xref: /freebsd/sys/net/if_lagg.c (revision 193d9e768ba63fcfb187cfd17f461f7d41345048)
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/taskqueue.h>
43 #include <sys/eventhandler.h>
44 
45 #include <net/ethernet.h>
46 #include <net/if.h>
47 #include <net/if_clone.h>
48 #include <net/if_arp.h>
49 #include <net/if_dl.h>
50 #include <net/if_media.h>
51 #include <net/if_types.h>
52 #include <net/if_var.h>
53 #include <net/bpf.h>
54 #include <net/vnet.h>
55 
56 #if defined(INET) || defined(INET6)
57 #include <netinet/in.h>
58 #include <netinet/ip.h>
59 #endif
60 #ifdef INET
61 #include <netinet/in_systm.h>
62 #include <netinet/if_ether.h>
63 #endif
64 
65 #ifdef INET6
66 #include <netinet/ip6.h>
67 #include <netinet6/in6_var.h>
68 #include <netinet6/in6_ifattach.h>
69 #endif
70 
71 #include <net/if_vlan_var.h>
72 #include <net/if_lagg.h>
73 #include <net/ieee8023ad_lacp.h>
74 
75 /* Special flags we should propagate to the lagg ports. */
76 static struct {
77 	int flag;
78 	int (*func)(struct ifnet *, int);
79 } lagg_pflags[] = {
80 	{IFF_PROMISC, ifpromisc},
81 	{IFF_ALLMULTI, if_allmulti},
82 	{0, NULL}
83 };
84 
85 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
86 #define	V_lagg_list	VNET(lagg_list)
87 static VNET_DEFINE(struct mtx, lagg_list_mtx);
88 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
89 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
90 					"if_lagg list", NULL, MTX_DEF)
91 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
92 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
93 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
94 eventhandler_tag	lagg_detach_cookie = NULL;
95 
96 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
97 static void	lagg_clone_destroy(struct ifnet *);
98 static VNET_DEFINE(struct if_clone *, lagg_cloner);
99 #define	V_lagg_cloner	VNET(lagg_cloner)
100 static const char laggname[] = "lagg";
101 
102 static void	lagg_lladdr(struct lagg_softc *, uint8_t *);
103 static void	lagg_capabilities(struct lagg_softc *);
104 static void	lagg_port_lladdr(struct lagg_port *, uint8_t *, lagg_llqtype);
105 static void	lagg_port_setlladdr(void *, int);
106 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
107 static int	lagg_port_destroy(struct lagg_port *, int);
108 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
109 static void	lagg_linkstate(struct lagg_softc *);
110 static void	lagg_port_state(struct ifnet *, int);
111 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
112 static int	lagg_port_output(struct ifnet *, struct mbuf *,
113 		    const struct sockaddr *, struct route *);
114 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
115 #ifdef LAGG_PORT_STACKING
116 static int	lagg_port_checkstacking(struct lagg_softc *);
117 #endif
118 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
119 static void	lagg_init(void *);
120 static void	lagg_stop(struct lagg_softc *);
121 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
122 #ifdef RATELIMIT
123 static int	lagg_snd_tag_alloc(struct ifnet *,
124 		    union if_snd_tag_alloc_params *,
125 		    struct m_snd_tag **);
126 #endif
127 static int	lagg_ether_setmulti(struct lagg_softc *);
128 static int	lagg_ether_cmdmulti(struct lagg_port *, int);
129 static	int	lagg_setflag(struct lagg_port *, int, int,
130 		    int (*func)(struct ifnet *, int));
131 static	int	lagg_setflags(struct lagg_port *, int status);
132 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
133 static int	lagg_transmit(struct ifnet *, struct mbuf *);
134 static void	lagg_qflush(struct ifnet *);
135 static int	lagg_media_change(struct ifnet *);
136 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
137 static struct lagg_port *lagg_link_active(struct lagg_softc *,
138 	    struct lagg_port *);
139 
140 /* Simple round robin */
141 static void	lagg_rr_attach(struct lagg_softc *);
142 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
143 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
144 		    struct mbuf *);
145 
146 /* Active failover */
147 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
148 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
149 		    struct mbuf *);
150 
151 /* Loadbalancing */
152 static void	lagg_lb_attach(struct lagg_softc *);
153 static void	lagg_lb_detach(struct lagg_softc *);
154 static int	lagg_lb_port_create(struct lagg_port *);
155 static void	lagg_lb_port_destroy(struct lagg_port *);
156 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
157 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
158 		    struct mbuf *);
159 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
160 
161 /* Broadcast */
162 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
163 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
164 		    struct mbuf *);
165 
166 /* 802.3ad LACP */
167 static void	lagg_lacp_attach(struct lagg_softc *);
168 static void	lagg_lacp_detach(struct lagg_softc *);
169 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
170 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
171 		    struct mbuf *);
172 static void	lagg_lacp_lladdr(struct lagg_softc *);
173 
174 /* lagg protocol table */
175 static const struct lagg_proto {
176 	lagg_proto	pr_num;
177 	void		(*pr_attach)(struct lagg_softc *);
178 	void		(*pr_detach)(struct lagg_softc *);
179 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
180 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
181 			    struct mbuf *);
182 	int		(*pr_addport)(struct lagg_port *);
183 	void		(*pr_delport)(struct lagg_port *);
184 	void		(*pr_linkstate)(struct lagg_port *);
185 	void 		(*pr_init)(struct lagg_softc *);
186 	void 		(*pr_stop)(struct lagg_softc *);
187 	void 		(*pr_lladdr)(struct lagg_softc *);
188 	void		(*pr_request)(struct lagg_softc *, void *);
189 	void		(*pr_portreq)(struct lagg_port *, void *);
190 } lagg_protos[] = {
191     {
192 	.pr_num = LAGG_PROTO_NONE
193     },
194     {
195 	.pr_num = LAGG_PROTO_ROUNDROBIN,
196 	.pr_attach = lagg_rr_attach,
197 	.pr_start = lagg_rr_start,
198 	.pr_input = lagg_rr_input,
199     },
200     {
201 	.pr_num = LAGG_PROTO_FAILOVER,
202 	.pr_start = lagg_fail_start,
203 	.pr_input = lagg_fail_input,
204     },
205     {
206 	.pr_num = LAGG_PROTO_LOADBALANCE,
207 	.pr_attach = lagg_lb_attach,
208 	.pr_detach = lagg_lb_detach,
209 	.pr_start = lagg_lb_start,
210 	.pr_input = lagg_lb_input,
211 	.pr_addport = lagg_lb_port_create,
212 	.pr_delport = lagg_lb_port_destroy,
213     },
214     {
215 	.pr_num = LAGG_PROTO_LACP,
216 	.pr_attach = lagg_lacp_attach,
217 	.pr_detach = lagg_lacp_detach,
218 	.pr_start = lagg_lacp_start,
219 	.pr_input = lagg_lacp_input,
220 	.pr_addport = lacp_port_create,
221 	.pr_delport = lacp_port_destroy,
222 	.pr_linkstate = lacp_linkstate,
223 	.pr_init = lacp_init,
224 	.pr_stop = lacp_stop,
225 	.pr_lladdr = lagg_lacp_lladdr,
226 	.pr_request = lacp_req,
227 	.pr_portreq = lacp_portreq,
228     },
229     {
230 	.pr_num = LAGG_PROTO_BROADCAST,
231 	.pr_start = lagg_bcast_start,
232 	.pr_input = lagg_bcast_input,
233     },
234 };
235 
236 SYSCTL_DECL(_net_link);
237 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
238     "Link Aggregation");
239 
240 /* Allow input on any failover links */
241 static VNET_DEFINE(int, lagg_failover_rx_all);
242 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
243 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
244     &VNET_NAME(lagg_failover_rx_all), 0,
245     "Accept input from any interface in a failover lagg");
246 
247 /* Default value for using flowid */
248 static VNET_DEFINE(int, def_use_flowid) = 1;
249 #define	V_def_use_flowid	VNET(def_use_flowid)
250 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
251     &VNET_NAME(def_use_flowid), 0,
252     "Default setting for using flow id for load sharing");
253 
254 /* Default value for flowid shift */
255 static VNET_DEFINE(int, def_flowid_shift) = 16;
256 #define	V_def_flowid_shift	VNET(def_flowid_shift)
257 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
258     &VNET_NAME(def_flowid_shift), 0,
259     "Default setting for flowid shift for load sharing");
260 
261 static void
262 vnet_lagg_init(const void *unused __unused)
263 {
264 
265 	LAGG_LIST_LOCK_INIT();
266 	SLIST_INIT(&V_lagg_list);
267 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
268 	    lagg_clone_destroy, 0);
269 }
270 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
271     vnet_lagg_init, NULL);
272 
273 static void
274 vnet_lagg_uninit(const void *unused __unused)
275 {
276 
277 	if_clone_detach(V_lagg_cloner);
278 	LAGG_LIST_LOCK_DESTROY();
279 }
280 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
281     vnet_lagg_uninit, NULL);
282 
283 static int
284 lagg_modevent(module_t mod, int type, void *data)
285 {
286 
287 	switch (type) {
288 	case MOD_LOAD:
289 		lagg_input_p = lagg_input;
290 		lagg_linkstate_p = lagg_port_state;
291 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
292 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
293 		    EVENTHANDLER_PRI_ANY);
294 		break;
295 	case MOD_UNLOAD:
296 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
297 		    lagg_detach_cookie);
298 		lagg_input_p = NULL;
299 		lagg_linkstate_p = NULL;
300 		break;
301 	default:
302 		return (EOPNOTSUPP);
303 	}
304 	return (0);
305 }
306 
307 static moduledata_t lagg_mod = {
308 	"if_lagg",
309 	lagg_modevent,
310 	0
311 };
