xref: /freebsd/sys/net/if_lagg.c (revision 4436b51dff5736e74da464946049ea6899a88938)
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 *);
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 
547 	ifmedia_removeall(&sc->sc_media);
548 	ether_ifdetach(ifp);
549 	if_free(ifp);
550 
551 	LAGG_LIST_LOCK();
552 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
553 	LAGG_LIST_UNLOCK();
554 
555 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
556 	LAGG_LOCK_DESTROY(sc);
557 	free(sc, M_DEVBUF);
558 }
559 
560 static void
561 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
562 {
563 	struct ifnet *ifp = sc->sc_ifp;
564 	struct lagg_port lp;
565 
566 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
567 		return;
568 
569 	LAGG_WLOCK_ASSERT(sc);
570 	/*
571 	 * Set the link layer address on the lagg interface.
572 	 * lagg_proto_lladdr() notifies the MAC change to
573 	 * the aggregation protocol.  iflladdr_event handler which
574 	 * may trigger gratuitous ARPs for INET will be handled in
575 	 * a taskqueue.
576 	 */
577 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
578 	lagg_proto_lladdr(sc);
579 
580 	bzero(&lp, sizeof(lp));
581 	lp.lp_ifp = sc->sc_ifp;
582 	lp.lp_softc = sc;
583 
584 	lagg_port_lladdr(&lp, lladdr);
585 }
586 
587 static void
588 lagg_capabilities(struct lagg_softc *sc)
589 {
590 	struct lagg_port *lp;
591 	int cap = ~0, ena = ~0;
592 	u_long hwa = ~0UL;
593 	struct ifnet_hw_tsomax hw_tsomax;
594 
595 	LAGG_WLOCK_ASSERT(sc);
596 
597 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
598 
599 	/* Get capabilities from the lagg ports */
600 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
601 		cap &= lp->lp_ifp->if_capabilities;
602 		ena &= lp->lp_ifp->if_capenable;
603 		hwa &= lp->lp_ifp->if_hwassist;
604 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
605 	}
606 	cap = (cap == ~0 ? 0 : cap);
607 	ena = (ena == ~0 ? 0 : ena);
608 	hwa = (hwa == ~0 ? 0 : hwa);
609 
610 	if (sc->sc_ifp->if_capabilities != cap ||
611 	    sc->sc_ifp->if_capenable != ena ||
612 	    sc->sc_ifp->if_hwassist != hwa ||
613 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
614 		sc->sc_ifp->if_capabilities = cap;
615 		sc->sc_ifp->if_capenable = ena;
616 		sc->sc_ifp->if_hwassist = hwa;
617 		getmicrotime(&sc->sc_ifp->if_lastchange);
618 
619 		if (sc->sc_ifflags & IFF_DEBUG)
620 			if_printf(sc->sc_ifp,
621 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
622 	}
623 }
624 
625 static void
626 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
627 {
628 	struct lagg_softc *sc = lp->lp_softc;
629 	struct ifnet *ifp = lp->lp_ifp;
630 	struct lagg_llq *llq;
631 	int pending = 0;
632 	int primary;
633 
634 	LAGG_WLOCK_ASSERT(sc);
635 
636 	primary = (sc->sc_primary->lp_ifp == ifp) ? 1 : 0;
637 	if (primary == 0 && (lp->lp_detaching ||
638 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0))
639 		return;
640 
641 	/* Check to make sure its not already queued to be changed */
642 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
643 		if (llq->llq_ifp == ifp) {
644 			pending = 1;
645 			break;
646 		}
647 	}
648 
649 	if (!pending) {
650 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
651 		if (llq == NULL)	/* XXX what to do */
652 			return;
653 	}
654 
655 	/* Update the lladdr even if pending, it may have changed */
656 	llq->llq_ifp = ifp;
657 	llq->llq_primary = primary;
658 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
659 
660 	if (!pending)
661 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
662 
663 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
664 }
665 
666 /*
667  * Set the interface MAC address from a taskqueue to avoid a LOR.
668  */
669 static void
670 lagg_port_setlladdr(void *arg, int pending)
671 {
672 	struct lagg_softc *sc = (struct lagg_softc *)arg;
673 	struct lagg_llq *llq, *head;
674 	struct ifnet *ifp;
675 	int error;
676 
677 	/* Grab a local reference of the queue and remove it from the softc */
678 	LAGG_WLOCK(sc);
679 	head = SLIST_FIRST(&sc->sc_llq_head);
680 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
681 	LAGG_WUNLOCK(sc);
682 
683 	/*
684 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
685 	 * unlocked as we have the only reference to it.
686 	 */
687 	for (llq = head; llq != NULL; llq = head) {
688 		ifp = llq->llq_ifp;
689 
690 		CURVNET_SET(ifp->if_vnet);
691 		if (llq->llq_primary == 0) {
692 			/*
693 			 * Set the link layer address on the laggport interface.
694 			 * if_setlladdr() triggers gratuitous ARPs for INET.
