xref: /freebsd/sys/net/if_lagg.c (revision 2bd9826995ca6b23f8b088cfa035c0ad1c578ac3)
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_kern_tls.h"
27 #include "opt_ratelimit.h"
28 
29 #include <sys/param.h>
30 #include <sys/kernel.h>
31 #include <sys/malloc.h>
32 #include <sys/mbuf.h>
33 #include <sys/queue.h>
34 #include <sys/socket.h>
35 #include <sys/sockio.h>
36 #include <sys/sysctl.h>
37 #include <sys/module.h>
38 #include <sys/priv.h>
39 #include <sys/systm.h>
40 #include <sys/proc.h>
41 #include <sys/lock.h>
42 #include <sys/rmlock.h>
43 #include <sys/sx.h>
44 #include <sys/taskqueue.h>
45 #include <sys/eventhandler.h>
46 
47 #include <net/ethernet.h>
48 #include <net/if.h>
49 #include <net/if_clone.h>
50 #include <net/if_arp.h>
51 #include <net/if_dl.h>
52 #include <net/if_media.h>
53 #include <net/if_types.h>
54 #include <net/if_var.h>
55 #include <net/bpf.h>
56 #include <net/route.h>
57 #include <net/vnet.h>
58 #include <net/infiniband.h>
59 
60 #if defined(INET) || defined(INET6)
61 #include <netinet/in.h>
62 #include <netinet/ip.h>
63 #endif
64 #ifdef INET
65 #include <netinet/in_systm.h>
66 #include <netinet/if_ether.h>
67 #endif
68 
69 #ifdef INET6
70 #include <netinet/ip6.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet6/in6_ifattach.h>
73 #endif
74 
75 #include <net/if_vlan_var.h>
76 #include <net/if_lagg.h>
77 #include <net/ieee8023ad_lacp.h>
78 
79 #ifdef INET6
80 /*
81  * XXX: declare here to avoid to include many inet6 related files..
82  * should be more generalized?
83  */
84 extern void	nd6_setmtu(struct ifnet *);
85 #endif
86 
87 #define	LAGG_SX_INIT(_sc)	sx_init(&(_sc)->sc_sx, "if_lagg sx")
88 #define	LAGG_SX_DESTROY(_sc)	sx_destroy(&(_sc)->sc_sx)
89 #define	LAGG_XLOCK(_sc)		sx_xlock(&(_sc)->sc_sx)
90 #define	LAGG_XUNLOCK(_sc)	sx_xunlock(&(_sc)->sc_sx)
91 #define	LAGG_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
92 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
93 
94 /* Special flags we should propagate to the lagg ports. */
95 static struct {
96 	int flag;
97 	int (*func)(struct ifnet *, int);
98 } lagg_pflags[] = {
99 	{IFF_PROMISC, ifpromisc},
100 	{IFF_ALLMULTI, if_allmulti},
101 	{0, NULL}
102 };
103 
104 struct lagg_snd_tag {
105 	struct m_snd_tag com;
106 	struct m_snd_tag *tag;
107 };
108 
109 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
110 #define	V_lagg_list	VNET(lagg_list)
111 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
112 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
113 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
114 					"if_lagg list", NULL, MTX_DEF)
115 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
116 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
117 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
118 eventhandler_tag	lagg_detach_cookie = NULL;
119 
120 static int	lagg_clone_create(struct if_clone *, int, caddr_t);
121 static void	lagg_clone_destroy(struct ifnet *);
122 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
123 #define	V_lagg_cloner	VNET(lagg_cloner)
124 static const char laggname[] = "lagg";
125 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
126 
127 static void	lagg_capabilities(struct lagg_softc *);
128 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
129 static int	lagg_port_destroy(struct lagg_port *, int);
130 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *);
131 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *);
132 static void	lagg_linkstate(struct lagg_softc *);
133 static void	lagg_port_state(struct ifnet *, int);
134 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
135 static int	lagg_port_output(struct ifnet *, struct mbuf *,
136 		    const struct sockaddr *, struct route *);
137 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
138 #ifdef LAGG_PORT_STACKING
139 static int	lagg_port_checkstacking(struct lagg_softc *);
140 #endif
141 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
142 static void	lagg_init(void *);
143 static void	lagg_stop(struct lagg_softc *);
144 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
145 #if defined(KERN_TLS) || defined(RATELIMIT)
146 static int	lagg_snd_tag_alloc(struct ifnet *,
147 		    union if_snd_tag_alloc_params *,
148 		    struct m_snd_tag **);
149 static int	lagg_snd_tag_modify(struct m_snd_tag *,
150 		    union if_snd_tag_modify_params *);
151 static int	lagg_snd_tag_query(struct m_snd_tag *,
152 		    union if_snd_tag_query_params *);
153 static void	lagg_snd_tag_free(struct m_snd_tag *);
154 static struct m_snd_tag *lagg_next_snd_tag(struct m_snd_tag *);
155 static void     lagg_ratelimit_query(struct ifnet *,
156 		    struct if_ratelimit_query_results *);
157 #endif
158 static int	lagg_setmulti(struct lagg_port *);
159 static int	lagg_clrmulti(struct lagg_port *);
160 static	int	lagg_setcaps(struct lagg_port *, int cap);
161 static	int	lagg_setflag(struct lagg_port *, int, int,
162 		    int (*func)(struct ifnet *, int));
163 static	int	lagg_setflags(struct lagg_port *, int status);
164 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
165 static int	lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
166 static int	lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
167 static void	lagg_qflush(struct ifnet *);
168 static int	lagg_media_change(struct ifnet *);
169 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
170 static struct lagg_port *lagg_link_active(struct lagg_softc *,
171 	    struct lagg_port *);
172 
173 /* Simple round robin */
174 static void	lagg_rr_attach(struct lagg_softc *);
175 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
176 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
177 		    struct mbuf *);
178 
179 /* Active failover */
180 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
181 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
182 		    struct mbuf *);
183 
184 /* Loadbalancing */
185 static void	lagg_lb_attach(struct lagg_softc *);
186 static void	lagg_lb_detach(struct lagg_softc *);
187 static int	lagg_lb_port_create(struct lagg_port *);
188 static void	lagg_lb_port_destroy(struct lagg_port *);
189 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
190 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
191 		    struct mbuf *);
192 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
193 
194 /* Broadcast */
195 static int    lagg_bcast_start(struct lagg_softc *, struct mbuf *);
196 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
197 		    struct mbuf *);
198 
199 /* 802.3ad LACP */
200 static void	lagg_lacp_attach(struct lagg_softc *);
201 static void	lagg_lacp_detach(struct lagg_softc *);
202 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
203 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
204 		    struct mbuf *);
205 static void	lagg_lacp_lladdr(struct lagg_softc *);
206 
207 /* lagg protocol table */
208 static const struct lagg_proto {
209 	lagg_proto	pr_num;
210 	void		(*pr_attach)(struct lagg_softc *);
211 	void		(*pr_detach)(struct lagg_softc *);
212 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
213 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
214 			    struct mbuf *);
215 	int		(*pr_addport)(struct lagg_port *);
216 	void		(*pr_delport)(struct lagg_port *);
217 	void		(*pr_linkstate)(struct lagg_port *);
218 	void 		(*pr_init)(struct lagg_softc *);
219 	void 		(*pr_stop)(struct lagg_softc *);
220 	void 		(*pr_lladdr)(struct lagg_softc *);
221 	void		(*pr_request)(struct lagg_softc *, void *);
222 	void		(*pr_portreq)(struct lagg_port *, void *);
223 } lagg_protos[] = {
224     {
225 	.pr_num = LAGG_PROTO_NONE
226     },
227     {
228 	.pr_num = LAGG_PROTO_ROUNDROBIN,
229 	.pr_attach = lagg_rr_attach,
230 	.pr_start = lagg_rr_start,
231 	.pr_input = lagg_rr_input,
232     },
233     {
234 	.pr_num = LAGG_PROTO_FAILOVER,
235 	.pr_start = lagg_fail_start,
236 	.pr_input = lagg_fail_input,
237     },
238     {
239 	.pr_num = LAGG_PROTO_LOADBALANCE,
240 	.pr_attach = lagg_lb_attach,
241 	.pr_detach = lagg_lb_detach,
242 	.pr_start = lagg_lb_start,
243 	.pr_input = lagg_lb_input,
244 	.pr_addport = lagg_lb_port_create,
245 	.pr_delport = lagg_lb_port_destroy,
246     },
247     {
248 	.pr_num = LAGG_PROTO_LACP,
249 	.pr_attach = lagg_lacp_attach,
250 	.pr_detach = lagg_lacp_detach,
251 	.pr_start = lagg_lacp_start,
252 	.pr_input = lagg_lacp_input,
253 	.pr_addport = lacp_port_create,
254 	.pr_delport = lacp_port_destroy,
255 	.pr_linkstate = lacp_linkstate,
256 	.pr_init = lacp_init,
257 	.pr_stop = lacp_stop,
258 	.pr_lladdr = lagg_lacp_lladdr,
259 	.pr_request = lacp_req,
260 	.pr_portreq = lacp_portreq,
261     },
262     {
263 	.pr_num = LAGG_PROTO_BROADCAST,
264 	.pr_start = lagg_bcast_start,
265 	.pr_input = lagg_bcast_input,
266     },
267 };
268 
269 SYSCTL_DECL(_net_link);
270 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
271     "Link Aggregation");
272 
273 /* Allow input on any failover links */
274 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
275 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
276 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
277     &VNET_NAME(lagg_failover_rx_all), 0,
278     "Accept input from any interface in a failover lagg");
279 
280 /* Default value for using flowid */
281 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
282 #define	V_def_use_flowid	VNET(def_use_flowid)
283 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
284     &VNET_NAME(def_use_flowid), 0,
285     "Default setting for using flow id for load sharing");
286 
287 /* Default value for using numa */
288 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
289 #define	V_def_use_numa	VNET(def_use_numa)
290 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN,
291     &VNET_NAME(def_use_numa), 0,
292     "Use numa to steer flows");
293 
294 /* Default value for flowid shift */
295 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
296 #define	V_def_flowid_shift	VNET(def_flowid_shift)
297 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
298     &VNET_NAME(def_flowid_shift), 0,
