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>
6 * Copyright (c) 2014, 2016 Marcelo Araujo <araujo@FreeBSD.org>
8 * Permission to use, copy, modify, and distribute this software for any
9 * purpose with or without fee is hereby granted, provided that the above
10 * copyright notice and this permission notice appear in all copies.
12 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
21 #include <sys/cdefs.h>
22 __FBSDID("$FreeBSD$");
25 #include "opt_inet6.h"
26 #include "opt_kern_tls.h"
27 #include "opt_ratelimit.h"
29 #include <sys/param.h>
30 #include <sys/kernel.h>
31 #include <sys/malloc.h>
33 #include <sys/queue.h>
34 #include <sys/socket.h>
35 #include <sys/sockio.h>
36 #include <sys/sysctl.h>
37 #include <sys/module.h>
39 #include <sys/systm.h>
42 #include <sys/rmlock.h>
44 #include <sys/taskqueue.h>
45 #include <sys/eventhandler.h>
47 #include <net/ethernet.h>
49 #include <net/if_clone.h>
50 #include <net/if_arp.h>
51 #include <net/if_dl.h>
52 #include <net/if_media.h>
53 #include <net/if_types.h>
54 #include <net/if_var.h>
56 #include <net/route.h>
58 #include <net/infiniband.h>
60 #if defined(INET) || defined(INET6)
61 #include <netinet/in.h>
62 #include <netinet/ip.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/if_ether.h>
70 #include <netinet/ip6.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet6/in6_ifattach.h>
75 #include <net/if_vlan_var.h>
76 #include <net/if_lagg.h>
77 #include <net/ieee8023ad_lacp.h>
81 * XXX: declare here to avoid to include many inet6 related files..
82 * should be more generalized?
84 extern void nd6_setmtu(struct ifnet *);
87 #define LAGG_RLOCK() struct epoch_tracker lagg_et; epoch_enter_preempt(net_epoch_preempt, &lagg_et)
88 #define LAGG_RUNLOCK() epoch_exit_preempt(net_epoch_preempt, &lagg_et)
89 #define LAGG_RLOCK_ASSERT() NET_EPOCH_ASSERT()
90 #define LAGG_UNLOCK_ASSERT() MPASS(!in_epoch(net_epoch_preempt))
92 #define LAGG_SX_INIT(_sc) sx_init(&(_sc)->sc_sx, "if_lagg sx")
93 #define LAGG_SX_DESTROY(_sc) sx_destroy(&(_sc)->sc_sx)
94 #define LAGG_XLOCK(_sc) sx_xlock(&(_sc)->sc_sx)
95 #define LAGG_XUNLOCK(_sc) sx_xunlock(&(_sc)->sc_sx)
96 #define LAGG_SXLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_LOCKED)
97 #define LAGG_XLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
99 /* Special flags we should propagate to the lagg ports. */
102 int (*func)(struct ifnet *, int);
104 {IFF_PROMISC, ifpromisc},
105 {IFF_ALLMULTI, if_allmulti},
109 struct lagg_snd_tag {
110 struct m_snd_tag com;
111 struct m_snd_tag *tag;
114 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
115 #define V_lagg_list VNET(lagg_list)
116 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
117 #define V_lagg_list_mtx VNET(lagg_list_mtx)
118 #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \
119 "if_lagg list", NULL, MTX_DEF)
120 #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx)
121 #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx)
122 #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx)
123 eventhandler_tag lagg_detach_cookie = NULL;
125 static int lagg_clone_create(struct if_clone *, int, caddr_t);
126 static void lagg_clone_destroy(struct ifnet *);
127 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
128 #define V_lagg_cloner VNET(lagg_cloner)
129 static const char laggname[] = "lagg";
130 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
132 static void lagg_capabilities(struct lagg_softc *);
133 static int lagg_port_create(struct lagg_softc *, struct ifnet *);
134 static int lagg_port_destroy(struct lagg_port *, int);
135 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *);
136 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *);
137 static void lagg_linkstate(struct lagg_softc *);
138 static void lagg_port_state(struct ifnet *, int);
139 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
140 static int lagg_port_output(struct ifnet *, struct mbuf *,
141 const struct sockaddr *, struct route *);
142 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *);
143 #ifdef LAGG_PORT_STACKING
144 static int lagg_port_checkstacking(struct lagg_softc *);
146 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *);
147 static void lagg_init(void *);
148 static void lagg_stop(struct lagg_softc *);
149 static int lagg_ioctl(struct ifnet *, u_long, caddr_t);
150 #if defined(KERN_TLS) || defined(RATELIMIT)
151 static int lagg_snd_tag_alloc(struct ifnet *,
152 union if_snd_tag_alloc_params *,
153 struct m_snd_tag **);
154 static int lagg_snd_tag_modify(struct m_snd_tag *,
155 union if_snd_tag_modify_params *);
156 static int lagg_snd_tag_query(struct m_snd_tag *,
157 union if_snd_tag_query_params *);
158 static void lagg_snd_tag_free(struct m_snd_tag *);
159 static void lagg_ratelimit_query(struct ifnet *,
160 struct if_ratelimit_query_results *);
162 static int lagg_setmulti(struct lagg_port *);
163 static int lagg_clrmulti(struct lagg_port *);
164 static int lagg_setcaps(struct lagg_port *, int cap);
165 static int lagg_setflag(struct lagg_port *, int, int,
166 int (*func)(struct ifnet *, int));
167 static int lagg_setflags(struct lagg_port *, int status);
168 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
169 static int lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
170 static int lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
171 static void lagg_qflush(struct ifnet *);
172 static int lagg_media_change(struct ifnet *);
173 static void lagg_media_status(struct ifnet *, struct ifmediareq *);
174 static struct lagg_port *lagg_link_active(struct lagg_softc *,
177 /* Simple round robin */
178 static void lagg_rr_attach(struct lagg_softc *);
179 static int lagg_rr_start(struct lagg_softc *, struct mbuf *);
180 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
183 /* Active failover */
184 static int lagg_fail_start(struct lagg_softc *, struct mbuf *);
185 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
189 static void lagg_lb_attach(struct lagg_softc *);
190 static void lagg_lb_detach(struct lagg_softc *);
191 static int lagg_lb_port_create(struct lagg_port *);
192 static void lagg_lb_port_destroy(struct lagg_port *);
193 static int lagg_lb_start(struct lagg_softc *, struct mbuf *);
194 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
196 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
199 static int lagg_bcast_start(struct lagg_softc *, struct mbuf *);
200 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
204 static void lagg_lacp_attach(struct lagg_softc *);
205 static void lagg_lacp_detach(struct lagg_softc *);
206 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *);
207 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
209 static void lagg_lacp_lladdr(struct lagg_softc *);
211 /* lagg protocol table */
212 static const struct lagg_proto {
214 void (*pr_attach)(struct lagg_softc *);
215 void (*pr_detach)(struct lagg_softc *);
216 int (*pr_start)(struct lagg_softc *, struct mbuf *);
217 struct mbuf * (*pr_input)(struct lagg_softc *, struct lagg_port *,
