xref: /freebsd/sys/net/if_lagg.c (revision 0bf48626aaa33768078f5872b922b1487b3a9296)
1 /*	$OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $	*/
2 
3 /*
4  * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org>
5  * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org>
6  * Copyright (c) 2014, 2016 Marcelo Araujo <araujo@FreeBSD.org>
7  *
8  * Permission to use, copy, modify, and distribute this software for any
9  * purpose with or without fee is hereby granted, provided that the above
10  * copyright notice and this permission notice appear in all copies.
11  *
12  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19  */
20 
21 #include <sys/cdefs.h>
22 __FBSDID("$FreeBSD$");
23 
24 #include "opt_inet.h"
25 #include "opt_inet6.h"
26 #include "opt_ratelimit.h"
27 
28 #include <sys/param.h>
29 #include <sys/kernel.h>
30 #include <sys/malloc.h>
31 #include <sys/mbuf.h>
32 #include <sys/queue.h>
33 #include <sys/socket.h>
34 #include <sys/sockio.h>
35 #include <sys/sysctl.h>
36 #include <sys/module.h>
37 #include <sys/priv.h>
38 #include <sys/systm.h>
39 #include <sys/proc.h>
40 #include <sys/lock.h>
41 #include <sys/rmlock.h>
42 #include <sys/sx.h>
43 #include <sys/taskqueue.h>
44 #include <sys/eventhandler.h>
45 
46 #include <net/ethernet.h>
47 #include <net/if.h>
48 #include <net/if_clone.h>
49 #include <net/if_arp.h>
50 #include <net/if_dl.h>
51 #include <net/if_media.h>
52 #include <net/if_types.h>
53 #include <net/if_var.h>
54 #include <net/bpf.h>
55 #include <net/vnet.h>
56 
57 #if defined(INET) || defined(INET6)
58 #include <netinet/in.h>
59 #include <netinet/ip.h>
60 #endif
61 #ifdef INET
62 #include <netinet/in_systm.h>
63 #include <netinet/if_ether.h>
64 #endif
65 
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #include <netinet6/in6_var.h>
69 #include <netinet6/in6_ifattach.h>
70 #endif
71 
72 #include <net/if_vlan_var.h>
73 #include <net/if_lagg.h>
74 #include <net/ieee8023ad_lacp.h>
75 
76 #define	LAGG_RLOCK()	struct epoch_tracker lagg_et; epoch_enter_preempt(net_epoch_preempt, &lagg_et)
77 #define	LAGG_RUNLOCK()	epoch_exit_preempt(net_epoch_preempt, &lagg_et)
78 #define	LAGG_RLOCK_ASSERT()	MPASS(in_epoch(net_epoch_preempt))
79 #define	LAGG_UNLOCK_ASSERT()	MPASS(!in_epoch(net_epoch_preempt))
80 
81 #define	LAGG_SX_INIT(_sc)	sx_init(&(_sc)->sc_sx, "if_lagg sx")
82 #define	LAGG_SX_DESTROY(_sc)	sx_destroy(&(_sc)->sc_sx)
83 #define	LAGG_XLOCK(_sc)		sx_xlock(&(_sc)->sc_sx)
84 #define	LAGG_XUNLOCK(_sc)	sx_xunlock(&(_sc)->sc_sx)
85 #define	LAGG_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
86 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
87 
88 /* Special flags we should propagate to the lagg ports. */
89 static struct {
90 	int flag;
91 	int (*func)(struct ifnet *, int);
92 } lagg_pflags[] = {
93 	{IFF_PROMISC, ifpromisc},
94 	{IFF_ALLMULTI, if_allmulti},
95 	{0, NULL}
96 };
97 
98 struct lagg_snd_tag {
99 	struct m_snd_tag com;
100 	struct m_snd_tag *tag;
101 };
102 
103 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
104 #define	V_lagg_list	VNET(lagg_list)
105 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
106 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
107 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
108 					"if_lagg list", NULL, MTX_DEF)
109 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
110 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
111 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
112 eventhandler_tag	lagg_detach_cookie = NULL;
113 
114 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
115 static void	lagg_clone_destroy(struct ifnet *);
116 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
117 #define	V_lagg_cloner	VNET(lagg_cloner)
118 static const char laggname[] = "lagg";
119 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
120 
121 static void	lagg_capabilities(struct lagg_softc *);
122 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
123 static int	lagg_port_destroy(struct lagg_port *, int);
124 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
125 static void	lagg_linkstate(struct lagg_softc *);
126 static void	lagg_port_state(struct ifnet *, int);
127 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
128 static int	lagg_port_output(struct ifnet *, struct mbuf *,
129 		    const struct sockaddr *, struct route *);
130 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
131 #ifdef LAGG_PORT_STACKING
132 static int	lagg_port_checkstacking(struct lagg_softc *);
133 #endif
134 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
135 static void	lagg_init(void *);
136 static void	lagg_stop(struct lagg_softc *);
137 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
138 #ifdef RATELIMIT
139 static int	lagg_snd_tag_alloc(struct ifnet *,
140 		    union if_snd_tag_alloc_params *,
141 		    struct m_snd_tag **);
142 static int	lagg_snd_tag_modify(struct m_snd_tag *,
143 		    union if_snd_tag_modify_params *);
144 static int	lagg_snd_tag_query(struct m_snd_tag *,
145 		    union if_snd_tag_query_params *);
146 static void	lagg_snd_tag_free(struct m_snd_tag *);
147 static void     lagg_ratelimit_query(struct ifnet *,
148 		    struct if_ratelimit_query_results *);
149 #endif
150 static int	lagg_setmulti(struct lagg_port *);
151 static int	lagg_clrmulti(struct lagg_port *);
152 static	int	lagg_setcaps(struct lagg_port *, int cap);
153 static	int	lagg_setflag(struct lagg_port *, int, int,
154 		    int (*func)(struct ifnet *, int));
155 static	int	lagg_setflags(struct lagg_port *, int status);
156 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
157 static int	lagg_transmit(struct ifnet *, struct mbuf *);
158 static void	lagg_qflush(struct ifnet *);
159 static int	lagg_media_change(struct ifnet *);
160 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
161 static struct lagg_port *lagg_link_active(struct lagg_softc *,
162 	    struct lagg_port *);
163 
164 /* Simple round robin */
165 static void	lagg_rr_attach(struct lagg_softc *);
166 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
167 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
168 		    struct mbuf *);
169 
170 /* Active failover */
171 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
172 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
173 		    struct mbuf *);
174 
175 /* Loadbalancing */
176 static void	lagg_lb_attach(struct lagg_softc *);
177 static void	lagg_lb_detach(struct lagg_softc *);
178 static int	lagg_lb_port_create(struct lagg_port *);
179 static void	lagg_lb_port_destroy(struct lagg_port *);
180 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
181 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
182 		    struct mbuf *);
183 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
184 
185 /* Broadcast */
186 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
187 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
188 		    struct mbuf *);
189 
190 /* 802.3ad LACP */
191 static void	lagg_lacp_attach(struct lagg_softc *);
192 static void	lagg_lacp_detach(struct lagg_softc *);
193 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
194 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
195 		    struct mbuf *);
196 static void	lagg_lacp_lladdr(struct lagg_softc *);
197 
198 /* lagg protocol table */
199 static const struct lagg_proto {
200 	lagg_proto	pr_num;
201 	void		(*pr_attach)(struct lagg_softc *);
202 	void		(*pr_detach)(struct lagg_softc *);
203 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
204 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
205 			    struct mbuf *);
206 	int		(*pr_addport)(struct lagg_port *);
207 	void		(*pr_delport)(struct lagg_port *);
208 	void		(*pr_linkstate)(struct lagg_port *);
209 	void 		(*pr_init)(struct lagg_softc *);
210 	void 		(*pr_stop)(struct lagg_softc *);
211 	void 		(*pr_lladdr)(struct lagg_softc *);
212 	void		(*pr_request)(struct lagg_softc *, void *);
213 	void		(*pr_portreq)(struct lagg_port *, void *);
214 } lagg_protos[] = {
215     {
216 	.pr_num = LAGG_PROTO_NONE
217     },
218     {
219 	.pr_num = LAGG_PROTO_ROUNDROBIN,
220 	.pr_attach = lagg_rr_attach,
221 	.pr_start = lagg_rr_start,
222 	.pr_input = lagg_rr_input,
223     },
224     {
225 	.pr_num = LAGG_PROTO_FAILOVER,
226 	.pr_start = lagg_fail_start,
227 	.pr_input = lagg_fail_input,
228     },
229     {
230 	.pr_num = LAGG_PROTO_LOADBALANCE,
231 	.pr_attach = lagg_lb_attach,
232 	.pr_detach = lagg_lb_detach,
233 	.pr_start = lagg_lb_start,
234 	.pr_input = lagg_lb_input,
235 	.pr_addport = lagg_lb_port_create,
236 	.pr_delport = lagg_lb_port_destroy,
237     },
238     {
239 	.pr_num = LAGG_PROTO_LACP,
240 	.pr_attach = lagg_lacp_attach,
241 	.pr_detach = lagg_lacp_detach,
242 	.pr_start = lagg_lacp_start,
243 	.pr_input = lagg_lacp_input,
244 	.pr_addport = lacp_port_create,
245 	.pr_delport = lacp_port_destroy,
246 	.pr_linkstate = lacp_linkstate,
247 	.pr_init = lacp_init,
248 	.pr_stop = lacp_stop,
249 	.pr_lladdr = lagg_lacp_lladdr,
250 	.pr_request = lacp_req,
251 	.pr_portreq = lacp_portreq,
252     },
253     {
254 	.pr_num = LAGG_PROTO_BROADCAST,
255 	.pr_start = lagg_bcast_start,
256 	.pr_input = lagg_bcast_input,
257     },
258 };
259 
260 SYSCTL_DECL(_net_link);
261 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
262     "Link Aggregation");
263 
264 /* Allow input on any failover links */
265 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
266 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
267 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
268     &VNET_NAME(lagg_failover_rx_all), 0,
269     "Accept input from any interface in a failover lagg");
