xref: /freebsd/sys/net/if_lagg.c (revision 10b59a9b4add0320d52c15ce057dd697261e7dfc)
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 #if defined(INET) || defined(INET6)
56 #include <netinet/in.h>
57 #endif
58 #ifdef INET
59 #include <netinet/in_systm.h>
60 #include <netinet/if_ether.h>
61 #include <netinet/ip.h>
62 #endif
63 
64 #ifdef INET6
65 #include <netinet/ip6.h>
66 #endif
67 
68 #include <net/if_vlan_var.h>
69 #include <net/if_lagg.h>
70 #include <net/ieee8023ad_lacp.h>
71 
72 /* Special flags we should propagate to the lagg ports. */
73 static struct {
74 	int flag;
75 	int (*func)(struct ifnet *, int);
76 } lagg_pflags[] = {
77 	{IFF_PROMISC, ifpromisc},
78 	{IFF_ALLMULTI, if_allmulti},
79 	{0, NULL}
80 };
81 
82 SLIST_HEAD(__trhead, lagg_softc) lagg_list;	/* list of laggs */
83 static struct mtx	lagg_list_mtx;
84 eventhandler_tag	lagg_detach_cookie = NULL;
85 
86 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
87 static void	lagg_clone_destroy(struct ifnet *);
88 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
89 static void	lagg_capabilities(struct lagg_softc *);
90 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
91 static void	lagg_port_setlladdr(void *, int);
92 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
93 static int	lagg_port_destroy(struct lagg_port *, int);
94 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
95 static void	lagg_linkstate(struct lagg_softc *);
96 static void	lagg_port_state(struct ifnet *, int);
97 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
98 static int	lagg_port_output(struct ifnet *, struct mbuf *,
99 		    struct sockaddr *, struct route *);
100 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
101 #ifdef LAGG_PORT_STACKING
102 static int	lagg_port_checkstacking(struct lagg_softc *);
103 #endif
104 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
105 static void	lagg_init(void *);
106 static void	lagg_stop(struct lagg_softc *);
107 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
108 static int	lagg_ether_setmulti(struct lagg_softc *);
109 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
110 static	int	lagg_setflag(struct lagg_port *, int, int,
111 		    int (*func)(struct ifnet *, int));
112 static	int	lagg_setflags(struct lagg_port *, int status);
113 static void	lagg_start(struct ifnet *);
114 static int	lagg_media_change(struct ifnet *);
115 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
116 static struct lagg_port *lagg_link_active(struct lagg_softc *,
117 	    struct lagg_port *);
118 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
119 
120 IFC_SIMPLE_DECLARE(lagg, 0);
121 
122 /* Simple round robin */
123 static int	lagg_rr_attach(struct lagg_softc *);
124 static int	lagg_rr_detach(struct lagg_softc *);
125 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
126 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
127 		    struct mbuf *);
128 
129 /* Active failover */
130 static int	lagg_fail_attach(struct lagg_softc *);
131 static int	lagg_fail_detach(struct lagg_softc *);
132 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
133 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
134 		    struct mbuf *);
135 
136 /* Loadbalancing */
137 static int	lagg_lb_attach(struct lagg_softc *);
138 static int	lagg_lb_detach(struct lagg_softc *);
139 static int	lagg_lb_port_create(struct lagg_port *);
140 static void	lagg_lb_port_destroy(struct lagg_port *);
141 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
142 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
143 		    struct mbuf *);
144 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
145 
146 /* 802.3ad LACP */
147 static int	lagg_lacp_attach(struct lagg_softc *);
148 static int	lagg_lacp_detach(struct lagg_softc *);
149 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
150 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
151 		    struct mbuf *);
152 static void	lagg_lacp_lladdr(struct lagg_softc *);
153 
154 /* lagg protocol table */
155 static const struct {
156 	int			ti_proto;
157 	int			(*ti_attach)(struct lagg_softc *);
158 } lagg_protos[] = {
159 	{ LAGG_PROTO_ROUNDROBIN,	lagg_rr_attach },
160 	{ LAGG_PROTO_FAILOVER,		lagg_fail_attach },
161 	{ LAGG_PROTO_LOADBALANCE,	lagg_lb_attach },
162 	{ LAGG_PROTO_ETHERCHANNEL,	lagg_lb_attach },
163 	{ LAGG_PROTO_LACP,		lagg_lacp_attach },
164 	{ LAGG_PROTO_NONE,		NULL }
165 };
166 
167 SYSCTL_DECL(_net_link);
168 static SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
169     "Link Aggregation");
170 
171 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */
172 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW,
173     &lagg_failover_rx_all, 0,
174     "Accept input from any interface in a failover lagg");
175 
176 static int
177 lagg_modevent(module_t mod, int type, void *data)
178 {
179 
180 	switch (type) {
181 	case MOD_LOAD:
182 		mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
183 		SLIST_INIT(&lagg_list);
184 		if_clone_attach(&lagg_cloner);
185 		lagg_input_p = lagg_input;
186 		lagg_linkstate_p = lagg_port_state;
187 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
188 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
189 		    EVENTHANDLER_PRI_ANY);
190 		break;
191 	case MOD_UNLOAD:
192 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
193 		    lagg_detach_cookie);
194 		if_clone_detach(&lagg_cloner);
195 		lagg_input_p = NULL;
196 		lagg_linkstate_p = NULL;
197 		mtx_destroy(&lagg_list_mtx);
198 		break;
199 	default:
200 		return (EOPNOTSUPP);
201 	}
202 	return (0);
203 }
204 
205 static moduledata_t lagg_mod = {
206 	"if_lagg",
207 	lagg_modevent,
208 	0
209 };
210 
211 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
212 MODULE_VERSION(if_lagg, 1);
213 
214 #if __FreeBSD_version >= 800000
215 /*
216  * This routine is run via an vlan
217  * config EVENT
218  */
219 static void
220 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
221 {
222         struct lagg_softc       *sc = ifp->if_softc;
223         struct lagg_port        *lp;
224 
225         if (ifp->if_softc !=  arg)   /* Not our event */
226                 return;
227 
228         LAGG_RLOCK(sc);
229         if (!SLIST_EMPTY(&sc->sc_ports)) {
230                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
231                         EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
232         }
233         LAGG_RUNLOCK(sc);
234 }
235 
236 /*
237  * This routine is run via an vlan
238  * unconfig EVENT
239  */
240 static void
241 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
242 {
243         struct lagg_softc       *sc = ifp->if_softc;
244         struct lagg_port        *lp;
245 
246         if (ifp->if_softc !=  arg)   /* Not our event */
247                 return;
248 
249         LAGG_RLOCK(sc);
250         if (!SLIST_EMPTY(&sc->sc_ports)) {
251                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
252                         EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
253         }
254         LAGG_RUNLOCK(sc);
255 }
256 #endif
257 
258 static int
259 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
260 {
261 	struct lagg_softc *sc;
262 	struct ifnet *ifp;
263 	int i, error = 0;
264 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
265 
266 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
267 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
268 	if (ifp == NULL) {
269 		free(sc, M_DEVBUF);
270 		return (ENOSPC);
271 	}
272 
273 	sc->sc_proto = LAGG_PROTO_NONE;
274 	for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
275 		if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
276 			sc->sc_proto = lagg_protos[i].ti_proto;
277 			if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
278 				if_free(ifp);
279 				free(sc, M_DEVBUF);
280 				return (error);
281 			}
282 			break;
283 		}
284 	}
285 	LAGG_LOCK_INIT(sc);
286 	SLIST_INIT(&sc->sc_ports);
287 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
288 
289 	/* Initialise pseudo media types */
290 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
291 	    lagg_media_status);
292 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
293 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
294 
295 	if_initname(ifp, ifc->ifc_name, unit);
296 	ifp->if_softc = sc;
297 	ifp->if_start = lagg_start;
298 	ifp->if_init = lagg_init;
299 	ifp->if_ioctl = lagg_ioctl;
300 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
301 
302 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
303 	ifp->if_snd.ifq_drv_maxlen = ifqmaxlen;
304 	IFQ_SET_READY(&ifp->if_snd);
305 
306 	/*
307 	 * Attach as an ordinary ethernet device, children will be attached
308 	 * as special device IFT_IEEE8023ADLAG.