312 
313 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
314 MODULE_VERSION(if_lagg, 1);
315 
316 static void
317 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
318 {
319 
320 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
321 	    __func__, sc));
322 
323 	if (sc->sc_ifflags & IFF_DEBUG)
324 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
325 
326 	if (lagg_protos[pr].pr_attach != NULL)
327 		lagg_protos[pr].pr_attach(sc);
328 	sc->sc_proto = pr;
329 }
330 
331 static void
332 lagg_proto_detach(struct lagg_softc *sc)
333 {
334 	lagg_proto pr;
335 
336 	LAGG_WLOCK_ASSERT(sc);
337 
338 	pr = sc->sc_proto;
339 	sc->sc_proto = LAGG_PROTO_NONE;
340 
341 	if (lagg_protos[pr].pr_detach != NULL)
342 		lagg_protos[pr].pr_detach(sc);
343 	else
344 		LAGG_WUNLOCK(sc);
345 }
346 
347 static int
348 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
349 {
350 
351 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
352 }
353 
354 static struct mbuf *
355 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
356 {
357 
358 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
359 }
360 
361 static int
362 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
363 {
364 
365 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
366 		return (0);
367 	else
368 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
369 }
370 
371 static void
372 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
373 {
374 
375 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
376 		lagg_protos[sc->sc_proto].pr_delport(lp);
377 }
378 
379 static void
380 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
381 {
382 
383 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
384 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
385 }
386 
387 static void
388 lagg_proto_init(struct lagg_softc *sc)
389 {
390 
391 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
392 		lagg_protos[sc->sc_proto].pr_init(sc);
393 }
394 
395 static void
396 lagg_proto_stop(struct lagg_softc *sc)
397 {
398 
399 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
400 		lagg_protos[sc->sc_proto].pr_stop(sc);
401 }
402 
403 static void
404 lagg_proto_lladdr(struct lagg_softc *sc)
405 {
406 
407 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
408 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
409 }
410 
411 static void
412 lagg_proto_request(struct lagg_softc *sc, void *v)
413 {
414 
415 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
416 		lagg_protos[sc->sc_proto].pr_request(sc, v);
417 }
418 
419 static void
420 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
421 {
422 
423 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
424 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
425 }
426 
427 /*
428  * This routine is run via an vlan
429  * config EVENT
430  */
431 static void
432 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
433 {
434 	struct lagg_softc *sc = ifp->if_softc;
435 	struct lagg_port *lp;
436 	struct rm_priotracker tracker;
437 
438 	if (ifp->if_softc !=  arg)   /* Not our event */
439 		return;
440 
441 	LAGG_RLOCK(sc, &tracker);
442 	if (!SLIST_EMPTY(&sc->sc_ports)) {
443 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
444 			EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
445 	}
446 	LAGG_RUNLOCK(sc, &tracker);
447 }
448 
449 /*
450  * This routine is run via an vlan
451  * unconfig EVENT
452  */
453 static void
454 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
455 {
456 	struct lagg_softc *sc = ifp->if_softc;
457 	struct lagg_port *lp;
458 	struct rm_priotracker tracker;
459 
460 	if (ifp->if_softc !=  arg)   /* Not our event */
461 		return;
462 
463 	LAGG_RLOCK(sc, &tracker);
464 	if (!SLIST_EMPTY(&sc->sc_ports)) {
465 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
466 			EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
467 	}
468 	LAGG_RUNLOCK(sc, &tracker);
469 }
470 
471 static int
472 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
473 {
474 	struct lagg_softc *sc;
475 	struct ifnet *ifp;
476 	static const u_char eaddr[6];	/* 00:00:00:00:00:00 */
477 
478 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
479 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
480 	if (ifp == NULL) {
481 		free(sc, M_DEVBUF);
482 		return (ENOSPC);
483 	}
484 
485 	if (V_def_use_flowid)
486 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
487 	sc->flowid_shift = V_def_flowid_shift;
488 
489 	/* Hash all layers by default */
490 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
491 
492 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
493 
494 	LAGG_LOCK_INIT(sc);
495 	SLIST_INIT(&sc->sc_ports);
496 	TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
497 
498 	/* Initialise pseudo media types */
499 	ifmedia_init(&sc->sc_media, 0, lagg_media_change,
500 	    lagg_media_status);
501 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
502 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
503 
504 	if_initname(ifp, laggname, unit);
505 	ifp->if_softc = sc;
506 	ifp->if_transmit = lagg_transmit;
507 	ifp->if_qflush = lagg_qflush;
508 	ifp->if_init = lagg_init;
509 	ifp->if_ioctl = lagg_ioctl;
510 	ifp->if_get_counter = lagg_get_counter;
511 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
512 #ifdef RATELIMIT
513 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
514 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS | IFCAP_TXRTLMT;
515 #else
516 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
517 #endif
518 
519 	/*
520 	 * Attach as an ordinary ethernet device, children will be attached
521 	 * as special device IFT_IEEE8023ADLAG.
522 	 */
523 	ether_ifattach(ifp, eaddr);
524 
525 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
526 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
527 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
528 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
529 
530 	/* Insert into the global list of laggs */
531 	LAGG_LIST_LOCK();
532 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
533 	LAGG_LIST_UNLOCK();
534 
535 	return (0);
536 }
537 
538 static void
539 lagg_clone_destroy(struct ifnet *ifp)
540 {
541 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
542 	struct lagg_port *lp;
543 
544 	LAGG_WLOCK(sc);
545 
546 	lagg_stop(sc);
547 	ifp->if_flags &= ~IFF_UP;
548 
549 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
550 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
551 
552 	/* Shutdown and remove lagg ports */
553 	while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) {
554 		lp->lp_detaching = LAGG_CLONE_DESTROY;
555 		lagg_port_destroy(lp, 1);
556 	}
557 	/* Unhook the aggregation protocol */
558 	lagg_proto_detach(sc);
559 	LAGG_UNLOCK_ASSERT(sc);
560 
561 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
562 	ifmedia_removeall(&sc->sc_media);
563 	ether_ifdetach(ifp);
564 	if_free(ifp);
565 
566 	LAGG_LIST_LOCK();
567 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
568 	LAGG_LIST_UNLOCK();
569 
570 	LAGG_LOCK_DESTROY(sc);
571 	free(sc, M_DEVBUF);
572 }
573 
574 /*
575  * Set link-layer address on the lagg interface itself.
576  *
577  * Set noinline to be dtrace-friendly
578  */
579 static __noinline void
580 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
581 {
582 	struct ifnet *ifp = sc->sc_ifp;
583 	struct lagg_port lp;
584 
585 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
586 		return;
587 
588 	LAGG_WLOCK_ASSERT(sc);
589 	/*
590 	 * Set the link layer address on the lagg interface.
591 	 * lagg_proto_lladdr() notifies the MAC change to
592 	 * the aggregation protocol.  iflladdr_event handler which
593 	 * may trigger gratuitous ARPs for INET will be handled in
594 	 * a taskqueue.
595 	 */
596 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
597 	lagg_proto_lladdr(sc);
598 
599 	/*
600 	 * Send notification request for lagg interface
601 	 * itself. Note that new lladdr is already set.
602 	 */
603 	bzero(&lp, sizeof(lp));
604 	lp.lp_ifp = sc->sc_ifp;
605 	lp.lp_softc = sc;
606 
607 	/* Do not request lladdr change */
608 	lagg_port_lladdr(&lp, lladdr, LAGG_LLQTYPE_VIRT);
609 }
610 
611 static void
612 lagg_capabilities(struct lagg_softc *sc)
613 {
614 	struct lagg_port *lp;
615 	int cap = ~0, ena = ~0;
616 	u_long hwa = ~0UL;
617 	struct ifnet_hw_tsomax hw_tsomax;
618 
619 	LAGG_WLOCK_ASSERT(sc);
620 
621 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
622 
623 	/* Get capabilities from the lagg ports */
624 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
625 		cap &= lp->lp_ifp->if_capabilities;
626 		ena &= lp->lp_ifp->if_capenable;
627 		hwa &= lp->lp_ifp->if_hwassist;
628 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
629 	}
630 	cap = (cap == ~0 ? 0 : cap);
631 	ena = (ena == ~0 ? 0 : ena);
632 	hwa = (hwa == ~0 ? 0 : hwa);
633 
634 	if (sc->sc_ifp->if_capabilities != cap ||
635 	    sc->sc_ifp->if_capenable != ena ||
636 	    sc->sc_ifp->if_hwassist != hwa ||
637 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
638 		sc->sc_ifp->if_capabilities = cap;
639 		sc->sc_ifp->if_capenable = ena;
640 		sc->sc_ifp->if_hwassist = hwa;
641 		getmicrotime(&sc->sc_ifp->if_lastchange);
642 
643 		if (sc->sc_ifflags & IFF_DEBUG)
644 			if_printf(sc->sc_ifp,
645 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
646 	}
647 }
648 
649 /*
650  * Enqueue interface lladdr notification.
