xref: /freebsd/sys/net/if_lagg.c (revision f02f7422801bb39f5eaab8fc383fa7b70c467ff9)
1 /*	$OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $	*/
2 
3 /*
4  * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org>
5  * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org>
6  * Copyright (c) 2014 Marcelo Araujo <araujo@FreeBSD.org>
7  *
8  * Permission to use, copy, modify, and distribute this software for any
9  * purpose with or without fee is hereby granted, provided that the above
10  * copyright notice and this permission notice appear in all copies.
11  *
12  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19  */
20 
21 #include <sys/cdefs.h>
22 __FBSDID("$FreeBSD$");
23 
24 #include "opt_inet.h"
25 #include "opt_inet6.h"
26 
27 #include <sys/param.h>
28 #include <sys/kernel.h>
29 #include <sys/malloc.h>
30 #include <sys/mbuf.h>
31 #include <sys/queue.h>
32 #include <sys/socket.h>
33 #include <sys/sockio.h>
34 #include <sys/sysctl.h>
35 #include <sys/module.h>
36 #include <sys/priv.h>
37 #include <sys/systm.h>
38 #include <sys/proc.h>
39 #include <sys/hash.h>
40 #include <sys/lock.h>
41 #include <sys/rmlock.h>
42 #include <sys/taskqueue.h>
43 #include <sys/eventhandler.h>
44 
45 #include <net/ethernet.h>
46 #include <net/if.h>
47 #include <net/if_clone.h>
48 #include <net/if_arp.h>
49 #include <net/if_dl.h>
50 #include <net/if_media.h>
51 #include <net/if_types.h>
52 #include <net/if_var.h>
53 #include <net/bpf.h>
54 
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 SLIST_HEAD(__trhead, lagg_softc) lagg_list;	/* list of laggs */
85 static struct mtx	lagg_list_mtx;
86 eventhandler_tag	lagg_detach_cookie = NULL;
87 
88 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
89 static void	lagg_clone_destroy(struct ifnet *);
90 static struct if_clone *lagg_cloner;
91 static const char laggname[] = "lagg";
92 
93 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
94 static void	lagg_capabilities(struct lagg_softc *);
95 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
96 static void	lagg_port_setlladdr(void *, int);
97 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
98 static int	lagg_port_destroy(struct lagg_port *, int);
99 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
100 static void	lagg_linkstate(struct lagg_softc *);
101 static void	lagg_port_state(struct ifnet *, int);
102 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
103 static int	lagg_port_output(struct ifnet *, struct mbuf *,
104 		    const struct sockaddr *, struct route *);
105 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
106 #ifdef LAGG_PORT_STACKING
107 static int	lagg_port_checkstacking(struct lagg_softc *);
108 #endif
109 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
110 static void	lagg_init(void *);
111 static void	lagg_stop(struct lagg_softc *);
112 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
113 static int	lagg_ether_setmulti(struct lagg_softc *);
114 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
115 static	int	lagg_setflag(struct lagg_port *, int, int,
116 		    int (*func)(struct ifnet *, int));
117 static	int	lagg_setflags(struct lagg_port *, int status);
118 static int	lagg_transmit(struct ifnet *, struct mbuf *);
119 static void	lagg_qflush(struct ifnet *);
120 static int	lagg_media_change(struct ifnet *);
121 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
122 static struct lagg_port *lagg_link_active(struct lagg_softc *,
123 	    struct lagg_port *);
124 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
125 static int	lagg_sysctl_active(SYSCTL_HANDLER_ARGS);
126 
127 /* Simple round robin */
128 static int	lagg_rr_attach(struct lagg_softc *);
129 static int	lagg_rr_detach(struct lagg_softc *);
130 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
131 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
132 		    struct mbuf *);
133 
134 /* Active failover */
135 static int	lagg_fail_attach(struct lagg_softc *);
136 static int	lagg_fail_detach(struct lagg_softc *);
137 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
138 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
139 		    struct mbuf *);
140 
141 /* Loadbalancing */
142 static int	lagg_lb_attach(struct lagg_softc *);
143 static int	lagg_lb_detach(struct lagg_softc *);
144 static int	lagg_lb_port_create(struct lagg_port *);
145 static void	lagg_lb_port_destroy(struct lagg_port *);
146 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
147 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
148 		    struct mbuf *);
149 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
150 
151 /* Broadcast */
152 static int    lagg_bcast_attach(struct lagg_softc *);
153 static int    lagg_bcast_detach(struct lagg_softc *);
154 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
155 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
156                   struct mbuf *);
157 
158 /* 802.3ad LACP */
159 static int	lagg_lacp_attach(struct lagg_softc *);
160 static int	lagg_lacp_detach(struct lagg_softc *);
161 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
162 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
163 		    struct mbuf *);
164 static void	lagg_lacp_lladdr(struct lagg_softc *);
165 
166 static void	lagg_callout(void *);
167 
168 /* lagg protocol table */
169 static const struct {
170 	int			ti_proto;
171 	int			(*ti_attach)(struct lagg_softc *);
172 } lagg_protos[] = {
173 	{ LAGG_PROTO_ROUNDROBIN,	lagg_rr_attach },
174 	{ LAGG_PROTO_FAILOVER,	lagg_fail_attach },
175 	{ LAGG_PROTO_LOADBALANCE,	lagg_lb_attach },
176 	{ LAGG_PROTO_ETHERCHANNEL,	lagg_lb_attach },
177 	{ LAGG_PROTO_LACP,		lagg_lacp_attach },
178 	{ LAGG_PROTO_BROADCAST,	lagg_bcast_attach },
179 	{ LAGG_PROTO_NONE,		NULL }
180 };
181 
182 SYSCTL_DECL(_net_link);
183 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
184     "Link Aggregation");
185 
186 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */
187 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW,
188     &lagg_failover_rx_all, 0,
189     "Accept input from any interface in a failover lagg");
190 static int def_use_flowid = 1; /* Default value for using M_FLOWID */
191 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
192     &def_use_flowid, 0,
193     "Default setting for using flow id for load sharing");
194 static int def_flowid_shift = 16; /* Default value for using M_FLOWID */
195 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
196     &def_flowid_shift, 0,
197     "Default setting for flowid shift for load sharing");
198 
199 static int
200 lagg_modevent(module_t mod, int type, void *data)
201 {
202 
203 	switch (type) {
204 	case MOD_LOAD:
205 		mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
206 		SLIST_INIT(&lagg_list);
207 		lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
208 		    lagg_clone_destroy, 0);
209 		lagg_input_p = lagg_input;
210 		lagg_linkstate_p = lagg_port_state;
211 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
212 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
213 		    EVENTHANDLER_PRI_ANY);
214 		break;
215 	case MOD_UNLOAD:
216 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
217 		    lagg_detach_cookie);
218 		if_clone_detach(lagg_cloner);
219 		lagg_input_p = NULL;
220 		lagg_linkstate_p = NULL;
221 		mtx_destroy(&lagg_list_mtx);
222 		break;
223 	default:
224 		return (EOPNOTSUPP);
225 	}
226 	return (0);
227 }
228 
229 static moduledata_t lagg_mod = {
230 	"if_lagg",
231 	lagg_modevent,
232 	0
233 };
234 
235 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
236 MODULE_VERSION(if_lagg, 1);
237 
238 /*
239  * This routine is run via an vlan
240  * config EVENT
241  */
242 static void
243 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
244 {
245         struct lagg_softc       *sc = ifp->if_softc;
246         struct lagg_port        *lp;
247         struct rm_priotracker   tracker;
248 
249         if (ifp->if_softc !=  arg)   /* Not our event */
250                 return;
251 
252         LAGG_RLOCK(sc, &tracker);
253         if (!SLIST_EMPTY(&sc->sc_ports)) {
254                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
255                         EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
256         }
257         LAGG_RUNLOCK(sc, &tracker);
258 }
259 
260 /*
261  * This routine is run via an vlan
262  * unconfig EVENT
263  */
264 static void
265 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
266 {
267         struct lagg_softc       *sc = ifp->if_softc;
268         struct lagg_port        *lp;
