xref: /freebsd/sys/net/if_lagg.c (revision e9ac41698b2f322d55ccf9da50a3596edb2c1800)
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 #include "opt_inet.h"
23 #include "opt_inet6.h"
24 #include "opt_kern_tls.h"
25 #include "opt_ratelimit.h"
26 
27 #include <sys/param.h>
28 #include <sys/kernel.h>
29 #include <sys/malloc.h>
30 #include <sys/mbuf.h>
31 #include <sys/queue.h>
32 #include <sys/socket.h>
33 #include <sys/sockio.h>
34 #include <sys/sysctl.h>
35 #include <sys/module.h>
36 #include <sys/priv.h>
37 #include <sys/systm.h>
38 #include <sys/proc.h>
39 #include <sys/lock.h>
40 #include <sys/rmlock.h>
41 #include <sys/sx.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/if_private.h>
54 #include <net/bpf.h>
55 #include <net/route.h>
56 #include <net/vnet.h>
57 #include <net/infiniband.h>
58 
59 #if defined(INET) || defined(INET6)
60 #include <netinet/in.h>
61 #include <netinet/ip.h>
62 #endif
63 #ifdef INET
64 #include <netinet/in_systm.h>
65 #include <netinet/if_ether.h>
66 #endif
67 
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet6/in6_ifattach.h>
72 #endif
73 
74 #include <net/if_vlan_var.h>
75 #include <net/if_lagg.h>
76 #include <net/ieee8023ad_lacp.h>
77 
78 #ifdef DEV_NETMAP
79 MODULE_DEPEND(if_lagg, netmap, 1, 1, 1);
80 #endif
81 
82 #define	LAGG_SX_INIT(_sc)	sx_init(&(_sc)->sc_sx, "if_lagg sx")
83 #define	LAGG_SX_DESTROY(_sc)	sx_destroy(&(_sc)->sc_sx)
84 #define	LAGG_XLOCK(_sc)		sx_xlock(&(_sc)->sc_sx)
85 #define	LAGG_XUNLOCK(_sc)	sx_xunlock(&(_sc)->sc_sx)
86 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
87 #define	LAGG_SLOCK(_sc)		sx_slock(&(_sc)->sc_sx)
88 #define	LAGG_SUNLOCK(_sc)	sx_sunlock(&(_sc)->sc_sx)
89 #define	LAGG_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
90 
91 /* Special flags we should propagate to the lagg ports. */
92 static struct {
93 	int flag;
94 	int (*func)(struct ifnet *, int);
95 } lagg_pflags[] = {
96 	{IFF_PROMISC, ifpromisc},
97 	{IFF_ALLMULTI, if_allmulti},
98 	{0, NULL}
99 };
100 
101 struct lagg_snd_tag {
102 	struct m_snd_tag com;
103 	struct m_snd_tag *tag;
104 };
105 
106 VNET_DEFINE_STATIC(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
107 #define	V_lagg_list	VNET(lagg_list)
108 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
109 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
110 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
111 					"if_lagg list", NULL, MTX_DEF)
112 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
113 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
114 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
115 static eventhandler_tag	lagg_detach_cookie = NULL;
116 
117 static int	lagg_clone_create(struct if_clone *, char *, size_t,
118 		    struct ifc_data *, struct ifnet **);
119 static int	lagg_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
120 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
121 #define	V_lagg_cloner	VNET(lagg_cloner)
122 static const char laggname[] = "lagg";
123 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
124 
125 static void	lagg_capabilities(struct lagg_softc *);
126 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
127 static int	lagg_port_destroy(struct lagg_port *, int);
128 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *);
129 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *);
130 static void	lagg_linkstate(struct lagg_softc *);
131 static void	lagg_port_state(struct ifnet *, int);
132 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
133 static int	lagg_port_output(struct ifnet *, struct mbuf *,
134 		    const struct sockaddr *, struct route *);
135 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
136 #ifdef LAGG_PORT_STACKING
137 static int	lagg_port_checkstacking(struct lagg_softc *);
138 #endif
139 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
140 static void	lagg_if_updown(struct lagg_softc *, bool);
141 static void	lagg_init(void *);
142 static void	lagg_stop(struct lagg_softc *);
143 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
144 #if defined(KERN_TLS) || defined(RATELIMIT)
145 static int	lagg_snd_tag_alloc(struct ifnet *,
146 		    union if_snd_tag_alloc_params *,
147 		    struct m_snd_tag **);
148 static int	lagg_snd_tag_modify(struct m_snd_tag *,
149 		    union if_snd_tag_modify_params *);
150 static int	lagg_snd_tag_query(struct m_snd_tag *,
151 		    union if_snd_tag_query_params *);
152 static void	lagg_snd_tag_free(struct m_snd_tag *);
153 static struct m_snd_tag *lagg_next_snd_tag(struct m_snd_tag *);
154 static void	lagg_ratelimit_query(struct ifnet *,
155 		    struct if_ratelimit_query_results *);
156 #endif
157 static int	lagg_setmulti(struct lagg_port *);
158 static int	lagg_clrmulti(struct lagg_port *);
159 static void	lagg_setcaps(struct lagg_port *, int cap, int cap2);
160 static int	lagg_setflag(struct lagg_port *, int, int,
161 		    int (*func)(struct ifnet *, int));
162 static int	lagg_setflags(struct lagg_port *, int status);
163 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
164 static int	lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
165 static int	lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
166 static void	lagg_qflush(struct ifnet *);
167 static int	lagg_media_change(struct ifnet *);
168 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
169 static struct lagg_port *lagg_link_active(struct lagg_softc *,
170 		    struct lagg_port *);
171 
172 /* Simple round robin */
173 static void	lagg_rr_attach(struct lagg_softc *);
174 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
175 
176 /* Active failover */
177 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
178 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
179 		    struct mbuf *);
180 
181 /* Loadbalancing */
182 static void	lagg_lb_attach(struct lagg_softc *);
183 static void	lagg_lb_detach(struct lagg_softc *);
184 static int	lagg_lb_port_create(struct lagg_port *);
185 static void	lagg_lb_port_destroy(struct lagg_port *);
186 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
187 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
188 
189 /* Broadcast */
190 static int	lagg_bcast_start(struct lagg_softc *, struct mbuf *);
191 
192 /* 802.3ad LACP */
193 static void	lagg_lacp_attach(struct lagg_softc *);
194 static void	lagg_lacp_detach(struct lagg_softc *);
195 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
196 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
197 		    struct mbuf *);
198 static void	lagg_lacp_lladdr(struct lagg_softc *);
199 
200 /* Default input */
201 static struct mbuf *lagg_default_input(struct lagg_softc *, struct lagg_port *,
202 		    struct mbuf *);
203 
204 /* lagg protocol table */
205 static const struct lagg_proto {
206 	lagg_proto	pr_num;
207 	void		(*pr_attach)(struct lagg_softc *);
208 	void		(*pr_detach)(struct lagg_softc *);
209 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
210 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
211 			    struct mbuf *);
212 	int		(*pr_addport)(struct lagg_port *);
213 	void		(*pr_delport)(struct lagg_port *);
214 	void		(*pr_linkstate)(struct lagg_port *);
215 	void 		(*pr_init)(struct lagg_softc *);
216 	void 		(*pr_stop)(struct lagg_softc *);
217 	void 		(*pr_lladdr)(struct lagg_softc *);
218 	void		(*pr_request)(struct lagg_softc *, void *);
219 	void		(*pr_portreq)(struct lagg_port *, void *);
220 } lagg_protos[] = {
221     {
222 	.pr_num = LAGG_PROTO_NONE
223     },
224     {
225 	.pr_num = LAGG_PROTO_ROUNDROBIN,
226 	.pr_attach = lagg_rr_attach,
227 	.pr_start = lagg_rr_start,
228 	.pr_input = lagg_default_input,
229     },
230     {
231 	.pr_num = LAGG_PROTO_FAILOVER,
232 	.pr_start = lagg_fail_start,
233 	.pr_input = lagg_fail_input,
234     },
235     {
236 	.pr_num = LAGG_PROTO_LOADBALANCE,
237 	.pr_attach = lagg_lb_attach,
238 	.pr_detach = lagg_lb_detach,
239 	.pr_start = lagg_lb_start,
240 	.pr_input = lagg_default_input,
241 	.pr_addport = lagg_lb_port_create,
242 	.pr_delport = lagg_lb_port_destroy,
243     },
244     {
245 	.pr_num = LAGG_PROTO_LACP,
246 	.pr_attach = lagg_lacp_attach,
247 	.pr_detach = lagg_lacp_detach,
248 	.pr_start = lagg_lacp_start,
249 	.pr_input = lagg_lacp_input,
250 	.pr_addport = lacp_port_create,
251 	.pr_delport = lacp_port_destroy,
252 	.pr_linkstate = lacp_linkstate,
253 	.pr_init = lacp_init,
254 	.pr_stop = lacp_stop,
255 	.pr_lladdr = lagg_lacp_lladdr,
256 	.pr_request = lacp_req,
257 	.pr_portreq = lacp_portreq,
258     },
259     {
260 	.pr_num = LAGG_PROTO_BROADCAST,
261 	.pr_start = lagg_bcast_start,
262 	.pr_input = lagg_default_input,
263     },
264 };
265 
266 SYSCTL_DECL(_net_link);
267 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
268     "Link Aggregation");
269 
270 /* Allow input on any failover links */
271 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
272 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
273 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
274     &VNET_NAME(lagg_failover_rx_all), 0,
275     "Accept input from any interface in a failover lagg");
276 
277 /* Default value for using flowid */
278 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
279 #define	V_def_use_flowid	VNET(def_use_flowid)
280 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid,
281     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_use_flowid), 0,
282     "Default setting for using flow id for load sharing");
283 
284 /* Default value for using numa */
285 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
286 #define	V_def_use_numa	VNET(def_use_numa)
287 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa,
288     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_use_numa), 0,
289     "Use numa to steer flows");
290 
291 /* Default value for flowid shift */
292 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
293 #define	V_def_flowid_shift	VNET(def_flowid_shift)
294 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift,
295     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_flowid_shift), 0,
296     "Default setting for flowid shift for load sharing");
297 
298 static void
299 vnet_lagg_init(const void *unused __unused)
300 {
301 
302 	LAGG_LIST_LOCK_INIT();
303 	SLIST_INIT(&V_lagg_list);
