xref: /freebsd/sys/net/if_lagg.c (revision 0b3105a37d7adcadcb720112fed4dc4e8040be99)
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 
27 #include <sys/param.h>
28 #include <sys/kernel.h>
29 #include <sys/malloc.h>
30 #include <sys/mbuf.h>
31 #include <sys/queue.h>
32 #include <sys/socket.h>
33 #include <sys/sockio.h>
34 #include <sys/sysctl.h>
35 #include <sys/module.h>
36 #include <sys/priv.h>
37 #include <sys/systm.h>
38 #include <sys/proc.h>
39 #include <sys/lock.h>
40 #include <sys/rmlock.h>
41 #include <sys/taskqueue.h>
42 #include <sys/eventhandler.h>
43 
44 #include <net/ethernet.h>
45 #include <net/if.h>
46 #include <net/if_clone.h>
47 #include <net/if_arp.h>
48 #include <net/if_dl.h>
49 #include <net/if_media.h>
50 #include <net/if_types.h>
51 #include <net/if_var.h>
52 #include <net/bpf.h>
53 #include <net/vnet.h>
54 
55 #if defined(INET) || defined(INET6)
56 #include <netinet/in.h>
57 #include <netinet/ip.h>
58 #endif
59 #ifdef INET
60 #include <netinet/in_systm.h>
61 #include <netinet/if_ether.h>
62 #endif
63 
64 #ifdef INET6
65 #include <netinet/ip6.h>
66 #include <netinet6/in6_var.h>
67 #include <netinet6/in6_ifattach.h>
68 #endif
69 
70 #include <net/if_vlan_var.h>
71 #include <net/if_lagg.h>
72 #include <net/ieee8023ad_lacp.h>
73 
74 /* Special flags we should propagate to the lagg ports. */
75 static struct {
76 	int flag;
77 	int (*func)(struct ifnet *, int);
78 } lagg_pflags[] = {
79 	{IFF_PROMISC, ifpromisc},
80 	{IFF_ALLMULTI, if_allmulti},
81 	{0, NULL}
82 };
83 
84 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
85 #define	V_lagg_list	VNET(lagg_list)
86 static VNET_DEFINE(struct mtx, lagg_list_mtx);
87 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
88 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
89 					"if_lagg list", NULL, MTX_DEF)
90 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
91 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
92 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
93 eventhandler_tag	lagg_detach_cookie = NULL;
94 
95 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
96 static void	lagg_clone_destroy(struct ifnet *);
97 static VNET_DEFINE(struct if_clone *, lagg_cloner);
98 #define	V_lagg_cloner	VNET(lagg_cloner)
99 static const char laggname[] = "lagg";
100 
101 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
102 static void	lagg_capabilities(struct lagg_softc *);
103 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *, lagg_llqtype);
104 static void	lagg_port_setlladdr(void *, int);
105 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
106 static int	lagg_port_destroy(struct lagg_port *, int);
107 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
108 static void	lagg_linkstate(struct lagg_softc *);
109 static void	lagg_port_state(struct ifnet *, int);
110 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
111 static int	lagg_port_output(struct ifnet *, struct mbuf *,
112 		    const struct sockaddr *, struct route *);
113 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
114 #ifdef LAGG_PORT_STACKING
115 static int	lagg_port_checkstacking(struct lagg_softc *);
116 #endif
117 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
118 static void	lagg_init(void *);
119 static void	lagg_stop(struct lagg_softc *);
120 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
121 static int	lagg_ether_setmulti(struct lagg_softc *);
122 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
123 static	int	lagg_setflag(struct lagg_port *, int, int,
124 		    int (*func)(struct ifnet *, int));
125 static	int	lagg_setflags(struct lagg_port *, int status);
126 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
127 static int	lagg_transmit(struct ifnet *, struct mbuf *);
128 static void	lagg_qflush(struct ifnet *);
129 static int	lagg_media_change(struct ifnet *);
130 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
131 static struct lagg_port *lagg_link_active(struct lagg_softc *,
132 	    struct lagg_port *);
133 
134 /* Simple round robin */
135 static void	lagg_rr_attach(struct lagg_softc *);
136 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
137 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
138 		    struct mbuf *);
139 
140 /* Active failover */
141 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
142 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
143 		    struct mbuf *);
144 
145 /* Loadbalancing */
146 static void	lagg_lb_attach(struct lagg_softc *);
147 static void	lagg_lb_detach(struct lagg_softc *);
148 static int	lagg_lb_port_create(struct lagg_port *);
149 static void	lagg_lb_port_destroy(struct lagg_port *);
150 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
151 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
152 		    struct mbuf *);
153 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
154 
155 /* Broadcast */
156 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
157 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
158 		    struct mbuf *);
159 
160 /* 802.3ad LACP */
161 static void	lagg_lacp_attach(struct lagg_softc *);
162 static void	lagg_lacp_detach(struct lagg_softc *);
163 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
164 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
165 		    struct mbuf *);
166 static void	lagg_lacp_lladdr(struct lagg_softc *);
167 
168 /* lagg protocol table */
169 static const struct lagg_proto {
170 	lagg_proto	pr_num;
171 	void		(*pr_attach)(struct lagg_softc *);
172 	void		(*pr_detach)(struct lagg_softc *);
173 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
174 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
175 			    struct mbuf *);
176 	int		(*pr_addport)(struct lagg_port *);
177 	void		(*pr_delport)(struct lagg_port *);
178 	void		(*pr_linkstate)(struct lagg_port *);
179 	void 		(*pr_init)(struct lagg_softc *);
180 	void 		(*pr_stop)(struct lagg_softc *);
181 	void 		(*pr_lladdr)(struct lagg_softc *);
182 	void		(*pr_request)(struct lagg_softc *, void *);
183 	void		(*pr_portreq)(struct lagg_port *, void *);
184 } lagg_protos[] = {
185     {
186 	.pr_num = LAGG_PROTO_NONE
187     },
188     {
189 	.pr_num = LAGG_PROTO_ROUNDROBIN,
190 	.pr_attach = lagg_rr_attach,
191 	.pr_start = lagg_rr_start,
192 	.pr_input = lagg_rr_input,
193     },
194     {
195 	.pr_num = LAGG_PROTO_FAILOVER,
196 	.pr_start = lagg_fail_start,
197 	.pr_input = lagg_fail_input,
198     },
199     {
200 	.pr_num = LAGG_PROTO_LOADBALANCE,
201 	.pr_attach = lagg_lb_attach,
202 	.pr_detach = lagg_lb_detach,
203 	.pr_start = lagg_lb_start,
204 	.pr_input = lagg_lb_input,
205 	.pr_addport = lagg_lb_port_create,
206 	.pr_delport = lagg_lb_port_destroy,
207     },
208     {
209 	.pr_num = LAGG_PROTO_LACP,
210 	.pr_attach = lagg_lacp_attach,
211 	.pr_detach = lagg_lacp_detach,
212 	.pr_start = lagg_lacp_start,
213 	.pr_input = lagg_lacp_input,
214 	.pr_addport = lacp_port_create,
215 	.pr_delport = lacp_port_destroy,
216 	.pr_linkstate = lacp_linkstate,
217 	.pr_init = lacp_init,
218 	.pr_stop = lacp_stop,
219 	.pr_lladdr = lagg_lacp_lladdr,
220 	.pr_request = lacp_req,
221 	.pr_portreq = lacp_portreq,
222     },
223     {
224 	.pr_num = LAGG_PROTO_BROADCAST,
225 	.pr_start = lagg_bcast_start,
226 	.pr_input = lagg_bcast_input,
227     },
228 };
229 
230 SYSCTL_DECL(_net_link);
231 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
232     "Link Aggregation");
233 
234 /* Allow input on any failover links */
235 static VNET_DEFINE(int, lagg_failover_rx_all);
236 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
237 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
238     &VNET_NAME(lagg_failover_rx_all), 0,
239     "Accept input from any interface in a failover lagg");
240 
241 /* Default value for using flowid */
242 static VNET_DEFINE(int, def_use_flowid) = 1;
243 #define	V_def_use_flowid	VNET(def_use_flowid)
244 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
245     &VNET_NAME(def_use_flowid), 0,
246     "Default setting for using flow id for load sharing");
247 
248 /* Default value for flowid shift */
249 static VNET_DEFINE(int, def_flowid_shift) = 16;
250 #define	V_def_flowid_shift	VNET(def_flowid_shift)
251 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
252     &VNET_NAME(def_flowid_shift), 0,
253     "Default setting for flowid shift for load sharing");
254 
255 static void
256 vnet_lagg_init(const void *unused __unused)
257 {
258 
259 	LAGG_LIST_LOCK_INIT();
260 	SLIST_INIT(&V_lagg_list);
