1 /* $NetBSD: ieee8023ad_lacp.c,v 1.3 2005/12/11 12:24:54 christos Exp $ */
2
3 /*-
4 * Copyright (c)2005 YAMAMOTO Takashi,
5 * Copyright (c)2008 Andrew Thompson <thompsa@FreeBSD.org>
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD: src/sys/net/ieee8023ad_lacp.c,v 1.15 2008/03/16 19:25:30 thompsa Exp $");
32
33 #include <sys/param.h>
34 #include <sys/callout.h>
35 #include <sys/mbuf.h>
36 #include <sys/systm.h>
37 #include <sys/malloc.h>
38 #include <sys/kernel.h> /* hz */
39 #include <sys/socket.h> /* for net/if.h */
40 #include <sys/sockio.h>
41 #include <machine/stdarg.h>
42 #include <sys/lock.h>
43 #include <sys/rwlock.h>
44
45 #include <net/if.h>
46 #include <net/if_dl.h>
47 #include <net/ethernet.h>
48 #include <net/if_media.h>
49 #include <net/if_types.h>
50
51 #include <net/if_lagg.h>
52 #include <net/ieee8023ad_lacp.h>
53
54 /*
55 * actor system priority and port priority.
56 * XXX should be configurable.
57 */
58
59 #define LACP_SYSTEM_PRIO 0x8000
60 #define LACP_PORT_PRIO 0x8000
61
62 const uint8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] =
63 { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 };
64
65 static const struct tlv_template lacp_info_tlv_template[] = {
66 { LACP_TYPE_ACTORINFO,
67 sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
68 { LACP_TYPE_PARTNERINFO,
69 sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
70 { LACP_TYPE_COLLECTORINFO,
71 sizeof(struct tlvhdr) + sizeof(struct lacp_collectorinfo) },
72 { 0, 0 },
73 };
74
75 static const struct tlv_template marker_info_tlv_template[] = {
76 { MARKER_TYPE_INFO,
77 sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
78 { 0, 0 },
79 };
80
81 static const struct tlv_template marker_response_tlv_template[] = {
82 { MARKER_TYPE_RESPONSE,
83 sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
84 { 0, 0 },
85 };
86
87 typedef void (*lacp_timer_func_t)(struct lacp_port *);
88
89 static void lacp_fill_actorinfo(struct lacp_port *, struct lacp_peerinfo *);
90 static void lacp_fill_markerinfo(struct lacp_port *,
91 struct lacp_markerinfo *);
92
93 static uint64_t lacp_aggregator_bandwidth(struct lacp_aggregator *);
94 static void lacp_suppress_distributing(struct lacp_softc *,
95 struct lacp_aggregator *);
96 static void lacp_transit_expire(void *);
97 static void lacp_update_portmap(struct lacp_softc *);
98 static void lacp_select_active_aggregator(struct lacp_softc *);
99 static uint16_t lacp_compose_key(struct lacp_port *);
100 static int tlv_check(const void *, size_t, const struct tlvhdr *,
101 const struct tlv_template *, boolean_t);
102 static void lacp_tick(void *);
103
104 static void lacp_fill_aggregator_id(struct lacp_aggregator *,
105 const struct lacp_port *);
106 static void lacp_fill_aggregator_id_peer(struct lacp_peerinfo *,
107 const struct lacp_peerinfo *);
108 static int lacp_aggregator_is_compatible(const struct lacp_aggregator *,
109 const struct lacp_port *);
110 static int lacp_peerinfo_is_compatible(const struct lacp_peerinfo *,
111 const struct lacp_peerinfo *);
112
113 static struct lacp_aggregator *lacp_aggregator_get(struct lacp_softc *,
114 struct lacp_port *);
115 static void lacp_aggregator_addref(struct lacp_softc *,
116 struct lacp_aggregator *);
117 static void lacp_aggregator_delref(struct lacp_softc *,
118 struct lacp_aggregator *);
