xref: /freebsd/sys/net/if_lagg.c (revision 5ca8e32633c4ffbbcd6762e5888b6a4ba0708c6c)
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_init(void *);
141 static void	lagg_stop(struct lagg_softc *);
142 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
143 #if defined(KERN_TLS) || defined(RATELIMIT)
144 static int	lagg_snd_tag_alloc(struct ifnet *,
145 		    union if_snd_tag_alloc_params *,
146 		    struct m_snd_tag **);
147 static int	lagg_snd_tag_modify(struct m_snd_tag *,
148 		    union if_snd_tag_modify_params *);
149 static int	lagg_snd_tag_query(struct m_snd_tag *,
150 		    union if_snd_tag_query_params *);
151 static void	lagg_snd_tag_free(struct m_snd_tag *);
152 static struct m_snd_tag *lagg_next_snd_tag(struct m_snd_tag *);
153 static void	lagg_ratelimit_query(struct ifnet *,
154 		    struct if_ratelimit_query_results *);
155 #endif
156 static int	lagg_setmulti(struct lagg_port *);
157 static int	lagg_clrmulti(struct lagg_port *);
158 static void	lagg_setcaps(struct lagg_port *, int cap, int cap2);
159 static int	lagg_setflag(struct lagg_port *, int, int,
160 		    int (*func)(struct ifnet *, int));
161 static int	lagg_setflags(struct lagg_port *, int status);
162 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
163 static int	lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
164 static int	lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
165 static void	lagg_qflush(struct ifnet *);
166 static int	lagg_media_change(struct ifnet *);
167 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
168 static struct lagg_port *lagg_link_active(struct lagg_softc *,
169 		    struct lagg_port *);
170 
171 /* Simple round robin */
172 static void	lagg_rr_attach(struct lagg_softc *);
173 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
174 
175 /* Active failover */
176 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
177 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
178 		    struct mbuf *);
179 
180 /* Loadbalancing */
181 static void	lagg_lb_attach(struct lagg_softc *);
182 static void	lagg_lb_detach(struct lagg_softc *);
183 static int	lagg_lb_port_create(struct lagg_port *);
184 static void	lagg_lb_port_destroy(struct lagg_port *);
185 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
186 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
187 
188 /* Broadcast */
189 static int	lagg_bcast_start(struct lagg_softc *, struct mbuf *);
190 
191 /* 802.3ad LACP */
192 static void	lagg_lacp_attach(struct lagg_softc *);
193 static void	lagg_lacp_detach(struct lagg_softc *);
194 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
195 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
196 		    struct mbuf *);
197 static void	lagg_lacp_lladdr(struct lagg_softc *);
198 
199 /* Default input */
200 static struct mbuf *lagg_default_input(struct lagg_softc *, struct lagg_port *,
201 		    struct mbuf *);
202 
203 /* lagg protocol table */
204 static const struct lagg_proto {
205 	lagg_proto	pr_num;
206 	void		(*pr_attach)(struct lagg_softc *);
207 	void		(*pr_detach)(struct lagg_softc *);
208 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
209 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
210 			    struct mbuf *);
211 	int		(*pr_addport)(struct lagg_port *);
212 	void		(*pr_delport)(struct lagg_port *);
213 	void		(*pr_linkstate)(struct lagg_port *);
214 	void 		(*pr_init)(struct lagg_softc *);
215 	void 		(*pr_stop)(struct lagg_softc *);
216 	void 		(*pr_lladdr)(struct lagg_softc *);
217 	void		(*pr_request)(struct lagg_softc *, void *);
218 	void		(*pr_portreq)(struct lagg_port *, void *);
219 } lagg_protos[] = {
220     {
221 	.pr_num = LAGG_PROTO_NONE
222     },
223     {
224 	.pr_num = LAGG_PROTO_ROUNDROBIN,
225 	.pr_attach = lagg_rr_attach,
226 	.pr_start = lagg_rr_start,
227 	.pr_input = lagg_default_input,
228     },
229     {
230 	.pr_num = LAGG_PROTO_FAILOVER,
231 	.pr_start = lagg_fail_start,
232 	.pr_input = lagg_fail_input,
233     },
234     {
235 	.pr_num = LAGG_PROTO_LOADBALANCE,
236 	.pr_attach = lagg_lb_attach,
237 	.pr_detach = lagg_lb_detach,
238 	.pr_start = lagg_lb_start,
239 	.pr_input = lagg_default_input,
240 	.pr_addport = lagg_lb_port_create,
241 	.pr_delport = lagg_lb_port_destroy,
242     },
243     {
244 	.pr_num = LAGG_PROTO_LACP,
245 	.pr_attach = lagg_lacp_attach,
246 	.pr_detach = lagg_lacp_detach,
247 	.pr_start = lagg_lacp_start,
248 	.pr_input = lagg_lacp_input,
249 	.pr_addport = lacp_port_create,
250 	.pr_delport = lacp_port_destroy,
251 	.pr_linkstate = lacp_linkstate,
252 	.pr_init = lacp_init,
253 	.pr_stop = lacp_stop,
254 	.pr_lladdr = lagg_lacp_lladdr,
255 	.pr_request = lacp_req,
256 	.pr_portreq = lacp_portreq,
257     },
258     {
259 	.pr_num = LAGG_PROTO_BROADCAST,
260 	.pr_start = lagg_bcast_start,
261 	.pr_input = lagg_default_input,
262     },
263 };
264 
265 SYSCTL_DECL(_net_link);
266 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
267     "Link Aggregation");
268 
269 /* Allow input on any failover links */
270 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
271 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
272 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
273     &VNET_NAME(lagg_failover_rx_all), 0,
274     "Accept input from any interface in a failover lagg");
275 
276 /* Default value for using flowid */
277 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
278 #define	V_def_use_flowid	VNET(def_use_flowid)
279 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid,
280     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_use_flowid), 0,
281     "Default setting for using flow id for load sharing");
282 
283 /* Default value for using numa */
284 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
285 #define	V_def_use_numa	VNET(def_use_numa)
286 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa,
287     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_use_numa), 0,
288     "Use numa to steer flows");
289 
290 /* Default value for flowid shift */
291 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
292 #define	V_def_flowid_shift	VNET(def_flowid_shift)
293 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift,
294     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_flowid_shift), 0,
295     "Default setting for flowid shift for load sharing");
296 
297 static void
298 vnet_lagg_init(const void *unused __unused)
299 {
300 
301 	LAGG_LIST_LOCK_INIT();
302 	SLIST_INIT(&V_lagg_list);
303 	struct if_clone_addreq req = {
304 		.create_f = lagg_clone_create,
305 		.destroy_f = lagg_clone_destroy,
306 		.flags = IFC_F_AUTOUNIT,
307 	};
308 	V_lagg_cloner = ifc_attach_cloner(laggname, &req);
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 	ifc_detach_cloner(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 inline 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 inline 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, char *name, size_t len,
508     struct ifc_data *ifd, struct ifnet **ifpp)
509 {
510 	struct iflaggparam iflp;
511 	struct lagg_softc *sc;
512 	struct ifnet *ifp;
513 	int if_type;
514 	int error;
515 	static const uint8_t eaddr[LAGG_ADDR_LEN];
516 
517 	if (ifd->params != NULL) {
518 		error = ifc_copyin(ifd, &iflp, sizeof(iflp));
519 		if (error)
520 			return (error);
521 
522 		switch (iflp.lagg_type) {
523 		case LAGG_TYPE_ETHERNET:
524 			if_type = IFT_ETHER;
525 			break;
526 		case LAGG_TYPE_INFINIBAND:
527 			if_type = IFT_INFINIBAND;
528 			break;
529 		default:
530 			return (EINVAL);
531 		}
532 	} else {
533 		if_type = IFT_ETHER;
534 	}
535 
536 	sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK | M_ZERO);
537 	ifp = sc->sc_ifp = if_alloc(if_type);
538 	if (ifp == NULL) {
539 		free(sc, M_LAGG);
540 		return (ENOSPC);
541 	}
542 	LAGG_SX_INIT(sc);
543 
544 	mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
545 	callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
546 
547 	LAGG_XLOCK(sc);
548 	if (V_def_use_flowid)
549 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
550 	if (V_def_use_numa)
551 		sc->sc_opts |= LAGG_OPT_USE_NUMA;
552 	sc->flowid_shift = V_def_flowid_shift;
553 
554 	/* Hash all layers by default */
555 	sc->sc_flags = MBUF_HASHFLAG_L2 | MBUF_HASHFLAG_L3 | MBUF_HASHFLAG_L4;
556 
557 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
558 
559 	CK_SLIST_INIT(&sc->sc_ports);
560 
561 	switch (if_type) {
562 	case IFT_ETHER:
563 		/* Initialise pseudo media types */
564 		ifmedia_init(&sc->sc_media, 0, lagg_media_change,
565 		    lagg_media_status);
566 		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
567 		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
568 
569 		if_initname(ifp, laggname, ifd->unit);
570 		ifp->if_transmit = lagg_transmit_ethernet;
571 		break;
572 	case IFT_INFINIBAND:
573 		if_initname(ifp, laggname, ifd->unit);
574 		ifp->if_transmit = lagg_transmit_infiniband;
575 		break;
576 	default:
577 		break;
578 	}
579 	ifp->if_softc = sc;
580 	ifp->if_qflush = lagg_qflush;
581 	ifp->if_init = lagg_init;
582 	ifp->if_ioctl = lagg_ioctl;
583 	ifp->if_get_counter = lagg_get_counter;
584 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
585 #if defined(KERN_TLS) || defined(RATELIMIT)
586 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
587 	ifp->if_ratelimit_query = lagg_ratelimit_query;
588 #endif
589 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
590 
591 	/*
592 	 * Attach as an ordinary ethernet device, children will be attached
593 	 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
594 	 */
595 	switch (if_type) {
596 	case IFT_ETHER:
597 		ether_ifattach(ifp, eaddr);
598 		break;
599 	case IFT_INFINIBAND:
600 		infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr);
601 		break;
602 	default:
603 		break;
604 	}
605 
606 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
607 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
608 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
609 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
610 
611 	/* Insert into the global list of laggs */
612 	LAGG_LIST_LOCK();
613 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
614 	LAGG_LIST_UNLOCK();
615 	LAGG_XUNLOCK(sc);
616 	*ifpp = ifp;
617 
618 	return (0);
619 }
620 
621 static int
622 lagg_clone_destroy(struct if_clone *ifc, struct ifnet *ifp, uint32_t flags)
623 {
624 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
625 	struct lagg_port *lp;
626 
627 	LAGG_XLOCK(sc);
628 	sc->sc_destroying = 1;
629 	lagg_stop(sc);
630 	ifp->if_flags &= ~IFF_UP;
631 
632 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
633 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
634 
635 	/* Shutdown and remove lagg ports */
636 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
637 		lagg_port_destroy(lp, 1);
638 
639 	/* Unhook the aggregation protocol */
640 	lagg_proto_detach(sc);
641 	LAGG_XUNLOCK(sc);
642 
643 	switch (ifp->if_type) {
644 	case IFT_ETHER:
645 		ifmedia_removeall(&sc->sc_media);
646 		ether_ifdetach(ifp);
647 		break;
648 	case IFT_INFINIBAND:
649 		infiniband_ifdetach(ifp);
650 		break;
651 	default:
652 		break;
653 	}
654 	if_free(ifp);
655 
656 	LAGG_LIST_LOCK();
657 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
658 	LAGG_LIST_UNLOCK();
659 
660 	mtx_destroy(&sc->sc_mtx);
661 	LAGG_SX_DESTROY(sc);
662 	free(sc, M_LAGG);
663 
664 	return (0);
665 }
666 
667 static void
668 lagg_capabilities(struct lagg_softc *sc)
669 {
670 	struct lagg_port *lp;
671 	int cap, cap2, ena, ena2, pena, pena2;
672 	uint64_t hwa;
673 	struct ifnet_hw_tsomax hw_tsomax;
674 
675 	LAGG_XLOCK_ASSERT(sc);
676 
677 	/* Get common enabled capabilities for the lagg ports */
678 	ena = ena2 = ~0;
679 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
680 		ena &= lp->lp_ifp->if_capenable;
681 		ena2 &= lp->lp_ifp->if_capenable2;
682 	}
683 	if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
684 		ena = ena2 = 0;
685 
686 	/*
687 	 * Apply common enabled capabilities back to the lagg ports.
