xref: /freebsd/sys/net/if_lagg.c (revision 282a3889ebf826db9839be296ff1dd903f6d6d6e)
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 int	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_port_state(struct ifnet *, int);
92 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
93 static int	lagg_port_output(struct ifnet *, struct mbuf *,
94 		    struct sockaddr *, struct rtentry *);
95 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
96 static int	lagg_port_checkstacking(struct lagg_softc *);
97 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
98 static void	lagg_init(void *);
99 static void	lagg_stop(struct lagg_softc *);
100 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
101 static int	lagg_ether_setmulti(struct lagg_softc *);
102 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
103 static	int	lagg_setflag(struct lagg_port *, int, int,
104 		    int (*func)(struct ifnet *, int));
105 static	int	lagg_setflags(struct lagg_port *, int status);
106 static void	lagg_start(struct ifnet *);
107 static int	lagg_media_change(struct ifnet *);
108 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
109 static struct lagg_port *lagg_link_active(struct lagg_softc *,
110 	    struct lagg_port *);
111 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
112 
113 IFC_SIMPLE_DECLARE(lagg, 0);
114 
115 /* Simple round robin */
116 static int	lagg_rr_attach(struct lagg_softc *);
117 static int	lagg_rr_detach(struct lagg_softc *);
118 static void	lagg_rr_port_destroy(struct lagg_port *);
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 	if (sc->sc_detach != NULL)
277 		(*sc->sc_detach)(sc);
278 
279 	LAGG_WUNLOCK(sc);
280 
281 	ifmedia_removeall(&sc->sc_media);
282 	ether_ifdetach(ifp);
283 	if_free_type(ifp, IFT_ETHER);
284 
285 	mtx_lock(&lagg_list_mtx);
286 	SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
287 	mtx_unlock(&lagg_list_mtx);
288 
289 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
290 	LAGG_LOCK_DESTROY(sc);
291 	free(sc, M_DEVBUF);
292 }
293 
294 static void
295 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
296 {
297 	struct ifnet *ifp = sc->sc_ifp;
298 
299 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
300 		return;
301 
302 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
303 	/* Let the protocol know the MAC has changed */
304 	if (sc->sc_lladdr != NULL)
305 		(*sc->sc_lladdr)(sc);
306 }
307 
308 static int
309 lagg_capabilities(struct lagg_softc *sc)
310 {
311 	struct lagg_port *lp;
312 	int cap = ~0, priv;
313 
314 	LAGG_WLOCK_ASSERT(sc);
315 
316 	/* Preserve private capabilities */
317 	priv = sc->sc_capabilities & IFCAP_LAGG_MASK;
318 
319 	/* Get capabilities from the lagg ports */
320 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
321 		cap &= lp->lp_capabilities;
322 
323 	if (sc->sc_ifflags & IFF_DEBUG) {
324 		printf("%s: capabilities 0x%08x\n",
325 		    sc->sc_ifname, cap == ~0 ? priv : (cap | priv));
326 	}
327 
328 	return (cap == ~0 ? priv : (cap | priv));
329 }
330 
331 static void
332 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
333 {
334 	struct lagg_softc *sc = lp->lp_softc;
335 	struct ifnet *ifp = lp->lp_ifp;
336 	struct lagg_llq *llq;
337 	int pending = 0;
338 
339 	LAGG_WLOCK_ASSERT(sc);
340 
341 	if (lp->lp_detaching ||
342 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
343 		return;
344 
345 	/* Check to make sure its not already queued to be changed */
346 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
347 		if (llq->llq_ifp == ifp) {
348 			pending = 1;
349 			break;
350 		}
351 	}
352 
353 	if (!pending) {
354 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
355 		if (llq == NULL)	/* XXX what to do */
356 			return;
357 	}
358 
359 	/* Update the lladdr even if pending, it may have changed */
360 	llq->llq_ifp = ifp;
361 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
362 
363 	if (!pending)
364 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
365 
366 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
367 }
368 
369 /*
370  * Set the interface MAC address from a taskqueue to avoid a LOR.
371  */
372 static void
373 lagg_port_setlladdr(void *arg, int pending)
374 {
375 	struct lagg_softc *sc = (struct lagg_softc *)arg;
376 	struct lagg_llq *llq, *head;
377 	struct ifnet *ifp;
378 	int error;
379 
380 	/* Grab a local reference of the queue and remove it from the softc */
381 	LAGG_WLOCK(sc);
382 	head = SLIST_FIRST(&sc->sc_llq_head);
383 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
384 	LAGG_WUNLOCK(sc);
385 
386 	/*
387 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
388 	 * unlocked as we have the only reference to it.
