xref: /freebsd/sys/net/if_lagg.c (revision 10b9d77bf1ccf2f3affafa6261692cb92cf7e992)
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/rwlock.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 #ifdef INET
56 #include <netinet/in.h>
57 #include <netinet/in_systm.h>
58 #include <netinet/if_ether.h>
59 #include <netinet/ip.h>
60 #endif
61 
62 #ifdef INET6
63 #include <netinet/ip6.h>
64 #endif
65 
66 #include <net/if_vlan_var.h>
67 #include <net/if_lagg.h>
68 #include <net/ieee8023ad_lacp.h>
69 
70 /* Special flags we should propagate to the lagg ports. */
71 static struct {
72 	int flag;
73 	int (*func)(struct ifnet *, int);
74 } lagg_pflags[] = {
75 	{IFF_PROMISC, ifpromisc},
76 	{IFF_ALLMULTI, if_allmulti},
77 	{0, NULL}
78 };
79 
80 SLIST_HEAD(__trhead, lagg_softc) lagg_list;	/* list of laggs */
81 static struct mtx	lagg_list_mtx;
82 eventhandler_tag	lagg_detach_cookie = NULL;
83 
84 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
85 static void	lagg_clone_destroy(struct ifnet *);
86 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
87 static void	lagg_capabilities(struct lagg_softc *);
88 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
89 static void	lagg_port_setlladdr(void *, int);
90 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
91 static int	lagg_port_destroy(struct lagg_port *, int);
92 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
93 static void	lagg_linkstate(struct lagg_softc *);
94 static void	lagg_port_state(struct ifnet *, int);
95 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
96 static int	lagg_port_output(struct ifnet *, struct mbuf *,
97 		    struct sockaddr *, struct route *);
98 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
99 #ifdef LAGG_PORT_STACKING
100 static int	lagg_port_checkstacking(struct lagg_softc *);
101 #endif
102 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
103 static void	lagg_init(void *);
104 static void	lagg_stop(struct lagg_softc *);
105 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
106 static int	lagg_ether_setmulti(struct lagg_softc *);
107 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
108 static	int	lagg_setflag(struct lagg_port *, int, int,
109 		    int (*func)(struct ifnet *, int));
110 static	int	lagg_setflags(struct lagg_port *, int status);
111 static void	lagg_start(struct ifnet *);
112 static int	lagg_media_change(struct ifnet *);
113 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
114 static struct lagg_port *lagg_link_active(struct lagg_softc *,
115 	    struct lagg_port *);
116 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
117 
118 IFC_SIMPLE_DECLARE(lagg, 0);
119 
120 /* Simple round robin */
121 static int	lagg_rr_attach(struct lagg_softc *);
122 static int	lagg_rr_detach(struct lagg_softc *);
123 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
124 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
125 		    struct mbuf *);
126 
127 /* Active failover */
128 static int	lagg_fail_attach(struct lagg_softc *);
129 static int	lagg_fail_detach(struct lagg_softc *);
130 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
131 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
132 		    struct mbuf *);
133 
134 /* Loadbalancing */
135 static int	lagg_lb_attach(struct lagg_softc *);
136 static int	lagg_lb_detach(struct lagg_softc *);
137 static int	lagg_lb_port_create(struct lagg_port *);
138 static void	lagg_lb_port_destroy(struct lagg_port *);
139 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
140 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
141 		    struct mbuf *);
142 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
143 
144 /* 802.3ad LACP */
145 static int	lagg_lacp_attach(struct lagg_softc *);
146 static int	lagg_lacp_detach(struct lagg_softc *);
147 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
148 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
149 		    struct mbuf *);
150 static void	lagg_lacp_lladdr(struct lagg_softc *);
151 
152 /* lagg protocol table */
153 static const struct {
154 	int			ti_proto;
155 	int			(*ti_attach)(struct lagg_softc *);
156 } lagg_protos[] = {
157 	{ LAGG_PROTO_ROUNDROBIN,	lagg_rr_attach },
158 	{ LAGG_PROTO_FAILOVER,		lagg_fail_attach },
159 	{ LAGG_PROTO_LOADBALANCE,	lagg_lb_attach },
160 	{ LAGG_PROTO_ETHERCHANNEL,	lagg_lb_attach },
161 	{ LAGG_PROTO_LACP,		lagg_lacp_attach },
162 	{ LAGG_PROTO_NONE,		NULL }
163 };
164 
165 SYSCTL_DECL(_net_link);
166 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, "Link Aggregation");
167 
168 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */
169 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW,
170     &lagg_failover_rx_all, 0,
171     "Accept input from any interface in a failover lagg");
172 
173 static int
174 lagg_modevent(module_t mod, int type, void *data)
175 {
176 
177 	switch (type) {
178 	case MOD_LOAD:
179 		mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
180 		SLIST_INIT(&lagg_list);
181 		if_clone_attach(&lagg_cloner);
182 		lagg_input_p = lagg_input;
183 		lagg_linkstate_p = lagg_port_state;
184 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
185 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
186 		    EVENTHANDLER_PRI_ANY);
187 		break;
188 	case MOD_UNLOAD:
189 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
190 		    lagg_detach_cookie);
191 		if_clone_detach(&lagg_cloner);
192 		lagg_input_p = NULL;
193 		lagg_linkstate_p = NULL;
194 		mtx_destroy(&lagg_list_mtx);
195 		break;
196 	default:
197 		return (EOPNOTSUPP);
198 	}
199 	return (0);
200 }
201 
202 static moduledata_t lagg_mod = {
203 	"if_lagg",
204 	lagg_modevent,
205 	0
206 };
207 
208 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
209 
210 #if __FreeBSD_version >= 800000
211 /*
212  * This routine is run via an vlan
213  * config EVENT
214  */
215 static void
216 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
217 {
218         struct lagg_softc       *sc = ifp->if_softc;
219         struct lagg_port        *lp;
220 
221         if (ifp->if_softc !=  arg)   /* Not our event */
222                 return;
223 
224         LAGG_RLOCK(sc);
225         if (!SLIST_EMPTY(&sc->sc_ports)) {
226                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
227                         EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
228         }
229         LAGG_RUNLOCK(sc);
230 }
231 
232 /*
233  * This routine is run via an vlan
234  * unconfig EVENT
235  */
236 static void
237 lagg_unregister_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 
242         if (ifp->if_softc !=  arg)   /* Not our event */
243                 return;
244 
245         LAGG_RLOCK(sc);
246         if (!SLIST_EMPTY(&sc->sc_ports)) {
247                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
248                         EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
249         }
250         LAGG_RUNLOCK(sc);
251 }
252 #endif
253 
254 static int
255 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
256 {
257 	struct lagg_softc *sc;
258 	struct ifnet *ifp;
259 	int i, error = 0;
260 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
261 
262 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
263 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
264 	if (ifp == NULL) {
265 		free(sc, M_DEVBUF);
266 		return (ENOSPC);
267 	}
268 
269 	sc->sc_proto = LAGG_PROTO_NONE;
270 	for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
271 		if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
272 			sc->sc_proto = lagg_protos[i].ti_proto;
273 			if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
274 				if_free_type(ifp, IFT_ETHER);
275 				free(sc, M_DEVBUF);
276 				return (error);
277 			}
278 			break;
279 		}
280 	}
281 	LAGG_LOCK_INIT(sc);
282 	SLIST_INIT(&sc->sc_ports);
283 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
284 
285 	/* Initialise pseudo media types */
286 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
287 	    lagg_media_status);
288 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
289 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
290 
291 	if_initname(ifp, ifc->ifc_name, unit);
292 	ifp->if_type = IFT_ETHER;
293 	ifp->if_softc = sc;
294 	ifp->if_start = lagg_start;
295 	ifp->if_init = lagg_init;
296 	ifp->if_ioctl = lagg_ioctl;
297 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
298 
299 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
300 	ifp->if_snd.ifq_drv_maxlen = ifqmaxlen;
301 	IFQ_SET_READY(&ifp->if_snd);
302 
303 	/*
304 	 * Attach as an ordinary ethernet device, childs will be attached
305 	 * as special device IFT_IEEE8023ADLAG.
