xref: /freebsd/sys/net/if_lagg.c (revision d5b0e70f7e04d971691517ce1304d86a1e367e2e)
1 /*	$OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $	*/
2 
3 /*
4  * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org>
5  * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org>
6  * Copyright (c) 2014, 2016 Marcelo Araujo <araujo@FreeBSD.org>
7  *
8  * Permission to use, copy, modify, and distribute this software for any
9  * purpose with or without fee is hereby granted, provided that the above
10  * copyright notice and this permission notice appear in all copies.
11  *
12  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19  */
20 
21 #include <sys/cdefs.h>
22 __FBSDID("$FreeBSD$");
23 
24 #include "opt_inet.h"
25 #include "opt_inet6.h"
26 #include "opt_kern_tls.h"
27 #include "opt_ratelimit.h"
28 
29 #include <sys/param.h>
30 #include <sys/kernel.h>
31 #include <sys/malloc.h>
32 #include <sys/mbuf.h>
33 #include <sys/queue.h>
34 #include <sys/socket.h>
35 #include <sys/sockio.h>
36 #include <sys/sysctl.h>
37 #include <sys/module.h>
38 #include <sys/priv.h>
39 #include <sys/systm.h>
40 #include <sys/proc.h>
41 #include <sys/lock.h>
42 #include <sys/rmlock.h>
43 #include <sys/sx.h>
44 #include <sys/taskqueue.h>
45 #include <sys/eventhandler.h>
46 
47 #include <net/ethernet.h>
48 #include <net/if.h>
49 #include <net/if_clone.h>
50 #include <net/if_arp.h>
51 #include <net/if_dl.h>
52 #include <net/if_media.h>
53 #include <net/if_types.h>
54 #include <net/if_var.h>
55 #include <net/bpf.h>
56 #include <net/route.h>
57 #include <net/vnet.h>
58 #include <net/infiniband.h>
59 
60 #if defined(INET) || defined(INET6)
61 #include <netinet/in.h>
62 #include <netinet/ip.h>
63 #endif
64 #ifdef INET
65 #include <netinet/in_systm.h>
66 #include <netinet/if_ether.h>
67 #endif
68 
69 #ifdef INET6
70 #include <netinet/ip6.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet6/in6_ifattach.h>
73 #endif
74 
75 #include <net/if_vlan_var.h>
76 #include <net/if_lagg.h>
77 #include <net/ieee8023ad_lacp.h>
78 
79 #ifdef INET6
80 /*
81  * XXX: declare here to avoid to include many inet6 related files..
82  * should be more generalized?
83  */
84 extern void	nd6_setmtu(struct ifnet *);
85 #endif
86 
87 #define	LAGG_SX_INIT(_sc)	sx_init(&(_sc)->sc_sx, "if_lagg sx")
88 #define	LAGG_SX_DESTROY(_sc)	sx_destroy(&(_sc)->sc_sx)
89 #define	LAGG_XLOCK(_sc)		sx_xlock(&(_sc)->sc_sx)
90 #define	LAGG_XUNLOCK(_sc)	sx_xunlock(&(_sc)->sc_sx)
91 #define	LAGG_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
92 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
93 
94 /* Special flags we should propagate to the lagg ports. */
95 static struct {
96 	int flag;
97 	int (*func)(struct ifnet *, int);
98 } lagg_pflags[] = {
99 	{IFF_PROMISC, ifpromisc},
100 	{IFF_ALLMULTI, if_allmulti},
101 	{0, NULL}
102 };
103 
104 struct lagg_snd_tag {
105 	struct m_snd_tag com;
106 	struct m_snd_tag *tag;
107 };
108 
109 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
110 #define	V_lagg_list	VNET(lagg_list)
111 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
112 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
113 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
114 					"if_lagg list", NULL, MTX_DEF)
115 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
116 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
117 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
118 eventhandler_tag	lagg_detach_cookie = NULL;
119 
120 static int	lagg_clone_create(struct if_clone *, char *, size_t,
121 		    struct ifc_data *, struct ifnet **);
122 static int	lagg_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
123 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
124 #define	V_lagg_cloner	VNET(lagg_cloner)
125 static const char laggname[] = "lagg";
126 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
127 
128 static void	lagg_capabilities(struct lagg_softc *);
129 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
130 static int	lagg_port_destroy(struct lagg_port *, int);
131 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *);
132 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *);
133 static void	lagg_linkstate(struct lagg_softc *);
134 static void	lagg_port_state(struct ifnet *, int);
135 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
136 static int	lagg_port_output(struct ifnet *, struct mbuf *,
137 		    const struct sockaddr *, struct route *);
138 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
139 #ifdef LAGG_PORT_STACKING
140 static int	lagg_port_checkstacking(struct lagg_softc *);
141 #endif
142 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
143 static void	lagg_init(void *);
144 static void	lagg_stop(struct lagg_softc *);
145 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
146 #if defined(KERN_TLS) || defined(RATELIMIT)
147 static int	lagg_snd_tag_alloc(struct ifnet *,
148 		    union if_snd_tag_alloc_params *,
149 		    struct m_snd_tag **);
150 static int	lagg_snd_tag_modify(struct m_snd_tag *,
151 		    union if_snd_tag_modify_params *);
152 static int	lagg_snd_tag_query(struct m_snd_tag *,
153 		    union if_snd_tag_query_params *);
154 static void	lagg_snd_tag_free(struct m_snd_tag *);
155 static struct m_snd_tag *lagg_next_snd_tag(struct m_snd_tag *);
156 static void     lagg_ratelimit_query(struct ifnet *,
157 		    struct if_ratelimit_query_results *);
158 #endif
159 static int	lagg_setmulti(struct lagg_port *);
160 static int	lagg_clrmulti(struct lagg_port *);
161 static	void	lagg_setcaps(struct lagg_port *, int cap, int cap2);
162 static	int	lagg_setflag(struct lagg_port *, int, int,
163 		    int (*func)(struct ifnet *, int));
164 static	int	lagg_setflags(struct lagg_port *, int status);
165 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
166 static int	lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
167 static int	lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
168 static void	lagg_qflush(struct ifnet *);
169 static int	lagg_media_change(struct ifnet *);
170 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
171 static struct lagg_port *lagg_link_active(struct lagg_softc *,
172 	    struct lagg_port *);
173 
174 /* Simple round robin */
175 static void	lagg_rr_attach(struct lagg_softc *);
176 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
177 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
178 		    struct mbuf *);
179 
180 /* Active failover */
181 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
182 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
183 		    struct mbuf *);
184 
185 /* Loadbalancing */
186 static void	lagg_lb_attach(struct lagg_softc *);
187 static void	lagg_lb_detach(struct lagg_softc *);
188 static int	lagg_lb_port_create(struct lagg_port *);
189 static void	lagg_lb_port_destroy(struct lagg_port *);
190 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
191 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
192 		    struct mbuf *);
193 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
194 
195 /* Broadcast */
196 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
197 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
198 		    struct mbuf *);
199 
200 /* 802.3ad LACP */
201 static void	lagg_lacp_attach(struct lagg_softc *);
202 static void	lagg_lacp_detach(struct lagg_softc *);
203 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
204 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
205 		    struct mbuf *);
206 static void	lagg_lacp_lladdr(struct lagg_softc *);
207 
208 /* lagg protocol table */
209 static const struct lagg_proto {
210 	lagg_proto	pr_num;
211 	void		(*pr_attach)(struct lagg_softc *);
212 	void		(*pr_detach)(struct lagg_softc *);
213 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
214 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
215 			    struct mbuf *);
216 	int		(*pr_addport)(struct lagg_port *);
217 	void		(*pr_delport)(struct lagg_port *);
218 	void		(*pr_linkstate)(struct lagg_port *);
219 	void 		(*pr_init)(struct lagg_softc *);
220 	void 		(*pr_stop)(struct lagg_softc *);
221 	void 		(*pr_lladdr)(struct lagg_softc *);
222 	void		(*pr_request)(struct lagg_softc *, void *);
223 	void		(*pr_portreq)(struct lagg_port *, void *);
224 } lagg_protos[] = {
225     {
226 	.pr_num = LAGG_PROTO_NONE
227     },
228     {
229 	.pr_num = LAGG_PROTO_ROUNDROBIN,
230 	.pr_attach = lagg_rr_attach,
231 	.pr_start = lagg_rr_start,
232 	.pr_input = lagg_rr_input,
233     },
234     {
235 	.pr_num = LAGG_PROTO_FAILOVER,
236 	.pr_start = lagg_fail_start,
237 	.pr_input = lagg_fail_input,
238     },
239     {
240 	.pr_num = LAGG_PROTO_LOADBALANCE,
241 	.pr_attach = lagg_lb_attach,
242 	.pr_detach = lagg_lb_detach,
243 	.pr_start = lagg_lb_start,
244 	.pr_input = lagg_lb_input,
245 	.pr_addport = lagg_lb_port_create,
246 	.pr_delport = lagg_lb_port_destroy,
247     },
248     {
249 	.pr_num = LAGG_PROTO_LACP,
250 	.pr_attach = lagg_lacp_attach,
251 	.pr_detach = lagg_lacp_detach,
252 	.pr_start = lagg_lacp_start,
253 	.pr_input = lagg_lacp_input,
254 	.pr_addport = lacp_port_create,
255 	.pr_delport = lacp_port_destroy,
256 	.pr_linkstate = lacp_linkstate,
257 	.pr_init = lacp_init,
258 	.pr_stop = lacp_stop,
259 	.pr_lladdr = lagg_lacp_lladdr,
260 	.pr_request = lacp_req,
261 	.pr_portreq = lacp_portreq,
262     },
263     {
264 	.pr_num = LAGG_PROTO_BROADCAST,
265 	.pr_start = lagg_bcast_start,
266 	.pr_input = lagg_bcast_input,
267     },
268 };
269 
270 SYSCTL_DECL(_net_link);
271 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
272     "Link Aggregation");
273 
274 /* Allow input on any failover links */
275 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
276 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
277 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
278     &VNET_NAME(lagg_failover_rx_all), 0,
279     "Accept input from any interface in a failover lagg");
280 
281 /* Default value for using flowid */
282 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
283 #define	V_def_use_flowid	VNET(def_use_flowid)
284 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
285     &VNET_NAME(def_use_flowid), 0,
286     "Default setting for using flow id for load sharing");
287 
288 /* Default value for using numa */
289 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
290 #define	V_def_use_numa	VNET(def_use_numa)
291 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN,
292     &VNET_NAME(def_use_numa), 0,
293     "Use numa to steer flows");
294 
295 /* Default value for flowid shift */
296 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
297 #define	V_def_flowid_shift	VNET(def_flowid_shift)
298 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
299     &VNET_NAME(def_flowid_shift), 0,
300     "Default setting for flowid shift for load sharing");
301 
302 static void
303 vnet_lagg_init(const void *unused __unused)
