xref: /freebsd/sys/net/if_lagg.c (revision dacc43df34a7da82747af82be62cb645eb36f6ca)
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()	struct epoch_tracker lagg_et; epoch_enter_preempt(net_epoch_preempt, &lagg_et)
77 #define	LAGG_RUNLOCK()	epoch_exit_preempt(net_epoch_preempt, &lagg_et)
78 #define	LAGG_RLOCK_ASSERT()	MPASS(in_epoch(net_epoch_preempt))
79 #define	LAGG_UNLOCK_ASSERT()	MPASS(!in_epoch(net_epoch_preempt))
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 VNET_DEFINE_STATIC(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 VNET_DEFINE_STATIC(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 VNET_DEFINE_STATIC(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 VNET_DEFINE_STATIC(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 VNET_DEFINE_STATIC(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 	struct ifreq ifr;
637 	int error, i, oldmtu;
638 	uint64_t *pval;
639 
640 	LAGG_XLOCK_ASSERT(sc);
641 
642 	if (sc->sc_ifp == ifp) {
643 		if_printf(sc->sc_ifp,
644 		    "cannot add a lagg to itself as a port\n");
645 		return (EINVAL);
646 	}
647 
648 	/* Limit the maximal number of lagg ports */
649 	if (sc->sc_count >= LAGG_MAX_PORTS)
650 		return (ENOSPC);
651 
652 	/* Check if port has already been associated to a lagg */
653 	if (ifp->if_lagg != NULL) {
654 		/* Port is already in the current lagg? */
655 		lp = (struct lagg_port *)ifp->if_lagg;
656 		if (lp->lp_softc == sc)
657 			return (EEXIST);
658 		return (EBUSY);
659 	}
660 
661 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
662 	if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
663 		return (EPROTONOSUPPORT);
664 
665 	/* Allow the first Ethernet member to define the MTU */
666 	oldmtu = -1;
667 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
668 		sc->sc_ifp->if_mtu = ifp->if_mtu;
669 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
670 		if (ifp->if_ioctl == NULL) {
671 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
672 			    ifp->if_xname);
673 			return (EINVAL);
674 		}
675 		oldmtu = ifp->if_mtu;
676 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
677 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
678 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
679 		if (error != 0) {
680 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
681 			    ifp->if_xname);
682 			return (error);
683 		}
684 		ifr.ifr_mtu = oldmtu;
685 	}
686 
687 	lp = malloc(sizeof(struct lagg_port), M_DEVBUF, M_WAITOK|M_ZERO);
688 	lp->lp_softc = sc;
689 
690 	/* Check if port is a stacked lagg */
691 	LAGG_LIST_LOCK();
692 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
693 		if (ifp == sc_ptr->sc_ifp) {
694 			LAGG_LIST_UNLOCK();
695 			free(lp, M_DEVBUF);
696 			if (oldmtu != -1)
697 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
698 				    (caddr_t)&ifr);
699 			return (EINVAL);
700 			/* XXX disable stacking for the moment, its untested */
701 #ifdef LAGG_PORT_STACKING
702 			lp->lp_flags |= LAGG_PORT_STACK;
703 			if (lagg_port_checkstacking(sc_ptr) >=
704 			    LAGG_MAX_STACKING) {
705 				LAGG_LIST_UNLOCK();
706 				free(lp, M_DEVBUF);
707 				if (oldmtu != -1)
708 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
709 					    (caddr_t)&ifr);
710 				return (E2BIG);
711 			}
712 #endif
713 		}
714 	}
715 	LAGG_LIST_UNLOCK();
716 
717 	if_ref(ifp);
718 	lp->lp_ifp = ifp;
719 
720 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
721 	lp->lp_ifcapenable = ifp->if_capenable;
722 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
723 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
724 		lagg_proto_lladdr(sc);
725 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
726 	} else {
727 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
728 	}
729 	lagg_setflags(lp, 1);
730 
731 	if (CK_SLIST_EMPTY(&sc->sc_ports))
732 		sc->sc_primary = lp;
733 
734 	/* Change the interface type */
735 	lp->lp_iftype = ifp->if_type;
736 	ifp->if_type = IFT_IEEE8023ADLAG;
737 	ifp->if_lagg = lp;
738 	lp->lp_ioctl = ifp->if_ioctl;
739 	ifp->if_ioctl = lagg_port_ioctl;
740 	lp->lp_output = ifp->if_output;
741 	ifp->if_output = lagg_port_output;
742 
743 	/* Read port counters */
744 	pval = lp->port_counters.val;
745 	for (i = 0; i < IFCOUNTERS; i++, pval++)
746 		*pval = ifp->if_get_counter(ifp, i);
747 
748 	/*
749 	 * Insert into the list of ports.
750 	 * Keep ports sorted by if_index. It is handy, when configuration
751 	 * is predictable and `ifconfig laggN create ...` command
752 	 * will lead to the same result each time.
