xref: /freebsd/sys/net/if_lagg.c (revision de7b456e596ff18032d2cbfdf244c66f36770da4)
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/rmlock.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 #include <netinet6/in6_var.h>
67 #include <netinet6/in6_ifattach.h>
68 #endif
69 
70 #include <net/if_vlan_var.h>
71 #include <net/if_lagg.h>
72 #include <net/ieee8023ad_lacp.h>
73 
74 /* Special flags we should propagate to the lagg ports. */
75 static struct {
76 	int flag;
77 	int (*func)(struct ifnet *, int);
78 } lagg_pflags[] = {
79 	{IFF_PROMISC, ifpromisc},
80 	{IFF_ALLMULTI, if_allmulti},
81 	{0, NULL}
82 };
83 
84 SLIST_HEAD(__trhead, lagg_softc) lagg_list;	/* list of laggs */
85 static struct mtx	lagg_list_mtx;
86 eventhandler_tag	lagg_detach_cookie = NULL;
87 
88 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
89 static void	lagg_clone_destroy(struct ifnet *);
90 static struct if_clone *lagg_cloner;
91 static const char laggname[] = "lagg";
92 
93 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
94 static void	lagg_capabilities(struct lagg_softc *);
95 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
96 static void	lagg_port_setlladdr(void *, int);
97 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
98 static int	lagg_port_destroy(struct lagg_port *, int);
99 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
100 static void	lagg_linkstate(struct lagg_softc *);
101 static void	lagg_port_state(struct ifnet *, int);
102 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
103 static int	lagg_port_output(struct ifnet *, struct mbuf *,
104 		    const struct sockaddr *, struct route *);
105 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
106 #ifdef LAGG_PORT_STACKING
107 static int	lagg_port_checkstacking(struct lagg_softc *);
108 #endif
109 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
110 static void	lagg_init(void *);
111 static void	lagg_stop(struct lagg_softc *);
112 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
113 static int	lagg_ether_setmulti(struct lagg_softc *);
114 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
115 static	int	lagg_setflag(struct lagg_port *, int, int,
116 		    int (*func)(struct ifnet *, int));
117 static	int	lagg_setflags(struct lagg_port *, int status);
118 static int	lagg_transmit(struct ifnet *, struct mbuf *);
119 static void	lagg_qflush(struct ifnet *);
120 static int	lagg_media_change(struct ifnet *);
121 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
122 static struct lagg_port *lagg_link_active(struct lagg_softc *,
123 	    struct lagg_port *);
124 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
125 static int	lagg_sysctl_active(SYSCTL_HANDLER_ARGS);
126 
127 /* Simple round robin */
128 static int	lagg_rr_attach(struct lagg_softc *);
129 static int	lagg_rr_detach(struct lagg_softc *);
130 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
131 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
132 		    struct mbuf *);
133 
134 /* Active failover */
135 static int	lagg_fail_attach(struct lagg_softc *);
136 static int	lagg_fail_detach(struct lagg_softc *);
137 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
138 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
139 		    struct mbuf *);
140 
141 /* Loadbalancing */
142 static int	lagg_lb_attach(struct lagg_softc *);
143 static int	lagg_lb_detach(struct lagg_softc *);
144 static int	lagg_lb_port_create(struct lagg_port *);
145 static void	lagg_lb_port_destroy(struct lagg_port *);
146 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
147 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
148 		    struct mbuf *);
149 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
150 
151 /* 802.3ad LACP */
152 static int	lagg_lacp_attach(struct lagg_softc *);
153 static int	lagg_lacp_detach(struct lagg_softc *);
154 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
155 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
156 		    struct mbuf *);
157 static void	lagg_lacp_lladdr(struct lagg_softc *);
158 
159 static void	lagg_callout(void *);
160 
161 /* lagg protocol table */
162 static const struct {
163 	int			ti_proto;
164 	int			(*ti_attach)(struct lagg_softc *);
165 } lagg_protos[] = {
166 	{ LAGG_PROTO_ROUNDROBIN,	lagg_rr_attach },
167 	{ LAGG_PROTO_FAILOVER,		lagg_fail_attach },
168 	{ LAGG_PROTO_LOADBALANCE,	lagg_lb_attach },
169 	{ LAGG_PROTO_ETHERCHANNEL,	lagg_lb_attach },
170 	{ LAGG_PROTO_LACP,		lagg_lacp_attach },
171 	{ LAGG_PROTO_NONE,		NULL }
172 };
173 
174 SYSCTL_DECL(_net_link);
175 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
176     "Link Aggregation");
177 
178 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */
179 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW,
180     &lagg_failover_rx_all, 0,
181     "Accept input from any interface in a failover lagg");
182 static int def_use_flowid = 1; /* Default value for using M_FLOWID */
183 TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid);
184 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RW,
185     &def_use_flowid, 0,
186     "Default setting for using flow id for load sharing");
187 
188 static int
189 lagg_modevent(module_t mod, int type, void *data)
190 {
191 
192 	switch (type) {
193 	case MOD_LOAD:
194 		mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
195 		SLIST_INIT(&lagg_list);
196 		lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
197 		    lagg_clone_destroy, 0);
198 		lagg_input_p = lagg_input;
199 		lagg_linkstate_p = lagg_port_state;
200 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
201 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
202 		    EVENTHANDLER_PRI_ANY);
203 		break;
204 	case MOD_UNLOAD:
205 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
206 		    lagg_detach_cookie);
207 		if_clone_detach(lagg_cloner);
208 		lagg_input_p = NULL;
209 		lagg_linkstate_p = NULL;
210 		mtx_destroy(&lagg_list_mtx);
211 		break;
212 	default:
213 		return (EOPNOTSUPP);
214 	}
215 	return (0);
216 }
217 
218 static moduledata_t lagg_mod = {
219 	"if_lagg",
220 	lagg_modevent,
221 	0
222 };
223 
224 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
225 MODULE_VERSION(if_lagg, 1);
226 
227 /*
228  * This routine is run via an vlan
229  * config EVENT
230  */
231 static void
232 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
233 {
234         struct lagg_softc       *sc = ifp->if_softc;
235         struct lagg_port        *lp;
236         struct rm_priotracker   tracker;
237 
238         if (ifp->if_softc !=  arg)   /* Not our event */
239                 return;
240 
241         LAGG_RLOCK(sc, &tracker);
242         if (!SLIST_EMPTY(&sc->sc_ports)) {
243                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
244                         EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
245         }
246         LAGG_RUNLOCK(sc, &tracker);
247 }
248 
249 /*
250  * This routine is run via an vlan
251  * unconfig EVENT
252  */
253 static void
254 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
255 {
256         struct lagg_softc       *sc = ifp->if_softc;
257         struct lagg_port        *lp;
258         struct rm_priotracker   tracker;
259 
260         if (ifp->if_softc !=  arg)   /* Not our event */
261                 return;
262 
263         LAGG_RLOCK(sc, &tracker);
264         if (!SLIST_EMPTY(&sc->sc_ports)) {
265                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
266                         EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
267         }
268         LAGG_RUNLOCK(sc, &tracker);
269 }
270 
271 static int
272 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
273 {
274 	struct lagg_softc *sc;
275 	struct ifnet *ifp;
276 	int i, error = 0;
277 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
278 	struct sysctl_oid *oid;
279 	char num[14];			/* sufficient for 32 bits */
280 
281 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
282 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
283 	if (ifp == NULL) {
284 		free(sc, M_DEVBUF);
285 		return (ENOSPC);
286 	}
287 
288 	sc->sc_ipackets = counter_u64_alloc(M_WAITOK);
289 	sc->sc_opackets = counter_u64_alloc(M_WAITOK);
290 	sc->sc_ibytes = counter_u64_alloc(M_WAITOK);
291 	sc->sc_obytes = counter_u64_alloc(M_WAITOK);
292 
293 	sysctl_ctx_init(&sc->ctx);
294 	snprintf(num, sizeof(num), "%u", unit);
295 	sc->use_flowid = def_use_flowid;
296 	sc->sc_oid = oid = SYSCTL_ADD_NODE(&sc->ctx,
297 		&SYSCTL_NODE_CHILDREN(_net_link, lagg),
298 		OID_AUTO, num, CTLFLAG_RD, NULL, "");
299 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
300 		"use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid, sc->use_flowid,
301 		"Use flow id for load sharing");
302 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
303 		"count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count,
304 		"Total number of ports");
305 	SYSCTL_ADD_PROC(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
306 		"active", CTLTYPE_INT|CTLFLAG_RD, sc, 0, lagg_sysctl_active,
307 		"I", "Total number of active ports");
308 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
309 		"flapping", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_flapping,
310 		sc->sc_flapping, "Total number of port change events");
311 	/* Hash all layers by default */
312 	sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4;
313 
314 	sc->sc_proto = LAGG_PROTO_NONE;
315 	for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
316 		if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
317 			sc->sc_proto = lagg_protos[i].ti_proto;
318 			if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
319 				if_free(ifp);
320 				free(sc, M_DEVBUF);
321 				return (error);
322 			}
323 			break;
324 		}
325 	}
326 	LAGG_LOCK_INIT(sc);
327 	LAGG_CALLOUT_LOCK_INIT(sc);
328 	SLIST_INIT(&sc->sc_ports);
329 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
330 
331 	/*
332 	 * This uses the callout lock rather than the rmlock; one can't
333 	 * hold said rmlock during SWI.
