xref: /freebsd/sys/net/if_lagg.c (revision 6d732c66bca5da4d261577aad2c8ea84519b0bea)
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  *
7  * Permission to use, copy, modify, and distribute this software for any
8  * purpose with or without fee is hereby granted, provided that the above
9  * copyright notice and this permission notice appear in all copies.
10  *
11  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18  */
19 
20 #include <sys/cdefs.h>
21 __FBSDID("$FreeBSD$");
22 
23 #include "opt_inet.h"
24 #include "opt_inet6.h"
25 
26 #include <sys/param.h>
27 #include <sys/kernel.h>
28 #include <sys/malloc.h>
29 #include <sys/mbuf.h>
30 #include <sys/queue.h>
31 #include <sys/socket.h>
32 #include <sys/sockio.h>
33 #include <sys/sysctl.h>
34 #include <sys/module.h>
35 #include <sys/priv.h>
36 #include <sys/systm.h>
37 #include <sys/proc.h>
38 #include <sys/hash.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_llc.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 #endif
58 #ifdef INET
59 #include <netinet/in_systm.h>
60 #include <netinet/if_ether.h>
61 #include <netinet/ip.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 /* 802.3ad LACP */
152 static int	lagg_lacp_attach(struct lagg_softc *);
153 static int	lagg_lacp_detach(struct lagg_softc *);
154 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
155 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
156 		    struct mbuf *);
157 static void	lagg_lacp_lladdr(struct lagg_softc *);
158 
159 static void	lagg_callout(void *);
160 
161 /* lagg protocol table */
162 static const struct {
163 	int			ti_proto;
164 	int			(*ti_attach)(struct lagg_softc *);
165 } lagg_protos[] = {
166 	{ LAGG_PROTO_ROUNDROBIN,	lagg_rr_attach },
167 	{ LAGG_PROTO_FAILOVER,		lagg_fail_attach },
168 	{ LAGG_PROTO_LOADBALANCE,	lagg_lb_attach },
169 	{ LAGG_PROTO_ETHERCHANNEL,	lagg_lb_attach },
170 	{ LAGG_PROTO_LACP,		lagg_lacp_attach },
171 	{ LAGG_PROTO_NONE,		NULL }
172 };
173 
174 SYSCTL_DECL(_net_link);
175 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
176     "Link Aggregation");
177 
178 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */
179 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW,
180     &lagg_failover_rx_all, 0,
181     "Accept input from any interface in a failover lagg");
182 static int def_use_flowid = 1; /* Default value for using M_FLOWID */
183 TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid);
184 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RW,
185     &def_use_flowid, 0,
186     "Default setting for using flow id for load sharing");
187 static int def_flowid_shift = 16; /* Default value for using M_FLOWID */
188 TUNABLE_INT("net.link.lagg.default_flowid_shift", &def_flowid_shift);
189 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RW,
190     &def_flowid_shift, 0,
191     "Default setting for flowid shift for load sharing");
192 
193 static int
194 lagg_modevent(module_t mod, int type, void *data)
195 {
196 
197 	switch (type) {
198 	case MOD_LOAD:
199 		mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
200 		SLIST_INIT(&lagg_list);
201 		lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
202 		    lagg_clone_destroy, 0);
203 		lagg_input_p = lagg_input;
204 		lagg_linkstate_p = lagg_port_state;
205 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
206 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
207 		    EVENTHANDLER_PRI_ANY);
208 		break;
209 	case MOD_UNLOAD:
210 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
211 		    lagg_detach_cookie);
212 		if_clone_detach(lagg_cloner);
213 		lagg_input_p = NULL;
214 		lagg_linkstate_p = NULL;
215 		mtx_destroy(&lagg_list_mtx);
216 		break;
217 	default:
218 		return (EOPNOTSUPP);
219 	}
220 	return (0);
221 }
222 
223 static moduledata_t lagg_mod = {
224 	"if_lagg",
225 	lagg_modevent,
226 	0
227 };
228 
229 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
230 MODULE_VERSION(if_lagg, 1);
231 
232 /*
233  * This routine is run via an vlan
234  * config EVENT
235  */
236 static void
237 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
238 {
239         struct lagg_softc       *sc = ifp->if_softc;
240         struct lagg_port        *lp;
241         struct rm_priotracker   tracker;
242 
243         if (ifp->if_softc !=  arg)   /* Not our event */
244                 return;
245 
246         LAGG_RLOCK(sc, &tracker);
247         if (!SLIST_EMPTY(&sc->sc_ports)) {
248                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
249                         EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
250         }
251         LAGG_RUNLOCK(sc, &tracker);
252 }
253 
254 /*
255  * This routine is run via an vlan
256  * unconfig EVENT
257  */
258 static void
259 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
260 {
261         struct lagg_softc       *sc = ifp->if_softc;
262         struct lagg_port        *lp;
263         struct rm_priotracker   tracker;
264 
265         if (ifp->if_softc !=  arg)   /* Not our event */
266                 return;
267 
268         LAGG_RLOCK(sc, &tracker);
269         if (!SLIST_EMPTY(&sc->sc_ports)) {
270                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
271                         EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
272         }
273         LAGG_RUNLOCK(sc, &tracker);
274 }
275 
276 static int
277 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
278 {
279 	struct lagg_softc *sc;
280 	struct ifnet *ifp;
281 	int i, error = 0;
282 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
283 	struct sysctl_oid *oid;
284 	char num[14];			/* sufficient for 32 bits */
285 
286 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
287 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
288 	if (ifp == NULL) {
289 		free(sc, M_DEVBUF);
290 		return (ENOSPC);
291 	}
292 
293 	sc->sc_ipackets = counter_u64_alloc(M_WAITOK);
294 	sc->sc_opackets = counter_u64_alloc(M_WAITOK);
295 	sc->sc_ibytes = counter_u64_alloc(M_WAITOK);
296 	sc->sc_obytes = counter_u64_alloc(M_WAITOK);
297 
298 	sysctl_ctx_init(&sc->ctx);
299 	snprintf(num, sizeof(num), "%u", unit);
300 	sc->use_flowid = def_use_flowid;
301 	sc->flowid_shift = def_flowid_shift;
302 	sc->sc_oid = oid = SYSCTL_ADD_NODE(&sc->ctx,
303 		&SYSCTL_NODE_CHILDREN(_net_link, lagg),
304 		OID_AUTO, num, CTLFLAG_RD, NULL, "");
305 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
306 		"use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid,
307 		sc->use_flowid, "Use flow id for load sharing");
308 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
309 		"flowid_shift", CTLTYPE_INT|CTLFLAG_RW, &sc->flowid_shift,
310 		sc->flowid_shift,
311 		"Shift flowid bits to prevent multiqueue collisions");
312 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
313 		"count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count,
314 		"Total number of ports");
315 	SYSCTL_ADD_PROC(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
316 		"active", CTLTYPE_INT|CTLFLAG_RD, sc, 0, lagg_sysctl_active,
317 		"I", "Total number of active ports");
318 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
319 		"flapping", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_flapping,
320 		sc->sc_flapping, "Total number of port change events");
321 	/* Hash all layers by default */
322 	sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4;
323 
324 	sc->sc_proto = LAGG_PROTO_NONE;
325 	for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
326 		if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
327 			sc->sc_proto = lagg_protos[i].ti_proto;
328 			if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
329 				if_free(ifp);
330 				free(sc, M_DEVBUF);
331 				return (error);
332 			}
333 			break;
334 		}
335 	}
336 	LAGG_LOCK_INIT(sc);
337 	LAGG_CALLOUT_LOCK_INIT(sc);
338 	SLIST_INIT(&sc->sc_ports);
339 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
340 
341 	/*
342 	 * This uses the callout lock rather than the rmlock; one can't
343 	 * hold said rmlock during SWI.
