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_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
607 ifsq->ifsq_prepended = NULL;
608 ifsq->ifsq_started = 0;
609 ifsq->ifsq_hw_oactive = 0;
610 ifsq_set_cpuid(ifsq, 0);
611 if (ifp->if_serializer != NULL)
612 ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
615 kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
616 M_DEVBUF, M_WAITOK | M_ZERO);
617 for (i = 0; i < ncpus; ++i)
618 ifsq->ifsq_stage[i].stg_subq = ifsq;
620 ifsq->ifsq_ifstart_nmsg =
621 kmalloc(ncpus * sizeof(struct netmsg_base),
622 M_LWKTMSG, M_WAITOK);
623 for (i = 0; i < ncpus; ++i) {
624 netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
625 &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
626 ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
629 ifq_set_classic(ifq);
631 if (!SLIST_EMPTY(&domains))
632 if_attachdomain1(ifp);
634 /* Announce the interface. */
635 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
637 mtx_unlock(&ifp->if_ioctl_mtx);
641 if_attachdomain(void *dummy)
646 TAILQ_FOREACH(ifp, &ifnet, if_list)
647 if_attachdomain1(ifp);
650 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
651 if_attachdomain, NULL);
654 if_attachdomain1(struct ifnet *ifp)
660 /* address family dependent data region */
661 bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
662 SLIST_FOREACH(dp, &domains, dom_next)
663 if (dp->dom_ifattach)
664 ifp->if_afdata[dp->dom_family] =
665 (*dp->dom_ifattach)(ifp);
670 * Purge all addresses whose type is _not_ AF_LINK
673 if_purgeaddrs_nolink(struct ifnet *ifp)
675 struct ifaddr_container *ifac, *next;
677 TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
679 struct ifaddr *ifa = ifac->ifa;
681 /* Leave link ifaddr as it is */
682 if (ifa->ifa_addr->sa_family == AF_LINK)
685 /* XXX: Ugly!! ad hoc just for INET */
686 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
687 struct ifaliasreq ifr;
688 #ifdef IFADDR_DEBUG_VERBOSE
691 kprintf("purge in4 addr %p: ", ifa);
692 for (i = 0; i < ncpus; ++i)
693 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
697 bzero(&ifr, sizeof ifr);
698 ifr.ifra_addr = *ifa->ifa_addr;
699 if (ifa->ifa_dstaddr)
700 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
701 if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
707 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
708 #ifdef IFADDR_DEBUG_VERBOSE
711 kprintf("purge in6 addr %p: ", ifa);
712 for (i = 0; i < ncpus; ++i)
713 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
718 /* ifp_addrhead is already updated */
722 ifa_ifunlink(ifa, ifp);
728 ifq_stage_detach_handler(netmsg_t nmsg)
730 struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
733 for (q = 0; q < ifq->altq_subq_cnt; ++q) {
734 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
735 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
737 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
738 ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
740 lwkt_replymsg(&nmsg->lmsg, 0);
744 ifq_stage_detach(struct ifaltq *ifq)
746 struct netmsg_base base;
749 netmsg_init(&base, NULL, &curthread->td_msgport, 0,
750 ifq_stage_detach_handler);
751 base.lmsg.u.ms_resultp = ifq;
753 for (cpu = 0; cpu < ncpus; ++cpu)
754 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
757 struct netmsg_if_rtdel {
758 struct netmsg_base base;
763 if_rtdel_dispatch(netmsg_t msg)
765 struct netmsg_if_rtdel *rmsg = (void *)msg;
769 for (i = 1; i <= AF_MAX; i++) {
770 struct radix_node_head *rnh;
772 if ((rnh = rt_tables[cpu][i]) == NULL)
774 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
779 lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg);
781 lwkt_replymsg(&rmsg->base.lmsg, 0);
785 * Detach an interface, removing it from the
786 * list of "active" interfaces.
789 if_detach(struct ifnet *ifp)
791 struct netmsg_if_rtdel msg;
795 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
798 * Remove routes and flush queues.
802 if (ifp->if_flags & IFF_NPOLLING)
803 ifpoll_deregister(ifp);
808 if (ifq_is_enabled(&ifp->if_snd))
809 altq_disable(&ifp->if_snd);
810 if (ifq_is_attached(&ifp->if_snd))
811 altq_detach(&ifp->if_snd);
815 * Clean up all addresses.
817 ifp->if_lladdr = NULL;
819 if_purgeaddrs_nolink(ifp);
820 if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
823 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
824 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
825 ("non-link ifaddr is left on if_addrheads"));
827 ifa_ifunlink(ifa, ifp);
829 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
830 ("there are still ifaddrs left on if_addrheads"));
835 * Remove all IPv4 kernel structures related to ifp.
842 * Remove all IPv6 kernel structs related to ifp. This should be done
843 * before removing routing entries below, since IPv6 interface direct
844 * routes are expected to be removed by the IPv6-specific kernel API.
845 * Otherwise, the kernel will detect some inconsistency and bark it.
851 * Delete all remaining routes using this interface
853 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
856 KASSERT(&curthread->td_msgport != rtable_portfn(0),
857 ("if_detach in rtable thread"));
858 lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0);
860 /* Announce that the interface is gone. */
861 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
862 devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
864 SLIST_FOREACH(dp, &domains, dom_next)
865 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
866 (*dp->dom_ifdetach)(ifp,
867 ifp->if_afdata[dp->dom_family]);
870 * Remove interface from ifindex2ifp[] and maybe decrement if_index.