334 	 */
335 	callout_init_mtx(&sc->sc_callout, &sc->sc_call_mtx, 0);
336 
337 	/* Initialise pseudo media types */
338 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
339 	    lagg_media_status);
340 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
341 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
342 
343 	if_initname(ifp, laggname, unit);
344 	ifp->if_softc = sc;
345 	ifp->if_transmit = lagg_transmit;
346 	ifp->if_qflush = lagg_qflush;
347 	ifp->if_init = lagg_init;
348 	ifp->if_ioctl = lagg_ioctl;
349 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
350 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
351 
352 	/*
353 	 * Attach as an ordinary ethernet device, children will be attached
354 	 * as special device IFT_IEEE8023ADLAG.
355 	 */
356 	ether_ifattach(ifp, eaddr);
357 
358 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
359 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
360 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
361 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
362 
363 	/* Insert into the global list of laggs */
364 	mtx_lock(&lagg_list_mtx);
365 	SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
366 	mtx_unlock(&lagg_list_mtx);
367 
368 	callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
369 
370 	return (0);
371 }
372 
373 static void
374 lagg_clone_destroy(struct ifnet *ifp)
375 {
376 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
377 	struct lagg_port *lp;
378 
379 	LAGG_WLOCK(sc);
380 
381 	lagg_stop(sc);
382 	ifp->if_flags &= ~IFF_UP;
383 
384 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
385 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
386 
387 	/* Shutdown and remove lagg ports */
388 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
389 		lagg_port_destroy(lp, 1);
390 	/* Unhook the aggregation protocol */
391 	if (sc->sc_detach != NULL)
392 		(*sc->sc_detach)(sc);
393 
394 	LAGG_WUNLOCK(sc);
395 
396 	sysctl_ctx_free(&sc->ctx);
397 	ifmedia_removeall(&sc->sc_media);
398 	ether_ifdetach(ifp);
399 	if_free(ifp);
400 
401 	/* This grabs sc_callout_mtx, serialising it correctly */
402 	callout_drain(&sc->sc_callout);
403 
404 	/* At this point it's drained; we can free this */
405 	counter_u64_free(sc->sc_ipackets);
406 	counter_u64_free(sc->sc_opackets);
407 	counter_u64_free(sc->sc_ibytes);
408 	counter_u64_free(sc->sc_obytes);
409 
410 	mtx_lock(&lagg_list_mtx);
411 	SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
412 	mtx_unlock(&lagg_list_mtx);
413 
414 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
415 	LAGG_LOCK_DESTROY(sc);
416 	LAGG_CALLOUT_LOCK_DESTROY(sc);
417 	free(sc, M_DEVBUF);
418 }
419 
420 static void
421 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
422 {
423 	struct ifnet *ifp = sc->sc_ifp;
424 
425 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
426 		return;
427 
428 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
429 	/* Let the protocol know the MAC has changed */
430 	if (sc->sc_lladdr != NULL)
431 		(*sc->sc_lladdr)(sc);
432 	EVENTHANDLER_INVOKE(iflladdr_event, ifp);
433 }
434 
435 static void
436 lagg_capabilities(struct lagg_softc *sc)
437 {
438 	struct lagg_port *lp;
439 	int cap = ~0, ena = ~0;
440 	u_long hwa = ~0UL;
441 
442 	LAGG_WLOCK_ASSERT(sc);
443 
444 	/* Get capabilities from the lagg ports */
445 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
446 		cap &= lp->lp_ifp->if_capabilities;
447 		ena &= lp->lp_ifp->if_capenable;
448 		hwa &= lp->lp_ifp->if_hwassist;
449 	}
450 	cap = (cap == ~0 ? 0 : cap);
451 	ena = (ena == ~0 ? 0 : ena);
452 	hwa = (hwa == ~0 ? 0 : hwa);
453 
454 	if (sc->sc_ifp->if_capabilities != cap ||
455 	    sc->sc_ifp->if_capenable != ena ||
456 	    sc->sc_ifp->if_hwassist != hwa) {
457 		sc->sc_ifp->if_capabilities = cap;
458 		sc->sc_ifp->if_capenable = ena;
459 		sc->sc_ifp->if_hwassist = hwa;
460 		getmicrotime(&sc->sc_ifp->if_lastchange);
461 
462 		if (sc->sc_ifflags & IFF_DEBUG)
463 			if_printf(sc->sc_ifp,
464 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
465 	}
466 }
467 
468 static void
469 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
470 {
471 	struct lagg_softc *sc = lp->lp_softc;
472 	struct ifnet *ifp = lp->lp_ifp;
473 	struct lagg_llq *llq;
474 	int pending = 0;
475 
476 	LAGG_WLOCK_ASSERT(sc);
477 
478 	if (lp->lp_detaching ||
479 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
480 		return;
481 
482 	/* Check to make sure its not already queued to be changed */
483 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
484 		if (llq->llq_ifp == ifp) {
485 			pending = 1;
486 			break;
487 		}
488 	}
489 
490 	if (!pending) {
491 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
492 		if (llq == NULL)	/* XXX what to do */
493 			return;
494 	}
495 
496 	/* Update the lladdr even if pending, it may have changed */
497 	llq->llq_ifp = ifp;
498 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
499 
500 	if (!pending)
501 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
502 
503 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
504 }
505 
506 /*
507  * Set the interface MAC address from a taskqueue to avoid a LOR.
508  */
509 static void
510 lagg_port_setlladdr(void *arg, int pending)
511 {
512 	struct lagg_softc *sc = (struct lagg_softc *)arg;
513 	struct lagg_llq *llq, *head;
514 	struct ifnet *ifp;
515 	int error;
516 
517 	/* Grab a local reference of the queue and remove it from the softc */
518 	LAGG_WLOCK(sc);
519 	head = SLIST_FIRST(&sc->sc_llq_head);
520 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
521 	LAGG_WUNLOCK(sc);
522 
523 	/*
524 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
525 	 * unlocked as we have the only reference to it.
