FreeBSD/Linux Kernel Cross Reference
sys/net/if_lagg.c
1 /* $OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $ */
2
3 /*
4 * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org>
5 * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org>
6 * Copyright (c) 2014, 2016 Marcelo Araujo <araujo@FreeBSD.org>
7 *
8 * Permission to use, copy, modify, and distribute this software for any
9 * purpose with or without fee is hereby granted, provided that the above
10 * copyright notice and this permission notice appear in all copies.
11 *
12 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19 */
20
21 #include <sys/cdefs.h>
22 __FBSDID("$FreeBSD$");
23
24 #include "opt_inet.h"
25 #include "opt_inet6.h"
26 #include "opt_kern_tls.h"
27 #include "opt_ratelimit.h"
28
29 #include <sys/param.h>
30 #include <sys/kernel.h>
31 #include <sys/malloc.h>
32 #include <sys/mbuf.h>
33 #include <sys/queue.h>
34 #include <sys/socket.h>
35 #include <sys/sockio.h>
36 #include <sys/sysctl.h>
37 #include <sys/module.h>
38 #include <sys/priv.h>
39 #include <sys/systm.h>
40 #include <sys/proc.h>
41 #include <sys/lock.h>
42 #include <sys/rmlock.h>
43 #include <sys/sx.h>
44 #include <sys/taskqueue.h>
45 #include <sys/eventhandler.h>
46
47 #include <net/ethernet.h>
48 #include <net/if.h>
49 #include <net/if_clone.h>
50 #include <net/if_arp.h>
51 #include <net/if_dl.h>
52 #include <net/if_media.h>
53 #include <net/if_types.h>
54 #include <net/if_var.h>
55 #include <net/if_private.h>
56 #include <net/bpf.h>
57 #include <net/route.h>
58 #include <net/vnet.h>
59 #include <net/infiniband.h>
60
61 #if defined(INET) || defined(INET6)
62 #include <netinet/in.h>
63 #include <netinet/ip.h>
64 #endif
65 #ifdef INET
66 #include <netinet/in_systm.h>
67 #include <netinet/if_ether.h>
68 #endif
69
70 #ifdef INET6
71 #include <netinet/ip6.h>
72 #include <netinet6/in6_var.h>
73 #include <netinet6/in6_ifattach.h>
74 #endif
75
76 #include <net/if_vlan_var.h>
77 #include <net/if_lagg.h>
78 #include <net/ieee8023ad_lacp.h>
79
80 #ifdef INET6
81 /*
82 * XXX: declare here to avoid to include many inet6 related files..
83 * should be more generalized?
84 */
85 extern void nd6_setmtu(struct ifnet *);
86 #endif
87
88 #ifdef DEV_NETMAP
89 MODULE_DEPEND(if_lagg, netmap, 1, 1, 1);
90 #endif
91
92 #define LAGG_SX_INIT(_sc) sx_init(&(_sc)->sc_sx, "if_lagg sx")
93 #define LAGG_SX_DESTROY(_sc) sx_destroy(&(_sc)->sc_sx)
94 #define LAGG_XLOCK(_sc) sx_xlock(&(_sc)->sc_sx)
95 #define LAGG_XUNLOCK(_sc) sx_xunlock(&(_sc)->sc_sx)
96 #define LAGG_SXLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_LOCKED)
97 #define LAGG_XLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
98
99 /* Special flags we should propagate to the lagg ports. */
100 static struct {
101 int flag;
102 int (*func)(struct ifnet *, int);
103 } lagg_pflags[] = {
104 {IFF_PROMISC, ifpromisc},
105 {IFF_ALLMULTI, if_allmulti},
106 {0, NULL}
107 };
108
109 struct lagg_snd_tag {
110 struct m_snd_tag com;
111 struct m_snd_tag *tag;
112 };
113
114 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
115 #define V_lagg_list VNET(lagg_list)
116 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
117 #define V_lagg_list_mtx VNET(lagg_list_mtx)
118 #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \
119 "if_lagg list", NULL, MTX_DEF)
120 #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx)
121 #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx)
122 #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx)
123 eventhandler_tag lagg_detach_cookie = NULL;
124
125 static int lagg_clone_create(struct if_clone *, char *, size_t,
126 struct ifc_data *, struct ifnet **);
127 static int lagg_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
128 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
129 #define V_lagg_cloner VNET(lagg_cloner)
130 static const char laggname[] = "lagg";
131 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
132
133 static void lagg_capabilities(struct lagg_softc *);
134 static int lagg_port_create(struct lagg_softc *, struct ifnet *);
135 static int lagg_port_destroy(struct lagg_port *, int);
136 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *);
137 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *);
138 static void lagg_linkstate(struct lagg_softc *);
139 static void lagg_port_state(struct ifnet *, int);
140 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
141 static int lagg_port_output(struct ifnet *, struct mbuf *,
142 const struct sockaddr *, struct route *);
143 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *);
144 #ifdef LAGG_PORT_STACKING
145 static int lagg_port_checkstacking(struct lagg_softc *);
146 #endif
147 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *);
148 static void lagg_init(void *);
149 static void lagg_stop(struct lagg_softc *);
150 static int lagg_ioctl(struct ifnet *, u_long, caddr_t);
151 #if defined(KERN_TLS) || defined(RATELIMIT)
152 static int lagg_snd_tag_alloc(struct ifnet *,
153 union if_snd_tag_alloc_params *,
154 struct m_snd_tag **);
155 static int lagg_snd_tag_modify(struct m_snd_tag *,
156 union if_snd_tag_modify_params *);
157 static int lagg_snd_tag_query(struct m_snd_tag *,
158 union if_snd_tag_query_params *);
159 static void lagg_snd_tag_free(struct m_snd_tag *);
160 static struct m_snd_tag *lagg_next_snd_tag(struct m_snd_tag *);
161 static void lagg_ratelimit_query(struct ifnet *,
162 struct if_ratelimit_query_results *);
163 #endif
164 static int lagg_setmulti(struct lagg_port *);
165 static int lagg_clrmulti(struct lagg_port *);
166 static void lagg_setcaps(struct lagg_port *, int cap, int cap2);
167 static int lagg_setflag(struct lagg_port *, int, int,
168 int (*func)(struct ifnet *, int));
169 static int lagg_setflags(struct lagg_port *, int status);
170 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
171 static int lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
172 static int lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
173 static void lagg_qflush(struct ifnet *);
174 static int lagg_media_change(struct ifnet *);
175 static void lagg_media_status(struct ifnet *, struct ifmediareq *);
176 static struct lagg_port *lagg_link_active(struct lagg_softc *,
177 struct lagg_port *);
178
179 /* Simple round robin */
180 static void lagg_rr_attach(struct lagg_softc *);
181 static int lagg_rr_start(struct lagg_softc *, struct mbuf *);
182 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
183 struct mbuf *);
184
185 /* Active failover */
186 static int lagg_fail_start(struct lagg_softc *, struct mbuf *);
187 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
188 struct mbuf *);
189
190 /* Loadbalancing */
191 static void lagg_lb_attach(struct lagg_softc *);
192 static void lagg_lb_detach(struct lagg_softc *);
193 static int lagg_lb_port_create(struct lagg_port *);
194 static void lagg_lb_port_destroy(struct lagg_port *);
195 static int lagg_lb_start(struct lagg_softc *, struct mbuf *);
196 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
197 struct mbuf *);
198 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
199
200 /* Broadcast */
201 static int lagg_bcast_start(struct lagg_softc *, struct mbuf *);
202 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
203 struct mbuf *);
204
205 /* 802.3ad LACP */
206 static void lagg_lacp_attach(struct lagg_softc *);
207 static void lagg_lacp_detach(struct lagg_softc *);
208 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *);
209 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
210 struct mbuf *);
211 static void lagg_lacp_lladdr(struct lagg_softc *);
212
213 /* lagg protocol table */
214 static const struct lagg_proto {
215 lagg_proto pr_num;
216 void (*pr_attach)(struct lagg_softc *);
217 void (*pr_detach)(struct lagg_softc *);
218 int (*pr_start)(struct lagg_softc *, struct mbuf *);
219 struct mbuf * (*pr_input)(struct lagg_softc *, struct lagg_port *,
220 struct mbuf *);
221 int (*pr_addport)(struct lagg_port *);
222 void (*pr_delport)(struct lagg_port *);
223 void (*pr_linkstate)(struct lagg_port *);
224 void (*pr_init)(struct lagg_softc *);
225 void (*pr_stop)(struct lagg_softc *);
226 void (*pr_lladdr)(struct lagg_softc *);
227 void (*pr_request)(struct lagg_softc *, void *);
228 void (*pr_portreq)(struct lagg_port *, void *);
229 } lagg_protos[] = {
230 {
231 .pr_num = LAGG_PROTO_NONE
232 },
233 {
234 .pr_num = LAGG_PROTO_ROUNDROBIN,
235 .pr_attach = lagg_rr_attach,
236 .pr_start = lagg_rr_start,
237 .pr_input = lagg_rr_input,
238 },
239 {
240 .pr_num = LAGG_PROTO_FAILOVER,
241 .pr_start = lagg_fail_start,
242 .pr_input = lagg_fail_input,
243 },
244 {
245 .pr_num = LAGG_PROTO_LOADBALANCE,
246 .pr_attach = lagg_lb_attach,
247 .pr_detach = lagg_lb_detach,
248 .pr_start = lagg_lb_start,
249 .pr_input = lagg_lb_input,
250 .pr_addport = lagg_lb_port_create,
251 .pr_delport = lagg_lb_port_destroy,
252 },
253 {
254 .pr_num = LAGG_PROTO_LACP,
255 .pr_attach = lagg_lacp_attach,
256 .pr_detach = lagg_lacp_detach,
257 .pr_start = lagg_lacp_start,
258 .pr_input = lagg_lacp_input,
259 .pr_addport = lacp_port_create,
260 .pr_delport = lacp_port_destroy,
261 .pr_linkstate = lacp_linkstate,
262 .pr_init = lacp_init,
263 .pr_stop = lacp_stop,
264 .pr_lladdr = lagg_lacp_lladdr,
265 .pr_request = lacp_req,
266 .pr_portreq = lacp_portreq,
267 },
268 {
269 .pr_num = LAGG_PROTO_BROADCAST,
270 .pr_start = lagg_bcast_start,
271 .pr_input = lagg_bcast_input,
272 },
273 };
274
275 SYSCTL_DECL(_net_link);
276 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
277 "Link Aggregation");
278
279 /* Allow input on any failover links */
280 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
281 #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all)
282 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
283 &VNET_NAME(lagg_failover_rx_all), 0,
284 "Accept input from any interface in a failover lagg");
285
286 /* Default value for using flowid */
287 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
288 #define V_def_use_flowid VNET(def_use_flowid)
289 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
290 &VNET_NAME(def_use_flowid), 0,
291 "Default setting for using flow id for load sharing");
292
293 /* Default value for using numa */
294 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
295 #define V_def_use_numa VNET(def_use_numa)