389 	 */
390 	for (llq = head; llq != NULL; llq = head) {
391 		ifp = llq->llq_ifp;
392 
393 		/* Set the link layer address */
394 		error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
395 		if (error)
396 			printf("%s: setlladdr failed on %s\n", __func__,
397 			    ifp->if_xname);
398 
399 		head = SLIST_NEXT(llq, llq_entries);
400 		free(llq, M_DEVBUF);
401 	}
402 }
403 
404 static int
405 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
406 {
407 	struct lagg_softc *sc_ptr;
408 	struct lagg_port *lp;
409 	int error = 0;
410 
411 	LAGG_WLOCK_ASSERT(sc);
412 
413 	/* Limit the maximal number of lagg ports */
414 	if (sc->sc_count >= LAGG_MAX_PORTS)
415 		return (ENOSPC);
416 
417 	/* New lagg port has to be in an idle state */
418 	if (ifp->if_drv_flags & IFF_DRV_OACTIVE)
419 		return (EBUSY);
420 
421 	/* Check if port has already been associated to a lagg */
422 	if (ifp->if_lagg != NULL)
423 		return (EBUSY);
424 
425 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
426 	if (ifp->if_type != IFT_ETHER)
427 		return (EPROTONOSUPPORT);
428 
429 	if ((lp = malloc(sizeof(struct lagg_port),
430 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
431 		return (ENOMEM);
432 
433 	/* Check if port is a stacked lagg */
434 	mtx_lock(&lagg_list_mtx);
435 	SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
436 		if (ifp == sc_ptr->sc_ifp) {
437 			mtx_unlock(&lagg_list_mtx);
438 			free(lp, M_DEVBUF);
439 			return (EINVAL);
440 			/* XXX disable stacking for the moment, its untested
441 			lp->lp_flags |= LAGG_PORT_STACK;
442 			if (lagg_port_checkstacking(sc_ptr) >=
443 			    LAGG_MAX_STACKING) {
444 				mtx_unlock(&lagg_list_mtx);
445 				free(lp, M_DEVBUF);
446 				return (E2BIG);
447 			}
448 			*/
449 		}
450 	}
451 	mtx_unlock(&lagg_list_mtx);
452 
453 	/* Change the interface type */
454 	lp->lp_iftype = ifp->if_type;
455 	ifp->if_type = IFT_IEEE8023ADLAG;
456 	ifp->if_lagg = lp;
457 	lp->lp_ioctl = ifp->if_ioctl;
458 	ifp->if_ioctl = lagg_port_ioctl;
459 	lp->lp_output = ifp->if_output;
460 	ifp->if_output = lagg_port_output;
461 
462 	lp->lp_ifp = ifp;
463 	lp->lp_softc = sc;
464 
465 	/* Save port link layer address */
466 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
467 
468 	if (SLIST_EMPTY(&sc->sc_ports)) {
469 		sc->sc_primary = lp;
470 		lagg_lladdr(sc, IF_LLADDR(ifp));
471 	} else {
472 		/* Update link layer address for this port */
473 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
474 	}
475 
476 	/* Insert into the list of ports */
477 	SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
478 	sc->sc_count++;
479 
480 	/* Update lagg capabilities */
481 	sc->sc_capabilities = lagg_capabilities(sc);
482 
483 	/* Add multicast addresses and interface flags to this port */
484 	lagg_ether_cmdmulti(lp, 1);
485 	lagg_setflags(lp, 1);
486 
487 	if (sc->sc_port_create != NULL)
488 		error = (*sc->sc_port_create)(lp);
489 	if (error) {
490 		/* remove the port again, without calling sc_port_destroy */
491 		lagg_port_destroy(lp, 0);
492 		return (error);
493 	}
494 
495 	return (error);
496 }
497 
498 static int
499 lagg_port_checkstacking(struct lagg_softc *sc)
500 {
501 	struct lagg_softc *sc_ptr;
502 	struct lagg_port *lp;
503 	int m = 0;
504 
505 	LAGG_WLOCK_ASSERT(sc);
506 
507 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
508 		if (lp->lp_flags & LAGG_PORT_STACK) {
509 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
510 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
511 		}
512 	}
513 
514 	return (m + 1);
515 }
516 
517 static int
518 lagg_port_destroy(struct lagg_port *lp, int runpd)
519 {
520 	struct lagg_softc *sc = lp->lp_softc;
521 	struct lagg_port *lp_ptr;
522 	struct lagg_llq *llq;
523 	struct ifnet *ifp = lp->lp_ifp;
524 
525 	LAGG_WLOCK_ASSERT(sc);
526 
527 	if (runpd && sc->sc_port_destroy != NULL)
528 		(*sc->sc_port_destroy)(lp);
529 
530 	/*
531 	 * Remove multicast addresses and interface flags from this port and
532 	 * reset the MAC address, skip if the interface is being detached.
