeout;
1708 
1709 	/*
1710 	 * check LACP_DISABLED first
1711 	 */
1712 
1713 	if (!(lp->lp_state & LACP_STATE_AGGREGATION)) {
1714 		return;
1715 	}
1716 
1717 	/*
1718 	 * check loopback condition.
1719 	 */
1720 
1721 	if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid,
1722 	    &lp->lp_actor.lip_systemid)) {
1723 		return;
1724 	}
1725 
1726 	/*
1727 	 * EXPIRED, DEFAULTED, CURRENT -> CURRENT
1728 	 */
1729 
1730 	microuptime(&lp->lp_last_lacpdu_rx);
1731 	lacp_sm_rx_update_selected(lp, du);
1732 	lacp_sm_rx_update_ntt(lp, du);
1733 	lacp_sm_rx_record_pdu(lp, du);
1734 
1735 	timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ?
1736 	    LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME;
1737 	LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout);
1738 
1739 	lp->lp_state &= ~LACP_STATE_EXPIRED;
1740 
1741 	/*
1742 	 * kick transmit machine without waiting the next tick.
1743 	 */
1744 
1745 	lacp_sm_tx(lp);
1746 }
1747 
1748 static void
1749 lacp_sm_rx_set_expired(struct lacp_port *lp)
1750 {
1751 	lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1752 	lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT;
1753 	LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME);
1754 	lp->lp_state |= LACP_STATE_EXPIRED;
1755 }
1756 
1757 static void
1758 lacp_sm_rx_timer(struct lacp_port *lp)
1759 {
1760 	if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) {
1761 		/* CURRENT -> EXPIRED */
1762 		LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__));
1763 		lacp_sm_rx_set_expired(lp);
1764 	} else {
1765 		/* EXPIRED -> DEFAULTED */
1766 		LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__));
1767 		lacp_sm_rx_update_default_selected(lp);
1768 		lacp_sm_rx_record_default(lp);
1769 		lp->lp_state &= ~LACP_STATE_EXPIRED;
1770 	}
1771 }
1772 
1773 static void
1774 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du)
1775 {
1776 	boolean_t active;
1777 	uint8_t oldpstate;
1778 	char buf[LACP_STATESTR_MAX+1];
1779 
1780 	LACP_TRACE(lp);
1781 
1782 	oldpstate = lp->lp_partner.lip_state;
1783 
1784 	active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY)
1785 	    || ((lp->lp_state & LACP_STATE_ACTIVITY) &&
1786 	    (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY));
1787 
1788 	lp->lp_partner = du->ldu_actor;
1789 	if (active &&
1790 	    ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1791 	    LACP_STATE_AGGREGATION) &&
1792 	    !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner))
1793 	    || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) {
1794 		/*
1795 		 * XXX Maintain legacy behavior of leaving the
1796 		 * LACP_STATE_SYNC bit unchanged from the partner's
1797 		 * advertisement if lsc_strict_mode is false.
1798 		 * TODO: We should re-examine the concept of the "strict mode"
1799 		 * to ensure it makes sense to maintain a non-strict mode.