309 	 */
310 	ether_ifattach(ifp, eaddr);
311 
312 #if __FreeBSD_version >= 800000
313 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
314 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
315 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
316 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
317 #endif
318 
319 	/* Insert into the global list of laggs */
320 	mtx_lock(&lagg_list_mtx);
321 	SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
322 	mtx_unlock(&lagg_list_mtx);
323 
324 	return (0);
325 }
326 
327 static void
328 lagg_clone_destroy(struct ifnet *ifp)
329 {
330 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
331 	struct lagg_port *lp;
332 
333 	LAGG_WLOCK(sc);
334 
335 	lagg_stop(sc);
336 	ifp->if_flags &= ~IFF_UP;
337 
338 #if __FreeBSD_version >= 800000
339 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
340 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
341 #endif
342 
343 	/* Shutdown and remove lagg ports */
344 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
345 		lagg_port_destroy(lp, 1);
346 	/* Unhook the aggregation protocol */
347 	if (sc->sc_detach != NULL)
348 		(*sc->sc_detach)(sc);
349 
350 	LAGG_WUNLOCK(sc);
351 
352 	ifmedia_removeall(&sc->sc_media);
353 	ether_ifdetach(ifp);
354 	if_free(ifp);
355 
356 	mtx_lock(&lagg_list_mtx);
357 	SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
358 	mtx_unlock(&lagg_list_mtx);
359 
360 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
361 	LAGG_LOCK_DESTROY(sc);
362 	free(sc, M_DEVBUF);
363 }
364 
365 static void
366 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
367 {
368 	struct ifnet *ifp = sc->sc_ifp;
369 
370 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
371 		return;
372 
373 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
374 	/* Let the protocol know the MAC has changed */
375 	if (sc->sc_lladdr != NULL)
376 		(*sc->sc_lladdr)(sc);
377 	EVENTHANDLER_INVOKE(iflladdr_event, ifp);
378 }
379 
380 static void
381 lagg_capabilities(struct lagg_softc *sc)
382 {
383 	struct lagg_port *lp;
384 	int cap = ~0, ena = ~0;
385 	u_long hwa = ~0UL;
386 
387 	LAGG_WLOCK_ASSERT(sc);
388 
389 	/* Get capabilities from the lagg ports */
390 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
391 		cap &= lp->lp_ifp->if_capabilities;
392 		ena &= lp->lp_ifp->if_capenable;
393 		hwa &= lp->lp_ifp->if_hwassist;
394 	}
395 	cap = (cap == ~0 ? 0 : cap);
396 	ena = (ena == ~0 ? 0 : ena);
397 	hwa = (hwa == ~0 ? 0 : hwa);
398 
399 	if (sc->sc_ifp->if_capabilities != cap ||
400 	    sc->sc_ifp->if_capenable != ena ||
401 	    sc->sc_ifp->if_hwassist != hwa) {
402 		sc->sc_ifp->if_capabilities = cap;
403 		sc->sc_ifp->if_capenable = ena;
404 		sc->sc_ifp->if_hwassist = hwa;
405 		getmicrotime(&sc->sc_ifp->if_lastchange);
406 
407 		if (sc->sc_ifflags & IFF_DEBUG)
408 			if_printf(sc->sc_ifp,
409 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
410 	}
411 }
412 
413 static void
414 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
415 {
416 	struct lagg_softc *sc = lp->lp_softc;
417 	struct ifnet *ifp = lp->lp_ifp;
418 	struct lagg_llq *llq;
419 	int pending = 0;
420 
421 	LAGG_WLOCK_ASSERT(sc);
422 
423 	if (lp->lp_detaching ||
424 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
425 		return;
426 
427 	/* Check to make sure its not already queued to be changed */
428 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
429 		if (llq->llq_ifp == ifp) {
430 			pending = 1;
431 			break;
432 		}
433 	}
434 
435 	if (!pending) {
436 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
437 		if (llq == NULL)	/* XXX what to do */
438 			return;
439 	}
440 
441 	/* Update the lladdr even if pending, it may have changed */
442 	llq->llq_ifp = ifp;
443 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
444 
445 	if (!pending)
446 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
447 
448 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
449 }
450 
451 /*
452  * Set the interface MAC address from a taskqueue to avoid a LOR.
453  */
454 static void
455 lagg_port_setlladdr(void *arg, int pending)
456 {
457 	struct lagg_softc *sc = (struct lagg_softc *)arg;
458 	struct lagg_llq *llq, *head;
459 	struct ifnet *ifp;
460 	int error;
461 
462 	/* Grab a local reference of the queue and remove it from the softc */
463 	LAGG_WLOCK(sc);
464 	head = SLIST_FIRST(&sc->sc_llq_head);
465 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
466 	LAGG_WUNLOCK(sc);
467 
468 	/*
469 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
470 	 * unlocked as we have the only reference to it.
