xref: /freebsd/sys/net/if_lagg.c (revision 4bfebc8d2c5d0e813dfdcbe7038b36ffc2bb9f1b)
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_RLOCK()	struct epoch_tracker lagg_et; epoch_enter_preempt(net_epoch_preempt, &lagg_et)
88 #define	LAGG_RUNLOCK()	epoch_exit_preempt(net_epoch_preempt, &lagg_et)
89 #define	LAGG_RLOCK_ASSERT()	NET_EPOCH_ASSERT()
90 #define	LAGG_UNLOCK_ASSERT()	MPASS(!in_epoch(net_epoch_preempt))
91 
92 #define	LAGG_SX_INIT(_sc)	sx_init(&(_sc)->sc_sx, "if_lagg sx")
93 #define	LAGG_SX_DESTROY(_sc)	sx_destroy(&(_sc)->sc_sx)
94 #define	LAGG_XLOCK(_sc)		sx_xlock(&(_sc)->sc_sx)
95 #define	LAGG_XUNLOCK(_sc)	sx_xunlock(&(_sc)->sc_sx)
96 #define	LAGG_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
97 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
98 
99 /* Special flags we should propagate to the lagg ports. */
100 static struct {
101 	int flag;
102 	int (*func)(struct ifnet *, int);
103 } lagg_pflags[] = {
104 	{IFF_PROMISC, ifpromisc},
105 	{IFF_ALLMULTI, if_allmulti},
106 	{0, NULL}
107 };
108 
109 struct lagg_snd_tag {
110 	struct m_snd_tag com;
111 	struct m_snd_tag *tag;
112 };
113 
114 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
115 #define	V_lagg_list	VNET(lagg_list)
116 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
117 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
118 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
119 					"if_lagg list", NULL, MTX_DEF)
120 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
121 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
122 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
123 eventhandler_tag	lagg_detach_cookie = NULL;
124 
125 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
126 static void	lagg_clone_destroy(struct ifnet *);
127 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
128 #define	V_lagg_cloner	VNET(lagg_cloner)
129 static const char laggname[] = "lagg";
130 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
131 
132 static void	lagg_capabilities(struct lagg_softc *);
133 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
134 static int	lagg_port_destroy(struct lagg_port *, int);
135 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *);
136 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *);
137 static void	lagg_linkstate(struct lagg_softc *);
138 static void	lagg_port_state(struct ifnet *, int);
139 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
140 static int	lagg_port_output(struct ifnet *, struct mbuf *,
141 		    const struct sockaddr *, struct route *);
142 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
143 #ifdef LAGG_PORT_STACKING
144 static int	lagg_port_checkstacking(struct lagg_softc *);
145 #endif
146 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
147 static void	lagg_init(void *);
148 static void	lagg_stop(struct lagg_softc *);
149 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
150 #if defined(KERN_TLS) || defined(RATELIMIT)
151 static int	lagg_snd_tag_alloc(struct ifnet *,
152 		    union if_snd_tag_alloc_params *,
153 		    struct m_snd_tag **);
154 static int	lagg_snd_tag_modify(struct m_snd_tag *,
155 		    union if_snd_tag_modify_params *);
156 static int	lagg_snd_tag_query(struct m_snd_tag *,
157 		    union if_snd_tag_query_params *);
158 static void	lagg_snd_tag_free(struct m_snd_tag *);
159 static void     lagg_ratelimit_query(struct ifnet *,
160 		    struct if_ratelimit_query_results *);
161 #endif
162 static int	lagg_setmulti(struct lagg_port *);
163 static int	lagg_clrmulti(struct lagg_port *);
164 static	int	lagg_setcaps(struct lagg_port *, int cap);
165 static	int	lagg_setflag(struct lagg_port *, int, int,
166 		    int (*func)(struct ifnet *, int));
167 static	int	lagg_setflags(struct lagg_port *, int status);
168 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
169 static int	lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
170 static int	lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
171 static void	lagg_qflush(struct ifnet *);
172 static int	lagg_media_change(struct ifnet *);
173 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
174 static struct lagg_port *lagg_link_active(struct lagg_softc *,
175 	    struct lagg_port *);
176 
177 /* Simple round robin */
178 static void	lagg_rr_attach(struct lagg_softc *);
179 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
180 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
181 		    struct mbuf *);
182 
183 /* Active failover */
184 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
185 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
186 		    struct mbuf *);
187 
188 /* Loadbalancing */
189 static void	lagg_lb_attach(struct lagg_softc *);
190 static void	lagg_lb_detach(struct lagg_softc *);
191 static int	lagg_lb_port_create(struct lagg_port *);
192 static void	lagg_lb_port_destroy(struct lagg_port *);
193 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
194 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
195 		    struct mbuf *);
196 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
197 
198 /* Broadcast */
199 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
200 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
201 		    struct mbuf *);
202 
203 /* 802.3ad LACP */
204 static void	lagg_lacp_attach(struct lagg_softc *);
205 static void	lagg_lacp_detach(struct lagg_softc *);
206 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
207 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
208 		    struct mbuf *);
209 static void	lagg_lacp_lladdr(struct lagg_softc *);
210 
211 /* lagg protocol table */
212 static const struct lagg_proto {
213 	lagg_proto	pr_num;
214 	void		(*pr_attach)(struct lagg_softc *);
215 	void		(*pr_detach)(struct lagg_softc *);
216 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
217 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
218 			    struct mbuf *);
219 	int		(*pr_addport)(struct lagg_port *);
220 	void		(*pr_delport)(struct lagg_port *);
221 	void		(*pr_linkstate)(struct lagg_port *);
222 	void 		(*pr_init)(struct lagg_softc *);
223 	void 		(*pr_stop)(struct lagg_softc *);
224 	void 		(*pr_lladdr)(struct lagg_softc *);
225 	void		(*pr_request)(struct lagg_softc *, void *);
226 	void		(*pr_portreq)(struct lagg_port *, void *);
227 } lagg_protos[] = {
228     {
229 	.pr_num = LAGG_PROTO_NONE
230     },
231     {
232 	.pr_num = LAGG_PROTO_ROUNDROBIN,
233 	.pr_attach = lagg_rr_attach,
234 	.pr_start = lagg_rr_start,
235 	.pr_input = lagg_rr_input,
236     },
237     {
238 	.pr_num = LAGG_PROTO_FAILOVER,
239 	.pr_start = lagg_fail_start,
240 	.pr_input = lagg_fail_input,
241     },
242     {
243 	.pr_num = LAGG_PROTO_LOADBALANCE,
244 	.pr_attach = lagg_lb_attach,
245 	.pr_detach = lagg_lb_detach,
246 	.pr_start = lagg_lb_start,
247 	.pr_input = lagg_lb_input,
248 	.pr_addport = lagg_lb_port_create,
249 	.pr_delport = lagg_lb_port_destroy,
250     },
251     {
252 	.pr_num = LAGG_PROTO_LACP,
253 	.pr_attach = lagg_lacp_attach,
254 	.pr_detach = lagg_lacp_detach,
255 	.pr_start = lagg_lacp_start,
256 	.pr_input = lagg_lacp_input,
257 	.pr_addport = lacp_port_create,
258 	.pr_delport = lacp_port_destroy,
259 	.pr_linkstate = lacp_linkstate,
260 	.pr_init = lacp_init,
261 	.pr_stop = lacp_stop,
262 	.pr_lladdr = lagg_lacp_lladdr,
263 	.pr_request = lacp_req,
264 	.pr_portreq = lacp_portreq,
265     },
266     {
267 	.pr_num = LAGG_PROTO_BROADCAST,
268 	.pr_start = lagg_bcast_start,
269 	.pr_input = lagg_bcast_input,
270     },
271 };
272 
273 SYSCTL_DECL(_net_link);
274 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
275     "Link Aggregation");
276 
277 /* Allow input on any failover links */
278 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
279 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
280 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
281     &VNET_NAME(lagg_failover_rx_all), 0,
282     "Accept input from any interface in a failover lagg");
283 
284 /* Default value for using flowid */
285 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
286 #define	V_def_use_flowid	VNET(def_use_flowid)
287 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
288     &VNET_NAME(def_use_flowid), 0,
289     "Default setting for using flow id for load sharing");
290 
291 /* Default value for using numa */
292 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
293 #define	V_def_use_numa	VNET(def_use_numa)
294 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN,
295     &VNET_NAME(def_use_numa), 0,
296     "Use numa to steer flows");
297 
298 /* Default value for flowid shift */
299 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
300 #define	V_def_flowid_shift	VNET(def_flowid_shift)
301 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
302     &VNET_NAME(def_flowid_shift), 0,
303     "Default setting for flowid shift for load sharing");
304 
305 static void
306 vnet_lagg_init(const void *unused __unused)
307 {
308 
309 	LAGG_LIST_LOCK_INIT();
310 	SLIST_INIT(&V_lagg_list);
311 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
312 	    lagg_clone_destroy, 0);
313 }
314 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
315     vnet_lagg_init, NULL);
316 
317 static void
318 vnet_lagg_uninit(const void *unused __unused)
319 {
320 
321 	if_clone_detach(V_lagg_cloner);
322 	LAGG_LIST_LOCK_DESTROY();
323 }
324 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
325     vnet_lagg_uninit, NULL);
326 
327 static int
328 lagg_modevent(module_t mod, int type, void *data)
329 {
330 
331 	switch (type) {
332 	case MOD_LOAD:
333 		lagg_input_ethernet_p = lagg_input_ethernet;