219 int (*pr_addport)(struct lagg_port *);
220 void (*pr_delport)(struct lagg_port *);
221 void (*pr_linkstate)(struct lagg_port *);
222 void (*pr_init)(struct lagg_softc *);
223 void (*pr_stop)(struct lagg_softc *);
224 void (*pr_lladdr)(struct lagg_softc *);
225 void (*pr_request)(struct lagg_softc *, void *);
226 void (*pr_portreq)(struct lagg_port *, void *);
229 .pr_num = LAGG_PROTO_NONE
232 .pr_num = LAGG_PROTO_ROUNDROBIN,
233 .pr_attach = lagg_rr_attach,
234 .pr_start = lagg_rr_start,
235 .pr_input = lagg_rr_input,
238 .pr_num = LAGG_PROTO_FAILOVER,
239 .pr_start = lagg_fail_start,
240 .pr_input = lagg_fail_input,
243 .pr_num = LAGG_PROTO_LOADBALANCE,
244 .pr_attach = lagg_lb_attach,
245 .pr_detach = lagg_lb_detach,
246 .pr_start = lagg_lb_start,
247 .pr_input = lagg_lb_input,
248 .pr_addport = lagg_lb_port_create,
249 .pr_delport = lagg_lb_port_destroy,
252 .pr_num = LAGG_PROTO_LACP,
253 .pr_attach = lagg_lacp_attach,
254 .pr_detach = lagg_lacp_detach,
255 .pr_start = lagg_lacp_start,
256 .pr_input = lagg_lacp_input,
257 .pr_addport = lacp_port_create,
258 .pr_delport = lacp_port_destroy,
259 .pr_linkstate = lacp_linkstate,
260 .pr_init = lacp_init,
261 .pr_stop = lacp_stop,
262 .pr_lladdr = lagg_lacp_lladdr,
263 .pr_request = lacp_req,
264 .pr_portreq = lacp_portreq,
267 .pr_num = LAGG_PROTO_BROADCAST,
268 .pr_start = lagg_bcast_start,
269 .pr_input = lagg_bcast_input,
273 SYSCTL_DECL(_net_link);
274 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
277 /* Allow input on any failover links */
278 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
279 #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all)
280 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
281 &VNET_NAME(lagg_failover_rx_all), 0,
282 "Accept input from any interface in a failover lagg");
284 /* Default value for using flowid */
285 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
286 #define V_def_use_flowid VNET(def_use_flowid)
287 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
288 &VNET_NAME(def_use_flowid), 0,
289 "Default setting for using flow id for load sharing");
291 /* Default value for using numa */
292 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
293 #define V_def_use_numa VNET(def_use_numa)
294 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN,
295 &VNET_NAME(def_use_numa), 0,
296 "Use numa to steer flows");
298 /* Default value for flowid shift */
299 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
300 #define V_def_flowid_shift VNET(def_flowid_shift)
301 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
302 &VNET_NAME(def_flowid_shift), 0,
303 "Default setting for flowid shift for load sharing");
306 vnet_lagg_init(const void *unused __unused)
309 LAGG_LIST_LOCK_INIT();
310 SLIST_INIT(&V_lagg_list);
311 V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
312 lagg_clone_destroy, 0);
314 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
315 vnet_lagg_init, NULL);
318 vnet_lagg_uninit(const void *unused __unused)
321 if_clone_detach(V_lagg_cloner);
322 LAGG_LIST_LOCK_DESTROY();
324 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
325 vnet_lagg_uninit, NULL);
328 lagg_modevent(module_t mod, int type, void *data)
333 lagg_input_ethernet_p = lagg_input_ethernet;
334 lagg_input_infiniband_p = lagg_input_infiniband;
335 lagg_linkstate_p = lagg_port_state;
336 lagg_detach_cookie = EVENTHANDLER_REGISTER(
337 ifnet_departure_event, lagg_port_ifdetach, NULL,
338 EVENTHANDLER_PRI_ANY);
341 EVENTHANDLER_DEREGISTER(ifnet_departure_event,
343 lagg_input_ethernet_p = NULL;
344 lagg_input_infiniband_p = NULL;
345 lagg_linkstate_p = NULL;
353 static moduledata_t lagg_mod = {
359 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
360 MODULE_VERSION(if_lagg, 1);
361 MODULE_DEPEND(if_lagg, if_infiniband, 1, 1, 1);
364 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
367 LAGG_XLOCK_ASSERT(sc);
368 KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
371 if (sc->sc_ifflags & IFF_DEBUG)
372 if_printf(sc->sc_ifp, "using proto %u\n", pr);
374 if (lagg_protos[pr].pr_attach != NULL)
375 lagg_protos[pr].pr_attach(sc);
380 lagg_proto_detach(struct lagg_softc *sc)
384 LAGG_XLOCK_ASSERT(sc);
386 sc->sc_proto = LAGG_PROTO_NONE;
388 if (lagg_protos[pr].pr_detach != NULL)
389 lagg_protos[pr].pr_detach(sc);
393 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
396 return (lagg_protos[sc->sc_proto].pr_start(sc, m));
400 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
403 return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
407 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
410 if (lagg_protos[sc->sc_proto].pr_addport == NULL)
413 return (lagg_protos[sc->sc_proto].pr_addport(lp));
417 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
420 if (lagg_protos[sc->sc_proto].pr_delport != NULL)
421 lagg_protos[sc->sc_proto].pr_delport(lp);
425 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
428 if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
429 lagg_protos[sc->sc_proto].pr_linkstate(lp);
433 lagg_proto_init(struct lagg_softc *sc)
436 if (lagg_protos[sc->sc_proto].pr_init != NULL)
437 lagg_protos[sc->sc_proto].pr_init(sc);
441 lagg_proto_stop(struct lagg_softc *sc)
444 if (lagg_protos[sc->sc_proto].pr_stop != NULL)
445 lagg_protos[sc->sc_proto].pr_stop(sc);
449 lagg_proto_lladdr(struct lagg_softc *sc)
452 if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
453 lagg_protos[sc->sc_proto].pr_lladdr(sc);
457 lagg_proto_request(struct lagg_softc *sc, void *v)
460 if (lagg_protos[sc->sc_proto].pr_request != NULL)
461 lagg_protos[sc->sc_proto].pr_request(sc, v);
465 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
468 if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
469 lagg_protos[sc->sc_proto].pr_portreq(lp, v);
473 * This routine is run via an vlan
477 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
479 struct lagg_softc *sc = ifp->if_softc;
480 struct lagg_port *lp;
482 if (ifp->if_softc != arg) /* Not our event */
486 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
487 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
492 * This routine is run via an vlan
496 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
498 struct lagg_softc *sc = ifp->if_softc;
499 struct lagg_port *lp;
501 if (ifp->if_softc != arg) /* Not our event */
505 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
506 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
511 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
513 struct iflaggparam iflp;
514 struct lagg_softc *sc;
518 static const uint8_t eaddr[LAGG_ADDR_LEN];
519 static const uint8_t ib_bcast_addr[INFINIBAND_ADDR_LEN] = {
520 0x00, 0xff, 0xff, 0xff,
521 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
522 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff
525 if (params != NULL) {
526 error = copyin(params, &iflp, sizeof(iflp));
530 switch (iflp.lagg_type) {
531 case LAGG_TYPE_ETHERNET:
534 case LAGG_TYPE_INFINIBAND:
535 if_type = IFT_INFINIBAND;
