1 /* $OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $
2 * $FreeBSD: head/sys/net/if_lagg.c 256218 2013-10-09 19:04:40Z glebius $
3 */
4 /*
5 * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org>
6 * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org>
7 * Copyright (c) 2014 Marcelo Araujo <araujo@FreeBSD.org>
8 *
9 * Permission to use, copy, modify, and distribute this software for any
10 * purpose with or without fee is hereby granted, provided that the above
11 * copyright notice and this permission notice appear in all copies.
12 *
13 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
14 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
15 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
16 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
17 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
18 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
19 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
20 */
21
22 #include <sys/cdefs.h>
23
24 #include "opt_inet.h"
25 #include "opt_inet6.h"
26
27 #include <sys/param.h>
28 #include <sys/kernel.h>
29 #include <sys/malloc.h>
30 #include <sys/mbuf.h>
31 #include <sys/queue.h>
32 #include <sys/socket.h>
33 #include <sys/sockio.h>
34 #include <sys/sysctl.h>
35 #include <sys/module.h>
36 #include <sys/caps.h>
37 #include <sys/systm.h>
38 #include <sys/proc.h>
39 #include <sys/hash.h>
40 #include <sys/lock.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/ifq_var.h>
54 #include <net/bpf.h>
55
56 #include <net/netmsg2.h>
57 #include <net/netisr2.h>
58
59 #if defined(INET) || defined(INET6)
60 #include <netinet/in.h>
61 #endif
62 #ifdef INET
63 #include <netinet/in_systm.h>
64 #include <netinet/if_ether.h>
65 #include <netinet/ip.h>
66 #endif
67
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet6/in6_ifattach.h>
72 #endif
73
74 #include <net/vlan/if_vlan_var.h>
75 #include <net/lagg/if_lagg.h>
76 #include <net/lagg/ieee8023ad_lacp.h>
77
78 /* Special flags we should propagate to the lagg ports. */
79 static struct {
80 int flag;
81 int (*func)(struct ifnet *, int);
82 } lagg_pflags[] = {
83 {IFF_PROMISC, ifpromisc},
84 {IFF_ALLMULTI, if_allmulti},
85 {0, NULL}
86 };
87
88 SLIST_HEAD(, lagg_softc) lagg_list; /* list of laggs */
89 static struct lock lagg_list_lock;
90 eventhandler_tag lagg_detach_cookie = NULL;
91
92 static int lagg_clone_create(struct if_clone *, int, caddr_t, caddr_t);
93 static int lagg_clone_destroy(struct ifnet *);
94 static const char * laggname = "lagg";
95 struct if_clone lagg_cloner = IF_CLONE_INITIALIZER("lagg",
96 lagg_clone_create,
97 lagg_clone_destroy,
98 0, IF_MAXUNIT);
99
100 static void lagg_lladdr(struct lagg_softc *, uint8_t *);
101 static void lagg_capabilities(struct lagg_softc *);
102 static void lagg_port_lladdr(struct lagg_port *, uint8_t *);
103 static void lagg_port_setlladdr(void *, int);
104 static int lagg_port_create(struct lagg_softc *, struct ifnet *);
105 static int lagg_port_destroy(struct lagg_port *, int);
106 static void lagg_input(struct ifnet *, struct mbuf *);
107 static void lagg_linkstate(struct lagg_softc *);
108 #if 0 /* XXX */
109 static void lagg_port_state(struct ifnet *, int);
110 #endif
111 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *cr);
112 static int lagg_port_output(struct ifnet *, struct mbuf *,
113 struct sockaddr *, struct rtentry *);
114 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *);
115 #ifdef LAGG_PORT_STACKING
116 static int lagg_port_checkstacking(struct lagg_softc *);
117 #endif
118 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *);
119 static void lagg_init(void *);
120 static void lagg_stop(struct lagg_softc *);
121 static int lagg_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *cr);
122 static int lagg_ether_setmulti(struct lagg_softc *);
123 static int lagg_ether_cmdmulti(struct lagg_port *, int);
124 static int lagg_setflag(struct lagg_port *, int, int,
125 int (*func)(struct ifnet *, int));
126 static int lagg_setflags(struct lagg_port *, int status);
127 static void lagg_start(struct ifnet *, struct ifaltq_subque *ifsq);
128 static void lagg_start_dispatch(netmsg_t msg);
129 /* Not needed?
130 static int lagg_output(struct ifnet *ifp, struct mbuf *m);
131 */
132 #if 0 /* XXX */
133 static int lagg_transmit(struct ifnet *, struct mbuf *);
134 static void lagg_qflush(struct ifnet *);
135 #endif
136 static int lagg_media_change(struct ifnet *);
137 static void lagg_media_status(struct ifnet *, struct ifmediareq *);
138 static struct lagg_port *lagg_link_active(struct lagg_softc *,
139 struct lagg_port *);
140 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
141 static int lagg_sysctl_active(SYSCTL_HANDLER_ARGS);
142
143 /* Simple round robin */
144 static int lagg_rr_attach(struct lagg_softc *);
145 static int lagg_rr_detach(struct lagg_softc *);
146 static struct ifnet *lagg_rr_select_tx_port(struct lagg_softc *sc,
147 struct mbuf *m);
148 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
149 struct mbuf *);
150
151 /* Active failover */
152 static int lagg_fail_attach(struct lagg_softc *);
153 static int lagg_fail_detach(struct lagg_softc *);
154 static struct ifnet *lagg_fail_select_tx_port(struct lagg_softc *,
155 struct mbuf *);
156 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
157 struct mbuf *);
158
159 /* Loadbalancing */
160 static int lagg_lb_attach(struct lagg_softc *);
161 static int lagg_lb_detach(struct lagg_softc *);
162 static int lagg_lb_port_create(struct lagg_port *);
163 static void lagg_lb_port_destroy(struct lagg_port *);
164 static struct ifnet *lagg_lb_select_tx_port(struct lagg_softc *,
165 struct mbuf *);
166 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
167 struct mbuf *);
168 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
169
170 /* 802.3ad LACP */
171 static int lagg_lacp_attach(struct lagg_softc *);
172 static int lagg_lacp_detach(struct lagg_softc *);
173 static struct ifnet *lagg_lacp_select_tx_port(struct lagg_softc *,
174 struct mbuf *);
175 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
176 struct mbuf *);
177 static void lagg_lacp_lladdr(struct lagg_softc *);
178
179 static void lagg_callout(void *);
180
181 /* lagg protocol table */
182 static const struct {
183 int ti_proto;
184 int (*ti_attach)(struct lagg_softc *);
185 } lagg_protos[] = {
186 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach },
187 { LAGG_PROTO_FAILOVER, lagg_fail_attach },
188 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach },
189 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach },
190 { LAGG_PROTO_LACP, lagg_lacp_attach },
191 { LAGG_PROTO_NONE, NULL }
192 };
193
194 SYSCTL_DECL(_net_link);
195 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
196 "Link Aggregation");
197
198 static int lagg_failover_rx_all = 0; /* Allow input on any failover links */
199 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW,
200 &lagg_failover_rx_all, 0,
201 "Accept input from any interface in a failover lagg");
202 static int def_use_flowid = 1; /* Default value for using M_FLOWID */
203 TUNABLE_INT("net.link.lagg.default_use_flowid", &def_use_flowid);
204 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RW,
205 &def_use_flowid, 0,
206 "Default setting for using flow id for load sharing");
207
208 static int
lagg_modevent(module_t mod,int type,void * data)209 lagg_modevent(module_t mod, int type, void *data)
210 {
211
212 switch (type) {
213 case MOD_LOAD:
214 lockinit(&lagg_list_lock, "if_lagg list", 0, 0);
215 SLIST_INIT(&lagg_list);
216 if_clone_attach(&lagg_cloner);
217 lagg_input_p = lagg_input;
218 lagg_detach_cookie = EVENTHANDLER_REGISTER(
219 ifnet_departure_event, lagg_port_ifdetach, NULL,
220 EVENTHANDLER_PRI_ANY);
221 break;
222 case MOD_UNLOAD:
223 EVENTHANDLER_DEREGISTER(ifnet_departure_event,
224 lagg_detach_cookie);
225 lagg_input_p = NULL;
226 if_clone_detach(&lagg_cloner);
227 lockuninit(&lagg_list_lock);
228 break;
229 default:
230 return (EOPNOTSUPP);
231 }
232 return (0);
233 }
234
235 static moduledata_t lagg_mod = {
236 "if_lagg",
237 lagg_modevent,
238 0
239 };
240
241 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
242 MODULE_VERSION(if_lagg, 1);
243
244 /*
245 * This routine is run via an vlan
246 * config EVENT
247 */
248 static void
lagg_register_vlan(void * arg,struct ifnet * ifp,u_int16_t vtag)249 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
250 {
251 #if 0 /* XXX */
252 struct lagg_softc *sc = ifp->if_softc;
253 struct lagg_port *lp;
254
255 if (ifp->if_softc != arg) /* Not our event */
256 return;
257
258 LAGG_RLOCK(sc);
259 if (!SLIST_EMPTY(&sc->sc_ports)) {
260 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
261 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp);
262 }
263 LAGG_RUNLOCK(sc);
264 #endif
265
266 }
267