334 		lagg_input_infiniband_p = lagg_input_infiniband;
335 		lagg_linkstate_p = lagg_port_state;
336 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
337 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
338 		    EVENTHANDLER_PRI_ANY);
339 		break;
340 	case MOD_UNLOAD:
341 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
342 		    lagg_detach_cookie);
343 		lagg_input_ethernet_p = NULL;
344 		lagg_input_infiniband_p = NULL;
345 		lagg_linkstate_p = NULL;
346 		break;
347 	default:
348 		return (EOPNOTSUPP);
349 	}
350 	return (0);
351 }
352 
353 static moduledata_t lagg_mod = {
354 	"if_lagg",
355 	lagg_modevent,
356 	0
357 };
358 
359 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
360 MODULE_VERSION(if_lagg, 1);
361 MODULE_DEPEND(if_lagg, if_infiniband, 1, 1, 1);
362 
363 static void
364 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
365 {
366 
367 	LAGG_XLOCK_ASSERT(sc);
368 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
369 	    __func__, sc));
370 
371 	if (sc->sc_ifflags & IFF_DEBUG)
372 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
373 
374 	if (lagg_protos[pr].pr_attach != NULL)
375 		lagg_protos[pr].pr_attach(sc);
376 	sc->sc_proto = pr;
377 }
378 
379 static void
380 lagg_proto_detach(struct lagg_softc *sc)
381 {
382 	lagg_proto pr;
383 
384 	LAGG_XLOCK_ASSERT(sc);
385 	pr = sc->sc_proto;
386 	sc->sc_proto = LAGG_PROTO_NONE;
387 
388 	if (lagg_protos[pr].pr_detach != NULL)
389 		lagg_protos[pr].pr_detach(sc);
390 }
391 
392 static int
393 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
394 {
395 
396 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
397 }
398 
399 static struct mbuf *
400 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
401 {
402 
403 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
404 }
405 
406 static int
407 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
408 {
409 
410 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
411 		return (0);
412 	else
413 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
414 }
415 
416 static void
417 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
418 {
419 
420 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
421 		lagg_protos[sc->sc_proto].pr_delport(lp);
422 }
423 
424 static void
425 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
426 {
427 
428 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
429 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
430 }
431 
432 static void
433 lagg_proto_init(struct lagg_softc *sc)
434 {
435 
436 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
437 		lagg_protos[sc->sc_proto].pr_init(sc);
438 }
439 
440 static void
441 lagg_proto_stop(struct lagg_softc *sc)
442 {
443 
444 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
445 		lagg_protos[sc->sc_proto].pr_stop(sc);
446 }
447 
448 static void
449 lagg_proto_lladdr(struct lagg_softc *sc)
450 {
451 
452 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
453 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
454 }
455 
456 static void
457 lagg_proto_request(struct lagg_softc *sc, void *v)
458 {
459 
460 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
461 		lagg_protos[sc->sc_proto].pr_request(sc, v);
462 }
463 
464 static void
465 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
466 {
467 
468 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
469 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
470 }
471 
472 /*
473  * This routine is run via an vlan
474  * config EVENT
475  */
476 static void
477 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
478 {
479 	struct lagg_softc *sc = ifp->if_softc;
480 	struct lagg_port *lp;
481 
482 	if (ifp->if_softc !=  arg)   /* Not our event */
483 		return;
484 
485 	LAGG_RLOCK();
486 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
487 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
488 	LAGG_RUNLOCK();
489 }
490 
491 /*
492  * This routine is run via an vlan
493  * unconfig EVENT
494  */
495 static void
496 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
497 {
498 	struct lagg_softc *sc = ifp->if_softc;
499 	struct lagg_port *lp;
500 
501 	if (ifp->if_softc !=  arg)   /* Not our event */
502 		return;
503 
504 	LAGG_RLOCK();
505 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
506 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
507 	LAGG_RUNLOCK();
508 }
509 
510 static int
511 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
512 {
513 	struct iflaggparam iflp;
514 	struct lagg_softc *sc;
515 	struct ifnet *ifp;
516 	int if_type;
517 	int error;
518 	static const uint8_t eaddr[LAGG_ADDR_LEN];
519 	static const uint8_t ib_bcast_addr[INFINIBAND_ADDR_LEN] = {
520 		0x00, 0xff, 0xff, 0xff,
521 		0xff, 0x12, 0x40, 0x1b,	0x00, 0x00, 0x00, 0x00,
522 		0x00, 0x00, 0x00, 0x00,	0xff, 0xff, 0xff, 0xff
523 	};
524 
525 	if (params != NULL) {
526 		error = copyin(params, &iflp, sizeof(iflp));
527 		if (error)
528 			return (error);
529 
530 		switch (iflp.lagg_type) {
531 		case LAGG_TYPE_ETHERNET:
532 			if_type = IFT_ETHER;
533 			break;
534 		case LAGG_TYPE_INFINIBAND:
535 			if_type = IFT_INFINIBAND;
536 			break;
537 		default:
538 			return (EINVAL);
539 		}
540 	} else {
541 		if_type = IFT_ETHER;
542 	}
543 
544 	sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO);
545 	ifp = sc->sc_ifp = if_alloc(if_type);
546 	if (ifp == NULL) {
547 		free(sc, M_LAGG);
548 		return (ENOSPC);
549 	}
550 	LAGG_SX_INIT(sc);
551 
552 	mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
553 	callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
554 
555 	LAGG_XLOCK(sc);
556 	if (V_def_use_flowid)
557 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
558 	if (V_def_use_numa)
559 		sc->sc_opts |= LAGG_OPT_USE_NUMA;
560 	sc->flowid_shift = V_def_flowid_shift;
561 
562 	/* Hash all layers by default */
563 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
564 
565 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
566 
567 	CK_SLIST_INIT(&sc->sc_ports);
568 
569 	switch (if_type) {
570 	case IFT_ETHER:
571 		/* Initialise pseudo media types */
572 		ifmedia_init(&sc->sc_media, 0, lagg_media_change,
573 		    lagg_media_status);
574 		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
575 		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
576 
577 		if_initname(ifp, laggname, unit);
578 		ifp->if_transmit = lagg_transmit_ethernet;
579 		break;
580 	case IFT_INFINIBAND:
581 		if_initname(ifp, laggname, unit);
582 		ifp->if_transmit = lagg_transmit_infiniband;
583 		break;
584 	default:
585 		break;
586 	}
587 	ifp->if_softc = sc;
588 	ifp->if_qflush = lagg_qflush;
589 	ifp->if_init = lagg_init;
590 	ifp->if_ioctl = lagg_ioctl;
591 	ifp->if_get_counter = lagg_get_counter;
592 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
593 #if defined(KERN_TLS) || defined(RATELIMIT)
594 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
595 	ifp->if_snd_tag_modify = lagg_snd_tag_modify;
596 	ifp->if_snd_tag_query = lagg_snd_tag_query;
597 	ifp->if_snd_tag_free = lagg_snd_tag_free;
598 	ifp->if_ratelimit_query = lagg_ratelimit_query;
599 #endif
600 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
601 
602 	/*
603 	 * Attach as an ordinary ethernet device, children will be attached
604 	 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
605 	 */
606 	switch (if_type) {
607 	case IFT_ETHER:
608 		ether_ifattach(ifp, eaddr);
609 		break;
610 	case IFT_INFINIBAND:
611 		infiniband_ifattach(ifp, eaddr, ib_bcast_addr);
612 		break;
613 	default:
614 		break;
615 	}
616 
617 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
618 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
619 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
620 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
621 
622 	/* Insert into the global list of laggs */
623 	LAGG_LIST_LOCK();
624 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
625 	LAGG_LIST_UNLOCK();
626 	LAGG_XUNLOCK(sc);
627 
628 	return (0);
629 }
630 
631 static void
632 lagg_clone_destroy(struct ifnet *ifp)
633 {
634 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
635 	struct lagg_port *lp;
636 
637 	LAGG_XLOCK(sc);
638 	sc->sc_destroying = 1;
639 	lagg_stop(sc);
640 	ifp->if_flags &= ~IFF_UP;
641 
642 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
643 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
644 
645 	/* Shutdown and remove lagg ports */
646 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
647 		lagg_port_destroy(lp, 1);
648 
649 	/* Unhook the aggregation protocol */
650 	lagg_proto_detach(sc);
651 	LAGG_XUNLOCK(sc);
652 
653 	switch (ifp->if_type) {
654 	case IFT_ETHER:
655 		ifmedia_removeall(&sc->sc_media);
656 		ether_ifdetach(ifp);
657 		break;
658 	case IFT_INFINIBAND:
659 		infiniband_ifdetach(ifp);
660 		break;
661 	default:
662 		break;
663 	}
664 	if_free(ifp);
665 
666 	LAGG_LIST_LOCK();
667 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
668 	LAGG_LIST_UNLOCK();
669 
670 	mtx_destroy(&sc->sc_mtx);
671 	LAGG_SX_DESTROY(sc);
672 	free(sc, M_LAGG);
673 }
674 
675 static void
676 lagg_capabilities(struct lagg_softc *sc)
677 {
678 	struct lagg_port *lp;
679 	int cap, ena, pena;
680 	uint64_t hwa;
681 	struct ifnet_hw_tsomax hw_tsomax;
682 
683 	LAGG_XLOCK_ASSERT(sc);
684 
685 	/* Get common enabled capabilities for the lagg ports */
686 	ena = ~0;
687 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
688 		ena &= lp->lp_ifp->if_capenable;
689 	ena = (ena == ~0 ? 0 : ena);
690 
691 	/*
692 	 * Apply common enabled capabilities back to the lagg ports.
693 	 * May require several iterations if they are dependent.
