xref: /freebsd/sys/net/if_lagg.c (revision 6186fd1857626de0f7cb1a9e4dff19082f9ebb11)
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  * Copyright (c) 2014 Marcelo Araujo <araujo@FreeBSD.org>
7  *
8  * Permission to use, copy, modify, and distribute this software for any
9  * purpose with or without fee is hereby granted, provided that the above
10  * copyright notice and this permission notice appear in all copies.
11  *
12  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19  */
20 
21 #include <sys/cdefs.h>
22 __FBSDID("$FreeBSD$");
23 
24 #include "opt_inet.h"
25 #include "opt_inet6.h"
26 
27 #include <sys/param.h>
28 #include <sys/kernel.h>
29 #include <sys/malloc.h>
30 #include <sys/mbuf.h>
31 #include <sys/queue.h>
32 #include <sys/socket.h>
33 #include <sys/sockio.h>
34 #include <sys/sysctl.h>
35 #include <sys/module.h>
36 #include <sys/priv.h>
37 #include <sys/systm.h>
38 #include <sys/proc.h>
39 #include <sys/hash.h>
40 #include <sys/lock.h>
41 #include <sys/rmlock.h>
42 #include <sys/taskqueue.h>
43 #include <sys/eventhandler.h>
44 
45 #include <net/ethernet.h>
46 #include <net/if.h>
47 #include <net/if_clone.h>
48 #include <net/if_arp.h>
49 #include <net/if_dl.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 #include <net/vnet.h>
55 
56 #if defined(INET) || defined(INET6)
57 #include <netinet/in.h>
58 #include <netinet/ip.h>
59 #endif
60 #ifdef INET
61 #include <netinet/in_systm.h>
62 #include <netinet/if_ether.h>
63 #endif
64 
65 #ifdef INET6
66 #include <netinet/ip6.h>
67 #include <netinet6/in6_var.h>
68 #include <netinet6/in6_ifattach.h>
69 #endif
70 
71 #include <net/if_vlan_var.h>
72 #include <net/if_lagg.h>
73 #include <net/ieee8023ad_lacp.h>
74 
75 /* Special flags we should propagate to the lagg ports. */
76 static struct {
77 	int flag;
78 	int (*func)(struct ifnet *, int);
79 } lagg_pflags[] = {
80 	{IFF_PROMISC, ifpromisc},
81 	{IFF_ALLMULTI, if_allmulti},
82 	{0, NULL}
83 };
84 
85 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
86 #define	V_lagg_list	VNET(lagg_list)
87 static VNET_DEFINE(struct mtx, lagg_list_mtx);
88 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
89 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
90 					"if_lagg list", NULL, MTX_DEF)
91 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
92 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
93 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
94 eventhandler_tag	lagg_detach_cookie = NULL;
95 
96 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
97 static void	lagg_clone_destroy(struct ifnet *);
98 static VNET_DEFINE(struct if_clone *, lagg_cloner);
99 #define	V_lagg_cloner	VNET(lagg_cloner)
100 static const char laggname[] = "lagg";
101 
102 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
103 static void	lagg_capabilities(struct lagg_softc *);
104 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
105 static void	lagg_port_setlladdr(void *, int);
106 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
107 static int	lagg_port_destroy(struct lagg_port *, int);
108 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
109 static void	lagg_linkstate(struct lagg_softc *);
110 static void	lagg_port_state(struct ifnet *, int);
111 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
112 static int	lagg_port_output(struct ifnet *, struct mbuf *,
113 		    const struct sockaddr *, struct route *);
114 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
115 #ifdef LAGG_PORT_STACKING
116 static int	lagg_port_checkstacking(struct lagg_softc *);
117 #endif
118 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
119 static void	lagg_init(void *);
120 static void	lagg_stop(struct lagg_softc *);
121 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
122 static int	lagg_ether_setmulti(struct lagg_softc *);
123 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
124 static	int	lagg_setflag(struct lagg_port *, int, int,
125 		    int (*func)(struct ifnet *, int));
126 static	int	lagg_setflags(struct lagg_port *, int status);
127 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
128 static int	lagg_transmit(struct ifnet *, struct mbuf *);
129 static void	lagg_qflush(struct ifnet *);
130 static int	lagg_media_change(struct ifnet *);
131 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
132 static struct lagg_port *lagg_link_active(struct lagg_softc *,
133 	    struct lagg_port *);
134 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
135 
136 /* Simple round robin */
137 static void	lagg_rr_attach(struct lagg_softc *);
138 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
139 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
140 		    struct mbuf *);
141 
142 /* Active failover */
143 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
144 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
145 		    struct mbuf *);
146 
147 /* Loadbalancing */
148 static void	lagg_lb_attach(struct lagg_softc *);
149 static void	lagg_lb_detach(struct lagg_softc *);
150 static int	lagg_lb_port_create(struct lagg_port *);
151 static void	lagg_lb_port_destroy(struct lagg_port *);
152 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
153 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
154 		    struct mbuf *);
155 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
156 
157 /* Broadcast */
158 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
159 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
160 		    struct mbuf *);
161 
162 /* 802.3ad LACP */
163 static void	lagg_lacp_attach(struct lagg_softc *);
164 static void	lagg_lacp_detach(struct lagg_softc *);
165 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
166 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
167 		    struct mbuf *);
168 static void	lagg_lacp_lladdr(struct lagg_softc *);
169 
170 /* lagg protocol table */
171 static const struct lagg_proto {
172 	lagg_proto	pr_num;
173 	void		(*pr_attach)(struct lagg_softc *);
174 	void		(*pr_detach)(struct lagg_softc *);
175 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
176 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
177 			    struct mbuf *);
178 	int		(*pr_addport)(struct lagg_port *);
179 	void		(*pr_delport)(struct lagg_port *);
180 	void		(*pr_linkstate)(struct lagg_port *);
181 	void 		(*pr_init)(struct lagg_softc *);
182 	void 		(*pr_stop)(struct lagg_softc *);
183 	void 		(*pr_lladdr)(struct lagg_softc *);
184 	void		(*pr_request)(struct lagg_softc *, void *);
185 	void		(*pr_portreq)(struct lagg_port *, void *);
186 } lagg_protos[] = {
187     {
188 	.pr_num = LAGG_PROTO_NONE
189     },
190     {
191 	.pr_num = LAGG_PROTO_ROUNDROBIN,
192 	.pr_attach = lagg_rr_attach,
193 	.pr_start = lagg_rr_start,
194 	.pr_input = lagg_rr_input,
195     },
196     {
197 	.pr_num = LAGG_PROTO_FAILOVER,
198 	.pr_start = lagg_fail_start,
199 	.pr_input = lagg_fail_input,
200     },
201     {
202 	.pr_num = LAGG_PROTO_LOADBALANCE,
203 	.pr_attach = lagg_lb_attach,
204 	.pr_detach = lagg_lb_detach,
205 	.pr_start = lagg_lb_start,
206 	.pr_input = lagg_lb_input,
207 	.pr_addport = lagg_lb_port_create,
208 	.pr_delport = lagg_lb_port_destroy,
209     },
210     {
211 	.pr_num = LAGG_PROTO_LACP,
212 	.pr_attach = lagg_lacp_attach,
213 	.pr_detach = lagg_lacp_detach,
214 	.pr_start = lagg_lacp_start,
215 	.pr_input = lagg_lacp_input,
216 	.pr_addport = lacp_port_create,
217 	.pr_delport = lacp_port_destroy,
218 	.pr_linkstate = lacp_linkstate,
219 	.pr_init = lacp_init,
220 	.pr_stop = lacp_stop,
221 	.pr_lladdr = lagg_lacp_lladdr,
222 	.pr_request = lacp_req,
223 	.pr_portreq = lacp_portreq,
224     },
225     {
226 	.pr_num = LAGG_PROTO_ETHERCHANNEL,
227 	.pr_attach = lagg_lb_attach,
228 	.pr_detach = lagg_lb_detach,
229 	.pr_start = lagg_lb_start,
230 	.pr_input = lagg_lb_input,
231     },
232     {
233 	.pr_num = LAGG_PROTO_BROADCAST,
234 	.pr_start = lagg_bcast_start,
235 	.pr_input = lagg_bcast_input,
236     },
237 };
238 
239 SYSCTL_DECL(_net_link);
240 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
241     "Link Aggregation");
242 
243 /* Allow input on any failover links */
244 static VNET_DEFINE(int, lagg_failover_rx_all);
245 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
246 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
247     &VNET_NAME(lagg_failover_rx_all), 0,
248     "Accept input from any interface in a failover lagg");
249 
250 /* Default value for using M_FLOWID */
251 static VNET_DEFINE(int, def_use_flowid) = 1;
252 #define	V_def_use_flowid	VNET(def_use_flowid)
253 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
254     &VNET_NAME(def_use_flowid), 0,
255     "Default setting for using flow id for load sharing");
256 
257 /* Default value for using M_FLOWID */
258 static VNET_DEFINE(int, def_flowid_shift) = 16;
259 #define	V_def_flowid_shift	VNET(def_flowid_shift)
260 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
261     &VNET_NAME(def_flowid_shift), 0,
262     "Default setting for flowid shift for load sharing");
263 
264 static void
265 vnet_lagg_init(const void *unused __unused)
266 {
267 
268 	LAGG_LIST_LOCK_INIT();
269 	SLIST_INIT(&V_lagg_list);
270 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
271 	    lagg_clone_destroy, 0);
272 }
273 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
274     vnet_lagg_init, NULL);
275 
276 static void
277 vnet_lagg_uninit(const void *unused __unused)
278 {
279 
280 	if_clone_detach(V_lagg_cloner);
281 	LAGG_LIST_LOCK_DESTROY();
282 }
283 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
284     vnet_lagg_uninit, NULL);
285 
286 static int
287 lagg_modevent(module_t mod, int type, void *data)
288 {
289 
290 	switch (type) {
291 	case MOD_LOAD:
292 		lagg_input_p = lagg_input;
293 		lagg_linkstate_p = lagg_port_state;
294 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
295 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
