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>
5 * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org>
7 * Permission to use, copy, modify, and distribute this software for any
8 * purpose with or without fee is hereby granted, provided that the above
9 * copyright notice and this permission notice appear in all copies.
11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
20 #include <sys/cdefs.h>
21 __FBSDID("$FreeBSD$");
24 #include "opt_inet6.h"
26 #include <sys/param.h>
27 #include <sys/kernel.h>
28 #include <sys/malloc.h>
30 #include <sys/queue.h>
31 #include <sys/socket.h>
32 #include <sys/sockio.h>
33 #include <sys/sysctl.h>
34 #include <sys/module.h>
36 #include <sys/systm.h>
40 #include <sys/rmlock.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>
56 #if defined(INET) || defined(INET6)
57 #include <netinet/in.h>
58 #include <netinet/ip.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/if_ether.h>
66 #include <netinet/ip6.h>
67 #include <netinet6/in6_var.h>
68 #include <netinet6/in6_ifattach.h>
71 #include <net/if_vlan_var.h>
72 #include <net/if_lagg.h>
73 #include <net/ieee8023ad_lacp.h>
75 /* Special flags we should propagate to the lagg ports. */
78 int (*func)(struct ifnet *, int);
80 {IFF_PROMISC, ifpromisc},
81 {IFF_ALLMULTI, if_allmulti},
85 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
86 #define V_lagg_list VNET(lagg_list)
87 static VNET_DEFINE(struct mtx, lagg_list_mtx);
88 #define V_lagg_list_mtx VNET(lagg_list_mtx)
89 #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \
90 "if_lagg list", NULL, MTX_DEF)
91 #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx)
92 #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx)
93 #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx)
94 eventhandler_tag lagg_detach_cookie = NULL;
96 static int lagg_clone_create(struct if_clone *, int, caddr_t);
97 static void lagg_clone_destroy(struct ifnet *);
98 static VNET_DEFINE(struct if_clone *, lagg_cloner);
99 #define V_lagg_cloner VNET(lagg_cloner)
100 static const char laggname[] = "lagg";
102 static void lagg_lladdr(struct lagg_softc *, uint8_t *);
103 static void lagg_capabilities(struct lagg_softc *);
104 static void lagg_port_lladdr(struct lagg_port *, uint8_t *);
105 static void lagg_port_setlladdr(void *, int);
106 static int lagg_port_create(struct lagg_softc *, struct ifnet *);
107 static int lagg_port_destroy(struct lagg_port *, int);
108 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
109 static void lagg_linkstate(struct lagg_softc *);
110 static void lagg_port_state(struct ifnet *, int);
111 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
112 static int lagg_port_output(struct ifnet *, struct mbuf *,
113 const struct sockaddr *, struct route *);
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);
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 int lagg_transmit(struct ifnet *, struct mbuf *);
128 static void lagg_qflush(struct ifnet *);
129 static int lagg_media_change(struct ifnet *);
130 static void lagg_media_status(struct ifnet *, struct ifmediareq *);
131 static struct lagg_port *lagg_link_active(struct lagg_softc *,
133 static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
135 /* Simple round robin */
136 static void lagg_rr_attach(struct lagg_softc *);
137 static int lagg_rr_detach(struct lagg_softc *);
138 static int lagg_rr_start(struct lagg_softc *, struct mbuf *);
139 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
142 /* Active failover */
143 static void lagg_fail_attach(struct lagg_softc *);
144 static int lagg_fail_detach(struct lagg_softc *);
145 static int lagg_fail_start(struct lagg_softc *, struct mbuf *);
146 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
150 static void lagg_lb_attach(struct lagg_softc *);
151 static int lagg_lb_detach(struct lagg_softc *);
152 static int lagg_lb_port_create(struct lagg_port *);
153 static void lagg_lb_port_destroy(struct lagg_port *);
154 static int lagg_lb_start(struct lagg_softc *, struct mbuf *);
155 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
157 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
160 static void lagg_lacp_attach(struct lagg_softc *);
161 static int lagg_lacp_detach(struct lagg_softc *);
162 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *);
163 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
165 static void lagg_lacp_lladdr(struct lagg_softc *);
167 static void lagg_callout(void *);
169 /* lagg protocol table */
170 static const struct lagg_proto {
172 void (*ti_attach)(struct lagg_softc *);
174 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach },
175 { LAGG_PROTO_FAILOVER, lagg_fail_attach },
176 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach },
177 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach },
178 { LAGG_PROTO_LACP, lagg_lacp_attach },
179 { LAGG_PROTO_NONE, NULL }
182 SYSCTL_DECL(_net_link);
183 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
186 /* Allow input on any failover links */
187 static VNET_DEFINE(int, lagg_failover_rx_all);
188 #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all)
189 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
190 &VNET_NAME(lagg_failover_rx_all), 0,
191 "Accept input from any interface in a failover lagg");
193 /* Default value for using M_FLOWID */
194 static VNET_DEFINE(int, def_use_flowid) = 1;
195 #define V_def_use_flowid VNET(def_use_flowid)
196 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
197 &VNET_NAME(def_use_flowid), 0,
198 "Default setting for using flow id for load sharing");
200 /* Default value for using M_FLOWID */
201 static VNET_DEFINE(int, def_flowid_shift) = 16;
202 #define V_def_flowid_shift VNET(def_flowid_shift)
203 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
204 &VNET_NAME(def_flowid_shift), 0,
205 "Default setting for flowid shift for load sharing");
208 vnet_lagg_init(const void *unused __unused)
211 LAGG_LIST_LOCK_INIT();
212 SLIST_INIT(&V_lagg_list);
213 V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
214 lagg_clone_destroy, 0);
216 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
217 vnet_lagg_init, NULL);
220 vnet_lagg_uninit(const void *unused __unused)
223 if_clone_detach(V_lagg_cloner);
224 LAGG_LIST_LOCK_DESTROY();
226 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
227 vnet_lagg_uninit, NULL);
230 lagg_modevent(module_t mod, int type, void *data)
235 lagg_input_p = lagg_input;
236 lagg_linkstate_p = lagg_port_state;
237 lagg_detach_cookie = EVENTHANDLER_REGISTER(
238 ifnet_departure_event, lagg_port_ifdetach, NULL,
239 EVENTHANDLER_PRI_ANY);
242 EVENTHANDLER_DEREGISTER(ifnet_departure_event,
245 lagg_linkstate_p = NULL;
253 static moduledata_t lagg_mod = {
259 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
260 MODULE_VERSION(if_lagg, 1);
