xref: /freebsd/sys/net/if_lagg.c (revision 8b25e8410533a6e69cceff910546b2dc485a5059)
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 		lagg_port_destroy(lp, 1);
555 	/* Unhook the aggregation protocol */
556 	lagg_proto_detach(sc);
557 	LAGG_UNLOCK_ASSERT(sc);
558 
559 	ifmedia_removeall(&sc->sc_media);
560 	ether_ifdetach(ifp);
561 	if_free(ifp);
562 
563 	LAGG_LIST_LOCK();
564 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
565 	LAGG_LIST_UNLOCK();
566 
567 	taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
568 	LAGG_LOCK_DESTROY(sc);
569 	free(sc, M_DEVBUF);
570 }
571 
572 /*
573  * Set link-layer address on the lagg interface itself.
574  *
575  * Set noinline to be dtrace-friendly
576  */
577 static __noinline void
578 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
579 {
580 	struct ifnet *ifp = sc->sc_ifp;
581 	struct lagg_port lp;
582 
583 	if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
584 		return;
585 
586 	LAGG_WLOCK_ASSERT(sc);
587 	/*
588 	 * Set the link layer address on the lagg interface.
589 	 * lagg_proto_lladdr() notifies the MAC change to
590 	 * the aggregation protocol.  iflladdr_event handler which
591 	 * may trigger gratuitous ARPs for INET will be handled in
592 	 * a taskqueue.
593 	 */
594 	bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
595 	lagg_proto_lladdr(sc);
596 
597 	/*
598 	 * Send notification request for lagg interface
599 	 * itself. Note that new lladdr is already set.
600 	 */
601 	bzero(&lp, sizeof(lp));
602 	lp.lp_ifp = sc->sc_ifp;
603 	lp.lp_softc = sc;
604 
605 	/* Do not request lladdr change */
606 	lagg_port_lladdr(&lp, lladdr, LAGG_LLQTYPE_VIRT);
607 }
608 
609 static void
610 lagg_capabilities(struct lagg_softc *sc)
611 {
612 	struct lagg_port *lp;
613 	int cap = ~0, ena = ~0;
614 	u_long hwa = ~0UL;
615 	struct ifnet_hw_tsomax hw_tsomax;
616 
617 	LAGG_WLOCK_ASSERT(sc);
618 
619 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
620 
621 	/* Get capabilities from the lagg ports */
622 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
623 		cap &= lp->lp_ifp->if_capabilities;
624 		ena &= lp->lp_ifp->if_capenable;
625 		hwa &= lp->lp_ifp->if_hwassist;
626 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
627 	}
628 	cap = (cap == ~0 ? 0 : cap);
629 	ena = (ena == ~0 ? 0 : ena);
630 	hwa = (hwa == ~0 ? 0 : hwa);
631 
632 	if (sc->sc_ifp->if_capabilities != cap ||
633 	    sc->sc_ifp->if_capenable != ena ||
634 	    sc->sc_ifp->if_hwassist != hwa ||
635 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
636 		sc->sc_ifp->if_capabilities = cap;
637 		sc->sc_ifp->if_capenable = ena;
638 		sc->sc_ifp->if_hwassist = hwa;
639 		getmicrotime(&sc->sc_ifp->if_lastchange);
640 
641 		if (sc->sc_ifflags & IFF_DEBUG)
642 			if_printf(sc->sc_ifp,
643 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
644 	}
645 }
646 
647 /*
648  * Enqueue interface lladdr notification.
649  * If request is already queued, it is updated.
650  * If setting lladdr is also desired, @do_change has to be set to 1.
651  *
652  * Set noinline to be dtrace-friendly
653  */
654 static __noinline void
655 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr, lagg_llqtype llq_type)
656 {
657 	struct lagg_softc *sc = lp->lp_softc;
658 	struct ifnet *ifp = lp->lp_ifp;
659 	struct lagg_llq *llq;
660 
661 	LAGG_WLOCK_ASSERT(sc);
662 
663 	/*
664 	 * Do not enqueue requests where lladdr is the same for
665 	 * "physical" interfaces (e.g. ports in lagg)
666 	 */
667 	if (llq_type == LAGG_LLQTYPE_PHYS &&
668 	    memcmp(IF_LLADDR(ifp), lladdr, ETHER_ADDR_LEN) == 0)
669 		return;
670 
671 	/* Check to make sure its not already queued to be changed */
672 	SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
673 		if (llq->llq_ifp == ifp) {
674 			/* Update lladdr, it may have changed */
675 			bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
676 			return;
677 		}
678 	}
679 
680 	llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT | M_ZERO);
681 	if (llq == NULL)	/* XXX what to do */
682 		return;
683 
684 	llq->llq_ifp = ifp;
685 	llq->llq_type = llq_type;
686 	bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
687 	/* XXX: We should insert to tail */
688 	SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
689 
690 	taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
691 }
692 
693 /*
694  * Set the interface MAC address from a taskqueue to avoid a LOR.
695  *
696  * Set noinline to be dtrace-friendly
697  */
698 static __noinline void
699 lagg_port_setlladdr(void *arg, int pending)
700 {
701 	struct lagg_softc *sc = (struct lagg_softc *)arg;
702 	struct lagg_llq *llq, *head;
703 	struct ifnet *ifp;
704 
705 	/* Grab a local reference of the queue and remove it from the softc */
706 	LAGG_WLOCK(sc);
707 	head = SLIST_FIRST(&sc->sc_llq_head);
708 	SLIST_FIRST(&sc->sc_llq_head) = NULL;
709 	LAGG_WUNLOCK(sc);
710 
711 	/*
712 	 * Traverse the queue and set the lladdr on each ifp. It is safe to do
713 	 * unlocked as we have the only reference to it.
714 	 */
715 	for (llq = head; llq != NULL; llq = head) {
716 		ifp = llq->llq_ifp;
717 
718 		CURVNET_SET(ifp->if_vnet);
719 
720 		/*
721 		 * Set the link layer address on the laggport interface.
722 		 * Note that if_setlladdr() or iflladdr_event handler
723 		 * may result in arp transmission / lltable updates.
