xref: /freebsd/sys/net/if_lagg.c (revision 3c4ba5f55438f7afd4f4b0b56f88f2bb505fd6a6)
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/if_private.h>
56 #include <net/bpf.h>
57 #include <net/route.h>
58 #include <net/vnet.h>
59 #include <net/infiniband.h>
60 
61 #if defined(INET) || defined(INET6)
62 #include <netinet/in.h>
63 #include <netinet/ip.h>
64 #endif
65 #ifdef INET
66 #include <netinet/in_systm.h>
67 #include <netinet/if_ether.h>
68 #endif
69 
70 #ifdef INET6
71 #include <netinet/ip6.h>
72 #include <netinet6/in6_var.h>
73 #include <netinet6/in6_ifattach.h>
74 #endif
75 
76 #include <net/if_vlan_var.h>
77 #include <net/if_lagg.h>
78 #include <net/ieee8023ad_lacp.h>
79 
80 #ifdef DEV_NETMAP
81 MODULE_DEPEND(if_lagg, netmap, 1, 1, 1);
82 #endif
83 
84 #define	LAGG_SX_INIT(_sc)	sx_init(&(_sc)->sc_sx, "if_lagg sx")
85 #define	LAGG_SX_DESTROY(_sc)	sx_destroy(&(_sc)->sc_sx)
86 #define	LAGG_XLOCK(_sc)		sx_xlock(&(_sc)->sc_sx)
87 #define	LAGG_XUNLOCK(_sc)	sx_xunlock(&(_sc)->sc_sx)
88 #define	LAGG_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
89 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
90 
91 /* Special flags we should propagate to the lagg ports. */
92 static struct {
93 	int flag;
94 	int (*func)(struct ifnet *, int);
95 } lagg_pflags[] = {
96 	{IFF_PROMISC, ifpromisc},
97 	{IFF_ALLMULTI, if_allmulti},
98 	{0, NULL}
99 };
100 
101 struct lagg_snd_tag {
102 	struct m_snd_tag com;
103 	struct m_snd_tag *tag;
104 };
105 
106 VNET_DEFINE_STATIC(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
107 #define	V_lagg_list	VNET(lagg_list)
108 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
109 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
110 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
111 					"if_lagg list", NULL, MTX_DEF)
112 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
113 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
114 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
115 static eventhandler_tag	lagg_detach_cookie = NULL;
116 
117 static int	lagg_clone_create(struct if_clone *, char *, size_t,
118 		    struct ifc_data *, struct ifnet **);
119 static int	lagg_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
120 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
121 #define	V_lagg_cloner	VNET(lagg_cloner)
122 static const char laggname[] = "lagg";
123 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
124 
125 static void	lagg_capabilities(struct lagg_softc *);
126 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
127 static int	lagg_port_destroy(struct lagg_port *, int);
128 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *);
129 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *);
130 static void	lagg_linkstate(struct lagg_softc *);
131 static void	lagg_port_state(struct ifnet *, int);
132 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
133 static int	lagg_port_output(struct ifnet *, struct mbuf *,
134 		    const struct sockaddr *, struct route *);
135 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
136 #ifdef LAGG_PORT_STACKING
137 static int	lagg_port_checkstacking(struct lagg_softc *);
138 #endif
139 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
140 static void	lagg_init(void *);
141 static void	lagg_stop(struct lagg_softc *);
142 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
143 #if defined(KERN_TLS) || defined(RATELIMIT)
144 static int	lagg_snd_tag_alloc(struct ifnet *,
145 		    union if_snd_tag_alloc_params *,
146 		    struct m_snd_tag **);
147 static int	lagg_snd_tag_modify(struct m_snd_tag *,
148 		    union if_snd_tag_modify_params *);
149 static int	lagg_snd_tag_query(struct m_snd_tag *,
150 		    union if_snd_tag_query_params *);
151 static void	lagg_snd_tag_free(struct m_snd_tag *);
152 static struct m_snd_tag *lagg_next_snd_tag(struct m_snd_tag *);
153 static void	lagg_ratelimit_query(struct ifnet *,
154 		    struct if_ratelimit_query_results *);
155 #endif
156 static int	lagg_setmulti(struct lagg_port *);
157 static int	lagg_clrmulti(struct lagg_port *);
158 static void	lagg_setcaps(struct lagg_port *, int cap, int cap2);
159 static int	lagg_setflag(struct lagg_port *, int, int,
160 		    int (*func)(struct ifnet *, int));
161 static int	lagg_setflags(struct lagg_port *, int status);
162 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
163 static int	lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
164 static int	lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
165 static void	lagg_qflush(struct ifnet *);
166 static int	lagg_media_change(struct ifnet *);
167 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
168 static struct lagg_port *lagg_link_active(struct lagg_softc *,
169 		    struct lagg_port *);
170 
171 /* Simple round robin */
172 static void	lagg_rr_attach(struct lagg_softc *);
173 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
174 
175 /* Active failover */
176 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
177 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
178 		    struct mbuf *);
179 
180 /* Loadbalancing */
181 static void	lagg_lb_attach(struct lagg_softc *);
182 static void	lagg_lb_detach(struct lagg_softc *);
183 static int	lagg_lb_port_create(struct lagg_port *);
184 static void	lagg_lb_port_destroy(struct lagg_port *);
185 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
186 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
187 
188 /* Broadcast */
189 static int	lagg_bcast_start(struct lagg_softc *, struct mbuf *);
190 
191 /* 802.3ad LACP */
192 static void	lagg_lacp_attach(struct lagg_softc *);
193 static void	lagg_lacp_detach(struct lagg_softc *);
194 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
195 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
196 		    struct mbuf *);
197 static void	lagg_lacp_lladdr(struct lagg_softc *);
198 
199 /* Default input */
200 static struct mbuf *lagg_default_input(struct lagg_softc *, struct lagg_port *,
201 		    struct mbuf *);
202 
203 /* lagg protocol table */
204 static const struct lagg_proto {
205 	lagg_proto	pr_num;
206 	void		(*pr_attach)(struct lagg_softc *);
207 	void		(*pr_detach)(struct lagg_softc *);
208 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
209 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
210 			    struct mbuf *);
211 	int		(*pr_addport)(struct lagg_port *);
212 	void		(*pr_delport)(struct lagg_port *);
213 	void		(*pr_linkstate)(struct lagg_port *);
214 	void 		(*pr_init)(struct lagg_softc *);
215 	void 		(*pr_stop)(struct lagg_softc *);
216 	void 		(*pr_lladdr)(struct lagg_softc *);
217 	void		(*pr_request)(struct lagg_softc *, void *);
218 	void		(*pr_portreq)(struct lagg_port *, void *);
219 } lagg_protos[] = {
220     {
221 	.pr_num = LAGG_PROTO_NONE
222     },
223     {
224 	.pr_num = LAGG_PROTO_ROUNDROBIN,
225 	.pr_attach = lagg_rr_attach,
226 	.pr_start = lagg_rr_start,
227 	.pr_input = lagg_default_input,
228     },
229     {
230 	.pr_num = LAGG_PROTO_FAILOVER,
231 	.pr_start = lagg_fail_start,
232 	.pr_input = lagg_fail_input,
233     },
234     {
235 	.pr_num = LAGG_PROTO_LOADBALANCE,
236 	.pr_attach = lagg_lb_attach,
237 	.pr_detach = lagg_lb_detach,
238 	.pr_start = lagg_lb_start,
239 	.pr_input = lagg_default_input,
240 	.pr_addport = lagg_lb_port_create,
241 	.pr_delport = lagg_lb_port_destroy,
242     },
243     {
244 	.pr_num = LAGG_PROTO_LACP,
245 	.pr_attach = lagg_lacp_attach,
246 	.pr_detach = lagg_lacp_detach,
247 	.pr_start = lagg_lacp_start,
248 	.pr_input = lagg_lacp_input,
249 	.pr_addport = lacp_port_create,
250 	.pr_delport = lacp_port_destroy,
251 	.pr_linkstate = lacp_linkstate,
252 	.pr_init = lacp_init,
253 	.pr_stop = lacp_stop,
254 	.pr_lladdr = lagg_lacp_lladdr,
255 	.pr_request = lacp_req,
256 	.pr_portreq = lacp_portreq,
257     },
258     {
259 	.pr_num = LAGG_PROTO_BROADCAST,
260 	.pr_start = lagg_bcast_start,
261 	.pr_input = lagg_default_input,
262     },
263 };
264 
265 SYSCTL_DECL(_net_link);
266 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
267     "Link Aggregation");
268 
269 /* Allow input on any failover links */
270 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
271 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
272 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
273     &VNET_NAME(lagg_failover_rx_all), 0,
274     "Accept input from any interface in a failover lagg");
275 
276 /* Default value for using flowid */
277 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
278 #define	V_def_use_flowid	VNET(def_use_flowid)
279 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid,
280     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_use_flowid), 0,
281     "Default setting for using flow id for load sharing");
282 
283 /* Default value for using numa */
284 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
285 #define	V_def_use_numa	VNET(def_use_numa)
286 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa,
287     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_use_numa), 0,
288     "Use numa to steer flows");
289 
290 /* Default value for flowid shift */
291 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
292 #define	V_def_flowid_shift	VNET(def_flowid_shift)
293 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift,
294     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_flowid_shift), 0,
295     "Default setting for flowid shift for load sharing");
296 
297 static void
