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 *
7 * Permission to use, copy, modify, and distribute this software for any
8 * purpose with or without fee is hereby granted, provided that the above
9 * copyright notice and this permission notice appear in all copies.
10 *
11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18 */
19
20 #include <sys/cdefs.h>
21 __FBSDID("$FreeBSD: releng/8.4/sys/net/if_lagg.c 238048 2012-07-03 01:45:38Z thompsa $");
22
23 #include "opt_inet.h"
24 #include "opt_inet6.h"
25
26 #include <sys/param.h>
27 #include <sys/kernel.h>
28 #include <sys/malloc.h>
29 #include <sys/mbuf.h>
30 #include <sys/queue.h>
31 #include <sys/socket.h>
32 #include <sys/sockio.h>
33 #include <sys/sysctl.h>
34 #include <sys/module.h>
35 #include <sys/priv.h>
36 #include <sys/systm.h>
37 #include <sys/proc.h>
38 #include <sys/hash.h>
39 #include <sys/lock.h>
40 #include <sys/rwlock.h>
41 #include <sys/taskqueue.h>
42 #include <sys/eventhandler.h>
43
44 #include <net/ethernet.h>
45 #include <net/if.h>
46 #include <net/if_clone.h>
47 #include <net/if_arp.h>
48 #include <net/if_dl.h>
49 #include <net/if_llc.h>
50 #include <net/if_media.h>
51 #include <net/if_types.h>
52 #include <net/if_var.h>
53 #include <net/bpf.h>
54
55 #ifdef INET
56 #include <netinet/in.h>
57 #include <netinet/in_systm.h>
58 #include <netinet/if_ether.h>
59 #include <netinet/ip.h>
60 #endif
61
62 #ifdef INET6
63 #include <netinet/ip6.h>
64 #endif
65
66 #include <net/if_vlan_var.h>
67 #include <net/if_lagg.h>
68 #include <net/ieee8023ad_lacp.h>
69
70 /* Special flags we should propagate to the lagg ports. */
71 static struct {
72 int flag;
73 int (*func)(struct ifnet *, int);
74 } lagg_pflags[] = {
75 {IFF_PROMISC, ifpromisc},
76 {IFF_ALLMULTI, if_allmulti},
77 {0, NULL}
78 };
79
80 SLIST_HEAD(__trhead, lagg_softc) lagg_list; /* list of laggs */
81 static struct mtx lagg_list_mtx;
82 eventhandler_tag lagg_detach_cookie = NULL;
83
84 static int lagg_clone_create(struct if_clone *, int, caddr_t);
85 static void lagg_clone_destroy(struct ifnet *);
86 static void lagg_lladdr(struct lagg_softc *, uint8_t *);
87 static void lagg_capabilities(struct lagg_softc *);
88 static void lagg_port_lladdr(struct lagg_port *, uint8_t *);
89 static void lagg_port_setlladdr(void *, int);
90 static int lagg_port_create(struct lagg_softc *, struct ifnet *);
91 static int lagg_port_destroy(struct lagg_port *, int);
92 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
93 static void lagg_linkstate(struct lagg_softc *);
94 static void lagg_port_state(struct ifnet *, int);
95 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
96 static int lagg_port_output(struct ifnet *, struct mbuf *,
97 struct sockaddr *, struct route *);
98 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *);
99 static int lagg_port_checkstacking(struct lagg_softc *);
100 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *);
101 static void lagg_init(void *);
102 static void lagg_stop(struct lagg_softc *);
103 static int lagg_ioctl(struct ifnet *, u_long, caddr_t);
104 static int lagg_ether_setmulti(struct lagg_softc *);
105 static int lagg_ether_cmdmulti(struct lagg_port *, int);
106 static int lagg_setflag(struct lagg_port *, int, int,
107 int (*func)(struct ifnet *, int));
108 static int lagg_setflags(struct lagg_port *, int status);
109 static void lagg_start(struct ifnet *);
110 static int lagg_media_change(struct ifnet *);
111 static void lagg_media_status(struct ifnet *, struct ifmediareq *);
112 static struct lagg_port *lagg_link_active(struct lagg_softc *,
113 struct lagg_port *);
114 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
115
116 IFC_SIMPLE_DECLARE(lagg, 0);
117
118 /* Simple round robin */
119 static int lagg_rr_attach(struct lagg_softc *);
120 static int lagg_rr_detach(struct lagg_softc *);
121 static int lagg_rr_start(struct lagg_softc *, struct mbuf *);
122 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
123 struct mbuf *);
124
125 /* Active failover */
126 static int lagg_fail_attach(struct lagg_softc *);
127 static int lagg_fail_detach(struct lagg_softc *);
128 static int lagg_fail_start(struct lagg_softc *, struct mbuf *);
129 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
130 struct mbuf *);
131
132 /* Loadbalancing */
133 static int lagg_lb_attach(struct lagg_softc *);
134 static int lagg_lb_detach(struct lagg_softc *);
135 static int lagg_lb_port_create(struct lagg_port *);
136 static void lagg_lb_port_destroy(struct lagg_port *);
137 static int lagg_lb_start(struct lagg_softc *, struct mbuf *);
138 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
139 struct mbuf *);
140 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
141
142 /* 802.3ad LACP */
143 static int lagg_lacp_attach(struct lagg_softc *);
144 static int lagg_lacp_detach(struct lagg_softc *);
145 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *);
146 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
147 struct mbuf *);
148 static void lagg_lacp_lladdr(struct lagg_softc *);
149
150 /* lagg protocol table */
151 static const struct {
152 int ti_proto;
153 int (*ti_attach)(struct lagg_softc *);
154 } lagg_protos[] = {
155 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach },
156 { LAGG_PROTO_FAILOVER, lagg_fail_attach },
157 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach },
158 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach },
159 { LAGG_PROTO_LACP, lagg_lacp_attach },
160 { LAGG_PROTO_NONE, NULL }
161 };
162
163 SYSCTL_DECL(_net_link);
164 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, "Link Aggregation");
165
166 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */
167 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW,
168 &lagg_failover_rx_all, 0,
169 "Accept input from any interface in a failover lagg");
170 static int def_use_flowid = 1; /* Default value for using M_FLOWID */
171 TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid);
172 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RW,
173 &def_use_flowid, 0,
174 "Default setting for using flow id for load sharing");
175
176 static int
177 lagg_modevent(module_t mod, int type, void *data)
178 {
179
180 switch (type) {
181 case MOD_LOAD:
182 mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
183 SLIST_INIT(&lagg_list);
184 if_clone_attach(&lagg_cloner);
185 lagg_input_p = lagg_input;
186 lagg_linkstate_p = lagg_port_state;
187 lagg_detach_cookie = EVENTHANDLER_REGISTER(
188 ifnet_departure_event, lagg_port_ifdetach, NULL,
189 EVENTHANDLER_PRI_ANY);
190 break;
191 case MOD_UNLOAD:
192 EVENTHANDLER_DEREGISTER(ifnet_departure_event,
193 lagg_detach_cookie);
194 if_clone_detach(&lagg_cloner);
195 lagg_input_p = NULL;
196 lagg_linkstate_p = NULL;
197 mtx_destroy(&lagg_list_mtx);
198 break;
199 default:
200 return (EOPNOTSUPP);
201 }
202 return (0);
203 }
204
205 static moduledata_t lagg_mod = {
206 "if_lagg",
207 lagg_modevent,
208 0
209 };
210
211 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
212 MODULE_VERSION(if_lagg, 1);
213
214 #if __FreeBSD_version >= 800000
215 /*
216 * This routine is run via an vlan