753 	 */
754 	LAGG_RLOCK();
755 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
756 		if (tlp->lp_ifp->if_index < ifp->if_index && (
757 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
758 		    ((struct  lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
759 		    ifp->if_index))
760 			break;
761 	}
762 	LAGG_RUNLOCK();
763 	if (tlp != NULL)
764 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
765 	else
766 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
767 	sc->sc_count++;
768 
769 	lagg_setmulti(lp);
770 
771 
772 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
773 		/* Remove the port, without calling pr_delport. */
774 		lagg_port_destroy(lp, 0);
775 		if (oldmtu != -1)
776 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
777 		return (error);
778 	}
779 
780 	/* Update lagg capabilities */
781 	lagg_capabilities(sc);
782 	lagg_linkstate(sc);
783 
784 	return (0);
785 }
786 
787 #ifdef LAGG_PORT_STACKING
788 static int
789 lagg_port_checkstacking(struct lagg_softc *sc)
790 {
791 	struct lagg_softc *sc_ptr;
792 	struct lagg_port *lp;
793 	int m = 0;
794 
795 	LAGG_SXLOCK_ASSERT(sc);
796 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
797 		if (lp->lp_flags & LAGG_PORT_STACK) {
798 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
799 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
800 		}
801 	}
802 
803 	return (m + 1);
804 }
805 #endif
806 
807 static void
808 lagg_port_destroy_cb(epoch_context_t ec)
809 {
810 	struct lagg_port *lp;
811 	struct ifnet *ifp;
812 
813 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
814 	ifp = lp->lp_ifp;
815 
816 	if_rele(ifp);
817 	free(lp, M_DEVBUF);
818 }
819 
820 static int
821 lagg_port_destroy(struct lagg_port *lp, int rundelport)
822 {
823 	struct lagg_softc *sc = lp->lp_softc;
824 	struct lagg_port *lp_ptr, *lp0;
825 	struct ifnet *ifp = lp->lp_ifp;
826 	uint64_t *pval, vdiff;
827 	int i;
828 
829 	LAGG_XLOCK_ASSERT(sc);
830 
831 	if (rundelport)
832 		lagg_proto_delport(sc, lp);
833 
834 	if (lp->lp_detaching == 0)
835 		lagg_clrmulti(lp);
836 
837 	/* Restore interface */
838 	ifp->if_type = lp->lp_iftype;
839 	ifp->if_ioctl = lp->lp_ioctl;
840 	ifp->if_output = lp->lp_output;
841 	ifp->if_lagg = NULL;
842 
843 	/* Update detached port counters */
844 	pval = lp->port_counters.val;
845 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
846 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
847 		sc->detached_counters.val[i] += vdiff;
848 	}
849 
850 	/* Finally, remove the port from the lagg */
851 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
852 	sc->sc_count--;
853 
854 	/* Update the primary interface */
855 	if (lp == sc->sc_primary) {
856 		uint8_t lladdr[ETHER_ADDR_LEN];
857 
858 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
859 			bzero(&lladdr, ETHER_ADDR_LEN);
860 		else
861 			bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN);
862 		sc->sc_primary = lp0;
863 		if (sc->sc_destroying == 0) {
864 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
865 			lagg_proto_lladdr(sc);
866 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
867 		}
868 
869 		/*
870 		 * Update lladdr for each port (new primary needs update
871 		 * as well, to switch from old lladdr to its 'real' one)
872 		 */
873 		CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
874 			if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN);
875 	}
876 
877 	if (lp->lp_ifflags)
878 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
879 
880 	if (lp->lp_detaching == 0) {
881 		lagg_setflags(lp, 0);
882 		lagg_setcaps(lp, lp->lp_ifcapenable);
883 		if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN);
884 	}
885 
886 	/*
887 	 * free port and release it's ifnet reference after a grace period has
888 	 * elapsed.
889 	 */
890 	epoch_call(net_epoch_preempt, &lp->lp_epoch_ctx, lagg_port_destroy_cb);
891 	/* Update lagg capabilities */
892 	lagg_capabilities(sc);
893 	lagg_linkstate(sc);
894 
895 	return (0);
896 }
897 
898 static int
899 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
900 {
901 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
902 	struct lagg_softc *sc;
903 	struct lagg_port *lp = NULL;
904 	int error = 0;
905 
906 	/* Should be checked by the caller */
907 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
908 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
909 		goto fallback;
910 
911 	switch (cmd) {
912 	case SIOCGLAGGPORT:
913 		if (rp->rp_portname[0] == '\0' ||
914 		    ifunit(rp->rp_portname) != ifp) {
915 			error = EINVAL;
916 			break;
917 		}
918 
919 		LAGG_RLOCK();
920 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
921 			error = ENOENT;
922 			LAGG_RUNLOCK();
923 			break;
924 		}
925 
926 		lagg_port2req(lp, rp);
927 		LAGG_RUNLOCK();
928 		break;
929 
930 	case SIOCSIFCAP:
931 		if (lp->lp_ioctl == NULL) {
932 			error = EINVAL;
933 			break;
934 		}
935 		error = (*lp->lp_ioctl)(ifp, cmd, data);
936 		if (error)
937 			break;
938 
939 		/* Update lagg interface capabilities */
940 		LAGG_XLOCK(sc);
941 		lagg_capabilities(sc);
942 		LAGG_XUNLOCK(sc);
943 		VLAN_CAPABILITIES(sc->sc_ifp);
944 		break;
945 
946 	case SIOCSIFMTU:
947 		/* Do not allow the MTU to be changed once joined */
948 		error = EINVAL;
949 		break;
950 
951 	default:
952 		goto fallback;
953 	}
954 
955 	return (error);
956 
957 fallback:
958 	if (lp != NULL && lp->lp_ioctl != NULL)
959 		return ((*lp->lp_ioctl)(ifp, cmd, data));
960 
961 	return (EINVAL);
962 }
963 
964 /*
965  * Requests counter @cnt data.
966  *
967  * Counter value is calculated the following way:
968  * 1) for each port, sum  difference between current and "initial" measurements.
969  * 2) add lagg logical interface counters.
970  * 3) add data from detached_counters array.
971  *
972  * We also do the following things on ports attach/detach:
973  * 1) On port attach we store all counters it has into port_counter array.
974  * 2) On port detach we add the different between "initial" and
975  *   current counters data to detached_counters array.
976  */
977 static uint64_t
978 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
979 {
980 	struct lagg_softc *sc;
981 	struct lagg_port *lp;
982 	struct ifnet *lpifp;
983 	uint64_t newval, oldval, vsum;
984 
985 	/* Revise this when we've got non-generic counters. */
986 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
987 
988 	sc = (struct lagg_softc *)ifp->if_softc;
989 
990 	vsum = 0;
991 	LAGG_RLOCK();
992 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
993 		/* Saved attached value */
994 		oldval = lp->port_counters.val[cnt];
995 		/* current value */
996 		lpifp = lp->lp_ifp;
997 		newval = lpifp->if_get_counter(lpifp, cnt);
998 		/* Calculate diff and save new */
999 		vsum += newval - oldval;
1000 	}
1001 	LAGG_RUNLOCK();
1002 
1003 	/*
1004 	 * Add counter data which might be added by upper
1005 	 * layer protocols operating on logical interface.
1006 	 */
1007 	vsum += if_get_counter_default(ifp, cnt);
1008 
1009 	/*
1010 	 * Add counter data from detached ports counters
1011 	 */
1012 	vsum += sc->detached_counters.val[cnt];
1013 
1014 
1015 	return (vsum);
1016 }
1017 
1018 /*
1019  * For direct output to child ports.