872 ifindex2ifnet[ifp->if_index] = NULL;
873 while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
876 TAILQ_REMOVE(&ifnet, ifp, if_link);
877 kfree(ifp->if_addrheads, M_IFADDR);
879 lwkt_synchronize_ipiqs("if_detach");
880 ifq_stage_detach(&ifp->if_snd);
882 for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
883 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
885 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
886 kfree(ifsq->ifsq_stage, M_DEVBUF);
888 kfree(ifp->if_snd.altq_subq, M_DEVBUF);
890 kfree(ifp->if_data_pcpu, M_DEVBUF);
896 * Create interface group without members
899 if_creategroup(const char *groupname)
901 struct ifg_group *ifg = NULL;
903 if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
904 M_TEMP, M_NOWAIT)) == NULL)
907 strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
909 ifg->ifg_carp_demoted = 0;
910 TAILQ_INIT(&ifg->ifg_members);
912 pfi_attach_ifgroup(ifg);
914 TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
920 * Add a group to an interface
923 if_addgroup(struct ifnet *ifp, const char *groupname)
925 struct ifg_list *ifgl;
926 struct ifg_group *ifg = NULL;
927 struct ifg_member *ifgm;
929 if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
930 groupname[strlen(groupname) - 1] <= '9')
933 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
934 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
937 if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
940 if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
945 TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
946 if (!strcmp(ifg->ifg_group, groupname))
949 if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
956 ifgl->ifgl_group = ifg;
957 ifgm->ifgm_ifp = ifp;
959 TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
960 TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
963 pfi_group_change(groupname);
970 * Remove a group from an interface
973 if_delgroup(struct ifnet *ifp, const char *groupname)
975 struct ifg_list *ifgl;
976 struct ifg_member *ifgm;
978 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
979 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
984 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
986 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
987 if (ifgm->ifgm_ifp == ifp)
991 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
995 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
996 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
998 pfi_detach_ifgroup(ifgl->ifgl_group);
1000 kfree(ifgl->ifgl_group, M_TEMP);
1003 kfree(ifgl, M_TEMP);
1006 pfi_group_change(groupname);
1013 * Stores all groups from an interface in memory pointed
1017 if_getgroup(caddr_t data, struct ifnet *ifp)
1020 struct ifg_list *ifgl;
1021 struct ifg_req ifgrq, *ifgp;
1022 struct ifgroupreq *ifgr = (struct ifgroupreq *)data;
1024 if (ifgr->ifgr_len == 0) {
1025 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1026 ifgr->ifgr_len += sizeof(struct ifg_req);
1030 len = ifgr->ifgr_len;
1031 ifgp = ifgr->ifgr_groups;
1032 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1033 if (len < sizeof(ifgrq))
1035 bzero(&ifgrq, sizeof ifgrq);
1036 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1037 sizeof(ifgrq.ifgrq_group));
1038 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1039 sizeof(struct ifg_req))))
1041 len -= sizeof(ifgrq);
1049 * Stores all members of a group in memory pointed to by data
1052 if_getgroupmembers(caddr_t data)
1054 struct ifgroupreq *ifgr = (struct ifgroupreq *)data;
1055 struct ifg_group *ifg;
1056 struct ifg_member *ifgm;
1057 struct ifg_req ifgrq, *ifgp;
1060 TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1061 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1066 if (ifgr->ifgr_len == 0) {
1067 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1068 ifgr->ifgr_len += sizeof(ifgrq);
1072 len = ifgr->ifgr_len;
1073 ifgp = ifgr->ifgr_groups;
1074 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1075 if (len < sizeof(ifgrq))
1077 bzero(&ifgrq, sizeof ifgrq);
1078 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1079 sizeof(ifgrq.ifgrq_member));
1080 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1081 sizeof(struct ifg_req))))
1083 len -= sizeof(ifgrq);
1091 * Delete Routes for a Network Interface
1093 * Called for each routing entry via the rnh->rnh_walktree() call above
1094 * to delete all route entries referencing a detaching network interface.
1097 * rn pointer to node in the routing table
1098 * arg argument passed to rnh->rnh_walktree() - detaching interface
1102 * errno failed - reason indicated
1106 if_rtdel(struct radix_node *rn, void *arg)
1108 struct rtentry *rt = (struct rtentry *)rn;
1109 struct ifnet *ifp = arg;
1112 if (rt->rt_ifp == ifp) {
1115 * Protect (sorta) against walktree recursion problems
1116 * with cloned routes
1118 if (!(rt->rt_flags & RTF_UP))
1121 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1122 rt_mask(rt), rt->rt_flags,
1125 log(LOG_WARNING, "if_rtdel: error %d\n", err);
1133 * Locate an interface based on a complete address.
1136 ifa_ifwithaddr(struct sockaddr *addr)
1140 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1141 struct ifaddr_container *ifac;
1143 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1144 struct ifaddr *ifa = ifac->ifa;
1146 if (ifa->ifa_addr->sa_family != addr->sa_family)
1148 if (sa_equal(addr, ifa->ifa_addr))
1150 if ((ifp->if_flags & IFF_BROADCAST) &&
1151 ifa->ifa_broadaddr &&
1152 /* IPv6 doesn't have broadcast */
1153 ifa->ifa_broadaddr->sa_len != 0 &&
1154 sa_equal(ifa->ifa_broadaddr, addr))
1161 * Locate the point to point interface with a given destination address.
