xref: /freebsd/sys/net/if_lagg.c (revision 6472ac3d8a86336899b6cfb789a4cd9897e3fab5)
1 /*	$OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $	*/
2 
3 /*
4  * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org>
5  * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org>
6  *
7  * Permission to use, copy, modify, and distribute this software for any
8  * purpose with or without fee is hereby granted, provided that the above
9  * copyright notice and this permission notice appear in all copies.
10  *
11  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18  */
19 
20 #include <sys/cdefs.h>
21 __FBSDID("$FreeBSD$");
22 
23 #include "opt_inet.h"
24 #include "opt_inet6.h"
25 
26 #include <sys/param.h>
27 #include <sys/kernel.h>
28 #include <sys/malloc.h>
29 #include <sys/mbuf.h>
30 #include <sys/queue.h>
31 #include <sys/socket.h>
32 #include <sys/sockio.h>
33 #include <sys/sysctl.h>
34 #include <sys/module.h>
35 #include <sys/priv.h>
36 #include <sys/systm.h>
37 #include <sys/proc.h>
38 #include <sys/hash.h>
39 #include <sys/lock.h>
40 #include <sys/rwlock.h>
41 #include <sys/taskqueue.h>
42 #include <sys/eventhandler.h>
43 
44 #include <net/ethernet.h>
45 #include <net/if.h>
46 #include <net/if_clone.h>
47 #include <net/if_arp.h>
48 #include <net/if_dl.h>
49 #include <net/if_llc.h>
50 #include <net/if_media.h>
51 #include <net/if_types.h>
52 #include <net/if_var.h>
53 #include <net/bpf.h>
54 
55 #if defined(INET) || defined(INET6)
56 #include <netinet/in.h>
57 #endif
58 #ifdef INET
59 #include <netinet/in_systm.h>
60 #include <netinet/if_ether.h>
61 #include <netinet/ip.h>
62 #endif
63 
64 #ifdef INET6
65 #include <netinet/ip6.h>
66 #endif
67 
68 #include <net/if_vlan_var.h>
69 #include <net/if_lagg.h>
70 #include <net/ieee8023ad_lacp.h>
71 
72 /* Special flags we should propagate to the lagg ports. */
73 static struct {
74 	int flag;
75 	int (*func)(struct ifnet *, int);
76 } lagg_pflags[] = {
77 	{IFF_PROMISC, ifpromisc},
78 	{IFF_ALLMULTI, if_allmulti},
79 	{0, NULL}
80 };
81 
82 SLIST_HEAD(__trhead, lagg_softc) lagg_list;	/* list of laggs */
83 static struct mtx	lagg_list_mtx;
84 eventhandler_tag	lagg_detach_cookie = NULL;
85 
86 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
87 static void	lagg_clone_destroy(struct ifnet *);
88 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
89 static void	lagg_capabilities(struct lagg_softc *);
90 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
91 static void	lagg_port_setlladdr(void *, int);
92 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
93 static int	lagg_port_destroy(struct lagg_port *, int);
94 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
95 static void	lagg_linkstate(struct lagg_softc *);
96 static void	lagg_port_state(struct ifnet *, int);
97 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
98 static int	lagg_port_output(struct ifnet *, struct mbuf *,
99 		    struct sockaddr *, struct route *);
100 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
101 #ifdef LAGG_PORT_STACKING
102 static int	lagg_port_checkstacking(struct lagg_softc *);
103 #endif
104 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
105 static void	lagg_init(void *);
106 static void	lagg_stop(struct lagg_softc *);
107 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
108 static int	lagg_ether_setmulti(struct lagg_softc *);
109 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
110 static	int	lagg_setflag(struct lagg_port *, int, int,
111 		    int (*func)(struct ifnet *, int));
112 static	int	lagg_setflags(struct lagg_port *, int status);
113 static void	lagg_start(struct ifnet *);
114 static int	lagg_media_change(struct ifnet *);
115 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
116 static struct lagg_port *lagg_link_active(struct lagg_softc *,
117 	    struct lagg_port *);
118 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
119 
120 IFC_SIMPLE_DECLARE(lagg, 0);
121 
122 /* Simple round robin */
123 static int	lagg_rr_attach(struct lagg_softc *);
124 static int	lagg_rr_detach(struct lagg_softc *);
125 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
126 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
127 		    struct mbuf *);
128 
129 /* Active failover */
130 static int	lagg_fail_attach(struct lagg_softc *);
131 static int	lagg_fail_detach(struct lagg_softc *);
132 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
133 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
134 		    struct mbuf *);
135 
136 /* Loadbalancing */
137 static int	lagg_lb_attach(struct lagg_softc *);
138 static int	lagg_lb_detach(struct lagg_softc *);
139 static int	lagg_lb_port_create(struct lagg_port *);
140 static void	lagg_lb_port_destroy(struct lagg_port *);
141 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
142 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
143 		    struct mbuf *);
144 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
145 
146 /* 802.3ad LACP */
147 static int	lagg_lacp_attach(struct lagg_softc *);
148 static int	lagg_lacp_detach(struct lagg_softc *);
149 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
150 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
151 		    struct mbuf *);
152 static void	lagg_lacp_lladdr(struct lagg_softc *);
153 
154 /* lagg protocol table */
155 static const struct {
156 	int			ti_proto;
157 	int			(*ti_attach)(struct lagg_softc *);
158 } lagg_protos[] = {
159 	{ LAGG_PROTO_ROUNDROBIN,	lagg_rr_attach },
160 	{ LAGG_PROTO_FAILOVER,		lagg_fail_attach },
161 	{ LAGG_PROTO_LOADBALANCE,	lagg_lb_attach },
162 	{ LAGG_PROTO_ETHERCHANNEL,	lagg_lb_attach },
163 	{ LAGG_PROTO_LACP,		lagg_lacp_attach },
164 	{ LAGG_PROTO_NONE,		NULL }
165 };
166 
167 SYSCTL_DECL(_net_link);
168 static SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
169     "Link Aggregation");
170 
171 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */
172 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW,
173     &lagg_failover_rx_all, 0,
174     "Accept input from any interface in a failover lagg");
175 
176 static int
177 lagg_modevent(module_t mod, int type, void *data)
178 {
179 
180 	switch (type) {
181 	case MOD_LOAD:
182 		mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
183 		SLIST_INIT(&lagg_list);
184 		if_clone_attach(&lagg_cloner);
185 		lagg_input_p = lagg_input;
186 		lagg_linkstate_p = lagg_port_state;
187 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
188 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
189 		    EVENTHANDLER_PRI_ANY);
190 		break;
191 	case MOD_UNLOAD:
192 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
193 		    lagg_detach_cookie);
194 		if_clone_detach(&lagg_cloner);
195 		lagg_input_p = NULL;
196 		lagg_linkstate_p = NULL;
197 		mtx_destroy(&lagg_list_mtx);
198 		break;
199 	default:
200 		return (EOPNOTSUPP);
201 	}
202 	return (0);
203 }
204 
205 static moduledata_t lagg_mod = {
206 	"if_lagg",
207 	lagg_modevent,
208 	0
209 };
210 
211 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
212 MODULE_VERSION(if_lagg, 1);
213 