304 	struct if_clone_addreq req = {
305 		.create_f = lagg_clone_create,
306 		.destroy_f = lagg_clone_destroy,
307 		.flags = IFC_F_AUTOUNIT,
308 	};
309 	V_lagg_cloner = ifc_attach_cloner(laggname, &req);
310 }
311 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
312     vnet_lagg_init, NULL);
313 
314 static void
315 vnet_lagg_uninit(const void *unused __unused)
316 {
317 
318 	ifc_detach_cloner(V_lagg_cloner);
319 	LAGG_LIST_LOCK_DESTROY();
320 }
321 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
322     vnet_lagg_uninit, NULL);
323 
324 static int
325 lagg_modevent(module_t mod, int type, void *data)
326 {
327 
328 	switch (type) {
329 	case MOD_LOAD:
330 		lagg_input_ethernet_p = lagg_input_ethernet;
331 		lagg_input_infiniband_p = lagg_input_infiniband;
332 		lagg_linkstate_p = lagg_port_state;
333 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
334 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
335 		    EVENTHANDLER_PRI_ANY);
336 		break;
337 	case MOD_UNLOAD:
338 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
339 		    lagg_detach_cookie);
340 		lagg_input_ethernet_p = NULL;
341 		lagg_input_infiniband_p = NULL;
342 		lagg_linkstate_p = NULL;
343 		break;
344 	default:
345 		return (EOPNOTSUPP);
346 	}
347 	return (0);
348 }
349 
350 static moduledata_t lagg_mod = {
351 	"if_lagg",
352 	lagg_modevent,
353 	0
354 };
355 
356 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
357 MODULE_VERSION(if_lagg, 1);
358 MODULE_DEPEND(if_lagg, if_infiniband, 1, 1, 1);
359 
360 static void
361 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
362 {
363 
364 	LAGG_XLOCK_ASSERT(sc);
365 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
366 	    __func__, sc));
367 
368 	if (sc->sc_ifflags & IFF_DEBUG)
369 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
370 
371 	if (lagg_protos[pr].pr_attach != NULL)
372 		lagg_protos[pr].pr_attach(sc);
373 	sc->sc_proto = pr;
374 }
375 
376 static void
377 lagg_proto_detach(struct lagg_softc *sc)
378 {
379 	lagg_proto pr;
380 
381 	LAGG_XLOCK_ASSERT(sc);
382 	pr = sc->sc_proto;
383 	sc->sc_proto = LAGG_PROTO_NONE;
384 
385 	if (lagg_protos[pr].pr_detach != NULL)
386 		lagg_protos[pr].pr_detach(sc);
387 }
388 
389 static inline int
390 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
391 {
392 
393 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
394 }
395 
396 static inline struct mbuf *
397 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
398 {
399 
400 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
401 }
402 
403 static int
404 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
405 {
406 
407 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
408 		return (0);
409 	else
410 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
411 }
412 
413 static void
414 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
415 {
416 
417 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
418 		lagg_protos[sc->sc_proto].pr_delport(lp);
419 }
420 
421 static void
422 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
423 {
424 
425 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
426 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
427 }
428 
429 static void
430 lagg_proto_init(struct lagg_softc *sc)
431 {
432 
433 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
434 		lagg_protos[sc->sc_proto].pr_init(sc);
435 }
436 
437 static void
438 lagg_proto_stop(struct lagg_softc *sc)
439 {
440 
441 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
442 		lagg_protos[sc->sc_proto].pr_stop(sc);
443 }
444 
445 static void
446 lagg_proto_lladdr(struct lagg_softc *sc)
447 {
448 
449 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
450 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
451 }
452 
453 static void
454 lagg_proto_request(struct lagg_softc *sc, void *v)
455 {
456 
457 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
458 		lagg_protos[sc->sc_proto].pr_request(sc, v);
459 }
460 
461 static void
462 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
463 {
464 
465 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
466 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
467 }
468 
469 /*
470  * This routine is run via an vlan
471  * config EVENT
472  */
473 static void
474 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
475 {
476 	struct lagg_softc *sc = ifp->if_softc;
477 	struct lagg_port *lp;
478 
479 	if (ifp->if_softc != arg) /* Not our event */
480 		return;
481 
482 	LAGG_XLOCK(sc);
483 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
484 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
485 	LAGG_XUNLOCK(sc);
486 }
487 
488 /*
489  * This routine is run via an vlan
490  * unconfig EVENT
491  */
492 static void
493 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
494 {
495 	struct lagg_softc *sc = ifp->if_softc;
496 	struct lagg_port *lp;
497 
498 	if (ifp->if_softc != arg) /* Not our event */
499 		return;
500 
501 	LAGG_XLOCK(sc);
502 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
503 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
504 	LAGG_XUNLOCK(sc);
505 }
506 
507 static int
508 lagg_clone_create(struct if_clone *ifc, char *name, size_t len,
509     struct ifc_data *ifd, struct ifnet **ifpp)
510 {
511 	struct iflaggparam iflp;
512 	struct lagg_softc *sc;
513 	struct ifnet *ifp;
514 	int if_type;
515 	int error;
516 	static const uint8_t eaddr[LAGG_ADDR_LEN];
517 
518 	if (ifd->params != NULL) {
519 		error = ifc_copyin(ifd, &iflp, sizeof(iflp));
520 		if (error)
521 			return (error);
522 
523 		switch (iflp.lagg_type) {
524 		case LAGG_TYPE_ETHERNET:
525 			if_type = IFT_ETHER;
526 			break;
527 		case LAGG_TYPE_INFINIBAND:
528 			if_type = IFT_INFINIBAND;
529 			break;
530 		default:
531 			return (EINVAL);
532 		}
533 	} else {
534 		if_type = IFT_ETHER;
535 	}
536 
537 	sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK | M_ZERO);
538 	ifp = sc->sc_ifp = if_alloc(if_type);
539 	LAGG_SX_INIT(sc);
540 
541 	mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
542 	callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
543 
544 	LAGG_XLOCK(sc);
545 	if (V_def_use_flowid)
546 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
547 	if (V_def_use_numa)
548 		sc->sc_opts |= LAGG_OPT_USE_NUMA;
549 	sc->flowid_shift = V_def_flowid_shift;
550 
551 	/* Hash all layers by default */
552 	sc->sc_flags = MBUF_HASHFLAG_L2 | MBUF_HASHFLAG_L3 | MBUF_HASHFLAG_L4;
553 
554 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
555 
556 	CK_SLIST_INIT(&sc->sc_ports);
557 
558 	switch (if_type) {
559 	case IFT_ETHER:
560 		/* Initialise pseudo media types */
561 		ifmedia_init(&sc->sc_media, 0, lagg_media_change,
562 		    lagg_media_status);
563 		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
564 		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
565 
566 		if_initname(ifp, laggname, ifd->unit);
567 		ifp->if_transmit = lagg_transmit_ethernet;
568 		break;
569 	case IFT_INFINIBAND:
570 		if_initname(ifp, laggname, ifd->unit);
571 		ifp->if_transmit = lagg_transmit_infiniband;
572 		break;
573 	default:
574 		break;
575 	}
576 	ifp->if_softc = sc;
577 	ifp->if_qflush = lagg_qflush;
578 	ifp->if_init = lagg_init;
579 	ifp->if_ioctl = lagg_ioctl;
580 	ifp->if_get_counter = lagg_get_counter;
581 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
582 #if defined(KERN_TLS) || defined(RATELIMIT)
583 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
584 	ifp->if_ratelimit_query = lagg_ratelimit_query;
585 #endif
586 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
587 
588 	/*
589 	 * Attach as an ordinary ethernet device, children will be attached
590 	 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
591 	 */
592 	switch (if_type) {
593 	case IFT_ETHER:
594 		ether_ifattach(ifp, eaddr);
595 		break;
596 	case IFT_INFINIBAND:
597 		infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr);
598 		break;
599 	default:
600 		break;
601 	}
602 
603 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
604 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
605 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
606 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
607 
608 	/* Insert into the global list of laggs */
609 	LAGG_LIST_LOCK();
610 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
611 	LAGG_LIST_UNLOCK();
612 	LAGG_XUNLOCK(sc);
613 	*ifpp = ifp;
614 
615 	return (0);
616 }
617 
618 static int
619 lagg_clone_destroy(struct if_clone *ifc, struct ifnet *ifp, uint32_t flags)
620 {
621 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
622 	struct lagg_port *lp;
623 
624 	LAGG_XLOCK(sc);
625 	sc->sc_destroying = 1;
626 	lagg_stop(sc);
627 	ifp->if_flags &= ~IFF_UP;
628 
629 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
630 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
631 
632 	/* Shutdown and remove lagg ports */
633 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
634 		lagg_port_destroy(lp, 1);
635 
636 	/* Unhook the aggregation protocol */
637 	lagg_proto_detach(sc);
638 	LAGG_XUNLOCK(sc);
639 
640 	switch (ifp->if_type) {
641 	case IFT_ETHER:
642 		ether_ifdetach(ifp);
643 		ifmedia_removeall(&sc->sc_media);
644 		break;
645 	case IFT_INFINIBAND:
646 		infiniband_ifdetach(ifp);
647 		break;
648 	default:
649 		break;
650 	}
651 	if_free(ifp);
652 
653 	LAGG_LIST_LOCK();
654 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
655 	LAGG_LIST_UNLOCK();
656 
657 	mtx_destroy(&sc->sc_mtx);
658 	LAGG_SX_DESTROY(sc);
659 	free(sc, M_LAGG);
660 
661 	return (0);
662 }
663 
664 static void
665 lagg_capabilities(struct lagg_softc *sc)
666 {
667 	struct lagg_port *lp;
668 	int cap, cap2, ena, ena2, pena, pena2;
669 	uint64_t hwa;
670 	struct ifnet_hw_tsomax hw_tsomax;
671 
672 	LAGG_XLOCK_ASSERT(sc);
673 
674 	/* Get common enabled capabilities for the lagg ports */
675 	ena = ena2 = ~0;
676 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
677 		ena &= lp->lp_ifp->if_capenable;
678 		ena2 &= lp->lp_ifp->if_capenable2;
679 	}
680 	if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
681 		ena = ena2 = 0;
682 
683 	/*
684 	 * Apply common enabled capabilities back to the lagg ports.
685 	 * May require several iterations if they are dependent.
