xref: /freebsd/sys/net/if_lagg.c (revision 43d1e6ee299ad4e143d90d3ad374d1c24bd3306f)
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  * Copyright (c) 2014, 2016 Marcelo Araujo <araujo@FreeBSD.org>
7  *
8  * Permission to use, copy, modify, and distribute this software for any
9  * purpose with or without fee is hereby granted, provided that the above
10  * copyright notice and this permission notice appear in all copies.
11  *
12  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19  */
20 
21 #include <sys/cdefs.h>
22 __FBSDID("$FreeBSD$");
23 
24 #include "opt_inet.h"
25 #include "opt_inet6.h"
26 #include "opt_ratelimit.h"
27 
28 #include <sys/param.h>
29 #include <sys/kernel.h>
30 #include <sys/malloc.h>
31 #include <sys/mbuf.h>
32 #include <sys/queue.h>
33 #include <sys/socket.h>
34 #include <sys/sockio.h>
35 #include <sys/sysctl.h>
36 #include <sys/module.h>
37 #include <sys/priv.h>
38 #include <sys/systm.h>
39 #include <sys/proc.h>
40 #include <sys/lock.h>
41 #include <sys/rmlock.h>
42 #include <sys/sx.h>
43 #include <sys/taskqueue.h>
44 #include <sys/eventhandler.h>
45 
46 #include <net/ethernet.h>
47 #include <net/if.h>
48 #include <net/if_clone.h>
49 #include <net/if_arp.h>
50 #include <net/if_dl.h>
51 #include <net/if_media.h>
52 #include <net/if_types.h>
53 #include <net/if_var.h>
54 #include <net/bpf.h>
55 #include <net/vnet.h>
56 
57 #if defined(INET) || defined(INET6)
58 #include <netinet/in.h>
59 #include <netinet/ip.h>
60 #endif
61 #ifdef INET
62 #include <netinet/in_systm.h>
63 #include <netinet/if_ether.h>
64 #endif
65 
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #include <netinet6/in6_var.h>
69 #include <netinet6/in6_ifattach.h>
70 #endif
71 
72 #include <net/if_vlan_var.h>
73 #include <net/if_lagg.h>
74 #include <net/ieee8023ad_lacp.h>
75 
76 #define	LAGG_RLOCK()	struct epoch_tracker lagg_et; epoch_enter_preempt(net_epoch_preempt, &lagg_et)
77 #define	LAGG_RUNLOCK()	epoch_exit_preempt(net_epoch_preempt, &lagg_et)
78 #define	LAGG_RLOCK_ASSERT()	MPASS(in_epoch(net_epoch_preempt))
79 #define	LAGG_UNLOCK_ASSERT()	MPASS(!in_epoch(net_epoch_preempt))
80 
81 #define	LAGG_SX_INIT(_sc)	sx_init(&(_sc)->sc_sx, "if_lagg sx")
82 #define	LAGG_SX_DESTROY(_sc)	sx_destroy(&(_sc)->sc_sx)
83 #define	LAGG_XLOCK(_sc)		sx_xlock(&(_sc)->sc_sx)
84 #define	LAGG_XUNLOCK(_sc)	sx_xunlock(&(_sc)->sc_sx)
85 #define	LAGG_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
86 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
87 
88 /* Special flags we should propagate to the lagg ports. */
89 static struct {
90 	int flag;
91 	int (*func)(struct ifnet *, int);
92 } lagg_pflags[] = {
93 	{IFF_PROMISC, ifpromisc},
94 	{IFF_ALLMULTI, if_allmulti},
95 	{0, NULL}
96 };
97 
98 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
99 #define	V_lagg_list	VNET(lagg_list)
100 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
101 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
102 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
103 					"if_lagg list", NULL, MTX_DEF)
104 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
105 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
106 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
107 eventhandler_tag	lagg_detach_cookie = NULL;
108 
109 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
110 static void	lagg_clone_destroy(struct ifnet *);
111 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
112 #define	V_lagg_cloner	VNET(lagg_cloner)
113 static const char laggname[] = "lagg";
114 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
115 
116 static void	lagg_capabilities(struct lagg_softc *);
117 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
118 static int	lagg_port_destroy(struct lagg_port *, int);
119 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
120 static void	lagg_linkstate(struct lagg_softc *);
121 static void	lagg_port_state(struct ifnet *, int);
122 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
123 static int	lagg_port_output(struct ifnet *, struct mbuf *,
124 		    const struct sockaddr *, struct route *);
125 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
126 #ifdef LAGG_PORT_STACKING
127 static int	lagg_port_checkstacking(struct lagg_softc *);
128 #endif
129 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
130 static void	lagg_init(void *);
131 static void	lagg_stop(struct lagg_softc *);
132 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
133 #ifdef RATELIMIT
134 static int	lagg_snd_tag_alloc(struct ifnet *,
135 		    union if_snd_tag_alloc_params *,
136 		    struct m_snd_tag **);
137 static void	lagg_snd_tag_free(struct m_snd_tag *);
138 #endif
139 static int	lagg_setmulti(struct lagg_port *);
140 static int	lagg_clrmulti(struct lagg_port *);
141 static	int	lagg_setcaps(struct lagg_port *, int cap);
142 static	int	lagg_setflag(struct lagg_port *, int, int,
143 		    int (*func)(struct ifnet *, int));
144 static	int	lagg_setflags(struct lagg_port *, int status);
145 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
146 static int	lagg_transmit(struct ifnet *, struct mbuf *);
147 static void	lagg_qflush(struct ifnet *);
148 static int	lagg_media_change(struct ifnet *);
149 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
150 static struct lagg_port *lagg_link_active(struct lagg_softc *,
151 	    struct lagg_port *);
152 
153 /* Simple round robin */
154 static void	lagg_rr_attach(struct lagg_softc *);
155 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
156 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
157 		    struct mbuf *);
158 
159 /* Active failover */
160 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
161 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
162 		    struct mbuf *);
163 
164 /* Loadbalancing */
165 static void	lagg_lb_attach(struct lagg_softc *);
166 static void	lagg_lb_detach(struct lagg_softc *);
167 static int	lagg_lb_port_create(struct lagg_port *);
168 static void	lagg_lb_port_destroy(struct lagg_port *);
169 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
170 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
171 		    struct mbuf *);
172 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
173 
174 /* Broadcast */
175 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
176 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
177 		    struct mbuf *);
178 
179 /* 802.3ad LACP */
180 static void	lagg_lacp_attach(struct lagg_softc *);
181 static void	lagg_lacp_detach(struct lagg_softc *);
182 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
183 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
184 		    struct mbuf *);
185 static void	lagg_lacp_lladdr(struct lagg_softc *);
186 
187 /* lagg protocol table */
188 static const struct lagg_proto {
189 	lagg_proto	pr_num;
190 	void		(*pr_attach)(struct lagg_softc *);
191 	void		(*pr_detach)(struct lagg_softc *);
192 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
193 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
194 			    struct mbuf *);
195 	int		(*pr_addport)(struct lagg_port *);
196 	void		(*pr_delport)(struct lagg_port *);
197 	void		(*pr_linkstate)(struct lagg_port *);
198 	void 		(*pr_init)(struct lagg_softc *);
199 	void 		(*pr_stop)(struct lagg_softc *);
200 	void 		(*pr_lladdr)(struct lagg_softc *);
201 	void		(*pr_request)(struct lagg_softc *, void *);
202 	void		(*pr_portreq)(struct lagg_port *, void *);
203 } lagg_protos[] = {
204     {
205 	.pr_num = LAGG_PROTO_NONE
206     },
207     {
208 	.pr_num = LAGG_PROTO_ROUNDROBIN,
209 	.pr_attach = lagg_rr_attach,
210 	.pr_start = lagg_rr_start,
211 	.pr_input = lagg_rr_input,
212     },
213     {
214 	.pr_num = LAGG_PROTO_FAILOVER,
215 	.pr_start = lagg_fail_start,
216 	.pr_input = lagg_fail_input,
217     },
218     {
219 	.pr_num = LAGG_PROTO_LOADBALANCE,
220 	.pr_attach = lagg_lb_attach,
221 	.pr_detach = lagg_lb_detach,
222 	.pr_start = lagg_lb_start,
223 	.pr_input = lagg_lb_input,
224 	.pr_addport = lagg_lb_port_create,
225 	.pr_delport = lagg_lb_port_destroy,
226     },
227     {
228 	.pr_num = LAGG_PROTO_LACP,
229 	.pr_attach = lagg_lacp_attach,
230 	.pr_detach = lagg_lacp_detach,
231 	.pr_start = lagg_lacp_start,
232 	.pr_input = lagg_lacp_input,
233 	.pr_addport = lacp_port_create,
234 	.pr_delport = lacp_port_destroy,
235 	.pr_linkstate = lacp_linkstate,
236 	.pr_init = lacp_init,
237 	.pr_stop = lacp_stop,
238 	.pr_lladdr = lagg_lacp_lladdr,
239 	.pr_request = lacp_req,
240 	.pr_portreq = lacp_portreq,
241     },
242     {
243 	.pr_num = LAGG_PROTO_BROADCAST,
244 	.pr_start = lagg_bcast_start,
245 	.pr_input = lagg_bcast_input,
246     },
247 };
248 
249 SYSCTL_DECL(_net_link);
250 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
251     "Link Aggregation");
252 
253 /* Allow input on any failover links */
254 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
255 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
256 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
257     &VNET_NAME(lagg_failover_rx_all), 0,
258     "Accept input from any interface in a failover lagg");
259 
