xref: /freebsd-14.2/sys/net/if_lagg.c (revision 7126da66)
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 <[email protected]>
5  * Copyright (c) 2007 Andrew Thompson <[email protected]>
6  * Copyright (c) 2014, 2016 Marcelo Araujo <[email protected]>
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_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
87 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
88 
89 /* Special flags we should propagate to the lagg ports. */
90 static struct {
91 	int flag;
92 	int (*func)(struct ifnet *, int);
93 } lagg_pflags[] = {
94 	{IFF_PROMISC, ifpromisc},
95 	{IFF_ALLMULTI, if_allmulti},
96 	{0, NULL}
97 };
98 
99 struct lagg_snd_tag {
100 	struct m_snd_tag com;
101 	struct m_snd_tag *tag;
102 };
103 
104 VNET_DEFINE_STATIC(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
105 #define	V_lagg_list	VNET(lagg_list)
106 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
107 #define	V_lagg_list_mtx	VNET(lagg_list_mtx)
108 #define	LAGG_LIST_LOCK_INIT(x)		mtx_init(&V_lagg_list_mtx, \
109 					"if_lagg list", NULL, MTX_DEF)
110 #define	LAGG_LIST_LOCK_DESTROY(x)	mtx_destroy(&V_lagg_list_mtx)
111 #define	LAGG_LIST_LOCK(x)		mtx_lock(&V_lagg_list_mtx)
112 #define	LAGG_LIST_UNLOCK(x)		mtx_unlock(&V_lagg_list_mtx)
113 static eventhandler_tag	lagg_detach_cookie = NULL;
114 
115 static int	lagg_clone_create(struct if_clone *, char *, size_t,
116 		    struct ifc_data *, struct ifnet **);
117 static int	lagg_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
118 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
119 #define	V_lagg_cloner	VNET(lagg_cloner)
120 static const char laggname[] = "lagg";
121 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
122 
123 static void	lagg_capabilities(struct lagg_softc *);
124 static int	lagg_port_create(struct lagg_softc *, struct ifnet *);
125 static int	lagg_port_destroy(struct lagg_port *, int);
126 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *);
127 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *);
128 static void	lagg_linkstate(struct lagg_softc *);
129 static void	lagg_port_state(struct ifnet *, int);
130 static int	lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
131 static int	lagg_port_output(struct ifnet *, struct mbuf *,
132 		    const struct sockaddr *, struct route *);
133 static void	lagg_port_ifdetach(void *arg __unused, struct ifnet *);
134 #ifdef LAGG_PORT_STACKING
135 static int	lagg_port_checkstacking(struct lagg_softc *);
136 #endif
137 static void	lagg_port2req(struct lagg_port *, struct lagg_reqport *);
138 static void	lagg_init(void *);
139 static void	lagg_stop(struct lagg_softc *);
140 static int	lagg_ioctl(struct ifnet *, u_long, caddr_t);
141 #if defined(KERN_TLS) || defined(RATELIMIT)
142 static int	lagg_snd_tag_alloc(struct ifnet *,
143 		    union if_snd_tag_alloc_params *,
144 		    struct m_snd_tag **);
145 static int	lagg_snd_tag_modify(struct m_snd_tag *,
146 		    union if_snd_tag_modify_params *);
147 static int	lagg_snd_tag_query(struct m_snd_tag *,
148 		    union if_snd_tag_query_params *);
149 static void	lagg_snd_tag_free(struct m_snd_tag *);
150 static struct m_snd_tag *lagg_next_snd_tag(struct m_snd_tag *);
151 static void	lagg_ratelimit_query(struct ifnet *,
152 		    struct if_ratelimit_query_results *);
153 #endif
154 static int	lagg_setmulti(struct lagg_port *);
155 static int	lagg_clrmulti(struct lagg_port *);
156 static void	lagg_setcaps(struct lagg_port *, int cap, int cap2);
157 static int	lagg_setflag(struct lagg_port *, int, int,
158 		    int (*func)(struct ifnet *, int));
159 static int	lagg_setflags(struct lagg_port *, int status);
160 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
161 static int	lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
162 static int	lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
163 static void	lagg_qflush(struct ifnet *);
164 static int	lagg_media_change(struct ifnet *);
165 static void	lagg_media_status(struct ifnet *, struct ifmediareq *);
166 static struct lagg_port *lagg_link_active(struct lagg_softc *,
167 		    struct lagg_port *);
168 
169 /* Simple round robin */
170 static void	lagg_rr_attach(struct lagg_softc *);
171 static int	lagg_rr_start(struct lagg_softc *, struct mbuf *);
172 
173 /* Active failover */
174 static int	lagg_fail_start(struct lagg_softc *, struct mbuf *);
175 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
176 		    struct mbuf *);
177 
178 /* Loadbalancing */
179 static void	lagg_lb_attach(struct lagg_softc *);
180 static void	lagg_lb_detach(struct lagg_softc *);
181 static int	lagg_lb_port_create(struct lagg_port *);
182 static void	lagg_lb_port_destroy(struct lagg_port *);
183 static int	lagg_lb_start(struct lagg_softc *, struct mbuf *);
184 static int	lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
185 
186 /* Broadcast */
187 static int	lagg_bcast_start(struct lagg_softc *, struct mbuf *);
188 
189 /* 802.3ad LACP */
190 static void	lagg_lacp_attach(struct lagg_softc *);
191 static void	lagg_lacp_detach(struct lagg_softc *);
192 static int	lagg_lacp_start(struct lagg_softc *, struct mbuf *);
193 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
194 		    struct mbuf *);
195 static void	lagg_lacp_lladdr(struct lagg_softc *);
196 
197 /* Default input */
198 static struct mbuf *lagg_default_input(struct lagg_softc *, struct lagg_port *,
199 		    struct mbuf *);
200 
201 /* lagg protocol table */
202 static const struct lagg_proto {
203 	lagg_proto	pr_num;
204 	void		(*pr_attach)(struct lagg_softc *);
205 	void		(*pr_detach)(struct lagg_softc *);
206 	int		(*pr_start)(struct lagg_softc *, struct mbuf *);
207 	struct mbuf *	(*pr_input)(struct lagg_softc *, struct lagg_port *,
208 			    struct mbuf *);
209 	int		(*pr_addport)(struct lagg_port *);
210 	void		(*pr_delport)(struct lagg_port *);
211 	void		(*pr_linkstate)(struct lagg_port *);
212 	void 		(*pr_init)(struct lagg_softc *);
213 	void 		(*pr_stop)(struct lagg_softc *);
214 	void 		(*pr_lladdr)(struct lagg_softc *);
215 	void		(*pr_request)(struct lagg_softc *, void *);
216 	void		(*pr_portreq)(struct lagg_port *, void *);
217 } lagg_protos[] = {
218     {
219 	.pr_num = LAGG_PROTO_NONE
220     },
221     {
222 	.pr_num = LAGG_PROTO_ROUNDROBIN,
223 	.pr_attach = lagg_rr_attach,
224 	.pr_start = lagg_rr_start,
225 	.pr_input = lagg_default_input,
226     },
227     {
228 	.pr_num = LAGG_PROTO_FAILOVER,
229 	.pr_start = lagg_fail_start,
230 	.pr_input = lagg_fail_input,
231     },
232     {
233 	.pr_num = LAGG_PROTO_LOADBALANCE,
234 	.pr_attach = lagg_lb_attach,
235 	.pr_detach = lagg_lb_detach,
236 	.pr_start = lagg_lb_start,
237 	.pr_input = lagg_default_input,
238 	.pr_addport = lagg_lb_port_create,
239 	.pr_delport = lagg_lb_port_destroy,
240     },
241     {
242 	.pr_num = LAGG_PROTO_LACP,
243 	.pr_attach = lagg_lacp_attach,
244 	.pr_detach = lagg_lacp_detach,
245 	.pr_start = lagg_lacp_start,
246 	.pr_input = lagg_lacp_input,
247 	.pr_addport = lacp_port_create,
248 	.pr_delport = lacp_port_destroy,
249 	.pr_linkstate = lacp_linkstate,
250 	.pr_init = lacp_init,
251 	.pr_stop = lacp_stop,
252 	.pr_lladdr = lagg_lacp_lladdr,
253 	.pr_request = lacp_req,
254 	.pr_portreq = lacp_portreq,
255     },
256     {
257 	.pr_num = LAGG_PROTO_BROADCAST,
258 	.pr_start = lagg_bcast_start,
259 	.pr_input = lagg_default_input,
260     },
261 };
262 
263 SYSCTL_DECL(_net_link);
264 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
265     "Link Aggregation");
266 
267 /* Allow input on any failover links */
268 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
269 #define	V_lagg_failover_rx_all	VNET(lagg_failover_rx_all)
270 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
271     &VNET_NAME(lagg_failover_rx_all), 0,
272     "Accept input from any interface in a failover lagg");
273 
274 /* Default value for using flowid */
275 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
276 #define	V_def_use_flowid	VNET(def_use_flowid)
277 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid,
278     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_use_flowid), 0,
279     "Default setting for using flow id for load sharing");
280 
281 /* Default value for using numa */
282 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
283 #define	V_def_use_numa	VNET(def_use_numa)
284 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa,
285     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_use_numa), 0,
286     "Use numa to steer flows");
287 
288 /* Default value for flowid shift */
289 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
290 #define	V_def_flowid_shift	VNET(def_flowid_shift)
291 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift,
292     CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_flowid_shift), 0,
293     "Default setting for flowid shift for load sharing");
294 
295 static void
vnet_lagg_init(const void * unused __unused)296 vnet_lagg_init(const void *unused __unused)
297 {
298 
299 	LAGG_LIST_LOCK_INIT();
300 	SLIST_INIT(&V_lagg_list);
301 	struct if_clone_addreq req = {
302 		.create_f = lagg_clone_create,
303 		.destroy_f = lagg_clone_destroy,
304 		.flags = IFC_F_AUTOUNIT,
305 	};
306 	V_lagg_cloner = ifc_attach_cloner(laggname, &req);
307 }
308 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
309     vnet_lagg_init, NULL);
310 
311 static void
vnet_lagg_uninit(const void * unused __unused)312 vnet_lagg_uninit(const void *unused __unused)
313 {
314 
315 	ifc_detach_cloner(V_lagg_cloner);
316 	LAGG_LIST_LOCK_DESTROY();
317 }
318 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
319     vnet_lagg_uninit, NULL);
320 
321 static int
lagg_modevent(module_t mod,int type,void * data)322 lagg_modevent(module_t mod, int type, void *data)
323 {
324 
325 	switch (type) {
326 	case MOD_LOAD:
327 		lagg_input_ethernet_p = lagg_input_ethernet;
328 		lagg_input_infiniband_p = lagg_input_infiniband;
329 		lagg_linkstate_p = lagg_port_state;
330 		lagg_detach_cookie = EVENTHANDLER_REGISTER(
331 		    ifnet_departure_event, lagg_port_ifdetach, NULL,
332 		    EVENTHANDLER_PRI_ANY);
333 		break;
334 	case MOD_UNLOAD:
335 		EVENTHANDLER_DEREGISTER(ifnet_departure_event,
336 		    lagg_detach_cookie);
337 		lagg_input_ethernet_p = NULL;
338 		lagg_input_infiniband_p = NULL;
339 		lagg_linkstate_p = NULL;
340 		break;
341 	default:
342 		return (EOPNOTSUPP);
343 	}
344 	return (0);
345 }
346 
347 static moduledata_t lagg_mod = {
348 	"if_lagg",
349 	lagg_modevent,
350 	0
351 };
352 
353 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
354 MODULE_VERSION(if_lagg, 1);
355 MODULE_DEPEND(if_lagg, if_infiniband, 1, 1, 1);
356 
357 static void
lagg_proto_attach(struct lagg_softc * sc,lagg_proto pr)358 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
359 {
360 
361 	LAGG_XLOCK_ASSERT(sc);
362 	KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
363 	    __func__, sc));
364 
365 	if (sc->sc_ifflags & IFF_DEBUG)
366 		if_printf(sc->sc_ifp, "using proto %u\n", pr);
367 
368 	if (lagg_protos[pr].pr_attach != NULL)
369 		lagg_protos[pr].pr_attach(sc);
370 	sc->sc_proto = pr;
371 }
372 
373 static void
lagg_proto_detach(struct lagg_softc * sc)374 lagg_proto_detach(struct lagg_softc *sc)
375 {
376 	lagg_proto pr;
377 
378 	LAGG_XLOCK_ASSERT(sc);
379 	pr = sc->sc_proto;
380 	sc->sc_proto = LAGG_PROTO_NONE;
381 
382 	if (lagg_protos[pr].pr_detach != NULL)
383 		lagg_protos[pr].pr_detach(sc);
384 }
385 
386 static inline int
lagg_proto_start(struct lagg_softc * sc,struct mbuf * m)387 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
388 {
389 
390 	return (lagg_protos[sc->sc_proto].pr_start(sc, m));
391 }
392 
393 static inline struct mbuf *
lagg_proto_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)394 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
395 {
396 
397 	return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
398 }
399 
400 static int
lagg_proto_addport(struct lagg_softc * sc,struct lagg_port * lp)401 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
402 {
403 
404 	if (lagg_protos[sc->sc_proto].pr_addport == NULL)
405 		return (0);
406 	else
407 		return (lagg_protos[sc->sc_proto].pr_addport(lp));
408 }
409 
410 static void
lagg_proto_delport(struct lagg_softc * sc,struct lagg_port * lp)411 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
412 {
413 
414 	if (lagg_protos[sc->sc_proto].pr_delport != NULL)
415 		lagg_protos[sc->sc_proto].pr_delport(lp);
416 }
417 
418 static void
lagg_proto_linkstate(struct lagg_softc * sc,struct lagg_port * lp)419 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
420 {
421 
422 	if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
423 		lagg_protos[sc->sc_proto].pr_linkstate(lp);
424 }
425 
426 static void
lagg_proto_init(struct lagg_softc * sc)427 lagg_proto_init(struct lagg_softc *sc)
428 {
429 
430 	if (lagg_protos[sc->sc_proto].pr_init != NULL)
431 		lagg_protos[sc->sc_proto].pr_init(sc);
432 }
433 
434 static void
lagg_proto_stop(struct lagg_softc * sc)435 lagg_proto_stop(struct lagg_softc *sc)
436 {
437 
438 	if (lagg_protos[sc->sc_proto].pr_stop != NULL)
439 		lagg_protos[sc->sc_proto].pr_stop(sc);
440 }
441 
442 static void
lagg_proto_lladdr(struct lagg_softc * sc)443 lagg_proto_lladdr(struct lagg_softc *sc)
444 {
445 
446 	if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
447 		lagg_protos[sc->sc_proto].pr_lladdr(sc);
448 }
449 
450 static void
lagg_proto_request(struct lagg_softc * sc,void * v)451 lagg_proto_request(struct lagg_softc *sc, void *v)
452 {
453 
454 	if (lagg_protos[sc->sc_proto].pr_request != NULL)
455 		lagg_protos[sc->sc_proto].pr_request(sc, v);
456 }
457 
458 static void
lagg_proto_portreq(struct lagg_softc * sc,struct lagg_port * lp,void * v)459 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
460 {
461 
462 	if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
463 		lagg_protos[sc->sc_proto].pr_portreq(lp, v);
464 }
465 
466 /*
467  * This routine is run via an vlan
468  * config EVENT
469  */
470 static void
lagg_register_vlan(void * arg,struct ifnet * ifp,u_int16_t vtag)471 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
472 {
473 	struct lagg_softc *sc = ifp->if_softc;
474 	struct lagg_port *lp;
475 
476 	if (ifp->if_softc != arg) /* Not our event */
477 		return;
478 
479 	LAGG_XLOCK(sc);
480 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
481 		EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
482 	LAGG_XUNLOCK(sc);
483 }
484 
485 /*
486  * This routine is run via an vlan
487  * unconfig EVENT
488  */
489 static void
lagg_unregister_vlan(void * arg,struct ifnet * ifp,u_int16_t vtag)490 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
491 {
492 	struct lagg_softc *sc = ifp->if_softc;
493 	struct lagg_port *lp;
494 
495 	if (ifp->if_softc != arg) /* Not our event */
496 		return;
497 
498 	LAGG_XLOCK(sc);
499 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
500 		EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
501 	LAGG_XUNLOCK(sc);
502 }
503 
504 static int
lagg_clone_create(struct if_clone * ifc,char * name,size_t len,struct ifc_data * ifd,struct ifnet ** ifpp)505 lagg_clone_create(struct if_clone *ifc, char *name, size_t len,
506     struct ifc_data *ifd, struct ifnet **ifpp)
507 {
508 	struct iflaggparam iflp;
509 	struct lagg_softc *sc;
510 	struct ifnet *ifp;
511 	int if_type;
512 	int error;
513 	static const uint8_t eaddr[LAGG_ADDR_LEN];
514 
515 	if (ifd->params != NULL) {
516 		error = ifc_copyin(ifd, &iflp, sizeof(iflp));
517 		if (error)
518 			return (error);
519 
520 		switch (iflp.lagg_type) {
521 		case LAGG_TYPE_ETHERNET:
522 			if_type = IFT_ETHER;
523 			break;
524 		case LAGG_TYPE_INFINIBAND:
525 			if_type = IFT_INFINIBAND;
526 			break;
527 		default:
528 			return (EINVAL);
529 		}
530 	} else {
531 		if_type = IFT_ETHER;
532 	}
533 
534 	sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK | M_ZERO);
535 	ifp = sc->sc_ifp = if_alloc(if_type);
536 	LAGG_SX_INIT(sc);
537 
538 	mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
539 	callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
540 
541 	LAGG_XLOCK(sc);
542 	if (V_def_use_flowid)
543 		sc->sc_opts |= LAGG_OPT_USE_FLOWID;
544 	if (V_def_use_numa)
545 		sc->sc_opts |= LAGG_OPT_USE_NUMA;
546 	sc->flowid_shift = V_def_flowid_shift;
547 
548 	/* Hash all layers by default */
549 	sc->sc_flags = MBUF_HASHFLAG_L2 | MBUF_HASHFLAG_L3 | MBUF_HASHFLAG_L4;
550 
551 	lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
552 
553 	CK_SLIST_INIT(&sc->sc_ports);
554 
555 	switch (if_type) {
556 	case IFT_ETHER:
557 		/* Initialise pseudo media types */
558 		ifmedia_init(&sc->sc_media, 0, lagg_media_change,
559 		    lagg_media_status);
560 		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
561 		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
562 
563 		if_initname(ifp, laggname, ifd->unit);
564 		ifp->if_transmit = lagg_transmit_ethernet;
565 		break;
566 	case IFT_INFINIBAND:
567 		if_initname(ifp, laggname, ifd->unit);
568 		ifp->if_transmit = lagg_transmit_infiniband;
569 		break;
570 	default:
571 		break;
572 	}
573 	ifp->if_softc = sc;
574 	ifp->if_qflush = lagg_qflush;
575 	ifp->if_init = lagg_init;
576 	ifp->if_ioctl = lagg_ioctl;
577 	ifp->if_get_counter = lagg_get_counter;
578 	ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
579 #if defined(KERN_TLS) || defined(RATELIMIT)
580 	ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
581 	ifp->if_ratelimit_query = lagg_ratelimit_query;
582 #endif
583 	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
584 
585 	/*
586 	 * Attach as an ordinary ethernet device, children will be attached
587 	 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
588 	 */
589 	switch (if_type) {
590 	case IFT_ETHER:
591 		ether_ifattach(ifp, eaddr);
592 		break;
593 	case IFT_INFINIBAND:
594 		infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr);
595 		break;
596 	default:
597 		break;
598 	}
599 
600 	sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
601 		lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
602 	sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
603 		lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
604 
605 	/* Insert into the global list of laggs */
606 	LAGG_LIST_LOCK();
607 	SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
608 	LAGG_LIST_UNLOCK();
609 	LAGG_XUNLOCK(sc);
610 	*ifpp = ifp;
611 
612 	return (0);
613 }
614 
615 static int
lagg_clone_destroy(struct if_clone * ifc,struct ifnet * ifp,uint32_t flags)616 lagg_clone_destroy(struct if_clone *ifc, struct ifnet *ifp, uint32_t flags)
617 {
618 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
619 	struct lagg_port *lp;
620 
621 	LAGG_XLOCK(sc);
622 	sc->sc_destroying = 1;
623 	lagg_stop(sc);
624 	ifp->if_flags &= ~IFF_UP;
625 
626 	EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
627 	EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
628 
629 	/* Shutdown and remove lagg ports */
630 	while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
631 		lagg_port_destroy(lp, 1);
632 
633 	/* Unhook the aggregation protocol */
634 	lagg_proto_detach(sc);
635 	LAGG_XUNLOCK(sc);
636 
637 	switch (ifp->if_type) {
638 	case IFT_ETHER:
639 		ether_ifdetach(ifp);
640 		ifmedia_removeall(&sc->sc_media);
641 		break;
642 	case IFT_INFINIBAND:
643 		infiniband_ifdetach(ifp);
644 		break;
645 	default:
646 		break;
647 	}
648 	if_free(ifp);
649 
650 	LAGG_LIST_LOCK();
651 	SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
652 	LAGG_LIST_UNLOCK();
653 
654 	mtx_destroy(&sc->sc_mtx);
655 	LAGG_SX_DESTROY(sc);
656 	free(sc, M_LAGG);
657 
658 	return (0);
659 }
660 
661 static void
lagg_capabilities(struct lagg_softc * sc)662 lagg_capabilities(struct lagg_softc *sc)
663 {
664 	struct lagg_port *lp;
665 	int cap, cap2, ena, ena2, pena, pena2;
666 	uint64_t hwa;
667 	struct ifnet_hw_tsomax hw_tsomax;
668 
669 	LAGG_XLOCK_ASSERT(sc);
670 
671 	/* Get common enabled capabilities for the lagg ports */
672 	ena = ena2 = ~0;
673 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
674 		ena &= lp->lp_ifp->if_capenable;
675 		ena2 &= lp->lp_ifp->if_capenable2;
676 	}
677 	if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
678 		ena = ena2 = 0;
679 
680 	/*
681 	 * Apply common enabled capabilities back to the lagg ports.
