xref: /freebsd/sys/net/if_lagg.c (revision 1e4896b176ff664dc9c2fce5426bf2fdf8017a7d)
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 			if (lp != NULL)
1149 				return ((*lp->lp_output)(ifp, m, dst, ro));
1150 	}
1151 
1152 	/* drop any other frames */
1153 	m_freem(m);
1154 	return (ENETDOWN);
1155 }
1156 
1157 static void
1158 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1159 {
1160 	struct lagg_port *lp;
1161 	struct lagg_softc *sc;
1162 
1163 	if ((lp = ifp->if_lagg) == NULL)
1164 		return;
1165 	/* If the ifnet is just being renamed, don't do anything. */
1166 	if (ifp->if_flags & IFF_RENAMING)
1167 		return;
1168 
1169 	sc = lp->lp_softc;
1170 
1171 	LAGG_XLOCK(sc);
1172 	lp->lp_detaching = 1;
1173 	lagg_port_destroy(lp, 1);
1174 	LAGG_XUNLOCK(sc);
1175 	VLAN_CAPABILITIES(sc->sc_ifp);
1176 }
1177 
1178 static void
1179 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1180 {
1181 	struct lagg_softc *sc = lp->lp_softc;
1182 
1183 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1184 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1185 	rp->rp_prio = lp->lp_prio;
1186 	rp->rp_flags = lp->lp_flags;
1187 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1188 
1189 	/* Add protocol specific flags */
1190 	switch (sc->sc_proto) {
1191 		case LAGG_PROTO_FAILOVER:
1192 			if (lp == sc->sc_primary)
1193 				rp->rp_flags |= LAGG_PORT_MASTER;
1194 			if (lp == lagg_link_active(sc, sc->sc_primary))
1195 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1196 			break;
1197 
1198 		case LAGG_PROTO_ROUNDROBIN:
1199 		case LAGG_PROTO_LOADBALANCE:
1200 		case LAGG_PROTO_BROADCAST:
1201 			if (LAGG_PORTACTIVE(lp))
1202 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1203 			break;
1204 
1205 		case LAGG_PROTO_LACP:
1206 			/* LACP has a different definition of active */
1207 			if (lacp_isactive(lp))
1208 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1209 			if (lacp_iscollecting(lp))
1210 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1211 			if (lacp_isdistributing(lp))
1212 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1213 			break;
1214 	}
1215 
1216 }
1217 
1218 static void
1219 lagg_watchdog_infiniband(void *arg)
1220 {
1221 	struct lagg_softc *sc;
1222 	struct lagg_port *lp;
1223 	struct ifnet *ifp;
1224 	struct ifnet *lp_ifp;
1225 
1226 	sc = arg;
1227 
1228 	/*
1229 	 * Because infiniband nodes have a fixed MAC address, which is
1230 	 * generated by the so-called GID, we need to regularly update
1231 	 * the link level address of the parent lagg<N> device when
1232 	 * the active port changes. Possibly we could piggy-back on
1233 	 * link up/down events aswell, but using a timer also provides
1234 	 * a guarantee against too frequent events. This operation
1235 	 * does not have to be atomic.
1236 	 */
1237 	LAGG_RLOCK();
1238 	lp = lagg_link_active(sc, sc->sc_primary);
1239 	if (lp != NULL) {
1240 		ifp = sc->sc_ifp;
1241 		lp_ifp = lp->lp_ifp;
1242 
1243 		if (ifp != NULL && lp_ifp != NULL &&
1244 		    memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0) {
1245 			memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1246 			CURVNET_SET(ifp->if_vnet);
1247 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1248 			CURVNET_RESTORE();
1249 		}
1250 	}
1251 	LAGG_RUNLOCK();
1252 
1253 	callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1254 }
1255 
1256 static void
1257 lagg_init(void *xsc)
1258 {
1259 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1260 	struct ifnet *ifp = sc->sc_ifp;
1261 	struct lagg_port *lp;
1262 
1263 	LAGG_XLOCK(sc);
1264 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1265 		LAGG_XUNLOCK(sc);
1266 		return;
1267 	}
1268 
1269 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1270 
1271 	/*
1272 	 * Update the port lladdrs if needed.
1273 	 * This might be if_setlladdr() notification
1274 	 * that lladdr has been changed.