688 	 * May require several iterations if they are dependent.
689 	 */
690 	do {
691 		pena = ena;
692 		pena2 = ena2;
693 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
694 			lagg_setcaps(lp, ena, ena2);
695 			ena &= lp->lp_ifp->if_capenable;
696 			ena2 &= lp->lp_ifp->if_capenable2;
697 		}
698 	} while (pena != ena || pena2 != ena2);
699 
700 	/* Get other capabilities from the lagg ports */
701 	cap = cap2 = ~0;
702 	hwa = ~(uint64_t)0;
703 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
704 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
705 		cap &= lp->lp_ifp->if_capabilities;
706 		cap2 &= lp->lp_ifp->if_capabilities2;
707 		hwa &= lp->lp_ifp->if_hwassist;
708 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
709 	}
710 	if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
711 		cap = cap2 = hwa = 0;
712 
713 	if (sc->sc_ifp->if_capabilities != cap ||
714 	    sc->sc_ifp->if_capenable != ena ||
715 	    sc->sc_ifp->if_capenable2 != ena2 ||
716 	    sc->sc_ifp->if_hwassist != hwa ||
717 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
718 		sc->sc_ifp->if_capabilities = cap;
719 		sc->sc_ifp->if_capabilities2 = cap2;
720 		sc->sc_ifp->if_capenable = ena;
721 		sc->sc_ifp->if_capenable2 = ena2;
722 		sc->sc_ifp->if_hwassist = hwa;
723 		getmicrotime(&sc->sc_ifp->if_lastchange);
724 
725 		if (sc->sc_ifflags & IFF_DEBUG)
726 			if_printf(sc->sc_ifp,
727 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
728 	}
729 }
730 
731 static int
732 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
733 {
734 	struct lagg_softc *sc_ptr;
735 	struct lagg_port *lp, *tlp;
736 	struct ifreq ifr;
737 	int error, i, oldmtu;
738 	int if_type;
739 	uint64_t *pval;
740 
741 	LAGG_XLOCK_ASSERT(sc);
742 
743 	if (sc->sc_ifp == ifp) {
744 		if_printf(sc->sc_ifp,
745 		    "cannot add a lagg to itself as a port\n");
746 		return (EINVAL);
747 	}
748 
749 	if (sc->sc_destroying == 1)
750 		return (ENXIO);
751 
752 	/* Limit the maximal number of lagg ports */
753 	if (sc->sc_count >= LAGG_MAX_PORTS)
754 		return (ENOSPC);
755 
756 	/* Check if port has already been associated to a lagg */
757 	if (ifp->if_lagg != NULL) {
758 		/* Port is already in the current lagg? */
759 		lp = (struct lagg_port *)ifp->if_lagg;
760 		if (lp->lp_softc == sc)
761 			return (EEXIST);
762 		return (EBUSY);
763 	}
764 
765 	switch (sc->sc_ifp->if_type) {
766 	case IFT_ETHER:
767 		/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
768 		if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
769 			return (EPROTONOSUPPORT);
770 		if_type = IFT_IEEE8023ADLAG;
771 		break;
772 	case IFT_INFINIBAND:
773 		/* XXX Disallow non-infiniband interfaces */
774 		if (ifp->if_type != IFT_INFINIBAND)
775 			return (EPROTONOSUPPORT);
776 		if_type = IFT_INFINIBANDLAG;
777 		break;
778 	default:
779 		break;
780 	}
781 
782 	/* Allow the first Ethernet member to define the MTU */
783 	oldmtu = -1;
784 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
785 		sc->sc_ifp->if_mtu = ifp->if_mtu;
786 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
787 		if (ifp->if_ioctl == NULL) {
788 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
789 			    ifp->if_xname);
790 			return (EINVAL);
791 		}
792 		oldmtu = ifp->if_mtu;
793 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
794 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
795 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
796 		if (error != 0) {
797 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
798 			    ifp->if_xname);
799 			return (error);
800 		}
801 		ifr.ifr_mtu = oldmtu;
802 	}
803 
804 	lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK | M_ZERO);
805 	lp->lp_softc = sc;
806 
807 	/* Check if port is a stacked lagg */
808 	LAGG_LIST_LOCK();
809 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
810 		if (ifp == sc_ptr->sc_ifp) {
811 			LAGG_LIST_UNLOCK();
812 			free(lp, M_LAGG);
813 			if (oldmtu != -1)
814 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
815 				    (caddr_t)&ifr);
816 			return (EINVAL);
817 			/* XXX disable stacking for the moment, its untested */
818 #ifdef LAGG_PORT_STACKING
819 			lp->lp_flags |= LAGG_PORT_STACK;
820 			if (lagg_port_checkstacking(sc_ptr) >=
821 			    LAGG_MAX_STACKING) {
822 				LAGG_LIST_UNLOCK();
823 				free(lp, M_LAGG);
824 				if (oldmtu != -1)
825 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
826 					    (caddr_t)&ifr);
827 				return (E2BIG);
828 			}
829 #endif
830 		}
831 	}
832 	LAGG_LIST_UNLOCK();
833 
834 	if_ref(ifp);
835 	lp->lp_ifp = ifp;
836 
837 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
838 	lp->lp_ifcapenable = ifp->if_capenable;
839 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
840 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
841 		lagg_proto_lladdr(sc);
842 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
843 	} else {
844 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
845 	}
846 	lagg_setflags(lp, 1);
847 
848 	if (CK_SLIST_EMPTY(&sc->sc_ports))
849 		sc->sc_primary = lp;
850 
851 	/* Change the interface type */
852 	lp->lp_iftype = ifp->if_type;
853 	ifp->if_type = if_type;
854 	ifp->if_lagg = lp;
855 	lp->lp_ioctl = ifp->if_ioctl;
856 	ifp->if_ioctl = lagg_port_ioctl;
857 	lp->lp_output = ifp->if_output;
858 	ifp->if_output = lagg_port_output;
859 
860 	/* Read port counters */
861 	pval = lp->port_counters.val;
862 	for (i = 0; i < IFCOUNTERS; i++, pval++)
863 		*pval = ifp->if_get_counter(ifp, i);
864 
865 	/*
866 	 * Insert into the list of ports.
867 	 * Keep ports sorted by if_index. It is handy, when configuration
868 	 * is predictable and `ifconfig laggN create ...` command
869 	 * will lead to the same result each time.
