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 lockuninit(&lagg_list_lock);
234 static moduledata_t lagg_mod = {
240 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
241 MODULE_VERSION(if_lagg, 1);
244 * This routine is run via an vlan
248 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
251 struct lagg_softc *sc = ifp->if_softc;
252 struct lagg_port *lp;
254 if (ifp->if_softc != arg) /* Not our event */
258 if (!SLIST_EMPTY(&sc->sc_ports)) {
259 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
260 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp);
268 * This routine is run via an vlan
272 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
274 struct lagg_softc *sc = ifp->if_softc;
276 struct lagg_port *lp;
279 if (ifp->if_softc != arg) /* Not our event */
283 if (!SLIST_EMPTY(&sc->sc_ports)) {
285 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
286 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp);
293 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params __unused)
295 struct lagg_softc *sc;
298 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */
299 struct sysctl_oid *oid;
300 char num[14]; /* sufficient for 32 bits */
302 sc = kmalloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
303 ifp = sc->sc_ifp = &sc->sc_if; // XXX if_alloc(IFT_ETHER);
311 sc->sc_ipackets = counter_u64_alloc(M_WAITOK);
312 sc->sc_opackets = counter_u64_alloc(M_WAITOK);
313 sc->sc_ibytes = counter_u64_alloc(M_WAITOK);
314 sc->sc_obytes = counter_u64_alloc(M_WAITOK);
316 sysctl_ctx_init(&sc->ctx);
317 ksnprintf(num, sizeof(num), "%u", unit);
318 sc->use_flowid = def_use_flowid;
319 sc->sc_oid = oid = SYSCTL_ADD_NODE(&sc->ctx,
320 &SYSCTL_NODE_CHILDREN(_net_link, lagg),
321 OID_AUTO, num, CTLFLAG_RD, NULL, "");
322 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
323 "use_flowid", CTLTYPE_INT|CTLFLAG_RW, &sc->use_flowid, sc->use_flowid,
324 "Use flow id for load sharing");
325 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
326 "count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count,
327 "Total number of ports");
328 SYSCTL_ADD_PROC(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
329 "active", CTLTYPE_INT|CTLFLAG_RD, sc, 0, lagg_sysctl_active,
330 "I", "Total number of active ports");
331 SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
332 "flapping", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_flapping,
333 sc->sc_flapping, "Total number of port change events");
334 /* Hash all layers by default */
335 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4;
337 sc->sc_proto = LAGG_PROTO_NONE;
338 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
339 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
340 sc->sc_proto = lagg_protos[i].ti_proto;
341 if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
350 LAGG_CALLOUT_LOCK_INIT(sc);
351 SLIST_INIT(&sc->sc_ports);
352 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
355 * This uses the callout lock rather than the rmlock; one can't
356 * hold said rmlock during SWI.
358 callout_init_lk(&sc->sc_callout, &sc->sc_call_lock);
360 /* Initialise pseudo media types */
361 ifmedia_init(&sc->sc_media, 0, lagg_media_change,
363 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
364 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
366 if_initname(ifp, laggname, unit);
369 ifp->if_transmit = lagg_transmit;
370 ifp->if_qflush = lagg_qflush;
372 ifp->if_mtu = ETHERMTU;
373 ifp->if_init = lagg_init;
374 ifp->if_ioctl = lagg_ioctl;
375 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
376 ifp->if_start = lagg_start;
377 ifp->if_type = IFT_ETHER;
378 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
379 ifq_set_ready(&ifp->if_snd);
380 ifp->if_hdrlen = ETHER_HDR_LEN;
383 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
386 * Attach as an ordinary ethernet device, children will be attached
387 * as special device IFT_IEEE8023ADLAG.
390 ether_ifattach(ifp, eaddr, NULL);
392 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
393 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
394 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
395 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
397 /* Insert into the global list of laggs */
398 lockmgr(&lagg_list_lock, LK_EXCLUSIVE);
399 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
400 lockmgr(&lagg_list_lock, LK_RELEASE);
402 callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
408 lagg_clone_destroy(struct ifnet *ifp)
410 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
411 struct lagg_port *lp;
416 ifp->if_flags &= ~IFF_UP;
418 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
419 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
