xref: /freebsd/sys/net/if_lagg.c (revision f39bffc62c1395bde25d152c7f68fdf7cbaab414)
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_ratelimit.h"
27 
28 #include <sys/param.h>
29 #include <sys/kernel.h>
30 #include <sys/malloc.h>
31 #include <sys/mbuf.h>
32 #include <sys/queue.h>
33 #include <sys/socket.h>
34 #include <sys/sockio.h>
35 #include <sys/sysctl.h>
36 #include <sys/module.h>
37 #include <sys/priv.h>
38 #include <sys/systm.h>
39 #include <sys/proc.h>
40 #include <sys/lock.h>
41 #include <sys/rmlock.h>
42 #include <sys/sx.h>
43 #include <sys/taskqueue.h>
44 #include <sys/eventhandler.h>
45 
46 #include <net/ethernet.h>
47 #include <net/if.h>
48 #include <net/if_clone.h>
49 #include <net/if_arp.h>
50 #include <net/if_dl.h>
51 #include <net/if_media.h>
52 #include <net/if_types.h>
53 #include <net/if_var.h>
54 #include <net/bpf.h>
55 #include <net/vnet.h>
56 
57 #if defined(INET) || defined(INET6)
58 #include <netinet/in.h>
59 #include <netinet/ip.h>
60 #endif
61 #ifdef INET
62 #include <netinet/in_systm.h>
63 #include <netinet/if_ether.h>
64 #endif
65 
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #include <netinet6/in6_var.h>
69 #include <netinet6/in6_ifattach.h>
70 #endif
71 
72 #include <net/if_vlan_var.h>
73 #include <net/if_lagg.h>
74 #include <net/ieee8023ad_lacp.h>
75 
76 #define	LAGG_RLOCK()	epoch_enter_preempt(net_epoch_preempt)
77 #define	LAGG_RUNLOCK()	epoch_exit_preempt(net_epoch_preempt)
78 #define	LAGG_RLOCK_ASSERT()	MPASS(in_epoch())
79 #define	LAGG_UNLOCK_ASSERT()	MPASS(!in_epoch())
80 
81 #define	LAGG_SX_INIT(_sc)	sx_init(&(_sc)->sc_sx, "if_lagg sx")
82 #define	LAGG_SX_DESTROY(_sc)	sx_destroy(&(_sc)->sc_sx)
83 #define	LAGG_XLOCK(_sc)		sx_xlock(&(_sc)->sc_sx)
84 #define	LAGG_XUNLOCK(_sc)	sx_xunlock(&(_sc)->sc_sx)
85 #define	LAGG_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
86 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
87 
88 /* Special flags we should propagate to the lagg ports. */
89 static struct {
90 	int flag;
91 	int (*func)(struct ifnet *, int);
92 } lagg_pflags[] = {
93 	{IFF_PROMISC, ifpromisc},
94 	{IFF_ALLMULTI, if_allmulti},
95 	{0, NULL}
96 };
97 
98 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
99 #define	V_lagg_list	VNET(lagg_list)
100 static VNET_DEFINE(struct mtx, lagg_list_mtx);
101 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
102 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
103 					"if_lagg list", NULL, MTX_DEF)
104 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
105 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
106 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
107 eventhandler_tag	lagg_detach_cookie = NULL;
108 
109 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
110 static void	lagg_clone_destroy(struct ifnet *);
111 static VNET_DEFINE(struct if_clone *, lagg_cloner);
112 #define	V_lagg_cloner	VNET(lagg_cloner)
113 static const char laggname[] = "lagg";
114 
115 static void	lagg_capabilities(struct lagg_softc *);
116 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
117 static int	lagg_port_destroy(struct lagg_port *, int);
118 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
119 static void	lagg_linkstate(struct lagg_softc *);
120 static void	lagg_port_state(struct ifnet *, int);
121 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
122 static int	lagg_port_output(struct ifnet *, struct mbuf *,
123 		    const struct sockaddr *, struct route *);
124 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
125 #ifdef LAGG_PORT_STACKING
126 static int	lagg_port_checkstacking(struct lagg_softc *);
127 #endif
128 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
129 static void	lagg_init(void *);
130 static void	lagg_stop(struct lagg_softc *);
131 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
132 #ifdef RATELIMIT
133 static int	lagg_snd_tag_alloc(struct ifnet *,
134 		    union if_snd_tag_alloc_params *,
135 		    struct m_snd_tag **);
136 #endif
137 static int	lagg_setmulti(struct lagg_port *);
138 static int	lagg_clrmulti(struct lagg_port *);
139 static	int	lagg_setcaps(struct lagg_port *, int cap);
140 static	int	lagg_setflag(struct lagg_port *, int, int,
141 		    int (*func)(struct ifnet *, int));
142 static	int	lagg_setflags(struct lagg_port *, int status);
143 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
144 static int	lagg_transmit(struct ifnet *, struct mbuf *);
145 static void	lagg_qflush(struct ifnet *);
146 static int	lagg_media_change(struct ifnet *);
147 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
148 static struct lagg_port *lagg_link_active(struct lagg_softc *,
149 	    struct lagg_port *);
150 
151 /* Simple round robin */
152 static void	lagg_rr_attach(struct lagg_softc *);
153 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
154 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
155 		    struct mbuf *);
156 
157 /* Active failover */
158 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
159 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
160 		    struct mbuf *);
161 
162 /* Loadbalancing */
163 static void	lagg_lb_attach(struct lagg_softc *);
164 static void	lagg_lb_detach(struct lagg_softc *);
165 static int	lagg_lb_port_create(struct lagg_port *);
166 static void	lagg_lb_port_destroy(struct lagg_port *);
167 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
168 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
169 		    struct mbuf *);
170 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
171 
172 /* Broadcast */
173 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
174 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
175 		    struct mbuf *);
176 
177 /* 802.3ad LACP */
178 static void	lagg_lacp_attach(struct lagg_softc *);
179 static void	lagg_lacp_detach(struct lagg_softc *);
180 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
181 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
182 		    struct mbuf *);
183 static void	lagg_lacp_lladdr(struct lagg_softc *);
184 
185 /* lagg protocol table */
186 static const struct lagg_proto {
187 	lagg_proto	pr_num;
188 	void		(*pr_attach)(struct lagg_softc *);
189 	void		(*pr_detach)(struct lagg_softc *);
190 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
191 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
192 			    struct mbuf *);
193 	int		(*pr_addport)(struct lagg_port *);
194 	void		(*pr_delport)(struct lagg_port *);
195 	void		(*pr_linkstate)(struct lagg_port *);
196 	void 		(*pr_init)(struct lagg_softc *);
197 	void 		(*pr_stop)(struct lagg_softc *);
198 	void 		(*pr_lladdr)(struct lagg_softc *);
199 	void		(*pr_request)(struct lagg_softc *, void *);
200 	void		(*pr_portreq)(struct lagg_port *, void *);
201 } lagg_protos[] = {
202     {
203 	.pr_num = LAGG_PROTO_NONE
204     },
205     {
206 	.pr_num = LAGG_PROTO_ROUNDROBIN,
207 	.pr_attach = lagg_rr_attach,
208 	.pr_start = lagg_rr_start,
209 	.pr_input = lagg_rr_input,
210     },
211     {
212 	.pr_num = LAGG_PROTO_FAILOVER,
213 	.pr_start = lagg_fail_start,
214 	.pr_input = lagg_fail_input,
215     },
216     {
217 	.pr_num = LAGG_PROTO_LOADBALANCE,
218 	.pr_attach = lagg_lb_attach,
219 	.pr_detach = lagg_lb_detach,
220 	.pr_start = lagg_lb_start,
221 	.pr_input = lagg_lb_input,
222 	.pr_addport = lagg_lb_port_create,
223 	.pr_delport = lagg_lb_port_destroy,
224     },
225     {
226 	.pr_num = LAGG_PROTO_LACP,
227 	.pr_attach = lagg_lacp_attach,
228 	.pr_detach = lagg_lacp_detach,
229 	.pr_start = lagg_lacp_start,
230 	.pr_input = lagg_lacp_input,
231 	.pr_addport = lacp_port_create,
232 	.pr_delport = lacp_port_destroy,
233 	.pr_linkstate = lacp_linkstate,
234 	.pr_init = lacp_init,
235 	.pr_stop = lacp_stop,
236 	.pr_lladdr = lagg_lacp_lladdr,
237 	.pr_request = lacp_req,
238 	.pr_portreq = lacp_portreq,
239     },
240     {
241 	.pr_num = LAGG_PROTO_BROADCAST,
242 	.pr_start = lagg_bcast_start,
243 	.pr_input = lagg_bcast_input,
244     },
245 };
246 
247 SYSCTL_DECL(_net_link);
248 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
249     "Link Aggregation");
250 
251 /* Allow input on any failover links */
252 static VNET_DEFINE(int, lagg_failover_rx_all);
253 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
254 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
255     &VNET_NAME(lagg_failover_rx_all), 0,
256     "Accept input from any interface in a failover lagg");
257 
258 /* Default value for using flowid */
259 static VNET_DEFINE(int, def_use_flowid) = 0;
260 #define	V_def_use_flowid	VNET(def_use_flowid)
261 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
262     &VNET_NAME(def_use_flowid), 0,
263     "Default setting for using flow id for load sharing");
264 
265 /* Default value for flowid shift */
266 static VNET_DEFINE(int, def_flowid_shift) = 16;
