xref: /freebsd/sys/net/if_lagg.c (revision 656f49f8e2b0656824a5f10aeb760a00fdd3753f)
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/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_media.h>
50 #include <net/if_types.h>
51 #include <net/if_var.h>
52 #include <net/bpf.h>
53 #include <net/vnet.h>
54 
55 #if defined(INET) || defined(INET6)
56 #include <netinet/in.h>
57 #include <netinet/ip.h>
58 #endif
59 #ifdef INET
60 #include <netinet/in_systm.h>
61 #include <netinet/if_ether.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 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
85 #define	V_lagg_list	VNET(lagg_list)
86 static VNET_DEFINE(struct mtx, lagg_list_mtx);
87 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
88 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
89 					"if_lagg list", NULL, MTX_DEF)
90 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
91 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
92 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
93 eventhandler_tag	lagg_detach_cookie = NULL;
94 
95 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
96 static void	lagg_clone_destroy(struct ifnet *);
97 static VNET_DEFINE(struct if_clone *, lagg_cloner);
98 #define	V_lagg_cloner	VNET(lagg_cloner)
99 static const char laggname[] = "lagg";
100 
101 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
102 static void	lagg_capabilities(struct lagg_softc *);
103 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
104 static void	lagg_port_setlladdr(void *, int);
105 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
106 static int	lagg_port_destroy(struct lagg_port *, int);
107 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
108 static void	lagg_linkstate(struct lagg_softc *);
109 static void	lagg_port_state(struct ifnet *, int);
110 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
111 static int	lagg_port_output(struct ifnet *, struct mbuf *,
112 		    const struct sockaddr *, struct route *);
113 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
114 #ifdef LAGG_PORT_STACKING
115 static int	lagg_port_checkstacking(struct lagg_softc *);
116 #endif
117 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
118 static void	lagg_init(void *);
119 static void	lagg_stop(struct lagg_softc *);
120 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
121 static int	lagg_ether_setmulti(struct lagg_softc *);
122 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
123 static	int	lagg_setflag(struct lagg_port *, int, int,
124 		    int (*func)(struct ifnet *, int));
125 static	int	lagg_setflags(struct lagg_port *, int status);
126 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
127 static int	lagg_transmit(struct ifnet *, struct mbuf *);
128 static void	lagg_qflush(struct ifnet *);
129 static int	lagg_media_change(struct ifnet *);
130 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
131 static struct lagg_port *lagg_link_active(struct lagg_softc *,
132 	    struct lagg_port *);
133 
134 /* Simple round robin */
135 static void	lagg_rr_attach(struct lagg_softc *);
136 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
137 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
138 		    struct mbuf *);
139 
140 /* Active failover */
141 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
142 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
143 		    struct mbuf *);
144 
145 /* Loadbalancing */
146 static void	lagg_lb_attach(struct lagg_softc *);
147 static void	lagg_lb_detach(struct lagg_softc *);
148 static int	lagg_lb_port_create(struct lagg_port *);
149 static void	lagg_lb_port_destroy(struct lagg_port *);
150 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
151 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
152 		    struct mbuf *);
153 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
154 
155 /* Broadcast */
156 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
157 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
158 		    struct mbuf *);
159 
160 /* 802.3ad LACP */
161 static void	lagg_lacp_attach(struct lagg_softc *);
162 static void	lagg_lacp_detach(struct lagg_softc *);
163 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
164 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
165 		    struct mbuf *);
166 static void	lagg_lacp_lladdr(struct lagg_softc *);
167 
168 /* lagg protocol table */
169 static const struct lagg_proto {
170 	lagg_proto	pr_num;
171 	void		(*pr_attach)(struct lagg_softc *);
172 	void		(*pr_detach)(struct lagg_softc *);
173 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
174 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
175 			    struct mbuf *);
176 	int		(*pr_addport)(struct lagg_port *);
177 	void		(*pr_delport)(struct lagg_port *);
178 	void		(*pr_linkstate)(struct lagg_port *);
179 	void 		(*pr_init)(struct lagg_softc *);
180 	void 		(*pr_stop)(struct lagg_softc *);
181 	void 		(*pr_lladdr)(struct lagg_softc *);
182 	void		(*pr_request)(struct lagg_softc *, void *);
183 	void		(*pr_portreq)(struct lagg_port *, void *);
184 } lagg_protos[] = {
185     {
186 	.pr_num = LAGG_PROTO_NONE
187     },
188     {
189 	.pr_num = LAGG_PROTO_ROUNDROBIN,
190 	.pr_attach = lagg_rr_attach,
191 	.pr_start = lagg_rr_start,
192 	.pr_input = lagg_rr_input,
193     },
194     {
195 	.pr_num = LAGG_PROTO_FAILOVER,
196 	.pr_start = lagg_fail_start,
197 	.pr_input = lagg_fail_input,
198     },
199     {
200 	.pr_num = LAGG_PROTO_LOADBALANCE,
201 	.pr_attach = lagg_lb_attach,
202 	.pr_detach = lagg_lb_detach,
203 	.pr_start = lagg_lb_start,
204 	.pr_input = lagg_lb_input,
205 	.pr_addport = lagg_lb_port_create,
206 	.pr_delport = lagg_lb_port_destroy,
207     },
208     {
209 	.pr_num = LAGG_PROTO_LACP,
210 	.pr_attach = lagg_lacp_attach,
211 	.pr_detach = lagg_lacp_detach,
212 	.pr_start = lagg_lacp_start,
213 	.pr_input = lagg_lacp_input,
214 	.pr_addport = lacp_port_create,
215 	.pr_delport = lacp_port_destroy,
216 	.pr_linkstate = lacp_linkstate,
217 	.pr_init = lacp_init,
218 	.pr_stop = lacp_stop,
219 	.pr_lladdr = lagg_lacp_lladdr,
220 	.pr_request = lacp_req,
221 	.pr_portreq = lacp_portreq,
222     },
223     {
224 	.pr_num = LAGG_PROTO_BROADCAST,
225 	.pr_start = lagg_bcast_start,
226 	.pr_input = lagg_bcast_input,
227     },
228 };
229 
230 SYSCTL_DECL(_net_link);
231 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
232     "Link Aggregation");
233 
234 /* Allow input on any failover links */
235 static VNET_DEFINE(int, lagg_failover_rx_all);
236 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
237 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
238     &VNET_NAME(lagg_failover_rx_all), 0,
239     "Accept input from any interface in a failover lagg");
240 
241 /* Default value for using flowid */
242 static VNET_DEFINE(int, def_use_flowid) = 1;
243 #define	V_def_use_flowid	VNET(def_use_flowid)
244 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
245     &VNET_NAME(def_use_flowid), 0,
246     "Default setting for using flow id for load sharing");
247 
248 /* Default value for flowid shift */
249 static VNET_DEFINE(int, def_flowid_shift) = 16;
250 #define	V_def_flowid_shift	VNET(def_flowid_shift)
251 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
252     &VNET_NAME(def_flowid_shift), 0,
253     "Default setting for flowid shift for load sharing");
254 
255 static void
256 vnet_lagg_init(const void *unused __unused)
257 {
258 
259 	LAGG_LIST_LOCK_INIT();
260 	SLIST_INIT(&V_lagg_list);
261 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
262 	    lagg_clone_destroy, 0);
263 }
264 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
265     vnet_lagg_init, NULL);
266 
267 static void
268 vnet_lagg_uninit(const void *unused __unused)
269 {
270 
271 	if_clone_detach(V_lagg_cloner);
272 	LAGG_LIST_LOCK_DESTROY();
273 }
274 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
275     vnet_lagg_uninit, NULL);
276 
277 static int
278 lagg_modevent(module_t mod, int type, void *data)
279 {
280 
281 	switch (type) {
282 	case MOD_LOAD:
283 		lagg_input_p = lagg_input;
284 		lagg_linkstate_p = lagg_port_state;
285 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
286 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
287 		    EVENTHANDLER_PRI_ANY);
288 		break;
289 	case MOD_UNLOAD:
290 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
291 		    lagg_detach_cookie);
292 		lagg_input_p = NULL;
293 		lagg_linkstate_p = NULL;
294 		break;
295 	default:
296 		return (EOPNOTSUPP);
297 	}
298 	return (0);
299 }
300 
301 static moduledata_t lagg_mod = {
302 	"if_lagg",
303 	lagg_modevent,
304 	0
305 };
306 
307 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
308 MODULE_VERSION(if_lagg, 1);
309 
310 static void
311 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
312 {
313 
314 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
315 	    __func__, sc));
316 
317 	if (sc->sc_ifflags & IFF_DEBUG)
318 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
319 
320 	if (lagg_protos[pr].pr_attach != NULL)
321 		lagg_protos[pr].pr_attach(sc);
322 	sc->sc_proto = pr;
323 }
324 
325 static void
326 lagg_proto_detach(struct lagg_softc *sc)
327 {
328 	lagg_proto pr;
329 
330 	LAGG_WLOCK_ASSERT(sc);
331 
332 	pr = sc->sc_proto;
333 	sc->sc_proto = LAGG_PROTO_NONE;
334 
