xref: /freebsd/sys/net/if_lagg.c (revision 995dc984471c92c03daad19a1d35af46c086ef3e)
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_port_isactive(lp))
750 				rp->rp_flags |= LAGG_PORT_ACTIVE;
751 			break;
752 	}
753 
754 }
755 
756 static void
757 lagg_init(void *xsc)
758 {
759 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
760 	struct lagg_port *lp;
761 	struct ifnet *ifp = sc->sc_ifp;
762 
763 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
764 		return;
765 
766 	LAGG_WLOCK(sc);
767 
768 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
769 	/* Update the port lladdrs */
770 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
771 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
772 
773 	if (sc->sc_init != NULL)
774 		(*sc->sc_init)(sc);
775 
776 	LAGG_WUNLOCK(sc);
777 }
778 
779 static void
780 lagg_stop(struct lagg_softc *sc)
781 {
782 	struct ifnet *ifp = sc->sc_ifp;
783 
784 	LAGG_WLOCK_ASSERT(sc);
785 
786 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
787 		return;
788 
789 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
790 
791 	if (sc->sc_stop != NULL)
792 		(*sc->sc_stop)(sc);
793 }
794 
795 static int
796 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
797 {
798 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
799 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
800 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
801 	struct ifreq *ifr = (struct ifreq *)data;
802 	struct lagg_port *lp;
803 	struct ifnet *tpif;
804 	struct thread *td = curthread;
805 	char *buf, *outbuf;
806 	int count, buflen, len, error = 0;
807 
808 	bzero(&rpbuf, sizeof(rpbuf));
809 
810 	switch (cmd) {
811 	case SIOCGLAGG:
812 		LAGG_RLOCK(sc);
813 		count = 0;
814 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
815 			count++;
816 		buflen = count * sizeof(struct lagg_reqport);
817 		LAGG_RUNLOCK(sc);
818 
819 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
820 
821 		LAGG_RLOCK(sc);
822 		ra->ra_proto = sc->sc_proto;
823 		if (sc->sc_req != NULL)
824 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
825 
826 		count = 0;
827 		buf = outbuf;
828 		len = min(ra->ra_size, buflen);
829 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
830 			if (len < sizeof(rpbuf))
831 				break;
832 
833 			lagg_port2req(lp, &rpbuf);
834 			memcpy(buf, &rpbuf, sizeof(rpbuf));
835 			count++;
836 			buf += sizeof(rpbuf);
837 			len -= sizeof(rpbuf);
838 		}
839 		LAGG_RUNLOCK(sc);
840 		ra->ra_ports = count;
841 		ra->ra_size = count * sizeof(rpbuf);
842 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
843 		free(outbuf, M_TEMP);
844 		break;
845 	case SIOCSLAGG:
846 		error = priv_check(td, PRIV_NET_LAGG);
847 		if (error)
848 			break;
849 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
850 			error = EPROTONOSUPPORT;
851 			break;
852 		}
853 		if (sc->sc_proto != LAGG_PROTO_NONE) {
854 			LAGG_WLOCK(sc);
855 			error = sc->sc_detach(sc);
856 			/* Reset protocol and pointers */
857 			sc->sc_proto = LAGG_PROTO_NONE;
858 			sc->sc_detach = NULL;
859 			sc->sc_start = NULL;
860 			sc->sc_input = NULL;
861 			sc->sc_port_create = NULL;
862 			sc->sc_port_destroy = NULL;
863 			sc->sc_linkstate = NULL;
864 			sc->sc_init = NULL;
865 			sc->sc_stop = NULL;
866 			sc->sc_lladdr = NULL;
867 			sc->sc_req = NULL;
868 			sc->sc_portreq = NULL;
869 			LAGG_WUNLOCK(sc);
870 		}
871 		if (error != 0)
872 			break;
873 		for (int i = 0; i < (sizeof(lagg_protos) /
874 		    sizeof(lagg_protos[0])); i++) {
