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>
59 #if defined(INET) || defined(INET6)
60 #include <netinet/in.h>
61 #include <netinet/ip.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/if_ether.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet6/in6_ifattach.h>
74 #include <net/if_vlan_var.h>
75 #include <net/if_lagg.h>
76 #include <net/ieee8023ad_lacp.h>
80 * XXX: declare here to avoid to include many inet6 related files..
81 * should be more generalized?
83 extern void nd6_setmtu(struct ifnet *);
86 #define LAGG_RLOCK() struct epoch_tracker lagg_et; epoch_enter_preempt(net_epoch_preempt, &lagg_et)
87 #define LAGG_RUNLOCK() epoch_exit_preempt(net_epoch_preempt, &lagg_et)
88 #define LAGG_RLOCK_ASSERT() NET_EPOCH_ASSERT()
89 #define LAGG_UNLOCK_ASSERT() MPASS(!in_epoch(net_epoch_preempt))
91 #define LAGG_SX_INIT(_sc) sx_init(&(_sc)->sc_sx, "if_lagg sx")
92 #define LAGG_SX_DESTROY(_sc) sx_destroy(&(_sc)->sc_sx)
93 #define LAGG_XLOCK(_sc) sx_xlock(&(_sc)->sc_sx)
94 #define LAGG_XUNLOCK(_sc) sx_xunlock(&(_sc)->sc_sx)
95 #define LAGG_SXLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_LOCKED)
96 #define LAGG_XLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
98 /* Special flags we should propagate to the lagg ports. */
101 int (*func)(struct ifnet *, int);
103 {IFF_PROMISC, ifpromisc},
104 {IFF_ALLMULTI, if_allmulti},
108 struct lagg_snd_tag {
109 struct m_snd_tag com;
110 struct m_snd_tag *tag;
113 VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
114 #define V_lagg_list VNET(lagg_list)
115 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
116 #define V_lagg_list_mtx VNET(lagg_list_mtx)
117 #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \
118 "if_lagg list", NULL, MTX_DEF)
119 #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx)
120 #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx)
121 #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx)
122 eventhandler_tag lagg_detach_cookie = NULL;
124 static int lagg_clone_create(struct if_clone *, int, caddr_t);
125 static void lagg_clone_destroy(struct ifnet *);
126 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
127 #define V_lagg_cloner VNET(lagg_cloner)
128 static const char laggname[] = "lagg";
129 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
131 static void lagg_capabilities(struct lagg_softc *);
132 static int lagg_port_create(struct lagg_softc *, struct ifnet *);
133 static int lagg_port_destroy(struct lagg_port *, int);
134 static struct mbuf *lagg_input(struct ifnet *, struct mbuf *);
135 static void lagg_linkstate(struct lagg_softc *);
136 static void lagg_port_state(struct ifnet *, int);
137 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
138 static int lagg_port_output(struct ifnet *, struct mbuf *,
139 const struct sockaddr *, struct route *);
140 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *);
141 #ifdef LAGG_PORT_STACKING
142 static int lagg_port_checkstacking(struct lagg_softc *);
144 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *);
145 static void lagg_init(void *);
146 static void lagg_stop(struct lagg_softc *);
147 static int lagg_ioctl(struct ifnet *, u_long, caddr_t);
148 #if defined(KERN_TLS) || defined(RATELIMIT)
149 static int lagg_snd_tag_alloc(struct ifnet *,
150 union if_snd_tag_alloc_params *,
151 struct m_snd_tag **);
152 static int lagg_snd_tag_modify(struct m_snd_tag *,
153 union if_snd_tag_modify_params *);
154 static int lagg_snd_tag_query(struct m_snd_tag *,
155 union if_snd_tag_query_params *);
156 static void lagg_snd_tag_free(struct m_snd_tag *);
157 static void lagg_ratelimit_query(struct ifnet *,
158 struct if_ratelimit_query_results *);
160 static int lagg_setmulti(struct lagg_port *);
161 static int lagg_clrmulti(struct lagg_port *);
162 static int lagg_setcaps(struct lagg_port *, int cap);
163 static int lagg_setflag(struct lagg_port *, int, int,
164 int (*func)(struct ifnet *, int));
165 static int lagg_setflags(struct lagg_port *, int status);
166 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
167 static int lagg_transmit(struct ifnet *, struct mbuf *);
168 static void lagg_qflush(struct ifnet *);
169 static int lagg_media_change(struct ifnet *);
170 static void lagg_media_status(struct ifnet *, struct ifmediareq *);
171 static struct lagg_port *lagg_link_active(struct lagg_softc *,
174 /* Simple round robin */
175 static void lagg_rr_attach(struct lagg_softc *);
176 static int lagg_rr_start(struct lagg_softc *, struct mbuf *);
177 static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *,
180 /* Active failover */
181 static int lagg_fail_start(struct lagg_softc *, struct mbuf *);
182 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
186 static void lagg_lb_attach(struct lagg_softc *);
187 static void lagg_lb_detach(struct lagg_softc *);
188 static int lagg_lb_port_create(struct lagg_port *);
189 static void lagg_lb_port_destroy(struct lagg_port *);
190 static int lagg_lb_start(struct lagg_softc *, struct mbuf *);
191 static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *,
193 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
196 static int lagg_bcast_start(struct lagg_softc *, struct mbuf *);
197 static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *,
201 static void lagg_lacp_attach(struct lagg_softc *);
202 static void lagg_lacp_detach(struct lagg_softc *);
203 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *);
204 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
206 static void lagg_lacp_lladdr(struct lagg_softc *);
208 /* lagg protocol table */
209 static const struct lagg_proto {
211 void (*pr_attach)(struct lagg_softc *);
212 void (*pr_detach)(struct lagg_softc *);
213 int (*pr_start)(struct lagg_softc *, struct mbuf *);
214 struct mbuf * (*pr_input)(struct lagg_softc *, struct lagg_port *,
216 int (*pr_addport)(struct lagg_port *);
217 void (*pr_delport)(struct lagg_port *);
218 void (*pr_linkstate)(struct lagg_port *);
219 void (*pr_init)(struct lagg_softc *);
220 void (*pr_stop)(struct lagg_softc *);
221 void (*pr_lladdr)(struct lagg_softc *);
222 void (*pr_request)(struct lagg_softc *, void *);
223 void (*pr_portreq)(struct lagg_port *, void *);
226 .pr_num = LAGG_PROTO_NONE
229 .pr_num = LAGG_PROTO_ROUNDROBIN,
230 .pr_attach = lagg_rr_attach,
231 .pr_start = lagg_rr_start,
232 .pr_input = lagg_rr_input,
235 .pr_num = LAGG_PROTO_FAILOVER,
236 .pr_start = lagg_fail_start,
237 .pr_input = lagg_fail_input,
240 .pr_num = LAGG_PROTO_LOADBALANCE,
241 .pr_attach = lagg_lb_attach,
242 .pr_detach = lagg_lb_detach,
243 .pr_start = lagg_lb_start,
244 .pr_input = lagg_lb_input,
245 .pr_addport = lagg_lb_port_create,
246 .pr_delport = lagg_lb_port_destroy,
249 .pr_num = LAGG_PROTO_LACP,
250 .pr_attach = lagg_lacp_attach,
251 .pr_detach = lagg_lacp_detach,
252 .pr_start = lagg_lacp_start,
