1 /* $OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $ */
4 * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org>
6 * Permission to use, copy, modify, and distribute this software for any
7 * purpose with or without fee is hereby granted, provided that the above
8 * copyright notice and this permission notice appear in all copies.
10 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
11 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
13 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
14 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
15 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
16 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19 #include <sys/cdefs.h>
20 __FBSDID("$FreeBSD$");
23 #include "opt_inet6.h"
25 #include <sys/param.h>
26 #include <sys/kernel.h>
27 #include <sys/malloc.h>
29 #include <sys/queue.h>
30 #include <sys/socket.h>
31 #include <sys/sockio.h>
32 #include <sys/sysctl.h>
33 #include <sys/module.h>
35 #include <sys/systm.h>
39 #include <sys/rwlock.h>
40 #include <sys/taskqueue.h>
42 #include <net/ethernet.h>
44 #include <net/if_clone.h>
45 #include <net/if_arp.h>
46 #include <net/if_dl.h>
47 #include <net/if_llc.h>
48 #include <net/if_media.h>
49 #include <net/if_types.h>
50 #include <net/if_var.h>
54 #include <netinet/in.h>
55 #include <netinet/in_systm.h>
56 #include <netinet/if_ether.h>
57 #include <netinet/ip.h>
61 #include <netinet/ip6.h>
64 #include <net/if_vlan_var.h>
65 #include <net/if_lagg.h>
66 #include <net/ieee8023ad_lacp.h>
68 /* Special flags we should propagate to the lagg ports. */
71 int (*func)(struct ifnet *, int);
73 {IFF_PROMISC, ifpromisc},
74 {IFF_ALLMULTI, if_allmulti},
78 SLIST_HEAD(__trhead, lagg_softc) lagg_list; /* list of laggs */
79 static struct mtx lagg_list_mtx;
80 eventhandler_tag lagg_detach_cookie = NULL;
82 static int lagg_clone_create(struct if_clone *, int, caddr_t);
83 static void lagg_clone_destroy(struct ifnet *);
84 static void lagg_lladdr(struct lagg_softc *, uint8_t *);
85 static int lagg_capabilities(struct lagg_softc *);
86 static void lagg_port_lladdr(struct lagg_port *, uint8_t *);
87 static void lagg_port_setlladdr(void *, int);
88 static int lagg_port_create(struct lagg_softc *, struct ifnet *);
89 static int lagg_port_destroy(struct lagg_port *, int);
90 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
91 static void lagg_port_state(struct ifnet *, int);
92 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
93 static int lagg_port_output(struct ifnet *, struct mbuf *,
94 struct sockaddr *, struct rtentry *);
95 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *);
96 static int lagg_port_checkstacking(struct lagg_softc *);
97 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *);
98 static void lagg_init(void *);
99 static void lagg_stop(struct lagg_softc *);
100 static int lagg_ioctl(struct ifnet *, u_long, caddr_t);
101 static int lagg_ether_setmulti(struct lagg_softc *);
102 static int lagg_ether_cmdmulti(struct lagg_port *, int);
103 static int lagg_setflag(struct lagg_port *, int, int,
104 int (*func)(struct ifnet *, int));
105 static int lagg_setflags(struct lagg_port *, int status);
106 static void lagg_start(struct ifnet *);
107 static int lagg_media_change(struct ifnet *);
108 static void lagg_media_status(struct ifnet *, struct ifmediareq *);
109 static struct lagg_port *lagg_link_active(struct lagg_softc *,
111 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
113 IFC_SIMPLE_DECLARE(lagg, 0);
115 /* Simple round robin */
116 static int lagg_rr_attach(struct lagg_softc *);
117 static int lagg_rr_detach(struct lagg_softc *);
118 static void lagg_rr_port_destroy(struct lagg_port *);
119 static int lagg_rr_start(struct lagg_softc *, struct mbuf *);
120 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
123 /* Active failover */
124 static int lagg_fail_attach(struct lagg_softc *);
125 static int lagg_fail_detach(struct lagg_softc *);
126 static int lagg_fail_start(struct lagg_softc *, struct mbuf *);
127 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
131 static int lagg_lb_attach(struct lagg_softc *);
132 static int lagg_lb_detach(struct lagg_softc *);
133 static int lagg_lb_port_create(struct lagg_port *);
134 static void lagg_lb_port_destroy(struct lagg_port *);
135 static int lagg_lb_start(struct lagg_softc *, struct mbuf *);
136 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
138 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
141 static int lagg_lacp_attach(struct lagg_softc *);
142 static int lagg_lacp_detach(struct lagg_softc *);
143 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *);
144 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
146 static void lagg_lacp_lladdr(struct lagg_softc *);
148 /* lagg protocol table */
149 static const struct {
