2 * Copyright (c) 1980, 1986, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)if.c 8.3 (Berkeley) 1/4/94
30 * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
33 #include "opt_compat.h"
34 #include "opt_inet6.h"
36 #include "opt_ifpoll.h"
38 #include <sys/param.h>
39 #include <sys/malloc.h>
41 #include <sys/systm.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/socketops.h>
48 #include <sys/protosw.h>
49 #include <sys/kernel.h>
51 #include <sys/mutex.h>
52 #include <sys/sockio.h>
53 #include <sys/syslog.h>
54 #include <sys/sysctl.h>
55 #include <sys/domain.h>
56 #include <sys/thread.h>
57 #include <sys/serialize.h>
60 #include <sys/thread2.h>
61 #include <sys/msgport2.h>
62 #include <sys/mutex2.h>
65 #include <net/if_arp.h>
66 #include <net/if_dl.h>
67 #include <net/if_types.h>
68 #include <net/if_var.h>
69 #include <net/ifq_var.h>
70 #include <net/radix.h>
71 #include <net/route.h>
72 #include <net/if_clone.h>
73 #include <net/netisr2.h>
74 #include <net/netmsg2.h>
76 #include <machine/atomic.h>
77 #include <machine/stdarg.h>
78 #include <machine/smp.h>
80 #if defined(INET) || defined(INET6)
82 #include <netinet/in.h>
83 #include <netinet/in_var.h>
84 #include <netinet/if_ether.h>
86 #include <netinet6/in6_var.h>
87 #include <netinet6/in6_ifattach.h>
91 #if defined(COMPAT_43)
92 #include <emulation/43bsd/43bsd_socket.h>
93 #endif /* COMPAT_43 */
95 struct netmsg_ifaddr {
96 struct netmsg_base base;
102 struct ifsubq_stage_head {
103 TAILQ_HEAD(, ifsubq_stage) stg_head;
107 * System initialization
109 static void if_attachdomain(void *);
110 static void if_attachdomain1(struct ifnet *);
111 static int ifconf(u_long, caddr_t, struct ucred *);
112 static void ifinit(void *);
113 static void ifnetinit(void *);
114 static void if_slowtimo(void *);
115 static void link_rtrequest(int, struct rtentry *);
116 static int if_rtdel(struct radix_node *, void *);
118 /* Helper functions */
119 static void ifsq_watchdog_reset(struct ifsubq_watchdog *);
123 * XXX: declare here to avoid to include many inet6 related files..
124 * should be more generalized?
126 extern void nd6_setmtu(struct ifnet *);
129 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
130 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
132 static int ifsq_stage_cntmax = 4;
133 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
134 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
135 &ifsq_stage_cntmax, 0, "ifq staging packet count max");
137 static int if_stats_compat = 0;
138 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
139 &if_stats_compat, 0, "Compat the old ifnet stats");
141 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
142 /* Must be after netisr_init */
143 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
145 static if_com_alloc_t *if_com_alloc[256];
146 static if_com_free_t *if_com_free[256];
148 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
149 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
150 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
152 int ifqmaxlen = IFQ_MAXLEN;
153 struct ifnethead ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
155 struct callout if_slowtimo_timer;
158 struct ifnet **ifindex2ifnet = NULL;
159 static struct thread ifnet_threads[MAXCPU];
161 static struct ifsubq_stage_head ifsubq_stage_heads[MAXCPU];
164 #define IFQ_KTR_STRING "ifq=%p"
165 #define IFQ_KTR_ARGS struct ifaltq *ifq
167 #define KTR_IFQ KTR_ALL
169 KTR_INFO_MASTER(ifq);
170 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
171 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
172 #define logifq(name, arg) KTR_LOG(ifq_ ## name, arg)
174 #define IF_START_KTR_STRING "ifp=%p"
175 #define IF_START_KTR_ARGS struct ifnet *ifp
177 #define KTR_IF_START KTR_ALL
179 KTR_INFO_MASTER(if_start);
180 KTR_INFO(KTR_IF_START, if_start, run, 0,
181 IF_START_KTR_STRING, IF_START_KTR_ARGS);
182 KTR_INFO(KTR_IF_START, if_start, sched, 1,
183 IF_START_KTR_STRING, IF_START_KTR_ARGS);
184 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
185 IF_START_KTR_STRING, IF_START_KTR_ARGS);
186 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
187 IF_START_KTR_STRING, IF_START_KTR_ARGS);
188 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
189 IF_START_KTR_STRING, IF_START_KTR_ARGS);
190 #define logifstart(name, arg) KTR_LOG(if_start_ ## name, arg)
193 TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
196 * Network interface utility routines.
198 * Routines with ifa_ifwith* names take sockaddr *'s as
207 callout_init(&if_slowtimo_timer);
210 TAILQ_FOREACH(ifp, &ifnet, if_link) {
211 if (ifp->if_snd.altq_maxlen == 0) {
212 if_printf(ifp, "XXX: driver didn't set altq_maxlen\n");
213 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
222 ifsq_ifstart_ipifunc(void *arg)
224 struct ifaltq_subque *ifsq = arg;
225 struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
228 if (lmsg->ms_flags & MSGF_DONE)
229 lwkt_sendmsg(netisr_cpuport(mycpuid), lmsg);
234 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
236 KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
237 TAILQ_REMOVE(&head->stg_head, stage, stg_link);
238 stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
244 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
246 KKASSERT((stage->stg_flags &
247 (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
248 stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
249 TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
253 * Schedule ifnet.if_start on the subqueue owner CPU
256 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
260 if (!force && curthread->td_type == TD_TYPE_NETISR &&
261 ifsq_stage_cntmax > 0) {
262 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
266 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
267 ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
268 stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
272 cpu = ifsq_get_cpuid(ifsq);
274 lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
276 ifsq_ifstart_ipifunc(ifsq);
281 * This function will release ifnet.if_start subqueue interlock,
282 * if ifnet.if_start for the subqueue does not need to be scheduled
285 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
287 if (!running || ifsq_is_empty(ifsq)
289 || ifsq->ifsq_altq->altq_tbr != NULL
294 * ifnet.if_start subqueue interlock is released, if:
295 * 1) Hardware can not take any packets, due to
296 * o interface is marked down
297 * o hardware queue is full (ifsq_is_oactive)
298 * Under the second situation, hardware interrupt
299 * or polling(4) will call/schedule ifnet.if_start
300 * on the subqueue when hardware queue is ready
301 * 2) There is no packet in the subqueue.
302 * Further ifq_dispatch or ifq_handoff will call/
303 * schedule ifnet.if_start on the subqueue.
304 * 3) TBR is used and it does not allow further
306 * TBR callout will call ifnet.if_start on the
309 if (!running || !ifsq_data_ready(ifsq)) {
310 ifsq_clr_started(ifsq);
311 ALTQ_SQ_UNLOCK(ifsq);
314 ALTQ_SQ_UNLOCK(ifsq);
320 ifsq_ifstart_dispatch(netmsg_t msg)
322 struct lwkt_msg *lmsg = &msg->base.lmsg;
323 struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
324 struct ifnet *ifp = ifsq_get_ifp(ifsq);
325 int running = 0, need_sched;
328 lwkt_replymsg(lmsg, 0); /* reply ASAP */
331 if (mycpuid != ifsq_get_cpuid(ifsq)) {
333 * We need to chase the subqueue owner CPU change.
335 ifsq_ifstart_schedule(ifsq, 1);
339 ifsq_serialize_hw(ifsq);
340 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
341 ifp->if_start(ifp, ifsq);
342 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
345 need_sched = ifsq_ifstart_need_schedule(ifsq, running);
346 ifsq_deserialize_hw(ifsq);
350 * More data need to be transmitted, ifnet.if_start is
351 * scheduled on the subqueue owner CPU, and we keep going.
352 * NOTE: ifnet.if_start subqueue interlock is not released.
354 ifsq_ifstart_schedule(ifsq, 0);
358 /* Device driver ifnet.if_start helper function */
360 ifsq_devstart(struct ifaltq_subque *ifsq)
362 struct ifnet *ifp = ifsq_get_ifp(ifsq);
365 ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
368 if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
369 ALTQ_SQ_UNLOCK(ifsq);
372 ifsq_set_started(ifsq);
373 ALTQ_SQ_UNLOCK(ifsq);
375 ifp->if_start(ifp, ifsq);
377 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
380 if (ifsq_ifstart_need_schedule(ifsq, running)) {
382 * More data need to be transmitted, ifnet.if_start is
383 * scheduled on ifnet's CPU, and we keep going.