724 		 */
725 		if (llq->llq_type == LAGG_LLQTYPE_PHYS)
726 			if_setlladdr(ifp, llq->llq_lladdr,
727 			    ETHER_ADDR_LEN);
728 		else
729 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
730 		CURVNET_RESTORE();
731 		head = SLIST_NEXT(llq, llq_entries);
732 		free(llq, M_DEVBUF);
733 	}
734 }
735 
736 static int
737 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
738 {
739 	struct lagg_softc *sc_ptr;
740 	struct lagg_port *lp, *tlp;
741 	int error, i;
742 	uint64_t *pval;
743 
744 	LAGG_WLOCK_ASSERT(sc);
745 
746 	/* Limit the maximal number of lagg ports */
747 	if (sc->sc_count >= LAGG_MAX_PORTS)
748 		return (ENOSPC);
749 
750 	/* Check if port has already been associated to a lagg */
751 	if (ifp->if_lagg != NULL) {
752 		/* Port is already in the current lagg? */
753 		lp = (struct lagg_port *)ifp->if_lagg;
754 		if (lp->lp_softc == sc)
755 			return (EEXIST);
756 		return (EBUSY);
757 	}
758 
759 	/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
760 	if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
761 		return (EPROTONOSUPPORT);
762 
763 	/* Allow the first Ethernet member to define the MTU */
764 	if (SLIST_EMPTY(&sc->sc_ports))
765 		sc->sc_ifp->if_mtu = ifp->if_mtu;
766 	else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
767 		if_printf(sc->sc_ifp, "invalid MTU for %s\n",
768 		    ifp->if_xname);
769 		return (EINVAL);
770 	}
771 
772 	if ((lp = malloc(sizeof(struct lagg_port),
773 	    M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
774 		return (ENOMEM);
775 
776 	/* Check if port is a stacked lagg */
777 	LAGG_LIST_LOCK();
778 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
779 		if (ifp == sc_ptr->sc_ifp) {
780 			LAGG_LIST_UNLOCK();
781 			free(lp, M_DEVBUF);
782 			return (EINVAL);
783 			/* XXX disable stacking for the moment, its untested */
784 #ifdef LAGG_PORT_STACKING
785 			lp->lp_flags |= LAGG_PORT_STACK;
786 			if (lagg_port_checkstacking(sc_ptr) >=
787 			    LAGG_MAX_STACKING) {
788 				LAGG_LIST_UNLOCK();
789 				free(lp, M_DEVBUF);
790 				return (E2BIG);
791 			}
792 #endif
793 		}
794 	}
795 	LAGG_LIST_UNLOCK();
796 
797 	/* Change the interface type */
798 	lp->lp_iftype = ifp->if_type;
799 	ifp->if_type = IFT_IEEE8023ADLAG;
800 	ifp->if_lagg = lp;
801 	lp->lp_ioctl = ifp->if_ioctl;
802 	ifp->if_ioctl = lagg_port_ioctl;
803 	lp->lp_output = ifp->if_output;
804 	ifp->if_output = lagg_port_output;
805 
806 	lp->lp_ifp = ifp;
807 	lp->lp_softc = sc;
808 
809 	/* Save port link layer address */
810 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
811 
812 	if (SLIST_EMPTY(&sc->sc_ports)) {
813 		sc->sc_primary = lp;
814 		/* First port in lagg. Update/notify lagg lladdress */
815 		lagg_lladdr(sc, IF_LLADDR(ifp));
816 	} else {
817 
818 		/*
819 		 * Update link layer address for this port and
820 		 * send notifications to other subsystems.
821 		 */
822 		lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp), LAGG_LLQTYPE_PHYS);
823 	}
824 
825 	/*
826 	 * Insert into the list of ports.
827 	 * Keep ports sorted by if_index. It is handy, when configuration
828 	 * is predictable and `ifconfig laggN create ...` command
829 	 * will lead to the same result each time.
830 	 */
831 	SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
832 		if (tlp->lp_ifp->if_index < ifp->if_index && (
833 		    SLIST_NEXT(tlp, lp_entries) == NULL ||
834 		    SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index >
835 		    ifp->if_index))
836 			break;
837 	}
838 	if (tlp != NULL)
839 		SLIST_INSERT_AFTER(tlp, lp, lp_entries);
840 	else
841 		SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
842 	sc->sc_count++;
843 
844 	/* Update lagg capabilities */
845 	lagg_capabilities(sc);
846 	lagg_linkstate(sc);
847 
848 	/* Read port counters */
849 	pval = lp->port_counters.val;
850 	for (i = 0; i < IFCOUNTERS; i++, pval++)
851 		*pval = ifp->if_get_counter(ifp, i);
852 	/* Add multicast addresses and interface flags to this port */
853 	lagg_ether_cmdmulti(lp, 1);
854 	lagg_setflags(lp, 1);
855 
856 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
857 		/* Remove the port, without calling pr_delport. */
858 		lagg_port_destroy(lp, 0);
859 		return (error);
860 	}
861 
862 	return (0);
863 }
864 
865 #ifdef LAGG_PORT_STACKING
866 static int
867 lagg_port_checkstacking(struct lagg_softc *sc)
868 {
869 	struct lagg_softc *sc_ptr;
870 	struct lagg_port *lp;
871 	int m = 0;
872 
873 	LAGG_WLOCK_ASSERT(sc);
874 
875 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
876 		if (lp->lp_flags & LAGG_PORT_STACK) {
877 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
878 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
879 		}
880 	}
881 
882 	return (m + 1);
883 }
884 #endif
885 
886 static int
887 lagg_port_destroy(struct lagg_port *lp, int rundelport)
888 {
889 	struct lagg_softc *sc = lp->lp_softc;
890 	struct lagg_port *lp_ptr, *lp0;
891 	struct lagg_llq *llq;
892 	struct ifnet *ifp = lp->lp_ifp;
893 	uint64_t *pval, vdiff;
894 	int i;
895 
896 	LAGG_WLOCK_ASSERT(sc);
897 
898 	if (rundelport)
899 		lagg_proto_delport(sc, lp);
900 
901 	/*
902 	 * Remove multicast addresses and interface flags from this port and
903 	 * reset the MAC address, skip if the interface is being detached.
904 	 */
905 	if (!lp->lp_detaching) {
906 		lagg_ether_cmdmulti(lp, 0);
907 		lagg_setflags(lp, 0);
908 		lagg_port_lladdr(lp, lp->lp_lladdr, LAGG_LLQTYPE_PHYS);
909 	}
910 
911 	/* Restore interface */
912 	ifp->if_type = lp->lp_iftype;
913 	ifp->if_ioctl = lp->lp_ioctl;
914 	ifp->if_output = lp->lp_output;
915 	ifp->if_lagg = NULL;
916 
917 	/* Update detached port counters */
918 	pval = lp->port_counters.val;
919 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
920 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
921 		sc->detached_counters.val[i] += vdiff;
922 	}
923 
924 	/* Finally, remove the port from the lagg */
925 	SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
926 	sc->sc_count--;
927 
928 	/* Update the primary interface */
929 	if (lp == sc->sc_primary) {
930 		uint8_t lladdr[ETHER_ADDR_LEN];
931 
932 		if ((lp0 = SLIST_FIRST(&sc->sc_ports)) == NULL) {
933 			bzero(&lladdr, ETHER_ADDR_LEN);
934 		} else {
935 			bcopy(lp0->lp_lladdr,
936 			    lladdr, ETHER_ADDR_LEN);
937 		}
938 		lagg_lladdr(sc, lladdr);
939 
940 		/* Mark lp0 as new primary */
941 		sc->sc_primary = lp0;
942 
943 		/*
944 		 * Enqueue lladdr update/notification for each port
945 		 * (new primary needs update as well, to switch from
946 		 * old lladdr to its 'real' one).