270 
271 /* Default value for using flowid */
272 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
273 #define	V_def_use_flowid	VNET(def_use_flowid)
274 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
275     &VNET_NAME(def_use_flowid), 0,
276     "Default setting for using flow id for load sharing");
277 
278 /* Default value for using numa */
279 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
280 #define	V_def_use_numa	VNET(def_use_numa)
281 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN,
282     &VNET_NAME(def_use_numa), 0,
283     "Use numa to steer flows");
284 
285 /* Default value for flowid shift */
286 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
287 #define	V_def_flowid_shift	VNET(def_flowid_shift)
288 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
289     &VNET_NAME(def_flowid_shift), 0,
290     "Default setting for flowid shift for load sharing");
291 
292 static void
293 vnet_lagg_init(const void *unused __unused)
294 {
295 
296 	LAGG_LIST_LOCK_INIT();
297 	SLIST_INIT(&V_lagg_list);
298 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
299 	    lagg_clone_destroy, 0);
300 }
301 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
302     vnet_lagg_init, NULL);
303 
304 static void
305 vnet_lagg_uninit(const void *unused __unused)
306 {
307 
308 	if_clone_detach(V_lagg_cloner);
309 	LAGG_LIST_LOCK_DESTROY();
310 }
311 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
312     vnet_lagg_uninit, NULL);
313 
314 static int
315 lagg_modevent(module_t mod, int type, void *data)
316 {
317 
318 	switch (type) {
319 	case MOD_LOAD:
320 		lagg_input_p = lagg_input;
321 		lagg_linkstate_p = lagg_port_state;
322 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
323 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
324 		    EVENTHANDLER_PRI_ANY);
325 		break;
326 	case MOD_UNLOAD:
327 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
328 		    lagg_detach_cookie);
329 		lagg_input_p = NULL;
330 		lagg_linkstate_p = NULL;
331 		break;
332 	default:
333 		return (EOPNOTSUPP);
334 	}
335 	return (0);
336 }
337 
338 static moduledata_t lagg_mod = {
339 	"if_lagg",
340 	lagg_modevent,
341 	0
342 };
343 
344 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
345 MODULE_VERSION(if_lagg, 1);
346 
347 static void
348 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
349 {
350 
351 	LAGG_XLOCK_ASSERT(sc);
352 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
353 	    __func__, sc));
354 
355 	if (sc->sc_ifflags & IFF_DEBUG)
356 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
357 
358 	if (lagg_protos[pr].pr_attach != NULL)
359 		lagg_protos[pr].pr_attach(sc);
360 	sc->sc_proto = pr;
361 }
362 
363 static void
364 lagg_proto_detach(struct lagg_softc *sc)
365 {
366 	lagg_proto pr;
367 
368 	LAGG_XLOCK_ASSERT(sc);
369 	pr = sc->sc_proto;
370 	sc->sc_proto = LAGG_PROTO_NONE;
371 
372 	if (lagg_protos[pr].pr_detach != NULL)
373 		lagg_protos[pr].pr_detach(sc);
374 }
375 
376 static int
377 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
378 {
379 
380 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
381 }
382 
383 static struct mbuf *
384 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
385 {
386 
387 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
388 }
389 
390 static int
391 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
392 {
393 
394 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
395 		return (0);
396 	else
397 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
398 }
399 
400 static void
401 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
402 {
403 
404 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
405 		lagg_protos[sc->sc_proto].pr_delport(lp);
406 }
407 
408 static void
409 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
410 {
411 
412 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
413 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
414 }
415 
416 static void
417 lagg_proto_init(struct lagg_softc *sc)
418 {
419 
420 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
421 		lagg_protos[sc->sc_proto].pr_init(sc);
422 }
423 
424 static void
425 lagg_proto_stop(struct lagg_softc *sc)
426 {
427 
428 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
429 		lagg_protos[sc->sc_proto].pr_stop(sc);
430 }
431 
432 static void
433 lagg_proto_lladdr(struct lagg_softc *sc)
434 {
435 
436 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
437 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
438 }
439 
440 static void
441 lagg_proto_request(struct lagg_softc *sc, void *v)
442 {
443 
444 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
445 		lagg_protos[sc->sc_proto].pr_request(sc, v);
446 }
447 
448 static void
449 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
450 {
451 
452 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
453 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
454 }
455 
456 /*
457  * This routine is run via an vlan
458  * config EVENT
459  */
460 static void
461 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
462 {
463 	struct lagg_softc *sc = ifp->if_softc;
464 	struct lagg_port *lp;
465 
466 	if (ifp->if_softc !=  arg)   /* Not our event */
467 		return;
468 
469 	LAGG_RLOCK();
470 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
471 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
472 	LAGG_RUNLOCK();
473 }
474 
475 /*
476  * This routine is run via an vlan
477  * unconfig EVENT
478  */
479 static void
480 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
481 {
482 	struct lagg_softc *sc = ifp->if_softc;
483 	struct lagg_port *lp;
484 
485 	if (ifp->if_softc !=  arg)   /* Not our event */
486 		return;
487 
488 	LAGG_RLOCK();
489 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
490 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
491 	LAGG_RUNLOCK();
492 }
493 
494 static int
495 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
496 {
497 	struct lagg_softc *sc;
498 	struct ifnet *ifp;
499 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
500 
501 	sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO);
502 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
503 	if (ifp == NULL) {
504 		free(sc, M_LAGG);
505 		return (ENOSPC);
506 	}
507 	LAGG_SX_INIT(sc);
508 
509 	LAGG_XLOCK(sc);
510 	if (V_def_use_flowid)
511 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
512 	if (V_def_use_numa)
513 		sc->sc_opts |= LAGG_OPT_USE_NUMA;
514 	sc->flowid_shift = V_def_flowid_shift;
515 
516 	/* Hash all layers by default */
517 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
518 
519 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
520 
521 	CK_SLIST_INIT(&sc->sc_ports);
522 
523 	/* Initialise pseudo media types */
524 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
525 	    lagg_media_status);
526 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
527 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
528 
529 	if_initname(ifp, laggname, unit);
530 	ifp->if_softc = sc;
531 	ifp->if_transmit = lagg_transmit;
532 	ifp->if_qflush = lagg_qflush;
533 	ifp->if_init = lagg_init;
534 	ifp->if_ioctl = lagg_ioctl;
535 	ifp->if_get_counter = lagg_get_counter;
536 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
537 #ifdef RATELIMIT
538 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
539 	ifp->if_snd_tag_modify = lagg_snd_tag_modify;
540 	ifp->if_snd_tag_query = lagg_snd_tag_query;
541 	ifp->if_snd_tag_free = lagg_snd_tag_free;
542 	ifp->if_ratelimit_query = lagg_ratelimit_query;
543 #endif
544 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
545 
546 	/*
547 	 * Attach as an ordinary ethernet device, children will be attached
548 	 * as special device IFT_IEEE8023ADLAG.
549 	 */
550 	ether_ifattach(ifp, eaddr);
551 
552 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
553 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
554 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
555 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
556 
557 	/* Insert into the global list of laggs */
558 	LAGG_LIST_LOCK();
559 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
560 	LAGG_LIST_UNLOCK();
561 	LAGG_XUNLOCK(sc);
562 
563 	return (0);
564 }
565 
566 static void
567 lagg_clone_destroy(struct ifnet *ifp)
568 {
569 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
570 	struct lagg_port *lp;
571 
572 	LAGG_XLOCK(sc);
573 	sc->sc_destroying = 1;
574 	lagg_stop(sc);
575 	ifp->if_flags &= ~IFF_UP;
576 
577 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
578 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
579 
580 	/* Shutdown and remove lagg ports */
581 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
582 		lagg_port_destroy(lp, 1);
583 
584 	/* Unhook the aggregation protocol */
585 	lagg_proto_detach(sc);
586 	LAGG_XUNLOCK(sc);
587 
588 	ifmedia_removeall(&sc->sc_media);
589 	ether_ifdetach(ifp);
590 	if_free(ifp);
591 
592 	LAGG_LIST_LOCK();
593 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
594 	LAGG_LIST_UNLOCK();
595 
596 	LAGG_SX_DESTROY(sc);
597 	free(sc, M_LAGG);
598 }
599 
600 static void
601 lagg_capabilities(struct lagg_softc *sc)
602 {
603 	struct lagg_port *lp;
604 	int cap, ena, pena;
605 	uint64_t hwa;
606 	struct ifnet_hw_tsomax hw_tsomax;
607 
608 	LAGG_XLOCK_ASSERT(sc);
609 
610 	/* Get common enabled capabilities for the lagg ports */
611 	ena = ~0;
612 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
613 		ena &= lp->lp_ifp->if_capenable;
614 	ena = (ena == ~0 ? 0 : ena);
615 
616 	/*
617 	 * Apply common enabled capabilities back to the lagg ports.