260 /* Default value for using flowid */
261 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
262 #define	V_def_use_flowid	VNET(def_use_flowid)
263 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
264     &VNET_NAME(def_use_flowid), 0,
265     "Default setting for using flow id for load sharing");
266 
267 /* Default value for flowid shift */
268 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
269 #define	V_def_flowid_shift	VNET(def_flowid_shift)
270 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
271     &VNET_NAME(def_flowid_shift), 0,
272     "Default setting for flowid shift for load sharing");
273 
274 static void
275 vnet_lagg_init(const void *unused __unused)
276 {
277 
278 	LAGG_LIST_LOCK_INIT();
279 	SLIST_INIT(&V_lagg_list);
280 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
281 	    lagg_clone_destroy, 0);
282 }
283 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
284     vnet_lagg_init, NULL);
285 
286 static void
287 vnet_lagg_uninit(const void *unused __unused)
288 {
289 
290 	if_clone_detach(V_lagg_cloner);
291 	LAGG_LIST_LOCK_DESTROY();
292 }
293 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
294     vnet_lagg_uninit, NULL);
295 
296 static int
297 lagg_modevent(module_t mod, int type, void *data)
298 {
299 
300 	switch (type) {
301 	case MOD_LOAD:
302 		lagg_input_p = lagg_input;
303 		lagg_linkstate_p = lagg_port_state;
304 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
305 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
306 		    EVENTHANDLER_PRI_ANY);
307 		break;
308 	case MOD_UNLOAD:
309 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
310 		    lagg_detach_cookie);
311 		lagg_input_p = NULL;
312 		lagg_linkstate_p = NULL;
313 		break;
314 	default:
315 		return (EOPNOTSUPP);
316 	}
317 	return (0);
318 }
319 
320 static moduledata_t lagg_mod = {
321 	"if_lagg",
322 	lagg_modevent,
323 	0
324 };
325 
326 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
327 MODULE_VERSION(if_lagg, 1);
328 
329 static void
330 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
331 {
332 
333 	LAGG_XLOCK_ASSERT(sc);
334 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
335 	    __func__, sc));
336 
337 	if (sc->sc_ifflags & IFF_DEBUG)
338 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
339 
340 	if (lagg_protos[pr].pr_attach != NULL)
341 		lagg_protos[pr].pr_attach(sc);
342 	sc->sc_proto = pr;
343 }
344 
345 static void
346 lagg_proto_detach(struct lagg_softc *sc)
347 {
348 	lagg_proto pr;
349 
350 	LAGG_XLOCK_ASSERT(sc);
351 	pr = sc->sc_proto;
352 	sc->sc_proto = LAGG_PROTO_NONE;
353 
354 	if (lagg_protos[pr].pr_detach != NULL)
355 		lagg_protos[pr].pr_detach(sc);
356 }
357 
358 static int
359 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
360 {
361 
362 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
363 }
364 
365 static struct mbuf *
366 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
367 {
368 
369 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
370 }
371 
372 static int
373 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
374 {
375 
376 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
377 		return (0);
378 	else
379 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
380 }
381 
382 static void
383 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
384 {
385 
386 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
387 		lagg_protos[sc->sc_proto].pr_delport(lp);
388 }
389 
390 static void
391 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
392 {
393 
394 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
395 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
396 }
397 
398 static void
399 lagg_proto_init(struct lagg_softc *sc)
400 {
401 
402 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
403 		lagg_protos[sc->sc_proto].pr_init(sc);
404 }
405 
406 static void
407 lagg_proto_stop(struct lagg_softc *sc)
408 {
409 
410 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
411 		lagg_protos[sc->sc_proto].pr_stop(sc);
412 }
413 
414 static void
415 lagg_proto_lladdr(struct lagg_softc *sc)
416 {
417 
418 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
419 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
420 }
421 
422 static void
423 lagg_proto_request(struct lagg_softc *sc, void *v)
424 {
425 
426 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
427 		lagg_protos[sc->sc_proto].pr_request(sc, v);
428 }
429 
430 static void
431 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
432 {
433 
434 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
435 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
436 }
437 
438 /*
439  * This routine is run via an vlan
440  * config EVENT
441  */
442 static void
443 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
444 {
445 	struct lagg_softc *sc = ifp->if_softc;
446 	struct lagg_port *lp;
447 
448 	if (ifp->if_softc !=  arg)   /* Not our event */
449 		return;
450 
451 	LAGG_RLOCK();
452 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
453 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
454 	LAGG_RUNLOCK();
455 }
456 
457 /*
458  * This routine is run via an vlan
459  * unconfig EVENT
460  */
461 static void
462 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
463 {
464 	struct lagg_softc *sc = ifp->if_softc;
465 	struct lagg_port *lp;
466 
467 	if (ifp->if_softc !=  arg)   /* Not our event */
468 		return;
469 
470 	LAGG_RLOCK();
471 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
472 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
473 	LAGG_RUNLOCK();
474 }
475 
476 static int
477 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
478 {
479 	struct lagg_softc *sc;
480 	struct ifnet *ifp;
481 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
482 
483 	sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO);
484 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
485 	if (ifp == NULL) {
486 		free(sc, M_LAGG);
487 		return (ENOSPC);
488 	}
489 	LAGG_SX_INIT(sc);
490 
491 	LAGG_XLOCK(sc);
492 	if (V_def_use_flowid)
493 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
494 	sc->flowid_shift = V_def_flowid_shift;
495 
496 	/* Hash all layers by default */
497 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
498 
499 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
500 
501 	CK_SLIST_INIT(&sc->sc_ports);
502 
503 	/* Initialise pseudo media types */
504 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
505 	    lagg_media_status);
506 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
507 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
508 
509 	if_initname(ifp, laggname, unit);
510 	ifp->if_softc = sc;
511 	ifp->if_transmit = lagg_transmit;
512 	ifp->if_qflush = lagg_qflush;
513 	ifp->if_init = lagg_init;
514 	ifp->if_ioctl = lagg_ioctl;
515 	ifp->if_get_counter = lagg_get_counter;
516 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
517 #ifdef RATELIMIT
518 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
519 	ifp->if_snd_tag_free = lagg_snd_tag_free;
520 #endif
521 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
522 
523 	/*
524 	 * Attach as an ordinary ethernet device, children will be attached
525 	 * as special device IFT_IEEE8023ADLAG.
526 	 */
527 	ether_ifattach(ifp, eaddr);
528 
529 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
530 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
531 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
532 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
533 
534 	/* Insert into the global list of laggs */
535 	LAGG_LIST_LOCK();
536 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
537 	LAGG_LIST_UNLOCK();
538 	LAGG_XUNLOCK(sc);
539 
540 	return (0);
541 }
542 
543 static void
544 lagg_clone_destroy(struct ifnet *ifp)
545 {
546 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
547 	struct lagg_port *lp;
548 
549 	LAGG_XLOCK(sc);
550 	sc->sc_destroying = 1;
551 	lagg_stop(sc);
552 	ifp->if_flags &= ~IFF_UP;
553 
554 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
555 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
556 
557 	/* Shutdown and remove lagg ports */
558 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
559 		lagg_port_destroy(lp, 1);
560 
561 	/* Unhook the aggregation protocol */
562 	lagg_proto_detach(sc);
563 	LAGG_XUNLOCK(sc);
564 
565 	ifmedia_removeall(&sc->sc_media);
566 	ether_ifdetach(ifp);
567 	if_free(ifp);
568 
569 	LAGG_LIST_LOCK();
570 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
571 	LAGG_LIST_UNLOCK();
572 
573 	LAGG_SX_DESTROY(sc);
574 	free(sc, M_LAGG);
575 }
576 
577 static void
578 lagg_capabilities(struct lagg_softc *sc)
579 {
580 	struct lagg_port *lp;
581 	int cap, ena, pena;
582 	uint64_t hwa;
583 	struct ifnet_hw_tsomax hw_tsomax;
584 
585 	LAGG_XLOCK_ASSERT(sc);
586 
587 	/* Get common enabled capabilities for the lagg ports */
588 	ena = ~0;
589 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
590 		ena &= lp->lp_ifp->if_capenable;
591 	ena = (ena == ~0 ? 0 : ena);
592 
593 	/*
594 	 * Apply common enabled capabilities back to the lagg ports.