686 	 */
687 	do {
688 		pena = ena;
689 		pena2 = ena2;
690 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
691 			lagg_setcaps(lp, ena, ena2);
692 			ena &= lp->lp_ifp->if_capenable;
693 			ena2 &= lp->lp_ifp->if_capenable2;
694 		}
695 	} while (pena != ena || pena2 != ena2);
696 
697 	/* Get other capabilities from the lagg ports */
698 	cap = cap2 = ~0;
699 	hwa = ~(uint64_t)0;
700 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
701 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
702 		cap &= lp->lp_ifp->if_capabilities;
703 		cap2 &= lp->lp_ifp->if_capabilities2;
704 		hwa &= lp->lp_ifp->if_hwassist;
705 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
706 	}
707 	if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
708 		cap = cap2 = hwa = 0;
709 
710 	if (sc->sc_ifp->if_capabilities != cap ||
711 	    sc->sc_ifp->if_capenable != ena ||
712 	    sc->sc_ifp->if_capenable2 != ena2 ||
713 	    sc->sc_ifp->if_hwassist != hwa ||
714 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
715 		sc->sc_ifp->if_capabilities = cap;
716 		sc->sc_ifp->if_capabilities2 = cap2;
717 		sc->sc_ifp->if_capenable = ena;
718 		sc->sc_ifp->if_capenable2 = ena2;
719 		sc->sc_ifp->if_hwassist = hwa;
720 		getmicrotime(&sc->sc_ifp->if_lastchange);
721 
722 		if (sc->sc_ifflags & IFF_DEBUG)
723 			if_printf(sc->sc_ifp,
724 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
725 	}
726 }
727 
728 static int
729 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
730 {
731 	struct lagg_softc *sc_ptr;
732 	struct lagg_port *lp, *tlp;
733 	struct ifreq ifr;
734 	int error, i, oldmtu;
735 	int if_type;
736 	uint64_t *pval;
737 
738 	LAGG_XLOCK_ASSERT(sc);
739 
740 	if (sc->sc_ifp == ifp) {
741 		if_printf(sc->sc_ifp,
742 		    "cannot add a lagg to itself as a port\n");
743 		return (EINVAL);
744 	}
745 
746 	if (sc->sc_destroying == 1)
747 		return (ENXIO);
748 
749 	/* Limit the maximal number of lagg ports */
750 	if (sc->sc_count >= LAGG_MAX_PORTS)
751 		return (ENOSPC);
752 
753 	/* Check if port has already been associated to a lagg */
754 	if (ifp->if_lagg != NULL) {
755 		/* Port is already in the current lagg? */
756 		lp = (struct lagg_port *)ifp->if_lagg;
757 		if (lp->lp_softc == sc)
758 			return (EEXIST);
759 		return (EBUSY);
760 	}
761 
762 	switch (sc->sc_ifp->if_type) {
763 	case IFT_ETHER:
764 		/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
765 		if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
766 			return (EPROTONOSUPPORT);
767 		if_type = IFT_IEEE8023ADLAG;
768 		break;
769 	case IFT_INFINIBAND:
770 		/* XXX Disallow non-infiniband interfaces */
771 		if (ifp->if_type != IFT_INFINIBAND)
772 			return (EPROTONOSUPPORT);
773 		if_type = IFT_INFINIBANDLAG;
774 		break;
775 	default:
776 		break;
777 	}
778 
779 	/* Allow the first Ethernet member to define the MTU */
780 	oldmtu = -1;
781 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
782 		sc->sc_ifp->if_mtu = ifp->if_mtu;
783 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
784 		if (ifp->if_ioctl == NULL) {
785 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
786 			    ifp->if_xname);
787 			return (EINVAL);
788 		}
789 		oldmtu = ifp->if_mtu;
790 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
791 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
792 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
793 		if (error != 0) {
794 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
795 			    ifp->if_xname);
796 			return (error);
797 		}
798 		ifr.ifr_mtu = oldmtu;
799 	}
800 
801 	lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK | M_ZERO);
802 	lp->lp_softc = sc;
803 
804 	/* Check if port is a stacked lagg */
805 	LAGG_LIST_LOCK();
806 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
807 		if (ifp == sc_ptr->sc_ifp) {
808 			LAGG_LIST_UNLOCK();
809 			free(lp, M_LAGG);
810 			if (oldmtu != -1)
811 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
812 				    (caddr_t)&ifr);
813 			return (EINVAL);
814 			/* XXX disable stacking for the moment, its untested */
815 #ifdef LAGG_PORT_STACKING
816 			lp->lp_flags |= LAGG_PORT_STACK;
817 			if (lagg_port_checkstacking(sc_ptr) >=
818 			    LAGG_MAX_STACKING) {
819 				LAGG_LIST_UNLOCK();
820 				free(lp, M_LAGG);
821 				if (oldmtu != -1)
822 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
823 					    (caddr_t)&ifr);
824 				return (E2BIG);
825 			}
826 #endif
827 		}
828 	}
829 	LAGG_LIST_UNLOCK();
830 
831 	if_ref(ifp);
832 	lp->lp_ifp = ifp;
833 
834 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
835 	lp->lp_ifcapenable = ifp->if_capenable;
836 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
837 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
838 		lagg_proto_lladdr(sc);
839 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
840 	} else {
841 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
842 	}
843 	lagg_setflags(lp, 1);
844 
845 	if (CK_SLIST_EMPTY(&sc->sc_ports))
846 		sc->sc_primary = lp;
847 
848 	/* Change the interface type */
849 	lp->lp_iftype = ifp->if_type;
850 	ifp->if_type = if_type;
851 	ifp->if_lagg = lp;
852 	lp->lp_ioctl = ifp->if_ioctl;
853 	ifp->if_ioctl = lagg_port_ioctl;
854 	lp->lp_output = ifp->if_output;
855 	ifp->if_output = lagg_port_output;
856 
857 	/* Read port counters */
858 	pval = lp->port_counters.val;
859 	for (i = 0; i < IFCOUNTERS; i++, pval++)
860 		*pval = ifp->if_get_counter(ifp, i);
861 
862 	/*
863 	 * Insert into the list of ports.
864 	 * Keep ports sorted by if_index. It is handy, when configuration
865 	 * is predictable and `ifconfig laggN create ...` command
866 	 * will lead to the same result each time.
867 	 */
868 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
869 		if (tlp->lp_ifp->if_index < ifp->if_index && (
870 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
871 		    ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
872 		    ifp->if_index))
873 			break;
874 	}
875 	if (tlp != NULL)
876 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
877 	else
878 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
879 	sc->sc_count++;
880 
881 	lagg_setmulti(lp);
882 
883 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
884 		/* Remove the port, without calling pr_delport. */
885 		lagg_port_destroy(lp, 0);
886 		if (oldmtu != -1)
887 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
888 		return (error);
889 	}
890 
891 	/* Update lagg capabilities */
892 	lagg_capabilities(sc);
893 	lagg_linkstate(sc);
894 
895 	return (0);
896 }
897 
898 #ifdef LAGG_PORT_STACKING
899 static int
900 lagg_port_checkstacking(struct lagg_softc *sc)
901 {
902 	struct lagg_softc *sc_ptr;
903 	struct lagg_port *lp;
904 	int m = 0;
905 
906 	LAGG_SXLOCK_ASSERT(sc);
907 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
908 		if (lp->lp_flags & LAGG_PORT_STACK) {
909 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
910 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
911 		}
912 	}
913 
914 	return (m + 1);
915 }
916 #endif
917 
918 static void
919 lagg_port_destroy_cb(epoch_context_t ec)
920 {
921 	struct lagg_port *lp;
922 	struct ifnet *ifp;
923 
924 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
925 	ifp = lp->lp_ifp;
926 
927 	if_rele(ifp);
928 	free(lp, M_LAGG);
929 }
930 
931 static int
932 lagg_port_destroy(struct lagg_port *lp, int rundelport)
933 {
934 	struct lagg_softc *sc = lp->lp_softc;
935 	struct lagg_port *lp_ptr, *lp0;
936 	struct ifnet *ifp = lp->lp_ifp;
937 	uint64_t *pval, vdiff;
938 	int i;
939 
940 	LAGG_XLOCK_ASSERT(sc);
941 
942 	if (rundelport)
943 		lagg_proto_delport(sc, lp);
944 
945 	if (lp->lp_detaching == 0)
946 		lagg_clrmulti(lp);
947 
948 	/* Restore interface */
949 	ifp->if_type = lp->lp_iftype;
950 	ifp->if_ioctl = lp->lp_ioctl;
951 	ifp->if_output = lp->lp_output;
952 	ifp->if_lagg = NULL;
953 
954 	/* Update detached port counters */
955 	pval = lp->port_counters.val;
956 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
957 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
958 		sc->detached_counters.val[i] += vdiff;
959 	}
960 
961 	/* Finally, remove the port from the lagg */
962 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
963 	sc->sc_count--;
964 
965 	/* Update the primary interface */
966 	if (lp == sc->sc_primary) {
967 		uint8_t lladdr[LAGG_ADDR_LEN];
968 
969 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
970 			bzero(&lladdr, LAGG_ADDR_LEN);
971 		else
972 			bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
973 		sc->sc_primary = lp0;
974 		if (sc->sc_destroying == 0) {
975 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
976 			lagg_proto_lladdr(sc);
977 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
978 
979 			/*
980 			 * Update lladdr for each port (new primary needs update
981 			 * as well, to switch from old lladdr to its 'real' one).
982 			 * We can skip this if the lagg is being destroyed.
983 			 */
984 			CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
985 				if_setlladdr(lp_ptr->lp_ifp, lladdr,
986 				    lp_ptr->lp_ifp->if_addrlen);
987 		}
988 	}
989 
990 	if (lp->lp_ifflags)
991 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
992 
993 	if (lp->lp_detaching == 0) {
994 		lagg_setflags(lp, 0);
995 		lagg_setcaps(lp, lp->lp_ifcapenable, lp->lp_ifcapenable2);
996 		if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
997 	}
998 
999 	/*
1000 	 * free port and release it's ifnet reference after a grace period has
1001 	 * elapsed.
1002 	 */
1003 	NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
1004 	/* Update lagg capabilities */
1005 	lagg_capabilities(sc);
1006 	lagg_linkstate(sc);
1007 
1008 	return (0);
1009 }
1010 
1011 static int
1012 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1013 {
1014 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
1015 	struct lagg_softc *sc;
1016 	struct lagg_port *lp = NULL;
1017 	int error = 0;
1018 
1019 	/* Should be checked by the caller */
1020 	switch (ifp->if_type) {
1021 	case IFT_IEEE8023ADLAG:
1022 	case IFT_INFINIBANDLAG:
1023 		if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1024 			goto fallback;
1025 		break;
1026 	default:
1027 		goto fallback;
1028 	}
1029 
1030 	switch (cmd) {
1031 	case SIOCGLAGGPORT:
1032 		if (rp->rp_portname[0] == '\0' ||
1033 		    ifunit(rp->rp_portname) != ifp) {
1034 			error = EINVAL;
1035 			break;
1036 		}
1037 
1038 		LAGG_SLOCK(sc);
1039 		if (__predict_true((lp = ifp->if_lagg) != NULL &&
1040 		    lp->lp_softc == sc))
1041 			lagg_port2req(lp, rp);
1042 		else
1043 			error = ENOENT;	/* XXXGL: can happen? */
1044 		LAGG_SUNLOCK(sc);
1045 		break;
1046 
1047 	case SIOCSIFCAP:
1048 	case SIOCSIFCAPNV:
1049 		if (lp->lp_ioctl == NULL) {
1050 			error = EINVAL;
1051 			break;
1052 		}
1053 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1054 		if (error)
1055 			break;
1056 
1057 		/* Update lagg interface capabilities */
1058 		LAGG_XLOCK(sc);
1059 		lagg_capabilities(sc);
1060 		LAGG_XUNLOCK(sc);
1061 		VLAN_CAPABILITIES(sc->sc_ifp);
1062 		break;
1063 
1064 	case SIOCSIFMTU:
1065 		/* Do not allow the MTU to be changed once joined */
1066 		error = EINVAL;
1067 		break;
1068 
1069 	default:
1070 		goto fallback;
1071 	}
1072 
1073 	return (error);
1074 
1075 fallback:
1076 	if (lp != NULL && lp->lp_ioctl != NULL)
1077 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1078 
1079 	return (EINVAL);
1080 }
1081 
1082 /*
1083  * Requests counter @cnt data.