217 * config EVENT
218 */
219 static void
220 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
221 {
222 struct lagg_softc *sc = ifp->if_softc;
223 struct lagg_port *lp;
224
225 if (ifp->if_softc != arg) /* Not our event */
226 return;
227
228 LAGG_RLOCK(sc);
229 if (!SLIST_EMPTY(&sc->sc_ports)) {
230 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
231 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
232 }
233 LAGG_RUNLOCK(sc);
234 }
235
236 /*
237 * This routine is run via an vlan
238 * unconfig EVENT
239 */
240 static void
241 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
242 {
243 struct lagg_softc *sc = ifp->if_softc;
244 struct lagg_port *lp;
245
246 if (ifp->if_softc != arg) /* Not our event */
247 return;
248
249 LAGG_RLOCK(sc);
250 if (!SLIST_EMPTY(&sc->sc_ports)) {
251 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
252 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
253 }
254 LAGG_RUNLOCK(sc);
255 }
256 #endif
257
258 static int
259 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
260 {
261 struct lagg_softc *sc;
262 struct ifnet *ifp;
263 int i, error = 0;
264 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */
265 struct sysctl_oid *oid;
266 char num[14]; /* sufficient for 32 bits */
267
268 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
269 ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
270 if (ifp == NULL) {
271 free(sc, M_DEVBUF);
272 return (ENOSPC);
273 }
274
275 sysctl_ctx_init(&sc->ctx);
276 snprintf(num, sizeof(num), "%u", unit);
277 sc->use_flowid = def_use_flowid;
278 oid = SYSCTL_ADD_NODE(&sc->ctx, &SYSCTL_NODE_CHILDREN(_net_link, lagg),
279 OID_AUTO, num, CTLFLAG_RD, NULL, "");
280 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
281 "use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid, sc->use_flowid,
282 "Use flow id for load sharing");
283 /* Hash all layers by default */
284 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4;
285
286 sc->sc_proto = LAGG_PROTO_NONE;
287 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
288 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
289 sc->sc_proto = lagg_protos[i].ti_proto;
290 if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
291 if_free_type(ifp, IFT_ETHER);
292 free(sc, M_DEVBUF);
293 return (error);
294 }
295 break;
296 }
297 }
298 LAGG_LOCK_INIT(sc);
299 SLIST_INIT(&sc->sc_ports);
300 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
301
302 /* Initialise pseudo media types */
303 ifmedia_init(&sc->sc_media, 0, lagg_media_change,
304 lagg_media_status);
305 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
306 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
307
308 if_initname(ifp, ifc->ifc_name, unit);
309 ifp->if_type = IFT_ETHER;
310 ifp->if_softc = sc;
311 ifp->if_start = lagg_start;
312 ifp->if_init = lagg_init;
313 ifp->if_ioctl = lagg_ioctl;
314 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
315
316 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
317 ifp->if_snd.ifq_drv_maxlen = ifqmaxlen;
318 IFQ_SET_READY(&ifp->if_snd);
319
320 /*
321 * Attach as an ordinary ethernet device, childs will be attached
322 * as special device IFT_IEEE8023ADLAG.
323 */
324 ether_ifattach(ifp, eaddr);
325
326 #if __FreeBSD_version >= 800000
327 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
328 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
329 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
330 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
331 #endif
332
333 /* Insert into the global list of laggs */
334 mtx_lock(&lagg_list_mtx);
335 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
336 mtx_unlock(&lagg_list_mtx);
337
338 return (0);
339 }
340
341 static void
342 lagg_clone_destroy(struct ifnet *ifp)
343 {
344 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
345 struct lagg_port *lp;
346
347 LAGG_WLOCK(sc);
348
349 lagg_stop(sc);
350 ifp->if_flags &= ~IFF_UP;
351
352 #if __FreeBSD_version >= 800000
353 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
354 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
355 #endif
356
357 /* Shutdown and remove lagg ports */
358 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
359 lagg_port_destroy(lp, 1);
360 /* Unhook the aggregation protocol */
361 (*sc->sc_detach)(sc);
362
363 LAGG_WUNLOCK(sc);
364
365 sysctl_ctx_free(&sc->ctx);
366 ifmedia_removeall(&sc->sc_media);
367 ether_ifdetach(ifp);
368 if_free_type(ifp, IFT_ETHER);
369
370 mtx_lock(&lagg_list_mtx);
371 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
372 mtx_unlock(&lagg_list_mtx);
373
374 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
375 LAGG_LOCK_DESTROY(sc);
376 free(sc, M_DEVBUF);
377 }
378
379 static void
380 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
381 {
382 struct ifnet *ifp = sc->sc_ifp;
383
384 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
385 return;
386
387 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
388 /* Let the protocol know the MAC has changed */
389 if (sc->sc_lladdr != NULL)
390 (*sc->sc_lladdr)(sc);
391 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
392 }
393
394 static void
395 lagg_capabilities(struct lagg_softc *sc)
396 {
397 struct lagg_port *lp;
398 int cap = ~0, ena = ~0;
399 u_long hwa = ~0UL;
400
401 LAGG_WLOCK_ASSERT(sc);
402
403 /* Get capabilities from the lagg ports */
404 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
405 cap &= lp->lp_ifp->if_capabilities;
406 ena &= lp->lp_ifp->if_capenable;
407 hwa &= lp->lp_ifp->if_hwassist;
408 }
409 cap = (cap == ~0 ? 0 : cap);
410 ena = (ena == ~0 ? 0 : ena);
411 hwa = (hwa == ~0 ? 0 : hwa);
412
413 if (sc->sc_ifp->if_capabilities != cap ||
414 sc->sc_ifp->if_capenable != ena ||
415 sc->sc_ifp->if_hwassist != hwa) {
416 sc->sc_ifp->if_capabilities = cap;
417 sc->sc_ifp->if_capenable = ena;
418 sc->sc_ifp->if_hwassist = hwa;
419 getmicrotime(&sc->sc_ifp->if_lastchange);
420
421 if (sc->sc_ifflags & IFF_DEBUG)
422 if_printf(sc->sc_ifp,
423 "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
424 }
425 }
426
427 static void
428 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
429 {
430 struct lagg_softc *sc = lp->lp_softc;
431 struct ifnet *ifp = lp->lp_ifp;
432 struct lagg_llq *llq;
433 int pending = 0;
434
435 LAGG_WLOCK_ASSERT(sc);
436
437 if (lp->lp_detaching ||
438 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
439 return;
440
441 /* Check to make sure its not already queued to be changed */
442 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
443 if (llq->llq_ifp == ifp) {
444 pending = 1;
445 break;
446 }
447 }
448
449 if (!pending) {
450 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
451 if (llq == NULL) /* XXX what to do */
452 return;
453 }
454
455 /* Update the lladdr even if pending, it may have changed */
456 llq->llq_ifp = ifp;
457 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
458
459 if (!pending)
460 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
461
462 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
463 }
464
465 /*
466 * Set the interface MAC address from a taskqueue to avoid a LOR.
