xref: /freebsd/sys/net/if_lagg.c (revision 41059135ce931c0f1014a999ffabc6bc470ce856)
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 = ~0, ena = ~0;
576 	u_long hwa = ~0UL;
577 	struct ifnet_hw_tsomax hw_tsomax;
578 
579 	LAGG_XLOCK_ASSERT(sc);
580 
581 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
582 
583 	/* Get capabilities from the lagg ports */
584 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
585 		cap &= lp->lp_ifp->if_capabilities;
586 		ena &= lp->lp_ifp->if_capenable;
587 		hwa &= lp->lp_ifp->if_hwassist;
588 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
589 	}
590 	cap = (cap == ~0 ? 0 : cap);
591 	ena = (ena == ~0 ? 0 : ena);
592 	hwa = (hwa == ~0 ? 0 : hwa);
593 
594 	if (sc->sc_ifp->if_capabilities != cap ||
595 	    sc->sc_ifp->if_capenable != ena ||
596 	    sc->sc_ifp->if_hwassist != hwa ||
597 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
598 		sc->sc_ifp->if_capabilities = cap;
599 		sc->sc_ifp->if_capenable = ena;
600 		sc->sc_ifp->if_hwassist = hwa;
601 		getmicrotime(&sc->sc_ifp->if_lastchange);
602 
603 		if (sc->sc_ifflags & IFF_DEBUG)
604 			if_printf(sc->sc_ifp,
605 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
606 	}
607 
608 	/* Apply unified capabilities back to the lagg ports. */
609 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
610 		lagg_setcaps(lp, ena);
611 }
612 
613 static int
614 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
615 {
616 	struct lagg_softc *sc_ptr;
617 	struct lagg_port *lp, *tlp;
618 	int error, i;
619 	uint64_t *pval;
620 
621 	LAGG_XLOCK_ASSERT(sc);
622 
623 	/* Limit the maximal number of lagg ports */
624 	if (sc->sc_count >= LAGG_MAX_PORTS)
625 		return (ENOSPC);
626 
627 	/* Check if port has already been associated to a lagg */
628 	if (ifp->if_lagg != NULL) {
629 		/* Port is already in the current lagg? */
630 		lp = (struct lagg_port *)ifp->if_lagg;
631 		if (lp->lp_softc == sc)
632 			return (EEXIST);
633 		return (EBUSY);
634 	}
635 
636 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
637 	if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
638 		return (EPROTONOSUPPORT);
639 
640 	/* Allow the first Ethernet member to define the MTU */
641 	if (SLIST_EMPTY(&sc->sc_ports))
642 		sc->sc_ifp->if_mtu = ifp->if_mtu;
643 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
644 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
645 		    ifp->if_xname);
646 		return (EINVAL);
647 	}
648 
649 	lp = malloc(sizeof(struct lagg_port), M_DEVBUF, M_WAITOK|M_ZERO);
650 	lp->lp_softc = sc;
651 
652 	/* Check if port is a stacked lagg */
653 	LAGG_LIST_LOCK();
654 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
655 		if (ifp == sc_ptr->sc_ifp) {
656 			LAGG_LIST_UNLOCK();
657 			free(lp, M_DEVBUF);
658 			return (EINVAL);
659 			/* XXX disable stacking for the moment, its untested */
660 #ifdef LAGG_PORT_STACKING
661 			lp->lp_flags |= LAGG_PORT_STACK;
662 			if (lagg_port_checkstacking(sc_ptr) >=
663 			    LAGG_MAX_STACKING) {
664 				LAGG_LIST_UNLOCK();
665 				free(lp, M_DEVBUF);
666 				return (E2BIG);
667 			}
668 #endif
669 		}
670 	}
671 	LAGG_LIST_UNLOCK();
672 
673 	if_ref(ifp);
674 	lp->lp_ifp = ifp;
675 
676 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
677 	lp->lp_ifcapenable = ifp->if_capenable;
678 	if (SLIST_EMPTY(&sc->sc_ports)) {
679 		LAGG_WLOCK(sc);
680 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
681 		lagg_proto_lladdr(sc);
682 		LAGG_WUNLOCK(sc);
683 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
684 	} else {
685 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
686 	}
687 	lagg_setflags(lp, 1);
688 
689 	LAGG_WLOCK(sc);
690 	if (SLIST_EMPTY(&sc->sc_ports))
691 		sc->sc_primary = lp;
692 
693 	/* Change the interface type */
694 	lp->lp_iftype = ifp->if_type;
695 	ifp->if_type = IFT_IEEE8023ADLAG;
696 	ifp->if_lagg = lp;
697 	lp->lp_ioctl = ifp->if_ioctl;
698 	ifp->if_ioctl = lagg_port_ioctl;
699 	lp->lp_output = ifp->if_output;
700 	ifp->if_output = lagg_port_output;
701 
702 	/* Read port counters */
703 	pval = lp->port_counters.val;
704 	for (i = 0; i < IFCOUNTERS; i++, pval++)
705 		*pval = ifp->if_get_counter(ifp, i);
706 
707 	/*
708 	 * Insert into the list of ports.
709 	 * Keep ports sorted by if_index. It is handy, when configuration
710 	 * is predictable and `ifconfig laggN create ...` command
711 	 * will lead to the same result each time.