384 * NOTE: ifnet.if_start interlock is not released.
386 ifsq_ifstart_schedule(ifsq, 0);
391 if_devstart(struct ifnet *ifp)
393 ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
396 /* Device driver ifnet.if_start schedule helper function */
398 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
400 ifsq_ifstart_schedule(ifsq, 1);
404 if_devstart_sched(struct ifnet *ifp)
406 ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
410 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
412 lwkt_serialize_enter(ifp->if_serializer);
416 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
418 lwkt_serialize_exit(ifp->if_serializer);
422 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
424 return lwkt_serialize_try(ifp->if_serializer);
429 if_default_serialize_assert(struct ifnet *ifp,
430 enum ifnet_serialize slz __unused,
431 boolean_t serialized)
434 ASSERT_SERIALIZED(ifp->if_serializer);
436 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
441 * Attach an interface to the list of "active" interfaces.
443 * The serializer is optional.
446 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
448 unsigned socksize, ifasize;
449 int namelen, masklen;
450 struct sockaddr_dl *sdl;
455 static int if_indexlim = 8;
457 if (ifp->if_serialize != NULL) {
458 KASSERT(ifp->if_deserialize != NULL &&
459 ifp->if_tryserialize != NULL &&
460 ifp->if_serialize_assert != NULL,
461 ("serialize functions are partially setup"));
464 * If the device supplies serialize functions,
465 * then clear if_serializer to catch any invalid
466 * usage of this field.
468 KASSERT(serializer == NULL,
469 ("both serialize functions and default serializer "
471 ifp->if_serializer = NULL;
473 KASSERT(ifp->if_deserialize == NULL &&
474 ifp->if_tryserialize == NULL &&
475 ifp->if_serialize_assert == NULL,
476 ("serialize functions are partially setup"));
477 ifp->if_serialize = if_default_serialize;
478 ifp->if_deserialize = if_default_deserialize;
479 ifp->if_tryserialize = if_default_tryserialize;
481 ifp->if_serialize_assert = if_default_serialize_assert;
485 * The serializer can be passed in from the device,
486 * allowing the same serializer to be used for both
487 * the interrupt interlock and the device queue.
488 * If not specified, the netif structure will use an
489 * embedded serializer.
491 if (serializer == NULL) {
492 serializer = &ifp->if_default_serializer;
493 lwkt_serialize_init(serializer);
495 ifp->if_serializer = serializer;
498 mtx_init(&ifp->if_ioctl_mtx);
499 mtx_lock(&ifp->if_ioctl_mtx);
501 TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
502 ifp->if_index = ++if_index;
506 * The old code would work if the interface passed a pre-existing
507 * chain of ifaddrs to this code. We don't trust our callers to
508 * properly initialize the tailq, however, so we no longer allow
509 * this unlikely case.
511 ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
512 M_IFADDR, M_WAITOK | M_ZERO);
513 for (i = 0; i < ncpus; ++i)
514 TAILQ_INIT(&ifp->if_addrheads[i]);
516 TAILQ_INIT(&ifp->if_prefixhead);
517 TAILQ_INIT(&ifp->if_multiaddrs);
518 TAILQ_INIT(&ifp->if_groups);
519 getmicrotime(&ifp->if_lastchange);
520 if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
526 /* grow ifindex2ifnet */
527 n = if_indexlim * sizeof(*q);
528 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
530 bcopy(ifindex2ifnet, q, n/2);
531 kfree(ifindex2ifnet, M_IFADDR);
536 ifindex2ifnet[if_index] = ifp;
539 * create a Link Level name for this device
541 namelen = strlen(ifp->if_xname);
542 masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
543 socksize = masklen + ifp->if_addrlen;
544 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
545 if (socksize < sizeof(*sdl))
546 socksize = sizeof(*sdl);
547 socksize = ROUNDUP(socksize);
549 ifasize = sizeof(struct ifaddr) + 2 * socksize;
550 ifa = ifa_create(ifasize, M_WAITOK);
551 sdl = (struct sockaddr_dl *)(ifa + 1);
552 sdl->sdl_len = socksize;
553 sdl->sdl_family = AF_LINK;
554 bcopy(ifp->if_xname, sdl->sdl_data, namelen);
555 sdl->sdl_nlen = namelen;
556 sdl->sdl_index = ifp->if_index;
557 sdl->sdl_type = ifp->if_type;
558 ifp->if_lladdr = ifa;
560 ifa->ifa_rtrequest = link_rtrequest;
561 ifa->ifa_addr = (struct sockaddr *)sdl;
562 sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
563 ifa->ifa_netmask = (struct sockaddr *)sdl;
564 sdl->sdl_len = masklen;
566 sdl->sdl_data[--namelen] = 0xff;
567 ifa_iflink(ifa, ifp, 0 /* Insert head */);
569 ifp->if_data_pcpu = kmalloc_cachealign(
570 ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
572 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
573 devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
575 if (ifp->if_mapsubq == NULL)
576 ifp->if_mapsubq = ifq_mapsubq_default;
580 ifq->altq_disc = NULL;
581 ifq->altq_flags &= ALTQF_CANTCHANGE;
582 ifq->altq_tbr = NULL;
585 if (ifq->altq_subq_cnt <= 0)
586 ifq->altq_subq_cnt = 1;
587 ifq->altq_subq = kmalloc_cachealign(
588 ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
589 M_DEVBUF, M_WAITOK | M_ZERO);
591 if (ifq->altq_maxlen == 0) {
592 if_printf(ifp, "driver didn't set altq_maxlen\n");
593 ifq_set_maxlen(ifq, ifqmaxlen);
596 for (q = 0; q < ifq->altq_subq_cnt; ++q) {
597 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
599 ALTQ_SQ_LOCK_INIT(ifsq);
600 ifsq->ifsq_index = q;
602 ifsq->ifsq_altq = ifq;
603 ifsq->ifsq_ifp = ifp;
605 ifsq->ifsq_maxlen = ifq->altq_maxlen;
606 ifsq->ifsq_prepended = NULL;
607 ifsq->ifsq_started = 0;
608 ifsq->ifsq_hw_oactive = 0;
609 ifsq_set_cpuid(ifsq, 0);
610 if (ifp->if_serializer != NULL)
611 ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
614 kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
615 M_DEVBUF, M_WAITOK | M_ZERO);
616 for (i = 0; i < ncpus; ++i)
617 ifsq->ifsq_stage[i].stg_subq = ifsq;
619 ifsq->ifsq_ifstart_nmsg =
620 kmalloc(ncpus * sizeof(struct netmsg_base),
621 M_LWKTMSG, M_WAITOK);
622 for (i = 0; i < ncpus; ++i) {
623 netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
624 &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
625 ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
628 ifq_set_classic(ifq);
630 if (!SLIST_EMPTY(&domains))
631 if_attachdomain1(ifp);
633 /* Announce the interface. */
634 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
636 mtx_unlock(&ifp->if_ioctl_mtx);
640 if_attachdomain(void *dummy)
645 TAILQ_FOREACH(ifp, &ifnet, if_list)
646 if_attachdomain1(ifp);
649 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
650 if_attachdomain, NULL);
653 if_attachdomain1(struct ifnet *ifp)
659 /* address family dependent data region */
660 bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
661 SLIST_FOREACH(dp, &domains, dom_next)
662 if (dp->dom_ifattach)
663 ifp->if_afdata[dp->dom_family] =
664 (*dp->dom_ifattach)(ifp);
669 * Purge all addresses whose type is _not_ AF_LINK
672 if_purgeaddrs_nolink(struct ifnet *ifp)
674 struct ifaddr_container *ifac, *next;
676 TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
678 struct ifaddr *ifa = ifac->ifa;
680 /* Leave link ifaddr as it is */
681 if (ifa->ifa_addr->sa_family == AF_LINK)
684 /* XXX: Ugly!! ad hoc just for INET */
685 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
686 struct ifaliasreq ifr;
687 #ifdef IFADDR_DEBUG_VERBOSE
690 kprintf("purge in4 addr %p: ", ifa);
691 for (i = 0; i < ncpus; ++i)
692 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
696 bzero(&ifr, sizeof ifr);
697 ifr.ifra_addr = *ifa->ifa_addr;
698 if (ifa->ifa_dstaddr)
699 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
700 if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
706 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
707 #ifdef IFADDR_DEBUG_VERBOSE
710 kprintf("purge in6 addr %p: ", ifa);
711 for (i = 0; i < ncpus; ++i)
712 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
717 /* ifp_addrhead is already updated */
721 ifa_ifunlink(ifa, ifp);
727 ifq_stage_detach_handler(netmsg_t nmsg)
729 struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
732 for (q = 0; q < ifq->altq_subq_cnt; ++q) {
733 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
734 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
736 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
737 ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
739 lwkt_replymsg(&nmsg->lmsg, 0);
743 ifq_stage_detach(struct ifaltq *ifq)
745 struct netmsg_base base;
748 netmsg_init(&base, NULL, &curthread->td_msgport, 0,
749 ifq_stage_detach_handler);
750 base.lmsg.u.ms_resultp = ifq;
752 for (cpu = 0; cpu < ncpus; ++cpu)
753 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
756 struct netmsg_if_rtdel {
757 struct netmsg_base base;
762 if_rtdel_dispatch(netmsg_t msg)
764 struct netmsg_if_rtdel *rmsg = (void *)msg;
768 for (i = 1; i <= AF_MAX; i++) {
769 struct radix_node_head *rnh;
771 if ((rnh = rt_tables[cpu][i]) == NULL)
773 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
778 lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg);
780 lwkt_replymsg(&rmsg->base.lmsg, 0);
784 * Detach an interface, removing it from the
785 * list of "active" interfaces.
