xref: /freebsd/sys/net/if_lagg.c (revision b197d4b893974c9eb4d7b38704c6d5c486235d6f)
1 /*	$OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $	*/
2 
3 /*
4  * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org>
5  * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org>
6  * Copyright (c) 2014, 2016 Marcelo Araujo <araujo@FreeBSD.org>
7  *
8  * Permission to use, copy, modify, and distribute this software for any
9  * purpose with or without fee is hereby granted, provided that the above
10  * copyright notice and this permission notice appear in all copies.
11  *
12  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19  */
20 
21 #include <sys/cdefs.h>
22 __FBSDID("$FreeBSD$");
23 
24 #include "opt_inet.h"
25 #include "opt_inet6.h"
26 #include "opt_kern_tls.h"
27 #include "opt_ratelimit.h"
28 
29 #include <sys/param.h>
30 #include <sys/kernel.h>
31 #include <sys/malloc.h>
32 #include <sys/mbuf.h>
33 #include <sys/queue.h>
34 #include <sys/socket.h>
35 #include <sys/sockio.h>
36 #include <sys/sysctl.h>
37 #include <sys/module.h>
38 #include <sys/priv.h>
39 #include <sys/systm.h>
40 #include <sys/proc.h>
41 #include <sys/lock.h>
42 #include <sys/rmlock.h>
43 #include <sys/sx.h>
44 #include <sys/taskqueue.h>
45 #include <sys/eventhandler.h>
46 
47 #include <net/ethernet.h>
48 #include <net/if.h>
49 #include <net/if_clone.h>
50 #include <net/if_arp.h>
51 #include <net/if_dl.h>
52 #include <net/if_media.h>
53 #include <net/if_types.h>
54 #include <net/if_var.h>
55 #include <net/bpf.h>
56 #include <net/route.h>
57 #include <net/vnet.h>
58 #include <net/infiniband.h>
59 
60 #if defined(INET) || defined(INET6)
61 #include <netinet/in.h>
62 #include <netinet/ip.h>
63 #endif
64 #ifdef INET
65 #include <netinet/in_systm.h>
66 #include <netinet/if_ether.h>
67 #endif
68 
69 #ifdef INET6
70 #include <netinet/ip6.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet6/in6_ifattach.h>
73 #endif
74 
75 #include <net/if_vlan_var.h>
76 #include <net/if_lagg.h>
77 #include <net/ieee8023ad_lacp.h>
78 
79 #ifdef INET6
80 /*
81  * XXX: declare here to avoid to include many inet6 related files..
82  * should be more generalized?
83  */
84 extern void	nd6_setmtu(struct ifnet *);
85 #endif
86 
87 #define	LAGG_SX_INIT(_sc)	sx_init(&(_sc)->sc_sx, "if_lagg sx")
88 #define	LAGG_SX_DESTROY(_sc)	sx_destroy(&(_sc)->sc_sx)
89 #define	LAGG_XLOCK(_sc)		sx_xlock(&(_sc)->sc_sx)
90 #define	LAGG_XUNLOCK(_sc)	sx_xunlock(&(_sc)->sc_sx)
91 #define	LAGG_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
92 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
93 
94 /* Special flags we should propagate to the lagg ports. */
95 static struct {
96 	int flag;
97 	int (*func)(struct ifnet *, int);
98 } lagg_pflags[] = {
99 	{IFF_PROMISC, ifpromisc},
100 	{IFF_ALLMULTI, if_allmulti},
101 	{0, NULL}
102 };
103 
104 struct lagg_snd_tag {
105 	struct m_snd_tag com;
106 	struct m_snd_tag *tag;
107 };
108 
109 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
110 #define	V_lagg_list	VNET(lagg_list)
111 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
112 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
113 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
114 					"if_lagg list", NULL, MTX_DEF)
115 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
116 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
117 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
118 eventhandler_tag	lagg_detach_cookie = NULL;
119 
120 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
121 static void	lagg_clone_destroy(struct ifnet *);
122 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
123 #define	V_lagg_cloner	VNET(lagg_cloner)
124 static const char laggname[] = "lagg";
125 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
126 
127 static void	lagg_capabilities(struct lagg_softc *);
128 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
129 static int	lagg_port_destroy(struct lagg_port *, int);
130 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *);
131 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *);
132 static void	lagg_linkstate(struct lagg_softc *);
133 static void	lagg_port_state(struct ifnet *, int);
134 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
135 static int	lagg_port_output(struct ifnet *, struct mbuf *,
136 		    const struct sockaddr *, struct route *);
137 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
138 #ifdef LAGG_PORT_STACKING
139 static int	lagg_port_checkstacking(struct lagg_softc *);
140 #endif
141 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
142 static void	lagg_init(void *);
143 static void	lagg_stop(struct lagg_softc *);
144 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
145 #if defined(KERN_TLS) || defined(RATELIMIT)
146 static int	lagg_snd_tag_alloc(struct ifnet *,
147 		    union if_snd_tag_alloc_params *,
148 		    struct m_snd_tag **);
149 static int	lagg_snd_tag_modify(struct m_snd_tag *,
150 		    union if_snd_tag_modify_params *);
151 static int	lagg_snd_tag_query(struct m_snd_tag *,
152 		    union if_snd_tag_query_params *);
153 static void	lagg_snd_tag_free(struct m_snd_tag *);
154 static struct m_snd_tag *lagg_next_snd_tag(struct m_snd_tag *);
155 static void     lagg_ratelimit_query(struct ifnet *,
156 		    struct if_ratelimit_query_results *);
157 #endif
158 static int	lagg_setmulti(struct lagg_port *);
159 static int	lagg_clrmulti(struct lagg_port *);
160 static	void	lagg_setcaps(struct lagg_port *, int cap, int cap2);
161 static	int	lagg_setflag(struct lagg_port *, int, int,
162 		    int (*func)(struct ifnet *, int));
163 static	int	lagg_setflags(struct lagg_port *, int status);
164 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
165 static int	lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
166 static int	lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
167 static void	lagg_qflush(struct ifnet *);
168 static int	lagg_media_change(struct ifnet *);
169 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
170 static struct lagg_port *lagg_link_active(struct lagg_softc *,
171 	    struct lagg_port *);
172 
173 /* Simple round robin */
174 static void	lagg_rr_attach(struct lagg_softc *);
175 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
176 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
177 		    struct mbuf *);
178 
179 /* Active failover */
180 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
181 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
182 		    struct mbuf *);
183 
184 /* Loadbalancing */
185 static void	lagg_lb_attach(struct lagg_softc *);
186 static void	lagg_lb_detach(struct lagg_softc *);
187 static int	lagg_lb_port_create(struct lagg_port *);
188 static void	lagg_lb_port_destroy(struct lagg_port *);
189 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
190 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
191 		    struct mbuf *);
192 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
193 
194 /* Broadcast */
195 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
196 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
197 		    struct mbuf *);
198 
199 /* 802.3ad LACP */
200 static void	lagg_lacp_attach(struct lagg_softc *);
201 static void	lagg_lacp_detach(struct lagg_softc *);
202 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
203 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
204 		    struct mbuf *);
205 static void	lagg_lacp_lladdr(struct lagg_softc *);
206 
207 /* lagg protocol table */
208 static const struct lagg_proto {
209 	lagg_proto	pr_num;
210 	void		(*pr_attach)(struct lagg_softc *);
211 	void		(*pr_detach)(struct lagg_softc *);
212 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
213 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
214 			    struct mbuf *);
215 	int		(*pr_addport)(struct lagg_port *);
216 	void		(*pr_delport)(struct lagg_port *);
217 	void		(*pr_linkstate)(struct lagg_port *);
218 	void 		(*pr_init)(struct lagg_softc *);
219 	void 		(*pr_stop)(struct lagg_softc *);
220 	void 		(*pr_lladdr)(struct lagg_softc *);
221 	void		(*pr_request)(struct lagg_softc *, void *);