695 			 */
696 			error = if_setlladdr(ifp, llq->llq_lladdr,
697 			    ETHER_ADDR_LEN);
698 			if (error)
699 				printf("%s: setlladdr failed on %s\n", __func__,
700 				    ifp->if_xname);
701 		} else
702 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
703 		CURVNET_RESTORE();
704 		head = SLIST_NEXT(llq, llq_entries);
705 		free(llq, M_DEVBUF);
706 	}
707 }
708 
709 static int
710 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
711 {
712 	struct lagg_softc *sc_ptr;
713 	struct lagg_port *lp, *tlp;
714 	int error, i;
715 	uint64_t *pval;
716 
717 	LAGG_WLOCK_ASSERT(sc);
718 
719 	/* Limit the maximal number of lagg ports */
720 	if (sc->sc_count >= LAGG_MAX_PORTS)
721 		return (ENOSPC);
722 
723 	/* Check if port has already been associated to a lagg */
724 	if (ifp->if_lagg != NULL) {
725 		/* Port is already in the current lagg? */
726 		lp = (struct lagg_port *)ifp->if_lagg;
727 		if (lp->lp_softc == sc)
728 			return (EEXIST);
729 		return (EBUSY);
730 	}
731 
732 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
733 	if (ifp->if_type != IFT_ETHER)
734 		return (EPROTONOSUPPORT);
735 
736 	/* Allow the first Ethernet member to define the MTU */
737 	if (SLIST_EMPTY(&sc->sc_ports))
738 		sc->sc_ifp->if_mtu = ifp->if_mtu;
739 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
740 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
741 		    ifp->if_xname);
742 		return (EINVAL);
743 	}
744 
745 	if ((lp = malloc(sizeof(struct lagg_port),
746 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
747 		return (ENOMEM);
748 
749 	/* Check if port is a stacked lagg */
750 	LAGG_LIST_LOCK();
751 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
752 		if (ifp == sc_ptr->sc_ifp) {
753 			LAGG_LIST_UNLOCK();
754 			free(lp, M_DEVBUF);
755 			return (EINVAL);
756 			/* XXX disable stacking for the moment, its untested */
757 #ifdef LAGG_PORT_STACKING
758 			lp->lp_flags |= LAGG_PORT_STACK;
759 			if (lagg_port_checkstacking(sc_ptr) >=
760 			    LAGG_MAX_STACKING) {
761 				LAGG_LIST_UNLOCK();
762 				free(lp, M_DEVBUF);
763 				return (E2BIG);
764 			}
765 #endif
766 		}
767 	}
768 	LAGG_LIST_UNLOCK();
769 
770 	/* Change the interface type */
771 	lp->lp_iftype = ifp->if_type;
772 	ifp->if_type = IFT_IEEE8023ADLAG;
773 	ifp->if_lagg = lp;
774 	lp->lp_ioctl = ifp->if_ioctl;
775 	ifp->if_ioctl = lagg_port_ioctl;
776 	lp->lp_output = ifp->if_output;
777 	ifp->if_output = lagg_port_output;
778 
779 	lp->lp_ifp = ifp;
780 	lp->lp_softc = sc;
781 
782 	/* Save port link layer address */
783 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
784 
785 	if (SLIST_EMPTY(&sc->sc_ports)) {
786 		sc->sc_primary = lp;
787 		lagg_lladdr(sc, IF_LLADDR(ifp));
788 	} else {
789 		/* Update link layer address for this port */
790 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
791 	}
792 
793 	/*
794 	 * Insert into the list of ports.
795 	 * Keep ports sorted by if_index. It is handy, when configuration
796 	 * is predictable and `ifconfig laggN create ...` command
797 	 * will lead to the same result each time.
798 	 */
799 	SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
800 		if (tlp->lp_ifp->if_index < ifp->if_index && (
801 		    SLIST_NEXT(tlp, lp_entries) == NULL ||
802 		    SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index >
803 		    ifp->if_index))
804 			break;
805 	}
806 	if (tlp != NULL)
807 		SLIST_INSERT_AFTER(tlp, lp, lp_entries);
808 	else
809 		SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
810 	sc->sc_count++;
811 
812 	/* Update lagg capabilities */
813 	lagg_capabilities(sc);
814 	lagg_linkstate(sc);
815 
816 	/* Read port counters */
817 	pval = lp->port_counters.val;
818 	for (i = 0; i < IFCOUNTERS; i++, pval++)
819 		*pval = ifp->if_get_counter(ifp, i);
820 	/* Add multicast addresses and interface flags to this port */
821 	lagg_ether_cmdmulti(lp, 1);
822 	lagg_setflags(lp, 1);
823 
824 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
825 		/* Remove the port, without calling pr_delport. */
826 		lagg_port_destroy(lp, 0);
827 		return (error);
828 	}
829 
830 	return (0);
831 }
832 
833 #ifdef LAGG_PORT_STACKING
834 static int
835 lagg_port_checkstacking(struct lagg_softc *sc)
836 {
837 	struct lagg_softc *sc_ptr;
838 	struct lagg_port *lp;
839 	int m = 0;
840 
841 	LAGG_WLOCK_ASSERT(sc);
842 
843 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
844 		if (lp->lp_flags & LAGG_PORT_STACK) {
845 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
846 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
847 		}
848 	}
849 
850 	return (m + 1);
851 }
852 #endif
853 
854 static int
855 lagg_port_destroy(struct lagg_port *lp, int rundelport)
856 {
857 	struct lagg_softc *sc = lp->lp_softc;
858 	struct lagg_port *lp_ptr;
859 	struct lagg_llq *llq;
860 	struct ifnet *ifp = lp->lp_ifp;
861 	uint64_t *pval, vdiff;
862 	int i;
863 
864 	LAGG_WLOCK_ASSERT(sc);
865 
866 	if (rundelport)
867 		lagg_proto_delport(sc, lp);
868 
869 	/*
870 	 * Remove multicast addresses and interface flags from this port and
871 	 * reset the MAC address, skip if the interface is being detached.
872 	 */
873 	if (!lp->lp_detaching) {
874 		lagg_ether_cmdmulti(lp, 0);
875 		lagg_setflags(lp, 0);
876 		lagg_port_lladdr(lp, lp->lp_lladdr);
877 	}
878 
879 	/* Restore interface */
880 	ifp->if_type = lp->lp_iftype;
881 	ifp->if_ioctl = lp->lp_ioctl;
882 	ifp->if_output = lp->lp_output;
883 	ifp->if_lagg = NULL;
884 
885 	/* Update detached port counters */
886 	pval = lp->port_counters.val;
887 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
888 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
889 		sc->detached_counters.val[i] += vdiff;
890 	}
891 
892 	/* Finally, remove the port from the lagg */
893 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
894 	sc->sc_count--;
895 
896 	/* Update the primary interface */
897 	if (lp == sc->sc_primary) {
898 		uint8_t lladdr[ETHER_ADDR_LEN];
899 
900 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
901 			bzero(&lladdr, ETHER_ADDR_LEN);
902 		} else {
903 			bcopy(lp_ptr->lp_lladdr,
904 			    lladdr, ETHER_ADDR_LEN);
905 		}
906 		lagg_lladdr(sc, lladdr);
907 		sc->sc_primary = lp_ptr;
908 
909 		/* Update link layer address for each port */
910 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
911 			lagg_port_lladdr(lp_ptr, lladdr);
912 	}
913 
914 	/* Remove any pending lladdr changes from the queue */
915 	if (lp->lp_detaching) {
916 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
917 			if (llq->llq_ifp == ifp) {
918 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
919 				    llq_entries);
920 				free(llq, M_DEVBUF);
921 				break;	/* Only appears once */
922 			}
923 		}
924 	}
925 
926 	if (lp->lp_ifflags)
927 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
928 
929 	free(lp, M_DEVBUF);
930 
931 	/* Update lagg capabilities */
932 	lagg_capabilities(sc);
933 	lagg_linkstate(sc);
934 
935 	return (0);
936 }
937 
938 static int
939 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
940 {
941 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
942 	struct lagg_softc *sc;
943 	struct lagg_port *lp = NULL;
944 	int error = 0;
945 	struct rm_priotracker tracker;
946 
947 	/* Should be checked by the caller */
948 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
949 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
950 		goto fallback;
951 
952 	switch (cmd) {
953 	case SIOCGLAGGPORT:
954 		if (rp->rp_portname[0] == '\0' ||
955 		    ifunit(rp->rp_portname) != ifp) {
956 			error = EINVAL;
957 			break;
958 		}
959 
960 		LAGG_RLOCK(sc, &tracker);
961 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
962 			error = ENOENT;
963 			LAGG_RUNLOCK(sc, &tracker);
964 			break;
965 		}
966 
967 		lagg_port2req(lp, rp);
968 		LAGG_RUNLOCK(sc, &tracker);
969 		break;
970 
971 	case SIOCSIFCAP:
972 		if (lp->lp_ioctl == NULL) {
973 			error = EINVAL;
974 			break;
975 		}
976 		error = (*lp->lp_ioctl)(ifp, cmd, data);
977 		if (error)
978 			break;
979 
980 		/* Update lagg interface capabilities */
981 		LAGG_WLOCK(sc);
982 		lagg_capabilities(sc);
983 		LAGG_WUNLOCK(sc);
984 		break;
985 
986 	case SIOCSIFMTU:
987 		/* Do not allow the MTU to be changed once joined */
988 		error = EINVAL;
989 		break;
990 
991 	default:
992 		goto fallback;
993 	}
994 
995 	return (error);
996 
997 fallback:
998 	if (lp->lp_ioctl != NULL)
999 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1000 
1001 	return (EINVAL);
1002 }
1003 
1004 /*
1005  * Requests counter @cnt data.