467 */
468 static void
469 lagg_port_setlladdr(void *arg, int pending)
470 {
471 struct lagg_softc *sc = (struct lagg_softc *)arg;
472 struct lagg_llq *llq, *head;
473 struct ifnet *ifp;
474 int error;
475
476 /* Grab a local reference of the queue and remove it from the softc */
477 LAGG_WLOCK(sc);
478 head = SLIST_FIRST(&sc->sc_llq_head);
479 SLIST_FIRST(&sc->sc_llq_head) = NULL;
480 LAGG_WUNLOCK(sc);
481
482 /*
483 * Traverse the queue and set the lladdr on each ifp. It is safe to do
484 * unlocked as we have the only reference to it.
485 */
486 for (llq = head; llq != NULL; llq = head) {
487 ifp = llq->llq_ifp;
488
489 /* Set the link layer address */
490 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
491 if (error)
492 printf("%s: setlladdr failed on %s\n", __func__,
493 ifp->if_xname);
494
495 head = SLIST_NEXT(llq, llq_entries);
496 free(llq, M_DEVBUF);
497 }
498 }
499
500 static int
501 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
502 {
503 struct lagg_softc *sc_ptr;
504 struct lagg_port *lp;
505 int error = 0;
506
507 LAGG_WLOCK_ASSERT(sc);
508
509 /* Limit the maximal number of lagg ports */
510 if (sc->sc_count >= LAGG_MAX_PORTS)
511 return (ENOSPC);
512
513 /* Check if port has already been associated to a lagg */
514 if (ifp->if_lagg != NULL)
515 return (EBUSY);
516
517 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */
518 if (ifp->if_type != IFT_ETHER)
519 return (EPROTONOSUPPORT);
520
521 /* Allow the first Ethernet member to define the MTU */
522 if (SLIST_EMPTY(&sc->sc_ports))
523 sc->sc_ifp->if_mtu = ifp->if_mtu;
524 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
525 if_printf(sc->sc_ifp, "invalid MTU for %s\n",
526 ifp->if_xname);
527 return (EINVAL);
528 }
529
530 if ((lp = malloc(sizeof(struct lagg_port),
531 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
532 return (ENOMEM);
533
534 /* Check if port is a stacked lagg */
535 mtx_lock(&lagg_list_mtx);
536 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
537 if (ifp == sc_ptr->sc_ifp) {
538 mtx_unlock(&lagg_list_mtx);
539 free(lp, M_DEVBUF);
540 return (EINVAL);
541 /* XXX disable stacking for the moment, its untested
542 lp->lp_flags |= LAGG_PORT_STACK;
543 if (lagg_port_checkstacking(sc_ptr) >=
544 LAGG_MAX_STACKING) {
545 mtx_unlock(&lagg_list_mtx);
546 free(lp, M_DEVBUF);
547 return (E2BIG);
548 }
549 */
550 }
551 }
552 mtx_unlock(&lagg_list_mtx);
553
554 /* Change the interface type */
555 lp->lp_iftype = ifp->if_type;
556 ifp->if_type = IFT_IEEE8023ADLAG;
557 ifp->if_lagg = lp;
558 lp->lp_ioctl = ifp->if_ioctl;
559 ifp->if_ioctl = lagg_port_ioctl;
560 lp->lp_output = ifp->if_output;
561 ifp->if_output = lagg_port_output;
562
563 lp->lp_ifp = ifp;
564 lp->lp_softc = sc;
565
566 /* Save port link layer address */
567 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
568
569 if (SLIST_EMPTY(&sc->sc_ports)) {
570 sc->sc_primary = lp;
571 lagg_lladdr(sc, IF_LLADDR(ifp));
572 } else {
573 /* Update link layer address for this port */
574 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
575 }
576
577 /* Insert into the list of ports */
578 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
579 sc->sc_count++;
580
581 /* Update lagg capabilities */
582 lagg_capabilities(sc);
583 lagg_linkstate(sc);
584
585 /* Add multicast addresses and interface flags to this port */
586 lagg_ether_cmdmulti(lp, 1);
587 lagg_setflags(lp, 1);
588
589 if (sc->sc_port_create != NULL)
590 error = (*sc->sc_port_create)(lp);
591 if (error) {
592 /* remove the port again, without calling sc_port_destroy */
593 lagg_port_destroy(lp, 0);
594 return (error);
595 }
596
597 return (error);
598 }
599
600 static int
601 lagg_port_checkstacking(struct lagg_softc *sc)
602 {
603 struct lagg_softc *sc_ptr;
604 struct lagg_port *lp;
605 int m = 0;
606
607 LAGG_WLOCK_ASSERT(sc);
608
609 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
610 if (lp->lp_flags & LAGG_PORT_STACK) {
611 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
612 m = MAX(m, lagg_port_checkstacking(sc_ptr));
613 }
614 }
615
616 return (m + 1);
617 }
618
619 static int
620 lagg_port_destroy(struct lagg_port *lp, int runpd)
621 {
622 struct lagg_softc *sc = lp->lp_softc;
623 struct lagg_port *lp_ptr;
624 struct lagg_llq *llq;
625 struct ifnet *ifp = lp->lp_ifp;
626
627 LAGG_WLOCK_ASSERT(sc);
628
629 if (runpd && sc->sc_port_destroy != NULL)
630 (*sc->sc_port_destroy)(lp);
631
632 /*
633 * Remove multicast addresses and interface flags from this port and
634 * reset the MAC address, skip if the interface is being detached.
