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 __FBSDID("$FreeBSD$");
23
24 #include "opt_inet.h"
25 #include "opt_inet6.h"
26 #include "opt_kern_tls.h"
27 #include "opt_ratelimit.h"
28
29 #include <sys/param.h>
30 #include <sys/kernel.h>
31 #include <sys/malloc.h>
32 #include <sys/mbuf.h>
33 #include <sys/queue.h>
34 #include <sys/socket.h>
35 #include <sys/sockio.h>
36 #include <sys/sysctl.h>
37 #include <sys/module.h>
38 #include <sys/priv.h>
39 #include <sys/systm.h>
40 #include <sys/proc.h>
41 #include <sys/lock.h>
42 #include <sys/rmlock.h>
43 #include <sys/sx.h>
44 #include <sys/taskqueue.h>
45 #include <sys/eventhandler.h>
46
47 #include <net/ethernet.h>
48 #include <net/if.h>
49 #include <net/if_clone.h>
50 #include <net/if_arp.h>
51 #include <net/if_dl.h>
52 #include <net/if_media.h>
53 #include <net/if_types.h>
54 #include <net/if_var.h>
55 #include <net/bpf.h>
56 #include <net/route.h>
57 #include <net/vnet.h>
58 #include <net/infiniband.h>
59
60 #if defined(INET) || defined(INET6)
61 #include <netinet/in.h>
62 #include <netinet/ip.h>
63 #endif
64 #ifdef INET
65 #include <netinet/in_systm.h>
66 #include <netinet/if_ether.h>
67 #endif
68
69 #ifdef INET6
70 #include <netinet/ip6.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet6/in6_ifattach.h>
73 #endif
74
75 #include <net/if_vlan_var.h>
76 #include <net/if_lagg.h>
77 #include <net/ieee8023ad_lacp.h>
78
79 #ifdef INET6
80 /*
81 * XXX: declare here to avoid to include many inet6 related files..
82 * should be more generalized?
83 */
84 extern void nd6_setmtu(struct ifnet *);
85 #endif
86
87 #define LAGG_SX_INIT(_sc) sx_init(&(_sc)->sc_sx, "if_lagg sx")
88 #define LAGG_SX_DESTROY(_sc) sx_destroy(&(_sc)->sc_sx)
89 #define LAGG_XLOCK(_sc) sx_xlock(&(_sc)->sc_sx)
90 #define LAGG_XUNLOCK(_sc) sx_xunlock(&(_sc)->sc_sx)
91 #define LAGG_SXLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_LOCKED)
92 #define LAGG_XLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
93
94 /* Special flags we should propagate to the lagg ports. */
95 static struct {
96 int flag;
97 int (*func)(struct ifnet *, int);
98 } lagg_pflags[] = {
99 {IFF_PROMISC, ifpromisc},
100 {IFF_ALLMULTI, if_allmulti},
101 {0, NULL}
102 };
103
104 struct lagg_snd_tag {
105 struct m_snd_tag com;
106 struct m_snd_tag *tag;
107 };
108
109 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
110 #define V_lagg_list VNET(lagg_list)
111 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
112 #define V_lagg_list_mtx VNET(lagg_list_mtx)
113 #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \
114 "if_lagg list", NULL, MTX_DEF)
115 #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx)
116 #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx)
117 #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx)
118 eventhandler_tag lagg_detach_cookie = NULL;
119
120 static int lagg_clone_create(struct if_clone *, int, caddr_t);
121 static void lagg_clone_destroy(struct ifnet *);
122 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
123 #define V_lagg_cloner VNET(lagg_cloner)
124 static const char laggname[] = "lagg";
125 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
126
127 static void lagg_capabilities(struct lagg_softc *);
128 static int lagg_port_create(struct lagg_softc *, struct ifnet *);
129 static int lagg_port_destroy(struct lagg_port *, int);
130 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *);
131 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *);
132 static void lagg_linkstate(struct lagg_softc *);
133 static void lagg_port_state(struct ifnet *, int);
134 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
135 static int lagg_port_output(struct ifnet *, struct mbuf *,
136 const struct sockaddr *, struct route *);
137 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *);
138 #ifdef LAGG_PORT_STACKING
139 static int lagg_port_checkstacking(struct lagg_softc *);
140 #endif
141 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *);
142 static void lagg_init(void *);
143 static void lagg_stop(struct lagg_softc *);
144 static int lagg_ioctl(struct ifnet *, u_long, caddr_t);
145 #if defined(KERN_TLS) || defined(RATELIMIT)
146 static int lagg_snd_tag_alloc(struct ifnet *,
147 union if_snd_tag_alloc_params *,
148 struct m_snd_tag **);
149 static int lagg_snd_tag_modify(struct m_snd_tag *,
150 union if_snd_tag_modify_params *);
151 static int lagg_snd_tag_query(struct m_snd_tag *,
152 union if_snd_tag_query_params *);
153 static void lagg_snd_tag_free(struct m_snd_tag *);
154 static struct m_snd_tag *lagg_next_snd_tag(struct m_snd_tag *);
155 static void lagg_ratelimit_query(struct ifnet *,
156 struct if_ratelimit_query_results *);
157 #endif
158 static int lagg_setmulti(struct lagg_port *);
159 static int lagg_clrmulti(struct lagg_port *);
160 static int lagg_setcaps(struct lagg_port *, int cap);
161 static int lagg_setflag(struct lagg_port *, int, int,
162 int (*func)(struct ifnet *, int));
163 static int lagg_setflags(struct lagg_port *, int status);
164 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
165 static int lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
166 static int lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
167 static void lagg_qflush(struct ifnet *);
168 static int lagg_media_change(struct ifnet *);
169 static void lagg_media_status(struct ifnet *, struct ifmediareq *);
170 static struct lagg_port *lagg_link_active(struct lagg_softc *,
171 struct lagg_port *);
172
173 /* Simple round robin */
174 static void lagg_rr_attach(struct lagg_softc *);
175 static int lagg_rr_start(struct lagg_softc *, struct mbuf *);
176 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
177 struct mbuf *);
178
179 /* Active failover */
180 static int lagg_fail_start(struct lagg_softc *, struct mbuf *);
181 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
182 struct mbuf *);
183
184 /* Loadbalancing */
185 static void lagg_lb_attach(struct lagg_softc *);
186 static void lagg_lb_detach(struct lagg_softc *);
187 static int lagg_lb_port_create(struct lagg_port *);
188 static void lagg_lb_port_destroy(struct lagg_port *);
189 static int lagg_lb_start(struct lagg_softc *, struct mbuf *);
190 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
191 struct mbuf *);
192 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
193
194 /* Broadcast */
195 static int lagg_bcast_start(struct lagg_softc *, struct mbuf *);
196 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
197 struct mbuf *);
198
199 /* 802.3ad LACP */
200 static void lagg_lacp_attach(struct lagg_softc *);
201 static void lagg_lacp_detach(struct lagg_softc *);
202 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *);
203 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
204 struct mbuf *);
205 static void lagg_lacp_lladdr(struct lagg_softc *);
206
207 /* lagg protocol table */
208 static const struct lagg_proto {
209 lagg_proto pr_num;
210 void (*pr_attach)(struct lagg_softc *);
211 void (*pr_detach)(struct lagg_softc *);
212 int (*pr_start)(struct lagg_softc *, struct mbuf *);
213 struct mbuf * (*pr_input)(struct lagg_softc *, struct lagg_port *,
214 struct mbuf *);
215 int (*pr_addport)(struct lagg_port *);
216 void (*pr_delport)(struct lagg_port *);
217 void (*pr_linkstate)(struct lagg_port *);
218 void (*pr_init)(struct lagg_softc *);
219 void (*pr_stop)(struct lagg_softc *);
220 void (*pr_lladdr)(struct lagg_softc *);
221 void (*pr_request)(struct lagg_softc *, void *);
222 void (*pr_portreq)(struct lagg_port *, void *);
223 } lagg_protos[] = {
224 {
225 .pr_num = LAGG_PROTO_NONE
226 },
227 {
228 .pr_num = LAGG_PROTO_ROUNDROBIN,
229 .pr_attach = lagg_rr_attach,
230 .pr_start = lagg_rr_start,
231 .pr_input = lagg_rr_input,
232 },
233 {
234 .pr_num = LAGG_PROTO_FAILOVER,
235 .pr_start = lagg_fail_start,
236 .pr_input = lagg_fail_input,
237 },
238 {
239 .pr_num = LAGG_PROTO_LOADBALANCE,
240 .pr_attach = lagg_lb_attach,
241 .pr_detach = lagg_lb_detach,
242 .pr_start = lagg_lb_start,
243 .pr_input = lagg_lb_input,
244 .pr_addport = lagg_lb_port_create,
245 .pr_delport = lagg_lb_port_destroy,
246 },
247 {
248 .pr_num = LAGG_PROTO_LACP,
249 .pr_attach = lagg_lacp_attach,
250 .pr_detach = lagg_lacp_detach,
251 .pr_start = lagg_lacp_start,
252 .pr_input = lagg_lacp_input,
253 .pr_addport = lacp_port_create,
254 .pr_delport = lacp_port_destroy,
255 .pr_linkstate = lacp_linkstate,
256 .pr_init = lacp_init,
257 .pr_stop = lacp_stop,
258 .pr_lladdr = lagg_lacp_lladdr,
259 .pr_request = lacp_req,
260 .pr_portreq = lacp_portreq,
261 },
262 {
263 .pr_num = LAGG_PROTO_BROADCAST,
264 .pr_start = lagg_bcast_start,
265 .pr_input = lagg_bcast_input,
266 },
267 };
268
269 SYSCTL_DECL(_net_link);
270 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
271 "Link Aggregation");
