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>
23 #include "opt_inet6.h"
24 #include "opt_kern_tls.h"
25 #include "opt_ratelimit.h"
27 #include <sys/param.h>
28 #include <sys/kernel.h>
29 #include <sys/malloc.h>
31 #include <sys/queue.h>
32 #include <sys/socket.h>
33 #include <sys/sockio.h>
34 #include <sys/sysctl.h>
35 #include <sys/module.h>
37 #include <sys/systm.h>
40 #include <sys/rmlock.h>
42 #include <sys/taskqueue.h>
43 #include <sys/eventhandler.h>
45 #include <net/ethernet.h>
47 #include <net/if_clone.h>
48 #include <net/if_arp.h>
49 #include <net/if_dl.h>
50 #include <net/if_media.h>
51 #include <net/if_types.h>
52 #include <net/if_var.h>
53 #include <net/if_private.h>
55 #include <net/route.h>
57 #include <net/infiniband.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>
79 MODULE_DEPEND(if_lagg, netmap, 1, 1, 1);
82 #define LAGG_SX_INIT(_sc) sx_init(&(_sc)->sc_sx, "if_lagg sx")
83 #define LAGG_SX_DESTROY(_sc) sx_destroy(&(_sc)->sc_sx)
84 #define LAGG_XLOCK(_sc) sx_xlock(&(_sc)->sc_sx)
85 #define LAGG_XUNLOCK(_sc) sx_xunlock(&(_sc)->sc_sx)
86 #define LAGG_XLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
87 #define LAGG_SLOCK(_sc) sx_slock(&(_sc)->sc_sx)
88 #define LAGG_SUNLOCK(_sc) sx_sunlock(&(_sc)->sc_sx)
89 #define LAGG_SXLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_LOCKED)
91 /* Special flags we should propagate to the lagg ports. */
94 int (*func)(struct ifnet *, int);
96 {IFF_PROMISC, ifpromisc},
97 {IFF_ALLMULTI, if_allmulti},
101 struct lagg_snd_tag {
102 struct m_snd_tag com;
103 struct m_snd_tag *tag;
106 VNET_DEFINE_STATIC(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */
107 #define V_lagg_list VNET(lagg_list)
108 VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx);
109 #define V_lagg_list_mtx VNET(lagg_list_mtx)
110 #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \
111 "if_lagg list", NULL, MTX_DEF)
112 #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx)
113 #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx)
114 #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx)
115 static eventhandler_tag lagg_detach_cookie = NULL;
117 static int lagg_clone_create(struct if_clone *, char *, size_t,
118 struct ifc_data *, struct ifnet **);
119 static int lagg_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
120 VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner);
121 #define V_lagg_cloner VNET(lagg_cloner)
122 static const char laggname[] = "lagg";
123 static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface");
125 static void lagg_capabilities(struct lagg_softc *);
126 static int lagg_port_create(struct lagg_softc *, struct ifnet *);
127 static int lagg_port_destroy(struct lagg_port *, int);
128 static struct mbuf *lagg_input_ethernet(struct ifnet *, struct mbuf *);
129 static struct mbuf *lagg_input_infiniband(struct ifnet *, struct mbuf *);
130 static void lagg_linkstate(struct lagg_softc *);
131 static void lagg_port_state(struct ifnet *, int);
132 static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t);
133 static int lagg_port_output(struct ifnet *, struct mbuf *,
134 const struct sockaddr *, struct route *);
135 static void lagg_port_ifdetach(void *arg __unused, struct ifnet *);
136 #ifdef LAGG_PORT_STACKING
137 static int lagg_port_checkstacking(struct lagg_softc *);
139 static void lagg_port2req(struct lagg_port *, struct lagg_reqport *);
140 static void lagg_init(void *);
141 static void lagg_stop(struct lagg_softc *);
142 static int lagg_ioctl(struct ifnet *, u_long, caddr_t);
143 #if defined(KERN_TLS) || defined(RATELIMIT)
144 static int lagg_snd_tag_alloc(struct ifnet *,
145 union if_snd_tag_alloc_params *,
146 struct m_snd_tag **);
147 static int lagg_snd_tag_modify(struct m_snd_tag *,
148 union if_snd_tag_modify_params *);
149 static int lagg_snd_tag_query(struct m_snd_tag *,
150 union if_snd_tag_query_params *);
151 static void lagg_snd_tag_free(struct m_snd_tag *);
152 static struct m_snd_tag *lagg_next_snd_tag(struct m_snd_tag *);
153 static void lagg_ratelimit_query(struct ifnet *,
154 struct if_ratelimit_query_results *);
156 static int lagg_setmulti(struct lagg_port *);
157 static int lagg_clrmulti(struct lagg_port *);
158 static void lagg_setcaps(struct lagg_port *, int cap, int cap2);
159 static int lagg_setflag(struct lagg_port *, int, int,
160 int (*func)(struct ifnet *, int));
161 static int lagg_setflags(struct lagg_port *, int status);
162 static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt);
163 static int lagg_transmit_ethernet(struct ifnet *, struct mbuf *);
164 static int lagg_transmit_infiniband(struct ifnet *, struct mbuf *);
165 static void lagg_qflush(struct ifnet *);
166 static int lagg_media_change(struct ifnet *);
167 static void lagg_media_status(struct ifnet *, struct ifmediareq *);
168 static struct lagg_port *lagg_link_active(struct lagg_softc *,
171 /* Simple round robin */
172 static void lagg_rr_attach(struct lagg_softc *);
173 static int lagg_rr_start(struct lagg_softc *, struct mbuf *);
175 /* Active failover */
176 static int lagg_fail_start(struct lagg_softc *, struct mbuf *);
177 static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *,
181 static void lagg_lb_attach(struct lagg_softc *);
182 static void lagg_lb_detach(struct lagg_softc *);
183 static int lagg_lb_port_create(struct lagg_port *);
184 static void lagg_lb_port_destroy(struct lagg_port *);
185 static int lagg_lb_start(struct lagg_softc *, struct mbuf *);
186 static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *);
189 static int lagg_bcast_start(struct lagg_softc *, struct mbuf *);
192 static void lagg_lacp_attach(struct lagg_softc *);
193 static void lagg_lacp_detach(struct lagg_softc *);
194 static int lagg_lacp_start(struct lagg_softc *, struct mbuf *);
195 static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *,
197 static void lagg_lacp_lladdr(struct lagg_softc *);
200 static struct mbuf *lagg_default_input(struct lagg_softc *, struct lagg_port *,
203 /* lagg protocol table */
204 static const struct lagg_proto {
206 void (*pr_attach)(struct lagg_softc *);
207 void (*pr_detach)(struct lagg_softc *);
208 int (*pr_start)(struct lagg_softc *, struct mbuf *);
209 struct mbuf * (*pr_input)(struct lagg_softc *, struct lagg_port *,
211 int (*pr_addport)(struct lagg_port *);
212 void (*pr_delport)(struct lagg_port *);
213 void (*pr_linkstate)(struct lagg_port *);
214 void (*pr_init)(struct lagg_softc *);
215 void (*pr_stop)(struct lagg_softc *);
216 void (*pr_lladdr)(struct lagg_softc *);
217 void (*pr_request)(struct lagg_softc *, void *);
218 void (*pr_portreq)(struct lagg_port *, void *);
221 .pr_num = LAGG_PROTO_NONE
224 .pr_num = LAGG_PROTO_ROUNDROBIN,
225 .pr_attach = lagg_rr_attach,
226 .pr_start = lagg_rr_start,
227 .pr_input = lagg_default_input,
230 .pr_num = LAGG_PROTO_FAILOVER,
231 .pr_start = lagg_fail_start,
232 .pr_input = lagg_fail_input,
235 .pr_num = LAGG_PROTO_LOADBALANCE,
236 .pr_attach = lagg_lb_attach,
237 .pr_detach = lagg_lb_detach,
238 .pr_start = lagg_lb_start,
239 .pr_input = lagg_default_input,
240 .pr_addport = lagg_lb_port_create,
241 .pr_delport = lagg_lb_port_destroy,
244 .pr_num = LAGG_PROTO_LACP,
245 .pr_attach = lagg_lacp_attach,
246 .pr_detach = lagg_lacp_detach,
247 .pr_start = lagg_lacp_start,
248 .pr_input = lagg_lacp_input,
249 .pr_addport = lacp_port_create,