263 * This routine is run via an vlan
267 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
269 struct lagg_softc *sc = ifp->if_softc;
270 struct lagg_port *lp;
271 struct rm_priotracker tracker;
273 if (ifp->if_softc != arg) /* Not our event */
276 LAGG_RLOCK(sc, &tracker);
277 if (!SLIST_EMPTY(&sc->sc_ports)) {
278 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
279 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
281 LAGG_RUNLOCK(sc, &tracker);
285 * This routine is run via an vlan
289 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
291 struct lagg_softc *sc = ifp->if_softc;
292 struct lagg_port *lp;
293 struct rm_priotracker tracker;
295 if (ifp->if_softc != arg) /* Not our event */
298 LAGG_RLOCK(sc, &tracker);
299 if (!SLIST_EMPTY(&sc->sc_ports)) {
300 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
301 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
303 LAGG_RUNLOCK(sc, &tracker);
307 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
309 struct lagg_softc *sc;
311 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */
314 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
315 ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
321 sc->sc_ipackets = counter_u64_alloc(M_WAITOK);
322 sc->sc_opackets = counter_u64_alloc(M_WAITOK);
323 sc->sc_ibytes = counter_u64_alloc(M_WAITOK);
324 sc->sc_obytes = counter_u64_alloc(M_WAITOK);
326 if (V_def_use_flowid)
327 sc->sc_opts |= LAGG_OPT_USE_FLOWID;
328 sc->flowid_shift = V_def_flowid_shift;
330 /* Hash all layers by default */
331 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4;
333 sc->sc_proto = LAGG_PROTO_NONE;
334 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) {
335 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) {
336 sc->sc_proto = lagg_protos[i].ti_proto;
337 lagg_protos[i].ti_attach(sc);
342 LAGG_CALLOUT_LOCK_INIT(sc);
343 SLIST_INIT(&sc->sc_ports);
344 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc);
347 * This uses the callout lock rather than the rmlock; one can't
348 * hold said rmlock during SWI.
350 callout_init_mtx(&sc->sc_callout, &sc->sc_call_mtx, 0);
352 /* Initialise pseudo media types */
353 ifmedia_init(&sc->sc_media, 0, lagg_media_change,
355 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
356 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
358 if_initname(ifp, laggname, unit);
360 ifp->if_transmit = lagg_transmit;
361 ifp->if_qflush = lagg_qflush;
362 ifp->if_init = lagg_init;
363 ifp->if_ioctl = lagg_ioctl;
364 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
365 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
368 * Attach as an ordinary ethernet device, children will be attached
369 * as special device IFT_IEEE8023ADLAG.
371 ether_ifattach(ifp, eaddr);
373 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
374 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
375 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
376 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
378 /* Insert into the global list of laggs */
380 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
383 callout_reset(&sc->sc_callout, hz, lagg_callout, sc);
389 lagg_clone_destroy(struct ifnet *ifp)
391 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
392 struct lagg_port *lp;
397 ifp->if_flags &= ~IFF_UP;
399 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
400 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
402 /* Shutdown and remove lagg ports */
403 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL)
404 lagg_port_destroy(lp, 1);
405 /* Unhook the aggregation protocol */
406 if (sc->sc_detach != NULL)
407 (*sc->sc_detach)(sc);
411 ifmedia_removeall(&sc->sc_media);
415 /* This grabs sc_callout_mtx, serialising it correctly */
416 callout_drain(&sc->sc_callout);
418 /* At this point it's drained; we can free this */
419 counter_u64_free(sc->sc_ipackets);
420 counter_u64_free(sc->sc_opackets);
421 counter_u64_free(sc->sc_ibytes);
422 counter_u64_free(sc->sc_obytes);
425 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
428 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task);
429 LAGG_LOCK_DESTROY(sc);
430 LAGG_CALLOUT_LOCK_DESTROY(sc);
435 lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr)
437 struct ifnet *ifp = sc->sc_ifp;
440 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)
443 LAGG_WLOCK_ASSERT(sc);
445 * Set the link layer address on the lagg interface.
446 * sc_lladdr() notifies the MAC change to
447 * the aggregation protocol. iflladdr_event handler which
448 * may trigger gratuitous ARPs for INET will be handled in
451 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN);
452 if (sc->sc_lladdr != NULL)
453 (*sc->sc_lladdr)(sc);
455 bzero(&lp, sizeof(lp));
456 lp.lp_ifp = sc->sc_ifp;
459 lagg_port_lladdr(&lp, lladdr);
463 lagg_capabilities(struct lagg_softc *sc)
465 struct lagg_port *lp;
466 int cap = ~0, ena = ~0;
468 struct ifnet_hw_tsomax hw_tsomax;
470 LAGG_WLOCK_ASSERT(sc);
472 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
474 /* Get capabilities from the lagg ports */
475 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
476 cap &= lp->lp_ifp->if_capabilities;
477 ena &= lp->lp_ifp->if_capenable;
478 hwa &= lp->lp_ifp->if_hwassist;
479 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
481 cap = (cap == ~0 ? 0 : cap);
482 ena = (ena == ~0 ? 0 : ena);
483 hwa = (hwa == ~0 ? 0 : hwa);
485 if (sc->sc_ifp->if_capabilities != cap ||
486 sc->sc_ifp->if_capenable != ena ||
487 sc->sc_ifp->if_hwassist != hwa ||
488 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
489 sc->sc_ifp->if_capabilities = cap;
490 sc->sc_ifp->if_capenable = ena;
491 sc->sc_ifp->if_hwassist = hwa;
492 getmicrotime(&sc->sc_ifp->if_lastchange);
494 if (sc->sc_ifflags & IFF_DEBUG)
495 if_printf(sc->sc_ifp,
496 "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
501 lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr)
503 struct lagg_softc *sc = lp->lp_softc;
504 struct ifnet *ifp = lp->lp_ifp;
505 struct lagg_llq *llq;
509 LAGG_WLOCK_ASSERT(sc);
511 primary = (sc->sc_primary->lp_ifp == ifp) ? 1 : 0;
512 if (primary == 0 && (lp->lp_detaching ||
513 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0))
516 /* Check to make sure its not already queued to be changed */
517 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
518 if (llq->llq_ifp == ifp) {
525 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT);
526 if (llq == NULL) /* XXX what to do */
530 /* Update the lladdr even if pending, it may have changed */
532 llq->llq_primary = primary;
533 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN);
536 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries);
538 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task);
542 * Set the interface MAC address from a taskqueue to avoid a LOR.