635 */
636 if (!lp->lp_detaching) {
637 lagg_ether_cmdmulti(lp, 0);
638 lagg_setflags(lp, 0);
639 lagg_port_lladdr(lp, lp->lp_lladdr);
640 }
641
642 /* Restore interface */
643 ifp->if_type = lp->lp_iftype;
644 ifp->if_ioctl = lp->lp_ioctl;
645 ifp->if_output = lp->lp_output;
646 ifp->if_lagg = NULL;
647
648 /* Finally, remove the port from the lagg */
649 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
650 sc->sc_count--;
651
652 /* Update the primary interface */
653 if (lp == sc->sc_primary) {
654 uint8_t lladdr[ETHER_ADDR_LEN];
655
656 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
657 bzero(&lladdr, ETHER_ADDR_LEN);
658 } else {
659 bcopy(lp_ptr->lp_lladdr,
660 lladdr, ETHER_ADDR_LEN);
661 }
662 lagg_lladdr(sc, lladdr);
663 sc->sc_primary = lp_ptr;
664
665 /* Update link layer address for each port */
666 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
667 lagg_port_lladdr(lp_ptr, lladdr);
668 }
669
670 /* Remove any pending lladdr changes from the queue */
671 if (lp->lp_detaching) {
672 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
673 if (llq->llq_ifp == ifp) {
674 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
675 llq_entries);
676 free(llq, M_DEVBUF);
677 break; /* Only appears once */
678 }
679 }
680 }
681
682 if (lp->lp_ifflags)
683 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
684
685 free(lp, M_DEVBUF);
686
687 /* Update lagg capabilities */
688 lagg_capabilities(sc);
689 lagg_linkstate(sc);
690
691 return (0);
692 }
693
694 static int
695 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
696 {
697 struct lagg_reqport *rp = (struct lagg_reqport *)data;
698 struct lagg_softc *sc;
699 struct lagg_port *lp = NULL;
700 int error = 0;
701
702 /* Should be checked by the caller */
703 if (ifp->if_type != IFT_IEEE8023ADLAG ||
704 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
705 goto fallback;
706
707 switch (cmd) {
708 case SIOCGLAGGPORT:
709 if (rp->rp_portname[0] == '\0' ||
710 ifunit(rp->rp_portname) != ifp) {
711 error = EINVAL;
712 break;
713 }
714
715 LAGG_RLOCK(sc);
716 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
717 error = ENOENT;
718 LAGG_RUNLOCK(sc);
719 break;
720 }
721
722 lagg_port2req(lp, rp);
723 LAGG_RUNLOCK(sc);
724 break;
725
726 case SIOCSIFCAP:
727 if (lp->lp_ioctl == NULL) {
728 error = EINVAL;
729 break;
730 }
731 error = (*lp->lp_ioctl)(ifp, cmd, data);
732 if (error)
733 break;
734
735 /* Update lagg interface capabilities */
736 LAGG_WLOCK(sc);
737 lagg_capabilities(sc);
738 LAGG_WUNLOCK(sc);
739 break;
740
741 case SIOCSIFMTU:
742 /* Do not allow the MTU to be changed once joined */
743 error = EINVAL;
744 break;
745
746 default:
747 goto fallback;
748 }
749
750 return (error);
751
752 fallback:
753 if (lp->lp_ioctl != NULL)
754 return ((*lp->lp_ioctl)(ifp, cmd, data));
755
756 return (EINVAL);
757 }
758
759 /*
760 * For direct output to child ports.
761 */
762 static int
763 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
764 struct sockaddr *dst, struct route *ro)
765 {
766 struct lagg_port *lp = ifp->if_lagg;
767
768 switch (dst->sa_family) {
769 case pseudo_AF_HDRCMPLT:
770 case AF_UNSPEC:
771 return ((*lp->lp_output)(ifp, m, dst, ro));
772 }
773
774 /* drop any other frames */
775 m_freem(m);
776 return (EBUSY);
777 }
778
779 static void
780 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
781 {
782 struct lagg_port *lp;
783 struct lagg_softc *sc;
784
785 if ((lp = ifp->if_lagg) == NULL)
786 return;
787 /* If the ifnet is just being renamed, don't do anything. */
788 if (ifp->if_flags & IFF_RENAMING)
789 return;
790
791 sc = lp->lp_softc;
792
793 LAGG_WLOCK(sc);
794 lp->lp_detaching = 1;
795 lagg_port_destroy(lp, 1);
796 LAGG_WUNLOCK(sc);
797 }
798
799 static void
800 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
801 {
802 struct lagg_softc *sc = lp->lp_softc;
803
804 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
805 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
806 rp->rp_prio = lp->lp_prio;
807 rp->rp_flags = lp->lp_flags;
808 if (sc->sc_portreq != NULL)
809 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
810
811 /* Add protocol specific flags */
812 switch (sc->sc_proto) {
813 case LAGG_PROTO_FAILOVER:
814 if (lp == sc->sc_primary)
815 rp->rp_flags |= LAGG_PORT_MASTER;
816 if (lp == lagg_link_active(sc, sc->sc_primary))
817 rp->rp_flags |= LAGG_PORT_ACTIVE;
818 break;
819
820 case LAGG_PROTO_ROUNDROBIN:
821 case LAGG_PROTO_LOADBALANCE:
822 case LAGG_PROTO_ETHERCHANNEL:
823 if (LAGG_PORTACTIVE(lp))
824 rp->rp_flags |= LAGG_PORT_ACTIVE;
825 break;
826
827 case LAGG_PROTO_LACP:
828 /* LACP has a different definition of active */
829 if (lacp_isactive(lp))
830 rp->rp_flags |= LAGG_PORT_ACTIVE;
831 if (lacp_iscollecting(lp))
832 rp->rp_flags |= LAGG_PORT_COLLECTING;
833 if (lacp_isdistributing(lp))
834 rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
835 break;
836 }
837
838 }
839
840 static void
841 lagg_init(void *xsc)
842 {
843 struct lagg_softc *sc = (struct lagg_softc *)xsc;
844 struct lagg_port *lp;
845 struct ifnet *ifp = sc->sc_ifp;
846
847 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
848 return;
849
850 LAGG_WLOCK(sc);
851
852 ifp->if_drv_flags |= IFF_DRV_RUNNING;
853 /* Update the port lladdrs */
854 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
855 lagg_port_lladdr(lp, IF_LLADDR(ifp));
856
857 if (sc->sc_init != NULL)
858 (*sc->sc_init)(sc);
859
860 LAGG_WUNLOCK(sc);
861 }
862
863 static void
864 lagg_stop(struct lagg_softc *sc)
865 {
866 struct ifnet *ifp = sc->sc_ifp;
867
868 LAGG_WLOCK_ASSERT(sc);
869
870 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
871 return;
872
873 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
874
875 if (sc->sc_stop != NULL)
876 (*sc->sc_stop)(sc);
877 }
878
879 static int
880 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
881 {
882 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
883 struct lagg_reqall *ra = (struct lagg_reqall *)data;
884 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
885 struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
886 struct ifreq *ifr = (struct ifreq *)data;
887 struct lagg_port *lp;
888 struct ifnet *tpif;
889 struct thread *td = curthread;
890 char *buf, *outbuf;
891 int count, buflen, len, error = 0;
892
893 bzero(&rpbuf, sizeof(rpbuf));
894
895 switch (cmd) {
896 case SIOCGLAGG:
897 LAGG_RLOCK(sc);
898 count = 0;
899 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
900 count++;
901 buflen = count * sizeof(struct lagg_reqport);
902 LAGG_RUNLOCK(sc);
903
904 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
905
906 LAGG_RLOCK(sc);
907 ra->ra_proto = sc->sc_proto;
908 if (sc->sc_req != NULL)
909 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
910
911 count = 0;
912 buf = outbuf;
913 len = min(ra->ra_size, buflen);
914 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
915 if (len < sizeof(rpbuf))
916 break;
917
918 lagg_port2req(lp, &rpbuf);
