xref: /freebsd/sys/net/if_lagg.c (revision cc3f4b99653c34ae64f8a1fddea370abefef680e)
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 #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 /* 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 #if defined(INET) || defined(INET6)
452 	u_int hw_tsomax = IP_MAXPACKET;	/* Initialize to the maximum value. */
453 #else
454 	u_int hw_tsomax = ~0;	/* if_hw_tsomax is only for INET/INET6, but.. */
455 #endif
456 
457 	LAGG_WLOCK_ASSERT(sc);
458 
459 	/* Get capabilities from the lagg ports */
460 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
461 		cap &= lp->lp_ifp->if_capabilities;
462 		ena &= lp->lp_ifp->if_capenable;
463 		hwa &= lp->lp_ifp->if_hwassist;
464 		/* Set to the minimum value of the lagg ports. */
465 		if (lp->lp_ifp->if_hw_tsomax < hw_tsomax &&
466 		    lp->lp_ifp->if_hw_tsomax > 0)
467 			hw_tsomax = lp->lp_ifp->if_hw_tsomax;
468 	}
469 	cap = (cap == ~0 ? 0 : cap);
470 	ena = (ena == ~0 ? 0 : ena);
471 	hwa = (hwa == ~0 ? 0 : hwa);
472 
473 	if (sc->sc_ifp->if_capabilities != cap ||
474 	    sc->sc_ifp->if_capenable != ena ||
475 	    sc->sc_ifp->if_hwassist != hwa ||
476 	    sc->sc_ifp->if_hw_tsomax != hw_tsomax) {
477 		sc->sc_ifp->if_capabilities = cap;
478 		sc->sc_ifp->if_capenable = ena;
479 		sc->sc_ifp->if_hwassist = hwa;
480 		sc->sc_ifp->if_hw_tsomax = hw_tsomax;
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 			if (LAGG_PORTACTIVE(lp))
925 				rp->rp_flags |= LAGG_PORT_ACTIVE;
926 			break;
927 
928 		case LAGG_PROTO_LACP:
929 			/* LACP has a different definition of active */
930 			if (lacp_isactive(lp))
931 				rp->rp_flags |= LAGG_PORT_ACTIVE;
932 			if (lacp_iscollecting(lp))
933 				rp->rp_flags |= LAGG_PORT_COLLECTING;
934 			if (lacp_isdistributing(lp))
935 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
936 			break;
937 	}
938 
939 }
940 
941 static void
942 lagg_init(void *xsc)
943 {
944 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
945 	struct lagg_port *lp;
946 	struct ifnet *ifp = sc->sc_ifp;
947 
948 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
949 		return;
950 
951 	LAGG_WLOCK(sc);
952 
953 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
954 	/* Update the port lladdrs */
955 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
956 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
957 
958 	if (sc->sc_init != NULL)
959 		(*sc->sc_init)(sc);
960 
961 	LAGG_WUNLOCK(sc);
962 }
963 
964 static void
965 lagg_stop(struct lagg_softc *sc)
966 {
967 	struct ifnet *ifp = sc->sc_ifp;
968 
969 	LAGG_WLOCK_ASSERT(sc);
970 
971 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
972 		return;
973 
974 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
975 
976 	if (sc->sc_stop != NULL)
977 		(*sc->sc_stop)(sc);
978 }
979 
980 static int
981 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
982 {
983 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
984 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
985 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
986 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
987 	struct ifreq *ifr = (struct ifreq *)data;
988 	struct lagg_port *lp;
989 	struct ifnet *tpif;
990 	struct thread *td = curthread;
991 	char *buf, *outbuf;
992 	int count, buflen, len, error = 0;
993 	struct rm_priotracker tracker;
994 
995 	bzero(&rpbuf, sizeof(rpbuf));
996 
997 	switch (cmd) {
998 	case SIOCGLAGG:
999 		LAGG_RLOCK(sc, &tracker);
1000 		count = 0;
1001 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1002 			count++;
1003 		buflen = count * sizeof(struct lagg_reqport);
1004 		LAGG_RUNLOCK(sc, &tracker);
1005 
1006 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1007 
1008 		LAGG_RLOCK(sc, &tracker);
1009 		ra->ra_proto = sc->sc_proto;
1010 		if (sc->sc_req != NULL)
1011 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
1012 
1013 		count = 0;
1014 		buf = outbuf;
1015 		len = min(ra->ra_size, buflen);
1016 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1017 			if (len < sizeof(rpbuf))
1018 				break;
1019 
1020 			lagg_port2req(lp, &rpbuf);
