xref: /freebsd/sys/net/if_lagg.c (revision 2ca7a12a81a5376ef17d3a3ea70f7000c025c579)
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  *
6  * Permission to use, copy, modify, and distribute this software for any
7  * purpose with or without fee is hereby granted, provided that the above
8  * copyright notice and this permission notice appear in all copies.
9  *
10  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
11  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
13  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
14  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
15  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
16  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
17  */
18 
19 #include <sys/cdefs.h>
20 __FBSDID("$FreeBSD$");
21 
22 #include "opt_inet.h"
23 #include "opt_inet6.h"
24 
25 #include <sys/param.h>
26 #include <sys/kernel.h>
27 #include <sys/malloc.h>
28 #include <sys/mbuf.h>
29 #include <sys/queue.h>
30 #include <sys/socket.h>
31 #include <sys/sockio.h>
32 #include <sys/sysctl.h>
33 #include <sys/module.h>
34 #include <sys/priv.h>
35 #include <sys/systm.h>
36 #include <sys/proc.h>
37 #include <sys/hash.h>
38 #include <sys/lock.h>
39 #include <sys/rwlock.h>
40 #include <sys/taskqueue.h>
41 
42 #include <net/ethernet.h>
43 #include <net/if.h>
44 #include <net/if_clone.h>
45 #include <net/if_arp.h>
46 #include <net/if_dl.h>
47 #include <net/if_llc.h>
48 #include <net/if_media.h>
49 #include <net/if_types.h>
50 #include <net/if_var.h>
51 #include <net/bpf.h>
52 
53 #ifdef INET
54 #include <netinet/in.h>
55 #include <netinet/in_systm.h>
56 #include <netinet/if_ether.h>
57 #include <netinet/ip.h>
58 #endif
59 
60 #ifdef INET6
61 #include <netinet/ip6.h>
62 #endif
63 
64 #include <net/if_vlan_var.h>
65 #include <net/if_lagg.h>
66 #include <net/ieee8023ad_lacp.h>
67 
68 /* Special flags we should propagate to the lagg ports. */
69 static struct {
70 	int flag;
71 	int (*func)(struct ifnet *, int);
72 } lagg_pflags[] = {
73 	{IFF_PROMISC, ifpromisc},
74 	{IFF_ALLMULTI, if_allmulti},
75 	{0, NULL}
76 };
77 
78 SLIST_HEAD(__trhead, lagg_softc) lagg_list;	/* list of laggs */
79 static struct mtx	lagg_list_mtx;
80 eventhandler_tag	lagg_detach_cookie = NULL;
81 
82 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
83 static void	lagg_clone_destroy(struct ifnet *);
84 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
85 static void	lagg_capabilities(struct lagg_softc *);
86 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
87 static void	lagg_port_setlladdr(void *, int);
88 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
89 static int	lagg_port_destroy(struct lagg_port *, int);
90 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
91 static void	lagg_linkstate(struct lagg_softc *);
92 static void	lagg_port_state(struct ifnet *, int);
93 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
94 static int	lagg_port_output(struct ifnet *, struct mbuf *,
95 		    struct sockaddr *, struct rtentry *);
96 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
97 static int	lagg_port_checkstacking(struct lagg_softc *);
98 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
99 static void	lagg_init(void *);
100 static void	lagg_stop(struct lagg_softc *);
101 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
102 static int	lagg_ether_setmulti(struct lagg_softc *);
103 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
104 static	int	lagg_setflag(struct lagg_port *, int, int,
105 		    int (*func)(struct ifnet *, int));
106 static	int	lagg_setflags(struct lagg_port *, int status);
107 static void	lagg_start(struct ifnet *);
108 static int	lagg_media_change(struct ifnet *);
109 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
110 static struct lagg_port *lagg_link_active(struct lagg_softc *,
111 	    struct lagg_port *);
112 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
113 
114 IFC_SIMPLE_DECLARE(lagg, 0);
115 
116 /* Simple round robin */
117 static int	lagg_rr_attach(struct lagg_softc *);
118 static int	lagg_rr_detach(struct lagg_softc *);
119 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
120 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
121 		    struct mbuf *);
122 
123 /* Active failover */
124 static int	lagg_fail_attach(struct lagg_softc *);
125 static int	lagg_fail_detach(struct lagg_softc *);
126 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
127 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
128 		    struct mbuf *);
129 
130 /* Loadbalancing */
131 static int	lagg_lb_attach(struct lagg_softc *);
132 static int	lagg_lb_detach(struct lagg_softc *);
133 static int	lagg_lb_port_create(struct lagg_port *);
134 static void	lagg_lb_port_destroy(struct lagg_port *);
135 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
136 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
137 		    struct mbuf *);
138 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
139 
140 /* 802.3ad LACP */
141 static int	lagg_lacp_attach(struct lagg_softc *);
142 static int	lagg_lacp_detach(struct lagg_softc *);
143 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
144 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
145 		    struct mbuf *);
146 static void	lagg_lacp_lladdr(struct lagg_softc *);
147 
148 /* lagg protocol table */
149 static const struct {
150 	int			ti_proto;
151 	int			(*ti_attach)(struct lagg_softc *);
152 } lagg_protos[] = {
153 	{ LAGG_PROTO_ROUNDROBIN,	lagg_rr_attach },
154 	{ LAGG_PROTO_FAILOVER,		lagg_fail_attach },
155 	{ LAGG_PROTO_LOADBALANCE,	lagg_lb_attach },
156 	{ LAGG_PROTO_ETHERCHANNEL,	lagg_lb_attach },
157 	{ LAGG_PROTO_LACP,		lagg_lacp_attach },
158 	{ LAGG_PROTO_NONE,		NULL }
159 };
160 
161 static int
162 lagg_modevent(module_t mod, int type, void *data)
163 {
164 
165 	switch (type) {
166 	case MOD_LOAD:
167 		mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
168 		SLIST_INIT(&lagg_list);
169 		if_clone_attach(&lagg_cloner);
170 		lagg_input_p = lagg_input;
171 		lagg_linkstate_p = lagg_port_state;
172 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
173 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
174 		    EVENTHANDLER_PRI_ANY);
175 		break;
176 	case MOD_UNLOAD:
177 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
178 		    lagg_detach_cookie);
179 		if_clone_detach(&lagg_cloner);
180 		lagg_input_p = NULL;
181 		lagg_linkstate_p = NULL;
182 		mtx_destroy(&lagg_list_mtx);
183 		break;
184 	default:
185 		return (EOPNOTSUPP);
186 	}
187 	return (0);
188 }
189 
190 static moduledata_t lagg_mod = {
191 	"if_lagg",
192 	lagg_modevent,
193 	0
194 };
195 
196 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
197 
198 static int
199 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
200 {
201 	struct lagg_softc *sc;
202 	struct ifnet *ifp;
203 	int i, error = 0;
204 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
205 
206 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
207 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
208 	if (ifp == NULL) {
209 		free(sc, M_DEVBUF);
210 		return (ENOSPC);
211 	}
212 
213 	sc->sc_proto = LAGG_PROTO_NONE;
214 	for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
215 		if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
216 			sc->sc_proto = lagg_protos[i].ti_proto;
217 			if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
218 				if_free_type(ifp, IFT_ETHER);
219 				free(sc, M_DEVBUF);
220 				return (error);
221 			}
222 			break;
223 		}
224 	}
225 	LAGG_LOCK_INIT(sc);
226 	SLIST_INIT(&sc->sc_ports);
227 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
228 
229 	/* Initialise pseudo media types */
230 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
231 	    lagg_media_status);
232 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
233 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
234 
235 	if_initname(ifp, ifc->ifc_name, unit);
236 	ifp->if_type = IFT_ETHER;
237 	ifp->if_softc = sc;
238 	ifp->if_start = lagg_start;
239 	ifp->if_init = lagg_init;
240 	ifp->if_ioctl = lagg_ioctl;
241 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
242 
243 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
244 	ifp->if_snd.ifq_drv_maxlen = ifqmaxlen;
245 	IFQ_SET_READY(&ifp->if_snd);
246 
247 	/*
248 	 * Attach as an ordinary ethernet device, childs will be attached
249 	 * as special device IFT_IEEE8023ADLAG.
