xref: /freebsd/sys/net/if_lagg.c (revision b89a7cc2ed6e4398d5be502f5bb5885d1ec6ff0f)
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, 2016 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 #include "opt_ratelimit.h"
27 
28 #include <sys/param.h>
29 #include <sys/kernel.h>
30 #include <sys/malloc.h>
31 #include <sys/mbuf.h>
32 #include <sys/queue.h>
33 #include <sys/socket.h>
34 #include <sys/sockio.h>
35 #include <sys/sysctl.h>
36 #include <sys/module.h>
37 #include <sys/priv.h>
38 #include <sys/systm.h>
39 #include <sys/proc.h>
40 #include <sys/lock.h>
41 #include <sys/rmlock.h>
42 #include <sys/sx.h>
43 #include <sys/taskqueue.h>
44 #include <sys/eventhandler.h>
45 
46 #include <net/ethernet.h>
47 #include <net/if.h>
48 #include <net/if_clone.h>
49 #include <net/if_arp.h>
50 #include <net/if_dl.h>
51 #include <net/if_media.h>
52 #include <net/if_types.h>
53 #include <net/if_var.h>
54 #include <net/bpf.h>
55 #include <net/vnet.h>
56 
57 #if defined(INET) || defined(INET6)
58 #include <netinet/in.h>
59 #include <netinet/ip.h>
60 #endif
61 #ifdef INET
62 #include <netinet/in_systm.h>
63 #include <netinet/if_ether.h>
64 #endif
65 
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #include <netinet6/in6_var.h>
69 #include <netinet6/in6_ifattach.h>
70 #endif
71 
72 #include <net/if_vlan_var.h>
73 #include <net/if_lagg.h>
74 #include <net/ieee8023ad_lacp.h>
75 
76 #define	LAGG_RLOCK()	struct epoch_tracker lagg_et; epoch_enter_preempt(net_epoch_preempt, &lagg_et)
77 #define	LAGG_RUNLOCK()	epoch_exit_preempt(net_epoch_preempt, &lagg_et)
78 #define	LAGG_RLOCK_ASSERT()	MPASS(in_epoch(net_epoch_preempt))
79 #define	LAGG_UNLOCK_ASSERT()	MPASS(!in_epoch(net_epoch_preempt))
80 
81 #define	LAGG_SX_INIT(_sc)	sx_init(&(_sc)->sc_sx, "if_lagg sx")
82 #define	LAGG_SX_DESTROY(_sc)	sx_destroy(&(_sc)->sc_sx)
83 #define	LAGG_XLOCK(_sc)		sx_xlock(&(_sc)->sc_sx)
84 #define	LAGG_XUNLOCK(_sc)	sx_xunlock(&(_sc)->sc_sx)
85 #define	LAGG_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
86 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
87 
88 /* Special flags we should propagate to the lagg ports. */
89 static struct {
90 	int flag;
91 	int (*func)(struct ifnet *, int);
92 } lagg_pflags[] = {
93 	{IFF_PROMISC, ifpromisc},
94 	{IFF_ALLMULTI, if_allmulti},
95 	{0, NULL}
96 };
97 
98 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
99 #define	V_lagg_list	VNET(lagg_list)
100 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
101 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
102 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
103 					"if_lagg list", NULL, MTX_DEF)
104 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
105 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
106 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
107 eventhandler_tag	lagg_detach_cookie = NULL;
108 
109 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
110 static void	lagg_clone_destroy(struct ifnet *);
111 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
112 #define	V_lagg_cloner	VNET(lagg_cloner)
113 static const char laggname[] = "lagg";
114 
115 static void	lagg_capabilities(struct lagg_softc *);
116 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
117 static int	lagg_port_destroy(struct lagg_port *, int);
118 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
119 static void	lagg_linkstate(struct lagg_softc *);
120 static void	lagg_port_state(struct ifnet *, int);
121 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
122 static int	lagg_port_output(struct ifnet *, struct mbuf *,
123 		    const struct sockaddr *, struct route *);
124 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
125 #ifdef LAGG_PORT_STACKING
126 static int	lagg_port_checkstacking(struct lagg_softc *);
127 #endif
128 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
129 static void	lagg_init(void *);
130 static void	lagg_stop(struct lagg_softc *);
131 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
132 #ifdef RATELIMIT
133 static int	lagg_snd_tag_alloc(struct ifnet *,
134 		    union if_snd_tag_alloc_params *,
135 		    struct m_snd_tag **);
136 #endif
137 static int	lagg_setmulti(struct lagg_port *);
138 static int	lagg_clrmulti(struct lagg_port *);
139 static	int	lagg_setcaps(struct lagg_port *, int cap);
140 static	int	lagg_setflag(struct lagg_port *, int, int,
141 		    int (*func)(struct ifnet *, int));
142 static	int	lagg_setflags(struct lagg_port *, int status);
143 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
144 static int	lagg_transmit(struct ifnet *, struct mbuf *);
145 static void	lagg_qflush(struct ifnet *);
146 static int	lagg_media_change(struct ifnet *);
147 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
148 static struct lagg_port *lagg_link_active(struct lagg_softc *,
149 	    struct lagg_port *);
150 
151 /* Simple round robin */
152 static void	lagg_rr_attach(struct lagg_softc *);
153 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
154 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
155 		    struct mbuf *);
156 
157 /* Active failover */
158 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
159 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
160 		    struct mbuf *);
161 
162 /* Loadbalancing */
163 static void	lagg_lb_attach(struct lagg_softc *);
164 static void	lagg_lb_detach(struct lagg_softc *);
165 static int	lagg_lb_port_create(struct lagg_port *);
166 static void	lagg_lb_port_destroy(struct lagg_port *);
167 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
168 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
169 		    struct mbuf *);
170 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
171 
172 /* Broadcast */
173 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
174 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
175 		    struct mbuf *);
176 
177 /* 802.3ad LACP */
178 static void	lagg_lacp_attach(struct lagg_softc *);
179 static void	lagg_lacp_detach(struct lagg_softc *);
180 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
181 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
182 		    struct mbuf *);
183 static void	lagg_lacp_lladdr(struct lagg_softc *);
184 
185 /* lagg protocol table */
186 static const struct lagg_proto {
187 	lagg_proto	pr_num;
188 	void		(*pr_attach)(struct lagg_softc *);
189 	void		(*pr_detach)(struct lagg_softc *);
190 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
191 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
192 			    struct mbuf *);
193 	int		(*pr_addport)(struct lagg_port *);
194 	void		(*pr_delport)(struct lagg_port *);
195 	void		(*pr_linkstate)(struct lagg_port *);
196 	void 		(*pr_init)(struct lagg_softc *);
197 	void 		(*pr_stop)(struct lagg_softc *);
198 	void 		(*pr_lladdr)(struct lagg_softc *);
199 	void		(*pr_request)(struct lagg_softc *, void *);
200 	void		(*pr_portreq)(struct lagg_port *, void *);
201 } lagg_protos[] = {
202     {
203 	.pr_num = LAGG_PROTO_NONE
204     },
205     {
206 	.pr_num = LAGG_PROTO_ROUNDROBIN,
207 	.pr_attach = lagg_rr_attach,
208 	.pr_start = lagg_rr_start,
209 	.pr_input = lagg_rr_input,
210     },
211     {
212 	.pr_num = LAGG_PROTO_FAILOVER,
213 	.pr_start = lagg_fail_start,
214 	.pr_input = lagg_fail_input,
215     },
216     {
217 	.pr_num = LAGG_PROTO_LOADBALANCE,
218 	.pr_attach = lagg_lb_attach,
219 	.pr_detach = lagg_lb_detach,
220 	.pr_start = lagg_lb_start,
221 	.pr_input = lagg_lb_input,
222 	.pr_addport = lagg_lb_port_create,
223 	.pr_delport = lagg_lb_port_destroy,
224     },
225     {
226 	.pr_num = LAGG_PROTO_LACP,
227 	.pr_attach = lagg_lacp_attach,
228 	.pr_detach = lagg_lacp_detach,
229 	.pr_start = lagg_lacp_start,
230 	.pr_input = lagg_lacp_input,
231 	.pr_addport = lacp_port_create,
232 	.pr_delport = lacp_port_destroy,
233 	.pr_linkstate = lacp_linkstate,
234 	.pr_init = lacp_init,
235 	.pr_stop = lacp_stop,
236 	.pr_lladdr = lagg_lacp_lladdr,
237 	.pr_request = lacp_req,
238 	.pr_portreq = lacp_portreq,
239     },
240     {
241 	.pr_num = LAGG_PROTO_BROADCAST,
242 	.pr_start = lagg_bcast_start,
243 	.pr_input = lagg_bcast_input,
244     },
245 };
246 
247 SYSCTL_DECL(_net_link);
248 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
249     "Link Aggregation");
250 
251 /* Allow input on any failover links */
252 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
253 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