947 		 */
948 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
949 			lagg_port_lladdr(lp_ptr, lladdr, LAGG_LLQTYPE_PHYS);
950 	}
951 
952 	/* Remove any pending lladdr changes from the queue */
953 	if (lp->lp_detaching) {
954 		SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
955 			if (llq->llq_ifp == ifp) {
956 				SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
957 				    llq_entries);
958 				free(llq, M_DEVBUF);
959 				break;	/* Only appears once */
960 			}
961 		}
962 	}
963 
964 	if (lp->lp_ifflags)
965 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
966 
967 	free(lp, M_DEVBUF);
968 
969 	/* Update lagg capabilities */
970 	lagg_capabilities(sc);
971 	lagg_linkstate(sc);
972 
973 	return (0);
974 }
975 
976 static int
977 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
978 {
979 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
980 	struct lagg_softc *sc;
981 	struct lagg_port *lp = NULL;
982 	int error = 0;
983 	struct rm_priotracker tracker;
984 
985 	/* Should be checked by the caller */
986 	if (ifp->if_type != IFT_IEEE8023ADLAG ||
987 	    (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
988 		goto fallback;
989 
990 	switch (cmd) {
991 	case SIOCGLAGGPORT:
992 		if (rp->rp_portname[0] == '\0' ||
993 		    ifunit(rp->rp_portname) != ifp) {
994 			error = EINVAL;
995 			break;
996 		}
997 
998 		LAGG_RLOCK(sc, &tracker);
999 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1000 			error = ENOENT;
1001 			LAGG_RUNLOCK(sc, &tracker);
1002 			break;
1003 		}
1004 
1005 		lagg_port2req(lp, rp);
1006 		LAGG_RUNLOCK(sc, &tracker);
1007 		break;
1008 
1009 	case SIOCSIFCAP:
1010 		if (lp->lp_ioctl == NULL) {
1011 			error = EINVAL;
1012 			break;
1013 		}
1014 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1015 		if (error)
1016 			break;
1017 
1018 		/* Update lagg interface capabilities */
1019 		LAGG_WLOCK(sc);
1020 		lagg_capabilities(sc);
1021 		LAGG_WUNLOCK(sc);
1022 		break;
1023 
1024 	case SIOCSIFMTU:
1025 		/* Do not allow the MTU to be changed once joined */
1026 		error = EINVAL;
1027 		break;
1028 
1029 	default:
1030 		goto fallback;
1031 	}
1032 
1033 	return (error);
1034 
1035 fallback:
1036 	if (lp != NULL && lp->lp_ioctl != NULL)
1037 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1038 
1039 	return (EINVAL);
1040 }
1041 
1042 /*
1043  * Requests counter @cnt data.
1044  *
1045  * Counter value is calculated the following way:
1046  * 1) for each port, sum  difference between current and "initial" measurements.
1047  * 2) add lagg logical interface counters.
1048  * 3) add data from detached_counters array.
1049  *
1050  * We also do the following things on ports attach/detach:
1051  * 1) On port attach we store all counters it has into port_counter array.
1052  * 2) On port detach we add the different between "initial" and
1053  *   current counters data to detached_counters array.
1054  */
1055 static uint64_t
1056 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1057 {
1058 	struct lagg_softc *sc;
1059 	struct lagg_port *lp;
1060 	struct ifnet *lpifp;
1061 	struct rm_priotracker tracker;
1062 	uint64_t newval, oldval, vsum;
1063 
1064 	/* Revise this when we've got non-generic counters. */
1065 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1066 
1067 	sc = (struct lagg_softc *)ifp->if_softc;
1068 	LAGG_RLOCK(sc, &tracker);
1069 
1070 	vsum = 0;
1071 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1072 		/* Saved attached value */
1073 		oldval = lp->port_counters.val[cnt];
1074 		/* current value */
1075 		lpifp = lp->lp_ifp;
1076 		newval = lpifp->if_get_counter(lpifp, cnt);
1077 		/* Calculate diff and save new */
1078 		vsum += newval - oldval;
1079 	}
1080 
1081 	/*
1082 	 * Add counter data which might be added by upper
1083 	 * layer protocols operating on logical interface.
1084 	 */
1085 	vsum += if_get_counter_default(ifp, cnt);
1086 
1087 	/*
1088 	 * Add counter data from detached ports counters
1089 	 */
1090 	vsum += sc->detached_counters.val[cnt];
1091 
1092 	LAGG_RUNLOCK(sc, &tracker);
1093 
1094 	return (vsum);
1095 }
1096 
1097 /*
1098  * For direct output to child ports.
1099  */
1100 static int
1101 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1102 	const struct sockaddr *dst, struct route *ro)
1103 {
1104 	struct lagg_port *lp = ifp->if_lagg;
1105 
1106 	switch (dst->sa_family) {
1107 		case pseudo_AF_HDRCMPLT:
1108 		case AF_UNSPEC:
1109 			return ((*lp->lp_output)(ifp, m, dst, ro));
1110 	}
1111 
1112 	/* drop any other frames */
1113 	m_freem(m);
1114 	return (ENETDOWN);
1115 }
1116 
1117 static void
1118 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1119 {
1120 	struct lagg_port *lp;
1121 	struct lagg_softc *sc;
1122 
1123 	if ((lp = ifp->if_lagg) == NULL)
1124 		return;
1125 	/* If the ifnet is just being renamed, don't do anything. */
1126 	if (ifp->if_flags & IFF_RENAMING)
1127 		return;
1128 
1129 	sc = lp->lp_softc;
1130 
1131 	LAGG_WLOCK(sc);
1132 	lp->lp_detaching = 1;
1133 	lagg_port_destroy(lp, 1);
1134 	LAGG_WUNLOCK(sc);
1135 }
1136 
1137 static void
1138 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1139 {
1140 	struct lagg_softc *sc = lp->lp_softc;
1141 
1142 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1143 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1144 	rp->rp_prio = lp->lp_prio;
1145 	rp->rp_flags = lp->lp_flags;
1146 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1147 
1148 	/* Add protocol specific flags */
1149 	switch (sc->sc_proto) {
1150 		case LAGG_PROTO_FAILOVER:
1151 			if (lp == sc->sc_primary)
1152 				rp->rp_flags |= LAGG_PORT_MASTER;
1153 			if (lp == lagg_link_active(sc, sc->sc_primary))
1154 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1155 			break;
1156 
1157 		case LAGG_PROTO_ROUNDROBIN:
1158 		case LAGG_PROTO_LOADBALANCE:
1159 		case LAGG_PROTO_BROADCAST:
1160 			if (LAGG_PORTACTIVE(lp))
1161 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1162 			break;
1163 
1164 		case LAGG_PROTO_LACP:
1165 			/* LACP has a different definition of active */
1166 			if (lacp_isactive(lp))
1167 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1168 			if (lacp_iscollecting(lp))
1169 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1170 			if (lacp_isdistributing(lp))
1171 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1172 			break;
1173 	}
1174 
1175 }
1176 
1177 static void
1178 lagg_init(void *xsc)
1179 {
1180 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1181 	struct ifnet *ifp = sc->sc_ifp;
1182 	struct lagg_port *lp;
1183 
1184 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1185 		return;
1186 
1187 	LAGG_WLOCK(sc);
1188 
1189 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1190 
1191 	/*
1192 	 * Update the port lladdrs if needed.
1193 	 * This might be if_setlladdr() notification
1194 	 * that lladdr has been changed.
