xref: /freebsd/sys/net/if_lagg.c (revision a0409676120c1e558d0ade943019934e0f15118d)
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_kern_tls.h"
27 #include "opt_ratelimit.h"
28 
29 #include <sys/param.h>
30 #include <sys/kernel.h>
31 #include <sys/malloc.h>
32 #include <sys/mbuf.h>
33 #include <sys/queue.h>
34 #include <sys/socket.h>
35 #include <sys/sockio.h>
36 #include <sys/sysctl.h>
37 #include <sys/module.h>
38 #include <sys/priv.h>
39 #include <sys/systm.h>
40 #include <sys/proc.h>
41 #include <sys/lock.h>
42 #include <sys/rmlock.h>
43 #include <sys/sx.h>
44 #include <sys/taskqueue.h>
45 #include <sys/eventhandler.h>
46 
47 #include <net/ethernet.h>
48 #include <net/if.h>
49 #include <net/if_clone.h>
50 #include <net/if_arp.h>
51 #include <net/if_dl.h>
52 #include <net/if_media.h>
53 #include <net/if_types.h>
54 #include <net/if_var.h>
55 #include <net/bpf.h>
56 #include <net/route.h>
57 #include <net/vnet.h>
58 #include <net/infiniband.h>
59 
60 #if defined(INET) || defined(INET6)
61 #include <netinet/in.h>
62 #include <netinet/ip.h>
63 #endif
64 #ifdef INET
65 #include <netinet/in_systm.h>
66 #include <netinet/if_ether.h>
67 #endif
68 
69 #ifdef INET6
70 #include <netinet/ip6.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet6/in6_ifattach.h>
73 #endif
74 
75 #include <net/if_vlan_var.h>
76 #include <net/if_lagg.h>
77 #include <net/ieee8023ad_lacp.h>
78 
79 #ifdef INET6
80 /*
81  * XXX: declare here to avoid to include many inet6 related files..
82  * should be more generalized?
83  */
84 extern void	nd6_setmtu(struct ifnet *);
85 #endif
86 
87 #define	LAGG_SX_INIT(_sc)	sx_init(&(_sc)->sc_sx, "if_lagg sx")
88 #define	LAGG_SX_DESTROY(_sc)	sx_destroy(&(_sc)->sc_sx)
89 #define	LAGG_XLOCK(_sc)		sx_xlock(&(_sc)->sc_sx)
90 #define	LAGG_XUNLOCK(_sc)	sx_xunlock(&(_sc)->sc_sx)
91 #define	LAGG_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
92 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
93 
94 /* Special flags we should propagate to the lagg ports. */
95 static struct {
96 	int flag;
97 	int (*func)(struct ifnet *, int);
98 } lagg_pflags[] = {
99 	{IFF_PROMISC, ifpromisc},
100 	{IFF_ALLMULTI, if_allmulti},
101 	{0, NULL}
102 };
103 
104 struct lagg_snd_tag {
105 	struct m_snd_tag com;
106 	struct m_snd_tag *tag;
107 };
108 
109 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
110 #define	V_lagg_list	VNET(lagg_list)
111 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
112 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
113 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
114 					"if_lagg list", NULL, MTX_DEF)
115 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
116 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
117 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
118 eventhandler_tag	lagg_detach_cookie = NULL;
119 
120 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
121 static void	lagg_clone_destroy(struct ifnet *);
122 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
123 #define	V_lagg_cloner	VNET(lagg_cloner)
124 static const char laggname[] = "lagg";
125 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
126 
127 static void	lagg_capabilities(struct lagg_softc *);
128 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
129 static int	lagg_port_destroy(struct lagg_port *, int);
130 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *);
131 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *);
132 static void	lagg_linkstate(struct lagg_softc *);
133 static void	lagg_port_state(struct ifnet *, int);
134 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
135 static int	lagg_port_output(struct ifnet *, struct mbuf *,
136 		    const struct sockaddr *, struct route *);
137 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
138 #ifdef LAGG_PORT_STACKING
139 static int	lagg_port_checkstacking(struct lagg_softc *);
140 #endif
141 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
142 static void	lagg_init(void *);
143 static void	lagg_stop(struct lagg_softc *);
144 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
145 #if defined(KERN_TLS) || defined(RATELIMIT)
146 static int	lagg_snd_tag_alloc(struct ifnet *,
147 		    union if_snd_tag_alloc_params *,
148 		    struct m_snd_tag **);
149 static int	lagg_snd_tag_modify(struct m_snd_tag *,
150 		    union if_snd_tag_modify_params *);
151 static int	lagg_snd_tag_query(struct m_snd_tag *,
152 		    union if_snd_tag_query_params *);
153 static void	lagg_snd_tag_free(struct m_snd_tag *);
154 static struct m_snd_tag *lagg_next_snd_tag(struct m_snd_tag *);
155 static void     lagg_ratelimit_query(struct ifnet *,
156 		    struct if_ratelimit_query_results *);
157 #endif
158 static int	lagg_setmulti(struct lagg_port *);
159 static int	lagg_clrmulti(struct lagg_port *);
160 static	int	lagg_setcaps(struct lagg_port *, int cap);
161 static	int	lagg_setflag(struct lagg_port *, int, int,
162 		    int (*func)(struct ifnet *, int));
163 static	int	lagg_setflags(struct lagg_port *, int status);
164 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
165 static int	lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
166 static int	lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
167 static void	lagg_qflush(struct ifnet *);
168 static int	lagg_media_change(struct ifnet *);
169 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
170 static struct lagg_port *lagg_link_active(struct lagg_softc *,
171 	    struct lagg_port *);
172 
173 /* Simple round robin */
174 static void	lagg_rr_attach(struct lagg_softc *);
175 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
176 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
177 		    struct mbuf *);
178 
179 /* Active failover */
180 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
181 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
182 		    struct mbuf *);
183 
184 /* Loadbalancing */
185 static void	lagg_lb_attach(struct lagg_softc *);
186 static void	lagg_lb_detach(struct lagg_softc *);
187 static int	lagg_lb_port_create(struct lagg_port *);
188 static void	lagg_lb_port_destroy(struct lagg_port *);
189 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
190 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
191 		    struct mbuf *);
192 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
193 
194 /* Broadcast */
195 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
196 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
197 		    struct mbuf *);
198 
199 /* 802.3ad LACP */
200 static void	lagg_lacp_attach(struct lagg_softc *);
201 static void	lagg_lacp_detach(struct lagg_softc *);
202 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
203 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
204 		    struct mbuf *);
205 static void	lagg_lacp_lladdr(struct lagg_softc *);
206 
207 /* lagg protocol table */
208 static const struct lagg_proto {
209 	lagg_proto	pr_num;
210 	void		(*pr_attach)(struct lagg_softc *);
211 	void		(*pr_detach)(struct lagg_softc *);
212 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
213 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
214 			    struct mbuf *);
215 	int		(*pr_addport)(struct lagg_port *);
216 	void		(*pr_delport)(struct lagg_port *);
217 	void		(*pr_linkstate)(struct lagg_port *);
218 	void 		(*pr_init)(struct lagg_softc *);
219 	void 		(*pr_stop)(struct lagg_softc *);
220 	void 		(*pr_lladdr)(struct lagg_softc *);
221 	void		(*pr_request)(struct lagg_softc *, void *);
222 	void		(*pr_portreq)(struct lagg_port *, void *);
223 } lagg_protos[] = {
224     {
225 	.pr_num = LAGG_PROTO_NONE
226     },
227     {
228 	.pr_num = LAGG_PROTO_ROUNDROBIN,
229 	.pr_attach = lagg_rr_attach,
230 	.pr_start = lagg_rr_start,
231 	.pr_input = lagg_rr_input,
232     },
233     {
234 	.pr_num = LAGG_PROTO_FAILOVER,
235 	.pr_start = lagg_fail_start,
236 	.pr_input = lagg_fail_input,
237     },
238     {
239 	.pr_num = LAGG_PROTO_LOADBALANCE,
240 	.pr_attach = lagg_lb_attach,
241 	.pr_detach = lagg_lb_detach,
242 	.pr_start = lagg_lb_start,
243 	.pr_input = lagg_lb_input,
244 	.pr_addport = lagg_lb_port_create,
245 	.pr_delport = lagg_lb_port_destroy,
246     },
247     {
248 	.pr_num = LAGG_PROTO_LACP,
249 	.pr_attach = lagg_lacp_attach,
250 	.pr_detach = lagg_lacp_detach,
251 	.pr_start = lagg_lacp_start,
252 	.pr_input = lagg_lacp_input,
253 	.pr_addport = lacp_port_create,
254 	.pr_delport = lacp_port_destroy,
255 	.pr_linkstate = lacp_linkstate,
256 	.pr_init = lacp_init,
257 	.pr_stop = lacp_stop,
258 	.pr_lladdr = lagg_lacp_lladdr,
259 	.pr_request = lacp_req,
260 	.pr_portreq = lacp_portreq,
261     },
262     {
263 	.pr_num = LAGG_PROTO_BROADCAST,
264 	.pr_start = lagg_bcast_start,
265 	.pr_input = lagg_bcast_input,
266     },
267 };
268 
269 SYSCTL_DECL(_net_link);
270 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
271     "Link Aggregation");
272 
273 /* Allow input on any failover links */
274 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
275 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
276 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
277     &VNET_NAME(lagg_failover_rx_all), 0,
278     "Accept input from any interface in a failover lagg");
279 
280 /* Default value for using flowid */
281 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
282 #define	V_def_use_flowid	VNET(def_use_flowid)
283 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
284     &VNET_NAME(def_use_flowid), 0,
285     "Default setting for using flow id for load sharing");
286 
287 /* Default value for using numa */
288 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
289 #define	V_def_use_numa	VNET(def_use_numa)
290 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN,
291     &VNET_NAME(def_use_numa), 0,
292     "Use numa to steer flows");
293 
294 /* Default value for flowid shift */
295 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
296 #define	V_def_flowid_shift	VNET(def_flowid_shift)
297 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
298     &VNET_NAME(def_flowid_shift), 0,
299     "Default setting for flowid shift for load sharing");
300 
301 static void
302 vnet_lagg_init(const void *unused __unused)