335 	if (lagg_protos[pr].pr_detach != NULL)
336 		lagg_protos[pr].pr_detach(sc);
337 	else
338 		LAGG_WUNLOCK(sc);
339 }
340 
341 static int
342 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
343 {
344 
345 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
346 }
347 
348 static struct mbuf *
349 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
350 {
351 
352 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
353 }
354 
355 static int
356 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
357 {
358 
359 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
360 		return (0);
361 	else
362 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
363 }
364 
365 static void
366 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
367 {
368 
369 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
370 		lagg_protos[sc->sc_proto].pr_delport(lp);
371 }
372 
373 static void
374 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
375 {
376 
377 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
378 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
379 }
380 
381 static void
382 lagg_proto_init(struct lagg_softc *sc)
383 {
384 
385 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
386 		lagg_protos[sc->sc_proto].pr_init(sc);
387 }
388 
389 static void
390 lagg_proto_stop(struct lagg_softc *sc)
391 {
392 
393 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
394 		lagg_protos[sc->sc_proto].pr_stop(sc);
395 }
396 
397 static void
398 lagg_proto_lladdr(struct lagg_softc *sc)
399 {
400 
401 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
402 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
403 }
404 
405 static void
406 lagg_proto_request(struct lagg_softc *sc, void *v)
407 {
408 
409 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
410 		lagg_protos[sc->sc_proto].pr_request(sc, v);
411 }
412 
413 static void
414 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
415 {
416 
417 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
418 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
419 }
420 
421 /*
422  * This routine is run via an vlan
423  * config EVENT
424  */
425 static void
426 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
427 {
428 	struct lagg_softc *sc = ifp->if_softc;
429 	struct lagg_port *lp;
430 	struct rm_priotracker tracker;
431 
432 	if (ifp->if_softc !=  arg)   /* Not our event */
433 		return;
434 
435 	LAGG_RLOCK(sc, &tracker);
436 	if (!SLIST_EMPTY(&sc->sc_ports)) {
437 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
438 			EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
439 	}
440 	LAGG_RUNLOCK(sc, &tracker);
441 }
442 
443 /*
444  * This routine is run via an vlan
445  * unconfig EVENT
446  */
447 static void
448 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
449 {
450 	struct lagg_softc *sc = ifp->if_softc;
451 	struct lagg_port *lp;
452 	struct rm_priotracker tracker;
453 
454 	if (ifp->if_softc !=  arg)   /* Not our event */
455 		return;
456 
457 	LAGG_RLOCK(sc, &tracker);
458 	if (!SLIST_EMPTY(&sc->sc_ports)) {
459 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
460 			EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
461 	}
462 	LAGG_RUNLOCK(sc, &tracker);
463 }
464 
465 static int
466 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
467 {
468 	struct lagg_softc *sc;
469 	struct ifnet *ifp;
470 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
471 
472 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
473 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
474 	if (ifp == NULL) {
475 		free(sc, M_DEVBUF);
476 		return (ENOSPC);
477 	}
478 
479 	if (V_def_use_flowid)
480 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
481 	sc->flowid_shift = V_def_flowid_shift;
482 
483 	/* Hash all layers by default */
484 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
485 
486 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
487 
488 	LAGG_LOCK_INIT(sc);
489 	SLIST_INIT(&sc->sc_ports);
490 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
491 
492 	/* Initialise pseudo media types */
493 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
494 	    lagg_media_status);
495 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
496 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
497 
498 	if_initname(ifp, laggname, unit);
499 	ifp->if_softc = sc;
500 	ifp->if_transmit = lagg_transmit;
501 	ifp->if_qflush = lagg_qflush;
502 	ifp->if_init = lagg_init;
503 	ifp->if_ioctl = lagg_ioctl;
504 	ifp->if_get_counter = lagg_get_counter;
505 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
506 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
507 
508 	/*
509 	 * Attach as an ordinary ethernet device, children will be attached
510 	 * as special device IFT_IEEE8023ADLAG.
511 	 */
512 	ether_ifattach(ifp, eaddr);
513 
514 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
515 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
516 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
517 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
518 
519 	/* Insert into the global list of laggs */
520 	LAGG_LIST_LOCK();
521 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
522 	LAGG_LIST_UNLOCK();
523 
524 	return (0);
525 }
526 
527 static void
528 lagg_clone_destroy(struct ifnet *ifp)
529 {
530 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
531 	struct lagg_port *lp;
532 
533 	LAGG_WLOCK(sc);
534 
535 	lagg_stop(sc);
536 	ifp->if_flags &= ~IFF_UP;
537 
538 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
539 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
540 
541 	/* Shutdown and remove lagg ports */
542 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
543 		lagg_port_destroy(lp, 1);
544 	/* Unhook the aggregation protocol */
545 	lagg_proto_detach(sc);
546 
547 	ifmedia_removeall(&sc->sc_media);
548 	ether_ifdetach(ifp);
549 	if_free(ifp);
550 
551 	LAGG_LIST_LOCK();
552 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
553 	LAGG_LIST_UNLOCK();
554 
555 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
556 	LAGG_LOCK_DESTROY(sc);
557 	free(sc, M_DEVBUF);
558 }
559 
560 static void
561 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
562 {
563 	struct ifnet *ifp = sc->sc_ifp;
564 	struct lagg_port lp;
565 
566 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
567 		return;
568 
569 	LAGG_WLOCK_ASSERT(sc);
570 	/*
571 	 * Set the link layer address on the lagg interface.
572 	 * lagg_proto_lladdr() notifies the MAC change to
573 	 * the aggregation protocol.  iflladdr_event handler which
574 	 * may trigger gratuitous ARPs for INET will be handled in
575 	 * a taskqueue.
576 	 */
577 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
578 	lagg_proto_lladdr(sc);
579 
580 	bzero(&lp, sizeof(lp));
581 	lp.lp_ifp = sc->sc_ifp;
582 	lp.lp_softc = sc;
583 
584 	lagg_port_lladdr(&lp, lladdr);
585 }
586 
587 static void
588 lagg_capabilities(struct lagg_softc *sc)
589 {
590 	struct lagg_port *lp;
591 	int cap = ~0, ena = ~0;
592 	u_long hwa = ~0UL;
593 	struct ifnet_hw_tsomax hw_tsomax;
594 
595 	LAGG_WLOCK_ASSERT(sc);
596 
597 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
598 
599 	/* Get capabilities from the lagg ports */
600 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
601 		cap &= lp->lp_ifp->if_capabilities;
602 		ena &= lp->lp_ifp->if_capenable;
603 		hwa &= lp->lp_ifp->if_hwassist;
604 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
605 	}
606 	cap = (cap == ~0 ? 0 : cap);
607 	ena = (ena == ~0 ? 0 : ena);
608 	hwa = (hwa == ~0 ? 0 : hwa);
609 
610 	if (sc->sc_ifp->if_capabilities != cap ||
611 	    sc->sc_ifp->if_capenable != ena ||
612 	    sc->sc_ifp->if_hwassist != hwa ||
613 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
614 		sc->sc_ifp->if_capabilities = cap;
615 		sc->sc_ifp->if_capenable = ena;
616 		sc->sc_ifp->if_hwassist = hwa;
617 		getmicrotime(&sc->sc_ifp->if_lastchange);
618 
619 		if (sc->sc_ifflags & IFF_DEBUG)
620 			if_printf(sc->sc_ifp,
621 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
622 	}
623 }
624 
625 static void
626 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
627 {
628 	struct lagg_softc *sc = lp->lp_softc;
629 	struct ifnet *ifp = lp->lp_ifp;
630 	struct lagg_llq *llq;
631 	int pending = 0;
632 	int primary;
633 
634 	LAGG_WLOCK_ASSERT(sc);
635 
636 	primary = (sc->sc_primary->lp_ifp == ifp) ? 1 : 0;
637 	if (primary == 0 && (lp->lp_detaching ||
638 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0))
639 		return;
640 
641 	/* Check to make sure its not already queued to be changed */
642 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
643 		if (llq->llq_ifp == ifp && llq->llq_primary == primary) {
644 			pending = 1;
645 			break;
646 		}
647 	}
648 
649 	if (!pending) {
650 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
651 		if (llq == NULL)	/* XXX what to do */
652 			return;
653 	}
654 
655 	/* Update the lladdr even if pending, it may have changed */
656 	llq->llq_ifp = ifp;
657 	llq->llq_primary = primary;
658 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
659 
660 	if (!pending)
661 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
662 
663 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
664 }
665 
666 /*
667  * Set the interface MAC address from a taskqueue to avoid a LOR.