1020  */
1021 static int
1022 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1023 	const struct sockaddr *dst, struct route *ro)
1024 {
1025 	struct lagg_port *lp = ifp->if_lagg;
1026 
1027 	switch (dst->sa_family) {
1028 		case pseudo_AF_HDRCMPLT:
1029 		case AF_UNSPEC:
1030 			return ((*lp->lp_output)(ifp, m, dst, ro));
1031 	}
1032 
1033 	/* drop any other frames */
1034 	m_freem(m);
1035 	return (ENETDOWN);
1036 }
1037 
1038 static void
1039 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1040 {
1041 	struct lagg_port *lp;
1042 	struct lagg_softc *sc;
1043 
1044 	if ((lp = ifp->if_lagg) == NULL)
1045 		return;
1046 	/* If the ifnet is just being renamed, don't do anything. */
1047 	if (ifp->if_flags & IFF_RENAMING)
1048 		return;
1049 
1050 	sc = lp->lp_softc;
1051 
1052 	LAGG_XLOCK(sc);
1053 	lp->lp_detaching = 1;
1054 	lagg_port_destroy(lp, 1);
1055 	LAGG_XUNLOCK(sc);
1056 	VLAN_CAPABILITIES(sc->sc_ifp);
1057 }
1058 
1059 static void
1060 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1061 {
1062 	struct lagg_softc *sc = lp->lp_softc;
1063 
1064 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1065 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1066 	rp->rp_prio = lp->lp_prio;
1067 	rp->rp_flags = lp->lp_flags;
1068 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1069 
1070 	/* Add protocol specific flags */
1071 	switch (sc->sc_proto) {
1072 		case LAGG_PROTO_FAILOVER:
1073 			if (lp == sc->sc_primary)
1074 				rp->rp_flags |= LAGG_PORT_MASTER;
1075 			if (lp == lagg_link_active(sc, sc->sc_primary))
1076 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1077 			break;
1078 
1079 		case LAGG_PROTO_ROUNDROBIN:
1080 		case LAGG_PROTO_LOADBALANCE:
1081 		case LAGG_PROTO_BROADCAST:
1082 			if (LAGG_PORTACTIVE(lp))
1083 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1084 			break;
1085 
1086 		case LAGG_PROTO_LACP:
1087 			/* LACP has a different definition of active */
1088 			if (lacp_isactive(lp))
1089 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1090 			if (lacp_iscollecting(lp))
1091 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1092 			if (lacp_isdistributing(lp))
1093 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1094 			break;
1095 	}
1096 
1097 }
1098 
1099 static void
1100 lagg_init(void *xsc)
1101 {
1102 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1103 	struct ifnet *ifp = sc->sc_ifp;
1104 	struct lagg_port *lp;
1105 
1106 	LAGG_XLOCK(sc);
1107 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1108 		LAGG_XUNLOCK(sc);
1109 		return;
1110 	}
1111 
1112 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1113 
1114 	/*
1115 	 * Update the port lladdrs if needed.
1116 	 * This might be if_setlladdr() notification
1117 	 * that lladdr has been changed.
1118 	 */
1119 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1120 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1121 		    ETHER_ADDR_LEN) != 0)
1122 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN);
1123 	}
1124 
1125 	lagg_proto_init(sc);
1126 
1127 	LAGG_XUNLOCK(sc);
1128 }
1129 
1130 static void
1131 lagg_stop(struct lagg_softc *sc)
1132 {
1133 	struct ifnet *ifp = sc->sc_ifp;
1134 
1135 	LAGG_XLOCK_ASSERT(sc);
1136 
1137 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1138 		return;
1139 
1140 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1141 
1142 	lagg_proto_stop(sc);
1143 }
1144 
1145 static int
1146 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1147 {
1148 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1149 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1150 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1151 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1152 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1153 	struct ifreq *ifr = (struct ifreq *)data;
1154 	struct lagg_port *lp;
1155 	struct ifnet *tpif;
1156 	struct thread *td = curthread;
1157 	char *buf, *outbuf;
1158 	int count, buflen, len, error = 0;
1159 
1160 	bzero(&rpbuf, sizeof(rpbuf));
1161 
1162 	switch (cmd) {
1163 	case SIOCGLAGG:
1164 		LAGG_XLOCK(sc);
1165 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1166 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1167 		ra->ra_proto = sc->sc_proto;
1168 		lagg_proto_request(sc, &ra->ra_psc);
1169 		count = 0;
1170 		buf = outbuf;
1171 		len = min(ra->ra_size, buflen);
1172 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1173 			if (len < sizeof(rpbuf))
1174 				break;
1175 
1176 			lagg_port2req(lp, &rpbuf);
1177 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1178 			count++;
1179 			buf += sizeof(rpbuf);
1180 			len -= sizeof(rpbuf);
1181 		}
1182 		LAGG_XUNLOCK(sc);
1183 		ra->ra_ports = count;
1184 		ra->ra_size = count * sizeof(rpbuf);
1185 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1186 		free(outbuf, M_TEMP);
1187 		break;
1188 	case SIOCSLAGG:
1189 		error = priv_check(td, PRIV_NET_LAGG);
1190 		if (error)
1191 			break;
1192 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1193 			error = EPROTONOSUPPORT;
1194 			break;
1195 		}
1196 
1197 		LAGG_XLOCK(sc);
1198 		lagg_proto_detach(sc);
1199 		LAGG_UNLOCK_ASSERT();
1200 		lagg_proto_attach(sc, ra->ra_proto);
1201 		LAGG_XUNLOCK(sc);
1202 		break;
1203 	case SIOCGLAGGOPTS:
1204 		LAGG_XLOCK(sc);
1205 		ro->ro_opts = sc->sc_opts;
1206 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1207 			struct lacp_softc *lsc;
1208 
1209 			lsc = (struct lacp_softc *)sc->sc_psc;
1210 			if (lsc->lsc_debug.lsc_tx_test != 0)
1211 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1212 			if (lsc->lsc_debug.lsc_rx_test != 0)
1213 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1214 			if (lsc->lsc_strict_mode != 0)
1215 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1216 			if (lsc->lsc_fast_timeout != 0)
1217 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1218 
1219 			ro->ro_active = sc->sc_active;
1220 		} else {
1221 			ro->ro_active = 0;
1222 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1223 				ro->ro_active += LAGG_PORTACTIVE(lp);
1224 		}
1225 		ro->ro_bkt = sc->sc_bkt;
1226 		ro->ro_flapping = sc->sc_flapping;
1227 		ro->ro_flowid_shift = sc->flowid_shift;
1228 		LAGG_XUNLOCK(sc);
1229 		break;
1230 	case SIOCSLAGGOPTS:
1231 		if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) {
1232 			if (ro->ro_bkt == 0)
1233 				sc->sc_bkt = 1; // Minimum 1 packet per iface.