618 	 * May require several iterations if they are dependent.
619 	 */
620 	do {
621 		pena = ena;
622 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
623 			lagg_setcaps(lp, ena);
624 			ena &= lp->lp_ifp->if_capenable;
625 		}
626 	} while (pena != ena);
627 
628 	/* Get other capabilities from the lagg ports */
629 	cap = ~0;
630 	hwa = ~(uint64_t)0;
631 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
632 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
633 		cap &= lp->lp_ifp->if_capabilities;
634 		hwa &= lp->lp_ifp->if_hwassist;
635 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
636 	}
637 	cap = (cap == ~0 ? 0 : cap);
638 	hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
639 
640 	if (sc->sc_ifp->if_capabilities != cap ||
641 	    sc->sc_ifp->if_capenable != ena ||
642 	    sc->sc_ifp->if_hwassist != hwa ||
643 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
644 		sc->sc_ifp->if_capabilities = cap;
645 		sc->sc_ifp->if_capenable = ena;
646 		sc->sc_ifp->if_hwassist = hwa;
647 		getmicrotime(&sc->sc_ifp->if_lastchange);
648 
649 		if (sc->sc_ifflags & IFF_DEBUG)
650 			if_printf(sc->sc_ifp,
651 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
652 	}
653 }
654 
655 static int
656 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
657 {
658 	struct lagg_softc *sc_ptr;
659 	struct lagg_port *lp, *tlp;
660 	struct ifreq ifr;
661 	int error, i, oldmtu;
662 	uint64_t *pval;
663 
664 	LAGG_XLOCK_ASSERT(sc);
665 
666 	if (sc->sc_ifp == ifp) {
667 		if_printf(sc->sc_ifp,
668 		    "cannot add a lagg to itself as a port\n");
669 		return (EINVAL);
670 	}
671 
672 	/* Limit the maximal number of lagg ports */
673 	if (sc->sc_count >= LAGG_MAX_PORTS)
674 		return (ENOSPC);
675 
676 	/* Check if port has already been associated to a lagg */
677 	if (ifp->if_lagg != NULL) {
678 		/* Port is already in the current lagg? */
679 		lp = (struct lagg_port *)ifp->if_lagg;
680 		if (lp->lp_softc == sc)
681 			return (EEXIST);
682 		return (EBUSY);
683 	}
684 
685 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
686 	if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
687 		return (EPROTONOSUPPORT);
688 
689 	/* Allow the first Ethernet member to define the MTU */
690 	oldmtu = -1;
691 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
692 		sc->sc_ifp->if_mtu = ifp->if_mtu;
693 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
694 		if (ifp->if_ioctl == NULL) {
695 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
696 			    ifp->if_xname);
697 			return (EINVAL);
698 		}
699 		oldmtu = ifp->if_mtu;
700 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
701 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
702 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
703 		if (error != 0) {
704 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
705 			    ifp->if_xname);
706 			return (error);
707 		}
708 		ifr.ifr_mtu = oldmtu;
709 	}
710 
711 	lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO);
712 	lp->lp_softc = sc;
713 
714 	/* Check if port is a stacked lagg */
715 	LAGG_LIST_LOCK();
716 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
717 		if (ifp == sc_ptr->sc_ifp) {
718 			LAGG_LIST_UNLOCK();
719 			free(lp, M_LAGG);
720 			if (oldmtu != -1)
721 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
722 				    (caddr_t)&ifr);
723 			return (EINVAL);
724 			/* XXX disable stacking for the moment, its untested */
725 #ifdef LAGG_PORT_STACKING
726 			lp->lp_flags |= LAGG_PORT_STACK;
727 			if (lagg_port_checkstacking(sc_ptr) >=
728 			    LAGG_MAX_STACKING) {
729 				LAGG_LIST_UNLOCK();
730 				free(lp, M_LAGG);
731 				if (oldmtu != -1)
732 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
733 					    (caddr_t)&ifr);
734 				return (E2BIG);
735 			}
736 #endif
737 		}
738 	}
739 	LAGG_LIST_UNLOCK();
740 
741 	if_ref(ifp);
742 	lp->lp_ifp = ifp;
743 
744 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
745 	lp->lp_ifcapenable = ifp->if_capenable;
746 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
747 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
748 		lagg_proto_lladdr(sc);
749 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
750 	} else {
751 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
752 	}
753 	lagg_setflags(lp, 1);
754 
755 	if (CK_SLIST_EMPTY(&sc->sc_ports))
756 		sc->sc_primary = lp;
757 
758 	/* Change the interface type */
759 	lp->lp_iftype = ifp->if_type;
760 	ifp->if_type = IFT_IEEE8023ADLAG;
761 	ifp->if_lagg = lp;
762 	lp->lp_ioctl = ifp->if_ioctl;
763 	ifp->if_ioctl = lagg_port_ioctl;
764 	lp->lp_output = ifp->if_output;
765 	ifp->if_output = lagg_port_output;
766 
767 	/* Read port counters */
768 	pval = lp->port_counters.val;
769 	for (i = 0; i < IFCOUNTERS; i++, pval++)
770 		*pval = ifp->if_get_counter(ifp, i);
771 
772 	/*
773 	 * Insert into the list of ports.
774 	 * Keep ports sorted by if_index. It is handy, when configuration
775 	 * is predictable and `ifconfig laggN create ...` command
776 	 * will lead to the same result each time.
777 	 */
778 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
779 		if (tlp->lp_ifp->if_index < ifp->if_index && (
780 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
781 		    ((struct  lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
782 		    ifp->if_index))
783 			break;
784 	}
785 	if (tlp != NULL)
786 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
787 	else
788 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
789 	sc->sc_count++;
790 
791 	lagg_setmulti(lp);
792 
793 
794 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
795 		/* Remove the port, without calling pr_delport. */
796 		lagg_port_destroy(lp, 0);
797 		if (oldmtu != -1)
798 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
799 		return (error);
800 	}
801 
802 	/* Update lagg capabilities */
803 	lagg_capabilities(sc);
804 	lagg_linkstate(sc);
805 
806 	return (0);
807 }
808 
809 #ifdef LAGG_PORT_STACKING
810 static int
811 lagg_port_checkstacking(struct lagg_softc *sc)
812 {
813 	struct lagg_softc *sc_ptr;
814 	struct lagg_port *lp;
815 	int m = 0;
816 
817 	LAGG_SXLOCK_ASSERT(sc);
818 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
819 		if (lp->lp_flags & LAGG_PORT_STACK) {
820 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
821 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
822 		}
823 	}
824 
825 	return (m + 1);
826 }
827 #endif
828 
829 static void
830 lagg_port_destroy_cb(epoch_context_t ec)
831 {
832 	struct lagg_port *lp;
833 	struct ifnet *ifp;
834 
835 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
836 	ifp = lp->lp_ifp;
837 
838 	if_rele(ifp);
839 	free(lp, M_LAGG);
840 }
841 
842 static int
843 lagg_port_destroy(struct lagg_port *lp, int rundelport)
844 {
845 	struct lagg_softc *sc = lp->lp_softc;
846 	struct lagg_port *lp_ptr, *lp0;
847 	struct ifnet *ifp = lp->lp_ifp;
848 	uint64_t *pval, vdiff;
849 	int i;
850 
851 	LAGG_XLOCK_ASSERT(sc);
852 
853 	if (rundelport)
854 		lagg_proto_delport(sc, lp);
855 
856 	if (lp->lp_detaching == 0)
857 		lagg_clrmulti(lp);
858 
859 	/* Restore interface */
860 	ifp->if_type = lp->lp_iftype;
861 	ifp->if_ioctl = lp->lp_ioctl;
862 	ifp->if_output = lp->lp_output;
863 	ifp->if_lagg = NULL;
864 
865 	/* Update detached port counters */
866 	pval = lp->port_counters.val;
867 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
868 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
869 		sc->detached_counters.val[i] += vdiff;
870 	}
871 
872 	/* Finally, remove the port from the lagg */
873 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
874 	sc->sc_count--;
875 
876 	/* Update the primary interface */
877 	if (lp == sc->sc_primary) {
878 		uint8_t lladdr[ETHER_ADDR_LEN];
879 
880 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
881 			bzero(&lladdr, ETHER_ADDR_LEN);
882 		else
883 			bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN);
884 		sc->sc_primary = lp0;
885 		if (sc->sc_destroying == 0) {
886 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
887 			lagg_proto_lladdr(sc);
888 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
889 		}
890 
891 		/*
892 		 * Update lladdr for each port (new primary needs update
893 		 * as well, to switch from old lladdr to its 'real' one)
894 		 */
895 		CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
896 			if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN);
897 	}
898 
899 	if (lp->lp_ifflags)
900 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
901 
902 	if (lp->lp_detaching == 0) {
903 		lagg_setflags(lp, 0);
904 		lagg_setcaps(lp, lp->lp_ifcapenable);
905 		if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN);
906 	}
907 
908 	/*
909 	 * free port and release it's ifnet reference after a grace period has
910 	 * elapsed.