151 int (*ti_attach)(struct lagg_softc *);
153 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach },
154 { LAGG_PROTO_FAILOVER, lagg_fail_attach },
155 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach },
156 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach },
157 { LAGG_PROTO_LACP, lagg_lacp_attach },
158 { LAGG_PROTO_NONE, NULL }
162 lagg_modevent(module_t mod, int type, void *data)
167 mtx_init(&lagg_list_mtx, "if_lagg list", NULL, MTX_DEF);
168 SLIST_INIT(&lagg_list);
169 if_clone_attach(&lagg_cloner);
170 lagg_input_p = lagg_input;
171 lagg_linkstate_p = lagg_port_state;
172 lagg_detach_cookie = EVENTHANDLER_REGISTER(
173 ifnet_departure_event, lagg_port_ifdetach, NULL,
174 EVENTHANDLER_PRI_ANY);
177 EVENTHANDLER_DEREGISTER(ifnet_departure_event,
179 if_clone_detach(&lagg_cloner);
181 lagg_linkstate_p = NULL;
182 mtx_destroy(&lagg_list_mtx);
190 static moduledata_t lagg_mod = {
196 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
199 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
201 struct lagg_softc *sc;
204 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */
206 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
207 ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
213 sc->sc_proto = LAGG_PROTO_NONE;
214 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
215 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
216 sc->sc_proto = lagg_protos[i].ti_proto;
217 if ((error = lagg_protos[i].ti_attach(sc)) != 0) {
218 if_free_type(ifp, IFT_ETHER);
226 SLIST_INIT(&sc->sc_ports);
227 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
229 /* Initialise pseudo media types */
230 ifmedia_init(&sc->sc_media, 0, lagg_media_change,
232 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
233 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
235 if_initname(ifp, ifc->ifc_name, unit);
236 ifp->if_type = IFT_ETHER;
238 ifp->if_start = lagg_start;
239 ifp->if_init = lagg_init;
240 ifp->if_ioctl = lagg_ioctl;
241 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
243 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
244 ifp->if_snd.ifq_drv_maxlen = ifqmaxlen;
245 IFQ_SET_READY(&ifp->if_snd);
248 * Attach as an ordinary ethernet device, childs will be attached
249 * as special device IFT_IEEE8023ADLAG.
251 ether_ifattach(ifp, eaddr);
253 /* Insert into the global list of laggs */
254 mtx_lock(&lagg_list_mtx);
255 SLIST_INSERT_HEAD(&lagg_list, sc, sc_entries);
256 mtx_unlock(&lagg_list_mtx);
262 lagg_clone_destroy(struct ifnet *ifp)
264 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
265 struct lagg_port *lp;
270 ifp->if_flags &= ~IFF_UP;
272 /* Shutdown and remove lagg ports */
273 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
274 lagg_port_destroy(lp, 1);
275 /* Unhook the aggregation protocol */
276 if (sc->sc_detach != NULL)
277 (*sc->sc_detach)(sc);
281 ifmedia_removeall(&sc->sc_media);
283 if_free_type(ifp, IFT_ETHER);
285 mtx_lock(&lagg_list_mtx);
286 SLIST_REMOVE(&lagg_list, sc, lagg_softc, sc_entries);
287 mtx_unlock(&lagg_list_mtx);
289 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
290 LAGG_LOCK_DESTROY(sc);
295 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
297 struct ifnet *ifp = sc->sc_ifp;
299 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
302 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
303 /* Let the protocol know the MAC has changed */
304 if (sc->sc_lladdr != NULL)
305 (*sc->sc_lladdr)(sc);
309 lagg_capabilities(struct lagg_softc *sc)
311 struct lagg_port *lp;
314 LAGG_WLOCK_ASSERT(sc);
316 /* Preserve private capabilities */
317 priv = sc->sc_capabilities & IFCAP_LAGG_MASK;
319 /* Get capabilities from the lagg ports */
320 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
321 cap &= lp->lp_capabilities;
323 if (sc->sc_ifflags & IFF_DEBUG) {
324 printf("%s: capabilities 0x%08x\n",
325 sc->sc_ifname, cap == ~0 ? priv : (cap | priv));
328 return (cap == ~0 ? priv : (cap | priv));
332 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
334 struct lagg_softc *sc = lp->lp_lagg;
335 struct ifnet *ifp = lp->lp_ifp;
336 struct lagg_llq *llq;
339 LAGG_WLOCK_ASSERT(sc);
341 if (lp->lp_detaching ||
342 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
345 /* Check to make sure its not already queued to be changed */
346 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
347 if (llq->llq_ifp == ifp) {
354 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
355 if (llq == NULL) /* XXX what to do */
359 /* Update the lladdr even if pending, it may have changed */
361 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
364 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
366 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
370 * Set the interface MAC address from a taskqueue to avoid a LOR.