304 {
305 
306 	LAGG_LIST_LOCK_INIT();
307 	SLIST_INIT(&V_lagg_list);
308 	struct if_clone_addreq req = {
309 		.create_f = lagg_clone_create,
310 		.destroy_f = lagg_clone_destroy,
311 		.flags = IFC_F_AUTOUNIT,
312 	};
313 	V_lagg_cloner = ifc_attach_cloner(laggname, &req);
314 }
315 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
316     vnet_lagg_init, NULL);
317 
318 static void
319 vnet_lagg_uninit(const void *unused __unused)
320 {
321 
322 	ifc_detach_cloner(V_lagg_cloner);
323 	LAGG_LIST_LOCK_DESTROY();
324 }
325 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
326     vnet_lagg_uninit, NULL);
327 
328 static int
329 lagg_modevent(module_t mod, int type, void *data)
330 {
331 
332 	switch (type) {
333 	case MOD_LOAD:
334 		lagg_input_ethernet_p = lagg_input_ethernet;
335 		lagg_input_infiniband_p = lagg_input_infiniband;
336 		lagg_linkstate_p = lagg_port_state;
337 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
338 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
339 		    EVENTHANDLER_PRI_ANY);
340 		break;
341 	case MOD_UNLOAD:
342 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
343 		    lagg_detach_cookie);
344 		lagg_input_ethernet_p = NULL;
345 		lagg_input_infiniband_p = NULL;
346 		lagg_linkstate_p = NULL;
347 		break;
348 	default:
349 		return (EOPNOTSUPP);
350 	}
351 	return (0);
352 }
353 
354 static moduledata_t lagg_mod = {
355 	"if_lagg",
356 	lagg_modevent,
357 	0
358 };
359 
360 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
361 MODULE_VERSION(if_lagg, 1);
362 MODULE_DEPEND(if_lagg, if_infiniband, 1, 1, 1);
363 
364 static void
365 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
366 {
367 
368 	LAGG_XLOCK_ASSERT(sc);
369 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
370 	    __func__, sc));
371 
372 	if (sc->sc_ifflags & IFF_DEBUG)
373 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
374 
375 	if (lagg_protos[pr].pr_attach != NULL)
376 		lagg_protos[pr].pr_attach(sc);
377 	sc->sc_proto = pr;
378 }
379 
380 static void
381 lagg_proto_detach(struct lagg_softc *sc)
382 {
383 	lagg_proto pr;
384 
385 	LAGG_XLOCK_ASSERT(sc);
386 	pr = sc->sc_proto;
387 	sc->sc_proto = LAGG_PROTO_NONE;
388 
389 	if (lagg_protos[pr].pr_detach != NULL)
390 		lagg_protos[pr].pr_detach(sc);
391 }
392 
393 static int
394 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
395 {
396 
397 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
398 }
399 
400 static struct mbuf *
401 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
402 {
403 
404 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
405 }
406 
407 static int
408 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
409 {
410 
411 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
412 		return (0);
413 	else
414 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
415 }
416 
417 static void
418 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
419 {
420 
421 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
422 		lagg_protos[sc->sc_proto].pr_delport(lp);
423 }
424 
425 static void
426 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
427 {
428 
429 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
430 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
431 }
432 
433 static void
434 lagg_proto_init(struct lagg_softc *sc)
435 {
436 
437 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
438 		lagg_protos[sc->sc_proto].pr_init(sc);
439 }
440 
441 static void
442 lagg_proto_stop(struct lagg_softc *sc)
443 {
444 
445 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
446 		lagg_protos[sc->sc_proto].pr_stop(sc);
447 }
448 
449 static void
450 lagg_proto_lladdr(struct lagg_softc *sc)
451 {
452 
453 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
454 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
455 }
456 
457 static void
458 lagg_proto_request(struct lagg_softc *sc, void *v)
459 {
460 
461 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
462 		lagg_protos[sc->sc_proto].pr_request(sc, v);
463 }
464 
465 static void
466 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
467 {
468 
469 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
470 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
471 }
472 
473 /*
474  * This routine is run via an vlan
475  * config EVENT
476  */
477 static void
478 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
479 {
480 	struct lagg_softc *sc = ifp->if_softc;
481 	struct lagg_port *lp;
482 
483 	if (ifp->if_softc !=  arg)   /* Not our event */
484 		return;
485 
486 	LAGG_XLOCK(sc);
487 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
488 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
489 	LAGG_XUNLOCK(sc);
490 }
491 
492 /*
493  * This routine is run via an vlan
494  * unconfig EVENT
495  */
496 static void
497 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
498 {
499 	struct lagg_softc *sc = ifp->if_softc;
500 	struct lagg_port *lp;
501 
502 	if (ifp->if_softc !=  arg)   /* Not our event */
503 		return;
504 
505 	LAGG_XLOCK(sc);
506 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
507 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
508 	LAGG_XUNLOCK(sc);
509 }
510 
511 static int
512 lagg_clone_create(struct if_clone *ifc, char *name, size_t len,
513     struct ifc_data *ifd, struct ifnet **ifpp)
514 {
515 	struct iflaggparam iflp;
516 	struct lagg_softc *sc;
517 	struct ifnet *ifp;
518 	int if_type;
519 	int error;
520 	static const uint8_t eaddr[LAGG_ADDR_LEN];
521 
522 	if (ifd->params != NULL) {
523 		error = ifc_copyin(ifd, &iflp, sizeof(iflp));
524 		if (error)
525 			return (error);
526 
527 		switch (iflp.lagg_type) {
528 		case LAGG_TYPE_ETHERNET:
529 			if_type = IFT_ETHER;
530 			break;
531 		case LAGG_TYPE_INFINIBAND:
532 			if_type = IFT_INFINIBAND;
533 			break;
534 		default:
535 			return (EINVAL);
536 		}
537 	} else {
538 		if_type = IFT_ETHER;
539 	}
540 
541 	sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO);
542 	ifp = sc->sc_ifp = if_alloc(if_type);
543 	if (ifp == NULL) {
544 		free(sc, M_LAGG);
545 		return (ENOSPC);
546 	}
547 	LAGG_SX_INIT(sc);
548 
549 	mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
550 	callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
551 
552 	LAGG_XLOCK(sc);
553 	if (V_def_use_flowid)
554 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
555 	if (V_def_use_numa)
556 		sc->sc_opts |= LAGG_OPT_USE_NUMA;
557 	sc->flowid_shift = V_def_flowid_shift;
558 
559 	/* Hash all layers by default */
560 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
561 
562 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
563 
564 	CK_SLIST_INIT(&sc->sc_ports);
565 
566 	switch (if_type) {
567 	case IFT_ETHER:
568 		/* Initialise pseudo media types */
569 		ifmedia_init(&sc->sc_media, 0, lagg_media_change,
570 		    lagg_media_status);
571 		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
572 		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
573 
574 		if_initname(ifp, laggname, ifd->unit);
575 		ifp->if_transmit = lagg_transmit_ethernet;
576 		break;
577 	case IFT_INFINIBAND:
578 		if_initname(ifp, laggname, ifd->unit);
579 		ifp->if_transmit = lagg_transmit_infiniband;
580 		break;
581 	default:
582 		break;
583 	}
584 	ifp->if_softc = sc;
585 	ifp->if_qflush = lagg_qflush;
586 	ifp->if_init = lagg_init;
587 	ifp->if_ioctl = lagg_ioctl;
588 	ifp->if_get_counter = lagg_get_counter;
589 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
590 #if defined(KERN_TLS) || defined(RATELIMIT)
591 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
592 	ifp->if_ratelimit_query = lagg_ratelimit_query;
593 #endif
594 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
595 
596 	/*
597 	 * Attach as an ordinary ethernet device, children will be attached
598 	 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
599 	 */
600 	switch (if_type) {
601 	case IFT_ETHER:
602 		ether_ifattach(ifp, eaddr);
603 		break;
604 	case IFT_INFINIBAND:
605 		infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr);
606 		break;
607 	default:
608 		break;
609 	}
610 
611 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
612 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
613 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
614 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
615 
616 	/* Insert into the global list of laggs */
617 	LAGG_LIST_LOCK();
618 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
619 	LAGG_LIST_UNLOCK();
620 	LAGG_XUNLOCK(sc);
621 	*ifpp = ifp;
622 
623 	return (0);
624 }
625 
626 static int
627 lagg_clone_destroy(struct if_clone *ifc, struct ifnet *ifp, uint32_t flags)
628 {
629 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
630 	struct lagg_port *lp;
631 
632 	LAGG_XLOCK(sc);
633 	sc->sc_destroying = 1;
634 	lagg_stop(sc);
635 	ifp->if_flags &= ~IFF_UP;
636 
637 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
638 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
639 
640 	/* Shutdown and remove lagg ports */
641 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
642 		lagg_port_destroy(lp, 1);
643 
644 	/* Unhook the aggregation protocol */
645 	lagg_proto_detach(sc);
646 	LAGG_XUNLOCK(sc);
647 
648 	switch (ifp->if_type) {
649 	case IFT_ETHER:
650 		ifmedia_removeall(&sc->sc_media);
651 		ether_ifdetach(ifp);
652 		break;
653 	case IFT_INFINIBAND:
654 		infiniband_ifdetach(ifp);
655 		break;
656 	default:
657 		break;
658 	}
659 	if_free(ifp);
660 
661 	LAGG_LIST_LOCK();
662 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
663 	LAGG_LIST_UNLOCK();
664 
665 	mtx_destroy(&sc->sc_mtx);
666 	LAGG_SX_DESTROY(sc);
667 	free(sc, M_LAGG);
668 
669 	return (0);
670 }
671 
672 static void
673 lagg_capabilities(struct lagg_softc *sc)
674 {
675 	struct lagg_port *lp;
676 	int cap, cap2, ena, ena2, pena, pena2;
677 	uint64_t hwa;
678 	struct ifnet_hw_tsomax hw_tsomax;
679 
680 	LAGG_XLOCK_ASSERT(sc);
681 
682 	/* Get common enabled capabilities for the lagg ports */
683 	ena = ena2 = ~0;
684 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
685 		ena &= lp->lp_ifp->if_capenable;
686 		ena2 &= lp->lp_ifp->if_capenable2;
687 	}
688 	if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
689 		ena = ena2 = 0;
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 		pena2 = ena2;
698 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
699 			lagg_setcaps(lp, ena, ena2);
700 			ena &= lp->lp_ifp->if_capenable;
701 			ena2 &= lp->lp_ifp->if_capenable2;
702 		}
703 	} while (pena != ena || pena2 != ena2);
704 
705 	/* Get other capabilities from the lagg ports */
706 	cap = cap2 = ~0;
707 	hwa = ~(uint64_t)0;
708 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
709 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
710 		cap &= lp->lp_ifp->if_capabilities;
711 		cap2 &= lp->lp_ifp->if_capabilities2;
712 		hwa &= lp->lp_ifp->if_hwassist;
