xref: /freebsd/sys/net/if_lagg.c (revision aa0a1e58f0189b0fde359a8bda032887e72057fa)
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 #include <sys/eventhandler.h>
43 
44 #include <net/ethernet.h>
45 #include <net/if.h>
46 #include <net/if_clone.h>
47 #include <net/if_arp.h>
48 #include <net/if_dl.h>
49 #include <net/if_llc.h>
50 #include <net/if_media.h>
51 #include <net/if_types.h>
52 #include <net/if_var.h>
53 #include <net/bpf.h>
54 
55 #ifdef INET
56 #include <netinet/in.h>
57 #include <netinet/in_systm.h>
58 #include <netinet/if_ether.h>
59 #include <netinet/ip.h>
60 #endif
61 
62 #ifdef INET6
63 #include <netinet/ip6.h>
64 #endif
65 
66 #include <net/if_vlan_var.h>
67 #include <net/if_lagg.h>
68 #include <net/ieee8023ad_lacp.h>
69 
70 /* Special flags we should propagate to the lagg ports. */
71 static struct {
72 	int flag;
73 	int (*func)(struct ifnet *, int);
74 } lagg_pflags[] = {
75 	{IFF_PROMISC, ifpromisc},
76 	{IFF_ALLMULTI, if_allmulti},
77 	{0, NULL}
78 };
79 
80 SLIST_HEAD(__trhead, lagg_softc) lagg_list;	/* list of laggs */
81 static struct mtx	lagg_list_mtx;
82 eventhandler_tag	lagg_detach_cookie = NULL;
83 
84 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
85 static void	lagg_clone_destroy(struct ifnet *);
86 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
87 static void	lagg_capabilities(struct lagg_softc *);
88 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
89 static void	lagg_port_setlladdr(void *, int);
90 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
91 static int	lagg_port_destroy(struct lagg_port *, int);
92 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
93 static void	lagg_linkstate(struct lagg_softc *);
94 static void	lagg_port_state(struct ifnet *, int);
95 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
96 static int	lagg_port_output(struct ifnet *, struct mbuf *,
97 		    struct sockaddr *, struct route *);
98 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
99 #ifdef LAGG_PORT_STACKING
100 static int	lagg_port_checkstacking(struct lagg_softc *);
101 #endif
102 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
103 static void	lagg_init(void *);
104 static void	lagg_stop(struct lagg_softc *);
105 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
106 static int	lagg_ether_setmulti(struct lagg_softc *);
107 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
108 static	int	lagg_setflag(struct lagg_port *, int, int,
109 		    int (*func)(struct ifnet *, int));
110 static	int	lagg_setflags(struct lagg_port *, int status);
111 static void	lagg_start(struct ifnet *);
112 static int	lagg_media_change(struct ifnet *);
113 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
114 static struct lagg_port *lagg_link_active(struct lagg_softc *,
115 	    struct lagg_port *);
116 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
117 
118 IFC_SIMPLE_DECLARE(lagg, 0);
119 
120 /* Simple round robin */
121 static int	lagg_rr_attach(struct lagg_softc *);
122 static int	lagg_rr_detach(struct lagg_softc *);
123 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
124 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
125 		    struct mbuf *);
126 
127 /* Active failover */
128 static int	lagg_fail_attach(struct lagg_softc *);
129 static int	lagg_fail_detach(struct lagg_softc *);
130 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
131 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
132 		    struct mbuf *);
133 
134 /* Loadbalancing */
135 static int	lagg_lb_attach(struct lagg_softc *);
136 static int	lagg_lb_detach(struct lagg_softc *);
137 static int	lagg_lb_port_create(struct lagg_port *);
138 static void	lagg_lb_port_destroy(struct lagg_port *);
139 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
140 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
141 		    struct mbuf *);
142 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
143 
144 /* 802.3ad LACP */
145 static int	lagg_lacp_attach(struct lagg_softc *);
146 static int	lagg_lacp_detach(struct lagg_softc *);
147 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
148 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
149 		    struct mbuf *);
150 static void	lagg_lacp_lladdr(struct lagg_softc *);
151 
152 /* lagg protocol table */
153 static const struct {
154 	int			ti_proto;
155 	int			(*ti_attach)(struct lagg_softc *);
156 } lagg_protos[] = {
157 	{ LAGG_PROTO_ROUNDROBIN,	lagg_rr_attach },
158 	{ LAGG_PROTO_FAILOVER,		lagg_fail_attach },
159 	{ LAGG_PROTO_LOADBALANCE,	lagg_lb_attach },
160 	{ LAGG_PROTO_ETHERCHANNEL,	lagg_lb_attach },
161 	{ LAGG_PROTO_LACP,		lagg_lacp_attach },
162 	{ LAGG_PROTO_NONE,		NULL }
163 };
164 
165 SYSCTL_DECL(_net_link);
166 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, "Link Aggregation");
167 
168 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */
169 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW,
170     &lagg_failover_rx_all, 0,
171     "Accept input from any interface in a failover lagg");
172 
173 static int
174 lagg_modevent(module_t mod, int type, void *data)
175 {
176 
177 	switch (type) {
178 	case MOD_LOAD:
179 		mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
180 		SLIST_INIT(&lagg_list);
181 		if_clone_attach(&lagg_cloner);
182 		lagg_input_p = lagg_input;
183 		lagg_linkstate_p = lagg_port_state;
184 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
185 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
186 		    EVENTHANDLER_PRI_ANY);
187 		break;
188 	case MOD_UNLOAD:
189 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
190 		    lagg_detach_cookie);
191 		if_clone_detach(&lagg_cloner);
192 		lagg_input_p = NULL;
193 		lagg_linkstate_p = NULL;
194 		mtx_destroy(&lagg_list_mtx);
195 		break;
196 	default:
197 		return (EOPNOTSUPP);
198 	}
199 	return (0);
200 }
201 
202 static moduledata_t lagg_mod = {
203 	"if_lagg",
204 	lagg_modevent,
205 	0
206 };
207 
208 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
209 
210 #if __FreeBSD_version >= 800000
211 /*
212  * This routine is run via an vlan
213  * config EVENT
214  */
215 static void
216 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
217 {
218         struct lagg_softc       *sc = ifp->if_softc;
219         struct lagg_port        *lp;
220 
221         if (ifp->if_softc !=  arg)   /* Not our event */
222                 return;
223 
224         LAGG_RLOCK(sc);
225         if (!SLIST_EMPTY(&sc->sc_ports)) {
226                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
227                         EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
228         }
229         LAGG_RUNLOCK(sc);
230 }
231 
232 /*
233  * This routine is run via an vlan
234  * unconfig EVENT
235  */
236 static void
237 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
238 {
239         struct lagg_softc       *sc = ifp->if_softc;
240         struct lagg_port        *lp;
241 
242         if (ifp->if_softc !=  arg)   /* Not our event */
243                 return;
244 
245         LAGG_RLOCK(sc);
246         if (!SLIST_EMPTY(&sc->sc_ports)) {
247                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
248                         EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
249         }
250         LAGG_RUNLOCK(sc);
251 }
252 #endif
253 
254 static int
255 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
256 {
257 	struct lagg_softc *sc;
258 	struct ifnet *ifp;
259 	int i, error = 0;
260 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
261 
262 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
263 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
264 	if (ifp == NULL) {
265 		free(sc, M_DEVBUF);
266 		return (ENOSPC);
267 	}
268 
269 	sc->sc_proto = LAGG_PROTO_NONE;
270 	for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
271 		if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
272 			sc->sc_proto = lagg_protos[i].ti_proto;
273 			if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
274 				if_free_type(ifp, IFT_ETHER);
275 				free(sc, M_DEVBUF);
276 				return (error);
277 			}
278 			break;
279 		}
280 	}
281 	LAGG_LOCK_INIT(sc);
282 	SLIST_INIT(&sc->sc_ports);
283 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
284 
285 	/* Initialise pseudo media types */
286 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
287 	    lagg_media_status);
288 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
289 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
290 
291 	if_initname(ifp, ifc->ifc_name, unit);
292 	ifp->if_type = IFT_ETHER;
293 	ifp->if_softc = sc;
294 	ifp->if_start = lagg_start;
295 	ifp->if_init = lagg_init;
296 	ifp->if_ioctl = lagg_ioctl;
297 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
298 
299 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
300 	ifp->if_snd.ifq_drv_maxlen = ifqmaxlen;
301 	IFQ_SET_READY(&ifp->if_snd);
302 
303 	/*
304 	 * Attach as an ordinary ethernet device, childs will be attached
305 	 * as special device IFT_IEEE8023ADLAG.