254 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
255     &VNET_NAME(lagg_failover_rx_all), 0,
256     "Accept input from any interface in a failover lagg");
257 
258 /* Default value for using flowid */
259 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
260 #define	V_def_use_flowid	VNET(def_use_flowid)
261 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
262     &VNET_NAME(def_use_flowid), 0,
263     "Default setting for using flow id for load sharing");
264 
265 /* Default value for flowid shift */
266 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
267 #define	V_def_flowid_shift	VNET(def_flowid_shift)
268 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
269     &VNET_NAME(def_flowid_shift), 0,
270     "Default setting for flowid shift for load sharing");
271 
272 static void
273 vnet_lagg_init(const void *unused __unused)
274 {
275 
276 	LAGG_LIST_LOCK_INIT();
277 	SLIST_INIT(&V_lagg_list);
278 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
279 	    lagg_clone_destroy, 0);
280 }
281 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
282     vnet_lagg_init, NULL);
283 
284 static void
285 vnet_lagg_uninit(const void *unused __unused)
286 {
287 
288 	if_clone_detach(V_lagg_cloner);
289 	LAGG_LIST_LOCK_DESTROY();
290 }
291 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
292     vnet_lagg_uninit, NULL);
293 
294 static int
295 lagg_modevent(module_t mod, int type, void *data)
296 {
297 
298 	switch (type) {
299 	case MOD_LOAD:
300 		lagg_input_p = lagg_input;
301 		lagg_linkstate_p = lagg_port_state;
302 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
303 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
304 		    EVENTHANDLER_PRI_ANY);
305 		break;
306 	case MOD_UNLOAD:
307 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
308 		    lagg_detach_cookie);
309 		lagg_input_p = NULL;
310 		lagg_linkstate_p = NULL;
311 		break;
312 	default:
313 		return (EOPNOTSUPP);
314 	}
315 	return (0);
316 }
317 
318 static moduledata_t lagg_mod = {
319 	"if_lagg",
320 	lagg_modevent,
321 	0
322 };
323 
324 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
325 MODULE_VERSION(if_lagg, 1);
326 
327 static void
328 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
329 {
330 
331 	LAGG_XLOCK_ASSERT(sc);
332 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
333 	    __func__, sc));
334 
335 	if (sc->sc_ifflags & IFF_DEBUG)
336 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
337 
338 	if (lagg_protos[pr].pr_attach != NULL)
339 		lagg_protos[pr].pr_attach(sc);
340 	sc->sc_proto = pr;
341 }
342 
343 static void
344 lagg_proto_detach(struct lagg_softc *sc)
345 {
346 	lagg_proto pr;
347 
348 	LAGG_XLOCK_ASSERT(sc);
349 	pr = sc->sc_proto;
350 	sc->sc_proto = LAGG_PROTO_NONE;
351 
352 	if (lagg_protos[pr].pr_detach != NULL)
353 		lagg_protos[pr].pr_detach(sc);
354 }
355 
356 static int
357 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
358 {
359 
360 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
361 }
362 
363 static struct mbuf *
364 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
365 {
366 
367 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
368 }
369 
370 static int
371 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
372 {
373 
374 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
375 		return (0);
376 	else
377 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
378 }
379 
380 static void
381 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
382 {
383 
384 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
385 		lagg_protos[sc->sc_proto].pr_delport(lp);
386 }
387 
388 static void
389 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
390 {
391 
392 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
393 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
394 }
395 
396 static void
397 lagg_proto_init(struct lagg_softc *sc)
398 {
399 
400 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
401 		lagg_protos[sc->sc_proto].pr_init(sc);
402 }
403 
404 static void
405 lagg_proto_stop(struct lagg_softc *sc)
406 {
407 
408 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
409 		lagg_protos[sc->sc_proto].pr_stop(sc);
410 }
411 
412 static void
413 lagg_proto_lladdr(struct lagg_softc *sc)
414 {
415 
416 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
417 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
418 }
419 
420 static void
421 lagg_proto_request(struct lagg_softc *sc, void *v)
422 {
423 
424 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
425 		lagg_protos[sc->sc_proto].pr_request(sc, v);
426 }
427 
428 static void
429 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
430 {
431 
432 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
433 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
434 }
435 
436 /*
437  * This routine is run via an vlan
438  * config EVENT
439  */
440 static void
441 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
442 {
443 	struct lagg_softc *sc = ifp->if_softc;
444 	struct lagg_port *lp;
445 
446 	if (ifp->if_softc !=  arg)   /* Not our event */
447 		return;
448 
449 	LAGG_RLOCK();
450 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
451 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
452 	LAGG_RUNLOCK();
453 }
454 
455 /*
456  * This routine is run via an vlan
457  * unconfig EVENT
458  */
459 static void
460 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
461 {
462 	struct lagg_softc *sc = ifp->if_softc;
463 	struct lagg_port *lp;
464 
465 	if (ifp->if_softc !=  arg)   /* Not our event */
466 		return;
467 
468 	LAGG_RLOCK();
469 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
470 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
471 	LAGG_RUNLOCK();
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 	LAGG_SX_INIT(sc);
488 
489 	LAGG_XLOCK(sc);
490 	if (V_def_use_flowid)
491 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
492 	sc->flowid_shift = V_def_flowid_shift;
493 
494 	/* Hash all layers by default */
495 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
496 
497 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
498 
499 	CK_SLIST_INIT(&sc->sc_ports);
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 #ifdef RATELIMIT
516 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
517 #endif
518 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
519 
520 	/*
521 	 * Attach as an ordinary ethernet device, children will be attached
522 	 * as special device IFT_IEEE8023ADLAG.
523 	 */
524 	ether_ifattach(ifp, eaddr);
525 
526 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
527 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
528 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
529 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
530 
531 	/* Insert into the global list of laggs */
532 	LAGG_LIST_LOCK();
533 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
534 	LAGG_LIST_UNLOCK();
535 	LAGG_XUNLOCK(sc);
536 
537 	return (0);
538 }
539 
540 static void
541 lagg_clone_destroy(struct ifnet *ifp)
542 {
543 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
544 	struct lagg_port *lp;
545 
546 	LAGG_XLOCK(sc);
547 	sc->sc_destroying = 1;
548 	lagg_stop(sc);
549 	ifp->if_flags &= ~IFF_UP;
550 
551 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
552 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
553 
554 	/* Shutdown and remove lagg ports */
555 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
556 		lagg_port_destroy(lp, 1);
557 
558 	/* Unhook the aggregation protocol */
559 	lagg_proto_detach(sc);
560 	LAGG_XUNLOCK(sc);
561 
562 	ifmedia_removeall(&sc->sc_media);
563 	ether_ifdetach(ifp);
564 	if_free(ifp);
565 
566 	LAGG_LIST_LOCK();
567 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
568 	LAGG_LIST_UNLOCK();
569 
570 	LAGG_SX_DESTROY(sc);
571 	free(sc, M_DEVBUF);
572 }
573 
574 static void
575 lagg_capabilities(struct lagg_softc *sc)
576 {
577 	struct lagg_port *lp;
578 	int cap, ena, pena;
579 	uint64_t hwa;
580 	struct ifnet_hw_tsomax hw_tsomax;
581 
582 	LAGG_XLOCK_ASSERT(sc);
583 
584 	/* Get common enabled capabilities for the lagg ports */
585 	ena = ~0;
586 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
587 		ena &= lp->lp_ifp->if_capenable;
588 	ena = (ena == ~0 ? 0 : ena);
589 
590 	/*
591 	 * Apply common enabled capabilities back to the lagg ports.
592 	 * May require several iterations if they are dependent.