1275 	 */
1276 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1277 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1278 		    ifp->if_addrlen) != 0)
1279 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1280 	}
1281 
1282 	lagg_proto_init(sc);
1283 
1284 	if (ifp->if_type == IFT_INFINIBAND) {
1285 		mtx_lock(&sc->sc_mtx);
1286 		lagg_watchdog_infiniband(sc);
1287 		mtx_unlock(&sc->sc_mtx);
1288 	}
1289 
1290 	LAGG_XUNLOCK(sc);
1291 }
1292 
1293 static void
1294 lagg_stop(struct lagg_softc *sc)
1295 {
1296 	struct ifnet *ifp = sc->sc_ifp;
1297 
1298 	LAGG_XLOCK_ASSERT(sc);
1299 
1300 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1301 		return;
1302 
1303 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1304 
1305 	lagg_proto_stop(sc);
1306 
1307 	mtx_lock(&sc->sc_mtx);
1308 	callout_stop(&sc->sc_watchdog);
1309 	mtx_unlock(&sc->sc_mtx);
1310 
1311 	callout_drain(&sc->sc_watchdog);
1312 }
1313 
1314 static int
1315 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1316 {
1317 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1318 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1319 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1320 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1321 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1322 	struct ifreq *ifr = (struct ifreq *)data;
1323 	struct lagg_port *lp;
1324 	struct ifnet *tpif;
1325 	struct thread *td = curthread;
1326 	char *buf, *outbuf;
1327 	int count, buflen, len, error = 0, oldmtu;
1328 
1329 	bzero(&rpbuf, sizeof(rpbuf));
1330 
1331 	/* XXX: This can race with lagg_clone_destroy. */
1332 
1333 	switch (cmd) {
1334 	case SIOCGLAGG:
1335 		LAGG_XLOCK(sc);
1336 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1337 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1338 		ra->ra_proto = sc->sc_proto;
1339 		lagg_proto_request(sc, &ra->ra_psc);
1340 		count = 0;
1341 		buf = outbuf;
1342 		len = min(ra->ra_size, buflen);
1343 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1344 			if (len < sizeof(rpbuf))
1345 				break;
1346 
1347 			lagg_port2req(lp, &rpbuf);
1348 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1349 			count++;
1350 			buf += sizeof(rpbuf);
1351 			len -= sizeof(rpbuf);
1352 		}
1353 		LAGG_XUNLOCK(sc);
1354 		ra->ra_ports = count;
1355 		ra->ra_size = count * sizeof(rpbuf);
1356 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1357 		free(outbuf, M_TEMP);
1358 		break;
1359 	case SIOCSLAGG:
1360 		error = priv_check(td, PRIV_NET_LAGG);
1361 		if (error)
1362 			break;
1363 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1364 			error = EPROTONOSUPPORT;
1365 			break;
1366 		}
1367 		/* Infiniband only supports the failover protocol. */
1368 		if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1369 		    ifp->if_type == IFT_INFINIBAND) {
1370 			error = EPROTONOSUPPORT;
1371 			break;
1372 		}
1373 		LAGG_XLOCK(sc);
1374 		lagg_proto_detach(sc);
1375 		LAGG_UNLOCK_ASSERT();
1376 		lagg_proto_attach(sc, ra->ra_proto);
1377 		LAGG_XUNLOCK(sc);
1378 		break;
1379 	case SIOCGLAGGOPTS:
1380 		LAGG_XLOCK(sc);
1381 		ro->ro_opts = sc->sc_opts;
1382 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1383 			struct lacp_softc *lsc;
1384 
1385 			lsc = (struct lacp_softc *)sc->sc_psc;
1386 			if (lsc->lsc_debug.lsc_tx_test != 0)
1387 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1388 			if (lsc->lsc_debug.lsc_rx_test != 0)
1389 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1390 			if (lsc->lsc_strict_mode != 0)
1391 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1392 			if (lsc->lsc_fast_timeout != 0)
1393 				ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1394 
1395 			ro->ro_active = sc->sc_active;
1396 		} else {
1397 			ro->ro_active = 0;
1398 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1399 				ro->ro_active += LAGG_PORTACTIVE(lp);
1400 		}
1401 		ro->ro_bkt = sc->sc_stride;
1402 		ro->ro_flapping = sc->sc_flapping;
1403 		ro->ro_flowid_shift = sc->flowid_shift;
1404 		LAGG_XUNLOCK(sc);
1405 		break;
1406 	case SIOCSLAGGOPTS:
1407 		error = priv_check(td, PRIV_NET_LAGG);
1408 		if (error)
1409 			break;
1410 
1411 		/*
1412 		 * The stride option was added without defining a corresponding
1413 		 * LAGG_OPT flag, so handle a non-zero value before checking
1414 		 * anything else to preserve compatibility.
1415 		 */
1416 		LAGG_XLOCK(sc);
1417 		if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1418 			if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1419 				LAGG_XUNLOCK(sc);
1420 				error = EINVAL;
1421 				break;
1422 			}
1423 			sc->sc_stride = ro->ro_bkt;
1424 		}
1425 		if (ro->ro_opts == 0) {
1426 			LAGG_XUNLOCK(sc);
1427 			break;
1428 		}
1429 
1430 		/*
1431 		 * Set options.  LACP options are stored in sc->sc_psc,
1432 		 * not in sc_opts.
1433 		 */
1434 		int valid, lacp;
1435 
1436 		switch (ro->ro_opts) {
1437 		case LAGG_OPT_USE_FLOWID:
1438 		case -LAGG_OPT_USE_FLOWID:
1439 		case LAGG_OPT_USE_NUMA:
1440 		case -LAGG_OPT_USE_NUMA:
1441 		case LAGG_OPT_FLOWIDSHIFT:
1442 		case LAGG_OPT_RR_LIMIT:
1443 			valid = 1;
1444 			lacp = 0;
1445 			break;
1446 		case LAGG_OPT_LACP_TXTEST:
1447 		case -LAGG_OPT_LACP_TXTEST:
1448 		case LAGG_OPT_LACP_RXTEST:
1449 		case -LAGG_OPT_LACP_RXTEST:
1450 		case LAGG_OPT_LACP_STRICT:
1451 		case -LAGG_OPT_LACP_STRICT:
1452 		case LAGG_OPT_LACP_FAST_TIMO:
1453 		case -LAGG_OPT_LACP_FAST_TIMO:
1454 			valid = lacp = 1;
1455 			break;
1456 		default:
1457 			valid = lacp = 0;
1458 			break;
1459 		}
1460 
1461 		if (valid == 0 ||
1462 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1463 			/* Invalid combination of options specified. */
1464 			error = EINVAL;
1465 			LAGG_XUNLOCK(sc);
1466 			break;	/* Return from SIOCSLAGGOPTS. */
1467 		}
1468 
1469 		/*
1470 		 * Store new options into sc->sc_opts except for
1471 		 * FLOWIDSHIFT, RR and LACP options.