299     "Default setting for flowid shift for load sharing");
300 
301 static void
302 vnet_lagg_init(const void *unused __unused)
303 {
304 
305 	LAGG_LIST_LOCK_INIT();
306 	SLIST_INIT(&V_lagg_list);
307 	V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
308 	    lagg_clone_destroy, 0);
309 }
310 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
311     vnet_lagg_init, NULL);
312 
313 static void
314 vnet_lagg_uninit(const void *unused __unused)
315 {
316 
317 	if_clone_detach(V_lagg_cloner);
318 	LAGG_LIST_LOCK_DESTROY();
319 }
320 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
321     vnet_lagg_uninit, NULL);
322 
323 static int
324 lagg_modevent(module_t mod, int type, void *data)
325 {
326 
327 	switch (type) {
328 	case MOD_LOAD:
329 		lagg_input_ethernet_p = lagg_input_ethernet;
330 		lagg_input_infiniband_p = lagg_input_infiniband;
331 		lagg_linkstate_p = lagg_port_state;
332 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
333 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
334 		    EVENTHANDLER_PRI_ANY);
335 		break;
336 	case MOD_UNLOAD:
337 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
338 		    lagg_detach_cookie);
339 		lagg_input_ethernet_p = NULL;
340 		lagg_input_infiniband_p = NULL;
341 		lagg_linkstate_p = NULL;
342 		break;
343 	default:
344 		return (EOPNOTSUPP);
345 	}
346 	return (0);
347 }
348 
349 static moduledata_t lagg_mod = {
350 	"if_lagg",
351 	lagg_modevent,
352 	0
353 };
354 
355 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
356 MODULE_VERSION(if_lagg, 1);
357 MODULE_DEPEND(if_lagg, if_infiniband, 1, 1, 1);
358 
359 static void
360 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
361 {
362 
363 	LAGG_XLOCK_ASSERT(sc);
364 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
365 	    __func__, sc));
366 
367 	if (sc->sc_ifflags & IFF_DEBUG)
368 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
369 
370 	if (lagg_protos[pr].pr_attach != NULL)
371 		lagg_protos[pr].pr_attach(sc);
372 	sc->sc_proto = pr;
373 }
374 
375 static void
376 lagg_proto_detach(struct lagg_softc *sc)
377 {
378 	lagg_proto pr;
379 
380 	LAGG_XLOCK_ASSERT(sc);
381 	pr = sc->sc_proto;
382 	sc->sc_proto = LAGG_PROTO_NONE;
383 
384 	if (lagg_protos[pr].pr_detach != NULL)
385 		lagg_protos[pr].pr_detach(sc);
386 }
387 
388 static int
389 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
390 {
391 
392 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
393 }
394 
395 static struct mbuf *
396 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
397 {
398 
399 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
400 }
401 
402 static int
403 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
404 {
405 
406 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
407 		return (0);
408 	else
409 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
410 }
411 
412 static void
413 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
414 {
415 
416 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
417 		lagg_protos[sc->sc_proto].pr_delport(lp);
418 }
419 
420 static void
421 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
422 {
423 
424 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
425 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
426 }
427 
428 static void
429 lagg_proto_init(struct lagg_softc *sc)
430 {
431 
432 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
433 		lagg_protos[sc->sc_proto].pr_init(sc);
434 }
435 
436 static void
437 lagg_proto_stop(struct lagg_softc *sc)
438 {
439 
440 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
441 		lagg_protos[sc->sc_proto].pr_stop(sc);
442 }
443 
444 static void
445 lagg_proto_lladdr(struct lagg_softc *sc)
446 {
447 
448 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
449 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
450 }
451 
452 static void
453 lagg_proto_request(struct lagg_softc *sc, void *v)
454 {
455 
456 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
457 		lagg_protos[sc->sc_proto].pr_request(sc, v);
458 }
459 
460 static void
461 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
462 {
463 
464 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
465 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
466 }
467 
468 /*
469  * This routine is run via an vlan
470  * config EVENT
471  */
472 static void
473 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
474 {
475 	struct lagg_softc *sc = ifp->if_softc;
476 	struct lagg_port *lp;
477 
478 	if (ifp->if_softc !=  arg)   /* Not our event */
479 		return;
480 
481 	LAGG_XLOCK(sc);
482 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
483 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
484 	LAGG_XUNLOCK(sc);
485 }
486 
487 /*
488  * This routine is run via an vlan
489  * unconfig EVENT
490  */
491 static void
492 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
493 {
494 	struct lagg_softc *sc = ifp->if_softc;
495 	struct lagg_port *lp;
496 
497 	if (ifp->if_softc !=  arg)   /* Not our event */
498 		return;
499 
500 	LAGG_XLOCK(sc);
501 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
502 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
503 	LAGG_XUNLOCK(sc);
504 }
505 
506 static int
507 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
508 {
509 	struct iflaggparam iflp;
510 	struct lagg_softc *sc;
511 	struct ifnet *ifp;
512 	int if_type;
513 	int error;
514 	static const uint8_t eaddr[LAGG_ADDR_LEN];
515 
516 	if (params != NULL) {
517 		error = copyin(params, &iflp, sizeof(iflp));
518 		if (error)
519 			return (error);
520 
521 		switch (iflp.lagg_type) {
522 		case LAGG_TYPE_ETHERNET:
523 			if_type = IFT_ETHER;
524 			break;
525 		case LAGG_TYPE_INFINIBAND:
526 			if_type = IFT_INFINIBAND;
527 			break;
528 		default:
529 			return (EINVAL);
530 		}
531 	} else {
532 		if_type = IFT_ETHER;
533 	}
534 
535 	sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO);
536 	ifp = sc->sc_ifp = if_alloc(if_type);
537 	if (ifp == NULL) {
538 		free(sc, M_LAGG);
539 		return (ENOSPC);
540 	}
541 	LAGG_SX_INIT(sc);
542 
543 	mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
544 	callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
545 
546 	LAGG_XLOCK(sc);
547 	if (V_def_use_flowid)
548 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
549 	if (V_def_use_numa)
550 		sc->sc_opts |= LAGG_OPT_USE_NUMA;
551 	sc->flowid_shift = V_def_flowid_shift;
552 
553 	/* Hash all layers by default */
554 	sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
555 
556 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
557 
558 	CK_SLIST_INIT(&sc->sc_ports);
559 
560 	switch (if_type) {
561 	case IFT_ETHER:
562 		/* Initialise pseudo media types */
563 		ifmedia_init(&sc->sc_media, 0, lagg_media_change,
564 		    lagg_media_status);
565 		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
566 		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
567 
568 		if_initname(ifp, laggname, unit);
569 		ifp->if_transmit = lagg_transmit_ethernet;
570 		break;
571 	case IFT_INFINIBAND:
572 		if_initname(ifp, laggname, unit);
573 		ifp->if_transmit = lagg_transmit_infiniband;
574 		break;
575 	default:
576 		break;
577 	}
578 	ifp->if_softc = sc;
579 	ifp->if_qflush = lagg_qflush;
580 	ifp->if_init = lagg_init;
581 	ifp->if_ioctl = lagg_ioctl;
582 	ifp->if_get_counter = lagg_get_counter;
583 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
584 #if defined(KERN_TLS) || defined(RATELIMIT)
585 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
586 	ifp->if_ratelimit_query = lagg_ratelimit_query;
587 #endif
588 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
589 
590 	/*
591 	 * Attach as an ordinary ethernet device, children will be attached
592 	 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
593 	 */
594 	switch (if_type) {
595 	case IFT_ETHER:
596 		ether_ifattach(ifp, eaddr);
597 		break;
598 	case IFT_INFINIBAND:
599 		infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr);
600 		break;
601 	default:
602 		break;
603 	}
604 
605 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
606 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
607 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
608 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
609 
610 	/* Insert into the global list of laggs */
611 	LAGG_LIST_LOCK();
612 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
613 	LAGG_LIST_UNLOCK();
614 	LAGG_XUNLOCK(sc);
615 
616 	return (0);
617 }
618 
619 static void
620 lagg_clone_destroy(struct ifnet *ifp)
621 {
622 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
623 	struct lagg_port *lp;
624 
625 	LAGG_XLOCK(sc);
626 	sc->sc_destroying = 1;
627 	lagg_stop(sc);
628 	ifp->if_flags &= ~IFF_UP;
629 
630 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
631 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
632 
633 	/* Shutdown and remove lagg ports */
634 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
635 		lagg_port_destroy(lp, 1);
636 
637 	/* Unhook the aggregation protocol */
638 	lagg_proto_detach(sc);
639 	LAGG_XUNLOCK(sc);
640 
641 	switch (ifp->if_type) {
642 	case IFT_ETHER:
643 		ifmedia_removeall(&sc->sc_media);
644 		ether_ifdetach(ifp);
645 		break;
646 	case IFT_INFINIBAND:
647 		infiniband_ifdetach(ifp);
648 		break;
649 	default:
650 		break;
651 	}
652 	if_free(ifp);
653 
654 	LAGG_LIST_LOCK();
655 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
656 	LAGG_LIST_UNLOCK();
657 
658 	mtx_destroy(&sc->sc_mtx);
659 	LAGG_SX_DESTROY(sc);
660 	free(sc, M_LAGG);
661 }
662 
663 static void
664 lagg_capabilities(struct lagg_softc *sc)
665 {
666 	struct lagg_port *lp;
667 	int cap, ena, pena;
668 	uint64_t hwa;
669 	struct ifnet_hw_tsomax hw_tsomax;
670 
671 	LAGG_XLOCK_ASSERT(sc);
672 
673 	/* Get common enabled capabilities for the lagg ports */
674 	ena = ~0;
675 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
676 		ena &= lp->lp_ifp->if_capenable;
677 	ena = (ena == ~0 ? 0 : ena);
678 
679 	/*
680 	 * Apply common enabled capabilities back to the lagg ports.
681 	 * May require several iterations if they are dependent.