593 	 */
594 	do {
595 		pena = ena;
596 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
597 			lagg_setcaps(lp, ena);
598 			ena &= lp->lp_ifp->if_capenable;
599 		}
600 	} while (pena != ena);
601 
602 	/* Get other capabilities from the lagg ports */
603 	cap = ~0;
604 	hwa = ~(uint64_t)0;
605 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
606 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
607 		cap &= lp->lp_ifp->if_capabilities;
608 		hwa &= lp->lp_ifp->if_hwassist;
609 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
610 	}
611 	cap = (cap == ~0 ? 0 : cap);
612 	hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
613 
614 	if (sc->sc_ifp->if_capabilities != cap ||
615 	    sc->sc_ifp->if_capenable != ena ||
616 	    sc->sc_ifp->if_hwassist != hwa ||
617 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
618 		sc->sc_ifp->if_capabilities = cap;
619 		sc->sc_ifp->if_capenable = ena;
620 		sc->sc_ifp->if_hwassist = hwa;
621 		getmicrotime(&sc->sc_ifp->if_lastchange);
622 
623 		if (sc->sc_ifflags & IFF_DEBUG)
624 			if_printf(sc->sc_ifp,
625 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
626 	}
627 }
628 
629 static int
630 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
631 {
632 	struct lagg_softc *sc_ptr;
633 	struct lagg_port *lp, *tlp;
634 	struct ifreq ifr;
635 	int error, i, oldmtu;
636 	uint64_t *pval;
637 
638 	LAGG_XLOCK_ASSERT(sc);
639 
640 	if (sc->sc_ifp == ifp) {
641 		if_printf(sc->sc_ifp,
642 		    "cannot add a lagg to itself as a port\n");
643 		return (EINVAL);
644 	}
645 
646 	/* Limit the maximal number of lagg ports */
647 	if (sc->sc_count >= LAGG_MAX_PORTS)
648 		return (ENOSPC);
649 
650 	/* Check if port has already been associated to a lagg */
651 	if (ifp->if_lagg != NULL) {
652 		/* Port is already in the current lagg? */
653 		lp = (struct lagg_port *)ifp->if_lagg;
654 		if (lp->lp_softc == sc)
655 			return (EEXIST);
656 		return (EBUSY);
657 	}
658 
659 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
660 	if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
661 		return (EPROTONOSUPPORT);
662 
663 	/* Allow the first Ethernet member to define the MTU */
664 	oldmtu = -1;
665 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
666 		sc->sc_ifp->if_mtu = ifp->if_mtu;
667 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
668 		if (ifp->if_ioctl == NULL) {
669 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
670 			    ifp->if_xname);
671 			return (EINVAL);
672 		}
673 		oldmtu = ifp->if_mtu;
674 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
675 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
676 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
677 		if (error != 0) {
678 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
679 			    ifp->if_xname);
680 			return (error);
681 		}
682 		ifr.ifr_mtu = oldmtu;
683 	}
684 
685 	lp = malloc(sizeof(struct lagg_port), M_DEVBUF, M_WAITOK|M_ZERO);
686 	lp->lp_softc = sc;
687 
688 	/* Check if port is a stacked lagg */
689 	LAGG_LIST_LOCK();
690 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
691 		if (ifp == sc_ptr->sc_ifp) {
692 			LAGG_LIST_UNLOCK();
693 			free(lp, M_DEVBUF);
694 			if (oldmtu != -1)
695 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
696 				    (caddr_t)&ifr);
697 			return (EINVAL);
698 			/* XXX disable stacking for the moment, its untested */
699 #ifdef LAGG_PORT_STACKING
700 			lp->lp_flags |= LAGG_PORT_STACK;
701 			if (lagg_port_checkstacking(sc_ptr) >=
702 			    LAGG_MAX_STACKING) {
703 				LAGG_LIST_UNLOCK();
704 				free(lp, M_DEVBUF);
705 				if (oldmtu != -1)
706 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
707 					    (caddr_t)&ifr);
708 				return (E2BIG);
709 			}
710 #endif
711 		}
712 	}
713 	LAGG_LIST_UNLOCK();
714 
715 	if_ref(ifp);
716 	lp->lp_ifp = ifp;
717 
718 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
719 	lp->lp_ifcapenable = ifp->if_capenable;
720 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
721 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
722 		lagg_proto_lladdr(sc);
723 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
724 	} else {
725 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
726 	}
727 	lagg_setflags(lp, 1);
728 
729 	if (CK_SLIST_EMPTY(&sc->sc_ports))
730 		sc->sc_primary = lp;
731 
732 	/* Change the interface type */
733 	lp->lp_iftype = ifp->if_type;
734 	ifp->if_type = IFT_IEEE8023ADLAG;
735 	ifp->if_lagg = lp;
736 	lp->lp_ioctl = ifp->if_ioctl;
737 	ifp->if_ioctl = lagg_port_ioctl;
738 	lp->lp_output = ifp->if_output;
739 	ifp->if_output = lagg_port_output;
740 
741 	/* Read port counters */
742 	pval = lp->port_counters.val;
743 	for (i = 0; i < IFCOUNTERS; i++, pval++)
744 		*pval = ifp->if_get_counter(ifp, i);
745 
746 	/*
747 	 * Insert into the list of ports.
748 	 * Keep ports sorted by if_index. It is handy, when configuration
749 	 * is predictable and `ifconfig laggN create ...` command
750 	 * will lead to the same result each time.
751 	 */
752 	LAGG_RLOCK();
753 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
754 		if (tlp->lp_ifp->if_index < ifp->if_index && (
755 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
756 		    ((struct  lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
757 		    ifp->if_index))
758 			break;
759 	}
760 	LAGG_RUNLOCK();
761 	if (tlp != NULL)
762 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
763 	else
764 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
765 	sc->sc_count++;
766 
767 	lagg_setmulti(lp);
768 
769 
770 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
771 		/* Remove the port, without calling pr_delport. */
772 		lagg_port_destroy(lp, 0);
773 		if (oldmtu != -1)
774 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
775 		return (error);
776 	}
777 
778 	/* Update lagg capabilities */
779 	lagg_capabilities(sc);
780 	lagg_linkstate(sc);
781 
782 	return (0);
783 }
784 
785 #ifdef LAGG_PORT_STACKING
786 static int
787 lagg_port_checkstacking(struct lagg_softc *sc)
788 {
789 	struct lagg_softc *sc_ptr;
790 	struct lagg_port *lp;
791 	int m = 0;
792 
793 	LAGG_SXLOCK_ASSERT(sc);
794 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
795 		if (lp->lp_flags & LAGG_PORT_STACK) {
796 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
797 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
798 		}
799 	}
800 
801 	return (m + 1);
802 }
803 #endif
804 
805 static void
806 lagg_port_destroy_cb(epoch_context_t ec)
807 {
808 	struct lagg_port *lp;
809 	struct ifnet *ifp;
810 
811 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
812 	ifp = lp->lp_ifp;
813 
814 	if_rele(ifp);
815 	free(lp, M_DEVBUF);
816 }
817 
818 static int
819 lagg_port_destroy(struct lagg_port *lp, int rundelport)
820 {
821 	struct lagg_softc *sc = lp->lp_softc;
822 	struct lagg_port *lp_ptr, *lp0;
823 	struct ifnet *ifp = lp->lp_ifp;
824 	uint64_t *pval, vdiff;
825 	int i;
826 
827 	LAGG_XLOCK_ASSERT(sc);
828 
829 	if (rundelport)
830 		lagg_proto_delport(sc, lp);
831 
832 	if (lp->lp_detaching == 0)
833 		lagg_clrmulti(lp);
834 
835 	/* Restore interface */
836 	ifp->if_type = lp->lp_iftype;
837 	ifp->if_ioctl = lp->lp_ioctl;
838 	ifp->if_output = lp->lp_output;
839 	ifp->if_lagg = NULL;
840 
841 	/* Update detached port counters */
842 	pval = lp->port_counters.val;
843 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
844 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
845 		sc->detached_counters.val[i] += vdiff;
846 	}
847 
848 	/* Finally, remove the port from the lagg */
849 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
850 	sc->sc_count--;
851 
852 	/* Update the primary interface */
853 	if (lp == sc->sc_primary) {
854 		uint8_t lladdr[ETHER_ADDR_LEN];
855 
856 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
857 			bzero(&lladdr, ETHER_ADDR_LEN);
858 		else
859 			bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN);
860 		sc->sc_primary = lp0;
861 		if (sc->sc_destroying == 0) {
862 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
863 			lagg_proto_lladdr(sc);
864 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
865 		}
866 
867 		/*
868 		 * Update lladdr for each port (new primary needs update
869 		 * as well, to switch from old lladdr to its 'real' one)
870 		 */
871 		CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
872 			if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN);
873 	}
874 
875 	if (lp->lp_ifflags)
876 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
877 
878 	if (lp->lp_detaching == 0) {
879 		lagg_setflags(lp, 0);
880 		lagg_setcaps(lp, lp->lp_ifcapenable);
881 		if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN);
882 	}
883 
884 	/*
885 	 * free port and release it's ifnet reference after a grace period has
886 	 * elapsed.