268 /*
269 * This routine is run via an vlan
270 * unconfig EVENT
271 */
272 static void
lagg_unregister_vlan(void * arg,struct ifnet * ifp,u_int16_t vtag)273 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
274 {
275 struct lagg_softc *sc = ifp->if_softc;
276 /*
277 struct lagg_port *lp;
278 */
279
280 if (ifp->if_softc != arg) /* Not our event */
281 return;
282
283 LAGG_RLOCK(sc);
284 if (!SLIST_EMPTY(&sc->sc_ports)) {
285 #if 0 /* XXX */
286 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
287 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp);
288 #endif
289 }
290 LAGG_RUNLOCK(sc);
291 }
292
293 static int
lagg_clone_create(struct if_clone * ifc,int unit,caddr_t params __unused,caddr_t data __unused)294 lagg_clone_create(struct if_clone *ifc, int unit,
295 caddr_t params __unused, caddr_t data __unused)
296 {
297 struct lagg_softc *sc;
298 struct ifnet *ifp;
299 int i, error = 0;
300 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */
301 struct sysctl_oid *oid;
302 char num[14]; /* sufficient for 32 bits */
303
304 sc = kmalloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
305 ifp = sc->sc_ifp = &sc->sc_if; // XXX if_alloc(IFT_ETHER);
306 /*
307 if (ifp == NULL) {
308 kfree(sc, M_DEVBUF);
309 return (ENOSPC);
310 }
311 */
312 /*
313 sc->sc_ipackets = counter_u64_alloc(M_WAITOK);
314 sc->sc_opackets = counter_u64_alloc(M_WAITOK);
315 sc->sc_ibytes = counter_u64_alloc(M_WAITOK);
316 sc->sc_obytes = counter_u64_alloc(M_WAITOK);
317 */
318 sysctl_ctx_init(&sc->ctx);
319 ksnprintf(num, sizeof(num), "%u", unit);
320 sc->use_flowid = def_use_flowid;
321 sc->sc_oid = oid = SYSCTL_ADD_NODE(&sc->ctx,
322 &SYSCTL_NODE_CHILDREN(_net_link, lagg),
323 OID_AUTO, num, CTLFLAG_RD, NULL, "");
324 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
325 "use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid, sc->use_flowid,
326 "Use flow id for load sharing");
327 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
328 "count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count,
329 "Total number of ports");
330 SYSCTL_ADD_PROC(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
331 "active", CTLTYPE_INT|CTLFLAG_RD, sc, 0, lagg_sysctl_active,
332 "I", "Total number of active ports");
333 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
334 "flapping", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_flapping,
335 sc->sc_flapping, "Total number of port change events");
336 /* Hash all layers by default */
337 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4;
338
339 sc->sc_proto = LAGG_PROTO_NONE;
340 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
341 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
342 sc->sc_proto = lagg_protos[i].ti_proto;
343 if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
344 if_free(ifp);
345 kfree(sc, M_DEVBUF);
346 return (error);
347 }
348 break;
349 }
350 }
351 LAGG_LOCK_INIT(sc);
352 LAGG_CALLOUT_LOCK_INIT(sc);
353 SLIST_INIT(&sc->sc_ports);
354 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
355
356 /*
357 * This uses the callout lock rather than the rmlock; one can't
358 * hold said rmlock during SWI.
359 */
360 callout_init_lk(&sc->sc_callout, &sc->sc_call_lock);
361
362 /* Initialise pseudo media types */
363 ifmedia_init(&sc->sc_media, 0, lagg_media_change,
364 lagg_media_status);
365 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
366 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
367
368 if_initname(ifp, laggname, unit);
369 ifp->if_softc = sc;
370 #if 0 /* XXX */
371 ifp->if_transmit = lagg_transmit;
372 ifp->if_qflush = lagg_qflush;
373 #endif
374 ifp->if_mtu = ETHERMTU;
375 ifp->if_init = lagg_init;
376 ifp->if_ioctl = lagg_ioctl;
377 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
378 ifp->if_start = lagg_start;
379 ifp->if_type = IFT_ETHER;
380 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
381 ifq_set_ready(&ifp->if_snd);
382 ifp->if_hdrlen = ETHER_HDR_LEN;
383
384 #if 0 /* XXX */
385 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
386 #endif
387 /*
388 * Attach as an ordinary ethernet device, children will be attached
389 * as special device IFT_IEEE8023ADLAG.
390 */
391
392 ether_ifattach(ifp, eaddr, NULL);
393
394 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
395 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
396 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
397 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
398
399 /* Insert into the global list of laggs */
400 lockmgr(&lagg_list_lock, LK_EXCLUSIVE);
401 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
402 lockmgr(&lagg_list_lock, LK_RELEASE);
403
404 callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
405
406 return (0);
407 }
408
409 static int
lagg_clone_destroy(struct ifnet * ifp)410 lagg_clone_destroy(struct ifnet *ifp)
411 {
412 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
413 struct lagg_port *lp;
414
415 LAGG_WLOCK(sc);
416
417 lagg_stop(sc);
418 ifp->if_flags &= ~IFF_UP;
419
420 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
421 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
422
423 /* Shutdown and remove lagg ports */
424 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
425 lagg_port_destroy(lp, 1);
426 /* Unhook the aggregation protocol */
427 if (sc->sc_detach != NULL)
428 (*sc->sc_detach)(sc);
429
430 LAGG_WUNLOCK(sc);
431
432 sysctl_ctx_free(&sc->ctx);
433 ifmedia_removeall(&sc->sc_media);
434 ether_ifdetach(ifp);
435 /* This ifp is part of lagg_softc, don't free it! */
436 /* if_free(ifp); */
437
438 /* This grabs sc_callout_mtx, serialising it correctly */
439 callout_drain(&sc->sc_callout);
440
441 #if 0
442 /* At this point it's drained; we can free this */
443 counter_u64_free(sc->sc_ipackets);
444 counter_u64_free(sc->sc_opackets);
445 counter_u64_free(sc->sc_ibytes);
446 counter_u64_free(sc->sc_obytes);
447 #endif
448
449 lockmgr(&lagg_list_lock, LK_EXCLUSIVE);
450 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
451 lockmgr(&lagg_list_lock, LK_RELEASE);
452
453 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
454 LAGG_LOCK_DESTROY(sc);
455 LAGG_CALLOUT_LOCK_DESTROY(sc);
456 kfree(sc, M_DEVBUF);
457
458 return 0;
459 }
460
461 static void
lagg_lladdr(struct lagg_softc * sc,uint8_t * lladdr)462 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
463 {
464 struct ifnet *ifp = sc->sc_ifp;
465
466 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
467 return;
468
469 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
470 /* Let the protocol know the MAC has changed */
471 if (sc->sc_lladdr != NULL)
472 (*sc->sc_lladdr)(sc);
473 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
474 }
475
476 static void
lagg_capabilities(struct lagg_softc * sc)477 lagg_capabilities(struct lagg_softc *sc)
478 {
479 struct lagg_port *lp;
480 int cap = ~0, ena = ~0;
481 u_long hwa = ~0UL;
482
483 LAGG_WLOCK_ASSERT(sc);
484
485 /* Get capabilities from the lagg ports */
486 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
487 cap &= lp->lp_ifp->if_capabilities;
488 ena &= lp->lp_ifp->if_capenable;
489 hwa &= lp->lp_ifp->if_hwassist;
490 }
491 cap = (cap == ~0 ? 0 : cap);
492 ena = (ena == ~0 ? 0 : ena);
493 hwa = (hwa == ~0 ? 0 : hwa);
494
495 if (sc->sc_ifp->if_capabilities != cap ||
496 sc->sc_ifp->if_capenable != ena ||
497 sc->sc_ifp->if_hwassist != hwa) {
498 sc->sc_ifp->if_capabilities = cap;
499 sc->sc_ifp->if_capenable = ena;
500 sc->sc_ifp->if_hwassist = hwa;
501 getmicrotime(&sc->sc_ifp->if_lastchange);
502
503 if (sc->sc_ifflags & IFF_DEBUG)
504 if_printf(sc->sc_ifp,
505 "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
506 }
507 }
508
509 static void
lagg_port_lladdr(struct lagg_port * lp,uint8_t * lladdr)510 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
511 {
512 struct lagg_softc *sc = lp->lp_softc;
513 struct ifnet *ifp = lp->lp_ifp;
514 struct lagg_llq *llq;
515 int pending = 0;
516
517 LAGG_WLOCK_ASSERT(sc);
518
519 if (lp->lp_detaching ||
520 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
521 return;
522
523 /* Check to make sure its not already queued to be changed */
524 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
525 if (llq->llq_ifp == ifp) {
526 pending = 1;
527 break;
528 }
529 }
530
531 if (!pending) {
532 llq = kmalloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
533 if (llq == NULL) /* XXX what to do */
534 return;
535 }
536
537 /* Update the lladdr even if pending, it may have changed */
538 llq->llq_ifp = ifp;
539 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
540
541 if (!pending)
542 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
543
544 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
545 }
546
547 /*
548 * Set the interface MAC address from a taskqueue to avoid a LOR.