875 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
876 				if (sc->sc_ifflags & IFF_DEBUG)
877 					printf("%s: using proto %u\n",
878 					    sc->sc_ifname,
879 					    lagg_protos[i].ti_proto);
880 				LAGG_WLOCK(sc);
881 				sc->sc_proto = lagg_protos[i].ti_proto;
882 				if (sc->sc_proto != LAGG_PROTO_NONE)
883 					error = lagg_protos[i].ti_attach(sc);
884 				LAGG_WUNLOCK(sc);
885 				return (error);
886 			}
887 		}
888 		error = EPROTONOSUPPORT;
889 		break;
890 	case SIOCGLAGGPORT:
891 		if (rp->rp_portname[0] == '\0' ||
892 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
893 			error = EINVAL;
894 			break;
895 		}
896 
897 		LAGG_RLOCK(sc);
898 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
899 		    lp->lp_softc != sc) {
900 			error = ENOENT;
901 			LAGG_RUNLOCK(sc);
902 			break;
903 		}
904 
905 		lagg_port2req(lp, rp);
906 		LAGG_RUNLOCK(sc);
907 		break;
908 	case SIOCSLAGGPORT:
909 		error = priv_check(td, PRIV_NET_LAGG);
910 		if (error)
911 			break;
912 		if (rp->rp_portname[0] == '\0' ||
913 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
914 			error = EINVAL;
915 			break;
916 		}
917 		LAGG_WLOCK(sc);
918 		error = lagg_port_create(sc, tpif);
919 		LAGG_WUNLOCK(sc);
920 		break;
921 	case SIOCSLAGGDELPORT:
922 		error = priv_check(td, PRIV_NET_LAGG);
923 		if (error)
924 			break;
925 		if (rp->rp_portname[0] == '\0' ||
926 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
927 			error = EINVAL;
928 			break;
929 		}
930 
931 		LAGG_WLOCK(sc);
932 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
933 		    lp->lp_softc != sc) {
934 			error = ENOENT;
935 			LAGG_WUNLOCK(sc);
936 			break;
937 		}
938 
939 		error = lagg_port_destroy(lp, 1);
940 		LAGG_WUNLOCK(sc);
941 		break;
942 	case SIOCSIFFLAGS:
943 		/* Set flags on ports too */
944 		LAGG_WLOCK(sc);
945 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
946 			lagg_setflags(lp, 1);
947 		}
948 		LAGG_WUNLOCK(sc);
949 
950 		if (!(ifp->if_flags & IFF_UP) &&
951 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
952 			/*
953 			 * If interface is marked down and it is running,
954 			 * then stop and disable it.
955 			 */
956 			LAGG_WLOCK(sc);
957 			lagg_stop(sc);
958 			LAGG_WUNLOCK(sc);
959 		} else if ((ifp->if_flags & IFF_UP) &&
960 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
961 			/*
962 			 * If interface is marked up and it is stopped, then
963 			 * start it.
964 			 */
965 			(*ifp->if_init)(sc);
966 		}
967 		break;
968 	case SIOCADDMULTI:
969 	case SIOCDELMULTI:
970 		LAGG_WLOCK(sc);
971 		error = lagg_ether_setmulti(sc);
972 		LAGG_WUNLOCK(sc);
973 		break;
974 	case SIOCSIFMEDIA:
975 	case SIOCGIFMEDIA:
976 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
977 		break;
978 
979 	case SIOCSIFCAP:
980 	case SIOCSIFMTU:
981 		/* Do not allow the MTU or caps to be directly changed */
982 		error = EINVAL;
983 		break;
984 
985 	default:
986 		error = ether_ioctl(ifp, cmd, data);
987 		break;
988 	}
989 	return (error);
990 }
991 
992 static int
993 lagg_ether_setmulti(struct lagg_softc *sc)
994 {
995 	struct lagg_port *lp;
996 
997 	LAGG_WLOCK_ASSERT(sc);
998 
999 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1000 		/* First, remove any existing filter entries. */
1001 		lagg_ether_cmdmulti(lp, 0);
1002 		/* copy all addresses from the lagg interface to the port */
1003 		lagg_ether_cmdmulti(lp, 1);
1004 	}
1005 	return (0);
1006 }
1007 
1008 static int
1009 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1010 {
1011 	struct lagg_softc *sc = lp->lp_softc;
