xref: /freebsd/sys/net/if_lagg.c (revision bc96366c864c07ef352edb92017357917c75b36c)
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 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/hash.h>
40 #include <sys/lock.h>
41 #include <sys/rmlock.h>
42 #include <sys/taskqueue.h>
43 #include <sys/eventhandler.h>
44 
45 #include <net/ethernet.h>
46 #include <net/if.h>
47 #include <net/if_clone.h>
48 #include <net/if_arp.h>
49 #include <net/if_dl.h>
50 #include <net/if_media.h>
51 #include <net/if_types.h>
52 #include <net/if_var.h>
53 #include <net/bpf.h>
54 #include <net/vnet.h>
55 
56 #if defined(INET) || defined(INET6)
57 #include <netinet/in.h>
58 #include <netinet/ip.h>
59 #endif
60 #ifdef INET
61 #include <netinet/in_systm.h>
62 #include <netinet/if_ether.h>
63 #endif
64 
65 #ifdef INET6
66 #include <netinet/ip6.h>
67 #include <netinet6/in6_var.h>
68 #include <netinet6/in6_ifattach.h>
69 #endif
70 
71 #include <net/if_vlan_var.h>
72 #include <net/if_lagg.h>
73 #include <net/ieee8023ad_lacp.h>
74 
75 /* Special flags we should propagate to the lagg ports. */
76 static struct {
77 	int flag;
78 	int (*func)(struct ifnet *, int);
79 } lagg_pflags[] = {
80 	{IFF_PROMISC, ifpromisc},
81 	{IFF_ALLMULTI, if_allmulti},
82 	{0, NULL}
83 };
84 
85 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
86 #define	V_lagg_list	VNET(lagg_list)
87 static VNET_DEFINE(struct mtx, lagg_list_mtx);
88 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
89 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
90 					"if_lagg list", NULL, MTX_DEF)
91 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
92 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
93 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
94 eventhandler_tag	lagg_detach_cookie = NULL;
95 
96 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
97 static void	lagg_clone_destroy(struct ifnet *);
98 static VNET_DEFINE(struct if_clone *, lagg_cloner);
99 #define	V_lagg_cloner	VNET(lagg_cloner)
100 static const char laggname[] = "lagg";
101 
102 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
103 static void	lagg_capabilities(struct lagg_softc *);
104 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
105 static void	lagg_port_setlladdr(void *, int);
106 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
107 static int	lagg_port_destroy(struct lagg_port *, int);
108 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
109 static void	lagg_linkstate(struct lagg_softc *);
110 static void	lagg_port_state(struct ifnet *, int);
111 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
112 static int	lagg_port_output(struct ifnet *, struct mbuf *,
113 		    const struct sockaddr *, struct route *);
114 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
115 #ifdef LAGG_PORT_STACKING
116 static int	lagg_port_checkstacking(struct lagg_softc *);
117 #endif
118 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
119 static void	lagg_init(void *);
120 static void	lagg_stop(struct lagg_softc *);
121 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
122 static int	lagg_ether_setmulti(struct lagg_softc *);
123 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
124 static	int	lagg_setflag(struct lagg_port *, int, int,
125 		    int (*func)(struct ifnet *, int));
126 static	int	lagg_setflags(struct lagg_port *, int status);
127 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
128 static int	lagg_transmit(struct ifnet *, struct mbuf *);
129 static void	lagg_qflush(struct ifnet *);
130 static int	lagg_media_change(struct ifnet *);
131 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
132 static struct lagg_port *lagg_link_active(struct lagg_softc *,
133 	    struct lagg_port *);
134 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
135 
136 /* Simple round robin */
137 static void	lagg_rr_attach(struct lagg_softc *);
138 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
139 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
140 		    struct mbuf *);
141 
142 /* Active failover */
143 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
144 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
145 		    struct mbuf *);
146 
147 /* Loadbalancing */
148 static void	lagg_lb_attach(struct lagg_softc *);
149 static void	lagg_lb_detach(struct lagg_softc *);
150 static int	lagg_lb_port_create(struct lagg_port *);
151 static void	lagg_lb_port_destroy(struct lagg_port *);
152 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
153 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
154 		    struct mbuf *);
155 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
156 
157 /* Broadcast */
158 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
159 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
160 		    struct mbuf *);
161 
162 /* 802.3ad LACP */
163 static void	lagg_lacp_attach(struct lagg_softc *);
164 static void	lagg_lacp_detach(struct lagg_softc *);
165 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
166 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
167 		    struct mbuf *);
168 static void	lagg_lacp_lladdr(struct lagg_softc *);
169 
170 /* lagg protocol table */
171 static const struct lagg_proto {
172 	lagg_proto	pr_num;
173 	void		(*pr_attach)(struct lagg_softc *);
174 	void		(*pr_detach)(struct lagg_softc *);
175 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
176 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
177 			    struct mbuf *);
178 	int		(*pr_addport)(struct lagg_port *);
179 	void		(*pr_delport)(struct lagg_port *);
180 	void		(*pr_linkstate)(struct lagg_port *);
181 	void 		(*pr_init)(struct lagg_softc *);
182 	void 		(*pr_stop)(struct lagg_softc *);
183 	void 		(*pr_lladdr)(struct lagg_softc *);
184 	void		(*pr_request)(struct lagg_softc *, void *);
185 	void		(*pr_portreq)(struct lagg_port *, void *);
186 } lagg_protos[] = {
187     {
188 	.pr_num = LAGG_PROTO_NONE
189     },
190     {
191 	.pr_num = LAGG_PROTO_ROUNDROBIN,
192 	.pr_attach = lagg_rr_attach,
193 	.pr_start = lagg_rr_start,
194 	.pr_input = lagg_rr_input,
195     },
196     {
197 	.pr_num = LAGG_PROTO_FAILOVER,
198 	.pr_start = lagg_fail_start,
199 	.pr_input = lagg_fail_input,
200     },
201     {
202 	.pr_num = LAGG_PROTO_LOADBALANCE,
203 	.pr_attach = lagg_lb_attach,
204 	.pr_detach = lagg_lb_detach,
205 	.pr_start = lagg_lb_start,
206 	.pr_input = lagg_lb_input,
207 	.pr_addport = lagg_lb_port_create,
208 	.pr_delport = lagg_lb_port_destroy,
209     },
210     {
211 	.pr_num = LAGG_PROTO_LACP,
212 	.pr_attach = lagg_lacp_attach,
213 	.pr_detach = lagg_lacp_detach,
214 	.pr_start = lagg_lacp_start,
215 	.pr_input = lagg_lacp_input,
216 	.pr_addport = lacp_port_create,
217 	.pr_delport = lacp_port_destroy,
218 	.pr_linkstate = lacp_linkstate,
219 	.pr_init = lacp_init,
220 	.pr_stop = lacp_stop,
221 	.pr_lladdr = lagg_lacp_lladdr,
222 	.pr_request = lacp_req,
223 	.pr_portreq = lacp_portreq,
224     },
225     {
226 	.pr_num = LAGG_PROTO_ETHERCHANNEL,
227 	.pr_attach = lagg_lb_attach,
228 	.pr_detach = lagg_lb_detach,
229 	.pr_start = lagg_lb_start,
230 	.pr_input = lagg_lb_input,
231     },
232     {
233 	.pr_num = LAGG_PROTO_BROADCAST,
234 	.pr_start = lagg_bcast_start,
235 	.pr_input = lagg_bcast_input,
236     },
237 };
238 
239 SYSCTL_DECL(_net_link);
240 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
241     "Link Aggregation");
242 
243 /* Allow input on any failover links */
244 static VNET_DEFINE(int, lagg_failover_rx_all);
245 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
246 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
247     &VNET_NAME(lagg_failover_rx_all), 0,
248     "Accept input from any interface in a failover lagg");
249 
250 /* Default value for using flowid */
251 static VNET_DEFINE(int, def_use_flowid) = 1;
252 #define	V_def_use_flowid	VNET(def_use_flowid)
253 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
254     &VNET_NAME(def_use_flowid), 0,
255     "Default setting for using flow id for load sharing");
256 
257 /* Default value for flowid shift */
258 static VNET_DEFINE(int, def_flowid_shift) = 16;
259 #define	V_def_flowid_shift	VNET(def_flowid_shift)
260 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
261     &VNET_NAME(def_flowid_shift), 0,
262     "Default setting for flowid shift for load sharing");
263 
264 static void
265 vnet_lagg_init(const void *unused __unused)
266 {
267 
268 	LAGG_LIST_LOCK_INIT();
269 	SLIST_INIT(&V_lagg_list);