214 #if __FreeBSD_version >= 800000
215 /*
216  * This routine is run via an vlan
217  * config EVENT
218  */
219 static void
220 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
221 {
222         struct lagg_softc       *sc = ifp->if_softc;
223         struct lagg_port        *lp;
224 
225         if (ifp->if_softc !=  arg)   /* Not our event */
226                 return;
227 
228         LAGG_RLOCK(sc);
229         if (!SLIST_EMPTY(&sc->sc_ports)) {
230                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
231                         EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
232         }
233         LAGG_RUNLOCK(sc);
234 }
235 
236 /*
237  * This routine is run via an vlan
238  * unconfig EVENT
239  */
240 static void
241 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
242 {
243         struct lagg_softc       *sc = ifp->if_softc;
244         struct lagg_port        *lp;
245 
246         if (ifp->if_softc !=  arg)   /* Not our event */
247                 return;
248 
249         LAGG_RLOCK(sc);
250         if (!SLIST_EMPTY(&sc->sc_ports)) {
251                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
252                         EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
253         }
254         LAGG_RUNLOCK(sc);
255 }
256 #endif
257 
258 static int
259 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
260 {
261 	struct lagg_softc *sc;
262 	struct ifnet *ifp;
263 	int i, error = 0;
264 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
265 
266 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
267 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
268 	if (ifp == NULL) {
269 		free(sc, M_DEVBUF);
270 		return (ENOSPC);
271 	}
272 
273 	sc->sc_proto = LAGG_PROTO_NONE;
274 	for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
275 		if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
276 			sc->sc_proto = lagg_protos[i].ti_proto;
277 			if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
278 				if_free_type(ifp, IFT_ETHER);
279 				free(sc, M_DEVBUF);
280 				return (error);
281 			}
282 			break;
283 		}
284 	}
285 	LAGG_LOCK_INIT(sc);
286 	SLIST_INIT(&sc->sc_ports);
287 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
288 
289 	/* Initialise pseudo media types */
290 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
291 	    lagg_media_status);
292 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
293 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
294 
295 	if_initname(ifp, ifc->ifc_name, unit);
296 	ifp->if_type = IFT_ETHER;
297 	ifp->if_softc = sc;
298 	ifp->if_start = lagg_start;
299 	ifp->if_init = lagg_init;
300 	ifp->if_ioctl = lagg_ioctl;
301 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
302 
303 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
304 	ifp->if_snd.ifq_drv_maxlen = ifqmaxlen;
305 	IFQ_SET_READY(&ifp->if_snd);
306 
307 	/*
308 	 * Attach as an ordinary ethernet device, childs will be attached
309 	 * as special device IFT_IEEE8023ADLAG.
310 	 */
311 	ether_ifattach(ifp, eaddr);
312 
313 #if __FreeBSD_version >= 800000
314 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
315 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
316 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
317 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
318 #endif
319 
320 	/* Insert into the global list of laggs */
321 	mtx_lock(&lagg_list_mtx);
322 	SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
323 	mtx_unlock(&lagg_list_mtx);
324 
325 	return (0);
326 }
327 
328 static void
329 lagg_clone_destroy(struct ifnet *ifp)
330 {
331 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
332 	struct lagg_port *lp;
333 
334 	LAGG_WLOCK(sc);
335 
336 	lagg_stop(sc);
337 	ifp->if_flags &= ~IFF_UP;
338 
339 #if __FreeBSD_version >= 800000
340 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
341 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
342 #endif
343 
344 	/* Shutdown and remove lagg ports */
345 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
346 		lagg_port_destroy(lp, 1);
347 	/* Unhook the aggregation protocol */
348 	if (sc->sc_detach != NULL)
349 		(*sc->sc_detach)(sc);
350 
351 	LAGG_WUNLOCK(sc);
352 
353 	ifmedia_removeall(&sc->sc_media);
354 	ether_ifdetach(ifp);
355 	if_free_type(ifp, IFT_ETHER);
356 
357 	mtx_lock(&lagg_list_mtx);
358 	SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
359 	mtx_unlock(&lagg_list_mtx);
360 
361 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
362 	LAGG_LOCK_DESTROY(sc);
363 	free(sc, M_DEVBUF);
364 }
365 
366 static void
367 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
368 {
369 	struct ifnet *ifp = sc->sc_ifp;
370 
371 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
372 		return;
373 
374 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
375 	/* Let the protocol know the MAC has changed */
376 	if (sc->sc_lladdr != NULL)
377 		(*sc->sc_lladdr)(sc);
378 	EVENTHANDLER_INVOKE(iflladdr_event, ifp);
379 }
380 
381 static void
382 lagg_capabilities(struct lagg_softc *sc)
383 {
384 	struct lagg_port *lp;
385 	int cap = ~0, ena = ~0;
386 	u_long hwa = ~0UL;
387 
388 	LAGG_WLOCK_ASSERT(sc);
389 
390 	/* Get capabilities from the lagg ports */
391 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
392 		cap &= lp->lp_ifp->if_capabilities;
393 		ena &= lp->lp_ifp->if_capenable;
394 		hwa &= lp->lp_ifp->if_hwassist;
395 	}
396 	cap = (cap == ~0 ? 0 : cap);
397 	ena = (ena == ~0 ? 0 : ena);
398 	hwa = (hwa == ~0 ? 0 : hwa);
399 
400 	if (sc->sc_ifp->if_capabilities != cap ||
401 	    sc->sc_ifp->if_capenable != ena ||
402 	    sc->sc_ifp->if_hwassist != hwa) {
403 		sc->sc_ifp->if_capabilities = cap;
404 		sc->sc_ifp->if_capenable = ena;
405 		sc->sc_ifp->if_hwassist = hwa;
406 		getmicrotime(&sc->sc_ifp->if_lastchange);
407 
408 		if (sc->sc_ifflags & IFF_DEBUG)
409 			if_printf(sc->sc_ifp,
410 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
411 	}
412 }
413 
414 static void
415 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
416 {
417 	struct lagg_softc *sc = lp->lp_softc;
418 	struct ifnet *ifp = lp->lp_ifp;
419 	struct lagg_llq *llq;
420 	int pending = 0;
421 
422 	LAGG_WLOCK_ASSERT(sc);
423 
424 	if (lp->lp_detaching ||
425 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
426 		return;
427 
428 	/* Check to make sure its not already queued to be changed */
429 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
430 		if (llq->llq_ifp == ifp) {
431 			pending = 1;
432 			break;
433 		}
434 	}
435 
436 	if (!pending) {
437 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
438 		if (llq == NULL)	/* XXX what to do */
439 			return;
440 	}
441 
442 	/* Update the lladdr even if pending, it may have changed */
443 	llq->llq_ifp = ifp;
444 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
445 
446 	if (!pending)
447 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
448 
449 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
450 }
451 
452 /*
453  * Set the interface MAC address from a taskqueue to avoid a LOR.