549 */
550 static void
lagg_port_setlladdr(void * arg,int pending)551 lagg_port_setlladdr(void *arg, int pending)
552 {
553 struct lagg_softc *sc = (struct lagg_softc *)arg;
554 struct lagg_llq *llq, *head;
555 struct ifnet *ifp;
556 int error;
557
558 /* Grab a local reference of the queue and remove it from the softc */
559 LAGG_WLOCK(sc);
560 head = SLIST_FIRST(&sc->sc_llq_head);
561 SLIST_FIRST(&sc->sc_llq_head) = NULL;
562 LAGG_WUNLOCK(sc);
563
564 /*
565 * Traverse the queue and set the lladdr on each ifp. It is safe to do
566 * unlocked as we have the only reference to it.
567 */
568 for (llq = head; llq != NULL; llq = head) {
569 ifp = llq->llq_ifp;
570
571 /* Set the link layer address */
572 /* CURVNET_SET(ifp->if_vnet); */
573 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
574 /* CURVNET_RESTORE(); */
575 if (error)
576 kprintf("%s: setlladdr failed on %s\n", __func__,
577 ifp->if_xname);
578
579 head = SLIST_NEXT(llq, llq_entries);
580 kfree(llq, M_DEVBUF);
581 }
582 }
583
584 static int
lagg_port_create(struct lagg_softc * sc,struct ifnet * ifp)585 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
586 {
587 struct lagg_softc *sc_ptr;
588 struct lagg_port *lp;
589 int error = 0;
590
591 LAGG_WLOCK_ASSERT(sc);
592
593 /* Limit the maximal number of lagg ports */
594 if (sc->sc_count >= LAGG_MAX_PORTS)
595 return (ENOSPC);
596
597 /* Check if port has already been associated to a lagg */
598 if (ifp->if_lagg != NULL) {
599 /* Port is already in the current lagg? */
600 lp = (struct lagg_port *)ifp->if_lagg;
601 if (lp->lp_softc == sc)
602 return (EEXIST);
603 return (EBUSY);
604 }
605
606 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */
607 if (ifp->if_type != IFT_ETHER)
608 return (EPROTONOSUPPORT);
609
610 #ifdef INET6
611 /*
612 * The member interface should not have inet6 address because
613 * two interfaces with a valid link-local scope zone must not be
614 * merged in any form. This restriction is needed to
615 * prevent violation of link-local scope zone. Attempts to
616 * add a member interface which has inet6 addresses triggers
617 * removal of all inet6 addresses on the member interface.
618 */
619 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
620 if (in6ifa_llaonifp(lp->lp_ifp)) {
621 in6_ifdetach(lp->lp_ifp);
622 if_printf(sc->sc_ifp,
623 "IPv6 addresses on %s have been removed "
624 "before adding it as a member to prevent "
625 "IPv6 address scope violation.\n",
626 lp->lp_ifp->if_xname);
627 }
628 }
629 if (in6ifa_llaonifp(ifp)) {
630 in6_ifdetach(ifp);
631 if_printf(sc->sc_ifp,
632 "IPv6 addresses on %s have been removed "
633 "before adding it as a member to prevent "
634 "IPv6 address scope violation.\n",
635 ifp->if_xname);
636 }
637 #endif
638 /* Allow the first Ethernet member to define the MTU */
639 if (SLIST_EMPTY(&sc->sc_ports))
640 sc->sc_ifp->if_mtu = ifp->if_mtu;
641 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
642 if_printf(sc->sc_ifp, "invalid MTU for %s\n",
643 ifp->if_xname);
644 return (EINVAL);
645 }
646
647 if ((lp = kmalloc(sizeof(struct lagg_port),
648 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
649 return (ENOMEM);
650
651 /* Check if port is a stacked lagg */
652 lockmgr(&lagg_list_lock, LK_EXCLUSIVE);
653 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
654 if (ifp == sc_ptr->sc_ifp) {
655 lockmgr(&lagg_list_lock, LK_RELEASE);
656 kfree(lp, M_DEVBUF);
657 return (EINVAL);
658 /* XXX disable stacking for the moment, its untested */
659 #ifdef LAGG_PORT_STACKING
660 lp->lp_flags |= LAGG_PORT_STACK;
661 if (lagg_port_checkstacking(sc_ptr) >=
662 LAGG_MAX_STACKING) {
663 lockmgr(&lagg_list_lock, LK_RELEASE);
664 kfree(lp, M_DEVBUF);
665 return (E2BIG);
666 }
667 #endif
668 }
669 }
670 lockmgr(&lagg_list_lock, LK_RELEASE);
671
672 /* Change the interface type */
673 lp->lp_iftype = ifp->if_type;
674 ifp->if_type = IFT_IEEE8023ADLAG;
675 ifp->if_lagg = lp;
676 lp->lp_ioctl = ifp->if_ioctl;
677 ifp->if_ioctl = lagg_port_ioctl;
678 lp->lp_output = ifp->if_output;
679 ifp->if_output = lagg_port_output;
680
681 lp->lp_ifp = ifp;
682 lp->lp_softc = sc;
683
684 /* Save port link layer address */
685 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
686
687 if (SLIST_EMPTY(&sc->sc_ports)) {
688 sc->sc_primary = lp;
689 lagg_lladdr(sc, IF_LLADDR(ifp));
690 } else {
691 /* Update link layer address for this port */
692 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
693 }
694
695 /* Insert into the list of ports */
696 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
697 sc->sc_count++;
698
699 /* Update lagg capabilities */
700 lagg_capabilities(sc);
701 lagg_linkstate(sc);
702
703 /* Add multicast addresses and interface flags to this port */
704 lagg_ether_cmdmulti(lp, 1);
705 lagg_setflags(lp, 1);
706
707 if (sc->sc_port_create != NULL)
708 error = (*sc->sc_port_create)(lp);
709 if (error) {
710 /* remove the port again, without calling sc_port_destroy */
711 lagg_port_destroy(lp, 0);
712 return (error);
713 }
714
715 return (error);
716 }
717
718 #ifdef LAGG_PORT_STACKING
719 static int
lagg_port_checkstacking(struct lagg_softc * sc)720 lagg_port_checkstacking(struct lagg_softc *sc)
721 {
722 struct lagg_softc *sc_ptr;
723 struct lagg_port *lp;
724 int m = 0;
725
726 LAGG_WLOCK_ASSERT(sc);
727
728 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
729 if (lp->lp_flags & LAGG_PORT_STACK) {
730 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
731 m = MAX(m, lagg_port_checkstacking(sc_ptr));
732 }
733 }
734
735 return (m + 1);
736 }
737 #endif
738
739 static int
lagg_port_destroy(struct lagg_port * lp,int runpd)740 lagg_port_destroy(struct lagg_port *lp, int runpd)
741 {
742 struct lagg_softc *sc = lp->lp_softc;
743 struct lagg_port *lp_ptr;
744 struct lagg_llq *llq;
745 struct ifnet *ifp = lp->lp_ifp;
746
747 LAGG_WLOCK_ASSERT(sc);
748
749 if (runpd && sc->sc_port_destroy != NULL)
750 (*sc->sc_port_destroy)(lp);
751
752 /*
753 * Remove multicast addresses and interface flags from this port and
754 * reset the MAC address, skip if the interface is being detached.
755 */
756 if (!lp->lp_detaching) {
757 lagg_ether_cmdmulti(lp, 0);
758 lagg_setflags(lp, 0);
759 lagg_port_lladdr(lp, lp->lp_lladdr);
760 }
761
762 /* Restore interface */
763 ifp->if_type = lp->lp_iftype;
764 ifp->if_ioctl = lp->lp_ioctl;
765 ifp->if_output = lp->lp_output;
766 ifp->if_lagg = NULL;
767
768 /* Finally, remove the port from the lagg */
769 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
770 sc->sc_count--;
771
772 /* Update the primary interface */
773 if (lp == sc->sc_primary) {
774 uint8_t lladdr[ETHER_ADDR_LEN];
775
776 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
777 bzero(&lladdr, ETHER_ADDR_LEN);
778 } else {
779 bcopy(lp_ptr->lp_lladdr,
780 lladdr, ETHER_ADDR_LEN);
781 }
782 lagg_lladdr(sc, lladdr);
783 sc->sc_primary = lp_ptr;
784
785 /* Update link layer address for each port */
786 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
787 lagg_port_lladdr(lp_ptr, lladdr);
788 }
789
790 /* Remove any pending lladdr changes from the queue */
791 if (lp->lp_detaching) {
792 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
793 if (llq->llq_ifp == ifp) {
794 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
795 llq_entries);
796 kfree(llq, M_DEVBUF);
797 break; /* Only appears once */
798 }
799 }
800 }
801
802 if (lp->lp_ifflags)
803 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
804
805 kfree(lp, M_DEVBUF);
806
807 /* Update lagg capabilities */
808 lagg_capabilities(sc);
809 lagg_linkstate(sc);
810
811 return (0);
812 }
813
814 static int
lagg_port_ioctl(struct ifnet * ifp,u_long cmd,caddr_t data,struct ucred * cr)815 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
816 {
817 struct lagg_reqport *rp = (struct lagg_reqport *)data;
818 struct lagg_softc *sc;
819 struct lagg_port *lp = NULL;
820 int error = 0;
821
822 ASSERT_IFNET_SERIALIZED_ALL(ifp);
823
824 /* Should be checked by the caller */
825 if (ifp->if_type != IFT_IEEE8023ADLAG ||
826 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
827 goto fallback;
828
829 switch (cmd) {
830 case SIOCGLAGGPORT:
831 if (rp->rp_portname[0] == '\0') {
832 error = EINVAL;
833 break;
834 }
835
836 /*
837 * Release ifp serializers before ifnet_lock
838 * to prevent lock order reversal.