222 	void		(*pr_portreq)(struct lagg_port *, void *);
223 } lagg_protos[] = {
224     {
225 	.pr_num = LAGG_PROTO_NONE
226     },
227     {
228 	.pr_num = LAGG_PROTO_ROUNDROBIN,
229 	.pr_attach = lagg_rr_attach,
230 	.pr_start = lagg_rr_start,
231 	.pr_input = lagg_rr_input,
232     },
233     {
234 	.pr_num = LAGG_PROTO_FAILOVER,
235 	.pr_start = lagg_fail_start,
236 	.pr_input = lagg_fail_input,
237     },
238     {
239 	.pr_num = LAGG_PROTO_LOADBALANCE,
240 	.pr_attach = lagg_lb_attach,
241 	.pr_detach = lagg_lb_detach,
242 	.pr_start = lagg_lb_start,
243 	.pr_input = lagg_lb_input,
244 	.pr_addport = lagg_lb_port_create,
245 	.pr_delport = lagg_lb_port_destroy,
246     },
247     {
248 	.pr_num = LAGG_PROTO_LACP,
249 	.pr_attach = lagg_lacp_attach,
250 	.pr_detach = lagg_lacp_detach,
251 	.pr_start = lagg_lacp_start,
252 	.pr_input = lagg_lacp_input,
253 	.pr_addport = lacp_port_create,
254 	.pr_delport = lacp_port_destroy,
255 	.pr_linkstate = lacp_linkstate,
256 	.pr_init = lacp_init,
257 	.pr_stop = lacp_stop,
258 	.pr_lladdr = lagg_lacp_lladdr,
259 	.pr_request = lacp_req,
260 	.pr_portreq = lacp_portreq,
261     },
262     {
263 	.pr_num = LAGG_PROTO_BROADCAST,
264 	.pr_start = lagg_bcast_start,
265 	.pr_input = lagg_bcast_input,
266     },
267 };
268 
269 SYSCTL_DECL(_net_link);
270 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
271     "Link Aggregation");
272 
273 /* Allow input on any failover links */
274 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
275 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
276 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
277     &VNET_NAME(lagg_failover_rx_all), 0,
278     "Accept input from any interface in a failover lagg");
279 
280 /* Default value for using flowid */
281 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
282 #define	V_def_use_flowid	VNET(def_use_flowid)
283 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
284     &VNET_NAME(def_use_flowid), 0,
285     "Default setting for using flow id for load sharing");
286 
287 /* Default value for using numa */
288 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
289 #define	V_def_use_numa	VNET(def_use_numa)
290 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN,
291     &VNET_NAME(def_use_numa), 0,
292     "Use numa to steer flows");
293 
294 /* Default value for flowid shift */
295 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
296 #define	V_def_flowid_shift	VNET(def_flowid_shift)
297 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
298     &VNET_NAME(def_flowid_shift), 0,
299     "Default setting for flowid shift for load sharing");
300 
301 static void
302 vnet_lagg_init(const void *unused __unused)
303 {
304 
305 	LAGG_LIST_LOCK_INIT();
306 	SLIST_INIT(&V_lagg_list);
307 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
308 	    lagg_clone_destroy, 0);
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 	if_clone_detach(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 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 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, int unit, caddr_t params)
508 {
509 	struct iflaggparam iflp;
510 	struct lagg_softc *sc;
511 	struct ifnet *ifp;
512 	int if_type;
513 	int error;
514 	static const uint8_t eaddr[LAGG_ADDR_LEN];
515 
516 	if (params != NULL) {
517 		error = copyin(params, &iflp, sizeof(iflp));
518 		if (error)
519 			return (error);
520 
521 		switch (iflp.lagg_type) {
522 		case LAGG_TYPE_ETHERNET:
523 			if_type = IFT_ETHER;
524 			break;
525 		case LAGG_TYPE_INFINIBAND:
526 			if_type = IFT_INFINIBAND;
527 			break;
528 		default:
529 			return (EINVAL);
530 		}
531 	} else {
532 		if_type = IFT_ETHER;
533 	}
534 
535 	sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO);
536 	ifp = sc->sc_ifp = if_alloc(if_type);
537 	if (ifp == NULL) {
538 		free(sc, M_LAGG);
539 		return (ENOSPC);
540 	}
541 	LAGG_SX_INIT(sc);
542 
543 	mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
544 	callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
545 
546 	LAGG_XLOCK(sc);
547 	if (V_def_use_flowid)
548 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
549 	if (V_def_use_numa)
550 		sc->sc_opts |= LAGG_OPT_USE_NUMA;
551 	sc->flowid_shift = V_def_flowid_shift;
552 
553 	/* Hash all layers by default */
554 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
555 
556 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
557 
558 	CK_SLIST_INIT(&sc->sc_ports);
559 
560 	switch (if_type) {
561 	case IFT_ETHER:
562 		/* Initialise pseudo media types */
563 		ifmedia_init(&sc->sc_media, 0, lagg_media_change,
564 		    lagg_media_status);
565 		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
566 		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
567 
568 		if_initname(ifp, laggname, unit);
569 		ifp->if_transmit = lagg_transmit_ethernet;
570 		break;
571 	case IFT_INFINIBAND:
572 		if_initname(ifp, laggname, unit);
573 		ifp->if_transmit = lagg_transmit_infiniband;
574 		break;
575 	default:
576 		break;
577 	}
578 	ifp->if_softc = sc;
579 	ifp->if_qflush = lagg_qflush;
580 	ifp->if_init = lagg_init;
581 	ifp->if_ioctl = lagg_ioctl;
582 	ifp->if_get_counter = lagg_get_counter;
583 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
584 #if defined(KERN_TLS) || defined(RATELIMIT)
585 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
586 	ifp->if_ratelimit_query = lagg_ratelimit_query;
587 #endif
588 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
589 
590 	/*
591 	 * Attach as an ordinary ethernet device, children will be attached
592 	 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
593 	 */
594 	switch (if_type) {
595 	case IFT_ETHER:
596 		ether_ifattach(ifp, eaddr);
597 		break;
598 	case IFT_INFINIBAND:
599 		infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr);
600 		break;
601 	default:
602 		break;
603 	}
604 
605 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
606 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
607 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
608 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
609 
610 	/* Insert into the global list of laggs */
611 	LAGG_LIST_LOCK();
612 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
613 	LAGG_LIST_UNLOCK();
614 	LAGG_XUNLOCK(sc);
615 
616 	return (0);
617 }
618 
619 static void
620 lagg_clone_destroy(struct ifnet *ifp)
621 {
622 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
623 	struct lagg_port *lp;
624 
625 	LAGG_XLOCK(sc);
626 	sc->sc_destroying = 1;
627 	lagg_stop(sc);
628 	ifp->if_flags &= ~IFF_UP;
629 
630 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
631 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
632 
633 	/* Shutdown and remove lagg ports */
634 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
635 		lagg_port_destroy(lp, 1);
636 
637 	/* Unhook the aggregation protocol */
638 	lagg_proto_detach(sc);
639 	LAGG_XUNLOCK(sc);
640 
641 	switch (ifp->if_type) {
642 	case IFT_ETHER:
643 		ifmedia_removeall(&sc->sc_media);
644 		ether_ifdetach(ifp);
645 		break;
646 	case IFT_INFINIBAND:
647 		infiniband_ifdetach(ifp);
648 		break;
649 	default:
650 		break;
651 	}
652 	if_free(ifp);
653 
654 	LAGG_LIST_LOCK();
655 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
656 	LAGG_LIST_UNLOCK();
657 
658 	mtx_destroy(&sc->sc_mtx);
659 	LAGG_SX_DESTROY(sc);
660 	free(sc, M_LAGG);
661 }
662 
663 static void
664 lagg_capabilities(struct lagg_softc *sc)
665 {
666 	struct lagg_port *lp;
667 	int cap, cap2, ena, ena2, pena, pena2;
668 	uint64_t hwa;
669 	struct ifnet_hw_tsomax hw_tsomax;
670 
671 	LAGG_XLOCK_ASSERT(sc);
672 
673 	/* Get common enabled capabilities for the lagg ports */
674 	ena = ena2 = ~0;
675 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
676 		ena &= lp->lp_ifp->if_capenable;
677 		ena2 &= lp->lp_ifp->if_capenable2;
678 	}
679 	if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
680 		ena = ena2 = 0;
681 
682 	/*
683 	 * Apply common enabled capabilities back to the lagg ports.
684 	 * May require several iterations if they are dependent.