296 		    EVENTHANDLER_PRI_ANY);
297 		break;
298 	case MOD_UNLOAD:
299 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
300 		    lagg_detach_cookie);
301 		lagg_input_p = NULL;
302 		lagg_linkstate_p = NULL;
303 		break;
304 	default:
305 		return (EOPNOTSUPP);
306 	}
307 	return (0);
308 }
309 
310 static moduledata_t lagg_mod = {
311 	"if_lagg",
312 	lagg_modevent,
313 	0
314 };
315 
316 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
317 MODULE_VERSION(if_lagg, 1);
318 
319 static void
320 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
321 {
322 
323 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
324 	    __func__, sc));
325 
326 	if (sc->sc_ifflags & IFF_DEBUG)
327 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
328 
329 	if (lagg_protos[pr].pr_attach != NULL)
330 		lagg_protos[pr].pr_attach(sc);
331 	sc->sc_proto = pr;
332 }
333 
334 static void
335 lagg_proto_detach(struct lagg_softc *sc)
336 {
337 	lagg_proto pr;
338 
339 	LAGG_WLOCK_ASSERT(sc);
340 
341 	pr = sc->sc_proto;
342 	sc->sc_proto = LAGG_PROTO_NONE;
343 
344 	if (lagg_protos[pr].pr_detach != NULL)
345 		lagg_protos[pr].pr_detach(sc);
346 	else
347 		LAGG_WUNLOCK(sc);
348 }
349 
350 static int
351 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
352 {
353 
354 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
355 }
356 
357 static struct mbuf *
358 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
359 {
360 
361 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
362 }
363 
364 static int
365 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
366 {
367 
368 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
369 		return (0);
370 	else
371 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
372 }
373 
374 static void
375 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
376 {
377 
378 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
379 		lagg_protos[sc->sc_proto].pr_delport(lp);
380 }
381 
382 static void
383 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
384 {
385 
386 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
387 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
388 }
389 
390 static void
391 lagg_proto_init(struct lagg_softc *sc)
392 {
393 
394 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
395 		lagg_protos[sc->sc_proto].pr_init(sc);
396 }
397 
398 static void
399 lagg_proto_stop(struct lagg_softc *sc)
400 {
401 
402 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
403 		lagg_protos[sc->sc_proto].pr_stop(sc);
404 }
405 
406 static void
407 lagg_proto_lladdr(struct lagg_softc *sc)
408 {
409 
410 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
411 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
412 }
413 
414 static void
415 lagg_proto_request(struct lagg_softc *sc, void *v)
416 {
417 
418 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
419 		lagg_protos[sc->sc_proto].pr_request(sc, v);
420 }
421 
422 static void
423 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
424 {
425 
426 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
427 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
428 }
429 
430 /*
431  * This routine is run via an vlan
432  * config EVENT
433  */
434 static void
435 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
436 {
437 	struct lagg_softc *sc = ifp->if_softc;
438 	struct lagg_port *lp;
439 	struct rm_priotracker tracker;
440 
441 	if (ifp->if_softc !=  arg)   /* Not our event */
442 		return;
443 
444 	LAGG_RLOCK(sc, &tracker);
445 	if (!SLIST_EMPTY(&sc->sc_ports)) {
446 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
447 			EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
448 	}
449 	LAGG_RUNLOCK(sc, &tracker);
450 }
451 
452 /*
453  * This routine is run via an vlan
454  * unconfig EVENT
455  */
456 static void
457 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
458 {
459 	struct lagg_softc *sc = ifp->if_softc;
460 	struct lagg_port *lp;
461 	struct rm_priotracker tracker;
462 
463 	if (ifp->if_softc !=  arg)   /* Not our event */
464 		return;
465 
466 	LAGG_RLOCK(sc, &tracker);
467 	if (!SLIST_EMPTY(&sc->sc_ports)) {
468 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
469 			EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
470 	}
471 	LAGG_RUNLOCK(sc, &tracker);
472 }
473 
474 static int
475 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
476 {
477 	struct lagg_softc *sc;
478 	struct ifnet *ifp;
479 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
480 
481 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
482 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
483 	if (ifp == NULL) {
484 		free(sc, M_DEVBUF);
485 		return (ENOSPC);
486 	}
487 
488 	if (V_def_use_flowid)
489 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
490 	sc->flowid_shift = V_def_flowid_shift;
491 
492 	/* Hash all layers by default */
493 	sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4;
494 
495 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
496 
497 	LAGG_LOCK_INIT(sc);
498 	SLIST_INIT(&sc->sc_ports);
499 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
500 
501 	/* Initialise pseudo media types */
502 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
503 	    lagg_media_status);
504 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
505 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
506 
507 	if_initname(ifp, laggname, unit);
508 	ifp->if_softc = sc;
509 	ifp->if_transmit = lagg_transmit;
510 	ifp->if_qflush = lagg_qflush;
511 	ifp->if_init = lagg_init;
512 	ifp->if_ioctl = lagg_ioctl;
513 	ifp->if_get_counter = lagg_get_counter;
514 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
515 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
516 
517 	/*
518 	 * Attach as an ordinary ethernet device, children will be attached
519 	 * as special device IFT_IEEE8023ADLAG.
520 	 */
521 	ether_ifattach(ifp, eaddr);
522 
523 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
524 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
525 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
526 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
527 
528 	/* Insert into the global list of laggs */
529 	LAGG_LIST_LOCK();
530 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
531 	LAGG_LIST_UNLOCK();
532 
533 	return (0);
534 }
535 
536 static void
537 lagg_clone_destroy(struct ifnet *ifp)
538 {
539 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
540 	struct lagg_port *lp;
541 
542 	LAGG_WLOCK(sc);
543 
544 	lagg_stop(sc);
545 	ifp->if_flags &= ~IFF_UP;
546 
547 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
548 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
549 
550 	/* Shutdown and remove lagg ports */
551 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
552 		lagg_port_destroy(lp, 1);
553 	/* Unhook the aggregation protocol */
554 	lagg_proto_detach(sc);
555 
556 	ifmedia_removeall(&sc->sc_media);
557 	ether_ifdetach(ifp);
558 	if_free(ifp);
559 
560 	LAGG_LIST_LOCK();
561 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
562 	LAGG_LIST_UNLOCK();
563 
564 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
565 	LAGG_LOCK_DESTROY(sc);
566 	free(sc, M_DEVBUF);
567 }
568 
569 static void
570 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
571 {
572 	struct lagg_port lp;
573 
574 	LAGG_WLOCK_ASSERT(sc);
575 
576 	bzero(&lp, sizeof(lp));
577 	lp.lp_ifp = sc->sc_ifp;
578 	lp.lp_softc = sc;
579 
580 	lagg_port_lladdr(&lp, lladdr);
581 }
582 
583 static void
584 lagg_capabilities(struct lagg_softc *sc)
585 {
586 	struct lagg_port *lp;
587 	int cap = ~0, ena = ~0;
588 	u_long hwa = ~0UL;
589 	struct ifnet_hw_tsomax hw_tsomax;
590 
591 	LAGG_WLOCK_ASSERT(sc);
592 
593 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
594 
595 	/* Get capabilities from the lagg ports */
596 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
597 		cap &= lp->lp_ifp->if_capabilities;
598 		ena &= lp->lp_ifp->if_capenable;
599 		hwa &= lp->lp_ifp->if_hwassist;
600 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
601 	}
602 	cap = (cap == ~0 ? 0 : cap);
603 	ena = (ena == ~0 ? 0 : ena);
604 	hwa = (hwa == ~0 ? 0 : hwa);
605 
606 	if (sc->sc_ifp->if_capabilities != cap ||
607 	    sc->sc_ifp->if_capenable != ena ||
608 	    sc->sc_ifp->if_hwassist != hwa ||
609 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
610 		sc->sc_ifp->if_capabilities = cap;
611 		sc->sc_ifp->if_capenable = ena;
612 		sc->sc_ifp->if_hwassist = hwa;
613 		getmicrotime(&sc->sc_ifp->if_lastchange);
614 
615 		if (sc->sc_ifflags & IFF_DEBUG)
616 			if_printf(sc->sc_ifp,
617 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
618 	}
619 }
620 
621 static void
622 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
623 {
624 	struct lagg_softc *sc = lp->lp_softc;
625 	struct ifnet *ifp = lp->lp_ifp;
626 	struct lagg_llq *llq;
627 	int pending = 0;
628 
629 	LAGG_WLOCK_ASSERT(sc);
630 
631 	if (lp->lp_detaching ||
632 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
633 		return;
634 
635 	/* Check to make sure its not already queued to be changed */
636 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
637 		if (llq->llq_ifp == ifp) {
638 			pending = 1;
639 			break;
640 		}
641 	}
642 
643 	if (!pending) {
644 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
645 		if (llq == NULL)	/* XXX what to do */
646 			return;
647 	}
648 
649 	/* Update the lladdr even if pending, it may have changed */
650 	llq->llq_ifp = ifp;
651 	llq->llq_primary = (sc->sc_primary->lp_ifp == ifp) ? 1 : 0;
652 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
653 
654 	if (!pending)
655 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
656 
657 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
658 }
659 
660 /*
661  * Set the interface MAC address from a taskqueue to avoid a LOR.