250 	 */
251 	ether_ifattach(ifp, eaddr);
252 
253 	/* Insert into the global list of laggs */
254 	mtx_lock(&lagg_list_mtx);
255 	SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
256 	mtx_unlock(&lagg_list_mtx);
257 
258 	return (0);
259 }
260 
261 static void
262 lagg_clone_destroy(struct ifnet *ifp)
263 {
264 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
265 	struct lagg_port *lp;
266 
267 	LAGG_WLOCK(sc);
268 
269 	lagg_stop(sc);
270 	ifp->if_flags &= ~IFF_UP;
271 
272 	/* Shutdown and remove lagg ports */
273 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
274 		lagg_port_destroy(lp, 1);
275 	/* Unhook the aggregation protocol */
276 	(*sc->sc_detach)(sc);
277 
278 	LAGG_WUNLOCK(sc);
279 
280 	ifmedia_removeall(&sc->sc_media);
281 	ether_ifdetach(ifp);
282 	if_free_type(ifp, IFT_ETHER);
283 
284 	mtx_lock(&lagg_list_mtx);
285 	SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
286 	mtx_unlock(&lagg_list_mtx);
287 
288 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
289 	LAGG_LOCK_DESTROY(sc);
290 	free(sc, M_DEVBUF);
291 }
292 
293 static void
294 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
295 {
296 	struct ifnet *ifp = sc->sc_ifp;
297 
298 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
299 		return;
300 
301 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
302 	/* Let the protocol know the MAC has changed */
303 	if (sc->sc_lladdr != NULL)
304 		(*sc->sc_lladdr)(sc);
305 }
306 
307 static void
308 lagg_capabilities(struct lagg_softc *sc)
309 {
310 	struct lagg_port *lp;
311 	int cap = ~0, ena = ~0;
312 
313 	LAGG_WLOCK_ASSERT(sc);
314 
315 	/* Get capabilities from the lagg ports */
316 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
317 		cap &= lp->lp_ifp->if_capabilities;
318 		ena &= lp->lp_ifp->if_capenable;
319 	}
320 	cap = (cap == ~0 ? 0 : cap);
321 	ena = (ena == ~0 ? 0 : ena);
322 
323 	if (sc->sc_ifp->if_capabilities != cap ||
324 	    sc->sc_ifp->if_capenable != ena) {
325 		sc->sc_ifp->if_capabilities = cap;
326 		sc->sc_ifp->if_capenable = ena;
327 		getmicrotime(&sc->sc_ifp->if_lastchange);
328 
329 		if (sc->sc_ifflags & IFF_DEBUG)
330 			if_printf(sc->sc_ifp,
331 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
332 	}
333 }
334 
335 static void
336 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
337 {
338 	struct lagg_softc *sc = lp->lp_softc;
339 	struct ifnet *ifp = lp->lp_ifp;
340 	struct lagg_llq *llq;
341 	int pending = 0;
342 
343 	LAGG_WLOCK_ASSERT(sc);
344 
345 	if (lp->lp_detaching ||
346 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
347 		return;
348 
349 	/* Check to make sure its not already queued to be changed */
350 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
351 		if (llq->llq_ifp == ifp) {
352 			pending = 1;
353 			break;
354 		}
355 	}
356 
357 	if (!pending) {
358 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
359 		if (llq == NULL)	/* XXX what to do */
360 			return;
361 	}
362 
363 	/* Update the lladdr even if pending, it may have changed */
364 	llq->llq_ifp = ifp;
365 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
366 
367 	if (!pending)
368 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
369 
370 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
371 }
372 
373 /*
374  * Set the interface MAC address from a taskqueue to avoid a LOR.
375  */
376 static void
377 lagg_port_setlladdr(void *arg, int pending)
378 {
379 	struct lagg_softc *sc = (struct lagg_softc *)arg;
380 	struct lagg_llq *llq, *head;
381 	struct ifnet *ifp;
382 	int error;
383 
384 	/* Grab a local reference of the queue and remove it from the softc */
385 	LAGG_WLOCK(sc);
386 	head = SLIST_FIRST(&sc->sc_llq_head);
387 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
388 	LAGG_WUNLOCK(sc);
389 
390 	/*
391 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
392 	 * unlocked as we have the only reference to it.