533 	 */
534 	if (!lp->lp_detaching) {
535 		lagg_ether_cmdmulti(lp, 0);
536 		lagg_setflags(lp, 0);
537 		lagg_port_lladdr(lp, lp->lp_lladdr);
538 	}
539 
540 	/* Restore interface */
541 	ifp->if_type = lp->lp_iftype;
542 	ifp->if_ioctl = lp->lp_ioctl;
543 	ifp->if_output = lp->lp_output;
544 	ifp->if_lagg = NULL;
545 
546 	/* Finally, remove the port from the lagg */
547 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
548 	sc->sc_count--;
549 
550 	/* Update the primary interface */
551 	if (lp == sc->sc_primary) {
552 		uint8_t lladdr[ETHER_ADDR_LEN];
553 
554 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
555 			bzero(&lladdr, ETHER_ADDR_LEN);
556 		} else {
557 			bcopy(lp_ptr->lp_lladdr,
558 			    lladdr, ETHER_ADDR_LEN);
559 		}
560 		lagg_lladdr(sc, lladdr);
561 		sc->sc_primary = lp_ptr;
562 
563 		/* Update link layer address for each port */
564 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
565 			lagg_port_lladdr(lp_ptr, lladdr);
566 	}
567 
568 	/* Remove any pending lladdr changes from the queue */
569 	if (lp->lp_detaching) {
570 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
571 			if (llq->llq_ifp == ifp) {
572 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
573 				    llq_entries);
574 				free(llq, M_DEVBUF);
575 				break;	/* Only appears once */
576 			}
577 		}
578 	}
579 
580 	if (lp->lp_ifflags)
581 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
582 
583 	free(lp, M_DEVBUF);
584 
585 	/* Update lagg capabilities */
586 	sc->sc_capabilities = lagg_capabilities(sc);
587 
588 	return (0);
589 }
590 
591 static int
592 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
593 {
594 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
595 	struct lagg_softc *sc;
596 	struct lagg_port *lp = NULL;
597 	int error = 0;
598 
599 	/* Should be checked by the caller */
600 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
601 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
602 		goto fallback;
603 
604 	switch (cmd) {
605 	case SIOCGLAGGPORT:
606 		LAGG_RLOCK(sc);
607 		if (rp->rp_portname[0] == '\0' ||
608 		    ifunit(rp->rp_portname) != ifp) {
609 			error = EINVAL;
610 			break;
611 		}
612 
613 		if (lp->lp_softc != sc) {
614 			error = ENOENT;
615 			break;
616 		}
617 
618 		lagg_port2req(lp, rp);
619 		LAGG_RUNLOCK(sc);
620 		break;
621 	default:
622 		goto fallback;
623 	}
624 
625 	return (error);
626 
627 fallback:
628 	if (lp->lp_ioctl != NULL)
629 		return ((*lp->lp_ioctl)(ifp, cmd, data));
630 
631 	return (EINVAL);
632 }
633 
634 static int
635 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
636 	struct sockaddr *dst, struct rtentry *rt0)
637 {
638 	struct lagg_port *lp = ifp->if_lagg;
639 	struct ether_header *eh;
640 	short type = 0;
641 
642 	switch (dst->sa_family) {
643 		case pseudo_AF_HDRCMPLT:
644 		case AF_UNSPEC:
645 			eh = (struct ether_header *)dst->sa_data;
646 			type = eh->ether_type;
647 			break;
648 	}
649 
650 	/*
651 	 * Only allow ethernet types required to initiate or maintain the link,
652 	 * aggregated frames take a different path.
653 	 */
654 	switch (ntohs(type)) {
655 		case ETHERTYPE_PAE:	/* EAPOL PAE/802.1x */
656 			return ((*lp->lp_output)(ifp, m, dst, rt0));
657 	}
658 
659 	/* drop any other frames */
660 	m_freem(m);
661 	return (EBUSY);
662 }
663 
664 static void
665 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
666 {
667 	struct lagg_port *lp;
668 	struct lagg_softc *sc;
669 
670 	if ((lp = ifp->if_lagg) == NULL)
671 		return;
672 
673 	sc = lp->lp_softc;
674 
675 	LAGG_WLOCK(sc);
676 	lp->lp_detaching = 1;
677 	lagg_port_destroy(lp, 1);
678 	LAGG_WUNLOCK(sc);
679 }
680 
681 static void
682 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
683 {
684 	struct lagg_softc *sc = lp->lp_softc;
685 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
686 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
687 	rp->rp_prio = lp->lp_prio;
688 	rp->rp_flags = lp->lp_flags;
689 	if (sc->sc_portreq != NULL)
690 		(*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
691 
692 	/* Add protocol specific flags */
693 	switch (sc->sc_proto) {
694 		case LAGG_PROTO_FAILOVER:
695 			if (lp == sc->sc_primary)
696 				rp->rp_flags |= LAGG_PORT_MASTER;
697 			/* FALLTHROUGH */
698 		case LAGG_PROTO_ROUNDROBIN:
699 		case LAGG_PROTO_LOADBALANCE:
700 		case LAGG_PROTO_ETHERCHANNEL:
701 			if (LAGG_PORTACTIVE(lp))
702 				rp->rp_flags |= LAGG_PORT_ACTIVE;
703 			break;
704 
705 		case LAGG_PROTO_LACP:
706 			/* LACP has a different definition of active */
707 			if (lacp_port_isactive(lp))
708 				rp->rp_flags |= LAGG_PORT_ACTIVE;
709 			break;
710 	}
711 
712 }
713 
714 static void
715 lagg_init(void *xsc)
716 {
717 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
718 	struct lagg_port *lp;