272
273 /* Allow input on any failover links */
274 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
275 #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all)
276 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
277 &VNET_NAME(lagg_failover_rx_all), 0,
278 "Accept input from any interface in a failover lagg");
279
280 /* Default value for using flowid */
281 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
282 #define V_def_use_flowid VNET(def_use_flowid)
283 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
284 &VNET_NAME(def_use_flowid), 0,
285 "Default setting for using flow id for load sharing");
286
287 /* Default value for using numa */
288 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
289 #define V_def_use_numa VNET(def_use_numa)
290 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN,
291 &VNET_NAME(def_use_numa), 0,
292 "Use numa to steer flows");
293
294 /* Default value for flowid shift */
295 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
296 #define V_def_flowid_shift VNET(def_flowid_shift)
297 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
298 &VNET_NAME(def_flowid_shift), 0,
299 "Default setting for flowid shift for load sharing");
300
301 static void
vnet_lagg_init(const void * unused __unused)302 vnet_lagg_init(const void *unused __unused)
303 {
304
305 LAGG_LIST_LOCK_INIT();
306 SLIST_INIT(&V_lagg_list);
307 V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
308 lagg_clone_destroy, 0);
309 }
310 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
311 vnet_lagg_init, NULL);
312
313 static void
vnet_lagg_uninit(const void * unused __unused)314 vnet_lagg_uninit(const void *unused __unused)
315 {
316
317 if_clone_detach(V_lagg_cloner);
318 LAGG_LIST_LOCK_DESTROY();
319 }
320 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
321 vnet_lagg_uninit, NULL);
322
323 static int
lagg_modevent(module_t mod,int type,void * data)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
lagg_proto_attach(struct lagg_softc * sc,lagg_proto pr)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
lagg_proto_detach(struct lagg_softc * sc)376 lagg_proto_detach(struct lagg_softc *sc)
377 {
378 lagg_proto pr;
379
380 LAGG_XLOCK_ASSERT(sc);
381 pr = sc->sc_proto;
382 sc->sc_proto = LAGG_PROTO_NONE;
383
384 if (lagg_protos[pr].pr_detach != NULL)
385 lagg_protos[pr].pr_detach(sc);
386 }
387
388 static int
lagg_proto_start(struct lagg_softc * sc,struct mbuf * m)389 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
390 {
391
392 return (lagg_protos[sc->sc_proto].pr_start(sc, m));
393 }
394
395 static struct mbuf *
lagg_proto_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)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
lagg_proto_addport(struct lagg_softc * sc,struct lagg_port * lp)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
lagg_proto_delport(struct lagg_softc * sc,struct lagg_port * lp)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
lagg_proto_linkstate(struct lagg_softc * sc,struct lagg_port * lp)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
lagg_proto_init(struct lagg_softc * sc)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
lagg_proto_stop(struct lagg_softc * sc)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
lagg_proto_lladdr(struct lagg_softc * sc)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
lagg_proto_request(struct lagg_softc * sc,void * v)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
lagg_proto_portreq(struct lagg_softc * sc,struct lagg_port * lp,void * v)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
lagg_register_vlan(void * arg,struct ifnet * ifp,u_int16_t vtag)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
lagg_unregister_vlan(void * arg,struct ifnet * ifp,u_int16_t vtag)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
lagg_clone_create(struct if_clone * ifc,int unit,caddr_t params)507 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
508 {
509 struct iflaggparam iflp;
510 struct lagg_softc *sc;
511 struct ifnet *ifp;
512 int if_type;
513 int error;
514 static const uint8_t eaddr[LAGG_ADDR_LEN];
515
516 if (params != NULL) {
517 error = copyin(params, &iflp, sizeof(iflp));
518 if (error)
519 return (error);
520
521 switch (iflp.lagg_type) {
522 case LAGG_TYPE_ETHERNET:
523 if_type = IFT_ETHER;
524 break;
525 case LAGG_TYPE_INFINIBAND:
526 if_type = IFT_INFINIBAND;
527 break;
528 default:
529 return (EINVAL);
530 }
531 } else {
532 if_type = IFT_ETHER;
533 }
534
535 sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO);
536 ifp = sc->sc_ifp = if_alloc(if_type);
537 if (ifp == NULL) {
538 free(sc, M_LAGG);
539 return (ENOSPC);
540 }
541 LAGG_SX_INIT(sc);
542
543 mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
544 callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
545
546 LAGG_XLOCK(sc);
547 if (V_def_use_flowid)
548 sc->sc_opts |= LAGG_OPT_USE_FLOWID;
549 if (V_def_use_numa)
550 sc->sc_opts |= LAGG_OPT_USE_NUMA;
551 sc->flowid_shift = V_def_flowid_shift;
552
553 /* Hash all layers by default */
554 sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
555
556 lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
557
558 CK_SLIST_INIT(&sc->sc_ports);
559
560 switch (if_type) {
561 case IFT_ETHER:
562 /* Initialise pseudo media types */
563 ifmedia_init(&sc->sc_media, 0, lagg_media_change,
564 lagg_media_status);
565 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
566 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
567
568 if_initname(ifp, laggname, unit);
569 ifp->if_transmit = lagg_transmit_ethernet;
570 break;
571 case IFT_INFINIBAND:
572 if_initname(ifp, laggname, unit);
573 ifp->if_transmit = lagg_transmit_infiniband;
574 break;
575 default:
576 break;
577 }
578 ifp->if_softc = sc;
579 ifp->if_qflush = lagg_qflush;
580 ifp->if_init = lagg_init;
581 ifp->if_ioctl = lagg_ioctl;
582 ifp->if_get_counter = lagg_get_counter;
583 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
584 #if defined(KERN_TLS) || defined(RATELIMIT)
585 ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
586 ifp->if_snd_tag_modify = lagg_snd_tag_modify;
587 ifp->if_snd_tag_query = lagg_snd_tag_query;
588 ifp->if_snd_tag_free = lagg_snd_tag_free;
589 ifp->if_next_snd_tag = lagg_next_snd_tag;
590 ifp->if_ratelimit_query = lagg_ratelimit_query;
591 #endif
592 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
593
594 /*
595 * Attach as an ordinary ethernet device, children will be attached
596 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
597 */
598 switch (if_type) {
599 case IFT_ETHER:
600 ether_ifattach(ifp, eaddr);
601 break;
602 case IFT_INFINIBAND:
603 infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr);
604 break;
605 default:
606 break;
607 }
608
609 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
610 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
611 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
612 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
613
614 /* Insert into the global list of laggs */
615 LAGG_LIST_LOCK();
616 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
617 LAGG_LIST_UNLOCK();
618 LAGG_XUNLOCK(sc);
619
620 return (0);
621 }
622
623 static void
lagg_clone_destroy(struct ifnet * ifp)624 lagg_clone_destroy(struct ifnet *ifp)
625 {
626 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
627 struct lagg_port *lp;
628
629 LAGG_XLOCK(sc);
630 sc->sc_destroying = 1;
631 lagg_stop(sc);
632 ifp->if_flags &= ~IFF_UP;
633
634 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
635 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
636
637 /* Shutdown and remove lagg ports */
638 while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
639 lagg_port_destroy(lp, 1);
640
641 /* Unhook the aggregation protocol */
642 lagg_proto_detach(sc);
643 LAGG_XUNLOCK(sc);
644
645 switch (ifp->if_type) {
646 case IFT_ETHER:
647 ifmedia_removeall(&sc->sc_media);
648 ether_ifdetach(ifp);
649 break;
650 case IFT_INFINIBAND:
651 infiniband_ifdetach(ifp);
652 break;
653 default:
654 break;
655 }
656 if_free(ifp);
657
658 LAGG_LIST_LOCK();
659 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
660 LAGG_LIST_UNLOCK();
661
662 mtx_destroy(&sc->sc_mtx);
663 LAGG_SX_DESTROY(sc);
664 free(sc, M_LAGG);
665 }
666
667 static void
lagg_capabilities(struct lagg_softc * sc)668 lagg_capabilities(struct lagg_softc *sc)
669 {
670 struct lagg_port *lp;
671 int cap, ena, pena;
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 = ~0;
679 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
680 ena &= lp->lp_ifp->if_capenable;
681 ena = (ena == ~0 ? 0 : ena);
682
683 /*
684 * Apply common enabled capabilities back to the lagg ports.
685 * May require several iterations if they are dependent.