269         struct rm_priotracker   tracker;
270 
271         if (ifp->if_softc !=  arg)   /* Not our event */
272                 return;
273 
274         LAGG_RLOCK(sc, &tracker);
275         if (!SLIST_EMPTY(&sc->sc_ports)) {
276                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
277                         EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
278         }
279         LAGG_RUNLOCK(sc, &tracker);
280 }
281 
282 static int
283 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
284 {
285 	struct lagg_softc *sc;
286 	struct ifnet *ifp;
287 	int i, error = 0;
288 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
289 	struct sysctl_oid *oid;
290 	char num[14];			/* sufficient for 32 bits */
291 
292 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
293 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
294 	if (ifp == NULL) {
295 		free(sc, M_DEVBUF);
296 		return (ENOSPC);
297 	}
298 
299 	sc->sc_ipackets = counter_u64_alloc(M_WAITOK);
300 	sc->sc_opackets = counter_u64_alloc(M_WAITOK);
301 	sc->sc_ibytes = counter_u64_alloc(M_WAITOK);
302 	sc->sc_obytes = counter_u64_alloc(M_WAITOK);
303 
304 	sysctl_ctx_init(&sc->ctx);
305 	snprintf(num, sizeof(num), "%u", unit);
306 	sc->use_flowid = def_use_flowid;
307 	sc->flowid_shift = def_flowid_shift;
308 	sc->sc_oid = oid = SYSCTL_ADD_NODE(&sc->ctx,
309 		&SYSCTL_NODE_CHILDREN(_net_link, lagg),
310 		OID_AUTO, num, CTLFLAG_RD, NULL, "");
311 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
312 		"use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid,
313 		sc->use_flowid, "Use flow id for load sharing");
314 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
315 		"flowid_shift", CTLTYPE_INT|CTLFLAG_RW, &sc->flowid_shift,
316 		sc->flowid_shift,
317 		"Shift flowid bits to prevent multiqueue collisions");
318 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
319 		"count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count,
320 		"Total number of ports");
321 	SYSCTL_ADD_PROC(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
322 		"active", CTLTYPE_INT|CTLFLAG_RD, sc, 0, lagg_sysctl_active,
323 		"I", "Total number of active ports");
324 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
325 		"flapping", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_flapping,
326 		sc->sc_flapping, "Total number of port change events");
327 	/* Hash all layers by default */
328 	sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4;
329 
330 	sc->sc_proto = LAGG_PROTO_NONE;
331 	for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
332 		if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
333 			sc->sc_proto = lagg_protos[i].ti_proto;
334 			if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
335 				if_free(ifp);
336 				free(sc, M_DEVBUF);
337 				return (error);
338 			}
339 			break;
340 		}
341 	}
342 	LAGG_LOCK_INIT(sc);
343 	LAGG_CALLOUT_LOCK_INIT(sc);
344 	SLIST_INIT(&sc->sc_ports);
345 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
346 
347 	/*
348 	 * This uses the callout lock rather than the rmlock; one can't
349 	 * hold said rmlock during SWI.
350 	 */
351 	callout_init_mtx(&sc->sc_callout, &sc->sc_call_mtx, 0);
352 
353 	/* Initialise pseudo media types */
354 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
355 	    lagg_media_status);
356 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
357 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
358 
359 	if_initname(ifp, laggname, unit);
360 	ifp->if_softc = sc;
361 	ifp->if_transmit = lagg_transmit;
362 	ifp->if_qflush = lagg_qflush;
363 	ifp->if_init = lagg_init;
364 	ifp->if_ioctl = lagg_ioctl;
365 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
366 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
367 
368 	/*
369 	 * Attach as an ordinary ethernet device, children will be attached
370 	 * as special device IFT_IEEE8023ADLAG.
371 	 */
372 	ether_ifattach(ifp, eaddr);
373 
374 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
375 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
376 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
377 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
378 
379 	/* Insert into the global list of laggs */
380 	mtx_lock(&lagg_list_mtx);
381 	SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
382 	mtx_unlock(&lagg_list_mtx);
383 
384 	callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
385 
386 	return (0);
387 }
388 
389 static void
390 lagg_clone_destroy(struct ifnet *ifp)
391 {
392 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
393 	struct lagg_port *lp;
394 
395 	LAGG_WLOCK(sc);
396 
397 	lagg_stop(sc);
398 	ifp->if_flags &= ~IFF_UP;
399 
400 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
401 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
402 
403 	/* Shutdown and remove lagg ports */
404 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
405 		lagg_port_destroy(lp, 1);
406 	/* Unhook the aggregation protocol */
407 	if (sc->sc_detach != NULL)
408 		(*sc->sc_detach)(sc);
409 
410 	LAGG_WUNLOCK(sc);
411 
412 	sysctl_ctx_free(&sc->ctx);
413 	ifmedia_removeall(&sc->sc_media);
414 	ether_ifdetach(ifp);
415 	if_free(ifp);
416 
417 	/* This grabs sc_callout_mtx, serialising it correctly */
418 	callout_drain(&sc->sc_callout);
419 
420 	/* At this point it's drained; we can free this */
421 	counter_u64_free(sc->sc_ipackets);
422 	counter_u64_free(sc->sc_opackets);
423 	counter_u64_free(sc->sc_ibytes);
424 	counter_u64_free(sc->sc_obytes);
425 
426 	mtx_lock(&lagg_list_mtx);
427 	SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
428 	mtx_unlock(&lagg_list_mtx);
429 
430 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
431 	LAGG_LOCK_DESTROY(sc);
432 	LAGG_CALLOUT_LOCK_DESTROY(sc);
433 	free(sc, M_DEVBUF);
434 }
435 
436 static void
437 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
438 {
439 	struct ifnet *ifp = sc->sc_ifp;
440 
441 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
442 		return;
443 
444 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
445 	/* Let the protocol know the MAC has changed */
446 	if (sc->sc_lladdr != NULL)
447 		(*sc->sc_lladdr)(sc);
448 	EVENTHANDLER_INVOKE(iflladdr_event, ifp);
449 }
450 
451 static void
452 lagg_capabilities(struct lagg_softc *sc)
453 {
454 	struct lagg_port *lp;
455 	int cap = ~0, ena = ~0;
456 	u_long hwa = ~0UL;
457 	struct ifnet_hw_tsomax hw_tsomax;
458 
459 	LAGG_WLOCK_ASSERT(sc);
460 
461 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
462 
463 	/* Get capabilities from the lagg ports */
464 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
465 		cap &= lp->lp_ifp->if_capabilities;
466 		ena &= lp->lp_ifp->if_capenable;
467 		hwa &= lp->lp_ifp->if_hwassist;
468 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
469 	}
470 	cap = (cap == ~0 ? 0 : cap);
471 	ena = (ena == ~0 ? 0 : ena);
472 	hwa = (hwa == ~0 ? 0 : hwa);
473 
474 	if (sc->sc_ifp->if_capabilities != cap ||
475 	    sc->sc_ifp->if_capenable != ena ||
476 	    sc->sc_ifp->if_hwassist != hwa ||
477 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
478 		sc->sc_ifp->if_capabilities = cap;
479 		sc->sc_ifp->if_capenable = ena;
480 		sc->sc_ifp->if_hwassist = hwa;
481 		getmicrotime(&sc->sc_ifp->if_lastchange);
482 
483 		if (sc->sc_ifflags & IFF_DEBUG)
484 			if_printf(sc->sc_ifp,
485 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
486 	}
487 }
488 
489 static void
490 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
491 {
492 	struct lagg_softc *sc = lp->lp_softc;
493 	struct ifnet *ifp = lp->lp_ifp;
494 	struct lagg_llq *llq;
495 	int pending = 0;
496 
497 	LAGG_WLOCK_ASSERT(sc);
498 
499 	if (lp->lp_detaching ||
500 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
501 		return;
502 
503 	/* Check to make sure its not already queued to be changed */
504 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
505 		if (llq->llq_ifp == ifp) {
506 			pending = 1;
507 			break;
508 		}
509 	}
510 
511 	if (!pending) {
512 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
513 		if (llq == NULL)	/* XXX what to do */
514 			return;
515 	}
516 
517 	/* Update the lladdr even if pending, it may have changed */
518 	llq->llq_ifp = ifp;
519 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
520 
521 	if (!pending)
522 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
523 
524 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
525 }
526 
527 /*
528  * Set the interface MAC address from a taskqueue to avoid a LOR.