788 if_detach(struct ifnet *ifp)
790 struct netmsg_if_rtdel msg;
794 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
797 * Remove routes and flush queues.
801 if (ifp->if_flags & IFF_NPOLLING)
802 ifpoll_deregister(ifp);
807 if (ifq_is_enabled(&ifp->if_snd))
808 altq_disable(&ifp->if_snd);
809 if (ifq_is_attached(&ifp->if_snd))
810 altq_detach(&ifp->if_snd);
814 * Clean up all addresses.
816 ifp->if_lladdr = NULL;
818 if_purgeaddrs_nolink(ifp);
819 if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
822 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
823 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
824 ("non-link ifaddr is left on if_addrheads"));
826 ifa_ifunlink(ifa, ifp);
828 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
829 ("there are still ifaddrs left on if_addrheads"));
834 * Remove all IPv4 kernel structures related to ifp.
841 * Remove all IPv6 kernel structs related to ifp. This should be done
842 * before removing routing entries below, since IPv6 interface direct
843 * routes are expected to be removed by the IPv6-specific kernel API.
844 * Otherwise, the kernel will detect some inconsistency and bark it.
850 * Delete all remaining routes using this interface
852 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
855 KASSERT(&curthread->td_msgport != rtable_portfn(0),
856 ("if_detach in rtable thread"));
857 lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0);
859 /* Announce that the interface is gone. */
860 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
861 devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
863 SLIST_FOREACH(dp, &domains, dom_next)
864 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
865 (*dp->dom_ifdetach)(ifp,
866 ifp->if_afdata[dp->dom_family]);
869 * Remove interface from ifindex2ifp[] and maybe decrement if_index.
871 ifindex2ifnet[ifp->if_index] = NULL;
872 while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
875 TAILQ_REMOVE(&ifnet, ifp, if_link);
876 kfree(ifp->if_addrheads, M_IFADDR);
878 lwkt_synchronize_ipiqs("if_detach");
879 ifq_stage_detach(&ifp->if_snd);
881 for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
882 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
884 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
885 kfree(ifsq->ifsq_stage, M_DEVBUF);
887 kfree(ifp->if_snd.altq_subq, M_DEVBUF);
889 kfree(ifp->if_data_pcpu, M_DEVBUF);
895 * Create interface group without members
898 if_creategroup(const char *groupname)
900 struct ifg_group *ifg = NULL;
902 if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
903 M_TEMP, M_NOWAIT)) == NULL)
906 strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
908 ifg->ifg_carp_demoted = 0;
909 TAILQ_INIT(&ifg->ifg_members);
911 pfi_attach_ifgroup(ifg);
913 TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
919 * Add a group to an interface
922 if_addgroup(struct ifnet *ifp, const char *groupname)
924 struct ifg_list *ifgl;
925 struct ifg_group *ifg = NULL;
926 struct ifg_member *ifgm;
928 if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
929 groupname[strlen(groupname) - 1] <= '9')
932 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
933 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
936 if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
939 if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
944 TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
945 if (!strcmp(ifg->ifg_group, groupname))
948 if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
955 ifgl->ifgl_group = ifg;
956 ifgm->ifgm_ifp = ifp;
958 TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
959 TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
962 pfi_group_change(groupname);
969 * Remove a group from an interface
972 if_delgroup(struct ifnet *ifp, const char *groupname)
974 struct ifg_list *ifgl;
975 struct ifg_member *ifgm;
977 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
978 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
983 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
985 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
986 if (ifgm->ifgm_ifp == ifp)
990 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
994 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
995 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
997 pfi_detach_ifgroup(ifgl->ifgl_group);
999 kfree(ifgl->ifgl_group, M_TEMP);
1002 kfree(ifgl, M_TEMP);
1005 pfi_group_change(groupname);
1012 * Stores all groups from an interface in memory pointed
1016 if_getgroup(caddr_t data, struct ifnet *ifp)
1019 struct ifg_list *ifgl;
1020 struct ifg_req ifgrq, *ifgp;
1021 struct ifgroupreq *ifgr = (struct ifgroupreq *)data;
1023 if (ifgr->ifgr_len == 0) {
1024 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1025 ifgr->ifgr_len += sizeof(struct ifg_req);
1029 len = ifgr->ifgr_len;
1030 ifgp = ifgr->ifgr_groups;
1031 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1032 if (len < sizeof(ifgrq))
1034 bzero(&ifgrq, sizeof ifgrq);
1035 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1036 sizeof(ifgrq.ifgrq_group));
1037 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1038 sizeof(struct ifg_req))))
1040 len -= sizeof(ifgrq);
1048 * Stores all members of a group in memory pointed to by data
1051 if_getgroupmembers(caddr_t data)
1053 struct ifgroupreq *ifgr = (struct ifgroupreq *)data;
1054 struct ifg_group *ifg;
1055 struct ifg_member *ifgm;
1056 struct ifg_req ifgrq, *ifgp;
1059 TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1060 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1065 if (ifgr->ifgr_len == 0) {
1066 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1067 ifgr->ifgr_len += sizeof(ifgrq);
1071 len = ifgr->ifgr_len;
1072 ifgp = ifgr->ifgr_groups;
1073 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1074 if (len < sizeof(ifgrq))
1076 bzero(&ifgrq, sizeof ifgrq);
1077 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1078 sizeof(ifgrq.ifgrq_member));
1079 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1080 sizeof(struct ifg_req))))
1082 len -= sizeof(ifgrq);
1090 * Delete Routes for a Network Interface
1092 * Called for each routing entry via the rnh->rnh_walktree() call above
1093 * to delete all route entries referencing a detaching network interface.
1096 * rn pointer to node in the routing table
1097 * arg argument passed to rnh->rnh_walktree() - detaching interface
1101 * errno failed - reason indicated
1105 if_rtdel(struct radix_node *rn, void *arg)
1107 struct rtentry *rt = (struct rtentry *)rn;
1108 struct ifnet *ifp = arg;
1111 if (rt->rt_ifp == ifp) {
1114 * Protect (sorta) against walktree recursion problems
1115 * with cloned routes
1117 if (!(rt->rt_flags & RTF_UP))
1120 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1121 rt_mask(rt), rt->rt_flags,
1124 log(LOG_WARNING, "if_rtdel: error %d\n", err);
1132 * Locate an interface based on a complete address.