719 	struct ifnet *ifp = sc->sc_ifp;
720 
721 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
722 		return;
723 
724 	LAGG_WLOCK(sc);
725 
726 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
727 	/* Update the port lladdrs */
728 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
729 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
730 
731 	if (sc->sc_init != NULL)
732 		(*sc->sc_init)(sc);
733 
734 	LAGG_WUNLOCK(sc);
735 }
736 
737 static void
738 lagg_stop(struct lagg_softc *sc)
739 {
740 	struct ifnet *ifp = sc->sc_ifp;
741 
742 	LAGG_WLOCK_ASSERT(sc);
743 
744 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
745 		return;
746 
747 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
748 
749 	if (sc->sc_stop != NULL)
750 		(*sc->sc_stop)(sc);
751 }
752 
753 static int
754 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
755 {
756 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
757 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
758 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
759 	struct ifreq *ifr = (struct ifreq *)data;
760 	struct lagg_port *lp;
761 	struct ifnet *tpif;
762 	struct thread *td = curthread;
763 	int i, error = 0, unlock = 1;
764 
765 	LAGG_WLOCK(sc);
766 
767 	bzero(&rpbuf, sizeof(rpbuf));
768 
769 	switch (cmd) {
770 	case SIOCGLAGG:
771 		ra->ra_proto = sc->sc_proto;
772 		ra->ra_ports = i = 0;
773 		if (sc->sc_req != NULL)
774 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
775 		lp = SLIST_FIRST(&sc->sc_ports);
776 		while (lp && ra->ra_size >=
777 		    i + sizeof(struct lagg_reqport)) {
778 			lagg_port2req(lp, &rpbuf);
779 			error = copyout(&rpbuf, (caddr_t)ra->ra_port + i,
780 			    sizeof(struct lagg_reqport));
781 			if (error)
782 				break;
783 			i += sizeof(struct lagg_reqport);
784 			ra->ra_ports++;
785 			lp = SLIST_NEXT(lp, lp_entries);
786 		}
787 		break;
788 	case SIOCSLAGG:
789 		error = priv_check(td, PRIV_NET_LAGG);
790 		if (error)
791 			break;
792 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
793 			error = EPROTONOSUPPORT;
794 			break;
795 		}
796 		if (sc->sc_proto != LAGG_PROTO_NONE) {
797 			error = sc->sc_detach(sc);
798 			/* Reset protocol and pointers */
799 			sc->sc_proto = LAGG_PROTO_NONE;
800 			sc->sc_detach = NULL;
801 			sc->sc_start = NULL;
802 			sc->sc_input = NULL;
803 			sc->sc_port_create = NULL;
804 			sc->sc_port_destroy = NULL;
805 			sc->sc_linkstate = NULL;
806 			sc->sc_init = NULL;
807 			sc->sc_stop = NULL;
808 			sc->sc_lladdr = NULL;
809 			sc->sc_req = NULL;
810 			sc->sc_portreq = NULL;
811 		}
812 		if (error != 0)
813 			break;
814 		for (i = 0; i < (sizeof(lagg_protos) /
815 		    sizeof(lagg_protos[0])); i++) {
816 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
817 				if (sc->sc_ifflags & IFF_DEBUG)
818 					printf("%s: using proto %u\n",
819 					    sc->sc_ifname,
820 					    lagg_protos[i].ti_proto);
821 				sc->sc_proto = lagg_protos[i].ti_proto;
822 				if (sc->sc_proto != LAGG_PROTO_NONE)
823 					error = lagg_protos[i].ti_attach(sc);
824 				goto out;
825 			}
826 		}
827 		error = EPROTONOSUPPORT;
828 		break;
829 	case SIOCGLAGGPORT:
830 		if (rp->rp_portname[0] == '\0' ||
831 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
832 			error = EINVAL;
833 			break;
834 		}
835 
836 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
837 		    lp->lp_softc != sc) {
838 			error = ENOENT;
839 			break;
840 		}
841 
842 		lagg_port2req(lp, rp);
843 		break;
844 	case SIOCSLAGGPORT:
845 		error = priv_check(td, PRIV_NET_LAGG);
846 		if (error)
847 			break;
848 		if (rp->rp_portname[0] == '\0' ||
849 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
850 			error = EINVAL;
851 			break;
852 		}
853 		error = lagg_port_create(sc, tpif);
854 		break;
855 	case SIOCSLAGGDELPORT:
856 		error = priv_check(td, PRIV_NET_LAGG);
857 		if (error)
858 			break;
859 		if (rp->rp_portname[0] == '\0' ||
860 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
861 			error = EINVAL;
862 			break;
863 		}
864 
865 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
866 		    lp->lp_softc != sc) {
867 			error = ENOENT;
868 			break;
869 		}
870 
871 		error = lagg_port_destroy(lp, 1);
872 		break;
873 	case SIOCSIFFLAGS:
874 		/* Set flags on ports too */
875 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
876 			lagg_setflags(lp, 1);
877 		}
878 
879 		if (!(ifp->if_flags & IFF_UP) &&
880 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
881 			/*
882 			 * If interface is marked down and it is running,
883 			 * then stop and disable it.
884 			 */
885 			lagg_stop(sc);
886 		} else if ((ifp->if_flags & IFF_UP) &&
887 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
888 			/*
889 			 * If interface is marked up and it is stopped, then
890 			 * start it.
