xref: /freebsd/sys/net/if_lagg.c (revision 6ff6d951ade3f3379932df7f878ef3ea272cfc59)
1 /*	$OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $	*/
2 
3 /*
4  * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org>
5  * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org>
6  *
7  * Permission to use, copy, modify, and distribute this software for any
8  * purpose with or without fee is hereby granted, provided that the above
9  * copyright notice and this permission notice appear in all copies.
10  *
11  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18  */
19 
20 #include <sys/cdefs.h>
21 __FBSDID("$FreeBSD$");
22 
23 #include "opt_inet.h"
24 #include "opt_inet6.h"
25 
26 #include <sys/param.h>
27 #include <sys/kernel.h>
28 #include <sys/malloc.h>
29 #include <sys/mbuf.h>
30 #include <sys/queue.h>
31 #include <sys/socket.h>
32 #include <sys/sockio.h>
33 #include <sys/sysctl.h>
34 #include <sys/module.h>
35 #include <sys/priv.h>
36 #include <sys/systm.h>
37 #include <sys/proc.h>
38 #include <sys/hash.h>
39 #include <sys/lock.h>
40 #include <sys/rwlock.h>
41 #include <sys/taskqueue.h>
42 
43 #include <net/ethernet.h>
44 #include <net/if.h>
45 #include <net/if_clone.h>
46 #include <net/if_arp.h>
47 #include <net/if_dl.h>
48 #include <net/if_llc.h>
49 #include <net/if_media.h>
50 #include <net/if_types.h>
51 #include <net/if_var.h>
52 #include <net/bpf.h>
53 
54 #ifdef INET
55 #include <netinet/in.h>
56 #include <netinet/in_systm.h>
57 #include <netinet/if_ether.h>
58 #include <netinet/ip.h>
59 #endif
60 
61 #ifdef INET6
62 #include <netinet/ip6.h>
63 #endif
64 
65 #include <net/if_vlan_var.h>
66 #include <net/if_lagg.h>
67 #include <net/ieee8023ad_lacp.h>
68 
69 /* Special flags we should propagate to the lagg ports. */
70 static struct {
71 	int flag;
72 	int (*func)(struct ifnet *, int);
73 } lagg_pflags[] = {
74 	{IFF_PROMISC, ifpromisc},
75 	{IFF_ALLMULTI, if_allmulti},
76 	{0, NULL}
77 };
78 
79 SLIST_HEAD(__trhead, lagg_softc) lagg_list;	/* list of laggs */
80 static struct mtx	lagg_list_mtx;
81 eventhandler_tag	lagg_detach_cookie = NULL;
82 
83 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
84 static void	lagg_clone_destroy(struct ifnet *);
85 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
86 static void	lagg_capabilities(struct lagg_softc *);
87 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
88 static void	lagg_port_setlladdr(void *, int);
89 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
90 static int	lagg_port_destroy(struct lagg_port *, int);
91 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
92 static void	lagg_linkstate(struct lagg_softc *);
93 static void	lagg_port_state(struct ifnet *, int);
94 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
95 static int	lagg_port_output(struct ifnet *, struct mbuf *,
96 		    struct sockaddr *, struct rtentry *);
97 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
98 static int	lagg_port_checkstacking(struct lagg_softc *);
99 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
100 static void	lagg_init(void *);
101 static void	lagg_stop(struct lagg_softc *);
102 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
103 static int	lagg_ether_setmulti(struct lagg_softc *);
104 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
105 static	int	lagg_setflag(struct lagg_port *, int, int,
106 		    int (*func)(struct ifnet *, int));
107 static	int	lagg_setflags(struct lagg_port *, int status);
108 static void	lagg_start(struct ifnet *);
109 static int	lagg_media_change(struct ifnet *);
110 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
111 static struct lagg_port *lagg_link_active(struct lagg_softc *,
112 	    struct lagg_port *);
113 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
114 
115 IFC_SIMPLE_DECLARE(lagg, 0);
116 
117 /* Simple round robin */
118 static int	lagg_rr_attach(struct lagg_softc *);
119 static int	lagg_rr_detach(struct lagg_softc *);
120 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
121 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
122 		    struct mbuf *);
123 
124 /* Active failover */
125 static int	lagg_fail_attach(struct lagg_softc *);
126 static int	lagg_fail_detach(struct lagg_softc *);
127 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
128 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
129 		    struct mbuf *);
130 
131 /* Loadbalancing */
132 static int	lagg_lb_attach(struct lagg_softc *);
133 static int	lagg_lb_detach(struct lagg_softc *);
134 static int	lagg_lb_port_create(struct lagg_port *);
135 static void	lagg_lb_port_destroy(struct lagg_port *);
136 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
137 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
138 		    struct mbuf *);
139 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
140 
141 /* 802.3ad LACP */
142 static int	lagg_lacp_attach(struct lagg_softc *);
143 static int	lagg_lacp_detach(struct lagg_softc *);
144 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
145 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
146 		    struct mbuf *);
147 static void	lagg_lacp_lladdr(struct lagg_softc *);
148 
149 /* lagg protocol table */
150 static const struct {
151 	int			ti_proto;
152 	int			(*ti_attach)(struct lagg_softc *);
153 } lagg_protos[] = {
154 	{ LAGG_PROTO_ROUNDROBIN,	lagg_rr_attach },
155 	{ LAGG_PROTO_FAILOVER,		lagg_fail_attach },
156 	{ LAGG_PROTO_LOADBALANCE,	lagg_lb_attach },
157 	{ LAGG_PROTO_ETHERCHANNEL,	lagg_lb_attach },
158 	{ LAGG_PROTO_LACP,		lagg_lacp_attach },
159 	{ LAGG_PROTO_NONE,		NULL }
160 };
161 
162 static int
163 lagg_modevent(module_t mod, int type, void *data)
164 {
165 
166 	switch (type) {
167 	case MOD_LOAD:
168 		mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
169 		SLIST_INIT(&lagg_list);
170 		if_clone_attach(&lagg_cloner);
171 		lagg_input_p = lagg_input;
172 		lagg_linkstate_p = lagg_port_state;
173 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
174 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
175 		    EVENTHANDLER_PRI_ANY);
176 		break;
177 	case MOD_UNLOAD:
178 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
179 		    lagg_detach_cookie);
180 		if_clone_detach(&lagg_cloner);
181 		lagg_input_p = NULL;
182 		lagg_linkstate_p = NULL;
183 		mtx_destroy(&lagg_list_mtx);
184 		break;
185 	default:
186 		return (EOPNOTSUPP);
187 	}
188 	return (0);
189 }
190 
191 static moduledata_t lagg_mod = {
192 	"if_lagg",
193 	lagg_modevent,
194 	0
195 };
196 
197 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
198 
199 static int
200 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
201 {
202 	struct lagg_softc *sc;
203 	struct ifnet *ifp;
204 	int i, error = 0;
205 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
206 
207 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
208 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
209 	if (ifp == NULL) {
210 		free(sc, M_DEVBUF);
211 		return (ENOSPC);
212 	}
213 
214 	sc->sc_proto = LAGG_PROTO_NONE;
215 	for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
216 		if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
217 			sc->sc_proto = lagg_protos[i].ti_proto;
218 			if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
219 				if_free_type(ifp, IFT_ETHER);
220 				free(sc, M_DEVBUF);
221 				return (error);
222 			}
223 			break;
224 		}
225 	}
226 	LAGG_LOCK_INIT(sc);
227 	SLIST_INIT(&sc->sc_ports);
228 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
229 
230 	/* Initialise pseudo media types */
231 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
232 	    lagg_media_status);
233 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
234 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
235 
236 	if_initname(ifp, ifc->ifc_name, unit);
237 	ifp->if_type = IFT_ETHER;
238 	ifp->if_softc = sc;
239 	ifp->if_start = lagg_start;
240 	ifp->if_init = lagg_init;
241 	ifp->if_ioctl = lagg_ioctl;
242 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
243 
244 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
245 	ifp->if_snd.ifq_drv_maxlen = ifqmaxlen;
246 	IFQ_SET_READY(&ifp->if_snd);
247 
248 	/*
249 	 * Attach as an ordinary ethernet device, childs will be attached
250 	 * as special device IFT_IEEE8023ADLAG.
