xref: /freebsd/sys/net/if_lagg.c (revision d6b92ffa)
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 /* Special flags we should propagate to the lagg ports. */
77 static struct {
78 	int flag;
79 	int (*func)(struct ifnet *, int);
80 } lagg_pflags[] = {
81 	{IFF_PROMISC, ifpromisc},
82 	{IFF_ALLMULTI, if_allmulti},
83 	{0, NULL}
84 };
85 
86 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
87 #define	V_lagg_list	VNET(lagg_list)
88 static VNET_DEFINE(struct mtx, lagg_list_mtx);
89 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
90 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
91 					"if_lagg list", NULL, MTX_DEF)
92 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
93 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
94 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
95 eventhandler_tag	lagg_detach_cookie = NULL;
96 
97 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
98 static void	lagg_clone_destroy(struct ifnet *);
99 static VNET_DEFINE(struct if_clone *, lagg_cloner);
100 #define	V_lagg_cloner	VNET(lagg_cloner)
101 static const char laggname[] = "lagg";
102 
103 static void	lagg_capabilities(struct lagg_softc *);
104 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
105 static int	lagg_port_destroy(struct lagg_port *, int);
106 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
107 static void	lagg_linkstate(struct lagg_softc *);
108 static void	lagg_port_state(struct ifnet *, int);
109 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
110 static int	lagg_port_output(struct ifnet *, struct mbuf *,
111 		    const struct sockaddr *, struct route *);
112 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
113 #ifdef LAGG_PORT_STACKING
114 static int	lagg_port_checkstacking(struct lagg_softc *);
115 #endif
116 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
117 static void	lagg_init(void *);
118 static void	lagg_stop(struct lagg_softc *);
119 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
120 #ifdef RATELIMIT
121 static int	lagg_snd_tag_alloc(struct ifnet *,
122 		    union if_snd_tag_alloc_params *,
123 		    struct m_snd_tag **);
124 #endif
125 static int	lagg_setmulti(struct lagg_port *);
126 static int	lagg_clrmulti(struct lagg_port *);
127 static	int	lagg_setcaps(struct lagg_port *, int cap);
128 static	int	lagg_setflag(struct lagg_port *, int, int,
129 		    int (*func)(struct ifnet *, int));
130 static	int	lagg_setflags(struct lagg_port *, int status);
131 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
132 static int	lagg_transmit(struct ifnet *, struct mbuf *);
133 static void	lagg_qflush(struct ifnet *);
134 static int	lagg_media_change(struct ifnet *);
135 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
136 static struct lagg_port *lagg_link_active(struct lagg_softc *,
137 	    struct lagg_port *);
138 
139 /* Simple round robin */
140 static void	lagg_rr_attach(struct lagg_softc *);
141 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
142 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
143 		    struct mbuf *);
144 
145 /* Active failover */
146 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
147 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
148 		    struct mbuf *);
149 
150 /* Loadbalancing */
151 static void	lagg_lb_attach(struct lagg_softc *);
152 static void	lagg_lb_detach(struct lagg_softc *);
153 static int	lagg_lb_port_create(struct lagg_port *);
154 static void	lagg_lb_port_destroy(struct lagg_port *);
155 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
156 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
157 		    struct mbuf *);
158 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
159 
160 /* Broadcast */
161 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
162 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
163 		    struct mbuf *);
164 
165 /* 802.3ad LACP */
166 static void	lagg_lacp_attach(struct lagg_softc *);
167 static void	lagg_lacp_detach(struct lagg_softc *);
168 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
169 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
170 		    struct mbuf *);
171 static void	lagg_lacp_lladdr(struct lagg_softc *);
172 
173 /* lagg protocol table */
174 static const struct lagg_proto {
175 	lagg_proto	pr_num;
176 	void		(*pr_attach)(struct lagg_softc *);
177 	void		(*pr_detach)(struct lagg_softc *);
178 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
179 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
180 			    struct mbuf *);
181 	int		(*pr_addport)(struct lagg_port *);
182 	void		(*pr_delport)(struct lagg_port *);
183 	void		(*pr_linkstate)(struct lagg_port *);
184 	void 		(*pr_init)(struct lagg_softc *);
185 	void 		(*pr_stop)(struct lagg_softc *);
186 	void 		(*pr_lladdr)(struct lagg_softc *);
187 	void		(*pr_request)(struct lagg_softc *, void *);
188 	void		(*pr_portreq)(struct lagg_port *, void *);
189 } lagg_protos[] = {
190     {
191 	.pr_num = LAGG_PROTO_NONE
192     },
193     {
194 	.pr_num = LAGG_PROTO_ROUNDROBIN,
195 	.pr_attach = lagg_rr_attach,
196 	.pr_start = lagg_rr_start,
197 	.pr_input = lagg_rr_input,
198     },
199     {
200 	.pr_num = LAGG_PROTO_FAILOVER,
201 	.pr_start = lagg_fail_start,
202 	.pr_input = lagg_fail_input,
203     },
204     {
205 	.pr_num = LAGG_PROTO_LOADBALANCE,
206 	.pr_attach = lagg_lb_attach,
207 	.pr_detach = lagg_lb_detach,
208 	.pr_start = lagg_lb_start,
209 	.pr_input = lagg_lb_input,
210 	.pr_addport = lagg_lb_port_create,
211 	.pr_delport = lagg_lb_port_destroy,
212     },
213     {
214 	.pr_num = LAGG_PROTO_LACP,
215 	.pr_attach = lagg_lacp_attach,
216 	.pr_detach = lagg_lacp_detach,
217 	.pr_start = lagg_lacp_start,
218 	.pr_input = lagg_lacp_input,
219 	.pr_addport = lacp_port_create,
220 	.pr_delport = lacp_port_destroy,
221 	.pr_linkstate = lacp_linkstate,
222 	.pr_init = lacp_init,
223 	.pr_stop = lacp_stop,
224 	.pr_lladdr = lagg_lacp_lladdr,
225 	.pr_request = lacp_req,
226 	.pr_portreq = lacp_portreq,
227     },
228     {
229 	.pr_num = LAGG_PROTO_BROADCAST,
230 	.pr_start = lagg_bcast_start,
231 	.pr_input = lagg_bcast_input,
232     },
233 };
234 
235 SYSCTL_DECL(_net_link);
236 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
237     "Link Aggregation");
238 
239 /* Allow input on any failover links */
240 static VNET_DEFINE(int, lagg_failover_rx_all);
241 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
242 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
243     &VNET_NAME(lagg_failover_rx_all), 0,
244     "Accept input from any interface in a failover lagg");
245 
246 /* Default value for using flowid */
247 static VNET_DEFINE(int, def_use_flowid) = 1;
248 #define	V_def_use_flowid	VNET(def_use_flowid)
249 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
250     &VNET_NAME(def_use_flowid), 0,
251     "Default setting for using flow id for load sharing");
252 
253 /* Default value for flowid shift */
254 static VNET_DEFINE(int, def_flowid_shift) = 16;
255 #define	V_def_flowid_shift	VNET(def_flowid_shift)
256 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
257     &VNET_NAME(def_flowid_shift), 0,
258     "Default setting for flowid shift for load sharing");
259 
260 static void
261 vnet_lagg_init(const void *unused __unused)
262 {
263 
264 	LAGG_LIST_LOCK_INIT();
265 	SLIST_INIT(&V_lagg_list);
266 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
267 	    lagg_clone_destroy, 0);
268 }
269 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
270     vnet_lagg_init, NULL);
271 
272 static void
273 vnet_lagg_uninit(const void *unused __unused)
274 {
275 
276 	if_clone_detach(V_lagg_cloner);
277 	LAGG_LIST_LOCK_DESTROY();
278 }
279 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
280     vnet_lagg_uninit, NULL);
281 
282 static int
283 lagg_modevent(module_t mod, int type, void *data)
284 {
285 
286 	switch (type) {
287 	case MOD_LOAD:
288 		lagg_input_p = lagg_input;
289 		lagg_linkstate_p = lagg_port_state;