891 			 */
892 			LAGG_WUNLOCK(sc);
893 			unlock = 0;
894 			(*ifp->if_init)(sc);
895 		}
896 		break;
897 	case SIOCADDMULTI:
898 	case SIOCDELMULTI:
899 		error = lagg_ether_setmulti(sc);
900 		break;
901 	case SIOCSIFMEDIA:
902 	case SIOCGIFMEDIA:
903 		LAGG_WUNLOCK(sc);
904 		unlock = 0;
905 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
906 		break;
907 	default:
908 		LAGG_WUNLOCK(sc);
909 		unlock = 0;
910 		error = ether_ioctl(ifp, cmd, data);
911 		break;
912 	}
913 
914 out:
915 	if (unlock)
916 		LAGG_WUNLOCK(sc);
917 	return (error);
918 }
919 
920 static int
921 lagg_ether_setmulti(struct lagg_softc *sc)
922 {
923 	struct lagg_port *lp;
924 
925 	LAGG_WLOCK_ASSERT(sc);
926 
927 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
928 		/* First, remove any existing filter entries. */
929 		lagg_ether_cmdmulti(lp, 0);
930 		/* copy all addresses from the lagg interface to the port */
931 		lagg_ether_cmdmulti(lp, 1);
932 	}
933 	return (0);
934 }
935 
936 static int
937 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
938 {
939 	struct lagg_softc *sc = lp->lp_softc;
940 	struct ifnet *ifp = lp->lp_ifp;
941 	struct ifnet *scifp = sc->sc_ifp;
942 	struct lagg_mc *mc;
943 	struct ifmultiaddr *ifma, *rifma = NULL;
944 	struct sockaddr_dl sdl;
945 	int error;
946 
947 	LAGG_WLOCK_ASSERT(sc);
948 
949 	bzero((char *)&sdl, sizeof(sdl));
950 	sdl.sdl_len = sizeof(sdl);
951 	sdl.sdl_family = AF_LINK;
952 	sdl.sdl_type = IFT_ETHER;
953 	sdl.sdl_alen = ETHER_ADDR_LEN;
954 	sdl.sdl_index = ifp->if_index;
955 
956 	if (set) {
957 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
958 			if (ifma->ifma_addr->sa_family != AF_LINK)
959 				continue;
960 			bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
961 			    LLADDR(&sdl), ETHER_ADDR_LEN);
962 
963 			error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
964 			if (error)
965 				return (error);
966 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
967 			if (mc == NULL)
968 				return (ENOMEM);
969 			mc->mc_ifma = rifma;
970 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
971 		}
972 	} else {
973 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
974 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
975 			if_delmulti_ifma(mc->mc_ifma);
976 			free(mc, M_DEVBUF);
977 		}
978 	}
979 	return (0);
980 }
981 
982 /* Handle a ref counted flag that should be set on the lagg port as well */
983 static int
984 lagg_setflag(struct lagg_port *lp, int flag, int status,
985 	     int (*func)(struct ifnet *, int))
986 {
987 	struct lagg_softc *sc = lp->lp_softc;
988 	struct ifnet *scifp = sc->sc_ifp;
989 	struct ifnet *ifp = lp->lp_ifp;
990 	int error;
991 
992 	LAGG_WLOCK_ASSERT(sc);
993 
994 	status = status ? (scifp->if_flags & flag) : 0;
995 	/* Now "status" contains the flag value or 0 */
996 
997 	/*
998 	 * See if recorded ports status is different from what
999 	 * we want it to be.  If it is, flip it.  We record ports
1000 	 * status in lp_ifflags so that we won't clear ports flag
1001 	 * we haven't set.  In fact, we don't clear or set ports
1002 	 * flags directly, but get or release references to them.
1003 	 * That's why we can be sure that recorded flags still are
1004 	 * in accord with actual ports flags.
1005 	 */
1006 	if (status != (lp->lp_ifflags & flag)) {
1007 		error = (*func)(ifp, status);
1008 		if (error)
1009 			return (error);
1010 		lp->lp_ifflags &= ~flag;
1011 		lp->lp_ifflags |= status;
1012 	}
1013 	return (0);
1014 }
1015 
1016 /*
1017  * Handle IFF_* flags that require certain changes on the lagg port
1018  * if "status" is true, update ports flags respective to the lagg
1019  * if "status" is false, forcedly clear the flags set on port.
1020  */
1021 static int
1022 lagg_setflags(struct lagg_port *lp, int status)
1023 {
1024 	int error, i;
1025 
1026 	for (i = 0; lagg_pflags[i].flag; i++) {
1027 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1028 		    status, lagg_pflags[i].func);
1029 		if (error)
1030 			return (error);
1031 	}
1032 	return (0);
1033 }
1034 
1035 static void
1036 lagg_start(struct ifnet *ifp)
1037 {
1038 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1039 	struct mbuf *m;
1040 	int error = 0;
1041 
1042 	LAGG_RLOCK(sc);
1043 	for (;; error = 0) {
1044 		IFQ_DEQUEUE(&ifp->if_snd, m);
1045 		if (m == NULL)
1046 			break;
1047 
1048 		BPF_MTAP(ifp, m);
1049 
1050 		if (sc->sc_proto != LAGG_PROTO_NONE)