685 	 */
686 	do {
687 		pena = ena;
688 		pena2 = ena2;
689 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
690 			lagg_setcaps(lp, ena, ena2);
691 			ena &= lp->lp_ifp->if_capenable;
692 			ena2 &= lp->lp_ifp->if_capenable2;
693 		}
694 	} while (pena != ena || pena2 != ena2);
695 
696 	/* Get other capabilities from the lagg ports */
697 	cap = cap2 = ~0;
698 	hwa = ~(uint64_t)0;
699 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
700 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
701 		cap &= lp->lp_ifp->if_capabilities;
702 		cap2 &= lp->lp_ifp->if_capabilities2;
703 		hwa &= lp->lp_ifp->if_hwassist;
704 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
705 	}
706 	if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
707 		cap = cap2 = hwa = 0;
708 
709 	if (sc->sc_ifp->if_capabilities != cap ||
710 	    sc->sc_ifp->if_capenable != ena ||
711 	    sc->sc_ifp->if_capenable2 != ena2 ||
712 	    sc->sc_ifp->if_hwassist != hwa ||
713 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
714 		sc->sc_ifp->if_capabilities = cap;
715 		sc->sc_ifp->if_capabilities2 = cap2;
716 		sc->sc_ifp->if_capenable = ena;
717 		sc->sc_ifp->if_capenable2 = ena2;
718 		sc->sc_ifp->if_hwassist = hwa;
719 		getmicrotime(&sc->sc_ifp->if_lastchange);
720 
721 		if (sc->sc_ifflags & IFF_DEBUG)
722 			if_printf(sc->sc_ifp,
723 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
724 	}
725 }
726 
727 static int
728 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
729 {
730 	struct lagg_softc *sc_ptr;
731 	struct lagg_port *lp, *tlp;
732 	struct ifreq ifr;
733 	int error, i, oldmtu;
734 	int if_type;
735 	uint64_t *pval;
736 
737 	LAGG_XLOCK_ASSERT(sc);
738 
739 	if (sc->sc_ifp == ifp) {
740 		if_printf(sc->sc_ifp,
741 		    "cannot add a lagg to itself as a port\n");
742 		return (EINVAL);
743 	}
744 
745 	if (sc->sc_destroying == 1)
746 		return (ENXIO);
747 
748 	/* Limit the maximal number of lagg ports */
749 	if (sc->sc_count >= LAGG_MAX_PORTS)
750 		return (ENOSPC);
751 
752 	/* Check if port has already been associated to a lagg */
753 	if (ifp->if_lagg != NULL) {
754 		/* Port is already in the current lagg? */
755 		lp = (struct lagg_port *)ifp->if_lagg;
756 		if (lp->lp_softc == sc)
757 			return (EEXIST);
758 		return (EBUSY);
759 	}
760 
761 	switch (sc->sc_ifp->if_type) {
762 	case IFT_ETHER:
763 		/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
764 		if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
765 			return (EPROTONOSUPPORT);
766 		if_type = IFT_IEEE8023ADLAG;
767 		break;
768 	case IFT_INFINIBAND:
769 		/* XXX Disallow non-infiniband interfaces */
770 		if (ifp->if_type != IFT_INFINIBAND)
771 			return (EPROTONOSUPPORT);
772 		if_type = IFT_INFINIBANDLAG;
773 		break;
774 	default:
775 		break;
776 	}
777 
778 	/* Allow the first Ethernet member to define the MTU */
779 	oldmtu = -1;
780 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
781 		sc->sc_ifp->if_mtu = ifp->if_mtu;
782 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
783 		if (ifp->if_ioctl == NULL) {
784 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
785 			    ifp->if_xname);
786 			return (EINVAL);
787 		}
788 		oldmtu = ifp->if_mtu;
789 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
790 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
791 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
792 		if (error != 0) {
793 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
794 			    ifp->if_xname);
795 			return (error);
796 		}
797 		ifr.ifr_mtu = oldmtu;
798 	}
799 
800 	lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO);
801 	lp->lp_softc = sc;
802 
803 	/* Check if port is a stacked lagg */
804 	LAGG_LIST_LOCK();
805 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
806 		if (ifp == sc_ptr->sc_ifp) {
807 			LAGG_LIST_UNLOCK();
808 			free(lp, M_LAGG);
809 			if (oldmtu != -1)
810 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
811 				    (caddr_t)&ifr);
812 			return (EINVAL);
813 			/* XXX disable stacking for the moment, its untested */
814 #ifdef LAGG_PORT_STACKING
815 			lp->lp_flags |= LAGG_PORT_STACK;
816 			if (lagg_port_checkstacking(sc_ptr) >=
817 			    LAGG_MAX_STACKING) {
818 				LAGG_LIST_UNLOCK();
819 				free(lp, M_LAGG);
820 				if (oldmtu != -1)
821 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
822 					    (caddr_t)&ifr);
823 				return (E2BIG);
824 			}
825 #endif
826 		}
827 	}
828 	LAGG_LIST_UNLOCK();
829 
830 	if_ref(ifp);
831 	lp->lp_ifp = ifp;
832 
833 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
834 	lp->lp_ifcapenable = ifp->if_capenable;
835 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
836 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
837 		lagg_proto_lladdr(sc);
838 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
839 	} else {
840 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
841 	}
842 	lagg_setflags(lp, 1);
843 
844 	if (CK_SLIST_EMPTY(&sc->sc_ports))
845 		sc->sc_primary = lp;
846 
847 	/* Change the interface type */
848 	lp->lp_iftype = ifp->if_type;
849 	ifp->if_type = if_type;
850 	ifp->if_lagg = lp;
851 	lp->lp_ioctl = ifp->if_ioctl;
852 	ifp->if_ioctl = lagg_port_ioctl;
853 	lp->lp_output = ifp->if_output;
854 	ifp->if_output = lagg_port_output;
855 
856 	/* Read port counters */
857 	pval = lp->port_counters.val;
858 	for (i = 0; i < IFCOUNTERS; i++, pval++)
859 		*pval = ifp->if_get_counter(ifp, i);
860 
861 	/*
862 	 * Insert into the list of ports.
863 	 * Keep ports sorted by if_index. It is handy, when configuration
864 	 * is predictable and `ifconfig laggN create ...` command
865 	 * will lead to the same result each time.
866 	 */
867 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
868 		if (tlp->lp_ifp->if_index < ifp->if_index && (
869 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
870 		    ((struct  lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
871 		    ifp->if_index))
872 			break;
873 	}
874 	if (tlp != NULL)
875 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
876 	else
877 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
878 	sc->sc_count++;
879 
880 	lagg_setmulti(lp);
881 
882 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
883 		/* Remove the port, without calling pr_delport. */
884 		lagg_port_destroy(lp, 0);
885 		if (oldmtu != -1)
886 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
887 		return (error);
888 	}
889 
890 	/* Update lagg capabilities */
891 	lagg_capabilities(sc);
892 	lagg_linkstate(sc);
893 
894 	return (0);
895 }
896 
897 #ifdef LAGG_PORT_STACKING
898 static int
899 lagg_port_checkstacking(struct lagg_softc *sc)
900 {
901 	struct lagg_softc *sc_ptr;
902 	struct lagg_port *lp;
903 	int m = 0;
904 
905 	LAGG_SXLOCK_ASSERT(sc);
906 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
907 		if (lp->lp_flags & LAGG_PORT_STACK) {
908 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
909 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
910 		}
911 	}
912 
913 	return (m + 1);
914 }
915 #endif
916 
917 static void
918 lagg_port_destroy_cb(epoch_context_t ec)
919 {
920 	struct lagg_port *lp;
921 	struct ifnet *ifp;
922 
923 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
924 	ifp = lp->lp_ifp;
925 
926 	if_rele(ifp);
927 	free(lp, M_LAGG);
928 }
929 
930 static int
931 lagg_port_destroy(struct lagg_port *lp, int rundelport)
932 {
933 	struct lagg_softc *sc = lp->lp_softc;
934 	struct lagg_port *lp_ptr, *lp0;
935 	struct ifnet *ifp = lp->lp_ifp;
936 	uint64_t *pval, vdiff;
937 	int i;
938 
939 	LAGG_XLOCK_ASSERT(sc);
940 
941 	if (rundelport)
942 		lagg_proto_delport(sc, lp);
943 
944 	if (lp->lp_detaching == 0)
945 		lagg_clrmulti(lp);
946 
947 	/* Restore interface */
948 	ifp->if_type = lp->lp_iftype;
949 	ifp->if_ioctl = lp->lp_ioctl;
950 	ifp->if_output = lp->lp_output;
951 	ifp->if_lagg = NULL;
952 
953 	/* Update detached port counters */
954 	pval = lp->port_counters.val;
955 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
956 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
957 		sc->detached_counters.val[i] += vdiff;
958 	}
959 
960 	/* Finally, remove the port from the lagg */
961 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
962 	sc->sc_count--;
963 
964 	/* Update the primary interface */
965 	if (lp == sc->sc_primary) {
966 		uint8_t lladdr[LAGG_ADDR_LEN];
967 
968 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
969 			bzero(&lladdr, LAGG_ADDR_LEN);
970 		else
971 			bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
972 		sc->sc_primary = lp0;
973 		if (sc->sc_destroying == 0) {
974 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
975 			lagg_proto_lladdr(sc);
976 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
977 
978 			/*
979 			 * Update lladdr for each port (new primary needs update
980 			 * as well, to switch from old lladdr to its 'real' one).
981 			 * We can skip this if the lagg is being destroyed.