1472 		 */
1473 		if (lacp == 0) {
1474 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1475 				sc->flowid_shift = ro->ro_flowid_shift;
1476 			else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1477 				if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1478 				    ro->ro_bkt == 0) {
1479 					error = EINVAL;
1480 					LAGG_XUNLOCK(sc);
1481 					break;
1482 				}
1483 				sc->sc_stride = ro->ro_bkt;
1484 			} else if (ro->ro_opts > 0)
1485 				sc->sc_opts |= ro->ro_opts;
1486 			else
1487 				sc->sc_opts &= ~ro->ro_opts;
1488 		} else {
1489 			struct lacp_softc *lsc;
1490 			struct lacp_port *lp;
1491 
1492 			lsc = (struct lacp_softc *)sc->sc_psc;
1493 
1494 			switch (ro->ro_opts) {
1495 			case LAGG_OPT_LACP_TXTEST:
1496 				lsc->lsc_debug.lsc_tx_test = 1;
1497 				break;
1498 			case -LAGG_OPT_LACP_TXTEST:
1499 				lsc->lsc_debug.lsc_tx_test = 0;
1500 				break;
1501 			case LAGG_OPT_LACP_RXTEST:
1502 				lsc->lsc_debug.lsc_rx_test = 1;
1503 				break;
1504 			case -LAGG_OPT_LACP_RXTEST:
1505 				lsc->lsc_debug.lsc_rx_test = 0;
1506 				break;
1507 			case LAGG_OPT_LACP_STRICT:
1508 				lsc->lsc_strict_mode = 1;
1509 				break;
1510 			case -LAGG_OPT_LACP_STRICT:
1511 				lsc->lsc_strict_mode = 0;
1512 				break;
1513 			case LAGG_OPT_LACP_FAST_TIMO:
1514 				LACP_LOCK(lsc);
1515         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1516                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1517 				LACP_UNLOCK(lsc);
1518 				lsc->lsc_fast_timeout = 1;
1519 				break;
1520 			case -LAGG_OPT_LACP_FAST_TIMO:
1521 				LACP_LOCK(lsc);
1522         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1523                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1524 				LACP_UNLOCK(lsc);
1525 				lsc->lsc_fast_timeout = 0;
1526 				break;
1527 			}
1528 		}
1529 		LAGG_XUNLOCK(sc);
1530 		break;
1531 	case SIOCGLAGGFLAGS:
1532 		rf->rf_flags = 0;
1533 		LAGG_XLOCK(sc);
1534 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1535 			rf->rf_flags |= LAGG_F_HASHL2;
1536 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1537 			rf->rf_flags |= LAGG_F_HASHL3;
1538 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1539 			rf->rf_flags |= LAGG_F_HASHL4;
1540 		LAGG_XUNLOCK(sc);
1541 		break;
1542 	case SIOCSLAGGHASH:
1543 		error = priv_check(td, PRIV_NET_LAGG);
1544 		if (error)
1545 			break;
1546 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1547 			error = EINVAL;
1548 			break;
1549 		}
1550 		LAGG_XLOCK(sc);
1551 		sc->sc_flags = 0;
1552 		if (rf->rf_flags & LAGG_F_HASHL2)
1553 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1554 		if (rf->rf_flags & LAGG_F_HASHL3)
1555 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1556 		if (rf->rf_flags & LAGG_F_HASHL4)
1557 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1558 		LAGG_XUNLOCK(sc);
1559 		break;
1560 	case SIOCGLAGGPORT:
1561 		if (rp->rp_portname[0] == '\0' ||
1562 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1563 			error = EINVAL;
1564 			break;
1565 		}
1566 
1567 		LAGG_RLOCK();
1568 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1569 		    lp->lp_softc != sc) {
1570 			error = ENOENT;
1571 			LAGG_RUNLOCK();
1572 			if_rele(tpif);
1573 			break;
1574 		}
1575 
1576 		lagg_port2req(lp, rp);
1577 		LAGG_RUNLOCK();
1578 		if_rele(tpif);
1579 		break;
1580 	case SIOCSLAGGPORT:
1581 		error = priv_check(td, PRIV_NET_LAGG);
1582 		if (error)
1583 			break;
1584 		if (rp->rp_portname[0] == '\0' ||
1585 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1586 			error = EINVAL;
1587 			break;
1588 		}
1589 #ifdef INET6
1590 		/*
1591 		 * A laggport interface should not have inet6 address
1592 		 * because two interfaces with a valid link-local
1593 		 * scope zone must not be merged in any form.  This
1594 		 * restriction is needed to prevent violation of
1595 		 * link-local scope zone.  Attempts to add a laggport
1596 		 * interface which has inet6 addresses triggers
1597 		 * removal of all inet6 addresses on the member
1598 		 * interface.
1599 		 */
1600 		if (in6ifa_llaonifp(tpif)) {
1601 			in6_ifdetach(tpif);
1602 				if_printf(sc->sc_ifp,
1603 				    "IPv6 addresses on %s have been removed "
1604 				    "before adding it as a member to prevent "
1605 				    "IPv6 address scope violation.\n",
1606 				    tpif->if_xname);
1607 		}
1608 #endif
1609 		oldmtu = ifp->if_mtu;
1610 		LAGG_XLOCK(sc);
1611 		error = lagg_port_create(sc, tpif);
1612 		LAGG_XUNLOCK(sc);
1613 		if_rele(tpif);
1614 
1615 		/*
1616 		 * LAGG MTU may change during addition of the first port.
1617 		 * If it did, do network layer specific procedure.
1618 		 */
1619 		if (ifp->if_mtu != oldmtu) {
1620 #ifdef INET6
1621 			nd6_setmtu(ifp);
1622 #endif
1623 			rt_updatemtu(ifp);
1624 		}
1625 
1626 		VLAN_CAPABILITIES(ifp);
1627 		break;
1628 	case SIOCSLAGGDELPORT:
1629 		error = priv_check(td, PRIV_NET_LAGG);
1630 		if (error)
1631 			break;
1632 		if (rp->rp_portname[0] == '\0' ||
1633 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1634 			error = EINVAL;
1635 			break;
1636 		}
1637 
1638 		LAGG_XLOCK(sc);
1639 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1640 		    lp->lp_softc != sc) {
1641 			error = ENOENT;
1642 			LAGG_XUNLOCK(sc);
1643 			if_rele(tpif);
1644 			break;
1645 		}
1646 
1647 		error = lagg_port_destroy(lp, 1);
1648 		LAGG_XUNLOCK(sc);
1649 		if_rele(tpif);
1650 		VLAN_CAPABILITIES(ifp);
1651 		break;
1652 	case SIOCSIFFLAGS:
1653 		/* Set flags on ports too */
1654 		LAGG_XLOCK(sc);
1655 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1656 			lagg_setflags(lp, 1);
1657 		}
1658 
1659 		if (!(ifp->if_flags & IFF_UP) &&
1660 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1661 			/*
1662 			 * If interface is marked down and it is running,
1663 			 * then stop and disable it.