250 .pr_delport = lacp_port_destroy,
251 .pr_linkstate = lacp_linkstate,
252 .pr_init = lacp_init,
253 .pr_stop = lacp_stop,
254 .pr_lladdr = lagg_lacp_lladdr,
255 .pr_request = lacp_req,
256 .pr_portreq = lacp_portreq,
259 .pr_num = LAGG_PROTO_BROADCAST,
260 .pr_start = lagg_bcast_start,
261 .pr_input = lagg_default_input,
265 SYSCTL_DECL(_net_link);
266 SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
269 /* Allow input on any failover links */
270 VNET_DEFINE_STATIC(int, lagg_failover_rx_all);
271 #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all)
272 SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET,
273 &VNET_NAME(lagg_failover_rx_all), 0,
274 "Accept input from any interface in a failover lagg");
276 /* Default value for using flowid */
277 VNET_DEFINE_STATIC(int, def_use_flowid) = 0;
278 #define V_def_use_flowid VNET(def_use_flowid)
279 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid,
280 CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_use_flowid), 0,
281 "Default setting for using flow id for load sharing");
283 /* Default value for using numa */
284 VNET_DEFINE_STATIC(int, def_use_numa) = 1;
285 #define V_def_use_numa VNET(def_use_numa)
286 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa,
287 CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_use_numa), 0,
288 "Use numa to steer flows");
290 /* Default value for flowid shift */
291 VNET_DEFINE_STATIC(int, def_flowid_shift) = 16;
292 #define V_def_flowid_shift VNET(def_flowid_shift)
293 SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift,
294 CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(def_flowid_shift), 0,
295 "Default setting for flowid shift for load sharing");
298 vnet_lagg_init(const void *unused __unused)
301 LAGG_LIST_LOCK_INIT();
302 SLIST_INIT(&V_lagg_list);
303 struct if_clone_addreq req = {
304 .create_f = lagg_clone_create,
305 .destroy_f = lagg_clone_destroy,
306 .flags = IFC_F_AUTOUNIT,
308 V_lagg_cloner = ifc_attach_cloner(laggname, &req);
310 VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
311 vnet_lagg_init, NULL);
314 vnet_lagg_uninit(const void *unused __unused)
317 ifc_detach_cloner(V_lagg_cloner);
318 LAGG_LIST_LOCK_DESTROY();
320 VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
321 vnet_lagg_uninit, NULL);
324 lagg_modevent(module_t mod, int type, void *data)
329 lagg_input_ethernet_p = lagg_input_ethernet;
330 lagg_input_infiniband_p = lagg_input_infiniband;
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,
339 lagg_input_ethernet_p = NULL;
340 lagg_input_infiniband_p = NULL;
341 lagg_linkstate_p = NULL;
349 static moduledata_t lagg_mod = {
355 DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
356 MODULE_VERSION(if_lagg, 1);
357 MODULE_DEPEND(if_lagg, if_infiniband, 1, 1, 1);
360 lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr)
363 LAGG_XLOCK_ASSERT(sc);
364 KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto",
367 if (sc->sc_ifflags & IFF_DEBUG)
368 if_printf(sc->sc_ifp, "using proto %u\n", pr);
370 if (lagg_protos[pr].pr_attach != NULL)
371 lagg_protos[pr].pr_attach(sc);
376 lagg_proto_detach(struct lagg_softc *sc)
380 LAGG_XLOCK_ASSERT(sc);
382 sc->sc_proto = LAGG_PROTO_NONE;
384 if (lagg_protos[pr].pr_detach != NULL)
385 lagg_protos[pr].pr_detach(sc);
389 lagg_proto_start(struct lagg_softc *sc, struct mbuf *m)
392 return (lagg_protos[sc->sc_proto].pr_start(sc, m));
395 static inline struct mbuf *
396 lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
399 return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m));
403 lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp)
406 if (lagg_protos[sc->sc_proto].pr_addport == NULL)
409 return (lagg_protos[sc->sc_proto].pr_addport(lp));
413 lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp)
416 if (lagg_protos[sc->sc_proto].pr_delport != NULL)
417 lagg_protos[sc->sc_proto].pr_delport(lp);
421 lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp)
424 if (lagg_protos[sc->sc_proto].pr_linkstate != NULL)
425 lagg_protos[sc->sc_proto].pr_linkstate(lp);
429 lagg_proto_init(struct lagg_softc *sc)
432 if (lagg_protos[sc->sc_proto].pr_init != NULL)
433 lagg_protos[sc->sc_proto].pr_init(sc);
437 lagg_proto_stop(struct lagg_softc *sc)
440 if (lagg_protos[sc->sc_proto].pr_stop != NULL)
441 lagg_protos[sc->sc_proto].pr_stop(sc);
445 lagg_proto_lladdr(struct lagg_softc *sc)
448 if (lagg_protos[sc->sc_proto].pr_lladdr != NULL)
449 lagg_protos[sc->sc_proto].pr_lladdr(sc);
453 lagg_proto_request(struct lagg_softc *sc, void *v)
456 if (lagg_protos[sc->sc_proto].pr_request != NULL)
457 lagg_protos[sc->sc_proto].pr_request(sc, v);
461 lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v)
464 if (lagg_protos[sc->sc_proto].pr_portreq != NULL)
465 lagg_protos[sc->sc_proto].pr_portreq(lp, v);
469 * This routine is run via an vlan
473 lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
475 struct lagg_softc *sc = ifp->if_softc;
476 struct lagg_port *lp;
478 if (ifp->if_softc != arg) /* Not our event */
482 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
483 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag);
488 * This routine is run via an vlan
492 lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
494 struct lagg_softc *sc = ifp->if_softc;
495 struct lagg_port *lp;
497 if (ifp->if_softc != arg) /* Not our event */
501 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
502 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag);
507 lagg_clone_create(struct if_clone *ifc, char *name, size_t len,
508 struct ifc_data *ifd, struct ifnet **ifpp)
510 struct iflaggparam iflp;
511 struct lagg_softc *sc;
515 static const uint8_t eaddr[LAGG_ADDR_LEN];
517 if (ifd->params != NULL) {
518 error = ifc_copyin(ifd, &iflp, sizeof(iflp));
522 switch (iflp.lagg_type) {
523 case LAGG_TYPE_ETHERNET:
526 case LAGG_TYPE_INFINIBAND:
527 if_type = IFT_INFINIBAND;
536 sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK | M_ZERO);
537 ifp = sc->sc_ifp = if_alloc(if_type);
544 mtx_init(&sc->sc_mtx, "lagg-mtx", NULL, MTX_DEF);
545 callout_init_mtx(&sc->sc_watchdog, &sc->sc_mtx, 0);
548 if (V_def_use_flowid)
549 sc->sc_opts |= LAGG_OPT_USE_FLOWID;
551 sc->sc_opts |= LAGG_OPT_USE_NUMA;
552 sc->flowid_shift = V_def_flowid_shift;
554 /* Hash all layers by default */
555 sc->sc_flags = MBUF_HASHFLAG_L2 | MBUF_HASHFLAG_L3 | MBUF_HASHFLAG_L4;
557 lagg_proto_attach(sc, LAGG_PROTO_DEFAULT);
559 CK_SLIST_INIT(&sc->sc_ports);
563 /* Initialise pseudo media types */
564 ifmedia_init(&sc->sc_media, 0, lagg_media_change,
566 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
567 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
569 if_initname(ifp, laggname, ifd->unit);
570 ifp->if_transmit = lagg_transmit_ethernet;
573 if_initname(ifp, laggname, ifd->unit);
574 ifp->if_transmit = lagg_transmit_infiniband;
580 ifp->if_qflush = lagg_qflush;
581 ifp->if_init = lagg_init;
582 ifp->if_ioctl = lagg_ioctl;
583 ifp->if_get_counter = lagg_get_counter;
584 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST;
585 #if defined(KERN_TLS) || defined(RATELIMIT)
586 ifp->if_snd_tag_alloc = lagg_snd_tag_alloc;
587 ifp->if_ratelimit_query = lagg_ratelimit_query;
589 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
592 * Attach as an ordinary ethernet device, children will be attached
593 * as special device IFT_IEEE8023ADLAG or IFT_INFINIBANDLAG.
