xref: /freebsd/sys/net/if_lagg.c (revision 458b591fb0ccbeaf80f7116cd145fe896cbf7421)
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/lock.h>
40 #include <sys/rmlock.h>
41 #include <sys/taskqueue.h>
42 #include <sys/eventhandler.h>
43 
44 #include <net/ethernet.h>
45 #include <net/if.h>
46 #include <net/if_clone.h>
47 #include <net/if_arp.h>
48 #include <net/if_dl.h>
49 #include <net/if_media.h>
50 #include <net/if_types.h>
51 #include <net/if_var.h>
52 #include <net/bpf.h>
53 #include <net/vnet.h>
54 
55 #if defined(INET) || defined(INET6)
56 #include <netinet/in.h>
57 #include <netinet/ip.h>
58 #endif
59 #ifdef INET
60 #include <netinet/in_systm.h>
61 #include <netinet/if_ether.h>
62 #endif
63 
64 #ifdef INET6
65 #include <netinet/ip6.h>
66 #include <netinet6/in6_var.h>
67 #include <netinet6/in6_ifattach.h>
68 #endif
69 
70 #include <net/if_vlan_var.h>
71 #include <net/if_lagg.h>
72 #include <net/ieee8023ad_lacp.h>
73 
74 /* Special flags we should propagate to the lagg ports. */
75 static struct {
76 	int flag;
77 	int (*func)(struct ifnet *, int);
78 } lagg_pflags[] = {
79 	{IFF_PROMISC, ifpromisc},
80 	{IFF_ALLMULTI, if_allmulti},
81 	{0, NULL}
82 };
83 
84 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
85 #define	V_lagg_list	VNET(lagg_list)
86 static VNET_DEFINE(struct mtx, lagg_list_mtx);
87 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
88 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
89 					"if_lagg list", NULL, MTX_DEF)
90 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
91 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
92 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
93 eventhandler_tag	lagg_detach_cookie = NULL;
94 
95 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
96 static void	lagg_clone_destroy(struct ifnet *);
97 static VNET_DEFINE(struct if_clone *, lagg_cloner);
98 #define	V_lagg_cloner	VNET(lagg_cloner)
99 static const char laggname[] = "lagg";
100 
101 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
102 static void	lagg_capabilities(struct lagg_softc *);
103 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *, lagg_llqtype);
104 static void	lagg_port_setlladdr(void *, int);
105 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
106 static int	lagg_port_destroy(struct lagg_port *, int);
107 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
108 static void	lagg_linkstate(struct lagg_softc *);
109 static void	lagg_port_state(struct ifnet *, int);
110 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
111 static int	lagg_port_output(struct ifnet *, struct mbuf *,
112 		    const struct sockaddr *, struct route *);
113 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
114 #ifdef LAGG_PORT_STACKING
115 static int	lagg_port_checkstacking(struct lagg_softc *);
116 #endif
117 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
118 static void	lagg_init(void *);
119 static void	lagg_stop(struct lagg_softc *);
120 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
121 static int	lagg_ether_setmulti(struct lagg_softc *);
122 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
123 static	int	lagg_setflag(struct lagg_port *, int, int,
124 		    int (*func)(struct ifnet *, int));
125 static	int	lagg_setflags(struct lagg_port *, int status);
126 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
127 static int	lagg_transmit(struct ifnet *, struct mbuf *);
128 static void	lagg_qflush(struct ifnet *);
129 static int	lagg_media_change(struct ifnet *);
130 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
131 static struct lagg_port *lagg_link_active(struct lagg_softc *,
132 	    struct lagg_port *);
133 
134 /* Simple round robin */
135 static void	lagg_rr_attach(struct lagg_softc *);
136 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
137 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
138 		    struct mbuf *);
139 
140 /* Active failover */
141 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
142 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
143 		    struct mbuf *);
144 
145 /* Loadbalancing */
146 static void	lagg_lb_attach(struct lagg_softc *);
147 static void	lagg_lb_detach(struct lagg_softc *);
148 static int	lagg_lb_port_create(struct lagg_port *);
149 static void	lagg_lb_port_destroy(struct lagg_port *);
150 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
151 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
152 		    struct mbuf *);
153 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
154 
155 /* Broadcast */
156 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
157 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
158 		    struct mbuf *);
159 
160 /* 802.3ad LACP */
161 static void	lagg_lacp_attach(struct lagg_softc *);
162 static void	lagg_lacp_detach(struct lagg_softc *);
163 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
164 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
165 		    struct mbuf *);
166 static void	lagg_lacp_lladdr(struct lagg_softc *);
167 
168 /* lagg protocol table */
169 static const struct lagg_proto {
170 	lagg_proto	pr_num;
171 	void		(*pr_attach)(struct lagg_softc *);
172 	void		(*pr_detach)(struct lagg_softc *);
173 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
174 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
175 			    struct mbuf *);
176 	int		(*pr_addport)(struct lagg_port *);
177 	void		(*pr_delport)(struct lagg_port *);
178 	void		(*pr_linkstate)(struct lagg_port *);
179 	void 		(*pr_init)(struct lagg_softc *);
180 	void 		(*pr_stop)(struct lagg_softc *);
181 	void 		(*pr_lladdr)(struct lagg_softc *);
182 	void		(*pr_request)(struct lagg_softc *, void *);
183 	void		(*pr_portreq)(struct lagg_port *, void *);
184 } lagg_protos[] = {
185     {
186 	.pr_num = LAGG_PROTO_NONE
187     },
188     {
189 	.pr_num = LAGG_PROTO_ROUNDROBIN,
190 	.pr_attach = lagg_rr_attach,
191 	.pr_start = lagg_rr_start,
192 	.pr_input = lagg_rr_input,
193     },
194     {
195 	.pr_num = LAGG_PROTO_FAILOVER,
196 	.pr_start = lagg_fail_start,
197 	.pr_input = lagg_fail_input,
198     },
199     {
200 	.pr_num = LAGG_PROTO_LOADBALANCE,
201 	.pr_attach = lagg_lb_attach,
202 	.pr_detach = lagg_lb_detach,
203 	.pr_start = lagg_lb_start,
204 	.pr_input = lagg_lb_input,
205 	.pr_addport = lagg_lb_port_create,
206 	.pr_delport = lagg_lb_port_destroy,
207     },
208     {
209 	.pr_num = LAGG_PROTO_LACP,
210 	.pr_attach = lagg_lacp_attach,
211 	.pr_detach = lagg_lacp_detach,
212 	.pr_start = lagg_lacp_start,
213 	.pr_input = lagg_lacp_input,
214 	.pr_addport = lacp_port_create,
215 	.pr_delport = lacp_port_destroy,
216 	.pr_linkstate = lacp_linkstate,
217 	.pr_init = lacp_init,
218 	.pr_stop = lacp_stop,
219 	.pr_lladdr = lagg_lacp_lladdr,
220 	.pr_request = lacp_req,
221 	.pr_portreq = lacp_portreq,
222     },
223     {
224 	.pr_num = LAGG_PROTO_BROADCAST,
225 	.pr_start = lagg_bcast_start,
226 	.pr_input = lagg_bcast_input,
227     },
228 };
229 
230 SYSCTL_DECL(_net_link);
231 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
232     "Link Aggregation");
233 
234 /* Allow input on any failover links */
235 static VNET_DEFINE(int, lagg_failover_rx_all);
236 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
237 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
238     &VNET_NAME(lagg_failover_rx_all), 0,
239     "Accept input from any interface in a failover lagg");
240 
241 /* Default value for using flowid */
242 static VNET_DEFINE(int, def_use_flowid) = 1;
243 #define	V_def_use_flowid	VNET(def_use_flowid)
244 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
245     &VNET_NAME(def_use_flowid), 0,
246     "Default setting for using flow id for load sharing");
247 
248 /* Default value for flowid shift */
249 static VNET_DEFINE(int, def_flowid_shift) = 16;
250 #define	V_def_flowid_shift	VNET(def_flowid_shift)
251 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
252     &VNET_NAME(def_flowid_shift), 0,
253     "Default setting for flowid shift for load sharing");
254 
255 static void
256 vnet_lagg_init(const void *unused __unused)
257 {
258 
259 	LAGG_LIST_LOCK_INIT();
260 	SLIST_INIT(&V_lagg_list);
261 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
262 	    lagg_clone_destroy, 0);
263 }
264 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
265     vnet_lagg_init, NULL);
266 
267 static void
268 vnet_lagg_uninit(const void *unused __unused)
269 {
270 
271 	if_clone_detach(V_lagg_cloner);
272 	LAGG_LIST_LOCK_DESTROY();
273 }