253 .pr_input = lagg_lacp_input,
254 .pr_addport = lacp_port_create,
255 .pr_delport = lacp_port_destroy,
256 .pr_linkstate = lacp_linkstate,
257 .pr_init = lacp_init,
258 .pr_stop = lacp_stop,
259 .pr_lladdr = lagg_lacp_lladdr,
260 .pr_request = lacp_req,
261 .pr_portreq = lacp_portreq,
264 .pr_num = LAGG_PROTO_BROADCAST,
265 .pr_start = lagg_bcast_start,
266 .pr_input = lagg_bcast_input,
270 SYSCTL_DECL(_net_link);
271 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
274 /* Allow input on any failover links */
275 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
276 #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all)
277 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
278 &VNET_NAME(lagg_failover_rx_all), 0,
279 "Accept input from any interface in a failover lagg");
281 /* Default value for using flowid */
282 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
283 #define V_def_use_flowid VNET(def_use_flowid)
284 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN,
285 &VNET_NAME(def_use_flowid), 0,
286 "Default setting for using flow id for load sharing");
288 /* Default value for using numa */
289 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
290 #define V_def_use_numa VNET(def_use_numa)
291 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN,
292 &VNET_NAME(def_use_numa), 0,
293 "Use numa to steer flows");
295 /* Default value for flowid shift */
296 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
297 #define V_def_flowid_shift VNET(def_flowid_shift)
298 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN,
299 &VNET_NAME(def_flowid_shift), 0,
300 "Default setting for flowid shift for load sharing");
303 vnet_lagg_init(const void *unused __unused)
306 LAGG_LIST_LOCK_INIT();
307 SLIST_INIT(&V_lagg_list);
308 V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create,
309 lagg_clone_destroy, 0);
311 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
312 vnet_lagg_init, NULL);
315 vnet_lagg_uninit(const void *unused __unused)
318 if_clone_detach(V_lagg_cloner);
319 LAGG_LIST_LOCK_DESTROY();
321 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
322 vnet_lagg_uninit, NULL);
325 lagg_modevent(module_t mod, int type, void *data)
330 lagg_input_p = lagg_input;
331 lagg_linkstate_p = lagg_port_state;
332 lagg_detach_cookie = EVENTHANDLER_REGISTER(
333 ifnet_departure_event, lagg_port_ifdetach, NULL,
334 EVENTHANDLER_PRI_ANY);
337 EVENTHANDLER_DEREGISTER(ifnet_departure_event,
340 lagg_linkstate_p = NULL;
348 static moduledata_t lagg_mod = {
354 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
355 MODULE_VERSION(if_lagg, 1);
358 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
361 LAGG_XLOCK_ASSERT(sc);
362 KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
365 if (sc->sc_ifflags & IFF_DEBUG)
366 if_printf(sc->sc_ifp, "using proto %u\n", pr);
368 if (lagg_protos[pr].pr_attach != NULL)
369 lagg_protos[pr].pr_attach(sc);
374 lagg_proto_detach(struct lagg_softc *sc)
378 LAGG_XLOCK_ASSERT(sc);
380 sc->sc_proto = LAGG_PROTO_NONE;
382 if (lagg_protos[pr].pr_detach != NULL)
383 lagg_protos[pr].pr_detach(sc);
387 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
390 return (lagg_protos[sc->sc_proto].pr_start(sc, m));
394 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
397 return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
401 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
404 if (lagg_protos[sc->sc_proto].pr_addport == NULL)
407 return (lagg_protos[sc->sc_proto].pr_addport(lp));
411 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
414 if (lagg_protos[sc->sc_proto].pr_delport != NULL)
415 lagg_protos[sc->sc_proto].pr_delport(lp);
419 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
422 if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
423 lagg_protos[sc->sc_proto].pr_linkstate(lp);
427 lagg_proto_init(struct lagg_softc *sc)
430 if (lagg_protos[sc->sc_proto].pr_init != NULL)
431 lagg_protos[sc->sc_proto].pr_init(sc);
435 lagg_proto_stop(struct lagg_softc *sc)
438 if (lagg_protos[sc->sc_proto].pr_stop != NULL)
439 lagg_protos[sc->sc_proto].pr_stop(sc);
443 lagg_proto_lladdr(struct lagg_softc *sc)
446 if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
447 lagg_protos[sc->sc_proto].pr_lladdr(sc);
451 lagg_proto_request(struct lagg_softc *sc, void *v)
454 if (lagg_protos[sc->sc_proto].pr_request != NULL)
455 lagg_protos[sc->sc_proto].pr_request(sc, v);
459 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
462 if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
463 lagg_protos[sc->sc_proto].pr_portreq(lp, v);
467 * This routine is run via an vlan
471 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
473 struct lagg_softc *sc = ifp->if_softc;
474 struct lagg_port *lp;
476 if (ifp->if_softc != arg) /* Not our event */
480 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
481 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
486 * This routine is run via an vlan
490 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
492 struct lagg_softc *sc = ifp->if_softc;
493 struct lagg_port *lp;
495 if (ifp->if_softc != arg) /* Not our event */
499 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
500 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
505 lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
507 struct lagg_softc *sc;
509 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */
511 sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO);
512 ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
520 if (V_def_use_flowid)
521 sc->sc_opts |= LAGG_OPT_USE_FLOWID;
523 sc->sc_opts |= LAGG_OPT_USE_NUMA;
524 sc->flowid_shift = V_def_flowid_shift;
526 /* Hash all layers by default */
527 sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4;
529 lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
531 CK_SLIST_INIT(&sc->sc_ports);
533 /* Initialise pseudo media types */
534 ifmedia_init(&sc->sc_media, 0, lagg_media_change,
536 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
537 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
539 if_initname(ifp, laggname, unit);
541 ifp->if_transmit = lagg_transmit;
542 ifp->if_qflush = lagg_qflush;
543 ifp->if_init = lagg_init;
544 ifp->if_ioctl = lagg_ioctl;
545 ifp->if_get_counter = lagg_get_counter;
546 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
547 #if defined(KERN_TLS) || defined(RATELIMIT)
548 ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
549 ifp->if_snd_tag_modify = lagg_snd_tag_modify;
550 ifp->if_snd_tag_query = lagg_snd_tag_query;
551 ifp->if_snd_tag_free = lagg_snd_tag_free;
552 ifp->if_ratelimit_query = lagg_ratelimit_query;
554 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
557 * Attach as an ordinary ethernet device, children will be attached
558 * as special device IFT_IEEE8023ADLAG.