686 */
687 do {
688 pena = ena;
689 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
690 lagg_setcaps(lp, ena);
691 ena &= lp->lp_ifp->if_capenable;
692 }
693 } while (pena != ena);
694
695 /* Get other capabilities from the lagg ports */
696 cap = ~0;
697 hwa = ~(uint64_t)0;
698 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
699 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
700 cap &= lp->lp_ifp->if_capabilities;
701 hwa &= lp->lp_ifp->if_hwassist;
702 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
703 }
704 cap = (cap == ~0 ? 0 : cap);
705 hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
706
707 if (sc->sc_ifp->if_capabilities != cap ||
708 sc->sc_ifp->if_capenable != ena ||
709 sc->sc_ifp->if_hwassist != hwa ||
710 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
711 sc->sc_ifp->if_capabilities = cap;
712 sc->sc_ifp->if_capenable = ena;
713 sc->sc_ifp->if_hwassist = hwa;
714 getmicrotime(&sc->sc_ifp->if_lastchange);
715
716 if (sc->sc_ifflags & IFF_DEBUG)
717 if_printf(sc->sc_ifp,
718 "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
719 }
720 }
721
722 static int
lagg_port_create(struct lagg_softc * sc,struct ifnet * ifp)723 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
724 {
725 struct lagg_softc *sc_ptr;
726 struct lagg_port *lp, *tlp;
727 struct ifreq ifr;
728 int error, i, oldmtu;
729 int if_type;
730 uint64_t *pval;
731
732 LAGG_XLOCK_ASSERT(sc);
733
734 if (sc->sc_ifp == ifp) {
735 if_printf(sc->sc_ifp,
736 "cannot add a lagg to itself as a port\n");
737 return (EINVAL);
738 }
739
740 if (sc->sc_destroying == 1)
741 return (ENXIO);
742
743 /* Limit the maximal number of lagg ports */
744 if (sc->sc_count >= LAGG_MAX_PORTS)
745 return (ENOSPC);
746
747 /* Check if port has already been associated to a lagg */
748 if (ifp->if_lagg != NULL) {
749 /* Port is already in the current lagg? */
750 lp = (struct lagg_port *)ifp->if_lagg;
751 if (lp->lp_softc == sc)
752 return (EEXIST);
753 return (EBUSY);
754 }
755
756 switch (sc->sc_ifp->if_type) {
757 case IFT_ETHER:
758 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */
759 if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
760 return (EPROTONOSUPPORT);
761 if_type = IFT_IEEE8023ADLAG;
762 break;
763 case IFT_INFINIBAND:
764 /* XXX Disallow non-infiniband interfaces */
765 if (ifp->if_type != IFT_INFINIBAND)
766 return (EPROTONOSUPPORT);
767 if_type = IFT_INFINIBANDLAG;
768 break;
769 default:
770 break;
771 }
772
773 /* Allow the first Ethernet member to define the MTU */
774 oldmtu = -1;
775 if (CK_SLIST_EMPTY(&sc->sc_ports)) {
776 sc->sc_ifp->if_mtu = ifp->if_mtu;
777 } else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
778 if (ifp->if_ioctl == NULL) {
779 if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
780 ifp->if_xname);
781 return (EINVAL);
782 }
783 oldmtu = ifp->if_mtu;
784 strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
785 ifr.ifr_mtu = sc->sc_ifp->if_mtu;
786 error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
787 if (error != 0) {
788 if_printf(sc->sc_ifp, "invalid MTU for %s\n",
789 ifp->if_xname);
790 return (error);
791 }
792 ifr.ifr_mtu = oldmtu;
793 }
794
795 lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO);
796 lp->lp_softc = sc;
797
798 /* Check if port is a stacked lagg */
799 LAGG_LIST_LOCK();
800 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
801 if (ifp == sc_ptr->sc_ifp) {
802 LAGG_LIST_UNLOCK();
803 free(lp, M_LAGG);
804 if (oldmtu != -1)
805 (*ifp->if_ioctl)(ifp, SIOCSIFMTU,
806 (caddr_t)&ifr);
807 return (EINVAL);
808 /* XXX disable stacking for the moment, its untested */
809 #ifdef LAGG_PORT_STACKING
810 lp->lp_flags |= LAGG_PORT_STACK;
811 if (lagg_port_checkstacking(sc_ptr) >=
812 LAGG_MAX_STACKING) {
813 LAGG_LIST_UNLOCK();
814 free(lp, M_LAGG);
815 if (oldmtu != -1)
816 (*ifp->if_ioctl)(ifp, SIOCSIFMTU,
817 (caddr_t)&ifr);
818 return (E2BIG);
819 }
820 #endif
821 }
822 }
823 LAGG_LIST_UNLOCK();
824
825 if_ref(ifp);
826 lp->lp_ifp = ifp;
827
828 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
829 lp->lp_ifcapenable = ifp->if_capenable;
830 if (CK_SLIST_EMPTY(&sc->sc_ports)) {
831 bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
832 lagg_proto_lladdr(sc);
833 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
834 } else {
835 if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
836 }
837 lagg_setflags(lp, 1);
838
839 if (CK_SLIST_EMPTY(&sc->sc_ports))
840 sc->sc_primary = lp;
841
842 /* Change the interface type */
843 lp->lp_iftype = ifp->if_type;
844 ifp->if_type = if_type;
845 ifp->if_lagg = lp;
846 lp->lp_ioctl = ifp->if_ioctl;
847 ifp->if_ioctl = lagg_port_ioctl;
848 lp->lp_output = ifp->if_output;
849 ifp->if_output = lagg_port_output;
850
851 /* Read port counters */
852 pval = lp->port_counters.val;
853 for (i = 0; i < IFCOUNTERS; i++, pval++)
854 *pval = ifp->if_get_counter(ifp, i);
855
856 /*
857 * Insert into the list of ports.
858 * Keep ports sorted by if_index. It is handy, when configuration
859 * is predictable and `ifconfig laggN create ...` command
860 * will lead to the same result each time.
861 */
862 CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
863 if (tlp->lp_ifp->if_index < ifp->if_index && (
864 CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
865 ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
866 ifp->if_index))
867 break;
868 }
869 if (tlp != NULL)
870 CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
871 else
872 CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
873 sc->sc_count++;
874
875 lagg_setmulti(lp);
876
877 if ((error = lagg_proto_addport(sc, lp)) != 0) {
878 /* Remove the port, without calling pr_delport. */
879 lagg_port_destroy(lp, 0);
880 if (oldmtu != -1)
881 (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
882 return (error);
883 }
884
885 /* Update lagg capabilities */
886 lagg_capabilities(sc);
887 lagg_linkstate(sc);
888
889 return (0);
890 }
891
892 #ifdef LAGG_PORT_STACKING
893 static int
lagg_port_checkstacking(struct lagg_softc * sc)894 lagg_port_checkstacking(struct lagg_softc *sc)
895 {
896 struct lagg_softc *sc_ptr;
897 struct lagg_port *lp;
898 int m = 0;
899
900 LAGG_SXLOCK_ASSERT(sc);
901 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
902 if (lp->lp_flags & LAGG_PORT_STACK) {
903 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
904 m = MAX(m, lagg_port_checkstacking(sc_ptr));
905 }
906 }
907
908 return (m + 1);
909 }
910 #endif
911
912 static void
lagg_port_destroy_cb(epoch_context_t ec)913 lagg_port_destroy_cb(epoch_context_t ec)
914 {
915 struct lagg_port *lp;
916 struct ifnet *ifp;
917
918 lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
919 ifp = lp->lp_ifp;
920
921 if_rele(ifp);
922 free(lp, M_LAGG);
923 }
924
925 static int
lagg_port_destroy(struct lagg_port * lp,int rundelport)926 lagg_port_destroy(struct lagg_port *lp, int rundelport)
927 {
928 struct lagg_softc *sc = lp->lp_softc;
929 struct lagg_port *lp_ptr, *lp0;
930 struct ifnet *ifp = lp->lp_ifp;
931 uint64_t *pval, vdiff;
932 int i;
933
934 LAGG_XLOCK_ASSERT(sc);
935
936 if (rundelport)
937 lagg_proto_delport(sc, lp);
938
939 if (lp->lp_detaching == 0)
940 lagg_clrmulti(lp);
941
942 /* Restore interface */
943 ifp->if_type = lp->lp_iftype;
944 ifp->if_ioctl = lp->lp_ioctl;
945 ifp->if_output = lp->lp_output;
946 ifp->if_lagg = NULL;
947
948 /* Update detached port counters */
949 pval = lp->port_counters.val;
950 for (i = 0; i < IFCOUNTERS; i++, pval++) {
951 vdiff = ifp->if_get_counter(ifp, i) - *pval;
952 sc->detached_counters.val[i] += vdiff;
953 }
954
955 /* Finally, remove the port from the lagg */
956 CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
957 sc->sc_count--;
958
959 /* Update the primary interface */
960 if (lp == sc->sc_primary) {
961 uint8_t lladdr[LAGG_ADDR_LEN];
962
963 if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
964 bzero(&lladdr, LAGG_ADDR_LEN);
965 else
966 bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
967 sc->sc_primary = lp0;
968 if (sc->sc_destroying == 0) {
969 bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
970 lagg_proto_lladdr(sc);
971 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
972
973 /*
974 * Update lladdr for each port (new primary needs update
975 * as well, to switch from old lladdr to its 'real' one).
976 * We can skip this if the lagg is being destroyed.
977 */
978 CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
979 if_setlladdr(lp_ptr->lp_ifp, lladdr,
980 lp_ptr->lp_ifp->if_addrlen);
981 }
982 }
983
984 if (lp->lp_ifflags)
985 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
986
987 if (lp->lp_detaching == 0) {
988 lagg_setflags(lp, 0);
989 lagg_setcaps(lp, lp->lp_ifcapenable);
990 if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
991 }
992
993 /*
994 * free port and release it's ifnet reference after a grace period has
995 * elapsed.
996 */
997 NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
998 /* Update lagg capabilities */
999 lagg_capabilities(sc);
1000 lagg_linkstate(sc);
1001
1002 return (0);
1003 }
1004
1005 static int
lagg_port_ioctl(struct ifnet * ifp,u_long cmd,caddr_t data)1006 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1007 {
1008 struct epoch_tracker et;
1009 struct lagg_reqport *rp = (struct lagg_reqport *)data;
1010 struct lagg_softc *sc;
1011 struct lagg_port *lp = NULL;
1012 int error = 0;
1013
1014 /* Should be checked by the caller */
1015 switch (ifp->if_type) {
1016 case IFT_IEEE8023ADLAG:
1017 case IFT_INFINIBANDLAG:
1018 if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1019 goto fallback;
1020 break;
1021 default:
1022 goto fallback;
1023 }
1024
1025 switch (cmd) {
1026 case SIOCGLAGGPORT:
1027 if (rp->rp_portname[0] == '\0' ||
1028 ifunit(rp->rp_portname) != ifp) {
1029 error = EINVAL;
1030 break;
1031 }
1032
1033 NET_EPOCH_ENTER(et);
1034 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1035 error = ENOENT;
1036 NET_EPOCH_EXIT(et);
1037 break;
1038 }
1039
1040 lagg_port2req(lp, rp);
1041 NET_EPOCH_EXIT(et);
1042 break;
1043
1044 case SIOCSIFCAP:
1045 if (lp->lp_ioctl == NULL) {
1046 error = EINVAL;
1047 break;
1048 }
1049 error = (*lp->lp_ioctl)(ifp, cmd, data);
1050 if (error)
1051 break;
1052
1053 /* Update lagg interface capabilities */
1054 LAGG_XLOCK(sc);
1055 lagg_capabilities(sc);
1056 LAGG_XUNLOCK(sc);
1057 VLAN_CAPABILITIES(sc->sc_ifp);
1058 break;
1059
1060 case SIOCSIFMTU:
1061 /* Do not allow the MTU to be changed once joined */
1062 error = EINVAL;
1063 break;
1064
1065 default:
1066 goto fallback;
1067 }
1068
1069 return (error);
1070
1071 fallback:
1072 if (lp != NULL && lp->lp_ioctl != NULL)
1073 return ((*lp->lp_ioctl)(ifp, cmd, data));
1074
1075 return (EINVAL);
1076 }
1077
1078 /*
1079 * Requests counter @cnt data.