421 /* Shutdown and remove lagg ports */
422 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
423 lagg_port_destroy(lp, 1);
424 /* Unhook the aggregation protocol */
425 if (sc->sc_detach != NULL)
426 (*sc->sc_detach)(sc);
430 sysctl_ctx_free(&sc->ctx);
431 ifmedia_removeall(&sc->sc_media);
433 /* This ifp is part of lagg_softc, don't free it! */
436 /* This grabs sc_callout_mtx, serialising it correctly */
437 callout_drain(&sc->sc_callout);
440 /* At this point it's drained; we can free this */
441 counter_u64_free(sc->sc_ipackets);
442 counter_u64_free(sc->sc_opackets);
443 counter_u64_free(sc->sc_ibytes);
444 counter_u64_free(sc->sc_obytes);
447 lockmgr(&lagg_list_lock, LK_EXCLUSIVE);
448 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
449 lockmgr(&lagg_list_lock, LK_RELEASE);
451 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
452 LAGG_LOCK_DESTROY(sc);
453 LAGG_CALLOUT_LOCK_DESTROY(sc);
460 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
462 struct ifnet *ifp = sc->sc_ifp;
464 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
467 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
468 /* Let the protocol know the MAC has changed */
469 if (sc->sc_lladdr != NULL)
470 (*sc->sc_lladdr)(sc);
471 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
475 lagg_capabilities(struct lagg_softc *sc)
477 struct lagg_port *lp;
478 int cap = ~0, ena = ~0;
481 LAGG_WLOCK_ASSERT(sc);
483 /* Get capabilities from the lagg ports */
484 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
485 cap &= lp->lp_ifp->if_capabilities;
486 ena &= lp->lp_ifp->if_capenable;
487 hwa &= lp->lp_ifp->if_hwassist;
489 cap = (cap == ~0 ? 0 : cap);
490 ena = (ena == ~0 ? 0 : ena);
491 hwa = (hwa == ~0 ? 0 : hwa);
493 if (sc->sc_ifp->if_capabilities != cap ||
494 sc->sc_ifp->if_capenable != ena ||
495 sc->sc_ifp->if_hwassist != hwa) {
496 sc->sc_ifp->if_capabilities = cap;
497 sc->sc_ifp->if_capenable = ena;
498 sc->sc_ifp->if_hwassist = hwa;
499 getmicrotime(&sc->sc_ifp->if_lastchange);
501 if (sc->sc_ifflags & IFF_DEBUG)
502 if_printf(sc->sc_ifp,
503 "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
508 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
510 struct lagg_softc *sc = lp->lp_softc;
511 struct ifnet *ifp = lp->lp_ifp;
512 struct lagg_llq *llq;
515 LAGG_WLOCK_ASSERT(sc);
517 if (lp->lp_detaching ||
518 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
521 /* Check to make sure its not already queued to be changed */
522 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
523 if (llq->llq_ifp == ifp) {
530 llq = kmalloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
531 if (llq == NULL) /* XXX what to do */
535 /* Update the lladdr even if pending, it may have changed */
537 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
540 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
542 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
546 * Set the interface MAC address from a taskqueue to avoid a LOR.
549 lagg_port_setlladdr(void *arg, int pending)
551 struct lagg_softc *sc = (struct lagg_softc *)arg;
552 struct lagg_llq *llq, *head;
556 /* Grab a local reference of the queue and remove it from the softc */
558 head = SLIST_FIRST(&sc->sc_llq_head);
559 SLIST_FIRST(&sc->sc_llq_head) = NULL;
563 * Traverse the queue and set the lladdr on each ifp. It is safe to do
564 * unlocked as we have the only reference to it.
566 for (llq = head; llq != NULL; llq = head) {
569 /* Set the link layer address */
570 /* CURVNET_SET(ifp->if_vnet); */
571 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
572 /* CURVNET_RESTORE(); */
574 kprintf("%s: setlladdr failed on %s\n", __func__,
577 head = SLIST_NEXT(llq, llq_entries);
578 kfree(llq, M_DEVBUF);
583 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
585 struct lagg_softc *sc_ptr;
586 struct lagg_port *lp;
589 LAGG_WLOCK_ASSERT(sc);
591 /* Limit the maximal number of lagg ports */
592 if (sc->sc_count >= LAGG_MAX_PORTS)
595 /* Check if port has already been associated to a lagg */
596 if (ifp->if_lagg != NULL) {
597 /* Port is already in the current lagg? */
598 lp = (struct lagg_port *)ifp->if_lagg;
599 if (lp->lp_softc == sc)
604 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */
605 if (ifp->if_type != IFT_ETHER)
606 return (EPROTONOSUPPORT);
610 * The member interface should not have inet6 address because
611 * two interfaces with a valid link-local scope zone must not be
612 * merged in any form. This restriction is needed to
613 * prevent violation of link-local scope zone. Attempts to
614 * add a member interface which has inet6 addresses triggers
615 * removal of all inet6 addresses on the member interface.