560 ether_ifattach(ifp, eaddr);
562 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
563 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
564 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
565 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
567 /* Insert into the global list of laggs */
569 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
577 lagg_clone_destroy(struct ifnet *ifp)
579 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
580 struct lagg_port *lp;
583 sc->sc_destroying = 1;
585 ifp->if_flags &= ~IFF_UP;
587 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
588 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
590 /* Shutdown and remove lagg ports */
591 while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
592 lagg_port_destroy(lp, 1);
594 /* Unhook the aggregation protocol */
595 lagg_proto_detach(sc);
598 ifmedia_removeall(&sc->sc_media);
603 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
611 lagg_capabilities(struct lagg_softc *sc)
613 struct lagg_port *lp;
616 struct ifnet_hw_tsomax hw_tsomax;
618 LAGG_XLOCK_ASSERT(sc);
620 /* Get common enabled capabilities for the lagg ports */
622 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
623 ena &= lp->lp_ifp->if_capenable;
624 ena = (ena == ~0 ? 0 : ena);
627 * Apply common enabled capabilities back to the lagg ports.
628 * May require several iterations if they are dependent.
632 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
633 lagg_setcaps(lp, ena);
634 ena &= lp->lp_ifp->if_capenable;
636 } while (pena != ena);
638 /* Get other capabilities from the lagg ports */
641 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
642 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
643 cap &= lp->lp_ifp->if_capabilities;
644 hwa &= lp->lp_ifp->if_hwassist;
645 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
647 cap = (cap == ~0 ? 0 : cap);
648 hwa = (hwa == ~(uint64_t)0 ? 0 : hwa);
650 if (sc->sc_ifp->if_capabilities != cap ||
651 sc->sc_ifp->if_capenable != ena ||
652 sc->sc_ifp->if_hwassist != hwa ||
653 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
654 sc->sc_ifp->if_capabilities = cap;
655 sc->sc_ifp->if_capenable = ena;
656 sc->sc_ifp->if_hwassist = hwa;
657 getmicrotime(&sc->sc_ifp->if_lastchange);
659 if (sc->sc_ifflags & IFF_DEBUG)
660 if_printf(sc->sc_ifp,
661 "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
666 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
668 struct lagg_softc *sc_ptr;
669 struct lagg_port *lp, *tlp;
671 int error, i, oldmtu;
674 LAGG_XLOCK_ASSERT(sc);
676 if (sc->sc_ifp == ifp) {
677 if_printf(sc->sc_ifp,
678 "cannot add a lagg to itself as a port\n");
682 /* Limit the maximal number of lagg ports */
683 if (sc->sc_count >= LAGG_MAX_PORTS)
686 /* Check if port has already been associated to a lagg */
687 if (ifp->if_lagg != NULL) {
688 /* Port is already in the current lagg? */
689 lp = (struct lagg_port *)ifp->if_lagg;
690 if (lp->lp_softc == sc)
695 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */
696 if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
697 return (EPROTONOSUPPORT);
699 /* Allow the first Ethernet member to define the MTU */
701 if (CK_SLIST_EMPTY(&sc->sc_ports)) {
702 sc->sc_ifp->if_mtu = ifp->if_mtu;
703 } else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
704 if (ifp->if_ioctl == NULL) {
705 if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
709 oldmtu = ifp->if_mtu;
710 strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
711 ifr.ifr_mtu = sc->sc_ifp->if_mtu;
712 error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
714 if_printf(sc->sc_ifp, "invalid MTU for %s\n",
718 ifr.ifr_mtu = oldmtu;
721 lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO);
724 /* Check if port is a stacked lagg */
726 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
727 if (ifp == sc_ptr->sc_ifp) {
731 (*ifp->if_ioctl)(ifp, SIOCSIFMTU,
734 /* XXX disable stacking for the moment, its untested */
735 #ifdef LAGG_PORT_STACKING
736 lp->lp_flags |= LAGG_PORT_STACK;
737 if (lagg_port_checkstacking(sc_ptr) >=
742 (*ifp->if_ioctl)(ifp, SIOCSIFMTU,
754 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
755 lp->lp_ifcapenable = ifp->if_capenable;
756 if (CK_SLIST_EMPTY(&sc->sc_ports)) {
757 bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
758 lagg_proto_lladdr(sc);
759 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
761 if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
763 lagg_setflags(lp, 1);
765 if (CK_SLIST_EMPTY(&sc->sc_ports))
768 /* Change the interface type */
769 lp->lp_iftype = ifp->if_type;
770 ifp->if_type = IFT_IEEE8023ADLAG;
772 lp->lp_ioctl = ifp->if_ioctl;
773 ifp->if_ioctl = lagg_port_ioctl;
774 lp->lp_output = ifp->if_output;
775 ifp->if_output = lagg_port_output;
777 /* Read port counters */
778 pval = lp->port_counters.val;
779 for (i = 0; i < IFCOUNTERS; i++, pval++)
780 *pval = ifp->if_get_counter(ifp, i);
783 * Insert into the list of ports.
784 * Keep ports sorted by if_index. It is handy, when configuration
785 * is predictable and `ifconfig laggN create ...` command
786 * will lead to the same result each time.