1021 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1022 			count++;
1023 			buf += sizeof(rpbuf);
1024 			len -= sizeof(rpbuf);
1025 		}
1026 		LAGG_RUNLOCK(sc, &tracker);
1027 		ra->ra_ports = count;
1028 		ra->ra_size = count * sizeof(rpbuf);
1029 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1030 		free(outbuf, M_TEMP);
1031 		break;
1032 	case SIOCSLAGG:
1033 		error = priv_check(td, PRIV_NET_LAGG);
1034 		if (error)
1035 			break;
1036 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1037 			error = EPROTONOSUPPORT;
1038 			break;
1039 		}
1040 		LAGG_WLOCK(sc);
1041 		if (sc->sc_proto != LAGG_PROTO_NONE) {
1042 			/* Reset protocol first in case detach unlocks */
1043 			sc->sc_proto = LAGG_PROTO_NONE;
1044 			error = sc->sc_detach(sc);
1045 			sc->sc_detach = NULL;
1046 			sc->sc_start = NULL;
1047 			sc->sc_input = NULL;
1048 			sc->sc_port_create = NULL;
1049 			sc->sc_port_destroy = NULL;
1050 			sc->sc_linkstate = NULL;
1051 			sc->sc_init = NULL;
1052 			sc->sc_stop = NULL;
1053 			sc->sc_lladdr = NULL;
1054 			sc->sc_req = NULL;
1055 			sc->sc_portreq = NULL;
1056 		} else if (sc->sc_input != NULL) {
1057 			/* Still detaching */
1058 			error = EBUSY;
1059 		}
1060 		if (error != 0) {
1061 			LAGG_WUNLOCK(sc);
1062 			break;
1063 		}
1064 		for (int i = 0; i < (sizeof(lagg_protos) /
1065 		    sizeof(lagg_protos[0])); i++) {
1066 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
1067 				if (sc->sc_ifflags & IFF_DEBUG)
1068 					printf("%s: using proto %u\n",
1069 					    sc->sc_ifname,
1070 					    lagg_protos[i].ti_proto);
1071 				sc->sc_proto = lagg_protos[i].ti_proto;
1072 				if (sc->sc_proto != LAGG_PROTO_NONE)
1073 					error = lagg_protos[i].ti_attach(sc);
1074 				LAGG_WUNLOCK(sc);
1075 				return (error);
1076 			}
1077 		}
1078 		LAGG_WUNLOCK(sc);
1079 		error = EPROTONOSUPPORT;
1080 		break;
1081 	case SIOCGLAGGFLAGS:
1082 		rf->rf_flags = sc->sc_flags;
1083 		break;
1084 	case SIOCSLAGGHASH:
1085 		error = priv_check(td, PRIV_NET_LAGG);
1086 		if (error)
1087 			break;
1088 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1089 			error = EINVAL;
1090 			break;
1091 		}
1092 		LAGG_WLOCK(sc);
1093 		sc->sc_flags &= ~LAGG_F_HASHMASK;
1094 		sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK;
1095 		LAGG_WUNLOCK(sc);
1096 		break;
1097 	case SIOCGLAGGPORT:
1098 		if (rp->rp_portname[0] == '\0' ||
1099 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1100 			error = EINVAL;
1101 			break;
1102 		}
1103 
1104 		LAGG_RLOCK(sc, &tracker);
1105 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1106 		    lp->lp_softc != sc) {
1107 			error = ENOENT;
1108 			LAGG_RUNLOCK(sc, &tracker);
1109 			break;
1110 		}
1111 
1112 		lagg_port2req(lp, rp);
1113 		LAGG_RUNLOCK(sc, &tracker);
1114 		break;
1115 	case SIOCSLAGGPORT:
1116 		error = priv_check(td, PRIV_NET_LAGG);
1117 		if (error)
1118 			break;
1119 		if (rp->rp_portname[0] == '\0' ||
1120 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1121 			error = EINVAL;
1122 			break;
1123 		}
1124 		LAGG_WLOCK(sc);
1125 		error = lagg_port_create(sc, tpif);
1126 		LAGG_WUNLOCK(sc);
1127 		break;
1128 	case SIOCSLAGGDELPORT:
1129 		error = priv_check(td, PRIV_NET_LAGG);
1130 		if (error)
1131 			break;
1132 		if (rp->rp_portname[0] == '\0' ||
1133 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1134 			error = EINVAL;
1135 			break;
1136 		}
1137 
1138 		LAGG_WLOCK(sc);
1139 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1140 		    lp->lp_softc != sc) {
1141 			error = ENOENT;
1142 			LAGG_WUNLOCK(sc);
1143 			break;
1144 		}
1145 
1146 		error = lagg_port_destroy(lp, 1);
1147 		LAGG_WUNLOCK(sc);
1148 		break;
1149 	case SIOCSIFFLAGS:
1150 		/* Set flags on ports too */
1151 		LAGG_WLOCK(sc);
1152 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1153 			lagg_setflags(lp, 1);
1154 		}
1155 		LAGG_WUNLOCK(sc);
1156 
1157 		if (!(ifp->if_flags & IFF_UP) &&
1158 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1159 			/*
1160 			 * If interface is marked down and it is running,
1161 			 * then stop and disable it.