713 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
714 	}
715 	if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
716 		cap = cap2 = hwa = 0;
717 
718 	if (sc->sc_ifp->if_capabilities != cap ||
719 	    sc->sc_ifp->if_capenable != ena ||
720 	    sc->sc_ifp->if_capenable2 != ena2 ||
721 	    sc->sc_ifp->if_hwassist != hwa ||
722 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
723 		sc->sc_ifp->if_capabilities = cap;
724 		sc->sc_ifp->if_capabilities2 = cap2;
725 		sc->sc_ifp->if_capenable = ena;
726 		sc->sc_ifp->if_capenable2 = ena2;
727 		sc->sc_ifp->if_hwassist = hwa;
728 		getmicrotime(&sc->sc_ifp->if_lastchange);
729 
730 		if (sc->sc_ifflags & IFF_DEBUG)
731 			if_printf(sc->sc_ifp,
732 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
733 	}
734 }
735 
736 static int
737 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
738 {
739 	struct lagg_softc *sc_ptr;
740 	struct lagg_port *lp, *tlp;
741 	struct ifreq ifr;
742 	int error, i, oldmtu;
743 	int if_type;
744 	uint64_t *pval;
745 
746 	LAGG_XLOCK_ASSERT(sc);
747 
748 	if (sc->sc_ifp == ifp) {
749 		if_printf(sc->sc_ifp,
750 		    "cannot add a lagg to itself as a port\n");
751 		return (EINVAL);
752 	}
753 
754 	if (sc->sc_destroying == 1)
755 		return (ENXIO);
756 
757 	/* Limit the maximal number of lagg ports */
758 	if (sc->sc_count >= LAGG_MAX_PORTS)
759 		return (ENOSPC);
760 
761 	/* Check if port has already been associated to a lagg */
762 	if (ifp->if_lagg != NULL) {
763 		/* Port is already in the current lagg? */
764 		lp = (struct lagg_port *)ifp->if_lagg;
765 		if (lp->lp_softc == sc)
766 			return (EEXIST);
767 		return (EBUSY);
768 	}
769 
770 	switch (sc->sc_ifp->if_type) {
771 	case IFT_ETHER:
772 		/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
773 		if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
774 			return (EPROTONOSUPPORT);
775 		if_type = IFT_IEEE8023ADLAG;
776 		break;
777 	case IFT_INFINIBAND:
778 		/* XXX Disallow non-infiniband interfaces */
779 		if (ifp->if_type != IFT_INFINIBAND)
780 			return (EPROTONOSUPPORT);
781 		if_type = IFT_INFINIBANDLAG;
782 		break;
783 	default:
784 		break;
785 	}
786 
787 	/* Allow the first Ethernet member to define the MTU */
788 	oldmtu = -1;
789 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
790 		sc->sc_ifp->if_mtu = ifp->if_mtu;
791 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
792 		if (ifp->if_ioctl == NULL) {
793 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
794 			    ifp->if_xname);
795 			return (EINVAL);
796 		}
797 		oldmtu = ifp->if_mtu;
798 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
799 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
800 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
801 		if (error != 0) {
802 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
803 			    ifp->if_xname);
804 			return (error);
805 		}
806 		ifr.ifr_mtu = oldmtu;
807 	}
808 
809 	lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO);
810 	lp->lp_softc = sc;
811 
812 	/* Check if port is a stacked lagg */
813 	LAGG_LIST_LOCK();
814 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
815 		if (ifp == sc_ptr->sc_ifp) {
816 			LAGG_LIST_UNLOCK();
817 			free(lp, M_LAGG);
818 			if (oldmtu != -1)
819 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
820 				    (caddr_t)&ifr);
821 			return (EINVAL);
822 			/* XXX disable stacking for the moment, its untested */
823 #ifdef LAGG_PORT_STACKING
824 			lp->lp_flags |= LAGG_PORT_STACK;
825 			if (lagg_port_checkstacking(sc_ptr) >=
826 			    LAGG_MAX_STACKING) {
827 				LAGG_LIST_UNLOCK();
828 				free(lp, M_LAGG);
829 				if (oldmtu != -1)
830 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
831 					    (caddr_t)&ifr);
832 				return (E2BIG);
833 			}
834 #endif
835 		}
836 	}
837 	LAGG_LIST_UNLOCK();
838 
839 	if_ref(ifp);
840 	lp->lp_ifp = ifp;
841 
842 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
843 	lp->lp_ifcapenable = ifp->if_capenable;
844 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
845 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
846 		lagg_proto_lladdr(sc);
847 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
848 	} else {
849 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
850 	}
851 	lagg_setflags(lp, 1);
852 
853 	if (CK_SLIST_EMPTY(&sc->sc_ports))
854 		sc->sc_primary = lp;
855 
856 	/* Change the interface type */
857 	lp->lp_iftype = ifp->if_type;
858 	ifp->if_type = if_type;
859 	ifp->if_lagg = lp;
860 	lp->lp_ioctl = ifp->if_ioctl;
861 	ifp->if_ioctl = lagg_port_ioctl;
862 	lp->lp_output = ifp->if_output;
863 	ifp->if_output = lagg_port_output;
864 
865 	/* Read port counters */
866 	pval = lp->port_counters.val;
867 	for (i = 0; i < IFCOUNTERS; i++, pval++)
868 		*pval = ifp->if_get_counter(ifp, i);
869 
870 	/*
871 	 * Insert into the list of ports.
872 	 * Keep ports sorted by if_index. It is handy, when configuration
873 	 * is predictable and `ifconfig laggN create ...` command
874 	 * will lead to the same result each time.
875 	 */
876 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
877 		if (tlp->lp_ifp->if_index < ifp->if_index && (
878 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
879 		    ((struct  lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
880 		    ifp->if_index))
881 			break;
882 	}
883 	if (tlp != NULL)
884 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
885 	else
886 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
887 	sc->sc_count++;
888 
889 	lagg_setmulti(lp);
890 
891 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
892 		/* Remove the port, without calling pr_delport. */
893 		lagg_port_destroy(lp, 0);
894 		if (oldmtu != -1)
895 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
896 		return (error);
897 	}
898 
899 	/* Update lagg capabilities */
900 	lagg_capabilities(sc);
901 	lagg_linkstate(sc);
902 
903 	return (0);
904 }
905 
906 #ifdef LAGG_PORT_STACKING
907 static int
908 lagg_port_checkstacking(struct lagg_softc *sc)
909 {
910 	struct lagg_softc *sc_ptr;
911 	struct lagg_port *lp;
912 	int m = 0;
913 
914 	LAGG_SXLOCK_ASSERT(sc);
915 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
916 		if (lp->lp_flags & LAGG_PORT_STACK) {
917 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
918 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
919 		}
920 	}
921 
922 	return (m + 1);
923 }
924 #endif
925 
926 static void
927 lagg_port_destroy_cb(epoch_context_t ec)
928 {
929 	struct lagg_port *lp;
930 	struct ifnet *ifp;
931 
932 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
933 	ifp = lp->lp_ifp;
934 
935 	if_rele(ifp);
936 	free(lp, M_LAGG);
937 }
938 
939 static int
940 lagg_port_destroy(struct lagg_port *lp, int rundelport)
941 {
942 	struct lagg_softc *sc = lp->lp_softc;
943 	struct lagg_port *lp_ptr, *lp0;
944 	struct ifnet *ifp = lp->lp_ifp;
945 	uint64_t *pval, vdiff;
946 	int i;
947 
948 	LAGG_XLOCK_ASSERT(sc);
949 
950 	if (rundelport)
951 		lagg_proto_delport(sc, lp);
952 
953 	if (lp->lp_detaching == 0)
954 		lagg_clrmulti(lp);
955 
956 	/* Restore interface */
957 	ifp->if_type = lp->lp_iftype;
958 	ifp->if_ioctl = lp->lp_ioctl;
959 	ifp->if_output = lp->lp_output;
960 	ifp->if_lagg = NULL;
961 
962 	/* Update detached port counters */
963 	pval = lp->port_counters.val;
964 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
965 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
966 		sc->detached_counters.val[i] += vdiff;
967 	}
968 
969 	/* Finally, remove the port from the lagg */
970 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
971 	sc->sc_count--;
972 
973 	/* Update the primary interface */
974 	if (lp == sc->sc_primary) {
975 		uint8_t lladdr[LAGG_ADDR_LEN];
976 
977 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
978 			bzero(&lladdr, LAGG_ADDR_LEN);
979 		else
980 			bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
981 		sc->sc_primary = lp0;
982 		if (sc->sc_destroying == 0) {
983 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
984 			lagg_proto_lladdr(sc);
985 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
986 
987 			/*
988 			 * Update lladdr for each port (new primary needs update
989 			 * as well, to switch from old lladdr to its 'real' one).
990 			 * We can skip this if the lagg is being destroyed.
991 			 */
992 			CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
993 				if_setlladdr(lp_ptr->lp_ifp, lladdr,
994 				    lp_ptr->lp_ifp->if_addrlen);
995 		}
996 	}
997 
998 	if (lp->lp_ifflags)
999 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
1000 
1001 	if (lp->lp_detaching == 0) {
1002 		lagg_setflags(lp, 0);
1003 		lagg_setcaps(lp, lp->lp_ifcapenable, lp->lp_ifcapenable2);
1004 		if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
1005 	}
1006 
1007 	/*
1008 	 * free port and release it's ifnet reference after a grace period has
1009 	 * elapsed.
1010 	 */
1011 	NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
1012 	/* Update lagg capabilities */
1013 	lagg_capabilities(sc);
1014 	lagg_linkstate(sc);
1015 
1016 	return (0);
1017 }
1018 
1019 static int
1020 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1021 {
1022 	struct epoch_tracker et;
1023 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
1024 	struct lagg_softc *sc;
1025 	struct lagg_port *lp = NULL;
1026 	int error = 0;
1027 
1028 	/* Should be checked by the caller */
1029 	switch (ifp->if_type) {
1030 	case IFT_IEEE8023ADLAG:
1031 	case IFT_INFINIBANDLAG:
1032 		if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1033 			goto fallback;
1034 		break;
1035 	default:
1036 		goto fallback;
1037 	}
1038 
1039 	switch (cmd) {
1040 	case SIOCGLAGGPORT:
1041 		if (rp->rp_portname[0] == '\0' ||
1042 		    ifunit(rp->rp_portname) != ifp) {
1043 			error = EINVAL;
1044 			break;
1045 		}
1046 
1047 		NET_EPOCH_ENTER(et);
1048 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1049 			error = ENOENT;
1050 			NET_EPOCH_EXIT(et);
1051 			break;
1052 		}
1053 
1054 		lagg_port2req(lp, rp);
1055 		NET_EPOCH_EXIT(et);
1056 		break;
1057 
1058 	case SIOCSIFCAP:
1059 	case SIOCSIFCAPNV:
1060 		if (lp->lp_ioctl == NULL) {
1061 			error = EINVAL;
1062 			break;
1063 		}
1064 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1065 		if (error)
1066 			break;
1067 
1068 		/* Update lagg interface capabilities */
1069 		LAGG_XLOCK(sc);
1070 		lagg_capabilities(sc);
1071 		LAGG_XUNLOCK(sc);
1072 		VLAN_CAPABILITIES(sc->sc_ifp);
1073 		break;
1074 
1075 	case SIOCSIFMTU:
1076 		/* Do not allow the MTU to be changed once joined */
1077 		error = EINVAL;
1078 		break;
1079 
1080 	default:
1081 		goto fallback;
1082 	}
1083 
1084 	return (error);
1085 
1086 fallback:
1087 	if (lp != NULL && lp->lp_ioctl != NULL)
1088 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1089 
1090 	return (EINVAL);
1091 }
1092 
1093 /*
1094  * Requests counter @cnt data.