1084  *
1085  * Counter value is calculated the following way:
1086  * 1) for each port, sum difference between current and "initial" measurements.
1087  * 2) add lagg logical interface counters.
1088  * 3) add data from detached_counters array.
1089  *
1090  * We also do the following things on ports attach/detach:
1091  * 1) On port attach we store all counters it has into port_counter array.
1092  * 2) On port detach we add the different between "initial" and
1093  *   current counters data to detached_counters array.
1094  */
1095 static uint64_t
1096 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1097 {
1098 	struct epoch_tracker et;
1099 	struct lagg_softc *sc;
1100 	struct lagg_port *lp;
1101 	struct ifnet *lpifp;
1102 	uint64_t newval, oldval, vsum;
1103 
1104 	/* Revise this when we've got non-generic counters. */
1105 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1106 
1107 	sc = (struct lagg_softc *)ifp->if_softc;
1108 
1109 	vsum = 0;
1110 	NET_EPOCH_ENTER(et);
1111 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1112 		/* Saved attached value */
1113 		oldval = lp->port_counters.val[cnt];
1114 		/* current value */
1115 		lpifp = lp->lp_ifp;
1116 		newval = lpifp->if_get_counter(lpifp, cnt);
1117 		/* Calculate diff and save new */
1118 		vsum += newval - oldval;
1119 	}
1120 	NET_EPOCH_EXIT(et);
1121 
1122 	/*
1123 	 * Add counter data which might be added by upper
1124 	 * layer protocols operating on logical interface.
1125 	 */
1126 	vsum += if_get_counter_default(ifp, cnt);
1127 
1128 	/*
1129 	 * Add counter data from detached ports counters
1130 	 */
1131 	vsum += sc->detached_counters.val[cnt];
1132 
1133 	return (vsum);
1134 }
1135 
1136 /*
1137  * For direct output to child ports.
1138  */
1139 static int
1140 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1141 	const struct sockaddr *dst, struct route *ro)
1142 {
1143 	struct lagg_port *lp = ifp->if_lagg;
1144 
1145 	switch (dst->sa_family) {
1146 		case pseudo_AF_HDRCMPLT:
1147 		case AF_UNSPEC:
1148 			if (lp != NULL)
1149 				return ((*lp->lp_output)(ifp, m, dst, ro));
1150 	}
1151 
1152 	/* drop any other frames */
1153 	m_freem(m);
1154 	return (ENETDOWN);
1155 }
1156 
1157 static void
1158 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1159 {
1160 	struct lagg_port *lp;
1161 	struct lagg_softc *sc;
1162 
1163 	if ((lp = ifp->if_lagg) == NULL)
1164 		return;
1165 	/* If the ifnet is just being renamed, don't do anything. */
1166 	if (ifp->if_flags & IFF_RENAMING)
1167 		return;
1168 
1169 	sc = lp->lp_softc;
1170 
1171 	LAGG_XLOCK(sc);
1172 	lp->lp_detaching = 1;
1173 	lagg_port_destroy(lp, 1);
1174 	LAGG_XUNLOCK(sc);
1175 	VLAN_CAPABILITIES(sc->sc_ifp);
1176 }
1177 
1178 static void
1179 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1180 {
1181 	struct lagg_softc *sc = lp->lp_softc;
1182 
1183 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1184 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1185 	rp->rp_prio = lp->lp_prio;
1186 	rp->rp_flags = lp->lp_flags;
1187 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1188 
1189 	/* Add protocol specific flags */
1190 	switch (sc->sc_proto) {
1191 		case LAGG_PROTO_FAILOVER:
1192 			if (lp == sc->sc_primary)
1193 				rp->rp_flags |= LAGG_PORT_MASTER;
1194 			if (lp == lagg_link_active(sc, sc->sc_primary))
1195 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1196 			break;
1197 
1198 		case LAGG_PROTO_ROUNDROBIN:
1199 		case LAGG_PROTO_LOADBALANCE:
1200 		case LAGG_PROTO_BROADCAST:
1201 			if (LAGG_PORTACTIVE(lp))
1202 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1203 			break;
1204 
1205 		case LAGG_PROTO_LACP:
1206 			/* LACP has a different definition of active */
1207 			if (lacp_isactive(lp))
1208 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1209 			if (lacp_iscollecting(lp))
1210 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1211 			if (lacp_isdistributing(lp))
1212 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1213 			break;
1214 	}
1215 
1216 }
1217 
1218 static void
1219 lagg_watchdog_infiniband(void *arg)
1220 {
1221 	struct epoch_tracker et;
1222 	struct lagg_softc *sc;
1223 	struct lagg_port *lp;
1224 	struct ifnet *ifp;
1225 	struct ifnet *lp_ifp;
1226 
1227 	sc = arg;
1228 
1229 	/*
1230 	 * Because infiniband nodes have a fixed MAC address, which is
1231 	 * generated by the so-called GID, we need to regularly update
1232 	 * the link level address of the parent lagg<N> device when
1233 	 * the active port changes. Possibly we could piggy-back on
1234 	 * link up/down events aswell, but using a timer also provides
1235 	 * a guarantee against too frequent events. This operation
1236 	 * does not have to be atomic.
1237 	 */
1238 	NET_EPOCH_ENTER(et);
1239 	lp = lagg_link_active(sc, sc->sc_primary);
1240 	if (lp != NULL) {
1241 		ifp = sc->sc_ifp;
1242 		lp_ifp = lp->lp_ifp;
1243 
1244 		if (ifp != NULL && lp_ifp != NULL &&
1245 		    (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0 ||
1246 		     memcmp(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen) != 0)) {
1247 			memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1248 			memcpy(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen);
1249 
1250 			CURVNET_SET(ifp->if_vnet);
1251 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1252 			CURVNET_RESTORE();
1253 		}
1254 	}
1255 	NET_EPOCH_EXIT(et);
1256 
1257 	callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1258 }
1259 
1260 static void
1261 lagg_if_updown(struct lagg_softc *sc, bool up)
1262 {
1263 	struct ifreq ifr = {};
1264 	struct lagg_port *lp;
1265 
1266 	LAGG_XLOCK_ASSERT(sc);
1267 
1268 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1269 		if (up)
1270 			if_up(lp->lp_ifp);
1271 		else
1272 			if_down(lp->lp_ifp);
1273 
1274 		if (lp->lp_ioctl != NULL)
1275 			lp->lp_ioctl(lp->lp_ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
1276 	}
1277 }
1278 
1279 static void
1280 lagg_init(void *xsc)
1281 {
1282 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1283 	struct ifnet *ifp = sc->sc_ifp;
1284 	struct lagg_port *lp;
1285 
1286 	LAGG_XLOCK(sc);
1287 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1288 		LAGG_XUNLOCK(sc);
1289 		return;
1290 	}
1291 
1292 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1293 
1294 	/*
1295 	 * Update the port lladdrs if needed.
1296 	 * This might be if_setlladdr() notification
1297 	 * that lladdr has been changed.
1298 	 */
1299 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1300 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1301 		    ifp->if_addrlen) != 0)
1302 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1303 	}
1304 
1305 	lagg_if_updown(sc, true);
1306 
1307 	lagg_proto_init(sc);
1308 
1309 	if (ifp->if_type == IFT_INFINIBAND) {
1310 		mtx_lock(&sc->sc_mtx);
1311 		lagg_watchdog_infiniband(sc);
1312 		mtx_unlock(&sc->sc_mtx);
1313 	}
1314 
1315 	LAGG_XUNLOCK(sc);
1316 }
1317 
1318 static void
1319 lagg_stop(struct lagg_softc *sc)
1320 {
1321 	struct ifnet *ifp = sc->sc_ifp;
1322 
1323 	LAGG_XLOCK_ASSERT(sc);
1324 
1325 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1326 		return;
1327 
1328 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1329 
1330 	lagg_proto_stop(sc);
1331 
1332 	mtx_lock(&sc->sc_mtx);
1333 	callout_stop(&sc->sc_watchdog);
1334 	mtx_unlock(&sc->sc_mtx);
1335 
1336 	lagg_if_updown(sc, false);
1337 
1338 	callout_drain(&sc->sc_watchdog);
1339 }
1340 
1341 static int
1342 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1343 {
1344 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1345 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1346 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1347 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1348 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1349 	struct ifreq *ifr = (struct ifreq *)data;
1350 	struct lagg_port *lp;
1351 	struct ifnet *tpif;
1352 	struct thread *td = curthread;
1353 	char *buf, *outbuf;
1354 	int count, buflen, len, error = 0, oldmtu;
1355 
1356 	bzero(&rpbuf, sizeof(rpbuf));
1357 
1358 	/* XXX: This can race with lagg_clone_destroy. */
1359 
1360 	switch (cmd) {
1361 	case SIOCGLAGG:
1362 		LAGG_XLOCK(sc);
1363 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1364 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1365 		ra->ra_proto = sc->sc_proto;
1366 		lagg_proto_request(sc, &ra->ra_psc);
1367 		count = 0;
1368 		buf = outbuf;
1369 		len = min(ra->ra_size, buflen);
1370 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1371 			if (len < sizeof(rpbuf))
1372 				break;
1373 
1374 			lagg_port2req(lp, &rpbuf);
1375 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1376 			count++;
1377 			buf += sizeof(rpbuf);
1378 			len -= sizeof(rpbuf);
1379 		}
1380 		LAGG_XUNLOCK(sc);
1381 		ra->ra_ports = count;
1382 		ra->ra_size = count * sizeof(rpbuf);
1383 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1384 		free(outbuf, M_TEMP);
1385 		break;
1386 	case SIOCSLAGG:
1387 		error = priv_check(td, PRIV_NET_LAGG);
1388 		if (error)
1389 			break;
1390 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1391 			error = EPROTONOSUPPORT;
1392 			break;
1393 		}
1394 		/* Infiniband only supports the failover protocol. */
1395 		if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1396 		    ifp->if_type == IFT_INFINIBAND) {
1397 			error = EPROTONOSUPPORT;
1398 			break;
1399 		}
1400 		LAGG_XLOCK(sc);
1401 		lagg_proto_detach(sc);
1402 		lagg_proto_attach(sc, ra->ra_proto);
1403 		LAGG_XUNLOCK(sc);
1404 		break;
1405 	case SIOCGLAGGOPTS:
1406 		LAGG_XLOCK(sc);
1407 		ro->ro_opts = sc->sc_opts;
1408 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1409 			struct lacp_softc *lsc;
1410 
1411 			lsc = (struct lacp_softc *)sc->sc_psc;
1412 			if (lsc->lsc_debug.lsc_tx_test != 0)
1413 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1414 			if (lsc->lsc_debug.lsc_rx_test != 0)
1415 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1416 			if (lsc->lsc_strict_mode != 0)
1417 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1418 			if (lsc->lsc_fast_timeout != 0)
1419 				ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1420 
1421 			ro->ro_active = sc->sc_active;
1422 		} else {
1423 			ro->ro_active = 0;
1424 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1425 				ro->ro_active += LAGG_PORTACTIVE(lp);
1426 		}
1427 		ro->ro_bkt = sc->sc_stride;
1428 		ro->ro_flapping = sc->sc_flapping;
1429 		ro->ro_flowid_shift = sc->flowid_shift;
1430 		LAGG_XUNLOCK(sc);
1431 		break;
1432 	case SIOCSLAGGOPTS:
1433 		error = priv_check(td, PRIV_NET_LAGG);
1434 		if (error)
1435 			break;
1436 
1437 		/*
1438 		 * The stride option was added without defining a corresponding
1439 		 * LAGG_OPT flag, so handle a non-zero value before checking
1440 		 * anything else to preserve compatibility.