454  */
455 static void
456 lagg_port_setlladdr(void *arg, int pending)
457 {
458 	struct lagg_softc *sc = (struct lagg_softc *)arg;
459 	struct lagg_llq *llq, *head;
460 	struct ifnet *ifp;
461 	int error;
462 
463 	/* Grab a local reference of the queue and remove it from the softc */
464 	LAGG_WLOCK(sc);
465 	head = SLIST_FIRST(&sc->sc_llq_head);
466 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
467 	LAGG_WUNLOCK(sc);
468 
469 	/*
470 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
471 	 * unlocked as we have the only reference to it.
472 	 */
473 	for (llq = head; llq != NULL; llq = head) {
474 		ifp = llq->llq_ifp;
475 
476 		/* Set the link layer address */
477 		error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
478 		if (error)
479 			printf("%s: setlladdr failed on %s\n", __func__,
480 			    ifp->if_xname);
481 
482 		head = SLIST_NEXT(llq, llq_entries);
483 		free(llq, M_DEVBUF);
484 	}
485 }
486 
487 static int
488 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
489 {
490 	struct lagg_softc *sc_ptr;
491 	struct lagg_port *lp;
492 	int error = 0;
493 
494 	LAGG_WLOCK_ASSERT(sc);
495 
496 	/* Limit the maximal number of lagg ports */
497 	if (sc->sc_count >= LAGG_MAX_PORTS)
498 		return (ENOSPC);
499 
500 	/* Check if port has already been associated to a lagg */
501 	if (ifp->if_lagg != NULL)
502 		return (EBUSY);
503 
504 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
505 	if (ifp->if_type != IFT_ETHER)
506 		return (EPROTONOSUPPORT);
507 
508 	/* Allow the first Ethernet member to define the MTU */
509 	if (SLIST_EMPTY(&sc->sc_ports))
510 		sc->sc_ifp->if_mtu = ifp->if_mtu;
511 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
512 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
513 		    ifp->if_xname);
514 		return (EINVAL);
515 	}
516 
517 	if ((lp = malloc(sizeof(struct lagg_port),
518 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
519 		return (ENOMEM);
520 
521 	/* Check if port is a stacked lagg */
522 	mtx_lock(&lagg_list_mtx);
523 	SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
524 		if (ifp == sc_ptr->sc_ifp) {
525 			mtx_unlock(&lagg_list_mtx);
526 			free(lp, M_DEVBUF);
527 			return (EINVAL);
528 			/* XXX disable stacking for the moment, its untested */
529 #ifdef LAGG_PORT_STACKING
530 			lp->lp_flags |= LAGG_PORT_STACK;
531 			if (lagg_port_checkstacking(sc_ptr) >=
532 			    LAGG_MAX_STACKING) {
533 				mtx_unlock(&lagg_list_mtx);
534 				free(lp, M_DEVBUF);
535 				return (E2BIG);
536 			}
537 #endif
538 		}
539 	}
540 	mtx_unlock(&lagg_list_mtx);
541 
542 	/* Change the interface type */
543 	lp->lp_iftype = ifp->if_type;
544 	ifp->if_type = IFT_IEEE8023ADLAG;
545 	ifp->if_lagg = lp;
546 	lp->lp_ioctl = ifp->if_ioctl;
547 	ifp->if_ioctl = lagg_port_ioctl;
548 	lp->lp_output = ifp->if_output;
549 	ifp->if_output = lagg_port_output;
550 
551 	lp->lp_ifp = ifp;
552 	lp->lp_softc = sc;
553 
554 	/* Save port link layer address */
555 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
556 
557 	if (SLIST_EMPTY(&sc->sc_ports)) {
558 		sc->sc_primary = lp;
559 		lagg_lladdr(sc, IF_LLADDR(ifp));
560 	} else {
561 		/* Update link layer address for this port */
562 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
563 	}
564 
565 	/* Insert into the list of ports */
566 	SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
567 	sc->sc_count++;
568 
569 	/* Update lagg capabilities */
570 	lagg_capabilities(sc);
571 	lagg_linkstate(sc);
572 
573 	/* Add multicast addresses and interface flags to this port */
574 	lagg_ether_cmdmulti(lp, 1);
575 	lagg_setflags(lp, 1);
576 
577 	if (sc->sc_port_create != NULL)
578 		error = (*sc->sc_port_create)(lp);
579 	if (error) {
580 		/* remove the port again, without calling sc_port_destroy */
581 		lagg_port_destroy(lp, 0);
582 		return (error);
583 	}
584 
585 	return (error);
586 }
587 
588 #ifdef LAGG_PORT_STACKING
589 static int
590 lagg_port_checkstacking(struct lagg_softc *sc)
591 {
592 	struct lagg_softc *sc_ptr;
593 	struct lagg_port *lp;
594 	int m = 0;
595 
596 	LAGG_WLOCK_ASSERT(sc);
597 
598 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
599 		if (lp->lp_flags & LAGG_PORT_STACK) {
600 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
601 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
602 		}
603 	}
604 
605 	return (m + 1);
606 }
607 #endif
608 
609 static int
610 lagg_port_destroy(struct lagg_port *lp, int runpd)
611 {
612 	struct lagg_softc *sc = lp->lp_softc;
613 	struct lagg_port *lp_ptr;
614 	struct lagg_llq *llq;
615 	struct ifnet *ifp = lp->lp_ifp;
616 
617 	LAGG_WLOCK_ASSERT(sc);
618 
619 	if (runpd && sc->sc_port_destroy != NULL)
620 		(*sc->sc_port_destroy)(lp);
621 
622 	/*
623 	 * Remove multicast addresses and interface flags from this port and
624 	 * reset the MAC address, skip if the interface is being detached.