306 	 */
307 	ether_ifattach(ifp, eaddr);
308 
309 #if __FreeBSD_version >= 800000
310 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
311 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
312 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
313 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
314 #endif
315 
316 	/* Insert into the global list of laggs */
317 	mtx_lock(&lagg_list_mtx);
318 	SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
319 	mtx_unlock(&lagg_list_mtx);
320 
321 	return (0);
322 }
323 
324 static void
325 lagg_clone_destroy(struct ifnet *ifp)
326 {
327 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
328 	struct lagg_port *lp;
329 
330 	LAGG_WLOCK(sc);
331 
332 	lagg_stop(sc);
333 	ifp->if_flags &= ~IFF_UP;
334 
335 #if __FreeBSD_version >= 800000
336 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
337 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
338 #endif
339 
340 	/* Shutdown and remove lagg ports */
341 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
342 		lagg_port_destroy(lp, 1);
343 	/* Unhook the aggregation protocol */
344 	(*sc->sc_detach)(sc);
345 
346 	LAGG_WUNLOCK(sc);
347 
348 	ifmedia_removeall(&sc->sc_media);
349 	ether_ifdetach(ifp);
350 	if_free_type(ifp, IFT_ETHER);
351 
352 	mtx_lock(&lagg_list_mtx);
353 	SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
354 	mtx_unlock(&lagg_list_mtx);
355 
356 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
357 	LAGG_LOCK_DESTROY(sc);
358 	free(sc, M_DEVBUF);
359 }
360 
361 static void
362 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
363 {
364 	struct ifnet *ifp = sc->sc_ifp;
365 
366 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
367 		return;
368 
369 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
370 	/* Let the protocol know the MAC has changed */
371 	if (sc->sc_lladdr != NULL)
372 		(*sc->sc_lladdr)(sc);
373 	EVENTHANDLER_INVOKE(iflladdr_event, ifp);
374 }
375 
376 static void
377 lagg_capabilities(struct lagg_softc *sc)
378 {
379 	struct lagg_port *lp;
380 	int cap = ~0, ena = ~0;
381 	u_long hwa = ~0UL;
382 
383 	LAGG_WLOCK_ASSERT(sc);
384 
385 	/* Get capabilities from the lagg ports */
386 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
387 		cap &= lp->lp_ifp->if_capabilities;
388 		ena &= lp->lp_ifp->if_capenable;
389 		hwa &= lp->lp_ifp->if_hwassist;
390 	}
391 	cap = (cap == ~0 ? 0 : cap);
392 	ena = (ena == ~0 ? 0 : ena);
393 	hwa = (hwa == ~0 ? 0 : hwa);
394 
395 	if (sc->sc_ifp->if_capabilities != cap ||
396 	    sc->sc_ifp->if_capenable != ena ||
397 	    sc->sc_ifp->if_hwassist != hwa) {
398 		sc->sc_ifp->if_capabilities = cap;
399 		sc->sc_ifp->if_capenable = ena;
400 		sc->sc_ifp->if_hwassist = hwa;
401 		getmicrotime(&sc->sc_ifp->if_lastchange);
402 
403 		if (sc->sc_ifflags & IFF_DEBUG)
404 			if_printf(sc->sc_ifp,
405 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
406 	}
407 }
408 
409 static void
410 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
411 {
412 	struct lagg_softc *sc = lp->lp_softc;
413 	struct ifnet *ifp = lp->lp_ifp;
414 	struct lagg_llq *llq;
415 	int pending = 0;
416 
417 	LAGG_WLOCK_ASSERT(sc);
418 
419 	if (lp->lp_detaching ||
420 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
421 		return;
422 
423 	/* Check to make sure its not already queued to be changed */
424 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
425 		if (llq->llq_ifp == ifp) {
426 			pending = 1;
427 			break;
428 		}
429 	}
430 
431 	if (!pending) {
432 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
433 		if (llq == NULL)	/* XXX what to do */
434 			return;
435 	}
436 
437 	/* Update the lladdr even if pending, it may have changed */
438 	llq->llq_ifp = ifp;
439 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
440 
441 	if (!pending)
442 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
443 
444 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
445 }
446 
447 /*
448  * Set the interface MAC address from a taskqueue to avoid a LOR.
449  */
450 static void
451 lagg_port_setlladdr(void *arg, int pending)
452 {
453 	struct lagg_softc *sc = (struct lagg_softc *)arg;
454 	struct lagg_llq *llq, *head;
455 	struct ifnet *ifp;
456 	int error;
457 
458 	/* Grab a local reference of the queue and remove it from the softc */
459 	LAGG_WLOCK(sc);
460 	head = SLIST_FIRST(&sc->sc_llq_head);
461 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
462 	LAGG_WUNLOCK(sc);
463 
464 	/*
465 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
466 	 * unlocked as we have the only reference to it.