373 lagg_port_setlladdr(void *arg, int pending)
375 struct lagg_softc *sc = (struct lagg_softc *)arg;
376 struct lagg_llq *llq, *head;
380 /* Grab a local reference of the queue and remove it from the softc */
382 head = SLIST_FIRST(&sc->sc_llq_head);
383 SLIST_FIRST(&sc->sc_llq_head) = NULL;
387 * Traverse the queue and set the lladdr on each ifp. It is safe to do
388 * unlocked as we have the only reference to it.
390 for (llq = head; llq != NULL; llq = head) {
393 /* Set the link layer address */
394 error = if_setlladdr(ifp, llq->llq_lladdr, ETHER_ADDR_LEN);
396 printf("%s: setlladdr failed on %s\n", __func__,
399 head = SLIST_NEXT(llq, llq_entries);
405 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
407 struct lagg_softc *sc_ptr;
408 struct lagg_port *lp;
411 LAGG_WLOCK_ASSERT(sc);
413 /* Limit the maximal number of lagg ports */
414 if (sc->sc_count >= LAGG_MAX_PORTS)
417 /* New lagg port has to be in an idle state */
418 if (ifp->if_drv_flags & IFF_DRV_OACTIVE)
421 /* Check if port has already been associated to a lagg */
422 if (ifp->if_lagg != NULL)
425 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */
426 if (ifp->if_type != IFT_ETHER)
427 return (EPROTONOSUPPORT);
429 if ((lp = malloc(sizeof(struct lagg_port),
430 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
433 /* Check if port is a stacked lagg */
434 mtx_lock(&lagg_list_mtx);
435 SLIST_FOREACH(sc_ptr, &lagg_list, sc_entries) {
436 if (ifp == sc_ptr->sc_ifp) {
437 mtx_unlock(&lagg_list_mtx);
440 /* XXX disable stacking for the moment, its untested
441 lp->lp_flags |= LAGG_PORT_STACK;
442 if (lagg_port_checkstacking(sc_ptr) >=
444 mtx_unlock(&lagg_list_mtx);
451 mtx_unlock(&lagg_list_mtx);
453 /* Change the interface type */
454 lp->lp_iftype = ifp->if_type;
455 ifp->if_type = IFT_IEEE8023ADLAG;
457 lp->lp_ioctl = ifp->if_ioctl;
458 ifp->if_ioctl = lagg_port_ioctl;
459 lp->lp_output = ifp->if_output;
460 ifp->if_output = lagg_port_output;
465 /* Save port link layer address */
466 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
468 if (SLIST_EMPTY(&sc->sc_ports)) {
470 lagg_lladdr(sc, IF_LLADDR(ifp));
472 /* Update link layer address for this port */
473 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
476 /* Insert into the list of ports */
477 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
480 /* Update lagg capabilities */
481 sc->sc_capabilities = lagg_capabilities(sc);
483 /* Add multicast addresses and interface flags to this port */
484 lagg_ether_cmdmulti(lp, 1);
485 lagg_setflags(lp, 1);
487 if (sc->sc_port_create != NULL)
488 error = (*sc->sc_port_create)(lp);
490 /* remove the port again, without calling sc_port_destroy */
491 lagg_port_destroy(lp, 0);
499 lagg_port_checkstacking(struct lagg_softc *sc)
501 struct lagg_softc *sc_ptr;
502 struct lagg_port *lp;
505 LAGG_WLOCK_ASSERT(sc);
507 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
508 if (lp->lp_flags & LAGG_PORT_STACK) {
509 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
510 m = MAX(m, lagg_port_checkstacking(sc_ptr));
518 lagg_port_destroy(struct lagg_port *lp, int runpd)
520 struct lagg_softc *sc = lp->lp_lagg;
521 struct lagg_port *lp_ptr;
522 struct lagg_llq *llq;
523 struct ifnet *ifp = lp->lp_ifp;
525 LAGG_WLOCK_ASSERT(sc);
527 if (runpd && sc->sc_port_destroy != NULL)
528 (*sc->sc_port_destroy)(lp);
531 * Remove multicast addresses and interface flags from this port and
532 * reset the MAC address, skip if the interface is being detached.