545 lagg_port_setlladdr(void *arg, int pending)
547 struct lagg_softc *sc = (struct lagg_softc *)arg;
548 struct lagg_llq *llq, *head;
552 /* Grab a local reference of the queue and remove it from the softc */
554 head = SLIST_FIRST(&sc->sc_llq_head);
555 SLIST_FIRST(&sc->sc_llq_head) = NULL;
559 * Traverse the queue and set the lladdr on each ifp. It is safe to do
560 * unlocked as we have the only reference to it.
562 for (llq = head; llq != NULL; llq = head) {
565 CURVNET_SET(ifp->if_vnet);
566 if (llq->llq_primary == 0) {
568 * Set the link layer address on the laggport interface.
569 * if_setlladdr() triggers gratuitous ARPs for INET.
571 error = if_setlladdr(ifp, llq->llq_lladdr,
574 printf("%s: setlladdr failed on %s\n", __func__,
577 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
579 head = SLIST_NEXT(llq, llq_entries);
585 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
587 struct lagg_softc *sc_ptr;
588 struct lagg_port *lp, *tlp;
591 LAGG_WLOCK_ASSERT(sc);
593 /* Limit the maximal number of lagg ports */
594 if (sc->sc_count >= LAGG_MAX_PORTS)
597 /* Check if port has already been associated to a lagg */
598 if (ifp->if_lagg != NULL) {
599 /* Port is already in the current lagg? */
600 lp = (struct lagg_port *)ifp->if_lagg;
601 if (lp->lp_softc == sc)
606 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */
607 if (ifp->if_type != IFT_ETHER)
608 return (EPROTONOSUPPORT);
610 /* Allow the first Ethernet member to define the MTU */
611 if (SLIST_EMPTY(&sc->sc_ports))
612 sc->sc_ifp->if_mtu = ifp->if_mtu;
613 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
614 if_printf(sc->sc_ifp, "invalid MTU for %s\n",
619 if ((lp = malloc(sizeof(struct lagg_port),
620 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL)
623 /* Check if port is a stacked lagg */
625 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
626 if (ifp == sc_ptr->sc_ifp) {
630 /* XXX disable stacking for the moment, its untested */
631 #ifdef LAGG_PORT_STACKING
632 lp->lp_flags |= LAGG_PORT_STACK;
633 if (lagg_port_checkstacking(sc_ptr) >=
644 /* Change the interface type */
645 lp->lp_iftype = ifp->if_type;
646 ifp->if_type = IFT_IEEE8023ADLAG;
648 lp->lp_ioctl = ifp->if_ioctl;
649 ifp->if_ioctl = lagg_port_ioctl;
650 lp->lp_output = ifp->if_output;
651 ifp->if_output = lagg_port_output;
656 /* Save port link layer address */
657 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
659 if (SLIST_EMPTY(&sc->sc_ports)) {
661 lagg_lladdr(sc, IF_LLADDR(ifp));
663 /* Update link layer address for this port */
664 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp));
668 * Insert into the list of ports.
669 * Keep ports sorted by if_index. It is handy, when configuration
670 * is predictable and `ifconfig laggN create ...` command
671 * will lead to the same result each time.
673 SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
674 if (tlp->lp_ifp->if_index < ifp->if_index && (
675 SLIST_NEXT(tlp, lp_entries) == NULL ||
676 SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index >
681 SLIST_INSERT_AFTER(tlp, lp, lp_entries);
683 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
686 /* Update lagg capabilities */
687 lagg_capabilities(sc);
690 /* Add multicast addresses and interface flags to this port */
691 lagg_ether_cmdmulti(lp, 1);
692 lagg_setflags(lp, 1);
694 if (sc->sc_port_create != NULL)
695 error = (*sc->sc_port_create)(lp);
697 /* remove the port again, without calling sc_port_destroy */
698 lagg_port_destroy(lp, 0);
705 #ifdef LAGG_PORT_STACKING
707 lagg_port_checkstacking(struct lagg_softc *sc)
709 struct lagg_softc *sc_ptr;
710 struct lagg_port *lp;
713 LAGG_WLOCK_ASSERT(sc);
715 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
716 if (lp->lp_flags & LAGG_PORT_STACK) {
717 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
718 m = MAX(m, lagg_port_checkstacking(sc_ptr));
727 lagg_port_destroy(struct lagg_port *lp, int runpd)
729 struct lagg_softc *sc = lp->lp_softc;
730 struct lagg_port *lp_ptr;
731 struct lagg_llq *llq;
732 struct ifnet *ifp = lp->lp_ifp;
734 LAGG_WLOCK_ASSERT(sc);
736 if (runpd && sc->sc_port_destroy != NULL)
737 (*sc->sc_port_destroy)(lp);
740 * Remove multicast addresses and interface flags from this port and
741 * reset the MAC address, skip if the interface is being detached.