270 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
271 	    lagg_clone_destroy, 0);
272 }
273 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
274     vnet_lagg_init, NULL);
275 
276 static void
277 vnet_lagg_uninit(const void *unused __unused)
278 {
279 
280 	if_clone_detach(V_lagg_cloner);
281 	LAGG_LIST_LOCK_DESTROY();
282 }
283 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
284     vnet_lagg_uninit, NULL);
285 
286 static int
287 lagg_modevent(module_t mod, int type, void *data)
288 {
289 
290 	switch (type) {
291 	case MOD_LOAD:
292 		lagg_input_p = lagg_input;
293 		lagg_linkstate_p = lagg_port_state;
294 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
295 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
296 		    EVENTHANDLER_PRI_ANY);
297 		break;
298 	case MOD_UNLOAD:
299 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
300 		    lagg_detach_cookie);
301 		lagg_input_p = NULL;
302 		lagg_linkstate_p = NULL;
303 		break;
304 	default:
305 		return (EOPNOTSUPP);
306 	}
307 	return (0);
308 }
309 
310 static moduledata_t lagg_mod = {
311 	"if_lagg",
312 	lagg_modevent,
313 	0
314 };
315 
316 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
317 MODULE_VERSION(if_lagg, 1);
318 
319 static void
320 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
321 {
322 
323 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
324 	    __func__, sc));
325 
326 	if (sc->sc_ifflags & IFF_DEBUG)
327 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
328 
329 	if (lagg_protos[pr].pr_attach != NULL)
330 		lagg_protos[pr].pr_attach(sc);
331 	sc->sc_proto = pr;
332 }
333 
334 static void
335 lagg_proto_detach(struct lagg_softc *sc)
336 {
337 	lagg_proto pr;
338 
339 	LAGG_WLOCK_ASSERT(sc);
340 
341 	pr = sc->sc_proto;
342 	sc->sc_proto = LAGG_PROTO_NONE;
343 
344 	if (lagg_protos[pr].pr_detach != NULL)
345 		lagg_protos[pr].pr_detach(sc);
346 	else
347 		LAGG_WUNLOCK(sc);
348 }
349 
350 static int
351 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
352 {
353 
354 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
355 }
356 
357 static struct mbuf *
358 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
359 {
360 
361 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
362 }
363 
364 static int
365 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
366 {
367 
368 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
369 		return (0);
370 	else
371 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
372 }
373 
374 static void
375 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
376 {
377 
378 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
379 		lagg_protos[sc->sc_proto].pr_delport(lp);
380 }
381 
382 static void
383 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
384 {
385 
386 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
387 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
388 }
389 
390 static void
391 lagg_proto_init(struct lagg_softc *sc)
392 {
393 
394 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
395 		lagg_protos[sc->sc_proto].pr_init(sc);
396 }
397 
398 static void
399 lagg_proto_stop(struct lagg_softc *sc)
400 {
401 
402 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
403 		lagg_protos[sc->sc_proto].pr_stop(sc);
404 }
405 
406 static void
407 lagg_proto_lladdr(struct lagg_softc *sc)
408 {
409 
410 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
411 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
412 }
413 
414 static void
415 lagg_proto_request(struct lagg_softc *sc, void *v)
416 {
417 
418 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
419 		lagg_protos[sc->sc_proto].pr_request(sc, v);
420 }
421 
422 static void
423 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
424 {
425 
426 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
427 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
428 }
429 
430 /*
431  * This routine is run via an vlan
432  * config EVENT
433  */
434 static void
435 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
436 {
437 	struct lagg_softc *sc = ifp->if_softc;
438 	struct lagg_port *lp;
439 	struct rm_priotracker tracker;
440 
441 	if (ifp->if_softc !=  arg)   /* Not our event */
442 		return;
443 
444 	LAGG_RLOCK(sc, &tracker);
445 	if (!SLIST_EMPTY(&sc->sc_ports)) {
446 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
447 			EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
448 	}
449 	LAGG_RUNLOCK(sc, &tracker);
450 }
451 
452 /*
453  * This routine is run via an vlan
454  * unconfig EVENT
455  */
456 static void
457 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
458 {
459 	struct lagg_softc *sc = ifp->if_softc;
460 	struct lagg_port *lp;
461 	struct rm_priotracker tracker;
462 
463 	if (ifp->if_softc !=  arg)   /* Not our event */
464 		return;
465 
466 	LAGG_RLOCK(sc, &tracker);
467 	if (!SLIST_EMPTY(&sc->sc_ports)) {
468 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
469 			EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
470 	}
471 	LAGG_RUNLOCK(sc, &tracker);
472 }
473 
474 static int
475 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
476 {
477 	struct lagg_softc *sc;
478 	struct ifnet *ifp;
479 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
480 
481 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
482 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
483 	if (ifp == NULL) {
484 		free(sc, M_DEVBUF);
485 		return (ENOSPC);
486 	}
487 
488 	if (V_def_use_flowid)
489 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
490 	sc->flowid_shift = V_def_flowid_shift;
491 
492 	/* Hash all layers by default */
493 	sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4;
494 
495 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
496 
497 	LAGG_LOCK_INIT(sc);
498 	SLIST_INIT(&sc->sc_ports);
499 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
500 
501 	/* Initialise pseudo media types */
502 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
503 	    lagg_media_status);
504 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
505 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
506 
507 	if_initname(ifp, laggname, unit);
508 	ifp->if_softc = sc;
509 	ifp->if_transmit = lagg_transmit;
510 	ifp->if_qflush = lagg_qflush;
511 	ifp->if_init = lagg_init;
512 	ifp->if_ioctl = lagg_ioctl;
513 	ifp->if_get_counter = lagg_get_counter;
514 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
515 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
516 
517 	/*
518 	 * Attach as an ordinary ethernet device, children will be attached
519 	 * as special device IFT_IEEE8023ADLAG.
520 	 */
521 	ether_ifattach(ifp, eaddr);
522 
523 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
524 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
525 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
526 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
527 
528 	/* Insert into the global list of laggs */
529 	LAGG_LIST_LOCK();
530 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
531 	LAGG_LIST_UNLOCK();
532 
533 	return (0);
534 }
535 
536 static void
537 lagg_clone_destroy(struct ifnet *ifp)
538 {
539 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
540 	struct lagg_port *lp;
541 
542 	LAGG_WLOCK(sc);
543 
544 	lagg_stop(sc);
545 	ifp->if_flags &= ~IFF_UP;
546 
547 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
548 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
549 
550 	/* Shutdown and remove lagg ports */
551 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
552 		lagg_port_destroy(lp, 1);
553 	/* Unhook the aggregation protocol */
554 	lagg_proto_detach(sc);
555 
556 	ifmedia_removeall(&sc->sc_media);
557 	ether_ifdetach(ifp);
558 	if_free(ifp);
559 
560 	LAGG_LIST_LOCK();
561 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
562 	LAGG_LIST_UNLOCK();
563 
564 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
565 	LAGG_LOCK_DESTROY(sc);
566 	free(sc, M_DEVBUF);
567 }
568 
569 static void
570 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
571 {
572 	struct ifnet *ifp = sc->sc_ifp;
573 	struct lagg_port lp;
574 
575 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
576 		return;
577 
578 	LAGG_WLOCK_ASSERT(sc);
579 	/*
580 	 * Set the link layer address on the lagg interface.
581 	 * lagg_proto_lladdr() notifies the MAC change to
582 	 * the aggregation protocol.  iflladdr_event handler which
583 	 * may trigger gratuitous ARPs for INET will be handled in
584 	 * a taskqueue.
