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 $
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>
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.
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.
22 #include <sys/cdefs.h>
25 #include "opt_inet6.h"
27 #include <sys/param.h>
28 #include <sys/kernel.h>
29 #include <sys/malloc.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>
37 #include <sys/systm.h>
41 #include <sys/taskqueue.h>
42 #include <sys/eventhandler.h>
44 #include <net/ethernet.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>
56 #include <net/netmsg2.h>
57 #include <net/netisr2.h>
59 #if defined(INET) || defined(INET6)
60 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/if_ether.h>
65 #include <netinet/ip.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet6/in6_ifattach.h>
74 #include <net/vlan/if_vlan_var.h>
75 #include <net/lagg/if_lagg.h>
76 #include <net/lagg/ieee8023ad_lacp.h>
78 /* Special flags we should propagate to the lagg ports. */
81 int (*func)(struct ifnet *, int);
83 {IFF_PROMISC, ifpromisc},
84 {IFF_ALLMULTI, if_allmulti},
88 SLIST_HEAD(, lagg_softc) lagg_list; /* list of laggs */
89 static struct lock lagg_list_lock;
90 eventhandler_tag lagg_detach_cookie = NULL;
92 static int lagg_clone_create(struct if_clone *, int, 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",
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 *);
109 static void lagg_port_state(struct ifnet *, int);
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 *);
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);
130 static int lagg_output(struct ifnet *ifp, struct mbuf *m);
133 static int lagg_transmit(struct ifnet *, struct mbuf *);
134 static void lagg_qflush(struct ifnet *);
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 *,
140 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
141 static int lagg_sysctl_active(SYSCTL_HANDLER_ARGS);
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,
148 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
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 *,
156 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
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 *,
166 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
168 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
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 *,
175 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
177 static void lagg_lacp_lladdr(struct lagg_softc *);
179 static void lagg_callout(void *);
181 /* lagg protocol table */
182 static const struct {
184 int (*ti_attach)(struct lagg_softc *);
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 }
194 SYSCTL_DECL(_net_link);
195 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
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,
206 "Default setting for using flow id for load sharing");
209 lagg_modevent(module_t mod, int type, void *data)
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);
223 EVENTHANDLER_DEREGISTER(ifnet_departure_event,
226 if_clone_detach(&lagg_cloner);
227 lockuninit(&lagg_list_lock);
235 static moduledata_t lagg_mod = {
241 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
242 MODULE_VERSION(if_lagg, 1);
245 * This routine is run via an vlan
249 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
252 struct lagg_softc *sc = ifp->if_softc;
253 struct lagg_port *lp;
255 if (ifp->if_softc != arg) /* Not our event */
259 if (!SLIST_EMPTY(&sc->sc_ports)) {
260 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
261 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp);
269 * This routine is run via an vlan
273 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
275 struct lagg_softc *sc = ifp->if_softc;
277 struct lagg_port *lp;
280 if (ifp->if_softc != arg) /* Not our event */
284 if (!SLIST_EMPTY(&sc->sc_ports)) {
286 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
287 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp);
294 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params __unused)
296 struct lagg_softc *sc;
299 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */
300 struct sysctl_oid *oid;
301 char num[14]; /* sufficient for 32 bits */
303 sc = kmalloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
304 ifp = sc->sc_ifp = &sc->sc_if; // XXX if_alloc(IFT_ETHER);
312 sc->sc_ipackets = counter_u64_alloc(M_WAITOK);
313 sc->sc_opackets = counter_u64_alloc(M_WAITOK);
314 sc->sc_ibytes = counter_u64_alloc(M_WAITOK);
315 sc->sc_obytes = counter_u64_alloc(M_WAITOK);
317 sysctl_ctx_init(&sc->ctx);
318 ksnprintf(num, sizeof(num), "%u", unit);
319 sc->use_flowid = def_use_flowid;
320 sc->sc_oid = oid = SYSCTL_ADD_NODE(&sc->ctx,
321 &SYSCTL_NODE_CHILDREN(_net_link, lagg),
322 OID_AUTO, num, CTLFLAG_RD, NULL, "");
323 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
324 "use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid, sc->use_flowid,
325 "Use flow id for load sharing");
326 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
327 "count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count,
328 "Total number of ports");
329 SYSCTL_ADD_PROC(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
330 "active", CTLTYPE_INT|CTLFLAG_RD, sc, 0, lagg_sysctl_active,
331 "I", "Total number of active ports");
332 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
333 "flapping", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_flapping,
334 sc->sc_flapping, "Total number of port change events");
335 /* Hash all layers by default */
336 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4;
338 sc->sc_proto = LAGG_PROTO_NONE;
339 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
340 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
341 sc->sc_proto = lagg_protos[i].ti_proto;
342 if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
351 LAGG_CALLOUT_LOCK_INIT(sc);
352 SLIST_INIT(&sc->sc_ports);
353 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
356 * This uses the callout lock rather than the rmlock; one can't
357 * hold said rmlock during SWI.
359 callout_init_lk(&sc->sc_callout, &sc->sc_call_lock);
361 /* Initialise pseudo media types */
362 ifmedia_init(&sc->sc_media, 0, lagg_media_change,
364 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
365 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
367 if_initname(ifp, laggname, unit);
370 ifp->if_transmit = lagg_transmit;
371 ifp->if_qflush = lagg_qflush;
373 ifp->if_mtu = ETHERMTU;
374 ifp->if_init = lagg_init;
375 ifp->if_ioctl = lagg_ioctl;
376 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
377 ifp->if_start = lagg_start;
378 ifp->if_type = IFT_ETHER;
379 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
380 ifq_set_ready(&ifp->if_snd);
381 ifp->if_hdrlen = ETHER_HDR_LEN;
384 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
387 * Attach as an ordinary ethernet device, children will be attached
388 * as special device IFT_IEEE8023ADLAG.