743 if (!lp->lp_detaching) {
744 lagg_ether_cmdmulti(lp, 0);
745 lagg_setflags(lp, 0);
746 lagg_port_lladdr(lp, lp->lp_lladdr);
749 /* Restore interface */
750 ifp->if_type = lp->lp_iftype;
751 ifp->if_ioctl = lp->lp_ioctl;
752 ifp->if_output = lp->lp_output;
755 /* Finally, remove the port from the lagg */
756 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
759 /* Update the primary interface */
760 if (lp == sc->sc_primary) {
761 uint8_t lladdr[ETHER_ADDR_LEN];
763 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) {
764 bzero(&lladdr, ETHER_ADDR_LEN);
766 bcopy(lp_ptr->lp_lladdr,
767 lladdr, ETHER_ADDR_LEN);
769 lagg_lladdr(sc, lladdr);
770 sc->sc_primary = lp_ptr;
772 /* Update link layer address for each port */
773 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
774 lagg_port_lladdr(lp_ptr, lladdr);
777 /* Remove any pending lladdr changes from the queue */
778 if (lp->lp_detaching) {
779 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) {
780 if (llq->llq_ifp == ifp) {
781 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq,
784 break; /* Only appears once */
790 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
794 /* Update lagg capabilities */
795 lagg_capabilities(sc);
802 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
804 struct lagg_reqport *rp = (struct lagg_reqport *)data;
805 struct lagg_softc *sc;
806 struct lagg_port *lp = NULL;
808 struct rm_priotracker tracker;
810 /* Should be checked by the caller */
811 if (ifp->if_type != IFT_IEEE8023ADLAG ||
812 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
817 if (rp->rp_portname[0] == '\0' ||
818 ifunit(rp->rp_portname) != ifp) {
823 LAGG_RLOCK(sc, &tracker);
824 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
826 LAGG_RUNLOCK(sc, &tracker);
830 lagg_port2req(lp, rp);
831 LAGG_RUNLOCK(sc, &tracker);
835 if (lp->lp_ioctl == NULL) {
839 error = (*lp->lp_ioctl)(ifp, cmd, data);
843 /* Update lagg interface capabilities */
845 lagg_capabilities(sc);
850 /* Do not allow the MTU to be changed once joined */
861 if (lp->lp_ioctl != NULL)
862 return ((*lp->lp_ioctl)(ifp, cmd, data));
868 * For direct output to child ports.
871 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
872 const struct sockaddr *dst, struct route *ro)
874 struct lagg_port *lp = ifp->if_lagg;
876 switch (dst->sa_family) {
877 case pseudo_AF_HDRCMPLT:
879 return ((*lp->lp_output)(ifp, m, dst, ro));
882 /* drop any other frames */
888 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
890 struct lagg_port *lp;
891 struct lagg_softc *sc;
893 if ((lp = ifp->if_lagg) == NULL)
895 /* If the ifnet is just being renamed, don't do anything. */
896 if (ifp->if_flags & IFF_RENAMING)
902 lp->lp_detaching = 1;
903 lagg_port_destroy(lp, 1);
908 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
910 struct lagg_softc *sc = lp->lp_softc;
912 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
913 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
914 rp->rp_prio = lp->lp_prio;
915 rp->rp_flags = lp->lp_flags;
916 if (sc->sc_portreq != NULL)
917 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc);
919 /* Add protocol specific flags */
920 switch (sc->sc_proto) {
921 case LAGG_PROTO_FAILOVER:
922 if (lp == sc->sc_primary)
923 rp->rp_flags |= LAGG_PORT_MASTER;
924 if (lp == lagg_link_active(sc, sc->sc_primary))
925 rp->rp_flags |= LAGG_PORT_ACTIVE;
928 case LAGG_PROTO_ROUNDROBIN:
929 case LAGG_PROTO_LOADBALANCE:
930 case LAGG_PROTO_ETHERCHANNEL:
931 if (LAGG_PORTACTIVE(lp))
932 rp->rp_flags |= LAGG_PORT_ACTIVE;
935 case LAGG_PROTO_LACP:
936 /* LACP has a different definition of active */
937 if (lacp_isactive(lp))
938 rp->rp_flags |= LAGG_PORT_ACTIVE;
939 if (lacp_iscollecting(lp))
940 rp->rp_flags |= LAGG_PORT_COLLECTING;
941 if (lacp_isdistributing(lp))
942 rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
951 struct lagg_softc *sc = (struct lagg_softc *)xsc;
952 struct lagg_port *lp;
953 struct ifnet *ifp = sc->sc_ifp;
955 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
960 ifp->if_drv_flags |= IFF_DRV_RUNNING;
961 /* Update the port lladdrs */
962 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
963 lagg_port_lladdr(lp, IF_LLADDR(ifp));
965 if (sc->sc_init != NULL)
972 lagg_stop(struct lagg_softc *sc)
974 struct ifnet *ifp = sc->sc_ifp;
976 LAGG_WLOCK_ASSERT(sc);
978 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
981 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
983 if (sc->sc_stop != NULL)
988 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
990 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
991 struct lagg_reqall *ra = (struct lagg_reqall *)data;
992 struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
993 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
994 struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
995 struct ifreq *ifr = (struct ifreq *)data;
996 struct lagg_port *lp;
997 const struct lagg_proto *proto = NULL;
999 struct thread *td = curthread;
1001 int count, buflen, len, error = 0;
1002 struct rm_priotracker tracker;
1004 bzero(&rpbuf, sizeof(rpbuf));
1008 LAGG_RLOCK(sc, &tracker);
1010 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1012 buflen = count * sizeof(struct lagg_reqport);
1013 LAGG_RUNLOCK(sc, &tracker);
1015 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1017 LAGG_RLOCK(sc, &tracker);
1018 ra->ra_proto = sc->sc_proto;
1019 if (sc->sc_req != NULL)
1020 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc);