1664 			 */
1665 			lagg_stop(sc);
1666 			LAGG_XUNLOCK(sc);
1667 		} else if ((ifp->if_flags & IFF_UP) &&
1668 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1669 			/*
1670 			 * If interface is marked up and it is stopped, then
1671 			 * start it.
1672 			 */
1673 			LAGG_XUNLOCK(sc);
1674 			(*ifp->if_init)(sc);
1675 		} else
1676 			LAGG_XUNLOCK(sc);
1677 		break;
1678 	case SIOCADDMULTI:
1679 	case SIOCDELMULTI:
1680 		LAGG_XLOCK(sc);
1681 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1682 			lagg_clrmulti(lp);
1683 			lagg_setmulti(lp);
1684 		}
1685 		LAGG_XUNLOCK(sc);
1686 		error = 0;
1687 		break;
1688 	case SIOCSIFMEDIA:
1689 	case SIOCGIFMEDIA:
1690 		if (ifp->if_type == IFT_INFINIBAND)
1691 			error = EINVAL;
1692 		else
1693 			error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1694 		break;
1695 
1696 	case SIOCSIFCAP:
1697 		LAGG_XLOCK(sc);
1698 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1699 			if (lp->lp_ioctl != NULL)
1700 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1701 		}
1702 		lagg_capabilities(sc);
1703 		LAGG_XUNLOCK(sc);
1704 		VLAN_CAPABILITIES(ifp);
1705 		error = 0;
1706 		break;
1707 
1708 	case SIOCSIFMTU:
1709 		LAGG_XLOCK(sc);
1710 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1711 			if (lp->lp_ioctl != NULL)
1712 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1713 			else
1714 				error = EINVAL;
1715 			if (error != 0) {
1716 				if_printf(ifp,
1717 				    "failed to change MTU to %d on port %s, "
1718 				    "reverting all ports to original MTU (%d)\n",
1719 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1720 				break;
1721 			}
1722 		}
1723 		if (error == 0) {
1724 			ifp->if_mtu = ifr->ifr_mtu;
1725 		} else {
1726 			/* set every port back to the original MTU */
1727 			ifr->ifr_mtu = ifp->if_mtu;
1728 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1729 				if (lp->lp_ioctl != NULL)
1730 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1731 			}
1732 		}
1733 		LAGG_XUNLOCK(sc);
1734 		break;
1735 
1736 	default:
1737 		error = ether_ioctl(ifp, cmd, data);
1738 		break;
1739 	}
1740 	return (error);
1741 }
1742 
1743 #if defined(KERN_TLS) || defined(RATELIMIT)
1744 static inline struct lagg_snd_tag *
1745 mst_to_lst(struct m_snd_tag *mst)
1746 {
1747 
1748 	return (__containerof(mst, struct lagg_snd_tag, com));
1749 }
1750 
1751 /*
1752  * Look up the port used by a specific flow.  This only works for lagg
1753  * protocols with deterministic port mappings (e.g. not roundrobin).
1754  * In addition protocols which use a hash to map flows to ports must
1755  * be configured to use the mbuf flowid rather than hashing packet
1756  * contents.
1757  */
1758 static struct lagg_port *
1759 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1760     uint8_t numa_domain)
1761 {
1762 	struct lagg_softc *sc;
1763 	struct lagg_port *lp;
1764 	struct lagg_lb *lb;
1765 	uint32_t hash, p;
1766 	int err;
1767 
1768 	sc = ifp->if_softc;
1769 
1770 	switch (sc->sc_proto) {
1771 	case LAGG_PROTO_FAILOVER:
1772 		return (lagg_link_active(sc, sc->sc_primary));
1773 	case LAGG_PROTO_LOADBALANCE:
1774 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1775 		    flowtype == M_HASHTYPE_NONE)
1776 			return (NULL);
1777 		p = flowid >> sc->flowid_shift;
1778 		p %= sc->sc_count;
1779 		lb = (struct lagg_lb *)sc->sc_psc;
1780 		lp = lb->lb_ports[p];
1781 		return (lagg_link_active(sc, lp));
1782 	case LAGG_PROTO_LACP:
1783 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1784 		    flowtype == M_HASHTYPE_NONE)
1785 			return (NULL);
1786 		hash = flowid >> sc->flowid_shift;
1787 		return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, &err));
1788 	default:
1789 		return (NULL);
1790 	}
1791 }
1792 
1793 static int
1794 lagg_snd_tag_alloc(struct ifnet *ifp,
1795     union if_snd_tag_alloc_params *params,
1796     struct m_snd_tag **ppmt)
1797 {
1798 	struct lagg_snd_tag *lst;
1799 	struct lagg_softc *sc;
1800 	struct lagg_port *lp;
1801 	struct ifnet *lp_ifp;
1802 	int error;
1803 
1804 	sc = ifp->if_softc;
1805 
1806 	LAGG_RLOCK();
1807 	lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1808 	    params->hdr.flowtype, params->hdr.numa_domain);
1809 	if (lp == NULL) {
1810 		LAGG_RUNLOCK();
1811 		return (EOPNOTSUPP);
1812 	}
1813 	if (lp->lp_ifp == NULL) {
1814 		LAGG_RUNLOCK();
1815 		return (EOPNOTSUPP);
1816 	}
1817 	lp_ifp = lp->lp_ifp;
1818 	if_ref(lp_ifp);
1819 	LAGG_RUNLOCK();
1820 
1821 	lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1822 	if (lst == NULL) {
1823 		if_rele(lp_ifp);
1824 		return (ENOMEM);