471 	 */
472 	for (llq = head; llq != NULL; llq = head) {
473 		ifp = llq->llq_ifp;
474 
475 		/* Set the link layer address */
476 		error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
477 		if (error)
478 			printf("%s: setlladdr failed on %s\n", __func__,
479 			    ifp->if_xname);
480 
481 		head = SLIST_NEXT(llq, llq_entries);
482 		free(llq, M_DEVBUF);
483 	}
484 }
485 
486 static int
487 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
488 {
489 	struct lagg_softc *sc_ptr;
490 	struct lagg_port *lp;
491 	int error = 0;
492 
493 	LAGG_WLOCK_ASSERT(sc);
494 
495 	/* Limit the maximal number of lagg ports */
496 	if (sc->sc_count >= LAGG_MAX_PORTS)
497 		return (ENOSPC);
498 
499 	/* Check if port has already been associated to a lagg */
500 	if (ifp->if_lagg != NULL)
501 		return (EBUSY);
502 
503 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
504 	if (ifp->if_type != IFT_ETHER)
505 		return (EPROTONOSUPPORT);
506 
507 	/* Allow the first Ethernet member to define the MTU */
508 	if (SLIST_EMPTY(&sc->sc_ports))
509 		sc->sc_ifp->if_mtu = ifp->if_mtu;
510 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
511 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
512 		    ifp->if_xname);
513 		return (EINVAL);
514 	}
515 
516 	if ((lp = malloc(sizeof(struct lagg_port),
517 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
518 		return (ENOMEM);
519 
520 	/* Check if port is a stacked lagg */
521 	mtx_lock(&lagg_list_mtx);
522 	SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
523 		if (ifp == sc_ptr->sc_ifp) {
524 			mtx_unlock(&lagg_list_mtx);
525 			free(lp, M_DEVBUF);
526 			return (EINVAL);
527 			/* XXX disable stacking for the moment, its untested */
528 #ifdef LAGG_PORT_STACKING
529 			lp->lp_flags |= LAGG_PORT_STACK;
530 			if (lagg_port_checkstacking(sc_ptr) >=
531 			    LAGG_MAX_STACKING) {
532 				mtx_unlock(&lagg_list_mtx);
533 				free(lp, M_DEVBUF);
534 				return (E2BIG);
535 			}
536 #endif
537 		}
538 	}
539 	mtx_unlock(&lagg_list_mtx);
540 
541 	/* Change the interface type */
542 	lp->lp_iftype = ifp->if_type;
543 	ifp->if_type = IFT_IEEE8023ADLAG;
544 	ifp->if_lagg = lp;
545 	lp->lp_ioctl = ifp->if_ioctl;
546 	ifp->if_ioctl = lagg_port_ioctl;
547 	lp->lp_output = ifp->if_output;
548 	ifp->if_output = lagg_port_output;
549 
550 	lp->lp_ifp = ifp;
551 	lp->lp_softc = sc;
552 
553 	/* Save port link layer address */
554 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
555 
556 	if (SLIST_EMPTY(&sc->sc_ports)) {
557 		sc->sc_primary = lp;
558 		lagg_lladdr(sc, IF_LLADDR(ifp));
559 	} else {
560 		/* Update link layer address for this port */
561 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
562 	}
563 
564 	/* Insert into the list of ports */
565 	SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
566 	sc->sc_count++;
567 
568 	/* Update lagg capabilities */
569 	lagg_capabilities(sc);
570 	lagg_linkstate(sc);
571 
572 	/* Add multicast addresses and interface flags to this port */
573 	lagg_ether_cmdmulti(lp, 1);
574 	lagg_setflags(lp, 1);
575 
576 	if (sc->sc_port_create != NULL)
577 		error = (*sc->sc_port_create)(lp);
578 	if (error) {
579 		/* remove the port again, without calling sc_port_destroy */
580 		lagg_port_destroy(lp, 0);
581 		return (error);
582 	}
583 
584 	return (error);
585 }
586 
587 #ifdef LAGG_PORT_STACKING
588 static int
589 lagg_port_checkstacking(struct lagg_softc *sc)
590 {
591 	struct lagg_softc *sc_ptr;
592 	struct lagg_port *lp;
593 	int m = 0;
594 
595 	LAGG_WLOCK_ASSERT(sc);
596 
597 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
598 		if (lp->lp_flags & LAGG_PORT_STACK) {
599 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
600 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
601 		}
602 	}
603 
604 	return (m + 1);
605 }
606 #endif
607 
608 static int
609 lagg_port_destroy(struct lagg_port *lp, int runpd)
610 {
611 	struct lagg_softc *sc = lp->lp_softc;
612 	struct lagg_port *lp_ptr;
613 	struct lagg_llq *llq;
614 	struct ifnet *ifp = lp->lp_ifp;
615 
616 	LAGG_WLOCK_ASSERT(sc);
617 
618 	if (runpd && sc->sc_port_destroy != NULL)
619 		(*sc->sc_port_destroy)(lp);
620 
621 	/*
622 	 * Remove multicast addresses and interface flags from this port and
623 	 * reset the MAC address, skip if the interface is being detached.
624 	 */
625 	if (!lp->lp_detaching) {
626 		lagg_ether_cmdmulti(lp, 0);
627 		lagg_setflags(lp, 0);
628 		lagg_port_lladdr(lp, lp->lp_lladdr);
629 	}
630 
631 	/* Restore interface */
632 	ifp->if_type = lp->lp_iftype;
633 	ifp->if_ioctl = lp->lp_ioctl;
634 	ifp->if_output = lp->lp_output;
635 	ifp->if_lagg = NULL;
636 
637 	/* Finally, remove the port from the lagg */
638 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
639 	sc->sc_count--;
640 
641 	/* Update the primary interface */
642 	if (lp == sc->sc_primary) {
643 		uint8_t lladdr[ETHER_ADDR_LEN];
644 
645 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
646 			bzero(&lladdr, ETHER_ADDR_LEN);
647 		} else {
648 			bcopy(lp_ptr->lp_lladdr,
649 			    lladdr, ETHER_ADDR_LEN);
650 		}
651 		lagg_lladdr(sc, lladdr);
652 		sc->sc_primary = lp_ptr;
653 
654 		/* Update link layer address for each port */
655 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
656 			lagg_port_lladdr(lp_ptr, lladdr);
657 	}
658 
659 	/* Remove any pending lladdr changes from the queue */
660 	if (lp->lp_detaching) {
661 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
662 			if (llq->llq_ifp == ifp) {
663 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
664 				    llq_entries);
665 				free(llq, M_DEVBUF);
666 				break;	/* Only appears once */
667 			}
668 		}
669 	}
670 
671 	if (lp->lp_ifflags)
672 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
673 
674 	free(lp, M_DEVBUF);
675 
676 	/* Update lagg capabilities */
677 	lagg_capabilities(sc);
678 	lagg_linkstate(sc);
679 
680 	return (0);
681 }
682 
683 static int
684 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
685 {
686 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
687 	struct lagg_softc *sc;
688 	struct lagg_port *lp = NULL;
689 	int error = 0;
690 
691 	/* Should be checked by the caller */
692 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
693 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
694 		goto fallback;
695 
696 	switch (cmd) {
697 	case SIOCGLAGGPORT:
698 		if (rp->rp_portname[0] == '\0' ||
699 		    ifunit(rp->rp_portname) != ifp) {
700 			error = EINVAL;
701 			break;
702 		}
703 
704 		LAGG_RLOCK(sc);
705 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
706 			error = ENOENT;
707 			LAGG_RUNLOCK(sc);
708 			break;
709 		}
710 
711 		lagg_port2req(lp, rp);
712 		LAGG_RUNLOCK(sc);
713 		break;
714 
715 	case SIOCSIFCAP:
716 		if (lp->lp_ioctl == NULL) {
717 			error = EINVAL;
718 			break;
719 		}
720 		error = (*lp->lp_ioctl)(ifp, cmd, data);
721 		if (error)
722 			break;
723 
724 		/* Update lagg interface capabilities */
725 		LAGG_WLOCK(sc);
726 		lagg_capabilities(sc);
727 		LAGG_WUNLOCK(sc);
728 		break;
729 
730 	case SIOCSIFMTU:
731 		/* Do not allow the MTU to be changed once joined */
732 		error = EINVAL;
733 		break;
734 
735 	default:
736 		goto fallback;
737 	}
738 
739 	return (error);
740 
741 fallback:
742 	if (lp->lp_ioctl != NULL)
743 		return ((*lp->lp_ioctl)(ifp, cmd, data));
744 
745 	return (EINVAL);
746 }
747 
748 static int
749 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
750 	struct sockaddr *dst, struct route *ro)
751 {
752 	struct lagg_port *lp = ifp->if_lagg;
753 	struct ether_header *eh;
754 	short type = 0;
755 
756 	switch (dst->sa_family) {
757 		case pseudo_AF_HDRCMPLT:
758 		case AF_UNSPEC:
759 			eh = (struct ether_header *)dst->sa_data;
760 			type = eh->ether_type;
761 			break;
762 	}
763 
764 	/*
765 	 * Only allow ethernet types required to initiate or maintain the link,
766 	 * aggregated frames take a different path.