682 	 */
683 	do {
684 		pena = ena;
685 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
686 			lagg_setcaps(lp, ena);
687 			ena &= lp->lp_ifp->if_capenable;
688 		}
689 	} while (pena != ena);
690 
691 	/* Get other capabilities from the lagg ports */
692 	cap = ~0;
693 	hwa = ~(uint64_t)0;
694 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
695 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
696 		cap &= lp->lp_ifp->if_capabilities;
697 		hwa &= lp->lp_ifp->if_hwassist;
698 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
699 	}
700 	cap = (cap == ~0 ? 0 : cap);
701 	hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
702 
703 	if (sc->sc_ifp->if_capabilities != cap ||
704 	    sc->sc_ifp->if_capenable != ena ||
705 	    sc->sc_ifp->if_hwassist != hwa ||
706 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
707 		sc->sc_ifp->if_capabilities = cap;
708 		sc->sc_ifp->if_capenable = ena;
709 		sc->sc_ifp->if_hwassist = hwa;
710 		getmicrotime(&sc->sc_ifp->if_lastchange);
711 
712 		if (sc->sc_ifflags & IFF_DEBUG)
713 			if_printf(sc->sc_ifp,
714 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
715 	}
716 }
717 
718 static int
719 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
720 {
721 	struct lagg_softc *sc_ptr;
722 	struct lagg_port *lp, *tlp;
723 	struct ifreq ifr;
724 	int error, i, oldmtu;
725 	int if_type;
726 	uint64_t *pval;
727 
728 	LAGG_XLOCK_ASSERT(sc);
729 
730 	if (sc->sc_ifp == ifp) {
731 		if_printf(sc->sc_ifp,
732 		    "cannot add a lagg to itself as a port\n");
733 		return (EINVAL);
734 	}
735 
736 	if (sc->sc_destroying == 1)
737 		return (ENXIO);
738 
739 	/* Limit the maximal number of lagg ports */
740 	if (sc->sc_count >= LAGG_MAX_PORTS)
741 		return (ENOSPC);
742 
743 	/* Check if port has already been associated to a lagg */
744 	if (ifp->if_lagg != NULL) {
745 		/* Port is already in the current lagg? */
746 		lp = (struct lagg_port *)ifp->if_lagg;
747 		if (lp->lp_softc == sc)
748 			return (EEXIST);
749 		return (EBUSY);
750 	}
751 
752 	switch (sc->sc_ifp->if_type) {
753 	case IFT_ETHER:
754 		/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
755 		if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
756 			return (EPROTONOSUPPORT);
757 		if_type = IFT_IEEE8023ADLAG;
758 		break;
759 	case IFT_INFINIBAND:
760 		/* XXX Disallow non-infiniband interfaces */
761 		if (ifp->if_type != IFT_INFINIBAND)
762 			return (EPROTONOSUPPORT);
763 		if_type = IFT_INFINIBANDLAG;
764 		break;
765 	default:
766 		break;
767 	}
768 
769 	/* Allow the first Ethernet member to define the MTU */
770 	oldmtu = -1;
771 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
772 		sc->sc_ifp->if_mtu = ifp->if_mtu;
773 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
774 		if (ifp->if_ioctl == NULL) {
775 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
776 			    ifp->if_xname);
777 			return (EINVAL);
778 		}
779 		oldmtu = ifp->if_mtu;
780 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
781 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
782 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
783 		if (error != 0) {
784 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
785 			    ifp->if_xname);
786 			return (error);
787 		}
788 		ifr.ifr_mtu = oldmtu;
789 	}
790 
791 	lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO);
792 	lp->lp_softc = sc;
793 
794 	/* Check if port is a stacked lagg */
795 	LAGG_LIST_LOCK();
796 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
797 		if (ifp == sc_ptr->sc_ifp) {
798 			LAGG_LIST_UNLOCK();
799 			free(lp, M_LAGG);
800 			if (oldmtu != -1)
801 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
802 				    (caddr_t)&ifr);
803 			return (EINVAL);
804 			/* XXX disable stacking for the moment, its untested */
805 #ifdef LAGG_PORT_STACKING
806 			lp->lp_flags |= LAGG_PORT_STACK;
807 			if (lagg_port_checkstacking(sc_ptr) >=
808 			    LAGG_MAX_STACKING) {
809 				LAGG_LIST_UNLOCK();
810 				free(lp, M_LAGG);
811 				if (oldmtu != -1)
812 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
813 					    (caddr_t)&ifr);
814 				return (E2BIG);
815 			}
816 #endif
817 		}
818 	}
819 	LAGG_LIST_UNLOCK();
820 
821 	if_ref(ifp);
822 	lp->lp_ifp = ifp;
823 
824 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
825 	lp->lp_ifcapenable = ifp->if_capenable;
826 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
827 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
828 		lagg_proto_lladdr(sc);
829 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
830 	} else {
831 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
832 	}
833 	lagg_setflags(lp, 1);
834 
835 	if (CK_SLIST_EMPTY(&sc->sc_ports))
836 		sc->sc_primary = lp;
837 
838 	/* Change the interface type */
839 	lp->lp_iftype = ifp->if_type;
840 	ifp->if_type = if_type;
841 	ifp->if_lagg = lp;
842 	lp->lp_ioctl = ifp->if_ioctl;
843 	ifp->if_ioctl = lagg_port_ioctl;
844 	lp->lp_output = ifp->if_output;
845 	ifp->if_output = lagg_port_output;
846 
847 	/* Read port counters */
848 	pval = lp->port_counters.val;
849 	for (i = 0; i < IFCOUNTERS; i++, pval++)
850 		*pval = ifp->if_get_counter(ifp, i);
851 
852 	/*
853 	 * Insert into the list of ports.
854 	 * Keep ports sorted by if_index. It is handy, when configuration
855 	 * is predictable and `ifconfig laggN create ...` command
856 	 * will lead to the same result each time.
857 	 */
858 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
859 		if (tlp->lp_ifp->if_index < ifp->if_index && (
860 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
861 		    ((struct  lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
862 		    ifp->if_index))
863 			break;
864 	}
865 	if (tlp != NULL)
866 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
867 	else
868 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
869 	sc->sc_count++;
870 
871 	lagg_setmulti(lp);
872 
873 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
874 		/* Remove the port, without calling pr_delport. */
875 		lagg_port_destroy(lp, 0);
876 		if (oldmtu != -1)
877 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
878 		return (error);
879 	}
880 
881 	/* Update lagg capabilities */
882 	lagg_capabilities(sc);
883 	lagg_linkstate(sc);
884 
885 	return (0);
886 }
887 
888 #ifdef LAGG_PORT_STACKING
889 static int
890 lagg_port_checkstacking(struct lagg_softc *sc)
891 {
892 	struct lagg_softc *sc_ptr;
893 	struct lagg_port *lp;
894 	int m = 0;
895 
896 	LAGG_SXLOCK_ASSERT(sc);
897 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
898 		if (lp->lp_flags & LAGG_PORT_STACK) {
899 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
900 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
901 		}
902 	}
903 
904 	return (m + 1);
905 }
906 #endif
907 
908 static void
909 lagg_port_destroy_cb(epoch_context_t ec)
910 {
911 	struct lagg_port *lp;
912 	struct ifnet *ifp;
913 
914 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
915 	ifp = lp->lp_ifp;
916 
917 	if_rele(ifp);
918 	free(lp, M_LAGG);
919 }
920 
921 static int
922 lagg_port_destroy(struct lagg_port *lp, int rundelport)
923 {
924 	struct lagg_softc *sc = lp->lp_softc;
925 	struct lagg_port *lp_ptr, *lp0;
926 	struct ifnet *ifp = lp->lp_ifp;
927 	uint64_t *pval, vdiff;
928 	int i;
929 
930 	LAGG_XLOCK_ASSERT(sc);
931 
932 	if (rundelport)
933 		lagg_proto_delport(sc, lp);
934 
935 	if (lp->lp_detaching == 0)
936 		lagg_clrmulti(lp);
937 
938 	/* Restore interface */
939 	ifp->if_type = lp->lp_iftype;
940 	ifp->if_ioctl = lp->lp_ioctl;
941 	ifp->if_output = lp->lp_output;
942 	ifp->if_lagg = NULL;
943 
944 	/* Update detached port counters */
945 	pval = lp->port_counters.val;
946 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
947 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
948 		sc->detached_counters.val[i] += vdiff;
949 	}
950 
951 	/* Finally, remove the port from the lagg */
952 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
953 	sc->sc_count--;
954 
955 	/* Update the primary interface */
956 	if (lp == sc->sc_primary) {
957 		uint8_t lladdr[LAGG_ADDR_LEN];
958 
959 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
960 			bzero(&lladdr, LAGG_ADDR_LEN);
961 		else
962 			bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
963 		sc->sc_primary = lp0;
964 		if (sc->sc_destroying == 0) {
965 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
966 			lagg_proto_lladdr(sc);
967 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
968 
969 			/*
970 			 * Update lladdr for each port (new primary needs update
971 			 * as well, to switch from old lladdr to its 'real' one).
972 			 * We can skip this if the lagg is being destroyed.
973 			 */
974 			CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
975 				if_setlladdr(lp_ptr->lp_ifp, lladdr,
976 				    lp_ptr->lp_ifp->if_addrlen);
977 		}
978 	}
979 
980 	if (lp->lp_ifflags)
981 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
982 
983 	if (lp->lp_detaching == 0) {
984 		lagg_setflags(lp, 0);
985 		lagg_setcaps(lp, lp->lp_ifcapenable);
986 		if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
987 	}
988 
989 	/*
990 	 * free port and release it's ifnet reference after a grace period has
991 	 * elapsed.
992 	 */
993 	NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
994 	/* Update lagg capabilities */
995 	lagg_capabilities(sc);
996 	lagg_linkstate(sc);
997 
998 	return (0);
999 }
1000 
1001 static int
1002 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1003 {
1004 	struct epoch_tracker et;
1005 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
1006 	struct lagg_softc *sc;
1007 	struct lagg_port *lp = NULL;
1008 	int error = 0;
1009 
1010 	/* Should be checked by the caller */
1011 	switch (ifp->if_type) {
1012 	case IFT_IEEE8023ADLAG:
1013 	case IFT_INFINIBANDLAG:
1014 		if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1015 			goto fallback;
1016 		break;
1017 	default:
1018 		goto fallback;
1019 	}
1020 
1021 	switch (cmd) {
1022 	case SIOCGLAGGPORT:
1023 		if (rp->rp_portname[0] == '\0' ||
1024 		    ifunit(rp->rp_portname) != ifp) {
1025 			error = EINVAL;
1026 			break;
1027 		}
1028 
1029 		NET_EPOCH_ENTER(et);
1030 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1031 			error = ENOENT;
1032 			NET_EPOCH_EXIT(et);
1033 			break;
1034 		}
1035 
1036 		lagg_port2req(lp, rp);
1037 		NET_EPOCH_EXIT(et);
1038 		break;
1039 
1040 	case SIOCSIFCAP:
1041 		if (lp->lp_ioctl == NULL) {
1042 			error = EINVAL;
1043 			break;
1044 		}
1045 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1046 		if (error)
1047 			break;
1048 
1049 		/* Update lagg interface capabilities */
1050 		LAGG_XLOCK(sc);
1051 		lagg_capabilities(sc);
1052 		LAGG_XUNLOCK(sc);
1053 		VLAN_CAPABILITIES(sc->sc_ifp);
1054 		break;
1055 
1056 	case SIOCSIFMTU:
1057 		/* Do not allow the MTU to be changed once joined */
1058 		error = EINVAL;
1059 		break;
1060 
1061 	default:
1062 		goto fallback;
1063 	}
1064 
1065 	return (error);
1066 
1067 fallback:
1068 	if (lp != NULL && lp->lp_ioctl != NULL)
1069 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1070 
1071 	return (EINVAL);
1072 }
1073 
1074 /*
1075  * Requests counter @cnt data.