306 	 */
307 	ether_ifattach(ifp, eaddr);
308 
309 #if __FreeBSD_version >= 800000
310 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
311 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
312 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
313 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
314 #endif
315 
316 	/* Insert into the global list of laggs */
317 	mtx_lock(&lagg_list_mtx);
318 	SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
319 	mtx_unlock(&lagg_list_mtx);
320 
321 	return (0);
322 }
323 
324 static void
325 lagg_clone_destroy(struct ifnet *ifp)
326 {
327 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
328 	struct lagg_port *lp;
329 
330 	LAGG_WLOCK(sc);
331 
332 	lagg_stop(sc);
333 	ifp->if_flags &= ~IFF_UP;
334 
335 #if __FreeBSD_version >= 800000
336 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
337 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
338 #endif
339 
340 	/* Shutdown and remove lagg ports */
341 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
342 		lagg_port_destroy(lp, 1);
343 	/* Unhook the aggregation protocol */
344 	if (sc->sc_detach != NULL)
345 		(*sc->sc_detach)(sc);
346 
347 	LAGG_WUNLOCK(sc);
348 
349 	ifmedia_removeall(&sc->sc_media);
350 	ether_ifdetach(ifp);
351 	if_free_type(ifp, IFT_ETHER);
352 
353 	mtx_lock(&lagg_list_mtx);
354 	SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
355 	mtx_unlock(&lagg_list_mtx);
356 
357 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
358 	LAGG_LOCK_DESTROY(sc);
359 	free(sc, M_DEVBUF);
360 }
361 
362 static void
363 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
364 {
365 	struct ifnet *ifp = sc->sc_ifp;
366 
367 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
368 		return;
369 
370 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
371 	/* Let the protocol know the MAC has changed */
372 	if (sc->sc_lladdr != NULL)
373 		(*sc->sc_lladdr)(sc);
374 	EVENTHANDLER_INVOKE(iflladdr_event, ifp);
375 }
376 
377 static void
378 lagg_capabilities(struct lagg_softc *sc)
379 {
380 	struct lagg_port *lp;
381 	int cap = ~0, ena = ~0;
382 	u_long hwa = ~0UL;
383 
384 	LAGG_WLOCK_ASSERT(sc);
385 
386 	/* Get capabilities from the lagg ports */
387 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
388 		cap &= lp->lp_ifp->if_capabilities;
389 		ena &= lp->lp_ifp->if_capenable;
390 		hwa &= lp->lp_ifp->if_hwassist;
391 	}
392 	cap = (cap == ~0 ? 0 : cap);
393 	ena = (ena == ~0 ? 0 : ena);
394 	hwa = (hwa == ~0 ? 0 : hwa);
395 
396 	if (sc->sc_ifp->if_capabilities != cap ||
397 	    sc->sc_ifp->if_capenable != ena ||
398 	    sc->sc_ifp->if_hwassist != hwa) {
399 		sc->sc_ifp->if_capabilities = cap;
400 		sc->sc_ifp->if_capenable = ena;
401 		sc->sc_ifp->if_hwassist = hwa;
402 		getmicrotime(&sc->sc_ifp->if_lastchange);
403 
404 		if (sc->sc_ifflags & IFF_DEBUG)
405 			if_printf(sc->sc_ifp,
406 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
407 	}
408 }
409 
410 static void
411 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
412 {
413 	struct lagg_softc *sc = lp->lp_softc;
414 	struct ifnet *ifp = lp->lp_ifp;
415 	struct lagg_llq *llq;
416 	int pending = 0;
417 
418 	LAGG_WLOCK_ASSERT(sc);
419 
420 	if (lp->lp_detaching ||
421 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
422 		return;
423 
424 	/* Check to make sure its not already queued to be changed */
425 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
426 		if (llq->llq_ifp == ifp) {
427 			pending = 1;
428 			break;
429 		}
430 	}
431 
432 	if (!pending) {
433 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
434 		if (llq == NULL)	/* XXX what to do */
435 			return;
436 	}
437 
438 	/* Update the lladdr even if pending, it may have changed */
439 	llq->llq_ifp = ifp;
440 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
441 
442 	if (!pending)
443 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
444 
445 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
446 }
447 
448 /*
449  * Set the interface MAC address from a taskqueue to avoid a LOR.
450  */
451 static void
452 lagg_port_setlladdr(void *arg, int pending)
453 {
454 	struct lagg_softc *sc = (struct lagg_softc *)arg;
455 	struct lagg_llq *llq, *head;
456 	struct ifnet *ifp;
457 	int error;
458 
459 	/* Grab a local reference of the queue and remove it from the softc */
460 	LAGG_WLOCK(sc);
461 	head = SLIST_FIRST(&sc->sc_llq_head);
462 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
463 	LAGG_WUNLOCK(sc);
464 
465 	/*
466 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
467 	 * unlocked as we have the only reference to it.