267 #define	V_def_flowid_shift	VNET(def_flowid_shift)
268 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
269     &VNET_NAME(def_flowid_shift), 0,
270     "Default setting for flowid shift for load sharing");
271 
272 static void
273 vnet_lagg_init(const void *unused __unused)
274 {
275 
276 	LAGG_LIST_LOCK_INIT();
277 	SLIST_INIT(&V_lagg_list);
278 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
279 	    lagg_clone_destroy, 0);
280 }
281 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
282     vnet_lagg_init, NULL);
283 
284 static void
285 vnet_lagg_uninit(const void *unused __unused)
286 {
287 
288 	if_clone_detach(V_lagg_cloner);
289 	LAGG_LIST_LOCK_DESTROY();
290 }
291 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
292     vnet_lagg_uninit, NULL);
293 
294 static int
295 lagg_modevent(module_t mod, int type, void *data)
296 {
297 
298 	switch (type) {
299 	case MOD_LOAD:
300 		lagg_input_p = lagg_input;
301 		lagg_linkstate_p = lagg_port_state;
302 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
303 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
304 		    EVENTHANDLER_PRI_ANY);
305 		break;
306 	case MOD_UNLOAD:
307 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
308 		    lagg_detach_cookie);
309 		lagg_input_p = NULL;
310 		lagg_linkstate_p = NULL;
311 		break;
312 	default:
313 		return (EOPNOTSUPP);
314 	}
315 	return (0);
316 }
317 
318 static moduledata_t lagg_mod = {
319 	"if_lagg",
320 	lagg_modevent,
321 	0
322 };
323 
324 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
325 MODULE_VERSION(if_lagg, 1);
326 
327 static void
328 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
329 {
330 
331 	LAGG_XLOCK_ASSERT(sc);
332 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
333 	    __func__, sc));
334 
335 	if (sc->sc_ifflags & IFF_DEBUG)
336 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
337 
338 	if (lagg_protos[pr].pr_attach != NULL)
339 		lagg_protos[pr].pr_attach(sc);
340 	sc->sc_proto = pr;
341 }
342 
343 static void
344 lagg_proto_detach(struct lagg_softc *sc)
345 {
346 	lagg_proto pr;
347 
348 	LAGG_XLOCK_ASSERT(sc);
349 	pr = sc->sc_proto;
350 	sc->sc_proto = LAGG_PROTO_NONE;
351 
352 	if (lagg_protos[pr].pr_detach != NULL)
353 		lagg_protos[pr].pr_detach(sc);
354 }
355 
356 static int
357 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
358 {
359 
360 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
361 }
362 
363 static struct mbuf *
364 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
365 {
366 
367 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
368 }
369 
370 static int
371 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
372 {
373 
374 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
375 		return (0);
376 	else
377 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
378 }
379 
380 static void
381 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
382 {
383 
384 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
385 		lagg_protos[sc->sc_proto].pr_delport(lp);
386 }
387 
388 static void
389 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
390 {
391 
392 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
393 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
394 }
395 
396 static void
397 lagg_proto_init(struct lagg_softc *sc)
398 {
399 
400 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
401 		lagg_protos[sc->sc_proto].pr_init(sc);
402 }
403 
404 static void
405 lagg_proto_stop(struct lagg_softc *sc)
406 {
407 
408 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
409 		lagg_protos[sc->sc_proto].pr_stop(sc);
410 }
411 
412 static void
413 lagg_proto_lladdr(struct lagg_softc *sc)
414 {
415 
416 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
417 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
418 }
419 
420 static void
421 lagg_proto_request(struct lagg_softc *sc, void *v)
422 {
423 
424 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
425 		lagg_protos[sc->sc_proto].pr_request(sc, v);
426 }
427 
428 static void
429 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
430 {
431 
432 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
433 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
434 }
435 
436 /*
437  * This routine is run via an vlan
438  * config EVENT
439  */
440 static void
441 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
442 {
443 	struct lagg_softc *sc = ifp->if_softc;
444 	struct lagg_port *lp;
445 
446 	if (ifp->if_softc !=  arg)   /* Not our event */
447 		return;
448 
449 	LAGG_RLOCK();
450 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
451 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
452 	LAGG_RUNLOCK();
453 }
454 
455 /*
456  * This routine is run via an vlan
457  * unconfig EVENT
458  */
459 static void
460 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
461 {
462 	struct lagg_softc *sc = ifp->if_softc;
463 	struct lagg_port *lp;
464 
465 	if (ifp->if_softc !=  arg)   /* Not our event */
466 		return;
467 
468 	LAGG_RLOCK();
469 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
470 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
471 	LAGG_RUNLOCK();
472 }
473 
474 static int
475 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
476 {
477 	struct lagg_softc *sc;
478 	struct ifnet *ifp;
479 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
480 
481 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
482 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
483 	if (ifp == NULL) {
484 		free(sc, M_DEVBUF);
485 		return (ENOSPC);
486 	}
487 	LAGG_SX_INIT(sc);
488 
489 	LAGG_XLOCK(sc);
490 	if (V_def_use_flowid)
491 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
492 	sc->flowid_shift = V_def_flowid_shift;
493 
494 	/* Hash all layers by default */
495 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
496 
497 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
498 
499 	CK_SLIST_INIT(&sc->sc_ports);
500 
501 	/* Initialise pseudo media types */
502 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
503 	    lagg_media_status);
504 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
505 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
506 
507 	if_initname(ifp, laggname, unit);
508 	ifp->if_softc = sc;
509 	ifp->if_transmit = lagg_transmit;
510 	ifp->if_qflush = lagg_qflush;
511 	ifp->if_init = lagg_init;
512 	ifp->if_ioctl = lagg_ioctl;
513 	ifp->if_get_counter = lagg_get_counter;
514 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
515 #ifdef RATELIMIT
516 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
517 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS | IFCAP_TXRTLMT;
518 #else
519 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
520 #endif
521 
522 	/*
523 	 * Attach as an ordinary ethernet device, children will be attached
524 	 * as special device IFT_IEEE8023ADLAG.
525 	 */
526 	ether_ifattach(ifp, eaddr);
527 
528 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
529 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
530 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
531 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
532 
533 	/* Insert into the global list of laggs */
534 	LAGG_LIST_LOCK();
535 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
536 	LAGG_LIST_UNLOCK();
537 	LAGG_XUNLOCK(sc);
538 
539 	return (0);
540 }
541 
542 static void
543 lagg_clone_destroy(struct ifnet *ifp)
544 {
545 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
546 	struct lagg_port *lp;
547 
548 	LAGG_XLOCK(sc);
549 	sc->sc_destroying = 1;
550 	lagg_stop(sc);
551 	ifp->if_flags &= ~IFF_UP;
552 
553 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
554 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
555 
556 	/* Shutdown and remove lagg ports */
557 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
558 		lagg_port_destroy(lp, 1);
559 
560 	/* Unhook the aggregation protocol */
561 	lagg_proto_detach(sc);
562 	LAGG_XUNLOCK(sc);
563 
564 	ifmedia_removeall(&sc->sc_media);
565 	ether_ifdetach(ifp);
566 	if_free(ifp);
567 
568 	LAGG_LIST_LOCK();
569 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
570 	LAGG_LIST_UNLOCK();
571 
572 	LAGG_SX_DESTROY(sc);
573 	free(sc, M_DEVBUF);
574 }
575 
576 static void
577 lagg_capabilities(struct lagg_softc *sc)
578 {
579 	struct lagg_port *lp;
580 	int cap, ena, pena;
581 	uint64_t hwa;
582 	struct ifnet_hw_tsomax hw_tsomax;
583 
584 	LAGG_XLOCK_ASSERT(sc);
585 
586 	/* Get common enabled capabilities for the lagg ports */
587 	ena = ~0;
588 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
589 		ena &= lp->lp_ifp->if_capenable;
590 	ena = (ena == ~0 ? 0 : ena);
591 
592 	/*
593 	 * Apply common enabled capabilities back to the lagg ports.