982 			 */
983 			CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
984 				if_setlladdr(lp_ptr->lp_ifp, lladdr,
985 				    lp_ptr->lp_ifp->if_addrlen);
986 		}
987 	}
988 
989 	if (lp->lp_ifflags)
990 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
991 
992 	if (lp->lp_detaching == 0) {
993 		lagg_setflags(lp, 0);
994 		lagg_setcaps(lp, lp->lp_ifcapenable, lp->lp_ifcapenable2);
995 		if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
996 	}
997 
998 	/*
999 	 * free port and release it's ifnet reference after a grace period has
1000 	 * elapsed.
1001 	 */
1002 	NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
1003 	/* Update lagg capabilities */
1004 	lagg_capabilities(sc);
1005 	lagg_linkstate(sc);
1006 
1007 	return (0);
1008 }
1009 
1010 static int
1011 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1012 {
1013 	struct epoch_tracker et;
1014 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
1015 	struct lagg_softc *sc;
1016 	struct lagg_port *lp = NULL;
1017 	int error = 0;
1018 
1019 	/* Should be checked by the caller */
1020 	switch (ifp->if_type) {
1021 	case IFT_IEEE8023ADLAG:
1022 	case IFT_INFINIBANDLAG:
1023 		if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1024 			goto fallback;
1025 		break;
1026 	default:
1027 		goto fallback;
1028 	}
1029 
1030 	switch (cmd) {
1031 	case SIOCGLAGGPORT:
1032 		if (rp->rp_portname[0] == '\0' ||
1033 		    ifunit(rp->rp_portname) != ifp) {
1034 			error = EINVAL;
1035 			break;
1036 		}
1037 
1038 		NET_EPOCH_ENTER(et);
1039 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1040 			error = ENOENT;
1041 			NET_EPOCH_EXIT(et);
1042 			break;
1043 		}
1044 
1045 		lagg_port2req(lp, rp);
1046 		NET_EPOCH_EXIT(et);
1047 		break;
1048 
1049 	case SIOCSIFCAP:
1050 	case SIOCSIFCAPNV:
1051 		if (lp->lp_ioctl == NULL) {
1052 			error = EINVAL;
1053 			break;
1054 		}
1055 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1056 		if (error)
1057 			break;
1058 
1059 		/* Update lagg interface capabilities */
1060 		LAGG_XLOCK(sc);
1061 		lagg_capabilities(sc);
1062 		LAGG_XUNLOCK(sc);
1063 		VLAN_CAPABILITIES(sc->sc_ifp);
1064 		break;
1065 
1066 	case SIOCSIFMTU:
1067 		/* Do not allow the MTU to be changed once joined */
1068 		error = EINVAL;
1069 		break;
1070 
1071 	default:
1072 		goto fallback;
1073 	}
1074 
1075 	return (error);
1076 
1077 fallback:
1078 	if (lp != NULL && lp->lp_ioctl != NULL)
1079 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1080 
1081 	return (EINVAL);
1082 }
1083 
1084 /*
1085  * Requests counter @cnt data.
1086  *
1087  * Counter value is calculated the following way:
1088  * 1) for each port, sum  difference between current and "initial" measurements.
1089  * 2) add lagg logical interface counters.
1090  * 3) add data from detached_counters array.
1091  *
1092  * We also do the following things on ports attach/detach:
1093  * 1) On port attach we store all counters it has into port_counter array.
1094  * 2) On port detach we add the different between "initial" and
1095  *   current counters data to detached_counters array.
1096  */
1097 static uint64_t
1098 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1099 {
1100 	struct epoch_tracker et;
1101 	struct lagg_softc *sc;
1102 	struct lagg_port *lp;
1103 	struct ifnet *lpifp;
1104 	uint64_t newval, oldval, vsum;
1105 
1106 	/* Revise this when we've got non-generic counters. */
1107 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1108 
1109 	sc = (struct lagg_softc *)ifp->if_softc;
1110 
1111 	vsum = 0;
1112 	NET_EPOCH_ENTER(et);
1113 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1114 		/* Saved attached value */
1115 		oldval = lp->port_counters.val[cnt];
1116 		/* current value */
1117 		lpifp = lp->lp_ifp;
1118 		newval = lpifp->if_get_counter(lpifp, cnt);
1119 		/* Calculate diff and save new */
1120 		vsum += newval - oldval;
1121 	}
1122 	NET_EPOCH_EXIT(et);
1123 
1124 	/*
1125 	 * Add counter data which might be added by upper
1126 	 * layer protocols operating on logical interface.
1127 	 */
1128 	vsum += if_get_counter_default(ifp, cnt);
1129 
1130 	/*
1131 	 * Add counter data from detached ports counters
1132 	 */
1133 	vsum += sc->detached_counters.val[cnt];
1134 
1135 	return (vsum);
1136 }
1137 
1138 /*
1139  * For direct output to child ports.
1140  */
1141 static int
1142 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1143 	const struct sockaddr *dst, struct route *ro)
1144 {
1145 	struct lagg_port *lp = ifp->if_lagg;
1146 
1147 	switch (dst->sa_family) {
1148 		case pseudo_AF_HDRCMPLT:
1149 		case AF_UNSPEC:
1150 			if (lp != NULL)
1151 				return ((*lp->lp_output)(ifp, m, dst, ro));
1152 	}
1153 
1154 	/* drop any other frames */
1155 	m_freem(m);
1156 	return (ENETDOWN);
1157 }
1158 
1159 static void
1160 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1161 {
1162 	struct lagg_port *lp;
1163 	struct lagg_softc *sc;
1164 
1165 	if ((lp = ifp->if_lagg) == NULL)
1166 		return;
1167 	/* If the ifnet is just being renamed, don't do anything. */
1168 	if (ifp->if_flags & IFF_RENAMING)
1169 		return;
1170 
1171 	sc = lp->lp_softc;
1172 
1173 	LAGG_XLOCK(sc);
1174 	lp->lp_detaching = 1;
1175 	lagg_port_destroy(lp, 1);
1176 	LAGG_XUNLOCK(sc);
1177 	VLAN_CAPABILITIES(sc->sc_ifp);
1178 }
1179 
1180 static void
1181 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1182 {
1183 	struct lagg_softc *sc = lp->lp_softc;
1184 
1185 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1186 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1187 	rp->rp_prio = lp->lp_prio;
1188 	rp->rp_flags = lp->lp_flags;
1189 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1190 
1191 	/* Add protocol specific flags */
1192 	switch (sc->sc_proto) {
1193 		case LAGG_PROTO_FAILOVER:
1194 			if (lp == sc->sc_primary)
1195 				rp->rp_flags |= LAGG_PORT_MASTER;
1196 			if (lp == lagg_link_active(sc, sc->sc_primary))
1197 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1198 			break;
1199 
1200 		case LAGG_PROTO_ROUNDROBIN:
1201 		case LAGG_PROTO_LOADBALANCE:
1202 		case LAGG_PROTO_BROADCAST:
1203 			if (LAGG_PORTACTIVE(lp))
1204 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1205 			break;
1206 
1207 		case LAGG_PROTO_LACP:
1208 			/* LACP has a different definition of active */
1209 			if (lacp_isactive(lp))
1210 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1211 			if (lacp_iscollecting(lp))
1212 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1213 			if (lacp_isdistributing(lp))
1214 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1215 			break;
1216 	}
1217 
1218 }
1219 
1220 static void
1221 lagg_watchdog_infiniband(void *arg)
1222 {
1223 	struct epoch_tracker et;
1224 	struct lagg_softc *sc;
1225 	struct lagg_port *lp;
1226 	struct ifnet *ifp;
1227 	struct ifnet *lp_ifp;
1228 
1229 	sc = arg;
1230 
1231 	/*
1232 	 * Because infiniband nodes have a fixed MAC address, which is
1233 	 * generated by the so-called GID, we need to regularly update
1234 	 * the link level address of the parent lagg<N> device when
1235 	 * the active port changes. Possibly we could piggy-back on
1236 	 * link up/down events aswell, but using a timer also provides
1237 	 * a guarantee against too frequent events. This operation
1238 	 * does not have to be atomic.
1239 	 */
1240 	NET_EPOCH_ENTER(et);
1241 	lp = lagg_link_active(sc, sc->sc_primary);
1242 	if (lp != NULL) {
1243 		ifp = sc->sc_ifp;
1244 		lp_ifp = lp->lp_ifp;
1245 
1246 		if (ifp != NULL && lp_ifp != NULL &&
1247 		    (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0 ||
1248 		     memcmp(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen) != 0)) {
1249 			memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1250 			memcpy(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen);
1251 
1252 			CURVNET_SET(ifp->if_vnet);
1253 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1254 			CURVNET_RESTORE();
1255 		}
1256 	}
1257 	NET_EPOCH_EXIT(et);
1258 
1259 	callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1260 }
1261 
1262 static void
1263 lagg_init(void *xsc)
1264 {
1265 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1266 	struct ifnet *ifp = sc->sc_ifp;
1267 	struct lagg_port *lp;
1268 
1269 	LAGG_XLOCK(sc);
1270 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1271 		LAGG_XUNLOCK(sc);
1272 		return;
1273 	}
1274 
1275 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1276 
1277 	/*
1278 	 * Update the port lladdrs if needed.