1135 ifa_ifwithaddr(struct sockaddr *addr)
1139 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1140 struct ifaddr_container *ifac;
1142 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1143 struct ifaddr *ifa = ifac->ifa;
1145 if (ifa->ifa_addr->sa_family != addr->sa_family)
1147 if (sa_equal(addr, ifa->ifa_addr))
1149 if ((ifp->if_flags & IFF_BROADCAST) &&
1150 ifa->ifa_broadaddr &&
1151 /* IPv6 doesn't have broadcast */
1152 ifa->ifa_broadaddr->sa_len != 0 &&
1153 sa_equal(ifa->ifa_broadaddr, addr))
1160 * Locate the point to point interface with a given destination address.
1163 ifa_ifwithdstaddr(struct sockaddr *addr)
1167 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1168 struct ifaddr_container *ifac;
1170 if (!(ifp->if_flags & IFF_POINTOPOINT))
1173 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1174 struct ifaddr *ifa = ifac->ifa;
1176 if (ifa->ifa_addr->sa_family != addr->sa_family)
1178 if (ifa->ifa_dstaddr &&
1179 sa_equal(addr, ifa->ifa_dstaddr))
1187 * Find an interface on a specific network. If many, choice
1188 * is most specific found.
1191 ifa_ifwithnet(struct sockaddr *addr)
1194 struct ifaddr *ifa_maybe = NULL;
1195 u_int af = addr->sa_family;
1196 char *addr_data = addr->sa_data, *cplim;
1199 * AF_LINK addresses can be looked up directly by their index number,
1200 * so do that if we can.
1202 if (af == AF_LINK) {
1203 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1205 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1206 return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1210 * Scan though each interface, looking for ones that have
1211 * addresses in this address family.
1213 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1214 struct ifaddr_container *ifac;
1216 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1217 struct ifaddr *ifa = ifac->ifa;
1218 char *cp, *cp2, *cp3;
1220 if (ifa->ifa_addr->sa_family != af)
1222 if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1224 * This is a bit broken as it doesn't
1225 * take into account that the remote end may
1226 * be a single node in the network we are
1228 * The trouble is that we don't know the
1229 * netmask for the remote end.
1231 if (ifa->ifa_dstaddr != NULL &&
1232 sa_equal(addr, ifa->ifa_dstaddr))
1236 * if we have a special address handler,
1237 * then use it instead of the generic one.
1239 if (ifa->ifa_claim_addr) {
1240 if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1248 * Scan all the bits in the ifa's address.
1249 * If a bit dissagrees with what we are
1250 * looking for, mask it with the netmask
1251 * to see if it really matters.
1252 * (A byte at a time)
1254 if (ifa->ifa_netmask == 0)
1257 cp2 = ifa->ifa_addr->sa_data;
1258 cp3 = ifa->ifa_netmask->sa_data;
1259 cplim = ifa->ifa_netmask->sa_len +
1260 (char *)ifa->ifa_netmask;
1262 if ((*cp++ ^ *cp2++) & *cp3++)
1263 goto next; /* next address! */
1265 * If the netmask of what we just found
1266 * is more specific than what we had before
1267 * (if we had one) then remember the new one
1268 * before continuing to search
1269 * for an even better one.
1271 if (ifa_maybe == NULL ||
1272 rn_refines((char *)ifa->ifa_netmask,
1273 (char *)ifa_maybe->ifa_netmask))
1282 * Find an interface address specific to an interface best matching
1286 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1288 struct ifaddr_container *ifac;
1289 char *cp, *cp2, *cp3;
1291 struct ifaddr *ifa_maybe = NULL;
1292 u_int af = addr->sa_family;
1296 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1297 struct ifaddr *ifa = ifac->ifa;
1299 if (ifa->ifa_addr->sa_family != af)
1301 if (ifa_maybe == NULL)
1303 if (ifa->ifa_netmask == NULL) {
1304 if (sa_equal(addr, ifa->ifa_addr) ||
1305 (ifa->ifa_dstaddr != NULL &&
1306 sa_equal(addr, ifa->ifa_dstaddr)))
1310 if (ifp->if_flags & IFF_POINTOPOINT) {
1311 if (sa_equal(addr, ifa->ifa_dstaddr))
1315 cp2 = ifa->ifa_addr->sa_data;
1316 cp3 = ifa->ifa_netmask->sa_data;
1317 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1318 for (; cp3 < cplim; cp3++)
1319 if ((*cp++ ^ *cp2++) & *cp3)
1329 * Default action when installing a route with a Link Level gateway.
1330 * Lookup an appropriate real ifa to point to.
1331 * This should be moved to /sys/net/link.c eventually.
1334 link_rtrequest(int cmd, struct rtentry *rt)
1337 struct sockaddr *dst;
1340 if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1341 (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1343 ifa = ifaof_ifpforaddr(dst, ifp);
1345 IFAFREE(rt->rt_ifa);
1348 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1349 ifa->ifa_rtrequest(cmd, rt);
1354 * Mark an interface down and notify protocols of
1356 * NOTE: must be called at splnet or eqivalent.
1359 if_unroute(struct ifnet *ifp, int flag, int fam)
1361 struct ifaddr_container *ifac;
1363 ifp->if_flags &= ~flag;
1364 getmicrotime(&ifp->if_lastchange);
1365 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1366 struct ifaddr *ifa = ifac->ifa;
1368 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1369 kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1371 ifq_purge_all(&ifp->if_snd);
1376 * Mark an interface up and notify protocols of
1378 * NOTE: must be called at splnet or eqivalent.
1381 if_route(struct ifnet *ifp, int flag, int fam)
1383 struct ifaddr_container *ifac;
1385 ifq_purge_all(&ifp->if_snd);
1386 ifp->if_flags |= flag;
1387 getmicrotime(&ifp->if_lastchange);
1388 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1389 struct ifaddr *ifa = ifac->ifa;
1391 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1392 kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1401 * Mark an interface down and notify protocols of the transition. An
1402 * interface going down is also considered to be a synchronizing event.
1403 * We must ensure that all packet processing related to the interface
1404 * has completed before we return so e.g. the caller can free the ifnet
1405 * structure that the mbufs may be referencing.
1407 * NOTE: must be called at splnet or eqivalent.
1410 if_down(struct ifnet *ifp)
1412 if_unroute(ifp, IFF_UP, AF_UNSPEC);
1413 netmsg_service_sync();
1417 * Mark an interface up and notify protocols of
1419 * NOTE: must be called at splnet or eqivalent.
1422 if_up(struct ifnet *ifp)
1424 if_route(ifp, IFF_UP, AF_UNSPEC);
1428 * Process a link state change.
1429 * NOTE: must be called at splsoftnet or equivalent.
1432 if_link_state_change(struct ifnet *ifp)
1434 int link_state = ifp->if_link_state;
1437 devctl_notify("IFNET", ifp->if_xname,
1438 (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1442 * Handle interface watchdog timer routines. Called
1443 * from softclock, we decrement timers (if set) and
1444 * call the appropriate interface routine on expiration.
1447 if_slowtimo(void *arg)
1453 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1454 if (if_stats_compat) {
1455 IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1456 IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1457 IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1458 IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1459 IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1460 IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1461 IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1462 IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1463 IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1464 IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1465 IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1468 if (ifp->if_timer == 0 || --ifp->if_timer)
1470 if (ifp->if_watchdog) {
1471 if (ifnet_tryserialize_all(ifp)) {
1472 (*ifp->if_watchdog)(ifp);
1473 ifnet_deserialize_all(ifp);
1475 /* try again next timeout */
1483 callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1487 * Map interface name to
1488 * interface structure pointer.
1491 ifunit(const char *name)
1496 * Search all the interfaces for this name/number
1499 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1500 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1508 * Map interface name in a sockaddr_dl to
1509 * interface structure pointer.
1512 if_withname(struct sockaddr *sa)
1514 char ifname[IFNAMSIZ+1];
1515 struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1517 if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1518 (sdl->sdl_nlen > IFNAMSIZ) )
1522 * ifunit wants a null-terminated name. It may not be null-terminated
1523 * in the sockaddr. We don't want to change the caller's sockaddr,
1524 * and there might not be room to put the trailing null anyway, so we
1525 * make a local copy that we know we can null terminate safely.