595 	 * May require several iterations if they are dependent.
596 	 */
597 	do {
598 		pena = ena;
599 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
600 			lagg_setcaps(lp, ena);
601 			ena &= lp->lp_ifp->if_capenable;
602 		}
603 	} while (pena != ena);
604 
605 	/* Get other capabilities from the lagg ports */
606 	cap = ~0;
607 	hwa = ~(uint64_t)0;
608 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
609 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
610 		cap &= lp->lp_ifp->if_capabilities;
611 		hwa &= lp->lp_ifp->if_hwassist;
612 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
613 	}
614 	cap = (cap == ~0 ? 0 : cap);
615 	hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
616 
617 	if (sc->sc_ifp->if_capabilities != cap ||
618 	    sc->sc_ifp->if_capenable != ena ||
619 	    sc->sc_ifp->if_hwassist != hwa ||
620 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
621 		sc->sc_ifp->if_capabilities = cap;
622 		sc->sc_ifp->if_capenable = ena;
623 		sc->sc_ifp->if_hwassist = hwa;
624 		getmicrotime(&sc->sc_ifp->if_lastchange);
625 
626 		if (sc->sc_ifflags & IFF_DEBUG)
627 			if_printf(sc->sc_ifp,
628 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
629 	}
630 }
631 
632 static int
633 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
634 {
635 	struct lagg_softc *sc_ptr;
636 	struct lagg_port *lp, *tlp;
637 	struct ifreq ifr;
638 	int error, i, oldmtu;
639 	uint64_t *pval;
640 
641 	LAGG_XLOCK_ASSERT(sc);
642 
643 	if (sc->sc_ifp == ifp) {
644 		if_printf(sc->sc_ifp,
645 		    "cannot add a lagg to itself as a port\n");
646 		return (EINVAL);
647 	}
648 
649 	/* Limit the maximal number of lagg ports */
650 	if (sc->sc_count >= LAGG_MAX_PORTS)
651 		return (ENOSPC);
652 
653 	/* Check if port has already been associated to a lagg */
654 	if (ifp->if_lagg != NULL) {
655 		/* Port is already in the current lagg? */
656 		lp = (struct lagg_port *)ifp->if_lagg;
657 		if (lp->lp_softc == sc)
658 			return (EEXIST);
659 		return (EBUSY);
660 	}
661 
662 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
663 	if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
664 		return (EPROTONOSUPPORT);
665 
666 	/* Allow the first Ethernet member to define the MTU */
667 	oldmtu = -1;
668 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
669 		sc->sc_ifp->if_mtu = ifp->if_mtu;
670 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
671 		if (ifp->if_ioctl == NULL) {
672 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
673 			    ifp->if_xname);
674 			return (EINVAL);
675 		}
676 		oldmtu = ifp->if_mtu;
677 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
678 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
679 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
680 		if (error != 0) {
681 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
682 			    ifp->if_xname);
683 			return (error);
684 		}
685 		ifr.ifr_mtu = oldmtu;
686 	}
687 
688 	lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO);
689 	lp->lp_softc = sc;
690 
691 	/* Check if port is a stacked lagg */
692 	LAGG_LIST_LOCK();
693 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
694 		if (ifp == sc_ptr->sc_ifp) {
695 			LAGG_LIST_UNLOCK();
696 			free(lp, M_LAGG);
697 			if (oldmtu != -1)
698 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
699 				    (caddr_t)&ifr);
700 			return (EINVAL);
701 			/* XXX disable stacking for the moment, its untested */
702 #ifdef LAGG_PORT_STACKING
703 			lp->lp_flags |= LAGG_PORT_STACK;
704 			if (lagg_port_checkstacking(sc_ptr) >=
705 			    LAGG_MAX_STACKING) {
706 				LAGG_LIST_UNLOCK();
707 				free(lp, M_LAGG);
708 				if (oldmtu != -1)
709 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
710 					    (caddr_t)&ifr);
711 				return (E2BIG);
712 			}
713 #endif
714 		}
715 	}
716 	LAGG_LIST_UNLOCK();
717 
718 	if_ref(ifp);
719 	lp->lp_ifp = ifp;
720 
721 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
722 	lp->lp_ifcapenable = ifp->if_capenable;
723 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
724 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
725 		lagg_proto_lladdr(sc);
726 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
727 	} else {
728 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
729 	}
730 	lagg_setflags(lp, 1);
731 
732 	if (CK_SLIST_EMPTY(&sc->sc_ports))
733 		sc->sc_primary = lp;
734 
735 	/* Change the interface type */
736 	lp->lp_iftype = ifp->if_type;
737 	ifp->if_type = IFT_IEEE8023ADLAG;
738 	ifp->if_lagg = lp;
739 	lp->lp_ioctl = ifp->if_ioctl;
740 	ifp->if_ioctl = lagg_port_ioctl;
741 	lp->lp_output = ifp->if_output;
742 	ifp->if_output = lagg_port_output;
743 
744 	/* Read port counters */
745 	pval = lp->port_counters.val;
746 	for (i = 0; i < IFCOUNTERS; i++, pval++)
747 		*pval = ifp->if_get_counter(ifp, i);
748 
749 	/*
750 	 * Insert into the list of ports.
751 	 * Keep ports sorted by if_index. It is handy, when configuration
752 	 * is predictable and `ifconfig laggN create ...` command
753 	 * will lead to the same result each time.
754 	 */
755 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
756 		if (tlp->lp_ifp->if_index < ifp->if_index && (
757 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
758 		    ((struct  lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
759 		    ifp->if_index))
760 			break;
761 	}
762 	if (tlp != NULL)
763 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
764 	else
765 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
766 	sc->sc_count++;
767 
768 	lagg_setmulti(lp);
769 
770 
771 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
772 		/* Remove the port, without calling pr_delport. */
773 		lagg_port_destroy(lp, 0);
774 		if (oldmtu != -1)
775 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
776 		return (error);
777 	}
778 
779 	/* Update lagg capabilities */
780 	lagg_capabilities(sc);
781 	lagg_linkstate(sc);
782 
783 	return (0);
784 }
785 
786 #ifdef LAGG_PORT_STACKING
787 static int
788 lagg_port_checkstacking(struct lagg_softc *sc)
789 {
790 	struct lagg_softc *sc_ptr;
791 	struct lagg_port *lp;
792 	int m = 0;
793 
794 	LAGG_SXLOCK_ASSERT(sc);
795 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
796 		if (lp->lp_flags & LAGG_PORT_STACK) {
797 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
798 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
799 		}
800 	}
801 
802 	return (m + 1);
803 }
804 #endif
805 
806 static void
807 lagg_port_destroy_cb(epoch_context_t ec)
808 {
809 	struct lagg_port *lp;
810 	struct ifnet *ifp;
811 
812 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
813 	ifp = lp->lp_ifp;
814 
815 	if_rele(ifp);
816 	free(lp, M_LAGG);
817 }
818 
819 static int
820 lagg_port_destroy(struct lagg_port *lp, int rundelport)
821 {
822 	struct lagg_softc *sc = lp->lp_softc;
823 	struct lagg_port *lp_ptr, *lp0;
824 	struct ifnet *ifp = lp->lp_ifp;
825 	uint64_t *pval, vdiff;
826 	int i;
827 
828 	LAGG_XLOCK_ASSERT(sc);
829 
830 	if (rundelport)
831 		lagg_proto_delport(sc, lp);
832 
833 	if (lp->lp_detaching == 0)
834 		lagg_clrmulti(lp);
835 
836 	/* Restore interface */
837 	ifp->if_type = lp->lp_iftype;
838 	ifp->if_ioctl = lp->lp_ioctl;
839 	ifp->if_output = lp->lp_output;
840 	ifp->if_lagg = NULL;
841 
842 	/* Update detached port counters */
843 	pval = lp->port_counters.val;
844 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
845 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
846 		sc->detached_counters.val[i] += vdiff;
847 	}
848 
849 	/* Finally, remove the port from the lagg */
850 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
851 	sc->sc_count--;
852 
853 	/* Update the primary interface */
854 	if (lp == sc->sc_primary) {
855 		uint8_t lladdr[ETHER_ADDR_LEN];
856 
857 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
858 			bzero(&lladdr, ETHER_ADDR_LEN);
859 		else
860 			bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN);
861 		sc->sc_primary = lp0;
862 		if (sc->sc_destroying == 0) {
863 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
864 			lagg_proto_lladdr(sc);
865 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
866 		}
867 
868 		/*
869 		 * Update lladdr for each port (new primary needs update
870 		 * as well, to switch from old lladdr to its 'real' one)
871 		 */
872 		CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
873 			if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN);
874 	}
875 
876 	if (lp->lp_ifflags)
877 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
878 
879 	if (lp->lp_detaching == 0) {
880 		lagg_setflags(lp, 0);
881 		lagg_setcaps(lp, lp->lp_ifcapenable);
882 		if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN);
883 	}
884 
885 	/*
886 	 * free port and release it's ifnet reference after a grace period has
887 	 * elapsed.
