xref: /freebsd/sys/net/if_lagg.c (revision 6574b8ed19b093f0af09501d2c9676c28993cb97)
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/rmlock.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_media.h>
50 #include <net/if_types.h>
51 #include <net/if_var.h>
52 #include <net/bpf.h>
53 
54 #if defined(INET) || defined(INET6)
55 #include <netinet/in.h>
56 #include <netinet/ip.h>
57 #endif
58 #ifdef INET
59 #include <netinet/in_systm.h>
60 #include <netinet/if_ether.h>
61 #endif
62 
63 #ifdef INET6
64 #include <netinet/ip6.h>
65 #include <netinet6/in6_var.h>
66 #include <netinet6/in6_ifattach.h>
67 #endif
68 
69 #include <net/if_vlan_var.h>
70 #include <net/if_lagg.h>
71 #include <net/ieee8023ad_lacp.h>
72 
73 /* Special flags we should propagate to the lagg ports. */
74 static struct {
75 	int flag;
76 	int (*func)(struct ifnet *, int);
77 } lagg_pflags[] = {
78 	{IFF_PROMISC, ifpromisc},
79 	{IFF_ALLMULTI, if_allmulti},
80 	{0, NULL}
81 };
82 
83 SLIST_HEAD(__trhead, lagg_softc) lagg_list;	/* list of laggs */
84 static struct mtx	lagg_list_mtx;
85 eventhandler_tag	lagg_detach_cookie = NULL;
86 
87 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
88 static void	lagg_clone_destroy(struct ifnet *);
89 static struct if_clone *lagg_cloner;
90 static const char laggname[] = "lagg";
91 
92 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
93 static void	lagg_capabilities(struct lagg_softc *);
94 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *);
95 static void	lagg_port_setlladdr(void *, int);
96 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
97 static int	lagg_port_destroy(struct lagg_port *, int);
98 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
99 static void	lagg_linkstate(struct lagg_softc *);
100 static void	lagg_port_state(struct ifnet *, int);
101 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
102 static int	lagg_port_output(struct ifnet *, struct mbuf *,
103 		    const struct sockaddr *, struct route *);
104 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
105 #ifdef LAGG_PORT_STACKING
106 static int	lagg_port_checkstacking(struct lagg_softc *);
107 #endif
108 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
109 static void	lagg_init(void *);
110 static void	lagg_stop(struct lagg_softc *);
111 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
112 static int	lagg_ether_setmulti(struct lagg_softc *);
113 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
114 static	int	lagg_setflag(struct lagg_port *, int, int,
115 		    int (*func)(struct ifnet *, int));
116 static	int	lagg_setflags(struct lagg_port *, int status);
117 static int	lagg_transmit(struct ifnet *, struct mbuf *);
118 static void	lagg_qflush(struct ifnet *);
119 static int	lagg_media_change(struct ifnet *);
120 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
121 static struct lagg_port *lagg_link_active(struct lagg_softc *,
122 	    struct lagg_port *);
123 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
124 static int	lagg_sysctl_active(SYSCTL_HANDLER_ARGS);
125 
126 /* Simple round robin */
127 static int	lagg_rr_attach(struct lagg_softc *);
128 static int	lagg_rr_detach(struct lagg_softc *);
129 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
130 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
131 		    struct mbuf *);
132 
133 /* Active failover */
134 static int	lagg_fail_attach(struct lagg_softc *);
135 static int	lagg_fail_detach(struct lagg_softc *);
136 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
137 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
138 		    struct mbuf *);
139 
140 /* Loadbalancing */
141 static int	lagg_lb_attach(struct lagg_softc *);
142 static int	lagg_lb_detach(struct lagg_softc *);
143 static int	lagg_lb_port_create(struct lagg_port *);
144 static void	lagg_lb_port_destroy(struct lagg_port *);
145 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
146 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
147 		    struct mbuf *);
148 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
149 
150 /* 802.3ad LACP */
151 static int	lagg_lacp_attach(struct lagg_softc *);
152 static int	lagg_lacp_detach(struct lagg_softc *);
153 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
154 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
155 		    struct mbuf *);
156 static void	lagg_lacp_lladdr(struct lagg_softc *);
157 
158 static void	lagg_callout(void *);
159 
160 /* lagg protocol table */
161 static const struct {
162 	int			ti_proto;
163 	int			(*ti_attach)(struct lagg_softc *);
164 } lagg_protos[] = {
165 	{ LAGG_PROTO_ROUNDROBIN,	lagg_rr_attach },
166 	{ LAGG_PROTO_FAILOVER,		lagg_fail_attach },
167 	{ LAGG_PROTO_LOADBALANCE,	lagg_lb_attach },
168 	{ LAGG_PROTO_ETHERCHANNEL,	lagg_lb_attach },
169 	{ LAGG_PROTO_LACP,		lagg_lacp_attach },
170 	{ LAGG_PROTO_NONE,		NULL }
171 };
172 
173 SYSCTL_DECL(_net_link);
174 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
175     "Link Aggregation");
176 
177 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */
178 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW,
179     &lagg_failover_rx_all, 0,
180     "Accept input from any interface in a failover lagg");
181 static int def_use_flowid = 1; /* Default value for using M_FLOWID */
182 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
183     &def_use_flowid, 0,
184     "Default setting for using flow id for load sharing");
185 static int def_flowid_shift = 16; /* Default value for using M_FLOWID */
186 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
187     &def_flowid_shift, 0,
188     "Default setting for flowid shift for load sharing");
189 
190 static int
191 lagg_modevent(module_t mod, int type, void *data)
192 {
193 
194 	switch (type) {
195 	case MOD_LOAD:
196 		mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
197 		SLIST_INIT(&lagg_list);
198 		lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
199 		    lagg_clone_destroy, 0);
200 		lagg_input_p = lagg_input;
201 		lagg_linkstate_p = lagg_port_state;
202 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
203 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
204 		    EVENTHANDLER_PRI_ANY);
205 		break;
206 	case MOD_UNLOAD:
207 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
208 		    lagg_detach_cookie);
209 		if_clone_detach(lagg_cloner);
210 		lagg_input_p = NULL;
211 		lagg_linkstate_p = NULL;
212 		mtx_destroy(&lagg_list_mtx);
213 		break;
214 	default:
215 		return (EOPNOTSUPP);
216 	}
217 	return (0);
218 }
219 
220 static moduledata_t lagg_mod = {
221 	"if_lagg",
222 	lagg_modevent,
223 	0
224 };
225 
226 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
227 MODULE_VERSION(if_lagg, 1);
228 
229 /*
230  * This routine is run via an vlan
231  * config EVENT
232  */
233 static void
234 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
235 {
236         struct lagg_softc       *sc = ifp->if_softc;
237         struct lagg_port        *lp;
238         struct rm_priotracker   tracker;
239 
240         if (ifp->if_softc !=  arg)   /* Not our event */
241                 return;
242 
243         LAGG_RLOCK(sc, &tracker);
244         if (!SLIST_EMPTY(&sc->sc_ports)) {
245                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
246                         EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
247         }
248         LAGG_RUNLOCK(sc, &tracker);
249 }
250 
251 /*
252  * This routine is run via an vlan
253  * unconfig EVENT
254  */
255 static void
256 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
257 {
258         struct lagg_softc       *sc = ifp->if_softc;
259         struct lagg_port        *lp;
260         struct rm_priotracker   tracker;
261 
262         if (ifp->if_softc !=  arg)   /* Not our event */
263                 return;
264 
265         LAGG_RLOCK(sc, &tracker);
266         if (!SLIST_EMPTY(&sc->sc_ports)) {
267                 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
268                         EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
269         }
270         LAGG_RUNLOCK(sc, &tracker);
271 }
272 
273 static int
274 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
275 {
276 	struct lagg_softc *sc;
277 	struct ifnet *ifp;
278 	int i, error = 0;
279 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
280 	struct sysctl_oid *oid;
281 	char num[14];			/* sufficient for 32 bits */
282 
283 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
284 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
285 	if (ifp == NULL) {
286 		free(sc, M_DEVBUF);
287 		return (ENOSPC);
288 	}
289 
290 	sc->sc_ipackets = counter_u64_alloc(M_WAITOK);
291 	sc->sc_opackets = counter_u64_alloc(M_WAITOK);
292 	sc->sc_ibytes = counter_u64_alloc(M_WAITOK);
293 	sc->sc_obytes = counter_u64_alloc(M_WAITOK);
294 
295 	sysctl_ctx_init(&sc->ctx);
296 	snprintf(num, sizeof(num), "%u", unit);
297 	sc->use_flowid = def_use_flowid;
298 	sc->flowid_shift = def_flowid_shift;
299 	sc->sc_oid = oid = SYSCTL_ADD_NODE(&sc->ctx,
300 		&SYSCTL_NODE_CHILDREN(_net_link, lagg),
301 		OID_AUTO, num, CTLFLAG_RD, NULL, "");
302 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
303 		"use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid,
304 		sc->use_flowid, "Use flow id for load sharing");
305 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
306 		"flowid_shift", CTLTYPE_INT|CTLFLAG_RW, &sc->flowid_shift,
307 		sc->flowid_shift,
308 		"Shift flowid bits to prevent multiqueue collisions");
309 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
310 		"count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count,
311 		"Total number of ports");
312 	SYSCTL_ADD_PROC(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
313 		"active", CTLTYPE_INT|CTLFLAG_RD, sc, 0, lagg_sysctl_active,
314 		"I", "Total number of active ports");
315 	SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
316 		"flapping", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_flapping,
317 		sc->sc_flapping, "Total number of port change events");
318 	/* Hash all layers by default */
319 	sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4;
320 
321 	sc->sc_proto = LAGG_PROTO_NONE;
322 	for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
323 		if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
324 			sc->sc_proto = lagg_protos[i].ti_proto;
325 			if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
326 				if_free(ifp);
327 				free(sc, M_DEVBUF);
328 				return (error);
329 			}
330 			break;
331 		}
332 	}
333 	LAGG_LOCK_INIT(sc);
334 	LAGG_CALLOUT_LOCK_INIT(sc);
335 	SLIST_INIT(&sc->sc_ports);
336 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
337 
338 	/*
339 	 * This uses the callout lock rather than the rmlock; one can't
340 	 * hold said rmlock during SWI.