1528 bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1529 ifname[sdl->sdl_nlen] = '\0';
1530 return ifunit(ifname);
1538 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1549 size_t namelen, onamelen;
1550 char new_name[IFNAMSIZ];
1552 struct sockaddr_dl *sdl;
1557 return (ifconf(cmd, data, cred));
1562 ifr = (struct ifreq *)data;
1567 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1569 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1570 cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1572 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1574 return (if_clone_destroy(ifr->ifr_name));
1575 case SIOCIFGCLONERS:
1576 return (if_clone_list((struct if_clonereq *)data));
1582 * Nominal ioctl through interface, lookup the ifp and obtain a
1583 * lock to serialize the ifconfig ioctl operation.
1585 ifp = ifunit(ifr->ifr_name);
1589 mtx_lock(&ifp->if_ioctl_mtx);
1593 ifr->ifr_index = ifp->if_index;
1597 ifr->ifr_flags = ifp->if_flags;
1598 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1602 ifr->ifr_reqcap = ifp->if_capabilities;
1603 ifr->ifr_curcap = ifp->if_capenable;
1607 ifr->ifr_metric = ifp->if_metric;
1611 ifr->ifr_mtu = ifp->if_mtu;
1615 ifr->ifr_tsolen = ifp->if_tsolen;
1619 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1620 sizeof(ifp->if_data));
1624 ifr->ifr_phys = ifp->if_physical;
1627 case SIOCGIFPOLLCPU:
1628 ifr->ifr_pollcpu = -1;
1631 case SIOCSIFPOLLCPU:
1635 error = priv_check_cred(cred, PRIV_ROOT, 0);
1638 new_flags = (ifr->ifr_flags & 0xffff) |
1639 (ifr->ifr_flagshigh << 16);
1640 if (ifp->if_flags & IFF_SMART) {
1641 /* Smart drivers twiddle their own routes */
1642 } else if (ifp->if_flags & IFF_UP &&
1643 (new_flags & IFF_UP) == 0) {
1647 } else if (new_flags & IFF_UP &&
1648 (ifp->if_flags & IFF_UP) == 0) {
1654 #ifdef IFPOLL_ENABLE
1655 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1656 if (new_flags & IFF_NPOLLING)
1657 ifpoll_register(ifp);
1659 ifpoll_deregister(ifp);
1663 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1664 (new_flags &~ IFF_CANTCHANGE);
1665 if (new_flags & IFF_PPROMISC) {
1666 /* Permanently promiscuous mode requested */
1667 ifp->if_flags |= IFF_PROMISC;
1668 } else if (ifp->if_pcount == 0) {
1669 ifp->if_flags &= ~IFF_PROMISC;
1671 if (ifp->if_ioctl) {
1672 ifnet_serialize_all(ifp);
1673 ifp->if_ioctl(ifp, cmd, data, cred);
1674 ifnet_deserialize_all(ifp);
1676 getmicrotime(&ifp->if_lastchange);
1680 error = priv_check_cred(cred, PRIV_ROOT, 0);
1683 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1687 ifnet_serialize_all(ifp);
1688 ifp->if_ioctl(ifp, cmd, data, cred);
1689 ifnet_deserialize_all(ifp);
1693 error = priv_check_cred(cred, PRIV_ROOT, 0);
1696 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1699 if (new_name[0] == '\0') {
1703 if (ifunit(new_name) != NULL) {
1708 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1710 /* Announce the departure of the interface. */
1711 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1713 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1714 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1715 /* XXX IFA_LOCK(ifa); */
1716 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1717 namelen = strlen(new_name);
1718 onamelen = sdl->sdl_nlen;
1720 * Move the address if needed. This is safe because we
1721 * allocate space for a name of length IFNAMSIZ when we
1722 * create this in if_attach().
1724 if (namelen != onamelen) {
1725 bcopy(sdl->sdl_data + onamelen,
1726 sdl->sdl_data + namelen, sdl->sdl_alen);
1728 bcopy(new_name, sdl->sdl_data, namelen);
1729 sdl->sdl_nlen = namelen;
1730 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1731 bzero(sdl->sdl_data, onamelen);
1732 while (namelen != 0)
1733 sdl->sdl_data[--namelen] = 0xff;
1734 /* XXX IFA_UNLOCK(ifa) */
1736 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1738 /* Announce the return of the interface. */
1739 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1743 error = priv_check_cred(cred, PRIV_ROOT, 0);
1746 ifp->if_metric = ifr->ifr_metric;
1747 getmicrotime(&ifp->if_lastchange);
1751 error = priv_check_cred(cred, PRIV_ROOT, 0);
1754 if (ifp->if_ioctl == NULL) {
1758 ifnet_serialize_all(ifp);
1759 error = ifp->if_ioctl(ifp, cmd, data, cred);
1760 ifnet_deserialize_all(ifp);
1762 getmicrotime(&ifp->if_lastchange);
1767 u_long oldmtu = ifp->if_mtu;
1769 error = priv_check_cred(cred, PRIV_ROOT, 0);
1772 if (ifp->if_ioctl == NULL) {
1776 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
1780 ifnet_serialize_all(ifp);
1781 error = ifp->if_ioctl(ifp, cmd, data, cred);
1782 ifnet_deserialize_all(ifp);
1784 getmicrotime(&ifp->if_lastchange);
1788 * If the link MTU changed, do network layer specific procedure.
1790 if (ifp->if_mtu != oldmtu) {
1799 error = priv_check_cred(cred, PRIV_ROOT, 0);
1803 /* XXX need driver supplied upper limit */
1804 if (ifr->ifr_tsolen <= 0) {
1808 ifp->if_tsolen = ifr->ifr_tsolen;
1813 error = priv_check_cred(cred, PRIV_ROOT, 0);
1817 /* Don't allow group membership on non-multicast interfaces. */
1818 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1823 /* Don't let users screw up protocols' entries. */
1824 if (ifr->ifr_addr.sa_family != AF_LINK) {
1829 if (cmd == SIOCADDMULTI) {
1830 struct ifmultiaddr *ifma;
1831 error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1833 error = if_delmulti(ifp, &ifr->ifr_addr);
1836 getmicrotime(&ifp->if_lastchange);
1839 case SIOCSIFPHYADDR:
1840 case SIOCDIFPHYADDR:
1842 case SIOCSIFPHYADDR_IN6:
1844 case SIOCSLIFPHYADDR:
1846 case SIOCSIFGENERIC:
1847 error = priv_check_cred(cred, PRIV_ROOT, 0);
1850 if (ifp->if_ioctl == 0) {
1854 ifnet_serialize_all(ifp);
1855 error = ifp->if_ioctl(ifp, cmd, data, cred);
1856 ifnet_deserialize_all(ifp);
1858 getmicrotime(&ifp->if_lastchange);
1862 ifs = (struct ifstat *)data;
1863 ifs->ascii[0] = '\0';
1865 case SIOCGIFPSRCADDR:
1866 case SIOCGIFPDSTADDR:
1867 case SIOCGLIFPHYADDR:
1869 case SIOCGIFGENERIC:
1870 if (ifp->if_ioctl == NULL) {
1874 ifnet_serialize_all(ifp);
1875 error = ifp->if_ioctl(ifp, cmd, data, cred);
1876 ifnet_deserialize_all(ifp);
1880 error = priv_check_cred(cred, PRIV_ROOT, 0);
1883 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
1884 ifr->ifr_addr.sa_len);
1885 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1889 oif_flags = ifp->if_flags;
1890 if (so->so_proto == 0) {
1895 error = so_pru_control_direct(so, cmd, data, ifp);
1900 case SIOCSIFDSTADDR:
1902 case SIOCSIFBRDADDR:
1903 case SIOCSIFNETMASK:
1904 #if BYTE_ORDER != BIG_ENDIAN
1905 if (ifr->ifr_addr.sa_family == 0 &&
1906 ifr->ifr_addr.sa_len < 16) {
1907 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1908 ifr->ifr_addr.sa_len = 16;
1911 if (ifr->ifr_addr.sa_len == 0)
1912 ifr->ifr_addr.sa_len = 16;
1918 case OSIOCGIFDSTADDR:
1919 cmd = SIOCGIFDSTADDR;
1921 case OSIOCGIFBRDADDR:
1922 cmd = SIOCGIFBRDADDR;
1924 case OSIOCGIFNETMASK:
1925 cmd = SIOCGIFNETMASK;
1931 error = so_pru_control_direct(so, cmd, data, ifp);
1935 case OSIOCGIFDSTADDR:
1936 case OSIOCGIFBRDADDR:
1937 case OSIOCGIFNETMASK:
1938 *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1941 #endif /* COMPAT_43 */
1943 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1945 DELAY(100);/* XXX: temporary workaround for fxp issue*/
1946 if (ifp->if_flags & IFF_UP) {
1956 mtx_unlock(&ifp->if_ioctl_mtx);
1961 * Set/clear promiscuous mode on interface ifp based on the truth value
1962 * of pswitch. The calls are reference counted so that only the first
1963 * "on" request actually has an effect, as does the final "off" request.