585 	 */
586 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
587 	lagg_proto_lladdr(sc);
588 
589 	bzero(&lp, sizeof(lp));
590 	lp.lp_ifp = sc->sc_ifp;
591 	lp.lp_softc = sc;
592 
593 	lagg_port_lladdr(&lp, lladdr);
594 }
595 
596 static void
597 lagg_capabilities(struct lagg_softc *sc)
598 {
599 	struct lagg_port *lp;
600 	int cap = ~0, ena = ~0;
601 	u_long hwa = ~0UL;
602 	struct ifnet_hw_tsomax hw_tsomax;
603 
604 	LAGG_WLOCK_ASSERT(sc);
605 
606 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
607 
608 	/* Get capabilities from the lagg ports */
609 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
610 		cap &= lp->lp_ifp->if_capabilities;
611 		ena &= lp->lp_ifp->if_capenable;
612 		hwa &= lp->lp_ifp->if_hwassist;
613 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
614 	}
615 	cap = (cap == ~0 ? 0 : cap);
616 	ena = (ena == ~0 ? 0 : ena);
617 	hwa = (hwa == ~0 ? 0 : hwa);
618 
619 	if (sc->sc_ifp->if_capabilities != cap ||
620 	    sc->sc_ifp->if_capenable != ena ||
621 	    sc->sc_ifp->if_hwassist != hwa ||
622 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
623 		sc->sc_ifp->if_capabilities = cap;
624 		sc->sc_ifp->if_capenable = ena;
625 		sc->sc_ifp->if_hwassist = hwa;
626 		getmicrotime(&sc->sc_ifp->if_lastchange);
627 
628 		if (sc->sc_ifflags & IFF_DEBUG)
629 			if_printf(sc->sc_ifp,
630 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
631 	}
632 }
633 
634 static void
635 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
636 {
637 	struct lagg_softc *sc = lp->lp_softc;
638 	struct ifnet *ifp = lp->lp_ifp;
639 	struct lagg_llq *llq;
640 	int pending = 0;
641 	int primary;
642 
643 	LAGG_WLOCK_ASSERT(sc);
644 
645 	primary = (sc->sc_primary->lp_ifp == ifp) ? 1 : 0;
646 	if (primary == 0 && (lp->lp_detaching ||
647 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0))
648 		return;
649 
650 	/* Check to make sure its not already queued to be changed */
651 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
652 		if (llq->llq_ifp == ifp) {
653 			pending = 1;
654 			break;
655 		}
656 	}
657 
658 	if (!pending) {
659 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
660 		if (llq == NULL)	/* XXX what to do */
661 			return;
662 	}
663 
664 	/* Update the lladdr even if pending, it may have changed */
665 	llq->llq_ifp = ifp;
666 	llq->llq_primary = primary;
667 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
668 
669 	if (!pending)
670 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
671 
672 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
673 }
674 
675 /*
676  * Set the interface MAC address from a taskqueue to avoid a LOR.
677  */
678 static void
679 lagg_port_setlladdr(void *arg, int pending)
680 {
681 	struct lagg_softc *sc = (struct lagg_softc *)arg;
682 	struct lagg_llq *llq, *head;
683 	struct ifnet *ifp;
684 	int error;
685 
686 	/* Grab a local reference of the queue and remove it from the softc */
687 	LAGG_WLOCK(sc);
688 	head = SLIST_FIRST(&sc->sc_llq_head);
689 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
690 	LAGG_WUNLOCK(sc);
691 
692 	/*
693 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
694 	 * unlocked as we have the only reference to it.
695 	 */
696 	for (llq = head; llq != NULL; llq = head) {
697 		ifp = llq->llq_ifp;
698 
699 		CURVNET_SET(ifp->if_vnet);
700 		if (llq->llq_primary == 0) {
701 			/*
702 			 * Set the link layer address on the laggport interface.
703 			 * if_setlladdr() triggers gratuitous ARPs for INET.
704 			 */
705 			error = if_setlladdr(ifp, llq->llq_lladdr,
706 			    ETHER_ADDR_LEN);
707 			if (error)
708 				printf("%s: setlladdr failed on %s\n", __func__,
709 				    ifp->if_xname);
710 		} else
711 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
712 		CURVNET_RESTORE();
713 		head = SLIST_NEXT(llq, llq_entries);
714 		free(llq, M_DEVBUF);
715 	}
716 }
717 
718 static int
719 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
720 {
721 	struct lagg_softc *sc_ptr;
722 	struct lagg_port *lp, *tlp;
723 	int error, i;
724 	uint64_t *pval;
725 
726 	LAGG_WLOCK_ASSERT(sc);
727 
728 	/* Limit the maximal number of lagg ports */
729 	if (sc->sc_count >= LAGG_MAX_PORTS)
730 		return (ENOSPC);
731 
732 	/* Check if port has already been associated to a lagg */
733 	if (ifp->if_lagg != NULL) {
734 		/* Port is already in the current lagg? */
735 		lp = (struct lagg_port *)ifp->if_lagg;
736 		if (lp->lp_softc == sc)
737 			return (EEXIST);
738 		return (EBUSY);
739 	}
740 
741 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
742 	if (ifp->if_type != IFT_ETHER)
743 		return (EPROTONOSUPPORT);
744 
745 	/* Allow the first Ethernet member to define the MTU */
746 	if (SLIST_EMPTY(&sc->sc_ports))
747 		sc->sc_ifp->if_mtu = ifp->if_mtu;
748 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
749 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
750 		    ifp->if_xname);
751 		return (EINVAL);
752 	}
753 
754 	if ((lp = malloc(sizeof(struct lagg_port),
755 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
756 		return (ENOMEM);
757 
758 	/* Check if port is a stacked lagg */
759 	LAGG_LIST_LOCK();
760 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
761 		if (ifp == sc_ptr->sc_ifp) {
762 			LAGG_LIST_UNLOCK();
763 			free(lp, M_DEVBUF);
764 			return (EINVAL);
765 			/* XXX disable stacking for the moment, its untested */
766 #ifdef LAGG_PORT_STACKING
767 			lp->lp_flags |= LAGG_PORT_STACK;
768 			if (lagg_port_checkstacking(sc_ptr) >=
769 			    LAGG_MAX_STACKING) {
770 				LAGG_LIST_UNLOCK();
771 				free(lp, M_DEVBUF);
772 				return (E2BIG);
773 			}
774 #endif
775 		}
776 	}
777 	LAGG_LIST_UNLOCK();
778 
779 	/* Change the interface type */
780 	lp->lp_iftype = ifp->if_type;
781 	ifp->if_type = IFT_IEEE8023ADLAG;
782 	ifp->if_lagg = lp;
783 	lp->lp_ioctl = ifp->if_ioctl;
784 	ifp->if_ioctl = lagg_port_ioctl;
785 	lp->lp_output = ifp->if_output;
786 	ifp->if_output = lagg_port_output;
787 
788 	lp->lp_ifp = ifp;
789 	lp->lp_softc = sc;
790 
791 	/* Save port link layer address */
792 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
793 
794 	if (SLIST_EMPTY(&sc->sc_ports)) {
795 		sc->sc_primary = lp;
796 		lagg_lladdr(sc, IF_LLADDR(ifp));
797 	} else {
798 		/* Update link layer address for this port */
799 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
800 	}
801 
802 	/*
803 	 * Insert into the list of ports.
804 	 * Keep ports sorted by if_index. It is handy, when configuration
805 	 * is predictable and `ifconfig laggN create ...` command
806 	 * will lead to the same result each time.
807 	 */
808 	SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
809 		if (tlp->lp_ifp->if_index < ifp->if_index && (
810 		    SLIST_NEXT(tlp, lp_entries) == NULL ||
811 		    SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index >
812 		    ifp->if_index))
813 			break;
814 	}
815 	if (tlp != NULL)
816 		SLIST_INSERT_AFTER(tlp, lp, lp_entries);
817 	else
818 		SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
819 	sc->sc_count++;
820 
821 	/* Update lagg capabilities */
822 	lagg_capabilities(sc);
823 	lagg_linkstate(sc);
824 
825 	/* Read port counters */
826 	pval = lp->port_counters.val;
827 	for (i = 0; i < IFCOUNTERS; i++, pval++)
828 		*pval = ifp->if_get_counter(ifp, i);
829 	/* Add multicast addresses and interface flags to this port */
830 	lagg_ether_cmdmulti(lp, 1);
831 	lagg_setflags(lp, 1);
832 
833 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
834 		/* Remove the port, without calling pr_delport. */
835 		lagg_port_destroy(lp, 0);
836 		return (error);
837 	}
838 
839 	return (0);
840 }
841 
842 #ifdef LAGG_PORT_STACKING
843 static int
844 lagg_port_checkstacking(struct lagg_softc *sc)
845 {
846 	struct lagg_softc *sc_ptr;
847 	struct lagg_port *lp;
848 	int m = 0;
849 
850 	LAGG_WLOCK_ASSERT(sc);
851 
852 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
853 		if (lp->lp_flags & LAGG_PORT_STACK) {
854 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
855 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
856 		}
857 	}
858 
859 	return (m + 1);
860 }
861 #endif
862 
863 static int
864 lagg_port_destroy(struct lagg_port *lp, int rundelport)
865 {
866 	struct lagg_softc *sc = lp->lp_softc;
867 	struct lagg_port *lp_ptr;
868 	struct lagg_llq *llq;
869 	struct ifnet *ifp = lp->lp_ifp;
870 	uint64_t *pval, vdiff;
871 	int i;
872 
873 	LAGG_WLOCK_ASSERT(sc);
874 
875 	if (rundelport)
876 		lagg_proto_delport(sc, lp);
877 
878 	/*
879 	 * Remove multicast addresses and interface flags from this port and
880 	 * reset the MAC address, skip if the interface is being detached.