839 */
840 ifnet_deserialize_all(ifp);
841 ifnet_lock();
842 if (ifunit(rp->rp_portname) != ifp) {
843 ifnet_unlock();
844 ifnet_serialize_all(ifp);
845 error = EINVAL;
846 break;
847 }
848 ifnet_unlock();
849 ifnet_serialize_all(ifp);
850
851 LAGG_RLOCK(sc);
852 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
853 error = ENOENT;
854 LAGG_RUNLOCK(sc);
855 break;
856 }
857
858 lagg_port2req(lp, rp);
859 LAGG_RUNLOCK(sc);
860 break;
861
862 case SIOCSIFCAP:
863 if (lp->lp_ioctl == NULL) {
864 error = EINVAL;
865 break;
866 }
867
868 error = (*lp->lp_ioctl)(ifp, cmd, data, cr);
869 if (error)
870 break;
871
872 /* Update lagg interface capabilities */
873 LAGG_WLOCK(sc);
874 lagg_capabilities(sc);
875 LAGG_WUNLOCK(sc);
876 break;
877 case SIOCGIFMEDIA:
878 if (lp->lp_ioctl == NULL) {
879 error = EINVAL;
880 break;
881 }
882
883 error = (*lp->lp_ioctl)(ifp, cmd, data, cr);
884 break;
885 case SIOCSIFMTU:
886 /* Do not allow the MTU to be changed once joined */
887 error = EINVAL;
888 break;
889
890 default:
891 goto fallback;
892 }
893
894 return (error);
895
896 fallback:
897 if (lp->lp_ioctl != NULL) {
898 int result;
899 result = ((*lp->lp_ioctl)(ifp, cmd, data, cr));
900 }
901 return (EINVAL);
902 }
903
904 /*
905 * For direct output to child ports.
906 */
907 static int
lagg_port_output(struct ifnet * ifp,struct mbuf * m,struct sockaddr * dst,struct rtentry * ro)908 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
909 struct sockaddr *dst, struct rtentry *ro)
910 {
911 struct lagg_port *lp = ifp->if_lagg;
912
913 switch (dst->sa_family) {
914 case pseudo_AF_HDRCMPLT:
915 case AF_UNSPEC:
916 return ((*lp->lp_output)(ifp, m, dst, ro));
917 }
918
919 /* drop any other frames */
920 m_freem(m);
921 return (ENETDOWN);
922 }
923
924 static void
lagg_port_ifdetach(void * arg __unused,struct ifnet * ifp)925 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
926 {
927 struct lagg_port *lp;
928 struct lagg_softc *sc;
929
930 if ((lp = ifp->if_lagg) == NULL)
931 return;
932 #if 0 /* XXX */
933 /* If the ifnet is just being renamed, don't do anything. */
934 if (ifp->if_flags & IFF_RENAMING)
935 return;
936 #endif
937 sc = lp->lp_softc;
938
939 LAGG_WLOCK(sc);
940 lp->lp_detaching = 1;
941 lagg_port_destroy(lp, 1);
942 LAGG_WUNLOCK(sc);
943 }
944
945 static void
lagg_port2req(struct lagg_port * lp,struct lagg_reqport * rp)946 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
947 {
948 struct lagg_softc *sc = lp->lp_softc;
949
950 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
951 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
952 rp->rp_prio = lp->lp_prio;
953 rp->rp_flags = lp->lp_flags;
954 if (sc->sc_portreq != NULL)
955 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
956
957 /* Add protocol specific flags */
958 switch (sc->sc_proto) {
959 case LAGG_PROTO_FAILOVER:
960 if (lp == sc->sc_primary)
961 rp->rp_flags |= LAGG_PORT_MASTER;
962 if (lp == lagg_link_active(sc, sc->sc_primary))
963 rp->rp_flags |= LAGG_PORT_ACTIVE;
964 break;
965
966 case LAGG_PROTO_ROUNDROBIN:
967 case LAGG_PROTO_LOADBALANCE:
968 case LAGG_PROTO_ETHERCHANNEL:
969 if (LAGG_PORTACTIVE(lp))
970 rp->rp_flags |= LAGG_PORT_ACTIVE;
971 break;
972
973 case LAGG_PROTO_LACP:
974 /* LACP has a different definition of active */
975 if (lacp_isactive(lp))
976 rp->rp_flags |= LAGG_PORT_ACTIVE;
977 if (lacp_iscollecting(lp))
978 rp->rp_flags |= LAGG_PORT_COLLECTING;
979 if (lacp_isdistributing(lp))
980 rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
981 break;
982 }
983
984 }
985
986 static void
lagg_init(void * xsc)987 lagg_init(void *xsc)
988 {
989 struct lagg_softc *sc = (struct lagg_softc *)xsc;
990 struct lagg_port *lp;
991 struct ifnet *ifp = sc->sc_ifp;
992
993 if (ifp->if_flags & IFF_RUNNING)
994 return;
995
996 LAGG_WLOCK(sc);
997
998 ifp->if_flags |= IFF_RUNNING;
999 /* Update the port lladdrs */
1000 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1001 lagg_port_lladdr(lp, IF_LLADDR(ifp));
1002
1003 if (sc->sc_init != NULL)
1004 (*sc->sc_init)(sc);
1005
1006 LAGG_WUNLOCK(sc);
1007 }
1008
1009 static void
lagg_stop(struct lagg_softc * sc)1010 lagg_stop(struct lagg_softc *sc)
1011 {
1012 struct ifnet *ifp = sc->sc_ifp;
1013
1014 LAGG_WLOCK_ASSERT(sc);
1015
1016 if ((ifp->if_flags & IFF_RUNNING) == 0)
1017 return;
1018
1019 ifp->if_flags &= ~IFF_RUNNING;
1020
1021 if (sc->sc_stop != NULL)
1022 (*sc->sc_stop)(sc);
1023 }
1024
1025 static int
lagg_ioctl(struct ifnet * ifp,u_long cmd,caddr_t data,struct ucred * cr)1026 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
1027 {
1028 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1029 struct lagg_reqall *ra = (struct lagg_reqall *)data;
1030 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1031 struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1032 struct ifreq *ifr = (struct ifreq *)data;
1033 struct lagg_port *lp;
1034 struct ifnet *tpif;
1035 char *buf, *outbuf;
1036 int count, buflen, len, error = 0;
1037
1038 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1039
1040 bzero(&rpbuf, sizeof(rpbuf));
1041
1042 switch (cmd) {
1043 case SIOCGLAGG:
1044 LAGG_RLOCK(sc);
1045 count = 0;
1046 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1047 count++;
1048 buflen = count * sizeof(struct lagg_reqport);
1049 LAGG_RUNLOCK(sc);
1050
1051 outbuf = kmalloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1052
1053 LAGG_RLOCK(sc);
1054 ra->ra_proto = sc->sc_proto;
1055 if (sc->sc_req != NULL)
1056 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
1057
1058 count = 0;
1059 buf = outbuf;
1060 len = min(ra->ra_size, buflen);
1061 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1062 if (len < sizeof(rpbuf))
1063 break;
1064
1065 lagg_port2req(lp, &rpbuf);
1066 memcpy(buf, &rpbuf, sizeof(rpbuf));
1067 count++;
1068 buf += sizeof(rpbuf);
1069 len -= sizeof(rpbuf);
1070 }
1071 LAGG_RUNLOCK(sc);
1072 ra->ra_ports = count;
1073 ra->ra_size = count * sizeof(rpbuf);
1074 error = copyout(outbuf, ra->ra_port, ra->ra_size);
1075 kfree(outbuf, M_TEMP);
1076 break;
1077 case SIOCSLAGG:
1078 error = caps_priv_check_self(SYSCAP_NONET_LAGG);
1079 if (error)
1080 break;
1081 if (ra->ra_proto >= LAGG_PROTO_MAX) {
1082 error = EPROTONOSUPPORT;
1083 break;
1084 }
1085 LAGG_WLOCK(sc);
1086 if (sc->sc_proto != LAGG_PROTO_NONE) {
1087 /* Reset protocol first in case detach unlocks */
1088 sc->sc_proto = LAGG_PROTO_NONE;
1089 error = sc->sc_detach(sc);
1090 sc->sc_detach = NULL;
1091 sc->sc_start = NULL;
1092 sc->sc_input = NULL;
1093 sc->sc_port_create = NULL;
1094 sc->sc_port_destroy = NULL;
1095 sc->sc_linkstate = NULL;
1096 sc->sc_init = NULL;
1097 sc->sc_stop = NULL;
1098 sc->sc_lladdr = NULL;
1099 sc->sc_req = NULL;
1100 sc->sc_portreq = NULL;
1101 } else if (sc->sc_input != NULL) {
1102 /* Still detaching */
1103 error = EBUSY;
1104 }
1105 if (error != 0) {
1106 LAGG_WUNLOCK(sc);
1107 break;
1108 }
1109 for (int i = 0; i < (NELEM(lagg_protos)); i++) {
1110 if (lagg_protos[i].ti_proto == ra->ra_proto) {
1111 if (sc->sc_ifflags & IFF_DEBUG)
1112 kprintf("%s: using proto %u\n",
1113 sc->sc_ifname,
1114 lagg_protos[i].ti_proto);
1115 sc->sc_proto = lagg_protos[i].ti_proto;
1116 if (sc->sc_proto != LAGG_PROTO_NONE)
1117 error = lagg_protos[i].ti_attach(sc);
1118 LAGG_WUNLOCK(sc);
1119 return (error);
1120 }
1121 }
1122 LAGG_WUNLOCK(sc);
1123 error = EPROTONOSUPPORT;
1124 break;
1125 case SIOCGLAGGFLAGS:
1126 rf->rf_flags = sc->sc_flags;
1127 break;
1128 case SIOCSLAGGHASH:
1129 error = caps_priv_check_self(SYSCAP_NONET_LAGG);
1130 if (error)
1131 break;
1132 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1133 error = EINVAL;
1134 break;
1135 }
1136 LAGG_WLOCK(sc);
1137 sc->sc_flags &= ~LAGG_F_HASHMASK;
1138 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK;
1139 LAGG_WUNLOCK(sc);
1140 break;
1141 case SIOCGLAGGPORT:
1142 /*
1143 * Release ifp serializers before ifnet_lock
1144 * to prevent lock order reversal.