1024 len = min(ra->ra_size, buflen);
1025 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1026 if (len < sizeof(rpbuf))
1029 lagg_port2req(lp, &rpbuf);
1030 memcpy(buf, &rpbuf, sizeof(rpbuf));
1032 buf += sizeof(rpbuf);
1033 len -= sizeof(rpbuf);
1035 LAGG_RUNLOCK(sc, &tracker);
1036 ra->ra_ports = count;
1037 ra->ra_size = count * sizeof(rpbuf);
1038 error = copyout(outbuf, ra->ra_port, ra->ra_size);
1039 free(outbuf, M_TEMP);
1042 error = priv_check(td, PRIV_NET_LAGG);
1045 for (proto = lagg_protos; proto->ti_proto != LAGG_PROTO_NONE;
1047 if (proto->ti_proto == ra->ra_proto) {
1048 if (sc->sc_ifflags & IFF_DEBUG)
1049 printf("%s: using proto %u\n",
1050 sc->sc_ifname, proto->ti_proto);
1054 if (proto->ti_proto == LAGG_PROTO_NONE) {
1055 error = EPROTONOSUPPORT;
1058 /* Set to LAGG_PROTO_NONE during the attach. */
1060 if (sc->sc_proto != LAGG_PROTO_NONE) {
1061 sc->sc_proto = LAGG_PROTO_NONE;
1062 if (sc->sc_detach != NULL)
1068 proto->ti_attach(sc);
1070 sc->sc_proto = proto->ti_proto;
1074 ro->ro_opts = sc->sc_opts;
1075 if (sc->sc_proto == LAGG_PROTO_LACP) {
1076 struct lacp_softc *lsc;
1078 lsc = (struct lacp_softc *)sc->sc_psc;
1079 if (lsc->lsc_debug.lsc_tx_test != 0)
1080 ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1081 if (lsc->lsc_debug.lsc_rx_test != 0)
1082 ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1083 if (lsc->lsc_strict_mode != 0)
1084 ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1085 if (lsc->lsc_fast_timeout != 0)
1086 ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1088 ro->ro_active = sc->sc_active;
1091 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1092 ro->ro_active += LAGG_PORTACTIVE(lp);
1094 ro->ro_flapping = sc->sc_flapping;
1095 ro->ro_flowid_shift = sc->flowid_shift;
1098 error = priv_check(td, PRIV_NET_LAGG);
1101 if (ro->ro_opts == 0)
1104 * Set options. LACP options are stored in sc->sc_psc,
1109 switch (ro->ro_opts) {
1110 case LAGG_OPT_USE_FLOWID:
1111 case -LAGG_OPT_USE_FLOWID:
1112 case LAGG_OPT_FLOWIDSHIFT:
1116 case LAGG_OPT_LACP_TXTEST:
1117 case -LAGG_OPT_LACP_TXTEST:
1118 case LAGG_OPT_LACP_RXTEST:
1119 case -LAGG_OPT_LACP_RXTEST:
1120 case LAGG_OPT_LACP_STRICT:
1121 case -LAGG_OPT_LACP_STRICT:
1122 case LAGG_OPT_LACP_TIMEOUT:
1123 case -LAGG_OPT_LACP_TIMEOUT:
1133 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1134 /* Invalid combination of options specified. */
1137 break; /* Return from SIOCSLAGGOPTS. */
1140 * Store new options into sc->sc_opts except for
1141 * FLOWIDSHIFT and LACP options.
1144 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1145 sc->flowid_shift = ro->ro_flowid_shift;
1146 else if (ro->ro_opts > 0)
1147 sc->sc_opts |= ro->ro_opts;
1149 sc->sc_opts &= ~ro->ro_opts;
1151 struct lacp_softc *lsc;
1152 struct lacp_port *lp;
1154 lsc = (struct lacp_softc *)sc->sc_psc;
1156 switch (ro->ro_opts) {
1157 case LAGG_OPT_LACP_TXTEST:
1158 lsc->lsc_debug.lsc_tx_test = 1;
1160 case -LAGG_OPT_LACP_TXTEST:
1161 lsc->lsc_debug.lsc_tx_test = 0;
1163 case LAGG_OPT_LACP_RXTEST:
1164 lsc->lsc_debug.lsc_rx_test = 1;
1166 case -LAGG_OPT_LACP_RXTEST:
1167 lsc->lsc_debug.lsc_rx_test = 0;
1169 case LAGG_OPT_LACP_STRICT:
1170 lsc->lsc_strict_mode = 1;
1172 case -LAGG_OPT_LACP_STRICT:
1173 lsc->lsc_strict_mode = 0;
1175 case LAGG_OPT_LACP_TIMEOUT:
1177 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1178 lp->lp_state |= LACP_STATE_TIMEOUT;
1180 lsc->lsc_fast_timeout = 1;
1182 case -LAGG_OPT_LACP_TIMEOUT:
1184 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1185 lp->lp_state &= ~LACP_STATE_TIMEOUT;
1187 lsc->lsc_fast_timeout = 0;
1193 case SIOCGLAGGFLAGS:
1194 rf->rf_flags = sc->sc_flags;
1197 error = priv_check(td, PRIV_NET_LAGG);
1200 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1205 sc->sc_flags &= ~LAGG_F_HASHMASK;
1206 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK;
1210 if (rp->rp_portname[0] == '\0' ||
1211 (tpif = ifunit(rp->rp_portname)) == NULL) {
1216 LAGG_RLOCK(sc, &tracker);
1217 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1218 lp->lp_softc != sc) {
1220 LAGG_RUNLOCK(sc, &tracker);
1224 lagg_port2req(lp, rp);
1225 LAGG_RUNLOCK(sc, &tracker);
1228 error = priv_check(td, PRIV_NET_LAGG);
1231 if (rp->rp_portname[0] == '\0' ||
1232 (tpif = ifunit(rp->rp_portname)) == NULL) {
1238 * A laggport interface should not have inet6 address
1239 * because two interfaces with a valid link-local
1240 * scope zone must not be merged in any form. This
1241 * restriction is needed to prevent violation of
1242 * link-local scope zone. Attempts to add a laggport
1243 * interface which has inet6 addresses triggers
1244 * removal of all inet6 addresses on the member
1247 if (in6ifa_llaonifp(tpif)) {
1249 if_printf(sc->sc_ifp,
1250 "IPv6 addresses on %s have been removed "
1251 "before adding it as a member to prevent "
1252 "IPv6 address scope violation.\n",
1257 error = lagg_port_create(sc, tpif);
1260 case SIOCSLAGGDELPORT:
1261 error = priv_check(td, PRIV_NET_LAGG);
1264 if (rp->rp_portname[0] == '\0' ||
1265 (tpif = ifunit(rp->rp_portname)) == NULL) {
1271 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1272 lp->lp_softc != sc) {
1278 error = lagg_port_destroy(lp, 1);
1282 /* Set flags on ports too */
1284 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1285 lagg_setflags(lp, 1);
1289 if (!(ifp->if_flags & IFF_UP) &&
1290 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1292 * If interface is marked down and it is running,
1293 * then stop and disable it.