274 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
275     vnet_lagg_uninit, NULL);
276 
277 static int
278 lagg_modevent(module_t mod, int type, void *data)
279 {
280 
281 	switch (type) {
282 	case MOD_LOAD:
283 		lagg_input_p = lagg_input;
284 		lagg_linkstate_p = lagg_port_state;
285 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
286 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
287 		    EVENTHANDLER_PRI_ANY);
288 		break;
289 	case MOD_UNLOAD:
290 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
291 		    lagg_detach_cookie);
292 		lagg_input_p = NULL;
293 		lagg_linkstate_p = NULL;
294 		break;
295 	default:
296 		return (EOPNOTSUPP);
297 	}
298 	return (0);
299 }
300 
301 static moduledata_t lagg_mod = {
302 	"if_lagg",
303 	lagg_modevent,
304 	0
305 };
306 
307 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
308 MODULE_VERSION(if_lagg, 1);
309 
310 static void
311 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
312 {
313 
314 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
315 	    __func__, sc));
316 
317 	if (sc->sc_ifflags & IFF_DEBUG)
318 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
319 
320 	if (lagg_protos[pr].pr_attach != NULL)
321 		lagg_protos[pr].pr_attach(sc);
322 	sc->sc_proto = pr;
323 }
324 
325 static void
326 lagg_proto_detach(struct lagg_softc *sc)
327 {
328 	lagg_proto pr;
329 
330 	LAGG_WLOCK_ASSERT(sc);
331 
332 	pr = sc->sc_proto;
333 	sc->sc_proto = LAGG_PROTO_NONE;
334 
335 	if (lagg_protos[pr].pr_detach != NULL)
336 		lagg_protos[pr].pr_detach(sc);
337 	else
338 		LAGG_WUNLOCK(sc);
339 }
340 
341 static int
342 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
343 {
344 
345 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
346 }
347 
348 static struct mbuf *
349 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
350 {
351 
352 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
353 }
354 
355 static int
356 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
357 {
358 
359 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
360 		return (0);
361 	else
362 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
363 }
364 
365 static void
366 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
367 {
368 
369 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
370 		lagg_protos[sc->sc_proto].pr_delport(lp);
371 }
372 
373 static void
374 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
375 {
376 
377 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
378 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
379 }
380 
381 static void
382 lagg_proto_init(struct lagg_softc *sc)
383 {
384 
385 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
386 		lagg_protos[sc->sc_proto].pr_init(sc);
387 }
388 
389 static void
390 lagg_proto_stop(struct lagg_softc *sc)
391 {
392 
393 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
394 		lagg_protos[sc->sc_proto].pr_stop(sc);
395 }
396 
397 static void
398 lagg_proto_lladdr(struct lagg_softc *sc)
399 {
400 
401 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
402 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
403 }
404 
405 static void
406 lagg_proto_request(struct lagg_softc *sc, void *v)
407 {
408 
409 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
410 		lagg_protos[sc->sc_proto].pr_request(sc, v);
411 }
412 
413 static void
414 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
415 {
416 
417 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
418 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
419 }
420 
421 /*
422  * This routine is run via an vlan
423  * config EVENT
424  */
425 static void
426 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
427 {
428 	struct lagg_softc *sc = ifp->if_softc;
429 	struct lagg_port *lp;
430 	struct rm_priotracker tracker;
431 
432 	if (ifp->if_softc !=  arg)   /* Not our event */
433 		return;
434 
435 	LAGG_RLOCK(sc, &tracker);
436 	if (!SLIST_EMPTY(&sc->sc_ports)) {
437 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
438 			EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
439 	}
440 	LAGG_RUNLOCK(sc, &tracker);
441 }
442 
443 /*
444  * This routine is run via an vlan
445  * unconfig EVENT
446  */
447 static void
448 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
449 {
450 	struct lagg_softc *sc = ifp->if_softc;
451 	struct lagg_port *lp;
452 	struct rm_priotracker tracker;
453 
454 	if (ifp->if_softc !=  arg)   /* Not our event */
455 		return;
456 
457 	LAGG_RLOCK(sc, &tracker);
458 	if (!SLIST_EMPTY(&sc->sc_ports)) {
459 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
460 			EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
461 	}
462 	LAGG_RUNLOCK(sc, &tracker);
463 }
464 
465 static int
466 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
467 {
468 	struct lagg_softc *sc;
469 	struct ifnet *ifp;
470 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
471 
472 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
473 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
474 	if (ifp == NULL) {
475 		free(sc, M_DEVBUF);
476 		return (ENOSPC);
477 	}
478 
479 	if (V_def_use_flowid)
480 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
481 	sc->flowid_shift = V_def_flowid_shift;
482 
483 	/* Hash all layers by default */
484 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
485 
486 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
487 
488 	LAGG_LOCK_INIT(sc);
489 	SLIST_INIT(&sc->sc_ports);
490 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
491 
492 	/* Initialise pseudo media types */
493 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
494 	    lagg_media_status);
495 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
496 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
497 
498 	if_initname(ifp, laggname, unit);
499 	ifp->if_softc = sc;
500 	ifp->if_transmit = lagg_transmit;
501 	ifp->if_qflush = lagg_qflush;
502 	ifp->if_init = lagg_init;
503 	ifp->if_ioctl = lagg_ioctl;
504 	ifp->if_get_counter = lagg_get_counter;
505 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
506 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
507 
508 	/*
509 	 * Attach as an ordinary ethernet device, children will be attached
510 	 * as special device IFT_IEEE8023ADLAG.
511 	 */
512 	ether_ifattach(ifp, eaddr);
513 
514 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
515 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
516 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
517 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
518 
519 	/* Insert into the global list of laggs */
520 	LAGG_LIST_LOCK();
521 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
522 	LAGG_LIST_UNLOCK();
523 
524 	return (0);
525 }
526 
527 static void
528 lagg_clone_destroy(struct ifnet *ifp)
529 {
530 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
531 	struct lagg_port *lp;
532 
533 	LAGG_WLOCK(sc);
534 
535 	lagg_stop(sc);
536 	ifp->if_flags &= ~IFF_UP;
537 
538 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
539 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
540 
541 	/* Shutdown and remove lagg ports */
542 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
543 		lagg_port_destroy(lp, 1);
544 	/* Unhook the aggregation protocol */
545 	lagg_proto_detach(sc);
546 	LAGG_UNLOCK_ASSERT(sc);
547 
548 	ifmedia_removeall(&sc->sc_media);
549 	ether_ifdetach(ifp);
550 	if_free(ifp);
551 
552 	LAGG_LIST_LOCK();
553 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
554 	LAGG_LIST_UNLOCK();
555 
556 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
557 	LAGG_LOCK_DESTROY(sc);
558 	free(sc, M_DEVBUF);
559 }
560 
561 /*
562  * Set link-layer address on the lagg interface itself.
563  *
564  * Set noinline to be dtrace-friendly
565  */
566 static __noinline void
567 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
568 {
569 	struct ifnet *ifp = sc->sc_ifp;
570 	struct lagg_port lp;
571 
572 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
573 		return;
574 
575 	LAGG_WLOCK_ASSERT(sc);
576 	/*
577 	 * Set the link layer address on the lagg interface.
578 	 * lagg_proto_lladdr() notifies the MAC change to
579 	 * the aggregation protocol.  iflladdr_event handler which
580 	 * may trigger gratuitous ARPs for INET will be handled in
581 	 * a taskqueue.
582 	 */
583 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
584 	lagg_proto_lladdr(sc);
585 
586 	/*
587 	 * Send notification request for lagg interface
588 	 * itself. Note that new lladdr is already set.