526 	 */
527 	for (llq = head; llq != NULL; llq = head) {
528 		ifp = llq->llq_ifp;
529 
530 		/* Set the link layer address */
531 		CURVNET_SET(ifp->if_vnet);
532 		error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
533 		CURVNET_RESTORE();
534 		if (error)
535 			printf("%s: setlladdr failed on %s\n", __func__,
536 			    ifp->if_xname);
537 
538 		head = SLIST_NEXT(llq, llq_entries);
539 		free(llq, M_DEVBUF);
540 	}
541 }
542 
543 static int
544 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
545 {
546 	struct lagg_softc *sc_ptr;
547 	struct lagg_port *lp;
548 	int error = 0;
549 
550 	LAGG_WLOCK_ASSERT(sc);
551 
552 	/* Limit the maximal number of lagg ports */
553 	if (sc->sc_count >= LAGG_MAX_PORTS)
554 		return (ENOSPC);
555 
556 	/* Check if port has already been associated to a lagg */
557 	if (ifp->if_lagg != NULL) {
558 		/* Port is already in the current lagg? */
559 		lp = (struct lagg_port *)ifp->if_lagg;
560 		if (lp->lp_softc == sc)
561 			return (EEXIST);
562 		return (EBUSY);
563 	}
564 
565 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
566 	if (ifp->if_type != IFT_ETHER)
567 		return (EPROTONOSUPPORT);
568 
569 #ifdef INET6
570 	/*
571 	 * The member interface should not have inet6 address because
572 	 * two interfaces with a valid link-local scope zone must not be
573 	 * merged in any form.  This restriction is needed to
574 	 * prevent violation of link-local scope zone.  Attempts to
575 	 * add a member interface which has inet6 addresses triggers
576 	 * removal of all inet6 addresses on the member interface.
577 	 */
578 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
579 		if (in6ifa_llaonifp(lp->lp_ifp)) {
580 			in6_ifdetach(lp->lp_ifp);
581 			if_printf(sc->sc_ifp,
582 			    "IPv6 addresses on %s have been removed "
583 			    "before adding it as a member to prevent "
584 			    "IPv6 address scope violation.\n",
585 			    lp->lp_ifp->if_xname);
586 		}
587 	}
588 	if (in6ifa_llaonifp(ifp)) {
589 		in6_ifdetach(ifp);
590 		if_printf(sc->sc_ifp,
591 		    "IPv6 addresses on %s have been removed "
592 		    "before adding it as a member to prevent "
593 		    "IPv6 address scope violation.\n",
594 		    ifp->if_xname);
595 	}
596 #endif
597 	/* Allow the first Ethernet member to define the MTU */
598 	if (SLIST_EMPTY(&sc->sc_ports))
599 		sc->sc_ifp->if_mtu = ifp->if_mtu;
600 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
601 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
602 		    ifp->if_xname);
603 		return (EINVAL);
604 	}
605 
606 	if ((lp = malloc(sizeof(struct lagg_port),
607 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
608 		return (ENOMEM);
609 
610 	/* Check if port is a stacked lagg */
611 	mtx_lock(&lagg_list_mtx);
612 	SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
613 		if (ifp == sc_ptr->sc_ifp) {
614 			mtx_unlock(&lagg_list_mtx);
615 			free(lp, M_DEVBUF);
616 			return (EINVAL);
617 			/* XXX disable stacking for the moment, its untested */
618 #ifdef LAGG_PORT_STACKING
619 			lp->lp_flags |= LAGG_PORT_STACK;
620 			if (lagg_port_checkstacking(sc_ptr) >=
621 			    LAGG_MAX_STACKING) {
622 				mtx_unlock(&lagg_list_mtx);
623 				free(lp, M_DEVBUF);
624 				return (E2BIG);
625 			}
626 #endif
627 		}
628 	}
629 	mtx_unlock(&lagg_list_mtx);
630 
631 	/* Change the interface type */
632 	lp->lp_iftype = ifp->if_type;
633 	ifp->if_type = IFT_IEEE8023ADLAG;
634 	ifp->if_lagg = lp;
635 	lp->lp_ioctl = ifp->if_ioctl;
636 	ifp->if_ioctl = lagg_port_ioctl;
637 	lp->lp_output = ifp->if_output;
638 	ifp->if_output = lagg_port_output;
639 
640 	lp->lp_ifp = ifp;
641 	lp->lp_softc = sc;
642 
643 	/* Save port link layer address */
644 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
645 
646 	if (SLIST_EMPTY(&sc->sc_ports)) {
647 		sc->sc_primary = lp;
648 		lagg_lladdr(sc, IF_LLADDR(ifp));
649 	} else {
650 		/* Update link layer address for this port */
651 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
652 	}
653 
654 	/* Insert into the list of ports */
655 	SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
656 	sc->sc_count++;
657 
658 	/* Update lagg capabilities */
659 	lagg_capabilities(sc);
660 	lagg_linkstate(sc);
661 
662 	/* Add multicast addresses and interface flags to this port */
663 	lagg_ether_cmdmulti(lp, 1);
664 	lagg_setflags(lp, 1);
665 
666 	if (sc->sc_port_create != NULL)
667 		error = (*sc->sc_port_create)(lp);
668 	if (error) {
669 		/* remove the port again, without calling sc_port_destroy */
670 		lagg_port_destroy(lp, 0);
671 		return (error);
672 	}
673 
674 	return (error);
675 }
676 
677 #ifdef LAGG_PORT_STACKING
678 static int
679 lagg_port_checkstacking(struct lagg_softc *sc)
680 {
681 	struct lagg_softc *sc_ptr;
682 	struct lagg_port *lp;
683 	int m = 0;
684 
685 	LAGG_WLOCK_ASSERT(sc);
686 
687 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
688 		if (lp->lp_flags & LAGG_PORT_STACK) {
689 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
690 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
691 		}
692 	}
693 
694 	return (m + 1);
695 }
696 #endif
697 
698 static int
699 lagg_port_destroy(struct lagg_port *lp, int runpd)
700 {
701 	struct lagg_softc *sc = lp->lp_softc;
702 	struct lagg_port *lp_ptr;
703 	struct lagg_llq *llq;
704 	struct ifnet *ifp = lp->lp_ifp;
705 
706 	LAGG_WLOCK_ASSERT(sc);
707 
708 	if (runpd && sc->sc_port_destroy != NULL)
709 		(*sc->sc_port_destroy)(lp);
710 
711 	/*
712 	 * Remove multicast addresses and interface flags from this port and
713 	 * reset the MAC address, skip if the interface is being detached.