1162 			 */
1163 			LAGG_WLOCK(sc);
1164 			lagg_stop(sc);
1165 			LAGG_WUNLOCK(sc);
1166 		} else if ((ifp->if_flags & IFF_UP) &&
1167 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1168 			/*
1169 			 * If interface is marked up and it is stopped, then
1170 			 * start it.
1171 			 */
1172 			(*ifp->if_init)(sc);
1173 		}
1174 		break;
1175 	case SIOCADDMULTI:
1176 	case SIOCDELMULTI:
1177 		LAGG_WLOCK(sc);
1178 		error = lagg_ether_setmulti(sc);
1179 		LAGG_WUNLOCK(sc);
1180 		break;
1181 	case SIOCSIFMEDIA:
1182 	case SIOCGIFMEDIA:
1183 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1184 		break;
1185 
1186 	case SIOCSIFCAP:
1187 	case SIOCSIFMTU:
1188 		/* Do not allow the MTU or caps to be directly changed */
1189 		error = EINVAL;
1190 		break;
1191 
1192 	default:
1193 		error = ether_ioctl(ifp, cmd, data);
1194 		break;
1195 	}
1196 	return (error);
1197 }
1198 
1199 static int
1200 lagg_ether_setmulti(struct lagg_softc *sc)
1201 {
1202 	struct lagg_port *lp;
1203 
1204 	LAGG_WLOCK_ASSERT(sc);
1205 
1206 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1207 		/* First, remove any existing filter entries. */
1208 		lagg_ether_cmdmulti(lp, 0);
1209 		/* copy all addresses from the lagg interface to the port */
1210 		lagg_ether_cmdmulti(lp, 1);
1211 	}
1212 	return (0);
1213 }
1214 
1215 static int
1216 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1217 {
1218 	struct lagg_softc *sc = lp->lp_softc;
1219 	struct ifnet *ifp = lp->lp_ifp;
1220 	struct ifnet *scifp = sc->sc_ifp;
1221 	struct lagg_mc *mc;
1222 	struct ifmultiaddr *ifma, *rifma = NULL;
1223 	struct sockaddr_dl sdl;
1224 	int error;
1225 
1226 	LAGG_WLOCK_ASSERT(sc);
1227 
1228 	link_init_sdl(ifp, (struct sockaddr *)&sdl, IFT_ETHER);
1229 	sdl.sdl_alen = ETHER_ADDR_LEN;
1230 
1231 	if (set) {
1232 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1233 			if (ifma->ifma_addr->sa_family != AF_LINK)
1234 				continue;
1235 			bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1236 			    LLADDR(&sdl), ETHER_ADDR_LEN);
1237 
1238 			error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1239 			if (error)
1240 				return (error);
1241 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1242 			if (mc == NULL)
1243 				return (ENOMEM);
1244 			mc->mc_ifma = rifma;
1245 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1246 		}
1247 	} else {
1248 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1249 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1250 			if_delmulti_ifma(mc->mc_ifma);
1251 			free(mc, M_DEVBUF);
1252 		}
1253 	}
1254 	return (0);
1255 }
1256 
1257 /* Handle a ref counted flag that should be set on the lagg port as well */
1258 static int
1259 lagg_setflag(struct lagg_port *lp, int flag, int status,
1260 	     int (*func)(struct ifnet *, int))
1261 {
1262 	struct lagg_softc *sc = lp->lp_softc;
1263 	struct ifnet *scifp = sc->sc_ifp;
1264 	struct ifnet *ifp = lp->lp_ifp;
1265 	int error;
1266 
1267 	LAGG_WLOCK_ASSERT(sc);
1268 
1269 	status = status ? (scifp->if_flags & flag) : 0;
1270 	/* Now "status" contains the flag value or 0 */
1271 
1272 	/*
1273 	 * See if recorded ports status is different from what
1274 	 * we want it to be.  If it is, flip it.  We record ports
1275 	 * status in lp_ifflags so that we won't clear ports flag
1276 	 * we haven't set.  In fact, we don't clear or set ports
1277 	 * flags directly, but get or release references to them.