1964 * Results are undefined if the "off" and "on" requests are not matched.
1967 ifpromisc(struct ifnet *ifp, int pswitch)
1973 oldflags = ifp->if_flags;
1974 if (ifp->if_flags & IFF_PPROMISC) {
1975 /* Do nothing if device is in permanently promiscuous mode */
1976 ifp->if_pcount += pswitch ? 1 : -1;
1981 * If the device is not configured up, we cannot put it in
1984 if ((ifp->if_flags & IFF_UP) == 0)
1986 if (ifp->if_pcount++ != 0)
1988 ifp->if_flags |= IFF_PROMISC;
1989 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1992 if (--ifp->if_pcount > 0)
1994 ifp->if_flags &= ~IFF_PROMISC;
1995 log(LOG_INFO, "%s: promiscuous mode disabled\n",
1998 ifr.ifr_flags = ifp->if_flags;
1999 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2000 ifnet_serialize_all(ifp);
2001 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2002 ifnet_deserialize_all(ifp);
2006 ifp->if_flags = oldflags;
2011 * Return interface configuration
2012 * of system. List may be used
2013 * in later ioctl's (above) to get
2014 * other information.
2017 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2019 struct ifconf *ifc = (struct ifconf *)data;
2021 struct sockaddr *sa;
2022 struct ifreq ifr, *ifrp;
2023 int space = ifc->ifc_len, error = 0;
2025 ifrp = ifc->ifc_req;
2026 TAILQ_FOREACH(ifp, &ifnet, if_link) {
2027 struct ifaddr_container *ifac;
2030 if (space <= sizeof ifr)
2034 * Zero the stack declared structure first to prevent
2035 * memory disclosure.
2037 bzero(&ifr, sizeof(ifr));
2038 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2039 >= sizeof(ifr.ifr_name)) {
2040 error = ENAMETOOLONG;
2045 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2046 struct ifaddr *ifa = ifac->ifa;
2048 if (space <= sizeof ifr)
2051 if (cred->cr_prison &&
2052 prison_if(cred, sa))
2056 if (cmd == OSIOCGIFCONF) {
2057 struct osockaddr *osa =
2058 (struct osockaddr *)&ifr.ifr_addr;
2060 osa->sa_family = sa->sa_family;
2061 error = copyout(&ifr, ifrp, sizeof ifr);
2065 if (sa->sa_len <= sizeof(*sa)) {
2067 error = copyout(&ifr, ifrp, sizeof ifr);
2070 if (space < (sizeof ifr) + sa->sa_len -
2073 space -= sa->sa_len - sizeof(*sa);
2074 error = copyout(&ifr, ifrp,
2075 sizeof ifr.ifr_name);
2077 error = copyout(sa, &ifrp->ifr_addr,
2079 ifrp = (struct ifreq *)
2080 (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2084 space -= sizeof ifr;
2089 bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2090 error = copyout(&ifr, ifrp, sizeof ifr);
2093 space -= sizeof ifr;
2097 ifc->ifc_len -= space;
2102 * Just like if_promisc(), but for all-multicast-reception mode.
2105 if_allmulti(struct ifnet *ifp, int onswitch)
2113 if (ifp->if_amcount++ == 0) {
2114 ifp->if_flags |= IFF_ALLMULTI;
2115 ifr.ifr_flags = ifp->if_flags;
2116 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2117 ifnet_serialize_all(ifp);
2118 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2120 ifnet_deserialize_all(ifp);
2123 if (ifp->if_amcount > 1) {
2126 ifp->if_amcount = 0;
2127 ifp->if_flags &= ~IFF_ALLMULTI;
2128 ifr.ifr_flags = ifp->if_flags;
2129 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2130 ifnet_serialize_all(ifp);
2131 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2133 ifnet_deserialize_all(ifp);
2145 * Add a multicast listenership to the interface in question.
2146 * The link layer provides a routine which converts
2150 struct ifnet *ifp, /* interface to manipulate */
2151 struct sockaddr *sa, /* address to add */
2152 struct ifmultiaddr **retifma)
2154 struct sockaddr *llsa, *dupsa;
2156 struct ifmultiaddr *ifma;
2159 * If the matching multicast address already exists
2160 * then don't add a new one, just add a reference
2162 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2163 if (sa_equal(sa, ifma->ifma_addr)) {
2164 ifma->ifma_refcount++;
2172 * Give the link layer a chance to accept/reject it, and also
2173 * find out which AF_LINK address this maps to, if it isn't one
2176 if (ifp->if_resolvemulti) {
2177 ifnet_serialize_all(ifp);
2178 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2179 ifnet_deserialize_all(ifp);
2186 ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2187 dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_WAITOK);
2188 bcopy(sa, dupsa, sa->sa_len);
2190 ifma->ifma_addr = dupsa;
2191 ifma->ifma_lladdr = llsa;
2192 ifma->ifma_ifp = ifp;
2193 ifma->ifma_refcount = 1;
2194 ifma->ifma_protospec = 0;
2195 rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2198 * Some network interfaces can scan the address list at
2199 * interrupt time; lock them out.
2202 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2208 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2209 if (sa_equal(ifma->ifma_addr, llsa))
2213 ifma->ifma_refcount++;
2215 ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2216 dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_WAITOK);
2217 bcopy(llsa, dupsa, llsa->sa_len);
2218 ifma->ifma_addr = dupsa;
2219 ifma->ifma_ifp = ifp;
2220 ifma->ifma_refcount = 1;
2222 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2227 * We are certain we have added something, so call down to the
2228 * interface to let them know about it.
2231 ifnet_serialize_all(ifp);
2233 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2234 ifnet_deserialize_all(ifp);
2241 * Remove a reference to a multicast address on this interface. Yell
2242 * if the request does not match an existing membership.
2245 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2247 struct ifmultiaddr *ifma;
2249 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2250 if (sa_equal(sa, ifma->ifma_addr))
2255 if (ifma->ifma_refcount > 1) {
2256 ifma->ifma_refcount--;
2260 rt_newmaddrmsg(RTM_DELMADDR, ifma);
2261 sa = ifma->ifma_lladdr;
2263 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2265 * Make sure the interface driver is notified
2266 * in the case of a link layer mcast group being left.
2268 if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL) {
2269 ifnet_serialize_all(ifp);
2270 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2271 ifnet_deserialize_all(ifp);
2274 kfree(ifma->ifma_addr, M_IFMADDR);
2275 kfree(ifma, M_IFMADDR);
2280 * Now look for the link-layer address which corresponds to
2281 * this network address. It had been squirreled away in
2282 * ifma->ifma_lladdr for this purpose (so we don't have
2283 * to call ifp->if_resolvemulti() again), and we saved that
2284 * value in sa above. If some nasty deleted the
2285 * link-layer address out from underneath us, we can deal because
2286 * the address we stored was is not the same as the one which was
2287 * in the record for the link-layer address. (So we don't complain
2290 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2291 if (sa_equal(sa, ifma->ifma_addr))
2296 if (ifma->ifma_refcount > 1) {
2297 ifma->ifma_refcount--;
2302 ifnet_serialize_all(ifp);
2303 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2304 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2305 ifnet_deserialize_all(ifp);
2307 kfree(ifma->ifma_addr, M_IFMADDR);
2308 kfree(sa, M_IFMADDR);
2309 kfree(ifma, M_IFMADDR);
2315 * Delete all multicast group membership for an interface.
2316 * Should be used to quickly flush all multicast filters.