303 {
304 
305 	LAGG_LIST_LOCK_INIT();
306 	SLIST_INIT(&V_lagg_list);
307 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
308 	    lagg_clone_destroy, 0);
309 }
310 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
311     vnet_lagg_init, NULL);
312 
313 static void
314 vnet_lagg_uninit(const void *unused __unused)
315 {
316 
317 	if_clone_detach(V_lagg_cloner);
318 	LAGG_LIST_LOCK_DESTROY();
319 }
320 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
321     vnet_lagg_uninit, NULL);
322 
323 static int
324 lagg_modevent(module_t mod, int type, void *data)
325 {
326 
327 	switch (type) {
328 	case MOD_LOAD:
329 		lagg_input_ethernet_p = lagg_input_ethernet;
330 		lagg_input_infiniband_p = lagg_input_infiniband;
331 		lagg_linkstate_p = lagg_port_state;
332 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
333 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
334 		    EVENTHANDLER_PRI_ANY);
335 		break;
336 	case MOD_UNLOAD:
337 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
338 		    lagg_detach_cookie);
339 		lagg_input_ethernet_p = NULL;
340 		lagg_input_infiniband_p = NULL;
341 		lagg_linkstate_p = NULL;
342 		break;
343 	default:
344 		return (EOPNOTSUPP);
345 	}
346 	return (0);
347 }
348 
349 static moduledata_t lagg_mod = {
350 	"if_lagg",
351 	lagg_modevent,
352 	0
353 };
354 
355 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
356 MODULE_VERSION(if_lagg, 1);
357 MODULE_DEPEND(if_lagg, if_infiniband, 1, 1, 1);
358 
359 static void
360 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
361 {
362 
363 	LAGG_XLOCK_ASSERT(sc);
364 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
365 	    __func__, sc));
366 
367 	if (sc->sc_ifflags & IFF_DEBUG)
368 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
369 
370 	if (lagg_protos[pr].pr_attach != NULL)
371 		lagg_protos[pr].pr_attach(sc);
372 	sc->sc_proto = pr;
373 }
374 
375 static void
376 lagg_proto_detach(struct lagg_softc *sc)
377 {
378 	lagg_proto pr;
379 
380 	LAGG_XLOCK_ASSERT(sc);
381 	pr = sc->sc_proto;
382 	sc->sc_proto = LAGG_PROTO_NONE;
383 
384 	if (lagg_protos[pr].pr_detach != NULL)
385 		lagg_protos[pr].pr_detach(sc);
386 }
387 
388 static int
389 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
390 {
391 
392 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
393 }
394 
395 static struct mbuf *
396 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
397 {
398 
399 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
400 }
401 
402 static int
403 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
404 {
405 
406 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
407 		return (0);
408 	else
409 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
410 }
411 
412 static void
413 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
414 {
415 
416 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
417 		lagg_protos[sc->sc_proto].pr_delport(lp);
418 }
419 
420 static void
421 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
422 {
423 
424 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
425 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
426 }
427 
428 static void
429 lagg_proto_init(struct lagg_softc *sc)
430 {
431 
432 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
433 		lagg_protos[sc->sc_proto].pr_init(sc);
434 }
435 
436 static void
437 lagg_proto_stop(struct lagg_softc *sc)
438 {
439 
440 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
441 		lagg_protos[sc->sc_proto].pr_stop(sc);
442 }
443 
444 static void
445 lagg_proto_lladdr(struct lagg_softc *sc)
446 {
447 
448 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
449 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
450 }
451 
452 static void
453 lagg_proto_request(struct lagg_softc *sc, void *v)
454 {
455 
456 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
457 		lagg_protos[sc->sc_proto].pr_request(sc, v);
458 }
459 
460 static void
461 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
462 {
463 
464 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
465 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
466 }
467 
468 /*
469  * This routine is run via an vlan
470  * config EVENT
471  */
472 static void
473 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
474 {
475 	struct lagg_softc *sc = ifp->if_softc;
476 	struct lagg_port *lp;
477 
478 	if (ifp->if_softc !=  arg)   /* Not our event */
479 		return;
480 
481 	LAGG_XLOCK(sc);
482 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
483 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
484 	LAGG_XUNLOCK(sc);
485 }
486 
487 /*
488  * This routine is run via an vlan
489  * unconfig EVENT
490  */
491 static void
492 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
493 {
494 	struct lagg_softc *sc = ifp->if_softc;
495 	struct lagg_port *lp;
496 
497 	if (ifp->if_softc !=  arg)   /* Not our event */
498 		return;
499 
500 	LAGG_XLOCK(sc);
501 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
502 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
503 	LAGG_XUNLOCK(sc);
504 }
505 
506 static int
507 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
508 {
509 	struct iflaggparam iflp;
510 	struct lagg_softc *sc;
511 	struct ifnet *ifp;
512 	int if_type;
513 	int error;
514 	static const uint8_t eaddr[LAGG_ADDR_LEN];
515 
516 	if (params != NULL) {
517 		error = copyin(params, &iflp, sizeof(iflp));
518 		if (error)
519 			return (error);
520 
521 		switch (iflp.lagg_type) {
522 		case LAGG_TYPE_ETHERNET:
523 			if_type = IFT_ETHER;
524 			break;
525 		case LAGG_TYPE_INFINIBAND:
526 			if_type = IFT_INFINIBAND;
527 			break;
528 		default:
529 			return (EINVAL);
530 		}
531 	} else {
532 		if_type = IFT_ETHER;
533 	}
534 
535 	sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO);
536 	ifp = sc->sc_ifp = if_alloc(if_type);
537 	if (ifp == NULL) {
538 		free(sc, M_LAGG);
539 		return (ENOSPC);
540 	}
541 	LAGG_SX_INIT(sc);
542 
543 	mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
544 	callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
545 
546 	LAGG_XLOCK(sc);
547 	if (V_def_use_flowid)
548 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
549 	if (V_def_use_numa)
550 		sc->sc_opts |= LAGG_OPT_USE_NUMA;
551 	sc->flowid_shift = V_def_flowid_shift;
552 
553 	/* Hash all layers by default */
554 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
555 
556 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
557 
558 	CK_SLIST_INIT(&sc->sc_ports);
559 
560 	switch (if_type) {
561 	case IFT_ETHER:
562 		/* Initialise pseudo media types */
563 		ifmedia_init(&sc->sc_media, 0, lagg_media_change,
564 		    lagg_media_status);
565 		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
566 		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
567 
568 		if_initname(ifp, laggname, unit);
569 		ifp->if_transmit = lagg_transmit_ethernet;
570 		break;
571 	case IFT_INFINIBAND:
572 		if_initname(ifp, laggname, unit);
573 		ifp->if_transmit = lagg_transmit_infiniband;
574 		break;
575 	default:
576 		break;
577 	}
578 	ifp->if_softc = sc;
579 	ifp->if_qflush = lagg_qflush;
580 	ifp->if_init = lagg_init;
581 	ifp->if_ioctl = lagg_ioctl;
582 	ifp->if_get_counter = lagg_get_counter;
583 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
584 #if defined(KERN_TLS) || defined(RATELIMIT)
585 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
586 	ifp->if_snd_tag_modify = lagg_snd_tag_modify;
587 	ifp->if_snd_tag_query = lagg_snd_tag_query;
588 	ifp->if_snd_tag_free = lagg_snd_tag_free;
589 	ifp->if_next_snd_tag = lagg_next_snd_tag;
590 	ifp->if_ratelimit_query = lagg_ratelimit_query;
591 #endif
592 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
593 
594 	/*
595 	 * Attach as an ordinary ethernet device, children will be attached
596 	 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
597 	 */
598 	switch (if_type) {
599 	case IFT_ETHER:
600 		ether_ifattach(ifp, eaddr);
601 		break;
602 	case IFT_INFINIBAND:
603 		infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr);
604 		break;
605 	default:
606 		break;
607 	}
608 
609 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
610 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
611 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
612 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
613 
614 	/* Insert into the global list of laggs */
615 	LAGG_LIST_LOCK();
616 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
617 	LAGG_LIST_UNLOCK();
618 	LAGG_XUNLOCK(sc);
619 
620 	return (0);
621 }
622 
623 static void
624 lagg_clone_destroy(struct ifnet *ifp)
625 {
626 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
627 	struct lagg_port *lp;
628 
629 	LAGG_XLOCK(sc);
630 	sc->sc_destroying = 1;
631 	lagg_stop(sc);
632 	ifp->if_flags &= ~IFF_UP;
633 
634 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
635 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
636 
637 	/* Shutdown and remove lagg ports */
638 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
639 		lagg_port_destroy(lp, 1);
640 
641 	/* Unhook the aggregation protocol */
642 	lagg_proto_detach(sc);
643 	LAGG_XUNLOCK(sc);
644 
645 	switch (ifp->if_type) {
646 	case IFT_ETHER:
647 		ifmedia_removeall(&sc->sc_media);
648 		ether_ifdetach(ifp);
649 		break;
650 	case IFT_INFINIBAND:
651 		infiniband_ifdetach(ifp);
652 		break;
653 	default:
654 		break;
655 	}
656 	if_free(ifp);
657 
658 	LAGG_LIST_LOCK();
659 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
660 	LAGG_LIST_UNLOCK();
661 
662 	mtx_destroy(&sc->sc_mtx);
663 	LAGG_SX_DESTROY(sc);
664 	free(sc, M_LAGG);
665 }
666 
667 static void
668 lagg_capabilities(struct lagg_softc *sc)
669 {
670 	struct lagg_port *lp;
671 	int cap, ena, pena;
672 	uint64_t hwa;
673 	struct ifnet_hw_tsomax hw_tsomax;
674 
675 	LAGG_XLOCK_ASSERT(sc);
676 
677 	/* Get common enabled capabilities for the lagg ports */
678 	ena = ~0;
679 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
680 		ena &= lp->lp_ifp->if_capenable;
681 	ena = (ena == ~0 ? 0 : ena);
682 
683 	/*
684 	 * Apply common enabled capabilities back to the lagg ports.
685 	 * May require several iterations if they are dependent.