651  * If request is already queued, it is updated.
652  * If setting lladdr is also desired, @do_change has to be set to 1.
653  *
654  * Set noinline to be dtrace-friendly
655  */
656 static __noinline void
657 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr, lagg_llqtype llq_type)
658 {
659 	struct lagg_softc *sc = lp->lp_softc;
660 	struct ifnet *ifp = lp->lp_ifp;
661 	struct lagg_llq *llq;
662 
663 	LAGG_WLOCK_ASSERT(sc);
664 
665 	/*
666 	 * Do not enqueue requests where lladdr is the same for
667 	 * "physical" interfaces (e.g. ports in lagg)
668 	 */
669 	if (llq_type == LAGG_LLQTYPE_PHYS &&
670 	    memcmp(IF_LLADDR(ifp), lladdr, ETHER_ADDR_LEN) == 0)
671 		return;
672 
673 	/* Check to make sure its not already queued to be changed */
674 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
675 		if (llq->llq_ifp == ifp) {
676 			/* Update lladdr, it may have changed */
677 			bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
678 			return;
679 		}
680 	}
681 
682 	llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT | M_ZERO);
683 	if (llq == NULL)	/* XXX what to do */
684 		return;
685 
686 	llq->llq_ifp = ifp;
687 	llq->llq_type = llq_type;
688 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
689 	/* XXX: We should insert to tail */
690 	SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
691 
692 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
693 }
694 
695 /*
696  * Set the interface MAC address from a taskqueue to avoid a LOR.
697  *
698  * Set noinline to be dtrace-friendly
699  */
700 static __noinline void
701 lagg_port_setlladdr(void *arg, int pending)
702 {
703 	struct lagg_softc *sc = (struct lagg_softc *)arg;
704 	struct lagg_llq *llq, *head;
705 	struct ifnet *ifp;
706 
707 	/* Grab a local reference of the queue and remove it from the softc */
708 	LAGG_WLOCK(sc);
709 	head = SLIST_FIRST(&sc->sc_llq_head);
710 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
711 	LAGG_WUNLOCK(sc);
712 
713 	/*
714 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
715 	 * unlocked as we have the only reference to it.
716 	 */
717 	for (llq = head; llq != NULL; llq = head) {
718 		ifp = llq->llq_ifp;
719 
720 		CURVNET_SET(ifp->if_vnet);
721 
722 		/*
723 		 * Set the link layer address on the laggport interface.
724 		 * Note that if_setlladdr() or iflladdr_event handler
725 		 * may result in arp transmission / lltable updates.
726 		 */
727 		if (llq->llq_type == LAGG_LLQTYPE_PHYS)
728 			if_setlladdr(ifp, llq->llq_lladdr,
729 			    ETHER_ADDR_LEN);
730 		else
731 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
732 		CURVNET_RESTORE();
733 		head = SLIST_NEXT(llq, llq_entries);
734 		free(llq, M_DEVBUF);
735 	}
736 }
737 
738 static int
739 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
740 {
741 	struct lagg_softc *sc_ptr;
742 	struct lagg_port *lp, *tlp;
743 	int error, i;
744 	uint64_t *pval;
745 
746 	LAGG_WLOCK_ASSERT(sc);
747 
748 	/* Limit the maximal number of lagg ports */
749 	if (sc->sc_count >= LAGG_MAX_PORTS)
750 		return (ENOSPC);
751 
752 	/* Check if port has already been associated to a lagg */
753 	if (ifp->if_lagg != NULL) {
754 		/* Port is already in the current lagg? */
755 		lp = (struct lagg_port *)ifp->if_lagg;
756 		if (lp->lp_softc == sc)
757 			return (EEXIST);
758 		return (EBUSY);
759 	}
760 
761 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
762 	if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
763 		return (EPROTONOSUPPORT);
764 
765 	/* Allow the first Ethernet member to define the MTU */
766 	if (SLIST_EMPTY(&sc->sc_ports))
767 		sc->sc_ifp->if_mtu = ifp->if_mtu;
768 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
769 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
770 		    ifp->if_xname);
771 		return (EINVAL);
772 	}
773 
774 	if ((lp = malloc(sizeof(struct lagg_port),
775 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
776 		return (ENOMEM);
777 
778 	/* Check if port is a stacked lagg */
779 	LAGG_LIST_LOCK();
780 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
781 		if (ifp == sc_ptr->sc_ifp) {
782 			LAGG_LIST_UNLOCK();
783 			free(lp, M_DEVBUF);
784 			return (EINVAL);
785 			/* XXX disable stacking for the moment, its untested */
786 #ifdef LAGG_PORT_STACKING
787 			lp->lp_flags |= LAGG_PORT_STACK;
788 			if (lagg_port_checkstacking(sc_ptr) >=
789 			    LAGG_MAX_STACKING) {
790 				LAGG_LIST_UNLOCK();
791 				free(lp, M_DEVBUF);
792 				return (E2BIG);
793 			}
794 #endif
795 		}
796 	}
797 	LAGG_LIST_UNLOCK();
798 
799 	/* Change the interface type */
800 	lp->lp_iftype = ifp->if_type;
801 	ifp->if_type = IFT_IEEE8023ADLAG;
802 	ifp->if_lagg = lp;
803 	lp->lp_ioctl = ifp->if_ioctl;
804 	ifp->if_ioctl = lagg_port_ioctl;
805 	lp->lp_output = ifp->if_output;
806 	ifp->if_output = lagg_port_output;
807 
808 	lp->lp_ifp = ifp;
809 	lp->lp_softc = sc;
810 
811 	/* Save port link layer address */
812 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
813 
814 	if (SLIST_EMPTY(&sc->sc_ports)) {
815 		sc->sc_primary = lp;
816 		/* First port in lagg. Update/notify lagg lladdress */
817 		lagg_lladdr(sc, IF_LLADDR(ifp));
818 	} else {
819 
820 		/*
821 		 * Update link layer address for this port and
822 		 * send notifications to other subsystems.
823 		 */
824 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp), LAGG_LLQTYPE_PHYS);
825 	}
826 
827 	/*
828 	 * Insert into the list of ports.
829 	 * Keep ports sorted by if_index. It is handy, when configuration
830 	 * is predictable and `ifconfig laggN create ...` command
831 	 * will lead to the same result each time.
832 	 */
833 	SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
834 		if (tlp->lp_ifp->if_index < ifp->if_index && (
835 		    SLIST_NEXT(tlp, lp_entries) == NULL ||
836 		    SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index >
837 		    ifp->if_index))
838 			break;
839 	}
840 	if (tlp != NULL)
841 		SLIST_INSERT_AFTER(tlp, lp, lp_entries);
842 	else
843 		SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
844 	sc->sc_count++;
845 
846 	/* Update lagg capabilities */
847 	lagg_capabilities(sc);
848 	lagg_linkstate(sc);
849 
850 	/* Read port counters */
851 	pval = lp->port_counters.val;
852 	for (i = 0; i < IFCOUNTERS; i++, pval++)
853 		*pval = ifp->if_get_counter(ifp, i);
854 	/* Add multicast addresses and interface flags to this port */
855 	lagg_ether_cmdmulti(lp, 1);
856 	lagg_setflags(lp, 1);
857 
858 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
859 		/* Remove the port, without calling pr_delport. */
860 		lagg_port_destroy(lp, 0);
861 		return (error);
862 	}
863 
864 	return (0);
865 }
866 
867 #ifdef LAGG_PORT_STACKING
868 static int
869 lagg_port_checkstacking(struct lagg_softc *sc)
870 {
871 	struct lagg_softc *sc_ptr;
872 	struct lagg_port *lp;
873 	int m = 0;
874 
875 	LAGG_WLOCK_ASSERT(sc);
876 
877 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
878 		if (lp->lp_flags & LAGG_PORT_STACK) {
879 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
880 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
881 		}
882 	}
883 
884 	return (m + 1);
885 }
886 #endif
887 
888 static int
889 lagg_port_destroy(struct lagg_port *lp, int rundelport)
890 {
891 	struct lagg_softc *sc = lp->lp_softc;
892 	struct lagg_port *lp_ptr, *lp0;
893 	struct lagg_llq *llq;
894 	struct ifnet *ifp = lp->lp_ifp;
895 	uint64_t *pval, vdiff;
896 	int i;
897 
898 	LAGG_WLOCK_ASSERT(sc);
899 
900 	if (rundelport)
901 		lagg_proto_delport(sc, lp);
902 
903 	/*
904 	 * Remove multicast addresses and interface flags from this port and
905 	 * reset the MAC address, skip if the interface is being detached.