261 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
262 	    lagg_clone_destroy, 0);
263 }
264 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
265     vnet_lagg_init, NULL);
266 
267 static void
268 vnet_lagg_uninit(const void *unused __unused)
269 {
270 
271 	if_clone_detach(V_lagg_cloner);
272 	LAGG_LIST_LOCK_DESTROY();
273 }
274 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
275     vnet_lagg_uninit, NULL);
276 
277 static int
278 lagg_modevent(module_t mod, int type, void *data)
279 {
280 
281 	switch (type) {
282 	case MOD_LOAD:
283 		lagg_input_p = lagg_input;
284 		lagg_linkstate_p = lagg_port_state;
285 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
286 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
287 		    EVENTHANDLER_PRI_ANY);
288 		break;
289 	case MOD_UNLOAD:
290 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
291 		    lagg_detach_cookie);
292 		lagg_input_p = NULL;
293 		lagg_linkstate_p = NULL;
294 		break;
295 	default:
296 		return (EOPNOTSUPP);
297 	}
298 	return (0);
299 }
300 
301 static moduledata_t lagg_mod = {
302 	"if_lagg",
303 	lagg_modevent,
304 	0
305 };
306 
307 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
308 MODULE_VERSION(if_lagg, 1);
309 
310 static void
311 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
312 {
313 
314 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
315 	    __func__, sc));
316 
317 	if (sc->sc_ifflags & IFF_DEBUG)
318 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
319 
320 	if (lagg_protos[pr].pr_attach != NULL)
321 		lagg_protos[pr].pr_attach(sc);
322 	sc->sc_proto = pr;
323 }
324 
325 static void
326 lagg_proto_detach(struct lagg_softc *sc)
327 {
328 	lagg_proto pr;
329 
330 	LAGG_WLOCK_ASSERT(sc);
331 
332 	pr = sc->sc_proto;
333 	sc->sc_proto = LAGG_PROTO_NONE;
334 
335 	if (lagg_protos[pr].pr_detach != NULL)
336 		lagg_protos[pr].pr_detach(sc);
337 	else
338 		LAGG_WUNLOCK(sc);
339 }
340 
341 static int
342 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
343 {
344 
345 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
346 }
347 
348 static struct mbuf *
349 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
350 {
351 
352 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
353 }
354 
355 static int
356 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
357 {
358 
359 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
360 		return (0);
361 	else
362 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
363 }
364 
365 static void
366 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
367 {
368 
369 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
370 		lagg_protos[sc->sc_proto].pr_delport(lp);
371 }
372 
373 static void
374 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
375 {
376 
377 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
378 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
379 }
380 
381 static void
382 lagg_proto_init(struct lagg_softc *sc)
383 {
384 
385 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
386 		lagg_protos[sc->sc_proto].pr_init(sc);
387 }
388 
389 static void
390 lagg_proto_stop(struct lagg_softc *sc)
391 {
392 
393 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
394 		lagg_protos[sc->sc_proto].pr_stop(sc);
395 }
396 
397 static void
398 lagg_proto_lladdr(struct lagg_softc *sc)
399 {
400 
401 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
402 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
403 }
404 
405 static void
406 lagg_proto_request(struct lagg_softc *sc, void *v)
407 {
408 
409 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
410 		lagg_protos[sc->sc_proto].pr_request(sc, v);
411 }
412 
413 static void
414 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
415 {
416 
417 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
418 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
419 }
420 
421 /*
422  * This routine is run via an vlan
423  * config EVENT
424  */
425 static void
426 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
427 {
428 	struct lagg_softc *sc = ifp->if_softc;
429 	struct lagg_port *lp;
430 	struct rm_priotracker tracker;
431 
432 	if (ifp->if_softc !=  arg)   /* Not our event */
433 		return;
434 
435 	LAGG_RLOCK(sc, &tracker);
436 	if (!SLIST_EMPTY(&sc->sc_ports)) {
437 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
438 			EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
439 	}
440 	LAGG_RUNLOCK(sc, &tracker);
441 }
442 
443 /*
444  * This routine is run via an vlan
445  * unconfig EVENT
446  */
447 static void
448 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
449 {
450 	struct lagg_softc *sc = ifp->if_softc;
451 	struct lagg_port *lp;
452 	struct rm_priotracker tracker;
453 
454 	if (ifp->if_softc !=  arg)   /* Not our event */
455 		return;
456 
457 	LAGG_RLOCK(sc, &tracker);
458 	if (!SLIST_EMPTY(&sc->sc_ports)) {
459 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
460 			EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
461 	}
462 	LAGG_RUNLOCK(sc, &tracker);
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 
479 	if (V_def_use_flowid)
480 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
481 	sc->flowid_shift = V_def_flowid_shift;
482 
483 	/* Hash all layers by default */
484 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
485 
486 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
487 
488 	LAGG_LOCK_INIT(sc);
489 	SLIST_INIT(&sc->sc_ports);
490 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
491 
492 	/* Initialise pseudo media types */
493 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
494 	    lagg_media_status);
495 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
496 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
497 
498 	if_initname(ifp, laggname, unit);
499 	ifp->if_softc = sc;
500 	ifp->if_transmit = lagg_transmit;
501 	ifp->if_qflush = lagg_qflush;
502 	ifp->if_init = lagg_init;
503 	ifp->if_ioctl = lagg_ioctl;
504 	ifp->if_get_counter = lagg_get_counter;
505 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
506 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
507 
508 	/*
509 	 * Attach as an ordinary ethernet device, children will be attached
510 	 * as special device IFT_IEEE8023ADLAG.
511 	 */
512 	ether_ifattach(ifp, eaddr);
513 
514 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
515 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
516 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
517 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
518 
519 	/* Insert into the global list of laggs */
520 	LAGG_LIST_LOCK();
521 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
522 	LAGG_LIST_UNLOCK();
523 
524 	return (0);
525 }
526 
527 static void
528 lagg_clone_destroy(struct ifnet *ifp)
529 {
530 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
531 	struct lagg_port *lp;
532 
533 	LAGG_WLOCK(sc);
534 
535 	lagg_stop(sc);
536 	ifp->if_flags &= ~IFF_UP;
537 
538 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
539 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
540 
541 	/* Shutdown and remove lagg ports */
542 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
543 		lagg_port_destroy(lp, 1);
544 	/* Unhook the aggregation protocol */
545 	lagg_proto_detach(sc);
546 	LAGG_UNLOCK_ASSERT(sc);
547 
548 	ifmedia_removeall(&sc->sc_media);
549 	ether_ifdetach(ifp);
550 	if_free(ifp);
551 
552 	LAGG_LIST_LOCK();
553 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
554 	LAGG_LIST_UNLOCK();
555 
556 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
557 	LAGG_LOCK_DESTROY(sc);
558 	free(sc, M_DEVBUF);
559 }
560 
561 /*
562  * Set link-layer address on the lagg interface itself.
563  *
564  * Set noinline to be dtrace-friendly
565  */
566 static __noinline void
567 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
568 {
569 	struct ifnet *ifp = sc->sc_ifp;
570 	struct lagg_port lp;
571 
572 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
573 		return;
574 
575 	LAGG_WLOCK_ASSERT(sc);
576 	/*
577 	 * Set the link layer address on the lagg interface.
578 	 * lagg_proto_lladdr() notifies the MAC change to
579 	 * the aggregation protocol.  iflladdr_event handler which
580 	 * may trigger gratuitous ARPs for INET will be handled in
581 	 * a taskqueue.
582 	 */
583 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
584 	lagg_proto_lladdr(sc);
585 
586 	/*
587 	 * Send notification request for lagg interface
588 	 * itself. Note that new lladdr is already set.