544 sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO);
545 ifp = sc->sc_ifp = if_alloc(if_type);
552 mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
553 callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
556 if (V_def_use_flowid)
557 sc->sc_opts |= LAGG_OPT_USE_FLOWID;
559 sc->sc_opts |= LAGG_OPT_USE_NUMA;
560 sc->flowid_shift = V_def_flowid_shift;
562 /* Hash all layers by default */
563 sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
565 lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
567 CK_SLIST_INIT(&sc->sc_ports);
571 /* Initialise pseudo media types */
572 ifmedia_init(&sc->sc_media, 0, lagg_media_change,
574 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
575 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
577 if_initname(ifp, laggname, unit);
578 ifp->if_transmit = lagg_transmit_ethernet;
581 if_initname(ifp, laggname, unit);
582 ifp->if_transmit = lagg_transmit_infiniband;
588 ifp->if_qflush = lagg_qflush;
589 ifp->if_init = lagg_init;
590 ifp->if_ioctl = lagg_ioctl;
591 ifp->if_get_counter = lagg_get_counter;
592 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
593 #if defined(KERN_TLS) || defined(RATELIMIT)
594 ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
595 ifp->if_snd_tag_modify = lagg_snd_tag_modify;
596 ifp->if_snd_tag_query = lagg_snd_tag_query;
597 ifp->if_snd_tag_free = lagg_snd_tag_free;
598 ifp->if_ratelimit_query = lagg_ratelimit_query;
600 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
603 * Attach as an ordinary ethernet device, children will be attached
604 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
608 ether_ifattach(ifp, eaddr);
611 infiniband_ifattach(ifp, eaddr, ib_bcast_addr);
617 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
618 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
619 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
620 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
622 /* Insert into the global list of laggs */
624 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
632 lagg_clone_destroy(struct ifnet *ifp)
634 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
635 struct lagg_port *lp;
638 sc->sc_destroying = 1;
640 ifp->if_flags &= ~IFF_UP;
642 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
643 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
645 /* Shutdown and remove lagg ports */
646 while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
647 lagg_port_destroy(lp, 1);
649 /* Unhook the aggregation protocol */
650 lagg_proto_detach(sc);
653 switch (ifp->if_type) {
655 ifmedia_removeall(&sc->sc_media);
659 infiniband_ifdetach(ifp);
667 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
670 mtx_destroy(&sc->sc_mtx);
676 lagg_capabilities(struct lagg_softc *sc)
678 struct lagg_port *lp;
681 struct ifnet_hw_tsomax hw_tsomax;
683 LAGG_XLOCK_ASSERT(sc);
685 /* Get common enabled capabilities for the lagg ports */
687 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
688 ena &= lp->lp_ifp->if_capenable;
689 ena = (ena == ~0 ? 0 : ena);
692 * Apply common enabled capabilities back to the lagg ports.
693 * May require several iterations if they are dependent.
697 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
698 lagg_setcaps(lp, ena);
699 ena &= lp->lp_ifp->if_capenable;
701 } while (pena != ena);
703 /* Get other capabilities from the lagg ports */
706 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
707 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
708 cap &= lp->lp_ifp->if_capabilities;
709 hwa &= lp->lp_ifp->if_hwassist;
710 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
712 cap = (cap == ~0 ? 0 : cap);
713 hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
715 if (sc->sc_ifp->if_capabilities != cap ||
716 sc->sc_ifp->if_capenable != ena ||
717 sc->sc_ifp->if_hwassist != hwa ||
718 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
719 sc->sc_ifp->if_capabilities = cap;
720 sc->sc_ifp->if_capenable = ena;
721 sc->sc_ifp->if_hwassist = hwa;
722 getmicrotime(&sc->sc_ifp->if_lastchange);
724 if (sc->sc_ifflags & IFF_DEBUG)
725 if_printf(sc->sc_ifp,
726 "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
731 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
733 struct lagg_softc *sc_ptr;
734 struct lagg_port *lp, *tlp;
736 int error, i, oldmtu;
740 LAGG_XLOCK_ASSERT(sc);
742 if (sc->sc_ifp == ifp) {
743 if_printf(sc->sc_ifp,
744 "cannot add a lagg to itself as a port\n");
748 if (sc->sc_destroying == 1)
751 /* Limit the maximal number of lagg ports */
752 if (sc->sc_count >= LAGG_MAX_PORTS)
755 /* Check if port has already been associated to a lagg */
756 if (ifp->if_lagg != NULL) {
757 /* Port is already in the current lagg? */
758 lp = (struct lagg_port *)ifp->if_lagg;
759 if (lp->lp_softc == sc)
764 switch (sc->sc_ifp->if_type) {
766 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */
767 if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
768 return (EPROTONOSUPPORT);
769 if_type = IFT_IEEE8023ADLAG;
772 /* XXX Disallow non-infiniband interfaces */
773 if (ifp->if_type != IFT_INFINIBAND)
774 return (EPROTONOSUPPORT);
775 if_type = IFT_INFINIBANDLAG;
781 /* Allow the first Ethernet member to define the MTU */
783 if (CK_SLIST_EMPTY(&sc->sc_ports)) {
784 sc->sc_ifp->if_mtu = ifp->if_mtu;
785 } else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
786 if (ifp->if_ioctl == NULL) {
787 if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
791 oldmtu = ifp->if_mtu;
792 strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
793 ifr.ifr_mtu = sc->sc_ifp->if_mtu;
794 error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
796 if_printf(sc->sc_ifp, "invalid MTU for %s\n",
800 ifr.ifr_mtu = oldmtu;
803 lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO);
806 /* Check if port is a stacked lagg */
808 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
809 if (ifp == sc_ptr->sc_ifp) {
813 (*ifp->if_ioctl)(ifp, SIOCSIFMTU,
816 /* XXX disable stacking for the moment, its untested */
817 #ifdef LAGG_PORT_STACKING
818 lp->lp_flags |= LAGG_PORT_STACK;
819 if (lagg_port_checkstacking(sc_ptr) >=
824 (*ifp->if_ioctl)(ifp, SIOCSIFMTU,
836 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
837 lp->lp_ifcapenable = ifp->if_capenable;
838 if (CK_SLIST_EMPTY(&sc->sc_ports)) {
839 bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
840 lagg_proto_lladdr(sc);
841 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
843 if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
845 lagg_setflags(lp, 1);
847 if (CK_SLIST_EMPTY(&sc->sc_ports))
850 /* Change the interface type */
851 lp->lp_iftype = ifp->if_type;
852 ifp->if_type = if_type;
854 lp->lp_ioctl = ifp->if_ioctl;
855 ifp->if_ioctl = lagg_port_ioctl;
856 lp->lp_output = ifp->if_output;
857 ifp->if_output = lagg_port_output;
859 /* Read port counters */
860 pval = lp->port_counters.val;
861 for (i = 0; i < IFCOUNTERS; i++, pval++)
862 *pval = ifp->if_get_counter(ifp, i);
865 * Insert into the list of ports.