1234 			else
1235 				sc->sc_bkt = ro->ro_bkt;
1236 		}
1237 		error = priv_check(td, PRIV_NET_LAGG);
1238 		if (error)
1239 			break;
1240 		if (ro->ro_opts == 0)
1241 			break;
1242 		/*
1243 		 * Set options.  LACP options are stored in sc->sc_psc,
1244 		 * not in sc_opts.
1245 		 */
1246 		int valid, lacp;
1247 
1248 		switch (ro->ro_opts) {
1249 		case LAGG_OPT_USE_FLOWID:
1250 		case -LAGG_OPT_USE_FLOWID:
1251 		case LAGG_OPT_FLOWIDSHIFT:
1252 			valid = 1;
1253 			lacp = 0;
1254 			break;
1255 		case LAGG_OPT_LACP_TXTEST:
1256 		case -LAGG_OPT_LACP_TXTEST:
1257 		case LAGG_OPT_LACP_RXTEST:
1258 		case -LAGG_OPT_LACP_RXTEST:
1259 		case LAGG_OPT_LACP_STRICT:
1260 		case -LAGG_OPT_LACP_STRICT:
1261 		case LAGG_OPT_LACP_TIMEOUT:
1262 		case -LAGG_OPT_LACP_TIMEOUT:
1263 			valid = lacp = 1;
1264 			break;
1265 		default:
1266 			valid = lacp = 0;
1267 			break;
1268 		}
1269 
1270 		LAGG_XLOCK(sc);
1271 
1272 		if (valid == 0 ||
1273 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1274 			/* Invalid combination of options specified. */
1275 			error = EINVAL;
1276 			LAGG_XUNLOCK(sc);
1277 			break;	/* Return from SIOCSLAGGOPTS. */
1278 		}
1279 		/*
1280 		 * Store new options into sc->sc_opts except for
1281 		 * FLOWIDSHIFT and LACP options.
1282 		 */
1283 		if (lacp == 0) {
1284 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1285 				sc->flowid_shift = ro->ro_flowid_shift;
1286 			else if (ro->ro_opts > 0)
1287 				sc->sc_opts |= ro->ro_opts;
1288 			else
1289 				sc->sc_opts &= ~ro->ro_opts;
1290 		} else {
1291 			struct lacp_softc *lsc;
1292 			struct lacp_port *lp;
1293 
1294 			lsc = (struct lacp_softc *)sc->sc_psc;
1295 
1296 			switch (ro->ro_opts) {
1297 			case LAGG_OPT_LACP_TXTEST:
1298 				lsc->lsc_debug.lsc_tx_test = 1;
1299 				break;
1300 			case -LAGG_OPT_LACP_TXTEST:
1301 				lsc->lsc_debug.lsc_tx_test = 0;
1302 				break;
1303 			case LAGG_OPT_LACP_RXTEST:
1304 				lsc->lsc_debug.lsc_rx_test = 1;
1305 				break;
1306 			case -LAGG_OPT_LACP_RXTEST:
1307 				lsc->lsc_debug.lsc_rx_test = 0;
1308 				break;
1309 			case LAGG_OPT_LACP_STRICT:
1310 				lsc->lsc_strict_mode = 1;
1311 				break;
1312 			case -LAGG_OPT_LACP_STRICT:
1313 				lsc->lsc_strict_mode = 0;
1314 				break;
1315 			case LAGG_OPT_LACP_TIMEOUT:
1316 				LACP_LOCK(lsc);
1317         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1318                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1319 				LACP_UNLOCK(lsc);
1320 				lsc->lsc_fast_timeout = 1;
1321 				break;
1322 			case -LAGG_OPT_LACP_TIMEOUT:
1323 				LACP_LOCK(lsc);
1324         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1325                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1326 				LACP_UNLOCK(lsc);
1327 				lsc->lsc_fast_timeout = 0;
1328 				break;
1329 			}
1330 		}
1331 		LAGG_XUNLOCK(sc);
1332 		break;
1333 	case SIOCGLAGGFLAGS:
1334 		rf->rf_flags = 0;
1335 		LAGG_XLOCK(sc);
1336 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1337 			rf->rf_flags |= LAGG_F_HASHL2;
1338 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1339 			rf->rf_flags |= LAGG_F_HASHL3;
1340 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1341 			rf->rf_flags |= LAGG_F_HASHL4;
1342 		LAGG_XUNLOCK(sc);
1343 		break;
1344 	case SIOCSLAGGHASH:
1345 		error = priv_check(td, PRIV_NET_LAGG);
1346 		if (error)
1347 			break;
1348 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1349 			error = EINVAL;
1350 			break;
1351 		}
1352 		LAGG_XLOCK(sc);
1353 		sc->sc_flags = 0;
1354 		if (rf->rf_flags & LAGG_F_HASHL2)
1355 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1356 		if (rf->rf_flags & LAGG_F_HASHL3)
1357 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1358 		if (rf->rf_flags & LAGG_F_HASHL4)
1359 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1360 		LAGG_XUNLOCK(sc);
1361 		break;
1362 	case SIOCGLAGGPORT:
1363 		if (rp->rp_portname[0] == '\0' ||
1364 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1365 			error = EINVAL;
1366 			break;
1367 		}
1368 
1369 		LAGG_RLOCK();
1370 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1371 		    lp->lp_softc != sc) {
1372 			error = ENOENT;
1373 			LAGG_RUNLOCK();
1374 			if_rele(tpif);
1375 			break;
1376 		}
1377 
1378 		lagg_port2req(lp, rp);
1379 		LAGG_RUNLOCK();
1380 		if_rele(tpif);
1381 		break;
1382 	case SIOCSLAGGPORT:
1383 		error = priv_check(td, PRIV_NET_LAGG);
1384 		if (error)
1385 			break;
1386 		if (rp->rp_portname[0] == '\0' ||
1387 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1388 			error = EINVAL;
1389 			break;
1390 		}
1391 #ifdef INET6
1392 		/*
1393 		 * A laggport interface should not have inet6 address
1394 		 * because two interfaces with a valid link-local
1395 		 * scope zone must not be merged in any form.  This
1396 		 * restriction is needed to prevent violation of
1397 		 * link-local scope zone.  Attempts to add a laggport
1398 		 * interface which has inet6 addresses triggers
1399 		 * removal of all inet6 addresses on the member
1400 		 * interface.