1012 	struct ifnet *ifp = lp->lp_ifp;
1013 	struct ifnet *scifp = sc->sc_ifp;
1014 	struct lagg_mc *mc;
1015 	struct ifmultiaddr *ifma, *rifma = NULL;
1016 	struct sockaddr_dl sdl;
1017 	int error;
1018 
1019 	LAGG_WLOCK_ASSERT(sc);
1020 
1021 	bzero((char *)&sdl, sizeof(sdl));
1022 	sdl.sdl_len = sizeof(sdl);
1023 	sdl.sdl_family = AF_LINK;
1024 	sdl.sdl_type = IFT_ETHER;
1025 	sdl.sdl_alen = ETHER_ADDR_LEN;
1026 	sdl.sdl_index = ifp->if_index;
1027 
1028 	if (set) {
1029 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1030 			if (ifma->ifma_addr->sa_family != AF_LINK)
1031 				continue;
1032 			bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1033 			    LLADDR(&sdl), ETHER_ADDR_LEN);
1034 
1035 			error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1036 			if (error)
1037 				return (error);
1038 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1039 			if (mc == NULL)
1040 				return (ENOMEM);
1041 			mc->mc_ifma = rifma;
1042 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1043 		}
1044 	} else {
1045 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1046 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1047 			if_delmulti_ifma(mc->mc_ifma);
1048 			free(mc, M_DEVBUF);
1049 		}
1050 	}
1051 	return (0);
1052 }
1053 
1054 /* Handle a ref counted flag that should be set on the lagg port as well */
1055 static int
1056 lagg_setflag(struct lagg_port *lp, int flag, int status,
1057 	     int (*func)(struct ifnet *, int))
1058 {
1059 	struct lagg_softc *sc = lp->lp_softc;
1060 	struct ifnet *scifp = sc->sc_ifp;
1061 	struct ifnet *ifp = lp->lp_ifp;
1062 	int error;
1063 
1064 	LAGG_WLOCK_ASSERT(sc);
1065 
1066 	status = status ? (scifp->if_flags & flag) : 0;
1067 	/* Now "status" contains the flag value or 0 */
1068 
1069 	/*
1070 	 * See if recorded ports status is different from what
1071 	 * we want it to be.  If it is, flip it.  We record ports
1072 	 * status in lp_ifflags so that we won't clear ports flag
1073 	 * we haven't set.  In fact, we don't clear or set ports
1074 	 * flags directly, but get or release references to them.
1075 	 * That's why we can be sure that recorded flags still are
1076 	 * in accord with actual ports flags.
1077 	 */
1078 	if (status != (lp->lp_ifflags & flag)) {
1079 		error = (*func)(ifp, status);
1080 		if (error)
1081 			return (error);
1082 		lp->lp_ifflags &= ~flag;
1083 		lp->lp_ifflags |= status;
1084 	}
1085 	return (0);
1086 }
1087 
1088 /*
1089  * Handle IFF_* flags that require certain changes on the lagg port
1090  * if "status" is true, update ports flags respective to the lagg
1091  * if "status" is false, forcedly clear the flags set on port.
1092  */
1093 static int
1094 lagg_setflags(struct lagg_port *lp, int status)
1095 {
1096 	int error, i;
1097 
1098 	for (i = 0; lagg_pflags[i].flag; i++) {
1099 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1100 		    status, lagg_pflags[i].func);
1101 		if (error)
1102 			return (error);
1103 	}
1104 	return (0);
1105 }
1106 
1107 static void
1108 lagg_start(struct ifnet *ifp)
1109 {
1110 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1111 	struct mbuf *m;
1112 	int error = 0;
1113 
1114 	LAGG_RLOCK(sc);
1115 	for (;; error = 0) {
1116 		IFQ_DEQUEUE(&ifp->if_snd, m);
1117 		if (m == NULL)
1118 			break;
1119 
1120 		ETHER_BPF_MTAP(ifp, m);
1121 
1122 		if (sc->sc_proto != LAGG_PROTO_NONE)
1123 			error = (*sc->sc_start)(sc, m);
1124 		else
1125 			m_freem(m);
1126 
1127 		if (error == 0)
1128 			ifp->if_opackets++;
1129 		else
1130 			ifp->if_oerrors++;