597 ether_ifattach(ifp, eaddr);
600 infiniband_ifattach(ifp, eaddr, sc->sc_bcast_addr);
606 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
607 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
608 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
609 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
611 /* Insert into the global list of laggs */
613 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries);
622 lagg_clone_destroy(struct if_clone *ifc, struct ifnet *ifp, uint32_t flags)
624 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
625 struct lagg_port *lp;
628 sc->sc_destroying = 1;
630 ifp->if_flags &= ~IFF_UP;
632 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
633 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
635 /* Shutdown and remove lagg ports */
636 while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL)
637 lagg_port_destroy(lp, 1);
639 /* Unhook the aggregation protocol */
640 lagg_proto_detach(sc);
643 switch (ifp->if_type) {
645 ifmedia_removeall(&sc->sc_media);
649 infiniband_ifdetach(ifp);
657 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries);
660 mtx_destroy(&sc->sc_mtx);
668 lagg_capabilities(struct lagg_softc *sc)
670 struct lagg_port *lp;
671 int cap, cap2, ena, ena2, pena, pena2;
673 struct ifnet_hw_tsomax hw_tsomax;
675 LAGG_XLOCK_ASSERT(sc);
677 /* Get common enabled capabilities for the lagg ports */
679 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
680 ena &= lp->lp_ifp->if_capenable;
681 ena2 &= lp->lp_ifp->if_capenable2;
683 if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
687 * Apply common enabled capabilities back to the lagg ports.
688 * May require several iterations if they are dependent.
693 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
694 lagg_setcaps(lp, ena, ena2);
695 ena &= lp->lp_ifp->if_capenable;
696 ena2 &= lp->lp_ifp->if_capenable2;
698 } while (pena != ena || pena2 != ena2);
700 /* Get other capabilities from the lagg ports */
703 memset(&hw_tsomax, 0, sizeof(hw_tsomax));
704 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
705 cap &= lp->lp_ifp->if_capabilities;
706 cap2 &= lp->lp_ifp->if_capabilities2;
707 hwa &= lp->lp_ifp->if_hwassist;
708 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax);
710 if (CK_SLIST_FIRST(&sc->sc_ports) == NULL)
711 cap = cap2 = hwa = 0;
713 if (sc->sc_ifp->if_capabilities != cap ||
714 sc->sc_ifp->if_capenable != ena ||
715 sc->sc_ifp->if_capenable2 != ena2 ||
716 sc->sc_ifp->if_hwassist != hwa ||
717 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) {
718 sc->sc_ifp->if_capabilities = cap;
719 sc->sc_ifp->if_capabilities2 = cap2;
720 sc->sc_ifp->if_capenable = ena;
721 sc->sc_ifp->if_capenable2 = ena2;
722 sc->sc_ifp->if_hwassist = hwa;
723 getmicrotime(&sc->sc_ifp->if_lastchange);
725 if (sc->sc_ifflags & IFF_DEBUG)
726 if_printf(sc->sc_ifp,
727 "capabilities 0x%08x enabled 0x%08x\n", cap, ena);
732 lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp)
734 struct lagg_softc *sc_ptr;
735 struct lagg_port *lp, *tlp;
737 int error, i, oldmtu;
741 LAGG_XLOCK_ASSERT(sc);
743 if (sc->sc_ifp == ifp) {
744 if_printf(sc->sc_ifp,
745 "cannot add a lagg to itself as a port\n");
749 if (sc->sc_destroying == 1)
752 /* Limit the maximal number of lagg ports */
753 if (sc->sc_count >= LAGG_MAX_PORTS)
756 /* Check if port has already been associated to a lagg */
757 if (ifp->if_lagg != NULL) {
758 /* Port is already in the current lagg? */
759 lp = (struct lagg_port *)ifp->if_lagg;
760 if (lp->lp_softc == sc)
765 switch (sc->sc_ifp->if_type) {
767 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */
768 if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN)
769 return (EPROTONOSUPPORT);
770 if_type = IFT_IEEE8023ADLAG;
773 /* XXX Disallow non-infiniband interfaces */
774 if (ifp->if_type != IFT_INFINIBAND)
775 return (EPROTONOSUPPORT);
776 if_type = IFT_INFINIBANDLAG;
782 /* Allow the first Ethernet member to define the MTU */
784 if (CK_SLIST_EMPTY(&sc->sc_ports)) {
785 sc->sc_ifp->if_mtu = ifp->if_mtu;
786 } else if (sc->sc_ifp->if_mtu != ifp->if_mtu) {
787 if (ifp->if_ioctl == NULL) {
788 if_printf(sc->sc_ifp, "cannot change MTU for %s\n",
792 oldmtu = ifp->if_mtu;
793 strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
794 ifr.ifr_mtu = sc->sc_ifp->if_mtu;
795 error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
797 if_printf(sc->sc_ifp, "invalid MTU for %s\n",
801 ifr.ifr_mtu = oldmtu;
804 lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK | M_ZERO);
807 /* Check if port is a stacked lagg */
809 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) {
810 if (ifp == sc_ptr->sc_ifp) {
814 (*ifp->if_ioctl)(ifp, SIOCSIFMTU,
817 /* XXX disable stacking for the moment, its untested */
818 #ifdef LAGG_PORT_STACKING
819 lp->lp_flags |= LAGG_PORT_STACK;
820 if (lagg_port_checkstacking(sc_ptr) >=
825 (*ifp->if_ioctl)(ifp, SIOCSIFMTU,
837 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ifp->if_addrlen);
838 lp->lp_ifcapenable = ifp->if_capenable;
839 if (CK_SLIST_EMPTY(&sc->sc_ports)) {
840 bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
841 lagg_proto_lladdr(sc);
842 EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
844 if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ifp->if_addrlen);
846 lagg_setflags(lp, 1);
848 if (CK_SLIST_EMPTY(&sc->sc_ports))
851 /* Change the interface type */
852 lp->lp_iftype = ifp->if_type;
853 ifp->if_type = if_type;
855 lp->lp_ioctl = ifp->if_ioctl;
856 ifp->if_ioctl = lagg_port_ioctl;
857 lp->lp_output = ifp->if_output;
858 ifp->if_output = lagg_port_output;
860 /* Read port counters */
861 pval = lp->port_counters.val;
862 for (i = 0; i < IFCOUNTERS; i++, pval++)
863 *pval = ifp->if_get_counter(ifp, i);
866 * Insert into the list of ports.
867 * Keep ports sorted by if_index. It is handy, when configuration
868 * is predictable and `ifconfig laggN create ...` command
869 * will lead to the same result each time.