1145 */
1146 ifnet_deserialize_all(ifp);
1147 ifnet_lock();
1148 if (rp->rp_portname[0] == '\0' ||
1149 (tpif = ifunit(rp->rp_portname)) == NULL) {
1150 ifnet_unlock();
1151 ifnet_serialize_all(ifp);
1152 error = EINVAL;
1153 break;
1154 }
1155
1156 LAGG_RLOCK(sc);
1157 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1158 lp->lp_softc != sc) {
1159 error = ENOENT;
1160 LAGG_RUNLOCK(sc);
1161 ifnet_unlock();
1162 ifnet_serialize_all(ifp);
1163 break;
1164 }
1165
1166 lagg_port2req(lp, rp);
1167 LAGG_RUNLOCK(sc);
1168 ifnet_unlock();
1169 ifnet_serialize_all(ifp);
1170 break;
1171 case SIOCSLAGGPORT:
1172 error = caps_priv_check_self(SYSCAP_NONET_LAGG);
1173 if (error)
1174 break;
1175 /*
1176 * Release ifp serializers before ifnet_lock
1177 * to prevent lock order reversal.
1178 */
1179 ifnet_deserialize_all(ifp);
1180 ifnet_lock();
1181 if (rp->rp_portname[0] == '\0' ||
1182 (tpif = ifunit(rp->rp_portname)) == NULL) {
1183 ifnet_unlock();
1184 ifnet_serialize_all(ifp);
1185 error = EINVAL;
1186 break;
1187 }
1188 LAGG_WLOCK(sc);
1189 error = lagg_port_create(sc, tpif);
1190 LAGG_WUNLOCK(sc);
1191 ifnet_unlock();
1192 ifnet_serialize_all(ifp);
1193 break;
1194 case SIOCSLAGGDELPORT:
1195 error = caps_priv_check_self(SYSCAP_NONET_LAGG);
1196 if (error)
1197 break;
1198 /*
1199 * Release ifp serializers before ifnet_lock
1200 * to prevent lock order reversal.
1201 */
1202 ifnet_deserialize_all(ifp);
1203 ifnet_lock();
1204 if (rp->rp_portname[0] == '\0' ||
1205 (tpif = ifunit(rp->rp_portname)) == NULL) {
1206 ifnet_unlock();
1207 ifnet_serialize_all(ifp);
1208 error = EINVAL;
1209 break;
1210 }
1211
1212 LAGG_WLOCK(sc);
1213 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1214 lp->lp_softc != sc) {
1215 error = ENOENT;
1216 LAGG_WUNLOCK(sc);
1217 ifnet_unlock();
1218 ifnet_serialize_all(ifp);
1219 break;
1220 }
1221
1222 error = lagg_port_destroy(lp, 1);
1223 LAGG_WUNLOCK(sc);
1224 ifnet_unlock();
1225 ifnet_serialize_all(ifp);
1226 break;
1227 case SIOCSIFFLAGS:
1228 /* Set flags on ports too */
1229 LAGG_WLOCK(sc);
1230 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1231 lagg_setflags(lp, 1);
1232 }
1233 LAGG_WUNLOCK(sc);
1234
1235 if (!(ifp->if_flags & IFF_UP) &&
1236 (ifp->if_flags & IFF_RUNNING)) {
1237 /*
1238 * If interface is marked down and it is running,
1239 * then stop and disable it.
1240 */
1241 LAGG_WLOCK(sc);
1242 lagg_stop(sc);
1243 LAGG_WUNLOCK(sc);
1244 } else if ((ifp->if_flags & IFF_UP) &&
1245 !(ifp->if_flags & IFF_RUNNING)) {
1246 /*
1247 * If interface is marked up and it is stopped, then
1248 * start it.
1249 */
1250 (*ifp->if_init)(sc);
1251 }
1252 break;
1253 case SIOCADDMULTI:
1254 case SIOCDELMULTI:
1255 LAGG_WLOCK(sc);
1256 error = lagg_ether_setmulti(sc);
1257 LAGG_WUNLOCK(sc);
1258 break;
1259 case SIOCSIFMEDIA:
1260 case SIOCGIFMEDIA:
1261 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1262 break;
1263
1264 case SIOCSIFCAP:
1265 case SIOCSIFMTU:
1266 /* Do not allow the MTU or caps to be directly changed */
1267 error = EINVAL;
1268 break;
1269
1270 default:
1271 error = ether_ioctl(ifp, cmd, data);
1272 break;
1273 }
1274 return (error);
1275 }
1276
1277 static int
lagg_ether_setmulti(struct lagg_softc * sc)1278 lagg_ether_setmulti(struct lagg_softc *sc)
1279 {
1280 struct lagg_port *lp;
1281
1282 LAGG_WLOCK_ASSERT(sc);
1283
1284 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1285 /* First, remove any existing filter entries. */
1286 lagg_ether_cmdmulti(lp, 0);
1287 /* copy all addresses from the lagg interface to the port */
1288 lagg_ether_cmdmulti(lp, 1);
1289 }
1290 return (0);
1291 }
1292
1293 static int
lagg_ether_cmdmulti(struct lagg_port * lp,int set)1294 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1295 {
1296 struct lagg_softc *sc = lp->lp_softc;
1297 struct ifnet *ifp = lp->lp_ifp;
1298 struct ifnet *scifp = sc->sc_ifp;
1299 struct lagg_mc *mc;
1300 struct ifmultiaddr *ifma, *rifma = NULL;
1301 struct sockaddr_dl sdl;
1302 int error;
1303
1304 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp);
1305 LAGG_WLOCK_ASSERT(sc);
1306
1307 bzero((char *)&sdl, sizeof(sdl));
1308 sdl.sdl_len = sizeof(sdl);
1309 sdl.sdl_family = AF_LINK;
1310 sdl.sdl_type = IFT_ETHER;
1311 sdl.sdl_alen = ETHER_ADDR_LEN;
1312 sdl.sdl_index = ifp->if_index;
1313
1314 if (set) {
1315 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1316 if (ifma->ifma_addr->sa_family != AF_LINK)
1317 continue;
1318 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1319 LLADDR(&sdl), ETHER_ADDR_LEN);
1320
1321 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1322 if (error)
1323 return (error);
1324 mc = kmalloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1325 if (mc == NULL)
1326 return (ENOMEM);
1327 mc->mc_ifma = rifma;
1328 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1329 }
1330 } else {
1331 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1332 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1333 if_delmulti(ifp, (struct sockaddr *)mc->mc_ifma);
1334 kfree(mc, M_DEVBUF);
1335 }
1336 }
1337 return (0);
1338 }
1339
1340 /* Handle a ref counted flag that should be set on the lagg port as well */
1341 static int
lagg_setflag(struct lagg_port * lp,int flag,int status,int (* func)(struct ifnet *,int))1342 lagg_setflag(struct lagg_port *lp, int flag, int status,
1343 int (*func)(struct ifnet *, int))
1344 {
1345 struct lagg_softc *sc = lp->lp_softc;
1346 struct ifnet *scifp = sc->sc_ifp;
1347 struct ifnet *ifp = lp->lp_ifp;
1348 int error;
1349
1350 LAGG_WLOCK_ASSERT(sc);
1351 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp);
1352
1353 status = status ? (scifp->if_flags & flag) : 0;
1354 /* Now "status" contains the flag value or 0 */
1355
1356 /*
1357 * See if recorded ports status is different from what
1358 * we want it to be. If it is, flip it. We record ports
1359 * status in lp_ifflags so that we won't clear ports flag
1360 * we haven't set. In fact, we don't clear or set ports
1361 * flags directly, but get or release references to them.