1095  *
1096  * Counter value is calculated the following way:
1097  * 1) for each port, sum  difference between current and "initial" measurements.
1098  * 2) add lagg logical interface counters.
1099  * 3) add data from detached_counters array.
1100  *
1101  * We also do the following things on ports attach/detach:
1102  * 1) On port attach we store all counters it has into port_counter array.
1103  * 2) On port detach we add the different between "initial" and
1104  *   current counters data to detached_counters array.
1105  */
1106 static uint64_t
1107 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1108 {
1109 	struct epoch_tracker et;
1110 	struct lagg_softc *sc;
1111 	struct lagg_port *lp;
1112 	struct ifnet *lpifp;
1113 	uint64_t newval, oldval, vsum;
1114 
1115 	/* Revise this when we've got non-generic counters. */
1116 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1117 
1118 	sc = (struct lagg_softc *)ifp->if_softc;
1119 
1120 	vsum = 0;
1121 	NET_EPOCH_ENTER(et);
1122 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1123 		/* Saved attached value */
1124 		oldval = lp->port_counters.val[cnt];
1125 		/* current value */
1126 		lpifp = lp->lp_ifp;
1127 		newval = lpifp->if_get_counter(lpifp, cnt);
1128 		/* Calculate diff and save new */
1129 		vsum += newval - oldval;
1130 	}
1131 	NET_EPOCH_EXIT(et);
1132 
1133 	/*
1134 	 * Add counter data which might be added by upper
1135 	 * layer protocols operating on logical interface.
1136 	 */
1137 	vsum += if_get_counter_default(ifp, cnt);
1138 
1139 	/*
1140 	 * Add counter data from detached ports counters
1141 	 */
1142 	vsum += sc->detached_counters.val[cnt];
1143 
1144 	return (vsum);
1145 }
1146 
1147 /*
1148  * For direct output to child ports.
1149  */
1150 static int
1151 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1152 	const struct sockaddr *dst, struct route *ro)
1153 {
1154 	struct lagg_port *lp = ifp->if_lagg;
1155 
1156 	switch (dst->sa_family) {
1157 		case pseudo_AF_HDRCMPLT:
1158 		case AF_UNSPEC:
1159 			if (lp != NULL)
1160 				return ((*lp->lp_output)(ifp, m, dst, ro));
1161 	}
1162 
1163 	/* drop any other frames */
1164 	m_freem(m);
1165 	return (ENETDOWN);
1166 }
1167 
1168 static void
1169 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1170 {
1171 	struct lagg_port *lp;
1172 	struct lagg_softc *sc;
1173 
1174 	if ((lp = ifp->if_lagg) == NULL)
1175 		return;
1176 	/* If the ifnet is just being renamed, don't do anything. */
1177 	if (ifp->if_flags & IFF_RENAMING)
1178 		return;
1179 
1180 	sc = lp->lp_softc;
1181 
1182 	LAGG_XLOCK(sc);
1183 	lp->lp_detaching = 1;
1184 	lagg_port_destroy(lp, 1);
1185 	LAGG_XUNLOCK(sc);
1186 	VLAN_CAPABILITIES(sc->sc_ifp);
1187 }
1188 
1189 static void
1190 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1191 {
1192 	struct lagg_softc *sc = lp->lp_softc;
1193 
1194 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1195 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1196 	rp->rp_prio = lp->lp_prio;
1197 	rp->rp_flags = lp->lp_flags;
1198 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1199 
1200 	/* Add protocol specific flags */
1201 	switch (sc->sc_proto) {
1202 		case LAGG_PROTO_FAILOVER:
1203 			if (lp == sc->sc_primary)
1204 				rp->rp_flags |= LAGG_PORT_MASTER;
1205 			if (lp == lagg_link_active(sc, sc->sc_primary))
1206 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1207 			break;
1208 
1209 		case LAGG_PROTO_ROUNDROBIN:
1210 		case LAGG_PROTO_LOADBALANCE:
1211 		case LAGG_PROTO_BROADCAST:
1212 			if (LAGG_PORTACTIVE(lp))
1213 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1214 			break;
1215 
1216 		case LAGG_PROTO_LACP:
1217 			/* LACP has a different definition of active */
1218 			if (lacp_isactive(lp))
1219 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1220 			if (lacp_iscollecting(lp))
1221 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1222 			if (lacp_isdistributing(lp))
1223 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1224 			break;
1225 	}
1226 
1227 }
1228 
1229 static void
1230 lagg_watchdog_infiniband(void *arg)
1231 {
1232 	struct epoch_tracker et;
1233 	struct lagg_softc *sc;
1234 	struct lagg_port *lp;
1235 	struct ifnet *ifp;
1236 	struct ifnet *lp_ifp;
1237 
1238 	sc = arg;
1239 
1240 	/*
1241 	 * Because infiniband nodes have a fixed MAC address, which is
1242 	 * generated by the so-called GID, we need to regularly update
1243 	 * the link level address of the parent lagg<N> device when
1244 	 * the active port changes. Possibly we could piggy-back on
1245 	 * link up/down events aswell, but using a timer also provides
1246 	 * a guarantee against too frequent events. This operation
1247 	 * does not have to be atomic.
1248 	 */
1249 	NET_EPOCH_ENTER(et);
1250 	lp = lagg_link_active(sc, sc->sc_primary);
1251 	if (lp != NULL) {
1252 		ifp = sc->sc_ifp;
1253 		lp_ifp = lp->lp_ifp;
1254 
1255 		if (ifp != NULL && lp_ifp != NULL &&
1256 		    (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0 ||
1257 		     memcmp(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen) != 0)) {
1258 			memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1259 			memcpy(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen);
1260 
1261 			CURVNET_SET(ifp->if_vnet);
1262 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1263 			CURVNET_RESTORE();
1264 		}
1265 	}
1266 	NET_EPOCH_EXIT(et);
1267 
1268 	callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1269 }
1270 
1271 static void
1272 lagg_init(void *xsc)
1273 {
1274 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1275 	struct ifnet *ifp = sc->sc_ifp;
1276 	struct lagg_port *lp;
1277 
1278 	LAGG_XLOCK(sc);
1279 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1280 		LAGG_XUNLOCK(sc);
1281 		return;
1282 	}
1283 
1284 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1285 
1286 	/*
1287 	 * Update the port lladdrs if needed.
1288 	 * This might be if_setlladdr() notification
1289 	 * that lladdr has been changed.
1290 	 */
1291 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1292 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1293 		    ifp->if_addrlen) != 0)
1294 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1295 	}
1296 
1297 	lagg_proto_init(sc);
1298 
1299 	if (ifp->if_type == IFT_INFINIBAND) {
1300 		mtx_lock(&sc->sc_mtx);
1301 		lagg_watchdog_infiniband(sc);
1302 		mtx_unlock(&sc->sc_mtx);
1303 	}
1304 
1305 	LAGG_XUNLOCK(sc);
1306 }
1307 
1308 static void
1309 lagg_stop(struct lagg_softc *sc)
1310 {
1311 	struct ifnet *ifp = sc->sc_ifp;
1312 
1313 	LAGG_XLOCK_ASSERT(sc);
1314 
1315 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1316 		return;
1317 
1318 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1319 
1320 	lagg_proto_stop(sc);
1321 
1322 	mtx_lock(&sc->sc_mtx);
1323 	callout_stop(&sc->sc_watchdog);
1324 	mtx_unlock(&sc->sc_mtx);
1325 
1326 	callout_drain(&sc->sc_watchdog);
1327 }
1328 
1329 static int
1330 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1331 {
1332 	struct epoch_tracker et;
1333 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1334 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1335 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1336 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1337 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1338 	struct ifreq *ifr = (struct ifreq *)data;
1339 	struct lagg_port *lp;
1340 	struct ifnet *tpif;
1341 	struct thread *td = curthread;
1342 	char *buf, *outbuf;
1343 	int count, buflen, len, error = 0, oldmtu;
1344 
1345 	bzero(&rpbuf, sizeof(rpbuf));
1346 
1347 	/* XXX: This can race with lagg_clone_destroy. */
1348 
1349 	switch (cmd) {
1350 	case SIOCGLAGG:
1351 		LAGG_XLOCK(sc);
1352 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1353 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1354 		ra->ra_proto = sc->sc_proto;
1355 		lagg_proto_request(sc, &ra->ra_psc);
1356 		count = 0;
1357 		buf = outbuf;
1358 		len = min(ra->ra_size, buflen);
1359 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1360 			if (len < sizeof(rpbuf))
1361 				break;
1362 
1363 			lagg_port2req(lp, &rpbuf);
1364 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1365 			count++;
1366 			buf += sizeof(rpbuf);
1367 			len -= sizeof(rpbuf);
1368 		}
1369 		LAGG_XUNLOCK(sc);
1370 		ra->ra_ports = count;
1371 		ra->ra_size = count * sizeof(rpbuf);
1372 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1373 		free(outbuf, M_TEMP);
1374 		break;
1375 	case SIOCSLAGG:
1376 		error = priv_check(td, PRIV_NET_LAGG);
1377 		if (error)
1378 			break;
1379 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1380 			error = EPROTONOSUPPORT;
1381 			break;
1382 		}
1383 		/* Infiniband only supports the failover protocol. */
1384 		if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1385 		    ifp->if_type == IFT_INFINIBAND) {
1386 			error = EPROTONOSUPPORT;
1387 			break;
1388 		}
1389 		LAGG_XLOCK(sc);
1390 		lagg_proto_detach(sc);
1391 		lagg_proto_attach(sc, ra->ra_proto);
1392 		LAGG_XUNLOCK(sc);
1393 		break;
1394 	case SIOCGLAGGOPTS:
1395 		LAGG_XLOCK(sc);
1396 		ro->ro_opts = sc->sc_opts;
1397 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1398 			struct lacp_softc *lsc;
1399 
1400 			lsc = (struct lacp_softc *)sc->sc_psc;
1401 			if (lsc->lsc_debug.lsc_tx_test != 0)
1402 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1403 			if (lsc->lsc_debug.lsc_rx_test != 0)
1404 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1405 			if (lsc->lsc_strict_mode != 0)
1406 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1407 			if (lsc->lsc_fast_timeout != 0)
1408 				ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1409 
1410 			ro->ro_active = sc->sc_active;
1411 		} else {
1412 			ro->ro_active = 0;
1413 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1414 				ro->ro_active += LAGG_PORTACTIVE(lp);
1415 		}
1416 		ro->ro_bkt = sc->sc_stride;
1417 		ro->ro_flapping = sc->sc_flapping;
1418 		ro->ro_flowid_shift = sc->flowid_shift;
1419 		LAGG_XUNLOCK(sc);
1420 		break;
1421 	case SIOCSLAGGOPTS:
1422 		error = priv_check(td, PRIV_NET_LAGG);
1423 		if (error)
1424 			break;
1425 
1426 		/*
1427 		 * The stride option was added without defining a corresponding
1428 		 * LAGG_OPT flag, so handle a non-zero value before checking
1429 		 * anything else to preserve compatibility.