919 memcpy(buf, &rpbuf, sizeof(rpbuf));
920 count++;
921 buf += sizeof(rpbuf);
922 len -= sizeof(rpbuf);
923 }
924 LAGG_RUNLOCK(sc);
925 ra->ra_ports = count;
926 ra->ra_size = count * sizeof(rpbuf);
927 error = copyout(outbuf, ra->ra_port, ra->ra_size);
928 free(outbuf, M_TEMP);
929 break;
930 case SIOCSLAGG:
931 error = priv_check(td, PRIV_NET_LAGG);
932 if (error)
933 break;
934 if (ra->ra_proto >= LAGG_PROTO_MAX) {
935 error = EPROTONOSUPPORT;
936 break;
937 }
938 LAGG_WLOCK(sc);
939 if (sc->sc_proto != LAGG_PROTO_NONE) {
940 /* Reset protocol first in case detach unlocks */
941 sc->sc_proto = LAGG_PROTO_NONE;
942 error = sc->sc_detach(sc);
943 sc->sc_detach = NULL;
944 sc->sc_start = NULL;
945 sc->sc_input = NULL;
946 sc->sc_port_create = NULL;
947 sc->sc_port_destroy = NULL;
948 sc->sc_linkstate = NULL;
949 sc->sc_init = NULL;
950 sc->sc_stop = NULL;
951 sc->sc_lladdr = NULL;
952 sc->sc_req = NULL;
953 sc->sc_portreq = NULL;
954 } else if (sc->sc_input != NULL) {
955 /* Still detaching */
956 error = EBUSY;
957 }
958 if (error != 0) {
959 LAGG_WUNLOCK(sc);
960 break;
961 }
962 for (int i = 0; i < (sizeof(lagg_protos) /
963 sizeof(lagg_protos[0])); i++) {
964 if (lagg_protos[i].ti_proto == ra->ra_proto) {
965 if (sc->sc_ifflags & IFF_DEBUG)
966 printf("%s: using proto %u\n",
967 sc->sc_ifname,
968 lagg_protos[i].ti_proto);
969 sc->sc_proto = lagg_protos[i].ti_proto;
970 if (sc->sc_proto != LAGG_PROTO_NONE)
971 error = lagg_protos[i].ti_attach(sc);
972 LAGG_WUNLOCK(sc);
973 return (error);
974 }
975 }
976 LAGG_WUNLOCK(sc);
977 error = EPROTONOSUPPORT;
978 break;
979 case SIOCGLAGGFLAGS:
980 rf->rf_flags = sc->sc_flags;
981 break;
982 case SIOCSLAGGHASH:
983 error = priv_check(td, PRIV_NET_LAGG);
984 if (error)
985 break;
986 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
987 error = EINVAL;
988 break;
989 }
990 LAGG_WLOCK(sc);
991 sc->sc_flags &= ~LAGG_F_HASHMASK;
992 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK;
993 LAGG_WUNLOCK(sc);
994 break;
995 case SIOCGLAGGPORT:
996 if (rp->rp_portname[0] == '\0' ||
997 (tpif = ifunit(rp->rp_portname)) == NULL) {
998 error = EINVAL;
999 break;
1000 }
1001
1002 LAGG_RLOCK(sc);
1003 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1004 lp->lp_softc != sc) {
1005 error = ENOENT;
1006 LAGG_RUNLOCK(sc);
1007 break;
1008 }
1009
1010 lagg_port2req(lp, rp);
1011 LAGG_RUNLOCK(sc);
1012 break;
1013 case SIOCSLAGGPORT:
1014 error = priv_check(td, PRIV_NET_LAGG);
1015 if (error)
1016 break;
1017 if (rp->rp_portname[0] == '\0' ||
1018 (tpif = ifunit(rp->rp_portname)) == NULL) {
1019 error = EINVAL;
1020 break;
1021 }
1022 LAGG_WLOCK(sc);
1023 error = lagg_port_create(sc, tpif);
1024 LAGG_WUNLOCK(sc);
1025 break;
1026 case SIOCSLAGGDELPORT:
1027 error = priv_check(td, PRIV_NET_LAGG);
1028 if (error)
1029 break;
1030 if (rp->rp_portname[0] == '\0' ||
1031 (tpif = ifunit(rp->rp_portname)) == NULL) {
1032 error = EINVAL;
1033 break;
1034 }
1035
1036 LAGG_WLOCK(sc);
1037 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1038 lp->lp_softc != sc) {
1039 error = ENOENT;
1040 LAGG_WUNLOCK(sc);
1041 break;
1042 }
1043
1044 error = lagg_port_destroy(lp, 1);
1045 LAGG_WUNLOCK(sc);
1046 break;
1047 case SIOCSIFFLAGS:
1048 /* Set flags on ports too */
1049 LAGG_WLOCK(sc);
1050 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1051 lagg_setflags(lp, 1);
1052 }
1053 LAGG_WUNLOCK(sc);
1054
1055 if (!(ifp->if_flags & IFF_UP) &&
1056 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1057 /*
1058 * If interface is marked down and it is running,
1059 * then stop and disable it.
1060 */
1061 LAGG_WLOCK(sc);
1062 lagg_stop(sc);
1063 LAGG_WUNLOCK(sc);
1064 } else if ((ifp->if_flags & IFF_UP) &&
1065 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1066 /*
1067 * If interface is marked up and it is stopped, then
1068 * start it.
1069 */
1070 (*ifp->if_init)(sc);
1071 }
1072 break;
1073 case SIOCADDMULTI:
1074 case SIOCDELMULTI:
1075 LAGG_WLOCK(sc);
1076 error = lagg_ether_setmulti(sc);
1077 LAGG_WUNLOCK(sc);
1078 break;
1079 case SIOCSIFMEDIA:
1080 case SIOCGIFMEDIA:
1081 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1082 break;
1083
1084 case SIOCSIFCAP:
1085 case SIOCSIFMTU:
1086 /* Do not allow the MTU or caps to be directly changed */
1087 error = EINVAL;
1088 break;
1089
1090 default:
1091 error = ether_ioctl(ifp, cmd, data);
1092 break;
1093 }
1094 return (error);
1095 }
1096
1097 static int
1098 lagg_ether_setmulti(struct lagg_softc *sc)
1099 {
1100 struct lagg_port *lp;
1101
1102 LAGG_WLOCK_ASSERT(sc);
1103
1104 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1105 /* First, remove any existing filter entries. */
1106 lagg_ether_cmdmulti(lp, 0);
1107 /* copy all addresses from the lagg interface to the port */
1108 lagg_ether_cmdmulti(lp, 1);
1109 }
1110 return (0);
1111 }
1112
1113 static int
1114 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1115 {
1116 struct lagg_softc *sc = lp->lp_softc;
1117 struct ifnet *ifp = lp->lp_ifp;
1118 struct ifnet *scifp = sc->sc_ifp;
1119 struct lagg_mc *mc;
1120 struct ifmultiaddr *ifma, *rifma = NULL;
1121 struct sockaddr_dl sdl;
1122 int error;
1123
1124 LAGG_WLOCK_ASSERT(sc);
1125
1126 bzero((char *)&sdl, sizeof(sdl));
1127 sdl.sdl_len = sizeof(sdl);
1128 sdl.sdl_family = AF_LINK;
1129 sdl.sdl_type = IFT_ETHER;
1130 sdl.sdl_alen = ETHER_ADDR_LEN;
1131 sdl.sdl_index = ifp->if_index;
1132
1133 if (set) {
1134 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1135 if (ifma->ifma_addr->sa_family != AF_LINK)
1136 continue;
1137 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1138 LLADDR(&sdl), ETHER_ADDR_LEN);
1139
1140 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1141 if (error)
1142 return (error);
1143 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1144 if (mc == NULL)
1145 return (ENOMEM);
1146 mc->mc_ifma = rifma;
1147 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1148 }
1149 } else {
1150 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1151 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1152 if_delmulti_ifma(mc->mc_ifma);
1153 free(mc, M_DEVBUF);
1154 }
1155 }
1156 return (0);
1157 }
1158
1159 /* Handle a ref counted flag that should be set on the lagg port as well */
1160 static int
1161 lagg_setflag(struct lagg_port *lp, int flag, int status,
1162 int (*func)(struct ifnet *, int))
1163 {
1164 struct lagg_softc *sc = lp->lp_softc;
1165 struct ifnet *scifp = sc->sc_ifp;
1166 struct ifnet *ifp = lp->lp_ifp;
1167 int error;
1168
1169 LAGG_WLOCK_ASSERT(sc);
1170
1171 status = status ? (scifp->if_flags & flag) : 0;
1172 /* Now "status" contains the flag value or 0 */
1173
1174 /*
1175 * See if recorded ports status is different from what
1176 * we want it to be. If it is, flip it. We record ports
1177 * status in lp_ifflags so that we won't clear ports flag
1178 * we haven't set. In fact, we don't clear or set ports
1179 * flags directly, but get or release references to them.