662  */
663 static void
664 lagg_port_setlladdr(void *arg, int pending)
665 {
666 	struct lagg_softc *sc = (struct lagg_softc *)arg;
667 	struct lagg_llq *llq, *head;
668 	struct ifnet *ifp;
669 	int error;
670 
671 	/* Grab a local reference of the queue and remove it from the softc */
672 	LAGG_WLOCK(sc);
673 	head = SLIST_FIRST(&sc->sc_llq_head);
674 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
675 	LAGG_WUNLOCK(sc);
676 
677 	/*
678 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
679 	 * unlocked as we have the only reference to it.
680 	 */
681 	for (llq = head; llq != NULL; llq = head) {
682 		ifp = llq->llq_ifp;
683 
684 		CURVNET_SET(ifp->if_vnet);
685 		if (llq->llq_primary == 0) {
686 			/*
687 			 * Set the link layer address on the laggport interface.
688 			 * if_setlladdr() triggers gratuitous ARPs for INET.
689 			 */
690 			error = if_setlladdr(ifp, llq->llq_lladdr,
691 			    ETHER_ADDR_LEN);
692 			if (error)
693 				printf("%s: setlladdr failed on %s\n", __func__,
694 				    ifp->if_xname);
695 		} else {
696 			/*
697 			 * Set the link layer address on the lagg interface.
698 			 * lagg_proto_lladdr() notifies the MAC change to
699 			 * the aggregation protocol.  iflladdr_event handler
700 			 * may trigger gratuitous ARPs for INET.
701 			 */
702 			if (memcmp(llq->llq_lladdr, IF_LLADDR(ifp),
703 			    ETHER_ADDR_LEN) != 0) {
704 				bcopy(llq->llq_lladdr, IF_LLADDR(ifp),
705 				    ETHER_ADDR_LEN);
706 				LAGG_WLOCK(sc);
707 				lagg_proto_lladdr(sc);
708 				LAGG_WUNLOCK(sc);
709 				EVENTHANDLER_INVOKE(iflladdr_event, ifp);
710 			}
711 		}
712 		CURVNET_RESTORE();
713 		head = SLIST_NEXT(llq, llq_entries);
714 		free(llq, M_DEVBUF);
715 	}
716 }
717 
718 static int
719 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
720 {
721 	struct lagg_softc *sc_ptr;
722 	struct lagg_port *lp, *tlp;
723 	int error, i;
724 	uint64_t *pval;
725 
726 	LAGG_WLOCK_ASSERT(sc);
727 
728 	/* Limit the maximal number of lagg ports */
729 	if (sc->sc_count >= LAGG_MAX_PORTS)
730 		return (ENOSPC);
731 
732 	/* Check if port has already been associated to a lagg */
733 	if (ifp->if_lagg != NULL) {
734 		/* Port is already in the current lagg? */
735 		lp = (struct lagg_port *)ifp->if_lagg;
736 		if (lp->lp_softc == sc)
737 			return (EEXIST);
738 		return (EBUSY);
739 	}
740 
741 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
742 	if (ifp->if_type != IFT_ETHER)
743 		return (EPROTONOSUPPORT);
744 
745 #ifdef INET6
746 	/*
747 	 * The member interface should not have inet6 address because
748 	 * two interfaces with a valid link-local scope zone must not be
749 	 * merged in any form.  This restriction is needed to
750 	 * prevent violation of link-local scope zone.  Attempts to
751 	 * add a member interface which has inet6 addresses triggers
752 	 * removal of all inet6 addresses on the member interface.
753 	 */
754 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
755 		if (in6ifa_llaonifp(lp->lp_ifp)) {
756 			in6_ifdetach(lp->lp_ifp);
757 			if_printf(sc->sc_ifp,
758 			    "IPv6 addresses on %s have been removed "
759 			    "before adding it as a member to prevent "
760 			    "IPv6 address scope violation.\n",
761 			    lp->lp_ifp->if_xname);
762 		}
763 	}
764 	if (in6ifa_llaonifp(ifp)) {
765 		in6_ifdetach(ifp);
766 		if_printf(sc->sc_ifp,
767 		    "IPv6 addresses on %s have been removed "
768 		    "before adding it as a member to prevent "
769 		    "IPv6 address scope violation.\n",
770 		    ifp->if_xname);
771 	}
772 #endif
773 	/* Allow the first Ethernet member to define the MTU */
774 	if (SLIST_EMPTY(&sc->sc_ports))
775 		sc->sc_ifp->if_mtu = ifp->if_mtu;
776 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
777 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
778 		    ifp->if_xname);
779 		return (EINVAL);
780 	}
781 
782 	if ((lp = malloc(sizeof(struct lagg_port),
783 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
784 		return (ENOMEM);
785 
786 	/* Check if port is a stacked lagg */
787 	LAGG_LIST_LOCK();
788 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
789 		if (ifp == sc_ptr->sc_ifp) {
790 			LAGG_LIST_UNLOCK();
791 			free(lp, M_DEVBUF);
792 			return (EINVAL);
793 			/* XXX disable stacking for the moment, its untested */
794 #ifdef LAGG_PORT_STACKING
795 			lp->lp_flags |= LAGG_PORT_STACK;
796 			if (lagg_port_checkstacking(sc_ptr) >=
797 			    LAGG_MAX_STACKING) {
798 				LAGG_LIST_UNLOCK();
799 				free(lp, M_DEVBUF);
800 				return (E2BIG);
801 			}
802 #endif
803 		}
804 	}
805 	LAGG_LIST_UNLOCK();
806 
807 	/* Change the interface type */
808 	lp->lp_iftype = ifp->if_type;
809 	ifp->if_type = IFT_IEEE8023ADLAG;
810 	ifp->if_lagg = lp;
811 	lp->lp_ioctl = ifp->if_ioctl;
812 	ifp->if_ioctl = lagg_port_ioctl;
813 	lp->lp_output = ifp->if_output;
814 	ifp->if_output = lagg_port_output;
815 
816 	lp->lp_ifp = ifp;
817 	lp->lp_softc = sc;
818 
819 	/* Save port link layer address */
820 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
821 
822 	if (SLIST_EMPTY(&sc->sc_ports)) {
823 		sc->sc_primary = lp;
824 		lagg_lladdr(sc, IF_LLADDR(ifp));
825 	} else {
826 		/* Update link layer address for this port */
827 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
828 	}
829 
830 	/* Insert into the list of ports. Keep ports sorted by if_index. */
831 	SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
832 		if (tlp->lp_ifp->if_index < ifp->if_index && (
833 		    SLIST_NEXT(tlp, lp_entries) == NULL ||
834 		    SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index <
835 		    ifp->if_index))
836 			break;
837 	}
838 	if (tlp != NULL)
839 		SLIST_INSERT_AFTER(tlp, lp, lp_entries);
840 	else
841 		SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
842 	sc->sc_count++;
843 
844 	/* Update lagg capabilities */
845 	lagg_capabilities(sc);
846 	lagg_linkstate(sc);
847 
848 	/* Read port counters */
849 	pval = lp->port_counters.val;
850 	for (i = 0; i < IFCOUNTERS; i++, pval++)
851 		*pval = ifp->if_get_counter(ifp, i);
852 	/* Add multicast addresses and interface flags to this port */
853 	lagg_ether_cmdmulti(lp, 1);
854 	lagg_setflags(lp, 1);
855 
856 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
857 		/* Remove the port, without calling pr_delport. */
858 		lagg_port_destroy(lp, 0);
859 		return (error);
860 	}
861 
862 	return (0);
863 }
864 
865 #ifdef LAGG_PORT_STACKING
866 static int
867 lagg_port_checkstacking(struct lagg_softc *sc)
868 {
869 	struct lagg_softc *sc_ptr;
870 	struct lagg_port *lp;
871 	int m = 0;
872 
873 	LAGG_WLOCK_ASSERT(sc);
874 
875 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
876 		if (lp->lp_flags & LAGG_PORT_STACK) {
877 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
878 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
879 		}
880 	}
881 
882 	return (m + 1);
883 }
884 #endif
885 
886 static int
887 lagg_port_destroy(struct lagg_port *lp, int rundelport)
888 {
889 	struct lagg_softc *sc = lp->lp_softc;
890 	struct lagg_port *lp_ptr;
891 	struct lagg_llq *llq;
892 	struct ifnet *ifp = lp->lp_ifp;
893 	uint64_t *pval, vdiff;
894 	int i;
895 
896 	LAGG_WLOCK_ASSERT(sc);
897 
898 	if (rundelport)
899 		lagg_proto_delport(sc, lp);
900 
901 	/*
902 	 * Remove multicast addresses and interface flags from this port and
903 	 * reset the MAC address, skip if the interface is being detached.