467 	 */
468 	for (llq = head; llq != NULL; llq = head) {
469 		ifp = llq->llq_ifp;
470 
471 		/* Set the link layer address */
472 		error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
473 		if (error)
474 			printf("%s: setlladdr failed on %s\n", __func__,
475 			    ifp->if_xname);
476 
477 		head = SLIST_NEXT(llq, llq_entries);
478 		free(llq, M_DEVBUF);
479 	}
480 }
481 
482 static int
483 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
484 {
485 	struct lagg_softc *sc_ptr;
486 	struct lagg_port *lp;
487 	int error = 0;
488 
489 	LAGG_WLOCK_ASSERT(sc);
490 
491 	/* Limit the maximal number of lagg ports */
492 	if (sc->sc_count >= LAGG_MAX_PORTS)
493 		return (ENOSPC);
494 
495 	/* Check if port has already been associated to a lagg */
496 	if (ifp->if_lagg != NULL)
497 		return (EBUSY);
498 
499 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
500 	if (ifp->if_type != IFT_ETHER)
501 		return (EPROTONOSUPPORT);
502 
503 	/* Allow the first Ethernet member to define the MTU */
504 	if (SLIST_EMPTY(&sc->sc_ports))
505 		sc->sc_ifp->if_mtu = ifp->if_mtu;
506 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
507 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
508 		    ifp->if_xname);
509 		return (EINVAL);
510 	}
511 
512 	if ((lp = malloc(sizeof(struct lagg_port),
513 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
514 		return (ENOMEM);
515 
516 	/* Check if port is a stacked lagg */
517 	mtx_lock(&lagg_list_mtx);
518 	SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
519 		if (ifp == sc_ptr->sc_ifp) {
520 			mtx_unlock(&lagg_list_mtx);
521 			free(lp, M_DEVBUF);
522 			return (EINVAL);
523 			/* XXX disable stacking for the moment, its untested */
524 #ifdef LAGG_PORT_STACKING
525 			lp->lp_flags |= LAGG_PORT_STACK;
526 			if (lagg_port_checkstacking(sc_ptr) >=
527 			    LAGG_MAX_STACKING) {
528 				mtx_unlock(&lagg_list_mtx);
529 				free(lp, M_DEVBUF);
530 				return (E2BIG);
531 			}
532 #endif
533 		}
534 	}
535 	mtx_unlock(&lagg_list_mtx);
536 
537 	/* Change the interface type */
538 	lp->lp_iftype = ifp->if_type;
539 	ifp->if_type = IFT_IEEE8023ADLAG;
540 	ifp->if_lagg = lp;
541 	lp->lp_ioctl = ifp->if_ioctl;
542 	ifp->if_ioctl = lagg_port_ioctl;
543 	lp->lp_output = ifp->if_output;
544 	ifp->if_output = lagg_port_output;
545 
546 	lp->lp_ifp = ifp;
547 	lp->lp_softc = sc;
548 
549 	/* Save port link layer address */
550 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
551 
552 	if (SLIST_EMPTY(&sc->sc_ports)) {
553 		sc->sc_primary = lp;
554 		lagg_lladdr(sc, IF_LLADDR(ifp));
555 	} else {
556 		/* Update link layer address for this port */
557 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
558 	}
559 
560 	/* Insert into the list of ports */
561 	SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
562 	sc->sc_count++;
563 
564 	/* Update lagg capabilities */
565 	lagg_capabilities(sc);
566 	lagg_linkstate(sc);
567 
568 	/* Add multicast addresses and interface flags to this port */
569 	lagg_ether_cmdmulti(lp, 1);
570 	lagg_setflags(lp, 1);
571 
572 	if (sc->sc_port_create != NULL)
573 		error = (*sc->sc_port_create)(lp);
574 	if (error) {
575 		/* remove the port again, without calling sc_port_destroy */
576 		lagg_port_destroy(lp, 0);
577 		return (error);
578 	}
579 
580 	return (error);
581 }
582 
583 #ifdef LAGG_PORT_STACKING
584 static int
585 lagg_port_checkstacking(struct lagg_softc *sc)
586 {
587 	struct lagg_softc *sc_ptr;
588 	struct lagg_port *lp;
589 	int m = 0;
590 
591 	LAGG_WLOCK_ASSERT(sc);
592 
593 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
594 		if (lp->lp_flags & LAGG_PORT_STACK) {
595 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
596 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
597 		}
598 	}
599 
600 	return (m + 1);
601 }
602 #endif
603 
604 static int
605 lagg_port_destroy(struct lagg_port *lp, int runpd)
606 {
607 	struct lagg_softc *sc = lp->lp_softc;
608 	struct lagg_port *lp_ptr;
609 	struct lagg_llq *llq;
610 	struct ifnet *ifp = lp->lp_ifp;
611 
612 	LAGG_WLOCK_ASSERT(sc);
613 
614 	if (runpd && sc->sc_port_destroy != NULL)
615 		(*sc->sc_port_destroy)(lp);
616 
617 	/*
618 	 * Remove multicast addresses and interface flags from this port and
619 	 * reset the MAC address, skip if the interface is being detached.
620 	 */
621 	if (!lp->lp_detaching) {
622 		lagg_ether_cmdmulti(lp, 0);
623 		lagg_setflags(lp, 0);
624 		lagg_port_lladdr(lp, lp->lp_lladdr);
625 	}
626 
627 	/* Restore interface */
628 	ifp->if_type = lp->lp_iftype;
629 	ifp->if_ioctl = lp->lp_ioctl;
630 	ifp->if_output = lp->lp_output;
631 	ifp->if_lagg = NULL;
632 
633 	/* Finally, remove the port from the lagg */
634 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
635 	sc->sc_count--;
636 
637 	/* Update the primary interface */
638 	if (lp == sc->sc_primary) {
639 		uint8_t lladdr[ETHER_ADDR_LEN];
640 
641 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
642 			bzero(&lladdr, ETHER_ADDR_LEN);
643 		} else {
644 			bcopy(lp_ptr->lp_lladdr,
645 			    lladdr, ETHER_ADDR_LEN);
646 		}
647 		lagg_lladdr(sc, lladdr);
648 		sc->sc_primary = lp_ptr;
649 
650 		/* Update link layer address for each port */
651 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
652 			lagg_port_lladdr(lp_ptr, lladdr);
653 	}
654 
655 	/* Remove any pending lladdr changes from the queue */
656 	if (lp->lp_detaching) {
657 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
658 			if (llq->llq_ifp == ifp) {
659 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
660 				    llq_entries);
661 				free(llq, M_DEVBUF);
662 				break;	/* Only appears once */
663 			}
664 		}
665 	}
666 
667 	if (lp->lp_ifflags)
668 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
669 
670 	free(lp, M_DEVBUF);
671 
672 	/* Update lagg capabilities */
673 	lagg_capabilities(sc);
674 	lagg_linkstate(sc);
675 
676 	return (0);
677 }
678 
679 static int
680 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
681 {
682 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
683 	struct lagg_softc *sc;
684 	struct lagg_port *lp = NULL;
685 	int error = 0;
686 
687 	/* Should be checked by the caller */
688 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
689 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
690 		goto fallback;
691 
692 	switch (cmd) {
693 	case SIOCGLAGGPORT:
694 		if (rp->rp_portname[0] == '\0' ||
695 		    ifunit(rp->rp_portname) != ifp) {
696 			error = EINVAL;
697 			break;
698 		}
699 
700 		LAGG_RLOCK(sc);
701 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
702 			error = ENOENT;
703 			LAGG_RUNLOCK(sc);
704 			break;
705 		}
706 
707 		lagg_port2req(lp, rp);
708 		LAGG_RUNLOCK(sc);
709 		break;
710 
711 	case SIOCSIFCAP:
712 		if (lp->lp_ioctl == NULL) {
713 			error = EINVAL;
714 			break;
715 		}
716 		error = (*lp->lp_ioctl)(ifp, cmd, data);
717 		if (error)
718 			break;
719 
720 		/* Update lagg interface capabilities */
721 		LAGG_WLOCK(sc);
722 		lagg_capabilities(sc);
723 		LAGG_WUNLOCK(sc);
724 		break;
725 
726 	case SIOCSIFMTU:
727 		/* Do not allow the MTU to be changed once joined */
728 		error = EINVAL;
729 		break;
730 
731 	default:
732 		goto fallback;
733 	}
734 
735 	return (error);
736 
737 fallback:
738 	if (lp->lp_ioctl != NULL)
739 		return ((*lp->lp_ioctl)(ifp, cmd, data));
740 
741 	return (EINVAL);
742 }
743 
744 static int
745 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
746 	struct sockaddr *dst, struct route *ro)
747 {
748 	struct lagg_port *lp = ifp->if_lagg;
749 	struct ether_header *eh;
750 	short type = 0;
751 
752 	switch (dst->sa_family) {
753 		case pseudo_AF_HDRCMPLT:
754 		case AF_UNSPEC:
755 			eh = (struct ether_header *)dst->sa_data;
756 			type = eh->ether_type;
757 			break;
758 	}
759 
760 	/*
761 	 * Only allow ethernet types required to initiate or maintain the link,
762 	 * aggregated frames take a different path.