911 	 */
912 	epoch_call(net_epoch_preempt, &lp->lp_epoch_ctx, lagg_port_destroy_cb);
913 	/* Update lagg capabilities */
914 	lagg_capabilities(sc);
915 	lagg_linkstate(sc);
916 
917 	return (0);
918 }
919 
920 static int
921 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
922 {
923 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
924 	struct lagg_softc *sc;
925 	struct lagg_port *lp = NULL;
926 	int error = 0;
927 
928 	/* Should be checked by the caller */
929 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
930 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
931 		goto fallback;
932 
933 	switch (cmd) {
934 	case SIOCGLAGGPORT:
935 		if (rp->rp_portname[0] == '\0' ||
936 		    ifunit(rp->rp_portname) != ifp) {
937 			error = EINVAL;
938 			break;
939 		}
940 
941 		LAGG_RLOCK();
942 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
943 			error = ENOENT;
944 			LAGG_RUNLOCK();
945 			break;
946 		}
947 
948 		lagg_port2req(lp, rp);
949 		LAGG_RUNLOCK();
950 		break;
951 
952 	case SIOCSIFCAP:
953 		if (lp->lp_ioctl == NULL) {
954 			error = EINVAL;
955 			break;
956 		}
957 		error = (*lp->lp_ioctl)(ifp, cmd, data);
958 		if (error)
959 			break;
960 
961 		/* Update lagg interface capabilities */
962 		LAGG_XLOCK(sc);
963 		lagg_capabilities(sc);
964 		LAGG_XUNLOCK(sc);
965 		VLAN_CAPABILITIES(sc->sc_ifp);
966 		break;
967 
968 	case SIOCSIFMTU:
969 		/* Do not allow the MTU to be changed once joined */
970 		error = EINVAL;
971 		break;
972 
973 	default:
974 		goto fallback;
975 	}
976 
977 	return (error);
978 
979 fallback:
980 	if (lp != NULL && lp->lp_ioctl != NULL)
981 		return ((*lp->lp_ioctl)(ifp, cmd, data));
982 
983 	return (EINVAL);
984 }
985 
986 /*
987  * Requests counter @cnt data.
988  *
989  * Counter value is calculated the following way:
990  * 1) for each port, sum  difference between current and "initial" measurements.
991  * 2) add lagg logical interface counters.
992  * 3) add data from detached_counters array.
993  *
994  * We also do the following things on ports attach/detach:
995  * 1) On port attach we store all counters it has into port_counter array.
996  * 2) On port detach we add the different between "initial" and
997  *   current counters data to detached_counters array.
998  */
999 static uint64_t
1000 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1001 {
1002 	struct lagg_softc *sc;
1003 	struct lagg_port *lp;
1004 	struct ifnet *lpifp;
1005 	uint64_t newval, oldval, vsum;
1006 
1007 	/* Revise this when we've got non-generic counters. */
1008 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1009 
1010 	sc = (struct lagg_softc *)ifp->if_softc;
1011 
1012 	vsum = 0;
1013 	LAGG_RLOCK();
1014 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1015 		/* Saved attached value */
1016 		oldval = lp->port_counters.val[cnt];
1017 		/* current value */
1018 		lpifp = lp->lp_ifp;
1019 		newval = lpifp->if_get_counter(lpifp, cnt);
1020 		/* Calculate diff and save new */
1021 		vsum += newval - oldval;
1022 	}
1023 	LAGG_RUNLOCK();
1024 
1025 	/*
1026 	 * Add counter data which might be added by upper
1027 	 * layer protocols operating on logical interface.
1028 	 */
1029 	vsum += if_get_counter_default(ifp, cnt);
1030 
1031 	/*
1032 	 * Add counter data from detached ports counters
1033 	 */
1034 	vsum += sc->detached_counters.val[cnt];
1035 
1036 
1037 	return (vsum);
1038 }
1039 
1040 /*
1041  * For direct output to child ports.
1042  */
1043 static int
1044 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1045 	const struct sockaddr *dst, struct route *ro)
1046 {
1047 	struct lagg_port *lp = ifp->if_lagg;
1048 
1049 	switch (dst->sa_family) {
1050 		case pseudo_AF_HDRCMPLT:
1051 		case AF_UNSPEC:
1052 			return ((*lp->lp_output)(ifp, m, dst, ro));
1053 	}
1054 
1055 	/* drop any other frames */
1056 	m_freem(m);
1057 	return (ENETDOWN);
1058 }
1059 
1060 static void
1061 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1062 {
1063 	struct lagg_port *lp;
1064 	struct lagg_softc *sc;
1065 
1066 	if ((lp = ifp->if_lagg) == NULL)
1067 		return;
1068 	/* If the ifnet is just being renamed, don't do anything. */
1069 	if (ifp->if_flags & IFF_RENAMING)
1070 		return;
1071 
1072 	sc = lp->lp_softc;
1073 
1074 	LAGG_XLOCK(sc);
1075 	lp->lp_detaching = 1;
1076 	lagg_port_destroy(lp, 1);
1077 	LAGG_XUNLOCK(sc);
1078 	VLAN_CAPABILITIES(sc->sc_ifp);
1079 }
1080 
1081 static void
1082 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1083 {
1084 	struct lagg_softc *sc = lp->lp_softc;
1085 
1086 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1087 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1088 	rp->rp_prio = lp->lp_prio;
1089 	rp->rp_flags = lp->lp_flags;
1090 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1091 
1092 	/* Add protocol specific flags */
1093 	switch (sc->sc_proto) {
1094 		case LAGG_PROTO_FAILOVER:
1095 			if (lp == sc->sc_primary)
1096 				rp->rp_flags |= LAGG_PORT_MASTER;
1097 			if (lp == lagg_link_active(sc, sc->sc_primary))
1098 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1099 			break;
1100 
1101 		case LAGG_PROTO_ROUNDROBIN:
1102 		case LAGG_PROTO_LOADBALANCE:
1103 		case LAGG_PROTO_BROADCAST:
1104 			if (LAGG_PORTACTIVE(lp))
1105 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1106 			break;
1107 
1108 		case LAGG_PROTO_LACP:
1109 			/* LACP has a different definition of active */
1110 			if (lacp_isactive(lp))
1111 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1112 			if (lacp_iscollecting(lp))
1113 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1114 			if (lacp_isdistributing(lp))
1115 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1116 			break;
1117 	}
1118 
1119 }
1120 
1121 static void
1122 lagg_init(void *xsc)
1123 {
1124 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1125 	struct ifnet *ifp = sc->sc_ifp;
1126 	struct lagg_port *lp;
1127 
1128 	LAGG_XLOCK(sc);
1129 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1130 		LAGG_XUNLOCK(sc);
1131 		return;
1132 	}
1133 
1134 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1135 
1136 	/*
1137 	 * Update the port lladdrs if needed.
1138 	 * This might be if_setlladdr() notification
1139 	 * that lladdr has been changed.
1140 	 */
1141 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1142 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1143 		    ETHER_ADDR_LEN) != 0)
1144 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN);
1145 	}
1146 
1147 	lagg_proto_init(sc);
1148 
1149 	LAGG_XUNLOCK(sc);
1150 }
1151 
1152 static void
1153 lagg_stop(struct lagg_softc *sc)
1154 {
1155 	struct ifnet *ifp = sc->sc_ifp;
1156 
1157 	LAGG_XLOCK_ASSERT(sc);
1158 
1159 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1160 		return;
1161 
1162 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1163 
1164 	lagg_proto_stop(sc);
1165 }
1166 
1167 static int
1168 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1169 {
1170 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1171 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1172 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1173 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1174 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1175 	struct ifreq *ifr = (struct ifreq *)data;
1176 	struct lagg_port *lp;
1177 	struct ifnet *tpif;
1178 	struct thread *td = curthread;
1179 	char *buf, *outbuf;
1180 	int count, buflen, len, error = 0;
1181 
1182 	bzero(&rpbuf, sizeof(rpbuf));
1183 
1184 	switch (cmd) {
1185 	case SIOCGLAGG:
1186 		LAGG_XLOCK(sc);
1187 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1188 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1189 		ra->ra_proto = sc->sc_proto;
1190 		lagg_proto_request(sc, &ra->ra_psc);
1191 		count = 0;
1192 		buf = outbuf;
1193 		len = min(ra->ra_size, buflen);
1194 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1195 			if (len < sizeof(rpbuf))
1196 				break;
1197 
1198 			lagg_port2req(lp, &rpbuf);
1199 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1200 			count++;
1201 			buf += sizeof(rpbuf);
1202 			len -= sizeof(rpbuf);
1203 		}
1204 		LAGG_XUNLOCK(sc);
1205 		ra->ra_ports = count;
1206 		ra->ra_size = count * sizeof(rpbuf);
1207 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1208 		free(outbuf, M_TEMP);
1209 		break;
1210 	case SIOCSLAGG:
1211 		error = priv_check(td, PRIV_NET_LAGG);
1212 		if (error)
1213 			break;
1214 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1215 			error = EPROTONOSUPPORT;
1216 			break;
1217 		}
1218 
1219 		LAGG_XLOCK(sc);
1220 		lagg_proto_detach(sc);
1221 		LAGG_UNLOCK_ASSERT();
1222 		lagg_proto_attach(sc, ra->ra_proto);
1223 		LAGG_XUNLOCK(sc);
1224 		break;
1225 	case SIOCGLAGGOPTS:
1226 		LAGG_XLOCK(sc);
1227 		ro->ro_opts = sc->sc_opts;
1228 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1229 			struct lacp_softc *lsc;
1230 
1231 			lsc = (struct lacp_softc *)sc->sc_psc;
1232 			if (lsc->lsc_debug.lsc_tx_test != 0)
1233 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1234 			if (lsc->lsc_debug.lsc_rx_test != 0)
1235 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1236 			if (lsc->lsc_strict_mode != 0)
1237 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1238 			if (lsc->lsc_fast_timeout != 0)
1239 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1240 
1241 			ro->ro_active = sc->sc_active;
1242 		} else {
1243 			ro->ro_active = 0;
1244 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1245 				ro->ro_active += LAGG_PORTACTIVE(lp);
1246 		}
1247 		ro->ro_bkt = sc->sc_bkt;
1248 		ro->ro_flapping = sc->sc_flapping;
1249 		ro->ro_flowid_shift = sc->flowid_shift;
1250 		LAGG_XUNLOCK(sc);
1251 		break;
1252 	case SIOCSLAGGOPTS:
1253 		if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) {
1254 			if (ro->ro_bkt == 0)
1255 				sc->sc_bkt = 1; // Minimum 1 packet per iface.