296 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN,
297 &VNET_NAME(def_use_numa), 0,
298 "Use numa to steer flows");
299
300 /* Default value for flowid shift */
301 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
302 #define V_def_flowid_shift VNET(def_flowid_shift)
303 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
304 &VNET_NAME(def_flowid_shift), 0,
305 "Default setting for flowid shift for load sharing");
306
307 static void
308 vnet_lagg_init(const void *unused __unused)
309 {
310
311 LAGG_LIST_LOCK_INIT();
312 SLIST_INIT(&V_lagg_list);
313 struct if_clone_addreq req = {
314 .create_f = lagg_clone_create,
315 .destroy_f = lagg_clone_destroy,
316 .flags = IFC_F_AUTOUNIT,
317 };
318 V_lagg_cloner = ifc_attach_cloner(laggname, &req);
319 }
320 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
321 vnet_lagg_init, NULL);
322
323 static void
324 vnet_lagg_uninit(const void *unused __unused)
325 {
326
327 ifc_detach_cloner(V_lagg_cloner);
328 LAGG_LIST_LOCK_DESTROY();
329 }
330 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
331 vnet_lagg_uninit, NULL);
332
333 static int
334 lagg_modevent(module_t mod, int type, void *data)
335 {
336
337 switch (type) {
338 case MOD_LOAD:
339 lagg_input_ethernet_p = lagg_input_ethernet;
340 lagg_input_infiniband_p = lagg_input_infiniband;
341 lagg_linkstate_p = lagg_port_state;
342 lagg_detach_cookie = EVENTHANDLER_REGISTER(
343 ifnet_departure_event, lagg_port_ifdetach, NULL,
344 EVENTHANDLER_PRI_ANY);
345 break;
346 case MOD_UNLOAD:
347 EVENTHANDLER_DEREGISTER(ifnet_departure_event,
348 lagg_detach_cookie);
349 lagg_input_ethernet_p = NULL;
350 lagg_input_infiniband_p = NULL;
351 lagg_linkstate_p = NULL;
352 break;
353 default:
354 return (EOPNOTSUPP);
355 }
356 return (0);
357 }
358
359 static moduledata_t lagg_mod = {
360 "if_lagg",
361 lagg_modevent,
362 0
363 };
364
365 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
366 MODULE_VERSION(if_lagg, 1);
367 MODULE_DEPEND(if_lagg, if_infiniband, 1, 1, 1);
368
369 static void
370 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
371 {
372
373 LAGG_XLOCK_ASSERT(sc);
374 KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
375 __func__, sc));
376
377 if (sc->sc_ifflags & IFF_DEBUG)
378 if_printf(sc->sc_ifp, "using proto %u\n", pr);
379
380 if (lagg_protos[pr].pr_attach != NULL)
381 lagg_protos[pr].pr_attach(sc);
382 sc->sc_proto = pr;
383 }
384
385 static void
386 lagg_proto_detach(struct lagg_softc *sc)
387 {
388 lagg_proto pr;
389
390 LAGG_XLOCK_ASSERT(sc);
391 pr = sc->sc_proto;
392 sc->sc_proto = LAGG_PROTO_NONE;
393
394 if (lagg_protos[pr].pr_detach != NULL)
395 lagg_protos[pr].pr_detach(sc);
396 }
397
398 static int
399 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
400 {
401
402 return (lagg_protos[sc->sc_proto].pr_start(sc, m));
403 }
404
405 static struct mbuf *
406 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
407 {
408
409 return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
410 }
411
412 static int
413 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
414 {
415
416 if (lagg_protos[sc->sc_proto].pr_addport == NULL)
417 return (0);
418 else
419 return (lagg_protos[sc->sc_proto].pr_addport(lp));
420 }
421
422 static void
423 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
424 {
425
426 if (lagg_protos[sc->sc_proto].pr_delport != NULL)
427 lagg_protos[sc->sc_proto].pr_delport(lp);
428 }
429
430 static void
431 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
432 {
433
434 if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
435 lagg_protos[sc->sc_proto].pr_linkstate(lp);
436 }
437
438 static void
439 lagg_proto_init(struct lagg_softc *sc)
440 {
441
442 if (lagg_protos[sc->sc_proto].pr_init != NULL)
443 lagg_protos[sc->sc_proto].pr_init(sc);
444 }
445
446 static void
447 lagg_proto_stop(struct lagg_softc *sc)
448 {
449
450 if (lagg_protos[sc->sc_proto].pr_stop != NULL)
451 lagg_protos[sc->sc_proto].pr_stop(sc);
452 }
453
454 static void
455 lagg_proto_lladdr(struct lagg_softc *sc)
456 {
457
458 if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
459 lagg_protos[sc->sc_proto].pr_lladdr(sc);
460 }
461
462 static void
463 lagg_proto_request(struct lagg_softc *sc, void *v)
464 {
465
466 if (lagg_protos[sc->sc_proto].pr_request != NULL)
467 lagg_protos[sc->sc_proto].pr_request(sc, v);
468 }
469
470 static void
471 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
472 {
473
474 if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
475 lagg_protos[sc->sc_proto].pr_portreq(lp, v);
476 }
477
478 /*
479 * This routine is run via an vlan
480 * config EVENT
481 */
482 static void
483 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
484 {
485 struct lagg_softc *sc = ifp->if_softc;
486 struct lagg_port *lp;
487
488 if (ifp->if_softc != arg) /* Not our event */
489 return;
490
491 LAGG_XLOCK(sc);
492 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
493 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
494 LAGG_XUNLOCK(sc);
495 }
496
497 /*
498 * This routine is run via an vlan
499 * unconfig EVENT
500 */
501 static void
502 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
503 {
504 struct lagg_softc *sc = ifp->if_softc;
505 struct lagg_port *lp;
506
507 if (ifp->if_softc != arg) /* Not our event */
508 return;
509
510 LAGG_XLOCK(sc);
511 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
512 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
513 LAGG_XUNLOCK(sc);
514 }
515
516 static int
517 lagg_clone_create(struct if_clone *ifc, char *name, size_t len,
518 struct ifc_data *ifd, struct ifnet **ifpp)
519 {
520 struct iflaggparam iflp;
521 struct lagg_softc *sc;
522 struct ifnet *ifp;
523 int if_type;
524 int error;
525 static const uint8_t eaddr[LAGG_ADDR_LEN];
526
527 if (ifd->params != NULL) {
528 error = ifc_copyin(ifd, &iflp, sizeof(iflp));
529 if (error)
530 return (error);
531
532 switch (iflp.lagg_type) {
533 case LAGG_TYPE_ETHERNET:
534 if_type = IFT_ETHER;
535 break;
536 case LAGG_TYPE_INFINIBAND:
537 if_type = IFT_INFINIBAND;
538 break;
539 default:
540 return (EINVAL);
541 }
542 } else {
543 if_type = IFT_ETHER;
544 }
545
546 sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO);
547 ifp = sc->sc_ifp = if_alloc(if_type);
548 if (ifp == NULL) {
549 free(sc, M_LAGG);
550 return (ENOSPC);
551 }
552 LAGG_SX_INIT(sc);
553
554 mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
555 callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
556
557 LAGG_XLOCK(sc);
558 if (V_def_use_flowid)
559 sc->sc_opts |= LAGG_OPT_USE_FLOWID;
560 if (V_def_use_numa)
561 sc->sc_opts |= LAGG_OPT_USE_NUMA;
562 sc->flowid_shift = V_def_flowid_shift;
563
564 /* Hash all layers by default */
565 sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
566
567 lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
568
569 CK_SLIST_INIT(&sc->sc_ports);
570
571 switch (if_type) {
572 case IFT_ETHER:
573 /* Initialise pseudo media types */
574 ifmedia_init(&sc->sc_media, 0, lagg_media_change,
575 lagg_media_status);
576 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
577 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
578
579 if_initname(ifp, laggname, ifd->unit);
580 ifp->if_transmit = lagg_transmit_ethernet;
581 break;
582 case IFT_INFINIBAND:
583 if_initname(ifp, laggname, ifd->unit);
584 ifp->if_transmit = lagg_transmit_infiniband;
585 break;
586 default:
587 break;
588 }
589 ifp->if_softc = sc;
590 ifp->if_qflush = lagg_qflush;
591 ifp->if_init = lagg_init;
592 ifp->if_ioctl = lagg_ioctl;
593 ifp->if_get_counter = lagg_get_counter;
594 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
595 #if defined(KERN_TLS) || defined(RATELIMIT)
596 ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
597 ifp->if_ratelimit_query = lagg_ratelimit_query;
598 #endif
599 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
600
601 /*
602 * Attach as an ordinary ethernet device, children will be attached
603 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
604 */
605 switch (if_type) {
606 case IFT_ETHER:
607 ether_ifattach(ifp, eaddr);
608 break;
609 case IFT_INFINIBAND:
610 infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr);
611 break;
612 default:
613 break;
614 }
615
616 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
617 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
618 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
619 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
620
621 /* Insert into the global list of laggs */
622 LAGG_LIST_LOCK();
623 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
624 LAGG_LIST_UNLOCK();
625 LAGG_XUNLOCK(sc);
626 *ifpp = ifp;
627
628 return (0);
629 }
630
631 static int
632 lagg_clone_destroy(struct if_clone *ifc, struct ifnet *ifp, uint32_t flags)
633 {
634 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
635 struct lagg_port *lp;
636
637 LAGG_XLOCK(sc);
638 sc->sc_destroying = 1;
639 lagg_stop(sc);
640 ifp->if_flags &= ~IFF_UP;
641
642 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
643 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
644
645 /* Shutdown and remove lagg ports */
646 while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
647 lagg_port_destroy(lp, 1);
648
649 /* Unhook the aggregation protocol */
650 lagg_proto_detach(sc);
651 LAGG_XUNLOCK(sc);
652
653 switch (ifp->if_type) {
654 case IFT_ETHER:
655 ifmedia_removeall(&sc->sc_media);
656 ether_ifdetach(ifp);
657 break;
658 case IFT_INFINIBAND:
659 infiniband_ifdetach(ifp);
660 break;
661 default:
662 break;
663 }
664 if_free(ifp);
665
666 LAGG_LIST_LOCK();
667 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
668 LAGG_LIST_UNLOCK();
669
670 mtx_destroy(&sc->sc_mtx);
671 LAGG_SX_DESTROY(sc);
672 free(sc, M_LAGG);
673
674 return (0);
675 }
676
677 static void
678 lagg_capabilities(struct lagg_softc *sc)
679 {
680 struct lagg_port *lp;
681 int cap, cap2, ena, ena2, pena, pena2;
682 uint64_t hwa;
683 struct ifnet_hw_tsomax hw_tsomax;
684
685 LAGG_XLOCK_ASSERT(sc);
686
687 /* Get common enabled capabilities for the lagg ports */
688 ena = ena2 = ~0;
689 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
690 ena &= lp->lp_ifp->if_capenable;
691 ena2 &= lp->lp_ifp->if_capenable2;
692 }
693 if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
694 ena = ena2 = 0;
695
696 /*
697 * Apply common enabled capabilities back to the lagg ports.