298 vnet_lagg_init(const void *unused __unused)
299 {
300 
301 	LAGG_LIST_LOCK_INIT();
302 	SLIST_INIT(&V_lagg_list);
303 	struct if_clone_addreq req = {
304 		.create_f = lagg_clone_create,
305 		.destroy_f = lagg_clone_destroy,
306 		.flags = IFC_F_AUTOUNIT,
307 	};
308 	V_lagg_cloner = ifc_attach_cloner(laggname, &req);
309 }
310 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
311     vnet_lagg_init, NULL);
312 
313 static void
314 vnet_lagg_uninit(const void *unused __unused)
315 {
316 
317 	ifc_detach_cloner(V_lagg_cloner);
318 	LAGG_LIST_LOCK_DESTROY();
319 }
320 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
321     vnet_lagg_uninit, NULL);
322 
323 static int
324 lagg_modevent(module_t mod, int type, void *data)
325 {
326 
327 	switch (type) {
328 	case MOD_LOAD:
329 		lagg_input_ethernet_p = lagg_input_ethernet;
330 		lagg_input_infiniband_p = lagg_input_infiniband;
331 		lagg_linkstate_p = lagg_port_state;
332 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
333 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
334 		    EVENTHANDLER_PRI_ANY);
335 		break;
336 	case MOD_UNLOAD:
337 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
338 		    lagg_detach_cookie);
339 		lagg_input_ethernet_p = NULL;
340 		lagg_input_infiniband_p = NULL;
341 		lagg_linkstate_p = NULL;
342 		break;
343 	default:
344 		return (EOPNOTSUPP);
345 	}
346 	return (0);
347 }
348 
349 static moduledata_t lagg_mod = {
350 	"if_lagg",
351 	lagg_modevent,
352 	0
353 };
354 
355 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
356 MODULE_VERSION(if_lagg, 1);
357 MODULE_DEPEND(if_lagg, if_infiniband, 1, 1, 1);
358 
359 static void
360 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
361 {
362 
363 	LAGG_XLOCK_ASSERT(sc);
364 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
365 	    __func__, sc));
366 
367 	if (sc->sc_ifflags & IFF_DEBUG)
368 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
369 
370 	if (lagg_protos[pr].pr_attach != NULL)
371 		lagg_protos[pr].pr_attach(sc);
372 	sc->sc_proto = pr;
373 }
374 
375 static void
376 lagg_proto_detach(struct lagg_softc *sc)
377 {
378 	lagg_proto pr;
379 
380 	LAGG_XLOCK_ASSERT(sc);
381 	pr = sc->sc_proto;
382 	sc->sc_proto = LAGG_PROTO_NONE;
383 
384 	if (lagg_protos[pr].pr_detach != NULL)
385 		lagg_protos[pr].pr_detach(sc);
386 }
387 
388 static inline int
389 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
390 {
391 
392 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
393 }
394 
395 static inline struct mbuf *
396 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
397 {
398 
399 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
400 }
401 
402 static int
403 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
404 {
405 
406 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
407 		return (0);
408 	else
409 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
410 }
411 
412 static void
413 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
414 {
415 
416 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
417 		lagg_protos[sc->sc_proto].pr_delport(lp);
418 }
419 
420 static void
421 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
422 {
423 
424 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
425 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
426 }
427 
428 static void
429 lagg_proto_init(struct lagg_softc *sc)
430 {
431 
432 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
433 		lagg_protos[sc->sc_proto].pr_init(sc);
434 }
435 
436 static void
437 lagg_proto_stop(struct lagg_softc *sc)
438 {
439 
440 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
441 		lagg_protos[sc->sc_proto].pr_stop(sc);
442 }
443 
444 static void
445 lagg_proto_lladdr(struct lagg_softc *sc)
446 {
447 
448 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
449 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
450 }
451 
452 static void
453 lagg_proto_request(struct lagg_softc *sc, void *v)
454 {
455 
456 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
457 		lagg_protos[sc->sc_proto].pr_request(sc, v);
458 }
459 
460 static void
461 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
462 {
463 
464 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
465 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
466 }
467 
468 /*
469  * This routine is run via an vlan
470  * config EVENT
471  */
472 static void
473 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
474 {
475 	struct lagg_softc *sc = ifp->if_softc;
476 	struct lagg_port *lp;
477 
478 	if (ifp->if_softc != arg) /* Not our event */
479 		return;
480 
481 	LAGG_XLOCK(sc);
482 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
483 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
484 	LAGG_XUNLOCK(sc);
485 }
486 
487 /*
488  * This routine is run via an vlan
489  * unconfig EVENT
490  */
491 static void
492 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
493 {
494 	struct lagg_softc *sc = ifp->if_softc;
495 	struct lagg_port *lp;
496 
497 	if (ifp->if_softc != arg) /* Not our event */
498 		return;
499 
500 	LAGG_XLOCK(sc);
501 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
502 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
503 	LAGG_XUNLOCK(sc);
504 }
505 
506 static int
507 lagg_clone_create(struct if_clone *ifc, char *name, size_t len,
508     struct ifc_data *ifd, struct ifnet **ifpp)
509 {
510 	struct iflaggparam iflp;
511 	struct lagg_softc *sc;
512 	struct ifnet *ifp;
513 	int if_type;
514 	int error;
515 	static const uint8_t eaddr[LAGG_ADDR_LEN];
516 
517 	if (ifd->params != NULL) {
518 		error = ifc_copyin(ifd, &iflp, sizeof(iflp));
519 		if (error)
520 			return (error);
521 
522 		switch (iflp.lagg_type) {
523 		case LAGG_TYPE_ETHERNET:
524 			if_type = IFT_ETHER;
525 			break;
526 		case LAGG_TYPE_INFINIBAND:
527 			if_type = IFT_INFINIBAND;
528 			break;
529 		default:
530 			return (EINVAL);
531 		}
532 	} else {
533 		if_type = IFT_ETHER;
534 	}
535 
536 	sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK | M_ZERO);
537 	ifp = sc->sc_ifp = if_alloc(if_type);
538 	if (ifp == NULL) {
539 		free(sc, M_LAGG);
540 		return (ENOSPC);
541 	}
542 	LAGG_SX_INIT(sc);
543 
544 	mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
545 	callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
546 
547 	LAGG_XLOCK(sc);
548 	if (V_def_use_flowid)
549 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
550 	if (V_def_use_numa)
551 		sc->sc_opts |= LAGG_OPT_USE_NUMA;
552 	sc->flowid_shift = V_def_flowid_shift;
553 
554 	/* Hash all layers by default */
555 	sc->sc_flags = MBUF_HASHFLAG_L2 | MBUF_HASHFLAG_L3 | MBUF_HASHFLAG_L4;
556 
557 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
558 
559 	CK_SLIST_INIT(&sc->sc_ports);
560 
561 	switch (if_type) {
562 	case IFT_ETHER:
563 		/* Initialise pseudo media types */
564 		ifmedia_init(&sc->sc_media, 0, lagg_media_change,
565 		    lagg_media_status);
566 		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
567 		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
568 
569 		if_initname(ifp, laggname, ifd->unit);
570 		ifp->if_transmit = lagg_transmit_ethernet;
571 		break;
572 	case IFT_INFINIBAND:
573 		if_initname(ifp, laggname, ifd->unit);
574 		ifp->if_transmit = lagg_transmit_infiniband;
575 		break;
576 	default:
577 		break;
578 	}
579 	ifp->if_softc = sc;
580 	ifp->if_qflush = lagg_qflush;
581 	ifp->if_init = lagg_init;
582 	ifp->if_ioctl = lagg_ioctl;
583 	ifp->if_get_counter = lagg_get_counter;
584 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
585 #if defined(KERN_TLS) || defined(RATELIMIT)
586 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
587 	ifp->if_ratelimit_query = lagg_ratelimit_query;
588 #endif
589 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
590 
591 	/*
592 	 * Attach as an ordinary ethernet device, children will be attached
593 	 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
594 	 */
595 	switch (if_type) {
596 	case IFT_ETHER:
597 		ether_ifattach(ifp, eaddr);
598 		break;
599 	case IFT_INFINIBAND:
600 		infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr);
601 		break;
602 	default:
603 		break;
604 	}
605 
606 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
607 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
608 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
609 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
610 
611 	/* Insert into the global list of laggs */
612 	LAGG_LIST_LOCK();
613 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
614 	LAGG_LIST_UNLOCK();
615 	LAGG_XUNLOCK(sc);
616 	*ifpp = ifp;
617 
618 	return (0);
619 }
620 
621 static int
622 lagg_clone_destroy(struct if_clone *ifc, struct ifnet *ifp, uint32_t flags)
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 	return (0);
665 }
666 
667 static void
668 lagg_capabilities(struct lagg_softc *sc)
669 {
670 	struct lagg_port *lp;
671 	int cap, cap2, ena, ena2, pena, pena2;
672 	uint64_t hwa;
673 	struct ifnet_hw_tsomax hw_tsomax;
674 
675 	LAGG_XLOCK_ASSERT(sc);
676 
677 	/* Get common enabled capabilities for the lagg ports */
678 	ena = ena2 = ~0;
679 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
680 		ena &= lp->lp_ifp->if_capenable;
681 		ena2 &= lp->lp_ifp->if_capenable2;
682 	}
683 	if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
684 		ena = ena2 = 0;
685 
686 	/*
687 	 * Apply common enabled capabilities back to the lagg ports.
688 	 * May require several iterations if they are dependent.