344 	 */
345 	callout_init_mtx(&sc->sc_callout, &sc->sc_call_mtx, 0);
346 
347 	/* Initialise pseudo media types */
348 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
349 	    lagg_media_status);
350 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
351 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
352 
353 	if_initname(ifp, laggname, unit);
354 	ifp->if_softc = sc;
355 	ifp->if_transmit = lagg_transmit;
356 	ifp->if_qflush = lagg_qflush;
357 	ifp->if_init = lagg_init;
358 	ifp->if_ioctl = lagg_ioctl;
359 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
360 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
361 
362 	/*
363 	 * Attach as an ordinary ethernet device, children will be attached
364 	 * as special device IFT_IEEE8023ADLAG.
365 	 */
366 	ether_ifattach(ifp, eaddr);
367 
368 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
369 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
370 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
371 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
372 
373 	/* Insert into the global list of laggs */
374 	mtx_lock(&lagg_list_mtx);
375 	SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
376 	mtx_unlock(&lagg_list_mtx);
377 
378 	callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
379 
380 	return (0);
381 }
382 
383 static void
384 lagg_clone_destroy(struct ifnet *ifp)
385 {
386 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
387 	struct lagg_port *lp;
388 
389 	LAGG_WLOCK(sc);
390 
391 	lagg_stop(sc);
392 	ifp->if_flags &= ~IFF_UP;
393 
394 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
395 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
396 
397 	/* Shutdown and remove lagg ports */
398 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
399 		lagg_port_destroy(lp, 1);
400 	/* Unhook the aggregation protocol */
401 	if (sc->sc_detach != NULL)
402 		(*sc->sc_detach)(sc);
403 
404 	LAGG_WUNLOCK(sc);
405 
406 	sysctl_ctx_free(&sc->ctx);
407 	ifmedia_removeall(&sc->sc_media);
408 	ether_ifdetach(ifp);
409 	if_free(ifp);
410 
411 	/* This grabs sc_callout_mtx, serialising it correctly */
412 	callout_drain(&sc->sc_callout);
413 
414 	/* At this point it's drained; we can free this */
415 	counter_u64_free(sc->sc_ipackets);
416 	counter_u64_free(sc->sc_opackets);
417 	counter_u64_free(sc->sc_ibytes);
418 	counter_u64_free(sc->sc_obytes);
419 
420 	mtx_lock(&lagg_list_mtx);
421 	SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
422 	mtx_unlock(&lagg_list_mtx);
423 
424 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
425 	LAGG_LOCK_DESTROY(sc);
426 	LAGG_CALLOUT_LOCK_DESTROY(sc);
427 	free(sc, M_DEVBUF);
428 }
429 
430 static void
431 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
432 {
433 	struct ifnet *ifp = sc->sc_ifp;
434 
435 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
436 		return;
437 
438 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
439 	/* Let the protocol know the MAC has changed */
440 	if (sc->sc_lladdr != NULL)
441 		(*sc->sc_lladdr)(sc);
442 	EVENTHANDLER_INVOKE(iflladdr_event, ifp);
443 }
444 
445 static void
446 lagg_capabilities(struct lagg_softc *sc)
447 {
448 	struct lagg_port *lp;
449 	int cap = ~0, ena = ~0;
450 	u_long hwa = ~0UL;
451 
452 	LAGG_WLOCK_ASSERT(sc);
453 
454 	/* Get capabilities from the lagg ports */
455 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
456 		cap &= lp->lp_ifp->if_capabilities;
457 		ena &= lp->lp_ifp->if_capenable;
458 		hwa &= lp->lp_ifp->if_hwassist;
459 	}
460 	cap = (cap == ~0 ? 0 : cap);
461 	ena = (ena == ~0 ? 0 : ena);
462 	hwa = (hwa == ~0 ? 0 : hwa);
463 
464 	if (sc->sc_ifp->if_capabilities != cap ||
465 	    sc->sc_ifp->if_capenable != ena ||
466 	    sc->sc_ifp->if_hwassist != hwa) {
467 		sc->sc_ifp->if_capabilities = cap;
468 		sc->sc_ifp->if_capenable = ena;
469 		sc->sc_ifp->if_hwassist = hwa;
470 		getmicrotime(&sc->sc_ifp->if_lastchange);
471 
472 		if (sc->sc_ifflags & IFF_DEBUG)
473 			if_printf(sc->sc_ifp,
474 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
475 	}
476 }
477 
478 static void
479 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
480 {
481 	struct lagg_softc *sc = lp->lp_softc;
482 	struct ifnet *ifp = lp->lp_ifp;
483 	struct lagg_llq *llq;
484 	int pending = 0;
485 
486 	LAGG_WLOCK_ASSERT(sc);
487 
488 	if (lp->lp_detaching ||
489 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
490 		return;
491 
492 	/* Check to make sure its not already queued to be changed */
493 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
494 		if (llq->llq_ifp == ifp) {
495 			pending = 1;
496 			break;
497 		}
498 	}
499 
500 	if (!pending) {
501 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
502 		if (llq == NULL)	/* XXX what to do */
503 			return;
504 	}
505 
506 	/* Update the lladdr even if pending, it may have changed */
507 	llq->llq_ifp = ifp;
508 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
509 
510 	if (!pending)
511 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
512 
513 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
514 }
515 
516 /*
517  * Set the interface MAC address from a taskqueue to avoid a LOR.
518  */
519 static void
520 lagg_port_setlladdr(void *arg, int pending)
521 {
522 	struct lagg_softc *sc = (struct lagg_softc *)arg;
523 	struct lagg_llq *llq, *head;
524 	struct ifnet *ifp;
525 	int error;
526 
527 	/* Grab a local reference of the queue and remove it from the softc */
528 	LAGG_WLOCK(sc);
529 	head = SLIST_FIRST(&sc->sc_llq_head);
530 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
531 	LAGG_WUNLOCK(sc);
532 
533 	/*
534 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
535 	 * unlocked as we have the only reference to it.
536 	 */
537 	for (llq = head; llq != NULL; llq = head) {
538 		ifp = llq->llq_ifp;
539 
540 		/* Set the link layer address */
541 		CURVNET_SET(ifp->if_vnet);
542 		error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
543 		CURVNET_RESTORE();
544 		if (error)
545 			printf("%s: setlladdr failed on %s\n", __func__,
546 			    ifp->if_xname);
547 
548 		head = SLIST_NEXT(llq, llq_entries);
549 		free(llq, M_DEVBUF);
550 	}
551 }
552 
553 static int
554 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
555 {
556 	struct lagg_softc *sc_ptr;
557 	struct lagg_port *lp;
558 	int error = 0;
559 
560 	LAGG_WLOCK_ASSERT(sc);
561 
562 	/* Limit the maximal number of lagg ports */
563 	if (sc->sc_count >= LAGG_MAX_PORTS)
564 		return (ENOSPC);
565 
566 	/* Check if port has already been associated to a lagg */
567 	if (ifp->if_lagg != NULL) {
568 		/* Port is already in the current lagg? */
569 		lp = (struct lagg_port *)ifp->if_lagg;
570 		if (lp->lp_softc == sc)
571 			return (EEXIST);
572 		return (EBUSY);
573 	}
574 
575 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
576 	if (ifp->if_type != IFT_ETHER)
577 		return (EPROTONOSUPPORT);
578 
579 #ifdef INET6
580 	/*
581 	 * The member interface should not have inet6 address because
582 	 * two interfaces with a valid link-local scope zone must not be
583 	 * merged in any form.  This restriction is needed to
584 	 * prevent violation of link-local scope zone.  Attempts to
585 	 * add a member interface which has inet6 addresses triggers
586 	 * removal of all inet6 addresses on the member interface.