712 	 */
713 	SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
714 		if (tlp->lp_ifp->if_index < ifp->if_index && (
715 		    SLIST_NEXT(tlp, lp_entries) == NULL ||
716 		    SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index >
717 		    ifp->if_index))
718 			break;
719 	}
720 	if (tlp != NULL)
721 		SLIST_INSERT_AFTER(tlp, lp, lp_entries);
722 	else
723 		SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
724 	sc->sc_count++;
725 
726 	lagg_setmulti(lp);
727 
728 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
729 		/* Remove the port, without calling pr_delport. */
730 		lagg_port_destroy(lp, 0);
731 		LAGG_UNLOCK_ASSERT(sc);
732 		return (error);
733 	}
734 
735 	LAGG_WUNLOCK(sc);
736 
737 	/* Update lagg capabilities */
738 	lagg_capabilities(sc);
739 	lagg_linkstate(sc);
740 
741 	return (0);
742 }
743 
744 #ifdef LAGG_PORT_STACKING
745 static int
746 lagg_port_checkstacking(struct lagg_softc *sc)
747 {
748 	struct lagg_softc *sc_ptr;
749 	struct lagg_port *lp;
750 	int m = 0;
751 
752 	LAGG_SXLOCK_ASSERT(sc);
753 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
754 		if (lp->lp_flags & LAGG_PORT_STACK) {
755 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
756 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
757 		}
758 	}
759 
760 	return (m + 1);
761 }
762 #endif
763 
764 static int
765 lagg_port_destroy(struct lagg_port *lp, int rundelport)
766 {
767 	struct lagg_softc *sc = lp->lp_softc;
768 	struct lagg_port *lp_ptr, *lp0;
769 	struct ifnet *ifp = lp->lp_ifp;
770 	uint64_t *pval, vdiff;
771 	int i;
772 
773 	LAGG_XLOCK_ASSERT(sc);
774 
775 	if (rundelport) {
776 		LAGG_WLOCK(sc);
777 		lagg_proto_delport(sc, lp);
778 	} else
779 		LAGG_WLOCK_ASSERT(sc);
780 
781 	if (lp->lp_detaching == 0)
782 		lagg_clrmulti(lp);
783 
784 	/* Restore interface */
785 	ifp->if_type = lp->lp_iftype;
786 	ifp->if_ioctl = lp->lp_ioctl;
787 	ifp->if_output = lp->lp_output;
788 	ifp->if_lagg = NULL;
789 
790 	/* Update detached port counters */
791 	pval = lp->port_counters.val;
792 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
793 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
794 		sc->detached_counters.val[i] += vdiff;
795 	}
796 
797 	/* Finally, remove the port from the lagg */
798 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
799 	sc->sc_count--;
800 
801 	/* Update the primary interface */
802 	if (lp == sc->sc_primary) {
803 		uint8_t lladdr[ETHER_ADDR_LEN];
804 
805 		if ((lp0 = SLIST_FIRST(&sc->sc_ports)) == NULL)
806 			bzero(&lladdr, ETHER_ADDR_LEN);
807 		else
808 			bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN);
809 		sc->sc_primary = lp0;
810 		if (sc->sc_destroying == 0) {
811 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
812 			lagg_proto_lladdr(sc);
813 			LAGG_WUNLOCK(sc);
814 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
815 		} else
816 			LAGG_WUNLOCK(sc);
817 
818 		/*
819 		 * Update lladdr for each port (new primary needs update
820 		 * as well, to switch from old lladdr to its 'real' one)
821 		 */
822 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
823 			if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN);
824 	} else
825 		LAGG_WUNLOCK(sc);
826 
827 	if (lp->lp_ifflags)
828 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
829 
830 	if (lp->lp_detaching == 0) {
831 		lagg_setflags(lp, 0);
832 		lagg_setcaps(lp, lp->lp_ifcapenable);
833 		if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN);
834 	}
835 
836 	if_rele(ifp);
837 	free(lp, M_DEVBUF);
838 
839 	/* Update lagg capabilities */
840 	lagg_capabilities(sc);
841 	lagg_linkstate(sc);
842 
843 	return (0);
844 }
845 
846 static int
847 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
848 {
849 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
850 	struct lagg_softc *sc;
851 	struct lagg_port *lp = NULL;
852 	int error = 0;
853 
854 	/* Should be checked by the caller */
855 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
856 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
857 		goto fallback;
858 
859 	switch (cmd) {
860 	case SIOCGLAGGPORT:
861 		if (rp->rp_portname[0] == '\0' ||
862 		    ifunit(rp->rp_portname) != ifp) {
863 			error = EINVAL;
864 			break;
865 		}
866 
867 		LAGG_SLOCK(sc);
868 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
869 			error = ENOENT;
870 			LAGG_SUNLOCK(sc);
871 			break;
872 		}
873 
874 		lagg_port2req(lp, rp);
875 		LAGG_SUNLOCK(sc);
876 		break;
877 
878 	case SIOCSIFCAP:
879 		if (lp->lp_ioctl == NULL) {
880 			error = EINVAL;
881 			break;
882 		}
883 		error = (*lp->lp_ioctl)(ifp, cmd, data);
884 		if (error)
885 			break;
886 
887 		/* Update lagg interface capabilities */
888 		LAGG_XLOCK(sc);
889 		lagg_capabilities(sc);
890 		LAGG_XUNLOCK(sc);
891 		break;
892 
893 	case SIOCSIFMTU:
894 		/* Do not allow the MTU to be changed once joined */
895 		error = EINVAL;
896 		break;
897 
898 	default:
899 		goto fallback;
900 	}
901 
902 	return (error);
903 
904 fallback:
905 	if (lp != NULL && lp->lp_ioctl != NULL)
906 		return ((*lp->lp_ioctl)(ifp, cmd, data));
907 
908 	return (EINVAL);
909 }
910 
911 /*
912  * Requests counter @cnt data.
913  *
914  * Counter value is calculated the following way:
915  * 1) for each port, sum  difference between current and "initial" measurements.
916  * 2) add lagg logical interface counters.
917  * 3) add data from detached_counters array.
918  *
919  * We also do the following things on ports attach/detach:
920  * 1) On port attach we store all counters it has into port_counter array.
921  * 2) On port detach we add the different between "initial" and
922  *   current counters data to detached_counters array.