871 CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) {
872 if (tlp->lp_ifp->if_index < ifp->if_index && (
873 CK_SLIST_NEXT(tlp, lp_entries) == NULL ||
874 ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index >
879 CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries);
881 CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries);
886 if ((error = lagg_proto_addport(sc, lp)) != 0) {
887 /* Remove the port, without calling pr_delport. */
888 lagg_port_destroy(lp, 0);
890 (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr);
894 /* Update lagg capabilities */
895 lagg_capabilities(sc);
901 #ifdef LAGG_PORT_STACKING
903 lagg_port_checkstacking(struct lagg_softc *sc)
905 struct lagg_softc *sc_ptr;
906 struct lagg_port *lp;
909 LAGG_SXLOCK_ASSERT(sc);
910 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
911 if (lp->lp_flags & LAGG_PORT_STACK) {
912 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc;
913 m = MAX(m, lagg_port_checkstacking(sc_ptr));
922 lagg_port_destroy_cb(epoch_context_t ec)
924 struct lagg_port *lp;
927 lp = __containerof(ec, struct lagg_port, lp_epoch_ctx);
935 lagg_port_destroy(struct lagg_port *lp, int rundelport)
937 struct lagg_softc *sc = lp->lp_softc;
938 struct lagg_port *lp_ptr, *lp0;
939 struct ifnet *ifp = lp->lp_ifp;
940 uint64_t *pval, vdiff;
943 LAGG_XLOCK_ASSERT(sc);
946 lagg_proto_delport(sc, lp);
948 if (lp->lp_detaching == 0)
951 /* Restore interface */
952 ifp->if_type = lp->lp_iftype;
953 ifp->if_ioctl = lp->lp_ioctl;
954 ifp->if_output = lp->lp_output;
957 /* Update detached port counters */
958 pval = lp->port_counters.val;
959 for (i = 0; i < IFCOUNTERS; i++, pval++) {
960 vdiff = ifp->if_get_counter(ifp, i) - *pval;
961 sc->detached_counters.val[i] += vdiff;
964 /* Finally, remove the port from the lagg */
965 CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries);
968 /* Update the primary interface */
969 if (lp == sc->sc_primary) {
970 uint8_t lladdr[LAGG_ADDR_LEN];
972 if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL)
973 bzero(&lladdr, LAGG_ADDR_LEN);
975 bcopy(lp0->lp_lladdr, lladdr, LAGG_ADDR_LEN);
976 sc->sc_primary = lp0;
977 if (sc->sc_destroying == 0) {
978 bcopy(lladdr, IF_LLADDR(sc->sc_ifp), sc->sc_ifp->if_addrlen);
979 lagg_proto_lladdr(sc);
980 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).
985 * We can skip this if the lagg is being destroyed.
987 CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
988 if_setlladdr(lp_ptr->lp_ifp, lladdr,
989 lp_ptr->lp_ifp->if_addrlen);
994 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
996 if (lp->lp_detaching == 0) {
997 lagg_setflags(lp, 0);
998 lagg_setcaps(lp, lp->lp_ifcapenable, lp->lp_ifcapenable2);
999 if_setlladdr(ifp, lp->lp_lladdr, ifp->if_addrlen);
1003 * free port and release it's ifnet reference after a grace period has
1006 NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx);
1007 /* Update lagg capabilities */
1008 lagg_capabilities(sc);
1015 lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1017 struct epoch_tracker et;
1018 struct lagg_reqport *rp = (struct lagg_reqport *)data;
1019 struct lagg_softc *sc;
1020 struct lagg_port *lp = NULL;
1023 /* Should be checked by the caller */
1024 switch (ifp->if_type) {
1025 case IFT_IEEE8023ADLAG:
1026 case IFT_INFINIBANDLAG:
1027 if ((lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL)
1036 if (rp->rp_portname[0] == '\0' ||
1037 ifunit(rp->rp_portname) != ifp) {
1042 NET_EPOCH_ENTER(et);
1043 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) {
1050 lagg_port2req(lp, rp);
1057 if (lp->lp_ioctl == NULL) {
1061 error = (*lp->lp_ioctl)(ifp, cmd, data);
1065 /* Update lagg interface capabilities */
1067 lagg_capabilities(sc);
1069 VLAN_CAPABILITIES(sc->sc_ifp);
1073 /* Do not allow the MTU to be changed once joined */
1084 if (lp != NULL && lp->lp_ioctl != NULL)
1085 return ((*lp->lp_ioctl)(ifp, cmd, data));
1091 * Requests counter @cnt data.
1093 * Counter value is calculated the following way:
1094 * 1) for each port, sum difference between current and "initial" measurements.
1095 * 2) add lagg logical interface counters.
1096 * 3) add data from detached_counters array.
1098 * We also do the following things on ports attach/detach:
1099 * 1) On port attach we store all counters it has into port_counter array.
1100 * 2) On port detach we add the different between "initial" and
1101 * current counters data to detached_counters array.
1104 lagg_get_counter(struct ifnet *ifp, ift_counter cnt)
1106 struct epoch_tracker et;
1107 struct lagg_softc *sc;
1108 struct lagg_port *lp;
1109 struct ifnet *lpifp;
1110 uint64_t newval, oldval, vsum;
1112 /* Revise this when we've got non-generic counters. */
1113 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1115 sc = (struct lagg_softc *)ifp->if_softc;
1118 NET_EPOCH_ENTER(et);
1119 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1120 /* Saved attached value */
1121 oldval = lp->port_counters.val[cnt];
1124 newval = lpifp->if_get_counter(lpifp, cnt);
1125 /* Calculate diff and save new */
1126 vsum += newval - oldval;
1131 * Add counter data which might be added by upper
1132 * layer protocols operating on logical interface.
1134 vsum += if_get_counter_default(ifp, cnt);
1137 * Add counter data from detached ports counters
1139 vsum += sc->detached_counters.val[cnt];
1145 * For direct output to child ports.
1148 lagg_port_output(struct ifnet *ifp, struct mbuf *m,
1149 const struct sockaddr *dst, struct route *ro)
1151 struct lagg_port *lp = ifp->if_lagg;
1153 switch (dst->sa_family) {
1154 case pseudo_AF_HDRCMPLT:
1157 return ((*lp->lp_output)(ifp, m, dst, ro));
1160 /* drop any other frames */
1166 lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp)
1168 struct lagg_port *lp;
1169 struct lagg_softc *sc;
1171 if ((lp = ifp->if_lagg) == NULL)
1173 /* If the ifnet is just being renamed, don't do anything. */
1174 if (ifp->if_flags & IFF_RENAMING)
1180 lp->lp_detaching = 1;
1181 lagg_port_destroy(lp, 1);
1183 VLAN_CAPABILITIES(sc->sc_ifp);
1187 lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp)
1189 struct lagg_softc *sc = lp->lp_softc;
1191 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname));
1192 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname));
1193 rp->rp_prio = lp->lp_prio;
1194 rp->rp_flags = lp->lp_flags;
1195 lagg_proto_portreq(sc, lp, &rp->rp_psc);
1197 /* Add protocol specific flags */
1198 switch (sc->sc_proto) {
1199 case LAGG_PROTO_FAILOVER:
1200 if (lp == sc->sc_primary)
1201 rp->rp_flags |= LAGG_PORT_MASTER;
1202 if (lp == lagg_link_active(sc, sc->sc_primary))
1203 rp->rp_flags |= LAGG_PORT_ACTIVE;
1206 case LAGG_PROTO_ROUNDROBIN:
1207 case LAGG_PROTO_LOADBALANCE:
1208 case LAGG_PROTO_BROADCAST:
1209 if (LAGG_PORTACTIVE(lp))
1210 rp->rp_flags |= LAGG_PORT_ACTIVE;
1213 case LAGG_PROTO_LACP:
1214 /* LACP has a different definition of active */
1215 if (lacp_isactive(lp))
1216 rp->rp_flags |= LAGG_PORT_ACTIVE;
1217 if (lacp_iscollecting(lp))
1218 rp->rp_flags |= LAGG_PORT_COLLECTING;
1219 if (lacp_isdistributing(lp))
1220 rp->rp_flags |= LAGG_PORT_DISTRIBUTING;
1227 lagg_watchdog_infiniband(void *arg)
1229 struct epoch_tracker et;
1230 struct lagg_softc *sc;
1231 struct lagg_port *lp;
1233 struct ifnet *lp_ifp;
1238 * Because infiniband nodes have a fixed MAC address, which is
1239 * generated by the so-called GID, we need to regularly update
1240 * the link level address of the parent lagg<N> device when
1241 * the active port changes. Possibly we could piggy-back on
1242 * link up/down events aswell, but using a timer also provides
1243 * a guarantee against too frequent events. This operation
1244 * does not have to be atomic.