881 	 */
882 	if (!lp->lp_detaching) {
883 		lagg_ether_cmdmulti(lp, 0);
884 		lagg_setflags(lp, 0);
885 		lagg_port_lladdr(lp, lp->lp_lladdr);
886 	}
887 
888 	/* Restore interface */
889 	ifp->if_type = lp->lp_iftype;
890 	ifp->if_ioctl = lp->lp_ioctl;
891 	ifp->if_output = lp->lp_output;
892 	ifp->if_lagg = NULL;
893 
894 	/* Update detached port counters */
895 	pval = lp->port_counters.val;
896 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
897 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
898 		sc->detached_counters.val[i] += vdiff;
899 	}
900 
901 	/* Finally, remove the port from the lagg */
902 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
903 	sc->sc_count--;
904 
905 	/* Update the primary interface */
906 	if (lp == sc->sc_primary) {
907 		uint8_t lladdr[ETHER_ADDR_LEN];
908 
909 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
910 			bzero(&lladdr, ETHER_ADDR_LEN);
911 		} else {
912 			bcopy(lp_ptr->lp_lladdr,
913 			    lladdr, ETHER_ADDR_LEN);
914 		}
915 		lagg_lladdr(sc, lladdr);
916 		sc->sc_primary = lp_ptr;
917 
918 		/* Update link layer address for each port */
919 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
920 			lagg_port_lladdr(lp_ptr, lladdr);
921 	}
922 
923 	/* Remove any pending lladdr changes from the queue */
924 	if (lp->lp_detaching) {
925 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
926 			if (llq->llq_ifp == ifp) {
927 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
928 				    llq_entries);
929 				free(llq, M_DEVBUF);
930 				break;	/* Only appears once */
931 			}
932 		}
933 	}
934 
935 	if (lp->lp_ifflags)
936 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
937 
938 	free(lp, M_DEVBUF);
939 
940 	/* Update lagg capabilities */
941 	lagg_capabilities(sc);
942 	lagg_linkstate(sc);
943 
944 	return (0);
945 }
946 
947 static int
948 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
949 {
950 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
951 	struct lagg_softc *sc;
952 	struct lagg_port *lp = NULL;
953 	int error = 0;
954 	struct rm_priotracker tracker;
955 
956 	/* Should be checked by the caller */
957 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
958 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
959 		goto fallback;
960 
961 	switch (cmd) {
962 	case SIOCGLAGGPORT:
963 		if (rp->rp_portname[0] == '\0' ||
964 		    ifunit(rp->rp_portname) != ifp) {
965 			error = EINVAL;
966 			break;
967 		}
968 
969 		LAGG_RLOCK(sc, &tracker);
970 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
971 			error = ENOENT;
972 			LAGG_RUNLOCK(sc, &tracker);
973 			break;
974 		}
975 
976 		lagg_port2req(lp, rp);
977 		LAGG_RUNLOCK(sc, &tracker);
978 		break;
979 
980 	case SIOCSIFCAP:
981 		if (lp->lp_ioctl == NULL) {
982 			error = EINVAL;
983 			break;
984 		}
985 		error = (*lp->lp_ioctl)(ifp, cmd, data);
986 		if (error)
987 			break;
988 
989 		/* Update lagg interface capabilities */
990 		LAGG_WLOCK(sc);
991 		lagg_capabilities(sc);
992 		LAGG_WUNLOCK(sc);
993 		break;
994 
995 	case SIOCSIFMTU:
996 		/* Do not allow the MTU to be changed once joined */
997 		error = EINVAL;
998 		break;
999 
1000 	default:
1001 		goto fallback;
1002 	}
1003 
1004 	return (error);
1005 
1006 fallback:
1007 	if (lp->lp_ioctl != NULL)
1008 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1009 
1010 	return (EINVAL);
1011 }
1012 
1013 /*
1014  * Requests counter @cnt data.
1015  *
1016  * Counter value is calculated the following way:
1017  * 1) for each port, sum  difference between current and "initial" measurements.
1018  * 2) add lagg logical interface counters.
1019  * 3) add data from detached_counters array.
1020  *
1021  * We also do the following things on ports attach/detach:
1022  * 1) On port attach we store all counters it has into port_counter array.
1023  * 2) On port detach we add the different between "initial" and
1024  *   current counters data to detached_counters array.
1025  */
1026 static uint64_t
1027 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1028 {
1029 	struct lagg_softc *sc;
1030 	struct lagg_port *lp;
1031 	struct ifnet *lpifp;
1032 	struct rm_priotracker tracker;
1033 	uint64_t newval, oldval, vsum;
1034 
1035 	/* Revise this when we've got non-generic counters. */
1036 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1037 
1038 	sc = (struct lagg_softc *)ifp->if_softc;
1039 	LAGG_RLOCK(sc, &tracker);
1040 
1041 	vsum = 0;
1042 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1043 		/* Saved attached value */
1044 		oldval = lp->port_counters.val[cnt];
1045 		/* current value */
1046 		lpifp = lp->lp_ifp;
1047 		newval = lpifp->if_get_counter(lpifp, cnt);
1048 		/* Calculate diff and save new */
1049 		vsum += newval - oldval;
1050 	}
1051 
1052 	/*
1053 	 * Add counter data which might be added by upper
1054 	 * layer protocols operating on logical interface.
1055 	 */
1056 	vsum += if_get_counter_default(ifp, cnt);
1057 
1058 	/*
1059 	 * Add counter data from detached ports counters
1060 	 */
1061 	vsum += sc->detached_counters.val[cnt];
1062 
1063 	LAGG_RUNLOCK(sc, &tracker);
1064 
1065 	return (vsum);
1066 }
1067 
1068 /*
1069  * For direct output to child ports.
1070  */
1071 static int
1072 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1073 	const struct sockaddr *dst, struct route *ro)
1074 {
1075 	struct lagg_port *lp = ifp->if_lagg;
1076 
1077 	switch (dst->sa_family) {
1078 		case pseudo_AF_HDRCMPLT:
1079 		case AF_UNSPEC:
1080 			return ((*lp->lp_output)(ifp, m, dst, ro));
1081 	}
1082 
1083 	/* drop any other frames */
1084 	m_freem(m);
1085 	return (ENETDOWN);
1086 }
1087 
1088 static void
1089 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1090 {
1091 	struct lagg_port *lp;
1092 	struct lagg_softc *sc;
1093 
1094 	if ((lp = ifp->if_lagg) == NULL)
1095 		return;
1096 	/* If the ifnet is just being renamed, don't do anything. */
1097 	if (ifp->if_flags & IFF_RENAMING)
1098 		return;
1099 
1100 	sc = lp->lp_softc;
1101 
1102 	LAGG_WLOCK(sc);
1103 	lp->lp_detaching = 1;
1104 	lagg_port_destroy(lp, 1);
1105 	LAGG_WUNLOCK(sc);
1106 }
1107 
1108 static void
1109 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1110 {
1111 	struct lagg_softc *sc = lp->lp_softc;
1112 
1113 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1114 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1115 	rp->rp_prio = lp->lp_prio;
1116 	rp->rp_flags = lp->lp_flags;
1117 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1118 
1119 	/* Add protocol specific flags */
1120 	switch (sc->sc_proto) {
1121 		case LAGG_PROTO_FAILOVER:
1122 			if (lp == sc->sc_primary)
1123 				rp->rp_flags |= LAGG_PORT_MASTER;
1124 			if (lp == lagg_link_active(sc, sc->sc_primary))
1125 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1126 			break;
1127 
1128 		case LAGG_PROTO_ROUNDROBIN:
1129 		case LAGG_PROTO_LOADBALANCE:
1130 		case LAGG_PROTO_ETHERCHANNEL:
1131 		case LAGG_PROTO_BROADCAST:
1132 			if (LAGG_PORTACTIVE(lp))
1133 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1134 			break;
1135 
1136 		case LAGG_PROTO_LACP:
1137 			/* LACP has a different definition of active */
1138 			if (lacp_isactive(lp))
1139 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1140 			if (lacp_iscollecting(lp))
1141 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1142 			if (lacp_isdistributing(lp))
1143 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1144 			break;
1145 	}
1146 
1147 }
1148 
1149 static void