694 	 */
695 	do {
696 		pena = ena;
697 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
698 			lagg_setcaps(lp, ena);
699 			ena &= lp->lp_ifp->if_capenable;
700 		}
701 	} while (pena != ena);
702 
703 	/* Get other capabilities from the lagg ports */
704 	cap = ~0;
705 	hwa = ~(uint64_t)0;
706 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
707 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
708 		cap &= lp->lp_ifp->if_capabilities;
709 		hwa &= lp->lp_ifp->if_hwassist;
710 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
711 	}
712 	cap = (cap == ~0 ? 0 : cap);
713 	hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
714 
715 	if (sc->sc_ifp->if_capabilities != cap ||
716 	    sc->sc_ifp->if_capenable != ena ||
717 	    sc->sc_ifp->if_hwassist != hwa ||
718 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
719 		sc->sc_ifp->if_capabilities = cap;
720 		sc->sc_ifp->if_capenable = ena;
721 		sc->sc_ifp->if_hwassist = hwa;
722 		getmicrotime(&sc->sc_ifp->if_lastchange);
723 
724 		if (sc->sc_ifflags & IFF_DEBUG)
725 			if_printf(sc->sc_ifp,
726 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
727 	}
728 }
729 
730 static int
731 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
732 {
733 	struct lagg_softc *sc_ptr;
734 	struct lagg_port *lp, *tlp;
735 	struct ifreq ifr;
736 	int error, i, oldmtu;
737 	int if_type;
738 	uint64_t *pval;
739 
740 	LAGG_XLOCK_ASSERT(sc);
741 
742 	if (sc->sc_ifp == ifp) {
743 		if_printf(sc->sc_ifp,
744 		    "cannot add a lagg to itself as a port\n");
745 		return (EINVAL);
746 	}
747 
748 	if (sc->sc_destroying == 1)
749 		return (ENXIO);
750 
751 	/* Limit the maximal number of lagg ports */
752 	if (sc->sc_count >= LAGG_MAX_PORTS)
753 		return (ENOSPC);
754 
755 	/* Check if port has already been associated to a lagg */
756 	if (ifp->if_lagg != NULL) {
757 		/* Port is already in the current lagg? */
758 		lp = (struct lagg_port *)ifp->if_lagg;
759 		if (lp->lp_softc == sc)
760 			return (EEXIST);
761 		return (EBUSY);
762 	}
763 
764 	switch (sc->sc_ifp->if_type) {
765 	case IFT_ETHER:
766 		/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
767 		if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
768 			return (EPROTONOSUPPORT);
769 		if_type = IFT_IEEE8023ADLAG;
770 		break;
771 	case IFT_INFINIBAND:
772 		/* XXX Disallow non-infiniband interfaces */
773 		if (ifp->if_type != IFT_INFINIBAND)
774 			return (EPROTONOSUPPORT);
775 		if_type = IFT_INFINIBANDLAG;
776 		break;
777 	default:
778 		break;
779 	}
780 
781 	/* Allow the first Ethernet member to define the MTU */
782 	oldmtu = -1;
783 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
784 		sc->sc_ifp->if_mtu = ifp->if_mtu;
785 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
786 		if (ifp->if_ioctl == NULL) {
787 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
788 			    ifp->if_xname);
789 			return (EINVAL);
790 		}
791 		oldmtu = ifp->if_mtu;
792 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
793 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
794 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
795 		if (error != 0) {
796 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
797 			    ifp->if_xname);
798 			return (error);
799 		}
800 		ifr.ifr_mtu = oldmtu;
801 	}
802 
803 	lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO);
804 	lp->lp_softc = sc;
805 
806 	/* Check if port is a stacked lagg */
807 	LAGG_LIST_LOCK();
808 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
809 		if (ifp == sc_ptr->sc_ifp) {
810 			LAGG_LIST_UNLOCK();
811 			free(lp, M_LAGG);
812 			if (oldmtu != -1)
813 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
814 				    (caddr_t)&ifr);
815 			return (EINVAL);
816 			/* XXX disable stacking for the moment, its untested */
817 #ifdef LAGG_PORT_STACKING
818 			lp->lp_flags |= LAGG_PORT_STACK;
819 			if (lagg_port_checkstacking(sc_ptr) >=
820 			    LAGG_MAX_STACKING) {
821 				LAGG_LIST_UNLOCK();
822 				free(lp, M_LAGG);
823 				if (oldmtu != -1)
824 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
825 					    (caddr_t)&ifr);
826 				return (E2BIG);
827 			}
828 #endif
829 		}
830 	}
831 	LAGG_LIST_UNLOCK();
832 
833 	if_ref(ifp);
834 	lp->lp_ifp = ifp;
835 
836 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
837 	lp->lp_ifcapenable = ifp->if_capenable;
838 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
839 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
840 		lagg_proto_lladdr(sc);
841 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
842 	} else {
843 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
844 	}
845 	lagg_setflags(lp, 1);
846 
847 	if (CK_SLIST_EMPTY(&sc->sc_ports))
848 		sc->sc_primary = lp;
849 
850 	/* Change the interface type */
851 	lp->lp_iftype = ifp->if_type;
852 	ifp->if_type = if_type;
853 	ifp->if_lagg = lp;
854 	lp->lp_ioctl = ifp->if_ioctl;
855 	ifp->if_ioctl = lagg_port_ioctl;
856 	lp->lp_output = ifp->if_output;
857 	ifp->if_output = lagg_port_output;
858 
859 	/* Read port counters */
860 	pval = lp->port_counters.val;
861 	for (i = 0; i < IFCOUNTERS; i++, pval++)
862 		*pval = ifp->if_get_counter(ifp, i);
863 
864 	/*
865 	 * Insert into the list of ports.
866 	 * Keep ports sorted by if_index. It is handy, when configuration
867 	 * is predictable and `ifconfig laggN create ...` command
868 	 * will lead to the same result each time.
869 	 */
870 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
871 		if (tlp->lp_ifp->if_index < ifp->if_index && (
872 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
873 		    ((struct  lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
874 		    ifp->if_index))
875 			break;
876 	}
877 	if (tlp != NULL)
878 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
879 	else
880 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
881 	sc->sc_count++;
882 
883 	lagg_setmulti(lp);
884 
885 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
886 		/* Remove the port, without calling pr_delport. */
887 		lagg_port_destroy(lp, 0);
888 		if (oldmtu != -1)
889 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
890 		return (error);
891 	}
892 
893 	/* Update lagg capabilities */
894 	lagg_capabilities(sc);
895 	lagg_linkstate(sc);
896 
897 	return (0);
898 }
899 
900 #ifdef LAGG_PORT_STACKING
901 static int
902 lagg_port_checkstacking(struct lagg_softc *sc)
903 {
904 	struct lagg_softc *sc_ptr;
905 	struct lagg_port *lp;
906 	int m = 0;
907 
908 	LAGG_SXLOCK_ASSERT(sc);
909 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
910 		if (lp->lp_flags & LAGG_PORT_STACK) {
911 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
912 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
913 		}
914 	}
915 
916 	return (m + 1);
917 }
918 #endif
919 
920 static void
921 lagg_port_destroy_cb(epoch_context_t ec)
922 {
923 	struct lagg_port *lp;
924 	struct ifnet *ifp;
925 
926 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
927 	ifp = lp->lp_ifp;
928 
929 	if_rele(ifp);
930 	free(lp, M_LAGG);
931 }
932 
933 static int
934 lagg_port_destroy(struct lagg_port *lp, int rundelport)
935 {
936 	struct lagg_softc *sc = lp->lp_softc;
937 	struct lagg_port *lp_ptr, *lp0;
938 	struct ifnet *ifp = lp->lp_ifp;
939 	uint64_t *pval, vdiff;
940 	int i;
941 
942 	LAGG_XLOCK_ASSERT(sc);
943 
944 	if (rundelport)
945 		lagg_proto_delport(sc, lp);
946 
947 	if (lp->lp_detaching == 0)
948 		lagg_clrmulti(lp);
949 
950 	/* Restore interface */
951 	ifp->if_type = lp->lp_iftype;
952 	ifp->if_ioctl = lp->lp_ioctl;
953 	ifp->if_output = lp->lp_output;
954 	ifp->if_lagg = NULL;
955 
956 	/* Update detached port counters */
957 	pval = lp->port_counters.val;
958 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
959 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
960 		sc->detached_counters.val[i] += vdiff;
961 	}
962 
963 	/* Finally, remove the port from the lagg */
964 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
965 	sc->sc_count--;
966 
967 	/* Update the primary interface */
968 	if (lp == sc->sc_primary) {
969 		uint8_t lladdr[LAGG_ADDR_LEN];
970 
971 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
972 			bzero(&lladdr, LAGG_ADDR_LEN);
973 		else
974 			bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
975 		sc->sc_primary = lp0;
976 		if (sc->sc_destroying == 0) {
977 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
978 			lagg_proto_lladdr(sc);
979 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
980 		}
981 
982 		/*
983 		 * Update lladdr for each port (new primary needs update
984 		 * as well, to switch from old lladdr to its 'real' one)
985 		 */
986 		CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
987 			if_setlladdr(lp_ptr->lp_ifp, lladdr, lp_ptr->lp_ifp->if_addrlen);
988 	}
989 
990 	if (lp->lp_ifflags)
991 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
992 
993 	if (lp->lp_detaching == 0) {
994 		lagg_setflags(lp, 0);
995 		lagg_setcaps(lp, lp->lp_ifcapenable);
996 		if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
997 	}
998 
999 	/*
1000 	 * free port and release it's ifnet reference after a grace period has
1001 	 * elapsed.