763 	 */
764 	switch (ntohs(type)) {
765 		case ETHERTYPE_PAE:	/* EAPOL PAE/802.1x */
766 			return ((*lp->lp_output)(ifp, m, dst, ro));
767 	}
768 
769 	/* drop any other frames */
770 	m_freem(m);
771 	return (EBUSY);
772 }
773 
774 static void
775 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
776 {
777 	struct lagg_port *lp;
778 	struct lagg_softc *sc;
779 
780 	if ((lp = ifp->if_lagg) == NULL)
781 		return;
782 
783 	sc = lp->lp_softc;
784 
785 	LAGG_WLOCK(sc);
786 	lp->lp_detaching = 1;
787 	lagg_port_destroy(lp, 1);
788 	LAGG_WUNLOCK(sc);
789 }
790 
791 static void
792 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
793 {
794 	struct lagg_softc *sc = lp->lp_softc;
795 
796 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
797 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
798 	rp->rp_prio = lp->lp_prio;
799 	rp->rp_flags = lp->lp_flags;
800 	if (sc->sc_portreq != NULL)
801 		(*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
802 
803 	/* Add protocol specific flags */
804 	switch (sc->sc_proto) {
805 		case LAGG_PROTO_FAILOVER:
806 			if (lp == sc->sc_primary)
807 				rp->rp_flags |= LAGG_PORT_MASTER;
808 			if (lp == lagg_link_active(sc, sc->sc_primary))
809 				rp->rp_flags |= LAGG_PORT_ACTIVE;
810 			break;
811 
812 		case LAGG_PROTO_ROUNDROBIN:
813 		case LAGG_PROTO_LOADBALANCE:
814 		case LAGG_PROTO_ETHERCHANNEL:
815 			if (LAGG_PORTACTIVE(lp))
816 				rp->rp_flags |= LAGG_PORT_ACTIVE;
817 			break;
818 
819 		case LAGG_PROTO_LACP:
820 			/* LACP has a different definition of active */
821 			if (lacp_isactive(lp))
822 				rp->rp_flags |= LAGG_PORT_ACTIVE;
823 			if (lacp_iscollecting(lp))
824 				rp->rp_flags |= LAGG_PORT_COLLECTING;
825 			if (lacp_isdistributing(lp))
826 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
827 			break;
828 	}
829 
830 }
831 
832 static void
833 lagg_init(void *xsc)
834 {
835 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
836 	struct lagg_port *lp;
837 	struct ifnet *ifp = sc->sc_ifp;
838 
839 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
840 		return;
841 
842 	LAGG_WLOCK(sc);
843 
844 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
845 	/* Update the port lladdrs */
846 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
847 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
848 
849 	if (sc->sc_init != NULL)
850 		(*sc->sc_init)(sc);
851 
852 	LAGG_WUNLOCK(sc);
853 }
854 
855 static void
856 lagg_stop(struct lagg_softc *sc)
857 {
858 	struct ifnet *ifp = sc->sc_ifp;
859 
860 	LAGG_WLOCK_ASSERT(sc);
861 
862 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
863 		return;
864 
865 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
866 
867 	if (sc->sc_stop != NULL)
868 		(*sc->sc_stop)(sc);
869 }
870 
871 static int
872 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
873 {
874 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
875 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
876 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
877 	struct ifreq *ifr = (struct ifreq *)data;
878 	struct lagg_port *lp;
879 	struct ifnet *tpif;
880 	struct thread *td = curthread;
881 	char *buf, *outbuf;
882 	int count, buflen, len, error = 0;
883 
884 	bzero(&rpbuf, sizeof(rpbuf));
885 
886 	switch (cmd) {
887 	case SIOCGLAGG:
888 		LAGG_RLOCK(sc);
889 		count = 0;
890 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
891 			count++;
892 		buflen = count * sizeof(struct lagg_reqport);
893 		LAGG_RUNLOCK(sc);
894 
895 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
896 
897 		LAGG_RLOCK(sc);
898 		ra->ra_proto = sc->sc_proto;
899 		if (sc->sc_req != NULL)
900 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
901 
902 		count = 0;
903 		buf = outbuf;
904 		len = min(ra->ra_size, buflen);
905 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
906 			if (len < sizeof(rpbuf))
907 				break;
908 
909 			lagg_port2req(lp, &rpbuf);
910 			memcpy(buf, &rpbuf, sizeof(rpbuf));
911 			count++;
912 			buf += sizeof(rpbuf);
913 			len -= sizeof(rpbuf);
914 		}
915 		LAGG_RUNLOCK(sc);
916 		ra->ra_ports = count;
917 		ra->ra_size = count * sizeof(rpbuf);
918 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
919 		free(outbuf, M_TEMP);
920 		break;
921 	case SIOCSLAGG:
922 		error = priv_check(td, PRIV_NET_LAGG);
923 		if (error)
924 			break;
925 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
926 			error = EPROTONOSUPPORT;
927 			break;
928 		}
929 		if (sc->sc_proto != LAGG_PROTO_NONE) {
930 			LAGG_WLOCK(sc);
931 			error = sc->sc_detach(sc);
932 			/* Reset protocol and pointers */
933 			sc->sc_proto = LAGG_PROTO_NONE;
934 			sc->sc_detach = NULL;
935 			sc->sc_start = NULL;
936 			sc->sc_input = NULL;
937 			sc->sc_port_create = NULL;
938 			sc->sc_port_destroy = NULL;
939 			sc->sc_linkstate = NULL;
940 			sc->sc_init = NULL;
941 			sc->sc_stop = NULL;
942 			sc->sc_lladdr = NULL;
943 			sc->sc_req = NULL;
944 			sc->sc_portreq = NULL;
945 			LAGG_WUNLOCK(sc);
946 		}
947 		if (error != 0)
948 			break;
949 		for (int i = 0; i < (sizeof(lagg_protos) /
950 		    sizeof(lagg_protos[0])); i++) {
951 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
952 				if (sc->sc_ifflags & IFF_DEBUG)
953 					printf("%s: using proto %u\n",
954 					    sc->sc_ifname,
955 					    lagg_protos[i].ti_proto);
956 				LAGG_WLOCK(sc);
957 				sc->sc_proto = lagg_protos[i].ti_proto;
958 				if (sc->sc_proto != LAGG_PROTO_NONE)