251 	 */
252 	ether_ifattach(ifp, eaddr);
253 
254 	/* Insert into the global list of laggs */
255 	mtx_lock(&lagg_list_mtx);
256 	SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
257 	mtx_unlock(&lagg_list_mtx);
258 
259 	return (0);
260 }
261 
262 static void
263 lagg_clone_destroy(struct ifnet *ifp)
264 {
265 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
266 	struct lagg_port *lp;
267 
268 	LAGG_WLOCK(sc);
269 
270 	lagg_stop(sc);
271 	ifp->if_flags &= ~IFF_UP;
272 
273 	/* Shutdown and remove lagg ports */
274 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
275 		lagg_port_destroy(lp, 1);
276 	/* Unhook the aggregation protocol */
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 void
309 lagg_capabilities(struct lagg_softc *sc)
310 {
311 	struct lagg_port *lp;
312 	int cap = ~0, ena = ~0;
313 
314 	LAGG_WLOCK_ASSERT(sc);
315 
316 	/* Get capabilities from the lagg ports */
317 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
318 		cap &= lp->lp_ifp->if_capabilities;
319 		ena &= lp->lp_ifp->if_capenable;
320 	}
321 	cap = (cap == ~0 ? 0 : cap);
322 	ena = (ena == ~0 ? 0 : ena);
323 
324 	if (sc->sc_ifp->if_capabilities != cap ||
325 	    sc->sc_ifp->if_capenable != ena) {
326 		sc->sc_ifp->if_capabilities = cap;
327 		sc->sc_ifp->if_capenable = ena;
328 		getmicrotime(&sc->sc_ifp->if_lastchange);
329 
330 		if (sc->sc_ifflags & IFF_DEBUG)
331 			if_printf(sc->sc_ifp,
332 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
333 	}
334 }
335 
336 static void
337 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
338 {
339 	struct lagg_softc *sc = lp->lp_softc;
340 	struct ifnet *ifp = lp->lp_ifp;
341 	struct lagg_llq *llq;
342 	int pending = 0;
343 
344 	LAGG_WLOCK_ASSERT(sc);
345 
346 	if (lp->lp_detaching ||
347 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
348 		return;
349 
350 	/* Check to make sure its not already queued to be changed */
351 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
352 		if (llq->llq_ifp == ifp) {
353 			pending = 1;
354 			break;
355 		}
356 	}
357 
358 	if (!pending) {
359 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
360 		if (llq == NULL)	/* XXX what to do */
361 			return;
362 	}
363 
364 	/* Update the lladdr even if pending, it may have changed */
365 	llq->llq_ifp = ifp;
366 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
367 
368 	if (!pending)
369 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
370 
371 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
372 }
373 
374 /*
375  * Set the interface MAC address from a taskqueue to avoid a LOR.
376  */
377 static void
378 lagg_port_setlladdr(void *arg, int pending)
379 {
380 	struct lagg_softc *sc = (struct lagg_softc *)arg;
381 	struct lagg_llq *llq, *head;
382 	struct ifnet *ifp;
383 	int error;
384 
385 	/* Grab a local reference of the queue and remove it from the softc */
386 	LAGG_WLOCK(sc);
387 	head = SLIST_FIRST(&sc->sc_llq_head);
388 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
389 	LAGG_WUNLOCK(sc);
390 
391 	/*
392 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
393 	 * unlocked as we have the only reference to it.