1006  *
1007  * Counter value is calculated the following way:
1008  * 1) for each port, sum  difference between current and "initial" measurements.
1009  * 2) add lagg logical interface counters.
1010  * 3) add data from detached_counters array.
1011  *
1012  * We also do the following things on ports attach/detach:
1013  * 1) On port attach we store all counters it has into port_counter array.
1014  * 2) On port detach we add the different between "initial" and
1015  *   current counters data to detached_counters array.
1016  */
1017 static uint64_t
1018 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1019 {
1020 	struct lagg_softc *sc;
1021 	struct lagg_port *lp;
1022 	struct ifnet *lpifp;
1023 	struct rm_priotracker tracker;
1024 	uint64_t newval, oldval, vsum;
1025 
1026 	/* Revise this when we've got non-generic counters. */
1027 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1028 
1029 	sc = (struct lagg_softc *)ifp->if_softc;
1030 	LAGG_RLOCK(sc, &tracker);
1031 
1032 	vsum = 0;
1033 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1034 		/* Saved attached value */
1035 		oldval = lp->port_counters.val[cnt];
1036 		/* current value */
1037 		lpifp = lp->lp_ifp;
1038 		newval = lpifp->if_get_counter(lpifp, cnt);
1039 		/* Calculate diff and save new */
1040 		vsum += newval - oldval;
1041 	}
1042 
1043 	/*
1044 	 * Add counter data which might be added by upper
1045 	 * layer protocols operating on logical interface.
1046 	 */
1047 	vsum += if_get_counter_default(ifp, cnt);
1048 
1049 	/*
1050 	 * Add counter data from detached ports counters
1051 	 */
1052 	vsum += sc->detached_counters.val[cnt];
1053 
1054 	LAGG_RUNLOCK(sc, &tracker);
1055 
1056 	return (vsum);
1057 }
1058 
1059 /*
1060  * For direct output to child ports.
1061  */
1062 static int
1063 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1064 	const struct sockaddr *dst, struct route *ro)
1065 {
1066 	struct lagg_port *lp = ifp->if_lagg;
1067 
1068 	switch (dst->sa_family) {
1069 		case pseudo_AF_HDRCMPLT:
1070 		case AF_UNSPEC:
1071 			return ((*lp->lp_output)(ifp, m, dst, ro));
1072 	}
1073 
1074 	/* drop any other frames */
1075 	m_freem(m);
1076 	return (ENETDOWN);
1077 }
1078 
1079 static void
1080 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1081 {
1082 	struct lagg_port *lp;
1083 	struct lagg_softc *sc;
1084 
1085 	if ((lp = ifp->if_lagg) == NULL)
1086 		return;
1087 	/* If the ifnet is just being renamed, don't do anything. */
1088 	if (ifp->if_flags & IFF_RENAMING)
1089 		return;
1090 
1091 	sc = lp->lp_softc;
1092 
1093 	LAGG_WLOCK(sc);
1094 	lp->lp_detaching = 1;
1095 	lagg_port_destroy(lp, 1);
1096 	LAGG_WUNLOCK(sc);
1097 }
1098 
1099 static void
1100 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1101 {
1102 	struct lagg_softc *sc = lp->lp_softc;
1103 
1104 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1105 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1106 	rp->rp_prio = lp->lp_prio;
1107 	rp->rp_flags = lp->lp_flags;
1108 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1109 
1110 	/* Add protocol specific flags */
1111 	switch (sc->sc_proto) {
1112 		case LAGG_PROTO_FAILOVER:
1113 			if (lp == sc->sc_primary)
1114 				rp->rp_flags |= LAGG_PORT_MASTER;
1115 			if (lp == lagg_link_active(sc, sc->sc_primary))
1116 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1117 			break;
1118 
1119 		case LAGG_PROTO_ROUNDROBIN:
1120 		case LAGG_PROTO_LOADBALANCE:
1121 		case LAGG_PROTO_BROADCAST:
1122 			if (LAGG_PORTACTIVE(lp))
1123 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1124 			break;
1125 
1126 		case LAGG_PROTO_LACP:
1127 			/* LACP has a different definition of active */
1128 			if (lacp_isactive(lp))
1129 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1130 			if (lacp_iscollecting(lp))
1131 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1132 			if (lacp_isdistributing(lp))
1133 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1134 			break;
1135 	}
1136 
1137 }
1138 
1139 static void
1140 lagg_init(void *xsc)
1141 {
1142 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1143 	struct lagg_port *lp;
1144 	struct ifnet *ifp = sc->sc_ifp;
1145 
1146 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1147 		return;
1148 
1149 	LAGG_WLOCK(sc);
1150 
1151 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1152 	/* Update the port lladdrs */
1153 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1154 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
1155 
1156 	lagg_proto_init(sc);