1080 *
1081 * Counter value is calculated the following way:
1082 * 1) for each port, sum difference between current and "initial" measurements.
1083 * 2) add lagg logical interface counters.
1084 * 3) add data from detached_counters array.
1085 *
1086 * We also do the following things on ports attach/detach:
1087 * 1) On port attach we store all counters it has into port_counter array.
1088 * 2) On port detach we add the different between "initial" and
1089 * current counters data to detached_counters array.
1090 */
1091 static uint64_t
lagg_get_counter(struct ifnet * ifp,ift_counter cnt)1092 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1093 {
1094 struct epoch_tracker et;
1095 struct lagg_softc *sc;
1096 struct lagg_port *lp;
1097 struct ifnet *lpifp;
1098 uint64_t newval, oldval, vsum;
1099
1100 /* Revise this when we've got non-generic counters. */
1101 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1102
1103 sc = (struct lagg_softc *)ifp->if_softc;
1104
1105 vsum = 0;
1106 NET_EPOCH_ENTER(et);
1107 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1108 /* Saved attached value */
1109 oldval = lp->port_counters.val[cnt];
1110 /* current value */
1111 lpifp = lp->lp_ifp;
1112 newval = lpifp->if_get_counter(lpifp, cnt);
1113 /* Calculate diff and save new */
1114 vsum += newval - oldval;
1115 }
1116 NET_EPOCH_EXIT(et);
1117
1118 /*
1119 * Add counter data which might be added by upper
1120 * layer protocols operating on logical interface.
1121 */
1122 vsum += if_get_counter_default(ifp, cnt);
1123
1124 /*
1125 * Add counter data from detached ports counters
1126 */
1127 vsum += sc->detached_counters.val[cnt];
1128
1129 return (vsum);
1130 }
1131
1132 /*
1133 * For direct output to child ports.
1134 */
1135 static int
lagg_port_output(struct ifnet * ifp,struct mbuf * m,const struct sockaddr * dst,struct route * ro)1136 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1137 const struct sockaddr *dst, struct route *ro)
1138 {
1139 struct lagg_port *lp = ifp->if_lagg;
1140
1141 switch (dst->sa_family) {
1142 case pseudo_AF_HDRCMPLT:
1143 case AF_UNSPEC:
1144 if (lp != NULL)
1145 return ((*lp->lp_output)(ifp, m, dst, ro));
1146 }
1147
1148 /* drop any other frames */
1149 m_freem(m);
1150 return (ENETDOWN);
1151 }
1152
1153 static void
lagg_port_ifdetach(void * arg __unused,struct ifnet * ifp)1154 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1155 {
1156 struct lagg_port *lp;
1157 struct lagg_softc *sc;
1158
1159 if ((lp = ifp->if_lagg) == NULL)
1160 return;
1161 /* If the ifnet is just being renamed, don't do anything. */
1162 if (ifp->if_flags & IFF_RENAMING)
1163 return;
1164
1165 sc = lp->lp_softc;
1166
1167 LAGG_XLOCK(sc);
1168 lp->lp_detaching = 1;
1169 lagg_port_destroy(lp, 1);
1170 LAGG_XUNLOCK(sc);
1171 VLAN_CAPABILITIES(sc->sc_ifp);
1172 }
1173
1174 static void
lagg_port2req(struct lagg_port * lp,struct lagg_reqport * rp)1175 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1176 {
1177 struct lagg_softc *sc = lp->lp_softc;
1178
1179 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1180 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1181 rp->rp_prio = lp->lp_prio;
1182 rp->rp_flags = lp->lp_flags;
1183 lagg_proto_portreq(sc, lp, &rp->rp_psc);
1184
1185 /* Add protocol specific flags */
1186 switch (sc->sc_proto) {
1187 case LAGG_PROTO_FAILOVER:
1188 if (lp == sc->sc_primary)
1189 rp->rp_flags |= LAGG_PORT_MASTER;
1190 if (lp == lagg_link_active(sc, sc->sc_primary))
1191 rp->rp_flags |= LAGG_PORT_ACTIVE;
1192 break;
1193
1194 case LAGG_PROTO_ROUNDROBIN:
1195 case LAGG_PROTO_LOADBALANCE:
1196 case LAGG_PROTO_BROADCAST:
1197 if (LAGG_PORTACTIVE(lp))
1198 rp->rp_flags |= LAGG_PORT_ACTIVE;
1199 break;
1200
1201 case LAGG_PROTO_LACP:
1202 /* LACP has a different definition of active */
1203 if (lacp_isactive(lp))
1204 rp->rp_flags |= LAGG_PORT_ACTIVE;
1205 if (lacp_iscollecting(lp))
1206 rp->rp_flags |= LAGG_PORT_COLLECTING;
1207 if (lacp_isdistributing(lp))
1208 rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1209 break;
1210 }
1211
1212 }
1213
1214 static void
lagg_watchdog_infiniband(void * arg)1215 lagg_watchdog_infiniband(void *arg)
1216 {
1217 struct epoch_tracker et;
1218 struct lagg_softc *sc;
1219 struct lagg_port *lp;
1220 struct ifnet *ifp;
1221 struct ifnet *lp_ifp;
1222
1223 sc = arg;
1224
1225 /*
1226 * Because infiniband nodes have a fixed MAC address, which is
1227 * generated by the so-called GID, we need to regularly update
1228 * the link level address of the parent lagg<N> device when
1229 * the active port changes. Possibly we could piggy-back on
1230 * link up/down events aswell, but using a timer also provides
1231 * a guarantee against too frequent events. This operation
1232 * does not have to be atomic.
1233 */
1234 NET_EPOCH_ENTER(et);
1235 lp = lagg_link_active(sc, sc->sc_primary);
1236 if (lp != NULL) {
1237 ifp = sc->sc_ifp;
1238 lp_ifp = lp->lp_ifp;
1239
1240 if (ifp != NULL && lp_ifp != NULL &&
1241 (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0 ||
1242 memcmp(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen) != 0)) {
1243 memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1244 memcpy(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen);
1245
1246 CURVNET_SET(ifp->if_vnet);
1247 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1248 CURVNET_RESTORE();
1249 }
1250 }
1251 NET_EPOCH_EXIT(et);
1252
1253 callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1254 }
1255
1256 static void
lagg_init(void * xsc)1257 lagg_init(void *xsc)
1258 {
1259 struct lagg_softc *sc = (struct lagg_softc *)xsc;
1260 struct ifnet *ifp = sc->sc_ifp;
1261 struct lagg_port *lp;
1262
1263 LAGG_XLOCK(sc);
1264 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1265 LAGG_XUNLOCK(sc);
1266 return;
1267 }
1268
1269 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1270
1271 /*
1272 * Update the port lladdrs if needed.
1273 * This might be if_setlladdr() notification
1274 * that lladdr has been changed.
1275 */
1276 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1277 if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1278 ifp->if_addrlen) != 0)
1279 if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1280 }
1281
1282 lagg_proto_init(sc);
1283
1284 if (ifp->if_type == IFT_INFINIBAND) {
1285 mtx_lock(&sc->sc_mtx);
1286 lagg_watchdog_infiniband(sc);
1287 mtx_unlock(&sc->sc_mtx);
1288 }
1289
1290 LAGG_XUNLOCK(sc);
1291 }
1292
1293 static void
lagg_stop(struct lagg_softc * sc)1294 lagg_stop(struct lagg_softc *sc)
1295 {
1296 struct ifnet *ifp = sc->sc_ifp;
1297
1298 LAGG_XLOCK_ASSERT(sc);
1299
1300 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1301 return;
1302
1303 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1304
1305 lagg_proto_stop(sc);
1306
1307 mtx_lock(&sc->sc_mtx);
1308 callout_stop(&sc->sc_watchdog);
1309 mtx_unlock(&sc->sc_mtx);
1310
1311 callout_drain(&sc->sc_watchdog);
1312 }
1313
1314 static int
lagg_ioctl(struct ifnet * ifp,u_long cmd,caddr_t data)1315 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1316 {
1317 struct epoch_tracker et;
1318 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1319 struct lagg_reqall *ra = (struct lagg_reqall *)data;
1320 struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1321 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1322 struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1323 struct ifreq *ifr = (struct ifreq *)data;
1324 struct lagg_port *lp;
1325 struct ifnet *tpif;
1326 struct thread *td = curthread;
1327 char *buf, *outbuf;
1328 int count, buflen, len, error = 0, oldmtu;
1329
1330 bzero(&rpbuf, sizeof(rpbuf));
1331
1332 /* XXX: This can race with lagg_clone_destroy. */
1333
1334 switch (cmd) {
1335 case SIOCGLAGG:
1336 LAGG_XLOCK(sc);
1337 buflen = sc->sc_count * sizeof(struct lagg_reqport);
1338 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1339 ra->ra_proto = sc->sc_proto;
1340 lagg_proto_request(sc, &ra->ra_psc);
1341 count = 0;
1342 buf = outbuf;
1343 len = min(ra->ra_size, buflen);
1344 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1345 if (len < sizeof(rpbuf))
1346 break;
1347
1348 lagg_port2req(lp, &rpbuf);
1349 memcpy(buf, &rpbuf, sizeof(rpbuf));
1350 count++;
1351 buf += sizeof(rpbuf);
1352 len -= sizeof(rpbuf);
1353 }
1354 LAGG_XUNLOCK(sc);
1355 ra->ra_ports = count;
1356 ra->ra_size = count * sizeof(rpbuf);
1357 error = copyout(outbuf, ra->ra_port, ra->ra_size);
1358 free(outbuf, M_TEMP);
1359 break;
1360 case SIOCSLAGG:
1361 error = priv_check(td, PRIV_NET_LAGG);
1362 if (error)
1363 break;
1364 if (ra->ra_proto >= LAGG_PROTO_MAX) {
1365 error = EPROTONOSUPPORT;
1366 break;
1367 }
1368 /* Infiniband only supports the failover protocol. */
1369 if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1370 ifp->if_type == IFT_INFINIBAND) {
1371 error = EPROTONOSUPPORT;
1372 break;
1373 }
1374 LAGG_XLOCK(sc);
1375 lagg_proto_detach(sc);
1376 lagg_proto_attach(sc, ra->ra_proto);
1377 LAGG_XUNLOCK(sc);
1378 break;
1379 case SIOCGLAGGOPTS:
1380 LAGG_XLOCK(sc);
1381 ro->ro_opts = sc->sc_opts;
1382 if (sc->sc_proto == LAGG_PROTO_LACP) {
1383 struct lacp_softc *lsc;
1384
1385 lsc = (struct lacp_softc *)sc->sc_psc;
1386 if (lsc->lsc_debug.lsc_tx_test != 0)
1387 ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1388 if (lsc->lsc_debug.lsc_rx_test != 0)
1389 ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1390 if (lsc->lsc_strict_mode != 0)
1391 ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1392 if (lsc->lsc_fast_timeout != 0)
1393 ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1394
1395 ro->ro_active = sc->sc_active;
1396 } else {
1397 ro->ro_active = 0;
1398 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1399 ro->ro_active += LAGG_PORTACTIVE(lp);
1400 }
1401 ro->ro_bkt = sc->sc_stride;
1402 ro->ro_flapping = sc->sc_flapping;
1403 ro->ro_flowid_shift = sc->flowid_shift;
1404 LAGG_XUNLOCK(sc);
1405 break;
1406 case SIOCSLAGGOPTS:
1407 error = priv_check(td, PRIV_NET_LAGG);
1408 if (error)
1409 break;
1410
1411 /*
1412 * The stride option was added without defining a corresponding
1413 * LAGG_OPT flag, so handle a non-zero value before checking
1414 * anything else to preserve compatibility.