394 	 */
395 	for (llq = head; llq != NULL; llq = head) {
396 		ifp = llq->llq_ifp;
397 
398 		/* Set the link layer address */
399 		error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
400 		if (error)
401 			printf("%s: setlladdr failed on %s\n", __func__,
402 			    ifp->if_xname);
403 
404 		head = SLIST_NEXT(llq, llq_entries);
405 		free(llq, M_DEVBUF);
406 	}
407 }
408 
409 static int
410 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
411 {
412 	struct lagg_softc *sc_ptr;
413 	struct lagg_port *lp;
414 	int error = 0;
415 
416 	LAGG_WLOCK_ASSERT(sc);
417 
418 	/* Limit the maximal number of lagg ports */
419 	if (sc->sc_count >= LAGG_MAX_PORTS)
420 		return (ENOSPC);
421 
422 	/* New lagg port has to be in an idle state */
423 	if (ifp->if_drv_flags & IFF_DRV_OACTIVE)
424 		return (EBUSY);
425 
426 	/* Check if port has already been associated to a lagg */
427 	if (ifp->if_lagg != NULL)
428 		return (EBUSY);
429 
430 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
431 	if (ifp->if_type != IFT_ETHER)
432 		return (EPROTONOSUPPORT);
433 
434 	/* Allow the first Ethernet member to define the MTU */
435 	if (SLIST_EMPTY(&sc->sc_ports))
436 		sc->sc_ifp->if_mtu = ifp->if_mtu;
437 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
438 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
439 		    ifp->if_xname);
440 		return (EINVAL);
441 	}
442 
443 	if ((lp = malloc(sizeof(struct lagg_port),
444 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
445 		return (ENOMEM);
446 
447 	/* Check if port is a stacked lagg */
448 	mtx_lock(&lagg_list_mtx);
449 	SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
450 		if (ifp == sc_ptr->sc_ifp) {
451 			mtx_unlock(&lagg_list_mtx);
452 			free(lp, M_DEVBUF);
453 			return (EINVAL);
454 			/* XXX disable stacking for the moment, its untested
455 			lp->lp_flags |= LAGG_PORT_STACK;
456 			if (lagg_port_checkstacking(sc_ptr) >=
457 			    LAGG_MAX_STACKING) {
458 				mtx_unlock(&lagg_list_mtx);
459 				free(lp, M_DEVBUF);
460 				return (E2BIG);
461 			}
462 			*/
463 		}
464 	}
465 	mtx_unlock(&lagg_list_mtx);
466 
467 	/* Change the interface type */
468 	lp->lp_iftype = ifp->if_type;
469 	ifp->if_type = IFT_IEEE8023ADLAG;
470 	ifp->if_lagg = lp;
471 	lp->lp_ioctl = ifp->if_ioctl;
472 	ifp->if_ioctl = lagg_port_ioctl;
473 	lp->lp_output = ifp->if_output;
474 	ifp->if_output = lagg_port_output;
475 
476 	lp->lp_ifp = ifp;
477 	lp->lp_softc = sc;
478 
479 	/* Save port link layer address */
480 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
481 
482 	if (SLIST_EMPTY(&sc->sc_ports)) {
483 		sc->sc_primary = lp;
484 		lagg_lladdr(sc, IF_LLADDR(ifp));
485 	} else {
486 		/* Update link layer address for this port */
487 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
488 	}
489 
490 	/* Insert into the list of ports */
491 	SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
492 	sc->sc_count++;
493 
494 	/* Update lagg capabilities */
495 	lagg_capabilities(sc);
496 	lagg_linkstate(sc);
497 
498 	/* Add multicast addresses and interface flags to this port */
499 	lagg_ether_cmdmulti(lp, 1);
500 	lagg_setflags(lp, 1);
501 
502 	if (sc->sc_port_create != NULL)
503 		error = (*sc->sc_port_create)(lp);
504 	if (error) {
505 		/* remove the port again, without calling sc_port_destroy */
506 		lagg_port_destroy(lp, 0);
507 		return (error);
508 	}
509 
510 	return (error);
511 }
512 
513 static int
514 lagg_port_checkstacking(struct lagg_softc *sc)
515 {
516 	struct lagg_softc *sc_ptr;
517 	struct lagg_port *lp;
518 	int m = 0;
519 
520 	LAGG_WLOCK_ASSERT(sc);
521 
522 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
523 		if (lp->lp_flags & LAGG_PORT_STACK) {
524 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
525 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
526 		}
527 	}
528 
529 	return (m + 1);
530 }
531 
532 static int
533 lagg_port_destroy(struct lagg_port *lp, int runpd)
534 {
535 	struct lagg_softc *sc = lp->lp_softc;
536 	struct lagg_port *lp_ptr;
537 	struct lagg_llq *llq;
538 	struct ifnet *ifp = lp->lp_ifp;
539 
540 	LAGG_WLOCK_ASSERT(sc);
541 
542 	if (runpd && sc->sc_port_destroy != NULL)
543 		(*sc->sc_port_destroy)(lp);
544 
545 	/*
546 	 * Remove multicast addresses and interface flags from this port and
547 	 * reset the MAC address, skip if the interface is being detached.
548 	 */
549 	if (!lp->lp_detaching) {
550 		lagg_ether_cmdmulti(lp, 0);
551 		lagg_setflags(lp, 0);
552 		lagg_port_lladdr(lp, lp->lp_lladdr);
553 	}
554 
555 	/* Restore interface */
556 	ifp->if_type = lp->lp_iftype;
557 	ifp->if_ioctl = lp->lp_ioctl;
558 	ifp->if_output = lp->lp_output;
559 	ifp->if_lagg = NULL;
560 
561 	/* Finally, remove the port from the lagg */
562 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
563 	sc->sc_count--;
564 
565 	/* Update the primary interface */
566 	if (lp == sc->sc_primary) {
567 		uint8_t lladdr[ETHER_ADDR_LEN];
568 
569 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
570 			bzero(&lladdr, ETHER_ADDR_LEN);
571 		} else {
572 			bcopy(lp_ptr->lp_lladdr,
573 			    lladdr, ETHER_ADDR_LEN);
574 		}
575 		lagg_lladdr(sc, lladdr);
576 		sc->sc_primary = lp_ptr;
577 
578 		/* Update link layer address for each port */
579 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
580 			lagg_port_lladdr(lp_ptr, lladdr);
581 	}
582 
583 	/* Remove any pending lladdr changes from the queue */
584 	if (lp->lp_detaching) {
585 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
586 			if (llq->llq_ifp == ifp) {
587 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
588 				    llq_entries);
589 				free(llq, M_DEVBUF);
590 				break;	/* Only appears once */
591 			}
592 		}
593 	}
594 
595 	if (lp->lp_ifflags)
596 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
597 
598 	free(lp, M_DEVBUF);
599 
600 	/* Update lagg capabilities */
601 	lagg_capabilities(sc);
602 	lagg_linkstate(sc);
603 
604 	return (0);
605 }
606 
607 static int
608 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
609 {
610 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
611 	struct lagg_softc *sc;
612 	struct lagg_port *lp = NULL;
613 	int error = 0;
614 
615 	/* Should be checked by the caller */
616 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
617 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
618 		goto fallback;
619 
620 	switch (cmd) {
621 	case SIOCGLAGGPORT:
622 		if (rp->rp_portname[0] == '\0' ||
623 		    ifunit(rp->rp_portname) != ifp) {
624 			error = EINVAL;
625 			break;
626 		}
627 
628 		LAGG_RLOCK(sc);
629 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
630 			error = ENOENT;
631 			LAGG_RUNLOCK(sc);
632 			break;
633 		}
634 
635 		lagg_port2req(lp, rp);
636 		LAGG_RUNLOCK(sc);
637 		break;
638 
639 	case SIOCSIFCAP:
640 		if (lp->lp_ioctl == NULL) {
641 			error = EINVAL;
642 			break;
643 		}
644 		error = (*lp->lp_ioctl)(ifp, cmd, data);
645 		if (error)
646 			break;
647 
648 		/* Update lagg interface capabilities */
649 		LAGG_WLOCK(sc);
650 		lagg_capabilities(sc);
651 		LAGG_WUNLOCK(sc);
652 		break;
653 
654 	case SIOCSIFMTU:
655 		/* Do not allow the MTU to be changed once joined */
656 		error = EINVAL;
657 		break;
658 
659 	default:
660 		goto fallback;
661 	}
662 
663 	return (error);
664 
665 fallback:
666 	if (lp->lp_ioctl != NULL)
667 		return ((*lp->lp_ioctl)(ifp, cmd, data));
668 
669 	return (EINVAL);
670 }
671 
672 static int
673 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
674 	struct sockaddr *dst, struct rtentry *rt0)
675 {
676 	struct lagg_port *lp = ifp->if_lagg;
677 	struct ether_header *eh;
678 	short type = 0;
679 
680 	switch (dst->sa_family) {
681 		case pseudo_AF_HDRCMPLT:
682 		case AF_UNSPEC:
683 			eh = (struct ether_header *)dst->sa_data;
684 			type = eh->ether_type;
685 			break;
686 	}
687 
688 	/*
689 	 * Only allow ethernet types required to initiate or maintain the link,
690 	 * aggregated frames take a different path.