1401 		 */
1402 		if (in6ifa_llaonifp(tpif)) {
1403 			in6_ifdetach(tpif);
1404 				if_printf(sc->sc_ifp,
1405 				    "IPv6 addresses on %s have been removed "
1406 				    "before adding it as a member to prevent "
1407 				    "IPv6 address scope violation.\n",
1408 				    tpif->if_xname);
1409 		}
1410 #endif
1411 		LAGG_XLOCK(sc);
1412 		error = lagg_port_create(sc, tpif);
1413 		LAGG_XUNLOCK(sc);
1414 		if_rele(tpif);
1415 		VLAN_CAPABILITIES(ifp);
1416 		break;
1417 	case SIOCSLAGGDELPORT:
1418 		error = priv_check(td, PRIV_NET_LAGG);
1419 		if (error)
1420 			break;
1421 		if (rp->rp_portname[0] == '\0' ||
1422 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1423 			error = EINVAL;
1424 			break;
1425 		}
1426 
1427 		LAGG_XLOCK(sc);
1428 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1429 		    lp->lp_softc != sc) {
1430 			error = ENOENT;
1431 			LAGG_XUNLOCK(sc);
1432 			if_rele(tpif);
1433 			break;
1434 		}
1435 
1436 		error = lagg_port_destroy(lp, 1);
1437 		LAGG_XUNLOCK(sc);
1438 		if_rele(tpif);
1439 		VLAN_CAPABILITIES(ifp);
1440 		break;
1441 	case SIOCSIFFLAGS:
1442 		/* Set flags on ports too */
1443 		LAGG_XLOCK(sc);
1444 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1445 			lagg_setflags(lp, 1);
1446 		}
1447 
1448 		if (!(ifp->if_flags & IFF_UP) &&
1449 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1450 			/*
1451 			 * If interface is marked down and it is running,
1452 			 * then stop and disable it.
1453 			 */
1454 			lagg_stop(sc);
1455 			LAGG_XUNLOCK(sc);
1456 		} else if ((ifp->if_flags & IFF_UP) &&
1457 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1458 			/*
1459 			 * If interface is marked up and it is stopped, then
1460 			 * start it.
1461 			 */
1462 			LAGG_XUNLOCK(sc);
1463 			(*ifp->if_init)(sc);
1464 		} else
1465 			LAGG_XUNLOCK(sc);
1466 		break;
1467 	case SIOCADDMULTI:
1468 	case SIOCDELMULTI:
1469 		LAGG_XLOCK(sc);
1470 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1471 			lagg_clrmulti(lp);
1472 			lagg_setmulti(lp);
1473 		}
1474 		LAGG_XUNLOCK(sc);
1475 		error = 0;
1476 		break;
1477 	case SIOCSIFMEDIA:
1478 	case SIOCGIFMEDIA:
1479 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1480 		break;
1481 
1482 	case SIOCSIFCAP:
1483 		LAGG_XLOCK(sc);
1484 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1485 			if (lp->lp_ioctl != NULL)
1486 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1487 		}
1488 		lagg_capabilities(sc);
1489 		LAGG_XUNLOCK(sc);
1490 		VLAN_CAPABILITIES(ifp);
1491 		error = 0;
1492 		break;
1493 
1494 	case SIOCSIFMTU:
1495 		LAGG_XLOCK(sc);
1496 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1497 			if (lp->lp_ioctl != NULL)
1498 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1499 			else
1500 				error = EINVAL;
1501 			if (error != 0) {
1502 				if_printf(ifp,
1503 				    "failed to change MTU to %d on port %s, "
1504 				    "reverting all ports to original MTU (%d)\n",
1505 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1506 				break;
1507 			}
1508 		}
1509 		if (error == 0) {
1510 			ifp->if_mtu = ifr->ifr_mtu;
1511 		} else {
1512 			/* set every port back to the original MTU */
1513 			ifr->ifr_mtu = ifp->if_mtu;
1514 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1515 				if (lp->lp_ioctl != NULL)
1516 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1517 			}
1518 		}
1519 		LAGG_XUNLOCK(sc);
1520 		break;
1521 
1522 	default:
1523 		error = ether_ioctl(ifp, cmd, data);
1524 		break;
1525 	}
1526 	return (error);
1527 }
1528 
1529 #ifdef RATELIMIT
1530 static int
1531 lagg_snd_tag_alloc(struct ifnet *ifp,
1532     union if_snd_tag_alloc_params *params,
1533     struct m_snd_tag **ppmt)
1534 {
1535 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1536 	struct lagg_port *lp;
1537 	struct lagg_lb *lb;
1538 	uint32_t p;
1539 
1540 	switch (sc->sc_proto) {
1541 	case LAGG_PROTO_FAILOVER:
1542 		lp = lagg_link_active(sc, sc->sc_primary);
1543 		break;
1544 	case LAGG_PROTO_LOADBALANCE:
1545 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1546 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1547 			return (EOPNOTSUPP);
1548 		p = params->hdr.flowid >> sc->flowid_shift;
1549 		p %= sc->sc_count;
1550 		lb = (struct lagg_lb *)sc->sc_psc;
1551 		lp = lb->lb_ports[p];
1552 		lp = lagg_link_active(sc, lp);
1553 		break;
1554 	case LAGG_PROTO_LACP:
1555 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1556 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1557 			return (EOPNOTSUPP);
1558 		lp = lacp_select_tx_port_by_hash(sc, params->hdr.flowid);
1559 		break;
1560 	default:
1561 		return (EOPNOTSUPP);
1562 	}
1563 	if (lp == NULL)
1564 		return (EOPNOTSUPP);
1565 	ifp = lp->lp_ifp;
1566 	if (ifp == NULL || ifp->if_snd_tag_alloc == NULL ||
1567 	    (ifp->if_capenable & IFCAP_TXRTLMT) == 0)
1568 		return (EOPNOTSUPP);
1569 
1570 	/* forward allocation request */
1571 	return (ifp->if_snd_tag_alloc(ifp, params, ppmt));
1572 }
1573 #endif
1574 
1575 static int
1576 lagg_setmulti(struct lagg_port *lp)
1577 {
1578 	struct lagg_softc *sc = lp->lp_softc;
1579 	struct ifnet *ifp = lp->lp_ifp;
1580 	struct ifnet *scifp = sc->sc_ifp;
1581 	struct lagg_mc *mc;
1582 	struct ifmultiaddr *ifma;
1583 	int error;
1584 
1585 	IF_ADDR_WLOCK(scifp);
1586 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1587 		if (ifma->ifma_addr->sa_family != AF_LINK)
1588 			continue;
1589 		mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1590 		if (mc == NULL) {
1591 			IF_ADDR_WUNLOCK(scifp);
1592 			return (ENOMEM);
1593 		}
1594 		bcopy(ifma->ifma_addr, &mc->mc_addr,
1595 		    ifma->ifma_addr->sa_len);
1596 		mc->mc_addr.sdl_index = ifp->if_index;
1597 		mc->mc_ifma = NULL;
1598 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1599 	}
1600 	IF_ADDR_WUNLOCK(scifp);
1601 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1602 		error = if_addmulti(ifp,
1603 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1604 		if (error)
1605 			return (error);
1606 	}
1607 	return (0);
1608 }
1609 
1610 static int
1611 lagg_clrmulti(struct lagg_port *lp)
1612 {
1613 	struct lagg_mc *mc;
1614 
1615 	LAGG_XLOCK_ASSERT(lp->lp_softc);
1616 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1617 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1618 		if (mc->mc_ifma && lp->lp_detaching == 0)