904 	 */
905 	if (!lp->lp_detaching) {
906 		lagg_ether_cmdmulti(lp, 0);
907 		lagg_setflags(lp, 0);
908 		lagg_port_lladdr(lp, lp->lp_lladdr);
909 	}
910 
911 	/* Restore interface */
912 	ifp->if_type = lp->lp_iftype;
913 	ifp->if_ioctl = lp->lp_ioctl;
914 	ifp->if_output = lp->lp_output;
915 	ifp->if_lagg = NULL;
916 
917 	/* Update detached port counters */
918 	pval = lp->port_counters.val;
919 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
920 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
921 		sc->detached_counters.val[i] += vdiff;
922 	}
923 
924 	/* Finally, remove the port from the lagg */
925 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
926 	sc->sc_count--;
927 
928 	/* Update the primary interface */
929 	if (lp == sc->sc_primary) {
930 		uint8_t lladdr[ETHER_ADDR_LEN];
931 
932 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
933 			bzero(&lladdr, ETHER_ADDR_LEN);
934 		} else {
935 			bcopy(lp_ptr->lp_lladdr,
936 			    lladdr, ETHER_ADDR_LEN);
937 		}
938 		lagg_lladdr(sc, lladdr);
939 		sc->sc_primary = lp_ptr;
940 
941 		/* Update link layer address for each port */
942 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
943 			lagg_port_lladdr(lp_ptr, lladdr);
944 	}
945 
946 	/* Remove any pending lladdr changes from the queue */
947 	if (lp->lp_detaching) {
948 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
949 			if (llq->llq_ifp == ifp) {
950 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
951 				    llq_entries);
952 				free(llq, M_DEVBUF);
953 				break;	/* Only appears once */
954 			}
955 		}
956 	}
957 
958 	if (lp->lp_ifflags)
959 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
960 
961 	free(lp, M_DEVBUF);
962 
963 	/* Update lagg capabilities */
964 	lagg_capabilities(sc);
965 	lagg_linkstate(sc);
966 
967 	return (0);
968 }
969 
970 static int
971 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
972 {
973 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
974 	struct lagg_softc *sc;
975 	struct lagg_port *lp = NULL;
976 	int error = 0;
977 	struct rm_priotracker tracker;
978 
979 	/* Should be checked by the caller */
980 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
981 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
982 		goto fallback;
983 
984 	switch (cmd) {
985 	case SIOCGLAGGPORT:
986 		if (rp->rp_portname[0] == '\0' ||
987 		    ifunit(rp->rp_portname) != ifp) {
988 			error = EINVAL;
989 			break;
990 		}
991 
992 		LAGG_RLOCK(sc, &tracker);
993 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
994 			error = ENOENT;
995 			LAGG_RUNLOCK(sc, &tracker);
996 			break;
997 		}
998 
999 		lagg_port2req(lp, rp);
1000 		LAGG_RUNLOCK(sc, &tracker);
1001 		break;
1002 
1003 	case SIOCSIFCAP:
1004 		if (lp->lp_ioctl == NULL) {
1005 			error = EINVAL;
1006 			break;
1007 		}
1008 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1009 		if (error)
1010 			break;
1011 
1012 		/* Update lagg interface capabilities */
1013 		LAGG_WLOCK(sc);
1014 		lagg_capabilities(sc);
1015 		LAGG_WUNLOCK(sc);
1016 		break;
1017 
1018 	case SIOCSIFMTU:
1019 		/* Do not allow the MTU to be changed once joined */
1020 		error = EINVAL;
1021 		break;
1022 
1023 	default:
1024 		goto fallback;
1025 	}
1026 
1027 	return (error);
1028 
1029 fallback:
1030 	if (lp->lp_ioctl != NULL)
1031 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1032 
1033 	return (EINVAL);
1034 }
1035 
1036 /*
1037  * Requests counter @cnt data.
1038  *
1039  * Counter value is calculated the following way:
1040  * 1) for each port, sum  difference between current and "initial" measurements.
1041  * 2) add lagg logical interface counters.
1042  * 3) add data from detached_counters array.
1043  *
1044  * We also do the following things on ports attach/detach:
1045  * 1) On port attach we store all counters it has into port_counter array.
1046  * 2) On port detach we add the different between "initial" and
1047  *   current counters data to detached_counters array.
1048  */
1049 static uint64_t
1050 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1051 {
1052 	struct lagg_softc *sc;
1053 	struct lagg_port *lp;
1054 	struct ifnet *lpifp;
1055 	struct rm_priotracker tracker;
1056 	uint64_t newval, oldval, vsum;
1057 
1058 	/* Revise this when we've got non-generic counters. */
1059 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1060 
1061 	sc = (struct lagg_softc *)ifp->if_softc;
1062 	LAGG_RLOCK(sc, &tracker);
1063 
1064 	vsum = 0;
1065 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1066 		/* Saved attached value */
1067 		oldval = lp->port_counters.val[cnt];
1068 		/* current value */
1069 		lpifp = lp->lp_ifp;
1070 		newval = lpifp->if_get_counter(lpifp, cnt);
1071 		/* Calculate diff and save new */
1072 		vsum += newval - oldval;
1073 	}
1074 
1075 	/*
1076 	 * Add counter data which might be added by upper
1077 	 * layer protocols operating on logical interface.
1078 	 */
1079 	vsum += if_get_counter_default(ifp, cnt);
1080 
1081 	/*
1082 	 * Add counter data from detached ports counters
1083 	 */
1084 	vsum += sc->detached_counters.val[cnt];
1085 
1086 	LAGG_RUNLOCK(sc, &tracker);
1087 
1088 	return (vsum);
1089 }
1090 
1091 /*
1092  * For direct output to child ports.