1278 	 * That's why we can be sure that recorded flags still are
1279 	 * in accord with actual ports flags.
1280 	 */
1281 	if (status != (lp->lp_ifflags & flag)) {
1282 		error = (*func)(ifp, status);
1283 		if (error)
1284 			return (error);
1285 		lp->lp_ifflags &= ~flag;
1286 		lp->lp_ifflags |= status;
1287 	}
1288 	return (0);
1289 }
1290 
1291 /*
1292  * Handle IFF_* flags that require certain changes on the lagg port
1293  * if "status" is true, update ports flags respective to the lagg
1294  * if "status" is false, forcedly clear the flags set on port.
1295  */
1296 static int
1297 lagg_setflags(struct lagg_port *lp, int status)
1298 {
1299 	int error, i;
1300 
1301 	for (i = 0; lagg_pflags[i].flag; i++) {
1302 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1303 		    status, lagg_pflags[i].func);
1304 		if (error)
1305 			return (error);
1306 	}
1307 	return (0);
1308 }
1309 
1310 static int
1311 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1312 {
1313 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1314 	int error, len, mcast;
1315 	struct rm_priotracker tracker;
1316 
1317 	len = m->m_pkthdr.len;
1318 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1319 
1320 	LAGG_RLOCK(sc, &tracker);
1321 	/* We need a Tx algorithm and at least one port */
1322 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1323 		LAGG_RUNLOCK(sc, &tracker);
1324 		m_freem(m);
1325 		ifp->if_oerrors++;
1326 		return (ENXIO);
1327 	}
1328 
1329 	ETHER_BPF_MTAP(ifp, m);
1330 
1331 	error = (*sc->sc_start)(sc, m);
1332 	LAGG_RUNLOCK(sc, &tracker);
1333 
1334 	if (error == 0) {
1335 		counter_u64_add(sc->sc_opackets, 1);
1336 		counter_u64_add(sc->sc_obytes, len);
1337 		ifp->if_omcasts += mcast;
1338 	} else
1339 		ifp->if_oerrors++;
1340 
1341 	return (error);
1342 }
1343 
1344 /*
1345  * The ifp->if_qflush entry point for lagg(4) is no-op.
1346  */
1347 static void
1348 lagg_qflush(struct ifnet *ifp __unused)
1349 {
1350 }
1351 
1352 static struct mbuf *
1353 lagg_input(struct ifnet *ifp, struct mbuf *m)
1354 {
1355 	struct lagg_port *lp = ifp->if_lagg;
1356 	struct lagg_softc *sc = lp->lp_softc;
1357 	struct ifnet *scifp = sc->sc_ifp;
1358 	struct rm_priotracker tracker;
1359 
1360 	LAGG_RLOCK(sc, &tracker);
1361 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1362 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1363 	    sc->sc_proto == LAGG_PROTO_NONE) {
1364 		LAGG_RUNLOCK(sc, &tracker);
1365 		m_freem(m);
1366 		return (NULL);
1367 	}
1368 
1369 	ETHER_BPF_MTAP(scifp, m);
1370 
1371 	m = (*sc->sc_input)(sc, lp, m);
1372 
1373 	if (m != NULL) {
1374 		counter_u64_add(sc->sc_ipackets, 1);
1375 		counter_u64_add(sc->sc_ibytes, m->m_pkthdr.len);
1376 
1377 		if (scifp->if_flags & IFF_MONITOR) {
1378 			m_freem(m);
1379 			m = NULL;
1380 		}
1381 	}
1382 
1383 	LAGG_RUNLOCK(sc, &tracker);
1384 	return (m);
1385 }
1386 
1387 static int
1388 lagg_media_change(struct ifnet *ifp)
1389 {
1390 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1391 
1392 	if (sc->sc_ifflags & IFF_DEBUG)
1393 		printf("%s\n", __func__);
1394 
1395 	/* Ignore */
1396 	return (0);
1397 }
1398 
1399 static void
1400 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1401 {
1402 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1403 	struct lagg_port *lp;
1404 	struct rm_priotracker tracker;
1405 
1406 	imr->ifm_status = IFM_AVALID;
1407 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1408 
1409 	LAGG_RLOCK(sc, &tracker);
1410 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1411 		if (LAGG_PORTACTIVE(lp))
1412 			imr->ifm_status |= IFM_ACTIVE;
1413 	}
1414 	LAGG_RUNLOCK(sc, &tracker);
1415 }
1416 
1417 static void
1418 lagg_linkstate(struct lagg_softc *sc)
1419 {
1420 	struct lagg_port *lp;
1421 	int new_link = LINK_STATE_DOWN;