1430 		 */
1431 		LAGG_XLOCK(sc);
1432 		if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1433 			if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1434 				LAGG_XUNLOCK(sc);
1435 				error = EINVAL;
1436 				break;
1437 			}
1438 			sc->sc_stride = ro->ro_bkt;
1439 		}
1440 		if (ro->ro_opts == 0) {
1441 			LAGG_XUNLOCK(sc);
1442 			break;
1443 		}
1444 
1445 		/*
1446 		 * Set options.  LACP options are stored in sc->sc_psc,
1447 		 * not in sc_opts.
1448 		 */
1449 		int valid, lacp;
1450 
1451 		switch (ro->ro_opts) {
1452 		case LAGG_OPT_USE_FLOWID:
1453 		case -LAGG_OPT_USE_FLOWID:
1454 		case LAGG_OPT_USE_NUMA:
1455 		case -LAGG_OPT_USE_NUMA:
1456 		case LAGG_OPT_FLOWIDSHIFT:
1457 		case LAGG_OPT_RR_LIMIT:
1458 			valid = 1;
1459 			lacp = 0;
1460 			break;
1461 		case LAGG_OPT_LACP_TXTEST:
1462 		case -LAGG_OPT_LACP_TXTEST:
1463 		case LAGG_OPT_LACP_RXTEST:
1464 		case -LAGG_OPT_LACP_RXTEST:
1465 		case LAGG_OPT_LACP_STRICT:
1466 		case -LAGG_OPT_LACP_STRICT:
1467 		case LAGG_OPT_LACP_FAST_TIMO:
1468 		case -LAGG_OPT_LACP_FAST_TIMO:
1469 			valid = lacp = 1;
1470 			break;
1471 		default:
1472 			valid = lacp = 0;
1473 			break;
1474 		}
1475 
1476 		if (valid == 0 ||
1477 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1478 			/* Invalid combination of options specified. */
1479 			error = EINVAL;
1480 			LAGG_XUNLOCK(sc);
1481 			break;	/* Return from SIOCSLAGGOPTS. */
1482 		}
1483 
1484 		/*
1485 		 * Store new options into sc->sc_opts except for
1486 		 * FLOWIDSHIFT, RR and LACP options.
1487 		 */
1488 		if (lacp == 0) {
1489 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1490 				sc->flowid_shift = ro->ro_flowid_shift;
1491 			else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1492 				if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1493 				    ro->ro_bkt == 0) {
1494 					error = EINVAL;
1495 					LAGG_XUNLOCK(sc);
1496 					break;
1497 				}
1498 				sc->sc_stride = ro->ro_bkt;
1499 			} else if (ro->ro_opts > 0)
1500 				sc->sc_opts |= ro->ro_opts;
1501 			else
1502 				sc->sc_opts &= ~ro->ro_opts;
1503 		} else {
1504 			struct lacp_softc *lsc;
1505 			struct lacp_port *lp;
1506 
1507 			lsc = (struct lacp_softc *)sc->sc_psc;
1508 
1509 			switch (ro->ro_opts) {
1510 			case LAGG_OPT_LACP_TXTEST:
1511 				lsc->lsc_debug.lsc_tx_test = 1;
1512 				break;
1513 			case -LAGG_OPT_LACP_TXTEST:
1514 				lsc->lsc_debug.lsc_tx_test = 0;
1515 				break;
1516 			case LAGG_OPT_LACP_RXTEST:
1517 				lsc->lsc_debug.lsc_rx_test = 1;
1518 				break;
1519 			case -LAGG_OPT_LACP_RXTEST:
1520 				lsc->lsc_debug.lsc_rx_test = 0;
1521 				break;
1522 			case LAGG_OPT_LACP_STRICT:
1523 				lsc->lsc_strict_mode = 1;
1524 				break;
1525 			case -LAGG_OPT_LACP_STRICT:
1526 				lsc->lsc_strict_mode = 0;
1527 				break;
1528 			case LAGG_OPT_LACP_FAST_TIMO:
1529 				LACP_LOCK(lsc);
1530         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1531                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1532 				LACP_UNLOCK(lsc);
1533 				lsc->lsc_fast_timeout = 1;
1534 				break;
1535 			case -LAGG_OPT_LACP_FAST_TIMO:
1536 				LACP_LOCK(lsc);
1537         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1538                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1539 				LACP_UNLOCK(lsc);
1540 				lsc->lsc_fast_timeout = 0;
1541 				break;
1542 			}
1543 		}
1544 		LAGG_XUNLOCK(sc);
1545 		break;
1546 	case SIOCGLAGGFLAGS:
1547 		rf->rf_flags = 0;
1548 		LAGG_XLOCK(sc);
1549 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1550 			rf->rf_flags |= LAGG_F_HASHL2;
1551 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1552 			rf->rf_flags |= LAGG_F_HASHL3;
1553 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1554 			rf->rf_flags |= LAGG_F_HASHL4;
1555 		LAGG_XUNLOCK(sc);
1556 		break;
1557 	case SIOCSLAGGHASH:
1558 		error = priv_check(td, PRIV_NET_LAGG);
1559 		if (error)
1560 			break;
1561 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1562 			error = EINVAL;
1563 			break;
1564 		}
1565 		LAGG_XLOCK(sc);
1566 		sc->sc_flags = 0;
1567 		if (rf->rf_flags & LAGG_F_HASHL2)
1568 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1569 		if (rf->rf_flags & LAGG_F_HASHL3)
1570 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1571 		if (rf->rf_flags & LAGG_F_HASHL4)
1572 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1573 		LAGG_XUNLOCK(sc);
1574 		break;
1575 	case SIOCGLAGGPORT:
1576 		if (rp->rp_portname[0] == '\0' ||
1577 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1578 			error = EINVAL;
1579 			break;
1580 		}
1581 
1582 		NET_EPOCH_ENTER(et);
1583 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1584 		    lp->lp_softc != sc) {
1585 			error = ENOENT;
1586 			NET_EPOCH_EXIT(et);
1587 			if_rele(tpif);
1588 			break;
1589 		}
1590 
1591 		lagg_port2req(lp, rp);
1592 		NET_EPOCH_EXIT(et);
1593 		if_rele(tpif);
1594 		break;
1595 	case SIOCSLAGGPORT:
1596 		error = priv_check(td, PRIV_NET_LAGG);
1597 		if (error)
1598 			break;
1599 		if (rp->rp_portname[0] == '\0' ||
1600 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1601 			error = EINVAL;
1602 			break;
1603 		}
1604 #ifdef INET6
1605 		/*
1606 		 * A laggport interface should not have inet6 address
1607 		 * because two interfaces with a valid link-local
1608 		 * scope zone must not be merged in any form.  This
1609 		 * restriction is needed to prevent violation of
1610 		 * link-local scope zone.  Attempts to add a laggport
1611 		 * interface which has inet6 addresses triggers
1612 		 * removal of all inet6 addresses on the member
1613 		 * interface.
1614 		 */
1615 		if (in6ifa_llaonifp(tpif)) {
1616 			in6_ifdetach(tpif);
1617 				if_printf(sc->sc_ifp,
1618 				    "IPv6 addresses on %s have been removed "
1619 				    "before adding it as a member to prevent "
1620 				    "IPv6 address scope violation.\n",
1621 				    tpif->if_xname);
1622 		}
1623 #endif
1624 		oldmtu = ifp->if_mtu;
1625 		LAGG_XLOCK(sc);
1626 		error = lagg_port_create(sc, tpif);
1627 		LAGG_XUNLOCK(sc);
1628 		if_rele(tpif);
1629 
1630 		/*
1631 		 * LAGG MTU may change during addition of the first port.
1632 		 * If it did, do network layer specific procedure.
1633 		 */
1634 		if (ifp->if_mtu != oldmtu) {
1635 #ifdef INET6
1636 			nd6_setmtu(ifp);
1637 #endif
1638 			rt_updatemtu(ifp);
1639 		}
1640 
1641 		VLAN_CAPABILITIES(ifp);
1642 		break;
1643 	case SIOCSLAGGDELPORT:
1644 		error = priv_check(td, PRIV_NET_LAGG);
1645 		if (error)
1646 			break;
1647 		if (rp->rp_portname[0] == '\0' ||
1648 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1649 			error = EINVAL;
1650 			break;
1651 		}
1652 
1653 		LAGG_XLOCK(sc);
1654 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1655 		    lp->lp_softc != sc) {
1656 			error = ENOENT;
1657 			LAGG_XUNLOCK(sc);
1658 			if_rele(tpif);
1659 			break;
1660 		}
1661 
1662 		error = lagg_port_destroy(lp, 1);
1663 		LAGG_XUNLOCK(sc);
1664 		if_rele(tpif);
1665 		VLAN_CAPABILITIES(ifp);
1666 		break;
1667 	case SIOCSIFFLAGS:
1668 		/* Set flags on ports too */
1669 		LAGG_XLOCK(sc);
1670 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1671 			lagg_setflags(lp, 1);
1672 		}
1673 
1674 		if (!(ifp->if_flags & IFF_UP) &&
1675 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1676 			/*
1677 			 * If interface is marked down and it is running,
1678 			 * then stop and disable it.
1679 			 */
1680 			lagg_stop(sc);
1681 			LAGG_XUNLOCK(sc);
1682 		} else if ((ifp->if_flags & IFF_UP) &&
1683 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1684 			/*
1685 			 * If interface is marked up and it is stopped, then
1686 			 * start it.