923  */
924 static uint64_t
925 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
926 {
927 	struct lagg_softc *sc;
928 	struct lagg_port *lp;
929 	struct ifnet *lpifp;
930 	struct rm_priotracker tracker;
931 	uint64_t newval, oldval, vsum;
932 
933 	/* Revise this when we've got non-generic counters. */
934 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
935 
936 	sc = (struct lagg_softc *)ifp->if_softc;
937 	LAGG_RLOCK(sc, &tracker);
938 
939 	vsum = 0;
940 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
941 		/* Saved attached value */
942 		oldval = lp->port_counters.val[cnt];
943 		/* current value */
944 		lpifp = lp->lp_ifp;
945 		newval = lpifp->if_get_counter(lpifp, cnt);
946 		/* Calculate diff and save new */
947 		vsum += newval - oldval;
948 	}
949 
950 	/*
951 	 * Add counter data which might be added by upper
952 	 * layer protocols operating on logical interface.
953 	 */
954 	vsum += if_get_counter_default(ifp, cnt);
955 
956 	/*
957 	 * Add counter data from detached ports counters
958 	 */
959 	vsum += sc->detached_counters.val[cnt];
960 
961 	LAGG_RUNLOCK(sc, &tracker);
962 
963 	return (vsum);
964 }
965 
966 /*
967  * For direct output to child ports.
968  */
969 static int
970 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
971 	const struct sockaddr *dst, struct route *ro)
972 {
973 	struct lagg_port *lp = ifp->if_lagg;
974 
975 	switch (dst->sa_family) {
976 		case pseudo_AF_HDRCMPLT:
977 		case AF_UNSPEC:
978 			return ((*lp->lp_output)(ifp, m, dst, ro));
979 	}
980 
981 	/* drop any other frames */
982 	m_freem(m);
983 	return (ENETDOWN);
984 }
985 
986 static void
987 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
988 {
989 	struct lagg_port *lp;
990 	struct lagg_softc *sc;
991 
992 	if ((lp = ifp->if_lagg) == NULL)
993 		return;
994 	/* If the ifnet is just being renamed, don't do anything. */
995 	if (ifp->if_flags & IFF_RENAMING)
996 		return;
997 
998 	sc = lp->lp_softc;
999 
1000 	LAGG_XLOCK(sc);
1001 	lp->lp_detaching = 1;
1002 	lagg_port_destroy(lp, 1);
1003 	LAGG_XUNLOCK(sc);
1004 }
1005 
1006 static void
1007 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1008 {
1009 	struct lagg_softc *sc = lp->lp_softc;
1010 
1011 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1012 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1013 	rp->rp_prio = lp->lp_prio;
1014 	rp->rp_flags = lp->lp_flags;
1015 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1016 
1017 	/* Add protocol specific flags */
1018 	switch (sc->sc_proto) {
1019 		case LAGG_PROTO_FAILOVER:
1020 			if (lp == sc->sc_primary)
1021 				rp->rp_flags |= LAGG_PORT_MASTER;
1022 			if (lp == lagg_link_active(sc, sc->sc_primary))
1023 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1024 			break;
1025 
1026 		case LAGG_PROTO_ROUNDROBIN:
1027 		case LAGG_PROTO_LOADBALANCE:
1028 		case LAGG_PROTO_BROADCAST:
1029 			if (LAGG_PORTACTIVE(lp))
1030 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1031 			break;
1032 
1033 		case LAGG_PROTO_LACP:
1034 			/* LACP has a different definition of active */
1035 			if (lacp_isactive(lp))
1036 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1037 			if (lacp_iscollecting(lp))
1038 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1039 			if (lacp_isdistributing(lp))
1040 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1041 			break;
1042 	}
1043 
1044 }
1045 
1046 static void
1047 lagg_init(void *xsc)
1048 {
1049 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1050 	struct ifnet *ifp = sc->sc_ifp;
1051 	struct lagg_port *lp;
1052 
1053 	LAGG_XLOCK(sc);
1054 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1055 		LAGG_XUNLOCK(sc);
1056 		return;
1057 	}
1058 
1059 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1060 
1061 	/*
1062 	 * Update the port lladdrs if needed.
1063 	 * This might be if_setlladdr() notification
1064 	 * that lladdr has been changed.
1065 	 */
1066 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1067 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1068 		    ETHER_ADDR_LEN) != 0)
1069 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN);
1070 	}
1071 
1072 	lagg_proto_init(sc);
1073 
1074 	LAGG_XUNLOCK(sc);
1075 }
1076 
1077 static void
1078 lagg_stop(struct lagg_softc *sc)
1079 {
1080 	struct ifnet *ifp = sc->sc_ifp;
1081 
1082 	LAGG_XLOCK_ASSERT(sc);
1083 
1084 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1085 		return;
1086 
1087 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1088 
1089 	lagg_proto_stop(sc);
1090 }
1091 
1092 static int
1093 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1094 {