589 	 */
590 	bzero(&lp, sizeof(lp));
591 	lp.lp_ifp = sc->sc_ifp;
592 	lp.lp_softc = sc;
593 
594 	/* Do not request lladdr change */
595 	lagg_port_lladdr(&lp, lladdr, LAGG_LLQTYPE_VIRT);
596 }
597 
598 static void
599 lagg_capabilities(struct lagg_softc *sc)
600 {
601 	struct lagg_port *lp;
602 	int cap = ~0, ena = ~0;
603 	u_long hwa = ~0UL;
604 	struct ifnet_hw_tsomax hw_tsomax;
605 
606 	LAGG_WLOCK_ASSERT(sc);
607 
608 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
609 
610 	/* Get capabilities from the lagg ports */
611 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
612 		cap &= lp->lp_ifp->if_capabilities;
613 		ena &= lp->lp_ifp->if_capenable;
614 		hwa &= lp->lp_ifp->if_hwassist;
615 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
616 	}
617 	cap = (cap == ~0 ? 0 : cap);
618 	ena = (ena == ~0 ? 0 : ena);
619 	hwa = (hwa == ~0 ? 0 : hwa);
620 
621 	if (sc->sc_ifp->if_capabilities != cap ||
622 	    sc->sc_ifp->if_capenable != ena ||
623 	    sc->sc_ifp->if_hwassist != hwa ||
624 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
625 		sc->sc_ifp->if_capabilities = cap;
626 		sc->sc_ifp->if_capenable = ena;
627 		sc->sc_ifp->if_hwassist = hwa;
628 		getmicrotime(&sc->sc_ifp->if_lastchange);
629 
630 		if (sc->sc_ifflags & IFF_DEBUG)
631 			if_printf(sc->sc_ifp,
632 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
633 	}
634 }
635 
636 /*
637  * Enqueue interface lladdr notification.
638  * If request is already queued, it is updated.
639  * If setting lladdr is also desired, @do_change has to be set to 1.
640  *
641  * Set noinline to be dtrace-friendly
642  */
643 static __noinline void
644 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr, lagg_llqtype llq_type)
645 {
646 	struct lagg_softc *sc = lp->lp_softc;
647 	struct ifnet *ifp = lp->lp_ifp;
648 	struct lagg_llq *llq;
649 
650 	LAGG_WLOCK_ASSERT(sc);
651 
652 	/*
653 	 * Do not enqueue requests where lladdr is the same for
654 	 * "physical" interfaces (e.g. ports in lagg)
655 	 */
656 	if (llq_type == LAGG_LLQTYPE_PHYS &&
657 	    memcmp(IF_LLADDR(ifp), lladdr, ETHER_ADDR_LEN) == 0)
658 		return;
659 
660 	/* Check to make sure its not already queued to be changed */
661 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
662 		if (llq->llq_ifp == ifp) {
663 			/* Update lladdr, it may have changed */
664 			bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
665 			return;
666 		}
667 	}
668 
669 	llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT | M_ZERO);
670 	if (llq == NULL)	/* XXX what to do */
671 		return;
672 
673 	llq->llq_ifp = ifp;
674 	llq->llq_type = llq_type;
675 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
676 	/* XXX: We should insert to tail */
677 	SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
678 
679 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
680 }
681 
682 /*
683  * Set the interface MAC address from a taskqueue to avoid a LOR.
684  *
685  * Set noinline to be dtrace-friendly
686  */
687 static __noinline void
688 lagg_port_setlladdr(void *arg, int pending)
689 {
690 	struct lagg_softc *sc = (struct lagg_softc *)arg;
691 	struct lagg_llq *llq, *head;
692 	struct ifnet *ifp;
693 
694 	/* Grab a local reference of the queue and remove it from the softc */
695 	LAGG_WLOCK(sc);
696 	head = SLIST_FIRST(&sc->sc_llq_head);
697 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
698 	LAGG_WUNLOCK(sc);
699 
700 	/*
701 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
702 	 * unlocked as we have the only reference to it.
703 	 */
704 	for (llq = head; llq != NULL; llq = head) {
705 		ifp = llq->llq_ifp;
706 
707 		CURVNET_SET(ifp->if_vnet);
708 
709 		/*
710 		 * Set the link layer address on the laggport interface.
711 		 * Note that if_setlladdr() or iflladdr_event handler
712 		 * may result in arp transmission / lltable updates.
713 		 */
714 		if (llq->llq_type == LAGG_LLQTYPE_PHYS)
715 			if_setlladdr(ifp, llq->llq_lladdr,
716 			    ETHER_ADDR_LEN);
717 		else
718 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
719 		CURVNET_RESTORE();
720 		head = SLIST_NEXT(llq, llq_entries);
721 		free(llq, M_DEVBUF);
722 	}
723 }
724 
725 static int
726 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
727 {
728 	struct lagg_softc *sc_ptr;
729 	struct lagg_port *lp, *tlp;
730 	int error, i;
731 	uint64_t *pval;
732 
733 	LAGG_WLOCK_ASSERT(sc);
734 
735 	/* Limit the maximal number of lagg ports */
736 	if (sc->sc_count >= LAGG_MAX_PORTS)
737 		return (ENOSPC);
738 
739 	/* Check if port has already been associated to a lagg */
740 	if (ifp->if_lagg != NULL) {
741 		/* Port is already in the current lagg? */
742 		lp = (struct lagg_port *)ifp->if_lagg;
743 		if (lp->lp_softc == sc)
744 			return (EEXIST);
745 		return (EBUSY);
746 	}
747 
748 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
749 	if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
750 		return (EPROTONOSUPPORT);
751 
752 	/* Allow the first Ethernet member to define the MTU */
753 	if (SLIST_EMPTY(&sc->sc_ports))
754 		sc->sc_ifp->if_mtu = ifp->if_mtu;
755 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
756 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
757 		    ifp->if_xname);
758 		return (EINVAL);
759 	}
760 
761 	if ((lp = malloc(sizeof(struct lagg_port),
762 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
763 		return (ENOMEM);
764 
765 	/* Check if port is a stacked lagg */
766 	LAGG_LIST_LOCK();
767 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
768 		if (ifp == sc_ptr->sc_ifp) {
769 			LAGG_LIST_UNLOCK();
770 			free(lp, M_DEVBUF);
771 			return (EINVAL);
772 			/* XXX disable stacking for the moment, its untested */
773 #ifdef LAGG_PORT_STACKING
774 			lp->lp_flags |= LAGG_PORT_STACK;
775 			if (lagg_port_checkstacking(sc_ptr) >=
776 			    LAGG_MAX_STACKING) {
777 				LAGG_LIST_UNLOCK();
778 				free(lp, M_DEVBUF);
779 				return (E2BIG);
780 			}
781 #endif
782 		}
783 	}
784 	LAGG_LIST_UNLOCK();
785 
786 	/* Change the interface type */
787 	lp->lp_iftype = ifp->if_type;
788 	ifp->if_type = IFT_IEEE8023ADLAG;
789 	ifp->if_lagg = lp;
790 	lp->lp_ioctl = ifp->if_ioctl;
791 	ifp->if_ioctl = lagg_port_ioctl;
792 	lp->lp_output = ifp->if_output;
793 	ifp->if_output = lagg_port_output;
794 
795 	lp->lp_ifp = ifp;
796 	lp->lp_softc = sc;
797 
798 	/* Save port link layer address */
799 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
800 
801 	if (SLIST_EMPTY(&sc->sc_ports)) {
802 		sc->sc_primary = lp;
803 		/* First port in lagg. Update/notify lagg lladdress */
804 		lagg_lladdr(sc, IF_LLADDR(ifp));
805 	} else {
806 
807 		/*
808 		 * Update link layer address for this port and
809 		 * send notifications to other subsystems.
810 		 */
811 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp), LAGG_LLQTYPE_PHYS);
812 	}
813 
814 	/*
815 	 * Insert into the list of ports.
816 	 * Keep ports sorted by if_index. It is handy, when configuration
817 	 * is predictable and `ifconfig laggN create ...` command
818 	 * will lead to the same result each time.
819 	 */
820 	SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
821 		if (tlp->lp_ifp->if_index < ifp->if_index && (
822 		    SLIST_NEXT(tlp, lp_entries) == NULL ||
823 		    SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index >
824 		    ifp->if_index))
825 			break;
826 	}
827 	if (tlp != NULL)
828 		SLIST_INSERT_AFTER(tlp, lp, lp_entries);
829 	else
830 		SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
831 	sc->sc_count++;
832 
833 	/* Update lagg capabilities */
834 	lagg_capabilities(sc);
835 	lagg_linkstate(sc);
836 
837 	/* Read port counters */
838 	pval = lp->port_counters.val;
839 	for (i = 0; i < IFCOUNTERS; i++, pval++)
840 		*pval = ifp->if_get_counter(ifp, i);
841 	/* Add multicast addresses and interface flags to this port */
842 	lagg_ether_cmdmulti(lp, 1);
843 	lagg_setflags(lp, 1);
844 
845 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
846 		/* Remove the port, without calling pr_delport. */
847 		lagg_port_destroy(lp, 0);
848 		return (error);
849 	}
850 
851 	return (0);
852 }
853 
854 #ifdef LAGG_PORT_STACKING
855 static int
856 lagg_port_checkstacking(struct lagg_softc *sc)
857 {
858 	struct lagg_softc *sc_ptr;
859 	struct lagg_port *lp;
860 	int m = 0;
861 
862 	LAGG_WLOCK_ASSERT(sc);
863 
864 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
865 		if (lp->lp_flags & LAGG_PORT_STACK) {
866 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
867 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
868 		}
869 	}
870 
871 	return (m + 1);
872 }
873 #endif
874 
875 static int
876 lagg_port_destroy(struct lagg_port *lp, int rundelport)
877 {
878 	struct lagg_softc *sc = lp->lp_softc;
879 	struct lagg_port *lp_ptr, *lp0;
880 	struct lagg_llq *llq;
881 	struct ifnet *ifp = lp->lp_ifp;
882 	uint64_t *pval, vdiff;
883 	int i;
884 
885 	LAGG_WLOCK_ASSERT(sc);
886 
887 	if (rundelport)
888 		lagg_proto_delport(sc, lp);
889 
890 	/*
891 	 * Remove multicast addresses and interface flags from this port and
892 	 * reset the MAC address, skip if the interface is being detached.