468 	 */
469 	for (llq = head; llq != NULL; llq = head) {
470 		ifp = llq->llq_ifp;
471 
472 		/* Set the link layer address */
473 		error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
474 		if (error)
475 			printf("%s: setlladdr failed on %s\n", __func__,
476 			    ifp->if_xname);
477 
478 		head = SLIST_NEXT(llq, llq_entries);
479 		free(llq, M_DEVBUF);
480 	}
481 }
482 
483 static int
484 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
485 {
486 	struct lagg_softc *sc_ptr;
487 	struct lagg_port *lp;
488 	int error = 0;
489 
490 	LAGG_WLOCK_ASSERT(sc);
491 
492 	/* Limit the maximal number of lagg ports */
493 	if (sc->sc_count >= LAGG_MAX_PORTS)
494 		return (ENOSPC);
495 
496 	/* Check if port has already been associated to a lagg */
497 	if (ifp->if_lagg != NULL)
498 		return (EBUSY);
499 
500 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
501 	if (ifp->if_type != IFT_ETHER)
502 		return (EPROTONOSUPPORT);
503 
504 	/* Allow the first Ethernet member to define the MTU */
505 	if (SLIST_EMPTY(&sc->sc_ports))
506 		sc->sc_ifp->if_mtu = ifp->if_mtu;
507 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
508 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
509 		    ifp->if_xname);
510 		return (EINVAL);
511 	}
512 
513 	if ((lp = malloc(sizeof(struct lagg_port),
514 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
515 		return (ENOMEM);
516 
517 	/* Check if port is a stacked lagg */
518 	mtx_lock(&lagg_list_mtx);
519 	SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
520 		if (ifp == sc_ptr->sc_ifp) {
521 			mtx_unlock(&lagg_list_mtx);
522 			free(lp, M_DEVBUF);
523 			return (EINVAL);
524 			/* XXX disable stacking for the moment, its untested */
525 #ifdef LAGG_PORT_STACKING
526 			lp->lp_flags |= LAGG_PORT_STACK;
527 			if (lagg_port_checkstacking(sc_ptr) >=
528 			    LAGG_MAX_STACKING) {
529 				mtx_unlock(&lagg_list_mtx);
530 				free(lp, M_DEVBUF);
531 				return (E2BIG);
532 			}
533 #endif
534 		}
535 	}
536 	mtx_unlock(&lagg_list_mtx);
537 
538 	/* Change the interface type */
539 	lp->lp_iftype = ifp->if_type;
540 	ifp->if_type = IFT_IEEE8023ADLAG;
541 	ifp->if_lagg = lp;
542 	lp->lp_ioctl = ifp->if_ioctl;
543 	ifp->if_ioctl = lagg_port_ioctl;
544 	lp->lp_output = ifp->if_output;
545 	ifp->if_output = lagg_port_output;
546 
547 	lp->lp_ifp = ifp;
548 	lp->lp_softc = sc;
549 
550 	/* Save port link layer address */
551 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
552 
553 	if (SLIST_EMPTY(&sc->sc_ports)) {
554 		sc->sc_primary = lp;
555 		lagg_lladdr(sc, IF_LLADDR(ifp));
556 	} else {
557 		/* Update link layer address for this port */
558 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
559 	}
560 
561 	/* Insert into the list of ports */
562 	SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
563 	sc->sc_count++;
564 
565 	/* Update lagg capabilities */
566 	lagg_capabilities(sc);
567 	lagg_linkstate(sc);
568 
569 	/* Add multicast addresses and interface flags to this port */
570 	lagg_ether_cmdmulti(lp, 1);
571 	lagg_setflags(lp, 1);
572 
573 	if (sc->sc_port_create != NULL)
574 		error = (*sc->sc_port_create)(lp);
575 	if (error) {
576 		/* remove the port again, without calling sc_port_destroy */
577 		lagg_port_destroy(lp, 0);
578 		return (error);
579 	}
580 
581 	return (error);
582 }
583 
584 #ifdef LAGG_PORT_STACKING
585 static int
586 lagg_port_checkstacking(struct lagg_softc *sc)
587 {
588 	struct lagg_softc *sc_ptr;
589 	struct lagg_port *lp;
590 	int m = 0;
591 
592 	LAGG_WLOCK_ASSERT(sc);
593 
594 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
595 		if (lp->lp_flags & LAGG_PORT_STACK) {
596 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
597 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
598 		}
599 	}
600 
601 	return (m + 1);
602 }
603 #endif
604 
605 static int
606 lagg_port_destroy(struct lagg_port *lp, int runpd)
607 {
608 	struct lagg_softc *sc = lp->lp_softc;
609 	struct lagg_port *lp_ptr;
610 	struct lagg_llq *llq;
611 	struct ifnet *ifp = lp->lp_ifp;
612 
613 	LAGG_WLOCK_ASSERT(sc);
614 
615 	if (runpd && sc->sc_port_destroy != NULL)
616 		(*sc->sc_port_destroy)(lp);
617 
618 	/*
619 	 * Remove multicast addresses and interface flags from this port and
620 	 * reset the MAC address, skip if the interface is being detached.
621 	 */
622 	if (!lp->lp_detaching) {
623 		lagg_ether_cmdmulti(lp, 0);
624 		lagg_setflags(lp, 0);
625 		lagg_port_lladdr(lp, lp->lp_lladdr);
626 	}
627 
628 	/* Restore interface */
629 	ifp->if_type = lp->lp_iftype;
630 	ifp->if_ioctl = lp->lp_ioctl;
631 	ifp->if_output = lp->lp_output;
632 	ifp->if_lagg = NULL;
633 
634 	/* Finally, remove the port from the lagg */
635 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
636 	sc->sc_count--;
637 
638 	/* Update the primary interface */
639 	if (lp == sc->sc_primary) {
640 		uint8_t lladdr[ETHER_ADDR_LEN];
641 
642 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
643 			bzero(&lladdr, ETHER_ADDR_LEN);
644 		} else {
645 			bcopy(lp_ptr->lp_lladdr,
646 			    lladdr, ETHER_ADDR_LEN);
647 		}
648 		lagg_lladdr(sc, lladdr);
649 		sc->sc_primary = lp_ptr;
650 
651 		/* Update link layer address for each port */
652 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
653 			lagg_port_lladdr(lp_ptr, lladdr);
654 	}
655 
656 	/* Remove any pending lladdr changes from the queue */
657 	if (lp->lp_detaching) {
658 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
659 			if (llq->llq_ifp == ifp) {
660 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
661 				    llq_entries);
662 				free(llq, M_DEVBUF);
663 				break;	/* Only appears once */
664 			}
665 		}
666 	}
667 
668 	if (lp->lp_ifflags)
669 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
670 
671 	free(lp, M_DEVBUF);
672 
673 	/* Update lagg capabilities */
674 	lagg_capabilities(sc);
675 	lagg_linkstate(sc);
676 
677 	return (0);
678 }
679 
680 static int
681 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
682 {
683 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
684 	struct lagg_softc *sc;
685 	struct lagg_port *lp = NULL;
686 	int error = 0;
687 
688 	/* Should be checked by the caller */
689 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
690 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
691 		goto fallback;
692 
693 	switch (cmd) {
694 	case SIOCGLAGGPORT:
695 		if (rp->rp_portname[0] == '\0' ||
696 		    ifunit(rp->rp_portname) != ifp) {
697 			error = EINVAL;
698 			break;
699 		}
700 
701 		LAGG_RLOCK(sc);
702 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
703 			error = ENOENT;
704 			LAGG_RUNLOCK(sc);
705 			break;
706 		}
707 
708 		lagg_port2req(lp, rp);
709 		LAGG_RUNLOCK(sc);
710 		break;
711 
712 	case SIOCSIFCAP:
713 		if (lp->lp_ioctl == NULL) {
714 			error = EINVAL;
715 			break;
716 		}
717 		error = (*lp->lp_ioctl)(ifp, cmd, data);
718 		if (error)
719 			break;
720 
721 		/* Update lagg interface capabilities */
722 		LAGG_WLOCK(sc);
723 		lagg_capabilities(sc);
724 		LAGG_WUNLOCK(sc);
725 		break;
726 
727 	case SIOCSIFMTU:
728 		/* Do not allow the MTU to be changed once joined */
729 		error = EINVAL;
730 		break;
731 
732 	default:
733 		goto fallback;
734 	}
735 
736 	return (error);
737 
738 fallback:
739 	if (lp->lp_ioctl != NULL)
740 		return ((*lp->lp_ioctl)(ifp, cmd, data));
741 
742 	return (EINVAL);
743 }
744 
745 static int
746 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
747 	struct sockaddr *dst, struct route *ro)
748 {
749 	struct lagg_port *lp = ifp->if_lagg;
750 	struct ether_header *eh;
751 	short type = 0;
752 
753 	switch (dst->sa_family) {
754 		case pseudo_AF_HDRCMPLT:
755 		case AF_UNSPEC:
756 			eh = (struct ether_header *)dst->sa_data;
757 			type = eh->ether_type;
758 			break;
759 	}
760 
761 	/*
762 	 * Only allow ethernet types required to initiate or maintain the link,
763 	 * aggregated frames take a different path.