1825 	}
1826 
1827 	error = m_snd_tag_alloc(lp_ifp, params, &lst->tag);
1828 	if_rele(lp_ifp);
1829 	if (error) {
1830 		free(lst, M_LAGG);
1831 		return (error);
1832 	}
1833 
1834 	m_snd_tag_init(&lst->com, ifp, lst->tag->type);
1835 
1836 	*ppmt = &lst->com;
1837 	return (0);
1838 }
1839 
1840 static int
1841 lagg_snd_tag_modify(struct m_snd_tag *mst,
1842     union if_snd_tag_modify_params *params)
1843 {
1844 	struct lagg_snd_tag *lst;
1845 
1846 	lst = mst_to_lst(mst);
1847 	return (lst->tag->ifp->if_snd_tag_modify(lst->tag, params));
1848 }
1849 
1850 static int
1851 lagg_snd_tag_query(struct m_snd_tag *mst,
1852     union if_snd_tag_query_params *params)
1853 {
1854 	struct lagg_snd_tag *lst;
1855 
1856 	lst = mst_to_lst(mst);
1857 	return (lst->tag->ifp->if_snd_tag_query(lst->tag, params));
1858 }
1859 
1860 static void
1861 lagg_snd_tag_free(struct m_snd_tag *mst)
1862 {
1863 	struct lagg_snd_tag *lst;
1864 
1865 	lst = mst_to_lst(mst);
1866 	m_snd_tag_rele(lst->tag);
1867 	free(lst, M_LAGG);
1868 }
1869 
1870 static void
1871 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1872 {
1873 	/*
1874 	 * For lagg, we have an indirect
1875 	 * interface. The caller needs to
1876 	 * get a ratelimit tag on the actual
1877 	 * interface the flow will go on.
1878 	 */
1879 	q->rate_table = NULL;
1880 	q->flags = RT_IS_INDIRECT;
1881 	q->max_flows = 0;
1882 	q->number_of_rates = 0;
1883 }
1884 #endif
1885 
1886 static int
1887 lagg_setmulti(struct lagg_port *lp)
1888 {
1889 	struct lagg_softc *sc = lp->lp_softc;
1890 	struct ifnet *ifp = lp->lp_ifp;
1891 	struct ifnet *scifp = sc->sc_ifp;
1892 	struct lagg_mc *mc;
1893 	struct ifmultiaddr *ifma;
1894 	int error;
1895 
1896 	IF_ADDR_WLOCK(scifp);
1897 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1898 		if (ifma->ifma_addr->sa_family != AF_LINK)
1899 			continue;
1900 		mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
1901 		if (mc == NULL) {
1902 			IF_ADDR_WUNLOCK(scifp);
1903 			return (ENOMEM);
1904 		}
1905 		bcopy(ifma->ifma_addr, &mc->mc_addr,
1906 		    ifma->ifma_addr->sa_len);
1907 		mc->mc_addr.sdl_index = ifp->if_index;
1908 		mc->mc_ifma = NULL;
1909 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1910 	}
1911 	IF_ADDR_WUNLOCK(scifp);
1912 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1913 		error = if_addmulti(ifp,
1914 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1915 		if (error)
1916 			return (error);
1917 	}
1918 	return (0);
1919 }
1920 
1921 static int
1922 lagg_clrmulti(struct lagg_port *lp)
1923 {
1924 	struct lagg_mc *mc;
1925 
1926 	LAGG_XLOCK_ASSERT(lp->lp_softc);
1927 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1928 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1929 		if (mc->mc_ifma && lp->lp_detaching == 0)
1930 			if_delmulti_ifma(mc->mc_ifma);
1931 		free(mc, M_LAGG);
1932 	}
1933 	return (0);
1934 }
1935 
1936 static int
1937 lagg_setcaps(struct lagg_port *lp, int cap)
1938 {
1939 	struct ifreq ifr;
1940 
1941 	if (lp->lp_ifp->if_capenable == cap)
1942 		return (0);
1943 	if (lp->lp_ioctl == NULL)
1944 		return (ENXIO);
1945 	ifr.ifr_reqcap = cap;
1946 	return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1947 }
1948 
1949 /* Handle a ref counted flag that should be set on the lagg port as well */
1950 static int
1951 lagg_setflag(struct lagg_port *lp, int flag, int status,
1952     int (*func)(struct ifnet *, int))
1953 {
1954 	struct lagg_softc *sc = lp->lp_softc;
1955 	struct ifnet *scifp = sc->sc_ifp;
1956 	struct ifnet *ifp = lp->lp_ifp;
1957 	int error;
1958 
1959 	LAGG_XLOCK_ASSERT(sc);
1960 
1961 	status = status ? (scifp->if_flags & flag) : 0;
1962 	/* Now "status" contains the flag value or 0 */
1963 
1964 	/*
1965 	 * See if recorded ports status is different from what
1966 	 * we want it to be.  If it is, flip it.  We record ports
1967 	 * status in lp_ifflags so that we won't clear ports flag
1968 	 * we haven't set.  In fact, we don't clear or set ports
1969 	 * flags directly, but get or release references to them.
1970 	 * That's why we can be sure that recorded flags still are
1971 	 * in accord with actual ports flags.
1972 	 */
1973 	if (status != (lp->lp_ifflags & flag)) {
1974 		error = (*func)(ifp, status);
1975 		if (error)
1976 			return (error);
1977 		lp->lp_ifflags &= ~flag;
1978 		lp->lp_ifflags |= status;
1979 	}
1980 	return (0);
1981 }
1982 
1983 /*
1984  * Handle IFF_* flags that require certain changes on the lagg port
1985  * if "status" is true, update ports flags respective to the lagg
1986  * if "status" is false, forcedly clear the flags set on port.