893 	 */
894 	if (!lp->lp_detaching) {
895 		lagg_ether_cmdmulti(lp, 0);
896 		lagg_setflags(lp, 0);
897 		lagg_port_lladdr(lp, lp->lp_lladdr, LAGG_LLQTYPE_PHYS);
898 	}
899 
900 	/* Restore interface */
901 	ifp->if_type = lp->lp_iftype;
902 	ifp->if_ioctl = lp->lp_ioctl;
903 	ifp->if_output = lp->lp_output;
904 	ifp->if_lagg = NULL;
905 
906 	/* Update detached port counters */
907 	pval = lp->port_counters.val;
908 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
909 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
910 		sc->detached_counters.val[i] += vdiff;
911 	}
912 
913 	/* Finally, remove the port from the lagg */
914 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
915 	sc->sc_count--;
916 
917 	/* Update the primary interface */
918 	if (lp == sc->sc_primary) {
919 		uint8_t lladdr[ETHER_ADDR_LEN];
920 
921 		if ((lp0 = SLIST_FIRST(&sc->sc_ports)) == NULL) {
922 			bzero(&lladdr, ETHER_ADDR_LEN);
923 		} else {
924 			bcopy(lp0->lp_lladdr,
925 			    lladdr, ETHER_ADDR_LEN);
926 		}
927 		lagg_lladdr(sc, lladdr);
928 
929 		/* Mark lp0 as new primary */
930 		sc->sc_primary = lp0;
931 
932 		/*
933 		 * Enqueue lladdr update/notification for each port
934 		 * (new primary needs update as well, to switch from
935 		 * old lladdr to its 'real' one).
936 		 */
937 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
938 			lagg_port_lladdr(lp_ptr, lladdr, LAGG_LLQTYPE_PHYS);
939 	}
940 
941 	/* Remove any pending lladdr changes from the queue */
942 	if (lp->lp_detaching) {
943 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
944 			if (llq->llq_ifp == ifp) {
945 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
946 				    llq_entries);
947 				free(llq, M_DEVBUF);
948 				break;	/* Only appears once */
949 			}
950 		}
951 	}
952 
953 	if (lp->lp_ifflags)
954 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
955 
956 	free(lp, M_DEVBUF);
957 
958 	/* Update lagg capabilities */
959 	lagg_capabilities(sc);
960 	lagg_linkstate(sc);
961 
962 	return (0);
963 }
964 
965 static int
966 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
967 {
968 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
969 	struct lagg_softc *sc;
970 	struct lagg_port *lp = NULL;
971 	int error = 0;
972 	struct rm_priotracker tracker;
973 
974 	/* Should be checked by the caller */
975 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
976 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
977 		goto fallback;
978 
979 	switch (cmd) {
980 	case SIOCGLAGGPORT:
981 		if (rp->rp_portname[0] == '\0' ||
982 		    ifunit(rp->rp_portname) != ifp) {
983 			error = EINVAL;
984 			break;
985 		}
986 
987 		LAGG_RLOCK(sc, &tracker);
988 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
989 			error = ENOENT;
990 			LAGG_RUNLOCK(sc, &tracker);
991 			break;
992 		}
993 
994 		lagg_port2req(lp, rp);
995 		LAGG_RUNLOCK(sc, &tracker);
996 		break;
997 
998 	case SIOCSIFCAP:
999 		if (lp->lp_ioctl == NULL) {
1000 			error = EINVAL;
1001 			break;
1002 		}
1003 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1004 		if (error)
1005 			break;
1006 
1007 		/* Update lagg interface capabilities */
1008 		LAGG_WLOCK(sc);
1009 		lagg_capabilities(sc);
1010 		LAGG_WUNLOCK(sc);
1011 		break;
1012 
1013 	case SIOCSIFMTU:
1014 		/* Do not allow the MTU to be changed once joined */
1015 		error = EINVAL;
1016 		break;
1017 
1018 	default:
1019 		goto fallback;
1020 	}
1021 
1022 	return (error);
1023 
1024 fallback:
1025 	if (lp->lp_ioctl != NULL)
1026 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1027 
1028 	return (EINVAL);
1029 }
1030 
1031 /*
1032  * Requests counter @cnt data.
1033  *
1034  * Counter value is calculated the following way:
1035  * 1) for each port, sum  difference between current and "initial" measurements.
1036  * 2) add lagg logical interface counters.
1037  * 3) add data from detached_counters array.
1038  *
1039  * We also do the following things on ports attach/detach:
1040  * 1) On port attach we store all counters it has into port_counter array.
1041  * 2) On port detach we add the different between "initial" and
1042  *   current counters data to detached_counters array.
1043  */
1044 static uint64_t
1045 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1046 {
1047 	struct lagg_softc *sc;
1048 	struct lagg_port *lp;
1049 	struct ifnet *lpifp;
1050 	struct rm_priotracker tracker;
1051 	uint64_t newval, oldval, vsum;
1052 
1053 	/* Revise this when we've got non-generic counters. */
1054 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1055 
1056 	sc = (struct lagg_softc *)ifp->if_softc;
1057 	LAGG_RLOCK(sc, &tracker);
1058 
1059 	vsum = 0;
1060 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1061 		/* Saved attached value */
1062 		oldval = lp->port_counters.val[cnt];
1063 		/* current value */
1064 		lpifp = lp->lp_ifp;
1065 		newval = lpifp->if_get_counter(lpifp, cnt);
1066 		/* Calculate diff and save new */
1067 		vsum += newval - oldval;
1068 	}
1069 
1070 	/*
1071 	 * Add counter data which might be added by upper
1072 	 * layer protocols operating on logical interface.
1073 	 */
1074 	vsum += if_get_counter_default(ifp, cnt);
1075 
1076 	/*
1077 	 * Add counter data from detached ports counters
1078 	 */
1079 	vsum += sc->detached_counters.val[cnt];
1080 
1081 	LAGG_RUNLOCK(sc, &tracker);
1082 
1083 	return (vsum);
1084 }
1085 
1086 /*
1087  * For direct output to child ports.
1088  */
1089 static int
1090 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1091 	const struct sockaddr *dst, struct route *ro)
1092 {
1093 	struct lagg_port *lp = ifp->if_lagg;
1094 
1095 	switch (dst->sa_family) {
1096 		case pseudo_AF_HDRCMPLT:
1097 		case AF_UNSPEC:
1098 			return ((*lp->lp_output)(ifp, m, dst, ro));
1099 	}
1100 
1101 	/* drop any other frames */
1102 	m_freem(m);
1103 	return (ENETDOWN);
1104 }
1105 
1106 static void
1107 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1108 {
1109 	struct lagg_port *lp;
1110 	struct lagg_softc *sc;
1111 
1112 	if ((lp = ifp->if_lagg) == NULL)
1113 		return;
1114 	/* If the ifnet is just being renamed, don't do anything. */
1115 	if (ifp->if_flags & IFF_RENAMING)
1116 		return;
1117 
1118 	sc = lp->lp_softc;
1119 
1120 	LAGG_WLOCK(sc);
1121 	lp->lp_detaching = 1;
1122 	lagg_port_destroy(lp, 1);
1123 	LAGG_WUNLOCK(sc);
1124 }
1125 
1126 static void
1127 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1128 {
1129 	struct lagg_softc *sc = lp->lp_softc;
1130 
1131 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1132 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1133 	rp->rp_prio = lp->lp_prio;
1134 	rp->rp_flags = lp->lp_flags;
1135 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1136 
1137 	/* Add protocol specific flags */
1138 	switch (sc->sc_proto) {
1139 		case LAGG_PROTO_FAILOVER:
1140 			if (lp == sc->sc_primary)
1141 				rp->rp_flags |= LAGG_PORT_MASTER;
1142 			if (lp == lagg_link_active(sc, sc->sc_primary))
1143 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1144 			break;
1145 
1146 		case LAGG_PROTO_ROUNDROBIN:
1147 		case LAGG_PROTO_LOADBALANCE:
1148 		case LAGG_PROTO_BROADCAST:
1149 			if (LAGG_PORTACTIVE(lp))
1150 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1151 			break;
1152 
1153 		case LAGG_PROTO_LACP:
1154 			/* LACP has a different definition of active */
1155 			if (lacp_isactive(lp))
1156 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1157 			if (lacp_iscollecting(lp))
1158 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1159 			if (lacp_isdistributing(lp))
1160 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1161 			break;
1162 	}
1163 
1164 }
1165 
1166 static void
1167 lagg_init(void *xsc)
1168 {
1169 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1170 	struct ifnet *ifp = sc->sc_ifp;
1171 	struct lagg_port *lp;
1172 
1173 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1174 		return;
1175 
1176 	LAGG_WLOCK(sc);
1177 
1178 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1179 
1180 	/*
1181 	 * Update the port lladdrs if needed.