887 	 */
888 	epoch_call(net_epoch_preempt, &lp->lp_epoch_ctx, lagg_port_destroy_cb);
889 	/* Update lagg capabilities */
890 	lagg_capabilities(sc);
891 	lagg_linkstate(sc);
892 
893 	return (0);
894 }
895 
896 static int
897 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
898 {
899 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
900 	struct lagg_softc *sc;
901 	struct lagg_port *lp = NULL;
902 	int error = 0;
903 
904 	/* Should be checked by the caller */
905 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
906 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
907 		goto fallback;
908 
909 	switch (cmd) {
910 	case SIOCGLAGGPORT:
911 		if (rp->rp_portname[0] == '\0' ||
912 		    ifunit(rp->rp_portname) != ifp) {
913 			error = EINVAL;
914 			break;
915 		}
916 
917 		LAGG_RLOCK();
918 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
919 			error = ENOENT;
920 			LAGG_RUNLOCK();
921 			break;
922 		}
923 
924 		lagg_port2req(lp, rp);
925 		LAGG_RUNLOCK();
926 		break;
927 
928 	case SIOCSIFCAP:
929 		if (lp->lp_ioctl == NULL) {
930 			error = EINVAL;
931 			break;
932 		}
933 		error = (*lp->lp_ioctl)(ifp, cmd, data);
934 		if (error)
935 			break;
936 
937 		/* Update lagg interface capabilities */
938 		LAGG_XLOCK(sc);
939 		lagg_capabilities(sc);
940 		LAGG_XUNLOCK(sc);
941 		VLAN_CAPABILITIES(sc->sc_ifp);
942 		break;
943 
944 	case SIOCSIFMTU:
945 		/* Do not allow the MTU to be changed once joined */
946 		error = EINVAL;
947 		break;
948 
949 	default:
950 		goto fallback;
951 	}
952 
953 	return (error);
954 
955 fallback:
956 	if (lp != NULL && lp->lp_ioctl != NULL)
957 		return ((*lp->lp_ioctl)(ifp, cmd, data));
958 
959 	return (EINVAL);
960 }
961 
962 /*
963  * Requests counter @cnt data.
964  *
965  * Counter value is calculated the following way:
966  * 1) for each port, sum  difference between current and "initial" measurements.
967  * 2) add lagg logical interface counters.
968  * 3) add data from detached_counters array.
969  *
970  * We also do the following things on ports attach/detach:
971  * 1) On port attach we store all counters it has into port_counter array.
972  * 2) On port detach we add the different between "initial" and
973  *   current counters data to detached_counters array.
974  */
975 static uint64_t
976 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
977 {
978 	struct lagg_softc *sc;
979 	struct lagg_port *lp;
980 	struct ifnet *lpifp;
981 	uint64_t newval, oldval, vsum;
982 
983 	/* Revise this when we've got non-generic counters. */
984 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
985 
986 	sc = (struct lagg_softc *)ifp->if_softc;
987 
988 	vsum = 0;
989 	LAGG_RLOCK();
990 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
991 		/* Saved attached value */
992 		oldval = lp->port_counters.val[cnt];
993 		/* current value */
994 		lpifp = lp->lp_ifp;
995 		newval = lpifp->if_get_counter(lpifp, cnt);
996 		/* Calculate diff and save new */
997 		vsum += newval - oldval;
998 	}
999 	LAGG_RUNLOCK();
1000 
1001 	/*
1002 	 * Add counter data which might be added by upper
1003 	 * layer protocols operating on logical interface.
1004 	 */
1005 	vsum += if_get_counter_default(ifp, cnt);
1006 
1007 	/*
1008 	 * Add counter data from detached ports counters
1009 	 */
1010 	vsum += sc->detached_counters.val[cnt];
1011 
1012 
1013 	return (vsum);
1014 }
1015 
1016 /*
1017  * For direct output to child ports.
1018  */
1019 static int
1020 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1021 	const struct sockaddr *dst, struct route *ro)
1022 {
1023 	struct lagg_port *lp = ifp->if_lagg;
1024 
1025 	switch (dst->sa_family) {
1026 		case pseudo_AF_HDRCMPLT:
1027 		case AF_UNSPEC:
1028 			return ((*lp->lp_output)(ifp, m, dst, ro));
1029 	}
1030 
1031 	/* drop any other frames */
1032 	m_freem(m);
1033 	return (ENETDOWN);
1034 }
1035 
1036 static void
1037 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1038 {
1039 	struct lagg_port *lp;
1040 	struct lagg_softc *sc;
1041 
1042 	if ((lp = ifp->if_lagg) == NULL)
1043 		return;
1044 	/* If the ifnet is just being renamed, don't do anything. */
1045 	if (ifp->if_flags & IFF_RENAMING)
1046 		return;
1047 
1048 	sc = lp->lp_softc;
1049 
1050 	LAGG_XLOCK(sc);
1051 	lp->lp_detaching = 1;
1052 	lagg_port_destroy(lp, 1);
1053 	LAGG_XUNLOCK(sc);
1054 	VLAN_CAPABILITIES(sc->sc_ifp);
1055 }
1056 
1057 static void
1058 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1059 {
1060 	struct lagg_softc *sc = lp->lp_softc;
1061 
1062 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1063 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1064 	rp->rp_prio = lp->lp_prio;
1065 	rp->rp_flags = lp->lp_flags;
1066 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1067 
1068 	/* Add protocol specific flags */
1069 	switch (sc->sc_proto) {
1070 		case LAGG_PROTO_FAILOVER:
1071 			if (lp == sc->sc_primary)
1072 				rp->rp_flags |= LAGG_PORT_MASTER;
1073 			if (lp == lagg_link_active(sc, sc->sc_primary))
1074 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1075 			break;
1076 
1077 		case LAGG_PROTO_ROUNDROBIN:
1078 		case LAGG_PROTO_LOADBALANCE:
1079 		case LAGG_PROTO_BROADCAST:
1080 			if (LAGG_PORTACTIVE(lp))
1081 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1082 			break;
1083 
1084 		case LAGG_PROTO_LACP:
1085 			/* LACP has a different definition of active */
1086 			if (lacp_isactive(lp))
1087 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1088 			if (lacp_iscollecting(lp))
1089 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1090 			if (lacp_isdistributing(lp))
1091 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1092 			break;
1093 	}
1094 
1095 }
1096 
1097 static void
1098 lagg_init(void *xsc)
1099 {
1100 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1101 	struct ifnet *ifp = sc->sc_ifp;
1102 	struct lagg_port *lp;
1103 
1104 	LAGG_XLOCK(sc);
1105 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1106 		LAGG_XUNLOCK(sc);
1107 		return;
1108 	}
1109 
1110 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1111 
1112 	/*
1113 	 * Update the port lladdrs if needed.
1114 	 * This might be if_setlladdr() notification
1115 	 * that lladdr has been changed.