959 					error = lagg_protos[i].ti_attach(sc);
960 				LAGG_WUNLOCK(sc);
961 				return (error);
962 			}
963 		}
964 		error = EPROTONOSUPPORT;
965 		break;
966 	case SIOCGLAGGPORT:
967 		if (rp->rp_portname[0] == '\0' ||
968 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
969 			error = EINVAL;
970 			break;
971 		}
972 
973 		LAGG_RLOCK(sc);
974 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
975 		    lp->lp_softc != sc) {
976 			error = ENOENT;
977 			LAGG_RUNLOCK(sc);
978 			break;
979 		}
980 
981 		lagg_port2req(lp, rp);
982 		LAGG_RUNLOCK(sc);
983 		break;
984 	case SIOCSLAGGPORT:
985 		error = priv_check(td, PRIV_NET_LAGG);
986 		if (error)
987 			break;
988 		if (rp->rp_portname[0] == '\0' ||
989 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
990 			error = EINVAL;
991 			break;
992 		}
993 		LAGG_WLOCK(sc);
994 		error = lagg_port_create(sc, tpif);
995 		LAGG_WUNLOCK(sc);
996 		break;
997 	case SIOCSLAGGDELPORT:
998 		error = priv_check(td, PRIV_NET_LAGG);
999 		if (error)
1000 			break;
1001 		if (rp->rp_portname[0] == '\0' ||
1002 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1003 			error = EINVAL;
1004 			break;
1005 		}
1006 
1007 		LAGG_WLOCK(sc);
1008 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1009 		    lp->lp_softc != sc) {
1010 			error = ENOENT;
1011 			LAGG_WUNLOCK(sc);
1012 			break;
1013 		}
1014 
1015 		error = lagg_port_destroy(lp, 1);
1016 		LAGG_WUNLOCK(sc);
1017 		break;
1018 	case SIOCSIFFLAGS:
1019 		/* Set flags on ports too */
1020 		LAGG_WLOCK(sc);
1021 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1022 			lagg_setflags(lp, 1);
1023 		}
1024 		LAGG_WUNLOCK(sc);
1025 
1026 		if (!(ifp->if_flags & IFF_UP) &&
1027 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1028 			/*
1029 			 * If interface is marked down and it is running,
1030 			 * then stop and disable it.
1031 			 */
1032 			LAGG_WLOCK(sc);
1033 			lagg_stop(sc);
1034 			LAGG_WUNLOCK(sc);
1035 		} else if ((ifp->if_flags & IFF_UP) &&
1036 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1037 			/*
1038 			 * If interface is marked up and it is stopped, then
1039 			 * start it.
1040 			 */
1041 			(*ifp->if_init)(sc);
1042 		}
1043 		break;
1044 	case SIOCADDMULTI:
1045 	case SIOCDELMULTI:
1046 		LAGG_WLOCK(sc);
1047 		error = lagg_ether_setmulti(sc);
1048 		LAGG_WUNLOCK(sc);
1049 		break;
1050 	case SIOCSIFMEDIA:
1051 	case SIOCGIFMEDIA:
1052 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1053 		break;
1054 
1055 	case SIOCSIFCAP:
1056 	case SIOCSIFMTU:
1057 		/* Do not allow the MTU or caps to be directly changed */
1058 		error = EINVAL;
1059 		break;
1060 
1061 	default:
1062 		error = ether_ioctl(ifp, cmd, data);
1063 		break;
1064 	}
1065 	return (error);
1066 }
1067 
1068 static int
1069 lagg_ether_setmulti(struct lagg_softc *sc)
1070 {
1071 	struct lagg_port *lp;
1072 
1073 	LAGG_WLOCK_ASSERT(sc);
1074 
1075 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1076 		/* First, remove any existing filter entries. */
1077 		lagg_ether_cmdmulti(lp, 0);
1078 		/* copy all addresses from the lagg interface to the port */
1079 		lagg_ether_cmdmulti(lp, 1);
1080 	}
1081 	return (0);
1082 }
1083 
1084 static int
1085 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1086 {
1087 	struct lagg_softc *sc = lp->lp_softc;
1088 	struct ifnet *ifp = lp->lp_ifp;
1089 	struct ifnet *scifp = sc->sc_ifp;
1090 	struct lagg_mc *mc;
1091 	struct ifmultiaddr *ifma, *rifma = NULL;
1092 	struct sockaddr_dl sdl;
1093 	int error;
1094 
1095 	LAGG_WLOCK_ASSERT(sc);
1096 
1097 	bzero((char *)&sdl, sizeof(sdl));
1098 	sdl.sdl_len = sizeof(sdl);
1099 	sdl.sdl_family = AF_LINK;
1100 	sdl.sdl_type = IFT_ETHER;
1101 	sdl.sdl_alen = ETHER_ADDR_LEN;
1102 	sdl.sdl_index = ifp->if_index;
1103 
1104 	if (set) {
1105 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1106 			if (ifma->ifma_addr->sa_family != AF_LINK)
1107 				continue;
1108 			bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1109 			    LLADDR(&sdl), ETHER_ADDR_LEN);
1110 
1111 			error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1112 			if (error)
1113 				return (error);
1114 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1115 			if (mc == NULL)
1116 				return (ENOMEM);
1117 			mc->mc_ifma = rifma;
1118 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1119 		}
1120 	} else {
1121 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1122 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1123 			if_delmulti_ifma(mc->mc_ifma);
1124 			free(mc, M_DEVBUF);
1125 		}
1126 	}
1127 	return (0);
1128 }
1129 
1130 /* Handle a ref counted flag that should be set on the lagg port as well */
1131 static int
1132 lagg_setflag(struct lagg_port *lp, int flag, int status,
1133 	     int (*func)(struct ifnet *, int))
1134 {
1135 	struct lagg_softc *sc = lp->lp_softc;
1136 	struct ifnet *scifp = sc->sc_ifp;
1137 	struct ifnet *ifp = lp->lp_ifp;
1138 	int error;
1139 
1140 	LAGG_WLOCK_ASSERT(sc);
1141 
1142 	status = status ? (scifp->if_flags & flag) : 0;
1143 	/* Now "status" contains the flag value or 0 */
1144 
1145 	/*
1146 	 * See if recorded ports status is different from what
1147 	 * we want it to be.  If it is, flip it.  We record ports
1148 	 * status in lp_ifflags so that we won't clear ports flag
1149 	 * we haven't set.  In fact, we don't clear or set ports
1150 	 * flags directly, but get or release references to them.