764 	 */
765 	switch (ntohs(type)) {
766 		case ETHERTYPE_PAE:	/* EAPOL PAE/802.1x */
767 			return ((*lp->lp_output)(ifp, m, dst, ro));
768 	}
769 
770 	/* drop any other frames */
771 	m_freem(m);
772 	return (EBUSY);
773 }
774 
775 static void
776 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
777 {
778 	struct lagg_port *lp;
779 	struct lagg_softc *sc;
780 
781 	if ((lp = ifp->if_lagg) == NULL)
782 		return;
783 
784 	sc = lp->lp_softc;
785 
786 	LAGG_WLOCK(sc);
787 	lp->lp_detaching = 1;
788 	lagg_port_destroy(lp, 1);
789 	LAGG_WUNLOCK(sc);
790 }
791 
792 static void
793 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
794 {
795 	struct lagg_softc *sc = lp->lp_softc;
796 
797 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
798 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
799 	rp->rp_prio = lp->lp_prio;
800 	rp->rp_flags = lp->lp_flags;
801 	if (sc->sc_portreq != NULL)
802 		(*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
803 
804 	/* Add protocol specific flags */
805 	switch (sc->sc_proto) {
806 		case LAGG_PROTO_FAILOVER:
807 			if (lp == sc->sc_primary)
808 				rp->rp_flags |= LAGG_PORT_MASTER;
809 			if (lp == lagg_link_active(sc, sc->sc_primary))
810 				rp->rp_flags |= LAGG_PORT_ACTIVE;
811 			break;
812 
813 		case LAGG_PROTO_ROUNDROBIN:
814 		case LAGG_PROTO_LOADBALANCE:
815 		case LAGG_PROTO_ETHERCHANNEL:
816 			if (LAGG_PORTACTIVE(lp))
817 				rp->rp_flags |= LAGG_PORT_ACTIVE;
818 			break;
819 
820 		case LAGG_PROTO_LACP:
821 			/* LACP has a different definition of active */
822 			if (lacp_isactive(lp))
823 				rp->rp_flags |= LAGG_PORT_ACTIVE;
824 			if (lacp_iscollecting(lp))
825 				rp->rp_flags |= LAGG_PORT_COLLECTING;
826 			if (lacp_isdistributing(lp))
827 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
828 			break;
829 	}
830 
831 }
832 
833 static void
834 lagg_init(void *xsc)
835 {
836 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
837 	struct lagg_port *lp;
838 	struct ifnet *ifp = sc->sc_ifp;
839 
840 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
841 		return;
842 
843 	LAGG_WLOCK(sc);
844 
845 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
846 	/* Update the port lladdrs */
847 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
848 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
849 
850 	if (sc->sc_init != NULL)
851 		(*sc->sc_init)(sc);
852 
853 	LAGG_WUNLOCK(sc);
854 }
855 
856 static void
857 lagg_stop(struct lagg_softc *sc)
858 {
859 	struct ifnet *ifp = sc->sc_ifp;
860 
861 	LAGG_WLOCK_ASSERT(sc);
862 
863 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
864 		return;
865 
866 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
867 
868 	if (sc->sc_stop != NULL)
869 		(*sc->sc_stop)(sc);
870 }
871 
872 static int
873 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
874 {
875 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
876 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
877 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
878 	struct ifreq *ifr = (struct ifreq *)data;
879 	struct lagg_port *lp;
880 	struct ifnet *tpif;
881 	struct thread *td = curthread;
882 	char *buf, *outbuf;
883 	int count, buflen, len, error = 0;
884 
885 	bzero(&rpbuf, sizeof(rpbuf));
886 
887 	switch (cmd) {
888 	case SIOCGLAGG:
889 		LAGG_RLOCK(sc);
890 		count = 0;
891 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
892 			count++;
893 		buflen = count * sizeof(struct lagg_reqport);
894 		LAGG_RUNLOCK(sc);
895 
896 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
897 
898 		LAGG_RLOCK(sc);
899 		ra->ra_proto = sc->sc_proto;
900 		if (sc->sc_req != NULL)
901 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
902 
903 		count = 0;
904 		buf = outbuf;
905 		len = min(ra->ra_size, buflen);
906 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
907 			if (len < sizeof(rpbuf))
908 				break;
909 
910 			lagg_port2req(lp, &rpbuf);
911 			memcpy(buf, &rpbuf, sizeof(rpbuf));
912 			count++;
913 			buf += sizeof(rpbuf);
914 			len -= sizeof(rpbuf);
915 		}
916 		LAGG_RUNLOCK(sc);
917 		ra->ra_ports = count;
918 		ra->ra_size = count * sizeof(rpbuf);
919 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
920 		free(outbuf, M_TEMP);
921 		break;
922 	case SIOCSLAGG:
923 		error = priv_check(td, PRIV_NET_LAGG);
924 		if (error)
925 			break;
926 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
927 			error = EPROTONOSUPPORT;
928 			break;
929 		}
930 		if (sc->sc_proto != LAGG_PROTO_NONE) {
931 			LAGG_WLOCK(sc);
932 			error = sc->sc_detach(sc);
933 			/* Reset protocol and pointers */
934 			sc->sc_proto = LAGG_PROTO_NONE;
935 			sc->sc_detach = NULL;
936 			sc->sc_start = NULL;
937 			sc->sc_input = NULL;
938 			sc->sc_port_create = NULL;
939 			sc->sc_port_destroy = NULL;
940 			sc->sc_linkstate = NULL;
941 			sc->sc_init = NULL;
942 			sc->sc_stop = NULL;
943 			sc->sc_lladdr = NULL;
944 			sc->sc_req = NULL;
945 			sc->sc_portreq = NULL;
946 			LAGG_WUNLOCK(sc);
947 		}
948 		if (error != 0)
949 			break;
950 		for (int i = 0; i < (sizeof(lagg_protos) /
951 		    sizeof(lagg_protos[0])); i++) {
952 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
953 				if (sc->sc_ifflags & IFF_DEBUG)
954 					printf("%s: using proto %u\n",
955 					    sc->sc_ifname,
956 					    lagg_protos[i].ti_proto);
957 				LAGG_WLOCK(sc);
958 				sc->sc_proto = lagg_protos[i].ti_proto;
959 				if (sc->sc_proto != LAGG_PROTO_NONE)
