xref: /freebsd/sys/net/if_lagg.c (revision c66ec88fed842fbaad62c30d510644ceb7bd2d71)
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 void     lagg_ratelimit_query(struct ifnet *,
155 		    struct if_ratelimit_query_results *);
156 #endif
157 static int	lagg_setmulti(struct lagg_port *);
158 static int	lagg_clrmulti(struct lagg_port *);
159 static	int	lagg_setcaps(struct lagg_port *, int cap);
160 static	int	lagg_setflag(struct lagg_port *, int, int,
161 		    int (*func)(struct ifnet *, int));
162 static	int	lagg_setflags(struct lagg_port *, int status);
163 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
164 static int	lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
165 static int	lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
166 static void	lagg_qflush(struct ifnet *);
167 static int	lagg_media_change(struct ifnet *);
168 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
169 static struct lagg_port *lagg_link_active(struct lagg_softc *,
170 	    struct lagg_port *);
171 
172 /* Simple round robin */
173 static void	lagg_rr_attach(struct lagg_softc *);
174 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
175 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
176 		    struct mbuf *);
177 
178 /* Active failover */
179 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
180 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
181 		    struct mbuf *);
182 
183 /* Loadbalancing */
184 static void	lagg_lb_attach(struct lagg_softc *);
185 static void	lagg_lb_detach(struct lagg_softc *);
186 static int	lagg_lb_port_create(struct lagg_port *);
187 static void	lagg_lb_port_destroy(struct lagg_port *);
188 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
189 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
190 		    struct mbuf *);
191 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
192 
193 /* Broadcast */
194 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
195 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
196 		    struct mbuf *);
197 
198 /* 802.3ad LACP */
199 static void	lagg_lacp_attach(struct lagg_softc *);
200 static void	lagg_lacp_detach(struct lagg_softc *);
201 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
202 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
203 		    struct mbuf *);
204 static void	lagg_lacp_lladdr(struct lagg_softc *);
205 
206 /* lagg protocol table */
207 static const struct lagg_proto {
208 	lagg_proto	pr_num;
209 	void		(*pr_attach)(struct lagg_softc *);
210 	void		(*pr_detach)(struct lagg_softc *);
211 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
212 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
213 			    struct mbuf *);
214 	int		(*pr_addport)(struct lagg_port *);
215 	void		(*pr_delport)(struct lagg_port *);
216 	void		(*pr_linkstate)(struct lagg_port *);
217 	void 		(*pr_init)(struct lagg_softc *);
218 	void 		(*pr_stop)(struct lagg_softc *);
219 	void 		(*pr_lladdr)(struct lagg_softc *);
220 	void		(*pr_request)(struct lagg_softc *, void *);
221 	void		(*pr_portreq)(struct lagg_port *, void *);
222 } lagg_protos[] = {
223     {
224 	.pr_num = LAGG_PROTO_NONE
225     },
226     {
227 	.pr_num = LAGG_PROTO_ROUNDROBIN,
228 	.pr_attach = lagg_rr_attach,
229 	.pr_start = lagg_rr_start,
230 	.pr_input = lagg_rr_input,
231     },
232     {
233 	.pr_num = LAGG_PROTO_FAILOVER,
234 	.pr_start = lagg_fail_start,
235 	.pr_input = lagg_fail_input,
236     },
237     {
238 	.pr_num = LAGG_PROTO_LOADBALANCE,
239 	.pr_attach = lagg_lb_attach,
240 	.pr_detach = lagg_lb_detach,
241 	.pr_start = lagg_lb_start,
242 	.pr_input = lagg_lb_input,
243 	.pr_addport = lagg_lb_port_create,
244 	.pr_delport = lagg_lb_port_destroy,
245     },
246     {
247 	.pr_num = LAGG_PROTO_LACP,
248 	.pr_attach = lagg_lacp_attach,
249 	.pr_detach = lagg_lacp_detach,
250 	.pr_start = lagg_lacp_start,
251 	.pr_input = lagg_lacp_input,
252 	.pr_addport = lacp_port_create,
253 	.pr_delport = lacp_port_destroy,
254 	.pr_linkstate = lacp_linkstate,
255 	.pr_init = lacp_init,
256 	.pr_stop = lacp_stop,
257 	.pr_lladdr = lagg_lacp_lladdr,
258 	.pr_request = lacp_req,
259 	.pr_portreq = lacp_portreq,
260     },
261     {
262 	.pr_num = LAGG_PROTO_BROADCAST,
263 	.pr_start = lagg_bcast_start,
264 	.pr_input = lagg_bcast_input,
265     },
266 };
267 
268 SYSCTL_DECL(_net_link);
269 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
270     "Link Aggregation");
271 
272 /* Allow input on any failover links */
273 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
274 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
275 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
276     &VNET_NAME(lagg_failover_rx_all), 0,
277     "Accept input from any interface in a failover lagg");
278 
279 /* Default value for using flowid */
280 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
281 #define	V_def_use_flowid	VNET(def_use_flowid)
282 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
283     &VNET_NAME(def_use_flowid), 0,
284     "Default setting for using flow id for load sharing");
285 
286 /* Default value for using numa */
287 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
288 #define	V_def_use_numa	VNET(def_use_numa)
289 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN,
290     &VNET_NAME(def_use_numa), 0,
291     "Use numa to steer flows");
292 
293 /* Default value for flowid shift */
294 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
295 #define	V_def_flowid_shift	VNET(def_flowid_shift)
296 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
297     &VNET_NAME(def_flowid_shift), 0,
298     "Default setting for flowid shift for load sharing");
299 
300 static void
301 vnet_lagg_init(const void *unused __unused)
302 {
303 
304 	LAGG_LIST_LOCK_INIT();
305 	SLIST_INIT(&V_lagg_list);
306 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
307 	    lagg_clone_destroy, 0);
308 }
309 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
310     vnet_lagg_init, NULL);
311 
312 static void
313 vnet_lagg_uninit(const void *unused __unused)
314 {
315 
316 	if_clone_detach(V_lagg_cloner);
317 	LAGG_LIST_LOCK_DESTROY();
318 }
319 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
320     vnet_lagg_uninit, NULL);
321 
322 static int
323 lagg_modevent(module_t mod, int type, void *data)
324 {
325 
326 	switch (type) {
327 	case MOD_LOAD:
328 		lagg_input_ethernet_p = lagg_input_ethernet;
329 		lagg_input_infiniband_p = lagg_input_infiniband;
330 		lagg_linkstate_p = lagg_port_state;
331 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
332 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
333 		    EVENTHANDLER_PRI_ANY);
334 		break;
335 	case MOD_UNLOAD:
336 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
337 		    lagg_detach_cookie);
338 		lagg_input_ethernet_p = NULL;
339 		lagg_input_infiniband_p = NULL;
340 		lagg_linkstate_p = NULL;
341 		break;
342 	default:
343 		return (EOPNOTSUPP);
344 	}
345 	return (0);
346 }
347 
348 static moduledata_t lagg_mod = {
349 	"if_lagg",
350 	lagg_modevent,
351 	0
352 };
353 
354 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
355 MODULE_VERSION(if_lagg, 1);
356 MODULE_DEPEND(if_lagg, if_infiniband, 1, 1, 1);
357 
358 static void
359 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
360 {
361 
362 	LAGG_XLOCK_ASSERT(sc);
363 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
364 	    __func__, sc));
365 
366 	if (sc->sc_ifflags & IFF_DEBUG)
367 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
368 
369 	if (lagg_protos[pr].pr_attach != NULL)
370 		lagg_protos[pr].pr_attach(sc);
371 	sc->sc_proto = pr;
372 }
373 
374 static void
375 lagg_proto_detach(struct lagg_softc *sc)
376 {
377 	lagg_proto pr;
378 
379 	LAGG_XLOCK_ASSERT(sc);
380 	pr = sc->sc_proto;
381 	sc->sc_proto = LAGG_PROTO_NONE;
382 
383 	if (lagg_protos[pr].pr_detach != NULL)
384 		lagg_protos[pr].pr_detach(sc);
385 }
386 
387 static int
388 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
389 {
390 
391 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
392 }
393 
394 static struct mbuf *
395 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
396 {
397 
398 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
399 }
400 
401 static int
402 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
403 {
404 
405 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
406 		return (0);
407 	else
408 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
409 }
410 
411 static void
412 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
413 {
414 
415 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
416 		lagg_protos[sc->sc_proto].pr_delport(lp);
417 }
418 
419 static void
420 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
421 {
422 
423 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
424 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
425 }
426 
427 static void
428 lagg_proto_init(struct lagg_softc *sc)
429 {
430 
431 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
432 		lagg_protos[sc->sc_proto].pr_init(sc);
433 }
434 
435 static void
436 lagg_proto_stop(struct lagg_softc *sc)
437 {
438 
439 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
440 		lagg_protos[sc->sc_proto].pr_stop(sc);
441 }
442 
443 static void
444 lagg_proto_lladdr(struct lagg_softc *sc)
445 {
446 
447 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
448 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
449 }
450 
451 static void
452 lagg_proto_request(struct lagg_softc *sc, void *v)
453 {
454 
455 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
456 		lagg_protos[sc->sc_proto].pr_request(sc, v);
457 }
458 
459 static void
460 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
461 {
462 
463 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
464 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
465 }
466 
467 /*
468  * This routine is run via an vlan
469  * config EVENT
470  */
471 static void
472 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
473 {
474 	struct lagg_softc *sc = ifp->if_softc;
475 	struct lagg_port *lp;
476 
477 	if (ifp->if_softc !=  arg)   /* Not our event */
478 		return;
479 
480 	LAGG_XLOCK(sc);
481 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
482 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
483 	LAGG_XUNLOCK(sc);
484 }
485 
486 /*
487  * This routine is run via an vlan
488  * unconfig EVENT
489  */
490 static void
491 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
492 {
493 	struct lagg_softc *sc = ifp->if_softc;
494 	struct lagg_port *lp;