1151 	 * That's why we can be sure that recorded flags still are
1152 	 * in accord with actual ports flags.
1153 	 */
1154 	if (status != (lp->lp_ifflags & flag)) {
1155 		error = (*func)(ifp, status);
1156 		if (error)
1157 			return (error);
1158 		lp->lp_ifflags &= ~flag;
1159 		lp->lp_ifflags |= status;
1160 	}
1161 	return (0);
1162 }
1163 
1164 /*
1165  * Handle IFF_* flags that require certain changes on the lagg port
1166  * if "status" is true, update ports flags respective to the lagg
1167  * if "status" is false, forcedly clear the flags set on port.
1168  */
1169 static int
1170 lagg_setflags(struct lagg_port *lp, int status)
1171 {
1172 	int error, i;
1173 
1174 	for (i = 0; lagg_pflags[i].flag; i++) {
1175 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1176 		    status, lagg_pflags[i].func);
1177 		if (error)
1178 			return (error);
1179 	}
1180 	return (0);
1181 }
1182 
1183 static void
1184 lagg_start(struct ifnet *ifp)
1185 {
1186 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1187 	struct mbuf *m;
1188 	int error = 0;
1189 
1190 	LAGG_RLOCK(sc);
1191 	/* We need a Tx algorithm and at least one port */
1192 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1193 		IF_DRAIN(&ifp->if_snd);
1194 		LAGG_RUNLOCK(sc);
1195 		return;
1196 	}
1197 
1198 	for (;; error = 0) {
1199 		IFQ_DEQUEUE(&ifp->if_snd, m);
1200 		if (m == NULL)
1201 			break;
1202 
1203 		ETHER_BPF_MTAP(ifp, m);
1204 
1205 		error = (*sc->sc_start)(sc, m);
1206 		if (error == 0)
1207 			ifp->if_opackets++;
1208 		else
1209 			ifp->if_oerrors++;
1210 	}
1211 	LAGG_RUNLOCK(sc);
1212 }
1213 
1214 static struct mbuf *
1215 lagg_input(struct ifnet *ifp, struct mbuf *m)
1216 {
1217 	struct lagg_port *lp = ifp->if_lagg;
1218 	struct lagg_softc *sc = lp->lp_softc;
1219 	struct ifnet *scifp = sc->sc_ifp;
1220 
1221 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1222 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1223 	    sc->sc_proto == LAGG_PROTO_NONE) {
1224 		m_freem(m);
1225 		return (NULL);
1226 	}
1227 
1228 	LAGG_RLOCK(sc);
1229 	ETHER_BPF_MTAP(scifp, m);
1230 
1231 	m = (*sc->sc_input)(sc, lp, m);
1232 
1233 	if (m != NULL) {
1234 		scifp->if_ipackets++;
1235 		scifp->if_ibytes += m->m_pkthdr.len;
1236 
1237 		if (scifp->if_flags & IFF_MONITOR) {
1238 			m_freem(m);
1239 			m = NULL;
1240 		}
1241 	}
1242 
1243 	LAGG_RUNLOCK(sc);
1244 	return (m);
1245 }
1246 
1247 static int
1248 lagg_media_change(struct ifnet *ifp)
1249 {
1250 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1251 
1252 	if (sc->sc_ifflags & IFF_DEBUG)
1253 		printf("%s\n", __func__);
1254 
1255 	/* Ignore */
1256 	return (0);
1257 }
1258 
1259 static void
1260 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1261 {
1262 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1263 	struct lagg_port *lp;
1264 
1265 	imr->ifm_status = IFM_AVALID;
1266 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1267 
1268 	LAGG_RLOCK(sc);
1269 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1270 		if (LAGG_PORTACTIVE(lp))
1271 			imr->ifm_status |= IFM_ACTIVE;
1272 	}
1273 	LAGG_RUNLOCK(sc);
1274 }
1275 
1276 static void
1277 lagg_linkstate(struct lagg_softc *sc)
1278 {
1279 	struct lagg_port *lp;
1280 	int new_link = LINK_STATE_DOWN;
1281 	uint64_t speed;
1282 
1283 	/* Our link is considered up if at least one of our ports is active */
1284 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1285 		if (lp->lp_link_state == LINK_STATE_UP) {
1286 			new_link = LINK_STATE_UP;
1287 			break;
1288 		}
1289 	}
1290 	if_link_state_change(sc->sc_ifp, new_link);
1291 
1292 	/* Update if_baudrate to reflect the max possible speed */
1293 	switch (sc->sc_proto) {
1294 		case LAGG_PROTO_FAILOVER:
1295 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1296 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1297 			break;
1298 		case LAGG_PROTO_ROUNDROBIN:
1299 		case LAGG_PROTO_LOADBALANCE:
1300 		case LAGG_PROTO_ETHERCHANNEL:
1301 			speed = 0;
1302 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1303 				speed += lp->lp_ifp->if_baudrate;
1304 			sc->sc_ifp->if_baudrate = speed;
1305 			break;
1306 		case LAGG_PROTO_LACP:
1307 			/* LACP updates if_baudrate itself */
1308 			break;
1309 	}
1310 }
1311 
1312 static void
1313 lagg_port_state(struct ifnet *ifp, int state)
1314 {
1315 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1316 	struct lagg_softc *sc = NULL;
1317 
1318 	if (lp != NULL)
1319 		sc = lp->lp_softc;
1320 	if (sc == NULL)
1321 		return;
1322 
1323 	LAGG_WLOCK(sc);
1324 	lagg_linkstate(sc);
1325 	if (sc->sc_linkstate != NULL)
1326 		(*sc->sc_linkstate)(lp);
1327 	LAGG_WUNLOCK(sc);
1328 }
1329 
1330 struct lagg_port *
1331 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1332 {
1333 	struct lagg_port *lp_next, *rval = NULL;
1334 	// int new_link = LINK_STATE_DOWN;
1335 
1336 	LAGG_RLOCK_ASSERT(sc);
1337 	/*
1338 	 * Search a port which reports an active link state.