691 	 */
692 	switch (ntohs(type)) {
693 		case ETHERTYPE_PAE:	/* EAPOL PAE/802.1x */
694 			return ((*lp->lp_output)(ifp, m, dst, rt0));
695 	}
696 
697 	/* drop any other frames */
698 	m_freem(m);
699 	return (EBUSY);
700 }
701 
702 static void
703 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
704 {
705 	struct lagg_port *lp;
706 	struct lagg_softc *sc;
707 
708 	if ((lp = ifp->if_lagg) == NULL)
709 		return;
710 
711 	sc = lp->lp_softc;
712 
713 	LAGG_WLOCK(sc);
714 	lp->lp_detaching = 1;
715 	lagg_port_destroy(lp, 1);
716 	LAGG_WUNLOCK(sc);
717 }
718 
719 static void
720 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
721 {
722 	struct lagg_softc *sc = lp->lp_softc;
723 
724 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
725 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
726 	rp->rp_prio = lp->lp_prio;
727 	rp->rp_flags = lp->lp_flags;
728 	if (sc->sc_portreq != NULL)
729 		(*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
730 
731 	/* Add protocol specific flags */
732 	switch (sc->sc_proto) {
733 		case LAGG_PROTO_FAILOVER:
734 			if (lp == sc->sc_primary)
735 				rp->rp_flags |= LAGG_PORT_MASTER;
736 			if (lp == lagg_link_active(sc, sc->sc_primary))
737 				rp->rp_flags |= LAGG_PORT_ACTIVE;
738 			break;
739 
740 		case LAGG_PROTO_ROUNDROBIN:
741 		case LAGG_PROTO_LOADBALANCE:
742 		case LAGG_PROTO_ETHERCHANNEL:
743 			if (LAGG_PORTACTIVE(lp))
744 				rp->rp_flags |= LAGG_PORT_ACTIVE;
745 			break;
746 
747 		case LAGG_PROTO_LACP:
748 			/* LACP has a different definition of active */
749 			if (lacp_isactive(lp))
750 				rp->rp_flags |= LAGG_PORT_ACTIVE;
751 			if (lacp_iscollecting(lp))
752 				rp->rp_flags |= LAGG_PORT_COLLECTING;
753 			if (lacp_isdistributing(lp))
754 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
755 			break;
756 	}
757 
758 }
759 
760 static void
761 lagg_init(void *xsc)
762 {
763 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
764 	struct lagg_port *lp;
765 	struct ifnet *ifp = sc->sc_ifp;
766 
767 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
768 		return;
769 
770 	LAGG_WLOCK(sc);
771 
772 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
773 	/* Update the port lladdrs */
774 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
775 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
776 
777 	if (sc->sc_init != NULL)
778 		(*sc->sc_init)(sc);
779 
780 	LAGG_WUNLOCK(sc);
781 }
782 
783 static void
784 lagg_stop(struct lagg_softc *sc)
785 {
786 	struct ifnet *ifp = sc->sc_ifp;
787 
788 	LAGG_WLOCK_ASSERT(sc);
789 
790 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
791 		return;
792 
793 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
794 
795 	if (sc->sc_stop != NULL)
796 		(*sc->sc_stop)(sc);
797 }
798 
799 static int
800 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
801 {
802 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
803 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
804 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
805 	struct ifreq *ifr = (struct ifreq *)data;
806 	struct lagg_port *lp;
807 	struct ifnet *tpif;
808 	struct thread *td = curthread;
809 	char *buf, *outbuf;
810 	int count, buflen, len, error = 0;
811 
812 	bzero(&rpbuf, sizeof(rpbuf));
813 
814 	switch (cmd) {
815 	case SIOCGLAGG:
816 		LAGG_RLOCK(sc);
817 		count = 0;
818 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
819 			count++;
820 		buflen = count * sizeof(struct lagg_reqport);
821 		LAGG_RUNLOCK(sc);
822 
823 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
824 
825 		LAGG_RLOCK(sc);
826 		ra->ra_proto = sc->sc_proto;
827 		if (sc->sc_req != NULL)
828 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
829 
830 		count = 0;
831 		buf = outbuf;
832 		len = min(ra->ra_size, buflen);
833 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
834 			if (len < sizeof(rpbuf))
835 				break;
836 
837 			lagg_port2req(lp, &rpbuf);
838 			memcpy(buf, &rpbuf, sizeof(rpbuf));
839 			count++;
840 			buf += sizeof(rpbuf);
841 			len -= sizeof(rpbuf);
842 		}
843 		LAGG_RUNLOCK(sc);
844 		ra->ra_ports = count;
845 		ra->ra_size = count * sizeof(rpbuf);
846 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
847 		free(outbuf, M_TEMP);
848 		break;
849 	case SIOCSLAGG:
850 		error = priv_check(td, PRIV_NET_LAGG);
851 		if (error)
852 			break;
853 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
854 			error = EPROTONOSUPPORT;
855 			break;
856 		}
857 		if (sc->sc_proto != LAGG_PROTO_NONE) {
858 			LAGG_WLOCK(sc);
859 			error = sc->sc_detach(sc);
860 			/* Reset protocol and pointers */
861 			sc->sc_proto = LAGG_PROTO_NONE;
862 			sc->sc_detach = NULL;
863 			sc->sc_start = NULL;
864 			sc->sc_input = NULL;
865 			sc->sc_port_create = NULL;
866 			sc->sc_port_destroy = NULL;
867 			sc->sc_linkstate = NULL;
868 			sc->sc_init = NULL;
869 			sc->sc_stop = NULL;
870 			sc->sc_lladdr = NULL;
871 			sc->sc_req = NULL;
872 			sc->sc_portreq = NULL;
873 			LAGG_WUNLOCK(sc);
874 		}
875 		if (error != 0)
876 			break;
877 		for (int i = 0; i < (sizeof(lagg_protos) /
878 		    sizeof(lagg_protos[0])); i++) {
879 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
880 				if (sc->sc_ifflags & IFF_DEBUG)
881 					printf("%s: using proto %u\n",
882 					    sc->sc_ifname,
883 					    lagg_protos[i].ti_proto);
884 				LAGG_WLOCK(sc);
885 				sc->sc_proto = lagg_protos[i].ti_proto;