1619 			if_delmulti_ifma(mc->mc_ifma);
1620 		free(mc, M_DEVBUF);
1621 	}
1622 	return (0);
1623 }
1624 
1625 static int
1626 lagg_setcaps(struct lagg_port *lp, int cap)
1627 {
1628 	struct ifreq ifr;
1629 
1630 	if (lp->lp_ifp->if_capenable == cap)
1631 		return (0);
1632 	if (lp->lp_ioctl == NULL)
1633 		return (ENXIO);
1634 	ifr.ifr_reqcap = cap;
1635 	return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1636 }
1637 
1638 /* Handle a ref counted flag that should be set on the lagg port as well */
1639 static int
1640 lagg_setflag(struct lagg_port *lp, int flag, int status,
1641     int (*func)(struct ifnet *, int))
1642 {
1643 	struct lagg_softc *sc = lp->lp_softc;
1644 	struct ifnet *scifp = sc->sc_ifp;
1645 	struct ifnet *ifp = lp->lp_ifp;
1646 	int error;
1647 
1648 	LAGG_XLOCK_ASSERT(sc);
1649 
1650 	status = status ? (scifp->if_flags & flag) : 0;
1651 	/* Now "status" contains the flag value or 0 */
1652 
1653 	/*
1654 	 * See if recorded ports status is different from what
1655 	 * we want it to be.  If it is, flip it.  We record ports
1656 	 * status in lp_ifflags so that we won't clear ports flag
1657 	 * we haven't set.  In fact, we don't clear or set ports
1658 	 * flags directly, but get or release references to them.
1659 	 * That's why we can be sure that recorded flags still are
1660 	 * in accord with actual ports flags.
1661 	 */
1662 	if (status != (lp->lp_ifflags & flag)) {
1663 		error = (*func)(ifp, status);
1664 		if (error)
1665 			return (error);
1666 		lp->lp_ifflags &= ~flag;
1667 		lp->lp_ifflags |= status;
1668 	}
1669 	return (0);
1670 }
1671 
1672 /*
1673  * Handle IFF_* flags that require certain changes on the lagg port
1674  * if "status" is true, update ports flags respective to the lagg
1675  * if "status" is false, forcedly clear the flags set on port.
1676  */
1677 static int
1678 lagg_setflags(struct lagg_port *lp, int status)
1679 {
1680 	int error, i;
1681 
1682 	for (i = 0; lagg_pflags[i].flag; i++) {
1683 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1684 		    status, lagg_pflags[i].func);
1685 		if (error)
1686 			return (error);
1687 	}
1688 	return (0);
1689 }
1690 
1691 static int
1692 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1693 {
1694 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1695 	int error;
1696 
1697 	LAGG_RLOCK();
1698 	/* We need a Tx algorithm and at least one port */
1699 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1700 		LAGG_RUNLOCK();
1701 		m_freem(m);
1702 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1703 		return (ENXIO);
1704 	}
1705 
1706 	ETHER_BPF_MTAP(ifp, m);
1707 
1708 	error = lagg_proto_start(sc, m);
1709 	LAGG_RUNLOCK();
1710 
1711 	if (error != 0)
1712 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1713 
1714 	return (error);
1715 }
1716 
1717 /*
1718  * The ifp->if_qflush entry point for lagg(4) is no-op.
1719  */
1720 static void
1721 lagg_qflush(struct ifnet *ifp __unused)
1722 {
1723 }
1724 
1725 static struct mbuf *
1726 lagg_input(struct ifnet *ifp, struct mbuf *m)
1727 {
1728 	struct lagg_port *lp = ifp->if_lagg;
1729 	struct lagg_softc *sc = lp->lp_softc;
1730 	struct ifnet *scifp = sc->sc_ifp;
1731 
1732 	LAGG_RLOCK();
1733 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1734 	    lp->lp_detaching != 0 ||
1735 	    sc->sc_proto == LAGG_PROTO_NONE) {
1736 		LAGG_RUNLOCK();
1737 		m_freem(m);
1738 		return (NULL);
1739 	}
1740 
1741 	ETHER_BPF_MTAP(scifp, m);
1742 
1743 	m = lagg_proto_input(sc, lp, m);
1744 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
1745 		m_freem(m);
1746 		m = NULL;
1747 	}
1748 
1749 	LAGG_RUNLOCK();
1750 	return (m);
1751 }
1752 
1753 static int
1754 lagg_media_change(struct ifnet *ifp)
1755 {
1756 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1757 
1758 	if (sc->sc_ifflags & IFF_DEBUG)
1759 		printf("%s\n", __func__);
1760 
1761 	/* Ignore */
1762 	return (0);
1763 }
1764 
1765 static void
1766 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1767 {
1768 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1769 	struct lagg_port *lp;
1770 
1771 	imr->ifm_status = IFM_AVALID;
1772 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1773 
1774 	LAGG_RLOCK();
1775 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1776 		if (LAGG_PORTACTIVE(lp))
1777 			imr->ifm_status |= IFM_ACTIVE;
1778 	}
1779 	LAGG_RUNLOCK();
1780 }
1781 
1782 static void
1783 lagg_linkstate(struct lagg_softc *sc)
1784 {
1785 	struct lagg_port *lp;
1786 	int new_link = LINK_STATE_DOWN;
1787 	uint64_t speed;
1788 
1789 	LAGG_XLOCK_ASSERT(sc);
1790 
1791 	/* LACP handles link state itself */
1792 	if (sc->sc_proto == LAGG_PROTO_LACP)
1793 		return;
1794 
1795 	/* Our link is considered up if at least one of our ports is active */
1796 	LAGG_RLOCK();
1797 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1798 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1799 			new_link = LINK_STATE_UP;
1800 			break;
1801 		}
1802 	}
1803 	LAGG_RUNLOCK();
1804 	if_link_state_change(sc->sc_ifp, new_link);
1805 
1806 	/* Update if_baudrate to reflect the max possible speed */
1807 	switch (sc->sc_proto) {
1808 		case LAGG_PROTO_FAILOVER:
1809 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1810 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1811 			break;
1812 		case LAGG_PROTO_ROUNDROBIN:
1813 		case LAGG_PROTO_LOADBALANCE:
1814 		case LAGG_PROTO_BROADCAST:
1815 			speed = 0;
1816 			LAGG_RLOCK();
1817 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1818 				speed += lp->lp_ifp->if_baudrate;
1819 			LAGG_RUNLOCK();
1820 			sc->sc_ifp->if_baudrate = speed;
1821 			break;
1822 		case LAGG_PROTO_LACP:
1823 			/* LACP updates if_baudrate itself */
1824 			break;
1825 	}
1826 }
1827 
1828 static void
1829 lagg_port_state(struct ifnet *ifp, int state)
1830 {
1831 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1832 	struct lagg_softc *sc = NULL;
1833 
1834 	if (lp != NULL)
1835 		sc = lp->lp_softc;
1836 	if (sc == NULL)
1837 		return;
1838 
1839 	LAGG_XLOCK(sc);
1840 	lagg_linkstate(sc);
1841 	lagg_proto_linkstate(sc, lp);
1842 	LAGG_XUNLOCK(sc);
1843 }
1844 
1845 struct lagg_port *
1846 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1847 {
1848 	struct lagg_port *lp_next, *rval = NULL;
1849 	struct epoch_tracker net_et;
1850 
1851 	/*
1852 	 * Search a port which reports an active link state.