960 					error = lagg_protos[i].ti_attach(sc);
961 				LAGG_WUNLOCK(sc);
962 				return (error);
963 			}
964 		}
965 		error = EPROTONOSUPPORT;
966 		break;
967 	case SIOCGLAGGPORT:
968 		if (rp->rp_portname[0] == '\0' ||
969 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
970 			error = EINVAL;
971 			break;
972 		}
973 
974 		LAGG_RLOCK(sc);
975 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
976 		    lp->lp_softc != sc) {
977 			error = ENOENT;
978 			LAGG_RUNLOCK(sc);
979 			break;
980 		}
981 
982 		lagg_port2req(lp, rp);
983 		LAGG_RUNLOCK(sc);
984 		break;
985 	case SIOCSLAGGPORT:
986 		error = priv_check(td, PRIV_NET_LAGG);
987 		if (error)
988 			break;
989 		if (rp->rp_portname[0] == '\0' ||
990 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
991 			error = EINVAL;
992 			break;
993 		}
994 		LAGG_WLOCK(sc);
995 		error = lagg_port_create(sc, tpif);
996 		LAGG_WUNLOCK(sc);
997 		break;
998 	case SIOCSLAGGDELPORT:
999 		error = priv_check(td, PRIV_NET_LAGG);
1000 		if (error)
1001 			break;
1002 		if (rp->rp_portname[0] == '\0' ||
1003 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1004 			error = EINVAL;
1005 			break;
1006 		}
1007 
1008 		LAGG_WLOCK(sc);
1009 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1010 		    lp->lp_softc != sc) {
1011 			error = ENOENT;
1012 			LAGG_WUNLOCK(sc);
1013 			break;
1014 		}
1015 
1016 		error = lagg_port_destroy(lp, 1);
1017 		LAGG_WUNLOCK(sc);
1018 		break;
1019 	case SIOCSIFFLAGS:
1020 		/* Set flags on ports too */
1021 		LAGG_WLOCK(sc);
1022 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1023 			lagg_setflags(lp, 1);
1024 		}
1025 		LAGG_WUNLOCK(sc);
1026 
1027 		if (!(ifp->if_flags & IFF_UP) &&
1028 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1029 			/*
1030 			 * If interface is marked down and it is running,
1031 			 * then stop and disable it.
1032 			 */
1033 			LAGG_WLOCK(sc);
1034 			lagg_stop(sc);
1035 			LAGG_WUNLOCK(sc);
1036 		} else if ((ifp->if_flags & IFF_UP) &&
1037 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1038 			/*
1039 			 * If interface is marked up and it is stopped, then
1040 			 * start it.
1041 			 */
1042 			(*ifp->if_init)(sc);
1043 		}
1044 		break;
1045 	case SIOCADDMULTI:
1046 	case SIOCDELMULTI:
1047 		LAGG_WLOCK(sc);
1048 		error = lagg_ether_setmulti(sc);
1049 		LAGG_WUNLOCK(sc);
1050 		break;
1051 	case SIOCSIFMEDIA:
1052 	case SIOCGIFMEDIA:
1053 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1054 		break;
1055 
1056 	case SIOCSIFCAP:
1057 	case SIOCSIFMTU:
1058 		/* Do not allow the MTU or caps to be directly changed */
1059 		error = EINVAL;
1060 		break;
1061 
1062 	default:
1063 		error = ether_ioctl(ifp, cmd, data);
1064 		break;
1065 	}
1066 	return (error);
1067 }
1068 
1069 static int
1070 lagg_ether_setmulti(struct lagg_softc *sc)
1071 {
1072 	struct lagg_port *lp;
1073 
1074 	LAGG_WLOCK_ASSERT(sc);
1075 
1076 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1077 		/* First, remove any existing filter entries. */
1078 		lagg_ether_cmdmulti(lp, 0);
1079 		/* copy all addresses from the lagg interface to the port */
1080 		lagg_ether_cmdmulti(lp, 1);
1081 	}
1082 	return (0);
1083 }
1084 
1085 static int
1086 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1087 {
1088 	struct lagg_softc *sc = lp->lp_softc;
1089 	struct ifnet *ifp = lp->lp_ifp;
1090 	struct ifnet *scifp = sc->sc_ifp;
1091 	struct lagg_mc *mc;
1092 	struct ifmultiaddr *ifma, *rifma = NULL;
1093 	struct sockaddr_dl sdl;
1094 	int error;
1095 
1096 	LAGG_WLOCK_ASSERT(sc);
1097 
1098 	bzero((char *)&sdl, sizeof(sdl));
1099 	sdl.sdl_len = sizeof(sdl);
1100 	sdl.sdl_family = AF_LINK;
1101 	sdl.sdl_type = IFT_ETHER;
1102 	sdl.sdl_alen = ETHER_ADDR_LEN;
1103 	sdl.sdl_index = ifp->if_index;
1104 
1105 	if (set) {
1106 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1107 			if (ifma->ifma_addr->sa_family != AF_LINK)
1108 				continue;
1109 			bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1110 			    LLADDR(&sdl), ETHER_ADDR_LEN);
1111 
1112 			error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1113 			if (error)
1114 				return (error);
1115 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1116 			if (mc == NULL)
1117 				return (ENOMEM);
1118 			mc->mc_ifma = rifma;
1119 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1120 		}
1121 	} else {
1122 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1123 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1124 			if_delmulti_ifma(mc->mc_ifma);
1125 			free(mc, M_DEVBUF);
1126 		}
1127 	}
1128 	return (0);
1129 }
1130 
1131 /* Handle a ref counted flag that should be set on the lagg port as well */
1132 static int
1133 lagg_setflag(struct lagg_port *lp, int flag, int status,
1134 	     int (*func)(struct ifnet *, int))
1135 {
1136 	struct lagg_softc *sc = lp->lp_softc;
1137 	struct ifnet *scifp = sc->sc_ifp;
1138 	struct ifnet *ifp = lp->lp_ifp;
1139 	int error;
1140 
1141 	LAGG_WLOCK_ASSERT(sc);
1142 
1143 	status = status ? (scifp->if_flags & flag) : 0;
1144 	/* Now "status" contains the flag value or 0 */
1145 
1146 	/*
1147 	 * See if recorded ports status is different from what
1148 	 * we want it to be.  If it is, flip it.  We record ports
1149 	 * status in lp_ifflags so that we won't clear ports flag
1150 	 * we haven't set.  In fact, we don't clear or set ports
1151 	 * flags directly, but get or release references to them.