119
120 /* receive machine */
121
122 static int lacp_pdu_input(struct lacp_port *, struct mbuf *);
123 static int lacp_marker_input(struct lacp_port *, struct mbuf *);
124 static void lacp_sm_rx(struct lacp_port *, const struct lacpdu *);
125 static void lacp_sm_rx_timer(struct lacp_port *);
126 static void lacp_sm_rx_set_expired(struct lacp_port *);
127 static void lacp_sm_rx_update_ntt(struct lacp_port *,
128 const struct lacpdu *);
129 static void lacp_sm_rx_record_pdu(struct lacp_port *,
130 const struct lacpdu *);
131 static void lacp_sm_rx_update_selected(struct lacp_port *,
132 const struct lacpdu *);
133 static void lacp_sm_rx_record_default(struct lacp_port *);
134 static void lacp_sm_rx_update_default_selected(struct lacp_port *);
135 static void lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *,
136 const struct lacp_peerinfo *);
137
138 /* mux machine */
139
140 static void lacp_sm_mux(struct lacp_port *);
141 static void lacp_set_mux(struct lacp_port *, enum lacp_mux_state);
142 static void lacp_sm_mux_timer(struct lacp_port *);
143
144 /* periodic transmit machine */
145
146 static void lacp_sm_ptx_update_timeout(struct lacp_port *, uint8_t);
147 static void lacp_sm_ptx_tx_schedule(struct lacp_port *);
148 static void lacp_sm_ptx_timer(struct lacp_port *);
149
150 /* transmit machine */
151
152 static void lacp_sm_tx(struct lacp_port *);
153 static void lacp_sm_assert_ntt(struct lacp_port *);
154
155 static void lacp_run_timers(struct lacp_port *);
156 static int lacp_compare_peerinfo(const struct lacp_peerinfo *,
157 const struct lacp_peerinfo *);
158 static int lacp_compare_systemid(const struct lacp_systemid *,
159 const struct lacp_systemid *);
160 static void lacp_port_enable(struct lacp_port *);
161 static void lacp_port_disable(struct lacp_port *);
162 static void lacp_select(struct lacp_port *);
163 static void lacp_unselect(struct lacp_port *);
164 static void lacp_disable_collecting(struct lacp_port *);
165 static void lacp_enable_collecting(struct lacp_port *);
166 static void lacp_disable_distributing(struct lacp_port *);
167 static void lacp_enable_distributing(struct lacp_port *);
168 static int lacp_xmit_lacpdu(struct lacp_port *);
169 static int lacp_xmit_marker(struct lacp_port *);
170
171 #if defined(LACP_DEBUG)
172 static void lacp_dump_lacpdu(const struct lacpdu *);
173 static const char *lacp_format_partner(const struct lacp_peerinfo *, char *,
174 size_t);
175 static const char *lacp_format_lagid(const struct lacp_peerinfo *,
176 const struct lacp_peerinfo *, char *, size_t);
177 static const char *lacp_format_lagid_aggregator(const struct lacp_aggregator *,
178 char *, size_t);
179 static const char *lacp_format_state(uint8_t, char *, size_t);
180 static const char *lacp_format_mac(const uint8_t *, char *, size_t);
181 static const char *lacp_format_systemid(const struct lacp_systemid *, char *,
182 size_t);
183 static const char *lacp_format_portid(const struct lacp_portid *, char *,
184 size_t);
185 static void lacp_dprintf(const struct lacp_port *, const char *, ...)
186 __attribute__((__format__(__printf__, 2, 3)));
187 #define LACP_DPRINTF(a) lacp_dprintf a
188 #else
189 #define LACP_DPRINTF(a) /* nothing */
190 #endif
191
192 /*
193 * partner administration variables.
194 * XXX should be configurable.