689 	 */
690 	do {
691 		pena = ena;
692 		pena2 = ena2;
693 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
694 			lagg_setcaps(lp, ena, ena2);
695 			ena &= lp->lp_ifp->if_capenable;
696 			ena2 &= lp->lp_ifp->if_capenable2;
697 		}
698 	} while (pena != ena || pena2 != ena2);
699 
700 	/* Get other capabilities from the lagg ports */
701 	cap = cap2 = ~0;
702 	hwa = ~(uint64_t)0;
703 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
704 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
705 		cap &= lp->lp_ifp->if_capabilities;
706 		cap2 &= lp->lp_ifp->if_capabilities2;
707 		hwa &= lp->lp_ifp->if_hwassist;
708 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
709 	}
710 	if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
711 		cap = cap2 = hwa = 0;
712 
713 	if (sc->sc_ifp->if_capabilities != cap ||
714 	    sc->sc_ifp->if_capenable != ena ||
715 	    sc->sc_ifp->if_capenable2 != ena2 ||
716 	    sc->sc_ifp->if_hwassist != hwa ||
717 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
718 		sc->sc_ifp->if_capabilities = cap;
719 		sc->sc_ifp->if_capabilities2 = cap2;
720 		sc->sc_ifp->if_capenable = ena;
721 		sc->sc_ifp->if_capenable2 = ena2;
722 		sc->sc_ifp->if_hwassist = hwa;
723 		getmicrotime(&sc->sc_ifp->if_lastchange);
724 
725 		if (sc->sc_ifflags & IFF_DEBUG)
726 			if_printf(sc->sc_ifp,
727 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
728 	}
729 }
730 
731 static int
732 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
733 {
734 	struct lagg_softc *sc_ptr;
735 	struct lagg_port *lp, *tlp;
736 	struct ifreq ifr;
737 	int error, i, oldmtu;
738 	int if_type;
739 	uint64_t *pval;
740 
741 	LAGG_XLOCK_ASSERT(sc);
742 
743 	if (sc->sc_ifp == ifp) {
744 		if_printf(sc->sc_ifp,
745 		    "cannot add a lagg to itself as a port\n");
746 		return (EINVAL);
747 	}
748 
749 	if (sc->sc_destroying == 1)
750 		return (ENXIO);
751 
752 	/* Limit the maximal number of lagg ports */
753 	if (sc->sc_count >= LAGG_MAX_PORTS)
754 		return (ENOSPC);
755 
756 	/* Check if port has already been associated to a lagg */
757 	if (ifp->if_lagg != NULL) {
758 		/* Port is already in the current lagg? */
759 		lp = (struct lagg_port *)ifp->if_lagg;
760 		if (lp->lp_softc == sc)
761 			return (EEXIST);
762 		return (EBUSY);
763 	}
764 
765 	switch (sc->sc_ifp->if_type) {
766 	case IFT_ETHER:
767 		/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
768 		if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
769 			return (EPROTONOSUPPORT);
770 		if_type = IFT_IEEE8023ADLAG;
771 		break;
772 	case IFT_INFINIBAND:
773 		/* XXX Disallow non-infiniband interfaces */
774 		if (ifp->if_type != IFT_INFINIBAND)
775 			return (EPROTONOSUPPORT);
776 		if_type = IFT_INFINIBANDLAG;
777 		break;
778 	default:
779 		break;
780 	}
781 
782 	/* Allow the first Ethernet member to define the MTU */
783 	oldmtu = -1;
784 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
785 		sc->sc_ifp->if_mtu = ifp->if_mtu;
786 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
787 		if (ifp->if_ioctl == NULL) {
788 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
789 			    ifp->if_xname);
790 			return (EINVAL);
791 		}
792 		oldmtu = ifp->if_mtu;
793 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
794 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
795 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
796 		if (error != 0) {
797 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
798 			    ifp->if_xname);
799 			return (error);
800 		}
801 		ifr.ifr_mtu = oldmtu;
802 	}
803 
804 	lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK | M_ZERO);
805 	lp->lp_softc = sc;
806 
807 	/* Check if port is a stacked lagg */
808 	LAGG_LIST_LOCK();
809 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
810 		if (ifp == sc_ptr->sc_ifp) {
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 (EINVAL);
817 			/* XXX disable stacking for the moment, its untested */
818 #ifdef LAGG_PORT_STACKING
819 			lp->lp_flags |= LAGG_PORT_STACK;
820 			if (lagg_port_checkstacking(sc_ptr) >=
821 			    LAGG_MAX_STACKING) {
822 				LAGG_LIST_UNLOCK();
823 				free(lp, M_LAGG);
824 				if (oldmtu != -1)
825 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
826 					    (caddr_t)&ifr);
827 				return (E2BIG);
828 			}
829 #endif
830 		}
831 	}
832 	LAGG_LIST_UNLOCK();
833 
834 	if_ref(ifp);
835 	lp->lp_ifp = ifp;
836 
837 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
838 	lp->lp_ifcapenable = ifp->if_capenable;
839 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
840 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
841 		lagg_proto_lladdr(sc);
842 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
843 	} else {
844 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
845 	}
846 	lagg_setflags(lp, 1);
847 
848 	if (CK_SLIST_EMPTY(&sc->sc_ports))
849 		sc->sc_primary = lp;
850 
851 	/* Change the interface type */
852 	lp->lp_iftype = ifp->if_type;
853 	ifp->if_type = if_type;
854 	ifp->if_lagg = lp;
855 	lp->lp_ioctl = ifp->if_ioctl;
856 	ifp->if_ioctl = lagg_port_ioctl;
857 	lp->lp_output = ifp->if_output;
858 	ifp->if_output = lagg_port_output;
859 
860 	/* Read port counters */
861 	pval = lp->port_counters.val;
862 	for (i = 0; i < IFCOUNTERS; i++, pval++)
863 		*pval = ifp->if_get_counter(ifp, i);
864 
865 	/*
866 	 * Insert into the list of ports.
867 	 * Keep ports sorted by if_index. It is handy, when configuration
868 	 * is predictable and `ifconfig laggN create ...` command
869 	 * will lead to the same result each time.
870 	 */
871 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
872 		if (tlp->lp_ifp->if_index < ifp->if_index && (
873 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
874 		    ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
875 		    ifp->if_index))
876 			break;
877 	}
878 	if (tlp != NULL)
879 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
880 	else
881 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
882 	sc->sc_count++;
883 
884 	lagg_setmulti(lp);
885 
886 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
887 		/* Remove the port, without calling pr_delport. */
888 		lagg_port_destroy(lp, 0);
889 		if (oldmtu != -1)
890 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
891 		return (error);
892 	}
893 
894 	/* Update lagg capabilities */
895 	lagg_capabilities(sc);
896 	lagg_linkstate(sc);
897 
898 	return (0);
899 }
900 
901 #ifdef LAGG_PORT_STACKING
902 static int
903 lagg_port_checkstacking(struct lagg_softc *sc)
904 {
905 	struct lagg_softc *sc_ptr;
906 	struct lagg_port *lp;
907 	int m = 0;
908 
909 	LAGG_SXLOCK_ASSERT(sc);
910 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
911 		if (lp->lp_flags & LAGG_PORT_STACK) {
912 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
913 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
914 		}
915 	}
916 
917 	return (m + 1);
918 }
919 #endif
920 
921 static void
922 lagg_port_destroy_cb(epoch_context_t ec)
923 {
924 	struct lagg_port *lp;
925 	struct ifnet *ifp;
926 
927 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
928 	ifp = lp->lp_ifp;
929 
930 	if_rele(ifp);
931 	free(lp, M_LAGG);
932 }
933 
934 static int
935 lagg_port_destroy(struct lagg_port *lp, int rundelport)
936 {
937 	struct lagg_softc *sc = lp->lp_softc;
938 	struct lagg_port *lp_ptr, *lp0;
939 	struct ifnet *ifp = lp->lp_ifp;
940 	uint64_t *pval, vdiff;
941 	int i;
942 
943 	LAGG_XLOCK_ASSERT(sc);
944 
945 	if (rundelport)
946 		lagg_proto_delport(sc, lp);
947 
948 	if (lp->lp_detaching == 0)
949 		lagg_clrmulti(lp);
950 
951 	/* Restore interface */
952 	ifp->if_type = lp->lp_iftype;
953 	ifp->if_ioctl = lp->lp_ioctl;
954 	ifp->if_output = lp->lp_output;
955 	ifp->if_lagg = NULL;
956 
957 	/* Update detached port counters */
958 	pval = lp->port_counters.val;
959 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
960 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
961 		sc->detached_counters.val[i] += vdiff;
962 	}
963 
964 	/* Finally, remove the port from the lagg */
965 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
966 	sc->sc_count--;
967 
968 	/* Update the primary interface */
969 	if (lp == sc->sc_primary) {
970 		uint8_t lladdr[LAGG_ADDR_LEN];
971 
972 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
973 			bzero(&lladdr, LAGG_ADDR_LEN);
974 		else
975 			bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
976 		sc->sc_primary = lp0;
977 		if (sc->sc_destroying == 0) {
978 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
979 			lagg_proto_lladdr(sc);
980 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
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 			 * We can skip this if the lagg is being destroyed.
986 			 */
987 			CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
988 				if_setlladdr(lp_ptr->lp_ifp, lladdr,
989 				    lp_ptr->lp_ifp->if_addrlen);
990 		}
991 	}
992 
993 	if (lp->lp_ifflags)
994 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
995 
996 	if (lp->lp_detaching == 0) {
997 		lagg_setflags(lp, 0);
998 		lagg_setcaps(lp, lp->lp_ifcapenable, lp->lp_ifcapenable2);
999 		if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
1000 	}
1001 
1002 	/*
1003 	 * free port and release it's ifnet reference after a grace period has
1004 	 * elapsed.
1005 	 */
1006 	NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
1007 	/* Update lagg capabilities */
1008 	lagg_capabilities(sc);
1009 	lagg_linkstate(sc);
1010 
1011 	return (0);
1012 }
1013 
1014 static int
1015 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1016 {
1017 	struct epoch_tracker et;
1018 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
1019 	struct lagg_softc *sc;
1020 	struct lagg_port *lp = NULL;
1021 	int error = 0;
1022 
1023 	/* Should be checked by the caller */
1024 	switch (ifp->if_type) {
1025 	case IFT_IEEE8023ADLAG:
1026 	case IFT_INFINIBANDLAG:
1027 		if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1028 			goto fallback;
1029 		break;
1030 	default:
1031 		goto fallback;
1032 	}
1033 
1034 	switch (cmd) {
1035 	case SIOCGLAGGPORT:
1036 		if (rp->rp_portname[0] == '\0' ||
1037 		    ifunit(rp->rp_portname) != ifp) {
1038 			error = EINVAL;
1039 			break;
1040 		}
1041 
1042 		NET_EPOCH_ENTER(et);
1043 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1044 			error = ENOENT;
1045 			NET_EPOCH_EXIT(et);
1046 			break;
1047 		}
1048 
1049 		lagg_port2req(lp, rp);
1050 		NET_EPOCH_EXIT(et);
1051 		break;
1052 
1053 	case SIOCSIFCAP:
1054 	case SIOCSIFCAPNV:
1055 		if (lp->lp_ioctl == NULL) {
1056 			error = EINVAL;
1057 			break;
1058 		}
1059 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1060 		if (error)
1061 			break;
1062 
1063 		/* Update lagg interface capabilities */
1064 		LAGG_XLOCK(sc);
1065 		lagg_capabilities(sc);
1066 		LAGG_XUNLOCK(sc);
1067 		VLAN_CAPABILITIES(sc->sc_ifp);
1068 		break;
1069 
1070 	case SIOCSIFMTU:
1071 		/* Do not allow the MTU to be changed once joined */
1072 		error = EINVAL;
1073 		break;
1074 
1075 	default:
1076 		goto fallback;
1077 	}
1078 
1079 	return (error);
1080 
1081 fallback:
1082 	if (lp != NULL && lp->lp_ioctl != NULL)
1083 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1084 
1085 	return (EINVAL);
1086 }
1087 
1088 /*
1089  * Requests counter @cnt data.
1090  *
1091  * Counter value is calculated the following way:
1092  * 1) for each port, sum difference between current and "initial" measurements.
1093  * 2) add lagg logical interface counters.
1094  * 3) add data from detached_counters array.
1095  *
1096  * We also do the following things on ports attach/detach:
1097  * 1) On port attach we store all counters it has into port_counter array.
1098  * 2) On port detach we add the different between "initial" and
1099  *   current counters data to detached_counters array.