495 
496 	if (ifp->if_softc !=  arg)   /* Not our event */
497 		return;
498 
499 	LAGG_XLOCK(sc);
500 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
501 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
502 	LAGG_XUNLOCK(sc);
503 }
504 
505 static int
506 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
507 {
508 	struct iflaggparam iflp;
509 	struct lagg_softc *sc;
510 	struct ifnet *ifp;
511 	int if_type;
512 	int error;
513 	static const uint8_t eaddr[LAGG_ADDR_LEN];
514 
515 	if (params != NULL) {
516 		error = copyin(params, &iflp, sizeof(iflp));
517 		if (error)
518 			return (error);
519 
520 		switch (iflp.lagg_type) {
521 		case LAGG_TYPE_ETHERNET:
522 			if_type = IFT_ETHER;
523 			break;
524 		case LAGG_TYPE_INFINIBAND:
525 			if_type = IFT_INFINIBAND;
526 			break;
527 		default:
528 			return (EINVAL);
529 		}
530 	} else {
531 		if_type = IFT_ETHER;
532 	}
533 
534 	sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO);
535 	ifp = sc->sc_ifp = if_alloc(if_type);
536 	if (ifp == NULL) {
537 		free(sc, M_LAGG);
538 		return (ENOSPC);
539 	}
540 	LAGG_SX_INIT(sc);
541 
542 	mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
543 	callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
544 
545 	LAGG_XLOCK(sc);
546 	if (V_def_use_flowid)
547 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
548 	if (V_def_use_numa)
549 		sc->sc_opts |= LAGG_OPT_USE_NUMA;
550 	sc->flowid_shift = V_def_flowid_shift;
551 
552 	/* Hash all layers by default */
553 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
554 
555 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
556 
557 	CK_SLIST_INIT(&sc->sc_ports);
558 
559 	switch (if_type) {
560 	case IFT_ETHER:
561 		/* Initialise pseudo media types */
562 		ifmedia_init(&sc->sc_media, 0, lagg_media_change,
563 		    lagg_media_status);
564 		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
565 		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
566 
567 		if_initname(ifp, laggname, unit);
568 		ifp->if_transmit = lagg_transmit_ethernet;
569 		break;
570 	case IFT_INFINIBAND:
571 		if_initname(ifp, laggname, unit);
572 		ifp->if_transmit = lagg_transmit_infiniband;
573 		break;
574 	default:
575 		break;
576 	}
577 	ifp->if_softc = sc;
578 	ifp->if_qflush = lagg_qflush;
579 	ifp->if_init = lagg_init;
580 	ifp->if_ioctl = lagg_ioctl;
581 	ifp->if_get_counter = lagg_get_counter;
582 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
583 #if defined(KERN_TLS) || defined(RATELIMIT)
584 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
585 	ifp->if_snd_tag_modify = lagg_snd_tag_modify;
586 	ifp->if_snd_tag_query = lagg_snd_tag_query;
587 	ifp->if_snd_tag_free = lagg_snd_tag_free;
588 	ifp->if_ratelimit_query = lagg_ratelimit_query;
589 #endif
590 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
591 
592 	/*
593 	 * Attach as an ordinary ethernet device, children will be attached
594 	 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
595 	 */
596 	switch (if_type) {
597 	case IFT_ETHER:
598 		ether_ifattach(ifp, eaddr);
599 		break;
600 	case IFT_INFINIBAND:
601 		infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr);
602 		break;
603 	default:
604 		break;
605 	}
606 
607 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
608 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
609 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
610 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
611 
612 	/* Insert into the global list of laggs */
613 	LAGG_LIST_LOCK();
614 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
615 	LAGG_LIST_UNLOCK();
616 	LAGG_XUNLOCK(sc);
617 
618 	return (0);
619 }
620 
621 static void
622 lagg_clone_destroy(struct ifnet *ifp)
623 {
624 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
625 	struct lagg_port *lp;
626 
627 	LAGG_XLOCK(sc);
628 	sc->sc_destroying = 1;
629 	lagg_stop(sc);
630 	ifp->if_flags &= ~IFF_UP;
631 
632 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
633 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
634 
635 	/* Shutdown and remove lagg ports */
636 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
637 		lagg_port_destroy(lp, 1);
638 
639 	/* Unhook the aggregation protocol */
640 	lagg_proto_detach(sc);
641 	LAGG_XUNLOCK(sc);
642 
643 	switch (ifp->if_type) {
644 	case IFT_ETHER:
645 		ifmedia_removeall(&sc->sc_media);
646 		ether_ifdetach(ifp);
647 		break;
648 	case IFT_INFINIBAND:
649 		infiniband_ifdetach(ifp);
650 		break;
651 	default:
652 		break;
653 	}
654 	if_free(ifp);
655 
656 	LAGG_LIST_LOCK();
657 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
658 	LAGG_LIST_UNLOCK();
659 
660 	mtx_destroy(&sc->sc_mtx);
661 	LAGG_SX_DESTROY(sc);
662 	free(sc, M_LAGG);
663 }
664 
665 static void
666 lagg_capabilities(struct lagg_softc *sc)
667 {
668 	struct lagg_port *lp;
669 	int cap, ena, pena;
670 	uint64_t hwa;
671 	struct ifnet_hw_tsomax hw_tsomax;
672 
673 	LAGG_XLOCK_ASSERT(sc);
674 
675 	/* Get common enabled capabilities for the lagg ports */
676 	ena = ~0;
677 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
678 		ena &= lp->lp_ifp->if_capenable;
679 	ena = (ena == ~0 ? 0 : ena);
680 
681 	/*
682 	 * Apply common enabled capabilities back to the lagg ports.
683 	 * May require several iterations if they are dependent.
684 	 */
685 	do {
686 		pena = ena;
687 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
688 			lagg_setcaps(lp, ena);
689 			ena &= lp->lp_ifp->if_capenable;
690 		}
691 	} while (pena != ena);
692 
693 	/* Get other capabilities from the lagg ports */
694 	cap = ~0;
695 	hwa = ~(uint64_t)0;
696 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
697 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
698 		cap &= lp->lp_ifp->if_capabilities;
699 		hwa &= lp->lp_ifp->if_hwassist;
700 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
701 	}
702 	cap = (cap == ~0 ? 0 : cap);
703 	hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
704 
705 	if (sc->sc_ifp->if_capabilities != cap ||
706 	    sc->sc_ifp->if_capenable != ena ||
707 	    sc->sc_ifp->if_hwassist != hwa ||
708 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
709 		sc->sc_ifp->if_capabilities = cap;
710 		sc->sc_ifp->if_capenable = ena;
711 		sc->sc_ifp->if_hwassist = hwa;
712 		getmicrotime(&sc->sc_ifp->if_lastchange);
713 
714 		if (sc->sc_ifflags & IFF_DEBUG)
715 			if_printf(sc->sc_ifp,
716 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
717 	}
718 }
719 
720 static int
721 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
722 {
723 	struct lagg_softc *sc_ptr;
724 	struct lagg_port *lp, *tlp;
725 	struct ifreq ifr;
726 	int error, i, oldmtu;
727 	int if_type;
728 	uint64_t *pval;
729 
730 	LAGG_XLOCK_ASSERT(sc);
731 
732 	if (sc->sc_ifp == ifp) {
733 		if_printf(sc->sc_ifp,
734 		    "cannot add a lagg to itself as a port\n");
735 		return (EINVAL);
736 	}
737 
738 	if (sc->sc_destroying == 1)
739 		return (ENXIO);
740 
741 	/* Limit the maximal number of lagg ports */
742 	if (sc->sc_count >= LAGG_MAX_PORTS)
743 		return (ENOSPC);
744 
745 	/* Check if port has already been associated to a lagg */
746 	if (ifp->if_lagg != NULL) {
747 		/* Port is already in the current lagg? */
748 		lp = (struct lagg_port *)ifp->if_lagg;
749 		if (lp->lp_softc == sc)
750 			return (EEXIST);
751 		return (EBUSY);
752 	}
753 
754 	switch (sc->sc_ifp->if_type) {
755 	case IFT_ETHER:
756 		/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
757 		if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
758 			return (EPROTONOSUPPORT);
759 		if_type = IFT_IEEE8023ADLAG;
760 		break;
761 	case IFT_INFINIBAND:
762 		/* XXX Disallow non-infiniband interfaces */
763 		if (ifp->if_type != IFT_INFINIBAND)
764 			return (EPROTONOSUPPORT);
765 		if_type = IFT_INFINIBANDLAG;
766 		break;
767 	default:
768 		break;
769 	}
770 
771 	/* Allow the first Ethernet member to define the MTU */
772 	oldmtu = -1;
773 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
774 		sc->sc_ifp->if_mtu = ifp->if_mtu;
775 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
776 		if (ifp->if_ioctl == NULL) {
777 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
778 			    ifp->if_xname);
779 			return (EINVAL);
780 		}
781 		oldmtu = ifp->if_mtu;
782 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
783 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
784 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
785 		if (error != 0) {
786 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
787 			    ifp->if_xname);
788 			return (error);
789 		}
790 		ifr.ifr_mtu = oldmtu;
791 	}
792 
793 	lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO);
794 	lp->lp_softc = sc;
795 
796 	/* Check if port is a stacked lagg */
797 	LAGG_LIST_LOCK();
798 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
799 		if (ifp == sc_ptr->sc_ifp) {
800 			LAGG_LIST_UNLOCK();
801 			free(lp, M_LAGG);
802 			if (oldmtu != -1)
803 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
804 				    (caddr_t)&ifr);
805 			return (EINVAL);
806 			/* XXX disable stacking for the moment, its untested */
807 #ifdef LAGG_PORT_STACKING
808 			lp->lp_flags |= LAGG_PORT_STACK;
809 			if (lagg_port_checkstacking(sc_ptr) >=
810 			    LAGG_MAX_STACKING) {
811 				LAGG_LIST_UNLOCK();
812 				free(lp, M_LAGG);
813 				if (oldmtu != -1)
814 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
815 					    (caddr_t)&ifr);
816 				return (E2BIG);
817 			}
818 #endif
819 		}
820 	}
821 	LAGG_LIST_UNLOCK();
822 
823 	if_ref(ifp);
824 	lp->lp_ifp = ifp;
825 
826 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
827 	lp->lp_ifcapenable = ifp->if_capenable;
828 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
829 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
830 		lagg_proto_lladdr(sc);
831 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
832 	} else {
833 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
834 	}
835 	lagg_setflags(lp, 1);
836 
837 	if (CK_SLIST_EMPTY(&sc->sc_ports))
838 		sc->sc_primary = lp;
839 
840 	/* Change the interface type */
841 	lp->lp_iftype = ifp->if_type;
842 	ifp->if_type = if_type;
843 	ifp->if_lagg = lp;
844 	lp->lp_ioctl = ifp->if_ioctl;
845 	ifp->if_ioctl = lagg_port_ioctl;
846 	lp->lp_output = ifp->if_output;
847 	ifp->if_output = lagg_port_output;
848 
849 	/* Read port counters */
850 	pval = lp->port_counters.val;
851 	for (i = 0; i < IFCOUNTERS; i++, pval++)
852 		*pval = ifp->if_get_counter(ifp, i);
853 
854 	/*
855 	 * Insert into the list of ports.