290 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
291 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
292 		    EVENTHANDLER_PRI_ANY);
293 		break;
294 	case MOD_UNLOAD:
295 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
296 		    lagg_detach_cookie);
297 		lagg_input_p = NULL;
298 		lagg_linkstate_p = NULL;
299 		break;
300 	default:
301 		return (EOPNOTSUPP);
302 	}
303 	return (0);
304 }
305 
306 static moduledata_t lagg_mod = {
307 	"if_lagg",
308 	lagg_modevent,
309 	0
310 };
311 
312 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
313 MODULE_VERSION(if_lagg, 1);
314 
315 static void
316 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
317 {
318 
319 	LAGG_XLOCK_ASSERT(sc);
320 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
321 	    __func__, sc));
322 
323 	if (sc->sc_ifflags & IFF_DEBUG)
324 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
325 
326 	if (lagg_protos[pr].pr_attach != NULL)
327 		lagg_protos[pr].pr_attach(sc);
328 	sc->sc_proto = pr;
329 }
330 
331 static void
332 lagg_proto_detach(struct lagg_softc *sc)
333 {
334 	lagg_proto pr;
335 
336 	LAGG_XLOCK_ASSERT(sc);
337 	LAGG_WLOCK_ASSERT(sc);
338 	pr = sc->sc_proto;
339 	sc->sc_proto = LAGG_PROTO_NONE;
340 
341 	if (lagg_protos[pr].pr_detach != NULL)
342 		lagg_protos[pr].pr_detach(sc);
343 	else
344 		LAGG_WUNLOCK(sc);
345 }
346 
347 static int
348 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
349 {
350 
351 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
352 }
353 
354 static struct mbuf *
355 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
356 {
357 
358 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
359 }
360 
361 static int
362 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
363 {
364 
365 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
366 		return (0);
367 	else
368 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
369 }
370 
371 static void
372 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
373 {
374 
375 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
376 		lagg_protos[sc->sc_proto].pr_delport(lp);
377 }
378 
379 static void
380 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
381 {
382 
383 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
384 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
385 }
386 
387 static void
388 lagg_proto_init(struct lagg_softc *sc)
389 {
390 
391 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
392 		lagg_protos[sc->sc_proto].pr_init(sc);
393 }
394 
395 static void
396 lagg_proto_stop(struct lagg_softc *sc)
397 {
398 
399 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
400 		lagg_protos[sc->sc_proto].pr_stop(sc);
401 }
402 
403 static void
404 lagg_proto_lladdr(struct lagg_softc *sc)
405 {
406 
407 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
408 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
409 }
410 
411 static void
412 lagg_proto_request(struct lagg_softc *sc, void *v)
413 {
414 
415 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
416 		lagg_protos[sc->sc_proto].pr_request(sc, v);
417 }
418 
419 static void
420 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
421 {
422 
423 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
424 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
425 }
426 
427 /*
428  * This routine is run via an vlan
429  * config EVENT
430  */
431 static void
432 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
433 {
434 	struct lagg_softc *sc = ifp->if_softc;
435 	struct lagg_port *lp;
436 
437 	if (ifp->if_softc !=  arg)   /* Not our event */
438 		return;
439 
440 	LAGG_SLOCK(sc);
441 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
442 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
443 	LAGG_SUNLOCK(sc);
444 }
445 
446 /*
447  * This routine is run via an vlan
448  * unconfig EVENT
449  */
450 static void
451 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
452 {
453 	struct lagg_softc *sc = ifp->if_softc;
454 	struct lagg_port *lp;
455 
456 	if (ifp->if_softc !=  arg)   /* Not our event */
457 		return;
458 
459 	LAGG_SLOCK(sc);
460 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
461 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
462 	LAGG_SUNLOCK(sc);
463 }
464 
465 static int
466 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
467 {
468 	struct lagg_softc *sc;
469 	struct ifnet *ifp;
470 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
471 
472 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
473 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
474 	if (ifp == NULL) {
475 		free(sc, M_DEVBUF);
476 		return (ENOSPC);
477 	}
478 	LAGG_LOCK_INIT(sc);
479 	LAGG_SX_INIT(sc);
480 
481 	LAGG_XLOCK(sc);
482 	if (V_def_use_flowid)
483 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
484 	sc->flowid_shift = V_def_flowid_shift;
485 
486 	/* Hash all layers by default */
487 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
488 
489 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
490 
491 	SLIST_INIT(&sc->sc_ports);
492 
493 	/* Initialise pseudo media types */
494 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
495 	    lagg_media_status);
496 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
497 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
498 
499 	if_initname(ifp, laggname, unit);
500 	ifp->if_softc = sc;
501 	ifp->if_transmit = lagg_transmit;
502 	ifp->if_qflush = lagg_qflush;
503 	ifp->if_init = lagg_init;
504 	ifp->if_ioctl = lagg_ioctl;
505 	ifp->if_get_counter = lagg_get_counter;
506 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
507 #ifdef RATELIMIT
508 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
509 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS | IFCAP_TXRTLMT;
510 #else
511 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
512 #endif
513 
514 	/*
515 	 * Attach as an ordinary ethernet device, children will be attached
516 	 * as special device IFT_IEEE8023ADLAG.
517 	 */
518 	ether_ifattach(ifp, eaddr);
519 
520 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
521 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
522 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
523 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
524 
525 	/* Insert into the global list of laggs */
526 	LAGG_LIST_LOCK();
527 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
528 	LAGG_LIST_UNLOCK();
529 	LAGG_XUNLOCK(sc);
530 
531 	return (0);
532 }
533 
534 static void
535 lagg_clone_destroy(struct ifnet *ifp)
536 {
537 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
538 	struct lagg_port *lp;
539 
540 	LAGG_XLOCK(sc);
541 	sc->sc_destroying = 1;
542 	lagg_stop(sc);
543 	ifp->if_flags &= ~IFF_UP;
544 
545 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
546 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
547 
548 	/* Shutdown and remove lagg ports */
549 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
550 		lagg_port_destroy(lp, 1);
551 
552 	/* Unhook the aggregation protocol */
553 	LAGG_WLOCK(sc);
554 	lagg_proto_detach(sc);
555 	LAGG_UNLOCK_ASSERT(sc);
556 	LAGG_XUNLOCK(sc);
557 
558 	ifmedia_removeall(&sc->sc_media);
559 	ether_ifdetach(ifp);
560 	if_free(ifp);
561 
562 	LAGG_LIST_LOCK();
563 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
564 	LAGG_LIST_UNLOCK();
565 
566 	LAGG_SX_DESTROY(sc);
567 	LAGG_LOCK_DESTROY(sc);
568 	free(sc, M_DEVBUF);
569 }
570 
571 static void
572 lagg_capabilities(struct lagg_softc *sc)
573 {
574 	struct lagg_port *lp;
575 	int cap, ena, pena;
576 	uint64_t hwa;
577 	struct ifnet_hw_tsomax hw_tsomax;
578 
579 	LAGG_XLOCK_ASSERT(sc);
580 
581 	/* Get common enabled capabilities for the lagg ports */
582 	ena = ~0;
583 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
584 		ena &= lp->lp_ifp->if_capenable;
585 	ena = (ena == ~0 ? 0 : ena);
586 
587 	/*
588 	 * Apply common enabled capabilities back to the lagg ports.
589 	 * May require several iterations if they are dependent.