714 	 */
715 	if (!lp->lp_detaching) {
716 		lagg_ether_cmdmulti(lp, 0);
717 		lagg_setflags(lp, 0);
718 		lagg_port_lladdr(lp, lp->lp_lladdr);
719 	}
720 
721 	/* Restore interface */
722 	ifp->if_type = lp->lp_iftype;
723 	ifp->if_ioctl = lp->lp_ioctl;
724 	ifp->if_output = lp->lp_output;
725 	ifp->if_lagg = NULL;
726 
727 	/* Finally, remove the port from the lagg */
728 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
729 	sc->sc_count--;
730 
731 	/* Update the primary interface */
732 	if (lp == sc->sc_primary) {
733 		uint8_t lladdr[ETHER_ADDR_LEN];
734 
735 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
736 			bzero(&lladdr, ETHER_ADDR_LEN);
737 		} else {
738 			bcopy(lp_ptr->lp_lladdr,
739 			    lladdr, ETHER_ADDR_LEN);
740 		}
741 		lagg_lladdr(sc, lladdr);
742 		sc->sc_primary = lp_ptr;
743 
744 		/* Update link layer address for each port */
745 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
746 			lagg_port_lladdr(lp_ptr, lladdr);
747 	}
748 
749 	/* Remove any pending lladdr changes from the queue */
750 	if (lp->lp_detaching) {
751 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
752 			if (llq->llq_ifp == ifp) {
753 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
754 				    llq_entries);
755 				free(llq, M_DEVBUF);
756 				break;	/* Only appears once */
757 			}
758 		}
759 	}
760 
761 	if (lp->lp_ifflags)
762 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
763 
764 	free(lp, M_DEVBUF);
765 
766 	/* Update lagg capabilities */
767 	lagg_capabilities(sc);
768 	lagg_linkstate(sc);
769 
770 	return (0);
771 }
772 
773 static int
774 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
775 {
776 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
777 	struct lagg_softc *sc;
778 	struct lagg_port *lp = NULL;
779 	int error = 0;
780 	struct rm_priotracker tracker;
781 
782 	/* Should be checked by the caller */
783 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
784 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
785 		goto fallback;
786 
787 	switch (cmd) {
788 	case SIOCGLAGGPORT:
789 		if (rp->rp_portname[0] == '\0' ||
790 		    ifunit(rp->rp_portname) != ifp) {
791 			error = EINVAL;
792 			break;
793 		}
794 
795 		LAGG_RLOCK(sc, &tracker);
796 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
797 			error = ENOENT;
798 			LAGG_RUNLOCK(sc, &tracker);
799 			break;
800 		}
801 
802 		lagg_port2req(lp, rp);
803 		LAGG_RUNLOCK(sc, &tracker);
804 		break;
805 
806 	case SIOCSIFCAP:
807 		if (lp->lp_ioctl == NULL) {
808 			error = EINVAL;
809 			break;
810 		}
811 		error = (*lp->lp_ioctl)(ifp, cmd, data);
812 		if (error)
813 			break;
814 
815 		/* Update lagg interface capabilities */
816 		LAGG_WLOCK(sc);
817 		lagg_capabilities(sc);
818 		LAGG_WUNLOCK(sc);
819 		break;
820 
821 	case SIOCSIFMTU:
822 		/* Do not allow the MTU to be changed once joined */
823 		error = EINVAL;
824 		break;
825 
826 	default:
827 		goto fallback;
828 	}
829 
830 	return (error);
831 
832 fallback:
833 	if (lp->lp_ioctl != NULL)
834 		return ((*lp->lp_ioctl)(ifp, cmd, data));
835 
836 	return (EINVAL);
837 }
838 
839 /*
840  * For direct output to child ports.
841  */
842 static int
843 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
844 	const struct sockaddr *dst, struct route *ro)
845 {
846 	struct lagg_port *lp = ifp->if_lagg;
847 
848 	switch (dst->sa_family) {
849 		case pseudo_AF_HDRCMPLT:
850 		case AF_UNSPEC:
851 			return ((*lp->lp_output)(ifp, m, dst, ro));
852 	}
853 
854 	/* drop any other frames */
855 	m_freem(m);
856 	return (ENETDOWN);
857 }
858 
859 static void
860 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
861 {
862 	struct lagg_port *lp;
863 	struct lagg_softc *sc;
864 
865 	if ((lp = ifp->if_lagg) == NULL)
866 		return;
867 	/* If the ifnet is just being renamed, don't do anything. */
868 	if (ifp->if_flags & IFF_RENAMING)
869 		return;
870 
871 	sc = lp->lp_softc;
872 
873 	LAGG_WLOCK(sc);
874 	lp->lp_detaching = 1;
875 	lagg_port_destroy(lp, 1);
876 	LAGG_WUNLOCK(sc);
877 }
878 
879 static void
880 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
881 {
882 	struct lagg_softc *sc = lp->lp_softc;
883 
884 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
885 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
886 	rp->rp_prio = lp->lp_prio;
887 	rp->rp_flags = lp->lp_flags;
888 	if (sc->sc_portreq != NULL)
889 		(*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
890 
891 	/* Add protocol specific flags */
892 	switch (sc->sc_proto) {
893 		case LAGG_PROTO_FAILOVER:
894 			if (lp == sc->sc_primary)
895 				rp->rp_flags |= LAGG_PORT_MASTER;
896 			if (lp == lagg_link_active(sc, sc->sc_primary))
897 				rp->rp_flags |= LAGG_PORT_ACTIVE;
898 			break;
899 
900 		case LAGG_PROTO_ROUNDROBIN:
901 		case LAGG_PROTO_LOADBALANCE:
902 		case LAGG_PROTO_ETHERCHANNEL:
903 			if (LAGG_PORTACTIVE(lp))
904 				rp->rp_flags |= LAGG_PORT_ACTIVE;
905 			break;
906 
907 		case LAGG_PROTO_LACP:
908 			/* LACP has a different definition of active */
909 			if (lacp_isactive(lp))
910 				rp->rp_flags |= LAGG_PORT_ACTIVE;
911 			if (lacp_iscollecting(lp))
912 				rp->rp_flags |= LAGG_PORT_COLLECTING;
913 			if (lacp_isdistributing(lp))
914 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
915 			break;
916 	}
917 
918 }
919 
920 static void
921 lagg_init(void *xsc)
922 {
923 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
924 	struct lagg_port *lp;
925 	struct ifnet *ifp = sc->sc_ifp;
926 
927 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
928 		return;
929 
930 	LAGG_WLOCK(sc);
931 
932 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
933 	/* Update the port lladdrs */
934 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
935 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
936 
937 	if (sc->sc_init != NULL)
938 		(*sc->sc_init)(sc);
939 
940 	LAGG_WUNLOCK(sc);
941 }
942 
943 static void
944 lagg_stop(struct lagg_softc *sc)
945 {
946 	struct ifnet *ifp = sc->sc_ifp;
947 
948 	LAGG_WLOCK_ASSERT(sc);
949 
950 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
951 		return;
952 
953 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
954 
955 	if (sc->sc_stop != NULL)
956 		(*sc->sc_stop)(sc);
957 }
958 
959 static int
960 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
961 {
962 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
963 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
964 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
965 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
966 	struct ifreq *ifr = (struct ifreq *)data;
967 	struct lagg_port *lp;
968 	struct ifnet *tpif;
969 	struct thread *td = curthread;
970 	char *buf, *outbuf;
971 	int count, buflen, len, error = 0;
972 	struct rm_priotracker tracker;
973 
974 	bzero(&rpbuf, sizeof(rpbuf));
975 
976 	switch (cmd) {
977 	case SIOCGLAGG:
978 		LAGG_RLOCK(sc, &tracker);
979 		count = 0;
980 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
981 			count++;
982 		buflen = count * sizeof(struct lagg_reqport);
983 		LAGG_RUNLOCK(sc, &tracker);
984 
985 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
986 
987 		LAGG_RLOCK(sc, &tracker);
988 		ra->ra_proto = sc->sc_proto;
989 		if (sc->sc_req != NULL)
990 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
991 
992 		count = 0;
993 		buf = outbuf;
994 		len = min(ra->ra_size, buflen);
995 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
996 			if (len < sizeof(rpbuf))
997 				break;
998 
999 			lagg_port2req(lp, &rpbuf);
1000 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1001 			count++;