1180 * That's why we can be sure that recorded flags still are
1181 * in accord with actual ports flags.
1182 */
1183 if (status != (lp->lp_ifflags & flag)) {
1184 error = (*func)(ifp, status);
1185 if (error)
1186 return (error);
1187 lp->lp_ifflags &= ~flag;
1188 lp->lp_ifflags |= status;
1189 }
1190 return (0);
1191 }
1192
1193 /*
1194 * Handle IFF_* flags that require certain changes on the lagg port
1195 * if "status" is true, update ports flags respective to the lagg
1196 * if "status" is false, forcedly clear the flags set on port.
1197 */
1198 static int
1199 lagg_setflags(struct lagg_port *lp, int status)
1200 {
1201 int error, i;
1202
1203 for (i = 0; lagg_pflags[i].flag; i++) {
1204 error = lagg_setflag(lp, lagg_pflags[i].flag,
1205 status, lagg_pflags[i].func);
1206 if (error)
1207 return (error);
1208 }
1209 return (0);
1210 }
1211
1212 static void
1213 lagg_start(struct ifnet *ifp)
1214 {
1215 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1216 struct mbuf *m;
1217 int error = 0;
1218
1219 LAGG_RLOCK(sc);
1220 /* We need a Tx algorithm and at least one port */
1221 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1222 IF_DRAIN(&ifp->if_snd);
1223 LAGG_RUNLOCK(sc);
1224 return;
1225 }
1226
1227 for (;; error = 0) {
1228 IFQ_DEQUEUE(&ifp->if_snd, m);
1229 if (m == NULL)
1230 break;
1231
1232 ETHER_BPF_MTAP(ifp, m);
1233
1234 error = (*sc->sc_start)(sc, m);
1235 if (error == 0)
1236 ifp->if_opackets++;
1237 else
1238 ifp->if_oerrors++;
1239 }
1240 LAGG_RUNLOCK(sc);
1241 }
1242
1243 static struct mbuf *
1244 lagg_input(struct ifnet *ifp, struct mbuf *m)
1245 {
1246 struct lagg_port *lp = ifp->if_lagg;
1247 struct lagg_softc *sc = lp->lp_softc;
1248 struct ifnet *scifp = sc->sc_ifp;
1249
1250 LAGG_RLOCK(sc);
1251 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1252 (lp->lp_flags & LAGG_PORT_DISABLED) ||
1253 sc->sc_proto == LAGG_PROTO_NONE) {
1254 LAGG_RUNLOCK(sc);
1255 m_freem(m);
1256 return (NULL);
1257 }
1258
1259 ETHER_BPF_MTAP(scifp, m);
1260
1261 m = (*sc->sc_input)(sc, lp, m);
1262
1263 if (m != NULL) {
1264 scifp->if_ipackets++;
1265 scifp->if_ibytes += m->m_pkthdr.len;
1266
1267 if (scifp->if_flags & IFF_MONITOR) {
1268 m_freem(m);
1269 m = NULL;
1270 }
1271 }
1272
1273 LAGG_RUNLOCK(sc);
1274 return (m);
1275 }
1276
1277 static int
1278 lagg_media_change(struct ifnet *ifp)
1279 {
1280 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1281
1282 if (sc->sc_ifflags & IFF_DEBUG)
1283 printf("%s\n", __func__);
1284
1285 /* Ignore */
1286 return (0);
1287 }
1288
1289 static void
1290 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1291 {
1292 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1293 struct lagg_port *lp;
1294
1295 imr->ifm_status = IFM_AVALID;
1296 imr->ifm_active = IFM_ETHER | IFM_AUTO;
1297
1298 LAGG_RLOCK(sc);
1299 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1300 if (LAGG_PORTACTIVE(lp))
1301 imr->ifm_status |= IFM_ACTIVE;
1302 }
1303 LAGG_RUNLOCK(sc);
1304 }
1305
1306 static void
1307 lagg_linkstate(struct lagg_softc *sc)
1308 {
1309 struct lagg_port *lp;
1310 int new_link = LINK_STATE_DOWN;
1311 uint64_t speed;
1312
1313 /* Our link is considered up if at least one of our ports is active */
1314 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1315 if (lp->lp_link_state == LINK_STATE_UP) {
1316 new_link = LINK_STATE_UP;
1317 break;
1318 }
1319 }
1320 if_link_state_change(sc->sc_ifp, new_link);
1321
1322 /* Update if_baudrate to reflect the max possible speed */
1323 switch (sc->sc_proto) {
1324 case LAGG_PROTO_FAILOVER:
1325 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1326 sc->sc_primary->lp_ifp->if_baudrate : 0;
1327 break;
1328 case LAGG_PROTO_ROUNDROBIN:
1329 case LAGG_PROTO_LOADBALANCE:
1330 case LAGG_PROTO_ETHERCHANNEL:
1331 speed = 0;
1332 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1333 speed += lp->lp_ifp->if_baudrate;
1334 sc->sc_ifp->if_baudrate = speed;
1335 break;
1336 case LAGG_PROTO_LACP:
1337 /* LACP updates if_baudrate itself */
1338 break;
1339 }
1340 }
1341
1342 static void
1343 lagg_port_state(struct ifnet *ifp, int state)
1344 {
1345 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1346 struct lagg_softc *sc = NULL;
1347
1348 if (lp != NULL)
1349 sc = lp->lp_softc;
1350 if (sc == NULL)
1351 return;
1352
1353 LAGG_WLOCK(sc);
1354 lagg_linkstate(sc);
1355 if (sc->sc_linkstate != NULL)
1356 (*sc->sc_linkstate)(lp);
1357 LAGG_WUNLOCK(sc);
1358 }
1359
1360 struct lagg_port *
1361 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1362 {
1363 struct lagg_port *lp_next, *rval = NULL;
1364 // int new_link = LINK_STATE_DOWN;
1365
1366 LAGG_RLOCK_ASSERT(sc);
1367 /*
1368 * Search a port which reports an active link state.