529  */
530 static void
531 lagg_port_setlladdr(void *arg, int pending)
532 {
533 	struct lagg_softc *sc = (struct lagg_softc *)arg;
534 	struct lagg_llq *llq, *head;
535 	struct ifnet *ifp;
536 	int error;
537 
538 	/* Grab a local reference of the queue and remove it from the softc */
539 	LAGG_WLOCK(sc);
540 	head = SLIST_FIRST(&sc->sc_llq_head);
541 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
542 	LAGG_WUNLOCK(sc);
543 
544 	/*
545 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
546 	 * unlocked as we have the only reference to it.
547 	 */
548 	for (llq = head; llq != NULL; llq = head) {
549 		ifp = llq->llq_ifp;
550 
551 		/* Set the link layer address */
552 		CURVNET_SET(ifp->if_vnet);
553 		error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
554 		CURVNET_RESTORE();
555 		if (error)
556 			printf("%s: setlladdr failed on %s\n", __func__,
557 			    ifp->if_xname);
558 
559 		head = SLIST_NEXT(llq, llq_entries);
560 		free(llq, M_DEVBUF);
561 	}
562 }
563 
564 static int
565 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
566 {
567 	struct lagg_softc *sc_ptr;
568 	struct lagg_port *lp;
569 	int error = 0;
570 
571 	LAGG_WLOCK_ASSERT(sc);
572 
573 	/* Limit the maximal number of lagg ports */
574 	if (sc->sc_count >= LAGG_MAX_PORTS)
575 		return (ENOSPC);
576 
577 	/* Check if port has already been associated to a lagg */
578 	if (ifp->if_lagg != NULL) {
579 		/* Port is already in the current lagg? */
580 		lp = (struct lagg_port *)ifp->if_lagg;
581 		if (lp->lp_softc == sc)
582 			return (EEXIST);
583 		return (EBUSY);
584 	}
585 
586 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
587 	if (ifp->if_type != IFT_ETHER)
588 		return (EPROTONOSUPPORT);
589 
590 #ifdef INET6
591 	/*
592 	 * The member interface should not have inet6 address because
593 	 * two interfaces with a valid link-local scope zone must not be
594 	 * merged in any form.  This restriction is needed to
595 	 * prevent violation of link-local scope zone.  Attempts to
596 	 * add a member interface which has inet6 addresses triggers
597 	 * removal of all inet6 addresses on the member interface.
598 	 */
599 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
600 		if (in6ifa_llaonifp(lp->lp_ifp)) {
601 			in6_ifdetach(lp->lp_ifp);
602 			if_printf(sc->sc_ifp,
603 			    "IPv6 addresses on %s have been removed "
604 			    "before adding it as a member to prevent "
605 			    "IPv6 address scope violation.\n",
606 			    lp->lp_ifp->if_xname);
607 		}
608 	}
609 	if (in6ifa_llaonifp(ifp)) {
610 		in6_ifdetach(ifp);
611 		if_printf(sc->sc_ifp,
612 		    "IPv6 addresses on %s have been removed "
613 		    "before adding it as a member to prevent "
614 		    "IPv6 address scope violation.\n",
615 		    ifp->if_xname);
616 	}
617 #endif
618 	/* Allow the first Ethernet member to define the MTU */
619 	if (SLIST_EMPTY(&sc->sc_ports))
620 		sc->sc_ifp->if_mtu = ifp->if_mtu;
621 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
622 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
623 		    ifp->if_xname);
624 		return (EINVAL);
625 	}
626 
627 	if ((lp = malloc(sizeof(struct lagg_port),
628 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
629 		return (ENOMEM);
630 
631 	/* Check if port is a stacked lagg */
632 	mtx_lock(&lagg_list_mtx);
633 	SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
634 		if (ifp == sc_ptr->sc_ifp) {
635 			mtx_unlock(&lagg_list_mtx);
636 			free(lp, M_DEVBUF);
637 			return (EINVAL);
638 			/* XXX disable stacking for the moment, its untested */
639 #ifdef LAGG_PORT_STACKING
640 			lp->lp_flags |= LAGG_PORT_STACK;
641 			if (lagg_port_checkstacking(sc_ptr) >=
642 			    LAGG_MAX_STACKING) {
643 				mtx_unlock(&lagg_list_mtx);
644 				free(lp, M_DEVBUF);
645 				return (E2BIG);
646 			}
647 #endif
648 		}
649 	}
650 	mtx_unlock(&lagg_list_mtx);
651 
652 	/* Change the interface type */
653 	lp->lp_iftype = ifp->if_type;
654 	ifp->if_type = IFT_IEEE8023ADLAG;
655 	ifp->if_lagg = lp;
656 	lp->lp_ioctl = ifp->if_ioctl;
657 	ifp->if_ioctl = lagg_port_ioctl;
658 	lp->lp_output = ifp->if_output;
659 	ifp->if_output = lagg_port_output;
660 
661 	lp->lp_ifp = ifp;
662 	lp->lp_softc = sc;
663 
664 	/* Save port link layer address */
665 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
666 
667 	if (SLIST_EMPTY(&sc->sc_ports)) {
668 		sc->sc_primary = lp;
669 		lagg_lladdr(sc, IF_LLADDR(ifp));
670 	} else {
671 		/* Update link layer address for this port */
672 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
673 	}
674 
675 	/* Insert into the list of ports */
676 	SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
677 	sc->sc_count++;
678 
679 	/* Update lagg capabilities */
680 	lagg_capabilities(sc);
681 	lagg_linkstate(sc);
682 
683 	/* Add multicast addresses and interface flags to this port */
684 	lagg_ether_cmdmulti(lp, 1);
685 	lagg_setflags(lp, 1);
686 
687 	if (sc->sc_port_create != NULL)
688 		error = (*sc->sc_port_create)(lp);
689 	if (error) {
690 		/* remove the port again, without calling sc_port_destroy */
691 		lagg_port_destroy(lp, 0);
692 		return (error);
693 	}
694 
695 	return (error);
696 }
697 
698 #ifdef LAGG_PORT_STACKING
699 static int
700 lagg_port_checkstacking(struct lagg_softc *sc)
701 {
702 	struct lagg_softc *sc_ptr;
703 	struct lagg_port *lp;
704 	int m = 0;
705 
706 	LAGG_WLOCK_ASSERT(sc);
707 
708 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
709 		if (lp->lp_flags & LAGG_PORT_STACK) {
710 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
711 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
712 		}
713 	}
714 
715 	return (m + 1);
716 }
717 #endif
718 
719 static int
720 lagg_port_destroy(struct lagg_port *lp, int runpd)
721 {
722 	struct lagg_softc *sc = lp->lp_softc;
723 	struct lagg_port *lp_ptr;
724 	struct lagg_llq *llq;
725 	struct ifnet *ifp = lp->lp_ifp;
726 
727 	LAGG_WLOCK_ASSERT(sc);
728 
729 	if (runpd && sc->sc_port_destroy != NULL)
730 		(*sc->sc_port_destroy)(lp);
731 
732 	/*
733 	 * Remove multicast addresses and interface flags from this port and
734 	 * reset the MAC address, skip if the interface is being detached.