1279 	 * This might be if_setlladdr() notification
1280 	 * that lladdr has been changed.
1281 	 */
1282 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1283 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1284 		    ifp->if_addrlen) != 0)
1285 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1286 	}
1287 
1288 	lagg_proto_init(sc);
1289 
1290 	if (ifp->if_type == IFT_INFINIBAND) {
1291 		mtx_lock(&sc->sc_mtx);
1292 		lagg_watchdog_infiniband(sc);
1293 		mtx_unlock(&sc->sc_mtx);
1294 	}
1295 
1296 	LAGG_XUNLOCK(sc);
1297 }
1298 
1299 static void
1300 lagg_stop(struct lagg_softc *sc)
1301 {
1302 	struct ifnet *ifp = sc->sc_ifp;
1303 
1304 	LAGG_XLOCK_ASSERT(sc);
1305 
1306 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1307 		return;
1308 
1309 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1310 
1311 	lagg_proto_stop(sc);
1312 
1313 	mtx_lock(&sc->sc_mtx);
1314 	callout_stop(&sc->sc_watchdog);
1315 	mtx_unlock(&sc->sc_mtx);
1316 
1317 	callout_drain(&sc->sc_watchdog);
1318 }
1319 
1320 static int
1321 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1322 {
1323 	struct epoch_tracker et;
1324 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1325 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1326 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1327 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1328 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1329 	struct ifreq *ifr = (struct ifreq *)data;
1330 	struct lagg_port *lp;
1331 	struct ifnet *tpif;
1332 	struct thread *td = curthread;
1333 	char *buf, *outbuf;
1334 	int count, buflen, len, error = 0, oldmtu;
1335 
1336 	bzero(&rpbuf, sizeof(rpbuf));
1337 
1338 	/* XXX: This can race with lagg_clone_destroy. */
1339 
1340 	switch (cmd) {
1341 	case SIOCGLAGG:
1342 		LAGG_XLOCK(sc);
1343 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1344 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1345 		ra->ra_proto = sc->sc_proto;
1346 		lagg_proto_request(sc, &ra->ra_psc);
1347 		count = 0;
1348 		buf = outbuf;
1349 		len = min(ra->ra_size, buflen);
1350 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1351 			if (len < sizeof(rpbuf))
1352 				break;
1353 
1354 			lagg_port2req(lp, &rpbuf);
1355 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1356 			count++;
1357 			buf += sizeof(rpbuf);
1358 			len -= sizeof(rpbuf);
1359 		}
1360 		LAGG_XUNLOCK(sc);
1361 		ra->ra_ports = count;
1362 		ra->ra_size = count * sizeof(rpbuf);
1363 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1364 		free(outbuf, M_TEMP);
1365 		break;
1366 	case SIOCSLAGG:
1367 		error = priv_check(td, PRIV_NET_LAGG);
1368 		if (error)
1369 			break;
1370 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1371 			error = EPROTONOSUPPORT;
1372 			break;
1373 		}
1374 		/* Infiniband only supports the failover protocol. */
1375 		if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1376 		    ifp->if_type == IFT_INFINIBAND) {
1377 			error = EPROTONOSUPPORT;
1378 			break;
1379 		}
1380 		LAGG_XLOCK(sc);
1381 		lagg_proto_detach(sc);
1382 		lagg_proto_attach(sc, ra->ra_proto);
1383 		LAGG_XUNLOCK(sc);
1384 		break;
1385 	case SIOCGLAGGOPTS:
1386 		LAGG_XLOCK(sc);
1387 		ro->ro_opts = sc->sc_opts;
1388 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1389 			struct lacp_softc *lsc;
1390 
1391 			lsc = (struct lacp_softc *)sc->sc_psc;
1392 			if (lsc->lsc_debug.lsc_tx_test != 0)
1393 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1394 			if (lsc->lsc_debug.lsc_rx_test != 0)
1395 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1396 			if (lsc->lsc_strict_mode != 0)
1397 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1398 			if (lsc->lsc_fast_timeout != 0)
1399 				ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1400 
1401 			ro->ro_active = sc->sc_active;
1402 		} else {
1403 			ro->ro_active = 0;
1404 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1405 				ro->ro_active += LAGG_PORTACTIVE(lp);
1406 		}
1407 		ro->ro_bkt = sc->sc_stride;
1408 		ro->ro_flapping = sc->sc_flapping;
1409 		ro->ro_flowid_shift = sc->flowid_shift;
1410 		LAGG_XUNLOCK(sc);
1411 		break;
1412 	case SIOCSLAGGOPTS:
1413 		error = priv_check(td, PRIV_NET_LAGG);
1414 		if (error)
1415 			break;
1416 
1417 		/*
1418 		 * The stride option was added without defining a corresponding
1419 		 * LAGG_OPT flag, so handle a non-zero value before checking
1420 		 * anything else to preserve compatibility.
1421 		 */
1422 		LAGG_XLOCK(sc);
1423 		if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1424 			if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1425 				LAGG_XUNLOCK(sc);
1426 				error = EINVAL;
1427 				break;
1428 			}
1429 			sc->sc_stride = ro->ro_bkt;
1430 		}
1431 		if (ro->ro_opts == 0) {
1432 			LAGG_XUNLOCK(sc);
1433 			break;
1434 		}
1435 
1436 		/*
1437 		 * Set options.  LACP options are stored in sc->sc_psc,
1438 		 * not in sc_opts.
1439 		 */
1440 		int valid, lacp;
1441 
1442 		switch (ro->ro_opts) {
1443 		case LAGG_OPT_USE_FLOWID:
1444 		case -LAGG_OPT_USE_FLOWID:
1445 		case LAGG_OPT_USE_NUMA:
1446 		case -LAGG_OPT_USE_NUMA:
1447 		case LAGG_OPT_FLOWIDSHIFT:
1448 		case LAGG_OPT_RR_LIMIT:
1449 			valid = 1;
1450 			lacp = 0;
1451 			break;
1452 		case LAGG_OPT_LACP_TXTEST:
1453 		case -LAGG_OPT_LACP_TXTEST:
1454 		case LAGG_OPT_LACP_RXTEST:
1455 		case -LAGG_OPT_LACP_RXTEST:
1456 		case LAGG_OPT_LACP_STRICT:
1457 		case -LAGG_OPT_LACP_STRICT:
1458 		case LAGG_OPT_LACP_FAST_TIMO:
1459 		case -LAGG_OPT_LACP_FAST_TIMO:
1460 			valid = lacp = 1;
1461 			break;
1462 		default:
1463 			valid = lacp = 0;
1464 			break;
1465 		}
1466 
1467 		if (valid == 0 ||
1468 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1469 			/* Invalid combination of options specified. */
1470 			error = EINVAL;
1471 			LAGG_XUNLOCK(sc);
1472 			break;	/* Return from SIOCSLAGGOPTS. */
1473 		}
1474 
1475 		/*
1476 		 * Store new options into sc->sc_opts except for
1477 		 * FLOWIDSHIFT, RR and LACP options.