1116 	 */
1117 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1118 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1119 		    ETHER_ADDR_LEN) != 0)
1120 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN);
1121 	}
1122 
1123 	lagg_proto_init(sc);
1124 
1125 	LAGG_XUNLOCK(sc);
1126 }
1127 
1128 static void
1129 lagg_stop(struct lagg_softc *sc)
1130 {
1131 	struct ifnet *ifp = sc->sc_ifp;
1132 
1133 	LAGG_XLOCK_ASSERT(sc);
1134 
1135 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1136 		return;
1137 
1138 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1139 
1140 	lagg_proto_stop(sc);
1141 }
1142 
1143 static int
1144 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1145 {
1146 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1147 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1148 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1149 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1150 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1151 	struct ifreq *ifr = (struct ifreq *)data;
1152 	struct lagg_port *lp;
1153 	struct ifnet *tpif;
1154 	struct thread *td = curthread;
1155 	char *buf, *outbuf;
1156 	int count, buflen, len, error = 0;
1157 
1158 	bzero(&rpbuf, sizeof(rpbuf));
1159 
1160 	switch (cmd) {
1161 	case SIOCGLAGG:
1162 		LAGG_XLOCK(sc);
1163 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1164 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1165 		ra->ra_proto = sc->sc_proto;
1166 		lagg_proto_request(sc, &ra->ra_psc);
1167 		count = 0;
1168 		buf = outbuf;
1169 		len = min(ra->ra_size, buflen);
1170 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1171 			if (len < sizeof(rpbuf))
1172 				break;
1173 
1174 			lagg_port2req(lp, &rpbuf);
1175 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1176 			count++;
1177 			buf += sizeof(rpbuf);
1178 			len -= sizeof(rpbuf);
1179 		}
1180 		LAGG_XUNLOCK(sc);
1181 		ra->ra_ports = count;
1182 		ra->ra_size = count * sizeof(rpbuf);
1183 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1184 		free(outbuf, M_TEMP);
1185 		break;
1186 	case SIOCSLAGG:
1187 		error = priv_check(td, PRIV_NET_LAGG);
1188 		if (error)
1189 			break;
1190 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1191 			error = EPROTONOSUPPORT;
1192 			break;
1193 		}
1194 
1195 		LAGG_XLOCK(sc);
1196 		lagg_proto_detach(sc);
1197 		LAGG_UNLOCK_ASSERT();
1198 		lagg_proto_attach(sc, ra->ra_proto);
1199 		LAGG_XUNLOCK(sc);
1200 		break;
1201 	case SIOCGLAGGOPTS:
1202 		LAGG_XLOCK(sc);
1203 		ro->ro_opts = sc->sc_opts;
1204 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1205 			struct lacp_softc *lsc;
1206 
1207 			lsc = (struct lacp_softc *)sc->sc_psc;
1208 			if (lsc->lsc_debug.lsc_tx_test != 0)
1209 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1210 			if (lsc->lsc_debug.lsc_rx_test != 0)
1211 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1212 			if (lsc->lsc_strict_mode != 0)
1213 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1214 			if (lsc->lsc_fast_timeout != 0)
1215 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1216 
1217 			ro->ro_active = sc->sc_active;
1218 		} else {
1219 			ro->ro_active = 0;
1220 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1221 				ro->ro_active += LAGG_PORTACTIVE(lp);
1222 		}
1223 		ro->ro_bkt = sc->sc_bkt;
1224 		ro->ro_flapping = sc->sc_flapping;
1225 		ro->ro_flowid_shift = sc->flowid_shift;
1226 		LAGG_XUNLOCK(sc);
1227 		break;
1228 	case SIOCSLAGGOPTS:
1229 		if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) {
1230 			if (ro->ro_bkt == 0)
1231 				sc->sc_bkt = 1; // Minimum 1 packet per iface.
1232 			else
1233 				sc->sc_bkt = ro->ro_bkt;
1234 		}
1235 		error = priv_check(td, PRIV_NET_LAGG);
1236 		if (error)
1237 			break;
1238 		if (ro->ro_opts == 0)
1239 			break;
1240 		/*
1241 		 * Set options.  LACP options are stored in sc->sc_psc,
1242 		 * not in sc_opts.
1243 		 */
1244 		int valid, lacp;
1245 
1246 		switch (ro->ro_opts) {
1247 		case LAGG_OPT_USE_FLOWID:
1248 		case -LAGG_OPT_USE_FLOWID:
1249 		case LAGG_OPT_FLOWIDSHIFT:
1250 			valid = 1;
1251 			lacp = 0;
1252 			break;
1253 		case LAGG_OPT_LACP_TXTEST:
1254 		case -LAGG_OPT_LACP_TXTEST:
1255 		case LAGG_OPT_LACP_RXTEST:
1256 		case -LAGG_OPT_LACP_RXTEST:
1257 		case LAGG_OPT_LACP_STRICT:
1258 		case -LAGG_OPT_LACP_STRICT:
1259 		case LAGG_OPT_LACP_TIMEOUT:
1260 		case -LAGG_OPT_LACP_TIMEOUT:
1261 			valid = lacp = 1;
1262 			break;
1263 		default:
1264 			valid = lacp = 0;
1265 			break;
1266 		}
1267 
1268 		LAGG_XLOCK(sc);
1269 
1270 		if (valid == 0 ||
1271 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1272 			/* Invalid combination of options specified. */
1273 			error = EINVAL;
1274 			LAGG_XUNLOCK(sc);
1275 			break;	/* Return from SIOCSLAGGOPTS. */
1276 		}
1277 		/*
1278 		 * Store new options into sc->sc_opts except for
1279 		 * FLOWIDSHIFT and LACP options.
1280 		 */
1281 		if (lacp == 0) {
1282 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1283 				sc->flowid_shift = ro->ro_flowid_shift;
1284 			else if (ro->ro_opts > 0)
1285 				sc->sc_opts |= ro->ro_opts;
1286 			else
1287 				sc->sc_opts &= ~ro->ro_opts;
1288 		} else {
1289 			struct lacp_softc *lsc;
1290 			struct lacp_port *lp;
1291 
1292 			lsc = (struct lacp_softc *)sc->sc_psc;
1293 
1294 			switch (ro->ro_opts) {
1295 			case LAGG_OPT_LACP_TXTEST:
1296 				lsc->lsc_debug.lsc_tx_test = 1;
1297 				break;
1298 			case -LAGG_OPT_LACP_TXTEST:
1299 				lsc->lsc_debug.lsc_tx_test = 0;
1300 				break;
1301 			case LAGG_OPT_LACP_RXTEST:
1302 				lsc->lsc_debug.lsc_rx_test = 1;
1303 				break;
1304 			case -LAGG_OPT_LACP_RXTEST:
1305 				lsc->lsc_debug.lsc_rx_test = 0;
1306 				break;
1307 			case LAGG_OPT_LACP_STRICT:
1308 				lsc->lsc_strict_mode = 1;
1309 				break;
1310 			case -LAGG_OPT_LACP_STRICT:
1311 				lsc->lsc_strict_mode = 0;
1312 				break;
1313 			case LAGG_OPT_LACP_TIMEOUT:
1314 				LACP_LOCK(lsc);
1315         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1316                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1317 				LACP_UNLOCK(lsc);
1318 				lsc->lsc_fast_timeout = 1;
1319 				break;
1320 			case -LAGG_OPT_LACP_TIMEOUT:
1321 				LACP_LOCK(lsc);
1322         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1323                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1324 				LACP_UNLOCK(lsc);
1325 				lsc->lsc_fast_timeout = 0;
1326 				break;
1327 			}
1328 		}
1329 		LAGG_XUNLOCK(sc);
1330 		break;
1331 	case SIOCGLAGGFLAGS:
1332 		rf->rf_flags = 0;
1333 		LAGG_XLOCK(sc);
1334 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1335 			rf->rf_flags |= LAGG_F_HASHL2;
1336 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1337 			rf->rf_flags |= LAGG_F_HASHL3;
1338 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1339 			rf->rf_flags |= LAGG_F_HASHL4;
1340 		LAGG_XUNLOCK(sc);
1341 		break;
1342 	case SIOCSLAGGHASH:
1343 		error = priv_check(td, PRIV_NET_LAGG);
1344 		if (error)
1345 			break;
1346 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1347 			error = EINVAL;
1348 			break;
1349 		}
1350 		LAGG_XLOCK(sc);
1351 		sc->sc_flags = 0;
1352 		if (rf->rf_flags & LAGG_F_HASHL2)
1353 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1354 		if (rf->rf_flags & LAGG_F_HASHL3)
1355 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1356 		if (rf->rf_flags & LAGG_F_HASHL4)
1357 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1358 		LAGG_XUNLOCK(sc);
1359 		break;
1360 	case SIOCGLAGGPORT:
1361 		if (rp->rp_portname[0] == '\0' ||
1362 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1363 			error = EINVAL;
1364 			break;
1365 		}
1366 
1367 		LAGG_RLOCK();
1368 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1369 		    lp->lp_softc != sc) {
1370 			error = ENOENT;
1371 			LAGG_RUNLOCK();
1372 			if_rele(tpif);
1373 			break;
1374 		}
1375 
1376 		lagg_port2req(lp, rp);
1377 		LAGG_RUNLOCK();
1378 		if_rele(tpif);
1379 		break;
1380 	case SIOCSLAGGPORT:
1381 		error = priv_check(td, PRIV_NET_LAGG);
1382 		if (error)
1383 			break;
1384 		if (rp->rp_portname[0] == '\0' ||
1385 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1386 			error = EINVAL;
1387 			break;
1388 		}
1389 #ifdef INET6
1390 		/*
1391 		 * A laggport interface should not have inet6 address
1392 		 * because two interfaces with a valid link-local
1393 		 * scope zone must not be merged in any form.  This
1394 		 * restriction is needed to prevent violation of
1395 		 * link-local scope zone.  Attempts to add a laggport
1396 		 * interface which has inet6 addresses triggers
1397 		 * removal of all inet6 addresses on the member
1398 		 * interface.