1195 	 */
1196 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1197 		lagg_port_lladdr(lp, IF_LLADDR(ifp), LAGG_LLQTYPE_PHYS);
1198 
1199 	lagg_proto_init(sc);
1200 
1201 	LAGG_WUNLOCK(sc);
1202 }
1203 
1204 static void
1205 lagg_stop(struct lagg_softc *sc)
1206 {
1207 	struct ifnet *ifp = sc->sc_ifp;
1208 
1209 	LAGG_WLOCK_ASSERT(sc);
1210 
1211 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1212 		return;
1213 
1214 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1215 
1216 	lagg_proto_stop(sc);
1217 }
1218 
1219 static int
1220 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1221 {
1222 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1223 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1224 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1225 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1226 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1227 	struct ifreq *ifr = (struct ifreq *)data;
1228 	struct lagg_port *lp;
1229 	struct ifnet *tpif;
1230 	struct thread *td = curthread;
1231 	char *buf, *outbuf;
1232 	int count, buflen, len, error = 0;
1233 	struct rm_priotracker tracker;
1234 
1235 	bzero(&rpbuf, sizeof(rpbuf));
1236 
1237 	switch (cmd) {
1238 	case SIOCGLAGG:
1239 		LAGG_RLOCK(sc, &tracker);
1240 		count = 0;
1241 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1242 			count++;
1243 		buflen = count * sizeof(struct lagg_reqport);
1244 		LAGG_RUNLOCK(sc, &tracker);
1245 
1246 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1247 
1248 		LAGG_RLOCK(sc, &tracker);
1249 		ra->ra_proto = sc->sc_proto;
1250 		lagg_proto_request(sc, &ra->ra_psc);
1251 		count = 0;
1252 		buf = outbuf;
1253 		len = min(ra->ra_size, buflen);
1254 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1255 			if (len < sizeof(rpbuf))
1256 				break;
1257 
1258 			lagg_port2req(lp, &rpbuf);
1259 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1260 			count++;
1261 			buf += sizeof(rpbuf);
1262 			len -= sizeof(rpbuf);
1263 		}
1264 		LAGG_RUNLOCK(sc, &tracker);
1265 		ra->ra_ports = count;
1266 		ra->ra_size = count * sizeof(rpbuf);
1267 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1268 		free(outbuf, M_TEMP);
1269 		break;
1270 	case SIOCSLAGG:
1271 		error = priv_check(td, PRIV_NET_LAGG);
1272 		if (error)
1273 			break;
1274 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1275 			error = EPROTONOSUPPORT;
1276 			break;
1277 		}
1278 
1279 		LAGG_WLOCK(sc);
1280 		lagg_proto_detach(sc);
1281 		LAGG_UNLOCK_ASSERT(sc);
1282 		lagg_proto_attach(sc, ra->ra_proto);
1283 		break;
1284 	case SIOCGLAGGOPTS:
1285 		ro->ro_opts = sc->sc_opts;
1286 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1287 			struct lacp_softc *lsc;
1288 
1289 			lsc = (struct lacp_softc *)sc->sc_psc;
1290 			if (lsc->lsc_debug.lsc_tx_test != 0)
1291 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1292 			if (lsc->lsc_debug.lsc_rx_test != 0)
1293 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1294 			if (lsc->lsc_strict_mode != 0)
1295 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1296 			if (lsc->lsc_fast_timeout != 0)
1297 				ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1298 
1299 			ro->ro_active = sc->sc_active;
1300 		} else {
1301 			ro->ro_active = 0;
1302 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1303 				ro->ro_active += LAGG_PORTACTIVE(lp);
1304 		}
1305 		ro->ro_bkt = sc->sc_bkt;
1306 		ro->ro_flapping = sc->sc_flapping;
1307 		ro->ro_flowid_shift = sc->flowid_shift;
1308 		break;
1309 	case SIOCSLAGGOPTS:
1310 		if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) {
1311 			if (ro->ro_bkt == 0)
1312 				sc->sc_bkt = 1; // Minimum 1 packet per iface.
1313 			else
1314 				sc->sc_bkt = ro->ro_bkt;
1315 		}
1316 		error = priv_check(td, PRIV_NET_LAGG);
1317 		if (error)
1318 			break;
1319 		if (ro->ro_opts == 0)
1320 			break;
1321 		/*
1322 		 * Set options.  LACP options are stored in sc->sc_psc,
1323 		 * not in sc_opts.
1324 		 */
1325 		int valid, lacp;
1326 
1327 		switch (ro->ro_opts) {
1328 		case LAGG_OPT_USE_FLOWID:
1329 		case -LAGG_OPT_USE_FLOWID:
1330 		case LAGG_OPT_FLOWIDSHIFT:
1331 			valid = 1;
1332 			lacp = 0;
1333 			break;
1334 		case LAGG_OPT_LACP_TXTEST:
1335 		case -LAGG_OPT_LACP_TXTEST:
1336 		case LAGG_OPT_LACP_RXTEST:
1337 		case -LAGG_OPT_LACP_RXTEST:
1338 		case LAGG_OPT_LACP_STRICT:
1339 		case -LAGG_OPT_LACP_STRICT:
1340 		case LAGG_OPT_LACP_TIMEOUT:
1341 		case -LAGG_OPT_LACP_TIMEOUT:
1342 			valid = lacp = 1;
1343 			break;
1344 		default:
1345 			valid = lacp = 0;
1346 			break;
1347 		}
1348 
1349 		LAGG_WLOCK(sc);
1350 
1351 		if (valid == 0 ||
1352 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1353 			/* Invalid combination of options specified. */
1354 			error = EINVAL;
1355 			LAGG_WUNLOCK(sc);
1356 			break;	/* Return from SIOCSLAGGOPTS. */
1357 		}
1358 		/*
1359 		 * Store new options into sc->sc_opts except for
1360 		 * FLOWIDSHIFT and LACP options.