1687 			 */
1688 			LAGG_XUNLOCK(sc);
1689 			(*ifp->if_init)(sc);
1690 		} else
1691 			LAGG_XUNLOCK(sc);
1692 		break;
1693 	case SIOCADDMULTI:
1694 	case SIOCDELMULTI:
1695 		LAGG_XLOCK(sc);
1696 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1697 			lagg_clrmulti(lp);
1698 			lagg_setmulti(lp);
1699 		}
1700 		LAGG_XUNLOCK(sc);
1701 		error = 0;
1702 		break;
1703 	case SIOCSIFMEDIA:
1704 	case SIOCGIFMEDIA:
1705 		if (ifp->if_type == IFT_INFINIBAND)
1706 			error = EINVAL;
1707 		else
1708 			error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1709 		break;
1710 
1711 	case SIOCSIFCAP:
1712 	case SIOCSIFCAPNV:
1713 		LAGG_XLOCK(sc);
1714 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1715 			if (lp->lp_ioctl != NULL)
1716 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1717 		}
1718 		lagg_capabilities(sc);
1719 		LAGG_XUNLOCK(sc);
1720 		VLAN_CAPABILITIES(ifp);
1721 		error = 0;
1722 		break;
1723 
1724 	case SIOCGIFCAPNV:
1725 		error = 0;
1726 		break;
1727 
1728 	case SIOCSIFMTU:
1729 		LAGG_XLOCK(sc);
1730 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1731 			if (lp->lp_ioctl != NULL)
1732 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1733 			else
1734 				error = EINVAL;
1735 			if (error != 0) {
1736 				if_printf(ifp,
1737 				    "failed to change MTU to %d on port %s, "
1738 				    "reverting all ports to original MTU (%d)\n",
1739 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1740 				break;
1741 			}
1742 		}
1743 		if (error == 0) {
1744 			ifp->if_mtu = ifr->ifr_mtu;
1745 		} else {
1746 			/* set every port back to the original MTU */
1747 			ifr->ifr_mtu = ifp->if_mtu;
1748 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1749 				if (lp->lp_ioctl != NULL)
1750 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1751 			}
1752 		}
1753 		lagg_capabilities(sc);
1754 		LAGG_XUNLOCK(sc);
1755 		VLAN_CAPABILITIES(ifp);
1756 		break;
1757 
1758 	default:
1759 		error = ether_ioctl(ifp, cmd, data);
1760 		break;
1761 	}
1762 	return (error);
1763 }
1764 
1765 #if defined(KERN_TLS) || defined(RATELIMIT)
1766 #ifdef RATELIMIT
1767 static const struct if_snd_tag_sw lagg_snd_tag_ul_sw = {
1768 	.snd_tag_modify = lagg_snd_tag_modify,
1769 	.snd_tag_query = lagg_snd_tag_query,
1770 	.snd_tag_free = lagg_snd_tag_free,
1771 	.next_snd_tag = lagg_next_snd_tag,
1772 	.type = IF_SND_TAG_TYPE_UNLIMITED
1773 };
1774 
1775 static const struct if_snd_tag_sw lagg_snd_tag_rl_sw = {
1776 	.snd_tag_modify = lagg_snd_tag_modify,
1777 	.snd_tag_query = lagg_snd_tag_query,
1778 	.snd_tag_free = lagg_snd_tag_free,
1779 	.next_snd_tag = lagg_next_snd_tag,
1780 	.type = IF_SND_TAG_TYPE_RATE_LIMIT
1781 };
1782 #endif
1783 
1784 #ifdef KERN_TLS
1785 static const struct if_snd_tag_sw lagg_snd_tag_tls_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
1791 };
1792 
1793 #ifdef RATELIMIT
1794 static const struct if_snd_tag_sw lagg_snd_tag_tls_rl_sw = {
1795 	.snd_tag_modify = lagg_snd_tag_modify,
1796 	.snd_tag_query = lagg_snd_tag_query,
1797 	.snd_tag_free = lagg_snd_tag_free,
1798 	.next_snd_tag = lagg_next_snd_tag,
1799 	.type = IF_SND_TAG_TYPE_TLS_RATE_LIMIT
1800 };
1801 #endif
1802 #endif
1803 
1804 static inline struct lagg_snd_tag *
1805 mst_to_lst(struct m_snd_tag *mst)
1806 {
1807 
1808 	return (__containerof(mst, struct lagg_snd_tag, com));
1809 }
1810 
1811 /*
1812  * Look up the port used by a specific flow.  This only works for lagg
1813  * protocols with deterministic port mappings (e.g. not roundrobin).
1814  * In addition protocols which use a hash to map flows to ports must
1815  * be configured to use the mbuf flowid rather than hashing packet
1816  * contents.
1817  */
1818 static struct lagg_port *
1819 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1820     uint8_t numa_domain)
1821 {
1822 	struct lagg_softc *sc;
1823 	struct lagg_port *lp;
1824 	struct lagg_lb *lb;
1825 	uint32_t hash, p;
1826 	int err;
1827 
1828 	sc = ifp->if_softc;
1829 
1830 	switch (sc->sc_proto) {
1831 	case LAGG_PROTO_FAILOVER:
1832 		return (lagg_link_active(sc, sc->sc_primary));
1833 	case LAGG_PROTO_LOADBALANCE:
1834 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1835 		    flowtype == M_HASHTYPE_NONE)
1836 			return (NULL);
1837 		p = flowid >> sc->flowid_shift;
1838 		p %= sc->sc_count;
1839 		lb = (struct lagg_lb *)sc->sc_psc;
1840 		lp = lb->lb_ports[p];
1841 		return (lagg_link_active(sc, lp));
1842 	case LAGG_PROTO_LACP:
1843 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1844 		    flowtype == M_HASHTYPE_NONE)
1845 			return (NULL);
1846 		hash = flowid >> sc->flowid_shift;
1847 		return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, &err));
1848 	default:
1849 		return (NULL);
1850 	}
1851 }
1852 
1853 static int
1854 lagg_snd_tag_alloc(struct ifnet *ifp,
1855     union if_snd_tag_alloc_params *params,
1856     struct m_snd_tag **ppmt)
1857 {
1858 	struct epoch_tracker et;
1859 	const struct if_snd_tag_sw *sw;
1860 	struct lagg_snd_tag *lst;
1861 	struct lagg_port *lp;
1862 	struct ifnet *lp_ifp;
1863 	struct m_snd_tag *mst;
1864 	int error;
1865 
1866 	switch (params->hdr.type) {
1867 #ifdef RATELIMIT
1868 	case IF_SND_TAG_TYPE_UNLIMITED:
1869 		sw = &lagg_snd_tag_ul_sw;
1870 		break;
1871 	case IF_SND_TAG_TYPE_RATE_LIMIT:
1872 		sw = &lagg_snd_tag_rl_sw;
1873 		break;
1874 #endif
1875 #ifdef KERN_TLS
1876 	case IF_SND_TAG_TYPE_TLS:
1877 		sw = &lagg_snd_tag_tls_sw;
1878 		break;
1879 	case IF_SND_TAG_TYPE_TLS_RX:
1880 		/* Return tag from port interface directly. */
1881 		sw = NULL;
1882 		break;
1883 #ifdef RATELIMIT
1884 	case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
1885 		sw = &lagg_snd_tag_tls_rl_sw;
1886 		break;
1887 #endif
1888 #endif
1889 	default:
1890 		return (EOPNOTSUPP);
1891 	}
1892 
1893 	NET_EPOCH_ENTER(et);
1894 	lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1895 	    params->hdr.flowtype, params->hdr.numa_domain);
1896 	if (lp == NULL) {
1897 		NET_EPOCH_EXIT(et);
1898 		return (EOPNOTSUPP);
1899 	}
1900 	if (lp->lp_ifp == NULL) {
1901 		NET_EPOCH_EXIT(et);
1902 		return (EOPNOTSUPP);
1903 	}
1904 	lp_ifp = lp->lp_ifp;
1905 	if_ref(lp_ifp);
1906 	NET_EPOCH_EXIT(et);
1907 
1908 	if (sw != NULL) {
1909 		lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1910 		if (lst == NULL) {
1911 			if_rele(lp_ifp);
1912 			return (ENOMEM);
1913 		}
1914 	} else
1915 		lst = NULL;
1916 
1917 	error = m_snd_tag_alloc(lp_ifp, params, &mst);
1918 	if_rele(lp_ifp);
1919 	if (error) {
1920 		free(lst, M_LAGG);
1921 		return (error);
1922 	}
1923 
1924 	if (sw != NULL) {
1925 		m_snd_tag_init(&lst->com, ifp, sw);
1926 		lst->tag = mst;
1927 
1928 		*ppmt = &lst->com;
1929 	} else
1930 		*ppmt = mst;
1931 
1932 	return (0);
1933 }
1934 
1935 static struct m_snd_tag *
1936 lagg_next_snd_tag(struct m_snd_tag *mst)
1937 {
1938 	struct lagg_snd_tag *lst;
1939 
1940 	lst = mst_to_lst(mst);
1941 	return (lst->tag);
1942 }
1943 
1944 static int
1945 lagg_snd_tag_modify(struct m_snd_tag *mst,
1946     union if_snd_tag_modify_params *params)
1947 {
1948 	struct lagg_snd_tag *lst;
1949 
1950 	lst = mst_to_lst(mst);
1951 	return (lst->tag->sw->snd_tag_modify(lst->tag, params));
1952 }
1953 
1954 static int
1955 lagg_snd_tag_query(struct m_snd_tag *mst,
1956     union if_snd_tag_query_params *params)
1957 {
1958 	struct lagg_snd_tag *lst;
1959 
1960 	lst = mst_to_lst(mst);
1961 	return (lst->tag->sw->snd_tag_query(lst->tag, params));
1962 }
1963 
1964 static void
1965 lagg_snd_tag_free(struct m_snd_tag *mst)
1966 {
1967 	struct lagg_snd_tag *lst;
1968 
1969 	lst = mst_to_lst(mst);
1970 	m_snd_tag_rele(lst->tag);
1971 	free(lst, M_LAGG);
1972 }
1973 
1974 static void
1975 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1976 {
1977 	/*
1978 	 * For lagg, we have an indirect
1979 	 * interface. The caller needs to
1980 	 * get a ratelimit tag on the actual
1981 	 * interface the flow will go on.