866 * Keep ports sorted by if_index. It is handy, when configuration
867 * is predictable and `ifconfig laggN create ...` command
868 * will lead to the same result each time.
870 CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
871 if (tlp->lp_ifp->if_index < ifp->if_index && (
872 CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
873 ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
878 CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
880 CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
885 if ((error = lagg_proto_addport(sc, lp)) != 0) {
886 /* Remove the port, without calling pr_delport. */
887 lagg_port_destroy(lp, 0);
889 (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
893 /* Update lagg capabilities */
894 lagg_capabilities(sc);
900 #ifdef LAGG_PORT_STACKING
902 lagg_port_checkstacking(struct lagg_softc *sc)
904 struct lagg_softc *sc_ptr;
905 struct lagg_port *lp;
908 LAGG_SXLOCK_ASSERT(sc);
909 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
910 if (lp->lp_flags & LAGG_PORT_STACK) {
911 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
912 m = MAX(m, lagg_port_checkstacking(sc_ptr));
921 lagg_port_destroy_cb(epoch_context_t ec)
923 struct lagg_port *lp;
926 lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
934 lagg_port_destroy(struct lagg_port *lp, int rundelport)
936 struct lagg_softc *sc = lp->lp_softc;
937 struct lagg_port *lp_ptr, *lp0;
938 struct ifnet *ifp = lp->lp_ifp;
939 uint64_t *pval, vdiff;
942 LAGG_XLOCK_ASSERT(sc);
945 lagg_proto_delport(sc, lp);
947 if (lp->lp_detaching == 0)
950 /* Restore interface */
951 ifp->if_type = lp->lp_iftype;
952 ifp->if_ioctl = lp->lp_ioctl;
953 ifp->if_output = lp->lp_output;
956 /* Update detached port counters */
957 pval = lp->port_counters.val;
958 for (i = 0; i < IFCOUNTERS; i++, pval++) {
959 vdiff = ifp->if_get_counter(ifp, i) - *pval;
960 sc->detached_counters.val[i] += vdiff;
963 /* Finally, remove the port from the lagg */
964 CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
967 /* Update the primary interface */
968 if (lp == sc->sc_primary) {
969 uint8_t lladdr[LAGG_ADDR_LEN];
971 if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
972 bzero(&lladdr, LAGG_ADDR_LEN);
974 bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
975 sc->sc_primary = lp0;
976 if (sc->sc_destroying == 0) {
977 bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
978 lagg_proto_lladdr(sc);
979 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
983 * Update lladdr for each port (new primary needs update
984 * as well, to switch from old lladdr to its 'real' one)
986 CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
987 if_setlladdr(lp_ptr->lp_ifp, lladdr, lp_ptr->lp_ifp->if_addrlen);
991 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
993 if (lp->lp_detaching == 0) {
994 lagg_setflags(lp, 0);
995 lagg_setcaps(lp, lp->lp_ifcapenable);
996 if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
1000 * free port and release it's ifnet reference after a grace period has
1003 NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
1004 /* Update lagg capabilities */
1005 lagg_capabilities(sc);
1012 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1014 struct lagg_reqport *rp = (struct lagg_reqport *)data;
1015 struct lagg_softc *sc;
1016 struct lagg_port *lp = NULL;
1019 /* Should be checked by the caller */
1020 switch (ifp->if_type) {
1021 case IFT_IEEE8023ADLAG:
1022 case IFT_INFINIBANDLAG:
1023 if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1032 if (rp->rp_portname[0] == '\0' ||
1033 ifunit(rp->rp_portname) != ifp) {
1039 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1045 lagg_port2req(lp, rp);
1050 if (lp->lp_ioctl == NULL) {
1054 error = (*lp->lp_ioctl)(ifp, cmd, data);
1058 /* Update lagg interface capabilities */
1060 lagg_capabilities(sc);
1062 VLAN_CAPABILITIES(sc->sc_ifp);
1066 /* Do not allow the MTU to be changed once joined */
1077 if (lp != NULL && lp->lp_ioctl != NULL)
1078 return ((*lp->lp_ioctl)(ifp, cmd, data));
1084 * Requests counter @cnt data.
1086 * Counter value is calculated the following way:
1087 * 1) for each port, sum difference between current and "initial" measurements.
1088 * 2) add lagg logical interface counters.
1089 * 3) add data from detached_counters array.
1091 * We also do the following things on ports attach/detach:
1092 * 1) On port attach we store all counters it has into port_counter array.
1093 * 2) On port detach we add the different between "initial" and
1094 * current counters data to detached_counters array.
1097 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1099 struct lagg_softc *sc;
1100 struct lagg_port *lp;
1101 struct ifnet *lpifp;
1102 uint64_t newval, oldval, vsum;
1104 /* Revise this when we've got non-generic counters. */
1105 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1107 sc = (struct lagg_softc *)ifp->if_softc;
1111 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1112 /* Saved attached value */
1113 oldval = lp->port_counters.val[cnt];
1116 newval = lpifp->if_get_counter(lpifp, cnt);
1117 /* Calculate diff and save new */
1118 vsum += newval - oldval;
1123 * Add counter data which might be added by upper
1124 * layer protocols operating on logical interface.
1126 vsum += if_get_counter_default(ifp, cnt);
1129 * Add counter data from detached ports counters
1131 vsum += sc->detached_counters.val[cnt];
1137 * For direct output to child ports.
1140 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1141 const struct sockaddr *dst, struct route *ro)
1143 struct lagg_port *lp = ifp->if_lagg;
1145 switch (dst->sa_family) {
1146 case pseudo_AF_HDRCMPLT:
1148 return ((*lp->lp_output)(ifp, m, dst, ro));
1151 /* drop any other frames */
1157 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1159 struct lagg_port *lp;
1160 struct lagg_softc *sc;
1162 if ((lp = ifp->if_lagg) == NULL)
1164 /* If the ifnet is just being renamed, don't do anything. */
1165 if (ifp->if_flags & IFF_RENAMING)
1171 lp->lp_detaching = 1;
1172 lagg_port_destroy(lp, 1);
1174 VLAN_CAPABILITIES(sc->sc_ifp);
1178 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1180 struct lagg_softc *sc = lp->lp_softc;
1182 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1183 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1184 rp->rp_prio = lp->lp_prio;
1185 rp->rp_flags = lp->lp_flags;
1186 lagg_proto_portreq(sc, lp, &rp->rp_psc);
1188 /* Add protocol specific flags */
1189 switch (sc->sc_proto) {
1190 case LAGG_PROTO_FAILOVER:
1191 if (lp == sc->sc_primary)
1192 rp->rp_flags |= LAGG_PORT_MASTER;
1193 if (lp == lagg_link_active(sc, sc->sc_primary))
1194 rp->rp_flags |= LAGG_PORT_ACTIVE;
1197 case LAGG_PROTO_ROUNDROBIN:
1198 case LAGG_PROTO_LOADBALANCE:
1199 case LAGG_PROTO_BROADCAST:
1200 if (LAGG_PORTACTIVE(lp))
1201 rp->rp_flags |= LAGG_PORT_ACTIVE;
1204 case LAGG_PROTO_LACP:
1205 /* LACP has a different definition of active */
1206 if (lacp_isactive(lp))
1207 rp->rp_flags |= LAGG_PORT_ACTIVE;
1208 if (lacp_iscollecting(lp))
1209 rp->rp_flags |= LAGG_PORT_COLLECTING;
1210 if (lacp_isdistributing(lp))
1211 rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1218 lagg_watchdog_infiniband(void *arg)
1220 struct lagg_softc *sc;
1221 struct lagg_port *lp;
1223 struct ifnet *lp_ifp;
1228 * Because infiniband nodes have a fixed MAC address, which is
1229 * generated by the so-called GID, we need to regularly update
1230 * the link level address of the parent lagg<N> device when
1231 * the active port changes. Possibly we could piggy-back on
1232 * link up/down events aswell, but using a timer also provides
1233 * a guarantee against too frequent events. This operation
1234 * does not have to be atomic.