1131 	}
1132 	LAGG_RUNLOCK(sc);
1133 
1134 	return;
1135 }
1136 
1137 static struct mbuf *
1138 lagg_input(struct ifnet *ifp, struct mbuf *m)
1139 {
1140 	struct lagg_port *lp = ifp->if_lagg;
1141 	struct lagg_softc *sc = lp->lp_softc;
1142 	struct ifnet *scifp = sc->sc_ifp;
1143 
1144 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1145 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1146 	    sc->sc_proto == LAGG_PROTO_NONE) {
1147 		m_freem(m);
1148 		return (NULL);
1149 	}
1150 
1151 	LAGG_RLOCK(sc);
1152 	ETHER_BPF_MTAP(scifp, m);
1153 
1154 	m = (*sc->sc_input)(sc, lp, m);
1155 
1156 	if (m != NULL) {
1157 		scifp->if_ipackets++;
1158 		scifp->if_ibytes += m->m_pkthdr.len;
1159 
1160 		if (scifp->if_flags & IFF_MONITOR) {
1161 			m_freem(m);
1162 			m = NULL;
1163 		}
1164 	}
1165 
1166 	LAGG_RUNLOCK(sc);
1167 	return (m);
1168 }
1169 
1170 static int
1171 lagg_media_change(struct ifnet *ifp)
1172 {
1173 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1174 
1175 	if (sc->sc_ifflags & IFF_DEBUG)
1176 		printf("%s\n", __func__);
1177 
1178 	/* Ignore */
1179 	return (0);
1180 }
1181 
1182 static void
1183 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1184 {
1185 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1186 	struct lagg_port *lp;
1187 
1188 	imr->ifm_status = IFM_AVALID;
1189 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1190 
1191 	LAGG_RLOCK(sc);
1192 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1193 		if (LAGG_PORTACTIVE(lp))
1194 			imr->ifm_status |= IFM_ACTIVE;
1195 	}
1196 	LAGG_RUNLOCK(sc);
1197 }
1198 
1199 static void
1200 lagg_linkstate(struct lagg_softc *sc)
1201 {
1202 	struct lagg_port *lp;
1203 	int new_link = LINK_STATE_DOWN;
1204 
1205 	/* Our link is considered up if at least one of our ports is active */
1206 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1207 		if (lp->lp_link_state == LINK_STATE_UP) {
1208 			new_link = LINK_STATE_UP;
1209 			break;
1210 		}
1211 	}
1212 	if_link_state_change(sc->sc_ifp, new_link);
1213 }
1214 
1215 static void
1216 lagg_port_state(struct ifnet *ifp, int state)
1217 {
1218 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1219 	struct lagg_softc *sc = NULL;
1220 
1221 	if (lp != NULL)
1222 		sc = lp->lp_softc;
1223 	if (sc == NULL)
1224 		return;
1225 
1226 	LAGG_WLOCK(sc);
1227 	lagg_linkstate(sc);
1228 	if (sc->sc_linkstate != NULL)
1229 		(*sc->sc_linkstate)(lp);
1230 	LAGG_WUNLOCK(sc);
1231 }
1232 
1233 struct lagg_port *
1234 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1235 {
1236 	struct lagg_port *lp_next, *rval = NULL;
1237 	// int new_link = LINK_STATE_DOWN;
1238 
1239 	LAGG_RLOCK_ASSERT(sc);
1240 	/*
1241 	 * Search a port which reports an active link state.
1242 	 */
1243 
1244 	if (lp == NULL)
1245 		goto search;
1246 	if (LAGG_PORTACTIVE(lp)) {
1247 		rval = lp;
1248 		goto found;
1249 	}
1250 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1251 	    LAGG_PORTACTIVE(lp_next)) {
1252 		rval = lp_next;
1253 		goto found;
1254 	}
1255 
1256 search:
1257 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1258 		if (LAGG_PORTACTIVE(lp_next)) {
1259 			rval = lp_next;
1260 			goto found;
1261 		}
1262 	}
1263 
1264 found:
1265 	if (rval != NULL) {
1266 		/*
1267 		 * The IEEE 802.1D standard assumes that a lagg with
1268 		 * multiple ports is always full duplex. This is valid
1269 		 * for load sharing laggs and if at least two links
1270 		 * are active. Unfortunately, checking the latter would
1271 		 * be too expensive at this point.