686 	 */
687 	do {
688 		pena = ena;
689 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
690 			lagg_setcaps(lp, ena);
691 			ena &= lp->lp_ifp->if_capenable;
692 		}
693 	} while (pena != ena);
694 
695 	/* Get other capabilities from the lagg ports */
696 	cap = ~0;
697 	hwa = ~(uint64_t)0;
698 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
699 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
700 		cap &= lp->lp_ifp->if_capabilities;
701 		hwa &= lp->lp_ifp->if_hwassist;
702 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
703 	}
704 	cap = (cap == ~0 ? 0 : cap);
705 	hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
706 
707 	if (sc->sc_ifp->if_capabilities != cap ||
708 	    sc->sc_ifp->if_capenable != ena ||
709 	    sc->sc_ifp->if_hwassist != hwa ||
710 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
711 		sc->sc_ifp->if_capabilities = cap;
712 		sc->sc_ifp->if_capenable = ena;
713 		sc->sc_ifp->if_hwassist = hwa;
714 		getmicrotime(&sc->sc_ifp->if_lastchange);
715 
716 		if (sc->sc_ifflags & IFF_DEBUG)
717 			if_printf(sc->sc_ifp,
718 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
719 	}
720 }
721 
722 static int
723 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
724 {
725 	struct lagg_softc *sc_ptr;
726 	struct lagg_port *lp, *tlp;
727 	struct ifreq ifr;
728 	int error, i, oldmtu;
729 	int if_type;
730 	uint64_t *pval;
731 
732 	LAGG_XLOCK_ASSERT(sc);
733 
734 	if (sc->sc_ifp == ifp) {
735 		if_printf(sc->sc_ifp,
736 		    "cannot add a lagg to itself as a port\n");
737 		return (EINVAL);
738 	}
739 
740 	if (sc->sc_destroying == 1)
741 		return (ENXIO);
742 
743 	/* Limit the maximal number of lagg ports */
744 	if (sc->sc_count >= LAGG_MAX_PORTS)
745 		return (ENOSPC);
746 
747 	/* Check if port has already been associated to a lagg */
748 	if (ifp->if_lagg != NULL) {
749 		/* Port is already in the current lagg? */
750 		lp = (struct lagg_port *)ifp->if_lagg;
751 		if (lp->lp_softc == sc)
752 			return (EEXIST);
753 		return (EBUSY);
754 	}
755 
756 	switch (sc->sc_ifp->if_type) {
757 	case IFT_ETHER:
758 		/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
759 		if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
760 			return (EPROTONOSUPPORT);
761 		if_type = IFT_IEEE8023ADLAG;
762 		break;
763 	case IFT_INFINIBAND:
764 		/* XXX Disallow non-infiniband interfaces */
765 		if (ifp->if_type != IFT_INFINIBAND)
766 			return (EPROTONOSUPPORT);
767 		if_type = IFT_INFINIBANDLAG;
768 		break;
769 	default:
770 		break;
771 	}
772 
773 	/* Allow the first Ethernet member to define the MTU */
774 	oldmtu = -1;
775 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
776 		sc->sc_ifp->if_mtu = ifp->if_mtu;
777 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
778 		if (ifp->if_ioctl == NULL) {
779 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
780 			    ifp->if_xname);
781 			return (EINVAL);
782 		}
783 		oldmtu = ifp->if_mtu;
784 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
785 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
786 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
787 		if (error != 0) {
788 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
789 			    ifp->if_xname);
790 			return (error);
791 		}
792 		ifr.ifr_mtu = oldmtu;
793 	}
794 
795 	lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO);
796 	lp->lp_softc = sc;
797 
798 	/* Check if port is a stacked lagg */
799 	LAGG_LIST_LOCK();
800 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
801 		if (ifp == sc_ptr->sc_ifp) {
802 			LAGG_LIST_UNLOCK();
803 			free(lp, M_LAGG);
804 			if (oldmtu != -1)
805 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
806 				    (caddr_t)&ifr);
807 			return (EINVAL);
808 			/* XXX disable stacking for the moment, its untested */
809 #ifdef LAGG_PORT_STACKING
810 			lp->lp_flags |= LAGG_PORT_STACK;
811 			if (lagg_port_checkstacking(sc_ptr) >=
812 			    LAGG_MAX_STACKING) {
813 				LAGG_LIST_UNLOCK();
814 				free(lp, M_LAGG);
815 				if (oldmtu != -1)
816 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
817 					    (caddr_t)&ifr);
818 				return (E2BIG);
819 			}
820 #endif
821 		}
822 	}
823 	LAGG_LIST_UNLOCK();
824 
825 	if_ref(ifp);
826 	lp->lp_ifp = ifp;
827 
828 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
829 	lp->lp_ifcapenable = ifp->if_capenable;
830 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
831 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
832 		lagg_proto_lladdr(sc);
833 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
834 	} else {
835 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
836 	}
837 	lagg_setflags(lp, 1);
838 
839 	if (CK_SLIST_EMPTY(&sc->sc_ports))
840 		sc->sc_primary = lp;
841 
842 	/* Change the interface type */
843 	lp->lp_iftype = ifp->if_type;
844 	ifp->if_type = if_type;
845 	ifp->if_lagg = lp;
846 	lp->lp_ioctl = ifp->if_ioctl;
847 	ifp->if_ioctl = lagg_port_ioctl;
848 	lp->lp_output = ifp->if_output;
849 	ifp->if_output = lagg_port_output;
850 
851 	/* Read port counters */
852 	pval = lp->port_counters.val;
853 	for (i = 0; i < IFCOUNTERS; i++, pval++)
854 		*pval = ifp->if_get_counter(ifp, i);
855 
856 	/*
857 	 * Insert into the list of ports.
858 	 * Keep ports sorted by if_index. It is handy, when configuration
859 	 * is predictable and `ifconfig laggN create ...` command
860 	 * will lead to the same result each time.
861 	 */
862 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
863 		if (tlp->lp_ifp->if_index < ifp->if_index && (
864 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
865 		    ((struct  lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
866 		    ifp->if_index))
867 			break;
868 	}
869 	if (tlp != NULL)
870 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
871 	else
872 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
873 	sc->sc_count++;
874 
875 	lagg_setmulti(lp);
876 
877 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
878 		/* Remove the port, without calling pr_delport. */
879 		lagg_port_destroy(lp, 0);
880 		if (oldmtu != -1)
881 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
882 		return (error);
883 	}
884 
885 	/* Update lagg capabilities */
886 	lagg_capabilities(sc);
887 	lagg_linkstate(sc);
888 
889 	return (0);
890 }
891 
892 #ifdef LAGG_PORT_STACKING
893 static int
894 lagg_port_checkstacking(struct lagg_softc *sc)
895 {
896 	struct lagg_softc *sc_ptr;
897 	struct lagg_port *lp;
898 	int m = 0;
899 
900 	LAGG_SXLOCK_ASSERT(sc);
901 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
902 		if (lp->lp_flags & LAGG_PORT_STACK) {
903 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
904 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
905 		}
906 	}
907 
908 	return (m + 1);
909 }
910 #endif
911 
912 static void
913 lagg_port_destroy_cb(epoch_context_t ec)
914 {
915 	struct lagg_port *lp;
916 	struct ifnet *ifp;
917 
918 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
919 	ifp = lp->lp_ifp;
920 
921 	if_rele(ifp);
922 	free(lp, M_LAGG);
923 }
924 
925 static int
926 lagg_port_destroy(struct lagg_port *lp, int rundelport)
927 {
928 	struct lagg_softc *sc = lp->lp_softc;
929 	struct lagg_port *lp_ptr, *lp0;
930 	struct ifnet *ifp = lp->lp_ifp;
931 	uint64_t *pval, vdiff;
932 	int i;
933 
934 	LAGG_XLOCK_ASSERT(sc);
935 
936 	if (rundelport)
937 		lagg_proto_delport(sc, lp);
938 
939 	if (lp->lp_detaching == 0)
940 		lagg_clrmulti(lp);
941 
942 	/* Restore interface */
943 	ifp->if_type = lp->lp_iftype;
944 	ifp->if_ioctl = lp->lp_ioctl;
945 	ifp->if_output = lp->lp_output;
946 	ifp->if_lagg = NULL;
947 
948 	/* Update detached port counters */
949 	pval = lp->port_counters.val;
950 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
951 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
952 		sc->detached_counters.val[i] += vdiff;
953 	}
954 
955 	/* Finally, remove the port from the lagg */
956 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
957 	sc->sc_count--;
958 
959 	/* Update the primary interface */
960 	if (lp == sc->sc_primary) {
961 		uint8_t lladdr[LAGG_ADDR_LEN];
962 
963 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
964 			bzero(&lladdr, LAGG_ADDR_LEN);
965 		else
966 			bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
967 		sc->sc_primary = lp0;
968 		if (sc->sc_destroying == 0) {
969 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
970 			lagg_proto_lladdr(sc);
971 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
972 		}
973 
974 		/*
975 		 * Update lladdr for each port (new primary needs update
976 		 * as well, to switch from old lladdr to its 'real' one)
977 		 */
978 		CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
979 			if_setlladdr(lp_ptr->lp_ifp, lladdr, lp_ptr->lp_ifp->if_addrlen);
980 	}
981 
982 	if (lp->lp_ifflags)
983 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
984 
985 	if (lp->lp_detaching == 0) {
986 		lagg_setflags(lp, 0);
987 		lagg_setcaps(lp, lp->lp_ifcapenable);
988 		if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
989 	}
990 
991 	/*
992 	 * free port and release it's ifnet reference after a grace period has
993 	 * elapsed.
