FreeBSD/Linux Kernel Cross Reference
sys/net/if_lagg.c
1 /* $OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $ */
2
3 /*
4 * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org>
5 * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org>
6 * Copyright (c) 2014, 2016 Marcelo Araujo <araujo@FreeBSD.org>
7 *
8 * Permission to use, copy, modify, and distribute this software for any
9 * purpose with or without fee is hereby granted, provided that the above
10 * copyright notice and this permission notice appear in all copies.
11 *
12 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19 */
20
21 #include <sys/cdefs.h>
22 __FBSDID("$FreeBSD: releng/11.1/sys/net/if_lagg.c 319697 2017-06-08 14:34:39Z mav $");
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/priv.h>
37 #include <sys/systm.h>
38 #include <sys/proc.h>
39 #include <sys/lock.h>
40 #include <sys/rmlock.h>
41 #include <sys/sx.h>
42 #include <sys/taskqueue.h>
43 #include <sys/eventhandler.h>
44
45 #include <net/ethernet.h>
46 #include <net/if.h>
47 #include <net/if_clone.h>
48 #include <net/if_arp.h>
49 #include <net/if_dl.h>
50 #include <net/if_media.h>
51 #include <net/if_types.h>
52 #include <net/if_var.h>
53 #include <net/bpf.h>
54 #include <net/vnet.h>
55
56 #if defined(INET) || defined(INET6)
57 #include <netinet/in.h>
58 #include <netinet/ip.h>
59 #endif
60 #ifdef INET
61 #include <netinet/in_systm.h>
62 #include <netinet/if_ether.h>
63 #endif
64
65 #ifdef INET6
66 #include <netinet/ip6.h>
67 #include <netinet6/in6_var.h>
68 #include <netinet6/in6_ifattach.h>
69 #endif
70
71 #include <net/if_vlan_var.h>
72 #include <net/if_lagg.h>
73 #include <net/ieee8023ad_lacp.h>
74
75 /* Special flags we should propagate to the lagg ports. */
76 static struct {
77 int flag;
78 int (*func)(struct ifnet *, int);
79 } lagg_pflags[] = {
80 {IFF_PROMISC, ifpromisc},
81 {IFF_ALLMULTI, if_allmulti},
82 {0, NULL}
83 };
84
85 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
86 #define V_lagg_list VNET(lagg_list)
87 static VNET_DEFINE(struct mtx, lagg_list_mtx);
88 #define V_lagg_list_mtx VNET(lagg_list_mtx)
89 #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \
90 "if_lagg list", NULL, MTX_DEF)
91 #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx)
92 #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx)
93 #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx)
94 eventhandler_tag lagg_detach_cookie = NULL;
95
96 static int lagg_clone_create(struct if_clone *, int, caddr_t);
97 static void lagg_clone_destroy(struct ifnet *);
98 static VNET_DEFINE(struct if_clone *, lagg_cloner);
99 #define V_lagg_cloner VNET(lagg_cloner)
100 static const char laggname[] = "lagg";
101
102 static void lagg_capabilities(struct lagg_softc *);
103 static int lagg_port_create(struct lagg_softc *, struct ifnet *);
104 static int lagg_port_destroy(struct lagg_port *, int);
105 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
106 static void lagg_linkstate(struct lagg_softc *);
107 static void lagg_port_state(struct ifnet *, int);
108 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
109 static int lagg_port_output(struct ifnet *, struct mbuf *,
110 const struct sockaddr *, struct route *);
111 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *);
112 #ifdef LAGG_PORT_STACKING
113 static int lagg_port_checkstacking(struct lagg_softc *);
114 #endif
115 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *);
116 static void lagg_init(void *);
117 static void lagg_stop(struct lagg_softc *);
118 static int lagg_ioctl(struct ifnet *, u_long, caddr_t);
119 static int lagg_setmulti(struct lagg_port *);
120 static int lagg_clrmulti(struct lagg_port *);
121 static int lagg_setcaps(struct lagg_port *, int cap);
122 static int lagg_setflag(struct lagg_port *, int, int,
123 int (*func)(struct ifnet *, int));
124 static int lagg_setflags(struct lagg_port *, int status);
125 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
126 static int lagg_transmit(struct ifnet *, struct mbuf *);
127 static void lagg_qflush(struct ifnet *);
128 static int lagg_media_change(struct ifnet *);
129 static void lagg_media_status(struct ifnet *, struct ifmediareq *);
130 static struct lagg_port *lagg_link_active(struct lagg_softc *,
131 struct lagg_port *);
132
133 /* Simple round robin */
134 static void lagg_rr_attach(struct lagg_softc *);
135 static int lagg_rr_start(struct lagg_softc *, struct mbuf *);
136 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
137 struct mbuf *);
138
139 /* Active failover */
140 static int lagg_fail_start(struct lagg_softc *, struct mbuf *);
141 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
142 struct mbuf *);
143
144 /* Loadbalancing */
145 static void lagg_lb_attach(struct lagg_softc *);
146 static void lagg_lb_detach(struct lagg_softc *);
147 static int lagg_lb_port_create(struct lagg_port *);
148 static void lagg_lb_port_destroy(struct lagg_port *);
149 static int lagg_lb_start(struct lagg_softc *, struct mbuf *);
150 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
151 struct mbuf *);
152 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
153
154 /* Broadcast */
155 static int lagg_bcast_start(struct lagg_softc *, struct mbuf *);
156 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
157 struct mbuf *);
158
159 /* 802.3ad LACP */
160 static void lagg_lacp_attach(struct lagg_softc *);
161 static void lagg_lacp_detach(struct lagg_softc *);
162 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *);
163 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
164 struct mbuf *);
165 static void lagg_lacp_lladdr(struct lagg_softc *);
166
167 /* lagg protocol table */
168 static const struct lagg_proto {
169 lagg_proto pr_num;
170 void (*pr_attach)(struct lagg_softc *);
171 void (*pr_detach)(struct lagg_softc *);
172 int (*pr_start)(struct lagg_softc *, struct mbuf *);
173 struct mbuf * (*pr_input)(struct lagg_softc *, struct lagg_port *,
174 struct mbuf *);
175 int (*pr_addport)(struct lagg_port *);
176 void (*pr_delport)(struct lagg_port *);
177 void (*pr_linkstate)(struct lagg_port *);
178 void (*pr_init)(struct lagg_softc *);
179 void (*pr_stop)(struct lagg_softc *);
180 void (*pr_lladdr)(struct lagg_softc *);
181 void (*pr_request)(struct lagg_softc *, void *);
182 void (*pr_portreq)(struct lagg_port *, void *);
183 } lagg_protos[] = {
184 {
185 .pr_num = LAGG_PROTO_NONE
186 },
187 {
188 .pr_num = LAGG_PROTO_ROUNDROBIN,
189 .pr_attach = lagg_rr_attach,
190 .pr_start = lagg_rr_start,
191 .pr_input = lagg_rr_input,
192 },
193 {
194 .pr_num = LAGG_PROTO_FAILOVER,
195 .pr_start = lagg_fail_start,
196 .pr_input = lagg_fail_input,
197 },
198 {
199 .pr_num = LAGG_PROTO_LOADBALANCE,
200 .pr_attach = lagg_lb_attach,
201 .pr_detach = lagg_lb_detach,
202 .pr_start = lagg_lb_start,
203 .pr_input = lagg_lb_input,
204 .pr_addport = lagg_lb_port_create,
205 .pr_delport = lagg_lb_port_destroy,
206 },
207 {
208 .pr_num = LAGG_PROTO_LACP,
209 .pr_attach = lagg_lacp_attach,
210 .pr_detach = lagg_lacp_detach,
211 .pr_start = lagg_lacp_start,
212 .pr_input = lagg_lacp_input,
213 .pr_addport = lacp_port_create,
214 .pr_delport = lacp_port_destroy,
215 .pr_linkstate = lacp_linkstate,
216 .pr_init = lacp_init,
217 .pr_stop = lacp_stop,
218 .pr_lladdr = lagg_lacp_lladdr,
219 .pr_request = lacp_req,
220 .pr_portreq = lacp_portreq,
221 },
222 {
223 .pr_num = LAGG_PROTO_BROADCAST,