1369 */
1370
1371 if (lp == NULL)
1372 goto search;
1373 if (LAGG_PORTACTIVE(lp)) {
1374 rval = lp;
1375 goto found;
1376 }
1377 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1378 LAGG_PORTACTIVE(lp_next)) {
1379 rval = lp_next;
1380 goto found;
1381 }
1382
1383 search:
1384 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1385 if (LAGG_PORTACTIVE(lp_next)) {
1386 rval = lp_next;
1387 goto found;
1388 }
1389 }
1390
1391 found:
1392 if (rval != NULL) {
1393 /*
1394 * The IEEE 802.1D standard assumes that a lagg with
1395 * multiple ports is always full duplex. This is valid
1396 * for load sharing laggs and if at least two links
1397 * are active. Unfortunately, checking the latter would
1398 * be too expensive at this point.
1399 XXX
1400 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1401 (sc->sc_count > 1))
1402 new_link = LINK_STATE_FULL_DUPLEX;
1403 else
1404 new_link = rval->lp_link_state;
1405 */
1406 }
1407
1408 return (rval);
1409 }
1410
1411 static const void *
1412 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1413 {
1414 if (m->m_pkthdr.len < (off + len)) {
1415 return (NULL);
1416 } else if (m->m_len < (off + len)) {
1417 m_copydata(m, off, len, buf);
1418 return (buf);
1419 }
1420 return (mtod(m, char *) + off);
1421 }
1422
1423 uint32_t
1424 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
1425 {
1426 uint16_t etype;
1427 uint32_t p = key;
1428 int off;
1429 struct ether_header *eh;
1430 const struct ether_vlan_header *vlan;
1431 #ifdef INET
1432 const struct ip *ip;
1433 const uint32_t *ports;
1434 int iphlen;
1435 #endif
1436 #ifdef INET6
1437 const struct ip6_hdr *ip6;
1438 uint32_t flow;
1439 #endif
1440 union {
1441 #ifdef INET
1442 struct ip ip;
1443 #endif
1444 #ifdef INET6
1445 struct ip6_hdr ip6;
1446 #endif
1447 struct ether_vlan_header vlan;
1448 uint32_t port;
1449 } buf;
1450
1451
1452 off = sizeof(*eh);
1453 if (m->m_len < off)
1454 goto out;
1455 eh = mtod(m, struct ether_header *);
1456 etype = ntohs(eh->ether_type);
1457 if (sc->sc_flags & LAGG_F_HASHL2) {
1458 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p);
1459 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1460 }
1461
1462 /* Special handling for encapsulating VLAN frames */
1463 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) {
1464 p = hash32_buf(&m->m_pkthdr.ether_vtag,
1465 sizeof(m->m_pkthdr.ether_vtag), p);
1466 } else if (etype == ETHERTYPE_VLAN) {
1467 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf);
1468 if (vlan == NULL)
1469 goto out;
1470
1471 if (sc->sc_flags & LAGG_F_HASHL2)
1472 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1473 etype = ntohs(vlan->evl_proto);
1474 off += sizeof(*vlan) - sizeof(*eh);
1475 }
1476
1477 switch (etype) {
1478 #ifdef INET
1479 case ETHERTYPE_IP:
1480 ip = lagg_gethdr(m, off, sizeof(*ip), &buf);
1481 if (ip == NULL)
1482 goto out;
1483
1484 if (sc->sc_flags & LAGG_F_HASHL3) {
1485 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1486 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1487 }
1488 if (!(sc->sc_flags & LAGG_F_HASHL4))
1489 break;
1490 switch (ip->ip_p) {
1491 case IPPROTO_TCP:
1492 case IPPROTO_UDP:
1493 case IPPROTO_SCTP:
1494 iphlen = ip->ip_hl << 2;
1495 if (iphlen < sizeof(*ip))
1496 break;
1497 off += iphlen;
1498 ports = lagg_gethdr(m, off, sizeof(*ports), &buf);
1499 if (ports == NULL)
1500 break;
1501 p = hash32_buf(ports, sizeof(*ports), p);
1502 break;
1503 }
1504 break;
1505 #endif
1506 #ifdef INET6
1507 case ETHERTYPE_IPV6:
1508 if (!(sc->sc_flags & LAGG_F_HASHL3))
1509 break;
1510 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf);
1511 if (ip6 == NULL)
1512 goto out;
1513
1514 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1515 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1516 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1517 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */
1518 break;
1519 #endif
1520 }
1521 out:
1522 return (p);
1523 }
1524
1525 int
1526 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1527 {
1528
1529 return (ifp->if_transmit)(ifp, m);
1530 }
1531
1532 /*
1533 * Simple round robin aggregation
1534 */
1535
1536 static int
1537 lagg_rr_attach(struct lagg_softc *sc)
1538 {
1539 sc->sc_detach = lagg_rr_detach;
1540 sc->sc_start = lagg_rr_start;
1541 sc->sc_input = lagg_rr_input;
1542 sc->sc_port_create = NULL;
1543 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1544 sc->sc_seq = 0;
1545
1546 return (0);
1547 }
1548
1549 static int
1550 lagg_rr_detach(struct lagg_softc *sc)
1551 {
1552 return (0);
1553 }
1554
1555 static int
1556 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1557 {
1558 struct lagg_port *lp;
1559 uint32_t p;
1560
1561 p = atomic_fetchadd_32(&sc->sc_seq, 1);
1562 p %= sc->sc_count;
1563 lp = SLIST_FIRST(&sc->sc_ports);
1564 while (p--)
1565 lp = SLIST_NEXT(lp, lp_entries);
1566
1567 /*
1568 * Check the port's link state. This will return the next active
1569 * port if the link is down or the port is NULL.
1570 */
1571 if ((lp = lagg_link_active(sc, lp)) == NULL) {
1572 m_freem(m);
1573 return (ENOENT);
1574 }
1575
1576 /* Send mbuf */
1577 return (lagg_enqueue(lp->lp_ifp, m));
1578 }
1579
1580 static struct mbuf *
1581 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1582 {
1583 struct ifnet *ifp = sc->sc_ifp;
1584
1585 /* Just pass in the packet to our lagg device */
1586 m->m_pkthdr.rcvif = ifp;
1587
1588 return (m);
1589 }
1590
1591 /*
1592 * Active failover
1593 */
1594
1595 static int
1596 lagg_fail_attach(struct lagg_softc *sc)
1597 {
1598 sc->sc_detach = lagg_fail_detach;
1599 sc->sc_start = lagg_fail_start;
1600 sc->sc_input = lagg_fail_input;
1601 sc->sc_port_create = NULL;
1602 sc->sc_port_destroy = NULL;
1603
1604 return (0);
1605 }
1606
1607 static int
1608 lagg_fail_detach(struct lagg_softc *sc)
1609 {
1610 return (0);
1611 }
1612
1613 static int
1614 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1615 {
1616 struct lagg_port *lp;
1617
1618 /* Use the master port if active or the next available port */
1619 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1620 m_freem(m);
1621 return (ENOENT);
1622 }
1623
1624 /* Send mbuf */
1625 return (lagg_enqueue(lp->lp_ifp, m));
1626 }
1627
1628 static struct mbuf *
1629 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1630 {
1631 struct ifnet *ifp = sc->sc_ifp;
1632 struct lagg_port *tmp_tp;
1633
1634 if (lp == sc->sc_primary || lagg_failover_rx_all) {
1635 m->m_pkthdr.rcvif = ifp;
1636 return (m);
1637 }
1638
1639 if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1640 tmp_tp = lagg_link_active(sc, sc->sc_primary);
1641 /*
1642 * If tmp_tp is null, we've recieved a packet when all
1643 * our links are down. Weird, but process it anyways.