534 if (!lp->lp_detaching) {
535 lagg_ether_cmdmulti(lp, 0);
536 lagg_setflags(lp, 0);
537 lagg_port_lladdr(lp, lp->lp_lladdr);
540 /* Restore interface */
541 ifp->if_type = lp->lp_iftype;
542 ifp->if_ioctl = lp->lp_ioctl;
543 ifp->if_output = lp->lp_output;
546 /* Finally, remove the port from the lagg */
547 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
550 /* Update the primary interface */
551 if (lp == sc->sc_primary) {
552 uint8_t lladdr[ETHER_ADDR_LEN];
554 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
555 bzero(&lladdr, ETHER_ADDR_LEN);
557 bcopy(lp_ptr->lp_lladdr,
558 lladdr, ETHER_ADDR_LEN);
560 lagg_lladdr(sc, lladdr);
561 sc->sc_primary = lp_ptr;
563 /* Update link layer address for each port */
564 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
565 lagg_port_lladdr(lp_ptr, lladdr);
568 /* Remove any pending lladdr changes from the queue */
569 if (lp->lp_detaching) {
570 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
571 if (llq->llq_ifp == ifp) {
572 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
575 break; /* Only appears once */
581 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
585 /* Update lagg capabilities */
586 sc->sc_capabilities = lagg_capabilities(sc);
592 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
594 struct lagg_reqport *rp = (struct lagg_reqport *)data;
595 struct lagg_softc *sc;
596 struct lagg_port *lp = NULL;
599 /* Should be checked by the caller */
600 if (ifp->if_type != IFT_IEEE8023ADLAG ||
601 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_lagg) == NULL)
607 if (rp->rp_portname[0] == '\0' ||
608 ifunit(rp->rp_portname) != ifp) {
613 if (lp->lp_lagg != sc) {
618 lagg_port2req(lp, rp);
628 if (lp->lp_ioctl != NULL)
629 return ((*lp->lp_ioctl)(ifp, cmd, data));
635 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
636 struct sockaddr *dst, struct rtentry *rt0)
638 struct lagg_port *lp = ifp->if_lagg;
639 struct ether_header *eh;
642 switch (dst->sa_family) {
643 case pseudo_AF_HDRCMPLT:
645 eh = (struct ether_header *)dst->sa_data;
646 type = eh->ether_type;
651 * Only allow ethernet types required to initiate or maintain the link,
652 * aggregated frames take a different path.
654 switch (ntohs(type)) {
655 case ETHERTYPE_PAE: /* EAPOL PAE/802.1x */
656 return ((*lp->lp_output)(ifp, m, dst, rt0));
659 /* drop any other frames */
665 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
667 struct lagg_port *lp;
668 struct lagg_softc *sc;
670 if ((lp = ifp->if_lagg) == NULL)
676 lp->lp_detaching = 1;
677 lagg_port_destroy(lp, 1);
682 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
684 struct lagg_softc *sc = lp->lp_lagg;
685 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
686 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
687 rp->rp_prio = lp->lp_prio;
688 rp->rp_flags = lp->lp_flags;
690 /* Add protocol specific flags */
691 switch (sc->sc_proto) {
692 case LAGG_PROTO_FAILOVER:
693 if (lp == sc->sc_primary)
694 rp->rp_flags |= LAGG_PORT_MASTER;
696 case LAGG_PROTO_ROUNDROBIN:
697 case LAGG_PROTO_LOADBALANCE:
698 case LAGG_PROTO_ETHERCHANNEL:
699 if (LAGG_PORTACTIVE(lp))
700 rp->rp_flags |= LAGG_PORT_ACTIVE;
703 case LAGG_PROTO_LACP:
704 /* LACP has a different definition of active */
705 if (lacp_port_isactive(lp))
706 rp->rp_flags |= LAGG_PORT_ACTIVE;
715 struct lagg_softc *sc = (struct lagg_softc *)xsc;
716 struct lagg_port *lp;
717 struct ifnet *ifp = sc->sc_ifp;
719 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
724 ifp->if_drv_flags |= IFF_DRV_RUNNING;
725 /* Update the port lladdrs */
726 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
727 lagg_port_lladdr(lp, IF_LLADDR(ifp));
729 if (sc->sc_init != NULL)
736 lagg_stop(struct lagg_softc *sc)
738 struct ifnet *ifp = sc->sc_ifp;
740 LAGG_WLOCK_ASSERT(sc);
742 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
745 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
747 if (sc->sc_stop != NULL)
752 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
754 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
755 struct lagg_reqall *ra = (struct lagg_reqall *)data;
756 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
757 struct ifreq *ifr = (struct ifreq *)data;
758 struct lagg_port *lp;
760 struct thread *td = curthread;
761 int i, error = 0, unlock = 1;
765 bzero(&rpbuf, sizeof(rpbuf));
769 ra->ra_proto = sc->sc_proto;
770 ra->ra_ports = i = 0;
771 lp = SLIST_FIRST(&sc->sc_ports);
772 while (lp && ra->ra_size >=
773 i + sizeof(struct lagg_reqport)) {
774 lagg_port2req(lp, &rpbuf);
775 error = copyout(&rpbuf, (caddr_t)ra->ra_port + i,
776 sizeof(struct lagg_reqport));
779 i += sizeof(struct lagg_reqport);
781 lp = SLIST_NEXT(lp, lp_entries);
785 error = priv_check(td, PRIV_NET_LAGG);
788 if (ra->ra_proto >= LAGG_PROTO_MAX) {
789 error = EPROTONOSUPPORT;
792 if (sc->sc_proto != LAGG_PROTO_NONE) {
793 error = sc->sc_detach(sc);
794 /* Reset protocol and pointers */
795 sc->sc_proto = LAGG_PROTO_NONE;
796 sc->sc_detach = NULL;
799 sc->sc_port_create = NULL;
800 sc->sc_port_destroy = NULL;
801 sc->sc_linkstate = NULL;
804 sc->sc_lladdr = NULL;
808 for (i = 0; i < (sizeof(lagg_protos) /
809 sizeof(lagg_protos[0])); i++) {
810 if (lagg_protos[i].ti_proto == ra->ra_proto) {
811 if (sc->sc_ifflags & IFF_DEBUG)
812 printf("%s: using proto %u\n",
814 lagg_protos[i].ti_proto);
815 sc->sc_proto = lagg_protos[i].ti_proto;
816 if (sc->sc_proto != LAGG_PROTO_NONE)
817 error = lagg_protos[i].ti_attach(sc);
821 error = EPROTONOSUPPORT;
824 if (rp->rp_portname[0] == '\0' ||
825 (tpif = ifunit(rp->rp_portname)) == NULL) {
830 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
836 lagg_port2req(lp, rp);
839 error = priv_check(td, PRIV_NET_LAGG);
842 if (rp->rp_portname[0] == '\0' ||
843 (tpif = ifunit(rp->rp_portname)) == NULL) {
847 error = lagg_port_create(sc, tpif);
849 case SIOCSLAGGDELPORT:
850 error = priv_check(td, PRIV_NET_LAGG);
853 if (rp->rp_portname[0] == '\0' ||
854 (tpif = ifunit(rp->rp_portname)) == NULL) {
859 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
865 error = lagg_port_destroy(lp, 1);
868 /* Set flags on ports too */
869 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
870 lagg_setflags(lp, 1);
873 if (!(ifp->if_flags & IFF_UP) &&
874 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
876 * If interface is marked down and it is running,
877 * then stop and disable it.