625 	 */
626 	if (!lp->lp_detaching) {
627 		lagg_ether_cmdmulti(lp, 0);
628 		lagg_setflags(lp, 0);
629 		lagg_port_lladdr(lp, lp->lp_lladdr);
630 	}
631 
632 	/* Restore interface */
633 	ifp->if_type = lp->lp_iftype;
634 	ifp->if_ioctl = lp->lp_ioctl;
635 	ifp->if_output = lp->lp_output;
636 	ifp->if_lagg = NULL;
637 
638 	/* Finally, remove the port from the lagg */
639 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
640 	sc->sc_count--;
641 
642 	/* Update the primary interface */
643 	if (lp == sc->sc_primary) {
644 		uint8_t lladdr[ETHER_ADDR_LEN];
645 
646 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
647 			bzero(&lladdr, ETHER_ADDR_LEN);
648 		} else {
649 			bcopy(lp_ptr->lp_lladdr,
650 			    lladdr, ETHER_ADDR_LEN);
651 		}
652 		lagg_lladdr(sc, lladdr);
653 		sc->sc_primary = lp_ptr;
654 
655 		/* Update link layer address for each port */
656 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
657 			lagg_port_lladdr(lp_ptr, lladdr);
658 	}
659 
660 	/* Remove any pending lladdr changes from the queue */
661 	if (lp->lp_detaching) {
662 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
663 			if (llq->llq_ifp == ifp) {
664 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
665 				    llq_entries);
666 				free(llq, M_DEVBUF);
667 				break;	/* Only appears once */
668 			}
669 		}
670 	}
671 
672 	if (lp->lp_ifflags)
673 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
674 
675 	free(lp, M_DEVBUF);
676 
677 	/* Update lagg capabilities */
678 	lagg_capabilities(sc);
679 	lagg_linkstate(sc);
680 
681 	return (0);
682 }
683 
684 static int
685 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
686 {
687 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
688 	struct lagg_softc *sc;
689 	struct lagg_port *lp = NULL;
690 	int error = 0;
691 
692 	/* Should be checked by the caller */
693 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
694 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
695 		goto fallback;
696 
697 	switch (cmd) {
698 	case SIOCGLAGGPORT:
699 		if (rp->rp_portname[0] == '\0' ||
700 		    ifunit(rp->rp_portname) != ifp) {
701 			error = EINVAL;
702 			break;
703 		}
704 
705 		LAGG_RLOCK(sc);
706 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
707 			error = ENOENT;
708 			LAGG_RUNLOCK(sc);
709 			break;
710 		}
711 
712 		lagg_port2req(lp, rp);
713 		LAGG_RUNLOCK(sc);
714 		break;
715 
716 	case SIOCSIFCAP:
717 		if (lp->lp_ioctl == NULL) {
718 			error = EINVAL;
719 			break;
720 		}
721 		error = (*lp->lp_ioctl)(ifp, cmd, data);
722 		if (error)
723 			break;
724 
725 		/* Update lagg interface capabilities */
726 		LAGG_WLOCK(sc);
727 		lagg_capabilities(sc);
728 		LAGG_WUNLOCK(sc);
729 		break;
730 
731 	case SIOCSIFMTU:
732 		/* Do not allow the MTU to be changed once joined */
733 		error = EINVAL;
734 		break;
735 
736 	default:
737 		goto fallback;
738 	}
739 
740 	return (error);
741 
742 fallback:
743 	if (lp->lp_ioctl != NULL)
744 		return ((*lp->lp_ioctl)(ifp, cmd, data));
745 
746 	return (EINVAL);
747 }
748 
749 static int
750 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
751 	struct sockaddr *dst, struct route *ro)
752 {
753 	struct lagg_port *lp = ifp->if_lagg;
754 	struct ether_header *eh;
755 	short type = 0;
756 
757 	switch (dst->sa_family) {
758 		case pseudo_AF_HDRCMPLT:
759 		case AF_UNSPEC:
760 			eh = (struct ether_header *)dst->sa_data;
761 			type = eh->ether_type;
762 			break;
763 	}
764 
765 	/*
766 	 * Only allow ethernet types required to initiate or maintain the link,
767 	 * aggregated frames take a different path.
768 	 */
769 	switch (ntohs(type)) {
770 		case ETHERTYPE_PAE:	/* EAPOL PAE/802.1x */
771 			return ((*lp->lp_output)(ifp, m, dst, ro));
772 	}
773 
774 	/* drop any other frames */
775 	m_freem(m);
776 	return (EBUSY);
777 }
778 
779 static void
780 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
781 {
782 	struct lagg_port *lp;
783 	struct lagg_softc *sc;
784 
785 	if ((lp = ifp->if_lagg) == NULL)
786 		return;
787 
788 	sc = lp->lp_softc;
789 
790 	LAGG_WLOCK(sc);
791 	lp->lp_detaching = 1;
792 	lagg_port_destroy(lp, 1);
793 	LAGG_WUNLOCK(sc);
794 }
795 
796 static void
797 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
798 {
799 	struct lagg_softc *sc = lp->lp_softc;
800 
801 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
802 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
803 	rp->rp_prio = lp->lp_prio;
804 	rp->rp_flags = lp->lp_flags;
805 	if (sc->sc_portreq != NULL)
806 		(*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
807 
808 	/* Add protocol specific flags */
809 	switch (sc->sc_proto) {
810 		case LAGG_PROTO_FAILOVER:
811 			if (lp == sc->sc_primary)
812 				rp->rp_flags |= LAGG_PORT_MASTER;
813 			if (lp == lagg_link_active(sc, sc->sc_primary))
814 				rp->rp_flags |= LAGG_PORT_ACTIVE;
815 			break;
816 
817 		case LAGG_PROTO_ROUNDROBIN:
818 		case LAGG_PROTO_LOADBALANCE:
819 		case LAGG_PROTO_ETHERCHANNEL:
820 			if (LAGG_PORTACTIVE(lp))
821 				rp->rp_flags |= LAGG_PORT_ACTIVE;
822 			break;
823 
824 		case LAGG_PROTO_LACP:
825 			/* LACP has a different definition of active */
826 			if (lacp_isactive(lp))
827 				rp->rp_flags |= LAGG_PORT_ACTIVE;
828 			if (lacp_iscollecting(lp))
829 				rp->rp_flags |= LAGG_PORT_COLLECTING;
830 			if (lacp_isdistributing(lp))
831 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
832 			break;
833 	}
834 
835 }
836 
837 static void
838 lagg_init(void *xsc)
839 {
840 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
841 	struct lagg_port *lp;
842 	struct ifnet *ifp = sc->sc_ifp;
843 
844 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
845 		return;
846 
847 	LAGG_WLOCK(sc);
848 
849 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
850 	/* Update the port lladdrs */
851 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
852 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
853 
854 	if (sc->sc_init != NULL)
855 		(*sc->sc_init)(sc);
856 
857 	LAGG_WUNLOCK(sc);
858 }
859 
860 static void
861 lagg_stop(struct lagg_softc *sc)
862 {
863 	struct ifnet *ifp = sc->sc_ifp;
864 
865 	LAGG_WLOCK_ASSERT(sc);
866 
867 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
868 		return;
869 
870 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
871 
872 	if (sc->sc_stop != NULL)
873 		(*sc->sc_stop)(sc);
874 }
875 
876 static int
877 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
878 {
879 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
880 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
881 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
882 	struct ifreq *ifr = (struct ifreq *)data;
883 	struct lagg_port *lp;
884 	struct ifnet *tpif;
885 	struct thread *td = curthread;
886 	char *buf, *outbuf;
887 	int count, buflen, len, error = 0;
888 
889 	bzero(&rpbuf, sizeof(rpbuf));
890 
891 	switch (cmd) {
892 	case SIOCGLAGG:
893 		LAGG_RLOCK(sc);
894 		count = 0;
895 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
896 			count++;
897 		buflen = count * sizeof(struct lagg_reqport);
898 		LAGG_RUNLOCK(sc);
899 
900 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
901 
902 		LAGG_RLOCK(sc);
903 		ra->ra_proto = sc->sc_proto;
904 		if (sc->sc_req != NULL)