587 	 */
588 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
589 		if (in6ifa_llaonifp(lp->lp_ifp)) {
590 			in6_ifdetach(lp->lp_ifp);
591 			if_printf(sc->sc_ifp,
592 			    "IPv6 addresses on %s have been removed "
593 			    "before adding it as a member to prevent "
594 			    "IPv6 address scope violation.\n",
595 			    lp->lp_ifp->if_xname);
596 		}
597 	}
598 	if (in6ifa_llaonifp(ifp)) {
599 		in6_ifdetach(ifp);
600 		if_printf(sc->sc_ifp,
601 		    "IPv6 addresses on %s have been removed "
602 		    "before adding it as a member to prevent "
603 		    "IPv6 address scope violation.\n",
604 		    ifp->if_xname);
605 	}
606 #endif
607 	/* Allow the first Ethernet member to define the MTU */
608 	if (SLIST_EMPTY(&sc->sc_ports))
609 		sc->sc_ifp->if_mtu = ifp->if_mtu;
610 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
611 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
612 		    ifp->if_xname);
613 		return (EINVAL);
614 	}
615 
616 	if ((lp = malloc(sizeof(struct lagg_port),
617 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
618 		return (ENOMEM);
619 
620 	/* Check if port is a stacked lagg */
621 	mtx_lock(&lagg_list_mtx);
622 	SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
623 		if (ifp == sc_ptr->sc_ifp) {
624 			mtx_unlock(&lagg_list_mtx);
625 			free(lp, M_DEVBUF);
626 			return (EINVAL);
627 			/* XXX disable stacking for the moment, its untested */
628 #ifdef LAGG_PORT_STACKING
629 			lp->lp_flags |= LAGG_PORT_STACK;
630 			if (lagg_port_checkstacking(sc_ptr) >=
631 			    LAGG_MAX_STACKING) {
632 				mtx_unlock(&lagg_list_mtx);
633 				free(lp, M_DEVBUF);
634 				return (E2BIG);
635 			}
636 #endif
637 		}
638 	}
639 	mtx_unlock(&lagg_list_mtx);
640 
641 	/* Change the interface type */
642 	lp->lp_iftype = ifp->if_type;
643 	ifp->if_type = IFT_IEEE8023ADLAG;
644 	ifp->if_lagg = lp;
645 	lp->lp_ioctl = ifp->if_ioctl;
646 	ifp->if_ioctl = lagg_port_ioctl;
647 	lp->lp_output = ifp->if_output;
648 	ifp->if_output = lagg_port_output;
649 
650 	lp->lp_ifp = ifp;
651 	lp->lp_softc = sc;
652 
653 	/* Save port link layer address */
654 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
655 
656 	if (SLIST_EMPTY(&sc->sc_ports)) {
657 		sc->sc_primary = lp;
658 		lagg_lladdr(sc, IF_LLADDR(ifp));
659 	} else {
660 		/* Update link layer address for this port */
661 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
662 	}
663 
664 	/* Insert into the list of ports */
665 	SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
666 	sc->sc_count++;
667 
668 	/* Update lagg capabilities */
669 	lagg_capabilities(sc);
670 	lagg_linkstate(sc);
671 
672 	/* Add multicast addresses and interface flags to this port */
673 	lagg_ether_cmdmulti(lp, 1);
674 	lagg_setflags(lp, 1);
675 
676 	if (sc->sc_port_create != NULL)
677 		error = (*sc->sc_port_create)(lp);
678 	if (error) {
679 		/* remove the port again, without calling sc_port_destroy */
680 		lagg_port_destroy(lp, 0);
681 		return (error);
682 	}
683 
684 	return (error);
685 }
686 
687 #ifdef LAGG_PORT_STACKING
688 static int
689 lagg_port_checkstacking(struct lagg_softc *sc)
690 {
691 	struct lagg_softc *sc_ptr;
692 	struct lagg_port *lp;
693 	int m = 0;
694 
695 	LAGG_WLOCK_ASSERT(sc);
696 
697 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
698 		if (lp->lp_flags & LAGG_PORT_STACK) {
699 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
700 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
701 		}
702 	}
703 
704 	return (m + 1);
705 }
706 #endif
707 
708 static int
709 lagg_port_destroy(struct lagg_port *lp, int runpd)
710 {
711 	struct lagg_softc *sc = lp->lp_softc;
712 	struct lagg_port *lp_ptr;
713 	struct lagg_llq *llq;
714 	struct ifnet *ifp = lp->lp_ifp;
715 
716 	LAGG_WLOCK_ASSERT(sc);
717 
718 	if (runpd && sc->sc_port_destroy != NULL)
719 		(*sc->sc_port_destroy)(lp);
720 
721 	/*
722 	 * Remove multicast addresses and interface flags from this port and
723 	 * reset the MAC address, skip if the interface is being detached.
724 	 */
725 	if (!lp->lp_detaching) {
726 		lagg_ether_cmdmulti(lp, 0);
727 		lagg_setflags(lp, 0);
728 		lagg_port_lladdr(lp, lp->lp_lladdr);
729 	}
730 
731 	/* Restore interface */
732 	ifp->if_type = lp->lp_iftype;
733 	ifp->if_ioctl = lp->lp_ioctl;
734 	ifp->if_output = lp->lp_output;
735 	ifp->if_lagg = NULL;
736 
737 	/* Finally, remove the port from the lagg */
738 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
739 	sc->sc_count--;
740 
741 	/* Update the primary interface */
742 	if (lp == sc->sc_primary) {
743 		uint8_t lladdr[ETHER_ADDR_LEN];
744 
745 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
746 			bzero(&lladdr, ETHER_ADDR_LEN);
747 		} else {
748 			bcopy(lp_ptr->lp_lladdr,
749 			    lladdr, ETHER_ADDR_LEN);
750 		}
751 		lagg_lladdr(sc, lladdr);
752 		sc->sc_primary = lp_ptr;
753 
754 		/* Update link layer address for each port */
755 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
756 			lagg_port_lladdr(lp_ptr, lladdr);
757 	}
758 
759 	/* Remove any pending lladdr changes from the queue */
760 	if (lp->lp_detaching) {
761 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
762 			if (llq->llq_ifp == ifp) {
763 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
764 				    llq_entries);
765 				free(llq, M_DEVBUF);
766 				break;	/* Only appears once */
767 			}
768 		}
769 	}
770 
771 	if (lp->lp_ifflags)
772 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
773 
774 	free(lp, M_DEVBUF);
775 
776 	/* Update lagg capabilities */
777 	lagg_capabilities(sc);
778 	lagg_linkstate(sc);
779 
780 	return (0);
781 }
782 
783 static int
784 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
785 {
786 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
787 	struct lagg_softc *sc;
788 	struct lagg_port *lp = NULL;
789 	int error = 0;
790 	struct rm_priotracker tracker;
791 
792 	/* Should be checked by the caller */
793 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
794 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
795 		goto fallback;
796 
797 	switch (cmd) {
798 	case SIOCGLAGGPORT:
799 		if (rp->rp_portname[0] == '\0' ||
800 		    ifunit(rp->rp_portname) != ifp) {
801 			error = EINVAL;
802 			break;
803 		}
804 
805 		LAGG_RLOCK(sc, &tracker);
806 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
807 			error = ENOENT;
808 			LAGG_RUNLOCK(sc, &tracker);
809 			break;
810 		}
811 
812 		lagg_port2req(lp, rp);
813 		LAGG_RUNLOCK(sc, &tracker);
814 		break;
815 
816 	case SIOCSIFCAP:
817 		if (lp->lp_ioctl == NULL) {
818 			error = EINVAL;
819 			break;
820 		}
821 		error = (*lp->lp_ioctl)(ifp, cmd, data);
822 		if (error)
823 			break;
824 
825 		/* Update lagg interface capabilities */
826 		LAGG_WLOCK(sc);
827 		lagg_capabilities(sc);
828 		LAGG_WUNLOCK(sc);
829 		break;
830 
831 	case SIOCSIFMTU:
832 		/* Do not allow the MTU to be changed once joined */
833 		error = EINVAL;
834 		break;
835 
836 	default:
837 		goto fallback;
838 	}
839 
840 	return (error);
841 
842 fallback:
843 	if (lp->lp_ioctl != NULL)
844 		return ((*lp->lp_ioctl)(ifp, cmd, data));
845 
846 	return (EINVAL);
847 }
848 
849 /*
850  * For direct output to child ports.