1051 			error = (*sc->sc_start)(sc, m);
1052 		else
1053 			m_freem(m);
1054 
1055 		if (error == 0)
1056 			ifp->if_opackets++;
1057 		else {
1058 			m_freem(m); /* sc_start failed */
1059 			ifp->if_oerrors++;
1060 		}
1061 	}
1062 	LAGG_RUNLOCK(sc);
1063 
1064 	return;
1065 }
1066 
1067 static struct mbuf *
1068 lagg_input(struct ifnet *ifp, struct mbuf *m)
1069 {
1070 	struct lagg_port *lp = ifp->if_lagg;
1071 	struct lagg_softc *sc = lp->lp_softc;
1072 	struct ifnet *scifp = sc->sc_ifp;
1073 
1074 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1075 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1076 	    sc->sc_proto == LAGG_PROTO_NONE) {
1077 		m_freem(m);
1078 		return (NULL);
1079 	}
1080 
1081 	LAGG_RLOCK(sc);
1082 	BPF_MTAP(scifp, m);
1083 
1084 	m = (*sc->sc_input)(sc, lp, m);
1085 
1086 	if (m != NULL) {
1087 		scifp->if_ipackets++;
1088 		scifp->if_ibytes += m->m_pkthdr.len;
1089 	}
1090 
1091 	LAGG_RUNLOCK(sc);
1092 	return (m);
1093 }
1094 
1095 static int
1096 lagg_media_change(struct ifnet *ifp)
1097 {
1098 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1099 
1100 	if (sc->sc_ifflags & IFF_DEBUG)
1101 		printf("%s\n", __func__);
1102 
1103 	/* Ignore */
1104 	return (0);
1105 }
1106 
1107 static void
1108 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1109 {
1110 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1111 	struct lagg_port *lp;
1112 
1113 	imr->ifm_status = IFM_AVALID;
1114 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1115 
1116 	LAGG_RLOCK(sc);
1117 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1118 		if (LAGG_PORTACTIVE(lp))
1119 			imr->ifm_status |= IFM_ACTIVE;
1120 	}
1121 	LAGG_RUNLOCK(sc);
1122 }
1123 
1124 static void
1125 lagg_port_state(struct ifnet *ifp, int state)
1126 {
1127 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1128 	struct lagg_softc *sc = NULL;
1129 
1130 	if (lp != NULL)
1131 		sc = lp->lp_softc;
1132 	if (sc == NULL)
1133 		return;
1134 
1135 	LAGG_WLOCK(sc);
1136 	if (sc->sc_linkstate != NULL)
1137 		(*sc->sc_linkstate)(lp);
1138 	LAGG_WUNLOCK(sc);
1139 }
1140 
1141 struct lagg_port *
1142 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1143 {
1144 	struct lagg_port *lp_next, *rval = NULL;
1145 	// int new_link = LINK_STATE_DOWN;
1146 
1147 	LAGG_RLOCK_ASSERT(sc);
1148 	/*
1149 	 * Search a port which reports an active link state.
1150 	 */
1151 
1152 	if (lp == NULL)
1153 		goto search;
1154 	if (LAGG_PORTACTIVE(lp)) {
1155 		rval = lp;
1156 		goto found;
1157 	}
1158 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1159 	    LAGG_PORTACTIVE(lp_next)) {
1160 		rval = lp_next;
1161 		goto found;
1162 	}
1163 
1164 search:
1165 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1166 		if (LAGG_PORTACTIVE(lp_next)) {
1167 			rval = lp_next;
1168 			goto found;
1169 		}
1170 	}
1171 
1172 found:
1173 	if (rval != NULL) {
1174 		/*
1175 		 * The IEEE 802.1D standard assumes that a lagg with
1176 		 * multiple ports is always full duplex. This is valid
1177 		 * for load sharing laggs and if at least two links
1178 		 * are active. Unfortunately, checking the latter would
1179 		 * be too expensive at this point.
1180 		 XXX
1181 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1182 		    (sc->sc_count > 1))
1183 			new_link = LINK_STATE_FULL_DUPLEX;
1184 		else
1185 			new_link = rval->lp_link_state;
1186 		 */
1187 	}
1188 
1189 	return (rval);
1190 }
1191 
1192 static const void *
1193 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1194 {
1195 	if (m->m_pkthdr.len < (off + len)) {
1196 		return (NULL);
1197 	} else if (m->m_len < (off + len)) {
1198 		m_copydata(m, off, len, buf);
1199 		return (buf);
1200 	}
1201 	return (mtod(m, char *) + off);
1202 }
1203 
1204 uint32_t
1205 lagg_hashmbuf(struct mbuf *m, uint32_t key)
1206 {
1207 	uint16_t etype;
1208 	uint32_t p = 0;
1209 	int off;
1210 	struct ether_header *eh;
1211 	struct ether_vlan_header vlanbuf;
1212 	const struct ether_vlan_header *vlan;
1213 #ifdef INET
1214 	const struct ip *ip;
1215 	struct ip ipbuf;
1216 #endif
1217 #ifdef INET6
1218 	const struct ip6_hdr *ip6;
1219 	struct ip6_hdr ip6buf;
1220 	uint32_t flow;
1221 #endif
1222 
1223 	off = sizeof(*eh);
1224 	if (m->m_len < off)
1225 		goto out;
1226 	eh = mtod(m, struct ether_header *);
1227 	etype = ntohs(eh->ether_type);
1228 	p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, key);