870 	 */
871 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
872 		if (tlp->lp_ifp->if_index < ifp->if_index && (
873 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
874 		    ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
875 		    ifp->if_index))
876 			break;
877 	}
878 	if (tlp != NULL)
879 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
880 	else
881 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
882 	sc->sc_count++;
883 
884 	lagg_setmulti(lp);
885 
886 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
887 		/* Remove the port, without calling pr_delport. */
888 		lagg_port_destroy(lp, 0);
889 		if (oldmtu != -1)
890 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
891 		return (error);
892 	}
893 
894 	/* Update lagg capabilities */
895 	lagg_capabilities(sc);
896 	lagg_linkstate(sc);
897 
898 	return (0);
899 }
900 
901 #ifdef LAGG_PORT_STACKING
902 static int
903 lagg_port_checkstacking(struct lagg_softc *sc)
904 {
905 	struct lagg_softc *sc_ptr;
906 	struct lagg_port *lp;
907 	int m = 0;
908 
909 	LAGG_SXLOCK_ASSERT(sc);
910 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
911 		if (lp->lp_flags & LAGG_PORT_STACK) {
912 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
913 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
914 		}
915 	}
916 
917 	return (m + 1);
918 }
919 #endif
920 
921 static void
922 lagg_port_destroy_cb(epoch_context_t ec)
923 {
924 	struct lagg_port *lp;
925 	struct ifnet *ifp;
926 
927 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
928 	ifp = lp->lp_ifp;
929 
930 	if_rele(ifp);
931 	free(lp, M_LAGG);
932 }
933 
934 static int
935 lagg_port_destroy(struct lagg_port *lp, int rundelport)
936 {
937 	struct lagg_softc *sc = lp->lp_softc;
938 	struct lagg_port *lp_ptr, *lp0;
939 	struct ifnet *ifp = lp->lp_ifp;
940 	uint64_t *pval, vdiff;
941 	int i;
942 
943 	LAGG_XLOCK_ASSERT(sc);
944 
945 	if (rundelport)
946 		lagg_proto_delport(sc, lp);
947 
948 	if (lp->lp_detaching == 0)
949 		lagg_clrmulti(lp);
950 
951 	/* Restore interface */
952 	ifp->if_type = lp->lp_iftype;
953 	ifp->if_ioctl = lp->lp_ioctl;
954 	ifp->if_output = lp->lp_output;
955 	ifp->if_lagg = NULL;
956 
957 	/* Update detached port counters */
958 	pval = lp->port_counters.val;
959 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
960 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
961 		sc->detached_counters.val[i] += vdiff;
962 	}
963 
964 	/* Finally, remove the port from the lagg */
965 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
966 	sc->sc_count--;
967 
968 	/* Update the primary interface */
969 	if (lp == sc->sc_primary) {
970 		uint8_t lladdr[LAGG_ADDR_LEN];
971 
972 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
973 			bzero(&lladdr, LAGG_ADDR_LEN);
974 		else
975 			bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
976 		sc->sc_primary = lp0;
977 		if (sc->sc_destroying == 0) {
978 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
979 			lagg_proto_lladdr(sc);
980 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
981 
982 			/*
983 			 * Update lladdr for each port (new primary needs update
984 			 * as well, to switch from old lladdr to its 'real' one).
985 			 * We can skip this if the lagg is being destroyed.
986 			 */
987 			CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
988 				if_setlladdr(lp_ptr->lp_ifp, lladdr,
989 				    lp_ptr->lp_ifp->if_addrlen);
990 		}
991 	}
992 
993 	if (lp->lp_ifflags)
994 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
995 
996 	if (lp->lp_detaching == 0) {
997 		lagg_setflags(lp, 0);
998 		lagg_setcaps(lp, lp->lp_ifcapenable, lp->lp_ifcapenable2);
999 		if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
1000 	}
1001 
1002 	/*
1003 	 * free port and release it's ifnet reference after a grace period has
1004 	 * elapsed.
1005 	 */
1006 	NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
1007 	/* Update lagg capabilities */
1008 	lagg_capabilities(sc);
1009 	lagg_linkstate(sc);
1010 
1011 	return (0);
1012 }
1013 
1014 static int
1015 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1016 {
1017 	struct epoch_tracker et;
1018 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
1019 	struct lagg_softc *sc;
1020 	struct lagg_port *lp = NULL;
1021 	int error = 0;
1022 
1023 	/* Should be checked by the caller */
1024 	switch (ifp->if_type) {
1025 	case IFT_IEEE8023ADLAG:
1026 	case IFT_INFINIBANDLAG:
1027 		if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1028 			goto fallback;
1029 		break;
1030 	default:
1031 		goto fallback;
1032 	}
1033 
1034 	switch (cmd) {
1035 	case SIOCGLAGGPORT:
1036 		if (rp->rp_portname[0] == '\0' ||
1037 		    ifunit(rp->rp_portname) != ifp) {
1038 			error = EINVAL;
1039 			break;
1040 		}
1041 
1042 		NET_EPOCH_ENTER(et);
1043 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1044 			error = ENOENT;
1045 			NET_EPOCH_EXIT(et);
1046 			break;
1047 		}
1048 
1049 		LAGG_SLOCK(sc);
1050 		lagg_port2req(lp, rp);
1051 		LAGG_SUNLOCK(sc);
1052 		NET_EPOCH_EXIT(et);
1053 		break;
1054 
1055 	case SIOCSIFCAP:
1056 	case SIOCSIFCAPNV:
1057 		if (lp->lp_ioctl == NULL) {
1058 			error = EINVAL;
1059 			break;
1060 		}
1061 		error = (*lp->lp_ioctl)(ifp, cmd, data);
1062 		if (error)
1063 			break;
1064 
1065 		/* Update lagg interface capabilities */
1066 		LAGG_XLOCK(sc);
1067 		lagg_capabilities(sc);
1068 		LAGG_XUNLOCK(sc);
1069 		VLAN_CAPABILITIES(sc->sc_ifp);
1070 		break;
1071 
1072 	case SIOCSIFMTU:
1073 		/* Do not allow the MTU to be changed once joined */
1074 		error = EINVAL;
1075 		break;
1076 
1077 	default:
1078 		goto fallback;
1079 	}
1080 
1081 	return (error);
1082 
1083 fallback:
1084 	if (lp != NULL && lp->lp_ioctl != NULL)
1085 		return ((*lp->lp_ioctl)(ifp, cmd, data));
1086 
1087 	return (EINVAL);
1088 }
1089 
1090 /*
1091  * Requests counter @cnt data.
1092  *
1093  * Counter value is calculated the following way:
1094  * 1) for each port, sum difference between current and "initial" measurements.
1095  * 2) add lagg logical interface counters.
1096  * 3) add data from detached_counters array.
1097  *
1098  * We also do the following things on ports attach/detach:
1099  * 1) On port attach we store all counters it has into port_counter array.
1100  * 2) On port detach we add the different between "initial" and
1101  *   current counters data to detached_counters array.
1102  */
1103 static uint64_t
1104 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1105 {
1106 	struct epoch_tracker et;
1107 	struct lagg_softc *sc;
1108 	struct lagg_port *lp;
1109 	struct ifnet *lpifp;
1110 	uint64_t newval, oldval, vsum;
1111 
1112 	/* Revise this when we've got non-generic counters. */
1113 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1114 
1115 	sc = (struct lagg_softc *)ifp->if_softc;
1116 
1117 	vsum = 0;
1118 	NET_EPOCH_ENTER(et);
1119 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1120 		/* Saved attached value */
1121 		oldval = lp->port_counters.val[cnt];
1122 		/* current value */
1123 		lpifp = lp->lp_ifp;
1124 		newval = lpifp->if_get_counter(lpifp, cnt);
1125 		/* Calculate diff and save new */
1126 		vsum += newval - oldval;
1127 	}
1128 	NET_EPOCH_EXIT(et);
1129 
1130 	/*
1131 	 * Add counter data which might be added by upper
1132 	 * layer protocols operating on logical interface.
1133 	 */
1134 	vsum += if_get_counter_default(ifp, cnt);
1135 
1136 	/*
1137 	 * Add counter data from detached ports counters
1138 	 */
1139 	vsum += sc->detached_counters.val[cnt];
1140 
1141 	return (vsum);
1142 }
1143 
1144 /*
1145  * For direct output to child ports.
1146  */
1147 static int
1148 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1149 	const struct sockaddr *dst, struct route *ro)
1150 {
1151 	struct lagg_port *lp = ifp->if_lagg;
1152 
1153 	switch (dst->sa_family) {
1154 		case pseudo_AF_HDRCMPLT:
1155 		case AF_UNSPEC:
1156 			if (lp != NULL)
1157 				return ((*lp->lp_output)(ifp, m, dst, ro));
1158 	}
1159 
1160 	/* drop any other frames */
1161 	m_freem(m);
1162 	return (ENETDOWN);
1163 }
1164 
1165 static void
1166 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1167 {
1168 	struct lagg_port *lp;
1169 	struct lagg_softc *sc;
1170 
1171 	if ((lp = ifp->if_lagg) == NULL)
1172 		return;
1173 	/* If the ifnet is just being renamed, don't do anything. */
1174 	if (ifp->if_flags & IFF_RENAMING)
1175 		return;
1176 
1177 	sc = lp->lp_softc;
1178 
1179 	LAGG_XLOCK(sc);
1180 	lp->lp_detaching = 1;
1181 	lagg_port_destroy(lp, 1);
1182 	LAGG_XUNLOCK(sc);
1183 	VLAN_CAPABILITIES(sc->sc_ifp);
1184 }
1185 
1186 static void
1187 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1188 {
1189 	struct lagg_softc *sc = lp->lp_softc;
1190 
1191 	strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1192 	strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1193 	rp->rp_prio = lp->lp_prio;
1194 	rp->rp_flags = lp->lp_flags;
1195 	lagg_proto_portreq(sc, lp, &rp->rp_psc);
1196 
1197 	/* Add protocol specific flags */
1198 	switch (sc->sc_proto) {
1199 		case LAGG_PROTO_FAILOVER:
1200 			if (lp == sc->sc_primary)
1201 				rp->rp_flags |= LAGG_PORT_MASTER;
1202 			if (lp == lagg_link_active(sc, sc->sc_primary))
1203 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1204 			break;
1205 
1206 		case LAGG_PROTO_ROUNDROBIN:
1207 		case LAGG_PROTO_LOADBALANCE:
1208 		case LAGG_PROTO_BROADCAST:
1209 			if (LAGG_PORTACTIVE(lp))
1210 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1211 			break;
1212 
1213 		case LAGG_PROTO_LACP:
1214 			/* LACP has a different definition of active */
1215 			if (lacp_isactive(lp))
1216 				rp->rp_flags |= LAGG_PORT_ACTIVE;
1217 			if (lacp_iscollecting(lp))
1218 				rp->rp_flags |= LAGG_PORT_COLLECTING;
1219 			if (lacp_isdistributing(lp))
1220 				rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1221 			break;
1222 	}
1223 
1224 }
1225 
1226 static void
1227 lagg_watchdog_infiniband(void *arg)
1228 {
1229 	struct epoch_tracker et;
1230 	struct lagg_softc *sc;
1231 	struct lagg_port *lp;
1232 	struct ifnet *ifp;
1233 	struct ifnet *lp_ifp;
1234 
1235 	sc = arg;
1236 
1237 	/*
1238 	 * Because infiniband nodes have a fixed MAC address, which is
1239 	 * generated by the so-called GID, we need to regularly update
1240 	 * the link level address of the parent lagg<N> device when
1241 	 * the active port changes. Possibly we could piggy-back on
1242 	 * link up/down events aswell, but using a timer also provides
1243 	 * a guarantee against too frequent events. This operation
1244 	 * does not have to be atomic.