1093  */
1094 static int
1095 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1096 	const struct sockaddr *dst, struct route *ro)
1097 {
1098 	struct lagg_port *lp = ifp->if_lagg;
1099 
1100 	switch (dst->sa_family) {
1101 		case pseudo_AF_HDRCMPLT:
1102 		case AF_UNSPEC:
1103 			return ((*lp->lp_output)(ifp, m, dst, ro));
1104 	}
1105 
1106 	/* drop any other frames */
1107 	m_freem(m);
1108 	return (ENETDOWN);
1109 }
1110 
1111 static void
1112 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1113 {
1114 	struct lagg_port *lp;
1115 	struct lagg_softc *sc;
1116 
1117 	if ((lp = ifp->if_lagg) == NULL)
1118 		return;
1119 	/* If the ifnet is just being renamed, don't do anything. */
1120 	if (ifp->if_flags & IFF_RENAMING)
1121 		return;
1122 
1123 	sc = lp->lp_softc;
1124 
1125 	LAGG_WLOCK(sc);
1126 	lp->lp_detaching = 1;
1127 	lagg_port_destroy(lp, 1);
1128 	LAGG_WUNLOCK(sc);
1129 }
1130 
1131 static void
1132 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1133 {
1134 	struct lagg_softc *sc = lp->lp_softc;
1135 
1136 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1137 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1138 	rp->rp_prio = lp->lp_prio;
1139 	rp->rp_flags = lp->lp_flags;
1140 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1141 
1142 	/* Add protocol specific flags */
1143 	switch (sc->sc_proto) {
1144 		case LAGG_PROTO_FAILOVER:
1145 			if (lp == sc->sc_primary)
1146 				rp->rp_flags |= LAGG_PORT_MASTER;
1147 			if (lp == lagg_link_active(sc, sc->sc_primary))
1148 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1149 			break;
1150 
1151 		case LAGG_PROTO_ROUNDROBIN:
1152 		case LAGG_PROTO_LOADBALANCE:
1153 		case LAGG_PROTO_ETHERCHANNEL:
1154 		case LAGG_PROTO_BROADCAST:
1155 			if (LAGG_PORTACTIVE(lp))
1156 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1157 			break;
1158 
1159 		case LAGG_PROTO_LACP:
1160 			/* LACP has a different definition of active */
1161 			if (lacp_isactive(lp))
1162 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1163 			if (lacp_iscollecting(lp))
1164 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1165 			if (lacp_isdistributing(lp))
1166 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1167 			break;
1168 	}
1169 
1170 }
1171 
1172 static void
1173 lagg_init(void *xsc)
1174 {
1175 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1176 	struct lagg_port *lp;
1177 	struct ifnet *ifp = sc->sc_ifp;
1178 
1179 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1180 		return;
1181 
1182 	LAGG_WLOCK(sc);
1183 
1184 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1185 	/* Update the port lladdrs */
1186 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1187 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
1188 
1189 	lagg_proto_init(sc);
1190 
1191 	LAGG_WUNLOCK(sc);
1192 }
1193 
1194 static void
1195 lagg_stop(struct lagg_softc *sc)
1196 {
1197 	struct ifnet *ifp = sc->sc_ifp;
1198 
1199 	LAGG_WLOCK_ASSERT(sc);
1200 
1201 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1202 		return;
1203 
1204 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1205 
1206 	lagg_proto_stop(sc);
1207 }
1208 
1209 static int
1210 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1211 {
1212 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1213 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1214 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1215 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1216 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1217 	struct ifreq *ifr = (struct ifreq *)data;
1218 	struct lagg_port *lp;
1219 	struct ifnet *tpif;
1220 	struct thread *td = curthread;
1221 	char *buf, *outbuf;
1222 	int count, buflen, len, error = 0;
1223 	struct rm_priotracker tracker;
1224 
1225 	bzero(&rpbuf, sizeof(rpbuf));
1226 
1227 	switch (cmd) {
1228 	case SIOCGLAGG:
1229 		LAGG_RLOCK(sc, &tracker);
1230 		count = 0;
1231 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1232 			count++;
1233 		buflen = count * sizeof(struct lagg_reqport);
1234 		LAGG_RUNLOCK(sc, &tracker);
1235 
1236 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1237 
1238 		LAGG_RLOCK(sc, &tracker);
1239 		ra->ra_proto = sc->sc_proto;
1240 		lagg_proto_request(sc, &ra->ra_psc);
1241 		count = 0;
1242 		buf = outbuf;
1243 		len = min(ra->ra_size, buflen);
1244 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1245 			if (len < sizeof(rpbuf))
1246 				break;
1247 
1248 			lagg_port2req(lp, &rpbuf);
1249 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1250 			count++;
1251 			buf += sizeof(rpbuf);
1252 			len -= sizeof(rpbuf);
1253 		}
1254 		LAGG_RUNLOCK(sc, &tracker);
1255 		ra->ra_ports = count;
1256 		ra->ra_size = count * sizeof(rpbuf);
1257 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1258 		free(outbuf, M_TEMP);
1259 		break;
1260 	case SIOCSLAGG:
1261 		error = priv_check(td, PRIV_NET_LAGG);
1262 		if (error)
1263 			break;
1264 		if (ra->ra_proto < 1 || ra->ra_proto >= LAGG_PROTO_MAX) {
1265 			error = EPROTONOSUPPORT;
1266 			break;
1267 		}
1268 
1269 		LAGG_WLOCK(sc);
1270 		lagg_proto_detach(sc);
1271 		lagg_proto_attach(sc, ra->ra_proto);
1272 		break;
1273 	case SIOCGLAGGOPTS:
1274 		ro->ro_opts = sc->sc_opts;
1275 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1276 			struct lacp_softc *lsc;
1277 
1278 			lsc = (struct lacp_softc *)sc->sc_psc;
1279 			if (lsc->lsc_debug.lsc_tx_test != 0)
1280 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1281 			if (lsc->lsc_debug.lsc_rx_test != 0)
1282 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1283 			if (lsc->lsc_strict_mode != 0)
1284 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1285 
1286 			ro->ro_active = sc->sc_active;
1287 		} else {
1288 			ro->ro_active = 0;
1289 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1290 				ro->ro_active += LAGG_PORTACTIVE(lp);
1291 		}
1292 		ro->ro_flapping = sc->sc_flapping;
1293 		ro->ro_flowid_shift = sc->flowid_shift;
1294 		break;
1295 	case SIOCSLAGGOPTS:
1296 		error = priv_check(td, PRIV_NET_LAGG);
1297 		if (error)
1298 			break;
1299 		if (ro->ro_opts == 0)
1300 			break;
1301 		/*
1302 		 * Set options.  LACP options are stored in sc->sc_psc,
1303 		 * not in sc_opts.
1304 		 */
1305 		int valid, lacp;
1306 
1307 		switch (ro->ro_opts) {
1308 		case LAGG_OPT_USE_FLOWID:
1309 		case -LAGG_OPT_USE_FLOWID:
1310 		case LAGG_OPT_FLOWIDSHIFT:
1311 			valid = 1;
1312 			lacp = 0;
1313 			break;
1314 		case LAGG_OPT_LACP_TXTEST:
1315 		case -LAGG_OPT_LACP_TXTEST:
1316 		case LAGG_OPT_LACP_RXTEST:
1317 		case -LAGG_OPT_LACP_RXTEST:
1318 		case LAGG_OPT_LACP_STRICT:
1319 		case -LAGG_OPT_LACP_STRICT:
1320 			valid = lacp = 1;
1321 			break;
1322 		default:
1323 			valid = lacp = 0;
1324 			break;
1325 		}
1326 
1327 		LAGG_WLOCK(sc);
1328 		if (valid == 0 ||
1329 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1330 			/* Invalid combination of options specified. */
1331 			error = EINVAL;
1332 			LAGG_WUNLOCK(sc);
1333 			break;	/* Return from SIOCSLAGGOPTS. */
1334 		}
1335 		/*
1336 		 * Store new options into sc->sc_opts except for
1337 		 * FLOWIDSHIFT and LACP options.
1338 		 */
1339 		if (lacp == 0) {
1340 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1341 				sc->flowid_shift = ro->ro_flowid_shift;
1342 			else if (ro->ro_opts > 0)
1343 				sc->sc_opts |= ro->ro_opts;
1344 			else
1345 				sc->sc_opts &= ~ro->ro_opts;
1346 		} else {
1347 			struct lacp_softc *lsc;
1348 
1349 			lsc = (struct lacp_softc *)sc->sc_psc;
1350 
1351 			switch (ro->ro_opts) {
1352 			case LAGG_OPT_LACP_TXTEST:
1353 				lsc->lsc_debug.lsc_tx_test = 1;
1354 				break;
1355 			case -LAGG_OPT_LACP_TXTEST:
1356 				lsc->lsc_debug.lsc_tx_test = 0;
1357 				break;
1358 			case LAGG_OPT_LACP_RXTEST:
1359 				lsc->lsc_debug.lsc_rx_test = 1;
1360 				break;
1361 			case -LAGG_OPT_LACP_RXTEST:
1362 				lsc->lsc_debug.lsc_rx_test = 0;
1363 				break;
1364 			case LAGG_OPT_LACP_STRICT:
1365 				lsc->lsc_strict_mode = 1;
1366 				break;
1367 			case -LAGG_OPT_LACP_STRICT:
1368 				lsc->lsc_strict_mode = 0;
1369 				break;
1370 			}
1371 		}
1372 		LAGG_WUNLOCK(sc);
1373 		break;
1374 	case SIOCGLAGGFLAGS:
1375 		rf->rf_flags = sc->sc_flags;
1376 		break;
1377 	case SIOCSLAGGHASH:
1378 		error = priv_check(td, PRIV_NET_LAGG);
1379 		if (error)
1380 			break;
1381 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1382 			error = EINVAL;
1383 			break;
1384 		}
1385 		LAGG_WLOCK(sc);
1386 		sc->sc_flags &= ~LAGG_F_HASHMASK;
1387 		sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK;
1388 		LAGG_WUNLOCK(sc);
1389 		break;
1390 	case SIOCGLAGGPORT:
1391 		if (rp->rp_portname[0] == '\0' ||
1392 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1393 			error = EINVAL;
1394 			break;
1395 		}
1396 
1397 		LAGG_RLOCK(sc, &tracker);
1398 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1399 		    lp->lp_softc != sc) {
1400 			error = ENOENT;
1401 			LAGG_RUNLOCK(sc, &tracker);
1402 			break;
1403 		}
1404 
1405 		lagg_port2req(lp, rp);
1406 		LAGG_RUNLOCK(sc, &tracker);
1407 		break;
1408 	case SIOCSLAGGPORT:
1409 		error = priv_check(td, PRIV_NET_LAGG);
1410 		if (error)
1411 			break;
1412 		if (rp->rp_portname[0] == '\0' ||
1413 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1414 			error = EINVAL;
1415 			break;
1416 		}
1417 		LAGG_WLOCK(sc);
1418 		error = lagg_port_create(sc, tpif);
1419 		LAGG_WUNLOCK(sc);
1420 		break;
1421 	case SIOCSLAGGDELPORT:
1422 		error = priv_check(td, PRIV_NET_LAGG);
1423 		if (error)
1424 			break;
1425 		if (rp->rp_portname[0] == '\0' ||
1426 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1427 			error = EINVAL;
1428 			break;
1429 		}
1430 
1431 		LAGG_WLOCK(sc);
1432 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1433 		    lp->lp_softc != sc) {
1434 			error = ENOENT;
1435 			LAGG_WUNLOCK(sc);
1436 			break;
1437 		}
1438 
1439 		error = lagg_port_destroy(lp, 1);
1440 		LAGG_WUNLOCK(sc);
1441 		break;
1442 	case SIOCSIFFLAGS:
1443 		/* Set flags on ports too */
1444 		LAGG_WLOCK(sc);
1445 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1446 			lagg_setflags(lp, 1);
1447 		}
1448 		LAGG_WUNLOCK(sc);
1449 
1450 		if (!(ifp->if_flags & IFF_UP) &&
1451 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1452 			/*
1453 			 * If interface is marked down and it is running,
1454 			 * then stop and disable it.