589 	 */
590 	bzero(&lp, sizeof(lp));
591 	lp.lp_ifp = sc->sc_ifp;
592 	lp.lp_softc = sc;
593 
594 	/* Do not request lladdr change */
595 	lagg_port_lladdr(&lp, lladdr, LAGG_LLQTYPE_VIRT);
596 }
597 
598 static void
599 lagg_capabilities(struct lagg_softc *sc)
600 {
601 	struct lagg_port *lp;
602 	int cap = ~0, ena = ~0;
603 	u_long hwa = ~0UL;
604 	struct ifnet_hw_tsomax hw_tsomax;
605 
606 	LAGG_WLOCK_ASSERT(sc);
607 
608 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
609 
610 	/* Get capabilities from the lagg ports */
611 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
612 		cap &= lp->lp_ifp->if_capabilities;
613 		ena &= lp->lp_ifp->if_capenable;
614 		hwa &= lp->lp_ifp->if_hwassist;
615 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
616 	}
617 	cap = (cap == ~0 ? 0 : cap);
618 	ena = (ena == ~0 ? 0 : ena);
619 	hwa = (hwa == ~0 ? 0 : hwa);
620 
621 	if (sc->sc_ifp->if_capabilities != cap ||
622 	    sc->sc_ifp->if_capenable != ena ||
623 	    sc->sc_ifp->if_hwassist != hwa ||
624 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
625 		sc->sc_ifp->if_capabilities = cap;
626 		sc->sc_ifp->if_capenable = ena;
627 		sc->sc_ifp->if_hwassist = hwa;
628 		getmicrotime(&sc->sc_ifp->if_lastchange);
629 
630 		if (sc->sc_ifflags & IFF_DEBUG)
631 			if_printf(sc->sc_ifp,
632 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
633 	}
634 }
635 
636 /*
637  * Enqueue interface lladdr notification.
638  * If request is already queued, it is updated.
639  * If setting lladdr is also desired, @do_change has to be set to 1.
640  *
641  * Set noinline to be dtrace-friendly
642  */
643 static __noinline void
644 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr, lagg_llqtype llq_type)
645 {
646 	struct lagg_softc *sc = lp->lp_softc;
647 	struct ifnet *ifp = lp->lp_ifp;
648 	struct lagg_llq *llq;
649 
650 	LAGG_WLOCK_ASSERT(sc);
651 
652 	/*
653 	 * Do not enqueue requests where lladdr is the same for
654 	 * "physical" interfaces (e.g. ports in lagg)
655 	 */
656 	if (llq_type == LAGG_LLQTYPE_PHYS &&
657 	    memcmp(IF_LLADDR(ifp), lladdr, ETHER_ADDR_LEN) == 0)
658 		return;
659 
660 	/* Check to make sure its not already queued to be changed */
661 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
662 		if (llq->llq_ifp == ifp) {
663 			/* Update lladdr, it may have changed */
664 			bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
665 			return;
666 		}
667 	}
668 
669 	llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT | M_ZERO);
670 	if (llq == NULL)	/* XXX what to do */
671 		return;
672 
673 	llq->llq_ifp = ifp;
674 	llq->llq_type = llq_type;
675 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
676 	/* XXX: We should insert to tail */
677 	SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
678 
679 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
680 }
681 
682 /*
683  * Set the interface MAC address from a taskqueue to avoid a LOR.
684  *
685  * Set noinline to be dtrace-friendly
686  */
687 static __noinline void
688 lagg_port_setlladdr(void *arg, int pending)
689 {
690 	struct lagg_softc *sc = (struct lagg_softc *)arg;
691 	struct lagg_llq *llq, *head;
692 	struct ifnet *ifp;
693 	int error;
694 
695 	/* Grab a local reference of the queue and remove it from the softc */
696 	LAGG_WLOCK(sc);
697 	head = SLIST_FIRST(&sc->sc_llq_head);
698 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
699 	LAGG_WUNLOCK(sc);
700 
701 	/*
702 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
703 	 * unlocked as we have the only reference to it.
704 	 */
705 	for (llq = head; llq != NULL; llq = head) {
706 		ifp = llq->llq_ifp;
707 
708 		CURVNET_SET(ifp->if_vnet);
709 		error = 0;
710 
711 		/*
712 		 * Set the link layer address on the laggport interface.
713 		 * Note that if_setlladdr() or iflladdr_event handler
714 		 * may result in arp transmission / lltable updates.
715 		 */
716 		if (llq->llq_type == LAGG_LLQTYPE_PHYS)
717 			error = if_setlladdr(ifp, llq->llq_lladdr,
718 			    ETHER_ADDR_LEN);
719 		if (error)
720 			printf("%s: setlladdr failed on %s\n", __func__,
721 			    ifp->if_xname);
722 		else
723 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
724 		CURVNET_RESTORE();
725 		head = SLIST_NEXT(llq, llq_entries);
726 		free(llq, M_DEVBUF);
727 	}
728 }
729 
730 static int
731 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
732 {
733 	struct lagg_softc *sc_ptr;
734 	struct lagg_port *lp, *tlp;
735 	int error, i;
736 	uint64_t *pval;
737 
738 	LAGG_WLOCK_ASSERT(sc);
739 
740 	/* Limit the maximal number of lagg ports */
741 	if (sc->sc_count >= LAGG_MAX_PORTS)
742 		return (ENOSPC);
743 
744 	/* Check if port has already been associated to a lagg */
745 	if (ifp->if_lagg != NULL) {
746 		/* Port is already in the current lagg? */
747 		lp = (struct lagg_port *)ifp->if_lagg;
748 		if (lp->lp_softc == sc)
749 			return (EEXIST);
750 		return (EBUSY);
751 	}
752 
753 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
754 	if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
755 		return (EPROTONOSUPPORT);
756 
757 	/* Allow the first Ethernet member to define the MTU */
758 	if (SLIST_EMPTY(&sc->sc_ports))
759 		sc->sc_ifp->if_mtu = ifp->if_mtu;
760 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
761 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
762 		    ifp->if_xname);
763 		return (EINVAL);
764 	}
765 
766 	if ((lp = malloc(sizeof(struct lagg_port),
767 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
768 		return (ENOMEM);
769 
770 	/* Check if port is a stacked lagg */
771 	LAGG_LIST_LOCK();
772 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
773 		if (ifp == sc_ptr->sc_ifp) {
774 			LAGG_LIST_UNLOCK();
775 			free(lp, M_DEVBUF);
776 			return (EINVAL);
777 			/* XXX disable stacking for the moment, its untested */
778 #ifdef LAGG_PORT_STACKING
779 			lp->lp_flags |= LAGG_PORT_STACK;
780 			if (lagg_port_checkstacking(sc_ptr) >=
781 			    LAGG_MAX_STACKING) {
782 				LAGG_LIST_UNLOCK();
783 				free(lp, M_DEVBUF);
784 				return (E2BIG);
785 			}
786 #endif
787 		}
788 	}
789 	LAGG_LIST_UNLOCK();
790 
791 	/* Change the interface type */
792 	lp->lp_iftype = ifp->if_type;
793 	ifp->if_type = IFT_IEEE8023ADLAG;
794 	ifp->if_lagg = lp;
795 	lp->lp_ioctl = ifp->if_ioctl;
796 	ifp->if_ioctl = lagg_port_ioctl;
797 	lp->lp_output = ifp->if_output;
798 	ifp->if_output = lagg_port_output;
799 
800 	lp->lp_ifp = ifp;
801 	lp->lp_softc = sc;
802 
803 	/* Save port link layer address */
804 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
805 
806 	if (SLIST_EMPTY(&sc->sc_ports)) {
807 		sc->sc_primary = lp;
808 		/* First port in lagg. Update/notify lagg lladdress */
809 		lagg_lladdr(sc, IF_LLADDR(ifp));
810 	} else {
811 
812 		/*
813 		 * Update link layer address for this port and
814 		 * send notifications to other subsystems.
815 		 */
816 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp), LAGG_LLQTYPE_PHYS);
817 	}
818 
819 	/*
820 	 * Insert into the list of ports.
821 	 * Keep ports sorted by if_index. It is handy, when configuration
822 	 * is predictable and `ifconfig laggN create ...` command
823 	 * will lead to the same result each time.