224 .pr_start = lagg_bcast_start,
225 .pr_input = lagg_bcast_input,
226 },
227 };
228
229 SYSCTL_DECL(_net_link);
230 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
231 "Link Aggregation");
232
233 /* Allow input on any failover links */
234 static VNET_DEFINE(int, lagg_failover_rx_all);
235 #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all)
236 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
237 &VNET_NAME(lagg_failover_rx_all), 0,
238 "Accept input from any interface in a failover lagg");
239
240 /* Default value for using flowid */
241 static VNET_DEFINE(int, def_use_flowid) = 1;
242 #define V_def_use_flowid VNET(def_use_flowid)
243 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
244 &VNET_NAME(def_use_flowid), 0,
245 "Default setting for using flow id for load sharing");
246
247 /* Default value for flowid shift */
248 static VNET_DEFINE(int, def_flowid_shift) = 16;
249 #define V_def_flowid_shift VNET(def_flowid_shift)
250 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
251 &VNET_NAME(def_flowid_shift), 0,
252 "Default setting for flowid shift for load sharing");
253
254 static void
255 vnet_lagg_init(const void *unused __unused)
256 {
257
258 LAGG_LIST_LOCK_INIT();
259 SLIST_INIT(&V_lagg_list);
260 V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
261 lagg_clone_destroy, 0);
262 }
263 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
264 vnet_lagg_init, NULL);
265
266 static void
267 vnet_lagg_uninit(const void *unused __unused)
268 {
269
270 if_clone_detach(V_lagg_cloner);
271 LAGG_LIST_LOCK_DESTROY();
272 }
273 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
274 vnet_lagg_uninit, NULL);
275
276 static int
277 lagg_modevent(module_t mod, int type, void *data)
278 {
279
280 switch (type) {
281 case MOD_LOAD:
282 lagg_input_p = lagg_input;
283 lagg_linkstate_p = lagg_port_state;
284 lagg_detach_cookie = EVENTHANDLER_REGISTER(
285 ifnet_departure_event, lagg_port_ifdetach, NULL,
286 EVENTHANDLER_PRI_ANY);
287 break;
288 case MOD_UNLOAD:
289 EVENTHANDLER_DEREGISTER(ifnet_departure_event,
290 lagg_detach_cookie);
291 lagg_input_p = NULL;
292 lagg_linkstate_p = NULL;
293 break;
294 default:
295 return (EOPNOTSUPP);
296 }
297 return (0);
298 }
299
300 static moduledata_t lagg_mod = {
301 "if_lagg",
302 lagg_modevent,
303 0
304 };
305
306 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
307 MODULE_VERSION(if_lagg, 1);
308
309 static void
310 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
311 {
312
313 LAGG_XLOCK_ASSERT(sc);
314 KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
315 __func__, sc));
316
317 if (sc->sc_ifflags & IFF_DEBUG)
318 if_printf(sc->sc_ifp, "using proto %u\n", pr);
319
320 if (lagg_protos[pr].pr_attach != NULL)
321 lagg_protos[pr].pr_attach(sc);
322 sc->sc_proto = pr;
323 }
324
325 static void
326 lagg_proto_detach(struct lagg_softc *sc)
327 {
328 lagg_proto pr;
329
330 LAGG_XLOCK_ASSERT(sc);
331 LAGG_WLOCK_ASSERT(sc);
332 pr = sc->sc_proto;
333 sc->sc_proto = LAGG_PROTO_NONE;
334
335 if (lagg_protos[pr].pr_detach != NULL)
336 lagg_protos[pr].pr_detach(sc);
337 else
338 LAGG_WUNLOCK(sc);
339 }
340
341 static int
342 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
343 {
344
345 return (lagg_protos[sc->sc_proto].pr_start(sc, m));
346 }
347
348 static struct mbuf *
349 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
350 {
351
352 return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
353 }
354
355 static int
356 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
357 {
358
359 if (lagg_protos[sc->sc_proto].pr_addport == NULL)
360 return (0);
361 else
362 return (lagg_protos[sc->sc_proto].pr_addport(lp));
363 }
364
365 static void
366 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
367 {
368
369 if (lagg_protos[sc->sc_proto].pr_delport != NULL)
370 lagg_protos[sc->sc_proto].pr_delport(lp);
371 }
372
373 static void
374 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
375 {
376
377 if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
378 lagg_protos[sc->sc_proto].pr_linkstate(lp);
379 }
380
381 static void
382 lagg_proto_init(struct lagg_softc *sc)
383 {
384
385 if (lagg_protos[sc->sc_proto].pr_init != NULL)
386 lagg_protos[sc->sc_proto].pr_init(sc);
387 }
388
389 static void
390 lagg_proto_stop(struct lagg_softc *sc)
391 {
392
393 if (lagg_protos[sc->sc_proto].pr_stop != NULL)
394 lagg_protos[sc->sc_proto].pr_stop(sc);
395 }
396
397 static void
398 lagg_proto_lladdr(struct lagg_softc *sc)
399 {
400
401 if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
402 lagg_protos[sc->sc_proto].pr_lladdr(sc);
403 }
404
405 static void
406 lagg_proto_request(struct lagg_softc *sc, void *v)
407 {
408
409 if (lagg_protos[sc->sc_proto].pr_request != NULL)
410 lagg_protos[sc->sc_proto].pr_request(sc, v);
411 }
412
413 static void
414 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
415 {
416
417 if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
418 lagg_protos[sc->sc_proto].pr_portreq(lp, v);
419 }
420
421 /*
422 * This routine is run via an vlan
423 * config EVENT
424 */
425 static void
426 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
427 {
428 struct lagg_softc *sc = ifp->if_softc;
429 struct lagg_port *lp;
430
431 if (ifp->if_softc != arg) /* Not our event */
432 return;
433
434 LAGG_SLOCK(sc);
435 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
436 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
437 LAGG_SUNLOCK(sc);
438 }
439
440 /*
441 * This routine is run via an vlan
442 * unconfig EVENT
443 */
444 static void
445 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
446 {
447 struct lagg_softc *sc = ifp->if_softc;
448 struct lagg_port *lp;
449
450 if (ifp->if_softc != arg) /* Not our event */
451 return;
452
453 LAGG_SLOCK(sc);
454 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
455 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
456 LAGG_SUNLOCK(sc);
457 }
458
459 static int
460 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
461 {
462 struct lagg_softc *sc;
463 struct ifnet *ifp;
464 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */
465
466 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
467 ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
468 if (ifp == NULL) {
469 free(sc, M_DEVBUF);
470 return (ENOSPC);
471 }
472 LAGG_LOCK_INIT(sc);
473 LAGG_SX_INIT(sc);
474
475 LAGG_XLOCK(sc);
476 if (V_def_use_flowid)
477 sc->sc_opts |= LAGG_OPT_USE_FLOWID;
478 sc->flowid_shift = V_def_flowid_shift;
479
480 /* Hash all layers by default */
481 sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
482
483 lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
484
485 SLIST_INIT(&sc->sc_ports);
486
487 /* Initialise pseudo media types */
488 ifmedia_init(&sc->sc_media, 0, lagg_media_change,
489 lagg_media_status);
490 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
491 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
492
493 if_initname(ifp, laggname, unit);
494 ifp->if_softc = sc;
495 ifp->if_transmit = lagg_transmit;
496 ifp->if_qflush = lagg_qflush;
497 ifp->if_init = lagg_init;
498 ifp->if_ioctl = lagg_ioctl;
499 ifp->if_get_counter = lagg_get_counter;
500 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
501 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
502
503 /*
504 * Attach as an ordinary ethernet device, children will be attached
505 * as special device IFT_IEEE8023ADLAG.