767 	 */
768 	switch (ntohs(type)) {
769 		case ETHERTYPE_PAE:	/* EAPOL PAE/802.1x */
770 			return ((*lp->lp_output)(ifp, m, dst, ro));
771 	}
772 
773 	/* drop any other frames */
774 	m_freem(m);
775 	return (EBUSY);
776 }
777 
778 static void
779 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
780 {
781 	struct lagg_port *lp;
782 	struct lagg_softc *sc;
783 
784 	if ((lp = ifp->if_lagg) == NULL)
785 		return;
786 
787 	sc = lp->lp_softc;
788 
789 	LAGG_WLOCK(sc);
790 	lp->lp_detaching = 1;
791 	lagg_port_destroy(lp, 1);
792 	LAGG_WUNLOCK(sc);
793 }
794 
795 static void
796 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
797 {
798 	struct lagg_softc *sc = lp->lp_softc;
799 
800 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
801 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
802 	rp->rp_prio = lp->lp_prio;
803 	rp->rp_flags = lp->lp_flags;
804 	if (sc->sc_portreq != NULL)
805 		(*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
806 
807 	/* Add protocol specific flags */
808 	switch (sc->sc_proto) {
809 		case LAGG_PROTO_FAILOVER:
810 			if (lp == sc->sc_primary)
811 				rp->rp_flags |= LAGG_PORT_MASTER;
812 			if (lp == lagg_link_active(sc, sc->sc_primary))
813 				rp->rp_flags |= LAGG_PORT_ACTIVE;
814 			break;
815 
816 		case LAGG_PROTO_ROUNDROBIN:
817 		case LAGG_PROTO_LOADBALANCE:
818 		case LAGG_PROTO_ETHERCHANNEL:
819 			if (LAGG_PORTACTIVE(lp))
820 				rp->rp_flags |= LAGG_PORT_ACTIVE;
821 			break;
822 
823 		case LAGG_PROTO_LACP:
824 			/* LACP has a different definition of active */
825 			if (lacp_isactive(lp))
826 				rp->rp_flags |= LAGG_PORT_ACTIVE;
827 			if (lacp_iscollecting(lp))
828 				rp->rp_flags |= LAGG_PORT_COLLECTING;
829 			if (lacp_isdistributing(lp))
830 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
831 			break;
832 	}
833 
834 }
835 
836 static void
837 lagg_init(void *xsc)
838 {
839 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
840 	struct lagg_port *lp;
841 	struct ifnet *ifp = sc->sc_ifp;
842 
843 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
844 		return;
845 
846 	LAGG_WLOCK(sc);
847 
848 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
849 	/* Update the port lladdrs */
850 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
851 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
852 
853 	if (sc->sc_init != NULL)
854 		(*sc->sc_init)(sc);
855 
856 	LAGG_WUNLOCK(sc);
857 }
858 
859 static void
860 lagg_stop(struct lagg_softc *sc)
861 {
862 	struct ifnet *ifp = sc->sc_ifp;
863 
864 	LAGG_WLOCK_ASSERT(sc);
865 
866 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
867 		return;
868 
869 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
870 
871 	if (sc->sc_stop != NULL)
872 		(*sc->sc_stop)(sc);
873 }
874 
875 static int
876 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
877 {
878 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
879 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
880 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
881 	struct ifreq *ifr = (struct ifreq *)data;
882 	struct lagg_port *lp;
883 	struct ifnet *tpif;
884 	struct thread *td = curthread;
885 	char *buf, *outbuf;
886 	int count, buflen, len, error = 0;
887 
888 	bzero(&rpbuf, sizeof(rpbuf));
889 
890 	switch (cmd) {
891 	case SIOCGLAGG:
892 		LAGG_RLOCK(sc);
893 		count = 0;
894 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
895 			count++;
896 		buflen = count * sizeof(struct lagg_reqport);
897 		LAGG_RUNLOCK(sc);
898 
899 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
900 
901 		LAGG_RLOCK(sc);
902 		ra->ra_proto = sc->sc_proto;
903 		if (sc->sc_req != NULL)
904 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
905 
906 		count = 0;
907 		buf = outbuf;
908 		len = min(ra->ra_size, buflen);
909 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
910 			if (len < sizeof(rpbuf))
911 				break;
912 
913 			lagg_port2req(lp, &rpbuf);
914 			memcpy(buf, &rpbuf, sizeof(rpbuf));
915 			count++;
916 			buf += sizeof(rpbuf);
917 			len -= sizeof(rpbuf);
918 		}
919 		LAGG_RUNLOCK(sc);
920 		ra->ra_ports = count;
921 		ra->ra_size = count * sizeof(rpbuf);
922 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
923 		free(outbuf, M_TEMP);
924 		break;
925 	case SIOCSLAGG:
926 		error = priv_check(td, PRIV_NET_LAGG);
927 		if (error)
928 			break;
929 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
930 			error = EPROTONOSUPPORT;
931 			break;
932 		}
933 		if (sc->sc_proto != LAGG_PROTO_NONE) {
934 			LAGG_WLOCK(sc);
935 			error = sc->sc_detach(sc);
936 			/* Reset protocol and pointers */
937 			sc->sc_proto = LAGG_PROTO_NONE;
938 			sc->sc_detach = NULL;
939 			sc->sc_start = NULL;
940 			sc->sc_input = NULL;
941 			sc->sc_port_create = NULL;
942 			sc->sc_port_destroy = NULL;
943 			sc->sc_linkstate = NULL;
944 			sc->sc_init = NULL;
945 			sc->sc_stop = NULL;
946 			sc->sc_lladdr = NULL;
947 			sc->sc_req = NULL;
948 			sc->sc_portreq = NULL;
949 			LAGG_WUNLOCK(sc);
950 		}
951 		if (error != 0)
952 			break;
953 		for (int i = 0; i < (sizeof(lagg_protos) /
954 		    sizeof(lagg_protos[0])); i++) {
955 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
956 				if (sc->sc_ifflags & IFF_DEBUG)
957 					printf("%s: using proto %u\n",
958 					    sc->sc_ifname,
959 					    lagg_protos[i].ti_proto);
960 				LAGG_WLOCK(sc);
961 				sc->sc_proto = lagg_protos[i].ti_proto;
962 				if (sc->sc_proto != LAGG_PROTO_NONE)