1229 	p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1230 
1231 	/* Special handling for encapsulating VLAN frames */
1232 	if (m->m_flags & M_VLANTAG) {
1233 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1234 		    sizeof(m->m_pkthdr.ether_vtag), p);
1235 	} else if (etype == ETHERTYPE_VLAN) {
1236 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &vlanbuf);
1237 		if (vlan == NULL)
1238 			goto out;
1239 
1240 		p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1241 		etype = ntohs(vlan->evl_proto);
1242 		off += sizeof(*vlan) - sizeof(*eh);
1243 	}
1244 
1245 	switch (etype) {
1246 #ifdef INET
1247 	case ETHERTYPE_IP:
1248 		ip = lagg_gethdr(m, off, sizeof(*ip), &ipbuf);
1249 		if (ip == NULL)
1250 			goto out;
1251 
1252 		p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1253 		p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1254 		break;
1255 #endif
1256 #ifdef INET6
1257 	case ETHERTYPE_IPV6:
1258 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &ip6buf);
1259 		if (ip6 == NULL)
1260 			goto out;
1261 
1262 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1263 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1264 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1265 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1266 		break;
1267 #endif
1268 	}
1269 out:
1270 	return (p);
1271 }
1272 
1273 int
1274 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1275 {
1276 	int error = 0;
1277 
1278 	IFQ_HANDOFF(ifp, m, error);
1279 	return (error);
1280 }
1281 
1282 /*
1283  * Simple round robin aggregation
1284  */
1285 
1286 static int
1287 lagg_rr_attach(struct lagg_softc *sc)
1288 {
1289 	struct lagg_port *lp;
1290 
1291 	sc->sc_detach = lagg_rr_detach;
1292 	sc->sc_start = lagg_rr_start;
1293 	sc->sc_input = lagg_rr_input;
1294 	sc->sc_port_create = NULL;
1295 	sc->sc_port_destroy = lagg_rr_port_destroy;
1296 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1297 
1298 	lp = SLIST_FIRST(&sc->sc_ports);
1299 	sc->sc_psc = (caddr_t)lp;
1300 
1301 	return (0);
1302 }
1303 
1304 static int
1305 lagg_rr_detach(struct lagg_softc *sc)
1306 {
1307 	sc->sc_psc = NULL;
1308 	return (0);
1309 }
1310 
1311 static void
1312 lagg_rr_port_destroy(struct lagg_port *lp)
1313 {
1314 	struct lagg_softc *sc = lp->lp_softc;
1315 
1316 	if (lp == (struct lagg_port *)sc->sc_psc)
1317 		sc->sc_psc = NULL;
1318 }
1319 
1320 static int
1321 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1322 {
1323 	struct lagg_port *lp = (struct lagg_port *)sc->sc_psc, *lp_next;
1324 	int error = 0;
1325 
1326 	if (lp == NULL && (lp = lagg_link_active(sc, NULL)) == NULL)
1327 		return (ENOENT);
1328 
1329 	/* Send mbuf */
1330 	error = lagg_enqueue(lp->lp_ifp, m);
1331 
1332 	/* Get next active port */
1333 	lp_next = lagg_link_active(sc, SLIST_NEXT(lp, lp_entries));
1334 	sc->sc_psc = (caddr_t)lp_next;
1335 
1336 	return (error);
1337 }
1338 
1339 static struct mbuf *
1340 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1341 {
1342 	struct ifnet *ifp = sc->sc_ifp;
1343 
1344 	/* Just pass in the packet to our lagg device */
1345 	m->m_pkthdr.rcvif = ifp;
1346 
1347 	return (m);
1348 }
1349 
1350 /*
1351  * Active failover
1352  */
1353 
1354 static int
1355 lagg_fail_attach(struct lagg_softc *sc)
1356 {
1357 	sc->sc_detach = lagg_fail_detach;
1358 	sc->sc_start = lagg_fail_start;
1359 	sc->sc_input = lagg_fail_input;
1360 	sc->sc_port_create = NULL;
1361 	sc->sc_port_destroy = NULL;
1362 
1363 	return (0);
1364 }
1365 
1366 static int
1367 lagg_fail_detach(struct lagg_softc *sc)
1368 {
1369 	return (0);
1370 }
1371 
1372 static int
1373 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1374 {
1375 	struct lagg_port *lp;
1376 
1377 	/* Use the master port if active or the next available port */
1378 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL)
1379 		return (ENOENT);
1380 
1381 	/* Send mbuf */
1382 	return (lagg_enqueue(lp->lp_ifp, m));
1383 }
1384 
1385 static struct mbuf *
1386 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1387 {
1388 	struct ifnet *ifp = sc->sc_ifp;
1389 	struct lagg_port *tmp_tp;
1390 
1391 	if (lp == sc->sc_primary) {
1392 		m->m_pkthdr.rcvif = ifp;
1393 		return (m);
1394 	}
1395 
1396 	if (sc->sc_primary->lp_link_state == LINK_STATE_DOWN) {
1397 		tmp_tp = lagg_link_active(sc, NULL);
1398 		/*
1399 		 * If tmp_tp is null, we've recieved a packet when all
1400 		 * our links are down. Weird, but process it anyways.