1256 			else
1257 				sc->sc_bkt = ro->ro_bkt;
1258 		}
1259 		error = priv_check(td, PRIV_NET_LAGG);
1260 		if (error)
1261 			break;
1262 		if (ro->ro_opts == 0)
1263 			break;
1264 		/*
1265 		 * Set options.  LACP options are stored in sc->sc_psc,
1266 		 * not in sc_opts.
1267 		 */
1268 		int valid, lacp;
1269 
1270 		switch (ro->ro_opts) {
1271 		case LAGG_OPT_USE_FLOWID:
1272 		case -LAGG_OPT_USE_FLOWID:
1273 		case LAGG_OPT_USE_NUMA:
1274 		case -LAGG_OPT_USE_NUMA:
1275 		case LAGG_OPT_FLOWIDSHIFT:
1276 			valid = 1;
1277 			lacp = 0;
1278 			break;
1279 		case LAGG_OPT_LACP_TXTEST:
1280 		case -LAGG_OPT_LACP_TXTEST:
1281 		case LAGG_OPT_LACP_RXTEST:
1282 		case -LAGG_OPT_LACP_RXTEST:
1283 		case LAGG_OPT_LACP_STRICT:
1284 		case -LAGG_OPT_LACP_STRICT:
1285 		case LAGG_OPT_LACP_TIMEOUT:
1286 		case -LAGG_OPT_LACP_TIMEOUT:
1287 			valid = lacp = 1;
1288 			break;
1289 		default:
1290 			valid = lacp = 0;
1291 			break;
1292 		}
1293 
1294 		LAGG_XLOCK(sc);
1295 
1296 		if (valid == 0 ||
1297 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1298 			/* Invalid combination of options specified. */
1299 			error = EINVAL;
1300 			LAGG_XUNLOCK(sc);
1301 			break;	/* Return from SIOCSLAGGOPTS. */
1302 		}
1303 		/*
1304 		 * Store new options into sc->sc_opts except for
1305 		 * FLOWIDSHIFT and LACP options.
1306 		 */
1307 		if (lacp == 0) {
1308 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1309 				sc->flowid_shift = ro->ro_flowid_shift;
1310 			else if (ro->ro_opts > 0)
1311 				sc->sc_opts |= ro->ro_opts;
1312 			else
1313 				sc->sc_opts &= ~ro->ro_opts;
1314 		} else {
1315 			struct lacp_softc *lsc;
1316 			struct lacp_port *lp;
1317 
1318 			lsc = (struct lacp_softc *)sc->sc_psc;
1319 
1320 			switch (ro->ro_opts) {
1321 			case LAGG_OPT_LACP_TXTEST:
1322 				lsc->lsc_debug.lsc_tx_test = 1;
1323 				break;
1324 			case -LAGG_OPT_LACP_TXTEST:
1325 				lsc->lsc_debug.lsc_tx_test = 0;
1326 				break;
1327 			case LAGG_OPT_LACP_RXTEST:
1328 				lsc->lsc_debug.lsc_rx_test = 1;
1329 				break;
1330 			case -LAGG_OPT_LACP_RXTEST:
1331 				lsc->lsc_debug.lsc_rx_test = 0;
1332 				break;
1333 			case LAGG_OPT_LACP_STRICT:
1334 				lsc->lsc_strict_mode = 1;
1335 				break;
1336 			case -LAGG_OPT_LACP_STRICT:
1337 				lsc->lsc_strict_mode = 0;
1338 				break;
1339 			case LAGG_OPT_LACP_TIMEOUT:
1340 				LACP_LOCK(lsc);
1341         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1342                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1343 				LACP_UNLOCK(lsc);
1344 				lsc->lsc_fast_timeout = 1;
1345 				break;
1346 			case -LAGG_OPT_LACP_TIMEOUT:
1347 				LACP_LOCK(lsc);
1348         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1349                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1350 				LACP_UNLOCK(lsc);
1351 				lsc->lsc_fast_timeout = 0;
1352 				break;
1353 			}
1354 		}
1355 		LAGG_XUNLOCK(sc);
1356 		break;
1357 	case SIOCGLAGGFLAGS:
1358 		rf->rf_flags = 0;
1359 		LAGG_XLOCK(sc);
1360 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1361 			rf->rf_flags |= LAGG_F_HASHL2;
1362 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1363 			rf->rf_flags |= LAGG_F_HASHL3;
1364 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1365 			rf->rf_flags |= LAGG_F_HASHL4;
1366 		LAGG_XUNLOCK(sc);
1367 		break;
1368 	case SIOCSLAGGHASH:
1369 		error = priv_check(td, PRIV_NET_LAGG);
1370 		if (error)
1371 			break;
1372 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1373 			error = EINVAL;
1374 			break;
1375 		}
1376 		LAGG_XLOCK(sc);
1377 		sc->sc_flags = 0;
1378 		if (rf->rf_flags & LAGG_F_HASHL2)
1379 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1380 		if (rf->rf_flags & LAGG_F_HASHL3)
1381 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1382 		if (rf->rf_flags & LAGG_F_HASHL4)
1383 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1384 		LAGG_XUNLOCK(sc);
1385 		break;
1386 	case SIOCGLAGGPORT:
1387 		if (rp->rp_portname[0] == '\0' ||
1388 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1389 			error = EINVAL;
1390 			break;
1391 		}
1392 
1393 		LAGG_RLOCK();
1394 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1395 		    lp->lp_softc != sc) {
1396 			error = ENOENT;
1397 			LAGG_RUNLOCK();
1398 			if_rele(tpif);
1399 			break;
1400 		}
1401 
1402 		lagg_port2req(lp, rp);
1403 		LAGG_RUNLOCK();
1404 		if_rele(tpif);
1405 		break;
1406 	case SIOCSLAGGPORT:
1407 		error = priv_check(td, PRIV_NET_LAGG);
1408 		if (error)
1409 			break;
1410 		if (rp->rp_portname[0] == '\0' ||
1411 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1412 			error = EINVAL;
1413 			break;
1414 		}
1415 #ifdef INET6
1416 		/*
1417 		 * A laggport interface should not have inet6 address
1418 		 * because two interfaces with a valid link-local
1419 		 * scope zone must not be merged in any form.  This
1420 		 * restriction is needed to prevent violation of
1421 		 * link-local scope zone.  Attempts to add a laggport
1422 		 * interface which has inet6 addresses triggers
1423 		 * removal of all inet6 addresses on the member
1424 		 * interface.
1425 		 */
1426 		if (in6ifa_llaonifp(tpif)) {
1427 			in6_ifdetach(tpif);
1428 				if_printf(sc->sc_ifp,
1429 				    "IPv6 addresses on %s have been removed "
1430 				    "before adding it as a member to prevent "
1431 				    "IPv6 address scope violation.\n",
1432 				    tpif->if_xname);
1433 		}
1434 #endif
1435 		LAGG_XLOCK(sc);
1436 		error = lagg_port_create(sc, tpif);
1437 		LAGG_XUNLOCK(sc);
1438 		if_rele(tpif);
1439 		VLAN_CAPABILITIES(ifp);
1440 		break;
1441 	case SIOCSLAGGDELPORT:
1442 		error = priv_check(td, PRIV_NET_LAGG);
1443 		if (error)
1444 			break;
1445 		if (rp->rp_portname[0] == '\0' ||
1446 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1447 			error = EINVAL;
1448 			break;
1449 		}
1450 
1451 		LAGG_XLOCK(sc);
1452 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1453 		    lp->lp_softc != sc) {
1454 			error = ENOENT;
1455 			LAGG_XUNLOCK(sc);
1456 			if_rele(tpif);
1457 			break;
1458 		}
1459 
1460 		error = lagg_port_destroy(lp, 1);
1461 		LAGG_XUNLOCK(sc);
1462 		if_rele(tpif);
1463 		VLAN_CAPABILITIES(ifp);
1464 		break;
1465 	case SIOCSIFFLAGS:
1466 		/* Set flags on ports too */
1467 		LAGG_XLOCK(sc);
1468 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1469 			lagg_setflags(lp, 1);
1470 		}
1471 
1472 		if (!(ifp->if_flags & IFF_UP) &&
1473 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1474 			/*
1475 			 * If interface is marked down and it is running,
1476 			 * then stop and disable it.