1455 			 */
1456 			LAGG_WLOCK(sc);
1457 			lagg_stop(sc);
1458 			LAGG_WUNLOCK(sc);
1459 		} else if ((ifp->if_flags & IFF_UP) &&
1460 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1461 			/*
1462 			 * If interface is marked up and it is stopped, then
1463 			 * start it.
1464 			 */
1465 			(*ifp->if_init)(sc);
1466 		}
1467 		break;
1468 	case SIOCADDMULTI:
1469 	case SIOCDELMULTI:
1470 		LAGG_WLOCK(sc);
1471 		error = lagg_ether_setmulti(sc);
1472 		LAGG_WUNLOCK(sc);
1473 		break;
1474 	case SIOCSIFMEDIA:
1475 	case SIOCGIFMEDIA:
1476 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1477 		break;
1478 
1479 	case SIOCSIFCAP:
1480 	case SIOCSIFMTU:
1481 		/* Do not allow the MTU or caps to be directly changed */
1482 		error = EINVAL;
1483 		break;
1484 
1485 	default:
1486 		error = ether_ioctl(ifp, cmd, data);
1487 		break;
1488 	}
1489 	return (error);
1490 }
1491 
1492 static int
1493 lagg_ether_setmulti(struct lagg_softc *sc)
1494 {
1495 	struct lagg_port *lp;
1496 
1497 	LAGG_WLOCK_ASSERT(sc);
1498 
1499 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1500 		/* First, remove any existing filter entries. */
1501 		lagg_ether_cmdmulti(lp, 0);
1502 		/* copy all addresses from the lagg interface to the port */
1503 		lagg_ether_cmdmulti(lp, 1);
1504 	}
1505 	return (0);
1506 }
1507 
1508 static int
1509 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1510 {
1511 	struct lagg_softc *sc = lp->lp_softc;
1512 	struct ifnet *ifp = lp->lp_ifp;
1513 	struct ifnet *scifp = sc->sc_ifp;
1514 	struct lagg_mc *mc;
1515 	struct ifmultiaddr *ifma;
1516 	int error;
1517 
1518 	LAGG_WLOCK_ASSERT(sc);
1519 
1520 	if (set) {
1521 		IF_ADDR_WLOCK(scifp);
1522 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1523 			if (ifma->ifma_addr->sa_family != AF_LINK)
1524 				continue;
1525 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1526 			if (mc == NULL) {
1527 				IF_ADDR_WUNLOCK(scifp);
1528 				return (ENOMEM);
1529 			}
1530 			bcopy(ifma->ifma_addr, &mc->mc_addr,
1531 			    ifma->ifma_addr->sa_len);
1532 			mc->mc_addr.sdl_index = ifp->if_index;
1533 			mc->mc_ifma = NULL;
1534 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1535 		}
1536 		IF_ADDR_WUNLOCK(scifp);
1537 		SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1538 			error = if_addmulti(ifp,
1539 			    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1540 			if (error)
1541 				return (error);
1542 		}
1543 	} else {
1544 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1545 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1546 			if (mc->mc_ifma && !lp->lp_detaching)
1547 				if_delmulti_ifma(mc->mc_ifma);
1548 			free(mc, M_DEVBUF);
1549 		}
1550 	}
1551 	return (0);
1552 }
1553 
1554 /* Handle a ref counted flag that should be set on the lagg port as well */
1555 static int
1556 lagg_setflag(struct lagg_port *lp, int flag, int status,
1557     int (*func)(struct ifnet *, int))
1558 {
1559 	struct lagg_softc *sc = lp->lp_softc;
1560 	struct ifnet *scifp = sc->sc_ifp;
1561 	struct ifnet *ifp = lp->lp_ifp;
1562 	int error;
1563 
1564 	LAGG_WLOCK_ASSERT(sc);
1565 
1566 	status = status ? (scifp->if_flags & flag) : 0;
1567 	/* Now "status" contains the flag value or 0 */
1568 
1569 	/*
1570 	 * See if recorded ports status is different from what
1571 	 * we want it to be.  If it is, flip it.  We record ports
1572 	 * status in lp_ifflags so that we won't clear ports flag
1573 	 * we haven't set.  In fact, we don't clear or set ports
1574 	 * flags directly, but get or release references to them.
1575 	 * That's why we can be sure that recorded flags still are
1576 	 * in accord with actual ports flags.
1577 	 */
1578 	if (status != (lp->lp_ifflags & flag)) {
1579 		error = (*func)(ifp, status);
1580 		if (error)
1581 			return (error);
1582 		lp->lp_ifflags &= ~flag;
1583 		lp->lp_ifflags |= status;
1584 	}
1585 	return (0);
1586 }
1587 
1588 /*
1589  * Handle IFF_* flags that require certain changes on the lagg port
1590  * if "status" is true, update ports flags respective to the lagg
1591  * if "status" is false, forcedly clear the flags set on port.
1592  */
1593 static int
1594 lagg_setflags(struct lagg_port *lp, int status)
1595 {
1596 	int error, i;
1597 
1598 	for (i = 0; lagg_pflags[i].flag; i++) {
1599 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1600 		    status, lagg_pflags[i].func);
1601 		if (error)
1602 			return (error);
1603 	}
1604 	return (0);
1605 }
1606 
1607 static int
1608 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1609 {
1610 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1611 	int error, len, mcast;
1612 	struct rm_priotracker tracker;
1613 
1614 	len = m->m_pkthdr.len;
1615 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1616 
1617 	LAGG_RLOCK(sc, &tracker);
1618 	/* We need a Tx algorithm and at least one port */
1619 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1620 		LAGG_RUNLOCK(sc, &tracker);
1621 		m_freem(m);
1622 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1623 		return (ENXIO);
1624 	}
1625 
1626 	ETHER_BPF_MTAP(ifp, m);
1627 
1628 	error = lagg_proto_start(sc, m);
1629 	LAGG_RUNLOCK(sc, &tracker);
1630 
1631 	if (error != 0)
1632 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1633 
1634 	return (error);
1635 }
1636 
1637 /*
1638  * The ifp->if_qflush entry point for lagg(4) is no-op.