668  */
669 static void
670 lagg_port_setlladdr(void *arg, int pending)
671 {
672 	struct lagg_softc *sc = (struct lagg_softc *)arg;
673 	struct lagg_llq *llq, *head;
674 	struct ifnet *ifp;
675 	int error;
676 
677 	/* Grab a local reference of the queue and remove it from the softc */
678 	LAGG_WLOCK(sc);
679 	head = SLIST_FIRST(&sc->sc_llq_head);
680 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
681 	LAGG_WUNLOCK(sc);
682 
683 	/*
684 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
685 	 * unlocked as we have the only reference to it.
686 	 */
687 	for (llq = head; llq != NULL; llq = head) {
688 		ifp = llq->llq_ifp;
689 
690 		CURVNET_SET(ifp->if_vnet);
691 		if (llq->llq_primary == 0) {
692 			/*
693 			 * Set the link layer address on the laggport interface.
694 			 * if_setlladdr() triggers gratuitous ARPs for INET.
695 			 */
696 			error = if_setlladdr(ifp, llq->llq_lladdr,
697 			    ETHER_ADDR_LEN);
698 			if (error)
699 				printf("%s: setlladdr failed on %s\n", __func__,
700 				    ifp->if_xname);
701 		} else
702 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
703 		CURVNET_RESTORE();
704 		head = SLIST_NEXT(llq, llq_entries);
705 		free(llq, M_DEVBUF);
706 	}
707 }
708 
709 static int
710 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
711 {
712 	struct lagg_softc *sc_ptr;
713 	struct lagg_port *lp, *tlp;
714 	int error, i;
715 	uint64_t *pval;
716 
717 	LAGG_WLOCK_ASSERT(sc);
718 
719 	/* Limit the maximal number of lagg ports */
720 	if (sc->sc_count >= LAGG_MAX_PORTS)
721 		return (ENOSPC);
722 
723 	/* Check if port has already been associated to a lagg */
724 	if (ifp->if_lagg != NULL) {
725 		/* Port is already in the current lagg? */
726 		lp = (struct lagg_port *)ifp->if_lagg;
727 		if (lp->lp_softc == sc)
728 			return (EEXIST);
729 		return (EBUSY);
730 	}
731 
732 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
733 	if (ifp->if_type != IFT_ETHER)
734 		return (EPROTONOSUPPORT);
735 
736 	/* Allow the first Ethernet member to define the MTU */
737 	if (SLIST_EMPTY(&sc->sc_ports))
738 		sc->sc_ifp->if_mtu = ifp->if_mtu;
739 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
740 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
741 		    ifp->if_xname);
742 		return (EINVAL);
743 	}
744 
745 	if ((lp = malloc(sizeof(struct lagg_port),
746 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
747 		return (ENOMEM);
748 
749 	/* Check if port is a stacked lagg */
750 	LAGG_LIST_LOCK();
751 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
752 		if (ifp == sc_ptr->sc_ifp) {
753 			LAGG_LIST_UNLOCK();
754 			free(lp, M_DEVBUF);
755 			return (EINVAL);
756 			/* XXX disable stacking for the moment, its untested */
757 #ifdef LAGG_PORT_STACKING
758 			lp->lp_flags |= LAGG_PORT_STACK;
759 			if (lagg_port_checkstacking(sc_ptr) >=
760 			    LAGG_MAX_STACKING) {
761 				LAGG_LIST_UNLOCK();
762 				free(lp, M_DEVBUF);
763 				return (E2BIG);
764 			}
765 #endif
766 		}
767 	}
768 	LAGG_LIST_UNLOCK();
769 
770 	/* Change the interface type */
771 	lp->lp_iftype = ifp->if_type;
772 	ifp->if_type = IFT_IEEE8023ADLAG;
773 	ifp->if_lagg = lp;
774 	lp->lp_ioctl = ifp->if_ioctl;
775 	ifp->if_ioctl = lagg_port_ioctl;
776 	lp->lp_output = ifp->if_output;
777 	ifp->if_output = lagg_port_output;
778 
779 	lp->lp_ifp = ifp;
780 	lp->lp_softc = sc;
781 
782 	/* Save port link layer address */
783 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
784 
785 	if (SLIST_EMPTY(&sc->sc_ports)) {
786 		sc->sc_primary = lp;
787 		lagg_lladdr(sc, IF_LLADDR(ifp));
788 	} else {
789 		/* Update link layer address for this port */
790 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
791 	}
792 
793 	/*
794 	 * Insert into the list of ports.
795 	 * Keep ports sorted by if_index. It is handy, when configuration
796 	 * is predictable and `ifconfig laggN create ...` command
797 	 * will lead to the same result each time.
798 	 */
799 	SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
800 		if (tlp->lp_ifp->if_index < ifp->if_index && (
801 		    SLIST_NEXT(tlp, lp_entries) == NULL ||
802 		    SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index >
803 		    ifp->if_index))
804 			break;
805 	}
806 	if (tlp != NULL)
807 		SLIST_INSERT_AFTER(tlp, lp, lp_entries);
808 	else
809 		SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
810 	sc->sc_count++;
811 
812 	/* Update lagg capabilities */
813 	lagg_capabilities(sc);
814 	lagg_linkstate(sc);
815 
816 	/* Read port counters */
817 	pval = lp->port_counters.val;
818 	for (i = 0; i < IFCOUNTERS; i++, pval++)
819 		*pval = ifp->if_get_counter(ifp, i);
820 	/* Add multicast addresses and interface flags to this port */
821 	lagg_ether_cmdmulti(lp, 1);
822 	lagg_setflags(lp, 1);
823 
824 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
825 		/* Remove the port, without calling pr_delport. */
826 		lagg_port_destroy(lp, 0);
827 		return (error);
828 	}
829 
830 	return (0);
831 }
832 
833 #ifdef LAGG_PORT_STACKING
834 static int
835 lagg_port_checkstacking(struct lagg_softc *sc)
836 {
837 	struct lagg_softc *sc_ptr;
838 	struct lagg_port *lp;
839 	int m = 0;
840 
841 	LAGG_WLOCK_ASSERT(sc);
842 
843 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
844 		if (lp->lp_flags & LAGG_PORT_STACK) {
845 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
846 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
847 		}
848 	}
849 
850 	return (m + 1);
851 }
852 #endif
853 
854 static int
855 lagg_port_destroy(struct lagg_port *lp, int rundelport)
856 {
857 	struct lagg_softc *sc = lp->lp_softc;
858 	struct lagg_port *lp_ptr, *lp0;
859 	struct lagg_llq *llq;
860 	struct ifnet *ifp = lp->lp_ifp;
861 	uint64_t *pval, vdiff;
862 	int i;
863 
864 	LAGG_WLOCK_ASSERT(sc);
865 
866 	if (rundelport)
867 		lagg_proto_delport(sc, lp);
868 
869 	/*
870 	 * Remove multicast addresses and interface flags from this port and
871 	 * reset the MAC address, skip if the interface is being detached.