1150 lagg_init(void *xsc)
1151 {
1152 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1153 	struct lagg_port *lp;
1154 	struct ifnet *ifp = sc->sc_ifp;
1155 
1156 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1157 		return;
1158 
1159 	LAGG_WLOCK(sc);
1160 
1161 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1162 	/* Update the port lladdrs */
1163 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1164 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
1165 
1166 	lagg_proto_init(sc);
1167 
1168 	LAGG_WUNLOCK(sc);
1169 }
1170 
1171 static void
1172 lagg_stop(struct lagg_softc *sc)
1173 {
1174 	struct ifnet *ifp = sc->sc_ifp;
1175 
1176 	LAGG_WLOCK_ASSERT(sc);
1177 
1178 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1179 		return;
1180 
1181 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1182 
1183 	lagg_proto_stop(sc);
1184 }
1185 
1186 static int
1187 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1188 {
1189 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1190 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1191 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1192 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1193 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1194 	struct ifreq *ifr = (struct ifreq *)data;
1195 	struct lagg_port *lp;
1196 	struct ifnet *tpif;
1197 	struct thread *td = curthread;
1198 	char *buf, *outbuf;
1199 	int count, buflen, len, error = 0;
1200 	struct rm_priotracker tracker;
1201 
1202 	bzero(&rpbuf, sizeof(rpbuf));
1203 
1204 	switch (cmd) {
1205 	case SIOCGLAGG:
1206 		LAGG_RLOCK(sc, &tracker);
1207 		count = 0;
1208 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1209 			count++;
1210 		buflen = count * sizeof(struct lagg_reqport);
1211 		LAGG_RUNLOCK(sc, &tracker);
1212 
1213 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1214 
1215 		LAGG_RLOCK(sc, &tracker);
1216 		ra->ra_proto = sc->sc_proto;
1217 		lagg_proto_request(sc, &ra->ra_psc);
1218 		count = 0;
1219 		buf = outbuf;
1220 		len = min(ra->ra_size, buflen);
1221 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1222 			if (len < sizeof(rpbuf))
1223 				break;
1224 
1225 			lagg_port2req(lp, &rpbuf);
1226 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1227 			count++;
1228 			buf += sizeof(rpbuf);
1229 			len -= sizeof(rpbuf);
1230 		}
1231 		LAGG_RUNLOCK(sc, &tracker);
1232 		ra->ra_ports = count;
1233 		ra->ra_size = count * sizeof(rpbuf);
1234 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1235 		free(outbuf, M_TEMP);
1236 		break;
1237 	case SIOCSLAGG:
1238 		error = priv_check(td, PRIV_NET_LAGG);
1239 		if (error)
1240 			break;
1241 		if (ra->ra_proto < 1 || ra->ra_proto >= LAGG_PROTO_MAX) {
1242 			error = EPROTONOSUPPORT;
1243 			break;
1244 		}
1245 
1246 		LAGG_WLOCK(sc);
1247 		lagg_proto_detach(sc);
1248 		lagg_proto_attach(sc, ra->ra_proto);
1249 		break;
1250 	case SIOCGLAGGOPTS:
1251 		ro->ro_opts = sc->sc_opts;
1252 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1253 			struct lacp_softc *lsc;
1254 
1255 			lsc = (struct lacp_softc *)sc->sc_psc;
1256 			if (lsc->lsc_debug.lsc_tx_test != 0)
1257 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1258 			if (lsc->lsc_debug.lsc_rx_test != 0)
1259 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1260 			if (lsc->lsc_strict_mode != 0)
1261 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1262 
1263 			ro->ro_active = sc->sc_active;
1264 		} else {
1265 			ro->ro_active = 0;
1266 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1267 				ro->ro_active += LAGG_PORTACTIVE(lp);
1268 		}
1269 		ro->ro_flapping = sc->sc_flapping;
1270 		ro->ro_flowid_shift = sc->flowid_shift;
1271 		break;
1272 	case SIOCSLAGGOPTS:
1273 		error = priv_check(td, PRIV_NET_LAGG);
1274 		if (error)
1275 			break;
1276 		if (ro->ro_opts == 0)
1277 			break;
1278 		/*
1279 		 * Set options.  LACP options are stored in sc->sc_psc,
1280 		 * not in sc_opts.
1281 		 */
1282 		int valid, lacp;
1283 
1284 		switch (ro->ro_opts) {
1285 		case LAGG_OPT_USE_FLOWID:
1286 		case -LAGG_OPT_USE_FLOWID:
1287 		case LAGG_OPT_FLOWIDSHIFT:
1288 			valid = 1;
1289 			lacp = 0;
1290 			break;
1291 		case LAGG_OPT_LACP_TXTEST:
1292 		case -LAGG_OPT_LACP_TXTEST:
1293 		case LAGG_OPT_LACP_RXTEST:
1294 		case -LAGG_OPT_LACP_RXTEST:
1295 		case LAGG_OPT_LACP_STRICT:
1296 		case -LAGG_OPT_LACP_STRICT:
1297 			valid = lacp = 1;
1298 			break;
1299 		default:
1300 			valid = lacp = 0;
1301 			break;
1302 		}
1303 
1304 		LAGG_WLOCK(sc);
1305 		if (valid == 0 ||
1306 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1307 			/* Invalid combination of options specified. */
1308 			error = EINVAL;
1309 			LAGG_WUNLOCK(sc);
1310 			break;	/* Return from SIOCSLAGGOPTS. */
1311 		}
1312 		/*
1313 		 * Store new options into sc->sc_opts except for
1314 		 * FLOWIDSHIFT and LACP options.
1315 		 */
1316 		if (lacp == 0) {
1317 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1318 				sc->flowid_shift = ro->ro_flowid_shift;
1319 			else if (ro->ro_opts > 0)
1320 				sc->sc_opts |= ro->ro_opts;
1321 			else
1322 				sc->sc_opts &= ~ro->ro_opts;
1323 		} else {
1324 			struct lacp_softc *lsc;
1325 
1326 			lsc = (struct lacp_softc *)sc->sc_psc;
1327 
1328 			switch (ro->ro_opts) {
1329 			case LAGG_OPT_LACP_TXTEST:
1330 				lsc->lsc_debug.lsc_tx_test = 1;
1331 				break;
1332 			case -LAGG_OPT_LACP_TXTEST:
1333 				lsc->lsc_debug.lsc_tx_test = 0;
1334 				break;
1335 			case LAGG_OPT_LACP_RXTEST:
1336 				lsc->lsc_debug.lsc_rx_test = 1;
1337 				break;
1338 			case -LAGG_OPT_LACP_RXTEST:
1339 				lsc->lsc_debug.lsc_rx_test = 0;
1340 				break;
1341 			case LAGG_OPT_LACP_STRICT:
1342 				lsc->lsc_strict_mode = 1;
1343 				break;
1344 			case -LAGG_OPT_LACP_STRICT:
1345 				lsc->lsc_strict_mode = 0;
1346 				break;
1347 			}
1348 		}
1349 		LAGG_WUNLOCK(sc);
1350 		break;
1351 	case SIOCGLAGGFLAGS:
1352 		rf->rf_flags = sc->sc_flags;
1353 		break;
1354 	case SIOCSLAGGHASH:
1355 		error = priv_check(td, PRIV_NET_LAGG);
1356 		if (error)
1357 			break;
1358 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1359 			error = EINVAL;
1360 			break;
1361 		}
1362 		LAGG_WLOCK(sc);
1363 		sc->sc_flags &= ~LAGG_F_HASHMASK;
1364 		sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK;
1365 		LAGG_WUNLOCK(sc);
1366 		break;
1367 	case SIOCGLAGGPORT:
1368 		if (rp->rp_portname[0] == '\0' ||
1369 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1370 			error = EINVAL;
1371 			break;
1372 		}
1373 
1374 		LAGG_RLOCK(sc, &tracker);
1375 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1376 		    lp->lp_softc != sc) {
1377 			error = ENOENT;
1378 			LAGG_RUNLOCK(sc, &tracker);
1379 			break;
1380 		}
1381 
1382 		lagg_port2req(lp, rp);
1383 		LAGG_RUNLOCK(sc, &tracker);
1384 		break;
1385 	case SIOCSLAGGPORT:
1386 		error = priv_check(td, PRIV_NET_LAGG);
1387 		if (error)
1388 			break;
1389 		if (rp->rp_portname[0] == '\0' ||
1390 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1391 			error = EINVAL;
1392 			break;
1393 		}
1394 #ifdef INET6
1395 		/*
1396 		 * A laggport interface should not have inet6 address
1397 		 * because two interfaces with a valid link-local
1398 		 * scope zone must not be merged in any form.  This
1399 		 * restriction is needed to prevent violation of
1400 		 * link-local scope zone.  Attempts to add a laggport
1401 		 * interface which has inet6 addresses triggers
1402 		 * removal of all inet6 addresses on the member
1403 		 * interface.