195 */
196
197 static const struct lacp_peerinfo lacp_partner_admin = {
198 .lip_systemid = { .lsi_prio = 0xffff },
199 .lip_portid = { .lpi_prio = 0xffff },
200 #if 1
201 /* optimistic */
202 .lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION |
203 LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING,
204 #else
205 /* pessimistic */
206 .lip_state = 0,
207 #endif
208 };
209
210 static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = {
211 [LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer,
212 [LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer,
213 [LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer,
214 };
215
216 struct mbuf *
217 lacp_input(struct lagg_port *lgp, struct mbuf *m)
218 {
219 struct lacp_port *lp = LACP_PORT(lgp);
220 uint8_t subtype;
221
222 if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) {
223 m_freem(m);
224 return (NULL);
225 }
226
227 m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype);
228 switch (subtype) {
229 case SLOWPROTOCOLS_SUBTYPE_LACP:
230 lacp_pdu_input(lp, m);
231 return (NULL);
232
233 case SLOWPROTOCOLS_SUBTYPE_MARKER:
234 lacp_marker_input(lp, m);
235 return (NULL);
236 }
237
238 /* Not a subtype we are interested in */
239 return (m);
240 }
241
242 /*
243 * lacp_pdu_input: process lacpdu
244 */
245 static int
246 lacp_pdu_input(struct lacp_port *lp, struct mbuf *m)
247 {
248 struct lacp_softc *lsc = lp->lp_lsc;
249 struct lacpdu *du;
250 int error = 0;
251
252 if (m->m_pkthdr.len != sizeof(*du)) {
253 goto bad;
254 }
255
256 if ((m->m_flags & M_MCAST) == 0) {
257 goto bad;
258 }
259
260 if (m->m_len < sizeof(*du)) {
261 m = m_pullup(m, sizeof(*du));
262 if (m == NULL) {
263 return (ENOMEM);
264 }
265 }
266
267 du = mtod(m, struct lacpdu *);
268
269 if (memcmp(&du->ldu_eh.ether_dhost,
270 ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
271 goto bad;
272 }
273
274 /*
275 * ignore the version for compatibility with
276 * the future protocol revisions.
277 */
278 #if 0
279 if (du->ldu_sph.sph_version != 1) {
280 goto bad;
281 }
282 #endif
283
284 /*
285 * ignore tlv types for compatibility with
286 * the future protocol revisions.
287 */
288 if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor,
289 lacp_info_tlv_template, FALSE)) {
290 goto bad;
291 }
292
293 #if defined(LACP_DEBUG)
294 LACP_DPRINTF((lp, "lacpdu receive\n"));
295 lacp_dump_lacpdu(du);
296 #endif /* defined(LACP_DEBUG) */
297
298 LACP_LOCK(lsc);
299 lacp_sm_rx(lp, du);
300 LACP_UNLOCK(lsc);
301
302 m_freem(m);
303 return (error);
304
305 bad:
306 m_freem(m);
307 return (EINVAL);
308 }
309
310 static void
311 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info)
312 {
313 struct lagg_port *lgp = lp->lp_lagg;
314 struct lagg_softc *sc = lgp->lp_softc;
315
316 info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO);
317 memcpy(&info->lip_systemid.lsi_mac,
318 IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
319 info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO);
320 info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index);
321 info->lip_state = lp->lp_state;
322 }
323
324 static void
325 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info)
326 {
327 struct ifnet *ifp = lp->lp_ifp;
328
329 /* Fill in the port index and system id (encoded as the MAC) */
330 info->mi_rq_port = htons(ifp->if_index);
331 memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN);
332 info->mi_rq_xid = htonl(0);
333 }
334
335 static int
336 lacp_xmit_lacpdu(struct lacp_port *lp)
337 {
338 struct lagg_port *lgp = lp->lp_lagg;
339 struct mbuf *m;
340 struct lacpdu *du;
341 int error;
342
343 LACP_LOCK_ASSERT(lp->lp_lsc);
344
345 m = m_gethdr(M_DONTWAIT, MT_DATA);
346 if (m == NULL) {
347 return (ENOMEM);
348 }
349 m->m_len = m->m_pkthdr.len = sizeof(*du);
350
351 du = mtod(m, struct lacpdu *);
352 memset(du, 0, sizeof(*du));
353
354 memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
355 ETHER_ADDR_LEN);
356 memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
357 du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW);
358
359 du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP;
360 du->ldu_sph.sph_version = 1;
361
362 TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor));
363 du->ldu_actor = lp->lp_actor;
364
365 TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO,
366 sizeof(du->ldu_partner));
367 du->ldu_partner = lp->lp_partner;
368
369 TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO,
370 sizeof(du->ldu_collector));
371 du->ldu_collector.lci_maxdelay = 0;
372
373 #if defined(LACP_DEBUG)
374 LACP_DPRINTF((lp, "lacpdu transmit\n"));
375 lacp_dump_lacpdu(du);
376 #endif /* defined(LACP_DEBUG) */
377
378 m->m_flags |= M_MCAST;
379
380 /*
381 * XXX should use higher priority queue.