1002 	 */
1003 	NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
1004 	/* Update lagg capabilities */
1005 	lagg_capabilities(sc);
1006 	lagg_linkstate(sc);
1007 
1008 	return (0);
1009 }
1010 
1011 static int
1012 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1013 {
1014 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
1015 	struct lagg_softc *sc;
1016 	struct lagg_port *lp = NULL;
1017 	int error = 0;
1018 
1019 	/* Should be checked by the caller */
1020 	switch (ifp->if_type) {
1021 	case IFT_IEEE8023ADLAG:
1022 	case IFT_INFINIBANDLAG:
1023 		if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1024 			goto fallback;
1025 		break;
1026 	default:
1027 		goto fallback;
1028 	}
1029 
1030 	switch (cmd) {
1031 	case SIOCGLAGGPORT:
1032 		if (rp->rp_portname[0] == '\0' ||
1033 		    ifunit(rp->rp_portname) != ifp) {
1034 			error = EINVAL;
1035 			break;
1036 		}
1037 
1038 		LAGG_RLOCK();
1039 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1040 			error = ENOENT;
1041 			LAGG_RUNLOCK();
1042 			break;
1043 		}
1044 
1045 		lagg_port2req(lp, rp);
1046 		LAGG_RUNLOCK();
1047 		break;
1048 
1049 	case SIOCSIFCAP:
1050 		if (lp->lp_ioctl == NULL) {
1051 			error = EINVAL;
1052 			break;
1053 		}
1054 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1055 		if (error)
1056 			break;
1057 
1058 		/* Update lagg interface capabilities */
1059 		LAGG_XLOCK(sc);
1060 		lagg_capabilities(sc);
1061 		LAGG_XUNLOCK(sc);
1062 		VLAN_CAPABILITIES(sc->sc_ifp);
1063 		break;
1064 
1065 	case SIOCSIFMTU:
1066 		/* Do not allow the MTU to be changed once joined */
1067 		error = EINVAL;
1068 		break;
1069 
1070 	default:
1071 		goto fallback;
1072 	}
1073 
1074 	return (error);
1075 
1076 fallback:
1077 	if (lp != NULL && lp->lp_ioctl != NULL)
1078 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1079 
1080 	return (EINVAL);
1081 }
1082 
1083 /*
1084  * Requests counter @cnt data.
1085  *
1086  * Counter value is calculated the following way:
1087  * 1) for each port, sum  difference between current and "initial" measurements.
1088  * 2) add lagg logical interface counters.
1089  * 3) add data from detached_counters array.
1090  *
1091  * We also do the following things on ports attach/detach:
1092  * 1) On port attach we store all counters it has into port_counter array.
1093  * 2) On port detach we add the different between "initial" and
1094  *   current counters data to detached_counters array.
1095  */
1096 static uint64_t
1097 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1098 {
1099 	struct lagg_softc *sc;
1100 	struct lagg_port *lp;
1101 	struct ifnet *lpifp;
1102 	uint64_t newval, oldval, vsum;
1103 
1104 	/* Revise this when we've got non-generic counters. */
1105 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1106 
1107 	sc = (struct lagg_softc *)ifp->if_softc;
1108 
1109 	vsum = 0;
1110 	LAGG_RLOCK();
1111 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1112 		/* Saved attached value */
1113 		oldval = lp->port_counters.val[cnt];
1114 		/* current value */
1115 		lpifp = lp->lp_ifp;
1116 		newval = lpifp->if_get_counter(lpifp, cnt);
1117 		/* Calculate diff and save new */
1118 		vsum += newval - oldval;
1119 	}
1120 	LAGG_RUNLOCK();
1121 
1122 	/*
1123 	 * Add counter data which might be added by upper
1124 	 * layer protocols operating on logical interface.
1125 	 */
1126 	vsum += if_get_counter_default(ifp, cnt);
1127 
1128 	/*
1129 	 * Add counter data from detached ports counters
1130 	 */
1131 	vsum += sc->detached_counters.val[cnt];
1132 
1133 	return (vsum);
1134 }
1135 
1136 /*
1137  * For direct output to child ports.
1138  */
1139 static int
1140 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1141 	const struct sockaddr *dst, struct route *ro)
1142 {
1143 	struct lagg_port *lp = ifp->if_lagg;
1144 
1145 	switch (dst->sa_family) {
1146 		case pseudo_AF_HDRCMPLT:
1147 		case AF_UNSPEC:
1148 			return ((*lp->lp_output)(ifp, m, dst, ro));
1149 	}
1150 
1151 	/* drop any other frames */
1152 	m_freem(m);
1153 	return (ENETDOWN);
1154 }
1155 
1156 static void
1157 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1158 {
1159 	struct lagg_port *lp;
1160 	struct lagg_softc *sc;
1161 
1162 	if ((lp = ifp->if_lagg) == NULL)
1163 		return;
1164 	/* If the ifnet is just being renamed, don't do anything. */
1165 	if (ifp->if_flags & IFF_RENAMING)
1166 		return;
1167 
1168 	sc = lp->lp_softc;
1169 
1170 	LAGG_XLOCK(sc);
1171 	lp->lp_detaching = 1;
1172 	lagg_port_destroy(lp, 1);
1173 	LAGG_XUNLOCK(sc);
1174 	VLAN_CAPABILITIES(sc->sc_ifp);
1175 }
1176 
1177 static void
1178 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1179 {
1180 	struct lagg_softc *sc = lp->lp_softc;
1181 
1182 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1183 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1184 	rp->rp_prio = lp->lp_prio;
1185 	rp->rp_flags = lp->lp_flags;
1186 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1187 
1188 	/* Add protocol specific flags */
1189 	switch (sc->sc_proto) {
1190 		case LAGG_PROTO_FAILOVER:
1191 			if (lp == sc->sc_primary)
1192 				rp->rp_flags |= LAGG_PORT_MASTER;
1193 			if (lp == lagg_link_active(sc, sc->sc_primary))
1194 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1195 			break;
1196 
1197 		case LAGG_PROTO_ROUNDROBIN:
1198 		case LAGG_PROTO_LOADBALANCE:
1199 		case LAGG_PROTO_BROADCAST:
1200 			if (LAGG_PORTACTIVE(lp))
1201 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1202 			break;
1203 
1204 		case LAGG_PROTO_LACP:
1205 			/* LACP has a different definition of active */
1206 			if (lacp_isactive(lp))
1207 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1208 			if (lacp_iscollecting(lp))
1209 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1210 			if (lacp_isdistributing(lp))
1211 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1212 			break;
1213 	}
1214 
1215 }
1216 
1217 static void
1218 lagg_watchdog_infiniband(void *arg)
1219 {
1220 	struct lagg_softc *sc;
1221 	struct lagg_port *lp;
1222 	struct ifnet *ifp;
1223 	struct ifnet *lp_ifp;
1224 
1225 	sc = arg;
1226 
1227 	/*
1228 	 * Because infiniband nodes have a fixed MAC address, which is
1229 	 * generated by the so-called GID, we need to regularly update
1230 	 * the link level address of the parent lagg<N> device when
1231 	 * the active port changes. Possibly we could piggy-back on
1232 	 * link up/down events aswell, but using a timer also provides
1233 	 * a guarantee against too frequent events. This operation
1234 	 * does not have to be atomic.
1235 	 */
1236 	LAGG_RLOCK();
1237 	lp = lagg_link_active(sc, sc->sc_primary);
1238 	if (lp != NULL) {
1239 		ifp = sc->sc_ifp;
1240 		lp_ifp = lp->lp_ifp;
1241 
1242 		if (ifp != NULL && lp_ifp != NULL &&
1243 		    memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0) {
1244 			memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1245 			CURVNET_SET(ifp->if_vnet);
1246 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1247 			CURVNET_RESTORE();
1248 		}
1249 	}
1250 	LAGG_RUNLOCK();
1251 
1252 	callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1253 }
1254 
1255 static void
1256 lagg_init(void *xsc)
1257 {
1258 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1259 	struct ifnet *ifp = sc->sc_ifp;
1260 	struct lagg_port *lp;
1261 
1262 	LAGG_XLOCK(sc);
1263 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1264 		LAGG_XUNLOCK(sc);
1265 		return;
1266 	}
1267 
1268 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1269 
1270 	/*
1271 	 * Update the port lladdrs if needed.
1272 	 * This might be if_setlladdr() notification
1273 	 * that lladdr has been changed.