735 	 */
736 	if (!lp->lp_detaching) {
737 		lagg_ether_cmdmulti(lp, 0);
738 		lagg_setflags(lp, 0);
739 		lagg_port_lladdr(lp, lp->lp_lladdr);
740 	}
741 
742 	/* Restore interface */
743 	ifp->if_type = lp->lp_iftype;
744 	ifp->if_ioctl = lp->lp_ioctl;
745 	ifp->if_output = lp->lp_output;
746 	ifp->if_lagg = NULL;
747 
748 	/* Finally, remove the port from the lagg */
749 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
750 	sc->sc_count--;
751 
752 	/* Update the primary interface */
753 	if (lp == sc->sc_primary) {
754 		uint8_t lladdr[ETHER_ADDR_LEN];
755 
756 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
757 			bzero(&lladdr, ETHER_ADDR_LEN);
758 		} else {
759 			bcopy(lp_ptr->lp_lladdr,
760 			    lladdr, ETHER_ADDR_LEN);
761 		}
762 		lagg_lladdr(sc, lladdr);
763 		sc->sc_primary = lp_ptr;
764 
765 		/* Update link layer address for each port */
766 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
767 			lagg_port_lladdr(lp_ptr, lladdr);
768 	}
769 
770 	/* Remove any pending lladdr changes from the queue */
771 	if (lp->lp_detaching) {
772 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
773 			if (llq->llq_ifp == ifp) {
774 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
775 				    llq_entries);
776 				free(llq, M_DEVBUF);
777 				break;	/* Only appears once */
778 			}
779 		}
780 	}
781 
782 	if (lp->lp_ifflags)
783 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
784 
785 	free(lp, M_DEVBUF);
786 
787 	/* Update lagg capabilities */
788 	lagg_capabilities(sc);
789 	lagg_linkstate(sc);
790 
791 	return (0);
792 }
793 
794 static int
795 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
796 {
797 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
798 	struct lagg_softc *sc;
799 	struct lagg_port *lp = NULL;
800 	int error = 0;
801 	struct rm_priotracker tracker;
802 
803 	/* Should be checked by the caller */
804 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
805 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
806 		goto fallback;
807 
808 	switch (cmd) {
809 	case SIOCGLAGGPORT:
810 		if (rp->rp_portname[0] == '\0' ||
811 		    ifunit(rp->rp_portname) != ifp) {
812 			error = EINVAL;
813 			break;
814 		}
815 
816 		LAGG_RLOCK(sc, &tracker);
817 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
818 			error = ENOENT;
819 			LAGG_RUNLOCK(sc, &tracker);
820 			break;
821 		}
822 
823 		lagg_port2req(lp, rp);
824 		LAGG_RUNLOCK(sc, &tracker);
825 		break;
826 
827 	case SIOCSIFCAP:
828 		if (lp->lp_ioctl == NULL) {
829 			error = EINVAL;
830 			break;
831 		}
832 		error = (*lp->lp_ioctl)(ifp, cmd, data);
833 		if (error)
834 			break;
835 
836 		/* Update lagg interface capabilities */
837 		LAGG_WLOCK(sc);
838 		lagg_capabilities(sc);
839 		LAGG_WUNLOCK(sc);
840 		break;
841 
842 	case SIOCSIFMTU:
843 		/* Do not allow the MTU to be changed once joined */
844 		error = EINVAL;
845 		break;
846 
847 	default:
848 		goto fallback;
849 	}
850 
851 	return (error);
852 
853 fallback:
854 	if (lp->lp_ioctl != NULL)
855 		return ((*lp->lp_ioctl)(ifp, cmd, data));
856 
857 	return (EINVAL);
858 }
859 
860 /*
861  * For direct output to child ports.
862  */
863 static int
864 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
865 	const struct sockaddr *dst, struct route *ro)
866 {
867 	struct lagg_port *lp = ifp->if_lagg;
868 
869 	switch (dst->sa_family) {
870 		case pseudo_AF_HDRCMPLT:
871 		case AF_UNSPEC:
872 			return ((*lp->lp_output)(ifp, m, dst, ro));
873 	}
874 
875 	/* drop any other frames */
876 	m_freem(m);
877 	return (ENETDOWN);
878 }
879 
880 static void
881 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
882 {
883 	struct lagg_port *lp;
884 	struct lagg_softc *sc;
885 
886 	if ((lp = ifp->if_lagg) == NULL)
887 		return;
888 	/* If the ifnet is just being renamed, don't do anything. */
889 	if (ifp->if_flags & IFF_RENAMING)
890 		return;
891 
892 	sc = lp->lp_softc;
893 
894 	LAGG_WLOCK(sc);
895 	lp->lp_detaching = 1;
896 	lagg_port_destroy(lp, 1);
897 	LAGG_WUNLOCK(sc);
898 }
899 
900 static void
901 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
902 {
903 	struct lagg_softc *sc = lp->lp_softc;
904 
905 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
906 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
907 	rp->rp_prio = lp->lp_prio;
908 	rp->rp_flags = lp->lp_flags;
909 	if (sc->sc_portreq != NULL)
910 		(*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
911 
912 	/* Add protocol specific flags */
913 	switch (sc->sc_proto) {
914 		case LAGG_PROTO_FAILOVER:
915 			if (lp == sc->sc_primary)
916 				rp->rp_flags |= LAGG_PORT_MASTER;
917 			if (lp == lagg_link_active(sc, sc->sc_primary))
918 				rp->rp_flags |= LAGG_PORT_ACTIVE;
919 			break;
920 
921 		case LAGG_PROTO_ROUNDROBIN:
922 		case LAGG_PROTO_LOADBALANCE:
923 		case LAGG_PROTO_ETHERCHANNEL:
924               case LAGG_PROTO_BROADCAST:
925 			if (LAGG_PORTACTIVE(lp))
926 				rp->rp_flags |= LAGG_PORT_ACTIVE;
927 			break;
928 
929 		case LAGG_PROTO_LACP:
930 			/* LACP has a different definition of active */
931 			if (lacp_isactive(lp))
932 				rp->rp_flags |= LAGG_PORT_ACTIVE;
933 			if (lacp_iscollecting(lp))
934 				rp->rp_flags |= LAGG_PORT_COLLECTING;
935 			if (lacp_isdistributing(lp))
936 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
937 			break;
938 	}
939 
940 }
941 
942 static void
943 lagg_init(void *xsc)
944 {
945 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
946 	struct lagg_port *lp;
947 	struct ifnet *ifp = sc->sc_ifp;
948 
949 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
950 		return;
951 
952 	LAGG_WLOCK(sc);
953 
954 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
955 	/* Update the port lladdrs */
956 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
957 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
958 
959 	if (sc->sc_init != NULL)
960 		(*sc->sc_init)(sc);
961 
962 	LAGG_WUNLOCK(sc);
963 }
964 
965 static void
966 lagg_stop(struct lagg_softc *sc)
967 {
968 	struct ifnet *ifp = sc->sc_ifp;
969 
970 	LAGG_WLOCK_ASSERT(sc);
971 
972 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
973 		return;
974 
975 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
976 
977 	if (sc->sc_stop != NULL)
978 		(*sc->sc_stop)(sc);
979 }
980 
981 static int
982 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
983 {
984 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
985 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
986 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
987 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
988 	struct ifreq *ifr = (struct ifreq *)data;
989 	struct lagg_port *lp;
990 	struct ifnet *tpif;
991 	struct thread *td = curthread;
992 	char *buf, *outbuf;
993 	int count, buflen, len, error = 0;
994 	struct rm_priotracker tracker;
995 
996 	bzero(&rpbuf, sizeof(rpbuf));
997 
998 	switch (cmd) {
999 	case SIOCGLAGG:
1000 		LAGG_RLOCK(sc, &tracker);
1001 		count = 0;
1002 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1003 			count++;
1004 		buflen = count * sizeof(struct lagg_reqport);
1005 		LAGG_RUNLOCK(sc, &tracker);
1006 
1007 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1008 
1009 		LAGG_RLOCK(sc, &tracker);
1010 		ra->ra_proto = sc->sc_proto;
1011 		if (sc->sc_req != NULL)
1012 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
1013 
1014 		count = 0;
1015 		buf = outbuf;
1016 		len = min(ra->ra_size, buflen);
1017 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1018 			if (len < sizeof(rpbuf))
1019 				break;
1020 
1021 			lagg_port2req(lp, &rpbuf);
1022 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1023 			count++;