963 					error = lagg_protos[i].ti_attach(sc);
964 				LAGG_WUNLOCK(sc);
965 				return (error);
966 			}
967 		}
968 		error = EPROTONOSUPPORT;
969 		break;
970 	case SIOCGLAGGPORT:
971 		if (rp->rp_portname[0] == '\0' ||
972 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
973 			error = EINVAL;
974 			break;
975 		}
976 
977 		LAGG_RLOCK(sc);
978 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
979 		    lp->lp_softc != sc) {
980 			error = ENOENT;
981 			LAGG_RUNLOCK(sc);
982 			break;
983 		}
984 
985 		lagg_port2req(lp, rp);
986 		LAGG_RUNLOCK(sc);
987 		break;
988 	case SIOCSLAGGPORT:
989 		error = priv_check(td, PRIV_NET_LAGG);
990 		if (error)
991 			break;
992 		if (rp->rp_portname[0] == '\0' ||
993 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
994 			error = EINVAL;
995 			break;
996 		}
997 		LAGG_WLOCK(sc);
998 		error = lagg_port_create(sc, tpif);
999 		LAGG_WUNLOCK(sc);
1000 		break;
1001 	case SIOCSLAGGDELPORT:
1002 		error = priv_check(td, PRIV_NET_LAGG);
1003 		if (error)
1004 			break;
1005 		if (rp->rp_portname[0] == '\0' ||
1006 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1007 			error = EINVAL;
1008 			break;
1009 		}
1010 
1011 		LAGG_WLOCK(sc);
1012 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1013 		    lp->lp_softc != sc) {
1014 			error = ENOENT;
1015 			LAGG_WUNLOCK(sc);
1016 			break;
1017 		}
1018 
1019 		error = lagg_port_destroy(lp, 1);
1020 		LAGG_WUNLOCK(sc);
1021 		break;
1022 	case SIOCSIFFLAGS:
1023 		/* Set flags on ports too */
1024 		LAGG_WLOCK(sc);
1025 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1026 			lagg_setflags(lp, 1);
1027 		}
1028 		LAGG_WUNLOCK(sc);
1029 
1030 		if (!(ifp->if_flags & IFF_UP) &&
1031 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1032 			/*
1033 			 * If interface is marked down and it is running,
1034 			 * then stop and disable it.
1035 			 */
1036 			LAGG_WLOCK(sc);
1037 			lagg_stop(sc);
1038 			LAGG_WUNLOCK(sc);
1039 		} else if ((ifp->if_flags & IFF_UP) &&
1040 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1041 			/*
1042 			 * If interface is marked up and it is stopped, then
1043 			 * start it.
1044 			 */
1045 			(*ifp->if_init)(sc);
1046 		}
1047 		break;
1048 	case SIOCADDMULTI:
1049 	case SIOCDELMULTI:
1050 		LAGG_WLOCK(sc);
1051 		error = lagg_ether_setmulti(sc);
1052 		LAGG_WUNLOCK(sc);
1053 		break;
1054 	case SIOCSIFMEDIA:
1055 	case SIOCGIFMEDIA:
1056 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1057 		break;
1058 
1059 	case SIOCSIFCAP:
1060 	case SIOCSIFMTU:
1061 		/* Do not allow the MTU or caps to be directly changed */
1062 		error = EINVAL;
1063 		break;
1064 
1065 	default:
1066 		error = ether_ioctl(ifp, cmd, data);
1067 		break;
1068 	}
1069 	return (error);
1070 }
1071 
1072 static int
1073 lagg_ether_setmulti(struct lagg_softc *sc)
1074 {
1075 	struct lagg_port *lp;
1076 
1077 	LAGG_WLOCK_ASSERT(sc);
1078 
1079 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1080 		/* First, remove any existing filter entries. */
1081 		lagg_ether_cmdmulti(lp, 0);
1082 		/* copy all addresses from the lagg interface to the port */
1083 		lagg_ether_cmdmulti(lp, 1);
1084 	}
1085 	return (0);
1086 }
1087 
1088 static int
1089 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1090 {
1091 	struct lagg_softc *sc = lp->lp_softc;
1092 	struct ifnet *ifp = lp->lp_ifp;
1093 	struct ifnet *scifp = sc->sc_ifp;
1094 	struct lagg_mc *mc;
1095 	struct ifmultiaddr *ifma, *rifma = NULL;
1096 	struct sockaddr_dl sdl;
1097 	int error;
1098 
1099 	LAGG_WLOCK_ASSERT(sc);
1100 
1101 	bzero((char *)&sdl, sizeof(sdl));
1102 	sdl.sdl_len = sizeof(sdl);
1103 	sdl.sdl_family = AF_LINK;
1104 	sdl.sdl_type = IFT_ETHER;
1105 	sdl.sdl_alen = ETHER_ADDR_LEN;
1106 	sdl.sdl_index = ifp->if_index;
1107 
1108 	if (set) {
1109 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1110 			if (ifma->ifma_addr->sa_family != AF_LINK)
1111 				continue;
1112 			bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1113 			    LLADDR(&sdl), ETHER_ADDR_LEN);
1114 
1115 			error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1116 			if (error)
1117 				return (error);
1118 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1119 			if (mc == NULL)
1120 				return (ENOMEM);
1121 			mc->mc_ifma = rifma;
1122 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1123 		}
1124 	} else {
1125 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1126 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1127 			if_delmulti_ifma(mc->mc_ifma);
1128 			free(mc, M_DEVBUF);
1129 		}
1130 	}
1131 	return (0);
1132 }
1133 
1134 /* Handle a ref counted flag that should be set on the lagg port as well */
1135 static int
1136 lagg_setflag(struct lagg_port *lp, int flag, int status,
1137 	     int (*func)(struct ifnet *, int))
1138 {
1139 	struct lagg_softc *sc = lp->lp_softc;
1140 	struct ifnet *scifp = sc->sc_ifp;
1141 	struct ifnet *ifp = lp->lp_ifp;
1142 	int error;
1143 
1144 	LAGG_WLOCK_ASSERT(sc);
1145 
1146 	status = status ? (scifp->if_flags & flag) : 0;
1147 	/* Now "status" contains the flag value or 0 */
1148 
1149 	/*
1150 	 * See if recorded ports status is different from what
1151 	 * we want it to be.  If it is, flip it.  We record ports
1152 	 * status in lp_ifflags so that we won't clear ports flag
1153 	 * we haven't set.  In fact, we don't clear or set ports
1154 	 * flags directly, but get or release references to them.