1987  */
1988 static int
1989 lagg_setflags(struct lagg_port *lp, int status)
1990 {
1991 	int error, i;
1992 
1993 	for (i = 0; lagg_pflags[i].flag; i++) {
1994 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1995 		    status, lagg_pflags[i].func);
1996 		if (error)
1997 			return (error);
1998 	}
1999 	return (0);
2000 }
2001 
2002 static int
2003 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2004 {
2005 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2006 	int error;
2007 
2008 #if defined(KERN_TLS) || defined(RATELIMIT)
2009 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2010 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2011 #endif
2012 	LAGG_RLOCK();
2013 	/* We need a Tx algorithm and at least one port */
2014 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2015 		LAGG_RUNLOCK();
2016 		m_freem(m);
2017 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2018 		return (ENXIO);
2019 	}
2020 
2021 	ETHER_BPF_MTAP(ifp, m);
2022 
2023 	error = lagg_proto_start(sc, m);
2024 	LAGG_RUNLOCK();
2025 	return (error);
2026 }
2027 
2028 static int
2029 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2030 {
2031 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2032 	int error;
2033 
2034 #if defined(KERN_TLS) || defined(RATELIMIT)
2035 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2036 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2037 #endif
2038 	LAGG_RLOCK();
2039 	/* We need a Tx algorithm and at least one port */
2040 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2041 		LAGG_RUNLOCK();
2042 		m_freem(m);
2043 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2044 		return (ENXIO);
2045 	}
2046 
2047 	INFINIBAND_BPF_MTAP(ifp, m);
2048 
2049 	error = lagg_proto_start(sc, m);
2050 	LAGG_RUNLOCK();
2051 	return (error);
2052 }
2053 
2054 /*
2055  * The ifp->if_qflush entry point for lagg(4) is no-op.
2056  */
2057 static void
2058 lagg_qflush(struct ifnet *ifp __unused)
2059 {
2060 }
2061 
2062 static struct mbuf *
2063 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2064 {
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 	LAGG_RLOCK();
2070 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2071 	    lp->lp_detaching != 0 ||
2072 	    sc->sc_proto == LAGG_PROTO_NONE) {
2073 		LAGG_RUNLOCK();
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 	LAGG_RUNLOCK();
2087 	return (m);
2088 }
2089 
2090 static struct mbuf *
2091 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2092 {
2093 	struct lagg_port *lp = ifp->if_lagg;
2094 	struct lagg_softc *sc = lp->lp_softc;
2095 	struct ifnet *scifp = sc->sc_ifp;
2096 
2097 	LAGG_RLOCK();
2098 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2099 	    lp->lp_detaching != 0 ||
2100 	    sc->sc_proto == LAGG_PROTO_NONE) {
2101 		LAGG_RUNLOCK();
2102 		m_freem(m);
2103 		return (NULL);
2104 	}
2105 
2106 	INFINIBAND_BPF_MTAP(scifp, m);
2107 
2108 	m = lagg_proto_input(sc, lp, m);
2109 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2110 		m_freem(m);
2111 		m = NULL;
2112 	}
2113 
2114 	LAGG_RUNLOCK();
2115 	return (m);
2116 }
2117 
2118 static int
2119 lagg_media_change(struct ifnet *ifp)
2120 {
2121 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2122 
2123 	if (sc->sc_ifflags & IFF_DEBUG)
2124 		printf("%s\n", __func__);
2125 
2126 	/* Ignore */
2127 	return (0);
2128 }
2129 
2130 static void
2131 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2132 {
2133 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2134 	struct lagg_port *lp;
2135 
2136 	imr->ifm_status = IFM_AVALID;
2137 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
2138 
2139 	LAGG_RLOCK();
2140 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2141 		if (LAGG_PORTACTIVE(lp))
2142 			imr->ifm_status |= IFM_ACTIVE;
2143 	}
2144 	LAGG_RUNLOCK();
2145 }
2146 
2147 static void
2148 lagg_linkstate(struct lagg_softc *sc)
2149 {
2150 	struct lagg_port *lp;
2151 	int new_link = LINK_STATE_DOWN;
2152 	uint64_t speed;
2153 
2154 	LAGG_XLOCK_ASSERT(sc);
2155 
2156 	/* LACP handles link state itself */
2157 	if (sc->sc_proto == LAGG_PROTO_LACP)
2158 		return;
2159 
2160 	/* Our link is considered up if at least one of our ports is active */
2161 	LAGG_RLOCK();
2162 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2163 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2164 			new_link = LINK_STATE_UP;
2165 			break;
2166 		}
2167 	}
2168 	LAGG_RUNLOCK();
2169 	if_link_state_change(sc->sc_ifp, new_link);
2170 
2171 	/* Update if_baudrate to reflect the max possible speed */
2172 	switch (sc->sc_proto) {
2173 		case LAGG_PROTO_FAILOVER:
2174 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2175 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
2176 			break;
2177 		case LAGG_PROTO_ROUNDROBIN:
2178 		case LAGG_PROTO_LOADBALANCE:
2179 		case LAGG_PROTO_BROADCAST:
2180 			speed = 0;
2181 			LAGG_RLOCK();
2182 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2183 				speed += lp->lp_ifp->if_baudrate;
2184 			LAGG_RUNLOCK();
2185 			sc->sc_ifp->if_baudrate = speed;
2186 			break;
2187 		case LAGG_PROTO_LACP:
2188 			/* LACP updates if_baudrate itself */
2189 			break;
2190 	}
2191 }
2192 
2193 static void
2194 lagg_port_state(struct ifnet *ifp, int state)
2195 {
2196 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2197 	struct lagg_softc *sc = NULL;
2198 
2199 	if (lp != NULL)
2200 		sc = lp->lp_softc;
2201 	if (sc == NULL)
2202 		return;
2203 
2204 	LAGG_XLOCK(sc);
2205 	lagg_linkstate(sc);
2206 	lagg_proto_linkstate(sc, lp);
2207 	LAGG_XUNLOCK(sc);
2208 }
2209 
2210 struct lagg_port *
2211 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2212 {
2213 	struct lagg_port *lp_next, *rval = NULL;
2214 
2215 	/*
2216 	 * Search a port which reports an active link state.