1298 } else if ((ifp->if_flags & IFF_UP) &&
1299 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1301 * If interface is marked up and it is stopped, then
1304 (*ifp->if_init)(sc);
1310 error = lagg_ether_setmulti(sc);
1315 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1320 /* Do not allow the MTU or caps to be directly changed */
1325 error = ether_ioctl(ifp, cmd, data);
1332 lagg_ether_setmulti(struct lagg_softc *sc)
1334 struct lagg_port *lp;
1336 LAGG_WLOCK_ASSERT(sc);
1338 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1339 /* First, remove any existing filter entries. */
1340 lagg_ether_cmdmulti(lp, 0);
1341 /* copy all addresses from the lagg interface to the port */
1342 lagg_ether_cmdmulti(lp, 1);
1348 lagg_ether_cmdmulti(struct lagg_port *lp, int set)
1350 struct lagg_softc *sc = lp->lp_softc;
1351 struct ifnet *ifp = lp->lp_ifp;
1352 struct ifnet *scifp = sc->sc_ifp;
1354 struct ifmultiaddr *ifma;
1357 LAGG_WLOCK_ASSERT(sc);
1360 IF_ADDR_WLOCK(scifp);
1361 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1362 if (ifma->ifma_addr->sa_family != AF_LINK)
1364 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT);
1366 IF_ADDR_WUNLOCK(scifp);
1369 bcopy(ifma->ifma_addr, &mc->mc_addr,
1370 ifma->ifma_addr->sa_len);
1371 mc->mc_addr.sdl_index = ifp->if_index;
1373 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1375 IF_ADDR_WUNLOCK(scifp);
1376 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1377 error = if_addmulti(ifp,
1378 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1383 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1384 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1385 if (mc->mc_ifma && !lp->lp_detaching)
1386 if_delmulti_ifma(mc->mc_ifma);
1393 /* Handle a ref counted flag that should be set on the lagg port as well */
1395 lagg_setflag(struct lagg_port *lp, int flag, int status,
1396 int (*func)(struct ifnet *, int))
1398 struct lagg_softc *sc = lp->lp_softc;
1399 struct ifnet *scifp = sc->sc_ifp;
1400 struct ifnet *ifp = lp->lp_ifp;
1403 LAGG_WLOCK_ASSERT(sc);
1405 status = status ? (scifp->if_flags & flag) : 0;
1406 /* Now "status" contains the flag value or 0 */
1409 * See if recorded ports status is different from what
1410 * we want it to be. If it is, flip it. We record ports
1411 * status in lp_ifflags so that we won't clear ports flag
1412 * we haven't set. In fact, we don't clear or set ports
1413 * flags directly, but get or release references to them.
1414 * That's why we can be sure that recorded flags still are
1415 * in accord with actual ports flags.
1417 if (status != (lp->lp_ifflags & flag)) {
1418 error = (*func)(ifp, status);
1421 lp->lp_ifflags &= ~flag;
1422 lp->lp_ifflags |= status;
1428 * Handle IFF_* flags that require certain changes on the lagg port
1429 * if "status" is true, update ports flags respective to the lagg
1430 * if "status" is false, forcedly clear the flags set on port.
1433 lagg_setflags(struct lagg_port *lp, int status)
1437 for (i = 0; lagg_pflags[i].flag; i++) {
1438 error = lagg_setflag(lp, lagg_pflags[i].flag,
1439 status, lagg_pflags[i].func);
1447 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1449 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1450 int error, len, mcast;
1451 struct rm_priotracker tracker;
1453 len = m->m_pkthdr.len;
1454 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0;
1456 LAGG_RLOCK(sc, &tracker);
1457 /* We need a Tx algorithm and at least one port */
1458 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1459 LAGG_RUNLOCK(sc, &tracker);
1465 ETHER_BPF_MTAP(ifp, m);
1467 error = (*sc->sc_start)(sc, m);
1468 LAGG_RUNLOCK(sc, &tracker);
1471 counter_u64_add(sc->sc_opackets, 1);
1472 counter_u64_add(sc->sc_obytes, len);
1473 ifp->if_omcasts += mcast;
1481 * The ifp->if_qflush entry point for lagg(4) is no-op.