856 	 * Keep ports sorted by if_index. It is handy, when configuration
857 	 * is predictable and `ifconfig laggN create ...` command
858 	 * will lead to the same result each time.
859 	 */
860 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
861 		if (tlp->lp_ifp->if_index < ifp->if_index && (
862 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
863 		    ((struct  lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
864 		    ifp->if_index))
865 			break;
866 	}
867 	if (tlp != NULL)
868 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
869 	else
870 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
871 	sc->sc_count++;
872 
873 	lagg_setmulti(lp);
874 
875 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
876 		/* Remove the port, without calling pr_delport. */
877 		lagg_port_destroy(lp, 0);
878 		if (oldmtu != -1)
879 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
880 		return (error);
881 	}
882 
883 	/* Update lagg capabilities */
884 	lagg_capabilities(sc);
885 	lagg_linkstate(sc);
886 
887 	return (0);
888 }
889 
890 #ifdef LAGG_PORT_STACKING
891 static int
892 lagg_port_checkstacking(struct lagg_softc *sc)
893 {
894 	struct lagg_softc *sc_ptr;
895 	struct lagg_port *lp;
896 	int m = 0;
897 
898 	LAGG_SXLOCK_ASSERT(sc);
899 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
900 		if (lp->lp_flags & LAGG_PORT_STACK) {
901 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
902 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
903 		}
904 	}
905 
906 	return (m + 1);
907 }
908 #endif
909 
910 static void
911 lagg_port_destroy_cb(epoch_context_t ec)
912 {
913 	struct lagg_port *lp;
914 	struct ifnet *ifp;
915 
916 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
917 	ifp = lp->lp_ifp;
918 
919 	if_rele(ifp);
920 	free(lp, M_LAGG);
921 }
922 
923 static int
924 lagg_port_destroy(struct lagg_port *lp, int rundelport)
925 {
926 	struct lagg_softc *sc = lp->lp_softc;
927 	struct lagg_port *lp_ptr, *lp0;
928 	struct ifnet *ifp = lp->lp_ifp;
929 	uint64_t *pval, vdiff;
930 	int i;
931 
932 	LAGG_XLOCK_ASSERT(sc);
933 
934 	if (rundelport)
935 		lagg_proto_delport(sc, lp);
936 
937 	if (lp->lp_detaching == 0)
938 		lagg_clrmulti(lp);
939 
940 	/* Restore interface */
941 	ifp->if_type = lp->lp_iftype;
942 	ifp->if_ioctl = lp->lp_ioctl;
943 	ifp->if_output = lp->lp_output;
944 	ifp->if_lagg = NULL;
945 
946 	/* Update detached port counters */
947 	pval = lp->port_counters.val;
948 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
949 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
950 		sc->detached_counters.val[i] += vdiff;
951 	}
952 
953 	/* Finally, remove the port from the lagg */
954 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
955 	sc->sc_count--;
956 
957 	/* Update the primary interface */
958 	if (lp == sc->sc_primary) {
959 		uint8_t lladdr[LAGG_ADDR_LEN];
960 
961 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
962 			bzero(&lladdr, LAGG_ADDR_LEN);
963 		else
964 			bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
965 		sc->sc_primary = lp0;
966 		if (sc->sc_destroying == 0) {
967 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
968 			lagg_proto_lladdr(sc);
969 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
970 		}
971 
972 		/*
973 		 * Update lladdr for each port (new primary needs update
974 		 * as well, to switch from old lladdr to its 'real' one)
975 		 */
976 		CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
977 			if_setlladdr(lp_ptr->lp_ifp, lladdr, lp_ptr->lp_ifp->if_addrlen);
978 	}
979 
980 	if (lp->lp_ifflags)
981 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
982 
983 	if (lp->lp_detaching == 0) {
984 		lagg_setflags(lp, 0);
985 		lagg_setcaps(lp, lp->lp_ifcapenable);
986 		if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
987 	}
988 
989 	/*
990 	 * free port and release it's ifnet reference after a grace period has
991 	 * elapsed.
992 	 */
993 	NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
994 	/* Update lagg capabilities */
995 	lagg_capabilities(sc);
996 	lagg_linkstate(sc);
997 
998 	return (0);
999 }
1000 
1001 static int
1002 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1003 {
1004 	struct epoch_tracker et;
1005 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
1006 	struct lagg_softc *sc;
1007 	struct lagg_port *lp = NULL;
1008 	int error = 0;
1009 
1010 	/* Should be checked by the caller */
1011 	switch (ifp->if_type) {
1012 	case IFT_IEEE8023ADLAG:
1013 	case IFT_INFINIBANDLAG:
1014 		if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1015 			goto fallback;
1016 		break;
1017 	default:
1018 		goto fallback;
1019 	}
1020 
1021 	switch (cmd) {
1022 	case SIOCGLAGGPORT:
1023 		if (rp->rp_portname[0] == '\0' ||
1024 		    ifunit(rp->rp_portname) != ifp) {
1025 			error = EINVAL;
1026 			break;
1027 		}
1028 
1029 		NET_EPOCH_ENTER(et);
1030 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1031 			error = ENOENT;
1032 			NET_EPOCH_EXIT(et);
1033 			break;
1034 		}
1035 
1036 		lagg_port2req(lp, rp);
1037 		NET_EPOCH_EXIT(et);
1038 		break;
1039 
1040 	case SIOCSIFCAP:
1041 		if (lp->lp_ioctl == NULL) {
1042 			error = EINVAL;
1043 			break;
1044 		}
1045 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1046 		if (error)
1047 			break;
1048 
1049 		/* Update lagg interface capabilities */
1050 		LAGG_XLOCK(sc);
1051 		lagg_capabilities(sc);
1052 		LAGG_XUNLOCK(sc);
1053 		VLAN_CAPABILITIES(sc->sc_ifp);
1054 		break;
1055 
1056 	case SIOCSIFMTU:
1057 		/* Do not allow the MTU to be changed once joined */
1058 		error = EINVAL;
1059 		break;
1060 
1061 	default:
1062 		goto fallback;
1063 	}
1064 
1065 	return (error);
1066 
1067 fallback:
1068 	if (lp != NULL && lp->lp_ioctl != NULL)
1069 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1070 
1071 	return (EINVAL);
1072 }
1073 
1074 /*
1075  * Requests counter @cnt data.
1076  *
1077  * Counter value is calculated the following way:
1078  * 1) for each port, sum  difference between current and "initial" measurements.
1079  * 2) add lagg logical interface counters.
1080  * 3) add data from detached_counters array.
1081  *
1082  * We also do the following things on ports attach/detach:
1083  * 1) On port attach we store all counters it has into port_counter array.
1084  * 2) On port detach we add the different between "initial" and
1085  *   current counters data to detached_counters array.
1086  */
1087 static uint64_t
1088 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1089 {
1090 	struct epoch_tracker et;
1091 	struct lagg_softc *sc;
1092 	struct lagg_port *lp;
1093 	struct ifnet *lpifp;
1094 	uint64_t newval, oldval, vsum;
1095 
1096 	/* Revise this when we've got non-generic counters. */
1097 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1098 
1099 	sc = (struct lagg_softc *)ifp->if_softc;
1100 
1101 	vsum = 0;
1102 	NET_EPOCH_ENTER(et);
1103 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1104 		/* Saved attached value */
1105 		oldval = lp->port_counters.val[cnt];
1106 		/* current value */
1107 		lpifp = lp->lp_ifp;
1108 		newval = lpifp->if_get_counter(lpifp, cnt);
1109 		/* Calculate diff and save new */
1110 		vsum += newval - oldval;
1111 	}
1112 	NET_EPOCH_EXIT(et);
1113 
1114 	/*
1115 	 * Add counter data which might be added by upper
1116 	 * layer protocols operating on logical interface.