1339 	 */
1340 
1341 	if (lp == NULL)
1342 		goto search;
1343 	if (LAGG_PORTACTIVE(lp)) {
1344 		rval = lp;
1345 		goto found;
1346 	}
1347 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1348 	    LAGG_PORTACTIVE(lp_next)) {
1349 		rval = lp_next;
1350 		goto found;
1351 	}
1352 
1353 search:
1354 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1355 		if (LAGG_PORTACTIVE(lp_next)) {
1356 			rval = lp_next;
1357 			goto found;
1358 		}
1359 	}
1360 
1361 found:
1362 	if (rval != NULL) {
1363 		/*
1364 		 * The IEEE 802.1D standard assumes that a lagg with
1365 		 * multiple ports is always full duplex. This is valid
1366 		 * for load sharing laggs and if at least two links
1367 		 * are active. Unfortunately, checking the latter would
1368 		 * be too expensive at this point.
1369 		 XXX
1370 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1371 		    (sc->sc_count > 1))
1372 			new_link = LINK_STATE_FULL_DUPLEX;
1373 		else
1374 			new_link = rval->lp_link_state;
1375 		 */
1376 	}
1377 
1378 	return (rval);
1379 }
1380 
1381 static const void *
1382 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1383 {
1384 	if (m->m_pkthdr.len < (off + len)) {
1385 		return (NULL);
1386 	} else if (m->m_len < (off + len)) {
1387 		m_copydata(m, off, len, buf);
1388 		return (buf);
1389 	}
1390 	return (mtod(m, char *) + off);
1391 }
1392 
1393 uint32_t
1394 lagg_hashmbuf(struct mbuf *m, uint32_t key)
1395 {
1396 	uint16_t etype;
1397 	uint32_t p = 0;
1398 	int off;
1399 	struct ether_header *eh;
1400 	struct ether_vlan_header vlanbuf;
1401 	const struct ether_vlan_header *vlan;
1402 #ifdef INET
1403 	const struct ip *ip;
1404 	struct ip ipbuf;
1405 #endif
1406 #ifdef INET6
1407 	const struct ip6_hdr *ip6;
1408 	struct ip6_hdr ip6buf;
1409 	uint32_t flow;
1410 #endif
1411 
1412 	off = sizeof(*eh);
1413 	if (m->m_len < off)
1414 		goto out;
1415 	eh = mtod(m, struct ether_header *);
1416 	etype = ntohs(eh->ether_type);
1417 	p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, key);
1418 	p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1419 
1420 	/* Special handling for encapsulating VLAN frames */
1421 	if (m->m_flags & M_VLANTAG) {
1422 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1423 		    sizeof(m->m_pkthdr.ether_vtag), p);
1424 	} else if (etype == ETHERTYPE_VLAN) {
1425 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &vlanbuf);
1426 		if (vlan == NULL)
1427 			goto out;
1428 
1429 		p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1430 		etype = ntohs(vlan->evl_proto);
1431 		off += sizeof(*vlan) - sizeof(*eh);
1432 	}
1433 
1434 	switch (etype) {
1435 #ifdef INET
1436 	case ETHERTYPE_IP:
1437 		ip = lagg_gethdr(m, off, sizeof(*ip), &ipbuf);
1438 		if (ip == NULL)
1439 			goto out;
1440 
1441 		p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1442 		p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1443 		break;
1444 #endif
1445 #ifdef INET6
1446 	case ETHERTYPE_IPV6:
1447 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &ip6buf);
1448 		if (ip6 == NULL)
1449 			goto out;
1450 
1451 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1452 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1453 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1454 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1455 		break;
1456 #endif
1457 	}
1458 out:
1459 	return (p);
1460 }
1461 
1462 int
1463 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1464 {
1465 
1466 	return (ifp->if_transmit)(ifp, m);
1467 }
1468 
1469 /*
1470  * Simple round robin aggregation
1471  */
1472 
1473 static int
1474 lagg_rr_attach(struct lagg_softc *sc)
1475 {
1476 	sc->sc_detach = lagg_rr_detach;
1477 	sc->sc_start = lagg_rr_start;
1478 	sc->sc_input = lagg_rr_input;
1479 	sc->sc_port_create = NULL;
1480 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1481 	sc->sc_seq = 0;
1482 
1483 	return (0);
1484 }
1485 
1486 static int
1487 lagg_rr_detach(struct lagg_softc *sc)
1488 {
1489 	return (0);
1490 }
1491 
1492 static int
1493 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1494 {
1495 	struct lagg_port *lp;
1496 	uint32_t p;
1497 
1498 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1499 	p %= sc->sc_count;
1500 	lp = SLIST_FIRST(&sc->sc_ports);
1501 	while (p--)
1502 		lp = SLIST_NEXT(lp, lp_entries);
1503 
1504 	/*
1505 	 * Check the port's link state. This will return the next active
1506 	 * port if the link is down or the port is NULL.
1507 	 */
1508 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1509 		m_freem(m);
1510 		return (ENOENT);
1511 	}
1512 
1513 	/* Send mbuf */
1514 	return (lagg_enqueue(lp->lp_ifp, m));
1515 }
1516 
1517 static struct mbuf *
1518 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1519 {
1520 	struct ifnet *ifp = sc->sc_ifp;
1521 
1522 	/* Just pass in the packet to our lagg device */
1523 	m->m_pkthdr.rcvif = ifp;
1524 
1525 	return (m);
1526 }
1527 
1528 /*
1529  * Active failover
1530  */
1531 
1532 static int
1533 lagg_fail_attach(struct lagg_softc *sc)
1534 {
1535 	sc->sc_detach = lagg_fail_detach;
1536 	sc->sc_start = lagg_fail_start;
1537 	sc->sc_input = lagg_fail_input;
1538 	sc->sc_port_create = NULL;
1539 	sc->sc_port_destroy = NULL;
1540 
1541 	return (0);
1542 }
1543 
1544 static int
1545 lagg_fail_detach(struct lagg_softc *sc)
1546 {
1547 	return (0);
1548 }
1549 
1550 static int
1551 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1552 {
1553 	struct lagg_port *lp;
1554 
1555 	/* Use the master port if active or the next available port */
1556 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1557 		m_freem(m);
1558 		return (ENOENT);
1559 	}
1560 
1561 	/* Send mbuf */
1562 	return (lagg_enqueue(lp->lp_ifp, m));
1563 }
1564 
1565 static struct mbuf *
1566 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1567 {
1568 	struct ifnet *ifp = sc->sc_ifp;
1569 	struct lagg_port *tmp_tp;
1570 
1571 	if (lp == sc->sc_primary || lagg_failover_rx_all) {
1572 		m->m_pkthdr.rcvif = ifp;
1573 		return (m);
1574 	}
1575 
1576 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1577 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1578 		/*
1579 		 * If tmp_tp is null, we've recieved a packet when all
1580 		 * our links are down. Weird, but process it anyways.