1800 		 */
1801 		if (lp->lp_lsc->lsc_strict_mode)
1802 			lp->lp_partner.lip_state |= LACP_STATE_SYNC;
1803 	} else {
1804 		lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
1805 	}
1806 
1807 	lp->lp_state &= ~LACP_STATE_DEFAULTED;
1808 
1809 	if (oldpstate != lp->lp_partner.lip_state) {
1810 		LACP_DPRINTF((lp, "old pstate %s\n",
1811 		    lacp_format_state(oldpstate, buf, sizeof(buf))));
1812 		LACP_DPRINTF((lp, "new pstate %s\n",
1813 		    lacp_format_state(lp->lp_partner.lip_state, buf,
1814 		    sizeof(buf))));
1815 	}
1816 
1817 	lacp_sm_ptx_update_timeout(lp, oldpstate);
1818 }
1819 
1820 static void
1821 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du)
1822 {
1823 
1824 	LACP_TRACE(lp);
1825 
1826 	if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) ||
1827 	    !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
1828 	    LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) {
1829 		LACP_DPRINTF((lp, "%s: assert ntt\n", __func__));
1830 		lacp_sm_assert_ntt(lp);
1831 	}
1832 }
1833 
1834 static void
1835 lacp_sm_rx_record_default(struct lacp_port *lp)
1836 {
1837 	uint8_t oldpstate;
1838 
1839 	LACP_TRACE(lp);
1840 
1841 	oldpstate = lp->lp_partner.lip_state;
1842 	if (lp->lp_lsc->lsc_strict_mode)
1843 		lp->lp_partner = lacp_partner_admin_strict;
1844 	else
1845 		lp->lp_partner = lacp_partner_admin_optimistic;
1846 	lp->lp_state |= LACP_STATE_DEFAULTED;
1847 	lacp_sm_ptx_update_timeout(lp, oldpstate);
1848 }
1849 
1850 static void
1851 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp,
1852     const struct lacp_peerinfo *info)
1853 {
1854 
1855 	LACP_TRACE(lp);
1856 
1857 	if (lacp_compare_peerinfo(&lp->lp_partner, info) ||
1858 	    !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state,
1859 	    LACP_STATE_AGGREGATION)) {
1860 		lp->lp_selected = LACP_UNSELECTED;
1861 		/* mux machine will clean up lp->lp_aggregator */
1862 	}
1863 }
1864 
1865 static void
1866 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du)
1867 {
1868 
1869 	LACP_TRACE(lp);
1870 
1871 	lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor);
1872 }
1873 
1874 static void
1875 lacp_sm_rx_update_default_selected(struct lacp_port *lp)
1876 {
1877 
1878 	LACP_TRACE(lp);
1879 
1880 	if (lp->lp_lsc->lsc_strict_mode)
1881 		lacp_sm_rx_update_selected_from_peerinfo(lp,
1882 		    &lacp_partner_admin_strict);
1883 	else
1884 		lacp_sm_rx_update_selected_from_peerinfo(lp,
1885 		    &lacp_partner_admin_optimistic);
1886 }
1887 
1888 /* transmit machine */
1889 
1890 static void
1891 lacp_sm_tx(struct lacp_port *lp)
1892 {
1893 	int error = 0;
1894 
1895 	if (!(lp->lp_state & LACP_STATE_AGGREGATION)
1896 #if 1
1897 	    || (!(lp->lp_state & LACP_STATE_ACTIVITY)
1898 	    && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY))
1899 #endif
1900 	    ) {
1901 		lp->lp_flags &= ~LACP_PORT_NTT;
1902 	}
1903 
1904 	if (!(lp->lp_flags & LACP_PORT_NTT)) {
1905 		return;
1906 	}
1907 
1908 	/* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */
1909 	if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent,
1910 		    (3 / LACP_FAST_PERIODIC_TIME)) == 0) {
1911 		LACP_DPRINTF((lp, "rate limited pdu\n"));
1912 		return;
1913 	}
1914 
1915 	if (((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_tx_test) == 0) {
1916 		error = lacp_xmit_lacpdu(lp);
1917 	} else {
1918 		LACP_TPRINTF((lp, "Dropping TX PDU\n"));
1919 	}
1920 
1921 	if (error == 0) {
1922 		lp->lp_flags &= ~LACP_PORT_NTT;
1923 	} else {
1924 		LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n",
1925 		    error));
1926 	}
1927 }
1928 
1929 static void
1930 lacp_sm_assert_ntt(struct lacp_port *lp)
1931 {
1932 
1933 	lp->lp_flags |= LACP_PORT_NTT;
1934 }
1935 
1936 static void
1937 lacp_run_timers(struct lacp_port *lp)