905 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
906 
907 		count = 0;
908 		buf = outbuf;
909 		len = min(ra->ra_size, buflen);
910 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
911 			if (len < sizeof(rpbuf))
912 				break;
913 
914 			lagg_port2req(lp, &rpbuf);
915 			memcpy(buf, &rpbuf, sizeof(rpbuf));
916 			count++;
917 			buf += sizeof(rpbuf);
918 			len -= sizeof(rpbuf);
919 		}
920 		LAGG_RUNLOCK(sc);
921 		ra->ra_ports = count;
922 		ra->ra_size = count * sizeof(rpbuf);
923 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
924 		free(outbuf, M_TEMP);
925 		break;
926 	case SIOCSLAGG:
927 		error = priv_check(td, PRIV_NET_LAGG);
928 		if (error)
929 			break;
930 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
931 			error = EPROTONOSUPPORT;
932 			break;
933 		}
934 		if (sc->sc_proto != LAGG_PROTO_NONE) {
935 			LAGG_WLOCK(sc);
936 			error = sc->sc_detach(sc);
937 			/* Reset protocol and pointers */
938 			sc->sc_proto = LAGG_PROTO_NONE;
939 			sc->sc_detach = NULL;
940 			sc->sc_start = NULL;
941 			sc->sc_input = NULL;
942 			sc->sc_port_create = NULL;
943 			sc->sc_port_destroy = NULL;
944 			sc->sc_linkstate = NULL;
945 			sc->sc_init = NULL;
946 			sc->sc_stop = NULL;
947 			sc->sc_lladdr = NULL;
948 			sc->sc_req = NULL;
949 			sc->sc_portreq = NULL;
950 			LAGG_WUNLOCK(sc);
951 		}
952 		if (error != 0)
953 			break;
954 		for (int i = 0; i < (sizeof(lagg_protos) /
955 		    sizeof(lagg_protos[0])); i++) {
956 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
957 				if (sc->sc_ifflags & IFF_DEBUG)
958 					printf("%s: using proto %u\n",
959 					    sc->sc_ifname,
960 					    lagg_protos[i].ti_proto);
961 				LAGG_WLOCK(sc);
962 				sc->sc_proto = lagg_protos[i].ti_proto;
963 				if (sc->sc_proto != LAGG_PROTO_NONE)
964 					error = lagg_protos[i].ti_attach(sc);
965 				LAGG_WUNLOCK(sc);
966 				return (error);
967 			}
968 		}
969 		error = EPROTONOSUPPORT;
970 		break;
971 	case SIOCGLAGGPORT:
972 		if (rp->rp_portname[0] == '\0' ||
973 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
974 			error = EINVAL;
975 			break;
976 		}
977 
978 		LAGG_RLOCK(sc);
979 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
980 		    lp->lp_softc != sc) {
981 			error = ENOENT;
982 			LAGG_RUNLOCK(sc);
983 			break;
984 		}
985 
986 		lagg_port2req(lp, rp);
987 		LAGG_RUNLOCK(sc);
988 		break;
989 	case SIOCSLAGGPORT:
990 		error = priv_check(td, PRIV_NET_LAGG);
991 		if (error)
992 			break;
993 		if (rp->rp_portname[0] == '\0' ||
994 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
995 			error = EINVAL;
996 			break;
997 		}
998 		LAGG_WLOCK(sc);
999 		error = lagg_port_create(sc, tpif);
1000 		LAGG_WUNLOCK(sc);
1001 		break;
1002 	case SIOCSLAGGDELPORT:
1003 		error = priv_check(td, PRIV_NET_LAGG);
1004 		if (error)
1005 			break;
1006 		if (rp->rp_portname[0] == '\0' ||
1007 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1008 			error = EINVAL;
1009 			break;
1010 		}
1011 
1012 		LAGG_WLOCK(sc);
1013 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1014 		    lp->lp_softc != sc) {
1015 			error = ENOENT;
1016 			LAGG_WUNLOCK(sc);
1017 			break;
1018 		}
1019 
1020 		error = lagg_port_destroy(lp, 1);
1021 		LAGG_WUNLOCK(sc);
1022 		break;
1023 	case SIOCSIFFLAGS:
1024 		/* Set flags on ports too */
1025 		LAGG_WLOCK(sc);
1026 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1027 			lagg_setflags(lp, 1);
1028 		}
1029 		LAGG_WUNLOCK(sc);
1030 
1031 		if (!(ifp->if_flags & IFF_UP) &&
1032 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1033 			/*
1034 			 * If interface is marked down and it is running,
1035 			 * then stop and disable it.
1036 			 */
1037 			LAGG_WLOCK(sc);
1038 			lagg_stop(sc);
1039 			LAGG_WUNLOCK(sc);
1040 		} else if ((ifp->if_flags & IFF_UP) &&
1041 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1042 			/*
1043 			 * If interface is marked up and it is stopped, then
1044 			 * start it.
1045 			 */
1046 			(*ifp->if_init)(sc);
1047 		}
1048 		break;
1049 	case SIOCADDMULTI:
1050 	case SIOCDELMULTI:
1051 		LAGG_WLOCK(sc);
1052 		error = lagg_ether_setmulti(sc);
1053 		LAGG_WUNLOCK(sc);
1054 		break;
1055 	case SIOCSIFMEDIA:
1056 	case SIOCGIFMEDIA:
1057 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1058 		break;
1059 
1060 	case SIOCSIFCAP:
1061 	case SIOCSIFMTU:
1062 		/* Do not allow the MTU or caps to be directly changed */
1063 		error = EINVAL;
1064 		break;
1065 
1066 	default:
1067 		error = ether_ioctl(ifp, cmd, data);
1068 		break;
1069 	}
1070 	return (error);
1071 }
1072 
1073 static int
1074 lagg_ether_setmulti(struct lagg_softc *sc)
1075 {
1076 	struct lagg_port *lp;
1077 
1078 	LAGG_WLOCK_ASSERT(sc);
1079 
1080 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1081 		/* First, remove any existing filter entries. */
1082 		lagg_ether_cmdmulti(lp, 0);
1083 		/* copy all addresses from the lagg interface to the port */
1084 		lagg_ether_cmdmulti(lp, 1);
1085 	}
1086 	return (0);
1087 }
1088 
1089 static int
1090 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1091 {
1092 	struct lagg_softc *sc = lp->lp_softc;
1093 	struct ifnet *ifp = lp->lp_ifp;
1094 	struct ifnet *scifp = sc->sc_ifp;
1095 	struct lagg_mc *mc;
1096 	struct ifmultiaddr *ifma, *rifma = NULL;
1097 	struct sockaddr_dl sdl;
1098 	int error;
1099 
1100 	LAGG_WLOCK_ASSERT(sc);
1101 
1102 	bzero((char *)&sdl, sizeof(sdl));
1103 	sdl.sdl_len = sizeof(sdl);
1104 	sdl.sdl_family = AF_LINK;
1105 	sdl.sdl_type = IFT_ETHER;
1106 	sdl.sdl_alen = ETHER_ADDR_LEN;
1107 	sdl.sdl_index = ifp->if_index;
1108 
1109 	if (set) {
1110 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1111 			if (ifma->ifma_addr->sa_family != AF_LINK)
1112 				continue;
1113 			bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1114 			    LLADDR(&sdl), ETHER_ADDR_LEN);
1115 
1116 			error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1117 			if (error)
1118 				return (error);
1119 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1120 			if (mc == NULL)
1121 				return (ENOMEM);
1122 			mc->mc_ifma = rifma;
1123 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1124 		}
1125 	} else {
1126 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1127 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1128 			if_delmulti_ifma(mc->mc_ifma);
1129 			free(mc, M_DEVBUF);
1130 		}
1131 	}
1132 	return (0);
1133 }
1134 
1135 /* Handle a ref counted flag that should be set on the lagg port as well */
1136 static int
1137 lagg_setflag(struct lagg_port *lp, int flag, int status,
1138 	     int (*func)(struct ifnet *, int))
1139 {
1140 	struct lagg_softc *sc = lp->lp_softc;
1141 	struct ifnet *scifp = sc->sc_ifp;
1142 	struct ifnet *ifp = lp->lp_ifp;
1143 	int error;
1144 
1145 	LAGG_WLOCK_ASSERT(sc);
1146 
1147 	status = status ? (scifp->if_flags & flag) : 0;
1148 	/* Now "status" contains the flag value or 0 */
1149 
1150 	/*
1151 	 * See if recorded ports status is different from what
1152 	 * we want it to be.  If it is, flip it.  We record ports
1153 	 * status in lp_ifflags so that we won't clear ports flag
1154 	 * we haven't set.  In fact, we don't clear or set ports
1155 	 * flags directly, but get or release references to them.