682 	 * May require several iterations if they are dependent.
683 	 */
684 	do {
685 		pena = ena;
686 		pena2 = ena2;
687 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
688 			lagg_setcaps(lp, ena, ena2);
689 			ena &= lp->lp_ifp->if_capenable;
690 			ena2 &= lp->lp_ifp->if_capenable2;
691 		}
692 	} while (pena != ena || pena2 != ena2);
693 
694 	/* Get other capabilities from the lagg ports */
695 	cap = cap2 = ~0;
696 	hwa = ~(uint64_t)0;
697 	memset(&hw_tsomax, 0, sizeof(hw_tsomax));
698 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
699 		cap &= lp->lp_ifp->if_capabilities;
700 		cap2 &= lp->lp_ifp->if_capabilities2;
701 		hwa &= lp->lp_ifp->if_hwassist;
702 		if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
703 	}
704 	if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
705 		cap = cap2 = hwa = 0;
706 
707 	if (sc->sc_ifp->if_capabilities != cap ||
708 	    sc->sc_ifp->if_capenable != ena ||
709 	    sc->sc_ifp->if_capenable2 != ena2 ||
710 	    sc->sc_ifp->if_hwassist != hwa ||
711 	    if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
712 		sc->sc_ifp->if_capabilities = cap;
713 		sc->sc_ifp->if_capabilities2 = cap2;
714 		sc->sc_ifp->if_capenable = ena;
715 		sc->sc_ifp->if_capenable2 = ena2;
716 		sc->sc_ifp->if_hwassist = hwa;
717 		getmicrotime(&sc->sc_ifp->if_lastchange);
718 
719 		if (sc->sc_ifflags & IFF_DEBUG)
720 			if_printf(sc->sc_ifp,
721 			    "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
722 	}
723 }
724 
725 static int
lagg_port_create(struct lagg_softc * sc,struct ifnet * ifp)726 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
727 {
728 	struct lagg_softc *sc_ptr;
729 	struct lagg_port *lp, *tlp;
730 	struct ifreq ifr;
731 	int error, i, oldmtu;
732 	int if_type;
733 	uint64_t *pval;
734 
735 	LAGG_XLOCK_ASSERT(sc);
736 
737 	if (sc->sc_ifp == ifp) {
738 		if_printf(sc->sc_ifp,
739 		    "cannot add a lagg to itself as a port\n");
740 		return (EINVAL);
741 	}
742 
743 	if (sc->sc_destroying == 1)
744 		return (ENXIO);
745 
746 	/* Limit the maximal number of lagg ports */
747 	if (sc->sc_count >= LAGG_MAX_PORTS)
748 		return (ENOSPC);
749 
750 	/* Check if port has already been associated to a lagg */
751 	if (ifp->if_lagg != NULL) {
752 		/* Port is already in the current lagg? */
753 		lp = (struct lagg_port *)ifp->if_lagg;
754 		if (lp->lp_softc == sc)
755 			return (EEXIST);
756 		return (EBUSY);
757 	}
758 
759 	switch (sc->sc_ifp->if_type) {
760 	case IFT_ETHER:
761 		/* XXX Disallow non-ethernet interfaces (this should be any of 802) */
762 		if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
763 			return (EPROTONOSUPPORT);
764 		if_type = IFT_IEEE8023ADLAG;
765 		break;
766 	case IFT_INFINIBAND:
767 		/* XXX Disallow non-infiniband interfaces */
768 		if (ifp->if_type != IFT_INFINIBAND)
769 			return (EPROTONOSUPPORT);
770 		if_type = IFT_INFINIBANDLAG;
771 		break;
772 	default:
773 		break;
774 	}
775 
776 	/* Allow the first Ethernet member to define the MTU */
777 	oldmtu = -1;
778 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
779 		sc->sc_ifp->if_mtu = ifp->if_mtu;
780 	} else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
781 		if (ifp->if_ioctl == NULL) {
782 			if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
783 			    ifp->if_xname);
784 			return (EINVAL);
785 		}
786 		oldmtu = ifp->if_mtu;
787 		strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
788 		ifr.ifr_mtu = sc->sc_ifp->if_mtu;
789 		error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
790 		if (error != 0) {
791 			if_printf(sc->sc_ifp, "invalid MTU for %s\n",
792 			    ifp->if_xname);
793 			return (error);
794 		}
795 		ifr.ifr_mtu = oldmtu;
796 	}
797 
798 	lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK | M_ZERO);
799 	lp->lp_softc = sc;
800 
801 	/* Check if port is a stacked lagg */
802 	LAGG_LIST_LOCK();
803 	SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
804 		if (ifp == sc_ptr->sc_ifp) {
805 			LAGG_LIST_UNLOCK();
806 			free(lp, M_LAGG);
807 			if (oldmtu != -1)
808 				(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
809 				    (caddr_t)&ifr);
810 			return (EINVAL);
811 			/* XXX disable stacking for the moment, its untested */
812 #ifdef LAGG_PORT_STACKING
813 			lp->lp_flags |= LAGG_PORT_STACK;
814 			if (lagg_port_checkstacking(sc_ptr) >=
815 			    LAGG_MAX_STACKING) {
816 				LAGG_LIST_UNLOCK();
817 				free(lp, M_LAGG);
818 				if (oldmtu != -1)
819 					(*ifp->if_ioctl)(ifp, SIOCSIFMTU,
820 					    (caddr_t)&ifr);
821 				return (E2BIG);
822 			}
823 #endif
824 		}
825 	}
826 	LAGG_LIST_UNLOCK();
827 
828 	if_ref(ifp);
829 	lp->lp_ifp = ifp;
830 
831 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
832 	lp->lp_ifcapenable = ifp->if_capenable;
833 	if (CK_SLIST_EMPTY(&sc->sc_ports)) {
834 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
835 		lagg_proto_lladdr(sc);
836 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
837 	} else {
838 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
839 	}
840 	lagg_setflags(lp, 1);
841 
842 	if (CK_SLIST_EMPTY(&sc->sc_ports))
843 		sc->sc_primary = lp;
844 
845 	/* Change the interface type */
846 	lp->lp_iftype = ifp->if_type;
847 	ifp->if_type = if_type;
848 	ifp->if_lagg = lp;
849 	lp->lp_ioctl = ifp->if_ioctl;
850 	ifp->if_ioctl = lagg_port_ioctl;
851 	lp->lp_output = ifp->if_output;
852 	ifp->if_output = lagg_port_output;
853 
854 	/* Read port counters */
855 	pval = lp->port_counters.val;
856 	for (i = 0; i < IFCOUNTERS; i++, pval++)
857 		*pval = ifp->if_get_counter(ifp, i);
858 
859 	/*
860 	 * Insert into the list of ports.
861 	 * Keep ports sorted by if_index. It is handy, when configuration
862 	 * is predictable and `ifconfig laggN create ...` command
863 	 * will lead to the same result each time.
864 	 */
865 	CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
866 		if (tlp->lp_ifp->if_index < ifp->if_index && (
867 		    CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
868 		    ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
869 		    ifp->if_index))
870 			break;
871 	}
872 	if (tlp != NULL)
873 		CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
874 	else
875 		CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
876 	sc->sc_count++;
877 
878 	lagg_setmulti(lp);
879 
880 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
881 		/* Remove the port, without calling pr_delport. */
882 		lagg_port_destroy(lp, 0);
883 		if (oldmtu != -1)
884 			(*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
885 		return (error);
886 	}
887 
888 	/* Update lagg capabilities */
889 	lagg_capabilities(sc);
890 	lagg_linkstate(sc);
891 
892 	return (0);
893 }
894 
895 #ifdef LAGG_PORT_STACKING
896 static int
lagg_port_checkstacking(struct lagg_softc * sc)897 lagg_port_checkstacking(struct lagg_softc *sc)
898 {
899 	struct lagg_softc *sc_ptr;
900 	struct lagg_port *lp;
901 	int m = 0;
902 
903 	LAGG_SXLOCK_ASSERT(sc);
904 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
905 		if (lp->lp_flags & LAGG_PORT_STACK) {
906 			sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
907 			m = MAX(m, lagg_port_checkstacking(sc_ptr));
908 		}
909 	}
910 
911 	return (m + 1);
912 }
913 #endif
914 
915 static void
lagg_port_destroy_cb(epoch_context_t ec)916 lagg_port_destroy_cb(epoch_context_t ec)
917 {
918 	struct lagg_port *lp;
919 	struct ifnet *ifp;
920 
921 	lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
922 	ifp = lp->lp_ifp;
923 
924 	if_rele(ifp);
925 	free(lp, M_LAGG);
926 }
927 
928 static int
lagg_port_destroy(struct lagg_port * lp,int rundelport)929 lagg_port_destroy(struct lagg_port *lp, int rundelport)
930 {
931 	struct lagg_softc *sc = lp->lp_softc;
932 	struct lagg_port *lp_ptr, *lp0;
933 	struct ifnet *ifp = lp->lp_ifp;
934 	uint64_t *pval, vdiff;
935 	int i;
936 
937 	LAGG_XLOCK_ASSERT(sc);
938 
939 	if (rundelport)
940 		lagg_proto_delport(sc, lp);
941 
942 	if (lp->lp_detaching == 0)
943 		lagg_clrmulti(lp);
944 
945 	/* Restore interface */
946 	ifp->if_type = lp->lp_iftype;
947 	ifp->if_ioctl = lp->lp_ioctl;
948 	ifp->if_output = lp->lp_output;
949 	ifp->if_lagg = NULL;
950 
951 	/* Update detached port counters */
952 	pval = lp->port_counters.val;
953 	for (i = 0; i < IFCOUNTERS; i++, pval++) {
954 		vdiff = ifp->if_get_counter(ifp, i) - *pval;
955 		sc->detached_counters.val[i] += vdiff;
956 	}
957 
958 	/* Finally, remove the port from the lagg */
959 	CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
960 	sc->sc_count--;
961 
962 	/* Update the primary interface */
963 	if (lp == sc->sc_primary) {
964 		uint8_t lladdr[LAGG_ADDR_LEN];
965 
966 		if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
967 			bzero(&lladdr, LAGG_ADDR_LEN);
968 		else
969 			bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
970 		sc->sc_primary = lp0;
971 		if (sc->sc_destroying == 0) {
972 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
973 			lagg_proto_lladdr(sc);
974 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
975 
976 			/*
977 			 * Update lladdr for each port (new primary needs update
978 			 * as well, to switch from old lladdr to its 'real' one).