1164 ifa_ifwithdstaddr(struct sockaddr *addr)
1168 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1169 struct ifaddr_container *ifac;
1171 if (!(ifp->if_flags & IFF_POINTOPOINT))
1174 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1175 struct ifaddr *ifa = ifac->ifa;
1177 if (ifa->ifa_addr->sa_family != addr->sa_family)
1179 if (ifa->ifa_dstaddr &&
1180 sa_equal(addr, ifa->ifa_dstaddr))
1188 * Find an interface on a specific network. If many, choice
1189 * is most specific found.
1192 ifa_ifwithnet(struct sockaddr *addr)
1195 struct ifaddr *ifa_maybe = NULL;
1196 u_int af = addr->sa_family;
1197 char *addr_data = addr->sa_data, *cplim;
1200 * AF_LINK addresses can be looked up directly by their index number,
1201 * so do that if we can.
1203 if (af == AF_LINK) {
1204 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1206 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1207 return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1211 * Scan though each interface, looking for ones that have
1212 * addresses in this address family.
1214 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1215 struct ifaddr_container *ifac;
1217 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1218 struct ifaddr *ifa = ifac->ifa;
1219 char *cp, *cp2, *cp3;
1221 if (ifa->ifa_addr->sa_family != af)
1223 if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1225 * This is a bit broken as it doesn't
1226 * take into account that the remote end may
1227 * be a single node in the network we are
1229 * The trouble is that we don't know the
1230 * netmask for the remote end.
1232 if (ifa->ifa_dstaddr != NULL &&
1233 sa_equal(addr, ifa->ifa_dstaddr))
1237 * if we have a special address handler,
1238 * then use it instead of the generic one.
1240 if (ifa->ifa_claim_addr) {
1241 if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1249 * Scan all the bits in the ifa's address.
1250 * If a bit dissagrees with what we are
1251 * looking for, mask it with the netmask
1252 * to see if it really matters.
1253 * (A byte at a time)
1255 if (ifa->ifa_netmask == 0)
1258 cp2 = ifa->ifa_addr->sa_data;
1259 cp3 = ifa->ifa_netmask->sa_data;
1260 cplim = ifa->ifa_netmask->sa_len +
1261 (char *)ifa->ifa_netmask;
1263 if ((*cp++ ^ *cp2++) & *cp3++)
1264 goto next; /* next address! */
1266 * If the netmask of what we just found
1267 * is more specific than what we had before
1268 * (if we had one) then remember the new one
1269 * before continuing to search
1270 * for an even better one.
1272 if (ifa_maybe == NULL ||
1273 rn_refines((char *)ifa->ifa_netmask,
1274 (char *)ifa_maybe->ifa_netmask))
1283 * Find an interface address specific to an interface best matching
1287 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1289 struct ifaddr_container *ifac;
1290 char *cp, *cp2, *cp3;
1292 struct ifaddr *ifa_maybe = NULL;
1293 u_int af = addr->sa_family;
1297 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1298 struct ifaddr *ifa = ifac->ifa;
1300 if (ifa->ifa_addr->sa_family != af)
1302 if (ifa_maybe == NULL)
1304 if (ifa->ifa_netmask == NULL) {
1305 if (sa_equal(addr, ifa->ifa_addr) ||
1306 (ifa->ifa_dstaddr != NULL &&
1307 sa_equal(addr, ifa->ifa_dstaddr)))
1311 if (ifp->if_flags & IFF_POINTOPOINT) {
1312 if (sa_equal(addr, ifa->ifa_dstaddr))
1316 cp2 = ifa->ifa_addr->sa_data;
1317 cp3 = ifa->ifa_netmask->sa_data;
1318 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1319 for (; cp3 < cplim; cp3++)
1320 if ((*cp++ ^ *cp2++) & *cp3)
1330 * Default action when installing a route with a Link Level gateway.
1331 * Lookup an appropriate real ifa to point to.
1332 * This should be moved to /sys/net/link.c eventually.
1335 link_rtrequest(int cmd, struct rtentry *rt)
1338 struct sockaddr *dst;
1341 if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1342 (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1344 ifa = ifaof_ifpforaddr(dst, ifp);
1346 IFAFREE(rt->rt_ifa);
1349 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1350 ifa->ifa_rtrequest(cmd, rt);
1355 * Mark an interface down and notify protocols of
1357 * NOTE: must be called at splnet or eqivalent.
1360 if_unroute(struct ifnet *ifp, int flag, int fam)
1362 struct ifaddr_container *ifac;
1364 ifp->if_flags &= ~flag;
1365 getmicrotime(&ifp->if_lastchange);
1366 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1367 struct ifaddr *ifa = ifac->ifa;
1369 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1370 kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1372 ifq_purge_all(&ifp->if_snd);
1377 * Mark an interface up and notify protocols of
1379 * NOTE: must be called at splnet or eqivalent.
1382 if_route(struct ifnet *ifp, int flag, int fam)
1384 struct ifaddr_container *ifac;
1386 ifq_purge_all(&ifp->if_snd);
1387 ifp->if_flags |= flag;
1388 getmicrotime(&ifp->if_lastchange);
1389 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1390 struct ifaddr *ifa = ifac->ifa;
1392 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1393 kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1402 * Mark an interface down and notify protocols of the transition. An
1403 * interface going down is also considered to be a synchronizing event.