1245 	 */
1246 	NET_EPOCH_ENTER(et);
1247 	lp = lagg_link_active(sc, sc->sc_primary);
1248 	if (lp != NULL) {
1249 		ifp = sc->sc_ifp;
1250 		lp_ifp = lp->lp_ifp;
1251 
1252 		if (ifp != NULL && lp_ifp != NULL &&
1253 		    (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0 ||
1254 		     memcmp(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen) != 0)) {
1255 			memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1256 			memcpy(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen);
1257 
1258 			CURVNET_SET(ifp->if_vnet);
1259 			EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1260 			CURVNET_RESTORE();
1261 		}
1262 	}
1263 	NET_EPOCH_EXIT(et);
1264 
1265 	callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1266 }
1267 
1268 static void
1269 lagg_init(void *xsc)
1270 {
1271 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1272 	struct ifnet *ifp = sc->sc_ifp;
1273 	struct lagg_port *lp;
1274 
1275 	LAGG_XLOCK(sc);
1276 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1277 		LAGG_XUNLOCK(sc);
1278 		return;
1279 	}
1280 
1281 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1282 
1283 	/*
1284 	 * Update the port lladdrs if needed.
1285 	 * This might be if_setlladdr() notification
1286 	 * that lladdr has been changed.
1287 	 */
1288 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1289 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1290 		    ifp->if_addrlen) != 0)
1291 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1292 	}
1293 
1294 	lagg_proto_init(sc);
1295 
1296 	if (ifp->if_type == IFT_INFINIBAND) {
1297 		mtx_lock(&sc->sc_mtx);
1298 		lagg_watchdog_infiniband(sc);
1299 		mtx_unlock(&sc->sc_mtx);
1300 	}
1301 
1302 	LAGG_XUNLOCK(sc);
1303 }
1304 
1305 static void
1306 lagg_stop(struct lagg_softc *sc)
1307 {
1308 	struct ifnet *ifp = sc->sc_ifp;
1309 
1310 	LAGG_XLOCK_ASSERT(sc);
1311 
1312 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1313 		return;
1314 
1315 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1316 
1317 	lagg_proto_stop(sc);
1318 
1319 	mtx_lock(&sc->sc_mtx);
1320 	callout_stop(&sc->sc_watchdog);
1321 	mtx_unlock(&sc->sc_mtx);
1322 
1323 	callout_drain(&sc->sc_watchdog);
1324 }
1325 
1326 static int
1327 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1328 {
1329 	struct epoch_tracker et;
1330 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1331 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1332 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1333 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1334 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1335 	struct ifreq *ifr = (struct ifreq *)data;
1336 	struct lagg_port *lp;
1337 	struct ifnet *tpif;
1338 	struct thread *td = curthread;
1339 	char *buf, *outbuf;
1340 	int count, buflen, len, error = 0, oldmtu;
1341 
1342 	bzero(&rpbuf, sizeof(rpbuf));
1343 
1344 	/* XXX: This can race with lagg_clone_destroy. */
1345 
1346 	switch (cmd) {
1347 	case SIOCGLAGG:
1348 		LAGG_XLOCK(sc);
1349 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1350 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1351 		ra->ra_proto = sc->sc_proto;
1352 		lagg_proto_request(sc, &ra->ra_psc);
1353 		count = 0;
1354 		buf = outbuf;
1355 		len = min(ra->ra_size, buflen);
1356 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1357 			if (len < sizeof(rpbuf))
1358 				break;
1359 
1360 			lagg_port2req(lp, &rpbuf);
1361 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1362 			count++;
1363 			buf += sizeof(rpbuf);
1364 			len -= sizeof(rpbuf);
1365 		}
1366 		LAGG_XUNLOCK(sc);
1367 		ra->ra_ports = count;
1368 		ra->ra_size = count * sizeof(rpbuf);
1369 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1370 		free(outbuf, M_TEMP);
1371 		break;
1372 	case SIOCSLAGG:
1373 		error = priv_check(td, PRIV_NET_LAGG);
1374 		if (error)
1375 			break;
1376 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1377 			error = EPROTONOSUPPORT;
1378 			break;
1379 		}
1380 		/* Infiniband only supports the failover protocol. */
1381 		if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1382 		    ifp->if_type == IFT_INFINIBAND) {
1383 			error = EPROTONOSUPPORT;
1384 			break;
1385 		}
1386 		LAGG_XLOCK(sc);
1387 		lagg_proto_detach(sc);
1388 		lagg_proto_attach(sc, ra->ra_proto);
1389 		LAGG_XUNLOCK(sc);
1390 		break;
1391 	case SIOCGLAGGOPTS:
1392 		LAGG_XLOCK(sc);
1393 		ro->ro_opts = sc->sc_opts;
1394 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1395 			struct lacp_softc *lsc;
1396 
1397 			lsc = (struct lacp_softc *)sc->sc_psc;
1398 			if (lsc->lsc_debug.lsc_tx_test != 0)
1399 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1400 			if (lsc->lsc_debug.lsc_rx_test != 0)
1401 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1402 			if (lsc->lsc_strict_mode != 0)
1403 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1404 			if (lsc->lsc_fast_timeout != 0)
1405 				ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1406 
1407 			ro->ro_active = sc->sc_active;
1408 		} else {
1409 			ro->ro_active = 0;
1410 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1411 				ro->ro_active += LAGG_PORTACTIVE(lp);
1412 		}
1413 		ro->ro_bkt = sc->sc_stride;
1414 		ro->ro_flapping = sc->sc_flapping;
1415 		ro->ro_flowid_shift = sc->flowid_shift;
1416 		LAGG_XUNLOCK(sc);
1417 		break;
1418 	case SIOCSLAGGOPTS:
1419 		error = priv_check(td, PRIV_NET_LAGG);
1420 		if (error)
1421 			break;
1422 
1423 		/*
1424 		 * The stride option was added without defining a corresponding
1425 		 * LAGG_OPT flag, so handle a non-zero value before checking
1426 		 * anything else to preserve compatibility.
1427 		 */
1428 		LAGG_XLOCK(sc);
1429 		if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1430 			if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1431 				LAGG_XUNLOCK(sc);
1432 				error = EINVAL;
1433 				break;
1434 			}
1435 			sc->sc_stride = ro->ro_bkt;
1436 		}
1437 		if (ro->ro_opts == 0) {
1438 			LAGG_XUNLOCK(sc);
1439 			break;
1440 		}
1441 
1442 		/*
1443 		 * Set options.  LACP options are stored in sc->sc_psc,
1444 		 * not in sc_opts.
1445 		 */
1446 		int valid, lacp;
1447 
1448 		switch (ro->ro_opts) {
1449 		case LAGG_OPT_USE_FLOWID:
1450 		case -LAGG_OPT_USE_FLOWID:
1451 		case LAGG_OPT_USE_NUMA:
1452 		case -LAGG_OPT_USE_NUMA:
1453 		case LAGG_OPT_FLOWIDSHIFT:
1454 		case LAGG_OPT_RR_LIMIT:
1455 			valid = 1;
1456 			lacp = 0;
1457 			break;
1458 		case LAGG_OPT_LACP_TXTEST:
1459 		case -LAGG_OPT_LACP_TXTEST:
1460 		case LAGG_OPT_LACP_RXTEST:
1461 		case -LAGG_OPT_LACP_RXTEST:
1462 		case LAGG_OPT_LACP_STRICT:
1463 		case -LAGG_OPT_LACP_STRICT:
1464 		case LAGG_OPT_LACP_FAST_TIMO:
1465 		case -LAGG_OPT_LACP_FAST_TIMO:
1466 			valid = lacp = 1;
1467 			break;
1468 		default:
1469 			valid = lacp = 0;
1470 			break;
1471 		}
1472 
1473 		if (valid == 0 ||
1474 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1475 			/* Invalid combination of options specified. */
1476 			error = EINVAL;
1477 			LAGG_XUNLOCK(sc);
1478 			break;	/* Return from SIOCSLAGGOPTS. */
1479 		}
1480 
1481 		/*
1482 		 * Store new options into sc->sc_opts except for
1483 		 * FLOWIDSHIFT, RR and LACP options.