2217 	 */
2218 
2219 #ifdef INVARIANTS
2220 	/*
2221 	 * This is called with either LAGG_RLOCK() held or
2222 	 * LAGG_XLOCK(sc) held.
2223 	 */
2224 	if (!in_epoch(net_epoch_preempt))
2225 		LAGG_XLOCK_ASSERT(sc);
2226 #endif
2227 
2228 	if (lp == NULL)
2229 		goto search;
2230 	if (LAGG_PORTACTIVE(lp)) {
2231 		rval = lp;
2232 		goto found;
2233 	}
2234 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2235 	    LAGG_PORTACTIVE(lp_next)) {
2236 		rval = lp_next;
2237 		goto found;
2238 	}
2239 
2240 search:
2241 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2242 		if (LAGG_PORTACTIVE(lp_next)) {
2243 			return (lp_next);
2244 		}
2245 	}
2246 found:
2247 	return (rval);
2248 }
2249 
2250 int
2251 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2252 {
2253 
2254 #if defined(KERN_TLS) || defined(RATELIMIT)
2255 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2256 		struct lagg_snd_tag *lst;
2257 		struct m_snd_tag *mst;
2258 
2259 		mst = m->m_pkthdr.snd_tag;
2260 		lst = mst_to_lst(mst);
2261 		if (lst->tag->ifp != ifp) {
2262 			m_freem(m);
2263 			return (EAGAIN);
2264 		}
2265 		m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2266 		m_snd_tag_rele(mst);
2267 	}
2268 #endif
2269 	return (ifp->if_transmit)(ifp, m);
2270 }
2271 
2272 /*
2273  * Simple round robin aggregation
2274  */
2275 static void
2276 lagg_rr_attach(struct lagg_softc *sc)
2277 {
2278 	sc->sc_seq = 0;
2279 	sc->sc_stride = 1;
2280 }
2281 
2282 static int
2283 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2284 {
2285 	struct lagg_port *lp;
2286 	uint32_t p;
2287 
2288 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
2289 	p /= sc->sc_stride;
2290 	p %= sc->sc_count;
2291 	lp = CK_SLIST_FIRST(&sc->sc_ports);
2292 
2293 	while (p--)
2294 		lp = CK_SLIST_NEXT(lp, lp_entries);
2295 
2296 	/*
2297 	 * Check the port's link state. This will return the next active
2298 	 * port if the link is down or the port is NULL.
2299 	 */
2300 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2301 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2302 		m_freem(m);
2303 		return (ENETDOWN);
2304 	}
2305 
2306 	/* Send mbuf */
2307 	return (lagg_enqueue(lp->lp_ifp, m));
2308 }
2309 
2310 static struct mbuf *
2311 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2312 {
2313 	struct ifnet *ifp = sc->sc_ifp;
2314 
2315 	/* Just pass in the packet to our lagg device */
2316 	m->m_pkthdr.rcvif = ifp;
2317 
2318 	return (m);
2319 }
2320 
2321 /*
2322  * Broadcast mode
2323  */
2324 static int
2325 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2326 {
2327 	int active_ports = 0;
2328 	int errors = 0;
2329 	int ret;
2330 	struct lagg_port *lp, *last = NULL;
2331 	struct mbuf *m0;
2332 
2333 	LAGG_RLOCK_ASSERT();
2334 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2335 		if (!LAGG_PORTACTIVE(lp))
2336 			continue;
2337 
2338 		active_ports++;
2339 
2340 		if (last != NULL) {
2341 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2342 			if (m0 == NULL) {
2343 				ret = ENOBUFS;
2344 				errors++;
2345 				break;
2346 			}
2347 			lagg_enqueue(last->lp_ifp, m0);
2348 		}
2349 		last = lp;
2350 	}
2351 
2352 	if (last == NULL) {
2353 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2354 		m_freem(m);
2355 		return (ENOENT);
2356 	}
2357 	if ((last = lagg_link_active(sc, last)) == NULL) {
2358 		errors++;
2359 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2360 		m_freem(m);
2361 		return (ENETDOWN);
2362 	}
2363 
2364 	ret = lagg_enqueue(last->lp_ifp, m);
2365 	if (errors != 0)
2366 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2367 
2368 	return (ret);
2369 }
2370 
2371 static struct mbuf*
2372 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2373 {
2374 	struct ifnet *ifp = sc->sc_ifp;
2375 
2376 	/* Just pass in the packet to our lagg device */
2377 	m->m_pkthdr.rcvif = ifp;
2378 	return (m);
2379 }
2380 
2381 /*
2382  * Active failover
2383  */
2384 static int
2385 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2386 {
2387 	struct lagg_port *lp;
2388 
2389 	/* Use the master port if active or the next available port */
2390 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2391 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2392 		m_freem(m);
2393 		return (ENETDOWN);
2394 	}
2395 
2396 	/* Send mbuf */
2397 	return (lagg_enqueue(lp->lp_ifp, m));
2398 }
2399 
2400 static struct mbuf *
2401 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2402 {
2403 	struct ifnet *ifp = sc->sc_ifp;
2404 	struct lagg_port *tmp_tp;
2405 
2406 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2407 		m->m_pkthdr.rcvif = ifp;
2408 		return (m);
2409 	}
2410 
2411 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2412 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2413 		/*
2414 		 * If tmp_tp is null, we've received a packet when all
2415 		 * our links are down. Weird, but process it anyways.