1404 * We must ensure that all packet processing related to the interface
1405 * has completed before we return so e.g. the caller can free the ifnet
1406 * structure that the mbufs may be referencing.
1408 * NOTE: must be called at splnet or eqivalent.
1411 if_down(struct ifnet *ifp)
1413 if_unroute(ifp, IFF_UP, AF_UNSPEC);
1414 netmsg_service_sync();
1418 * Mark an interface up and notify protocols of
1420 * NOTE: must be called at splnet or eqivalent.
1423 if_up(struct ifnet *ifp)
1425 if_route(ifp, IFF_UP, AF_UNSPEC);
1429 * Process a link state change.
1430 * NOTE: must be called at splsoftnet or equivalent.
1433 if_link_state_change(struct ifnet *ifp)
1435 int link_state = ifp->if_link_state;
1438 devctl_notify("IFNET", ifp->if_xname,
1439 (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1443 * Handle interface watchdog timer routines. Called
1444 * from softclock, we decrement timers (if set) and
1445 * call the appropriate interface routine on expiration.
1448 if_slowtimo(void *arg)
1454 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1455 if (if_stats_compat) {
1456 IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1457 IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1458 IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1459 IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1460 IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1461 IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1462 IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1463 IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1464 IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1465 IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1466 IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1469 if (ifp->if_timer == 0 || --ifp->if_timer)
1471 if (ifp->if_watchdog) {
1472 if (ifnet_tryserialize_all(ifp)) {
1473 (*ifp->if_watchdog)(ifp);
1474 ifnet_deserialize_all(ifp);
1476 /* try again next timeout */
1484 callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1488 * Map interface name to
1489 * interface structure pointer.
1492 ifunit(const char *name)
1497 * Search all the interfaces for this name/number
1500 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1501 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1509 * Map interface name in a sockaddr_dl to
1510 * interface structure pointer.
1513 if_withname(struct sockaddr *sa)
1515 char ifname[IFNAMSIZ+1];
1516 struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1518 if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1519 (sdl->sdl_nlen > IFNAMSIZ) )
1523 * ifunit wants a null-terminated name. It may not be null-terminated
1524 * in the sockaddr. We don't want to change the caller's sockaddr,
1525 * and there might not be room to put the trailing null anyway, so we
1526 * make a local copy that we know we can null terminate safely.
1529 bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1530 ifname[sdl->sdl_nlen] = '\0';
1531 return ifunit(ifname);
1539 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1550 size_t namelen, onamelen;
1551 char new_name[IFNAMSIZ];
1553 struct sockaddr_dl *sdl;
1558 return (ifconf(cmd, data, cred));
1563 ifr = (struct ifreq *)data;
1568 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1570 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1571 cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1573 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1575 return (if_clone_destroy(ifr->ifr_name));
1576 case SIOCIFGCLONERS:
1577 return (if_clone_list((struct if_clonereq *)data));
1583 * Nominal ioctl through interface, lookup the ifp and obtain a
1584 * lock to serialize the ifconfig ioctl operation.
1586 ifp = ifunit(ifr->ifr_name);
1590 mtx_lock(&ifp->if_ioctl_mtx);
1594 ifr->ifr_index = ifp->if_index;
1598 ifr->ifr_flags = ifp->if_flags;
1599 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1603 ifr->ifr_reqcap = ifp->if_capabilities;
1604 ifr->ifr_curcap = ifp->if_capenable;
1608 ifr->ifr_metric = ifp->if_metric;
1612 ifr->ifr_mtu = ifp->if_mtu;
1616 ifr->ifr_tsolen = ifp->if_tsolen;
1620 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1621 sizeof(ifp->if_data));
1625 ifr->ifr_phys = ifp->if_physical;
1628 case SIOCGIFPOLLCPU:
1629 ifr->ifr_pollcpu = -1;
1632 case SIOCSIFPOLLCPU:
1636 error = priv_check_cred(cred, PRIV_ROOT, 0);
1639 new_flags = (ifr->ifr_flags & 0xffff) |
1640 (ifr->ifr_flagshigh << 16);
1641 if (ifp->if_flags & IFF_SMART) {
1642 /* Smart drivers twiddle their own routes */
1643 } else if (ifp->if_flags & IFF_UP &&
1644 (new_flags & IFF_UP) == 0) {
1648 } else if (new_flags & IFF_UP &&
1649 (ifp->if_flags & IFF_UP) == 0) {
1655 #ifdef IFPOLL_ENABLE
1656 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1657 if (new_flags & IFF_NPOLLING)
1658 ifpoll_register(ifp);
1660 ifpoll_deregister(ifp);
1664 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1665 (new_flags &~ IFF_CANTCHANGE);
1666 if (new_flags & IFF_PPROMISC) {
1667 /* Permanently promiscuous mode requested */
1668 ifp->if_flags |= IFF_PROMISC;
1669 } else if (ifp->if_pcount == 0) {
1670 ifp->if_flags &= ~IFF_PROMISC;
1672 if (ifp->if_ioctl) {
1673 ifnet_serialize_all(ifp);
1674 ifp->if_ioctl(ifp, cmd, data, cred);
1675 ifnet_deserialize_all(ifp);
1677 getmicrotime(&ifp->if_lastchange);
1681 error = priv_check_cred(cred, PRIV_ROOT, 0);
1684 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1688 ifnet_serialize_all(ifp);
1689 ifp->if_ioctl(ifp, cmd, data, cred);
1690 ifnet_deserialize_all(ifp);
1694 error = priv_check_cred(cred, PRIV_ROOT, 0);
1697 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1700 if (new_name[0] == '\0') {
1704 if (ifunit(new_name) != NULL) {
1709 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1711 /* Announce the departure of the interface. */
1712 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1714 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1715 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1716 /* XXX IFA_LOCK(ifa); */
1717 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1718 namelen = strlen(new_name);
1719 onamelen = sdl->sdl_nlen;
1721 * Move the address if needed. This is safe because we
1722 * allocate space for a name of length IFNAMSIZ when we
1723 * create this in if_attach().