1100  */
1101 static uint64_t
1102 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1103 {
1104 	struct epoch_tracker et;
1105 	struct lagg_softc *sc;
1106 	struct lagg_port *lp;
1107 	struct ifnet *lpifp;
1108 	uint64_t newval, oldval, vsum;
1109 
1110 	/* Revise this when we've got non-generic counters. */
1111 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1112 
1113 	sc = (struct lagg_softc *)ifp->if_softc;
1114 
1115 	vsum = 0;
1116 	NET_EPOCH_ENTER(et);
1117 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1118 		/* Saved attached value */
1119 		oldval = lp->port_counters.val[cnt];
1120 		/* current value */
1121 		lpifp = lp->lp_ifp;
1122 		newval = lpifp->if_get_counter(lpifp, cnt);
1123 		/* Calculate diff and save new */
1124 		vsum += newval - oldval;
1125 	}
1126 	NET_EPOCH_EXIT(et);
1127 
1128 	/*
1129 	 * Add counter data which might be added by upper
1130 	 * layer protocols operating on logical interface.
1131 	 */
1132 	vsum += if_get_counter_default(ifp, cnt);
1133 
1134 	/*
1135 	 * Add counter data from detached ports counters
1136 	 */
1137 	vsum += sc->detached_counters.val[cnt];
1138 
1139 	return (vsum);
1140 }
1141 
1142 /*
1143  * For direct output to child ports.
1144  */
1145 static int
1146 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1147 	const struct sockaddr *dst, struct route *ro)
1148 {
1149 	struct lagg_port *lp = ifp->if_lagg;
1150 
1151 	switch (dst->sa_family) {
1152 		case pseudo_AF_HDRCMPLT:
1153 		case AF_UNSPEC:
1154 			if (lp != NULL)
1155 				return ((*lp->lp_output)(ifp, m, dst, ro));
1156 	}
1157 
1158 	/* drop any other frames */
1159 	m_freem(m);
1160 	return (ENETDOWN);
1161 }
1162 
1163 static void
1164 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1165 {
1166 	struct lagg_port *lp;
1167 	struct lagg_softc *sc;
1168 
1169 	if ((lp = ifp->if_lagg) == NULL)
1170 		return;
1171 	/* If the ifnet is just being renamed, don't do anything. */
1172 	if (ifp->if_flags & IFF_RENAMING)
1173 		return;
1174 
1175 	sc = lp->lp_softc;
1176 
1177 	LAGG_XLOCK(sc);
1178 	lp->lp_detaching = 1;
1179 	lagg_port_destroy(lp, 1);
1180 	LAGG_XUNLOCK(sc);
1181 	VLAN_CAPABILITIES(sc->sc_ifp);
1182 }
1183 
1184 static void
1185 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1186 {
1187 	struct lagg_softc *sc = lp->lp_softc;
1188 
1189 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1190 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1191 	rp->rp_prio = lp->lp_prio;
1192 	rp->rp_flags = lp->lp_flags;
1193 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1194 
1195 	/* Add protocol specific flags */
1196 	switch (sc->sc_proto) {
1197 		case LAGG_PROTO_FAILOVER:
1198 			if (lp == sc->sc_primary)
1199 				rp->rp_flags |= LAGG_PORT_MASTER;
1200 			if (lp == lagg_link_active(sc, sc->sc_primary))
1201 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1202 			break;
1203 
1204 		case LAGG_PROTO_ROUNDROBIN:
1205 		case LAGG_PROTO_LOADBALANCE:
1206 		case LAGG_PROTO_BROADCAST:
1207 			if (LAGG_PORTACTIVE(lp))
1208 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1209 			break;
1210 
1211 		case LAGG_PROTO_LACP:
1212 			/* LACP has a different definition of active */
1213 			if (lacp_isactive(lp))
1214 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1215 			if (lacp_iscollecting(lp))
1216 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1217 			if (lacp_isdistributing(lp))
1218 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1219 			break;
1220 	}
1221 
1222 }
1223 
1224 static void
1225 lagg_watchdog_infiniband(void *arg)
1226 {
1227 	struct epoch_tracker et;
1228 	struct lagg_softc *sc;
1229 	struct lagg_port *lp;
1230 	struct ifnet *ifp;
1231 	struct ifnet *lp_ifp;
1232 
1233 	sc = arg;
1234 
1235 	/*
1236 	 * Because infiniband nodes have a fixed MAC address, which is
1237 	 * generated by the so-called GID, we need to regularly update
1238 	 * the link level address of the parent lagg<N> device when
1239 	 * the active port changes. Possibly we could piggy-back on
1240 	 * link up/down events aswell, but using a timer also provides
1241 	 * a guarantee against too frequent events. This operation
1242 	 * does not have to be atomic.
1243 	 */
1244 	NET_EPOCH_ENTER(et);
1245 	lp = lagg_link_active(sc, sc->sc_primary);
1246 	if (lp != NULL) {
1247 		ifp = sc->sc_ifp;
1248 		lp_ifp = lp->lp_ifp;
1249 
1250 		if (ifp != NULL && lp_ifp != NULL &&
1251 		    (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0 ||
1252 		     memcmp(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen) != 0)) {
1253 			memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1254 			memcpy(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen);
1255 
1256 			CURVNET_SET(ifp->if_vnet);
1257 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1258 			CURVNET_RESTORE();
1259 		}
1260 	}
1261 	NET_EPOCH_EXIT(et);
1262 
1263 	callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1264 }
1265 
1266 static void
1267 lagg_init(void *xsc)
1268 {
1269 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1270 	struct ifnet *ifp = sc->sc_ifp;
1271 	struct lagg_port *lp;
1272 
1273 	LAGG_XLOCK(sc);
1274 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1275 		LAGG_XUNLOCK(sc);
1276 		return;
1277 	}
1278 
1279 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1280 
1281 	/*
1282 	 * Update the port lladdrs if needed.
1283 	 * This might be if_setlladdr() notification
1284 	 * that lladdr has been changed.
1285 	 */
1286 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1287 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1288 		    ifp->if_addrlen) != 0)
1289 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1290 	}
1291 
1292 	lagg_proto_init(sc);
1293 
1294 	if (ifp->if_type == IFT_INFINIBAND) {
1295 		mtx_lock(&sc->sc_mtx);
1296 		lagg_watchdog_infiniband(sc);
1297 		mtx_unlock(&sc->sc_mtx);
1298 	}
1299 
1300 	LAGG_XUNLOCK(sc);
1301 }
1302 
1303 static void
1304 lagg_stop(struct lagg_softc *sc)
1305 {
1306 	struct ifnet *ifp = sc->sc_ifp;
1307 
1308 	LAGG_XLOCK_ASSERT(sc);
1309 
1310 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1311 		return;
1312 
1313 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1314 
1315 	lagg_proto_stop(sc);
1316 
1317 	mtx_lock(&sc->sc_mtx);
1318 	callout_stop(&sc->sc_watchdog);
1319 	mtx_unlock(&sc->sc_mtx);
1320 
1321 	callout_drain(&sc->sc_watchdog);
1322 }
1323 
1324 static int
1325 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1326 {
1327 	struct epoch_tracker et;
1328 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1329 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1330 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1331 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1332 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1333 	struct ifreq *ifr = (struct ifreq *)data;
1334 	struct lagg_port *lp;
1335 	struct ifnet *tpif;
1336 	struct thread *td = curthread;
1337 	char *buf, *outbuf;
1338 	int count, buflen, len, error = 0, oldmtu;
1339 
1340 	bzero(&rpbuf, sizeof(rpbuf));
1341 
1342 	/* XXX: This can race with lagg_clone_destroy. */
1343 
1344 	switch (cmd) {
1345 	case SIOCGLAGG:
1346 		LAGG_XLOCK(sc);
1347 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1348 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1349 		ra->ra_proto = sc->sc_proto;
1350 		lagg_proto_request(sc, &ra->ra_psc);
1351 		count = 0;
1352 		buf = outbuf;
1353 		len = min(ra->ra_size, buflen);
1354 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1355 			if (len < sizeof(rpbuf))
1356 				break;
1357 
1358 			lagg_port2req(lp, &rpbuf);
1359 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1360 			count++;
1361 			buf += sizeof(rpbuf);
1362 			len -= sizeof(rpbuf);
1363 		}
1364 		LAGG_XUNLOCK(sc);
1365 		ra->ra_ports = count;
1366 		ra->ra_size = count * sizeof(rpbuf);
1367 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1368 		free(outbuf, M_TEMP);
1369 		break;
1370 	case SIOCSLAGG:
1371 		error = priv_check(td, PRIV_NET_LAGG);
1372 		if (error)
1373 			break;
1374 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1375 			error = EPROTONOSUPPORT;
1376 			break;
1377 		}
1378 		/* Infiniband only supports the failover protocol. */
1379 		if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1380 		    ifp->if_type == IFT_INFINIBAND) {
1381 			error = EPROTONOSUPPORT;
1382 			break;
1383 		}
1384 		LAGG_XLOCK(sc);
1385 		lagg_proto_detach(sc);
1386 		lagg_proto_attach(sc, ra->ra_proto);
1387 		LAGG_XUNLOCK(sc);
1388 		break;
1389 	case SIOCGLAGGOPTS:
1390 		LAGG_XLOCK(sc);
1391 		ro->ro_opts = sc->sc_opts;
1392 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1393 			struct lacp_softc *lsc;
1394 
1395 			lsc = (struct lacp_softc *)sc->sc_psc;
1396 			if (lsc->lsc_debug.lsc_tx_test != 0)
1397 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1398 			if (lsc->lsc_debug.lsc_rx_test != 0)
1399 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1400 			if (lsc->lsc_strict_mode != 0)
1401 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1402 			if (lsc->lsc_fast_timeout != 0)
1403 				ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1404 
1405 			ro->ro_active = sc->sc_active;
1406 		} else {
1407 			ro->ro_active = 0;
1408 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1409 				ro->ro_active += LAGG_PORTACTIVE(lp);
1410 		}
1411 		ro->ro_bkt = sc->sc_stride;
1412 		ro->ro_flapping = sc->sc_flapping;
1413 		ro->ro_flowid_shift = sc->flowid_shift;
1414 		LAGG_XUNLOCK(sc);
1415 		break;
1416 	case SIOCSLAGGOPTS:
1417 		error = priv_check(td, PRIV_NET_LAGG);
1418 		if (error)
1419 			break;
1420 
1421 		/*
1422 		 * The stride option was added without defining a corresponding
1423 		 * LAGG_OPT flag, so handle a non-zero value before checking
1424 		 * anything else to preserve compatibility.