382 * otherwise network congestion can break aggregation.
383 */
384
385 error = lagg_enqueue(lp->lp_ifp, m);
386 return (error);
387 }
388
389 static int
390 lacp_xmit_marker(struct lacp_port *lp)
391 {
392 struct lagg_port *lgp = lp->lp_lagg;
393 struct mbuf *m;
394 struct markerdu *mdu;
395 int error;
396
397 LACP_LOCK_ASSERT(lp->lp_lsc);
398
399 m = m_gethdr(M_DONTWAIT, MT_DATA);
400 if (m == NULL) {
401 return (ENOMEM);
402 }
403 m->m_len = m->m_pkthdr.len = sizeof(*mdu);
404
405 mdu = mtod(m, struct markerdu *);
406 memset(mdu, 0, sizeof(*mdu));
407
408 memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
409 ETHER_ADDR_LEN);
410 memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
411 mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW);
412
413 mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER;
414 mdu->mdu_sph.sph_version = 1;
415
416 /* Bump the transaction id and copy over the marker info */
417 lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1);
418 TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info));
419 mdu->mdu_info = lp->lp_marker;
420
421 LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n",
422 ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":",
423 ntohl(mdu->mdu_info.mi_rq_xid)));
424
425 m->m_flags |= M_MCAST;
426 error = lagg_enqueue(lp->lp_ifp, m);
427 return (error);
428 }
429
430 void
431 lacp_linkstate(struct lagg_port *lgp)
432 {
433 struct lacp_port *lp = LACP_PORT(lgp);
434 struct lacp_softc *lsc = lp->lp_lsc;
435 struct ifnet *ifp = lgp->lp_ifp;
436 struct ifmediareq ifmr;
437 int error = 0;
438 u_int media;
439 uint8_t old_state;
440 uint16_t old_key;
441
442 bzero((char *)&ifmr, sizeof(ifmr));
443 error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr);
444 if (error != 0)
445 return;
446
447 LACP_LOCK(lsc);
448 media = ifmr.ifm_active;
449 LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, "
450 "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER,
451 (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP));
452 old_state = lp->lp_state;
453 old_key = lp->lp_key;
454
455 lp->lp_media = media;
456 /*
457 * If the port is not an active full duplex Ethernet link then it can
458 * not be aggregated.