506 */
507 ether_ifattach(ifp, eaddr);
508
509 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
510 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
511 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
512 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
513
514 /* Insert into the global list of laggs */
515 LAGG_LIST_LOCK();
516 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
517 LAGG_LIST_UNLOCK();
518 LAGG_XUNLOCK(sc);
519
520 return (0);
521 }
522
523 static void
524 lagg_clone_destroy(struct ifnet *ifp)
525 {
526 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
527 struct lagg_port *lp;
528
529 LAGG_XLOCK(sc);
530 sc->sc_destroying = 1;
531 lagg_stop(sc);
532 ifp->if_flags &= ~IFF_UP;
533
534 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
535 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
536
537 /* Shutdown and remove lagg ports */
538 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
539 lagg_port_destroy(lp, 1);
540
541 /* Unhook the aggregation protocol */
542 LAGG_WLOCK(sc);
543 lagg_proto_detach(sc);
544 LAGG_UNLOCK_ASSERT(sc);
545 LAGG_XUNLOCK(sc);
546
547 ifmedia_removeall(&sc->sc_media);
548 ether_ifdetach(ifp);
549 if_free(ifp);
550
551 LAGG_LIST_LOCK();
552 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
553 LAGG_LIST_UNLOCK();
554
555 LAGG_SX_DESTROY(sc);
556 LAGG_LOCK_DESTROY(sc);
557 free(sc, M_DEVBUF);
558 }
559
560 static void
561 lagg_capabilities(struct lagg_softc *sc)
562 {
563 struct lagg_port *lp;
564 int cap, ena, pena;
565 uint64_t hwa;
566 struct ifnet_hw_tsomax hw_tsomax;
567
568 LAGG_XLOCK_ASSERT(sc);
569
570 /* Get common enabled capabilities for the lagg ports */
571 ena = ~0;
572 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
573 ena &= lp->lp_ifp->if_capenable;
574 ena = (ena == ~0 ? 0 : ena);
575
576 /*
577 * Apply common enabled capabilities back to the lagg ports.
578 * May require several iterations if they are dependent.
579 */
580 do {
581 pena = ena;
582 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
583 lagg_setcaps(lp, ena);
584 ena &= lp->lp_ifp->if_capenable;
585 }
586 } while (pena != ena);
587
588 /* Get other capabilities from the lagg ports */
589 cap = ~0;
590 hwa = ~(uint64_t)0;
591 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
592 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
593 cap &= lp->lp_ifp->if_capabilities;
594 hwa &= lp->lp_ifp->if_hwassist;
595 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
596 }
597 cap = (cap == ~0 ? 0 : cap);
598 hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
599
600 if (sc->sc_ifp->if_capabilities != cap ||
601 sc->sc_ifp->if_capenable != ena ||
602 sc->sc_ifp->if_hwassist != hwa ||
603 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
604 sc->sc_ifp->if_capabilities = cap;
605 sc->sc_ifp->if_capenable = ena;
606 sc->sc_ifp->if_hwassist = hwa;
607 getmicrotime(&sc->sc_ifp->if_lastchange);
608
609 if (sc->sc_ifflags & IFF_DEBUG)
610 if_printf(sc->sc_ifp,
611 "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
612 }
613 }
614
615 static int
616 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
617 {
618 struct lagg_softc *sc_ptr;
619 struct lagg_port *lp, *tlp;
620 int error, i;
621 uint64_t *pval;
622
623 LAGG_XLOCK_ASSERT(sc);
624
625 /* Limit the maximal number of lagg ports */
626 if (sc->sc_count >= LAGG_MAX_PORTS)
627 return (ENOSPC);
628
629 /* Check if port has already been associated to a lagg */
630 if (ifp->if_lagg != NULL) {
631 /* Port is already in the current lagg? */
632 lp = (struct lagg_port *)ifp->if_lagg;
633 if (lp->lp_softc == sc)
634 return (EEXIST);
635 return (EBUSY);
636 }
637
638 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */
639 if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
640 return (EPROTONOSUPPORT);
641
642 /* Allow the first Ethernet member to define the MTU */
643 if (SLIST_EMPTY(&sc->sc_ports))
644 sc->sc_ifp->if_mtu = ifp->if_mtu;
645 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
646 if_printf(sc->sc_ifp, "invalid MTU for %s\n",
647 ifp->if_xname);
648 return (EINVAL);
649 }
650
651 lp = malloc(sizeof(struct lagg_port), M_DEVBUF, M_WAITOK|M_ZERO);
652 lp->lp_softc = sc;
653
654 /* Check if port is a stacked lagg */
655 LAGG_LIST_LOCK();
656 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
657 if (ifp == sc_ptr->sc_ifp) {
658 LAGG_LIST_UNLOCK();
659 free(lp, M_DEVBUF);
660 return (EINVAL);
661 /* XXX disable stacking for the moment, its untested */
662 #ifdef LAGG_PORT_STACKING
663 lp->lp_flags |= LAGG_PORT_STACK;
664 if (lagg_port_checkstacking(sc_ptr) >=
665 LAGG_MAX_STACKING) {
666 LAGG_LIST_UNLOCK();
667 free(lp, M_DEVBUF);
668 return (E2BIG);
669 }
670 #endif
671 }
672 }
673 LAGG_LIST_UNLOCK();
674
675 if_ref(ifp);
676 lp->lp_ifp = ifp;
677
678 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
679 lp->lp_ifcapenable = ifp->if_capenable;
680 if (SLIST_EMPTY(&sc->sc_ports)) {
681 LAGG_WLOCK(sc);
682 bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
683 lagg_proto_lladdr(sc);
684 LAGG_WUNLOCK(sc);
685 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
686 } else {
687 if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
688 }
689 lagg_setflags(lp, 1);
690
691 LAGG_WLOCK(sc);
692 if (SLIST_EMPTY(&sc->sc_ports))
693 sc->sc_primary = lp;
694
695 /* Change the interface type */
696 lp->lp_iftype = ifp->if_type;
697 ifp->if_type = IFT_IEEE8023ADLAG;
698 ifp->if_lagg = lp;
699 lp->lp_ioctl = ifp->if_ioctl;
700 ifp->if_ioctl = lagg_port_ioctl;
701 lp->lp_output = ifp->if_output;
702 ifp->if_output = lagg_port_output;
703
704 /* Read port counters */
705 pval = lp->port_counters.val;
706 for (i = 0; i < IFCOUNTERS; i++, pval++)
707 *pval = ifp->if_get_counter(ifp, i);
708
709 /*
710 * Insert into the list of ports.
711 * Keep ports sorted by if_index. It is handy, when configuration
712 * is predictable and `ifconfig laggN create ...` command
713 * will lead to the same result each time.