393 	 */
394 	for (llq = head; llq != NULL; llq = head) {
395 		ifp = llq->llq_ifp;
396 
397 		/* Set the link layer address */
398 		error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
399 		if (error)
400 			printf("%s: setlladdr failed on %s\n", __func__,
401 			    ifp->if_xname);
402 
403 		head = SLIST_NEXT(llq, llq_entries);
404 		free(llq, M_DEVBUF);
405 	}
406 }
407 
408 static int
409 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
410 {
411 	struct lagg_softc *sc_ptr;
412 	struct lagg_port *lp;
413 	int error = 0;
414 
415 	LAGG_WLOCK_ASSERT(sc);
416 
417 	/* Limit the maximal number of lagg ports */
418 	if (sc->sc_count >= LAGG_MAX_PORTS)
419 		return (ENOSPC);
420 
421 	/* New lagg port has to be in an idle state */
422 	if (ifp->if_drv_flags & IFF_DRV_OACTIVE)
423 		return (EBUSY);
424 
425 	/* Check if port has already been associated to a lagg */
426 	if (ifp->if_lagg != NULL)
427 		return (EBUSY);
428 
429 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
430 	if (ifp->if_type != IFT_ETHER)
431 		return (EPROTONOSUPPORT);
432 
433 	/* Allow the first Ethernet member to define the MTU */
434 	if (SLIST_EMPTY(&sc->sc_ports))
435 		sc->sc_ifp->if_mtu = ifp->if_mtu;
436 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
437 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
438 		    ifp->if_xname);
439 		return (EINVAL);
440 	}
441 
442 	if ((lp = malloc(sizeof(struct lagg_port),
443 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
444 		return (ENOMEM);
445 
446 	/* Check if port is a stacked lagg */
447 	mtx_lock(&lagg_list_mtx);
448 	SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
449 		if (ifp == sc_ptr->sc_ifp) {
450 			mtx_unlock(&lagg_list_mtx);
451 			free(lp, M_DEVBUF);
452 			return (EINVAL);
453 			/* XXX disable stacking for the moment, its untested
454 			lp->lp_flags |= LAGG_PORT_STACK;
455 			if (lagg_port_checkstacking(sc_ptr) >=
456 			    LAGG_MAX_STACKING) {
457 				mtx_unlock(&lagg_list_mtx);
458 				free(lp, M_DEVBUF);
459 				return (E2BIG);
460 			}
461 			*/
462 		}
463 	}
464 	mtx_unlock(&lagg_list_mtx);
465 
466 	/* Change the interface type */
467 	lp->lp_iftype = ifp->if_type;
468 	ifp->if_type = IFT_IEEE8023ADLAG;
469 	ifp->if_lagg = lp;
470 	lp->lp_ioctl = ifp->if_ioctl;
471 	ifp->if_ioctl = lagg_port_ioctl;
472 	lp->lp_output = ifp->if_output;
473 	ifp->if_output = lagg_port_output;
474 
475 	lp->lp_ifp = ifp;
476 	lp->lp_softc = sc;
477 
478 	/* Save port link layer address */
479 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
480 
481 	if (SLIST_EMPTY(&sc->sc_ports)) {
482 		sc->sc_primary = lp;
483 		lagg_lladdr(sc, IF_LLADDR(ifp));
484 	} else {
485 		/* Update link layer address for this port */
486 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
487 	}
488 
489 	/* Insert into the list of ports */
490 	SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
491 	sc->sc_count++;
492 
493 	/* Update lagg capabilities */
494 	lagg_capabilities(sc);
495 	lagg_linkstate(sc);
496 
497 	/* Add multicast addresses and interface flags to this port */
498 	lagg_ether_cmdmulti(lp, 1);
499 	lagg_setflags(lp, 1);
500 
501 	if (sc->sc_port_create != NULL)
502 		error = (*sc->sc_port_create)(lp);
503 	if (error) {
504 		/* remove the port again, without calling sc_port_destroy */
505 		lagg_port_destroy(lp, 0);
506 		return (error);
507 	}
508 
509 	return (error);
510 }
511 
512 static int
513 lagg_port_checkstacking(struct lagg_softc *sc)
514 {
515 	struct lagg_softc *sc_ptr;
516 	struct lagg_port *lp;
517 	int m = 0;
518 
519 	LAGG_WLOCK_ASSERT(sc);
520 
521 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
522 		if (lp->lp_flags & LAGG_PORT_STACK) {
523 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
524 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
525 		}
526 	}
527 
528 	return (m + 1);
529 }
530 
531 static int
532 lagg_port_destroy(struct lagg_port *lp, int runpd)
533 {
534 	struct lagg_softc *sc = lp->lp_softc;
535 	struct lagg_port *lp_ptr;
536 	struct lagg_llq *llq;
537 	struct ifnet *ifp = lp->lp_ifp;
538 
539 	LAGG_WLOCK_ASSERT(sc);
540 
541 	if (runpd && sc->sc_port_destroy != NULL)
542 		(*sc->sc_port_destroy)(lp);
543 
544 	/*
545 	 * Remove multicast addresses and interface flags from this port and
546 	 * reset the MAC address, skip if the interface is being detached.
547 	 */
548 	if (!lp->lp_detaching) {
549 		lagg_ether_cmdmulti(lp, 0);
550 		lagg_setflags(lp, 0);
551 		lagg_port_lladdr(lp, lp->lp_lladdr);
552 	}
553 
554 	/* Restore interface */
555 	ifp->if_type = lp->lp_iftype;
556 	ifp->if_ioctl = lp->lp_ioctl;
557 	ifp->if_output = lp->lp_output;
558 	ifp->if_lagg = NULL;
559 
560 	/* Finally, remove the port from the lagg */
561 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
562 	sc->sc_count--;
563 
564 	/* Update the primary interface */
565 	if (lp == sc->sc_primary) {
566 		uint8_t lladdr[ETHER_ADDR_LEN];
567 
568 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
569 			bzero(&lladdr, ETHER_ADDR_LEN);
570 		} else {
571 			bcopy(lp_ptr->lp_lladdr,
572 			    lladdr, ETHER_ADDR_LEN);
573 		}
574 		lagg_lladdr(sc, lladdr);
575 		sc->sc_primary = lp_ptr;
576 
577 		/* Update link layer address for each port */
578 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
579 			lagg_port_lladdr(lp_ptr, lladdr);
580 	}
581 
582 	/* Remove any pending lladdr changes from the queue */
583 	if (lp->lp_detaching) {
584 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
585 			if (llq->llq_ifp == ifp) {
586 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
587 				    llq_entries);
588 				free(llq, M_DEVBUF);
589 				break;	/* Only appears once */
590 			}
591 		}
592 	}
593 
594 	if (lp->lp_ifflags)
595 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
596 
597 	free(lp, M_DEVBUF);
598 
599 	/* Update lagg capabilities */
600 	lagg_capabilities(sc);
601 	lagg_linkstate(sc);
602 
603 	return (0);
604 }
605 
606 static int
607 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
608 {
609 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
610 	struct lagg_softc *sc;
611 	struct lagg_port *lp = NULL;
612 	int error = 0;
613 
614 	/* Should be checked by the caller */
615 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
616 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
617 		goto fallback;
618 
619 	switch (cmd) {
620 	case SIOCGLAGGPORT:
621 		if (rp->rp_portname[0] == '\0' ||
622 		    ifunit(rp->rp_portname) != ifp) {
623 			error = EINVAL;
624 			break;
625 		}
626 
627 		LAGG_RLOCK(sc);
628 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
629 			error = ENOENT;
630 			LAGG_RUNLOCK(sc);
631 			break;
632 		}
633 
634 		lagg_port2req(lp, rp);
635 		LAGG_RUNLOCK(sc);
636 		break;
637 
638 	case SIOCSIFCAP:
639 		if (lp->lp_ioctl == NULL) {
640 			error = EINVAL;
641 			break;
642 		}
643 		error = (*lp->lp_ioctl)(ifp, cmd, data);
644 		if (error)
645 			break;
646 
647 		/* Update lagg interface capabilities */
648 		LAGG_WLOCK(sc);
649 		lagg_capabilities(sc);
650 		LAGG_WUNLOCK(sc);
651 		break;
652 
653 	case SIOCSIFMTU:
654 		/* Do not allow the MTU to be changed once joined */
655 		error = EINVAL;
656 		break;
657 
658 	default:
659 		goto fallback;
660 	}
661 
662 	return (error);
663 
664 fallback:
665 	if (lp->lp_ioctl != NULL)
666 		return ((*lp->lp_ioctl)(ifp, cmd, data));
667 
668 	return (EINVAL);
669 }
670 
671 static int
672 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
673 	struct sockaddr *dst, struct rtentry *rt0)
674 {
675 	struct lagg_port *lp = ifp->if_lagg;
676 	struct ether_header *eh;
677 	short type = 0;
678 
679 	switch (dst->sa_family) {
680 		case pseudo_AF_HDRCMPLT:
681 		case AF_UNSPEC:
682 			eh = (struct ether_header *)dst->sa_data;
683 			type = eh->ether_type;
684 			break;
685 	}
686 
687 	/*
688 	 * Only allow ethernet types required to initiate or maintain the link,
689 	 * aggregated frames take a different path.