886 				if (sc->sc_proto != LAGG_PROTO_NONE)
887 					error = lagg_protos[i].ti_attach(sc);
888 				LAGG_WUNLOCK(sc);
889 				return (error);
890 			}
891 		}
892 		error = EPROTONOSUPPORT;
893 		break;
894 	case SIOCGLAGGPORT:
895 		if (rp->rp_portname[0] == '\0' ||
896 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
897 			error = EINVAL;
898 			break;
899 		}
900 
901 		LAGG_RLOCK(sc);
902 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
903 		    lp->lp_softc != sc) {
904 			error = ENOENT;
905 			LAGG_RUNLOCK(sc);
906 			break;
907 		}
908 
909 		lagg_port2req(lp, rp);
910 		LAGG_RUNLOCK(sc);
911 		break;
912 	case SIOCSLAGGPORT:
913 		error = priv_check(td, PRIV_NET_LAGG);
914 		if (error)
915 			break;
916 		if (rp->rp_portname[0] == '\0' ||
917 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
918 			error = EINVAL;
919 			break;
920 		}
921 		LAGG_WLOCK(sc);
922 		error = lagg_port_create(sc, tpif);
923 		LAGG_WUNLOCK(sc);
924 		break;
925 	case SIOCSLAGGDELPORT:
926 		error = priv_check(td, PRIV_NET_LAGG);
927 		if (error)
928 			break;
929 		if (rp->rp_portname[0] == '\0' ||
930 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
931 			error = EINVAL;
932 			break;
933 		}
934 
935 		LAGG_WLOCK(sc);
936 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
937 		    lp->lp_softc != sc) {
938 			error = ENOENT;
939 			LAGG_WUNLOCK(sc);
940 			break;
941 		}
942 
943 		error = lagg_port_destroy(lp, 1);
944 		LAGG_WUNLOCK(sc);
945 		break;
946 	case SIOCSIFFLAGS:
947 		/* Set flags on ports too */
948 		LAGG_WLOCK(sc);
949 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
950 			lagg_setflags(lp, 1);
951 		}
952 		LAGG_WUNLOCK(sc);
953 
954 		if (!(ifp->if_flags & IFF_UP) &&
955 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
956 			/*
957 			 * If interface is marked down and it is running,
958 			 * then stop and disable it.
959 			 */
960 			LAGG_WLOCK(sc);
961 			lagg_stop(sc);
962 			LAGG_WUNLOCK(sc);
963 		} else if ((ifp->if_flags & IFF_UP) &&
964 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
965 			/*
966 			 * If interface is marked up and it is stopped, then
967 			 * start it.
968 			 */
969 			(*ifp->if_init)(sc);
970 		}
971 		break;
972 	case SIOCADDMULTI:
973 	case SIOCDELMULTI:
974 		LAGG_WLOCK(sc);
975 		error = lagg_ether_setmulti(sc);
976 		LAGG_WUNLOCK(sc);
977 		break;
978 	case SIOCSIFMEDIA:
979 	case SIOCGIFMEDIA:
980 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
981 		break;
982 
983 	case SIOCSIFCAP:
984 	case SIOCSIFMTU:
985 		/* Do not allow the MTU or caps to be directly changed */
986 		error = EINVAL;
987 		break;
988 
989 	default:
990 		error = ether_ioctl(ifp, cmd, data);
991 		break;
992 	}
993 	return (error);
994 }
995 
996 static int
997 lagg_ether_setmulti(struct lagg_softc *sc)
998 {
999 	struct lagg_port *lp;
1000 
1001 	LAGG_WLOCK_ASSERT(sc);
1002 
1003 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1004 		/* First, remove any existing filter entries. */
1005 		lagg_ether_cmdmulti(lp, 0);
1006 		/* copy all addresses from the lagg interface to the port */
1007 		lagg_ether_cmdmulti(lp, 1);
1008 	}
1009 	return (0);
1010 }
1011 
1012 static int
1013 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1014 {
1015 	struct lagg_softc *sc = lp->lp_softc;
1016 	struct ifnet *ifp = lp->lp_ifp;
1017 	struct ifnet *scifp = sc->sc_ifp;
1018 	struct lagg_mc *mc;
1019 	struct ifmultiaddr *ifma, *rifma = NULL;
1020 	struct sockaddr_dl sdl;
1021 	int error;
1022 
1023 	LAGG_WLOCK_ASSERT(sc);
1024 
1025 	bzero((char *)&sdl, sizeof(sdl));
1026 	sdl.sdl_len = sizeof(sdl);
1027 	sdl.sdl_family = AF_LINK;
1028 	sdl.sdl_type = IFT_ETHER;
1029 	sdl.sdl_alen = ETHER_ADDR_LEN;
1030 	sdl.sdl_index = ifp->if_index;
1031 
1032 	if (set) {
1033 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1034 			if (ifma->ifma_addr->sa_family != AF_LINK)
1035 				continue;
1036 			bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1037 			    LLADDR(&sdl), ETHER_ADDR_LEN);
1038 
1039 			error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1040 			if (error)
1041 				return (error);
1042 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1043 			if (mc == NULL)
1044 				return (ENOMEM);
1045 			mc->mc_ifma = rifma;
1046 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1047 		}
1048 	} else {
1049 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1050 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1051 			if_delmulti_ifma(mc->mc_ifma);
1052 			free(mc, M_DEVBUF);
1053 		}
1054 	}
1055 	return (0);
1056 }
1057 
1058 /* Handle a ref counted flag that should be set on the lagg port as well */
1059 static int
1060 lagg_setflag(struct lagg_port *lp, int flag, int status,
1061 	     int (*func)(struct ifnet *, int))
1062 {
1063 	struct lagg_softc *sc = lp->lp_softc;
1064 	struct ifnet *scifp = sc->sc_ifp;
1065 	struct ifnet *ifp = lp->lp_ifp;
1066 	int error;
1067 
1068 	LAGG_WLOCK_ASSERT(sc);
1069 
1070 	status = status ? (scifp->if_flags & flag) : 0;
1071 	/* Now "status" contains the flag value or 0 */
1072 
1073 	/*
1074 	 * See if recorded ports status is different from what
1075 	 * we want it to be.  If it is, flip it.  We record ports
1076 	 * status in lp_ifflags so that we won't clear ports flag
1077 	 * we haven't set.  In fact, we don't clear or set ports
1078 	 * flags directly, but get or release references to them.