1117 	 */
1118 	vsum += if_get_counter_default(ifp, cnt);
1119 
1120 	/*
1121 	 * Add counter data from detached ports counters
1122 	 */
1123 	vsum += sc->detached_counters.val[cnt];
1124 
1125 	return (vsum);
1126 }
1127 
1128 /*
1129  * For direct output to child ports.
1130  */
1131 static int
1132 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1133 	const struct sockaddr *dst, struct route *ro)
1134 {
1135 	struct lagg_port *lp = ifp->if_lagg;
1136 
1137 	switch (dst->sa_family) {
1138 		case pseudo_AF_HDRCMPLT:
1139 		case AF_UNSPEC:
1140 			if (lp != NULL)
1141 				return ((*lp->lp_output)(ifp, m, dst, ro));
1142 	}
1143 
1144 	/* drop any other frames */
1145 	m_freem(m);
1146 	return (ENETDOWN);
1147 }
1148 
1149 static void
1150 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1151 {
1152 	struct lagg_port *lp;
1153 	struct lagg_softc *sc;
1154 
1155 	if ((lp = ifp->if_lagg) == NULL)
1156 		return;
1157 	/* If the ifnet is just being renamed, don't do anything. */
1158 	if (ifp->if_flags & IFF_RENAMING)
1159 		return;
1160 
1161 	sc = lp->lp_softc;
1162 
1163 	LAGG_XLOCK(sc);
1164 	lp->lp_detaching = 1;
1165 	lagg_port_destroy(lp, 1);
1166 	LAGG_XUNLOCK(sc);
1167 	VLAN_CAPABILITIES(sc->sc_ifp);
1168 }
1169 
1170 static void
1171 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1172 {
1173 	struct lagg_softc *sc = lp->lp_softc;
1174 
1175 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1176 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1177 	rp->rp_prio = lp->lp_prio;
1178 	rp->rp_flags = lp->lp_flags;
1179 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1180 
1181 	/* Add protocol specific flags */
1182 	switch (sc->sc_proto) {
1183 		case LAGG_PROTO_FAILOVER:
1184 			if (lp == sc->sc_primary)
1185 				rp->rp_flags |= LAGG_PORT_MASTER;
1186 			if (lp == lagg_link_active(sc, sc->sc_primary))
1187 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1188 			break;
1189 
1190 		case LAGG_PROTO_ROUNDROBIN:
1191 		case LAGG_PROTO_LOADBALANCE:
1192 		case LAGG_PROTO_BROADCAST:
1193 			if (LAGG_PORTACTIVE(lp))
1194 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1195 			break;
1196 
1197 		case LAGG_PROTO_LACP:
1198 			/* LACP has a different definition of active */
1199 			if (lacp_isactive(lp))
1200 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1201 			if (lacp_iscollecting(lp))
1202 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1203 			if (lacp_isdistributing(lp))
1204 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1205 			break;
1206 	}
1207 
1208 }
1209 
1210 static void
1211 lagg_watchdog_infiniband(void *arg)
1212 {
1213 	struct epoch_tracker et;
1214 	struct lagg_softc *sc;
1215 	struct lagg_port *lp;
1216 	struct ifnet *ifp;
1217 	struct ifnet *lp_ifp;
1218 
1219 	sc = arg;
1220 
1221 	/*
1222 	 * Because infiniband nodes have a fixed MAC address, which is
1223 	 * generated by the so-called GID, we need to regularly update
1224 	 * the link level address of the parent lagg<N> device when
1225 	 * the active port changes. Possibly we could piggy-back on
1226 	 * link up/down events aswell, but using a timer also provides
1227 	 * a guarantee against too frequent events. This operation
1228 	 * does not have to be atomic.
1229 	 */
1230 	NET_EPOCH_ENTER(et);
1231 	lp = lagg_link_active(sc, sc->sc_primary);
1232 	if (lp != NULL) {
1233 		ifp = sc->sc_ifp;
1234 		lp_ifp = lp->lp_ifp;
1235 
1236 		if (ifp != NULL && lp_ifp != NULL &&
1237 		    (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0 ||
1238 		     memcmp(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen) != 0)) {
1239 			memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1240 			memcpy(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen);
1241 
1242 			CURVNET_SET(ifp->if_vnet);
1243 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1244 			CURVNET_RESTORE();
1245 		}
1246 	}
1247 	NET_EPOCH_EXIT(et);
1248 
1249 	callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1250 }
1251 
1252 static void
1253 lagg_init(void *xsc)
1254 {
1255 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1256 	struct ifnet *ifp = sc->sc_ifp;
1257 	struct lagg_port *lp;
1258 
1259 	LAGG_XLOCK(sc);
1260 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1261 		LAGG_XUNLOCK(sc);
1262 		return;
1263 	}
1264 
1265 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1266 
1267 	/*
1268 	 * Update the port lladdrs if needed.
1269 	 * This might be if_setlladdr() notification
1270 	 * that lladdr has been changed.
1271 	 */
1272 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1273 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1274 		    ifp->if_addrlen) != 0)
1275 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1276 	}
1277 
1278 	lagg_proto_init(sc);
1279 
1280 	if (ifp->if_type == IFT_INFINIBAND) {
1281 		mtx_lock(&sc->sc_mtx);
1282 		lagg_watchdog_infiniband(sc);
1283 		mtx_unlock(&sc->sc_mtx);
1284 	}
1285 
1286 	LAGG_XUNLOCK(sc);
1287 }
1288 
1289 static void
1290 lagg_stop(struct lagg_softc *sc)
1291 {
1292 	struct ifnet *ifp = sc->sc_ifp;
1293 
1294 	LAGG_XLOCK_ASSERT(sc);
1295 
1296 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1297 		return;
1298 
1299 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1300 
1301 	lagg_proto_stop(sc);
1302 
1303 	mtx_lock(&sc->sc_mtx);
1304 	callout_stop(&sc->sc_watchdog);
1305 	mtx_unlock(&sc->sc_mtx);
1306 
1307 	callout_drain(&sc->sc_watchdog);
1308 }
1309 
1310 static int
1311 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1312 {
1313 	struct epoch_tracker et;
1314 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1315 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1316 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1317 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1318 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1319 	struct ifreq *ifr = (struct ifreq *)data;
1320 	struct lagg_port *lp;
1321 	struct ifnet *tpif;
1322 	struct thread *td = curthread;
1323 	char *buf, *outbuf;
1324 	int count, buflen, len, error = 0, oldmtu;
1325 
1326 	bzero(&rpbuf, sizeof(rpbuf));
1327 
1328 	/* XXX: This can race with lagg_clone_destroy. */
1329 
1330 	switch (cmd) {
1331 	case SIOCGLAGG:
1332 		LAGG_XLOCK(sc);
1333 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1334 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1335 		ra->ra_proto = sc->sc_proto;
1336 		lagg_proto_request(sc, &ra->ra_psc);
1337 		count = 0;
1338 		buf = outbuf;
1339 		len = min(ra->ra_size, buflen);
1340 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1341 			if (len < sizeof(rpbuf))
1342 				break;
1343 
1344 			lagg_port2req(lp, &rpbuf);
1345 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1346 			count++;
1347 			buf += sizeof(rpbuf);
1348 			len -= sizeof(rpbuf);
1349 		}
1350 		LAGG_XUNLOCK(sc);
1351 		ra->ra_ports = count;
1352 		ra->ra_size = count * sizeof(rpbuf);
1353 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1354 		free(outbuf, M_TEMP);
1355 		break;
1356 	case SIOCSLAGG:
1357 		error = priv_check(td, PRIV_NET_LAGG);
1358 		if (error)
1359 			break;
1360 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1361 			error = EPROTONOSUPPORT;
1362 			break;
1363 		}
1364 		/* Infiniband only supports the failover protocol. */
1365 		if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1366 		    ifp->if_type == IFT_INFINIBAND) {
1367 			error = EPROTONOSUPPORT;
1368 			break;
1369 		}
1370 		LAGG_XLOCK(sc);
1371 		lagg_proto_detach(sc);
1372 		lagg_proto_attach(sc, ra->ra_proto);
1373 		LAGG_XUNLOCK(sc);
1374 		break;
1375 	case SIOCGLAGGOPTS:
1376 		LAGG_XLOCK(sc);
1377 		ro->ro_opts = sc->sc_opts;
1378 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1379 			struct lacp_softc *lsc;
1380 
1381 			lsc = (struct lacp_softc *)sc->sc_psc;
1382 			if (lsc->lsc_debug.lsc_tx_test != 0)
1383 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1384 			if (lsc->lsc_debug.lsc_rx_test != 0)
1385 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1386 			if (lsc->lsc_strict_mode != 0)
1387 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1388 			if (lsc->lsc_fast_timeout != 0)
1389 				ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1390 
1391 			ro->ro_active = sc->sc_active;
1392 		} else {
1393 			ro->ro_active = 0;
1394 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1395 				ro->ro_active += LAGG_PORTACTIVE(lp);
1396 		}
1397 		ro->ro_bkt = sc->sc_stride;
1398 		ro->ro_flapping = sc->sc_flapping;
1399 		ro->ro_flowid_shift = sc->flowid_shift;
1400 		LAGG_XUNLOCK(sc);
1401 		break;
1402 	case SIOCSLAGGOPTS:
1403 		error = priv_check(td, PRIV_NET_LAGG);
1404 		if (error)
1405 			break;
1406 
1407 		/*
1408 		 * The stride option was added without defining a corresponding
1409 		 * LAGG_OPT flag, so handle a non-zero value before checking
1410 		 * anything else to preserve compatibility.