1272 		 XXX
1273 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1274 		    (sc->sc_count > 1))
1275 			new_link = LINK_STATE_FULL_DUPLEX;
1276 		else
1277 			new_link = rval->lp_link_state;
1278 		 */
1279 	}
1280 
1281 	return (rval);
1282 }
1283 
1284 static const void *
1285 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1286 {
1287 	if (m->m_pkthdr.len < (off + len)) {
1288 		return (NULL);
1289 	} else if (m->m_len < (off + len)) {
1290 		m_copydata(m, off, len, buf);
1291 		return (buf);
1292 	}
1293 	return (mtod(m, char *) + off);
1294 }
1295 
1296 uint32_t
1297 lagg_hashmbuf(struct mbuf *m, uint32_t key)
1298 {
1299 	uint16_t etype;
1300 	uint32_t p = 0;
1301 	int off;
1302 	struct ether_header *eh;
1303 	struct ether_vlan_header vlanbuf;
1304 	const struct ether_vlan_header *vlan;
1305 #ifdef INET
1306 	const struct ip *ip;
1307 	struct ip ipbuf;
1308 #endif
1309 #ifdef INET6
1310 	const struct ip6_hdr *ip6;
1311 	struct ip6_hdr ip6buf;
1312 	uint32_t flow;
1313 #endif
1314 
1315 	off = sizeof(*eh);
1316 	if (m->m_len < off)
1317 		goto out;
1318 	eh = mtod(m, struct ether_header *);
1319 	etype = ntohs(eh->ether_type);
1320 	p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, key);
1321 	p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1322 
1323 	/* Special handling for encapsulating VLAN frames */
1324 	if (m->m_flags & M_VLANTAG) {
1325 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1326 		    sizeof(m->m_pkthdr.ether_vtag), p);
1327 	} else if (etype == ETHERTYPE_VLAN) {
1328 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &vlanbuf);
1329 		if (vlan == NULL)
1330 			goto out;
1331 
1332 		p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1333 		etype = ntohs(vlan->evl_proto);
1334 		off += sizeof(*vlan) - sizeof(*eh);
1335 	}
1336 
1337 	switch (etype) {
1338 #ifdef INET
1339 	case ETHERTYPE_IP:
1340 		ip = lagg_gethdr(m, off, sizeof(*ip), &ipbuf);
1341 		if (ip == NULL)
1342 			goto out;
1343 
1344 		p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1345 		p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1346 		break;
1347 #endif
1348 #ifdef INET6
1349 	case ETHERTYPE_IPV6:
1350 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &ip6buf);
1351 		if (ip6 == NULL)
1352 			goto out;
1353 
1354 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1355 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1356 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1357 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1358 		break;
1359 #endif
1360 	}
1361 out:
1362 	return (p);
1363 }
1364 
1365 int
1366 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1367 {
1368 	int error = 0;
1369 
1370 	IFQ_HANDOFF(ifp, m, error);
1371 	if (error)
1372 		ifp->if_oerrors++;
1373 	return (error);
1374 }
1375 
1376 /*
1377  * Simple round robin aggregation
1378  */
1379 
1380 static int
1381 lagg_rr_attach(struct lagg_softc *sc)
1382 {
1383 	sc->sc_detach = lagg_rr_detach;
1384 	sc->sc_start = lagg_rr_start;
1385 	sc->sc_input = lagg_rr_input;
1386 	sc->sc_port_create = NULL;
1387 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1388 	sc->sc_seq = 0;
1389 
1390 	return (0);
1391 }
1392 
1393 static int
1394 lagg_rr_detach(struct lagg_softc *sc)
1395 {
1396 	return (0);
1397 }
1398 
1399 static int
1400 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1401 {
1402 	struct lagg_port *lp;
1403 	uint32_t p;
1404 
1405 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1406 	p %= sc->sc_count;
1407 	lp = SLIST_FIRST(&sc->sc_ports);
1408 	while (p--)
1409 		lp = SLIST_NEXT(lp, lp_entries);
1410 
1411 	/*
1412 	 * Check the port's link state. This will return the next active
1413 	 * port if the link is down or the port is NULL.