1156 	 * That's why we can be sure that recorded flags still are
1157 	 * in accord with actual ports flags.
1158 	 */
1159 	if (status != (lp->lp_ifflags & flag)) {
1160 		error = (*func)(ifp, status);
1161 		if (error)
1162 			return (error);
1163 		lp->lp_ifflags &= ~flag;
1164 		lp->lp_ifflags |= status;
1165 	}
1166 	return (0);
1167 }
1168 
1169 /*
1170  * Handle IFF_* flags that require certain changes on the lagg port
1171  * if "status" is true, update ports flags respective to the lagg
1172  * if "status" is false, forcedly clear the flags set on port.
1173  */
1174 static int
1175 lagg_setflags(struct lagg_port *lp, int status)
1176 {
1177 	int error, i;
1178 
1179 	for (i = 0; lagg_pflags[i].flag; i++) {
1180 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1181 		    status, lagg_pflags[i].func);
1182 		if (error)
1183 			return (error);
1184 	}
1185 	return (0);
1186 }
1187 
1188 static void
1189 lagg_start(struct ifnet *ifp)
1190 {
1191 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1192 	struct mbuf *m;
1193 	int error = 0;
1194 
1195 	LAGG_RLOCK(sc);
1196 	/* We need a Tx algorithm and at least one port */
1197 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1198 		IF_DRAIN(&ifp->if_snd);
1199 		LAGG_RUNLOCK(sc);
1200 		return;
1201 	}
1202 
1203 	for (;; error = 0) {
1204 		IFQ_DEQUEUE(&ifp->if_snd, m);
1205 		if (m == NULL)
1206 			break;
1207 
1208 		ETHER_BPF_MTAP(ifp, m);
1209 
1210 		error = (*sc->sc_start)(sc, m);
1211 		if (error == 0)
1212 			ifp->if_opackets++;
1213 		else
1214 			ifp->if_oerrors++;
1215 	}
1216 	LAGG_RUNLOCK(sc);
1217 }
1218 
1219 static struct mbuf *
1220 lagg_input(struct ifnet *ifp, struct mbuf *m)
1221 {
1222 	struct lagg_port *lp = ifp->if_lagg;
1223 	struct lagg_softc *sc = lp->lp_softc;
1224 	struct ifnet *scifp = sc->sc_ifp;
1225 
1226 	LAGG_RLOCK(sc);
1227 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1228 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1229 	    sc->sc_proto == LAGG_PROTO_NONE) {
1230 		LAGG_RUNLOCK(sc);
1231 		m_freem(m);
1232 		return (NULL);
1233 	}
1234 
1235 	ETHER_BPF_MTAP(scifp, m);
1236 
1237 	m = (*sc->sc_input)(sc, lp, m);
1238 
1239 	if (m != NULL) {
1240 		scifp->if_ipackets++;
1241 		scifp->if_ibytes += m->m_pkthdr.len;
1242 
1243 		if (scifp->if_flags & IFF_MONITOR) {
1244 			m_freem(m);
1245 			m = NULL;
1246 		}
1247 	}
1248 
1249 	LAGG_RUNLOCK(sc);
1250 	return (m);
1251 }
1252 
1253 static int
1254 lagg_media_change(struct ifnet *ifp)
1255 {
1256 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1257 
1258 	if (sc->sc_ifflags & IFF_DEBUG)
1259 		printf("%s\n", __func__);
1260 
1261 	/* Ignore */
1262 	return (0);
1263 }
1264 
1265 static void
1266 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1267 {
1268 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1269 	struct lagg_port *lp;
1270 
1271 	imr->ifm_status = IFM_AVALID;
1272 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1273 
1274 	LAGG_RLOCK(sc);
1275 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1276 		if (LAGG_PORTACTIVE(lp))
1277 			imr->ifm_status |= IFM_ACTIVE;
1278 	}
1279 	LAGG_RUNLOCK(sc);
1280 }
1281 
1282 static void
1283 lagg_linkstate(struct lagg_softc *sc)
1284 {
1285 	struct lagg_port *lp;
1286 	int new_link = LINK_STATE_DOWN;
1287 	uint64_t speed;
1288 
1289 	/* Our link is considered up if at least one of our ports is active */
1290 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1291 		if (lp->lp_link_state == LINK_STATE_UP) {
1292 			new_link = LINK_STATE_UP;
1293 			break;
1294 		}
1295 	}
1296 	if_link_state_change(sc->sc_ifp, new_link);
1297 
1298 	/* Update if_baudrate to reflect the max possible speed */
1299 	switch (sc->sc_proto) {
1300 		case LAGG_PROTO_FAILOVER:
1301 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1302 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1303 			break;
1304 		case LAGG_PROTO_ROUNDROBIN:
1305 		case LAGG_PROTO_LOADBALANCE:
1306 		case LAGG_PROTO_ETHERCHANNEL:
1307 			speed = 0;
1308 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1309 				speed += lp->lp_ifp->if_baudrate;
1310 			sc->sc_ifp->if_baudrate = speed;
1311 			break;
1312 		case LAGG_PROTO_LACP:
1313 			/* LACP updates if_baudrate itself */
1314 			break;
1315 	}
1316 }
1317 
1318 static void
1319 lagg_port_state(struct ifnet *ifp, int state)
1320 {
1321 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1322 	struct lagg_softc *sc = NULL;
1323 
1324 	if (lp != NULL)
1325 		sc = lp->lp_softc;
1326 	if (sc == NULL)
1327 		return;
1328 
1329 	LAGG_WLOCK(sc);
1330 	lagg_linkstate(sc);
1331 	if (sc->sc_linkstate != NULL)
1332 		(*sc->sc_linkstate)(lp);
1333 	LAGG_WUNLOCK(sc);
1334 }
1335 
1336 struct lagg_port *
1337 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1338 {
1339 	struct lagg_port *lp_next, *rval = NULL;
1340 	// int new_link = LINK_STATE_DOWN;
1341 
1342 	LAGG_RLOCK_ASSERT(sc);
1343 	/*
1344 	 * Search a port which reports an active link state.