979 			 * We can skip this if the lagg is being destroyed.
980 			 */
981 			CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
982 				if_setlladdr(lp_ptr->lp_ifp, lladdr,
983 				    lp_ptr->lp_ifp->if_addrlen);
984 		}
985 	}
986 
987 	if (lp->lp_ifflags)
988 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
989 
990 	if (lp->lp_detaching == 0) {
991 		lagg_setflags(lp, 0);
992 		lagg_setcaps(lp, lp->lp_ifcapenable, lp->lp_ifcapenable2);
993 		if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
994 	}
995 
996 	/*
997 	 * free port and release it's ifnet reference after a grace period has
998 	 * elapsed.
999 	 */
1000 	NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
1001 	/* Update lagg capabilities */
1002 	lagg_capabilities(sc);
1003 	lagg_linkstate(sc);
1004 
1005 	return (0);
1006 }
1007 
1008 static int
lagg_port_ioctl(struct ifnet * ifp,u_long cmd,caddr_t data)1009 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1010 {
1011 	struct epoch_tracker et;
1012 	struct lagg_reqport *rp = (struct lagg_reqport *)data;
1013 	struct lagg_softc *sc;
1014 	struct lagg_port *lp = NULL;
1015 	int error = 0;
1016 
1017 	/* Should be checked by the caller */
1018 	switch (ifp->if_type) {
1019 	case IFT_IEEE8023ADLAG:
1020 	case IFT_INFINIBANDLAG:
1021 		if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1022 			goto fallback;
1023 		break;
1024 	default:
1025 		goto fallback;
1026 	}
1027 
1028 	switch (cmd) {
1029 	case SIOCGLAGGPORT:
1030 		if (rp->rp_portname[0] == '\0' ||
1031 		    ifunit(rp->rp_portname) != ifp) {
1032 			error = EINVAL;
1033 			break;
1034 		}
1035 
1036 		NET_EPOCH_ENTER(et);
1037 		if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1038 			error = ENOENT;
1039 			NET_EPOCH_EXIT(et);
1040 			break;
1041 		}
1042 
1043 		lagg_port2req(lp, rp);
1044 		NET_EPOCH_EXIT(et);
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
lagg_get_counter(struct ifnet * ifp,ift_counter cnt)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
lagg_port_output(struct ifnet * ifp,struct mbuf * m,const struct sockaddr * dst,struct route * ro)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
lagg_port_ifdetach(void * arg __unused,struct ifnet * ifp)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
lagg_port2req(struct lagg_port * lp,struct lagg_reqport * rp)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
lagg_watchdog_infiniband(void * arg)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
lagg_init(void * xsc)1261 lagg_init(void *xsc)
1262 {
1263 	struct lagg_softc *sc = (struct lagg_softc *)xsc;
1264 	struct ifnet *ifp = sc->sc_ifp;
1265 	struct lagg_port *lp;
1266 
1267 	LAGG_XLOCK(sc);
1268 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1269 		LAGG_XUNLOCK(sc);
1270 		return;
1271 	}
1272 
1273 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1274 
1275 	/*
1276 	 * Update the port lladdrs if needed.
1277 	 * This might be if_setlladdr() notification
1278 	 * that lladdr has been changed.
1279 	 */
1280 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1281 		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1282 		    ifp->if_addrlen) != 0)
1283 			if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1284 	}
1285 
1286 	lagg_proto_init(sc);
1287 
1288 	if (ifp->if_type == IFT_INFINIBAND) {
1289 		mtx_lock(&sc->sc_mtx);
1290 		lagg_watchdog_infiniband(sc);
1291 		mtx_unlock(&sc->sc_mtx);
1292 	}
1293 
1294 	LAGG_XUNLOCK(sc);
1295 }
1296 
1297 static void
lagg_stop(struct lagg_softc * sc)1298 lagg_stop(struct lagg_softc *sc)
1299 {
1300 	struct ifnet *ifp = sc->sc_ifp;
1301 
1302 	LAGG_XLOCK_ASSERT(sc);
1303 
1304 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1305 		return;
1306 
1307 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1308 
1309 	lagg_proto_stop(sc);
1310 
1311 	mtx_lock(&sc->sc_mtx);
1312 	callout_stop(&sc->sc_watchdog);
1313 	mtx_unlock(&sc->sc_mtx);
1314 
1315 	callout_drain(&sc->sc_watchdog);
1316 }
1317 
1318 static int
lagg_ioctl(struct ifnet * ifp,u_long cmd,caddr_t data)1319 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1320 {
1321 	struct epoch_tracker et;
1322 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1323 	struct lagg_reqall *ra = (struct lagg_reqall *)data;
1324 	struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1325 	struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1326 	struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1327 	struct ifreq *ifr = (struct ifreq *)data;
1328 	struct lagg_port *lp;
1329 	struct ifnet *tpif;
1330 	struct thread *td = curthread;
1331 	char *buf, *outbuf;
1332 	int count, buflen, len, error = 0, oldmtu;
1333 
1334 	bzero(&rpbuf, sizeof(rpbuf));
1335 
1336 	/* XXX: This can race with lagg_clone_destroy. */
1337 
1338 	switch (cmd) {
1339 	case SIOCGLAGG:
1340 		LAGG_XLOCK(sc);
1341 		buflen = sc->sc_count * sizeof(struct lagg_reqport);
1342 		outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1343 		ra->ra_proto = sc->sc_proto;
1344 		lagg_proto_request(sc, &ra->ra_psc);
1345 		count = 0;
1346 		buf = outbuf;
1347 		len = min(ra->ra_size, buflen);
1348 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1349 			if (len < sizeof(rpbuf))
1350 				break;
1351 
1352 			lagg_port2req(lp, &rpbuf);
1353 			memcpy(buf, &rpbuf, sizeof(rpbuf));
1354 			count++;
1355 			buf += sizeof(rpbuf);
1356 			len -= sizeof(rpbuf);
1357 		}
1358 		LAGG_XUNLOCK(sc);
1359 		ra->ra_ports = count;
1360 		ra->ra_size = count * sizeof(rpbuf);
1361 		error = copyout(outbuf, ra->ra_port, ra->ra_size);
1362 		free(outbuf, M_TEMP);
1363 		break;
1364 	case SIOCSLAGG:
1365 		error = priv_check(td, PRIV_NET_LAGG);
1366 		if (error)
1367 			break;
1368 		if (ra->ra_proto >= LAGG_PROTO_MAX) {
1369 			error = EPROTONOSUPPORT;
1370 			break;
1371 		}
1372 		/* Infiniband only supports the failover protocol. */
1373 		if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1374 		    ifp->if_type == IFT_INFINIBAND) {
1375 			error = EPROTONOSUPPORT;
1376 			break;
1377 		}
1378 		LAGG_XLOCK(sc);
1379 		lagg_proto_detach(sc);
1380 		lagg_proto_attach(sc, ra->ra_proto);
1381 		LAGG_XUNLOCK(sc);
1382 		break;
1383 	case SIOCGLAGGOPTS:
1384 		LAGG_XLOCK(sc);
1385 		ro->ro_opts = sc->sc_opts;
1386 		if (sc->sc_proto == LAGG_PROTO_LACP) {
1387 			struct lacp_softc *lsc;
1388 
1389 			lsc = (struct lacp_softc *)sc->sc_psc;
1390 			if (lsc->lsc_debug.lsc_tx_test != 0)
1391 				ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1392 			if (lsc->lsc_debug.lsc_rx_test != 0)
1393 				ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1394 			if (lsc->lsc_strict_mode != 0)
1395 				ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1396 			if (lsc->lsc_fast_timeout != 0)
1397 				ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1398 
1399 			ro->ro_active = sc->sc_active;
1400 		} else {
1401 			ro->ro_active = 0;
1402 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1403 				ro->ro_active += LAGG_PORTACTIVE(lp);
1404 		}
1405 		ro->ro_bkt = sc->sc_stride;
1406 		ro->ro_flapping = sc->sc_flapping;
1407 		ro->ro_flowid_shift = sc->flowid_shift;
1408 		LAGG_XUNLOCK(sc);
1409 		break;
1410 	case SIOCSLAGGOPTS:
1411 		error = priv_check(td, PRIV_NET_LAGG);
1412 		if (error)
1413 			break;
1414 
1415 		/*
1416 		 * The stride option was added without defining a corresponding
1417 		 * LAGG_OPT flag, so handle a non-zero value before checking
1418 		 * anything else to preserve compatibility.
1419 		 */
1420 		LAGG_XLOCK(sc);
1421 		if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1422 			if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1423 				LAGG_XUNLOCK(sc);
1424 				error = EINVAL;
1425 				break;
1426 			}
1427 			sc->sc_stride = ro->ro_bkt;
1428 		}
1429 		if (ro->ro_opts == 0) {
1430 			LAGG_XUNLOCK(sc);
1431 			break;
1432 		}
1433 
1434 		/*
1435 		 * Set options.  LACP options are stored in sc->sc_psc,
1436 		 * not in sc_opts.
1437 		 */
1438 		int valid, lacp;
1439 
1440 		switch (ro->ro_opts) {
1441 		case LAGG_OPT_USE_FLOWID:
1442 		case -LAGG_OPT_USE_FLOWID:
1443 		case LAGG_OPT_USE_NUMA:
1444 		case -LAGG_OPT_USE_NUMA:
1445 		case LAGG_OPT_FLOWIDSHIFT:
1446 		case LAGG_OPT_RR_LIMIT:
1447 			valid = 1;
1448 			lacp = 0;
1449 			break;
1450 		case LAGG_OPT_LACP_TXTEST:
1451 		case -LAGG_OPT_LACP_TXTEST:
1452 		case LAGG_OPT_LACP_RXTEST:
1453 		case -LAGG_OPT_LACP_RXTEST:
1454 		case LAGG_OPT_LACP_STRICT:
1455 		case -LAGG_OPT_LACP_STRICT:
1456 		case LAGG_OPT_LACP_FAST_TIMO:
1457 		case -LAGG_OPT_LACP_FAST_TIMO:
1458 			valid = lacp = 1;
1459 			break;
1460 		default:
1461 			valid = lacp = 0;
1462 			break;
1463 		}
1464 
1465 		if (valid == 0 ||
1466 		    (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1467 			/* Invalid combination of options specified. */
1468 			error = EINVAL;
1469 			LAGG_XUNLOCK(sc);
1470 			break;	/* Return from SIOCSLAGGOPTS. */
1471 		}
1472 
1473 		/*
1474 		 * Store new options into sc->sc_opts except for
1475 		 * FLOWIDSHIFT, RR and LACP options.