1076  *
1077  * Counter value is calculated the following way:
1078  * 1) for each port, sum  difference between current and "initial" measurements.
1079  * 2) add lagg logical interface counters.
1080  * 3) add data from detached_counters array.
1081  *
1082  * We also do the following things on ports attach/detach:
1083  * 1) On port attach we store all counters it has into port_counter array.
1084  * 2) On port detach we add the different between "initial" and
1085  *   current counters data to detached_counters array.
1086  */
1087 static uint64_t
1088 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1089 {
1090 	struct epoch_tracker et;
1091 	struct lagg_softc *sc;
1092 	struct lagg_port *lp;
1093 	struct ifnet *lpifp;
1094 	uint64_t newval, oldval, vsum;
1095 
1096 	/* Revise this when we've got non-generic counters. */
1097 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1098 
1099 	sc = (struct lagg_softc *)ifp->if_softc;
1100 
1101 	vsum = 0;
1102 	NET_EPOCH_ENTER(et);
1103 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1104 		/* Saved attached value */
1105 		oldval = lp->port_counters.val[cnt];
1106 		/* current value */
1107 		lpifp = lp->lp_ifp;
1108 		newval = lpifp->if_get_counter(lpifp, cnt);
1109 		/* Calculate diff and save new */
1110 		vsum += newval - oldval;
1111 	}
1112 	NET_EPOCH_EXIT(et);
1113 
1114 	/*
1115 	 * Add counter data which might be added by upper
1116 	 * layer protocols operating on logical interface.
1117 	 */
1118 	vsum += if_get_counter_default(ifp, cnt);
1119 
1120 	/*
1121 	 * Add counter data from detached ports counters
1122 	 */
1123 	vsum += sc->detached_counters.val[cnt];
1124 
1125 	return (vsum);
1126 }
1127 
1128 /*
1129  * For direct output to child ports.
1130  */
1131 static int
1132 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1133 	const struct sockaddr *dst, struct route *ro)
1134 {
1135 	struct lagg_port *lp = ifp->if_lagg;
1136 
1137 	switch (dst->sa_family) {
1138 		case pseudo_AF_HDRCMPLT:
1139 		case AF_UNSPEC:
1140 			if (lp != NULL)
1141 				return ((*lp->lp_output)(ifp, m, dst, ro));
1142 	}
1143 
1144 	/* drop any other frames */
1145 	m_freem(m);
1146 	return (ENETDOWN);
1147 }
1148 
1149 static void
1150 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1151 {
1152 	struct lagg_port *lp;
1153 	struct lagg_softc *sc;
1154 
1155 	if ((lp = ifp->if_lagg) == NULL)
1156 		return;
1157 	/* If the ifnet is just being renamed, don't do anything. */
1158 	if (ifp->if_flags & IFF_RENAMING)
1159 		return;
1160 
1161 	sc = lp->lp_softc;
1162 
1163 	LAGG_XLOCK(sc);
1164 	lp->lp_detaching = 1;
1165 	lagg_port_destroy(lp, 1);
1166 	LAGG_XUNLOCK(sc);
1167 	VLAN_CAPABILITIES(sc->sc_ifp);
1168 }
1169 
1170 static void
1171 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1172 {
1173 	struct lagg_softc *sc = lp->lp_softc;
1174 
1175 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1176 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1177 	rp->rp_prio = lp->lp_prio;
1178 	rp->rp_flags = lp->lp_flags;
1179 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1180 
1181 	/* Add protocol specific flags */
1182 	switch (sc->sc_proto) {
1183 		case LAGG_PROTO_FAILOVER:
1184 			if (lp == sc->sc_primary)
1185 				rp->rp_flags |= LAGG_PORT_MASTER;
1186 			if (lp == lagg_link_active(sc, sc->sc_primary))
1187 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1188 			break;
1189 
1190 		case LAGG_PROTO_ROUNDROBIN:
1191 		case LAGG_PROTO_LOADBALANCE:
1192 		case LAGG_PROTO_BROADCAST:
1193 			if (LAGG_PORTACTIVE(lp))
1194 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1195 			break;
1196 
1197 		case LAGG_PROTO_LACP:
1198 			/* LACP has a different definition of active */
1199 			if (lacp_isactive(lp))
1200 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1201 			if (lacp_iscollecting(lp))
1202 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1203 			if (lacp_isdistributing(lp))
1204 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1205 			break;
1206 	}
1207 
1208 }
1209 
1210 static void
1211 lagg_watchdog_infiniband(void *arg)
1212 {
1213 	struct epoch_tracker et;
1214 	struct lagg_softc *sc;
1215 	struct lagg_port *lp;
1216 	struct ifnet *ifp;
1217 	struct ifnet *lp_ifp;
1218 
1219 	sc = arg;
1220 
1221 	/*
1222 	 * Because infiniband nodes have a fixed MAC address, which is
1223 	 * generated by the so-called GID, we need to regularly update
1224 	 * the link level address of the parent lagg<N> device when
1225 	 * the active port changes. Possibly we could piggy-back on
1226 	 * link up/down events aswell, but using a timer also provides
1227 	 * a guarantee against too frequent events. This operation
1228 	 * does not have to be atomic.
1229 	 */
1230 	NET_EPOCH_ENTER(et);
1231 	lp = lagg_link_active(sc, sc->sc_primary);
1232 	if (lp != NULL) {
1233 		ifp = sc->sc_ifp;
1234 		lp_ifp = lp->lp_ifp;
1235 
1236 		if (ifp != NULL && lp_ifp != NULL &&
1237 		    (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0 ||
1238 		     memcmp(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen) != 0)) {
1239 			memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1240 			memcpy(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen);
1241 
1242 			CURVNET_SET(ifp->if_vnet);
1243 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1244 			CURVNET_RESTORE();
1245 		}
1246 	}
1247 	NET_EPOCH_EXIT(et);
1248 
1249 	callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1250 }
1251 
1252 static void
1253 lagg_init(void *xsc)
1254 {
1255 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1256 	struct ifnet *ifp = sc->sc_ifp;
1257 	struct lagg_port *lp;
1258 
1259 	LAGG_XLOCK(sc);
1260 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1261 		LAGG_XUNLOCK(sc);
1262 		return;
1263 	}
1264 
1265 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1266 
1267 	/*
1268 	 * Update the port lladdrs if needed.
1269 	 * This might be if_setlladdr() notification
1270 	 * that lladdr has been changed.
1271 	 */
1272 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1273 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1274 		    ifp->if_addrlen) != 0)
1275 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1276 	}
1277 
1278 	lagg_proto_init(sc);
1279 
1280 	if (ifp->if_type == IFT_INFINIBAND) {
1281 		mtx_lock(&sc->sc_mtx);
1282 		lagg_watchdog_infiniband(sc);
1283 		mtx_unlock(&sc->sc_mtx);
1284 	}
1285 
1286 	LAGG_XUNLOCK(sc);
1287 }
1288 
1289 static void
1290 lagg_stop(struct lagg_softc *sc)
1291 {
1292 	struct ifnet *ifp = sc->sc_ifp;
1293 
1294 	LAGG_XLOCK_ASSERT(sc);
1295 
1296 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1297 		return;
1298 
1299 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1300 
1301 	lagg_proto_stop(sc);
1302 
1303 	mtx_lock(&sc->sc_mtx);
1304 	callout_stop(&sc->sc_watchdog);
1305 	mtx_unlock(&sc->sc_mtx);
1306 
1307 	callout_drain(&sc->sc_watchdog);
1308 }
1309 
1310 static int
1311 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1312 {
1313 	struct epoch_tracker et;
1314 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1315 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1316 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1317 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1318 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1319 	struct ifreq *ifr = (struct ifreq *)data;
1320 	struct lagg_port *lp;
1321 	struct ifnet *tpif;
1322 	struct thread *td = curthread;
1323 	char *buf, *outbuf;
1324 	int count, buflen, len, error = 0, oldmtu;
1325 
1326 	bzero(&rpbuf, sizeof(rpbuf));
1327 
1328 	/* XXX: This can race with lagg_clone_destroy. */
1329 
1330 	switch (cmd) {
1331 	case SIOCGLAGG:
1332 		LAGG_XLOCK(sc);
1333 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1334 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1335 		ra->ra_proto = sc->sc_proto;
1336 		lagg_proto_request(sc, &ra->ra_psc);
1337 		count = 0;
1338 		buf = outbuf;
1339 		len = min(ra->ra_size, buflen);
1340 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1341 			if (len < sizeof(rpbuf))
1342 				break;
1343 
1344 			lagg_port2req(lp, &rpbuf);
1345 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1346 			count++;
1347 			buf += sizeof(rpbuf);
1348 			len -= sizeof(rpbuf);
1349 		}
1350 		LAGG_XUNLOCK(sc);
1351 		ra->ra_ports = count;
1352 		ra->ra_size = count * sizeof(rpbuf);
1353 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1354 		free(outbuf, M_TEMP);
1355 		break;
1356 	case SIOCSLAGG:
1357 		error = priv_check(td, PRIV_NET_LAGG);
1358 		if (error)
1359 			break;
1360 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1361 			error = EPROTONOSUPPORT;
1362 			break;
1363 		}
1364 		/* Infiniband only supports the failover protocol. */
1365 		if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1366 		    ifp->if_type == IFT_INFINIBAND) {
1367 			error = EPROTONOSUPPORT;
1368 			break;
1369 		}
1370 		LAGG_XLOCK(sc);
1371 		lagg_proto_detach(sc);
1372 		lagg_proto_attach(sc, ra->ra_proto);
1373 		LAGG_XUNLOCK(sc);
1374 		break;
1375 	case SIOCGLAGGOPTS:
1376 		LAGG_XLOCK(sc);
1377 		ro->ro_opts = sc->sc_opts;
1378 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1379 			struct lacp_softc *lsc;
1380 
1381 			lsc = (struct lacp_softc *)sc->sc_psc;
1382 			if (lsc->lsc_debug.lsc_tx_test != 0)
1383 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1384 			if (lsc->lsc_debug.lsc_rx_test != 0)
1385 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1386 			if (lsc->lsc_strict_mode != 0)
1387 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1388 			if (lsc->lsc_fast_timeout != 0)
1389 				ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1390 
1391 			ro->ro_active = sc->sc_active;
1392 		} else {
1393 			ro->ro_active = 0;
1394 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1395 				ro->ro_active += LAGG_PORTACTIVE(lp);
1396 		}
1397 		ro->ro_bkt = sc->sc_stride;
1398 		ro->ro_flapping = sc->sc_flapping;
1399 		ro->ro_flowid_shift = sc->flowid_shift;
1400 		LAGG_XUNLOCK(sc);
1401 		break;
1402 	case SIOCSLAGGOPTS:
1403 		error = priv_check(td, PRIV_NET_LAGG);
1404 		if (error)
1405 			break;
1406 
1407 		/*
1408 		 * The stride option was added without defining a corresponding
1409 		 * LAGG_OPT flag, so handle a non-zero value before checking
1410 		 * anything else to preserve compatibility.