1404 		 */
1405 		if (in6ifa_llaonifp(tpif)) {
1406 			in6_ifdetach(tpif);
1407 				if_printf(sc->sc_ifp,
1408 				    "IPv6 addresses on %s have been removed "
1409 				    "before adding it as a member to prevent "
1410 				    "IPv6 address scope violation.\n",
1411 				    tpif->if_xname);
1412 		}
1413 #endif
1414 		LAGG_WLOCK(sc);
1415 		error = lagg_port_create(sc, tpif);
1416 		LAGG_WUNLOCK(sc);
1417 		break;
1418 	case SIOCSLAGGDELPORT:
1419 		error = priv_check(td, PRIV_NET_LAGG);
1420 		if (error)
1421 			break;
1422 		if (rp->rp_portname[0] == '\0' ||
1423 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1424 			error = EINVAL;
1425 			break;
1426 		}
1427 
1428 		LAGG_WLOCK(sc);
1429 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1430 		    lp->lp_softc != sc) {
1431 			error = ENOENT;
1432 			LAGG_WUNLOCK(sc);
1433 			break;
1434 		}
1435 
1436 		error = lagg_port_destroy(lp, 1);
1437 		LAGG_WUNLOCK(sc);
1438 		break;
1439 	case SIOCSIFFLAGS:
1440 		/* Set flags on ports too */
1441 		LAGG_WLOCK(sc);
1442 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1443 			lagg_setflags(lp, 1);
1444 		}
1445 		LAGG_WUNLOCK(sc);
1446 
1447 		if (!(ifp->if_flags & IFF_UP) &&
1448 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1449 			/*
1450 			 * If interface is marked down and it is running,
1451 			 * then stop and disable it.
1452 			 */
1453 			LAGG_WLOCK(sc);
1454 			lagg_stop(sc);
1455 			LAGG_WUNLOCK(sc);
1456 		} else if ((ifp->if_flags & IFF_UP) &&
1457 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1458 			/*
1459 			 * If interface is marked up and it is stopped, then
1460 			 * start it.
1461 			 */
1462 			(*ifp->if_init)(sc);
1463 		}
1464 		break;
1465 	case SIOCADDMULTI:
1466 	case SIOCDELMULTI:
1467 		LAGG_WLOCK(sc);
1468 		error = lagg_ether_setmulti(sc);
1469 		LAGG_WUNLOCK(sc);
1470 		break;
1471 	case SIOCSIFMEDIA:
1472 	case SIOCGIFMEDIA:
1473 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1474 		break;
1475 
1476 	case SIOCSIFCAP:
1477 	case SIOCSIFMTU:
1478 		/* Do not allow the MTU or caps to be directly changed */
1479 		error = EINVAL;
1480 		break;
1481 
1482 	default:
1483 		error = ether_ioctl(ifp, cmd, data);
1484 		break;
1485 	}
1486 	return (error);
1487 }
1488 
1489 static int
1490 lagg_ether_setmulti(struct lagg_softc *sc)
1491 {
1492 	struct lagg_port *lp;
1493 
1494 	LAGG_WLOCK_ASSERT(sc);
1495 
1496 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1497 		/* First, remove any existing filter entries. */
1498 		lagg_ether_cmdmulti(lp, 0);
1499 		/* copy all addresses from the lagg interface to the port */
1500 		lagg_ether_cmdmulti(lp, 1);
1501 	}
1502 	return (0);
1503 }
1504 
1505 static int
1506 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1507 {
1508 	struct lagg_softc *sc = lp->lp_softc;
1509 	struct ifnet *ifp = lp->lp_ifp;
1510 	struct ifnet *scifp = sc->sc_ifp;
1511 	struct lagg_mc *mc;
1512 	struct ifmultiaddr *ifma;
1513 	int error;
1514 
1515 	LAGG_WLOCK_ASSERT(sc);
1516 
1517 	if (set) {
1518 		IF_ADDR_WLOCK(scifp);
1519 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1520 			if (ifma->ifma_addr->sa_family != AF_LINK)
1521 				continue;
1522 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1523 			if (mc == NULL) {
1524 				IF_ADDR_WUNLOCK(scifp);
1525 				return (ENOMEM);
1526 			}
1527 			bcopy(ifma->ifma_addr, &mc->mc_addr,
1528 			    ifma->ifma_addr->sa_len);
1529 			mc->mc_addr.sdl_index = ifp->if_index;
1530 			mc->mc_ifma = NULL;
1531 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1532 		}
1533 		IF_ADDR_WUNLOCK(scifp);
1534 		SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1535 			error = if_addmulti(ifp,
1536 			    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1537 			if (error)
1538 				return (error);
1539 		}
1540 	} else {
1541 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1542 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1543 			if (mc->mc_ifma && !lp->lp_detaching)
1544 				if_delmulti_ifma(mc->mc_ifma);
1545 			free(mc, M_DEVBUF);
1546 		}
1547 	}
1548 	return (0);
1549 }
1550 
1551 /* Handle a ref counted flag that should be set on the lagg port as well */
1552 static int
1553 lagg_setflag(struct lagg_port *lp, int flag, int status,
1554     int (*func)(struct ifnet *, int))
1555 {
1556 	struct lagg_softc *sc = lp->lp_softc;
1557 	struct ifnet *scifp = sc->sc_ifp;
1558 	struct ifnet *ifp = lp->lp_ifp;
1559 	int error;
1560 
1561 	LAGG_WLOCK_ASSERT(sc);
1562 
1563 	status = status ? (scifp->if_flags & flag) : 0;
1564 	/* Now "status" contains the flag value or 0 */
1565 
1566 	/*
1567 	 * See if recorded ports status is different from what
1568 	 * we want it to be.  If it is, flip it.  We record ports
1569 	 * status in lp_ifflags so that we won't clear ports flag
1570 	 * we haven't set.  In fact, we don't clear or set ports
1571 	 * flags directly, but get or release references to them.
1572 	 * That's why we can be sure that recorded flags still are
1573 	 * in accord with actual ports flags.
1574 	 */
1575 	if (status != (lp->lp_ifflags & flag)) {
1576 		error = (*func)(ifp, status);
1577 		if (error)
1578 			return (error);
1579 		lp->lp_ifflags &= ~flag;
1580 		lp->lp_ifflags |= status;
1581 	}
1582 	return (0);
1583 }
1584 
1585 /*
1586  * Handle IFF_* flags that require certain changes on the lagg port
1587  * if "status" is true, update ports flags respective to the lagg
1588  * if "status" is false, forcedly clear the flags set on port.
1589  */
1590 static int
1591 lagg_setflags(struct lagg_port *lp, int status)
1592 {
1593 	int error, i;
1594 
1595 	for (i = 0; lagg_pflags[i].flag; i++) {
1596 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1597 		    status, lagg_pflags[i].func);
1598 		if (error)
1599 			return (error);
1600 	}
1601 	return (0);
1602 }
1603 
1604 static int
1605 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1606 {
1607 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1608 	int error, len, mcast;
1609 	struct rm_priotracker tracker;
1610 
1611 	len = m->m_pkthdr.len;
1612 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1613 
1614 	LAGG_RLOCK(sc, &tracker);
1615 	/* We need a Tx algorithm and at least one port */
1616 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1617 		LAGG_RUNLOCK(sc, &tracker);
1618 		m_freem(m);
1619 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1620 		return (ENXIO);
1621 	}
1622 
1623 	ETHER_BPF_MTAP(ifp, m);
1624 
1625 	error = lagg_proto_start(sc, m);
1626 	LAGG_RUNLOCK(sc, &tracker);
1627 
1628 	if (error != 0)
1629 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1630 
1631 	return (error);
1632 }
1633 
1634 /*
1635  * The ifp->if_qflush entry point for lagg(4) is no-op.