788 CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
789 if (tlp->lp_ifp->if_index < ifp->if_index && (
790 CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
791 ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
796 CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
798 CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
804 if ((error = lagg_proto_addport(sc, lp)) != 0) {
805 /* Remove the port, without calling pr_delport. */
806 lagg_port_destroy(lp, 0);
808 (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
812 /* Update lagg capabilities */
813 lagg_capabilities(sc);
819 #ifdef LAGG_PORT_STACKING
821 lagg_port_checkstacking(struct lagg_softc *sc)
823 struct lagg_softc *sc_ptr;
824 struct lagg_port *lp;
827 LAGG_SXLOCK_ASSERT(sc);
828 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
829 if (lp->lp_flags & LAGG_PORT_STACK) {
830 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
831 m = MAX(m, lagg_port_checkstacking(sc_ptr));
840 lagg_port_destroy_cb(epoch_context_t ec)
842 struct lagg_port *lp;
845 lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
853 lagg_port_destroy(struct lagg_port *lp, int rundelport)
855 struct lagg_softc *sc = lp->lp_softc;
856 struct lagg_port *lp_ptr, *lp0;
857 struct ifnet *ifp = lp->lp_ifp;
858 uint64_t *pval, vdiff;
861 LAGG_XLOCK_ASSERT(sc);
864 lagg_proto_delport(sc, lp);
866 if (lp->lp_detaching == 0)
869 /* Restore interface */
870 ifp->if_type = lp->lp_iftype;
871 ifp->if_ioctl = lp->lp_ioctl;
872 ifp->if_output = lp->lp_output;
875 /* Update detached port counters */
876 pval = lp->port_counters.val;
877 for (i = 0; i < IFCOUNTERS; i++, pval++) {
878 vdiff = ifp->if_get_counter(ifp, i) - *pval;
879 sc->detached_counters.val[i] += vdiff;
882 /* Finally, remove the port from the lagg */
883 CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
886 /* Update the primary interface */
887 if (lp == sc->sc_primary) {
888 uint8_t lladdr[ETHER_ADDR_LEN];
890 if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
891 bzero(&lladdr, ETHER_ADDR_LEN);
893 bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN);
894 sc->sc_primary = lp0;
895 if (sc->sc_destroying == 0) {
896 bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
897 lagg_proto_lladdr(sc);
898 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
902 * Update lladdr for each port (new primary needs update
903 * as well, to switch from old lladdr to its 'real' one)
905 CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
906 if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN);
910 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
912 if (lp->lp_detaching == 0) {
913 lagg_setflags(lp, 0);
914 lagg_setcaps(lp, lp->lp_ifcapenable);
915 if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN);
919 * free port and release it's ifnet reference after a grace period has
922 NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
923 /* Update lagg capabilities */
924 lagg_capabilities(sc);
931 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
933 struct lagg_reqport *rp = (struct lagg_reqport *)data;
934 struct lagg_softc *sc;
935 struct lagg_port *lp = NULL;
938 /* Should be checked by the caller */
939 if (ifp->if_type != IFT_IEEE8023ADLAG ||
940 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
945 if (rp->rp_portname[0] == '\0' ||
946 ifunit(rp->rp_portname) != ifp) {
952 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
958 lagg_port2req(lp, rp);
963 if (lp->lp_ioctl == NULL) {
967 error = (*lp->lp_ioctl)(ifp, cmd, data);
971 /* Update lagg interface capabilities */
973 lagg_capabilities(sc);
975 VLAN_CAPABILITIES(sc->sc_ifp);
979 /* Do not allow the MTU to be changed once joined */
990 if (lp != NULL && lp->lp_ioctl != NULL)
991 return ((*lp->lp_ioctl)(ifp, cmd, data));
997 * Requests counter @cnt data.
999 * Counter value is calculated the following way:
1000 * 1) for each port, sum difference between current and "initial" measurements.
1001 * 2) add lagg logical interface counters.
1002 * 3) add data from detached_counters array.
1004 * We also do the following things on ports attach/detach:
1005 * 1) On port attach we store all counters it has into port_counter array.
1006 * 2) On port detach we add the different between "initial" and
1007 * current counters data to detached_counters array.
1010 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1012 struct lagg_softc *sc;
1013 struct lagg_port *lp;
1014 struct ifnet *lpifp;
1015 uint64_t newval, oldval, vsum;
1017 /* Revise this when we've got non-generic counters. */
1018 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1020 sc = (struct lagg_softc *)ifp->if_softc;
1024 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1025 /* Saved attached value */
1026 oldval = lp->port_counters.val[cnt];
1029 newval = lpifp->if_get_counter(lpifp, cnt);
1030 /* Calculate diff and save new */
1031 vsum += newval - oldval;
1036 * Add counter data which might be added by upper
1037 * layer protocols operating on logical interface.
1039 vsum += if_get_counter_default(ifp, cnt);
1042 * Add counter data from detached ports counters
1044 vsum += sc->detached_counters.val[cnt];
1051 * For direct output to child ports.
1054 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1055 const struct sockaddr *dst, struct route *ro)
1057 struct lagg_port *lp = ifp->if_lagg;
1059 switch (dst->sa_family) {
1060 case pseudo_AF_HDRCMPLT:
1062 return ((*lp->lp_output)(ifp, m, dst, ro));
1065 /* drop any other frames */
1071 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1073 struct lagg_port *lp;
1074 struct lagg_softc *sc;
1076 if ((lp = ifp->if_lagg) == NULL)
1078 /* If the ifnet is just being renamed, don't do anything. */
1079 if (ifp->if_flags & IFF_RENAMING)
1085 lp->lp_detaching = 1;
1086 lagg_port_destroy(lp, 1);
1088 VLAN_CAPABILITIES(sc->sc_ifp);
1092 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1094 struct lagg_softc *sc = lp->lp_softc;
1096 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1097 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1098 rp->rp_prio = lp->lp_prio;
1099 rp->rp_flags = lp->lp_flags;
1100 lagg_proto_portreq(sc, lp, &rp->rp_psc);
1102 /* Add protocol specific flags */
1103 switch (sc->sc_proto) {
1104 case LAGG_PROTO_FAILOVER:
1105 if (lp == sc->sc_primary)
1106 rp->rp_flags |= LAGG_PORT_MASTER;
1107 if (lp == lagg_link_active(sc, sc->sc_primary))
1108 rp->rp_flags |= LAGG_PORT_ACTIVE;
1111 case LAGG_PROTO_ROUNDROBIN:
1112 case LAGG_PROTO_LOADBALANCE:
1113 case LAGG_PROTO_BROADCAST:
1114 if (LAGG_PORTACTIVE(lp))
1115 rp->rp_flags |= LAGG_PORT_ACTIVE;
1118 case LAGG_PROTO_LACP:
1119 /* LACP has a different definition of active */
1120 if (lacp_isactive(lp))
1121 rp->rp_flags |= LAGG_PORT_ACTIVE;
1122 if (lacp_iscollecting(lp))
1123 rp->rp_flags |= LAGG_PORT_COLLECTING;
1124 if (lacp_isdistributing(lp))
1125 rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1132 lagg_init(void *xsc)
1134 struct lagg_softc *sc = (struct lagg_softc *)xsc;
1135 struct ifnet *ifp = sc->sc_ifp;
1136 struct lagg_port *lp;
1139 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1144 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1147 * Update the port lladdrs if needed.