872 	 */
873 	if (!lp->lp_detaching) {
874 		lagg_ether_cmdmulti(lp, 0);
875 		lagg_setflags(lp, 0);
876 		lagg_port_lladdr(lp, lp->lp_lladdr);
877 	}
878 
879 	/* Restore interface */
880 	ifp->if_type = lp->lp_iftype;
881 	ifp->if_ioctl = lp->lp_ioctl;
882 	ifp->if_output = lp->lp_output;
883 	ifp->if_lagg = NULL;
884 
885 	/* Update detached port counters */
886 	pval = lp->port_counters.val;
887 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
888 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
889 		sc->detached_counters.val[i] += vdiff;
890 	}
891 
892 	/* Finally, remove the port from the lagg */
893 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
894 	sc->sc_count--;
895 
896 	/* Update the primary interface */
897 	if (lp == sc->sc_primary) {
898 		uint8_t lladdr[ETHER_ADDR_LEN];
899 
900 		if ((lp0 = SLIST_FIRST(&sc->sc_ports)) == NULL) {
901 			bzero(&lladdr, ETHER_ADDR_LEN);
902 		} else {
903 			bcopy(lp0->lp_lladdr,
904 			    lladdr, ETHER_ADDR_LEN);
905 		}
906 		lagg_lladdr(sc, lladdr);
907 
908 		/*
909 		 * Update link layer address for each port.  No port is
910 		 * marked as primary at this moment.
911 		 */
912 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
913 			lagg_port_lladdr(lp_ptr, lladdr);
914 		/*
915 		 * Mark lp0 as the new primary.  This invokes an
916 		 * iflladdr_event.
917 		 */
918 		sc->sc_primary = lp0;
919 		lagg_port_lladdr(lp0, lladdr);
920 	}
921 
922 	/* Remove any pending lladdr changes from the queue */
923 	if (lp->lp_detaching) {
924 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
925 			if (llq->llq_ifp == ifp) {
926 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
927 				    llq_entries);
928 				free(llq, M_DEVBUF);
929 				break;	/* Only appears once */
930 			}
931 		}
932 	}
933 
934 	if (lp->lp_ifflags)
935 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
936 
937 	free(lp, M_DEVBUF);
938 
939 	/* Update lagg capabilities */
940 	lagg_capabilities(sc);
941 	lagg_linkstate(sc);
942 
943 	return (0);
944 }
945 
946 static int
947 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
948 {
949 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
950 	struct lagg_softc *sc;
951 	struct lagg_port *lp = NULL;
952 	int error = 0;
953 	struct rm_priotracker tracker;
954 
955 	/* Should be checked by the caller */
956 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
957 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
958 		goto fallback;
959 
960 	switch (cmd) {
961 	case SIOCGLAGGPORT:
962 		if (rp->rp_portname[0] == '\0' ||
963 		    ifunit(rp->rp_portname) != ifp) {
964 			error = EINVAL;
965 			break;
966 		}
967 
968 		LAGG_RLOCK(sc, &tracker);
969 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
970 			error = ENOENT;
971 			LAGG_RUNLOCK(sc, &tracker);
972 			break;
973 		}
974 
975 		lagg_port2req(lp, rp);
976 		LAGG_RUNLOCK(sc, &tracker);
977 		break;
978 
979 	case SIOCSIFCAP:
980 		if (lp->lp_ioctl == NULL) {
981 			error = EINVAL;
982 			break;
983 		}
984 		error = (*lp->lp_ioctl)(ifp, cmd, data);
985 		if (error)
986 			break;
987 
988 		/* Update lagg interface capabilities */
989 		LAGG_WLOCK(sc);
990 		lagg_capabilities(sc);
991 		LAGG_WUNLOCK(sc);
992 		break;
993 
994 	case SIOCSIFMTU:
995 		/* Do not allow the MTU to be changed once joined */
996 		error = EINVAL;
997 		break;
998 
999 	default:
1000 		goto fallback;
1001 	}
1002 
1003 	return (error);
1004 
1005 fallback:
1006 	if (lp->lp_ioctl != NULL)
1007 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1008 
1009 	return (EINVAL);
1010 }
1011 
1012 /*
1013  * Requests counter @cnt data.
1014  *
1015  * Counter value is calculated the following way:
1016  * 1) for each port, sum  difference between current and "initial" measurements.
1017  * 2) add lagg logical interface counters.
1018  * 3) add data from detached_counters array.
1019  *
1020  * We also do the following things on ports attach/detach:
1021  * 1) On port attach we store all counters it has into port_counter array.
1022  * 2) On port detach we add the different between "initial" and
1023  *   current counters data to detached_counters array.
1024  */
1025 static uint64_t
1026 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1027 {
1028 	struct lagg_softc *sc;
1029 	struct lagg_port *lp;
1030 	struct ifnet *lpifp;
1031 	struct rm_priotracker tracker;
1032 	uint64_t newval, oldval, vsum;
1033 
1034 	/* Revise this when we've got non-generic counters. */
1035 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1036 
1037 	sc = (struct lagg_softc *)ifp->if_softc;
1038 	LAGG_RLOCK(sc, &tracker);
1039 
1040 	vsum = 0;
1041 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1042 		/* Saved attached value */
1043 		oldval = lp->port_counters.val[cnt];
1044 		/* current value */
1045 		lpifp = lp->lp_ifp;
1046 		newval = lpifp->if_get_counter(lpifp, cnt);
1047 		/* Calculate diff and save new */
1048 		vsum += newval - oldval;
1049 	}
1050 
1051 	/*
1052 	 * Add counter data which might be added by upper
1053 	 * layer protocols operating on logical interface.
1054 	 */
1055 	vsum += if_get_counter_default(ifp, cnt);
1056 
1057 	/*
1058 	 * Add counter data from detached ports counters
1059 	 */
1060 	vsum += sc->detached_counters.val[cnt];
1061 
1062 	LAGG_RUNLOCK(sc, &tracker);
1063 
1064 	return (vsum);
1065 }
1066 
1067 /*
1068  * For direct output to child ports.