341 	 */
342 	callout_init_mtx(&sc->sc_callout, &sc->sc_call_mtx, 0);
343 
344 	/* Initialise pseudo media types */
345 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
346 	    lagg_media_status);
347 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
348 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
349 
350 	if_initname(ifp, laggname, unit);
351 	ifp->if_softc = sc;
352 	ifp->if_transmit = lagg_transmit;
353 	ifp->if_qflush = lagg_qflush;
354 	ifp->if_init = lagg_init;
355 	ifp->if_ioctl = lagg_ioctl;
356 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
357 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
358 
359 	/*
360 	 * Attach as an ordinary ethernet device, children will be attached
361 	 * as special device IFT_IEEE8023ADLAG.
362 	 */
363 	ether_ifattach(ifp, eaddr);
364 
365 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
366 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
367 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
368 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
369 
370 	/* Insert into the global list of laggs */
371 	mtx_lock(&lagg_list_mtx);
372 	SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
373 	mtx_unlock(&lagg_list_mtx);
374 
375 	callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
376 
377 	return (0);
378 }
379 
380 static void
381 lagg_clone_destroy(struct ifnet *ifp)
382 {
383 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
384 	struct lagg_port *lp;
385 
386 	LAGG_WLOCK(sc);
387 
388 	lagg_stop(sc);
389 	ifp->if_flags &= ~IFF_UP;
390 
391 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
392 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
393 
394 	/* Shutdown and remove lagg ports */
395 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
396 		lagg_port_destroy(lp, 1);
397 	/* Unhook the aggregation protocol */
398 	if (sc->sc_detach != NULL)
399 		(*sc->sc_detach)(sc);
400 
401 	LAGG_WUNLOCK(sc);
402 
403 	sysctl_ctx_free(&sc->ctx);
404 	ifmedia_removeall(&sc->sc_media);
405 	ether_ifdetach(ifp);
406 	if_free(ifp);
407 
408 	/* This grabs sc_callout_mtx, serialising it correctly */
409 	callout_drain(&sc->sc_callout);
410 
411 	/* At this point it's drained; we can free this */
412 	counter_u64_free(sc->sc_ipackets);
413 	counter_u64_free(sc->sc_opackets);
414 	counter_u64_free(sc->sc_ibytes);
415 	counter_u64_free(sc->sc_obytes);
416 
417 	mtx_lock(&lagg_list_mtx);
418 	SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
419 	mtx_unlock(&lagg_list_mtx);
420 
421 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
422 	LAGG_LOCK_DESTROY(sc);
423 	LAGG_CALLOUT_LOCK_DESTROY(sc);
424 	free(sc, M_DEVBUF);
425 }
426 
427 static void
428 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
429 {
430 	struct ifnet *ifp = sc->sc_ifp;
431 
432 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
433 		return;
434 
435 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
436 	/* Let the protocol know the MAC has changed */
437 	if (sc->sc_lladdr != NULL)
438 		(*sc->sc_lladdr)(sc);
439 	EVENTHANDLER_INVOKE(iflladdr_event, ifp);
440 }
441 
442 static void
443 lagg_capabilities(struct lagg_softc *sc)
444 {
445 	struct lagg_port *lp;
446 	int cap = ~0, ena = ~0;
447 	u_long hwa = ~0UL;
448 #if defined(INET) || defined(INET6)
449 	u_int hw_tsomax = IP_MAXPACKET;	/* Initialize to the maximum value. */
450 #else
451 	u_int hw_tsomax = ~0;	/* if_hw_tsomax is only for INET/INET6, but.. */
452 #endif
453 
454 	LAGG_WLOCK_ASSERT(sc);
455 
456 	/* Get capabilities from the lagg ports */
457 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
458 		cap &= lp->lp_ifp->if_capabilities;
459 		ena &= lp->lp_ifp->if_capenable;
460 		hwa &= lp->lp_ifp->if_hwassist;
461 		/* Set to the minimum value of the lagg ports. */
462 		if (lp->lp_ifp->if_hw_tsomax < hw_tsomax &&
463 		    lp->lp_ifp->if_hw_tsomax > 0)
464 			hw_tsomax = lp->lp_ifp->if_hw_tsomax;
465 	}
466 	cap = (cap == ~0 ? 0 : cap);
467 	ena = (ena == ~0 ? 0 : ena);
468 	hwa = (hwa == ~0 ? 0 : hwa);
469 
470 	if (sc->sc_ifp->if_capabilities != cap ||
471 	    sc->sc_ifp->if_capenable != ena ||
472 	    sc->sc_ifp->if_hwassist != hwa ||
473 	    sc->sc_ifp->if_hw_tsomax != hw_tsomax) {
474 		sc->sc_ifp->if_capabilities = cap;
475 		sc->sc_ifp->if_capenable = ena;
476 		sc->sc_ifp->if_hwassist = hwa;
477 		sc->sc_ifp->if_hw_tsomax = hw_tsomax;
478 		getmicrotime(&sc->sc_ifp->if_lastchange);
479 
480 		if (sc->sc_ifflags & IFF_DEBUG)
481 			if_printf(sc->sc_ifp,
482 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
483 	}
484 }
485 
486 static void
487 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
488 {
489 	struct lagg_softc *sc = lp->lp_softc;
490 	struct ifnet *ifp = lp->lp_ifp;
491 	struct lagg_llq *llq;
492 	int pending = 0;
493 
494 	LAGG_WLOCK_ASSERT(sc);
495 
496 	if (lp->lp_detaching ||
497 	    memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
498 		return;
499 
500 	/* Check to make sure its not already queued to be changed */
501 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
502 		if (llq->llq_ifp == ifp) {
503 			pending = 1;
504 			break;
505 		}
506 	}
507 
508 	if (!pending) {
509 		llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
510 		if (llq == NULL)	/* XXX what to do */
511 			return;
512 	}
513 
514 	/* Update the lladdr even if pending, it may have changed */
515 	llq->llq_ifp = ifp;
516 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
517 
518 	if (!pending)
519 		SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
520 
521 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
522 }
523 
524 /*
525  * Set the interface MAC address from a taskqueue to avoid a LOR.