690 	 */
691 	switch (ntohs(type)) {
692 		case ETHERTYPE_PAE:	/* EAPOL PAE/802.1x */
693 			return ((*lp->lp_output)(ifp, m, dst, rt0));
694 	}
695 
696 	/* drop any other frames */
697 	m_freem(m);
698 	return (EBUSY);
699 }
700 
701 static void
702 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
703 {
704 	struct lagg_port *lp;
705 	struct lagg_softc *sc;
706 
707 	if ((lp = ifp->if_lagg) == NULL)
708 		return;
709 
710 	sc = lp->lp_softc;
711 
712 	LAGG_WLOCK(sc);
713 	lp->lp_detaching = 1;
714 	lagg_port_destroy(lp, 1);
715 	LAGG_WUNLOCK(sc);
716 }
717 
718 static void
719 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
720 {
721 	struct lagg_softc *sc = lp->lp_softc;
722 
723 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
724 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
725 	rp->rp_prio = lp->lp_prio;
726 	rp->rp_flags = lp->lp_flags;
727 	if (sc->sc_portreq != NULL)
728 		(*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
729 
730 	/* Add protocol specific flags */
731 	switch (sc->sc_proto) {
732 		case LAGG_PROTO_FAILOVER:
733 			if (lp == sc->sc_primary)
734 				rp->rp_flags |= LAGG_PORT_MASTER;
735 			if (lp == lagg_link_active(sc, sc->sc_primary))
736 				rp->rp_flags |= LAGG_PORT_ACTIVE;
737 			break;
738 
739 		case LAGG_PROTO_ROUNDROBIN:
740 		case LAGG_PROTO_LOADBALANCE:
741 		case LAGG_PROTO_ETHERCHANNEL:
742 			if (LAGG_PORTACTIVE(lp))
743 				rp->rp_flags |= LAGG_PORT_ACTIVE;
744 			break;
745 
746 		case LAGG_PROTO_LACP:
747 			/* LACP has a different definition of active */
748 			if (lacp_port_isactive(lp))
749 				rp->rp_flags |= LAGG_PORT_ACTIVE;
750 			break;
751 	}
752 
753 }
754 
755 static void
756 lagg_init(void *xsc)
757 {
758 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
759 	struct lagg_port *lp;
760 	struct ifnet *ifp = sc->sc_ifp;
761 
762 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
763 		return;
764 
765 	LAGG_WLOCK(sc);
766 
767 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
768 	/* Update the port lladdrs */
769 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
770 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
771 
772 	if (sc->sc_init != NULL)
773 		(*sc->sc_init)(sc);
774 
775 	LAGG_WUNLOCK(sc);
776 }
777 
778 static void
779 lagg_stop(struct lagg_softc *sc)
780 {
781 	struct ifnet *ifp = sc->sc_ifp;
782 
783 	LAGG_WLOCK_ASSERT(sc);
784 
785 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
786 		return;
787 
788 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
789 
790 	if (sc->sc_stop != NULL)
791 		(*sc->sc_stop)(sc);
792 }
793 
794 static int
795 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
796 {
797 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
798 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
799 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
800 	struct ifreq *ifr = (struct ifreq *)data;
801 	struct lagg_port *lp;
802 	struct ifnet *tpif;
803 	struct thread *td = curthread;
804 	char *buf, *outbuf;
805 	int count, buflen, len, error = 0;
806 
807 	bzero(&rpbuf, sizeof(rpbuf));
808 
809 	switch (cmd) {
810 	case SIOCGLAGG:
811 		LAGG_RLOCK(sc);
812 		count = 0;
813 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
814 			count++;
815 		buflen = count * sizeof(struct lagg_reqport);
816 		LAGG_RUNLOCK(sc);
817 
818 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
819 
820 		LAGG_RLOCK(sc);
821 		ra->ra_proto = sc->sc_proto;
822 		if (sc->sc_req != NULL)
823 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
824 
825 		count = 0;
826 		buf = outbuf;
827 		len = min(ra->ra_size, buflen);
828 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
829 			if (len < sizeof(rpbuf))
830 				break;
831 
832 			lagg_port2req(lp, &rpbuf);
833 			memcpy(buf, &rpbuf, sizeof(rpbuf));
834 			count++;
835 			buf += sizeof(rpbuf);
836 			len -= sizeof(rpbuf);
837 		}
838 		LAGG_RUNLOCK(sc);
839 		ra->ra_ports = count;
840 		ra->ra_size = count * sizeof(rpbuf);
841 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
842 		free(outbuf, M_TEMP);
843 		break;
844 	case SIOCSLAGG:
845 		error = priv_check(td, PRIV_NET_LAGG);
846 		if (error)
847 			break;
848 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
849 			error = EPROTONOSUPPORT;
850 			break;
851 		}
852 		if (sc->sc_proto != LAGG_PROTO_NONE) {
853 			LAGG_WLOCK(sc);
854 			error = sc->sc_detach(sc);
855 			/* Reset protocol and pointers */
856 			sc->sc_proto = LAGG_PROTO_NONE;
857 			sc->sc_detach = NULL;
858 			sc->sc_start = NULL;
859 			sc->sc_input = NULL;
860 			sc->sc_port_create = NULL;
861 			sc->sc_port_destroy = NULL;
862 			sc->sc_linkstate = NULL;
863 			sc->sc_init = NULL;
864 			sc->sc_stop = NULL;
865 			sc->sc_lladdr = NULL;
866 			sc->sc_req = NULL;
867 			sc->sc_portreq = NULL;
868 			LAGG_WUNLOCK(sc);
869 		}
870 		if (error != 0)
871 			break;
872 		for (int i = 0; i < (sizeof(lagg_protos) /
873 		    sizeof(lagg_protos[0])); i++) {
874 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
875 				if (sc->sc_ifflags & IFF_DEBUG)
876 					printf("%s: using proto %u\n",
877 					    sc->sc_ifname,
878 					    lagg_protos[i].ti_proto);
879 				LAGG_WLOCK(sc);
880 				sc->sc_proto = lagg_protos[i].ti_proto;
881 				if (sc->sc_proto != LAGG_PROTO_NONE)
882 					error = lagg_protos[i].ti_attach(sc);