1237 lp = lagg_link_active(sc, sc->sc_primary);
1240 lp_ifp = lp->lp_ifp;
1242 if (ifp != NULL && lp_ifp != NULL &&
1243 memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0) {
1244 memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1245 CURVNET_SET(ifp->if_vnet);
1246 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1252 callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1256 lagg_init(void *xsc)
1258 struct lagg_softc *sc = (struct lagg_softc *)xsc;
1259 struct ifnet *ifp = sc->sc_ifp;
1260 struct lagg_port *lp;
1263 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1268 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1271 * Update the port lladdrs if needed.
1272 * This might be if_setlladdr() notification
1273 * that lladdr has been changed.
1275 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1276 if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1277 ifp->if_addrlen) != 0)
1278 if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1281 lagg_proto_init(sc);
1283 if (ifp->if_type == IFT_INFINIBAND) {
1284 mtx_lock(&sc->sc_mtx);
1285 lagg_watchdog_infiniband(sc);
1286 mtx_unlock(&sc->sc_mtx);
1293 lagg_stop(struct lagg_softc *sc)
1295 struct ifnet *ifp = sc->sc_ifp;
1297 LAGG_XLOCK_ASSERT(sc);
1299 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1302 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1304 lagg_proto_stop(sc);
1306 mtx_lock(&sc->sc_mtx);
1307 callout_stop(&sc->sc_watchdog);
1308 mtx_unlock(&sc->sc_mtx);
1310 callout_drain(&sc->sc_watchdog);
1314 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1316 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1317 struct lagg_reqall *ra = (struct lagg_reqall *)data;
1318 struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1319 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1320 struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1321 struct ifreq *ifr = (struct ifreq *)data;
1322 struct lagg_port *lp;
1324 struct thread *td = curthread;
1326 int count, buflen, len, error = 0, oldmtu;
1328 bzero(&rpbuf, sizeof(rpbuf));
1330 /* XXX: This can race with lagg_clone_destroy. */
1335 buflen = sc->sc_count * sizeof(struct lagg_reqport);
1336 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1337 ra->ra_proto = sc->sc_proto;
1338 lagg_proto_request(sc, &ra->ra_psc);
1341 len = min(ra->ra_size, buflen);
1342 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1343 if (len < sizeof(rpbuf))
1346 lagg_port2req(lp, &rpbuf);
1347 memcpy(buf, &rpbuf, sizeof(rpbuf));
1349 buf += sizeof(rpbuf);
1350 len -= sizeof(rpbuf);
1353 ra->ra_ports = count;
1354 ra->ra_size = count * sizeof(rpbuf);
1355 error = copyout(outbuf, ra->ra_port, ra->ra_size);
1356 free(outbuf, M_TEMP);
1359 error = priv_check(td, PRIV_NET_LAGG);
1362 if (ra->ra_proto >= LAGG_PROTO_MAX) {
1363 error = EPROTONOSUPPORT;
1366 /* Infiniband only supports the failover protocol. */
1367 if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1368 ifp->if_type == IFT_INFINIBAND) {
1369 error = EPROTONOSUPPORT;
1373 lagg_proto_detach(sc);
1374 LAGG_UNLOCK_ASSERT();
1375 lagg_proto_attach(sc, ra->ra_proto);
1380 ro->ro_opts = sc->sc_opts;
1381 if (sc->sc_proto == LAGG_PROTO_LACP) {
1382 struct lacp_softc *lsc;
1384 lsc = (struct lacp_softc *)sc->sc_psc;
1385 if (lsc->lsc_debug.lsc_tx_test != 0)
1386 ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1387 if (lsc->lsc_debug.lsc_rx_test != 0)
1388 ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1389 if (lsc->lsc_strict_mode != 0)
1390 ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1391 if (lsc->lsc_fast_timeout != 0)
1392 ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1394 ro->ro_active = sc->sc_active;
1397 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1398 ro->ro_active += LAGG_PORTACTIVE(lp);
1400 ro->ro_bkt = sc->sc_stride;
1401 ro->ro_flapping = sc->sc_flapping;
1402 ro->ro_flowid_shift = sc->flowid_shift;
1406 error = priv_check(td, PRIV_NET_LAGG);
1411 * The stride option was added without defining a corresponding
1412 * LAGG_OPT flag, so handle a non-zero value before checking
1413 * anything else to preserve compatibility.
1416 if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1417 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1422 sc->sc_stride = ro->ro_bkt;
1424 if (ro->ro_opts == 0) {
1430 * Set options. LACP options are stored in sc->sc_psc,
1435 switch (ro->ro_opts) {
1436 case LAGG_OPT_USE_FLOWID:
1437 case -LAGG_OPT_USE_FLOWID:
1438 case LAGG_OPT_USE_NUMA:
1439 case -LAGG_OPT_USE_NUMA:
1440 case LAGG_OPT_FLOWIDSHIFT:
1441 case LAGG_OPT_RR_LIMIT:
1445 case LAGG_OPT_LACP_TXTEST:
1446 case -LAGG_OPT_LACP_TXTEST:
1447 case LAGG_OPT_LACP_RXTEST:
1448 case -LAGG_OPT_LACP_RXTEST:
1449 case LAGG_OPT_LACP_STRICT:
1450 case -LAGG_OPT_LACP_STRICT:
1451 case LAGG_OPT_LACP_FAST_TIMO:
1452 case -LAGG_OPT_LACP_FAST_TIMO:
1461 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1462 /* Invalid combination of options specified. */
1465 break; /* Return from SIOCSLAGGOPTS. */
1469 * Store new options into sc->sc_opts except for
1470 * FLOWIDSHIFT, RR and LACP options.