1155 	 * That's why we can be sure that recorded flags still are
1156 	 * in accord with actual ports flags.
1157 	 */
1158 	if (status != (lp->lp_ifflags & flag)) {
1159 		error = (*func)(ifp, status);
1160 		if (error)
1161 			return (error);
1162 		lp->lp_ifflags &= ~flag;
1163 		lp->lp_ifflags |= status;
1164 	}
1165 	return (0);
1166 }
1167 
1168 /*
1169  * Handle IFF_* flags that require certain changes on the lagg port
1170  * if "status" is true, update ports flags respective to the lagg
1171  * if "status" is false, forcedly clear the flags set on port.
1172  */
1173 static int
1174 lagg_setflags(struct lagg_port *lp, int status)
1175 {
1176 	int error, i;
1177 
1178 	for (i = 0; lagg_pflags[i].flag; i++) {
1179 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1180 		    status, lagg_pflags[i].func);
1181 		if (error)
1182 			return (error);
1183 	}
1184 	return (0);
1185 }
1186 
1187 static void
1188 lagg_start(struct ifnet *ifp)
1189 {
1190 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1191 	struct mbuf *m;
1192 	int error = 0;
1193 
1194 	LAGG_RLOCK(sc);
1195 	/* We need a Tx algorithm and at least one port */
1196 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1197 		IF_DRAIN(&ifp->if_snd);
1198 		LAGG_RUNLOCK(sc);
1199 		return;
1200 	}
1201 
1202 	for (;; error = 0) {
1203 		IFQ_DEQUEUE(&ifp->if_snd, m);
1204 		if (m == NULL)
1205 			break;
1206 
1207 		ETHER_BPF_MTAP(ifp, m);
1208 
1209 		error = (*sc->sc_start)(sc, m);
1210 		if (error == 0)
1211 			ifp->if_opackets++;
1212 		else
1213 			ifp->if_oerrors++;
1214 	}
1215 	LAGG_RUNLOCK(sc);
1216 }
1217 
1218 static struct mbuf *
1219 lagg_input(struct ifnet *ifp, struct mbuf *m)
1220 {
1221 	struct lagg_port *lp = ifp->if_lagg;
1222 	struct lagg_softc *sc = lp->lp_softc;
1223 	struct ifnet *scifp = sc->sc_ifp;
1224 
1225 	LAGG_RLOCK(sc);
1226 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1227 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1228 	    sc->sc_proto == LAGG_PROTO_NONE) {
1229 		LAGG_RUNLOCK(sc);
1230 		m_freem(m);
1231 		return (NULL);
1232 	}
1233 
1234 	ETHER_BPF_MTAP(scifp, m);
1235 
1236 	m = (*sc->sc_input)(sc, lp, m);
1237 
1238 	if (m != NULL) {
1239 		scifp->if_ipackets++;
1240 		scifp->if_ibytes += m->m_pkthdr.len;
1241 
1242 		if (scifp->if_flags & IFF_MONITOR) {
1243 			m_freem(m);
1244 			m = NULL;
1245 		}
1246 	}
1247 
1248 	LAGG_RUNLOCK(sc);
1249 	return (m);
1250 }
1251 
1252 static int
1253 lagg_media_change(struct ifnet *ifp)
1254 {
1255 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1256 
1257 	if (sc->sc_ifflags & IFF_DEBUG)
1258 		printf("%s\n", __func__);
1259 
1260 	/* Ignore */
1261 	return (0);
1262 }
1263 
1264 static void
1265 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1266 {
1267 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1268 	struct lagg_port *lp;
1269 
1270 	imr->ifm_status = IFM_AVALID;
1271 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1272 
1273 	LAGG_RLOCK(sc);
1274 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1275 		if (LAGG_PORTACTIVE(lp))
1276 			imr->ifm_status |= IFM_ACTIVE;
1277 	}
1278 	LAGG_RUNLOCK(sc);
1279 }
1280 
1281 static void
1282 lagg_linkstate(struct lagg_softc *sc)
1283 {
1284 	struct lagg_port *lp;
1285 	int new_link = LINK_STATE_DOWN;
1286 	uint64_t speed;
1287 
1288 	/* Our link is considered up if at least one of our ports is active */
1289 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1290 		if (lp->lp_link_state == LINK_STATE_UP) {
1291 			new_link = LINK_STATE_UP;
1292 			break;
1293 		}
1294 	}
1295 	if_link_state_change(sc->sc_ifp, new_link);
1296 
1297 	/* Update if_baudrate to reflect the max possible speed */
1298 	switch (sc->sc_proto) {
1299 		case LAGG_PROTO_FAILOVER:
1300 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1301 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1302 			break;
1303 		case LAGG_PROTO_ROUNDROBIN:
1304 		case LAGG_PROTO_LOADBALANCE:
1305 		case LAGG_PROTO_ETHERCHANNEL:
1306 			speed = 0;
1307 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1308 				speed += lp->lp_ifp->if_baudrate;
1309 			sc->sc_ifp->if_baudrate = speed;
1310 			break;
1311 		case LAGG_PROTO_LACP:
1312 			/* LACP updates if_baudrate itself */
1313 			break;
1314 	}
1315 }
1316 
1317 static void
1318 lagg_port_state(struct ifnet *ifp, int state)
1319 {
1320 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1321 	struct lagg_softc *sc = NULL;
1322 
1323 	if (lp != NULL)
1324 		sc = lp->lp_softc;
1325 	if (sc == NULL)
1326 		return;
1327 
1328 	LAGG_WLOCK(sc);
1329 	lagg_linkstate(sc);
1330 	if (sc->sc_linkstate != NULL)
1331 		(*sc->sc_linkstate)(lp);
1332 	LAGG_WUNLOCK(sc);
1333 }
1334 
1335 struct lagg_port *
1336 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1337 {
1338 	struct lagg_port *lp_next, *rval = NULL;
1339 	// int new_link = LINK_STATE_DOWN;
1340 
1341 	LAGG_RLOCK_ASSERT(sc);
1342 	/*
1343 	 * Search a port which reports an active link state.