888 	 */
889 	epoch_call(net_epoch_preempt, &lp->lp_epoch_ctx, lagg_port_destroy_cb);
890 	/* Update lagg capabilities */
891 	lagg_capabilities(sc);
892 	lagg_linkstate(sc);
893 
894 	return (0);
895 }
896 
897 static int
898 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
899 {
900 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
901 	struct lagg_softc *sc;
902 	struct lagg_port *lp = NULL;
903 	int error = 0;
904 
905 	/* Should be checked by the caller */
906 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
907 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
908 		goto fallback;
909 
910 	switch (cmd) {
911 	case SIOCGLAGGPORT:
912 		if (rp->rp_portname[0] == '\0' ||
913 		    ifunit(rp->rp_portname) != ifp) {
914 			error = EINVAL;
915 			break;
916 		}
917 
918 		LAGG_RLOCK();
919 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
920 			error = ENOENT;
921 			LAGG_RUNLOCK();
922 			break;
923 		}
924 
925 		lagg_port2req(lp, rp);
926 		LAGG_RUNLOCK();
927 		break;
928 
929 	case SIOCSIFCAP:
930 		if (lp->lp_ioctl == NULL) {
931 			error = EINVAL;
932 			break;
933 		}
934 		error = (*lp->lp_ioctl)(ifp, cmd, data);
935 		if (error)
936 			break;
937 
938 		/* Update lagg interface capabilities */
939 		LAGG_XLOCK(sc);
940 		lagg_capabilities(sc);
941 		LAGG_XUNLOCK(sc);
942 		VLAN_CAPABILITIES(sc->sc_ifp);
943 		break;
944 
945 	case SIOCSIFMTU:
946 		/* Do not allow the MTU to be changed once joined */
947 		error = EINVAL;
948 		break;
949 
950 	default:
951 		goto fallback;
952 	}
953 
954 	return (error);
955 
956 fallback:
957 	if (lp != NULL && lp->lp_ioctl != NULL)
958 		return ((*lp->lp_ioctl)(ifp, cmd, data));
959 
960 	return (EINVAL);
961 }
962 
963 /*
964  * Requests counter @cnt data.
965  *
966  * Counter value is calculated the following way:
967  * 1) for each port, sum  difference between current and "initial" measurements.
968  * 2) add lagg logical interface counters.
969  * 3) add data from detached_counters array.
970  *
971  * We also do the following things on ports attach/detach:
972  * 1) On port attach we store all counters it has into port_counter array.
973  * 2) On port detach we add the different between "initial" and
974  *   current counters data to detached_counters array.
975  */
976 static uint64_t
977 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
978 {
979 	struct lagg_softc *sc;
980 	struct lagg_port *lp;
981 	struct ifnet *lpifp;
982 	uint64_t newval, oldval, vsum;
983 
984 	/* Revise this when we've got non-generic counters. */
985 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
986 
987 	sc = (struct lagg_softc *)ifp->if_softc;
988 
989 	vsum = 0;
990 	LAGG_RLOCK();
991 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
992 		/* Saved attached value */
993 		oldval = lp->port_counters.val[cnt];
994 		/* current value */
995 		lpifp = lp->lp_ifp;
996 		newval = lpifp->if_get_counter(lpifp, cnt);
997 		/* Calculate diff and save new */
998 		vsum += newval - oldval;
999 	}
1000 	LAGG_RUNLOCK();
1001 
1002 	/*
1003 	 * Add counter data which might be added by upper
1004 	 * layer protocols operating on logical interface.
1005 	 */
1006 	vsum += if_get_counter_default(ifp, cnt);
1007 
1008 	/*
1009 	 * Add counter data from detached ports counters
1010 	 */
1011 	vsum += sc->detached_counters.val[cnt];
1012 
1013 
1014 	return (vsum);
1015 }
1016 
1017 /*
1018  * For direct output to child ports.
1019  */
1020 static int
1021 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1022 	const struct sockaddr *dst, struct route *ro)
1023 {
1024 	struct lagg_port *lp = ifp->if_lagg;
1025 
1026 	switch (dst->sa_family) {
1027 		case pseudo_AF_HDRCMPLT:
1028 		case AF_UNSPEC:
1029 			return ((*lp->lp_output)(ifp, m, dst, ro));
1030 	}
1031 
1032 	/* drop any other frames */
1033 	m_freem(m);
1034 	return (ENETDOWN);
1035 }
1036 
1037 static void
1038 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1039 {
1040 	struct lagg_port *lp;
1041 	struct lagg_softc *sc;
1042 
1043 	if ((lp = ifp->if_lagg) == NULL)
1044 		return;
1045 	/* If the ifnet is just being renamed, don't do anything. */
1046 	if (ifp->if_flags & IFF_RENAMING)
1047 		return;
1048 
1049 	sc = lp->lp_softc;
1050 
1051 	LAGG_XLOCK(sc);
1052 	lp->lp_detaching = 1;
1053 	lagg_port_destroy(lp, 1);
1054 	LAGG_XUNLOCK(sc);
1055 	VLAN_CAPABILITIES(sc->sc_ifp);
1056 }
1057 
1058 static void
1059 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1060 {
1061 	struct lagg_softc *sc = lp->lp_softc;
1062 
1063 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1064 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1065 	rp->rp_prio = lp->lp_prio;
1066 	rp->rp_flags = lp->lp_flags;
1067 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1068 
1069 	/* Add protocol specific flags */
1070 	switch (sc->sc_proto) {
1071 		case LAGG_PROTO_FAILOVER:
1072 			if (lp == sc->sc_primary)
1073 				rp->rp_flags |= LAGG_PORT_MASTER;
1074 			if (lp == lagg_link_active(sc, sc->sc_primary))
1075 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1076 			break;
1077 
1078 		case LAGG_PROTO_ROUNDROBIN:
1079 		case LAGG_PROTO_LOADBALANCE:
1080 		case LAGG_PROTO_BROADCAST:
1081 			if (LAGG_PORTACTIVE(lp))
1082 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1083 			break;
1084 
1085 		case LAGG_PROTO_LACP:
1086 			/* LACP has a different definition of active */
1087 			if (lacp_isactive(lp))
1088 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1089 			if (lacp_iscollecting(lp))
1090 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1091 			if (lacp_isdistributing(lp))
1092 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1093 			break;
1094 	}
1095 
1096 }
1097 
1098 static void
1099 lagg_init(void *xsc)
1100 {
1101 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1102 	struct ifnet *ifp = sc->sc_ifp;
1103 	struct lagg_port *lp;
1104 
1105 	LAGG_XLOCK(sc);
1106 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1107 		LAGG_XUNLOCK(sc);
1108 		return;
1109 	}
1110 
1111 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1112 
1113 	/*
1114 	 * Update the port lladdrs if needed.
1115 	 * This might be if_setlladdr() notification
1116 	 * that lladdr has been changed.