1095 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1096 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1097 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1098 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1099 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1100 	struct ifreq *ifr = (struct ifreq *)data;
1101 	struct lagg_port *lp;
1102 	struct ifnet *tpif;
1103 	struct thread *td = curthread;
1104 	char *buf, *outbuf;
1105 	int count, buflen, len, error = 0;
1106 
1107 	bzero(&rpbuf, sizeof(rpbuf));
1108 
1109 	switch (cmd) {
1110 	case SIOCGLAGG:
1111 		LAGG_SLOCK(sc);
1112 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1113 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1114 		ra->ra_proto = sc->sc_proto;
1115 		lagg_proto_request(sc, &ra->ra_psc);
1116 		count = 0;
1117 		buf = outbuf;
1118 		len = min(ra->ra_size, buflen);
1119 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1120 			if (len < sizeof(rpbuf))
1121 				break;
1122 
1123 			lagg_port2req(lp, &rpbuf);
1124 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1125 			count++;
1126 			buf += sizeof(rpbuf);
1127 			len -= sizeof(rpbuf);
1128 		}
1129 		LAGG_SUNLOCK(sc);
1130 		ra->ra_ports = count;
1131 		ra->ra_size = count * sizeof(rpbuf);
1132 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1133 		free(outbuf, M_TEMP);
1134 		break;
1135 	case SIOCSLAGG:
1136 		error = priv_check(td, PRIV_NET_LAGG);
1137 		if (error)
1138 			break;
1139 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1140 			error = EPROTONOSUPPORT;
1141 			break;
1142 		}
1143 
1144 		LAGG_XLOCK(sc);
1145 		LAGG_WLOCK(sc);
1146 		lagg_proto_detach(sc);
1147 		LAGG_UNLOCK_ASSERT(sc);
1148 		lagg_proto_attach(sc, ra->ra_proto);
1149 		LAGG_XUNLOCK(sc);
1150 		break;
1151 	case SIOCGLAGGOPTS:
1152 		LAGG_SLOCK(sc);
1153 		ro->ro_opts = sc->sc_opts;
1154 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1155 			struct lacp_softc *lsc;
1156 
1157 			lsc = (struct lacp_softc *)sc->sc_psc;
1158 			if (lsc->lsc_debug.lsc_tx_test != 0)
1159 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1160 			if (lsc->lsc_debug.lsc_rx_test != 0)
1161 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1162 			if (lsc->lsc_strict_mode != 0)
1163 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1164 			if (lsc->lsc_fast_timeout != 0)
1165 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1166 
1167 			ro->ro_active = sc->sc_active;
1168 		} else {
1169 			ro->ro_active = 0;
1170 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1171 				ro->ro_active += LAGG_PORTACTIVE(lp);
1172 		}
1173 		ro->ro_bkt = sc->sc_bkt;
1174 		ro->ro_flapping = sc->sc_flapping;
1175 		ro->ro_flowid_shift = sc->flowid_shift;
1176 		LAGG_SUNLOCK(sc);
1177 		break;
1178 	case SIOCSLAGGOPTS:
1179 		if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) {
1180 			if (ro->ro_bkt == 0)
1181 				sc->sc_bkt = 1; // Minimum 1 packet per iface.
1182 			else
1183 				sc->sc_bkt = ro->ro_bkt;
1184 		}
1185 		error = priv_check(td, PRIV_NET_LAGG);
1186 		if (error)
1187 			break;
1188 		if (ro->ro_opts == 0)
1189 			break;
1190 		/*
1191 		 * Set options.  LACP options are stored in sc->sc_psc,
1192 		 * not in sc_opts.
1193 		 */
1194 		int valid, lacp;
1195 
1196 		switch (ro->ro_opts) {
1197 		case LAGG_OPT_USE_FLOWID:
1198 		case -LAGG_OPT_USE_FLOWID:
1199 		case LAGG_OPT_FLOWIDSHIFT:
1200 			valid = 1;
1201 			lacp = 0;
1202 			break;
1203 		case LAGG_OPT_LACP_TXTEST:
1204 		case -LAGG_OPT_LACP_TXTEST:
1205 		case LAGG_OPT_LACP_RXTEST:
1206 		case -LAGG_OPT_LACP_RXTEST:
1207 		case LAGG_OPT_LACP_STRICT:
1208 		case -LAGG_OPT_LACP_STRICT:
1209 		case LAGG_OPT_LACP_TIMEOUT:
1210 		case -LAGG_OPT_LACP_TIMEOUT:
1211 			valid = lacp = 1;
1212 			break;
1213 		default:
1214 			valid = lacp = 0;
1215 			break;
1216 		}
1217 
1218 		LAGG_XLOCK(sc);
1219 
1220 		if (valid == 0 ||
1221 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1222 			/* Invalid combination of options specified. */
1223 			error = EINVAL;
1224 			LAGG_XUNLOCK(sc);
1225 			break;	/* Return from SIOCSLAGGOPTS. */
1226 		}
1227 		/*
1228 		 * Store new options into sc->sc_opts except for
1229 		 * FLOWIDSHIFT and LACP options.
1230 		 */
1231 		if (lacp == 0) {
1232 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1233 				sc->flowid_shift = ro->ro_flowid_shift;
1234 			else if (ro->ro_opts > 0)
1235 				sc->sc_opts |= ro->ro_opts;
1236 			else
1237 				sc->sc_opts &= ~ro->ro_opts;
1238 		} else {
1239 			struct lacp_softc *lsc;
1240 			struct lacp_port *lp;
1241 
1242 			lsc = (struct lacp_softc *)sc->sc_psc;
1243 
1244 			switch (ro->ro_opts) {
1245 			case LAGG_OPT_LACP_TXTEST:
1246 				lsc->lsc_debug.lsc_tx_test = 1;
1247 				break;
1248 			case -LAGG_OPT_LACP_TXTEST:
1249 				lsc->lsc_debug.lsc_tx_test = 0;
1250 				break;
1251 			case LAGG_OPT_LACP_RXTEST:
1252 				lsc->lsc_debug.lsc_rx_test = 1;