1399 		 */
1400 		if (in6ifa_llaonifp(tpif)) {
1401 			in6_ifdetach(tpif);
1402 				if_printf(sc->sc_ifp,
1403 				    "IPv6 addresses on %s have been removed "
1404 				    "before adding it as a member to prevent "
1405 				    "IPv6 address scope violation.\n",
1406 				    tpif->if_xname);
1407 		}
1408 #endif
1409 		LAGG_XLOCK(sc);
1410 		error = lagg_port_create(sc, tpif);
1411 		LAGG_XUNLOCK(sc);
1412 		if_rele(tpif);
1413 		VLAN_CAPABILITIES(ifp);
1414 		break;
1415 	case SIOCSLAGGDELPORT:
1416 		error = priv_check(td, PRIV_NET_LAGG);
1417 		if (error)
1418 			break;
1419 		if (rp->rp_portname[0] == '\0' ||
1420 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1421 			error = EINVAL;
1422 			break;
1423 		}
1424 
1425 		LAGG_XLOCK(sc);
1426 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1427 		    lp->lp_softc != sc) {
1428 			error = ENOENT;
1429 			LAGG_XUNLOCK(sc);
1430 			if_rele(tpif);
1431 			break;
1432 		}
1433 
1434 		error = lagg_port_destroy(lp, 1);
1435 		LAGG_XUNLOCK(sc);
1436 		if_rele(tpif);
1437 		VLAN_CAPABILITIES(ifp);
1438 		break;
1439 	case SIOCSIFFLAGS:
1440 		/* Set flags on ports too */
1441 		LAGG_XLOCK(sc);
1442 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1443 			lagg_setflags(lp, 1);
1444 		}
1445 
1446 		if (!(ifp->if_flags & IFF_UP) &&
1447 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1448 			/*
1449 			 * If interface is marked down and it is running,
1450 			 * then stop and disable it.
1451 			 */
1452 			lagg_stop(sc);
1453 			LAGG_XUNLOCK(sc);
1454 		} else if ((ifp->if_flags & IFF_UP) &&
1455 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1456 			/*
1457 			 * If interface is marked up and it is stopped, then
1458 			 * start it.
1459 			 */
1460 			LAGG_XUNLOCK(sc);
1461 			(*ifp->if_init)(sc);
1462 		} else
1463 			LAGG_XUNLOCK(sc);
1464 		break;
1465 	case SIOCADDMULTI:
1466 	case SIOCDELMULTI:
1467 		LAGG_XLOCK(sc);
1468 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1469 			lagg_clrmulti(lp);
1470 			lagg_setmulti(lp);
1471 		}
1472 		LAGG_XUNLOCK(sc);
1473 		error = 0;
1474 		break;
1475 	case SIOCSIFMEDIA:
1476 	case SIOCGIFMEDIA:
1477 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1478 		break;
1479 
1480 	case SIOCSIFCAP:
1481 		LAGG_XLOCK(sc);
1482 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1483 			if (lp->lp_ioctl != NULL)
1484 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1485 		}
1486 		lagg_capabilities(sc);
1487 		LAGG_XUNLOCK(sc);
1488 		VLAN_CAPABILITIES(ifp);
1489 		error = 0;
1490 		break;
1491 
1492 	case SIOCSIFMTU:
1493 		LAGG_XLOCK(sc);
1494 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1495 			if (lp->lp_ioctl != NULL)
1496 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1497 			else
1498 				error = EINVAL;
1499 			if (error != 0) {
1500 				if_printf(ifp,
1501 				    "failed to change MTU to %d on port %s, "
1502 				    "reverting all ports to original MTU (%d)\n",
1503 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1504 				break;
1505 			}
1506 		}
1507 		if (error == 0) {
1508 			ifp->if_mtu = ifr->ifr_mtu;
1509 		} else {
1510 			/* set every port back to the original MTU */
1511 			ifr->ifr_mtu = ifp->if_mtu;
1512 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1513 				if (lp->lp_ioctl != NULL)
1514 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1515 			}
1516 		}
1517 		LAGG_XUNLOCK(sc);
1518 		break;
1519 
1520 	default:
1521 		error = ether_ioctl(ifp, cmd, data);
1522 		break;
1523 	}
1524 	return (error);
1525 }
1526 
1527 #ifdef RATELIMIT
1528 static int
1529 lagg_snd_tag_alloc(struct ifnet *ifp,
1530     union if_snd_tag_alloc_params *params,
1531     struct m_snd_tag **ppmt)
1532 {
1533 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1534 	struct lagg_port *lp;
1535 	struct lagg_lb *lb;
1536 	uint32_t p;
1537 
1538 	switch (sc->sc_proto) {
1539 	case LAGG_PROTO_FAILOVER:
1540 		lp = lagg_link_active(sc, sc->sc_primary);
1541 		break;
1542 	case LAGG_PROTO_LOADBALANCE:
1543 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1544 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1545 			return (EOPNOTSUPP);
1546 		p = params->hdr.flowid >> sc->flowid_shift;
1547 		p %= sc->sc_count;
1548 		lb = (struct lagg_lb *)sc->sc_psc;
1549 		lp = lb->lb_ports[p];
1550 		lp = lagg_link_active(sc, lp);
1551 		break;
1552 	case LAGG_PROTO_LACP:
1553 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1554 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1555 			return (EOPNOTSUPP);
1556 		lp = lacp_select_tx_port_by_hash(sc, params->hdr.flowid);
1557 		break;
1558 	default:
1559 		return (EOPNOTSUPP);
1560 	}
1561 	if (lp == NULL)
1562 		return (EOPNOTSUPP);
1563 	ifp = lp->lp_ifp;
1564 	if (ifp == NULL || ifp->if_snd_tag_alloc == NULL ||
1565 	    (ifp->if_capenable & IFCAP_TXRTLMT) == 0)
1566 		return (EOPNOTSUPP);
1567 
1568 	/* forward allocation request */
1569 	return (ifp->if_snd_tag_alloc(ifp, params, ppmt));
1570 }
1571 #endif
1572 
1573 static int
1574 lagg_setmulti(struct lagg_port *lp)
1575 {
1576 	struct lagg_softc *sc = lp->lp_softc;
1577 	struct ifnet *ifp = lp->lp_ifp;
1578 	struct ifnet *scifp = sc->sc_ifp;
1579 	struct lagg_mc *mc;
1580 	struct ifmultiaddr *ifma;
1581 	int error;
1582 
1583 	IF_ADDR_WLOCK(scifp);
1584 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1585 		if (ifma->ifma_addr->sa_family != AF_LINK)
1586 			continue;
1587 		mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1588 		if (mc == NULL) {
1589 			IF_ADDR_WUNLOCK(scifp);
1590 			return (ENOMEM);
1591 		}
1592 		bcopy(ifma->ifma_addr, &mc->mc_addr,
1593 		    ifma->ifma_addr->sa_len);
1594 		mc->mc_addr.sdl_index = ifp->if_index;
1595 		mc->mc_ifma = NULL;
1596 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1597 	}
1598 	IF_ADDR_WUNLOCK(scifp);
1599 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1600 		error = if_addmulti(ifp,
1601 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1602 		if (error)
1603 			return (error);
1604 	}
1605 	return (0);
1606 }
1607 
1608 static int
1609 lagg_clrmulti(struct lagg_port *lp)
1610 {
1611 	struct lagg_mc *mc;
1612 
1613 	LAGG_XLOCK_ASSERT(lp->lp_softc);
1614 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1615 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1616 		if (mc->mc_ifma && lp->lp_detaching == 0)
1617 			if_delmulti_ifma(mc->mc_ifma);
1618 		free(mc, M_DEVBUF);
1619 	}
1620 	return (0);
1621 }
1622 
1623 static int
1624 lagg_setcaps(struct lagg_port *lp, int cap)
1625 {
1626 	struct ifreq ifr;
1627 
1628 	if (lp->lp_ifp->if_capenable == cap)
1629 		return (0);
1630 	if (lp->lp_ioctl == NULL)
1631 		return (ENXIO);
1632 	ifr.ifr_reqcap = cap;
1633 	return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1634 }
1635 
1636 /* Handle a ref counted flag that should be set on the lagg port as well */
1637 static int
1638 lagg_setflag(struct lagg_port *lp, int flag, int status,
1639     int (*func)(struct ifnet *, int))
1640 {
1641 	struct lagg_softc *sc = lp->lp_softc;
1642 	struct ifnet *scifp = sc->sc_ifp;
1643 	struct ifnet *ifp = lp->lp_ifp;
1644 	int error;
1645 
1646 	LAGG_XLOCK_ASSERT(sc);
1647 
1648 	status = status ? (scifp->if_flags & flag) : 0;
1649 	/* Now "status" contains the flag value or 0 */
1650 
1651 	/*
1652 	 * See if recorded ports status is different from what
1653 	 * we want it to be.  If it is, flip it.  We record ports
1654 	 * status in lp_ifflags so that we won't clear ports flag
1655 	 * we haven't set.  In fact, we don't clear or set ports
1656 	 * flags directly, but get or release references to them.