883 				LAGG_WUNLOCK(sc);
884 				return (error);
885 			}
886 		}
887 		error = EPROTONOSUPPORT;
888 		break;
889 	case SIOCGLAGGPORT:
890 		if (rp->rp_portname[0] == '\0' ||
891 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
892 			error = EINVAL;
893 			break;
894 		}
895 
896 		LAGG_RLOCK(sc);
897 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
898 		    lp->lp_softc != sc) {
899 			error = ENOENT;
900 			LAGG_RUNLOCK(sc);
901 			break;
902 		}
903 
904 		lagg_port2req(lp, rp);
905 		LAGG_RUNLOCK(sc);
906 		break;
907 	case SIOCSLAGGPORT:
908 		error = priv_check(td, PRIV_NET_LAGG);
909 		if (error)
910 			break;
911 		if (rp->rp_portname[0] == '\0' ||
912 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
913 			error = EINVAL;
914 			break;
915 		}
916 		LAGG_WLOCK(sc);
917 		error = lagg_port_create(sc, tpif);
918 		LAGG_WUNLOCK(sc);
919 		break;
920 	case SIOCSLAGGDELPORT:
921 		error = priv_check(td, PRIV_NET_LAGG);
922 		if (error)
923 			break;
924 		if (rp->rp_portname[0] == '\0' ||
925 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
926 			error = EINVAL;
927 			break;
928 		}
929 
930 		LAGG_WLOCK(sc);
931 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
932 		    lp->lp_softc != sc) {
933 			error = ENOENT;
934 			LAGG_WUNLOCK(sc);
935 			break;
936 		}
937 
938 		error = lagg_port_destroy(lp, 1);
939 		LAGG_WUNLOCK(sc);
940 		break;
941 	case SIOCSIFFLAGS:
942 		/* Set flags on ports too */
943 		LAGG_WLOCK(sc);
944 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
945 			lagg_setflags(lp, 1);
946 		}
947 		LAGG_WUNLOCK(sc);
948 
949 		if (!(ifp->if_flags & IFF_UP) &&
950 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
951 			/*
952 			 * If interface is marked down and it is running,
953 			 * then stop and disable it.
954 			 */
955 			LAGG_WLOCK(sc);
956 			lagg_stop(sc);
957 			LAGG_WUNLOCK(sc);
958 		} else if ((ifp->if_flags & IFF_UP) &&
959 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
960 			/*
961 			 * If interface is marked up and it is stopped, then
962 			 * start it.
963 			 */
964 			(*ifp->if_init)(sc);
965 		}
966 		break;
967 	case SIOCADDMULTI:
968 	case SIOCDELMULTI:
969 		LAGG_WLOCK(sc);
970 		error = lagg_ether_setmulti(sc);
971 		LAGG_WUNLOCK(sc);
972 		break;
973 	case SIOCSIFMEDIA:
974 	case SIOCGIFMEDIA:
975 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
976 		break;
977 
978 	case SIOCSIFCAP:
979 	case SIOCSIFMTU:
980 		/* Do not allow the MTU or caps to be directly changed */
981 		error = EINVAL;
982 		break;
983 
984 	default:
985 		error = ether_ioctl(ifp, cmd, data);
986 		break;
987 	}
988 	return (error);
989 }
990 
991 static int
992 lagg_ether_setmulti(struct lagg_softc *sc)
993 {
994 	struct lagg_port *lp;
995 
996 	LAGG_WLOCK_ASSERT(sc);
997 
998 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
999 		/* First, remove any existing filter entries. */
1000 		lagg_ether_cmdmulti(lp, 0);
1001 		/* copy all addresses from the lagg interface to the port */
1002 		lagg_ether_cmdmulti(lp, 1);
1003 	}
1004 	return (0);
1005 }
1006 
1007 static int
1008 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1009 {
1010 	struct lagg_softc *sc = lp->lp_softc;
1011 	struct ifnet *ifp = lp->lp_ifp;
1012 	struct ifnet *scifp = sc->sc_ifp;
1013 	struct lagg_mc *mc;
1014 	struct ifmultiaddr *ifma, *rifma = NULL;
1015 	struct sockaddr_dl sdl;
1016 	int error;
1017 
1018 	LAGG_WLOCK_ASSERT(sc);
1019 
1020 	bzero((char *)&sdl, sizeof(sdl));
1021 	sdl.sdl_len = sizeof(sdl);
1022 	sdl.sdl_family = AF_LINK;
1023 	sdl.sdl_type = IFT_ETHER;
1024 	sdl.sdl_alen = ETHER_ADDR_LEN;
1025 	sdl.sdl_index = ifp->if_index;
1026 
1027 	if (set) {
1028 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1029 			if (ifma->ifma_addr->sa_family != AF_LINK)
1030 				continue;
1031 			bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1032 			    LLADDR(&sdl), ETHER_ADDR_LEN);
1033 
1034 			error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1035 			if (error)
1036 				return (error);
1037 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1038 			if (mc == NULL)
1039 				return (ENOMEM);
1040 			mc->mc_ifma = rifma;
1041 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1042 		}
1043 	} else {
1044 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1045 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1046 			if_delmulti_ifma(mc->mc_ifma);
1047 			free(mc, M_DEVBUF);
1048 		}
1049 	}
1050 	return (0);
1051 }
1052 
1053 /* Handle a ref counted flag that should be set on the lagg port as well */
1054 static int
1055 lagg_setflag(struct lagg_port *lp, int flag, int status,
1056 	     int (*func)(struct ifnet *, int))
1057 {
1058 	struct lagg_softc *sc = lp->lp_softc;
1059 	struct ifnet *scifp = sc->sc_ifp;
1060 	struct ifnet *ifp = lp->lp_ifp;
1061 	int error;
1062 
1063 	LAGG_WLOCK_ASSERT(sc);
1064 
1065 	status = status ? (scifp->if_flags & flag) : 0;
1066 	/* Now "status" contains the flag value or 0 */
1067 
1068 	/*
1069 	 * See if recorded ports status is different from what
1070 	 * we want it to be.  If it is, flip it.  We record ports
1071 	 * status in lp_ifflags so that we won't clear ports flag
1072 	 * we haven't set.  In fact, we don't clear or set ports
1073 	 * flags directly, but get or release references to them.