1725 if (namelen != onamelen) {
1726 bcopy(sdl->sdl_data + onamelen,
1727 sdl->sdl_data + namelen, sdl->sdl_alen);
1729 bcopy(new_name, sdl->sdl_data, namelen);
1730 sdl->sdl_nlen = namelen;
1731 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1732 bzero(sdl->sdl_data, onamelen);
1733 while (namelen != 0)
1734 sdl->sdl_data[--namelen] = 0xff;
1735 /* XXX IFA_UNLOCK(ifa) */
1737 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1739 /* Announce the return of the interface. */
1740 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1744 error = priv_check_cred(cred, PRIV_ROOT, 0);
1747 ifp->if_metric = ifr->ifr_metric;
1748 getmicrotime(&ifp->if_lastchange);
1752 error = priv_check_cred(cred, PRIV_ROOT, 0);
1755 if (ifp->if_ioctl == NULL) {
1759 ifnet_serialize_all(ifp);
1760 error = ifp->if_ioctl(ifp, cmd, data, cred);
1761 ifnet_deserialize_all(ifp);
1763 getmicrotime(&ifp->if_lastchange);
1768 u_long oldmtu = ifp->if_mtu;
1770 error = priv_check_cred(cred, PRIV_ROOT, 0);
1773 if (ifp->if_ioctl == NULL) {
1777 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
1781 ifnet_serialize_all(ifp);
1782 error = ifp->if_ioctl(ifp, cmd, data, cred);
1783 ifnet_deserialize_all(ifp);
1785 getmicrotime(&ifp->if_lastchange);
1789 * If the link MTU changed, do network layer specific procedure.
1791 if (ifp->if_mtu != oldmtu) {
1800 error = priv_check_cred(cred, PRIV_ROOT, 0);
1804 /* XXX need driver supplied upper limit */
1805 if (ifr->ifr_tsolen <= 0) {
1809 ifp->if_tsolen = ifr->ifr_tsolen;
1814 error = priv_check_cred(cred, PRIV_ROOT, 0);
1818 /* Don't allow group membership on non-multicast interfaces. */
1819 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1824 /* Don't let users screw up protocols' entries. */
1825 if (ifr->ifr_addr.sa_family != AF_LINK) {
1830 if (cmd == SIOCADDMULTI) {
1831 struct ifmultiaddr *ifma;
1832 error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1834 error = if_delmulti(ifp, &ifr->ifr_addr);
1837 getmicrotime(&ifp->if_lastchange);
1840 case SIOCSIFPHYADDR:
1841 case SIOCDIFPHYADDR:
1843 case SIOCSIFPHYADDR_IN6:
1845 case SIOCSLIFPHYADDR:
1847 case SIOCSIFGENERIC:
1848 error = priv_check_cred(cred, PRIV_ROOT, 0);
1851 if (ifp->if_ioctl == 0) {
1855 ifnet_serialize_all(ifp);
1856 error = ifp->if_ioctl(ifp, cmd, data, cred);
1857 ifnet_deserialize_all(ifp);
1859 getmicrotime(&ifp->if_lastchange);
1863 ifs = (struct ifstat *)data;
1864 ifs->ascii[0] = '\0';
1866 case SIOCGIFPSRCADDR:
1867 case SIOCGIFPDSTADDR:
1868 case SIOCGLIFPHYADDR:
1870 case SIOCGIFGENERIC:
1871 if (ifp->if_ioctl == NULL) {
1875 ifnet_serialize_all(ifp);
1876 error = ifp->if_ioctl(ifp, cmd, data, cred);
1877 ifnet_deserialize_all(ifp);
1881 error = priv_check_cred(cred, PRIV_ROOT, 0);
1884 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
1885 ifr->ifr_addr.sa_len);
1886 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1890 oif_flags = ifp->if_flags;
1891 if (so->so_proto == 0) {
1896 error = so_pru_control_direct(so, cmd, data, ifp);
1901 case SIOCSIFDSTADDR:
1903 case SIOCSIFBRDADDR:
1904 case SIOCSIFNETMASK:
1905 #if BYTE_ORDER != BIG_ENDIAN
1906 if (ifr->ifr_addr.sa_family == 0 &&
1907 ifr->ifr_addr.sa_len < 16) {
1908 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1909 ifr->ifr_addr.sa_len = 16;
1912 if (ifr->ifr_addr.sa_len == 0)
1913 ifr->ifr_addr.sa_len = 16;
1919 case OSIOCGIFDSTADDR:
1920 cmd = SIOCGIFDSTADDR;
1922 case OSIOCGIFBRDADDR:
1923 cmd = SIOCGIFBRDADDR;
1925 case OSIOCGIFNETMASK:
1926 cmd = SIOCGIFNETMASK;
1932 error = so_pru_control_direct(so, cmd, data, ifp);
1936 case OSIOCGIFDSTADDR:
1937 case OSIOCGIFBRDADDR:
1938 case OSIOCGIFNETMASK:
1939 *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1942 #endif /* COMPAT_43 */
1944 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1946 DELAY(100);/* XXX: temporary workaround for fxp issue*/
1947 if (ifp->if_flags & IFF_UP) {
1957 mtx_unlock(&ifp->if_ioctl_mtx);
1962 * Set/clear promiscuous mode on interface ifp based on the truth value
1963 * of pswitch. The calls are reference counted so that only the first
1964 * "on" request actually has an effect, as does the final "off" request.