1657 	 * That's why we can be sure that recorded flags still are
1658 	 * in accord with actual ports flags.
1659 	 */
1660 	if (status != (lp->lp_ifflags & flag)) {
1661 		error = (*func)(ifp, status);
1662 		if (error)
1663 			return (error);
1664 		lp->lp_ifflags &= ~flag;
1665 		lp->lp_ifflags |= status;
1666 	}
1667 	return (0);
1668 }
1669 
1670 /*
1671  * Handle IFF_* flags that require certain changes on the lagg port
1672  * if "status" is true, update ports flags respective to the lagg
1673  * if "status" is false, forcedly clear the flags set on port.
1674  */
1675 static int
1676 lagg_setflags(struct lagg_port *lp, int status)
1677 {
1678 	int error, i;
1679 
1680 	for (i = 0; lagg_pflags[i].flag; i++) {
1681 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1682 		    status, lagg_pflags[i].func);
1683 		if (error)
1684 			return (error);
1685 	}
1686 	return (0);
1687 }
1688 
1689 static int
1690 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1691 {
1692 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1693 	int error;
1694 
1695 	LAGG_RLOCK();
1696 	/* We need a Tx algorithm and at least one port */
1697 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1698 		LAGG_RUNLOCK();
1699 		m_freem(m);
1700 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1701 		return (ENXIO);
1702 	}
1703 
1704 	ETHER_BPF_MTAP(ifp, m);
1705 
1706 	error = lagg_proto_start(sc, m);
1707 	LAGG_RUNLOCK();
1708 
1709 	if (error != 0)
1710 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1711 
1712 	return (error);
1713 }
1714 
1715 /*
1716  * The ifp->if_qflush entry point for lagg(4) is no-op.
1717  */
1718 static void
1719 lagg_qflush(struct ifnet *ifp __unused)
1720 {
1721 }
1722 
1723 static struct mbuf *
1724 lagg_input(struct ifnet *ifp, struct mbuf *m)
1725 {
1726 	struct lagg_port *lp = ifp->if_lagg;
1727 	struct lagg_softc *sc = lp->lp_softc;
1728 	struct ifnet *scifp = sc->sc_ifp;
1729 
1730 	LAGG_RLOCK();
1731 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1732 	    lp->lp_detaching != 0 ||
1733 	    sc->sc_proto == LAGG_PROTO_NONE) {
1734 		LAGG_RUNLOCK();
1735 		m_freem(m);
1736 		return (NULL);
1737 	}
1738 
1739 	ETHER_BPF_MTAP(scifp, m);
1740 
1741 	m = lagg_proto_input(sc, lp, m);
1742 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
1743 		m_freem(m);
1744 		m = NULL;
1745 	}
1746 
1747 	LAGG_RUNLOCK();
1748 	return (m);
1749 }
1750 
1751 static int
1752 lagg_media_change(struct ifnet *ifp)
1753 {
1754 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1755 
1756 	if (sc->sc_ifflags & IFF_DEBUG)
1757 		printf("%s\n", __func__);
1758 
1759 	/* Ignore */
1760 	return (0);
1761 }
1762 
1763 static void
1764 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1765 {
1766 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1767 	struct lagg_port *lp;
1768 
1769 	imr->ifm_status = IFM_AVALID;
1770 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1771 
1772 	LAGG_RLOCK();
1773 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1774 		if (LAGG_PORTACTIVE(lp))
1775 			imr->ifm_status |= IFM_ACTIVE;
1776 	}
1777 	LAGG_RUNLOCK();
1778 }
1779 
1780 static void
1781 lagg_linkstate(struct lagg_softc *sc)
1782 {
1783 	struct lagg_port *lp;
1784 	int new_link = LINK_STATE_DOWN;
1785 	uint64_t speed;
1786 
1787 	LAGG_XLOCK_ASSERT(sc);
1788 
1789 	/* LACP handles link state itself */
1790 	if (sc->sc_proto == LAGG_PROTO_LACP)
1791 		return;
1792 
1793 	/* Our link is considered up if at least one of our ports is active */
1794 	LAGG_RLOCK();
1795 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1796 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1797 			new_link = LINK_STATE_UP;
1798 			break;
1799 		}
1800 	}
1801 	LAGG_RUNLOCK();
1802 	if_link_state_change(sc->sc_ifp, new_link);
1803 
1804 	/* Update if_baudrate to reflect the max possible speed */
1805 	switch (sc->sc_proto) {
1806 		case LAGG_PROTO_FAILOVER:
1807 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1808 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1809 			break;
1810 		case LAGG_PROTO_ROUNDROBIN:
1811 		case LAGG_PROTO_LOADBALANCE:
1812 		case LAGG_PROTO_BROADCAST:
1813 			speed = 0;
1814 			LAGG_RLOCK();
1815 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1816 				speed += lp->lp_ifp->if_baudrate;
1817 			LAGG_RUNLOCK();
1818 			sc->sc_ifp->if_baudrate = speed;
1819 			break;
1820 		case LAGG_PROTO_LACP:
1821 			/* LACP updates if_baudrate itself */
1822 			break;
1823 	}
1824 }
1825 
1826 static void
1827 lagg_port_state(struct ifnet *ifp, int state)
1828 {
1829 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1830 	struct lagg_softc *sc = NULL;
1831 
1832 	if (lp != NULL)
1833 		sc = lp->lp_softc;
1834 	if (sc == NULL)
1835 		return;
1836 
1837 	LAGG_XLOCK(sc);
1838 	lagg_linkstate(sc);
1839 	lagg_proto_linkstate(sc, lp);
1840 	LAGG_XUNLOCK(sc);
1841 }
1842 
1843 struct lagg_port *
1844 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1845 {
1846 	struct lagg_port *lp_next, *rval = NULL;
1847 	struct epoch_tracker net_et;
1848 
1849 	/*
1850 	 * Search a port which reports an active link state.
1851 	 */
1852 
1853 	if (lp == NULL)
1854 		goto search;
1855 	if (LAGG_PORTACTIVE(lp)) {
1856 		rval = lp;
1857 		goto found;
1858 	}
1859 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
1860 	    LAGG_PORTACTIVE(lp_next)) {
1861 		rval = lp_next;
1862 		goto found;
1863 	}
1864 
1865  search:
1866 	epoch_enter_preempt(net_epoch_preempt, &net_et);
1867 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1868 		if (LAGG_PORTACTIVE(lp_next)) {
1869 			epoch_exit_preempt(net_epoch_preempt, &net_et);
1870 			return (lp_next);
1871 		}
1872 	}
1873 	epoch_exit_preempt(net_epoch_preempt, &net_et);
1874 found:
1875 	return (rval);
1876 }
1877 
1878 int
1879 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1880 {
1881 
1882 	return (ifp->if_transmit)(ifp, m);
1883 }
1884 
1885 /*
1886  * Simple round robin aggregation
1887  */
1888 static void
1889 lagg_rr_attach(struct lagg_softc *sc)
1890 {
1891 	sc->sc_seq = 0;
1892 	sc->sc_bkt_count = sc->sc_bkt;
1893 }
1894 
1895 static int
1896 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1897 {
1898 	struct lagg_port *lp;
1899 	uint32_t p;
1900 
1901 	if (sc->sc_bkt_count == 0 && sc->sc_bkt > 0)
1902 		sc->sc_bkt_count = sc->sc_bkt;
1903 
1904 	if (sc->sc_bkt > 0) {
1905 		atomic_subtract_int(&sc->sc_bkt_count, 1);
1906 	if (atomic_cmpset_int(&sc->sc_bkt_count, 0, sc->sc_bkt))
1907 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1908 	else
1909 		p = sc->sc_seq;
1910 	} else
1911 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1912 
1913 	p %= sc->sc_count;
1914 	lp = CK_SLIST_FIRST(&sc->sc_ports);
1915 
1916 	while (p--)
1917 		lp = CK_SLIST_NEXT(lp, lp_entries);
1918 
1919 	/*
1920 	 * Check the port's link state. This will return the next active
1921 	 * port if the link is down or the port is NULL.