1401 		 */
1402 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1403 			m->m_pkthdr.rcvif = ifp;
1404 			return (m);
1405 		}
1406 	}
1407 
1408 	m_freem(m);
1409 	return (NULL);
1410 }
1411 
1412 /*
1413  * Loadbalancing
1414  */
1415 
1416 static int
1417 lagg_lb_attach(struct lagg_softc *sc)
1418 {
1419 	struct lagg_port *lp;
1420 	struct lagg_lb *lb;
1421 
1422 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1423 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1424 		return (ENOMEM);
1425 
1426 	sc->sc_detach = lagg_lb_detach;
1427 	sc->sc_start = lagg_lb_start;
1428 	sc->sc_input = lagg_lb_input;
1429 	sc->sc_port_create = lagg_lb_port_create;
1430 	sc->sc_port_destroy = lagg_lb_port_destroy;
1431 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1432 
1433 	lb->lb_key = arc4random();
1434 	sc->sc_psc = (caddr_t)lb;
1435 
1436 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1437 		lagg_lb_port_create(lp);
1438 
1439 	return (0);
1440 }
1441 
1442 static int
1443 lagg_lb_detach(struct lagg_softc *sc)
1444 {
1445 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1446 	if (lb != NULL)
1447 		free(lb, M_DEVBUF);
1448 	return (0);
1449 }
1450 
1451 static int
1452 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1453 {
1454 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1455 	struct lagg_port *lp_next;
1456 	int i = 0;
1457 
1458 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1459 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1460 		if (lp_next == lp)
1461 			continue;
1462 		if (i >= LAGG_MAX_PORTS)
1463 			return (EINVAL);
1464 		if (sc->sc_ifflags & IFF_DEBUG)
1465 			printf("%s: port %s at index %d\n",
1466 			    sc->sc_ifname, lp_next->lp_ifname, i);
1467 		lb->lb_ports[i++] = lp_next;
1468 	}
1469 
1470 	return (0);
1471 }
1472 
1473 static int
1474 lagg_lb_port_create(struct lagg_port *lp)
1475 {
1476 	struct lagg_softc *sc = lp->lp_softc;
1477 	return (lagg_lb_porttable(sc, NULL));
1478 }
1479 
1480 static void
1481 lagg_lb_port_destroy(struct lagg_port *lp)
1482 {
1483 	struct lagg_softc *sc = lp->lp_softc;
1484 	lagg_lb_porttable(sc, lp);
1485 }
1486 
1487 static int
1488 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1489 {
1490 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1491 	struct lagg_port *lp = NULL;
1492 	uint32_t p = 0;
1493 	int idx;
1494 
1495 	p = lagg_hashmbuf(m, lb->lb_key);
1496 	if ((idx = p % sc->sc_count) >= LAGG_MAX_PORTS)
1497 		return (EINVAL);
1498 	lp = lb->lb_ports[idx];
1499 
1500 	/*
1501 	 * Check the port's link state. This will return the next active
1502 	 * port if the link is down or the port is NULL.
1503 	 */
1504 	if ((lp = lagg_link_active(sc, lp)) == NULL)
1505 		return (ENOENT);
1506 
1507 	/* Send mbuf */
1508 	return (lagg_enqueue(lp->lp_ifp, m));
1509 }
1510 
1511 static struct mbuf *
1512 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1513 {
1514 	struct ifnet *ifp = sc->sc_ifp;
1515 
1516 	/* Just pass in the packet to our lagg device */
1517 	m->m_pkthdr.rcvif = ifp;
1518 
1519 	return (m);
1520 }
1521 
1522 /*
1523  * 802.3ad LACP
1524  */
1525 
1526 static int
1527 lagg_lacp_attach(struct lagg_softc *sc)
1528 {
1529 	struct lagg_port *lp;
1530 	int error;
1531 
1532 	sc->sc_detach = lagg_lacp_detach;
1533 	sc->sc_port_create = lacp_port_create;
1534 	sc->sc_port_destroy = lacp_port_destroy;
1535 	sc->sc_linkstate = lacp_linkstate;
1536 	sc->sc_start = lagg_lacp_start;
1537 	sc->sc_input = lagg_lacp_input;
1538 	sc->sc_init = lacp_init;
1539 	sc->sc_stop = lacp_stop;
1540 	sc->sc_lladdr = lagg_lacp_lladdr;
1541 	sc->sc_req = lacp_req;
1542 	sc->sc_portreq = lacp_portreq;
1543 
1544 	error = lacp_attach(sc);
1545 	if (error)
1546 		return (error);
1547 
1548 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1549 		lacp_port_create(lp);
1550 
1551 	return (error);
1552 }
1553 
1554 static int
1555 lagg_lacp_detach(struct lagg_softc *sc)
1556 {
1557 	struct lagg_port *lp;
1558 	int error;
1559 
1560 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1561 		lacp_port_destroy(lp);
1562 
1563 	/* unlocking is safe here */
1564 	LAGG_WUNLOCK(sc);
1565 	error = lacp_detach(sc);
1566 	LAGG_WLOCK(sc);
1567 
1568 	return (error);
1569 }
1570 
1571 static void
1572 lagg_lacp_lladdr(struct lagg_softc *sc)
1573 {
1574 	struct lagg_port *lp;
1575 
1576 	/* purge all the lacp ports */
1577 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1578 		lacp_port_destroy(lp);
1579 
1580 	/* add them back in */
1581 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1582 		lacp_port_create(lp);
1583 }
1584 
1585 static int
1586 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1587 {
1588 	struct lagg_port *lp;
1589 
1590 	lp = lacp_select_tx_port(sc, m);
1591 	if (lp == NULL)
1592 		return (EBUSY);
1593 
1594 	/* Send mbuf */
1595 	return (lagg_enqueue(lp->lp_ifp, m));
1596 }
1597 
1598 static struct mbuf *
1599 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1600 {
1601 	struct ifnet *ifp = sc->sc_ifp;
1602 	struct ether_header *eh;
1603 	u_short etype;
1604 
1605 	eh = mtod(m, struct ether_header *);
1606 	etype = ntohs(eh->ether_type);
1607 
1608 	/* Tap off LACP control messages */
1609 	if (etype == ETHERTYPE_SLOW) {
1610 		lacp_input(lp, m);
1611 		return (NULL);
1612 	}
1613 
1614 	/*
1615 	 * If the port is not collecting or not in the active aggregator then
1616 	 * free and return.
1617 	 */
1618 	if ((lp->lp_flags & LAGG_PORT_COLLECTING) == 0 ||
1619 	    lacp_port_isactive(lp) == 0) {
1620 		m_freem(m);
1621 		return (NULL);
1622 	}
1623 
1624 	m->m_pkthdr.rcvif = ifp;
1625 	return (m);
1626 }
1627