1246 NET_EPOCH_ENTER(et);
1247 lp = lagg_link_active(sc, sc->sc_primary);
1250 lp_ifp = lp->lp_ifp;
1252 if (ifp != NULL && lp_ifp != NULL &&
1253 (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen) != 0 ||
1254 memcmp(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen) != 0)) {
1255 memcpy(IF_LLADDR(ifp), IF_LLADDR(lp_ifp), ifp->if_addrlen);
1256 memcpy(sc->sc_bcast_addr, lp_ifp->if_broadcastaddr, ifp->if_addrlen);
1258 CURVNET_SET(ifp->if_vnet);
1259 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1265 callout_reset(&sc->sc_watchdog, hz, &lagg_watchdog_infiniband, arg);
1269 lagg_init(void *xsc)
1271 struct lagg_softc *sc = (struct lagg_softc *)xsc;
1272 struct ifnet *ifp = sc->sc_ifp;
1273 struct lagg_port *lp;
1276 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1281 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1284 * Update the port lladdrs if needed.
1285 * This might be if_setlladdr() notification
1286 * that lladdr has been changed.
1288 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1289 if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp),
1290 ifp->if_addrlen) != 0)
1291 if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ifp->if_addrlen);
1294 lagg_proto_init(sc);
1296 if (ifp->if_type == IFT_INFINIBAND) {
1297 mtx_lock(&sc->sc_mtx);
1298 lagg_watchdog_infiniband(sc);
1299 mtx_unlock(&sc->sc_mtx);
1306 lagg_stop(struct lagg_softc *sc)
1308 struct ifnet *ifp = sc->sc_ifp;
1310 LAGG_XLOCK_ASSERT(sc);
1312 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1315 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1317 lagg_proto_stop(sc);
1319 mtx_lock(&sc->sc_mtx);
1320 callout_stop(&sc->sc_watchdog);
1321 mtx_unlock(&sc->sc_mtx);
1323 callout_drain(&sc->sc_watchdog);
1327 lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1329 struct epoch_tracker et;
1330 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
1331 struct lagg_reqall *ra = (struct lagg_reqall *)data;
1332 struct lagg_reqopts *ro = (struct lagg_reqopts *)data;
1333 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf;
1334 struct lagg_reqflags *rf = (struct lagg_reqflags *)data;
1335 struct ifreq *ifr = (struct ifreq *)data;
1336 struct lagg_port *lp;
1338 struct thread *td = curthread;
1340 int count, buflen, len, error = 0, oldmtu;
1342 bzero(&rpbuf, sizeof(rpbuf));
1344 /* XXX: This can race with lagg_clone_destroy. */
1349 buflen = sc->sc_count * sizeof(struct lagg_reqport);
1350 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
1351 ra->ra_proto = sc->sc_proto;
1352 lagg_proto_request(sc, &ra->ra_psc);
1355 len = min(ra->ra_size, buflen);
1356 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1357 if (len < sizeof(rpbuf))
1360 lagg_port2req(lp, &rpbuf);
1361 memcpy(buf, &rpbuf, sizeof(rpbuf));
1363 buf += sizeof(rpbuf);
1364 len -= sizeof(rpbuf);
1367 ra->ra_ports = count;
1368 ra->ra_size = count * sizeof(rpbuf);
1369 error = copyout(outbuf, ra->ra_port, ra->ra_size);
1370 free(outbuf, M_TEMP);
1373 error = priv_check(td, PRIV_NET_LAGG);
1376 if (ra->ra_proto >= LAGG_PROTO_MAX) {
1377 error = EPROTONOSUPPORT;
1380 /* Infiniband only supports the failover protocol. */
1381 if (ra->ra_proto != LAGG_PROTO_FAILOVER &&
1382 ifp->if_type == IFT_INFINIBAND) {
1383 error = EPROTONOSUPPORT;
1387 lagg_proto_detach(sc);
1388 lagg_proto_attach(sc, ra->ra_proto);
1393 ro->ro_opts = sc->sc_opts;
1394 if (sc->sc_proto == LAGG_PROTO_LACP) {
1395 struct lacp_softc *lsc;
1397 lsc = (struct lacp_softc *)sc->sc_psc;
1398 if (lsc->lsc_debug.lsc_tx_test != 0)
1399 ro->ro_opts |= LAGG_OPT_LACP_TXTEST;
1400 if (lsc->lsc_debug.lsc_rx_test != 0)
1401 ro->ro_opts |= LAGG_OPT_LACP_RXTEST;
1402 if (lsc->lsc_strict_mode != 0)
1403 ro->ro_opts |= LAGG_OPT_LACP_STRICT;
1404 if (lsc->lsc_fast_timeout != 0)
1405 ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO;
1407 ro->ro_active = sc->sc_active;
1410 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
1411 ro->ro_active += LAGG_PORTACTIVE(lp);
1413 ro->ro_bkt = sc->sc_stride;
1414 ro->ro_flapping = sc->sc_flapping;
1415 ro->ro_flowid_shift = sc->flowid_shift;
1419 error = priv_check(td, PRIV_NET_LAGG);
1424 * The stride option was added without defining a corresponding
1425 * LAGG_OPT flag, so handle a non-zero value before checking
1426 * anything else to preserve compatibility.
1429 if (ro->ro_opts == 0 && ro->ro_bkt != 0) {
1430 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) {
1435 sc->sc_stride = ro->ro_bkt;
1437 if (ro->ro_opts == 0) {
1443 * Set options. LACP options are stored in sc->sc_psc,
1448 switch (ro->ro_opts) {
1449 case LAGG_OPT_USE_FLOWID:
1450 case -LAGG_OPT_USE_FLOWID:
1451 case LAGG_OPT_USE_NUMA:
1452 case -LAGG_OPT_USE_NUMA:
1453 case LAGG_OPT_FLOWIDSHIFT:
1454 case LAGG_OPT_RR_LIMIT:
1458 case LAGG_OPT_LACP_TXTEST:
1459 case -LAGG_OPT_LACP_TXTEST:
1460 case LAGG_OPT_LACP_RXTEST:
1461 case -LAGG_OPT_LACP_RXTEST:
1462 case LAGG_OPT_LACP_STRICT:
1463 case -LAGG_OPT_LACP_STRICT:
1464 case LAGG_OPT_LACP_FAST_TIMO:
1465 case -LAGG_OPT_LACP_FAST_TIMO:
1474 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) {
1475 /* Invalid combination of options specified. */
1478 break; /* Return from SIOCSLAGGOPTS. */
1482 * Store new options into sc->sc_opts except for
1483 * FLOWIDSHIFT, RR and LACP options.
1486 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT)
1487 sc->flowid_shift = ro->ro_flowid_shift;
1488 else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) {
1489 if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN ||
1495 sc->sc_stride = ro->ro_bkt;
1496 } else if (ro->ro_opts > 0)
1497 sc->sc_opts |= ro->ro_opts;
1499 sc->sc_opts &= ~ro->ro_opts;
1501 struct lacp_softc *lsc;
1502 struct lacp_port *lp;
1504 lsc = (struct lacp_softc *)sc->sc_psc;
1506 switch (ro->ro_opts) {
1507 case LAGG_OPT_LACP_TXTEST:
1508 lsc->lsc_debug.lsc_tx_test = 1;
1510 case -LAGG_OPT_LACP_TXTEST:
1511 lsc->lsc_debug.lsc_tx_test = 0;
1513 case LAGG_OPT_LACP_RXTEST:
1514 lsc->lsc_debug.lsc_rx_test = 1;
1516 case -LAGG_OPT_LACP_RXTEST:
1517 lsc->lsc_debug.lsc_rx_test = 0;
1519 case LAGG_OPT_LACP_STRICT:
1520 lsc->lsc_strict_mode = 1;
1522 case -LAGG_OPT_LACP_STRICT:
1523 lsc->lsc_strict_mode = 0;
1525 case LAGG_OPT_LACP_FAST_TIMO:
1527 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1528 lp->lp_state |= LACP_STATE_TIMEOUT;
1530 lsc->lsc_fast_timeout = 1;
1532 case -LAGG_OPT_LACP_FAST_TIMO:
1534 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next)
1535 lp->lp_state &= ~LACP_STATE_TIMEOUT;
1537 lsc->lsc_fast_timeout = 0;
1543 case SIOCGLAGGFLAGS:
1546 if (sc->sc_flags & MBUF_HASHFLAG_L2)
1547 rf->rf_flags |= LAGG_F_HASHL2;
1548 if (sc->sc_flags & MBUF_HASHFLAG_L3)
1549 rf->rf_flags |= LAGG_F_HASHL3;
1550 if (sc->sc_flags & MBUF_HASHFLAG_L4)
1551 rf->rf_flags |= LAGG_F_HASHL4;
1555 error = priv_check(td, PRIV_NET_LAGG);
1558 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) {
1564 if (rf->rf_flags & LAGG_F_HASHL2)
1565 sc->sc_flags |= MBUF_HASHFLAG_L2;
1566 if (rf->rf_flags & LAGG_F_HASHL3)
1567 sc->sc_flags |= MBUF_HASHFLAG_L3;
1568 if (rf->rf_flags & LAGG_F_HASHL4)
1569 sc->sc_flags |= MBUF_HASHFLAG_L4;
1573 if (rp->rp_portname[0] == '\0' ||
1574 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1579 NET_EPOCH_ENTER(et);
1580 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1581 lp->lp_softc != sc) {
1589 lagg_port2req(lp, rp);
1595 error = priv_check(td, PRIV_NET_LAGG);
1598 if (rp->rp_portname[0] == '\0' ||
1599 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1605 * A laggport interface should not have inet6 address
1606 * because two interfaces with a valid link-local
1607 * scope zone must not be merged in any form. This
1608 * restriction is needed to prevent violation of
1609 * link-local scope zone. Attempts to add a laggport
1610 * interface which has inet6 addresses triggers
1611 * removal of all inet6 addresses on the member
1614 if (in6ifa_llaonifp(tpif)) {
1616 if_printf(sc->sc_ifp,
1617 "IPv6 addresses on %s have been removed "
1618 "before adding it as a member to prevent "
1619 "IPv6 address scope violation.\n",
1623 oldmtu = ifp->if_mtu;
1625 error = lagg_port_create(sc, tpif);
1630 * LAGG MTU may change during addition of the first port.