1411 		 */
1412 		LAGG_XLOCK(sc);
1413 		if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1414 			if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1415 				LAGG_XUNLOCK(sc);
1416 				error = EINVAL;
1417 				break;
1418 			}
1419 			sc->sc_stride = ro->ro_bkt;
1420 		}
1421 		if (ro->ro_opts == 0) {
1422 			LAGG_XUNLOCK(sc);
1423 			break;
1424 		}
1425 
1426 		/*
1427 		 * Set options.  LACP options are stored in sc->sc_psc,
1428 		 * not in sc_opts.
1429 		 */
1430 		int valid, lacp;
1431 
1432 		switch (ro->ro_opts) {
1433 		case LAGG_OPT_USE_FLOWID:
1434 		case -LAGG_OPT_USE_FLOWID:
1435 		case LAGG_OPT_USE_NUMA:
1436 		case -LAGG_OPT_USE_NUMA:
1437 		case LAGG_OPT_FLOWIDSHIFT:
1438 		case LAGG_OPT_RR_LIMIT:
1439 			valid = 1;
1440 			lacp = 0;
1441 			break;
1442 		case LAGG_OPT_LACP_TXTEST:
1443 		case -LAGG_OPT_LACP_TXTEST:
1444 		case LAGG_OPT_LACP_RXTEST:
1445 		case -LAGG_OPT_LACP_RXTEST:
1446 		case LAGG_OPT_LACP_STRICT:
1447 		case -LAGG_OPT_LACP_STRICT:
1448 		case LAGG_OPT_LACP_FAST_TIMO:
1449 		case -LAGG_OPT_LACP_FAST_TIMO:
1450 			valid = lacp = 1;
1451 			break;
1452 		default:
1453 			valid = lacp = 0;
1454 			break;
1455 		}
1456 
1457 		if (valid == 0 ||
1458 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1459 			/* Invalid combination of options specified. */
1460 			error = EINVAL;
1461 			LAGG_XUNLOCK(sc);
1462 			break;	/* Return from SIOCSLAGGOPTS. */
1463 		}
1464 
1465 		/*
1466 		 * Store new options into sc->sc_opts except for
1467 		 * FLOWIDSHIFT, RR and LACP options.
1468 		 */
1469 		if (lacp == 0) {
1470 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1471 				sc->flowid_shift = ro->ro_flowid_shift;
1472 			else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1473 				if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1474 				    ro->ro_bkt == 0) {
1475 					error = EINVAL;
1476 					LAGG_XUNLOCK(sc);
1477 					break;
1478 				}
1479 				sc->sc_stride = ro->ro_bkt;
1480 			} else if (ro->ro_opts > 0)
1481 				sc->sc_opts |= ro->ro_opts;
1482 			else
1483 				sc->sc_opts &= ~ro->ro_opts;
1484 		} else {
1485 			struct lacp_softc *lsc;
1486 			struct lacp_port *lp;
1487 
1488 			lsc = (struct lacp_softc *)sc->sc_psc;
1489 
1490 			switch (ro->ro_opts) {
1491 			case LAGG_OPT_LACP_TXTEST:
1492 				lsc->lsc_debug.lsc_tx_test = 1;
1493 				break;
1494 			case -LAGG_OPT_LACP_TXTEST:
1495 				lsc->lsc_debug.lsc_tx_test = 0;
1496 				break;
1497 			case LAGG_OPT_LACP_RXTEST:
1498 				lsc->lsc_debug.lsc_rx_test = 1;
1499 				break;
1500 			case -LAGG_OPT_LACP_RXTEST:
1501 				lsc->lsc_debug.lsc_rx_test = 0;
1502 				break;
1503 			case LAGG_OPT_LACP_STRICT:
1504 				lsc->lsc_strict_mode = 1;
1505 				break;
1506 			case -LAGG_OPT_LACP_STRICT:
1507 				lsc->lsc_strict_mode = 0;
1508 				break;
1509 			case LAGG_OPT_LACP_FAST_TIMO:
1510 				LACP_LOCK(lsc);
1511         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1512                         		lp->lp_state |= LACP_STATE_TIMEOUT;
1513 				LACP_UNLOCK(lsc);
1514 				lsc->lsc_fast_timeout = 1;
1515 				break;
1516 			case -LAGG_OPT_LACP_FAST_TIMO:
1517 				LACP_LOCK(lsc);
1518         			LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1519                         		lp->lp_state &= ~LACP_STATE_TIMEOUT;
1520 				LACP_UNLOCK(lsc);
1521 				lsc->lsc_fast_timeout = 0;
1522 				break;
1523 			}
1524 		}
1525 		LAGG_XUNLOCK(sc);
1526 		break;
1527 	case SIOCGLAGGFLAGS:
1528 		rf->rf_flags = 0;
1529 		LAGG_XLOCK(sc);
1530 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1531 			rf->rf_flags |= LAGG_F_HASHL2;
1532 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1533 			rf->rf_flags |= LAGG_F_HASHL3;
1534 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1535 			rf->rf_flags |= LAGG_F_HASHL4;
1536 		LAGG_XUNLOCK(sc);
1537 		break;
1538 	case SIOCSLAGGHASH:
1539 		error = priv_check(td, PRIV_NET_LAGG);
1540 		if (error)
1541 			break;
1542 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1543 			error = EINVAL;
1544 			break;
1545 		}
1546 		LAGG_XLOCK(sc);
1547 		sc->sc_flags = 0;
1548 		if (rf->rf_flags & LAGG_F_HASHL2)
1549 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1550 		if (rf->rf_flags & LAGG_F_HASHL3)
1551 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1552 		if (rf->rf_flags & LAGG_F_HASHL4)
1553 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1554 		LAGG_XUNLOCK(sc);
1555 		break;
1556 	case SIOCGLAGGPORT:
1557 		if (rp->rp_portname[0] == '\0' ||
1558 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1559 			error = EINVAL;
1560 			break;
1561 		}
1562 
1563 		NET_EPOCH_ENTER(et);
1564 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1565 		    lp->lp_softc != sc) {
1566 			error = ENOENT;
1567 			NET_EPOCH_EXIT(et);
1568 			if_rele(tpif);
1569 			break;
1570 		}
1571 
1572 		lagg_port2req(lp, rp);
1573 		NET_EPOCH_EXIT(et);
1574 		if_rele(tpif);
1575 		break;
1576 	case SIOCSLAGGPORT:
1577 		error = priv_check(td, PRIV_NET_LAGG);
1578 		if (error)
1579 			break;
1580 		if (rp->rp_portname[0] == '\0' ||
1581 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1582 			error = EINVAL;
1583 			break;
1584 		}
1585 #ifdef INET6
1586 		/*
1587 		 * A laggport interface should not have inet6 address
1588 		 * because two interfaces with a valid link-local
1589 		 * scope zone must not be merged in any form.  This
1590 		 * restriction is needed to prevent violation of
1591 		 * link-local scope zone.  Attempts to add a laggport
1592 		 * interface which has inet6 addresses triggers
1593 		 * removal of all inet6 addresses on the member
1594 		 * interface.
1595 		 */
1596 		if (in6ifa_llaonifp(tpif)) {
1597 			in6_ifdetach(tpif);
1598 				if_printf(sc->sc_ifp,
1599 				    "IPv6 addresses on %s have been removed "
1600 				    "before adding it as a member to prevent "
1601 				    "IPv6 address scope violation.\n",
1602 				    tpif->if_xname);
1603 		}
1604 #endif
1605 		oldmtu = ifp->if_mtu;
1606 		LAGG_XLOCK(sc);
1607 		error = lagg_port_create(sc, tpif);
1608 		LAGG_XUNLOCK(sc);
1609 		if_rele(tpif);
1610 
1611 		/*
1612 		 * LAGG MTU may change during addition of the first port.
1613 		 * If it did, do network layer specific procedure.
1614 		 */
1615 		if (ifp->if_mtu != oldmtu) {
1616 #ifdef INET6
1617 			nd6_setmtu(ifp);
1618 #endif
1619 			rt_updatemtu(ifp);
1620 		}
1621 
1622 		VLAN_CAPABILITIES(ifp);
1623 		break;
1624 	case SIOCSLAGGDELPORT:
1625 		error = priv_check(td, PRIV_NET_LAGG);
1626 		if (error)
1627 			break;
1628 		if (rp->rp_portname[0] == '\0' ||
1629 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1630 			error = EINVAL;
1631 			break;
1632 		}
1633 
1634 		LAGG_XLOCK(sc);
1635 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1636 		    lp->lp_softc != sc) {
1637 			error = ENOENT;
1638 			LAGG_XUNLOCK(sc);
1639 			if_rele(tpif);
1640 			break;
1641 		}
1642 
1643 		error = lagg_port_destroy(lp, 1);
1644 		LAGG_XUNLOCK(sc);
1645 		if_rele(tpif);
1646 		VLAN_CAPABILITIES(ifp);
1647 		break;
1648 	case SIOCSIFFLAGS:
1649 		/* Set flags on ports too */
1650 		LAGG_XLOCK(sc);
1651 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1652 			lagg_setflags(lp, 1);
1653 		}
1654 
1655 		if (!(ifp->if_flags & IFF_UP) &&
1656 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1657 			/*
1658 			 * If interface is marked down and it is running,
1659 			 * then stop and disable it.
1660 			 */
1661 			lagg_stop(sc);
1662 			LAGG_XUNLOCK(sc);
1663 		} else if ((ifp->if_flags & IFF_UP) &&
1664 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1665 			/*
1666 			 * If interface is marked up and it is stopped, then
1667 			 * start it.