1024 			buf += sizeof(rpbuf);
1025 			len -= sizeof(rpbuf);
1026 		}
1027 		LAGG_RUNLOCK(sc, &tracker);
1028 		ra->ra_ports = count;
1029 		ra->ra_size = count * sizeof(rpbuf);
1030 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1031 		free(outbuf, M_TEMP);
1032 		break;
1033 	case SIOCSLAGG:
1034 		error = priv_check(td, PRIV_NET_LAGG);
1035 		if (error)
1036 			break;
1037 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1038 			error = EPROTONOSUPPORT;
1039 			break;
1040 		}
1041 		LAGG_WLOCK(sc);
1042 		if (sc->sc_proto != LAGG_PROTO_NONE) {
1043 			/* Reset protocol first in case detach unlocks */
1044 			sc->sc_proto = LAGG_PROTO_NONE;
1045 			error = sc->sc_detach(sc);
1046 			sc->sc_detach = NULL;
1047 			sc->sc_start = NULL;
1048 			sc->sc_input = NULL;
1049 			sc->sc_port_create = NULL;
1050 			sc->sc_port_destroy = NULL;
1051 			sc->sc_linkstate = NULL;
1052 			sc->sc_init = NULL;
1053 			sc->sc_stop = NULL;
1054 			sc->sc_lladdr = NULL;
1055 			sc->sc_req = NULL;
1056 			sc->sc_portreq = NULL;
1057 		} else if (sc->sc_input != NULL) {
1058 			/* Still detaching */
1059 			error = EBUSY;
1060 		}
1061 		if (error != 0) {
1062 			LAGG_WUNLOCK(sc);
1063 			break;
1064 		}
1065 		for (int i = 0; i < (sizeof(lagg_protos) /
1066 		    sizeof(lagg_protos[0])); i++) {
1067 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
1068 				if (sc->sc_ifflags & IFF_DEBUG)
1069 					printf("%s: using proto %u\n",
1070 					    sc->sc_ifname,
1071 					    lagg_protos[i].ti_proto);
1072 				sc->sc_proto = lagg_protos[i].ti_proto;
1073 				if (sc->sc_proto != LAGG_PROTO_NONE)
1074 					error = lagg_protos[i].ti_attach(sc);
1075 				LAGG_WUNLOCK(sc);
1076 				return (error);
1077 			}
1078 		}
1079 		LAGG_WUNLOCK(sc);
1080 		error = EPROTONOSUPPORT;
1081 		break;
1082 	case SIOCGLAGGFLAGS:
1083 		rf->rf_flags = sc->sc_flags;
1084 		break;
1085 	case SIOCSLAGGHASH:
1086 		error = priv_check(td, PRIV_NET_LAGG);
1087 		if (error)
1088 			break;
1089 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1090 			error = EINVAL;
1091 			break;
1092 		}
1093 		LAGG_WLOCK(sc);
1094 		sc->sc_flags &= ~LAGG_F_HASHMASK;
1095 		sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK;
1096 		LAGG_WUNLOCK(sc);
1097 		break;
1098 	case SIOCGLAGGPORT:
1099 		if (rp->rp_portname[0] == '\0' ||
1100 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1101 			error = EINVAL;
1102 			break;
1103 		}
1104 
1105 		LAGG_RLOCK(sc, &tracker);
1106 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1107 		    lp->lp_softc != sc) {
1108 			error = ENOENT;
1109 			LAGG_RUNLOCK(sc, &tracker);
1110 			break;
1111 		}
1112 
1113 		lagg_port2req(lp, rp);
1114 		LAGG_RUNLOCK(sc, &tracker);
1115 		break;
1116 	case SIOCSLAGGPORT:
1117 		error = priv_check(td, PRIV_NET_LAGG);
1118 		if (error)
1119 			break;
1120 		if (rp->rp_portname[0] == '\0' ||
1121 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1122 			error = EINVAL;
1123 			break;
1124 		}
1125 		LAGG_WLOCK(sc);
1126 		error = lagg_port_create(sc, tpif);
1127 		LAGG_WUNLOCK(sc);
1128 		break;
1129 	case SIOCSLAGGDELPORT:
1130 		error = priv_check(td, PRIV_NET_LAGG);
1131 		if (error)
1132 			break;
1133 		if (rp->rp_portname[0] == '\0' ||
1134 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1135 			error = EINVAL;
1136 			break;
1137 		}
1138 
1139 		LAGG_WLOCK(sc);
1140 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1141 		    lp->lp_softc != sc) {
1142 			error = ENOENT;
1143 			LAGG_WUNLOCK(sc);
1144 			break;
1145 		}
1146 
1147 		error = lagg_port_destroy(lp, 1);
1148 		LAGG_WUNLOCK(sc);
1149 		break;
1150 	case SIOCSIFFLAGS:
1151 		/* Set flags on ports too */
1152 		LAGG_WLOCK(sc);
1153 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1154 			lagg_setflags(lp, 1);
1155 		}
1156 		LAGG_WUNLOCK(sc);
1157 
1158 		if (!(ifp->if_flags & IFF_UP) &&
1159 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1160 			/*
1161 			 * If interface is marked down and it is running,
1162 			 * then stop and disable it.
1163 			 */
1164 			LAGG_WLOCK(sc);
1165 			lagg_stop(sc);
1166 			LAGG_WUNLOCK(sc);
1167 		} else if ((ifp->if_flags & IFF_UP) &&
1168 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1169 			/*
1170 			 * If interface is marked up and it is stopped, then
1171 			 * start it.
1172 			 */
1173 			(*ifp->if_init)(sc);
1174 		}
1175 		break;
1176 	case SIOCADDMULTI:
1177 	case SIOCDELMULTI:
1178 		LAGG_WLOCK(sc);
1179 		error = lagg_ether_setmulti(sc);
1180 		LAGG_WUNLOCK(sc);
1181 		break;
1182 	case SIOCSIFMEDIA:
1183 	case SIOCGIFMEDIA:
1184 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1185 		break;
1186 
1187 	case SIOCSIFCAP:
1188 	case SIOCSIFMTU:
1189 		/* Do not allow the MTU or caps to be directly changed */
1190 		error = EINVAL;
1191 		break;
1192 
1193 	default:
1194 		error = ether_ioctl(ifp, cmd, data);
1195 		break;
1196 	}
1197 	return (error);
1198 }
1199 
1200 static int
1201 lagg_ether_setmulti(struct lagg_softc *sc)
1202 {
1203 	struct lagg_port *lp;
1204 
1205 	LAGG_WLOCK_ASSERT(sc);
1206 
1207 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1208 		/* First, remove any existing filter entries. */
1209 		lagg_ether_cmdmulti(lp, 0);
1210 		/* copy all addresses from the lagg interface to the port */
1211 		lagg_ether_cmdmulti(lp, 1);
1212 	}
1213 	return (0);
1214 }
1215 
1216 static int
1217 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1218 {
1219 	struct lagg_softc *sc = lp->lp_softc;
1220 	struct ifnet *ifp = lp->lp_ifp;
1221 	struct ifnet *scifp = sc->sc_ifp;
1222 	struct lagg_mc *mc;
1223 	struct ifmultiaddr *ifma;
1224 	int error;
1225 
1226 	LAGG_WLOCK_ASSERT(sc);
1227 
1228 	if (set) {
1229 		IF_ADDR_WLOCK(scifp);
1230 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1231 			if (ifma->ifma_addr->sa_family != AF_LINK)
1232 				continue;
1233 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1234 			if (mc == NULL) {
1235 				IF_ADDR_WUNLOCK(scifp);
1236 				return (ENOMEM);
1237 			}
1238 			bcopy(ifma->ifma_addr, &mc->mc_addr,
1239 			    ifma->ifma_addr->sa_len);
1240 			mc->mc_addr.sdl_index = ifp->if_index;
1241 			mc->mc_ifma = NULL;
1242 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1243 		}
1244 		IF_ADDR_WUNLOCK(scifp);
1245 		SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1246 			error = if_addmulti(ifp,
1247 			    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1248 			if (error)
1249 				return (error);
1250 		}
1251 	} else {
1252 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1253 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1254 			if (mc->mc_ifma && !lp->lp_detaching)
1255 				if_delmulti_ifma(mc->mc_ifma);
1256 			free(mc, M_DEVBUF);
1257 		}
1258 	}
1259 	return (0);
1260 }
1261 
1262 /* Handle a ref counted flag that should be set on the lagg port as well */
1263 static int
1264 lagg_setflag(struct lagg_port *lp, int flag, int status,
1265 	     int (*func)(struct ifnet *, int))
1266 {
1267 	struct lagg_softc *sc = lp->lp_softc;
1268 	struct ifnet *scifp = sc->sc_ifp;
1269 	struct ifnet *ifp = lp->lp_ifp;
1270 	int error;
1271 
1272 	LAGG_WLOCK_ASSERT(sc);
1273 
1274 	status = status ? (scifp->if_flags & flag) : 0;
1275 	/* Now "status" contains the flag value or 0 */
1276 
1277 	/*
1278 	 * See if recorded ports status is different from what
1279 	 * we want it to be.  If it is, flip it.  We record ports
1280 	 * status in lp_ifflags so that we won't clear ports flag
1281 	 * we haven't set.  In fact, we don't clear or set ports
1282 	 * flags directly, but get or release references to them.