994 	 */
995 	NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
996 	/* Update lagg capabilities */
997 	lagg_capabilities(sc);
998 	lagg_linkstate(sc);
999 
1000 	return (0);
1001 }
1002 
1003 static int
1004 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1005 {
1006 	struct epoch_tracker et;
1007 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
1008 	struct lagg_softc *sc;
1009 	struct lagg_port *lp = NULL;
1010 	int error = 0;
1011 
1012 	/* Should be checked by the caller */
1013 	switch (ifp->if_type) {
1014 	case IFT_IEEE8023ADLAG:
1015 	case IFT_INFINIBANDLAG:
1016 		if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1017 			goto fallback;
1018 		break;
1019 	default:
1020 		goto fallback;
1021 	}
1022 
1023 	switch (cmd) {
1024 	case SIOCGLAGGPORT:
1025 		if (rp->rp_portname[0] == '\0' ||
1026 		    ifunit(rp->rp_portname) != ifp) {
1027 			error = EINVAL;
1028 			break;
1029 		}
1030 
1031 		NET_EPOCH_ENTER(et);
1032 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1033 			error = ENOENT;
1034 			NET_EPOCH_EXIT(et);
1035 			break;
1036 		}
1037 
1038 		lagg_port2req(lp, rp);
1039 		NET_EPOCH_EXIT(et);
1040 		break;
1041 
1042 	case SIOCSIFCAP:
1043 		if (lp->lp_ioctl == NULL) {
1044 			error = EINVAL;
1045 			break;
1046 		}
1047 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1048 		if (error)
1049 			break;
1050 
1051 		/* Update lagg interface capabilities */
1052 		LAGG_XLOCK(sc);
1053 		lagg_capabilities(sc);
1054 		LAGG_XUNLOCK(sc);
1055 		VLAN_CAPABILITIES(sc->sc_ifp);
1056 		break;
1057 
1058 	case SIOCSIFMTU:
1059 		/* Do not allow the MTU to be changed once joined */
1060 		error = EINVAL;
1061 		break;
1062 
1063 	default:
1064 		goto fallback;
1065 	}
1066 
1067 	return (error);
1068 
1069 fallback:
1070 	if (lp != NULL && lp->lp_ioctl != NULL)
1071 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1072 
1073 	return (EINVAL);
1074 }
1075 
1076 /*
1077  * Requests counter @cnt data.
1078  *
1079  * Counter value is calculated the following way:
1080  * 1) for each port, sum  difference between current and "initial" measurements.
1081  * 2) add lagg logical interface counters.
1082  * 3) add data from detached_counters array.
1083  *
1084  * We also do the following things on ports attach/detach:
1085  * 1) On port attach we store all counters it has into port_counter array.
1086  * 2) On port detach we add the different between "initial" and
1087  *   current counters data to detached_counters array.
1088  */
1089 static uint64_t
1090 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1091 {
1092 	struct epoch_tracker et;
1093 	struct lagg_softc *sc;
1094 	struct lagg_port *lp;
1095 	struct ifnet *lpifp;
1096 	uint64_t newval, oldval, vsum;
1097 
1098 	/* Revise this when we've got non-generic counters. */
1099 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1100 
1101 	sc = (struct lagg_softc *)ifp->if_softc;
1102 
1103 	vsum = 0;
1104 	NET_EPOCH_ENTER(et);
1105 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1106 		/* Saved attached value */
1107 		oldval = lp->port_counters.val[cnt];
1108 		/* current value */
1109 		lpifp = lp->lp_ifp;
1110 		newval = lpifp->if_get_counter(lpifp, cnt);
1111 		/* Calculate diff and save new */
1112 		vsum += newval - oldval;
1113 	}
1114 	NET_EPOCH_EXIT(et);
1115 
1116 	/*
1117 	 * Add counter data which might be added by upper
1118 	 * layer protocols operating on logical interface.
1119 	 */
1120 	vsum += if_get_counter_default(ifp, cnt);
1121 
1122 	/*
1123 	 * Add counter data from detached ports counters
1124 	 */
1125 	vsum += sc->detached_counters.val[cnt];
1126 
1127 	return (vsum);
1128 }
1129 
1130 /*
1131  * For direct output to child ports.
1132  */
1133 static int
1134 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1135 	const struct sockaddr *dst, struct route *ro)
1136 {
1137 	struct lagg_port *lp = ifp->if_lagg;
1138 
1139 	switch (dst->sa_family) {
1140 		case pseudo_AF_HDRCMPLT:
1141 		case AF_UNSPEC:
1142 			if (lp != NULL)
1143 				return ((*lp->lp_output)(ifp, m, dst, ro));
1144 	}
1145 
1146 	/* drop any other frames */
1147 	m_freem(m);
1148 	return (ENETDOWN);
1149 }
1150 
1151 static void
1152 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1153 {
1154 	struct lagg_port *lp;
1155 	struct lagg_softc *sc;
1156 
1157 	if ((lp = ifp->if_lagg) == NULL)
1158 		return;
1159 	/* If the ifnet is just being renamed, don't do anything. */
1160 	if (ifp->if_flags & IFF_RENAMING)
1161 		return;
1162 
1163 	sc = lp->lp_softc;
1164 
1165 	LAGG_XLOCK(sc);
1166 	lp->lp_detaching = 1;
1167 	lagg_port_destroy(lp, 1);
1168 	LAGG_XUNLOCK(sc);
1169 	VLAN_CAPABILITIES(sc->sc_ifp);
1170 }
1171 
1172 static void
1173 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1174 {
1175 	struct lagg_softc *sc = lp->lp_softc;
1176 
1177 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1178 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1179 	rp->rp_prio = lp->lp_prio;
1180 	rp->rp_flags = lp->lp_flags;
1181 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1182 
1183 	/* Add protocol specific flags */
1184 	switch (sc->sc_proto) {
1185 		case LAGG_PROTO_FAILOVER:
1186 			if (lp == sc->sc_primary)
1187 				rp->rp_flags |= LAGG_PORT_MASTER;
1188 			if (lp == lagg_link_active(sc, sc->sc_primary))
1189 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1190 			break;
1191 
1192 		case LAGG_PROTO_ROUNDROBIN:
1193 		case LAGG_PROTO_LOADBALANCE:
1194 		case LAGG_PROTO_BROADCAST:
1195 			if (LAGG_PORTACTIVE(lp))
1196 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1197 			break;
1198 
1199 		case LAGG_PROTO_LACP:
1200 			/* LACP has a different definition of active */
1201 			if (lacp_isactive(lp))
1202 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1203 			if (lacp_iscollecting(lp))
1204 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1205 			if (lacp_isdistributing(lp))
1206 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1207 			break;
1208 	}
1209 
1210 }
1211 
1212 static void
1213 lagg_watchdog_infiniband(void *arg)
1214 {
1215 	struct epoch_tracker et;
1216 	struct lagg_softc *sc;
1217 	struct lagg_port *lp;
1218 	struct ifnet *ifp;
1219 	struct ifnet *lp_ifp;
1220 
1221 	sc = arg;
1222 
1223 	/*
1224 	 * Because infiniband nodes have a fixed MAC address, which is
1225 	 * generated by the so-called GID, we need to regularly update
1226 	 * the link level address of the parent lagg<N> device when
1227 	 * the active port changes. Possibly we could piggy-back on
1228 	 * link up/down events aswell, but using a timer also provides
1229 	 * a guarantee against too frequent events. This operation
1230 	 * does not have to be atomic.
1231 	 */
1232 	NET_EPOCH_ENTER(et);
1233 	lp = lagg_link_active(sc, sc->sc_primary);
1234 	if (lp != NULL) {
1235 		ifp = sc->sc_ifp;
1236 		lp_ifp = lp->lp_ifp;
1237 
1238 		if (ifp != NULL && lp_ifp != NULL &&
1239 		    (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0 ||
1240 		     memcmp(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen) != 0)) {
1241 			memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1242 			memcpy(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen);
1243 
1244 			CURVNET_SET(ifp->if_vnet);
1245 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1246 			CURVNET_RESTORE();
1247 		}
1248 	}
1249 	NET_EPOCH_EXIT(et);
1250 
1251 	callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1252 }
1253 
1254 static void
1255 lagg_init(void *xsc)
1256 {
1257 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1258 	struct ifnet *ifp = sc->sc_ifp;
1259 	struct lagg_port *lp;
1260 
1261 	LAGG_XLOCK(sc);
1262 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1263 		LAGG_XUNLOCK(sc);
1264 		return;
1265 	}
1266 
1267 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1268 
1269 	/*
1270 	 * Update the port lladdrs if needed.
1271 	 * This might be if_setlladdr() notification
1272 	 * that lladdr has been changed.
1273 	 */
1274 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1275 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1276 		    ifp->if_addrlen) != 0)
1277 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1278 	}
1279 
1280 	lagg_proto_init(sc);
1281 
1282 	if (ifp->if_type == IFT_INFINIBAND) {
1283 		mtx_lock(&sc->sc_mtx);
1284 		lagg_watchdog_infiniband(sc);
1285 		mtx_unlock(&sc->sc_mtx);
1286 	}
1287 
1288 	LAGG_XUNLOCK(sc);
1289 }
1290 
1291 static void
1292 lagg_stop(struct lagg_softc *sc)
1293 {
1294 	struct ifnet *ifp = sc->sc_ifp;
1295 
1296 	LAGG_XLOCK_ASSERT(sc);
1297 
1298 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1299 		return;
1300 
1301 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1302 
1303 	lagg_proto_stop(sc);
1304 
1305 	mtx_lock(&sc->sc_mtx);
1306 	callout_stop(&sc->sc_watchdog);
1307 	mtx_unlock(&sc->sc_mtx);
1308 
1309 	callout_drain(&sc->sc_watchdog);
1310 }
1311 
1312 static int
1313 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1314 {
1315 	struct epoch_tracker et;
1316 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1317 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1318 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1319 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1320 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1321 	struct ifreq *ifr = (struct ifreq *)data;
1322 	struct lagg_port *lp;
1323 	struct ifnet *tpif;
1324 	struct thread *td = curthread;
1325 	char *buf, *outbuf;
1326 	int count, buflen, len, error = 0, oldmtu;
1327 
1328 	bzero(&rpbuf, sizeof(rpbuf));
1329 
1330 	/* XXX: This can race with lagg_clone_destroy. */
1331 
1332 	switch (cmd) {
1333 	case SIOCGLAGG:
1334 		LAGG_XLOCK(sc);
1335 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1336 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1337 		ra->ra_proto = sc->sc_proto;
1338 		lagg_proto_request(sc, &ra->ra_psc);
1339 		count = 0;
1340 		buf = outbuf;
1341 		len = min(ra->ra_size, buflen);
1342 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1343 			if (len < sizeof(rpbuf))
1344 				break;
1345 
1346 			lagg_port2req(lp, &rpbuf);
1347 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1348 			count++;
1349 			buf += sizeof(rpbuf);
1350 			len -= sizeof(rpbuf);
1351 		}
1352 		LAGG_XUNLOCK(sc);
1353 		ra->ra_ports = count;
1354 		ra->ra_size = count * sizeof(rpbuf);
1355 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1356 		free(outbuf, M_TEMP);
1357 		break;
1358 	case SIOCSLAGG:
1359 		error = priv_check(td, PRIV_NET_LAGG);
1360 		if (error)
1361 			break;
1362 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1363 			error = EPROTONOSUPPORT;
1364 			break;
1365 		}
1366 		/* Infiniband only supports the failover protocol. */
1367 		if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1368 		    ifp->if_type == IFT_INFINIBAND) {
1369 			error = EPROTONOSUPPORT;
1370 			break;
1371 		}
1372 		LAGG_XLOCK(sc);
1373 		lagg_proto_detach(sc);
1374 		lagg_proto_attach(sc, ra->ra_proto);
1375 		LAGG_XUNLOCK(sc);
1376 		break;
1377 	case SIOCGLAGGOPTS:
1378 		LAGG_XLOCK(sc);
1379 		ro->ro_opts = sc->sc_opts;
1380 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1381 			struct lacp_softc *lsc;
1382 
1383 			lsc = (struct lacp_softc *)sc->sc_psc;
1384 			if (lsc->lsc_debug.lsc_tx_test != 0)
1385 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1386 			if (lsc->lsc_debug.lsc_rx_test != 0)
1387 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1388 			if (lsc->lsc_strict_mode != 0)
1389 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1390 			if (lsc->lsc_fast_timeout != 0)
1391 				ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1392 
1393 			ro->ro_active = sc->sc_active;
1394 		} else {
1395 			ro->ro_active = 0;
1396 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1397 				ro->ro_active += LAGG_PORTACTIVE(lp);
1398 		}
1399 		ro->ro_bkt = sc->sc_stride;
1400 		ro->ro_flapping = sc->sc_flapping;
1401 		ro->ro_flowid_shift = sc->flowid_shift;
1402 		LAGG_XUNLOCK(sc);
1403 		break;
1404 	case SIOCSLAGGOPTS:
1405 		error = priv_check(td, PRIV_NET_LAGG);
1406 		if (error)
1407 			break;
1408 
1409 		/*
1410 		 * The stride option was added without defining a corresponding
1411 		 * LAGG_OPT flag, so handle a non-zero value before checking
1412 		 * anything else to preserve compatibility.