1425 		 */
1426 		LAGG_XLOCK(sc);
1427 		if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1428 			if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1429 				LAGG_XUNLOCK(sc);
1430 				error = EINVAL;
1431 				break;
1432 			}
1433 			sc->sc_stride = ro->ro_bkt;
1434 		}
1435 		if (ro->ro_opts == 0) {
1436 			LAGG_XUNLOCK(sc);
1437 			break;
1438 		}
1439 
1440 		/*
1441 		 * Set options.  LACP options are stored in sc->sc_psc,
1442 		 * not in sc_opts.
1443 		 */
1444 		int valid, lacp;
1445 
1446 		switch (ro->ro_opts) {
1447 		case LAGG_OPT_USE_FLOWID:
1448 		case -LAGG_OPT_USE_FLOWID:
1449 		case LAGG_OPT_USE_NUMA:
1450 		case -LAGG_OPT_USE_NUMA:
1451 		case LAGG_OPT_FLOWIDSHIFT:
1452 		case LAGG_OPT_RR_LIMIT:
1453 			valid = 1;
1454 			lacp = 0;
1455 			break;
1456 		case LAGG_OPT_LACP_TXTEST:
1457 		case -LAGG_OPT_LACP_TXTEST:
1458 		case LAGG_OPT_LACP_RXTEST:
1459 		case -LAGG_OPT_LACP_RXTEST:
1460 		case LAGG_OPT_LACP_STRICT:
1461 		case -LAGG_OPT_LACP_STRICT:
1462 		case LAGG_OPT_LACP_FAST_TIMO:
1463 		case -LAGG_OPT_LACP_FAST_TIMO:
1464 			valid = lacp = 1;
1465 			break;
1466 		default:
1467 			valid = lacp = 0;
1468 			break;
1469 		}
1470 
1471 		if (valid == 0 ||
1472 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1473 			/* Invalid combination of options specified. */
1474 			error = EINVAL;
1475 			LAGG_XUNLOCK(sc);
1476 			break;	/* Return from SIOCSLAGGOPTS. */
1477 		}
1478 
1479 		/*
1480 		 * Store new options into sc->sc_opts except for
1481 		 * FLOWIDSHIFT, RR and LACP options.
1482 		 */
1483 		if (lacp == 0) {
1484 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1485 				sc->flowid_shift = ro->ro_flowid_shift;
1486 			else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1487 				if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1488 				    ro->ro_bkt == 0) {
1489 					error = EINVAL;
1490 					LAGG_XUNLOCK(sc);
1491 					break;
1492 				}
1493 				sc->sc_stride = ro->ro_bkt;
1494 			} else if (ro->ro_opts > 0)
1495 				sc->sc_opts |= ro->ro_opts;
1496 			else
1497 				sc->sc_opts &= ~ro->ro_opts;
1498 		} else {
1499 			struct lacp_softc *lsc;
1500 			struct lacp_port *lp;
1501 
1502 			lsc = (struct lacp_softc *)sc->sc_psc;
1503 
1504 			switch (ro->ro_opts) {
1505 			case LAGG_OPT_LACP_TXTEST:
1506 				lsc->lsc_debug.lsc_tx_test = 1;
1507 				break;
1508 			case -LAGG_OPT_LACP_TXTEST:
1509 				lsc->lsc_debug.lsc_tx_test = 0;
1510 				break;
1511 			case LAGG_OPT_LACP_RXTEST:
1512 				lsc->lsc_debug.lsc_rx_test = 1;
1513 				break;
1514 			case -LAGG_OPT_LACP_RXTEST:
1515 				lsc->lsc_debug.lsc_rx_test = 0;
1516 				break;
1517 			case LAGG_OPT_LACP_STRICT:
1518 				lsc->lsc_strict_mode = 1;
1519 				break;
1520 			case -LAGG_OPT_LACP_STRICT:
1521 				lsc->lsc_strict_mode = 0;
1522 				break;
1523 			case LAGG_OPT_LACP_FAST_TIMO:
1524 				LACP_LOCK(lsc);
1525 				LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1526 					lp->lp_state |= LACP_STATE_TIMEOUT;
1527 				LACP_UNLOCK(lsc);
1528 				lsc->lsc_fast_timeout = 1;
1529 				break;
1530 			case -LAGG_OPT_LACP_FAST_TIMO:
1531 				LACP_LOCK(lsc);
1532 				LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1533 					lp->lp_state &= ~LACP_STATE_TIMEOUT;
1534 				LACP_UNLOCK(lsc);
1535 				lsc->lsc_fast_timeout = 0;
1536 				break;
1537 			}
1538 		}
1539 		LAGG_XUNLOCK(sc);
1540 		break;
1541 	case SIOCGLAGGFLAGS:
1542 		rf->rf_flags = 0;
1543 		LAGG_XLOCK(sc);
1544 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1545 			rf->rf_flags |= LAGG_F_HASHL2;
1546 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1547 			rf->rf_flags |= LAGG_F_HASHL3;
1548 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1549 			rf->rf_flags |= LAGG_F_HASHL4;
1550 		LAGG_XUNLOCK(sc);
1551 		break;
1552 	case SIOCSLAGGHASH:
1553 		error = priv_check(td, PRIV_NET_LAGG);
1554 		if (error)
1555 			break;
1556 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1557 			error = EINVAL;
1558 			break;
1559 		}
1560 		LAGG_XLOCK(sc);
1561 		sc->sc_flags = 0;
1562 		if (rf->rf_flags & LAGG_F_HASHL2)
1563 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1564 		if (rf->rf_flags & LAGG_F_HASHL3)
1565 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1566 		if (rf->rf_flags & LAGG_F_HASHL4)
1567 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1568 		LAGG_XUNLOCK(sc);
1569 		break;
1570 	case SIOCGLAGGPORT:
1571 		if (rp->rp_portname[0] == '\0' ||
1572 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1573 			error = EINVAL;
1574 			break;
1575 		}
1576 
1577 		NET_EPOCH_ENTER(et);
1578 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1579 		    lp->lp_softc != sc) {
1580 			error = ENOENT;
1581 			NET_EPOCH_EXIT(et);
1582 			if_rele(tpif);
1583 			break;
1584 		}
1585 
1586 		lagg_port2req(lp, rp);
1587 		NET_EPOCH_EXIT(et);
1588 		if_rele(tpif);
1589 		break;
1590 	case SIOCSLAGGPORT:
1591 		error = priv_check(td, PRIV_NET_LAGG);
1592 		if (error)
1593 			break;
1594 		if (rp->rp_portname[0] == '\0' ||
1595 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1596 			error = EINVAL;
1597 			break;
1598 		}
1599 #ifdef INET6
1600 		/*
1601 		 * A laggport interface should not have inet6 address
1602 		 * because two interfaces with a valid link-local
1603 		 * scope zone must not be merged in any form.  This
1604 		 * restriction is needed to prevent violation of
1605 		 * link-local scope zone.  Attempts to add a laggport
1606 		 * interface which has inet6 addresses triggers
1607 		 * removal of all inet6 addresses on the member
1608 		 * interface.
1609 		 */
1610 		if (in6ifa_llaonifp(tpif)) {
1611 			in6_ifdetach(tpif);
1612 				if_printf(sc->sc_ifp,
1613 				    "IPv6 addresses on %s have been removed "
1614 				    "before adding it as a member to prevent "
1615 				    "IPv6 address scope violation.\n",
1616 				    tpif->if_xname);
1617 		}
1618 #endif
1619 		oldmtu = ifp->if_mtu;
1620 		LAGG_XLOCK(sc);
1621 		error = lagg_port_create(sc, tpif);
1622 		LAGG_XUNLOCK(sc);
1623 		if_rele(tpif);
1624 
1625 		/*
1626 		 * LAGG MTU may change during addition of the first port.
1627 		 * If it did, do network layer specific procedure.
1628 		 */
1629 		if (ifp->if_mtu != oldmtu)
1630 			if_notifymtu(ifp);
1631 
1632 		VLAN_CAPABILITIES(ifp);
1633 		break;
1634 	case SIOCSLAGGDELPORT:
1635 		error = priv_check(td, PRIV_NET_LAGG);
1636 		if (error)
1637 			break;
1638 		if (rp->rp_portname[0] == '\0' ||
1639 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1640 			error = EINVAL;
1641 			break;
1642 		}
1643 
1644 		LAGG_XLOCK(sc);
1645 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1646 		    lp->lp_softc != sc) {
1647 			error = ENOENT;
1648 			LAGG_XUNLOCK(sc);
1649 			if_rele(tpif);
1650 			break;
1651 		}
1652 
1653 		error = lagg_port_destroy(lp, 1);
1654 		LAGG_XUNLOCK(sc);
1655 		if_rele(tpif);
1656 		VLAN_CAPABILITIES(ifp);
1657 		break;
1658 	case SIOCSIFFLAGS:
1659 		/* Set flags on ports too */
1660 		LAGG_XLOCK(sc);
1661 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1662 			lagg_setflags(lp, 1);
1663 		}
1664 
1665 		if (!(ifp->if_flags & IFF_UP) &&
1666 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1667 			/*
1668 			 * If interface is marked down and it is running,
1669 			 * then stop and disable it.
1670 			 */
1671 			lagg_stop(sc);
1672 			LAGG_XUNLOCK(sc);
1673 		} else if ((ifp->if_flags & IFF_UP) &&
1674 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1675 			/*
1676 			 * If interface is marked up and it is stopped, then
1677 			 * start it.