714 */
715 SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
716 if (tlp->lp_ifp->if_index < ifp->if_index && (
717 SLIST_NEXT(tlp, lp_entries) == NULL ||
718 SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index >
719 ifp->if_index))
720 break;
721 }
722 if (tlp != NULL)
723 SLIST_INSERT_AFTER(tlp, lp, lp_entries);
724 else
725 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
726 sc->sc_count++;
727
728 lagg_setmulti(lp);
729
730 if ((error = lagg_proto_addport(sc, lp)) != 0) {
731 /* Remove the port, without calling pr_delport. */
732 lagg_port_destroy(lp, 0);
733 LAGG_UNLOCK_ASSERT(sc);
734 return (error);
735 }
736
737 LAGG_WUNLOCK(sc);
738
739 /* Update lagg capabilities */
740 lagg_capabilities(sc);
741 lagg_linkstate(sc);
742
743 return (0);
744 }
745
746 #ifdef LAGG_PORT_STACKING
747 static int
748 lagg_port_checkstacking(struct lagg_softc *sc)
749 {
750 struct lagg_softc *sc_ptr;
751 struct lagg_port *lp;
752 int m = 0;
753
754 LAGG_SXLOCK_ASSERT(sc);
755 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
756 if (lp->lp_flags & LAGG_PORT_STACK) {
757 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
758 m = MAX(m, lagg_port_checkstacking(sc_ptr));
759 }
760 }
761
762 return (m + 1);
763 }
764 #endif
765
766 static int
767 lagg_port_destroy(struct lagg_port *lp, int rundelport)
768 {
769 struct lagg_softc *sc = lp->lp_softc;
770 struct lagg_port *lp_ptr, *lp0;
771 struct ifnet *ifp = lp->lp_ifp;
772 uint64_t *pval, vdiff;
773 int i;
774
775 LAGG_XLOCK_ASSERT(sc);
776
777 if (rundelport) {
778 LAGG_WLOCK(sc);
779 lagg_proto_delport(sc, lp);
780 } else
781 LAGG_WLOCK_ASSERT(sc);
782
783 if (lp->lp_detaching == 0)
784 lagg_clrmulti(lp);
785
786 /* Restore interface */
787 ifp->if_type = lp->lp_iftype;
788 ifp->if_ioctl = lp->lp_ioctl;
789 ifp->if_output = lp->lp_output;
790 ifp->if_lagg = NULL;
791
792 /* Update detached port counters */
793 pval = lp->port_counters.val;
794 for (i = 0; i < IFCOUNTERS; i++, pval++) {
795 vdiff = ifp->if_get_counter(ifp, i) - *pval;
796 sc->detached_counters.val[i] += vdiff;
797 }
798
799 /* Finally, remove the port from the lagg */
800 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
801 sc->sc_count--;
802
803 /* Update the primary interface */
804 if (lp == sc->sc_primary) {
805 uint8_t lladdr[ETHER_ADDR_LEN];
806
807 if ((lp0 = SLIST_FIRST(&sc->sc_ports)) == NULL)
808 bzero(&lladdr, ETHER_ADDR_LEN);
809 else
810 bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN);
811 sc->sc_primary = lp0;
812 if (sc->sc_destroying == 0) {
813 bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
814 lagg_proto_lladdr(sc);
815 LAGG_WUNLOCK(sc);
816 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
817 } else
818 LAGG_WUNLOCK(sc);
819
820 /*
821 * Update lladdr for each port (new primary needs update
822 * as well, to switch from old lladdr to its 'real' one)
823 */
824 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
825 if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN);
826 } else
827 LAGG_WUNLOCK(sc);
828
829 if (lp->lp_ifflags)
830 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
831
832 if (lp->lp_detaching == 0) {
833 lagg_setflags(lp, 0);
834 lagg_setcaps(lp, lp->lp_ifcapenable);
835 if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN);
836 }
837
838 if_rele(ifp);
839 free(lp, M_DEVBUF);
840
841 /* Update lagg capabilities */
842 lagg_capabilities(sc);
843 lagg_linkstate(sc);
844
845 return (0);
846 }
847
848 static int
849 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
850 {
851 struct lagg_reqport *rp = (struct lagg_reqport *)data;
852 struct lagg_softc *sc;
853 struct lagg_port *lp = NULL;
854 int error = 0;
855
856 /* Should be checked by the caller */
857 if (ifp->if_type != IFT_IEEE8023ADLAG ||
858 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
859 goto fallback;
860
861 switch (cmd) {
862 case SIOCGLAGGPORT:
863 if (rp->rp_portname[0] == '\0' ||
864 ifunit(rp->rp_portname) != ifp) {
865 error = EINVAL;
866 break;
867 }
868
869 LAGG_SLOCK(sc);
870 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
871 error = ENOENT;
872 LAGG_SUNLOCK(sc);
873 break;
874 }
875
876 lagg_port2req(lp, rp);
877 LAGG_SUNLOCK(sc);
878 break;
879
880 case SIOCSIFCAP:
881 if (lp->lp_ioctl == NULL) {
882 error = EINVAL;
883 break;
884 }
885 error = (*lp->lp_ioctl)(ifp, cmd, data);
886 if (error)
887 break;
888
889 /* Update lagg interface capabilities */
890 LAGG_XLOCK(sc);
891 lagg_capabilities(sc);
892 LAGG_XUNLOCK(sc);
893 VLAN_CAPABILITIES(sc->sc_ifp);
894 break;
895
896 case SIOCSIFMTU:
897 /* Do not allow the MTU to be changed once joined */
898 error = EINVAL;
899 break;
900
901 default:
902 goto fallback;
903 }
904
905 return (error);
906
907 fallback:
908 if (lp != NULL && lp->lp_ioctl != NULL)
909 return ((*lp->lp_ioctl)(ifp, cmd, data));
910
911 return (EINVAL);
912 }
913
914 /*
915 * Requests counter @cnt data.
916 *
917 * Counter value is calculated the following way:
918 * 1) for each port, sum difference between current and "initial" measurements.
919 * 2) add lagg logical interface counters.
920 * 3) add data from detached_counters array.
921 *
922 * We also do the following things on ports attach/detach:
923 * 1) On port attach we store all counters it has into port_counter array.
924 * 2) On port detach we add the different between "initial" and
925 * current counters data to detached_counters array.
926 */
927 static uint64_t
928 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
929 {
930 struct lagg_softc *sc;
931 struct lagg_port *lp;
932 struct ifnet *lpifp;
933 struct rm_priotracker tracker;
934 uint64_t newval, oldval, vsum;
935
936 /* Revise this when we've got non-generic counters. */
937 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
938
939 sc = (struct lagg_softc *)ifp->if_softc;
940 LAGG_RLOCK(sc, &tracker);
941
942 vsum = 0;
943 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
944 /* Saved attached value */
945 oldval = lp->port_counters.val[cnt];
946 /* current value */
947 lpifp = lp->lp_ifp;
948 newval = lpifp->if_get_counter(lpifp, cnt);
949 /* Calculate diff and save new */
950 vsum += newval - oldval;
951 }
952
953 /*
954 * Add counter data which might be added by upper
955 * layer protocols operating on logical interface.
956 */
957 vsum += if_get_counter_default(ifp, cnt);
958
959 /*
960 * Add counter data from detached ports counters
961 */
962 vsum += sc->detached_counters.val[cnt];
963
964 LAGG_RUNLOCK(sc, &tracker);
965
966 return (vsum);
967 }
968
969 /*
970 * For direct output to child ports.
971 */
972 static int
973 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
974 const struct sockaddr *dst, struct route *ro)
975 {
976 struct lagg_port *lp = ifp->if_lagg;
977
978 switch (dst->sa_family) {
979 case pseudo_AF_HDRCMPLT:
980 case AF_UNSPEC:
981 return ((*lp->lp_output)(ifp, m, dst, ro));
982 }
983
984 /* drop any other frames */
985 m_freem(m);
986 return (ENETDOWN);
987 }
988
989 static void
990 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
991 {
992 struct lagg_port *lp;
993 struct lagg_softc *sc;
994
995 if ((lp = ifp->if_lagg) == NULL)
996 return;
997 /* If the ifnet is just being renamed, don't do anything. */
998 if (ifp->if_flags & IFF_RENAMING)
999 return;
1000
1001 sc = lp->lp_softc;
1002
1003 LAGG_XLOCK(sc);
1004 lp->lp_detaching = 1;
1005 lagg_port_destroy(lp, 1);
1006 LAGG_XUNLOCK(sc);
1007 VLAN_CAPABILITIES(sc->sc_ifp);
1008 }
1009
1010 static void
1011 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1012 {
1013 struct lagg_softc *sc = lp->lp_softc;
1014
1015 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1016 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1017 rp->rp_prio = lp->lp_prio;
1018 rp->rp_flags = lp->lp_flags;
1019 lagg_proto_portreq(sc, lp, &rp->rp_psc);
1020
1021 /* Add protocol specific flags */
1022 switch (sc->sc_proto) {
1023 case LAGG_PROTO_FAILOVER:
1024 if (lp == sc->sc_primary)
1025 rp->rp_flags |= LAGG_PORT_MASTER;
1026 if (lp == lagg_link_active(sc, sc->sc_primary))
1027 rp->rp_flags |= LAGG_PORT_ACTIVE;
1028 break;
1029
1030 case LAGG_PROTO_ROUNDROBIN:
1031 case LAGG_PROTO_LOADBALANCE:
1032 case LAGG_PROTO_BROADCAST:
1033 if (LAGG_PORTACTIVE(lp))
1034 rp->rp_flags |= LAGG_PORT_ACTIVE;
1035 break;
1036
1037 case LAGG_PROTO_LACP:
1038 /* LACP has a different definition of active */
1039 if (lacp_isactive(lp))
1040 rp->rp_flags |= LAGG_PORT_ACTIVE;
1041 if (lacp_iscollecting(lp))
1042 rp->rp_flags |= LAGG_PORT_COLLECTING;
1043 if (lacp_isdistributing(lp))
1044 rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1045 break;
1046 }
1047
1048 }
1049
1050 static void
1051 lagg_init(void *xsc)
1052 {
1053 struct lagg_softc *sc = (struct lagg_softc *)xsc;
1054 struct ifnet *ifp = sc->sc_ifp;
1055 struct lagg_port *lp;
1056
1057 LAGG_XLOCK(sc);
1058 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1059 LAGG_XUNLOCK(sc);
1060 return;
1061 }
1062
1063 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1064
1065 /*
1066 * Update the port lladdrs if needed.