1441 		 */
1442 		LAGG_XLOCK(sc);
1443 		if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1444 			if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1445 				LAGG_XUNLOCK(sc);
1446 				error = EINVAL;
1447 				break;
1448 			}
1449 			sc->sc_stride = ro->ro_bkt;
1450 		}
1451 		if (ro->ro_opts == 0) {
1452 			LAGG_XUNLOCK(sc);
1453 			break;
1454 		}
1455 
1456 		/*
1457 		 * Set options.  LACP options are stored in sc->sc_psc,
1458 		 * not in sc_opts.
1459 		 */
1460 		int valid, lacp;
1461 
1462 		switch (ro->ro_opts) {
1463 		case LAGG_OPT_USE_FLOWID:
1464 		case -LAGG_OPT_USE_FLOWID:
1465 		case LAGG_OPT_USE_NUMA:
1466 		case -LAGG_OPT_USE_NUMA:
1467 		case LAGG_OPT_FLOWIDSHIFT:
1468 		case LAGG_OPT_RR_LIMIT:
1469 			valid = 1;
1470 			lacp = 0;
1471 			break;
1472 		case LAGG_OPT_LACP_TXTEST:
1473 		case -LAGG_OPT_LACP_TXTEST:
1474 		case LAGG_OPT_LACP_RXTEST:
1475 		case -LAGG_OPT_LACP_RXTEST:
1476 		case LAGG_OPT_LACP_STRICT:
1477 		case -LAGG_OPT_LACP_STRICT:
1478 		case LAGG_OPT_LACP_FAST_TIMO:
1479 		case -LAGG_OPT_LACP_FAST_TIMO:
1480 			valid = lacp = 1;
1481 			break;
1482 		default:
1483 			valid = lacp = 0;
1484 			break;
1485 		}
1486 
1487 		if (valid == 0 ||
1488 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1489 			/* Invalid combination of options specified. */
1490 			error = EINVAL;
1491 			LAGG_XUNLOCK(sc);
1492 			break;	/* Return from SIOCSLAGGOPTS. */
1493 		}
1494 
1495 		/*
1496 		 * Store new options into sc->sc_opts except for
1497 		 * FLOWIDSHIFT, RR and LACP options.
1498 		 */
1499 		if (lacp == 0) {
1500 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1501 				sc->flowid_shift = ro->ro_flowid_shift;
1502 			else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1503 				if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1504 				    ro->ro_bkt == 0) {
1505 					error = EINVAL;
1506 					LAGG_XUNLOCK(sc);
1507 					break;
1508 				}
1509 				sc->sc_stride = ro->ro_bkt;
1510 			} else if (ro->ro_opts > 0)
1511 				sc->sc_opts |= ro->ro_opts;
1512 			else
1513 				sc->sc_opts &= ~ro->ro_opts;
1514 		} else {
1515 			struct lacp_softc *lsc;
1516 			struct lacp_port *lp;
1517 
1518 			lsc = (struct lacp_softc *)sc->sc_psc;
1519 
1520 			switch (ro->ro_opts) {
1521 			case LAGG_OPT_LACP_TXTEST:
1522 				lsc->lsc_debug.lsc_tx_test = 1;
1523 				break;
1524 			case -LAGG_OPT_LACP_TXTEST:
1525 				lsc->lsc_debug.lsc_tx_test = 0;
1526 				break;
1527 			case LAGG_OPT_LACP_RXTEST:
1528 				lsc->lsc_debug.lsc_rx_test = 1;
1529 				break;
1530 			case -LAGG_OPT_LACP_RXTEST:
1531 				lsc->lsc_debug.lsc_rx_test = 0;
1532 				break;
1533 			case LAGG_OPT_LACP_STRICT:
1534 				lsc->lsc_strict_mode = 1;
1535 				break;
1536 			case -LAGG_OPT_LACP_STRICT:
1537 				lsc->lsc_strict_mode = 0;
1538 				break;
1539 			case LAGG_OPT_LACP_FAST_TIMO:
1540 				LACP_LOCK(lsc);
1541 				LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1542 					lp->lp_state |= LACP_STATE_TIMEOUT;
1543 				LACP_UNLOCK(lsc);
1544 				lsc->lsc_fast_timeout = 1;
1545 				break;
1546 			case -LAGG_OPT_LACP_FAST_TIMO:
1547 				LACP_LOCK(lsc);
1548 				LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1549 					lp->lp_state &= ~LACP_STATE_TIMEOUT;
1550 				LACP_UNLOCK(lsc);
1551 				lsc->lsc_fast_timeout = 0;
1552 				break;
1553 			}
1554 		}
1555 		LAGG_XUNLOCK(sc);
1556 		break;
1557 	case SIOCGLAGGFLAGS:
1558 		rf->rf_flags = 0;
1559 		LAGG_XLOCK(sc);
1560 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1561 			rf->rf_flags |= LAGG_F_HASHL2;
1562 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1563 			rf->rf_flags |= LAGG_F_HASHL3;
1564 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1565 			rf->rf_flags |= LAGG_F_HASHL4;
1566 		LAGG_XUNLOCK(sc);
1567 		break;
1568 	case SIOCSLAGGHASH:
1569 		error = priv_check(td, PRIV_NET_LAGG);
1570 		if (error)
1571 			break;
1572 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1573 			error = EINVAL;
1574 			break;
1575 		}
1576 		LAGG_XLOCK(sc);
1577 		sc->sc_flags = 0;
1578 		if (rf->rf_flags & LAGG_F_HASHL2)
1579 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1580 		if (rf->rf_flags & LAGG_F_HASHL3)
1581 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1582 		if (rf->rf_flags & LAGG_F_HASHL4)
1583 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1584 		LAGG_XUNLOCK(sc);
1585 		break;
1586 	case SIOCGLAGGPORT:
1587 		if (rp->rp_portname[0] == '\0' ||
1588 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1589 			error = EINVAL;
1590 			break;
1591 		}
1592 
1593 		LAGG_SLOCK(sc);
1594 		if (__predict_true((lp = tpif->if_lagg) != NULL &&
1595 		    lp->lp_softc == sc))
1596 			lagg_port2req(lp, rp);
1597 		else
1598 			error = ENOENT;	/* XXXGL: can happen? */
1599 		LAGG_SUNLOCK(sc);
1600 		if_rele(tpif);
1601 		break;
1602 
1603 	case SIOCSLAGGPORT:
1604 		error = priv_check(td, PRIV_NET_LAGG);
1605 		if (error)
1606 			break;
1607 		if (rp->rp_portname[0] == '\0' ||
1608 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1609 			error = EINVAL;
1610 			break;
1611 		}
1612 #ifdef INET6
1613 		/*
1614 		 * A laggport interface should not have inet6 address
1615 		 * because two interfaces with a valid link-local
1616 		 * scope zone must not be merged in any form.  This
1617 		 * restriction is needed to prevent violation of
1618 		 * link-local scope zone.  Attempts to add a laggport
1619 		 * interface which has inet6 addresses triggers
1620 		 * removal of all inet6 addresses on the member
1621 		 * interface.
1622 		 */
1623 		if (in6ifa_llaonifp(tpif)) {
1624 			in6_ifdetach(tpif);
1625 				if_printf(sc->sc_ifp,
1626 				    "IPv6 addresses on %s have been removed "
1627 				    "before adding it as a member to prevent "
1628 				    "IPv6 address scope violation.\n",
1629 				    tpif->if_xname);
1630 		}
1631 #endif
1632 		oldmtu = ifp->if_mtu;
1633 		LAGG_XLOCK(sc);
1634 		error = lagg_port_create(sc, tpif);
1635 		LAGG_XUNLOCK(sc);
1636 		if_rele(tpif);
1637 
1638 		/*
1639 		 * LAGG MTU may change during addition of the first port.
1640 		 * If it did, do network layer specific procedure.
1641 		 */
1642 		if (ifp->if_mtu != oldmtu)
1643 			if_notifymtu(ifp);
1644 
1645 		VLAN_CAPABILITIES(ifp);
1646 		break;
1647 	case SIOCSLAGGDELPORT:
1648 		error = priv_check(td, PRIV_NET_LAGG);
1649 		if (error)
1650 			break;
1651 		if (rp->rp_portname[0] == '\0' ||
1652 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1653 			error = EINVAL;
1654 			break;
1655 		}
1656 
1657 		LAGG_XLOCK(sc);
1658 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1659 		    lp->lp_softc != sc) {
1660 			error = ENOENT;
1661 			LAGG_XUNLOCK(sc);
1662 			if_rele(tpif);
1663 			break;
1664 		}
1665 
1666 		error = lagg_port_destroy(lp, 1);
1667 		LAGG_XUNLOCK(sc);
1668 		if_rele(tpif);
1669 		VLAN_CAPABILITIES(ifp);
1670 		break;
1671 	case SIOCSIFFLAGS:
1672 		/* Set flags on ports too */
1673 		LAGG_XLOCK(sc);
1674 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1675 			lagg_setflags(lp, 1);
1676 		}
1677 
1678 		if (!(ifp->if_flags & IFF_UP) &&
1679 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1680 			/*
1681 			 * If interface is marked down and it is running,
1682 			 * then stop and disable it.
1683 			 */
1684 			lagg_stop(sc);
1685 			LAGG_XUNLOCK(sc);
1686 		} else if ((ifp->if_flags & IFF_UP) &&
1687 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1688 			/*
1689 			 * If interface is marked up and it is stopped, then
1690 			 * start it.