1678 			 */
1679 			LAGG_XUNLOCK(sc);
1680 			(*ifp->if_init)(sc);
1681 		} else
1682 			LAGG_XUNLOCK(sc);
1683 		break;
1684 	case SIOCADDMULTI:
1685 	case SIOCDELMULTI:
1686 		LAGG_XLOCK(sc);
1687 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1688 			lagg_clrmulti(lp);
1689 			lagg_setmulti(lp);
1690 		}
1691 		LAGG_XUNLOCK(sc);
1692 		error = 0;
1693 		break;
1694 	case SIOCSIFMEDIA:
1695 	case SIOCGIFMEDIA:
1696 		if (ifp->if_type == IFT_INFINIBAND)
1697 			error = EINVAL;
1698 		else
1699 			error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1700 		break;
1701 
1702 	case SIOCSIFCAP:
1703 	case SIOCSIFCAPNV:
1704 		LAGG_XLOCK(sc);
1705 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1706 			if (lp->lp_ioctl != NULL)
1707 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1708 		}
1709 		lagg_capabilities(sc);
1710 		LAGG_XUNLOCK(sc);
1711 		VLAN_CAPABILITIES(ifp);
1712 		error = 0;
1713 		break;
1714 
1715 	case SIOCGIFCAPNV:
1716 		error = 0;
1717 		break;
1718 
1719 	case SIOCSIFMTU:
1720 		LAGG_XLOCK(sc);
1721 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1722 			if (lp->lp_ioctl != NULL)
1723 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1724 			else
1725 				error = EINVAL;
1726 			if (error != 0) {
1727 				if_printf(ifp,
1728 				    "failed to change MTU to %d on port %s, "
1729 				    "reverting all ports to original MTU (%d)\n",
1730 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1731 				break;
1732 			}
1733 		}
1734 		if (error == 0) {
1735 			ifp->if_mtu = ifr->ifr_mtu;
1736 		} else {
1737 			/* set every port back to the original MTU */
1738 			ifr->ifr_mtu = ifp->if_mtu;
1739 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1740 				if (lp->lp_ioctl != NULL)
1741 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1742 			}
1743 		}
1744 		lagg_capabilities(sc);
1745 		LAGG_XUNLOCK(sc);
1746 		VLAN_CAPABILITIES(ifp);
1747 		break;
1748 
1749 	default:
1750 		error = ether_ioctl(ifp, cmd, data);
1751 		break;
1752 	}
1753 	return (error);
1754 }
1755 
1756 #if defined(KERN_TLS) || defined(RATELIMIT)
1757 #ifdef RATELIMIT
1758 static const struct if_snd_tag_sw lagg_snd_tag_ul_sw = {
1759 	.snd_tag_modify = lagg_snd_tag_modify,
1760 	.snd_tag_query = lagg_snd_tag_query,
1761 	.snd_tag_free = lagg_snd_tag_free,
1762 	.next_snd_tag = lagg_next_snd_tag,
1763 	.type = IF_SND_TAG_TYPE_UNLIMITED
1764 };
1765 
1766 static const struct if_snd_tag_sw lagg_snd_tag_rl_sw = {
1767 	.snd_tag_modify = lagg_snd_tag_modify,
1768 	.snd_tag_query = lagg_snd_tag_query,
1769 	.snd_tag_free = lagg_snd_tag_free,
1770 	.next_snd_tag = lagg_next_snd_tag,
1771 	.type = IF_SND_TAG_TYPE_RATE_LIMIT
1772 };
1773 #endif
1774 
1775 #ifdef KERN_TLS
1776 static const struct if_snd_tag_sw lagg_snd_tag_tls_sw = {
1777 	.snd_tag_modify = lagg_snd_tag_modify,
1778 	.snd_tag_query = lagg_snd_tag_query,
1779 	.snd_tag_free = lagg_snd_tag_free,
1780 	.next_snd_tag = lagg_next_snd_tag,
1781 	.type = IF_SND_TAG_TYPE_TLS
1782 };
1783 
1784 #ifdef RATELIMIT
1785 static const struct if_snd_tag_sw lagg_snd_tag_tls_rl_sw = {
1786 	.snd_tag_modify = lagg_snd_tag_modify,
1787 	.snd_tag_query = lagg_snd_tag_query,
1788 	.snd_tag_free = lagg_snd_tag_free,
1789 	.next_snd_tag = lagg_next_snd_tag,
1790 	.type = IF_SND_TAG_TYPE_TLS_RATE_LIMIT
1791 };
1792 #endif
1793 #endif
1794 
1795 static inline struct lagg_snd_tag *
1796 mst_to_lst(struct m_snd_tag *mst)
1797 {
1798 
1799 	return (__containerof(mst, struct lagg_snd_tag, com));
1800 }
1801 
1802 /*
1803  * Look up the port used by a specific flow.  This only works for lagg
1804  * protocols with deterministic port mappings (e.g. not roundrobin).
1805  * In addition protocols which use a hash to map flows to ports must
1806  * be configured to use the mbuf flowid rather than hashing packet
1807  * contents.
1808  */
1809 static struct lagg_port *
1810 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1811     uint8_t numa_domain)
1812 {
1813 	struct lagg_softc *sc;
1814 	struct lagg_port *lp;
1815 	struct lagg_lb *lb;
1816 	uint32_t hash, p;
1817 	int err;
1818 
1819 	sc = ifp->if_softc;
1820 
1821 	switch (sc->sc_proto) {
1822 	case LAGG_PROTO_FAILOVER:
1823 		return (lagg_link_active(sc, sc->sc_primary));
1824 	case LAGG_PROTO_LOADBALANCE:
1825 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1826 		    flowtype == M_HASHTYPE_NONE)
1827 			return (NULL);
1828 		p = flowid >> sc->flowid_shift;
1829 		p %= sc->sc_count;
1830 		lb = (struct lagg_lb *)sc->sc_psc;
1831 		lp = lb->lb_ports[p];
1832 		return (lagg_link_active(sc, lp));
1833 	case LAGG_PROTO_LACP:
1834 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1835 		    flowtype == M_HASHTYPE_NONE)
1836 			return (NULL);
1837 		hash = flowid >> sc->flowid_shift;
1838 		return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, &err));
1839 	default:
1840 		return (NULL);
1841 	}
1842 }
1843 
1844 static int
1845 lagg_snd_tag_alloc(struct ifnet *ifp,
1846     union if_snd_tag_alloc_params *params,
1847     struct m_snd_tag **ppmt)
1848 {
1849 	struct epoch_tracker et;
1850 	const struct if_snd_tag_sw *sw;
1851 	struct lagg_snd_tag *lst;
1852 	struct lagg_port *lp;
1853 	struct ifnet *lp_ifp;
1854 	struct m_snd_tag *mst;
1855 	int error;
1856 
1857 	switch (params->hdr.type) {
1858 #ifdef RATELIMIT
1859 	case IF_SND_TAG_TYPE_UNLIMITED:
1860 		sw = &lagg_snd_tag_ul_sw;
1861 		break;
1862 	case IF_SND_TAG_TYPE_RATE_LIMIT:
1863 		sw = &lagg_snd_tag_rl_sw;
1864 		break;
1865 #endif
1866 #ifdef KERN_TLS
1867 	case IF_SND_TAG_TYPE_TLS:
1868 		sw = &lagg_snd_tag_tls_sw;
1869 		break;
1870 	case IF_SND_TAG_TYPE_TLS_RX:
1871 		/* Return tag from port interface directly. */
1872 		sw = NULL;
1873 		break;
1874 #ifdef RATELIMIT
1875 	case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
1876 		sw = &lagg_snd_tag_tls_rl_sw;
1877 		break;
1878 #endif
1879 #endif
1880 	default:
1881 		return (EOPNOTSUPP);
1882 	}
1883 
1884 	NET_EPOCH_ENTER(et);
1885 	lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1886 	    params->hdr.flowtype, params->hdr.numa_domain);
1887 	if (lp == NULL) {
1888 		NET_EPOCH_EXIT(et);
1889 		return (EOPNOTSUPP);
1890 	}
1891 	if (lp->lp_ifp == NULL) {
1892 		NET_EPOCH_EXIT(et);
1893 		return (EOPNOTSUPP);
1894 	}
1895 	lp_ifp = lp->lp_ifp;
1896 	if_ref(lp_ifp);
1897 	NET_EPOCH_EXIT(et);
1898 
1899 	if (sw != NULL) {
1900 		lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1901 		if (lst == NULL) {
1902 			if_rele(lp_ifp);
1903 			return (ENOMEM);
1904 		}
1905 	} else
1906 		lst = NULL;
1907 
1908 	error = m_snd_tag_alloc(lp_ifp, params, &mst);
1909 	if_rele(lp_ifp);
1910 	if (error) {
1911 		free(lst, M_LAGG);
1912 		return (error);
1913 	}
1914 
1915 	if (sw != NULL) {
1916 		m_snd_tag_init(&lst->com, ifp, sw);
1917 		lst->tag = mst;
1918 
1919 		*ppmt = &lst->com;
1920 	} else
1921 		*ppmt = mst;
1922 
1923 	return (0);
1924 }
1925 
1926 static struct m_snd_tag *
1927 lagg_next_snd_tag(struct m_snd_tag *mst)
1928 {
1929 	struct lagg_snd_tag *lst;
1930 
1931 	lst = mst_to_lst(mst);
1932 	return (lst->tag);
1933 }
1934 
1935 static int
1936 lagg_snd_tag_modify(struct m_snd_tag *mst,
1937     union if_snd_tag_modify_params *params)
1938 {
1939 	struct lagg_snd_tag *lst;
1940 
1941 	lst = mst_to_lst(mst);
1942 	return (lst->tag->sw->snd_tag_modify(lst->tag, params));
1943 }
1944 
1945 static int
1946 lagg_snd_tag_query(struct m_snd_tag *mst,
1947     union if_snd_tag_query_params *params)
1948 {
1949 	struct lagg_snd_tag *lst;
1950 
1951 	lst = mst_to_lst(mst);
1952 	return (lst->tag->sw->snd_tag_query(lst->tag, params));
1953 }
1954 
1955 static void
1956 lagg_snd_tag_free(struct m_snd_tag *mst)
1957 {
1958 	struct lagg_snd_tag *lst;
1959 
1960 	lst = mst_to_lst(mst);
1961 	m_snd_tag_rele(lst->tag);
1962 	free(lst, M_LAGG);
1963 }
1964 
1965 static void
1966 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1967 {
1968 	/*
1969 	 * For lagg, we have an indirect
1970 	 * interface. The caller needs to
1971 	 * get a ratelimit tag on the actual
1972 	 * interface the flow will go on.