1476 		 */
1477 		if (lacp == 0) {
1478 			if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1479 				sc->flowid_shift = ro->ro_flowid_shift;
1480 			else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1481 				if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1482 				    ro->ro_bkt == 0) {
1483 					error = EINVAL;
1484 					LAGG_XUNLOCK(sc);
1485 					break;
1486 				}
1487 				sc->sc_stride = ro->ro_bkt;
1488 			} else if (ro->ro_opts > 0)
1489 				sc->sc_opts |= ro->ro_opts;
1490 			else
1491 				sc->sc_opts &= ~ro->ro_opts;
1492 		} else {
1493 			struct lacp_softc *lsc;
1494 			struct lacp_port *lp;
1495 
1496 			lsc = (struct lacp_softc *)sc->sc_psc;
1497 
1498 			switch (ro->ro_opts) {
1499 			case LAGG_OPT_LACP_TXTEST:
1500 				lsc->lsc_debug.lsc_tx_test = 1;
1501 				break;
1502 			case -LAGG_OPT_LACP_TXTEST:
1503 				lsc->lsc_debug.lsc_tx_test = 0;
1504 				break;
1505 			case LAGG_OPT_LACP_RXTEST:
1506 				lsc->lsc_debug.lsc_rx_test = 1;
1507 				break;
1508 			case -LAGG_OPT_LACP_RXTEST:
1509 				lsc->lsc_debug.lsc_rx_test = 0;
1510 				break;
1511 			case LAGG_OPT_LACP_STRICT:
1512 				lsc->lsc_strict_mode = 1;
1513 				break;
1514 			case -LAGG_OPT_LACP_STRICT:
1515 				lsc->lsc_strict_mode = 0;
1516 				break;
1517 			case LAGG_OPT_LACP_FAST_TIMO:
1518 				LACP_LOCK(lsc);
1519 				LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1520 					lp->lp_state |= LACP_STATE_TIMEOUT;
1521 				LACP_UNLOCK(lsc);
1522 				lsc->lsc_fast_timeout = 1;
1523 				break;
1524 			case -LAGG_OPT_LACP_FAST_TIMO:
1525 				LACP_LOCK(lsc);
1526 				LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1527 					lp->lp_state &= ~LACP_STATE_TIMEOUT;
1528 				LACP_UNLOCK(lsc);
1529 				lsc->lsc_fast_timeout = 0;
1530 				break;
1531 			}
1532 		}
1533 		LAGG_XUNLOCK(sc);
1534 		break;
1535 	case SIOCGLAGGFLAGS:
1536 		rf->rf_flags = 0;
1537 		LAGG_XLOCK(sc);
1538 		if (sc->sc_flags & MBUF_HASHFLAG_L2)
1539 			rf->rf_flags |= LAGG_F_HASHL2;
1540 		if (sc->sc_flags & MBUF_HASHFLAG_L3)
1541 			rf->rf_flags |= LAGG_F_HASHL3;
1542 		if (sc->sc_flags & MBUF_HASHFLAG_L4)
1543 			rf->rf_flags |= LAGG_F_HASHL4;
1544 		LAGG_XUNLOCK(sc);
1545 		break;
1546 	case SIOCSLAGGHASH:
1547 		error = priv_check(td, PRIV_NET_LAGG);
1548 		if (error)
1549 			break;
1550 		if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1551 			error = EINVAL;
1552 			break;
1553 		}
1554 		LAGG_XLOCK(sc);
1555 		sc->sc_flags = 0;
1556 		if (rf->rf_flags & LAGG_F_HASHL2)
1557 			sc->sc_flags |= MBUF_HASHFLAG_L2;
1558 		if (rf->rf_flags & LAGG_F_HASHL3)
1559 			sc->sc_flags |= MBUF_HASHFLAG_L3;
1560 		if (rf->rf_flags & LAGG_F_HASHL4)
1561 			sc->sc_flags |= MBUF_HASHFLAG_L4;
1562 		LAGG_XUNLOCK(sc);
1563 		break;
1564 	case SIOCGLAGGPORT:
1565 		if (rp->rp_portname[0] == '\0' ||
1566 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1567 			error = EINVAL;
1568 			break;
1569 		}
1570 
1571 		NET_EPOCH_ENTER(et);
1572 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1573 		    lp->lp_softc != sc) {
1574 			error = ENOENT;
1575 			NET_EPOCH_EXIT(et);
1576 			if_rele(tpif);
1577 			break;
1578 		}
1579 
1580 		lagg_port2req(lp, rp);
1581 		NET_EPOCH_EXIT(et);
1582 		if_rele(tpif);
1583 		break;
1584 	case SIOCSLAGGPORT:
1585 		error = priv_check(td, PRIV_NET_LAGG);
1586 		if (error)
1587 			break;
1588 		if (rp->rp_portname[0] == '\0' ||
1589 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1590 			error = EINVAL;
1591 			break;
1592 		}
1593 #ifdef INET6
1594 		/*
1595 		 * A laggport interface should not have inet6 address
1596 		 * because two interfaces with a valid link-local
1597 		 * scope zone must not be merged in any form.  This
1598 		 * restriction is needed to prevent violation of
1599 		 * link-local scope zone.  Attempts to add a laggport
1600 		 * interface which has inet6 addresses triggers
1601 		 * removal of all inet6 addresses on the member
1602 		 * interface.
1603 		 */
1604 		if (in6ifa_llaonifp(tpif)) {
1605 			in6_ifdetach(tpif);
1606 				if_printf(sc->sc_ifp,
1607 				    "IPv6 addresses on %s have been removed "
1608 				    "before adding it as a member to prevent "
1609 				    "IPv6 address scope violation.\n",
1610 				    tpif->if_xname);
1611 		}
1612 #endif
1613 		oldmtu = ifp->if_mtu;
1614 		LAGG_XLOCK(sc);
1615 		error = lagg_port_create(sc, tpif);
1616 		LAGG_XUNLOCK(sc);
1617 		if_rele(tpif);
1618 
1619 		/*
1620 		 * LAGG MTU may change during addition of the first port.
1621 		 * If it did, do network layer specific procedure.
1622 		 */
1623 		if (ifp->if_mtu != oldmtu)
1624 			if_notifymtu(ifp);
1625 
1626 		VLAN_CAPABILITIES(ifp);
1627 		break;
1628 	case SIOCSLAGGDELPORT:
1629 		error = priv_check(td, PRIV_NET_LAGG);
1630 		if (error)
1631 			break;
1632 		if (rp->rp_portname[0] == '\0' ||
1633 		    (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1634 			error = EINVAL;
1635 			break;
1636 		}
1637 
1638 		LAGG_XLOCK(sc);
1639 		if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1640 		    lp->lp_softc != sc) {
1641 			error = ENOENT;
1642 			LAGG_XUNLOCK(sc);
1643 			if_rele(tpif);
1644 			break;
1645 		}
1646 
1647 		error = lagg_port_destroy(lp, 1);
1648 		LAGG_XUNLOCK(sc);
1649 		if_rele(tpif);
1650 		VLAN_CAPABILITIES(ifp);
1651 		break;
1652 	case SIOCSIFFLAGS:
1653 		/* Set flags on ports too */
1654 		LAGG_XLOCK(sc);
1655 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1656 			lagg_setflags(lp, 1);
1657 		}
1658 
1659 		if (!(ifp->if_flags & IFF_UP) &&
1660 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1661 			/*
1662 			 * If interface is marked down and it is running,
1663 			 * then stop and disable it.
1664 			 */
1665 			lagg_stop(sc);
1666 			LAGG_XUNLOCK(sc);
1667 		} else if ((ifp->if_flags & IFF_UP) &&
1668 		    !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1669 			/*
1670 			 * If interface is marked up and it is stopped, then
1671 			 * start it.
1672 			 */
1673 			LAGG_XUNLOCK(sc);
1674 			(*ifp->if_init)(sc);
1675 		} else
1676 			LAGG_XUNLOCK(sc);
1677 		break;
1678 	case SIOCADDMULTI:
1679 	case SIOCDELMULTI:
1680 		LAGG_XLOCK(sc);
1681 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1682 			lagg_clrmulti(lp);
1683 			lagg_setmulti(lp);
1684 		}
1685 		LAGG_XUNLOCK(sc);
1686 		error = 0;
1687 		break;
1688 	case SIOCSIFMEDIA:
1689 	case SIOCGIFMEDIA:
1690 		if (ifp->if_type == IFT_INFINIBAND)
1691 			error = EINVAL;
1692 		else
1693 			error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1694 		break;
1695 
1696 	case SIOCSIFCAP:
1697 	case SIOCSIFCAPNV:
1698 		LAGG_XLOCK(sc);
1699 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1700 			if (lp->lp_ioctl != NULL)
1701 				(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1702 		}
1703 		lagg_capabilities(sc);
1704 		LAGG_XUNLOCK(sc);
1705 		VLAN_CAPABILITIES(ifp);
1706 		error = 0;
1707 		break;
1708 
1709 	case SIOCGIFCAPNV:
1710 		error = 0;
1711 		break;
1712 
1713 	case SIOCSIFMTU:
1714 		LAGG_XLOCK(sc);
1715 		CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1716 			if (lp->lp_ioctl != NULL)
1717 				error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1718 			else
1719 				error = EINVAL;
1720 			if (error != 0) {
1721 				if_printf(ifp,
1722 				    "failed to change MTU to %d on port %s, "
1723 				    "reverting all ports to original MTU (%d)\n",
1724 				    ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1725 				break;
1726 			}
1727 		}
1728 		if (error == 0) {
1729 			ifp->if_mtu = ifr->ifr_mtu;
1730 		} else {
1731 			/* set every port back to the original MTU */
1732 			ifr->ifr_mtu = ifp->if_mtu;
1733 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1734 				if (lp->lp_ioctl != NULL)
1735 					(*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1736 			}
1737 		}
1738 		lagg_capabilities(sc);
1739 		LAGG_XUNLOCK(sc);
1740 		VLAN_CAPABILITIES(ifp);
1741 		break;
1742 
1743 	default:
1744 		error = ether_ioctl(ifp, cmd, data);
1745 		break;
1746 	}
1747 	return (error);
1748 }
1749 
1750 #if defined(KERN_TLS) || defined(RATELIMIT)
1751 #ifdef RATELIMIT
1752 static const struct if_snd_tag_sw lagg_snd_tag_ul_sw = {
1753 	.snd_tag_modify = lagg_snd_tag_modify,
1754 	.snd_tag_query = lagg_snd_tag_query,
1755 	.snd_tag_free = lagg_snd_tag_free,
1756 	.next_snd_tag = lagg_next_snd_tag,
1757 	.type = IF_SND_TAG_TYPE_UNLIMITED
1758 };
1759 
1760 static const struct if_snd_tag_sw lagg_snd_tag_rl_sw = {
1761 	.snd_tag_modify = lagg_snd_tag_modify,
1762 	.snd_tag_query = lagg_snd_tag_query,
1763 	.snd_tag_free = lagg_snd_tag_free,
1764 	.next_snd_tag = lagg_next_snd_tag,
1765 	.type = IF_SND_TAG_TYPE_RATE_LIMIT
1766 };
1767 #endif
1768 
1769 #ifdef KERN_TLS
1770 static const struct if_snd_tag_sw lagg_snd_tag_tls_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_TLS
1776 };
1777 
1778 #ifdef RATELIMIT
1779 static const struct if_snd_tag_sw lagg_snd_tag_tls_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_TLS_RATE_LIMIT
1785 };
1786 #endif
1787 #endif
1788 
1789 static inline struct lagg_snd_tag *
mst_to_lst(struct m_snd_tag * mst)1790 mst_to_lst(struct m_snd_tag *mst)
1791 {
1792 
1793 	return (__containerof(mst, struct lagg_snd_tag, com));
1794 }
1795 
1796 /*
1797  * Look up the port used by a specific flow.  This only works for lagg
1798  * protocols with deterministic port mappings (e.g. not roundrobin).