1691 			 */
1692 			LAGG_XUNLOCK(sc);
1693 			(*ifp->if_init)(sc);
1694 		} else
1695 			LAGG_XUNLOCK(sc);
1696 		break;
1697 	case SIOCADDMULTI:
1698 	case SIOCDELMULTI:
1699 		LAGG_XLOCK(sc);
1700 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1701 			lagg_clrmulti(lp);
1702 			lagg_setmulti(lp);
1703 		}
1704 		LAGG_XUNLOCK(sc);
1705 		error = 0;
1706 		break;
1707 	case SIOCSIFMEDIA:
1708 	case SIOCGIFMEDIA:
1709 		if (ifp->if_type == IFT_INFINIBAND)
1710 			error = EINVAL;
1711 		else
1712 			error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1713 		break;
1714 
1715 	case SIOCSIFCAP:
1716 	case SIOCSIFCAPNV:
1717 		LAGG_XLOCK(sc);
1718 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1719 			if (lp->lp_ioctl != NULL)
1720 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1721 		}
1722 		lagg_capabilities(sc);
1723 		LAGG_XUNLOCK(sc);
1724 		VLAN_CAPABILITIES(ifp);
1725 		error = 0;
1726 		break;
1727 
1728 	case SIOCGIFCAPNV:
1729 		error = 0;
1730 		break;
1731 
1732 	case SIOCSIFMTU:
1733 		LAGG_XLOCK(sc);
1734 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1735 			if (lp->lp_ioctl != NULL)
1736 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1737 			else
1738 				error = EINVAL;
1739 			if (error != 0) {
1740 				if_printf(ifp,
1741 				    "failed to change MTU to %d on port %s, "
1742 				    "reverting all ports to original MTU (%d)\n",
1743 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1744 				break;
1745 			}
1746 		}
1747 		if (error == 0) {
1748 			ifp->if_mtu = ifr->ifr_mtu;
1749 		} else {
1750 			/* set every port back to the original MTU */
1751 			ifr->ifr_mtu = ifp->if_mtu;
1752 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1753 				if (lp->lp_ioctl != NULL)
1754 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1755 			}
1756 		}
1757 		lagg_capabilities(sc);
1758 		LAGG_XUNLOCK(sc);
1759 		VLAN_CAPABILITIES(ifp);
1760 		break;
1761 
1762 	default:
1763 		error = ether_ioctl(ifp, cmd, data);
1764 		break;
1765 	}
1766 	return (error);
1767 }
1768 
1769 #if defined(KERN_TLS) || defined(RATELIMIT)
1770 #ifdef RATELIMIT
1771 static const struct if_snd_tag_sw lagg_snd_tag_ul_sw = {
1772 	.snd_tag_modify = lagg_snd_tag_modify,
1773 	.snd_tag_query = lagg_snd_tag_query,
1774 	.snd_tag_free = lagg_snd_tag_free,
1775 	.next_snd_tag = lagg_next_snd_tag,
1776 	.type = IF_SND_TAG_TYPE_UNLIMITED
1777 };
1778 
1779 static const struct if_snd_tag_sw lagg_snd_tag_rl_sw = {
1780 	.snd_tag_modify = lagg_snd_tag_modify,
1781 	.snd_tag_query = lagg_snd_tag_query,
1782 	.snd_tag_free = lagg_snd_tag_free,
1783 	.next_snd_tag = lagg_next_snd_tag,
1784 	.type = IF_SND_TAG_TYPE_RATE_LIMIT
1785 };
1786 #endif
1787 
1788 #ifdef KERN_TLS
1789 static const struct if_snd_tag_sw lagg_snd_tag_tls_sw = {
1790 	.snd_tag_modify = lagg_snd_tag_modify,
1791 	.snd_tag_query = lagg_snd_tag_query,
1792 	.snd_tag_free = lagg_snd_tag_free,
1793 	.next_snd_tag = lagg_next_snd_tag,
1794 	.type = IF_SND_TAG_TYPE_TLS
1795 };
1796 
1797 #ifdef RATELIMIT
1798 static const struct if_snd_tag_sw lagg_snd_tag_tls_rl_sw = {
1799 	.snd_tag_modify = lagg_snd_tag_modify,
1800 	.snd_tag_query = lagg_snd_tag_query,
1801 	.snd_tag_free = lagg_snd_tag_free,
1802 	.next_snd_tag = lagg_next_snd_tag,
1803 	.type = IF_SND_TAG_TYPE_TLS_RATE_LIMIT
1804 };
1805 #endif
1806 #endif
1807 
1808 static inline struct lagg_snd_tag *
1809 mst_to_lst(struct m_snd_tag *mst)
1810 {
1811 
1812 	return (__containerof(mst, struct lagg_snd_tag, com));
1813 }
1814 
1815 /*
1816  * Look up the port used by a specific flow.  This only works for lagg
1817  * protocols with deterministic port mappings (e.g. not roundrobin).
1818  * In addition protocols which use a hash to map flows to ports must
1819  * be configured to use the mbuf flowid rather than hashing packet
1820  * contents.
1821  */
1822 static struct lagg_port *
1823 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1824     uint8_t numa_domain)
1825 {
1826 	struct lagg_softc *sc;
1827 	struct lagg_port *lp;
1828 	struct lagg_lb *lb;
1829 	uint32_t hash, p;
1830 	int err;
1831 
1832 	sc = ifp->if_softc;
1833 
1834 	switch (sc->sc_proto) {
1835 	case LAGG_PROTO_FAILOVER:
1836 		return (lagg_link_active(sc, sc->sc_primary));
1837 	case LAGG_PROTO_LOADBALANCE:
1838 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1839 		    flowtype == M_HASHTYPE_NONE)
1840 			return (NULL);
1841 		p = flowid >> sc->flowid_shift;
1842 		p %= sc->sc_count;
1843 		lb = (struct lagg_lb *)sc->sc_psc;
1844 		lp = lb->lb_ports[p];
1845 		return (lagg_link_active(sc, lp));
1846 	case LAGG_PROTO_LACP:
1847 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1848 		    flowtype == M_HASHTYPE_NONE)
1849 			return (NULL);
1850 		hash = flowid >> sc->flowid_shift;
1851 		return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, &err));
1852 	default:
1853 		return (NULL);
1854 	}
1855 }
1856 
1857 static int
1858 lagg_snd_tag_alloc(struct ifnet *ifp,
1859     union if_snd_tag_alloc_params *params,
1860     struct m_snd_tag **ppmt)
1861 {
1862 	struct epoch_tracker et;
1863 	const struct if_snd_tag_sw *sw;
1864 	struct lagg_snd_tag *lst;
1865 	struct lagg_port *lp;
1866 	struct ifnet *lp_ifp;
1867 	struct m_snd_tag *mst;
1868 	int error;
1869 
1870 	switch (params->hdr.type) {
1871 #ifdef RATELIMIT
1872 	case IF_SND_TAG_TYPE_UNLIMITED:
1873 		sw = &lagg_snd_tag_ul_sw;
1874 		break;
1875 	case IF_SND_TAG_TYPE_RATE_LIMIT:
1876 		sw = &lagg_snd_tag_rl_sw;
1877 		break;
1878 #endif
1879 #ifdef KERN_TLS
1880 	case IF_SND_TAG_TYPE_TLS:
1881 		sw = &lagg_snd_tag_tls_sw;
1882 		break;
1883 	case IF_SND_TAG_TYPE_TLS_RX:
1884 		/* Return tag from port interface directly. */
1885 		sw = NULL;
1886 		break;
1887 #ifdef RATELIMIT
1888 	case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
1889 		sw = &lagg_snd_tag_tls_rl_sw;
1890 		break;
1891 #endif
1892 #endif
1893 	default:
1894 		return (EOPNOTSUPP);
1895 	}
1896 
1897 	NET_EPOCH_ENTER(et);
1898 	lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1899 	    params->hdr.flowtype, params->hdr.numa_domain);
1900 	if (lp == NULL) {
1901 		NET_EPOCH_EXIT(et);
1902 		return (EOPNOTSUPP);
1903 	}
1904 	if (lp->lp_ifp == NULL) {
1905 		NET_EPOCH_EXIT(et);
1906 		return (EOPNOTSUPP);
1907 	}
1908 	lp_ifp = lp->lp_ifp;
1909 	if_ref(lp_ifp);
1910 	NET_EPOCH_EXIT(et);
1911 
1912 	if (sw != NULL) {
1913 		lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1914 		if (lst == NULL) {
1915 			if_rele(lp_ifp);
1916 			return (ENOMEM);
1917 		}
1918 	} else
1919 		lst = NULL;
1920 
1921 	error = m_snd_tag_alloc(lp_ifp, params, &mst);
1922 	if_rele(lp_ifp);
1923 	if (error) {
1924 		free(lst, M_LAGG);
1925 		return (error);
1926 	}
1927 
1928 	if (sw != NULL) {
1929 		m_snd_tag_init(&lst->com, ifp, sw);
1930 		lst->tag = mst;
1931 
1932 		*ppmt = &lst->com;
1933 	} else
1934 		*ppmt = mst;
1935 
1936 	return (0);
1937 }
1938 
1939 static struct m_snd_tag *
1940 lagg_next_snd_tag(struct m_snd_tag *mst)
1941 {
1942 	struct lagg_snd_tag *lst;
1943 
1944 	lst = mst_to_lst(mst);
1945 	return (lst->tag);
1946 }
1947 
1948 static int
1949 lagg_snd_tag_modify(struct m_snd_tag *mst,
1950     union if_snd_tag_modify_params *params)
1951 {
1952 	struct lagg_snd_tag *lst;
1953 
1954 	lst = mst_to_lst(mst);
1955 	return (lst->tag->sw->snd_tag_modify(lst->tag, params));
1956 }
1957 
1958 static int
1959 lagg_snd_tag_query(struct m_snd_tag *mst,
1960     union if_snd_tag_query_params *params)
1961 {
1962 	struct lagg_snd_tag *lst;
1963 
1964 	lst = mst_to_lst(mst);
1965 	return (lst->tag->sw->snd_tag_query(lst->tag, params));
1966 }
1967 
1968 static void
1969 lagg_snd_tag_free(struct m_snd_tag *mst)
1970 {
1971 	struct lagg_snd_tag *lst;
1972 
1973 	lst = mst_to_lst(mst);
1974 	m_snd_tag_rele(lst->tag);
1975 	free(lst, M_LAGG);
1976 }
1977 
1978 static void
1979 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1980 {
1981 	/*
1982 	 * For lagg, we have an indirect
1983 	 * interface. The caller needs to
1984 	 * get a ratelimit tag on the actual
1985 	 * interface the flow will go on.