1182 	 * This might be if_setlladdr() notification
1183 	 * that lladdr has been changed.
1184 	 */
1185 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1186 		lagg_port_lladdr(lp, IF_LLADDR(ifp), LAGG_LLQTYPE_PHYS);
1187 
1188 	lagg_proto_init(sc);
1189 
1190 	LAGG_WUNLOCK(sc);
1191 }
1192 
1193 static void
1194 lagg_stop(struct lagg_softc *sc)
1195 {
1196 	struct ifnet *ifp = sc->sc_ifp;
1197 
1198 	LAGG_WLOCK_ASSERT(sc);
1199 
1200 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1201 		return;
1202 
1203 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1204 
1205 	lagg_proto_stop(sc);
1206 }
1207 
1208 static int
1209 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1210 {
1211 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1212 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1213 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1214 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1215 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1216 	struct ifreq *ifr = (struct ifreq *)data;
1217 	struct lagg_port *lp;
1218 	struct ifnet *tpif;
1219 	struct thread *td = curthread;
1220 	char *buf, *outbuf;
1221 	int count, buflen, len, error = 0;
1222 	struct rm_priotracker tracker;
1223 
1224 	bzero(&rpbuf, sizeof(rpbuf));
1225 
1226 	switch (cmd) {
1227 	case SIOCGLAGG:
1228 		LAGG_RLOCK(sc, &tracker);
1229 		count = 0;
1230 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1231 			count++;
1232 		buflen = count * sizeof(struct lagg_reqport);
1233 		LAGG_RUNLOCK(sc, &tracker);
1234 
1235 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1236 
1237 		LAGG_RLOCK(sc, &tracker);
1238 		ra->ra_proto = sc->sc_proto;
1239 		lagg_proto_request(sc, &ra->ra_psc);
1240 		count = 0;
1241 		buf = outbuf;
1242 		len = min(ra->ra_size, buflen);
1243 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1244 			if (len < sizeof(rpbuf))
1245 				break;
1246 
1247 			lagg_port2req(lp, &rpbuf);
1248 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1249 			count++;
1250 			buf += sizeof(rpbuf);
1251 			len -= sizeof(rpbuf);
1252 		}
1253 		LAGG_RUNLOCK(sc, &tracker);
1254 		ra->ra_ports = count;
1255 		ra->ra_size = count * sizeof(rpbuf);
1256 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1257 		free(outbuf, M_TEMP);
1258 		break;
1259 	case SIOCSLAGG:
1260 		error = priv_check(td, PRIV_NET_LAGG);
1261 		if (error)
1262 			break;
1263 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1264 			error = EPROTONOSUPPORT;
1265 			break;
1266 		}
1267 
1268 		LAGG_WLOCK(sc);
1269 		lagg_proto_detach(sc);
1270 		LAGG_UNLOCK_ASSERT(sc);
1271 		lagg_proto_attach(sc, ra->ra_proto);
1272 		break;
1273 	case SIOCGLAGGOPTS:
1274 		ro->ro_opts = sc->sc_opts;
1275 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1276 			struct lacp_softc *lsc;
1277 
1278 			lsc = (struct lacp_softc *)sc->sc_psc;
1279 			if (lsc->lsc_debug.lsc_tx_test != 0)
1280 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1281 			if (lsc->lsc_debug.lsc_rx_test != 0)
1282 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1283 			if (lsc->lsc_strict_mode != 0)
1284 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1285 			if (lsc->lsc_fast_timeout != 0)
1286 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1287 
1288 			ro->ro_active = sc->sc_active;
1289 		} else {
1290 			ro->ro_active = 0;
1291 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1292 				ro->ro_active += LAGG_PORTACTIVE(lp);
1293 		}
1294 		ro->ro_bkt = sc->sc_bkt;
1295 		ro->ro_flapping = sc->sc_flapping;
1296 		ro->ro_flowid_shift = sc->flowid_shift;
1297 		break;
1298 	case SIOCSLAGGOPTS:
1299 		if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) {
1300 			if (ro->ro_bkt == 0)
1301 				sc->sc_bkt = 1; // Minimum 1 packet per iface.
1302 			else
1303 				sc->sc_bkt = ro->ro_bkt;
1304 		}
1305 		error = priv_check(td, PRIV_NET_LAGG);
1306 		if (error)
1307 			break;
1308 		if (ro->ro_opts == 0)
1309 			break;
1310 		/*
1311 		 * Set options.  LACP options are stored in sc->sc_psc,
1312 		 * not in sc_opts.
1313 		 */
1314 		int valid, lacp;
1315 
1316 		switch (ro->ro_opts) {
1317 		case LAGG_OPT_USE_FLOWID:
1318 		case -LAGG_OPT_USE_FLOWID:
1319 		case LAGG_OPT_FLOWIDSHIFT:
1320 			valid = 1;
1321 			lacp = 0;
1322 			break;
1323 		case LAGG_OPT_LACP_TXTEST:
1324 		case -LAGG_OPT_LACP_TXTEST:
1325 		case LAGG_OPT_LACP_RXTEST:
1326 		case -LAGG_OPT_LACP_RXTEST:
1327 		case LAGG_OPT_LACP_STRICT:
1328 		case -LAGG_OPT_LACP_STRICT:
1329 		case LAGG_OPT_LACP_TIMEOUT:
1330 		case -LAGG_OPT_LACP_TIMEOUT:
1331 			valid = lacp = 1;
1332 			break;
1333 		default:
1334 			valid = lacp = 0;
1335 			break;
1336 		}
1337 
1338 		LAGG_WLOCK(sc);
1339 
1340 		if (valid == 0 ||
1341 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1342 			/* Invalid combination of options specified. */
1343 			error = EINVAL;
1344 			LAGG_WUNLOCK(sc);
1345 			break;	/* Return from SIOCSLAGGOPTS. */
1346 		}
1347 		/*
1348 		 * Store new options into sc->sc_opts except for
1349 		 * FLOWIDSHIFT and LACP options.