1067 * This might be if_setlladdr() notification
1068 * that lladdr has been changed.
1069 */
1070 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1071 if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1072 ETHER_ADDR_LEN) != 0)
1073 if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN);
1074 }
1075
1076 lagg_proto_init(sc);
1077
1078 LAGG_XUNLOCK(sc);
1079 }
1080
1081 static void
1082 lagg_stop(struct lagg_softc *sc)
1083 {
1084 struct ifnet *ifp = sc->sc_ifp;
1085
1086 LAGG_XLOCK_ASSERT(sc);
1087
1088 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1089 return;
1090
1091 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1092
1093 lagg_proto_stop(sc);
1094 }
1095
1096 static int
1097 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1098 {
1099 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1100 struct lagg_reqall *ra = (struct lagg_reqall *)data;
1101 struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1102 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1103 struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1104 struct ifreq *ifr = (struct ifreq *)data;
1105 struct lagg_port *lp;
1106 struct ifnet *tpif;
1107 struct thread *td = curthread;
1108 char *buf, *outbuf;
1109 int count, buflen, len, error = 0;
1110
1111 bzero(&rpbuf, sizeof(rpbuf));
1112
1113 switch (cmd) {
1114 case SIOCGLAGG:
1115 LAGG_SLOCK(sc);
1116 buflen = sc->sc_count * sizeof(struct lagg_reqport);
1117 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1118 ra->ra_proto = sc->sc_proto;
1119 lagg_proto_request(sc, &ra->ra_psc);
1120 count = 0;
1121 buf = outbuf;
1122 len = min(ra->ra_size, buflen);
1123 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1124 if (len < sizeof(rpbuf))
1125 break;
1126
1127 lagg_port2req(lp, &rpbuf);
1128 memcpy(buf, &rpbuf, sizeof(rpbuf));
1129 count++;
1130 buf += sizeof(rpbuf);
1131 len -= sizeof(rpbuf);
1132 }
1133 LAGG_SUNLOCK(sc);
1134 ra->ra_ports = count;
1135 ra->ra_size = count * sizeof(rpbuf);
1136 error = copyout(outbuf, ra->ra_port, ra->ra_size);
1137 free(outbuf, M_TEMP);
1138 break;
1139 case SIOCSLAGG:
1140 error = priv_check(td, PRIV_NET_LAGG);
1141 if (error)
1142 break;
1143 if (ra->ra_proto >= LAGG_PROTO_MAX) {
1144 error = EPROTONOSUPPORT;
1145 break;
1146 }
1147
1148 LAGG_XLOCK(sc);
1149 LAGG_WLOCK(sc);
1150 lagg_proto_detach(sc);
1151 LAGG_UNLOCK_ASSERT(sc);
1152 lagg_proto_attach(sc, ra->ra_proto);
1153 LAGG_XUNLOCK(sc);
1154 break;
1155 case SIOCGLAGGOPTS:
1156 LAGG_SLOCK(sc);
1157 ro->ro_opts = sc->sc_opts;
1158 if (sc->sc_proto == LAGG_PROTO_LACP) {
1159 struct lacp_softc *lsc;
1160
1161 lsc = (struct lacp_softc *)sc->sc_psc;
1162 if (lsc->lsc_debug.lsc_tx_test != 0)
1163 ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1164 if (lsc->lsc_debug.lsc_rx_test != 0)
1165 ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1166 if (lsc->lsc_strict_mode != 0)
1167 ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1168 if (lsc->lsc_fast_timeout != 0)
1169 ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1170
1171 ro->ro_active = sc->sc_active;
1172 } else {
1173 ro->ro_active = 0;
1174 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1175 ro->ro_active += LAGG_PORTACTIVE(lp);
1176 }
1177 ro->ro_bkt = sc->sc_bkt;
1178 ro->ro_flapping = sc->sc_flapping;
1179 ro->ro_flowid_shift = sc->flowid_shift;
1180 LAGG_SUNLOCK(sc);
1181 break;
1182 case SIOCSLAGGOPTS:
1183 if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) {
1184 if (ro->ro_bkt == 0)
1185 sc->sc_bkt = 1; // Minimum 1 packet per iface.
1186 else
1187 sc->sc_bkt = ro->ro_bkt;
1188 }
1189 error = priv_check(td, PRIV_NET_LAGG);
1190 if (error)
1191 break;
1192 if (ro->ro_opts == 0)
1193 break;
1194 /*
1195 * Set options. LACP options are stored in sc->sc_psc,
1196 * not in sc_opts.
1197 */
1198 int valid, lacp;
1199
1200 switch (ro->ro_opts) {
1201 case LAGG_OPT_USE_FLOWID:
1202 case -LAGG_OPT_USE_FLOWID:
1203 case LAGG_OPT_FLOWIDSHIFT:
1204 valid = 1;
1205 lacp = 0;
1206 break;
1207 case LAGG_OPT_LACP_TXTEST:
1208 case -LAGG_OPT_LACP_TXTEST:
1209 case LAGG_OPT_LACP_RXTEST:
1210 case -LAGG_OPT_LACP_RXTEST:
1211 case LAGG_OPT_LACP_STRICT:
1212 case -LAGG_OPT_LACP_STRICT:
1213 case LAGG_OPT_LACP_TIMEOUT:
1214 case -LAGG_OPT_LACP_TIMEOUT:
1215 valid = lacp = 1;
1216 break;
1217 default:
1218 valid = lacp = 0;
1219 break;
1220 }
1221
1222 LAGG_XLOCK(sc);
1223
1224 if (valid == 0 ||
1225 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1226 /* Invalid combination of options specified. */
1227 error = EINVAL;
1228 LAGG_XUNLOCK(sc);
1229 break; /* Return from SIOCSLAGGOPTS. */
1230 }
1231 /*
1232 * Store new options into sc->sc_opts except for
1233 * FLOWIDSHIFT and LACP options.
1234 */
1235 if (lacp == 0) {
1236 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1237 sc->flowid_shift = ro->ro_flowid_shift;
1238 else if (ro->ro_opts > 0)
1239 sc->sc_opts |= ro->ro_opts;
1240 else
1241 sc->sc_opts &= ~ro->ro_opts;
1242 } else {
1243 struct lacp_softc *lsc;
1244 struct lacp_port *lp;
1245
1246 lsc = (struct lacp_softc *)sc->sc_psc;
1247
1248 switch (ro->ro_opts) {
1249 case LAGG_OPT_LACP_TXTEST:
1250 lsc->lsc_debug.lsc_tx_test = 1;
1251 break;
1252 case -LAGG_OPT_LACP_TXTEST:
1253 lsc->lsc_debug.lsc_tx_test = 0;
1254 break;
1255 case LAGG_OPT_LACP_RXTEST:
1256 lsc->lsc_debug.lsc_rx_test = 1;
1257 break;
1258 case -LAGG_OPT_LACP_RXTEST:
1259 lsc->lsc_debug.lsc_rx_test = 0;
1260 break;
1261 case LAGG_OPT_LACP_STRICT:
1262 lsc->lsc_strict_mode = 1;
1263 break;
1264 case -LAGG_OPT_LACP_STRICT:
1265 lsc->lsc_strict_mode = 0;
1266 break;
1267 case LAGG_OPT_LACP_TIMEOUT:
1268 LACP_LOCK(lsc);
1269 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1270 lp->lp_state |= LACP_STATE_TIMEOUT;
1271 LACP_UNLOCK(lsc);
1272 lsc->lsc_fast_timeout = 1;
1273 break;
1274 case -LAGG_OPT_LACP_TIMEOUT:
1275 LACP_LOCK(lsc);
1276 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1277 lp->lp_state &= ~LACP_STATE_TIMEOUT;
1278 LACP_UNLOCK(lsc);
1279 lsc->lsc_fast_timeout = 0;
1280 break;
1281 }
1282 }
1283 LAGG_XUNLOCK(sc);
1284 break;
1285 case SIOCGLAGGFLAGS:
1286 rf->rf_flags = 0;
1287 LAGG_SLOCK(sc);
1288 if (sc->sc_flags & MBUF_HASHFLAG_L2)
1289 rf->rf_flags |= LAGG_F_HASHL2;
1290 if (sc->sc_flags & MBUF_HASHFLAG_L3)
1291 rf->rf_flags |= LAGG_F_HASHL3;
1292 if (sc->sc_flags & MBUF_HASHFLAG_L4)
1293 rf->rf_flags |= LAGG_F_HASHL4;
1294 LAGG_SUNLOCK(sc);
1295 break;
1296 case SIOCSLAGGHASH:
1297 error = priv_check(td, PRIV_NET_LAGG);
1298 if (error)
1299 break;
1300 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1301 error = EINVAL;
1302 break;
1303 }
1304 LAGG_XLOCK(sc);
1305 sc->sc_flags = 0;
1306 if (rf->rf_flags & LAGG_F_HASHL2)
1307 sc->sc_flags |= MBUF_HASHFLAG_L2;
1308 if (rf->rf_flags & LAGG_F_HASHL3)
1309 sc->sc_flags |= MBUF_HASHFLAG_L3;
1310 if (rf->rf_flags & LAGG_F_HASHL4)
1311 sc->sc_flags |= MBUF_HASHFLAG_L4;
1312 LAGG_XUNLOCK(sc);
1313 break;
1314 case SIOCGLAGGPORT:
1315 if (rp->rp_portname[0] == '\0' ||
1316 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1317 error = EINVAL;
1318 break;
1319 }
1320
1321 LAGG_SLOCK(sc);
1322 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1323 lp->lp_softc != sc) {
1324 error = ENOENT;
1325 LAGG_SUNLOCK(sc);
1326 if_rele(tpif);
1327 break;
1328 }
1329
1330 lagg_port2req(lp, rp);
1331 LAGG_SUNLOCK(sc);
1332 if_rele(tpif);
1333 break;
1334 case SIOCSLAGGPORT:
1335 error = priv_check(td, PRIV_NET_LAGG);
1336 if (error)
1337 break;
1338 if (rp->rp_portname[0] == '\0' ||
1339 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1340 error = EINVAL;
1341 break;
1342 }
1343 #ifdef INET6
1344 /*
1345 * A laggport interface should not have inet6 address
1346 * because two interfaces with a valid link-local
1347 * scope zone must not be merged in any form. This
1348 * restriction is needed to prevent violation of
1349 * link-local scope zone. Attempts to add a laggport
1350 * interface which has inet6 addresses triggers
1351 * removal of all inet6 addresses on the member
1352 * interface.