1478 		 */
1479 		if (lacp == 0) {
1480 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1481 				sc->flowid_shift = ro->ro_flowid_shift;
1482 			else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1483 				if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1484 				    ro->ro_bkt == 0) {
1485 					error = EINVAL;
1486 					LAGG_XUNLOCK(sc);
1487 					break;
1488 				}
1489 				sc->sc_stride = ro->ro_bkt;
1490 			} else if (ro->ro_opts > 0)
1491 				sc->sc_opts |= ro->ro_opts;
1492 			else
1493 				sc->sc_opts &= ~ro->ro_opts;
1494 		} else {
1495 			struct lacp_softc *lsc;
1496 			struct lacp_port *lp;
1497 
1498 			lsc = (struct lacp_softc *)sc->sc_psc;
1499 
1500 			switch (ro->ro_opts) {
1501 			case LAGG_OPT_LACP_TXTEST:
1502 				lsc->lsc_debug.lsc_tx_test = 1;
1503 				break;
1504 			case -LAGG_OPT_LACP_TXTEST:
1505 				lsc->lsc_debug.lsc_tx_test = 0;
1506 				break;
1507 			case LAGG_OPT_LACP_RXTEST:
1508 				lsc->lsc_debug.lsc_rx_test = 1;
1509 				break;
1510 			case -LAGG_OPT_LACP_RXTEST:
1511 				lsc->lsc_debug.lsc_rx_test = 0;
1512 				break;
1513 			case LAGG_OPT_LACP_STRICT:
1514 				lsc->lsc_strict_mode = 1;
1515 				break;
1516 			case -LAGG_OPT_LACP_STRICT:
1517 				lsc->lsc_strict_mode = 0;
1518 				break;
1519 			case LAGG_OPT_LACP_FAST_TIMO:
1520 				LACP_LOCK(lsc);
1521         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1522                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1523 				LACP_UNLOCK(lsc);
1524 				lsc->lsc_fast_timeout = 1;
1525 				break;
1526 			case -LAGG_OPT_LACP_FAST_TIMO:
1527 				LACP_LOCK(lsc);
1528         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1529                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1530 				LACP_UNLOCK(lsc);
1531 				lsc->lsc_fast_timeout = 0;
1532 				break;
1533 			}
1534 		}
1535 		LAGG_XUNLOCK(sc);
1536 		break;
1537 	case SIOCGLAGGFLAGS:
1538 		rf->rf_flags = 0;
1539 		LAGG_XLOCK(sc);
1540 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1541 			rf->rf_flags |= LAGG_F_HASHL2;
1542 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1543 			rf->rf_flags |= LAGG_F_HASHL3;
1544 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1545 			rf->rf_flags |= LAGG_F_HASHL4;
1546 		LAGG_XUNLOCK(sc);
1547 		break;
1548 	case SIOCSLAGGHASH:
1549 		error = priv_check(td, PRIV_NET_LAGG);
1550 		if (error)
1551 			break;
1552 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1553 			error = EINVAL;
1554 			break;
1555 		}
1556 		LAGG_XLOCK(sc);
1557 		sc->sc_flags = 0;
1558 		if (rf->rf_flags & LAGG_F_HASHL2)
1559 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1560 		if (rf->rf_flags & LAGG_F_HASHL3)
1561 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1562 		if (rf->rf_flags & LAGG_F_HASHL4)
1563 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1564 		LAGG_XUNLOCK(sc);
1565 		break;
1566 	case SIOCGLAGGPORT:
1567 		if (rp->rp_portname[0] == '\0' ||
1568 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1569 			error = EINVAL;
1570 			break;
1571 		}
1572 
1573 		NET_EPOCH_ENTER(et);
1574 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1575 		    lp->lp_softc != sc) {
1576 			error = ENOENT;
1577 			NET_EPOCH_EXIT(et);
1578 			if_rele(tpif);
1579 			break;
1580 		}
1581 
1582 		lagg_port2req(lp, rp);
1583 		NET_EPOCH_EXIT(et);
1584 		if_rele(tpif);
1585 		break;
1586 	case SIOCSLAGGPORT:
1587 		error = priv_check(td, PRIV_NET_LAGG);
1588 		if (error)
1589 			break;
1590 		if (rp->rp_portname[0] == '\0' ||
1591 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1592 			error = EINVAL;
1593 			break;
1594 		}
1595 #ifdef INET6
1596 		/*
1597 		 * A laggport interface should not have inet6 address
1598 		 * because two interfaces with a valid link-local
1599 		 * scope zone must not be merged in any form.  This
1600 		 * restriction is needed to prevent violation of
1601 		 * link-local scope zone.  Attempts to add a laggport
1602 		 * interface which has inet6 addresses triggers
1603 		 * removal of all inet6 addresses on the member
1604 		 * interface.
1605 		 */
1606 		if (in6ifa_llaonifp(tpif)) {
1607 			in6_ifdetach(tpif);
1608 				if_printf(sc->sc_ifp,
1609 				    "IPv6 addresses on %s have been removed "
1610 				    "before adding it as a member to prevent "
1611 				    "IPv6 address scope violation.\n",
1612 				    tpif->if_xname);
1613 		}
1614 #endif
1615 		oldmtu = ifp->if_mtu;
1616 		LAGG_XLOCK(sc);
1617 		error = lagg_port_create(sc, tpif);
1618 		LAGG_XUNLOCK(sc);
1619 		if_rele(tpif);
1620 
1621 		/*
1622 		 * LAGG MTU may change during addition of the first port.
1623 		 * If it did, do network layer specific procedure.
1624 		 */
1625 		if (ifp->if_mtu != oldmtu) {
1626 #ifdef INET6
1627 			nd6_setmtu(ifp);
1628 #endif
1629 			rt_updatemtu(ifp);
1630 		}
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,
2001 		    ifma->ifma_addr->sa_len);
2002 		mc->mc_addr.sdl_index = ifp->if_index;
2003 		mc->mc_ifma = NULL;
2004 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
2005 	}
2006 	IF_ADDR_WUNLOCK(scifp);
2007 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
2008 		error = if_addmulti(ifp,
2009 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
2010 		if (error)
2011 			return (error);
2012 	}
2013 	return (0);
2014 }
2015 
2016 static int
2017 lagg_clrmulti(struct lagg_port *lp)
2018 {
2019 	struct lagg_mc *mc;
2020 
2021 	LAGG_XLOCK_ASSERT(lp->lp_softc);
2022 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
2023 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
2024 		if (mc->mc_ifma && lp->lp_detaching == 0)
2025 			if_delmulti_ifma(mc->mc_ifma);
2026 		free(mc, M_LAGG);
2027 	}
2028 	return (0);
2029 }
2030 
2031 static void
2032 lagg_setcaps(struct lagg_port *lp, int cap, int cap2)
2033 {
2034 	struct ifreq ifr;
2035 	struct siocsifcapnv_driver_data drv_ioctl_data;
2036 
2037 	if (lp->lp_ifp->if_capenable == cap &&
2038 	    lp->lp_ifp->if_capenable2 == cap2)
2039 		return;
2040 	if (lp->lp_ioctl == NULL)
2041 		return;
2042 	/* XXX */
2043 	if ((lp->lp_ifp->if_capabilities & IFCAP_NV) != 0) {
2044 		drv_ioctl_data.reqcap = cap;
2045 		drv_ioctl_data.reqcap2 = cap2;
2046 		drv_ioctl_data.nvcap = NULL;
2047 		(*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAPNV,
2048 		    (caddr_t)&drv_ioctl_data);
2049 	} else {
2050 		ifr.ifr_reqcap = cap;
2051 		(*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr);
2052 	}
2053 }
2054 
2055 /* Handle a ref counted flag that should be set on the lagg port as well */
2056 static int
2057 lagg_setflag(struct lagg_port *lp, int flag, int status,
2058     int (*func)(struct ifnet *, int))
2059 {
2060 	struct lagg_softc *sc = lp->lp_softc;
2061 	struct ifnet *scifp = sc->sc_ifp;
2062 	struct ifnet *ifp = lp->lp_ifp;
2063 	int error;
2064 
2065 	LAGG_XLOCK_ASSERT(sc);
2066 
2067 	status = status ? (scifp->if_flags & flag) : 0;
2068 	/* Now "status" contains the flag value or 0 */
2069 
2070 	/*
2071 	 * See if recorded ports status is different from what
2072 	 * we want it to be.  If it is, flip it.  We record ports
2073 	 * status in lp_ifflags so that we won't clear ports flag
2074 	 * we haven't set.  In fact, we don't clear or set ports
2075 	 * flags directly, but get or release references to them.
2076 	 * That's why we can be sure that recorded flags still are
2077 	 * in accord with actual ports flags.
2078 	 */
2079 	if (status != (lp->lp_ifflags & flag)) {
2080 		error = (*func)(ifp, status);
2081 		if (error)
2082 			return (error);
2083 		lp->lp_ifflags &= ~flag;
2084 		lp->lp_ifflags |= status;
2085 	}
2086 	return (0);
2087 }
2088 
2089 /*
2090  * Handle IFF_* flags that require certain changes on the lagg port
2091  * if "status" is true, update ports flags respective to the lagg
2092  * if "status" is false, forcedly clear the flags set on port.