1362 * That's why we can be sure that recorded flags still are
1363 * in accord with actual ports flags.
1364 */
1365 if (status != (lp->lp_ifflags & flag)) {
1366 error = (*func)(ifp, status);
1367 if (error)
1368 return (error);
1369 lp->lp_ifflags &= ~flag;
1370 lp->lp_ifflags |= status;
1371 }
1372 return (0);
1373 }
1374
1375 /*
1376 * Handle IFF_* flags that require certain changes on the lagg port
1377 * if "status" is true, update ports flags respective to the lagg
1378 * if "status" is false, forcedly clear the flags set on port.
1379 */
1380 static int
lagg_setflags(struct lagg_port * lp,int status)1381 lagg_setflags(struct lagg_port *lp, int status)
1382 {
1383 int error, i;
1384
1385 ASSERT_IFNET_NOT_SERIALIZED_ALL(lp->lp_ifp);
1386
1387 for (i = 0; lagg_pflags[i].flag; i++) {
1388 error = lagg_setflag(lp, lagg_pflags[i].flag,
1389 status, lagg_pflags[i].func);
1390 if (error)
1391 return (error);
1392 }
1393 return (0);
1394 }
1395
1396
1397 #if 0 /* XXX not needed? */
1398 static int
1399 lagg_output(struct ifnet *ifp, struct mbuf *m)
1400 {
1401 struct lagg_softc *sc = ifp->if_softc;
1402 int error, len, mcast;
1403
1404 len = m->m_pkthdr.len;
1405 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1406
1407 LAGG_RLOCK(sc);
1408 /* We need a Tx algorithm and at least one port */
1409 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1410 LAGG_RUNLOCK(sc);
1411 m_freem(m);
1412 ifp->if_oerrors++;
1413 return (ENXIO);
1414 }
1415
1416 BPF_MTAP(ifp, m);
1417
1418 error = (*sc->sc_start)(sc, m);
1419 LAGG_RUNLOCK(sc);
1420
1421 if (error == 0) {
1422 IFNET_STAT_INC(ifp, opackets, 1);
1423 IFNET_STAT_INC(ifp, obytes, len);
1424 ifp->if_omcasts += mcast;
1425 } else
1426 ifp->if_oerrors++;
1427
1428 return error;
1429 }
1430 #endif
1431
1432 #if 0 /* XXX */
1433 /*
1434 * The ifp->if_qflush entry point for lagg(4) is no-op.
1435 */
1436 static void
1437 lagg_qflush(struct ifnet *ifp __unused)
1438 {
1439 }
1440 #endif
1441 static void
lagg_input(struct ifnet * ifp,struct mbuf * m)1442 lagg_input(struct ifnet *ifp, struct mbuf *m)
1443 {
1444 struct lagg_port *lp = ifp->if_lagg;
1445 struct lagg_softc *sc = lp->lp_softc;
1446 struct ifnet *scifp = sc->sc_ifp;
1447
1448 LAGG_RLOCK(sc);
1449 if ((scifp->if_flags & IFF_RUNNING) == 0 ||
1450 (lp->lp_flags & LAGG_PORT_DISABLED) ||
1451 sc->sc_proto == LAGG_PROTO_NONE) {
1452 LAGG_RUNLOCK(sc);
1453 m_freem(m);
1454 return;
1455 }
1456
1457 BPF_MTAP(scifp, m);
1458 m = (*sc->sc_input)(sc, lp, m);
1459
1460 LAGG_RUNLOCK(sc);
1461
1462 if (m != NULL) {
1463 IFNET_STAT_INC(ifp, ipackets, 1);
1464 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len);
1465
1466 if (scifp->if_flags & IFF_MONITOR) {
1467 m_freem(m);
1468 m = NULL;
1469 }
1470 ether_reinput_oncpu(scifp, m, REINPUT_RUNBPF);
1471 }
1472 }
1473
1474 static int
lagg_media_change(struct ifnet * ifp)1475 lagg_media_change(struct ifnet *ifp)
1476 {
1477 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1478
1479 if (sc->sc_ifflags & IFF_DEBUG)
1480 kprintf("%s\n", __func__);
1481 /* Ignore */
1482 return (0);
1483 }
1484
1485 static void
lagg_media_status(struct ifnet * ifp,struct ifmediareq * imr)1486 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1487 {
1488 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1489 struct lagg_port *lp;
1490
1491 imr->ifm_status = IFM_AVALID;
1492 imr->ifm_active = IFM_ETHER | IFM_AUTO;
1493
1494 LAGG_RLOCK(sc);
1495 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1496 if (LAGG_PORTACTIVE(lp))
1497 imr->ifm_status |= IFM_ACTIVE;
1498 }
1499 LAGG_RUNLOCK(sc);
1500 }
1501
1502 static void
lagg_linkstate(struct lagg_softc * sc)1503 lagg_linkstate(struct lagg_softc *sc)
1504 {
1505 struct lagg_port *lp;
1506 int new_link = LINK_STATE_DOWN;
1507 uint64_t speed;
1508
1509 /* Our link is considered up if at least one of our ports is active */
1510 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1511 if (lp->lp_link_state == LINK_STATE_UP) {
1512 new_link = LINK_STATE_UP;
1513 break;
1514 }
1515 }
1516 if_link_state_change(sc->sc_ifp);
1517
1518 /* Update if_baudrate to reflect the max possible speed */
1519 switch (sc->sc_proto) {
1520 case LAGG_PROTO_FAILOVER:
1521 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1522 sc->sc_primary->lp_ifp->if_baudrate : 0;
1523 break;
1524 case LAGG_PROTO_ROUNDROBIN:
1525 case LAGG_PROTO_LOADBALANCE:
1526 case LAGG_PROTO_ETHERCHANNEL:
1527 speed = 0;
1528 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1529 speed += lp->lp_ifp->if_baudrate;
1530 sc->sc_ifp->if_baudrate = speed;
1531 break;
1532 case LAGG_PROTO_LACP:
1533 /* LACP updates if_baudrate itself */
1534 break;
1535 }
1536 }
1537
1538 #if 0 /* XXX */
1539 static void
1540 lagg_port_state(struct ifnet *ifp, int state)
1541 {
1542 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1543 struct lagg_softc *sc = NULL;
1544
1545 if (lp != NULL)
1546 sc = lp->lp_softc;
1547 if (sc == NULL)
1548 return;
1549
1550 LAGG_WLOCK(sc);
1551 lagg_linkstate(sc);
1552 if (sc->sc_linkstate != NULL)
1553 (*sc->sc_linkstate)(lp);
1554 LAGG_WUNLOCK(sc);
1555 }
1556 #endif
1557
1558 struct lagg_port *
lagg_link_active(struct lagg_softc * sc,struct lagg_port * lp)1559 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1560 {
1561 struct lagg_port *lp_next, *rval = NULL;
1562 // int new_link = LINK_STATE_DOWN;
1563
1564 LAGG_RLOCK_ASSERT(sc);
1565 /*
1566 * Search a port which reports an active link state.
1567 */
1568
1569 if (lp == NULL)
1570 goto search;
1571 if (LAGG_PORTACTIVE(lp)) {
1572 rval = lp;
1573 goto found;
1574 }
1575 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1576 LAGG_PORTACTIVE(lp_next)) {
1577 rval = lp_next;
1578 goto found;
1579 }
1580
1581 search:
1582 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1583 if (LAGG_PORTACTIVE(lp_next)) {
1584 rval = lp_next;
1585 goto found;
1586 }
1587 }
1588
1589 found:
1590 if (rval != NULL) {
1591 /*
1592 * The IEEE 802.1D standard assumes that a lagg with
1593 * multiple ports is always full duplex. This is valid
1594 * for load sharing laggs and if at least two links
1595 * are active. Unfortunately, checking the latter would
1596 * be too expensive at this point.