1353 */
1354 if (in6ifa_llaonifp(tpif)) {
1355 in6_ifdetach(tpif);
1356 if_printf(sc->sc_ifp,
1357 "IPv6 addresses on %s have been removed "
1358 "before adding it as a member to prevent "
1359 "IPv6 address scope violation.\n",
1360 tpif->if_xname);
1361 }
1362 #endif
1363 LAGG_XLOCK(sc);
1364 error = lagg_port_create(sc, tpif);
1365 LAGG_XUNLOCK(sc);
1366 if_rele(tpif);
1367 VLAN_CAPABILITIES(ifp);
1368 break;
1369 case SIOCSLAGGDELPORT:
1370 error = priv_check(td, PRIV_NET_LAGG);
1371 if (error)
1372 break;
1373 if (rp->rp_portname[0] == '\0' ||
1374 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1375 error = EINVAL;
1376 break;
1377 }
1378
1379 LAGG_XLOCK(sc);
1380 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1381 lp->lp_softc != sc) {
1382 error = ENOENT;
1383 LAGG_XUNLOCK(sc);
1384 if_rele(tpif);
1385 break;
1386 }
1387
1388 error = lagg_port_destroy(lp, 1);
1389 LAGG_XUNLOCK(sc);
1390 if_rele(tpif);
1391 VLAN_CAPABILITIES(ifp);
1392 break;
1393 case SIOCSIFFLAGS:
1394 /* Set flags on ports too */
1395 LAGG_XLOCK(sc);
1396 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1397 lagg_setflags(lp, 1);
1398 }
1399
1400 if (!(ifp->if_flags & IFF_UP) &&
1401 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1402 /*
1403 * If interface is marked down and it is running,
1404 * then stop and disable it.
1405 */
1406 lagg_stop(sc);
1407 LAGG_XUNLOCK(sc);
1408 } else if ((ifp->if_flags & IFF_UP) &&
1409 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1410 /*
1411 * If interface is marked up and it is stopped, then
1412 * start it.
1413 */
1414 LAGG_XUNLOCK(sc);
1415 (*ifp->if_init)(sc);
1416 } else
1417 LAGG_XUNLOCK(sc);
1418 break;
1419 case SIOCADDMULTI:
1420 case SIOCDELMULTI:
1421 LAGG_WLOCK(sc);
1422 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1423 lagg_clrmulti(lp);
1424 lagg_setmulti(lp);
1425 }
1426 LAGG_WUNLOCK(sc);
1427 error = 0;
1428 break;
1429 case SIOCSIFMEDIA:
1430 case SIOCGIFMEDIA:
1431 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1432 break;
1433
1434 case SIOCSIFCAP:
1435 LAGG_XLOCK(sc);
1436 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1437 if (lp->lp_ioctl != NULL)
1438 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1439 }
1440 lagg_capabilities(sc);
1441 LAGG_XUNLOCK(sc);
1442 VLAN_CAPABILITIES(ifp);
1443 error = 0;
1444 break;
1445
1446 case SIOCSIFMTU:
1447 /* Do not allow the MTU to be directly changed */
1448 error = EINVAL;
1449 break;
1450
1451 default:
1452 error = ether_ioctl(ifp, cmd, data);
1453 break;
1454 }
1455 return (error);
1456 }
1457
1458 static int
1459 lagg_setmulti(struct lagg_port *lp)
1460 {
1461 struct lagg_softc *sc = lp->lp_softc;
1462 struct ifnet *ifp = lp->lp_ifp;
1463 struct ifnet *scifp = sc->sc_ifp;
1464 struct lagg_mc *mc;
1465 struct ifmultiaddr *ifma;
1466 int error;
1467
1468 LAGG_WLOCK_ASSERT(sc);
1469 IF_ADDR_WLOCK(scifp);
1470 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1471 if (ifma->ifma_addr->sa_family != AF_LINK)
1472 continue;
1473 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1474 if (mc == NULL) {
1475 IF_ADDR_WUNLOCK(scifp);
1476 return (ENOMEM);
1477 }
1478 bcopy(ifma->ifma_addr, &mc->mc_addr,
1479 ifma->ifma_addr->sa_len);
1480 mc->mc_addr.sdl_index = ifp->if_index;
1481 mc->mc_ifma = NULL;
1482 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1483 }
1484 IF_ADDR_WUNLOCK(scifp);
1485 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1486 error = if_addmulti(ifp,
1487 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1488 if (error)
1489 return (error);
1490 }
1491 return (0);
1492 }
1493
1494 static int
1495 lagg_clrmulti(struct lagg_port *lp)
1496 {
1497 struct lagg_mc *mc;
1498
1499 LAGG_WLOCK_ASSERT(lp->lp_softc);
1500 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1501 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1502 if (mc->mc_ifma && lp->lp_detaching == 0)
1503 if_delmulti_ifma(mc->mc_ifma);
1504 free(mc, M_DEVBUF);
1505 }
1506 return (0);
1507 }
1508
1509 static int
1510 lagg_setcaps(struct lagg_port *lp, int cap)
1511 {
1512 struct ifreq ifr;
1513
1514 if (lp->lp_ifp->if_capenable == cap)
1515 return (0);
1516 if (lp->lp_ioctl == NULL)
1517 return (ENXIO);
1518 ifr.ifr_reqcap = cap;
1519 return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1520 }
1521
1522 /* Handle a ref counted flag that should be set on the lagg port as well */
1523 static int
1524 lagg_setflag(struct lagg_port *lp, int flag, int status,
1525 int (*func)(struct ifnet *, int))
1526 {
1527 struct lagg_softc *sc = lp->lp_softc;
1528 struct ifnet *scifp = sc->sc_ifp;
1529 struct ifnet *ifp = lp->lp_ifp;
1530 int error;
1531
1532 LAGG_XLOCK_ASSERT(sc);
1533
1534 status = status ? (scifp->if_flags & flag) : 0;
1535 /* Now "status" contains the flag value or 0 */
1536
1537 /*
1538 * See if recorded ports status is different from what
1539 * we want it to be. If it is, flip it. We record ports
1540 * status in lp_ifflags so that we won't clear ports flag
1541 * we haven't set. In fact, we don't clear or set ports
1542 * flags directly, but get or release references to them.