1253 				break;
1254 			case -LAGG_OPT_LACP_RXTEST:
1255 				lsc->lsc_debug.lsc_rx_test = 0;
1256 				break;
1257 			case LAGG_OPT_LACP_STRICT:
1258 				lsc->lsc_strict_mode = 1;
1259 				break;
1260 			case -LAGG_OPT_LACP_STRICT:
1261 				lsc->lsc_strict_mode = 0;
1262 				break;
1263 			case LAGG_OPT_LACP_TIMEOUT:
1264 				LACP_LOCK(lsc);
1265         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1266                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1267 				LACP_UNLOCK(lsc);
1268 				lsc->lsc_fast_timeout = 1;
1269 				break;
1270 			case -LAGG_OPT_LACP_TIMEOUT:
1271 				LACP_LOCK(lsc);
1272         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1273                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1274 				LACP_UNLOCK(lsc);
1275 				lsc->lsc_fast_timeout = 0;
1276 				break;
1277 			}
1278 		}
1279 		LAGG_XUNLOCK(sc);
1280 		break;
1281 	case SIOCGLAGGFLAGS:
1282 		rf->rf_flags = 0;
1283 		LAGG_SLOCK(sc);
1284 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1285 			rf->rf_flags |= LAGG_F_HASHL2;
1286 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1287 			rf->rf_flags |= LAGG_F_HASHL3;
1288 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1289 			rf->rf_flags |= LAGG_F_HASHL4;
1290 		LAGG_SUNLOCK(sc);
1291 		break;
1292 	case SIOCSLAGGHASH:
1293 		error = priv_check(td, PRIV_NET_LAGG);
1294 		if (error)
1295 			break;
1296 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1297 			error = EINVAL;
1298 			break;
1299 		}
1300 		LAGG_XLOCK(sc);
1301 		sc->sc_flags = 0;
1302 		if (rf->rf_flags & LAGG_F_HASHL2)
1303 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1304 		if (rf->rf_flags & LAGG_F_HASHL3)
1305 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1306 		if (rf->rf_flags & LAGG_F_HASHL4)
1307 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1308 		LAGG_XUNLOCK(sc);
1309 		break;
1310 	case SIOCGLAGGPORT:
1311 		if (rp->rp_portname[0] == '\0' ||
1312 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1313 			error = EINVAL;
1314 			break;
1315 		}
1316 
1317 		LAGG_SLOCK(sc);
1318 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1319 		    lp->lp_softc != sc) {
1320 			error = ENOENT;
1321 			LAGG_SUNLOCK(sc);
1322 			if_rele(tpif);
1323 			break;
1324 		}
1325 
1326 		lagg_port2req(lp, rp);
1327 		LAGG_SUNLOCK(sc);
1328 		if_rele(tpif);
1329 		break;
1330 	case SIOCSLAGGPORT:
1331 		error = priv_check(td, PRIV_NET_LAGG);
1332 		if (error)
1333 			break;
1334 		if (rp->rp_portname[0] == '\0' ||
1335 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1336 			error = EINVAL;
1337 			break;
1338 		}
1339 #ifdef INET6
1340 		/*
1341 		 * A laggport interface should not have inet6 address
1342 		 * because two interfaces with a valid link-local
1343 		 * scope zone must not be merged in any form.  This
1344 		 * restriction is needed to prevent violation of
1345 		 * link-local scope zone.  Attempts to add a laggport
1346 		 * interface which has inet6 addresses triggers
1347 		 * removal of all inet6 addresses on the member
1348 		 * interface.
1349 		 */
1350 		if (in6ifa_llaonifp(tpif)) {
1351 			in6_ifdetach(tpif);
1352 				if_printf(sc->sc_ifp,
1353 				    "IPv6 addresses on %s have been removed "
1354 				    "before adding it as a member to prevent "
1355 				    "IPv6 address scope violation.\n",
1356 				    tpif->if_xname);
1357 		}
1358 #endif
1359 		LAGG_XLOCK(sc);
1360 		error = lagg_port_create(sc, tpif);
1361 		LAGG_XUNLOCK(sc);
1362 		if_rele(tpif);
1363 		break;
1364 	case SIOCSLAGGDELPORT:
1365 		error = priv_check(td, PRIV_NET_LAGG);
1366 		if (error)
1367 			break;
1368 		if (rp->rp_portname[0] == '\0' ||
1369 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1370 			error = EINVAL;
1371 			break;
1372 		}
1373 
1374 		LAGG_XLOCK(sc);
1375 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1376 		    lp->lp_softc != sc) {
1377 			error = ENOENT;
1378 			LAGG_XUNLOCK(sc);
1379 			if_rele(tpif);
1380 			break;
1381 		}
1382 
1383 		error = lagg_port_destroy(lp, 1);
1384 		LAGG_XUNLOCK(sc);
1385 		if_rele(tpif);
1386 		break;
1387 	case SIOCSIFFLAGS:
1388 		/* Set flags on ports too */
1389 		LAGG_XLOCK(sc);
1390 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1391 			lagg_setflags(lp, 1);
1392 		}
1393 
1394 		if (!(ifp->if_flags & IFF_UP) &&
1395 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1396 			/*
1397 			 * If interface is marked down and it is running,
1398 			 * then stop and disable it.
1399 			 */
1400 			lagg_stop(sc);
1401 			LAGG_XUNLOCK(sc);
1402 		} else if ((ifp->if_flags & IFF_UP) &&
1403 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1404 			/*
1405 			 * If interface is marked up and it is stopped, then
1406 			 * start it.