1117 	 */
1118 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1119 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1120 		    ETHER_ADDR_LEN) != 0)
1121 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN);
1122 	}
1123 
1124 	lagg_proto_init(sc);
1125 
1126 	LAGG_XUNLOCK(sc);
1127 }
1128 
1129 static void
1130 lagg_stop(struct lagg_softc *sc)
1131 {
1132 	struct ifnet *ifp = sc->sc_ifp;
1133 
1134 	LAGG_XLOCK_ASSERT(sc);
1135 
1136 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1137 		return;
1138 
1139 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1140 
1141 	lagg_proto_stop(sc);
1142 }
1143 
1144 static int
1145 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1146 {
1147 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1148 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1149 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1150 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1151 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1152 	struct ifreq *ifr = (struct ifreq *)data;
1153 	struct lagg_port *lp;
1154 	struct ifnet *tpif;
1155 	struct thread *td = curthread;
1156 	char *buf, *outbuf;
1157 	int count, buflen, len, error = 0;
1158 
1159 	bzero(&rpbuf, sizeof(rpbuf));
1160 
1161 	switch (cmd) {
1162 	case SIOCGLAGG:
1163 		LAGG_XLOCK(sc);
1164 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1165 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1166 		ra->ra_proto = sc->sc_proto;
1167 		lagg_proto_request(sc, &ra->ra_psc);
1168 		count = 0;
1169 		buf = outbuf;
1170 		len = min(ra->ra_size, buflen);
1171 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1172 			if (len < sizeof(rpbuf))
1173 				break;
1174 
1175 			lagg_port2req(lp, &rpbuf);
1176 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1177 			count++;
1178 			buf += sizeof(rpbuf);
1179 			len -= sizeof(rpbuf);
1180 		}
1181 		LAGG_XUNLOCK(sc);
1182 		ra->ra_ports = count;
1183 		ra->ra_size = count * sizeof(rpbuf);
1184 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1185 		free(outbuf, M_TEMP);
1186 		break;
1187 	case SIOCSLAGG:
1188 		error = priv_check(td, PRIV_NET_LAGG);
1189 		if (error)
1190 			break;
1191 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1192 			error = EPROTONOSUPPORT;
1193 			break;
1194 		}
1195 
1196 		LAGG_XLOCK(sc);
1197 		lagg_proto_detach(sc);
1198 		LAGG_UNLOCK_ASSERT();
1199 		lagg_proto_attach(sc, ra->ra_proto);
1200 		LAGG_XUNLOCK(sc);
1201 		break;
1202 	case SIOCGLAGGOPTS:
1203 		LAGG_XLOCK(sc);
1204 		ro->ro_opts = sc->sc_opts;
1205 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1206 			struct lacp_softc *lsc;
1207 
1208 			lsc = (struct lacp_softc *)sc->sc_psc;
1209 			if (lsc->lsc_debug.lsc_tx_test != 0)
1210 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1211 			if (lsc->lsc_debug.lsc_rx_test != 0)
1212 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1213 			if (lsc->lsc_strict_mode != 0)
1214 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1215 			if (lsc->lsc_fast_timeout != 0)
1216 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1217 
1218 			ro->ro_active = sc->sc_active;
1219 		} else {
1220 			ro->ro_active = 0;
1221 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1222 				ro->ro_active += LAGG_PORTACTIVE(lp);
1223 		}
1224 		ro->ro_bkt = sc->sc_bkt;
1225 		ro->ro_flapping = sc->sc_flapping;
1226 		ro->ro_flowid_shift = sc->flowid_shift;
1227 		LAGG_XUNLOCK(sc);
1228 		break;
1229 	case SIOCSLAGGOPTS:
1230 		if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) {
1231 			if (ro->ro_bkt == 0)
1232 				sc->sc_bkt = 1; // Minimum 1 packet per iface.
1233 			else
1234 				sc->sc_bkt = ro->ro_bkt;
1235 		}
1236 		error = priv_check(td, PRIV_NET_LAGG);
1237 		if (error)
1238 			break;
1239 		if (ro->ro_opts == 0)
1240 			break;
1241 		/*
1242 		 * Set options.  LACP options are stored in sc->sc_psc,
1243 		 * not in sc_opts.
1244 		 */
1245 		int valid, lacp;
1246 
1247 		switch (ro->ro_opts) {
1248 		case LAGG_OPT_USE_FLOWID:
1249 		case -LAGG_OPT_USE_FLOWID:
1250 		case LAGG_OPT_FLOWIDSHIFT:
1251 			valid = 1;
1252 			lacp = 0;
1253 			break;
1254 		case LAGG_OPT_LACP_TXTEST:
1255 		case -LAGG_OPT_LACP_TXTEST:
1256 		case LAGG_OPT_LACP_RXTEST:
1257 		case -LAGG_OPT_LACP_RXTEST:
1258 		case LAGG_OPT_LACP_STRICT:
1259 		case -LAGG_OPT_LACP_STRICT:
1260 		case LAGG_OPT_LACP_TIMEOUT:
1261 		case -LAGG_OPT_LACP_TIMEOUT:
1262 			valid = lacp = 1;
1263 			break;
1264 		default:
1265 			valid = lacp = 0;
1266 			break;
1267 		}
1268 
1269 		LAGG_XLOCK(sc);
1270 
1271 		if (valid == 0 ||
1272 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1273 			/* Invalid combination of options specified. */
1274 			error = EINVAL;
1275 			LAGG_XUNLOCK(sc);
1276 			break;	/* Return from SIOCSLAGGOPTS. */
1277 		}
1278 		/*
1279 		 * Store new options into sc->sc_opts except for
1280 		 * FLOWIDSHIFT and LACP options.
1281 		 */
1282 		if (lacp == 0) {
1283 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1284 				sc->flowid_shift = ro->ro_flowid_shift;
1285 			else if (ro->ro_opts > 0)
1286 				sc->sc_opts |= ro->ro_opts;
1287 			else
1288 				sc->sc_opts &= ~ro->ro_opts;
1289 		} else {
1290 			struct lacp_softc *lsc;
1291 			struct lacp_port *lp;
1292 
1293 			lsc = (struct lacp_softc *)sc->sc_psc;
1294 
1295 			switch (ro->ro_opts) {
1296 			case LAGG_OPT_LACP_TXTEST:
1297 				lsc->lsc_debug.lsc_tx_test = 1;
1298 				break;
1299 			case -LAGG_OPT_LACP_TXTEST:
1300 				lsc->lsc_debug.lsc_tx_test = 0;
1301 				break;
1302 			case LAGG_OPT_LACP_RXTEST:
1303 				lsc->lsc_debug.lsc_rx_test = 1;
1304 				break;
1305 			case -LAGG_OPT_LACP_RXTEST:
1306 				lsc->lsc_debug.lsc_rx_test = 0;
1307 				break;
1308 			case LAGG_OPT_LACP_STRICT:
1309 				lsc->lsc_strict_mode = 1;
1310 				break;
1311 			case -LAGG_OPT_LACP_STRICT:
1312 				lsc->lsc_strict_mode = 0;
1313 				break;
1314 			case LAGG_OPT_LACP_TIMEOUT:
1315 				LACP_LOCK(lsc);
1316         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1317                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1318 				LACP_UNLOCK(lsc);
1319 				lsc->lsc_fast_timeout = 1;
1320 				break;
1321 			case -LAGG_OPT_LACP_TIMEOUT:
1322 				LACP_LOCK(lsc);
1323         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1324                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1325 				LACP_UNLOCK(lsc);
1326 				lsc->lsc_fast_timeout = 0;
1327 				break;
1328 			}
1329 		}
1330 		LAGG_XUNLOCK(sc);
1331 		break;
1332 	case SIOCGLAGGFLAGS:
1333 		rf->rf_flags = 0;
1334 		LAGG_XLOCK(sc);
1335 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1336 			rf->rf_flags |= LAGG_F_HASHL2;
1337 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1338 			rf->rf_flags |= LAGG_F_HASHL3;
1339 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1340 			rf->rf_flags |= LAGG_F_HASHL4;
1341 		LAGG_XUNLOCK(sc);
1342 		break;
1343 	case SIOCSLAGGHASH:
1344 		error = priv_check(td, PRIV_NET_LAGG);
1345 		if (error)
1346 			break;
1347 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1348 			error = EINVAL;
1349 			break;
1350 		}
1351 		LAGG_XLOCK(sc);
1352 		sc->sc_flags = 0;
1353 		if (rf->rf_flags & LAGG_F_HASHL2)
1354 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1355 		if (rf->rf_flags & LAGG_F_HASHL3)
1356 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1357 		if (rf->rf_flags & LAGG_F_HASHL4)
1358 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1359 		LAGG_XUNLOCK(sc);
1360 		break;
1361 	case SIOCGLAGGPORT:
1362 		if (rp->rp_portname[0] == '\0' ||
1363 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1364 			error = EINVAL;
1365 			break;
1366 		}
1367 
1368 		LAGG_RLOCK();
1369 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1370 		    lp->lp_softc != sc) {
1371 			error = ENOENT;
1372 			LAGG_RUNLOCK();
1373 			if_rele(tpif);
1374 			break;
1375 		}
1376 
1377 		lagg_port2req(lp, rp);
1378 		LAGG_RUNLOCK();
1379 		if_rele(tpif);
1380 		break;
1381 	case SIOCSLAGGPORT:
1382 		error = priv_check(td, PRIV_NET_LAGG);
1383 		if (error)
1384 			break;
1385 		if (rp->rp_portname[0] == '\0' ||
1386 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1387 			error = EINVAL;
1388 			break;
1389 		}
1390 #ifdef INET6
1391 		/*
1392 		 * A laggport interface should not have inet6 address
1393 		 * because two interfaces with a valid link-local
1394 		 * scope zone must not be merged in any form.  This
1395 		 * restriction is needed to prevent violation of
1396 		 * link-local scope zone.  Attempts to add a laggport
1397 		 * interface which has inet6 addresses triggers
1398 		 * removal of all inet6 addresses on the member
1399 		 * interface.