1002 			buf += sizeof(rpbuf);
1003 			len -= sizeof(rpbuf);
1004 		}
1005 		LAGG_RUNLOCK(sc, &tracker);
1006 		ra->ra_ports = count;
1007 		ra->ra_size = count * sizeof(rpbuf);
1008 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1009 		free(outbuf, M_TEMP);
1010 		break;
1011 	case SIOCSLAGG:
1012 		error = priv_check(td, PRIV_NET_LAGG);
1013 		if (error)
1014 			break;
1015 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1016 			error = EPROTONOSUPPORT;
1017 			break;
1018 		}
1019 		LAGG_WLOCK(sc);
1020 		if (sc->sc_proto != LAGG_PROTO_NONE) {
1021 			/* Reset protocol first in case detach unlocks */
1022 			sc->sc_proto = LAGG_PROTO_NONE;
1023 			error = sc->sc_detach(sc);
1024 			sc->sc_detach = NULL;
1025 			sc->sc_start = NULL;
1026 			sc->sc_input = NULL;
1027 			sc->sc_port_create = NULL;
1028 			sc->sc_port_destroy = NULL;
1029 			sc->sc_linkstate = NULL;
1030 			sc->sc_init = NULL;
1031 			sc->sc_stop = NULL;
1032 			sc->sc_lladdr = NULL;
1033 			sc->sc_req = NULL;
1034 			sc->sc_portreq = NULL;
1035 		} else if (sc->sc_input != NULL) {
1036 			/* Still detaching */
1037 			error = EBUSY;
1038 		}
1039 		if (error != 0) {
1040 			LAGG_WUNLOCK(sc);
1041 			break;
1042 		}
1043 		for (int i = 0; i < (sizeof(lagg_protos) /
1044 		    sizeof(lagg_protos[0])); i++) {
1045 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
1046 				if (sc->sc_ifflags & IFF_DEBUG)
1047 					printf("%s: using proto %u\n",
1048 					    sc->sc_ifname,
1049 					    lagg_protos[i].ti_proto);
1050 				sc->sc_proto = lagg_protos[i].ti_proto;
1051 				if (sc->sc_proto != LAGG_PROTO_NONE)
1052 					error = lagg_protos[i].ti_attach(sc);
1053 				LAGG_WUNLOCK(sc);
1054 				return (error);
1055 			}
1056 		}
1057 		LAGG_WUNLOCK(sc);
1058 		error = EPROTONOSUPPORT;
1059 		break;
1060 	case SIOCGLAGGFLAGS:
1061 		rf->rf_flags = sc->sc_flags;
1062 		break;
1063 	case SIOCSLAGGHASH:
1064 		error = priv_check(td, PRIV_NET_LAGG);
1065 		if (error)
1066 			break;
1067 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1068 			error = EINVAL;
1069 			break;
1070 		}
1071 		LAGG_WLOCK(sc);
1072 		sc->sc_flags &= ~LAGG_F_HASHMASK;
1073 		sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK;
1074 		LAGG_WUNLOCK(sc);
1075 		break;
1076 	case SIOCGLAGGPORT:
1077 		if (rp->rp_portname[0] == '\0' ||
1078 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1079 			error = EINVAL;
1080 			break;
1081 		}
1082 
1083 		LAGG_RLOCK(sc, &tracker);
1084 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1085 		    lp->lp_softc != sc) {
1086 			error = ENOENT;
1087 			LAGG_RUNLOCK(sc, &tracker);
1088 			break;
1089 		}
1090 
1091 		lagg_port2req(lp, rp);
1092 		LAGG_RUNLOCK(sc, &tracker);
1093 		break;
1094 	case SIOCSLAGGPORT:
1095 		error = priv_check(td, PRIV_NET_LAGG);
1096 		if (error)
1097 			break;
1098 		if (rp->rp_portname[0] == '\0' ||
1099 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1100 			error = EINVAL;
1101 			break;
1102 		}
1103 		LAGG_WLOCK(sc);
1104 		error = lagg_port_create(sc, tpif);
1105 		LAGG_WUNLOCK(sc);
1106 		break;
1107 	case SIOCSLAGGDELPORT:
1108 		error = priv_check(td, PRIV_NET_LAGG);
1109 		if (error)
1110 			break;
1111 		if (rp->rp_portname[0] == '\0' ||
1112 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1113 			error = EINVAL;
1114 			break;
1115 		}
1116 
1117 		LAGG_WLOCK(sc);
1118 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1119 		    lp->lp_softc != sc) {
1120 			error = ENOENT;
1121 			LAGG_WUNLOCK(sc);
1122 			break;
1123 		}
1124 
1125 		error = lagg_port_destroy(lp, 1);
1126 		LAGG_WUNLOCK(sc);
1127 		break;
1128 	case SIOCSIFFLAGS:
1129 		/* Set flags on ports too */
1130 		LAGG_WLOCK(sc);
1131 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1132 			lagg_setflags(lp, 1);
1133 		}
1134 		LAGG_WUNLOCK(sc);
1135 
1136 		if (!(ifp->if_flags & IFF_UP) &&
1137 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1138 			/*
1139 			 * If interface is marked down and it is running,
1140 			 * then stop and disable it.
1141 			 */
1142 			LAGG_WLOCK(sc);
1143 			lagg_stop(sc);
1144 			LAGG_WUNLOCK(sc);
1145 		} else if ((ifp->if_flags & IFF_UP) &&
1146 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1147 			/*
1148 			 * If interface is marked up and it is stopped, then
1149 			 * start it.
1150 			 */
1151 			(*ifp->if_init)(sc);
1152 		}
1153 		break;
1154 	case SIOCADDMULTI:
1155 	case SIOCDELMULTI:
1156 		LAGG_WLOCK(sc);
1157 		error = lagg_ether_setmulti(sc);
1158 		LAGG_WUNLOCK(sc);
1159 		break;
1160 	case SIOCSIFMEDIA:
1161 	case SIOCGIFMEDIA:
1162 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1163 		break;
1164 
1165 	case SIOCSIFCAP:
1166 	case SIOCSIFMTU:
1167 		/* Do not allow the MTU or caps to be directly changed */
1168 		error = EINVAL;
1169 		break;
1170 
1171 	default:
1172 		error = ether_ioctl(ifp, cmd, data);
1173 		break;
1174 	}
1175 	return (error);
1176 }
1177 
1178 static int
1179 lagg_ether_setmulti(struct lagg_softc *sc)
1180 {
1181 	struct lagg_port *lp;
1182 
1183 	LAGG_WLOCK_ASSERT(sc);
1184 
1185 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1186 		/* First, remove any existing filter entries. */
1187 		lagg_ether_cmdmulti(lp, 0);
1188 		/* copy all addresses from the lagg interface to the port */
1189 		lagg_ether_cmdmulti(lp, 1);
1190 	}
1191 	return (0);
1192 }
1193 
1194 static int
1195 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1196 {
1197 	struct lagg_softc *sc = lp->lp_softc;
1198 	struct ifnet *ifp = lp->lp_ifp;
1199 	struct ifnet *scifp = sc->sc_ifp;
1200 	struct lagg_mc *mc;
1201 	struct ifmultiaddr *ifma, *rifma = NULL;
1202 	struct sockaddr_dl sdl;
1203 	int error;
1204 
1205 	LAGG_WLOCK_ASSERT(sc);
1206 
1207 	bzero((char *)&sdl, sizeof(sdl));
1208 	sdl.sdl_len = sizeof(sdl);
1209 	sdl.sdl_family = AF_LINK;
1210 	sdl.sdl_type = IFT_ETHER;
1211 	sdl.sdl_alen = ETHER_ADDR_LEN;
1212 	sdl.sdl_index = ifp->if_index;
1213 
1214 	if (set) {
1215 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1216 			if (ifma->ifma_addr->sa_family != AF_LINK)
1217 				continue;
1218 			bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1219 			    LLADDR(&sdl), ETHER_ADDR_LEN);
1220 
1221 			error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1222 			if (error)
1223 				return (error);
1224 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1225 			if (mc == NULL)
1226 				return (ENOMEM);
1227 			mc->mc_ifma = rifma;
1228 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1229 		}
1230 	} else {
1231 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1232 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1233 			if_delmulti_ifma(mc->mc_ifma);
1234 			free(mc, M_DEVBUF);
1235 		}
1236 	}
1237 	return (0);
1238 }
1239 
1240 /* Handle a ref counted flag that should be set on the lagg port as well */
1241 static int
1242 lagg_setflag(struct lagg_port *lp, int flag, int status,
1243 	     int (*func)(struct ifnet *, int))
1244 {
1245 	struct lagg_softc *sc = lp->lp_softc;
1246 	struct ifnet *scifp = sc->sc_ifp;
1247 	struct ifnet *ifp = lp->lp_ifp;
1248 	int error;
1249 
1250 	LAGG_WLOCK_ASSERT(sc);
1251 
1252 	status = status ? (scifp->if_flags & flag) : 0;
1253 	/* Now "status" contains the flag value or 0 */
1254 
1255 	/*
1256 	 * See if recorded ports status is different from what
1257 	 * we want it to be.  If it is, flip it.  We record ports
1258 	 * status in lp_ifflags so that we won't clear ports flag
1259 	 * we haven't set.  In fact, we don't clear or set ports
1260 	 * flags directly, but get or release references to them.