851  */
852 static int
853 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
854 	const struct sockaddr *dst, struct route *ro)
855 {
856 	struct lagg_port *lp = ifp->if_lagg;
857 
858 	switch (dst->sa_family) {
859 		case pseudo_AF_HDRCMPLT:
860 		case AF_UNSPEC:
861 			return ((*lp->lp_output)(ifp, m, dst, ro));
862 	}
863 
864 	/* drop any other frames */
865 	m_freem(m);
866 	return (ENETDOWN);
867 }
868 
869 static void
870 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
871 {
872 	struct lagg_port *lp;
873 	struct lagg_softc *sc;
874 
875 	if ((lp = ifp->if_lagg) == NULL)
876 		return;
877 	/* If the ifnet is just being renamed, don't do anything. */
878 	if (ifp->if_flags & IFF_RENAMING)
879 		return;
880 
881 	sc = lp->lp_softc;
882 
883 	LAGG_WLOCK(sc);
884 	lp->lp_detaching = 1;
885 	lagg_port_destroy(lp, 1);
886 	LAGG_WUNLOCK(sc);
887 }
888 
889 static void
890 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
891 {
892 	struct lagg_softc *sc = lp->lp_softc;
893 
894 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
895 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
896 	rp->rp_prio = lp->lp_prio;
897 	rp->rp_flags = lp->lp_flags;
898 	if (sc->sc_portreq != NULL)
899 		(*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
900 
901 	/* Add protocol specific flags */
902 	switch (sc->sc_proto) {
903 		case LAGG_PROTO_FAILOVER:
904 			if (lp == sc->sc_primary)
905 				rp->rp_flags |= LAGG_PORT_MASTER;
906 			if (lp == lagg_link_active(sc, sc->sc_primary))
907 				rp->rp_flags |= LAGG_PORT_ACTIVE;
908 			break;
909 
910 		case LAGG_PROTO_ROUNDROBIN:
911 		case LAGG_PROTO_LOADBALANCE:
912 		case LAGG_PROTO_ETHERCHANNEL:
913 			if (LAGG_PORTACTIVE(lp))
914 				rp->rp_flags |= LAGG_PORT_ACTIVE;
915 			break;
916 
917 		case LAGG_PROTO_LACP:
918 			/* LACP has a different definition of active */
919 			if (lacp_isactive(lp))
920 				rp->rp_flags |= LAGG_PORT_ACTIVE;
921 			if (lacp_iscollecting(lp))
922 				rp->rp_flags |= LAGG_PORT_COLLECTING;
923 			if (lacp_isdistributing(lp))
924 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
925 			break;
926 	}
927 
928 }
929 
930 static void
931 lagg_init(void *xsc)
932 {
933 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
934 	struct lagg_port *lp;
935 	struct ifnet *ifp = sc->sc_ifp;
936 
937 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
938 		return;
939 
940 	LAGG_WLOCK(sc);
941 
942 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
943 	/* Update the port lladdrs */
944 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
945 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
946 
947 	if (sc->sc_init != NULL)
948 		(*sc->sc_init)(sc);
949 
950 	LAGG_WUNLOCK(sc);
951 }
952 
953 static void
954 lagg_stop(struct lagg_softc *sc)
955 {
956 	struct ifnet *ifp = sc->sc_ifp;
957 
958 	LAGG_WLOCK_ASSERT(sc);
959 
960 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
961 		return;
962 
963 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
964 
965 	if (sc->sc_stop != NULL)
966 		(*sc->sc_stop)(sc);
967 }
968 
969 static int
970 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
971 {
972 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
973 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
974 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
975 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
976 	struct ifreq *ifr = (struct ifreq *)data;
977 	struct lagg_port *lp;
978 	struct ifnet *tpif;
979 	struct thread *td = curthread;
980 	char *buf, *outbuf;
981 	int count, buflen, len, error = 0;
982 	struct rm_priotracker tracker;
983 
984 	bzero(&rpbuf, sizeof(rpbuf));
985 
986 	switch (cmd) {
987 	case SIOCGLAGG:
988 		LAGG_RLOCK(sc, &tracker);
989 		count = 0;
990 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
991 			count++;
992 		buflen = count * sizeof(struct lagg_reqport);
993 		LAGG_RUNLOCK(sc, &tracker);
994 
995 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
996 
997 		LAGG_RLOCK(sc, &tracker);
998 		ra->ra_proto = sc->sc_proto;
999 		if (sc->sc_req != NULL)
1000 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
1001 
1002 		count = 0;
1003 		buf = outbuf;
1004 		len = min(ra->ra_size, buflen);
1005 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1006 			if (len < sizeof(rpbuf))
1007 				break;
1008 
1009 			lagg_port2req(lp, &rpbuf);
1010 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1011 			count++;
1012 			buf += sizeof(rpbuf);
1013 			len -= sizeof(rpbuf);
1014 		}
1015 		LAGG_RUNLOCK(sc, &tracker);
1016 		ra->ra_ports = count;
1017 		ra->ra_size = count * sizeof(rpbuf);
1018 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1019 		free(outbuf, M_TEMP);
1020 		break;
1021 	case SIOCSLAGG:
1022 		error = priv_check(td, PRIV_NET_LAGG);
1023 		if (error)
1024 			break;
1025 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1026 			error = EPROTONOSUPPORT;
1027 			break;
1028 		}
1029 		LAGG_WLOCK(sc);
1030 		if (sc->sc_proto != LAGG_PROTO_NONE) {
1031 			/* Reset protocol first in case detach unlocks */
1032 			sc->sc_proto = LAGG_PROTO_NONE;
1033 			error = sc->sc_detach(sc);
1034 			sc->sc_detach = NULL;
1035 			sc->sc_start = NULL;
1036 			sc->sc_input = NULL;
1037 			sc->sc_port_create = NULL;
1038 			sc->sc_port_destroy = NULL;
1039 			sc->sc_linkstate = NULL;
1040 			sc->sc_init = NULL;
1041 			sc->sc_stop = NULL;
1042 			sc->sc_lladdr = NULL;
1043 			sc->sc_req = NULL;
1044 			sc->sc_portreq = NULL;
1045 		} else if (sc->sc_input != NULL) {
1046 			/* Still detaching */
1047 			error = EBUSY;
1048 		}
1049 		if (error != 0) {
1050 			LAGG_WUNLOCK(sc);
1051 			break;
1052 		}
1053 		for (int i = 0; i < (sizeof(lagg_protos) /
1054 		    sizeof(lagg_protos[0])); i++) {
1055 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
1056 				if (sc->sc_ifflags & IFF_DEBUG)
1057 					printf("%s: using proto %u\n",
1058 					    sc->sc_ifname,
1059 					    lagg_protos[i].ti_proto);
1060 				sc->sc_proto = lagg_protos[i].ti_proto;