1982 	 */
1983 	q->rate_table = NULL;
1984 	q->flags = RT_IS_INDIRECT;
1985 	q->max_flows = 0;
1986 	q->number_of_rates = 0;
1987 }
1988 #endif
1989 
1990 static int
1991 lagg_setmulti(struct lagg_port *lp)
1992 {
1993 	struct lagg_softc *sc = lp->lp_softc;
1994 	struct ifnet *ifp = lp->lp_ifp;
1995 	struct ifnet *scifp = sc->sc_ifp;
1996 	struct lagg_mc *mc;
1997 	struct ifmultiaddr *ifma;
1998 	int error;
1999 
2000 	IF_ADDR_WLOCK(scifp);
2001 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
2002 		if (ifma->ifma_addr->sa_family != AF_LINK)
2003 			continue;
2004 		mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
2005 		if (mc == NULL) {
2006 			IF_ADDR_WUNLOCK(scifp);
2007 			return (ENOMEM);
2008 		}
2009 		bcopy(ifma->ifma_addr, &mc->mc_addr,
2010 		    ifma->ifma_addr->sa_len);
2011 		mc->mc_addr.sdl_index = ifp->if_index;
2012 		mc->mc_ifma = NULL;
2013 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
2014 	}
2015 	IF_ADDR_WUNLOCK(scifp);
2016 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
2017 		error = if_addmulti(ifp,
2018 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
2019 		if (error)
2020 			return (error);
2021 	}
2022 	return (0);
2023 }
2024 
2025 static int
2026 lagg_clrmulti(struct lagg_port *lp)
2027 {
2028 	struct lagg_mc *mc;
2029 
2030 	LAGG_XLOCK_ASSERT(lp->lp_softc);
2031 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
2032 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
2033 		if (mc->mc_ifma && lp->lp_detaching == 0)
2034 			if_delmulti_ifma(mc->mc_ifma);
2035 		free(mc, M_LAGG);
2036 	}
2037 	return (0);
2038 }
2039 
2040 static void
2041 lagg_setcaps(struct lagg_port *lp, int cap, int cap2)
2042 {
2043 	struct ifreq ifr;
2044 	struct siocsifcapnv_driver_data drv_ioctl_data;
2045 
2046 	if (lp->lp_ifp->if_capenable == cap &&
2047 	    lp->lp_ifp->if_capenable2 == cap2)
2048 		return;
2049 	if (lp->lp_ioctl == NULL)
2050 		return;
2051 	/* XXX */
2052 	if ((lp->lp_ifp->if_capabilities & IFCAP_NV) != 0) {
2053 		drv_ioctl_data.reqcap = cap;
2054 		drv_ioctl_data.reqcap2 = cap2;
2055 		drv_ioctl_data.nvcap = NULL;
2056 		(*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAPNV,
2057 		    (caddr_t)&drv_ioctl_data);
2058 	} else {
2059 		ifr.ifr_reqcap = cap;
2060 		(*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr);
2061 	}
2062 }
2063 
2064 /* Handle a ref counted flag that should be set on the lagg port as well */
2065 static int
2066 lagg_setflag(struct lagg_port *lp, int flag, int status,
2067     int (*func)(struct ifnet *, int))
2068 {
2069 	struct lagg_softc *sc = lp->lp_softc;
2070 	struct ifnet *scifp = sc->sc_ifp;
2071 	struct ifnet *ifp = lp->lp_ifp;
2072 	int error;
2073 
2074 	LAGG_XLOCK_ASSERT(sc);
2075 
2076 	status = status ? (scifp->if_flags & flag) : 0;
2077 	/* Now "status" contains the flag value or 0 */
2078 
2079 	/*
2080 	 * See if recorded ports status is different from what
2081 	 * we want it to be.  If it is, flip it.  We record ports
2082 	 * status in lp_ifflags so that we won't clear ports flag
2083 	 * we haven't set.  In fact, we don't clear or set ports
2084 	 * flags directly, but get or release references to them.
2085 	 * That's why we can be sure that recorded flags still are
2086 	 * in accord with actual ports flags.
2087 	 */
2088 	if (status != (lp->lp_ifflags & flag)) {
2089 		error = (*func)(ifp, status);
2090 		if (error)
2091 			return (error);
2092 		lp->lp_ifflags &= ~flag;
2093 		lp->lp_ifflags |= status;
2094 	}
2095 	return (0);
2096 }
2097 
2098 /*
2099  * Handle IFF_* flags that require certain changes on the lagg port
2100  * if "status" is true, update ports flags respective to the lagg
2101  * if "status" is false, forcedly clear the flags set on port.
2102  */
2103 static int
2104 lagg_setflags(struct lagg_port *lp, int status)
2105 {
2106 	int error, i;
2107 
2108 	for (i = 0; lagg_pflags[i].flag; i++) {
2109 		error = lagg_setflag(lp, lagg_pflags[i].flag,
2110 		    status, lagg_pflags[i].func);
2111 		if (error)
2112 			return (error);
2113 	}
2114 	return (0);
2115 }
2116 
2117 static int
2118 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2119 {
2120 	struct epoch_tracker et;
2121 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2122 	int error;
2123 
2124 #if defined(KERN_TLS) || defined(RATELIMIT)
2125 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2126 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2127 #endif
2128 	NET_EPOCH_ENTER(et);
2129 	/* We need a Tx algorithm and at least one port */
2130 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2131 		NET_EPOCH_EXIT(et);
2132 		m_freem(m);
2133 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2134 		return (ENXIO);
2135 	}
2136 
2137 	ETHER_BPF_MTAP(ifp, m);
2138 
2139 	error = lagg_proto_start(sc, m);
2140 	NET_EPOCH_EXIT(et);
2141 	return (error);
2142 }
2143 
2144 static int
2145 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2146 {
2147 	struct epoch_tracker et;
2148 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2149 	int error;
2150 
2151 #if defined(KERN_TLS) || defined(RATELIMIT)
2152 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2153 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2154 #endif
2155 	NET_EPOCH_ENTER(et);
2156 	/* We need a Tx algorithm and at least one port */
2157 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2158 		NET_EPOCH_EXIT(et);
2159 		m_freem(m);
2160 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2161 		return (ENXIO);
2162 	}
2163 
2164 	INFINIBAND_BPF_MTAP(ifp, m);
2165 
2166 	error = lagg_proto_start(sc, m);
2167 	NET_EPOCH_EXIT(et);
2168 	return (error);
2169 }
2170 
2171 /*
2172  * The ifp->if_qflush entry point for lagg(4) is no-op.
2173  */
2174 static void
2175 lagg_qflush(struct ifnet *ifp __unused)
2176 {
2177 }
2178 
2179 static struct mbuf *
2180 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2181 {
2182 	struct epoch_tracker et;
2183 	struct lagg_port *lp = ifp->if_lagg;
2184 	struct lagg_softc *sc = lp->lp_softc;
2185 	struct ifnet *scifp = sc->sc_ifp;
2186 
2187 	NET_EPOCH_ENTER(et);
2188 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2189 	    lp->lp_detaching != 0 ||
2190 	    sc->sc_proto == LAGG_PROTO_NONE) {
2191 		NET_EPOCH_EXIT(et);
2192 		m_freem(m);
2193 		return (NULL);
2194 	}
2195 
2196 	ETHER_BPF_MTAP(scifp, m);
2197 
2198 	m = lagg_proto_input(sc, lp, m);
2199 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2200 		m_freem(m);
2201 		m = NULL;
2202 	}
2203 
2204 	NET_EPOCH_EXIT(et);
2205 	return (m);
2206 }
2207 
2208 static struct mbuf *
2209 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2210 {
2211 	struct epoch_tracker et;
2212 	struct lagg_port *lp = ifp->if_lagg;
2213 	struct lagg_softc *sc = lp->lp_softc;
2214 	struct ifnet *scifp = sc->sc_ifp;
2215 
2216 	NET_EPOCH_ENTER(et);
2217 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2218 	    lp->lp_detaching != 0 ||
2219 	    sc->sc_proto == LAGG_PROTO_NONE) {
2220 		NET_EPOCH_EXIT(et);
2221 		m_freem(m);
2222 		return (NULL);
2223 	}
2224 
2225 	INFINIBAND_BPF_MTAP(scifp, m);
2226 
2227 	m = lagg_proto_input(sc, lp, m);
2228 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2229 		m_freem(m);
2230 		m = NULL;
2231 	}
2232 
2233 	NET_EPOCH_EXIT(et);
2234 	return (m);
2235 }
2236 
2237 static int
2238 lagg_media_change(struct ifnet *ifp)
2239 {
2240 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2241 
2242 	if (sc->sc_ifflags & IFF_DEBUG)
2243 		printf("%s\n", __func__);
2244 
2245 	/* Ignore */
2246 	return (0);
2247 }
2248 
2249 static void
2250 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2251 {
2252 	struct epoch_tracker et;
2253 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2254 	struct lagg_port *lp;
2255 
2256 	imr->ifm_status = IFM_AVALID;
2257 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
2258 
2259 	NET_EPOCH_ENTER(et);
2260 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2261 		if (LAGG_PORTACTIVE(lp))
2262 			imr->ifm_status |= IFM_ACTIVE;
2263 	}
2264 	NET_EPOCH_EXIT(et);
2265 }
2266 
2267 static void
2268 lagg_linkstate(struct lagg_softc *sc)
2269 {
2270 	struct epoch_tracker et;
2271 	struct lagg_port *lp;
2272 	int new_link = LINK_STATE_DOWN;
2273 	uint64_t speed;
2274 
2275 	LAGG_XLOCK_ASSERT(sc);
2276 
2277 	/* LACP handles link state itself */
2278 	if (sc->sc_proto == LAGG_PROTO_LACP)
2279 		return;
2280 
2281 	/* Our link is considered up if at least one of our ports is active */
2282 	NET_EPOCH_ENTER(et);
2283 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2284 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2285 			new_link = LINK_STATE_UP;
2286 			break;
2287 		}
2288 	}
2289 	NET_EPOCH_EXIT(et);
2290 	if_link_state_change(sc->sc_ifp, new_link);
2291 
2292 	/* Update if_baudrate to reflect the max possible speed */
2293 	switch (sc->sc_proto) {
2294 		case LAGG_PROTO_FAILOVER:
2295 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2296 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
2297 			break;
2298 		case LAGG_PROTO_ROUNDROBIN:
2299 		case LAGG_PROTO_LOADBALANCE:
2300 		case LAGG_PROTO_BROADCAST:
2301 			speed = 0;
2302 			NET_EPOCH_ENTER(et);
2303 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2304 				speed += lp->lp_ifp->if_baudrate;
2305 			NET_EPOCH_EXIT(et);
2306 			sc->sc_ifp->if_baudrate = speed;
2307 			break;
2308 		case LAGG_PROTO_LACP:
2309 			/* LACP updates if_baudrate itself */
2310 			break;
2311 	}
2312 }
2313 
2314 static void
2315 lagg_port_state(struct ifnet *ifp, int state)
2316 {
2317 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2318 	struct lagg_softc *sc = NULL;
2319 
2320 	if (lp != NULL)
2321 		sc = lp->lp_softc;
2322 	if (sc == NULL)
2323 		return;
2324 
2325 	LAGG_XLOCK(sc);
2326 	lagg_linkstate(sc);
2327 	lagg_proto_linkstate(sc, lp);
2328 	LAGG_XUNLOCK(sc);
2329 }
2330 
2331 struct lagg_port *
2332 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2333 {
2334 	struct lagg_port *lp_next, *rval = NULL;
2335 
2336 	/*
2337 	 * Search a port which reports an active link state.