1477 			 */
1478 			lagg_stop(sc);
1479 			LAGG_XUNLOCK(sc);
1480 		} else if ((ifp->if_flags & IFF_UP) &&
1481 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1482 			/*
1483 			 * If interface is marked up and it is stopped, then
1484 			 * start it.
1485 			 */
1486 			LAGG_XUNLOCK(sc);
1487 			(*ifp->if_init)(sc);
1488 		} else
1489 			LAGG_XUNLOCK(sc);
1490 		break;
1491 	case SIOCADDMULTI:
1492 	case SIOCDELMULTI:
1493 		LAGG_XLOCK(sc);
1494 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1495 			lagg_clrmulti(lp);
1496 			lagg_setmulti(lp);
1497 		}
1498 		LAGG_XUNLOCK(sc);
1499 		error = 0;
1500 		break;
1501 	case SIOCSIFMEDIA:
1502 	case SIOCGIFMEDIA:
1503 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1504 		break;
1505 
1506 	case SIOCSIFCAP:
1507 		LAGG_XLOCK(sc);
1508 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1509 			if (lp->lp_ioctl != NULL)
1510 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1511 		}
1512 		lagg_capabilities(sc);
1513 		LAGG_XUNLOCK(sc);
1514 		VLAN_CAPABILITIES(ifp);
1515 		error = 0;
1516 		break;
1517 
1518 	case SIOCSIFMTU:
1519 		LAGG_XLOCK(sc);
1520 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1521 			if (lp->lp_ioctl != NULL)
1522 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1523 			else
1524 				error = EINVAL;
1525 			if (error != 0) {
1526 				if_printf(ifp,
1527 				    "failed to change MTU to %d on port %s, "
1528 				    "reverting all ports to original MTU (%d)\n",
1529 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1530 				break;
1531 			}
1532 		}
1533 		if (error == 0) {
1534 			ifp->if_mtu = ifr->ifr_mtu;
1535 		} else {
1536 			/* set every port back to the original MTU */
1537 			ifr->ifr_mtu = ifp->if_mtu;
1538 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1539 				if (lp->lp_ioctl != NULL)
1540 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1541 			}
1542 		}
1543 		LAGG_XUNLOCK(sc);
1544 		break;
1545 
1546 	default:
1547 		error = ether_ioctl(ifp, cmd, data);
1548 		break;
1549 	}
1550 	return (error);
1551 }
1552 
1553 #ifdef RATELIMIT
1554 static inline struct lagg_snd_tag *
1555 mst_to_lst(struct m_snd_tag *mst)
1556 {
1557 
1558 	return (__containerof(mst, struct lagg_snd_tag, com));
1559 }
1560 
1561 /*
1562  * Look up the port used by a specific flow.  This only works for lagg
1563  * protocols with deterministic port mappings (e.g. not roundrobin).
1564  * In addition protocols which use a hash to map flows to ports must
1565  * be configured to use the mbuf flowid rather than hashing packet
1566  * contents.
1567  */
1568 static struct lagg_port *
1569 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype)
1570 {
1571 	struct lagg_softc *sc;
1572 	struct lagg_port *lp;
1573 	struct lagg_lb *lb;
1574 	uint32_t p;
1575 
1576 	sc = ifp->if_softc;
1577 
1578 	switch (sc->sc_proto) {
1579 	case LAGG_PROTO_FAILOVER:
1580 		return (lagg_link_active(sc, sc->sc_primary));
1581 	case LAGG_PROTO_LOADBALANCE:
1582 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1583 		    flowtype == M_HASHTYPE_NONE)
1584 			return (NULL);
1585 		p = flowid >> sc->flowid_shift;
1586 		p %= sc->sc_count;
1587 		lb = (struct lagg_lb *)sc->sc_psc;
1588 		lp = lb->lb_ports[p];
1589 		return (lagg_link_active(sc, lp));
1590 	case LAGG_PROTO_LACP:
1591 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1592 		    flowtype == M_HASHTYPE_NONE)
1593 			return (NULL);
1594 		return (lacp_select_tx_port_by_hash(sc, flowid));
1595 	default:
1596 		return (NULL);
1597 	}
1598 }
1599 
1600 static int
1601 lagg_snd_tag_alloc(struct ifnet *ifp,
1602     union if_snd_tag_alloc_params *params,
1603     struct m_snd_tag **ppmt)
1604 {
1605 	struct lagg_snd_tag *lst;
1606 	struct lagg_softc *sc;
1607 	struct lagg_port *lp;
1608 	struct ifnet *lp_ifp;
1609 	int error;
1610 
1611 	sc = ifp->if_softc;
1612 
1613 	LAGG_RLOCK();
1614 	lp = lookup_snd_tag_port(ifp, params->hdr.flowid, params->hdr.flowtype);
1615 	if (lp == NULL) {
1616 		LAGG_RUNLOCK();
1617 		return (EOPNOTSUPP);
1618 	}
1619 	if (lp->lp_ifp == NULL || lp->lp_ifp->if_snd_tag_alloc == NULL) {
1620 		LAGG_RUNLOCK();
1621 		return (EOPNOTSUPP);
1622 	}
1623 	lp_ifp = lp->lp_ifp;
1624 	if_ref(lp_ifp);
1625 	LAGG_RUNLOCK();
1626 
1627 	lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1628 	if (lst == NULL) {
1629 		if_rele(lp_ifp);
1630 		return (ENOMEM);
1631 	}
1632 
1633 	error = lp_ifp->if_snd_tag_alloc(lp_ifp, params, &lst->tag);
1634 	if_rele(lp_ifp);
1635 	if (error) {
1636 		free(lst, M_LAGG);
1637 		return (error);
1638 	}
1639 
1640 	m_snd_tag_init(&lst->com, ifp);
1641 
1642 	*ppmt = &lst->com;
1643 	return (0);
1644 }
1645 
1646 static int
1647 lagg_snd_tag_modify(struct m_snd_tag *mst,
1648     union if_snd_tag_modify_params *params)
1649 {
1650 	struct lagg_snd_tag *lst;
1651 
1652 	lst = mst_to_lst(mst);
1653 	return (lst->tag->ifp->if_snd_tag_modify(lst->tag, params));
1654 }
1655 
1656 static int
1657 lagg_snd_tag_query(struct m_snd_tag *mst,
1658     union if_snd_tag_query_params *params)
1659 {
1660 	struct lagg_snd_tag *lst;
1661 
1662 	lst = mst_to_lst(mst);
1663 	return (lst->tag->ifp->if_snd_tag_query(lst->tag, params));
1664 }
1665 
1666 static void
1667 lagg_snd_tag_free(struct m_snd_tag *mst)
1668 {
1669 	struct lagg_snd_tag *lst;
1670 
1671 	lst = mst_to_lst(mst);
1672 	m_snd_tag_rele(lst->tag);
1673 	free(lst, M_LAGG);
1674 }
1675 
1676 static void
1677 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1678 {
1679 	/*
1680 	 * For lagg, we have an indirect
1681 	 * interface. The caller needs to
1682 	 * get a ratelimit tag on the actual
1683 	 * interface the flow will go on.
1684 	 */
1685 	q->rate_table = NULL;
1686 	q->flags = RT_IS_INDIRECT;
1687 	q->max_flows = 0;
1688 	q->number_of_rates = 0;
1689 }
1690 #endif
1691 
1692 static int
1693 lagg_setmulti(struct lagg_port *lp)
1694 {
1695 	struct lagg_softc *sc = lp->lp_softc;
1696 	struct ifnet *ifp = lp->lp_ifp;
1697 	struct ifnet *scifp = sc->sc_ifp;
1698 	struct lagg_mc *mc;
1699 	struct ifmultiaddr *ifma;
1700 	int error;
1701 
1702 	IF_ADDR_WLOCK(scifp);
1703 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1704 		if (ifma->ifma_addr->sa_family != AF_LINK)
1705 			continue;
1706 		mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
1707 		if (mc == NULL) {
1708 			IF_ADDR_WUNLOCK(scifp);
1709 			return (ENOMEM);
1710 		}
1711 		bcopy(ifma->ifma_addr, &mc->mc_addr,
1712 		    ifma->ifma_addr->sa_len);
1713 		mc->mc_addr.sdl_index = ifp->if_index;
1714 		mc->mc_ifma = NULL;
1715 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1716 	}
1717 	IF_ADDR_WUNLOCK(scifp);
1718 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1719 		error = if_addmulti(ifp,
1720 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1721 		if (error)
1722 			return (error);
1723 	}
1724 	return (0);
1725 }
1726 
1727 static int
1728 lagg_clrmulti(struct lagg_port *lp)
1729 {
1730 	struct lagg_mc *mc;
1731 
1732 	LAGG_XLOCK_ASSERT(lp->lp_softc);
1733 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1734 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1735 		if (mc->mc_ifma && lp->lp_detaching == 0)
1736 			if_delmulti_ifma(mc->mc_ifma);
1737 		free(mc, M_LAGG);
1738 	}
1739 	return (0);
1740 }
1741 
1742 static int
1743 lagg_setcaps(struct lagg_port *lp, int cap)
1744 {
1745 	struct ifreq ifr;
1746 
1747 	if (lp->lp_ifp->if_capenable == cap)
1748 		return (0);
1749 	if (lp->lp_ioctl == NULL)
1750 		return (ENXIO);
1751 	ifr.ifr_reqcap = cap;
1752 	return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1753 }
1754 
1755 /* Handle a ref counted flag that should be set on the lagg port as well */
1756 static int
1757 lagg_setflag(struct lagg_port *lp, int flag, int status,
1758     int (*func)(struct ifnet *, int))
1759 {
1760 	struct lagg_softc *sc = lp->lp_softc;
1761 	struct ifnet *scifp = sc->sc_ifp;
1762 	struct ifnet *ifp = lp->lp_ifp;
1763 	int error;
1764 
1765 	LAGG_XLOCK_ASSERT(sc);
1766 
1767 	status = status ? (scifp->if_flags & flag) : 0;
1768 	/* Now "status" contains the flag value or 0 */
1769 
1770 	/*
1771 	 * See if recorded ports status is different from what
1772 	 * we want it to be.  If it is, flip it.  We record ports
1773 	 * status in lp_ifflags so that we won't clear ports flag
1774 	 * we haven't set.  In fact, we don't clear or set ports
1775 	 * flags directly, but get or release references to them.