1484 lagg_qflush(struct ifnet *ifp __unused)
1488 static struct mbuf *
1489 lagg_input(struct ifnet *ifp, struct mbuf *m)
1491 struct lagg_port *lp = ifp->if_lagg;
1492 struct lagg_softc *sc = lp->lp_softc;
1493 struct ifnet *scifp = sc->sc_ifp;
1494 struct rm_priotracker tracker;
1496 LAGG_RLOCK(sc, &tracker);
1497 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1498 (lp->lp_flags & LAGG_PORT_DISABLED) ||
1499 sc->sc_proto == LAGG_PROTO_NONE) {
1500 LAGG_RUNLOCK(sc, &tracker);
1505 ETHER_BPF_MTAP(scifp, m);
1507 m = (lp->lp_detaching == 0) ? (*sc->sc_input)(sc, lp, m) : NULL;
1510 counter_u64_add(sc->sc_ipackets, 1);
1511 counter_u64_add(sc->sc_ibytes, m->m_pkthdr.len);
1513 if (scifp->if_flags & IFF_MONITOR) {
1519 LAGG_RUNLOCK(sc, &tracker);
1524 lagg_media_change(struct ifnet *ifp)
1526 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1528 if (sc->sc_ifflags & IFF_DEBUG)
1529 printf("%s\n", __func__);
1536 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1538 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1539 struct lagg_port *lp;
1540 struct rm_priotracker tracker;
1542 imr->ifm_status = IFM_AVALID;
1543 imr->ifm_active = IFM_ETHER | IFM_AUTO;
1545 LAGG_RLOCK(sc, &tracker);
1546 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1547 if (LAGG_PORTACTIVE(lp))
1548 imr->ifm_status |= IFM_ACTIVE;
1550 LAGG_RUNLOCK(sc, &tracker);
1554 lagg_linkstate(struct lagg_softc *sc)
1556 struct lagg_port *lp;
1557 int new_link = LINK_STATE_DOWN;
1560 /* Our link is considered up if at least one of our ports is active */
1561 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1562 if (lp->lp_link_state == LINK_STATE_UP) {
1563 new_link = LINK_STATE_UP;
1567 if_link_state_change(sc->sc_ifp, new_link);
1569 /* Update if_baudrate to reflect the max possible speed */
1570 switch (sc->sc_proto) {
1571 case LAGG_PROTO_FAILOVER:
1572 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1573 sc->sc_primary->lp_ifp->if_baudrate : 0;
1575 case LAGG_PROTO_ROUNDROBIN:
1576 case LAGG_PROTO_LOADBALANCE:
1577 case LAGG_PROTO_ETHERCHANNEL:
1579 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1580 speed += lp->lp_ifp->if_baudrate;
1581 sc->sc_ifp->if_baudrate = speed;
1583 case LAGG_PROTO_LACP:
1584 /* LACP updates if_baudrate itself */
1590 lagg_port_state(struct ifnet *ifp, int state)
1592 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1593 struct lagg_softc *sc = NULL;
1602 if (sc->sc_linkstate != NULL)
1603 (*sc->sc_linkstate)(lp);
1608 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
1610 struct lagg_port *lp_next, *rval = NULL;
1611 // int new_link = LINK_STATE_DOWN;
1613 LAGG_RLOCK_ASSERT(sc);
1615 * Search a port which reports an active link state.
1620 if (LAGG_PORTACTIVE(lp)) {
1624 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL &&
1625 LAGG_PORTACTIVE(lp_next)) {
1631 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1632 if (LAGG_PORTACTIVE(lp_next)) {
1641 * The IEEE 802.1D standard assumes that a lagg with
1642 * multiple ports is always full duplex. This is valid
1643 * for load sharing laggs and if at least two links
1644 * are active. Unfortunately, checking the latter would
1645 * be too expensive at this point.
1647 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) &&
1649 new_link = LINK_STATE_FULL_DUPLEX;
1651 new_link = rval->lp_link_state;
1659 lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
1661 if (m->m_pkthdr.len < (off + len)) {
1663 } else if (m->m_len < (off + len)) {
1664 m_copydata(m, off, len, buf);
1667 return (mtod(m, char *) + off);
1671 lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
1676 struct ether_header *eh;
1677 const struct ether_vlan_header *vlan;
1679 const struct ip *ip;
1680 const uint32_t *ports;
1684 const struct ip6_hdr *ip6;
1694 struct ether_vlan_header vlan;
1702 eh = mtod(m, struct ether_header *);
1703 etype = ntohs(eh->ether_type);
1704 if (sc->sc_flags & LAGG_F_HASHL2) {
1705 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p);
1706 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p);
1709 /* Special handling for encapsulating VLAN frames */
1710 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) {
1711 p = hash32_buf(&m->m_pkthdr.ether_vtag,
1712 sizeof(m->m_pkthdr.ether_vtag), p);
1713 } else if (etype == ETHERTYPE_VLAN) {
1714 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf);
1718 if (sc->sc_flags & LAGG_F_HASHL2)
1719 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p);
1720 etype = ntohs(vlan->evl_proto);
1721 off += sizeof(*vlan) - sizeof(*eh);
1727 ip = lagg_gethdr(m, off, sizeof(*ip), &buf);
1731 if (sc->sc_flags & LAGG_F_HASHL3) {
1732 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p);
1733 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p);
1735 if (!(sc->sc_flags & LAGG_F_HASHL4))
1741 iphlen = ip->ip_hl << 2;
1742 if (iphlen < sizeof(*ip))
1745 ports = lagg_gethdr(m, off, sizeof(*ports), &buf);
1748 p = hash32_buf(ports, sizeof(*ports), p);
1754 case ETHERTYPE_IPV6:
1755 if (!(sc->sc_flags & LAGG_F_HASHL3))
1757 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf);
1761 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p);
1762 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p);
1763 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK;
1764 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */
1773 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
1776 return (ifp->if_transmit)(ifp, m);
1780 * Simple round robin aggregation
1783 lagg_rr_attach(struct lagg_softc *sc)
1785 sc->sc_detach = lagg_rr_detach;
1786 sc->sc_start = lagg_rr_start;
1787 sc->sc_input = lagg_rr_input;
1788 sc->sc_detach = NULL;
1789 sc->sc_port_create = NULL;
1790 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1795 lagg_rr_detach(struct lagg_softc *sc)
1801 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
1803 struct lagg_port *lp;
1806 p = atomic_fetchadd_32(&sc->sc_seq, 1);
1808 lp = SLIST_FIRST(&sc->sc_ports);
1810 lp = SLIST_NEXT(lp, lp_entries);
1813 * Check the port's link state. This will return the next active
1814 * port if the link is down or the port is NULL.