1414 	 */
1415 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1416 		m_freem(m);
1417 		return (ENOENT);
1418 	}
1419 
1420 	/* Send mbuf */
1421 	return (lagg_enqueue(lp->lp_ifp, m));
1422 }
1423 
1424 static struct mbuf *
1425 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1426 {
1427 	struct ifnet *ifp = sc->sc_ifp;
1428 
1429 	/* Just pass in the packet to our lagg device */
1430 	m->m_pkthdr.rcvif = ifp;
1431 
1432 	return (m);
1433 }
1434 
1435 /*
1436  * Active failover
1437  */
1438 
1439 static int
1440 lagg_fail_attach(struct lagg_softc *sc)
1441 {
1442 	sc->sc_detach = lagg_fail_detach;
1443 	sc->sc_start = lagg_fail_start;
1444 	sc->sc_input = lagg_fail_input;
1445 	sc->sc_port_create = NULL;
1446 	sc->sc_port_destroy = NULL;
1447 
1448 	return (0);
1449 }
1450 
1451 static int
1452 lagg_fail_detach(struct lagg_softc *sc)
1453 {
1454 	return (0);
1455 }
1456 
1457 static int
1458 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1459 {
1460 	struct lagg_port *lp;
1461 
1462 	/* Use the master port if active or the next available port */
1463 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1464 		m_freem(m);
1465 		return (ENOENT);
1466 	}
1467 
1468 	/* Send mbuf */
1469 	return (lagg_enqueue(lp->lp_ifp, m));
1470 }
1471 
1472 static struct mbuf *
1473 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1474 {
1475 	struct ifnet *ifp = sc->sc_ifp;
1476 	struct lagg_port *tmp_tp;
1477 
1478 	if (lp == sc->sc_primary) {
1479 		m->m_pkthdr.rcvif = ifp;
1480 		return (m);
1481 	}
1482 
1483 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1484 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1485 		/*
1486 		 * If tmp_tp is null, we've recieved a packet when all
1487 		 * our links are down. Weird, but process it anyways.
1488 		 */
1489 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1490 			m->m_pkthdr.rcvif = ifp;
1491 			return (m);
1492 		}
1493 	}
1494 
1495 	m_freem(m);
1496 	return (NULL);
1497 }
1498 
1499 /*
1500  * Loadbalancing
1501  */
1502 
1503 static int
1504 lagg_lb_attach(struct lagg_softc *sc)
1505 {
1506 	struct lagg_port *lp;
1507 	struct lagg_lb *lb;
1508 
1509 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1510 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1511 		return (ENOMEM);
1512 
1513 	sc->sc_detach = lagg_lb_detach;
1514 	sc->sc_start = lagg_lb_start;
1515 	sc->sc_input = lagg_lb_input;
1516 	sc->sc_port_create = lagg_lb_port_create;
1517 	sc->sc_port_destroy = lagg_lb_port_destroy;
1518 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1519 
1520 	lb->lb_key = arc4random();
1521 	sc->sc_psc = (caddr_t)lb;
1522 
1523 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1524 		lagg_lb_port_create(lp);
1525 
1526 	return (0);
1527 }
1528 
1529 static int
1530 lagg_lb_detach(struct lagg_softc *sc)
1531 {
1532 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1533 	if (lb != NULL)
1534 		free(lb, M_DEVBUF);
1535 	return (0);
1536 }
1537 
1538 static int
1539 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1540 {
1541 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1542 	struct lagg_port *lp_next;
1543 	int i = 0;
1544 
1545 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1546 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1547 		if (lp_next == lp)
1548 			continue;
1549 		if (i >= LAGG_MAX_PORTS)
1550 			return (EINVAL);
1551 		if (sc->sc_ifflags & IFF_DEBUG)
1552 			printf("%s: port %s at index %d\n",
1553 			    sc->sc_ifname, lp_next->lp_ifname, i);
1554 		lb->lb_ports[i++] = lp_next;
1555 	}
1556 
1557 	return (0);
1558 }
1559 
1560 static int
1561 lagg_lb_port_create(struct lagg_port *lp)
1562 {
1563 	struct lagg_softc *sc = lp->lp_softc;
1564 	return (lagg_lb_porttable(sc, NULL));
1565 }
1566 
1567 static void
1568 lagg_lb_port_destroy(struct lagg_port *lp)
1569 {
1570 	struct lagg_softc *sc = lp->lp_softc;
1571 	lagg_lb_porttable(sc, lp);
1572 }
1573 
1574 static int
1575 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1576 {
1577 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1578 	struct lagg_port *lp = NULL;
1579 	uint32_t p = 0;
1580 	int idx;
1581 
1582 	p = lagg_hashmbuf(m, lb->lb_key);
1583 	if ((idx = p % sc->sc_count) >= LAGG_MAX_PORTS) {
1584 		m_freem(m);
1585 		return (EINVAL);
1586 	}
1587 	lp = lb->lb_ports[idx];
1588 
1589 	/*
1590 	 * Check the port's link state. This will return the next active
1591 	 * port if the link is down or the port is NULL.