391 ether_ifattach(ifp, eaddr, NULL);
393 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
394 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
395 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
396 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
398 /* Insert into the global list of laggs */
399 lockmgr(&lagg_list_lock, LK_EXCLUSIVE);
400 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
401 lockmgr(&lagg_list_lock, LK_RELEASE);
403 callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
409 lagg_clone_destroy(struct ifnet *ifp)
411 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
412 struct lagg_port *lp;
417 ifp->if_flags &= ~IFF_UP;
419 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
420 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
422 /* Shutdown and remove lagg ports */
423 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
424 lagg_port_destroy(lp, 1);
425 /* Unhook the aggregation protocol */
426 if (sc->sc_detach != NULL)
427 (*sc->sc_detach)(sc);
431 sysctl_ctx_free(&sc->ctx);
432 ifmedia_removeall(&sc->sc_media);
434 /* This ifp is part of lagg_softc, don't free it! */
437 /* This grabs sc_callout_mtx, serialising it correctly */
438 callout_drain(&sc->sc_callout);
441 /* At this point it's drained; we can free this */
442 counter_u64_free(sc->sc_ipackets);
443 counter_u64_free(sc->sc_opackets);
444 counter_u64_free(sc->sc_ibytes);
445 counter_u64_free(sc->sc_obytes);
448 lockmgr(&lagg_list_lock, LK_EXCLUSIVE);
449 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
450 lockmgr(&lagg_list_lock, LK_RELEASE);
452 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
453 LAGG_LOCK_DESTROY(sc);
454 LAGG_CALLOUT_LOCK_DESTROY(sc);
461 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
463 struct ifnet *ifp = sc->sc_ifp;
465 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
468 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
469 /* Let the protocol know the MAC has changed */
470 if (sc->sc_lladdr != NULL)
471 (*sc->sc_lladdr)(sc);
472 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
476 lagg_capabilities(struct lagg_softc *sc)
478 struct lagg_port *lp;
479 int cap = ~0, ena = ~0;
482 LAGG_WLOCK_ASSERT(sc);
484 /* Get capabilities from the lagg ports */
485 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
486 cap &= lp->lp_ifp->if_capabilities;
487 ena &= lp->lp_ifp->if_capenable;
488 hwa &= lp->lp_ifp->if_hwassist;
490 cap = (cap == ~0 ? 0 : cap);
491 ena = (ena == ~0 ? 0 : ena);
492 hwa = (hwa == ~0 ? 0 : hwa);
494 if (sc->sc_ifp->if_capabilities != cap ||
495 sc->sc_ifp->if_capenable != ena ||
496 sc->sc_ifp->if_hwassist != hwa) {
497 sc->sc_ifp->if_capabilities = cap;
498 sc->sc_ifp->if_capenable = ena;
499 sc->sc_ifp->if_hwassist = hwa;
500 getmicrotime(&sc->sc_ifp->if_lastchange);
502 if (sc->sc_ifflags & IFF_DEBUG)
503 if_printf(sc->sc_ifp,
504 "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
509 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
511 struct lagg_softc *sc = lp->lp_softc;
512 struct ifnet *ifp = lp->lp_ifp;
513 struct lagg_llq *llq;
516 LAGG_WLOCK_ASSERT(sc);
518 if (lp->lp_detaching ||
519 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
522 /* Check to make sure its not already queued to be changed */
523 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
524 if (llq->llq_ifp == ifp) {
531 llq = kmalloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
532 if (llq == NULL) /* XXX what to do */
536 /* Update the lladdr even if pending, it may have changed */
538 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
541 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
543 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
547 * Set the interface MAC address from a taskqueue to avoid a LOR.
550 lagg_port_setlladdr(void *arg, int pending)
552 struct lagg_softc *sc = (struct lagg_softc *)arg;
553 struct lagg_llq *llq, *head;
557 /* Grab a local reference of the queue and remove it from the softc */
559 head = SLIST_FIRST(&sc->sc_llq_head);
560 SLIST_FIRST(&sc->sc_llq_head) = NULL;
564 * Traverse the queue and set the lladdr on each ifp. It is safe to do
565 * unlocked as we have the only reference to it.
567 for (llq = head; llq != NULL; llq = head) {
570 /* Set the link layer address */
571 /* CURVNET_SET(ifp->if_vnet); */
572 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
573 /* CURVNET_RESTORE(); */
575 kprintf("%s: setlladdr failed on %s\n", __func__,
578 head = SLIST_NEXT(llq, llq_entries);
579 kfree(llq, M_DEVBUF);
584 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
586 struct lagg_softc *sc_ptr;
587 struct lagg_port *lp;
590 LAGG_WLOCK_ASSERT(sc);
592 /* Limit the maximal number of lagg ports */
593 if (sc->sc_count >= LAGG_MAX_PORTS)
596 /* Check if port has already been associated to a lagg */
597 if (ifp->if_lagg != NULL) {
598 /* Port is already in the current lagg? */
599 lp = (struct lagg_port *)ifp->if_lagg;
600 if (lp->lp_softc == sc)
605 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */
606 if (ifp->if_type != IFT_ETHER)
607 return (EPROTONOSUPPORT);
611 * The member interface should not have inet6 address because
612 * two interfaces with a valid link-local scope zone must not be
613 * merged in any form. This restriction is needed to
614 * prevent violation of link-local scope zone. Attempts to
615 * add a member interface which has inet6 addresses triggers
616 * removal of all inet6 addresses on the member interface.