1422 	uint64_t speed;
1423 
1424 	/* Our link is considered up if at least one of our ports is active */
1425 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1426 		if (lp->lp_link_state == LINK_STATE_UP) {
1427 			new_link = LINK_STATE_UP;
1428 			break;
1429 		}
1430 	}
1431 	if_link_state_change(sc->sc_ifp, new_link);
1432 
1433 	/* Update if_baudrate to reflect the max possible speed */
1434 	switch (sc->sc_proto) {
1435 		case LAGG_PROTO_FAILOVER:
1436 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1437 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1438 			break;
1439 		case LAGG_PROTO_ROUNDROBIN:
1440 		case LAGG_PROTO_LOADBALANCE:
1441 		case LAGG_PROTO_ETHERCHANNEL:
1442 			speed = 0;
1443 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1444 				speed += lp->lp_ifp->if_baudrate;
1445 			sc->sc_ifp->if_baudrate = speed;
1446 			break;
1447 		case LAGG_PROTO_LACP:
1448 			/* LACP updates if_baudrate itself */
1449 			break;
1450 	}
1451 }
1452 
1453 static void
1454 lagg_port_state(struct ifnet *ifp, int state)
1455 {
1456 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1457 	struct lagg_softc *sc = NULL;
1458 
1459 	if (lp != NULL)
1460 		sc = lp->lp_softc;
1461 	if (sc == NULL)
1462 		return;
1463 
1464 	LAGG_WLOCK(sc);
1465 	lagg_linkstate(sc);
1466 	if (sc->sc_linkstate != NULL)
1467 		(*sc->sc_linkstate)(lp);
1468 	LAGG_WUNLOCK(sc);
1469 }
1470 
1471 struct lagg_port *
1472 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1473 {
1474 	struct lagg_port *lp_next, *rval = NULL;
1475 	// int new_link = LINK_STATE_DOWN;
1476 
1477 	LAGG_RLOCK_ASSERT(sc);
1478 	/*
1479 	 * Search a port which reports an active link state.
1480 	 */
1481 
1482 	if (lp == NULL)
1483 		goto search;
1484 	if (LAGG_PORTACTIVE(lp)) {
1485 		rval = lp;
1486 		goto found;
1487 	}
1488 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1489 	    LAGG_PORTACTIVE(lp_next)) {
1490 		rval = lp_next;
1491 		goto found;
1492 	}
1493 
1494 search:
1495 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1496 		if (LAGG_PORTACTIVE(lp_next)) {
1497 			rval = lp_next;
1498 			goto found;
1499 		}
1500 	}
1501 
1502 found:
1503 	if (rval != NULL) {
1504 		/*
1505 		 * The IEEE 802.1D standard assumes that a lagg with
1506 		 * multiple ports is always full duplex. This is valid
1507 		 * for load sharing laggs and if at least two links
1508 		 * are active. Unfortunately, checking the latter would
1509 		 * be too expensive at this point.
1510 		 XXX
1511 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1512 		    (sc->sc_count > 1))
1513 			new_link = LINK_STATE_FULL_DUPLEX;
1514 		else
1515 			new_link = rval->lp_link_state;
1516 		 */
1517 	}
1518 
1519 	return (rval);
1520 }
1521 
1522 static const void *
1523 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1524 {
1525 	if (m->m_pkthdr.len < (off + len)) {
1526 		return (NULL);
1527 	} else if (m->m_len < (off + len)) {
1528 		m_copydata(m, off, len, buf);
1529 		return (buf);
1530 	}
1531 	return (mtod(m, char *) + off);
1532 }
1533 
1534 static int
1535 lagg_sysctl_active(SYSCTL_HANDLER_ARGS)
1536 {
1537 	struct lagg_softc *sc = (struct lagg_softc *)arg1;
1538 	struct lagg_port *lp;
1539 	int error;
1540 
1541 	/* LACP tracks active links automatically, the others do not */
1542 	if (sc->sc_proto != LAGG_PROTO_LACP) {
1543 		sc->sc_active = 0;
1544 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1545 			sc->sc_active += LAGG_PORTACTIVE(lp);
1546 	}
1547 
1548 	error = sysctl_handle_int(oidp, &sc->sc_active, 0, req);
1549 	if ((error) || (req->newptr == NULL))
1550 		return (error);
1551 
1552 	return (0);
1553 }
1554 
1555 uint32_t
1556 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
1557 {
1558 	uint16_t etype;
1559 	uint32_t p = key;
1560 	int off;