1074 	 * That's why we can be sure that recorded flags still are
1075 	 * in accord with actual ports flags.
1076 	 */
1077 	if (status != (lp->lp_ifflags & flag)) {
1078 		error = (*func)(ifp, status);
1079 		if (error)
1080 			return (error);
1081 		lp->lp_ifflags &= ~flag;
1082 		lp->lp_ifflags |= status;
1083 	}
1084 	return (0);
1085 }
1086 
1087 /*
1088  * Handle IFF_* flags that require certain changes on the lagg port
1089  * if "status" is true, update ports flags respective to the lagg
1090  * if "status" is false, forcedly clear the flags set on port.
1091  */
1092 static int
1093 lagg_setflags(struct lagg_port *lp, int status)
1094 {
1095 	int error, i;
1096 
1097 	for (i = 0; lagg_pflags[i].flag; i++) {
1098 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1099 		    status, lagg_pflags[i].func);
1100 		if (error)
1101 			return (error);
1102 	}
1103 	return (0);
1104 }
1105 
1106 static void
1107 lagg_start(struct ifnet *ifp)
1108 {
1109 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1110 	struct mbuf *m;
1111 	int error = 0;
1112 
1113 	LAGG_RLOCK(sc);
1114 	for (;; error = 0) {
1115 		IFQ_DEQUEUE(&ifp->if_snd, m);
1116 		if (m == NULL)
1117 			break;
1118 
1119 		ETHER_BPF_MTAP(ifp, m);
1120 
1121 		if (sc->sc_proto != LAGG_PROTO_NONE)
1122 			error = (*sc->sc_start)(sc, m);
1123 		else
1124 			m_freem(m);
1125 
1126 		if (error == 0)
1127 			ifp->if_opackets++;
1128 		else
1129 			ifp->if_oerrors++;
1130 	}
1131 	LAGG_RUNLOCK(sc);
1132 
1133 	return;
1134 }
1135 
1136 static struct mbuf *
1137 lagg_input(struct ifnet *ifp, struct mbuf *m)
1138 {
1139 	struct lagg_port *lp = ifp->if_lagg;
1140 	struct lagg_softc *sc = lp->lp_softc;
1141 	struct ifnet *scifp = sc->sc_ifp;
1142 
1143 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1144 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1145 	    sc->sc_proto == LAGG_PROTO_NONE) {
1146 		m_freem(m);
1147 		return (NULL);
1148 	}
1149 
1150 	LAGG_RLOCK(sc);
1151 	ETHER_BPF_MTAP(scifp, m);
1152 
1153 	m = (*sc->sc_input)(sc, lp, m);
1154 
1155 	if (m != NULL) {
1156 		scifp->if_ipackets++;
1157 		scifp->if_ibytes += m->m_pkthdr.len;
1158 
1159 		if (scifp->if_flags & IFF_MONITOR) {
1160 			m_freem(m);
1161 			m = NULL;
1162 		}
1163 	}
1164 
1165 	LAGG_RUNLOCK(sc);
1166 	return (m);
1167 }
1168 
1169 static int
1170 lagg_media_change(struct ifnet *ifp)
1171 {
1172 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1173 
1174 	if (sc->sc_ifflags & IFF_DEBUG)
1175 		printf("%s\n", __func__);
1176 
1177 	/* Ignore */
1178 	return (0);
1179 }
1180 
1181 static void
1182 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1183 {
1184 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1185 	struct lagg_port *lp;
1186 
1187 	imr->ifm_status = IFM_AVALID;
1188 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1189 
1190 	LAGG_RLOCK(sc);
1191 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1192 		if (LAGG_PORTACTIVE(lp))
1193 			imr->ifm_status |= IFM_ACTIVE;
1194 	}
1195 	LAGG_RUNLOCK(sc);
1196 }
1197 
1198 static void
1199 lagg_linkstate(struct lagg_softc *sc)
1200 {
1201 	struct lagg_port *lp;
1202 	int new_link = LINK_STATE_DOWN;
1203 
1204 	/* Our link is considered up if at least one of our ports is active */
1205 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1206 		if (lp->lp_link_state == LINK_STATE_UP) {
1207 			new_link = LINK_STATE_UP;
1208 			break;
1209 		}
1210 	}
1211 	if_link_state_change(sc->sc_ifp, new_link);
1212 }
1213 
1214 static void
1215 lagg_port_state(struct ifnet *ifp, int state)
1216 {
1217 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1218 	struct lagg_softc *sc = NULL;
1219 
1220 	if (lp != NULL)
1221 		sc = lp->lp_softc;
1222 	if (sc == NULL)
1223 		return;
1224 
1225 	LAGG_WLOCK(sc);
1226 	lagg_linkstate(sc);
1227 	if (sc->sc_linkstate != NULL)
1228 		(*sc->sc_linkstate)(lp);
1229 	LAGG_WUNLOCK(sc);
1230 }
1231 
1232 struct lagg_port *
1233 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1234 {
1235 	struct lagg_port *lp_next, *rval = NULL;
1236 	// int new_link = LINK_STATE_DOWN;
1237 
1238 	LAGG_RLOCK_ASSERT(sc);
1239 	/*
1240 	 * Search a port which reports an active link state.
1241 	 */
1242 
1243 	if (lp == NULL)
1244 		goto search;
1245 	if (LAGG_PORTACTIVE(lp)) {
1246 		rval = lp;
1247 		goto found;
1248 	}
1249 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1250 	    LAGG_PORTACTIVE(lp_next)) {
1251 		rval = lp_next;
1252 		goto found;
1253 	}
1254 
1255 search:
1256 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1257 		if (LAGG_PORTACTIVE(lp_next)) {
1258 			rval = lp_next;
1259 			goto found;
1260 		}
1261 	}
1262 
1263 found:
1264 	if (rval != NULL) {
1265 		/*
1266 		 * The IEEE 802.1D standard assumes that a lagg with
1267 		 * multiple ports is always full duplex. This is valid
1268 		 * for load sharing laggs and if at least two links
1269 		 * are active. Unfortunately, checking the latter would
1270 		 * be too expensive at this point.