618 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
619 if (in6ifa_llaonifp(lp->lp_ifp)) {
620 in6_ifdetach(lp->lp_ifp);
621 if_printf(sc->sc_ifp,
622 "IPv6 addresses on %s have been removed "
623 "before adding it as a member to prevent "
624 "IPv6 address scope violation.\n",
625 lp->lp_ifp->if_xname);
628 if (in6ifa_llaonifp(ifp)) {
630 if_printf(sc->sc_ifp,
631 "IPv6 addresses on %s have been removed "
632 "before adding it as a member to prevent "
633 "IPv6 address scope violation.\n",
637 /* Allow the first Ethernet member to define the MTU */
638 if (SLIST_EMPTY(&sc->sc_ports))
639 sc->sc_ifp->if_mtu = ifp->if_mtu;
640 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
641 if_printf(sc->sc_ifp, "invalid MTU for %s\n",
646 if ((lp = kmalloc(sizeof(struct lagg_port),
647 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
650 /* Check if port is a stacked lagg */
651 lockmgr(&lagg_list_lock, LK_EXCLUSIVE);
652 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
653 if (ifp == sc_ptr->sc_ifp) {
654 lockmgr(&lagg_list_lock, LK_RELEASE);
657 /* XXX disable stacking for the moment, its untested */
658 #ifdef LAGG_PORT_STACKING
659 lp->lp_flags |= LAGG_PORT_STACK;
660 if (lagg_port_checkstacking(sc_ptr) >=
662 lockmgr(&lagg_list_lock, LK_RELEASE);
669 lockmgr(&lagg_list_lock, LK_RELEASE);
671 /* Change the interface type */
672 lp->lp_iftype = ifp->if_type;
673 ifp->if_type = IFT_IEEE8023ADLAG;
675 lp->lp_ioctl = ifp->if_ioctl;
676 ifp->if_ioctl = lagg_port_ioctl;
677 lp->lp_output = ifp->if_output;
678 ifp->if_output = lagg_port_output;
683 /* Save port link layer address */
684 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
686 if (SLIST_EMPTY(&sc->sc_ports)) {
688 lagg_lladdr(sc, IF_LLADDR(ifp));
690 /* Update link layer address for this port */
691 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
694 /* Insert into the list of ports */
695 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
698 /* Update lagg capabilities */
699 lagg_capabilities(sc);
702 /* Add multicast addresses and interface flags to this port */
703 lagg_ether_cmdmulti(lp, 1);
704 lagg_setflags(lp, 1);
706 if (sc->sc_port_create != NULL)
707 error = (*sc->sc_port_create)(lp);
709 /* remove the port again, without calling sc_port_destroy */
710 lagg_port_destroy(lp, 0);
717 #ifdef LAGG_PORT_STACKING
719 lagg_port_checkstacking(struct lagg_softc *sc)
721 struct lagg_softc *sc_ptr;
722 struct lagg_port *lp;
725 LAGG_WLOCK_ASSERT(sc);
727 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
728 if (lp->lp_flags & LAGG_PORT_STACK) {
729 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
730 m = MAX(m, lagg_port_checkstacking(sc_ptr));
739 lagg_port_destroy(struct lagg_port *lp, int runpd)
741 struct lagg_softc *sc = lp->lp_softc;
742 struct lagg_port *lp_ptr;
743 struct lagg_llq *llq;
744 struct ifnet *ifp = lp->lp_ifp;
746 LAGG_WLOCK_ASSERT(sc);
748 if (runpd && sc->sc_port_destroy != NULL)
749 (*sc->sc_port_destroy)(lp);
752 * Remove multicast addresses and interface flags from this port and
753 * reset the MAC address, skip if the interface is being detached.
755 if (!lp->lp_detaching) {
756 lagg_ether_cmdmulti(lp, 0);
757 lagg_setflags(lp, 0);
758 lagg_port_lladdr(lp, lp->lp_lladdr);
761 /* Restore interface */
762 ifp->if_type = lp->lp_iftype;
763 ifp->if_ioctl = lp->lp_ioctl;
764 ifp->if_output = lp->lp_output;
767 /* Finally, remove the port from the lagg */
768 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
771 /* Update the primary interface */
772 if (lp == sc->sc_primary) {
773 uint8_t lladdr[ETHER_ADDR_LEN];
775 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
776 bzero(&lladdr, ETHER_ADDR_LEN);
778 bcopy(lp_ptr->lp_lladdr,
779 lladdr, ETHER_ADDR_LEN);
781 lagg_lladdr(sc, lladdr);
782 sc->sc_primary = lp_ptr;
784 /* Update link layer address for each port */
785 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
786 lagg_port_lladdr(lp_ptr, lladdr);
789 /* Remove any pending lladdr changes from the queue */
790 if (lp->lp_detaching) {
791 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
792 if (llq->llq_ifp == ifp) {
793 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
795 kfree(llq, M_DEVBUF);
796 break; /* Only appears once */
802 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
806 /* Update lagg capabilities */
807 lagg_capabilities(sc);
814 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
816 struct lagg_reqport *rp = (struct lagg_reqport *)data;
817 struct lagg_softc *sc;
818 struct lagg_port *lp = NULL;
821 ASSERT_IFNET_SERIALIZED_ALL(ifp);
823 /* Should be checked by the caller */
824 if (ifp->if_type != IFT_IEEE8023ADLAG ||
825 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
830 if (rp->rp_portname[0] == '\0') {
836 * Release ifp serializers before ifnet_lock
837 * to prevent lock order reversal.