1069  */
1070 static int
1071 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1072 	const struct sockaddr *dst, struct route *ro)
1073 {
1074 	struct lagg_port *lp = ifp->if_lagg;
1075 
1076 	switch (dst->sa_family) {
1077 		case pseudo_AF_HDRCMPLT:
1078 		case AF_UNSPEC:
1079 			return ((*lp->lp_output)(ifp, m, dst, ro));
1080 	}
1081 
1082 	/* drop any other frames */
1083 	m_freem(m);
1084 	return (ENETDOWN);
1085 }
1086 
1087 static void
1088 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1089 {
1090 	struct lagg_port *lp;
1091 	struct lagg_softc *sc;
1092 
1093 	if ((lp = ifp->if_lagg) == NULL)
1094 		return;
1095 	/* If the ifnet is just being renamed, don't do anything. */
1096 	if (ifp->if_flags & IFF_RENAMING)
1097 		return;
1098 
1099 	sc = lp->lp_softc;
1100 
1101 	LAGG_WLOCK(sc);
1102 	lp->lp_detaching = 1;
1103 	lagg_port_destroy(lp, 1);
1104 	LAGG_WUNLOCK(sc);
1105 }
1106 
1107 static void
1108 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1109 {
1110 	struct lagg_softc *sc = lp->lp_softc;
1111 
1112 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1113 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1114 	rp->rp_prio = lp->lp_prio;
1115 	rp->rp_flags = lp->lp_flags;
1116 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1117 
1118 	/* Add protocol specific flags */
1119 	switch (sc->sc_proto) {
1120 		case LAGG_PROTO_FAILOVER:
1121 			if (lp == sc->sc_primary)
1122 				rp->rp_flags |= LAGG_PORT_MASTER;
1123 			if (lp == lagg_link_active(sc, sc->sc_primary))
1124 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1125 			break;
1126 
1127 		case LAGG_PROTO_ROUNDROBIN:
1128 		case LAGG_PROTO_LOADBALANCE:
1129 		case LAGG_PROTO_BROADCAST:
1130 			if (LAGG_PORTACTIVE(lp))
1131 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1132 			break;
1133 
1134 		case LAGG_PROTO_LACP:
1135 			/* LACP has a different definition of active */
1136 			if (lacp_isactive(lp))
1137 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1138 			if (lacp_iscollecting(lp))
1139 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1140 			if (lacp_isdistributing(lp))
1141 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1142 			break;
1143 	}
1144 
1145 }
1146 
1147 static void
1148 lagg_init(void *xsc)
1149 {
1150 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1151 	struct lagg_port *lp;
1152 	struct ifnet *ifp = sc->sc_ifp;
1153 
1154 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1155 		return;
1156 
1157 	LAGG_WLOCK(sc);
1158 
1159 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1160 	/* Update the port lladdrs */
1161 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1162 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
1163 
1164 	lagg_proto_init(sc);
1165 
1166 	LAGG_WUNLOCK(sc);
1167 }
1168 
1169 static void
1170 lagg_stop(struct lagg_softc *sc)
1171 {
1172 	struct ifnet *ifp = sc->sc_ifp;
1173 
1174 	LAGG_WLOCK_ASSERT(sc);
1175 
1176 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1177 		return;
1178 
1179 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1180 
1181 	lagg_proto_stop(sc);
1182 }
1183 
1184 static int
1185 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1186 {
1187 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1188 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1189 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1190 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1191 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1192 	struct ifreq *ifr = (struct ifreq *)data;
1193 	struct lagg_port *lp;
1194 	struct ifnet *tpif;
1195 	struct thread *td = curthread;
1196 	char *buf, *outbuf;
1197 	int count, buflen, len, error = 0;
1198 	struct rm_priotracker tracker;
1199 
1200 	bzero(&rpbuf, sizeof(rpbuf));
1201 
1202 	switch (cmd) {
1203 	case SIOCGLAGG:
1204 		LAGG_RLOCK(sc, &tracker);
1205 		count = 0;
1206 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1207 			count++;
1208 		buflen = count * sizeof(struct lagg_reqport);
1209 		LAGG_RUNLOCK(sc, &tracker);
1210 
1211 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1212 
1213 		LAGG_RLOCK(sc, &tracker);
1214 		ra->ra_proto = sc->sc_proto;
1215 		lagg_proto_request(sc, &ra->ra_psc);
1216 		count = 0;
1217 		buf = outbuf;
1218 		len = min(ra->ra_size, buflen);
1219 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1220 			if (len < sizeof(rpbuf))
1221 				break;
1222 
1223 			lagg_port2req(lp, &rpbuf);
1224 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1225 			count++;
1226 			buf += sizeof(rpbuf);
1227 			len -= sizeof(rpbuf);
1228 		}
1229 		LAGG_RUNLOCK(sc, &tracker);
1230 		ra->ra_ports = count;
1231 		ra->ra_size = count * sizeof(rpbuf);
1232 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1233 		free(outbuf, M_TEMP);
1234 		break;
1235 	case SIOCSLAGG:
1236 		error = priv_check(td, PRIV_NET_LAGG);
1237 		if (error)
1238 			break;
1239 		if (ra->ra_proto < 1 || ra->ra_proto >= LAGG_PROTO_MAX) {
1240 			error = EPROTONOSUPPORT;
1241 			break;
1242 		}
1243 
1244 		LAGG_WLOCK(sc);
1245 		lagg_proto_detach(sc);
1246 		lagg_proto_attach(sc, ra->ra_proto);
1247 		break;
1248 	case SIOCGLAGGOPTS:
1249 		ro->ro_opts = sc->sc_opts;
1250 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1251 			struct lacp_softc *lsc;
1252 
1253 			lsc = (struct lacp_softc *)sc->sc_psc;
1254 			if (lsc->lsc_debug.lsc_tx_test != 0)
1255 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1256 			if (lsc->lsc_debug.lsc_rx_test != 0)
1257 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1258 			if (lsc->lsc_strict_mode != 0)
1259 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1260 			if (lsc->lsc_fast_timeout != 0)
1261 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1262 
1263 			ro->ro_active = sc->sc_active;
1264 		} else {
1265 			ro->ro_active = 0;
1266 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1267 				ro->ro_active += LAGG_PORTACTIVE(lp);
1268 		}
1269 		ro->ro_flapping = sc->sc_flapping;
1270 		ro->ro_flowid_shift = sc->flowid_shift;
1271 		break;
1272 	case SIOCSLAGGOPTS:
1273 		error = priv_check(td, PRIV_NET_LAGG);
1274 		if (error)
1275 			break;
1276 		if (ro->ro_opts == 0)
1277 			break;
1278 		/*
1279 		 * Set options.  LACP options are stored in sc->sc_psc,
1280 		 * not in sc_opts.
1281 		 */
1282 		int valid, lacp;
1283 
1284 		switch (ro->ro_opts) {
1285 		case LAGG_OPT_USE_FLOWID:
1286 		case -LAGG_OPT_USE_FLOWID:
1287 		case LAGG_OPT_FLOWIDSHIFT:
1288 			valid = 1;
1289 			lacp = 0;
1290 			break;
1291 		case LAGG_OPT_LACP_TXTEST:
1292 		case -LAGG_OPT_LACP_TXTEST:
1293 		case LAGG_OPT_LACP_RXTEST:
1294 		case -LAGG_OPT_LACP_RXTEST:
1295 		case LAGG_OPT_LACP_STRICT:
1296 		case -LAGG_OPT_LACP_STRICT:
1297 		case LAGG_OPT_LACP_TIMEOUT:
1298 		case -LAGG_OPT_LACP_TIMEOUT:
1299 			valid = lacp = 1;
1300 			break;
1301 		default:
1302 			valid = lacp = 0;
1303 			break;
1304 		}
1305 
1306 		LAGG_WLOCK(sc);
1307 		if (valid == 0 ||
1308 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1309 			/* Invalid combination of options specified. */
1310 			error = EINVAL;
1311 			LAGG_WUNLOCK(sc);
1312 			break;	/* Return from SIOCSLAGGOPTS. */
1313 		}
1314 		/*
1315 		 * Store new options into sc->sc_opts except for
1316 		 * FLOWIDSHIFT and LACP options.