1361 		 */
1362 		if (lacp == 0) {
1363 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1364 				sc->flowid_shift = ro->ro_flowid_shift;
1365 			else if (ro->ro_opts > 0)
1366 				sc->sc_opts |= ro->ro_opts;
1367 			else
1368 				sc->sc_opts &= ~ro->ro_opts;
1369 		} else {
1370 			struct lacp_softc *lsc;
1371 			struct lacp_port *lp;
1372 
1373 			lsc = (struct lacp_softc *)sc->sc_psc;
1374 
1375 			switch (ro->ro_opts) {
1376 			case LAGG_OPT_LACP_TXTEST:
1377 				lsc->lsc_debug.lsc_tx_test = 1;
1378 				break;
1379 			case -LAGG_OPT_LACP_TXTEST:
1380 				lsc->lsc_debug.lsc_tx_test = 0;
1381 				break;
1382 			case LAGG_OPT_LACP_RXTEST:
1383 				lsc->lsc_debug.lsc_rx_test = 1;
1384 				break;
1385 			case -LAGG_OPT_LACP_RXTEST:
1386 				lsc->lsc_debug.lsc_rx_test = 0;
1387 				break;
1388 			case LAGG_OPT_LACP_STRICT:
1389 				lsc->lsc_strict_mode = 1;
1390 				break;
1391 			case -LAGG_OPT_LACP_STRICT:
1392 				lsc->lsc_strict_mode = 0;
1393 				break;
1394 			case LAGG_OPT_LACP_TIMEOUT:
1395 				LACP_LOCK(lsc);
1396         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1397                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1398 				LACP_UNLOCK(lsc);
1399 				lsc->lsc_fast_timeout = 1;
1400 				break;
1401 			case -LAGG_OPT_LACP_TIMEOUT:
1402 				LACP_LOCK(lsc);
1403         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1404                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1405 				LACP_UNLOCK(lsc);
1406 				lsc->lsc_fast_timeout = 0;
1407 				break;
1408 			}
1409 		}
1410 		LAGG_WUNLOCK(sc);
1411 		break;
1412 	case SIOCGLAGGFLAGS:
1413 		rf->rf_flags = 0;
1414 		LAGG_RLOCK(sc, &tracker);
1415 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1416 			rf->rf_flags |= LAGG_F_HASHL2;
1417 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1418 			rf->rf_flags |= LAGG_F_HASHL3;
1419 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1420 			rf->rf_flags |= LAGG_F_HASHL4;
1421 		LAGG_RUNLOCK(sc, &tracker);
1422 		break;
1423 	case SIOCSLAGGHASH:
1424 		error = priv_check(td, PRIV_NET_LAGG);
1425 		if (error)
1426 			break;
1427 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1428 			error = EINVAL;
1429 			break;
1430 		}
1431 		LAGG_WLOCK(sc);
1432 		sc->sc_flags = 0;
1433 		if (rf->rf_flags & LAGG_F_HASHL2)
1434 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1435 		if (rf->rf_flags & LAGG_F_HASHL3)
1436 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1437 		if (rf->rf_flags & LAGG_F_HASHL4)
1438 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1439 		LAGG_WUNLOCK(sc);
1440 		break;
1441 	case SIOCGLAGGPORT:
1442 		if (rp->rp_portname[0] == '\0' ||
1443 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1444 			error = EINVAL;
1445 			break;
1446 		}
1447 
1448 		LAGG_RLOCK(sc, &tracker);
1449 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1450 		    lp->lp_softc != sc) {
1451 			error = ENOENT;
1452 			LAGG_RUNLOCK(sc, &tracker);
1453 			break;
1454 		}
1455 
1456 		lagg_port2req(lp, rp);
1457 		LAGG_RUNLOCK(sc, &tracker);
1458 		break;
1459 	case SIOCSLAGGPORT:
1460 		error = priv_check(td, PRIV_NET_LAGG);
1461 		if (error)
1462 			break;
1463 		if (rp->rp_portname[0] == '\0' ||
1464 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1465 			error = EINVAL;
1466 			break;
1467 		}
1468 #ifdef INET6
1469 		/*
1470 		 * A laggport interface should not have inet6 address
1471 		 * because two interfaces with a valid link-local
1472 		 * scope zone must not be merged in any form.  This
1473 		 * restriction is needed to prevent violation of
1474 		 * link-local scope zone.  Attempts to add a laggport
1475 		 * interface which has inet6 addresses triggers
1476 		 * removal of all inet6 addresses on the member
1477 		 * interface.
1478 		 */
1479 		if (in6ifa_llaonifp(tpif)) {
1480 			in6_ifdetach(tpif);
1481 				if_printf(sc->sc_ifp,
1482 				    "IPv6 addresses on %s have been removed "
1483 				    "before adding it as a member to prevent "
1484 				    "IPv6 address scope violation.\n",
1485 				    tpif->if_xname);
1486 		}
1487 #endif
1488 		LAGG_WLOCK(sc);
1489 		error = lagg_port_create(sc, tpif);
1490 		LAGG_WUNLOCK(sc);
1491 		break;
1492 	case SIOCSLAGGDELPORT:
1493 		error = priv_check(td, PRIV_NET_LAGG);
1494 		if (error)
1495 			break;
1496 		if (rp->rp_portname[0] == '\0' ||
1497 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
1498 			error = EINVAL;
1499 			break;
1500 		}
1501 
1502 		LAGG_WLOCK(sc);
1503 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1504 		    lp->lp_softc != sc) {
1505 			error = ENOENT;
1506 			LAGG_WUNLOCK(sc);
1507 			break;
1508 		}
1509 
1510 		error = lagg_port_destroy(lp, 1);
1511 		LAGG_WUNLOCK(sc);
1512 		break;
1513 	case SIOCSIFFLAGS:
1514 		/* Set flags on ports too */
1515 		LAGG_WLOCK(sc);
1516 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1517 			lagg_setflags(lp, 1);
1518 		}
1519 		LAGG_WUNLOCK(sc);
1520 
1521 		if (!(ifp->if_flags & IFF_UP) &&
1522 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1523 			/*
1524 			 * If interface is marked down and it is running,
1525 			 * then stop and disable it.
1526 			 */
1527 			LAGG_WLOCK(sc);
1528 			lagg_stop(sc);
1529 			LAGG_WUNLOCK(sc);
1530 		} else if ((ifp->if_flags & IFF_UP) &&
1531 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1532 			/*
1533 			 * If interface is marked up and it is stopped, then
1534 			 * start it.