1561 	struct ether_header *eh;
1562 	const struct ether_vlan_header *vlan;
1563 #ifdef INET
1564 	const struct ip *ip;
1565 	const uint32_t *ports;
1566 	int iphlen;
1567 #endif
1568 #ifdef INET6
1569 	const struct ip6_hdr *ip6;
1570 	uint32_t flow;
1571 #endif
1572 	union {
1573 #ifdef INET
1574 		struct ip ip;
1575 #endif
1576 #ifdef INET6
1577 		struct ip6_hdr ip6;
1578 #endif
1579 		struct ether_vlan_header vlan;
1580 		uint32_t port;
1581 	} buf;
1582 
1583 
1584 	off = sizeof(*eh);
1585 	if (m->m_len < off)
1586 		goto out;
1587 	eh = mtod(m, struct ether_header *);
1588 	etype = ntohs(eh->ether_type);
1589 	if (sc->sc_flags & LAGG_F_HASHL2) {
1590 		p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p);
1591 		p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1592 	}
1593 
1594 	/* Special handling for encapsulating VLAN frames */
1595 	if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) {
1596 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1597 		    sizeof(m->m_pkthdr.ether_vtag), p);
1598 	} else if (etype == ETHERTYPE_VLAN) {
1599 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &buf);
1600 		if (vlan == NULL)
1601 			goto out;
1602 
1603 		if (sc->sc_flags & LAGG_F_HASHL2)
1604 			p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1605 		etype = ntohs(vlan->evl_proto);
1606 		off += sizeof(*vlan) - sizeof(*eh);
1607 	}
1608 
1609 	switch (etype) {
1610 #ifdef INET
1611 	case ETHERTYPE_IP:
1612 		ip = lagg_gethdr(m, off, sizeof(*ip), &buf);
1613 		if (ip == NULL)
1614 			goto out;
1615 
1616 		if (sc->sc_flags & LAGG_F_HASHL3) {
1617 			p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1618 			p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1619 		}
1620 		if (!(sc->sc_flags & LAGG_F_HASHL4))
1621 			break;
1622 		switch (ip->ip_p) {
1623 			case IPPROTO_TCP:
1624 			case IPPROTO_UDP:
1625 			case IPPROTO_SCTP:
1626 				iphlen = ip->ip_hl << 2;
1627 				if (iphlen < sizeof(*ip))
1628 					break;
1629 				off += iphlen;
1630 				ports = lagg_gethdr(m, off, sizeof(*ports), &buf);
1631 				if (ports == NULL)
1632 					break;
1633 				p = hash32_buf(ports, sizeof(*ports), p);
1634 				break;
1635 		}
1636 		break;
1637 #endif
1638 #ifdef INET6
1639 	case ETHERTYPE_IPV6:
1640 		if (!(sc->sc_flags & LAGG_F_HASHL3))
1641 			break;
1642 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf);
1643 		if (ip6 == NULL)
1644 			goto out;
1645 
1646 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1647 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1648 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1649 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1650 		break;
1651 #endif
1652 	}
1653 out:
1654 	return (p);
1655 }
1656 
1657 int
1658 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1659 {
1660 
1661 	return (ifp->if_transmit)(ifp, m);
1662 }
1663 
1664 /*
1665  * Simple round robin aggregation
1666  */
1667 
1668 static int
1669 lagg_rr_attach(struct lagg_softc *sc)
1670 {
1671 	sc->sc_detach = lagg_rr_detach;
1672 	sc->sc_start = lagg_rr_start;
1673 	sc->sc_input = lagg_rr_input;
1674 	sc->sc_port_create = NULL;
1675 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1676 	sc->sc_seq = 0;
1677 
1678 	return (0);
1679 }
1680 
1681 static int
1682 lagg_rr_detach(struct lagg_softc *sc)
1683 {
1684 	return (0);
1685 }
1686 
1687 static int
1688 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1689 {
1690 	struct lagg_port *lp;
1691 	uint32_t p;
1692 
1693 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1694 	p %= sc->sc_count;
1695 	lp = SLIST_FIRST(&sc->sc_ports);
1696 	while (p--)
1697 		lp = SLIST_NEXT(lp, lp_entries);
1698 
1699 	/*
1700 	 * Check the port's link state. This will return the next active
1701 	 * port if the link is down or the port is NULL.