1061 				if (sc->sc_proto != LAGG_PROTO_NONE)
1062 					error = lagg_protos[i].ti_attach(sc);
1063 				LAGG_WUNLOCK(sc);
1064 				return (error);
1065 			}
1066 		}
1067 		LAGG_WUNLOCK(sc);
1068 		error = EPROTONOSUPPORT;
1069 		break;
1070 	case SIOCGLAGGFLAGS:
1071 		rf->rf_flags = sc->sc_flags;
1072 		break;
1073 	case SIOCSLAGGHASH:
1074 		error = priv_check(td, PRIV_NET_LAGG);
1075 		if (error)
1076 			break;
1077 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1078 			error = EINVAL;
1079 			break;
1080 		}
1081 		LAGG_WLOCK(sc);
1082 		sc->sc_flags &= ~LAGG_F_HASHMASK;
1083 		sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK;
1084 		LAGG_WUNLOCK(sc);
1085 		break;
1086 	case SIOCGLAGGPORT:
1087 		if (rp->rp_portname[0] == '\0' ||
1088 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1089 			error = EINVAL;
1090 			break;
1091 		}
1092 
1093 		LAGG_RLOCK(sc, &tracker);
1094 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1095 		    lp->lp_softc != sc) {
1096 			error = ENOENT;
1097 			LAGG_RUNLOCK(sc, &tracker);
1098 			break;
1099 		}
1100 
1101 		lagg_port2req(lp, rp);
1102 		LAGG_RUNLOCK(sc, &tracker);
1103 		break;
1104 	case SIOCSLAGGPORT:
1105 		error = priv_check(td, PRIV_NET_LAGG);
1106 		if (error)
1107 			break;
1108 		if (rp->rp_portname[0] == '\0' ||
1109 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1110 			error = EINVAL;
1111 			break;
1112 		}
1113 		LAGG_WLOCK(sc);
1114 		error = lagg_port_create(sc, tpif);
1115 		LAGG_WUNLOCK(sc);
1116 		break;
1117 	case SIOCSLAGGDELPORT:
1118 		error = priv_check(td, PRIV_NET_LAGG);
1119 		if (error)
1120 			break;
1121 		if (rp->rp_portname[0] == '\0' ||
1122 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1123 			error = EINVAL;
1124 			break;
1125 		}
1126 
1127 		LAGG_WLOCK(sc);
1128 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1129 		    lp->lp_softc != sc) {
1130 			error = ENOENT;
1131 			LAGG_WUNLOCK(sc);
1132 			break;
1133 		}
1134 
1135 		error = lagg_port_destroy(lp, 1);
1136 		LAGG_WUNLOCK(sc);
1137 		break;
1138 	case SIOCSIFFLAGS:
1139 		/* Set flags on ports too */
1140 		LAGG_WLOCK(sc);
1141 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1142 			lagg_setflags(lp, 1);
1143 		}
1144 		LAGG_WUNLOCK(sc);
1145 
1146 		if (!(ifp->if_flags & IFF_UP) &&
1147 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1148 			/*
1149 			 * If interface is marked down and it is running,
1150 			 * then stop and disable it.
1151 			 */
1152 			LAGG_WLOCK(sc);
1153 			lagg_stop(sc);
1154 			LAGG_WUNLOCK(sc);
1155 		} else if ((ifp->if_flags & IFF_UP) &&
1156 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1157 			/*
1158 			 * If interface is marked up and it is stopped, then
1159 			 * start it.
1160 			 */
1161 			(*ifp->if_init)(sc);
1162 		}
1163 		break;
1164 	case SIOCADDMULTI:
1165 	case SIOCDELMULTI:
1166 		LAGG_WLOCK(sc);
1167 		error = lagg_ether_setmulti(sc);
1168 		LAGG_WUNLOCK(sc);
1169 		break;
1170 	case SIOCSIFMEDIA:
1171 	case SIOCGIFMEDIA:
1172 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1173 		break;
1174 
1175 	case SIOCSIFCAP:
1176 	case SIOCSIFMTU:
1177 		/* Do not allow the MTU or caps to be directly changed */
1178 		error = EINVAL;
1179 		break;
1180 
1181 	default:
1182 		error = ether_ioctl(ifp, cmd, data);
1183 		break;
1184 	}
1185 	return (error);
1186 }
1187 
1188 static int
1189 lagg_ether_setmulti(struct lagg_softc *sc)
1190 {
1191 	struct lagg_port *lp;
1192 
1193 	LAGG_WLOCK_ASSERT(sc);
1194 
1195 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1196 		/* First, remove any existing filter entries. */
1197 		lagg_ether_cmdmulti(lp, 0);
1198 		/* copy all addresses from the lagg interface to the port */
1199 		lagg_ether_cmdmulti(lp, 1);
1200 	}
1201 	return (0);
1202 }
1203 
1204 static int
1205 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1206 {
1207 	struct lagg_softc *sc = lp->lp_softc;
1208 	struct ifnet *ifp = lp->lp_ifp;
1209 	struct ifnet *scifp = sc->sc_ifp;
1210 	struct lagg_mc *mc;
1211 	struct ifmultiaddr *ifma, *rifma = NULL;
1212 	struct sockaddr_dl sdl;
1213 	int error;
1214 
1215 	LAGG_WLOCK_ASSERT(sc);
1216 
1217 	bzero((char *)&sdl, sizeof(sdl));
1218 	sdl.sdl_len = sizeof(sdl);
1219 	sdl.sdl_family = AF_LINK;
1220 	sdl.sdl_type = IFT_ETHER;
1221 	sdl.sdl_alen = ETHER_ADDR_LEN;
1222 	sdl.sdl_index = ifp->if_index;
1223 
1224 	if (set) {
1225 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1226 			if (ifma->ifma_addr->sa_family != AF_LINK)
1227 				continue;
1228 			bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1229 			    LLADDR(&sdl), ETHER_ADDR_LEN);
1230 
1231 			error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1232 			if (error)
1233 				return (error);
1234 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1235 			if (mc == NULL)
1236 				return (ENOMEM);
1237 			mc->mc_ifma = rifma;
1238 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1239 		}
1240 	} else {
1241 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1242 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1243 			if_delmulti_ifma(mc->mc_ifma);
1244 			free(mc, M_DEVBUF);
1245 		}
1246 	}
1247 	return (0);
1248 }
1249 
1250 /* Handle a ref counted flag that should be set on the lagg port as well */
1251 static int
1252 lagg_setflag(struct lagg_port *lp, int flag, int status,
1253 	     int (*func)(struct ifnet *, int))
1254 {
1255 	struct lagg_softc *sc = lp->lp_softc;
1256 	struct ifnet *scifp = sc->sc_ifp;
1257 	struct ifnet *ifp = lp->lp_ifp;
1258 	int error;
1259 
1260 	LAGG_WLOCK_ASSERT(sc);
1261 
1262 	status = status ? (scifp->if_flags & flag) : 0;
1263 	/* Now "status" contains the flag value or 0 */
1264 
1265 	/*
1266 	 * See if recorded ports status is different from what
1267 	 * we want it to be.  If it is, flip it.  We record ports
1268 	 * status in lp_ifflags so that we won't clear ports flag
1269 	 * we haven't set.  In fact, we don't clear or set ports
1270 	 * flags directly, but get or release references to them.