839 ifnet_deserialize_all(ifp);
841 if (ifunit(rp->rp_portname) != ifp) {
843 ifnet_serialize_all(ifp);
848 ifnet_serialize_all(ifp);
851 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
857 lagg_port2req(lp, rp);
862 if (lp->lp_ioctl == NULL) {
867 error = (*lp->lp_ioctl)(ifp, cmd, data, cr);
871 /* Update lagg interface capabilities */
873 lagg_capabilities(sc);
877 if (lp->lp_ioctl == NULL) {
882 error = (*lp->lp_ioctl)(ifp, cmd, data, cr);
885 /* Do not allow the MTU to be changed once joined */
896 if (lp->lp_ioctl != NULL) {
898 result = ((*lp->lp_ioctl)(ifp, cmd, data, cr));
904 * For direct output to child ports.
907 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
908 struct sockaddr *dst, struct rtentry *ro)
910 struct lagg_port *lp = ifp->if_lagg;
912 switch (dst->sa_family) {
913 case pseudo_AF_HDRCMPLT:
915 return ((*lp->lp_output)(ifp, m, dst, ro));
918 /* drop any other frames */
924 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
926 struct lagg_port *lp;
927 struct lagg_softc *sc;
929 if ((lp = ifp->if_lagg) == NULL)
932 /* If the ifnet is just being renamed, don't do anything. */
933 if (ifp->if_flags & IFF_RENAMING)
939 lp->lp_detaching = 1;
940 lagg_port_destroy(lp, 1);
945 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
947 struct lagg_softc *sc = lp->lp_softc;
949 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
950 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
951 rp->rp_prio = lp->lp_prio;
952 rp->rp_flags = lp->lp_flags;
953 if (sc->sc_portreq != NULL)
954 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
956 /* Add protocol specific flags */
957 switch (sc->sc_proto) {
958 case LAGG_PROTO_FAILOVER:
959 if (lp == sc->sc_primary)
960 rp->rp_flags |= LAGG_PORT_MASTER;
961 if (lp == lagg_link_active(sc, sc->sc_primary))
962 rp->rp_flags |= LAGG_PORT_ACTIVE;
965 case LAGG_PROTO_ROUNDROBIN:
966 case LAGG_PROTO_LOADBALANCE:
967 case LAGG_PROTO_ETHERCHANNEL:
968 if (LAGG_PORTACTIVE(lp))
969 rp->rp_flags |= LAGG_PORT_ACTIVE;
972 case LAGG_PROTO_LACP:
973 /* LACP has a different definition of active */
974 if (lacp_isactive(lp))
975 rp->rp_flags |= LAGG_PORT_ACTIVE;
976 if (lacp_iscollecting(lp))
977 rp->rp_flags |= LAGG_PORT_COLLECTING;
978 if (lacp_isdistributing(lp))
979 rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
988 struct lagg_softc *sc = (struct lagg_softc *)xsc;
989 struct lagg_port *lp;
990 struct ifnet *ifp = sc->sc_ifp;
992 if (ifp->if_flags & IFF_RUNNING)
997 ifp->if_flags |= IFF_RUNNING;
998 /* Update the port lladdrs */
999 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1000 lagg_port_lladdr(lp, IF_LLADDR(ifp));
1002 if (sc->sc_init != NULL)
1009 lagg_stop(struct lagg_softc *sc)
1011 struct ifnet *ifp = sc->sc_ifp;
1013 LAGG_WLOCK_ASSERT(sc);
1015 if ((ifp->if_flags & IFF_RUNNING) == 0)
1018 ifp->if_flags &= ~IFF_RUNNING;
1020 if (sc->sc_stop != NULL)
1025 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
1027 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1028 struct lagg_reqall *ra = (struct lagg_reqall *)data;
1029 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1030 struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1031 struct ifreq *ifr = (struct ifreq *)data;
1032 struct lagg_port *lp;
1034 struct thread *td = curthread;
1036 int count, buflen, len, error = 0;
1038 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1040 bzero(&rpbuf, sizeof(rpbuf));
1046 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1048 buflen = count * sizeof(struct lagg_reqport);
1051 outbuf = kmalloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1054 ra->ra_proto = sc->sc_proto;
1055 if (sc->sc_req != NULL)
1056 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
1060 len = min(ra->ra_size, buflen);
1061 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1062 if (len < sizeof(rpbuf))
1065 lagg_port2req(lp, &rpbuf);
1066 memcpy(buf, &rpbuf, sizeof(rpbuf));
1068 buf += sizeof(rpbuf);
1069 len -= sizeof(rpbuf);
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);
1078 error = priv_check(td, PRIV_NET_LAGG);
1081 if (ra->ra_proto >= LAGG_PROTO_MAX) {
1082 error = EPROTONOSUPPORT;
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;
1098 sc->sc_lladdr = NULL;
1100 sc->sc_portreq = NULL;
1101 } else if (sc->sc_input != NULL) {
1102 /* Still detaching */
1109 for (int i = 0; i < (sizeof(lagg_protos) /
1110 sizeof(lagg_protos[0])); i++) {
1111 if (lagg_protos[i].ti_proto == ra->ra_proto) {
1112 if (sc->sc_ifflags & IFF_DEBUG)
1113 kprintf("%s: using proto %u\n",
1115 lagg_protos[i].ti_proto);
1116 sc->sc_proto = lagg_protos[i].ti_proto;
1117 if (sc->sc_proto != LAGG_PROTO_NONE)
1118 error = lagg_protos[i].ti_attach(sc);
1124 error = EPROTONOSUPPORT;
1126 case SIOCGLAGGFLAGS:
1127 rf->rf_flags = sc->sc_flags;
1130 error = priv_check(td, PRIV_NET_LAGG);
1133 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1138 sc->sc_flags &= ~LAGG_F_HASHMASK;
1139 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK;
1144 * Release ifp serializers before ifnet_lock
1145 * to prevent lock order reversal.