1473 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1474 sc->flowid_shift = ro->ro_flowid_shift;
1475 else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1476 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1482 sc->sc_stride = ro->ro_bkt;
1483 } else if (ro->ro_opts > 0)
1484 sc->sc_opts |= ro->ro_opts;
1486 sc->sc_opts &= ~ro->ro_opts;
1488 struct lacp_softc *lsc;
1489 struct lacp_port *lp;
1491 lsc = (struct lacp_softc *)sc->sc_psc;
1493 switch (ro->ro_opts) {
1494 case LAGG_OPT_LACP_TXTEST:
1495 lsc->lsc_debug.lsc_tx_test = 1;
1497 case -LAGG_OPT_LACP_TXTEST:
1498 lsc->lsc_debug.lsc_tx_test = 0;
1500 case LAGG_OPT_LACP_RXTEST:
1501 lsc->lsc_debug.lsc_rx_test = 1;
1503 case -LAGG_OPT_LACP_RXTEST:
1504 lsc->lsc_debug.lsc_rx_test = 0;
1506 case LAGG_OPT_LACP_STRICT:
1507 lsc->lsc_strict_mode = 1;
1509 case -LAGG_OPT_LACP_STRICT:
1510 lsc->lsc_strict_mode = 0;
1512 case LAGG_OPT_LACP_FAST_TIMO:
1514 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1515 lp->lp_state |= LACP_STATE_TIMEOUT;
1517 lsc->lsc_fast_timeout = 1;
1519 case -LAGG_OPT_LACP_FAST_TIMO:
1521 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1522 lp->lp_state &= ~LACP_STATE_TIMEOUT;
1524 lsc->lsc_fast_timeout = 0;
1530 case SIOCGLAGGFLAGS:
1533 if (sc->sc_flags & MBUF_HASHFLAG_L2)
1534 rf->rf_flags |= LAGG_F_HASHL2;
1535 if (sc->sc_flags & MBUF_HASHFLAG_L3)
1536 rf->rf_flags |= LAGG_F_HASHL3;
1537 if (sc->sc_flags & MBUF_HASHFLAG_L4)
1538 rf->rf_flags |= LAGG_F_HASHL4;
1542 error = priv_check(td, PRIV_NET_LAGG);
1545 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1551 if (rf->rf_flags & LAGG_F_HASHL2)
1552 sc->sc_flags |= MBUF_HASHFLAG_L2;
1553 if (rf->rf_flags & LAGG_F_HASHL3)
1554 sc->sc_flags |= MBUF_HASHFLAG_L3;
1555 if (rf->rf_flags & LAGG_F_HASHL4)
1556 sc->sc_flags |= MBUF_HASHFLAG_L4;
1560 if (rp->rp_portname[0] == '\0' ||
1561 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1567 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1568 lp->lp_softc != sc) {
1575 lagg_port2req(lp, rp);
1580 error = priv_check(td, PRIV_NET_LAGG);
1583 if (rp->rp_portname[0] == '\0' ||
1584 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1590 * A laggport interface should not have inet6 address
1591 * because two interfaces with a valid link-local
1592 * scope zone must not be merged in any form. This
1593 * restriction is needed to prevent violation of
1594 * link-local scope zone. Attempts to add a laggport
1595 * interface which has inet6 addresses triggers
1596 * removal of all inet6 addresses on the member
1599 if (in6ifa_llaonifp(tpif)) {
1601 if_printf(sc->sc_ifp,
1602 "IPv6 addresses on %s have been removed "
1603 "before adding it as a member to prevent "
1604 "IPv6 address scope violation.\n",
1608 oldmtu = ifp->if_mtu;
1610 error = lagg_port_create(sc, tpif);
1615 * LAGG MTU may change during addition of the first port.
1616 * If it did, do network layer specific procedure.
1618 if (ifp->if_mtu != oldmtu) {
1625 VLAN_CAPABILITIES(ifp);
1627 case SIOCSLAGGDELPORT:
1628 error = priv_check(td, PRIV_NET_LAGG);
1631 if (rp->rp_portname[0] == '\0' ||
1632 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1638 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1639 lp->lp_softc != sc) {
1646 error = lagg_port_destroy(lp, 1);
1649 VLAN_CAPABILITIES(ifp);
1652 /* Set flags on ports too */
1654 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1655 lagg_setflags(lp, 1);
1658 if (!(ifp->if_flags & IFF_UP) &&
1659 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1661 * If interface is marked down and it is running,
1662 * then stop and disable it.
1666 } else if ((ifp->if_flags & IFF_UP) &&
1667 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1669 * If interface is marked up and it is stopped, then
1673 (*ifp->if_init)(sc);
1680 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1689 if (ifp->if_type == IFT_INFINIBAND)
1692 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1697 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1698 if (lp->lp_ioctl != NULL)
1699 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1701 lagg_capabilities(sc);
1703 VLAN_CAPABILITIES(ifp);
1709 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1710 if (lp->lp_ioctl != NULL)
1711 error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1716 "failed to change MTU to %d on port %s, "
1717 "reverting all ports to original MTU (%d)\n",
1718 ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1723 ifp->if_mtu = ifr->ifr_mtu;
1725 /* set every port back to the original MTU */
1726 ifr->ifr_mtu = ifp->if_mtu;
1727 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1728 if (lp->lp_ioctl != NULL)
1729 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1736 error = ether_ioctl(ifp, cmd, data);
1742 #if defined(KERN_TLS) || defined(RATELIMIT)
1743 static inline struct lagg_snd_tag *
1744 mst_to_lst(struct m_snd_tag *mst)
1747 return (__containerof(mst, struct lagg_snd_tag, com));
1751 * Look up the port used by a specific flow. This only works for lagg
1752 * protocols with deterministic port mappings (e.g. not roundrobin).
1753 * In addition protocols which use a hash to map flows to ports must
1754 * be configured to use the mbuf flowid rather than hashing packet
1757 static struct lagg_port *
1758 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1759 uint8_t numa_domain)
1761 struct lagg_softc *sc;
1762 struct lagg_port *lp;
1768 switch (sc->sc_proto) {
1769 case LAGG_PROTO_FAILOVER:
1770 return (lagg_link_active(sc, sc->sc_primary));
1771 case LAGG_PROTO_LOADBALANCE:
1772 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1773 flowtype == M_HASHTYPE_NONE)
1775 p = flowid >> sc->flowid_shift;
1777 lb = (struct lagg_lb *)sc->sc_psc;
1778 lp = lb->lb_ports[p];
1779 return (lagg_link_active(sc, lp));
1780 case LAGG_PROTO_LACP:
1781 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1782 flowtype == M_HASHTYPE_NONE)
1784 hash = flowid >> sc->flowid_shift;
1785 return (lacp_select_tx_port_by_hash(sc, hash, numa_domain));
1792 lagg_snd_tag_alloc(struct ifnet *ifp,
1793 union if_snd_tag_alloc_params *params,
1794 struct m_snd_tag **ppmt)
1796 struct lagg_snd_tag *lst;
1797 struct lagg_softc *sc;
1798 struct lagg_port *lp;
1799 struct ifnet *lp_ifp;
1805 lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1806 params->hdr.flowtype, params->hdr.numa_domain);
1809 return (EOPNOTSUPP);
1811 if (lp->lp_ifp == NULL || lp->lp_ifp->if_snd_tag_alloc == NULL) {
1813 return (EOPNOTSUPP);
1815 lp_ifp = lp->lp_ifp;
1819 lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1825 error = lp_ifp->if_snd_tag_alloc(lp_ifp, params, &lst->tag);
1832 m_snd_tag_init(&lst->com, ifp, lst->tag->type);
1839 lagg_snd_tag_modify(struct m_snd_tag *mst,
1840 union if_snd_tag_modify_params *params)
1842 struct lagg_snd_tag *lst;
1844 lst = mst_to_lst(mst);
1845 return (lst->tag->ifp->if_snd_tag_modify(lst->tag, params));
1849 lagg_snd_tag_query(struct m_snd_tag *mst,
1850 union if_snd_tag_query_params *params)
1852 struct lagg_snd_tag *lst;
1854 lst = mst_to_lst(mst);
1855 return (lst->tag->ifp->if_snd_tag_query(lst->tag, params));
1859 lagg_snd_tag_free(struct m_snd_tag *mst)
1861 struct lagg_snd_tag *lst;
1863 lst = mst_to_lst(mst);
1864 m_snd_tag_rele(lst->tag);
1869 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1872 * For lagg, we have an indirect
1873 * interface. The caller needs to
1874 * get a ratelimit tag on the actual
1875 * interface the flow will go on.