1799  * In addition protocols which use a hash to map flows to ports must
1800  * be configured to use the mbuf flowid rather than hashing packet
1801  * contents.
1802  */
1803 static struct lagg_port *
lookup_snd_tag_port(struct ifnet * ifp,uint32_t flowid,uint32_t flowtype,uint8_t numa_domain)1804 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1805     uint8_t numa_domain)
1806 {
1807 	struct lagg_softc *sc;
1808 	struct lagg_port *lp;
1809 	struct lagg_lb *lb;
1810 	uint32_t hash, p;
1811 	int err;
1812 
1813 	sc = ifp->if_softc;
1814 
1815 	switch (sc->sc_proto) {
1816 	case LAGG_PROTO_FAILOVER:
1817 		return (lagg_link_active(sc, sc->sc_primary));
1818 	case LAGG_PROTO_LOADBALANCE:
1819 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1820 		    flowtype == M_HASHTYPE_NONE)
1821 			return (NULL);
1822 		p = flowid >> sc->flowid_shift;
1823 		p %= sc->sc_count;
1824 		lb = (struct lagg_lb *)sc->sc_psc;
1825 		lp = lb->lb_ports[p];
1826 		return (lagg_link_active(sc, lp));
1827 	case LAGG_PROTO_LACP:
1828 		if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1829 		    flowtype == M_HASHTYPE_NONE)
1830 			return (NULL);
1831 		hash = flowid >> sc->flowid_shift;
1832 		return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, &err));
1833 	default:
1834 		return (NULL);
1835 	}
1836 }
1837 
1838 static int
lagg_snd_tag_alloc(struct ifnet * ifp,union if_snd_tag_alloc_params * params,struct m_snd_tag ** ppmt)1839 lagg_snd_tag_alloc(struct ifnet *ifp,
1840     union if_snd_tag_alloc_params *params,
1841     struct m_snd_tag **ppmt)
1842 {
1843 	struct epoch_tracker et;
1844 	const struct if_snd_tag_sw *sw;
1845 	struct lagg_snd_tag *lst;
1846 	struct lagg_port *lp;
1847 	struct ifnet *lp_ifp;
1848 	struct m_snd_tag *mst;
1849 	int error;
1850 
1851 	switch (params->hdr.type) {
1852 #ifdef RATELIMIT
1853 	case IF_SND_TAG_TYPE_UNLIMITED:
1854 		sw = &lagg_snd_tag_ul_sw;
1855 		break;
1856 	case IF_SND_TAG_TYPE_RATE_LIMIT:
1857 		sw = &lagg_snd_tag_rl_sw;
1858 		break;
1859 #endif
1860 #ifdef KERN_TLS
1861 	case IF_SND_TAG_TYPE_TLS:
1862 		sw = &lagg_snd_tag_tls_sw;
1863 		break;
1864 	case IF_SND_TAG_TYPE_TLS_RX:
1865 		/* Return tag from port interface directly. */
1866 		sw = NULL;
1867 		break;
1868 #ifdef RATELIMIT
1869 	case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
1870 		sw = &lagg_snd_tag_tls_rl_sw;
1871 		break;
1872 #endif
1873 #endif
1874 	default:
1875 		return (EOPNOTSUPP);
1876 	}
1877 
1878 	NET_EPOCH_ENTER(et);
1879 	lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1880 	    params->hdr.flowtype, params->hdr.numa_domain);
1881 	if (lp == NULL) {
1882 		NET_EPOCH_EXIT(et);
1883 		return (EOPNOTSUPP);
1884 	}
1885 	if (lp->lp_ifp == NULL) {
1886 		NET_EPOCH_EXIT(et);
1887 		return (EOPNOTSUPP);
1888 	}
1889 	lp_ifp = lp->lp_ifp;
1890 	if_ref(lp_ifp);
1891 	NET_EPOCH_EXIT(et);
1892 
1893 	if (sw != NULL) {
1894 		lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1895 		if (lst == NULL) {
1896 			if_rele(lp_ifp);
1897 			return (ENOMEM);
1898 		}
1899 	} else
1900 		lst = NULL;
1901 
1902 	error = m_snd_tag_alloc(lp_ifp, params, &mst);
1903 	if_rele(lp_ifp);
1904 	if (error) {
1905 		free(lst, M_LAGG);
1906 		return (error);
1907 	}
1908 
1909 	if (sw != NULL) {
1910 		m_snd_tag_init(&lst->com, ifp, sw);
1911 		lst->tag = mst;
1912 
1913 		*ppmt = &lst->com;
1914 	} else
1915 		*ppmt = mst;
1916 
1917 	return (0);
1918 }
1919 
1920 static struct m_snd_tag *
lagg_next_snd_tag(struct m_snd_tag * mst)1921 lagg_next_snd_tag(struct m_snd_tag *mst)
1922 {
1923 	struct lagg_snd_tag *lst;
1924 
1925 	lst = mst_to_lst(mst);
1926 	return (lst->tag);
1927 }
1928 
1929 static int
lagg_snd_tag_modify(struct m_snd_tag * mst,union if_snd_tag_modify_params * params)1930 lagg_snd_tag_modify(struct m_snd_tag *mst,
1931     union if_snd_tag_modify_params *params)
1932 {
1933 	struct lagg_snd_tag *lst;
1934 
1935 	lst = mst_to_lst(mst);
1936 	return (lst->tag->sw->snd_tag_modify(lst->tag, params));
1937 }
1938 
1939 static int
lagg_snd_tag_query(struct m_snd_tag * mst,union if_snd_tag_query_params * params)1940 lagg_snd_tag_query(struct m_snd_tag *mst,
1941     union if_snd_tag_query_params *params)
1942 {
1943 	struct lagg_snd_tag *lst;
1944 
1945 	lst = mst_to_lst(mst);
1946 	return (lst->tag->sw->snd_tag_query(lst->tag, params));
1947 }
1948 
1949 static void
lagg_snd_tag_free(struct m_snd_tag * mst)1950 lagg_snd_tag_free(struct m_snd_tag *mst)
1951 {
1952 	struct lagg_snd_tag *lst;
1953 
1954 	lst = mst_to_lst(mst);
1955 	m_snd_tag_rele(lst->tag);
1956 	free(lst, M_LAGG);
1957 }
1958 
1959 static void
lagg_ratelimit_query(struct ifnet * ifp __unused,struct if_ratelimit_query_results * q)1960 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1961 {
1962 	/*
1963 	 * For lagg, we have an indirect
1964 	 * interface. The caller needs to
1965 	 * get a ratelimit tag on the actual
1966 	 * interface the flow will go on.
1967 	 */
1968 	q->rate_table = NULL;
1969 	q->flags = RT_IS_INDIRECT;
1970 	q->max_flows = 0;
1971 	q->number_of_rates = 0;
1972 }
1973 #endif
1974 
1975 static int
lagg_setmulti(struct lagg_port * lp)1976 lagg_setmulti(struct lagg_port *lp)
1977 {
1978 	struct lagg_softc *sc = lp->lp_softc;
1979 	struct ifnet *ifp = lp->lp_ifp;
1980 	struct ifnet *scifp = sc->sc_ifp;
1981 	struct lagg_mc *mc;
1982 	struct ifmultiaddr *ifma;
1983 	int error;
1984 
1985 	IF_ADDR_WLOCK(scifp);
1986 	CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1987 		if (ifma->ifma_addr->sa_family != AF_LINK)
1988 			continue;
1989 		mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
1990 		if (mc == NULL) {
1991 			IF_ADDR_WUNLOCK(scifp);
1992 			return (ENOMEM);
1993 		}
1994 		bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
1995 		mc->mc_addr.sdl_index = ifp->if_index;
1996 		mc->mc_ifma = NULL;
1997 		SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1998 	}
1999 	IF_ADDR_WUNLOCK(scifp);
2000 	SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
2001 		error = if_addmulti(ifp,
2002 		    (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
2003 		if (error)
2004 			return (error);
2005 	}
2006 	return (0);
2007 }
2008 
2009 static int
lagg_clrmulti(struct lagg_port * lp)2010 lagg_clrmulti(struct lagg_port *lp)
2011 {
2012 	struct lagg_mc *mc;
2013 
2014 	LAGG_XLOCK_ASSERT(lp->lp_softc);
2015 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
2016 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
2017 		if (mc->mc_ifma && lp->lp_detaching == 0)
2018 			if_delmulti_ifma(mc->mc_ifma);
2019 		free(mc, M_LAGG);
2020 	}
2021 	return (0);
2022 }
2023 
2024 static void
lagg_setcaps(struct lagg_port * lp,int cap,int cap2)2025 lagg_setcaps(struct lagg_port *lp, int cap, int cap2)
2026 {
2027 	struct ifreq ifr;
2028 	struct siocsifcapnv_driver_data drv_ioctl_data;
2029 
2030 	if (lp->lp_ifp->if_capenable == cap &&
2031 	    lp->lp_ifp->if_capenable2 == cap2)
2032 		return;
2033 	if (lp->lp_ioctl == NULL)
2034 		return;
2035 	/* XXX */
2036 	if ((lp->lp_ifp->if_capabilities & IFCAP_NV) != 0) {
2037 		drv_ioctl_data.reqcap = cap;
2038 		drv_ioctl_data.reqcap2 = cap2;
2039 		drv_ioctl_data.nvcap = NULL;
2040 		(*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAPNV,
2041 		    (caddr_t)&drv_ioctl_data);
2042 	} else {
2043 		ifr.ifr_reqcap = cap;
2044 		(*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr);
2045 	}
2046 }
2047 
2048 /* Handle a ref counted flag that should be set on the lagg port as well */
2049 static int
lagg_setflag(struct lagg_port * lp,int flag,int status,int (* func)(struct ifnet *,int))2050 lagg_setflag(struct lagg_port *lp, int flag, int status,
2051     int (*func)(struct ifnet *, int))
2052 {
2053 	struct lagg_softc *sc = lp->lp_softc;
2054 	struct ifnet *scifp = sc->sc_ifp;
2055 	struct ifnet *ifp = lp->lp_ifp;
2056 	int error;
2057 
2058 	LAGG_XLOCK_ASSERT(sc);
2059 
2060 	status = status ? (scifp->if_flags & flag) : 0;
2061 	/* Now "status" contains the flag value or 0 */
2062 
2063 	/*
2064 	 * See if recorded ports status is different from what
2065 	 * we want it to be.  If it is, flip it.  We record ports
2066 	 * status in lp_ifflags so that we won't clear ports flag
2067 	 * we haven't set.  In fact, we don't clear or set ports
2068 	 * flags directly, but get or release references to them.