590 	 */
591 	do {
592 		pena = ena;
593 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
594 			lagg_setcaps(lp, ena);
595 			ena &= lp->lp_ifp->if_capenable;
596 		}
597 	} while (pena != ena);
598 
599 	/* Get other capabilities from the lagg ports */
600 	cap = ~0;
601 	hwa = ~(uint64_t)0;
602 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
603 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
604 		cap &= lp->lp_ifp->if_capabilities;
605 		hwa &= lp->lp_ifp->if_hwassist;
606 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
607 	}
608 	cap = (cap == ~0 ? 0 : cap);
609 	hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
610 
611 	if (sc->sc_ifp->if_capabilities != cap ||
612 	    sc->sc_ifp->if_capenable != ena ||
613 	    sc->sc_ifp->if_hwassist != hwa ||
614 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
615 		sc->sc_ifp->if_capabilities = cap;
616 		sc->sc_ifp->if_capenable = ena;
617 		sc->sc_ifp->if_hwassist = hwa;
618 		getmicrotime(&sc->sc_ifp->if_lastchange);
619 
620 		if (sc->sc_ifflags & IFF_DEBUG)
621 			if_printf(sc->sc_ifp,
622 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
623 	}
624 }
625 
626 static int
627 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
628 {
629 	struct lagg_softc *sc_ptr;
630 	struct lagg_port *lp, *tlp;
631 	int error, i;
632 	uint64_t *pval;
633 
634 	LAGG_XLOCK_ASSERT(sc);
635 
636 	/* Limit the maximal number of lagg ports */
637 	if (sc->sc_count >= LAGG_MAX_PORTS)
638 		return (ENOSPC);
639 
640 	/* Check if port has already been associated to a lagg */
641 	if (ifp->if_lagg != NULL) {
642 		/* Port is already in the current lagg? */
643 		lp = (struct lagg_port *)ifp->if_lagg;
644 		if (lp->lp_softc == sc)
645 			return (EEXIST);
646 		return (EBUSY);
647 	}
648 
649 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
650 	if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
651 		return (EPROTONOSUPPORT);
652 
653 	/* Allow the first Ethernet member to define the MTU */
654 	if (SLIST_EMPTY(&sc->sc_ports))
655 		sc->sc_ifp->if_mtu = ifp->if_mtu;
656 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
657 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
658 		    ifp->if_xname);
659 		return (EINVAL);
660 	}
661 
662 	lp = malloc(sizeof(struct lagg_port), M_DEVBUF, M_WAITOK|M_ZERO);
663 	lp->lp_softc = sc;
664 
665 	/* Check if port is a stacked lagg */
666 	LAGG_LIST_LOCK();
667 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
668 		if (ifp == sc_ptr->sc_ifp) {
669 			LAGG_LIST_UNLOCK();
670 			free(lp, M_DEVBUF);
671 			return (EINVAL);
672 			/* XXX disable stacking for the moment, its untested */
673 #ifdef LAGG_PORT_STACKING
674 			lp->lp_flags |= LAGG_PORT_STACK;
675 			if (lagg_port_checkstacking(sc_ptr) >=
676 			    LAGG_MAX_STACKING) {
677 				LAGG_LIST_UNLOCK();
678 				free(lp, M_DEVBUF);
679 				return (E2BIG);
680 			}
681 #endif
682 		}
683 	}
684 	LAGG_LIST_UNLOCK();
685 
686 	if_ref(ifp);
687 	lp->lp_ifp = ifp;
688 
689 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
690 	lp->lp_ifcapenable = ifp->if_capenable;
691 	if (SLIST_EMPTY(&sc->sc_ports)) {
692 		LAGG_WLOCK(sc);
693 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
694 		lagg_proto_lladdr(sc);
695 		LAGG_WUNLOCK(sc);
696 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
697 	} else {
698 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
699 	}
700 	lagg_setflags(lp, 1);
701 
702 	LAGG_WLOCK(sc);
703 	if (SLIST_EMPTY(&sc->sc_ports))
704 		sc->sc_primary = lp;
705 
706 	/* Change the interface type */
707 	lp->lp_iftype = ifp->if_type;
708 	ifp->if_type = IFT_IEEE8023ADLAG;
709 	ifp->if_lagg = lp;
710 	lp->lp_ioctl = ifp->if_ioctl;
711 	ifp->if_ioctl = lagg_port_ioctl;
712 	lp->lp_output = ifp->if_output;
713 	ifp->if_output = lagg_port_output;
714 
715 	/* Read port counters */
716 	pval = lp->port_counters.val;
717 	for (i = 0; i < IFCOUNTERS; i++, pval++)
718 		*pval = ifp->if_get_counter(ifp, i);
719 
720 	/*
721 	 * Insert into the list of ports.
722 	 * Keep ports sorted by if_index. It is handy, when configuration
723 	 * is predictable and `ifconfig laggN create ...` command
724 	 * will lead to the same result each time.
725 	 */
726 	SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
727 		if (tlp->lp_ifp->if_index < ifp->if_index && (
728 		    SLIST_NEXT(tlp, lp_entries) == NULL ||
729 		    SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index >
730 		    ifp->if_index))
731 			break;
732 	}
733 	if (tlp != NULL)
734 		SLIST_INSERT_AFTER(tlp, lp, lp_entries);
735 	else
736 		SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
737 	sc->sc_count++;
738 
739 	lagg_setmulti(lp);
740 
741 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
742 		/* Remove the port, without calling pr_delport. */
743 		lagg_port_destroy(lp, 0);
744 		LAGG_UNLOCK_ASSERT(sc);
745 		return (error);
746 	}
747 
748 	LAGG_WUNLOCK(sc);
749 
750 	/* Update lagg capabilities */
751 	lagg_capabilities(sc);
752 	lagg_linkstate(sc);
753 
754 	return (0);
755 }
756 
757 #ifdef LAGG_PORT_STACKING
758 static int
759 lagg_port_checkstacking(struct lagg_softc *sc)
760 {
761 	struct lagg_softc *sc_ptr;
762 	struct lagg_port *lp;
763 	int m = 0;
764 
765 	LAGG_SXLOCK_ASSERT(sc);
766 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
767 		if (lp->lp_flags & LAGG_PORT_STACK) {
768 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
769 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
770 		}
771 	}
772 
773 	return (m + 1);
774 }
775 #endif
776 
777 static int
778 lagg_port_destroy(struct lagg_port *lp, int rundelport)
779 {
780 	struct lagg_softc *sc = lp->lp_softc;
781 	struct lagg_port *lp_ptr, *lp0;
782 	struct ifnet *ifp = lp->lp_ifp;
783 	uint64_t *pval, vdiff;
784 	int i;
785 
786 	LAGG_XLOCK_ASSERT(sc);
787 
788 	if (rundelport) {
789 		LAGG_WLOCK(sc);
790 		lagg_proto_delport(sc, lp);
791 	} else
792 		LAGG_WLOCK_ASSERT(sc);
793 
794 	if (lp->lp_detaching == 0)
795 		lagg_clrmulti(lp);
796 
797 	/* Restore interface */
798 	ifp->if_type = lp->lp_iftype;
799 	ifp->if_ioctl = lp->lp_ioctl;
800 	ifp->if_output = lp->lp_output;
801 	ifp->if_lagg = NULL;
802 
803 	/* Update detached port counters */
804 	pval = lp->port_counters.val;
805 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
806 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
807 		sc->detached_counters.val[i] += vdiff;
808 	}
809 
810 	/* Finally, remove the port from the lagg */
811 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
812 	sc->sc_count--;
813 
814 	/* Update the primary interface */
815 	if (lp == sc->sc_primary) {
816 		uint8_t lladdr[ETHER_ADDR_LEN];
817 
818 		if ((lp0 = SLIST_FIRST(&sc->sc_ports)) == NULL)
819 			bzero(&lladdr, ETHER_ADDR_LEN);
820 		else
821 			bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN);
822 		sc->sc_primary = lp0;
823 		if (sc->sc_destroying == 0) {
824 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
825 			lagg_proto_lladdr(sc);
826 			LAGG_WUNLOCK(sc);
827 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
828 		} else
829 			LAGG_WUNLOCK(sc);
830 
831 		/*
832 		 * Update lladdr for each port (new primary needs update
833 		 * as well, to switch from old lladdr to its 'real' one)
834 		 */
835 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
836 			if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN);
837 	} else
838 		LAGG_WUNLOCK(sc);
839 
840 	if (lp->lp_ifflags)
841 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
842 
843 	if (lp->lp_detaching == 0) {
844 		lagg_setflags(lp, 0);
845 		lagg_setcaps(lp, lp->lp_ifcapenable);
846 		if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN);
847 	}
848 
849 	if_rele(ifp);
850 	free(lp, M_DEVBUF);
851 
852 	/* Update lagg capabilities */
853 	lagg_capabilities(sc);
854 	lagg_linkstate(sc);
855 
856 	return (0);
857 }
858 
859 static int
860 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
861 {
862 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
863 	struct lagg_softc *sc;
864 	struct lagg_port *lp = NULL;
865 	int error = 0;
866 
867 	/* Should be checked by the caller */
868 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
869 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
870 		goto fallback;
871 
872 	switch (cmd) {
873 	case SIOCGLAGGPORT:
874 		if (rp->rp_portname[0] == '\0' ||
875 		    ifunit(rp->rp_portname) != ifp) {
876 			error = EINVAL;
877 			break;
878 		}
879 
880 		LAGG_SLOCK(sc);
881 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
882 			error = ENOENT;
883 			LAGG_SUNLOCK(sc);
884 			break;
885 		}
886 
887 		lagg_port2req(lp, rp);
888 		LAGG_SUNLOCK(sc);
889 		break;
890 
891 	case SIOCSIFCAP:
892 		if (lp->lp_ioctl == NULL) {
893 			error = EINVAL;
894 			break;
895 		}
896 		error = (*lp->lp_ioctl)(ifp, cmd, data);
897 		if (error)
898 			break;
899 
900 		/* Update lagg interface capabilities */
901 		LAGG_XLOCK(sc);
902 		lagg_capabilities(sc);
903 		LAGG_XUNLOCK(sc);
904 		VLAN_CAPABILITIES(sc->sc_ifp);
905 		break;
906 
907 	case SIOCSIFMTU:
908 		/* Do not allow the MTU to be changed once joined */
909 		error = EINVAL;
910 		break;
911 
912 	default:
913 		goto fallback;
914 	}
915 
916 	return (error);
917 
918 fallback:
919 	if (lp != NULL && lp->lp_ioctl != NULL)
920 		return ((*lp->lp_ioctl)(ifp, cmd, data));
921 
922 	return (EINVAL);
923 }
924 
925 /*
926  * Requests counter @cnt data.