1345 	 */
1346 
1347 	if (lp == NULL)
1348 		goto search;
1349 	if (LAGG_PORTACTIVE(lp)) {
1350 		rval = lp;
1351 		goto found;
1352 	}
1353 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1354 	    LAGG_PORTACTIVE(lp_next)) {
1355 		rval = lp_next;
1356 		goto found;
1357 	}
1358 
1359 search:
1360 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1361 		if (LAGG_PORTACTIVE(lp_next)) {
1362 			rval = lp_next;
1363 			goto found;
1364 		}
1365 	}
1366 
1367 found:
1368 	if (rval != NULL) {
1369 		/*
1370 		 * The IEEE 802.1D standard assumes that a lagg with
1371 		 * multiple ports is always full duplex. This is valid
1372 		 * for load sharing laggs and if at least two links
1373 		 * are active. Unfortunately, checking the latter would
1374 		 * be too expensive at this point.
1375 		 XXX
1376 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1377 		    (sc->sc_count > 1))
1378 			new_link = LINK_STATE_FULL_DUPLEX;
1379 		else
1380 			new_link = rval->lp_link_state;
1381 		 */
1382 	}
1383 
1384 	return (rval);
1385 }
1386 
1387 static const void *
1388 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1389 {
1390 	if (m->m_pkthdr.len < (off + len)) {
1391 		return (NULL);
1392 	} else if (m->m_len < (off + len)) {
1393 		m_copydata(m, off, len, buf);
1394 		return (buf);
1395 	}
1396 	return (mtod(m, char *) + off);
1397 }
1398 
1399 uint32_t
1400 lagg_hashmbuf(struct mbuf *m, uint32_t key)
1401 {
1402 	uint16_t etype;
1403 	uint32_t p = 0;
1404 	int off;
1405 	struct ether_header *eh;
1406 	struct ether_vlan_header vlanbuf;
1407 	const struct ether_vlan_header *vlan;
1408 #ifdef INET
1409 	const struct ip *ip;
1410 	struct ip ipbuf;
1411 #endif
1412 #ifdef INET6
1413 	const struct ip6_hdr *ip6;
1414 	struct ip6_hdr ip6buf;
1415 	uint32_t flow;
1416 #endif
1417 
1418 	off = sizeof(*eh);
1419 	if (m->m_len < off)
1420 		goto out;
1421 	eh = mtod(m, struct ether_header *);
1422 	etype = ntohs(eh->ether_type);
1423 	p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, key);
1424 	p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1425 
1426 	/* Special handling for encapsulating VLAN frames */
1427 	if (m->m_flags & M_VLANTAG) {
1428 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1429 		    sizeof(m->m_pkthdr.ether_vtag), p);
1430 	} else if (etype == ETHERTYPE_VLAN) {
1431 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &vlanbuf);
1432 		if (vlan == NULL)
1433 			goto out;
1434 
1435 		p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1436 		etype = ntohs(vlan->evl_proto);
1437 		off += sizeof(*vlan) - sizeof(*eh);
1438 	}
1439 
1440 	switch (etype) {
1441 #ifdef INET
1442 	case ETHERTYPE_IP:
1443 		ip = lagg_gethdr(m, off, sizeof(*ip), &ipbuf);
1444 		if (ip == NULL)
1445 			goto out;
1446 
1447 		p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1448 		p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1449 		break;
1450 #endif
1451 #ifdef INET6
1452 	case ETHERTYPE_IPV6:
1453 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &ip6buf);
1454 		if (ip6 == NULL)
1455 			goto out;
1456 
1457 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1458 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1459 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1460 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1461 		break;
1462 #endif
1463 	}
1464 out:
1465 	return (p);
1466 }
1467 
1468 int
1469 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1470 {
1471 
1472 	return (ifp->if_transmit)(ifp, m);
1473 }
1474 
1475 /*
1476  * Simple round robin aggregation
1477  */
1478 
1479 static int
1480 lagg_rr_attach(struct lagg_softc *sc)
1481 {
1482 	sc->sc_detach = lagg_rr_detach;
1483 	sc->sc_start = lagg_rr_start;
1484 	sc->sc_input = lagg_rr_input;
1485 	sc->sc_port_create = NULL;
1486 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1487 	sc->sc_seq = 0;
1488 
1489 	return (0);
1490 }
1491 
1492 static int
1493 lagg_rr_detach(struct lagg_softc *sc)
1494 {
1495 	return (0);
1496 }
1497 
1498 static int
1499 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1500 {
1501 	struct lagg_port *lp;
1502 	uint32_t p;
1503 
1504 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1505 	p %= sc->sc_count;
1506 	lp = SLIST_FIRST(&sc->sc_ports);
1507 	while (p--)
1508 		lp = SLIST_NEXT(lp, lp_entries);
1509 
1510 	/*
1511 	 * Check the port's link state. This will return the next active
1512 	 * port if the link is down or the port is NULL.
1513 	 */
1514 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1515 		m_freem(m);
1516 		return (ENOENT);
1517 	}
1518 
1519 	/* Send mbuf */
1520 	return (lagg_enqueue(lp->lp_ifp, m));
1521 }
1522 
1523 static struct mbuf *
1524 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1525 {
1526 	struct ifnet *ifp = sc->sc_ifp;
1527 
1528 	/* Just pass in the packet to our lagg device */
1529 	m->m_pkthdr.rcvif = ifp;
1530 
1531 	return (m);
1532 }
1533 
1534 /*
1535  * Active failover
1536  */
1537 
1538 static int
1539 lagg_fail_attach(struct lagg_softc *sc)
1540 {
1541 	sc->sc_detach = lagg_fail_detach;
1542 	sc->sc_start = lagg_fail_start;
1543 	sc->sc_input = lagg_fail_input;
1544 	sc->sc_port_create = NULL;
1545 	sc->sc_port_destroy = NULL;
1546 
1547 	return (0);
1548 }
1549 
1550 static int
1551 lagg_fail_detach(struct lagg_softc *sc)
1552 {
1553 	return (0);
1554 }
1555 
1556 static int
1557 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1558 {
1559 	struct lagg_port *lp;
1560 
1561 	/* Use the master port if active or the next available port */
1562 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1563 		m_freem(m);
1564 		return (ENOENT);
1565 	}
1566 
1567 	/* Send mbuf */
1568 	return (lagg_enqueue(lp->lp_ifp, m));
1569 }
1570 
1571 static struct mbuf *
1572 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1573 {
1574 	struct ifnet *ifp = sc->sc_ifp;
1575 	struct lagg_port *tmp_tp;
1576 
1577 	if (lp == sc->sc_primary || lagg_failover_rx_all) {
1578 		m->m_pkthdr.rcvif = ifp;
1579 		return (m);
1580 	}
1581 
1582 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1583 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1584 		/*
1585 		 * If tmp_tp is null, we've recieved a packet when all
1586 		 * our links are down. Weird, but process it anyways.