2093  */
2094 static int
2095 lagg_setflags(struct lagg_port *lp, int status)
2096 {
2097 	int error, i;
2098 
2099 	for (i = 0; lagg_pflags[i].flag; i++) {
2100 		error = lagg_setflag(lp, lagg_pflags[i].flag,
2101 		    status, lagg_pflags[i].func);
2102 		if (error)
2103 			return (error);
2104 	}
2105 	return (0);
2106 }
2107 
2108 static int
2109 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2110 {
2111 	struct epoch_tracker et;
2112 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2113 	int error;
2114 
2115 #if defined(KERN_TLS) || defined(RATELIMIT)
2116 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2117 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2118 #endif
2119 	NET_EPOCH_ENTER(et);
2120 	/* We need a Tx algorithm and at least one port */
2121 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2122 		NET_EPOCH_EXIT(et);
2123 		m_freem(m);
2124 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2125 		return (ENXIO);
2126 	}
2127 
2128 	ETHER_BPF_MTAP(ifp, m);
2129 
2130 	error = lagg_proto_start(sc, m);
2131 	NET_EPOCH_EXIT(et);
2132 	return (error);
2133 }
2134 
2135 static int
2136 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2137 {
2138 	struct epoch_tracker et;
2139 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2140 	int error;
2141 
2142 #if defined(KERN_TLS) || defined(RATELIMIT)
2143 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2144 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2145 #endif
2146 	NET_EPOCH_ENTER(et);
2147 	/* We need a Tx algorithm and at least one port */
2148 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2149 		NET_EPOCH_EXIT(et);
2150 		m_freem(m);
2151 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2152 		return (ENXIO);
2153 	}
2154 
2155 	INFINIBAND_BPF_MTAP(ifp, m);
2156 
2157 	error = lagg_proto_start(sc, m);
2158 	NET_EPOCH_EXIT(et);
2159 	return (error);
2160 }
2161 
2162 /*
2163  * The ifp->if_qflush entry point for lagg(4) is no-op.
2164  */
2165 static void
2166 lagg_qflush(struct ifnet *ifp __unused)
2167 {
2168 }
2169 
2170 static struct mbuf *
2171 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2172 {
2173 	struct epoch_tracker et;
2174 	struct lagg_port *lp = ifp->if_lagg;
2175 	struct lagg_softc *sc = lp->lp_softc;
2176 	struct ifnet *scifp = sc->sc_ifp;
2177 
2178 	NET_EPOCH_ENTER(et);
2179 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2180 	    lp->lp_detaching != 0 ||
2181 	    sc->sc_proto == LAGG_PROTO_NONE) {
2182 		NET_EPOCH_EXIT(et);
2183 		m_freem(m);
2184 		return (NULL);
2185 	}
2186 
2187 	ETHER_BPF_MTAP(scifp, m);
2188 
2189 	m = lagg_proto_input(sc, lp, m);
2190 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2191 		m_freem(m);
2192 		m = NULL;
2193 	}
2194 
2195 	NET_EPOCH_EXIT(et);
2196 	return (m);
2197 }
2198 
2199 static struct mbuf *
2200 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2201 {
2202 	struct epoch_tracker et;
2203 	struct lagg_port *lp = ifp->if_lagg;
2204 	struct lagg_softc *sc = lp->lp_softc;
2205 	struct ifnet *scifp = sc->sc_ifp;
2206 
2207 	NET_EPOCH_ENTER(et);
2208 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2209 	    lp->lp_detaching != 0 ||
2210 	    sc->sc_proto == LAGG_PROTO_NONE) {
2211 		NET_EPOCH_EXIT(et);
2212 		m_freem(m);
2213 		return (NULL);
2214 	}
2215 
2216 	INFINIBAND_BPF_MTAP(scifp, m);
2217 
2218 	m = lagg_proto_input(sc, lp, m);
2219 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2220 		m_freem(m);
2221 		m = NULL;
2222 	}
2223 
2224 	NET_EPOCH_EXIT(et);
2225 	return (m);
2226 }
2227 
2228 static int
2229 lagg_media_change(struct ifnet *ifp)
2230 {
2231 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2232 
2233 	if (sc->sc_ifflags & IFF_DEBUG)
2234 		printf("%s\n", __func__);
2235 
2236 	/* Ignore */
2237 	return (0);
2238 }
2239 
2240 static void
2241 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2242 {
2243 	struct epoch_tracker et;
2244 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2245 	struct lagg_port *lp;
2246 
2247 	imr->ifm_status = IFM_AVALID;
2248 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
2249 
2250 	NET_EPOCH_ENTER(et);
2251 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2252 		if (LAGG_PORTACTIVE(lp))
2253 			imr->ifm_status |= IFM_ACTIVE;
2254 	}
2255 	NET_EPOCH_EXIT(et);
2256 }
2257 
2258 static void
2259 lagg_linkstate(struct lagg_softc *sc)
2260 {
2261 	struct epoch_tracker et;
2262 	struct lagg_port *lp;
2263 	int new_link = LINK_STATE_DOWN;
2264 	uint64_t speed;
2265 
2266 	LAGG_XLOCK_ASSERT(sc);
2267 
2268 	/* LACP handles link state itself */
2269 	if (sc->sc_proto == LAGG_PROTO_LACP)
2270 		return;
2271 
2272 	/* Our link is considered up if at least one of our ports is active */
2273 	NET_EPOCH_ENTER(et);
2274 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2275 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2276 			new_link = LINK_STATE_UP;
2277 			break;
2278 		}
2279 	}
2280 	NET_EPOCH_EXIT(et);
2281 	if_link_state_change(sc->sc_ifp, new_link);
2282 
2283 	/* Update if_baudrate to reflect the max possible speed */
2284 	switch (sc->sc_proto) {
2285 		case LAGG_PROTO_FAILOVER:
2286 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2287 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
2288 			break;
2289 		case LAGG_PROTO_ROUNDROBIN:
2290 		case LAGG_PROTO_LOADBALANCE:
2291 		case LAGG_PROTO_BROADCAST:
2292 			speed = 0;
2293 			NET_EPOCH_ENTER(et);
2294 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2295 				speed += lp->lp_ifp->if_baudrate;
2296 			NET_EPOCH_EXIT(et);
2297 			sc->sc_ifp->if_baudrate = speed;
2298 			break;
2299 		case LAGG_PROTO_LACP:
2300 			/* LACP updates if_baudrate itself */
2301 			break;
2302 	}
2303 }
2304 
2305 static void
2306 lagg_port_state(struct ifnet *ifp, int state)
2307 {
2308 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2309 	struct lagg_softc *sc = NULL;
2310 
2311 	if (lp != NULL)
2312 		sc = lp->lp_softc;
2313 	if (sc == NULL)
2314 		return;
2315 
2316 	LAGG_XLOCK(sc);
2317 	lagg_linkstate(sc);
2318 	lagg_proto_linkstate(sc, lp);
2319 	LAGG_XUNLOCK(sc);
2320 }
2321 
2322 struct lagg_port *
2323 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2324 {
2325 	struct lagg_port *lp_next, *rval = NULL;
2326 
2327 	/*
2328 	 * Search a port which reports an active link state.
2329 	 */
2330 
2331 #ifdef INVARIANTS
2332 	/*
2333 	 * This is called with either in the network epoch
2334 	 * or with LAGG_XLOCK(sc) held.
2335 	 */
2336 	if (!in_epoch(net_epoch_preempt))
2337 		LAGG_XLOCK_ASSERT(sc);
2338 #endif
2339 
2340 	if (lp == NULL)
2341 		goto search;
2342 	if (LAGG_PORTACTIVE(lp)) {
2343 		rval = lp;
2344 		goto found;
2345 	}
2346 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2347 	    LAGG_PORTACTIVE(lp_next)) {
2348 		rval = lp_next;
2349 		goto found;
2350 	}
2351 
2352 search:
2353 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2354 		if (LAGG_PORTACTIVE(lp_next)) {
2355 			return (lp_next);
2356 		}
2357 	}
2358 found:
2359 	return (rval);
2360 }
2361 
2362 int
2363 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2364 {
2365 
2366 #if defined(KERN_TLS) || defined(RATELIMIT)
2367 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2368 		struct lagg_snd_tag *lst;
2369 		struct m_snd_tag *mst;
2370 
2371 		mst = m->m_pkthdr.snd_tag;
2372 		lst = mst_to_lst(mst);
2373 		if (lst->tag->ifp != ifp) {
2374 			m_freem(m);
2375 			return (EAGAIN);
2376 		}
2377 		m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2378 		m_snd_tag_rele(mst);
2379 	}
2380 #endif
2381 	return (ifp->if_transmit)(ifp, m);
2382 }
2383 
2384 /*
2385  * Simple round robin aggregation
2386  */
2387 static void
2388 lagg_rr_attach(struct lagg_softc *sc)
2389 {
2390 	sc->sc_seq = 0;
2391 	sc->sc_stride = 1;
2392 }
2393 
2394 static int
2395 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2396 {
2397 	struct lagg_port *lp;
2398 	uint32_t p;
2399 
2400 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
2401 	p /= sc->sc_stride;
2402 	p %= sc->sc_count;
2403 	lp = CK_SLIST_FIRST(&sc->sc_ports);
2404 
2405 	while (p--)
2406 		lp = CK_SLIST_NEXT(lp, lp_entries);
2407 
2408 	/*
2409 	 * Check the port's link state. This will return the next active
2410 	 * port if the link is down or the port is NULL.