927  *
928  * Counter value is calculated the following way:
929  * 1) for each port, sum  difference between current and "initial" measurements.
930  * 2) add lagg logical interface counters.
931  * 3) add data from detached_counters array.
932  *
933  * We also do the following things on ports attach/detach:
934  * 1) On port attach we store all counters it has into port_counter array.
935  * 2) On port detach we add the different between "initial" and
936  *   current counters data to detached_counters array.
937  */
938 static uint64_t
939 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
940 {
941 	struct lagg_softc *sc;
942 	struct lagg_port *lp;
943 	struct ifnet *lpifp;
944 	struct rm_priotracker tracker;
945 	uint64_t newval, oldval, vsum;
946 
947 	/* Revise this when we've got non-generic counters. */
948 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
949 
950 	sc = (struct lagg_softc *)ifp->if_softc;
951 	LAGG_RLOCK(sc, &tracker);
952 
953 	vsum = 0;
954 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
955 		/* Saved attached value */
956 		oldval = lp->port_counters.val[cnt];
957 		/* current value */
958 		lpifp = lp->lp_ifp;
959 		newval = lpifp->if_get_counter(lpifp, cnt);
960 		/* Calculate diff and save new */
961 		vsum += newval - oldval;
962 	}
963 
964 	/*
965 	 * Add counter data which might be added by upper
966 	 * layer protocols operating on logical interface.
967 	 */
968 	vsum += if_get_counter_default(ifp, cnt);
969 
970 	/*
971 	 * Add counter data from detached ports counters
972 	 */
973 	vsum += sc->detached_counters.val[cnt];
974 
975 	LAGG_RUNLOCK(sc, &tracker);
976 
977 	return (vsum);
978 }
979 
980 /*
981  * For direct output to child ports.
982  */
983 static int
984 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
985 	const struct sockaddr *dst, struct route *ro)
986 {
987 	struct lagg_port *lp = ifp->if_lagg;
988 
989 	switch (dst->sa_family) {
990 		case pseudo_AF_HDRCMPLT:
991 		case AF_UNSPEC:
992 			return ((*lp->lp_output)(ifp, m, dst, ro));
993 	}
994 
995 	/* drop any other frames */
996 	m_freem(m);
997 	return (ENETDOWN);
998 }
999 
1000 static void
1001 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1002 {
1003 	struct lagg_port *lp;
1004 	struct lagg_softc *sc;
1005 
1006 	if ((lp = ifp->if_lagg) == NULL)
1007 		return;
1008 	/* If the ifnet is just being renamed, don't do anything. */
1009 	if (ifp->if_flags & IFF_RENAMING)
1010 		return;
1011 
1012 	sc = lp->lp_softc;
1013 
1014 	LAGG_XLOCK(sc);
1015 	lp->lp_detaching = 1;
1016 	lagg_port_destroy(lp, 1);
1017 	LAGG_XUNLOCK(sc);
1018 	VLAN_CAPABILITIES(sc->sc_ifp);
1019 }
1020 
1021 static void
1022 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1023 {
1024 	struct lagg_softc *sc = lp->lp_softc;
1025 
1026 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1027 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1028 	rp->rp_prio = lp->lp_prio;
1029 	rp->rp_flags = lp->lp_flags;
1030 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1031 
1032 	/* Add protocol specific flags */
1033 	switch (sc->sc_proto) {
1034 		case LAGG_PROTO_FAILOVER:
1035 			if (lp == sc->sc_primary)
1036 				rp->rp_flags |= LAGG_PORT_MASTER;
1037 			if (lp == lagg_link_active(sc, sc->sc_primary))
1038 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1039 			break;
1040 
1041 		case LAGG_PROTO_ROUNDROBIN:
1042 		case LAGG_PROTO_LOADBALANCE:
1043 		case LAGG_PROTO_BROADCAST:
1044 			if (LAGG_PORTACTIVE(lp))
1045 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1046 			break;
1047 
1048 		case LAGG_PROTO_LACP:
1049 			/* LACP has a different definition of active */
1050 			if (lacp_isactive(lp))
1051 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1052 			if (lacp_iscollecting(lp))
1053 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1054 			if (lacp_isdistributing(lp))
1055 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1056 			break;
1057 	}
1058 
1059 }
1060 
1061 static void
1062 lagg_init(void *xsc)
1063 {
1064 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1065 	struct ifnet *ifp = sc->sc_ifp;
1066 	struct lagg_port *lp;
1067 
1068 	LAGG_XLOCK(sc);
1069 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1070 		LAGG_XUNLOCK(sc);
1071 		return;
1072 	}
1073 
1074 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1075 
1076 	/*
1077 	 * Update the port lladdrs if needed.
1078 	 * This might be if_setlladdr() notification
1079 	 * that lladdr has been changed.