1157 
1158 	LAGG_WUNLOCK(sc);
1159 }
1160 
1161 static void
1162 lagg_stop(struct lagg_softc *sc)
1163 {
1164 	struct ifnet *ifp = sc->sc_ifp;
1165 
1166 	LAGG_WLOCK_ASSERT(sc);
1167 
1168 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1169 		return;
1170 
1171 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1172 
1173 	lagg_proto_stop(sc);
1174 }
1175 
1176 static int
1177 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1178 {
1179 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1180 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1181 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1182 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1183 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1184 	struct ifreq *ifr = (struct ifreq *)data;
1185 	struct lagg_port *lp;
1186 	struct ifnet *tpif;
1187 	struct thread *td = curthread;
1188 	char *buf, *outbuf;
1189 	int count, buflen, len, error = 0;
1190 	struct rm_priotracker tracker;
1191 
1192 	bzero(&rpbuf, sizeof(rpbuf));
1193 
1194 	switch (cmd) {
1195 	case SIOCGLAGG:
1196 		LAGG_RLOCK(sc, &tracker);
1197 		count = 0;
1198 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1199 			count++;
1200 		buflen = count * sizeof(struct lagg_reqport);
1201 		LAGG_RUNLOCK(sc, &tracker);
1202 
1203 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1204 
1205 		LAGG_RLOCK(sc, &tracker);
1206 		ra->ra_proto = sc->sc_proto;
1207 		lagg_proto_request(sc, &ra->ra_psc);
1208 		count = 0;
1209 		buf = outbuf;
1210 		len = min(ra->ra_size, buflen);
1211 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1212 			if (len < sizeof(rpbuf))
1213 				break;
1214 
1215 			lagg_port2req(lp, &rpbuf);
1216 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1217 			count++;
1218 			buf += sizeof(rpbuf);
1219 			len -= sizeof(rpbuf);
1220 		}
1221 		LAGG_RUNLOCK(sc, &tracker);
1222 		ra->ra_ports = count;
1223 		ra->ra_size = count * sizeof(rpbuf);
1224 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1225 		free(outbuf, M_TEMP);
1226 		break;
1227 	case SIOCSLAGG:
1228 		error = priv_check(td, PRIV_NET_LAGG);
1229 		if (error)
1230 			break;
1231 		if (ra->ra_proto < 1 || ra->ra_proto >= LAGG_PROTO_MAX) {
1232 			error = EPROTONOSUPPORT;
1233 			break;
1234 		}
1235 
1236 		LAGG_WLOCK(sc);
1237 		lagg_proto_detach(sc);
1238 		lagg_proto_attach(sc, ra->ra_proto);
1239 		break;
1240 	case SIOCGLAGGOPTS:
1241 		ro->ro_opts = sc->sc_opts;
1242 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1243 			struct lacp_softc *lsc;
1244 
1245 			lsc = (struct lacp_softc *)sc->sc_psc;
1246 			if (lsc->lsc_debug.lsc_tx_test != 0)
1247 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1248 			if (lsc->lsc_debug.lsc_rx_test != 0)
1249 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1250 			if (lsc->lsc_strict_mode != 0)
1251 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1252 			if (lsc->lsc_fast_timeout != 0)
1253 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1254 
1255 			ro->ro_active = sc->sc_active;
1256 		} else {
1257 			ro->ro_active = 0;
1258 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1259 				ro->ro_active += LAGG_PORTACTIVE(lp);
1260 		}
1261 		ro->ro_flapping = sc->sc_flapping;
1262 		ro->ro_flowid_shift = sc->flowid_shift;
1263 		break;
1264 	case SIOCSLAGGOPTS:
1265 		error = priv_check(td, PRIV_NET_LAGG);
1266 		if (error)
1267 			break;
1268 		if (ro->ro_opts == 0)
1269 			break;
1270 		/*
1271 		 * Set options.  LACP options are stored in sc->sc_psc,
1272 		 * not in sc_opts.
1273 		 */
1274 		int valid, lacp;
1275 
1276 		switch (ro->ro_opts) {
1277 		case LAGG_OPT_USE_FLOWID:
1278 		case -LAGG_OPT_USE_FLOWID:
1279 		case LAGG_OPT_FLOWIDSHIFT:
1280 			valid = 1;
1281 			lacp = 0;
1282 			break;
1283 		case LAGG_OPT_LACP_TXTEST:
1284 		case -LAGG_OPT_LACP_TXTEST:
1285 		case LAGG_OPT_LACP_RXTEST:
1286 		case -LAGG_OPT_LACP_RXTEST:
1287 		case LAGG_OPT_LACP_STRICT:
1288 		case -LAGG_OPT_LACP_STRICT:
1289 		case LAGG_OPT_LACP_TIMEOUT:
1290 		case -LAGG_OPT_LACP_TIMEOUT:
1291 			valid = lacp = 1;
1292 			break;
1293 		default:
1294 			valid = lacp = 0;
1295 			break;
1296 		}
1297 
1298 		LAGG_WLOCK(sc);
1299 		if (valid == 0 ||
1300 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1301 			/* Invalid combination of options specified. */
1302 			error = EINVAL;
1303 			LAGG_WUNLOCK(sc);
1304 			break;	/* Return from SIOCSLAGGOPTS. */
1305 		}
1306 		/*
1307 		 * Store new options into sc->sc_opts except for
1308 		 * FLOWIDSHIFT and LACP options.