459 */
460 if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 ||
461 ifp->if_link_state != LINK_STATE_UP) {
462 lacp_port_disable(lp);
463 } else {
464 lacp_port_enable(lp);
465 }
466 lp->lp_key = lacp_compose_key(lp);
467
468 if (old_state != lp->lp_state || old_key != lp->lp_key) {
469 LACP_DPRINTF((lp, "-> UNSELECTED\n"));
470 lp->lp_selected = LACP_UNSELECTED;
471 }
472 LACP_UNLOCK(lsc);
473 }
474
475 static void
476 lacp_tick(void *arg)
477 {
478 struct lacp_softc *lsc = arg;
479 struct lacp_port *lp;
480
481 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
482 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0)
483 continue;
484
485 lacp_run_timers(lp);
486
487 lacp_select(lp);
488 lacp_sm_mux(lp);
489 lacp_sm_tx(lp);
490 lacp_sm_ptx_tx_schedule(lp);
491 }
492 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
493 }
494
495 int
496 lacp_port_create(struct lagg_port *lgp)
497 {
498 struct lagg_softc *sc = lgp->lp_softc;
499 struct lacp_softc *lsc = LACP_SOFTC(sc);
500 struct lacp_port *lp;
501 struct ifnet *ifp = lgp->lp_ifp;
502 struct sockaddr_dl sdl;
503 struct ifmultiaddr *rifma = NULL;
504 int error;
505
506 boolean_t active = TRUE; /* XXX should be configurable */
507 boolean_t fast = FALSE; /* XXX should be configurable */
508
509 bzero((char *)&sdl, sizeof(sdl));
510 sdl.sdl_len = sizeof(sdl);
511 sdl.sdl_family = AF_LINK;
512 sdl.sdl_index = ifp->if_index;
513 sdl.sdl_type = IFT_ETHER;
514 sdl.sdl_alen = ETHER_ADDR_LEN;
515
516 bcopy(ðermulticastaddr_slowprotocols,
517 LLADDR(&sdl), ETHER_ADDR_LEN);
518 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
519 if (error) {
520 printf("%s: ADDMULTI failed on %s\n", __func__, lgp->lp_ifname);
521 return (error);
522 }
523
524 lp = malloc(sizeof(struct lacp_port),
525 M_DEVBUF, M_NOWAIT|M_ZERO);
526 if (lp == NULL)
527 return (ENOMEM);
528
529 LACP_LOCK(lsc);
530 lgp->lp_psc = (caddr_t)lp;
531 lp->lp_ifp = ifp;
532 lp->lp_lagg = lgp;
533 lp->lp_lsc = lsc;
534 lp->lp_ifma = rifma;
535
536 LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next);
537
538 lacp_fill_actorinfo(lp, &lp->lp_actor);
539 lacp_fill_markerinfo(lp, &lp->lp_marker);
540 lp->lp_state =
541 (active ? LACP_STATE_ACTIVITY : 0) |
542 (fast ? LACP_STATE_TIMEOUT : 0);
543 lp->lp_aggregator = NULL;
544 lacp_sm_rx_set_expired(lp);
545 LACP_UNLOCK(lsc);
546 lacp_linkstate(lgp);
547
548 return (0);
549 }
550
551 void
552 lacp_port_destroy(struct lagg_port *lgp)
553 {
554 struct lacp_port *lp = LACP_PORT(lgp);
555 struct lacp_softc *lsc = lp->lp_lsc;
556 int i;
557
558 LACP_LOCK(lsc);
559 for (i = 0; i < LACP_NTIMER; i++) {
560 LACP_TIMER_DISARM(lp, i);
561 }
562
563 lacp_disable_collecting(lp);
564 lacp_disable_distributing(lp);
565 lacp_unselect(lp);
566
567 /* The address may have already been removed by if_purgemaddrs() */
568 if (!lgp->lp_detaching)
569 if_delmulti_ifma(lp->lp_ifma);
570
571 LIST_REMOVE(lp, lp_next);
572 LACP_UNLOCK(lsc);
573 free(lp, M_DEVBUF);
574 }
575
576 void
577 lacp_req(struct lagg_softc *sc, caddr_t data)
578 {
579 struct lacp_opreq *req = (struct lacp_opreq *)data;
580 struct lacp_softc *lsc = LACP_SOFTC(sc);
581 struct lacp_aggregator *la = lsc->lsc_active_aggregator;
582
583 LACP_LOCK(lsc);
584 bzero(req, sizeof(struct lacp_opreq));
585 if (la != NULL) {
586 req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio);
587 memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac,
588 ETHER_ADDR_LEN);
589 req->actor_key = ntohs(la->la_actor.lip_key);