1702 	 */
1703 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1704 		m_freem(m);
1705 		return (ENETDOWN);
1706 	}
1707 
1708 	/* Send mbuf */
1709 	return (lagg_enqueue(lp->lp_ifp, m));
1710 }
1711 
1712 static struct mbuf *
1713 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1714 {
1715 	struct ifnet *ifp = sc->sc_ifp;
1716 
1717 	/* Just pass in the packet to our lagg device */
1718 	m->m_pkthdr.rcvif = ifp;
1719 
1720 	return (m);
1721 }
1722 
1723 /*
1724  * Active failover
1725  */
1726 
1727 static int
1728 lagg_fail_attach(struct lagg_softc *sc)
1729 {
1730 	sc->sc_detach = lagg_fail_detach;
1731 	sc->sc_start = lagg_fail_start;
1732 	sc->sc_input = lagg_fail_input;
1733 	sc->sc_port_create = NULL;
1734 	sc->sc_port_destroy = NULL;
1735 
1736 	return (0);
1737 }
1738 
1739 static int
1740 lagg_fail_detach(struct lagg_softc *sc)
1741 {
1742 	return (0);
1743 }
1744 
1745 static int
1746 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1747 {
1748 	struct lagg_port *lp;
1749 
1750 	/* Use the master port if active or the next available port */
1751 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1752 		m_freem(m);
1753 		return (ENETDOWN);
1754 	}
1755 
1756 	/* Send mbuf */
1757 	return (lagg_enqueue(lp->lp_ifp, m));
1758 }
1759 
1760 static struct mbuf *
1761 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1762 {
1763 	struct ifnet *ifp = sc->sc_ifp;
1764 	struct lagg_port *tmp_tp;
1765 
1766 	if (lp == sc->sc_primary || lagg_failover_rx_all) {
1767 		m->m_pkthdr.rcvif = ifp;
1768 		return (m);
1769 	}
1770 
1771 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1772 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1773 		/*
1774 		 * If tmp_tp is null, we've recieved a packet when all
1775 		 * our links are down. Weird, but process it anyways.
1776 		 */
1777 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1778 			m->m_pkthdr.rcvif = ifp;
1779 			return (m);
1780 		}
1781 	}
1782 
1783 	m_freem(m);
1784 	return (NULL);
1785 }
1786 
1787 /*
1788  * Loadbalancing
1789  */
1790 
1791 static int
1792 lagg_lb_attach(struct lagg_softc *sc)
1793 {
1794 	struct lagg_port *lp;
1795 	struct lagg_lb *lb;
1796 
1797 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1798 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1799 		return (ENOMEM);
1800 
1801 	sc->sc_detach = lagg_lb_detach;
1802 	sc->sc_start = lagg_lb_start;
1803 	sc->sc_input = lagg_lb_input;
1804 	sc->sc_port_create = lagg_lb_port_create;
1805 	sc->sc_port_destroy = lagg_lb_port_destroy;
1806 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1807 
1808 	lb->lb_key = arc4random();
1809 	sc->sc_psc = (caddr_t)lb;
1810 
1811 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1812 		lagg_lb_port_create(lp);
1813 
1814 	return (0);
1815 }
1816 
1817 static int
1818 lagg_lb_detach(struct lagg_softc *sc)
1819 {
1820 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1821 	if (lb != NULL)
1822 		free(lb, M_DEVBUF);
1823 	return (0);
1824 }
1825 
1826 static int
1827 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1828 {
1829 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1830 	struct lagg_port *lp_next;
1831 	int i = 0;
1832 
1833 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1834 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1835 		if (lp_next == lp)
1836 			continue;
1837 		if (i >= LAGG_MAX_PORTS)
1838 			return (EINVAL);
1839 		if (sc->sc_ifflags & IFF_DEBUG)
1840 			printf("%s: port %s at index %d\n",
1841 			    sc->sc_ifname, lp_next->lp_ifname, i);
1842 		lb->lb_ports[i++] = lp_next;
1843 	}
1844 
1845 	return (0);
1846 }
1847 
1848 static int
1849 lagg_lb_port_create(struct lagg_port *lp)
1850 {
1851 	struct lagg_softc *sc = lp->lp_softc;
1852 	return (lagg_lb_porttable(sc, NULL));
1853 }
1854 
1855 static void
1856 lagg_lb_port_destroy(struct lagg_port *lp)
1857 {
1858 	struct lagg_softc *sc = lp->lp_softc;
1859 	lagg_lb_porttable(sc, lp);
1860 }
1861 
1862 static int
1863 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1864 {
1865 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1866 	struct lagg_port *lp = NULL;
1867 	uint32_t p = 0;
1868 
1869 	if (sc->use_flowid && (m->m_flags & M_FLOWID))
1870 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
1871 	else
1872 		p = lagg_hashmbuf(sc, m, lb->lb_key);
1873 	p %= sc->sc_count;
1874 	lp = lb->lb_ports[p];
1875 
1876 	/*
1877 	 * Check the port's link state. This will return the next active
1878 	 * port if the link is down or the port is NULL.