1411 		 */
1412 		LAGG_XLOCK(sc);
1413 		if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1414 			if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1415 				LAGG_XUNLOCK(sc);
1416 				error = EINVAL;
1417 				break;
1418 			}
1419 			sc->sc_stride = ro->ro_bkt;
1420 		}
1421 		if (ro->ro_opts == 0) {
1422 			LAGG_XUNLOCK(sc);
1423 			break;
1424 		}
1425 
1426 		/*
1427 		 * Set options.  LACP options are stored in sc->sc_psc,
1428 		 * not in sc_opts.
1429 		 */
1430 		int valid, lacp;
1431 
1432 		switch (ro->ro_opts) {
1433 		case LAGG_OPT_USE_FLOWID:
1434 		case -LAGG_OPT_USE_FLOWID:
1435 		case LAGG_OPT_USE_NUMA:
1436 		case -LAGG_OPT_USE_NUMA:
1437 		case LAGG_OPT_FLOWIDSHIFT:
1438 		case LAGG_OPT_RR_LIMIT:
1439 			valid = 1;
1440 			lacp = 0;
1441 			break;
1442 		case LAGG_OPT_LACP_TXTEST:
1443 		case -LAGG_OPT_LACP_TXTEST:
1444 		case LAGG_OPT_LACP_RXTEST:
1445 		case -LAGG_OPT_LACP_RXTEST:
1446 		case LAGG_OPT_LACP_STRICT:
1447 		case -LAGG_OPT_LACP_STRICT:
1448 		case LAGG_OPT_LACP_FAST_TIMO:
1449 		case -LAGG_OPT_LACP_FAST_TIMO:
1450 			valid = lacp = 1;
1451 			break;
1452 		default:
1453 			valid = lacp = 0;
1454 			break;
1455 		}
1456 
1457 		if (valid == 0 ||
1458 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1459 			/* Invalid combination of options specified. */
1460 			error = EINVAL;
1461 			LAGG_XUNLOCK(sc);
1462 			break;	/* Return from SIOCSLAGGOPTS. */
1463 		}
1464 
1465 		/*
1466 		 * Store new options into sc->sc_opts except for
1467 		 * FLOWIDSHIFT, RR and LACP options.
1468 		 */
1469 		if (lacp == 0) {
1470 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1471 				sc->flowid_shift = ro->ro_flowid_shift;
1472 			else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1473 				if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1474 				    ro->ro_bkt == 0) {
1475 					error = EINVAL;
1476 					LAGG_XUNLOCK(sc);
1477 					break;
1478 				}
1479 				sc->sc_stride = ro->ro_bkt;
1480 			} else if (ro->ro_opts > 0)
1481 				sc->sc_opts |= ro->ro_opts;
1482 			else
1483 				sc->sc_opts &= ~ro->ro_opts;
1484 		} else {
1485 			struct lacp_softc *lsc;
1486 			struct lacp_port *lp;
1487 
1488 			lsc = (struct lacp_softc *)sc->sc_psc;
1489 
1490 			switch (ro->ro_opts) {
1491 			case LAGG_OPT_LACP_TXTEST:
1492 				lsc->lsc_debug.lsc_tx_test = 1;
1493 				break;
1494 			case -LAGG_OPT_LACP_TXTEST:
1495 				lsc->lsc_debug.lsc_tx_test = 0;
1496 				break;
1497 			case LAGG_OPT_LACP_RXTEST:
1498 				lsc->lsc_debug.lsc_rx_test = 1;
1499 				break;
1500 			case -LAGG_OPT_LACP_RXTEST:
1501 				lsc->lsc_debug.lsc_rx_test = 0;
1502 				break;
1503 			case LAGG_OPT_LACP_STRICT:
1504 				lsc->lsc_strict_mode = 1;
1505 				break;
1506 			case -LAGG_OPT_LACP_STRICT:
1507 				lsc->lsc_strict_mode = 0;
1508 				break;
1509 			case LAGG_OPT_LACP_FAST_TIMO:
1510 				LACP_LOCK(lsc);
1511         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1512                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1513 				LACP_UNLOCK(lsc);
1514 				lsc->lsc_fast_timeout = 1;
1515 				break;
1516 			case -LAGG_OPT_LACP_FAST_TIMO:
1517 				LACP_LOCK(lsc);
1518         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1519                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1520 				LACP_UNLOCK(lsc);
1521 				lsc->lsc_fast_timeout = 0;
1522 				break;
1523 			}
1524 		}
1525 		LAGG_XUNLOCK(sc);
1526 		break;
1527 	case SIOCGLAGGFLAGS:
1528 		rf->rf_flags = 0;
1529 		LAGG_XLOCK(sc);
1530 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1531 			rf->rf_flags |= LAGG_F_HASHL2;
1532 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1533 			rf->rf_flags |= LAGG_F_HASHL3;
1534 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1535 			rf->rf_flags |= LAGG_F_HASHL4;
1536 		LAGG_XUNLOCK(sc);
1537 		break;
1538 	case SIOCSLAGGHASH:
1539 		error = priv_check(td, PRIV_NET_LAGG);
1540 		if (error)
1541 			break;
1542 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1543 			error = EINVAL;
1544 			break;
1545 		}
1546 		LAGG_XLOCK(sc);
1547 		sc->sc_flags = 0;
1548 		if (rf->rf_flags & LAGG_F_HASHL2)
1549 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1550 		if (rf->rf_flags & LAGG_F_HASHL3)
1551 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1552 		if (rf->rf_flags & LAGG_F_HASHL4)
1553 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1554 		LAGG_XUNLOCK(sc);
1555 		break;
1556 	case SIOCGLAGGPORT:
1557 		if (rp->rp_portname[0] == '\0' ||
1558 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1559 			error = EINVAL;
1560 			break;
1561 		}
1562 
1563 		NET_EPOCH_ENTER(et);
1564 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1565 		    lp->lp_softc != sc) {
1566 			error = ENOENT;
1567 			NET_EPOCH_EXIT(et);
1568 			if_rele(tpif);
1569 			break;
1570 		}
1571 
1572 		lagg_port2req(lp, rp);
1573 		NET_EPOCH_EXIT(et);
1574 		if_rele(tpif);
1575 		break;
1576 	case SIOCSLAGGPORT:
1577 		error = priv_check(td, PRIV_NET_LAGG);
1578 		if (error)
1579 			break;
1580 		if (rp->rp_portname[0] == '\0' ||
1581 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1582 			error = EINVAL;
1583 			break;
1584 		}
1585 #ifdef INET6
1586 		/*
1587 		 * A laggport interface should not have inet6 address
1588 		 * because two interfaces with a valid link-local
1589 		 * scope zone must not be merged in any form.  This
1590 		 * restriction is needed to prevent violation of
1591 		 * link-local scope zone.  Attempts to add a laggport
1592 		 * interface which has inet6 addresses triggers
1593 		 * removal of all inet6 addresses on the member
1594 		 * interface.
1595 		 */
1596 		if (in6ifa_llaonifp(tpif)) {
1597 			in6_ifdetach(tpif);
1598 				if_printf(sc->sc_ifp,
1599 				    "IPv6 addresses on %s have been removed "
1600 				    "before adding it as a member to prevent "
1601 				    "IPv6 address scope violation.\n",
1602 				    tpif->if_xname);
1603 		}
1604 #endif
1605 		oldmtu = ifp->if_mtu;
1606 		LAGG_XLOCK(sc);
1607 		error = lagg_port_create(sc, tpif);
1608 		LAGG_XUNLOCK(sc);
1609 		if_rele(tpif);
1610 
1611 		/*
1612 		 * LAGG MTU may change during addition of the first port.
1613 		 * If it did, do network layer specific procedure.
1614 		 */
1615 		if (ifp->if_mtu != oldmtu) {
1616 #ifdef INET6
1617 			nd6_setmtu(ifp);
1618 #endif
1619 			rt_updatemtu(ifp);
1620 		}
1621 
1622 		VLAN_CAPABILITIES(ifp);
1623 		break;
1624 	case SIOCSLAGGDELPORT:
1625 		error = priv_check(td, PRIV_NET_LAGG);
1626 		if (error)
1627 			break;
1628 		if (rp->rp_portname[0] == '\0' ||
1629 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1630 			error = EINVAL;
1631 			break;
1632 		}
1633 
1634 		LAGG_XLOCK(sc);
1635 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1636 		    lp->lp_softc != sc) {
1637 			error = ENOENT;
1638 			LAGG_XUNLOCK(sc);
1639 			if_rele(tpif);
1640 			break;
1641 		}
1642 
1643 		error = lagg_port_destroy(lp, 1);
1644 		LAGG_XUNLOCK(sc);
1645 		if_rele(tpif);
1646 		VLAN_CAPABILITIES(ifp);
1647 		break;
1648 	case SIOCSIFFLAGS:
1649 		/* Set flags on ports too */
1650 		LAGG_XLOCK(sc);
1651 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1652 			lagg_setflags(lp, 1);
1653 		}
1654 
1655 		if (!(ifp->if_flags & IFF_UP) &&
1656 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1657 			/*
1658 			 * If interface is marked down and it is running,
1659 			 * then stop and disable it.
1660 			 */
1661 			lagg_stop(sc);
1662 			LAGG_XUNLOCK(sc);
1663 		} else if ((ifp->if_flags & IFF_UP) &&
1664 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1665 			/*
1666 			 * If interface is marked up and it is stopped, then
1667 			 * start it.