1631 * If it did, do network layer specific procedure.
1633 if (ifp->if_mtu != oldmtu)
1636 VLAN_CAPABILITIES(ifp);
1638 case SIOCSLAGGDELPORT:
1639 error = priv_check(td, PRIV_NET_LAGG);
1642 if (rp->rp_portname[0] == '\0' ||
1643 (tpif = ifunit_ref(rp->rp_portname)) == NULL) {
1649 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL ||
1650 lp->lp_softc != sc) {
1657 error = lagg_port_destroy(lp, 1);
1660 VLAN_CAPABILITIES(ifp);
1663 /* Set flags on ports too */
1665 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1666 lagg_setflags(lp, 1);
1669 if (!(ifp->if_flags & IFF_UP) &&
1670 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1672 * If interface is marked down and it is running,
1673 * then stop and disable it.
1677 } else if ((ifp->if_flags & IFF_UP) &&
1678 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1680 * If interface is marked up and it is stopped, then
1684 (*ifp->if_init)(sc);
1691 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1700 if (ifp->if_type == IFT_INFINIBAND)
1703 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1709 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1710 if (lp->lp_ioctl != NULL)
1711 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1713 lagg_capabilities(sc);
1715 VLAN_CAPABILITIES(ifp);
1725 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1726 if (lp->lp_ioctl != NULL)
1727 error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1732 "failed to change MTU to %d on port %s, "
1733 "reverting all ports to original MTU (%d)\n",
1734 ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu);
1739 ifp->if_mtu = ifr->ifr_mtu;
1741 /* set every port back to the original MTU */
1742 ifr->ifr_mtu = ifp->if_mtu;
1743 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
1744 if (lp->lp_ioctl != NULL)
1745 (*lp->lp_ioctl)(lp->lp_ifp, cmd, data);
1748 lagg_capabilities(sc);
1750 VLAN_CAPABILITIES(ifp);
1754 error = ether_ioctl(ifp, cmd, data);
1760 #if defined(KERN_TLS) || defined(RATELIMIT)
1762 static const struct if_snd_tag_sw lagg_snd_tag_ul_sw = {
1763 .snd_tag_modify = lagg_snd_tag_modify,
1764 .snd_tag_query = lagg_snd_tag_query,
1765 .snd_tag_free = lagg_snd_tag_free,
1766 .next_snd_tag = lagg_next_snd_tag,
1767 .type = IF_SND_TAG_TYPE_UNLIMITED
1770 static const struct if_snd_tag_sw lagg_snd_tag_rl_sw = {
1771 .snd_tag_modify = lagg_snd_tag_modify,
1772 .snd_tag_query = lagg_snd_tag_query,
1773 .snd_tag_free = lagg_snd_tag_free,
1774 .next_snd_tag = lagg_next_snd_tag,
1775 .type = IF_SND_TAG_TYPE_RATE_LIMIT
1780 static const struct if_snd_tag_sw lagg_snd_tag_tls_sw = {
1781 .snd_tag_modify = lagg_snd_tag_modify,
1782 .snd_tag_query = lagg_snd_tag_query,
1783 .snd_tag_free = lagg_snd_tag_free,
1784 .next_snd_tag = lagg_next_snd_tag,
1785 .type = IF_SND_TAG_TYPE_TLS
1789 static const struct if_snd_tag_sw lagg_snd_tag_tls_rl_sw = {
1790 .snd_tag_modify = lagg_snd_tag_modify,
1791 .snd_tag_query = lagg_snd_tag_query,
1792 .snd_tag_free = lagg_snd_tag_free,
1793 .next_snd_tag = lagg_next_snd_tag,
1794 .type = IF_SND_TAG_TYPE_TLS_RATE_LIMIT
1799 static inline struct lagg_snd_tag *
1800 mst_to_lst(struct m_snd_tag *mst)
1803 return (__containerof(mst, struct lagg_snd_tag, com));
1807 * Look up the port used by a specific flow. This only works for lagg
1808 * protocols with deterministic port mappings (e.g. not roundrobin).
1809 * In addition protocols which use a hash to map flows to ports must
1810 * be configured to use the mbuf flowid rather than hashing packet
1813 static struct lagg_port *
1814 lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype,
1815 uint8_t numa_domain)
1817 struct lagg_softc *sc;
1818 struct lagg_port *lp;
1825 switch (sc->sc_proto) {
1826 case LAGG_PROTO_FAILOVER:
1827 return (lagg_link_active(sc, sc->sc_primary));
1828 case LAGG_PROTO_LOADBALANCE:
1829 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1830 flowtype == M_HASHTYPE_NONE)
1832 p = flowid >> sc->flowid_shift;
1834 lb = (struct lagg_lb *)sc->sc_psc;
1835 lp = lb->lb_ports[p];
1836 return (lagg_link_active(sc, lp));
1837 case LAGG_PROTO_LACP:
1838 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 ||
1839 flowtype == M_HASHTYPE_NONE)
1841 hash = flowid >> sc->flowid_shift;
1842 return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, &err));
1849 lagg_snd_tag_alloc(struct ifnet *ifp,
1850 union if_snd_tag_alloc_params *params,
1851 struct m_snd_tag **ppmt)
1853 struct epoch_tracker et;
1854 const struct if_snd_tag_sw *sw;
1855 struct lagg_snd_tag *lst;
1856 struct lagg_port *lp;
1857 struct ifnet *lp_ifp;
1858 struct m_snd_tag *mst;
1861 switch (params->hdr.type) {
1863 case IF_SND_TAG_TYPE_UNLIMITED:
1864 sw = &lagg_snd_tag_ul_sw;
1866 case IF_SND_TAG_TYPE_RATE_LIMIT:
1867 sw = &lagg_snd_tag_rl_sw;
1871 case IF_SND_TAG_TYPE_TLS:
1872 sw = &lagg_snd_tag_tls_sw;
1874 case IF_SND_TAG_TYPE_TLS_RX:
1875 /* Return tag from port interface directly. */
1879 case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
1880 sw = &lagg_snd_tag_tls_rl_sw;
1885 return (EOPNOTSUPP);
1888 NET_EPOCH_ENTER(et);
1889 lp = lookup_snd_tag_port(ifp, params->hdr.flowid,
1890 params->hdr.flowtype, params->hdr.numa_domain);
1893 return (EOPNOTSUPP);
1895 if (lp->lp_ifp == NULL) {
1897 return (EOPNOTSUPP);
1899 lp_ifp = lp->lp_ifp;
1904 lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT);
1912 error = m_snd_tag_alloc(lp_ifp, params, &mst);
1920 m_snd_tag_init(&lst->com, ifp, sw);
1930 static struct m_snd_tag *
1931 lagg_next_snd_tag(struct m_snd_tag *mst)
1933 struct lagg_snd_tag *lst;
1935 lst = mst_to_lst(mst);
1940 lagg_snd_tag_modify(struct m_snd_tag *mst,
1941 union if_snd_tag_modify_params *params)
1943 struct lagg_snd_tag *lst;
1945 lst = mst_to_lst(mst);
1946 return (lst->tag->sw->snd_tag_modify(lst->tag, params));
1950 lagg_snd_tag_query(struct m_snd_tag *mst,
1951 union if_snd_tag_query_params *params)
1953 struct lagg_snd_tag *lst;
1955 lst = mst_to_lst(mst);
1956 return (lst->tag->sw->snd_tag_query(lst->tag, params));
1960 lagg_snd_tag_free(struct m_snd_tag *mst)
1962 struct lagg_snd_tag *lst;
1964 lst = mst_to_lst(mst);
1965 m_snd_tag_rele(lst->tag);
1970 lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
1973 * For lagg, we have an indirect
1974 * interface. The caller needs to
1975 * get a ratelimit tag on the actual
1976 * interface the flow will go on.