1484 		 */
1485 		if (lacp == 0) {
1486 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1487 				sc->flowid_shift = ro->ro_flowid_shift;
1488 			else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1489 				if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1490 				    ro->ro_bkt == 0) {
1491 					error = EINVAL;
1492 					LAGG_XUNLOCK(sc);
1493 					break;
1494 				}
1495 				sc->sc_stride = ro->ro_bkt;
1496 			} else if (ro->ro_opts > 0)
1497 				sc->sc_opts |= ro->ro_opts;
1498 			else
1499 				sc->sc_opts &= ~ro->ro_opts;
1500 		} else {
1501 			struct lacp_softc *lsc;
1502 			struct lacp_port *lp;
1503 
1504 			lsc = (struct lacp_softc *)sc->sc_psc;
1505 
1506 			switch (ro->ro_opts) {
1507 			case LAGG_OPT_LACP_TXTEST:
1508 				lsc->lsc_debug.lsc_tx_test = 1;
1509 				break;
1510 			case -LAGG_OPT_LACP_TXTEST:
1511 				lsc->lsc_debug.lsc_tx_test = 0;
1512 				break;
1513 			case LAGG_OPT_LACP_RXTEST:
1514 				lsc->lsc_debug.lsc_rx_test = 1;
1515 				break;
1516 			case -LAGG_OPT_LACP_RXTEST:
1517 				lsc->lsc_debug.lsc_rx_test = 0;
1518 				break;
1519 			case LAGG_OPT_LACP_STRICT:
1520 				lsc->lsc_strict_mode = 1;
1521 				break;
1522 			case -LAGG_OPT_LACP_STRICT:
1523 				lsc->lsc_strict_mode = 0;
1524 				break;
1525 			case LAGG_OPT_LACP_FAST_TIMO:
1526 				LACP_LOCK(lsc);
1527 				LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1528 					lp->lp_state |= LACP_STATE_TIMEOUT;
1529 				LACP_UNLOCK(lsc);
1530 				lsc->lsc_fast_timeout = 1;
1531 				break;
1532 			case -LAGG_OPT_LACP_FAST_TIMO:
1533 				LACP_LOCK(lsc);
1534 				LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1535 					lp->lp_state &= ~LACP_STATE_TIMEOUT;
1536 				LACP_UNLOCK(lsc);
1537 				lsc->lsc_fast_timeout = 0;
1538 				break;
1539 			}
1540 		}
1541 		LAGG_XUNLOCK(sc);
1542 		break;
1543 	case SIOCGLAGGFLAGS:
1544 		rf->rf_flags = 0;
1545 		LAGG_XLOCK(sc);
1546 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1547 			rf->rf_flags |= LAGG_F_HASHL2;
1548 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1549 			rf->rf_flags |= LAGG_F_HASHL3;
1550 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1551 			rf->rf_flags |= LAGG_F_HASHL4;
1552 		LAGG_XUNLOCK(sc);
1553 		break;
1554 	case SIOCSLAGGHASH:
1555 		error = priv_check(td, PRIV_NET_LAGG);
1556 		if (error)
1557 			break;
1558 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1559 			error = EINVAL;
1560 			break;
1561 		}
1562 		LAGG_XLOCK(sc);
1563 		sc->sc_flags = 0;
1564 		if (rf->rf_flags & LAGG_F_HASHL2)
1565 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1566 		if (rf->rf_flags & LAGG_F_HASHL3)
1567 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1568 		if (rf->rf_flags & LAGG_F_HASHL4)
1569 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1570 		LAGG_XUNLOCK(sc);
1571 		break;
1572 	case SIOCGLAGGPORT:
1573 		if (rp->rp_portname[0] == '\0' ||
1574 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1575 			error = EINVAL;
1576 			break;
1577 		}
1578 
1579 		NET_EPOCH_ENTER(et);
1580 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1581 		    lp->lp_softc != sc) {
1582 			error = ENOENT;
1583 			NET_EPOCH_EXIT(et);
1584 			if_rele(tpif);
1585 			break;
1586 		}
1587 
1588 		LAGG_SLOCK(sc);
1589 		lagg_port2req(lp, rp);
1590 		LAGG_SUNLOCK(sc);
1591 		NET_EPOCH_EXIT(et);
1592 		if_rele(tpif);
1593 		break;
1594 	case SIOCSLAGGPORT:
1595 		error = priv_check(td, PRIV_NET_LAGG);
1596 		if (error)
1597 			break;
1598 		if (rp->rp_portname[0] == '\0' ||
1599 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1600 			error = EINVAL;
1601 			break;
1602 		}
1603 #ifdef INET6
1604 		/*
1605 		 * A laggport interface should not have inet6 address
1606 		 * because two interfaces with a valid link-local
1607 		 * scope zone must not be merged in any form.  This
1608 		 * restriction is needed to prevent violation of
1609 		 * link-local scope zone.  Attempts to add a laggport
1610 		 * interface which has inet6 addresses triggers
1611 		 * removal of all inet6 addresses on the member
1612 		 * interface.
1613 		 */
1614 		if (in6ifa_llaonifp(tpif)) {
1615 			in6_ifdetach(tpif);
1616 				if_printf(sc->sc_ifp,
1617 				    "IPv6 addresses on %s have been removed "
1618 				    "before adding it as a member to prevent "
1619 				    "IPv6 address scope violation.\n",
1620 				    tpif->if_xname);
1621 		}
1622 #endif
1623 		oldmtu = ifp->if_mtu;
1624 		LAGG_XLOCK(sc);
1625 		error = lagg_port_create(sc, tpif);
1626 		LAGG_XUNLOCK(sc);
1627 		if_rele(tpif);
1628 
1629 		/*
1630 		 * LAGG MTU may change during addition of the first port.
1631 		 * If it did, do network layer specific procedure.
1632 		 */
1633 		if (ifp->if_mtu != oldmtu)
1634 			if_notifymtu(ifp);
1635 
1636 		VLAN_CAPABILITIES(ifp);
1637 		break;
1638 	case SIOCSLAGGDELPORT:
1639 		error = priv_check(td, PRIV_NET_LAGG);
1640 		if (error)
1641 			break;
1642 		if (rp->rp_portname[0] == '\0' ||
1643 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1644 			error = EINVAL;
1645 			break;
1646 		}
1647 
1648 		LAGG_XLOCK(sc);
1649 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1650 		    lp->lp_softc != sc) {
1651 			error = ENOENT;
1652 			LAGG_XUNLOCK(sc);
1653 			if_rele(tpif);
1654 			break;
1655 		}
1656 
1657 		error = lagg_port_destroy(lp, 1);
1658 		LAGG_XUNLOCK(sc);
1659 		if_rele(tpif);
1660 		VLAN_CAPABILITIES(ifp);
1661 		break;
1662 	case SIOCSIFFLAGS:
1663 		/* Set flags on ports too */
1664 		LAGG_XLOCK(sc);
1665 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1666 			lagg_setflags(lp, 1);
1667 		}
1668 
1669 		if (!(ifp->if_flags & IFF_UP) &&
1670 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1671 			/*
1672 			 * If interface is marked down and it is running,
1673 			 * then stop and disable it.
1674 			 */
1675 			lagg_stop(sc);
1676 			LAGG_XUNLOCK(sc);
1677 		} else if ((ifp->if_flags & IFF_UP) &&
1678 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1679 			/*
1680 			 * If interface is marked up and it is stopped, then
1681 			 * start it.
1682 			 */
1683 			LAGG_XUNLOCK(sc);
1684 			(*ifp->if_init)(sc);
1685 		} else
1686 			LAGG_XUNLOCK(sc);
1687 		break;
1688 	case SIOCADDMULTI:
1689 	case SIOCDELMULTI:
1690 		LAGG_XLOCK(sc);
1691 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1692 			lagg_clrmulti(lp);
1693 			lagg_setmulti(lp);
1694 		}
1695 		LAGG_XUNLOCK(sc);
1696 		error = 0;
1697 		break;
1698 	case SIOCSIFMEDIA:
1699 	case SIOCGIFMEDIA:
1700 		if (ifp->if_type == IFT_INFINIBAND)
1701 			error = EINVAL;
1702 		else
1703 			error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1704 		break;
1705 
1706 	case SIOCSIFCAP:
1707 	case SIOCSIFCAPNV:
1708 		LAGG_XLOCK(sc);
1709 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1710 			if (lp->lp_ioctl != NULL)
1711 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1712 		}
1713 		lagg_capabilities(sc);
1714 		LAGG_XUNLOCK(sc);
1715 		VLAN_CAPABILITIES(ifp);
1716 		error = 0;
1717 		break;
1718 
1719 	case SIOCGIFCAPNV:
1720 		error = 0;
1721 		break;
1722 
1723 	case SIOCSIFMTU:
1724 		LAGG_XLOCK(sc);
1725 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1726 			if (lp->lp_ioctl != NULL)
1727 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1728 			else
1729 				error = EINVAL;
1730 			if (error != 0) {
1731 				if_printf(ifp,
1732 				    "failed to change MTU to %d on port %s, "
1733 				    "reverting all ports to original MTU (%d)\n",
1734 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1735 				break;
1736 			}
1737 		}
1738 		if (error == 0) {
1739 			ifp->if_mtu = ifr->ifr_mtu;
1740 		} else {
1741 			/* set every port back to the original MTU */
1742 			ifr->ifr_mtu = ifp->if_mtu;
1743 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1744 				if (lp->lp_ioctl != NULL)
1745 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1746 			}
1747 		}
1748 		lagg_capabilities(sc);
1749 		LAGG_XUNLOCK(sc);
1750 		VLAN_CAPABILITIES(ifp);
1751 		break;
1752 
1753 	default:
1754 		error = ether_ioctl(ifp, cmd, data);
1755 		break;
1756 	}
1757 	return (error);
1758 }
1759 
1760 #if defined(KERN_TLS) || defined(RATELIMIT)
1761 #ifdef RATELIMIT
1762 static const struct if_snd_tag_sw lagg_snd_tag_ul_sw = {
1763 	.snd_tag_modify = lagg_snd_tag_modify,
1764 	.snd_tag_query = lagg_snd_tag_query,
1765 	.snd_tag_free = lagg_snd_tag_free,
1766 	.next_snd_tag = lagg_next_snd_tag,
1767 	.type = IF_SND_TAG_TYPE_UNLIMITED
1768 };
1769 
1770 static const struct if_snd_tag_sw lagg_snd_tag_rl_sw = {
1771 	.snd_tag_modify = lagg_snd_tag_modify,
1772 	.snd_tag_query = lagg_snd_tag_query,
1773 	.snd_tag_free = lagg_snd_tag_free,
1774 	.next_snd_tag = lagg_next_snd_tag,
1775 	.type = IF_SND_TAG_TYPE_RATE_LIMIT
1776 };
1777 #endif
1778 
1779 #ifdef KERN_TLS
1780 static const struct if_snd_tag_sw lagg_snd_tag_tls_sw = {
1781 	.snd_tag_modify = lagg_snd_tag_modify,
1782 	.snd_tag_query = lagg_snd_tag_query,
1783 	.snd_tag_free = lagg_snd_tag_free,
1784 	.next_snd_tag = lagg_next_snd_tag,
1785 	.type = IF_SND_TAG_TYPE_TLS
1786 };
1787 
1788 #ifdef RATELIMIT
1789 static const struct if_snd_tag_sw lagg_snd_tag_tls_rl_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_RATE_LIMIT
1795 };
1796 #endif
1797 #endif
1798 
1799 static inline struct lagg_snd_tag *
1800 mst_to_lst(struct m_snd_tag *mst)
1801 {
1802 
1803 	return (__containerof(mst, struct lagg_snd_tag, com));
1804 }
1805 
1806 /*
1807  * Look up the port used by a specific flow.  This only works for lagg
1808  * protocols with deterministic port mappings (e.g. not roundrobin).