1344 	 */
1345 
1346 	if (lp == NULL)
1347 		goto search;
1348 	if (LAGG_PORTACTIVE(lp)) {
1349 		rval = lp;
1350 		goto found;
1351 	}
1352 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1353 	    LAGG_PORTACTIVE(lp_next)) {
1354 		rval = lp_next;
1355 		goto found;
1356 	}
1357 
1358 search:
1359 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1360 		if (LAGG_PORTACTIVE(lp_next)) {
1361 			rval = lp_next;
1362 			goto found;
1363 		}
1364 	}
1365 
1366 found:
1367 	if (rval != NULL) {
1368 		/*
1369 		 * The IEEE 802.1D standard assumes that a lagg with
1370 		 * multiple ports is always full duplex. This is valid
1371 		 * for load sharing laggs and if at least two links
1372 		 * are active. Unfortunately, checking the latter would
1373 		 * be too expensive at this point.
1374 		 XXX
1375 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1376 		    (sc->sc_count > 1))
1377 			new_link = LINK_STATE_FULL_DUPLEX;
1378 		else
1379 			new_link = rval->lp_link_state;
1380 		 */
1381 	}
1382 
1383 	return (rval);
1384 }
1385 
1386 static const void *
1387 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1388 {
1389 	if (m->m_pkthdr.len < (off + len)) {
1390 		return (NULL);
1391 	} else if (m->m_len < (off + len)) {
1392 		m_copydata(m, off, len, buf);
1393 		return (buf);
1394 	}
1395 	return (mtod(m, char *) + off);
1396 }
1397 
1398 uint32_t
1399 lagg_hashmbuf(struct mbuf *m, uint32_t key)
1400 {
1401 	uint16_t etype;
1402 	uint32_t p = 0;
1403 	int off;
1404 	struct ether_header *eh;
1405 	struct ether_vlan_header vlanbuf;
1406 	const struct ether_vlan_header *vlan;
1407 #ifdef INET
1408 	const struct ip *ip;
1409 	struct ip ipbuf;
1410 #endif
1411 #ifdef INET6
1412 	const struct ip6_hdr *ip6;
1413 	struct ip6_hdr ip6buf;
1414 	uint32_t flow;
1415 #endif
1416 
1417 	off = sizeof(*eh);
1418 	if (m->m_len < off)
1419 		goto out;
1420 	eh = mtod(m, struct ether_header *);
1421 	etype = ntohs(eh->ether_type);
1422 	p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, key);
1423 	p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1424 
1425 	/* Special handling for encapsulating VLAN frames */
1426 	if (m->m_flags & M_VLANTAG) {
1427 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1428 		    sizeof(m->m_pkthdr.ether_vtag), p);
1429 	} else if (etype == ETHERTYPE_VLAN) {
1430 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &vlanbuf);
1431 		if (vlan == NULL)
1432 			goto out;
1433 
1434 		p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1435 		etype = ntohs(vlan->evl_proto);
1436 		off += sizeof(*vlan) - sizeof(*eh);
1437 	}
1438 
1439 	switch (etype) {
1440 #ifdef INET
1441 	case ETHERTYPE_IP:
1442 		ip = lagg_gethdr(m, off, sizeof(*ip), &ipbuf);
1443 		if (ip == NULL)
1444 			goto out;
1445 
1446 		p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1447 		p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1448 		break;
1449 #endif
1450 #ifdef INET6
1451 	case ETHERTYPE_IPV6:
1452 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &ip6buf);
1453 		if (ip6 == NULL)
1454 			goto out;
1455 
1456 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1457 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1458 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1459 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1460 		break;
1461 #endif
1462 	}
1463 out:
1464 	return (p);
1465 }
1466 
1467 int
1468 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1469 {
1470 
1471 	return (ifp->if_transmit)(ifp, m);
1472 }
1473 
1474 /*
1475  * Simple round robin aggregation
1476  */
1477 
1478 static int
1479 lagg_rr_attach(struct lagg_softc *sc)
1480 {
1481 	sc->sc_detach = lagg_rr_detach;
1482 	sc->sc_start = lagg_rr_start;
1483 	sc->sc_input = lagg_rr_input;
1484 	sc->sc_port_create = NULL;
1485 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1486 	sc->sc_seq = 0;
1487 
1488 	return (0);
1489 }
1490 
1491 static int
1492 lagg_rr_detach(struct lagg_softc *sc)
1493 {
1494 	return (0);
1495 }
1496 
1497 static int
1498 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1499 {
1500 	struct lagg_port *lp;
1501 	uint32_t p;
1502 
1503 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1504 	p %= sc->sc_count;
1505 	lp = SLIST_FIRST(&sc->sc_ports);
1506 	while (p--)
1507 		lp = SLIST_NEXT(lp, lp_entries);
1508 
1509 	/*
1510 	 * Check the port's link state. This will return the next active
1511 	 * port if the link is down or the port is NULL.
1512 	 */
1513 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1514 		m_freem(m);
1515 		return (ENOENT);
1516 	}
1517 
1518 	/* Send mbuf */
1519 	return (lagg_enqueue(lp->lp_ifp, m));
1520 }
1521 
1522 static struct mbuf *
1523 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1524 {
1525 	struct ifnet *ifp = sc->sc_ifp;
1526 
1527 	/* Just pass in the packet to our lagg device */
1528 	m->m_pkthdr.rcvif = ifp;
1529 
1530 	return (m);
1531 }
1532 
1533 /*
1534  * Active failover
1535  */
1536 
1537 static int
1538 lagg_fail_attach(struct lagg_softc *sc)
1539 {
1540 	sc->sc_detach = lagg_fail_detach;
1541 	sc->sc_start = lagg_fail_start;
1542 	sc->sc_input = lagg_fail_input;
1543 	sc->sc_port_create = NULL;
1544 	sc->sc_port_destroy = NULL;
1545 
1546 	return (0);
1547 }
1548 
1549 static int
1550 lagg_fail_detach(struct lagg_softc *sc)
1551 {
1552 	return (0);
1553 }
1554 
1555 static int
1556 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1557 {
1558 	struct lagg_port *lp;
1559 
1560 	/* Use the master port if active or the next available port */
1561 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1562 		m_freem(m);
1563 		return (ENOENT);
1564 	}
1565 
1566 	/* Send mbuf */
1567 	return (lagg_enqueue(lp->lp_ifp, m));
1568 }
1569 
1570 static struct mbuf *
1571 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1572 {
1573 	struct ifnet *ifp = sc->sc_ifp;
1574 	struct lagg_port *tmp_tp;
1575 
1576 	if (lp == sc->sc_primary || lagg_failover_rx_all) {
1577 		m->m_pkthdr.rcvif = ifp;
1578 		return (m);
1579 	}
1580 
1581 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1582 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1583 		/*
1584 		 * If tmp_tp is null, we've recieved a packet when all
1585 		 * our links are down. Weird, but process it anyways.