906 	 */
907 	if (lp->lp_detaching == 0) {
908 		lagg_ether_cmdmulti(lp, 0);
909 		lagg_setflags(lp, 0);
910 		lagg_port_lladdr(lp, lp->lp_lladdr, LAGG_LLQTYPE_PHYS);
911 	}
912 
913 	/* Restore interface */
914 	ifp->if_type = lp->lp_iftype;
915 	ifp->if_ioctl = lp->lp_ioctl;
916 	ifp->if_output = lp->lp_output;
917 	ifp->if_lagg = NULL;
918 
919 	/* Update detached port counters */
920 	pval = lp->port_counters.val;
921 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
922 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
923 		sc->detached_counters.val[i] += vdiff;
924 	}
925 
926 	/* Finally, remove the port from the lagg */
927 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
928 	sc->sc_count--;
929 
930 	/* Update the primary interface */
931 	if (lp == sc->sc_primary) {
932 		uint8_t lladdr[ETHER_ADDR_LEN];
933 
934 		if ((lp0 = SLIST_FIRST(&sc->sc_ports)) == NULL) {
935 			bzero(&lladdr, ETHER_ADDR_LEN);
936 		} else {
937 			bcopy(lp0->lp_lladdr,
938 			    lladdr, ETHER_ADDR_LEN);
939 		}
940 		if (lp->lp_detaching != LAGG_CLONE_DESTROY)
941 			lagg_lladdr(sc, lladdr);
942 
943 		/* Mark lp0 as new primary */
944 		sc->sc_primary = lp0;
945 
946 		/*
947 		 * Enqueue lladdr update/notification for each port
948 		 * (new primary needs update as well, to switch from
949 		 * old lladdr to its 'real' one).
950 		 */
951 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
952 			lagg_port_lladdr(lp_ptr, lladdr, LAGG_LLQTYPE_PHYS);
953 	}
954 
955 	/* Remove any pending lladdr changes from the queue */
956 	if (lp->lp_detaching != 0) {
957 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
958 			if (llq->llq_ifp == ifp) {
959 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
960 				    llq_entries);
961 				free(llq, M_DEVBUF);
962 				break;	/* Only appears once */
963 			}
964 		}
965 	}
966 
967 	if (lp->lp_ifflags)
968 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
969 
970 	free(lp, M_DEVBUF);
971 
972 	/* Update lagg capabilities */
973 	lagg_capabilities(sc);
974 	lagg_linkstate(sc);
975 
976 	return (0);
977 }
978 
979 static int
980 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
981 {
982 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
983 	struct lagg_softc *sc;
984 	struct lagg_port *lp = NULL;
985 	int error = 0;
986 	struct rm_priotracker tracker;
987 
988 	/* Should be checked by the caller */
989 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
990 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
991 		goto fallback;
992 
993 	switch (cmd) {
994 	case SIOCGLAGGPORT:
995 		if (rp->rp_portname[0] == '\0' ||
996 		    ifunit(rp->rp_portname) != ifp) {
997 			error = EINVAL;
998 			break;
999 		}
1000 
1001 		LAGG_RLOCK(sc, &tracker);
1002 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1003 			error = ENOENT;
1004 			LAGG_RUNLOCK(sc, &tracker);
1005 			break;
1006 		}
1007 
1008 		lagg_port2req(lp, rp);
1009 		LAGG_RUNLOCK(sc, &tracker);
1010 		break;
1011 
1012 	case SIOCSIFCAP:
1013 		if (lp->lp_ioctl == NULL) {
1014 			error = EINVAL;
1015 			break;
1016 		}
1017 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1018 		if (error)
1019 			break;
1020 
1021 		/* Update lagg interface capabilities */
1022 		LAGG_WLOCK(sc);
1023 		lagg_capabilities(sc);
1024 		LAGG_WUNLOCK(sc);
1025 		break;
1026 
1027 	case SIOCSIFMTU:
1028 		/* Do not allow the MTU to be changed once joined */
1029 		error = EINVAL;
1030 		break;
1031 
1032 	default:
1033 		goto fallback;
1034 	}
1035 
1036 	return (error);
1037 
1038 fallback:
1039 	if (lp != NULL && lp->lp_ioctl != NULL)
1040 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1041 
1042 	return (EINVAL);
1043 }
1044 
1045 /*
1046  * Requests counter @cnt data.
1047  *
1048  * Counter value is calculated the following way:
1049  * 1) for each port, sum  difference between current and "initial" measurements.
1050  * 2) add lagg logical interface counters.
1051  * 3) add data from detached_counters array.
1052  *
1053  * We also do the following things on ports attach/detach:
1054  * 1) On port attach we store all counters it has into port_counter array.
1055  * 2) On port detach we add the different between "initial" and
1056  *   current counters data to detached_counters array.
1057  */
1058 static uint64_t
1059 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1060 {
1061 	struct lagg_softc *sc;
1062 	struct lagg_port *lp;
1063 	struct ifnet *lpifp;
1064 	struct rm_priotracker tracker;
1065 	uint64_t newval, oldval, vsum;
1066 
1067 	/* Revise this when we've got non-generic counters. */
1068 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1069 
1070 	sc = (struct lagg_softc *)ifp->if_softc;
1071 	LAGG_RLOCK(sc, &tracker);
1072 
1073 	vsum = 0;
1074 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1075 		/* Saved attached value */
1076 		oldval = lp->port_counters.val[cnt];
1077 		/* current value */
1078 		lpifp = lp->lp_ifp;
1079 		newval = lpifp->if_get_counter(lpifp, cnt);
1080 		/* Calculate diff and save new */
1081 		vsum += newval - oldval;
1082 	}
1083 
1084 	/*
1085 	 * Add counter data which might be added by upper
1086 	 * layer protocols operating on logical interface.
1087 	 */
1088 	vsum += if_get_counter_default(ifp, cnt);
1089 
1090 	/*
1091 	 * Add counter data from detached ports counters
1092 	 */
1093 	vsum += sc->detached_counters.val[cnt];
1094 
1095 	LAGG_RUNLOCK(sc, &tracker);
1096 
1097 	return (vsum);
1098 }
1099 
1100 /*
1101  * For direct output to child ports.
1102  */
1103 static int
1104 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1105 	const struct sockaddr *dst, struct route *ro)
1106 {
1107 	struct lagg_port *lp = ifp->if_lagg;
1108 
1109 	switch (dst->sa_family) {
1110 		case pseudo_AF_HDRCMPLT:
1111 		case AF_UNSPEC:
1112 			return ((*lp->lp_output)(ifp, m, dst, ro));
1113 	}
1114 
1115 	/* drop any other frames */
1116 	m_freem(m);
1117 	return (ENETDOWN);
1118 }
1119 
1120 static void
1121 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1122 {
1123 	struct lagg_port *lp;
1124 	struct lagg_softc *sc;
1125 
1126 	if ((lp = ifp->if_lagg) == NULL)
1127 		return;
1128 	/* If the ifnet is just being renamed, don't do anything. */
1129 	if (ifp->if_flags & IFF_RENAMING)
1130 		return;
1131 
1132 	sc = lp->lp_softc;
1133 
1134 	LAGG_WLOCK(sc);
1135 	lp->lp_detaching = LAGG_PORT_DETACH;
1136 	lagg_port_destroy(lp, 1);
1137 	LAGG_WUNLOCK(sc);
1138 }
1139 
1140 static void
1141 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1142 {
1143 	struct lagg_softc *sc = lp->lp_softc;
1144 
1145 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1146 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1147 	rp->rp_prio = lp->lp_prio;
1148 	rp->rp_flags = lp->lp_flags;
1149 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1150 
1151 	/* Add protocol specific flags */
1152 	switch (sc->sc_proto) {
1153 		case LAGG_PROTO_FAILOVER:
1154 			if (lp == sc->sc_primary)
1155 				rp->rp_flags |= LAGG_PORT_MASTER;
1156 			if (lp == lagg_link_active(sc, sc->sc_primary))
1157 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1158 			break;
1159 
1160 		case LAGG_PROTO_ROUNDROBIN:
1161 		case LAGG_PROTO_LOADBALANCE:
1162 		case LAGG_PROTO_BROADCAST:
1163 			if (LAGG_PORTACTIVE(lp))
1164 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1165 			break;
1166 
1167 		case LAGG_PROTO_LACP:
1168 			/* LACP has a different definition of active */
1169 			if (lacp_isactive(lp))
1170 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1171 			if (lacp_iscollecting(lp))
1172 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1173 			if (lacp_isdistributing(lp))
1174 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1175 			break;
1176 	}
1177 
1178 }
1179 
1180 static void
1181 lagg_init(void *xsc)
1182 {
1183 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1184 	struct ifnet *ifp = sc->sc_ifp;
1185 	struct lagg_port *lp;
1186 
1187 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1188 		return;
1189 
1190 	LAGG_WLOCK(sc);
1191 
1192 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1193 
1194 	/*
1195 	 * Update the port lladdrs if needed.