1978 q->rate_table = NULL;
1979 q->flags = RT_IS_INDIRECT;
1981 q->number_of_rates = 0;
1986 lagg_setmulti(struct lagg_port *lp)
1988 struct lagg_softc *sc = lp->lp_softc;
1989 struct ifnet *ifp = lp->lp_ifp;
1990 struct ifnet *scifp = sc->sc_ifp;
1992 struct ifmultiaddr *ifma;
1995 IF_ADDR_WLOCK(scifp);
1996 CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
1997 if (ifma->ifma_addr->sa_family != AF_LINK)
1999 mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT);
2001 IF_ADDR_WUNLOCK(scifp);
2004 bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len);
2005 mc->mc_addr.sdl_index = ifp->if_index;
2007 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries);
2009 IF_ADDR_WUNLOCK(scifp);
2010 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) {
2011 error = if_addmulti(ifp,
2012 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma);
2020 lagg_clrmulti(struct lagg_port *lp)
2024 LAGG_XLOCK_ASSERT(lp->lp_softc);
2025 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
2026 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
2027 if (mc->mc_ifma && lp->lp_detaching == 0)
2028 if_delmulti_ifma(mc->mc_ifma);
2035 lagg_setcaps(struct lagg_port *lp, int cap, int cap2)
2038 struct siocsifcapnv_driver_data drv_ioctl_data;
2040 if (lp->lp_ifp->if_capenable == cap &&
2041 lp->lp_ifp->if_capenable2 == cap2)
2043 if (lp->lp_ioctl == NULL)
2046 if ((lp->lp_ifp->if_capabilities & IFCAP_NV) != 0) {
2047 drv_ioctl_data.reqcap = cap;
2048 drv_ioctl_data.reqcap2 = cap2;
2049 drv_ioctl_data.nvcap = NULL;
2050 (*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAPNV,
2051 (caddr_t)&drv_ioctl_data);
2053 ifr.ifr_reqcap = cap;
2054 (*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr);
2058 /* Handle a ref counted flag that should be set on the lagg port as well */
2060 lagg_setflag(struct lagg_port *lp, int flag, int status,
2061 int (*func)(struct ifnet *, int))
2063 struct lagg_softc *sc = lp->lp_softc;
2064 struct ifnet *scifp = sc->sc_ifp;
2065 struct ifnet *ifp = lp->lp_ifp;
2068 LAGG_XLOCK_ASSERT(sc);
2070 status = status ? (scifp->if_flags & flag) : 0;
2071 /* Now "status" contains the flag value or 0 */
2074 * See if recorded ports status is different from what
2075 * we want it to be. If it is, flip it. We record ports
2076 * status in lp_ifflags so that we won't clear ports flag
2077 * we haven't set. In fact, we don't clear or set ports
2078 * flags directly, but get or release references to them.
2079 * That's why we can be sure that recorded flags still are
2080 * in accord with actual ports flags.
2082 if (status != (lp->lp_ifflags & flag)) {
2083 error = (*func)(ifp, status);
2086 lp->lp_ifflags &= ~flag;
2087 lp->lp_ifflags |= status;
2093 * Handle IFF_* flags that require certain changes on the lagg port
2094 * if "status" is true, update ports flags respective to the lagg
2095 * if "status" is false, forcedly clear the flags set on port.
2098 lagg_setflags(struct lagg_port *lp, int status)
2102 for (i = 0; lagg_pflags[i].flag; i++) {
2103 error = lagg_setflag(lp, lagg_pflags[i].flag,
2104 status, lagg_pflags[i].func);
2112 lagg_transmit_ethernet(struct ifnet *ifp, struct mbuf *m)
2114 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2117 #if defined(KERN_TLS) || defined(RATELIMIT)
2118 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2119 MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2121 /* We need a Tx algorithm and at least one port */
2122 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2124 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2128 ETHER_BPF_MTAP(ifp, m);
2130 return (lagg_proto_start(sc, m));
2134 lagg_transmit_infiniband(struct ifnet *ifp, struct mbuf *m)
2136 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2139 #if defined(KERN_TLS) || defined(RATELIMIT)
2140 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2141 MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2143 /* We need a Tx algorithm and at least one port */
2144 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) {
2146 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2150 infiniband_bpf_mtap(ifp, m);
2152 return (lagg_proto_start(sc, m));
2156 * The ifp->if_qflush entry point for lagg(4) is no-op.
2159 lagg_qflush(struct ifnet *ifp __unused)
2163 static struct mbuf *
2164 lagg_input_ethernet(struct ifnet *ifp, struct mbuf *m)
2166 struct lagg_port *lp = ifp->if_lagg;
2167 struct lagg_softc *sc = lp->lp_softc;
2168 struct ifnet *scifp = sc->sc_ifp;
2171 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2172 lp->lp_detaching != 0 ||
2173 sc->sc_proto == LAGG_PROTO_NONE) {
2178 m = lagg_proto_input(sc, lp, m);
2180 ETHER_BPF_MTAP(scifp, m);
2182 if ((scifp->if_flags & IFF_MONITOR) != 0) {
2189 if (m != NULL && scifp->if_capenable & IFCAP_NETMAP) {
2190 scifp->if_input(scifp, m);
2193 #endif /* DEV_NETMAP */
2198 static struct mbuf *
2199 lagg_input_infiniband(struct ifnet *ifp, struct mbuf *m)
2201 struct lagg_port *lp = ifp->if_lagg;
2202 struct lagg_softc *sc = lp->lp_softc;
2203 struct ifnet *scifp = sc->sc_ifp;
2206 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
2207 lp->lp_detaching != 0 ||
2208 sc->sc_proto == LAGG_PROTO_NONE) {
2213 m = lagg_proto_input(sc, lp, m);
2215 infiniband_bpf_mtap(scifp, m);
2217 if ((scifp->if_flags & IFF_MONITOR) != 0) {
2227 lagg_media_change(struct ifnet *ifp)
2229 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2231 if (sc->sc_ifflags & IFF_DEBUG)
2232 printf("%s\n", __func__);
2239 lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr)
2241 struct epoch_tracker et;
2242 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc;
2243 struct lagg_port *lp;
2245 imr->ifm_status = IFM_AVALID;
2246 imr->ifm_active = IFM_ETHER | IFM_AUTO;
2248 NET_EPOCH_ENTER(et);
2249 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2250 if (LAGG_PORTACTIVE(lp))
2251 imr->ifm_status |= IFM_ACTIVE;
2257 lagg_linkstate(struct lagg_softc *sc)
2259 struct epoch_tracker et;
2260 struct lagg_port *lp;
2261 int new_link = LINK_STATE_DOWN;
2264 LAGG_XLOCK_ASSERT(sc);
2266 /* LACP handles link state itself */
2267 if (sc->sc_proto == LAGG_PROTO_LACP)
2270 /* Our link is considered up if at least one of our ports is active */
2271 NET_EPOCH_ENTER(et);
2272 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2273 if (lp->lp_ifp->if_link_state == LINK_STATE_UP) {
2274 new_link = LINK_STATE_UP;
2279 if_link_state_change(sc->sc_ifp, new_link);
2281 /* Update if_baudrate to reflect the max possible speed */
2282 switch (sc->sc_proto) {
2283 case LAGG_PROTO_FAILOVER:
2284 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ?