1986 	 */
1987 	q->rate_table = NULL;
1988 	q->flags = RT_IS_INDIRECT;
1989 	q->max_flows = 0;
1990 	q->number_of_rates = 0;
1991 }
1992 #endif
1993 
1994 static int
1995 lagg_setmulti(struct lagg_port *lp)
1996 {
1997 	struct lagg_softc *sc = lp->lp_softc;
1998 	struct ifnet *ifp = lp->lp_ifp;
1999 	struct ifnet *scifp = sc->sc_ifp;
2000 	struct lagg_mc *mc;
2001 	struct ifmultiaddr *ifma;
2002 	int error;
2003 
2004 	IF_ADDR_WLOCK(scifp);
2005 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
2006 		if (ifma->ifma_addr->sa_family != AF_LINK)
2007 			continue;
2008 		mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
2009 		if (mc == NULL) {
2010 			IF_ADDR_WUNLOCK(scifp);
2011 			return (ENOMEM);
2012 		}
2013 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
2014 		mc->mc_addr.sdl_index = ifp->if_index;
2015 		mc->mc_ifma = NULL;
2016 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
2017 	}
2018 	IF_ADDR_WUNLOCK(scifp);
2019 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
2020 		error = if_addmulti(ifp,
2021 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
2022 		if (error)
2023 			return (error);
2024 	}
2025 	return (0);
2026 }
2027 
2028 static int
2029 lagg_clrmulti(struct lagg_port *lp)
2030 {
2031 	struct lagg_mc *mc;
2032 
2033 	LAGG_XLOCK_ASSERT(lp->lp_softc);
2034 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
2035 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
2036 		if (mc->mc_ifma && lp->lp_detaching == 0)
2037 			if_delmulti_ifma(mc->mc_ifma);
2038 		free(mc, M_LAGG);
2039 	}
2040 	return (0);
2041 }
2042 
2043 static void
2044 lagg_setcaps(struct lagg_port *lp, int cap, int cap2)
2045 {
2046 	struct ifreq ifr;
2047 	struct siocsifcapnv_driver_data drv_ioctl_data;
2048 
2049 	if (lp->lp_ifp->if_capenable == cap &&
2050 	    lp->lp_ifp->if_capenable2 == cap2)
2051 		return;
2052 	if (lp->lp_ioctl == NULL)
2053 		return;
2054 	/* XXX */
2055 	if ((lp->lp_ifp->if_capabilities & IFCAP_NV) != 0) {
2056 		drv_ioctl_data.reqcap = cap;
2057 		drv_ioctl_data.reqcap2 = cap2;
2058 		drv_ioctl_data.nvcap = NULL;
2059 		(*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAPNV,
2060 		    (caddr_t)&drv_ioctl_data);
2061 	} else {
2062 		ifr.ifr_reqcap = cap;
2063 		(*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr);
2064 	}
2065 }
2066 
2067 /* Handle a ref counted flag that should be set on the lagg port as well */
2068 static int
2069 lagg_setflag(struct lagg_port *lp, int flag, int status,
2070     int (*func)(struct ifnet *, int))
2071 {
2072 	struct lagg_softc *sc = lp->lp_softc;
2073 	struct ifnet *scifp = sc->sc_ifp;
2074 	struct ifnet *ifp = lp->lp_ifp;
2075 	int error;
2076 
2077 	LAGG_XLOCK_ASSERT(sc);
2078 
2079 	status = status ? (scifp->if_flags & flag) : 0;
2080 	/* Now "status" contains the flag value or 0 */
2081 
2082 	/*
2083 	 * See if recorded ports status is different from what
2084 	 * we want it to be.  If it is, flip it.  We record ports
2085 	 * status in lp_ifflags so that we won't clear ports flag
2086 	 * we haven't set.  In fact, we don't clear or set ports
2087 	 * flags directly, but get or release references to them.
2088 	 * That's why we can be sure that recorded flags still are
2089 	 * in accord with actual ports flags.
2090 	 */
2091 	if (status != (lp->lp_ifflags & flag)) {
2092 		error = (*func)(ifp, status);
2093 		if (error)
2094 			return (error);
2095 		lp->lp_ifflags &= ~flag;
2096 		lp->lp_ifflags |= status;
2097 	}
2098 	return (0);
2099 }
2100 
2101 /*
2102  * Handle IFF_* flags that require certain changes on the lagg port
2103  * if "status" is true, update ports flags respective to the lagg
2104  * if "status" is false, forcedly clear the flags set on port.
2105  */
2106 static int
2107 lagg_setflags(struct lagg_port *lp, int status)
2108 {
2109 	int error, i;
2110 
2111 	for (i = 0; lagg_pflags[i].flag; i++) {
2112 		error = lagg_setflag(lp, lagg_pflags[i].flag,
2113 		    status, lagg_pflags[i].func);
2114 		if (error)
2115 			return (error);
2116 	}
2117 	return (0);
2118 }
2119 
2120 static int
2121 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2122 {
2123 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2124 
2125 	NET_EPOCH_ASSERT();
2126 #if defined(KERN_TLS) || defined(RATELIMIT)
2127 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2128 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2129 #endif
2130 	/* We need a Tx algorithm and at least one port */
2131 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2132 		m_freem(m);
2133 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2134 		return (ENXIO);
2135 	}
2136 
2137 	ETHER_BPF_MTAP(ifp, m);
2138 
2139 	return (lagg_proto_start(sc, m));
2140 }
2141 
2142 static int
2143 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2144 {
2145 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2146 
2147 	NET_EPOCH_ASSERT();
2148 #if defined(KERN_TLS) || defined(RATELIMIT)
2149 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2150 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2151 #endif
2152 	/* We need a Tx algorithm and at least one port */
2153 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2154 		m_freem(m);
2155 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2156 		return (ENXIO);
2157 	}
2158 
2159 	infiniband_bpf_mtap(ifp, m);
2160 
2161 	return (lagg_proto_start(sc, m));
2162 }
2163 
2164 /*
2165  * The ifp->if_qflush entry point for lagg(4) is no-op.
2166  */
2167 static void
2168 lagg_qflush(struct ifnet *ifp __unused)
2169 {
2170 }
2171 
2172 static struct mbuf *
2173 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2174 {
2175 	struct lagg_port *lp = ifp->if_lagg;
2176 	struct lagg_softc *sc = lp->lp_softc;
2177 	struct ifnet *scifp = sc->sc_ifp;
2178 
2179 	NET_EPOCH_ASSERT();
2180 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2181 	    lp->lp_detaching != 0 ||
2182 	    sc->sc_proto == LAGG_PROTO_NONE) {
2183 		m_freem(m);
2184 		return (NULL);
2185 	}
2186 
2187 	m = lagg_proto_input(sc, lp, m);
2188 	if (m != NULL) {
2189 		ETHER_BPF_MTAP(scifp, m);
2190 
2191 		if ((scifp->if_flags & IFF_MONITOR) != 0) {
2192 			m_freem(m);
2193 			m = NULL;
2194 		}
2195 	}
2196 
2197 #ifdef DEV_NETMAP
2198 	if (m != NULL && scifp->if_capenable & IFCAP_NETMAP) {
2199 		scifp->if_input(scifp, m);
2200 		m = NULL;
2201 	}
2202 #endif	/* DEV_NETMAP */
2203 
2204 	return (m);
2205 }
2206 
2207 static struct mbuf *
2208 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2209 {
2210 	struct lagg_port *lp = ifp->if_lagg;
2211 	struct lagg_softc *sc = lp->lp_softc;
2212 	struct ifnet *scifp = sc->sc_ifp;
2213 
2214 	NET_EPOCH_ASSERT();
2215 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2216 	    lp->lp_detaching != 0 ||
2217 	    sc->sc_proto == LAGG_PROTO_NONE) {
2218 		m_freem(m);
2219 		return (NULL);
2220 	}
2221 
2222 	m = lagg_proto_input(sc, lp, m);
2223 	if (m != NULL) {
2224 		infiniband_bpf_mtap(scifp, m);
2225 
2226 		if ((scifp->if_flags & IFF_MONITOR) != 0) {
2227 			m_freem(m);
2228 			m = NULL;
2229 		}
2230 	}
2231 
2232 	return (m);
2233 }
2234 
2235 static int
2236 lagg_media_change(struct ifnet *ifp)
2237 {
2238 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2239 
2240 	if (sc->sc_ifflags & IFF_DEBUG)
2241 		printf("%s\n", __func__);
2242 
2243 	/* Ignore */
2244 	return (0);
2245 }
2246 
2247 static void
2248 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2249 {
2250 	struct epoch_tracker et;
2251 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2252 	struct lagg_port *lp;
2253 
2254 	imr->ifm_status = IFM_AVALID;
2255 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
2256 
2257 	NET_EPOCH_ENTER(et);
2258 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2259 		if (LAGG_PORTACTIVE(lp))
2260 			imr->ifm_status |= IFM_ACTIVE;
2261 	}
2262 	NET_EPOCH_EXIT(et);
2263 }
2264 
2265 static void
2266 lagg_linkstate(struct lagg_softc *sc)
2267 {
2268 	struct epoch_tracker et;
2269 	struct lagg_port *lp;
2270 	int new_link = LINK_STATE_DOWN;
2271 	uint64_t speed;
2272 
2273 	LAGG_XLOCK_ASSERT(sc);
2274 
2275 	/* LACP handles link state itself */
2276 	if (sc->sc_proto == LAGG_PROTO_LACP)
2277 		return;
2278 
2279 	/* Our link is considered up if at least one of our ports is active */
2280 	NET_EPOCH_ENTER(et);
2281 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2282 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2283 			new_link = LINK_STATE_UP;
2284 			break;
2285 		}
2286 	}
2287 	NET_EPOCH_EXIT(et);
2288 	if_link_state_change(sc->sc_ifp, new_link);
2289 
2290 	/* Update if_baudrate to reflect the max possible speed */
2291 	switch (sc->sc_proto) {
2292 		case LAGG_PROTO_FAILOVER:
2293 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2294 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
2295 			break;
2296 		case LAGG_PROTO_ROUNDROBIN:
2297 		case LAGG_PROTO_LOADBALANCE:
2298 		case LAGG_PROTO_BROADCAST:
2299 			speed = 0;
2300 			NET_EPOCH_ENTER(et);
2301 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2302 				speed += lp->lp_ifp->if_baudrate;
2303 			NET_EPOCH_EXIT(et);
2304 			sc->sc_ifp->if_baudrate = speed;
2305 			break;
2306 		case LAGG_PROTO_LACP:
2307 			/* LACP updates if_baudrate itself */
2308 			break;
2309 	}
2310 }
2311 
2312 static void
2313 lagg_port_state(struct ifnet *ifp, int state)
2314 {
2315 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2316 	struct lagg_softc *sc = NULL;
2317 
2318 	if (lp != NULL)
2319 		sc = lp->lp_softc;
2320 	if (sc == NULL)
2321 		return;
2322 
2323 	LAGG_XLOCK(sc);
2324 	lagg_linkstate(sc);
2325 	lagg_proto_linkstate(sc, lp);
2326 	LAGG_XUNLOCK(sc);
2327 }
2328 
2329 struct lagg_port *
2330 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2331 {
2332 	struct lagg_port *lp_next, *rval = NULL;
2333 
2334 	/*
2335 	 * Search a port which reports an active link state.