1350 		 */
1351 		if (lacp == 0) {
1352 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1353 				sc->flowid_shift = ro->ro_flowid_shift;
1354 			else if (ro->ro_opts > 0)
1355 				sc->sc_opts |= ro->ro_opts;
1356 			else
1357 				sc->sc_opts &= ~ro->ro_opts;
1358 		} else {
1359 			struct lacp_softc *lsc;
1360 			struct lacp_port *lp;
1361 
1362 			lsc = (struct lacp_softc *)sc->sc_psc;
1363 
1364 			switch (ro->ro_opts) {
1365 			case LAGG_OPT_LACP_TXTEST:
1366 				lsc->lsc_debug.lsc_tx_test = 1;
1367 				break;
1368 			case -LAGG_OPT_LACP_TXTEST:
1369 				lsc->lsc_debug.lsc_tx_test = 0;
1370 				break;
1371 			case LAGG_OPT_LACP_RXTEST:
1372 				lsc->lsc_debug.lsc_rx_test = 1;
1373 				break;
1374 			case -LAGG_OPT_LACP_RXTEST:
1375 				lsc->lsc_debug.lsc_rx_test = 0;
1376 				break;
1377 			case LAGG_OPT_LACP_STRICT:
1378 				lsc->lsc_strict_mode = 1;
1379 				break;
1380 			case -LAGG_OPT_LACP_STRICT:
1381 				lsc->lsc_strict_mode = 0;
1382 				break;
1383 			case LAGG_OPT_LACP_TIMEOUT:
1384 				LACP_LOCK(lsc);
1385         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1386                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1387 				LACP_UNLOCK(lsc);
1388 				lsc->lsc_fast_timeout = 1;
1389 				break;
1390 			case -LAGG_OPT_LACP_TIMEOUT:
1391 				LACP_LOCK(lsc);
1392         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1393                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1394 				LACP_UNLOCK(lsc);
1395 				lsc->lsc_fast_timeout = 0;
1396 				break;
1397 			}
1398 		}
1399 		LAGG_WUNLOCK(sc);
1400 		break;
1401 	case SIOCGLAGGFLAGS:
1402 		rf->rf_flags = 0;
1403 		LAGG_RLOCK(sc, &tracker);
1404 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1405 			rf->rf_flags |= LAGG_F_HASHL2;
1406 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1407 			rf->rf_flags |= LAGG_F_HASHL3;
1408 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1409 			rf->rf_flags |= LAGG_F_HASHL4;
1410 		LAGG_RUNLOCK(sc, &tracker);
1411 		break;
1412 	case SIOCSLAGGHASH:
1413 		error = priv_check(td, PRIV_NET_LAGG);
1414 		if (error)
1415 			break;
1416 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1417 			error = EINVAL;
1418 			break;
1419 		}
1420 		LAGG_WLOCK(sc);
1421 		sc->sc_flags = 0;
1422 		if (rf->rf_flags & LAGG_F_HASHL2)
1423 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1424 		if (rf->rf_flags & LAGG_F_HASHL3)
1425 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1426 		if (rf->rf_flags & LAGG_F_HASHL4)
1427 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1428 		LAGG_WUNLOCK(sc);
1429 		break;
1430 	case SIOCGLAGGPORT:
1431 		if (rp->rp_portname[0] == '\0' ||
1432 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1433 			error = EINVAL;
1434 			break;
1435 		}
1436 
1437 		LAGG_RLOCK(sc, &tracker);
1438 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1439 		    lp->lp_softc != sc) {
1440 			error = ENOENT;
1441 			LAGG_RUNLOCK(sc, &tracker);
1442 			break;
1443 		}
1444 
1445 		lagg_port2req(lp, rp);
1446 		LAGG_RUNLOCK(sc, &tracker);
1447 		break;
1448 	case SIOCSLAGGPORT:
1449 		error = priv_check(td, PRIV_NET_LAGG);
1450 		if (error)
1451 			break;
1452 		if (rp->rp_portname[0] == '\0' ||
1453 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1454 			error = EINVAL;
1455 			break;
1456 		}
1457 #ifdef INET6
1458 		/*
1459 		 * A laggport interface should not have inet6 address
1460 		 * because two interfaces with a valid link-local
1461 		 * scope zone must not be merged in any form.  This
1462 		 * restriction is needed to prevent violation of
1463 		 * link-local scope zone.  Attempts to add a laggport
1464 		 * interface which has inet6 addresses triggers
1465 		 * removal of all inet6 addresses on the member
1466 		 * interface.
1467 		 */
1468 		if (in6ifa_llaonifp(tpif)) {
1469 			in6_ifdetach(tpif);
1470 				if_printf(sc->sc_ifp,
1471 				    "IPv6 addresses on %s have been removed "
1472 				    "before adding it as a member to prevent "
1473 				    "IPv6 address scope violation.\n",
1474 				    tpif->if_xname);
1475 		}
1476 #endif
1477 		LAGG_WLOCK(sc);
1478 		error = lagg_port_create(sc, tpif);
1479 		LAGG_WUNLOCK(sc);
1480 		break;
1481 	case SIOCSLAGGDELPORT:
1482 		error = priv_check(td, PRIV_NET_LAGG);
1483 		if (error)
1484 			break;
1485 		if (rp->rp_portname[0] == '\0' ||
1486 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1487 			error = EINVAL;
1488 			break;
1489 		}
1490 
1491 		LAGG_WLOCK(sc);
1492 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1493 		    lp->lp_softc != sc) {
1494 			error = ENOENT;
1495 			LAGG_WUNLOCK(sc);
1496 			break;
1497 		}
1498 
1499 		error = lagg_port_destroy(lp, 1);
1500 		LAGG_WUNLOCK(sc);
1501 		break;
1502 	case SIOCSIFFLAGS:
1503 		/* Set flags on ports too */
1504 		LAGG_WLOCK(sc);
1505 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1506 			lagg_setflags(lp, 1);
1507 		}
1508 		LAGG_WUNLOCK(sc);
1509 
1510 		if (!(ifp->if_flags & IFF_UP) &&
1511 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1512 			/*
1513 			 * If interface is marked down and it is running,
1514 			 * then stop and disable it.
1515 			 */
1516 			LAGG_WLOCK(sc);
1517 			lagg_stop(sc);
1518 			LAGG_WUNLOCK(sc);
1519 		} else if ((ifp->if_flags & IFF_UP) &&
1520 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1521 			/*
1522 			 * If interface is marked up and it is stopped, then
1523 			 * start it.
1524 			 */
1525 			(*ifp->if_init)(sc);
1526 		}
1527 		break;
1528 	case SIOCADDMULTI:
1529 	case SIOCDELMULTI:
1530 		LAGG_WLOCK(sc);
1531 		error = lagg_ether_setmulti(sc);
1532 		LAGG_WUNLOCK(sc);
1533 		break;
1534 	case SIOCSIFMEDIA:
1535 	case SIOCGIFMEDIA:
1536 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1537 		break;
1538 
1539 	case SIOCSIFCAP:
1540 	case SIOCSIFMTU:
1541 		/* Do not allow the MTU or caps to be directly changed */
1542 		error = EINVAL;
1543 		break;
1544 
1545 	default:
1546 		error = ether_ioctl(ifp, cmd, data);
1547 		break;
1548 	}
1549 	return (error);
1550 }
1551 
1552 static int
1553 lagg_ether_setmulti(struct lagg_softc *sc)
1554 {
1555 	struct lagg_port *lp;
1556 
1557 	LAGG_WLOCK_ASSERT(sc);
1558 
1559 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1560 		/* First, remove any existing filter entries. */
1561 		lagg_ether_cmdmulti(lp, 0);
1562 		/* copy all addresses from the lagg interface to the port */
1563 		lagg_ether_cmdmulti(lp, 1);
1564 	}
1565 	return (0);
1566 }
1567 
1568 static int
1569 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1570 {
1571 	struct lagg_softc *sc = lp->lp_softc;
1572 	struct ifnet *ifp = lp->lp_ifp;
1573 	struct ifnet *scifp = sc->sc_ifp;
1574 	struct lagg_mc *mc;
1575 	struct ifmultiaddr *ifma;
1576 	int error;
1577 
1578 	LAGG_WLOCK_ASSERT(sc);
1579 
1580 	if (set) {
1581 		IF_ADDR_WLOCK(scifp);
1582 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1583 			if (ifma->ifma_addr->sa_family != AF_LINK)
1584 				continue;
1585 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1586 			if (mc == NULL) {
1587 				IF_ADDR_WUNLOCK(scifp);
1588 				return (ENOMEM);
1589 			}
1590 			bcopy(ifma->ifma_addr, &mc->mc_addr,
1591 			    ifma->ifma_addr->sa_len);
1592 			mc->mc_addr.sdl_index = ifp->if_index;
1593 			mc->mc_ifma = NULL;
1594 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1595 		}
1596 		IF_ADDR_WUNLOCK(scifp);
1597 		SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1598 			error = if_addmulti(ifp,
1599 			    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1600 			if (error)
1601 				return (error);
1602 		}
1603 	} else {
1604 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1605 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1606 			if (mc->mc_ifma && !lp->lp_detaching)
1607 				if_delmulti_ifma(mc->mc_ifma);
1608 			free(mc, M_DEVBUF);
1609 		}
1610 	}
1611 	return (0);
1612 }
1613 
1614 /* Handle a ref counted flag that should be set on the lagg port as well */
1615 static int
1616 lagg_setflag(struct lagg_port *lp, int flag, int status,
1617     int (*func)(struct ifnet *, int))
1618 {
1619 	struct lagg_softc *sc = lp->lp_softc;
1620 	struct ifnet *scifp = sc->sc_ifp;
1621 	struct ifnet *ifp = lp->lp_ifp;
1622 	int error;
1623 
1624 	LAGG_WLOCK_ASSERT(sc);
1625 
1626 	status = status ? (scifp->if_flags & flag) : 0;
1627 	/* Now "status" contains the flag value or 0 */
1628 
1629 	/*
1630 	 * See if recorded ports status is different from what
1631 	 * we want it to be.  If it is, flip it.  We record ports
1632 	 * status in lp_ifflags so that we won't clear ports flag
1633 	 * we haven't set.  In fact, we don't clear or set ports
1634 	 * flags directly, but get or release references to them.
