xref: /freebsd/sys/net/if_lagg.c (revision 409a390c3341fb4f162cd7de1fd595a323ebbfd8)
1 /*	$OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $	*/
2 
3 /*
4  * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org>
5  * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org>
6  *
7  * Permission to use, copy, modify, and distribute this software for any
8  * purpose with or without fee is hereby granted, provided that the above
9  * copyright notice and this permission notice appear in all copies.
10  *
11  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18  */
19 
20 #include <sys/cdefs.h>
21 __FBSDID("$FreeBSD$");
22 
23 #include "opt_inet.h"
24 #include "opt_inet6.h"
25 
26 #include <sys/param.h>
27 #include <sys/kernel.h>
28 #include <sys/malloc.h>
29 #include <sys/mbuf.h>
30 #include <sys/queue.h>
31 #include <sys/socket.h>
32 #include <sys/sockio.h>
33 #include <sys/sysctl.h>
34 #include <sys/module.h>
35 #include <sys/priv.h>
36 #include <sys/systm.h>
37 #include <sys/proc.h>
38 #include <sys/hash.h>
39 #include <sys/lock.h>
40 #include <sys/rwlock.h>
41 #include <sys/taskqueue.h>
42 
43 #include <net/ethernet.h>
44 #include <net/if.h>
45 #include <net/if_clone.h>
46 #include <net/if_arp.h>
47 #include <net/if_dl.h>
48 #include <net/if_llc.h>
49 #include <net/if_media.h>
50 #include <net/if_types.h>
51 #include <net/if_var.h>
52 #include <net/bpf.h>
53 
54 #ifdef INET
55 #include <netinet/in.h>
56 #include <netinet/in_systm.h>
57 #include <netinet/if_ether.h>
58 #include <netinet/ip.h>
59 #endif
60 
61 #ifdef INET6
62 #include <netinet/ip6.h>
63 #endif
64 
65 #include <net/if_vlan_var.h>
66 #include <net/if_lagg.h>
67 #include <net/ieee8023ad_lacp.h>
68 
69 /* Special flags we should propagate to the lagg ports. */
70 static struct {
71 	int flag;
72 	int (*func)(struct ifnet *, int);
73 } lagg_pflags[] = {
74 	{IFF_PROMISC, ifpromisc},
75 	{IFF_ALLMULTI, if_allmulti},
76 	{0, NULL}
77 };
78 
79 SLIST_HEAD(__trhead, lagg_softc) lagg_list;	/* list of laggs */
80 static struct mtx	lagg_list_mtx;
81 eventhandler_tag	lagg_detach_cookie = NULL;
82 
83 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
84 static void	lagg_clone_destroy(struct ifnet *);
85 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
86 static void	lagg_capabilities(struct lagg_softc *);
87 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
88 static void	lagg_port_setlladdr(void *, int);
89 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
90 static int	lagg_port_destroy(struct lagg_port *, int);
91 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
92 static void	lagg_linkstate(struct lagg_softc *);
93 static void	lagg_port_state(struct ifnet *, int);
94 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
95 static int	lagg_port_output(struct ifnet *, struct mbuf *,
96 		    struct sockaddr *, struct route *);
97 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
98 #ifdef LAGG_PORT_STACKING
99 static int	lagg_port_checkstacking(struct lagg_softc *);
100 #endif
101 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
102 static void	lagg_init(void *);
103 static void	lagg_stop(struct lagg_softc *);
104 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
105 static int	lagg_ether_setmulti(struct lagg_softc *);
106 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
107 static	int	lagg_setflag(struct lagg_port *, int, int,
108 		    int (*func)(struct ifnet *, int));
109 static	int	lagg_setflags(struct lagg_port *, int status);
110 static void	lagg_start(struct ifnet *);
111 static int	lagg_media_change(struct ifnet *);
112 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
113 static struct lagg_port *lagg_link_active(struct lagg_softc *,
114 	    struct lagg_port *);
115 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
116 
117 IFC_SIMPLE_DECLARE(lagg, 0);
118 
119 /* Simple round robin */
120 static int	lagg_rr_attach(struct lagg_softc *);
121 static int	lagg_rr_detach(struct lagg_softc *);
122 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
123 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
124 		    struct mbuf *);
125 
126 /* Active failover */
127 static int	lagg_fail_attach(struct lagg_softc *);
128 static int	lagg_fail_detach(struct lagg_softc *);
129 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
130 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
131 		    struct mbuf *);
132 
133 /* Loadbalancing */
134 static int	lagg_lb_attach(struct lagg_softc *);
135 static int	lagg_lb_detach(struct lagg_softc *);
136 static int	lagg_lb_port_create(struct lagg_port *);
137 static void	lagg_lb_port_destroy(struct lagg_port *);
138 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
139 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
140 		    struct mbuf *);
141 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
142 
143 /* 802.3ad LACP */
144 static int	lagg_lacp_attach(struct lagg_softc *);
145 static int	lagg_lacp_detach(struct lagg_softc *);
146 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
147 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
148 		    struct mbuf *);
149 static void	lagg_lacp_lladdr(struct lagg_softc *);
150 
151 /* lagg protocol table */
152 static const struct {
153 	int			ti_proto;
154 	int			(*ti_attach)(struct lagg_softc *);
155 } lagg_protos[] = {
156 	{ LAGG_PROTO_ROUNDROBIN,	lagg_rr_attach },
157 	{ LAGG_PROTO_FAILOVER,		lagg_fail_attach },
158 	{ LAGG_PROTO_LOADBALANCE,	lagg_lb_attach },
159 	{ LAGG_PROTO_ETHERCHANNEL,	lagg_lb_attach },
160 	{ LAGG_PROTO_LACP,		lagg_lacp_attach },
161 	{ LAGG_PROTO_NONE,		NULL }
162 };
163 
164 static int
165 lagg_modevent(module_t mod, int type, void *data)
166 {
167 
168 	switch (type) {
169 	case MOD_LOAD:
170 		mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
171 		SLIST_INIT(&lagg_list);
172 		if_clone_attach(&lagg_cloner);
173 		lagg_input_p = lagg_input;
174 		lagg_linkstate_p = lagg_port_state;
175 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
176 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
177 		    EVENTHANDLER_PRI_ANY);
178 		break;
179 	case MOD_UNLOAD:
180 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
181 		    lagg_detach_cookie);
182 		if_clone_detach(&lagg_cloner);
183 		lagg_input_p = NULL;
184 		lagg_linkstate_p = NULL;
185 		mtx_destroy(&lagg_list_mtx);
186 		break;
187 	default:
188 		return (EOPNOTSUPP);
189 	}
190 	return (0);
191 }
192 
193 static moduledata_t lagg_mod = {
194 	"if_lagg",
195 	lagg_modevent,
196 	0
197 };
198 
199 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
200 
201 static int
202 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
203 {
204 	struct lagg_softc *sc;
205 	struct ifnet *ifp;
206 	int i, error = 0;
207 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
208 
209 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
210 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
211 	if (ifp == NULL) {
212 		free(sc, M_DEVBUF);
213 		return (ENOSPC);
214 	}
215 
216 	sc->sc_proto = LAGG_PROTO_NONE;
217 	for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
218 		if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
219 			sc->sc_proto = lagg_protos[i].ti_proto;
220 			if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
221 				if_free_type(ifp, IFT_ETHER);
222 				free(sc, M_DEVBUF);
223 				return (error);
224 			}
225 			break;
226 		}
227 	}
228 	LAGG_LOCK_INIT(sc);
229 	SLIST_INIT(&sc->sc_ports);
230 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
231 
232 	/* Initialise pseudo media types */
233 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
234 	    lagg_media_status);
235 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
236 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
237 
238 	if_initname(ifp, ifc->ifc_name, unit);
239 	ifp->if_type = IFT_ETHER;
240 	ifp->if_softc = sc;
241 	ifp->if_start = lagg_start;
242 	ifp->if_init = lagg_init;
243 	ifp->if_ioctl = lagg_ioctl;
244 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
245 
246 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
247 	ifp->if_snd.ifq_drv_maxlen = ifqmaxlen;
248 	IFQ_SET_READY(&ifp->if_snd);
249 
250 	/*
251 	 * Attach as an ordinary ethernet device, childs will be attached
252 	 * as special device IFT_IEEE8023ADLAG.