617 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
618 if (in6ifa_llaonifp(lp->lp_ifp)) {
619 in6_ifdetach(lp->lp_ifp);
620 if_printf(sc->sc_ifp,
621 "IPv6 addresses on %s have been removed "
622 "before adding it as a member to prevent "
623 "IPv6 address scope violation.\n",
624 lp->lp_ifp->if_xname);
627 if (in6ifa_llaonifp(ifp)) {
629 if_printf(sc->sc_ifp,
630 "IPv6 addresses on %s have been removed "
631 "before adding it as a member to prevent "
632 "IPv6 address scope violation.\n",
636 /* Allow the first Ethernet member to define the MTU */
637 if (SLIST_EMPTY(&sc->sc_ports))
638 sc->sc_ifp->if_mtu = ifp->if_mtu;
639 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
640 if_printf(sc->sc_ifp, "invalid MTU for %s\n",
645 if ((lp = kmalloc(sizeof(struct lagg_port),
646 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
649 /* Check if port is a stacked lagg */
650 lockmgr(&lagg_list_lock, LK_EXCLUSIVE);
651 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
652 if (ifp == sc_ptr->sc_ifp) {
653 lockmgr(&lagg_list_lock, LK_RELEASE);
656 /* XXX disable stacking for the moment, its untested */
657 #ifdef LAGG_PORT_STACKING
658 lp->lp_flags |= LAGG_PORT_STACK;
659 if (lagg_port_checkstacking(sc_ptr) >=
661 lockmgr(&lagg_list_lock, LK_RELEASE);
668 lockmgr(&lagg_list_lock, LK_RELEASE);
670 /* Change the interface type */
671 lp->lp_iftype = ifp->if_type;
672 ifp->if_type = IFT_IEEE8023ADLAG;
674 lp->lp_ioctl = ifp->if_ioctl;
675 ifp->if_ioctl = lagg_port_ioctl;
676 lp->lp_output = ifp->if_output;
677 ifp->if_output = lagg_port_output;
682 /* Save port link layer address */
683 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
685 if (SLIST_EMPTY(&sc->sc_ports)) {
687 lagg_lladdr(sc, IF_LLADDR(ifp));
689 /* Update link layer address for this port */
690 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
693 /* Insert into the list of ports */
694 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
697 /* Update lagg capabilities */
698 lagg_capabilities(sc);
701 /* Add multicast addresses and interface flags to this port */
702 lagg_ether_cmdmulti(lp, 1);
703 lagg_setflags(lp, 1);
705 if (sc->sc_port_create != NULL)
706 error = (*sc->sc_port_create)(lp);
708 /* remove the port again, without calling sc_port_destroy */
709 lagg_port_destroy(lp, 0);
716 #ifdef LAGG_PORT_STACKING
718 lagg_port_checkstacking(struct lagg_softc *sc)
720 struct lagg_softc *sc_ptr;
721 struct lagg_port *lp;
724 LAGG_WLOCK_ASSERT(sc);
726 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
727 if (lp->lp_flags & LAGG_PORT_STACK) {
728 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
729 m = MAX(m, lagg_port_checkstacking(sc_ptr));
738 lagg_port_destroy(struct lagg_port *lp, int runpd)
740 struct lagg_softc *sc = lp->lp_softc;
741 struct lagg_port *lp_ptr;
742 struct lagg_llq *llq;
743 struct ifnet *ifp = lp->lp_ifp;
745 LAGG_WLOCK_ASSERT(sc);
747 if (runpd && sc->sc_port_destroy != NULL)
748 (*sc->sc_port_destroy)(lp);
751 * Remove multicast addresses and interface flags from this port and
752 * reset the MAC address, skip if the interface is being detached.
754 if (!lp->lp_detaching) {
755 lagg_ether_cmdmulti(lp, 0);
756 lagg_setflags(lp, 0);
757 lagg_port_lladdr(lp, lp->lp_lladdr);
760 /* Restore interface */
761 ifp->if_type = lp->lp_iftype;
762 ifp->if_ioctl = lp->lp_ioctl;
763 ifp->if_output = lp->lp_output;
766 /* Finally, remove the port from the lagg */
767 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
770 /* Update the primary interface */
771 if (lp == sc->sc_primary) {
772 uint8_t lladdr[ETHER_ADDR_LEN];
774 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
775 bzero(&lladdr, ETHER_ADDR_LEN);
777 bcopy(lp_ptr->lp_lladdr,
778 lladdr, ETHER_ADDR_LEN);
780 lagg_lladdr(sc, lladdr);
781 sc->sc_primary = lp_ptr;
783 /* Update link layer address for each port */
784 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
785 lagg_port_lladdr(lp_ptr, lladdr);
788 /* Remove any pending lladdr changes from the queue */
789 if (lp->lp_detaching) {
790 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
791 if (llq->llq_ifp == ifp) {
792 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
794 kfree(llq, M_DEVBUF);
795 break; /* Only appears once */
801 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
805 /* Update lagg capabilities */
806 lagg_capabilities(sc);
813 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
815 struct lagg_reqport *rp = (struct lagg_reqport *)data;
816 struct lagg_softc *sc;
817 struct lagg_port *lp = NULL;
820 ASSERT_IFNET_SERIALIZED_ALL(ifp);
822 /* Should be checked by the caller */
823 if (ifp->if_type != IFT_IEEE8023ADLAG ||
824 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
829 if (rp->rp_portname[0] == '\0') {
835 * Release ifp serializers before ifnet_lock
836 * to prevent lock order reversal.
838 ifnet_deserialize_all(ifp);
840 if (ifunit(rp->rp_portname) != ifp) {
842 ifnet_serialize_all(ifp);
847 ifnet_serialize_all(ifp);
850 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
856 lagg_port2req(lp, rp);
861 if (lp->lp_ioctl == NULL) {
866 error = (*lp->lp_ioctl)(ifp, cmd, data, cr);
870 /* Update lagg interface capabilities */
872 lagg_capabilities(sc);
876 if (lp->lp_ioctl == NULL) {
881 error = (*lp->lp_ioctl)(ifp, cmd, data, cr);
884 /* Do not allow the MTU to be changed once joined */
895 if (lp->lp_ioctl != NULL) {
897 result = ((*lp->lp_ioctl)(ifp, cmd, data, cr));
903 * For direct output to child ports.