1415 */
1416 LAGG_XLOCK(sc);
1417 if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1418 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1419 LAGG_XUNLOCK(sc);
1420 error = EINVAL;
1421 break;
1422 }
1423 sc->sc_stride = ro->ro_bkt;
1424 }
1425 if (ro->ro_opts == 0) {
1426 LAGG_XUNLOCK(sc);
1427 break;
1428 }
1429
1430 /*
1431 * Set options. LACP options are stored in sc->sc_psc,
1432 * not in sc_opts.
1433 */
1434 int valid, lacp;
1435
1436 switch (ro->ro_opts) {
1437 case LAGG_OPT_USE_FLOWID:
1438 case -LAGG_OPT_USE_FLOWID:
1439 case LAGG_OPT_USE_NUMA:
1440 case -LAGG_OPT_USE_NUMA:
1441 case LAGG_OPT_FLOWIDSHIFT:
1442 case LAGG_OPT_RR_LIMIT:
1443 valid = 1;
1444 lacp = 0;
1445 break;
1446 case LAGG_OPT_LACP_TXTEST:
1447 case -LAGG_OPT_LACP_TXTEST:
1448 case LAGG_OPT_LACP_RXTEST:
1449 case -LAGG_OPT_LACP_RXTEST:
1450 case LAGG_OPT_LACP_STRICT:
1451 case -LAGG_OPT_LACP_STRICT:
1452 case LAGG_OPT_LACP_FAST_TIMO:
1453 case -LAGG_OPT_LACP_FAST_TIMO:
1454 valid = lacp = 1;
1455 break;
1456 default:
1457 valid = lacp = 0;
1458 break;
1459 }
1460
1461 if (valid == 0 ||
1462 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1463 /* Invalid combination of options specified. */
1464 error = EINVAL;
1465 LAGG_XUNLOCK(sc);
1466 break; /* Return from SIOCSLAGGOPTS. */
1467 }
1468
1469 /*
1470 * Store new options into sc->sc_opts except for
1471 * FLOWIDSHIFT, RR and LACP options.
1472 */
1473 if (lacp == 0) {
1474 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1475 sc->flowid_shift = ro->ro_flowid_shift;
1476 else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1477 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1478 ro->ro_bkt == 0) {
1479 error = EINVAL;
1480 LAGG_XUNLOCK(sc);
1481 break;
1482 }
1483 sc->sc_stride = ro->ro_bkt;
1484 } else if (ro->ro_opts > 0)
1485 sc->sc_opts |= ro->ro_opts;
1486 else
1487 sc->sc_opts &= ~ro->ro_opts;
1488 } else {
1489 struct lacp_softc *lsc;
1490 struct lacp_port *lp;
1491
1492 lsc = (struct lacp_softc *)sc->sc_psc;
1493
1494 switch (ro->ro_opts) {
1495 case LAGG_OPT_LACP_TXTEST:
1496 lsc->lsc_debug.lsc_tx_test = 1;
1497 break;
1498 case -LAGG_OPT_LACP_TXTEST:
1499 lsc->lsc_debug.lsc_tx_test = 0;
1500 break;
1501 case LAGG_OPT_LACP_RXTEST:
1502 lsc->lsc_debug.lsc_rx_test = 1;
1503 break;
1504 case -LAGG_OPT_LACP_RXTEST:
1505 lsc->lsc_debug.lsc_rx_test = 0;
1506 break;
1507 case LAGG_OPT_LACP_STRICT:
1508 lsc->lsc_strict_mode = 1;
1509 break;
1510 case -LAGG_OPT_LACP_STRICT:
1511 lsc->lsc_strict_mode = 0;
1512 break;
1513 case LAGG_OPT_LACP_FAST_TIMO:
1514 LACP_LOCK(lsc);
1515 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1516 lp->lp_state |= LACP_STATE_TIMEOUT;
1517 LACP_UNLOCK(lsc);
1518 lsc->lsc_fast_timeout = 1;
1519 break;
1520 case -LAGG_OPT_LACP_FAST_TIMO:
1521 LACP_LOCK(lsc);
1522 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1523 lp->lp_state &= ~LACP_STATE_TIMEOUT;
1524 LACP_UNLOCK(lsc);
1525 lsc->lsc_fast_timeout = 0;
1526 break;
1527 }
1528 }
1529 LAGG_XUNLOCK(sc);
1530 break;
1531 case SIOCGLAGGFLAGS:
1532 rf->rf_flags = 0;
1533 LAGG_XLOCK(sc);
1534 if (sc->sc_flags & MBUF_HASHFLAG_L2)
1535 rf->rf_flags |= LAGG_F_HASHL2;
1536 if (sc->sc_flags & MBUF_HASHFLAG_L3)
1537 rf->rf_flags |= LAGG_F_HASHL3;
1538 if (sc->sc_flags & MBUF_HASHFLAG_L4)
1539 rf->rf_flags |= LAGG_F_HASHL4;
1540 LAGG_XUNLOCK(sc);
1541 break;
1542 case SIOCSLAGGHASH:
1543 error = priv_check(td, PRIV_NET_LAGG);
1544 if (error)
1545 break;
1546 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1547 error = EINVAL;
1548 break;
1549 }
1550 LAGG_XLOCK(sc);
1551 sc->sc_flags = 0;
1552 if (rf->rf_flags & LAGG_F_HASHL2)
1553 sc->sc_flags |= MBUF_HASHFLAG_L2;
1554 if (rf->rf_flags & LAGG_F_HASHL3)
1555 sc->sc_flags |= MBUF_HASHFLAG_L3;
1556 if (rf->rf_flags & LAGG_F_HASHL4)
1557 sc->sc_flags |= MBUF_HASHFLAG_L4;
1558 LAGG_XUNLOCK(sc);
1559 break;
1560 case SIOCGLAGGPORT:
1561 if (rp->rp_portname[0] == '\0' ||
1562 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1563 error = EINVAL;
1564 break;
1565 }
1566
1567 NET_EPOCH_ENTER(et);
1568 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1569 lp->lp_softc != sc) {
1570 error = ENOENT;
1571 NET_EPOCH_EXIT(et);
1572 if_rele(tpif);
1573 break;
1574 }
1575
1576 lagg_port2req(lp, rp);
1577 NET_EPOCH_EXIT(et);
1578 if_rele(tpif);
1579 break;
1580 case SIOCSLAGGPORT:
1581 error = priv_check(td, PRIV_NET_LAGG);
1582 if (error)
1583 break;
1584 if (rp->rp_portname[0] == '\0' ||
1585 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1586 error = EINVAL;
1587 break;
1588 }
1589 #ifdef INET6
1590 /*
1591 * A laggport interface should not have inet6 address
1592 * because two interfaces with a valid link-local
1593 * scope zone must not be merged in any form. This
1594 * restriction is needed to prevent violation of
1595 * link-local scope zone. Attempts to add a laggport
1596 * interface which has inet6 addresses triggers
1597 * removal of all inet6 addresses on the member
1598 * interface.
1599 */
1600 if (in6ifa_llaonifp(tpif)) {
1601 in6_ifdetach(tpif);
1602 if_printf(sc->sc_ifp,
1603 "IPv6 addresses on %s have been removed "
1604 "before adding it as a member to prevent "
1605 "IPv6 address scope violation.\n",
1606 tpif->if_xname);
1607 }
1608 #endif
1609 oldmtu = ifp->if_mtu;
1610 LAGG_XLOCK(sc);
1611 error = lagg_port_create(sc, tpif);
1612 LAGG_XUNLOCK(sc);
1613 if_rele(tpif);
1614
1615 /*
1616 * LAGG MTU may change during addition of the first port.
1617 * If it did, do network layer specific procedure.