698 * May require several iterations if they are dependent.
699 */
700 do {
701 pena = ena;
702 pena2 = ena2;
703 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
704 lagg_setcaps(lp, ena, ena2);
705 ena &= lp->lp_ifp->if_capenable;
706 ena2 &= lp->lp_ifp->if_capenable2;
707 }
708 } while (pena != ena || pena2 != ena2);
709
710 /* Get other capabilities from the lagg ports */
711 cap = cap2 = ~0;
712 hwa = ~(uint64_t)0;
713 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
714 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
715 cap &= lp->lp_ifp->if_capabilities;
716 cap2 &= lp->lp_ifp->if_capabilities2;
717 hwa &= lp->lp_ifp->if_hwassist;
718 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
719 }
720 if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
721 cap = cap2 = hwa = 0;
722
723 if (sc->sc_ifp->if_capabilities != cap ||
724 sc->sc_ifp->if_capenable != ena ||
725 sc->sc_ifp->if_capenable2 != ena2 ||
726 sc->sc_ifp->if_hwassist != hwa ||
727 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
728 sc->sc_ifp->if_capabilities = cap;
729 sc->sc_ifp->if_capabilities2 = cap2;
730 sc->sc_ifp->if_capenable = ena;
731 sc->sc_ifp->if_capenable2 = ena2;
732 sc->sc_ifp->if_hwassist = hwa;
733 getmicrotime(&sc->sc_ifp->if_lastchange);
734
735 if (sc->sc_ifflags & IFF_DEBUG)
736 if_printf(sc->sc_ifp,
737 "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
738 }
739 }
740
741 static int
742 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
743 {
744 struct lagg_softc *sc_ptr;
745 struct lagg_port *lp, *tlp;
746 struct ifreq ifr;
747 int error, i, oldmtu;
748 int if_type;
749 uint64_t *pval;
750
751 LAGG_XLOCK_ASSERT(sc);
752
753 if (sc->sc_ifp == ifp) {
754 if_printf(sc->sc_ifp,
755 "cannot add a lagg to itself as a port\n");
756 return (EINVAL);
757 }
758
759 if (sc->sc_destroying == 1)
760 return (ENXIO);
761
762 /* Limit the maximal number of lagg ports */
763 if (sc->sc_count >= LAGG_MAX_PORTS)
764 return (ENOSPC);
765
766 /* Check if port has already been associated to a lagg */
767 if (ifp->if_lagg != NULL) {
768 /* Port is already in the current lagg? */
769 lp = (struct lagg_port *)ifp->if_lagg;
770 if (lp->lp_softc == sc)
771 return (EEXIST);
772 return (EBUSY);
773 }
774
775 switch (sc->sc_ifp->if_type) {
776 case IFT_ETHER:
777 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */
778 if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
779 return (EPROTONOSUPPORT);
780 if_type = IFT_IEEE8023ADLAG;
781 break;
782 case IFT_INFINIBAND:
783 /* XXX Disallow non-infiniband interfaces */
784 if (ifp->if_type != IFT_INFINIBAND)
785 return (EPROTONOSUPPORT);
786 if_type = IFT_INFINIBANDLAG;
787 break;
788 default:
789 break;
790 }
791
792 /* Allow the first Ethernet member to define the MTU */
793 oldmtu = -1;
794 if (CK_SLIST_EMPTY(&sc->sc_ports)) {
795 sc->sc_ifp->if_mtu = ifp->if_mtu;
796 } else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
797 if (ifp->if_ioctl == NULL) {
798 if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
799 ifp->if_xname);
800 return (EINVAL);
801 }
802 oldmtu = ifp->if_mtu;
803 strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
804 ifr.ifr_mtu = sc->sc_ifp->if_mtu;
805 error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
806 if (error != 0) {
807 if_printf(sc->sc_ifp, "invalid MTU for %s\n",
808 ifp->if_xname);
809 return (error);
810 }
811 ifr.ifr_mtu = oldmtu;
812 }
813
814 lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO);
815 lp->lp_softc = sc;
816
817 /* Check if port is a stacked lagg */
818 LAGG_LIST_LOCK();
819 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
820 if (ifp == sc_ptr->sc_ifp) {
821 LAGG_LIST_UNLOCK();
822 free(lp, M_LAGG);
823 if (oldmtu != -1)
824 (*ifp->if_ioctl)(ifp, SIOCSIFMTU,
825 (caddr_t)&ifr);
826 return (EINVAL);
827 /* XXX disable stacking for the moment, its untested */
828 #ifdef LAGG_PORT_STACKING
829 lp->lp_flags |= LAGG_PORT_STACK;
830 if (lagg_port_checkstacking(sc_ptr) >=
831 LAGG_MAX_STACKING) {
832 LAGG_LIST_UNLOCK();
833 free(lp, M_LAGG);
834 if (oldmtu != -1)
835 (*ifp->if_ioctl)(ifp, SIOCSIFMTU,
836 (caddr_t)&ifr);
837 return (E2BIG);
838 }
839 #endif
840 }
841 }
842 LAGG_LIST_UNLOCK();
843
844 if_ref(ifp);
845 lp->lp_ifp = ifp;
846
847 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
848 lp->lp_ifcapenable = ifp->if_capenable;
849 if (CK_SLIST_EMPTY(&sc->sc_ports)) {
850 bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
851 lagg_proto_lladdr(sc);
852 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
853 } else {
854 if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
855 }
856 lagg_setflags(lp, 1);
857
858 if (CK_SLIST_EMPTY(&sc->sc_ports))
859 sc->sc_primary = lp;
860
861 /* Change the interface type */
862 lp->lp_iftype = ifp->if_type;
863 ifp->if_type = if_type;
864 ifp->if_lagg = lp;
865 lp->lp_ioctl = ifp->if_ioctl;
866 ifp->if_ioctl = lagg_port_ioctl;
867 lp->lp_output = ifp->if_output;
868 ifp->if_output = lagg_port_output;
869
870 /* Read port counters */
871 pval = lp->port_counters.val;
872 for (i = 0; i < IFCOUNTERS; i++, pval++)
873 *pval = ifp->if_get_counter(ifp, i);
874
875 /*
876 * Insert into the list of ports.
877 * Keep ports sorted by if_index. It is handy, when configuration
878 * is predictable and `ifconfig laggN create ...` command
879 * will lead to the same result each time.
880 */
881 CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
882 if (tlp->lp_ifp->if_index < ifp->if_index && (
883 CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
884 ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
885 ifp->if_index))
886 break;
887 }
888 if (tlp != NULL)
889 CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
890 else
891 CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
892 sc->sc_count++;
893
894 lagg_setmulti(lp);
895
896 if ((error = lagg_proto_addport(sc, lp)) != 0) {
897 /* Remove the port, without calling pr_delport. */
898 lagg_port_destroy(lp, 0);
899 if (oldmtu != -1)
900 (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
901 return (error);
902 }
903
904 /* Update lagg capabilities */
905 lagg_capabilities(sc);
906 lagg_linkstate(sc);
907
908 return (0);
909 }
910
911 #ifdef LAGG_PORT_STACKING
912 static int
913 lagg_port_checkstacking(struct lagg_softc *sc)
914 {
915 struct lagg_softc *sc_ptr;
916 struct lagg_port *lp;
917 int m = 0;
918
919 LAGG_SXLOCK_ASSERT(sc);
920 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
921 if (lp->lp_flags & LAGG_PORT_STACK) {
922 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
923 m = MAX(m, lagg_port_checkstacking(sc_ptr));
924 }
925 }
926
927 return (m + 1);
928 }
929 #endif
930
931 static void
932 lagg_port_destroy_cb(epoch_context_t ec)
933 {
934 struct lagg_port *lp;
935 struct ifnet *ifp;
936
937 lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
938 ifp = lp->lp_ifp;
939
940 if_rele(ifp);
941 free(lp, M_LAGG);
942 }
943
944 static int
945 lagg_port_destroy(struct lagg_port *lp, int rundelport)
946 {
947 struct lagg_softc *sc = lp->lp_softc;
948 struct lagg_port *lp_ptr, *lp0;
949 struct ifnet *ifp = lp->lp_ifp;
950 uint64_t *pval, vdiff;
951 int i;
952
953 LAGG_XLOCK_ASSERT(sc);
954
955 if (rundelport)
956 lagg_proto_delport(sc, lp);
957
958 if (lp->lp_detaching == 0)
959 lagg_clrmulti(lp);
960
961 /* Restore interface */
962 ifp->if_type = lp->lp_iftype;
963 ifp->if_ioctl = lp->lp_ioctl;
964 ifp->if_output = lp->lp_output;
965 ifp->if_lagg = NULL;
966
967 /* Update detached port counters */
968 pval = lp->port_counters.val;
969 for (i = 0; i < IFCOUNTERS; i++, pval++) {
970 vdiff = ifp->if_get_counter(ifp, i) - *pval;
971 sc->detached_counters.val[i] += vdiff;
972 }
973
974 /* Finally, remove the port from the lagg */
975 CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
976 sc->sc_count--;
977
978 /* Update the primary interface */
979 if (lp == sc->sc_primary) {
980 uint8_t lladdr[LAGG_ADDR_LEN];
981
982 if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
983 bzero(&lladdr, LAGG_ADDR_LEN);
984 else
985 bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
986 sc->sc_primary = lp0;
987 if (sc->sc_destroying == 0) {
988 bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
989 lagg_proto_lladdr(sc);
990 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
991
992 /*
993 * Update lladdr for each port (new primary needs update
994 * as well, to switch from old lladdr to its 'real' one).
995 * We can skip this if the lagg is being destroyed.
996 */
997 CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
998 if_setlladdr(lp_ptr->lp_ifp, lladdr,
999 lp_ptr->lp_ifp->if_addrlen);
1000 }
1001 }
1002
1003 if (lp->lp_ifflags)
1004 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
1005
1006 if (lp->lp_detaching == 0) {
1007 lagg_setflags(lp, 0);
1008 lagg_setcaps(lp, lp->lp_ifcapenable, lp->lp_ifcapenable2);
1009 if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
1010 }
1011
1012 /*
1013 * free port and release it's ifnet reference after a grace period has
1014 * elapsed.