1407 			 */
1408 			LAGG_XUNLOCK(sc);
1409 			(*ifp->if_init)(sc);
1410 		} else
1411 			LAGG_XUNLOCK(sc);
1412 		break;
1413 	case SIOCADDMULTI:
1414 	case SIOCDELMULTI:
1415 		LAGG_WLOCK(sc);
1416 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1417 			lagg_clrmulti(lp);
1418 			lagg_setmulti(lp);
1419 		}
1420 		LAGG_WUNLOCK(sc);
1421 		error = 0;
1422 		break;
1423 	case SIOCSIFMEDIA:
1424 	case SIOCGIFMEDIA:
1425 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1426 		break;
1427 
1428 	case SIOCSIFCAP:
1429 		LAGG_XLOCK(sc);
1430 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1431 			if (lp->lp_ioctl != NULL)
1432 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1433 		}
1434 		lagg_capabilities(sc);
1435 		LAGG_XUNLOCK(sc);
1436 		error = 0;
1437 		break;
1438 
1439 	case SIOCSIFMTU:
1440 		/* Do not allow the MTU to be directly changed */
1441 		error = EINVAL;
1442 		break;
1443 
1444 	default:
1445 		error = ether_ioctl(ifp, cmd, data);
1446 		break;
1447 	}
1448 	return (error);
1449 }
1450 
1451 #ifdef RATELIMIT
1452 static int
1453 lagg_snd_tag_alloc(struct ifnet *ifp,
1454     union if_snd_tag_alloc_params *params,
1455     struct m_snd_tag **ppmt)
1456 {
1457 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1458 	struct lagg_port *lp;
1459 	struct lagg_lb *lb;
1460 	uint32_t p;
1461 
1462 	switch (sc->sc_proto) {
1463 	case LAGG_PROTO_FAILOVER:
1464 		lp = lagg_link_active(sc, sc->sc_primary);
1465 		break;
1466 	case LAGG_PROTO_LOADBALANCE:
1467 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1468 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1469 			return (EOPNOTSUPP);
1470 		p = params->hdr.flowid >> sc->flowid_shift;
1471 		p %= sc->sc_count;
1472 		lb = (struct lagg_lb *)sc->sc_psc;
1473 		lp = lb->lb_ports[p];
1474 		lp = lagg_link_active(sc, lp);
1475 		break;
1476 	case LAGG_PROTO_LACP:
1477 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1478 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1479 			return (EOPNOTSUPP);
1480 		lp = lacp_select_tx_port_by_hash(sc, params->hdr.flowid);
1481 		break;
1482 	default:
1483 		return (EOPNOTSUPP);
1484 	}
1485 	if (lp == NULL)
1486 		return (EOPNOTSUPP);
1487 	ifp = lp->lp_ifp;
1488 	if (ifp == NULL || ifp->if_snd_tag_alloc == NULL ||
1489 	    (ifp->if_capenable & IFCAP_TXRTLMT) == 0)
1490 		return (EOPNOTSUPP);
1491 
1492 	/* forward allocation request */
1493 	return (ifp->if_snd_tag_alloc(ifp, params, ppmt));
1494 }
1495 #endif
1496 
1497 static int
1498 lagg_setmulti(struct lagg_port *lp)
1499 {
1500 	struct lagg_softc *sc = lp->lp_softc;
1501 	struct ifnet *ifp = lp->lp_ifp;
1502 	struct ifnet *scifp = sc->sc_ifp;
1503 	struct lagg_mc *mc;
1504 	struct ifmultiaddr *ifma;
1505 	int error;
1506 
1507 	LAGG_WLOCK_ASSERT(sc);
1508 	IF_ADDR_WLOCK(scifp);
1509 	TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1510 		if (ifma->ifma_addr->sa_family != AF_LINK)
1511 			continue;
1512 		mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1513 		if (mc == NULL) {
1514 			IF_ADDR_WUNLOCK(scifp);
1515 			return (ENOMEM);
1516 		}
1517 		bcopy(ifma->ifma_addr, &mc->mc_addr,
1518 		    ifma->ifma_addr->sa_len);
1519 		mc->mc_addr.sdl_index = ifp->if_index;
1520 		mc->mc_ifma = NULL;
1521 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1522 	}
1523 	IF_ADDR_WUNLOCK(scifp);
1524 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1525 		error = if_addmulti(ifp,
1526 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1527 		if (error)
1528 			return (error);
1529 	}
1530 	return (0);
1531 }
1532 
1533 static int
1534 lagg_clrmulti(struct lagg_port *lp)
1535 {
1536 	struct lagg_mc *mc;
1537 
1538 	LAGG_WLOCK_ASSERT(lp->lp_softc);
1539 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1540 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1541 		if (mc->mc_ifma && lp->lp_detaching == 0)
1542 			if_delmulti_ifma(mc->mc_ifma);
1543 		free(mc, M_DEVBUF);
1544 	}
1545 	return (0);
1546 }
1547 
1548 static int
1549 lagg_setcaps(struct lagg_port *lp, int cap)
1550 {
1551 	struct ifreq ifr;
1552 
1553 	if (lp->lp_ifp->if_capenable == cap)
1554 		return (0);
1555 	if (lp->lp_ioctl == NULL)
1556 		return (ENXIO);
1557 	ifr.ifr_reqcap = cap;
1558 	return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1559 }
1560 
1561 /* Handle a ref counted flag that should be set on the lagg port as well */
1562 static int
1563 lagg_setflag(struct lagg_port *lp, int flag, int status,
1564     int (*func)(struct ifnet *, int))
1565 {
1566 	struct lagg_softc *sc = lp->lp_softc;
1567 	struct ifnet *scifp = sc->sc_ifp;
1568 	struct ifnet *ifp = lp->lp_ifp;
1569 	int error;
1570 
1571 	LAGG_XLOCK_ASSERT(sc);
1572 
1573 	status = status ? (scifp->if_flags & flag) : 0;
1574 	/* Now "status" contains the flag value or 0 */
1575 
1576 	/*
1577 	 * See if recorded ports status is different from what
1578 	 * we want it to be.  If it is, flip it.  We record ports
1579 	 * status in lp_ifflags so that we won't clear ports flag
1580 	 * we haven't set.  In fact, we don't clear or set ports
1581 	 * flags directly, but get or release references to them.
1582 	 * That's why we can be sure that recorded flags still are
1583 	 * in accord with actual ports flags.
1584 	 */
1585 	if (status != (lp->lp_ifflags & flag)) {
1586 		error = (*func)(ifp, status);
1587 		if (error)
1588 			return (error);
1589 		lp->lp_ifflags &= ~flag;
1590 		lp->lp_ifflags |= status;
1591 	}
1592 	return (0);
1593 }
1594 
1595 /*
1596  * Handle IFF_* flags that require certain changes on the lagg port
1597  * if "status" is true, update ports flags respective to the lagg
1598  * if "status" is false, forcedly clear the flags set on port.