1877 q->rate_table = NULL;
1878 q->flags = RT_IS_INDIRECT;
1880 q->number_of_rates = 0;
1885 lagg_setmulti(struct lagg_port *lp)
1887 struct lagg_softc *sc = lp->lp_softc;
1888 struct ifnet *ifp = lp->lp_ifp;
1889 struct ifnet *scifp = sc->sc_ifp;
1891 struct ifmultiaddr *ifma;
1894 IF_ADDR_WLOCK(scifp);
1895 CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1896 if (ifma->ifma_addr->sa_family != AF_LINK)
1898 mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
1900 IF_ADDR_WUNLOCK(scifp);
1903 bcopy(ifma->ifma_addr, &mc->mc_addr,
1904 ifma->ifma_addr->sa_len);
1905 mc->mc_addr.sdl_index = ifp->if_index;
1907 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1909 IF_ADDR_WUNLOCK(scifp);
1910 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1911 error = if_addmulti(ifp,
1912 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1920 lagg_clrmulti(struct lagg_port *lp)
1924 LAGG_XLOCK_ASSERT(lp->lp_softc);
1925 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1926 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1927 if (mc->mc_ifma && lp->lp_detaching == 0)
1928 if_delmulti_ifma(mc->mc_ifma);
1935 lagg_setcaps(struct lagg_port *lp, int cap)
1939 if (lp->lp_ifp->if_capenable == cap)
1941 if (lp->lp_ioctl == NULL)
1943 ifr.ifr_reqcap = cap;
1944 return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1947 /* Handle a ref counted flag that should be set on the lagg port as well */
1949 lagg_setflag(struct lagg_port *lp, int flag, int status,
1950 int (*func)(struct ifnet *, int))
1952 struct lagg_softc *sc = lp->lp_softc;
1953 struct ifnet *scifp = sc->sc_ifp;
1954 struct ifnet *ifp = lp->lp_ifp;
1957 LAGG_XLOCK_ASSERT(sc);
1959 status = status ? (scifp->if_flags & flag) : 0;
1960 /* Now "status" contains the flag value or 0 */
1963 * See if recorded ports status is different from what
1964 * we want it to be. If it is, flip it. We record ports
1965 * status in lp_ifflags so that we won't clear ports flag
1966 * we haven't set. In fact, we don't clear or set ports
1967 * flags directly, but get or release references to them.
1968 * That's why we can be sure that recorded flags still are
1969 * in accord with actual ports flags.
1971 if (status != (lp->lp_ifflags & flag)) {
1972 error = (*func)(ifp, status);
1975 lp->lp_ifflags &= ~flag;
1976 lp->lp_ifflags |= status;
1982 * Handle IFF_* flags that require certain changes on the lagg port
1983 * if "status" is true, update ports flags respective to the lagg
1984 * if "status" is false, forcedly clear the flags set on port.
1987 lagg_setflags(struct lagg_port *lp, int status)
1991 for (i = 0; lagg_pflags[i].flag; i++) {
1992 error = lagg_setflag(lp, lagg_pflags[i].flag,
1993 status, lagg_pflags[i].func);
2001 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2003 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2006 #if defined(KERN_TLS) || defined(RATELIMIT)
2007 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2008 MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2011 /* We need a Tx algorithm and at least one port */
2012 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2015 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2019 ETHER_BPF_MTAP(ifp, m);
2021 error = lagg_proto_start(sc, m);
2027 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2029 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2032 #if defined(KERN_TLS) || defined(RATELIMIT)
2033 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2034 MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2037 /* We need a Tx algorithm and at least one port */
2038 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2041 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2045 INFINIBAND_BPF_MTAP(ifp, m);
2047 error = lagg_proto_start(sc, m);
2053 * The ifp->if_qflush entry point for lagg(4) is no-op.
2056 lagg_qflush(struct ifnet *ifp __unused)
2060 static struct mbuf *
2061 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2063 struct lagg_port *lp = ifp->if_lagg;
2064 struct lagg_softc *sc = lp->lp_softc;
2065 struct ifnet *scifp = sc->sc_ifp;
2068 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2069 lp->lp_detaching != 0 ||
2070 sc->sc_proto == LAGG_PROTO_NONE) {
2076 ETHER_BPF_MTAP(scifp, m);
2078 m = lagg_proto_input(sc, lp, m);
2079 if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2088 static struct mbuf *
2089 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2091 struct lagg_port *lp = ifp->if_lagg;
2092 struct lagg_softc *sc = lp->lp_softc;
2093 struct ifnet *scifp = sc->sc_ifp;
2096 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2097 lp->lp_detaching != 0 ||
2098 sc->sc_proto == LAGG_PROTO_NONE) {
2104 INFINIBAND_BPF_MTAP(scifp, m);
2106 m = lagg_proto_input(sc, lp, m);
2107 if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
2117 lagg_media_change(struct ifnet *ifp)
2119 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2121 if (sc->sc_ifflags & IFF_DEBUG)
2122 printf("%s\n", __func__);
2129 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2131 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2132 struct lagg_port *lp;
2134 imr->ifm_status = IFM_AVALID;
2135 imr->ifm_active = IFM_ETHER | IFM_AUTO;
2138 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2139 if (LAGG_PORTACTIVE(lp))
2140 imr->ifm_status |= IFM_ACTIVE;
2146 lagg_linkstate(struct lagg_softc *sc)
2148 struct lagg_port *lp;
2149 int new_link = LINK_STATE_DOWN;
2152 LAGG_XLOCK_ASSERT(sc);
2154 /* LACP handles link state itself */
2155 if (sc->sc_proto == LAGG_PROTO_LACP)
2158 /* Our link is considered up if at least one of our ports is active */
2160 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2161 if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2162 new_link = LINK_STATE_UP;
2167 if_link_state_change(sc->sc_ifp, new_link);
2169 /* Update if_baudrate to reflect the max possible speed */
2170 switch (sc->sc_proto) {
2171 case LAGG_PROTO_FAILOVER:
2172 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2173 sc->sc_primary->lp_ifp->if_baudrate : 0;
2175 case LAGG_PROTO_ROUNDROBIN:
2176 case LAGG_PROTO_LOADBALANCE:
2177 case LAGG_PROTO_BROADCAST:
2180 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2181 speed += lp->lp_ifp->if_baudrate;
2183 sc->sc_ifp->if_baudrate = speed;
2185 case LAGG_PROTO_LACP:
2186 /* LACP updates if_baudrate itself */
2192 lagg_port_state(struct ifnet *ifp, int state)
2194 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2195 struct lagg_softc *sc = NULL;
2204 lagg_proto_linkstate(sc, lp);
2209 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2211 struct lagg_port *lp_next, *rval = NULL;
2214 * Search a port which reports an active link state.