1413 		 */
1414 		LAGG_XLOCK(sc);
1415 		if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1416 			if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1417 				LAGG_XUNLOCK(sc);
1418 				error = EINVAL;
1419 				break;
1420 			}
1421 			sc->sc_stride = ro->ro_bkt;
1422 		}
1423 		if (ro->ro_opts == 0) {
1424 			LAGG_XUNLOCK(sc);
1425 			break;
1426 		}
1427 
1428 		/*
1429 		 * Set options.  LACP options are stored in sc->sc_psc,
1430 		 * not in sc_opts.
1431 		 */
1432 		int valid, lacp;
1433 
1434 		switch (ro->ro_opts) {
1435 		case LAGG_OPT_USE_FLOWID:
1436 		case -LAGG_OPT_USE_FLOWID:
1437 		case LAGG_OPT_USE_NUMA:
1438 		case -LAGG_OPT_USE_NUMA:
1439 		case LAGG_OPT_FLOWIDSHIFT:
1440 		case LAGG_OPT_RR_LIMIT:
1441 			valid = 1;
1442 			lacp = 0;
1443 			break;
1444 		case LAGG_OPT_LACP_TXTEST:
1445 		case -LAGG_OPT_LACP_TXTEST:
1446 		case LAGG_OPT_LACP_RXTEST:
1447 		case -LAGG_OPT_LACP_RXTEST:
1448 		case LAGG_OPT_LACP_STRICT:
1449 		case -LAGG_OPT_LACP_STRICT:
1450 		case LAGG_OPT_LACP_FAST_TIMO:
1451 		case -LAGG_OPT_LACP_FAST_TIMO:
1452 			valid = lacp = 1;
1453 			break;
1454 		default:
1455 			valid = lacp = 0;
1456 			break;
1457 		}
1458 
1459 		if (valid == 0 ||
1460 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1461 			/* Invalid combination of options specified. */
1462 			error = EINVAL;
1463 			LAGG_XUNLOCK(sc);
1464 			break;	/* Return from SIOCSLAGGOPTS. */
1465 		}
1466 
1467 		/*
1468 		 * Store new options into sc->sc_opts except for
1469 		 * FLOWIDSHIFT, RR and LACP options.
1470 		 */
1471 		if (lacp == 0) {
1472 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1473 				sc->flowid_shift = ro->ro_flowid_shift;
1474 			else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1475 				if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1476 				    ro->ro_bkt == 0) {
1477 					error = EINVAL;
1478 					LAGG_XUNLOCK(sc);
1479 					break;
1480 				}
1481 				sc->sc_stride = ro->ro_bkt;
1482 			} else if (ro->ro_opts > 0)
1483 				sc->sc_opts |= ro->ro_opts;
1484 			else
1485 				sc->sc_opts &= ~ro->ro_opts;
1486 		} else {
1487 			struct lacp_softc *lsc;
1488 			struct lacp_port *lp;
1489 
1490 			lsc = (struct lacp_softc *)sc->sc_psc;
1491 
1492 			switch (ro->ro_opts) {
1493 			case LAGG_OPT_LACP_TXTEST:
1494 				lsc->lsc_debug.lsc_tx_test = 1;
1495 				break;
1496 			case -LAGG_OPT_LACP_TXTEST:
1497 				lsc->lsc_debug.lsc_tx_test = 0;
1498 				break;
1499 			case LAGG_OPT_LACP_RXTEST:
1500 				lsc->lsc_debug.lsc_rx_test = 1;
1501 				break;
1502 			case -LAGG_OPT_LACP_RXTEST:
1503 				lsc->lsc_debug.lsc_rx_test = 0;
1504 				break;
1505 			case LAGG_OPT_LACP_STRICT:
1506 				lsc->lsc_strict_mode = 1;
1507 				break;
1508 			case -LAGG_OPT_LACP_STRICT:
1509 				lsc->lsc_strict_mode = 0;
1510 				break;
1511 			case LAGG_OPT_LACP_FAST_TIMO:
1512 				LACP_LOCK(lsc);
1513         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1514                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1515 				LACP_UNLOCK(lsc);
1516 				lsc->lsc_fast_timeout = 1;
1517 				break;
1518 			case -LAGG_OPT_LACP_FAST_TIMO:
1519 				LACP_LOCK(lsc);
1520         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1521                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1522 				LACP_UNLOCK(lsc);
1523 				lsc->lsc_fast_timeout = 0;
1524 				break;
1525 			}
1526 		}
1527 		LAGG_XUNLOCK(sc);
1528 		break;
1529 	case SIOCGLAGGFLAGS:
1530 		rf->rf_flags = 0;
1531 		LAGG_XLOCK(sc);
1532 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1533 			rf->rf_flags |= LAGG_F_HASHL2;
1534 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1535 			rf->rf_flags |= LAGG_F_HASHL3;
1536 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1537 			rf->rf_flags |= LAGG_F_HASHL4;
1538 		LAGG_XUNLOCK(sc);
1539 		break;
1540 	case SIOCSLAGGHASH:
1541 		error = priv_check(td, PRIV_NET_LAGG);
1542 		if (error)
1543 			break;
1544 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1545 			error = EINVAL;
1546 			break;
1547 		}
1548 		LAGG_XLOCK(sc);
1549 		sc->sc_flags = 0;
1550 		if (rf->rf_flags & LAGG_F_HASHL2)
1551 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1552 		if (rf->rf_flags & LAGG_F_HASHL3)
1553 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1554 		if (rf->rf_flags & LAGG_F_HASHL4)
1555 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1556 		LAGG_XUNLOCK(sc);
1557 		break;
1558 	case SIOCGLAGGPORT:
1559 		if (rp->rp_portname[0] == '\0' ||
1560 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1561 			error = EINVAL;
1562 			break;
1563 		}
1564 
1565 		NET_EPOCH_ENTER(et);
1566 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1567 		    lp->lp_softc != sc) {
1568 			error = ENOENT;
1569 			NET_EPOCH_EXIT(et);
1570 			if_rele(tpif);
1571 			break;
1572 		}
1573 
1574 		lagg_port2req(lp, rp);
1575 		NET_EPOCH_EXIT(et);
1576 		if_rele(tpif);
1577 		break;
1578 	case SIOCSLAGGPORT:
1579 		error = priv_check(td, PRIV_NET_LAGG);
1580 		if (error)
1581 			break;
1582 		if (rp->rp_portname[0] == '\0' ||
1583 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1584 			error = EINVAL;
1585 			break;
1586 		}
1587 #ifdef INET6
1588 		/*
1589 		 * A laggport interface should not have inet6 address
1590 		 * because two interfaces with a valid link-local
1591 		 * scope zone must not be merged in any form.  This
1592 		 * restriction is needed to prevent violation of
1593 		 * link-local scope zone.  Attempts to add a laggport
1594 		 * interface which has inet6 addresses triggers
1595 		 * removal of all inet6 addresses on the member
1596 		 * interface.
1597 		 */
1598 		if (in6ifa_llaonifp(tpif)) {
1599 			in6_ifdetach(tpif);
1600 				if_printf(sc->sc_ifp,
1601 				    "IPv6 addresses on %s have been removed "
1602 				    "before adding it as a member to prevent "
1603 				    "IPv6 address scope violation.\n",
1604 				    tpif->if_xname);
1605 		}
1606 #endif
1607 		oldmtu = ifp->if_mtu;
1608 		LAGG_XLOCK(sc);
1609 		error = lagg_port_create(sc, tpif);
1610 		LAGG_XUNLOCK(sc);
1611 		if_rele(tpif);
1612 
1613 		/*
1614 		 * LAGG MTU may change during addition of the first port.
1615 		 * If it did, do network layer specific procedure.