1668 			 */
1669 			LAGG_XUNLOCK(sc);
1670 			(*ifp->if_init)(sc);
1671 		} else
1672 			LAGG_XUNLOCK(sc);
1673 		break;
1674 	case SIOCADDMULTI:
1675 	case SIOCDELMULTI:
1676 		LAGG_XLOCK(sc);
1677 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1678 			lagg_clrmulti(lp);
1679 			lagg_setmulti(lp);
1680 		}
1681 		LAGG_XUNLOCK(sc);
1682 		error = 0;
1683 		break;
1684 	case SIOCSIFMEDIA:
1685 	case SIOCGIFMEDIA:
1686 		if (ifp->if_type == IFT_INFINIBAND)
1687 			error = EINVAL;
1688 		else
1689 			error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1690 		break;
1691 
1692 	case SIOCSIFCAP:
1693 		LAGG_XLOCK(sc);
1694 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1695 			if (lp->lp_ioctl != NULL)
1696 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1697 		}
1698 		lagg_capabilities(sc);
1699 		LAGG_XUNLOCK(sc);
1700 		VLAN_CAPABILITIES(ifp);
1701 		error = 0;
1702 		break;
1703 
1704 	case SIOCSIFMTU:
1705 		LAGG_XLOCK(sc);
1706 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1707 			if (lp->lp_ioctl != NULL)
1708 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1709 			else
1710 				error = EINVAL;
1711 			if (error != 0) {
1712 				if_printf(ifp,
1713 				    "failed to change MTU to %d on port %s, "
1714 				    "reverting all ports to original MTU (%d)\n",
1715 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1716 				break;
1717 			}
1718 		}
1719 		if (error == 0) {
1720 			ifp->if_mtu = ifr->ifr_mtu;
1721 		} else {
1722 			/* set every port back to the original MTU */
1723 			ifr->ifr_mtu = ifp->if_mtu;
1724 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1725 				if (lp->lp_ioctl != NULL)
1726 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1727 			}
1728 		}
1729 		LAGG_XUNLOCK(sc);
1730 		break;
1731 
1732 	default:
1733 		error = ether_ioctl(ifp, cmd, data);
1734 		break;
1735 	}
1736 	return (error);
1737 }
1738 
1739 #if defined(KERN_TLS) || defined(RATELIMIT)
1740 static inline struct lagg_snd_tag *
1741 mst_to_lst(struct m_snd_tag *mst)
1742 {
1743 
1744 	return (__containerof(mst, struct lagg_snd_tag, com));
1745 }
1746 
1747 /*
1748  * Look up the port used by a specific flow.  This only works for lagg
1749  * protocols with deterministic port mappings (e.g. not roundrobin).
1750  * In addition protocols which use a hash to map flows to ports must
1751  * be configured to use the mbuf flowid rather than hashing packet
1752  * contents.
1753  */
1754 static struct lagg_port *
1755 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1756     uint8_t numa_domain)
1757 {
1758 	struct lagg_softc *sc;
1759 	struct lagg_port *lp;
1760 	struct lagg_lb *lb;
1761 	uint32_t hash, p;
1762 	int err;
1763 
1764 	sc = ifp->if_softc;
1765 
1766 	switch (sc->sc_proto) {
1767 	case LAGG_PROTO_FAILOVER:
1768 		return (lagg_link_active(sc, sc->sc_primary));
1769 	case LAGG_PROTO_LOADBALANCE:
1770 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1771 		    flowtype == M_HASHTYPE_NONE)
1772 			return (NULL);
1773 		p = flowid >> sc->flowid_shift;
1774 		p %= sc->sc_count;
1775 		lb = (struct lagg_lb *)sc->sc_psc;
1776 		lp = lb->lb_ports[p];
1777 		return (lagg_link_active(sc, lp));
1778 	case LAGG_PROTO_LACP:
1779 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1780 		    flowtype == M_HASHTYPE_NONE)
1781 			return (NULL);
1782 		hash = flowid >> sc->flowid_shift;
1783 		return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, &err));
1784 	default:
1785 		return (NULL);
1786 	}
1787 }
1788 
1789 static int
1790 lagg_snd_tag_alloc(struct ifnet *ifp,
1791     union if_snd_tag_alloc_params *params,
1792     struct m_snd_tag **ppmt)
1793 {
1794 	struct epoch_tracker et;
1795 	struct lagg_snd_tag *lst;
1796 	struct lagg_softc *sc;
1797 	struct lagg_port *lp;
1798 	struct ifnet *lp_ifp;
1799 	int error;
1800 
1801 	sc = ifp->if_softc;
1802 
1803 	NET_EPOCH_ENTER(et);
1804 	lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1805 	    params->hdr.flowtype, params->hdr.numa_domain);
1806 	if (lp == NULL) {
1807 		NET_EPOCH_EXIT(et);
1808 		return (EOPNOTSUPP);
1809 	}
1810 	if (lp->lp_ifp == NULL) {
1811 		NET_EPOCH_EXIT(et);
1812 		return (EOPNOTSUPP);
1813 	}
1814 	lp_ifp = lp->lp_ifp;
1815 	if_ref(lp_ifp);
1816 	NET_EPOCH_EXIT(et);
1817 
1818 	lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1819 	if (lst == NULL) {
1820 		if_rele(lp_ifp);
1821 		return (ENOMEM);
1822 	}
1823 
1824 	error = m_snd_tag_alloc(lp_ifp, params, &lst->tag);
1825 	if_rele(lp_ifp);
1826 	if (error) {
1827 		free(lst, M_LAGG);
1828 		return (error);
1829 	}
1830 
1831 	m_snd_tag_init(&lst->com, ifp, lst->tag->type);
1832 
1833 	*ppmt = &lst->com;
1834 	return (0);
1835 }
1836 
1837 static int
1838 lagg_snd_tag_modify(struct m_snd_tag *mst,
1839     union if_snd_tag_modify_params *params)
1840 {
1841 	struct lagg_snd_tag *lst;
1842 
1843 	lst = mst_to_lst(mst);
1844 	return (lst->tag->ifp->if_snd_tag_modify(lst->tag, params));
1845 }
1846 
1847 static int
1848 lagg_snd_tag_query(struct m_snd_tag *mst,
1849     union if_snd_tag_query_params *params)
1850 {
1851 	struct lagg_snd_tag *lst;
1852 
1853 	lst = mst_to_lst(mst);
1854 	return (lst->tag->ifp->if_snd_tag_query(lst->tag, params));
1855 }
1856 
1857 static void
1858 lagg_snd_tag_free(struct m_snd_tag *mst)
1859 {
1860 	struct lagg_snd_tag *lst;
1861 
1862 	lst = mst_to_lst(mst);
1863 	m_snd_tag_rele(lst->tag);
1864 	free(lst, M_LAGG);
1865 }
1866 
1867 static void
1868 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1869 {
1870 	/*
1871 	 * For lagg, we have an indirect
1872 	 * interface. The caller needs to
1873 	 * get a ratelimit tag on the actual
1874 	 * interface the flow will go on.
1875 	 */
1876 	q->rate_table = NULL;
1877 	q->flags = RT_IS_INDIRECT;
1878 	q->max_flows = 0;
1879 	q->number_of_rates = 0;
1880 }
1881 #endif
1882 
1883 static int
1884 lagg_setmulti(struct lagg_port *lp)
1885 {
1886 	struct lagg_softc *sc = lp->lp_softc;
1887 	struct ifnet *ifp = lp->lp_ifp;
1888 	struct ifnet *scifp = sc->sc_ifp;
1889 	struct lagg_mc *mc;
1890 	struct ifmultiaddr *ifma;
1891 	int error;
1892 
1893 	IF_ADDR_WLOCK(scifp);
1894 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1895 		if (ifma->ifma_addr->sa_family != AF_LINK)
1896 			continue;
1897 		mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
1898 		if (mc == NULL) {
1899 			IF_ADDR_WUNLOCK(scifp);
1900 			return (ENOMEM);
1901 		}
1902 		bcopy(ifma->ifma_addr, &mc->mc_addr,
1903 		    ifma->ifma_addr->sa_len);
1904 		mc->mc_addr.sdl_index = ifp->if_index;
1905 		mc->mc_ifma = NULL;
1906 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1907 	}
1908 	IF_ADDR_WUNLOCK(scifp);
1909 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1910 		error = if_addmulti(ifp,
1911 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1912 		if (error)
1913 			return (error);
1914 	}
1915 	return (0);
1916 }
1917 
1918 static int
1919 lagg_clrmulti(struct lagg_port *lp)
1920 {
1921 	struct lagg_mc *mc;
1922 
1923 	LAGG_XLOCK_ASSERT(lp->lp_softc);
1924 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1925 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1926 		if (mc->mc_ifma && lp->lp_detaching == 0)
1927 			if_delmulti_ifma(mc->mc_ifma);
1928 		free(mc, M_LAGG);
1929 	}
1930 	return (0);
1931 }
1932 
1933 static int
1934 lagg_setcaps(struct lagg_port *lp, int cap)
1935 {
1936 	struct ifreq ifr;
1937 
1938 	if (lp->lp_ifp->if_capenable == cap)
1939 		return (0);
1940 	if (lp->lp_ioctl == NULL)
1941 		return (ENXIO);
1942 	ifr.ifr_reqcap = cap;
1943 	return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1944 }
1945 
1946 /* Handle a ref counted flag that should be set on the lagg port as well */
1947 static int
1948 lagg_setflag(struct lagg_port *lp, int flag, int status,
1949     int (*func)(struct ifnet *, int))
1950 {
1951 	struct lagg_softc *sc = lp->lp_softc;
1952 	struct ifnet *scifp = sc->sc_ifp;
1953 	struct ifnet *ifp = lp->lp_ifp;
1954 	int error;
1955 
1956 	LAGG_XLOCK_ASSERT(sc);
1957 
1958 	status = status ? (scifp->if_flags & flag) : 0;
1959 	/* Now "status" contains the flag value or 0 */
1960 
1961 	/*
1962 	 * See if recorded ports status is different from what
1963 	 * we want it to be.  If it is, flip it.  We record ports
1964 	 * status in lp_ifflags so that we won't clear ports flag
1965 	 * we haven't set.  In fact, we don't clear or set ports
1966 	 * flags directly, but get or release references to them.