526  */
527 static void
528 lagg_port_setlladdr(void *arg, int pending)
529 {
530 	struct lagg_softc *sc = (struct lagg_softc *)arg;
531 	struct lagg_llq *llq, *head;
532 	struct ifnet *ifp;
533 	int error;
534 
535 	/* Grab a local reference of the queue and remove it from the softc */
536 	LAGG_WLOCK(sc);
537 	head = SLIST_FIRST(&sc->sc_llq_head);
538 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
539 	LAGG_WUNLOCK(sc);
540 
541 	/*
542 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
543 	 * unlocked as we have the only reference to it.
544 	 */
545 	for (llq = head; llq != NULL; llq = head) {
546 		ifp = llq->llq_ifp;
547 
548 		/* Set the link layer address */
549 		CURVNET_SET(ifp->if_vnet);
550 		error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
551 		CURVNET_RESTORE();
552 		if (error)
553 			printf("%s: setlladdr failed on %s\n", __func__,
554 			    ifp->if_xname);
555 
556 		head = SLIST_NEXT(llq, llq_entries);
557 		free(llq, M_DEVBUF);
558 	}
559 }
560 
561 static int
562 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
563 {
564 	struct lagg_softc *sc_ptr;
565 	struct lagg_port *lp;
566 	int error = 0;
567 
568 	LAGG_WLOCK_ASSERT(sc);
569 
570 	/* Limit the maximal number of lagg ports */
571 	if (sc->sc_count >= LAGG_MAX_PORTS)
572 		return (ENOSPC);
573 
574 	/* Check if port has already been associated to a lagg */
575 	if (ifp->if_lagg != NULL) {
576 		/* Port is already in the current lagg? */
577 		lp = (struct lagg_port *)ifp->if_lagg;
578 		if (lp->lp_softc == sc)
579 			return (EEXIST);
580 		return (EBUSY);
581 	}
582 
583 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
584 	if (ifp->if_type != IFT_ETHER)
585 		return (EPROTONOSUPPORT);
586 
587 #ifdef INET6
588 	/*
589 	 * The member interface should not have inet6 address because
590 	 * two interfaces with a valid link-local scope zone must not be
591 	 * merged in any form.  This restriction is needed to
592 	 * prevent violation of link-local scope zone.  Attempts to
593 	 * add a member interface which has inet6 addresses triggers
594 	 * removal of all inet6 addresses on the member interface.
595 	 */
596 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
597 		if (in6ifa_llaonifp(lp->lp_ifp)) {
598 			in6_ifdetach(lp->lp_ifp);
599 			if_printf(sc->sc_ifp,
600 			    "IPv6 addresses on %s have been removed "
601 			    "before adding it as a member to prevent "
602 			    "IPv6 address scope violation.\n",
603 			    lp->lp_ifp->if_xname);
604 		}
605 	}
606 	if (in6ifa_llaonifp(ifp)) {
607 		in6_ifdetach(ifp);
608 		if_printf(sc->sc_ifp,
609 		    "IPv6 addresses on %s have been removed "
610 		    "before adding it as a member to prevent "
611 		    "IPv6 address scope violation.\n",
612 		    ifp->if_xname);
613 	}
614 #endif
615 	/* Allow the first Ethernet member to define the MTU */
616 	if (SLIST_EMPTY(&sc->sc_ports))
617 		sc->sc_ifp->if_mtu = ifp->if_mtu;
618 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
619 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
620 		    ifp->if_xname);
621 		return (EINVAL);
622 	}
623 
624 	if ((lp = malloc(sizeof(struct lagg_port),
625 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
626 		return (ENOMEM);
627 
628 	/* Check if port is a stacked lagg */
629 	mtx_lock(&lagg_list_mtx);
630 	SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
631 		if (ifp == sc_ptr->sc_ifp) {
632 			mtx_unlock(&lagg_list_mtx);
633 			free(lp, M_DEVBUF);
634 			return (EINVAL);
635 			/* XXX disable stacking for the moment, its untested */
636 #ifdef LAGG_PORT_STACKING
637 			lp->lp_flags |= LAGG_PORT_STACK;
638 			if (lagg_port_checkstacking(sc_ptr) >=
639 			    LAGG_MAX_STACKING) {
640 				mtx_unlock(&lagg_list_mtx);
641 				free(lp, M_DEVBUF);
642 				return (E2BIG);
643 			}
644 #endif
645 		}
646 	}
647 	mtx_unlock(&lagg_list_mtx);
648 
649 	/* Change the interface type */
650 	lp->lp_iftype = ifp->if_type;
651 	ifp->if_type = IFT_IEEE8023ADLAG;
652 	ifp->if_lagg = lp;
653 	lp->lp_ioctl = ifp->if_ioctl;
654 	ifp->if_ioctl = lagg_port_ioctl;
655 	lp->lp_output = ifp->if_output;
656 	ifp->if_output = lagg_port_output;
657 
658 	lp->lp_ifp = ifp;
659 	lp->lp_softc = sc;
660 
661 	/* Save port link layer address */
662 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
663 
664 	if (SLIST_EMPTY(&sc->sc_ports)) {
665 		sc->sc_primary = lp;
666 		lagg_lladdr(sc, IF_LLADDR(ifp));
667 	} else {
668 		/* Update link layer address for this port */
669 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
670 	}
671 
672 	/* Insert into the list of ports */
673 	SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
674 	sc->sc_count++;
675 
676 	/* Update lagg capabilities */
677 	lagg_capabilities(sc);
678 	lagg_linkstate(sc);
679 
680 	/* Add multicast addresses and interface flags to this port */
681 	lagg_ether_cmdmulti(lp, 1);
682 	lagg_setflags(lp, 1);
683 
684 	if (sc->sc_port_create != NULL)
685 		error = (*sc->sc_port_create)(lp);
686 	if (error) {
687 		/* remove the port again, without calling sc_port_destroy */
688 		lagg_port_destroy(lp, 0);
689 		return (error);
690 	}
691 
692 	return (error);
693 }
694 
695 #ifdef LAGG_PORT_STACKING
696 static int
697 lagg_port_checkstacking(struct lagg_softc *sc)
698 {
699 	struct lagg_softc *sc_ptr;
700 	struct lagg_port *lp;
701 	int m = 0;
702 
703 	LAGG_WLOCK_ASSERT(sc);
704 
705 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
706 		if (lp->lp_flags & LAGG_PORT_STACK) {
707 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
708 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
709 		}
710 	}
711 
712 	return (m + 1);
713 }
714 #endif
715 
716 static int
717 lagg_port_destroy(struct lagg_port *lp, int runpd)
718 {
719 	struct lagg_softc *sc = lp->lp_softc;
720 	struct lagg_port *lp_ptr;
721 	struct lagg_llq *llq;
722 	struct ifnet *ifp = lp->lp_ifp;
723 
724 	LAGG_WLOCK_ASSERT(sc);
725 
726 	if (runpd && sc->sc_port_destroy != NULL)
727 		(*sc->sc_port_destroy)(lp);
728 
729 	/*
730 	 * Remove multicast addresses and interface flags from this port and
731 	 * reset the MAC address, skip if the interface is being detached.