880 } else if ((ifp->if_flags & IFF_UP) &&
881 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
883 * If interface is marked up and it is stopped, then
893 error = lagg_ether_setmulti(sc);
899 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
904 error = ether_ioctl(ifp, cmd, data);
915 lagg_ether_setmulti(struct lagg_softc *sc)
917 struct lagg_port *lp;
919 LAGG_WLOCK_ASSERT(sc);
921 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
922 /* First, remove any existing filter entries. */
923 lagg_ether_cmdmulti(lp, 0);
924 /* copy all addresses from the lagg interface to the port */
925 lagg_ether_cmdmulti(lp, 1);
931 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
933 struct lagg_softc *sc = lp->lp_lagg;
934 struct ifnet *ifp = lp->lp_ifp;
935 struct ifnet *trifp = sc->sc_ifp;
937 struct ifmultiaddr *ifma, *rifma = NULL;
938 struct sockaddr_dl sdl;
941 LAGG_WLOCK_ASSERT(sc);
943 bzero((char *)&sdl, sizeof(sdl));
944 sdl.sdl_len = sizeof(sdl);
945 sdl.sdl_family = AF_LINK;
946 sdl.sdl_type = IFT_ETHER;
947 sdl.sdl_alen = ETHER_ADDR_LEN;
948 sdl.sdl_index = ifp->if_index;
951 TAILQ_FOREACH(ifma, &trifp->if_multiaddrs, ifma_link) {
952 if (ifma->ifma_addr->sa_family != AF_LINK)
954 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
955 LLADDR(&sdl), ETHER_ADDR_LEN);
957 error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
960 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
964 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
967 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
968 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
969 if_delmulti_ifma(mc->mc_ifma);
976 /* Handle a ref counted flag that should be set on the lagg port as well */
978 lagg_setflag(struct lagg_port *lp, int flag, int status,
979 int (*func)(struct ifnet *, int))
981 struct lagg_softc *sc = lp->lp_lagg;
982 struct ifnet *trifp = sc->sc_ifp;
983 struct ifnet *ifp = lp->lp_ifp;
986 LAGG_WLOCK_ASSERT(sc);
988 status = status ? (trifp->if_flags & flag) : 0;
989 /* Now "status" contains the flag value or 0 */
992 * See if recorded ports status is different from what
993 * we want it to be. If it is, flip it. We record ports
994 * status in lp_ifflags so that we won't clear ports flag
995 * we haven't set. In fact, we don't clear or set ports
996 * flags directly, but get or release references to them.
997 * That's why we can be sure that recorded flags still are
998 * in accord with actual ports flags.
1000 if (status != (lp->lp_ifflags & flag)) {
1001 error = (*func)(ifp, status);
1004 lp->lp_ifflags &= ~flag;
1005 lp->lp_ifflags |= status;
1011 * Handle IFF_* flags that require certain changes on the lagg port
1012 * if "status" is true, update ports flags respective to the lagg
1013 * if "status" is false, forcedly clear the flags set on port.
1016 lagg_setflags(struct lagg_port *lp, int status)
1020 for (i = 0; lagg_pflags[i].flag; i++) {
1021 error = lagg_setflag(lp, lagg_pflags[i].flag,
1022 status, lagg_pflags[i].func);
1030 lagg_start(struct ifnet *ifp)
1032 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1037 for (;; error = 0) {
1038 IFQ_DEQUEUE(&ifp->if_snd, m);
1044 if (sc->sc_proto != LAGG_PROTO_NONE)
1045 error = (*sc->sc_start)(sc, m);
1052 m_freem(m); /* sc_start failed */
1061 static struct mbuf *
1062 lagg_input(struct ifnet *ifp, struct mbuf *m)
1064 struct lagg_port *lp = ifp->if_lagg;
1065 struct lagg_softc *sc = lp->lp_lagg;
1066 struct ifnet *trifp = sc->sc_ifp;
1068 if ((trifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1069 (lp->lp_flags & LAGG_PORT_DISABLED) ||
1070 sc->sc_proto == LAGG_PROTO_NONE) {
1078 m = (*sc->sc_input)(sc, lp, m);
1082 ifp->if_ibytes += m->m_pkthdr.len;
1083 trifp->if_ipackets++;
1084 trifp->if_ibytes += m->m_pkthdr.len;
1092 lagg_media_change(struct ifnet *ifp)
1094 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1096 if (sc->sc_ifflags & IFF_DEBUG)
1097 printf("%s\n", __func__);
1104 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1106 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1107 struct lagg_port *lp;
1109 imr->ifm_status = IFM_AVALID;
1110 imr->ifm_active = IFM_ETHER | IFM_AUTO;
1113 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1114 if (LAGG_PORTACTIVE(lp))
1115 imr->ifm_status |= IFM_ACTIVE;
1121 lagg_port_state(struct ifnet *ifp, int state)
1123 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1124 struct lagg_softc *sc = NULL;
1132 if (sc->sc_linkstate != NULL)
1133 (*sc->sc_linkstate)(lp);
1138 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1140 struct lagg_port *lp_next, *rval = NULL;
1141 // int new_link = LINK_STATE_DOWN;
1143 LAGG_RLOCK_ASSERT(sc);
1145 * Search a port which reports an active link state.
1150 if (LAGG_PORTACTIVE(lp)) {
1154 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1155 LAGG_PORTACTIVE(lp_next)) {
1161 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1162 if (LAGG_PORTACTIVE(lp_next)) {
1171 * The IEEE 802.1D standard assumes that a lagg with
1172 * multiple ports is always full duplex. This is valid
1173 * for load sharing laggs and if at least two links
1174 * are active. Unfortunately, checking the latter would
1175 * be too expensive at this point.