590 req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio);
591 req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno);
592 req->actor_state = la->la_actor.lip_state;
593
594 req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio);
595 memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac,
596 ETHER_ADDR_LEN);
597 req->partner_key = ntohs(la->la_partner.lip_key);
598 req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio);
599 req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno);
600 req->partner_state = la->la_partner.lip_state;
601 }
602 LACP_UNLOCK(lsc);
603 }
604
605 void
606 lacp_portreq(struct lagg_port *lgp, caddr_t data)
607 {
608 struct lacp_opreq *req = (struct lacp_opreq *)data;
609 struct lacp_port *lp = LACP_PORT(lgp);
610 struct lacp_softc *lsc = lp->lp_lsc;
611
612 LACP_LOCK(lsc);
613 req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio);
614 memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac,
615 ETHER_ADDR_LEN);
616 req->actor_key = ntohs(lp->lp_actor.lip_key);
617 req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio);
618 req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno);
619 req->actor_state = lp->lp_actor.lip_state;
620
621 req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio);
622 memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac,
623 ETHER_ADDR_LEN);
624 req->partner_key = ntohs(lp->lp_partner.lip_key);
625 req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio);
626 req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno);
627 req->partner_state = lp->lp_partner.lip_state;
628 LACP_UNLOCK(lsc);
629 }
630
631 static void
632 lacp_disable_collecting(struct lacp_port *lp)
633 {
634 LACP_DPRINTF((lp, "collecting disabled\n"));
635 lp->lp_state &= ~LACP_STATE_COLLECTING;
636 }
637
638 static void
639 lacp_enable_collecting(struct lacp_port *lp)
640 {
641 LACP_DPRINTF((lp, "collecting enabled\n"));
642 lp->lp_state |= LACP_STATE_COLLECTING;
643 }
644
645 static void
646 lacp_disable_distributing(struct lacp_port *lp)
647 {
648 struct lacp_aggregator *la = lp->lp_aggregator;
649 struct lacp_softc *lsc = lp->lp_lsc;
650 #if defined(LACP_DEBUG)
651 char buf[LACP_LAGIDSTR_MAX+1];
652 #endif /* defined(LACP_DEBUG) */
653
654 LACP_LOCK_ASSERT(lsc);
655
656 if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) {
657 return;
658 }
659
660 KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports"));
661 KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports));
662 KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid"));
663
664 LACP_DPRINTF((lp, "disable distributing on aggregator %s, "
665 "nports %d -> %d\n",
666 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
667 la->la_nports, la->la_nports - 1));
668
669 TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q);
670 la->la_nports--;
671
672 if (lsc->lsc_active_aggregator == la) {
673 lacp_suppress_distributing(lsc, la);
674 lacp_select_active_aggregator(lsc);
675 /* regenerate the port map, the active aggregator has changed */
676 lacp_update_portmap(lsc);
677 }
678
679 lp->lp_state &= ~LACP_STATE_DISTRIBUTING;
680 }
681
682 static void
683 lacp_enable_distributing(struct lacp_port *lp)
684 {
685 struct lacp_aggregator *la = lp->lp_aggregator;
686 struct lacp_softc *lsc = lp->lp_lsc;
687 #if defined(LACP_DEBUG)
688 char buf[LACP_LAGIDSTR_MAX+1];
689 #endif /* defined(LACP_DEBUG) */
690
691 LACP_LOCK_ASSERT(lsc);
692