253 	 */
254 	ether_ifattach(ifp, eaddr);
255 
256 	/* Insert into the global list of laggs */
257 	mtx_lock(&lagg_list_mtx);
258 	SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
259 	mtx_unlock(&lagg_list_mtx);
260 
261 	return (0);
262 }
263 
264 static void
265 lagg_clone_destroy(struct ifnet *ifp)
266 {
267 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
268 	struct lagg_port *lp;
269 
270 	LAGG_WLOCK(sc);
271 
272 	lagg_stop(sc);
273 	ifp->if_flags &= ~IFF_UP;
274 
275 	/* Shutdown and remove lagg ports */
276 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
277 		lagg_port_destroy(lp, 1);
278 	/* Unhook the aggregation protocol */
279 	(*sc->sc_detach)(sc);
280 
281 	LAGG_WUNLOCK(sc);
282 
283 	ifmedia_removeall(&sc->sc_media);
284 	ether_ifdetach(ifp);
285 	if_free_type(ifp, IFT_ETHER);
286 
287 	mtx_lock(&lagg_list_mtx);
288 	SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
289 	mtx_unlock(&lagg_list_mtx);
290 
291 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
292 	LAGG_LOCK_DESTROY(sc);
293 	free(sc, M_DEVBUF);
294 }
295 
296 static void
297 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
298 {
299 	struct ifnet *ifp = sc->sc_ifp;
300 
301 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
302 		return;
303 
304 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
305 	/* Let the protocol know the MAC has changed */
306 	if (sc->sc_lladdr != NULL)
307 		(*sc->sc_lladdr)(sc);
308 }
309 
310 static void
311 lagg_capabilities(struct lagg_softc *sc)
312 {
313 	struct lagg_port *lp;
314 	int cap = ~0, ena = ~0;
315 	u_long hwa = ~0UL;
316 
317 	LAGG_WLOCK_ASSERT(sc);
318 
319 	/* Get capabilities from the lagg ports */
320 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
321 		cap &= lp->lp_ifp->if_capabilities;
322 		ena &= lp->lp_ifp->if_capenable;
323 		hwa &= lp->lp_ifp->if_hwassist;
324 	}
325 	cap = (cap == ~0 ? 0 : cap);
326 	ena = (ena == ~0 ? 0 : ena);
327 	hwa = (hwa == ~0 ? 0 : hwa);
328 
329 	if (sc->sc_ifp->if_capabilities != cap ||
330 	    sc->sc_ifp->if_capenable != ena ||
331 	    sc->sc_ifp->if_hwassist != hwa) {
332 		sc->sc_ifp->if_capabilities = cap;
333 		sc->sc_ifp->if_capenable = ena;
334 		sc->sc_ifp->if_hwassist = hwa;
335 		getmicrotime(&sc->sc_ifp->if_lastchange);
336 
337 		if (sc->sc_ifflags & IFF_DEBUG)
338 			if_printf(sc->sc_ifp,
339 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
340 	}
341 }
342 
343 static void
344 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
345 {
346 	struct lagg_softc *sc = lp->lp_softc;
347 	struct ifnet *ifp = lp->lp_ifp;
348 	struct lagg_llq *llq;
349 	int pending = 0;
350 
351 	LAGG_WLOCK_ASSERT(sc);
352 
353 	if (lp->lp_detaching ||
354 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
355 		return;
356 
357 	/* Check to make sure its not already queued to be changed */
358 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
359 		if (llq->llq_ifp == ifp) {
360 			pending = 1;
361 			break;
362 		}
363 	}
364 
365 	if (!pending) {
366 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
367 		if (llq == NULL)	/* XXX what to do */
368 			return;
369 	}
370 
371 	/* Update the lladdr even if pending, it may have changed */
372 	llq->llq_ifp = ifp;
373 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
374 
375 	if (!pending)
376 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
377 
378 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
379 }
380 
381 /*
382  * Set the interface MAC address from a taskqueue to avoid a LOR.
383  */
384 static void
385 lagg_port_setlladdr(void *arg, int pending)
386 {
387 	struct lagg_softc *sc = (struct lagg_softc *)arg;
388 	struct lagg_llq *llq, *head;
389 	struct ifnet *ifp;
390 	int error;
391 
392 	/* Grab a local reference of the queue and remove it from the softc */
393 	LAGG_WLOCK(sc);
394 	head = SLIST_FIRST(&sc->sc_llq_head);
395 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
396 	LAGG_WUNLOCK(sc);
397 
398 	/*
399 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
400 	 * unlocked as we have the only reference to it.