1196 	 * This might be if_setlladdr() notification
1197 	 * that lladdr has been changed.
1198 	 */
1199 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1200 		lagg_port_lladdr(lp, IF_LLADDR(ifp), LAGG_LLQTYPE_PHYS);
1201 
1202 	lagg_proto_init(sc);
1203 
1204 	LAGG_WUNLOCK(sc);
1205 }
1206 
1207 static void
1208 lagg_stop(struct lagg_softc *sc)
1209 {
1210 	struct ifnet *ifp = sc->sc_ifp;
1211 
1212 	LAGG_WLOCK_ASSERT(sc);
1213 
1214 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1215 		return;
1216 
1217 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1218 
1219 	lagg_proto_stop(sc);
1220 }
1221 
1222 static int
1223 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1224 {
1225 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1226 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1227 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1228 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1229 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1230 	struct ifreq *ifr = (struct ifreq *)data;
1231 	struct lagg_port *lp;
1232 	struct ifnet *tpif;
1233 	struct thread *td = curthread;
1234 	char *buf, *outbuf;
1235 	int count, buflen, len, error = 0;
1236 	struct rm_priotracker tracker;
1237 
1238 	bzero(&rpbuf, sizeof(rpbuf));
1239 
1240 	switch (cmd) {
1241 	case SIOCGLAGG:
1242 		LAGG_RLOCK(sc, &tracker);
1243 		count = 0;
1244 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1245 			count++;
1246 		buflen = count * sizeof(struct lagg_reqport);
1247 		LAGG_RUNLOCK(sc, &tracker);
1248 
1249 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1250 
1251 		LAGG_RLOCK(sc, &tracker);
1252 		ra->ra_proto = sc->sc_proto;
1253 		lagg_proto_request(sc, &ra->ra_psc);
1254 		count = 0;
1255 		buf = outbuf;
1256 		len = min(ra->ra_size, buflen);
1257 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1258 			if (len < sizeof(rpbuf))
1259 				break;
1260 
1261 			lagg_port2req(lp, &rpbuf);
1262 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1263 			count++;
1264 			buf += sizeof(rpbuf);
1265 			len -= sizeof(rpbuf);
1266 		}
1267 		LAGG_RUNLOCK(sc, &tracker);
1268 		ra->ra_ports = count;
1269 		ra->ra_size = count * sizeof(rpbuf);
1270 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1271 		free(outbuf, M_TEMP);
1272 		break;
1273 	case SIOCSLAGG:
1274 		error = priv_check(td, PRIV_NET_LAGG);
1275 		if (error)
1276 			break;
1277 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1278 			error = EPROTONOSUPPORT;
1279 			break;
1280 		}
1281 
1282 		LAGG_WLOCK(sc);
1283 		lagg_proto_detach(sc);
1284 		LAGG_UNLOCK_ASSERT(sc);
1285 		lagg_proto_attach(sc, ra->ra_proto);
1286 		break;
1287 	case SIOCGLAGGOPTS:
1288 		ro->ro_opts = sc->sc_opts;
1289 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1290 			struct lacp_softc *lsc;
1291 
1292 			lsc = (struct lacp_softc *)sc->sc_psc;
1293 			if (lsc->lsc_debug.lsc_tx_test != 0)
1294 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1295 			if (lsc->lsc_debug.lsc_rx_test != 0)
1296 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1297 			if (lsc->lsc_strict_mode != 0)
1298 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1299 			if (lsc->lsc_fast_timeout != 0)
1300 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1301 
1302 			ro->ro_active = sc->sc_active;
1303 		} else {
1304 			ro->ro_active = 0;
1305 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1306 				ro->ro_active += LAGG_PORTACTIVE(lp);
1307 		}
1308 		ro->ro_bkt = sc->sc_bkt;
1309 		ro->ro_flapping = sc->sc_flapping;
1310 		ro->ro_flowid_shift = sc->flowid_shift;
1311 		break;
1312 	case SIOCSLAGGOPTS:
1313 		if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) {
1314 			if (ro->ro_bkt == 0)
1315 				sc->sc_bkt = 1; // Minimum 1 packet per iface.
1316 			else
1317 				sc->sc_bkt = ro->ro_bkt;
1318 		}
1319 		error = priv_check(td, PRIV_NET_LAGG);
1320 		if (error)
1321 			break;
1322 		if (ro->ro_opts == 0)
1323 			break;
1324 		/*
1325 		 * Set options.  LACP options are stored in sc->sc_psc,
1326 		 * not in sc_opts.
1327 		 */
1328 		int valid, lacp;
1329 
1330 		switch (ro->ro_opts) {
1331 		case LAGG_OPT_USE_FLOWID:
1332 		case -LAGG_OPT_USE_FLOWID:
1333 		case LAGG_OPT_FLOWIDSHIFT:
1334 			valid = 1;
1335 			lacp = 0;
1336 			break;
1337 		case LAGG_OPT_LACP_TXTEST:
1338 		case -LAGG_OPT_LACP_TXTEST:
1339 		case LAGG_OPT_LACP_RXTEST:
1340 		case -LAGG_OPT_LACP_RXTEST:
1341 		case LAGG_OPT_LACP_STRICT:
1342 		case -LAGG_OPT_LACP_STRICT:
1343 		case LAGG_OPT_LACP_TIMEOUT:
1344 		case -LAGG_OPT_LACP_TIMEOUT:
1345 			valid = lacp = 1;
1346 			break;
1347 		default:
1348 			valid = lacp = 0;
1349 			break;
1350 		}
1351 
1352 		LAGG_WLOCK(sc);
1353 
1354 		if (valid == 0 ||
1355 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1356 			/* Invalid combination of options specified. */
1357 			error = EINVAL;
1358 			LAGG_WUNLOCK(sc);
1359 			break;	/* Return from SIOCSLAGGOPTS. */
1360 		}
1361 		/*
1362 		 * Store new options into sc->sc_opts except for
1363 		 * FLOWIDSHIFT and LACP options.
1364 		 */
1365 		if (lacp == 0) {
1366 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1367 				sc->flowid_shift = ro->ro_flowid_shift;
1368 			else if (ro->ro_opts > 0)
1369 				sc->sc_opts |= ro->ro_opts;
1370 			else
1371 				sc->sc_opts &= ~ro->ro_opts;
1372 		} else {
1373 			struct lacp_softc *lsc;
1374 			struct lacp_port *lp;
1375 
1376 			lsc = (struct lacp_softc *)sc->sc_psc;
1377 
1378 			switch (ro->ro_opts) {
1379 			case LAGG_OPT_LACP_TXTEST:
1380 				lsc->lsc_debug.lsc_tx_test = 1;
1381 				break;
1382 			case -LAGG_OPT_LACP_TXTEST:
1383 				lsc->lsc_debug.lsc_tx_test = 0;
1384 				break;
1385 			case LAGG_OPT_LACP_RXTEST:
1386 				lsc->lsc_debug.lsc_rx_test = 1;
1387 				break;
1388 			case -LAGG_OPT_LACP_RXTEST:
1389 				lsc->lsc_debug.lsc_rx_test = 0;
1390 				break;
1391 			case LAGG_OPT_LACP_STRICT:
1392 				lsc->lsc_strict_mode = 1;
1393 				break;
1394 			case -LAGG_OPT_LACP_STRICT:
1395 				lsc->lsc_strict_mode = 0;
1396 				break;
1397 			case LAGG_OPT_LACP_TIMEOUT:
1398 				LACP_LOCK(lsc);
1399         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1400                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1401 				LACP_UNLOCK(lsc);
1402 				lsc->lsc_fast_timeout = 1;
1403 				break;
1404 			case -LAGG_OPT_LACP_TIMEOUT:
1405 				LACP_LOCK(lsc);
1406         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1407                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1408 				LACP_UNLOCK(lsc);
1409 				lsc->lsc_fast_timeout = 0;
1410 				break;
1411 			}
1412 		}
1413 		LAGG_WUNLOCK(sc);
1414 		break;
1415 	case SIOCGLAGGFLAGS:
1416 		rf->rf_flags = 0;
1417 		LAGG_RLOCK(sc, &tracker);
1418 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1419 			rf->rf_flags |= LAGG_F_HASHL2;
1420 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1421 			rf->rf_flags |= LAGG_F_HASHL3;
1422 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1423 			rf->rf_flags |= LAGG_F_HASHL4;
1424 		LAGG_RUNLOCK(sc, &tracker);
1425 		break;
1426 	case SIOCSLAGGHASH:
1427 		error = priv_check(td, PRIV_NET_LAGG);
1428 		if (error)
1429 			break;
1430 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1431 			error = EINVAL;
1432 			break;
1433 		}
1434 		LAGG_WLOCK(sc);
1435 		sc->sc_flags = 0;
1436 		if (rf->rf_flags & LAGG_F_HASHL2)
1437 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1438 		if (rf->rf_flags & LAGG_F_HASHL3)
1439 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1440 		if (rf->rf_flags & LAGG_F_HASHL4)
1441 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1442 		LAGG_WUNLOCK(sc);
1443 		break;
1444 	case SIOCGLAGGPORT:
1445 		if (rp->rp_portname[0] == '\0' ||
1446 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1447 			error = EINVAL;
1448 			break;
1449 		}
1450 
1451 		LAGG_RLOCK(sc, &tracker);
1452 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1453 		    lp->lp_softc != sc) {
1454 			error = ENOENT;
1455 			LAGG_RUNLOCK(sc, &tracker);
1456 			break;
1457 		}
1458 
1459 		lagg_port2req(lp, rp);
1460 		LAGG_RUNLOCK(sc, &tracker);
1461 		break;
1462 	case SIOCSLAGGPORT:
1463 		error = priv_check(td, PRIV_NET_LAGG);
1464 		if (error)
1465 			break;
1466 		if (rp->rp_portname[0] == '\0' ||
1467 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1468 			error = EINVAL;
1469 			break;
1470 		}
1471 #ifdef INET6
1472 		/*
1473 		 * A laggport interface should not have inet6 address
1474 		 * because two interfaces with a valid link-local
1475 		 * scope zone must not be merged in any form.  This
1476 		 * restriction is needed to prevent violation of
1477 		 * link-local scope zone.  Attempts to add a laggport
1478 		 * interface which has inet6 addresses triggers
1479 		 * removal of all inet6 addresses on the member
1480 		 * interface.