1543 * That's why we can be sure that recorded flags still are
1544 * in accord with actual ports flags.
1545 */
1546 if (status != (lp->lp_ifflags & flag)) {
1547 error = (*func)(ifp, status);
1548 if (error)
1549 return (error);
1550 lp->lp_ifflags &= ~flag;
1551 lp->lp_ifflags |= status;
1552 }
1553 return (0);
1554 }
1555
1556 /*
1557 * Handle IFF_* flags that require certain changes on the lagg port
1558 * if "status" is true, update ports flags respective to the lagg
1559 * if "status" is false, forcedly clear the flags set on port.
1560 */
1561 static int
1562 lagg_setflags(struct lagg_port *lp, int status)
1563 {
1564 int error, i;
1565
1566 for (i = 0; lagg_pflags[i].flag; i++) {
1567 error = lagg_setflag(lp, lagg_pflags[i].flag,
1568 status, lagg_pflags[i].func);
1569 if (error)
1570 return (error);
1571 }
1572 return (0);
1573 }
1574
1575 static int
1576 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1577 {
1578 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1579 int error, len, mcast;
1580 struct rm_priotracker tracker;
1581
1582 len = m->m_pkthdr.len;
1583 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1584
1585 LAGG_RLOCK(sc, &tracker);
1586 /* We need a Tx algorithm and at least one port */
1587 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1588 LAGG_RUNLOCK(sc, &tracker);
1589 m_freem(m);
1590 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1591 return (ENXIO);
1592 }
1593
1594 ETHER_BPF_MTAP(ifp, m);
1595
1596 error = lagg_proto_start(sc, m);
1597 LAGG_RUNLOCK(sc, &tracker);
1598
1599 if (error != 0)
1600 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1601
1602 return (error);
1603 }
1604
1605 /*
1606 * The ifp->if_qflush entry point for lagg(4) is no-op.
1607 */
1608 static void
1609 lagg_qflush(struct ifnet *ifp __unused)
1610 {
1611 }
1612
1613 static struct mbuf *
1614 lagg_input(struct ifnet *ifp, struct mbuf *m)
1615 {
1616 struct lagg_port *lp = ifp->if_lagg;
1617 struct lagg_softc *sc = lp->lp_softc;
1618 struct ifnet *scifp = sc->sc_ifp;
1619 struct rm_priotracker tracker;
1620
1621 LAGG_RLOCK(sc, &tracker);
1622 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1623 (lp->lp_flags & LAGG_PORT_DISABLED) ||
1624 sc->sc_proto == LAGG_PROTO_NONE) {
1625 LAGG_RUNLOCK(sc, &tracker);
1626 m_freem(m);
1627 return (NULL);
1628 }
1629
1630 ETHER_BPF_MTAP(scifp, m);
1631
1632 if (lp->lp_detaching != 0) {
1633 m_freem(m);
1634 m = NULL;
1635 } else
1636 m = lagg_proto_input(sc, lp, m);
1637
1638 if (m != NULL) {
1639 if (scifp->if_flags & IFF_MONITOR) {
1640 m_freem(m);
1641 m = NULL;
1642 }
1643 }
1644
1645 LAGG_RUNLOCK(sc, &tracker);
1646 return (m);
1647 }
1648
1649 static int
1650 lagg_media_change(struct ifnet *ifp)
1651 {
1652 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1653
1654 if (sc->sc_ifflags & IFF_DEBUG)
1655 printf("%s\n", __func__);
1656
1657 /* Ignore */
1658 return (0);
1659 }
1660
1661 static void
1662 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1663 {
1664 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1665 struct lagg_port *lp;
1666
1667 imr->ifm_status = IFM_AVALID;
1668 imr->ifm_active = IFM_ETHER | IFM_AUTO;
1669
1670 LAGG_SLOCK(sc);
1671 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1672 if (LAGG_PORTACTIVE(lp))
1673 imr->ifm_status |= IFM_ACTIVE;
1674 }
1675 LAGG_SUNLOCK(sc);
1676 }
1677
1678 static void
1679 lagg_linkstate(struct lagg_softc *sc)
1680 {
1681 struct lagg_port *lp;
1682 int new_link = LINK_STATE_DOWN;
1683 uint64_t speed;
1684
1685 LAGG_XLOCK_ASSERT(sc);
1686
1687 /* Our link is considered up if at least one of our ports is active */
1688 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1689 if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1690 new_link = LINK_STATE_UP;
1691 break;
1692 }
1693 }
1694 if_link_state_change(sc->sc_ifp, new_link);
1695
1696 /* Update if_baudrate to reflect the max possible speed */
1697 switch (sc->sc_proto) {
1698 case LAGG_PROTO_FAILOVER:
1699 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1700 sc->sc_primary->lp_ifp->if_baudrate : 0;
1701 break;
1702 case LAGG_PROTO_ROUNDROBIN:
1703 case LAGG_PROTO_LOADBALANCE:
1704 case LAGG_PROTO_BROADCAST:
1705 speed = 0;
1706 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1707 speed += lp->lp_ifp->if_baudrate;
1708 sc->sc_ifp->if_baudrate = speed;
1709 break;
1710 case LAGG_PROTO_LACP:
1711 /* LACP updates if_baudrate itself */
1712 break;
1713 }
1714 }
1715
1716 static void
1717 lagg_port_state(struct ifnet *ifp, int state)
1718 {
1719 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1720 struct lagg_softc *sc = NULL;
1721
1722 if (lp != NULL)
1723 sc = lp->lp_softc;
1724 if (sc == NULL)
1725 return;
1726
1727 LAGG_XLOCK(sc);
1728 lagg_linkstate(sc);
1729 lagg_proto_linkstate(sc, lp);
1730 LAGG_XUNLOCK(sc);
1731 }
1732
1733 struct lagg_port *
1734 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1735 {
1736 struct lagg_port *lp_next, *rval = NULL;
1737
1738 /*
1739 * Search a port which reports an active link state.
1740 */
1741
1742 if (lp == NULL)
1743 goto search;
1744 if (LAGG_PORTACTIVE(lp)) {
1745 rval = lp;
1746 goto found;
1747 }
1748 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1749 LAGG_PORTACTIVE(lp_next)) {
1750 rval = lp_next;
1751 goto found;
1752 }
1753
1754 search:
1755 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1756 if (LAGG_PORTACTIVE(lp_next)) {
1757 rval = lp_next;
1758 goto found;
1759 }
1760 }
1761
1762 found:
1763 return (rval);
1764 }
1765
1766 int
1767 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1768 {
1769
1770 return (ifp->if_transmit)(ifp, m);
1771 }
1772
1773 /*
1774 * Simple round robin aggregation
1775 */
1776 static void
1777 lagg_rr_attach(struct lagg_softc *sc)
1778 {
1779 sc->sc_seq = 0;
1780 sc->sc_bkt_count = sc->sc_bkt;
1781 }
1782
1783 static int
1784 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1785 {
1786 struct lagg_port *lp;
1787 uint32_t p;
1788
1789 if (sc->sc_bkt_count == 0 && sc->sc_bkt > 0)
1790 sc->sc_bkt_count = sc->sc_bkt;
1791
1792 if (sc->sc_bkt > 0) {
1793 atomic_subtract_int(&sc->sc_bkt_count, 1);
1794 if (atomic_cmpset_int(&sc->sc_bkt_count, 0, sc->sc_bkt))
1795 p = atomic_fetchadd_32(&sc->sc_seq, 1);
1796 else
1797 p = sc->sc_seq;
1798 } else
1799 p = atomic_fetchadd_32(&sc->sc_seq, 1);
1800
1801 p %= sc->sc_count;
1802 lp = SLIST_FIRST(&sc->sc_ports);
1803
1804 while (p--)
1805 lp = SLIST_NEXT(lp, lp_entries);
1806
1807 /*
1808 * Check the port's link state. This will return the next active
1809 * port if the link is down or the port is NULL.