1080 	 */
1081 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1082 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1083 		    ETHER_ADDR_LEN) != 0)
1084 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN);
1085 	}
1086 
1087 	lagg_proto_init(sc);
1088 
1089 	LAGG_XUNLOCK(sc);
1090 }
1091 
1092 static void
1093 lagg_stop(struct lagg_softc *sc)
1094 {
1095 	struct ifnet *ifp = sc->sc_ifp;
1096 
1097 	LAGG_XLOCK_ASSERT(sc);
1098 
1099 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1100 		return;
1101 
1102 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1103 
1104 	lagg_proto_stop(sc);
1105 }
1106 
1107 static int
1108 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1109 {
1110 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1111 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1112 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1113 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1114 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1115 	struct ifreq *ifr = (struct ifreq *)data;
1116 	struct lagg_port *lp;
1117 	struct ifnet *tpif;
1118 	struct thread *td = curthread;
1119 	char *buf, *outbuf;
1120 	int count, buflen, len, error = 0;
1121 
1122 	bzero(&rpbuf, sizeof(rpbuf));
1123 
1124 	switch (cmd) {
1125 	case SIOCGLAGG:
1126 		LAGG_SLOCK(sc);
1127 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1128 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1129 		ra->ra_proto = sc->sc_proto;
1130 		lagg_proto_request(sc, &ra->ra_psc);
1131 		count = 0;
1132 		buf = outbuf;
1133 		len = min(ra->ra_size, buflen);
1134 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1135 			if (len < sizeof(rpbuf))
1136 				break;
1137 
1138 			lagg_port2req(lp, &rpbuf);
1139 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1140 			count++;
1141 			buf += sizeof(rpbuf);
1142 			len -= sizeof(rpbuf);
1143 		}
1144 		LAGG_SUNLOCK(sc);
1145 		ra->ra_ports = count;
1146 		ra->ra_size = count * sizeof(rpbuf);
1147 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1148 		free(outbuf, M_TEMP);
1149 		break;
1150 	case SIOCSLAGG:
1151 		error = priv_check(td, PRIV_NET_LAGG);
1152 		if (error)
1153 			break;
1154 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1155 			error = EPROTONOSUPPORT;
1156 			break;
1157 		}
1158 
1159 		LAGG_XLOCK(sc);
1160 		LAGG_WLOCK(sc);
1161 		lagg_proto_detach(sc);
1162 		LAGG_UNLOCK_ASSERT(sc);
1163 		lagg_proto_attach(sc, ra->ra_proto);
1164 		LAGG_XUNLOCK(sc);
1165 		break;
1166 	case SIOCGLAGGOPTS:
1167 		LAGG_SLOCK(sc);
1168 		ro->ro_opts = sc->sc_opts;
1169 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1170 			struct lacp_softc *lsc;
1171 
1172 			lsc = (struct lacp_softc *)sc->sc_psc;
1173 			if (lsc->lsc_debug.lsc_tx_test != 0)
1174 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1175 			if (lsc->lsc_debug.lsc_rx_test != 0)
1176 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1177 			if (lsc->lsc_strict_mode != 0)
1178 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1179 			if (lsc->lsc_fast_timeout != 0)
1180 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1181 
1182 			ro->ro_active = sc->sc_active;
1183 		} else {
1184 			ro->ro_active = 0;
1185 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1186 				ro->ro_active += LAGG_PORTACTIVE(lp);
1187 		}
1188 		ro->ro_bkt = sc->sc_bkt;
1189 		ro->ro_flapping = sc->sc_flapping;
1190 		ro->ro_flowid_shift = sc->flowid_shift;
1191 		LAGG_SUNLOCK(sc);
1192 		break;
1193 	case SIOCSLAGGOPTS:
1194 		if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) {
1195 			if (ro->ro_bkt == 0)
1196 				sc->sc_bkt = 1; // Minimum 1 packet per iface.
1197 			else
1198 				sc->sc_bkt = ro->ro_bkt;
1199 		}
1200 		error = priv_check(td, PRIV_NET_LAGG);
1201 		if (error)
1202 			break;
1203 		if (ro->ro_opts == 0)
1204 			break;
1205 		/*
1206 		 * Set options.  LACP options are stored in sc->sc_psc,
1207 		 * not in sc_opts.
1208 		 */
1209 		int valid, lacp;
1210 
1211 		switch (ro->ro_opts) {
1212 		case LAGG_OPT_USE_FLOWID:
1213 		case -LAGG_OPT_USE_FLOWID:
1214 		case LAGG_OPT_FLOWIDSHIFT:
1215 			valid = 1;
1216 			lacp = 0;
1217 			break;
1218 		case LAGG_OPT_LACP_TXTEST:
1219 		case -LAGG_OPT_LACP_TXTEST:
1220 		case LAGG_OPT_LACP_RXTEST:
1221 		case -LAGG_OPT_LACP_RXTEST:
1222 		case LAGG_OPT_LACP_STRICT:
1223 		case -LAGG_OPT_LACP_STRICT:
1224 		case LAGG_OPT_LACP_TIMEOUT:
1225 		case -LAGG_OPT_LACP_TIMEOUT:
1226 			valid = lacp = 1;
1227 			break;
1228 		default:
1229 			valid = lacp = 0;
1230 			break;
1231 		}
1232 
1233 		LAGG_XLOCK(sc);
1234 
1235 		if (valid == 0 ||
1236 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1237 			/* Invalid combination of options specified. */
1238 			error = EINVAL;
1239 			LAGG_XUNLOCK(sc);
1240 			break;	/* Return from SIOCSLAGGOPTS. */
1241 		}
1242 		/*
1243 		 * Store new options into sc->sc_opts except for
1244 		 * FLOWIDSHIFT and LACP options.
1245 		 */
1246 		if (lacp == 0) {
1247 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1248 				sc->flowid_shift = ro->ro_flowid_shift;
1249 			else if (ro->ro_opts > 0)
1250 				sc->sc_opts |= ro->ro_opts;
1251 			else
1252 				sc->sc_opts &= ~ro->ro_opts;
1253 		} else {
1254 			struct lacp_softc *lsc;
1255 			struct lacp_port *lp;
1256 
1257 			lsc = (struct lacp_softc *)sc->sc_psc;
1258 
1259 			switch (ro->ro_opts) {
1260 			case LAGG_OPT_LACP_TXTEST:
1261 				lsc->lsc_debug.lsc_tx_test = 1;
1262 				break;
1263 			case -LAGG_OPT_LACP_TXTEST:
1264 				lsc->lsc_debug.lsc_tx_test = 0;
1265 				break;
1266 			case LAGG_OPT_LACP_RXTEST:
1267 				lsc->lsc_debug.lsc_rx_test = 1;
1268 				break;
1269 			case -LAGG_OPT_LACP_RXTEST:
1270 				lsc->lsc_debug.lsc_rx_test = 0;
1271 				break;
1272 			case LAGG_OPT_LACP_STRICT:
1273 				lsc->lsc_strict_mode = 1;
1274 				break;
1275 			case -LAGG_OPT_LACP_STRICT:
1276 				lsc->lsc_strict_mode = 0;
1277 				break;
1278 			case LAGG_OPT_LACP_TIMEOUT:
1279 				LACP_LOCK(lsc);
1280         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1281                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1282 				LACP_UNLOCK(lsc);
1283 				lsc->lsc_fast_timeout = 1;
1284 				break;
1285 			case -LAGG_OPT_LACP_TIMEOUT:
1286 				LACP_LOCK(lsc);
1287         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1288                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1289 				LACP_UNLOCK(lsc);
1290 				lsc->lsc_fast_timeout = 0;
1291 				break;
1292 			}
1293 		}
1294 		LAGG_XUNLOCK(sc);
1295 		break;
1296 	case SIOCGLAGGFLAGS:
1297 		rf->rf_flags = 0;
1298 		LAGG_SLOCK(sc);
1299 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1300 			rf->rf_flags |= LAGG_F_HASHL2;
1301 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1302 			rf->rf_flags |= LAGG_F_HASHL3;
1303 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1304 			rf->rf_flags |= LAGG_F_HASHL4;
1305 		LAGG_SUNLOCK(sc);
1306 		break;
1307 	case SIOCSLAGGHASH:
1308 		error = priv_check(td, PRIV_NET_LAGG);
1309 		if (error)
1310 			break;
1311 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1312 			error = EINVAL;
1313 			break;
1314 		}
1315 		LAGG_XLOCK(sc);
1316 		sc->sc_flags = 0;
1317 		if (rf->rf_flags & LAGG_F_HASHL2)
1318 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1319 		if (rf->rf_flags & LAGG_F_HASHL3)
1320 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1321 		if (rf->rf_flags & LAGG_F_HASHL4)
1322 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1323 		LAGG_XUNLOCK(sc);
1324 		break;
1325 	case SIOCGLAGGPORT:
1326 		if (rp->rp_portname[0] == '\0' ||
1327 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1328 			error = EINVAL;
1329 			break;
1330 		}
1331 
1332 		LAGG_SLOCK(sc);
1333 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1334 		    lp->lp_softc != sc) {
1335 			error = ENOENT;
1336 			LAGG_SUNLOCK(sc);
1337 			if_rele(tpif);
1338 			break;
1339 		}
1340 
1341 		lagg_port2req(lp, rp);
1342 		LAGG_SUNLOCK(sc);
1343 		if_rele(tpif);
1344 		break;
1345 	case SIOCSLAGGPORT:
1346 		error = priv_check(td, PRIV_NET_LAGG);
1347 		if (error)
1348 			break;
1349 		if (rp->rp_portname[0] == '\0' ||
1350 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1351 			error = EINVAL;
1352 			break;
1353 		}
1354 #ifdef INET6
1355 		/*
1356 		 * A laggport interface should not have inet6 address
1357 		 * because two interfaces with a valid link-local
1358 		 * scope zone must not be merged in any form.  This
1359 		 * restriction is needed to prevent violation of
1360 		 * link-local scope zone.  Attempts to add a laggport
1361 		 * interface which has inet6 addresses triggers
1362 		 * removal of all inet6 addresses on the member
1363 		 * interface.