824 	 */
825 	SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
826 		if (tlp->lp_ifp->if_index < ifp->if_index && (
827 		    SLIST_NEXT(tlp, lp_entries) == NULL ||
828 		    SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index >
829 		    ifp->if_index))
830 			break;
831 	}
832 	if (tlp != NULL)
833 		SLIST_INSERT_AFTER(tlp, lp, lp_entries);
834 	else
835 		SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
836 	sc->sc_count++;
837 
838 	/* Update lagg capabilities */
839 	lagg_capabilities(sc);
840 	lagg_linkstate(sc);
841 
842 	/* Read port counters */
843 	pval = lp->port_counters.val;
844 	for (i = 0; i < IFCOUNTERS; i++, pval++)
845 		*pval = ifp->if_get_counter(ifp, i);
846 	/* Add multicast addresses and interface flags to this port */
847 	lagg_ether_cmdmulti(lp, 1);
848 	lagg_setflags(lp, 1);
849 
850 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
851 		/* Remove the port, without calling pr_delport. */
852 		lagg_port_destroy(lp, 0);
853 		return (error);
854 	}
855 
856 	return (0);
857 }
858 
859 #ifdef LAGG_PORT_STACKING
860 static int
861 lagg_port_checkstacking(struct lagg_softc *sc)
862 {
863 	struct lagg_softc *sc_ptr;
864 	struct lagg_port *lp;
865 	int m = 0;
866 
867 	LAGG_WLOCK_ASSERT(sc);
868 
869 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
870 		if (lp->lp_flags & LAGG_PORT_STACK) {
871 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
872 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
873 		}
874 	}
875 
876 	return (m + 1);
877 }
878 #endif
879 
880 static int
881 lagg_port_destroy(struct lagg_port *lp, int rundelport)
882 {
883 	struct lagg_softc *sc = lp->lp_softc;
884 	struct lagg_port *lp_ptr, *lp0;
885 	struct lagg_llq *llq;
886 	struct ifnet *ifp = lp->lp_ifp;
887 	uint64_t *pval, vdiff;
888 	int i;
889 
890 	LAGG_WLOCK_ASSERT(sc);
891 
892 	if (rundelport)
893 		lagg_proto_delport(sc, lp);
894 
895 	/*
896 	 * Remove multicast addresses and interface flags from this port and
897 	 * reset the MAC address, skip if the interface is being detached.
898 	 */
899 	if (!lp->lp_detaching) {
900 		lagg_ether_cmdmulti(lp, 0);
901 		lagg_setflags(lp, 0);
902 		lagg_port_lladdr(lp, lp->lp_lladdr, LAGG_LLQTYPE_PHYS);
903 	}
904 
905 	/* Restore interface */
906 	ifp->if_type = lp->lp_iftype;
907 	ifp->if_ioctl = lp->lp_ioctl;
908 	ifp->if_output = lp->lp_output;
909 	ifp->if_lagg = NULL;
910 
911 	/* Update detached port counters */
912 	pval = lp->port_counters.val;
913 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
914 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
915 		sc->detached_counters.val[i] += vdiff;
916 	}
917 
918 	/* Finally, remove the port from the lagg */
919 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
920 	sc->sc_count--;
921 
922 	/* Update the primary interface */
923 	if (lp == sc->sc_primary) {
924 		uint8_t lladdr[ETHER_ADDR_LEN];
925 
926 		if ((lp0 = SLIST_FIRST(&sc->sc_ports)) == NULL) {
927 			bzero(&lladdr, ETHER_ADDR_LEN);
928 		} else {
929 			bcopy(lp0->lp_lladdr,
930 			    lladdr, ETHER_ADDR_LEN);
931 		}
932 		lagg_lladdr(sc, lladdr);
933 
934 		/* Mark lp0 as new primary */
935 		sc->sc_primary = lp0;
936 
937 		/*
938 		 * Enqueue lladdr update/notification for each port
939 		 * (new primary needs update as well, to switch from
940 		 * old lladdr to its 'real' one).
941 		 */
942 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
943 			lagg_port_lladdr(lp_ptr, lladdr, LAGG_LLQTYPE_PHYS);
944 	}
945 
946 	/* Remove any pending lladdr changes from the queue */
947 	if (lp->lp_detaching) {
948 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
949 			if (llq->llq_ifp == ifp) {
950 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
951 				    llq_entries);
952 				free(llq, M_DEVBUF);
953 				break;	/* Only appears once */
954 			}
955 		}
956 	}
957 
958 	if (lp->lp_ifflags)
959 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
960 
961 	free(lp, M_DEVBUF);
962 
963 	/* Update lagg capabilities */
964 	lagg_capabilities(sc);
965 	lagg_linkstate(sc);
966 
967 	return (0);
968 }
969 
970 static int
971 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
972 {
973 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
974 	struct lagg_softc *sc;
975 	struct lagg_port *lp = NULL;
976 	int error = 0;
977 	struct rm_priotracker tracker;
978 
979 	/* Should be checked by the caller */
980 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
981 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
982 		goto fallback;
983 
984 	switch (cmd) {
985 	case SIOCGLAGGPORT:
986 		if (rp->rp_portname[0] == '\0' ||
987 		    ifunit(rp->rp_portname) != ifp) {
988 			error = EINVAL;
989 			break;
990 		}
991 
992 		LAGG_RLOCK(sc, &tracker);
993 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
994 			error = ENOENT;
995 			LAGG_RUNLOCK(sc, &tracker);
996 			break;
997 		}
998 
999 		lagg_port2req(lp, rp);
1000 		LAGG_RUNLOCK(sc, &tracker);
1001 		break;
1002 
1003 	case SIOCSIFCAP:
1004 		if (lp->lp_ioctl == NULL) {
1005 			error = EINVAL;
1006 			break;
1007 		}
1008 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1009 		if (error)
1010 			break;
1011 
1012 		/* Update lagg interface capabilities */
1013 		LAGG_WLOCK(sc);
1014 		lagg_capabilities(sc);
1015 		LAGG_WUNLOCK(sc);
1016 		break;
1017 
1018 	case SIOCSIFMTU:
1019 		/* Do not allow the MTU to be changed once joined */
1020 		error = EINVAL;
1021 		break;
1022 
1023 	default:
1024 		goto fallback;
1025 	}
1026 
1027 	return (error);
1028 
1029 fallback:
1030 	if (lp->lp_ioctl != NULL)
1031 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1032 
1033 	return (EINVAL);
1034 }
1035 
1036 /*
1037  * Requests counter @cnt data.
1038  *
1039  * Counter value is calculated the following way:
1040  * 1) for each port, sum  difference between current and "initial" measurements.
1041  * 2) add lagg logical interface counters.
1042  * 3) add data from detached_counters array.
1043  *
1044  * We also do the following things on ports attach/detach:
1045  * 1) On port attach we store all counters it has into port_counter array.
1046  * 2) On port detach we add the different between "initial" and
1047  *   current counters data to detached_counters array.
1048  */
1049 static uint64_t
1050 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1051 {
1052 	struct lagg_softc *sc;
1053 	struct lagg_port *lp;
1054 	struct ifnet *lpifp;
1055 	struct rm_priotracker tracker;
1056 	uint64_t newval, oldval, vsum;
1057 
1058 	/* Revise this when we've got non-generic counters. */
1059 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1060 
1061 	sc = (struct lagg_softc *)ifp->if_softc;
1062 	LAGG_RLOCK(sc, &tracker);
1063 
1064 	vsum = 0;
1065 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1066 		/* Saved attached value */
1067 		oldval = lp->port_counters.val[cnt];
1068 		/* current value */
1069 		lpifp = lp->lp_ifp;
1070 		newval = lpifp->if_get_counter(lpifp, cnt);
1071 		/* Calculate diff and save new */
1072 		vsum += newval - oldval;
1073 	}
1074 
1075 	/*
1076 	 * Add counter data which might be added by upper
1077 	 * layer protocols operating on logical interface.
1078 	 */
1079 	vsum += if_get_counter_default(ifp, cnt);
1080 
1081 	/*
1082 	 * Add counter data from detached ports counters
1083 	 */
1084 	vsum += sc->detached_counters.val[cnt];
1085 
1086 	LAGG_RUNLOCK(sc, &tracker);
1087 
1088 	return (vsum);
1089 }
1090 
1091 /*
1092  * For direct output to child ports.