401 	 */
402 	for (llq = head; llq != NULL; llq = head) {
403 		ifp = llq->llq_ifp;
404 
405 		/* Set the link layer address */
406 		error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
407 		if (error)
408 			printf("%s: setlladdr failed on %s\n", __func__,
409 			    ifp->if_xname);
410 
411 		head = SLIST_NEXT(llq, llq_entries);
412 		free(llq, M_DEVBUF);
413 	}
414 }
415 
416 static int
417 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
418 {
419 	struct lagg_softc *sc_ptr;
420 	struct lagg_port *lp;
421 	int error = 0;
422 
423 	LAGG_WLOCK_ASSERT(sc);
424 
425 	/* Limit the maximal number of lagg ports */
426 	if (sc->sc_count >= LAGG_MAX_PORTS)
427 		return (ENOSPC);
428 
429 	/* New lagg port has to be in an idle state */
430 	if (ifp->if_drv_flags & IFF_DRV_OACTIVE)
431 		return (EBUSY);
432 
433 	/* Check if port has already been associated to a lagg */
434 	if (ifp->if_lagg != NULL)
435 		return (EBUSY);
436 
437 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
438 	if (ifp->if_type != IFT_ETHER)
439 		return (EPROTONOSUPPORT);
440 
441 	/* Allow the first Ethernet member to define the MTU */
442 	if (SLIST_EMPTY(&sc->sc_ports))
443 		sc->sc_ifp->if_mtu = ifp->if_mtu;
444 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
445 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
446 		    ifp->if_xname);
447 		return (EINVAL);
448 	}
449 
450 	if ((lp = malloc(sizeof(struct lagg_port),
451 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
452 		return (ENOMEM);
453 
454 	/* Check if port is a stacked lagg */
455 	mtx_lock(&lagg_list_mtx);
456 	SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
457 		if (ifp == sc_ptr->sc_ifp) {
458 			mtx_unlock(&lagg_list_mtx);
459 			free(lp, M_DEVBUF);
460 			return (EINVAL);
461 			/* XXX disable stacking for the moment, its untested */
462 #ifdef LAGG_PORT_STACKING
463 			lp->lp_flags |= LAGG_PORT_STACK;
464 			if (lagg_port_checkstacking(sc_ptr) >=
465 			    LAGG_MAX_STACKING) {
466 				mtx_unlock(&lagg_list_mtx);
467 				free(lp, M_DEVBUF);
468 				return (E2BIG);
469 			}
470 #endif
471 		}
472 	}
473 	mtx_unlock(&lagg_list_mtx);
474 
475 	/* Change the interface type */
476 	lp->lp_iftype = ifp->if_type;
477 	ifp->if_type = IFT_IEEE8023ADLAG;
478 	ifp->if_lagg = lp;
479 	lp->lp_ioctl = ifp->if_ioctl;
480 	ifp->if_ioctl = lagg_port_ioctl;
481 	lp->lp_output = ifp->if_output;
482 	ifp->if_output = lagg_port_output;
483 
484 	lp->lp_ifp = ifp;
485 	lp->lp_softc = sc;
486 
487 	/* Save port link layer address */
488 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
489 
490 	if (SLIST_EMPTY(&sc->sc_ports)) {
491 		sc->sc_primary = lp;
492 		lagg_lladdr(sc, IF_LLADDR(ifp));
493 	} else {
494 		/* Update link layer address for this port */
495 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
496 	}
497 
498 	/* Insert into the list of ports */
499 	SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
500 	sc->sc_count++;
501 
502 	/* Update lagg capabilities */
503 	lagg_capabilities(sc);
504 	lagg_linkstate(sc);
505 
506 	/* Add multicast addresses and interface flags to this port */
507 	lagg_ether_cmdmulti(lp, 1);
508 	lagg_setflags(lp, 1);
509 
510 	if (sc->sc_port_create != NULL)
511 		error = (*sc->sc_port_create)(lp);
512 	if (error) {
513 		/* remove the port again, without calling sc_port_destroy */
514 		lagg_port_destroy(lp, 0);
515 		return (error);
516 	}
517 
518 	return (error);
519 }
520 
521 #ifdef LAGG_PORT_STACKING
522 static int
523 lagg_port_checkstacking(struct lagg_softc *sc)
524 {
525 	struct lagg_softc *sc_ptr;
526 	struct lagg_port *lp;
527 	int m = 0;
528 
529 	LAGG_WLOCK_ASSERT(sc);
530 
531 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
532 		if (lp->lp_flags & LAGG_PORT_STACK) {
533 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
534 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
535 		}
536 	}
537 
538 	return (m + 1);
539 }
540 #endif
541 
542 static int
543 lagg_port_destroy(struct lagg_port *lp, int runpd)
544 {
545 	struct lagg_softc *sc = lp->lp_softc;
546 	struct lagg_port *lp_ptr;
547 	struct lagg_llq *llq;
548 	struct ifnet *ifp = lp->lp_ifp;
549 
550 	LAGG_WLOCK_ASSERT(sc);
551 
552 	if (runpd && sc->sc_port_destroy != NULL)
553 		(*sc->sc_port_destroy)(lp);
554 
555 	/*
556 	 * Remove multicast addresses and interface flags from this port and
557 	 * reset the MAC address, skip if the interface is being detached.
558 	 */
559 	if (!lp->lp_detaching) {
560 		lagg_ether_cmdmulti(lp, 0);
561 		lagg_setflags(lp, 0);
562 		lagg_port_lladdr(lp, lp->lp_lladdr);
563 	}
564 
565 	/* Restore interface */
566 	ifp->if_type = lp->lp_iftype;
567 	ifp->if_ioctl = lp->lp_ioctl;
568 	ifp->if_output = lp->lp_output;
569 	ifp->if_lagg = NULL;
570 
571 	/* Finally, remove the port from the lagg */
572 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
573 	sc->sc_count--;
574 
575 	/* Update the primary interface */
576 	if (lp == sc->sc_primary) {
577 		uint8_t lladdr[ETHER_ADDR_LEN];
578 
579 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
580 			bzero(&lladdr, ETHER_ADDR_LEN);
581 		} else {
582 			bcopy(lp_ptr->lp_lladdr,
583 			    lladdr, ETHER_ADDR_LEN);
584 		}
585 		lagg_lladdr(sc, lladdr);
586 		sc->sc_primary = lp_ptr;
587 
588 		/* Update link layer address for each port */
589 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
590 			lagg_port_lladdr(lp_ptr, lladdr);
591 	}
592 
593 	/* Remove any pending lladdr changes from the queue */
594 	if (lp->lp_detaching) {
595 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
596 			if (llq->llq_ifp == ifp) {
597 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
598 				    llq_entries);
599 				free(llq, M_DEVBUF);
600 				break;	/* Only appears once */
601 			}
602 		}
603 	}
604 
605 	if (lp->lp_ifflags)
606 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
607 
608 	free(lp, M_DEVBUF);
609 
610 	/* Update lagg capabilities */
611 	lagg_capabilities(sc);
612 	lagg_linkstate(sc);
613 
614 	return (0);
615 }
616 
617 static int
618 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
619 {
620 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
621 	struct lagg_softc *sc;
622 	struct lagg_port *lp = NULL;
623 	int error = 0;
624 
625 	/* Should be checked by the caller */
626 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
627 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
628 		goto fallback;
629 
630 	switch (cmd) {
631 	case SIOCGLAGGPORT:
632 		if (rp->rp_portname[0] == '\0' ||
633 		    ifunit(rp->rp_portname) != ifp) {
634 			error = EINVAL;
635 			break;
636 		}
637 
638 		LAGG_RLOCK(sc);
639 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
640 			error = ENOENT;
641 			LAGG_RUNLOCK(sc);
642 			break;
643 		}
644 
645 		lagg_port2req(lp, rp);
646 		LAGG_RUNLOCK(sc);
647 		break;
648 
649 	case SIOCSIFCAP:
650 		if (lp->lp_ioctl == NULL) {
651 			error = EINVAL;
652 			break;
653 		}
654 		error = (*lp->lp_ioctl)(ifp, cmd, data);
655 		if (error)
656 			break;
657 
658 		/* Update lagg interface capabilities */
659 		LAGG_WLOCK(sc);
660 		lagg_capabilities(sc);
661 		LAGG_WUNLOCK(sc);
662 		break;
663 
664 	case SIOCSIFMTU:
665 		/* Do not allow the MTU to be changed once joined */
666 		error = EINVAL;
667 		break;
668 
669 	default:
670 		goto fallback;
671 	}
672 
673 	return (error);
674 
675 fallback:
676 	if (lp->lp_ioctl != NULL)
677 		return ((*lp->lp_ioctl)(ifp, cmd, data));
678 
679 	return (EINVAL);
680 }
681 
682 static int
683 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
684 	struct sockaddr *dst, struct route *ro)
685 {
686 	struct lagg_port *lp = ifp->if_lagg;
687 	struct ether_header *eh;
688 	short type = 0;
689 
690 	switch (dst->sa_family) {
691 		case pseudo_AF_HDRCMPLT:
692 		case AF_UNSPEC:
693 			eh = (struct ether_header *)dst->sa_data;
694 			type = eh->ether_type;
695 			break;
696 	}
697 
698 	/*
699 	 * Only allow ethernet types required to initiate or maintain the link,
700 	 * aggregated frames take a different path.