1271 	 * That's why we can be sure that recorded flags still are
1272 	 * in accord with actual ports flags.
1273 	 */
1274 	if (status != (lp->lp_ifflags & flag)) {
1275 		error = (*func)(ifp, status);
1276 		if (error)
1277 			return (error);
1278 		lp->lp_ifflags &= ~flag;
1279 		lp->lp_ifflags |= status;
1280 	}
1281 	return (0);
1282 }
1283 
1284 /*
1285  * Handle IFF_* flags that require certain changes on the lagg port
1286  * if "status" is true, update ports flags respective to the lagg
1287  * if "status" is false, forcedly clear the flags set on port.
1288  */
1289 static int
1290 lagg_setflags(struct lagg_port *lp, int status)
1291 {
1292 	int error, i;
1293 
1294 	for (i = 0; lagg_pflags[i].flag; i++) {
1295 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1296 		    status, lagg_pflags[i].func);
1297 		if (error)
1298 			return (error);
1299 	}
1300 	return (0);
1301 }
1302 
1303 static int
1304 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1305 {
1306 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1307 	int error, len, mcast;
1308 	struct rm_priotracker tracker;
1309 
1310 	len = m->m_pkthdr.len;
1311 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1312 
1313 	LAGG_RLOCK(sc, &tracker);
1314 	/* We need a Tx algorithm and at least one port */
1315 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1316 		LAGG_RUNLOCK(sc, &tracker);
1317 		m_freem(m);
1318 		ifp->if_oerrors++;
1319 		return (ENXIO);
1320 	}
1321 
1322 	ETHER_BPF_MTAP(ifp, m);
1323 
1324 	error = (*sc->sc_start)(sc, m);
1325 	LAGG_RUNLOCK(sc, &tracker);
1326 
1327 	if (error == 0) {
1328 		counter_u64_add(sc->sc_opackets, 1);
1329 		counter_u64_add(sc->sc_obytes, len);
1330 		ifp->if_omcasts += mcast;
1331 	} else
1332 		ifp->if_oerrors++;
1333 
1334 	return (error);
1335 }
1336 
1337 /*
1338  * The ifp->if_qflush entry point for lagg(4) is no-op.
1339  */
1340 static void
1341 lagg_qflush(struct ifnet *ifp __unused)
1342 {
1343 }
1344 
1345 static struct mbuf *
1346 lagg_input(struct ifnet *ifp, struct mbuf *m)
1347 {
1348 	struct lagg_port *lp = ifp->if_lagg;
1349 	struct lagg_softc *sc = lp->lp_softc;
1350 	struct ifnet *scifp = sc->sc_ifp;
1351 	struct rm_priotracker tracker;
1352 
1353 	LAGG_RLOCK(sc, &tracker);
1354 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1355 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1356 	    sc->sc_proto == LAGG_PROTO_NONE) {
1357 		LAGG_RUNLOCK(sc, &tracker);
1358 		m_freem(m);
1359 		return (NULL);
1360 	}
1361 
1362 	ETHER_BPF_MTAP(scifp, m);
1363 
1364 	m = (*sc->sc_input)(sc, lp, m);
1365 
1366 	if (m != NULL) {
1367 		counter_u64_add(sc->sc_ipackets, 1);
1368 		counter_u64_add(sc->sc_ibytes, m->m_pkthdr.len);
1369 
1370 		if (scifp->if_flags & IFF_MONITOR) {
1371 			m_freem(m);
1372 			m = NULL;
1373 		}
1374 	}
1375 
1376 	LAGG_RUNLOCK(sc, &tracker);
1377 	return (m);
1378 }
1379 
1380 static int
1381 lagg_media_change(struct ifnet *ifp)
1382 {
1383 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1384 
1385 	if (sc->sc_ifflags & IFF_DEBUG)
1386 		printf("%s\n", __func__);
1387 
1388 	/* Ignore */
1389 	return (0);
1390 }
1391 
1392 static void
1393 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1394 {
1395 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1396 	struct lagg_port *lp;
1397 	struct rm_priotracker tracker;
1398 
1399 	imr->ifm_status = IFM_AVALID;
1400 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1401 
1402 	LAGG_RLOCK(sc, &tracker);
1403 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1404 		if (LAGG_PORTACTIVE(lp))
1405 			imr->ifm_status |= IFM_ACTIVE;
1406 	}
1407 	LAGG_RUNLOCK(sc, &tracker);
1408 }
1409 
1410 static void
1411 lagg_linkstate(struct lagg_softc *sc)
1412 {
1413 	struct lagg_port *lp;
1414 	int new_link = LINK_STATE_DOWN;
1415 	uint64_t speed;
1416 
1417 	/* Our link is considered up if at least one of our ports is active */
1418 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1419 		if (lp->lp_link_state == LINK_STATE_UP) {
1420 			new_link = LINK_STATE_UP;
1421 			break;
1422 		}
1423 	}
1424 	if_link_state_change(sc->sc_ifp, new_link);
1425 
1426 	/* Update if_baudrate to reflect the max possible speed */
1427 	switch (sc->sc_proto) {
1428 		case LAGG_PROTO_FAILOVER:
1429 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1430 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1431 			break;
1432 		case LAGG_PROTO_ROUNDROBIN:
1433 		case LAGG_PROTO_LOADBALANCE:
1434 		case LAGG_PROTO_ETHERCHANNEL:
1435 			speed = 0;
1436 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1437 				speed += lp->lp_ifp->if_baudrate;
1438 			sc->sc_ifp->if_baudrate = speed;
1439 			break;
1440 		case LAGG_PROTO_LACP:
1441 			/* LACP updates if_baudrate itself */
1442 			break;
1443 	}
1444 }
1445 
1446 static void
1447 lagg_port_state(struct ifnet *ifp, int state)
1448 {
1449 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1450 	struct lagg_softc *sc = NULL;
1451 
1452 	if (lp != NULL)
1453 		sc = lp->lp_softc;
1454 	if (sc == NULL)
1455 		return;
1456 
1457 	LAGG_WLOCK(sc);
1458 	lagg_linkstate(sc);
1459 	if (sc->sc_linkstate != NULL)
1460 		(*sc->sc_linkstate)(lp);
1461 	LAGG_WUNLOCK(sc);
1462 }
1463 
1464 struct lagg_port *
1465 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1466 {
1467 	struct lagg_port *lp_next, *rval = NULL;
1468 	// int new_link = LINK_STATE_DOWN;
1469 
1470 	LAGG_RLOCK_ASSERT(sc);
1471 	/*
1472 	 * Search a port which reports an active link state.
1473 	 */
1474 
1475 	if (lp == NULL)
1476 		goto search;
1477 	if (LAGG_PORTACTIVE(lp)) {
1478 		rval = lp;
1479 		goto found;
1480 	}
1481 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1482 	    LAGG_PORTACTIVE(lp_next)) {
1483 		rval = lp_next;
1484 		goto found;
1485 	}
1486 
1487 search:
1488 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1489 		if (LAGG_PORTACTIVE(lp_next)) {
1490 			rval = lp_next;
1491 			goto found;
1492 		}
1493 	}
1494 
1495 found:
1496 	if (rval != NULL) {
1497 		/*
1498 		 * The IEEE 802.1D standard assumes that a lagg with
1499 		 * multiple ports is always full duplex. This is valid
1500 		 * for load sharing laggs and if at least two links
1501 		 * are active. Unfortunately, checking the latter would
1502 		 * be too expensive at this point.