1015 */
1016 NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
1017 /* Update lagg capabilities */
1018 lagg_capabilities(sc);
1019 lagg_linkstate(sc);
1020
1021 return (0);
1022 }
1023
1024 static int
1025 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1026 {
1027 struct epoch_tracker et;
1028 struct lagg_reqport *rp = (struct lagg_reqport *)data;
1029 struct lagg_softc *sc;
1030 struct lagg_port *lp = NULL;
1031 int error = 0;
1032
1033 /* Should be checked by the caller */
1034 switch (ifp->if_type) {
1035 case IFT_IEEE8023ADLAG:
1036 case IFT_INFINIBANDLAG:
1037 if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1038 goto fallback;
1039 break;
1040 default:
1041 goto fallback;
1042 }
1043
1044 switch (cmd) {
1045 case SIOCGLAGGPORT:
1046 if (rp->rp_portname[0] == '\0' ||
1047 ifunit(rp->rp_portname) != ifp) {
1048 error = EINVAL;
1049 break;
1050 }
1051
1052 NET_EPOCH_ENTER(et);
1053 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1054 error = ENOENT;
1055 NET_EPOCH_EXIT(et);
1056 break;
1057 }
1058
1059 lagg_port2req(lp, rp);
1060 NET_EPOCH_EXIT(et);
1061 break;
1062
1063 case SIOCSIFCAP:
1064 case SIOCSIFCAPNV:
1065 if (lp->lp_ioctl == NULL) {
1066 error = EINVAL;
1067 break;
1068 }
1069 error = (*lp->lp_ioctl)(ifp, cmd, data);
1070 if (error)
1071 break;
1072
1073 /* Update lagg interface capabilities */
1074 LAGG_XLOCK(sc);
1075 lagg_capabilities(sc);
1076 LAGG_XUNLOCK(sc);
1077 VLAN_CAPABILITIES(sc->sc_ifp);
1078 break;
1079
1080 case SIOCSIFMTU:
1081 /* Do not allow the MTU to be changed once joined */
1082 error = EINVAL;
1083 break;
1084
1085 default:
1086 goto fallback;
1087 }
1088
1089 return (error);
1090
1091 fallback:
1092 if (lp != NULL && lp->lp_ioctl != NULL)
1093 return ((*lp->lp_ioctl)(ifp, cmd, data));
1094
1095 return (EINVAL);
1096 }
1097
1098 /*
1099 * Requests counter @cnt data.
1100 *
1101 * Counter value is calculated the following way:
1102 * 1) for each port, sum difference between current and "initial" measurements.
1103 * 2) add lagg logical interface counters.
1104 * 3) add data from detached_counters array.
1105 *
1106 * We also do the following things on ports attach/detach:
1107 * 1) On port attach we store all counters it has into port_counter array.
1108 * 2) On port detach we add the different between "initial" and
1109 * current counters data to detached_counters array.
1110 */
1111 static uint64_t
1112 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1113 {
1114 struct epoch_tracker et;
1115 struct lagg_softc *sc;
1116 struct lagg_port *lp;
1117 struct ifnet *lpifp;
1118 uint64_t newval, oldval, vsum;
1119
1120 /* Revise this when we've got non-generic counters. */
1121 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1122
1123 sc = (struct lagg_softc *)ifp->if_softc;
1124
1125 vsum = 0;
1126 NET_EPOCH_ENTER(et);
1127 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1128 /* Saved attached value */
1129 oldval = lp->port_counters.val[cnt];
1130 /* current value */
1131 lpifp = lp->lp_ifp;
1132 newval = lpifp->if_get_counter(lpifp, cnt);
1133 /* Calculate diff and save new */
1134 vsum += newval - oldval;
1135 }
1136 NET_EPOCH_EXIT(et);
1137
1138 /*
1139 * Add counter data which might be added by upper
1140 * layer protocols operating on logical interface.
1141 */
1142 vsum += if_get_counter_default(ifp, cnt);
1143
1144 /*
1145 * Add counter data from detached ports counters
1146 */
1147 vsum += sc->detached_counters.val[cnt];
1148
1149 return (vsum);
1150 }
1151
1152 /*
1153 * For direct output to child ports.
1154 */
1155 static int
1156 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1157 const struct sockaddr *dst, struct route *ro)
1158 {
1159 struct lagg_port *lp = ifp->if_lagg;
1160
1161 switch (dst->sa_family) {
1162 case pseudo_AF_HDRCMPLT:
1163 case AF_UNSPEC:
1164 if (lp != NULL)
1165 return ((*lp->lp_output)(ifp, m, dst, ro));
1166 }
1167
1168 /* drop any other frames */
1169 m_freem(m);
1170 return (ENETDOWN);
1171 }
1172
1173 static void
1174 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1175 {
1176 struct lagg_port *lp;
1177 struct lagg_softc *sc;
1178
1179 if ((lp = ifp->if_lagg) == NULL)
1180 return;
1181 /* If the ifnet is just being renamed, don't do anything. */
1182 if (ifp->if_flags & IFF_RENAMING)
1183 return;
1184
1185 sc = lp->lp_softc;
1186
1187 LAGG_XLOCK(sc);
1188 lp->lp_detaching = 1;
1189 lagg_port_destroy(lp, 1);
1190 LAGG_XUNLOCK(sc);
1191 VLAN_CAPABILITIES(sc->sc_ifp);
1192 }
1193
1194 static void
1195 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1196 {
1197 struct lagg_softc *sc = lp->lp_softc;
1198
1199 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1200 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1201 rp->rp_prio = lp->lp_prio;
1202 rp->rp_flags = lp->lp_flags;
1203 lagg_proto_portreq(sc, lp, &rp->rp_psc);
1204
1205 /* Add protocol specific flags */
1206 switch (sc->sc_proto) {
1207 case LAGG_PROTO_FAILOVER:
1208 if (lp == sc->sc_primary)
1209 rp->rp_flags |= LAGG_PORT_MASTER;
1210 if (lp == lagg_link_active(sc, sc->sc_primary))
1211 rp->rp_flags |= LAGG_PORT_ACTIVE;
1212 break;
1213
1214 case LAGG_PROTO_ROUNDROBIN:
1215 case LAGG_PROTO_LOADBALANCE:
1216 case LAGG_PROTO_BROADCAST:
1217 if (LAGG_PORTACTIVE(lp))
1218 rp->rp_flags |= LAGG_PORT_ACTIVE;
1219 break;
1220
1221 case LAGG_PROTO_LACP:
1222 /* LACP has a different definition of active */
1223 if (lacp_isactive(lp))
1224 rp->rp_flags |= LAGG_PORT_ACTIVE;
1225 if (lacp_iscollecting(lp))
1226 rp->rp_flags |= LAGG_PORT_COLLECTING;
1227 if (lacp_isdistributing(lp))
1228 rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1229 break;
1230 }
1231
1232 }
1233
1234 static void
1235 lagg_watchdog_infiniband(void *arg)
1236 {
1237 struct epoch_tracker et;
1238 struct lagg_softc *sc;
1239 struct lagg_port *lp;
1240 struct ifnet *ifp;
1241 struct ifnet *lp_ifp;
1242
1243 sc = arg;
1244
1245 /*
1246 * Because infiniband nodes have a fixed MAC address, which is
1247 * generated by the so-called GID, we need to regularly update
1248 * the link level address of the parent lagg<N> device when
1249 * the active port changes. Possibly we could piggy-back on
1250 * link up/down events aswell, but using a timer also provides
1251 * a guarantee against too frequent events. This operation
1252 * does not have to be atomic.
1253 */
1254 NET_EPOCH_ENTER(et);
1255 lp = lagg_link_active(sc, sc->sc_primary);
1256 if (lp != NULL) {
1257 ifp = sc->sc_ifp;
1258 lp_ifp = lp->lp_ifp;
1259
1260 if (ifp != NULL && lp_ifp != NULL &&
1261 (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0 ||
1262 memcmp(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen) != 0)) {
1263 memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1264 memcpy(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen);
1265
1266 CURVNET_SET(ifp->if_vnet);
1267 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1268 CURVNET_RESTORE();
1269 }
1270 }
1271 NET_EPOCH_EXIT(et);
1272
1273 callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1274 }
1275
1276 static void
1277 lagg_init(void *xsc)
1278 {
1279 struct lagg_softc *sc = (struct lagg_softc *)xsc;
1280 struct ifnet *ifp = sc->sc_ifp;
1281 struct lagg_port *lp;
1282
1283 LAGG_XLOCK(sc);
1284 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1285 LAGG_XUNLOCK(sc);
1286 return;
1287 }
1288
1289 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1290
1291 /*
1292 * Update the port lladdrs if needed.
1293 * This might be if_setlladdr() notification
1294 * that lladdr has been changed.
1295 */
1296 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1297 if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1298 ifp->if_addrlen) != 0)
1299 if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1300 }
1301
1302 lagg_proto_init(sc);
1303
1304 if (ifp->if_type == IFT_INFINIBAND) {
1305 mtx_lock(&sc->sc_mtx);
1306 lagg_watchdog_infiniband(sc);
1307 mtx_unlock(&sc->sc_mtx);
1308 }
1309
1310 LAGG_XUNLOCK(sc);
1311 }
1312
1313 static void
1314 lagg_stop(struct lagg_softc *sc)
1315 {
1316 struct ifnet *ifp = sc->sc_ifp;
1317
1318 LAGG_XLOCK_ASSERT(sc);
1319
1320 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1321 return;
1322
1323 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1324
1325 lagg_proto_stop(sc);
1326
1327 mtx_lock(&sc->sc_mtx);
1328 callout_stop(&sc->sc_watchdog);
1329 mtx_unlock(&sc->sc_mtx);
1330
1331 callout_drain(&sc->sc_watchdog);
1332 }
1333
1334 static int
1335 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1336 {
1337 struct epoch_tracker et;
1338 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1339 struct lagg_reqall *ra = (struct lagg_reqall *)data;
1340 struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1341 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1342 struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1343 struct ifreq *ifr = (struct ifreq *)data;
1344 struct lagg_port *lp;
1345 struct ifnet *tpif;
1346 struct thread *td = curthread;
1347 char *buf, *outbuf;
1348 int count, buflen, len, error = 0, oldmtu;
1349
1350 bzero(&rpbuf, sizeof(rpbuf));
1351
1352 /* XXX: This can race with lagg_clone_destroy. */
1353
1354 switch (cmd) {
1355 case SIOCGLAGG:
1356 LAGG_XLOCK(sc);
1357 buflen = sc->sc_count * sizeof(struct lagg_reqport);
1358 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1359 ra->ra_proto = sc->sc_proto;
1360 lagg_proto_request(sc, &ra->ra_psc);
1361 count = 0;
1362 buf = outbuf;
1363 len = min(ra->ra_size, buflen);
1364 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1365 if (len < sizeof(rpbuf))
1366 break;
1367
1368 lagg_port2req(lp, &rpbuf);
1369 memcpy(buf, &rpbuf, sizeof(rpbuf));
1370 count++;
1371 buf += sizeof(rpbuf);
1372 len -= sizeof(rpbuf);
1373 }
1374 LAGG_XUNLOCK(sc);
1375 ra->ra_ports = count;
1376 ra->ra_size = count * sizeof(rpbuf);
1377 error = copyout(outbuf, ra->ra_port, ra->ra_size);
1378 free(outbuf, M_TEMP);
1379 break;
1380 case SIOCSLAGG:
1381 error = priv_check(td, PRIV_NET_LAGG);
1382 if (error)
1383 break;
1384 if (ra->ra_proto >= LAGG_PROTO_MAX) {
1385 error = EPROTONOSUPPORT;
1386 break;
1387 }
1388 /* Infiniband only supports the failover protocol. */
1389 if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1390 ifp->if_type == IFT_INFINIBAND) {
1391 error = EPROTONOSUPPORT;
1392 break;
1393 }
1394 LAGG_XLOCK(sc);
1395 lagg_proto_detach(sc);
1396 lagg_proto_attach(sc, ra->ra_proto);
1397 LAGG_XUNLOCK(sc);
1398 break;
1399 case SIOCGLAGGOPTS:
1400 LAGG_XLOCK(sc);
1401 ro->ro_opts = sc->sc_opts;
1402 if (sc->sc_proto == LAGG_PROTO_LACP) {
1403 struct lacp_softc *lsc;
1404
1405 lsc = (struct lacp_softc *)sc->sc_psc;
1406 if (lsc->lsc_debug.lsc_tx_test != 0)
1407 ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1408 if (lsc->lsc_debug.lsc_rx_test != 0)
1409 ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1410 if (lsc->lsc_strict_mode != 0)
1411 ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1412 if (lsc->lsc_fast_timeout != 0)
1413 ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1414
1415 ro->ro_active = sc->sc_active;
1416 } else {
1417 ro->ro_active = 0;
1418 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1419 ro->ro_active += LAGG_PORTACTIVE(lp);
1420 }
1421 ro->ro_bkt = sc->sc_stride;
1422 ro->ro_flapping = sc->sc_flapping;
1423 ro->ro_flowid_shift = sc->flowid_shift;
1424 LAGG_XUNLOCK(sc);
1425 break;
1426 case SIOCSLAGGOPTS:
1427 error = priv_check(td, PRIV_NET_LAGG);
1428 if (error)
1429 break;
1430
1431 /*
1432 * The stride option was added without defining a corresponding
1433 * LAGG_OPT flag, so handle a non-zero value before checking
1434 * anything else to preserve compatibility.