1809  * In addition protocols which use a hash to map flows to ports must
1810  * be configured to use the mbuf flowid rather than hashing packet
1811  * contents.
1812  */
1813 static struct lagg_port *
1814 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1815     uint8_t numa_domain)
1816 {
1817 	struct lagg_softc *sc;
1818 	struct lagg_port *lp;
1819 	struct lagg_lb *lb;
1820 	uint32_t hash, p;
1821 	int err;
1822 
1823 	sc = ifp->if_softc;
1824 
1825 	switch (sc->sc_proto) {
1826 	case LAGG_PROTO_FAILOVER:
1827 		return (lagg_link_active(sc, sc->sc_primary));
1828 	case LAGG_PROTO_LOADBALANCE:
1829 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1830 		    flowtype == M_HASHTYPE_NONE)
1831 			return (NULL);
1832 		p = flowid >> sc->flowid_shift;
1833 		p %= sc->sc_count;
1834 		lb = (struct lagg_lb *)sc->sc_psc;
1835 		lp = lb->lb_ports[p];
1836 		return (lagg_link_active(sc, lp));
1837 	case LAGG_PROTO_LACP:
1838 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1839 		    flowtype == M_HASHTYPE_NONE)
1840 			return (NULL);
1841 		hash = flowid >> sc->flowid_shift;
1842 		return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, &err));
1843 	default:
1844 		return (NULL);
1845 	}
1846 }
1847 
1848 static int
1849 lagg_snd_tag_alloc(struct ifnet *ifp,
1850     union if_snd_tag_alloc_params *params,
1851     struct m_snd_tag **ppmt)
1852 {
1853 	struct epoch_tracker et;
1854 	const struct if_snd_tag_sw *sw;
1855 	struct lagg_snd_tag *lst;
1856 	struct lagg_port *lp;
1857 	struct ifnet *lp_ifp;
1858 	struct m_snd_tag *mst;
1859 	int error;
1860 
1861 	switch (params->hdr.type) {
1862 #ifdef RATELIMIT
1863 	case IF_SND_TAG_TYPE_UNLIMITED:
1864 		sw = &lagg_snd_tag_ul_sw;
1865 		break;
1866 	case IF_SND_TAG_TYPE_RATE_LIMIT:
1867 		sw = &lagg_snd_tag_rl_sw;
1868 		break;
1869 #endif
1870 #ifdef KERN_TLS
1871 	case IF_SND_TAG_TYPE_TLS:
1872 		sw = &lagg_snd_tag_tls_sw;
1873 		break;
1874 	case IF_SND_TAG_TYPE_TLS_RX:
1875 		/* Return tag from port interface directly. */
1876 		sw = NULL;
1877 		break;
1878 #ifdef RATELIMIT
1879 	case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
1880 		sw = &lagg_snd_tag_tls_rl_sw;
1881 		break;
1882 #endif
1883 #endif
1884 	default:
1885 		return (EOPNOTSUPP);
1886 	}
1887 
1888 	NET_EPOCH_ENTER(et);
1889 	lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1890 	    params->hdr.flowtype, params->hdr.numa_domain);
1891 	if (lp == NULL) {
1892 		NET_EPOCH_EXIT(et);
1893 		return (EOPNOTSUPP);
1894 	}
1895 	if (lp->lp_ifp == NULL) {
1896 		NET_EPOCH_EXIT(et);
1897 		return (EOPNOTSUPP);
1898 	}
1899 	lp_ifp = lp->lp_ifp;
1900 	if_ref(lp_ifp);
1901 	NET_EPOCH_EXIT(et);
1902 
1903 	if (sw != NULL) {
1904 		lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1905 		if (lst == NULL) {
1906 			if_rele(lp_ifp);
1907 			return (ENOMEM);
1908 		}
1909 	} else
1910 		lst = NULL;
1911 
1912 	error = m_snd_tag_alloc(lp_ifp, params, &mst);
1913 	if_rele(lp_ifp);
1914 	if (error) {
1915 		free(lst, M_LAGG);
1916 		return (error);
1917 	}
1918 
1919 	if (sw != NULL) {
1920 		m_snd_tag_init(&lst->com, ifp, sw);
1921 		lst->tag = mst;
1922 
1923 		*ppmt = &lst->com;
1924 	} else
1925 		*ppmt = mst;
1926 
1927 	return (0);
1928 }
1929 
1930 static struct m_snd_tag *
1931 lagg_next_snd_tag(struct m_snd_tag *mst)
1932 {
1933 	struct lagg_snd_tag *lst;
1934 
1935 	lst = mst_to_lst(mst);
1936 	return (lst->tag);
1937 }
1938 
1939 static int
1940 lagg_snd_tag_modify(struct m_snd_tag *mst,
1941     union if_snd_tag_modify_params *params)
1942 {
1943 	struct lagg_snd_tag *lst;
1944 
1945 	lst = mst_to_lst(mst);
1946 	return (lst->tag->sw->snd_tag_modify(lst->tag, params));
1947 }
1948 
1949 static int
1950 lagg_snd_tag_query(struct m_snd_tag *mst,
1951     union if_snd_tag_query_params *params)
1952 {
1953 	struct lagg_snd_tag *lst;
1954 
1955 	lst = mst_to_lst(mst);
1956 	return (lst->tag->sw->snd_tag_query(lst->tag, params));
1957 }
1958 
1959 static void
1960 lagg_snd_tag_free(struct m_snd_tag *mst)
1961 {
1962 	struct lagg_snd_tag *lst;
1963 
1964 	lst = mst_to_lst(mst);
1965 	m_snd_tag_rele(lst->tag);
1966 	free(lst, M_LAGG);
1967 }
1968 
1969 static void
1970 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1971 {
1972 	/*
1973 	 * For lagg, we have an indirect
1974 	 * interface. The caller needs to
1975 	 * get a ratelimit tag on the actual
1976 	 * interface the flow will go on.
1977 	 */
1978 	q->rate_table = NULL;
1979 	q->flags = RT_IS_INDIRECT;
1980 	q->max_flows = 0;
1981 	q->number_of_rates = 0;
1982 }
1983 #endif
1984 
1985 static int
1986 lagg_setmulti(struct lagg_port *lp)
1987 {
1988 	struct lagg_softc *sc = lp->lp_softc;
1989 	struct ifnet *ifp = lp->lp_ifp;
1990 	struct ifnet *scifp = sc->sc_ifp;
1991 	struct lagg_mc *mc;
1992 	struct ifmultiaddr *ifma;
1993 	int error;
1994 
1995 	IF_ADDR_WLOCK(scifp);
1996 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1997 		if (ifma->ifma_addr->sa_family != AF_LINK)
1998 			continue;
1999 		mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
2000 		if (mc == NULL) {
2001 			IF_ADDR_WUNLOCK(scifp);
2002 			return (ENOMEM);
2003 		}
2004 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
2005 		mc->mc_addr.sdl_index = ifp->if_index;
2006 		mc->mc_ifma = NULL;
2007 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
2008 	}
2009 	IF_ADDR_WUNLOCK(scifp);
2010 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
2011 		error = if_addmulti(ifp,
2012 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
2013 		if (error)
2014 			return (error);
2015 	}
2016 	return (0);
2017 }
2018 
2019 static int
2020 lagg_clrmulti(struct lagg_port *lp)
2021 {
2022 	struct lagg_mc *mc;
2023 
2024 	LAGG_XLOCK_ASSERT(lp->lp_softc);
2025 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
2026 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
2027 		if (mc->mc_ifma && lp->lp_detaching == 0)
2028 			if_delmulti_ifma(mc->mc_ifma);
2029 		free(mc, M_LAGG);
2030 	}
2031 	return (0);
2032 }
2033 
2034 static void
2035 lagg_setcaps(struct lagg_port *lp, int cap, int cap2)
2036 {
2037 	struct ifreq ifr;
2038 	struct siocsifcapnv_driver_data drv_ioctl_data;
2039 
2040 	if (lp->lp_ifp->if_capenable == cap &&
2041 	    lp->lp_ifp->if_capenable2 == cap2)
2042 		return;
2043 	if (lp->lp_ioctl == NULL)
2044 		return;
2045 	/* XXX */
2046 	if ((lp->lp_ifp->if_capabilities & IFCAP_NV) != 0) {
2047 		drv_ioctl_data.reqcap = cap;
2048 		drv_ioctl_data.reqcap2 = cap2;
2049 		drv_ioctl_data.nvcap = NULL;
2050 		(*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAPNV,
2051 		    (caddr_t)&drv_ioctl_data);
2052 	} else {
2053 		ifr.ifr_reqcap = cap;
2054 		(*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr);
2055 	}
2056 }
2057 
2058 /* Handle a ref counted flag that should be set on the lagg port as well */
2059 static int
2060 lagg_setflag(struct lagg_port *lp, int flag, int status,
2061     int (*func)(struct ifnet *, int))
2062 {
2063 	struct lagg_softc *sc = lp->lp_softc;
2064 	struct ifnet *scifp = sc->sc_ifp;
2065 	struct ifnet *ifp = lp->lp_ifp;
2066 	int error;
2067 
2068 	LAGG_XLOCK_ASSERT(sc);
2069 
2070 	status = status ? (scifp->if_flags & flag) : 0;
2071 	/* Now "status" contains the flag value or 0 */
2072 
2073 	/*
2074 	 * See if recorded ports status is different from what
2075 	 * we want it to be.  If it is, flip it.  We record ports
2076 	 * status in lp_ifflags so that we won't clear ports flag
2077 	 * we haven't set.  In fact, we don't clear or set ports
2078 	 * flags directly, but get or release references to them.