1283 	 * That's why we can be sure that recorded flags still are
1284 	 * in accord with actual ports flags.
1285 	 */
1286 	if (status != (lp->lp_ifflags & flag)) {
1287 		error = (*func)(ifp, status);
1288 		if (error)
1289 			return (error);
1290 		lp->lp_ifflags &= ~flag;
1291 		lp->lp_ifflags |= status;
1292 	}
1293 	return (0);
1294 }
1295 
1296 /*
1297  * Handle IFF_* flags that require certain changes on the lagg port
1298  * if "status" is true, update ports flags respective to the lagg
1299  * if "status" is false, forcedly clear the flags set on port.
1300  */
1301 static int
1302 lagg_setflags(struct lagg_port *lp, int status)
1303 {
1304 	int error, i;
1305 
1306 	for (i = 0; lagg_pflags[i].flag; i++) {
1307 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1308 		    status, lagg_pflags[i].func);
1309 		if (error)
1310 			return (error);
1311 	}
1312 	return (0);
1313 }
1314 
1315 static int
1316 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1317 {
1318 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1319 	int error, len, mcast;
1320 	struct rm_priotracker tracker;
1321 
1322 	len = m->m_pkthdr.len;
1323 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1324 
1325 	LAGG_RLOCK(sc, &tracker);
1326 	/* We need a Tx algorithm and at least one port */
1327 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1328 		LAGG_RUNLOCK(sc, &tracker);
1329 		m_freem(m);
1330 		ifp->if_oerrors++;
1331 		return (ENXIO);
1332 	}
1333 
1334 	ETHER_BPF_MTAP(ifp, m);
1335 
1336 	error = (*sc->sc_start)(sc, m);
1337 	LAGG_RUNLOCK(sc, &tracker);
1338 
1339 	if (error == 0) {
1340 		counter_u64_add(sc->sc_opackets, 1);
1341 		counter_u64_add(sc->sc_obytes, len);
1342 		ifp->if_omcasts += mcast;
1343 	} else
1344 		ifp->if_oerrors++;
1345 
1346 	return (error);
1347 }
1348 
1349 /*
1350  * The ifp->if_qflush entry point for lagg(4) is no-op.
1351  */
1352 static void
1353 lagg_qflush(struct ifnet *ifp __unused)
1354 {
1355 }
1356 
1357 static struct mbuf *
1358 lagg_input(struct ifnet *ifp, struct mbuf *m)
1359 {
1360 	struct lagg_port *lp = ifp->if_lagg;
1361 	struct lagg_softc *sc = lp->lp_softc;
1362 	struct ifnet *scifp = sc->sc_ifp;
1363 	struct rm_priotracker tracker;
1364 
1365 	LAGG_RLOCK(sc, &tracker);
1366 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1367 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1368 	    sc->sc_proto == LAGG_PROTO_NONE) {
1369 		LAGG_RUNLOCK(sc, &tracker);
1370 		m_freem(m);
1371 		return (NULL);
1372 	}
1373 
1374 	ETHER_BPF_MTAP(scifp, m);
1375 
1376 	m = (*sc->sc_input)(sc, lp, m);
1377 
1378 	if (m != NULL) {
1379 		counter_u64_add(sc->sc_ipackets, 1);
1380 		counter_u64_add(sc->sc_ibytes, m->m_pkthdr.len);
1381 
1382 		if (scifp->if_flags & IFF_MONITOR) {
1383 			m_freem(m);
1384 			m = NULL;
1385 		}
1386 	}
1387 
1388 	LAGG_RUNLOCK(sc, &tracker);
1389 	return (m);
1390 }
1391 
1392 static int
1393 lagg_media_change(struct ifnet *ifp)
1394 {
1395 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1396 
1397 	if (sc->sc_ifflags & IFF_DEBUG)
1398 		printf("%s\n", __func__);
1399 
1400 	/* Ignore */
1401 	return (0);
1402 }
1403 
1404 static void
1405 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1406 {
1407 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1408 	struct lagg_port *lp;
1409 	struct rm_priotracker tracker;
1410 
1411 	imr->ifm_status = IFM_AVALID;
1412 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1413 
1414 	LAGG_RLOCK(sc, &tracker);
1415 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1416 		if (LAGG_PORTACTIVE(lp))
1417 			imr->ifm_status |= IFM_ACTIVE;
1418 	}
1419 	LAGG_RUNLOCK(sc, &tracker);
1420 }
1421 
1422 static void
1423 lagg_linkstate(struct lagg_softc *sc)
1424 {
1425 	struct lagg_port *lp;
1426 	int new_link = LINK_STATE_DOWN;
1427 	uint64_t speed;
1428 
1429 	/* Our link is considered up if at least one of our ports is active */
1430 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1431 		if (lp->lp_link_state == LINK_STATE_UP) {
1432 			new_link = LINK_STATE_UP;
1433 			break;
1434 		}
1435 	}
1436 	if_link_state_change(sc->sc_ifp, new_link);
1437 
1438 	/* Update if_baudrate to reflect the max possible speed */
1439 	switch (sc->sc_proto) {
1440 		case LAGG_PROTO_FAILOVER:
1441 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1442 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1443 			break;
1444 		case LAGG_PROTO_ROUNDROBIN:
1445 		case LAGG_PROTO_LOADBALANCE:
1446 		case LAGG_PROTO_ETHERCHANNEL:
1447               case LAGG_PROTO_BROADCAST:
1448 			speed = 0;
1449 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1450 				speed += lp->lp_ifp->if_baudrate;
1451 			sc->sc_ifp->if_baudrate = speed;
1452 			break;
1453 		case LAGG_PROTO_LACP:
1454 			/* LACP updates if_baudrate itself */
1455 			break;
1456 	}
1457 }
1458 
1459 static void
1460 lagg_port_state(struct ifnet *ifp, int state)
1461 {
1462 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1463 	struct lagg_softc *sc = NULL;
1464 
1465 	if (lp != NULL)
1466 		sc = lp->lp_softc;
1467 	if (sc == NULL)
1468 		return;
1469 
1470 	LAGG_WLOCK(sc);
1471 	lagg_linkstate(sc);
1472 	if (sc->sc_linkstate != NULL)
1473 		(*sc->sc_linkstate)(lp);
1474 	LAGG_WUNLOCK(sc);
1475 }
1476 
1477 struct lagg_port *
1478 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1479 {
1480 	struct lagg_port *lp_next, *rval = NULL;
1481 	// int new_link = LINK_STATE_DOWN;
1482 
1483 	LAGG_RLOCK_ASSERT(sc);
1484 	/*
1485 	 * Search a port which reports an active link state.
1486 	 */
1487 
1488 	if (lp == NULL)
1489 		goto search;
1490 	if (LAGG_PORTACTIVE(lp)) {
1491 		rval = lp;
1492 		goto found;
1493 	}
1494 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1495 	    LAGG_PORTACTIVE(lp_next)) {
1496 		rval = lp_next;
1497 		goto found;
1498 	}
1499 
1500 search:
1501 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1502 		if (LAGG_PORTACTIVE(lp_next)) {
1503 			rval = lp_next;
1504 			goto found;
1505 		}
1506 	}
1507 
1508 found:
1509 	if (rval != NULL) {
1510 		/*
1511 		 * The IEEE 802.1D standard assumes that a lagg with
1512 		 * multiple ports is always full duplex. This is valid
1513 		 * for load sharing laggs and if at least two links
1514 		 * are active. Unfortunately, checking the latter would
1515 		 * be too expensive at this point.