906 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
907 struct sockaddr *dst, struct rtentry *ro)
909 struct lagg_port *lp = ifp->if_lagg;
911 switch (dst->sa_family) {
912 case pseudo_AF_HDRCMPLT:
914 return ((*lp->lp_output)(ifp, m, dst, ro));
917 /* drop any other frames */
923 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
925 struct lagg_port *lp;
926 struct lagg_softc *sc;
928 if ((lp = ifp->if_lagg) == NULL)
931 /* If the ifnet is just being renamed, don't do anything. */
932 if (ifp->if_flags & IFF_RENAMING)
938 lp->lp_detaching = 1;
939 lagg_port_destroy(lp, 1);
944 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
946 struct lagg_softc *sc = lp->lp_softc;
948 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
949 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
950 rp->rp_prio = lp->lp_prio;
951 rp->rp_flags = lp->lp_flags;
952 if (sc->sc_portreq != NULL)
953 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
955 /* Add protocol specific flags */
956 switch (sc->sc_proto) {
957 case LAGG_PROTO_FAILOVER:
958 if (lp == sc->sc_primary)
959 rp->rp_flags |= LAGG_PORT_MASTER;
960 if (lp == lagg_link_active(sc, sc->sc_primary))
961 rp->rp_flags |= LAGG_PORT_ACTIVE;
964 case LAGG_PROTO_ROUNDROBIN:
965 case LAGG_PROTO_LOADBALANCE:
966 case LAGG_PROTO_ETHERCHANNEL:
967 if (LAGG_PORTACTIVE(lp))
968 rp->rp_flags |= LAGG_PORT_ACTIVE;
971 case LAGG_PROTO_LACP:
972 /* LACP has a different definition of active */
973 if (lacp_isactive(lp))
974 rp->rp_flags |= LAGG_PORT_ACTIVE;
975 if (lacp_iscollecting(lp))
976 rp->rp_flags |= LAGG_PORT_COLLECTING;
977 if (lacp_isdistributing(lp))
978 rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
987 struct lagg_softc *sc = (struct lagg_softc *)xsc;
988 struct lagg_port *lp;
989 struct ifnet *ifp = sc->sc_ifp;
991 if (ifp->if_flags & IFF_RUNNING)
996 ifp->if_flags |= IFF_RUNNING;
997 /* Update the port lladdrs */
998 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
999 lagg_port_lladdr(lp, IF_LLADDR(ifp));
1001 if (sc->sc_init != NULL)
1008 lagg_stop(struct lagg_softc *sc)
1010 struct ifnet *ifp = sc->sc_ifp;
1012 LAGG_WLOCK_ASSERT(sc);
1014 if ((ifp->if_flags & IFF_RUNNING) == 0)
1017 ifp->if_flags &= ~IFF_RUNNING;
1019 if (sc->sc_stop != NULL)
1024 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
1026 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1027 struct lagg_reqall *ra = (struct lagg_reqall *)data;
1028 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1029 struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1030 struct ifreq *ifr = (struct ifreq *)data;
1031 struct lagg_port *lp;
1033 struct thread *td = curthread;
1035 int count, buflen, len, error = 0;
1037 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1039 bzero(&rpbuf, sizeof(rpbuf));
1045 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1047 buflen = count * sizeof(struct lagg_reqport);
1050 outbuf = kmalloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1053 ra->ra_proto = sc->sc_proto;
1054 if (sc->sc_req != NULL)
1055 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
1059 len = min(ra->ra_size, buflen);
1060 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1061 if (len < sizeof(rpbuf))
1064 lagg_port2req(lp, &rpbuf);
1065 memcpy(buf, &rpbuf, sizeof(rpbuf));
1067 buf += sizeof(rpbuf);
1068 len -= sizeof(rpbuf);
1071 ra->ra_ports = count;
1072 ra->ra_size = count * sizeof(rpbuf);
1073 error = copyout(outbuf, ra->ra_port, ra->ra_size);
1074 kfree(outbuf, M_TEMP);
1077 error = priv_check(td, PRIV_NET_LAGG);
1080 if (ra->ra_proto >= LAGG_PROTO_MAX) {
1081 error = EPROTONOSUPPORT;
1085 if (sc->sc_proto != LAGG_PROTO_NONE) {
1086 /* Reset protocol first in case detach unlocks */
1087 sc->sc_proto = LAGG_PROTO_NONE;
1088 error = sc->sc_detach(sc);
1089 sc->sc_detach = NULL;
1090 sc->sc_start = NULL;
1091 sc->sc_input = NULL;
1092 sc->sc_port_create = NULL;
1093 sc->sc_port_destroy = NULL;
1094 sc->sc_linkstate = NULL;
1097 sc->sc_lladdr = NULL;
1099 sc->sc_portreq = NULL;
1100 } else if (sc->sc_input != NULL) {
1101 /* Still detaching */
1108 for (int i = 0; i < (sizeof(lagg_protos) /
1109 sizeof(lagg_protos[0])); i++) {
1110 if (lagg_protos[i].ti_proto == ra->ra_proto) {
1111 if (sc->sc_ifflags & IFF_DEBUG)
1112 kprintf("%s: using proto %u\n",
1114 lagg_protos[i].ti_proto);
1115 sc->sc_proto = lagg_protos[i].ti_proto;
1116 if (sc->sc_proto != LAGG_PROTO_NONE)
1117 error = lagg_protos[i].ti_attach(sc);
1123 error = EPROTONOSUPPORT;
1125 case SIOCGLAGGFLAGS:
1126 rf->rf_flags = sc->sc_flags;
1129 error = priv_check(td, PRIV_NET_LAGG);
1132 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1137 sc->sc_flags &= ~LAGG_F_HASHMASK;
1138 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK;
1143 * Release ifp serializers before ifnet_lock
1144 * to prevent lock order reversal.