1938 {
1939 	int i;
1940 	struct timeval time_diff;
1941 
1942 	for (i = 0; i < LACP_NTIMER; i++) {
1943 		KASSERT(lp->lp_timer[i] >= 0,
1944 		    ("invalid timer value %d", lp->lp_timer[i]));
1945 		if (lp->lp_timer[i] == 0) {
1946 			continue;
1947 		} else {
1948 			if (i == LACP_TIMER_CURRENT_WHILE) {
1949 				microuptime(&time_diff);
1950 				timevalsub(&time_diff, &lp->lp_last_lacpdu_rx);
1951 				if (time_diff.tv_sec) {
1952 					/* At least one sec has elapsed since last LACP packet. */
1953 					--lp->lp_timer[i];
1954 				}
1955 			} else {
1956 				--lp->lp_timer[i];
1957 			}
1958 
1959 			if ((lp->lp_timer[i] <= 0) && (lacp_timer_funcs[i])) {
1960 				(*lacp_timer_funcs[i])(lp);
1961 			}
1962 		}
1963 	}
1964 }
1965 
1966 int
1967 lacp_marker_input(struct lacp_port *lp, struct mbuf *m)
1968 {
1969 	struct lacp_softc *lsc = lp->lp_lsc;
1970 	struct lagg_port *lgp = lp->lp_lagg;
1971 	struct lacp_port *lp2;
1972 	struct markerdu *mdu;
1973 	int error = 0;
1974 	int pending = 0;
1975 
1976 	if (m->m_pkthdr.len != sizeof(*mdu)) {
1977 		goto bad;
1978 	}
1979 
1980 	if ((m->m_flags & M_MCAST) == 0) {
1981 		goto bad;
1982 	}
1983 
1984 	if (m->m_len < sizeof(*mdu)) {
1985 		m = m_pullup(m, sizeof(*mdu));
1986 		if (m == NULL) {
1987 			return (ENOMEM);
1988 		}
1989 	}
1990 
1991 	mdu = mtod(m, struct markerdu *);
1992 
1993 	if (memcmp(&mdu->mdu_eh.ether_dhost,
1994 	    &ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
1995 		goto bad;
1996 	}
1997 
1998 	if (mdu->mdu_sph.sph_version != 1) {
1999 		goto bad;
2000 	}
2001 
2002 	switch (mdu->mdu_tlv.tlv_type) {
2003 	case MARKER_TYPE_INFO:
2004 		if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
2005 		    marker_info_tlv_template, TRUE)) {
2006 			goto bad;
2007 		}
2008 		mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE;
2009 		memcpy(&mdu->mdu_eh.ether_dhost,
2010 		    &ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN);
2011 		memcpy(&mdu->mdu_eh.ether_shost,
2012 		    lgp->lp_lladdr, ETHER_ADDR_LEN);
2013 		error = lagg_enqueue(lp->lp_ifp, m);
2014 		break;
2015 
2016 	case MARKER_TYPE_RESPONSE:
2017 		if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
2018 		    marker_response_tlv_template, TRUE)) {
2019 			goto bad;
2020 		}
2021 		LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n",
2022 		    ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system,
2023 		    ":", ntohl(mdu->mdu_info.mi_rq_xid)));
2024 
2025 		/* Verify that it is the last marker we sent out */
2026 		if (memcmp(&mdu->mdu_info, &lp->lp_marker,
2027 		    sizeof(struct lacp_markerinfo)))
2028 			goto bad;
2029 
2030 		LACP_LOCK(lsc);
2031 		lp->lp_flags &= ~LACP_PORT_MARK;
2032 
2033 		if (lsc->lsc_suppress_distributing) {
2034 			/* Check if any ports are waiting for a response */
2035 			LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) {
2036 				if (lp2->lp_flags & LACP_PORT_MARK) {
2037 					pending = 1;
2038 					break;
2039 				}
2040 			}
2041 
2042 			if (pending == 0) {
2043 				/* All interface queues are clear */
2044 				LACP_DPRINTF((NULL, "queue flush complete\n"));
2045 				lsc->lsc_suppress_distributing = FALSE;
2046 			}
2047 		}
2048 		LACP_UNLOCK(lsc);
2049 		m_freem(m);
2050 		break;
2051 
2052 	default:
2053 		goto bad;
2054 	}
2055 
2056 	return (error);
2057 
2058 bad:
2059 	LACP_DPRINTF((lp, "bad marker frame\n"));
2060 	m_freem(m);
2061 	return (EINVAL);
2062 }
2063 
2064 static int
2065 tlv_check(const void *p, size_t size, const struct tlvhdr *tlv,
2066     const struct tlv_template *tmpl, boolean_t check_type)
2067 {
2068 	while (/* CONSTCOND */ 1) {
2069 		if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) {
2070 			return (EINVAL);
2071 		}