1516 		 XXX
1517 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1518 		    (sc->sc_count > 1))
1519 			new_link = LINK_STATE_FULL_DUPLEX;
1520 		else
1521 			new_link = rval->lp_link_state;
1522 		 */
1523 	}
1524 
1525 	return (rval);
1526 }
1527 
1528 static const void *
1529 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1530 {
1531 	if (m->m_pkthdr.len < (off + len)) {
1532 		return (NULL);
1533 	} else if (m->m_len < (off + len)) {
1534 		m_copydata(m, off, len, buf);
1535 		return (buf);
1536 	}
1537 	return (mtod(m, char *) + off);
1538 }
1539 
1540 static int
1541 lagg_sysctl_active(SYSCTL_HANDLER_ARGS)
1542 {
1543 	struct lagg_softc *sc = (struct lagg_softc *)arg1;
1544 	struct lagg_port *lp;
1545 	int error;
1546 
1547 	/* LACP tracks active links automatically, the others do not */
1548 	if (sc->sc_proto != LAGG_PROTO_LACP) {
1549 		sc->sc_active = 0;
1550 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1551 			sc->sc_active += LAGG_PORTACTIVE(lp);
1552 	}
1553 
1554 	error = sysctl_handle_int(oidp, &sc->sc_active, 0, req);
1555 	if ((error) || (req->newptr == NULL))
1556 		return (error);
1557 
1558 	return (0);
1559 }
1560 
1561 uint32_t
1562 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
1563 {
1564 	uint16_t etype;
1565 	uint32_t p = key;
1566 	int off;
1567 	struct ether_header *eh;
1568 	const struct ether_vlan_header *vlan;
1569 #ifdef INET
1570 	const struct ip *ip;
1571 	const uint32_t *ports;
1572 	int iphlen;
1573 #endif
1574 #ifdef INET6
1575 	const struct ip6_hdr *ip6;
1576 	uint32_t flow;
1577 #endif
1578 	union {
1579 #ifdef INET
1580 		struct ip ip;
1581 #endif
1582 #ifdef INET6
1583 		struct ip6_hdr ip6;
1584 #endif
1585 		struct ether_vlan_header vlan;
1586 		uint32_t port;
1587 	} buf;
1588 
1589 
1590 	off = sizeof(*eh);
1591 	if (m->m_len < off)
1592 		goto out;
1593 	eh = mtod(m, struct ether_header *);
1594 	etype = ntohs(eh->ether_type);
1595 	if (sc->sc_flags & LAGG_F_HASHL2) {
1596 		p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p);
1597 		p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1598 	}
1599 
1600 	/* Special handling for encapsulating VLAN frames */
1601 	if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) {
1602 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1603 		    sizeof(m->m_pkthdr.ether_vtag), p);
1604 	} else if (etype == ETHERTYPE_VLAN) {
1605 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &buf);
1606 		if (vlan == NULL)
1607 			goto out;
1608 
1609 		if (sc->sc_flags & LAGG_F_HASHL2)
1610 			p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1611 		etype = ntohs(vlan->evl_proto);
1612 		off += sizeof(*vlan) - sizeof(*eh);
1613 	}
1614 
1615 	switch (etype) {
1616 #ifdef INET
1617 	case ETHERTYPE_IP:
1618 		ip = lagg_gethdr(m, off, sizeof(*ip), &buf);
1619 		if (ip == NULL)
1620 			goto out;
1621 
1622 		if (sc->sc_flags & LAGG_F_HASHL3) {
1623 			p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1624 			p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1625 		}
1626 		if (!(sc->sc_flags & LAGG_F_HASHL4))
1627 			break;
1628 		switch (ip->ip_p) {
1629 			case IPPROTO_TCP:
1630 			case IPPROTO_UDP:
1631 			case IPPROTO_SCTP:
1632 				iphlen = ip->ip_hl << 2;
1633 				if (iphlen < sizeof(*ip))
1634 					break;
1635 				off += iphlen;
1636 				ports = lagg_gethdr(m, off, sizeof(*ports), &buf);
1637 				if (ports == NULL)
1638 					break;
1639 				p = hash32_buf(ports, sizeof(*ports), p);
1640 				break;
1641 		}
1642 		break;
1643 #endif
1644 #ifdef INET6
1645 	case ETHERTYPE_IPV6:
1646 		if (!(sc->sc_flags & LAGG_F_HASHL3))
1647 			break;
1648 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf);
1649 		if (ip6 == NULL)
1650 			goto out;
1651 
1652 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1653 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1654 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1655 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1656 		break;
1657 #endif
1658 	}
1659 out:
1660 	return (p);
1661 }
1662 
1663 int
1664 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1665 {
1666 
1667 	return (ifp->if_transmit)(ifp, m);
1668 }
1669 
1670 /*
1671  * Simple round robin aggregation
1672  */
1673 
1674 static int
1675 lagg_rr_attach(struct lagg_softc *sc)
1676 {
1677 	sc->sc_detach = lagg_rr_detach;
1678 	sc->sc_start = lagg_rr_start;
1679 	sc->sc_input = lagg_rr_input;
1680 	sc->sc_port_create = NULL;
1681 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1682 	sc->sc_seq = 0;
1683 
1684 	return (0);
1685 }
1686 
1687 static int
1688 lagg_rr_detach(struct lagg_softc *sc)
1689 {
1690 	return (0);
1691 }
1692 
1693 static int
1694 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1695 {
1696 	struct lagg_port *lp;
1697 	uint32_t p;
1698 
1699 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1700 	p %= sc->sc_count;
1701 	lp = SLIST_FIRST(&sc->sc_ports);
1702 	while (p--)
1703 		lp = SLIST_NEXT(lp, lp_entries);
1704 
1705 	/*
1706 	 * Check the port's link state. This will return the next active
1707 	 * port if the link is down or the port is NULL.
1708 	 */
1709 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1710 		m_freem(m);
1711 		return (ENETDOWN);
1712 	}
1713 
1714 	/* Send mbuf */
1715 	return (lagg_enqueue(lp->lp_ifp, m));
1716 }
1717 
1718 static struct mbuf *
1719 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1720 {
1721 	struct ifnet *ifp = sc->sc_ifp;
1722 
1723 	/* Just pass in the packet to our lagg device */
1724 	m->m_pkthdr.rcvif = ifp;
1725 
1726 	return (m);
1727 }
1728 
1729 /*
1730  * Broadcast mode
1731  */
1732 
1733 static int
1734 lagg_bcast_attach(struct lagg_softc *sc)
1735 {
1736        sc->sc_detach = lagg_bcast_detach;
1737        sc->sc_start = lagg_bcast_start;
1738        sc->sc_input = lagg_bcast_input;
1739        sc->sc_port_create = NULL;
1740        sc->sc_port_destroy = NULL;
1741        sc->sc_linkstate = NULL;
1742        sc->sc_req = NULL;
1743        sc->sc_portreq = NULL;
1744 
1745        return (0);
1746 }
1747 
1748 static int
1749 lagg_bcast_detach(struct lagg_softc *sc)
1750 {
1751        return (0);
1752 }
1753 
1754 static int
1755 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1756 {
1757        int active_ports = 0;
1758        int errors = 0;
1759        int ret;
1760        struct lagg_port *lp, *last = NULL;
1761        struct mbuf *m0;
1762 
1763        SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1764               if (!LAGG_PORTACTIVE(lp))
1765                      continue;
1766 
1767               active_ports++;
1768 
1769               if (last != NULL) {
1770                      m0 = m_copym(m, 0, M_COPYALL,
1771                                    M_NOWAIT);
1772                      if (m0 == NULL) {
1773                             ret = ENOBUFS;
1774                             errors++;
1775                             break;
1776                      }
1777 
1778                      ret = lagg_enqueue(last->lp_ifp, m0);
1779                      if (ret != 0)
1780                             errors++;
1781               }
1782               last = lp;
1783        }
1784        if (last == NULL) {
1785               m_freem(m);
1786               return (ENOENT);
1787        }
1788        if ((last = lagg_link_active(sc, last)) == NULL) {
1789               m_freem(m);
1790               return (ENETDOWN);
1791        }
1792 
1793        ret = lagg_enqueue(last->lp_ifp, m);
1794        if (ret != 0)
1795               errors++;
1796 
1797        if (errors == 0)
1798               return (ret);
1799 
1800        return (0);
1801 }
1802 
1803 static struct mbuf*
1804 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp,
1805                      struct mbuf *m)
1806 {
1807        struct ifnet *ifp = sc->sc_ifp;
1808 
1809        /* Just pass in the packet to our lagg device */
1810        m->m_pkthdr.rcvif = ifp;
1811        return (m);
1812 }
1813 
1814 /*
1815  * Active failover
1816  */
1817 
1818 static int
1819 lagg_fail_attach(struct lagg_softc *sc)
1820 {
1821 	sc->sc_detach = lagg_fail_detach;
1822 	sc->sc_start = lagg_fail_start;
1823 	sc->sc_input = lagg_fail_input;
1824 	sc->sc_port_create = NULL;
1825 	sc->sc_port_destroy = NULL;
1826 
1827 	return (0);
1828 }
1829 
1830 static int
1831 lagg_fail_detach(struct lagg_softc *sc)
1832 {
1833 	return (0);
1834 }
1835 
1836 static int
1837 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1838 {
1839 	struct lagg_port *lp;
1840 
1841 	/* Use the master port if active or the next available port */
1842 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1843 		m_freem(m);
1844 		return (ENETDOWN);
1845 	}
1846 
1847 	/* Send mbuf */
1848 	return (lagg_enqueue(lp->lp_ifp, m));
1849 }
1850 
1851 static struct mbuf *
1852 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1853 {
1854 	struct ifnet *ifp = sc->sc_ifp;
1855 	struct lagg_port *tmp_tp;
1856 
1857 	if (lp == sc->sc_primary || lagg_failover_rx_all) {
1858 		m->m_pkthdr.rcvif = ifp;
1859 		return (m);
1860 	}
1861 
1862 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1863 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1864 		/*
1865 		 * If tmp_tp is null, we've recieved a packet when all
1866 		 * our links are down. Weird, but process it anyways.