2079 	 * That's why we can be sure that recorded flags still are
2080 	 * in accord with actual ports flags.
2081 	 */
2082 	if (status != (lp->lp_ifflags & flag)) {
2083 		error = (*func)(ifp, status);
2084 		if (error)
2085 			return (error);
2086 		lp->lp_ifflags &= ~flag;
2087 		lp->lp_ifflags |= status;
2088 	}
2089 	return (0);
2090 }
2091 
2092 /*
2093  * Handle IFF_* flags that require certain changes on the lagg port
2094  * if "status" is true, update ports flags respective to the lagg
2095  * if "status" is false, forcedly clear the flags set on port.
2096  */
2097 static int
2098 lagg_setflags(struct lagg_port *lp, int status)
2099 {
2100 	int error, i;
2101 
2102 	for (i = 0; lagg_pflags[i].flag; i++) {
2103 		error = lagg_setflag(lp, lagg_pflags[i].flag,
2104 		    status, lagg_pflags[i].func);
2105 		if (error)
2106 			return (error);
2107 	}
2108 	return (0);
2109 }
2110 
2111 static int
2112 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2113 {
2114 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2115 
2116 	NET_EPOCH_ASSERT();
2117 #if defined(KERN_TLS) || defined(RATELIMIT)
2118 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2119 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2120 #endif
2121 	/* We need a Tx algorithm and at least one port */
2122 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2123 		m_freem(m);
2124 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2125 		return (ENXIO);
2126 	}
2127 
2128 	ETHER_BPF_MTAP(ifp, m);
2129 
2130 	return (lagg_proto_start(sc, m));
2131 }
2132 
2133 static int
2134 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2135 {
2136 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2137 
2138 	NET_EPOCH_ASSERT();
2139 #if defined(KERN_TLS) || defined(RATELIMIT)
2140 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2141 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2142 #endif
2143 	/* We need a Tx algorithm and at least one port */
2144 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2145 		m_freem(m);
2146 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2147 		return (ENXIO);
2148 	}
2149 
2150 	infiniband_bpf_mtap(ifp, m);
2151 
2152 	return (lagg_proto_start(sc, m));
2153 }
2154 
2155 /*
2156  * The ifp->if_qflush entry point for lagg(4) is no-op.
2157  */
2158 static void
2159 lagg_qflush(struct ifnet *ifp __unused)
2160 {
2161 }
2162 
2163 static struct mbuf *
2164 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2165 {
2166 	struct lagg_port *lp = ifp->if_lagg;
2167 	struct lagg_softc *sc = lp->lp_softc;
2168 	struct ifnet *scifp = sc->sc_ifp;
2169 
2170 	NET_EPOCH_ASSERT();
2171 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2172 	    lp->lp_detaching != 0 ||
2173 	    sc->sc_proto == LAGG_PROTO_NONE) {
2174 		m_freem(m);
2175 		return (NULL);
2176 	}
2177 
2178 	m = lagg_proto_input(sc, lp, m);
2179 	if (m != NULL) {
2180 		ETHER_BPF_MTAP(scifp, m);
2181 
2182 		if ((scifp->if_flags & IFF_MONITOR) != 0) {
2183 			m_freem(m);
2184 			m = NULL;
2185 		}
2186 	}
2187 
2188 #ifdef DEV_NETMAP
2189 	if (m != NULL && scifp->if_capenable & IFCAP_NETMAP) {
2190 		scifp->if_input(scifp, m);
2191 		m = NULL;
2192 	}
2193 #endif	/* DEV_NETMAP */
2194 
2195 	return (m);
2196 }
2197 
2198 static struct mbuf *
2199 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2200 {
2201 	struct lagg_port *lp = ifp->if_lagg;
2202 	struct lagg_softc *sc = lp->lp_softc;
2203 	struct ifnet *scifp = sc->sc_ifp;
2204 
2205 	NET_EPOCH_ASSERT();
2206 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2207 	    lp->lp_detaching != 0 ||
2208 	    sc->sc_proto == LAGG_PROTO_NONE) {
2209 		m_freem(m);
2210 		return (NULL);
2211 	}
2212 
2213 	m = lagg_proto_input(sc, lp, m);
2214 	if (m != NULL) {
2215 		infiniband_bpf_mtap(scifp, m);
2216 
2217 		if ((scifp->if_flags & IFF_MONITOR) != 0) {
2218 			m_freem(m);
2219 			m = NULL;
2220 		}
2221 	}
2222 
2223 	return (m);
2224 }
2225 
2226 static int
2227 lagg_media_change(struct ifnet *ifp)
2228 {
2229 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2230 
2231 	if (sc->sc_ifflags & IFF_DEBUG)
2232 		printf("%s\n", __func__);
2233 
2234 	/* Ignore */
2235 	return (0);
2236 }
2237 
2238 static void
2239 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2240 {
2241 	struct epoch_tracker et;
2242 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2243 	struct lagg_port *lp;
2244 
2245 	imr->ifm_status = IFM_AVALID;
2246 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
2247 
2248 	NET_EPOCH_ENTER(et);
2249 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2250 		if (LAGG_PORTACTIVE(lp))
2251 			imr->ifm_status |= IFM_ACTIVE;
2252 	}
2253 	NET_EPOCH_EXIT(et);
2254 }
2255 
2256 static void
2257 lagg_linkstate(struct lagg_softc *sc)
2258 {
2259 	struct epoch_tracker et;
2260 	struct lagg_port *lp;
2261 	int new_link = LINK_STATE_DOWN;
2262 	uint64_t speed;
2263 
2264 	LAGG_XLOCK_ASSERT(sc);
2265 
2266 	/* LACP handles link state itself */
2267 	if (sc->sc_proto == LAGG_PROTO_LACP)
2268 		return;
2269 
2270 	/* Our link is considered up if at least one of our ports is active */
2271 	NET_EPOCH_ENTER(et);
2272 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2273 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2274 			new_link = LINK_STATE_UP;
2275 			break;
2276 		}
2277 	}
2278 	NET_EPOCH_EXIT(et);
2279 	if_link_state_change(sc->sc_ifp, new_link);
2280 
2281 	/* Update if_baudrate to reflect the max possible speed */
2282 	switch (sc->sc_proto) {
2283 		case LAGG_PROTO_FAILOVER:
2284 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2285 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
2286 			break;
2287 		case LAGG_PROTO_ROUNDROBIN:
2288 		case LAGG_PROTO_LOADBALANCE:
2289 		case LAGG_PROTO_BROADCAST:
2290 			speed = 0;
2291 			NET_EPOCH_ENTER(et);
2292 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2293 				speed += lp->lp_ifp->if_baudrate;
2294 			NET_EPOCH_EXIT(et);
2295 			sc->sc_ifp->if_baudrate = speed;
2296 			break;
2297 		case LAGG_PROTO_LACP:
2298 			/* LACP updates if_baudrate itself */
2299 			break;
2300 	}
2301 }
2302 
2303 static void
2304 lagg_port_state(struct ifnet *ifp, int state)
2305 {
2306 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2307 	struct lagg_softc *sc = NULL;
2308 
2309 	if (lp != NULL)
2310 		sc = lp->lp_softc;
2311 	if (sc == NULL)
2312 		return;
2313 
2314 	LAGG_XLOCK(sc);
2315 	lagg_linkstate(sc);
2316 	lagg_proto_linkstate(sc, lp);
2317 	LAGG_XUNLOCK(sc);
2318 }
2319 
2320 struct lagg_port *
2321 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2322 {
2323 	struct lagg_port *lp_next, *rval = NULL;
2324 
2325 	/*
2326 	 * Search a port which reports an active link state.
2327 	 */
2328 
2329 #ifdef INVARIANTS
2330 	/*
2331 	 * This is called with either in the network epoch
2332 	 * or with LAGG_XLOCK(sc) held.