1535 			 */
1536 			(*ifp->if_init)(sc);
1537 		}
1538 		break;
1539 	case SIOCADDMULTI:
1540 	case SIOCDELMULTI:
1541 		LAGG_WLOCK(sc);
1542 		error = lagg_ether_setmulti(sc);
1543 		LAGG_WUNLOCK(sc);
1544 		break;
1545 	case SIOCSIFMEDIA:
1546 	case SIOCGIFMEDIA:
1547 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1548 		break;
1549 
1550 	case SIOCSIFCAP:
1551 	case SIOCSIFMTU:
1552 		/* Do not allow the MTU or caps to be directly changed */
1553 		error = EINVAL;
1554 		break;
1555 
1556 	default:
1557 		error = ether_ioctl(ifp, cmd, data);
1558 		break;
1559 	}
1560 	return (error);
1561 }
1562 
1563 #ifdef RATELIMIT
1564 static int
1565 lagg_snd_tag_alloc(struct ifnet *ifp,
1566     union if_snd_tag_alloc_params *params,
1567     struct m_snd_tag **ppmt)
1568 {
1569 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1570 	struct lagg_port *lp;
1571 	struct lagg_lb *lb;
1572 	uint32_t p;
1573 
1574 	switch (sc->sc_proto) {
1575 	case LAGG_PROTO_FAILOVER:
1576 		lp = lagg_link_active(sc, sc->sc_primary);
1577 		break;
1578 	case LAGG_PROTO_LOADBALANCE:
1579 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1580 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1581 			return (EOPNOTSUPP);
1582 		p = params->hdr.flowid >> sc->flowid_shift;
1583 		p %= sc->sc_count;
1584 		lb = (struct lagg_lb *)sc->sc_psc;
1585 		lp = lb->lb_ports[p];
1586 		lp = lagg_link_active(sc, lp);
1587 		break;
1588 	case LAGG_PROTO_LACP:
1589 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1590 		    params->hdr.flowtype == M_HASHTYPE_NONE)
1591 			return (EOPNOTSUPP);
1592 		lp = lacp_select_tx_port_by_hash(sc, params->hdr.flowid);
1593 		break;
1594 	default:
1595 		return (EOPNOTSUPP);
1596 	}
1597 	if (lp == NULL)
1598 		return (EOPNOTSUPP);
1599 	ifp = lp->lp_ifp;
1600 	if (ifp == NULL || ifp->if_snd_tag_alloc == NULL ||
1601 	    (ifp->if_capenable & IFCAP_TXRTLMT) == 0)
1602 		return (EOPNOTSUPP);
1603 
1604 	/* forward allocation request */
1605 	return (ifp->if_snd_tag_alloc(ifp, params, ppmt));
1606 }
1607 #endif
1608 
1609 static int
1610 lagg_ether_setmulti(struct lagg_softc *sc)
1611 {
1612 	struct lagg_port *lp;
1613 
1614 	LAGG_WLOCK_ASSERT(sc);
1615 
1616 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1617 		/* First, remove any existing filter entries. */
1618 		lagg_ether_cmdmulti(lp, 0);
1619 		/* copy all addresses from the lagg interface to the port */
1620 		lagg_ether_cmdmulti(lp, 1);
1621 	}
1622 	return (0);
1623 }
1624 
1625 static int
1626 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1627 {
1628 	struct lagg_softc *sc = lp->lp_softc;
1629 	struct ifnet *ifp = lp->lp_ifp;
1630 	struct ifnet *scifp = sc->sc_ifp;
1631 	struct lagg_mc *mc;
1632 	struct ifmultiaddr *ifma;
1633 	int error;
1634 
1635 	LAGG_WLOCK_ASSERT(sc);
1636 
1637 	if (set) {
1638 		IF_ADDR_WLOCK(scifp);
1639 		TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1640 			if (ifma->ifma_addr->sa_family != AF_LINK)
1641 				continue;
1642 			mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1643 			if (mc == NULL) {
1644 				IF_ADDR_WUNLOCK(scifp);
1645 				return (ENOMEM);
1646 			}
1647 			bcopy(ifma->ifma_addr, &mc->mc_addr,
1648 			    ifma->ifma_addr->sa_len);
1649 			mc->mc_addr.sdl_index = ifp->if_index;
1650 			mc->mc_ifma = NULL;
1651 			SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1652 		}
1653 		IF_ADDR_WUNLOCK(scifp);
1654 		SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1655 			error = if_addmulti(ifp,
1656 			    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1657 			if (error)
1658 				return (error);
1659 		}
1660 	} else {
1661 		while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1662 			SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1663 			if (mc->mc_ifma && !lp->lp_detaching)
1664 				if_delmulti_ifma(mc->mc_ifma);
1665 			free(mc, M_DEVBUF);
1666 		}
1667 	}
1668 	return (0);
1669 }
1670 
1671 /* Handle a ref counted flag that should be set on the lagg port as well */
1672 static int
1673 lagg_setflag(struct lagg_port *lp, int flag, int status,
1674     int (*func)(struct ifnet *, int))
1675 {
1676 	struct lagg_softc *sc = lp->lp_softc;
1677 	struct ifnet *scifp = sc->sc_ifp;
1678 	struct ifnet *ifp = lp->lp_ifp;
1679 	int error;
1680 
1681 	LAGG_WLOCK_ASSERT(sc);
1682 
1683 	status = status ? (scifp->if_flags & flag) : 0;
1684 	/* Now "status" contains the flag value or 0 */
1685 
1686 	/*
1687 	 * See if recorded ports status is different from what
1688 	 * we want it to be.  If it is, flip it.  We record ports
1689 	 * status in lp_ifflags so that we won't clear ports flag
1690 	 * we haven't set.  In fact, we don't clear or set ports
1691 	 * flags directly, but get or release references to them.
1692 	 * That's why we can be sure that recorded flags still are
1693 	 * in accord with actual ports flags.
1694 	 */
1695 	if (status != (lp->lp_ifflags & flag)) {
1696 		error = (*func)(ifp, status);
1697 		if (error)
1698 			return (error);
1699 		lp->lp_ifflags &= ~flag;
1700 		lp->lp_ifflags |= status;
1701 	}
1702 	return (0);
1703 }
1704 
1705 /*
1706  * Handle IFF_* flags that require certain changes on the lagg port
1707  * if "status" is true, update ports flags respective to the lagg
1708  * if "status" is false, forcedly clear the flags set on port.
1709  */
1710 static int
1711 lagg_setflags(struct lagg_port *lp, int status)
1712 {
1713 	int error, i;
1714 
1715 	for (i = 0; lagg_pflags[i].flag; i++) {
1716 		error = lagg_setflag(lp, lagg_pflags[i].flag,
1717 		    status, lagg_pflags[i].func);
1718 		if (error)
1719 			return (error);
1720 	}
1721 	return (0);
1722 }
1723 
1724 static int
1725 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1726 {
1727 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1728 	int error, len, mcast;
1729 	struct rm_priotracker tracker;
1730 
1731 	len = m->m_pkthdr.len;
1732 	mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1733 
1734 	LAGG_RLOCK(sc, &tracker);
1735 	/* We need a Tx algorithm and at least one port */
1736 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1737 		LAGG_RUNLOCK(sc, &tracker);
1738 		m_freem(m);
1739 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1740 		return (ENXIO);
1741 	}
1742 
1743 	ETHER_BPF_MTAP(ifp, m);
1744 
1745 	error = lagg_proto_start(sc, m);
1746 	LAGG_RUNLOCK(sc, &tracker);
1747 
1748 	if (error != 0)
1749 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1750 
1751 	return (error);
1752 }
1753 
1754 /*
1755  * The ifp->if_qflush entry point for lagg(4) is no-op.