1400 		 */
1401 		if (in6ifa_llaonifp(tpif)) {
1402 			in6_ifdetach(tpif);
1403 				if_printf(sc->sc_ifp,
1404 				    "IPv6 addresses on %s have been removed "
1405 				    "before adding it as a member to prevent "
1406 				    "IPv6 address scope violation.\n",
1407 				    tpif->if_xname);
1408 		}
1409 #endif
1410 		LAGG_XLOCK(sc);
1411 		error = lagg_port_create(sc, tpif);
1412 		LAGG_XUNLOCK(sc);
1413 		if_rele(tpif);
1414 		VLAN_CAPABILITIES(ifp);
1415 		break;
1416 	case SIOCSLAGGDELPORT:
1417 		error = priv_check(td, PRIV_NET_LAGG);
1418 		if (error)
1419 			break;
1420 		if (rp->rp_portname[0] == '\0' ||
1421 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1422 			error = EINVAL;
1423 			break;
1424 		}
1425 
1426 		LAGG_XLOCK(sc);
1427 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1428 		    lp->lp_softc != sc) {
1429 			error = ENOENT;
1430 			LAGG_XUNLOCK(sc);
1431 			if_rele(tpif);
1432 			break;
1433 		}
1434 
1435 		error = lagg_port_destroy(lp, 1);
1436 		LAGG_XUNLOCK(sc);
1437 		if_rele(tpif);
1438 		VLAN_CAPABILITIES(ifp);
1439 		break;
1440 	case SIOCSIFFLAGS:
1441 		/* Set flags on ports too */
1442 		LAGG_XLOCK(sc);
1443 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1444 			lagg_setflags(lp, 1);
1445 		}
1446 
1447 		if (!(ifp->if_flags & IFF_UP) &&
1448 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1449 			/*
1450 			 * If interface is marked down and it is running,
1451 			 * then stop and disable it.
1452 			 */
1453 			lagg_stop(sc);
1454 			LAGG_XUNLOCK(sc);
1455 		} else if ((ifp->if_flags & IFF_UP) &&
1456 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1457 			/*
1458 			 * If interface is marked up and it is stopped, then
1459 			 * start it.
1460 			 */
1461 			LAGG_XUNLOCK(sc);
1462 			(*ifp->if_init)(sc);
1463 		} else
1464 			LAGG_XUNLOCK(sc);
1465 		break;
1466 	case SIOCADDMULTI:
1467 	case SIOCDELMULTI:
1468 		LAGG_XLOCK(sc);
1469 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1470 			lagg_clrmulti(lp);
1471 			lagg_setmulti(lp);
1472 		}
1473 		LAGG_XUNLOCK(sc);
1474 		error = 0;
1475 		break;
1476 	case SIOCSIFMEDIA:
1477 	case SIOCGIFMEDIA:
1478 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1479 		break;
1480 
1481 	case SIOCSIFCAP:
1482 		LAGG_XLOCK(sc);
1483 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1484 			if (lp->lp_ioctl != NULL)
1485 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1486 		}
1487 		lagg_capabilities(sc);
1488 		LAGG_XUNLOCK(sc);
1489 		VLAN_CAPABILITIES(ifp);
1490 		error = 0;
1491 		break;
1492 
1493 	case SIOCSIFMTU:
1494 		LAGG_XLOCK(sc);
1495 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1496 			if (lp->lp_ioctl != NULL)
1497 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1498 			else
1499 				error = EINVAL;
1500 			if (error != 0) {
1501 				if_printf(ifp,
1502 				    "failed to change MTU to %d on port %s, "
1503 				    "reverting all ports to original MTU (%d)\n",
1504 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1505 				break;
1506 			}
1507 		}
1508 		if (error == 0) {
1509 			ifp->if_mtu = ifr->ifr_mtu;
1510 		} else {
1511 			/* set every port back to the original MTU */
1512 			ifr->ifr_mtu = ifp->if_mtu;
1513 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1514 				if (lp->lp_ioctl != NULL)
1515 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1516 			}
1517 		}
1518 		LAGG_XUNLOCK(sc);
1519 		break;
1520 
1521 	default:
1522 		error = ether_ioctl(ifp, cmd, data);
1523 		break;
1524 	}
1525 	return (error);
1526 }
1527 
1528 #ifdef RATELIMIT
1529 static int
1530 lagg_snd_tag_alloc(struct ifnet *ifp,
1531     union if_snd_tag_alloc_params *params,
1532     struct m_snd_tag **ppmt)
1533 {
1534 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1535 	struct lagg_port *lp;
1536 	struct lagg_lb *lb;
1537 	uint32_t p;
1538 
1539 	LAGG_RLOCK();
1540 	switch (sc->sc_proto) {
1541 	case LAGG_PROTO_FAILOVER:
1542 		lp = lagg_link_active(sc, sc->sc_primary);
1543 		break;
1544 	case LAGG_PROTO_LOADBALANCE:
1545 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1546 		    params->hdr.flowtype == M_HASHTYPE_NONE) {
1547 			LAGG_RUNLOCK();
1548 			return (EOPNOTSUPP);
1549 		}
1550 		p = params->hdr.flowid >> sc->flowid_shift;
1551 		p %= sc->sc_count;
1552 		lb = (struct lagg_lb *)sc->sc_psc;
1553 		lp = lb->lb_ports[p];
1554 		lp = lagg_link_active(sc, lp);
1555 		break;
1556 	case LAGG_PROTO_LACP:
1557 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1558 		    params->hdr.flowtype == M_HASHTYPE_NONE) {
1559 			LAGG_RUNLOCK();
1560 			return (EOPNOTSUPP);
1561 		}
1562 		lp = lacp_select_tx_port_by_hash(sc, params->hdr.flowid);
1563 		break;
1564 	default:
1565 		LAGG_RUNLOCK();
1566 		return (EOPNOTSUPP);
1567 	}
1568 	if (lp == NULL) {
1569 		LAGG_RUNLOCK();
1570 		return (EOPNOTSUPP);
1571 	}
1572 	ifp = lp->lp_ifp;
1573 	LAGG_RUNLOCK();
1574 	if (ifp == NULL || ifp->if_snd_tag_alloc == NULL ||
1575 	    (ifp->if_capenable & IFCAP_TXRTLMT) == 0)
1576 		return (EOPNOTSUPP);
1577 
1578 	/* forward allocation request */
1579 	return (ifp->if_snd_tag_alloc(ifp, params, ppmt));
1580 }
1581 
1582 static void
1583 lagg_snd_tag_free(struct m_snd_tag *tag)
1584 {
1585 	tag->ifp->if_snd_tag_free(tag);
1586 }
1587 
1588 #endif
1589 
1590 static int
1591 lagg_setmulti(struct lagg_port *lp)
1592 {
1593 	struct lagg_softc *sc = lp->lp_softc;
1594 	struct ifnet *ifp = lp->lp_ifp;
1595 	struct ifnet *scifp = sc->sc_ifp;
1596 	struct lagg_mc *mc;
1597 	struct ifmultiaddr *ifma;
1598 	int error;
1599 
1600 	IF_ADDR_WLOCK(scifp);
1601 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1602 		if (ifma->ifma_addr->sa_family != AF_LINK)
1603 			continue;
1604 		mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
1605 		if (mc == NULL) {
1606 			IF_ADDR_WUNLOCK(scifp);
1607 			return (ENOMEM);
1608 		}
1609 		bcopy(ifma->ifma_addr, &mc->mc_addr,
1610 		    ifma->ifma_addr->sa_len);
1611 		mc->mc_addr.sdl_index = ifp->if_index;
1612 		mc->mc_ifma = NULL;
1613 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1614 	}
1615 	IF_ADDR_WUNLOCK(scifp);
1616 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1617 		error = if_addmulti(ifp,
1618 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1619 		if (error)
1620 			return (error);
1621 	}
1622 	return (0);
1623 }
1624 
1625 static int
1626 lagg_clrmulti(struct lagg_port *lp)
1627 {
1628 	struct lagg_mc *mc;
1629 
1630 	LAGG_XLOCK_ASSERT(lp->lp_softc);
1631 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1632 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1633 		if (mc->mc_ifma && lp->lp_detaching == 0)
1634 			if_delmulti_ifma(mc->mc_ifma);
1635 		free(mc, M_LAGG);
1636 	}
1637 	return (0);
1638 }
1639 
1640 static int
1641 lagg_setcaps(struct lagg_port *lp, int cap)
1642 {
1643 	struct ifreq ifr;
1644 
1645 	if (lp->lp_ifp->if_capenable == cap)
1646 		return (0);
1647 	if (lp->lp_ioctl == NULL)
1648 		return (ENXIO);
1649 	ifr.ifr_reqcap = cap;
1650 	return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1651 }
1652 
1653 /* Handle a ref counted flag that should be set on the lagg port as well */
1654 static int
1655 lagg_setflag(struct lagg_port *lp, int flag, int status,
1656     int (*func)(struct ifnet *, int))
1657 {
1658 	struct lagg_softc *sc = lp->lp_softc;
1659 	struct ifnet *scifp = sc->sc_ifp;
1660 	struct ifnet *ifp = lp->lp_ifp;
1661 	int error;
1662 
1663 	LAGG_XLOCK_ASSERT(sc);
1664 
1665 	status = status ? (scifp->if_flags & flag) : 0;
1666 	/* Now "status" contains the flag value or 0 */
1667 
1668 	/*
1669 	 * See if recorded ports status is different from what
1670 	 * we want it to be.  If it is, flip it.  We record ports
1671 	 * status in lp_ifflags so that we won't clear ports flag
1672 	 * we haven't set.  In fact, we don't clear or set ports
1673 	 * flags directly, but get or release references to them.