594 	 * May require several iterations if they are dependent.
595 	 */
596 	do {
597 		pena = ena;
598 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
599 			lagg_setcaps(lp, ena);
600 			ena &= lp->lp_ifp->if_capenable;
601 		}
602 	} while (pena != ena);
603 
604 	/* Get other capabilities from the lagg ports */
605 	cap = ~0;
606 	hwa = ~(uint64_t)0;
607 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
608 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
609 		cap &= lp->lp_ifp->if_capabilities;
610 		hwa &= lp->lp_ifp->if_hwassist;
611 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
612 	}
613 	cap = (cap == ~0 ? 0 : cap);
614 	hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
615 
616 	if (sc->sc_ifp->if_capabilities != cap ||
617 	    sc->sc_ifp->if_capenable != ena ||
618 	    sc->sc_ifp->if_hwassist != hwa ||
619 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
620 		sc->sc_ifp->if_capabilities = cap;
621 		sc->sc_ifp->if_capenable = ena;
622 		sc->sc_ifp->if_hwassist = hwa;
623 		getmicrotime(&sc->sc_ifp->if_lastchange);
624 
625 		if (sc->sc_ifflags & IFF_DEBUG)
626 			if_printf(sc->sc_ifp,
627 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
628 	}
629 }
630 
631 static int
632 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
633 {
634 	struct lagg_softc *sc_ptr;
635 	struct lagg_port *lp, *tlp;
636 	int error, i;
637 	uint64_t *pval;
638 
639 	LAGG_XLOCK_ASSERT(sc);
640 
641 	/* Limit the maximal number of lagg ports */
642 	if (sc->sc_count >= LAGG_MAX_PORTS)
643 		return (ENOSPC);
644 
645 	/* Check if port has already been associated to a lagg */
646 	if (ifp->if_lagg != NULL) {
647 		/* Port is already in the current lagg? */
648 		lp = (struct lagg_port *)ifp->if_lagg;
649 		if (lp->lp_softc == sc)
650 			return (EEXIST);
651 		return (EBUSY);
652 	}
653 
654 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
655 	if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
656 		return (EPROTONOSUPPORT);
657 
658 	/* Allow the first Ethernet member to define the MTU */
659 	if (CK_SLIST_EMPTY(&sc->sc_ports))
660 		sc->sc_ifp->if_mtu = ifp->if_mtu;
661 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
662 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
663 		    ifp->if_xname);
664 		return (EINVAL);
665 	}
666 
667 	lp = malloc(sizeof(struct lagg_port), M_DEVBUF, M_WAITOK|M_ZERO);
668 	lp->lp_softc = sc;
669 
670 	/* Check if port is a stacked lagg */
671 	LAGG_LIST_LOCK();
672 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
673 		if (ifp == sc_ptr->sc_ifp) {
674 			LAGG_LIST_UNLOCK();
675 			free(lp, M_DEVBUF);
676 			return (EINVAL);
677 			/* XXX disable stacking for the moment, its untested */
678 #ifdef LAGG_PORT_STACKING
679 			lp->lp_flags |= LAGG_PORT_STACK;
680 			if (lagg_port_checkstacking(sc_ptr) >=
681 			    LAGG_MAX_STACKING) {
682 				LAGG_LIST_UNLOCK();
683 				free(lp, M_DEVBUF);
684 				return (E2BIG);
685 			}
686 #endif
687 		}
688 	}
689 	LAGG_LIST_UNLOCK();
690 
691 	if_ref(ifp);
692 	lp->lp_ifp = ifp;
693 
694 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
695 	lp->lp_ifcapenable = ifp->if_capenable;
696 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
697 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
698 		lagg_proto_lladdr(sc);
699 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
700 	} else {
701 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
702 	}
703 	lagg_setflags(lp, 1);
704 
705 	if (CK_SLIST_EMPTY(&sc->sc_ports))
706 		sc->sc_primary = lp;
707 
708 	/* Change the interface type */
709 	lp->lp_iftype = ifp->if_type;
710 	ifp->if_type = IFT_IEEE8023ADLAG;
711 	ifp->if_lagg = lp;
712 	lp->lp_ioctl = ifp->if_ioctl;
713 	ifp->if_ioctl = lagg_port_ioctl;
714 	lp->lp_output = ifp->if_output;
715 	ifp->if_output = lagg_port_output;
716 
717 	/* Read port counters */
718 	pval = lp->port_counters.val;
719 	for (i = 0; i < IFCOUNTERS; i++, pval++)
720 		*pval = ifp->if_get_counter(ifp, i);
721 
722 	/*
723 	 * Insert into the list of ports.
724 	 * Keep ports sorted by if_index. It is handy, when configuration
725 	 * is predictable and `ifconfig laggN create ...` command
726 	 * will lead to the same result each time.
727 	 */
728 	LAGG_RLOCK();
729 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
730 		if (tlp->lp_ifp->if_index < ifp->if_index && (
731 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
732 		    ((struct  lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
733 		    ifp->if_index))
734 			break;
735 	}
736 	LAGG_RUNLOCK();
737 	if (tlp != NULL)
738 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
739 	else
740 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
741 	sc->sc_count++;
742 
743 	lagg_setmulti(lp);
744 
745 
746 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
747 		/* Remove the port, without calling pr_delport. */
748 		lagg_port_destroy(lp, 0);
749 		return (error);
750 	}
751 
752 	/* Update lagg capabilities */
753 	lagg_capabilities(sc);
754 	lagg_linkstate(sc);
755 
756 	return (0);
757 }
758 
759 #ifdef LAGG_PORT_STACKING
760 static int
761 lagg_port_checkstacking(struct lagg_softc *sc)
762 {
763 	struct lagg_softc *sc_ptr;
764 	struct lagg_port *lp;
765 	int m = 0;
766 
767 	LAGG_SXLOCK_ASSERT(sc);
768 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
769 		if (lp->lp_flags & LAGG_PORT_STACK) {
770 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
771 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
772 		}
773 	}
774 
775 	return (m + 1);
776 }
777 #endif
778 
779 static void
780 lagg_port_destroy_cb(epoch_context_t ec)
781 {
782 	struct lagg_port *lp;
783 	struct ifnet *ifp;
784 
785 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
786 	ifp = lp->lp_ifp;
787 
788 	if_rele(ifp);
789 	free(lp, M_DEVBUF);
790 }
791 
792 static int
793 lagg_port_destroy(struct lagg_port *lp, int rundelport)
794 {
795 	struct lagg_softc *sc = lp->lp_softc;
796 	struct lagg_port *lp_ptr, *lp0;
797 	struct ifnet *ifp = lp->lp_ifp;
798 	uint64_t *pval, vdiff;
799 	int i;
800 
801 	LAGG_XLOCK_ASSERT(sc);
802 
803 	if (rundelport)
804 		lagg_proto_delport(sc, lp);
805 
806 	if (lp->lp_detaching == 0)
807 		lagg_clrmulti(lp);
808 
809 	/* Restore interface */
810 	ifp->if_type = lp->lp_iftype;
811 	ifp->if_ioctl = lp->lp_ioctl;
812 	ifp->if_output = lp->lp_output;
813 	ifp->if_lagg = NULL;
814 
815 	/* Update detached port counters */
816 	pval = lp->port_counters.val;
817 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
818 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
819 		sc->detached_counters.val[i] += vdiff;
820 	}
821 
822 	/* Finally, remove the port from the lagg */
823 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
824 	sc->sc_count--;
825 
826 	/* Update the primary interface */
827 	if (lp == sc->sc_primary) {
828 		uint8_t lladdr[ETHER_ADDR_LEN];
829 
830 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
831 			bzero(&lladdr, ETHER_ADDR_LEN);
832 		else
833 			bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN);
834 		sc->sc_primary = lp0;
835 		if (sc->sc_destroying == 0) {
836 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
837 			lagg_proto_lladdr(sc);
838 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
839 		}
840 
841 		/*
842 		 * Update lladdr for each port (new primary needs update
843 		 * as well, to switch from old lladdr to its 'real' one)
844 		 */
845 		CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
846 			if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN);
847 	}
848 
849 	if (lp->lp_ifflags)
850 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
851 
852 	if (lp->lp_detaching == 0) {
853 		lagg_setflags(lp, 0);
854 		lagg_setcaps(lp, lp->lp_ifcapenable);
855 		if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN);
856 	}
857 
858 	/*
859 	 * free port and release it's ifnet reference after a grace period has
860 	 * elapsed.