1592 	 */
1593 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1594 		m_freem(m);
1595 		return (ENOENT);
1596 	}
1597 
1598 	/* Send mbuf */
1599 	return (lagg_enqueue(lp->lp_ifp, m));
1600 }
1601 
1602 static struct mbuf *
1603 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1604 {
1605 	struct ifnet *ifp = sc->sc_ifp;
1606 
1607 	/* Just pass in the packet to our lagg device */
1608 	m->m_pkthdr.rcvif = ifp;
1609 
1610 	return (m);
1611 }
1612 
1613 /*
1614  * 802.3ad LACP
1615  */
1616 
1617 static int
1618 lagg_lacp_attach(struct lagg_softc *sc)
1619 {
1620 	struct lagg_port *lp;
1621 	int error;
1622 
1623 	sc->sc_detach = lagg_lacp_detach;
1624 	sc->sc_port_create = lacp_port_create;
1625 	sc->sc_port_destroy = lacp_port_destroy;
1626 	sc->sc_linkstate = lacp_linkstate;
1627 	sc->sc_start = lagg_lacp_start;
1628 	sc->sc_input = lagg_lacp_input;
1629 	sc->sc_init = lacp_init;
1630 	sc->sc_stop = lacp_stop;
1631 	sc->sc_lladdr = lagg_lacp_lladdr;
1632 	sc->sc_req = lacp_req;
1633 	sc->sc_portreq = lacp_portreq;
1634 
1635 	error = lacp_attach(sc);
1636 	if (error)
1637 		return (error);
1638 
1639 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1640 		lacp_port_create(lp);
1641 
1642 	return (error);
1643 }
1644 
1645 static int
1646 lagg_lacp_detach(struct lagg_softc *sc)
1647 {
1648 	struct lagg_port *lp;
1649 	int error;
1650 
1651 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1652 		lacp_port_destroy(lp);
1653 
1654 	/* unlocking is safe here */
1655 	LAGG_WUNLOCK(sc);
1656 	error = lacp_detach(sc);
1657 	LAGG_WLOCK(sc);
1658 
1659 	return (error);
1660 }
1661 
1662 static void
1663 lagg_lacp_lladdr(struct lagg_softc *sc)
1664 {
1665 	struct lagg_port *lp;
1666 
1667 	/* purge all the lacp ports */
1668 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1669 		lacp_port_destroy(lp);
1670 
1671 	/* add them back in */
1672 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1673 		lacp_port_create(lp);
1674 }
1675 
1676 static int
1677 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1678 {
1679 	struct lagg_port *lp;
1680 
1681 	lp = lacp_select_tx_port(sc, m);
1682 	if (lp == NULL) {
1683 		m_freem(m);
1684 		return (EBUSY);
1685 	}
1686 
1687 	/* Send mbuf */
1688 	return (lagg_enqueue(lp->lp_ifp, m));
1689 }
1690 
1691 static struct mbuf *
1692 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1693 {
1694 	struct ifnet *ifp = sc->sc_ifp;
1695 	struct ether_header *eh;
1696 	u_short etype;
1697 
1698 	eh = mtod(m, struct ether_header *);
1699 	etype = ntohs(eh->ether_type);
1700 
1701 	/* Tap off LACP control messages */
1702 	if (etype == ETHERTYPE_SLOW) {
1703 		m = lacp_input(lp, m);
1704 		if (m == NULL)
1705 			return (NULL);
1706 	}
1707 
1708 	/*
1709 	 * If the port is not collecting or not in the active aggregator then
1710 	 * free and return.
1711 	 */
1712 	if ((lp->lp_flags & LAGG_PORT_COLLECTING) == 0 ||
1713 	    lacp_port_isactive(lp) == 0) {
1714 		m_freem(m);
1715 		return (NULL);
1716 	}
1717 
1718 	m->m_pkthdr.rcvif = ifp;
1719 	return (m);
1720 }
1721