1309 		 */
1310 		if (lacp == 0) {
1311 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1312 				sc->flowid_shift = ro->ro_flowid_shift;
1313 			else if (ro->ro_opts > 0)
1314 				sc->sc_opts |= ro->ro_opts;
1315 			else
1316 				sc->sc_opts &= ~ro->ro_opts;
1317 		} else {
1318 			struct lacp_softc *lsc;
1319 			struct lacp_port *lp;
1320 
1321 			lsc = (struct lacp_softc *)sc->sc_psc;
1322 
1323 			switch (ro->ro_opts) {
1324 			case LAGG_OPT_LACP_TXTEST:
1325 				lsc->lsc_debug.lsc_tx_test = 1;
1326 				break;
1327 			case -LAGG_OPT_LACP_TXTEST:
1328 				lsc->lsc_debug.lsc_tx_test = 0;
1329 				break;
1330 			case LAGG_OPT_LACP_RXTEST:
1331 				lsc->lsc_debug.lsc_rx_test = 1;
1332 				break;
1333 			case -LAGG_OPT_LACP_RXTEST:
1334 				lsc->lsc_debug.lsc_rx_test = 0;
1335 				break;
1336 			case LAGG_OPT_LACP_STRICT:
1337 				lsc->lsc_strict_mode = 1;
1338 				break;
1339 			case -LAGG_OPT_LACP_STRICT:
1340 				lsc->lsc_strict_mode = 0;
1341 				break;
1342 			case LAGG_OPT_LACP_TIMEOUT:
1343 				LACP_LOCK(lsc);
1344         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1345                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1346 				LACP_UNLOCK(lsc);
1347 				lsc->lsc_fast_timeout = 1;
1348 				break;
1349 			case -LAGG_OPT_LACP_TIMEOUT:
1350 				LACP_LOCK(lsc);
1351         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1352                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1353 				LACP_UNLOCK(lsc);
1354 				lsc->lsc_fast_timeout = 0;
1355 				break;
1356 			}
1357 		}
1358 		LAGG_WUNLOCK(sc);
1359 		break;
1360 	case SIOCGLAGGFLAGS:
1361 		rf->rf_flags = 0;
1362 		LAGG_RLOCK(sc, &tracker);
1363 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1364 			rf->rf_flags |= LAGG_F_HASHL2;
1365 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1366 			rf->rf_flags |= LAGG_F_HASHL3;
1367 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1368 			rf->rf_flags |= LAGG_F_HASHL4;
1369 		LAGG_RUNLOCK(sc, &tracker);
1370 		break;
1371 	case SIOCSLAGGHASH:
1372 		error = priv_check(td, PRIV_NET_LAGG);
1373 		if (error)
1374 			break;
1375 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1376 			error = EINVAL;
1377 			break;
1378 		}
1379 		LAGG_WLOCK(sc);
1380 		sc->sc_flags = 0;
1381 		if (rf->rf_flags & LAGG_F_HASHL2)
1382 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1383 		if (rf->rf_flags & LAGG_F_HASHL3)
1384 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1385 		if (rf->rf_flags & LAGG_F_HASHL4)
1386 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1387 		LAGG_WUNLOCK(sc);
1388 		break;
1389 	case SIOCGLAGGPORT:
1390 		if (rp->rp_portname[0] == '\0' ||
1391 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1392 			error = EINVAL;
1393 			break;
1394 		}
1395 
1396 		LAGG_RLOCK(sc, &tracker);
1397 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1398 		    lp->lp_softc != sc) {
1399 			error = ENOENT;
1400 			LAGG_RUNLOCK(sc, &tracker);
1401 			break;
1402 		}
1403 
1404 		lagg_port2req(lp, rp);
1405 		LAGG_RUNLOCK(sc, &tracker);
1406 		break;
1407 	case SIOCSLAGGPORT:
1408 		error = priv_check(td, PRIV_NET_LAGG);
1409 		if (error)
1410 			break;
1411 		if (rp->rp_portname[0] == '\0' ||
1412 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1413 			error = EINVAL;
1414 			break;
1415 		}
1416 #ifdef INET6
1417 		/*
1418 		 * A laggport interface should not have inet6 address
1419 		 * because two interfaces with a valid link-local
1420 		 * scope zone must not be merged in any form.  This
1421 		 * restriction is needed to prevent violation of
1422 		 * link-local scope zone.  Attempts to add a laggport
1423 		 * interface which has inet6 addresses triggers
1424 		 * removal of all inet6 addresses on the member
1425 		 * interface.
1426 		 */
1427 		if (in6ifa_llaonifp(tpif)) {
1428 			in6_ifdetach(tpif);
1429 				if_printf(sc->sc_ifp,
1430 				    "IPv6 addresses on %s have been removed "
1431 				    "before adding it as a member to prevent "
1432 				    "IPv6 address scope violation.\n",
1433 				    tpif->if_xname);
1434 		}
1435 #endif
1436 		LAGG_WLOCK(sc);
1437 		error = lagg_port_create(sc, tpif);
1438 		LAGG_WUNLOCK(sc);
1439 		break;
1440 	case SIOCSLAGGDELPORT:
1441 		error = priv_check(td, PRIV_NET_LAGG);
1442 		if (error)
1443 			break;
1444 		if (rp->rp_portname[0] == '\0' ||
1445 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1446 			error = EINVAL;
1447 			break;
1448 		}
1449 
1450 		LAGG_WLOCK(sc);
1451 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1452 		    lp->lp_softc != sc) {
1453 			error = ENOENT;
1454 			LAGG_WUNLOCK(sc);
1455 			break;
1456 		}
1457 
1458 		error = lagg_port_destroy(lp, 1);
1459 		LAGG_WUNLOCK(sc);
1460 		break;
1461 	case SIOCSIFFLAGS:
1462 		/* Set flags on ports too */
1463 		LAGG_WLOCK(sc);
1464 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1465 			lagg_setflags(lp, 1);
1466 		}
1467 		LAGG_WUNLOCK(sc);
1468 
1469 		if (!(ifp->if_flags & IFF_UP) &&
1470 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1471 			/*
1472 			 * If interface is marked down and it is running,
1473 			 * then stop and disable it.
1474 			 */
1475 			LAGG_WLOCK(sc);
1476 			lagg_stop(sc);
1477 			LAGG_WUNLOCK(sc);
1478 		} else if ((ifp->if_flags & IFF_UP) &&
1479 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1480 			/*
1481 			 * If interface is marked up and it is stopped, then
1482 			 * start it.