1967 	 * That's why we can be sure that recorded flags still are
1968 	 * in accord with actual ports flags.
1969 	 */
1970 	if (status != (lp->lp_ifflags & flag)) {
1971 		error = (*func)(ifp, status);
1972 		if (error)
1973 			return (error);
1974 		lp->lp_ifflags &= ~flag;
1975 		lp->lp_ifflags |= status;
1976 	}
1977 	return (0);
1978 }
1979 
1980 /*
1981  * Handle IFF_* flags that require certain changes on the lagg port
1982  * if "status" is true, update ports flags respective to the lagg
1983  * if "status" is false, forcedly clear the flags set on port.
1984  */
1985 static int
1986 lagg_setflags(struct lagg_port *lp, int status)
1987 {
1988 	int error, i;
1989 
1990 	for (i = 0; lagg_pflags[i].flag; i++) {
1991 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1992 		    status, lagg_pflags[i].func);
1993 		if (error)
1994 			return (error);
1995 	}
1996 	return (0);
1997 }
1998 
1999 static int
2000 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2001 {
2002 	struct epoch_tracker et;
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 	NET_EPOCH_ENTER(et);
2011 	/* We need a Tx algorithm and at least one port */
2012 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2013 		NET_EPOCH_EXIT(et);
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 	NET_EPOCH_EXIT(et);
2023 	return (error);
2024 }
2025 
2026 static int
2027 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2028 {
2029 	struct epoch_tracker et;
2030 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2031 	int error;
2032 
2033 #if defined(KERN_TLS) || defined(RATELIMIT)
2034 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2035 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2036 #endif
2037 	NET_EPOCH_ENTER(et);
2038 	/* We need a Tx algorithm and at least one port */
2039 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2040 		NET_EPOCH_EXIT(et);
2041 		m_freem(m);
2042 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2043 		return (ENXIO);
2044 	}
2045 
2046 	INFINIBAND_BPF_MTAP(ifp, m);
2047 
2048 	error = lagg_proto_start(sc, m);
2049 	NET_EPOCH_EXIT(et);
2050 	return (error);
2051 }
2052 
2053 /*
2054  * The ifp->if_qflush entry point for lagg(4) is no-op.
2055  */
2056 static void
2057 lagg_qflush(struct ifnet *ifp __unused)
2058 {
2059 }
2060 
2061 static struct mbuf *
2062 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2063 {
2064 	struct epoch_tracker et;
2065 	struct lagg_port *lp = ifp->if_lagg;
2066 	struct lagg_softc *sc = lp->lp_softc;
2067 	struct ifnet *scifp = sc->sc_ifp;
2068 
2069 	NET_EPOCH_ENTER(et);
2070 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2071 	    lp->lp_detaching != 0 ||
2072 	    sc->sc_proto == LAGG_PROTO_NONE) {
2073 		NET_EPOCH_EXIT(et);
2074 		m_freem(m);
2075 		return (NULL);
2076 	}
2077 
2078 	ETHER_BPF_MTAP(scifp, m);
2079 
2080 	m = lagg_proto_input(sc, lp, m);
2081 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2082 		m_freem(m);
2083 		m = NULL;
2084 	}
2085 
2086 	NET_EPOCH_EXIT(et);
2087 	return (m);
2088 }
2089 
2090 static struct mbuf *
2091 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2092 {
2093 	struct epoch_tracker et;
2094 	struct lagg_port *lp = ifp->if_lagg;
2095 	struct lagg_softc *sc = lp->lp_softc;
2096 	struct ifnet *scifp = sc->sc_ifp;
2097 
2098 	NET_EPOCH_ENTER(et);
2099 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2100 	    lp->lp_detaching != 0 ||
2101 	    sc->sc_proto == LAGG_PROTO_NONE) {
2102 		NET_EPOCH_EXIT(et);
2103 		m_freem(m);
2104 		return (NULL);
2105 	}
2106 
2107 	INFINIBAND_BPF_MTAP(scifp, m);
2108 
2109 	m = lagg_proto_input(sc, lp, m);
2110 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2111 		m_freem(m);
2112 		m = NULL;
2113 	}
2114 
2115 	NET_EPOCH_EXIT(et);
2116 	return (m);
2117 }
2118 
2119 static int
2120 lagg_media_change(struct ifnet *ifp)
2121 {
2122 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2123 
2124 	if (sc->sc_ifflags & IFF_DEBUG)
2125 		printf("%s\n", __func__);
2126 
2127 	/* Ignore */
2128 	return (0);
2129 }
2130 
2131 static void
2132 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2133 {
2134 	struct epoch_tracker et;
2135 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2136 	struct lagg_port *lp;
2137 
2138 	imr->ifm_status = IFM_AVALID;
2139 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
2140 
2141 	NET_EPOCH_ENTER(et);
2142 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2143 		if (LAGG_PORTACTIVE(lp))
2144 			imr->ifm_status |= IFM_ACTIVE;
2145 	}
2146 	NET_EPOCH_EXIT(et);
2147 }
2148 
2149 static void
2150 lagg_linkstate(struct lagg_softc *sc)
2151 {
2152 	struct epoch_tracker et;
2153 	struct lagg_port *lp;
2154 	int new_link = LINK_STATE_DOWN;
2155 	uint64_t speed;
2156 
2157 	LAGG_XLOCK_ASSERT(sc);
2158 
2159 	/* LACP handles link state itself */
2160 	if (sc->sc_proto == LAGG_PROTO_LACP)
2161 		return;
2162 
2163 	/* Our link is considered up if at least one of our ports is active */
2164 	NET_EPOCH_ENTER(et);
2165 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2166 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2167 			new_link = LINK_STATE_UP;
2168 			break;
2169 		}
2170 	}
2171 	NET_EPOCH_EXIT(et);
2172 	if_link_state_change(sc->sc_ifp, new_link);
2173 
2174 	/* Update if_baudrate to reflect the max possible speed */
2175 	switch (sc->sc_proto) {
2176 		case LAGG_PROTO_FAILOVER:
2177 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2178 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
2179 			break;
2180 		case LAGG_PROTO_ROUNDROBIN:
2181 		case LAGG_PROTO_LOADBALANCE:
2182 		case LAGG_PROTO_BROADCAST:
2183 			speed = 0;
2184 			NET_EPOCH_ENTER(et);
2185 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2186 				speed += lp->lp_ifp->if_baudrate;
2187 			NET_EPOCH_EXIT(et);
2188 			sc->sc_ifp->if_baudrate = speed;
2189 			break;
2190 		case LAGG_PROTO_LACP:
2191 			/* LACP updates if_baudrate itself */
2192 			break;
2193 	}
2194 }
2195 
2196 static void
2197 lagg_port_state(struct ifnet *ifp, int state)
2198 {
2199 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2200 	struct lagg_softc *sc = NULL;
2201 
2202 	if (lp != NULL)
2203 		sc = lp->lp_softc;
2204 	if (sc == NULL)
2205 		return;
2206 
2207 	LAGG_XLOCK(sc);
2208 	lagg_linkstate(sc);
2209 	lagg_proto_linkstate(sc, lp);
2210 	LAGG_XUNLOCK(sc);
2211 }
2212 
2213 struct lagg_port *
2214 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2215 {
2216 	struct lagg_port *lp_next, *rval = NULL;
2217 
2218 	/*
2219 	 * Search a port which reports an active link state.
2220 	 */
2221 
2222 #ifdef INVARIANTS
2223 	/*
2224 	 * This is called with either in the network epoch
2225 	 * or with LAGG_XLOCK(sc) held.
2226 	 */
2227 	if (!in_epoch(net_epoch_preempt))
2228 		LAGG_XLOCK_ASSERT(sc);
2229 #endif
2230 
2231 	if (lp == NULL)
2232 		goto search;
2233 	if (LAGG_PORTACTIVE(lp)) {
2234 		rval = lp;
2235 		goto found;
2236 	}
2237 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2238 	    LAGG_PORTACTIVE(lp_next)) {
2239 		rval = lp_next;
2240 		goto found;
2241 	}
2242 
2243 search:
2244 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2245 		if (LAGG_PORTACTIVE(lp_next)) {
2246 			return (lp_next);
2247 		}
2248 	}
2249 found:
2250 	return (rval);
2251 }
2252 
2253 int
2254 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2255 {
2256 
2257 #if defined(KERN_TLS) || defined(RATELIMIT)
2258 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2259 		struct lagg_snd_tag *lst;
2260 		struct m_snd_tag *mst;
2261 
2262 		mst = m->m_pkthdr.snd_tag;
2263 		lst = mst_to_lst(mst);
2264 		if (lst->tag->ifp != ifp) {
2265 			m_freem(m);
2266 			return (EAGAIN);
2267 		}
2268 		m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2269 		m_snd_tag_rele(mst);
2270 	}
2271 #endif
2272 	return (ifp->if_transmit)(ifp, m);
2273 }
2274 
2275 /*
2276  * Simple round robin aggregation
2277  */
2278 static void
2279 lagg_rr_attach(struct lagg_softc *sc)
2280 {
2281 	sc->sc_seq = 0;
2282 	sc->sc_stride = 1;
2283 }
2284 
2285 static int
2286 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2287 {
2288 	struct lagg_port *lp;
2289 	uint32_t p;
2290 
2291 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
2292 	p /= sc->sc_stride;
2293 	p %= sc->sc_count;
2294 	lp = CK_SLIST_FIRST(&sc->sc_ports);
2295 
2296 	while (p--)
2297 		lp = CK_SLIST_NEXT(lp, lp_entries);
2298 
2299 	/*
2300 	 * Check the port's link state. This will return the next active
2301 	 * port if the link is down or the port is NULL.