1639  */
1640 static void
1641 lagg_qflush(struct ifnet *ifp __unused)
1642 {
1643 }
1644 
1645 static struct mbuf *
1646 lagg_input(struct ifnet *ifp, struct mbuf *m)
1647 {
1648 	struct lagg_port *lp = ifp->if_lagg;
1649 	struct lagg_softc *sc = lp->lp_softc;
1650 	struct ifnet *scifp = sc->sc_ifp;
1651 	struct rm_priotracker tracker;
1652 
1653 	LAGG_RLOCK(sc, &tracker);
1654 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1655 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1656 	    sc->sc_proto == LAGG_PROTO_NONE) {
1657 		LAGG_RUNLOCK(sc, &tracker);
1658 		m_freem(m);
1659 		return (NULL);
1660 	}
1661 
1662 	ETHER_BPF_MTAP(scifp, m);
1663 
1664 	m = (lp->lp_detaching == 0) ? lagg_proto_input(sc, lp, m) : NULL;
1665 
1666 	if (m != NULL) {
1667 		if (scifp->if_flags & IFF_MONITOR) {
1668 			m_freem(m);
1669 			m = NULL;
1670 		}
1671 	}
1672 
1673 	LAGG_RUNLOCK(sc, &tracker);
1674 	return (m);
1675 }
1676 
1677 static int
1678 lagg_media_change(struct ifnet *ifp)
1679 {
1680 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1681 
1682 	if (sc->sc_ifflags & IFF_DEBUG)
1683 		printf("%s\n", __func__);
1684 
1685 	/* Ignore */
1686 	return (0);
1687 }
1688 
1689 static void
1690 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1691 {
1692 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1693 	struct lagg_port *lp;
1694 	struct rm_priotracker tracker;
1695 
1696 	imr->ifm_status = IFM_AVALID;
1697 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1698 
1699 	LAGG_RLOCK(sc, &tracker);
1700 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1701 		if (LAGG_PORTACTIVE(lp))
1702 			imr->ifm_status |= IFM_ACTIVE;
1703 	}
1704 	LAGG_RUNLOCK(sc, &tracker);
1705 }
1706 
1707 static void
1708 lagg_linkstate(struct lagg_softc *sc)
1709 {
1710 	struct lagg_port *lp;
1711 	int new_link = LINK_STATE_DOWN;
1712 	uint64_t speed;
1713 
1714 	/* Our link is considered up if at least one of our ports is active */
1715 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1716 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1717 			new_link = LINK_STATE_UP;
1718 			break;
1719 		}
1720 	}
1721 	if_link_state_change(sc->sc_ifp, new_link);
1722 
1723 	/* Update if_baudrate to reflect the max possible speed */
1724 	switch (sc->sc_proto) {
1725 		case LAGG_PROTO_FAILOVER:
1726 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1727 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1728 			break;
1729 		case LAGG_PROTO_ROUNDROBIN:
1730 		case LAGG_PROTO_LOADBALANCE:
1731 		case LAGG_PROTO_ETHERCHANNEL:
1732 		case LAGG_PROTO_BROADCAST:
1733 			speed = 0;
1734 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1735 				speed += lp->lp_ifp->if_baudrate;
1736 			sc->sc_ifp->if_baudrate = speed;
1737 			break;
1738 		case LAGG_PROTO_LACP:
1739 			/* LACP updates if_baudrate itself */
1740 			break;
1741 	}
1742 }
1743 
1744 static void
1745 lagg_port_state(struct ifnet *ifp, int state)
1746 {
1747 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1748 	struct lagg_softc *sc = NULL;
1749 
1750 	if (lp != NULL)
1751 		sc = lp->lp_softc;
1752 	if (sc == NULL)
1753 		return;
1754 
1755 	LAGG_WLOCK(sc);
1756 	lagg_linkstate(sc);
1757 	lagg_proto_linkstate(sc, lp);
1758 	LAGG_WUNLOCK(sc);
1759 }
1760 
1761 struct lagg_port *
1762 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1763 {
1764 	struct lagg_port *lp_next, *rval = NULL;
1765 	// int new_link = LINK_STATE_DOWN;
1766 
1767 	LAGG_RLOCK_ASSERT(sc);
1768 	/*
1769 	 * Search a port which reports an active link state.
1770 	 */
1771 
1772 	if (lp == NULL)
1773 		goto search;
1774 	if (LAGG_PORTACTIVE(lp)) {
1775 		rval = lp;
1776 		goto found;
1777 	}
1778 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1779 	    LAGG_PORTACTIVE(lp_next)) {
1780 		rval = lp_next;
1781 		goto found;
1782 	}
1783 
1784 search:
1785 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1786 		if (LAGG_PORTACTIVE(lp_next)) {
1787 			rval = lp_next;
1788 			goto found;
1789 		}
1790 	}
1791 
1792 found:
1793 	if (rval != NULL) {
1794 		/*
1795 		 * The IEEE 802.1D standard assumes that a lagg with
1796 		 * multiple ports is always full duplex. This is valid
1797 		 * for load sharing laggs and if at least two links
1798 		 * are active. Unfortunately, checking the latter would
1799 		 * be too expensive at this point.
1800 		 XXX
1801 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1802 		    (sc->sc_count > 1))
1803 			new_link = LINK_STATE_FULL_DUPLEX;
1804 		else
1805 			new_link = rval->lp_link_state;
1806 		 */
1807 	}
1808 
1809 	return (rval);
1810 }
1811 
1812 static const void *
1813 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1814 {
1815 	if (m->m_pkthdr.len < (off + len)) {
1816 		return (NULL);
1817 	} else if (m->m_len < (off + len)) {
1818 		m_copydata(m, off, len, buf);
1819 		return (buf);
1820 	}
1821 	return (mtod(m, char *) + off);
1822 }
1823 
1824 uint32_t
1825 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
1826 {
1827 	uint16_t etype;
1828 	uint32_t p = key;
1829 	int off;
1830 	struct ether_header *eh;
1831 	const struct ether_vlan_header *vlan;
1832 #ifdef INET
1833 	const struct ip *ip;
1834 	const uint32_t *ports;
1835 	int iphlen;
1836 #endif
1837 #ifdef INET6
1838 	const struct ip6_hdr *ip6;
1839 	uint32_t flow;
1840 #endif
1841 	union {
1842 #ifdef INET
1843 		struct ip ip;
1844 #endif
1845 #ifdef INET6
1846 		struct ip6_hdr ip6;
1847 #endif
1848 		struct ether_vlan_header vlan;
1849 		uint32_t port;
1850 	} buf;
1851 
1852 
1853 	off = sizeof(*eh);
1854 	if (m->m_len < off)
1855 		goto out;
1856 	eh = mtod(m, struct ether_header *);
1857 	etype = ntohs(eh->ether_type);
1858 	if (sc->sc_flags & LAGG_F_HASHL2) {
1859 		p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p);
1860 		p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1861 	}
1862 
1863 	/* Special handling for encapsulating VLAN frames */
1864 	if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) {
1865 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1866 		    sizeof(m->m_pkthdr.ether_vtag), p);
1867 	} else if (etype == ETHERTYPE_VLAN) {
1868 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &buf);
1869 		if (vlan == NULL)
1870 			goto out;
1871 
1872 		if (sc->sc_flags & LAGG_F_HASHL2)
1873 			p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1874 		etype = ntohs(vlan->evl_proto);
1875 		off += sizeof(*vlan) - sizeof(*eh);
1876 	}
1877 
1878 	switch (etype) {
1879 #ifdef INET
1880 	case ETHERTYPE_IP:
1881 		ip = lagg_gethdr(m, off, sizeof(*ip), &buf);
1882 		if (ip == NULL)
1883 			goto out;
1884 
1885 		if (sc->sc_flags & LAGG_F_HASHL3) {
1886 			p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1887 			p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1888 		}
1889 		if (!(sc->sc_flags & LAGG_F_HASHL4))
1890 			break;
1891 		switch (ip->ip_p) {
1892 			case IPPROTO_TCP:
1893 			case IPPROTO_UDP:
1894 			case IPPROTO_SCTP:
1895 				iphlen = ip->ip_hl << 2;
1896 				if (iphlen < sizeof(*ip))
1897 					break;
1898 				off += iphlen;
1899 				ports = lagg_gethdr(m, off, sizeof(*ports), &buf);
1900 				if (ports == NULL)
1901 					break;
1902 				p = hash32_buf(ports, sizeof(*ports), p);
1903 				break;
1904 		}
1905 		break;
1906 #endif
1907 #ifdef INET6
1908 	case ETHERTYPE_IPV6:
1909 		if (!(sc->sc_flags & LAGG_F_HASHL3))
1910 			break;
1911 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf);
1912 		if (ip6 == NULL)
1913 			goto out;
1914 
1915 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1916 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1917 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1918 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1919 		break;
1920 #endif
1921 	}
1922 out:
1923 	return (p);
1924 }
1925 
1926 int
1927 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1928 {
1929 
1930 	return (ifp->if_transmit)(ifp, m);
1931 }
1932 
1933 /*
1934  * Simple round robin aggregation
1935  */
1936 static void
1937 lagg_rr_attach(struct lagg_softc *sc)
1938 {
1939 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1940 	sc->sc_seq = 0;
1941 }
1942 
1943 static int
1944 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1945 {
1946 	struct lagg_port *lp;
1947 	uint32_t p;
1948 
1949 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1950 	p %= sc->sc_count;
1951 	lp = SLIST_FIRST(&sc->sc_ports);
1952 	while (p--)
1953 		lp = SLIST_NEXT(lp, lp_entries);
1954 
1955 	/*
1956 	 * Check the port's link state. This will return the next active
1957 	 * port if the link is down or the port is NULL.