2069 	 * That's why we can be sure that recorded flags still are
2070 	 * in accord with actual ports flags.
2071 	 */
2072 	if (status != (lp->lp_ifflags & flag)) {
2073 		error = (*func)(ifp, status);
2074 		if (error)
2075 			return (error);
2076 		lp->lp_ifflags &= ~flag;
2077 		lp->lp_ifflags |= status;
2078 	}
2079 	return (0);
2080 }
2081 
2082 /*
2083  * Handle IFF_* flags that require certain changes on the lagg port
2084  * if "status" is true, update ports flags respective to the lagg
2085  * if "status" is false, forcedly clear the flags set on port.
2086  */
2087 static int
lagg_setflags(struct lagg_port * lp,int status)2088 lagg_setflags(struct lagg_port *lp, int status)
2089 {
2090 	int error, i;
2091 
2092 	for (i = 0; lagg_pflags[i].flag; i++) {
2093 		error = lagg_setflag(lp, lagg_pflags[i].flag,
2094 		    status, lagg_pflags[i].func);
2095 		if (error)
2096 			return (error);
2097 	}
2098 	return (0);
2099 }
2100 
2101 static int
lagg_transmit_ethernet(struct ifnet * ifp,struct mbuf * m)2102 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2103 {
2104 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2105 
2106 	NET_EPOCH_ASSERT();
2107 #if defined(KERN_TLS) || defined(RATELIMIT)
2108 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2109 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2110 #endif
2111 	/* We need a Tx algorithm and at least one port */
2112 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2113 		m_freem(m);
2114 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2115 		return (ENXIO);
2116 	}
2117 
2118 	ETHER_BPF_MTAP(ifp, m);
2119 
2120 	return (lagg_proto_start(sc, m));
2121 }
2122 
2123 static int
lagg_transmit_infiniband(struct ifnet * ifp,struct mbuf * m)2124 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2125 {
2126 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2127 
2128 	NET_EPOCH_ASSERT();
2129 #if defined(KERN_TLS) || defined(RATELIMIT)
2130 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2131 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2132 #endif
2133 	/* We need a Tx algorithm and at least one port */
2134 	if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2135 		m_freem(m);
2136 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2137 		return (ENXIO);
2138 	}
2139 
2140 	infiniband_bpf_mtap(ifp, m);
2141 
2142 	return (lagg_proto_start(sc, m));
2143 }
2144 
2145 /*
2146  * The ifp->if_qflush entry point for lagg(4) is no-op.
2147  */
2148 static void
lagg_qflush(struct ifnet * ifp __unused)2149 lagg_qflush(struct ifnet *ifp __unused)
2150 {
2151 }
2152 
2153 static struct mbuf *
lagg_input_ethernet(struct ifnet * ifp,struct mbuf * m)2154 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2155 {
2156 	struct lagg_port *lp = ifp->if_lagg;
2157 	struct lagg_softc *sc = lp->lp_softc;
2158 	struct ifnet *scifp = sc->sc_ifp;
2159 
2160 	NET_EPOCH_ASSERT();
2161 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2162 	    lp->lp_detaching != 0 ||
2163 	    sc->sc_proto == LAGG_PROTO_NONE) {
2164 		m_freem(m);
2165 		return (NULL);
2166 	}
2167 
2168 	m = lagg_proto_input(sc, lp, m);
2169 	if (m != NULL) {
2170 		ETHER_BPF_MTAP(scifp, m);
2171 
2172 		if ((scifp->if_flags & IFF_MONITOR) != 0) {
2173 			m_freem(m);
2174 			m = NULL;
2175 		}
2176 	}
2177 
2178 #ifdef DEV_NETMAP
2179 	if (m != NULL && scifp->if_capenable & IFCAP_NETMAP) {
2180 		scifp->if_input(scifp, m);
2181 		m = NULL;
2182 	}
2183 #endif	/* DEV_NETMAP */
2184 
2185 	return (m);
2186 }
2187 
2188 static struct mbuf *
lagg_input_infiniband(struct ifnet * ifp,struct mbuf * m)2189 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2190 {
2191 	struct lagg_port *lp = ifp->if_lagg;
2192 	struct lagg_softc *sc = lp->lp_softc;
2193 	struct ifnet *scifp = sc->sc_ifp;
2194 
2195 	NET_EPOCH_ASSERT();
2196 	if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2197 	    lp->lp_detaching != 0 ||
2198 	    sc->sc_proto == LAGG_PROTO_NONE) {
2199 		m_freem(m);
2200 		return (NULL);
2201 	}
2202 
2203 	m = lagg_proto_input(sc, lp, m);
2204 	if (m != NULL) {
2205 		infiniband_bpf_mtap(scifp, m);
2206 
2207 		if ((scifp->if_flags & IFF_MONITOR) != 0) {
2208 			m_freem(m);
2209 			m = NULL;
2210 		}
2211 	}
2212 
2213 	return (m);
2214 }
2215 
2216 static int
lagg_media_change(struct ifnet * ifp)2217 lagg_media_change(struct ifnet *ifp)
2218 {
2219 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2220 
2221 	if (sc->sc_ifflags & IFF_DEBUG)
2222 		printf("%s\n", __func__);
2223 
2224 	/* Ignore */
2225 	return (0);
2226 }
2227 
2228 static void
lagg_media_status(struct ifnet * ifp,struct ifmediareq * imr)2229 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2230 {
2231 	struct epoch_tracker et;
2232 	struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2233 	struct lagg_port *lp;
2234 
2235 	imr->ifm_status = IFM_AVALID;
2236 	imr->ifm_active = IFM_ETHER | IFM_AUTO;
2237 
2238 	NET_EPOCH_ENTER(et);
2239 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2240 		if (LAGG_PORTACTIVE(lp))
2241 			imr->ifm_status |= IFM_ACTIVE;
2242 	}
2243 	NET_EPOCH_EXIT(et);
2244 }
2245 
2246 static void
lagg_linkstate(struct lagg_softc * sc)2247 lagg_linkstate(struct lagg_softc *sc)
2248 {
2249 	struct epoch_tracker et;
2250 	struct lagg_port *lp;
2251 	int new_link = LINK_STATE_DOWN;
2252 	uint64_t speed;
2253 
2254 	LAGG_XLOCK_ASSERT(sc);
2255 
2256 	/* LACP handles link state itself */
2257 	if (sc->sc_proto == LAGG_PROTO_LACP)
2258 		return;
2259 
2260 	/* Our link is considered up if at least one of our ports is active */
2261 	NET_EPOCH_ENTER(et);
2262 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2263 		if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2264 			new_link = LINK_STATE_UP;
2265 			break;
2266 		}
2267 	}
2268 	NET_EPOCH_EXIT(et);
2269 	if_link_state_change(sc->sc_ifp, new_link);
2270 
2271 	/* Update if_baudrate to reflect the max possible speed */
2272 	switch (sc->sc_proto) {
2273 		case LAGG_PROTO_FAILOVER:
2274 			sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2275 			    sc->sc_primary->lp_ifp->if_baudrate : 0;
2276 			break;
2277 		case LAGG_PROTO_ROUNDROBIN:
2278 		case LAGG_PROTO_LOADBALANCE:
2279 		case LAGG_PROTO_BROADCAST:
2280 			speed = 0;
2281 			NET_EPOCH_ENTER(et);
2282 			CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2283 				speed += lp->lp_ifp->if_baudrate;
2284 			NET_EPOCH_EXIT(et);
2285 			sc->sc_ifp->if_baudrate = speed;
2286 			break;
2287 		case LAGG_PROTO_LACP:
2288 			/* LACP updates if_baudrate itself */
2289 			break;
2290 	}
2291 }
2292 
2293 static void
lagg_port_state(struct ifnet * ifp,int state)2294 lagg_port_state(struct ifnet *ifp, int state)
2295 {
2296 	struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2297 	struct lagg_softc *sc = NULL;
2298 
2299 	if (lp != NULL)
2300 		sc = lp->lp_softc;
2301 	if (sc == NULL)
2302 		return;
2303 
2304 	LAGG_XLOCK(sc);
2305 	lagg_linkstate(sc);
2306 	lagg_proto_linkstate(sc, lp);
2307 	LAGG_XUNLOCK(sc);
2308 }
2309 
2310 struct lagg_port *
lagg_link_active(struct lagg_softc * sc,struct lagg_port * lp)2311 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2312 {
2313 	struct lagg_port *lp_next, *rval = NULL;
2314 
2315 	/*
2316 	 * Search a port which reports an active link state.
2317 	 */
2318 
2319 #ifdef INVARIANTS
2320 	/*
2321 	 * This is called with either in the network epoch
2322 	 * or with LAGG_XLOCK(sc) held.