1435 */
1436 LAGG_XLOCK(sc);
1437 if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1438 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1439 LAGG_XUNLOCK(sc);
1440 error = EINVAL;
1441 break;
1442 }
1443 sc->sc_stride = ro->ro_bkt;
1444 }
1445 if (ro->ro_opts == 0) {
1446 LAGG_XUNLOCK(sc);
1447 break;
1448 }
1449
1450 /*
1451 * Set options. LACP options are stored in sc->sc_psc,
1452 * not in sc_opts.
1453 */
1454 int valid, lacp;
1455
1456 switch (ro->ro_opts) {
1457 case LAGG_OPT_USE_FLOWID:
1458 case -LAGG_OPT_USE_FLOWID:
1459 case LAGG_OPT_USE_NUMA:
1460 case -LAGG_OPT_USE_NUMA:
1461 case LAGG_OPT_FLOWIDSHIFT:
1462 case LAGG_OPT_RR_LIMIT:
1463 valid = 1;
1464 lacp = 0;
1465 break;
1466 case LAGG_OPT_LACP_TXTEST:
1467 case -LAGG_OPT_LACP_TXTEST:
1468 case LAGG_OPT_LACP_RXTEST:
1469 case -LAGG_OPT_LACP_RXTEST:
1470 case LAGG_OPT_LACP_STRICT:
1471 case -LAGG_OPT_LACP_STRICT:
1472 case LAGG_OPT_LACP_FAST_TIMO:
1473 case -LAGG_OPT_LACP_FAST_TIMO:
1474 valid = lacp = 1;
1475 break;
1476 default:
1477 valid = lacp = 0;
1478 break;
1479 }
1480
1481 if (valid == 0 ||
1482 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1483 /* Invalid combination of options specified. */
1484 error = EINVAL;
1485 LAGG_XUNLOCK(sc);
1486 break; /* Return from SIOCSLAGGOPTS. */
1487 }
1488
1489 /*
1490 * Store new options into sc->sc_opts except for
1491 * FLOWIDSHIFT, RR and LACP options.
1492 */
1493 if (lacp == 0) {
1494 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1495 sc->flowid_shift = ro->ro_flowid_shift;
1496 else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1497 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1498 ro->ro_bkt == 0) {
1499 error = EINVAL;
1500 LAGG_XUNLOCK(sc);
1501 break;
1502 }
1503 sc->sc_stride = ro->ro_bkt;
1504 } else if (ro->ro_opts > 0)
1505 sc->sc_opts |= ro->ro_opts;
1506 else
1507 sc->sc_opts &= ~ro->ro_opts;
1508 } else {
1509 struct lacp_softc *lsc;
1510 struct lacp_port *lp;
1511
1512 lsc = (struct lacp_softc *)sc->sc_psc;
1513
1514 switch (ro->ro_opts) {
1515 case LAGG_OPT_LACP_TXTEST:
1516 lsc->lsc_debug.lsc_tx_test = 1;
1517 break;
1518 case -LAGG_OPT_LACP_TXTEST:
1519 lsc->lsc_debug.lsc_tx_test = 0;
1520 break;
1521 case LAGG_OPT_LACP_RXTEST:
1522 lsc->lsc_debug.lsc_rx_test = 1;
1523 break;
1524 case -LAGG_OPT_LACP_RXTEST:
1525 lsc->lsc_debug.lsc_rx_test = 0;
1526 break;
1527 case LAGG_OPT_LACP_STRICT:
1528 lsc->lsc_strict_mode = 1;
1529 break;
1530 case -LAGG_OPT_LACP_STRICT:
1531 lsc->lsc_strict_mode = 0;
1532 break;
1533 case LAGG_OPT_LACP_FAST_TIMO:
1534 LACP_LOCK(lsc);
1535 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1536 lp->lp_state |= LACP_STATE_TIMEOUT;
1537 LACP_UNLOCK(lsc);
1538 lsc->lsc_fast_timeout = 1;
1539 break;
1540 case -LAGG_OPT_LACP_FAST_TIMO:
1541 LACP_LOCK(lsc);
1542 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1543 lp->lp_state &= ~LACP_STATE_TIMEOUT;
1544 LACP_UNLOCK(lsc);
1545 lsc->lsc_fast_timeout = 0;
1546 break;
1547 }
1548 }
1549 LAGG_XUNLOCK(sc);
1550 break;
1551 case SIOCGLAGGFLAGS:
1552 rf->rf_flags = 0;
1553 LAGG_XLOCK(sc);
1554 if (sc->sc_flags & MBUF_HASHFLAG_L2)
1555 rf->rf_flags |= LAGG_F_HASHL2;
1556 if (sc->sc_flags & MBUF_HASHFLAG_L3)
1557 rf->rf_flags |= LAGG_F_HASHL3;
1558 if (sc->sc_flags & MBUF_HASHFLAG_L4)
1559 rf->rf_flags |= LAGG_F_HASHL4;
1560 LAGG_XUNLOCK(sc);
1561 break;
1562 case SIOCSLAGGHASH:
1563 error = priv_check(td, PRIV_NET_LAGG);
1564 if (error)
1565 break;
1566 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1567 error = EINVAL;
1568 break;
1569 }
1570 LAGG_XLOCK(sc);
1571 sc->sc_flags = 0;
1572 if (rf->rf_flags & LAGG_F_HASHL2)
1573 sc->sc_flags |= MBUF_HASHFLAG_L2;
1574 if (rf->rf_flags & LAGG_F_HASHL3)
1575 sc->sc_flags |= MBUF_HASHFLAG_L3;
1576 if (rf->rf_flags & LAGG_F_HASHL4)
1577 sc->sc_flags |= MBUF_HASHFLAG_L4;
1578 LAGG_XUNLOCK(sc);
1579 break;
1580 case SIOCGLAGGPORT:
1581 if (rp->rp_portname[0] == '\0' ||
1582 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1583 error = EINVAL;
1584 break;
1585 }
1586
1587 NET_EPOCH_ENTER(et);
1588 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1589 lp->lp_softc != sc) {
1590 error = ENOENT;
1591 NET_EPOCH_EXIT(et);
1592 if_rele(tpif);
1593 break;
1594 }
1595
1596 lagg_port2req(lp, rp);
1597 NET_EPOCH_EXIT(et);
1598 if_rele(tpif);
1599 break;
1600 case SIOCSLAGGPORT:
1601 error = priv_check(td, PRIV_NET_LAGG);
1602 if (error)
1603 break;
1604 if (rp->rp_portname[0] == '\0' ||
1605 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1606 error = EINVAL;
1607 break;
1608 }
1609 #ifdef INET6
1610 /*
1611 * A laggport interface should not have inet6 address
1612 * because two interfaces with a valid link-local
1613 * scope zone must not be merged in any form. This
1614 * restriction is needed to prevent violation of
1615 * link-local scope zone. Attempts to add a laggport
1616 * interface which has inet6 addresses triggers
1617 * removal of all inet6 addresses on the member
1618 * interface.
1619 */
1620 if (in6ifa_llaonifp(tpif)) {
1621 in6_ifdetach(tpif);
1622 if_printf(sc->sc_ifp,
1623 "IPv6 addresses on %s have been removed "
1624 "before adding it as a member to prevent "
1625 "IPv6 address scope violation.\n",
1626 tpif->if_xname);
1627 }
1628 #endif
1629 oldmtu = ifp->if_mtu;
1630 LAGG_XLOCK(sc);
1631 error = lagg_port_create(sc, tpif);
1632 LAGG_XUNLOCK(sc);
1633 if_rele(tpif);
1634
1635 /*
1636 * LAGG MTU may change during addition of the first port.
1637 * If it did, do network layer specific procedure.
1638 */
1639 if (ifp->if_mtu != oldmtu) {
1640 #ifdef INET6
1641 nd6_setmtu(ifp);
1642 #endif
1643 rt_updatemtu(ifp);
1644 }
1645
1646 VLAN_CAPABILITIES(ifp);
1647 break;
1648 case SIOCSLAGGDELPORT:
1649 error = priv_check(td, PRIV_NET_LAGG);
1650 if (error)
1651 break;
1652 if (rp->rp_portname[0] == '\0' ||
1653 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1654 error = EINVAL;
1655 break;
1656 }
1657
1658 LAGG_XLOCK(sc);
1659 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1660 lp->lp_softc != sc) {
1661 error = ENOENT;
1662 LAGG_XUNLOCK(sc);
1663 if_rele(tpif);
1664 break;
1665 }
1666
1667 error = lagg_port_destroy(lp, 1);
1668 LAGG_XUNLOCK(sc);
1669 if_rele(tpif);
1670 VLAN_CAPABILITIES(ifp);
1671 break;
1672 case SIOCSIFFLAGS:
1673 /* Set flags on ports too */
1674 LAGG_XLOCK(sc);
1675 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1676 lagg_setflags(lp, 1);
1677 }
1678
1679 if (!(ifp->if_flags & IFF_UP) &&
1680 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1681 /*
1682 * If interface is marked down and it is running,
1683 * then stop and disable it.