1618 */
1619 if (ifp->if_mtu != oldmtu) {
1620 #ifdef INET6
1621 nd6_setmtu(ifp);
1622 #endif
1623 rt_updatemtu(ifp);
1624 }
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 LAGG_XLOCK(sc);
1698 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1699 if (lp->lp_ioctl != NULL)
1700 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1701 }
1702 lagg_capabilities(sc);
1703 LAGG_XUNLOCK(sc);
1704 VLAN_CAPABILITIES(ifp);
1705 error = 0;
1706 break;
1707
1708 case SIOCSIFMTU:
1709 LAGG_XLOCK(sc);
1710 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1711 if (lp->lp_ioctl != NULL)
1712 error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1713 else
1714 error = EINVAL;
1715 if (error != 0) {
1716 if_printf(ifp,
1717 "failed to change MTU to %d on port %s, "
1718 "reverting all ports to original MTU (%d)\n",
1719 ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1720 break;
1721 }
1722 }
1723 if (error == 0) {
1724 ifp->if_mtu = ifr->ifr_mtu;
1725 } else {
1726 /* set every port back to the original MTU */
1727 ifr->ifr_mtu = ifp->if_mtu;
1728 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1729 if (lp->lp_ioctl != NULL)
1730 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1731 }
1732 }
1733 LAGG_XUNLOCK(sc);
1734 break;
1735
1736 default:
1737 error = ether_ioctl(ifp, cmd, data);
1738 break;
1739 }
1740 return (error);
1741 }
1742
1743 #if defined(KERN_TLS) || defined(RATELIMIT)
1744 static inline struct lagg_snd_tag *
mst_to_lst(struct m_snd_tag * mst)1745 mst_to_lst(struct m_snd_tag *mst)
1746 {
1747
1748 return (__containerof(mst, struct lagg_snd_tag, com));
1749 }
1750
1751 /*
1752 * Look up the port used by a specific flow. This only works for lagg
1753 * protocols with deterministic port mappings (e.g. not roundrobin).
1754 * In addition protocols which use a hash to map flows to ports must
1755 * be configured to use the mbuf flowid rather than hashing packet
1756 * contents.
1757 */
1758 static struct lagg_port *
lookup_snd_tag_port(struct ifnet * ifp,uint32_t flowid,uint32_t flowtype,uint8_t numa_domain)1759 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1760 uint8_t numa_domain)
1761 {
1762 struct lagg_softc *sc;
1763 struct lagg_port *lp;
1764 struct lagg_lb *lb;
1765 uint32_t hash, p;
1766 int err;
1767
1768 sc = ifp->if_softc;
1769
1770 switch (sc->sc_proto) {
1771 case LAGG_PROTO_FAILOVER:
1772 return (lagg_link_active(sc, sc->sc_primary));
1773 case LAGG_PROTO_LOADBALANCE:
1774 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1775 flowtype == M_HASHTYPE_NONE)
1776 return (NULL);
1777 p = flowid >> sc->flowid_shift;
1778 p %= sc->sc_count;
1779 lb = (struct lagg_lb *)sc->sc_psc;
1780 lp = lb->lb_ports[p];
1781 return (lagg_link_active(sc, lp));
1782 case LAGG_PROTO_LACP:
1783 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1784 flowtype == M_HASHTYPE_NONE)
1785 return (NULL);
1786 hash = flowid >> sc->flowid_shift;
1787 return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, &err));
1788 default:
1789 return (NULL);
1790 }
1791 }
1792
1793 static int
lagg_snd_tag_alloc(struct ifnet * ifp,union if_snd_tag_alloc_params * params,struct m_snd_tag ** ppmt)1794 lagg_snd_tag_alloc(struct ifnet *ifp,
1795 union if_snd_tag_alloc_params *params,
1796 struct m_snd_tag **ppmt)
1797 {
1798 struct epoch_tracker et;
1799 struct lagg_snd_tag *lst;
1800 struct lagg_port *lp;
1801 struct ifnet *lp_ifp;
1802 int error;
1803
1804 NET_EPOCH_ENTER(et);
1805 lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1806 params->hdr.flowtype, params->hdr.numa_domain);
1807 if (lp == NULL) {
1808 NET_EPOCH_EXIT(et);
1809 return (EOPNOTSUPP);
1810 }
1811 if (lp->lp_ifp == NULL) {
1812 NET_EPOCH_EXIT(et);
1813 return (EOPNOTSUPP);
1814 }
1815 lp_ifp = lp->lp_ifp;
1816 if_ref(lp_ifp);
1817 NET_EPOCH_EXIT(et);
1818
1819 lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1820 if (lst == NULL) {
1821 if_rele(lp_ifp);
1822 return (ENOMEM);
1823 }
1824
1825 error = m_snd_tag_alloc(lp_ifp, params, &lst->tag);
1826 if_rele(lp_ifp);
1827 if (error) {
1828 free(lst, M_LAGG);
1829 return (error);
1830 }
1831
1832 m_snd_tag_init(&lst->com, ifp, lst->tag->type);
1833
1834 *ppmt = &lst->com;
1835 return (0);
1836 }
1837
1838 static struct m_snd_tag *
lagg_next_snd_tag(struct m_snd_tag * mst)1839 lagg_next_snd_tag(struct m_snd_tag *mst)
1840 {
1841 struct lagg_snd_tag *lst;
1842
1843 lst = mst_to_lst(mst);
1844 return (lst->tag);
1845 }
1846
1847 static int
lagg_snd_tag_modify(struct m_snd_tag * mst,union if_snd_tag_modify_params * params)1848 lagg_snd_tag_modify(struct m_snd_tag *mst,
1849 union if_snd_tag_modify_params *params)
1850 {
1851 struct lagg_snd_tag *lst;
1852
1853 lst = mst_to_lst(mst);
1854 return (lst->tag->ifp->if_snd_tag_modify(lst->tag, params));
1855 }
1856
1857 static int
lagg_snd_tag_query(struct m_snd_tag * mst,union if_snd_tag_query_params * params)1858 lagg_snd_tag_query(struct m_snd_tag *mst,
1859 union if_snd_tag_query_params *params)
1860 {
1861 struct lagg_snd_tag *lst;
1862
1863 lst = mst_to_lst(mst);
1864 return (lst->tag->ifp->if_snd_tag_query(lst->tag, params));
1865 }
1866
1867 static void
lagg_snd_tag_free(struct m_snd_tag * mst)1868 lagg_snd_tag_free(struct m_snd_tag *mst)
1869 {
1870 struct lagg_snd_tag *lst;
1871
1872 lst = mst_to_lst(mst);
1873 m_snd_tag_rele(lst->tag);
1874 free(lst, M_LAGG);
1875 }
1876
1877 static void
lagg_ratelimit_query(struct ifnet * ifp __unused,struct if_ratelimit_query_results * q)1878 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1879 {
1880 /*
1881 * For lagg, we have an indirect
1882 * interface. The caller needs to
1883 * get a ratelimit tag on the actual
1884 * interface the flow will go on.
1885 */
1886 q->rate_table = NULL;
1887 q->flags = RT_IS_INDIRECT;
1888 q->max_flows = 0;
1889 q->number_of_rates = 0;
1890 }
1891 #endif
1892
1893 static int
lagg_setmulti(struct lagg_port * lp)1894 lagg_setmulti(struct lagg_port *lp)
1895 {
1896 struct lagg_softc *sc = lp->lp_softc;
1897 struct ifnet *ifp = lp->lp_ifp;
1898 struct ifnet *scifp = sc->sc_ifp;
1899 struct lagg_mc *mc;
1900 struct ifmultiaddr *ifma;
1901 int error;
1902
1903 IF_ADDR_WLOCK(scifp);
1904 CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1905 if (ifma->ifma_addr->sa_family != AF_LINK)
1906 continue;
1907 mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
1908 if (mc == NULL) {
1909 IF_ADDR_WUNLOCK(scifp);
1910 return (ENOMEM);
1911 }
1912 bcopy(ifma->ifma_addr, &mc->mc_addr,
1913 ifma->ifma_addr->sa_len);
1914 mc->mc_addr.sdl_index = ifp->if_index;
1915 mc->mc_ifma = NULL;
1916 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1917 }
1918 IF_ADDR_WUNLOCK(scifp);
1919 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1920 error = if_addmulti(ifp,
1921 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1922 if (error)
1923 return (error);
1924 }
1925 return (0);
1926 }
1927
1928 static int
lagg_clrmulti(struct lagg_port * lp)1929 lagg_clrmulti(struct lagg_port *lp)
1930 {
1931 struct lagg_mc *mc;
1932
1933 LAGG_XLOCK_ASSERT(lp->lp_softc);
1934 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1935 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1936 if (mc->mc_ifma && lp->lp_detaching == 0)
1937 if_delmulti_ifma(mc->mc_ifma);
1938 free(mc, M_LAGG);
1939 }
1940 return (0);
1941 }
1942
1943 static int
lagg_setcaps(struct lagg_port * lp,int cap)1944 lagg_setcaps(struct lagg_port *lp, int cap)
1945 {
1946 struct ifreq ifr;
1947
1948 if (lp->lp_ifp->if_capenable == cap)
1949 return (0);
1950 if (lp->lp_ioctl == NULL)
1951 return (ENXIO);
1952 ifr.ifr_reqcap = cap;
1953 return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1954 }
1955
1956 /* Handle a ref counted flag that should be set on the lagg port as well */
1957 static int
lagg_setflag(struct lagg_port * lp,int flag,int status,int (* func)(struct ifnet *,int))1958 lagg_setflag(struct lagg_port *lp, int flag, int status,
1959 int (*func)(struct ifnet *, int))
1960 {
1961 struct lagg_softc *sc = lp->lp_softc;
1962 struct ifnet *scifp = sc->sc_ifp;
1963 struct ifnet *ifp = lp->lp_ifp;
1964 int error;
1965
1966 LAGG_XLOCK_ASSERT(sc);
1967
1968 status = status ? (scifp->if_flags & flag) : 0;
1969 /* Now "status" contains the flag value or 0 */
1970
1971 /*
1972 * See if recorded ports status is different from what
1973 * we want it to be. If it is, flip it. We record ports
1974 * status in lp_ifflags so that we won't clear ports flag
1975 * we haven't set. In fact, we don't clear or set ports
1976 * flags directly, but get or release references to them.