1597 XXX
1598 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1599 (sc->sc_count > 1))
1600 new_link = LINK_STATE_FULL_DUPLEX;
1601 else
1602 new_link = rval->lp_link_state;
1603 */
1604 }
1605
1606 return (rval);
1607 }
1608
1609 static const void *
lagg_gethdr(struct mbuf * m,u_int off,u_int len,void * buf)1610 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1611 {
1612 if (m->m_pkthdr.len < (off + len)) {
1613 return (NULL);
1614 } else if (m->m_len < (off + len)) {
1615 m_copydata(m, off, len, buf);
1616 return (buf);
1617 }
1618 return (mtod(m, char *) + off);
1619 }
1620
1621 static int
lagg_sysctl_active(SYSCTL_HANDLER_ARGS)1622 lagg_sysctl_active(SYSCTL_HANDLER_ARGS)
1623 {
1624 struct lagg_softc *sc = (struct lagg_softc *)arg1;
1625 struct lagg_port *lp;
1626 int error;
1627
1628 /* LACP tracks active links automatically, the others do not */
1629 if (sc->sc_proto != LAGG_PROTO_LACP) {
1630 sc->sc_active = 0;
1631 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1632 sc->sc_active += LAGG_PORTACTIVE(lp);
1633 }
1634
1635 error = sysctl_handle_int(oidp, &sc->sc_active, 0, req);
1636 if ((error) || (req->newptr == NULL))
1637 return (error);
1638
1639 return (0);
1640 }
1641
1642 uint32_t
lagg_hashmbuf(struct lagg_softc * sc,struct mbuf * m,uint32_t key)1643 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
1644 {
1645 uint16_t etype;
1646 uint32_t p = key;
1647 int off;
1648 struct ether_header *eh;
1649 const struct ether_vlan_header *vlan;
1650 #ifdef INET
1651 const struct ip *ip;
1652 const uint32_t *ports;
1653 int iphlen;
1654 #endif
1655 #ifdef INET6
1656 const struct ip6_hdr *ip6;
1657 uint32_t flow;
1658 #endif
1659 union {
1660 #ifdef INET
1661 struct ip ip;
1662 #endif
1663 #ifdef INET6
1664 struct ip6_hdr ip6;
1665 #endif
1666 struct ether_vlan_header vlan;
1667 uint32_t port;
1668 } buf;
1669
1670
1671 off = sizeof(*eh);
1672 if (m->m_len < off)
1673 goto out;
1674 eh = mtod(m, struct ether_header *);
1675 etype = ntohs(eh->ether_type);
1676 if (sc->sc_flags & LAGG_F_HASHL2) {
1677 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p);
1678 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1679 }
1680
1681 /* Special handling for encapsulating VLAN frames */
1682 #if 0 /* XXX */
1683 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) {
1684 p = hash32_buf(&m->m_pkthdr.ether_vtag,
1685 sizeof(m->m_pkthdr.ether_vtag), p);
1686 } else
1687 #endif
1688 if (etype == ETHERTYPE_VLAN) {
1689 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf);
1690 if (vlan == NULL)
1691 goto out;
1692
1693 if (sc->sc_flags & LAGG_F_HASHL2)
1694 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1695 etype = ntohs(vlan->evl_proto);
1696 off += sizeof(*vlan) - sizeof(*eh);
1697 }
1698
1699 switch (etype) {
1700 #ifdef INET
1701 case ETHERTYPE_IP:
1702 ip = lagg_gethdr(m, off, sizeof(*ip), &buf);
1703 if (ip == NULL)
1704 goto out;
1705
1706 if (sc->sc_flags & LAGG_F_HASHL3) {
1707 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1708 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1709 }
1710 if (!(sc->sc_flags & LAGG_F_HASHL4))
1711 break;
1712 switch (ip->ip_p) {
1713 case IPPROTO_TCP:
1714 case IPPROTO_UDP:
1715 iphlen = ip->ip_hl << 2;
1716 if (iphlen < sizeof(*ip))
1717 break;
1718 off += iphlen;
1719 ports = lagg_gethdr(m, off, sizeof(*ports), &buf);
1720 if (ports == NULL)
1721 break;
1722 p = hash32_buf(ports, sizeof(*ports), p);
1723 break;
1724 }
1725 break;
1726 #endif
1727 #ifdef INET6
1728 case ETHERTYPE_IPV6:
1729 if (!(sc->sc_flags & LAGG_F_HASHL3))
1730 break;
1731 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf);
1732 if (ip6 == NULL)
1733 goto out;
1734
1735 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1736 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1737 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1738 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */
1739 break;
1740 #endif
1741 }
1742 out:
1743 return (p);
1744 }
1745
1746 static void
lagg_start(struct ifnet * ifp,struct ifaltq_subque * ifsq)1747 lagg_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1748 {
1749 struct lagg_softc *sc = ifp->if_softc;
1750 struct mbuf *m;
1751 struct ifnet *ifp_p;
1752 struct netmsg_packet *nmp;
1753 lwkt_port_t p_port;
1754
1755 ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
1756 ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
1757
1758 if (((ifp->if_flags & IFF_RUNNING) == 0)
1759 || (sc->sc_proto == LAGG_PROTO_NONE)
1760 || (sc->sc_count == 0)) {
1761 ifsq_purge(ifsq);
1762 return;
1763 }
1764
1765
1766 LAGG_RLOCK(sc);
1767 for (;;) {
1768 m = ifsq_dequeue(ifsq);
1769 if (m == NULL){
1770 break;
1771 }
1772 // Choose output port
1773 ifp_p = (*sc->sc_select_tx_port)(sc, m);
1774
1775 if (ifp_p == NULL) {
1776 ifsq_purge(ifsq);
1777 break;
1778 }
1779 p_port = netisr_cpuport(
1780 ifsq_get_cpuid(ifq_get_subq_default(&ifp_p->if_snd)));
1781
1782 BPF_MTAP(ifp, m);
1783
1784 nmp = &m->m_hdr.mh_netmsg;
1785
1786 netmsg_init(&nmp->base, NULL, &netisr_apanic_rport,
1787 0, lagg_start_dispatch);
1788 nmp->nm_packet = m;
1789 nmp->base.lmsg.u.ms_resultp = ifp_p;
1790
1791 lwkt_sendmsg(p_port, &nmp->base.lmsg);
1792 IFNET_STAT_INC(ifp, opackets, 1);
1793 }
1794 LAGG_RUNLOCK(sc);
1795 }
1796
1797
1798 static void
lagg_start_dispatch(netmsg_t msg)1799 lagg_start_dispatch(netmsg_t msg)
1800 {
1801 struct netmsg_packet *nmp = &msg->packet;
1802 struct mbuf *m;
1803 struct ifnet *ifp;
1804 struct altq_pktattr pktattr;
1805
1806 m = nmp->nm_packet;
1807 ifp = msg->lmsg.u.ms_resultp;
1808
1809 M_ASSERTPKTHDR(m);
1810
1811 /* Does altq mix with lacp? */
1812 if (ifq_is_enabled(&ifp->if_snd))
1813 altq_etherclassify(&ifp->if_snd, m, &pktattr);
1814
1815 ifq_dispatch(ifp, m, &pktattr);
1816 }
1817
1818
1819 int
lagg_enqueue(struct ifnet * ifp,struct mbuf * m)1820 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1821 {
1822 struct altq_pktattr pktattr;
1823
1824 if (ifq_is_enabled(&ifp->if_snd))
1825 altq_etherclassify(&ifp->if_snd, m, &pktattr);
1826
1827 ifq_dispatch(ifp, m, &pktattr);
1828 return 0;
1829 }
1830
1831 /*
1832 * Simple round robin aggregation
1833 */
1834 static int
lagg_rr_attach(struct lagg_softc * sc)1835 lagg_rr_attach(struct lagg_softc *sc)
1836 {
1837 sc->sc_detach = lagg_rr_detach;
1838 sc->sc_input = lagg_rr_input;
1839 sc->sc_select_tx_port = lagg_rr_select_tx_port;
1840 sc->sc_port_create = NULL;
1841 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1842 sc->sc_seq = 0;
1843
1844 return (0);
1845 }
1846
1847 static int
lagg_rr_detach(struct lagg_softc * sc)1848 lagg_rr_detach(struct lagg_softc *sc)
1849 {
1850 return (0);
1851 }
1852
1853 static struct ifnet *
lagg_rr_select_tx_port(struct lagg_softc * sc,struct mbuf * m)1854 lagg_rr_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
1855 {
1856 struct lagg_port *lp;
1857 uint32_t p;
1858
1859 p = atomic_fetchadd_32(&sc->sc_seq, 1);
1860 p %= sc->sc_count;
1861 lp = SLIST_FIRST(&sc->sc_ports);
1862 while (p--)
1863 lp = SLIST_NEXT(lp, lp_entries);
1864
1865 /*
1866 * Check the port's link state. This will return the next active
1867 * port if the link is down or the port is NULL.