1684 */
1685 lagg_stop(sc);
1686 LAGG_XUNLOCK(sc);
1687 } else if ((ifp->if_flags & IFF_UP) &&
1688 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1689 /*
1690 * If interface is marked up and it is stopped, then
1691 * start it.
1692 */
1693 LAGG_XUNLOCK(sc);
1694 (*ifp->if_init)(sc);
1695 } else
1696 LAGG_XUNLOCK(sc);
1697 break;
1698 case SIOCADDMULTI:
1699 case SIOCDELMULTI:
1700 LAGG_XLOCK(sc);
1701 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1702 lagg_clrmulti(lp);
1703 lagg_setmulti(lp);
1704 }
1705 LAGG_XUNLOCK(sc);
1706 error = 0;
1707 break;
1708 case SIOCSIFMEDIA:
1709 case SIOCGIFMEDIA:
1710 if (ifp->if_type == IFT_INFINIBAND)
1711 error = EINVAL;
1712 else
1713 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1714 break;
1715
1716 case SIOCSIFCAP:
1717 case SIOCSIFCAPNV:
1718 LAGG_XLOCK(sc);
1719 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1720 if (lp->lp_ioctl != NULL)
1721 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1722 }
1723 lagg_capabilities(sc);
1724 LAGG_XUNLOCK(sc);
1725 VLAN_CAPABILITIES(ifp);
1726 error = 0;
1727 break;
1728
1729 case SIOCGIFCAPNV:
1730 error = 0;
1731 break;
1732
1733 case SIOCSIFMTU:
1734 LAGG_XLOCK(sc);
1735 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1736 if (lp->lp_ioctl != NULL)
1737 error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1738 else
1739 error = EINVAL;
1740 if (error != 0) {
1741 if_printf(ifp,
1742 "failed to change MTU to %d on port %s, "
1743 "reverting all ports to original MTU (%d)\n",
1744 ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1745 break;
1746 }
1747 }
1748 if (error == 0) {
1749 ifp->if_mtu = ifr->ifr_mtu;
1750 } else {
1751 /* set every port back to the original MTU */
1752 ifr->ifr_mtu = ifp->if_mtu;
1753 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1754 if (lp->lp_ioctl != NULL)
1755 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1756 }
1757 }
1758 lagg_capabilities(sc);
1759 LAGG_XUNLOCK(sc);
1760 VLAN_CAPABILITIES(ifp);
1761 break;
1762
1763 default:
1764 error = ether_ioctl(ifp, cmd, data);
1765 break;
1766 }
1767 return (error);
1768 }
1769
1770 #if defined(KERN_TLS) || defined(RATELIMIT)
1771 #ifdef RATELIMIT
1772 static const struct if_snd_tag_sw lagg_snd_tag_ul_sw = {
1773 .snd_tag_modify = lagg_snd_tag_modify,
1774 .snd_tag_query = lagg_snd_tag_query,
1775 .snd_tag_free = lagg_snd_tag_free,
1776 .next_snd_tag = lagg_next_snd_tag,
1777 .type = IF_SND_TAG_TYPE_UNLIMITED
1778 };
1779
1780 static const struct if_snd_tag_sw lagg_snd_tag_rl_sw = {
1781 .snd_tag_modify = lagg_snd_tag_modify,
1782 .snd_tag_query = lagg_snd_tag_query,
1783 .snd_tag_free = lagg_snd_tag_free,
1784 .next_snd_tag = lagg_next_snd_tag,
1785 .type = IF_SND_TAG_TYPE_RATE_LIMIT
1786 };
1787 #endif
1788
1789 #ifdef KERN_TLS
1790 static const struct if_snd_tag_sw lagg_snd_tag_tls_sw = {
1791 .snd_tag_modify = lagg_snd_tag_modify,
1792 .snd_tag_query = lagg_snd_tag_query,
1793 .snd_tag_free = lagg_snd_tag_free,
1794 .next_snd_tag = lagg_next_snd_tag,
1795 .type = IF_SND_TAG_TYPE_TLS
1796 };
1797
1798 #ifdef RATELIMIT
1799 static const struct if_snd_tag_sw lagg_snd_tag_tls_rl_sw = {
1800 .snd_tag_modify = lagg_snd_tag_modify,
1801 .snd_tag_query = lagg_snd_tag_query,
1802 .snd_tag_free = lagg_snd_tag_free,
1803 .next_snd_tag = lagg_next_snd_tag,
1804 .type = IF_SND_TAG_TYPE_TLS_RATE_LIMIT
1805 };
1806 #endif
1807 #endif
1808
1809 static inline struct lagg_snd_tag *
1810 mst_to_lst(struct m_snd_tag *mst)
1811 {
1812
1813 return (__containerof(mst, struct lagg_snd_tag, com));
1814 }
1815
1816 /*
1817 * Look up the port used by a specific flow. This only works for lagg
1818 * protocols with deterministic port mappings (e.g. not roundrobin).
1819 * In addition protocols which use a hash to map flows to ports must
1820 * be configured to use the mbuf flowid rather than hashing packet
1821 * contents.
1822 */
1823 static struct lagg_port *
1824 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1825 uint8_t numa_domain)
1826 {
1827 struct lagg_softc *sc;
1828 struct lagg_port *lp;
1829 struct lagg_lb *lb;
1830 uint32_t hash, p;
1831 int err;
1832
1833 sc = ifp->if_softc;
1834
1835 switch (sc->sc_proto) {
1836 case LAGG_PROTO_FAILOVER:
1837 return (lagg_link_active(sc, sc->sc_primary));
1838 case LAGG_PROTO_LOADBALANCE:
1839 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1840 flowtype == M_HASHTYPE_NONE)
1841 return (NULL);
1842 p = flowid >> sc->flowid_shift;
1843 p %= sc->sc_count;
1844 lb = (struct lagg_lb *)sc->sc_psc;
1845 lp = lb->lb_ports[p];
1846 return (lagg_link_active(sc, lp));
1847 case LAGG_PROTO_LACP:
1848 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1849 flowtype == M_HASHTYPE_NONE)
1850 return (NULL);
1851 hash = flowid >> sc->flowid_shift;
1852 return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, &err));
1853 default:
1854 return (NULL);
1855 }
1856 }
1857
1858 static int
1859 lagg_snd_tag_alloc(struct ifnet *ifp,
1860 union if_snd_tag_alloc_params *params,
1861 struct m_snd_tag **ppmt)
1862 {
1863 struct epoch_tracker et;
1864 const struct if_snd_tag_sw *sw;
1865 struct lagg_snd_tag *lst;
1866 struct lagg_port *lp;
1867 struct ifnet *lp_ifp;
1868 struct m_snd_tag *mst;
1869 int error;
1870
1871 switch (params->hdr.type) {
1872 #ifdef RATELIMIT
1873 case IF_SND_TAG_TYPE_UNLIMITED:
1874 sw = &lagg_snd_tag_ul_sw;
1875 break;
1876 case IF_SND_TAG_TYPE_RATE_LIMIT:
1877 sw = &lagg_snd_tag_rl_sw;
1878 break;
1879 #endif
1880 #ifdef KERN_TLS
1881 case IF_SND_TAG_TYPE_TLS:
1882 sw = &lagg_snd_tag_tls_sw;
1883 break;
1884 case IF_SND_TAG_TYPE_TLS_RX:
1885 /* Return tag from port interface directly. */
1886 sw = NULL;
1887 break;
1888 #ifdef RATELIMIT
1889 case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
1890 sw = &lagg_snd_tag_tls_rl_sw;
1891 break;
1892 #endif
1893 #endif
1894 default:
1895 return (EOPNOTSUPP);
1896 }
1897
1898 NET_EPOCH_ENTER(et);
1899 lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1900 params->hdr.flowtype, params->hdr.numa_domain);
1901 if (lp == NULL) {
1902 NET_EPOCH_EXIT(et);
1903 return (EOPNOTSUPP);
1904 }
1905 if (lp->lp_ifp == NULL) {
1906 NET_EPOCH_EXIT(et);
1907 return (EOPNOTSUPP);
1908 }
1909 lp_ifp = lp->lp_ifp;
1910 if_ref(lp_ifp);
1911 NET_EPOCH_EXIT(et);
1912
1913 if (sw != NULL) {
1914 lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1915 if (lst == NULL) {
1916 if_rele(lp_ifp);
1917 return (ENOMEM);
1918 }
1919 } else
1920 lst = NULL;
1921
1922 error = m_snd_tag_alloc(lp_ifp, params, &mst);
1923 if_rele(lp_ifp);
1924 if (error) {
1925 free(lst, M_LAGG);
1926 return (error);
1927 }
1928
1929 if (sw != NULL) {
1930 m_snd_tag_init(&lst->com, ifp, sw);
1931 lst->tag = mst;
1932
1933 *ppmt = &lst->com;
1934 } else
1935 *ppmt = mst;
1936
1937 return (0);
1938 }
1939
1940 static struct m_snd_tag *
1941 lagg_next_snd_tag(struct m_snd_tag *mst)
1942 {
1943 struct lagg_snd_tag *lst;
1944
1945 lst = mst_to_lst(mst);
1946 return (lst->tag);
1947 }
1948
1949 static int
1950 lagg_snd_tag_modify(struct m_snd_tag *mst,
1951 union if_snd_tag_modify_params *params)
1952 {
1953 struct lagg_snd_tag *lst;
1954
1955 lst = mst_to_lst(mst);
1956 return (lst->tag->sw->snd_tag_modify(lst->tag, params));
1957 }
1958
1959 static int
1960 lagg_snd_tag_query(struct m_snd_tag *mst,
1961 union if_snd_tag_query_params *params)
1962 {
1963 struct lagg_snd_tag *lst;
1964
1965 lst = mst_to_lst(mst);
1966 return (lst->tag->sw->snd_tag_query(lst->tag, params));
1967 }
1968
1969 static void
1970 lagg_snd_tag_free(struct m_snd_tag *mst)
1971 {
1972 struct lagg_snd_tag *lst;
1973
1974 lst = mst_to_lst(mst);
1975 m_snd_tag_rele(lst->tag);
1976 free(lst, M_LAGG);
1977 }
1978
1979 static void
1980 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1981 {
1982 /*
1983 * For lagg, we have an indirect
1984 * interface. The caller needs to
1985 * get a ratelimit tag on the actual
1986 * interface the flow will go on.