1364 		 */
1365 		if (in6ifa_llaonifp(tpif)) {
1366 			in6_ifdetach(tpif);
1367 				if_printf(sc->sc_ifp,
1368 				    "IPv6 addresses on %s have been removed "
1369 				    "before adding it as a member to prevent "
1370 				    "IPv6 address scope violation.\n",
1371 				    tpif->if_xname);
1372 		}
1373 #endif
1374 		LAGG_XLOCK(sc);
1375 		error = lagg_port_create(sc, tpif);
1376 		LAGG_XUNLOCK(sc);
1377 		if_rele(tpif);
1378 		VLAN_CAPABILITIES(ifp);
1379 		break;
1380 	case SIOCSLAGGDELPORT:
1381 		error = priv_check(td, PRIV_NET_LAGG);
1382 		if (error)
1383 			break;
1384 		if (rp->rp_portname[0] == '\0' ||
1385 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1386 			error = EINVAL;
1387 			break;
1388 		}
1389 
1390 		LAGG_XLOCK(sc);
1391 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1392 		    lp->lp_softc != sc) {
1393 			error = ENOENT;
1394 			LAGG_XUNLOCK(sc);
1395 			if_rele(tpif);
1396 			break;
1397 		}
1398 
1399 		error = lagg_port_destroy(lp, 1);
1400 		LAGG_XUNLOCK(sc);
1401 		if_rele(tpif);
1402 		VLAN_CAPABILITIES(ifp);
1403 		break;
1404 	case SIOCSIFFLAGS:
1405 		/* Set flags on ports too */
1406 		LAGG_XLOCK(sc);
1407 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1408 			lagg_setflags(lp, 1);
1409 		}
1410 
1411 		if (!(ifp->if_flags & IFF_UP) &&
1412 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1413 			/*
1414 			 * If interface is marked down and it is running,
1415 			 * then stop and disable it.
1416 			 */
1417 			lagg_stop(sc);
1418 			LAGG_XUNLOCK(sc);
1419 		} else if ((ifp->if_flags & IFF_UP) &&
1420 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1421 			/*
1422 			 * If interface is marked up and it is stopped, then
1423 			 * start it.
1424 			 */
1425 			LAGG_XUNLOCK(sc);
1426 			(*ifp->if_init)(sc);
1427 		} else
1428 			LAGG_XUNLOCK(sc);
1429 		break;
1430 	case SIOCADDMULTI:
1431 	case SIOCDELMULTI:
1432 		LAGG_WLOCK(sc);
1433 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1434 			lagg_clrmulti(lp);
1435 			lagg_setmulti(lp);
1436 		}
1437 		LAGG_WUNLOCK(sc);
1438 		error = 0;
1439 		break;
1440 	case SIOCSIFMEDIA:
1441 	case SIOCGIFMEDIA:
1442 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1443 		break;
1444 
1445 	case SIOCSIFCAP:
1446 		LAGG_XLOCK(sc);
1447 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1448 			if (lp->lp_ioctl != NULL)
1449 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1450 		}
1451 		lagg_capabilities(sc);
1452 		LAGG_XUNLOCK(sc);
1453 		VLAN_CAPABILITIES(ifp);
1454 		error = 0;
1455 		break;
1456 
1457 	case SIOCSIFMTU:
1458 		/* Do not allow the MTU to be directly changed */
1459 		error = EINVAL;
1460 		break;
1461 
1462 	default:
1463 		error = ether_ioctl(ifp, cmd, data);
1464 		break;
1465 	}
1466 	return (error);
1467 }
1468 
1469 #ifdef RATELIMIT
1470 static int
1471 lagg_snd_tag_alloc(struct ifnet *ifp,
1472     union if_snd_tag_alloc_params *params,
1473     struct m_snd_tag **ppmt)
1474 {
1475 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1476 	struct lagg_port *lp;
1477 	struct lagg_lb *lb;
1478 	uint32_t p;
1479 
1480 	switch (sc->sc_proto) {
1481 	case LAGG_PROTO_FAILOVER:
1482 		lp = lagg_link_active(sc, sc->sc_primary);
1483 		break;
1484 	case LAGG_PROTO_LOADBALANCE:
1485 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1486 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1487 			return (EOPNOTSUPP);
1488 		p = params->hdr.flowid >> sc->flowid_shift;
1489 		p %= sc->sc_count;
1490 		lb = (struct lagg_lb *)sc->sc_psc;
1491 		lp = lb->lb_ports[p];
1492 		lp = lagg_link_active(sc, lp);
1493 		break;
1494 	case LAGG_PROTO_LACP:
1495 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1496 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1497 			return (EOPNOTSUPP);
1498 		lp = lacp_select_tx_port_by_hash(sc, params->hdr.flowid);
1499 		break;
1500 	default:
1501 		return (EOPNOTSUPP);
1502 	}
1503 	if (lp == NULL)
1504 		return (EOPNOTSUPP);
1505 	ifp = lp->lp_ifp;
1506 	if (ifp == NULL || ifp->if_snd_tag_alloc == NULL ||
1507 	    (ifp->if_capenable & IFCAP_TXRTLMT) == 0)
1508 		return (EOPNOTSUPP);
1509 
1510 	/* forward allocation request */
1511 	return (ifp->if_snd_tag_alloc(ifp, params, ppmt));
1512 }
1513 #endif
1514 
1515 static int
1516 lagg_setmulti(struct lagg_port *lp)
1517 {
1518 	struct lagg_softc *sc = lp->lp_softc;
1519 	struct ifnet *ifp = lp->lp_ifp;
1520 	struct ifnet *scifp = sc->sc_ifp;
1521 	struct lagg_mc *mc;
1522 	struct ifmultiaddr *ifma;
1523 	int error;
1524 
1525 	LAGG_WLOCK_ASSERT(sc);
1526 	IF_ADDR_WLOCK(scifp);
1527 	TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1528 		if (ifma->ifma_addr->sa_family != AF_LINK)
1529 			continue;
1530 		mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1531 		if (mc == NULL) {
1532 			IF_ADDR_WUNLOCK(scifp);
1533 			return (ENOMEM);
1534 		}
1535 		bcopy(ifma->ifma_addr, &mc->mc_addr,
1536 		    ifma->ifma_addr->sa_len);
1537 		mc->mc_addr.sdl_index = ifp->if_index;
1538 		mc->mc_ifma = NULL;
1539 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1540 	}
1541 	IF_ADDR_WUNLOCK(scifp);
1542 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1543 		error = if_addmulti(ifp,
1544 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1545 		if (error)
1546 			return (error);
1547 	}
1548 	return (0);
1549 }
1550 
1551 static int
1552 lagg_clrmulti(struct lagg_port *lp)
1553 {
1554 	struct lagg_mc *mc;
1555 
1556 	LAGG_WLOCK_ASSERT(lp->lp_softc);
1557 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1558 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1559 		if (mc->mc_ifma && lp->lp_detaching == 0)
1560 			if_delmulti_ifma(mc->mc_ifma);
1561 		free(mc, M_DEVBUF);
1562 	}
1563 	return (0);
1564 }
1565 
1566 static int
1567 lagg_setcaps(struct lagg_port *lp, int cap)
1568 {
1569 	struct ifreq ifr;
1570 
1571 	if (lp->lp_ifp->if_capenable == cap)
1572 		return (0);
1573 	if (lp->lp_ioctl == NULL)
1574 		return (ENXIO);
1575 	ifr.ifr_reqcap = cap;
1576 	return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1577 }
1578 
1579 /* Handle a ref counted flag that should be set on the lagg port as well */
1580 static int
1581 lagg_setflag(struct lagg_port *lp, int flag, int status,
1582     int (*func)(struct ifnet *, int))
1583 {
1584 	struct lagg_softc *sc = lp->lp_softc;
1585 	struct ifnet *scifp = sc->sc_ifp;
1586 	struct ifnet *ifp = lp->lp_ifp;
1587 	int error;
1588 
1589 	LAGG_XLOCK_ASSERT(sc);
1590 
1591 	status = status ? (scifp->if_flags & flag) : 0;
1592 	/* Now "status" contains the flag value or 0 */
1593 
1594 	/*
1595 	 * See if recorded ports status is different from what
1596 	 * we want it to be.  If it is, flip it.  We record ports
1597 	 * status in lp_ifflags so that we won't clear ports flag
1598 	 * we haven't set.  In fact, we don't clear or set ports
1599 	 * flags directly, but get or release references to them.