1674 	 * That's why we can be sure that recorded flags still are
1675 	 * in accord with actual ports flags.
1676 	 */
1677 	if (status != (lp->lp_ifflags & flag)) {
1678 		error = (*func)(ifp, status);
1679 		if (error)
1680 			return (error);
1681 		lp->lp_ifflags &= ~flag;
1682 		lp->lp_ifflags |= status;
1683 	}
1684 	return (0);
1685 }
1686 
1687 /*
1688  * Handle IFF_* flags that require certain changes on the lagg port
1689  * if "status" is true, update ports flags respective to the lagg
1690  * if "status" is false, forcedly clear the flags set on port.
1691  */
1692 static int
1693 lagg_setflags(struct lagg_port *lp, int status)
1694 {
1695 	int error, i;
1696 
1697 	for (i = 0; lagg_pflags[i].flag; i++) {
1698 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1699 		    status, lagg_pflags[i].func);
1700 		if (error)
1701 			return (error);
1702 	}
1703 	return (0);
1704 }
1705 
1706 static int
1707 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1708 {
1709 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1710 	int error;
1711 
1712 	LAGG_RLOCK();
1713 	/* We need a Tx algorithm and at least one port */
1714 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1715 		LAGG_RUNLOCK();
1716 		m_freem(m);
1717 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1718 		return (ENXIO);
1719 	}
1720 
1721 	ETHER_BPF_MTAP(ifp, m);
1722 
1723 	error = lagg_proto_start(sc, m);
1724 	LAGG_RUNLOCK();
1725 
1726 	if (error != 0)
1727 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1728 
1729 	return (error);
1730 }
1731 
1732 /*
1733  * The ifp->if_qflush entry point for lagg(4) is no-op.
1734  */
1735 static void
1736 lagg_qflush(struct ifnet *ifp __unused)
1737 {
1738 }
1739 
1740 static struct mbuf *
1741 lagg_input(struct ifnet *ifp, struct mbuf *m)
1742 {
1743 	struct lagg_port *lp = ifp->if_lagg;
1744 	struct lagg_softc *sc = lp->lp_softc;
1745 	struct ifnet *scifp = sc->sc_ifp;
1746 
1747 	LAGG_RLOCK();
1748 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1749 	    lp->lp_detaching != 0 ||
1750 	    sc->sc_proto == LAGG_PROTO_NONE) {
1751 		LAGG_RUNLOCK();
1752 		m_freem(m);
1753 		return (NULL);
1754 	}
1755 
1756 	ETHER_BPF_MTAP(scifp, m);
1757 
1758 	m = lagg_proto_input(sc, lp, m);
1759 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
1760 		m_freem(m);
1761 		m = NULL;
1762 	}
1763 
1764 	LAGG_RUNLOCK();
1765 	return (m);
1766 }
1767 
1768 static int
1769 lagg_media_change(struct ifnet *ifp)
1770 {
1771 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1772 
1773 	if (sc->sc_ifflags & IFF_DEBUG)
1774 		printf("%s\n", __func__);
1775 
1776 	/* Ignore */
1777 	return (0);
1778 }
1779 
1780 static void
1781 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1782 {
1783 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1784 	struct lagg_port *lp;
1785 
1786 	imr->ifm_status = IFM_AVALID;
1787 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1788 
1789 	LAGG_RLOCK();
1790 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1791 		if (LAGG_PORTACTIVE(lp))
1792 			imr->ifm_status |= IFM_ACTIVE;
1793 	}
1794 	LAGG_RUNLOCK();
1795 }
1796 
1797 static void
1798 lagg_linkstate(struct lagg_softc *sc)
1799 {
1800 	struct lagg_port *lp;
1801 	int new_link = LINK_STATE_DOWN;
1802 	uint64_t speed;
1803 
1804 	LAGG_XLOCK_ASSERT(sc);
1805 
1806 	/* LACP handles link state itself */
1807 	if (sc->sc_proto == LAGG_PROTO_LACP)
1808 		return;
1809 
1810 	/* Our link is considered up if at least one of our ports is active */
1811 	LAGG_RLOCK();
1812 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1813 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1814 			new_link = LINK_STATE_UP;
1815 			break;
1816 		}
1817 	}
1818 	LAGG_RUNLOCK();
1819 	if_link_state_change(sc->sc_ifp, new_link);
1820 
1821 	/* Update if_baudrate to reflect the max possible speed */
1822 	switch (sc->sc_proto) {
1823 		case LAGG_PROTO_FAILOVER:
1824 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1825 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1826 			break;
1827 		case LAGG_PROTO_ROUNDROBIN:
1828 		case LAGG_PROTO_LOADBALANCE:
1829 		case LAGG_PROTO_BROADCAST:
1830 			speed = 0;
1831 			LAGG_RLOCK();
1832 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1833 				speed += lp->lp_ifp->if_baudrate;
1834 			LAGG_RUNLOCK();
1835 			sc->sc_ifp->if_baudrate = speed;
1836 			break;
1837 		case LAGG_PROTO_LACP:
1838 			/* LACP updates if_baudrate itself */
1839 			break;
1840 	}
1841 }
1842 
1843 static void
1844 lagg_port_state(struct ifnet *ifp, int state)
1845 {
1846 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1847 	struct lagg_softc *sc = NULL;
1848 
1849 	if (lp != NULL)
1850 		sc = lp->lp_softc;
1851 	if (sc == NULL)
1852 		return;
1853 
1854 	LAGG_XLOCK(sc);
1855 	lagg_linkstate(sc);
1856 	lagg_proto_linkstate(sc, lp);
1857 	LAGG_XUNLOCK(sc);
1858 }
1859 
1860 struct lagg_port *
1861 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1862 {
1863 	struct lagg_port *lp_next, *rval = NULL;
1864 
1865 	/*
1866 	 * Search a port which reports an active link state.
1867 	 */
1868 
1869 	/*
1870 	 * This is called with either LAGG_RLOCK() held or
1871 	 * LAGG_XLOCK(sc) held.
1872 	 */
1873 	if (!in_epoch(net_epoch_preempt))
1874 		LAGG_XLOCK_ASSERT(sc);
1875 
1876 	if (lp == NULL)
1877 		goto search;
1878 	if (LAGG_PORTACTIVE(lp)) {
1879 		rval = lp;
1880 		goto found;
1881 	}
1882 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
1883 	    LAGG_PORTACTIVE(lp_next)) {
1884 		rval = lp_next;
1885 		goto found;
1886 	}
1887 
1888 search:
1889 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1890 		if (LAGG_PORTACTIVE(lp_next)) {
1891 			return (lp_next);
1892 		}
1893 	}
1894 found:
1895 	return (rval);
1896 }
1897 
1898 int
1899 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1900 {
1901 
1902 	return (ifp->if_transmit)(ifp, m);
1903 }
1904 
1905 /*
1906  * Simple round robin aggregation
1907  */
1908 static void
1909 lagg_rr_attach(struct lagg_softc *sc)
1910 {
1911 	sc->sc_seq = 0;
1912 	sc->sc_bkt_count = sc->sc_bkt;
1913 }
1914 
1915 static int
1916 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1917 {
1918 	struct lagg_port *lp;
1919 	uint32_t p;
1920 
1921 	if (sc->sc_bkt_count == 0 && sc->sc_bkt > 0)
1922 		sc->sc_bkt_count = sc->sc_bkt;
1923 
1924 	if (sc->sc_bkt > 0) {
1925 		atomic_subtract_int(&sc->sc_bkt_count, 1);
1926 	if (atomic_cmpset_int(&sc->sc_bkt_count, 0, sc->sc_bkt))
1927 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1928 	else
1929 		p = sc->sc_seq;
1930 	} else
1931 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1932 
1933 	p %= sc->sc_count;
1934 	lp = CK_SLIST_FIRST(&sc->sc_ports);
1935 
1936 	while (p--)
1937 		lp = CK_SLIST_NEXT(lp, lp_entries);
1938 
1939 	/*
1940 	 * Check the port's link state. This will return the next active
1941 	 * port if the link is down or the port is NULL.