1271 		 XXX
1272 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1273 		    (sc->sc_count > 1))
1274 			new_link = LINK_STATE_FULL_DUPLEX;
1275 		else
1276 			new_link = rval->lp_link_state;
1277 		 */
1278 	}
1279 
1280 	return (rval);
1281 }
1282 
1283 static const void *
1284 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1285 {
1286 	if (m->m_pkthdr.len < (off + len)) {
1287 		return (NULL);
1288 	} else if (m->m_len < (off + len)) {
1289 		m_copydata(m, off, len, buf);
1290 		return (buf);
1291 	}
1292 	return (mtod(m, char *) + off);
1293 }
1294 
1295 uint32_t
1296 lagg_hashmbuf(struct mbuf *m, uint32_t key)
1297 {
1298 	uint16_t etype;
1299 	uint32_t p = 0;
1300 	int off;
1301 	struct ether_header *eh;
1302 	struct ether_vlan_header vlanbuf;
1303 	const struct ether_vlan_header *vlan;
1304 #ifdef INET
1305 	const struct ip *ip;
1306 	struct ip ipbuf;
1307 #endif
1308 #ifdef INET6
1309 	const struct ip6_hdr *ip6;
1310 	struct ip6_hdr ip6buf;
1311 	uint32_t flow;
1312 #endif
1313 
1314 	off = sizeof(*eh);
1315 	if (m->m_len < off)
1316 		goto out;
1317 	eh = mtod(m, struct ether_header *);
1318 	etype = ntohs(eh->ether_type);
1319 	p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, key);
1320 	p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1321 
1322 	/* Special handling for encapsulating VLAN frames */
1323 	if (m->m_flags & M_VLANTAG) {
1324 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1325 		    sizeof(m->m_pkthdr.ether_vtag), p);
1326 	} else if (etype == ETHERTYPE_VLAN) {
1327 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &vlanbuf);
1328 		if (vlan == NULL)
1329 			goto out;
1330 
1331 		p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1332 		etype = ntohs(vlan->evl_proto);
1333 		off += sizeof(*vlan) - sizeof(*eh);
1334 	}
1335 
1336 	switch (etype) {
1337 #ifdef INET
1338 	case ETHERTYPE_IP:
1339 		ip = lagg_gethdr(m, off, sizeof(*ip), &ipbuf);
1340 		if (ip == NULL)
1341 			goto out;
1342 
1343 		p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1344 		p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1345 		break;
1346 #endif
1347 #ifdef INET6
1348 	case ETHERTYPE_IPV6:
1349 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &ip6buf);
1350 		if (ip6 == NULL)
1351 			goto out;
1352 
1353 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1354 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1355 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1356 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1357 		break;
1358 #endif
1359 	}
1360 out:
1361 	return (p);
1362 }
1363 
1364 int
1365 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1366 {
1367 	int error = 0;
1368 
1369 	IFQ_HANDOFF(ifp, m, error);
1370 	if (error)
1371 		ifp->if_oerrors++;
1372 	return (error);
1373 }
1374 
1375 /*
1376  * Simple round robin aggregation
1377  */
1378 
1379 static int
1380 lagg_rr_attach(struct lagg_softc *sc)
1381 {
1382 	sc->sc_detach = lagg_rr_detach;
1383 	sc->sc_start = lagg_rr_start;
1384 	sc->sc_input = lagg_rr_input;
1385 	sc->sc_port_create = NULL;
1386 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1387 	sc->sc_seq = 0;
1388 
1389 	return (0);
1390 }
1391 
1392 static int
1393 lagg_rr_detach(struct lagg_softc *sc)
1394 {
1395 	return (0);
1396 }
1397 
1398 static int
1399 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1400 {
1401 	struct lagg_port *lp;
1402 	uint32_t p;
1403 
1404 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1405 	p %= sc->sc_count;
1406 	lp = SLIST_FIRST(&sc->sc_ports);
1407 	while (p--)
1408 		lp = SLIST_NEXT(lp, lp_entries);
1409 
1410 	/*
1411 	 * Check the port's link state. This will return the next active
1412 	 * port if the link is down or the port is NULL.
1413 	 */
1414 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1415 		m_freem(m);
1416 		return (ENOENT);
1417 	}
1418 
1419 	/* Send mbuf */
1420 	return (lagg_enqueue(lp->lp_ifp, m));
1421 }
1422 
1423 static struct mbuf *
1424 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1425 {
1426 	struct ifnet *ifp = sc->sc_ifp;
1427 
1428 	/* Just pass in the packet to our lagg device */
1429 	m->m_pkthdr.rcvif = ifp;
1430 
1431 	return (m);
1432 }
1433 
1434 /*
1435  * Active failover
1436  */
1437 
1438 static int
1439 lagg_fail_attach(struct lagg_softc *sc)
1440 {
1441 	sc->sc_detach = lagg_fail_detach;
1442 	sc->sc_start = lagg_fail_start;
1443 	sc->sc_input = lagg_fail_input;
1444 	sc->sc_port_create = NULL;
1445 	sc->sc_port_destroy = NULL;
1446 
1447 	return (0);
1448 }
1449 
1450 static int
1451 lagg_fail_detach(struct lagg_softc *sc)
1452 {
1453 	return (0);
1454 }
1455 
1456 static int
1457 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1458 {
1459 	struct lagg_port *lp;
1460 
1461 	/* Use the master port if active or the next available port */
1462 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1463 		m_freem(m);
1464 		return (ENOENT);
1465 	}
1466 
1467 	/* Send mbuf */
1468 	return (lagg_enqueue(lp->lp_ifp, m));
1469 }
1470 
1471 static struct mbuf *
1472 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1473 {
1474 	struct ifnet *ifp = sc->sc_ifp;
1475 	struct lagg_port *tmp_tp;
1476 
1477 	if (lp == sc->sc_primary) {
1478 		m->m_pkthdr.rcvif = ifp;
1479 		return (m);
1480 	}
1481 
1482 	if (sc->sc_primary->lp_link_state == LINK_STATE_DOWN) {
1483 		tmp_tp = lagg_link_active(sc, NULL);
1484 		/*
1485 		 * If tmp_tp is null, we've recieved a packet when all
1486 		 * our links are down. Weird, but process it anyways.