2319 if_delallmulti(struct ifnet *ifp)
2321 struct ifmultiaddr *ifma;
2322 struct ifmultiaddr *next;
2324 TAILQ_FOREACH_MUTABLE(ifma, &ifp->if_multiaddrs, ifma_link, next)
2325 if_delmulti(ifp, ifma->ifma_addr);
2330 * Set the link layer address on an interface.
2332 * At this time we only support certain types of interfaces,
2333 * and we don't allow the length of the address to change.
2336 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2338 struct sockaddr_dl *sdl;
2341 sdl = IF_LLSOCKADDR(ifp);
2344 if (len != sdl->sdl_alen) /* don't allow length to change */
2346 switch (ifp->if_type) {
2347 case IFT_ETHER: /* these types use struct arpcom */
2350 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2351 bcopy(lladdr, LLADDR(sdl), len);
2357 * If the interface is already up, we need
2358 * to re-init it in order to reprogram its
2361 ifnet_serialize_all(ifp);
2362 if ((ifp->if_flags & IFF_UP) != 0) {
2364 struct ifaddr_container *ifac;
2367 ifp->if_flags &= ~IFF_UP;
2368 ifr.ifr_flags = ifp->if_flags;
2369 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2370 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2372 ifp->if_flags |= IFF_UP;
2373 ifr.ifr_flags = ifp->if_flags;
2374 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2375 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2379 * Also send gratuitous ARPs to notify other nodes about
2380 * the address change.
2382 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2383 struct ifaddr *ifa = ifac->ifa;
2385 if (ifa->ifa_addr != NULL &&
2386 ifa->ifa_addr->sa_family == AF_INET)
2387 arp_gratuitous(ifp, ifa);
2391 ifnet_deserialize_all(ifp);
2395 struct ifmultiaddr *
2396 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2398 struct ifmultiaddr *ifma;
2400 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2401 if (sa_equal(ifma->ifma_addr, sa))
2408 * This function locates the first real ethernet MAC from a network
2409 * card and loads it into node, returning 0 on success or ENOENT if
2410 * no suitable interfaces were found. It is used by the uuid code to
2411 * generate a unique 6-byte number.
2414 if_getanyethermac(uint16_t *node, int minlen)
2417 struct sockaddr_dl *sdl;
2419 TAILQ_FOREACH(ifp, &ifnet, if_link) {
2420 if (ifp->if_type != IFT_ETHER)
2422 sdl = IF_LLSOCKADDR(ifp);
2423 if (sdl->sdl_alen < minlen)
2425 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2433 * The name argument must be a pointer to storage which will last as
2434 * long as the interface does. For physical devices, the result of
2435 * device_get_name(dev) is a good choice and for pseudo-devices a
2436 * static string works well.
2439 if_initname(struct ifnet *ifp, const char *name, int unit)
2441 ifp->if_dname = name;
2442 ifp->if_dunit = unit;
2443 if (unit != IF_DUNIT_NONE)
2444 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2446 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2450 if_printf(struct ifnet *ifp, const char *fmt, ...)
2455 retval = kprintf("%s: ", ifp->if_xname);
2456 __va_start(ap, fmt);
2457 retval += kvprintf(fmt, ap);
2463 if_alloc(uint8_t type)
2469 * XXX temporary hack until arpcom is setup in if_l2com
2471 if (type == IFT_ETHER)
2472 size = sizeof(struct arpcom);
2474 size = sizeof(struct ifnet);
2476 ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2478 ifp->if_type = type;
2480 if (if_com_alloc[type] != NULL) {
2481 ifp->if_l2com = if_com_alloc[type](type, ifp);
2482 if (ifp->if_l2com == NULL) {
2483 kfree(ifp, M_IFNET);
2491 if_free(struct ifnet *ifp)
2493 kfree(ifp, M_IFNET);
2497 ifq_set_classic(struct ifaltq *ifq)
2499 ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2500 ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2504 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2505 ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2509 KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2510 KASSERT(enqueue != NULL, ("enqueue is not specified"));
2511 KASSERT(dequeue != NULL, ("dequeue is not specified"));
2512 KASSERT(request != NULL, ("request is not specified"));
2514 ifq->altq_mapsubq = mapsubq;
2515 for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2516 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2518 ifsq->ifsq_enqueue = enqueue;
2519 ifsq->ifsq_dequeue = dequeue;
2520 ifsq->ifsq_request = request;
2525 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2526 struct altq_pktattr *pa __unused)
2529 if (ifsq->ifsq_len >= ifsq->ifsq_maxlen) {
2533 m->m_nextpkt = NULL;
2534 if (ifsq->ifsq_tail == NULL)
2535 ifsq->ifsq_head = m;
2537 ifsq->ifsq_tail->m_nextpkt = m;
2538 ifsq->ifsq_tail = m;
2545 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2551 m = ifsq->ifsq_head;
2555 m = ifsq->ifsq_head;
2557 if ((ifsq->ifsq_head = m->m_nextpkt) == NULL)
2558 ifsq->ifsq_tail = NULL;
2559 m->m_nextpkt = NULL;
2565 panic("unsupported ALTQ dequeue op: %d", op);
2571 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2578 m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2586 panic("unsupported ALTQ request: %d", req);
2592 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2594 struct ifnet *ifp = ifsq_get_ifp(ifsq);
2595 int running = 0, need_sched;
2598 * Try to do direct ifnet.if_start on the subqueue first, if there is
2599 * contention on the subqueue hardware serializer, ifnet.if_start on
2600 * the subqueue will be scheduled on the subqueue owner CPU.
2602 if (!ifsq_tryserialize_hw(ifsq)) {
2604 * Subqueue hardware serializer contention happened,
2605 * ifnet.if_start on the subqueue is scheduled on
2606 * the subqueue owner CPU, and we keep going.
2608 ifsq_ifstart_schedule(ifsq, 1);
2612 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2613 ifp->if_start(ifp, ifsq);
2614 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2617 need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2619 ifsq_deserialize_hw(ifsq);
2623 * More data need to be transmitted, ifnet.if_start on the
2624 * subqueue is scheduled on the subqueue owner CPU, and we
2626 * NOTE: ifnet.if_start subqueue interlock is not released.
2628 ifsq_ifstart_schedule(ifsq, force_sched);
2633 * Subqeue packets staging mechanism:
2635 * The packets enqueued into the subqueue are staged to a certain amount
2636 * before the ifnet.if_start on the subqueue is called. In this way, the
2637 * driver could avoid writing to hardware registers upon every packet,
2638 * instead, hardware registers could be written when certain amount of
2639 * packets are put onto hardware TX ring. The measurement on several modern
2640 * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
2641 * registers writing aggregation could save ~20% CPU time when 18bytes UDP
2642 * datagrams are transmitted at 1.48Mpps. The performance improvement by
2643 * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
2644 * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
2646 * Subqueue packets staging is performed for two entry points into drivers'
2647 * transmission function:
2648 * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
2649 * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
2651 * Subqueue packets staging will be stopped upon any of the following
2653 * - If the count of packets enqueued on the current CPU is great than or
2654 * equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2655 * - If the total length of packets enqueued on the current CPU is great
2656 * than or equal to the hardware's MTU - max_protohdr. max_protohdr is
2657 * cut from the hardware's MTU mainly bacause a full TCP segment's size
2658 * is usually less than hardware's MTU.
2659 * - ifsq_ifstart_schedule() is not pending on the current CPU and
2660 * ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
2662 * - The if_start_rollup(), which is registered as low priority netisr
2663 * rollup function, is called; probably because no more work is pending
2667 * Currently subqueue packet staging is only performed in netisr threads.