1853 	 */
1854 
1855 	if (lp == NULL)
1856 		goto search;
1857 	if (LAGG_PORTACTIVE(lp)) {
1858 		rval = lp;
1859 		goto found;
1860 	}
1861 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
1862 	    LAGG_PORTACTIVE(lp_next)) {
1863 		rval = lp_next;
1864 		goto found;
1865 	}
1866 
1867  search:
1868 	epoch_enter_preempt(net_epoch_preempt, &net_et);
1869 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1870 		if (LAGG_PORTACTIVE(lp_next)) {
1871 			epoch_exit_preempt(net_epoch_preempt, &net_et);
1872 			return (lp_next);
1873 		}
1874 	}
1875 	epoch_exit_preempt(net_epoch_preempt, &net_et);
1876 found:
1877 	return (rval);
1878 }
1879 
1880 int
1881 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1882 {
1883 
1884 	return (ifp->if_transmit)(ifp, m);
1885 }
1886 
1887 /*
1888  * Simple round robin aggregation
1889  */
1890 static void
1891 lagg_rr_attach(struct lagg_softc *sc)
1892 {
1893 	sc->sc_seq = 0;
1894 	sc->sc_bkt_count = sc->sc_bkt;
1895 }
1896 
1897 static int
1898 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1899 {
1900 	struct lagg_port *lp;
1901 	uint32_t p;
1902 
1903 	if (sc->sc_bkt_count == 0 && sc->sc_bkt > 0)
1904 		sc->sc_bkt_count = sc->sc_bkt;
1905 
1906 	if (sc->sc_bkt > 0) {
1907 		atomic_subtract_int(&sc->sc_bkt_count, 1);
1908 	if (atomic_cmpset_int(&sc->sc_bkt_count, 0, sc->sc_bkt))
1909 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1910 	else
1911 		p = sc->sc_seq;
1912 	} else
1913 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1914 
1915 	p %= sc->sc_count;
1916 	lp = CK_SLIST_FIRST(&sc->sc_ports);
1917 
1918 	while (p--)
1919 		lp = CK_SLIST_NEXT(lp, lp_entries);
1920 
1921 	/*
1922 	 * Check the port's link state. This will return the next active
1923 	 * port if the link is down or the port is NULL.
1924 	 */
1925 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1926 		m_freem(m);
1927 		return (ENETDOWN);
1928 	}
1929 
1930 	/* Send mbuf */
1931 	return (lagg_enqueue(lp->lp_ifp, m));
1932 }
1933 
1934 static struct mbuf *
1935 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1936 {
1937 	struct ifnet *ifp = sc->sc_ifp;
1938 
1939 	/* Just pass in the packet to our lagg device */
1940 	m->m_pkthdr.rcvif = ifp;
1941 
1942 	return (m);
1943 }
1944 
1945 /*
1946  * Broadcast mode
1947  */
1948 static int
1949 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1950 {
1951 	int active_ports = 0;
1952 	int errors = 0;
1953 	int ret;
1954 	struct lagg_port *lp, *last = NULL;
1955 	struct mbuf *m0;
1956 
1957 	LAGG_RLOCK();
1958 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1959 		if (!LAGG_PORTACTIVE(lp))
1960 			continue;
1961 
1962 		active_ports++;
1963 
1964 		if (last != NULL) {
1965 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
1966 			if (m0 == NULL) {
1967 				ret = ENOBUFS;
1968 				errors++;
1969 				break;
1970 			}
1971 
1972 			ret = lagg_enqueue(last->lp_ifp, m0);
1973 			if (ret != 0)
1974 				errors++;
1975 		}
1976 		last = lp;
1977 	}
1978 	LAGG_RUNLOCK();
1979 
1980 	if (last == NULL) {
1981 		m_freem(m);
1982 		return (ENOENT);
1983 	}
1984 	if ((last = lagg_link_active(sc, last)) == NULL) {
1985 		m_freem(m);
1986 		return (ENETDOWN);
1987 	}
1988 
1989 	ret = lagg_enqueue(last->lp_ifp, m);
1990 	if (ret != 0)
1991 		errors++;
1992 
1993 	if (errors == 0)
1994 		return (ret);
1995 
1996 	return (0);
1997 }
1998 
1999 static struct mbuf*
2000 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2001 {
2002 	struct ifnet *ifp = sc->sc_ifp;
2003 
2004 	/* Just pass in the packet to our lagg device */
2005 	m->m_pkthdr.rcvif = ifp;
2006 	return (m);
2007 }
2008 
2009 /*
2010  * Active failover
2011  */
2012 static int
2013 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2014 {
2015 	struct lagg_port *lp;
2016 
2017 	/* Use the master port if active or the next available port */
2018 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2019 		m_freem(m);
2020 		return (ENETDOWN);
2021 	}
2022 
2023 	/* Send mbuf */
2024 	return (lagg_enqueue(lp->lp_ifp, m));
2025 }
2026 
2027 static struct mbuf *
2028 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2029 {
2030 	struct ifnet *ifp = sc->sc_ifp;
2031 	struct lagg_port *tmp_tp;
2032 
2033 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2034 		m->m_pkthdr.rcvif = ifp;
2035 		return (m);
2036 	}
2037 
2038 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2039 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2040 		/*
2041 		 * If tmp_tp is null, we've received a packet when all
2042 		 * our links are down. Weird, but process it anyways.