861 	 */
862 	epoch_call(net_epoch_preempt, &lp->lp_epoch_ctx, lagg_port_destroy_cb);
863 	/* Update lagg capabilities */
864 	lagg_capabilities(sc);
865 	lagg_linkstate(sc);
866 
867 	return (0);
868 }
869 
870 static int
871 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
872 {
873 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
874 	struct lagg_softc *sc;
875 	struct lagg_port *lp = NULL;
876 	int error = 0;
877 
878 	/* Should be checked by the caller */
879 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
880 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
881 		goto fallback;
882 
883 	switch (cmd) {
884 	case SIOCGLAGGPORT:
885 		if (rp->rp_portname[0] == '\0' ||
886 		    ifunit(rp->rp_portname) != ifp) {
887 			error = EINVAL;
888 			break;
889 		}
890 
891 		LAGG_RLOCK();
892 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
893 			error = ENOENT;
894 			LAGG_RUNLOCK();
895 			break;
896 		}
897 
898 		lagg_port2req(lp, rp);
899 		LAGG_RUNLOCK();
900 		break;
901 
902 	case SIOCSIFCAP:
903 		if (lp->lp_ioctl == NULL) {
904 			error = EINVAL;
905 			break;
906 		}
907 		error = (*lp->lp_ioctl)(ifp, cmd, data);
908 		if (error)
909 			break;
910 
911 		/* Update lagg interface capabilities */
912 		LAGG_XLOCK(sc);
913 		lagg_capabilities(sc);
914 		LAGG_XUNLOCK(sc);
915 		VLAN_CAPABILITIES(sc->sc_ifp);
916 		break;
917 
918 	case SIOCSIFMTU:
919 		/* Do not allow the MTU to be changed once joined */
920 		error = EINVAL;
921 		break;
922 
923 	default:
924 		goto fallback;
925 	}
926 
927 	return (error);
928 
929 fallback:
930 	if (lp != NULL && lp->lp_ioctl != NULL)
931 		return ((*lp->lp_ioctl)(ifp, cmd, data));
932 
933 	return (EINVAL);
934 }
935 
936 /*
937  * Requests counter @cnt data.
938  *
939  * Counter value is calculated the following way:
940  * 1) for each port, sum  difference between current and "initial" measurements.
941  * 2) add lagg logical interface counters.
942  * 3) add data from detached_counters array.
943  *
944  * We also do the following things on ports attach/detach:
945  * 1) On port attach we store all counters it has into port_counter array.
946  * 2) On port detach we add the different between "initial" and
947  *   current counters data to detached_counters array.
948  */
949 static uint64_t
950 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
951 {
952 	struct lagg_softc *sc;
953 	struct lagg_port *lp;
954 	struct ifnet *lpifp;
955 	uint64_t newval, oldval, vsum;
956 
957 	/* Revise this when we've got non-generic counters. */
958 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
959 
960 	sc = (struct lagg_softc *)ifp->if_softc;
961 
962 	vsum = 0;
963 	LAGG_RLOCK();
964 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
965 		/* Saved attached value */
966 		oldval = lp->port_counters.val[cnt];
967 		/* current value */
968 		lpifp = lp->lp_ifp;
969 		newval = lpifp->if_get_counter(lpifp, cnt);
970 		/* Calculate diff and save new */
971 		vsum += newval - oldval;
972 	}
973 	LAGG_RUNLOCK();
974 
975 	/*
976 	 * Add counter data which might be added by upper
977 	 * layer protocols operating on logical interface.
978 	 */
979 	vsum += if_get_counter_default(ifp, cnt);
980 
981 	/*
982 	 * Add counter data from detached ports counters
983 	 */
984 	vsum += sc->detached_counters.val[cnt];
985 
986 
987 	return (vsum);
988 }
989 
990 /*
991  * For direct output to child ports.
992  */
993 static int
994 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
995 	const struct sockaddr *dst, struct route *ro)
996 {
997 	struct lagg_port *lp = ifp->if_lagg;
998 
999 	switch (dst->sa_family) {
1000 		case pseudo_AF_HDRCMPLT:
1001 		case AF_UNSPEC:
1002 			return ((*lp->lp_output)(ifp, m, dst, ro));
1003 	}
1004 
1005 	/* drop any other frames */
1006 	m_freem(m);
1007 	return (ENETDOWN);
1008 }
1009 
1010 static void
1011 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1012 {
1013 	struct lagg_port *lp;
1014 	struct lagg_softc *sc;
1015 
1016 	if ((lp = ifp->if_lagg) == NULL)
1017 		return;
1018 	/* If the ifnet is just being renamed, don't do anything. */
1019 	if (ifp->if_flags & IFF_RENAMING)
1020 		return;
1021 
1022 	sc = lp->lp_softc;
1023 
1024 	LAGG_XLOCK(sc);
1025 	lp->lp_detaching = 1;
1026 	lagg_port_destroy(lp, 1);
1027 	LAGG_XUNLOCK(sc);
1028 	VLAN_CAPABILITIES(sc->sc_ifp);
1029 }
1030 
1031 static void
1032 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1033 {
1034 	struct lagg_softc *sc = lp->lp_softc;
1035 
1036 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1037 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1038 	rp->rp_prio = lp->lp_prio;
1039 	rp->rp_flags = lp->lp_flags;
1040 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1041 
1042 	/* Add protocol specific flags */
1043 	switch (sc->sc_proto) {
1044 		case LAGG_PROTO_FAILOVER:
1045 			if (lp == sc->sc_primary)
1046 				rp->rp_flags |= LAGG_PORT_MASTER;
1047 			if (lp == lagg_link_active(sc, sc->sc_primary))
1048 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1049 			break;
1050 
1051 		case LAGG_PROTO_ROUNDROBIN:
1052 		case LAGG_PROTO_LOADBALANCE:
1053 		case LAGG_PROTO_BROADCAST:
1054 			if (LAGG_PORTACTIVE(lp))
1055 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1056 			break;
1057 
1058 		case LAGG_PROTO_LACP:
1059 			/* LACP has a different definition of active */
1060 			if (lacp_isactive(lp))
1061 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1062 			if (lacp_iscollecting(lp))
1063 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1064 			if (lacp_isdistributing(lp))
1065 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1066 			break;
1067 	}
1068 
1069 }
1070 
1071 static void
1072 lagg_init(void *xsc)
1073 {
1074 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1075 	struct ifnet *ifp = sc->sc_ifp;
1076 	struct lagg_port *lp;
1077 
1078 	LAGG_XLOCK(sc);
1079 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1080 		LAGG_XUNLOCK(sc);
1081 		return;
1082 	}
1083 
1084 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1085 
1086 	/*
1087 	 * Update the port lladdrs if needed.
1088 	 * This might be if_setlladdr() notification
1089 	 * that lladdr has been changed.