1093  */
1094 static int
1095 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1096 	const struct sockaddr *dst, struct route *ro)
1097 {
1098 	struct lagg_port *lp = ifp->if_lagg;
1099 
1100 	switch (dst->sa_family) {
1101 		case pseudo_AF_HDRCMPLT:
1102 		case AF_UNSPEC:
1103 			return ((*lp->lp_output)(ifp, m, dst, ro));
1104 	}
1105 
1106 	/* drop any other frames */
1107 	m_freem(m);
1108 	return (ENETDOWN);
1109 }
1110 
1111 static void
1112 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1113 {
1114 	struct lagg_port *lp;
1115 	struct lagg_softc *sc;
1116 
1117 	if ((lp = ifp->if_lagg) == NULL)
1118 		return;
1119 	/* If the ifnet is just being renamed, don't do anything. */
1120 	if (ifp->if_flags & IFF_RENAMING)
1121 		return;
1122 
1123 	sc = lp->lp_softc;
1124 
1125 	LAGG_WLOCK(sc);
1126 	lp->lp_detaching = 1;
1127 	lagg_port_destroy(lp, 1);
1128 	LAGG_WUNLOCK(sc);
1129 }
1130 
1131 static void
1132 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1133 {
1134 	struct lagg_softc *sc = lp->lp_softc;
1135 
1136 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1137 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1138 	rp->rp_prio = lp->lp_prio;
1139 	rp->rp_flags = lp->lp_flags;
1140 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1141 
1142 	/* Add protocol specific flags */
1143 	switch (sc->sc_proto) {
1144 		case LAGG_PROTO_FAILOVER:
1145 			if (lp == sc->sc_primary)
1146 				rp->rp_flags |= LAGG_PORT_MASTER;
1147 			if (lp == lagg_link_active(sc, sc->sc_primary))
1148 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1149 			break;
1150 
1151 		case LAGG_PROTO_ROUNDROBIN:
1152 		case LAGG_PROTO_LOADBALANCE:
1153 		case LAGG_PROTO_BROADCAST:
1154 			if (LAGG_PORTACTIVE(lp))
1155 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1156 			break;
1157 
1158 		case LAGG_PROTO_LACP:
1159 			/* LACP has a different definition of active */
1160 			if (lacp_isactive(lp))
1161 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1162 			if (lacp_iscollecting(lp))
1163 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1164 			if (lacp_isdistributing(lp))
1165 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1166 			break;
1167 	}
1168 
1169 }
1170 
1171 static void
1172 lagg_init(void *xsc)
1173 {
1174 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1175 	struct ifnet *ifp = sc->sc_ifp;
1176 	struct lagg_port *lp;
1177 
1178 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1179 		return;
1180 
1181 	LAGG_WLOCK(sc);
1182 
1183 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1184 
1185 	/*
1186 	 * Update the port lladdrs if needed.
1187 	 * This might be if_setlladdr() notification
1188 	 * that lladdr has been changed.
1189 	 */
1190 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1191 		lagg_port_lladdr(lp, IF_LLADDR(ifp), LAGG_LLQTYPE_PHYS);
1192 
1193 	lagg_proto_init(sc);
1194 
1195 	LAGG_WUNLOCK(sc);
1196 }
1197 
1198 static void
1199 lagg_stop(struct lagg_softc *sc)
1200 {
1201 	struct ifnet *ifp = sc->sc_ifp;
1202 
1203 	LAGG_WLOCK_ASSERT(sc);
1204 
1205 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1206 		return;
1207 
1208 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1209 
1210 	lagg_proto_stop(sc);
1211 }
1212 
1213 static int
1214 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1215 {
1216 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1217 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1218 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1219 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1220 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1221 	struct ifreq *ifr = (struct ifreq *)data;
1222 	struct lagg_port *lp;
1223 	struct ifnet *tpif;
1224 	struct thread *td = curthread;
1225 	char *buf, *outbuf;
1226 	int count, buflen, len, error = 0;
1227 	struct rm_priotracker tracker;
1228 
1229 	bzero(&rpbuf, sizeof(rpbuf));
1230 
1231 	switch (cmd) {
1232 	case SIOCGLAGG:
1233 		LAGG_RLOCK(sc, &tracker);
1234 		count = 0;
1235 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1236 			count++;
1237 		buflen = count * sizeof(struct lagg_reqport);
1238 		LAGG_RUNLOCK(sc, &tracker);
1239 
1240 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1241 
1242 		LAGG_RLOCK(sc, &tracker);
1243 		ra->ra_proto = sc->sc_proto;
1244 		lagg_proto_request(sc, &ra->ra_psc);
1245 		count = 0;
1246 		buf = outbuf;
1247 		len = min(ra->ra_size, buflen);
1248 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1249 			if (len < sizeof(rpbuf))
1250 				break;
1251 
1252 			lagg_port2req(lp, &rpbuf);
1253 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1254 			count++;
1255 			buf += sizeof(rpbuf);
1256 			len -= sizeof(rpbuf);
1257 		}
1258 		LAGG_RUNLOCK(sc, &tracker);
1259 		ra->ra_ports = count;
1260 		ra->ra_size = count * sizeof(rpbuf);
1261 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1262 		free(outbuf, M_TEMP);
1263 		break;
1264 	case SIOCSLAGG:
1265 		error = priv_check(td, PRIV_NET_LAGG);
1266 		if (error)
1267 			break;
1268 		if (ra->ra_proto < 1 || ra->ra_proto >= LAGG_PROTO_MAX) {
1269 			error = EPROTONOSUPPORT;
1270 			break;
1271 		}
1272 
1273 		LAGG_WLOCK(sc);
1274 		lagg_proto_detach(sc);
1275 		LAGG_UNLOCK_ASSERT(sc);
1276 		lagg_proto_attach(sc, ra->ra_proto);
1277 		break;
1278 	case SIOCGLAGGOPTS:
1279 		ro->ro_opts = sc->sc_opts;
1280 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1281 			struct lacp_softc *lsc;
1282 
1283 			lsc = (struct lacp_softc *)sc->sc_psc;
1284 			if (lsc->lsc_debug.lsc_tx_test != 0)
1285 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1286 			if (lsc->lsc_debug.lsc_rx_test != 0)
1287 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1288 			if (lsc->lsc_strict_mode != 0)
1289 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1290 			if (lsc->lsc_fast_timeout != 0)
1291 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1292 
1293 			ro->ro_active = sc->sc_active;
1294 		} else {
1295 			ro->ro_active = 0;
1296 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1297 				ro->ro_active += LAGG_PORTACTIVE(lp);
1298 		}
1299 		ro->ro_flapping = sc->sc_flapping;
1300 		ro->ro_flowid_shift = sc->flowid_shift;
1301 		break;
1302 	case SIOCSLAGGOPTS:
1303 		error = priv_check(td, PRIV_NET_LAGG);
1304 		if (error)
1305 			break;
1306 		if (ro->ro_opts == 0)
1307 			break;
1308 		/*
1309 		 * Set options.  LACP options are stored in sc->sc_psc,
1310 		 * not in sc_opts.
1311 		 */
1312 		int valid, lacp;
1313 
1314 		switch (ro->ro_opts) {
1315 		case LAGG_OPT_USE_FLOWID:
1316 		case -LAGG_OPT_USE_FLOWID:
1317 		case LAGG_OPT_FLOWIDSHIFT:
1318 			valid = 1;
1319 			lacp = 0;
1320 			break;
1321 		case LAGG_OPT_LACP_TXTEST:
1322 		case -LAGG_OPT_LACP_TXTEST:
1323 		case LAGG_OPT_LACP_RXTEST:
1324 		case -LAGG_OPT_LACP_RXTEST:
1325 		case LAGG_OPT_LACP_STRICT:
1326 		case -LAGG_OPT_LACP_STRICT:
1327 		case LAGG_OPT_LACP_TIMEOUT:
1328 		case -LAGG_OPT_LACP_TIMEOUT:
1329 			valid = lacp = 1;
1330 			break;
1331 		default:
1332 			valid = lacp = 0;
1333 			break;
1334 		}
1335 
1336 		LAGG_WLOCK(sc);
1337 		if (valid == 0 ||
1338 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1339 			/* Invalid combination of options specified. */
1340 			error = EINVAL;
1341 			LAGG_WUNLOCK(sc);
1342 			break;	/* Return from SIOCSLAGGOPTS. */
1343 		}
1344 		/*
1345 		 * Store new options into sc->sc_opts except for
1346 		 * FLOWIDSHIFT and LACP options.