1317 		 */
1318 		if (lacp == 0) {
1319 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1320 				sc->flowid_shift = ro->ro_flowid_shift;
1321 			else if (ro->ro_opts > 0)
1322 				sc->sc_opts |= ro->ro_opts;
1323 			else
1324 				sc->sc_opts &= ~ro->ro_opts;
1325 		} else {
1326 			struct lacp_softc *lsc;
1327 			struct lacp_port *lp;
1328 
1329 			lsc = (struct lacp_softc *)sc->sc_psc;
1330 
1331 			switch (ro->ro_opts) {
1332 			case LAGG_OPT_LACP_TXTEST:
1333 				lsc->lsc_debug.lsc_tx_test = 1;
1334 				break;
1335 			case -LAGG_OPT_LACP_TXTEST:
1336 				lsc->lsc_debug.lsc_tx_test = 0;
1337 				break;
1338 			case LAGG_OPT_LACP_RXTEST:
1339 				lsc->lsc_debug.lsc_rx_test = 1;
1340 				break;
1341 			case -LAGG_OPT_LACP_RXTEST:
1342 				lsc->lsc_debug.lsc_rx_test = 0;
1343 				break;
1344 			case LAGG_OPT_LACP_STRICT:
1345 				lsc->lsc_strict_mode = 1;
1346 				break;
1347 			case -LAGG_OPT_LACP_STRICT:
1348 				lsc->lsc_strict_mode = 0;
1349 				break;
1350 			case LAGG_OPT_LACP_TIMEOUT:
1351 				LACP_LOCK(lsc);
1352         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1353                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1354 				LACP_UNLOCK(lsc);
1355 				lsc->lsc_fast_timeout = 1;
1356 				break;
1357 			case -LAGG_OPT_LACP_TIMEOUT:
1358 				LACP_LOCK(lsc);
1359         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1360                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1361 				LACP_UNLOCK(lsc);
1362 				lsc->lsc_fast_timeout = 0;
1363 				break;
1364 			}
1365 		}
1366 		LAGG_WUNLOCK(sc);
1367 		break;
1368 	case SIOCGLAGGFLAGS:
1369 		rf->rf_flags = 0;
1370 		LAGG_RLOCK(sc, &tracker);
1371 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1372 			rf->rf_flags |= LAGG_F_HASHL2;
1373 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1374 			rf->rf_flags |= LAGG_F_HASHL3;
1375 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1376 			rf->rf_flags |= LAGG_F_HASHL4;
1377 		LAGG_RUNLOCK(sc, &tracker);
1378 		break;
1379 	case SIOCSLAGGHASH:
1380 		error = priv_check(td, PRIV_NET_LAGG);
1381 		if (error)
1382 			break;
1383 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1384 			error = EINVAL;
1385 			break;
1386 		}
1387 		LAGG_WLOCK(sc);
1388 		sc->sc_flags = 0;
1389 		if (rf->rf_flags & LAGG_F_HASHL2)
1390 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1391 		if (rf->rf_flags & LAGG_F_HASHL3)
1392 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1393 		if (rf->rf_flags & LAGG_F_HASHL4)
1394 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1395 		LAGG_WUNLOCK(sc);
1396 		break;
1397 	case SIOCGLAGGPORT:
1398 		if (rp->rp_portname[0] == '\0' ||
1399 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1400 			error = EINVAL;
1401 			break;
1402 		}
1403 
1404 		LAGG_RLOCK(sc, &tracker);
1405 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1406 		    lp->lp_softc != sc) {
1407 			error = ENOENT;
1408 			LAGG_RUNLOCK(sc, &tracker);
1409 			break;
1410 		}
1411 
1412 		lagg_port2req(lp, rp);
1413 		LAGG_RUNLOCK(sc, &tracker);
1414 		break;
1415 	case SIOCSLAGGPORT:
1416 		error = priv_check(td, PRIV_NET_LAGG);
1417 		if (error)
1418 			break;
1419 		if (rp->rp_portname[0] == '\0' ||
1420 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1421 			error = EINVAL;
1422 			break;
1423 		}
1424 #ifdef INET6
1425 		/*
1426 		 * A laggport interface should not have inet6 address
1427 		 * because two interfaces with a valid link-local
1428 		 * scope zone must not be merged in any form.  This
1429 		 * restriction is needed to prevent violation of
1430 		 * link-local scope zone.  Attempts to add a laggport
1431 		 * interface which has inet6 addresses triggers
1432 		 * removal of all inet6 addresses on the member
1433 		 * interface.
1434 		 */
1435 		if (in6ifa_llaonifp(tpif)) {
1436 			in6_ifdetach(tpif);
1437 				if_printf(sc->sc_ifp,
1438 				    "IPv6 addresses on %s have been removed "
1439 				    "before adding it as a member to prevent "
1440 				    "IPv6 address scope violation.\n",
1441 				    tpif->if_xname);
1442 		}
1443 #endif
1444 		LAGG_WLOCK(sc);
1445 		error = lagg_port_create(sc, tpif);
1446 		LAGG_WUNLOCK(sc);
1447 		break;
1448 	case SIOCSLAGGDELPORT:
1449 		error = priv_check(td, PRIV_NET_LAGG);
1450 		if (error)
1451 			break;
1452 		if (rp->rp_portname[0] == '\0' ||
1453 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1454 			error = EINVAL;
1455 			break;
1456 		}
1457 
1458 		LAGG_WLOCK(sc);
1459 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1460 		    lp->lp_softc != sc) {
1461 			error = ENOENT;
1462 			LAGG_WUNLOCK(sc);
1463 			break;
1464 		}
1465 
1466 		error = lagg_port_destroy(lp, 1);
1467 		LAGG_WUNLOCK(sc);
1468 		break;
1469 	case SIOCSIFFLAGS:
1470 		/* Set flags on ports too */
1471 		LAGG_WLOCK(sc);
1472 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1473 			lagg_setflags(lp, 1);
1474 		}
1475 		LAGG_WUNLOCK(sc);
1476 
1477 		if (!(ifp->if_flags & IFF_UP) &&
1478 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1479 			/*
1480 			 * If interface is marked down and it is running,
1481 			 * then stop and disable it.
1482 			 */
1483 			LAGG_WLOCK(sc);
1484 			lagg_stop(sc);
1485 			LAGG_WUNLOCK(sc);
1486 		} else if ((ifp->if_flags & IFF_UP) &&
1487 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1488 			/*
1489 			 * If interface is marked up and it is stopped, then
1490 			 * start it.
1491 			 */
1492 			(*ifp->if_init)(sc);
1493 		}
1494 		break;
1495 	case SIOCADDMULTI:
1496 	case SIOCDELMULTI:
1497 		LAGG_WLOCK(sc);
1498 		error = lagg_ether_setmulti(sc);
1499 		LAGG_WUNLOCK(sc);
1500 		break;
1501 	case SIOCSIFMEDIA:
1502 	case SIOCGIFMEDIA:
1503 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1504 		break;
1505 
1506 	case SIOCSIFCAP:
1507 	case SIOCSIFMTU:
1508 		/* Do not allow the MTU or caps to be directly changed */
1509 		error = EINVAL;
1510 		break;
1511 
1512 	default:
1513 		error = ether_ioctl(ifp, cmd, data);
1514 		break;
1515 	}
1516 	return (error);
1517 }
1518 
1519 static int
1520 lagg_ether_setmulti(struct lagg_softc *sc)
1521 {
1522 	struct lagg_port *lp;
1523 
1524 	LAGG_WLOCK_ASSERT(sc);
1525 
1526 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1527 		/* First, remove any existing filter entries. */
1528 		lagg_ether_cmdmulti(lp, 0);
1529 		/* copy all addresses from the lagg interface to the port */
1530 		lagg_ether_cmdmulti(lp, 1);
1531 	}
1532 	return (0);
1533 }
1534 
1535 static int
1536 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1537 {
1538 	struct lagg_softc *sc = lp->lp_softc;
1539 	struct ifnet *ifp = lp->lp_ifp;
1540 	struct ifnet *scifp = sc->sc_ifp;
1541 	struct lagg_mc *mc;
1542 	struct ifmultiaddr *ifma;
1543 	int error;
1544 
1545 	LAGG_WLOCK_ASSERT(sc);
1546 
1547 	if (set) {
1548 		IF_ADDR_WLOCK(scifp);
1549 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1550 			if (ifma->ifma_addr->sa_family != AF_LINK)
1551 				continue;
1552 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1553 			if (mc == NULL) {
1554 				IF_ADDR_WUNLOCK(scifp);
1555 				return (ENOMEM);
1556 			}
1557 			bcopy(ifma->ifma_addr, &mc->mc_addr,
1558 			    ifma->ifma_addr->sa_len);
1559 			mc->mc_addr.sdl_index = ifp->if_index;
1560 			mc->mc_ifma = NULL;
1561 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1562 		}
1563 		IF_ADDR_WUNLOCK(scifp);
1564 		SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1565 			error = if_addmulti(ifp,
1566 			    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1567 			if (error)
1568 				return (error);
1569 		}
1570 	} else {
1571 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1572 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1573 			if (mc->mc_ifma && !lp->lp_detaching)
1574 				if_delmulti_ifma(mc->mc_ifma);
1575 			free(mc, M_DEVBUF);
1576 		}
1577 	}
1578 	return (0);
1579 }
1580 
1581 /* Handle a ref counted flag that should be set on the lagg port as well */
1582 static int
1583 lagg_setflag(struct lagg_port *lp, int flag, int status,
1584     int (*func)(struct ifnet *, int))
1585 {
1586 	struct lagg_softc *sc = lp->lp_softc;
1587 	struct ifnet *scifp = sc->sc_ifp;
1588 	struct ifnet *ifp = lp->lp_ifp;
1589 	int error;
1590 
1591 	LAGG_WLOCK_ASSERT(sc);
1592 
1593 	status = status ? (scifp->if_flags & flag) : 0;
1594 	/* Now "status" contains the flag value or 0 */
1595 
1596 	/*
1597 	 * See if recorded ports status is different from what
1598 	 * we want it to be.  If it is, flip it.  We record ports
1599 	 * status in lp_ifflags so that we won't clear ports flag
1600 	 * we haven't set.  In fact, we don't clear or set ports
1601 	 * flags directly, but get or release references to them.
1602 	 * That's why we can be sure that recorded flags still are
1603 	 * in accord with actual ports flags.