1274 	 */
1275 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1276 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1277 		    ifp->if_addrlen) != 0)
1278 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1279 	}
1280 
1281 	lagg_proto_init(sc);
1282 
1283 	if (ifp->if_type == IFT_INFINIBAND) {
1284 		mtx_lock(&sc->sc_mtx);
1285 		lagg_watchdog_infiniband(sc);
1286 		mtx_unlock(&sc->sc_mtx);
1287 	}
1288 
1289 	LAGG_XUNLOCK(sc);
1290 }
1291 
1292 static void
1293 lagg_stop(struct lagg_softc *sc)
1294 {
1295 	struct ifnet *ifp = sc->sc_ifp;
1296 
1297 	LAGG_XLOCK_ASSERT(sc);
1298 
1299 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1300 		return;
1301 
1302 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1303 
1304 	lagg_proto_stop(sc);
1305 
1306 	mtx_lock(&sc->sc_mtx);
1307 	callout_stop(&sc->sc_watchdog);
1308 	mtx_unlock(&sc->sc_mtx);
1309 
1310 	callout_drain(&sc->sc_watchdog);
1311 }
1312 
1313 static int
1314 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1315 {
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_UNLOCK_ASSERT();
1375 		lagg_proto_attach(sc, ra->ra_proto);
1376 		LAGG_XUNLOCK(sc);
1377 		break;
1378 	case SIOCGLAGGOPTS:
1379 		LAGG_XLOCK(sc);
1380 		ro->ro_opts = sc->sc_opts;
1381 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1382 			struct lacp_softc *lsc;
1383 
1384 			lsc = (struct lacp_softc *)sc->sc_psc;
1385 			if (lsc->lsc_debug.lsc_tx_test != 0)
1386 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1387 			if (lsc->lsc_debug.lsc_rx_test != 0)
1388 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1389 			if (lsc->lsc_strict_mode != 0)
1390 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1391 			if (lsc->lsc_fast_timeout != 0)
1392 				ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1393 
1394 			ro->ro_active = sc->sc_active;
1395 		} else {
1396 			ro->ro_active = 0;
1397 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1398 				ro->ro_active += LAGG_PORTACTIVE(lp);
1399 		}
1400 		ro->ro_bkt = sc->sc_stride;
1401 		ro->ro_flapping = sc->sc_flapping;
1402 		ro->ro_flowid_shift = sc->flowid_shift;
1403 		LAGG_XUNLOCK(sc);
1404 		break;
1405 	case SIOCSLAGGOPTS:
1406 		error = priv_check(td, PRIV_NET_LAGG);
1407 		if (error)
1408 			break;
1409 
1410 		/*
1411 		 * The stride option was added without defining a corresponding
1412 		 * LAGG_OPT flag, so handle a non-zero value before checking
1413 		 * anything else to preserve compatibility.
1414 		 */
1415 		LAGG_XLOCK(sc);
1416 		if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1417 			if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1418 				LAGG_XUNLOCK(sc);
1419 				error = EINVAL;
1420 				break;
1421 			}
1422 			sc->sc_stride = ro->ro_bkt;
1423 		}
1424 		if (ro->ro_opts == 0) {
1425 			LAGG_XUNLOCK(sc);
1426 			break;
1427 		}
1428 
1429 		/*
1430 		 * Set options.  LACP options are stored in sc->sc_psc,
1431 		 * not in sc_opts.
1432 		 */
1433 		int valid, lacp;
1434 
1435 		switch (ro->ro_opts) {
1436 		case LAGG_OPT_USE_FLOWID:
1437 		case -LAGG_OPT_USE_FLOWID:
1438 		case LAGG_OPT_USE_NUMA:
1439 		case -LAGG_OPT_USE_NUMA:
1440 		case LAGG_OPT_FLOWIDSHIFT:
1441 		case LAGG_OPT_RR_LIMIT:
1442 			valid = 1;
1443 			lacp = 0;
1444 			break;
1445 		case LAGG_OPT_LACP_TXTEST:
1446 		case -LAGG_OPT_LACP_TXTEST:
1447 		case LAGG_OPT_LACP_RXTEST:
1448 		case -LAGG_OPT_LACP_RXTEST:
1449 		case LAGG_OPT_LACP_STRICT:
1450 		case -LAGG_OPT_LACP_STRICT:
1451 		case LAGG_OPT_LACP_FAST_TIMO:
1452 		case -LAGG_OPT_LACP_FAST_TIMO:
1453 			valid = lacp = 1;
1454 			break;
1455 		default:
1456 			valid = lacp = 0;
1457 			break;
1458 		}
1459 
1460 		if (valid == 0 ||
1461 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1462 			/* Invalid combination of options specified. */
1463 			error = EINVAL;
1464 			LAGG_XUNLOCK(sc);
1465 			break;	/* Return from SIOCSLAGGOPTS. */
1466 		}
1467 
1468 		/*
1469 		 * Store new options into sc->sc_opts except for
1470 		 * FLOWIDSHIFT, RR and LACP options.
1471 		 */
1472 		if (lacp == 0) {
1473 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1474 				sc->flowid_shift = ro->ro_flowid_shift;
1475 			else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1476 				if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1477 				    ro->ro_bkt == 0) {
1478 					error = EINVAL;
1479 					LAGG_XUNLOCK(sc);
1480 					break;
1481 				}
1482 				sc->sc_stride = ro->ro_bkt;
1483 			} else if (ro->ro_opts > 0)
1484 				sc->sc_opts |= ro->ro_opts;
1485 			else
1486 				sc->sc_opts &= ~ro->ro_opts;
1487 		} else {
1488 			struct lacp_softc *lsc;
1489 			struct lacp_port *lp;
1490 
1491 			lsc = (struct lacp_softc *)sc->sc_psc;
1492 
1493 			switch (ro->ro_opts) {
1494 			case LAGG_OPT_LACP_TXTEST:
1495 				lsc->lsc_debug.lsc_tx_test = 1;
1496 				break;
1497 			case -LAGG_OPT_LACP_TXTEST:
1498 				lsc->lsc_debug.lsc_tx_test = 0;
1499 				break;
1500 			case LAGG_OPT_LACP_RXTEST:
1501 				lsc->lsc_debug.lsc_rx_test = 1;
1502 				break;
1503 			case -LAGG_OPT_LACP_RXTEST:
1504 				lsc->lsc_debug.lsc_rx_test = 0;
1505 				break;
1506 			case LAGG_OPT_LACP_STRICT:
1507 				lsc->lsc_strict_mode = 1;
1508 				break;
1509 			case -LAGG_OPT_LACP_STRICT:
1510 				lsc->lsc_strict_mode = 0;
1511 				break;
1512 			case LAGG_OPT_LACP_FAST_TIMO:
1513 				LACP_LOCK(lsc);
1514         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1515                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1516 				LACP_UNLOCK(lsc);
1517 				lsc->lsc_fast_timeout = 1;
1518 				break;
1519 			case -LAGG_OPT_LACP_FAST_TIMO:
1520 				LACP_LOCK(lsc);
1521         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1522                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1523 				LACP_UNLOCK(lsc);
1524 				lsc->lsc_fast_timeout = 0;
1525 				break;
1526 			}
1527 		}
1528 		LAGG_XUNLOCK(sc);
1529 		break;
1530 	case SIOCGLAGGFLAGS:
1531 		rf->rf_flags = 0;
1532 		LAGG_XLOCK(sc);
1533 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1534 			rf->rf_flags |= LAGG_F_HASHL2;
1535 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1536 			rf->rf_flags |= LAGG_F_HASHL3;
1537 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1538 			rf->rf_flags |= LAGG_F_HASHL4;
1539 		LAGG_XUNLOCK(sc);
1540 		break;
1541 	case SIOCSLAGGHASH:
1542 		error = priv_check(td, PRIV_NET_LAGG);
1543 		if (error)
1544 			break;
1545 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1546 			error = EINVAL;
1547 			break;
1548 		}
1549 		LAGG_XLOCK(sc);
1550 		sc->sc_flags = 0;
1551 		if (rf->rf_flags & LAGG_F_HASHL2)
1552 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1553 		if (rf->rf_flags & LAGG_F_HASHL3)
1554 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1555 		if (rf->rf_flags & LAGG_F_HASHL4)
1556 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1557 		LAGG_XUNLOCK(sc);
1558 		break;
1559 	case SIOCGLAGGPORT:
1560 		if (rp->rp_portname[0] == '\0' ||
1561 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1562 			error = EINVAL;
1563 			break;
1564 		}
1565 
1566 		LAGG_RLOCK();
1567 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1568 		    lp->lp_softc != sc) {
1569 			error = ENOENT;
1570 			LAGG_RUNLOCK();
1571 			if_rele(tpif);
1572 			break;
1573 		}
1574 
1575 		lagg_port2req(lp, rp);
1576 		LAGG_RUNLOCK();
1577 		if_rele(tpif);
1578 		break;
1579 	case SIOCSLAGGPORT:
1580 		error = priv_check(td, PRIV_NET_LAGG);
1581 		if (error)
1582 			break;
1583 		if (rp->rp_portname[0] == '\0' ||
1584 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1585 			error = EINVAL;
1586 			break;
1587 		}
1588 #ifdef INET6
1589 		/*
1590 		 * A laggport interface should not have inet6 address
1591 		 * because two interfaces with a valid link-local
1592 		 * scope zone must not be merged in any form.  This
1593 		 * restriction is needed to prevent violation of
1594 		 * link-local scope zone.  Attempts to add a laggport
1595 		 * interface which has inet6 addresses triggers
1596 		 * removal of all inet6 addresses on the member
1597 		 * interface.
1598 		 */
1599 		if (in6ifa_llaonifp(tpif)) {
1600 			in6_ifdetach(tpif);
1601 				if_printf(sc->sc_ifp,
1602 				    "IPv6 addresses on %s have been removed "
1603 				    "before adding it as a member to prevent "
1604 				    "IPv6 address scope violation.\n",
1605 				    tpif->if_xname);
1606 		}
1607 #endif
1608 		oldmtu = ifp->if_mtu;
1609 		LAGG_XLOCK(sc);
1610 		error = lagg_port_create(sc, tpif);
1611 		LAGG_XUNLOCK(sc);
1612 		if_rele(tpif);
1613 
1614 		/*
1615 		 * LAGG MTU may change during addition of the first port.
1616 		 * If it did, do network layer specific procedure.