1942 	 */
1943 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1944 		m_freem(m);
1945 		return (ENETDOWN);
1946 	}
1947 
1948 	/* Send mbuf */
1949 	return (lagg_enqueue(lp->lp_ifp, m));
1950 }
1951 
1952 static struct mbuf *
1953 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1954 {
1955 	struct ifnet *ifp = sc->sc_ifp;
1956 
1957 	/* Just pass in the packet to our lagg device */
1958 	m->m_pkthdr.rcvif = ifp;
1959 
1960 	return (m);
1961 }
1962 
1963 /*
1964  * Broadcast mode
1965  */
1966 static int
1967 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1968 {
1969 	int active_ports = 0;
1970 	int errors = 0;
1971 	int ret;
1972 	struct lagg_port *lp, *last = NULL;
1973 	struct mbuf *m0;
1974 
1975 	LAGG_RLOCK_ASSERT();
1976 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1977 		if (!LAGG_PORTACTIVE(lp))
1978 			continue;
1979 
1980 		active_ports++;
1981 
1982 		if (last != NULL) {
1983 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
1984 			if (m0 == NULL) {
1985 				ret = ENOBUFS;
1986 				errors++;
1987 				break;
1988 			}
1989 
1990 			ret = lagg_enqueue(last->lp_ifp, m0);
1991 			if (ret != 0)
1992 				errors++;
1993 		}
1994 		last = lp;
1995 	}
1996 
1997 	if (last == NULL) {
1998 		m_freem(m);
1999 		return (ENOENT);
2000 	}
2001 	if ((last = lagg_link_active(sc, last)) == NULL) {
2002 		m_freem(m);
2003 		return (ENETDOWN);
2004 	}
2005 
2006 	ret = lagg_enqueue(last->lp_ifp, m);
2007 	if (ret != 0)
2008 		errors++;
2009 
2010 	if (errors == 0)
2011 		return (ret);
2012 
2013 	return (0);
2014 }
2015 
2016 static struct mbuf*
2017 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2018 {
2019 	struct ifnet *ifp = sc->sc_ifp;
2020 
2021 	/* Just pass in the packet to our lagg device */
2022 	m->m_pkthdr.rcvif = ifp;
2023 	return (m);
2024 }
2025 
2026 /*
2027  * Active failover
2028  */
2029 static int
2030 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2031 {
2032 	struct lagg_port *lp;
2033 
2034 	/* Use the master port if active or the next available port */
2035 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2036 		m_freem(m);
2037 		return (ENETDOWN);
2038 	}
2039 
2040 	/* Send mbuf */
2041 	return (lagg_enqueue(lp->lp_ifp, m));
2042 }
2043 
2044 static struct mbuf *
2045 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2046 {
2047 	struct ifnet *ifp = sc->sc_ifp;
2048 	struct lagg_port *tmp_tp;
2049 
2050 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2051 		m->m_pkthdr.rcvif = ifp;
2052 		return (m);
2053 	}
2054 
2055 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2056 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2057 		/*
2058 		 * If tmp_tp is null, we've received a packet when all
2059 		 * our links are down. Weird, but process it anyways.
2060 		 */
2061 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2062 			m->m_pkthdr.rcvif = ifp;
2063 			return (m);
2064 		}
2065 	}
2066 
2067 	m_freem(m);
2068 	return (NULL);
2069 }
2070 
2071 /*
2072  * Loadbalancing
2073  */
2074 static void
2075 lagg_lb_attach(struct lagg_softc *sc)
2076 {
2077 	struct lagg_port *lp;
2078 	struct lagg_lb *lb;
2079 
2080 	LAGG_XLOCK_ASSERT(sc);
2081 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2082 	lb->lb_key = m_ether_tcpip_hash_init();
2083 	sc->sc_psc = lb;
2084 
2085 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2086 		lagg_lb_port_create(lp);
2087 }
2088 
2089 static void
2090 lagg_lb_detach(struct lagg_softc *sc)
2091 {
2092 	struct lagg_lb *lb;
2093 
2094 	lb = (struct lagg_lb *)sc->sc_psc;
2095 	if (lb != NULL)
2096 		free(lb, M_LAGG);
2097 }
2098 
2099 static int
2100 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2101 {
2102 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2103 	struct lagg_port *lp_next;
2104 	int i = 0, rv;
2105 
2106 	rv = 0;
2107 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2108 	LAGG_XLOCK_ASSERT(sc);
2109 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2110 		if (lp_next == lp)
2111 			continue;
2112 		if (i >= LAGG_MAX_PORTS) {
2113 			rv = EINVAL;
2114 			break;
2115 		}
2116 		if (sc->sc_ifflags & IFF_DEBUG)
2117 			printf("%s: port %s at index %d\n",
2118 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2119 		lb->lb_ports[i++] = lp_next;
2120 	}
2121 
2122 	return (rv);
2123 }
2124 
2125 static int
2126 lagg_lb_port_create(struct lagg_port *lp)
2127 {
2128 	struct lagg_softc *sc = lp->lp_softc;
2129 	return (lagg_lb_porttable(sc, NULL));
2130 }
2131 
2132 static void
2133 lagg_lb_port_destroy(struct lagg_port *lp)
2134 {
2135 	struct lagg_softc *sc = lp->lp_softc;
2136 	lagg_lb_porttable(sc, lp);
2137 }
2138 
2139 static int
2140 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2141 {
2142 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2143 	struct lagg_port *lp = NULL;
2144 	uint32_t p = 0;
2145 
2146 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2147 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2148 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2149 	else
2150 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2151 	p %= sc->sc_count;
2152 	lp = lb->lb_ports[p];
2153 
2154 	/*
2155 	 * Check the port's link state. This will return the next active
2156 	 * port if the link is down or the port is NULL.
2157 	 */
2158 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2159 		m_freem(m);
2160 		return (ENETDOWN);
2161 	}
2162 
2163 	/* Send mbuf */
2164 	return (lagg_enqueue(lp->lp_ifp, m));
2165 }
2166 
2167 static struct mbuf *
2168 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2169 {
2170 	struct ifnet *ifp = sc->sc_ifp;
2171 
2172 	/* Just pass in the packet to our lagg device */
2173 	m->m_pkthdr.rcvif = ifp;
2174 
2175 	return (m);
2176 }
2177 
2178 /*
2179  * 802.3ad LACP
2180  */
2181 static void
2182 lagg_lacp_attach(struct lagg_softc *sc)
2183 {
2184 	struct lagg_port *lp;
2185 
2186 	lacp_attach(sc);
2187 	LAGG_XLOCK_ASSERT(sc);
2188 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2189 		lacp_port_create(lp);
2190 }
2191 
2192 static void
2193 lagg_lacp_detach(struct lagg_softc *sc)
2194 {
2195 	struct lagg_port *lp;
2196 	void *psc;
2197 
2198 	LAGG_XLOCK_ASSERT(sc);
2199 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2200 		lacp_port_destroy(lp);
2201 
2202 	psc = sc->sc_psc;
2203 	sc->sc_psc = NULL;
2204 	lacp_detach(psc);
2205 }
2206 
2207 static void
2208 lagg_lacp_lladdr(struct lagg_softc *sc)
2209 {
2210 	struct lagg_port *lp;
2211 
2212 	LAGG_SXLOCK_ASSERT(sc);
2213 
2214 	/* purge all the lacp ports */
2215 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2216 		lacp_port_destroy(lp);
2217 
2218 	/* add them back in */
2219 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2220 		lacp_port_create(lp);
2221 }
2222 
2223 static int
2224 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2225 {
2226 	struct lagg_port *lp;
2227 
2228 	lp = lacp_select_tx_port(sc, m);
2229 	if (lp == NULL) {
2230 		m_freem(m);
2231 		return (ENETDOWN);
2232 	}
2233 
2234 	/* Send mbuf */
2235 	return (lagg_enqueue(lp->lp_ifp, m));
2236 }
2237 
2238 static struct mbuf *
2239 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2240 {
2241 	struct ifnet *ifp = sc->sc_ifp;
2242 	struct ether_header *eh;
2243 	u_short etype;
2244 
2245 	eh = mtod(m, struct ether_header *);
2246 	etype = ntohs(eh->ether_type);
2247 
2248 	/* Tap off LACP control messages */
2249 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2250 		m = lacp_input(lp, m);
2251 		if (m == NULL)
2252 			return (NULL);
2253 	}
2254 
2255 	/*
2256 	 * If the port is not collecting or not in the active aggregator then
2257 	 * free and return.
2258 	 */
2259 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2260 		m_freem(m);
2261 		return (NULL);
2262 	}
2263 
2264 	m->m_pkthdr.rcvif = ifp;
2265 	return (m);
2266 }
2267 
2268