1090 	 */
1091 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1092 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1093 		    ETHER_ADDR_LEN) != 0)
1094 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN);
1095 	}
1096 
1097 	lagg_proto_init(sc);
1098 
1099 	LAGG_XUNLOCK(sc);
1100 }
1101 
1102 static void
1103 lagg_stop(struct lagg_softc *sc)
1104 {
1105 	struct ifnet *ifp = sc->sc_ifp;
1106 
1107 	LAGG_XLOCK_ASSERT(sc);
1108 
1109 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1110 		return;
1111 
1112 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1113 
1114 	lagg_proto_stop(sc);
1115 }
1116 
1117 static int
1118 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1119 {
1120 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1121 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1122 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1123 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1124 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1125 	struct ifreq *ifr = (struct ifreq *)data;
1126 	struct lagg_port *lp;
1127 	struct ifnet *tpif;
1128 	struct thread *td = curthread;
1129 	char *buf, *outbuf;
1130 	int count, buflen, len, error = 0;
1131 
1132 	bzero(&rpbuf, sizeof(rpbuf));
1133 
1134 	switch (cmd) {
1135 	case SIOCGLAGG:
1136 		LAGG_XLOCK(sc);
1137 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1138 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1139 		ra->ra_proto = sc->sc_proto;
1140 		lagg_proto_request(sc, &ra->ra_psc);
1141 		count = 0;
1142 		buf = outbuf;
1143 		len = min(ra->ra_size, buflen);
1144 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1145 			if (len < sizeof(rpbuf))
1146 				break;
1147 
1148 			lagg_port2req(lp, &rpbuf);
1149 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1150 			count++;
1151 			buf += sizeof(rpbuf);
1152 			len -= sizeof(rpbuf);
1153 		}
1154 		LAGG_XUNLOCK(sc);
1155 		ra->ra_ports = count;
1156 		ra->ra_size = count * sizeof(rpbuf);
1157 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1158 		free(outbuf, M_TEMP);
1159 		break;
1160 	case SIOCSLAGG:
1161 		error = priv_check(td, PRIV_NET_LAGG);
1162 		if (error)
1163 			break;
1164 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1165 			error = EPROTONOSUPPORT;
1166 			break;
1167 		}
1168 
1169 		LAGG_XLOCK(sc);
1170 		lagg_proto_detach(sc);
1171 		LAGG_UNLOCK_ASSERT();
1172 		lagg_proto_attach(sc, ra->ra_proto);
1173 		LAGG_XUNLOCK(sc);
1174 		break;
1175 	case SIOCGLAGGOPTS:
1176 		LAGG_XLOCK(sc);
1177 		ro->ro_opts = sc->sc_opts;
1178 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1179 			struct lacp_softc *lsc;
1180 
1181 			lsc = (struct lacp_softc *)sc->sc_psc;
1182 			if (lsc->lsc_debug.lsc_tx_test != 0)
1183 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1184 			if (lsc->lsc_debug.lsc_rx_test != 0)
1185 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1186 			if (lsc->lsc_strict_mode != 0)
1187 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1188 			if (lsc->lsc_fast_timeout != 0)
1189 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1190 
1191 			ro->ro_active = sc->sc_active;
1192 		} else {
1193 			ro->ro_active = 0;
1194 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1195 				ro->ro_active += LAGG_PORTACTIVE(lp);
1196 		}
1197 		ro->ro_bkt = sc->sc_bkt;
1198 		ro->ro_flapping = sc->sc_flapping;
1199 		ro->ro_flowid_shift = sc->flowid_shift;
1200 		LAGG_XUNLOCK(sc);
1201 		break;
1202 	case SIOCSLAGGOPTS:
1203 		if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) {
1204 			if (ro->ro_bkt == 0)
1205 				sc->sc_bkt = 1; // Minimum 1 packet per iface.
1206 			else
1207 				sc->sc_bkt = ro->ro_bkt;
1208 		}
1209 		error = priv_check(td, PRIV_NET_LAGG);
1210 		if (error)
1211 			break;
1212 		if (ro->ro_opts == 0)
1213 			break;
1214 		/*
1215 		 * Set options.  LACP options are stored in sc->sc_psc,
1216 		 * not in sc_opts.
1217 		 */
1218 		int valid, lacp;
1219 
1220 		switch (ro->ro_opts) {
1221 		case LAGG_OPT_USE_FLOWID:
1222 		case -LAGG_OPT_USE_FLOWID:
1223 		case LAGG_OPT_FLOWIDSHIFT:
1224 			valid = 1;
1225 			lacp = 0;
1226 			break;
1227 		case LAGG_OPT_LACP_TXTEST:
1228 		case -LAGG_OPT_LACP_TXTEST:
1229 		case LAGG_OPT_LACP_RXTEST:
1230 		case -LAGG_OPT_LACP_RXTEST:
1231 		case LAGG_OPT_LACP_STRICT:
1232 		case -LAGG_OPT_LACP_STRICT:
1233 		case LAGG_OPT_LACP_TIMEOUT:
1234 		case -LAGG_OPT_LACP_TIMEOUT:
1235 			valid = lacp = 1;
1236 			break;
1237 		default:
1238 			valid = lacp = 0;
1239 			break;
1240 		}
1241 
1242 		LAGG_XLOCK(sc);
1243 
1244 		if (valid == 0 ||
1245 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1246 			/* Invalid combination of options specified. */
1247 			error = EINVAL;
1248 			LAGG_XUNLOCK(sc);
1249 			break;	/* Return from SIOCSLAGGOPTS. */
1250 		}
1251 		/*
1252 		 * Store new options into sc->sc_opts except for
1253 		 * FLOWIDSHIFT and LACP options.
1254 		 */
1255 		if (lacp == 0) {
1256 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1257 				sc->flowid_shift = ro->ro_flowid_shift;
1258 			else if (ro->ro_opts > 0)
1259 				sc->sc_opts |= ro->ro_opts;
1260 			else
1261 				sc->sc_opts &= ~ro->ro_opts;
1262 		} else {
1263 			struct lacp_softc *lsc;
1264 			struct lacp_port *lp;
1265 
1266 			lsc = (struct lacp_softc *)sc->sc_psc;
1267 
1268 			switch (ro->ro_opts) {
1269 			case LAGG_OPT_LACP_TXTEST:
1270 				lsc->lsc_debug.lsc_tx_test = 1;
1271 				break;
1272 			case -LAGG_OPT_LACP_TXTEST:
1273 				lsc->lsc_debug.lsc_tx_test = 0;
1274 				break;
1275 			case LAGG_OPT_LACP_RXTEST:
1276 				lsc->lsc_debug.lsc_rx_test = 1;
1277 				break;
1278 			case -LAGG_OPT_LACP_RXTEST:
1279 				lsc->lsc_debug.lsc_rx_test = 0;
1280 				break;
1281 			case LAGG_OPT_LACP_STRICT:
1282 				lsc->lsc_strict_mode = 1;
1283 				break;
1284 			case -LAGG_OPT_LACP_STRICT:
1285 				lsc->lsc_strict_mode = 0;
1286 				break;
1287 			case LAGG_OPT_LACP_TIMEOUT:
1288 				LACP_LOCK(lsc);
1289         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1290                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1291 				LACP_UNLOCK(lsc);
1292 				lsc->lsc_fast_timeout = 1;
1293 				break;
1294 			case -LAGG_OPT_LACP_TIMEOUT:
1295 				LACP_LOCK(lsc);
1296         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1297                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1298 				LACP_UNLOCK(lsc);
1299 				lsc->lsc_fast_timeout = 0;
1300 				break;
1301 			}
1302 		}
1303 		LAGG_XUNLOCK(sc);
1304 		break;
1305 	case SIOCGLAGGFLAGS:
1306 		rf->rf_flags = 0;
1307 		LAGG_XLOCK(sc);
1308 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1309 			rf->rf_flags |= LAGG_F_HASHL2;
1310 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1311 			rf->rf_flags |= LAGG_F_HASHL3;
1312 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1313 			rf->rf_flags |= LAGG_F_HASHL4;
1314 		LAGG_XUNLOCK(sc);
1315 		break;
1316 	case SIOCSLAGGHASH:
1317 		error = priv_check(td, PRIV_NET_LAGG);
1318 		if (error)
1319 			break;
1320 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1321 			error = EINVAL;
1322 			break;
1323 		}
1324 		LAGG_XLOCK(sc);
1325 		sc->sc_flags = 0;
1326 		if (rf->rf_flags & LAGG_F_HASHL2)
1327 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1328 		if (rf->rf_flags & LAGG_F_HASHL3)
1329 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1330 		if (rf->rf_flags & LAGG_F_HASHL4)
1331 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1332 		LAGG_XUNLOCK(sc);
1333 		break;
1334 	case SIOCGLAGGPORT:
1335 		if (rp->rp_portname[0] == '\0' ||
1336 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1337 			error = EINVAL;
1338 			break;
1339 		}
1340 
1341 		LAGG_RLOCK();
1342 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1343 		    lp->lp_softc != sc) {
1344 			error = ENOENT;
1345 			LAGG_RUNLOCK();
1346 			if_rele(tpif);
1347 			break;
1348 		}
1349 
1350 		lagg_port2req(lp, rp);
1351 		LAGG_RUNLOCK();
1352 		if_rele(tpif);
1353 		break;
1354 	case SIOCSLAGGPORT:
1355 		error = priv_check(td, PRIV_NET_LAGG);
1356 		if (error)
1357 			break;
1358 		if (rp->rp_portname[0] == '\0' ||
1359 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1360 			error = EINVAL;
1361 			break;
1362 		}
1363 #ifdef INET6
1364 		/*
1365 		 * A laggport interface should not have inet6 address
1366 		 * because two interfaces with a valid link-local
1367 		 * scope zone must not be merged in any form.  This
1368 		 * restriction is needed to prevent violation of
1369 		 * link-local scope zone.  Attempts to add a laggport
1370 		 * interface which has inet6 addresses triggers
1371 		 * removal of all inet6 addresses on the member
1372 		 * interface.