1987 */
1988 q->rate_table = NULL;
1989 q->flags = RT_IS_INDIRECT;
1990 q->max_flows = 0;
1991 q->number_of_rates = 0;
1992 }
1993 #endif
1994
1995 static int
1996 lagg_setmulti(struct lagg_port *lp)
1997 {
1998 struct lagg_softc *sc = lp->lp_softc;
1999 struct ifnet *ifp = lp->lp_ifp;
2000 struct ifnet *scifp = sc->sc_ifp;
2001 struct lagg_mc *mc;
2002 struct ifmultiaddr *ifma;
2003 int error;
2004
2005 IF_ADDR_WLOCK(scifp);
2006 CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
2007 if (ifma->ifma_addr->sa_family != AF_LINK)
2008 continue;
2009 mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
2010 if (mc == NULL) {
2011 IF_ADDR_WUNLOCK(scifp);
2012 return (ENOMEM);
2013 }
2014 bcopy(ifma->ifma_addr, &mc->mc_addr,
2015 ifma->ifma_addr->sa_len);
2016 mc->mc_addr.sdl_index = ifp->if_index;
2017 mc->mc_ifma = NULL;
2018 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
2019 }
2020 IF_ADDR_WUNLOCK(scifp);
2021 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
2022 error = if_addmulti(ifp,
2023 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
2024 if (error)
2025 return (error);
2026 }
2027 return (0);
2028 }
2029
2030 static int
2031 lagg_clrmulti(struct lagg_port *lp)
2032 {
2033 struct lagg_mc *mc;
2034
2035 LAGG_XLOCK_ASSERT(lp->lp_softc);
2036 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
2037 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
2038 if (mc->mc_ifma && lp->lp_detaching == 0)
2039 if_delmulti_ifma(mc->mc_ifma);
2040 free(mc, M_LAGG);
2041 }
2042 return (0);
2043 }
2044
2045 static void
2046 lagg_setcaps(struct lagg_port *lp, int cap, int cap2)
2047 {
2048 struct ifreq ifr;
2049 struct siocsifcapnv_driver_data drv_ioctl_data;
2050
2051 if (lp->lp_ifp->if_capenable == cap &&
2052 lp->lp_ifp->if_capenable2 == cap2)
2053 return;
2054 if (lp->lp_ioctl == NULL)
2055 return;
2056 /* XXX */
2057 if ((lp->lp_ifp->if_capabilities & IFCAP_NV) != 0) {
2058 drv_ioctl_data.reqcap = cap;
2059 drv_ioctl_data.reqcap2 = cap2;
2060 drv_ioctl_data.nvcap = NULL;
2061 (*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAPNV,
2062 (caddr_t)&drv_ioctl_data);
2063 } else {
2064 ifr.ifr_reqcap = cap;
2065 (*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr);
2066 }
2067 }
2068
2069 /* Handle a ref counted flag that should be set on the lagg port as well */
2070 static int
2071 lagg_setflag(struct lagg_port *lp, int flag, int status,
2072 int (*func)(struct ifnet *, int))
2073 {
2074 struct lagg_softc *sc = lp->lp_softc;
2075 struct ifnet *scifp = sc->sc_ifp;
2076 struct ifnet *ifp = lp->lp_ifp;
2077 int error;
2078
2079 LAGG_XLOCK_ASSERT(sc);
2080
2081 status = status ? (scifp->if_flags & flag) : 0;
2082 /* Now "status" contains the flag value or 0 */
2083
2084 /*
2085 * See if recorded ports status is different from what
2086 * we want it to be. If it is, flip it. We record ports
2087 * status in lp_ifflags so that we won't clear ports flag
2088 * we haven't set. In fact, we don't clear or set ports
2089 * flags directly, but get or release references to them.
2090 * That's why we can be sure that recorded flags still are
2091 * in accord with actual ports flags.
2092 */
2093 if (status != (lp->lp_ifflags & flag)) {
2094 error = (*func)(ifp, status);
2095 if (error)
2096 return (error);
2097 lp->lp_ifflags &= ~flag;
2098 lp->lp_ifflags |= status;
2099 }
2100 return (0);
2101 }
2102
2103 /*
2104 * Handle IFF_* flags that require certain changes on the lagg port
2105 * if "status" is true, update ports flags respective to the lagg
2106 * if "status" is false, forcedly clear the flags set on port.
2107 */
2108 static int
2109 lagg_setflags(struct lagg_port *lp, int status)
2110 {
2111 int error, i;
2112
2113 for (i = 0; lagg_pflags[i].flag; i++) {
2114 error = lagg_setflag(lp, lagg_pflags[i].flag,
2115 status, lagg_pflags[i].func);
2116 if (error)
2117 return (error);
2118 }
2119 return (0);
2120 }
2121
2122 static int
2123 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2124 {
2125 struct epoch_tracker et;
2126 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2127 int error;
2128
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 NET_EPOCH_ENTER(et);
2134 /* We need a Tx algorithm and at least one port */
2135 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2136 NET_EPOCH_EXIT(et);
2137 m_freem(m);
2138 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2139 return (ENXIO);
2140 }
2141
2142 ETHER_BPF_MTAP(ifp, m);
2143
2144 error = lagg_proto_start(sc, m);
2145 NET_EPOCH_EXIT(et);
2146 return (error);
2147 }
2148
2149 static int
2150 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2151 {
2152 struct epoch_tracker et;
2153 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2154 int error;
2155
2156 #if defined(KERN_TLS) || defined(RATELIMIT)
2157 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2158 MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2159 #endif
2160 NET_EPOCH_ENTER(et);
2161 /* We need a Tx algorithm and at least one port */
2162 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2163 NET_EPOCH_EXIT(et);
2164 m_freem(m);
2165 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2166 return (ENXIO);
2167 }
2168
2169 INFINIBAND_BPF_MTAP(ifp, m);
2170
2171 error = lagg_proto_start(sc, m);
2172 NET_EPOCH_EXIT(et);
2173 return (error);
2174 }
2175
2176 /*
2177 * The ifp->if_qflush entry point for lagg(4) is no-op.
2178 */
2179 static void
2180 lagg_qflush(struct ifnet *ifp __unused)
2181 {
2182 }
2183
2184 static struct mbuf *
2185 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2186 {
2187 struct epoch_tracker et;
2188 struct lagg_port *lp = ifp->if_lagg;
2189 struct lagg_softc *sc = lp->lp_softc;
2190 struct ifnet *scifp = sc->sc_ifp;
2191
2192 NET_EPOCH_ENTER(et);
2193 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2194 lp->lp_detaching != 0 ||
2195 sc->sc_proto == LAGG_PROTO_NONE) {
2196 NET_EPOCH_EXIT(et);
2197 m_freem(m);
2198 return (NULL);
2199 }
2200
2201 ETHER_BPF_MTAP(scifp, m);
2202
2203 m = lagg_proto_input(sc, lp, m);
2204 if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2205 m_freem(m);
2206 m = NULL;
2207 }
2208
2209 #ifdef DEV_NETMAP
2210 if (m != NULL && scifp->if_capenable & IFCAP_NETMAP) {
2211 scifp->if_input(scifp, m);
2212 m = NULL;
2213 }
2214 #endif /* DEV_NETMAP */
2215
2216 NET_EPOCH_EXIT(et);
2217 return (m);
2218 }
2219
2220 static struct mbuf *
2221 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2222 {
2223 struct epoch_tracker et;
2224 struct lagg_port *lp = ifp->if_lagg;
2225 struct lagg_softc *sc = lp->lp_softc;
2226 struct ifnet *scifp = sc->sc_ifp;
2227
2228 NET_EPOCH_ENTER(et);
2229 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2230 lp->lp_detaching != 0 ||
2231 sc->sc_proto == LAGG_PROTO_NONE) {
2232 NET_EPOCH_EXIT(et);
2233 m_freem(m);
2234 return (NULL);
2235 }
2236
2237 INFINIBAND_BPF_MTAP(scifp, m);
2238
2239 m = lagg_proto_input(sc, lp, m);
2240 if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2241 m_freem(m);
2242 m = NULL;
2243 }
2244
2245 NET_EPOCH_EXIT(et);
2246 return (m);
2247 }
2248
2249 static int
2250 lagg_media_change(struct ifnet *ifp)
2251 {
2252 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2253
2254 if (sc->sc_ifflags & IFF_DEBUG)
2255 printf("%s\n", __func__);
2256
2257 /* Ignore */
2258 return (0);
2259 }
2260
2261 static void
2262 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2263 {
2264 struct epoch_tracker et;
2265 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2266 struct lagg_port *lp;
2267
2268 imr->ifm_status = IFM_AVALID;
2269 imr->ifm_active = IFM_ETHER | IFM_AUTO;
2270
2271 NET_EPOCH_ENTER(et);
2272 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2273 if (LAGG_PORTACTIVE(lp))
2274 imr->ifm_status |= IFM_ACTIVE;
2275 }
2276 NET_EPOCH_EXIT(et);
2277 }
2278
2279 static void
2280 lagg_linkstate(struct lagg_softc *sc)
2281 {
2282 struct epoch_tracker et;
2283 struct lagg_port *lp;
2284 int new_link = LINK_STATE_DOWN;
2285 uint64_t speed;
2286
2287 LAGG_XLOCK_ASSERT(sc);
2288
2289 /* LACP handles link state itself */
2290 if (sc->sc_proto == LAGG_PROTO_LACP)
2291 return;
2292
2293 /* Our link is considered up if at least one of our ports is active */
2294 NET_EPOCH_ENTER(et);
2295 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2296 if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2297 new_link = LINK_STATE_UP;
2298 break;
2299 }
2300 }
2301 NET_EPOCH_EXIT(et);
2302 if_link_state_change(sc->sc_ifp, new_link);
2303
2304 /* Update if_baudrate to reflect the max possible speed */
2305 switch (sc->sc_proto) {
2306 case LAGG_PROTO_FAILOVER:
2307 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2308 sc->sc_primary->lp_ifp->if_baudrate : 0;
2309 break;
2310 case LAGG_PROTO_ROUNDROBIN:
2311 case LAGG_PROTO_LOADBALANCE:
2312 case LAGG_PROTO_BROADCAST:
2313 speed = 0;
2314 NET_EPOCH_ENTER(et);
2315 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2316 speed += lp->lp_ifp->if_baudrate;
2317 NET_EPOCH_EXIT(et);
2318 sc->sc_ifp->if_baudrate = speed;
2319 break;
2320 case LAGG_PROTO_LACP:
2321 /* LACP updates if_baudrate itself */
2322 break;
2323 }
2324 }
2325
2326 static void
2327 lagg_port_state(struct ifnet *ifp, int state)
2328 {
2329 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2330 struct lagg_softc *sc = NULL;
2331
2332 if (lp != NULL)
2333 sc = lp->lp_softc;
2334 if (sc == NULL)
2335 return;
2336
2337 LAGG_XLOCK(sc);
2338 lagg_linkstate(sc);
2339 lagg_proto_linkstate(sc, lp);
2340 LAGG_XUNLOCK(sc);
2341 }
2342
2343 struct lagg_port *
2344 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2345 {
2346 struct lagg_port *lp_next, *rval = NULL;
2347
2348 /*
2349 * Search a port which reports an active link state.