1973 	 */
1974 	q->rate_table = NULL;
1975 	q->flags = RT_IS_INDIRECT;
1976 	q->max_flows = 0;
1977 	q->number_of_rates = 0;
1978 }
1979 #endif
1980 
1981 static int
1982 lagg_setmulti(struct lagg_port *lp)
1983 {
1984 	struct lagg_softc *sc = lp->lp_softc;
1985 	struct ifnet *ifp = lp->lp_ifp;
1986 	struct ifnet *scifp = sc->sc_ifp;
1987 	struct lagg_mc *mc;
1988 	struct ifmultiaddr *ifma;
1989 	int error;
1990 
1991 	IF_ADDR_WLOCK(scifp);
1992 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1993 		if (ifma->ifma_addr->sa_family != AF_LINK)
1994 			continue;
1995 		mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
1996 		if (mc == NULL) {
1997 			IF_ADDR_WUNLOCK(scifp);
1998 			return (ENOMEM);
1999 		}
2000 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
2001 		mc->mc_addr.sdl_index = ifp->if_index;
2002 		mc->mc_ifma = NULL;
2003 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
2004 	}
2005 	IF_ADDR_WUNLOCK(scifp);
2006 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
2007 		error = if_addmulti(ifp,
2008 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
2009 		if (error)
2010 			return (error);
2011 	}
2012 	return (0);
2013 }
2014 
2015 static int
2016 lagg_clrmulti(struct lagg_port *lp)
2017 {
2018 	struct lagg_mc *mc;
2019 
2020 	LAGG_XLOCK_ASSERT(lp->lp_softc);
2021 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
2022 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
2023 		if (mc->mc_ifma && lp->lp_detaching == 0)
2024 			if_delmulti_ifma(mc->mc_ifma);
2025 		free(mc, M_LAGG);
2026 	}
2027 	return (0);
2028 }
2029 
2030 static void
2031 lagg_setcaps(struct lagg_port *lp, int cap, int cap2)
2032 {
2033 	struct ifreq ifr;
2034 	struct siocsifcapnv_driver_data drv_ioctl_data;
2035 
2036 	if (lp->lp_ifp->if_capenable == cap &&
2037 	    lp->lp_ifp->if_capenable2 == cap2)
2038 		return;
2039 	if (lp->lp_ioctl == NULL)
2040 		return;
2041 	/* XXX */
2042 	if ((lp->lp_ifp->if_capabilities & IFCAP_NV) != 0) {
2043 		drv_ioctl_data.reqcap = cap;
2044 		drv_ioctl_data.reqcap2 = cap2;
2045 		drv_ioctl_data.nvcap = NULL;
2046 		(*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAPNV,
2047 		    (caddr_t)&drv_ioctl_data);
2048 	} else {
2049 		ifr.ifr_reqcap = cap;
2050 		(*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr);
2051 	}
2052 }
2053 
2054 /* Handle a ref counted flag that should be set on the lagg port as well */
2055 static int
2056 lagg_setflag(struct lagg_port *lp, int flag, int status,
2057     int (*func)(struct ifnet *, int))
2058 {
2059 	struct lagg_softc *sc = lp->lp_softc;
2060 	struct ifnet *scifp = sc->sc_ifp;
2061 	struct ifnet *ifp = lp->lp_ifp;
2062 	int error;
2063 
2064 	LAGG_XLOCK_ASSERT(sc);
2065 
2066 	status = status ? (scifp->if_flags & flag) : 0;
2067 	/* Now "status" contains the flag value or 0 */
2068 
2069 	/*
2070 	 * See if recorded ports status is different from what
2071 	 * we want it to be.  If it is, flip it.  We record ports
2072 	 * status in lp_ifflags so that we won't clear ports flag
2073 	 * we haven't set.  In fact, we don't clear or set ports
2074 	 * flags directly, but get or release references to them.
2075 	 * That's why we can be sure that recorded flags still are
2076 	 * in accord with actual ports flags.
2077 	 */
2078 	if (status != (lp->lp_ifflags & flag)) {
2079 		error = (*func)(ifp, status);
2080 		if (error)
2081 			return (error);
2082 		lp->lp_ifflags &= ~flag;
2083 		lp->lp_ifflags |= status;
2084 	}
2085 	return (0);
2086 }
2087 
2088 /*
2089  * Handle IFF_* flags that require certain changes on the lagg port
2090  * if "status" is true, update ports flags respective to the lagg
2091  * if "status" is false, forcedly clear the flags set on port.
2092  */
2093 static int
2094 lagg_setflags(struct lagg_port *lp, int status)
2095 {
2096 	int error, i;
2097 
2098 	for (i = 0; lagg_pflags[i].flag; i++) {
2099 		error = lagg_setflag(lp, lagg_pflags[i].flag,
2100 		    status, lagg_pflags[i].func);
2101 		if (error)
2102 			return (error);
2103 	}
2104 	return (0);
2105 }
2106 
2107 static int
2108 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2109 {
2110 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2111 
2112 	NET_EPOCH_ASSERT();
2113 #if defined(KERN_TLS) || defined(RATELIMIT)
2114 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2115 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2116 #endif
2117 	/* We need a Tx algorithm and at least one port */
2118 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2119 		m_freem(m);
2120 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2121 		return (ENXIO);
2122 	}
2123 
2124 	ETHER_BPF_MTAP(ifp, m);
2125 
2126 	return (lagg_proto_start(sc, m));
2127 }
2128 
2129 static int
2130 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2131 {
2132 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2133 
2134 	NET_EPOCH_ASSERT();
2135 #if defined(KERN_TLS) || defined(RATELIMIT)
2136 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2137 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2138 #endif
2139 	/* We need a Tx algorithm and at least one port */
2140 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2141 		m_freem(m);
2142 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2143 		return (ENXIO);
2144 	}
2145 
2146 	infiniband_bpf_mtap(ifp, m);
2147 
2148 	return (lagg_proto_start(sc, m));
2149 }
2150 
2151 /*
2152  * The ifp->if_qflush entry point for lagg(4) is no-op.
2153  */
2154 static void
2155 lagg_qflush(struct ifnet *ifp __unused)
2156 {
2157 }
2158 
2159 static struct mbuf *
2160 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2161 {
2162 	struct lagg_port *lp = ifp->if_lagg;
2163 	struct lagg_softc *sc = lp->lp_softc;
2164 	struct ifnet *scifp = sc->sc_ifp;
2165 
2166 	NET_EPOCH_ASSERT();
2167 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2168 	    lp->lp_detaching != 0 ||
2169 	    sc->sc_proto == LAGG_PROTO_NONE) {
2170 		m_freem(m);
2171 		return (NULL);
2172 	}
2173 
2174 	m = lagg_proto_input(sc, lp, m);
2175 	if (m != NULL) {
2176 		ETHER_BPF_MTAP(scifp, m);
2177 
2178 		if ((scifp->if_flags & IFF_MONITOR) != 0) {
2179 			m_freem(m);
2180 			m = NULL;
2181 		}
2182 	}
2183 
2184 #ifdef DEV_NETMAP
2185 	if (m != NULL && scifp->if_capenable & IFCAP_NETMAP) {
2186 		scifp->if_input(scifp, m);
2187 		m = NULL;
2188 	}
2189 #endif	/* DEV_NETMAP */
2190 
2191 	return (m);
2192 }
2193 
2194 static struct mbuf *
2195 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2196 {
2197 	struct lagg_port *lp = ifp->if_lagg;
2198 	struct lagg_softc *sc = lp->lp_softc;
2199 	struct ifnet *scifp = sc->sc_ifp;
2200 
2201 	NET_EPOCH_ASSERT();
2202 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2203 	    lp->lp_detaching != 0 ||
2204 	    sc->sc_proto == LAGG_PROTO_NONE) {
2205 		m_freem(m);
2206 		return (NULL);
2207 	}
2208 
2209 	m = lagg_proto_input(sc, lp, m);
2210 	if (m != NULL) {
2211 		infiniband_bpf_mtap(scifp, m);
2212 
2213 		if ((scifp->if_flags & IFF_MONITOR) != 0) {
2214 			m_freem(m);
2215 			m = NULL;
2216 		}
2217 	}
2218 
2219 	return (m);
2220 }
2221 
2222 static int
2223 lagg_media_change(struct ifnet *ifp)
2224 {
2225 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2226 
2227 	if (sc->sc_ifflags & IFF_DEBUG)
2228 		printf("%s\n", __func__);
2229 
2230 	/* Ignore */
2231 	return (0);
2232 }
2233 
2234 static void
2235 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2236 {
2237 	struct epoch_tracker et;
2238 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2239 	struct lagg_port *lp;
2240 
2241 	imr->ifm_status = IFM_AVALID;
2242 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
2243 
2244 	NET_EPOCH_ENTER(et);
2245 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2246 		if (LAGG_PORTACTIVE(lp))
2247 			imr->ifm_status |= IFM_ACTIVE;
2248 	}
2249 	NET_EPOCH_EXIT(et);
2250 }
2251 
2252 static void
2253 lagg_linkstate(struct lagg_softc *sc)
2254 {
2255 	struct epoch_tracker et;
2256 	struct lagg_port *lp;
2257 	int new_link = LINK_STATE_DOWN;
2258 	uint64_t speed;
2259 
2260 	LAGG_XLOCK_ASSERT(sc);
2261 
2262 	/* LACP handles link state itself */
2263 	if (sc->sc_proto == LAGG_PROTO_LACP)
2264 		return;
2265 
2266 	/* Our link is considered up if at least one of our ports is active */
2267 	NET_EPOCH_ENTER(et);
2268 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2269 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2270 			new_link = LINK_STATE_UP;
2271 			break;
2272 		}
2273 	}
2274 	NET_EPOCH_EXIT(et);
2275 	if_link_state_change(sc->sc_ifp, new_link);
2276 
2277 	/* Update if_baudrate to reflect the max possible speed */
2278 	switch (sc->sc_proto) {
2279 		case LAGG_PROTO_FAILOVER:
2280 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2281 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
2282 			break;
2283 		case LAGG_PROTO_ROUNDROBIN:
2284 		case LAGG_PROTO_LOADBALANCE:
2285 		case LAGG_PROTO_BROADCAST:
2286 			speed = 0;
2287 			NET_EPOCH_ENTER(et);
2288 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2289 				speed += lp->lp_ifp->if_baudrate;
2290 			NET_EPOCH_EXIT(et);
2291 			sc->sc_ifp->if_baudrate = speed;
2292 			break;
2293 		case LAGG_PROTO_LACP:
2294 			/* LACP updates if_baudrate itself */
2295 			break;
2296 	}
2297 }
2298 
2299 static void
2300 lagg_port_state(struct ifnet *ifp, int state)
2301 {
2302 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2303 	struct lagg_softc *sc = NULL;
2304 
2305 	if (lp != NULL)
2306 		sc = lp->lp_softc;
2307 	if (sc == NULL)
2308 		return;
2309 
2310 	LAGG_XLOCK(sc);
2311 	lagg_linkstate(sc);
2312 	lagg_proto_linkstate(sc, lp);
2313 	LAGG_XUNLOCK(sc);
2314 }
2315 
2316 struct lagg_port *
2317 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2318 {
2319 	struct lagg_port *lp_next, *rval = NULL;
2320 
2321 	/*
2322 	 * Search a port which reports an active link state.