1604 	 */
1605 	if (status != (lp->lp_ifflags & flag)) {
1606 		error = (*func)(ifp, status);
1607 		if (error)
1608 			return (error);
1609 		lp->lp_ifflags &= ~flag;
1610 		lp->lp_ifflags |= status;
1611 	}
1612 	return (0);
1613 }
1614 
1615 /*
1616  * Handle IFF_* flags that require certain changes on the lagg port
1617  * if "status" is true, update ports flags respective to the lagg
1618  * if "status" is false, forcedly clear the flags set on port.
1619  */
1620 static int
1621 lagg_setflags(struct lagg_port *lp, int status)
1622 {
1623 	int error, i;
1624 
1625 	for (i = 0; lagg_pflags[i].flag; i++) {
1626 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1627 		    status, lagg_pflags[i].func);
1628 		if (error)
1629 			return (error);
1630 	}
1631 	return (0);
1632 }
1633 
1634 static int
1635 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1636 {
1637 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1638 	int error, len, mcast;
1639 	struct rm_priotracker tracker;
1640 
1641 	len = m->m_pkthdr.len;
1642 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1643 
1644 	LAGG_RLOCK(sc, &tracker);
1645 	/* We need a Tx algorithm and at least one port */
1646 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1647 		LAGG_RUNLOCK(sc, &tracker);
1648 		m_freem(m);
1649 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1650 		return (ENXIO);
1651 	}
1652 
1653 	ETHER_BPF_MTAP(ifp, m);
1654 
1655 	error = lagg_proto_start(sc, m);
1656 	LAGG_RUNLOCK(sc, &tracker);
1657 
1658 	if (error != 0)
1659 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1660 
1661 	return (error);
1662 }
1663 
1664 /*
1665  * The ifp->if_qflush entry point for lagg(4) is no-op.
1666  */
1667 static void
1668 lagg_qflush(struct ifnet *ifp __unused)
1669 {
1670 }
1671 
1672 static struct mbuf *
1673 lagg_input(struct ifnet *ifp, struct mbuf *m)
1674 {
1675 	struct lagg_port *lp = ifp->if_lagg;
1676 	struct lagg_softc *sc = lp->lp_softc;
1677 	struct ifnet *scifp = sc->sc_ifp;
1678 	struct rm_priotracker tracker;
1679 
1680 	LAGG_RLOCK(sc, &tracker);
1681 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1682 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1683 	    sc->sc_proto == LAGG_PROTO_NONE) {
1684 		LAGG_RUNLOCK(sc, &tracker);
1685 		m_freem(m);
1686 		return (NULL);
1687 	}
1688 
1689 	ETHER_BPF_MTAP(scifp, m);
1690 
1691 	if (lp->lp_detaching != 0) {
1692 		m_freem(m);
1693 		m = NULL;
1694 	} else
1695 		m = lagg_proto_input(sc, lp, m);
1696 
1697 	if (m != NULL) {
1698 		if (scifp->if_flags & IFF_MONITOR) {
1699 			m_freem(m);
1700 			m = NULL;
1701 		}
1702 	}
1703 
1704 	LAGG_RUNLOCK(sc, &tracker);
1705 	return (m);
1706 }
1707 
1708 static int
1709 lagg_media_change(struct ifnet *ifp)
1710 {
1711 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1712 
1713 	if (sc->sc_ifflags & IFF_DEBUG)
1714 		printf("%s\n", __func__);
1715 
1716 	/* Ignore */
1717 	return (0);
1718 }
1719 
1720 static void
1721 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1722 {
1723 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1724 	struct lagg_port *lp;
1725 	struct rm_priotracker tracker;
1726 
1727 	imr->ifm_status = IFM_AVALID;
1728 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1729 
1730 	LAGG_RLOCK(sc, &tracker);
1731 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1732 		if (LAGG_PORTACTIVE(lp))
1733 			imr->ifm_status |= IFM_ACTIVE;
1734 	}
1735 	LAGG_RUNLOCK(sc, &tracker);
1736 }
1737 
1738 static void
1739 lagg_linkstate(struct lagg_softc *sc)
1740 {
1741 	struct lagg_port *lp;
1742 	int new_link = LINK_STATE_DOWN;
1743 	uint64_t speed;
1744 
1745 	/* Our link is considered up if at least one of our ports is active */
1746 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1747 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1748 			new_link = LINK_STATE_UP;
1749 			break;
1750 		}
1751 	}
1752 	if_link_state_change(sc->sc_ifp, new_link);
1753 
1754 	/* Update if_baudrate to reflect the max possible speed */
1755 	switch (sc->sc_proto) {
1756 		case LAGG_PROTO_FAILOVER:
1757 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1758 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1759 			break;
1760 		case LAGG_PROTO_ROUNDROBIN:
1761 		case LAGG_PROTO_LOADBALANCE:
1762 		case LAGG_PROTO_BROADCAST:
1763 			speed = 0;
1764 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1765 				speed += lp->lp_ifp->if_baudrate;
1766 			sc->sc_ifp->if_baudrate = speed;
1767 			break;
1768 		case LAGG_PROTO_LACP:
1769 			/* LACP updates if_baudrate itself */
1770 			break;
1771 	}
1772 }
1773 
1774 static void
1775 lagg_port_state(struct ifnet *ifp, int state)
1776 {
1777 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1778 	struct lagg_softc *sc = NULL;
1779 
1780 	if (lp != NULL)
1781 		sc = lp->lp_softc;
1782 	if (sc == NULL)
1783 		return;
1784 
1785 	LAGG_WLOCK(sc);
1786 	lagg_linkstate(sc);
1787 	lagg_proto_linkstate(sc, lp);
1788 	LAGG_WUNLOCK(sc);
1789 }
1790 
1791 struct lagg_port *
1792 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1793 {
1794 	struct lagg_port *lp_next, *rval = NULL;
1795 	// int new_link = LINK_STATE_DOWN;
1796 
1797 	LAGG_RLOCK_ASSERT(sc);
1798 	/*
1799 	 * Search a port which reports an active link state.
1800 	 */
1801 
1802 	if (lp == NULL)
1803 		goto search;
1804 	if (LAGG_PORTACTIVE(lp)) {
1805 		rval = lp;
1806 		goto found;
1807 	}
1808 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1809 	    LAGG_PORTACTIVE(lp_next)) {
1810 		rval = lp_next;
1811 		goto found;
1812 	}
1813 
1814 search:
1815 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1816 		if (LAGG_PORTACTIVE(lp_next)) {
1817 			rval = lp_next;
1818 			goto found;
1819 		}
1820 	}
1821 
1822 found:
1823 	if (rval != NULL) {
1824 		/*
1825 		 * The IEEE 802.1D standard assumes that a lagg with
1826 		 * multiple ports is always full duplex. This is valid
1827 		 * for load sharing laggs and if at least two links
1828 		 * are active. Unfortunately, checking the latter would
1829 		 * be too expensive at this point.
1830 		 XXX
1831 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1832 		    (sc->sc_count > 1))
1833 			new_link = LINK_STATE_FULL_DUPLEX;
1834 		else
1835 			new_link = rval->lp_link_state;
1836 		 */
1837 	}
1838 
1839 	return (rval);
1840 }
1841 
1842 int
1843 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1844 {
1845 
1846 	return (ifp->if_transmit)(ifp, m);
1847 }
1848 
1849 /*
1850  * Simple round robin aggregation
1851  */
1852 static void
1853 lagg_rr_attach(struct lagg_softc *sc)
1854 {
1855 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1856 	sc->sc_seq = 0;
1857 }
1858 
1859 static int
1860 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1861 {
1862 	struct lagg_port *lp;
1863 	uint32_t p;
1864 
1865 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1866 	p %= sc->sc_count;
1867 	lp = SLIST_FIRST(&sc->sc_ports);
1868 	while (p--)
1869 		lp = SLIST_NEXT(lp, lp_entries);
1870 
1871 	/*
1872 	 * Check the port's link state. This will return the next active
1873 	 * port if the link is down or the port is NULL.