1600 	 * That's why we can be sure that recorded flags still are
1601 	 * in accord with actual ports flags.
1602 	 */
1603 	if (status != (lp->lp_ifflags & flag)) {
1604 		error = (*func)(ifp, status);
1605 		if (error)
1606 			return (error);
1607 		lp->lp_ifflags &= ~flag;
1608 		lp->lp_ifflags |= status;
1609 	}
1610 	return (0);
1611 }
1612 
1613 /*
1614  * Handle IFF_* flags that require certain changes on the lagg port
1615  * if "status" is true, update ports flags respective to the lagg
1616  * if "status" is false, forcedly clear the flags set on port.
1617  */
1618 static int
1619 lagg_setflags(struct lagg_port *lp, int status)
1620 {
1621 	int error, i;
1622 
1623 	for (i = 0; lagg_pflags[i].flag; i++) {
1624 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1625 		    status, lagg_pflags[i].func);
1626 		if (error)
1627 			return (error);
1628 	}
1629 	return (0);
1630 }
1631 
1632 static int
1633 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1634 {
1635 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1636 	int error, len, mcast;
1637 	struct rm_priotracker tracker;
1638 
1639 	len = m->m_pkthdr.len;
1640 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1641 
1642 	LAGG_RLOCK(sc, &tracker);
1643 	/* We need a Tx algorithm and at least one port */
1644 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1645 		LAGG_RUNLOCK(sc, &tracker);
1646 		m_freem(m);
1647 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1648 		return (ENXIO);
1649 	}
1650 
1651 	ETHER_BPF_MTAP(ifp, m);
1652 
1653 	error = lagg_proto_start(sc, m);
1654 	LAGG_RUNLOCK(sc, &tracker);
1655 
1656 	if (error != 0)
1657 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1658 
1659 	return (error);
1660 }
1661 
1662 /*
1663  * The ifp->if_qflush entry point for lagg(4) is no-op.
1664  */
1665 static void
1666 lagg_qflush(struct ifnet *ifp __unused)
1667 {
1668 }
1669 
1670 static struct mbuf *
1671 lagg_input(struct ifnet *ifp, struct mbuf *m)
1672 {
1673 	struct lagg_port *lp = ifp->if_lagg;
1674 	struct lagg_softc *sc = lp->lp_softc;
1675 	struct ifnet *scifp = sc->sc_ifp;
1676 	struct rm_priotracker tracker;
1677 
1678 	LAGG_RLOCK(sc, &tracker);
1679 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1680 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1681 	    sc->sc_proto == LAGG_PROTO_NONE) {
1682 		LAGG_RUNLOCK(sc, &tracker);
1683 		m_freem(m);
1684 		return (NULL);
1685 	}
1686 
1687 	ETHER_BPF_MTAP(scifp, m);
1688 
1689 	if (lp->lp_detaching != 0) {
1690 		m_freem(m);
1691 		m = NULL;
1692 	} else
1693 		m = lagg_proto_input(sc, lp, m);
1694 
1695 	if (m != NULL) {
1696 		if (scifp->if_flags & IFF_MONITOR) {
1697 			m_freem(m);
1698 			m = NULL;
1699 		}
1700 	}
1701 
1702 	LAGG_RUNLOCK(sc, &tracker);
1703 	return (m);
1704 }
1705 
1706 static int
1707 lagg_media_change(struct ifnet *ifp)
1708 {
1709 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1710 
1711 	if (sc->sc_ifflags & IFF_DEBUG)
1712 		printf("%s\n", __func__);
1713 
1714 	/* Ignore */
1715 	return (0);
1716 }
1717 
1718 static void
1719 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1720 {
1721 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1722 	struct lagg_port *lp;
1723 
1724 	imr->ifm_status = IFM_AVALID;
1725 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1726 
1727 	LAGG_SLOCK(sc);
1728 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1729 		if (LAGG_PORTACTIVE(lp))
1730 			imr->ifm_status |= IFM_ACTIVE;
1731 	}
1732 	LAGG_SUNLOCK(sc);
1733 }
1734 
1735 static void
1736 lagg_linkstate(struct lagg_softc *sc)
1737 {
1738 	struct lagg_port *lp;
1739 	int new_link = LINK_STATE_DOWN;
1740 	uint64_t speed;
1741 
1742 	LAGG_XLOCK_ASSERT(sc);
1743 
1744 	/* Our link is considered up if at least one of our ports is active */
1745 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1746 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1747 			new_link = LINK_STATE_UP;
1748 			break;
1749 		}
1750 	}
1751 	if_link_state_change(sc->sc_ifp, new_link);
1752 
1753 	/* Update if_baudrate to reflect the max possible speed */
1754 	switch (sc->sc_proto) {
1755 		case LAGG_PROTO_FAILOVER:
1756 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1757 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1758 			break;
1759 		case LAGG_PROTO_ROUNDROBIN:
1760 		case LAGG_PROTO_LOADBALANCE:
1761 		case LAGG_PROTO_BROADCAST:
1762 			speed = 0;
1763 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1764 				speed += lp->lp_ifp->if_baudrate;
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 = SLIST_NEXT(lp, lp_entries)) != NULL &&
1806 	    LAGG_PORTACTIVE(lp_next)) {
1807 		rval = lp_next;
1808 		goto found;
1809 	}
1810 
1811 search:
1812 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1813 		if (LAGG_PORTACTIVE(lp_next)) {
1814 			rval = lp_next;
1815 			goto found;
1816 		}
1817 	}
1818 
1819 found:
1820 	return (rval);
1821 }
1822 
1823 int
1824 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1825 {
1826 
1827 	return (ifp->if_transmit)(ifp, m);
1828 }
1829 
1830 /*
1831  * Simple round robin aggregation
1832  */
1833 static void
1834 lagg_rr_attach(struct lagg_softc *sc)
1835 {
1836 	sc->sc_seq = 0;
1837 	sc->sc_bkt_count = sc->sc_bkt;
1838 }
1839 
1840 static int
1841 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1842 {
1843 	struct lagg_port *lp;
1844 	uint32_t p;
1845 
1846 	if (sc->sc_bkt_count == 0 && sc->sc_bkt > 0)
1847 		sc->sc_bkt_count = sc->sc_bkt;
1848 
1849 	if (sc->sc_bkt > 0) {
1850 		atomic_subtract_int(&sc->sc_bkt_count, 1);
1851 	if (atomic_cmpset_int(&sc->sc_bkt_count, 0, sc->sc_bkt))
1852 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1853 	else
1854 		p = sc->sc_seq;
1855 	} else
1856 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1857 
1858 	p %= sc->sc_count;
1859 	lp = SLIST_FIRST(&sc->sc_ports);
1860 
1861 	while (p--)
1862 		lp = SLIST_NEXT(lp, lp_entries);
1863 
1864 	/*
1865 	 * Check the port's link state. This will return the next active
1866 	 * port if the link is down or the port is NULL.