2302 	 */
2303 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2304 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2305 		m_freem(m);
2306 		return (ENETDOWN);
2307 	}
2308 
2309 	/* Send mbuf */
2310 	return (lagg_enqueue(lp->lp_ifp, m));
2311 }
2312 
2313 static struct mbuf *
2314 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2315 {
2316 	struct ifnet *ifp = sc->sc_ifp;
2317 
2318 	/* Just pass in the packet to our lagg device */
2319 	m->m_pkthdr.rcvif = ifp;
2320 
2321 	return (m);
2322 }
2323 
2324 /*
2325  * Broadcast mode
2326  */
2327 static int
2328 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2329 {
2330 	int active_ports = 0;
2331 	int errors = 0;
2332 	int ret;
2333 	struct lagg_port *lp, *last = NULL;
2334 	struct mbuf *m0;
2335 
2336 	NET_EPOCH_ASSERT();
2337 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2338 		if (!LAGG_PORTACTIVE(lp))
2339 			continue;
2340 
2341 		active_ports++;
2342 
2343 		if (last != NULL) {
2344 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2345 			if (m0 == NULL) {
2346 				ret = ENOBUFS;
2347 				errors++;
2348 				break;
2349 			}
2350 			lagg_enqueue(last->lp_ifp, m0);
2351 		}
2352 		last = lp;
2353 	}
2354 
2355 	if (last == NULL) {
2356 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2357 		m_freem(m);
2358 		return (ENOENT);
2359 	}
2360 	if ((last = lagg_link_active(sc, last)) == NULL) {
2361 		errors++;
2362 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2363 		m_freem(m);
2364 		return (ENETDOWN);
2365 	}
2366 
2367 	ret = lagg_enqueue(last->lp_ifp, m);
2368 	if (errors != 0)
2369 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2370 
2371 	return (ret);
2372 }
2373 
2374 static struct mbuf*
2375 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2376 {
2377 	struct ifnet *ifp = sc->sc_ifp;
2378 
2379 	/* Just pass in the packet to our lagg device */
2380 	m->m_pkthdr.rcvif = ifp;
2381 	return (m);
2382 }
2383 
2384 /*
2385  * Active failover
2386  */
2387 static int
2388 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2389 {
2390 	struct lagg_port *lp;
2391 
2392 	/* Use the master port if active or the next available port */
2393 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2394 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2395 		m_freem(m);
2396 		return (ENETDOWN);
2397 	}
2398 
2399 	/* Send mbuf */
2400 	return (lagg_enqueue(lp->lp_ifp, m));
2401 }
2402 
2403 static struct mbuf *
2404 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2405 {
2406 	struct ifnet *ifp = sc->sc_ifp;
2407 	struct lagg_port *tmp_tp;
2408 
2409 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2410 		m->m_pkthdr.rcvif = ifp;
2411 		return (m);
2412 	}
2413 
2414 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2415 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2416 		/*
2417 		 * If tmp_tp is null, we've received a packet when all
2418 		 * our links are down. Weird, but process it anyways.
2419 		 */
2420 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2421 			m->m_pkthdr.rcvif = ifp;
2422 			return (m);
2423 		}
2424 	}
2425 
2426 	m_freem(m);
2427 	return (NULL);
2428 }
2429 
2430 /*
2431  * Loadbalancing
2432  */
2433 static void
2434 lagg_lb_attach(struct lagg_softc *sc)
2435 {
2436 	struct lagg_port *lp;
2437 	struct lagg_lb *lb;
2438 
2439 	LAGG_XLOCK_ASSERT(sc);
2440 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2441 	lb->lb_key = m_ether_tcpip_hash_init();
2442 	sc->sc_psc = lb;
2443 
2444 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2445 		lagg_lb_port_create(lp);
2446 }
2447 
2448 static void
2449 lagg_lb_detach(struct lagg_softc *sc)
2450 {
2451 	struct lagg_lb *lb;
2452 
2453 	lb = (struct lagg_lb *)sc->sc_psc;
2454 	if (lb != NULL)
2455 		free(lb, M_LAGG);
2456 }
2457 
2458 static int
2459 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2460 {
2461 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2462 	struct lagg_port *lp_next;
2463 	int i = 0, rv;
2464 
2465 	rv = 0;
2466 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2467 	LAGG_XLOCK_ASSERT(sc);
2468 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2469 		if (lp_next == lp)
2470 			continue;
2471 		if (i >= LAGG_MAX_PORTS) {
2472 			rv = EINVAL;
2473 			break;
2474 		}
2475 		if (sc->sc_ifflags & IFF_DEBUG)
2476 			printf("%s: port %s at index %d\n",
2477 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2478 		lb->lb_ports[i++] = lp_next;
2479 	}
2480 
2481 	return (rv);
2482 }
2483 
2484 static int
2485 lagg_lb_port_create(struct lagg_port *lp)
2486 {
2487 	struct lagg_softc *sc = lp->lp_softc;
2488 	return (lagg_lb_porttable(sc, NULL));
2489 }
2490 
2491 static void
2492 lagg_lb_port_destroy(struct lagg_port *lp)
2493 {
2494 	struct lagg_softc *sc = lp->lp_softc;
2495 	lagg_lb_porttable(sc, lp);
2496 }
2497 
2498 static int
2499 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2500 {
2501 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2502 	struct lagg_port *lp = NULL;
2503 	uint32_t p = 0;
2504 
2505 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2506 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2507 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2508 	else
2509 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2510 	p %= sc->sc_count;
2511 	lp = lb->lb_ports[p];
2512 
2513 	/*
2514 	 * Check the port's link state. This will return the next active
2515 	 * port if the link is down or the port is NULL.
2516 	 */
2517 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2518 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2519 		m_freem(m);
2520 		return (ENETDOWN);
2521 	}
2522 
2523 	/* Send mbuf */
2524 	return (lagg_enqueue(lp->lp_ifp, m));
2525 }
2526 
2527 static struct mbuf *
2528 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2529 {
2530 	struct ifnet *ifp = sc->sc_ifp;
2531 
2532 	/* Just pass in the packet to our lagg device */
2533 	m->m_pkthdr.rcvif = ifp;
2534 
2535 	return (m);
2536 }
2537 
2538 /*
2539  * 802.3ad LACP
2540  */
2541 static void
2542 lagg_lacp_attach(struct lagg_softc *sc)
2543 {
2544 	struct lagg_port *lp;
2545 
2546 	lacp_attach(sc);
2547 	LAGG_XLOCK_ASSERT(sc);
2548 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2549 		lacp_port_create(lp);
2550 }
2551 
2552 static void
2553 lagg_lacp_detach(struct lagg_softc *sc)
2554 {
2555 	struct lagg_port *lp;
2556 	void *psc;
2557 
2558 	LAGG_XLOCK_ASSERT(sc);
2559 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2560 		lacp_port_destroy(lp);
2561 
2562 	psc = sc->sc_psc;
2563 	sc->sc_psc = NULL;
2564 	lacp_detach(psc);
2565 }
2566 
2567 static void
2568 lagg_lacp_lladdr(struct lagg_softc *sc)
2569 {
2570 	struct lagg_port *lp;
2571 
2572 	LAGG_SXLOCK_ASSERT(sc);
2573 
2574 	/* purge all the lacp ports */
2575 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2576 		lacp_port_destroy(lp);
2577 
2578 	/* add them back in */
2579 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2580 		lacp_port_create(lp);
2581 }
2582 
2583 static int
2584 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2585 {
2586 	struct lagg_port *lp;
2587 	int err;
2588 
2589 	lp = lacp_select_tx_port(sc, m, &err);
2590 	if (lp == NULL) {
2591 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2592 		m_freem(m);
2593 		return (err);
2594 	}
2595 
2596 	/* Send mbuf */
2597 	return (lagg_enqueue(lp->lp_ifp, m));
2598 }
2599 
2600 static struct mbuf *
2601 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2602 {
2603 	struct ifnet *ifp = sc->sc_ifp;
2604 	struct ether_header *eh;
2605 	u_short etype;
2606 
2607 	eh = mtod(m, struct ether_header *);
2608 	etype = ntohs(eh->ether_type);
2609 
2610 	/* Tap off LACP control messages */
2611 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2612 		m = lacp_input(lp, m);
2613 		if (m == NULL)
2614 			return (NULL);
2615 	}
2616 
2617 	/*
2618 	 * If the port is not collecting or not in the active aggregator then
2619 	 * free and return.
2620 	 */
2621 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2622 		m_freem(m);
2623 		return (NULL);
2624 	}
2625 
2626 	m->m_pkthdr.rcvif = ifp;
2627 	return (m);
2628 }
2629