1977 * That's why we can be sure that recorded flags still are
1978 * in accord with actual ports flags.
1979 */
1980 if (status != (lp->lp_ifflags & flag)) {
1981 error = (*func)(ifp, status);
1982 if (error)
1983 return (error);
1984 lp->lp_ifflags &= ~flag;
1985 lp->lp_ifflags |= status;
1986 }
1987 return (0);
1988 }
1989
1990 /*
1991 * Handle IFF_* flags that require certain changes on the lagg port
1992 * if "status" is true, update ports flags respective to the lagg
1993 * if "status" is false, forcedly clear the flags set on port.
1994 */
1995 static int
lagg_setflags(struct lagg_port * lp,int status)1996 lagg_setflags(struct lagg_port *lp, int status)
1997 {
1998 int error, i;
1999
2000 for (i = 0; lagg_pflags[i].flag; i++) {
2001 error = lagg_setflag(lp, lagg_pflags[i].flag,
2002 status, lagg_pflags[i].func);
2003 if (error)
2004 return (error);
2005 }
2006 return (0);
2007 }
2008
2009 static int
lagg_transmit_ethernet(struct ifnet * ifp,struct mbuf * m)2010 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2011 {
2012 struct epoch_tracker et;
2013 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2014 int error;
2015
2016 #if defined(KERN_TLS) || defined(RATELIMIT)
2017 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2018 MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2019 #endif
2020 NET_EPOCH_ENTER(et);
2021 /* We need a Tx algorithm and at least one port */
2022 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2023 NET_EPOCH_EXIT(et);
2024 m_freem(m);
2025 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2026 return (ENXIO);
2027 }
2028
2029 ETHER_BPF_MTAP(ifp, m);
2030
2031 error = lagg_proto_start(sc, m);
2032 NET_EPOCH_EXIT(et);
2033 return (error);
2034 }
2035
2036 static int
lagg_transmit_infiniband(struct ifnet * ifp,struct mbuf * m)2037 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2038 {
2039 struct epoch_tracker et;
2040 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2041 int error;
2042
2043 #if defined(KERN_TLS) || defined(RATELIMIT)
2044 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2045 MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2046 #endif
2047 NET_EPOCH_ENTER(et);
2048 /* We need a Tx algorithm and at least one port */
2049 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2050 NET_EPOCH_EXIT(et);
2051 m_freem(m);
2052 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2053 return (ENXIO);
2054 }
2055
2056 INFINIBAND_BPF_MTAP(ifp, m);
2057
2058 error = lagg_proto_start(sc, m);
2059 NET_EPOCH_EXIT(et);
2060 return (error);
2061 }
2062
2063 /*
2064 * The ifp->if_qflush entry point for lagg(4) is no-op.
2065 */
2066 static void
lagg_qflush(struct ifnet * ifp __unused)2067 lagg_qflush(struct ifnet *ifp __unused)
2068 {
2069 }
2070
2071 static struct mbuf *
lagg_input_ethernet(struct ifnet * ifp,struct mbuf * m)2072 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2073 {
2074 struct epoch_tracker et;
2075 struct lagg_port *lp = ifp->if_lagg;
2076 struct lagg_softc *sc = lp->lp_softc;
2077 struct ifnet *scifp = sc->sc_ifp;
2078
2079 NET_EPOCH_ENTER(et);
2080 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2081 lp->lp_detaching != 0 ||
2082 sc->sc_proto == LAGG_PROTO_NONE) {
2083 NET_EPOCH_EXIT(et);
2084 m_freem(m);
2085 return (NULL);
2086 }
2087
2088 ETHER_BPF_MTAP(scifp, m);
2089
2090 m = lagg_proto_input(sc, lp, m);
2091 if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2092 m_freem(m);
2093 m = NULL;
2094 }
2095
2096 NET_EPOCH_EXIT(et);
2097 return (m);
2098 }
2099
2100 static struct mbuf *
lagg_input_infiniband(struct ifnet * ifp,struct mbuf * m)2101 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2102 {
2103 struct epoch_tracker et;
2104 struct lagg_port *lp = ifp->if_lagg;
2105 struct lagg_softc *sc = lp->lp_softc;
2106 struct ifnet *scifp = sc->sc_ifp;
2107
2108 NET_EPOCH_ENTER(et);
2109 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2110 lp->lp_detaching != 0 ||
2111 sc->sc_proto == LAGG_PROTO_NONE) {
2112 NET_EPOCH_EXIT(et);
2113 m_freem(m);
2114 return (NULL);
2115 }
2116
2117 INFINIBAND_BPF_MTAP(scifp, m);
2118
2119 m = lagg_proto_input(sc, lp, m);
2120 if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2121 m_freem(m);
2122 m = NULL;
2123 }
2124
2125 NET_EPOCH_EXIT(et);
2126 return (m);
2127 }
2128
2129 static int
lagg_media_change(struct ifnet * ifp)2130 lagg_media_change(struct ifnet *ifp)
2131 {
2132 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2133
2134 if (sc->sc_ifflags & IFF_DEBUG)
2135 printf("%s\n", __func__);
2136
2137 /* Ignore */
2138 return (0);
2139 }
2140
2141 static void
lagg_media_status(struct ifnet * ifp,struct ifmediareq * imr)2142 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2143 {
2144 struct epoch_tracker et;
2145 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2146 struct lagg_port *lp;
2147
2148 imr->ifm_status = IFM_AVALID;
2149 imr->ifm_active = IFM_ETHER | IFM_AUTO;
2150
2151 NET_EPOCH_ENTER(et);
2152 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2153 if (LAGG_PORTACTIVE(lp))
2154 imr->ifm_status |= IFM_ACTIVE;
2155 }
2156 NET_EPOCH_EXIT(et);
2157 }
2158
2159 static void
lagg_linkstate(struct lagg_softc * sc)2160 lagg_linkstate(struct lagg_softc *sc)
2161 {
2162 struct epoch_tracker et;
2163 struct lagg_port *lp;
2164 int new_link = LINK_STATE_DOWN;
2165 uint64_t speed;
2166
2167 LAGG_XLOCK_ASSERT(sc);
2168
2169 /* LACP handles link state itself */
2170 if (sc->sc_proto == LAGG_PROTO_LACP)
2171 return;
2172
2173 /* Our link is considered up if at least one of our ports is active */
2174 NET_EPOCH_ENTER(et);
2175 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2176 if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2177 new_link = LINK_STATE_UP;
2178 break;
2179 }
2180 }
2181 NET_EPOCH_EXIT(et);
2182 if_link_state_change(sc->sc_ifp, new_link);
2183
2184 /* Update if_baudrate to reflect the max possible speed */
2185 switch (sc->sc_proto) {
2186 case LAGG_PROTO_FAILOVER:
2187 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2188 sc->sc_primary->lp_ifp->if_baudrate : 0;
2189 break;
2190 case LAGG_PROTO_ROUNDROBIN:
2191 case LAGG_PROTO_LOADBALANCE:
2192 case LAGG_PROTO_BROADCAST:
2193 speed = 0;
2194 NET_EPOCH_ENTER(et);
2195 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2196 speed += lp->lp_ifp->if_baudrate;
2197 NET_EPOCH_EXIT(et);
2198 sc->sc_ifp->if_baudrate = speed;
2199 break;
2200 case LAGG_PROTO_LACP:
2201 /* LACP updates if_baudrate itself */
2202 break;
2203 }
2204 }
2205
2206 static void
lagg_port_state(struct ifnet * ifp,int state)2207 lagg_port_state(struct ifnet *ifp, int state)
2208 {
2209 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2210 struct lagg_softc *sc = NULL;
2211
2212 if (lp != NULL)
2213 sc = lp->lp_softc;
2214 if (sc == NULL)
2215 return;
2216
2217 LAGG_XLOCK(sc);
2218 lagg_linkstate(sc);
2219 lagg_proto_linkstate(sc, lp);
2220 LAGG_XUNLOCK(sc);
2221 }
2222
2223 struct lagg_port *
lagg_link_active(struct lagg_softc * sc,struct lagg_port * lp)2224 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2225 {
2226 struct lagg_port *lp_next, *rval = NULL;
2227
2228 /*
2229 * Search a port which reports an active link state.
2230 */
2231
2232 #ifdef INVARIANTS
2233 /*
2234 * This is called with either in the network epoch
2235 * or with LAGG_XLOCK(sc) held.
2236 */
2237 if (!in_epoch(net_epoch_preempt))
2238 LAGG_XLOCK_ASSERT(sc);
2239 #endif
2240
2241 if (lp == NULL)
2242 goto search;
2243 if (LAGG_PORTACTIVE(lp)) {
2244 rval = lp;
2245 goto found;
2246 }
2247 if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2248 LAGG_PORTACTIVE(lp_next)) {
2249 rval = lp_next;
2250 goto found;
2251 }
2252
2253 search:
2254 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2255 if (LAGG_PORTACTIVE(lp_next)) {
2256 return (lp_next);
2257 }
2258 }
2259 found:
2260 return (rval);
2261 }
2262
2263 int
lagg_enqueue(struct ifnet * ifp,struct mbuf * m)2264 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2265 {
2266
2267 #if defined(KERN_TLS) || defined(RATELIMIT)
2268 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2269 struct lagg_snd_tag *lst;
2270 struct m_snd_tag *mst;
2271
2272 mst = m->m_pkthdr.snd_tag;
2273 lst = mst_to_lst(mst);
2274 if (lst->tag->ifp != ifp) {
2275 m_freem(m);
2276 return (EAGAIN);
2277 }
2278 m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2279 m_snd_tag_rele(mst);
2280 }
2281 #endif
2282 return (ifp->if_transmit)(ifp, m);
2283 }
2284
2285 /*
2286 * Simple round robin aggregation
2287 */
2288 static void
lagg_rr_attach(struct lagg_softc * sc)2289 lagg_rr_attach(struct lagg_softc *sc)
2290 {
2291 sc->sc_seq = 0;
2292 sc->sc_stride = 1;
2293 }
2294
2295 static int
lagg_rr_start(struct lagg_softc * sc,struct mbuf * m)2296 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2297 {
2298 struct lagg_port *lp;
2299 uint32_t p;
2300
2301 p = atomic_fetchadd_32(&sc->sc_seq, 1);
2302 p /= sc->sc_stride;
2303 p %= sc->sc_count;
2304 lp = CK_SLIST_FIRST(&sc->sc_ports);
2305
2306 while (p--)
2307 lp = CK_SLIST_NEXT(lp, lp_entries);
2308
2309 /*
2310 * Check the port's link state. This will return the next active
2311 * port if the link is down or the port is NULL.