1616 		 */
1617 		if (ifp->if_mtu != oldmtu) {
1618 #ifdef INET6
1619 			nd6_setmtu(ifp);
1620 #endif
1621 			rt_updatemtu(ifp);
1622 		}
1623 
1624 		VLAN_CAPABILITIES(ifp);
1625 		break;
1626 	case SIOCSLAGGDELPORT:
1627 		error = priv_check(td, PRIV_NET_LAGG);
1628 		if (error)
1629 			break;
1630 		if (rp->rp_portname[0] == '\0' ||
1631 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1632 			error = EINVAL;
1633 			break;
1634 		}
1635 
1636 		LAGG_XLOCK(sc);
1637 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1638 		    lp->lp_softc != sc) {
1639 			error = ENOENT;
1640 			LAGG_XUNLOCK(sc);
1641 			if_rele(tpif);
1642 			break;
1643 		}
1644 
1645 		error = lagg_port_destroy(lp, 1);
1646 		LAGG_XUNLOCK(sc);
1647 		if_rele(tpif);
1648 		VLAN_CAPABILITIES(ifp);
1649 		break;
1650 	case SIOCSIFFLAGS:
1651 		/* Set flags on ports too */
1652 		LAGG_XLOCK(sc);
1653 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1654 			lagg_setflags(lp, 1);
1655 		}
1656 
1657 		if (!(ifp->if_flags & IFF_UP) &&
1658 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1659 			/*
1660 			 * If interface is marked down and it is running,
1661 			 * then stop and disable it.
1662 			 */
1663 			lagg_stop(sc);
1664 			LAGG_XUNLOCK(sc);
1665 		} else if ((ifp->if_flags & IFF_UP) &&
1666 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1667 			/*
1668 			 * If interface is marked up and it is stopped, then
1669 			 * start it.
1670 			 */
1671 			LAGG_XUNLOCK(sc);
1672 			(*ifp->if_init)(sc);
1673 		} else
1674 			LAGG_XUNLOCK(sc);
1675 		break;
1676 	case SIOCADDMULTI:
1677 	case SIOCDELMULTI:
1678 		LAGG_XLOCK(sc);
1679 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1680 			lagg_clrmulti(lp);
1681 			lagg_setmulti(lp);
1682 		}
1683 		LAGG_XUNLOCK(sc);
1684 		error = 0;
1685 		break;
1686 	case SIOCSIFMEDIA:
1687 	case SIOCGIFMEDIA:
1688 		if (ifp->if_type == IFT_INFINIBAND)
1689 			error = EINVAL;
1690 		else
1691 			error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1692 		break;
1693 
1694 	case SIOCSIFCAP:
1695 		LAGG_XLOCK(sc);
1696 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1697 			if (lp->lp_ioctl != NULL)
1698 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1699 		}
1700 		lagg_capabilities(sc);
1701 		LAGG_XUNLOCK(sc);
1702 		VLAN_CAPABILITIES(ifp);
1703 		error = 0;
1704 		break;
1705 
1706 	case SIOCSIFMTU:
1707 		LAGG_XLOCK(sc);
1708 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1709 			if (lp->lp_ioctl != NULL)
1710 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1711 			else
1712 				error = EINVAL;
1713 			if (error != 0) {
1714 				if_printf(ifp,
1715 				    "failed to change MTU to %d on port %s, "
1716 				    "reverting all ports to original MTU (%d)\n",
1717 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1718 				break;
1719 			}
1720 		}
1721 		if (error == 0) {
1722 			ifp->if_mtu = ifr->ifr_mtu;
1723 		} else {
1724 			/* set every port back to the original MTU */
1725 			ifr->ifr_mtu = ifp->if_mtu;
1726 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1727 				if (lp->lp_ioctl != NULL)
1728 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1729 			}
1730 		}
1731 		LAGG_XUNLOCK(sc);
1732 		break;
1733 
1734 	default:
1735 		error = ether_ioctl(ifp, cmd, data);
1736 		break;
1737 	}
1738 	return (error);
1739 }
1740 
1741 #if defined(KERN_TLS) || defined(RATELIMIT)
1742 static inline struct lagg_snd_tag *
1743 mst_to_lst(struct m_snd_tag *mst)
1744 {
1745 
1746 	return (__containerof(mst, struct lagg_snd_tag, com));
1747 }
1748 
1749 /*
1750  * Look up the port used by a specific flow.  This only works for lagg
1751  * protocols with deterministic port mappings (e.g. not roundrobin).
1752  * In addition protocols which use a hash to map flows to ports must
1753  * be configured to use the mbuf flowid rather than hashing packet
1754  * contents.
1755  */
1756 static struct lagg_port *
1757 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1758     uint8_t numa_domain)
1759 {
1760 	struct lagg_softc *sc;
1761 	struct lagg_port *lp;
1762 	struct lagg_lb *lb;
1763 	uint32_t hash, p;
1764 	int err;
1765 
1766 	sc = ifp->if_softc;
1767 
1768 	switch (sc->sc_proto) {
1769 	case LAGG_PROTO_FAILOVER:
1770 		return (lagg_link_active(sc, sc->sc_primary));
1771 	case LAGG_PROTO_LOADBALANCE:
1772 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1773 		    flowtype == M_HASHTYPE_NONE)
1774 			return (NULL);
1775 		p = flowid >> sc->flowid_shift;
1776 		p %= sc->sc_count;
1777 		lb = (struct lagg_lb *)sc->sc_psc;
1778 		lp = lb->lb_ports[p];
1779 		return (lagg_link_active(sc, lp));
1780 	case LAGG_PROTO_LACP:
1781 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1782 		    flowtype == M_HASHTYPE_NONE)
1783 			return (NULL);
1784 		hash = flowid >> sc->flowid_shift;
1785 		return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, &err));
1786 	default:
1787 		return (NULL);
1788 	}
1789 }
1790 
1791 static int
1792 lagg_snd_tag_alloc(struct ifnet *ifp,
1793     union if_snd_tag_alloc_params *params,
1794     struct m_snd_tag **ppmt)
1795 {
1796 	struct epoch_tracker et;
1797 	struct lagg_snd_tag *lst;
1798 	struct lagg_softc *sc;
1799 	struct lagg_port *lp;
1800 	struct ifnet *lp_ifp;
1801 	int error;
1802 
1803 	sc = ifp->if_softc;
1804 
1805 	NET_EPOCH_ENTER(et);
1806 	lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1807 	    params->hdr.flowtype, params->hdr.numa_domain);
1808 	if (lp == NULL) {
1809 		NET_EPOCH_EXIT(et);
1810 		return (EOPNOTSUPP);
1811 	}
1812 	if (lp->lp_ifp == NULL) {
1813 		NET_EPOCH_EXIT(et);
1814 		return (EOPNOTSUPP);
1815 	}
1816 	lp_ifp = lp->lp_ifp;
1817 	if_ref(lp_ifp);
1818 	NET_EPOCH_EXIT(et);
1819 
1820 	lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1821 	if (lst == NULL) {
1822 		if_rele(lp_ifp);
1823 		return (ENOMEM);
1824 	}
1825 
1826 	error = m_snd_tag_alloc(lp_ifp, params, &lst->tag);
1827 	if_rele(lp_ifp);
1828 	if (error) {
1829 		free(lst, M_LAGG);
1830 		return (error);
1831 	}
1832 
1833 	m_snd_tag_init(&lst->com, ifp, lst->tag->type);
1834 
1835 	*ppmt = &lst->com;
1836 	return (0);
1837 }
1838 
1839 static struct m_snd_tag *
1840 lagg_next_snd_tag(struct m_snd_tag *mst)
1841 {
1842 	struct lagg_snd_tag *lst;
1843 
1844 	lst = mst_to_lst(mst);
1845 	return (lst->tag);
1846 }
1847 
1848 static int
1849 lagg_snd_tag_modify(struct m_snd_tag *mst,
1850     union if_snd_tag_modify_params *params)
1851 {
1852 	struct lagg_snd_tag *lst;
1853 
1854 	lst = mst_to_lst(mst);
1855 	return (lst->tag->ifp->if_snd_tag_modify(lst->tag, params));
1856 }
1857 
1858 static int
1859 lagg_snd_tag_query(struct m_snd_tag *mst,
1860     union if_snd_tag_query_params *params)
1861 {
1862 	struct lagg_snd_tag *lst;
1863 
1864 	lst = mst_to_lst(mst);
1865 	return (lst->tag->ifp->if_snd_tag_query(lst->tag, params));
1866 }
1867 
1868 static void
1869 lagg_snd_tag_free(struct m_snd_tag *mst)
1870 {
1871 	struct lagg_snd_tag *lst;
1872 
1873 	lst = mst_to_lst(mst);
1874 	m_snd_tag_rele(lst->tag);
1875 	free(lst, M_LAGG);
1876 }
1877 
1878 static void
1879 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1880 {
1881 	/*
1882 	 * For lagg, we have an indirect
1883 	 * interface. The caller needs to
1884 	 * get a ratelimit tag on the actual
1885 	 * interface the flow will go on.