1146 ifnet_deserialize_all(ifp);
1148 if (rp->rp_portname[0] == '\0' ||
1149 (tpif = ifunit(rp->rp_portname)) == NULL) {
1151 ifnet_serialize_all(ifp);
1157 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1158 lp->lp_softc != sc) {
1162 ifnet_serialize_all(ifp);
1166 lagg_port2req(lp, rp);
1169 ifnet_serialize_all(ifp);
1172 error = priv_check(td, PRIV_NET_LAGG);
1176 * Release ifp serializers before ifnet_lock
1177 * to prevent lock order reversal.
1179 ifnet_deserialize_all(ifp);
1181 if (rp->rp_portname[0] == '\0' ||
1182 (tpif = ifunit(rp->rp_portname)) == NULL) {
1184 ifnet_serialize_all(ifp);
1189 error = lagg_port_create(sc, tpif);
1192 ifnet_serialize_all(ifp);
1194 case SIOCSLAGGDELPORT:
1195 error = priv_check(td, PRIV_NET_LAGG);
1199 * Release ifp serializers before ifnet_lock
1200 * to prevent lock order reversal.
1202 ifnet_deserialize_all(ifp);
1204 if (rp->rp_portname[0] == '\0' ||
1205 (tpif = ifunit(rp->rp_portname)) == NULL) {
1207 ifnet_serialize_all(ifp);
1213 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1214 lp->lp_softc != sc) {
1218 ifnet_serialize_all(ifp);
1222 error = lagg_port_destroy(lp, 1);
1225 ifnet_serialize_all(ifp);
1228 /* Set flags on ports too */
1230 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1231 lagg_setflags(lp, 1);
1235 if (!(ifp->if_flags & IFF_UP) &&
1236 (ifp->if_flags & IFF_RUNNING)) {
1238 * If interface is marked down and it is running,
1239 * then stop and disable it.
1244 } else if ((ifp->if_flags & IFF_UP) &&
1245 !(ifp->if_flags & IFF_RUNNING)) {
1247 * If interface is marked up and it is stopped, then
1250 (*ifp->if_init)(sc);
1256 error = lagg_ether_setmulti(sc);
1261 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1266 /* Do not allow the MTU or caps to be directly changed */
1271 error = ether_ioctl(ifp, cmd, data);
1278 lagg_ether_setmulti(struct lagg_softc *sc)
1280 struct lagg_port *lp;
1282 LAGG_WLOCK_ASSERT(sc);
1284 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1285 /* First, remove any existing filter entries. */
1286 lagg_ether_cmdmulti(lp, 0);
1287 /* copy all addresses from the lagg interface to the port */
1288 lagg_ether_cmdmulti(lp, 1);
1294 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1296 struct lagg_softc *sc = lp->lp_softc;
1297 struct ifnet *ifp = lp->lp_ifp;
1298 struct ifnet *scifp = sc->sc_ifp;
1300 struct ifmultiaddr *ifma, *rifma = NULL;
1301 struct sockaddr_dl sdl;
1304 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp);
1305 LAGG_WLOCK_ASSERT(sc);
1307 bzero((char *)&sdl, sizeof(sdl));
1308 sdl.sdl_len = sizeof(sdl);
1309 sdl.sdl_family = AF_LINK;
1310 sdl.sdl_type = IFT_ETHER;
1311 sdl.sdl_alen = ETHER_ADDR_LEN;
1312 sdl.sdl_index = ifp->if_index;
1315 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1316 if (ifma->ifma_addr->sa_family != AF_LINK)
1318 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1319 LLADDR(&sdl), ETHER_ADDR_LEN);
1321 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
1324 mc = kmalloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1327 mc->mc_ifma = rifma;
1328 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1331 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1332 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1333 if_delmulti(ifp, (struct sockaddr *)mc->mc_ifma);
1334 kfree(mc, M_DEVBUF);
1340 /* Handle a ref counted flag that should be set on the lagg port as well */
1342 lagg_setflag(struct lagg_port *lp, int flag, int status,
1343 int (*func)(struct ifnet *, int))
1345 struct lagg_softc *sc = lp->lp_softc;
1346 struct ifnet *scifp = sc->sc_ifp;
1347 struct ifnet *ifp = lp->lp_ifp;
1350 LAGG_WLOCK_ASSERT(sc);
1351 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp);
1353 status = status ? (scifp->if_flags & flag) : 0;
1354 /* Now "status" contains the flag value or 0 */
1357 * See if recorded ports status is different from what
1358 * we want it to be. If it is, flip it. We record ports
1359 * status in lp_ifflags so that we won't clear ports flag
1360 * we haven't set. In fact, we don't clear or set ports
1361 * flags directly, but get or release references to them.
1362 * That's why we can be sure that recorded flags still are
1363 * in accord with actual ports flags.
1365 if (status != (lp->lp_ifflags & flag)) {
1366 error = (*func)(ifp, status);
1369 lp->lp_ifflags &= ~flag;
1370 lp->lp_ifflags |= status;
1376 * Handle IFF_* flags that require certain changes on the lagg port
1377 * if "status" is true, update ports flags respective to the lagg
1378 * if "status" is false, forcedly clear the flags set on port.