732 	 */
733 	if (!lp->lp_detaching) {
734 		lagg_ether_cmdmulti(lp, 0);
735 		lagg_setflags(lp, 0);
736 		lagg_port_lladdr(lp, lp->lp_lladdr);
737 	}
738 
739 	/* Restore interface */
740 	ifp->if_type = lp->lp_iftype;
741 	ifp->if_ioctl = lp->lp_ioctl;
742 	ifp->if_output = lp->lp_output;
743 	ifp->if_lagg = NULL;
744 
745 	/* Finally, remove the port from the lagg */
746 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
747 	sc->sc_count--;
748 
749 	/* Update the primary interface */
750 	if (lp == sc->sc_primary) {
751 		uint8_t lladdr[ETHER_ADDR_LEN];
752 
753 		if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
754 			bzero(&lladdr, ETHER_ADDR_LEN);
755 		} else {
756 			bcopy(lp_ptr->lp_lladdr,
757 			    lladdr, ETHER_ADDR_LEN);
758 		}
759 		lagg_lladdr(sc, lladdr);
760 		sc->sc_primary = lp_ptr;
761 
762 		/* Update link layer address for each port */
763 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
764 			lagg_port_lladdr(lp_ptr, lladdr);
765 	}
766 
767 	/* Remove any pending lladdr changes from the queue */
768 	if (lp->lp_detaching) {
769 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
770 			if (llq->llq_ifp == ifp) {
771 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
772 				    llq_entries);
773 				free(llq, M_DEVBUF);
774 				break;	/* Only appears once */
775 			}
776 		}
777 	}
778 
779 	if (lp->lp_ifflags)
780 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
781 
782 	free(lp, M_DEVBUF);
783 
784 	/* Update lagg capabilities */
785 	lagg_capabilities(sc);
786 	lagg_linkstate(sc);
787 
788 	return (0);
789 }
790 
791 static int
792 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
793 {
794 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
795 	struct lagg_softc *sc;
796 	struct lagg_port *lp = NULL;
797 	int error = 0;
798 	struct rm_priotracker tracker;
799 
800 	/* Should be checked by the caller */
801 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
802 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
803 		goto fallback;
804 
805 	switch (cmd) {
806 	case SIOCGLAGGPORT:
807 		if (rp->rp_portname[0] == '\0' ||
808 		    ifunit(rp->rp_portname) != ifp) {
809 			error = EINVAL;
810 			break;
811 		}
812 
813 		LAGG_RLOCK(sc, &tracker);
814 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
815 			error = ENOENT;
816 			LAGG_RUNLOCK(sc, &tracker);
817 			break;
818 		}
819 
820 		lagg_port2req(lp, rp);
821 		LAGG_RUNLOCK(sc, &tracker);
822 		break;
823 
824 	case SIOCSIFCAP:
825 		if (lp->lp_ioctl == NULL) {
826 			error = EINVAL;
827 			break;
828 		}
829 		error = (*lp->lp_ioctl)(ifp, cmd, data);
830 		if (error)
831 			break;
832 
833 		/* Update lagg interface capabilities */
834 		LAGG_WLOCK(sc);
835 		lagg_capabilities(sc);
836 		LAGG_WUNLOCK(sc);
837 		break;
838 
839 	case SIOCSIFMTU:
840 		/* Do not allow the MTU to be changed once joined */
841 		error = EINVAL;
842 		break;
843 
844 	default:
845 		goto fallback;
846 	}
847 
848 	return (error);
849 
850 fallback:
851 	if (lp->lp_ioctl != NULL)
852 		return ((*lp->lp_ioctl)(ifp, cmd, data));
853 
854 	return (EINVAL);
855 }
856 
857 /*
858  * For direct output to child ports.
859  */
860 static int
861 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
862 	const struct sockaddr *dst, struct route *ro)
863 {
864 	struct lagg_port *lp = ifp->if_lagg;
865 
866 	switch (dst->sa_family) {
867 		case pseudo_AF_HDRCMPLT:
868 		case AF_UNSPEC:
869 			return ((*lp->lp_output)(ifp, m, dst, ro));
870 	}
871 
872 	/* drop any other frames */
873 	m_freem(m);
874 	return (ENETDOWN);
875 }
876 
877 static void
878 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
879 {
880 	struct lagg_port *lp;
881 	struct lagg_softc *sc;
882 
883 	if ((lp = ifp->if_lagg) == NULL)
884 		return;
885 	/* If the ifnet is just being renamed, don't do anything. */
886 	if (ifp->if_flags & IFF_RENAMING)
887 		return;
888 
889 	sc = lp->lp_softc;
890 
891 	LAGG_WLOCK(sc);
892 	lp->lp_detaching = 1;
893 	lagg_port_destroy(lp, 1);
894 	LAGG_WUNLOCK(sc);
895 }
896 
897 static void
898 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
899 {
900 	struct lagg_softc *sc = lp->lp_softc;
901 
902 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
903 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
904 	rp->rp_prio = lp->lp_prio;
905 	rp->rp_flags = lp->lp_flags;
906 	if (sc->sc_portreq != NULL)
907 		(*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
908 
909 	/* Add protocol specific flags */
910 	switch (sc->sc_proto) {
911 		case LAGG_PROTO_FAILOVER:
912 			if (lp == sc->sc_primary)
913 				rp->rp_flags |= LAGG_PORT_MASTER;
914 			if (lp == lagg_link_active(sc, sc->sc_primary))
915 				rp->rp_flags |= LAGG_PORT_ACTIVE;
916 			break;
917 
918 		case LAGG_PROTO_ROUNDROBIN:
919 		case LAGG_PROTO_LOADBALANCE:
920 		case LAGG_PROTO_ETHERCHANNEL:
921 			if (LAGG_PORTACTIVE(lp))
922 				rp->rp_flags |= LAGG_PORT_ACTIVE;
923 			break;
924 
925 		case LAGG_PROTO_LACP:
926 			/* LACP has a different definition of active */
927 			if (lacp_isactive(lp))
928 				rp->rp_flags |= LAGG_PORT_ACTIVE;
929 			if (lacp_iscollecting(lp))
930 				rp->rp_flags |= LAGG_PORT_COLLECTING;
931 			if (lacp_isdistributing(lp))
932 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
933 			break;
934 	}
935 
936 }
937 
938 static void
939 lagg_init(void *xsc)
940 {
941 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
942 	struct lagg_port *lp;
943 	struct ifnet *ifp = sc->sc_ifp;
944 
945 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
946 		return;
947 
948 	LAGG_WLOCK(sc);
949 
950 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
951 	/* Update the port lladdrs */
952 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
953 		lagg_port_lladdr(lp, IF_LLADDR(ifp));
954 
955 	if (sc->sc_init != NULL)
956 		(*sc->sc_init)(sc);
957 
958 	LAGG_WUNLOCK(sc);
959 }
960 
961 static void
962 lagg_stop(struct lagg_softc *sc)
963 {
964 	struct ifnet *ifp = sc->sc_ifp;
965 
966 	LAGG_WLOCK_ASSERT(sc);
967 
968 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
969 		return;
970 
971 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
972 
973 	if (sc->sc_stop != NULL)
974 		(*sc->sc_stop)(sc);
975 }
976 
977 static int
978 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
979 {
980 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
981 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
982 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
983 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
984 	struct ifreq *ifr = (struct ifreq *)data;
985 	struct lagg_port *lp;
986 	struct ifnet *tpif;
987 	struct thread *td = curthread;
988 	char *buf, *outbuf;
989 	int count, buflen, len, error = 0;
990 	struct rm_priotracker tracker;
991 
992 	bzero(&rpbuf, sizeof(rpbuf));
993 
994 	switch (cmd) {
995 	case SIOCGLAGG:
996 		LAGG_RLOCK(sc, &tracker);
997 		count = 0;
998 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
999 			count++;
1000 		buflen = count * sizeof(struct lagg_reqport);
1001 		LAGG_RUNLOCK(sc, &tracker);
1002 
1003 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1004 
1005 		LAGG_RLOCK(sc, &tracker);
1006 		ra->ra_proto = sc->sc_proto;
1007 		if (sc->sc_req != NULL)
1008 			(*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
1009 
1010 		count = 0;
1011 		buf = outbuf;
1012 		len = min(ra->ra_size, buflen);
1013 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1014 			if (len < sizeof(rpbuf))
1015 				break;
1016 
1017 			lagg_port2req(lp, &rpbuf);
1018 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1019 			count++;