1347 		 */
1348 		if (lacp == 0) {
1349 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1350 				sc->flowid_shift = ro->ro_flowid_shift;
1351 			else if (ro->ro_opts > 0)
1352 				sc->sc_opts |= ro->ro_opts;
1353 			else
1354 				sc->sc_opts &= ~ro->ro_opts;
1355 		} else {
1356 			struct lacp_softc *lsc;
1357 			struct lacp_port *lp;
1358 
1359 			lsc = (struct lacp_softc *)sc->sc_psc;
1360 
1361 			switch (ro->ro_opts) {
1362 			case LAGG_OPT_LACP_TXTEST:
1363 				lsc->lsc_debug.lsc_tx_test = 1;
1364 				break;
1365 			case -LAGG_OPT_LACP_TXTEST:
1366 				lsc->lsc_debug.lsc_tx_test = 0;
1367 				break;
1368 			case LAGG_OPT_LACP_RXTEST:
1369 				lsc->lsc_debug.lsc_rx_test = 1;
1370 				break;
1371 			case -LAGG_OPT_LACP_RXTEST:
1372 				lsc->lsc_debug.lsc_rx_test = 0;
1373 				break;
1374 			case LAGG_OPT_LACP_STRICT:
1375 				lsc->lsc_strict_mode = 1;
1376 				break;
1377 			case -LAGG_OPT_LACP_STRICT:
1378 				lsc->lsc_strict_mode = 0;
1379 				break;
1380 			case LAGG_OPT_LACP_TIMEOUT:
1381 				LACP_LOCK(lsc);
1382         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1383                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1384 				LACP_UNLOCK(lsc);
1385 				lsc->lsc_fast_timeout = 1;
1386 				break;
1387 			case -LAGG_OPT_LACP_TIMEOUT:
1388 				LACP_LOCK(lsc);
1389         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1390                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1391 				LACP_UNLOCK(lsc);
1392 				lsc->lsc_fast_timeout = 0;
1393 				break;
1394 			}
1395 		}
1396 		LAGG_WUNLOCK(sc);
1397 		break;
1398 	case SIOCGLAGGFLAGS:
1399 		rf->rf_flags = 0;
1400 		LAGG_RLOCK(sc, &tracker);
1401 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1402 			rf->rf_flags |= LAGG_F_HASHL2;
1403 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1404 			rf->rf_flags |= LAGG_F_HASHL3;
1405 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1406 			rf->rf_flags |= LAGG_F_HASHL4;
1407 		LAGG_RUNLOCK(sc, &tracker);
1408 		break;
1409 	case SIOCSLAGGHASH:
1410 		error = priv_check(td, PRIV_NET_LAGG);
1411 		if (error)
1412 			break;
1413 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1414 			error = EINVAL;
1415 			break;
1416 		}
1417 		LAGG_WLOCK(sc);
1418 		sc->sc_flags = 0;
1419 		if (rf->rf_flags & LAGG_F_HASHL2)
1420 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1421 		if (rf->rf_flags & LAGG_F_HASHL3)
1422 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1423 		if (rf->rf_flags & LAGG_F_HASHL4)
1424 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1425 		LAGG_WUNLOCK(sc);
1426 		break;
1427 	case SIOCGLAGGPORT:
1428 		if (rp->rp_portname[0] == '\0' ||
1429 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1430 			error = EINVAL;
1431 			break;
1432 		}
1433 
1434 		LAGG_RLOCK(sc, &tracker);
1435 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1436 		    lp->lp_softc != sc) {
1437 			error = ENOENT;
1438 			LAGG_RUNLOCK(sc, &tracker);
1439 			break;
1440 		}
1441 
1442 		lagg_port2req(lp, rp);
1443 		LAGG_RUNLOCK(sc, &tracker);
1444 		break;
1445 	case SIOCSLAGGPORT:
1446 		error = priv_check(td, PRIV_NET_LAGG);
1447 		if (error)
1448 			break;
1449 		if (rp->rp_portname[0] == '\0' ||
1450 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1451 			error = EINVAL;
1452 			break;
1453 		}
1454 #ifdef INET6
1455 		/*
1456 		 * A laggport interface should not have inet6 address
1457 		 * because two interfaces with a valid link-local
1458 		 * scope zone must not be merged in any form.  This
1459 		 * restriction is needed to prevent violation of
1460 		 * link-local scope zone.  Attempts to add a laggport
1461 		 * interface which has inet6 addresses triggers
1462 		 * removal of all inet6 addresses on the member
1463 		 * interface.
1464 		 */
1465 		if (in6ifa_llaonifp(tpif)) {
1466 			in6_ifdetach(tpif);
1467 				if_printf(sc->sc_ifp,
1468 				    "IPv6 addresses on %s have been removed "
1469 				    "before adding it as a member to prevent "
1470 				    "IPv6 address scope violation.\n",
1471 				    tpif->if_xname);
1472 		}
1473 #endif
1474 		LAGG_WLOCK(sc);
1475 		error = lagg_port_create(sc, tpif);
1476 		LAGG_WUNLOCK(sc);
1477 		break;
1478 	case SIOCSLAGGDELPORT:
1479 		error = priv_check(td, PRIV_NET_LAGG);
1480 		if (error)
1481 			break;
1482 		if (rp->rp_portname[0] == '\0' ||
1483 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1484 			error = EINVAL;
1485 			break;
1486 		}
1487 
1488 		LAGG_WLOCK(sc);
1489 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1490 		    lp->lp_softc != sc) {
1491 			error = ENOENT;
1492 			LAGG_WUNLOCK(sc);
1493 			break;
1494 		}
1495 
1496 		error = lagg_port_destroy(lp, 1);
1497 		LAGG_WUNLOCK(sc);
1498 		break;
1499 	case SIOCSIFFLAGS:
1500 		/* Set flags on ports too */
1501 		LAGG_WLOCK(sc);
1502 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1503 			lagg_setflags(lp, 1);
1504 		}
1505 		LAGG_WUNLOCK(sc);
1506 
1507 		if (!(ifp->if_flags & IFF_UP) &&
1508 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1509 			/*
1510 			 * If interface is marked down and it is running,
1511 			 * then stop and disable it.
1512 			 */
1513 			LAGG_WLOCK(sc);
1514 			lagg_stop(sc);
1515 			LAGG_WUNLOCK(sc);
1516 		} else if ((ifp->if_flags & IFF_UP) &&
1517 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1518 			/*
1519 			 * If interface is marked up and it is stopped, then
1520 			 * start it.
1521 			 */
1522 			(*ifp->if_init)(sc);
1523 		}
1524 		break;
1525 	case SIOCADDMULTI:
1526 	case SIOCDELMULTI:
1527 		LAGG_WLOCK(sc);
1528 		error = lagg_ether_setmulti(sc);
1529 		LAGG_WUNLOCK(sc);
1530 		break;
1531 	case SIOCSIFMEDIA:
1532 	case SIOCGIFMEDIA:
1533 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1534 		break;
1535 
1536 	case SIOCSIFCAP:
1537 	case SIOCSIFMTU:
1538 		/* Do not allow the MTU or caps to be directly changed */
1539 		error = EINVAL;
1540 		break;
1541 
1542 	default:
1543 		error = ether_ioctl(ifp, cmd, data);
1544 		break;
1545 	}
1546 	return (error);
1547 }
1548 
1549 static int
1550 lagg_ether_setmulti(struct lagg_softc *sc)
1551 {
1552 	struct lagg_port *lp;
1553 
1554 	LAGG_WLOCK_ASSERT(sc);
1555 
1556 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1557 		/* First, remove any existing filter entries. */
1558 		lagg_ether_cmdmulti(lp, 0);
1559 		/* copy all addresses from the lagg interface to the port */
1560 		lagg_ether_cmdmulti(lp, 1);
1561 	}
1562 	return (0);
1563 }
1564 
1565 static int
1566 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1567 {
1568 	struct lagg_softc *sc = lp->lp_softc;
1569 	struct ifnet *ifp = lp->lp_ifp;
1570 	struct ifnet *scifp = sc->sc_ifp;
1571 	struct lagg_mc *mc;
1572 	struct ifmultiaddr *ifma;
1573 	int error;
1574 
1575 	LAGG_WLOCK_ASSERT(sc);
1576 
1577 	if (set) {
1578 		IF_ADDR_WLOCK(scifp);
1579 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1580 			if (ifma->ifma_addr->sa_family != AF_LINK)
1581 				continue;
1582 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1583 			if (mc == NULL) {
1584 				IF_ADDR_WUNLOCK(scifp);
1585 				return (ENOMEM);
1586 			}
1587 			bcopy(ifma->ifma_addr, &mc->mc_addr,
1588 			    ifma->ifma_addr->sa_len);
1589 			mc->mc_addr.sdl_index = ifp->if_index;
1590 			mc->mc_ifma = NULL;
1591 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1592 		}
1593 		IF_ADDR_WUNLOCK(scifp);
1594 		SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1595 			error = if_addmulti(ifp,
1596 			    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1597 			if (error)
1598 				return (error);
1599 		}
1600 	} else {
1601 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1602 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1603 			if (mc->mc_ifma && !lp->lp_detaching)
1604 				if_delmulti_ifma(mc->mc_ifma);
1605 			free(mc, M_DEVBUF);
1606 		}
1607 	}
1608 	return (0);
1609 }
1610 
1611 /* Handle a ref counted flag that should be set on the lagg port as well */
1612 static int
1613 lagg_setflag(struct lagg_port *lp, int flag, int status,
1614     int (*func)(struct ifnet *, int))
1615 {
1616 	struct lagg_softc *sc = lp->lp_softc;
1617 	struct ifnet *scifp = sc->sc_ifp;
1618 	struct ifnet *ifp = lp->lp_ifp;
1619 	int error;
1620 
1621 	LAGG_WLOCK_ASSERT(sc);
1622 
1623 	status = status ? (scifp->if_flags & flag) : 0;
1624 	/* Now "status" contains the flag value or 0 */
1625 
1626 	/*
1627 	 * See if recorded ports status is different from what
1628 	 * we want it to be.  If it is, flip it.  We record ports
1629 	 * status in lp_ifflags so that we won't clear ports flag
1630 	 * we haven't set.  In fact, we don't clear or set ports
1631 	 * flags directly, but get or release references to them.