1874 	 */
1875 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1876 		m_freem(m);
1877 		return (ENETDOWN);
1878 	}
1879 
1880 	/* Send mbuf */
1881 	return (lagg_enqueue(lp->lp_ifp, m));
1882 }
1883 
1884 static struct mbuf *
1885 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1886 {
1887 	struct ifnet *ifp = sc->sc_ifp;
1888 
1889 	/* Just pass in the packet to our lagg device */
1890 	m->m_pkthdr.rcvif = ifp;
1891 
1892 	return (m);
1893 }
1894 
1895 /*
1896  * Broadcast mode
1897  */
1898 static int
1899 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1900 {
1901 	int active_ports = 0;
1902 	int errors = 0;
1903 	int ret;
1904 	struct lagg_port *lp, *last = NULL;
1905 	struct mbuf *m0;
1906 
1907 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1908 		if (!LAGG_PORTACTIVE(lp))
1909 			continue;
1910 
1911 		active_ports++;
1912 
1913 		if (last != NULL) {
1914 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
1915 			if (m0 == NULL) {
1916 				ret = ENOBUFS;
1917 				errors++;
1918 				break;
1919 			}
1920 
1921 			ret = lagg_enqueue(last->lp_ifp, m0);
1922 			if (ret != 0)
1923 				errors++;
1924 		}
1925 		last = lp;
1926 	}
1927 	if (last == NULL) {
1928 		m_freem(m);
1929 		return (ENOENT);
1930 	}
1931 	if ((last = lagg_link_active(sc, last)) == NULL) {
1932 		m_freem(m);
1933 		return (ENETDOWN);
1934 	}
1935 
1936 	ret = lagg_enqueue(last->lp_ifp, m);
1937 	if (ret != 0)
1938 		errors++;
1939 
1940 	if (errors == 0)
1941 		return (ret);
1942 
1943 	return (0);
1944 }
1945 
1946 static struct mbuf*
1947 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1948 {
1949 	struct ifnet *ifp = sc->sc_ifp;
1950 
1951 	/* Just pass in the packet to our lagg device */
1952 	m->m_pkthdr.rcvif = ifp;
1953 	return (m);
1954 }
1955 
1956 /*
1957  * Active failover
1958  */
1959 static int
1960 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1961 {
1962 	struct lagg_port *lp;
1963 
1964 	/* Use the master port if active or the next available port */
1965 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1966 		m_freem(m);
1967 		return (ENETDOWN);
1968 	}
1969 
1970 	/* Send mbuf */
1971 	return (lagg_enqueue(lp->lp_ifp, m));
1972 }
1973 
1974 static struct mbuf *
1975 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1976 {
1977 	struct ifnet *ifp = sc->sc_ifp;
1978 	struct lagg_port *tmp_tp;
1979 
1980 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
1981 		m->m_pkthdr.rcvif = ifp;
1982 		return (m);
1983 	}
1984 
1985 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1986 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1987 		/*
1988 		 * If tmp_tp is null, we've recieved a packet when all
1989 		 * our links are down. Weird, but process it anyways.
1990 		 */
1991 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1992 			m->m_pkthdr.rcvif = ifp;
1993 			return (m);
1994 		}
1995 	}
1996 
1997 	m_freem(m);
1998 	return (NULL);
1999 }
2000 
2001 /*
2002  * Loadbalancing
2003  */
2004 static void
2005 lagg_lb_attach(struct lagg_softc *sc)
2006 {
2007 	struct lagg_port *lp;
2008 	struct lagg_lb *lb;
2009 
2010 	lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
2011 
2012 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
2013 
2014 	lb->lb_key = m_ether_tcpip_hash_init();
2015 	sc->sc_psc = lb;
2016 
2017 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2018 		lagg_lb_port_create(lp);
2019 }
2020 
2021 static void
2022 lagg_lb_detach(struct lagg_softc *sc)
2023 {
2024 	struct lagg_lb *lb;
2025 
2026 	lb = (struct lagg_lb *)sc->sc_psc;
2027 	LAGG_WUNLOCK(sc);
2028 	if (lb != NULL)
2029 		free(lb, M_DEVBUF);
2030 }
2031 
2032 static int
2033 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2034 {
2035 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2036 	struct lagg_port *lp_next;
2037 	int i = 0;
2038 
2039 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2040 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2041 		if (lp_next == lp)
2042 			continue;
2043 		if (i >= LAGG_MAX_PORTS)
2044 			return (EINVAL);
2045 		if (sc->sc_ifflags & IFF_DEBUG)
2046 			printf("%s: port %s at index %d\n",
2047 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2048 		lb->lb_ports[i++] = lp_next;
2049 	}
2050 
2051 	return (0);
2052 }
2053 
2054 static int
2055 lagg_lb_port_create(struct lagg_port *lp)
2056 {
2057 	struct lagg_softc *sc = lp->lp_softc;
2058 	return (lagg_lb_porttable(sc, NULL));
2059 }
2060 
2061 static void
2062 lagg_lb_port_destroy(struct lagg_port *lp)
2063 {
2064 	struct lagg_softc *sc = lp->lp_softc;
2065 	lagg_lb_porttable(sc, lp);
2066 }
2067 
2068 static int
2069 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2070 {
2071 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2072 	struct lagg_port *lp = NULL;
2073 	uint32_t p = 0;
2074 
2075 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2076 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2077 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2078 	else
2079 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2080 	p %= sc->sc_count;
2081 	lp = lb->lb_ports[p];
2082 
2083 	/*
2084 	 * Check the port's link state. This will return the next active
2085 	 * port if the link is down or the port is NULL.
2086 	 */
2087 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2088 		m_freem(m);
2089 		return (ENETDOWN);
2090 	}
2091 
2092 	/* Send mbuf */
2093 	return (lagg_enqueue(lp->lp_ifp, m));
2094 }
2095 
2096 static struct mbuf *
2097 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2098 {
2099 	struct ifnet *ifp = sc->sc_ifp;
2100 
2101 	/* Just pass in the packet to our lagg device */
2102 	m->m_pkthdr.rcvif = ifp;
2103 
2104 	return (m);
2105 }
2106 
2107 /*
2108  * 802.3ad LACP
2109  */
2110 static void
2111 lagg_lacp_attach(struct lagg_softc *sc)
2112 {
2113 	struct lagg_port *lp;
2114 
2115 	lacp_attach(sc);
2116 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2117 		lacp_port_create(lp);
2118 }
2119 
2120 static void
2121 lagg_lacp_detach(struct lagg_softc *sc)
2122 {
2123 	struct lagg_port *lp;
2124 	void *psc;
2125 
2126 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2127 		lacp_port_destroy(lp);
2128 
2129 	psc = sc->sc_psc;
2130 	sc->sc_psc = NULL;
2131 	LAGG_WUNLOCK(sc);
2132 
2133 	lacp_detach(psc);
2134 }
2135 
2136 static void
2137 lagg_lacp_lladdr(struct lagg_softc *sc)
2138 {
2139 	struct lagg_port *lp;
2140 
2141 	/* purge all the lacp ports */
2142 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2143 		lacp_port_destroy(lp);
2144 
2145 	/* add them back in */
2146 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2147 		lacp_port_create(lp);
2148 }
2149 
2150 static int
2151 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2152 {
2153 	struct lagg_port *lp;
2154 
2155 	lp = lacp_select_tx_port(sc, m);
2156 	if (lp == NULL) {
2157 		m_freem(m);
2158 		return (ENETDOWN);
2159 	}
2160 
2161 	/* Send mbuf */
2162 	return (lagg_enqueue(lp->lp_ifp, m));
2163 }
2164 
2165 static struct mbuf *
2166 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2167 {
2168 	struct ifnet *ifp = sc->sc_ifp;
2169 	struct ether_header *eh;
2170 	u_short etype;
2171 
2172 	eh = mtod(m, struct ether_header *);
2173 	etype = ntohs(eh->ether_type);
2174 
2175 	/* Tap off LACP control messages */
2176 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2177 		m = lacp_input(lp, m);
2178 		if (m == NULL)
2179 			return (NULL);
2180 	}
2181 
2182 	/*
2183 	 * If the port is not collecting or not in the active aggregator then
2184 	 * free and return.
2185 	 */
2186 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2187 		m_freem(m);
2188 		return (NULL);
2189 	}
2190 
2191 	m->m_pkthdr.rcvif = ifp;
2192 	return (m);
2193 }
2194 
2195