1373 		 */
1374 		if (in6ifa_llaonifp(tpif)) {
1375 			in6_ifdetach(tpif);
1376 				if_printf(sc->sc_ifp,
1377 				    "IPv6 addresses on %s have been removed "
1378 				    "before adding it as a member to prevent "
1379 				    "IPv6 address scope violation.\n",
1380 				    tpif->if_xname);
1381 		}
1382 #endif
1383 		LAGG_XLOCK(sc);
1384 		error = lagg_port_create(sc, tpif);
1385 		LAGG_XUNLOCK(sc);
1386 		if_rele(tpif);
1387 		VLAN_CAPABILITIES(ifp);
1388 		break;
1389 	case SIOCSLAGGDELPORT:
1390 		error = priv_check(td, PRIV_NET_LAGG);
1391 		if (error)
1392 			break;
1393 		if (rp->rp_portname[0] == '\0' ||
1394 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1395 			error = EINVAL;
1396 			break;
1397 		}
1398 
1399 		LAGG_XLOCK(sc);
1400 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1401 		    lp->lp_softc != sc) {
1402 			error = ENOENT;
1403 			LAGG_XUNLOCK(sc);
1404 			if_rele(tpif);
1405 			break;
1406 		}
1407 
1408 		error = lagg_port_destroy(lp, 1);
1409 		LAGG_XUNLOCK(sc);
1410 		if_rele(tpif);
1411 		VLAN_CAPABILITIES(ifp);
1412 		break;
1413 	case SIOCSIFFLAGS:
1414 		/* Set flags on ports too */
1415 		LAGG_XLOCK(sc);
1416 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1417 			lagg_setflags(lp, 1);
1418 		}
1419 
1420 		if (!(ifp->if_flags & IFF_UP) &&
1421 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1422 			/*
1423 			 * If interface is marked down and it is running,
1424 			 * then stop and disable it.
1425 			 */
1426 			lagg_stop(sc);
1427 			LAGG_XUNLOCK(sc);
1428 		} else if ((ifp->if_flags & IFF_UP) &&
1429 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1430 			/*
1431 			 * If interface is marked up and it is stopped, then
1432 			 * start it.
1433 			 */
1434 			LAGG_XUNLOCK(sc);
1435 			(*ifp->if_init)(sc);
1436 		} else
1437 			LAGG_XUNLOCK(sc);
1438 		break;
1439 	case SIOCADDMULTI:
1440 	case SIOCDELMULTI:
1441 		LAGG_XLOCK(sc);
1442 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1443 			lagg_clrmulti(lp);
1444 			lagg_setmulti(lp);
1445 		}
1446 		LAGG_XUNLOCK(sc);
1447 		error = 0;
1448 		break;
1449 	case SIOCSIFMEDIA:
1450 	case SIOCGIFMEDIA:
1451 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1452 		break;
1453 
1454 	case SIOCSIFCAP:
1455 		LAGG_XLOCK(sc);
1456 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1457 			if (lp->lp_ioctl != NULL)
1458 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1459 		}
1460 		lagg_capabilities(sc);
1461 		LAGG_XUNLOCK(sc);
1462 		VLAN_CAPABILITIES(ifp);
1463 		error = 0;
1464 		break;
1465 
1466 	case SIOCSIFMTU:
1467 		/* Do not allow the MTU to be directly changed */
1468 		error = EINVAL;
1469 		break;
1470 
1471 	default:
1472 		error = ether_ioctl(ifp, cmd, data);
1473 		break;
1474 	}
1475 	return (error);
1476 }
1477 
1478 #ifdef RATELIMIT
1479 static int
1480 lagg_snd_tag_alloc(struct ifnet *ifp,
1481     union if_snd_tag_alloc_params *params,
1482     struct m_snd_tag **ppmt)
1483 {
1484 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1485 	struct lagg_port *lp;
1486 	struct lagg_lb *lb;
1487 	uint32_t p;
1488 
1489 	switch (sc->sc_proto) {
1490 	case LAGG_PROTO_FAILOVER:
1491 		lp = lagg_link_active(sc, sc->sc_primary);
1492 		break;
1493 	case LAGG_PROTO_LOADBALANCE:
1494 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1495 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1496 			return (EOPNOTSUPP);
1497 		p = params->hdr.flowid >> sc->flowid_shift;
1498 		p %= sc->sc_count;
1499 		lb = (struct lagg_lb *)sc->sc_psc;
1500 		lp = lb->lb_ports[p];
1501 		lp = lagg_link_active(sc, lp);
1502 		break;
1503 	case LAGG_PROTO_LACP:
1504 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1505 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1506 			return (EOPNOTSUPP);
1507 		lp = lacp_select_tx_port_by_hash(sc, params->hdr.flowid);
1508 		break;
1509 	default:
1510 		return (EOPNOTSUPP);
1511 	}
1512 	if (lp == NULL)
1513 		return (EOPNOTSUPP);
1514 	ifp = lp->lp_ifp;
1515 	if (ifp == NULL || ifp->if_snd_tag_alloc == NULL ||
1516 	    (ifp->if_capenable & IFCAP_TXRTLMT) == 0)
1517 		return (EOPNOTSUPP);
1518 
1519 	/* forward allocation request */
1520 	return (ifp->if_snd_tag_alloc(ifp, params, ppmt));
1521 }
1522 #endif
1523 
1524 static int
1525 lagg_setmulti(struct lagg_port *lp)
1526 {
1527 	struct lagg_softc *sc = lp->lp_softc;
1528 	struct ifnet *ifp = lp->lp_ifp;
1529 	struct ifnet *scifp = sc->sc_ifp;
1530 	struct lagg_mc *mc;
1531 	struct ifmultiaddr *ifma;
1532 	int error;
1533 
1534 	IF_ADDR_WLOCK(scifp);
1535 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1536 		if (ifma->ifma_addr->sa_family != AF_LINK)
1537 			continue;
1538 		mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1539 		if (mc == NULL) {
1540 			IF_ADDR_WUNLOCK(scifp);
1541 			return (ENOMEM);
1542 		}
1543 		bcopy(ifma->ifma_addr, &mc->mc_addr,
1544 		    ifma->ifma_addr->sa_len);
1545 		mc->mc_addr.sdl_index = ifp->if_index;
1546 		mc->mc_ifma = NULL;
1547 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1548 	}
1549 	IF_ADDR_WUNLOCK(scifp);
1550 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1551 		error = if_addmulti(ifp,
1552 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1553 		if (error)
1554 			return (error);
1555 	}
1556 	return (0);
1557 }
1558 
1559 static int
1560 lagg_clrmulti(struct lagg_port *lp)
1561 {
1562 	struct lagg_mc *mc;
1563 
1564 	LAGG_XLOCK_ASSERT(lp->lp_softc);
1565 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1566 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1567 		if (mc->mc_ifma && lp->lp_detaching == 0)
1568 			if_delmulti_ifma(mc->mc_ifma);
1569 		free(mc, M_DEVBUF);
1570 	}
1571 	return (0);
1572 }
1573 
1574 static int
1575 lagg_setcaps(struct lagg_port *lp, int cap)
1576 {
1577 	struct ifreq ifr;
1578 
1579 	if (lp->lp_ifp->if_capenable == cap)
1580 		return (0);
1581 	if (lp->lp_ioctl == NULL)
1582 		return (ENXIO);
1583 	ifr.ifr_reqcap = cap;
1584 	return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1585 }
1586 
1587 /* Handle a ref counted flag that should be set on the lagg port as well */
1588 static int
1589 lagg_setflag(struct lagg_port *lp, int flag, int status,
1590     int (*func)(struct ifnet *, int))
1591 {
1592 	struct lagg_softc *sc = lp->lp_softc;
1593 	struct ifnet *scifp = sc->sc_ifp;
1594 	struct ifnet *ifp = lp->lp_ifp;
1595 	int error;
1596 
1597 	LAGG_XLOCK_ASSERT(sc);
1598 
1599 	status = status ? (scifp->if_flags & flag) : 0;
1600 	/* Now "status" contains the flag value or 0 */
1601 
1602 	/*
1603 	 * See if recorded ports status is different from what
1604 	 * we want it to be.  If it is, flip it.  We record ports
1605 	 * status in lp_ifflags so that we won't clear ports flag
1606 	 * we haven't set.  In fact, we don't clear or set ports
1607 	 * flags directly, but get or release references to them.