2333 	 */
2334 	if (!in_epoch(net_epoch_preempt))
2335 		LAGG_XLOCK_ASSERT(sc);
2336 #endif
2337 
2338 	if (lp == NULL)
2339 		goto search;
2340 	if (LAGG_PORTACTIVE(lp)) {
2341 		rval = lp;
2342 		goto found;
2343 	}
2344 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2345 	    LAGG_PORTACTIVE(lp_next)) {
2346 		rval = lp_next;
2347 		goto found;
2348 	}
2349 
2350 search:
2351 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2352 		if (LAGG_PORTACTIVE(lp_next)) {
2353 			return (lp_next);
2354 		}
2355 	}
2356 found:
2357 	return (rval);
2358 }
2359 
2360 int
2361 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2362 {
2363 
2364 #if defined(KERN_TLS) || defined(RATELIMIT)
2365 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2366 		struct lagg_snd_tag *lst;
2367 		struct m_snd_tag *mst;
2368 
2369 		mst = m->m_pkthdr.snd_tag;
2370 		lst = mst_to_lst(mst);
2371 		if (lst->tag->ifp != ifp) {
2372 			m_freem(m);
2373 			return (EAGAIN);
2374 		}
2375 		m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2376 		m_snd_tag_rele(mst);
2377 	}
2378 #endif
2379 	return (ifp->if_transmit)(ifp, m);
2380 }
2381 
2382 /*
2383  * Simple round robin aggregation
2384  */
2385 static void
2386 lagg_rr_attach(struct lagg_softc *sc)
2387 {
2388 	sc->sc_seq = 0;
2389 	sc->sc_stride = 1;
2390 }
2391 
2392 static int
2393 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2394 {
2395 	struct lagg_port *lp;
2396 	uint32_t p;
2397 
2398 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
2399 	p /= sc->sc_stride;
2400 	p %= sc->sc_count;
2401 	lp = CK_SLIST_FIRST(&sc->sc_ports);
2402 
2403 	while (p--)
2404 		lp = CK_SLIST_NEXT(lp, lp_entries);
2405 
2406 	/*
2407 	 * Check the port's link state. This will return the next active
2408 	 * port if the link is down or the port is NULL.
2409 	 */
2410 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2411 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2412 		m_freem(m);
2413 		return (ENETDOWN);
2414 	}
2415 
2416 	/* Send mbuf */
2417 	return (lagg_enqueue(lp->lp_ifp, m));
2418 }
2419 
2420 /*
2421  * Broadcast mode
2422  */
2423 static int
2424 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2425 {
2426 	int errors = 0;
2427 	int ret;
2428 	struct lagg_port *lp, *last = NULL;
2429 	struct mbuf *m0;
2430 
2431 	NET_EPOCH_ASSERT();
2432 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2433 		if (!LAGG_PORTACTIVE(lp))
2434 			continue;
2435 
2436 		if (last != NULL) {
2437 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2438 			if (m0 == NULL) {
2439 				ret = ENOBUFS;
2440 				errors++;
2441 				break;
2442 			}
2443 			lagg_enqueue(last->lp_ifp, m0);
2444 		}
2445 		last = lp;
2446 	}
2447 
2448 	if (last == NULL) {
2449 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2450 		m_freem(m);
2451 		return (ENOENT);
2452 	}
2453 	if ((last = lagg_link_active(sc, last)) == NULL) {
2454 		errors++;
2455 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2456 		m_freem(m);
2457 		return (ENETDOWN);
2458 	}
2459 
2460 	ret = lagg_enqueue(last->lp_ifp, m);
2461 	if (errors != 0)
2462 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2463 
2464 	return (ret);
2465 }
2466 
2467 /*
2468  * Active failover
2469  */
2470 static int
2471 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2472 {
2473 	struct lagg_port *lp;
2474 
2475 	/* Use the master port if active or the next available port */
2476 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2477 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2478 		m_freem(m);
2479 		return (ENETDOWN);
2480 	}
2481 
2482 	/* Send mbuf */
2483 	return (lagg_enqueue(lp->lp_ifp, m));
2484 }
2485 
2486 static struct mbuf *
2487 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2488 {
2489 	struct ifnet *ifp = sc->sc_ifp;
2490 	struct lagg_port *tmp_tp;
2491 
2492 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2493 		m->m_pkthdr.rcvif = ifp;
2494 		return (m);
2495 	}
2496 
2497 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2498 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2499 		/*
2500 		 * If tmp_tp is null, we've received a packet when all
2501 		 * our links are down. Weird, but process it anyways.
2502 		 */
2503 		if (tmp_tp == NULL || tmp_tp == lp) {
2504 			m->m_pkthdr.rcvif = ifp;
2505 			return (m);
2506 		}
2507 	}
2508 
2509 	m_freem(m);
2510 	return (NULL);
2511 }
2512 
2513 /*
2514  * Loadbalancing
2515  */
2516 static void
2517 lagg_lb_attach(struct lagg_softc *sc)
2518 {
2519 	struct lagg_port *lp;
2520 	struct lagg_lb *lb;
2521 
2522 	LAGG_XLOCK_ASSERT(sc);
2523 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2524 	lb->lb_key = m_ether_tcpip_hash_init();
2525 	sc->sc_psc = lb;
2526 
2527 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2528 		lagg_lb_port_create(lp);
2529 }
2530 
2531 static void
2532 lagg_lb_detach(struct lagg_softc *sc)
2533 {
2534 	struct lagg_lb *lb;
2535 
2536 	lb = (struct lagg_lb *)sc->sc_psc;
2537 	if (lb != NULL)
2538 		free(lb, M_LAGG);
2539 }
2540 
2541 static int
2542 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2543 {
2544 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2545 	struct lagg_port *lp_next;
2546 	int i = 0, rv;
2547 
2548 	rv = 0;
2549 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2550 	LAGG_XLOCK_ASSERT(sc);
2551 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2552 		if (lp_next == lp)
2553 			continue;
2554 		if (i >= LAGG_MAX_PORTS) {
2555 			rv = EINVAL;
2556 			break;
2557 		}
2558 		if (sc->sc_ifflags & IFF_DEBUG)
2559 			printf("%s: port %s at index %d\n",
2560 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2561 		lb->lb_ports[i++] = lp_next;
2562 	}
2563 
2564 	return (rv);
2565 }
2566 
2567 static int
2568 lagg_lb_port_create(struct lagg_port *lp)
2569 {
2570 	struct lagg_softc *sc = lp->lp_softc;
2571 	return (lagg_lb_porttable(sc, NULL));
2572 }
2573 
2574 static void
2575 lagg_lb_port_destroy(struct lagg_port *lp)
2576 {
2577 	struct lagg_softc *sc = lp->lp_softc;
2578 	lagg_lb_porttable(sc, lp);
2579 }
2580 
2581 static int
2582 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2583 {
2584 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2585 	struct lagg_port *lp = NULL;
2586 	uint32_t p = 0;
2587 
2588 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2589 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2590 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2591 	else
2592 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2593 	p %= sc->sc_count;
2594 	lp = lb->lb_ports[p];
2595 
2596 	/*
2597 	 * Check the port's link state. This will return the next active
2598 	 * port if the link is down or the port is NULL.
2599 	 */
2600 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2601 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2602 		m_freem(m);
2603 		return (ENETDOWN);
2604 	}
2605 
2606 	/* Send mbuf */
2607 	return (lagg_enqueue(lp->lp_ifp, m));
2608 }
2609 
2610 /*
2611  * 802.3ad LACP
2612  */
2613 static void
2614 lagg_lacp_attach(struct lagg_softc *sc)
2615 {
2616 	struct lagg_port *lp;
2617 
2618 	lacp_attach(sc);
2619 	LAGG_XLOCK_ASSERT(sc);
2620 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2621 		lacp_port_create(lp);
2622 }
2623 
2624 static void
2625 lagg_lacp_detach(struct lagg_softc *sc)
2626 {
2627 	struct lagg_port *lp;
2628 	void *psc;
2629 
2630 	LAGG_XLOCK_ASSERT(sc);
2631 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2632 		lacp_port_destroy(lp);
2633 
2634 	psc = sc->sc_psc;
2635 	sc->sc_psc = NULL;
2636 	lacp_detach(psc);
2637 }
2638 
2639 static void
2640 lagg_lacp_lladdr(struct lagg_softc *sc)
2641 {
2642 	struct lagg_port *lp;
2643 
2644 	LAGG_SXLOCK_ASSERT(sc);
2645 
2646 	/* purge all the lacp ports */
2647 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2648 		lacp_port_destroy(lp);
2649 
2650 	/* add them back in */
2651 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2652 		lacp_port_create(lp);
2653 }
2654 
2655 static int
2656 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2657 {
2658 	struct lagg_port *lp;
2659 	int err;
2660 
2661 	lp = lacp_select_tx_port(sc, m, &err);
2662 	if (lp == NULL) {
2663 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2664 		m_freem(m);
2665 		return (err);
2666 	}
2667 
2668 	/* Send mbuf */
2669 	return (lagg_enqueue(lp->lp_ifp, m));
2670 }
2671 
2672 static struct mbuf *
2673 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2674 {
2675 	struct ifnet *ifp = sc->sc_ifp;
2676 	struct ether_header *eh;
2677 	u_short etype;
2678 
2679 	eh = mtod(m, struct ether_header *);
2680 	etype = ntohs(eh->ether_type);
2681 
2682 	/* Tap off LACP control messages */
2683 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2684 		m = lacp_input(lp, m);
2685 		if (m == NULL)
2686 			return (NULL);
2687 	}
2688 
2689 	/*
2690 	 * If the port is not collecting or not in the active aggregator then
2691 	 * free and return.
2692 	 */
2693 	if (!lacp_iscollecting(lp) || !lacp_isactive(lp)) {
2694 		m_freem(m);
2695 		return (NULL);
2696 	}
2697 
2698 	m->m_pkthdr.rcvif = ifp;
2699 	return (m);
2700 }
2701 
2702 /* Default input */
2703 static struct mbuf *
2704 lagg_default_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2705 {
2706 	struct ifnet *ifp = sc->sc_ifp;
2707 
2708 	/* Just pass in the packet to our lagg device */
2709 	m->m_pkthdr.rcvif = ifp;
2710 
2711 	return (m);
2712 }
2713