2043 		 */
2044 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2045 			m->m_pkthdr.rcvif = ifp;
2046 			return (m);
2047 		}
2048 	}
2049 
2050 	m_freem(m);
2051 	return (NULL);
2052 }
2053 
2054 /*
2055  * Loadbalancing
2056  */
2057 static void
2058 lagg_lb_attach(struct lagg_softc *sc)
2059 {
2060 	struct lagg_port *lp;
2061 	struct lagg_lb *lb;
2062 
2063 	LAGG_XLOCK_ASSERT(sc);
2064 	lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
2065 	lb->lb_key = m_ether_tcpip_hash_init();
2066 	sc->sc_psc = lb;
2067 
2068 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2069 		lagg_lb_port_create(lp);
2070 }
2071 
2072 static void
2073 lagg_lb_detach(struct lagg_softc *sc)
2074 {
2075 	struct lagg_lb *lb;
2076 
2077 	lb = (struct lagg_lb *)sc->sc_psc;
2078 	if (lb != NULL)
2079 		free(lb, M_DEVBUF);
2080 }
2081 
2082 static int
2083 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2084 {
2085 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2086 	struct lagg_port *lp_next;
2087 	int i = 0, rv;
2088 
2089 	rv = 0;
2090 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2091 	LAGG_RLOCK();
2092 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2093 		if (lp_next == lp)
2094 			continue;
2095 		if (i >= LAGG_MAX_PORTS) {
2096 			rv = EINVAL;
2097 			break;
2098 		}
2099 		if (sc->sc_ifflags & IFF_DEBUG)
2100 			printf("%s: port %s at index %d\n",
2101 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2102 		lb->lb_ports[i++] = lp_next;
2103 	}
2104 	LAGG_RUNLOCK();
2105 
2106 	return (rv);
2107 }
2108 
2109 static int
2110 lagg_lb_port_create(struct lagg_port *lp)
2111 {
2112 	struct lagg_softc *sc = lp->lp_softc;
2113 	return (lagg_lb_porttable(sc, NULL));
2114 }
2115 
2116 static void
2117 lagg_lb_port_destroy(struct lagg_port *lp)
2118 {
2119 	struct lagg_softc *sc = lp->lp_softc;
2120 	lagg_lb_porttable(sc, lp);
2121 }
2122 
2123 static int
2124 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2125 {
2126 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2127 	struct lagg_port *lp = NULL;
2128 	uint32_t p = 0;
2129 
2130 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2131 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2132 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2133 	else
2134 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2135 	p %= sc->sc_count;
2136 	lp = lb->lb_ports[p];
2137 
2138 	/*
2139 	 * Check the port's link state. This will return the next active
2140 	 * port if the link is down or the port is NULL.
2141 	 */
2142 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2143 		m_freem(m);
2144 		return (ENETDOWN);
2145 	}
2146 
2147 	/* Send mbuf */
2148 	return (lagg_enqueue(lp->lp_ifp, m));
2149 }
2150 
2151 static struct mbuf *
2152 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2153 {
2154 	struct ifnet *ifp = sc->sc_ifp;
2155 
2156 	/* Just pass in the packet to our lagg device */
2157 	m->m_pkthdr.rcvif = ifp;
2158 
2159 	return (m);
2160 }
2161 
2162 /*
2163  * 802.3ad LACP
2164  */
2165 static void
2166 lagg_lacp_attach(struct lagg_softc *sc)
2167 {
2168 	struct lagg_port *lp;
2169 
2170 	lacp_attach(sc);
2171 	LAGG_XLOCK_ASSERT(sc);
2172 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2173 		lacp_port_create(lp);
2174 }
2175 
2176 static void
2177 lagg_lacp_detach(struct lagg_softc *sc)
2178 {
2179 	struct lagg_port *lp;
2180 	void *psc;
2181 
2182 	LAGG_XLOCK_ASSERT(sc);
2183 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2184 		lacp_port_destroy(lp);
2185 
2186 	psc = sc->sc_psc;
2187 	sc->sc_psc = NULL;
2188 	lacp_detach(psc);
2189 }
2190 
2191 static void
2192 lagg_lacp_lladdr(struct lagg_softc *sc)
2193 {
2194 	struct lagg_port *lp;
2195 
2196 	LAGG_SXLOCK_ASSERT(sc);
2197 
2198 	/* purge all the lacp ports */
2199 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2200 		lacp_port_destroy(lp);
2201 
2202 	/* add them back in */
2203 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2204 		lacp_port_create(lp);
2205 }
2206 
2207 static int
2208 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2209 {
2210 	struct lagg_port *lp;
2211 
2212 	lp = lacp_select_tx_port(sc, m);
2213 	if (lp == NULL) {
2214 		m_freem(m);
2215 		return (ENETDOWN);
2216 	}
2217 
2218 	/* Send mbuf */
2219 	return (lagg_enqueue(lp->lp_ifp, m));
2220 }
2221 
2222 static struct mbuf *
2223 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2224 {
2225 	struct ifnet *ifp = sc->sc_ifp;
2226 	struct ether_header *eh;
2227 	u_short etype;
2228 
2229 	eh = mtod(m, struct ether_header *);
2230 	etype = ntohs(eh->ether_type);
2231 
2232 	/* Tap off LACP control messages */
2233 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2234 		m = lacp_input(lp, m);
2235 		if (m == NULL)
2236 			return (NULL);
2237 	}
2238 
2239 	/*
2240 	 * If the port is not collecting or not in the active aggregator then
2241 	 * free and return.
2242 	 */
2243 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2244 		m_freem(m);
2245 		return (NULL);
2246 	}
2247 
2248 	m->m_pkthdr.rcvif = ifp;
2249 	return (m);
2250 }
2251 
2252