693 if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) {
694 return;
695 }
696
697 LACP_DPRINTF((lp, "enable distributing on aggregator %s, "
698 "nports %d -> %d\n",
699 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
700 la->la_nports, la->la_nports + 1));
701
702 KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid"));
703 TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q);
704 la->la_nports++;
705
706 lp->lp_state |= LACP_STATE_DISTRIBUTING;
707
708 if (lsc->lsc_active_aggregator == la) {
709 lacp_suppress_distributing(lsc, la);
710 lacp_update_portmap(lsc);
711 } else
712 /* try to become the active aggregator */
713 lacp_select_active_aggregator(lsc);
714 }
715
716 static void
717 lacp_transit_expire(void *vp)
718 {
719 struct lacp_softc *lsc = vp;
720
721 LACP_LOCK_ASSERT(lsc);
722
723 LACP_DPRINTF((NULL, "%s\n", __func__));
724 lsc->lsc_suppress_distributing = FALSE;
725 }
726
727 int
728 lacp_attach(struct lagg_softc *sc)
729 {
730 struct lacp_softc *lsc;
731
732 lsc = malloc(sizeof(struct lacp_softc),
733 M_DEVBUF, M_NOWAIT|M_ZERO);
734 if (lsc == NULL)
735 return (ENOMEM);
736
737 sc->sc_psc = (caddr_t)lsc;
738 lsc->lsc_softc = sc;
739
740 lsc->lsc_hashkey = arc4random();
741 lsc->lsc_active_aggregator = NULL;
742 LACP_LOCK_INIT(lsc);
743 TAILQ_INIT(&lsc->lsc_aggregators);
744 LIST_INIT(&lsc->lsc_ports);
745
746 callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0);
747 callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0);
748
749 /* if the lagg is already up then do the same */
750 if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING)
751 lacp_init(sc);
752
753 return (0);
754 }
755
756 int
757 lacp_detach(struct lagg_softc *sc)
758 {
759 struct lacp_softc *lsc = LACP_SOFTC(sc);
760
761 KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators),
762 ("aggregators still active"));
763 KASSERT(lsc->lsc_active_aggregator == NULL,
764 ("aggregator still attached"));
765
766 sc->sc_psc = NULL;
767 callout_drain(&lsc->lsc_transit_callout);
768 callout_drain(&lsc->lsc_callout);
769
770 LACP_LOCK_DESTROY(lsc);
771 free(lsc, M_DEVBUF);
772 return (0);
773 }
774
775 void
776 lacp_init(struct lagg_softc *sc)
777 {
778 struct lacp_softc *lsc = LACP_SOFTC(sc);
779
780 LACP_LOCK(lsc);
781 callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
782 LACP_UNLOCK(lsc);
783 }
784
785 void
786 lacp_stop(struct lagg_softc *sc)
787 {
788 struct lacp_softc *lsc = LACP_SOFTC(sc);
789
790 LACP_LOCK(lsc);
791 callout_stop(&lsc->lsc_transit_callout);
792 callout_stop(&lsc->lsc_callout);
793 LACP_UNLOCK(lsc);
794 }
795
796 struct lagg_port *
797 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
798 {
799 struct lacp_softc *lsc = LACP_SOFTC(sc);
800 struct lacp_portmap *pm;
801 struct lacp_port *lp;
802 uint32_t hash;
803
804 if (__predict_false(lsc->lsc_suppress_distributing)) {
805 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__));
806 return (NULL);
807 }
808
809 pm = &lsc->lsc_pmap[lsc->lsc_activemap];
810 if (pm->pm_count == 0) {
811 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__));
812 return (NULL);
813 }
814
815 hash = lagg_hashmbuf(m, lsc->lsc_hashkey);
816 hash %= pm->pm_count;
817 lp = pm->pm_map[hash];
818
819 KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0,
820 ("aggregated port is not distributing"));
821
822 return (lp->lp_lagg);
823 }
824 /*
825 * lacp_suppress_distributing: drop transmit packets for a while
826 * to preserve packet ordering.