1776 	 * That's why we can be sure that recorded flags still are
1777 	 * in accord with actual ports flags.
1778 	 */
1779 	if (status != (lp->lp_ifflags & flag)) {
1780 		error = (*func)(ifp, status);
1781 		if (error)
1782 			return (error);
1783 		lp->lp_ifflags &= ~flag;
1784 		lp->lp_ifflags |= status;
1785 	}
1786 	return (0);
1787 }
1788 
1789 /*
1790  * Handle IFF_* flags that require certain changes on the lagg port
1791  * if "status" is true, update ports flags respective to the lagg
1792  * if "status" is false, forcedly clear the flags set on port.
1793  */
1794 static int
1795 lagg_setflags(struct lagg_port *lp, int status)
1796 {
1797 	int error, i;
1798 
1799 	for (i = 0; lagg_pflags[i].flag; i++) {
1800 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1801 		    status, lagg_pflags[i].func);
1802 		if (error)
1803 			return (error);
1804 	}
1805 	return (0);
1806 }
1807 
1808 static int
1809 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1810 {
1811 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1812 	int error;
1813 
1814 #ifdef RATELIMIT
1815 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1816 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
1817 #endif
1818 	LAGG_RLOCK();
1819 	/* We need a Tx algorithm and at least one port */
1820 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1821 		LAGG_RUNLOCK();
1822 		m_freem(m);
1823 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1824 		return (ENXIO);
1825 	}
1826 
1827 	ETHER_BPF_MTAP(ifp, m);
1828 
1829 	error = lagg_proto_start(sc, m);
1830 	LAGG_RUNLOCK();
1831 
1832 	if (error != 0)
1833 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1834 
1835 	return (error);
1836 }
1837 
1838 /*
1839  * The ifp->if_qflush entry point for lagg(4) is no-op.
1840  */
1841 static void
1842 lagg_qflush(struct ifnet *ifp __unused)
1843 {
1844 }
1845 
1846 static struct mbuf *
1847 lagg_input(struct ifnet *ifp, struct mbuf *m)
1848 {
1849 	struct lagg_port *lp = ifp->if_lagg;
1850 	struct lagg_softc *sc = lp->lp_softc;
1851 	struct ifnet *scifp = sc->sc_ifp;
1852 
1853 	LAGG_RLOCK();
1854 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1855 	    lp->lp_detaching != 0 ||
1856 	    sc->sc_proto == LAGG_PROTO_NONE) {
1857 		LAGG_RUNLOCK();
1858 		m_freem(m);
1859 		return (NULL);
1860 	}
1861 
1862 	ETHER_BPF_MTAP(scifp, m);
1863 
1864 	m = lagg_proto_input(sc, lp, m);
1865 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
1866 		m_freem(m);
1867 		m = NULL;
1868 	}
1869 
1870 	LAGG_RUNLOCK();
1871 	return (m);
1872 }
1873 
1874 static int
1875 lagg_media_change(struct ifnet *ifp)
1876 {
1877 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1878 
1879 	if (sc->sc_ifflags & IFF_DEBUG)
1880 		printf("%s\n", __func__);
1881 
1882 	/* Ignore */
1883 	return (0);
1884 }
1885 
1886 static void
1887 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1888 {
1889 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1890 	struct lagg_port *lp;
1891 
1892 	imr->ifm_status = IFM_AVALID;
1893 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1894 
1895 	LAGG_RLOCK();
1896 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1897 		if (LAGG_PORTACTIVE(lp))
1898 			imr->ifm_status |= IFM_ACTIVE;
1899 	}
1900 	LAGG_RUNLOCK();
1901 }
1902 
1903 static void
1904 lagg_linkstate(struct lagg_softc *sc)
1905 {
1906 	struct lagg_port *lp;
1907 	int new_link = LINK_STATE_DOWN;
1908 	uint64_t speed;
1909 
1910 	LAGG_XLOCK_ASSERT(sc);
1911 
1912 	/* LACP handles link state itself */
1913 	if (sc->sc_proto == LAGG_PROTO_LACP)
1914 		return;
1915 
1916 	/* Our link is considered up if at least one of our ports is active */
1917 	LAGG_RLOCK();
1918 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1919 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1920 			new_link = LINK_STATE_UP;
1921 			break;
1922 		}
1923 	}
1924 	LAGG_RUNLOCK();
1925 	if_link_state_change(sc->sc_ifp, new_link);
1926 
1927 	/* Update if_baudrate to reflect the max possible speed */
1928 	switch (sc->sc_proto) {
1929 		case LAGG_PROTO_FAILOVER:
1930 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1931 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1932 			break;
1933 		case LAGG_PROTO_ROUNDROBIN:
1934 		case LAGG_PROTO_LOADBALANCE:
1935 		case LAGG_PROTO_BROADCAST:
1936 			speed = 0;
1937 			LAGG_RLOCK();
1938 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1939 				speed += lp->lp_ifp->if_baudrate;
1940 			LAGG_RUNLOCK();
1941 			sc->sc_ifp->if_baudrate = speed;
1942 			break;
1943 		case LAGG_PROTO_LACP:
1944 			/* LACP updates if_baudrate itself */
1945 			break;
1946 	}
1947 }
1948 
1949 static void
1950 lagg_port_state(struct ifnet *ifp, int state)
1951 {
1952 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1953 	struct lagg_softc *sc = NULL;
1954 
1955 	if (lp != NULL)
1956 		sc = lp->lp_softc;
1957 	if (sc == NULL)
1958 		return;
1959 
1960 	LAGG_XLOCK(sc);
1961 	lagg_linkstate(sc);
1962 	lagg_proto_linkstate(sc, lp);
1963 	LAGG_XUNLOCK(sc);
1964 }
1965 
1966 struct lagg_port *
1967 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1968 {
1969 	struct lagg_port *lp_next, *rval = NULL;
1970 
1971 	/*
1972 	 * Search a port which reports an active link state.
1973 	 */
1974 
1975 #ifdef INVARIANTS
1976 	/*
1977 	 * This is called with either LAGG_RLOCK() held or
1978 	 * LAGG_XLOCK(sc) held.
1979 	 */
1980 	if (!in_epoch(net_epoch_preempt))
1981 		LAGG_XLOCK_ASSERT(sc);
1982 #endif
1983 
1984 	if (lp == NULL)
1985 		goto search;
1986 	if (LAGG_PORTACTIVE(lp)) {
1987 		rval = lp;
1988 		goto found;
1989 	}
1990 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
1991 	    LAGG_PORTACTIVE(lp_next)) {
1992 		rval = lp_next;
1993 		goto found;
1994 	}
1995 
1996 search:
1997 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1998 		if (LAGG_PORTACTIVE(lp_next)) {
1999 			return (lp_next);
2000 		}
2001 	}
2002 found:
2003 	return (rval);
2004 }
2005 
2006 int
2007 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2008 {
2009 
2010 #ifdef RATELIMIT
2011 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2012 		struct lagg_snd_tag *lst;
2013 		struct m_snd_tag *mst;
2014 
2015 		mst = m->m_pkthdr.snd_tag;
2016 		lst = mst_to_lst(mst);
2017 		if (lst->tag->ifp != ifp) {
2018 			m_freem(m);
2019 			return (EAGAIN);
2020 		}
2021 		m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2022 		m_snd_tag_rele(mst);
2023 	}
2024 #endif
2025 	return (ifp->if_transmit)(ifp, m);
2026 }
2027 
2028 /*
2029  * Simple round robin aggregation
2030  */
2031 static void
2032 lagg_rr_attach(struct lagg_softc *sc)
2033 {
2034 	sc->sc_seq = 0;
2035 	sc->sc_bkt_count = sc->sc_bkt;
2036 }
2037 
2038 static int
2039 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2040 {
2041 	struct lagg_port *lp;
2042 	uint32_t p;
2043 
2044 	if (sc->sc_bkt_count == 0 && sc->sc_bkt > 0)
2045 		sc->sc_bkt_count = sc->sc_bkt;
2046 
2047 	if (sc->sc_bkt > 0) {
2048 		atomic_subtract_int(&sc->sc_bkt_count, 1);
2049 	if (atomic_cmpset_int(&sc->sc_bkt_count, 0, sc->sc_bkt))
2050 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
2051 	else
2052 		p = sc->sc_seq;
2053 	} else
2054 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
2055 
2056 	p %= sc->sc_count;
2057 	lp = CK_SLIST_FIRST(&sc->sc_ports);
2058 
2059 	while (p--)
2060 		lp = CK_SLIST_NEXT(lp, lp_entries);
2061 
2062 	/*
2063 	 * Check the port's link state. This will return the next active
2064 	 * port if the link is down or the port is NULL.