1868 */
1869 if ((lp = lagg_link_active(sc, lp)) == NULL) {
1870 return (NULL);
1871 }
1872
1873 return (lp->lp_ifp);
1874 }
1875
1876 static struct mbuf *
lagg_rr_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)1877 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1878 {
1879 struct ifnet *ifp = sc->sc_ifp;
1880
1881 /* Just pass in the packet to our lagg device */
1882 m->m_pkthdr.rcvif = ifp;
1883
1884 return (m);
1885 }
1886
1887 /*
1888 * Active failover
1889 */
1890
1891 static int
lagg_fail_attach(struct lagg_softc * sc)1892 lagg_fail_attach(struct lagg_softc *sc)
1893 {
1894 sc->sc_detach = lagg_fail_detach;
1895 sc->sc_select_tx_port = lagg_fail_select_tx_port;
1896 sc->sc_input = lagg_fail_input;
1897 sc->sc_port_create = NULL;
1898 sc->sc_port_destroy = NULL;
1899
1900 return (0);
1901 }
1902
1903 static int
lagg_fail_detach(struct lagg_softc * sc)1904 lagg_fail_detach(struct lagg_softc *sc)
1905 {
1906 return (0);
1907 }
1908
1909 struct ifnet *
lagg_fail_select_tx_port(struct lagg_softc * sc,struct mbuf * m)1910 lagg_fail_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
1911 {
1912 struct lagg_port *lp;
1913
1914 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL)
1915 return NULL;
1916
1917 return lp->lp_ifp;
1918 }
1919
1920 static struct mbuf *
lagg_fail_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)1921 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1922 {
1923 struct ifnet *ifp = sc->sc_ifp;
1924 struct lagg_port *tmp_tp;
1925
1926 if (lp == sc->sc_primary || lagg_failover_rx_all) {
1927 m->m_pkthdr.rcvif = ifp;
1928 return (m);
1929 }
1930
1931 if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1932 tmp_tp = lagg_link_active(sc, sc->sc_primary);
1933 /*
1934 * If tmp_tp is null, we've recieved a packet when all
1935 * our links are down. Weird, but process it anyways.
1936 */
1937 if ((tmp_tp == NULL || tmp_tp == lp)) {
1938 m->m_pkthdr.rcvif = ifp;
1939 return (m);
1940 }
1941 }
1942
1943 m_freem(m);
1944 return (NULL);
1945 }
1946
1947 /*
1948 * Loadbalancing
1949 */
1950
1951 static int
lagg_lb_attach(struct lagg_softc * sc)1952 lagg_lb_attach(struct lagg_softc *sc)
1953 {
1954 struct lagg_port *lp;
1955 struct lagg_lb *lb;
1956
1957 if ((lb = (struct lagg_lb *)kmalloc(sizeof(struct lagg_lb),
1958 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1959 return (ENOMEM);
1960
1961 sc->sc_detach = lagg_lb_detach;
1962 sc->sc_select_tx_port = lagg_lb_select_tx_port;
1963 sc->sc_input = lagg_lb_input;
1964 sc->sc_port_create = lagg_lb_port_create;
1965 sc->sc_port_destroy = lagg_lb_port_destroy;
1966 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1967
1968 lb->lb_key = karc4random();
1969 sc->sc_psc = (caddr_t)lb;
1970
1971 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1972 lagg_lb_port_create(lp);
1973
1974 return (0);
1975 }
1976
1977 static int
lagg_lb_detach(struct lagg_softc * sc)1978 lagg_lb_detach(struct lagg_softc *sc)
1979 {
1980 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1981 if (lb != NULL)
1982 kfree(lb, M_DEVBUF);
1983 return (0);
1984 }
1985
1986 static int
lagg_lb_porttable(struct lagg_softc * sc,struct lagg_port * lp)1987 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1988 {
1989 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1990 struct lagg_port *lp_next;
1991 int i = 0;
1992
1993 bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1994 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1995 if (lp_next == lp)
1996 continue;
1997 if (i >= LAGG_MAX_PORTS)
1998 return (EINVAL);
1999 if (sc->sc_ifflags & IFF_DEBUG)
2000 kprintf("%s: port %s at index %d\n",
2001 sc->sc_ifname, lp_next->lp_ifname, i);
2002 lb->lb_ports[i++] = lp_next;
2003 }
2004
2005 return (0);
2006 }
2007
2008 static int
lagg_lb_port_create(struct lagg_port * lp)2009 lagg_lb_port_create(struct lagg_port *lp)
2010 {
2011 struct lagg_softc *sc = lp->lp_softc;
2012 return (lagg_lb_porttable(sc, NULL));
2013 }
2014
2015 static void
lagg_lb_port_destroy(struct lagg_port * lp)2016 lagg_lb_port_destroy(struct lagg_port *lp)
2017 {
2018 struct lagg_softc *sc = lp->lp_softc;
2019 lagg_lb_porttable(sc, lp);
2020 }
2021
2022
2023 struct ifnet *
lagg_lb_select_tx_port(struct lagg_softc * sc,struct mbuf * m)2024 lagg_lb_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
2025 {
2026 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2027 struct lagg_port *lp = NULL;
2028 uint32_t p = 0;
2029
2030 /* XXX
2031 if (sc->use_flowid && (m->m_flags & M_FLOWID))
2032 p = m->m_pkthdr.flowid;
2033 else
2034 */
2035 p = lagg_hashmbuf(sc, m, lb->lb_key);
2036 p %= sc->sc_count;
2037 lp = lb->lb_ports[p];
2038
2039 /*
2040 * Check the port's link state. This will return the next active
2041 * port if the link is down or the port is NULL.
2042 */
2043 if ((lp = lagg_link_active(sc, lp)) == NULL)
2044 return NULL;
2045
2046 return lp->lp_ifp;
2047 }
2048
2049 static struct mbuf *
lagg_lb_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)2050 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2051 {
2052 struct ifnet *ifp = sc->sc_ifp;
2053
2054 /* Just pass in the packet to our lagg device */
2055 m->m_pkthdr.rcvif = ifp;
2056
2057 return (m);
2058 }
2059
2060 /*
2061 * 802.3ad LACP
2062 */
2063 static int
lagg_lacp_attach(struct lagg_softc * sc)2064 lagg_lacp_attach(struct lagg_softc *sc)
2065 {
2066 struct lagg_port *lp;
2067 int error;
2068
2069 sc->sc_detach = lagg_lacp_detach;
2070 sc->sc_port_create = lacp_port_create;
2071 sc->sc_port_destroy = lacp_port_destroy;
2072 sc->sc_linkstate = lacp_linkstate;
2073 sc->sc_select_tx_port = lagg_lacp_select_tx_port;
2074 sc->sc_input = lagg_lacp_input;
2075 sc->sc_init = lacp_init;
2076 sc->sc_stop = lacp_stop;
2077 sc->sc_lladdr = lagg_lacp_lladdr;
2078 sc->sc_req = lacp_req;
2079 sc->sc_portreq = lacp_portreq;
2080
2081 error = lacp_attach(sc);
2082 if (error)
2083 return (error);
2084
2085 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2086 lacp_port_create(lp);
2087
2088 return (error);
2089 }
2090
2091 static int
lagg_lacp_detach(struct lagg_softc * sc)2092 lagg_lacp_detach(struct lagg_softc *sc)
2093 {
2094 struct lagg_port *lp;
2095 int error;
2096
2097 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2098 lacp_port_destroy(lp);
2099
2100 /* unlocking is safe here */
2101 LAGG_WUNLOCK(sc);
2102 error = lacp_detach(sc);
2103 LAGG_WLOCK(sc);
2104
2105 return (error);
2106 }
2107
2108 static void
lagg_lacp_lladdr(struct lagg_softc * sc)2109 lagg_lacp_lladdr(struct lagg_softc *sc)
2110 {
2111 struct lagg_port *lp;
2112
2113 /* purge all the lacp ports */
2114 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2115 lacp_port_destroy(lp);
2116
2117 /* add them back in */
2118 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2119 lacp_port_create(lp);
2120 }
2121
2122 struct ifnet *
lagg_lacp_select_tx_port(struct lagg_softc * sc,struct mbuf * m)2123 lagg_lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
2124 {
2125 struct lagg_port *lp;
2126 lp = lacp_select_tx_port(sc, m);
2127 if (lp == NULL)
2128 return NULL;
2129
2130 return lp->lp_ifp;
2131 }
2132
2133 static struct mbuf *
lagg_lacp_input(struct lagg_softc * sc,struct lagg_port * lp,struct mbuf * m)2134 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2135 {
2136 struct ifnet *ifp = sc->sc_ifp;
2137 struct ether_header *eh;
2138 u_short etype;
2139
2140 eh = mtod(m, struct ether_header *);
2141 etype = ntohs(eh->ether_type);
2142
2143 /* Tap off LACP control messages */
2144 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2145 m = lacp_input(lp, m);
2146 if (m == NULL)
2147 return (NULL);
2148 }
2149
2150 /*
2151 * If the port is not collecting or not in the active aggregator then
2152 * free and return.
2153 */
2154 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2155 m_freem(m);
2156 return (NULL);
2157 }
2158
2159 m->m_pkthdr.rcvif = ifp;
2160 return (m);
2161 }
2162
2163 static void
lagg_callout(void * arg)2164 lagg_callout(void *arg)
2165 {
2166 struct lagg_softc *sc = (struct lagg_softc *)arg;
2167 #if 0 /* XXX */
2168 struct ifnet *ifp = sc->sc_ifp;
2169
2170 ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets);
2171 ifp->if_opackets = counter_u64_fetch(sc->sc_opackets);
2172 ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes);
2173 ifp->if_obytes = counter_u64_fetch(sc->sc_obytes);
2174 #endif
2175
2176 callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
2177 }
2178