827 */
828
829 static void
830 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la)
831 {
832 struct lacp_port *lp;
833
834 if (lsc->lsc_active_aggregator != la) {
835 return;
836 }
837
838 LACP_DPRINTF((NULL, "%s\n", __func__));
839 lsc->lsc_suppress_distributing = TRUE;
840
841 /* send a marker frame down each port to verify the queues are empty */
842 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
843 lp->lp_flags |= LACP_PORT_MARK;
844 lacp_xmit_marker(lp);
845 }
846
847 /* set a timeout for the marker frames */
848 callout_reset(&lsc->lsc_transit_callout,
849 LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc);
850 }
851
852 static int
853 lacp_compare_peerinfo(const struct lacp_peerinfo *a,
854 const struct lacp_peerinfo *b)
855 {
856 return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state)));
857 }
858
859 static int
860 lacp_compare_systemid(const struct lacp_systemid *a,
861 const struct lacp_systemid *b)
862 {
863 return (memcmp(a, b, sizeof(*a)));
864 }
865
866 #if 0 /* unused */
867 static int
868 lacp_compare_portid(const struct lacp_portid *a,
869 const struct lacp_portid *b)
870 {
871 return (memcmp(a, b, sizeof(*a)));
872 }
873 #endif
874
875 static uint64_t
876 lacp_aggregator_bandwidth(struct lacp_aggregator *la)
877 {
878 struct lacp_port *lp;
879 uint64_t speed;
880
881 lp = TAILQ_FIRST(&la->la_ports);
882 if (lp == NULL) {
883 return (0);
884 }
885
886 speed = ifmedia_baudrate(lp->lp_media);
887 speed *= la->la_nports;
888 if (speed == 0) {
889 LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n",
890 lp->lp_media, la->la_nports));
891 }
892
893 return (speed);
894 }
895
896 /*
897 * lacp_select_active_aggregator: select an aggregator to be used to transmit
898 * packets from lagg(4) interface.
899 */
900
901 static void
902 lacp_select_active_aggregator(struct lacp_softc *lsc)
903 {
904 struct lacp_aggregator *la;
905 struct lacp_aggregator *best_la = NULL;
906 uint64_t best_speed = 0;
907 #if defined(LACP_DEBUG)
908 char buf[LACP_LAGIDSTR_MAX+1];
909 #endif /* defined(LACP_DEBUG) */
910
911 LACP_DPRINTF((NULL, "%s:\n", __func__));
912
913 TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
914 uint64_t speed;
915
916 if (la->la_nports == 0) {
917 continue;
918 }
919
920 speed = lacp_aggregator_bandwidth(la);
921 LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n",
922 lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
923 speed, la->la_nports));
924
925 /* This aggregator is chosen if
926 * the partner has a better system priority
927 * or, the total aggregated speed is higher
928 * or, it is already the chosen aggregator
929 */
930 if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) <
931 LACP_SYS_PRI(best_la->la_partner)) ||
932 speed > best_speed ||
933 (speed == best_speed &&
934 la == lsc->lsc_active_aggregator)) {
935 best_la = la;
936 best_speed = speed;
937 }
938 }
939
940 KASSERT(best_la == NULL || best_la->la_nports > 0,
941 ("invalid aggregator refcnt"));
942 KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports),
943 ("invalid aggregator list"));
944
945 #if defined(LACP_DEBUG)
946 if (lsc->lsc_active_aggregator != best_la) {
947 LACP_DPRINTF((NULL, "active aggregator changed\n"));
948 LACP_DPRINTF((NULL, "old %s\n",
949 lacp_format_lagid_aggregator(lsc->lsc_active_aggregator,
950 buf, sizeof(buf))));
951 } else {
952 LACP_DPRINTF((NULL, "active aggregator not changed\n"));
953 }
954 LACP_DPRINTF((NULL, "new %s\n",
955 lacp_format_lagid_aggregator(best_la, buf, sizeof(buf))));
956 #endif /* defined(LACP_DEBUG) */
957
958 if (lsc->lsc_active_aggregator != best_la) {
959 lsc->lsc_active_aggregator = best_la;
960 lacp_update_portmap(lsc);
961 if (best_la) {
962 lacp_suppress_distributing(lsc, best_la);
963 }
964 }
965 }
966
967 /*
968 * Updated the inactive portmap array with the new list of ports and
969 * make it live.
970 */
971 |