1020 			buf += sizeof(rpbuf);
1021 			len -= sizeof(rpbuf);
1022 		}
1023 		LAGG_RUNLOCK(sc, &tracker);
1024 		ra->ra_ports = count;
1025 		ra->ra_size = count * sizeof(rpbuf);
1026 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1027 		free(outbuf, M_TEMP);
1028 		break;
1029 	case SIOCSLAGG:
1030 		error = priv_check(td, PRIV_NET_LAGG);
1031 		if (error)
1032 			break;
1033 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1034 			error = EPROTONOSUPPORT;
1035 			break;
1036 		}
1037 		LAGG_WLOCK(sc);
1038 		if (sc->sc_proto != LAGG_PROTO_NONE) {
1039 			/* Reset protocol first in case detach unlocks */
1040 			sc->sc_proto = LAGG_PROTO_NONE;
1041 			error = sc->sc_detach(sc);
1042 			sc->sc_detach = NULL;
1043 			sc->sc_start = NULL;
1044 			sc->sc_input = NULL;
1045 			sc->sc_port_create = NULL;
1046 			sc->sc_port_destroy = NULL;
1047 			sc->sc_linkstate = NULL;
1048 			sc->sc_init = NULL;
1049 			sc->sc_stop = NULL;
1050 			sc->sc_lladdr = NULL;
1051 			sc->sc_req = NULL;
1052 			sc->sc_portreq = NULL;
1053 		} else if (sc->sc_input != NULL) {
1054 			/* Still detaching */
1055 			error = EBUSY;
1056 		}
1057 		if (error != 0) {
1058 			LAGG_WUNLOCK(sc);
1059 			break;
1060 		}
1061 		for (int i = 0; i < (sizeof(lagg_protos) /
1062 		    sizeof(lagg_protos[0])); i++) {
1063 			if (lagg_protos[i].ti_proto == ra->ra_proto) {
1064 				if (sc->sc_ifflags & IFF_DEBUG)
1065 					printf("%s: using proto %u\n",
1066 					    sc->sc_ifname,
1067 					    lagg_protos[i].ti_proto);
1068 				sc->sc_proto = lagg_protos[i].ti_proto;
1069 				if (sc->sc_proto != LAGG_PROTO_NONE)
1070 					error = lagg_protos[i].ti_attach(sc);
1071 				LAGG_WUNLOCK(sc);
1072 				return (error);
1073 			}
1074 		}
1075 		LAGG_WUNLOCK(sc);
1076 		error = EPROTONOSUPPORT;
1077 		break;
1078 	case SIOCGLAGGFLAGS:
1079 		rf->rf_flags = sc->sc_flags;
1080 		break;
1081 	case SIOCSLAGGHASH:
1082 		error = priv_check(td, PRIV_NET_LAGG);
1083 		if (error)
1084 			break;
1085 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1086 			error = EINVAL;
1087 			break;
1088 		}
1089 		LAGG_WLOCK(sc);
1090 		sc->sc_flags &= ~LAGG_F_HASHMASK;
1091 		sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK;
1092 		LAGG_WUNLOCK(sc);
1093 		break;
1094 	case SIOCGLAGGPORT:
1095 		if (rp->rp_portname[0] == '\0' ||
1096 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1097 			error = EINVAL;
1098 			break;
1099 		}
1100 
1101 		LAGG_RLOCK(sc, &tracker);
1102 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1103 		    lp->lp_softc != sc) {
1104 			error = ENOENT;
1105 			LAGG_RUNLOCK(sc, &tracker);
1106 			break;
1107 		}
1108 
1109 		lagg_port2req(lp, rp);
1110 		LAGG_RUNLOCK(sc, &tracker);
1111 		break;
1112 	case SIOCSLAGGPORT:
1113 		error = priv_check(td, PRIV_NET_LAGG);
1114 		if (error)
1115 			break;
1116 		if (rp->rp_portname[0] == '\0' ||
1117 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1118 			error = EINVAL;
1119 			break;
1120 		}
1121 		LAGG_WLOCK(sc);
1122 		error = lagg_port_create(sc, tpif);
1123 		LAGG_WUNLOCK(sc);
1124 		break;
1125 	case SIOCSLAGGDELPORT:
1126 		error = priv_check(td, PRIV_NET_LAGG);
1127 		if (error)
1128 			break;
1129 		if (rp->rp_portname[0] == '\0' ||
1130 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1131 			error = EINVAL;
1132 			break;
1133 		}
1134 
1135 		LAGG_WLOCK(sc);
1136 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1137 		    lp->lp_softc != sc) {
1138 			error = ENOENT;
1139 			LAGG_WUNLOCK(sc);
1140 			break;
1141 		}
1142 
1143 		error = lagg_port_destroy(lp, 1);
1144 		LAGG_WUNLOCK(sc);
1145 		break;
1146 	case SIOCSIFFLAGS:
1147 		/* Set flags on ports too */
1148 		LAGG_WLOCK(sc);
1149 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1150 			lagg_setflags(lp, 1);
1151 		}
1152 		LAGG_WUNLOCK(sc);
1153 
1154 		if (!(ifp->if_flags & IFF_UP) &&
1155 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1156 			/*
1157 			 * If interface is marked down and it is running,
1158 			 * then stop and disable it.
1159 			 */
1160 			LAGG_WLOCK(sc);
1161 			lagg_stop(sc);
1162 			LAGG_WUNLOCK(sc);
1163 		} else if ((ifp->if_flags & IFF_UP) &&
1164 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1165 			/*
1166 			 * If interface is marked up and it is stopped, then
1167 			 * start it.
1168 			 */
1169 			(*ifp->if_init)(sc);
1170 		}
1171 		break;
1172 	case SIOCADDMULTI:
1173 	case SIOCDELMULTI:
1174 		LAGG_WLOCK(sc);
1175 		error = lagg_ether_setmulti(sc);
1176 		LAGG_WUNLOCK(sc);
1177 		break;
1178 	case SIOCSIFMEDIA:
1179 	case SIOCGIFMEDIA:
1180 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1181 		break;
1182 
1183 	case SIOCSIFCAP:
1184 	case SIOCSIFMTU:
1185 		/* Do not allow the MTU or caps to be directly changed */
1186 		error = EINVAL;
1187 		break;
1188 
1189 	default:
1190 		error = ether_ioctl(ifp, cmd, data);
1191 		break;
1192 	}
1193 	return (error);
1194 }
1195 
1196 static int
1197 lagg_ether_setmulti(struct lagg_softc *sc)
1198 {
1199 	struct lagg_port *lp;
1200 
1201 	LAGG_WLOCK_ASSERT(sc);
1202 
1203 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1204 		/* First, remove any existing filter entries. */
1205 		lagg_ether_cmdmulti(lp, 0);
1206 		/* copy all addresses from the lagg interface to the port */
1207 		lagg_ether_cmdmulti(lp, 1);
1208 	}
1209 	return (0);
1210 }
1211 
1212 static int
1213 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1214 {
1215 	struct lagg_softc *sc = lp->lp_softc;
1216 	struct ifnet *ifp = lp->lp_ifp;
1217 	struct ifnet *scifp = sc->sc_ifp;
1218 	struct lagg_mc *mc;
1219 	struct ifmultiaddr *ifma;
1220 	int error;
1221 
1222 	LAGG_WLOCK_ASSERT(sc);
1223 
1224 	if (set) {
1225 		IF_ADDR_WLOCK(scifp);
1226 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1227 			if (ifma->ifma_addr->sa_family != AF_LINK)
1228 				continue;
1229 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1230 			if (mc == NULL) {
1231 				IF_ADDR_WUNLOCK(scifp);
1232 				return (ENOMEM);
1233 			}
1234 			bcopy(ifma->ifma_addr, &mc->mc_addr,
1235 			    ifma->ifma_addr->sa_len);
1236 			mc->mc_addr.sdl_index = ifp->if_index;
1237 			mc->mc_ifma = NULL;
1238 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1239 		}
1240 		IF_ADDR_WUNLOCK(scifp);
1241 		SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1242 			error = if_addmulti(ifp,
1243 			    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1244 			if (error)
1245 				return (error);
1246 		}
1247 	} else {
1248 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1249 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1250 			if (mc->mc_ifma && !lp->lp_detaching)
1251 				if_delmulti_ifma(mc->mc_ifma);
1252 			free(mc, M_DEVBUF);
1253 		}
1254 	}
1255 	return (0);
1256 }
1257 
1258 /* Handle a ref counted flag that should be set on the lagg port as well */
1259 static int
1260 lagg_setflag(struct lagg_port *lp, int flag, int status,
1261 	     int (*func)(struct ifnet *, int))
1262 {
1263 	struct lagg_softc *sc = lp->lp_softc;
1264 	struct ifnet *scifp = sc->sc_ifp;
1265 	struct ifnet *ifp = lp->lp_ifp;
1266 	int error;
1267 
1268 	LAGG_WLOCK_ASSERT(sc);
1269 
1270 	status = status ? (scifp->if_flags & flag) : 0;
1271 	/* Now "status" contains the flag value or 0 */
1272 
1273 	/*
1274 	 * See if recorded ports status is different from what
1275 	 * we want it to be.  If it is, flip it.  We record ports
1276 	 * status in lp_ifflags so that we won't clear ports flag
1277 	 * we haven't set.  In fact, we don't clear or set ports
1278 	 * flags directly, but get or release references to them.