1481 		 */
1482 		if (in6ifa_llaonifp(tpif)) {
1483 			in6_ifdetach(tpif);
1484 				if_printf(sc->sc_ifp,
1485 				    "IPv6 addresses on %s have been removed "
1486 				    "before adding it as a member to prevent "
1487 				    "IPv6 address scope violation.\n",
1488 				    tpif->if_xname);
1489 		}
1490 #endif
1491 		LAGG_WLOCK(sc);
1492 		error = lagg_port_create(sc, tpif);
1493 		LAGG_WUNLOCK(sc);
1494 		break;
1495 	case SIOCSLAGGDELPORT:
1496 		error = priv_check(td, PRIV_NET_LAGG);
1497 		if (error)
1498 			break;
1499 		if (rp->rp_portname[0] == '\0' ||
1500 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1501 			error = EINVAL;
1502 			break;
1503 		}
1504 
1505 		LAGG_WLOCK(sc);
1506 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1507 		    lp->lp_softc != sc) {
1508 			error = ENOENT;
1509 			LAGG_WUNLOCK(sc);
1510 			break;
1511 		}
1512 
1513 		error = lagg_port_destroy(lp, 1);
1514 		LAGG_WUNLOCK(sc);
1515 		break;
1516 	case SIOCSIFFLAGS:
1517 		/* Set flags on ports too */
1518 		LAGG_WLOCK(sc);
1519 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1520 			lagg_setflags(lp, 1);
1521 		}
1522 		LAGG_WUNLOCK(sc);
1523 
1524 		if (!(ifp->if_flags & IFF_UP) &&
1525 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1526 			/*
1527 			 * If interface is marked down and it is running,
1528 			 * then stop and disable it.
1529 			 */
1530 			LAGG_WLOCK(sc);
1531 			lagg_stop(sc);
1532 			LAGG_WUNLOCK(sc);
1533 		} else if ((ifp->if_flags & IFF_UP) &&
1534 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1535 			/*
1536 			 * If interface is marked up and it is stopped, then
1537 			 * start it.
1538 			 */
1539 			(*ifp->if_init)(sc);
1540 		}
1541 		break;
1542 	case SIOCADDMULTI:
1543 	case SIOCDELMULTI:
1544 		LAGG_WLOCK(sc);
1545 		error = lagg_ether_setmulti(sc);
1546 		LAGG_WUNLOCK(sc);
1547 		break;
1548 	case SIOCSIFMEDIA:
1549 	case SIOCGIFMEDIA:
1550 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1551 		break;
1552 
1553 	case SIOCSIFCAP:
1554 	case SIOCSIFMTU:
1555 		/* Do not allow the MTU or caps to be directly changed */
1556 		error = EINVAL;
1557 		break;
1558 
1559 	default:
1560 		error = ether_ioctl(ifp, cmd, data);
1561 		break;
1562 	}
1563 	return (error);
1564 }
1565 
1566 #ifdef RATELIMIT
1567 static int
1568 lagg_snd_tag_alloc(struct ifnet *ifp,
1569     union if_snd_tag_alloc_params *params,
1570     struct m_snd_tag **ppmt)
1571 {
1572 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1573 	struct lagg_port *lp;
1574 	struct lagg_lb *lb;
1575 	uint32_t p;
1576 
1577 	switch (sc->sc_proto) {
1578 	case LAGG_PROTO_FAILOVER:
1579 		lp = lagg_link_active(sc, sc->sc_primary);
1580 		break;
1581 	case LAGG_PROTO_LOADBALANCE:
1582 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1583 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1584 			return (EOPNOTSUPP);
1585 		p = params->hdr.flowid >> sc->flowid_shift;
1586 		p %= sc->sc_count;
1587 		lb = (struct lagg_lb *)sc->sc_psc;
1588 		lp = lb->lb_ports[p];
1589 		lp = lagg_link_active(sc, lp);
1590 		break;
1591 	case LAGG_PROTO_LACP:
1592 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1593 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1594 			return (EOPNOTSUPP);
1595 		lp = lacp_select_tx_port_by_hash(sc, params->hdr.flowid);
1596 		break;
1597 	default:
1598 		return (EOPNOTSUPP);
1599 	}
1600 	if (lp == NULL)
1601 		return (EOPNOTSUPP);
1602 	ifp = lp->lp_ifp;
1603 	if (ifp == NULL || ifp->if_snd_tag_alloc == NULL ||
1604 	    (ifp->if_capenable & IFCAP_TXRTLMT) == 0)
1605 		return (EOPNOTSUPP);
1606 
1607 	/* forward allocation request */
1608 	return (ifp->if_snd_tag_alloc(ifp, params, ppmt));
1609 }
1610 #endif
1611 
1612 static int
1613 lagg_ether_setmulti(struct lagg_softc *sc)
1614 {
1615 	struct lagg_port *lp;
1616 
1617 	LAGG_WLOCK_ASSERT(sc);
1618 
1619 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1620 		/* First, remove any existing filter entries. */
1621 		lagg_ether_cmdmulti(lp, 0);
1622 		/* copy all addresses from the lagg interface to the port */
1623 		lagg_ether_cmdmulti(lp, 1);
1624 	}
1625 	return (0);
1626 }
1627 
1628 static int
1629 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1630 {
1631 	struct lagg_softc *sc = lp->lp_softc;
1632 	struct ifnet *ifp = lp->lp_ifp;
1633 	struct ifnet *scifp = sc->sc_ifp;
1634 	struct lagg_mc *mc;
1635 	struct ifmultiaddr *ifma;
1636 	int error;
1637 
1638 	LAGG_WLOCK_ASSERT(sc);
1639 
1640 	if (set) {
1641 		IF_ADDR_WLOCK(scifp);
1642 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1643 			if (ifma->ifma_addr->sa_family != AF_LINK)
1644 				continue;
1645 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1646 			if (mc == NULL) {
1647 				IF_ADDR_WUNLOCK(scifp);
1648 				return (ENOMEM);
1649 			}
1650 			bcopy(ifma->ifma_addr, &mc->mc_addr,
1651 			    ifma->ifma_addr->sa_len);
1652 			mc->mc_addr.sdl_index = ifp->if_index;
1653 			mc->mc_ifma = NULL;
1654 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1655 		}
1656 		IF_ADDR_WUNLOCK(scifp);
1657 		SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1658 			error = if_addmulti(ifp,
1659 			    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1660 			if (error)
1661 				return (error);
1662 		}
1663 	} else {
1664 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1665 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1666 			if (mc->mc_ifma && lp->lp_detaching == 0)
1667 				if_delmulti_ifma(mc->mc_ifma);
1668 			free(mc, M_DEVBUF);
1669 		}
1670 	}
1671 	return (0);
1672 }
1673 
1674 /* Handle a ref counted flag that should be set on the lagg port as well */
1675 static int
1676 lagg_setflag(struct lagg_port *lp, int flag, int status,
1677     int (*func)(struct ifnet *, int))
1678 {
1679 	struct lagg_softc *sc = lp->lp_softc;
1680 	struct ifnet *scifp = sc->sc_ifp;
1681 	struct ifnet *ifp = lp->lp_ifp;
1682 	int error;
1683 
1684 	LAGG_WLOCK_ASSERT(sc);
1685 
1686 	status = status ? (scifp->if_flags & flag) : 0;
1687 	/* Now "status" contains the flag value or 0 */
1688 
1689 	/*
1690 	 * See if recorded ports status is different from what
1691 	 * we want it to be.  If it is, flip it.  We record ports
1692 	 * status in lp_ifflags so that we won't clear ports flag
1693 	 * we haven't set.  In fact, we don't clear or set ports
1694 	 * flags directly, but get or release references to them.