1581 		 */
1582 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1583 			m->m_pkthdr.rcvif = ifp;
1584 			return (m);
1585 		}
1586 	}
1587 
1588 	m_freem(m);
1589 	return (NULL);
1590 }
1591 
1592 /*
1593  * Loadbalancing
1594  */
1595 
1596 static int
1597 lagg_lb_attach(struct lagg_softc *sc)
1598 {
1599 	struct lagg_port *lp;
1600 	struct lagg_lb *lb;
1601 
1602 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1603 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1604 		return (ENOMEM);
1605 
1606 	sc->sc_detach = lagg_lb_detach;
1607 	sc->sc_start = lagg_lb_start;
1608 	sc->sc_input = lagg_lb_input;
1609 	sc->sc_port_create = lagg_lb_port_create;
1610 	sc->sc_port_destroy = lagg_lb_port_destroy;
1611 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1612 
1613 	lb->lb_key = arc4random();
1614 	sc->sc_psc = (caddr_t)lb;
1615 
1616 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1617 		lagg_lb_port_create(lp);
1618 
1619 	return (0);
1620 }
1621 
1622 static int
1623 lagg_lb_detach(struct lagg_softc *sc)
1624 {
1625 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1626 	if (lb != NULL)
1627 		free(lb, M_DEVBUF);
1628 	return (0);
1629 }
1630 
1631 static int
1632 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1633 {
1634 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1635 	struct lagg_port *lp_next;
1636 	int i = 0;
1637 
1638 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1639 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1640 		if (lp_next == lp)
1641 			continue;
1642 		if (i >= LAGG_MAX_PORTS)
1643 			return (EINVAL);
1644 		if (sc->sc_ifflags & IFF_DEBUG)
1645 			printf("%s: port %s at index %d\n",
1646 			    sc->sc_ifname, lp_next->lp_ifname, i);
1647 		lb->lb_ports[i++] = lp_next;
1648 	}
1649 
1650 	return (0);
1651 }
1652 
1653 static int
1654 lagg_lb_port_create(struct lagg_port *lp)
1655 {
1656 	struct lagg_softc *sc = lp->lp_softc;
1657 	return (lagg_lb_porttable(sc, NULL));
1658 }
1659 
1660 static void
1661 lagg_lb_port_destroy(struct lagg_port *lp)
1662 {
1663 	struct lagg_softc *sc = lp->lp_softc;
1664 	lagg_lb_porttable(sc, lp);
1665 }
1666 
1667 static int
1668 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1669 {
1670 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1671 	struct lagg_port *lp = NULL;
1672 	uint32_t p = 0;
1673 
1674 	if (m->m_flags & M_FLOWID)
1675 		p = m->m_pkthdr.flowid;
1676 	else
1677 		p = lagg_hashmbuf(m, lb->lb_key);
1678 	p %= sc->sc_count;
1679 	lp = lb->lb_ports[p];
1680 
1681 	/*
1682 	 * Check the port's link state. This will return the next active
1683 	 * port if the link is down or the port is NULL.
1684 	 */
1685 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1686 		m_freem(m);
1687 		return (ENOENT);
1688 	}
1689 
1690 	/* Send mbuf */
1691 	return (lagg_enqueue(lp->lp_ifp, m));
1692 }
1693 
1694 static struct mbuf *
1695 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1696 {
1697 	struct ifnet *ifp = sc->sc_ifp;
1698 
1699 	/* Just pass in the packet to our lagg device */
1700 	m->m_pkthdr.rcvif = ifp;
1701 
1702 	return (m);
1703 }
1704 
1705 /*
1706  * 802.3ad LACP
1707  */
1708 
1709 static int
1710 lagg_lacp_attach(struct lagg_softc *sc)
1711 {
1712 	struct lagg_port *lp;
1713 	int error;
1714 
1715 	sc->sc_detach = lagg_lacp_detach;
1716 	sc->sc_port_create = lacp_port_create;
1717 	sc->sc_port_destroy = lacp_port_destroy;
1718 	sc->sc_linkstate = lacp_linkstate;
1719 	sc->sc_start = lagg_lacp_start;
1720 	sc->sc_input = lagg_lacp_input;
1721 	sc->sc_init = lacp_init;
1722 	sc->sc_stop = lacp_stop;
1723 	sc->sc_lladdr = lagg_lacp_lladdr;
1724 	sc->sc_req = lacp_req;
1725 	sc->sc_portreq = lacp_portreq;
1726 
1727 	error = lacp_attach(sc);
1728 	if (error)
1729 		return (error);
1730 
1731 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1732 		lacp_port_create(lp);
1733 
1734 	return (error);
1735 }
1736 
1737 static int
1738 lagg_lacp_detach(struct lagg_softc *sc)
1739 {
1740 	struct lagg_port *lp;
1741 	int error;
1742 
1743 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1744 		lacp_port_destroy(lp);
1745 
1746 	/* unlocking is safe here */
1747 	LAGG_WUNLOCK(sc);
1748 	error = lacp_detach(sc);
1749 	LAGG_WLOCK(sc);
1750 
1751 	return (error);
1752 }
1753 
1754 static void
1755 lagg_lacp_lladdr(struct lagg_softc *sc)
1756 {
1757 	struct lagg_port *lp;
1758 
1759 	/* purge all the lacp ports */
1760 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1761 		lacp_port_destroy(lp);
1762 
1763 	/* add them back in */
1764 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1765 		lacp_port_create(lp);
1766 }
1767 
1768 static int
1769 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1770 {
1771 	struct lagg_port *lp;
1772 
1773 	lp = lacp_select_tx_port(sc, m);
1774 	if (lp == NULL) {
1775 		m_freem(m);
1776 		return (EBUSY);
1777 	}
1778 
1779 	/* Send mbuf */
1780 	return (lagg_enqueue(lp->lp_ifp, m));
1781 }
1782 
1783 static struct mbuf *
1784 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1785 {
1786 	struct ifnet *ifp = sc->sc_ifp;
1787 	struct ether_header *eh;
1788 	u_short etype;
1789 
1790 	eh = mtod(m, struct ether_header *);
1791 	etype = ntohs(eh->ether_type);
1792 
1793 	/* Tap off LACP control messages */
1794 	if (etype == ETHERTYPE_SLOW) {
1795 		m = lacp_input(lp, m);
1796 		if (m == NULL)
1797 			return (NULL);
1798 	}
1799 
1800 	/*
1801 	 * If the port is not collecting or not in the active aggregator then
1802 	 * free and return.
1803 	 */
1804 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
1805 		m_freem(m);
1806 		return (NULL);
1807 	}
1808 
1809 	m->m_pkthdr.rcvif = ifp;
1810 	return (m);
1811 }
1812