1483 			 */
1484 			(*ifp->if_init)(sc);
1485 		}
1486 		break;
1487 	case SIOCADDMULTI:
1488 	case SIOCDELMULTI:
1489 		LAGG_WLOCK(sc);
1490 		error = lagg_ether_setmulti(sc);
1491 		LAGG_WUNLOCK(sc);
1492 		break;
1493 	case SIOCSIFMEDIA:
1494 	case SIOCGIFMEDIA:
1495 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1496 		break;
1497 
1498 	case SIOCSIFCAP:
1499 	case SIOCSIFMTU:
1500 		/* Do not allow the MTU or caps to be directly changed */
1501 		error = EINVAL;
1502 		break;
1503 
1504 	default:
1505 		error = ether_ioctl(ifp, cmd, data);
1506 		break;
1507 	}
1508 	return (error);
1509 }
1510 
1511 static int
1512 lagg_ether_setmulti(struct lagg_softc *sc)
1513 {
1514 	struct lagg_port *lp;
1515 
1516 	LAGG_WLOCK_ASSERT(sc);
1517 
1518 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1519 		/* First, remove any existing filter entries. */
1520 		lagg_ether_cmdmulti(lp, 0);
1521 		/* copy all addresses from the lagg interface to the port */
1522 		lagg_ether_cmdmulti(lp, 1);
1523 	}
1524 	return (0);
1525 }
1526 
1527 static int
1528 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1529 {
1530 	struct lagg_softc *sc = lp->lp_softc;
1531 	struct ifnet *ifp = lp->lp_ifp;
1532 	struct ifnet *scifp = sc->sc_ifp;
1533 	struct lagg_mc *mc;
1534 	struct ifmultiaddr *ifma;
1535 	int error;
1536 
1537 	LAGG_WLOCK_ASSERT(sc);
1538 
1539 	if (set) {
1540 		IF_ADDR_WLOCK(scifp);
1541 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1542 			if (ifma->ifma_addr->sa_family != AF_LINK)
1543 				continue;
1544 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1545 			if (mc == NULL) {
1546 				IF_ADDR_WUNLOCK(scifp);
1547 				return (ENOMEM);
1548 			}
1549 			bcopy(ifma->ifma_addr, &mc->mc_addr,
1550 			    ifma->ifma_addr->sa_len);
1551 			mc->mc_addr.sdl_index = ifp->if_index;
1552 			mc->mc_ifma = NULL;
1553 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1554 		}
1555 		IF_ADDR_WUNLOCK(scifp);
1556 		SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1557 			error = if_addmulti(ifp,
1558 			    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1559 			if (error)
1560 				return (error);
1561 		}
1562 	} else {
1563 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1564 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1565 			if (mc->mc_ifma && !lp->lp_detaching)
1566 				if_delmulti_ifma(mc->mc_ifma);
1567 			free(mc, M_DEVBUF);
1568 		}
1569 	}
1570 	return (0);
1571 }
1572 
1573 /* Handle a ref counted flag that should be set on the lagg port as well */
1574 static int
1575 lagg_setflag(struct lagg_port *lp, int flag, int status,
1576     int (*func)(struct ifnet *, int))
1577 {
1578 	struct lagg_softc *sc = lp->lp_softc;
1579 	struct ifnet *scifp = sc->sc_ifp;
1580 	struct ifnet *ifp = lp->lp_ifp;
1581 	int error;
1582 
1583 	LAGG_WLOCK_ASSERT(sc);
1584 
1585 	status = status ? (scifp->if_flags & flag) : 0;
1586 	/* Now "status" contains the flag value or 0 */
1587 
1588 	/*
1589 	 * See if recorded ports status is different from what
1590 	 * we want it to be.  If it is, flip it.  We record ports
1591 	 * status in lp_ifflags so that we won't clear ports flag
1592 	 * we haven't set.  In fact, we don't clear or set ports
1593 	 * flags directly, but get or release references to them.
1594 	 * That's why we can be sure that recorded flags still are
1595 	 * in accord with actual ports flags.
1596 	 */
1597 	if (status != (lp->lp_ifflags & flag)) {
1598 		error = (*func)(ifp, status);
1599 		if (error)
1600 			return (error);
1601 		lp->lp_ifflags &= ~flag;
1602 		lp->lp_ifflags |= status;
1603 	}
1604 	return (0);
1605 }
1606 
1607 /*
1608  * Handle IFF_* flags that require certain changes on the lagg port
1609  * if "status" is true, update ports flags respective to the lagg
1610  * if "status" is false, forcedly clear the flags set on port.
1611  */
1612 static int
1613 lagg_setflags(struct lagg_port *lp, int status)
1614 {
1615 	int error, i;
1616 
1617 	for (i = 0; lagg_pflags[i].flag; i++) {
1618 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1619 		    status, lagg_pflags[i].func);
1620 		if (error)
1621 			return (error);
1622 	}
1623 	return (0);
1624 }
1625 
1626 static int
1627 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1628 {
1629 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1630 	int error, len, mcast;
1631 	struct rm_priotracker tracker;
1632 
1633 	len = m->m_pkthdr.len;
1634 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1635 
1636 	LAGG_RLOCK(sc, &tracker);
1637 	/* We need a Tx algorithm and at least one port */
1638 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1639 		LAGG_RUNLOCK(sc, &tracker);
1640 		m_freem(m);
1641 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1642 		return (ENXIO);
1643 	}
1644 
1645 	ETHER_BPF_MTAP(ifp, m);
1646 
1647 	error = lagg_proto_start(sc, m);
1648 	LAGG_RUNLOCK(sc, &tracker);
1649 
1650 	if (error != 0)
1651 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1652 
1653 	return (error);
1654 }
1655 
1656 /*
1657  * The ifp->if_qflush entry point for lagg(4) is no-op.