1177 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1179 new_link = LINK_STATE_FULL_DUPLEX;
1181 new_link = rval->lp_link_state;
1189 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1191 if (m->m_pkthdr.len < (off + len)) {
1193 } else if (m->m_len < (off + len)) {
1194 m_copydata(m, off, len, buf);
1197 return (mtod(m, char *) + off);
1201 lagg_hashmbuf(struct mbuf *m, uint32_t key)
1206 struct ether_header *eh;
1207 struct ether_vlan_header vlanbuf;
1208 const struct ether_vlan_header *vlan;
1210 const struct ip *ip;
1214 const struct ip6_hdr *ip6;
1215 struct ip6_hdr ip6buf;
1222 eh = mtod(m, struct ether_header *);
1223 etype = ntohs(eh->ether_type);
1224 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, key);
1225 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1227 /* Special handling for encapsulating VLAN frames */
1228 if (m->m_flags & M_VLANTAG) {
1229 p = hash32_buf(&m->m_pkthdr.ether_vtag,
1230 sizeof(m->m_pkthdr.ether_vtag), p);
1231 } else if (etype == ETHERTYPE_VLAN) {
1232 vlan = lagg_gethdr(m, off, sizeof(*vlan), &vlanbuf);
1236 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1237 etype = ntohs(vlan->evl_proto);
1238 off += sizeof(*vlan) - sizeof(*eh);
1244 ip = lagg_gethdr(m, off, sizeof(*ip), &ipbuf);
1248 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1249 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1253 case ETHERTYPE_IPV6:
1254 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &ip6buf);
1258 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1259 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1260 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1261 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */
1270 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1275 IFQ_ENQUEUE(&ifp->if_snd, m, error);
1278 if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) == 0)
1279 (*ifp->if_start)(ifp);
1281 ifp->if_obytes += m->m_pkthdr.len;
1282 if (m->m_flags & M_MCAST)
1289 * Simple round robin aggregation
1293 lagg_rr_attach(struct lagg_softc *sc)
1295 struct lagg_port *lp;
1297 sc->sc_detach = lagg_rr_detach;
1298 sc->sc_start = lagg_rr_start;
1299 sc->sc_input = lagg_rr_input;
1300 sc->sc_port_create = NULL;
1301 sc->sc_port_destroy = lagg_rr_port_destroy;
1302 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1304 lp = SLIST_FIRST(&sc->sc_ports);
1305 sc->sc_psc = (caddr_t)lp;
1311 lagg_rr_detach(struct lagg_softc *sc)
1318 lagg_rr_port_destroy(struct lagg_port *lp)
1320 struct lagg_softc *sc = lp->lp_lagg;
1322 if (lp == (struct lagg_port *)sc->sc_psc)
1327 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1329 struct lagg_port *lp = (struct lagg_port *)sc->sc_psc, *lp_next;
1332 if (lp == NULL && (lp = lagg_link_active(sc, NULL)) == NULL)
1336 error = lagg_enqueue(lp->lp_ifp, m);
1338 /* Get next active port */
1339 lp_next = lagg_link_active(sc, SLIST_NEXT(lp, lp_entries));
1340 sc->sc_psc = (caddr_t)lp_next;
1345 static struct mbuf *
1346 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1348 struct ifnet *ifp = sc->sc_ifp;
1350 /* Just pass in the packet to our lagg device */
1351 m->m_pkthdr.rcvif = ifp;
1361 lagg_fail_attach(struct lagg_softc *sc)
1363 sc->sc_detach = lagg_fail_detach;
1364 sc->sc_start = lagg_fail_start;
1365 sc->sc_input = lagg_fail_input;
1366 sc->sc_port_create = NULL;
1367 sc->sc_port_destroy = NULL;
1373 lagg_fail_detach(struct lagg_softc *sc)
1379 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1381 struct lagg_port *lp;
1383 /* Use the master port if active or the next available port */
1384 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL)
1388 return (lagg_enqueue(lp->lp_ifp, m));
1391 static struct mbuf *
1392 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1394 struct ifnet *ifp = sc->sc_ifp;
1395 struct lagg_port *tmp_tp;
1397 if (lp == sc->sc_primary) {
1398 m->m_pkthdr.rcvif = ifp;
1402 if (sc->sc_primary->lp_link_state == LINK_STATE_DOWN) {
1403 tmp_tp = lagg_link_active(sc, NULL);
1405 * If tmp_tp is null, we've recieved a packet when all
1406 * our links are down. Weird, but process it anyways.