1886 	 */
1887 	q->rate_table = NULL;
1888 	q->flags = RT_IS_INDIRECT;
1889 	q->max_flows = 0;
1890 	q->number_of_rates = 0;
1891 }
1892 #endif
1893 
1894 static int
1895 lagg_setmulti(struct lagg_port *lp)
1896 {
1897 	struct lagg_softc *sc = lp->lp_softc;
1898 	struct ifnet *ifp = lp->lp_ifp;
1899 	struct ifnet *scifp = sc->sc_ifp;
1900 	struct lagg_mc *mc;
1901 	struct ifmultiaddr *ifma;
1902 	int error;
1903 
1904 	IF_ADDR_WLOCK(scifp);
1905 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1906 		if (ifma->ifma_addr->sa_family != AF_LINK)
1907 			continue;
1908 		mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
1909 		if (mc == NULL) {
1910 			IF_ADDR_WUNLOCK(scifp);
1911 			return (ENOMEM);
1912 		}
1913 		bcopy(ifma->ifma_addr, &mc->mc_addr,
1914 		    ifma->ifma_addr->sa_len);
1915 		mc->mc_addr.sdl_index = ifp->if_index;
1916 		mc->mc_ifma = NULL;
1917 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1918 	}
1919 	IF_ADDR_WUNLOCK(scifp);
1920 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1921 		error = if_addmulti(ifp,
1922 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1923 		if (error)
1924 			return (error);
1925 	}
1926 	return (0);
1927 }
1928 
1929 static int
1930 lagg_clrmulti(struct lagg_port *lp)
1931 {
1932 	struct lagg_mc *mc;
1933 
1934 	LAGG_XLOCK_ASSERT(lp->lp_softc);
1935 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1936 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1937 		if (mc->mc_ifma && lp->lp_detaching == 0)
1938 			if_delmulti_ifma(mc->mc_ifma);
1939 		free(mc, M_LAGG);
1940 	}
1941 	return (0);
1942 }
1943 
1944 static int
1945 lagg_setcaps(struct lagg_port *lp, int cap)
1946 {
1947 	struct ifreq ifr;
1948 
1949 	if (lp->lp_ifp->if_capenable == cap)
1950 		return (0);
1951 	if (lp->lp_ioctl == NULL)
1952 		return (ENXIO);
1953 	ifr.ifr_reqcap = cap;
1954 	return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1955 }
1956 
1957 /* Handle a ref counted flag that should be set on the lagg port as well */
1958 static int
1959 lagg_setflag(struct lagg_port *lp, int flag, int status,
1960     int (*func)(struct ifnet *, int))
1961 {
1962 	struct lagg_softc *sc = lp->lp_softc;
1963 	struct ifnet *scifp = sc->sc_ifp;
1964 	struct ifnet *ifp = lp->lp_ifp;
1965 	int error;
1966 
1967 	LAGG_XLOCK_ASSERT(sc);
1968 
1969 	status = status ? (scifp->if_flags & flag) : 0;
1970 	/* Now "status" contains the flag value or 0 */
1971 
1972 	/*
1973 	 * See if recorded ports status is different from what
1974 	 * we want it to be.  If it is, flip it.  We record ports
1975 	 * status in lp_ifflags so that we won't clear ports flag
1976 	 * we haven't set.  In fact, we don't clear or set ports
1977 	 * flags directly, but get or release references to them.
1978 	 * That's why we can be sure that recorded flags still are
1979 	 * in accord with actual ports flags.
1980 	 */
1981 	if (status != (lp->lp_ifflags & flag)) {
1982 		error = (*func)(ifp, status);
1983 		if (error)
1984 			return (error);
1985 		lp->lp_ifflags &= ~flag;
1986 		lp->lp_ifflags |= status;
1987 	}
1988 	return (0);
1989 }
1990 
1991 /*
1992  * Handle IFF_* flags that require certain changes on the lagg port
1993  * if "status" is true, update ports flags respective to the lagg
1994  * if "status" is false, forcedly clear the flags set on port.
1995  */
1996 static int
1997 lagg_setflags(struct lagg_port *lp, int status)
1998 {
1999 	int error, i;
2000 
2001 	for (i = 0; lagg_pflags[i].flag; i++) {
2002 		error = lagg_setflag(lp, lagg_pflags[i].flag,
2003 		    status, lagg_pflags[i].func);
2004 		if (error)
2005 			return (error);
2006 	}
2007 	return (0);
2008 }
2009 
2010 static int
2011 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2012 {
2013 	struct epoch_tracker et;
2014 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2015 	int error;
2016 
2017 #if defined(KERN_TLS) || defined(RATELIMIT)
2018 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2019 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2020 #endif
2021 	NET_EPOCH_ENTER(et);
2022 	/* We need a Tx algorithm and at least one port */
2023 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2024 		NET_EPOCH_EXIT(et);
2025 		m_freem(m);
2026 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2027 		return (ENXIO);
2028 	}
2029 
2030 	ETHER_BPF_MTAP(ifp, m);
2031 
2032 	error = lagg_proto_start(sc, m);
2033 	NET_EPOCH_EXIT(et);
2034 	return (error);
2035 }
2036 
2037 static int
2038 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2039 {
2040 	struct epoch_tracker et;
2041 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2042 	int error;
2043 
2044 #if defined(KERN_TLS) || defined(RATELIMIT)
2045 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2046 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2047 #endif
2048 	NET_EPOCH_ENTER(et);
2049 	/* We need a Tx algorithm and at least one port */
2050 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2051 		NET_EPOCH_EXIT(et);
2052 		m_freem(m);
2053 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2054 		return (ENXIO);
2055 	}
2056 
2057 	INFINIBAND_BPF_MTAP(ifp, m);
2058 
2059 	error = lagg_proto_start(sc, m);
2060 	NET_EPOCH_EXIT(et);
2061 	return (error);
2062 }
2063 
2064 /*
2065  * The ifp->if_qflush entry point for lagg(4) is no-op.
2066  */
2067 static void
2068 lagg_qflush(struct ifnet *ifp __unused)
2069 {
2070 }
2071 
2072 static struct mbuf *
2073 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2074 {
2075 	struct epoch_tracker et;
2076 	struct lagg_port *lp = ifp->if_lagg;
2077 	struct lagg_softc *sc = lp->lp_softc;
2078 	struct ifnet *scifp = sc->sc_ifp;
2079 
2080 	NET_EPOCH_ENTER(et);
2081 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2082 	    lp->lp_detaching != 0 ||
2083 	    sc->sc_proto == LAGG_PROTO_NONE) {
2084 		NET_EPOCH_EXIT(et);
2085 		m_freem(m);
2086 		return (NULL);
2087 	}
2088 
2089 	ETHER_BPF_MTAP(scifp, m);
2090 
2091 	m = lagg_proto_input(sc, lp, m);
2092 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2093 		m_freem(m);
2094 		m = NULL;
2095 	}
2096 
2097 	NET_EPOCH_EXIT(et);
2098 	return (m);
2099 }
2100 
2101 static struct mbuf *
2102 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2103 {
2104 	struct epoch_tracker et;
2105 	struct lagg_port *lp = ifp->if_lagg;
2106 	struct lagg_softc *sc = lp->lp_softc;
2107 	struct ifnet *scifp = sc->sc_ifp;
2108 
2109 	NET_EPOCH_ENTER(et);
2110 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2111 	    lp->lp_detaching != 0 ||
2112 	    sc->sc_proto == LAGG_PROTO_NONE) {
2113 		NET_EPOCH_EXIT(et);
2114 		m_freem(m);
2115 		return (NULL);
2116 	}
2117 
2118 	INFINIBAND_BPF_MTAP(scifp, m);
2119 
2120 	m = lagg_proto_input(sc, lp, m);
2121 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2122 		m_freem(m);
2123 		m = NULL;
2124 	}
2125 
2126 	NET_EPOCH_EXIT(et);
2127 	return (m);
2128 }
2129 
2130 static int
2131 lagg_media_change(struct ifnet *ifp)
2132 {
2133 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2134 
2135 	if (sc->sc_ifflags & IFF_DEBUG)
2136 		printf("%s\n", __func__);
2137 
2138 	/* Ignore */
2139 	return (0);
2140 }
2141 
2142 static void
2143 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2144 {
2145 	struct epoch_tracker et;
2146 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2147 	struct lagg_port *lp;
2148 
2149 	imr->ifm_status = IFM_AVALID;
2150 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
2151 
2152 	NET_EPOCH_ENTER(et);
2153 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2154 		if (LAGG_PORTACTIVE(lp))
2155 			imr->ifm_status |= IFM_ACTIVE;
2156 	}
2157 	NET_EPOCH_EXIT(et);
2158 }
2159 
2160 static void
2161 lagg_linkstate(struct lagg_softc *sc)
2162 {
2163 	struct epoch_tracker et;
2164 	struct lagg_port *lp;
2165 	int new_link = LINK_STATE_DOWN;
2166 	uint64_t speed;
2167 
2168 	LAGG_XLOCK_ASSERT(sc);
2169 
2170 	/* LACP handles link state itself */
2171 	if (sc->sc_proto == LAGG_PROTO_LACP)
2172 		return;
2173 
2174 	/* Our link is considered up if at least one of our ports is active */
2175 	NET_EPOCH_ENTER(et);
2176 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2177 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2178 			new_link = LINK_STATE_UP;
2179 			break;
2180 		}
2181 	}
2182 	NET_EPOCH_EXIT(et);
2183 	if_link_state_change(sc->sc_ifp, new_link);
2184 
2185 	/* Update if_baudrate to reflect the max possible speed */
2186 	switch (sc->sc_proto) {
2187 		case LAGG_PROTO_FAILOVER:
2188 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2189 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
2190 			break;
2191 		case LAGG_PROTO_ROUNDROBIN:
2192 		case LAGG_PROTO_LOADBALANCE:
2193 		case LAGG_PROTO_BROADCAST:
2194 			speed = 0;
2195 			NET_EPOCH_ENTER(et);
2196 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2197 				speed += lp->lp_ifp->if_baudrate;
2198 			NET_EPOCH_EXIT(et);
2199 			sc->sc_ifp->if_baudrate = speed;
2200 			break;
2201 		case LAGG_PROTO_LACP:
2202 			/* LACP updates if_baudrate itself */
2203 			break;
2204 	}
2205 }
2206 
2207 static void
2208 lagg_port_state(struct ifnet *ifp, int state)
2209 {
2210 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2211 	struct lagg_softc *sc = NULL;
2212 
2213 	if (lp != NULL)
2214 		sc = lp->lp_softc;
2215 	if (sc == NULL)
2216 		return;
2217 
2218 	LAGG_XLOCK(sc);
2219 	lagg_linkstate(sc);
2220 	lagg_proto_linkstate(sc, lp);
2221 	LAGG_XUNLOCK(sc);
2222 }
2223 
2224 struct lagg_port *
2225 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2226 {
2227 	struct lagg_port *lp_next, *rval = NULL;
2228 
2229 	/*
2230 	 * Search a port which reports an active link state.
2231 	 */
2232 
2233 #ifdef INVARIANTS
2234 	/*
2235 	 * This is called with either in the network epoch
2236 	 * or with LAGG_XLOCK(sc) held.