1381 lagg_setflags(struct lagg_port *lp, int status)
1385 ASSERT_IFNET_NOT_SERIALIZED_ALL(lp->lp_ifp);
1387 for (i = 0; lagg_pflags[i].flag; i++) {
1388 error = lagg_setflag(lp, lagg_pflags[i].flag,
1389 status, lagg_pflags[i].func);
1397 #ifdef XXX /* not needed? */
1399 lagg_output(struct ifnet *ifp, struct mbuf *m)
1401 struct lagg_softc *sc = ifp->if_softc;
1402 int error, len, mcast;
1404 len = m->m_pkthdr.len;
1405 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1408 /* We need a Tx algorithm and at least one port */
1409 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1418 error = (*sc->sc_start)(sc, m);
1422 IFNET_STAT_INC(ifp, opackets, 1);
1423 IFNET_STAT_INC(ifp, obytes, len);
1424 ifp->if_omcasts += mcast;
1434 * The ifp->if_qflush entry point for lagg(4) is no-op.
1437 lagg_qflush(struct ifnet *ifp __unused)
1442 lagg_input(struct ifnet *ifp, struct mbuf *m)
1444 struct lagg_port *lp = ifp->if_lagg;
1445 struct lagg_softc *sc = lp->lp_softc;
1446 struct ifnet *scifp = sc->sc_ifp;
1449 if ((scifp->if_flags & IFF_RUNNING) == 0 ||
1450 (lp->lp_flags & LAGG_PORT_DISABLED) ||
1451 sc->sc_proto == LAGG_PROTO_NONE) {
1458 m = (*sc->sc_input)(sc, lp, m);
1463 IFNET_STAT_INC(ifp, ipackets, 1);
1464 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len);
1466 if (scifp->if_flags & IFF_MONITOR) {
1470 ether_reinput_oncpu(scifp, m, REINPUT_RUNBPF);
1475 lagg_media_change(struct ifnet *ifp)
1477 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1479 if (sc->sc_ifflags & IFF_DEBUG)
1480 kprintf("%s\n", __func__);
1486 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1488 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1489 struct lagg_port *lp;
1491 imr->ifm_status = IFM_AVALID;
1492 imr->ifm_active = IFM_ETHER | IFM_AUTO;
1495 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1496 if (LAGG_PORTACTIVE(lp))
1497 imr->ifm_status |= IFM_ACTIVE;
1503 lagg_linkstate(struct lagg_softc *sc)
1505 struct lagg_port *lp;
1506 int new_link = LINK_STATE_DOWN;
1509 /* Our link is considered up if at least one of our ports is active */
1510 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1511 if (lp->lp_link_state == LINK_STATE_UP) {
1512 new_link = LINK_STATE_UP;
1516 if_link_state_change(sc->sc_ifp);
1518 /* Update if_baudrate to reflect the max possible speed */
1519 switch (sc->sc_proto) {
1520 case LAGG_PROTO_FAILOVER:
1521 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1522 sc->sc_primary->lp_ifp->if_baudrate : 0;
1524 case LAGG_PROTO_ROUNDROBIN:
1525 case LAGG_PROTO_LOADBALANCE:
1526 case LAGG_PROTO_ETHERCHANNEL:
1528 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1529 speed += lp->lp_ifp->if_baudrate;
1530 sc->sc_ifp->if_baudrate = speed;
1532 case LAGG_PROTO_LACP:
1533 /* LACP updates if_baudrate itself */
1540 lagg_port_state(struct ifnet *ifp, int state)
1542 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1543 struct lagg_softc *sc = NULL;
1552 if (sc->sc_linkstate != NULL)
1553 (*sc->sc_linkstate)(lp);
1559 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1561 struct lagg_port *lp_next, *rval = NULL;
1562 // int new_link = LINK_STATE_DOWN;
1564 LAGG_RLOCK_ASSERT(sc);
1566 * Search a port which reports an active link state.
1571 if (LAGG_PORTACTIVE(lp)) {
1575 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1576 LAGG_PORTACTIVE(lp_next)) {
1582 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1583 if (LAGG_PORTACTIVE(lp_next)) {
1592 * The IEEE 802.1D standard assumes that a lagg with
1593 * multiple ports is always full duplex. This is valid
1594 * for load sharing laggs and if at least two links
1595 * are active. Unfortunately, checking the latter would
1596 * be too expensive at this point.