2336 	 */
2337 
2338 #ifdef INVARIANTS
2339 	/*
2340 	 * This is called with either in the network epoch
2341 	 * or with LAGG_XLOCK(sc) held.
2342 	 */
2343 	if (!in_epoch(net_epoch_preempt))
2344 		LAGG_XLOCK_ASSERT(sc);
2345 #endif
2346 
2347 	if (lp == NULL)
2348 		goto search;
2349 	if (LAGG_PORTACTIVE(lp)) {
2350 		rval = lp;
2351 		goto found;
2352 	}
2353 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2354 	    LAGG_PORTACTIVE(lp_next)) {
2355 		rval = lp_next;
2356 		goto found;
2357 	}
2358 
2359 search:
2360 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2361 		if (LAGG_PORTACTIVE(lp_next)) {
2362 			return (lp_next);
2363 		}
2364 	}
2365 found:
2366 	return (rval);
2367 }
2368 
2369 int
2370 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2371 {
2372 
2373 #if defined(KERN_TLS) || defined(RATELIMIT)
2374 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2375 		struct lagg_snd_tag *lst;
2376 		struct m_snd_tag *mst;
2377 
2378 		mst = m->m_pkthdr.snd_tag;
2379 		lst = mst_to_lst(mst);
2380 		if (lst->tag->ifp != ifp) {
2381 			m_freem(m);
2382 			return (EAGAIN);
2383 		}
2384 		m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2385 		m_snd_tag_rele(mst);
2386 	}
2387 #endif
2388 	return (ifp->if_transmit)(ifp, m);
2389 }
2390 
2391 /*
2392  * Simple round robin aggregation
2393  */
2394 static void
2395 lagg_rr_attach(struct lagg_softc *sc)
2396 {
2397 	sc->sc_seq = 0;
2398 	sc->sc_stride = 1;
2399 }
2400 
2401 static int
2402 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2403 {
2404 	struct lagg_port *lp;
2405 	uint32_t p;
2406 
2407 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
2408 	p /= sc->sc_stride;
2409 	p %= sc->sc_count;
2410 	lp = CK_SLIST_FIRST(&sc->sc_ports);
2411 
2412 	while (p--)
2413 		lp = CK_SLIST_NEXT(lp, lp_entries);
2414 
2415 	/*
2416 	 * Check the port's link state. This will return the next active
2417 	 * port if the link is down or the port is NULL.
2418 	 */
2419 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2420 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2421 		m_freem(m);
2422 		return (ENETDOWN);
2423 	}
2424 
2425 	/* Send mbuf */
2426 	return (lagg_enqueue(lp->lp_ifp, m));
2427 }
2428 
2429 /*
2430  * Broadcast mode
2431  */
2432 static int
2433 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2434 {
2435 	int errors = 0;
2436 	int ret;
2437 	struct lagg_port *lp, *last = NULL;
2438 	struct mbuf *m0;
2439 
2440 	NET_EPOCH_ASSERT();
2441 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2442 		if (!LAGG_PORTACTIVE(lp))
2443 			continue;
2444 
2445 		if (last != NULL) {
2446 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2447 			if (m0 == NULL) {
2448 				ret = ENOBUFS;
2449 				errors++;
2450 				break;
2451 			}
2452 			lagg_enqueue(last->lp_ifp, m0);
2453 		}
2454 		last = lp;
2455 	}
2456 
2457 	if (last == NULL) {
2458 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2459 		m_freem(m);
2460 		return (ENOENT);
2461 	}
2462 	if ((last = lagg_link_active(sc, last)) == NULL) {
2463 		errors++;
2464 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2465 		m_freem(m);
2466 		return (ENETDOWN);
2467 	}
2468 
2469 	ret = lagg_enqueue(last->lp_ifp, m);
2470 	if (errors != 0)
2471 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2472 
2473 	return (ret);
2474 }
2475 
2476 /*
2477  * Active failover
2478  */
2479 static int
2480 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2481 {
2482 	struct lagg_port *lp;
2483 
2484 	/* Use the master port if active or the next available port */
2485 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2486 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2487 		m_freem(m);
2488 		return (ENETDOWN);
2489 	}
2490 
2491 	/* Send mbuf */
2492 	return (lagg_enqueue(lp->lp_ifp, m));
2493 }
2494 
2495 static struct mbuf *
2496 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2497 {
2498 	struct ifnet *ifp = sc->sc_ifp;
2499 	struct lagg_port *tmp_tp;
2500 
2501 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2502 		m->m_pkthdr.rcvif = ifp;
2503 		return (m);
2504 	}
2505 
2506 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2507 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2508 		/*
2509 		 * If tmp_tp is null, we've received a packet when all
2510 		 * our links are down. Weird, but process it anyways.
2511 		 */
2512 		if (tmp_tp == NULL || tmp_tp == lp) {
2513 			m->m_pkthdr.rcvif = ifp;
2514 			return (m);
2515 		}
2516 	}
2517 
2518 	m_freem(m);
2519 	return (NULL);
2520 }
2521 
2522 /*
2523  * Loadbalancing
2524  */
2525 static void
2526 lagg_lb_attach(struct lagg_softc *sc)
2527 {
2528 	struct lagg_port *lp;
2529 	struct lagg_lb *lb;
2530 
2531 	LAGG_XLOCK_ASSERT(sc);
2532 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2533 	lb->lb_key = m_ether_tcpip_hash_init();
2534 	sc->sc_psc = lb;
2535 
2536 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2537 		lagg_lb_port_create(lp);
2538 }
2539 
2540 static void
2541 lagg_lb_detach(struct lagg_softc *sc)
2542 {
2543 	struct lagg_lb *lb;
2544 
2545 	lb = (struct lagg_lb *)sc->sc_psc;
2546 	if (lb != NULL)
2547 		free(lb, M_LAGG);
2548 }
2549 
2550 static int
2551 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2552 {
2553 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2554 	struct lagg_port *lp_next;
2555 	int i = 0, rv;
2556 
2557 	rv = 0;
2558 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2559 	LAGG_XLOCK_ASSERT(sc);
2560 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2561 		if (lp_next == lp)
2562 			continue;
2563 		if (i >= LAGG_MAX_PORTS) {
2564 			rv = EINVAL;
2565 			break;
2566 		}
2567 		if (sc->sc_ifflags & IFF_DEBUG)
2568 			printf("%s: port %s at index %d\n",
2569 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2570 		lb->lb_ports[i++] = lp_next;
2571 	}
2572 
2573 	return (rv);
2574 }
2575 
2576 static int
2577 lagg_lb_port_create(struct lagg_port *lp)
2578 {
2579 	struct lagg_softc *sc = lp->lp_softc;
2580 	return (lagg_lb_porttable(sc, NULL));
2581 }
2582 
2583 static void
2584 lagg_lb_port_destroy(struct lagg_port *lp)
2585 {
2586 	struct lagg_softc *sc = lp->lp_softc;
2587 	lagg_lb_porttable(sc, lp);
2588 }
2589 
2590 static int
2591 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2592 {
2593 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2594 	struct lagg_port *lp = NULL;
2595 	uint32_t p = 0;
2596 
2597 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2598 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2599 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2600 	else
2601 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2602 	p %= sc->sc_count;
2603 	lp = lb->lb_ports[p];
2604 
2605 	/*
2606 	 * Check the port's link state. This will return the next active
2607 	 * port if the link is down or the port is NULL.
2608 	 */
2609 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2610 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2611 		m_freem(m);
2612 		return (ENETDOWN);
2613 	}
2614 
2615 	/* Send mbuf */
2616 	return (lagg_enqueue(lp->lp_ifp, m));
2617 }
2618 
2619 /*
2620  * 802.3ad LACP
2621  */
2622 static void
2623 lagg_lacp_attach(struct lagg_softc *sc)
2624 {
2625 	struct lagg_port *lp;
2626 
2627 	lacp_attach(sc);
2628 	LAGG_XLOCK_ASSERT(sc);
2629 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2630 		lacp_port_create(lp);
2631 }
2632 
2633 static void
2634 lagg_lacp_detach(struct lagg_softc *sc)
2635 {
2636 	struct lagg_port *lp;
2637 	void *psc;
2638 
2639 	LAGG_XLOCK_ASSERT(sc);
2640 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2641 		lacp_port_destroy(lp);
2642 
2643 	psc = sc->sc_psc;
2644 	sc->sc_psc = NULL;
2645 	lacp_detach(psc);
2646 }
2647 
2648 static void
2649 lagg_lacp_lladdr(struct lagg_softc *sc)
2650 {
2651 	struct lagg_port *lp;
2652 
2653 	LAGG_SXLOCK_ASSERT(sc);
2654 
2655 	/* purge all the lacp ports */
2656 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2657 		lacp_port_destroy(lp);
2658 
2659 	/* add them back in */
2660 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2661 		lacp_port_create(lp);
2662 }
2663 
2664 static int
2665 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2666 {
2667 	struct lagg_port *lp;
2668 	int err;
2669 
2670 	lp = lacp_select_tx_port(sc, m, &err);
2671 	if (lp == NULL) {
2672 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2673 		m_freem(m);
2674 		return (err);
2675 	}
2676 
2677 	/* Send mbuf */
2678 	return (lagg_enqueue(lp->lp_ifp, m));
2679 }
2680 
2681 static struct mbuf *
2682 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2683 {
2684 	struct ifnet *ifp = sc->sc_ifp;
2685 	struct ether_header *eh;
2686 	u_short etype;
2687 
2688 	eh = mtod(m, struct ether_header *);
2689 	etype = ntohs(eh->ether_type);
2690 
2691 	/* Tap off LACP control messages */
2692 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2693 		m = lacp_input(lp, m);
2694 		if (m == NULL)
2695 			return (NULL);
2696 	}
2697 
2698 	/*
2699 	 * If the port is not collecting or not in the active aggregator then
2700 	 * free and return.
2701 	 */
2702 	if (!lacp_iscollecting(lp) || !lacp_isactive(lp)) {
2703 		m_freem(m);
2704 		return (NULL);
2705 	}
2706 
2707 	m->m_pkthdr.rcvif = ifp;
2708 	return (m);
2709 }
2710 
2711 /* Default input */
2712 static struct mbuf *
2713 lagg_default_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2714 {
2715 	struct ifnet *ifp = sc->sc_ifp;
2716 
2717 	/* Just pass in the packet to our lagg device */
2718 	m->m_pkthdr.rcvif = ifp;
2719 
2720 	return (m);
2721 }
2722