1965 * Results are undefined if the "off" and "on" requests are not matched.
1968 ifpromisc(struct ifnet *ifp, int pswitch)
1974 oldflags = ifp->if_flags;
1975 if (ifp->if_flags & IFF_PPROMISC) {
1976 /* Do nothing if device is in permanently promiscuous mode */
1977 ifp->if_pcount += pswitch ? 1 : -1;
1982 * If the device is not configured up, we cannot put it in
1985 if ((ifp->if_flags & IFF_UP) == 0)
1987 if (ifp->if_pcount++ != 0)
1989 ifp->if_flags |= IFF_PROMISC;
1990 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1993 if (--ifp->if_pcount > 0)
1995 ifp->if_flags &= ~IFF_PROMISC;
1996 log(LOG_INFO, "%s: promiscuous mode disabled\n",
1999 ifr.ifr_flags = ifp->if_flags;
2000 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2001 ifnet_serialize_all(ifp);
2002 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2003 ifnet_deserialize_all(ifp);
2007 ifp->if_flags = oldflags;
2012 * Return interface configuration
2013 * of system. List may be used
2014 * in later ioctl's (above) to get
2015 * other information.
2018 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2020 struct ifconf *ifc = (struct ifconf *)data;
2022 struct sockaddr *sa;
2023 struct ifreq ifr, *ifrp;
2024 int space = ifc->ifc_len, error = 0;
2026 ifrp = ifc->ifc_req;
2027 TAILQ_FOREACH(ifp, &ifnet, if_link) {
2028 struct ifaddr_container *ifac;
2031 if (space <= sizeof ifr)
2035 * Zero the stack declared structure first to prevent
2036 * memory disclosure.
2038 bzero(&ifr, sizeof(ifr));
2039 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2040 >= sizeof(ifr.ifr_name)) {
2041 error = ENAMETOOLONG;
2046 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2047 struct ifaddr *ifa = ifac->ifa;
2049 if (space <= sizeof ifr)
2052 if (cred->cr_prison &&
2053 prison_if(cred, sa))
2057 if (cmd == OSIOCGIFCONF) {
2058 struct osockaddr *osa =
2059 (struct osockaddr *)&ifr.ifr_addr;
2061 osa->sa_family = sa->sa_family;
2062 error = copyout(&ifr, ifrp, sizeof ifr);
2066 if (sa->sa_len <= sizeof(*sa)) {
2068 error = copyout(&ifr, ifrp, sizeof ifr);
2071 if (space < (sizeof ifr) + sa->sa_len -
2074 space -= sa->sa_len - sizeof(*sa);
2075 error = copyout(&ifr, ifrp,
2076 sizeof ifr.ifr_name);
2078 error = copyout(sa, &ifrp->ifr_addr,
2080 ifrp = (struct ifreq *)
2081 (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2085 space -= sizeof ifr;
2090 bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2091 error = copyout(&ifr, ifrp, sizeof ifr);
2094 space -= sizeof ifr;
2098 ifc->ifc_len -= space;
2103 * Just like if_promisc(), but for all-multicast-reception mode.
2106 if_allmulti(struct ifnet *ifp, int onswitch)
2114 if (ifp->if_amcount++ == 0) {
2115 ifp->if_flags |= IFF_ALLMULTI;
2116 ifr.ifr_flags = ifp->if_flags;
2117 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2118 ifnet_serialize_all(ifp);
2119 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2121 ifnet_deserialize_all(ifp);
2124 if (ifp->if_amcount > 1) {
2127 ifp->if_amcount = 0;
2128 ifp->if_flags &= ~IFF_ALLMULTI;
2129 ifr.ifr_flags = ifp->if_flags;
2130 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2131 ifnet_serialize_all(ifp);
2132 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2134 ifnet_deserialize_all(ifp);
2146 * Add a multicast listenership to the interface in question.
2147 * The link layer provides a routine which converts
2151 struct ifnet *ifp, /* interface to manipulate */
2152 struct sockaddr *sa, /* address to add */
2153 struct ifmultiaddr **retifma)
2155 struct sockaddr *llsa, *dupsa;
2157 struct ifmultiaddr *ifma;
2160 * If the matching multicast address already exists
2161 * then don't add a new one, just add a reference
2163 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2164 if (sa_equal(sa, ifma->ifma_addr)) {
2165 ifma->ifma_refcount++;
2173 * Give the link layer a chance to accept/reject it, and also
2174 * find out which AF_LINK address this maps to, if it isn't one
2177 if (ifp->if_resolvemulti) {
2178 ifnet_serialize_all(ifp);
2179 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2180 ifnet_deserialize_all(ifp);
2187 ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2188 dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_WAITOK);
2189 bcopy(sa, dupsa, sa->sa_len);
2191 ifma->ifma_addr = dupsa;
2192 ifma->ifma_lladdr = llsa;
2193 ifma->ifma_ifp = ifp;
2194 ifma->ifma_refcount = 1;
2195 ifma->ifma_protospec = 0;
2196 rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2199 * Some network interfaces can scan the address list at
2200 * interrupt time; lock them out.