1636  */
1637 static void
1638 lagg_qflush(struct ifnet *ifp __unused)
1639 {
1640 }
1641 
1642 static struct mbuf *
1643 lagg_input(struct ifnet *ifp, struct mbuf *m)
1644 {
1645 	struct lagg_port *lp = ifp->if_lagg;
1646 	struct lagg_softc *sc = lp->lp_softc;
1647 	struct ifnet *scifp = sc->sc_ifp;
1648 	struct rm_priotracker tracker;
1649 
1650 	LAGG_RLOCK(sc, &tracker);
1651 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1652 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1653 	    sc->sc_proto == LAGG_PROTO_NONE) {
1654 		LAGG_RUNLOCK(sc, &tracker);
1655 		m_freem(m);
1656 		return (NULL);
1657 	}
1658 
1659 	ETHER_BPF_MTAP(scifp, m);
1660 
1661 	m = (lp->lp_detaching == 0) ? lagg_proto_input(sc, lp, m) : NULL;
1662 
1663 	if (m != NULL) {
1664 		if (scifp->if_flags & IFF_MONITOR) {
1665 			m_freem(m);
1666 			m = NULL;
1667 		}
1668 	}
1669 
1670 	LAGG_RUNLOCK(sc, &tracker);
1671 	return (m);
1672 }
1673 
1674 static int
1675 lagg_media_change(struct ifnet *ifp)
1676 {
1677 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1678 
1679 	if (sc->sc_ifflags & IFF_DEBUG)
1680 		printf("%s\n", __func__);
1681 
1682 	/* Ignore */
1683 	return (0);
1684 }
1685 
1686 static void
1687 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1688 {
1689 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1690 	struct lagg_port *lp;
1691 	struct rm_priotracker tracker;
1692 
1693 	imr->ifm_status = IFM_AVALID;
1694 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1695 
1696 	LAGG_RLOCK(sc, &tracker);
1697 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1698 		if (LAGG_PORTACTIVE(lp))
1699 			imr->ifm_status |= IFM_ACTIVE;
1700 	}
1701 	LAGG_RUNLOCK(sc, &tracker);
1702 }
1703 
1704 static void
1705 lagg_linkstate(struct lagg_softc *sc)
1706 {
1707 	struct lagg_port *lp;
1708 	int new_link = LINK_STATE_DOWN;
1709 	uint64_t speed;
1710 
1711 	/* Our link is considered up if at least one of our ports is active */
1712 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1713 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1714 			new_link = LINK_STATE_UP;
1715 			break;
1716 		}
1717 	}
1718 	if_link_state_change(sc->sc_ifp, new_link);
1719 
1720 	/* Update if_baudrate to reflect the max possible speed */
1721 	switch (sc->sc_proto) {
1722 		case LAGG_PROTO_FAILOVER:
1723 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1724 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1725 			break;
1726 		case LAGG_PROTO_ROUNDROBIN:
1727 		case LAGG_PROTO_LOADBALANCE:
1728 		case LAGG_PROTO_ETHERCHANNEL:
1729 		case LAGG_PROTO_BROADCAST:
1730 			speed = 0;
1731 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1732 				speed += lp->lp_ifp->if_baudrate;
1733 			sc->sc_ifp->if_baudrate = speed;
1734 			break;
1735 		case LAGG_PROTO_LACP:
1736 			/* LACP updates if_baudrate itself */
1737 			break;
1738 	}
1739 }
1740 
1741 static void
1742 lagg_port_state(struct ifnet *ifp, int state)
1743 {
1744 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1745 	struct lagg_softc *sc = NULL;
1746 
1747 	if (lp != NULL)
1748 		sc = lp->lp_softc;
1749 	if (sc == NULL)
1750 		return;
1751 
1752 	LAGG_WLOCK(sc);
1753 	lagg_linkstate(sc);
1754 	lagg_proto_linkstate(sc, lp);
1755 	LAGG_WUNLOCK(sc);
1756 }
1757 
1758 struct lagg_port *
1759 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1760 {
1761 	struct lagg_port *lp_next, *rval = NULL;
1762 	// int new_link = LINK_STATE_DOWN;
1763 
1764 	LAGG_RLOCK_ASSERT(sc);
1765 	/*
1766 	 * Search a port which reports an active link state.
1767 	 */
1768 
1769 	if (lp == NULL)
1770 		goto search;
1771 	if (LAGG_PORTACTIVE(lp)) {
1772 		rval = lp;
1773 		goto found;
1774 	}
1775 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1776 	    LAGG_PORTACTIVE(lp_next)) {
1777 		rval = lp_next;
1778 		goto found;
1779 	}
1780 
1781 search:
1782 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1783 		if (LAGG_PORTACTIVE(lp_next)) {
1784 			rval = lp_next;
1785 			goto found;
1786 		}
1787 	}
1788 
1789 found:
1790 	if (rval != NULL) {
1791 		/*
1792 		 * The IEEE 802.1D standard assumes that a lagg with
1793 		 * multiple ports is always full duplex. This is valid
1794 		 * for load sharing laggs and if at least two links
1795 		 * are active. Unfortunately, checking the latter would
1796 		 * be too expensive at this point.
1797 		 XXX
1798 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1799 		    (sc->sc_count > 1))
1800 			new_link = LINK_STATE_FULL_DUPLEX;
1801 		else
1802 			new_link = rval->lp_link_state;
1803 		 */
1804 	}
1805 
1806 	return (rval);
1807 }
1808 
1809 static const void *
1810 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1811 {
1812 	if (m->m_pkthdr.len < (off + len)) {
1813 		return (NULL);
1814 	} else if (m->m_len < (off + len)) {
1815 		m_copydata(m, off, len, buf);
1816 		return (buf);
1817 	}
1818 	return (mtod(m, char *) + off);
1819 }
1820 
1821 uint32_t
1822 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
1823 {
1824 	uint16_t etype;
1825 	uint32_t p = key;
1826 	int off;
1827 	struct ether_header *eh;
1828 	const struct ether_vlan_header *vlan;
1829 #ifdef INET
1830 	const struct ip *ip;
1831 	const uint32_t *ports;
1832 	int iphlen;
1833 #endif
1834 #ifdef INET6
1835 	const struct ip6_hdr *ip6;
1836 	uint32_t flow;
1837 #endif
1838 	union {
1839 #ifdef INET
1840 		struct ip ip;
1841 #endif
1842 #ifdef INET6
1843 		struct ip6_hdr ip6;
1844 #endif
1845 		struct ether_vlan_header vlan;
1846 		uint32_t port;
1847 	} buf;
1848 
1849 
1850 	off = sizeof(*eh);
1851 	if (m->m_len < off)
1852 		goto out;
1853 	eh = mtod(m, struct ether_header *);
1854 	etype = ntohs(eh->ether_type);
1855 	if (sc->sc_flags & LAGG_F_HASHL2) {
1856 		p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p);
1857 		p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1858 	}
1859 
1860 	/* Special handling for encapsulating VLAN frames */
1861 	if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) {
1862 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1863 		    sizeof(m->m_pkthdr.ether_vtag), p);
1864 	} else if (etype == ETHERTYPE_VLAN) {
1865 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &buf);
1866 		if (vlan == NULL)
1867 			goto out;
1868 
1869 		if (sc->sc_flags & LAGG_F_HASHL2)
1870 			p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1871 		etype = ntohs(vlan->evl_proto);
1872 		off += sizeof(*vlan) - sizeof(*eh);
1873 	}
1874 
1875 	switch (etype) {
1876 #ifdef INET
1877 	case ETHERTYPE_IP:
1878 		ip = lagg_gethdr(m, off, sizeof(*ip), &buf);
1879 		if (ip == NULL)
1880 			goto out;
1881 
1882 		if (sc->sc_flags & LAGG_F_HASHL3) {
1883 			p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1884 			p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1885 		}
1886 		if (!(sc->sc_flags & LAGG_F_HASHL4))
1887 			break;
1888 		switch (ip->ip_p) {
1889 			case IPPROTO_TCP:
1890 			case IPPROTO_UDP:
1891 			case IPPROTO_SCTP:
1892 				iphlen = ip->ip_hl << 2;
1893 				if (iphlen < sizeof(*ip))
1894 					break;
1895 				off += iphlen;
1896 				ports = lagg_gethdr(m, off, sizeof(*ports), &buf);
1897 				if (ports == NULL)
1898 					break;
1899 				p = hash32_buf(ports, sizeof(*ports), p);
1900 				break;
1901 		}
1902 		break;
1903 #endif
1904 #ifdef INET6
1905 	case ETHERTYPE_IPV6:
1906 		if (!(sc->sc_flags & LAGG_F_HASHL3))
1907 			break;
1908 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf);
1909 		if (ip6 == NULL)
1910 			goto out;
1911 
1912 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1913 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1914 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1915 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1916 		break;
1917 #endif
1918 	}
1919 out:
1920 	return (p);
1921 }
1922 
1923 int
1924 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1925 {
1926 
1927 	return (ifp->if_transmit)(ifp, m);
1928 }
1929 
1930 /*
1931  * Simple round robin aggregation
1932  */
1933 static void
1934 lagg_rr_attach(struct lagg_softc *sc)
1935 {
1936 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1937 	sc->sc_seq = 0;
1938 }
1939 
1940 static int
1941 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1942 {
1943 	struct lagg_port *lp;
1944 	uint32_t p;
1945 
1946 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1947 	p %= sc->sc_count;
1948 	lp = SLIST_FIRST(&sc->sc_ports);
1949 	while (p--)
1950 		lp = SLIST_NEXT(lp, lp_entries);
1951 
1952 	/*
1953 	 * Check the port's link state. This will return the next active
1954 	 * port if the link is down or the port is NULL.