1668 			 */
1669 			LAGG_XUNLOCK(sc);
1670 			(*ifp->if_init)(sc);
1671 		} else
1672 			LAGG_XUNLOCK(sc);
1673 		break;
1674 	case SIOCADDMULTI:
1675 	case SIOCDELMULTI:
1676 		LAGG_XLOCK(sc);
1677 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1678 			lagg_clrmulti(lp);
1679 			lagg_setmulti(lp);
1680 		}
1681 		LAGG_XUNLOCK(sc);
1682 		error = 0;
1683 		break;
1684 	case SIOCSIFMEDIA:
1685 	case SIOCGIFMEDIA:
1686 		if (ifp->if_type == IFT_INFINIBAND)
1687 			error = EINVAL;
1688 		else
1689 			error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1690 		break;
1691 
1692 	case SIOCSIFCAP:
1693 		LAGG_XLOCK(sc);
1694 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1695 			if (lp->lp_ioctl != NULL)
1696 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1697 		}
1698 		lagg_capabilities(sc);
1699 		LAGG_XUNLOCK(sc);
1700 		VLAN_CAPABILITIES(ifp);
1701 		error = 0;
1702 		break;
1703 
1704 	case SIOCSIFMTU:
1705 		LAGG_XLOCK(sc);
1706 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1707 			if (lp->lp_ioctl != NULL)
1708 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1709 			else
1710 				error = EINVAL;
1711 			if (error != 0) {
1712 				if_printf(ifp,
1713 				    "failed to change MTU to %d on port %s, "
1714 				    "reverting all ports to original MTU (%d)\n",
1715 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1716 				break;
1717 			}
1718 		}
1719 		if (error == 0) {
1720 			ifp->if_mtu = ifr->ifr_mtu;
1721 		} else {
1722 			/* set every port back to the original MTU */
1723 			ifr->ifr_mtu = ifp->if_mtu;
1724 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1725 				if (lp->lp_ioctl != NULL)
1726 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1727 			}
1728 		}
1729 		LAGG_XUNLOCK(sc);
1730 		break;
1731 
1732 	default:
1733 		error = ether_ioctl(ifp, cmd, data);
1734 		break;
1735 	}
1736 	return (error);
1737 }
1738 
1739 #if defined(KERN_TLS) || defined(RATELIMIT)
1740 #ifdef RATELIMIT
1741 static const struct if_snd_tag_sw lagg_snd_tag_ul_sw = {
1742 	.snd_tag_modify = lagg_snd_tag_modify,
1743 	.snd_tag_query = lagg_snd_tag_query,
1744 	.snd_tag_free = lagg_snd_tag_free,
1745 	.next_snd_tag = lagg_next_snd_tag,
1746 	.type = IF_SND_TAG_TYPE_UNLIMITED
1747 };
1748 
1749 static const struct if_snd_tag_sw lagg_snd_tag_rl_sw = {
1750 	.snd_tag_modify = lagg_snd_tag_modify,
1751 	.snd_tag_query = lagg_snd_tag_query,
1752 	.snd_tag_free = lagg_snd_tag_free,
1753 	.next_snd_tag = lagg_next_snd_tag,
1754 	.type = IF_SND_TAG_TYPE_RATE_LIMIT
1755 };
1756 #endif
1757 
1758 #ifdef KERN_TLS
1759 static const struct if_snd_tag_sw lagg_snd_tag_tls_sw = {
1760 	.snd_tag_modify = lagg_snd_tag_modify,
1761 	.snd_tag_query = lagg_snd_tag_query,
1762 	.snd_tag_free = lagg_snd_tag_free,
1763 	.next_snd_tag = lagg_next_snd_tag,
1764 	.type = IF_SND_TAG_TYPE_TLS
1765 };
1766 
1767 #ifdef RATELIMIT
1768 static const struct if_snd_tag_sw lagg_snd_tag_tls_rl_sw = {
1769 	.snd_tag_modify = lagg_snd_tag_modify,
1770 	.snd_tag_query = lagg_snd_tag_query,
1771 	.snd_tag_free = lagg_snd_tag_free,
1772 	.next_snd_tag = lagg_next_snd_tag,
1773 	.type = IF_SND_TAG_TYPE_TLS_RATE_LIMIT
1774 };
1775 #endif
1776 #endif
1777 
1778 static inline struct lagg_snd_tag *
1779 mst_to_lst(struct m_snd_tag *mst)
1780 {
1781 
1782 	return (__containerof(mst, struct lagg_snd_tag, com));
1783 }
1784 
1785 /*
1786  * Look up the port used by a specific flow.  This only works for lagg
1787  * protocols with deterministic port mappings (e.g. not roundrobin).
1788  * In addition protocols which use a hash to map flows to ports must
1789  * be configured to use the mbuf flowid rather than hashing packet
1790  * contents.
1791  */
1792 static struct lagg_port *
1793 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1794     uint8_t numa_domain)
1795 {
1796 	struct lagg_softc *sc;
1797 	struct lagg_port *lp;
1798 	struct lagg_lb *lb;
1799 	uint32_t hash, p;
1800 	int err;
1801 
1802 	sc = ifp->if_softc;
1803 
1804 	switch (sc->sc_proto) {
1805 	case LAGG_PROTO_FAILOVER:
1806 		return (lagg_link_active(sc, sc->sc_primary));
1807 	case LAGG_PROTO_LOADBALANCE:
1808 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1809 		    flowtype == M_HASHTYPE_NONE)
1810 			return (NULL);
1811 		p = flowid >> sc->flowid_shift;
1812 		p %= sc->sc_count;
1813 		lb = (struct lagg_lb *)sc->sc_psc;
1814 		lp = lb->lb_ports[p];
1815 		return (lagg_link_active(sc, lp));
1816 	case LAGG_PROTO_LACP:
1817 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1818 		    flowtype == M_HASHTYPE_NONE)
1819 			return (NULL);
1820 		hash = flowid >> sc->flowid_shift;
1821 		return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, &err));
1822 	default:
1823 		return (NULL);
1824 	}
1825 }
1826 
1827 static int
1828 lagg_snd_tag_alloc(struct ifnet *ifp,
1829     union if_snd_tag_alloc_params *params,
1830     struct m_snd_tag **ppmt)
1831 {
1832 	struct epoch_tracker et;
1833 	const struct if_snd_tag_sw *sw;
1834 	struct lagg_snd_tag *lst;
1835 	struct lagg_softc *sc;
1836 	struct lagg_port *lp;
1837 	struct ifnet *lp_ifp;
1838 	int error;
1839 
1840 	sc = ifp->if_softc;
1841 
1842 	switch (params->hdr.type) {
1843 #ifdef RATELIMIT
1844 	case IF_SND_TAG_TYPE_UNLIMITED:
1845 		sw = &lagg_snd_tag_ul_sw;
1846 		break;
1847 	case IF_SND_TAG_TYPE_RATE_LIMIT:
1848 		sw = &lagg_snd_tag_rl_sw;
1849 		break;
1850 #endif
1851 #ifdef KERN_TLS
1852 	case IF_SND_TAG_TYPE_TLS:
1853 		sw = &lagg_snd_tag_tls_sw;
1854 		break;
1855 #ifdef RATELIMIT
1856 	case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
1857 		sw = &lagg_snd_tag_tls_rl_sw;
1858 		break;
1859 #endif
1860 #endif
1861 	default:
1862 		return (EOPNOTSUPP);
1863 	}
1864 
1865 	NET_EPOCH_ENTER(et);
1866 	lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1867 	    params->hdr.flowtype, params->hdr.numa_domain);
1868 	if (lp == NULL) {
1869 		NET_EPOCH_EXIT(et);
1870 		return (EOPNOTSUPP);
1871 	}
1872 	if (lp->lp_ifp == NULL) {
1873 		NET_EPOCH_EXIT(et);
1874 		return (EOPNOTSUPP);
1875 	}
1876 	lp_ifp = lp->lp_ifp;
1877 	if_ref(lp_ifp);
1878 	NET_EPOCH_EXIT(et);
1879 
1880 	lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1881 	if (lst == NULL) {
1882 		if_rele(lp_ifp);
1883 		return (ENOMEM);
1884 	}
1885 
1886 	error = m_snd_tag_alloc(lp_ifp, params, &lst->tag);
1887 	if_rele(lp_ifp);
1888 	if (error) {
1889 		free(lst, M_LAGG);
1890 		return (error);
1891 	}
1892 
1893 	m_snd_tag_init(&lst->com, ifp, sw);
1894 
1895 	*ppmt = &lst->com;
1896 	return (0);
1897 }
1898 
1899 static struct m_snd_tag *
1900 lagg_next_snd_tag(struct m_snd_tag *mst)
1901 {
1902 	struct lagg_snd_tag *lst;
1903 
1904 	lst = mst_to_lst(mst);
1905 	return (lst->tag);
1906 }
1907 
1908 static int
1909 lagg_snd_tag_modify(struct m_snd_tag *mst,
1910     union if_snd_tag_modify_params *params)
1911 {
1912 	struct lagg_snd_tag *lst;
1913 
1914 	lst = mst_to_lst(mst);
1915 	return (lst->tag->sw->snd_tag_modify(lst->tag, params));
1916 }
1917 
1918 static int
1919 lagg_snd_tag_query(struct m_snd_tag *mst,
1920     union if_snd_tag_query_params *params)
1921 {
1922 	struct lagg_snd_tag *lst;
1923 
1924 	lst = mst_to_lst(mst);
1925 	return (lst->tag->sw->snd_tag_query(lst->tag, params));
1926 }
1927 
1928 static void
1929 lagg_snd_tag_free(struct m_snd_tag *mst)
1930 {
1931 	struct lagg_snd_tag *lst;
1932 
1933 	lst = mst_to_lst(mst);
1934 	m_snd_tag_rele(lst->tag);
1935 	free(lst, M_LAGG);
1936 }
1937 
1938 static void
1939 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1940 {
1941 	/*
1942 	 * For lagg, we have an indirect
1943 	 * interface. The caller needs to
1944 	 * get a ratelimit tag on the actual
1945 	 * interface the flow will go on.