1148 * This might be if_setlladdr() notification
1149 * that lladdr has been changed.
1151 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1152 if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1153 ETHER_ADDR_LEN) != 0)
1154 if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN);
1157 lagg_proto_init(sc);
1163 lagg_stop(struct lagg_softc *sc)
1165 struct ifnet *ifp = sc->sc_ifp;
1167 LAGG_XLOCK_ASSERT(sc);
1169 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1172 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1174 lagg_proto_stop(sc);
1178 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1180 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1181 struct lagg_reqall *ra = (struct lagg_reqall *)data;
1182 struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1183 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1184 struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1185 struct ifreq *ifr = (struct ifreq *)data;
1186 struct lagg_port *lp;
1188 struct thread *td = curthread;
1190 int count, buflen, len, error = 0, oldmtu;
1192 bzero(&rpbuf, sizeof(rpbuf));
1197 buflen = sc->sc_count * sizeof(struct lagg_reqport);
1198 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1199 ra->ra_proto = sc->sc_proto;
1200 lagg_proto_request(sc, &ra->ra_psc);
1203 len = min(ra->ra_size, buflen);
1204 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1205 if (len < sizeof(rpbuf))
1208 lagg_port2req(lp, &rpbuf);
1209 memcpy(buf, &rpbuf, sizeof(rpbuf));
1211 buf += sizeof(rpbuf);
1212 len -= sizeof(rpbuf);
1215 ra->ra_ports = count;
1216 ra->ra_size = count * sizeof(rpbuf);
1217 error = copyout(outbuf, ra->ra_port, ra->ra_size);
1218 free(outbuf, M_TEMP);
1221 error = priv_check(td, PRIV_NET_LAGG);
1224 if (ra->ra_proto >= LAGG_PROTO_MAX) {
1225 error = EPROTONOSUPPORT;
1230 lagg_proto_detach(sc);
1231 LAGG_UNLOCK_ASSERT();
1232 lagg_proto_attach(sc, ra->ra_proto);
1237 ro->ro_opts = sc->sc_opts;
1238 if (sc->sc_proto == LAGG_PROTO_LACP) {
1239 struct lacp_softc *lsc;
1241 lsc = (struct lacp_softc *)sc->sc_psc;
1242 if (lsc->lsc_debug.lsc_tx_test != 0)
1243 ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1244 if (lsc->lsc_debug.lsc_rx_test != 0)
1245 ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1246 if (lsc->lsc_strict_mode != 0)
1247 ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1248 if (lsc->lsc_fast_timeout != 0)
1249 ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT;
1251 ro->ro_active = sc->sc_active;
1254 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1255 ro->ro_active += LAGG_PORTACTIVE(lp);
1257 ro->ro_bkt = sc->sc_stride;
1258 ro->ro_flapping = sc->sc_flapping;
1259 ro->ro_flowid_shift = sc->flowid_shift;
1263 error = priv_check(td, PRIV_NET_LAGG);
1268 * The stride option was added without defining a corresponding
1269 * LAGG_OPT flag, so handle a non-zero value before checking
1270 * anything else to preserve compatibility.
1273 if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1274 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1279 sc->sc_stride = ro->ro_bkt;
1281 if (ro->ro_opts == 0) {
1287 * Set options. LACP options are stored in sc->sc_psc,
1292 switch (ro->ro_opts) {
1293 case LAGG_OPT_USE_FLOWID:
1294 case -LAGG_OPT_USE_FLOWID:
1295 case LAGG_OPT_USE_NUMA:
1296 case -LAGG_OPT_USE_NUMA:
1297 case LAGG_OPT_FLOWIDSHIFT:
1298 case LAGG_OPT_RR_LIMIT:
1302 case LAGG_OPT_LACP_TXTEST:
1303 case -LAGG_OPT_LACP_TXTEST:
1304 case LAGG_OPT_LACP_RXTEST:
1305 case -LAGG_OPT_LACP_RXTEST:
1306 case LAGG_OPT_LACP_STRICT:
1307 case -LAGG_OPT_LACP_STRICT:
1308 case LAGG_OPT_LACP_TIMEOUT:
1309 case -LAGG_OPT_LACP_TIMEOUT:
1318 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1319 /* Invalid combination of options specified. */
1322 break; /* Return from SIOCSLAGGOPTS. */
1326 * Store new options into sc->sc_opts except for
1327 * FLOWIDSHIFT, RR and LACP options.