1487 		 */
1488 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1489 			m->m_pkthdr.rcvif = ifp;
1490 			return (m);
1491 		}
1492 	}
1493 
1494 	m_freem(m);
1495 	return (NULL);
1496 }
1497 
1498 /*
1499  * Loadbalancing
1500  */
1501 
1502 static int
1503 lagg_lb_attach(struct lagg_softc *sc)
1504 {
1505 	struct lagg_port *lp;
1506 	struct lagg_lb *lb;
1507 
1508 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1509 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1510 		return (ENOMEM);
1511 
1512 	sc->sc_detach = lagg_lb_detach;
1513 	sc->sc_start = lagg_lb_start;
1514 	sc->sc_input = lagg_lb_input;
1515 	sc->sc_port_create = lagg_lb_port_create;
1516 	sc->sc_port_destroy = lagg_lb_port_destroy;
1517 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1518 
1519 	lb->lb_key = arc4random();
1520 	sc->sc_psc = (caddr_t)lb;
1521 
1522 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1523 		lagg_lb_port_create(lp);
1524 
1525 	return (0);
1526 }
1527 
1528 static int
1529 lagg_lb_detach(struct lagg_softc *sc)
1530 {
1531 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1532 	if (lb != NULL)
1533 		free(lb, M_DEVBUF);
1534 	return (0);
1535 }
1536 
1537 static int
1538 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1539 {
1540 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1541 	struct lagg_port *lp_next;
1542 	int i = 0;
1543 
1544 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1545 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1546 		if (lp_next == lp)
1547 			continue;
1548 		if (i >= LAGG_MAX_PORTS)
1549 			return (EINVAL);
1550 		if (sc->sc_ifflags & IFF_DEBUG)
1551 			printf("%s: port %s at index %d\n",
1552 			    sc->sc_ifname, lp_next->lp_ifname, i);
1553 		lb->lb_ports[i++] = lp_next;
1554 	}
1555 
1556 	return (0);
1557 }
1558 
1559 static int
1560 lagg_lb_port_create(struct lagg_port *lp)
1561 {
1562 	struct lagg_softc *sc = lp->lp_softc;
1563 	return (lagg_lb_porttable(sc, NULL));
1564 }
1565 
1566 static void
1567 lagg_lb_port_destroy(struct lagg_port *lp)
1568 {
1569 	struct lagg_softc *sc = lp->lp_softc;
1570 	lagg_lb_porttable(sc, lp);
1571 }
1572 
1573 static int
1574 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1575 {
1576 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1577 	struct lagg_port *lp = NULL;
1578 	uint32_t p = 0;
1579 	int idx;
1580 
1581 	p = lagg_hashmbuf(m, lb->lb_key);
1582 	if ((idx = p % sc->sc_count) >= LAGG_MAX_PORTS) {
1583 		m_freem(m);
1584 		return (EINVAL);
1585 	}
1586 	lp = lb->lb_ports[idx];
1587 
1588 	/*
1589 	 * Check the port's link state. This will return the next active
1590 	 * port if the link is down or the port is NULL.
1591 	 */
1592 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1593 		m_freem(m);
1594 		return (ENOENT);
1595 	}
1596 
1597 	/* Send mbuf */
1598 	return (lagg_enqueue(lp->lp_ifp, m));
1599 }
1600 
1601 static struct mbuf *
1602 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1603 {
1604 	struct ifnet *ifp = sc->sc_ifp;
1605 
1606 	/* Just pass in the packet to our lagg device */
1607 	m->m_pkthdr.rcvif = ifp;
1608 
1609 	return (m);
1610 }
1611 
1612 /*
1613  * 802.3ad LACP
1614  */
1615 
1616 static int
1617 lagg_lacp_attach(struct lagg_softc *sc)
1618 {
1619 	struct lagg_port *lp;
1620 	int error;
1621 
1622 	sc->sc_detach = lagg_lacp_detach;
1623 	sc->sc_port_create = lacp_port_create;
1624 	sc->sc_port_destroy = lacp_port_destroy;
1625 	sc->sc_linkstate = lacp_linkstate;
1626 	sc->sc_start = lagg_lacp_start;
1627 	sc->sc_input = lagg_lacp_input;
1628 	sc->sc_init = lacp_init;
1629 	sc->sc_stop = lacp_stop;
1630 	sc->sc_lladdr = lagg_lacp_lladdr;
1631 	sc->sc_req = lacp_req;
1632 	sc->sc_portreq = lacp_portreq;
1633 
1634 	error = lacp_attach(sc);
1635 	if (error)
1636 		return (error);
1637 
1638 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1639 		lacp_port_create(lp);
1640 
1641 	return (error);
1642 }
1643 
1644 static int
1645 lagg_lacp_detach(struct lagg_softc *sc)
1646 {
1647 	struct lagg_port *lp;
1648 	int error;
1649 
1650 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1651 		lacp_port_destroy(lp);
1652 
1653 	/* unlocking is safe here */
1654 	LAGG_WUNLOCK(sc);
1655 	error = lacp_detach(sc);
1656 	LAGG_WLOCK(sc);
1657 
1658 	return (error);
1659 }
1660 
1661 static void
1662 lagg_lacp_lladdr(struct lagg_softc *sc)
1663 {
1664 	struct lagg_port *lp;
1665 
1666 	/* purge all the lacp ports */
1667 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1668 		lacp_port_destroy(lp);
1669 
1670 	/* add them back in */
1671 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1672 		lacp_port_create(lp);
1673 }
1674 
1675 static int
1676 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1677 {
1678 	struct lagg_port *lp;
1679 
1680 	lp = lacp_select_tx_port(sc, m);
1681 	if (lp == NULL) {
1682 		m_freem(m);
1683 		return (EBUSY);
1684 	}
1685 
1686 	/* Send mbuf */
1687 	return (lagg_enqueue(lp->lp_ifp, m));
1688 }
1689 
1690 static struct mbuf *
1691 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1692 {
1693 	struct ifnet *ifp = sc->sc_ifp;
1694 	struct ether_header *eh;
1695 	u_short etype;
1696 
1697 	eh = mtod(m, struct ether_header *);
1698 	etype = ntohs(eh->ether_type);
1699 
1700 	/* Tap off LACP control messages */
1701 	if (etype == ETHERTYPE_SLOW) {
1702 		lacp_input(lp, m);
1703 		return (NULL);
1704 	}
1705 
1706 	/*
1707 	 * If the port is not collecting or not in the active aggregator then
1708 	 * free and return.
1709 	 */
1710 	if ((lp->lp_flags & LAGG_PORT_COLLECTING) == 0 ||
1711 	    lacp_port_isactive(lp) == 0) {
1712 		m_freem(m);
1713 		return (NULL);
1714 	}
1715 
1716 	m->m_pkthdr.rcvif = ifp;
1717 	return (m);
1718 }
1719