701 	 */
702 	switch (ntohs(type)) {
703 		case ETHERTYPE_PAE:	/* EAPOL PAE/802.1x */
704 			return ((*lp->lp_output)(ifp, m, dst, ro));
705 	}
706 
707 	/* drop any other frames */
708 	m_freem(m);
709 	return (EBUSY);
710 }
711 
712 static void
713 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
714 {
715 	struct lagg_port *lp;
716 	struct lagg_softc *sc;
717 
718 	if ((lp = ifp->if_lagg) == NULL)
719 		return;
720 
721 	sc = lp->lp_softc;
722 
723 	LAGG_WLOCK(sc);
724 	lp->lp_detaching = 1;
725 	lagg_port_destroy(lp, 1);
726 	LAGG_WUNLOCK(sc);
727 }
728 
729 static void
730 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
731 {
732 	struct lagg_softc *sc = lp->lp_softc;
733 
734 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
735 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
736 	rp->rp_prio = lp->lp_prio;
737 	rp->rp_flags = lp->lp_flags;
738 	if (sc->sc_portreq != NULL)
739 		(*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
740 
741 	/* Add protocol specific flags */
742 	switch (sc->sc_proto) {
743 		case LAGG_PROTO_FAILOVER:
744 			if (lp == sc->sc_primary)
745 				rp->rp_flags |= LAGG_PORT_MASTER;
746 			if (lp == lagg_link_active(sc, sc->sc_primary))
747 				rp->rp_flags |= LAGG_PORT_ACTIVE;
748 			break;
749 
750 		case LAGG_PROTO_ROUNDROBIN:
751 		case LAGG_PROTO_LOADBALANCE:
752 		case LAGG_PROTO_ETHERCHANNEL:
753 			if (LAGG_PORTACTIVE(lp))
754 				rp->rp_flags |= LAGG_PORT_ACTIVE;
755 			break;
756 
757 		case LAGG_PROTO_LACP:
758 			/* LACP has a different definition of active */
759 			if (lacp_isactive(lp))
760 				rp->rp_flags |= LAGG_PORT_ACTIVE;
761 			if (lacp_iscollecting(lp))
762 				rp->rp_flags |= LAGG_PORT_COLLECTING;
763 			if (lacp_isdistributing(lp))
764 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
765 			break;
766 	}
767 
768 }
769 
770 static void
771 lagg_init(void *xsc)
772 {
773 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
774 	struct lagg_port *lp;
775 	struct ifnet *ifp = sc->sc_ifp;
776 
777 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
778 		return;
779 
780 	LAGG_WLOCK(sc);
781 
782 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
783 	/* Update the port lladdrs */
784 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
785 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
786 
787 	if (sc->sc_init != NULL)
788 		(*sc->sc_init)(sc);
789 
790 	LAGG_WUNLOCK(sc);
791 }
792 
793 static void
794 lagg_stop(struct lagg_softc *sc)
795 {
796 	struct ifnet *ifp = sc->sc_ifp;
797 
798 	LAGG_WLOCK_ASSERT(sc);
799 
800 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
801 		return;
802 
803 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
804 
805 	if (sc->sc_stop != NULL)
806 		(*sc->sc_stop)(sc);
807 }
808 
809 static int
810 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
811 {
812 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
813 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
814 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
815 	struct ifreq *ifr = (struct ifreq *)data;
816 	struct lagg_port *lp;
817 	struct ifnet *tpif;
818 	struct thread *td = curthread;
819 	char *buf, *outbuf;
820 	int count, buflen, len, error = 0;
821 
822 	bzero(&rpbuf, sizeof(rpbuf));
823 
824 	switch (cmd) {
825 	case SIOCGLAGG:
826 		LAGG_RLOCK(sc);
827 		count = 0;
828 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
829 			count++;
830 		buflen = count * sizeof(struct lagg_reqport);
831 		LAGG_RUNLOCK(sc);
832 
833 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
834 
835 		LAGG_RLOCK(sc);
836 		ra->ra_proto = sc->sc_proto;
837 		if (sc->sc_req != NULL)
838 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
839 
840 		count = 0;
841 		buf = outbuf;
842 		len = min(ra->ra_size, buflen);
843 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
844 			if (len < sizeof(rpbuf))
845 				break;
846 
847 			lagg_port2req(lp, &rpbuf);
848 			memcpy(buf, &rpbuf, sizeof(rpbuf));
849 			count++;
850 			buf += sizeof(rpbuf);
851 			len -= sizeof(rpbuf);
852 		}
853 		LAGG_RUNLOCK(sc);
854 		ra->ra_ports = count;
855 		ra->ra_size = count * sizeof(rpbuf);
856 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
857 		free(outbuf, M_TEMP);
858 		break;
859 	case SIOCSLAGG:
860 		error = priv_check(td, PRIV_NET_LAGG);
861 		if (error)
862 			break;
863 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
864 			error = EPROTONOSUPPORT;
865 			break;
866 		}
867 		if (sc->sc_proto != LAGG_PROTO_NONE) {
868 			LAGG_WLOCK(sc);
869 			error = sc->sc_detach(sc);
870 			/* Reset protocol and pointers */
871 			sc->sc_proto = LAGG_PROTO_NONE;
872 			sc->sc_detach = NULL;
873 			sc->sc_start = NULL;
874 			sc->sc_input = NULL;
875 			sc->sc_port_create = NULL;
876 			sc->sc_port_destroy = NULL;
877 			sc->sc_linkstate = NULL;
878 			sc->sc_init = NULL;
879 			sc->sc_stop = NULL;
880 			sc->sc_lladdr = NULL;
881 			sc->sc_req = NULL;
882 			sc->sc_portreq = NULL;
883 			LAGG_WUNLOCK(sc);
884 		}
885 		if (error != 0)
886 			break;
887 		for (int i = 0; i < (sizeof(lagg_protos) /
888 		    sizeof(lagg_protos[0])); i++) {
889 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
890 				if (sc->sc_ifflags & IFF_DEBUG)
891 					printf("%s: using proto %u\n",
892 					    sc->sc_ifname,
893 					    lagg_protos[i].ti_proto);
894 				LAGG_WLOCK(sc);
895 				sc->sc_proto = lagg_protos[i].ti_proto;