1152 	 * That's why we can be sure that recorded flags still are
1153 	 * in accord with actual ports flags.
1154 	 */
1155 	if (status != (lp->lp_ifflags & flag)) {
1156 		error = (*func)(ifp, status);
1157 		if (error)
1158 			return (error);
1159 		lp->lp_ifflags &= ~flag;
1160 		lp->lp_ifflags |= status;
1161 	}
1162 	return (0);
1163 }
1164 
1165 /*
1166  * Handle IFF_* flags that require certain changes on the lagg port
1167  * if "status" is true, update ports flags respective to the lagg
1168  * if "status" is false, forcedly clear the flags set on port.
1169  */
1170 static int
1171 lagg_setflags(struct lagg_port *lp, int status)
1172 {
1173 	int error, i;
1174 
1175 	for (i = 0; lagg_pflags[i].flag; i++) {
1176 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1177 		    status, lagg_pflags[i].func);
1178 		if (error)
1179 			return (error);
1180 	}
1181 	return (0);
1182 }
1183 
1184 static void
1185 lagg_start(struct ifnet *ifp)
1186 {
1187 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1188 	struct mbuf *m;
1189 	int error = 0;
1190 
1191 	LAGG_RLOCK(sc);
1192 	/* We need a Tx algorithm and at least one port */
1193 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1194 		IF_DRAIN(&ifp->if_snd);
1195 		LAGG_RUNLOCK(sc);
1196 		return;
1197 	}
1198 
1199 	for (;; error = 0) {
1200 		IFQ_DEQUEUE(&ifp->if_snd, m);
1201 		if (m == NULL)
1202 			break;
1203 
1204 		ETHER_BPF_MTAP(ifp, m);
1205 
1206 		error = (*sc->sc_start)(sc, m);
1207 		if (error == 0)
1208 			ifp->if_opackets++;
1209 		else
1210 			ifp->if_oerrors++;
1211 	}
1212 	LAGG_RUNLOCK(sc);
1213 }
1214 
1215 static struct mbuf *
1216 lagg_input(struct ifnet *ifp, struct mbuf *m)
1217 {
1218 	struct lagg_port *lp = ifp->if_lagg;
1219 	struct lagg_softc *sc = lp->lp_softc;
1220 	struct ifnet *scifp = sc->sc_ifp;
1221 
1222 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1223 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1224 	    sc->sc_proto == LAGG_PROTO_NONE) {
1225 		m_freem(m);
1226 		return (NULL);
1227 	}
1228 
1229 	LAGG_RLOCK(sc);
1230 	ETHER_BPF_MTAP(scifp, m);
1231 
1232 	m = (*sc->sc_input)(sc, lp, m);
1233 
1234 	if (m != NULL) {
1235 		scifp->if_ipackets++;
1236 		scifp->if_ibytes += m->m_pkthdr.len;
1237 
1238 		if (scifp->if_flags & IFF_MONITOR) {
1239 			m_freem(m);
1240 			m = NULL;
1241 		}
1242 	}
1243 
1244 	LAGG_RUNLOCK(sc);
1245 	return (m);
1246 }
1247 
1248 static int
1249 lagg_media_change(struct ifnet *ifp)
1250 {
1251 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1252 
1253 	if (sc->sc_ifflags & IFF_DEBUG)
1254 		printf("%s\n", __func__);
1255 
1256 	/* Ignore */
1257 	return (0);
1258 }
1259 
1260 static void
1261 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1262 {
1263 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1264 	struct lagg_port *lp;
1265 
1266 	imr->ifm_status = IFM_AVALID;
1267 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1268 
1269 	LAGG_RLOCK(sc);
1270 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1271 		if (LAGG_PORTACTIVE(lp))
1272 			imr->ifm_status |= IFM_ACTIVE;
1273 	}
1274 	LAGG_RUNLOCK(sc);
1275 }
1276 
1277 static void
1278 lagg_linkstate(struct lagg_softc *sc)
1279 {
1280 	struct lagg_port *lp;
1281 	int new_link = LINK_STATE_DOWN;
1282 	uint64_t speed;
1283 
1284 	/* Our link is considered up if at least one of our ports is active */
1285 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1286 		if (lp->lp_link_state == LINK_STATE_UP) {
1287 			new_link = LINK_STATE_UP;
1288 			break;
1289 		}
1290 	}
1291 	if_link_state_change(sc->sc_ifp, new_link);
1292 
1293 	/* Update if_baudrate to reflect the max possible speed */
1294 	switch (sc->sc_proto) {
1295 		case LAGG_PROTO_FAILOVER:
1296 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1297 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1298 			break;
1299 		case LAGG_PROTO_ROUNDROBIN:
1300 		case LAGG_PROTO_LOADBALANCE:
1301 		case LAGG_PROTO_ETHERCHANNEL:
1302 			speed = 0;
1303 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1304 				speed += lp->lp_ifp->if_baudrate;
1305 			sc->sc_ifp->if_baudrate = speed;
1306 			break;
1307 		case LAGG_PROTO_LACP:
1308 			/* LACP updates if_baudrate itself */
1309 			break;
1310 	}
1311 }
1312 
1313 static void
1314 lagg_port_state(struct ifnet *ifp, int state)
1315 {
1316 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1317 	struct lagg_softc *sc = NULL;
1318 
1319 	if (lp != NULL)
1320 		sc = lp->lp_softc;
1321 	if (sc == NULL)
1322 		return;
1323 
1324 	LAGG_WLOCK(sc);
1325 	lagg_linkstate(sc);
1326 	if (sc->sc_linkstate != NULL)
1327 		(*sc->sc_linkstate)(lp);
1328 	LAGG_WUNLOCK(sc);
1329 }
1330 
1331 struct lagg_port *
1332 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1333 {
1334 	struct lagg_port *lp_next, *rval = NULL;
1335 	// int new_link = LINK_STATE_DOWN;
1336 
1337 	LAGG_RLOCK_ASSERT(sc);
1338 	/*
1339 	 * Search a port which reports an active link state.