1632 	 * That's why we can be sure that recorded flags still are
1633 	 * in accord with actual ports flags.
1634 	 */
1635 	if (status != (lp->lp_ifflags & flag)) {
1636 		error = (*func)(ifp, status);
1637 		if (error)
1638 			return (error);
1639 		lp->lp_ifflags &= ~flag;
1640 		lp->lp_ifflags |= status;
1641 	}
1642 	return (0);
1643 }
1644 
1645 /*
1646  * Handle IFF_* flags that require certain changes on the lagg port
1647  * if "status" is true, update ports flags respective to the lagg
1648  * if "status" is false, forcedly clear the flags set on port.
1649  */
1650 static int
1651 lagg_setflags(struct lagg_port *lp, int status)
1652 {
1653 	int error, i;
1654 
1655 	for (i = 0; lagg_pflags[i].flag; i++) {
1656 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1657 		    status, lagg_pflags[i].func);
1658 		if (error)
1659 			return (error);
1660 	}
1661 	return (0);
1662 }
1663 
1664 static int
1665 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1666 {
1667 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1668 	int error, len, mcast;
1669 	struct rm_priotracker tracker;
1670 
1671 	len = m->m_pkthdr.len;
1672 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1673 
1674 	LAGG_RLOCK(sc, &tracker);
1675 	/* We need a Tx algorithm and at least one port */
1676 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1677 		LAGG_RUNLOCK(sc, &tracker);
1678 		m_freem(m);
1679 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1680 		return (ENXIO);
1681 	}
1682 
1683 	ETHER_BPF_MTAP(ifp, m);
1684 
1685 	error = lagg_proto_start(sc, m);
1686 	LAGG_RUNLOCK(sc, &tracker);
1687 
1688 	if (error != 0)
1689 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1690 
1691 	return (error);
1692 }
1693 
1694 /*
1695  * The ifp->if_qflush entry point for lagg(4) is no-op.
1696  */
1697 static void
1698 lagg_qflush(struct ifnet *ifp __unused)
1699 {
1700 }
1701 
1702 static struct mbuf *
1703 lagg_input(struct ifnet *ifp, struct mbuf *m)
1704 {
1705 	struct lagg_port *lp = ifp->if_lagg;
1706 	struct lagg_softc *sc = lp->lp_softc;
1707 	struct ifnet *scifp = sc->sc_ifp;
1708 	struct rm_priotracker tracker;
1709 
1710 	LAGG_RLOCK(sc, &tracker);
1711 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1712 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1713 	    sc->sc_proto == LAGG_PROTO_NONE) {
1714 		LAGG_RUNLOCK(sc, &tracker);
1715 		m_freem(m);
1716 		return (NULL);
1717 	}
1718 
1719 	ETHER_BPF_MTAP(scifp, m);
1720 
1721 	if (lp->lp_detaching != 0) {
1722 		m_freem(m);
1723 		m = NULL;
1724 	} else
1725 		m = lagg_proto_input(sc, lp, m);
1726 
1727 	if (m != NULL) {
1728 		if (scifp->if_flags & IFF_MONITOR) {
1729 			m_freem(m);
1730 			m = NULL;
1731 		}
1732 	}
1733 
1734 	LAGG_RUNLOCK(sc, &tracker);
1735 	return (m);
1736 }
1737 
1738 static int
1739 lagg_media_change(struct ifnet *ifp)
1740 {
1741 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1742 
1743 	if (sc->sc_ifflags & IFF_DEBUG)
1744 		printf("%s\n", __func__);
1745 
1746 	/* Ignore */
1747 	return (0);
1748 }
1749 
1750 static void
1751 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1752 {
1753 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1754 	struct lagg_port *lp;
1755 	struct rm_priotracker tracker;
1756 
1757 	imr->ifm_status = IFM_AVALID;
1758 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1759 
1760 	LAGG_RLOCK(sc, &tracker);
1761 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1762 		if (LAGG_PORTACTIVE(lp))
1763 			imr->ifm_status |= IFM_ACTIVE;
1764 	}
1765 	LAGG_RUNLOCK(sc, &tracker);
1766 }
1767 
1768 static void
1769 lagg_linkstate(struct lagg_softc *sc)
1770 {
1771 	struct lagg_port *lp;
1772 	int new_link = LINK_STATE_DOWN;
1773 	uint64_t speed;
1774 
1775 	/* Our link is considered up if at least one of our ports is active */
1776 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1777 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1778 			new_link = LINK_STATE_UP;
1779 			break;
1780 		}
1781 	}
1782 	if_link_state_change(sc->sc_ifp, new_link);
1783 
1784 	/* Update if_baudrate to reflect the max possible speed */
1785 	switch (sc->sc_proto) {
1786 		case LAGG_PROTO_FAILOVER:
1787 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1788 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1789 			break;
1790 		case LAGG_PROTO_ROUNDROBIN:
1791 		case LAGG_PROTO_LOADBALANCE:
1792 		case LAGG_PROTO_BROADCAST:
1793 			speed = 0;
1794 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1795 				speed += lp->lp_ifp->if_baudrate;
1796 			sc->sc_ifp->if_baudrate = speed;
1797 			break;
1798 		case LAGG_PROTO_LACP:
1799 			/* LACP updates if_baudrate itself */
1800 			break;
1801 	}
1802 }
1803 
1804 static void
1805 lagg_port_state(struct ifnet *ifp, int state)
1806 {
1807 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1808 	struct lagg_softc *sc = NULL;
1809 
1810 	if (lp != NULL)
1811 		sc = lp->lp_softc;
1812 	if (sc == NULL)
1813 		return;
1814 
1815 	LAGG_WLOCK(sc);
1816 	lagg_linkstate(sc);
1817 	lagg_proto_linkstate(sc, lp);
1818 	LAGG_WUNLOCK(sc);
1819 }
1820 
1821 struct lagg_port *
1822 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1823 {
1824 	struct lagg_port *lp_next, *rval = NULL;
1825 	// int new_link = LINK_STATE_DOWN;
1826 
1827 	LAGG_RLOCK_ASSERT(sc);
1828 	/*
1829 	 * Search a port which reports an active link state.
1830 	 */
1831 
1832 	if (lp == NULL)
1833 		goto search;
1834 	if (LAGG_PORTACTIVE(lp)) {
1835 		rval = lp;
1836 		goto found;
1837 	}
1838 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1839 	    LAGG_PORTACTIVE(lp_next)) {
1840 		rval = lp_next;
1841 		goto found;
1842 	}
1843 
1844 search:
1845 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1846 		if (LAGG_PORTACTIVE(lp_next)) {
1847 			rval = lp_next;
1848 			goto found;
1849 		}
1850 	}
1851 
1852 found:
1853 	if (rval != NULL) {
1854 		/*
1855 		 * The IEEE 802.1D standard assumes that a lagg with
1856 		 * multiple ports is always full duplex. This is valid
1857 		 * for load sharing laggs and if at least two links
1858 		 * are active. Unfortunately, checking the latter would
1859 		 * be too expensive at this point.
1860 		 XXX
1861 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1862 		    (sc->sc_count > 1))
1863 			new_link = LINK_STATE_FULL_DUPLEX;
1864 		else
1865 			new_link = rval->lp_link_state;
1866 		 */
1867 	}
1868 
1869 	return (rval);
1870 }
1871 
1872 int
1873 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1874 {
1875 
1876 	return (ifp->if_transmit)(ifp, m);
1877 }
1878 
1879 /*
1880  * Simple round robin aggregation
1881  */
1882 static void
1883 lagg_rr_attach(struct lagg_softc *sc)
1884 {
1885 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1886 	sc->sc_seq = 0;
1887 }
1888 
1889 static int
1890 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1891 {
1892 	struct lagg_port *lp;
1893 	uint32_t p;
1894 
1895 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1896 	p %= sc->sc_count;
1897 	lp = SLIST_FIRST(&sc->sc_ports);
1898 	while (p--)
1899 		lp = SLIST_NEXT(lp, lp_entries);
1900 
1901 	/*
1902 	 * Check the port's link state. This will return the next active
1903 	 * port if the link is down or the port is NULL.