896 				if (sc->sc_proto != LAGG_PROTO_NONE)
897 					error = lagg_protos[i].ti_attach(sc);
898 				LAGG_WUNLOCK(sc);
899 				return (error);
900 			}
901 		}
902 		error = EPROTONOSUPPORT;
903 		break;
904 	case SIOCGLAGGPORT:
905 		if (rp->rp_portname[0] == '\0' ||
906 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
907 			error = EINVAL;
908 			break;
909 		}
910 
911 		LAGG_RLOCK(sc);
912 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
913 		    lp->lp_softc != sc) {
914 			error = ENOENT;
915 			LAGG_RUNLOCK(sc);
916 			break;
917 		}
918 
919 		lagg_port2req(lp, rp);
920 		LAGG_RUNLOCK(sc);
921 		break;
922 	case SIOCSLAGGPORT:
923 		error = priv_check(td, PRIV_NET_LAGG);
924 		if (error)
925 			break;
926 		if (rp->rp_portname[0] == '\0' ||
927 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
928 			error = EINVAL;
929 			break;
930 		}
931 		LAGG_WLOCK(sc);
932 		error = lagg_port_create(sc, tpif);
933 		LAGG_WUNLOCK(sc);
934 		break;
935 	case SIOCSLAGGDELPORT:
936 		error = priv_check(td, PRIV_NET_LAGG);
937 		if (error)
938 			break;
939 		if (rp->rp_portname[0] == '\0' ||
940 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
941 			error = EINVAL;
942 			break;
943 		}
944 
945 		LAGG_WLOCK(sc);
946 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
947 		    lp->lp_softc != sc) {
948 			error = ENOENT;
949 			LAGG_WUNLOCK(sc);
950 			break;
951 		}
952 
953 		error = lagg_port_destroy(lp, 1);
954 		LAGG_WUNLOCK(sc);
955 		break;
956 	case SIOCSIFFLAGS:
957 		/* Set flags on ports too */
958 		LAGG_WLOCK(sc);
959 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
960 			lagg_setflags(lp, 1);
961 		}
962 		LAGG_WUNLOCK(sc);
963 
964 		if (!(ifp->if_flags & IFF_UP) &&
965 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
966 			/*
967 			 * If interface is marked down and it is running,
968 			 * then stop and disable it.
969 			 */
970 			LAGG_WLOCK(sc);
971 			lagg_stop(sc);
972 			LAGG_WUNLOCK(sc);
973 		} else if ((ifp->if_flags & IFF_UP) &&
974 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
975 			/*
976 			 * If interface is marked up and it is stopped, then
977 			 * start it.
978 			 */
979 			(*ifp->if_init)(sc);
980 		}
981 		break;
982 	case SIOCADDMULTI:
983 	case SIOCDELMULTI:
984 		LAGG_WLOCK(sc);
985 		error = lagg_ether_setmulti(sc);
986 		LAGG_WUNLOCK(sc);
987 		break;
988 	case SIOCSIFMEDIA:
989 	case SIOCGIFMEDIA:
990 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
991 		break;
992 
993 	case SIOCSIFCAP:
994 	case SIOCSIFMTU:
995 		/* Do not allow the MTU or caps to be directly changed */
996 		error = EINVAL;
997 		break;
998 
999 	default:
1000 		error = ether_ioctl(ifp, cmd, data);
1001 		break;
1002 	}
1003 	return (error);
1004 }
1005 
1006 static int
1007 lagg_ether_setmulti(struct lagg_softc *sc)
1008 {
1009 	struct lagg_port *lp;
1010 
1011 	LAGG_WLOCK_ASSERT(sc);
1012 
1013 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1014 		/* First, remove any existing filter entries. */
1015 		lagg_ether_cmdmulti(lp, 0);
1016 		/* copy all addresses from the lagg interface to the port */
1017 		lagg_ether_cmdmulti(lp, 1);
1018 	}
1019 	return (0);
1020 }
1021 
1022 static int
1023 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1024 {
1025 	struct lagg_softc *sc = lp->lp_softc;
1026 	struct ifnet *ifp = lp->lp_ifp;
1027 	struct ifnet *scifp = sc->sc_ifp;
1028 	struct lagg_mc *mc;
1029 	struct ifmultiaddr *ifma, *rifma = NULL;
1030 	struct sockaddr_dl sdl;
1031 	int error;
1032 
1033 	LAGG_WLOCK_ASSERT(sc);
1034 
1035 	bzero((char *)&sdl, sizeof(sdl));
1036 	sdl.sdl_len = sizeof(sdl);
1037 	sdl.sdl_family = AF_LINK;
1038 	sdl.sdl_type = IFT_ETHER;
1039 	sdl.sdl_alen = ETHER_ADDR_LEN;
1040 	sdl.sdl_index = ifp->if_index;
1041 
1042 	if (set) {
1043 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1044 			if (ifma->ifma_addr->sa_family != AF_LINK)
1045 				continue;
1046 			bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1047 			    LLADDR(&sdl), ETHER_ADDR_LEN);
1048 
1049 			error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1050 			if (error)
1051 				return (error);
1052 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1053 			if (mc == NULL)
1054 				return (ENOMEM);
1055 			mc->mc_ifma = rifma;
1056 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1057 		}
1058 	} else {
1059 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1060 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1061 			if_delmulti_ifma(mc->mc_ifma);
1062 			free(mc, M_DEVBUF);
1063 		}
1064 	}
1065 	return (0);
1066 }
1067 
1068 /* Handle a ref counted flag that should be set on the lagg port as well */
1069 static int
1070 lagg_setflag(struct lagg_port *lp, int flag, int status,
1071 	     int (*func)(struct ifnet *, int))
1072 {
1073 	struct lagg_softc *sc = lp->lp_softc;
1074 	struct ifnet *scifp = sc->sc_ifp;
1075 	struct ifnet *ifp = lp->lp_ifp;
1076 	int error;
1077 
1078 	LAGG_WLOCK_ASSERT(sc);
1079 
1080 	status = status ? (scifp->if_flags & flag) : 0;
1081 	/* Now "status" contains the flag value or 0 */
1082 
1083 	/*
1084 	 * See if recorded ports status is different from what
1085 	 * we want it to be.  If it is, flip it.  We record ports
1086 	 * status in lp_ifflags so that we won't clear ports flag
1087 	 * we haven't set.  In fact, we don't clear or set ports
1088 	 * flags directly, but get or release references to them.