2312 */
2313 if ((lp = lagg_link_active(sc, lp)) == NULL) {
2314 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2315 m_freem(m);
2316 return (ENETDOWN);
2317 }
2318
2319 /* Send mbuf */
2320 return (lagg_enqueue(lp->lp_ifp, m));
2321 }
2322
2323 static struct mbuf *
lagg_rr_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)2324 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2325 {
2326 struct ifnet *ifp = sc->sc_ifp;
2327
2328 /* Just pass in the packet to our lagg device */
2329 m->m_pkthdr.rcvif = ifp;
2330
2331 return (m);
2332 }
2333
2334 /*
2335 * Broadcast mode
2336 */
2337 static int
lagg_bcast_start(struct lagg_softc * sc,struct mbuf * m)2338 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2339 {
2340 int active_ports = 0;
2341 int errors = 0;
2342 int ret;
2343 struct lagg_port *lp, *last = NULL;
2344 struct mbuf *m0;
2345
2346 NET_EPOCH_ASSERT();
2347 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2348 if (!LAGG_PORTACTIVE(lp))
2349 continue;
2350
2351 active_ports++;
2352
2353 if (last != NULL) {
2354 m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2355 if (m0 == NULL) {
2356 ret = ENOBUFS;
2357 errors++;
2358 break;
2359 }
2360 lagg_enqueue(last->lp_ifp, m0);
2361 }
2362 last = lp;
2363 }
2364
2365 if (last == NULL) {
2366 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2367 m_freem(m);
2368 return (ENOENT);
2369 }
2370 if ((last = lagg_link_active(sc, last)) == NULL) {
2371 errors++;
2372 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2373 m_freem(m);
2374 return (ENETDOWN);
2375 }
2376
2377 ret = lagg_enqueue(last->lp_ifp, m);
2378 if (errors != 0)
2379 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2380
2381 return (ret);
2382 }
2383
2384 static struct mbuf*
lagg_bcast_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)2385 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2386 {
2387 struct ifnet *ifp = sc->sc_ifp;
2388
2389 /* Just pass in the packet to our lagg device */
2390 m->m_pkthdr.rcvif = ifp;
2391 return (m);
2392 }
2393
2394 /*
2395 * Active failover
2396 */
2397 static int
lagg_fail_start(struct lagg_softc * sc,struct mbuf * m)2398 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2399 {
2400 struct lagg_port *lp;
2401
2402 /* Use the master port if active or the next available port */
2403 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2404 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2405 m_freem(m);
2406 return (ENETDOWN);
2407 }
2408
2409 /* Send mbuf */
2410 return (lagg_enqueue(lp->lp_ifp, m));
2411 }
2412
2413 static struct mbuf *
lagg_fail_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)2414 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2415 {
2416 struct ifnet *ifp = sc->sc_ifp;
2417 struct lagg_port *tmp_tp;
2418
2419 if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2420 m->m_pkthdr.rcvif = ifp;
2421 return (m);
2422 }
2423
2424 if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2425 tmp_tp = lagg_link_active(sc, sc->sc_primary);
2426 /*
2427 * If tmp_tp is null, we've received a packet when all
2428 * our links are down. Weird, but process it anyways.
2429 */
2430 if ((tmp_tp == NULL || tmp_tp == lp)) {
2431 m->m_pkthdr.rcvif = ifp;
2432 return (m);
2433 }
2434 }
2435
2436 m_freem(m);
2437 return (NULL);
2438 }
2439
2440 /*
2441 * Loadbalancing
2442 */
2443 static void
lagg_lb_attach(struct lagg_softc * sc)2444 lagg_lb_attach(struct lagg_softc *sc)
2445 {
2446 struct lagg_port *lp;
2447 struct lagg_lb *lb;
2448
2449 LAGG_XLOCK_ASSERT(sc);
2450 lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2451 lb->lb_key = m_ether_tcpip_hash_init();
2452 sc->sc_psc = lb;
2453
2454 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2455 lagg_lb_port_create(lp);
2456 }
2457
2458 static void
lagg_lb_detach(struct lagg_softc * sc)2459 lagg_lb_detach(struct lagg_softc *sc)
2460 {
2461 struct lagg_lb *lb;
2462
2463 lb = (struct lagg_lb *)sc->sc_psc;
2464 if (lb != NULL)
2465 free(lb, M_LAGG);
2466 }
2467
2468 static int
lagg_lb_porttable(struct lagg_softc * sc,struct lagg_port * lp)2469 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2470 {
2471 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2472 struct lagg_port *lp_next;
2473 int i = 0, rv;
2474
2475 rv = 0;
2476 bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2477 LAGG_XLOCK_ASSERT(sc);
2478 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2479 if (lp_next == lp)
2480 continue;
2481 if (i >= LAGG_MAX_PORTS) {
2482 rv = EINVAL;
2483 break;
2484 }
2485 if (sc->sc_ifflags & IFF_DEBUG)
2486 printf("%s: port %s at index %d\n",
2487 sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2488 lb->lb_ports[i++] = lp_next;
2489 }
2490
2491 return (rv);
2492 }
2493
2494 static int
lagg_lb_port_create(struct lagg_port * lp)2495 lagg_lb_port_create(struct lagg_port *lp)
2496 {
2497 struct lagg_softc *sc = lp->lp_softc;
2498 return (lagg_lb_porttable(sc, NULL));
2499 }
2500
2501 static void
lagg_lb_port_destroy(struct lagg_port * lp)2502 lagg_lb_port_destroy(struct lagg_port *lp)
2503 {
2504 struct lagg_softc *sc = lp->lp_softc;
2505 lagg_lb_porttable(sc, lp);
2506 }
2507
2508 static int
lagg_lb_start(struct lagg_softc * sc,struct mbuf * m)2509 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2510 {
2511 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2512 struct lagg_port *lp = NULL;
2513 uint32_t p = 0;
2514
2515 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2516 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2517 p = m->m_pkthdr.flowid >> sc->flowid_shift;
2518 else
2519 p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2520 p %= sc->sc_count;
2521 lp = lb->lb_ports[p];
2522
2523 /*
2524 * Check the port's link state. This will return the next active
2525 * port if the link is down or the port is NULL.
2526 */
2527 if ((lp = lagg_link_active(sc, lp)) == NULL) {
2528 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2529 m_freem(m);
2530 return (ENETDOWN);
2531 }
2532
2533 /* Send mbuf */
2534 return (lagg_enqueue(lp->lp_ifp, m));
2535 }
2536
2537 static struct mbuf *
lagg_lb_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)2538 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2539 {
2540 struct ifnet *ifp = sc->sc_ifp;
2541
2542 /* Just pass in the packet to our lagg device */
2543 m->m_pkthdr.rcvif = ifp;
2544
2545 return (m);
2546 }
2547
2548 /*
2549 * 802.3ad LACP
2550 */
2551 static void
lagg_lacp_attach(struct lagg_softc * sc)2552 lagg_lacp_attach(struct lagg_softc *sc)
2553 {
2554 struct lagg_port *lp;
2555
2556 lacp_attach(sc);
2557 LAGG_XLOCK_ASSERT(sc);
2558 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2559 lacp_port_create(lp);
2560 }
2561
2562 static void
lagg_lacp_detach(struct lagg_softc * sc)2563 lagg_lacp_detach(struct lagg_softc *sc)
2564 {
2565 struct lagg_port *lp;
2566 void *psc;
2567
2568 LAGG_XLOCK_ASSERT(sc);
2569 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2570 lacp_port_destroy(lp);
2571
2572 psc = sc->sc_psc;
2573 sc->sc_psc = NULL;
2574 lacp_detach(psc);
2575 }
2576
2577 static void
lagg_lacp_lladdr(struct lagg_softc * sc)2578 lagg_lacp_lladdr(struct lagg_softc *sc)
2579 {
2580 struct lagg_port *lp;
2581
2582 LAGG_SXLOCK_ASSERT(sc);
2583
2584 /* purge all the lacp ports */
2585 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2586 lacp_port_destroy(lp);
2587
2588 /* add them back in */
2589 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2590 lacp_port_create(lp);
2591 }
2592
2593 static int
lagg_lacp_start(struct lagg_softc * sc,struct mbuf * m)2594 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2595 {
2596 struct lagg_port *lp;
2597 int err;
2598
2599 lp = lacp_select_tx_port(sc, m, &err);
2600 if (lp == NULL) {
2601 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2602 m_freem(m);
2603 return (err);
2604 }
2605
2606 /* Send mbuf */
2607 return (lagg_enqueue(lp->lp_ifp, m));
2608 }
2609
2610 static struct mbuf *
lagg_lacp_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)2611 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2612 {
2613 struct ifnet *ifp = sc->sc_ifp;
2614 struct ether_header *eh;
2615 u_short etype;
2616
2617 eh = mtod(m, struct ether_header *);
2618 etype = ntohs(eh->ether_type);
2619
2620 /* Tap off LACP control messages */
2621 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2622 m = lacp_input(lp, m);
2623 if (m == NULL)
2624 return (NULL);
2625 }
2626
2627 /*
2628 * If the port is not collecting or not in the active aggregator then
2629 * free and return.
2630 */
2631 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2632 m_freem(m);
2633 return (NULL);
2634 }
2635
2636 m->m_pkthdr.rcvif = ifp;
2637 return (m);
2638 }
2639