1922 	 */
1923 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1924 		m_freem(m);
1925 		return (ENETDOWN);
1926 	}
1927 
1928 	/* Send mbuf */
1929 	return (lagg_enqueue(lp->lp_ifp, m));
1930 }
1931 
1932 static struct mbuf *
1933 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1934 {
1935 	struct ifnet *ifp = sc->sc_ifp;
1936 
1937 	/* Just pass in the packet to our lagg device */
1938 	m->m_pkthdr.rcvif = ifp;
1939 
1940 	return (m);
1941 }
1942 
1943 /*
1944  * Broadcast mode
1945  */
1946 static int
1947 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1948 {
1949 	int active_ports = 0;
1950 	int errors = 0;
1951 	int ret;
1952 	struct lagg_port *lp, *last = NULL;
1953 	struct mbuf *m0;
1954 
1955 	LAGG_RLOCK();
1956 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1957 		if (!LAGG_PORTACTIVE(lp))
1958 			continue;
1959 
1960 		active_ports++;
1961 
1962 		if (last != NULL) {
1963 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
1964 			if (m0 == NULL) {
1965 				ret = ENOBUFS;
1966 				errors++;
1967 				break;
1968 			}
1969 
1970 			ret = lagg_enqueue(last->lp_ifp, m0);
1971 			if (ret != 0)
1972 				errors++;
1973 		}
1974 		last = lp;
1975 	}
1976 	LAGG_RUNLOCK();
1977 
1978 	if (last == NULL) {
1979 		m_freem(m);
1980 		return (ENOENT);
1981 	}
1982 	if ((last = lagg_link_active(sc, last)) == NULL) {
1983 		m_freem(m);
1984 		return (ENETDOWN);
1985 	}
1986 
1987 	ret = lagg_enqueue(last->lp_ifp, m);
1988 	if (ret != 0)
1989 		errors++;
1990 
1991 	if (errors == 0)
1992 		return (ret);
1993 
1994 	return (0);
1995 }
1996 
1997 static struct mbuf*
1998 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1999 {
2000 	struct ifnet *ifp = sc->sc_ifp;
2001 
2002 	/* Just pass in the packet to our lagg device */
2003 	m->m_pkthdr.rcvif = ifp;
2004 	return (m);
2005 }
2006 
2007 /*
2008  * Active failover
2009  */
2010 static int
2011 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2012 {
2013 	struct lagg_port *lp;
2014 
2015 	/* Use the master port if active or the next available port */
2016 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2017 		m_freem(m);
2018 		return (ENETDOWN);
2019 	}
2020 
2021 	/* Send mbuf */
2022 	return (lagg_enqueue(lp->lp_ifp, m));
2023 }
2024 
2025 static struct mbuf *
2026 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2027 {
2028 	struct ifnet *ifp = sc->sc_ifp;
2029 	struct lagg_port *tmp_tp;
2030 
2031 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2032 		m->m_pkthdr.rcvif = ifp;
2033 		return (m);
2034 	}
2035 
2036 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2037 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2038 		/*
2039 		 * If tmp_tp is null, we've received a packet when all
2040 		 * our links are down. Weird, but process it anyways.
2041 		 */
2042 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2043 			m->m_pkthdr.rcvif = ifp;
2044 			return (m);
2045 		}
2046 	}
2047 
2048 	m_freem(m);
2049 	return (NULL);
2050 }
2051 
2052 /*
2053  * Loadbalancing
2054  */
2055 static void
2056 lagg_lb_attach(struct lagg_softc *sc)
2057 {
2058 	struct lagg_port *lp;
2059 	struct lagg_lb *lb;
2060 
2061 	LAGG_XLOCK_ASSERT(sc);
2062 	lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
2063 	lb->lb_key = m_ether_tcpip_hash_init();
2064 	sc->sc_psc = lb;
2065 
2066 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2067 		lagg_lb_port_create(lp);
2068 }
2069 
2070 static void
2071 lagg_lb_detach(struct lagg_softc *sc)
2072 {
2073 	struct lagg_lb *lb;
2074 
2075 	lb = (struct lagg_lb *)sc->sc_psc;
2076 	if (lb != NULL)
2077 		free(lb, M_DEVBUF);
2078 }
2079 
2080 static int
2081 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2082 {
2083 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2084 	struct lagg_port *lp_next;
2085 	int i = 0, rv;
2086 
2087 	rv = 0;
2088 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2089 	LAGG_RLOCK();
2090 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2091 		if (lp_next == lp)
2092 			continue;
2093 		if (i >= LAGG_MAX_PORTS) {
2094 			rv = EINVAL;
2095 			break;
2096 		}
2097 		if (sc->sc_ifflags & IFF_DEBUG)
2098 			printf("%s: port %s at index %d\n",
2099 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2100 		lb->lb_ports[i++] = lp_next;
2101 	}
2102 	LAGG_RUNLOCK();
2103 
2104 	return (rv);
2105 }
2106 
2107 static int
2108 lagg_lb_port_create(struct lagg_port *lp)
2109 {
2110 	struct lagg_softc *sc = lp->lp_softc;
2111 	return (lagg_lb_porttable(sc, NULL));
2112 }
2113 
2114 static void
2115 lagg_lb_port_destroy(struct lagg_port *lp)
2116 {
2117 	struct lagg_softc *sc = lp->lp_softc;
2118 	lagg_lb_porttable(sc, lp);
2119 }
2120 
2121 static int
2122 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2123 {
2124 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2125 	struct lagg_port *lp = NULL;
2126 	uint32_t p = 0;
2127 
2128 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2129 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2130 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2131 	else
2132 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2133 	p %= sc->sc_count;
2134 	lp = lb->lb_ports[p];
2135 
2136 	/*
2137 	 * Check the port's link state. This will return the next active
2138 	 * port if the link is down or the port is NULL.
2139 	 */
2140 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2141 		m_freem(m);
2142 		return (ENETDOWN);
2143 	}
2144 
2145 	/* Send mbuf */
2146 	return (lagg_enqueue(lp->lp_ifp, m));
2147 }
2148 
2149 static struct mbuf *
2150 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2151 {
2152 	struct ifnet *ifp = sc->sc_ifp;
2153 
2154 	/* Just pass in the packet to our lagg device */
2155 	m->m_pkthdr.rcvif = ifp;
2156 
2157 	return (m);
2158 }
2159 
2160 /*
2161  * 802.3ad LACP
2162  */
2163 static void
2164 lagg_lacp_attach(struct lagg_softc *sc)
2165 {
2166 	struct lagg_port *lp;
2167 
2168 	lacp_attach(sc);
2169 	LAGG_XLOCK_ASSERT(sc);
2170 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2171 		lacp_port_create(lp);
2172 }
2173 
2174 static void
2175 lagg_lacp_detach(struct lagg_softc *sc)
2176 {
2177 	struct lagg_port *lp;
2178 	void *psc;
2179 
2180 	LAGG_XLOCK_ASSERT(sc);
2181 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2182 		lacp_port_destroy(lp);
2183 
2184 	psc = sc->sc_psc;
2185 	sc->sc_psc = NULL;
2186 	lacp_detach(psc);
2187 }
2188 
2189 static void
2190 lagg_lacp_lladdr(struct lagg_softc *sc)
2191 {
2192 	struct lagg_port *lp;
2193 
2194 	LAGG_SXLOCK_ASSERT(sc);
2195 
2196 	/* purge all the lacp ports */
2197 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2198 		lacp_port_destroy(lp);
2199 
2200 	/* add them back in */
2201 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2202 		lacp_port_create(lp);
2203 }
2204 
2205 static int
2206 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2207 {
2208 	struct lagg_port *lp;
2209 
2210 	lp = lacp_select_tx_port(sc, m);
2211 	if (lp == NULL) {
2212 		m_freem(m);
2213 		return (ENETDOWN);
2214 	}
2215 
2216 	/* Send mbuf */
2217 	return (lagg_enqueue(lp->lp_ifp, m));
2218 }
2219 
2220 static struct mbuf *
2221 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2222 {
2223 	struct ifnet *ifp = sc->sc_ifp;
2224 	struct ether_header *eh;
2225 	u_short etype;
2226 
2227 	eh = mtod(m, struct ether_header *);
2228 	etype = ntohs(eh->ether_type);
2229 
2230 	/* Tap off LACP control messages */
2231 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2232 		m = lacp_input(lp, m);
2233 		if (m == NULL)
2234 			return (NULL);
2235 	}
2236 
2237 	/*
2238 	 * If the port is not collecting or not in the active aggregator then
2239 	 * free and return.
2240 	 */
2241 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2242 		m_freem(m);
2243 		return (NULL);
2244 	}
2245 
2246 	m->m_pkthdr.rcvif = ifp;
2247 	return (m);
2248 }
2249 
2250