1089 	 * That's why we can be sure that recorded flags still are
1090 	 * in accord with actual ports flags.
1091 	 */
1092 	if (status != (lp->lp_ifflags & flag)) {
1093 		error = (*func)(ifp, status);
1094 		if (error)
1095 			return (error);
1096 		lp->lp_ifflags &= ~flag;
1097 		lp->lp_ifflags |= status;
1098 	}
1099 	return (0);
1100 }
1101 
1102 /*
1103  * Handle IFF_* flags that require certain changes on the lagg port
1104  * if "status" is true, update ports flags respective to the lagg
1105  * if "status" is false, forcedly clear the flags set on port.
1106  */
1107 static int
1108 lagg_setflags(struct lagg_port *lp, int status)
1109 {
1110 	int error, i;
1111 
1112 	for (i = 0; lagg_pflags[i].flag; i++) {
1113 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1114 		    status, lagg_pflags[i].func);
1115 		if (error)
1116 			return (error);
1117 	}
1118 	return (0);
1119 }
1120 
1121 static void
1122 lagg_start(struct ifnet *ifp)
1123 {
1124 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1125 	struct mbuf *m;
1126 	int error = 0;
1127 
1128 	LAGG_RLOCK(sc);
1129 	/* We need a Tx algorithm and at least one port */
1130 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1131 		IF_DRAIN(&ifp->if_snd);
1132 		LAGG_RUNLOCK(sc);
1133 		return;
1134 	}
1135 
1136 	for (;; error = 0) {
1137 		IFQ_DEQUEUE(&ifp->if_snd, m);
1138 		if (m == NULL)
1139 			break;
1140 
1141 		ETHER_BPF_MTAP(ifp, m);
1142 
1143 		error = (*sc->sc_start)(sc, m);
1144 		if (error == 0)
1145 			ifp->if_opackets++;
1146 		else
1147 			ifp->if_oerrors++;
1148 	}
1149 	LAGG_RUNLOCK(sc);
1150 }
1151 
1152 static struct mbuf *
1153 lagg_input(struct ifnet *ifp, struct mbuf *m)
1154 {
1155 	struct lagg_port *lp = ifp->if_lagg;
1156 	struct lagg_softc *sc = lp->lp_softc;
1157 	struct ifnet *scifp = sc->sc_ifp;
1158 
1159 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1160 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1161 	    sc->sc_proto == LAGG_PROTO_NONE) {
1162 		m_freem(m);
1163 		return (NULL);
1164 	}
1165 
1166 	LAGG_RLOCK(sc);
1167 	ETHER_BPF_MTAP(scifp, m);
1168 
1169 	m = (*sc->sc_input)(sc, lp, m);
1170 
1171 	if (m != NULL) {
1172 		scifp->if_ipackets++;
1173 		scifp->if_ibytes += m->m_pkthdr.len;
1174 
1175 		if (scifp->if_flags & IFF_MONITOR) {
1176 			m_freem(m);
1177 			m = NULL;
1178 		}
1179 	}
1180 
1181 	LAGG_RUNLOCK(sc);
1182 	return (m);
1183 }
1184 
1185 static int
1186 lagg_media_change(struct ifnet *ifp)
1187 {
1188 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1189 
1190 	if (sc->sc_ifflags & IFF_DEBUG)
1191 		printf("%s\n", __func__);
1192 
1193 	/* Ignore */
1194 	return (0);
1195 }
1196 
1197 static void
1198 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1199 {
1200 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1201 	struct lagg_port *lp;
1202 
1203 	imr->ifm_status = IFM_AVALID;
1204 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1205 
1206 	LAGG_RLOCK(sc);
1207 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1208 		if (LAGG_PORTACTIVE(lp))
1209 			imr->ifm_status |= IFM_ACTIVE;
1210 	}
1211 	LAGG_RUNLOCK(sc);
1212 }
1213 
1214 static void
1215 lagg_linkstate(struct lagg_softc *sc)
1216 {
1217 	struct lagg_port *lp;
1218 	int new_link = LINK_STATE_DOWN;
1219 	uint64_t speed;
1220 
1221 	/* Our link is considered up if at least one of our ports is active */
1222 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1223 		if (lp->lp_link_state == LINK_STATE_UP) {
1224 			new_link = LINK_STATE_UP;
1225 			break;
1226 		}
1227 	}
1228 	if_link_state_change(sc->sc_ifp, new_link);
1229 
1230 	/* Update if_baudrate to reflect the max possible speed */
1231 	switch (sc->sc_proto) {
1232 		case LAGG_PROTO_FAILOVER:
1233 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1234 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1235 			break;
1236 		case LAGG_PROTO_ROUNDROBIN:
1237 		case LAGG_PROTO_LOADBALANCE:
1238 		case LAGG_PROTO_ETHERCHANNEL:
1239 			speed = 0;
1240 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1241 				speed += lp->lp_ifp->if_baudrate;
1242 			sc->sc_ifp->if_baudrate = speed;
1243 			break;
1244 		case LAGG_PROTO_LACP:
1245 			/* LACP updates if_baudrate itself */
1246 			break;
1247 	}
1248 }
1249 
1250 static void
1251 lagg_port_state(struct ifnet *ifp, int state)
1252 {
1253 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1254 	struct lagg_softc *sc = NULL;
1255 
1256 	if (lp != NULL)
1257 		sc = lp->lp_softc;
1258 	if (sc == NULL)
1259 		return;
1260 
1261 	LAGG_WLOCK(sc);
1262 	lagg_linkstate(sc);
1263 	if (sc->sc_linkstate != NULL)
1264 		(*sc->sc_linkstate)(lp);
1265 	LAGG_WUNLOCK(sc);
1266 }
1267 
1268 struct lagg_port *
1269 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1270 {
1271 	struct lagg_port *lp_next, *rval = NULL;
1272 	// int new_link = LINK_STATE_DOWN;
1273 
1274 	LAGG_RLOCK_ASSERT(sc);
1275 	/*
1276 	 * Search a port which reports an active link state.