1279 	 * That's why we can be sure that recorded flags still are
1280 	 * in accord with actual ports flags.
1281 	 */
1282 	if (status != (lp->lp_ifflags & flag)) {
1283 		error = (*func)(ifp, status);
1284 		if (error)
1285 			return (error);
1286 		lp->lp_ifflags &= ~flag;
1287 		lp->lp_ifflags |= status;
1288 	}
1289 	return (0);
1290 }
1291 
1292 /*
1293  * Handle IFF_* flags that require certain changes on the lagg port
1294  * if "status" is true, update ports flags respective to the lagg
1295  * if "status" is false, forcedly clear the flags set on port.
1296  */
1297 static int
1298 lagg_setflags(struct lagg_port *lp, int status)
1299 {
1300 	int error, i;
1301 
1302 	for (i = 0; lagg_pflags[i].flag; i++) {
1303 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1304 		    status, lagg_pflags[i].func);
1305 		if (error)
1306 			return (error);
1307 	}
1308 	return (0);
1309 }
1310 
1311 static int
1312 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1313 {
1314 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1315 	int error, len, mcast;
1316 	struct rm_priotracker tracker;
1317 
1318 	len = m->m_pkthdr.len;
1319 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1320 
1321 	LAGG_RLOCK(sc, &tracker);
1322 	/* We need a Tx algorithm and at least one port */
1323 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1324 		LAGG_RUNLOCK(sc, &tracker);
1325 		m_freem(m);
1326 		ifp->if_oerrors++;
1327 		return (ENXIO);
1328 	}
1329 
1330 	ETHER_BPF_MTAP(ifp, m);
1331 
1332 	error = (*sc->sc_start)(sc, m);
1333 	LAGG_RUNLOCK(sc, &tracker);
1334 
1335 	if (error == 0) {
1336 		counter_u64_add(sc->sc_opackets, 1);
1337 		counter_u64_add(sc->sc_obytes, len);
1338 		ifp->if_omcasts += mcast;
1339 	} else
1340 		ifp->if_oerrors++;
1341 
1342 	return (error);
1343 }
1344 
1345 /*
1346  * The ifp->if_qflush entry point for lagg(4) is no-op.
1347  */
1348 static void
1349 lagg_qflush(struct ifnet *ifp __unused)
1350 {
1351 }
1352 
1353 static struct mbuf *
1354 lagg_input(struct ifnet *ifp, struct mbuf *m)
1355 {
1356 	struct lagg_port *lp = ifp->if_lagg;
1357 	struct lagg_softc *sc = lp->lp_softc;
1358 	struct ifnet *scifp = sc->sc_ifp;
1359 	struct rm_priotracker tracker;
1360 
1361 	LAGG_RLOCK(sc, &tracker);
1362 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1363 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1364 	    sc->sc_proto == LAGG_PROTO_NONE) {
1365 		LAGG_RUNLOCK(sc, &tracker);
1366 		m_freem(m);
1367 		return (NULL);
1368 	}
1369 
1370 	ETHER_BPF_MTAP(scifp, m);
1371 
1372 	m = (*sc->sc_input)(sc, lp, m);
1373 
1374 	if (m != NULL) {
1375 		counter_u64_add(sc->sc_ipackets, 1);
1376 		counter_u64_add(sc->sc_ibytes, m->m_pkthdr.len);
1377 
1378 		if (scifp->if_flags & IFF_MONITOR) {
1379 			m_freem(m);
1380 			m = NULL;
1381 		}
1382 	}
1383 
1384 	LAGG_RUNLOCK(sc, &tracker);
1385 	return (m);
1386 }
1387 
1388 static int
1389 lagg_media_change(struct ifnet *ifp)
1390 {
1391 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1392 
1393 	if (sc->sc_ifflags & IFF_DEBUG)
1394 		printf("%s\n", __func__);
1395 
1396 	/* Ignore */
1397 	return (0);
1398 }
1399 
1400 static void
1401 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1402 {
1403 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1404 	struct lagg_port *lp;
1405 	struct rm_priotracker tracker;
1406 
1407 	imr->ifm_status = IFM_AVALID;
1408 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1409 
1410 	LAGG_RLOCK(sc, &tracker);
1411 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1412 		if (LAGG_PORTACTIVE(lp))
1413 			imr->ifm_status |= IFM_ACTIVE;
1414 	}
1415 	LAGG_RUNLOCK(sc, &tracker);
1416 }
1417 
1418 static void
1419 lagg_linkstate(struct lagg_softc *sc)
1420 {
1421 	struct lagg_port *lp;
1422 	int new_link = LINK_STATE_DOWN;
1423 	uint64_t speed;
1424 
1425 	/* Our link is considered up if at least one of our ports is active */
1426 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1427 		if (lp->lp_link_state == LINK_STATE_UP) {
1428 			new_link = LINK_STATE_UP;
1429 			break;
1430 		}
1431 	}
1432 	if_link_state_change(sc->sc_ifp, new_link);
1433 
1434 	/* Update if_baudrate to reflect the max possible speed */
1435 	switch (sc->sc_proto) {
1436 		case LAGG_PROTO_FAILOVER:
1437 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1438 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1439 			break;
1440 		case LAGG_PROTO_ROUNDROBIN:
1441 		case LAGG_PROTO_LOADBALANCE:
1442 		case LAGG_PROTO_ETHERCHANNEL:
1443 			speed = 0;
1444 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1445 				speed += lp->lp_ifp->if_baudrate;
1446 			sc->sc_ifp->if_baudrate = speed;
1447 			break;
1448 		case LAGG_PROTO_LACP:
1449 			/* LACP updates if_baudrate itself */
1450 			break;
1451 	}
1452 }
1453 
1454 static void
1455 lagg_port_state(struct ifnet *ifp, int state)
1456 {
1457 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1458 	struct lagg_softc *sc = NULL;
1459 
1460 	if (lp != NULL)
1461 		sc = lp->lp_softc;
1462 	if (sc == NULL)
1463 		return;
1464 
1465 	LAGG_WLOCK(sc);
1466 	lagg_linkstate(sc);
1467 	if (sc->sc_linkstate != NULL)
1468 		(*sc->sc_linkstate)(lp);
1469 	LAGG_WUNLOCK(sc);
1470 }
1471 
1472 struct lagg_port *
1473 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1474 {
1475 	struct lagg_port *lp_next, *rval = NULL;
1476 	// int new_link = LINK_STATE_DOWN;
1477 
1478 	LAGG_RLOCK_ASSERT(sc);
1479 	/*
1480 	 * Search a port which reports an active link state.