1079 	 * That's why we can be sure that recorded flags still are
1080 	 * in accord with actual ports flags.
1081 	 */
1082 	if (status != (lp->lp_ifflags & flag)) {
1083 		error = (*func)(ifp, status);
1084 		if (error)
1085 			return (error);
1086 		lp->lp_ifflags &= ~flag;
1087 		lp->lp_ifflags |= status;
1088 	}
1089 	return (0);
1090 }
1091 
1092 /*
1093  * Handle IFF_* flags that require certain changes on the lagg port
1094  * if "status" is true, update ports flags respective to the lagg
1095  * if "status" is false, forcedly clear the flags set on port.
1096  */
1097 static int
1098 lagg_setflags(struct lagg_port *lp, int status)
1099 {
1100 	int error, i;
1101 
1102 	for (i = 0; lagg_pflags[i].flag; i++) {
1103 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1104 		    status, lagg_pflags[i].func);
1105 		if (error)
1106 			return (error);
1107 	}
1108 	return (0);
1109 }
1110 
1111 static void
1112 lagg_start(struct ifnet *ifp)
1113 {
1114 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1115 	struct mbuf *m;
1116 	int error = 0;
1117 
1118 	LAGG_RLOCK(sc);
1119 	for (;; error = 0) {
1120 		IFQ_DEQUEUE(&ifp->if_snd, m);
1121 		if (m == NULL)
1122 			break;
1123 
1124 		ETHER_BPF_MTAP(ifp, m);
1125 
1126 		if (sc->sc_proto != LAGG_PROTO_NONE)
1127 			error = (*sc->sc_start)(sc, m);
1128 		else
1129 			m_freem(m);
1130 
1131 		if (error == 0)
1132 			ifp->if_opackets++;
1133 		else
1134 			ifp->if_oerrors++;
1135 	}
1136 	LAGG_RUNLOCK(sc);
1137 
1138 	return;
1139 }
1140 
1141 static struct mbuf *
1142 lagg_input(struct ifnet *ifp, struct mbuf *m)
1143 {
1144 	struct lagg_port *lp = ifp->if_lagg;
1145 	struct lagg_softc *sc = lp->lp_softc;
1146 	struct ifnet *scifp = sc->sc_ifp;
1147 
1148 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1149 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1150 	    sc->sc_proto == LAGG_PROTO_NONE) {
1151 		m_freem(m);
1152 		return (NULL);
1153 	}
1154 
1155 	LAGG_RLOCK(sc);
1156 	ETHER_BPF_MTAP(scifp, m);
1157 
1158 	m = (*sc->sc_input)(sc, lp, m);
1159 
1160 	if (m != NULL) {
1161 		scifp->if_ipackets++;
1162 		scifp->if_ibytes += m->m_pkthdr.len;
1163 
1164 		if (scifp->if_flags & IFF_MONITOR) {
1165 			m_freem(m);
1166 			m = NULL;
1167 		}
1168 	}
1169 
1170 	LAGG_RUNLOCK(sc);
1171 	return (m);
1172 }
1173 
1174 static int
1175 lagg_media_change(struct ifnet *ifp)
1176 {
1177 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1178 
1179 	if (sc->sc_ifflags & IFF_DEBUG)
1180 		printf("%s\n", __func__);
1181 
1182 	/* Ignore */
1183 	return (0);
1184 }
1185 
1186 static void
1187 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1188 {
1189 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1190 	struct lagg_port *lp;
1191 
1192 	imr->ifm_status = IFM_AVALID;
1193 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1194 
1195 	LAGG_RLOCK(sc);
1196 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1197 		if (LAGG_PORTACTIVE(lp))
1198 			imr->ifm_status |= IFM_ACTIVE;
1199 	}
1200 	LAGG_RUNLOCK(sc);
1201 }
1202 
1203 static void
1204 lagg_linkstate(struct lagg_softc *sc)
1205 {
1206 	struct lagg_port *lp;
1207 	int new_link = LINK_STATE_DOWN;
1208 
1209 	/* Our link is considered up if at least one of our ports is active */
1210 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1211 		if (lp->lp_link_state == LINK_STATE_UP) {
1212 			new_link = LINK_STATE_UP;
1213 			break;
1214 		}
1215 	}
1216 	if_link_state_change(sc->sc_ifp, new_link);
1217 }
1218 
1219 static void
1220 lagg_port_state(struct ifnet *ifp, int state)
1221 {
1222 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1223 	struct lagg_softc *sc = NULL;
1224 
1225 	if (lp != NULL)
1226 		sc = lp->lp_softc;
1227 	if (sc == NULL)
1228 		return;
1229 
1230 	LAGG_WLOCK(sc);
1231 	lagg_linkstate(sc);
1232 	if (sc->sc_linkstate != NULL)
1233 		(*sc->sc_linkstate)(lp);
1234 	LAGG_WUNLOCK(sc);
1235 }
1236 
1237 struct lagg_port *
1238 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1239 {
1240 	struct lagg_port *lp_next, *rval = NULL;
1241 	// int new_link = LINK_STATE_DOWN;
1242 
1243 	LAGG_RLOCK_ASSERT(sc);
1244 	/*
1245 	 * Search a port which reports an active link state.
1246 	 */
1247 
1248 	if (lp == NULL)
1249 		goto search;
1250 	if (LAGG_PORTACTIVE(lp)) {
1251 		rval = lp;
1252 		goto found;
1253 	}
1254 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1255 	    LAGG_PORTACTIVE(lp_next)) {
1256 		rval = lp_next;
1257 		goto found;
1258 	}
1259 
1260 search:
1261 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1262 		if (LAGG_PORTACTIVE(lp_next)) {
1263 			rval = lp_next;
1264 			goto found;
1265 		}
1266 	}
1267 
1268 found:
1269 	if (rval != NULL) {
1270 		/*
1271 		 * The IEEE 802.1D standard assumes that a lagg with
1272 		 * multiple ports is always full duplex. This is valid
1273 		 * for load sharing laggs and if at least two links
1274 		 * are active. Unfortunately, checking the latter would
1275 		 * be too expensive at this point.