1503 		 XXX
1504 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1505 		    (sc->sc_count > 1))
1506 			new_link = LINK_STATE_FULL_DUPLEX;
1507 		else
1508 			new_link = rval->lp_link_state;
1509 		 */
1510 	}
1511 
1512 	return (rval);
1513 }
1514 
1515 static const void *
1516 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1517 {
1518 	if (m->m_pkthdr.len < (off + len)) {
1519 		return (NULL);
1520 	} else if (m->m_len < (off + len)) {
1521 		m_copydata(m, off, len, buf);
1522 		return (buf);
1523 	}
1524 	return (mtod(m, char *) + off);
1525 }
1526 
1527 static int
1528 lagg_sysctl_active(SYSCTL_HANDLER_ARGS)
1529 {
1530 	struct lagg_softc *sc = (struct lagg_softc *)arg1;
1531 	struct lagg_port *lp;
1532 	int error;
1533 
1534 	/* LACP tracks active links automatically, the others do not */
1535 	if (sc->sc_proto != LAGG_PROTO_LACP) {
1536 		sc->sc_active = 0;
1537 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1538 			sc->sc_active += LAGG_PORTACTIVE(lp);
1539 	}
1540 
1541 	error = sysctl_handle_int(oidp, &sc->sc_active, 0, req);
1542 	if ((error) || (req->newptr == NULL))
1543 		return (error);
1544 
1545 	return (0);
1546 }
1547 
1548 uint32_t
1549 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
1550 {
1551 	uint16_t etype;
1552 	uint32_t p = key;
1553 	int off;
1554 	struct ether_header *eh;
1555 	const struct ether_vlan_header *vlan;
1556 #ifdef INET
1557 	const struct ip *ip;
1558 	const uint32_t *ports;
1559 	int iphlen;
1560 #endif
1561 #ifdef INET6
1562 	const struct ip6_hdr *ip6;
1563 	uint32_t flow;
1564 #endif
1565 	union {
1566 #ifdef INET
1567 		struct ip ip;
1568 #endif
1569 #ifdef INET6
1570 		struct ip6_hdr ip6;
1571 #endif
1572 		struct ether_vlan_header vlan;
1573 		uint32_t port;
1574 	} buf;
1575 
1576 
1577 	off = sizeof(*eh);
1578 	if (m->m_len < off)
1579 		goto out;
1580 	eh = mtod(m, struct ether_header *);
1581 	etype = ntohs(eh->ether_type);
1582 	if (sc->sc_flags & LAGG_F_HASHL2) {
1583 		p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p);
1584 		p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1585 	}
1586 
1587 	/* Special handling for encapsulating VLAN frames */
1588 	if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) {
1589 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1590 		    sizeof(m->m_pkthdr.ether_vtag), p);
1591 	} else if (etype == ETHERTYPE_VLAN) {
1592 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &buf);
1593 		if (vlan == NULL)
1594 			goto out;
1595 
1596 		if (sc->sc_flags & LAGG_F_HASHL2)
1597 			p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1598 		etype = ntohs(vlan->evl_proto);
1599 		off += sizeof(*vlan) - sizeof(*eh);
1600 	}
1601 
1602 	switch (etype) {
1603 #ifdef INET
1604 	case ETHERTYPE_IP:
1605 		ip = lagg_gethdr(m, off, sizeof(*ip), &buf);
1606 		if (ip == NULL)
1607 			goto out;
1608 
1609 		if (sc->sc_flags & LAGG_F_HASHL3) {
1610 			p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1611 			p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1612 		}
1613 		if (!(sc->sc_flags & LAGG_F_HASHL4))
1614 			break;
1615 		switch (ip->ip_p) {
1616 			case IPPROTO_TCP:
1617 			case IPPROTO_UDP:
1618 			case IPPROTO_SCTP:
1619 				iphlen = ip->ip_hl << 2;
1620 				if (iphlen < sizeof(*ip))
1621 					break;
1622 				off += iphlen;
1623 				ports = lagg_gethdr(m, off, sizeof(*ports), &buf);
1624 				if (ports == NULL)
1625 					break;
1626 				p = hash32_buf(ports, sizeof(*ports), p);
1627 				break;
1628 		}
1629 		break;
1630 #endif
1631 #ifdef INET6
1632 	case ETHERTYPE_IPV6:
1633 		if (!(sc->sc_flags & LAGG_F_HASHL3))
1634 			break;
1635 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf);
1636 		if (ip6 == NULL)
1637 			goto out;
1638 
1639 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1640 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1641 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1642 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1643 		break;
1644 #endif
1645 	}
1646 out:
1647 	return (p);
1648 }
1649 
1650 int
1651 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1652 {
1653 
1654 	return (ifp->if_transmit)(ifp, m);
1655 }
1656 
1657 /*
1658  * Simple round robin aggregation
1659  */
1660 
1661 static int
1662 lagg_rr_attach(struct lagg_softc *sc)
1663 {
1664 	sc->sc_detach = lagg_rr_detach;
1665 	sc->sc_start = lagg_rr_start;
1666 	sc->sc_input = lagg_rr_input;
1667 	sc->sc_port_create = NULL;
1668 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1669 	sc->sc_seq = 0;
1670 
1671 	return (0);
1672 }
1673 
1674 static int
1675 lagg_rr_detach(struct lagg_softc *sc)
1676 {
1677 	return (0);
1678 }
1679 
1680 static int
1681 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1682 {
1683 	struct lagg_port *lp;
1684 	uint32_t p;
1685 
1686 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1687 	p %= sc->sc_count;
1688 	lp = SLIST_FIRST(&sc->sc_ports);
1689 	while (p--)
1690 		lp = SLIST_NEXT(lp, lp_entries);
1691 
1692 	/*
1693 	 * Check the port's link state. This will return the next active
1694 	 * port if the link is down or the port is NULL.
1695 	 */
1696 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1697 		m_freem(m);
1698 		return (ENETDOWN);
1699 	}
1700 
1701 	/* Send mbuf */
1702 	return (lagg_enqueue(lp->lp_ifp, m));
1703 }
1704 
1705 static struct mbuf *
1706 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1707 {
1708 	struct ifnet *ifp = sc->sc_ifp;
1709 
1710 	/* Just pass in the packet to our lagg device */
1711 	m->m_pkthdr.rcvif = ifp;
1712 
1713 	return (m);
1714 }
1715 
1716 /*
1717  * Active failover
1718  */
1719 
1720 static int
1721 lagg_fail_attach(struct lagg_softc *sc)
1722 {
1723 	sc->sc_detach = lagg_fail_detach;
1724 	sc->sc_start = lagg_fail_start;
1725 	sc->sc_input = lagg_fail_input;
1726 	sc->sc_port_create = NULL;
1727 	sc->sc_port_destroy = NULL;
1728 
1729 	return (0);
1730 }
1731 
1732 static int
1733 lagg_fail_detach(struct lagg_softc *sc)
1734 {
1735 	return (0);
1736 }
1737 
1738 static int
1739 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1740 {
1741 	struct lagg_port *lp;
1742 
1743 	/* Use the master port if active or the next available port */
1744 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1745 		m_freem(m);
1746 		return (ENETDOWN);
1747 	}
1748 
1749 	/* Send mbuf */
1750 	return (lagg_enqueue(lp->lp_ifp, m));
1751 }
1752 
1753 static struct mbuf *
1754 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1755 {
1756 	struct ifnet *ifp = sc->sc_ifp;
1757 	struct lagg_port *tmp_tp;
1758 
1759 	if (lp == sc->sc_primary || lagg_failover_rx_all) {
1760 		m->m_pkthdr.rcvif = ifp;
1761 		return (m);
1762 	}
1763 
1764 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1765 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1766 		/*
1767 		 * If tmp_tp is null, we've recieved a packet when all
1768 		 * our links are down. Weird, but process it anyways.