1147 ifnet_deserialize_all(ifp);
1149 if (rp->rp_portname[0] == '\0' ||
1150 (tpif = ifunit(rp->rp_portname)) == NULL) {
1152 ifnet_serialize_all(ifp);
1158 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1159 lp->lp_softc != sc) {
1163 ifnet_serialize_all(ifp);
1167 lagg_port2req(lp, rp);
1170 ifnet_serialize_all(ifp);
1173 error = priv_check(td, PRIV_NET_LAGG);
1177 * Release ifp serializers before ifnet_lock
1178 * to prevent lock order reversal.
1180 ifnet_deserialize_all(ifp);
1182 if (rp->rp_portname[0] == '\0' ||
1183 (tpif = ifunit(rp->rp_portname)) == NULL) {
1185 ifnet_serialize_all(ifp);
1190 error = lagg_port_create(sc, tpif);
1193 ifnet_serialize_all(ifp);
1195 case SIOCSLAGGDELPORT:
1196 error = priv_check(td, PRIV_NET_LAGG);
1200 * Release ifp serializers before ifnet_lock
1201 * to prevent lock order reversal.
1203 ifnet_deserialize_all(ifp);
1205 if (rp->rp_portname[0] == '\0' ||
1206 (tpif = ifunit(rp->rp_portname)) == NULL) {
1208 ifnet_serialize_all(ifp);
1214 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1215 lp->lp_softc != sc) {
1219 ifnet_serialize_all(ifp);
1223 error = lagg_port_destroy(lp, 1);
1226 ifnet_serialize_all(ifp);
1229 /* Set flags on ports too */
1231 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1232 lagg_setflags(lp, 1);
1236 if (!(ifp->if_flags & IFF_UP) &&
1237 (ifp->if_flags & IFF_RUNNING)) {
1239 * If interface is marked down and it is running,
1240 * then stop and disable it.
1245 } else if ((ifp->if_flags & IFF_UP) &&
1246 !(ifp->if_flags & IFF_RUNNING)) {
1248 * If interface is marked up and it is stopped, then
1251 (*ifp->if_init)(sc);
1257 error = lagg_ether_setmulti(sc);
1262 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1267 /* Do not allow the MTU or caps to be directly changed */
1272 error = ether_ioctl(ifp, cmd, data);
1279 lagg_ether_setmulti(struct lagg_softc *sc)
1281 struct lagg_port *lp;
1283 LAGG_WLOCK_ASSERT(sc);
1285 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1286 /* First, remove any existing filter entries. */
1287 lagg_ether_cmdmulti(lp, 0);
1288 /* copy all addresses from the lagg interface to the port */
1289 lagg_ether_cmdmulti(lp, 1);
1295 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1297 struct lagg_softc *sc = lp->lp_softc;
1298 struct ifnet *ifp = lp->lp_ifp;
1299 struct ifnet *scifp = sc->sc_ifp;
1301 struct ifmultiaddr *ifma, *rifma = NULL;
1302 struct sockaddr_dl sdl;
1305 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp);
1306 LAGG_WLOCK_ASSERT(sc);
1308 bzero((char *)&sdl, sizeof(sdl));
1309 sdl.sdl_len = sizeof(sdl);
1310 sdl.sdl_family = AF_LINK;
1311 sdl.sdl_type = IFT_ETHER;
1312 sdl.sdl_alen = ETHER_ADDR_LEN;
1313 sdl.sdl_index = ifp->if_index;
1316 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1317 if (ifma->ifma_addr->sa_family != AF_LINK)
1319 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1320 LLADDR(&sdl), ETHER_ADDR_LEN);
1322 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1325 mc = kmalloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1328 mc->mc_ifma = rifma;
1329 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1332 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1333 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1334 if_delmulti(ifp, (struct sockaddr *)mc->mc_ifma);
1335 kfree(mc, M_DEVBUF);
1341 /* Handle a ref counted flag that should be set on the lagg port as well */
1343 lagg_setflag(struct lagg_port *lp, int flag, int status,
1344 int (*func)(struct ifnet *, int))
1346 struct lagg_softc *sc = lp->lp_softc;
1347 struct ifnet *scifp = sc->sc_ifp;
1348 struct ifnet *ifp = lp->lp_ifp;
1351 LAGG_WLOCK_ASSERT(sc);
1352 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp);
1354 status = status ? (scifp->if_flags & flag) : 0;
1355 /* Now "status" contains the flag value or 0 */
1358 * See if recorded ports status is different from what
1359 * we want it to be. If it is, flip it. We record ports
1360 * status in lp_ifflags so that we won't clear ports flag
1361 * we haven't set. In fact, we don't clear or set ports
1362 * flags directly, but get or release references to them.