1810 */
1811 if ((lp = lagg_link_active(sc, lp)) == NULL) {
1812 m_freem(m);
1813 return (ENETDOWN);
1814 }
1815
1816 /* Send mbuf */
1817 return (lagg_enqueue(lp->lp_ifp, m));
1818 }
1819
1820 static struct mbuf *
1821 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1822 {
1823 struct ifnet *ifp = sc->sc_ifp;
1824
1825 /* Just pass in the packet to our lagg device */
1826 m->m_pkthdr.rcvif = ifp;
1827
1828 return (m);
1829 }
1830
1831 /*
1832 * Broadcast mode
1833 */
1834 static int
1835 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
1836 {
1837 int active_ports = 0;
1838 int errors = 0;
1839 int ret;
1840 struct lagg_port *lp, *last = NULL;
1841 struct mbuf *m0;
1842
1843 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1844 if (!LAGG_PORTACTIVE(lp))
1845 continue;
1846
1847 active_ports++;
1848
1849 if (last != NULL) {
1850 m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
1851 if (m0 == NULL) {
1852 ret = ENOBUFS;
1853 errors++;
1854 break;
1855 }
1856
1857 ret = lagg_enqueue(last->lp_ifp, m0);
1858 if (ret != 0)
1859 errors++;
1860 }
1861 last = lp;
1862 }
1863 if (last == NULL) {
1864 m_freem(m);
1865 return (ENOENT);
1866 }
1867 if ((last = lagg_link_active(sc, last)) == NULL) {
1868 m_freem(m);
1869 return (ENETDOWN);
1870 }
1871
1872 ret = lagg_enqueue(last->lp_ifp, m);
1873 if (ret != 0)
1874 errors++;
1875
1876 if (errors == 0)
1877 return (ret);
1878
1879 return (0);
1880 }
1881
1882 static struct mbuf*
1883 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1884 {
1885 struct ifnet *ifp = sc->sc_ifp;
1886
1887 /* Just pass in the packet to our lagg device */
1888 m->m_pkthdr.rcvif = ifp;
1889 return (m);
1890 }
1891
1892 /*
1893 * Active failover
1894 */
1895 static int
1896 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1897 {
1898 struct lagg_port *lp;
1899
1900 /* Use the master port if active or the next available port */
1901 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1902 m_freem(m);
1903 return (ENETDOWN);
1904 }
1905
1906 /* Send mbuf */
1907 return (lagg_enqueue(lp->lp_ifp, m));
1908 }
1909
1910 static struct mbuf *
1911 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1912 {
1913 struct ifnet *ifp = sc->sc_ifp;
1914 struct lagg_port *tmp_tp;
1915
1916 if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
1917 m->m_pkthdr.rcvif = ifp;
1918 return (m);
1919 }
1920
1921 if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1922 tmp_tp = lagg_link_active(sc, sc->sc_primary);
1923 /*
1924 * If tmp_tp is null, we've received a packet when all
1925 * our links are down. Weird, but process it anyways.
1926 */
1927 if ((tmp_tp == NULL || tmp_tp == lp)) {
1928 m->m_pkthdr.rcvif = ifp;
1929 return (m);
1930 }
1931 }
1932
1933 m_freem(m);
1934 return (NULL);
1935 }
1936
1937 /*
1938 * Loadbalancing
1939 */
1940 static void
1941 lagg_lb_attach(struct lagg_softc *sc)
1942 {
1943 struct lagg_port *lp;
1944 struct lagg_lb *lb;
1945
1946 lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
1947 lb->lb_key = m_ether_tcpip_hash_init();
1948 sc->sc_psc = lb;
1949
1950 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1951 lagg_lb_port_create(lp);
1952 }
1953
1954 static void
1955 lagg_lb_detach(struct lagg_softc *sc)
1956 {
1957 struct lagg_lb *lb;
1958
1959 lb = (struct lagg_lb *)sc->sc_psc;
1960 LAGG_WUNLOCK(sc);
1961 if (lb != NULL)
1962 free(lb, M_DEVBUF);
1963 }
1964
1965 static int
1966 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1967 {
1968 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1969 struct lagg_port *lp_next;
1970 int i = 0;
1971
1972 bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1973 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1974 if (lp_next == lp)
1975 continue;
1976 if (i >= LAGG_MAX_PORTS)
1977 return (EINVAL);
1978 if (sc->sc_ifflags & IFF_DEBUG)
1979 printf("%s: port %s at index %d\n",
1980 sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
1981 lb->lb_ports[i++] = lp_next;
1982 }
1983
1984 return (0);
1985 }
1986
1987 static int
1988 lagg_lb_port_create(struct lagg_port *lp)
1989 {
1990 struct lagg_softc *sc = lp->lp_softc;
1991 return (lagg_lb_porttable(sc, NULL));
1992 }
1993
1994 static void
1995 lagg_lb_port_destroy(struct lagg_port *lp)
1996 {
1997 struct lagg_softc *sc = lp->lp_softc;
1998 lagg_lb_porttable(sc, lp);
1999 }
2000
2001 static int
2002 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2003 {
2004 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2005 struct lagg_port *lp = NULL;
2006 uint32_t p = 0;
2007
2008 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2009 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2010 p = m->m_pkthdr.flowid >> sc->flowid_shift;
2011 else
2012 p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2013 p %= sc->sc_count;
2014 lp = lb->lb_ports[p];
2015
2016 /*
2017 * Check the port's link state. This will return the next active
2018 * port if the link is down or the port is NULL.
2019 */
2020 if ((lp = lagg_link_active(sc, lp)) == NULL) {
2021 m_freem(m);
2022 return (ENETDOWN);
2023 }
2024
2025 /* Send mbuf */
2026 return (lagg_enqueue(lp->lp_ifp, m));
2027 }
2028
2029 static struct mbuf *
2030 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2031 {
2032 struct ifnet *ifp = sc->sc_ifp;
2033
2034 /* Just pass in the packet to our lagg device */
2035 m->m_pkthdr.rcvif = ifp;
2036
2037 return (m);
2038 }
2039
2040 /*
2041 * 802.3ad LACP
2042 */
2043 static void
2044 lagg_lacp_attach(struct lagg_softc *sc)
2045 {
2046 struct lagg_port *lp;
2047
2048 lacp_attach(sc);
2049 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2050 lacp_port_create(lp);
2051 }
2052
2053 static void
2054 lagg_lacp_detach(struct lagg_softc *sc)
2055 {
2056 struct lagg_port *lp;
2057 void *psc;
2058
2059 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2060 lacp_port_destroy(lp);
2061
2062 psc = sc->sc_psc;
2063 sc->sc_psc = NULL;
2064 LAGG_WUNLOCK(sc);
2065
2066 lacp_detach(psc);
2067 }
2068
2069 static void
2070 lagg_lacp_lladdr(struct lagg_softc *sc)
2071 {
2072 struct lagg_port *lp;
2073
2074 LAGG_SXLOCK_ASSERT(sc);
2075
2076 /* purge all the lacp ports */
2077 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2078 lacp_port_destroy(lp);
2079
2080 /* add them back in */
2081 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2082 lacp_port_create(lp);
2083 }
2084
2085 static int
2086 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2087 {
2088 struct lagg_port *lp;
2089
2090 lp = lacp_select_tx_port(sc, m);
2091 if (lp == NULL) {
2092 m_freem(m);
2093 return (ENETDOWN);
2094 }
2095
2096 /* Send mbuf */
2097 return (lagg_enqueue(lp->lp_ifp, m));
2098 }
2099
2100 static struct mbuf *
2101 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2102 {
2103 struct ifnet *ifp = sc->sc_ifp;
2104 struct ether_header *eh;
2105 u_short etype;
2106
2107 eh = mtod(m, struct ether_header *);
2108 etype = ntohs(eh->ether_type);
2109
2110 /* Tap off LACP control messages */
2111 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2112 m = lacp_input(lp, m);
2113 if (m == NULL)
2114 return (NULL);
2115 }
2116
2117 /*
2118 * If the port is not collecting or not in the active aggregator then
2119 * free and return.
2120 */
2121 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2122 m_freem(m);
2123 return (NULL);
2124 }
2125
2126 m->m_pkthdr.rcvif = ifp;
2127 return (m);
2128 }
2129
Cache object: 8f0047df4bf067ac7fef182abe65cd52
|