1598 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1600 new_link = LINK_STATE_FULL_DUPLEX;
1602 new_link = rval->lp_link_state;
1610 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1612 if (m->m_pkthdr.len < (off + len)) {
1614 } else if (m->m_len < (off + len)) {
1615 m_copydata(m, off, len, buf);
1618 return (mtod(m, char *) + off);
1622 lagg_sysctl_active(SYSCTL_HANDLER_ARGS)
1624 struct lagg_softc *sc = (struct lagg_softc *)arg1;
1625 struct lagg_port *lp;
1628 /* LACP tracks active links automatically, the others do not */
1629 if (sc->sc_proto != LAGG_PROTO_LACP) {
1631 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1632 sc->sc_active += LAGG_PORTACTIVE(lp);
1635 error = sysctl_handle_int(oidp, &sc->sc_active, 0, req);
1636 if ((error) || (req->newptr == NULL))
1643 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
1648 struct ether_header *eh;
1649 const struct ether_vlan_header *vlan;
1651 const struct ip *ip;
1652 const uint32_t *ports;
1656 const struct ip6_hdr *ip6;
1666 struct ether_vlan_header vlan;
1674 eh = mtod(m, struct ether_header *);
1675 etype = ntohs(eh->ether_type);
1676 if (sc->sc_flags & LAGG_F_HASHL2) {
1677 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p);
1678 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1681 /* Special handling for encapsulating VLAN frames */
1683 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) {
1684 p = hash32_buf(&m->m_pkthdr.ether_vtag,
1685 sizeof(m->m_pkthdr.ether_vtag), p);
1688 if (etype == ETHERTYPE_VLAN) {
1689 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf);
1693 if (sc->sc_flags & LAGG_F_HASHL2)
1694 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1695 etype = ntohs(vlan->evl_proto);
1696 off += sizeof(*vlan) - sizeof(*eh);
1702 ip = lagg_gethdr(m, off, sizeof(*ip), &buf);
1706 if (sc->sc_flags & LAGG_F_HASHL3) {
1707 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1708 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1710 if (!(sc->sc_flags & LAGG_F_HASHL4))
1715 iphlen = ip->ip_hl << 2;
1716 if (iphlen < sizeof(*ip))
1719 ports = lagg_gethdr(m, off, sizeof(*ports), &buf);
1722 p = hash32_buf(ports, sizeof(*ports), p);
1728 case ETHERTYPE_IPV6:
1729 if (!(sc->sc_flags & LAGG_F_HASHL3))
1731 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf);
1735 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1736 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1737 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1738 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */
1747 lagg_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1749 struct lagg_softc *sc = ifp->if_softc;
1751 struct ifnet *ifp_p;
1752 struct netmsg_packet *nmp;
1755 ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
1756 ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
1758 if (((ifp->if_flags & IFF_RUNNING) == 0)
1759 || (sc->sc_proto == LAGG_PROTO_NONE)
1760 || (sc->sc_count == 0)) {
1768 m = ifsq_dequeue(ifsq);
1772 // Choose output port
1773 ifp_p = (*sc->sc_select_tx_port)(sc, m);
1775 if (ifp_p == NULL) {
1779 p_port = netisr_cpuport(
1780 ifsq_get_cpuid(ifq_get_subq_default(&ifp_p->if_snd)));
1784 nmp = &m->m_hdr.mh_netmsg;
1786 netmsg_init(&nmp->base, NULL, &netisr_apanic_rport,
1787 0, lagg_start_dispatch);
1789 nmp->base.lmsg.u.ms_resultp = ifp_p;
1791 lwkt_sendmsg(p_port, &nmp->base.lmsg);
1792 IFNET_STAT_INC(ifp, opackets, 1);
1799 lagg_start_dispatch(netmsg_t msg)
1801 struct netmsg_packet *nmp = &msg->packet;
1804 struct altq_pktattr pktattr;
1807 ifp = msg->lmsg.u.ms_resultp;
1811 /* Does altq mix with lacp? */
1812 if (ifq_is_enabled(&ifp->if_snd))
1813 altq_etherclassify(&ifp->if_snd, m, &pktattr);
1815 ifq_dispatch(ifp, m, &pktattr);
1820 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1822 struct altq_pktattr pktattr;
1824 if (ifq_is_enabled(&ifp->if_snd))
1825 altq_etherclassify(&ifp->if_snd, m, &pktattr);
1827 ifq_dispatch(ifp, m, &pktattr);
1832 * Simple round robin aggregation
1835 lagg_rr_attach(struct lagg_softc *sc)
1837 sc->sc_detach = lagg_rr_detach;
1838 sc->sc_input = lagg_rr_input;
1839 sc->sc_select_tx_port = lagg_rr_select_tx_port;
1840 sc->sc_port_create = NULL;
1841 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1848 lagg_rr_detach(struct lagg_softc *sc)
1853 static struct ifnet *
1854 lagg_rr_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
1856 struct lagg_port *lp;
1859 p = atomic_fetchadd_32(&sc->sc_seq, 1);
1861 lp = SLIST_FIRST(&sc->sc_ports);
1863 lp = SLIST_NEXT(lp, lp_entries);
1866 * Check the port's link state. This will return the next active
1867 * port if the link is down or the port is NULL.
1869 if ((lp = lagg_link_active(sc, lp)) == NULL) {
1873 return (lp->lp_ifp);
1876 static struct mbuf *
1877 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1879 struct ifnet *ifp = sc->sc_ifp;
1881 /* Just pass in the packet to our lagg device */
1882 m->m_pkthdr.rcvif = ifp;
1892 lagg_fail_attach(struct lagg_softc *sc)
1894 sc->sc_detach = lagg_fail_detach;
1895 sc->sc_select_tx_port = lagg_fail_select_tx_port;
1896 sc->sc_input = lagg_fail_input;
1897 sc->sc_port_create = NULL;
1898 sc->sc_port_destroy = NULL;
1904 lagg_fail_detach(struct lagg_softc *sc)
1910 lagg_fail_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
1912 struct lagg_port *lp;
1914 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL)
1920 static struct mbuf *
1921 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1923 struct ifnet *ifp = sc->sc_ifp;
1924 struct lagg_port *tmp_tp;
1926 if (lp == sc->sc_primary || lagg_failover_rx_all) {
1927 m->m_pkthdr.rcvif = ifp;
1931 if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1932 tmp_tp = lagg_link_active(sc, sc->sc_primary);
1934 * If tmp_tp is null, we've recieved a packet when all
1935 * our links are down. Weird, but process it anyways.