1599  */
1600 static int
1601 lagg_setflags(struct lagg_port *lp, int status)
1602 {
1603 	int error, i;
1604 
1605 	for (i = 0; lagg_pflags[i].flag; i++) {
1606 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1607 		    status, lagg_pflags[i].func);
1608 		if (error)
1609 			return (error);
1610 	}
1611 	return (0);
1612 }
1613 
1614 static int
1615 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1616 {
1617 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1618 	int error, len, mcast;
1619 	struct rm_priotracker tracker;
1620 
1621 	len = m->m_pkthdr.len;
1622 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1623 
1624 	LAGG_RLOCK(sc, &tracker);
1625 	/* We need a Tx algorithm and at least one port */
1626 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1627 		LAGG_RUNLOCK(sc, &tracker);
1628 		m_freem(m);
1629 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1630 		return (ENXIO);
1631 	}
1632 
1633 	ETHER_BPF_MTAP(ifp, m);
1634 
1635 	error = lagg_proto_start(sc, m);
1636 	LAGG_RUNLOCK(sc, &tracker);
1637 
1638 	if (error != 0)
1639 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1640 
1641 	return (error);
1642 }
1643 
1644 /*
1645  * The ifp->if_qflush entry point for lagg(4) is no-op.
1646  */
1647 static void
1648 lagg_qflush(struct ifnet *ifp __unused)
1649 {
1650 }
1651 
1652 static struct mbuf *
1653 lagg_input(struct ifnet *ifp, struct mbuf *m)
1654 {
1655 	struct lagg_port *lp = ifp->if_lagg;
1656 	struct lagg_softc *sc = lp->lp_softc;
1657 	struct ifnet *scifp = sc->sc_ifp;
1658 	struct rm_priotracker tracker;
1659 
1660 	LAGG_RLOCK(sc, &tracker);
1661 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1662 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1663 	    sc->sc_proto == LAGG_PROTO_NONE) {
1664 		LAGG_RUNLOCK(sc, &tracker);
1665 		m_freem(m);
1666 		return (NULL);
1667 	}
1668 
1669 	ETHER_BPF_MTAP(scifp, m);
1670 
1671 	if (lp->lp_detaching != 0) {
1672 		m_freem(m);
1673 		m = NULL;
1674 	} else
1675 		m = lagg_proto_input(sc, lp, m);
1676 
1677 	if (m != NULL) {
1678 		if (scifp->if_flags & IFF_MONITOR) {
1679 			m_freem(m);
1680 			m = NULL;
1681 		}
1682 	}
1683 
1684 	LAGG_RUNLOCK(sc, &tracker);
1685 	return (m);
1686 }
1687 
1688 static int
1689 lagg_media_change(struct ifnet *ifp)
1690 {
1691 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1692 
1693 	if (sc->sc_ifflags & IFF_DEBUG)
1694 		printf("%s\n", __func__);
1695 
1696 	/* Ignore */
1697 	return (0);
1698 }
1699 
1700 static void
1701 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1702 {
1703 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1704 	struct lagg_port *lp;
1705 
1706 	imr->ifm_status = IFM_AVALID;
1707 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1708 
1709 	LAGG_SLOCK(sc);
1710 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1711 		if (LAGG_PORTACTIVE(lp))
1712 			imr->ifm_status |= IFM_ACTIVE;
1713 	}
1714 	LAGG_SUNLOCK(sc);
1715 }
1716 
1717 static void
1718 lagg_linkstate(struct lagg_softc *sc)
1719 {
1720 	struct lagg_port *lp;
1721 	int new_link = LINK_STATE_DOWN;
1722 	uint64_t speed;
1723 
1724 	LAGG_XLOCK_ASSERT(sc);
1725 
1726 	/* Our link is considered up if at least one of our ports is active */
1727 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1728 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1729 			new_link = LINK_STATE_UP;
1730 			break;
1731 		}
1732 	}
1733 	if_link_state_change(sc->sc_ifp, new_link);
1734 
1735 	/* Update if_baudrate to reflect the max possible speed */
1736 	switch (sc->sc_proto) {
1737 		case LAGG_PROTO_FAILOVER:
1738 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1739 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1740 			break;
1741 		case LAGG_PROTO_ROUNDROBIN:
1742 		case LAGG_PROTO_LOADBALANCE:
1743 		case LAGG_PROTO_BROADCAST:
1744 			speed = 0;
1745 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1746 				speed += lp->lp_ifp->if_baudrate;
1747 			sc->sc_ifp->if_baudrate = speed;
1748 			break;
1749 		case LAGG_PROTO_LACP:
1750 			/* LACP updates if_baudrate itself */
1751 			break;
1752 	}
1753 }
1754 
1755 static void
1756 lagg_port_state(struct ifnet *ifp, int state)
1757 {
1758 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1759 	struct lagg_softc *sc = NULL;
1760 
1761 	if (lp != NULL)
1762 		sc = lp->lp_softc;
1763 	if (sc == NULL)
1764 		return;
1765 
1766 	LAGG_XLOCK(sc);
1767 	lagg_linkstate(sc);
1768 	lagg_proto_linkstate(sc, lp);
1769 	LAGG_XUNLOCK(sc);
1770 }
1771 
1772 struct lagg_port *
1773 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1774 {
1775 	struct lagg_port *lp_next, *rval = NULL;
1776 	// int new_link = LINK_STATE_DOWN;
1777 
1778 	/*
1779 	 * Search a port which reports an active link state.
1780 	 */
1781 
1782 	if (lp == NULL)
1783 		goto search;
1784 	if (LAGG_PORTACTIVE(lp)) {
1785 		rval = lp;
1786 		goto found;
1787 	}
1788 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1789 	    LAGG_PORTACTIVE(lp_next)) {
1790 		rval = lp_next;
1791 		goto found;
1792 	}
1793 
1794 search:
1795 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1796 		if (LAGG_PORTACTIVE(lp_next)) {
1797 			rval = lp_next;
1798 			goto found;
1799 		}
1800 	}
1801 
1802 found:
1803 	if (rval != NULL) {
1804 		/*
1805 		 * The IEEE 802.1D standard assumes that a lagg with
1806 		 * multiple ports is always full duplex. This is valid
1807 		 * for load sharing laggs and if at least two links
1808 		 * are active. Unfortunately, checking the latter would
1809 		 * be too expensive at this point.