1363 * That's why we can be sure that recorded flags still are
1364 * in accord with actual ports flags.
1366 if (status != (lp->lp_ifflags & flag)) {
1367 error = (*func)(ifp, status);
1370 lp->lp_ifflags &= ~flag;
1371 lp->lp_ifflags |= status;
1377 * Handle IFF_* flags that require certain changes on the lagg port
1378 * if "status" is true, update ports flags respective to the lagg
1379 * if "status" is false, forcedly clear the flags set on port.
1382 lagg_setflags(struct lagg_port *lp, int status)
1386 ASSERT_IFNET_NOT_SERIALIZED_ALL(lp->lp_ifp);
1388 for (i = 0; lagg_pflags[i].flag; i++) {
1389 error = lagg_setflag(lp, lagg_pflags[i].flag,
1390 status, lagg_pflags[i].func);
1398 #ifdef XXX /* not needed? */
1400 lagg_output(struct ifnet *ifp, struct mbuf *m)
1402 struct lagg_softc *sc = ifp->if_softc;
1403 int error, len, mcast;
1405 len = m->m_pkthdr.len;
1406 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1409 /* We need a Tx algorithm and at least one port */
1410 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1419 error = (*sc->sc_start)(sc, m);
1423 IFNET_STAT_INC(ifp, opackets, 1);
1424 IFNET_STAT_INC(ifp, obytes, len);
1425 ifp->if_omcasts += mcast;
1435 * The ifp->if_qflush entry point for lagg(4) is no-op.
1438 lagg_qflush(struct ifnet *ifp __unused)
1443 lagg_input(struct ifnet *ifp, struct mbuf *m)
1445 struct lagg_port *lp = ifp->if_lagg;
1446 struct lagg_softc *sc = lp->lp_softc;
1447 struct ifnet *scifp = sc->sc_ifp;
1450 if ((scifp->if_flags & IFF_RUNNING) == 0 ||
1451 (lp->lp_flags & LAGG_PORT_DISABLED) ||
1452 sc->sc_proto == LAGG_PROTO_NONE) {
1459 m = (*sc->sc_input)(sc, lp, m);
1464 IFNET_STAT_INC(ifp, ipackets, 1);
1465 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len);
1467 if (scifp->if_flags & IFF_MONITOR) {
1471 ether_reinput_oncpu(scifp, m, REINPUT_RUNBPF);
1476 lagg_media_change(struct ifnet *ifp)
1478 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1480 if (sc->sc_ifflags & IFF_DEBUG)
1481 kprintf("%s\n", __func__);
1487 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1489 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1490 struct lagg_port *lp;
1492 imr->ifm_status = IFM_AVALID;
1493 imr->ifm_active = IFM_ETHER | IFM_AUTO;
1496 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1497 if (LAGG_PORTACTIVE(lp))
1498 imr->ifm_status |= IFM_ACTIVE;
1504 lagg_linkstate(struct lagg_softc *sc)
1506 struct lagg_port *lp;
1507 int new_link = LINK_STATE_DOWN;
1510 /* Our link is considered up if at least one of our ports is active */
1511 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1512 if (lp->lp_link_state == LINK_STATE_UP) {
1513 new_link = LINK_STATE_UP;
1517 if_link_state_change(sc->sc_ifp);
1519 /* Update if_baudrate to reflect the max possible speed */
1520 switch (sc->sc_proto) {
1521 case LAGG_PROTO_FAILOVER:
1522 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1523 sc->sc_primary->lp_ifp->if_baudrate : 0;
1525 case LAGG_PROTO_ROUNDROBIN:
1526 case LAGG_PROTO_LOADBALANCE:
1527 case LAGG_PROTO_ETHERCHANNEL:
1529 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1530 speed += lp->lp_ifp->if_baudrate;
1531 sc->sc_ifp->if_baudrate = speed;
1533 case LAGG_PROTO_LACP:
1534 /* LACP updates if_baudrate itself */
1541 lagg_port_state(struct ifnet *ifp, int state)
1543 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1544 struct lagg_softc *sc = NULL;
1553 if (sc->sc_linkstate != NULL)
1554 (*sc->sc_linkstate)(lp);
1560 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1562 struct lagg_port *lp_next, *rval = NULL;
1563 // int new_link = LINK_STATE_DOWN;
1565 LAGG_RLOCK_ASSERT(sc);
1567 * Search a port which reports an active link state.
1572 if (LAGG_PORTACTIVE(lp)) {
1576 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1577 LAGG_PORTACTIVE(lp_next)) {
1583 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1584 if (LAGG_PORTACTIVE(lp_next)) {
1593 * The IEEE 802.1D standard assumes that a lagg with
1594 * multiple ports is always full duplex. This is valid
1595 * for load sharing laggs and if at least two links
1596 * are active. Unfortunately, checking the latter would
1597 * be too expensive at this point.