1277 	 */
1278 
1279 	if (lp == NULL)
1280 		goto search;
1281 	if (LAGG_PORTACTIVE(lp)) {
1282 		rval = lp;
1283 		goto found;
1284 	}
1285 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1286 	    LAGG_PORTACTIVE(lp_next)) {
1287 		rval = lp_next;
1288 		goto found;
1289 	}
1290 
1291 search:
1292 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1293 		if (LAGG_PORTACTIVE(lp_next)) {
1294 			rval = lp_next;
1295 			goto found;
1296 		}
1297 	}
1298 
1299 found:
1300 	if (rval != NULL) {
1301 		/*
1302 		 * The IEEE 802.1D standard assumes that a lagg with
1303 		 * multiple ports is always full duplex. This is valid
1304 		 * for load sharing laggs and if at least two links
1305 		 * are active. Unfortunately, checking the latter would
1306 		 * be too expensive at this point.
1307 		 XXX
1308 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1309 		    (sc->sc_count > 1))
1310 			new_link = LINK_STATE_FULL_DUPLEX;
1311 		else
1312 			new_link = rval->lp_link_state;
1313 		 */
1314 	}
1315 
1316 	return (rval);
1317 }
1318 
1319 static const void *
1320 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1321 {
1322 	if (m->m_pkthdr.len < (off + len)) {
1323 		return (NULL);
1324 	} else if (m->m_len < (off + len)) {
1325 		m_copydata(m, off, len, buf);
1326 		return (buf);
1327 	}
1328 	return (mtod(m, char *) + off);
1329 }
1330 
1331 uint32_t
1332 lagg_hashmbuf(struct mbuf *m, uint32_t key)
1333 {
1334 	uint16_t etype;
1335 	uint32_t p = 0;
1336 	int off;
1337 	struct ether_header *eh;
1338 	struct ether_vlan_header vlanbuf;
1339 	const struct ether_vlan_header *vlan;
1340 #ifdef INET
1341 	const struct ip *ip;
1342 	struct ip ipbuf;
1343 #endif
1344 #ifdef INET6
1345 	const struct ip6_hdr *ip6;
1346 	struct ip6_hdr ip6buf;
1347 	uint32_t flow;
1348 #endif
1349 
1350 	off = sizeof(*eh);
1351 	if (m->m_len < off)
1352 		goto out;
1353 	eh = mtod(m, struct ether_header *);
1354 	etype = ntohs(eh->ether_type);
1355 	p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, key);
1356 	p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1357 
1358 	/* Special handling for encapsulating VLAN frames */
1359 	if (m->m_flags & M_VLANTAG) {
1360 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1361 		    sizeof(m->m_pkthdr.ether_vtag), p);
1362 	} else if (etype == ETHERTYPE_VLAN) {
1363 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &vlanbuf);
1364 		if (vlan == NULL)
1365 			goto out;
1366 
1367 		p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1368 		etype = ntohs(vlan->evl_proto);
1369 		off += sizeof(*vlan) - sizeof(*eh);
1370 	}
1371 
1372 	switch (etype) {
1373 #ifdef INET
1374 	case ETHERTYPE_IP:
1375 		ip = lagg_gethdr(m, off, sizeof(*ip), &ipbuf);
1376 		if (ip == NULL)
1377 			goto out;
1378 
1379 		p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1380 		p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1381 		break;
1382 #endif
1383 #ifdef INET6
1384 	case ETHERTYPE_IPV6:
1385 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &ip6buf);
1386 		if (ip6 == NULL)
1387 			goto out;
1388 
1389 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1390 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1391 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1392 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1393 		break;
1394 #endif
1395 	}
1396 out:
1397 	return (p);
1398 }
1399 
1400 int
1401 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1402 {
1403 
1404 	return (ifp->if_transmit)(ifp, m);
1405 }
1406 
1407 /*
1408  * Simple round robin aggregation
1409  */
1410 
1411 static int
1412 lagg_rr_attach(struct lagg_softc *sc)
1413 {
1414 	sc->sc_detach = lagg_rr_detach;
1415 	sc->sc_start = lagg_rr_start;
1416 	sc->sc_input = lagg_rr_input;
1417 	sc->sc_port_create = NULL;
1418 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1419 	sc->sc_seq = 0;
1420 
1421 	return (0);
1422 }
1423 
1424 static int
1425 lagg_rr_detach(struct lagg_softc *sc)
1426 {
1427 	return (0);
1428 }
1429 
1430 static int
1431 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1432 {
1433 	struct lagg_port *lp;
1434 	uint32_t p;
1435 
1436 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1437 	p %= sc->sc_count;
1438 	lp = SLIST_FIRST(&sc->sc_ports);
1439 	while (p--)
1440 		lp = SLIST_NEXT(lp, lp_entries);
1441 
1442 	/*
1443 	 * Check the port's link state. This will return the next active
1444 	 * port if the link is down or the port is NULL.
1445 	 */
1446 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1447 		m_freem(m);
1448 		return (ENOENT);
1449 	}
1450 
1451 	/* Send mbuf */
1452 	return (lagg_enqueue(lp->lp_ifp, m));
1453 }
1454 
1455 static struct mbuf *
1456 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1457 {
1458 	struct ifnet *ifp = sc->sc_ifp;
1459 
1460 	/* Just pass in the packet to our lagg device */
1461 	m->m_pkthdr.rcvif = ifp;
1462 
1463 	return (m);
1464 }
1465 
1466 /*
1467  * Active failover
1468  */
1469 
1470 static int
1471 lagg_fail_attach(struct lagg_softc *sc)
1472 {
1473 	sc->sc_detach = lagg_fail_detach;
1474 	sc->sc_start = lagg_fail_start;
1475 	sc->sc_input = lagg_fail_input;
1476 	sc->sc_port_create = NULL;
1477 	sc->sc_port_destroy = NULL;
1478 
1479 	return (0);
1480 }
1481 
1482 static int
1483 lagg_fail_detach(struct lagg_softc *sc)
1484 {
1485 	return (0);
1486 }
1487 
1488 static int
1489 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1490 {
1491 	struct lagg_port *lp;
1492 
1493 	/* Use the master port if active or the next available port */
1494 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1495 		m_freem(m);
1496 		return (ENOENT);
1497 	}
1498 
1499 	/* Send mbuf */
1500 	return (lagg_enqueue(lp->lp_ifp, m));
1501 }
1502 
1503 static struct mbuf *
1504 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1505 {
1506 	struct ifnet *ifp = sc->sc_ifp;
1507 	struct lagg_port *tmp_tp;
1508 
1509 	if (lp == sc->sc_primary) {
1510 		m->m_pkthdr.rcvif = ifp;
1511 		return (m);
1512 	}
1513 
1514 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1515 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1516 		/*
1517 		 * If tmp_tp is null, we've recieved a packet when all
1518 		 * our links are down. Weird, but process it anyways.