1810 		 XXX
1811 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1812 		    (sc->sc_count > 1))
1813 			new_link = LINK_STATE_FULL_DUPLEX;
1814 		else
1815 			new_link = rval->lp_link_state;
1816 		 */
1817 	}
1818 
1819 	return (rval);
1820 }
1821 
1822 int
1823 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1824 {
1825 
1826 	return (ifp->if_transmit)(ifp, m);
1827 }
1828 
1829 /*
1830  * Simple round robin aggregation
1831  */
1832 static void
1833 lagg_rr_attach(struct lagg_softc *sc)
1834 {
1835 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
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 
2005 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
2006 
2007 	lb->lb_key = m_ether_tcpip_hash_init();
2008 	sc->sc_psc = lb;
2009 
2010 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2011 		lagg_lb_port_create(lp);
2012 }
2013 
2014 static void
2015 lagg_lb_detach(struct lagg_softc *sc)
2016 {
2017 	struct lagg_lb *lb;
2018 
2019 	lb = (struct lagg_lb *)sc->sc_psc;
2020 	LAGG_WUNLOCK(sc);
2021 	if (lb != NULL)
2022 		free(lb, M_DEVBUF);
2023 }
2024 
2025 static int
2026 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2027 {
2028 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2029 	struct lagg_port *lp_next;
2030 	int i = 0;
2031 
2032 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2033 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2034 		if (lp_next == lp)
2035 			continue;
2036 		if (i >= LAGG_MAX_PORTS)
2037 			return (EINVAL);
2038 		if (sc->sc_ifflags & IFF_DEBUG)
2039 			printf("%s: port %s at index %d\n",
2040 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2041 		lb->lb_ports[i++] = lp_next;
2042 	}
2043 
2044 	return (0);
2045 }
2046 
2047 static int
2048 lagg_lb_port_create(struct lagg_port *lp)
2049 {
2050 	struct lagg_softc *sc = lp->lp_softc;
2051 	return (lagg_lb_porttable(sc, NULL));
2052 }
2053 
2054 static void
2055 lagg_lb_port_destroy(struct lagg_port *lp)
2056 {
2057 	struct lagg_softc *sc = lp->lp_softc;
2058 	lagg_lb_porttable(sc, lp);
2059 }
2060 
2061 static int
2062 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2063 {
2064 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2065 	struct lagg_port *lp = NULL;
2066 	uint32_t p = 0;
2067 
2068 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2069 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2070 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2071 	else
2072 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2073 	p %= sc->sc_count;
2074 	lp = lb->lb_ports[p];
2075 
2076 	/*
2077 	 * Check the port's link state. This will return the next active
2078 	 * port if the link is down or the port is NULL.
2079 	 */
2080 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2081 		m_freem(m);
2082 		return (ENETDOWN);
2083 	}
2084 
2085 	/* Send mbuf */
2086 	return (lagg_enqueue(lp->lp_ifp, m));
2087 }
2088 
2089 static struct mbuf *
2090 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2091 {
2092 	struct ifnet *ifp = sc->sc_ifp;
2093 
2094 	/* Just pass in the packet to our lagg device */
2095 	m->m_pkthdr.rcvif = ifp;
2096 
2097 	return (m);
2098 }
2099 
2100 /*
2101  * 802.3ad LACP
2102  */
2103 static void
2104 lagg_lacp_attach(struct lagg_softc *sc)
2105 {
2106 	struct lagg_port *lp;
2107 
2108 	lacp_attach(sc);
2109 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2110 		lacp_port_create(lp);
2111 }
2112 
2113 static void
2114 lagg_lacp_detach(struct lagg_softc *sc)
2115 {
2116 	struct lagg_port *lp;
2117 	void *psc;
2118 
2119 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2120 		lacp_port_destroy(lp);
2121 
2122 	psc = sc->sc_psc;
2123 	sc->sc_psc = NULL;
2124 	LAGG_WUNLOCK(sc);
2125 
2126 	lacp_detach(psc);
2127 }
2128 
2129 static void
2130 lagg_lacp_lladdr(struct lagg_softc *sc)
2131 {
2132 	struct lagg_port *lp;
2133 
2134 	LAGG_SXLOCK_ASSERT(sc);
2135 
2136 	/* purge all the lacp ports */
2137 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2138 		lacp_port_destroy(lp);
2139 
2140 	/* add them back in */
2141 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2142 		lacp_port_create(lp);
2143 }
2144 
2145 static int
2146 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2147 {
2148 	struct lagg_port *lp;
2149 
2150 	lp = lacp_select_tx_port(sc, m);
2151 	if (lp == NULL) {
2152 		m_freem(m);
2153 		return (ENETDOWN);
2154 	}
2155 
2156 	/* Send mbuf */
2157 	return (lagg_enqueue(lp->lp_ifp, m));
2158 }
2159 
2160 static struct mbuf *
2161 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2162 {
2163 	struct ifnet *ifp = sc->sc_ifp;
2164 	struct ether_header *eh;
2165 	u_short etype;
2166 
2167 	eh = mtod(m, struct ether_header *);
2168 	etype = ntohs(eh->ether_type);
2169 
2170 	/* Tap off LACP control messages */
2171 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2172 		m = lacp_input(lp, m);
2173 		if (m == NULL)
2174 			return (NULL);
2175 	}
2176 
2177 	/*
2178 	 * If the port is not collecting or not in the active aggregator then
2179 	 * free and return.
2180 	 */
2181 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2182 		m_freem(m);
2183 		return (NULL);
2184 	}
2185 
2186 	m->m_pkthdr.rcvif = ifp;
2187 	return (m);
2188 }
2189 
2190