2323 	 */
2324 
2325 #ifdef INVARIANTS
2326 	/*
2327 	 * This is called with either in the network epoch
2328 	 * or with LAGG_XLOCK(sc) held.
2329 	 */
2330 	if (!in_epoch(net_epoch_preempt))
2331 		LAGG_XLOCK_ASSERT(sc);
2332 #endif
2333 
2334 	if (lp == NULL)
2335 		goto search;
2336 	if (LAGG_PORTACTIVE(lp)) {
2337 		rval = lp;
2338 		goto found;
2339 	}
2340 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2341 	    LAGG_PORTACTIVE(lp_next)) {
2342 		rval = lp_next;
2343 		goto found;
2344 	}
2345 
2346 search:
2347 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2348 		if (LAGG_PORTACTIVE(lp_next)) {
2349 			return (lp_next);
2350 		}
2351 	}
2352 found:
2353 	return (rval);
2354 }
2355 
2356 int
2357 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2358 {
2359 
2360 #if defined(KERN_TLS) || defined(RATELIMIT)
2361 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2362 		struct lagg_snd_tag *lst;
2363 		struct m_snd_tag *mst;
2364 
2365 		mst = m->m_pkthdr.snd_tag;
2366 		lst = mst_to_lst(mst);
2367 		if (lst->tag->ifp != ifp) {
2368 			m_freem(m);
2369 			return (EAGAIN);
2370 		}
2371 		m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2372 		m_snd_tag_rele(mst);
2373 	}
2374 #endif
2375 	return (ifp->if_transmit)(ifp, m);
2376 }
2377 
2378 /*
2379  * Simple round robin aggregation
2380  */
2381 static void
2382 lagg_rr_attach(struct lagg_softc *sc)
2383 {
2384 	sc->sc_seq = 0;
2385 	sc->sc_stride = 1;
2386 }
2387 
2388 static int
2389 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2390 {
2391 	struct lagg_port *lp;
2392 	uint32_t p;
2393 
2394 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
2395 	p /= sc->sc_stride;
2396 	p %= sc->sc_count;
2397 	lp = CK_SLIST_FIRST(&sc->sc_ports);
2398 
2399 	while (p--)
2400 		lp = CK_SLIST_NEXT(lp, lp_entries);
2401 
2402 	/*
2403 	 * Check the port's link state. This will return the next active
2404 	 * port if the link is down or the port is NULL.
2405 	 */
2406 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2407 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2408 		m_freem(m);
2409 		return (ENETDOWN);
2410 	}
2411 
2412 	/* Send mbuf */
2413 	return (lagg_enqueue(lp->lp_ifp, m));
2414 }
2415 
2416 /*
2417  * Broadcast mode
2418  */
2419 static int
2420 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2421 {
2422 	int errors = 0;
2423 	int ret;
2424 	struct lagg_port *lp, *last = NULL;
2425 	struct mbuf *m0;
2426 
2427 	NET_EPOCH_ASSERT();
2428 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2429 		if (!LAGG_PORTACTIVE(lp))
2430 			continue;
2431 
2432 		if (last != NULL) {
2433 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2434 			if (m0 == NULL) {
2435 				ret = ENOBUFS;
2436 				errors++;
2437 				break;
2438 			}
2439 			lagg_enqueue(last->lp_ifp, m0);
2440 		}
2441 		last = lp;
2442 	}
2443 
2444 	if (last == NULL) {
2445 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2446 		m_freem(m);
2447 		return (ENOENT);
2448 	}
2449 	if ((last = lagg_link_active(sc, last)) == NULL) {
2450 		errors++;
2451 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2452 		m_freem(m);
2453 		return (ENETDOWN);
2454 	}
2455 
2456 	ret = lagg_enqueue(last->lp_ifp, m);
2457 	if (errors != 0)
2458 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2459 
2460 	return (ret);
2461 }
2462 
2463 /*
2464  * Active failover
2465  */
2466 static int
2467 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2468 {
2469 	struct lagg_port *lp;
2470 
2471 	/* Use the master port if active or the next available port */
2472 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2473 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2474 		m_freem(m);
2475 		return (ENETDOWN);
2476 	}
2477 
2478 	/* Send mbuf */
2479 	return (lagg_enqueue(lp->lp_ifp, m));
2480 }
2481 
2482 static struct mbuf *
2483 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2484 {
2485 	struct ifnet *ifp = sc->sc_ifp;
2486 	struct lagg_port *tmp_tp;
2487 
2488 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2489 		m->m_pkthdr.rcvif = ifp;
2490 		return (m);
2491 	}
2492 
2493 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2494 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2495 		/*
2496 		 * If tmp_tp is null, we've received a packet when all
2497 		 * our links are down. Weird, but process it anyways.
2498 		 */
2499 		if (tmp_tp == NULL || tmp_tp == lp) {
2500 			m->m_pkthdr.rcvif = ifp;
2501 			return (m);
2502 		}
2503 	}
2504 
2505 	m_freem(m);
2506 	return (NULL);
2507 }
2508 
2509 /*
2510  * Loadbalancing
2511  */
2512 static void
2513 lagg_lb_attach(struct lagg_softc *sc)
2514 {
2515 	struct lagg_port *lp;
2516 	struct lagg_lb *lb;
2517 
2518 	LAGG_XLOCK_ASSERT(sc);
2519 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2520 	lb->lb_key = m_ether_tcpip_hash_init();
2521 	sc->sc_psc = lb;
2522 
2523 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2524 		lagg_lb_port_create(lp);
2525 }
2526 
2527 static void
2528 lagg_lb_detach(struct lagg_softc *sc)
2529 {
2530 	struct lagg_lb *lb;
2531 
2532 	lb = (struct lagg_lb *)sc->sc_psc;
2533 	if (lb != NULL)
2534 		free(lb, M_LAGG);
2535 }
2536 
2537 static int
2538 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2539 {
2540 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2541 	struct lagg_port *lp_next;
2542 	int i = 0, rv;
2543 
2544 	rv = 0;
2545 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2546 	LAGG_XLOCK_ASSERT(sc);
2547 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2548 		if (lp_next == lp)
2549 			continue;
2550 		if (i >= LAGG_MAX_PORTS) {
2551 			rv = EINVAL;
2552 			break;
2553 		}
2554 		if (sc->sc_ifflags & IFF_DEBUG)
2555 			printf("%s: port %s at index %d\n",
2556 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2557 		lb->lb_ports[i++] = lp_next;
2558 	}
2559 
2560 	return (rv);
2561 }
2562 
2563 static int
2564 lagg_lb_port_create(struct lagg_port *lp)
2565 {
2566 	struct lagg_softc *sc = lp->lp_softc;
2567 	return (lagg_lb_porttable(sc, NULL));
2568 }
2569 
2570 static void
2571 lagg_lb_port_destroy(struct lagg_port *lp)
2572 {
2573 	struct lagg_softc *sc = lp->lp_softc;
2574 	lagg_lb_porttable(sc, lp);
2575 }
2576 
2577 static int
2578 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2579 {
2580 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2581 	struct lagg_port *lp = NULL;
2582 	uint32_t p = 0;
2583 
2584 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2585 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2586 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2587 	else
2588 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2589 	p %= sc->sc_count;
2590 	lp = lb->lb_ports[p];
2591 
2592 	/*
2593 	 * Check the port's link state. This will return the next active
2594 	 * port if the link is down or the port is NULL.
2595 	 */
2596 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2597 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2598 		m_freem(m);
2599 		return (ENETDOWN);
2600 	}
2601 
2602 	/* Send mbuf */
2603 	return (lagg_enqueue(lp->lp_ifp, m));
2604 }
2605 
2606 /*
2607  * 802.3ad LACP
2608  */
2609 static void
2610 lagg_lacp_attach(struct lagg_softc *sc)
2611 {
2612 	struct lagg_port *lp;
2613 
2614 	lacp_attach(sc);
2615 	LAGG_XLOCK_ASSERT(sc);
2616 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2617 		lacp_port_create(lp);
2618 }
2619 
2620 static void
2621 lagg_lacp_detach(struct lagg_softc *sc)
2622 {
2623 	struct lagg_port *lp;
2624 	void *psc;
2625 
2626 	LAGG_XLOCK_ASSERT(sc);
2627 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2628 		lacp_port_destroy(lp);
2629 
2630 	psc = sc->sc_psc;
2631 	sc->sc_psc = NULL;
2632 	lacp_detach(psc);
2633 }
2634 
2635 static void
2636 lagg_lacp_lladdr(struct lagg_softc *sc)
2637 {
2638 	struct lagg_port *lp;
2639 
2640 	LAGG_SXLOCK_ASSERT(sc);
2641 
2642 	/* purge all the lacp ports */
2643 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2644 		lacp_port_destroy(lp);
2645 
2646 	/* add them back in */
2647 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2648 		lacp_port_create(lp);
2649 }
2650 
2651 static int
2652 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2653 {
2654 	struct lagg_port *lp;
2655 	int err;
2656 
2657 	lp = lacp_select_tx_port(sc, m, &err);
2658 	if (lp == NULL) {
2659 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2660 		m_freem(m);
2661 		return (err);
2662 	}
2663 
2664 	/* Send mbuf */
2665 	return (lagg_enqueue(lp->lp_ifp, m));
2666 }
2667 
2668 static struct mbuf *
2669 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2670 {
2671 	struct ifnet *ifp = sc->sc_ifp;
2672 	struct ether_header *eh;
2673 	u_short etype;
2674 
2675 	eh = mtod(m, struct ether_header *);
2676 	etype = ntohs(eh->ether_type);
2677 
2678 	/* Tap off LACP control messages */
2679 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2680 		m = lacp_input(lp, m);
2681 		if (m == NULL)
2682 			return (NULL);
2683 	}
2684 
2685 	/*
2686 	 * If the port is not collecting or not in the active aggregator then
2687 	 * free and return.
2688 	 */
2689 	if (!lacp_iscollecting(lp) || !lacp_isactive(lp)) {
2690 		m_freem(m);
2691 		return (NULL);
2692 	}
2693 
2694 	m->m_pkthdr.rcvif = ifp;
2695 	return (m);
2696 }
2697 
2698 /* Default input */
2699 static struct mbuf *
2700 lagg_default_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2701 {
2702 	struct ifnet *ifp = sc->sc_ifp;
2703 
2704 	/* Just pass in the packet to our lagg device */
2705 	m->m_pkthdr.rcvif = ifp;
2706 
2707 	return (m);
2708 }
2709