2411 	 */
2412 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2413 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2414 		m_freem(m);
2415 		return (ENETDOWN);
2416 	}
2417 
2418 	/* Send mbuf */
2419 	return (lagg_enqueue(lp->lp_ifp, m));
2420 }
2421 
2422 static struct mbuf *
2423 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2424 {
2425 	struct ifnet *ifp = sc->sc_ifp;
2426 
2427 	/* Just pass in the packet to our lagg device */
2428 	m->m_pkthdr.rcvif = ifp;
2429 
2430 	return (m);
2431 }
2432 
2433 /*
2434  * Broadcast mode
2435  */
2436 static int
2437 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2438 {
2439 	int errors = 0;
2440 	int ret;
2441 	struct lagg_port *lp, *last = NULL;
2442 	struct mbuf *m0;
2443 
2444 	NET_EPOCH_ASSERT();
2445 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2446 		if (!LAGG_PORTACTIVE(lp))
2447 			continue;
2448 
2449 		if (last != NULL) {
2450 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2451 			if (m0 == NULL) {
2452 				ret = ENOBUFS;
2453 				errors++;
2454 				break;
2455 			}
2456 			lagg_enqueue(last->lp_ifp, m0);
2457 		}
2458 		last = lp;
2459 	}
2460 
2461 	if (last == NULL) {
2462 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2463 		m_freem(m);
2464 		return (ENOENT);
2465 	}
2466 	if ((last = lagg_link_active(sc, last)) == NULL) {
2467 		errors++;
2468 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2469 		m_freem(m);
2470 		return (ENETDOWN);
2471 	}
2472 
2473 	ret = lagg_enqueue(last->lp_ifp, m);
2474 	if (errors != 0)
2475 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2476 
2477 	return (ret);
2478 }
2479 
2480 static struct mbuf*
2481 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2482 {
2483 	struct ifnet *ifp = sc->sc_ifp;
2484 
2485 	/* Just pass in the packet to our lagg device */
2486 	m->m_pkthdr.rcvif = ifp;
2487 	return (m);
2488 }
2489 
2490 /*
2491  * Active failover
2492  */
2493 static int
2494 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2495 {
2496 	struct lagg_port *lp;
2497 
2498 	/* Use the master port if active or the next available port */
2499 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2500 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2501 		m_freem(m);
2502 		return (ENETDOWN);
2503 	}
2504 
2505 	/* Send mbuf */
2506 	return (lagg_enqueue(lp->lp_ifp, m));
2507 }
2508 
2509 static struct mbuf *
2510 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2511 {
2512 	struct ifnet *ifp = sc->sc_ifp;
2513 	struct lagg_port *tmp_tp;
2514 
2515 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2516 		m->m_pkthdr.rcvif = ifp;
2517 		return (m);
2518 	}
2519 
2520 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2521 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2522 		/*
2523 		 * If tmp_tp is null, we've received a packet when all
2524 		 * our links are down. Weird, but process it anyways.
2525 		 */
2526 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2527 			m->m_pkthdr.rcvif = ifp;
2528 			return (m);
2529 		}
2530 	}
2531 
2532 	m_freem(m);
2533 	return (NULL);
2534 }
2535 
2536 /*
2537  * Loadbalancing
2538  */
2539 static void
2540 lagg_lb_attach(struct lagg_softc *sc)
2541 {
2542 	struct lagg_port *lp;
2543 	struct lagg_lb *lb;
2544 
2545 	LAGG_XLOCK_ASSERT(sc);
2546 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2547 	lb->lb_key = m_ether_tcpip_hash_init();
2548 	sc->sc_psc = lb;
2549 
2550 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2551 		lagg_lb_port_create(lp);
2552 }
2553 
2554 static void
2555 lagg_lb_detach(struct lagg_softc *sc)
2556 {
2557 	struct lagg_lb *lb;
2558 
2559 	lb = (struct lagg_lb *)sc->sc_psc;
2560 	if (lb != NULL)
2561 		free(lb, M_LAGG);
2562 }
2563 
2564 static int
2565 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2566 {
2567 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2568 	struct lagg_port *lp_next;
2569 	int i = 0, rv;
2570 
2571 	rv = 0;
2572 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2573 	LAGG_XLOCK_ASSERT(sc);
2574 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2575 		if (lp_next == lp)
2576 			continue;
2577 		if (i >= LAGG_MAX_PORTS) {
2578 			rv = EINVAL;
2579 			break;
2580 		}
2581 		if (sc->sc_ifflags & IFF_DEBUG)
2582 			printf("%s: port %s at index %d\n",
2583 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2584 		lb->lb_ports[i++] = lp_next;
2585 	}
2586 
2587 	return (rv);
2588 }
2589 
2590 static int
2591 lagg_lb_port_create(struct lagg_port *lp)
2592 {
2593 	struct lagg_softc *sc = lp->lp_softc;
2594 	return (lagg_lb_porttable(sc, NULL));
2595 }
2596 
2597 static void
2598 lagg_lb_port_destroy(struct lagg_port *lp)
2599 {
2600 	struct lagg_softc *sc = lp->lp_softc;
2601 	lagg_lb_porttable(sc, lp);
2602 }
2603 
2604 static int
2605 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2606 {
2607 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2608 	struct lagg_port *lp = NULL;
2609 	uint32_t p = 0;
2610 
2611 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2612 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2613 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2614 	else
2615 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2616 	p %= sc->sc_count;
2617 	lp = lb->lb_ports[p];
2618 
2619 	/*
2620 	 * Check the port's link state. This will return the next active
2621 	 * port if the link is down or the port is NULL.
2622 	 */
2623 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2624 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2625 		m_freem(m);
2626 		return (ENETDOWN);
2627 	}
2628 
2629 	/* Send mbuf */
2630 	return (lagg_enqueue(lp->lp_ifp, m));
2631 }
2632 
2633 static struct mbuf *
2634 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2635 {
2636 	struct ifnet *ifp = sc->sc_ifp;
2637 
2638 	/* Just pass in the packet to our lagg device */
2639 	m->m_pkthdr.rcvif = ifp;
2640 
2641 	return (m);
2642 }
2643 
2644 /*
2645  * 802.3ad LACP
2646  */
2647 static void
2648 lagg_lacp_attach(struct lagg_softc *sc)
2649 {
2650 	struct lagg_port *lp;
2651 
2652 	lacp_attach(sc);
2653 	LAGG_XLOCK_ASSERT(sc);
2654 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2655 		lacp_port_create(lp);
2656 }
2657 
2658 static void
2659 lagg_lacp_detach(struct lagg_softc *sc)
2660 {
2661 	struct lagg_port *lp;
2662 	void *psc;
2663 
2664 	LAGG_XLOCK_ASSERT(sc);
2665 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2666 		lacp_port_destroy(lp);
2667 
2668 	psc = sc->sc_psc;
2669 	sc->sc_psc = NULL;
2670 	lacp_detach(psc);
2671 }
2672 
2673 static void
2674 lagg_lacp_lladdr(struct lagg_softc *sc)
2675 {
2676 	struct lagg_port *lp;
2677 
2678 	LAGG_SXLOCK_ASSERT(sc);
2679 
2680 	/* purge all the lacp ports */
2681 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2682 		lacp_port_destroy(lp);
2683 
2684 	/* add them back in */
2685 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2686 		lacp_port_create(lp);
2687 }
2688 
2689 static int
2690 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2691 {
2692 	struct lagg_port *lp;
2693 	int err;
2694 
2695 	lp = lacp_select_tx_port(sc, m, &err);
2696 	if (lp == NULL) {
2697 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2698 		m_freem(m);
2699 		return (err);
2700 	}
2701 
2702 	/* Send mbuf */
2703 	return (lagg_enqueue(lp->lp_ifp, m));
2704 }
2705 
2706 static struct mbuf *
2707 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2708 {
2709 	struct ifnet *ifp = sc->sc_ifp;
2710 	struct ether_header *eh;
2711 	u_short etype;
2712 
2713 	eh = mtod(m, struct ether_header *);
2714 	etype = ntohs(eh->ether_type);
2715 
2716 	/* Tap off LACP control messages */
2717 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2718 		m = lacp_input(lp, m);
2719 		if (m == NULL)
2720 			return (NULL);
2721 	}
2722 
2723 	/*
2724 	 * If the port is not collecting or not in the active aggregator then
2725 	 * free and return.
2726 	 */
2727 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2728 		m_freem(m);
2729 		return (NULL);
2730 	}
2731 
2732 	m->m_pkthdr.rcvif = ifp;
2733 	return (m);
2734 }
2735