2065 	 */
2066 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2067 		m_freem(m);
2068 		return (ENETDOWN);
2069 	}
2070 
2071 	/* Send mbuf */
2072 	return (lagg_enqueue(lp->lp_ifp, m));
2073 }
2074 
2075 static struct mbuf *
2076 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2077 {
2078 	struct ifnet *ifp = sc->sc_ifp;
2079 
2080 	/* Just pass in the packet to our lagg device */
2081 	m->m_pkthdr.rcvif = ifp;
2082 
2083 	return (m);
2084 }
2085 
2086 /*
2087  * Broadcast mode
2088  */
2089 static int
2090 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2091 {
2092 	int active_ports = 0;
2093 	int errors = 0;
2094 	int ret;
2095 	struct lagg_port *lp, *last = NULL;
2096 	struct mbuf *m0;
2097 
2098 	LAGG_RLOCK_ASSERT();
2099 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2100 		if (!LAGG_PORTACTIVE(lp))
2101 			continue;
2102 
2103 		active_ports++;
2104 
2105 		if (last != NULL) {
2106 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2107 			if (m0 == NULL) {
2108 				ret = ENOBUFS;
2109 				errors++;
2110 				break;
2111 			}
2112 
2113 			ret = lagg_enqueue(last->lp_ifp, m0);
2114 			if (ret != 0)
2115 				errors++;
2116 		}
2117 		last = lp;
2118 	}
2119 
2120 	if (last == NULL) {
2121 		m_freem(m);
2122 		return (ENOENT);
2123 	}
2124 	if ((last = lagg_link_active(sc, last)) == NULL) {
2125 		m_freem(m);
2126 		return (ENETDOWN);
2127 	}
2128 
2129 	ret = lagg_enqueue(last->lp_ifp, m);
2130 	if (ret != 0)
2131 		errors++;
2132 
2133 	if (errors == 0)
2134 		return (ret);
2135 
2136 	return (0);
2137 }
2138 
2139 static struct mbuf*
2140 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2141 {
2142 	struct ifnet *ifp = sc->sc_ifp;
2143 
2144 	/* Just pass in the packet to our lagg device */
2145 	m->m_pkthdr.rcvif = ifp;
2146 	return (m);
2147 }
2148 
2149 /*
2150  * Active failover
2151  */
2152 static int
2153 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2154 {
2155 	struct lagg_port *lp;
2156 
2157 	/* Use the master port if active or the next available port */
2158 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2159 		m_freem(m);
2160 		return (ENETDOWN);
2161 	}
2162 
2163 	/* Send mbuf */
2164 	return (lagg_enqueue(lp->lp_ifp, m));
2165 }
2166 
2167 static struct mbuf *
2168 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2169 {
2170 	struct ifnet *ifp = sc->sc_ifp;
2171 	struct lagg_port *tmp_tp;
2172 
2173 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2174 		m->m_pkthdr.rcvif = ifp;
2175 		return (m);
2176 	}
2177 
2178 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2179 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2180 		/*
2181 		 * If tmp_tp is null, we've received a packet when all
2182 		 * our links are down. Weird, but process it anyways.
2183 		 */
2184 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2185 			m->m_pkthdr.rcvif = ifp;
2186 			return (m);
2187 		}
2188 	}
2189 
2190 	m_freem(m);
2191 	return (NULL);
2192 }
2193 
2194 /*
2195  * Loadbalancing
2196  */
2197 static void
2198 lagg_lb_attach(struct lagg_softc *sc)
2199 {
2200 	struct lagg_port *lp;
2201 	struct lagg_lb *lb;
2202 
2203 	LAGG_XLOCK_ASSERT(sc);
2204 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2205 	lb->lb_key = m_ether_tcpip_hash_init();
2206 	sc->sc_psc = lb;
2207 
2208 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2209 		lagg_lb_port_create(lp);
2210 }
2211 
2212 static void
2213 lagg_lb_detach(struct lagg_softc *sc)
2214 {
2215 	struct lagg_lb *lb;
2216 
2217 	lb = (struct lagg_lb *)sc->sc_psc;
2218 	if (lb != NULL)
2219 		free(lb, M_LAGG);
2220 }
2221 
2222 static int
2223 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2224 {
2225 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2226 	struct lagg_port *lp_next;
2227 	int i = 0, rv;
2228 
2229 	rv = 0;
2230 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2231 	LAGG_XLOCK_ASSERT(sc);
2232 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2233 		if (lp_next == lp)
2234 			continue;
2235 		if (i >= LAGG_MAX_PORTS) {
2236 			rv = EINVAL;
2237 			break;
2238 		}
2239 		if (sc->sc_ifflags & IFF_DEBUG)
2240 			printf("%s: port %s at index %d\n",
2241 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2242 		lb->lb_ports[i++] = lp_next;
2243 	}
2244 
2245 	return (rv);
2246 }
2247 
2248 static int
2249 lagg_lb_port_create(struct lagg_port *lp)
2250 {
2251 	struct lagg_softc *sc = lp->lp_softc;
2252 	return (lagg_lb_porttable(sc, NULL));
2253 }
2254 
2255 static void
2256 lagg_lb_port_destroy(struct lagg_port *lp)
2257 {
2258 	struct lagg_softc *sc = lp->lp_softc;
2259 	lagg_lb_porttable(sc, lp);
2260 }
2261 
2262 static int
2263 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2264 {
2265 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2266 	struct lagg_port *lp = NULL;
2267 	uint32_t p = 0;
2268 
2269 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2270 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2271 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2272 	else
2273 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2274 	p %= sc->sc_count;
2275 	lp = lb->lb_ports[p];
2276 
2277 	/*
2278 	 * Check the port's link state. This will return the next active
2279 	 * port if the link is down or the port is NULL.
2280 	 */
2281 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2282 		m_freem(m);
2283 		return (ENETDOWN);
2284 	}
2285 
2286 	/* Send mbuf */
2287 	return (lagg_enqueue(lp->lp_ifp, m));
2288 }
2289 
2290 static struct mbuf *
2291 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2292 {
2293 	struct ifnet *ifp = sc->sc_ifp;
2294 
2295 	/* Just pass in the packet to our lagg device */
2296 	m->m_pkthdr.rcvif = ifp;
2297 
2298 	return (m);
2299 }
2300 
2301 /*
2302  * 802.3ad LACP
2303  */
2304 static void
2305 lagg_lacp_attach(struct lagg_softc *sc)
2306 {
2307 	struct lagg_port *lp;
2308 
2309 	lacp_attach(sc);
2310 	LAGG_XLOCK_ASSERT(sc);
2311 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2312 		lacp_port_create(lp);
2313 }
2314 
2315 static void
2316 lagg_lacp_detach(struct lagg_softc *sc)
2317 {
2318 	struct lagg_port *lp;
2319 	void *psc;
2320 
2321 	LAGG_XLOCK_ASSERT(sc);
2322 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2323 		lacp_port_destroy(lp);
2324 
2325 	psc = sc->sc_psc;
2326 	sc->sc_psc = NULL;
2327 	lacp_detach(psc);
2328 }
2329 
2330 static void
2331 lagg_lacp_lladdr(struct lagg_softc *sc)
2332 {
2333 	struct lagg_port *lp;
2334 
2335 	LAGG_SXLOCK_ASSERT(sc);
2336 
2337 	/* purge all the lacp ports */
2338 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2339 		lacp_port_destroy(lp);
2340 
2341 	/* add them back in */
2342 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2343 		lacp_port_create(lp);
2344 }
2345 
2346 static int
2347 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2348 {
2349 	struct lagg_port *lp;
2350 
2351 	lp = lacp_select_tx_port(sc, m);
2352 	if (lp == NULL) {
2353 		m_freem(m);
2354 		return (ENETDOWN);
2355 	}
2356 
2357 	/* Send mbuf */
2358 	return (lagg_enqueue(lp->lp_ifp, m));
2359 }
2360 
2361 static struct mbuf *
2362 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2363 {
2364 	struct ifnet *ifp = sc->sc_ifp;
2365 	struct ether_header *eh;
2366 	u_short etype;
2367 
2368 	eh = mtod(m, struct ether_header *);
2369 	etype = ntohs(eh->ether_type);
2370 
2371 	/* Tap off LACP control messages */
2372 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2373 		m = lacp_input(lp, m);
2374 		if (m == NULL)
2375 			return (NULL);
2376 	}
2377 
2378 	/*
2379 	 * If the port is not collecting or not in the active aggregator then
2380 	 * free and return.
2381 	 */
2382 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2383 		m_freem(m);
2384 		return (NULL);
2385 	}
2386 
2387 	m->m_pkthdr.rcvif = ifp;
2388 	return (m);
2389 }
2390 
2391