2219 * This is called with either LAGG_RLOCK() held or
2220 * LAGG_XLOCK(sc) held.
2222 if (!in_epoch(net_epoch_preempt))
2223 LAGG_XLOCK_ASSERT(sc);
2228 if (LAGG_PORTACTIVE(lp)) {
2232 if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2233 LAGG_PORTACTIVE(lp_next)) {
2239 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2240 if (LAGG_PORTACTIVE(lp_next)) {
2249 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2252 #if defined(KERN_TLS) || defined(RATELIMIT)
2253 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2254 struct lagg_snd_tag *lst;
2255 struct m_snd_tag *mst;
2257 mst = m->m_pkthdr.snd_tag;
2258 lst = mst_to_lst(mst);
2259 if (lst->tag->ifp != ifp) {
2263 m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2264 m_snd_tag_rele(mst);
2267 return (ifp->if_transmit)(ifp, m);
2271 * Simple round robin aggregation
2274 lagg_rr_attach(struct lagg_softc *sc)
2281 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2283 struct lagg_port *lp;
2286 p = atomic_fetchadd_32(&sc->sc_seq, 1);
2289 lp = CK_SLIST_FIRST(&sc->sc_ports);
2292 lp = CK_SLIST_NEXT(lp, lp_entries);
2295 * Check the port's link state. This will return the next active
2296 * port if the link is down or the port is NULL.
2298 if ((lp = lagg_link_active(sc, lp)) == NULL) {
2299 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2305 return (lagg_enqueue(lp->lp_ifp, m));
2308 static struct mbuf *
2309 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2311 struct ifnet *ifp = sc->sc_ifp;
2313 /* Just pass in the packet to our lagg device */
2314 m->m_pkthdr.rcvif = ifp;
2323 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2325 int active_ports = 0;
2328 struct lagg_port *lp, *last = NULL;
2331 LAGG_RLOCK_ASSERT();
2332 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2333 if (!LAGG_PORTACTIVE(lp))
2339 m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2345 lagg_enqueue(last->lp_ifp, m0);
2351 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2355 if ((last = lagg_link_active(sc, last)) == NULL) {
2357 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2362 ret = lagg_enqueue(last->lp_ifp, m);
2364 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2370 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2372 struct ifnet *ifp = sc->sc_ifp;
2374 /* Just pass in the packet to our lagg device */
2375 m->m_pkthdr.rcvif = ifp;
2383 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2385 struct lagg_port *lp;
2387 /* Use the master port if active or the next available port */
2388 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2389 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2395 return (lagg_enqueue(lp->lp_ifp, m));
2398 static struct mbuf *
2399 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2401 struct ifnet *ifp = sc->sc_ifp;
2402 struct lagg_port *tmp_tp;
2404 if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2405 m->m_pkthdr.rcvif = ifp;
2409 if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2410 tmp_tp = lagg_link_active(sc, sc->sc_primary);
2412 * If tmp_tp is null, we've received a packet when all
2413 * our links are down. Weird, but process it anyways.
2415 if ((tmp_tp == NULL || tmp_tp == lp)) {
2416 m->m_pkthdr.rcvif = ifp;
2429 lagg_lb_attach(struct lagg_softc *sc)
2431 struct lagg_port *lp;
2434 LAGG_XLOCK_ASSERT(sc);
2435 lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2436 lb->lb_key = m_ether_tcpip_hash_init();
2439 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2440 lagg_lb_port_create(lp);
2444 lagg_lb_detach(struct lagg_softc *sc)
2448 lb = (struct lagg_lb *)sc->sc_psc;
2454 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2456 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2457 struct lagg_port *lp_next;
2461 bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2462 LAGG_XLOCK_ASSERT(sc);
2463 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2466 if (i >= LAGG_MAX_PORTS) {
2470 if (sc->sc_ifflags & IFF_DEBUG)
2471 printf("%s: port %s at index %d\n",
2472 sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2473 lb->lb_ports[i++] = lp_next;
2480 lagg_lb_port_create(struct lagg_port *lp)
2482 struct lagg_softc *sc = lp->lp_softc;
2483 return (lagg_lb_porttable(sc, NULL));
2487 lagg_lb_port_destroy(struct lagg_port *lp)
2489 struct lagg_softc *sc = lp->lp_softc;
2490 lagg_lb_porttable(sc, lp);
2494 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2496 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2497 struct lagg_port *lp = NULL;
2500 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2501 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2502 p = m->m_pkthdr.flowid >> sc->flowid_shift;
2504 p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2506 lp = lb->lb_ports[p];
2509 * Check the port's link state. This will return the next active
2510 * port if the link is down or the port is NULL.
2512 if ((lp = lagg_link_active(sc, lp)) == NULL) {
2513 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2519 return (lagg_enqueue(lp->lp_ifp, m));
2522 static struct mbuf *
2523 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2525 struct ifnet *ifp = sc->sc_ifp;
2527 /* Just pass in the packet to our lagg device */
2528 m->m_pkthdr.rcvif = ifp;
2537 lagg_lacp_attach(struct lagg_softc *sc)
2539 struct lagg_port *lp;
2542 LAGG_XLOCK_ASSERT(sc);
2543 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2544 lacp_port_create(lp);
2548 lagg_lacp_detach(struct lagg_softc *sc)
2550 struct lagg_port *lp;
2553 LAGG_XLOCK_ASSERT(sc);
2554 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2555 lacp_port_destroy(lp);
2563 lagg_lacp_lladdr(struct lagg_softc *sc)
2565 struct lagg_port *lp;
2567 LAGG_SXLOCK_ASSERT(sc);
2569 /* purge all the lacp ports */
2570 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2571 lacp_port_destroy(lp);
2573 /* add them back in */
2574 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2575 lacp_port_create(lp);
2579 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2581 struct lagg_port *lp;
2583 lp = lacp_select_tx_port(sc, m);
2585 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2591 return (lagg_enqueue(lp->lp_ifp, m));
2594 static struct mbuf *
2595 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2597 struct ifnet *ifp = sc->sc_ifp;
2598 struct ether_header *eh;
2601 eh = mtod(m, struct ether_header *);
2602 etype = ntohs(eh->ether_type);
2604 /* Tap off LACP control messages */
2605 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2606 m = lacp_input(lp, m);
2612 * If the port is not collecting or not in the active aggregator then
2615 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2620 m->m_pkthdr.rcvif = ifp;