1879 	 */
1880 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1881 		m_freem(m);
1882 		return (ENETDOWN);
1883 	}
1884 
1885 	/* Send mbuf */
1886 	return (lagg_enqueue(lp->lp_ifp, m));
1887 }
1888 
1889 static struct mbuf *
1890 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1891 {
1892 	struct ifnet *ifp = sc->sc_ifp;
1893 
1894 	/* Just pass in the packet to our lagg device */
1895 	m->m_pkthdr.rcvif = ifp;
1896 
1897 	return (m);
1898 }
1899 
1900 /*
1901  * 802.3ad LACP
1902  */
1903 
1904 static int
1905 lagg_lacp_attach(struct lagg_softc *sc)
1906 {
1907 	struct lagg_port *lp;
1908 	int error;
1909 
1910 	sc->sc_detach = lagg_lacp_detach;
1911 	sc->sc_port_create = lacp_port_create;
1912 	sc->sc_port_destroy = lacp_port_destroy;
1913 	sc->sc_linkstate = lacp_linkstate;
1914 	sc->sc_start = lagg_lacp_start;
1915 	sc->sc_input = lagg_lacp_input;
1916 	sc->sc_init = lacp_init;
1917 	sc->sc_stop = lacp_stop;
1918 	sc->sc_lladdr = lagg_lacp_lladdr;
1919 	sc->sc_req = lacp_req;
1920 	sc->sc_portreq = lacp_portreq;
1921 
1922 	error = lacp_attach(sc);
1923 	if (error)
1924 		return (error);
1925 
1926 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1927 		lacp_port_create(lp);
1928 
1929 	return (error);
1930 }
1931 
1932 static int
1933 lagg_lacp_detach(struct lagg_softc *sc)
1934 {
1935 	struct lagg_port *lp;
1936 	int error;
1937 
1938 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1939 		lacp_port_destroy(lp);
1940 
1941 	/* unlocking is safe here */
1942 	LAGG_WUNLOCK(sc);
1943 	error = lacp_detach(sc);
1944 	LAGG_WLOCK(sc);
1945 
1946 	return (error);
1947 }
1948 
1949 static void
1950 lagg_lacp_lladdr(struct lagg_softc *sc)
1951 {
1952 	struct lagg_port *lp;
1953 
1954 	/* purge all the lacp ports */
1955 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1956 		lacp_port_destroy(lp);
1957 
1958 	/* add them back in */
1959 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1960 		lacp_port_create(lp);
1961 }
1962 
1963 static int
1964 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1965 {
1966 	struct lagg_port *lp;
1967 
1968 	lp = lacp_select_tx_port(sc, m);
1969 	if (lp == NULL) {
1970 		m_freem(m);
1971 		return (ENETDOWN);
1972 	}
1973 
1974 	/* Send mbuf */
1975 	return (lagg_enqueue(lp->lp_ifp, m));
1976 }
1977 
1978 static struct mbuf *
1979 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1980 {
1981 	struct ifnet *ifp = sc->sc_ifp;
1982 	struct ether_header *eh;
1983 	u_short etype;
1984 
1985 	eh = mtod(m, struct ether_header *);
1986 	etype = ntohs(eh->ether_type);
1987 
1988 	/* Tap off LACP control messages */
1989 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
1990 		m = lacp_input(lp, m);
1991 		if (m == NULL)
1992 			return (NULL);
1993 	}
1994 
1995 	/*
1996 	 * If the port is not collecting or not in the active aggregator then
1997 	 * free and return.
1998 	 */
1999 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2000 		m_freem(m);
2001 		return (NULL);
2002 	}
2003 
2004 	m->m_pkthdr.rcvif = ifp;
2005 	return (m);
2006 }
2007 
2008 static void
2009 lagg_callout(void *arg)
2010 {
2011 	struct lagg_softc *sc = (struct lagg_softc *)arg;
2012 	struct ifnet *ifp = sc->sc_ifp;
2013 
2014 	ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets);
2015 	ifp->if_opackets = counter_u64_fetch(sc->sc_opackets);
2016 	ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes);
2017 	ifp->if_obytes = counter_u64_fetch(sc->sc_obytes);
2018 
2019 	callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
2020 }
2021