1617 		 */
1618 		if (ifp->if_mtu != oldmtu) {
1619 #ifdef INET6
1620 			nd6_setmtu(ifp);
1621 #endif
1622 			rt_updatemtu(ifp);
1623 		}
1624 
1625 		VLAN_CAPABILITIES(ifp);
1626 		break;
1627 	case SIOCSLAGGDELPORT:
1628 		error = priv_check(td, PRIV_NET_LAGG);
1629 		if (error)
1630 			break;
1631 		if (rp->rp_portname[0] == '\0' ||
1632 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1633 			error = EINVAL;
1634 			break;
1635 		}
1636 
1637 		LAGG_XLOCK(sc);
1638 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1639 		    lp->lp_softc != sc) {
1640 			error = ENOENT;
1641 			LAGG_XUNLOCK(sc);
1642 			if_rele(tpif);
1643 			break;
1644 		}
1645 
1646 		error = lagg_port_destroy(lp, 1);
1647 		LAGG_XUNLOCK(sc);
1648 		if_rele(tpif);
1649 		VLAN_CAPABILITIES(ifp);
1650 		break;
1651 	case SIOCSIFFLAGS:
1652 		/* Set flags on ports too */
1653 		LAGG_XLOCK(sc);
1654 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1655 			lagg_setflags(lp, 1);
1656 		}
1657 
1658 		if (!(ifp->if_flags & IFF_UP) &&
1659 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1660 			/*
1661 			 * If interface is marked down and it is running,
1662 			 * then stop and disable it.
1663 			 */
1664 			lagg_stop(sc);
1665 			LAGG_XUNLOCK(sc);
1666 		} else if ((ifp->if_flags & IFF_UP) &&
1667 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1668 			/*
1669 			 * If interface is marked up and it is stopped, then
1670 			 * start it.
1671 			 */
1672 			LAGG_XUNLOCK(sc);
1673 			(*ifp->if_init)(sc);
1674 		} else
1675 			LAGG_XUNLOCK(sc);
1676 		break;
1677 	case SIOCADDMULTI:
1678 	case SIOCDELMULTI:
1679 		LAGG_XLOCK(sc);
1680 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1681 			lagg_clrmulti(lp);
1682 			lagg_setmulti(lp);
1683 		}
1684 		LAGG_XUNLOCK(sc);
1685 		error = 0;
1686 		break;
1687 	case SIOCSIFMEDIA:
1688 	case SIOCGIFMEDIA:
1689 		if (ifp->if_type == IFT_INFINIBAND)
1690 			error = EINVAL;
1691 		else
1692 			error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1693 		break;
1694 
1695 	case SIOCSIFCAP:
1696 		LAGG_XLOCK(sc);
1697 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1698 			if (lp->lp_ioctl != NULL)
1699 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1700 		}
1701 		lagg_capabilities(sc);
1702 		LAGG_XUNLOCK(sc);
1703 		VLAN_CAPABILITIES(ifp);
1704 		error = 0;
1705 		break;
1706 
1707 	case SIOCSIFMTU:
1708 		LAGG_XLOCK(sc);
1709 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1710 			if (lp->lp_ioctl != NULL)
1711 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1712 			else
1713 				error = EINVAL;
1714 			if (error != 0) {
1715 				if_printf(ifp,
1716 				    "failed to change MTU to %d on port %s, "
1717 				    "reverting all ports to original MTU (%d)\n",
1718 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1719 				break;
1720 			}
1721 		}
1722 		if (error == 0) {
1723 			ifp->if_mtu = ifr->ifr_mtu;
1724 		} else {
1725 			/* set every port back to the original MTU */
1726 			ifr->ifr_mtu = ifp->if_mtu;
1727 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1728 				if (lp->lp_ioctl != NULL)
1729 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1730 			}
1731 		}
1732 		LAGG_XUNLOCK(sc);
1733 		break;
1734 
1735 	default:
1736 		error = ether_ioctl(ifp, cmd, data);
1737 		break;
1738 	}
1739 	return (error);
1740 }
1741 
1742 #if defined(KERN_TLS) || defined(RATELIMIT)
1743 static inline struct lagg_snd_tag *
1744 mst_to_lst(struct m_snd_tag *mst)
1745 {
1746 
1747 	return (__containerof(mst, struct lagg_snd_tag, com));
1748 }
1749 
1750 /*
1751  * Look up the port used by a specific flow.  This only works for lagg
1752  * protocols with deterministic port mappings (e.g. not roundrobin).
1753  * In addition protocols which use a hash to map flows to ports must
1754  * be configured to use the mbuf flowid rather than hashing packet
1755  * contents.
1756  */
1757 static struct lagg_port *
1758 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1759     uint8_t numa_domain)
1760 {
1761 	struct lagg_softc *sc;
1762 	struct lagg_port *lp;
1763 	struct lagg_lb *lb;
1764 	uint32_t hash, p;
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));
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 lagg_snd_tag *lst;
1797 	struct lagg_softc *sc;
1798 	struct lagg_port *lp;
1799 	struct ifnet *lp_ifp;
1800 	int error;
1801 
1802 	sc = ifp->if_softc;
1803 
1804 	LAGG_RLOCK();
1805 	lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1806 	    params->hdr.flowtype, params->hdr.numa_domain);
1807 	if (lp == NULL) {
1808 		LAGG_RUNLOCK();
1809 		return (EOPNOTSUPP);
1810 	}
1811 	if (lp->lp_ifp == NULL) {
1812 		LAGG_RUNLOCK();
1813 		return (EOPNOTSUPP);
1814 	}
1815 	lp_ifp = lp->lp_ifp;
1816 	if_ref(lp_ifp);
1817 	LAGG_RUNLOCK();
1818 
1819 	lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1820 	if (lst == NULL) {
1821 		if_rele(lp_ifp);
1822 		return (ENOMEM);
1823 	}
1824 
1825 	error = m_snd_tag_alloc(lp_ifp, params, &lst->tag);
1826 	if_rele(lp_ifp);
1827 	if (error) {
1828 		free(lst, M_LAGG);
1829 		return (error);
1830 	}
1831 
1832 	m_snd_tag_init(&lst->com, ifp, lst->tag->type);
1833 
1834 	*ppmt = &lst->com;
1835 	return (0);
1836 }
1837 
1838 static int
1839 lagg_snd_tag_modify(struct m_snd_tag *mst,
1840     union if_snd_tag_modify_params *params)
1841 {
1842 	struct lagg_snd_tag *lst;
1843 
1844 	lst = mst_to_lst(mst);
1845 	return (lst->tag->ifp->if_snd_tag_modify(lst->tag, params));
1846 }
1847 
1848 static int
1849 lagg_snd_tag_query(struct m_snd_tag *mst,
1850     union if_snd_tag_query_params *params)
1851 {
1852 	struct lagg_snd_tag *lst;
1853 
1854 	lst = mst_to_lst(mst);
1855 	return (lst->tag->ifp->if_snd_tag_query(lst->tag, params));
1856 }
1857 
1858 static void
1859 lagg_snd_tag_free(struct m_snd_tag *mst)
1860 {
1861 	struct lagg_snd_tag *lst;
1862 
1863 	lst = mst_to_lst(mst);
1864 	m_snd_tag_rele(lst->tag);
1865 	free(lst, M_LAGG);
1866 }
1867 
1868 static void
1869 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1870 {
1871 	/*
1872 	 * For lagg, we have an indirect
1873 	 * interface. The caller needs to
1874 	 * get a ratelimit tag on the actual
1875 	 * interface the flow will go on.
1876 	 */
1877 	q->rate_table = NULL;
1878 	q->flags = RT_IS_INDIRECT;
1879 	q->max_flows = 0;
1880 	q->number_of_rates = 0;
1881 }
1882 #endif
1883 
1884 static int
1885 lagg_setmulti(struct lagg_port *lp)
1886 {
1887 	struct lagg_softc *sc = lp->lp_softc;
1888 	struct ifnet *ifp = lp->lp_ifp;
1889 	struct ifnet *scifp = sc->sc_ifp;
1890 	struct lagg_mc *mc;
1891 	struct ifmultiaddr *ifma;
1892 	int error;
1893 
1894 	IF_ADDR_WLOCK(scifp);
1895 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1896 		if (ifma->ifma_addr->sa_family != AF_LINK)
1897 			continue;
1898 		mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
1899 		if (mc == NULL) {
1900 			IF_ADDR_WUNLOCK(scifp);
1901 			return (ENOMEM);
1902 		}
1903 		bcopy(ifma->ifma_addr, &mc->mc_addr,
1904 		    ifma->ifma_addr->sa_len);
1905 		mc->mc_addr.sdl_index = ifp->if_index;
1906 		mc->mc_ifma = NULL;
1907 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1908 	}
1909 	IF_ADDR_WUNLOCK(scifp);
1910 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1911 		error = if_addmulti(ifp,
1912 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1913 		if (error)
1914 			return (error);
1915 	}
1916 	return (0);
1917 }
1918 
1919 static int
1920 lagg_clrmulti(struct lagg_port *lp)
1921 {
1922 	struct lagg_mc *mc;
1923 
1924 	LAGG_XLOCK_ASSERT(lp->lp_softc);
1925 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1926 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1927 		if (mc->mc_ifma && lp->lp_detaching == 0)
1928 			if_delmulti_ifma(mc->mc_ifma);
1929 		free(mc, M_LAGG);
1930 	}
1931 	return (0);
1932 }
1933 
1934 static int
1935 lagg_setcaps(struct lagg_port *lp, int cap)
1936 {
1937 	struct ifreq ifr;
1938 
1939 	if (lp->lp_ifp->if_capenable == cap)
1940 		return (0);
1941 	if (lp->lp_ioctl == NULL)
1942 		return (ENXIO);
1943 	ifr.ifr_reqcap = cap;
1944 	return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1945 }
1946 
1947 /* Handle a ref counted flag that should be set on the lagg port as well */
1948 static int
1949 lagg_setflag(struct lagg_port *lp, int flag, int status,
1950     int (*func)(struct ifnet *, int))
1951 {
1952 	struct lagg_softc *sc = lp->lp_softc;
1953 	struct ifnet *scifp = sc->sc_ifp;
1954 	struct ifnet *ifp = lp->lp_ifp;
1955 	int error;
1956 
1957 	LAGG_XLOCK_ASSERT(sc);
1958 
1959 	status = status ? (scifp->if_flags & flag) : 0;
1960 	/* Now "status" contains the flag value or 0 */
1961 
1962 	/*
1963 	 * See if recorded ports status is different from what
1964 	 * we want it to be.  If it is, flip it.  We record ports
1965 	 * status in lp_ifflags so that we won't clear ports flag
1966 	 * we haven't set.  In fact, we don't clear or set ports
1967 	 * flags directly, but get or release references to them.