1644 */
1645 if ((tmp_tp == NULL || tmp_tp == lp)) {
1646 m->m_pkthdr.rcvif = ifp;
1647 return (m);
1648 }
1649 }
1650
1651 m_freem(m);
1652 return (NULL);
1653 }
1654
1655 /*
1656 * Loadbalancing
1657 */
1658
1659 static int
1660 lagg_lb_attach(struct lagg_softc *sc)
1661 {
1662 struct lagg_port *lp;
1663 struct lagg_lb *lb;
1664
1665 if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1666 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1667 return (ENOMEM);
1668
1669 sc->sc_detach = lagg_lb_detach;
1670 sc->sc_start = lagg_lb_start;
1671 sc->sc_input = lagg_lb_input;
1672 sc->sc_port_create = lagg_lb_port_create;
1673 sc->sc_port_destroy = lagg_lb_port_destroy;
1674 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1675
1676 lb->lb_key = arc4random();
1677 sc->sc_psc = (caddr_t)lb;
1678
1679 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1680 lagg_lb_port_create(lp);
1681
1682 return (0);
1683 }
1684
1685 static int
1686 lagg_lb_detach(struct lagg_softc *sc)
1687 {
1688 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1689 if (lb != NULL)
1690 free(lb, M_DEVBUF);
1691 return (0);
1692 }
1693
1694 static int
1695 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1696 {
1697 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1698 struct lagg_port *lp_next;
1699 int i = 0;
1700
1701 bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1702 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1703 if (lp_next == lp)
1704 continue;
1705 if (i >= LAGG_MAX_PORTS)
1706 return (EINVAL);
1707 if (sc->sc_ifflags & IFF_DEBUG)
1708 printf("%s: port %s at index %d\n",
1709 sc->sc_ifname, lp_next->lp_ifname, i);
1710 lb->lb_ports[i++] = lp_next;
1711 }
1712
1713 return (0);
1714 }
1715
1716 static int
1717 lagg_lb_port_create(struct lagg_port *lp)
1718 {
1719 struct lagg_softc *sc = lp->lp_softc;
1720 return (lagg_lb_porttable(sc, NULL));
1721 }
1722
1723 static void
1724 lagg_lb_port_destroy(struct lagg_port *lp)
1725 {
1726 struct lagg_softc *sc = lp->lp_softc;
1727 lagg_lb_porttable(sc, lp);
1728 }
1729
1730 static int
1731 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1732 {
1733 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1734 struct lagg_port *lp = NULL;
1735 uint32_t p = 0;
1736
1737 if (sc->use_flowid && (m->m_flags & M_FLOWID))
1738 p = m->m_pkthdr.flowid;
1739 else
1740 p = lagg_hashmbuf(sc, m, lb->lb_key);
1741 p %= sc->sc_count;
1742 lp = lb->lb_ports[p];
1743
1744 /*
1745 * Check the port's link state. This will return the next active
1746 * port if the link is down or the port is NULL.
1747 */
1748 if ((lp = lagg_link_active(sc, lp)) == NULL) {
1749 m_freem(m);
1750 return (ENOENT);
1751 }
1752
1753 /* Send mbuf */
1754 return (lagg_enqueue(lp->lp_ifp, m));
1755 }
1756
1757 static struct mbuf *
1758 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1759 {
1760 struct ifnet *ifp = sc->sc_ifp;
1761
1762 /* Just pass in the packet to our lagg device */
1763 m->m_pkthdr.rcvif = ifp;
1764
1765 return (m);
1766 }
1767
1768 /*
1769 * 802.3ad LACP
1770 */
1771
1772 static int
1773 lagg_lacp_attach(struct lagg_softc *sc)
1774 {
1775 struct lagg_port *lp;
1776 int error;
1777
1778 sc->sc_detach = lagg_lacp_detach;
1779 sc->sc_port_create = lacp_port_create;
1780 sc->sc_port_destroy = lacp_port_destroy;
1781 sc->sc_linkstate = lacp_linkstate;
1782 sc->sc_start = lagg_lacp_start;
1783 sc->sc_input = lagg_lacp_input;
1784 sc->sc_init = lacp_init;
1785 sc->sc_stop = lacp_stop;
1786 sc->sc_lladdr = lagg_lacp_lladdr;
1787 sc->sc_req = lacp_req;
1788 sc->sc_portreq = lacp_portreq;
1789
1790 error = lacp_attach(sc);
1791 if (error)
1792 return (error);
1793
1794 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1795 lacp_port_create(lp);
1796
1797 return (error);
1798 }
1799
1800 static int
1801 lagg_lacp_detach(struct lagg_softc *sc)
1802 {
1803 struct lagg_port *lp;
1804 int error;
1805
1806 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1807 lacp_port_destroy(lp);
1808
1809 /* unlocking is safe here */
1810 LAGG_WUNLOCK(sc);
1811 error = lacp_detach(sc);
1812 LAGG_WLOCK(sc);
1813
1814 return (error);
1815 }
1816
1817 static void
1818 lagg_lacp_lladdr(struct lagg_softc *sc)
1819 {
1820 struct lagg_port *lp;
1821
1822 /* purge all the lacp ports */
1823 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1824 lacp_port_destroy(lp);
1825
1826 /* add them back in */
1827 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1828 lacp_port_create(lp);
1829 }
1830
1831 static int
1832 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1833 {
1834 struct lagg_port *lp;
1835
1836 lp = lacp_select_tx_port(sc, m);
1837 if (lp == NULL) {
1838 m_freem(m);
1839 return (EBUSY);
1840 }
1841
1842 /* Send mbuf */
1843 return (lagg_enqueue(lp->lp_ifp, m));
1844 }
1845
1846 static struct mbuf *
1847 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1848 {
1849 struct ifnet *ifp = sc->sc_ifp;
1850 struct ether_header *eh;
1851 u_short etype;
1852
1853 eh = mtod(m, struct ether_header *);
1854 etype = ntohs(eh->ether_type);
1855
1856 /* Tap off LACP control messages */
1857 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
1858 m = lacp_input(lp, m);
1859 if (m == NULL)
1860 return (NULL);
1861 }
1862
1863 /*
1864 * If the port is not collecting or not in the active aggregator then
1865 * free and return.
1866 */
1867 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
1868 m_freem(m);
1869 return (NULL);
1870 }
1871
1872 m->m_pkthdr.rcvif = ifp;
1873 return (m);
1874 }
Cache object: 2f9086d70232a9f3a8b88ebd619a4db1
|