1261 	 * That's why we can be sure that recorded flags still are
1262 	 * in accord with actual ports flags.
1263 	 */
1264 	if (status != (lp->lp_ifflags & flag)) {
1265 		error = (*func)(ifp, status);
1266 		if (error)
1267 			return (error);
1268 		lp->lp_ifflags &= ~flag;
1269 		lp->lp_ifflags |= status;
1270 	}
1271 	return (0);
1272 }
1273 
1274 /*
1275  * Handle IFF_* flags that require certain changes on the lagg port
1276  * if "status" is true, update ports flags respective to the lagg
1277  * if "status" is false, forcedly clear the flags set on port.
1278  */
1279 static int
1280 lagg_setflags(struct lagg_port *lp, int status)
1281 {
1282 	int error, i;
1283 
1284 	for (i = 0; lagg_pflags[i].flag; i++) {
1285 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1286 		    status, lagg_pflags[i].func);
1287 		if (error)
1288 			return (error);
1289 	}
1290 	return (0);
1291 }
1292 
1293 static int
1294 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1295 {
1296 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1297 	int error, len, mcast;
1298 	struct rm_priotracker tracker;
1299 
1300 	len = m->m_pkthdr.len;
1301 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1302 
1303 	LAGG_RLOCK(sc, &tracker);
1304 	/* We need a Tx algorithm and at least one port */
1305 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1306 		LAGG_RUNLOCK(sc, &tracker);
1307 		m_freem(m);
1308 		ifp->if_oerrors++;
1309 		return (ENXIO);
1310 	}
1311 
1312 	ETHER_BPF_MTAP(ifp, m);
1313 
1314 	error = (*sc->sc_start)(sc, m);
1315 	LAGG_RUNLOCK(sc, &tracker);
1316 
1317 	if (error == 0) {
1318 		counter_u64_add(sc->sc_opackets, 1);
1319 		counter_u64_add(sc->sc_obytes, len);
1320 		ifp->if_omcasts += mcast;
1321 	} else
1322 		ifp->if_oerrors++;
1323 
1324 	return (error);
1325 }
1326 
1327 /*
1328  * The ifp->if_qflush entry point for lagg(4) is no-op.
1329  */
1330 static void
1331 lagg_qflush(struct ifnet *ifp __unused)
1332 {
1333 }
1334 
1335 static struct mbuf *
1336 lagg_input(struct ifnet *ifp, struct mbuf *m)
1337 {
1338 	struct lagg_port *lp = ifp->if_lagg;
1339 	struct lagg_softc *sc = lp->lp_softc;
1340 	struct ifnet *scifp = sc->sc_ifp;
1341 	struct rm_priotracker tracker;
1342 
1343 	LAGG_RLOCK(sc, &tracker);
1344 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1345 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1346 	    sc->sc_proto == LAGG_PROTO_NONE) {
1347 		LAGG_RUNLOCK(sc, &tracker);
1348 		m_freem(m);
1349 		return (NULL);
1350 	}
1351 
1352 	ETHER_BPF_MTAP(scifp, m);
1353 
1354 	m = (*sc->sc_input)(sc, lp, m);
1355 
1356 	if (m != NULL) {
1357 		counter_u64_add(sc->sc_ipackets, 1);
1358 		counter_u64_add(sc->sc_ibytes, m->m_pkthdr.len);
1359 
1360 		if (scifp->if_flags & IFF_MONITOR) {
1361 			m_freem(m);
1362 			m = NULL;
1363 		}
1364 	}
1365 
1366 	LAGG_RUNLOCK(sc, &tracker);
1367 	return (m);
1368 }
1369 
1370 static int
1371 lagg_media_change(struct ifnet *ifp)
1372 {
1373 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1374 
1375 	if (sc->sc_ifflags & IFF_DEBUG)
1376 		printf("%s\n", __func__);
1377 
1378 	/* Ignore */
1379 	return (0);
1380 }
1381 
1382 static void
1383 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1384 {
1385 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1386 	struct lagg_port *lp;
1387 	struct rm_priotracker tracker;
1388 
1389 	imr->ifm_status = IFM_AVALID;
1390 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1391 
1392 	LAGG_RLOCK(sc, &tracker);
1393 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1394 		if (LAGG_PORTACTIVE(lp))
1395 			imr->ifm_status |= IFM_ACTIVE;
1396 	}
1397 	LAGG_RUNLOCK(sc, &tracker);
1398 }
1399 
1400 static void
1401 lagg_linkstate(struct lagg_softc *sc)
1402 {
1403 	struct lagg_port *lp;
1404 	int new_link = LINK_STATE_DOWN;
1405 	uint64_t speed;
1406 
1407 	/* Our link is considered up if at least one of our ports is active */
1408 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1409 		if (lp->lp_link_state == LINK_STATE_UP) {
1410 			new_link = LINK_STATE_UP;
1411 			break;
1412 		}
1413 	}
1414 	if_link_state_change(sc->sc_ifp, new_link);
1415 
1416 	/* Update if_baudrate to reflect the max possible speed */
1417 	switch (sc->sc_proto) {
1418 		case LAGG_PROTO_FAILOVER:
1419 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1420 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1421 			break;
1422 		case LAGG_PROTO_ROUNDROBIN:
1423 		case LAGG_PROTO_LOADBALANCE:
1424 		case LAGG_PROTO_ETHERCHANNEL:
1425 			speed = 0;
1426 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1427 				speed += lp->lp_ifp->if_baudrate;
1428 			sc->sc_ifp->if_baudrate = speed;
1429 			break;
1430 		case LAGG_PROTO_LACP:
1431 			/* LACP updates if_baudrate itself */
1432 			break;
1433 	}
1434 }
1435 
1436 static void
1437 lagg_port_state(struct ifnet *ifp, int state)
1438 {
1439 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1440 	struct lagg_softc *sc = NULL;
1441 
1442 	if (lp != NULL)
1443 		sc = lp->lp_softc;
1444 	if (sc == NULL)
1445 		return;
1446 
1447 	LAGG_WLOCK(sc);
1448 	lagg_linkstate(sc);
1449 	if (sc->sc_linkstate != NULL)
1450 		(*sc->sc_linkstate)(lp);
1451 	LAGG_WUNLOCK(sc);
1452 }
1453 
1454 struct lagg_port *
1455 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1456 {
1457 	struct lagg_port *lp_next, *rval = NULL;
1458 	// int new_link = LINK_STATE_DOWN;
1459 
1460 	LAGG_RLOCK_ASSERT(sc);
1461 	/*
1462 	 * Search a port which reports an active link state.