2237 	 */
2238 	if (!in_epoch(net_epoch_preempt))
2239 		LAGG_XLOCK_ASSERT(sc);
2240 #endif
2241 
2242 	if (lp == NULL)
2243 		goto search;
2244 	if (LAGG_PORTACTIVE(lp)) {
2245 		rval = lp;
2246 		goto found;
2247 	}
2248 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2249 	    LAGG_PORTACTIVE(lp_next)) {
2250 		rval = lp_next;
2251 		goto found;
2252 	}
2253 
2254 search:
2255 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2256 		if (LAGG_PORTACTIVE(lp_next)) {
2257 			return (lp_next);
2258 		}
2259 	}
2260 found:
2261 	return (rval);
2262 }
2263 
2264 int
2265 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2266 {
2267 
2268 #if defined(KERN_TLS) || defined(RATELIMIT)
2269 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2270 		struct lagg_snd_tag *lst;
2271 		struct m_snd_tag *mst;
2272 
2273 		mst = m->m_pkthdr.snd_tag;
2274 		lst = mst_to_lst(mst);
2275 		if (lst->tag->ifp != ifp) {
2276 			m_freem(m);
2277 			return (EAGAIN);
2278 		}
2279 		m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2280 		m_snd_tag_rele(mst);
2281 	}
2282 #endif
2283 	return (ifp->if_transmit)(ifp, m);
2284 }
2285 
2286 /*
2287  * Simple round robin aggregation
2288  */
2289 static void
2290 lagg_rr_attach(struct lagg_softc *sc)
2291 {
2292 	sc->sc_seq = 0;
2293 	sc->sc_stride = 1;
2294 }
2295 
2296 static int
2297 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2298 {
2299 	struct lagg_port *lp;
2300 	uint32_t p;
2301 
2302 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
2303 	p /= sc->sc_stride;
2304 	p %= sc->sc_count;
2305 	lp = CK_SLIST_FIRST(&sc->sc_ports);
2306 
2307 	while (p--)
2308 		lp = CK_SLIST_NEXT(lp, lp_entries);
2309 
2310 	/*
2311 	 * Check the port's link state. This will return the next active
2312 	 * port if the link is down or the port is NULL.
2313 	 */
2314 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2315 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2316 		m_freem(m);
2317 		return (ENETDOWN);
2318 	}
2319 
2320 	/* Send mbuf */
2321 	return (lagg_enqueue(lp->lp_ifp, m));
2322 }
2323 
2324 static struct mbuf *
2325 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2326 {
2327 	struct ifnet *ifp = sc->sc_ifp;
2328 
2329 	/* Just pass in the packet to our lagg device */
2330 	m->m_pkthdr.rcvif = ifp;
2331 
2332 	return (m);
2333 }
2334 
2335 /*
2336  * Broadcast mode
2337  */
2338 static int
2339 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2340 {
2341 	int active_ports = 0;
2342 	int errors = 0;
2343 	int ret;
2344 	struct lagg_port *lp, *last = NULL;
2345 	struct mbuf *m0;
2346 
2347 	NET_EPOCH_ASSERT();
2348 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2349 		if (!LAGG_PORTACTIVE(lp))
2350 			continue;
2351 
2352 		active_ports++;
2353 
2354 		if (last != NULL) {
2355 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2356 			if (m0 == NULL) {
2357 				ret = ENOBUFS;
2358 				errors++;
2359 				break;
2360 			}
2361 			lagg_enqueue(last->lp_ifp, m0);
2362 		}
2363 		last = lp;
2364 	}
2365 
2366 	if (last == NULL) {
2367 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2368 		m_freem(m);
2369 		return (ENOENT);
2370 	}
2371 	if ((last = lagg_link_active(sc, last)) == NULL) {
2372 		errors++;
2373 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2374 		m_freem(m);
2375 		return (ENETDOWN);
2376 	}
2377 
2378 	ret = lagg_enqueue(last->lp_ifp, m);
2379 	if (errors != 0)
2380 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2381 
2382 	return (ret);
2383 }
2384 
2385 static struct mbuf*
2386 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2387 {
2388 	struct ifnet *ifp = sc->sc_ifp;
2389 
2390 	/* Just pass in the packet to our lagg device */
2391 	m->m_pkthdr.rcvif = ifp;
2392 	return (m);
2393 }
2394 
2395 /*
2396  * Active failover
2397  */
2398 static int
2399 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2400 {
2401 	struct lagg_port *lp;
2402 
2403 	/* Use the master port if active or the next available port */
2404 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2405 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2406 		m_freem(m);
2407 		return (ENETDOWN);
2408 	}
2409 
2410 	/* Send mbuf */
2411 	return (lagg_enqueue(lp->lp_ifp, m));
2412 }
2413 
2414 static struct mbuf *
2415 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2416 {
2417 	struct ifnet *ifp = sc->sc_ifp;
2418 	struct lagg_port *tmp_tp;
2419 
2420 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2421 		m->m_pkthdr.rcvif = ifp;
2422 		return (m);
2423 	}
2424 
2425 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2426 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2427 		/*
2428 		 * If tmp_tp is null, we've received a packet when all
2429 		 * our links are down. Weird, but process it anyways.
2430 		 */
2431 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2432 			m->m_pkthdr.rcvif = ifp;
2433 			return (m);
2434 		}
2435 	}
2436 
2437 	m_freem(m);
2438 	return (NULL);
2439 }
2440 
2441 /*
2442  * Loadbalancing
2443  */
2444 static void
2445 lagg_lb_attach(struct lagg_softc *sc)
2446 {
2447 	struct lagg_port *lp;
2448 	struct lagg_lb *lb;
2449 
2450 	LAGG_XLOCK_ASSERT(sc);
2451 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2452 	lb->lb_key = m_ether_tcpip_hash_init();
2453 	sc->sc_psc = lb;
2454 
2455 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2456 		lagg_lb_port_create(lp);
2457 }
2458 
2459 static void
2460 lagg_lb_detach(struct lagg_softc *sc)
2461 {
2462 	struct lagg_lb *lb;
2463 
2464 	lb = (struct lagg_lb *)sc->sc_psc;
2465 	if (lb != NULL)
2466 		free(lb, M_LAGG);
2467 }
2468 
2469 static int
2470 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2471 {
2472 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2473 	struct lagg_port *lp_next;
2474 	int i = 0, rv;
2475 
2476 	rv = 0;
2477 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2478 	LAGG_XLOCK_ASSERT(sc);
2479 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2480 		if (lp_next == lp)
2481 			continue;
2482 		if (i >= LAGG_MAX_PORTS) {
2483 			rv = EINVAL;
2484 			break;
2485 		}
2486 		if (sc->sc_ifflags & IFF_DEBUG)
2487 			printf("%s: port %s at index %d\n",
2488 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2489 		lb->lb_ports[i++] = lp_next;
2490 	}
2491 
2492 	return (rv);
2493 }
2494 
2495 static int
2496 lagg_lb_port_create(struct lagg_port *lp)
2497 {
2498 	struct lagg_softc *sc = lp->lp_softc;
2499 	return (lagg_lb_porttable(sc, NULL));
2500 }
2501 
2502 static void
2503 lagg_lb_port_destroy(struct lagg_port *lp)
2504 {
2505 	struct lagg_softc *sc = lp->lp_softc;
2506 	lagg_lb_porttable(sc, lp);
2507 }
2508 
2509 static int
2510 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2511 {
2512 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2513 	struct lagg_port *lp = NULL;
2514 	uint32_t p = 0;
2515 
2516 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2517 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2518 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2519 	else
2520 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2521 	p %= sc->sc_count;
2522 	lp = lb->lb_ports[p];
2523 
2524 	/*
2525 	 * Check the port's link state. This will return the next active
2526 	 * port if the link is down or the port is NULL.
2527 	 */
2528 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2529 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2530 		m_freem(m);
2531 		return (ENETDOWN);
2532 	}
2533 
2534 	/* Send mbuf */
2535 	return (lagg_enqueue(lp->lp_ifp, m));
2536 }
2537 
2538 static struct mbuf *
2539 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2540 {
2541 	struct ifnet *ifp = sc->sc_ifp;
2542 
2543 	/* Just pass in the packet to our lagg device */
2544 	m->m_pkthdr.rcvif = ifp;
2545 
2546 	return (m);
2547 }
2548 
2549 /*
2550  * 802.3ad LACP
2551  */
2552 static void
2553 lagg_lacp_attach(struct lagg_softc *sc)
2554 {
2555 	struct lagg_port *lp;
2556 
2557 	lacp_attach(sc);
2558 	LAGG_XLOCK_ASSERT(sc);
2559 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2560 		lacp_port_create(lp);
2561 }
2562 
2563 static void
2564 lagg_lacp_detach(struct lagg_softc *sc)
2565 {
2566 	struct lagg_port *lp;
2567 	void *psc;
2568 
2569 	LAGG_XLOCK_ASSERT(sc);
2570 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2571 		lacp_port_destroy(lp);
2572 
2573 	psc = sc->sc_psc;
2574 	sc->sc_psc = NULL;
2575 	lacp_detach(psc);
2576 }
2577 
2578 static void
2579 lagg_lacp_lladdr(struct lagg_softc *sc)
2580 {
2581 	struct lagg_port *lp;
2582 
2583 	LAGG_SXLOCK_ASSERT(sc);
2584 
2585 	/* purge all the lacp ports */
2586 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2587 		lacp_port_destroy(lp);
2588 
2589 	/* add them back in */
2590 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2591 		lacp_port_create(lp);
2592 }
2593 
2594 static int
2595 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2596 {
2597 	struct lagg_port *lp;
2598 	int err;
2599 
2600 	lp = lacp_select_tx_port(sc, m, &err);
2601 	if (lp == NULL) {
2602 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2603 		m_freem(m);
2604 		return (err);
2605 	}
2606 
2607 	/* Send mbuf */
2608 	return (lagg_enqueue(lp->lp_ifp, m));
2609 }
2610 
2611 static struct mbuf *
2612 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2613 {
2614 	struct ifnet *ifp = sc->sc_ifp;
2615 	struct ether_header *eh;
2616 	u_short etype;
2617 
2618 	eh = mtod(m, struct ether_header *);
2619 	etype = ntohs(eh->ether_type);
2620 
2621 	/* Tap off LACP control messages */
2622 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2623 		m = lacp_input(lp, m);
2624 		if (m == NULL)
2625 			return (NULL);
2626 	}
2627 
2628 	/*
2629 	 * If the port is not collecting or not in the active aggregator then
2630 	 * free and return.
2631 	 */
2632 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2633 		m_freem(m);
2634 		return (NULL);
2635 	}
2636 
2637 	m->m_pkthdr.rcvif = ifp;
2638 	return (m);
2639 }
2640