1658  */
1659 static void
1660 lagg_qflush(struct ifnet *ifp __unused)
1661 {
1662 }
1663 
1664 static struct mbuf *
1665 lagg_input(struct ifnet *ifp, struct mbuf *m)
1666 {
1667 	struct lagg_port *lp = ifp->if_lagg;
1668 	struct lagg_softc *sc = lp->lp_softc;
1669 	struct ifnet *scifp = sc->sc_ifp;
1670 	struct rm_priotracker tracker;
1671 
1672 	LAGG_RLOCK(sc, &tracker);
1673 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1674 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1675 	    sc->sc_proto == LAGG_PROTO_NONE) {
1676 		LAGG_RUNLOCK(sc, &tracker);
1677 		m_freem(m);
1678 		return (NULL);
1679 	}
1680 
1681 	ETHER_BPF_MTAP(scifp, m);
1682 
1683 	if (lp->lp_detaching != 0) {
1684 		m_freem(m);
1685 		m = NULL;
1686 	} else
1687 		m = lagg_proto_input(sc, lp, m);
1688 
1689 	if (m != NULL) {
1690 		if (scifp->if_flags & IFF_MONITOR) {
1691 			m_freem(m);
1692 			m = NULL;
1693 		}
1694 	}
1695 
1696 	LAGG_RUNLOCK(sc, &tracker);
1697 	return (m);
1698 }
1699 
1700 static int
1701 lagg_media_change(struct ifnet *ifp)
1702 {
1703 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1704 
1705 	if (sc->sc_ifflags & IFF_DEBUG)
1706 		printf("%s\n", __func__);
1707 
1708 	/* Ignore */
1709 	return (0);
1710 }
1711 
1712 static void
1713 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1714 {
1715 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1716 	struct lagg_port *lp;
1717 	struct rm_priotracker tracker;
1718 
1719 	imr->ifm_status = IFM_AVALID;
1720 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1721 
1722 	LAGG_RLOCK(sc, &tracker);
1723 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1724 		if (LAGG_PORTACTIVE(lp))
1725 			imr->ifm_status |= IFM_ACTIVE;
1726 	}
1727 	LAGG_RUNLOCK(sc, &tracker);
1728 }
1729 
1730 static void
1731 lagg_linkstate(struct lagg_softc *sc)
1732 {
1733 	struct lagg_port *lp;
1734 	int new_link = LINK_STATE_DOWN;
1735 	uint64_t speed;
1736 
1737 	/* Our link is considered up if at least one of our ports is active */
1738 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1739 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1740 			new_link = LINK_STATE_UP;
1741 			break;
1742 		}
1743 	}
1744 	if_link_state_change(sc->sc_ifp, new_link);
1745 
1746 	/* Update if_baudrate to reflect the max possible speed */
1747 	switch (sc->sc_proto) {
1748 		case LAGG_PROTO_FAILOVER:
1749 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1750 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1751 			break;
1752 		case LAGG_PROTO_ROUNDROBIN:
1753 		case LAGG_PROTO_LOADBALANCE:
1754 		case LAGG_PROTO_BROADCAST:
1755 			speed = 0;
1756 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1757 				speed += lp->lp_ifp->if_baudrate;
1758 			sc->sc_ifp->if_baudrate = speed;
1759 			break;
1760 		case LAGG_PROTO_LACP:
1761 			/* LACP updates if_baudrate itself */
1762 			break;
1763 	}
1764 }
1765 
1766 static void
1767 lagg_port_state(struct ifnet *ifp, int state)
1768 {
1769 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1770 	struct lagg_softc *sc = NULL;
1771 
1772 	if (lp != NULL)
1773 		sc = lp->lp_softc;
1774 	if (sc == NULL)
1775 		return;
1776 
1777 	LAGG_WLOCK(sc);
1778 	lagg_linkstate(sc);
1779 	lagg_proto_linkstate(sc, lp);
1780 	LAGG_WUNLOCK(sc);
1781 }
1782 
1783 struct lagg_port *
1784 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1785 {
1786 	struct lagg_port *lp_next, *rval = NULL;
1787 	// int new_link = LINK_STATE_DOWN;
1788 
1789 	LAGG_RLOCK_ASSERT(sc);
1790 	/*
1791 	 * Search a port which reports an active link state.
1792 	 */
1793 
1794 	if (lp == NULL)
1795 		goto search;
1796 	if (LAGG_PORTACTIVE(lp)) {
1797 		rval = lp;
1798 		goto found;
1799 	}
1800 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1801 	    LAGG_PORTACTIVE(lp_next)) {
1802 		rval = lp_next;
1803 		goto found;
1804 	}
1805 
1806 search:
1807 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1808 		if (LAGG_PORTACTIVE(lp_next)) {
1809 			rval = lp_next;
1810 			goto found;
1811 		}
1812 	}
1813 
1814 found:
1815 	if (rval != NULL) {
1816 		/*
1817 		 * The IEEE 802.1D standard assumes that a lagg with
1818 		 * multiple ports is always full duplex. This is valid
1819 		 * for load sharing laggs and if at least two links
1820 		 * are active. Unfortunately, checking the latter would
1821 		 * be too expensive at this point.
1822 		 XXX
1823 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1824 		    (sc->sc_count > 1))
1825 			new_link = LINK_STATE_FULL_DUPLEX;
1826 		else
1827 			new_link = rval->lp_link_state;
1828 		 */
1829 	}
1830 
1831 	return (rval);
1832 }
1833 
1834 int
1835 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1836 {
1837 
1838 	return (ifp->if_transmit)(ifp, m);
1839 }
1840 
1841 /*
1842  * Simple round robin aggregation
1843  */
1844 static void
1845 lagg_rr_attach(struct lagg_softc *sc)
1846 {
1847 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1848 	sc->sc_seq = 0;
1849 }
1850 
1851 static int
1852 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1853 {
1854 	struct lagg_port *lp;
1855 	uint32_t p;
1856 
1857 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1858 	p %= sc->sc_count;
1859 	lp = SLIST_FIRST(&sc->sc_ports);
1860 	while (p--)
1861 		lp = SLIST_NEXT(lp, lp_entries);
1862 
1863 	/*
1864 	 * Check the port's link state. This will return the next active
1865 	 * port if the link is down or the port is NULL.
1866 	 */
1867 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1868 		m_freem(m);
1869 		return (ENETDOWN);
1870 	}
1871 
1872 	/* Send mbuf */
1873 	return (lagg_enqueue(lp->lp_ifp, m));
1874 }
1875 
1876 static struct mbuf *
1877 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1878 {
1879 	struct ifnet *ifp = sc->sc_ifp;
1880 
1881 	/* Just pass in the packet to our lagg device */
1882 	m->m_pkthdr.rcvif = ifp;
1883 
1884 	return (m);
1885 }
1886 
1887 /*
1888  * Broadcast mode
1889  */
1890 static int
1891 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1892 {
1893 	int active_ports = 0;
1894 	int errors = 0;
1895 	int ret;
1896 	struct lagg_port *lp, *last = NULL;
1897 	struct mbuf *m0;
1898 
1899 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1900 		if (!LAGG_PORTACTIVE(lp))
1901 			continue;
1902 
1903 		active_ports++;
1904 
1905 		if (last != NULL) {
1906 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
1907 			if (m0 == NULL) {
1908 				ret = ENOBUFS;
1909 				errors++;
1910 				break;
1911 			}
1912 
1913 			ret = lagg_enqueue(last->lp_ifp, m0);
1914 			if (ret != 0)
1915 				errors++;
1916 		}
1917 		last = lp;
1918 	}
1919 	if (last == NULL) {
1920 		m_freem(m);
1921 		return (ENOENT);
1922 	}
1923 	if ((last = lagg_link_active(sc, last)) == NULL) {
1924 		m_freem(m);
1925 		return (ENETDOWN);
1926 	}
1927 
1928 	ret = lagg_enqueue(last->lp_ifp, m);
1929 	if (ret != 0)
1930 		errors++;
1931 
1932 	if (errors == 0)
1933 		return (ret);
1934 
1935 	return (0);
1936 }
1937 
1938 static struct mbuf*
1939 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1940 {
1941 	struct ifnet *ifp = sc->sc_ifp;
1942 
1943 	/* Just pass in the packet to our lagg device */
1944 	m->m_pkthdr.rcvif = ifp;
1945 	return (m);
1946 }
1947 
1948 /*
1949  * Active failover
1950  */
1951 static int
1952 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1953 {
1954 	struct lagg_port *lp;
1955 
1956 	/* Use the master port if active or the next available port */
1957 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1958 		m_freem(m);
1959 		return (ENETDOWN);
1960 	}
1961 
1962 	/* Send mbuf */
1963 	return (lagg_enqueue(lp->lp_ifp, m));
1964 }
1965 
1966 static struct mbuf *
1967 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1968 {
1969 	struct ifnet *ifp = sc->sc_ifp;
1970 	struct lagg_port *tmp_tp;
1971 
1972 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
1973 		m->m_pkthdr.rcvif = ifp;
1974 		return (m);
1975 	}
1976 
1977 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1978 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1979 		/*
1980 		 * If tmp_tp is null, we've recieved a packet when all
1981 		 * our links are down. Weird, but process it anyways.