1867 		 */
1868 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1869 			m->m_pkthdr.rcvif = ifp;
1870 			return (m);
1871 		}
1872 	}
1873 
1874 	m_freem(m);
1875 	return (NULL);
1876 }
1877 
1878 /*
1879  * Loadbalancing
1880  */
1881 
1882 static int
1883 lagg_lb_attach(struct lagg_softc *sc)
1884 {
1885 	struct lagg_port *lp;
1886 	struct lagg_lb *lb;
1887 
1888 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1889 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1890 		return (ENOMEM);
1891 
1892 	sc->sc_detach = lagg_lb_detach;
1893 	sc->sc_start = lagg_lb_start;
1894 	sc->sc_input = lagg_lb_input;
1895 	sc->sc_port_create = lagg_lb_port_create;
1896 	sc->sc_port_destroy = lagg_lb_port_destroy;
1897 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1898 
1899 	lb->lb_key = arc4random();
1900 	sc->sc_psc = (caddr_t)lb;
1901 
1902 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1903 		lagg_lb_port_create(lp);
1904 
1905 	return (0);
1906 }
1907 
1908 static int
1909 lagg_lb_detach(struct lagg_softc *sc)
1910 {
1911 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1912 	if (lb != NULL)
1913 		free(lb, M_DEVBUF);
1914 	return (0);
1915 }
1916 
1917 static int
1918 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1919 {
1920 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1921 	struct lagg_port *lp_next;
1922 	int i = 0;
1923 
1924 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1925 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1926 		if (lp_next == lp)
1927 			continue;
1928 		if (i >= LAGG_MAX_PORTS)
1929 			return (EINVAL);
1930 		if (sc->sc_ifflags & IFF_DEBUG)
1931 			printf("%s: port %s at index %d\n",
1932 			    sc->sc_ifname, lp_next->lp_ifname, i);
1933 		lb->lb_ports[i++] = lp_next;
1934 	}
1935 
1936 	return (0);
1937 }
1938 
1939 static int
1940 lagg_lb_port_create(struct lagg_port *lp)
1941 {
1942 	struct lagg_softc *sc = lp->lp_softc;
1943 	return (lagg_lb_porttable(sc, NULL));
1944 }
1945 
1946 static void
1947 lagg_lb_port_destroy(struct lagg_port *lp)
1948 {
1949 	struct lagg_softc *sc = lp->lp_softc;
1950 	lagg_lb_porttable(sc, lp);
1951 }
1952 
1953 static int
1954 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1955 {
1956 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1957 	struct lagg_port *lp = NULL;
1958 	uint32_t p = 0;
1959 
1960 	if (sc->use_flowid && (m->m_flags & M_FLOWID))
1961 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
1962 	else
1963 		p = lagg_hashmbuf(sc, m, lb->lb_key);
1964 	p %= sc->sc_count;
1965 	lp = lb->lb_ports[p];
1966 
1967 	/*
1968 	 * Check the port's link state. This will return the next active
1969 	 * port if the link is down or the port is NULL.
1970 	 */
1971 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1972 		m_freem(m);
1973 		return (ENETDOWN);
1974 	}
1975 
1976 	/* Send mbuf */
1977 	return (lagg_enqueue(lp->lp_ifp, m));
1978 }
1979 
1980 static struct mbuf *
1981 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1982 {
1983 	struct ifnet *ifp = sc->sc_ifp;
1984 
1985 	/* Just pass in the packet to our lagg device */
1986 	m->m_pkthdr.rcvif = ifp;
1987 
1988 	return (m);
1989 }
1990 
1991 /*
1992  * 802.3ad LACP
1993  */
1994 
1995 static int
1996 lagg_lacp_attach(struct lagg_softc *sc)
1997 {
1998 	struct lagg_port *lp;
1999 	int error;
2000 
2001 	sc->sc_detach = lagg_lacp_detach;
2002 	sc->sc_port_create = lacp_port_create;
2003 	sc->sc_port_destroy = lacp_port_destroy;
2004 	sc->sc_linkstate = lacp_linkstate;
2005 	sc->sc_start = lagg_lacp_start;
2006 	sc->sc_input = lagg_lacp_input;
2007 	sc->sc_init = lacp_init;
2008 	sc->sc_stop = lacp_stop;
2009 	sc->sc_lladdr = lagg_lacp_lladdr;
2010 	sc->sc_req = lacp_req;
2011 	sc->sc_portreq = lacp_portreq;
2012 
2013 	error = lacp_attach(sc);
2014 	if (error)
2015 		return (error);
2016 
2017 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2018 		lacp_port_create(lp);
2019 
2020 	return (error);
2021 }
2022 
2023 static int
2024 lagg_lacp_detach(struct lagg_softc *sc)
2025 {
2026 	struct lagg_port *lp;
2027 	int error;
2028 
2029 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2030 		lacp_port_destroy(lp);
2031 
2032 	/* unlocking is safe here */
2033 	LAGG_WUNLOCK(sc);
2034 	error = lacp_detach(sc);
2035 	LAGG_WLOCK(sc);
2036 
2037 	return (error);
2038 }
2039 
2040 static void
2041 lagg_lacp_lladdr(struct lagg_softc *sc)
2042 {
2043 	struct lagg_port *lp;
2044 
2045 	/* purge all the lacp ports */
2046 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2047 		lacp_port_destroy(lp);
2048 
2049 	/* add them back in */
2050 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2051 		lacp_port_create(lp);
2052 }
2053 
2054 static int
2055 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2056 {
2057 	struct lagg_port *lp;
2058 
2059 	lp = lacp_select_tx_port(sc, m);
2060 	if (lp == NULL) {
2061 		m_freem(m);
2062 		return (ENETDOWN);
2063 	}
2064 
2065 	/* Send mbuf */
2066 	return (lagg_enqueue(lp->lp_ifp, m));
2067 }
2068 
2069 static struct mbuf *
2070 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2071 {
2072 	struct ifnet *ifp = sc->sc_ifp;
2073 	struct ether_header *eh;
2074 	u_short etype;
2075 
2076 	eh = mtod(m, struct ether_header *);
2077 	etype = ntohs(eh->ether_type);
2078 
2079 	/* Tap off LACP control messages */
2080 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2081 		m = lacp_input(lp, m);
2082 		if (m == NULL)
2083 			return (NULL);
2084 	}
2085 
2086 	/*
2087 	 * If the port is not collecting or not in the active aggregator then
2088 	 * free and return.
2089 	 */
2090 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2091 		m_freem(m);
2092 		return (NULL);
2093 	}
2094 
2095 	m->m_pkthdr.rcvif = ifp;
2096 	return (m);
2097 }
2098 
2099 static void
2100 lagg_callout(void *arg)
2101 {
2102 	struct lagg_softc *sc = (struct lagg_softc *)arg;
2103 	struct ifnet *ifp = sc->sc_ifp;
2104 
2105 	ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets);
2106 	ifp->if_opackets = counter_u64_fetch(sc->sc_opackets);
2107 	ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes);
2108 	ifp->if_obytes = counter_u64_fetch(sc->sc_obytes);
2109 
2110 	callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
2111 }
2112