1946 	 */
1947 	q->rate_table = NULL;
1948 	q->flags = RT_IS_INDIRECT;
1949 	q->max_flows = 0;
1950 	q->number_of_rates = 0;
1951 }
1952 #endif
1953 
1954 static int
1955 lagg_setmulti(struct lagg_port *lp)
1956 {
1957 	struct lagg_softc *sc = lp->lp_softc;
1958 	struct ifnet *ifp = lp->lp_ifp;
1959 	struct ifnet *scifp = sc->sc_ifp;
1960 	struct lagg_mc *mc;
1961 	struct ifmultiaddr *ifma;
1962 	int error;
1963 
1964 	IF_ADDR_WLOCK(scifp);
1965 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1966 		if (ifma->ifma_addr->sa_family != AF_LINK)
1967 			continue;
1968 		mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
1969 		if (mc == NULL) {
1970 			IF_ADDR_WUNLOCK(scifp);
1971 			return (ENOMEM);
1972 		}
1973 		bcopy(ifma->ifma_addr, &mc->mc_addr,
1974 		    ifma->ifma_addr->sa_len);
1975 		mc->mc_addr.sdl_index = ifp->if_index;
1976 		mc->mc_ifma = NULL;
1977 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1978 	}
1979 	IF_ADDR_WUNLOCK(scifp);
1980 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1981 		error = if_addmulti(ifp,
1982 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1983 		if (error)
1984 			return (error);
1985 	}
1986 	return (0);
1987 }
1988 
1989 static int
1990 lagg_clrmulti(struct lagg_port *lp)
1991 {
1992 	struct lagg_mc *mc;
1993 
1994 	LAGG_XLOCK_ASSERT(lp->lp_softc);
1995 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1996 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1997 		if (mc->mc_ifma && lp->lp_detaching == 0)
1998 			if_delmulti_ifma(mc->mc_ifma);
1999 		free(mc, M_LAGG);
2000 	}
2001 	return (0);
2002 }
2003 
2004 static int
2005 lagg_setcaps(struct lagg_port *lp, int cap)
2006 {
2007 	struct ifreq ifr;
2008 
2009 	if (lp->lp_ifp->if_capenable == cap)
2010 		return (0);
2011 	if (lp->lp_ioctl == NULL)
2012 		return (ENXIO);
2013 	ifr.ifr_reqcap = cap;
2014 	return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
2015 }
2016 
2017 /* Handle a ref counted flag that should be set on the lagg port as well */
2018 static int
2019 lagg_setflag(struct lagg_port *lp, int flag, int status,
2020     int (*func)(struct ifnet *, int))
2021 {
2022 	struct lagg_softc *sc = lp->lp_softc;
2023 	struct ifnet *scifp = sc->sc_ifp;
2024 	struct ifnet *ifp = lp->lp_ifp;
2025 	int error;
2026 
2027 	LAGG_XLOCK_ASSERT(sc);
2028 
2029 	status = status ? (scifp->if_flags & flag) : 0;
2030 	/* Now "status" contains the flag value or 0 */
2031 
2032 	/*
2033 	 * See if recorded ports status is different from what
2034 	 * we want it to be.  If it is, flip it.  We record ports
2035 	 * status in lp_ifflags so that we won't clear ports flag
2036 	 * we haven't set.  In fact, we don't clear or set ports
2037 	 * flags directly, but get or release references to them.
2038 	 * That's why we can be sure that recorded flags still are
2039 	 * in accord with actual ports flags.
2040 	 */
2041 	if (status != (lp->lp_ifflags & flag)) {
2042 		error = (*func)(ifp, status);
2043 		if (error)
2044 			return (error);
2045 		lp->lp_ifflags &= ~flag;
2046 		lp->lp_ifflags |= status;
2047 	}
2048 	return (0);
2049 }
2050 
2051 /*
2052  * Handle IFF_* flags that require certain changes on the lagg port
2053  * if "status" is true, update ports flags respective to the lagg
2054  * if "status" is false, forcedly clear the flags set on port.
2055  */
2056 static int
2057 lagg_setflags(struct lagg_port *lp, int status)
2058 {
2059 	int error, i;
2060 
2061 	for (i = 0; lagg_pflags[i].flag; i++) {
2062 		error = lagg_setflag(lp, lagg_pflags[i].flag,
2063 		    status, lagg_pflags[i].func);
2064 		if (error)
2065 			return (error);
2066 	}
2067 	return (0);
2068 }
2069 
2070 static int
2071 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2072 {
2073 	struct epoch_tracker et;
2074 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2075 	int error;
2076 
2077 #if defined(KERN_TLS) || defined(RATELIMIT)
2078 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2079 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2080 #endif
2081 	NET_EPOCH_ENTER(et);
2082 	/* We need a Tx algorithm and at least one port */
2083 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2084 		NET_EPOCH_EXIT(et);
2085 		m_freem(m);
2086 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2087 		return (ENXIO);
2088 	}
2089 
2090 	ETHER_BPF_MTAP(ifp, m);
2091 
2092 	error = lagg_proto_start(sc, m);
2093 	NET_EPOCH_EXIT(et);
2094 	return (error);
2095 }
2096 
2097 static int
2098 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2099 {
2100 	struct epoch_tracker et;
2101 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2102 	int error;
2103 
2104 #if defined(KERN_TLS) || defined(RATELIMIT)
2105 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2106 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2107 #endif
2108 	NET_EPOCH_ENTER(et);
2109 	/* We need a Tx algorithm and at least one port */
2110 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2111 		NET_EPOCH_EXIT(et);
2112 		m_freem(m);
2113 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2114 		return (ENXIO);
2115 	}
2116 
2117 	INFINIBAND_BPF_MTAP(ifp, m);
2118 
2119 	error = lagg_proto_start(sc, m);
2120 	NET_EPOCH_EXIT(et);
2121 	return (error);
2122 }
2123 
2124 /*
2125  * The ifp->if_qflush entry point for lagg(4) is no-op.
2126  */
2127 static void
2128 lagg_qflush(struct ifnet *ifp __unused)
2129 {
2130 }
2131 
2132 static struct mbuf *
2133 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2134 {
2135 	struct epoch_tracker et;
2136 	struct lagg_port *lp = ifp->if_lagg;
2137 	struct lagg_softc *sc = lp->lp_softc;
2138 	struct ifnet *scifp = sc->sc_ifp;
2139 
2140 	NET_EPOCH_ENTER(et);
2141 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2142 	    lp->lp_detaching != 0 ||
2143 	    sc->sc_proto == LAGG_PROTO_NONE) {
2144 		NET_EPOCH_EXIT(et);
2145 		m_freem(m);
2146 		return (NULL);
2147 	}
2148 
2149 	ETHER_BPF_MTAP(scifp, m);
2150 
2151 	m = lagg_proto_input(sc, lp, m);
2152 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2153 		m_freem(m);
2154 		m = NULL;
2155 	}
2156 
2157 	NET_EPOCH_EXIT(et);
2158 	return (m);
2159 }
2160 
2161 static struct mbuf *
2162 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2163 {
2164 	struct epoch_tracker et;
2165 	struct lagg_port *lp = ifp->if_lagg;
2166 	struct lagg_softc *sc = lp->lp_softc;
2167 	struct ifnet *scifp = sc->sc_ifp;
2168 
2169 	NET_EPOCH_ENTER(et);
2170 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2171 	    lp->lp_detaching != 0 ||
2172 	    sc->sc_proto == LAGG_PROTO_NONE) {
2173 		NET_EPOCH_EXIT(et);
2174 		m_freem(m);
2175 		return (NULL);
2176 	}
2177 
2178 	INFINIBAND_BPF_MTAP(scifp, m);
2179 
2180 	m = lagg_proto_input(sc, lp, m);
2181 	if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2182 		m_freem(m);
2183 		m = NULL;
2184 	}
2185 
2186 	NET_EPOCH_EXIT(et);
2187 	return (m);
2188 }
2189 
2190 static int
2191 lagg_media_change(struct ifnet *ifp)
2192 {
2193 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2194 
2195 	if (sc->sc_ifflags & IFF_DEBUG)
2196 		printf("%s\n", __func__);
2197 
2198 	/* Ignore */
2199 	return (0);
2200 }
2201 
2202 static void
2203 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2204 {
2205 	struct epoch_tracker et;
2206 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2207 	struct lagg_port *lp;
2208 
2209 	imr->ifm_status = IFM_AVALID;
2210 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
2211 
2212 	NET_EPOCH_ENTER(et);
2213 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2214 		if (LAGG_PORTACTIVE(lp))
2215 			imr->ifm_status |= IFM_ACTIVE;
2216 	}
2217 	NET_EPOCH_EXIT(et);
2218 }
2219 
2220 static void
2221 lagg_linkstate(struct lagg_softc *sc)
2222 {
2223 	struct epoch_tracker et;
2224 	struct lagg_port *lp;
2225 	int new_link = LINK_STATE_DOWN;
2226 	uint64_t speed;
2227 
2228 	LAGG_XLOCK_ASSERT(sc);
2229 
2230 	/* LACP handles link state itself */
2231 	if (sc->sc_proto == LAGG_PROTO_LACP)
2232 		return;
2233 
2234 	/* Our link is considered up if at least one of our ports is active */
2235 	NET_EPOCH_ENTER(et);
2236 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2237 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2238 			new_link = LINK_STATE_UP;
2239 			break;
2240 		}
2241 	}
2242 	NET_EPOCH_EXIT(et);
2243 	if_link_state_change(sc->sc_ifp, new_link);
2244 
2245 	/* Update if_baudrate to reflect the max possible speed */
2246 	switch (sc->sc_proto) {
2247 		case LAGG_PROTO_FAILOVER:
2248 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2249 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
2250 			break;
2251 		case LAGG_PROTO_ROUNDROBIN:
2252 		case LAGG_PROTO_LOADBALANCE:
2253 		case LAGG_PROTO_BROADCAST:
2254 			speed = 0;
2255 			NET_EPOCH_ENTER(et);
2256 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2257 				speed += lp->lp_ifp->if_baudrate;
2258 			NET_EPOCH_EXIT(et);
2259 			sc->sc_ifp->if_baudrate = speed;
2260 			break;
2261 		case LAGG_PROTO_LACP:
2262 			/* LACP updates if_baudrate itself */
2263 			break;
2264 	}
2265 }
2266 
2267 static void
2268 lagg_port_state(struct ifnet *ifp, int state)
2269 {
2270 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2271 	struct lagg_softc *sc = NULL;
2272 
2273 	if (lp != NULL)
2274 		sc = lp->lp_softc;
2275 	if (sc == NULL)
2276 		return;
2277 
2278 	LAGG_XLOCK(sc);
2279 	lagg_linkstate(sc);
2280 	lagg_proto_linkstate(sc, lp);
2281 	LAGG_XUNLOCK(sc);
2282 }
2283 
2284 struct lagg_port *
2285 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2286 {
2287 	struct lagg_port *lp_next, *rval = NULL;
2288 
2289 	/*
2290 	 * Search a port which reports an active link state.
2291 	 */
2292 
2293 #ifdef INVARIANTS
2294 	/*
2295 	 * This is called with either in the network epoch
2296 	 * or with LAGG_XLOCK(sc) held.