2338 	 */
2339 
2340 #ifdef INVARIANTS
2341 	/*
2342 	 * This is called with either in the network epoch
2343 	 * or with LAGG_XLOCK(sc) held.
2344 	 */
2345 	if (!in_epoch(net_epoch_preempt))
2346 		LAGG_XLOCK_ASSERT(sc);
2347 #endif
2348 
2349 	if (lp == NULL)
2350 		goto search;
2351 	if (LAGG_PORTACTIVE(lp)) {
2352 		rval = lp;
2353 		goto found;
2354 	}
2355 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2356 	    LAGG_PORTACTIVE(lp_next)) {
2357 		rval = lp_next;
2358 		goto found;
2359 	}
2360 
2361 search:
2362 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2363 		if (LAGG_PORTACTIVE(lp_next)) {
2364 			return (lp_next);
2365 		}
2366 	}
2367 found:
2368 	return (rval);
2369 }
2370 
2371 int
2372 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2373 {
2374 
2375 #if defined(KERN_TLS) || defined(RATELIMIT)
2376 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2377 		struct lagg_snd_tag *lst;
2378 		struct m_snd_tag *mst;
2379 
2380 		mst = m->m_pkthdr.snd_tag;
2381 		lst = mst_to_lst(mst);
2382 		if (lst->tag->ifp != ifp) {
2383 			m_freem(m);
2384 			return (EAGAIN);
2385 		}
2386 		m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2387 		m_snd_tag_rele(mst);
2388 	}
2389 #endif
2390 	return (ifp->if_transmit)(ifp, m);
2391 }
2392 
2393 /*
2394  * Simple round robin aggregation
2395  */
2396 static void
2397 lagg_rr_attach(struct lagg_softc *sc)
2398 {
2399 	sc->sc_seq = 0;
2400 	sc->sc_stride = 1;
2401 }
2402 
2403 static int
2404 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2405 {
2406 	struct lagg_port *lp;
2407 	uint32_t p;
2408 
2409 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
2410 	p /= sc->sc_stride;
2411 	p %= sc->sc_count;
2412 	lp = CK_SLIST_FIRST(&sc->sc_ports);
2413 
2414 	while (p--)
2415 		lp = CK_SLIST_NEXT(lp, lp_entries);
2416 
2417 	/*
2418 	 * Check the port's link state. This will return the next active
2419 	 * port if the link is down or the port is NULL.
2420 	 */
2421 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2422 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2423 		m_freem(m);
2424 		return (ENETDOWN);
2425 	}
2426 
2427 	/* Send mbuf */
2428 	return (lagg_enqueue(lp->lp_ifp, m));
2429 }
2430 
2431 static struct mbuf *
2432 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2433 {
2434 	struct ifnet *ifp = sc->sc_ifp;
2435 
2436 	/* Just pass in the packet to our lagg device */
2437 	m->m_pkthdr.rcvif = ifp;
2438 
2439 	return (m);
2440 }
2441 
2442 /*
2443  * Broadcast mode
2444  */
2445 static int
2446 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2447 {
2448 	int errors = 0;
2449 	int ret;
2450 	struct lagg_port *lp, *last = NULL;
2451 	struct mbuf *m0;
2452 
2453 	NET_EPOCH_ASSERT();
2454 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2455 		if (!LAGG_PORTACTIVE(lp))
2456 			continue;
2457 
2458 		if (last != NULL) {
2459 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2460 			if (m0 == NULL) {
2461 				ret = ENOBUFS;
2462 				errors++;
2463 				break;
2464 			}
2465 			lagg_enqueue(last->lp_ifp, m0);
2466 		}
2467 		last = lp;
2468 	}
2469 
2470 	if (last == NULL) {
2471 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2472 		m_freem(m);
2473 		return (ENOENT);
2474 	}
2475 	if ((last = lagg_link_active(sc, last)) == NULL) {
2476 		errors++;
2477 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2478 		m_freem(m);
2479 		return (ENETDOWN);
2480 	}
2481 
2482 	ret = lagg_enqueue(last->lp_ifp, m);
2483 	if (errors != 0)
2484 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2485 
2486 	return (ret);
2487 }
2488 
2489 static struct mbuf*
2490 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2491 {
2492 	struct ifnet *ifp = sc->sc_ifp;
2493 
2494 	/* Just pass in the packet to our lagg device */
2495 	m->m_pkthdr.rcvif = ifp;
2496 	return (m);
2497 }
2498 
2499 /*
2500  * Active failover
2501  */
2502 static int
2503 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2504 {
2505 	struct lagg_port *lp;
2506 
2507 	/* Use the master port if active or the next available port */
2508 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2509 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2510 		m_freem(m);
2511 		return (ENETDOWN);
2512 	}
2513 
2514 	/* Send mbuf */
2515 	return (lagg_enqueue(lp->lp_ifp, m));
2516 }
2517 
2518 static struct mbuf *
2519 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2520 {
2521 	struct ifnet *ifp = sc->sc_ifp;
2522 	struct lagg_port *tmp_tp;
2523 
2524 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2525 		m->m_pkthdr.rcvif = ifp;
2526 		return (m);
2527 	}
2528 
2529 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2530 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2531 		/*
2532 		 * If tmp_tp is null, we've received a packet when all
2533 		 * our links are down. Weird, but process it anyways.
2534 		 */
2535 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2536 			m->m_pkthdr.rcvif = ifp;
2537 			return (m);
2538 		}
2539 	}
2540 
2541 	m_freem(m);
2542 	return (NULL);
2543 }
2544 
2545 /*
2546  * Loadbalancing
2547  */
2548 static void
2549 lagg_lb_attach(struct lagg_softc *sc)
2550 {
2551 	struct lagg_port *lp;
2552 	struct lagg_lb *lb;
2553 
2554 	LAGG_XLOCK_ASSERT(sc);
2555 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2556 	lb->lb_key = m_ether_tcpip_hash_init();
2557 	sc->sc_psc = lb;
2558 
2559 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2560 		lagg_lb_port_create(lp);
2561 }
2562 
2563 static void
2564 lagg_lb_detach(struct lagg_softc *sc)
2565 {
2566 	struct lagg_lb *lb;
2567 
2568 	lb = (struct lagg_lb *)sc->sc_psc;
2569 	if (lb != NULL)
2570 		free(lb, M_LAGG);
2571 }
2572 
2573 static int
2574 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2575 {
2576 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2577 	struct lagg_port *lp_next;
2578 	int i = 0, rv;
2579 
2580 	rv = 0;
2581 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2582 	LAGG_XLOCK_ASSERT(sc);
2583 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2584 		if (lp_next == lp)
2585 			continue;
2586 		if (i >= LAGG_MAX_PORTS) {
2587 			rv = EINVAL;
2588 			break;
2589 		}
2590 		if (sc->sc_ifflags & IFF_DEBUG)
2591 			printf("%s: port %s at index %d\n",
2592 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2593 		lb->lb_ports[i++] = lp_next;
2594 	}
2595 
2596 	return (rv);
2597 }
2598 
2599 static int
2600 lagg_lb_port_create(struct lagg_port *lp)
2601 {
2602 	struct lagg_softc *sc = lp->lp_softc;
2603 	return (lagg_lb_porttable(sc, NULL));
2604 }
2605 
2606 static void
2607 lagg_lb_port_destroy(struct lagg_port *lp)
2608 {
2609 	struct lagg_softc *sc = lp->lp_softc;
2610 	lagg_lb_porttable(sc, lp);
2611 }
2612 
2613 static int
2614 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2615 {
2616 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2617 	struct lagg_port *lp = NULL;
2618 	uint32_t p = 0;
2619 
2620 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2621 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2622 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2623 	else
2624 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2625 	p %= sc->sc_count;
2626 	lp = lb->lb_ports[p];
2627 
2628 	/*
2629 	 * Check the port's link state. This will return the next active
2630 	 * port if the link is down or the port is NULL.
2631 	 */
2632 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2633 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2634 		m_freem(m);
2635 		return (ENETDOWN);
2636 	}
2637 
2638 	/* Send mbuf */
2639 	return (lagg_enqueue(lp->lp_ifp, m));
2640 }
2641 
2642 static struct mbuf *
2643 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2644 {
2645 	struct ifnet *ifp = sc->sc_ifp;
2646 
2647 	/* Just pass in the packet to our lagg device */
2648 	m->m_pkthdr.rcvif = ifp;
2649 
2650 	return (m);
2651 }
2652 
2653 /*
2654  * 802.3ad LACP
2655  */
2656 static void
2657 lagg_lacp_attach(struct lagg_softc *sc)
2658 {
2659 	struct lagg_port *lp;
2660 
2661 	lacp_attach(sc);
2662 	LAGG_XLOCK_ASSERT(sc);
2663 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2664 		lacp_port_create(lp);
2665 }
2666 
2667 static void
2668 lagg_lacp_detach(struct lagg_softc *sc)
2669 {
2670 	struct lagg_port *lp;
2671 	void *psc;
2672 
2673 	LAGG_XLOCK_ASSERT(sc);
2674 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2675 		lacp_port_destroy(lp);
2676 
2677 	psc = sc->sc_psc;
2678 	sc->sc_psc = NULL;
2679 	lacp_detach(psc);
2680 }
2681 
2682 static void
2683 lagg_lacp_lladdr(struct lagg_softc *sc)
2684 {
2685 	struct lagg_port *lp;
2686 
2687 	LAGG_SXLOCK_ASSERT(sc);
2688 
2689 	/* purge all the lacp ports */
2690 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2691 		lacp_port_destroy(lp);
2692 
2693 	/* add them back in */
2694 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2695 		lacp_port_create(lp);
2696 }
2697 
2698 static int
2699 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2700 {
2701 	struct lagg_port *lp;
2702 	int err;
2703 
2704 	lp = lacp_select_tx_port(sc, m, &err);
2705 	if (lp == NULL) {
2706 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2707 		m_freem(m);
2708 		return (err);
2709 	}
2710 
2711 	/* Send mbuf */
2712 	return (lagg_enqueue(lp->lp_ifp, m));
2713 }
2714 
2715 static struct mbuf *
2716 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2717 {
2718 	struct ifnet *ifp = sc->sc_ifp;
2719 	struct ether_header *eh;
2720 	u_short etype;
2721 
2722 	eh = mtod(m, struct ether_header *);
2723 	etype = ntohs(eh->ether_type);
2724 
2725 	/* Tap off LACP control messages */
2726 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2727 		m = lacp_input(lp, m);
2728 		if (m == NULL)
2729 			return (NULL);
2730 	}
2731 
2732 	/*
2733 	 * If the port is not collecting or not in the active aggregator then
2734 	 * free and return.
2735 	 */
2736 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2737 		m_freem(m);
2738 		return (NULL);
2739 	}
2740 
2741 	m->m_pkthdr.rcvif = ifp;
2742 	return (m);
2743 }
2744