1867 	 */
1868 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1869 		m_freem(m);
1870 		return (ENETDOWN);
1871 	}
1872 
1873 	/* Send mbuf */
1874 	return (lagg_enqueue(lp->lp_ifp, m));
1875 }
1876 
1877 static struct mbuf *
1878 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1879 {
1880 	struct ifnet *ifp = sc->sc_ifp;
1881 
1882 	/* Just pass in the packet to our lagg device */
1883 	m->m_pkthdr.rcvif = ifp;
1884 
1885 	return (m);
1886 }
1887 
1888 /*
1889  * Broadcast mode
1890  */
1891 static int
1892 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1893 {
1894 	int active_ports = 0;
1895 	int errors = 0;
1896 	int ret;
1897 	struct lagg_port *lp, *last = NULL;
1898 	struct mbuf *m0;
1899 
1900 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1901 		if (!LAGG_PORTACTIVE(lp))
1902 			continue;
1903 
1904 		active_ports++;
1905 
1906 		if (last != NULL) {
1907 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
1908 			if (m0 == NULL) {
1909 				ret = ENOBUFS;
1910 				errors++;
1911 				break;
1912 			}
1913 
1914 			ret = lagg_enqueue(last->lp_ifp, m0);
1915 			if (ret != 0)
1916 				errors++;
1917 		}
1918 		last = lp;
1919 	}
1920 	if (last == NULL) {
1921 		m_freem(m);
1922 		return (ENOENT);
1923 	}
1924 	if ((last = lagg_link_active(sc, last)) == NULL) {
1925 		m_freem(m);
1926 		return (ENETDOWN);
1927 	}
1928 
1929 	ret = lagg_enqueue(last->lp_ifp, m);
1930 	if (ret != 0)
1931 		errors++;
1932 
1933 	if (errors == 0)
1934 		return (ret);
1935 
1936 	return (0);
1937 }
1938 
1939 static struct mbuf*
1940 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1941 {
1942 	struct ifnet *ifp = sc->sc_ifp;
1943 
1944 	/* Just pass in the packet to our lagg device */
1945 	m->m_pkthdr.rcvif = ifp;
1946 	return (m);
1947 }
1948 
1949 /*
1950  * Active failover
1951  */
1952 static int
1953 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1954 {
1955 	struct lagg_port *lp;
1956 
1957 	/* Use the master port if active or the next available port */
1958 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1959 		m_freem(m);
1960 		return (ENETDOWN);
1961 	}
1962 
1963 	/* Send mbuf */
1964 	return (lagg_enqueue(lp->lp_ifp, m));
1965 }
1966 
1967 static struct mbuf *
1968 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1969 {
1970 	struct ifnet *ifp = sc->sc_ifp;
1971 	struct lagg_port *tmp_tp;
1972 
1973 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
1974 		m->m_pkthdr.rcvif = ifp;
1975 		return (m);
1976 	}
1977 
1978 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1979 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1980 		/*
1981 		 * If tmp_tp is null, we've received a packet when all
1982 		 * our links are down. Weird, but process it anyways.
1983 		 */
1984 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1985 			m->m_pkthdr.rcvif = ifp;
1986 			return (m);
1987 		}
1988 	}
1989 
1990 	m_freem(m);
1991 	return (NULL);
1992 }
1993 
1994 /*
1995  * Loadbalancing
1996  */
1997 static void
1998 lagg_lb_attach(struct lagg_softc *sc)
1999 {
2000 	struct lagg_port *lp;
2001 	struct lagg_lb *lb;
2002 
2003 	lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
2004 	lb->lb_key = m_ether_tcpip_hash_init();
2005 	sc->sc_psc = lb;
2006 
2007 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2008 		lagg_lb_port_create(lp);
2009 }
2010 
2011 static void
2012 lagg_lb_detach(struct lagg_softc *sc)
2013 {
2014 	struct lagg_lb *lb;
2015 
2016 	lb = (struct lagg_lb *)sc->sc_psc;
2017 	LAGG_WUNLOCK(sc);
2018 	if (lb != NULL)
2019 		free(lb, M_DEVBUF);
2020 }
2021 
2022 static int
2023 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2024 {
2025 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2026 	struct lagg_port *lp_next;
2027 	int i = 0;
2028 
2029 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2030 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2031 		if (lp_next == lp)
2032 			continue;
2033 		if (i >= LAGG_MAX_PORTS)
2034 			return (EINVAL);
2035 		if (sc->sc_ifflags & IFF_DEBUG)
2036 			printf("%s: port %s at index %d\n",
2037 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2038 		lb->lb_ports[i++] = lp_next;
2039 	}
2040 
2041 	return (0);
2042 }
2043 
2044 static int
2045 lagg_lb_port_create(struct lagg_port *lp)
2046 {
2047 	struct lagg_softc *sc = lp->lp_softc;
2048 	return (lagg_lb_porttable(sc, NULL));
2049 }
2050 
2051 static void
2052 lagg_lb_port_destroy(struct lagg_port *lp)
2053 {
2054 	struct lagg_softc *sc = lp->lp_softc;
2055 	lagg_lb_porttable(sc, lp);
2056 }
2057 
2058 static int
2059 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2060 {
2061 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2062 	struct lagg_port *lp = NULL;
2063 	uint32_t p = 0;
2064 
2065 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2066 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2067 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2068 	else
2069 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2070 	p %= sc->sc_count;
2071 	lp = lb->lb_ports[p];
2072 
2073 	/*
2074 	 * Check the port's link state. This will return the next active
2075 	 * port if the link is down or the port is NULL.
2076 	 */
2077 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2078 		m_freem(m);
2079 		return (ENETDOWN);
2080 	}
2081 
2082 	/* Send mbuf */
2083 	return (lagg_enqueue(lp->lp_ifp, m));
2084 }
2085 
2086 static struct mbuf *
2087 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2088 {
2089 	struct ifnet *ifp = sc->sc_ifp;
2090 
2091 	/* Just pass in the packet to our lagg device */
2092 	m->m_pkthdr.rcvif = ifp;
2093 
2094 	return (m);
2095 }
2096 
2097 /*
2098  * 802.3ad LACP
2099  */
2100 static void
2101 lagg_lacp_attach(struct lagg_softc *sc)
2102 {
2103 	struct lagg_port *lp;
2104 
2105 	lacp_attach(sc);
2106 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2107 		lacp_port_create(lp);
2108 }
2109 
2110 static void
2111 lagg_lacp_detach(struct lagg_softc *sc)
2112 {
2113 	struct lagg_port *lp;
2114 	void *psc;
2115 
2116 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2117 		lacp_port_destroy(lp);
2118 
2119 	psc = sc->sc_psc;
2120 	sc->sc_psc = NULL;
2121 	LAGG_WUNLOCK(sc);
2122 
2123 	lacp_detach(psc);
2124 }
2125 
2126 static void
2127 lagg_lacp_lladdr(struct lagg_softc *sc)
2128 {
2129 	struct lagg_port *lp;
2130 
2131 	LAGG_SXLOCK_ASSERT(sc);
2132 
2133 	/* purge all the lacp ports */
2134 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2135 		lacp_port_destroy(lp);
2136 
2137 	/* add them back in */
2138 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2139 		lacp_port_create(lp);
2140 }
2141 
2142 static int
2143 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2144 {
2145 	struct lagg_port *lp;
2146 
2147 	lp = lacp_select_tx_port(sc, m);
2148 	if (lp == NULL) {
2149 		m_freem(m);
2150 		return (ENETDOWN);
2151 	}
2152 
2153 	/* Send mbuf */
2154 	return (lagg_enqueue(lp->lp_ifp, m));
2155 }
2156 
2157 static struct mbuf *
2158 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2159 {
2160 	struct ifnet *ifp = sc->sc_ifp;
2161 	struct ether_header *eh;
2162 	u_short etype;
2163 
2164 	eh = mtod(m, struct ether_header *);
2165 	etype = ntohs(eh->ether_type);
2166 
2167 	/* Tap off LACP control messages */
2168 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2169 		m = lacp_input(lp, m);
2170 		if (m == NULL)
2171 			return (NULL);
2172 	}
2173 
2174 	/*
2175 	 * If the port is not collecting or not in the active aggregator then
2176 	 * free and return.
2177 	 */
2178 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2179 		m_freem(m);
2180 		return (NULL);
2181 	}
2182 
2183 	m->m_pkthdr.rcvif = ifp;
2184 	return (m);
2185 }
2186 
2187