2416 		 */
2417 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2418 			m->m_pkthdr.rcvif = ifp;
2419 			return (m);
2420 		}
2421 	}
2422 
2423 	m_freem(m);
2424 	return (NULL);
2425 }
2426 
2427 /*
2428  * Loadbalancing
2429  */
2430 static void
2431 lagg_lb_attach(struct lagg_softc *sc)
2432 {
2433 	struct lagg_port *lp;
2434 	struct lagg_lb *lb;
2435 
2436 	LAGG_XLOCK_ASSERT(sc);
2437 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2438 	lb->lb_key = m_ether_tcpip_hash_init();
2439 	sc->sc_psc = lb;
2440 
2441 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2442 		lagg_lb_port_create(lp);
2443 }
2444 
2445 static void
2446 lagg_lb_detach(struct lagg_softc *sc)
2447 {
2448 	struct lagg_lb *lb;
2449 
2450 	lb = (struct lagg_lb *)sc->sc_psc;
2451 	if (lb != NULL)
2452 		free(lb, M_LAGG);
2453 }
2454 
2455 static int
2456 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2457 {
2458 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2459 	struct lagg_port *lp_next;
2460 	int i = 0, rv;
2461 
2462 	rv = 0;
2463 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2464 	LAGG_XLOCK_ASSERT(sc);
2465 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2466 		if (lp_next == lp)
2467 			continue;
2468 		if (i >= LAGG_MAX_PORTS) {
2469 			rv = EINVAL;
2470 			break;
2471 		}
2472 		if (sc->sc_ifflags & IFF_DEBUG)
2473 			printf("%s: port %s at index %d\n",
2474 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2475 		lb->lb_ports[i++] = lp_next;
2476 	}
2477 
2478 	return (rv);
2479 }
2480 
2481 static int
2482 lagg_lb_port_create(struct lagg_port *lp)
2483 {
2484 	struct lagg_softc *sc = lp->lp_softc;
2485 	return (lagg_lb_porttable(sc, NULL));
2486 }
2487 
2488 static void
2489 lagg_lb_port_destroy(struct lagg_port *lp)
2490 {
2491 	struct lagg_softc *sc = lp->lp_softc;
2492 	lagg_lb_porttable(sc, lp);
2493 }
2494 
2495 static int
2496 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2497 {
2498 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2499 	struct lagg_port *lp = NULL;
2500 	uint32_t p = 0;
2501 
2502 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2503 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2504 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2505 	else
2506 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2507 	p %= sc->sc_count;
2508 	lp = lb->lb_ports[p];
2509 
2510 	/*
2511 	 * Check the port's link state. This will return the next active
2512 	 * port if the link is down or the port is NULL.
2513 	 */
2514 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2515 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2516 		m_freem(m);
2517 		return (ENETDOWN);
2518 	}
2519 
2520 	/* Send mbuf */
2521 	return (lagg_enqueue(lp->lp_ifp, m));
2522 }
2523 
2524 static struct mbuf *
2525 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2526 {
2527 	struct ifnet *ifp = sc->sc_ifp;
2528 
2529 	/* Just pass in the packet to our lagg device */
2530 	m->m_pkthdr.rcvif = ifp;
2531 
2532 	return (m);
2533 }
2534 
2535 /*
2536  * 802.3ad LACP
2537  */
2538 static void
2539 lagg_lacp_attach(struct lagg_softc *sc)
2540 {
2541 	struct lagg_port *lp;
2542 
2543 	lacp_attach(sc);
2544 	LAGG_XLOCK_ASSERT(sc);
2545 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2546 		lacp_port_create(lp);
2547 }
2548 
2549 static void
2550 lagg_lacp_detach(struct lagg_softc *sc)
2551 {
2552 	struct lagg_port *lp;
2553 	void *psc;
2554 
2555 	LAGG_XLOCK_ASSERT(sc);
2556 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2557 		lacp_port_destroy(lp);
2558 
2559 	psc = sc->sc_psc;
2560 	sc->sc_psc = NULL;
2561 	lacp_detach(psc);
2562 }
2563 
2564 static void
2565 lagg_lacp_lladdr(struct lagg_softc *sc)
2566 {
2567 	struct lagg_port *lp;
2568 
2569 	LAGG_SXLOCK_ASSERT(sc);
2570 
2571 	/* purge all the lacp ports */
2572 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2573 		lacp_port_destroy(lp);
2574 
2575 	/* add them back in */
2576 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2577 		lacp_port_create(lp);
2578 }
2579 
2580 static int
2581 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2582 {
2583 	struct lagg_port *lp;
2584 	int err;
2585 
2586 	lp = lacp_select_tx_port(sc, m, &err);
2587 	if (lp == NULL) {
2588 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2589 		m_freem(m);
2590 		return (err);
2591 	}
2592 
2593 	/* Send mbuf */
2594 	return (lagg_enqueue(lp->lp_ifp, m));
2595 }
2596 
2597 static struct mbuf *
2598 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2599 {
2600 	struct ifnet *ifp = sc->sc_ifp;
2601 	struct ether_header *eh;
2602 	u_short etype;
2603 
2604 	eh = mtod(m, struct ether_header *);
2605 	etype = ntohs(eh->ether_type);
2606 
2607 	/* Tap off LACP control messages */
2608 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2609 		m = lacp_input(lp, m);
2610 		if (m == NULL)
2611 			return (NULL);
2612 	}
2613 
2614 	/*
2615 	 * If the port is not collecting or not in the active aggregator then
2616 	 * free and return.
2617 	 */
2618 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2619 		m_freem(m);
2620 		return (NULL);
2621 	}
2622 
2623 	m->m_pkthdr.rcvif = ifp;
2624 	return (m);
2625 }
2626