1340 	 */
1341 
1342 	if (lp == NULL)
1343 		goto search;
1344 	if (LAGG_PORTACTIVE(lp)) {
1345 		rval = lp;
1346 		goto found;
1347 	}
1348 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1349 	    LAGG_PORTACTIVE(lp_next)) {
1350 		rval = lp_next;
1351 		goto found;
1352 	}
1353 
1354 search:
1355 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1356 		if (LAGG_PORTACTIVE(lp_next)) {
1357 			rval = lp_next;
1358 			goto found;
1359 		}
1360 	}
1361 
1362 found:
1363 	if (rval != NULL) {
1364 		/*
1365 		 * The IEEE 802.1D standard assumes that a lagg with
1366 		 * multiple ports is always full duplex. This is valid
1367 		 * for load sharing laggs and if at least two links
1368 		 * are active. Unfortunately, checking the latter would
1369 		 * be too expensive at this point.
1370 		 XXX
1371 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1372 		    (sc->sc_count > 1))
1373 			new_link = LINK_STATE_FULL_DUPLEX;
1374 		else
1375 			new_link = rval->lp_link_state;
1376 		 */
1377 	}
1378 
1379 	return (rval);
1380 }
1381 
1382 static const void *
1383 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1384 {
1385 	if (m->m_pkthdr.len < (off + len)) {
1386 		return (NULL);
1387 	} else if (m->m_len < (off + len)) {
1388 		m_copydata(m, off, len, buf);
1389 		return (buf);
1390 	}
1391 	return (mtod(m, char *) + off);
1392 }
1393 
1394 uint32_t
1395 lagg_hashmbuf(struct mbuf *m, uint32_t key)
1396 {
1397 	uint16_t etype;
1398 	uint32_t p = 0;
1399 	int off;
1400 	struct ether_header *eh;
1401 	struct ether_vlan_header vlanbuf;
1402 	const struct ether_vlan_header *vlan;
1403 #ifdef INET
1404 	const struct ip *ip;
1405 	struct ip ipbuf;
1406 #endif
1407 #ifdef INET6
1408 	const struct ip6_hdr *ip6;
1409 	struct ip6_hdr ip6buf;
1410 	uint32_t flow;
1411 #endif
1412 
1413 	off = sizeof(*eh);
1414 	if (m->m_len < off)
1415 		goto out;
1416 	eh = mtod(m, struct ether_header *);
1417 	etype = ntohs(eh->ether_type);
1418 	p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, key);
1419 	p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1420 
1421 	/* Special handling for encapsulating VLAN frames */
1422 	if (m->m_flags & M_VLANTAG) {
1423 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1424 		    sizeof(m->m_pkthdr.ether_vtag), p);
1425 	} else if (etype == ETHERTYPE_VLAN) {
1426 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &vlanbuf);
1427 		if (vlan == NULL)
1428 			goto out;
1429 
1430 		p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1431 		etype = ntohs(vlan->evl_proto);
1432 		off += sizeof(*vlan) - sizeof(*eh);
1433 	}
1434 
1435 	switch (etype) {
1436 #ifdef INET
1437 	case ETHERTYPE_IP:
1438 		ip = lagg_gethdr(m, off, sizeof(*ip), &ipbuf);
1439 		if (ip == NULL)
1440 			goto out;
1441 
1442 		p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1443 		p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1444 		break;
1445 #endif
1446 #ifdef INET6
1447 	case ETHERTYPE_IPV6:
1448 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &ip6buf);
1449 		if (ip6 == NULL)
1450 			goto out;
1451 
1452 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1453 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1454 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1455 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1456 		break;
1457 #endif
1458 	}
1459 out:
1460 	return (p);
1461 }
1462 
1463 int
1464 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1465 {
1466 
1467 	return (ifp->if_transmit)(ifp, m);
1468 }
1469 
1470 /*
1471  * Simple round robin aggregation
1472  */
1473 
1474 static int
1475 lagg_rr_attach(struct lagg_softc *sc)
1476 {
1477 	sc->sc_detach = lagg_rr_detach;
1478 	sc->sc_start = lagg_rr_start;
1479 	sc->sc_input = lagg_rr_input;
1480 	sc->sc_port_create = NULL;
1481 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1482 	sc->sc_seq = 0;
1483 
1484 	return (0);
1485 }
1486 
1487 static int
1488 lagg_rr_detach(struct lagg_softc *sc)
1489 {
1490 	return (0);
1491 }
1492 
1493 static int
1494 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1495 {
1496 	struct lagg_port *lp;
1497 	uint32_t p;
1498 
1499 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1500 	p %= sc->sc_count;
1501 	lp = SLIST_FIRST(&sc->sc_ports);
1502 	while (p--)
1503 		lp = SLIST_NEXT(lp, lp_entries);
1504 
1505 	/*
1506 	 * Check the port's link state. This will return the next active
1507 	 * port if the link is down or the port is NULL.
1508 	 */
1509 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1510 		m_freem(m);
1511 		return (ENOENT);
1512 	}
1513 
1514 	/* Send mbuf */
1515 	return (lagg_enqueue(lp->lp_ifp, m));
1516 }
1517 
1518 static struct mbuf *
1519 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1520 {
1521 	struct ifnet *ifp = sc->sc_ifp;
1522 
1523 	/* Just pass in the packet to our lagg device */
1524 	m->m_pkthdr.rcvif = ifp;
1525 
1526 	return (m);
1527 }
1528 
1529 /*
1530  * Active failover
1531  */
1532 
1533 static int
1534 lagg_fail_attach(struct lagg_softc *sc)
1535 {
1536 	sc->sc_detach = lagg_fail_detach;
1537 	sc->sc_start = lagg_fail_start;
1538 	sc->sc_input = lagg_fail_input;
1539 	sc->sc_port_create = NULL;
1540 	sc->sc_port_destroy = NULL;
1541 
1542 	return (0);
1543 }
1544 
1545 static int
1546 lagg_fail_detach(struct lagg_softc *sc)
1547 {
1548 	return (0);
1549 }
1550 
1551 static int
1552 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1553 {
1554 	struct lagg_port *lp;
1555 
1556 	/* Use the master port if active or the next available port */
1557 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1558 		m_freem(m);
1559 		return (ENOENT);
1560 	}
1561 
1562 	/* Send mbuf */
1563 	return (lagg_enqueue(lp->lp_ifp, m));
1564 }
1565 
1566 static struct mbuf *
1567 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1568 {
1569 	struct ifnet *ifp = sc->sc_ifp;
1570 	struct lagg_port *tmp_tp;
1571 
1572 	if (lp == sc->sc_primary || lagg_failover_rx_all) {
1573 		m->m_pkthdr.rcvif = ifp;
1574 		return (m);
1575 	}
1576 
1577 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1578 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1579 		/*
1580 		 * If tmp_tp is null, we've recieved a packet when all
1581 		 * our links are down. Weird, but process it anyways.