1481 	 */
1482 
1483 	if (lp == NULL)
1484 		goto search;
1485 	if (LAGG_PORTACTIVE(lp)) {
1486 		rval = lp;
1487 		goto found;
1488 	}
1489 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1490 	    LAGG_PORTACTIVE(lp_next)) {
1491 		rval = lp_next;
1492 		goto found;
1493 	}
1494 
1495 search:
1496 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1497 		if (LAGG_PORTACTIVE(lp_next)) {
1498 			rval = lp_next;
1499 			goto found;
1500 		}
1501 	}
1502 
1503 found:
1504 	if (rval != NULL) {
1505 		/*
1506 		 * The IEEE 802.1D standard assumes that a lagg with
1507 		 * multiple ports is always full duplex. This is valid
1508 		 * for load sharing laggs and if at least two links
1509 		 * are active. Unfortunately, checking the latter would
1510 		 * be too expensive at this point.
1511 		 XXX
1512 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1513 		    (sc->sc_count > 1))
1514 			new_link = LINK_STATE_FULL_DUPLEX;
1515 		else
1516 			new_link = rval->lp_link_state;
1517 		 */
1518 	}
1519 
1520 	return (rval);
1521 }
1522 
1523 static const void *
1524 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1525 {
1526 	if (m->m_pkthdr.len < (off + len)) {
1527 		return (NULL);
1528 	} else if (m->m_len < (off + len)) {
1529 		m_copydata(m, off, len, buf);
1530 		return (buf);
1531 	}
1532 	return (mtod(m, char *) + off);
1533 }
1534 
1535 static int
1536 lagg_sysctl_active(SYSCTL_HANDLER_ARGS)
1537 {
1538 	struct lagg_softc *sc = (struct lagg_softc *)arg1;
1539 	struct lagg_port *lp;
1540 	int error;
1541 
1542 	/* LACP tracks active links automatically, the others do not */
1543 	if (sc->sc_proto != LAGG_PROTO_LACP) {
1544 		sc->sc_active = 0;
1545 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1546 			sc->sc_active += LAGG_PORTACTIVE(lp);
1547 	}
1548 
1549 	error = sysctl_handle_int(oidp, &sc->sc_active, 0, req);
1550 	if ((error) || (req->newptr == NULL))
1551 		return (error);
1552 
1553 	return (0);
1554 }
1555 
1556 uint32_t
1557 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
1558 {
1559 	uint16_t etype;
1560 	uint32_t p = key;
1561 	int off;
1562 	struct ether_header *eh;
1563 	const struct ether_vlan_header *vlan;
1564 #ifdef INET
1565 	const struct ip *ip;
1566 	const uint32_t *ports;
1567 	int iphlen;
1568 #endif
1569 #ifdef INET6
1570 	const struct ip6_hdr *ip6;
1571 	uint32_t flow;
1572 #endif
1573 	union {
1574 #ifdef INET
1575 		struct ip ip;
1576 #endif
1577 #ifdef INET6
1578 		struct ip6_hdr ip6;
1579 #endif
1580 		struct ether_vlan_header vlan;
1581 		uint32_t port;
1582 	} buf;
1583 
1584 
1585 	off = sizeof(*eh);
1586 	if (m->m_len < off)
1587 		goto out;
1588 	eh = mtod(m, struct ether_header *);
1589 	etype = ntohs(eh->ether_type);
1590 	if (sc->sc_flags & LAGG_F_HASHL2) {
1591 		p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p);
1592 		p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1593 	}
1594 
1595 	/* Special handling for encapsulating VLAN frames */
1596 	if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) {
1597 		p = hash32_buf(&m->m_pkthdr.ether_vtag,
1598 		    sizeof(m->m_pkthdr.ether_vtag), p);
1599 	} else if (etype == ETHERTYPE_VLAN) {
1600 		vlan = lagg_gethdr(m, off,  sizeof(*vlan), &buf);
1601 		if (vlan == NULL)
1602 			goto out;
1603 
1604 		if (sc->sc_flags & LAGG_F_HASHL2)
1605 			p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1606 		etype = ntohs(vlan->evl_proto);
1607 		off += sizeof(*vlan) - sizeof(*eh);
1608 	}
1609 
1610 	switch (etype) {
1611 #ifdef INET
1612 	case ETHERTYPE_IP:
1613 		ip = lagg_gethdr(m, off, sizeof(*ip), &buf);
1614 		if (ip == NULL)
1615 			goto out;
1616 
1617 		if (sc->sc_flags & LAGG_F_HASHL3) {
1618 			p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1619 			p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1620 		}
1621 		if (!(sc->sc_flags & LAGG_F_HASHL4))
1622 			break;
1623 		switch (ip->ip_p) {
1624 			case IPPROTO_TCP:
1625 			case IPPROTO_UDP:
1626 			case IPPROTO_SCTP:
1627 				iphlen = ip->ip_hl << 2;
1628 				if (iphlen < sizeof(*ip))
1629 					break;
1630 				off += iphlen;
1631 				ports = lagg_gethdr(m, off, sizeof(*ports), &buf);
1632 				if (ports == NULL)
1633 					break;
1634 				p = hash32_buf(ports, sizeof(*ports), p);
1635 				break;
1636 		}
1637 		break;
1638 #endif
1639 #ifdef INET6
1640 	case ETHERTYPE_IPV6:
1641 		if (!(sc->sc_flags & LAGG_F_HASHL3))
1642 			break;
1643 		ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf);
1644 		if (ip6 == NULL)
1645 			goto out;
1646 
1647 		p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1648 		p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1649 		flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1650 		p = hash32_buf(&flow, sizeof(flow), p);	/* IPv6 flow label */
1651 		break;
1652 #endif
1653 	}
1654 out:
1655 	return (p);
1656 }
1657 
1658 int
1659 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1660 {
1661 
1662 	return (ifp->if_transmit)(ifp, m);
1663 }
1664 
1665 /*
1666  * Simple round robin aggregation
1667  */
1668 
1669 static int
1670 lagg_rr_attach(struct lagg_softc *sc)
1671 {
1672 	sc->sc_detach = lagg_rr_detach;
1673 	sc->sc_start = lagg_rr_start;
1674 	sc->sc_input = lagg_rr_input;
1675 	sc->sc_port_create = NULL;
1676 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1677 	sc->sc_seq = 0;
1678 
1679 	return (0);
1680 }
1681 
1682 static int
1683 lagg_rr_detach(struct lagg_softc *sc)
1684 {
1685 	return (0);
1686 }
1687 
1688 static int
1689 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1690 {
1691 	struct lagg_port *lp;
1692 	uint32_t p;
1693 
1694 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
1695 	p %= sc->sc_count;
1696 	lp = SLIST_FIRST(&sc->sc_ports);
1697 	while (p--)
1698 		lp = SLIST_NEXT(lp, lp_entries);
1699 
1700 	/*
1701 	 * Check the port's link state. This will return the next active
1702 	 * port if the link is down or the port is NULL.
1703 	 */
1704 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1705 		m_freem(m);
1706 		return (ENETDOWN);
1707 	}
1708 
1709 	/* Send mbuf */
1710 	return (lagg_enqueue(lp->lp_ifp, m));
1711 }
1712 
1713 static struct mbuf *
1714 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1715 {
1716 	struct ifnet *ifp = sc->sc_ifp;
1717 
1718 	/* Just pass in the packet to our lagg device */
1719 	m->m_pkthdr.rcvif = ifp;
1720 
1721 	return (m);
1722 }
1723 
1724 /*
1725  * Active failover
1726  */
1727 
1728 static int
1729 lagg_fail_attach(struct lagg_softc *sc)
1730 {
1731 	sc->sc_detach = lagg_fail_detach;
1732 	sc->sc_start = lagg_fail_start;
1733 	sc->sc_input = lagg_fail_input;
1734 	sc->sc_port_create = NULL;
1735 	sc->sc_port_destroy = NULL;
1736 
1737 	return (0);
1738 }
1739 
1740 static int
1741 lagg_fail_detach(struct lagg_softc *sc)
1742 {
1743 	return (0);
1744 }
1745 
1746 static int
1747 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1748 {
1749 	struct lagg_port *lp;
1750 
1751 	/* Use the master port if active or the next available port */
1752 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1753 		m_freem(m);
1754 		return (ENETDOWN);
1755 	}
1756 
1757 	/* Send mbuf */
1758 	return (lagg_enqueue(lp->lp_ifp, m));
1759 }
1760 
1761 static struct mbuf *
1762 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1763 {
1764 	struct ifnet *ifp = sc->sc_ifp;
1765 	struct lagg_port *tmp_tp;
1766 
1767 	if (lp == sc->sc_primary || lagg_failover_rx_all) {
1768 		m->m_pkthdr.rcvif = ifp;
1769 		return (m);
1770 	}
1771 
1772 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1773 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
1774 		/*
1775 		 * If tmp_tp is null, we've recieved a packet when all
1776 		 * our links are down. Weird, but process it anyways.