1635 	 * That's why we can be sure that recorded flags still are
1636 	 * in accord with actual ports flags.
1637 	 */
1638 	if (status != (lp->lp_ifflags & flag)) {
1639 		error = (*func)(ifp, status);
1640 		if (error)
1641 			return (error);
1642 		lp->lp_ifflags &= ~flag;
1643 		lp->lp_ifflags |= status;
1644 	}
1645 	return (0);
1646 }
1647 
1648 /*
1649  * Handle IFF_* flags that require certain changes on the lagg port
1650  * if "status" is true, update ports flags respective to the lagg
1651  * if "status" is false, forcedly clear the flags set on port.
1652  */
1653 static int
1654 lagg_setflags(struct lagg_port *lp, int status)
1655 {
1656 	int error, i;
1657 
1658 	for (i = 0; lagg_pflags[i].flag; i++) {
1659 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1660 		    status, lagg_pflags[i].func);
1661 		if (error)
1662 			return (error);
1663 	}
1664 	return (0);
1665 }
1666 
1667 static int
1668 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1669 {
1670 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1671 	int error, len, mcast;
1672 	struct rm_priotracker tracker;
1673 
1674 	len = m->m_pkthdr.len;
1675 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1676 
1677 	LAGG_RLOCK(sc, &tracker);
1678 	/* We need a Tx algorithm and at least one port */
1679 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1680 		LAGG_RUNLOCK(sc, &tracker);
1681 		m_freem(m);
1682 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1683 		return (ENXIO);
1684 	}
1685 
1686 	ETHER_BPF_MTAP(ifp, m);
1687 
1688 	error = lagg_proto_start(sc, m);
1689 	LAGG_RUNLOCK(sc, &tracker);
1690 
1691 	if (error != 0)
1692 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1693 
1694 	return (error);
1695 }
1696 
1697 /*
1698  * The ifp->if_qflush entry point for lagg(4) is no-op.
1699  */
1700 static void
1701 lagg_qflush(struct ifnet *ifp __unused)
1702 {
1703 }
1704 
1705 static struct mbuf *
1706 lagg_input(struct ifnet *ifp, struct mbuf *m)
1707 {
1708 	struct lagg_port *lp = ifp->if_lagg;
1709 	struct lagg_softc *sc = lp->lp_softc;
1710 	struct ifnet *scifp = sc->sc_ifp;
1711 	struct rm_priotracker tracker;
1712 
1713 	LAGG_RLOCK(sc, &tracker);
1714 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1715 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1716 	    sc->sc_proto == LAGG_PROTO_NONE) {
1717 		LAGG_RUNLOCK(sc, &tracker);
1718 		m_freem(m);
1719 		return (NULL);
1720 	}
1721 
1722 	ETHER_BPF_MTAP(scifp, m);
1723 
1724 	if (lp->lp_detaching != 0) {
1725 		m_freem(m);
1726 		m = NULL;
1727 	} else
1728 		m = lagg_proto_input(sc, lp, m);
1729 
1730 	if (m != NULL) {
1731 		if (scifp->if_flags & IFF_MONITOR) {
1732 			m_freem(m);
1733 			m = NULL;
1734 		}
1735 	}
1736 
1737 	LAGG_RUNLOCK(sc, &tracker);
1738 	return (m);
1739 }
1740 
1741 static int
1742 lagg_media_change(struct ifnet *ifp)
1743 {
1744 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1745 
1746 	if (sc->sc_ifflags & IFF_DEBUG)
1747 		printf("%s\n", __func__);
1748 
1749 	/* Ignore */
1750 	return (0);
1751 }
1752 
1753 static void
1754 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1755 {
1756 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1757 	struct lagg_port *lp;
1758 	struct rm_priotracker tracker;
1759 
1760 	imr->ifm_status = IFM_AVALID;
1761 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1762 
1763 	LAGG_RLOCK(sc, &tracker);
1764 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1765 		if (LAGG_PORTACTIVE(lp))
1766 			imr->ifm_status |= IFM_ACTIVE;
1767 	}
1768 	LAGG_RUNLOCK(sc, &tracker);
1769 }
1770 
1771 static void
1772 lagg_linkstate(struct lagg_softc *sc)
1773 {
1774 	struct lagg_port *lp;
1775 	int new_link = LINK_STATE_DOWN;
1776 	uint64_t speed;
1777 
1778 	/* Our link is considered up if at least one of our ports is active */
1779 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1780 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1781 			new_link = LINK_STATE_UP;
1782 			break;
1783 		}
1784 	}
1785 	if_link_state_change(sc->sc_ifp, new_link);
1786 
1787 	/* Update if_baudrate to reflect the max possible speed */
1788 	switch (sc->sc_proto) {
1789 		case LAGG_PROTO_FAILOVER:
1790 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1791 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1792 			break;
1793 		case LAGG_PROTO_ROUNDROBIN:
1794 		case LAGG_PROTO_LOADBALANCE:
1795 		case LAGG_PROTO_BROADCAST:
1796 			speed = 0;
1797 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1798 				speed += lp->lp_ifp->if_baudrate;
1799 			sc->sc_ifp->if_baudrate = speed;
1800 			break;
1801 		case LAGG_PROTO_LACP:
1802 			/* LACP updates if_baudrate itself */
1803 			break;
1804 	}
1805 }
1806 
1807 static void
1808 lagg_port_state(struct ifnet *ifp, int state)
1809 {
1810 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1811 	struct lagg_softc *sc = NULL;
1812 
1813 	if (lp != NULL)
1814 		sc = lp->lp_softc;
1815 	if (sc == NULL)
1816 		return;
1817 
1818 	LAGG_WLOCK(sc);
1819 	lagg_linkstate(sc);
1820 	lagg_proto_linkstate(sc, lp);
1821 	LAGG_WUNLOCK(sc);
1822 }
1823 
1824 struct lagg_port *
1825 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1826 {
1827 	struct lagg_port *lp_next, *rval = NULL;
1828 	// int new_link = LINK_STATE_DOWN;
1829 
1830 	LAGG_RLOCK_ASSERT(sc);
1831 	/*
1832 	 * Search a port which reports an active link state.
1833 	 */
1834 
1835 	if (lp == NULL)
1836 		goto search;
1837 	if (LAGG_PORTACTIVE(lp)) {
1838 		rval = lp;
1839 		goto found;
1840 	}
1841 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1842 	    LAGG_PORTACTIVE(lp_next)) {
1843 		rval = lp_next;
1844 		goto found;
1845 	}
1846 
1847 search:
1848 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1849 		if (LAGG_PORTACTIVE(lp_next)) {
1850 			rval = lp_next;
1851 			goto found;
1852 		}
1853 	}
1854 
1855 found:
1856 	if (rval != NULL) {
1857 		/*
1858 		 * The IEEE 802.1D standard assumes that a lagg with
1859 		 * multiple ports is always full duplex. This is valid
1860 		 * for load sharing laggs and if at least two links
1861 		 * are active. Unfortunately, checking the latter would
1862 		 * be too expensive at this point.
1863 		 XXX
1864 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1865 		    (sc->sc_count > 1))
1866 			new_link = LINK_STATE_FULL_DUPLEX;
1867 		else
1868 			new_link = rval->lp_link_state;
1869 		 */
1870 	}
1871 
1872 	return (rval);
1873 }
1874 
1875 int
1876 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1877 {
1878 
1879 	return (ifp->if_transmit)(ifp, m);
1880 }
1881 
1882 /*
1883  * Simple round robin aggregation
1884  */
1885 static void
1886 lagg_rr_attach(struct lagg_softc *sc)
1887 {
1888 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1889 	sc->sc_seq = 0;
1890 	sc->sc_bkt_count = sc->sc_bkt;
1891 }
1892 
1893 static int
1894 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1895 {
1896 	struct lagg_port *lp;
1897 	uint32_t p;
1898 
1899 	if (sc->sc_bkt_count == 0 && sc->sc_bkt > 0)
1900 		sc->sc_bkt_count = sc->sc_bkt;
1901 
1902 	if (sc->sc_bkt > 0) {
1903 		atomic_subtract_int(&sc->sc_bkt_count, 1);
1904 	if (atomic_cmpset_int(&sc->sc_bkt_count, 0, sc->sc_bkt))
1905 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1906 	else
1907 		p = sc->sc_seq;
1908 	} else
1909 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1910 
1911 	p %= sc->sc_count;
1912 	lp = SLIST_FIRST(&sc->sc_ports);
1913 
1914 	while (p--)
1915 		lp = SLIST_NEXT(lp, lp_entries);
1916 
1917 	/*
1918 	 * Check the port's link state. This will return the next active
1919 	 * port if the link is down or the port is NULL.