2203 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2209 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2210 if (sa_equal(ifma->ifma_addr, llsa))
2214 ifma->ifma_refcount++;
2216 ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2217 dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_WAITOK);
2218 bcopy(llsa, dupsa, llsa->sa_len);
2219 ifma->ifma_addr = dupsa;
2220 ifma->ifma_ifp = ifp;
2221 ifma->ifma_refcount = 1;
2223 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2228 * We are certain we have added something, so call down to the
2229 * interface to let them know about it.
2232 ifnet_serialize_all(ifp);
2234 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2235 ifnet_deserialize_all(ifp);
2242 * Remove a reference to a multicast address on this interface. Yell
2243 * if the request does not match an existing membership.
2246 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2248 struct ifmultiaddr *ifma;
2250 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2251 if (sa_equal(sa, ifma->ifma_addr))
2256 if (ifma->ifma_refcount > 1) {
2257 ifma->ifma_refcount--;
2261 rt_newmaddrmsg(RTM_DELMADDR, ifma);
2262 sa = ifma->ifma_lladdr;
2264 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2266 * Make sure the interface driver is notified
2267 * in the case of a link layer mcast group being left.
2269 if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL) {
2270 ifnet_serialize_all(ifp);
2271 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2272 ifnet_deserialize_all(ifp);
2275 kfree(ifma->ifma_addr, M_IFMADDR);
2276 kfree(ifma, M_IFMADDR);
2281 * Now look for the link-layer address which corresponds to
2282 * this network address. It had been squirreled away in
2283 * ifma->ifma_lladdr for this purpose (so we don't have
2284 * to call ifp->if_resolvemulti() again), and we saved that
2285 * value in sa above. If some nasty deleted the
2286 * link-layer address out from underneath us, we can deal because
2287 * the address we stored was is not the same as the one which was
2288 * in the record for the link-layer address. (So we don't complain
2291 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2292 if (sa_equal(sa, ifma->ifma_addr))
2297 if (ifma->ifma_refcount > 1) {
2298 ifma->ifma_refcount--;
2303 ifnet_serialize_all(ifp);
2304 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2305 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2306 ifnet_deserialize_all(ifp);
2308 kfree(ifma->ifma_addr, M_IFMADDR);
2309 kfree(sa, M_IFMADDR);
2310 kfree(ifma, M_IFMADDR);
2316 * Delete all multicast group membership for an interface.
2317 * Should be used to quickly flush all multicast filters.
2320 if_delallmulti(struct ifnet *ifp)
2322 struct ifmultiaddr *ifma;
2323 struct ifmultiaddr *next;
2325 TAILQ_FOREACH_MUTABLE(ifma, &ifp->if_multiaddrs, ifma_link, next)
2326 if_delmulti(ifp, ifma->ifma_addr);
2331 * Set the link layer address on an interface.
2333 * At this time we only support certain types of interfaces,
2334 * and we don't allow the length of the address to change.
2337 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2339 struct sockaddr_dl *sdl;
2342 sdl = IF_LLSOCKADDR(ifp);
2345 if (len != sdl->sdl_alen) /* don't allow length to change */
2347 switch (ifp->if_type) {
2348 case IFT_ETHER: /* these types use struct arpcom */
2351 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2352 bcopy(lladdr, LLADDR(sdl), len);
2358 * If the interface is already up, we need
2359 * to re-init it in order to reprogram its
2362 ifnet_serialize_all(ifp);
2363 if ((ifp->if_flags & IFF_UP) != 0) {
2365 struct ifaddr_container *ifac;
2368 ifp->if_flags &= ~IFF_UP;
2369 ifr.ifr_flags = ifp->if_flags;
2370 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2371 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2373 ifp->if_flags |= IFF_UP;
2374 ifr.ifr_flags = ifp->if_flags;
2375 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2376 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2380 * Also send gratuitous ARPs to notify other nodes about
2381 * the address change.
2383 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2384 struct ifaddr *ifa = ifac->ifa;
2386 if (ifa->ifa_addr != NULL &&
2387 ifa->ifa_addr->sa_family == AF_INET)
2388 arp_gratuitous(ifp, ifa);
2392 ifnet_deserialize_all(ifp);
2396 struct ifmultiaddr *
2397 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2399 struct ifmultiaddr *ifma;
2401 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2402 if (sa_equal(ifma->ifma_addr, sa))
2409 * This function locates the first real ethernet MAC from a network
2410 * card and loads it into node, returning 0 on success or ENOENT if
2411 * no suitable interfaces were found. It is used by the uuid code to
2412 * generate a unique 6-byte number.
2415 if_getanyethermac(uint16_t *node, int minlen)
2418 struct sockaddr_dl *sdl;
2420 TAILQ_FOREACH(ifp, &ifnet, if_link) {
2421 if (ifp->if_type != IFT_ETHER)
2423 sdl = IF_LLSOCKADDR(ifp);
2424 if (sdl->sdl_alen < minlen)
2426 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2434 * The name argument must be a pointer to storage which will last as
2435 * long as the interface does. For physical devices, the result of
2436 * device_get_name(dev) is a good choice and for pseudo-devices a
2437 * static string works well.