1816 if ((lp = lagg_link_active(sc, lp)) == NULL) {
1822 return (lagg_enqueue(lp->lp_ifp, m));
1825 static struct mbuf *
1826 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1828 struct ifnet *ifp = sc->sc_ifp;
1830 /* Just pass in the packet to our lagg device */
1831 m->m_pkthdr.rcvif = ifp;
1840 lagg_fail_attach(struct lagg_softc *sc)
1842 sc->sc_detach = lagg_fail_detach;
1843 sc->sc_start = lagg_fail_start;
1844 sc->sc_input = lagg_fail_input;
1845 sc->sc_port_create = NULL;
1846 sc->sc_port_destroy = NULL;
1847 sc->sc_detach = NULL;
1851 lagg_fail_detach(struct lagg_softc *sc)
1857 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
1859 struct lagg_port *lp;
1861 /* Use the master port if active or the next available port */
1862 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
1868 return (lagg_enqueue(lp->lp_ifp, m));
1871 static struct mbuf *
1872 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
1874 struct ifnet *ifp = sc->sc_ifp;
1875 struct lagg_port *tmp_tp;
1877 if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
1878 m->m_pkthdr.rcvif = ifp;
1882 if (!LAGG_PORTACTIVE(sc->sc_primary)) {
1883 tmp_tp = lagg_link_active(sc, sc->sc_primary);
1885 * If tmp_tp is null, we've recieved a packet when all
1886 * our links are down. Weird, but process it anyways.
1888 if ((tmp_tp == NULL || tmp_tp == lp)) {
1889 m->m_pkthdr.rcvif = ifp;
1902 lagg_lb_attach(struct lagg_softc *sc)
1904 struct lagg_port *lp;
1907 lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO);
1909 sc->sc_detach = lagg_lb_detach;
1910 sc->sc_start = lagg_lb_start;
1911 sc->sc_input = lagg_lb_input;
1912 sc->sc_port_create = lagg_lb_port_create;
1913 sc->sc_port_destroy = lagg_lb_port_destroy;
1914 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX;
1916 lb->lb_key = arc4random();
1917 sc->sc_psc = (caddr_t)lb;
1919 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1920 lagg_lb_port_create(lp);
1924 lagg_lb_detach(struct lagg_softc *sc)
1926 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1934 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
1936 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1937 struct lagg_port *lp_next;
1940 bzero(&lb->lb_ports, sizeof(lb->lb_ports));
1941 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
1944 if (i >= LAGG_MAX_PORTS)
1946 if (sc->sc_ifflags & IFF_DEBUG)
1947 printf("%s: port %s at index %d\n",
1948 sc->sc_ifname, lp_next->lp_ifname, i);
1949 lb->lb_ports[i++] = lp_next;
1956 lagg_lb_port_create(struct lagg_port *lp)
1958 struct lagg_softc *sc = lp->lp_softc;
1959 return (lagg_lb_porttable(sc, NULL));
1963 lagg_lb_port_destroy(struct lagg_port *lp)
1965 struct lagg_softc *sc = lp->lp_softc;
1966 lagg_lb_porttable(sc, lp);
1970 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
1972 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
1973 struct lagg_port *lp = NULL;
1976 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
1977 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1978 p = m->m_pkthdr.flowid >> sc->flowid_shift;
1980 p = lagg_hashmbuf(sc, m, lb->lb_key);
1982 lp = lb->lb_ports[p];
1985 * Check the port's link state. This will return the next active
1986 * port if the link is down or the port is NULL.
1988 if ((lp = lagg_link_active(sc, lp)) == NULL) {
1994 return (lagg_enqueue(lp->lp_ifp, m));
1997 static struct mbuf *
1998 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2000 struct ifnet *ifp = sc->sc_ifp;
2002 /* Just pass in the packet to our lagg device */
2003 m->m_pkthdr.rcvif = ifp;
2012 lagg_lacp_attach(struct lagg_softc *sc)
2014 struct lagg_port *lp;
2016 sc->sc_detach = lagg_lacp_detach;
2017 sc->sc_port_create = lacp_port_create;
2018 sc->sc_port_destroy = lacp_port_destroy;
2019 sc->sc_linkstate = lacp_linkstate;
2020 sc->sc_start = lagg_lacp_start;
2021 sc->sc_input = lagg_lacp_input;
2022 sc->sc_init = lacp_init;
2023 sc->sc_stop = lacp_stop;
2024 sc->sc_lladdr = lagg_lacp_lladdr;
2025 sc->sc_req = lacp_req;
2026 sc->sc_portreq = lacp_portreq;
2030 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2031 lacp_port_create(lp);
2035 lagg_lacp_detach(struct lagg_softc *sc)
2037 struct lagg_port *lp;
2040 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2041 lacp_port_destroy(lp);
2053 lagg_lacp_lladdr(struct lagg_softc *sc)
2055 struct lagg_port *lp;
2057 /* purge all the lacp ports */
2058 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2059 lacp_port_destroy(lp);
2061 /* add them back in */
2062 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2063 lacp_port_create(lp);
2067 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2069 struct lagg_port *lp;
2071 lp = lacp_select_tx_port(sc, m);
2078 return (lagg_enqueue(lp->lp_ifp, m));
2081 static struct mbuf *
2082 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2084 struct ifnet *ifp = sc->sc_ifp;
2085 struct ether_header *eh;
2088 eh = mtod(m, struct ether_header *);
2089 etype = ntohs(eh->ether_type);
2091 /* Tap off LACP control messages */
2092 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2093 m = lacp_input(lp, m);
2099 * If the port is not collecting or not in the active aggregator then
2102 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2107 m->m_pkthdr.rcvif = ifp;
2112 lagg_callout(void *arg)
2114 struct lagg_softc *sc = (struct lagg_softc *)arg;
2115 struct ifnet *ifp = sc->sc_ifp;
2117 ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets);
2118 ifp->if_opackets = counter_u64_fetch(sc->sc_opackets);
2119 ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes);
2120 ifp->if_obytes = counter_u64_fetch(sc->sc_obytes);
2122 callout_reset(&sc->sc_callout, hz, lagg_callout, sc);