1695 	 * That's why we can be sure that recorded flags still are
1696 	 * in accord with actual ports flags.
1697 	 */
1698 	if (status != (lp->lp_ifflags & flag)) {
1699 		error = (*func)(ifp, status);
1700 		if (error)
1701 			return (error);
1702 		lp->lp_ifflags &= ~flag;
1703 		lp->lp_ifflags |= status;
1704 	}
1705 	return (0);
1706 }
1707 
1708 /*
1709  * Handle IFF_* flags that require certain changes on the lagg port
1710  * if "status" is true, update ports flags respective to the lagg
1711  * if "status" is false, forcedly clear the flags set on port.
1712  */
1713 static int
1714 lagg_setflags(struct lagg_port *lp, int status)
1715 {
1716 	int error, i;
1717 
1718 	for (i = 0; lagg_pflags[i].flag; i++) {
1719 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1720 		    status, lagg_pflags[i].func);
1721 		if (error)
1722 			return (error);
1723 	}
1724 	return (0);
1725 }
1726 
1727 static int
1728 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1729 {
1730 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1731 	int error, len, mcast;
1732 	struct rm_priotracker tracker;
1733 
1734 	len = m->m_pkthdr.len;
1735 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1736 
1737 	LAGG_RLOCK(sc, &tracker);
1738 	/* We need a Tx algorithm and at least one port */
1739 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1740 		LAGG_RUNLOCK(sc, &tracker);
1741 		m_freem(m);
1742 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1743 		return (ENXIO);
1744 	}
1745 
1746 	ETHER_BPF_MTAP(ifp, m);
1747 
1748 	error = lagg_proto_start(sc, m);
1749 	LAGG_RUNLOCK(sc, &tracker);
1750 
1751 	if (error != 0)
1752 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1753 
1754 	return (error);
1755 }
1756 
1757 /*
1758  * The ifp->if_qflush entry point for lagg(4) is no-op.
1759  */
1760 static void
1761 lagg_qflush(struct ifnet *ifp __unused)
1762 {
1763 }
1764 
1765 static struct mbuf *
1766 lagg_input(struct ifnet *ifp, struct mbuf *m)
1767 {
1768 	struct lagg_port *lp = ifp->if_lagg;
1769 	struct lagg_softc *sc = lp->lp_softc;
1770 	struct ifnet *scifp = sc->sc_ifp;
1771 	struct rm_priotracker tracker;
1772 
1773 	LAGG_RLOCK(sc, &tracker);
1774 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1775 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1776 	    sc->sc_proto == LAGG_PROTO_NONE) {
1777 		LAGG_RUNLOCK(sc, &tracker);
1778 		m_freem(m);
1779 		return (NULL);
1780 	}
1781 
1782 	ETHER_BPF_MTAP(scifp, m);
1783 
1784 	if (lp->lp_detaching != 0) {
1785 		m_freem(m);
1786 		m = NULL;
1787 	} else
1788 		m = lagg_proto_input(sc, lp, m);
1789 
1790 	if (m != NULL) {
1791 		if (scifp->if_flags & IFF_MONITOR) {
1792 			m_freem(m);
1793 			m = NULL;
1794 		}
1795 	}
1796 
1797 	LAGG_RUNLOCK(sc, &tracker);
1798 	return (m);
1799 }
1800 
1801 static int
1802 lagg_media_change(struct ifnet *ifp)
1803 {
1804 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1805 
1806 	if (sc->sc_ifflags & IFF_DEBUG)
1807 		printf("%s\n", __func__);
1808 
1809 	/* Ignore */
1810 	return (0);
1811 }
1812 
1813 static void
1814 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1815 {
1816 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1817 	struct lagg_port *lp;
1818 	struct rm_priotracker tracker;
1819 
1820 	imr->ifm_status = IFM_AVALID;
1821 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1822 
1823 	LAGG_RLOCK(sc, &tracker);
1824 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1825 		if (LAGG_PORTACTIVE(lp))
1826 			imr->ifm_status |= IFM_ACTIVE;
1827 	}
1828 	LAGG_RUNLOCK(sc, &tracker);
1829 }
1830 
1831 static void
1832 lagg_linkstate(struct lagg_softc *sc)
1833 {
1834 	struct lagg_port *lp;
1835 	int new_link = LINK_STATE_DOWN;
1836 	uint64_t speed;
1837 
1838 	/* Our link is considered up if at least one of our ports is active */
1839 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1840 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1841 			new_link = LINK_STATE_UP;
1842 			break;
1843 		}
1844 	}
1845 	if_link_state_change(sc->sc_ifp, new_link);
1846 
1847 	/* Update if_baudrate to reflect the max possible speed */
1848 	switch (sc->sc_proto) {
1849 		case LAGG_PROTO_FAILOVER:
1850 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1851 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1852 			break;
1853 		case LAGG_PROTO_ROUNDROBIN:
1854 		case LAGG_PROTO_LOADBALANCE:
1855 		case LAGG_PROTO_BROADCAST:
1856 			speed = 0;
1857 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1858 				speed += lp->lp_ifp->if_baudrate;
1859 			sc->sc_ifp->if_baudrate = speed;
1860 			break;
1861 		case LAGG_PROTO_LACP:
1862 			/* LACP updates if_baudrate itself */
1863 			break;
1864 	}
1865 }
1866 
1867 static void
1868 lagg_port_state(struct ifnet *ifp, int state)
1869 {
1870 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1871 	struct lagg_softc *sc = NULL;
1872 
1873 	if (lp != NULL)
1874 		sc = lp->lp_softc;
1875 	if (sc == NULL)
1876 		return;
1877 
1878 	LAGG_WLOCK(sc);
1879 	lagg_linkstate(sc);
1880 	lagg_proto_linkstate(sc, lp);
1881 	LAGG_WUNLOCK(sc);
1882 }
1883 
1884 struct lagg_port *
1885 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1886 {
1887 	struct lagg_port *lp_next, *rval = NULL;
1888 	// int new_link = LINK_STATE_DOWN;
1889 
1890 	LAGG_RLOCK_ASSERT(sc);
1891 	/*
1892 	 * Search a port which reports an active link state.
1893 	 */
1894 
1895 	if (lp == NULL)
1896 		goto search;
1897 	if (LAGG_PORTACTIVE(lp)) {
1898 		rval = lp;
1899 		goto found;
1900 	}
1901 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1902 	    LAGG_PORTACTIVE(lp_next)) {
1903 		rval = lp_next;
1904 		goto found;
1905 	}
1906 
1907 search:
1908 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1909 		if (LAGG_PORTACTIVE(lp_next)) {
1910 			rval = lp_next;
1911 			goto found;
1912 		}
1913 	}
1914 
1915 found:
1916 	if (rval != NULL) {
1917 		/*
1918 		 * The IEEE 802.1D standard assumes that a lagg with
1919 		 * multiple ports is always full duplex. This is valid
1920 		 * for load sharing laggs and if at least two links
1921 		 * are active. Unfortunately, checking the latter would
1922 		 * be too expensive at this point.
1923 		 XXX
1924 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1925 		    (sc->sc_count > 1))
1926 			new_link = LINK_STATE_FULL_DUPLEX;
1927 		else
1928 			new_link = rval->lp_link_state;
1929 		 */
1930 	}
1931 
1932 	return (rval);
1933 }
1934 
1935 int
1936 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1937 {
1938 
1939 	return (ifp->if_transmit)(ifp, m);
1940 }
1941 
1942 /*
1943  * Simple round robin aggregation
1944  */
1945 static void
1946 lagg_rr_attach(struct lagg_softc *sc)
1947 {
1948 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1949 	sc->sc_seq = 0;
1950 	sc->sc_bkt_count = sc->sc_bkt;
1951 }
1952 
1953 static int
1954 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1955 {
1956 	struct lagg_port *lp;
1957 	uint32_t p;
1958 
1959 	if (sc->sc_bkt_count == 0 && sc->sc_bkt > 0)
1960 		sc->sc_bkt_count = sc->sc_bkt;
1961 
1962 	if (sc->sc_bkt > 0) {
1963 		atomic_subtract_int(&sc->sc_bkt_count, 1);
1964 	if (atomic_cmpset_int(&sc->sc_bkt_count, 0, sc->sc_bkt))
1965 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1966 	else
1967 		p = sc->sc_seq;
1968 	} else
1969 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1970 
1971 	p %= sc->sc_count;
1972 	lp = SLIST_FIRST(&sc->sc_ports);
1973 
1974 	while (p--)
1975 		lp = SLIST_NEXT(lp, lp_entries);
1976 
1977 	/*
1978 	 * Check the port's link state. This will return the next active
1979 	 * port if the link is down or the port is NULL.