1608 	 * That's why we can be sure that recorded flags still are
1609 	 * in accord with actual ports flags.
1610 	 */
1611 	if (status != (lp->lp_ifflags & flag)) {
1612 		error = (*func)(ifp, status);
1613 		if (error)
1614 			return (error);
1615 		lp->lp_ifflags &= ~flag;
1616 		lp->lp_ifflags |= status;
1617 	}
1618 	return (0);
1619 }
1620 
1621 /*
1622  * Handle IFF_* flags that require certain changes on the lagg port
1623  * if "status" is true, update ports flags respective to the lagg
1624  * if "status" is false, forcedly clear the flags set on port.
1625  */
1626 static int
1627 lagg_setflags(struct lagg_port *lp, int status)
1628 {
1629 	int error, i;
1630 
1631 	for (i = 0; lagg_pflags[i].flag; i++) {
1632 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1633 		    status, lagg_pflags[i].func);
1634 		if (error)
1635 			return (error);
1636 	}
1637 	return (0);
1638 }
1639 
1640 static int
1641 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1642 {
1643 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1644 	int error;
1645 
1646 	LAGG_RLOCK();
1647 	/* We need a Tx algorithm and at least one port */
1648 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1649 		LAGG_RUNLOCK();
1650 		m_freem(m);
1651 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1652 		return (ENXIO);
1653 	}
1654 
1655 	ETHER_BPF_MTAP(ifp, m);
1656 
1657 	error = lagg_proto_start(sc, m);
1658 	LAGG_RUNLOCK();
1659 
1660 	if (error != 0)
1661 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1662 
1663 	return (error);
1664 }
1665 
1666 /*
1667  * The ifp->if_qflush entry point for lagg(4) is no-op.
1668  */
1669 static void
1670 lagg_qflush(struct ifnet *ifp __unused)
1671 {
1672 }
1673 
1674 static struct mbuf *
1675 lagg_input(struct ifnet *ifp, struct mbuf *m)
1676 {
1677 	struct lagg_port *lp = ifp->if_lagg;
1678 	struct lagg_softc *sc = lp->lp_softc;
1679 	struct ifnet *scifp = sc->sc_ifp;
1680 
1681 	LAGG_RLOCK();
1682 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1683 	    lp->lp_detaching != 0 ||
1684 	    sc->sc_proto == LAGG_PROTO_NONE) {
1685 		LAGG_RUNLOCK();
1686 		m_freem(m);
1687 		return (NULL);
1688 	}
1689 
1690 	ETHER_BPF_MTAP(scifp, m);
1691 
1692 	m = lagg_proto_input(sc, lp, m);
1693 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
1694 		m_freem(m);
1695 		m = NULL;
1696 	}
1697 
1698 	LAGG_RUNLOCK();
1699 	return (m);
1700 }
1701 
1702 static int
1703 lagg_media_change(struct ifnet *ifp)
1704 {
1705 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1706 
1707 	if (sc->sc_ifflags & IFF_DEBUG)
1708 		printf("%s\n", __func__);
1709 
1710 	/* Ignore */
1711 	return (0);
1712 }
1713 
1714 static void
1715 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1716 {
1717 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1718 	struct lagg_port *lp;
1719 
1720 	imr->ifm_status = IFM_AVALID;
1721 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1722 
1723 	LAGG_RLOCK();
1724 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1725 		if (LAGG_PORTACTIVE(lp))
1726 			imr->ifm_status |= IFM_ACTIVE;
1727 	}
1728 	LAGG_RUNLOCK();
1729 }
1730 
1731 static void
1732 lagg_linkstate(struct lagg_softc *sc)
1733 {
1734 	struct lagg_port *lp;
1735 	int new_link = LINK_STATE_DOWN;
1736 	uint64_t speed;
1737 
1738 	LAGG_XLOCK_ASSERT(sc);
1739 
1740 	/* Our link is considered up if at least one of our ports is active */
1741 	LAGG_RLOCK();
1742 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1743 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1744 			new_link = LINK_STATE_UP;
1745 			break;
1746 		}
1747 	}
1748 	LAGG_RUNLOCK();
1749 	if_link_state_change(sc->sc_ifp, new_link);
1750 
1751 	/* Update if_baudrate to reflect the max possible speed */
1752 	switch (sc->sc_proto) {
1753 		case LAGG_PROTO_FAILOVER:
1754 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1755 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1756 			break;
1757 		case LAGG_PROTO_ROUNDROBIN:
1758 		case LAGG_PROTO_LOADBALANCE:
1759 		case LAGG_PROTO_BROADCAST:
1760 			speed = 0;
1761 			LAGG_RLOCK();
1762 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1763 				speed += lp->lp_ifp->if_baudrate;
1764 			LAGG_RUNLOCK();
1765 			sc->sc_ifp->if_baudrate = speed;
1766 			break;
1767 		case LAGG_PROTO_LACP:
1768 			/* LACP updates if_baudrate itself */
1769 			break;
1770 	}
1771 }
1772 
1773 static void
1774 lagg_port_state(struct ifnet *ifp, int state)
1775 {
1776 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1777 	struct lagg_softc *sc = NULL;
1778 
1779 	if (lp != NULL)
1780 		sc = lp->lp_softc;
1781 	if (sc == NULL)
1782 		return;
1783 
1784 	LAGG_XLOCK(sc);
1785 	lagg_linkstate(sc);
1786 	lagg_proto_linkstate(sc, lp);
1787 	LAGG_XUNLOCK(sc);
1788 }
1789 
1790 struct lagg_port *
1791 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1792 {
1793 	struct lagg_port *lp_next, *rval = NULL;
1794 
1795 	/*
1796 	 * Search a port which reports an active link state.
1797 	 */
1798 
1799 	if (lp == NULL)
1800 		goto search;
1801 	if (LAGG_PORTACTIVE(lp)) {
1802 		rval = lp;
1803 		goto found;
1804 	}
1805 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
1806 	    LAGG_PORTACTIVE(lp_next)) {
1807 		rval = lp_next;
1808 		goto found;
1809 	}
1810 
1811  search:
1812 	LAGG_RLOCK();
1813 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1814 		if (LAGG_PORTACTIVE(lp_next)) {
1815 			LAGG_RUNLOCK();
1816 			rval = lp_next;
1817 			goto found;
1818 		}
1819 	}
1820 	LAGG_RUNLOCK();
1821 found:
1822 	return (rval);
1823 }
1824 
1825 int
1826 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1827 {
1828 
1829 	return (ifp->if_transmit)(ifp, m);
1830 }
1831 
1832 /*
1833  * Simple round robin aggregation
1834  */
1835 static void
1836 lagg_rr_attach(struct lagg_softc *sc)
1837 {
1838 	sc->sc_seq = 0;
1839 	sc->sc_bkt_count = sc->sc_bkt;
1840 }
1841 
1842 static int
1843 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1844 {
1845 	struct lagg_port *lp;
1846 	uint32_t p;
1847 
1848 	if (sc->sc_bkt_count == 0 && sc->sc_bkt > 0)
1849 		sc->sc_bkt_count = sc->sc_bkt;
1850 
1851 	if (sc->sc_bkt > 0) {
1852 		atomic_subtract_int(&sc->sc_bkt_count, 1);
1853 	if (atomic_cmpset_int(&sc->sc_bkt_count, 0, sc->sc_bkt))
1854 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1855 	else
1856 		p = sc->sc_seq;
1857 	} else
1858 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1859 
1860 	p %= sc->sc_count;
1861 	lp = CK_SLIST_FIRST(&sc->sc_ports);
1862 
1863 	while (p--)
1864 		lp = CK_SLIST_NEXT(lp, lp_entries);
1865 
1866 	/*
1867 	 * Check the port's link state. This will return the next active
1868 	 * port if the link is down or the port is NULL.