1955 	 */
1956 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1957 		m_freem(m);
1958 		return (ENETDOWN);
1959 	}
1960 
1961 	/* Send mbuf */
1962 	return (lagg_enqueue(lp->lp_ifp, m));
1963 }
1964 
1965 static struct mbuf *
1966 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1967 {
1968 	struct ifnet *ifp = sc->sc_ifp;
1969 
1970 	/* Just pass in the packet to our lagg device */
1971 	m->m_pkthdr.rcvif = ifp;
1972 
1973 	return (m);
1974 }
1975 
1976 /*
1977  * Broadcast mode
1978  */
1979 static int
1980 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1981 {
1982 	int active_ports = 0;
1983 	int errors = 0;
1984 	int ret;
1985 	struct lagg_port *lp, *last = NULL;
1986 	struct mbuf *m0;
1987 
1988 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1989 		if (!LAGG_PORTACTIVE(lp))
1990 			continue;
1991 
1992 		active_ports++;
1993 
1994 		if (last != NULL) {
1995 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
1996 			if (m0 == NULL) {
1997 				ret = ENOBUFS;
1998 				errors++;
1999 				break;
2000 			}
2001 
2002 			ret = lagg_enqueue(last->lp_ifp, m0);
2003 			if (ret != 0)
2004 				errors++;
2005 		}
2006 		last = lp;
2007 	}
2008 	if (last == NULL) {
2009 		m_freem(m);
2010 		return (ENOENT);
2011 	}
2012 	if ((last = lagg_link_active(sc, last)) == NULL) {
2013 		m_freem(m);
2014 		return (ENETDOWN);
2015 	}
2016 
2017 	ret = lagg_enqueue(last->lp_ifp, m);
2018 	if (ret != 0)
2019 		errors++;
2020 
2021 	if (errors == 0)
2022 		return (ret);
2023 
2024 	return (0);
2025 }
2026 
2027 static struct mbuf*
2028 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2029 {
2030 	struct ifnet *ifp = sc->sc_ifp;
2031 
2032 	/* Just pass in the packet to our lagg device */
2033 	m->m_pkthdr.rcvif = ifp;
2034 	return (m);
2035 }
2036 
2037 /*
2038  * Active failover
2039  */
2040 static int
2041 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2042 {
2043 	struct lagg_port *lp;
2044 
2045 	/* Use the master port if active or the next available port */
2046 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2047 		m_freem(m);
2048 		return (ENETDOWN);
2049 	}
2050 
2051 	/* Send mbuf */
2052 	return (lagg_enqueue(lp->lp_ifp, m));
2053 }
2054 
2055 static struct mbuf *
2056 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2057 {
2058 	struct ifnet *ifp = sc->sc_ifp;
2059 	struct lagg_port *tmp_tp;
2060 
2061 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2062 		m->m_pkthdr.rcvif = ifp;
2063 		return (m);
2064 	}
2065 
2066 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2067 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2068 		/*
2069 		 * If tmp_tp is null, we've recieved a packet when all
2070 		 * our links are down. Weird, but process it anyways.
2071 		 */
2072 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2073 			m->m_pkthdr.rcvif = ifp;
2074 			return (m);
2075 		}
2076 	}
2077 
2078 	m_freem(m);
2079 	return (NULL);
2080 }
2081 
2082 /*
2083  * Loadbalancing
2084  */
2085 static void
2086 lagg_lb_attach(struct lagg_softc *sc)
2087 {
2088 	struct lagg_port *lp;
2089 	struct lagg_lb *lb;
2090 
2091 	lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
2092 
2093 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
2094 
2095 	lb->lb_key = arc4random();
2096 	sc->sc_psc = lb;
2097 
2098 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2099 		lagg_lb_port_create(lp);
2100 }
2101 
2102 static void
2103 lagg_lb_detach(struct lagg_softc *sc)
2104 {
2105 	struct lagg_lb *lb;
2106 
2107 	lb = (struct lagg_lb *)sc->sc_psc;
2108 	LAGG_WUNLOCK(sc);
2109 	if (lb != NULL)
2110 		free(lb, M_DEVBUF);
2111 }
2112 
2113 static int
2114 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2115 {
2116 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2117 	struct lagg_port *lp_next;
2118 	int i = 0;
2119 
2120 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2121 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2122 		if (lp_next == lp)
2123 			continue;
2124 		if (i >= LAGG_MAX_PORTS)
2125 			return (EINVAL);
2126 		if (sc->sc_ifflags & IFF_DEBUG)
2127 			printf("%s: port %s at index %d\n",
2128 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2129 		lb->lb_ports[i++] = lp_next;
2130 	}
2131 
2132 	return (0);
2133 }
2134 
2135 static int
2136 lagg_lb_port_create(struct lagg_port *lp)
2137 {
2138 	struct lagg_softc *sc = lp->lp_softc;
2139 	return (lagg_lb_porttable(sc, NULL));
2140 }
2141 
2142 static void
2143 lagg_lb_port_destroy(struct lagg_port *lp)
2144 {
2145 	struct lagg_softc *sc = lp->lp_softc;
2146 	lagg_lb_porttable(sc, lp);
2147 }
2148 
2149 static int
2150 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2151 {
2152 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2153 	struct lagg_port *lp = NULL;
2154 	uint32_t p = 0;
2155 
2156 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2157 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2158 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2159 	else
2160 		p = lagg_hashmbuf(sc, m, lb->lb_key);
2161 	p %= sc->sc_count;
2162 	lp = lb->lb_ports[p];
2163 
2164 	/*
2165 	 * Check the port's link state. This will return the next active
2166 	 * port if the link is down or the port is NULL.
2167 	 */
2168 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2169 		m_freem(m);
2170 		return (ENETDOWN);
2171 	}
2172 
2173 	/* Send mbuf */
2174 	return (lagg_enqueue(lp->lp_ifp, m));
2175 }
2176 
2177 static struct mbuf *
2178 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2179 {
2180 	struct ifnet *ifp = sc->sc_ifp;
2181 
2182 	/* Just pass in the packet to our lagg device */
2183 	m->m_pkthdr.rcvif = ifp;
2184 
2185 	return (m);
2186 }
2187 
2188 /*
2189  * 802.3ad LACP
2190  */
2191 static void
2192 lagg_lacp_attach(struct lagg_softc *sc)
2193 {
2194 	struct lagg_port *lp;
2195 
2196 	lacp_attach(sc);
2197 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2198 		lacp_port_create(lp);
2199 }
2200 
2201 static void
2202 lagg_lacp_detach(struct lagg_softc *sc)
2203 {
2204 	struct lagg_port *lp;
2205 	void *psc;
2206 
2207 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2208 		lacp_port_destroy(lp);
2209 
2210 	psc = sc->sc_psc;
2211 	sc->sc_psc = NULL;
2212 	LAGG_WUNLOCK(sc);
2213 
2214 	lacp_detach(psc);
2215 }
2216 
2217 static void
2218 lagg_lacp_lladdr(struct lagg_softc *sc)
2219 {
2220 	struct lagg_port *lp;
2221 
2222 	/* purge all the lacp ports */
2223 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2224 		lacp_port_destroy(lp);
2225 
2226 	/* add them back in */
2227 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2228 		lacp_port_create(lp);
2229 }
2230 
2231 static int
2232 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2233 {
2234 	struct lagg_port *lp;
2235 
2236 	lp = lacp_select_tx_port(sc, m);
2237 	if (lp == NULL) {
2238 		m_freem(m);
2239 		return (ENETDOWN);
2240 	}
2241 
2242 	/* Send mbuf */
2243 	return (lagg_enqueue(lp->lp_ifp, m));
2244 }
2245 
2246 static struct mbuf *
2247 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2248 {
2249 	struct ifnet *ifp = sc->sc_ifp;
2250 	struct ether_header *eh;
2251 	u_short etype;
2252 
2253 	eh = mtod(m, struct ether_header *);
2254 	etype = ntohs(eh->ether_type);
2255 
2256 	/* Tap off LACP control messages */
2257 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2258 		m = lacp_input(lp, m);
2259 		if (m == NULL)
2260 			return (NULL);
2261 	}
2262 
2263 	/*
2264 	 * If the port is not collecting or not in the active aggregator then
2265 	 * free and return.
2266 	 */
2267 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2268 		m_freem(m);
2269 		return (NULL);
2270 	}
2271 
2272 	m->m_pkthdr.rcvif = ifp;
2273 	return (m);
2274 }
2275 
2276