1463 	 */
1464 
1465 	if (lp == NULL)
1466 		goto search;
1467 	if (LAGG_PORTACTIVE(lp)) {
1468 		rval = lp;
1469 		goto found;
1470 	}
1471 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1472 	    LAGG_PORTACTIVE(lp_next)) {
1473 		rval = lp_next;
1474 		goto found;
1475 	}
1476 
1477 search:
1478 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1479 		if (LAGG_PORTACTIVE(lp_next)) {
1480 			rval = lp_next;
1481 			goto found;
1482 		}
1483 	}
1484 
1485 found:
1486 	if (rval != NULL) {
1487 		/*
1488 		 * The IEEE 802.1D standard assumes that a lagg with
1489 		 * multiple ports is always full duplex. This is valid
1490 		 * for load sharing laggs and if at least two links
1491 		 * are active. Unfortunately, checking the latter would
1492 		 * be too expensive at this point.
1493 		 XXX
1494 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1495 		    (sc->sc_count > 1))
1496 			new_link = LINK_STATE_FULL_DUPLEX;
1497 		else
1498 			new_link = rval->lp_link_state;
1499 		 */
1500 	}
1501 
1502 	return (rval);
1503 }
1504 
1505 static const void *
1506 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1507 {
1508 	if (m->m_pkthdr.len < (off + len)) {
1509 		return (NULL);
1510 	} else if (m->m_len < (off + len)) {
1511 		m_copydata(m, off, len, buf);
1512 		return (buf);
1513 	}
1514 	return (mtod(m, char *) + off);
1515 }
1516 
1517 static int
1518 lagg_sysctl_active(SYSCTL_HANDLER_ARGS)
1519 {
1520 	struct lagg_softc *sc = (struct lagg_softc *)arg1;
1521 	struct lagg_port *lp;
1522 	int error;
1523 
1524 	/* LACP tracks active links automatically, the others do not */
1525 	if (sc->sc_proto != LAGG_PROTO_LACP) {
1526 		sc->sc_active = 0;
1527 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1528 			sc->sc_active += LAGG_PORTACTIVE(lp);
1529 	}
1530 
1531 	error = sysctl_handle_int(oidp, &sc->sc_active, 0, req);
1532 	if ((error) || (req->newptr == NULL))
1533 		return (error);
1534 
1535 	return (0);
1536 }
1537 
1538 uint32_t
1539 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
1540 {
1541 	uint16_t etype;
1542 	uint32_t p = key;
1543 	int off;
1544 	struct ether_header *eh;
1545 	const struct ether_vlan_header *vlan;
1546 #ifdef INET
1547 	const struct ip *ip;
1548 	const uint32_t *ports;
1549 	int iphlen;
1550 #endif
1551 #ifdef INET6
1552 	const struct ip6_hdr *ip6;
1553 	uint32_t flow;
1554 #endif
1555 	union {
1556 #ifdef INET
1557 		struct ip ip;
1558 #endif
1559 #ifdef INET6
1560 		struct ip6_hdr ip6;
1561 #endif
1562 		struct ether_vlan_header vlan;
1563 		uint32_t port;
1564 	} buf;
1565 
1566 
1567 	off = sizeof(*eh);
1568 	if (m->m_len < off)
1569 		goto out;
1570 	eh = mtod(m, struct ether_header *);
1571 	etype = ntohs(eh->ether_type);
1572 	if (sc->sc_flags & LAGG_F_HASHL2) {
1573 		p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p);
1574 		p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1575 	}
1576 
1577 	/* Special handling for encapsulating VLAN frames */
1578 	if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) {
1579 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1580 		    sizeof(m->m_pkthdr.ether_vtag), p);
1581 	} else if (etype == ETHERTYPE_VLAN) {
1582 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &buf);
1583 		if (vlan == NULL)
1584 			goto out;
1585 
1586 		if (sc->sc_flags & LAGG_F_HASHL2)
1587 			p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1588 		etype = ntohs(vlan->evl_proto);
1589 		off += sizeof(*vlan) - sizeof(*eh);
1590 	}
1591 
1592 	switch (etype) {
1593 #ifdef INET
1594 	case ETHERTYPE_IP:
1595 		ip = lagg_gethdr(m, off, sizeof(*ip), &buf);
1596 		if (ip == NULL)
1597 			goto out;
1598 
1599 		if (sc->sc_flags & LAGG_F_HASHL3) {
1600 			p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1601 			p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1602 		}
1603 		if (!(sc->sc_flags & LAGG_F_HASHL4))
1604 			break;
1605 		switch (ip->ip_p) {
1606 			case IPPROTO_TCP:
1607 			case IPPROTO_UDP:
1608 			case IPPROTO_SCTP:
1609 				iphlen = ip->ip_hl << 2;
1610 				if (iphlen < sizeof(*ip))
1611 					break;
1612 				off += iphlen;
1613 				ports = lagg_gethdr(m, off, sizeof(*ports), &buf);
1614 				if (ports == NULL)
1615 					break;
1616 				p = hash32_buf(ports, sizeof(*ports), p);
1617 				break;
1618 		}
1619 		break;
1620 #endif
1621 #ifdef INET6
1622 	case ETHERTYPE_IPV6:
1623 		if (!(sc->sc_flags & LAGG_F_HASHL3))
1624 			break;
1625 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf);
1626 		if (ip6 == NULL)
1627 			goto out;
1628 
1629 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1630 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1631 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1632 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1633 		break;
1634 #endif
1635 	}
1636 out:
1637 	return (p);
1638 }
1639 
1640 int
1641 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1642 {
1643 
1644 	return (ifp->if_transmit)(ifp, m);
1645 }
1646 
1647 /*
1648  * Simple round robin aggregation
1649  */
1650 
1651 static int
1652 lagg_rr_attach(struct lagg_softc *sc)
1653 {
1654 	sc->sc_detach = lagg_rr_detach;
1655 	sc->sc_start = lagg_rr_start;
1656 	sc->sc_input = lagg_rr_input;
1657 	sc->sc_port_create = NULL;
1658 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1659 	sc->sc_seq = 0;
1660 
1661 	return (0);
1662 }
1663 
1664 static int
1665 lagg_rr_detach(struct lagg_softc *sc)
1666 {
1667 	return (0);
1668 }
1669 
1670 static int
1671 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1672 {
1673 	struct lagg_port *lp;
1674 	uint32_t p;
1675 
1676 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1677 	p %= sc->sc_count;
1678 	lp = SLIST_FIRST(&sc->sc_ports);
1679 	while (p--)
1680 		lp = SLIST_NEXT(lp, lp_entries);
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 (ENETDOWN);
1689 	}
1690 
1691 	/* Send mbuf */
1692 	return (lagg_enqueue(lp->lp_ifp, m));
1693 }
1694 
1695 static struct mbuf *
1696 lagg_rr_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  * Active failover
1708  */
1709 
1710 static int
1711 lagg_fail_attach(struct lagg_softc *sc)
1712 {
1713 	sc->sc_detach = lagg_fail_detach;
1714 	sc->sc_start = lagg_fail_start;
1715 	sc->sc_input = lagg_fail_input;
1716 	sc->sc_port_create = NULL;
1717 	sc->sc_port_destroy = NULL;
1718 
1719 	return (0);
1720 }
1721 
1722 static int
1723 lagg_fail_detach(struct lagg_softc *sc)
1724 {
1725 	return (0);
1726 }
1727 
1728 static int
1729 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1730 {
1731 	struct lagg_port *lp;
1732 
1733 	/* Use the master port if active or the next available port */
1734 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1735 		m_freem(m);
1736 		return (ENETDOWN);
1737 	}
1738 
1739 	/* Send mbuf */
1740 	return (lagg_enqueue(lp->lp_ifp, m));
1741 }
1742 
1743 static struct mbuf *
1744 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1745 {
1746 	struct ifnet *ifp = sc->sc_ifp;
1747 	struct lagg_port *tmp_tp;
1748 
1749 	if (lp == sc->sc_primary || lagg_failover_rx_all) {
1750 		m->m_pkthdr.rcvif = ifp;
1751 		return (m);
1752 	}
1753 
1754 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1755 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1756 		/*
1757 		 * If tmp_tp is null, we've recieved a packet when all
1758 		 * our links are down. Weird, but process it anyways.