1869 	 */
1870 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1871 		m_freem(m);
1872 		return (ENETDOWN);
1873 	}
1874 
1875 	/* Send mbuf */
1876 	return (lagg_enqueue(lp->lp_ifp, m));
1877 }
1878 
1879 static struct mbuf *
1880 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1881 {
1882 	struct ifnet *ifp = sc->sc_ifp;
1883 
1884 	/* Just pass in the packet to our lagg device */
1885 	m->m_pkthdr.rcvif = ifp;
1886 
1887 	return (m);
1888 }
1889 
1890 /*
1891  * Broadcast mode
1892  */
1893 static int
1894 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1895 {
1896 	int active_ports = 0;
1897 	int errors = 0;
1898 	int ret;
1899 	struct lagg_port *lp, *last = NULL;
1900 	struct mbuf *m0;
1901 
1902 	LAGG_RLOCK();
1903 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1904 		if (!LAGG_PORTACTIVE(lp))
1905 			continue;
1906 
1907 		active_ports++;
1908 
1909 		if (last != NULL) {
1910 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
1911 			if (m0 == NULL) {
1912 				ret = ENOBUFS;
1913 				errors++;
1914 				break;
1915 			}
1916 
1917 			ret = lagg_enqueue(last->lp_ifp, m0);
1918 			if (ret != 0)
1919 				errors++;
1920 		}
1921 		last = lp;
1922 	}
1923 	LAGG_RUNLOCK();
1924 
1925 	if (last == NULL) {
1926 		m_freem(m);
1927 		return (ENOENT);
1928 	}
1929 	if ((last = lagg_link_active(sc, last)) == NULL) {
1930 		m_freem(m);
1931 		return (ENETDOWN);
1932 	}
1933 
1934 	ret = lagg_enqueue(last->lp_ifp, m);
1935 	if (ret != 0)
1936 		errors++;
1937 
1938 	if (errors == 0)
1939 		return (ret);
1940 
1941 	return (0);
1942 }
1943 
1944 static struct mbuf*
1945 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1946 {
1947 	struct ifnet *ifp = sc->sc_ifp;
1948 
1949 	/* Just pass in the packet to our lagg device */
1950 	m->m_pkthdr.rcvif = ifp;
1951 	return (m);
1952 }
1953 
1954 /*
1955  * Active failover
1956  */
1957 static int
1958 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1959 {
1960 	struct lagg_port *lp;
1961 
1962 	/* Use the master port if active or the next available port */
1963 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1964 		m_freem(m);
1965 		return (ENETDOWN);
1966 	}
1967 
1968 	/* Send mbuf */
1969 	return (lagg_enqueue(lp->lp_ifp, m));
1970 }
1971 
1972 static struct mbuf *
1973 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1974 {
1975 	struct ifnet *ifp = sc->sc_ifp;
1976 	struct lagg_port *tmp_tp;
1977 
1978 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
1979 		m->m_pkthdr.rcvif = ifp;
1980 		return (m);
1981 	}
1982 
1983 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1984 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1985 		/*
1986 		 * If tmp_tp is null, we've received a packet when all
1987 		 * our links are down. Weird, but process it anyways.
1988 		 */
1989 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1990 			m->m_pkthdr.rcvif = ifp;
1991 			return (m);
1992 		}
1993 	}
1994 
1995 	m_freem(m);
1996 	return (NULL);
1997 }
1998 
1999 /*
2000  * Loadbalancing
2001  */
2002 static void
2003 lagg_lb_attach(struct lagg_softc *sc)
2004 {
2005 	struct lagg_port *lp;
2006 	struct lagg_lb *lb;
2007 
2008 	LAGG_XLOCK_ASSERT(sc);
2009 	lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
2010 	lb->lb_key = m_ether_tcpip_hash_init();
2011 	sc->sc_psc = lb;
2012 
2013 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2014 		lagg_lb_port_create(lp);
2015 }
2016 
2017 static void
2018 lagg_lb_detach(struct lagg_softc *sc)
2019 {
2020 	struct lagg_lb *lb;
2021 
2022 	lb = (struct lagg_lb *)sc->sc_psc;
2023 	if (lb != NULL)
2024 		free(lb, M_DEVBUF);
2025 }
2026 
2027 static int
2028 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2029 {
2030 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2031 	struct lagg_port *lp_next;
2032 	int i = 0;
2033 
2034 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2035 	LAGG_RLOCK();
2036 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2037 		if (lp_next == lp)
2038 			continue;
2039 		if (i >= LAGG_MAX_PORTS)
2040 			return (EINVAL);
2041 		if (sc->sc_ifflags & IFF_DEBUG)
2042 			printf("%s: port %s at index %d\n",
2043 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2044 		lb->lb_ports[i++] = lp_next;
2045 	}
2046 	LAGG_RUNLOCK();
2047 
2048 	return (0);
2049 }
2050 
2051 static int
2052 lagg_lb_port_create(struct lagg_port *lp)
2053 {
2054 	struct lagg_softc *sc = lp->lp_softc;
2055 	return (lagg_lb_porttable(sc, NULL));
2056 }
2057 
2058 static void
2059 lagg_lb_port_destroy(struct lagg_port *lp)
2060 {
2061 	struct lagg_softc *sc = lp->lp_softc;
2062 	lagg_lb_porttable(sc, lp);
2063 }
2064 
2065 static int
2066 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2067 {
2068 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2069 	struct lagg_port *lp = NULL;
2070 	uint32_t p = 0;
2071 
2072 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2073 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2074 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2075 	else
2076 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2077 	p %= sc->sc_count;
2078 	lp = lb->lb_ports[p];
2079 
2080 	/*
2081 	 * Check the port's link state. This will return the next active
2082 	 * port if the link is down or the port is NULL.
2083 	 */
2084 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2085 		m_freem(m);
2086 		return (ENETDOWN);
2087 	}
2088 
2089 	/* Send mbuf */
2090 	return (lagg_enqueue(lp->lp_ifp, m));
2091 }
2092 
2093 static struct mbuf *
2094 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2095 {
2096 	struct ifnet *ifp = sc->sc_ifp;
2097 
2098 	/* Just pass in the packet to our lagg device */
2099 	m->m_pkthdr.rcvif = ifp;
2100 
2101 	return (m);
2102 }
2103 
2104 /*
2105  * 802.3ad LACP
2106  */
2107 static void
2108 lagg_lacp_attach(struct lagg_softc *sc)
2109 {
2110 	struct lagg_port *lp;
2111 
2112 	lacp_attach(sc);
2113 	LAGG_XLOCK_ASSERT(sc);
2114 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2115 		lacp_port_create(lp);
2116 }
2117 
2118 static void
2119 lagg_lacp_detach(struct lagg_softc *sc)
2120 {
2121 	struct lagg_port *lp;
2122 	void *psc;
2123 
2124 	LAGG_XLOCK_ASSERT(sc);
2125 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2126 		lacp_port_destroy(lp);
2127 
2128 	psc = sc->sc_psc;
2129 	sc->sc_psc = NULL;
2130 	lacp_detach(psc);
2131 }
2132 
2133 static void
2134 lagg_lacp_lladdr(struct lagg_softc *sc)
2135 {
2136 	struct lagg_port *lp;
2137 
2138 	LAGG_SXLOCK_ASSERT(sc);
2139 
2140 	/* purge all the lacp ports */
2141 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2142 		lacp_port_destroy(lp);
2143 
2144 	/* add them back in */
2145 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2146 		lacp_port_create(lp);
2147 }
2148 
2149 static int
2150 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2151 {
2152 	struct lagg_port *lp;
2153 
2154 	lp = lacp_select_tx_port(sc, m);
2155 	if (lp == NULL) {
2156 		m_freem(m);
2157 		return (ENETDOWN);
2158 	}
2159 
2160 	/* Send mbuf */
2161 	return (lagg_enqueue(lp->lp_ifp, m));
2162 }
2163 
2164 static struct mbuf *
2165 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2166 {
2167 	struct ifnet *ifp = sc->sc_ifp;
2168 	struct ether_header *eh;
2169 	u_short etype;
2170 
2171 	eh = mtod(m, struct ether_header *);
2172 	etype = ntohs(eh->ether_type);
2173 
2174 	/* Tap off LACP control messages */
2175 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2176 		m = lacp_input(lp, m);
2177 		if (m == NULL)
2178 			return (NULL);
2179 	}
2180 
2181 	/*
2182 	 * If the port is not collecting or not in the active aggregator then
2183 	 * free and return.
2184 	 */
2185 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2186 		m_freem(m);
2187 		return (NULL);
2188 	}
2189 
2190 	m->m_pkthdr.rcvif = ifp;
2191 	return (m);
2192 }
2193 
2194