1519 		 */
1520 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1521 			m->m_pkthdr.rcvif = ifp;
1522 			return (m);
1523 		}
1524 	}
1525 
1526 	m_freem(m);
1527 	return (NULL);
1528 }
1529 
1530 /*
1531  * Loadbalancing
1532  */
1533 
1534 static int
1535 lagg_lb_attach(struct lagg_softc *sc)
1536 {
1537 	struct lagg_port *lp;
1538 	struct lagg_lb *lb;
1539 
1540 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1541 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1542 		return (ENOMEM);
1543 
1544 	sc->sc_detach = lagg_lb_detach;
1545 	sc->sc_start = lagg_lb_start;
1546 	sc->sc_input = lagg_lb_input;
1547 	sc->sc_port_create = lagg_lb_port_create;
1548 	sc->sc_port_destroy = lagg_lb_port_destroy;
1549 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1550 
1551 	lb->lb_key = arc4random();
1552 	sc->sc_psc = (caddr_t)lb;
1553 
1554 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1555 		lagg_lb_port_create(lp);
1556 
1557 	return (0);
1558 }
1559 
1560 static int
1561 lagg_lb_detach(struct lagg_softc *sc)
1562 {
1563 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1564 	if (lb != NULL)
1565 		free(lb, M_DEVBUF);
1566 	return (0);
1567 }
1568 
1569 static int
1570 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1571 {
1572 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1573 	struct lagg_port *lp_next;
1574 	int i = 0;
1575 
1576 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1577 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1578 		if (lp_next == lp)
1579 			continue;
1580 		if (i >= LAGG_MAX_PORTS)
1581 			return (EINVAL);
1582 		if (sc->sc_ifflags & IFF_DEBUG)
1583 			printf("%s: port %s at index %d\n",
1584 			    sc->sc_ifname, lp_next->lp_ifname, i);
1585 		lb->lb_ports[i++] = lp_next;
1586 	}
1587 
1588 	return (0);
1589 }
1590 
1591 static int
1592 lagg_lb_port_create(struct lagg_port *lp)
1593 {
1594 	struct lagg_softc *sc = lp->lp_softc;
1595 	return (lagg_lb_porttable(sc, NULL));
1596 }
1597 
1598 static void
1599 lagg_lb_port_destroy(struct lagg_port *lp)
1600 {
1601 	struct lagg_softc *sc = lp->lp_softc;
1602 	lagg_lb_porttable(sc, lp);
1603 }
1604 
1605 static int
1606 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1607 {
1608 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1609 	struct lagg_port *lp = NULL;
1610 	uint32_t p = 0;
1611 
1612 	if (m->m_flags & M_FLOWID)
1613 		p = m->m_pkthdr.flowid;
1614 	else
1615 		p = lagg_hashmbuf(m, lb->lb_key);
1616 	p %= sc->sc_count;
1617 	lp = lb->lb_ports[p];
1618 
1619 	/*
1620 	 * Check the port's link state. This will return the next active
1621 	 * port if the link is down or the port is NULL.
1622 	 */
1623 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1624 		m_freem(m);
1625 		return (ENOENT);
1626 	}
1627 
1628 	/* Send mbuf */
1629 	return (lagg_enqueue(lp->lp_ifp, m));
1630 }
1631 
1632 static struct mbuf *
1633 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1634 {
1635 	struct ifnet *ifp = sc->sc_ifp;
1636 
1637 	/* Just pass in the packet to our lagg device */
1638 	m->m_pkthdr.rcvif = ifp;
1639 
1640 	return (m);
1641 }
1642 
1643 /*
1644  * 802.3ad LACP
1645  */
1646 
1647 static int
1648 lagg_lacp_attach(struct lagg_softc *sc)
1649 {
1650 	struct lagg_port *lp;
1651 	int error;
1652 
1653 	sc->sc_detach = lagg_lacp_detach;
1654 	sc->sc_port_create = lacp_port_create;
1655 	sc->sc_port_destroy = lacp_port_destroy;
1656 	sc->sc_linkstate = lacp_linkstate;
1657 	sc->sc_start = lagg_lacp_start;
1658 	sc->sc_input = lagg_lacp_input;
1659 	sc->sc_init = lacp_init;
1660 	sc->sc_stop = lacp_stop;
1661 	sc->sc_lladdr = lagg_lacp_lladdr;
1662 	sc->sc_req = lacp_req;
1663 	sc->sc_portreq = lacp_portreq;
1664 
1665 	error = lacp_attach(sc);
1666 	if (error)
1667 		return (error);
1668 
1669 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1670 		lacp_port_create(lp);
1671 
1672 	return (error);
1673 }
1674 
1675 static int
1676 lagg_lacp_detach(struct lagg_softc *sc)
1677 {
1678 	struct lagg_port *lp;
1679 	int error;
1680 
1681 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1682 		lacp_port_destroy(lp);
1683 
1684 	/* unlocking is safe here */
1685 	LAGG_WUNLOCK(sc);
1686 	error = lacp_detach(sc);
1687 	LAGG_WLOCK(sc);
1688 
1689 	return (error);
1690 }
1691 
1692 static void
1693 lagg_lacp_lladdr(struct lagg_softc *sc)
1694 {
1695 	struct lagg_port *lp;
1696 
1697 	/* purge all the lacp ports */
1698 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1699 		lacp_port_destroy(lp);
1700 
1701 	/* add them back in */
1702 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1703 		lacp_port_create(lp);
1704 }
1705 
1706 static int
1707 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1708 {
1709 	struct lagg_port *lp;
1710 
1711 	lp = lacp_select_tx_port(sc, m);
1712 	if (lp == NULL) {
1713 		m_freem(m);
1714 		return (EBUSY);
1715 	}
1716 
1717 	/* Send mbuf */
1718 	return (lagg_enqueue(lp->lp_ifp, m));
1719 }
1720 
1721 static struct mbuf *
1722 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1723 {
1724 	struct ifnet *ifp = sc->sc_ifp;
1725 	struct ether_header *eh;
1726 	u_short etype;
1727 
1728 	eh = mtod(m, struct ether_header *);
1729 	etype = ntohs(eh->ether_type);
1730 
1731 	/* Tap off LACP control messages */
1732 	if (etype == ETHERTYPE_SLOW) {
1733 		m = lacp_input(lp, m);
1734 		if (m == NULL)
1735 			return (NULL);
1736 	}
1737 
1738 	/*
1739 	 * If the port is not collecting or not in the active aggregator then
1740 	 * free and return.
1741 	 */
1742 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
1743 		m_freem(m);
1744 		return (NULL);
1745 	}
1746 
1747 	m->m_pkthdr.rcvif = ifp;
1748 	return (m);
1749 }
1750