2072 		if ((check_type && tlv->tlv_type != tmpl->tmpl_type) ||
2073 		    tlv->tlv_length != tmpl->tmpl_length) {
2074 			return (EINVAL);
2075 		}
2076 		if (tmpl->tmpl_type == 0) {
2077 			break;
2078 		}
2079 		tlv = (const struct tlvhdr *)
2080 		    ((const char *)tlv + tlv->tlv_length);
2081 		tmpl++;
2082 	}
2083 
2084 	return (0);
2085 }
2086 
2087 /* Debugging */
2088 const char *
2089 lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen)
2090 {
2091 	snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X",
2092 	    (int)mac[0],
2093 	    (int)mac[1],
2094 	    (int)mac[2],
2095 	    (int)mac[3],
2096 	    (int)mac[4],
2097 	    (int)mac[5]);
2098 
2099 	return (buf);
2100 }
2101 
2102 const char *
2103 lacp_format_systemid(const struct lacp_systemid *sysid,
2104     char *buf, size_t buflen)
2105 {
2106 	char macbuf[LACP_MACSTR_MAX+1];
2107 
2108 	snprintf(buf, buflen, "%04X,%s",
2109 	    ntohs(sysid->lsi_prio),
2110 	    lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf)));
2111 
2112 	return (buf);
2113 }
2114 
2115 const char *
2116 lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen)
2117 {
2118 	snprintf(buf, buflen, "%04X,%04X",
2119 	    ntohs(portid->lpi_prio),
2120 	    ntohs(portid->lpi_portno));
2121 
2122 	return (buf);
2123 }
2124 
2125 const char *
2126 lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen)
2127 {
2128 	char sysid[LACP_SYSTEMIDSTR_MAX+1];
2129 	char portid[LACP_PORTIDSTR_MAX+1];
2130 
2131 	snprintf(buf, buflen, "(%s,%04X,%s)",
2132 	    lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)),
2133 	    ntohs(peer->lip_key),
2134 	    lacp_format_portid(&peer->lip_portid, portid, sizeof(portid)));
2135 
2136 	return (buf);
2137 }
2138 
2139 const char *
2140 lacp_format_lagid(const struct lacp_peerinfo *a,
2141     const struct lacp_peerinfo *b, char *buf, size_t buflen)
2142 {
2143 	char astr[LACP_PARTNERSTR_MAX+1];
2144 	char bstr[LACP_PARTNERSTR_MAX+1];
2145 
2146 #if 0
2147 	/*
2148 	 * there's a convention to display small numbered peer
2149 	 * in the left.
2150 	 */
2151 
2152 	if (lacp_compare_peerinfo(a, b) > 0) {
2153 		const struct lacp_peerinfo *t;
2154 
2155 		t = a;
2156 		a = b;
2157 		b = t;
2158 	}
2159 #endif
2160 
2161 	snprintf(buf, buflen, "[%s,%s]",
2162 	    lacp_format_partner(a, astr, sizeof(astr)),
2163 	    lacp_format_partner(b, bstr, sizeof(bstr)));
2164 
2165 	return (buf);
2166 }
2167 
2168 const char *
2169 lacp_format_lagid_aggregator(const struct lacp_aggregator *la,
2170     char *buf, size_t buflen)
2171 {
2172 	if (la == NULL) {
2173 		return ("(none)");
2174 	}
2175 
2176 	return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen));
2177 }
2178 
2179 const char *
2180 lacp_format_state(uint8_t state, char *buf, size_t buflen)
2181 {
2182 	snprintf(buf, buflen, "%b", state, LACP_STATE_BITS);
2183 	return (buf);
2184 }
2185 
2186 static void
2187 lacp_dump_lacpdu(const struct lacpdu *du)
2188 {
2189 	char buf[LACP_PARTNERSTR_MAX+1];
2190 	char buf2[LACP_STATESTR_MAX+1];
2191 
2192 	printf("actor=%s\n",
2193 	    lacp_format_partner(&du->ldu_actor, buf, sizeof(buf)));
2194 	printf("actor.state=%s\n",
2195 	    lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2)));
2196 	printf("partner=%s\n",
2197 	    lacp_format_partner(&du->ldu_partner, buf, sizeof(buf)));
2198 	printf("partner.state=%s\n",
2199 	    lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2)));
2200 
2201 	printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay));
2202 }
2203 
2204 static void
2205 lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...)
2206 {
2207 	va_list va;
2208 
2209 	if (lp) {
2210 		printf("%s: ", lp->lp_ifp->if_xname);
2211 	}
2212 
2213 	va_start(va, fmt);
2214 	vprintf(fmt, va);
2215 	va_end(va);
2216 }
2217