1982 		 */
1983 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1984 			m->m_pkthdr.rcvif = ifp;
1985 			return (m);
1986 		}
1987 	}
1988 
1989 	m_freem(m);
1990 	return (NULL);
1991 }
1992 
1993 /*
1994  * Loadbalancing
1995  */
1996 static void
1997 lagg_lb_attach(struct lagg_softc *sc)
1998 {
1999 	struct lagg_port *lp;
2000 	struct lagg_lb *lb;
2001 
2002 	lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
2003 
2004 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
2005 
2006 	lb->lb_key = m_ether_tcpip_hash_init();
2007 	sc->sc_psc = lb;
2008 
2009 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2010 		lagg_lb_port_create(lp);
2011 }
2012 
2013 static void
2014 lagg_lb_detach(struct lagg_softc *sc)
2015 {
2016 	struct lagg_lb *lb;
2017 
2018 	lb = (struct lagg_lb *)sc->sc_psc;
2019 	LAGG_WUNLOCK(sc);
2020 	if (lb != NULL)
2021 		free(lb, M_DEVBUF);
2022 }
2023 
2024 static int
2025 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2026 {
2027 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2028 	struct lagg_port *lp_next;
2029 	int i = 0;
2030 
2031 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2032 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2033 		if (lp_next == lp)
2034 			continue;
2035 		if (i >= LAGG_MAX_PORTS)
2036 			return (EINVAL);
2037 		if (sc->sc_ifflags & IFF_DEBUG)
2038 			printf("%s: port %s at index %d\n",
2039 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2040 		lb->lb_ports[i++] = lp_next;
2041 	}
2042 
2043 	return (0);
2044 }
2045 
2046 static int
2047 lagg_lb_port_create(struct lagg_port *lp)
2048 {
2049 	struct lagg_softc *sc = lp->lp_softc;
2050 	return (lagg_lb_porttable(sc, NULL));
2051 }
2052 
2053 static void
2054 lagg_lb_port_destroy(struct lagg_port *lp)
2055 {
2056 	struct lagg_softc *sc = lp->lp_softc;
2057 	lagg_lb_porttable(sc, lp);
2058 }
2059 
2060 static int
2061 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2062 {
2063 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2064 	struct lagg_port *lp = NULL;
2065 	uint32_t p = 0;
2066 
2067 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2068 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2069 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2070 	else
2071 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2072 	p %= sc->sc_count;
2073 	lp = lb->lb_ports[p];
2074 
2075 	/*
2076 	 * Check the port's link state. This will return the next active
2077 	 * port if the link is down or the port is NULL.
2078 	 */
2079 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2080 		m_freem(m);
2081 		return (ENETDOWN);
2082 	}
2083 
2084 	/* Send mbuf */
2085 	return (lagg_enqueue(lp->lp_ifp, m));
2086 }
2087 
2088 static struct mbuf *
2089 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2090 {
2091 	struct ifnet *ifp = sc->sc_ifp;
2092 
2093 	/* Just pass in the packet to our lagg device */
2094 	m->m_pkthdr.rcvif = ifp;
2095 
2096 	return (m);
2097 }
2098 
2099 /*
2100  * 802.3ad LACP
2101  */
2102 static void
2103 lagg_lacp_attach(struct lagg_softc *sc)
2104 {
2105 	struct lagg_port *lp;
2106 
2107 	lacp_attach(sc);
2108 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2109 		lacp_port_create(lp);
2110 }
2111 
2112 static void
2113 lagg_lacp_detach(struct lagg_softc *sc)
2114 {
2115 	struct lagg_port *lp;
2116 	void *psc;
2117 
2118 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2119 		lacp_port_destroy(lp);
2120 
2121 	psc = sc->sc_psc;
2122 	sc->sc_psc = NULL;
2123 	LAGG_WUNLOCK(sc);
2124 
2125 	lacp_detach(psc);
2126 }
2127 
2128 static void
2129 lagg_lacp_lladdr(struct lagg_softc *sc)
2130 {
2131 	struct lagg_port *lp;
2132 
2133 	/* purge all the lacp ports */
2134 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2135 		lacp_port_destroy(lp);
2136 
2137 	/* add them back in */
2138 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2139 		lacp_port_create(lp);
2140 }
2141 
2142 static int
2143 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2144 {
2145 	struct lagg_port *lp;
2146 
2147 	lp = lacp_select_tx_port(sc, m);
2148 	if (lp == NULL) {
2149 		m_freem(m);
2150 		return (ENETDOWN);
2151 	}
2152 
2153 	/* Send mbuf */
2154 	return (lagg_enqueue(lp->lp_ifp, m));
2155 }
2156 
2157 static struct mbuf *
2158 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2159 {
2160 	struct ifnet *ifp = sc->sc_ifp;
2161 	struct ether_header *eh;
2162 	u_short etype;
2163 
2164 	eh = mtod(m, struct ether_header *);
2165 	etype = ntohs(eh->ether_type);
2166 
2167 	/* Tap off LACP control messages */
2168 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2169 		m = lacp_input(lp, m);
2170 		if (m == NULL)
2171 			return (NULL);
2172 	}
2173 
2174 	/*
2175 	 * If the port is not collecting or not in the active aggregator then
2176 	 * free and return.
2177 	 */
2178 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2179 		m_freem(m);
2180 		return (NULL);
2181 	}
2182 
2183 	m->m_pkthdr.rcvif = ifp;
2184 	return (m);
2185 }
2186 
2187