2670 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2672 struct ifaltq *ifq = &ifp->if_snd;
2673 struct ifaltq_subque *ifsq;
2674 int error, start = 0, len, mcast = 0, avoid_start = 0;
2675 struct ifsubq_stage_head *head = NULL;
2676 struct ifsubq_stage *stage = NULL;
2678 ifsq = ifq_map_subq(ifq, mycpuid);
2679 ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
2681 len = m->m_pkthdr.len;
2682 if (m->m_flags & M_MCAST)
2685 if (curthread->td_type == TD_TYPE_NETISR) {
2686 head = &ifsubq_stage_heads[mycpuid];
2687 stage = ifsq_get_stage(ifsq, mycpuid);
2690 stage->stg_len += len;
2691 if (stage->stg_cnt < ifsq_stage_cntmax &&
2692 stage->stg_len < (ifp->if_mtu - max_protohdr))
2697 error = ifsq_enqueue_locked(ifsq, m, pa);
2699 if (!ifsq_data_ready(ifsq)) {
2700 ALTQ_SQ_UNLOCK(ifsq);
2705 if (!ifsq_is_started(ifsq)) {
2707 ALTQ_SQ_UNLOCK(ifsq);
2710 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
2711 ifsq_stage_insert(head, stage);
2713 IFNET_STAT_INC(ifp, obytes, len);
2715 IFNET_STAT_INC(ifp, omcasts, 1);
2720 * Hold the subqueue interlock of ifnet.if_start
2722 ifsq_set_started(ifsq);
2725 ALTQ_SQ_UNLOCK(ifsq);
2728 IFNET_STAT_INC(ifp, obytes, len);
2730 IFNET_STAT_INC(ifp, omcasts, 1);
2733 if (stage != NULL) {
2734 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
2735 KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
2737 ifsq_stage_remove(head, stage);
2738 ifsq_ifstart_schedule(ifsq, 1);
2743 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
2744 ifsq_stage_remove(head, stage);
2754 ifsq_ifstart_try(ifsq, 0);
2759 ifa_create(int size, int flags)
2764 KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
2766 ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2770 ifa->ifa_containers =
2771 kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
2772 M_IFADDR, M_WAITOK | M_ZERO);
2773 ifa->ifa_ncnt = ncpus;
2774 for (i = 0; i < ncpus; ++i) {
2775 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2777 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2779 ifac->ifa_refcnt = 1;
2782 kprintf("alloc ifa %p %d\n", ifa, size);
2788 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2790 struct ifaddr *ifa = ifac->ifa;
2792 KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2793 KKASSERT(ifac->ifa_refcnt == 0);
2794 KASSERT(ifac->ifa_listmask == 0,
2795 ("ifa is still on %#x lists", ifac->ifa_listmask));
2797 ifac->ifa_magic = IFA_CONTAINER_DEAD;
2799 #ifdef IFADDR_DEBUG_VERBOSE
2800 kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2803 KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2804 ("invalid # of ifac, %d", ifa->ifa_ncnt));
2805 if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2807 kprintf("free ifa %p\n", ifa);
2809 kfree(ifa->ifa_containers, M_IFADDR);
2810 kfree(ifa, M_IFADDR);
2815 ifa_iflink_dispatch(netmsg_t nmsg)
2817 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2818 struct ifaddr *ifa = msg->ifa;
2819 struct ifnet *ifp = msg->ifp;
2821 struct ifaddr_container *ifac;
2825 ifac = &ifa->ifa_containers[cpu];
2826 ASSERT_IFAC_VALID(ifac);
2827 KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2828 ("ifaddr is on if_addrheads"));
2830 ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2832 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2834 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2838 ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2842 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2844 struct netmsg_ifaddr msg;
2846 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2847 0, ifa_iflink_dispatch);
2852 ifa_domsg(&msg.base.lmsg, 0);
2856 ifa_ifunlink_dispatch(netmsg_t nmsg)
2858 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2859 struct ifaddr *ifa = msg->ifa;
2860 struct ifnet *ifp = msg->ifp;
2862 struct ifaddr_container *ifac;
2866 ifac = &ifa->ifa_containers[cpu];
2867 ASSERT_IFAC_VALID(ifac);
2868 KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2869 ("ifaddr is not on if_addrhead"));
2871 TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2872 ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2876 ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2880 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2882 struct netmsg_ifaddr msg;
2884 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2885 0, ifa_ifunlink_dispatch);
2889 ifa_domsg(&msg.base.lmsg, 0);
2893 ifa_destroy_dispatch(netmsg_t nmsg)
2895 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2898 ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2902 ifa_destroy(struct ifaddr *ifa)
2904 struct netmsg_ifaddr msg;
2906 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2907 0, ifa_destroy_dispatch);
2910 ifa_domsg(&msg.base.lmsg, 0);
2914 ifnet_portfn(int cpu)
2916 return &ifnet_threads[cpu].td_msgport;
2920 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2922 KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2924 if (next_cpu < ncpus)
2925 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2927 lwkt_replymsg(lmsg, 0);
2931 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
2933 KKASSERT(cpu < ncpus);
2934 return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
2938 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
2940 KKASSERT(cpu < ncpus);
2941 lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
2945 * Generic netmsg service loop. Some protocols may roll their own but all
2946 * must do the basic command dispatch function call done here.
2949 ifnet_service_loop(void *arg __unused)
2953 while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
2954 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
2955 msg->base.nm_dispatch(msg);
2960 if_start_rollup(void)
2962 struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
2963 struct ifsubq_stage *stage;
2965 while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
2966 struct ifaltq_subque *ifsq = stage->stg_subq;
2969 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
2971 ifsq_stage_remove(head, stage);
2974 ifsq_ifstart_schedule(ifsq, 1);
2979 if (!ifsq_is_started(ifsq)) {
2981 * Hold the subqueue interlock of
2984 ifsq_set_started(ifsq);
2987 ALTQ_SQ_UNLOCK(ifsq);
2990 ifsq_ifstart_try(ifsq, 1);
2992 KKASSERT((stage->stg_flags &
2993 (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
2998 ifnetinit(void *dummy __unused)
3002 for (i = 0; i < ncpus; ++i) {
3003 struct thread *thr = &ifnet_threads[i];
3005 lwkt_create(ifnet_service_loop, NULL, NULL,
3006 thr, TDF_NOSTART|TDF_FORCE_SPINPORT,
3008 netmsg_service_port_init(&thr->td_msgport);
3012 for (i = 0; i < ncpus; ++i)
3013 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3014 netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3018 ifnet_byindex(unsigned short idx)
3022 return ifindex2ifnet[idx];
3026 ifaddr_byindex(unsigned short idx)
3030 ifp = ifnet_byindex(idx);
3033 return TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
3037 if_register_com_alloc(u_char type,
3038 if_com_alloc_t *a, if_com_free_t *f)
3041 KASSERT(if_com_alloc[type] == NULL,
3042 ("if_register_com_alloc: %d already registered", type));
3043 KASSERT(if_com_free[type] == NULL,
3044 ("if_register_com_alloc: %d free already registered", type));
3046 if_com_alloc[type] = a;
3047 if_com_free[type] = f;
3051 if_deregister_com_alloc(u_char type)
3054 KASSERT(if_com_alloc[type] != NULL,
3055 ("if_deregister_com_alloc: %d not registered", type));
3056 KASSERT(if_com_free[type] != NULL,
3057 ("if_deregister_com_alloc: %d free not registered", type));
3058 if_com_alloc[type] = NULL;
3059 if_com_free[type] = NULL;
3063 if_ring_count2(int cnt, int cnt_max)
3067 KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3068 ("invalid ring count max %d", cnt_max));
3077 while ((1 << (shift + 1)) <= cnt)
3081 KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3082 ("calculate cnt %d, ncpus2 %d, cnt max %d",
3083 cnt, ncpus2, cnt_max));
3088 ifq_set_maxlen(struct ifaltq *ifq, int len)
3090 ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3094 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3096 return ALTQ_SUBQ_INDEX_DEFAULT;
3100 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3102 return (cpuid & ifq->altq_subq_mask);
3106 ifsq_watchdog(void *arg)
3108 struct ifsubq_watchdog *wd = arg;
3111 if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3114 ifp = ifsq_get_ifp(wd->wd_subq);
3115 if (ifnet_tryserialize_all(ifp)) {
3116 wd->wd_watchdog(wd->wd_subq);
3117 ifnet_deserialize_all(ifp);
3119 /* try again next timeout */
3123 ifsq_watchdog_reset(wd);
3127 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3129 callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3130 ifsq_get_cpuid(wd->wd_subq));
3134 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3135 ifsq_watchdog_t watchdog)
3137 callout_init_mp(&wd->wd_callout);
3140 wd->wd_watchdog = watchdog;
3144 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3147 ifsq_watchdog_reset(wd);
3151 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3154 callout_stop(&wd->wd_callout);