1756  */
1757 static void
1758 lagg_qflush(struct ifnet *ifp __unused)
1759 {
1760 }
1761 
1762 static struct mbuf *
1763 lagg_input(struct ifnet *ifp, struct mbuf *m)
1764 {
1765 	struct lagg_port *lp = ifp->if_lagg;
1766 	struct lagg_softc *sc = lp->lp_softc;
1767 	struct ifnet *scifp = sc->sc_ifp;
1768 	struct rm_priotracker tracker;
1769 
1770 	LAGG_RLOCK(sc, &tracker);
1771 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1772 	    (lp->lp_flags & LAGG_PORT_DISABLED) ||
1773 	    sc->sc_proto == LAGG_PROTO_NONE) {
1774 		LAGG_RUNLOCK(sc, &tracker);
1775 		m_freem(m);
1776 		return (NULL);
1777 	}
1778 
1779 	ETHER_BPF_MTAP(scifp, m);
1780 
1781 	if (lp->lp_detaching != 0) {
1782 		m_freem(m);
1783 		m = NULL;
1784 	} else
1785 		m = lagg_proto_input(sc, lp, m);
1786 
1787 	if (m != NULL) {
1788 		if (scifp->if_flags & IFF_MONITOR) {
1789 			m_freem(m);
1790 			m = NULL;
1791 		}
1792 	}
1793 
1794 	LAGG_RUNLOCK(sc, &tracker);
1795 	return (m);
1796 }
1797 
1798 static int
1799 lagg_media_change(struct ifnet *ifp)
1800 {
1801 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1802 
1803 	if (sc->sc_ifflags & IFF_DEBUG)
1804 		printf("%s\n", __func__);
1805 
1806 	/* Ignore */
1807 	return (0);
1808 }
1809 
1810 static void
1811 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1812 {
1813 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1814 	struct lagg_port *lp;
1815 	struct rm_priotracker tracker;
1816 
1817 	imr->ifm_status = IFM_AVALID;
1818 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
1819 
1820 	LAGG_RLOCK(sc, &tracker);
1821 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1822 		if (LAGG_PORTACTIVE(lp))
1823 			imr->ifm_status |= IFM_ACTIVE;
1824 	}
1825 	LAGG_RUNLOCK(sc, &tracker);
1826 }
1827 
1828 static void
1829 lagg_linkstate(struct lagg_softc *sc)
1830 {
1831 	struct lagg_port *lp;
1832 	int new_link = LINK_STATE_DOWN;
1833 	uint64_t speed;
1834 
1835 	/* Our link is considered up if at least one of our ports is active */
1836 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1837 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1838 			new_link = LINK_STATE_UP;
1839 			break;
1840 		}
1841 	}
1842 	if_link_state_change(sc->sc_ifp, new_link);
1843 
1844 	/* Update if_baudrate to reflect the max possible speed */
1845 	switch (sc->sc_proto) {
1846 		case LAGG_PROTO_FAILOVER:
1847 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1848 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
1849 			break;
1850 		case LAGG_PROTO_ROUNDROBIN:
1851 		case LAGG_PROTO_LOADBALANCE:
1852 		case LAGG_PROTO_BROADCAST:
1853 			speed = 0;
1854 			SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1855 				speed += lp->lp_ifp->if_baudrate;
1856 			sc->sc_ifp->if_baudrate = speed;
1857 			break;
1858 		case LAGG_PROTO_LACP:
1859 			/* LACP updates if_baudrate itself */
1860 			break;
1861 	}
1862 }
1863 
1864 static void
1865 lagg_port_state(struct ifnet *ifp, int state)
1866 {
1867 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1868 	struct lagg_softc *sc = NULL;
1869 
1870 	if (lp != NULL)
1871 		sc = lp->lp_softc;
1872 	if (sc == NULL)
1873 		return;
1874 
1875 	LAGG_WLOCK(sc);
1876 	lagg_linkstate(sc);
1877 	lagg_proto_linkstate(sc, lp);
1878 	LAGG_WUNLOCK(sc);
1879 }
1880 
1881 struct lagg_port *
1882 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1883 {
1884 	struct lagg_port *lp_next, *rval = NULL;
1885 	// int new_link = LINK_STATE_DOWN;
1886 
1887 	LAGG_RLOCK_ASSERT(sc);
1888 	/*
1889 	 * Search a port which reports an active link state.
1890 	 */
1891 
1892 	if (lp == NULL)
1893 		goto search;
1894 	if (LAGG_PORTACTIVE(lp)) {
1895 		rval = lp;
1896 		goto found;
1897 	}
1898 	if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1899 	    LAGG_PORTACTIVE(lp_next)) {
1900 		rval = lp_next;
1901 		goto found;
1902 	}
1903 
1904 search:
1905 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1906 		if (LAGG_PORTACTIVE(lp_next)) {
1907 			rval = lp_next;
1908 			goto found;
1909 		}
1910 	}
1911 
1912 found:
1913 	if (rval != NULL) {
1914 		/*
1915 		 * The IEEE 802.1D standard assumes that a lagg with
1916 		 * multiple ports is always full duplex. This is valid
1917 		 * for load sharing laggs and if at least two links
1918 		 * are active. Unfortunately, checking the latter would
1919 		 * be too expensive at this point.
1920 		 XXX
1921 		if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1922 		    (sc->sc_count > 1))
1923 			new_link = LINK_STATE_FULL_DUPLEX;
1924 		else
1925 			new_link = rval->lp_link_state;
1926 		 */
1927 	}
1928 
1929 	return (rval);
1930 }
1931 
1932 int
1933 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1934 {
1935 
1936 	return (ifp->if_transmit)(ifp, m);
1937 }
1938 
1939 /*
1940  * Simple round robin aggregation
1941  */
1942 static void
1943 lagg_rr_attach(struct lagg_softc *sc)
1944 {
1945 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1946 	sc->sc_seq = 0;
1947 	sc->sc_bkt_count = sc->sc_bkt;
1948 }
1949 
1950 static int
1951 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1952 {
1953 	struct lagg_port *lp;
1954 	uint32_t p;
1955 
1956 	if (sc->sc_bkt_count == 0 && sc->sc_bkt > 0)
1957 		sc->sc_bkt_count = sc->sc_bkt;
1958 
1959 	if (sc->sc_bkt > 0) {
1960 		atomic_subtract_int(&sc->sc_bkt_count, 1);
1961 	if (atomic_cmpset_int(&sc->sc_bkt_count, 0, sc->sc_bkt))
1962 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1963 	else
1964 		p = sc->sc_seq;
1965 	} else
1966 		p = atomic_fetchadd_32(&sc->sc_seq, 1);
1967 
1968 	p %= sc->sc_count;
1969 	lp = SLIST_FIRST(&sc->sc_ports);
1970 
1971 	while (p--)
1972 		lp = SLIST_NEXT(lp, lp_entries);
1973 
1974 	/*
1975 	 * Check the port's link state. This will return the next active
1976 	 * port if the link is down or the port is NULL.