1330 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1331 sc->flowid_shift = ro->ro_flowid_shift;
1332 else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1333 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1339 sc->sc_stride = ro->ro_bkt;
1340 } else if (ro->ro_opts > 0)
1341 sc->sc_opts |= ro->ro_opts;
1343 sc->sc_opts &= ~ro->ro_opts;
1345 struct lacp_softc *lsc;
1346 struct lacp_port *lp;
1348 lsc = (struct lacp_softc *)sc->sc_psc;
1350 switch (ro->ro_opts) {
1351 case LAGG_OPT_LACP_TXTEST:
1352 lsc->lsc_debug.lsc_tx_test = 1;
1354 case -LAGG_OPT_LACP_TXTEST:
1355 lsc->lsc_debug.lsc_tx_test = 0;
1357 case LAGG_OPT_LACP_RXTEST:
1358 lsc->lsc_debug.lsc_rx_test = 1;
1360 case -LAGG_OPT_LACP_RXTEST:
1361 lsc->lsc_debug.lsc_rx_test = 0;
1363 case LAGG_OPT_LACP_STRICT:
1364 lsc->lsc_strict_mode = 1;
1366 case -LAGG_OPT_LACP_STRICT:
1367 lsc->lsc_strict_mode = 0;
1369 case LAGG_OPT_LACP_TIMEOUT:
1371 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1372 lp->lp_state |= LACP_STATE_TIMEOUT;
1374 lsc->lsc_fast_timeout = 1;
1376 case -LAGG_OPT_LACP_TIMEOUT:
1378 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1379 lp->lp_state &= ~LACP_STATE_TIMEOUT;
1381 lsc->lsc_fast_timeout = 0;
1387 case SIOCGLAGGFLAGS:
1390 if (sc->sc_flags & MBUF_HASHFLAG_L2)
1391 rf->rf_flags |= LAGG_F_HASHL2;
1392 if (sc->sc_flags & MBUF_HASHFLAG_L3)
1393 rf->rf_flags |= LAGG_F_HASHL3;
1394 if (sc->sc_flags & MBUF_HASHFLAG_L4)
1395 rf->rf_flags |= LAGG_F_HASHL4;
1399 error = priv_check(td, PRIV_NET_LAGG);
1402 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1408 if (rf->rf_flags & LAGG_F_HASHL2)
1409 sc->sc_flags |= MBUF_HASHFLAG_L2;
1410 if (rf->rf_flags & LAGG_F_HASHL3)
1411 sc->sc_flags |= MBUF_HASHFLAG_L3;
1412 if (rf->rf_flags & LAGG_F_HASHL4)
1413 sc->sc_flags |= MBUF_HASHFLAG_L4;
1417 if (rp->rp_portname[0] == '\0' ||
1418 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1424 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1425 lp->lp_softc != sc) {
1432 lagg_port2req(lp, rp);
1437 error = priv_check(td, PRIV_NET_LAGG);
1440 if (rp->rp_portname[0] == '\0' ||
1441 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1447 * A laggport interface should not have inet6 address
1448 * because two interfaces with a valid link-local
1449 * scope zone must not be merged in any form. This
1450 * restriction is needed to prevent violation of
1451 * link-local scope zone. Attempts to add a laggport
1452 * interface which has inet6 addresses triggers
1453 * removal of all inet6 addresses on the member
1456 if (in6ifa_llaonifp(tpif)) {
1458 if_printf(sc->sc_ifp,
1459 "IPv6 addresses on %s have been removed "
1460 "before adding it as a member to prevent "
1461 "IPv6 address scope violation.\n",
1465 oldmtu = ifp->if_mtu;
1467 error = lagg_port_create(sc, tpif);
1472 * LAGG MTU may change during addition of the first port.
1473 * If it did, do network layer specific procedure.
1475 if (ifp->if_mtu != oldmtu) {
1482 VLAN_CAPABILITIES(ifp);
1484 case SIOCSLAGGDELPORT:
1485 error = priv_check(td, PRIV_NET_LAGG);
1488 if (rp->rp_portname[0] == '\0' ||
1489 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1495 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1496 lp->lp_softc != sc) {
1503 error = lagg_port_destroy(lp, 1);
1506 VLAN_CAPABILITIES(ifp);
1509 /* Set flags on ports too */
1511 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1512 lagg_setflags(lp, 1);
1515 if (!(ifp->if_flags & IFF_UP) &&
1516 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1518 * If interface is marked down and it is running,
1519 * then stop and disable it.
1523 } else if ((ifp->if_flags & IFF_UP) &&
1524 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1526 * If interface is marked up and it is stopped, then
1530 (*ifp->if_init)(sc);
1537 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1546 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1551 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1552 if (lp->lp_ioctl != NULL)
1553 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1555 lagg_capabilities(sc);
1557 VLAN_CAPABILITIES(ifp);
1563 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1564 if (lp->lp_ioctl != NULL)
1565 error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1570 "failed to change MTU to %d on port %s, "
1571 "reverting all ports to original MTU (%d)\n",
1572 ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1577 ifp->if_mtu = ifr->ifr_mtu;
1579 /* set every port back to the original MTU */
1580 ifr->ifr_mtu = ifp->if_mtu;
1581 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1582 if (lp->lp_ioctl != NULL)
1583 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1590 error = ether_ioctl(ifp, cmd, data);
1596 #if defined(KERN_TLS) || defined(RATELIMIT)
1597 static inline struct lagg_snd_tag *
1598 mst_to_lst(struct m_snd_tag *mst)
1601 return (__containerof(mst, struct lagg_snd_tag, com));
1605 * Look up the port used by a specific flow. This only works for lagg
1606 * protocols with deterministic port mappings (e.g. not roundrobin).
1607 * In addition protocols which use a hash to map flows to ports must
1608 * be configured to use the mbuf flowid rather than hashing packet
1611 static struct lagg_port *
1612 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype)
1614 struct lagg_softc *sc;
1615 struct lagg_port *lp;
1621 switch (sc->sc_proto) {
1622 case LAGG_PROTO_FAILOVER:
1623 return (lagg_link_active(sc, sc->sc_primary));
1624 case LAGG_PROTO_LOADBALANCE:
1625 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1626 flowtype == M_HASHTYPE_NONE)
1628 p = flowid >> sc->flowid_shift;
1630 lb = (struct lagg_lb *)sc->sc_psc;
1631 lp = lb->lb_ports[p];
1632 return (lagg_link_active(sc, lp));
1633 case LAGG_PROTO_LACP:
1634 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1635 flowtype == M_HASHTYPE_NONE)
1637 return (lacp_select_tx_port_by_hash(sc, flowid));
1644 lagg_snd_tag_alloc(struct ifnet *ifp,
1645 union if_snd_tag_alloc_params *params,
1646 struct m_snd_tag **ppmt)
1648 struct lagg_snd_tag *lst;
1649 struct lagg_softc *sc;
1650 struct lagg_port *lp;
1651 struct ifnet *lp_ifp;
1657 lp = lookup_snd_tag_port(ifp, params->hdr.flowid, params->hdr.flowtype);
1660 return (EOPNOTSUPP);
1662 if (lp->lp_ifp == NULL || lp->lp_ifp->if_snd_tag_alloc == NULL) {
1664 return (EOPNOTSUPP);
1666 lp_ifp = lp->lp_ifp;
1670 lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1676 error = lp_ifp->if_snd_tag_alloc(lp_ifp, params, &lst->tag);
1683 m_snd_tag_init(&lst->com, ifp);
1690 lagg_snd_tag_modify(struct m_snd_tag *mst,
1691 union if_snd_tag_modify_params *params)
1693 struct lagg_snd_tag *lst;
1695 lst = mst_to_lst(mst);
1696 return (lst->tag->ifp->if_snd_tag_modify(lst->tag, params));
1700 lagg_snd_tag_query(struct m_snd_tag *mst,
1701 union if_snd_tag_query_params *params)
1703 struct lagg_snd_tag *lst;
1705 lst = mst_to_lst(mst);
1706 return (lst->tag->ifp->if_snd_tag_query(lst->tag, params));
1710 lagg_snd_tag_free(struct m_snd_tag *mst)
1712 struct lagg_snd_tag *lst;
1714 lst = mst_to_lst(mst);
1715 m_snd_tag_rele(lst->tag);
1720 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1723 * For lagg, we have an indirect
1724 * interface. The caller needs to
1725 * get a ratelimit tag on the actual
1726 * interface the flow will go on.