1958 	 */
1959 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1960 		m_freem(m);
1961 		return (ENETDOWN);
1962 	}
1963 
1964 	/* Send mbuf */
1965 	return (lagg_enqueue(lp->lp_ifp, m));
1966 }
1967 
1968 static struct mbuf *
1969 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1970 {
1971 	struct ifnet *ifp = sc->sc_ifp;
1972 
1973 	/* Just pass in the packet to our lagg device */
1974 	m->m_pkthdr.rcvif = ifp;
1975 
1976 	return (m);
1977 }
1978 
1979 /*
1980  * Broadcast mode
1981  */
1982 static int
1983 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1984 {
1985 	int active_ports = 0;
1986 	int errors = 0;
1987 	int ret;
1988 	struct lagg_port *lp, *last = NULL;
1989 	struct mbuf *m0;
1990 
1991 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1992 		if (!LAGG_PORTACTIVE(lp))
1993 			continue;
1994 
1995 		active_ports++;
1996 
1997 		if (last != NULL) {
1998 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
1999 			if (m0 == NULL) {
2000 				ret = ENOBUFS;
2001 				errors++;
2002 				break;
2003 			}
2004 
2005 			ret = lagg_enqueue(last->lp_ifp, m0);
2006 			if (ret != 0)
2007 				errors++;
2008 		}
2009 		last = lp;
2010 	}
2011 	if (last == NULL) {
2012 		m_freem(m);
2013 		return (ENOENT);
2014 	}
2015 	if ((last = lagg_link_active(sc, last)) == NULL) {
2016 		m_freem(m);
2017 		return (ENETDOWN);
2018 	}
2019 
2020 	ret = lagg_enqueue(last->lp_ifp, m);
2021 	if (ret != 0)
2022 		errors++;
2023 
2024 	if (errors == 0)
2025 		return (ret);
2026 
2027 	return (0);
2028 }
2029 
2030 static struct mbuf*
2031 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2032 {
2033 	struct ifnet *ifp = sc->sc_ifp;
2034 
2035 	/* Just pass in the packet to our lagg device */
2036 	m->m_pkthdr.rcvif = ifp;
2037 	return (m);
2038 }
2039 
2040 /*
2041  * Active failover
2042  */
2043 static int
2044 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2045 {
2046 	struct lagg_port *lp;
2047 
2048 	/* Use the master port if active or the next available port */
2049 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2050 		m_freem(m);
2051 		return (ENETDOWN);
2052 	}
2053 
2054 	/* Send mbuf */
2055 	return (lagg_enqueue(lp->lp_ifp, m));
2056 }
2057 
2058 static struct mbuf *
2059 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2060 {
2061 	struct ifnet *ifp = sc->sc_ifp;
2062 	struct lagg_port *tmp_tp;
2063 
2064 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2065 		m->m_pkthdr.rcvif = ifp;
2066 		return (m);
2067 	}
2068 
2069 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2070 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2071 		/*
2072 		 * If tmp_tp is null, we've recieved a packet when all
2073 		 * our links are down. Weird, but process it anyways.
2074 		 */
2075 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2076 			m->m_pkthdr.rcvif = ifp;
2077 			return (m);
2078 		}
2079 	}
2080 
2081 	m_freem(m);
2082 	return (NULL);
2083 }
2084 
2085 /*
2086  * Loadbalancing
2087  */
2088 static void
2089 lagg_lb_attach(struct lagg_softc *sc)
2090 {
2091 	struct lagg_port *lp;
2092 	struct lagg_lb *lb;
2093 
2094 	lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
2095 
2096 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
2097 
2098 	lb->lb_key = arc4random();
2099 	sc->sc_psc = lb;
2100 
2101 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2102 		lagg_lb_port_create(lp);
2103 }
2104 
2105 static void
2106 lagg_lb_detach(struct lagg_softc *sc)
2107 {
2108 	struct lagg_lb *lb;
2109 
2110 	lb = (struct lagg_lb *)sc->sc_psc;
2111 	LAGG_WUNLOCK(sc);
2112 	if (lb != NULL)
2113 		free(lb, M_DEVBUF);
2114 }
2115 
2116 static int
2117 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2118 {
2119 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2120 	struct lagg_port *lp_next;
2121 	int i = 0;
2122 
2123 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2124 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2125 		if (lp_next == lp)
2126 			continue;
2127 		if (i >= LAGG_MAX_PORTS)
2128 			return (EINVAL);
2129 		if (sc->sc_ifflags & IFF_DEBUG)
2130 			printf("%s: port %s at index %d\n",
2131 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2132 		lb->lb_ports[i++] = lp_next;
2133 	}
2134 
2135 	return (0);
2136 }
2137 
2138 static int
2139 lagg_lb_port_create(struct lagg_port *lp)
2140 {
2141 	struct lagg_softc *sc = lp->lp_softc;
2142 	return (lagg_lb_porttable(sc, NULL));
2143 }
2144 
2145 static void
2146 lagg_lb_port_destroy(struct lagg_port *lp)
2147 {
2148 	struct lagg_softc *sc = lp->lp_softc;
2149 	lagg_lb_porttable(sc, lp);
2150 }
2151 
2152 static int
2153 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2154 {
2155 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2156 	struct lagg_port *lp = NULL;
2157 	uint32_t p = 0;
2158 
2159 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && (m->m_flags & M_FLOWID))
2160 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2161 	else
2162 		p = lagg_hashmbuf(sc, m, lb->lb_key);
2163 	p %= sc->sc_count;
2164 	lp = lb->lb_ports[p];
2165 
2166 	/*
2167 	 * Check the port's link state. This will return the next active
2168 	 * port if the link is down or the port is NULL.
2169 	 */
2170 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2171 		m_freem(m);
2172 		return (ENETDOWN);
2173 	}
2174 
2175 	/* Send mbuf */
2176 	return (lagg_enqueue(lp->lp_ifp, m));
2177 }
2178 
2179 static struct mbuf *
2180 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2181 {
2182 	struct ifnet *ifp = sc->sc_ifp;
2183 
2184 	/* Just pass in the packet to our lagg device */
2185 	m->m_pkthdr.rcvif = ifp;
2186 
2187 	return (m);
2188 }
2189 
2190 /*
2191  * 802.3ad LACP
2192  */
2193 static void
2194 lagg_lacp_attach(struct lagg_softc *sc)
2195 {
2196 	struct lagg_port *lp;
2197 
2198 	lacp_attach(sc);
2199 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2200 		lacp_port_create(lp);
2201 }
2202 
2203 static void
2204 lagg_lacp_detach(struct lagg_softc *sc)
2205 {
2206 	struct lagg_port *lp;
2207 	void *psc;
2208 
2209 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2210 		lacp_port_destroy(lp);
2211 
2212 	psc = sc->sc_psc;
2213 	sc->sc_psc = NULL;
2214 	LAGG_WUNLOCK(sc);
2215 
2216 	lacp_detach(psc);
2217 }
2218 
2219 static void
2220 lagg_lacp_lladdr(struct lagg_softc *sc)
2221 {
2222 	struct lagg_port *lp;
2223 
2224 	/* purge all the lacp ports */
2225 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2226 		lacp_port_destroy(lp);
2227 
2228 	/* add them back in */
2229 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2230 		lacp_port_create(lp);
2231 }
2232 
2233 static int
2234 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2235 {
2236 	struct lagg_port *lp;
2237 
2238 	lp = lacp_select_tx_port(sc, m);
2239 	if (lp == NULL) {
2240 		m_freem(m);
2241 		return (ENETDOWN);
2242 	}
2243 
2244 	/* Send mbuf */
2245 	return (lagg_enqueue(lp->lp_ifp, m));
2246 }
2247 
2248 static struct mbuf *
2249 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2250 {
2251 	struct ifnet *ifp = sc->sc_ifp;
2252 	struct ether_header *eh;
2253 	u_short etype;
2254 
2255 	eh = mtod(m, struct ether_header *);
2256 	etype = ntohs(eh->ether_type);
2257 
2258 	/* Tap off LACP control messages */
2259 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2260 		m = lacp_input(lp, m);
2261 		if (m == NULL)
2262 			return (NULL);
2263 	}
2264 
2265 	/*
2266 	 * If the port is not collecting or not in the active aggregator then
2267 	 * free and return.
2268 	 */
2269 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2270 		m_freem(m);
2271 		return (NULL);
2272 	}
2273 
2274 	m->m_pkthdr.rcvif = ifp;
2275 	return (m);
2276 }
2277 
2278