2285 sc->sc_primary->lp_ifp->if_baudrate : 0;
2287 case LAGG_PROTO_ROUNDROBIN:
2288 case LAGG_PROTO_LOADBALANCE:
2289 case LAGG_PROTO_BROADCAST:
2291 NET_EPOCH_ENTER(et);
2292 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2293 speed += lp->lp_ifp->if_baudrate;
2295 sc->sc_ifp->if_baudrate = speed;
2297 case LAGG_PROTO_LACP:
2298 /* LACP updates if_baudrate itself */
2304 lagg_port_state(struct ifnet *ifp, int state)
2306 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg;
2307 struct lagg_softc *sc = NULL;
2316 lagg_proto_linkstate(sc, lp);
2321 lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp)
2323 struct lagg_port *lp_next, *rval = NULL;
2326 * Search a port which reports an active link state.
2331 * This is called with either in the network epoch
2332 * or with LAGG_XLOCK(sc) held.
2334 if (!in_epoch(net_epoch_preempt))
2335 LAGG_XLOCK_ASSERT(sc);
2340 if (LAGG_PORTACTIVE(lp)) {
2344 if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL &&
2345 LAGG_PORTACTIVE(lp_next)) {
2351 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2352 if (LAGG_PORTACTIVE(lp_next)) {
2361 lagg_enqueue(struct ifnet *ifp, struct mbuf *m)
2364 #if defined(KERN_TLS) || defined(RATELIMIT)
2365 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2366 struct lagg_snd_tag *lst;
2367 struct m_snd_tag *mst;
2369 mst = m->m_pkthdr.snd_tag;
2370 lst = mst_to_lst(mst);
2371 if (lst->tag->ifp != ifp) {
2375 m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag);
2376 m_snd_tag_rele(mst);
2379 return (ifp->if_transmit)(ifp, m);
2383 * Simple round robin aggregation
2386 lagg_rr_attach(struct lagg_softc *sc)
2393 lagg_rr_start(struct lagg_softc *sc, struct mbuf *m)
2395 struct lagg_port *lp;
2398 p = atomic_fetchadd_32(&sc->sc_seq, 1);
2401 lp = CK_SLIST_FIRST(&sc->sc_ports);
2404 lp = CK_SLIST_NEXT(lp, lp_entries);
2407 * Check the port's link state. This will return the next active
2408 * port if the link is down or the port is NULL.
2410 if ((lp = lagg_link_active(sc, lp)) == NULL) {
2411 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2417 return (lagg_enqueue(lp->lp_ifp, m));
2424 lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m)
2428 struct lagg_port *lp, *last = NULL;
2432 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
2433 if (!LAGG_PORTACTIVE(lp))
2437 m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT);
2443 lagg_enqueue(last->lp_ifp, m0);
2449 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2453 if ((last = lagg_link_active(sc, last)) == NULL) {
2455 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2460 ret = lagg_enqueue(last->lp_ifp, m);
2462 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, errors);
2471 lagg_fail_start(struct lagg_softc *sc, struct mbuf *m)
2473 struct lagg_port *lp;
2475 /* Use the master port if active or the next available port */
2476 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) {
2477 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2483 return (lagg_enqueue(lp->lp_ifp, m));
2486 static struct mbuf *
2487 lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2489 struct ifnet *ifp = sc->sc_ifp;
2490 struct lagg_port *tmp_tp;
2492 if (lp == sc->sc_primary || V_lagg_failover_rx_all) {
2493 m->m_pkthdr.rcvif = ifp;
2497 if (!LAGG_PORTACTIVE(sc->sc_primary)) {
2498 tmp_tp = lagg_link_active(sc, sc->sc_primary);
2500 * If tmp_tp is null, we've received a packet when all
2501 * our links are down. Weird, but process it anyways.
2503 if (tmp_tp == NULL || tmp_tp == lp) {
2504 m->m_pkthdr.rcvif = ifp;
2517 lagg_lb_attach(struct lagg_softc *sc)
2519 struct lagg_port *lp;
2522 LAGG_XLOCK_ASSERT(sc);
2523 lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO);
2524 lb->lb_key = m_ether_tcpip_hash_init();
2527 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2528 lagg_lb_port_create(lp);
2532 lagg_lb_detach(struct lagg_softc *sc)
2536 lb = (struct lagg_lb *)sc->sc_psc;
2542 lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp)
2544 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2545 struct lagg_port *lp_next;
2549 bzero(&lb->lb_ports, sizeof(lb->lb_ports));
2550 LAGG_XLOCK_ASSERT(sc);
2551 CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) {
2554 if (i >= LAGG_MAX_PORTS) {
2558 if (sc->sc_ifflags & IFF_DEBUG)
2559 printf("%s: port %s at index %d\n",
2560 sc->sc_ifname, lp_next->lp_ifp->if_xname, i);
2561 lb->lb_ports[i++] = lp_next;
2568 lagg_lb_port_create(struct lagg_port *lp)
2570 struct lagg_softc *sc = lp->lp_softc;
2571 return (lagg_lb_porttable(sc, NULL));
2575 lagg_lb_port_destroy(struct lagg_port *lp)
2577 struct lagg_softc *sc = lp->lp_softc;
2578 lagg_lb_porttable(sc, lp);
2582 lagg_lb_start(struct lagg_softc *sc, struct mbuf *m)
2584 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc;
2585 struct lagg_port *lp = NULL;
2588 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
2589 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2590 p = m->m_pkthdr.flowid >> sc->flowid_shift;
2592 p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key);
2594 lp = lb->lb_ports[p];
2597 * Check the port's link state. This will return the next active
2598 * port if the link is down or the port is NULL.
2600 if ((lp = lagg_link_active(sc, lp)) == NULL) {
2601 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2607 return (lagg_enqueue(lp->lp_ifp, m));
2614 lagg_lacp_attach(struct lagg_softc *sc)
2616 struct lagg_port *lp;
2619 LAGG_XLOCK_ASSERT(sc);
2620 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2621 lacp_port_create(lp);
2625 lagg_lacp_detach(struct lagg_softc *sc)
2627 struct lagg_port *lp;
2630 LAGG_XLOCK_ASSERT(sc);
2631 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2632 lacp_port_destroy(lp);
2640 lagg_lacp_lladdr(struct lagg_softc *sc)
2642 struct lagg_port *lp;
2644 LAGG_SXLOCK_ASSERT(sc);
2646 /* purge all the lacp ports */
2647 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2648 lacp_port_destroy(lp);
2650 /* add them back in */
2651 CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
2652 lacp_port_create(lp);
2656 lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m)
2658 struct lagg_port *lp;
2661 lp = lacp_select_tx_port(sc, m, &err);
2663 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
2669 return (lagg_enqueue(lp->lp_ifp, m));
2672 static struct mbuf *
2673 lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2675 struct ifnet *ifp = sc->sc_ifp;
2676 struct ether_header *eh;
2679 eh = mtod(m, struct ether_header *);
2680 etype = ntohs(eh->ether_type);
2682 /* Tap off LACP control messages */
2683 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) {
2684 m = lacp_input(lp, m);
2690 * If the port is not collecting or not in the active aggregator then
2693 if (!lacp_iscollecting(lp) || !lacp_isactive(lp)) {
2698 m->m_pkthdr.rcvif = ifp;
2703 static struct mbuf *
2704 lagg_default_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m)
2706 struct ifnet *ifp = sc->sc_ifp;
2708 /* Just pass in the packet to our lagg device */
2709 m->m_pkthdr.rcvif = ifp;