1276 		 XXX
1277 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1278 		    (sc->sc_count > 1))
1279 			new_link = LINK_STATE_FULL_DUPLEX;
1280 		else
1281 			new_link = rval->lp_link_state;
1282 		 */
1283 	}
1284 
1285 	return (rval);
1286 }
1287 
1288 static const void *
1289 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1290 {
1291 	if (m->m_pkthdr.len < (off + len)) {
1292 		return (NULL);
1293 	} else if (m->m_len < (off + len)) {
1294 		m_copydata(m, off, len, buf);
1295 		return (buf);
1296 	}
1297 	return (mtod(m, char *) + off);
1298 }
1299 
1300 uint32_t
1301 lagg_hashmbuf(struct mbuf *m, uint32_t key)
1302 {
1303 	uint16_t etype;
1304 	uint32_t p = 0;
1305 	int off;
1306 	struct ether_header *eh;
1307 	struct ether_vlan_header vlanbuf;
1308 	const struct ether_vlan_header *vlan;
1309 #ifdef INET
1310 	const struct ip *ip;
1311 	struct ip ipbuf;
1312 #endif
1313 #ifdef INET6
1314 	const struct ip6_hdr *ip6;
1315 	struct ip6_hdr ip6buf;
1316 	uint32_t flow;
1317 #endif
1318 
1319 	off = sizeof(*eh);
1320 	if (m->m_len < off)
1321 		goto out;
1322 	eh = mtod(m, struct ether_header *);
1323 	etype = ntohs(eh->ether_type);
1324 	p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, key);
1325 	p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1326 
1327 	/* Special handling for encapsulating VLAN frames */
1328 	if (m->m_flags & M_VLANTAG) {
1329 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1330 		    sizeof(m->m_pkthdr.ether_vtag), p);
1331 	} else if (etype == ETHERTYPE_VLAN) {
1332 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &vlanbuf);
1333 		if (vlan == NULL)
1334 			goto out;
1335 
1336 		p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1337 		etype = ntohs(vlan->evl_proto);
1338 		off += sizeof(*vlan) - sizeof(*eh);
1339 	}
1340 
1341 	switch (etype) {
1342 #ifdef INET
1343 	case ETHERTYPE_IP:
1344 		ip = lagg_gethdr(m, off, sizeof(*ip), &ipbuf);
1345 		if (ip == NULL)
1346 			goto out;
1347 
1348 		p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1349 		p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1350 		break;
1351 #endif
1352 #ifdef INET6
1353 	case ETHERTYPE_IPV6:
1354 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &ip6buf);
1355 		if (ip6 == NULL)
1356 			goto out;
1357 
1358 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1359 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1360 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1361 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1362 		break;
1363 #endif
1364 	}
1365 out:
1366 	return (p);
1367 }
1368 
1369 int
1370 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1371 {
1372 	int error = 0;
1373 
1374 	IFQ_HANDOFF(ifp, m, error);
1375 	if (error)
1376 		ifp->if_oerrors++;
1377 	return (error);
1378 }
1379 
1380 /*
1381  * Simple round robin aggregation
1382  */
1383 
1384 static int
1385 lagg_rr_attach(struct lagg_softc *sc)
1386 {
1387 	sc->sc_detach = lagg_rr_detach;
1388 	sc->sc_start = lagg_rr_start;
1389 	sc->sc_input = lagg_rr_input;
1390 	sc->sc_port_create = NULL;
1391 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1392 	sc->sc_seq = 0;
1393 
1394 	return (0);
1395 }
1396 
1397 static int
1398 lagg_rr_detach(struct lagg_softc *sc)
1399 {
1400 	return (0);
1401 }
1402 
1403 static int
1404 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1405 {
1406 	struct lagg_port *lp;
1407 	uint32_t p;
1408 
1409 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1410 	p %= sc->sc_count;
1411 	lp = SLIST_FIRST(&sc->sc_ports);
1412 	while (p--)
1413 		lp = SLIST_NEXT(lp, lp_entries);
1414 
1415 	/*
1416 	 * Check the port's link state. This will return the next active
1417 	 * port if the link is down or the port is NULL.
1418 	 */
1419 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1420 		m_freem(m);
1421 		return (ENOENT);
1422 	}
1423 
1424 	/* Send mbuf */
1425 	return (lagg_enqueue(lp->lp_ifp, m));
1426 }
1427 
1428 static struct mbuf *
1429 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1430 {
1431 	struct ifnet *ifp = sc->sc_ifp;
1432 
1433 	/* Just pass in the packet to our lagg device */
1434 	m->m_pkthdr.rcvif = ifp;
1435 
1436 	return (m);
1437 }
1438 
1439 /*
1440  * Active failover
1441  */
1442 
1443 static int
1444 lagg_fail_attach(struct lagg_softc *sc)
1445 {
1446 	sc->sc_detach = lagg_fail_detach;
1447 	sc->sc_start = lagg_fail_start;
1448 	sc->sc_input = lagg_fail_input;
1449 	sc->sc_port_create = NULL;
1450 	sc->sc_port_destroy = NULL;
1451 
1452 	return (0);
1453 }
1454 
1455 static int
1456 lagg_fail_detach(struct lagg_softc *sc)
1457 {
1458 	return (0);
1459 }
1460 
1461 static int
1462 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1463 {
1464 	struct lagg_port *lp;
1465 
1466 	/* Use the master port if active or the next available port */
1467 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1468 		m_freem(m);
1469 		return (ENOENT);
1470 	}
1471 
1472 	/* Send mbuf */
1473 	return (lagg_enqueue(lp->lp_ifp, m));
1474 }
1475 
1476 static struct mbuf *
1477 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1478 {
1479 	struct ifnet *ifp = sc->sc_ifp;
1480 	struct lagg_port *tmp_tp;
1481 
1482 	if (lp == sc->sc_primary) {
1483 		m->m_pkthdr.rcvif = ifp;
1484 		return (m);
1485 	}
1486 
1487 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1488 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1489 		/*
1490 		 * If tmp_tp is null, we've recieved a packet when all
1491 		 * our links are down. Weird, but process it anyways.