2350 */
2351
2352 #ifdef INVARIANTS
2353 /*
2354 * This is called with either in the network epoch
2355 * or with LAGG_XLOCK(sc) held.
2356 */
2357 if (!in_epoch(net_epoch_preempt))
2358 LAGG_XLOCK_ASSERT(sc);
2359 #endif
2360
2361 if (lp == NULL)
2362 goto search;
2363 if (LAGG_PORTACTIVE(lp)) {
2364 rval = lp;
2365 goto found;
2366 }
2367 if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2368 LAGG_PORTACTIVE(lp_next)) {
2369 rval = lp_next;
2370 goto found;
2371 }
2372
2373 search:
2374 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2375 if (LAGG_PORTACTIVE(lp_next)) {
2376 return (lp_next);
2377 }
2378 }
2379 found:
2380 return (rval);
2381 }
2382
2383 int
2384 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2385 {
2386
2387 #if defined(KERN_TLS) || defined(RATELIMIT)
2388 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2389 struct lagg_snd_tag *lst;
2390 struct m_snd_tag *mst;
2391
2392 mst = m->m_pkthdr.snd_tag;
2393 lst = mst_to_lst(mst);
2394 if (lst->tag->ifp != ifp) {
2395 m_freem(m);
2396 return (EAGAIN);
2397 }
2398 m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2399 m_snd_tag_rele(mst);
2400 }
2401 #endif
2402 return (ifp->if_transmit)(ifp, m);
2403 }
2404
2405 /*
2406 * Simple round robin aggregation
2407 */
2408 static void
2409 lagg_rr_attach(struct lagg_softc *sc)
2410 {
2411 sc->sc_seq = 0;
2412 sc->sc_stride = 1;
2413 }
2414
2415 static int
2416 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2417 {
2418 struct lagg_port *lp;
2419 uint32_t p;
2420
2421 p = atomic_fetchadd_32(&sc->sc_seq, 1);
2422 p /= sc->sc_stride;
2423 p %= sc->sc_count;
2424 lp = CK_SLIST_FIRST(&sc->sc_ports);
2425
2426 while (p--)
2427 lp = CK_SLIST_NEXT(lp, lp_entries);
2428
2429 /*
2430 * Check the port's link state. This will return the next active
2431 * port if the link is down or the port is NULL.
2432 */
2433 if ((lp = lagg_link_active(sc, lp)) == NULL) {
2434 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2435 m_freem(m);
2436 return (ENETDOWN);
2437 }
2438
2439 /* Send mbuf */
2440 return (lagg_enqueue(lp->lp_ifp, m));
2441 }
2442
2443 static struct mbuf *
2444 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2445 {
2446 struct ifnet *ifp = sc->sc_ifp;
2447
2448 /* Just pass in the packet to our lagg device */
2449 m->m_pkthdr.rcvif = ifp;
2450
2451 return (m);
2452 }
2453
2454 /*
2455 * Broadcast mode
2456 */
2457 static int
2458 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2459 {
2460 int errors = 0;
2461 int ret;
2462 struct lagg_port *lp, *last = NULL;
2463 struct mbuf *m0;
2464
2465 NET_EPOCH_ASSERT();
2466 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2467 if (!LAGG_PORTACTIVE(lp))
2468 continue;
2469
2470 if (last != NULL) {
2471 m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2472 if (m0 == NULL) {
2473 ret = ENOBUFS;
2474 errors++;
2475 break;
2476 }
2477 lagg_enqueue(last->lp_ifp, m0);
2478 }
2479 last = lp;
2480 }
2481
2482 if (last == NULL) {
2483 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2484 m_freem(m);
2485 return (ENOENT);
2486 }
2487 if ((last = lagg_link_active(sc, last)) == NULL) {
2488 errors++;
2489 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2490 m_freem(m);
2491 return (ENETDOWN);
2492 }
2493
2494 ret = lagg_enqueue(last->lp_ifp, m);
2495 if (errors != 0)
2496 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2497
2498 return (ret);
2499 }
2500
2501 static struct mbuf*
2502 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2503 {
2504 struct ifnet *ifp = sc->sc_ifp;
2505
2506 /* Just pass in the packet to our lagg device */
2507 m->m_pkthdr.rcvif = ifp;
2508 return (m);
2509 }
2510
2511 /*
2512 * Active failover
2513 */
2514 static int
2515 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2516 {
2517 struct lagg_port *lp;
2518
2519 /* Use the master port if active or the next available port */
2520 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2521 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2522 m_freem(m);
2523 return (ENETDOWN);
2524 }
2525
2526 /* Send mbuf */
2527 return (lagg_enqueue(lp->lp_ifp, m));
2528 }
2529
2530 static struct mbuf *
2531 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2532 {
2533 struct ifnet *ifp = sc->sc_ifp;
2534 struct lagg_port *tmp_tp;
2535
2536 if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2537 m->m_pkthdr.rcvif = ifp;
2538 return (m);
2539 }
2540
2541 if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2542 tmp_tp = lagg_link_active(sc, sc->sc_primary);
2543 /*
2544 * If tmp_tp is null, we've received a packet when all
2545 * our links are down. Weird, but process it anyways.
2546 */
2547 if ((tmp_tp == NULL || tmp_tp == lp)) {
2548 m->m_pkthdr.rcvif = ifp;
2549 return (m);
2550 }
2551 }
2552
2553 m_freem(m);
2554 return (NULL);
2555 }
2556
2557 /*
2558 * Loadbalancing
2559 */
2560 static void
2561 lagg_lb_attach(struct lagg_softc *sc)
2562 {
2563 struct lagg_port *lp;
2564 struct lagg_lb *lb;
2565
2566 LAGG_XLOCK_ASSERT(sc);
2567 lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2568 lb->lb_key = m_ether_tcpip_hash_init();
2569 sc->sc_psc = lb;
2570
2571 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2572 lagg_lb_port_create(lp);
2573 }
2574
2575 static void
2576 lagg_lb_detach(struct lagg_softc *sc)
2577 {
2578 struct lagg_lb *lb;
2579
2580 lb = (struct lagg_lb *)sc->sc_psc;
2581 if (lb != NULL)
2582 free(lb, M_LAGG);
2583 }
2584
2585 static int
2586 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2587 {
2588 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2589 struct lagg_port *lp_next;
2590 int i = 0, rv;
2591
2592 rv = 0;
2593 bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2594 LAGG_XLOCK_ASSERT(sc);
2595 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2596 if (lp_next == lp)
2597 continue;
2598 if (i >= LAGG_MAX_PORTS) {
2599 rv = EINVAL;
2600 break;
2601 }
2602 if (sc->sc_ifflags & IFF_DEBUG)
2603 printf("%s: port %s at index %d\n",
2604 sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2605 lb->lb_ports[i++] = lp_next;
2606 }
2607
2608 return (rv);
2609 }
2610
2611 static int
2612 lagg_lb_port_create(struct lagg_port *lp)
2613 {
2614 struct lagg_softc *sc = lp->lp_softc;
2615 return (lagg_lb_porttable(sc, NULL));
2616 }
2617
2618 static void
2619 lagg_lb_port_destroy(struct lagg_port *lp)
2620 {
2621 struct lagg_softc *sc = lp->lp_softc;
2622 lagg_lb_porttable(sc, lp);
2623 }
2624
2625 static int
2626 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2627 {
2628 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2629 struct lagg_port *lp = NULL;
2630 uint32_t p = 0;
2631
2632 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2633 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2634 p = m->m_pkthdr.flowid >> sc->flowid_shift;
2635 else
2636 p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2637 p %= sc->sc_count;
2638 lp = lb->lb_ports[p];
2639
2640 /*
2641 * Check the port's link state. This will return the next active
2642 * port if the link is down or the port is NULL.
2643 */
2644 if ((lp = lagg_link_active(sc, lp)) == NULL) {
2645 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2646 m_freem(m);
2647 return (ENETDOWN);
2648 }
2649
2650 /* Send mbuf */
2651 return (lagg_enqueue(lp->lp_ifp, m));
2652 }
2653
2654 static struct mbuf *
2655 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2656 {
2657 struct ifnet *ifp = sc->sc_ifp;
2658
2659 /* Just pass in the packet to our lagg device */
2660 m->m_pkthdr.rcvif = ifp;
2661
2662 return (m);
2663 }
2664
2665 /*
2666 * 802.3ad LACP
2667 */
2668 static void
2669 lagg_lacp_attach(struct lagg_softc *sc)
2670 {
2671 struct lagg_port *lp;
2672
2673 lacp_attach(sc);
2674 LAGG_XLOCK_ASSERT(sc);
2675 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2676 lacp_port_create(lp);
2677 }
2678
2679 static void
2680 lagg_lacp_detach(struct lagg_softc *sc)
2681 {
2682 struct lagg_port *lp;
2683 void *psc;
2684
2685 LAGG_XLOCK_ASSERT(sc);
2686 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2687 lacp_port_destroy(lp);
2688
2689 psc = sc->sc_psc;
2690 sc->sc_psc = NULL;
2691 lacp_detach(psc);
2692 }
2693
2694 static void
2695 lagg_lacp_lladdr(struct lagg_softc *sc)
2696 {
2697 struct lagg_port *lp;
2698
2699 LAGG_SXLOCK_ASSERT(sc);
2700
2701 /* purge all the lacp ports */
2702 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2703 lacp_port_destroy(lp);
2704
2705 /* add them back in */
2706 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2707 lacp_port_create(lp);
2708 }
2709
2710 static int
2711 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2712 {
2713 struct lagg_port *lp;
2714 int err;
2715
2716 lp = lacp_select_tx_port(sc, m, &err);
2717 if (lp == NULL) {
2718 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2719 m_freem(m);
2720 return (err);
2721 }
2722
2723 /* Send mbuf */
2724 return (lagg_enqueue(lp->lp_ifp, m));
2725 }
2726
2727 static struct mbuf *
2728 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2729 {
2730 struct ifnet *ifp = sc->sc_ifp;
2731 struct ether_header *eh;
2732 u_short etype;
2733
2734 eh = mtod(m, struct ether_header *);
2735 etype = ntohs(eh->ether_type);
2736
2737 /* Tap off LACP control messages */
2738 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2739 m = lacp_input(lp, m);
2740 if (m == NULL)
2741 return (NULL);
2742 }
2743
2744 /*
2745 * If the port is not collecting or not in the active aggregator then
2746 * free and return.
2747 */
2748 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2749 m_freem(m);
2750 return (NULL);
2751 }
2752
2753 m->m_pkthdr.rcvif = ifp;
2754 return (m);
2755 }
Cache object: f687001fced5c15091b4513096d2ba2e
|