2323 	 */
2324 	if (!in_epoch(net_epoch_preempt))
2325 		LAGG_XLOCK_ASSERT(sc);
2326 #endif
2327 
2328 	if (lp == NULL)
2329 		goto search;
2330 	if (LAGG_PORTACTIVE(lp)) {
2331 		rval = lp;
2332 		goto found;
2333 	}
2334 	if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2335 	    LAGG_PORTACTIVE(lp_next)) {
2336 		rval = lp_next;
2337 		goto found;
2338 	}
2339 
2340 search:
2341 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2342 		if (LAGG_PORTACTIVE(lp_next)) {
2343 			return (lp_next);
2344 		}
2345 	}
2346 found:
2347 	return (rval);
2348 }
2349 
2350 int
lagg_enqueue(struct ifnet * ifp,struct mbuf * m)2351 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2352 {
2353 
2354 #if defined(KERN_TLS) || defined(RATELIMIT)
2355 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2356 		struct lagg_snd_tag *lst;
2357 		struct m_snd_tag *mst;
2358 
2359 		mst = m->m_pkthdr.snd_tag;
2360 		lst = mst_to_lst(mst);
2361 		if (lst->tag->ifp != ifp) {
2362 			m_freem(m);
2363 			return (EAGAIN);
2364 		}
2365 		m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2366 		m_snd_tag_rele(mst);
2367 	}
2368 #endif
2369 	return (ifp->if_transmit)(ifp, m);
2370 }
2371 
2372 /*
2373  * Simple round robin aggregation
2374  */
2375 static void
lagg_rr_attach(struct lagg_softc * sc)2376 lagg_rr_attach(struct lagg_softc *sc)
2377 {
2378 	sc->sc_seq = 0;
2379 	sc->sc_stride = 1;
2380 }
2381 
2382 static int
lagg_rr_start(struct lagg_softc * sc,struct mbuf * m)2383 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2384 {
2385 	struct lagg_port *lp;
2386 	uint32_t p;
2387 
2388 	p = atomic_fetchadd_32(&sc->sc_seq, 1);
2389 	p /= sc->sc_stride;
2390 	p %= sc->sc_count;
2391 	lp = CK_SLIST_FIRST(&sc->sc_ports);
2392 
2393 	while (p--)
2394 		lp = CK_SLIST_NEXT(lp, lp_entries);
2395 
2396 	/*
2397 	 * Check the port's link state. This will return the next active
2398 	 * port if the link is down or the port is NULL.
2399 	 */
2400 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2401 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2402 		m_freem(m);
2403 		return (ENETDOWN);
2404 	}
2405 
2406 	/* Send mbuf */
2407 	return (lagg_enqueue(lp->lp_ifp, m));
2408 }
2409 
2410 /*
2411  * Broadcast mode
2412  */
2413 static int
lagg_bcast_start(struct lagg_softc * sc,struct mbuf * m)2414 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2415 {
2416 	int errors = 0;
2417 	int ret;
2418 	struct lagg_port *lp, *last = NULL;
2419 	struct mbuf *m0;
2420 
2421 	NET_EPOCH_ASSERT();
2422 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2423 		if (!LAGG_PORTACTIVE(lp))
2424 			continue;
2425 
2426 		if (last != NULL) {
2427 			m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2428 			if (m0 == NULL) {
2429 				ret = ENOBUFS;
2430 				errors++;
2431 				break;
2432 			}
2433 			lagg_enqueue(last->lp_ifp, m0);
2434 		}
2435 		last = lp;
2436 	}
2437 
2438 	if (last == NULL) {
2439 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2440 		m_freem(m);
2441 		return (ENOENT);
2442 	}
2443 	if ((last = lagg_link_active(sc, last)) == NULL) {
2444 		errors++;
2445 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2446 		m_freem(m);
2447 		return (ENETDOWN);
2448 	}
2449 
2450 	ret = lagg_enqueue(last->lp_ifp, m);
2451 	if (errors != 0)
2452 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2453 
2454 	return (ret);
2455 }
2456 
2457 /*
2458  * Active failover
2459  */
2460 static int
lagg_fail_start(struct lagg_softc * sc,struct mbuf * m)2461 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2462 {
2463 	struct lagg_port *lp;
2464 
2465 	/* Use the master port if active or the next available port */
2466 	if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2467 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2468 		m_freem(m);
2469 		return (ENETDOWN);
2470 	}
2471 
2472 	/* Send mbuf */
2473 	return (lagg_enqueue(lp->lp_ifp, m));
2474 }
2475 
2476 static struct mbuf *
lagg_fail_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)2477 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2478 {
2479 	struct ifnet *ifp = sc->sc_ifp;
2480 	struct lagg_port *tmp_tp;
2481 
2482 	if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2483 		m->m_pkthdr.rcvif = ifp;
2484 		return (m);
2485 	}
2486 
2487 	if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2488 		tmp_tp = lagg_link_active(sc, sc->sc_primary);
2489 		/*
2490 		 * If tmp_tp is null, we've received a packet when all
2491 		 * our links are down. Weird, but process it anyways.
2492 		 */
2493 		if (tmp_tp == NULL || tmp_tp == lp) {
2494 			m->m_pkthdr.rcvif = ifp;
2495 			return (m);
2496 		}
2497 	}
2498 
2499 	m_freem(m);
2500 	return (NULL);
2501 }
2502 
2503 /*
2504  * Loadbalancing
2505  */
2506 static void
lagg_lb_attach(struct lagg_softc * sc)2507 lagg_lb_attach(struct lagg_softc *sc)
2508 {
2509 	struct lagg_port *lp;
2510 	struct lagg_lb *lb;
2511 
2512 	LAGG_XLOCK_ASSERT(sc);
2513 	lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2514 	lb->lb_key = m_ether_tcpip_hash_init();
2515 	sc->sc_psc = lb;
2516 
2517 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2518 		lagg_lb_port_create(lp);
2519 }
2520 
2521 static void
lagg_lb_detach(struct lagg_softc * sc)2522 lagg_lb_detach(struct lagg_softc *sc)
2523 {
2524 	struct lagg_lb *lb;
2525 
2526 	lb = (struct lagg_lb *)sc->sc_psc;
2527 	if (lb != NULL)
2528 		free(lb, M_LAGG);
2529 }
2530 
2531 static int
lagg_lb_porttable(struct lagg_softc * sc,struct lagg_port * lp)2532 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2533 {
2534 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2535 	struct lagg_port *lp_next;
2536 	int i = 0, rv;
2537 
2538 	rv = 0;
2539 	bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2540 	LAGG_XLOCK_ASSERT(sc);
2541 	CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2542 		if (lp_next == lp)
2543 			continue;
2544 		if (i >= LAGG_MAX_PORTS) {
2545 			rv = EINVAL;
2546 			break;
2547 		}
2548 		if (sc->sc_ifflags & IFF_DEBUG)
2549 			printf("%s: port %s at index %d\n",
2550 			    sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2551 		lb->lb_ports[i++] = lp_next;
2552 	}
2553 
2554 	return (rv);
2555 }
2556 
2557 static int
lagg_lb_port_create(struct lagg_port * lp)2558 lagg_lb_port_create(struct lagg_port *lp)
2559 {
2560 	struct lagg_softc *sc = lp->lp_softc;
2561 	return (lagg_lb_porttable(sc, NULL));
2562 }
2563 
2564 static void
lagg_lb_port_destroy(struct lagg_port * lp)2565 lagg_lb_port_destroy(struct lagg_port *lp)
2566 {
2567 	struct lagg_softc *sc = lp->lp_softc;
2568 	lagg_lb_porttable(sc, lp);
2569 }
2570 
2571 static int
lagg_lb_start(struct lagg_softc * sc,struct mbuf * m)2572 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2573 {
2574 	struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2575 	struct lagg_port *lp = NULL;
2576 	uint32_t p = 0;
2577 
2578 	if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2579 	    M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2580 		p = m->m_pkthdr.flowid >> sc->flowid_shift;
2581 	else
2582 		p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2583 	p %= sc->sc_count;
2584 	lp = lb->lb_ports[p];
2585 
2586 	/*
2587 	 * Check the port's link state. This will return the next active
2588 	 * port if the link is down or the port is NULL.
2589 	 */
2590 	if ((lp = lagg_link_active(sc, lp)) == NULL) {
2591 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2592 		m_freem(m);
2593 		return (ENETDOWN);
2594 	}
2595 
2596 	/* Send mbuf */
2597 	return (lagg_enqueue(lp->lp_ifp, m));
2598 }
2599 
2600 /*
2601  * 802.3ad LACP
2602  */
2603 static void
lagg_lacp_attach(struct lagg_softc * sc)2604 lagg_lacp_attach(struct lagg_softc *sc)
2605 {
2606 	struct lagg_port *lp;
2607 
2608 	lacp_attach(sc);
2609 	LAGG_XLOCK_ASSERT(sc);
2610 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2611 		lacp_port_create(lp);
2612 }
2613 
2614 static void
lagg_lacp_detach(struct lagg_softc * sc)2615 lagg_lacp_detach(struct lagg_softc *sc)
2616 {
2617 	struct lagg_port *lp;
2618 	void *psc;
2619 
2620 	LAGG_XLOCK_ASSERT(sc);
2621 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2622 		lacp_port_destroy(lp);
2623 
2624 	psc = sc->sc_psc;
2625 	sc->sc_psc = NULL;
2626 	lacp_detach(psc);
2627 }
2628 
2629 static void
lagg_lacp_lladdr(struct lagg_softc * sc)2630 lagg_lacp_lladdr(struct lagg_softc *sc)
2631 {
2632 	struct lagg_port *lp;
2633 
2634 	LAGG_SXLOCK_ASSERT(sc);
2635 
2636 	/* purge all the lacp ports */
2637 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2638 		lacp_port_destroy(lp);
2639 
2640 	/* add them back in */
2641 	CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2642 		lacp_port_create(lp);
2643 }
2644 
2645 static int
lagg_lacp_start(struct lagg_softc * sc,struct mbuf * m)2646 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2647 {
2648 	struct lagg_port *lp;
2649 	int err;
2650 
2651 	lp = lacp_select_tx_port(sc, m, &err);
2652 	if (lp == NULL) {
2653 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2654 		m_freem(m);
2655 		return (err);
2656 	}
2657 
2658 	/* Send mbuf */
2659 	return (lagg_enqueue(lp->lp_ifp, m));
2660 }
2661 
2662 static struct mbuf *
lagg_lacp_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)2663 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2664 {
2665 	struct ifnet *ifp = sc->sc_ifp;
2666 	struct ether_header *eh;
2667 	u_short etype;
2668 
2669 	eh = mtod(m, struct ether_header *);
2670 	etype = ntohs(eh->ether_type);
2671 
2672 	/* Tap off LACP control messages */
2673 	if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2674 		m = lacp_input(lp, m);
2675 		if (m == NULL)
2676 			return (NULL);
2677 	}
2678 
2679 	/*
2680 	 * If the port is not collecting or not in the active aggregator then
2681 	 * free and return.
2682 	 */
2683 	if (!lacp_iscollecting(lp) || !lacp_isactive(lp)) {
2684 		m_freem(m);
2685 		return (NULL);
2686 	}
2687 
2688 	m->m_pkthdr.rcvif = ifp;
2689 	return (m);
2690 }
2691 
2692 /* Default input */
2693 static struct mbuf *
lagg_default_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)2694 lagg_default_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2695 {
2696 	struct ifnet *ifp = sc->sc_ifp;
2697 
2698 	/* Just pass in the packet to our lagg device */
2699 	m->m_pkthdr.rcvif = ifp;
2700 
2701 	return (m);
2702 }
2703