1769 		 */
1770 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1771 			m->m_pkthdr.rcvif = ifp;
1772 			return (m);
1773 		}
1774 	}
1775 
1776 	m_freem(m);
1777 	return (NULL);
1778 }
1779 
1780 /*
1781  * Loadbalancing
1782  */
1783 
1784 static int
1785 lagg_lb_attach(struct lagg_softc *sc)
1786 {
1787 	struct lagg_port *lp;
1788 	struct lagg_lb *lb;
1789 
1790 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1791 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1792 		return (ENOMEM);
1793 
1794 	sc->sc_detach = lagg_lb_detach;
1795 	sc->sc_start = lagg_lb_start;
1796 	sc->sc_input = lagg_lb_input;
1797 	sc->sc_port_create = lagg_lb_port_create;
1798 	sc->sc_port_destroy = lagg_lb_port_destroy;
1799 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1800 
1801 	lb->lb_key = arc4random();
1802 	sc->sc_psc = (caddr_t)lb;
1803 
1804 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1805 		lagg_lb_port_create(lp);
1806 
1807 	return (0);
1808 }
1809 
1810 static int
1811 lagg_lb_detach(struct lagg_softc *sc)
1812 {
1813 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1814 	if (lb != NULL)
1815 		free(lb, M_DEVBUF);
1816 	return (0);
1817 }
1818 
1819 static int
1820 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1821 {
1822 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1823 	struct lagg_port *lp_next;
1824 	int i = 0;
1825 
1826 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1827 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1828 		if (lp_next == lp)
1829 			continue;
1830 		if (i >= LAGG_MAX_PORTS)
1831 			return (EINVAL);
1832 		if (sc->sc_ifflags & IFF_DEBUG)
1833 			printf("%s: port %s at index %d\n",
1834 			    sc->sc_ifname, lp_next->lp_ifname, i);
1835 		lb->lb_ports[i++] = lp_next;
1836 	}
1837 
1838 	return (0);
1839 }
1840 
1841 static int
1842 lagg_lb_port_create(struct lagg_port *lp)
1843 {
1844 	struct lagg_softc *sc = lp->lp_softc;
1845 	return (lagg_lb_porttable(sc, NULL));
1846 }
1847 
1848 static void
1849 lagg_lb_port_destroy(struct lagg_port *lp)
1850 {
1851 	struct lagg_softc *sc = lp->lp_softc;
1852 	lagg_lb_porttable(sc, lp);
1853 }
1854 
1855 static int
1856 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1857 {
1858 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1859 	struct lagg_port *lp = NULL;
1860 	uint32_t p = 0;
1861 
1862 	if (sc->use_flowid && (m->m_flags & M_FLOWID))
1863 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
1864 	else
1865 		p = lagg_hashmbuf(sc, m, lb->lb_key);
1866 	p %= sc->sc_count;
1867 	lp = lb->lb_ports[p];
1868 
1869 	/*
1870 	 * Check the port's link state. This will return the next active
1871 	 * port if the link is down or the port is NULL.
1872 	 */
1873 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1874 		m_freem(m);
1875 		return (ENETDOWN);
1876 	}
1877 
1878 	/* Send mbuf */
1879 	return (lagg_enqueue(lp->lp_ifp, m));
1880 }
1881 
1882 static struct mbuf *
1883 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1884 {
1885 	struct ifnet *ifp = sc->sc_ifp;
1886 
1887 	/* Just pass in the packet to our lagg device */
1888 	m->m_pkthdr.rcvif = ifp;
1889 
1890 	return (m);
1891 }
1892 
1893 /*
1894  * 802.3ad LACP
1895  */
1896 
1897 static int
1898 lagg_lacp_attach(struct lagg_softc *sc)
1899 {
1900 	struct lagg_port *lp;
1901 	int error;
1902 
1903 	sc->sc_detach = lagg_lacp_detach;
1904 	sc->sc_port_create = lacp_port_create;
1905 	sc->sc_port_destroy = lacp_port_destroy;
1906 	sc->sc_linkstate = lacp_linkstate;
1907 	sc->sc_start = lagg_lacp_start;
1908 	sc->sc_input = lagg_lacp_input;
1909 	sc->sc_init = lacp_init;
1910 	sc->sc_stop = lacp_stop;
1911 	sc->sc_lladdr = lagg_lacp_lladdr;
1912 	sc->sc_req = lacp_req;
1913 	sc->sc_portreq = lacp_portreq;
1914 
1915 	error = lacp_attach(sc);
1916 	if (error)
1917 		return (error);
1918 
1919 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1920 		lacp_port_create(lp);
1921 
1922 	return (error);
1923 }
1924 
1925 static int
1926 lagg_lacp_detach(struct lagg_softc *sc)
1927 {
1928 	struct lagg_port *lp;
1929 	int error;
1930 
1931 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1932 		lacp_port_destroy(lp);
1933 
1934 	/* unlocking is safe here */
1935 	LAGG_WUNLOCK(sc);
1936 	error = lacp_detach(sc);
1937 	LAGG_WLOCK(sc);
1938 
1939 	return (error);
1940 }
1941 
1942 static void
1943 lagg_lacp_lladdr(struct lagg_softc *sc)
1944 {
1945 	struct lagg_port *lp;
1946 
1947 	/* purge all the lacp ports */
1948 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1949 		lacp_port_destroy(lp);
1950 
1951 	/* add them back in */
1952 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1953 		lacp_port_create(lp);
1954 }
1955 
1956 static int
1957 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1958 {
1959 	struct lagg_port *lp;
1960 
1961 	lp = lacp_select_tx_port(sc, m);
1962 	if (lp == NULL) {
1963 		m_freem(m);
1964 		return (ENETDOWN);
1965 	}
1966 
1967 	/* Send mbuf */
1968 	return (lagg_enqueue(lp->lp_ifp, m));
1969 }
1970 
1971 static struct mbuf *
1972 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1973 {
1974 	struct ifnet *ifp = sc->sc_ifp;
1975 	struct ether_header *eh;
1976 	u_short etype;
1977 
1978 	eh = mtod(m, struct ether_header *);
1979 	etype = ntohs(eh->ether_type);
1980 
1981 	/* Tap off LACP control messages */
1982 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
1983 		m = lacp_input(lp, m);
1984 		if (m == NULL)
1985 			return (NULL);
1986 	}
1987 
1988 	/*
1989 	 * If the port is not collecting or not in the active aggregator then
1990 	 * free and return.
1991 	 */
1992 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
1993 		m_freem(m);
1994 		return (NULL);
1995 	}
1996 
1997 	m->m_pkthdr.rcvif = ifp;
1998 	return (m);
1999 }
2000 
2001 static void
2002 lagg_callout(void *arg)
2003 {
2004 	struct lagg_softc *sc = (struct lagg_softc *)arg;
2005 	struct ifnet *ifp = sc->sc_ifp;
2006 
2007 	ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets);
2008 	ifp->if_opackets = counter_u64_fetch(sc->sc_opackets);
2009 	ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes);
2010 	ifp->if_obytes = counter_u64_fetch(sc->sc_obytes);
2011 
2012 	callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
2013 }
2014