1728 q->rate_table = NULL;
1729 q->flags = RT_IS_INDIRECT;
1731 q->number_of_rates = 0;
1736 lagg_setmulti(struct lagg_port *lp)
1738 struct lagg_softc *sc = lp->lp_softc;
1739 struct ifnet *ifp = lp->lp_ifp;
1740 struct ifnet *scifp = sc->sc_ifp;
1742 struct ifmultiaddr *ifma;
1745 IF_ADDR_WLOCK(scifp);
1746 CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1747 if (ifma->ifma_addr->sa_family != AF_LINK)
1749 mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
1751 IF_ADDR_WUNLOCK(scifp);
1754 bcopy(ifma->ifma_addr, &mc->mc_addr,
1755 ifma->ifma_addr->sa_len);
1756 mc->mc_addr.sdl_index = ifp->if_index;
1758 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
1760 IF_ADDR_WUNLOCK(scifp);
1761 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
1762 error = if_addmulti(ifp,
1763 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
1771 lagg_clrmulti(struct lagg_port *lp)
1775 LAGG_XLOCK_ASSERT(lp->lp_softc);
1776 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
1777 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
1778 if (mc->mc_ifma && lp->lp_detaching == 0)
1779 if_delmulti_ifma(mc->mc_ifma);
1786 lagg_setcaps(struct lagg_port *lp, int cap)
1790 if (lp->lp_ifp->if_capenable == cap)
1792 if (lp->lp_ioctl == NULL)
1794 ifr.ifr_reqcap = cap;
1795 return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr));
1798 /* Handle a ref counted flag that should be set on the lagg port as well */
1800 lagg_setflag(struct lagg_port *lp, int flag, int status,
1801 int (*func)(struct ifnet *, int))
1803 struct lagg_softc *sc = lp->lp_softc;
1804 struct ifnet *scifp = sc->sc_ifp;
1805 struct ifnet *ifp = lp->lp_ifp;
1808 LAGG_XLOCK_ASSERT(sc);
1810 status = status ? (scifp->if_flags & flag) : 0;
1811 /* Now "status" contains the flag value or 0 */
1814 * See if recorded ports status is different from what
1815 * we want it to be. If it is, flip it. We record ports
1816 * status in lp_ifflags so that we won't clear ports flag
1817 * we haven't set. In fact, we don't clear or set ports
1818 * flags directly, but get or release references to them.
1819 * That's why we can be sure that recorded flags still are
1820 * in accord with actual ports flags.
1822 if (status != (lp->lp_ifflags & flag)) {
1823 error = (*func)(ifp, status);
1826 lp->lp_ifflags &= ~flag;
1827 lp->lp_ifflags |= status;
1833 * Handle IFF_* flags that require certain changes on the lagg port
1834 * if "status" is true, update ports flags respective to the lagg
1835 * if "status" is false, forcedly clear the flags set on port.
1838 lagg_setflags(struct lagg_port *lp, int status)
1842 for (i = 0; lagg_pflags[i].flag; i++) {
1843 error = lagg_setflag(lp, lagg_pflags[i].flag,
1844 status, lagg_pflags[i].func);
1852 lagg_transmit(struct ifnet *ifp, struct mbuf *m)
1854 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1857 #if defined(KERN_TLS) || defined(RATELIMIT)
1858 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1859 MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
1862 /* We need a Tx algorithm and at least one port */
1863 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
1866 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1870 ETHER_BPF_MTAP(ifp, m);
1872 error = lagg_proto_start(sc, m);
1876 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1882 * The ifp->if_qflush entry point for lagg(4) is no-op.
1885 lagg_qflush(struct ifnet *ifp __unused)
1889 static struct mbuf *
1890 lagg_input(struct ifnet *ifp, struct mbuf *m)
1892 struct lagg_port *lp = ifp->if_lagg;
1893 struct lagg_softc *sc = lp->lp_softc;
1894 struct ifnet *scifp = sc->sc_ifp;
1897 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1898 lp->lp_detaching != 0 ||
1899 sc->sc_proto == LAGG_PROTO_NONE) {
1905 ETHER_BPF_MTAP(scifp, m);
1907 m = lagg_proto_input(sc, lp, m);
1908 if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) {
1918 lagg_media_change(struct ifnet *ifp)
1920 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1922 if (sc->sc_ifflags & IFF_DEBUG)
1923 printf("%s\n", __func__);
1930 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
1932 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1933 struct lagg_port *lp;
1935 imr->ifm_status = IFM_AVALID;
1936 imr->ifm_active = IFM_ETHER | IFM_AUTO;
1939 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1940 if (LAGG_PORTACTIVE(lp))
1941 imr->ifm_status |= IFM_ACTIVE;
1947 lagg_linkstate(struct lagg_softc *sc)
1949 struct lagg_port *lp;
1950 int new_link = LINK_STATE_DOWN;
1953 LAGG_XLOCK_ASSERT(sc);
1955 /* LACP handles link state itself */
1956 if (sc->sc_proto == LAGG_PROTO_LACP)
1959 /* Our link is considered up if at least one of our ports is active */
1961 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1962 if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
1963 new_link = LINK_STATE_UP;
1968 if_link_state_change(sc->sc_ifp, new_link);
1970 /* Update if_baudrate to reflect the max possible speed */
1971 switch (sc->sc_proto) {
1972 case LAGG_PROTO_FAILOVER:
1973 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
1974 sc->sc_primary->lp_ifp->if_baudrate : 0;
1976 case LAGG_PROTO_ROUNDROBIN:
1977 case LAGG_PROTO_LOADBALANCE:
1978 case LAGG_PROTO_BROADCAST:
1981 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1982 speed += lp->lp_ifp->if_baudrate;
1984 sc->sc_ifp->if_baudrate = speed;
1986 case LAGG_PROTO_LACP:
1987 /* LACP updates if_baudrate itself */
1993 lagg_port_state(struct ifnet *ifp, int state)
1995 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
1996 struct lagg_softc *sc = NULL;
2005 lagg_proto_linkstate(sc, lp);
2010 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2012 struct lagg_port *lp_next, *rval = NULL;
2015 * Search a port which reports an active link state.