1599 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1601 new_link = LINK_STATE_FULL_DUPLEX;
1603 new_link = rval->lp_link_state;
1611 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1613 if (m->m_pkthdr.len < (off + len)) {
1615 } else if (m->m_len < (off + len)) {
1616 m_copydata(m, off, len, buf);
1619 return (mtod(m, char *) + off);
1623 lagg_sysctl_active(SYSCTL_HANDLER_ARGS)
1625 struct lagg_softc *sc = (struct lagg_softc *)arg1;
1626 struct lagg_port *lp;
1629 /* LACP tracks active links automatically, the others do not */
1630 if (sc->sc_proto != LAGG_PROTO_LACP) {
1632 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1633 sc->sc_active += LAGG_PORTACTIVE(lp);
1636 error = sysctl_handle_int(oidp, &sc->sc_active, 0, req);
1637 if ((error) || (req->newptr == NULL))
1644 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
1649 struct ether_header *eh;
1650 const struct ether_vlan_header *vlan;
1652 const struct ip *ip;
1653 const uint32_t *ports;
1657 const struct ip6_hdr *ip6;
1667 struct ether_vlan_header vlan;
1675 eh = mtod(m, struct ether_header *);
1676 etype = ntohs(eh->ether_type);
1677 if (sc->sc_flags & LAGG_F_HASHL2) {
1678 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p);
1679 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1682 /* Special handling for encapsulating VLAN frames */
1684 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) {
1685 p = hash32_buf(&m->m_pkthdr.ether_vtag,
1686 sizeof(m->m_pkthdr.ether_vtag), p);
1689 if (etype == ETHERTYPE_VLAN) {
1690 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf);
1694 if (sc->sc_flags & LAGG_F_HASHL2)
1695 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1696 etype = ntohs(vlan->evl_proto);
1697 off += sizeof(*vlan) - sizeof(*eh);
1703 ip = lagg_gethdr(m, off, sizeof(*ip), &buf);
1707 if (sc->sc_flags & LAGG_F_HASHL3) {
1708 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1709 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1711 if (!(sc->sc_flags & LAGG_F_HASHL4))
1716 iphlen = ip->ip_hl << 2;
1717 if (iphlen < sizeof(*ip))
1720 ports = lagg_gethdr(m, off, sizeof(*ports), &buf);
1723 p = hash32_buf(ports, sizeof(*ports), p);
1729 case ETHERTYPE_IPV6:
1730 if (!(sc->sc_flags & LAGG_F_HASHL3))
1732 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf);
1736 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1737 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1738 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1739 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */
1748 lagg_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1750 struct lagg_softc *sc = ifp->if_softc;
1752 struct ifnet *ifp_p;
1753 struct netmsg_packet *nmp;
1756 ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
1757 ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
1759 if (((ifp->if_flags & IFF_RUNNING) == 0)
1760 || (sc->sc_proto == LAGG_PROTO_NONE)
1761 || (sc->sc_count == 0)) {
1769 m = ifsq_dequeue(ifsq);
1773 // Choose output port
1774 ifp_p = (*sc->sc_select_tx_port)(sc, m);
1776 if (ifp_p == NULL) {
1780 p_port = netisr_cpuport(
1781 ifsq_get_cpuid(ifq_get_subq_default(&ifp_p->if_snd)));
1785 nmp = &m->m_hdr.mh_netmsg;
1787 netmsg_init(&nmp->base, NULL, &netisr_apanic_rport,
1788 0, lagg_start_dispatch);
1790 nmp->base.lmsg.u.ms_resultp = ifp_p;
1792 lwkt_sendmsg(p_port, &nmp->base.lmsg);
1793 IFNET_STAT_INC(ifp, opackets, 1);
1800 lagg_start_dispatch(netmsg_t msg)
1802 struct netmsg_packet *nmp = &msg->packet;
1805 struct altq_pktattr pktattr;
1808 ifp = msg->lmsg.u.ms_resultp;
1812 /* Does altq mix with lacp? */
1813 if (ifq_is_enabled(&ifp->if_snd))
1814 altq_etherclassify(&ifp->if_snd, m, &pktattr);
1816 ifq_dispatch(ifp, m, &pktattr);
1821 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1823 struct altq_pktattr pktattr;
1825 if (ifq_is_enabled(&ifp->if_snd))
1826 altq_etherclassify(&ifp->if_snd, m, &pktattr);
1828 ifq_dispatch(ifp, m, &pktattr);
1833 * Simple round robin aggregation
1836 lagg_rr_attach(struct lagg_softc *sc)
1838 sc->sc_detach = lagg_rr_detach;
1839 sc->sc_input = lagg_rr_input;
1840 sc->sc_select_tx_port = lagg_rr_select_tx_port;
1841 sc->sc_port_create = NULL;
1842 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1849 lagg_rr_detach(struct lagg_softc *sc)
1854 static struct ifnet *
1855 lagg_rr_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
1857 struct lagg_port *lp;
1860 p = atomic_fetchadd_32(&sc->sc_seq, 1);
1862 lp = SLIST_FIRST(&sc->sc_ports);
1864 lp = SLIST_NEXT(lp, lp_entries);
1867 * Check the port's link state. This will return the next active
1868 * port if the link is down or the port is NULL.
1870 if ((lp = lagg_link_active(sc, lp)) == NULL) {
1874 return (lp->lp_ifp);
1877 static struct mbuf *
1878 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1880 struct ifnet *ifp = sc->sc_ifp;
1882 /* Just pass in the packet to our lagg device */
1883 m->m_pkthdr.rcvif = ifp;
1893 lagg_fail_attach(struct lagg_softc *sc)
1895 sc->sc_detach = lagg_fail_detach;
1896 sc->sc_select_tx_port = lagg_fail_select_tx_port;
1897 sc->sc_input = lagg_fail_input;
1898 sc->sc_port_create = NULL;
1899 sc->sc_port_destroy = NULL;
1905 lagg_fail_detach(struct lagg_softc *sc)
1911 lagg_fail_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
1913 struct lagg_port *lp;
1915 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL)
1921 static struct mbuf *
1922 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1924 struct ifnet *ifp = sc->sc_ifp;
1925 struct lagg_port *tmp_tp;
1927 if (lp == sc->sc_primary || lagg_failover_rx_all) {
1928 m->m_pkthdr.rcvif = ifp;
1932 if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1933 tmp_tp = lagg_link_active(sc, sc->sc_primary);
1935 * If tmp_tp is null, we've recieved a packet when all
1936 * our links are down. Weird, but process it anyways.