1492 		 */
1493 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1494 			m->m_pkthdr.rcvif = ifp;
1495 			return (m);
1496 		}
1497 	}
1498 
1499 	m_freem(m);
1500 	return (NULL);
1501 }
1502 
1503 /*
1504  * Loadbalancing
1505  */
1506 
1507 static int
1508 lagg_lb_attach(struct lagg_softc *sc)
1509 {
1510 	struct lagg_port *lp;
1511 	struct lagg_lb *lb;
1512 
1513 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1514 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1515 		return (ENOMEM);
1516 
1517 	sc->sc_detach = lagg_lb_detach;
1518 	sc->sc_start = lagg_lb_start;
1519 	sc->sc_input = lagg_lb_input;
1520 	sc->sc_port_create = lagg_lb_port_create;
1521 	sc->sc_port_destroy = lagg_lb_port_destroy;
1522 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1523 
1524 	lb->lb_key = arc4random();
1525 	sc->sc_psc = (caddr_t)lb;
1526 
1527 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1528 		lagg_lb_port_create(lp);
1529 
1530 	return (0);
1531 }
1532 
1533 static int
1534 lagg_lb_detach(struct lagg_softc *sc)
1535 {
1536 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1537 	if (lb != NULL)
1538 		free(lb, M_DEVBUF);
1539 	return (0);
1540 }
1541 
1542 static int
1543 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1544 {
1545 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1546 	struct lagg_port *lp_next;
1547 	int i = 0;
1548 
1549 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1550 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1551 		if (lp_next == lp)
1552 			continue;
1553 		if (i >= LAGG_MAX_PORTS)
1554 			return (EINVAL);
1555 		if (sc->sc_ifflags & IFF_DEBUG)
1556 			printf("%s: port %s at index %d\n",
1557 			    sc->sc_ifname, lp_next->lp_ifname, i);
1558 		lb->lb_ports[i++] = lp_next;
1559 	}
1560 
1561 	return (0);
1562 }
1563 
1564 static int
1565 lagg_lb_port_create(struct lagg_port *lp)
1566 {
1567 	struct lagg_softc *sc = lp->lp_softc;
1568 	return (lagg_lb_porttable(sc, NULL));
1569 }
1570 
1571 static void
1572 lagg_lb_port_destroy(struct lagg_port *lp)
1573 {
1574 	struct lagg_softc *sc = lp->lp_softc;
1575 	lagg_lb_porttable(sc, lp);
1576 }
1577 
1578 static int
1579 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1580 {
1581 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1582 	struct lagg_port *lp = NULL;
1583 	uint32_t p = 0;
1584 	int idx;
1585 
1586 	p = lagg_hashmbuf(m, lb->lb_key);
1587 	if ((idx = p % sc->sc_count) >= LAGG_MAX_PORTS) {
1588 		m_freem(m);
1589 		return (EINVAL);
1590 	}
1591 	lp = lb->lb_ports[idx];
1592 
1593 	/*
1594 	 * Check the port's link state. This will return the next active
1595 	 * port if the link is down or the port is NULL.
1596 	 */
1597 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1598 		m_freem(m);
1599 		return (ENOENT);
1600 	}
1601 
1602 	/* Send mbuf */
1603 	return (lagg_enqueue(lp->lp_ifp, m));
1604 }
1605 
1606 static struct mbuf *
1607 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1608 {
1609 	struct ifnet *ifp = sc->sc_ifp;
1610 
1611 	/* Just pass in the packet to our lagg device */
1612 	m->m_pkthdr.rcvif = ifp;
1613 
1614 	return (m);
1615 }
1616 
1617 /*
1618  * 802.3ad LACP
1619  */
1620 
1621 static int
1622 lagg_lacp_attach(struct lagg_softc *sc)
1623 {
1624 	struct lagg_port *lp;
1625 	int error;
1626 
1627 	sc->sc_detach = lagg_lacp_detach;
1628 	sc->sc_port_create = lacp_port_create;
1629 	sc->sc_port_destroy = lacp_port_destroy;
1630 	sc->sc_linkstate = lacp_linkstate;
1631 	sc->sc_start = lagg_lacp_start;
1632 	sc->sc_input = lagg_lacp_input;
1633 	sc->sc_init = lacp_init;
1634 	sc->sc_stop = lacp_stop;
1635 	sc->sc_lladdr = lagg_lacp_lladdr;
1636 	sc->sc_req = lacp_req;
1637 	sc->sc_portreq = lacp_portreq;
1638 
1639 	error = lacp_attach(sc);
1640 	if (error)
1641 		return (error);
1642 
1643 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1644 		lacp_port_create(lp);
1645 
1646 	return (error);
1647 }
1648 
1649 static int
1650 lagg_lacp_detach(struct lagg_softc *sc)
1651 {
1652 	struct lagg_port *lp;
1653 	int error;
1654 
1655 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1656 		lacp_port_destroy(lp);
1657 
1658 	/* unlocking is safe here */
1659 	LAGG_WUNLOCK(sc);
1660 	error = lacp_detach(sc);
1661 	LAGG_WLOCK(sc);
1662 
1663 	return (error);
1664 }
1665 
1666 static void
1667 lagg_lacp_lladdr(struct lagg_softc *sc)
1668 {
1669 	struct lagg_port *lp;
1670 
1671 	/* purge all the lacp ports */
1672 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1673 		lacp_port_destroy(lp);
1674 
1675 	/* add them back in */
1676 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1677 		lacp_port_create(lp);
1678 }
1679 
1680 static int
1681 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1682 {
1683 	struct lagg_port *lp;
1684 
1685 	lp = lacp_select_tx_port(sc, m);
1686 	if (lp == NULL) {
1687 		m_freem(m);
1688 		return (EBUSY);
1689 	}
1690 
1691 	/* Send mbuf */
1692 	return (lagg_enqueue(lp->lp_ifp, m));
1693 }
1694 
1695 static struct mbuf *
1696 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1697 {
1698 	struct ifnet *ifp = sc->sc_ifp;
1699 	struct ether_header *eh;
1700 	u_short etype;
1701 
1702 	eh = mtod(m, struct ether_header *);
1703 	etype = ntohs(eh->ether_type);
1704 
1705 	/* Tap off LACP control messages */
1706 	if (etype == ETHERTYPE_SLOW) {
1707 		m = lacp_input(lp, m);
1708 		if (m == NULL)
1709 			return (NULL);
1710 	}
1711 
1712 	/*
1713 	 * If the port is not collecting or not in the active aggregator then
1714 	 * free and return.
1715 	 */
1716 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
1717 		m_freem(m);
1718 		return (NULL);
1719 	}
1720 
1721 	m->m_pkthdr.rcvif = ifp;
1722 	return (m);
1723 }
1724