1582 		 */
1583 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1584 			m->m_pkthdr.rcvif = ifp;
1585 			return (m);
1586 		}
1587 	}
1588 
1589 	m_freem(m);
1590 	return (NULL);
1591 }
1592 
1593 /*
1594  * Loadbalancing
1595  */
1596 
1597 static int
1598 lagg_lb_attach(struct lagg_softc *sc)
1599 {
1600 	struct lagg_port *lp;
1601 	struct lagg_lb *lb;
1602 
1603 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1604 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1605 		return (ENOMEM);
1606 
1607 	sc->sc_detach = lagg_lb_detach;
1608 	sc->sc_start = lagg_lb_start;
1609 	sc->sc_input = lagg_lb_input;
1610 	sc->sc_port_create = lagg_lb_port_create;
1611 	sc->sc_port_destroy = lagg_lb_port_destroy;
1612 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1613 
1614 	lb->lb_key = arc4random();
1615 	sc->sc_psc = (caddr_t)lb;
1616 
1617 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1618 		lagg_lb_port_create(lp);
1619 
1620 	return (0);
1621 }
1622 
1623 static int
1624 lagg_lb_detach(struct lagg_softc *sc)
1625 {
1626 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1627 	if (lb != NULL)
1628 		free(lb, M_DEVBUF);
1629 	return (0);
1630 }
1631 
1632 static int
1633 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1634 {
1635 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1636 	struct lagg_port *lp_next;
1637 	int i = 0;
1638 
1639 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1640 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1641 		if (lp_next == lp)
1642 			continue;
1643 		if (i >= LAGG_MAX_PORTS)
1644 			return (EINVAL);
1645 		if (sc->sc_ifflags & IFF_DEBUG)
1646 			printf("%s: port %s at index %d\n",
1647 			    sc->sc_ifname, lp_next->lp_ifname, i);
1648 		lb->lb_ports[i++] = lp_next;
1649 	}
1650 
1651 	return (0);
1652 }
1653 
1654 static int
1655 lagg_lb_port_create(struct lagg_port *lp)
1656 {
1657 	struct lagg_softc *sc = lp->lp_softc;
1658 	return (lagg_lb_porttable(sc, NULL));
1659 }
1660 
1661 static void
1662 lagg_lb_port_destroy(struct lagg_port *lp)
1663 {
1664 	struct lagg_softc *sc = lp->lp_softc;
1665 	lagg_lb_porttable(sc, lp);
1666 }
1667 
1668 static int
1669 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1670 {
1671 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1672 	struct lagg_port *lp = NULL;
1673 	uint32_t p = 0;
1674 
1675 	if (m->m_flags & M_FLOWID)
1676 		p = m->m_pkthdr.flowid;
1677 	else
1678 		p = lagg_hashmbuf(m, lb->lb_key);
1679 	p %= sc->sc_count;
1680 	lp = lb->lb_ports[p];
1681 
1682 	/*
1683 	 * Check the port's link state. This will return the next active
1684 	 * port if the link is down or the port is NULL.
1685 	 */
1686 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1687 		m_freem(m);
1688 		return (ENOENT);
1689 	}
1690 
1691 	/* Send mbuf */
1692 	return (lagg_enqueue(lp->lp_ifp, m));
1693 }
1694 
1695 static struct mbuf *
1696 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1697 {
1698 	struct ifnet *ifp = sc->sc_ifp;
1699 
1700 	/* Just pass in the packet to our lagg device */
1701 	m->m_pkthdr.rcvif = ifp;
1702 
1703 	return (m);
1704 }
1705 
1706 /*
1707  * 802.3ad LACP
1708  */
1709 
1710 static int
1711 lagg_lacp_attach(struct lagg_softc *sc)
1712 {
1713 	struct lagg_port *lp;
1714 	int error;
1715 
1716 	sc->sc_detach = lagg_lacp_detach;
1717 	sc->sc_port_create = lacp_port_create;
1718 	sc->sc_port_destroy = lacp_port_destroy;
1719 	sc->sc_linkstate = lacp_linkstate;
1720 	sc->sc_start = lagg_lacp_start;
1721 	sc->sc_input = lagg_lacp_input;
1722 	sc->sc_init = lacp_init;
1723 	sc->sc_stop = lacp_stop;
1724 	sc->sc_lladdr = lagg_lacp_lladdr;
1725 	sc->sc_req = lacp_req;
1726 	sc->sc_portreq = lacp_portreq;
1727 
1728 	error = lacp_attach(sc);
1729 	if (error)
1730 		return (error);
1731 
1732 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1733 		lacp_port_create(lp);
1734 
1735 	return (error);
1736 }
1737 
1738 static int
1739 lagg_lacp_detach(struct lagg_softc *sc)
1740 {
1741 	struct lagg_port *lp;
1742 	int error;
1743 
1744 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1745 		lacp_port_destroy(lp);
1746 
1747 	/* unlocking is safe here */
1748 	LAGG_WUNLOCK(sc);
1749 	error = lacp_detach(sc);
1750 	LAGG_WLOCK(sc);
1751 
1752 	return (error);
1753 }
1754 
1755 static void
1756 lagg_lacp_lladdr(struct lagg_softc *sc)
1757 {
1758 	struct lagg_port *lp;
1759 
1760 	/* purge all the lacp ports */
1761 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1762 		lacp_port_destroy(lp);
1763 
1764 	/* add them back in */
1765 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1766 		lacp_port_create(lp);
1767 }
1768 
1769 static int
1770 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1771 {
1772 	struct lagg_port *lp;
1773 
1774 	lp = lacp_select_tx_port(sc, m);
1775 	if (lp == NULL) {
1776 		m_freem(m);
1777 		return (EBUSY);
1778 	}
1779 
1780 	/* Send mbuf */
1781 	return (lagg_enqueue(lp->lp_ifp, m));
1782 }
1783 
1784 static struct mbuf *
1785 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1786 {
1787 	struct ifnet *ifp = sc->sc_ifp;
1788 	struct ether_header *eh;
1789 	u_short etype;
1790 
1791 	eh = mtod(m, struct ether_header *);
1792 	etype = ntohs(eh->ether_type);
1793 
1794 	/* Tap off LACP control messages */
1795 	if (etype == ETHERTYPE_SLOW) {
1796 		m = lacp_input(lp, m);
1797 		if (m == NULL)
1798 			return (NULL);
1799 	}
1800 
1801 	/*
1802 	 * If the port is not collecting or not in the active aggregator then
1803 	 * free and return.
1804 	 */
1805 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
1806 		m_freem(m);
1807 		return (NULL);
1808 	}
1809 
1810 	m->m_pkthdr.rcvif = ifp;
1811 	return (m);
1812 }
1813