1759 		 */
1760 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1761 			m->m_pkthdr.rcvif = ifp;
1762 			return (m);
1763 		}
1764 	}
1765 
1766 	m_freem(m);
1767 	return (NULL);
1768 }
1769 
1770 /*
1771  * Loadbalancing
1772  */
1773 
1774 static int
1775 lagg_lb_attach(struct lagg_softc *sc)
1776 {
1777 	struct lagg_port *lp;
1778 	struct lagg_lb *lb;
1779 
1780 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1781 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1782 		return (ENOMEM);
1783 
1784 	sc->sc_detach = lagg_lb_detach;
1785 	sc->sc_start = lagg_lb_start;
1786 	sc->sc_input = lagg_lb_input;
1787 	sc->sc_port_create = lagg_lb_port_create;
1788 	sc->sc_port_destroy = lagg_lb_port_destroy;
1789 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1790 
1791 	lb->lb_key = arc4random();
1792 	sc->sc_psc = (caddr_t)lb;
1793 
1794 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1795 		lagg_lb_port_create(lp);
1796 
1797 	return (0);
1798 }
1799 
1800 static int
1801 lagg_lb_detach(struct lagg_softc *sc)
1802 {
1803 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1804 	if (lb != NULL)
1805 		free(lb, M_DEVBUF);
1806 	return (0);
1807 }
1808 
1809 static int
1810 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1811 {
1812 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1813 	struct lagg_port *lp_next;
1814 	int i = 0;
1815 
1816 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1817 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1818 		if (lp_next == lp)
1819 			continue;
1820 		if (i >= LAGG_MAX_PORTS)
1821 			return (EINVAL);
1822 		if (sc->sc_ifflags & IFF_DEBUG)
1823 			printf("%s: port %s at index %d\n",
1824 			    sc->sc_ifname, lp_next->lp_ifname, i);
1825 		lb->lb_ports[i++] = lp_next;
1826 	}
1827 
1828 	return (0);
1829 }
1830 
1831 static int
1832 lagg_lb_port_create(struct lagg_port *lp)
1833 {
1834 	struct lagg_softc *sc = lp->lp_softc;
1835 	return (lagg_lb_porttable(sc, NULL));
1836 }
1837 
1838 static void
1839 lagg_lb_port_destroy(struct lagg_port *lp)
1840 {
1841 	struct lagg_softc *sc = lp->lp_softc;
1842 	lagg_lb_porttable(sc, lp);
1843 }
1844 
1845 static int
1846 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1847 {
1848 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1849 	struct lagg_port *lp = NULL;
1850 	uint32_t p = 0;
1851 
1852 	if (sc->use_flowid && (m->m_flags & M_FLOWID))
1853 		p = m->m_pkthdr.flowid;
1854 	else
1855 		p = lagg_hashmbuf(sc, m, lb->lb_key);
1856 	p %= sc->sc_count;
1857 	lp = lb->lb_ports[p];
1858 
1859 	/*
1860 	 * Check the port's link state. This will return the next active
1861 	 * port if the link is down or the port is NULL.
1862 	 */
1863 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1864 		m_freem(m);
1865 		return (ENETDOWN);
1866 	}
1867 
1868 	/* Send mbuf */
1869 	return (lagg_enqueue(lp->lp_ifp, m));
1870 }
1871 
1872 static struct mbuf *
1873 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1874 {
1875 	struct ifnet *ifp = sc->sc_ifp;
1876 
1877 	/* Just pass in the packet to our lagg device */
1878 	m->m_pkthdr.rcvif = ifp;
1879 
1880 	return (m);
1881 }
1882 
1883 /*
1884  * 802.3ad LACP
1885  */
1886 
1887 static int
1888 lagg_lacp_attach(struct lagg_softc *sc)
1889 {
1890 	struct lagg_port *lp;
1891 	int error;
1892 
1893 	sc->sc_detach = lagg_lacp_detach;
1894 	sc->sc_port_create = lacp_port_create;
1895 	sc->sc_port_destroy = lacp_port_destroy;
1896 	sc->sc_linkstate = lacp_linkstate;
1897 	sc->sc_start = lagg_lacp_start;
1898 	sc->sc_input = lagg_lacp_input;
1899 	sc->sc_init = lacp_init;
1900 	sc->sc_stop = lacp_stop;
1901 	sc->sc_lladdr = lagg_lacp_lladdr;
1902 	sc->sc_req = lacp_req;
1903 	sc->sc_portreq = lacp_portreq;
1904 
1905 	error = lacp_attach(sc);
1906 	if (error)
1907 		return (error);
1908 
1909 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1910 		lacp_port_create(lp);
1911 
1912 	return (error);
1913 }
1914 
1915 static int
1916 lagg_lacp_detach(struct lagg_softc *sc)
1917 {
1918 	struct lagg_port *lp;
1919 	int error;
1920 
1921 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1922 		lacp_port_destroy(lp);
1923 
1924 	/* unlocking is safe here */
1925 	LAGG_WUNLOCK(sc);
1926 	error = lacp_detach(sc);
1927 	LAGG_WLOCK(sc);
1928 
1929 	return (error);
1930 }
1931 
1932 static void
1933 lagg_lacp_lladdr(struct lagg_softc *sc)
1934 {
1935 	struct lagg_port *lp;
1936 
1937 	/* purge all the lacp ports */
1938 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1939 		lacp_port_destroy(lp);
1940 
1941 	/* add them back in */
1942 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1943 		lacp_port_create(lp);
1944 }
1945 
1946 static int
1947 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1948 {
1949 	struct lagg_port *lp;
1950 
1951 	lp = lacp_select_tx_port(sc, m);
1952 	if (lp == NULL) {
1953 		m_freem(m);
1954 		return (ENETDOWN);
1955 	}
1956 
1957 	/* Send mbuf */
1958 	return (lagg_enqueue(lp->lp_ifp, m));
1959 }
1960 
1961 static struct mbuf *
1962 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1963 {
1964 	struct ifnet *ifp = sc->sc_ifp;
1965 	struct ether_header *eh;
1966 	u_short etype;
1967 
1968 	eh = mtod(m, struct ether_header *);
1969 	etype = ntohs(eh->ether_type);
1970 
1971 	/* Tap off LACP control messages */
1972 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
1973 		m = lacp_input(lp, m);
1974 		if (m == NULL)
1975 			return (NULL);
1976 	}
1977 
1978 	/*
1979 	 * If the port is not collecting or not in the active aggregator then
1980 	 * free and return.
1981 	 */
1982 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
1983 		m_freem(m);
1984 		return (NULL);
1985 	}
1986 
1987 	m->m_pkthdr.rcvif = ifp;
1988 	return (m);
1989 }
1990 
1991 static void
1992 lagg_callout(void *arg)
1993 {
1994 	struct lagg_softc *sc = (struct lagg_softc *)arg;
1995 	struct ifnet *ifp = sc->sc_ifp;
1996 
1997 	ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets);
1998 	ifp->if_opackets = counter_u64_fetch(sc->sc_opackets);
1999 	ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes);
2000 	ifp->if_obytes = counter_u64_fetch(sc->sc_obytes);
2001 
2002 	callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
2003 }
2004