1968 	 * That's why we can be sure that recorded flags still are
1969 	 * in accord with actual ports flags.
1970 	 */
1971 	if (status != (lp->lp_ifflags & flag)) {
1972 		error = (*func)(ifp, status);
1973 		if (error)
1974 			return (error);
1975 		lp->lp_ifflags &= ~flag;
1976 		lp->lp_ifflags |= status;
1977 	}
1978 	return (0);
1979 }
1980 
1981 /*
1982  * Handle IFF_* flags that require certain changes on the lagg port
1983  * if "status" is true, update ports flags respective to the lagg
1984  * if "status" is false, forcedly clear the flags set on port.
1985  */
1986 static int
1987 lagg_setflags(struct lagg_port *lp, int status)
1988 {
1989 	int error, i;
1990 
1991 	for (i = 0; lagg_pflags[i].flag; i++) {
1992 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1993 		    status, lagg_pflags[i].func);
1994 		if (error)
1995 			return (error);
1996 	}
1997 	return (0);
1998 }
1999 
2000 static int
2001 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2002 {
2003 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2004 	int error;
2005 
2006 #if defined(KERN_TLS) || defined(RATELIMIT)
2007 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2008 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2009 #endif
2010 	LAGG_RLOCK();
2011 	/* We need a Tx algorithm and at least one port */
2012 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2013 		LAGG_RUNLOCK();
2014 		m_freem(m);
2015 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2016 		return (ENXIO);
2017 	}
2018 
2019 	ETHER_BPF_MTAP(ifp, m);
2020 
2021 	error = lagg_proto_start(sc, m);
2022 	LAGG_RUNLOCK();
2023 	return (error);
2024 }
2025 
2026 static int
2027 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2028 {
2029 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2030 	int error;
2031 
2032 #if defined(KERN_TLS) || defined(RATELIMIT)
2033 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2034 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2035 #endif
2036 	LAGG_RLOCK();
2037 	/* We need a Tx algorithm and at least one port */
2038 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2039 		LAGG_RUNLOCK();
2040 		m_freem(m);
2041 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2042 		return (ENXIO);
2043 	}
2044 
2045 	INFINIBAND_BPF_MTAP(ifp, m);
2046 
2047 	error = lagg_proto_start(sc, m);
2048 	LAGG_RUNLOCK();
2049 	return (error);
2050 }
2051 
2052 /*
2053  * The ifp->if_qflush entry point for lagg(4) is no-op.
2054  */
2055 static void
2056 lagg_qflush(struct ifnet *ifp __unused)
2057 {
2058 }
2059 
2060 static struct mbuf *
2061 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2062 {
2063 	struct lagg_port *lp = ifp->if_lagg;
2064 	struct lagg_softc *sc = lp->lp_softc;
2065 	struct ifnet *scifp = sc->sc_ifp;
2066 
2067 	LAGG_RLOCK();
2068 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2069 	    lp->lp_detaching != 0 ||
2070 	    sc->sc_proto == LAGG_PROTO_NONE) {
2071 		LAGG_RUNLOCK();
2072 		m_freem(m);
2073 		return (NULL);
2074 	}
2075 
2076 	ETHER_BPF_MTAP(scifp, m);
2077 
2078 	m = lagg_proto_input(sc, lp, m);
2079 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2080 		m_freem(m);
2081 		m = NULL;
2082 	}
2083 
2084 	LAGG_RUNLOCK();
2085 	return (m);
2086 }
2087 
2088 static struct mbuf *
2089 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2090 {
2091 	struct lagg_port *lp = ifp->if_lagg;
2092 	struct lagg_softc *sc = lp->lp_softc;
2093 	struct ifnet *scifp = sc->sc_ifp;
2094 
2095 	LAGG_RLOCK();
2096 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2097 	    lp->lp_detaching != 0 ||
2098 	    sc->sc_proto == LAGG_PROTO_NONE) {
2099 		LAGG_RUNLOCK();
2100 		m_freem(m);
2101 		return (NULL);
2102 	}
2103 
2104 	INFINIBAND_BPF_MTAP(scifp, m);
2105 
2106 	m = lagg_proto_input(sc, lp, m);
2107 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2108 		m_freem(m);
2109 		m = NULL;
2110 	}
2111 
2112 	LAGG_RUNLOCK();
2113 	return (m);
2114 }
2115 
2116 static int
2117 lagg_media_change(struct ifnet *ifp)
2118 {
2119 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2120 
2121 	if (sc->sc_ifflags & IFF_DEBUG)
2122 		printf("%s\n", __func__);
2123 
2124 	/* Ignore */
2125 	return (0);
2126 }
2127 
2128 static void
2129 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2130 {
2131 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2132 	struct lagg_port *lp;
2133 
2134 	imr->ifm_status = IFM_AVALID;
2135 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
2136 
2137 	LAGG_RLOCK();
2138 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2139 		if (LAGG_PORTACTIVE(lp))
2140 			imr->ifm_status |= IFM_ACTIVE;
2141 	}
2142 	LAGG_RUNLOCK();
2143 }
2144 
2145 static void
2146 lagg_linkstate(struct lagg_softc *sc)
2147 {
2148 	struct lagg_port *lp;
2149 	int new_link = LINK_STATE_DOWN;
2150 	uint64_t speed;
2151 
2152 	LAGG_XLOCK_ASSERT(sc);
2153 
2154 	/* LACP handles link state itself */
2155 	if (sc->sc_proto == LAGG_PROTO_LACP)
2156 		return;
2157 
2158 	/* Our link is considered up if at least one of our ports is active */
2159 	LAGG_RLOCK();
2160 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2161 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2162 			new_link = LINK_STATE_UP;
2163 			break;
2164 		}
2165 	}
2166 	LAGG_RUNLOCK();
2167 	if_link_state_change(sc->sc_ifp, new_link);
2168 
2169 	/* Update if_baudrate to reflect the max possible speed */
2170 	switch (sc->sc_proto) {
2171 		case LAGG_PROTO_FAILOVER:
2172 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2173 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
2174 			break;
2175 		case LAGG_PROTO_ROUNDROBIN:
2176 		case LAGG_PROTO_LOADBALANCE:
2177 		case LAGG_PROTO_BROADCAST:
2178 			speed = 0;
2179 			LAGG_RLOCK();
2180 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2181 				speed += lp->lp_ifp->if_baudrate;
2182 			LAGG_RUNLOCK();
2183 			sc->sc_ifp->if_baudrate = speed;
2184 			break;
2185 		case LAGG_PROTO_LACP:
2186 			/* LACP updates if_baudrate itself */
2187 			break;
2188 	}
2189 }
2190 
2191 static void
2192 lagg_port_state(struct ifnet *ifp, int state)
2193 {
2194 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2195 	struct lagg_softc *sc = NULL;
2196 
2197 	if (lp != NULL)
2198 		sc = lp->lp_softc;
2199 	if (sc == NULL)
2200 		return;
2201 
2202 	LAGG_XLOCK(sc);
2203 	lagg_linkstate(sc);
2204 	lagg_proto_linkstate(sc, lp);
2205 	LAGG_XUNLOCK(sc);
2206 }
2207 
2208 struct lagg_port *
2209 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2210 {
2211 	struct lagg_port *lp_next, *rval = NULL;
2212 
2213 	/*
2214 	 * Search a port which reports an active link state.
2215 	 */
2216 
2217 #ifdef INVARIANTS
2218 	/*
2219 	 * This is called with either LAGG_RLOCK() held or
2220 	 * LAGG_XLOCK(sc) held.
2221 	 */
2222 	if (!in_epoch(net_epoch_preempt))
2223 		LAGG_XLOCK_ASSERT(sc);
2224 #endif
2225 
2226 	if (lp == NULL)
2227 		goto search;
2228 	if (LAGG_PORTACTIVE(lp)) {
2229 		rval = lp;
2230 		goto found;
2231 	}
2232 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2233 	    LAGG_PORTACTIVE(lp_next)) {
2234 		rval = lp_next;
2235 		goto found;
2236 	}
2237 
2238 search:
2239 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2240 		if (LAGG_PORTACTIVE(lp_next)) {
2241 			return (lp_next);
2242 		}
2243 	}
2244 found:
2245 	return (rval);
2246 }
2247 
2248 int
2249 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2250 {
2251 
2252 #if defined(KERN_TLS) || defined(RATELIMIT)
2253 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2254 		struct lagg_snd_tag *lst;
2255 		struct m_snd_tag *mst;
2256 
2257 		mst = m->m_pkthdr.snd_tag;
2258 		lst = mst_to_lst(mst);
2259 		if (lst->tag->ifp != ifp) {
2260 			m_freem(m);
2261 			return (EAGAIN);
2262 		}
2263 		m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2264 		m_snd_tag_rele(mst);
2265 	}
2266 #endif
2267 	return (ifp->if_transmit)(ifp, m);
2268 }
2269 
2270 /*
2271  * Simple round robin aggregation
2272  */
2273 static void
2274 lagg_rr_attach(struct lagg_softc *sc)
2275 {
2276 	sc->sc_seq = 0;
2277 	sc->sc_stride = 1;
2278 }
2279 
2280 static int
2281 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2282 {
2283 	struct lagg_port *lp;
2284 	uint32_t p;
2285 
2286 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
2287 	p /= sc->sc_stride;
2288 	p %= sc->sc_count;
2289 	lp = CK_SLIST_FIRST(&sc->sc_ports);
2290 
2291 	while (p--)
2292 		lp = CK_SLIST_NEXT(lp, lp_entries);
2293 
2294 	/*
2295 	 * Check the port's link state. This will return the next active
2296 	 * port if the link is down or the port is NULL.