1904 	 */
1905 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1906 		m_freem(m);
1907 		return (ENETDOWN);
1908 	}
1909 
1910 	/* Send mbuf */
1911 	return (lagg_enqueue(lp->lp_ifp, m));
1912 }
1913 
1914 static struct mbuf *
1915 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1916 {
1917 	struct ifnet *ifp = sc->sc_ifp;
1918 
1919 	/* Just pass in the packet to our lagg device */
1920 	m->m_pkthdr.rcvif = ifp;
1921 
1922 	return (m);
1923 }
1924 
1925 /*
1926  * Broadcast mode
1927  */
1928 static int
1929 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1930 {
1931 	int active_ports = 0;
1932 	int errors = 0;
1933 	int ret;
1934 	struct lagg_port *lp, *last = NULL;
1935 	struct mbuf *m0;
1936 
1937 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1938 		if (!LAGG_PORTACTIVE(lp))
1939 			continue;
1940 
1941 		active_ports++;
1942 
1943 		if (last != NULL) {
1944 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
1945 			if (m0 == NULL) {
1946 				ret = ENOBUFS;
1947 				errors++;
1948 				break;
1949 			}
1950 
1951 			ret = lagg_enqueue(last->lp_ifp, m0);
1952 			if (ret != 0)
1953 				errors++;
1954 		}
1955 		last = lp;
1956 	}
1957 	if (last == NULL) {
1958 		m_freem(m);
1959 		return (ENOENT);
1960 	}
1961 	if ((last = lagg_link_active(sc, last)) == NULL) {
1962 		m_freem(m);
1963 		return (ENETDOWN);
1964 	}
1965 
1966 	ret = lagg_enqueue(last->lp_ifp, m);
1967 	if (ret != 0)
1968 		errors++;
1969 
1970 	if (errors == 0)
1971 		return (ret);
1972 
1973 	return (0);
1974 }
1975 
1976 static struct mbuf*
1977 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1978 {
1979 	struct ifnet *ifp = sc->sc_ifp;
1980 
1981 	/* Just pass in the packet to our lagg device */
1982 	m->m_pkthdr.rcvif = ifp;
1983 	return (m);
1984 }
1985 
1986 /*
1987  * Active failover
1988  */
1989 static int
1990 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1991 {
1992 	struct lagg_port *lp;
1993 
1994 	/* Use the master port if active or the next available port */
1995 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1996 		m_freem(m);
1997 		return (ENETDOWN);
1998 	}
1999 
2000 	/* Send mbuf */
2001 	return (lagg_enqueue(lp->lp_ifp, m));
2002 }
2003 
2004 static struct mbuf *
2005 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2006 {
2007 	struct ifnet *ifp = sc->sc_ifp;
2008 	struct lagg_port *tmp_tp;
2009 
2010 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2011 		m->m_pkthdr.rcvif = ifp;
2012 		return (m);
2013 	}
2014 
2015 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2016 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2017 		/*
2018 		 * If tmp_tp is null, we've recieved a packet when all
2019 		 * our links are down. Weird, but process it anyways.
2020 		 */
2021 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2022 			m->m_pkthdr.rcvif = ifp;
2023 			return (m);
2024 		}
2025 	}
2026 
2027 	m_freem(m);
2028 	return (NULL);
2029 }
2030 
2031 /*
2032  * Loadbalancing
2033  */
2034 static void
2035 lagg_lb_attach(struct lagg_softc *sc)
2036 {
2037 	struct lagg_port *lp;
2038 	struct lagg_lb *lb;
2039 
2040 	lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
2041 
2042 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
2043 
2044 	lb->lb_key = m_ether_tcpip_hash_init();
2045 	sc->sc_psc = lb;
2046 
2047 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2048 		lagg_lb_port_create(lp);
2049 }
2050 
2051 static void
2052 lagg_lb_detach(struct lagg_softc *sc)
2053 {
2054 	struct lagg_lb *lb;
2055 
2056 	lb = (struct lagg_lb *)sc->sc_psc;
2057 	LAGG_WUNLOCK(sc);
2058 	if (lb != NULL)
2059 		free(lb, M_DEVBUF);
2060 }
2061 
2062 static int
2063 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2064 {
2065 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2066 	struct lagg_port *lp_next;
2067 	int i = 0;
2068 
2069 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2070 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2071 		if (lp_next == lp)
2072 			continue;
2073 		if (i >= LAGG_MAX_PORTS)
2074 			return (EINVAL);
2075 		if (sc->sc_ifflags & IFF_DEBUG)
2076 			printf("%s: port %s at index %d\n",
2077 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2078 		lb->lb_ports[i++] = lp_next;
2079 	}
2080 
2081 	return (0);
2082 }
2083 
2084 static int
2085 lagg_lb_port_create(struct lagg_port *lp)
2086 {
2087 	struct lagg_softc *sc = lp->lp_softc;
2088 	return (lagg_lb_porttable(sc, NULL));
2089 }
2090 
2091 static void
2092 lagg_lb_port_destroy(struct lagg_port *lp)
2093 {
2094 	struct lagg_softc *sc = lp->lp_softc;
2095 	lagg_lb_porttable(sc, lp);
2096 }
2097 
2098 static int
2099 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2100 {
2101 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2102 	struct lagg_port *lp = NULL;
2103 	uint32_t p = 0;
2104 
2105 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2106 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2107 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2108 	else
2109 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2110 	p %= sc->sc_count;
2111 	lp = lb->lb_ports[p];
2112 
2113 	/*
2114 	 * Check the port's link state. This will return the next active
2115 	 * port if the link is down or the port is NULL.
2116 	 */
2117 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2118 		m_freem(m);
2119 		return (ENETDOWN);
2120 	}
2121 
2122 	/* Send mbuf */
2123 	return (lagg_enqueue(lp->lp_ifp, m));
2124 }
2125 
2126 static struct mbuf *
2127 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2128 {
2129 	struct ifnet *ifp = sc->sc_ifp;
2130 
2131 	/* Just pass in the packet to our lagg device */
2132 	m->m_pkthdr.rcvif = ifp;
2133 
2134 	return (m);
2135 }
2136 
2137 /*
2138  * 802.3ad LACP
2139  */
2140 static void
2141 lagg_lacp_attach(struct lagg_softc *sc)
2142 {
2143 	struct lagg_port *lp;
2144 
2145 	lacp_attach(sc);
2146 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2147 		lacp_port_create(lp);
2148 }
2149 
2150 static void
2151 lagg_lacp_detach(struct lagg_softc *sc)
2152 {
2153 	struct lagg_port *lp;
2154 	void *psc;
2155 
2156 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2157 		lacp_port_destroy(lp);
2158 
2159 	psc = sc->sc_psc;
2160 	sc->sc_psc = NULL;
2161 	LAGG_WUNLOCK(sc);
2162 
2163 	lacp_detach(psc);
2164 }
2165 
2166 static void
2167 lagg_lacp_lladdr(struct lagg_softc *sc)
2168 {
2169 	struct lagg_port *lp;
2170 
2171 	/* purge all the lacp ports */
2172 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2173 		lacp_port_destroy(lp);
2174 
2175 	/* add them back in */
2176 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2177 		lacp_port_create(lp);
2178 }
2179 
2180 static int
2181 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2182 {
2183 	struct lagg_port *lp;
2184 
2185 	lp = lacp_select_tx_port(sc, m);
2186 	if (lp == NULL) {
2187 		m_freem(m);
2188 		return (ENETDOWN);
2189 	}
2190 
2191 	/* Send mbuf */
2192 	return (lagg_enqueue(lp->lp_ifp, m));
2193 }
2194 
2195 static struct mbuf *
2196 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2197 {
2198 	struct ifnet *ifp = sc->sc_ifp;
2199 	struct ether_header *eh;
2200 	u_short etype;
2201 
2202 	eh = mtod(m, struct ether_header *);
2203 	etype = ntohs(eh->ether_type);
2204 
2205 	/* Tap off LACP control messages */
2206 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2207 		m = lacp_input(lp, m);
2208 		if (m == NULL)
2209 			return (NULL);
2210 	}
2211 
2212 	/*
2213 	 * If the port is not collecting or not in the active aggregator then
2214 	 * free and return.
2215 	 */
2216 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2217 		m_freem(m);
2218 		return (NULL);
2219 	}
2220 
2221 	m->m_pkthdr.rcvif = ifp;
2222 	return (m);
2223 }
2224 
2225