2297 	 */
2298 	if (!in_epoch(net_epoch_preempt))
2299 		LAGG_XLOCK_ASSERT(sc);
2300 #endif
2301 
2302 	if (lp == NULL)
2303 		goto search;
2304 	if (LAGG_PORTACTIVE(lp)) {
2305 		rval = lp;
2306 		goto found;
2307 	}
2308 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2309 	    LAGG_PORTACTIVE(lp_next)) {
2310 		rval = lp_next;
2311 		goto found;
2312 	}
2313 
2314 search:
2315 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2316 		if (LAGG_PORTACTIVE(lp_next)) {
2317 			return (lp_next);
2318 		}
2319 	}
2320 found:
2321 	return (rval);
2322 }
2323 
2324 int
2325 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2326 {
2327 
2328 #if defined(KERN_TLS) || defined(RATELIMIT)
2329 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2330 		struct lagg_snd_tag *lst;
2331 		struct m_snd_tag *mst;
2332 
2333 		mst = m->m_pkthdr.snd_tag;
2334 		lst = mst_to_lst(mst);
2335 		if (lst->tag->ifp != ifp) {
2336 			m_freem(m);
2337 			return (EAGAIN);
2338 		}
2339 		m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2340 		m_snd_tag_rele(mst);
2341 	}
2342 #endif
2343 	return (ifp->if_transmit)(ifp, m);
2344 }
2345 
2346 /*
2347  * Simple round robin aggregation
2348  */
2349 static void
2350 lagg_rr_attach(struct lagg_softc *sc)
2351 {
2352 	sc->sc_seq = 0;
2353 	sc->sc_stride = 1;
2354 }
2355 
2356 static int
2357 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2358 {
2359 	struct lagg_port *lp;
2360 	uint32_t p;
2361 
2362 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
2363 	p /= sc->sc_stride;
2364 	p %= sc->sc_count;
2365 	lp = CK_SLIST_FIRST(&sc->sc_ports);
2366 
2367 	while (p--)
2368 		lp = CK_SLIST_NEXT(lp, lp_entries);
2369 
2370 	/*
2371 	 * Check the port's link state. This will return the next active
2372 	 * port if the link is down or the port is NULL.
2373 	 */
2374 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2375 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2376 		m_freem(m);
2377 		return (ENETDOWN);
2378 	}
2379 
2380 	/* Send mbuf */
2381 	return (lagg_enqueue(lp->lp_ifp, m));
2382 }
2383 
2384 static struct mbuf *
2385 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2386 {
2387 	struct ifnet *ifp = sc->sc_ifp;
2388 
2389 	/* Just pass in the packet to our lagg device */
2390 	m->m_pkthdr.rcvif = ifp;
2391 
2392 	return (m);
2393 }
2394 
2395 /*
2396  * Broadcast mode
2397  */
2398 static int
2399 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2400 {
2401 	int active_ports = 0;
2402 	int errors = 0;
2403 	int ret;
2404 	struct lagg_port *lp, *last = NULL;
2405 	struct mbuf *m0;
2406 
2407 	NET_EPOCH_ASSERT();
2408 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2409 		if (!LAGG_PORTACTIVE(lp))
2410 			continue;
2411 
2412 		active_ports++;
2413 
2414 		if (last != NULL) {
2415 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2416 			if (m0 == NULL) {
2417 				ret = ENOBUFS;
2418 				errors++;
2419 				break;
2420 			}
2421 			lagg_enqueue(last->lp_ifp, m0);
2422 		}
2423 		last = lp;
2424 	}
2425 
2426 	if (last == NULL) {
2427 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2428 		m_freem(m);
2429 		return (ENOENT);
2430 	}
2431 	if ((last = lagg_link_active(sc, last)) == NULL) {
2432 		errors++;
2433 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2434 		m_freem(m);
2435 		return (ENETDOWN);
2436 	}
2437 
2438 	ret = lagg_enqueue(last->lp_ifp, m);
2439 	if (errors != 0)
2440 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2441 
2442 	return (ret);
2443 }
2444 
2445 static struct mbuf*
2446 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2447 {
2448 	struct ifnet *ifp = sc->sc_ifp;
2449 
2450 	/* Just pass in the packet to our lagg device */
2451 	m->m_pkthdr.rcvif = ifp;
2452 	return (m);
2453 }
2454 
2455 /*
2456  * Active failover
2457  */
2458 static int
2459 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2460 {
2461 	struct lagg_port *lp;
2462 
2463 	/* Use the master port if active or the next available port */
2464 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2465 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2466 		m_freem(m);
2467 		return (ENETDOWN);
2468 	}
2469 
2470 	/* Send mbuf */
2471 	return (lagg_enqueue(lp->lp_ifp, m));
2472 }
2473 
2474 static struct mbuf *
2475 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2476 {
2477 	struct ifnet *ifp = sc->sc_ifp;
2478 	struct lagg_port *tmp_tp;
2479 
2480 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2481 		m->m_pkthdr.rcvif = ifp;
2482 		return (m);
2483 	}
2484 
2485 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2486 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2487 		/*
2488 		 * If tmp_tp is null, we've received a packet when all
2489 		 * our links are down. Weird, but process it anyways.
2490 		 */
2491 		if ((tmp_tp == NULL || tmp_tp == lp)) {
2492 			m->m_pkthdr.rcvif = ifp;
2493 			return (m);
2494 		}
2495 	}
2496 
2497 	m_freem(m);
2498 	return (NULL);
2499 }
2500 
2501 /*
2502  * Loadbalancing
2503  */
2504 static void
2505 lagg_lb_attach(struct lagg_softc *sc)
2506 {
2507 	struct lagg_port *lp;
2508 	struct lagg_lb *lb;
2509 
2510 	LAGG_XLOCK_ASSERT(sc);
2511 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2512 	lb->lb_key = m_ether_tcpip_hash_init();
2513 	sc->sc_psc = lb;
2514 
2515 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2516 		lagg_lb_port_create(lp);
2517 }
2518 
2519 static void
2520 lagg_lb_detach(struct lagg_softc *sc)
2521 {
2522 	struct lagg_lb *lb;
2523 
2524 	lb = (struct lagg_lb *)sc->sc_psc;
2525 	if (lb != NULL)
2526 		free(lb, M_LAGG);
2527 }
2528 
2529 static int
2530 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2531 {
2532 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2533 	struct lagg_port *lp_next;
2534 	int i = 0, rv;
2535 
2536 	rv = 0;
2537 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2538 	LAGG_XLOCK_ASSERT(sc);
2539 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2540 		if (lp_next == lp)
2541 			continue;
2542 		if (i >= LAGG_MAX_PORTS) {
2543 			rv = EINVAL;
2544 			break;
2545 		}
2546 		if (sc->sc_ifflags & IFF_DEBUG)
2547 			printf("%s: port %s at index %d\n",
2548 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2549 		lb->lb_ports[i++] = lp_next;
2550 	}
2551 
2552 	return (rv);
2553 }
2554 
2555 static int
2556 lagg_lb_port_create(struct lagg_port *lp)
2557 {
2558 	struct lagg_softc *sc = lp->lp_softc;
2559 	return (lagg_lb_porttable(sc, NULL));
2560 }
2561 
2562 static void
2563 lagg_lb_port_destroy(struct lagg_port *lp)
2564 {
2565 	struct lagg_softc *sc = lp->lp_softc;
2566 	lagg_lb_porttable(sc, lp);
2567 }
2568 
2569 static int
2570 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2571 {
2572 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2573 	struct lagg_port *lp = NULL;
2574 	uint32_t p = 0;
2575 
2576 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2577 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2578 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2579 	else
2580 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2581 	p %= sc->sc_count;
2582 	lp = lb->lb_ports[p];
2583 
2584 	/*
2585 	 * Check the port's link state. This will return the next active
2586 	 * port if the link is down or the port is NULL.
2587 	 */
2588 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2589 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2590 		m_freem(m);
2591 		return (ENETDOWN);
2592 	}
2593 
2594 	/* Send mbuf */
2595 	return (lagg_enqueue(lp->lp_ifp, m));
2596 }
2597 
2598 static struct mbuf *
2599 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2600 {
2601 	struct ifnet *ifp = sc->sc_ifp;
2602 
2603 	/* Just pass in the packet to our lagg device */
2604 	m->m_pkthdr.rcvif = ifp;
2605 
2606 	return (m);
2607 }
2608 
2609 /*
2610  * 802.3ad LACP
2611  */
2612 static void
2613 lagg_lacp_attach(struct lagg_softc *sc)
2614 {
2615 	struct lagg_port *lp;
2616 
2617 	lacp_attach(sc);
2618 	LAGG_XLOCK_ASSERT(sc);
2619 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2620 		lacp_port_create(lp);
2621 }
2622 
2623 static void
2624 lagg_lacp_detach(struct lagg_softc *sc)
2625 {
2626 	struct lagg_port *lp;
2627 	void *psc;
2628 
2629 	LAGG_XLOCK_ASSERT(sc);
2630 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2631 		lacp_port_destroy(lp);
2632 
2633 	psc = sc->sc_psc;
2634 	sc->sc_psc = NULL;
2635 	lacp_detach(psc);
2636 }
2637 
2638 static void
2639 lagg_lacp_lladdr(struct lagg_softc *sc)
2640 {
2641 	struct lagg_port *lp;
2642 
2643 	LAGG_SXLOCK_ASSERT(sc);
2644 
2645 	/* purge all the lacp ports */
2646 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2647 		lacp_port_destroy(lp);
2648 
2649 	/* add them back in */
2650 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2651 		lacp_port_create(lp);
2652 }
2653 
2654 static int
2655 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2656 {
2657 	struct lagg_port *lp;
2658 	int err;
2659 
2660 	lp = lacp_select_tx_port(sc, m, &err);
2661 	if (lp == NULL) {
2662 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2663 		m_freem(m);
2664 		return (err);
2665 	}
2666 
2667 	/* Send mbuf */
2668 	return (lagg_enqueue(lp->lp_ifp, m));
2669 }
2670 
2671 static struct mbuf *
2672 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2673 {
2674 	struct ifnet *ifp = sc->sc_ifp;
2675 	struct ether_header *eh;
2676 	u_short etype;
2677 
2678 	eh = mtod(m, struct ether_header *);
2679 	etype = ntohs(eh->ether_type);
2680 
2681 	/* Tap off LACP control messages */
2682 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2683 		m = lacp_input(lp, m);
2684 		if (m == NULL)
2685 			return (NULL);
2686 	}
2687 
2688 	/*
2689 	 * If the port is not collecting or not in the active aggregator then
2690 	 * free and return.
2691 	 */
2692 	if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2693 		m_freem(m);
2694 		return (NULL);
2695 	}
2696 
2697 	m->m_pkthdr.rcvif = ifp;
2698 	return (m);
2699 }
2700