1977 	 */
1978 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
1979 		m_freem(m);
1980 		return (ENETDOWN);
1981 	}
1982 
1983 	/* Send mbuf */
1984 	return (lagg_enqueue(lp->lp_ifp, m));
1985 }
1986 
1987 static struct mbuf *
1988 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1989 {
1990 	struct ifnet *ifp = sc->sc_ifp;
1991 
1992 	/* Just pass in the packet to our lagg device */
1993 	m->m_pkthdr.rcvif = ifp;
1994 
1995 	return (m);
1996 }
1997 
1998 /*
1999  * Broadcast mode
2000  */
2001 static int
2002 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2003 {
2004 	int active_ports = 0;
2005 	int errors = 0;
2006 	int ret;
2007 	struct lagg_port *lp, *last = NULL;
2008 	struct mbuf *m0;
2009 
2010 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2011 		if (!LAGG_PORTACTIVE(lp))
2012 			continue;
2013 
2014 		active_ports++;
2015 
2016 		if (last != NULL) {
2017 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2018 			if (m0 == NULL) {
2019 				ret = ENOBUFS;
2020 				errors++;
2021 				break;
2022 			}
2023 
2024 			ret = lagg_enqueue(last->lp_ifp, m0);
2025 			if (ret != 0)
2026 				errors++;
2027 		}
2028 		last = lp;
2029 	}
2030 	if (last == NULL) {
2031 		m_freem(m);
2032 		return (ENOENT);
2033 	}
2034 	if ((last = lagg_link_active(sc, last)) == NULL) {
2035 		m_freem(m);
2036 		return (ENETDOWN);
2037 	}
2038 
2039 	ret = lagg_enqueue(last->lp_ifp, m);
2040 	if (ret != 0)
2041 		errors++;
2042 
2043 	if (errors == 0)
2044 		return (ret);
2045 
2046 	return (0);
2047 }
2048 
2049 static struct mbuf*
2050 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2051 {
2052 	struct ifnet *ifp = sc->sc_ifp;
2053 
2054 	/* Just pass in the packet to our lagg device */
2055 	m->m_pkthdr.rcvif = ifp;
2056 	return (m);
2057 }
2058 
2059 /*
2060  * Active failover
2061  */
2062 static int
2063 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2064 {
2065 	struct lagg_port *lp;
2066 
2067 	/* Use the master port if active or the next available port */
2068 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2069 		m_freem(m);
2070 		return (ENETDOWN);
2071 	}
2072 
2073 	/* Send mbuf */
2074 	return (lagg_enqueue(lp->lp_ifp, m));
2075 }
2076 
2077 static struct mbuf *
2078 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2079 {
2080 	struct ifnet *ifp = sc->sc_ifp;
2081 	struct lagg_port *tmp_tp;
2082 
2083 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2084 		m->m_pkthdr.rcvif = ifp;
2085 		return (m);
2086 	}
2087 
2088 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2089 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2090 		/*
2091 		 * If tmp_tp is null, we've received a packet when all
2092 		 * our links are down. Weird, but process it anyways.
2093 		 */
2094 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2095 			m->m_pkthdr.rcvif = ifp;
2096 			return (m);
2097 		}
2098 	}
2099 
2100 	m_freem(m);
2101 	return (NULL);
2102 }
2103 
2104 /*
2105  * Loadbalancing
2106  */
2107 static void
2108 lagg_lb_attach(struct lagg_softc *sc)
2109 {
2110 	struct lagg_port *lp;
2111 	struct lagg_lb *lb;
2112 
2113 	lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
2114 
2115 	sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
2116 
2117 	lb->lb_key = m_ether_tcpip_hash_init();
2118 	sc->sc_psc = lb;
2119 
2120 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2121 		lagg_lb_port_create(lp);
2122 }
2123 
2124 static void
2125 lagg_lb_detach(struct lagg_softc *sc)
2126 {
2127 	struct lagg_lb *lb;
2128 
2129 	lb = (struct lagg_lb *)sc->sc_psc;
2130 	LAGG_WUNLOCK(sc);
2131 	if (lb != NULL)
2132 		free(lb, M_DEVBUF);
2133 }
2134 
2135 static int
2136 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2137 {
2138 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2139 	struct lagg_port *lp_next;
2140 	int i = 0;
2141 
2142 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2143 	SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2144 		if (lp_next == lp)
2145 			continue;
2146 		if (i >= LAGG_MAX_PORTS)
2147 			return (EINVAL);
2148 		if (sc->sc_ifflags & IFF_DEBUG)
2149 			printf("%s: port %s at index %d\n",
2150 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2151 		lb->lb_ports[i++] = lp_next;
2152 	}
2153 
2154 	return (0);
2155 }
2156 
2157 static int
2158 lagg_lb_port_create(struct lagg_port *lp)
2159 {
2160 	struct lagg_softc *sc = lp->lp_softc;
2161 	return (lagg_lb_porttable(sc, NULL));
2162 }
2163 
2164 static void
2165 lagg_lb_port_destroy(struct lagg_port *lp)
2166 {
2167 	struct lagg_softc *sc = lp->lp_softc;
2168 	lagg_lb_porttable(sc, lp);
2169 }
2170 
2171 static int
2172 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2173 {
2174 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2175 	struct lagg_port *lp = NULL;
2176 	uint32_t p = 0;
2177 
2178 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2179 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2180 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2181 	else
2182 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2183 	p %= sc->sc_count;
2184 	lp = lb->lb_ports[p];
2185 
2186 	/*
2187 	 * Check the port's link state. This will return the next active
2188 	 * port if the link is down or the port is NULL.
2189 	 */
2190 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2191 		m_freem(m);
2192 		return (ENETDOWN);
2193 	}
2194 
2195 	/* Send mbuf */
2196 	return (lagg_enqueue(lp->lp_ifp, m));
2197 }
2198 
2199 static struct mbuf *
2200 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2201 {
2202 	struct ifnet *ifp = sc->sc_ifp;
2203 
2204 	/* Just pass in the packet to our lagg device */
2205 	m->m_pkthdr.rcvif = ifp;
2206 
2207 	return (m);
2208 }
2209 
2210 /*
2211  * 802.3ad LACP
2212  */
2213 static void
2214 lagg_lacp_attach(struct lagg_softc *sc)
2215 {
2216 	struct lagg_port *lp;
2217 
2218 	lacp_attach(sc);
2219 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2220 		lacp_port_create(lp);
2221 }
2222 
2223 static void
2224 lagg_lacp_detach(struct lagg_softc *sc)
2225 {
2226 	struct lagg_port *lp;
2227 	void *psc;
2228 
2229 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2230 		lacp_port_destroy(lp);
2231 
2232 	psc = sc->sc_psc;
2233 	sc->sc_psc = NULL;
2234 	LAGG_WUNLOCK(sc);
2235 
2236 	lacp_detach(psc);
2237 }
2238 
2239 static void
2240 lagg_lacp_lladdr(struct lagg_softc *sc)
2241 {
2242 	struct lagg_port *lp;
2243 
2244 	/* purge all the lacp ports */
2245 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2246 		lacp_port_destroy(lp);
2247 
2248 	/* add them back in */
2249 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2250 		lacp_port_create(lp);
2251 }
2252 
2253 static int
2254 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2255 {
2256 	struct lagg_port *lp;
2257 
2258 	lp = lacp_select_tx_port(sc, m);
2259 	if (lp == NULL) {
2260 		m_freem(m);
2261 		return (ENETDOWN);
2262 	}
2263 
2264 	/* Send mbuf */
2265 	return (lagg_enqueue(lp->lp_ifp, m));
2266 }
2267 
2268 static struct mbuf *
2269 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2270 {
2271 	struct ifnet *ifp = sc->sc_ifp;
2272 	struct ether_header *eh;
2273 	u_short etype;
2274 
2275 	eh = mtod(m, struct ether_header *);
2276 	etype = ntohs(eh->ether_type);
2277 
2278 	/* Tap off LACP control messages */
2279 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2280 		m = lacp_input(lp, m);
2281 		if (m == NULL)
2282 			return (NULL);
2283 	}
2284 
2285 	/*
2286 	 * If the port is not collecting or not in the active aggregator then
2287 	 * free and return.
2288 	 */
2289 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2290 		m_freem(m);
2291 		return (NULL);
2292 	}
2293 
2294 	m->m_pkthdr.rcvif = ifp;
2295 	return (m);
2296 }
2297 
2298