1938 if ((tmp_tp == NULL || tmp_tp == lp)) {
1939 m->m_pkthdr.rcvif = ifp;
1953 lagg_lb_attach(struct lagg_softc *sc)
1955 struct lagg_port *lp;
1958 if ((lb = (struct lagg_lb *)kmalloc(sizeof(struct lagg_lb),
1959 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1962 sc->sc_detach = lagg_lb_detach;
1963 sc->sc_select_tx_port = lagg_lb_select_tx_port;
1964 sc->sc_input = lagg_lb_input;
1965 sc->sc_port_create = lagg_lb_port_create;
1966 sc->sc_port_destroy = lagg_lb_port_destroy;
1967 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1969 lb->lb_key = karc4random();
1970 sc->sc_psc = (caddr_t)lb;
1972 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1973 lagg_lb_port_create(lp);
1979 lagg_lb_detach(struct lagg_softc *sc)
1981 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1983 kfree(lb, M_DEVBUF);
1988 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1990 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1991 struct lagg_port *lp_next;
1994 bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1995 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1998 if (i >= LAGG_MAX_PORTS)
2000 if (sc->sc_ifflags & IFF_DEBUG)
2001 kprintf("%s: port %s at index %d\n",
2002 sc->sc_ifname, lp_next->lp_ifname, i);
2003 lb->lb_ports[i++] = lp_next;
2010 lagg_lb_port_create(struct lagg_port *lp)
2012 struct lagg_softc *sc = lp->lp_softc;
2013 return (lagg_lb_porttable(sc, NULL));
2017 lagg_lb_port_destroy(struct lagg_port *lp)
2019 struct lagg_softc *sc = lp->lp_softc;
2020 lagg_lb_porttable(sc, lp);
2025 lagg_lb_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
2027 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2028 struct lagg_port *lp = NULL;
2032 if (sc->use_flowid && (m->m_flags & M_FLOWID))
2033 p = m->m_pkthdr.flowid;
2036 p = lagg_hashmbuf(sc, m, lb->lb_key);
2038 lp = lb->lb_ports[p];
2041 * Check the port's link state. This will return the next active
2042 * port if the link is down or the port is NULL.
2044 if ((lp = lagg_link_active(sc, lp)) == NULL)
2050 static struct mbuf *
2051 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2053 struct ifnet *ifp = sc->sc_ifp;
2055 /* Just pass in the packet to our lagg device */
2056 m->m_pkthdr.rcvif = ifp;
2065 lagg_lacp_attach(struct lagg_softc *sc)
2067 struct lagg_port *lp;
2070 sc->sc_detach = lagg_lacp_detach;
2071 sc->sc_port_create = lacp_port_create;
2072 sc->sc_port_destroy = lacp_port_destroy;
2073 sc->sc_linkstate = lacp_linkstate;
2074 sc->sc_select_tx_port = lagg_lacp_select_tx_port;
2075 sc->sc_input = lagg_lacp_input;
2076 sc->sc_init = lacp_init;
2077 sc->sc_stop = lacp_stop;
2078 sc->sc_lladdr = lagg_lacp_lladdr;
2079 sc->sc_req = lacp_req;
2080 sc->sc_portreq = lacp_portreq;
2082 error = lacp_attach(sc);
2086 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2087 lacp_port_create(lp);
2093 lagg_lacp_detach(struct lagg_softc *sc)
2095 struct lagg_port *lp;
2098 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2099 lacp_port_destroy(lp);
2101 /* unlocking is safe here */
2103 error = lacp_detach(sc);
2110 lagg_lacp_lladdr(struct lagg_softc *sc)
2112 struct lagg_port *lp;
2114 /* purge all the lacp ports */
2115 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2116 lacp_port_destroy(lp);
2118 /* add them back in */
2119 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2120 lacp_port_create(lp);
2124 lagg_lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
2126 struct lagg_port *lp;
2127 lp = lacp_select_tx_port(sc, m);
2134 static struct mbuf *
2135 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2137 struct ifnet *ifp = sc->sc_ifp;
2138 struct ether_header *eh;
2141 eh = mtod(m, struct ether_header *);
2142 etype = ntohs(eh->ether_type);
2144 /* Tap off LACP control messages */
2145 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2146 m = lacp_input(lp, m);
2152 * If the port is not collecting or not in the active aggregator then
2155 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2160 m->m_pkthdr.rcvif = ifp;
2165 lagg_callout(void *arg)
2167 struct lagg_softc *sc = (struct lagg_softc *)arg;
2169 struct ifnet *ifp = sc->sc_ifp;
2171 ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets);
2172 ifp->if_opackets = counter_u64_fetch(sc->sc_opackets);
2173 ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes);
2174 ifp->if_obytes = counter_u64_fetch(sc->sc_obytes);
2177 callout_reset(&sc->sc_callout, hz, lagg_callout, sc);