1777 		 */
1778 		if ((tmp_tp == NULL || tmp_tp == lp)) {
1779 			m->m_pkthdr.rcvif = ifp;
1780 			return (m);
1781 		}
1782 	}
1783 
1784 	m_freem(m);
1785 	return (NULL);
1786 }
1787 
1788 /*
1789  * Loadbalancing
1790  */
1791 
1792 static int
1793 lagg_lb_attach(struct lagg_softc *sc)
1794 {
1795 	struct lagg_port *lp;
1796 	struct lagg_lb *lb;
1797 
1798 	if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1799 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1800 		return (ENOMEM);
1801 
1802 	sc->sc_detach = lagg_lb_detach;
1803 	sc->sc_start = lagg_lb_start;
1804 	sc->sc_input = lagg_lb_input;
1805 	sc->sc_port_create = lagg_lb_port_create;
1806 	sc->sc_port_destroy = lagg_lb_port_destroy;
1807 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1808 
1809 	lb->lb_key = arc4random();
1810 	sc->sc_psc = (caddr_t)lb;
1811 
1812 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1813 		lagg_lb_port_create(lp);
1814 
1815 	return (0);
1816 }
1817 
1818 static int
1819 lagg_lb_detach(struct lagg_softc *sc)
1820 {
1821 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1822 	if (lb != NULL)
1823 		free(lb, M_DEVBUF);
1824 	return (0);
1825 }
1826 
1827 static int
1828 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1829 {
1830 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1831 	struct lagg_port *lp_next;
1832 	int i = 0;
1833 
1834 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1835 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1836 		if (lp_next == lp)
1837 			continue;
1838 		if (i >= LAGG_MAX_PORTS)
1839 			return (EINVAL);
1840 		if (sc->sc_ifflags & IFF_DEBUG)
1841 			printf("%s: port %s at index %d\n",
1842 			    sc->sc_ifname, lp_next->lp_ifname, i);
1843 		lb->lb_ports[i++] = lp_next;
1844 	}
1845 
1846 	return (0);
1847 }
1848 
1849 static int
1850 lagg_lb_port_create(struct lagg_port *lp)
1851 {
1852 	struct lagg_softc *sc = lp->lp_softc;
1853 	return (lagg_lb_porttable(sc, NULL));
1854 }
1855 
1856 static void
1857 lagg_lb_port_destroy(struct lagg_port *lp)
1858 {
1859 	struct lagg_softc *sc = lp->lp_softc;
1860 	lagg_lb_porttable(sc, lp);
1861 }
1862 
1863 static int
1864 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1865 {
1866 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1867 	struct lagg_port *lp = NULL;
1868 	uint32_t p = 0;
1869 
1870 	if (sc->use_flowid && (m->m_flags & M_FLOWID))
1871 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
1872 	else
1873 		p = lagg_hashmbuf(sc, m, lb->lb_key);
1874 	p %= sc->sc_count;
1875 	lp = lb->lb_ports[p];
1876 
1877 	/*
1878 	 * Check the port's link state. This will return the next active
1879 	 * port if the link is down or the port is NULL.
1880 	 */
1881 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1882 		m_freem(m);
1883 		return (ENETDOWN);
1884 	}
1885 
1886 	/* Send mbuf */
1887 	return (lagg_enqueue(lp->lp_ifp, m));
1888 }
1889 
1890 static struct mbuf *
1891 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1892 {
1893 	struct ifnet *ifp = sc->sc_ifp;
1894 
1895 	/* Just pass in the packet to our lagg device */
1896 	m->m_pkthdr.rcvif = ifp;
1897 
1898 	return (m);
1899 }
1900 
1901 /*
1902  * 802.3ad LACP
1903  */
1904 
1905 static int
1906 lagg_lacp_attach(struct lagg_softc *sc)
1907 {
1908 	struct lagg_port *lp;
1909 	int error;
1910 
1911 	sc->sc_detach = lagg_lacp_detach;
1912 	sc->sc_port_create = lacp_port_create;
1913 	sc->sc_port_destroy = lacp_port_destroy;
1914 	sc->sc_linkstate = lacp_linkstate;
1915 	sc->sc_start = lagg_lacp_start;
1916 	sc->sc_input = lagg_lacp_input;
1917 	sc->sc_init = lacp_init;
1918 	sc->sc_stop = lacp_stop;
1919 	sc->sc_lladdr = lagg_lacp_lladdr;
1920 	sc->sc_req = lacp_req;
1921 	sc->sc_portreq = lacp_portreq;
1922 
1923 	error = lacp_attach(sc);
1924 	if (error)
1925 		return (error);
1926 
1927 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1928 		lacp_port_create(lp);
1929 
1930 	return (error);
1931 }
1932 
1933 static int
1934 lagg_lacp_detach(struct lagg_softc *sc)
1935 {
1936 	struct lagg_port *lp;
1937 	int error;
1938 
1939 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1940 		lacp_port_destroy(lp);
1941 
1942 	/* unlocking is safe here */
1943 	LAGG_WUNLOCK(sc);
1944 	error = lacp_detach(sc);
1945 	LAGG_WLOCK(sc);
1946 
1947 	return (error);
1948 }
1949 
1950 static void
1951 lagg_lacp_lladdr(struct lagg_softc *sc)
1952 {
1953 	struct lagg_port *lp;
1954 
1955 	/* purge all the lacp ports */
1956 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1957 		lacp_port_destroy(lp);
1958 
1959 	/* add them back in */
1960 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1961 		lacp_port_create(lp);
1962 }
1963 
1964 static int
1965 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1966 {
1967 	struct lagg_port *lp;
1968 
1969 	lp = lacp_select_tx_port(sc, m);
1970 	if (lp == NULL) {
1971 		m_freem(m);
1972 		return (ENETDOWN);
1973 	}
1974 
1975 	/* Send mbuf */
1976 	return (lagg_enqueue(lp->lp_ifp, m));
1977 }
1978 
1979 static struct mbuf *
1980 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1981 {
1982 	struct ifnet *ifp = sc->sc_ifp;
1983 	struct ether_header *eh;
1984 	u_short etype;
1985 
1986 	eh = mtod(m, struct ether_header *);
1987 	etype = ntohs(eh->ether_type);
1988 
1989 	/* Tap off LACP control messages */
1990 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
1991 		m = lacp_input(lp, m);
1992 		if (m == NULL)
1993 			return (NULL);
1994 	}
1995 
1996 	/*
1997 	 * If the port is not collecting or not in the active aggregator then
1998 	 * free and return.
1999 	 */
2000 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2001 		m_freem(m);
2002 		return (NULL);
2003 	}
2004 
2005 	m->m_pkthdr.rcvif = ifp;
2006 	return (m);
2007 }
2008 
2009 static void
2010 lagg_callout(void *arg)
2011 {
2012 	struct lagg_softc *sc = (struct lagg_softc *)arg;
2013 	struct ifnet *ifp = sc->sc_ifp;
2014 
2015 	ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets);
2016 	ifp->if_opackets = counter_u64_fetch(sc->sc_opackets);
2017 	ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes);
2018 	ifp->if_obytes = counter_u64_fetch(sc->sc_obytes);
2019 
2020 	callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
2021 }
2022