1408 if ((tmp_tp == NULL || tmp_tp == lp)) {
1409 m->m_pkthdr.rcvif = ifp;
1423 lagg_lb_attach(struct lagg_softc *sc)
1425 struct lagg_port *lp;
1428 if ((lb = (struct lagg_lb *)malloc(sizeof(struct lagg_lb),
1429 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
1432 sc->sc_detach = lagg_lb_detach;
1433 sc->sc_start = lagg_lb_start;
1434 sc->sc_input = lagg_lb_input;
1435 sc->sc_port_create = lagg_lb_port_create;
1436 sc->sc_port_destroy = lagg_lb_port_destroy;
1437 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1439 lb->lb_key = arc4random();
1440 sc->sc_psc = (caddr_t)lb;
1442 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1443 lagg_lb_port_create(lp);
1449 lagg_lb_detach(struct lagg_softc *sc)
1451 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1458 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1460 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1461 struct lagg_port *lp_next;
1464 bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1465 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1468 if (i >= LAGG_MAX_PORTS)
1470 if (sc->sc_ifflags & IFF_DEBUG)
1471 printf("%s: port %s at index %d\n",
1472 sc->sc_ifname, lp_next->lp_ifname, i);
1473 lb->lb_ports[i++] = lp_next;
1480 lagg_lb_port_create(struct lagg_port *lp)
1482 struct lagg_softc *sc = lp->lp_lagg;
1483 return (lagg_lb_porttable(sc, NULL));
1487 lagg_lb_port_destroy(struct lagg_port *lp)
1489 struct lagg_softc *sc = lp->lp_lagg;
1490 lagg_lb_porttable(sc, lp);
1494 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1496 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1497 struct lagg_port *lp = NULL;
1501 p = lagg_hashmbuf(m, lb->lb_key);
1502 if ((idx = p % sc->sc_count) >= LAGG_MAX_PORTS)
1504 lp = lb->lb_ports[idx];
1507 * Check the port's link state. This will return the next active
1508 * port if the link is down or the port is NULL.
1510 if ((lp = lagg_link_active(sc, lp)) == NULL)
1514 return (lagg_enqueue(lp->lp_ifp, m));
1517 static struct mbuf *
1518 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1520 struct ifnet *ifp = sc->sc_ifp;
1522 /* Just pass in the packet to our lagg device */
1523 m->m_pkthdr.rcvif = ifp;
1533 lagg_lacp_attach(struct lagg_softc *sc)
1535 struct lagg_port *lp;
1538 sc->sc_detach = lagg_lacp_detach;
1539 sc->sc_port_create = lacp_port_create;
1540 sc->sc_port_destroy = lacp_port_destroy;
1541 sc->sc_linkstate = lacp_linkstate;
1542 sc->sc_start = lagg_lacp_start;
1543 sc->sc_input = lagg_lacp_input;
1544 sc->sc_init = lacp_init;
1545 sc->sc_stop = lacp_stop;
1546 sc->sc_lladdr = lagg_lacp_lladdr;
1548 error = lacp_attach(sc);
1552 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1553 lacp_port_create(lp);
1559 lagg_lacp_detach(struct lagg_softc *sc)
1561 struct lagg_port *lp;
1564 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1565 lacp_port_destroy(lp);
1567 /* unlocking is safe here */
1569 error = lacp_detach(sc);
1576 lagg_lacp_lladdr(struct lagg_softc *sc)
1578 struct lagg_port *lp;
1580 /* purge all the lacp ports */
1581 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1582 lacp_port_destroy(lp);
1584 /* add them back in */
1585 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1586 lacp_port_create(lp);
1590 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
1592 struct lagg_port *lp;
1594 lp = lacp_select_tx_port(sc, m);
1599 return (lagg_enqueue(lp->lp_ifp, m));
1602 static struct mbuf *
1603 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1605 struct ifnet *ifp = sc->sc_ifp;
1606 struct ether_header *eh;
1609 eh = mtod(m, struct ether_header *);
1610 etype = ntohs(eh->ether_type);
1612 /* Tap off LACP control messages */
1613 if (etype == ETHERTYPE_SLOW) {
1619 * If the port is not collecting or not in the active aggregator then
1622 if ((lp->lp_flags & LAGG_PORT_COLLECTING) == 0 ||
1623 lacp_port_isactive(lp) == 0) {
1628 m->m_pkthdr.rcvif = ifp;