1920 	 */
1921 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1922 		m_freem(m);
1923 		return (ENETDOWN);
1924 	}
1925 
1926 	/* Send mbuf */
1927 	return (lagg_enqueue(lp->lp_ifp, m));
1928 }
1929 
1930 static struct mbuf *
1931 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1932 {
1933 	struct ifnet *ifp = sc->sc_ifp;
1934 
1935 	/* Just pass in the packet to our lagg device */
1936 	m->m_pkthdr.rcvif = ifp;
1937 
1938 	return (m);
1939 }
1940 
1941 /*
1942  * Broadcast mode
1943  */
1944 static int
1945 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1946 {
1947 	int active_ports = 0;
1948 	int errors = 0;
1949 	int ret;
1950 	struct lagg_port *lp, *last = NULL;
1951 	struct mbuf *m0;
1952 
1953 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1954 		if (!LAGG_PORTACTIVE(lp))
1955 			continue;
1956 
1957 		active_ports++;
1958 
1959 		if (last != NULL) {
1960 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
1961 			if (m0 == NULL) {
1962 				ret = ENOBUFS;
1963 				errors++;
1964 				break;
1965 			}
1966 
1967 			ret = lagg_enqueue(last->lp_ifp, m0);
1968 			if (ret != 0)
1969 				errors++;
1970 		}
1971 		last = lp;
1972 	}
1973 	if (last == NULL) {
1974 		m_freem(m);
1975 		return (ENOENT);
1976 	}
1977 	if ((last = lagg_link_active(sc, last)) == NULL) {
1978 		m_freem(m);
1979 		return (ENETDOWN);
1980 	}
1981 
1982 	ret = lagg_enqueue(last->lp_ifp, m);
1983 	if (ret != 0)
1984 		errors++;
1985 
1986 	if (errors == 0)
1987 		return (ret);
1988 
1989 	return (0);
1990 }
1991 
1992 static struct mbuf*
1993 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1994 {
1995 	struct ifnet *ifp = sc->sc_ifp;
1996 
1997 	/* Just pass in the packet to our lagg device */
1998 	m->m_pkthdr.rcvif = ifp;
1999 	return (m);
2000 }
2001 
2002 /*
2003  * Active failover
2004  */
2005 static int
2006 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2007 {
2008 	struct lagg_port *lp;
2009 
2010 	/* Use the master port if active or the next available port */
2011 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2012 		m_freem(m);
2013 		return (ENETDOWN);
2014 	}
2015 
2016 	/* Send mbuf */
2017 	return (lagg_enqueue(lp->lp_ifp, m));
2018 }
2019 
2020 static struct mbuf *
2021 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2022 {
2023 	struct ifnet *ifp = sc->sc_ifp;
2024 	struct lagg_port *tmp_tp;
2025 
2026 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2027 		m->m_pkthdr.rcvif = ifp;
2028 		return (m);
2029 	}
2030 
2031 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2032 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2033 		/*
2034 		 * If tmp_tp is null, we've recieved a packet when all
2035 		 * our links are down. Weird, but process it anyways.
2036 		 */
2037 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2038 			m->m_pkthdr.rcvif = ifp;
2039 			return (m);
2040 		}
2041 	}
2042 
2043 	m_freem(m);
2044 	return (NULL);
2045 }
2046 
2047 /*
2048  * Loadbalancing
2049  */
2050 static void
2051 lagg_lb_attach(struct lagg_softc *sc)
2052 {
2053 	struct lagg_port *lp;
2054 	struct lagg_lb *lb;
2055 
2056 	lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
2057 
2058 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
2059 
2060 	lb->lb_key = m_ether_tcpip_hash_init();
2061 	sc->sc_psc = lb;
2062 
2063 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2064 		lagg_lb_port_create(lp);
2065 }
2066 
2067 static void
2068 lagg_lb_detach(struct lagg_softc *sc)
2069 {
2070 	struct lagg_lb *lb;
2071 
2072 	lb = (struct lagg_lb *)sc->sc_psc;
2073 	LAGG_WUNLOCK(sc);
2074 	if (lb != NULL)
2075 		free(lb, M_DEVBUF);
2076 }
2077 
2078 static int
2079 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2080 {
2081 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2082 	struct lagg_port *lp_next;
2083 	int i = 0;
2084 
2085 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2086 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2087 		if (lp_next == lp)
2088 			continue;
2089 		if (i >= LAGG_MAX_PORTS)
2090 			return (EINVAL);
2091 		if (sc->sc_ifflags & IFF_DEBUG)
2092 			printf("%s: port %s at index %d\n",
2093 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2094 		lb->lb_ports[i++] = lp_next;
2095 	}
2096 
2097 	return (0);
2098 }
2099 
2100 static int
2101 lagg_lb_port_create(struct lagg_port *lp)
2102 {
2103 	struct lagg_softc *sc = lp->lp_softc;
2104 	return (lagg_lb_porttable(sc, NULL));
2105 }
2106 
2107 static void
2108 lagg_lb_port_destroy(struct lagg_port *lp)
2109 {
2110 	struct lagg_softc *sc = lp->lp_softc;
2111 	lagg_lb_porttable(sc, lp);
2112 }
2113 
2114 static int
2115 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2116 {
2117 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2118 	struct lagg_port *lp = NULL;
2119 	uint32_t p = 0;
2120 
2121 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2122 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2123 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2124 	else
2125 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2126 	p %= sc->sc_count;
2127 	lp = lb->lb_ports[p];
2128 
2129 	/*
2130 	 * Check the port's link state. This will return the next active
2131 	 * port if the link is down or the port is NULL.
2132 	 */
2133 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2134 		m_freem(m);
2135 		return (ENETDOWN);
2136 	}
2137 
2138 	/* Send mbuf */
2139 	return (lagg_enqueue(lp->lp_ifp, m));
2140 }
2141 
2142 static struct mbuf *
2143 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2144 {
2145 	struct ifnet *ifp = sc->sc_ifp;
2146 
2147 	/* Just pass in the packet to our lagg device */
2148 	m->m_pkthdr.rcvif = ifp;
2149 
2150 	return (m);
2151 }
2152 
2153 /*
2154  * 802.3ad LACP
2155  */
2156 static void
2157 lagg_lacp_attach(struct lagg_softc *sc)
2158 {
2159 	struct lagg_port *lp;
2160 
2161 	lacp_attach(sc);
2162 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2163 		lacp_port_create(lp);
2164 }
2165 
2166 static void
2167 lagg_lacp_detach(struct lagg_softc *sc)
2168 {
2169 	struct lagg_port *lp;
2170 	void *psc;
2171 
2172 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2173 		lacp_port_destroy(lp);
2174 
2175 	psc = sc->sc_psc;
2176 	sc->sc_psc = NULL;
2177 	LAGG_WUNLOCK(sc);
2178 
2179 	lacp_detach(psc);
2180 }
2181 
2182 static void
2183 lagg_lacp_lladdr(struct lagg_softc *sc)
2184 {
2185 	struct lagg_port *lp;
2186 
2187 	/* purge all the lacp ports */
2188 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2189 		lacp_port_destroy(lp);
2190 
2191 	/* add them back in */
2192 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2193 		lacp_port_create(lp);
2194 }
2195 
2196 static int
2197 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2198 {
2199 	struct lagg_port *lp;
2200 
2201 	lp = lacp_select_tx_port(sc, m);
2202 	if (lp == NULL) {
2203 		m_freem(m);
2204 		return (ENETDOWN);
2205 	}
2206 
2207 	/* Send mbuf */
2208 	return (lagg_enqueue(lp->lp_ifp, m));
2209 }
2210 
2211 static struct mbuf *
2212 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2213 {
2214 	struct ifnet *ifp = sc->sc_ifp;
2215 	struct ether_header *eh;
2216 	u_short etype;
2217 
2218 	eh = mtod(m, struct ether_header *);
2219 	etype = ntohs(eh->ether_type);
2220 
2221 	/* Tap off LACP control messages */
2222 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2223 		m = lacp_input(lp, m);
2224 		if (m == NULL)
2225 			return (NULL);
2226 	}
2227 
2228 	/*
2229 	 * If the port is not collecting or not in the active aggregator then
2230 	 * free and return.
2231 	 */
2232 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2233 		m_freem(m);
2234 		return (NULL);
2235 	}
2236 
2237 	m->m_pkthdr.rcvif = ifp;
2238 	return (m);
2239 }
2240 
2241