2020 * This is called with either LAGG_RLOCK() held or
2021 * LAGG_XLOCK(sc) held.
2023 if (!in_epoch(net_epoch_preempt))
2024 LAGG_XLOCK_ASSERT(sc);
2029 if (LAGG_PORTACTIVE(lp)) {
2033 if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2034 LAGG_PORTACTIVE(lp_next)) {
2040 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2041 if (LAGG_PORTACTIVE(lp_next)) {
2050 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2053 #if defined(KERN_TLS) || defined(RATELIMIT)
2054 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2055 struct lagg_snd_tag *lst;
2056 struct m_snd_tag *mst;
2058 mst = m->m_pkthdr.snd_tag;
2059 lst = mst_to_lst(mst);
2060 if (lst->tag->ifp != ifp) {
2064 m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2065 m_snd_tag_rele(mst);
2068 return (ifp->if_transmit)(ifp, m);
2072 * Simple round robin aggregation
2075 lagg_rr_attach(struct lagg_softc *sc)
2082 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2084 struct lagg_port *lp;
2087 p = atomic_fetchadd_32(&sc->sc_seq, 1);
2090 lp = CK_SLIST_FIRST(&sc->sc_ports);
2093 lp = CK_SLIST_NEXT(lp, lp_entries);
2096 * Check the port's link state. This will return the next active
2097 * port if the link is down or the port is NULL.
2099 if ((lp = lagg_link_active(sc, lp)) == NULL) {
2105 return (lagg_enqueue(lp->lp_ifp, m));
2108 static struct mbuf *
2109 lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2111 struct ifnet *ifp = sc->sc_ifp;
2113 /* Just pass in the packet to our lagg device */
2114 m->m_pkthdr.rcvif = ifp;
2123 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2125 int active_ports = 0;
2128 struct lagg_port *lp, *last = NULL;
2131 LAGG_RLOCK_ASSERT();
2132 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2133 if (!LAGG_PORTACTIVE(lp))
2139 m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2146 ret = lagg_enqueue(last->lp_ifp, m0);
2157 if ((last = lagg_link_active(sc, last)) == NULL) {
2162 ret = lagg_enqueue(last->lp_ifp, m);
2173 lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2175 struct ifnet *ifp = sc->sc_ifp;
2177 /* Just pass in the packet to our lagg device */
2178 m->m_pkthdr.rcvif = ifp;
2186 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2188 struct lagg_port *lp;
2190 /* Use the master port if active or the next available port */
2191 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2197 return (lagg_enqueue(lp->lp_ifp, m));
2200 static struct mbuf *
2201 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2203 struct ifnet *ifp = sc->sc_ifp;
2204 struct lagg_port *tmp_tp;
2206 if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2207 m->m_pkthdr.rcvif = ifp;
2211 if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2212 tmp_tp = lagg_link_active(sc, sc->sc_primary);
2214 * If tmp_tp is null, we've received a packet when all
2215 * our links are down. Weird, but process it anyways.
2217 if ((tmp_tp == NULL || tmp_tp == lp)) {
2218 m->m_pkthdr.rcvif = ifp;
2231 lagg_lb_attach(struct lagg_softc *sc)
2233 struct lagg_port *lp;
2236 LAGG_XLOCK_ASSERT(sc);
2237 lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2238 lb->lb_key = m_ether_tcpip_hash_init();
2241 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2242 lagg_lb_port_create(lp);
2246 lagg_lb_detach(struct lagg_softc *sc)
2250 lb = (struct lagg_lb *)sc->sc_psc;
2256 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2258 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2259 struct lagg_port *lp_next;
2263 bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2264 LAGG_XLOCK_ASSERT(sc);
2265 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2268 if (i >= LAGG_MAX_PORTS) {
2272 if (sc->sc_ifflags & IFF_DEBUG)
2273 printf("%s: port %s at index %d\n",
2274 sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2275 lb->lb_ports[i++] = lp_next;
2282 lagg_lb_port_create(struct lagg_port *lp)
2284 struct lagg_softc *sc = lp->lp_softc;
2285 return (lagg_lb_porttable(sc, NULL));
2289 lagg_lb_port_destroy(struct lagg_port *lp)
2291 struct lagg_softc *sc = lp->lp_softc;
2292 lagg_lb_porttable(sc, lp);
2296 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2298 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2299 struct lagg_port *lp = NULL;
2302 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2303 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2304 p = m->m_pkthdr.flowid >> sc->flowid_shift;
2306 p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2308 lp = lb->lb_ports[p];
2311 * Check the port's link state. This will return the next active
2312 * port if the link is down or the port is NULL.
2314 if ((lp = lagg_link_active(sc, lp)) == NULL) {
2320 return (lagg_enqueue(lp->lp_ifp, m));
2323 static struct mbuf *
2324 lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2326 struct ifnet *ifp = sc->sc_ifp;
2328 /* Just pass in the packet to our lagg device */
2329 m->m_pkthdr.rcvif = ifp;
2338 lagg_lacp_attach(struct lagg_softc *sc)
2340 struct lagg_port *lp;
2343 LAGG_XLOCK_ASSERT(sc);
2344 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2345 lacp_port_create(lp);
2349 lagg_lacp_detach(struct lagg_softc *sc)
2351 struct lagg_port *lp;
2354 LAGG_XLOCK_ASSERT(sc);
2355 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2356 lacp_port_destroy(lp);
2364 lagg_lacp_lladdr(struct lagg_softc *sc)
2366 struct lagg_port *lp;
2368 LAGG_SXLOCK_ASSERT(sc);
2370 /* purge all the lacp ports */
2371 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2372 lacp_port_destroy(lp);
2374 /* add them back in */
2375 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2376 lacp_port_create(lp);
2380 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2382 struct lagg_port *lp;
2384 lp = lacp_select_tx_port(sc, m);
2391 return (lagg_enqueue(lp->lp_ifp, m));
2394 static struct mbuf *
2395 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2397 struct ifnet *ifp = sc->sc_ifp;
2398 struct ether_header *eh;
2401 eh = mtod(m, struct ether_header *);
2402 etype = ntohs(eh->ether_type);
2404 /* Tap off LACP control messages */
2405 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2406 m = lacp_input(lp, m);
2412 * If the port is not collecting or not in the active aggregator then
2415 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) {
2420 m->m_pkthdr.rcvif = ifp;