1587 		 */
1588 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1589 			m->m_pkthdr.rcvif = ifp;
1590 			return (m);
1591 		}
1592 	}
1593 
1594 	m_freem(m);
1595 	return (NULL);
1596 }
1597 
1598 /*
1599  * Loadbalancing
1600  */
1601 
1602 static int
1603 lagg_lb_attach(struct lagg_softc *sc)
1604 {
1605 	struct lagg_port *lp;
1606 	struct lagg_lb *lb;
1607 
1608 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1609 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1610 		return (ENOMEM);
1611 
1612 	sc->sc_detach = lagg_lb_detach;
1613 	sc->sc_start = lagg_lb_start;
1614 	sc->sc_input = lagg_lb_input;
1615 	sc->sc_port_create = lagg_lb_port_create;
1616 	sc->sc_port_destroy = lagg_lb_port_destroy;
1617 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1618 
1619 	lb->lb_key = arc4random();
1620 	sc->sc_psc = (caddr_t)lb;
1621 
1622 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1623 		lagg_lb_port_create(lp);
1624 
1625 	return (0);
1626 }
1627 
1628 static int
1629 lagg_lb_detach(struct lagg_softc *sc)
1630 {
1631 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1632 	if (lb != NULL)
1633 		free(lb, M_DEVBUF);
1634 	return (0);
1635 }
1636 
1637 static int
1638 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1639 {
1640 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1641 	struct lagg_port *lp_next;
1642 	int i = 0;
1643 
1644 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1645 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1646 		if (lp_next == lp)
1647 			continue;
1648 		if (i >= LAGG_MAX_PORTS)
1649 			return (EINVAL);
1650 		if (sc->sc_ifflags & IFF_DEBUG)
1651 			printf("%s: port %s at index %d\n",
1652 			    sc->sc_ifname, lp_next->lp_ifname, i);
1653 		lb->lb_ports[i++] = lp_next;
1654 	}
1655 
1656 	return (0);
1657 }
1658 
1659 static int
1660 lagg_lb_port_create(struct lagg_port *lp)
1661 {
1662 	struct lagg_softc *sc = lp->lp_softc;
1663 	return (lagg_lb_porttable(sc, NULL));
1664 }
1665 
1666 static void
1667 lagg_lb_port_destroy(struct lagg_port *lp)
1668 {
1669 	struct lagg_softc *sc = lp->lp_softc;
1670 	lagg_lb_porttable(sc, lp);
1671 }
1672 
1673 static int
1674 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1675 {
1676 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1677 	struct lagg_port *lp = NULL;
1678 	uint32_t p = 0;
1679 
1680 	if (m->m_flags & M_FLOWID)
1681 		p = m->m_pkthdr.flowid;
1682 	else
1683 		p = lagg_hashmbuf(m, lb->lb_key);
1684 	p %= sc->sc_count;
1685 	lp = lb->lb_ports[p];
1686 
1687 	/*
1688 	 * Check the port's link state. This will return the next active
1689 	 * port if the link is down or the port is NULL.
1690 	 */
1691 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1692 		m_freem(m);
1693 		return (ENOENT);
1694 	}
1695 
1696 	/* Send mbuf */
1697 	return (lagg_enqueue(lp->lp_ifp, m));
1698 }
1699 
1700 static struct mbuf *
1701 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1702 {
1703 	struct ifnet *ifp = sc->sc_ifp;
1704 
1705 	/* Just pass in the packet to our lagg device */
1706 	m->m_pkthdr.rcvif = ifp;
1707 
1708 	return (m);
1709 }
1710 
1711 /*
1712  * 802.3ad LACP
1713  */
1714 
1715 static int
1716 lagg_lacp_attach(struct lagg_softc *sc)
1717 {
1718 	struct lagg_port *lp;
1719 	int error;
1720 
1721 	sc->sc_detach = lagg_lacp_detach;
1722 	sc->sc_port_create = lacp_port_create;
1723 	sc->sc_port_destroy = lacp_port_destroy;
1724 	sc->sc_linkstate = lacp_linkstate;
1725 	sc->sc_start = lagg_lacp_start;
1726 	sc->sc_input = lagg_lacp_input;
1727 	sc->sc_init = lacp_init;
1728 	sc->sc_stop = lacp_stop;
1729 	sc->sc_lladdr = lagg_lacp_lladdr;
1730 	sc->sc_req = lacp_req;
1731 	sc->sc_portreq = lacp_portreq;
1732 
1733 	error = lacp_attach(sc);
1734 	if (error)
1735 		return (error);
1736 
1737 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1738 		lacp_port_create(lp);
1739 
1740 	return (error);
1741 }
1742 
1743 static int
1744 lagg_lacp_detach(struct lagg_softc *sc)
1745 {
1746 	struct lagg_port *lp;
1747 	int error;
1748 
1749 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1750 		lacp_port_destroy(lp);
1751 
1752 	/* unlocking is safe here */
1753 	LAGG_WUNLOCK(sc);
1754 	error = lacp_detach(sc);
1755 	LAGG_WLOCK(sc);
1756 
1757 	return (error);
1758 }
1759 
1760 static void
1761 lagg_lacp_lladdr(struct lagg_softc *sc)
1762 {
1763 	struct lagg_port *lp;
1764 
1765 	/* purge all the lacp ports */
1766 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1767 		lacp_port_destroy(lp);
1768 
1769 	/* add them back in */
1770 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1771 		lacp_port_create(lp);
1772 }
1773 
1774 static int
1775 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1776 {
1777 	struct lagg_port *lp;
1778 
1779 	lp = lacp_select_tx_port(sc, m);
1780 	if (lp == NULL) {
1781 		m_freem(m);
1782 		return (EBUSY);
1783 	}
1784 
1785 	/* Send mbuf */
1786 	return (lagg_enqueue(lp->lp_ifp, m));
1787 }
1788 
1789 static struct mbuf *
1790 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1791 {
1792 	struct ifnet *ifp = sc->sc_ifp;
1793 	struct ether_header *eh;
1794 	u_short etype;
1795 
1796 	eh = mtod(m, struct ether_header *);
1797 	etype = ntohs(eh->ether_type);
1798 
1799 	/* Tap off LACP control messages */
1800 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
1801 		m = lacp_input(lp, m);
1802 		if (m == NULL)
1803 			return (NULL);
1804 	}
1805 
1806 	/*
1807 	 * If the port is not collecting or not in the active aggregator then
1808 	 * free and return.
1809 	 */
1810 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
1811 		m_freem(m);
1812 		return (NULL);
1813 	}
1814 
1815 	m->m_pkthdr.rcvif = ifp;
1816 	return (m);
1817 }
1818