1586 		 */
1587 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1588 			m->m_pkthdr.rcvif = ifp;
1589 			return (m);
1590 		}
1591 	}
1592 
1593 	m_freem(m);
1594 	return (NULL);
1595 }
1596 
1597 /*
1598  * Loadbalancing
1599  */
1600 
1601 static int
1602 lagg_lb_attach(struct lagg_softc *sc)
1603 {
1604 	struct lagg_port *lp;
1605 	struct lagg_lb *lb;
1606 
1607 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1608 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1609 		return (ENOMEM);
1610 
1611 	sc->sc_detach = lagg_lb_detach;
1612 	sc->sc_start = lagg_lb_start;
1613 	sc->sc_input = lagg_lb_input;
1614 	sc->sc_port_create = lagg_lb_port_create;
1615 	sc->sc_port_destroy = lagg_lb_port_destroy;
1616 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1617 
1618 	lb->lb_key = arc4random();
1619 	sc->sc_psc = (caddr_t)lb;
1620 
1621 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1622 		lagg_lb_port_create(lp);
1623 
1624 	return (0);
1625 }
1626 
1627 static int
1628 lagg_lb_detach(struct lagg_softc *sc)
1629 {
1630 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1631 	if (lb != NULL)
1632 		free(lb, M_DEVBUF);
1633 	return (0);
1634 }
1635 
1636 static int
1637 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1638 {
1639 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1640 	struct lagg_port *lp_next;
1641 	int i = 0;
1642 
1643 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1644 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1645 		if (lp_next == lp)
1646 			continue;
1647 		if (i >= LAGG_MAX_PORTS)
1648 			return (EINVAL);
1649 		if (sc->sc_ifflags & IFF_DEBUG)
1650 			printf("%s: port %s at index %d\n",
1651 			    sc->sc_ifname, lp_next->lp_ifname, i);
1652 		lb->lb_ports[i++] = lp_next;
1653 	}
1654 
1655 	return (0);
1656 }
1657 
1658 static int
1659 lagg_lb_port_create(struct lagg_port *lp)
1660 {
1661 	struct lagg_softc *sc = lp->lp_softc;
1662 	return (lagg_lb_porttable(sc, NULL));
1663 }
1664 
1665 static void
1666 lagg_lb_port_destroy(struct lagg_port *lp)
1667 {
1668 	struct lagg_softc *sc = lp->lp_softc;
1669 	lagg_lb_porttable(sc, lp);
1670 }
1671 
1672 static int
1673 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1674 {
1675 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1676 	struct lagg_port *lp = NULL;
1677 	uint32_t p = 0;
1678 
1679 	if (m->m_flags & M_FLOWID)
1680 		p = m->m_pkthdr.flowid;
1681 	else
1682 		p = lagg_hashmbuf(m, lb->lb_key);
1683 	p %= sc->sc_count;
1684 	lp = lb->lb_ports[p];
1685 
1686 	/*
1687 	 * Check the port's link state. This will return the next active
1688 	 * port if the link is down or the port is NULL.
1689 	 */
1690 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1691 		m_freem(m);
1692 		return (ENOENT);
1693 	}
1694 
1695 	/* Send mbuf */
1696 	return (lagg_enqueue(lp->lp_ifp, m));
1697 }
1698 
1699 static struct mbuf *
1700 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1701 {
1702 	struct ifnet *ifp = sc->sc_ifp;
1703 
1704 	/* Just pass in the packet to our lagg device */
1705 	m->m_pkthdr.rcvif = ifp;
1706 
1707 	return (m);
1708 }
1709 
1710 /*
1711  * 802.3ad LACP
1712  */
1713 
1714 static int
1715 lagg_lacp_attach(struct lagg_softc *sc)
1716 {
1717 	struct lagg_port *lp;
1718 	int error;
1719 
1720 	sc->sc_detach = lagg_lacp_detach;
1721 	sc->sc_port_create = lacp_port_create;
1722 	sc->sc_port_destroy = lacp_port_destroy;
1723 	sc->sc_linkstate = lacp_linkstate;
1724 	sc->sc_start = lagg_lacp_start;
1725 	sc->sc_input = lagg_lacp_input;
1726 	sc->sc_init = lacp_init;
1727 	sc->sc_stop = lacp_stop;
1728 	sc->sc_lladdr = lagg_lacp_lladdr;
1729 	sc->sc_req = lacp_req;
1730 	sc->sc_portreq = lacp_portreq;
1731 
1732 	error = lacp_attach(sc);
1733 	if (error)
1734 		return (error);
1735 
1736 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1737 		lacp_port_create(lp);
1738 
1739 	return (error);
1740 }
1741 
1742 static int
1743 lagg_lacp_detach(struct lagg_softc *sc)
1744 {
1745 	struct lagg_port *lp;
1746 	int error;
1747 
1748 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1749 		lacp_port_destroy(lp);
1750 
1751 	/* unlocking is safe here */
1752 	LAGG_WUNLOCK(sc);
1753 	error = lacp_detach(sc);
1754 	LAGG_WLOCK(sc);
1755 
1756 	return (error);
1757 }
1758 
1759 static void
1760 lagg_lacp_lladdr(struct lagg_softc *sc)
1761 {
1762 	struct lagg_port *lp;
1763 
1764 	/* purge all the lacp ports */
1765 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1766 		lacp_port_destroy(lp);
1767 
1768 	/* add them back in */
1769 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1770 		lacp_port_create(lp);
1771 }
1772 
1773 static int
1774 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1775 {
1776 	struct lagg_port *lp;
1777 
1778 	lp = lacp_select_tx_port(sc, m);
1779 	if (lp == NULL) {
1780 		m_freem(m);
1781 		return (EBUSY);
1782 	}
1783 
1784 	/* Send mbuf */
1785 	return (lagg_enqueue(lp->lp_ifp, m));
1786 }
1787 
1788 static struct mbuf *
1789 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1790 {
1791 	struct ifnet *ifp = sc->sc_ifp;
1792 	struct ether_header *eh;
1793 	u_short etype;
1794 
1795 	eh = mtod(m, struct ether_header *);
1796 	etype = ntohs(eh->ether_type);
1797 
1798 	/* Tap off LACP control messages */
1799 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
1800 		m = lacp_input(lp, m);
1801 		if (m == NULL)
1802 			return (NULL);
1803 	}
1804 
1805 	/*
1806 	 * If the port is not collecting or not in the active aggregator then
1807 	 * free and return.
1808 	 */
1809 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
1810 		m_freem(m);
1811 		return (NULL);
1812 	}
1813 
1814 	m->m_pkthdr.rcvif = ifp;
1815 	return (m);
1816 }
1817