1980 	 */
1981 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1982 		m_freem(m);
1983 		return (ENETDOWN);
1984 	}
1985 
1986 	/* Send mbuf */
1987 	return (lagg_enqueue(lp->lp_ifp, m));
1988 }
1989 
1990 static struct mbuf *
1991 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1992 {
1993 	struct ifnet *ifp = sc->sc_ifp;
1994 
1995 	/* Just pass in the packet to our lagg device */
1996 	m->m_pkthdr.rcvif = ifp;
1997 
1998 	return (m);
1999 }
2000 
2001 /*
2002  * Broadcast mode
2003  */
2004 static int
2005 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2006 {
2007 	int active_ports = 0;
2008 	int errors = 0;
2009 	int ret;
2010 	struct lagg_port *lp, *last = NULL;
2011 	struct mbuf *m0;
2012 
2013 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2014 		if (!LAGG_PORTACTIVE(lp))
2015 			continue;
2016 
2017 		active_ports++;
2018 
2019 		if (last != NULL) {
2020 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2021 			if (m0 == NULL) {
2022 				ret = ENOBUFS;
2023 				errors++;
2024 				break;
2025 			}
2026 
2027 			ret = lagg_enqueue(last->lp_ifp, m0);
2028 			if (ret != 0)
2029 				errors++;
2030 		}
2031 		last = lp;
2032 	}
2033 	if (last == NULL) {
2034 		m_freem(m);
2035 		return (ENOENT);
2036 	}
2037 	if ((last = lagg_link_active(sc, last)) == NULL) {
2038 		m_freem(m);
2039 		return (ENETDOWN);
2040 	}
2041 
2042 	ret = lagg_enqueue(last->lp_ifp, m);
2043 	if (ret != 0)
2044 		errors++;
2045 
2046 	if (errors == 0)
2047 		return (ret);
2048 
2049 	return (0);
2050 }
2051 
2052 static struct mbuf*
2053 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2054 {
2055 	struct ifnet *ifp = sc->sc_ifp;
2056 
2057 	/* Just pass in the packet to our lagg device */
2058 	m->m_pkthdr.rcvif = ifp;
2059 	return (m);
2060 }
2061 
2062 /*
2063  * Active failover
2064  */
2065 static int
2066 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2067 {
2068 	struct lagg_port *lp;
2069 
2070 	/* Use the master port if active or the next available port */
2071 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2072 		m_freem(m);
2073 		return (ENETDOWN);
2074 	}
2075 
2076 	/* Send mbuf */
2077 	return (lagg_enqueue(lp->lp_ifp, m));
2078 }
2079 
2080 static struct mbuf *
2081 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2082 {
2083 	struct ifnet *ifp = sc->sc_ifp;
2084 	struct lagg_port *tmp_tp;
2085 
2086 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2087 		m->m_pkthdr.rcvif = ifp;
2088 		return (m);
2089 	}
2090 
2091 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2092 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2093 		/*
2094 		 * If tmp_tp is null, we've received a packet when all
2095 		 * our links are down. Weird, but process it anyways.
2096 		 */
2097 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2098 			m->m_pkthdr.rcvif = ifp;
2099 			return (m);
2100 		}
2101 	}
2102 
2103 	m_freem(m);
2104 	return (NULL);
2105 }
2106 
2107 /*
2108  * Loadbalancing
2109  */
2110 static void
2111 lagg_lb_attach(struct lagg_softc *sc)
2112 {
2113 	struct lagg_port *lp;
2114 	struct lagg_lb *lb;
2115 
2116 	lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
2117 
2118 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
2119 
2120 	lb->lb_key = m_ether_tcpip_hash_init();
2121 	sc->sc_psc = lb;
2122 
2123 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2124 		lagg_lb_port_create(lp);
2125 }
2126 
2127 static void
2128 lagg_lb_detach(struct lagg_softc *sc)
2129 {
2130 	struct lagg_lb *lb;
2131 
2132 	lb = (struct lagg_lb *)sc->sc_psc;
2133 	LAGG_WUNLOCK(sc);
2134 	if (lb != NULL)
2135 		free(lb, M_DEVBUF);
2136 }
2137 
2138 static int
2139 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2140 {
2141 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2142 	struct lagg_port *lp_next;
2143 	int i = 0;
2144 
2145 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2146 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2147 		if (lp_next == lp)
2148 			continue;
2149 		if (i >= LAGG_MAX_PORTS)
2150 			return (EINVAL);
2151 		if (sc->sc_ifflags & IFF_DEBUG)
2152 			printf("%s: port %s at index %d\n",
2153 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2154 		lb->lb_ports[i++] = lp_next;
2155 	}
2156 
2157 	return (0);
2158 }
2159 
2160 static int
2161 lagg_lb_port_create(struct lagg_port *lp)
2162 {
2163 	struct lagg_softc *sc = lp->lp_softc;
2164 	return (lagg_lb_porttable(sc, NULL));
2165 }
2166 
2167 static void
2168 lagg_lb_port_destroy(struct lagg_port *lp)
2169 {
2170 	struct lagg_softc *sc = lp->lp_softc;
2171 	lagg_lb_porttable(sc, lp);
2172 }
2173 
2174 static int
2175 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2176 {
2177 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2178 	struct lagg_port *lp = NULL;
2179 	uint32_t p = 0;
2180 
2181 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2182 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2183 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2184 	else
2185 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2186 	p %= sc->sc_count;
2187 	lp = lb->lb_ports[p];
2188 
2189 	/*
2190 	 * Check the port's link state. This will return the next active
2191 	 * port if the link is down or the port is NULL.
2192 	 */
2193 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2194 		m_freem(m);
2195 		return (ENETDOWN);
2196 	}
2197 
2198 	/* Send mbuf */
2199 	return (lagg_enqueue(lp->lp_ifp, m));
2200 }
2201 
2202 static struct mbuf *
2203 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2204 {
2205 	struct ifnet *ifp = sc->sc_ifp;
2206 
2207 	/* Just pass in the packet to our lagg device */
2208 	m->m_pkthdr.rcvif = ifp;
2209 
2210 	return (m);
2211 }
2212 
2213 /*
2214  * 802.3ad LACP
2215  */
2216 static void
2217 lagg_lacp_attach(struct lagg_softc *sc)
2218 {
2219 	struct lagg_port *lp;
2220 
2221 	lacp_attach(sc);
2222 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2223 		lacp_port_create(lp);
2224 }
2225 
2226 static void
2227 lagg_lacp_detach(struct lagg_softc *sc)
2228 {
2229 	struct lagg_port *lp;
2230 	void *psc;
2231 
2232 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2233 		lacp_port_destroy(lp);
2234 
2235 	psc = sc->sc_psc;
2236 	sc->sc_psc = NULL;
2237 	LAGG_WUNLOCK(sc);
2238 
2239 	lacp_detach(psc);
2240 }
2241 
2242 static void
2243 lagg_lacp_lladdr(struct lagg_softc *sc)
2244 {
2245 	struct lagg_port *lp;
2246 
2247 	/* purge all the lacp ports */
2248 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2249 		lacp_port_destroy(lp);
2250 
2251 	/* add them back in */
2252 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2253 		lacp_port_create(lp);
2254 }
2255 
2256 static int
2257 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2258 {
2259 	struct lagg_port *lp;
2260 
2261 	lp = lacp_select_tx_port(sc, m);
2262 	if (lp == NULL) {
2263 		m_freem(m);
2264 		return (ENETDOWN);
2265 	}
2266 
2267 	/* Send mbuf */
2268 	return (lagg_enqueue(lp->lp_ifp, m));
2269 }
2270 
2271 static struct mbuf *
2272 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2273 {
2274 	struct ifnet *ifp = sc->sc_ifp;
2275 	struct ether_header *eh;
2276 	u_short etype;
2277 
2278 	eh = mtod(m, struct ether_header *);
2279 	etype = ntohs(eh->ether_type);
2280 
2281 	/* Tap off LACP control messages */
2282 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2283 		m = lacp_input(lp, m);
2284 		if (m == NULL)
2285 			return (NULL);
2286 	}
2287 
2288 	/*
2289 	 * If the port is not collecting or not in the active aggregator then
2290 	 * free and return.
2291 	 */
2292 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2293 		m_freem(m);
2294 		return (NULL);
2295 	}
2296 
2297 	m->m_pkthdr.rcvif = ifp;
2298 	return (m);
2299 }
2300 
2301