1937 if ((tmp_tp == NULL || tmp_tp == lp)) {
1938 m->m_pkthdr.rcvif = ifp;
1952 lagg_lb_attach(struct lagg_softc *sc)
1954 struct lagg_port *lp;
1957 if ((lb = (struct lagg_lb *)kmalloc(sizeof(struct lagg_lb),
1958 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1961 sc->sc_detach = lagg_lb_detach;
1962 sc->sc_select_tx_port = lagg_lb_select_tx_port;
1963 sc->sc_input = lagg_lb_input;
1964 sc->sc_port_create = lagg_lb_port_create;
1965 sc->sc_port_destroy = lagg_lb_port_destroy;
1966 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1968 lb->lb_key = karc4random();
1969 sc->sc_psc = (caddr_t)lb;
1971 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1972 lagg_lb_port_create(lp);
1978 lagg_lb_detach(struct lagg_softc *sc)
1980 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1982 kfree(lb, M_DEVBUF);
1987 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1989 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1990 struct lagg_port *lp_next;
1993 bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1994 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1997 if (i >= LAGG_MAX_PORTS)
1999 if (sc->sc_ifflags & IFF_DEBUG)
2000 kprintf("%s: port %s at index %d\n",
2001 sc->sc_ifname, lp_next->lp_ifname, i);
2002 lb->lb_ports[i++] = lp_next;
2009 lagg_lb_port_create(struct lagg_port *lp)
2011 struct lagg_softc *sc = lp->lp_softc;
2012 return (lagg_lb_porttable(sc, NULL));
2016 lagg_lb_port_destroy(struct lagg_port *lp)
2018 struct lagg_softc *sc = lp->lp_softc;
2019 lagg_lb_porttable(sc, lp);
2024 lagg_lb_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
2026 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2027 struct lagg_port *lp = NULL;
2031 if (sc->use_flowid && (m->m_flags & M_FLOWID))
2032 p = m->m_pkthdr.flowid;
2035 p = lagg_hashmbuf(sc, m, lb->lb_key);
2037 lp = lb->lb_ports[p];
2040 * Check the port's link state. This will return the next active
2041 * port if the link is down or the port is NULL.
2043 if ((lp = lagg_link_active(sc, lp)) == NULL)
2049 static struct mbuf *
2050 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2052 struct ifnet *ifp = sc->sc_ifp;
2054 /* Just pass in the packet to our lagg device */
2055 m->m_pkthdr.rcvif = ifp;
2064 lagg_lacp_attach(struct lagg_softc *sc)
2066 struct lagg_port *lp;
2069 sc->sc_detach = lagg_lacp_detach;
2070 sc->sc_port_create = lacp_port_create;
2071 sc->sc_port_destroy = lacp_port_destroy;
2072 sc->sc_linkstate = lacp_linkstate;
2073 sc->sc_select_tx_port = lagg_lacp_select_tx_port;
2074 sc->sc_input = lagg_lacp_input;
2075 sc->sc_init = lacp_init;
2076 sc->sc_stop = lacp_stop;
2077 sc->sc_lladdr = lagg_lacp_lladdr;
2078 sc->sc_req = lacp_req;
2079 sc->sc_portreq = lacp_portreq;
2081 error = lacp_attach(sc);
2085 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2086 lacp_port_create(lp);
2092 lagg_lacp_detach(struct lagg_softc *sc)
2094 struct lagg_port *lp;
2097 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2098 lacp_port_destroy(lp);
2100 /* unlocking is safe here */
2102 error = lacp_detach(sc);
2109 lagg_lacp_lladdr(struct lagg_softc *sc)
2111 struct lagg_port *lp;
2113 /* purge all the lacp ports */
2114 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2115 lacp_port_destroy(lp);
2117 /* add them back in */
2118 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2119 lacp_port_create(lp);
2123 lagg_lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
2125 struct lagg_port *lp;
2126 lp = lacp_select_tx_port(sc, m);
2133 static struct mbuf *
2134 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2136 struct ifnet *ifp = sc->sc_ifp;
2137 struct ether_header *eh;
2140 eh = mtod(m, struct ether_header *);
2141 etype = ntohs(eh->ether_type);
2143 /* Tap off LACP control messages */
2144 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2145 m = lacp_input(lp, m);
2151 * If the port is not collecting or not in the active aggregator then
2154 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2159 m->m_pkthdr.rcvif = ifp;
2164 lagg_callout(void *arg)
2166 struct lagg_softc *sc = (struct lagg_softc *)arg;
2168 struct ifnet *ifp = sc->sc_ifp;
2170 ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets);
2171 ifp->if_opackets = counter_u64_fetch(sc->sc_opackets);
2172 ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes);
2173 ifp->if_obytes = counter_u64_fetch(sc->sc_obytes);
2176 callout_reset(&sc->sc_callout, hz, lagg_callout, sc);