2297 	 */
2298 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2299 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2300 		m_freem(m);
2301 		return (ENETDOWN);
2302 	}
2303 
2304 	/* Send mbuf */
2305 	return (lagg_enqueue(lp->lp_ifp, m));
2306 }
2307 
2308 static struct mbuf *
2309 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2310 {
2311 	struct ifnet *ifp = sc->sc_ifp;
2312 
2313 	/* Just pass in the packet to our lagg device */
2314 	m->m_pkthdr.rcvif = ifp;
2315 
2316 	return (m);
2317 }
2318 
2319 /*
2320  * Broadcast mode
2321  */
2322 static int
2323 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2324 {
2325 	int active_ports = 0;
2326 	int errors = 0;
2327 	int ret;
2328 	struct lagg_port *lp, *last = NULL;
2329 	struct mbuf *m0;
2330 
2331 	LAGG_RLOCK_ASSERT();
2332 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2333 		if (!LAGG_PORTACTIVE(lp))
2334 			continue;
2335 
2336 		active_ports++;
2337 
2338 		if (last != NULL) {
2339 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2340 			if (m0 == NULL) {
2341 				ret = ENOBUFS;
2342 				errors++;
2343 				break;
2344 			}
2345 			lagg_enqueue(last->lp_ifp, m0);
2346 		}
2347 		last = lp;
2348 	}
2349 
2350 	if (last == NULL) {
2351 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2352 		m_freem(m);
2353 		return (ENOENT);
2354 	}
2355 	if ((last = lagg_link_active(sc, last)) == NULL) {
2356 		errors++;
2357 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2358 		m_freem(m);
2359 		return (ENETDOWN);
2360 	}
2361 
2362 	ret = lagg_enqueue(last->lp_ifp, m);
2363 	if (errors != 0)
2364 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2365 
2366 	return (ret);
2367 }
2368 
2369 static struct mbuf*
2370 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2371 {
2372 	struct ifnet *ifp = sc->sc_ifp;
2373 
2374 	/* Just pass in the packet to our lagg device */
2375 	m->m_pkthdr.rcvif = ifp;
2376 	return (m);
2377 }
2378 
2379 /*
2380  * Active failover
2381  */
2382 static int
2383 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2384 {
2385 	struct lagg_port *lp;
2386 
2387 	/* Use the master port if active or the next available port */
2388 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2389 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2390 		m_freem(m);
2391 		return (ENETDOWN);
2392 	}
2393 
2394 	/* Send mbuf */
2395 	return (lagg_enqueue(lp->lp_ifp, m));
2396 }
2397 
2398 static struct mbuf *
2399 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2400 {
2401 	struct ifnet *ifp = sc->sc_ifp;
2402 	struct lagg_port *tmp_tp;
2403 
2404 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2405 		m->m_pkthdr.rcvif = ifp;
2406 		return (m);
2407 	}
2408 
2409 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2410 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2411 		/*
2412 		 * If tmp_tp is null, we've received a packet when all
2413 		 * our links are down. Weird, but process it anyways.
2414 		 */
2415 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2416 			m->m_pkthdr.rcvif = ifp;
2417 			return (m);
2418 		}
2419 	}
2420 
2421 	m_freem(m);
2422 	return (NULL);
2423 }
2424 
2425 /*
2426  * Loadbalancing
2427  */
2428 static void
2429 lagg_lb_attach(struct lagg_softc *sc)
2430 {
2431 	struct lagg_port *lp;
2432 	struct lagg_lb *lb;
2433 
2434 	LAGG_XLOCK_ASSERT(sc);
2435 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2436 	lb->lb_key = m_ether_tcpip_hash_init();
2437 	sc->sc_psc = lb;
2438 
2439 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2440 		lagg_lb_port_create(lp);
2441 }
2442 
2443 static void
2444 lagg_lb_detach(struct lagg_softc *sc)
2445 {
2446 	struct lagg_lb *lb;
2447 
2448 	lb = (struct lagg_lb *)sc->sc_psc;
2449 	if (lb != NULL)
2450 		free(lb, M_LAGG);
2451 }
2452 
2453 static int
2454 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2455 {
2456 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2457 	struct lagg_port *lp_next;
2458 	int i = 0, rv;
2459 
2460 	rv = 0;
2461 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2462 	LAGG_XLOCK_ASSERT(sc);
2463 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2464 		if (lp_next == lp)
2465 			continue;
2466 		if (i >= LAGG_MAX_PORTS) {
2467 			rv = EINVAL;
2468 			break;
2469 		}
2470 		if (sc->sc_ifflags & IFF_DEBUG)
2471 			printf("%s: port %s at index %d\n",
2472 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2473 		lb->lb_ports[i++] = lp_next;
2474 	}
2475 
2476 	return (rv);
2477 }
2478 
2479 static int
2480 lagg_lb_port_create(struct lagg_port *lp)
2481 {
2482 	struct lagg_softc *sc = lp->lp_softc;
2483 	return (lagg_lb_porttable(sc, NULL));
2484 }
2485 
2486 static void
2487 lagg_lb_port_destroy(struct lagg_port *lp)
2488 {
2489 	struct lagg_softc *sc = lp->lp_softc;
2490 	lagg_lb_porttable(sc, lp);
2491 }
2492 
2493 static int
2494 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2495 {
2496 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2497 	struct lagg_port *lp = NULL;
2498 	uint32_t p = 0;
2499 
2500 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2501 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2502 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2503 	else
2504 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2505 	p %= sc->sc_count;
2506 	lp = lb->lb_ports[p];
2507 
2508 	/*
2509 	 * Check the port's link state. This will return the next active
2510 	 * port if the link is down or the port is NULL.
2511 	 */
2512 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2513 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2514 		m_freem(m);
2515 		return (ENETDOWN);
2516 	}
2517 
2518 	/* Send mbuf */
2519 	return (lagg_enqueue(lp->lp_ifp, m));
2520 }
2521 
2522 static struct mbuf *
2523 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2524 {
2525 	struct ifnet *ifp = sc->sc_ifp;
2526 
2527 	/* Just pass in the packet to our lagg device */
2528 	m->m_pkthdr.rcvif = ifp;
2529 
2530 	return (m);
2531 }
2532 
2533 /*
2534  * 802.3ad LACP
2535  */
2536 static void
2537 lagg_lacp_attach(struct lagg_softc *sc)
2538 {
2539 	struct lagg_port *lp;
2540 
2541 	lacp_attach(sc);
2542 	LAGG_XLOCK_ASSERT(sc);
2543 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2544 		lacp_port_create(lp);
2545 }
2546 
2547 static void
2548 lagg_lacp_detach(struct lagg_softc *sc)
2549 {
2550 	struct lagg_port *lp;
2551 	void *psc;
2552 
2553 	LAGG_XLOCK_ASSERT(sc);
2554 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2555 		lacp_port_destroy(lp);
2556 
2557 	psc = sc->sc_psc;
2558 	sc->sc_psc = NULL;
2559 	lacp_detach(psc);
2560 }
2561 
2562 static void
2563 lagg_lacp_lladdr(struct lagg_softc *sc)
2564 {
2565 	struct lagg_port *lp;
2566 
2567 	LAGG_SXLOCK_ASSERT(sc);
2568 
2569 	/* purge all the lacp ports */
2570 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2571 		lacp_port_destroy(lp);
2572 
2573 	/* add them back in */
2574 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2575 		lacp_port_create(lp);
2576 }
2577 
2578 static int
2579 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2580 {
2581 	struct lagg_port *lp;
2582 
2583 	lp = lacp_select_tx_port(sc, m);
2584 	if (lp == NULL) {
2585 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2586 		m_freem(m);
2587 		return (ENETDOWN);
2588 	}
2589 
2590 	/* Send mbuf */
2591 	return (lagg_enqueue(lp->lp_ifp, m));
2592 }
2593 
2594 static struct mbuf *
2595 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2596 {
2597 	struct ifnet *ifp = sc->sc_ifp;
2598 	struct ether_header *eh;
2599 	u_short etype;
2600 
2601 	eh = mtod(m, struct ether_header *);
2602 	etype = ntohs(eh->ether_type);
2603 
2604 	/* Tap off LACP control messages */
2605 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2606 		m = lacp_input(lp, m);
2607 		if (m == NULL)
2608 			return (NULL);
2609 	}
2610 
2611 	/*
2612 	 * If the port is not collecting or not in the active aggregator then
2613 	 * free and return.
2614 	 */
2615 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2616 		m_freem(m);
2617 		return (NULL);
2618 	}
2619 
2620 	m->m_pkthdr.rcvif = ifp;
2621 	return (m);
2622 }
2623