2440 if_initname(struct ifnet *ifp, const char *name, int unit)
2442 ifp->if_dname = name;
2443 ifp->if_dunit = unit;
2444 if (unit != IF_DUNIT_NONE)
2445 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2447 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2451 if_printf(struct ifnet *ifp, const char *fmt, ...)
2456 retval = kprintf("%s: ", ifp->if_xname);
2457 __va_start(ap, fmt);
2458 retval += kvprintf(fmt, ap);
2464 if_alloc(uint8_t type)
2470 * XXX temporary hack until arpcom is setup in if_l2com
2472 if (type == IFT_ETHER)
2473 size = sizeof(struct arpcom);
2475 size = sizeof(struct ifnet);
2477 ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2479 ifp->if_type = type;
2481 if (if_com_alloc[type] != NULL) {
2482 ifp->if_l2com = if_com_alloc[type](type, ifp);
2483 if (ifp->if_l2com == NULL) {
2484 kfree(ifp, M_IFNET);
2492 if_free(struct ifnet *ifp)
2494 kfree(ifp, M_IFNET);
2498 ifq_set_classic(struct ifaltq *ifq)
2500 ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2501 ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2505 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2506 ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2510 KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2511 KASSERT(enqueue != NULL, ("enqueue is not specified"));
2512 KASSERT(dequeue != NULL, ("dequeue is not specified"));
2513 KASSERT(request != NULL, ("request is not specified"));
2515 ifq->altq_mapsubq = mapsubq;
2516 for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2517 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2519 ifsq->ifsq_enqueue = enqueue;
2520 ifsq->ifsq_dequeue = dequeue;
2521 ifsq->ifsq_request = request;
2526 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2528 m->m_nextpkt = NULL;
2529 if (ifsq->ifsq_norm_tail == NULL)
2530 ifsq->ifsq_norm_head = m;
2532 ifsq->ifsq_norm_tail->m_nextpkt = m;
2533 ifsq->ifsq_norm_tail = m;
2534 ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2538 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2540 m->m_nextpkt = NULL;
2541 if (ifsq->ifsq_prio_tail == NULL)
2542 ifsq->ifsq_prio_head = m;
2544 ifsq->ifsq_prio_tail->m_nextpkt = m;
2545 ifsq->ifsq_prio_tail = m;
2546 ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2547 ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2550 static struct mbuf *
2551 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2555 m = ifsq->ifsq_norm_head;
2557 if ((ifsq->ifsq_norm_head = m->m_nextpkt) == NULL)
2558 ifsq->ifsq_norm_tail = NULL;
2559 m->m_nextpkt = NULL;
2560 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2565 static struct mbuf *
2566 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2570 m = ifsq->ifsq_prio_head;
2572 if ((ifsq->ifsq_prio_head = m->m_nextpkt) == NULL)
2573 ifsq->ifsq_prio_tail = NULL;
2574 m->m_nextpkt = NULL;
2575 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2576 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2582 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2583 struct altq_pktattr *pa __unused)
2586 if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2587 ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2588 if ((m->m_flags & M_PRIO) &&
2589 ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen / 2) &&
2590 ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt / 2)) {
2591 struct mbuf *m_drop;
2594 * Perform drop-head on normal queue
2596 m_drop = ifsq_norm_dequeue(ifsq);
2597 if (m_drop != NULL) {
2599 ifsq_prio_enqueue(ifsq, m);
2602 /* XXX nothing could be dropped? */
2607 if (m->m_flags & M_PRIO)
2608 ifsq_prio_enqueue(ifsq, m);
2610 ifsq_norm_enqueue(ifsq, m);
2616 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2622 m = ifsq->ifsq_prio_head;
2624 m = ifsq->ifsq_norm_head;
2628 m = ifsq_prio_dequeue(ifsq);
2630 m = ifsq_norm_dequeue(ifsq);
2634 panic("unsupported ALTQ dequeue op: %d", op);
2640 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2647 m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2655 panic("unsupported ALTQ request: %d", req);
2661 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2663 struct ifnet *ifp = ifsq_get_ifp(ifsq);
2664 int running = 0, need_sched;
2667 * Try to do direct ifnet.if_start on the subqueue first, if there is
2668 * contention on the subqueue hardware serializer, ifnet.if_start on
2669 * the subqueue will be scheduled on the subqueue owner CPU.
2671 if (!ifsq_tryserialize_hw(ifsq)) {
2673 * Subqueue hardware serializer contention happened,
2674 * ifnet.if_start on the subqueue is scheduled on
2675 * the subqueue owner CPU, and we keep going.
2677 ifsq_ifstart_schedule(ifsq, 1);
2681 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2682 ifp->if_start(ifp, ifsq);
2683 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2686 need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2688 ifsq_deserialize_hw(ifsq);
2692 * More data need to be transmitted, ifnet.if_start on the
2693 * subqueue is scheduled on the subqueue owner CPU, and we
2695 * NOTE: ifnet.if_start subqueue interlock is not released.
2697 ifsq_ifstart_schedule(ifsq, force_sched);
2702 * Subqeue packets staging mechanism:
2704 * The packets enqueued into the subqueue are staged to a certain amount
2705 * before the ifnet.if_start on the subqueue is called. In this way, the
2706 * driver could avoid writing to hardware registers upon every packet,
2707 * instead, hardware registers could be written when certain amount of
2708 * packets are put onto hardware TX ring. The measurement on several modern
2709 * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
2710 * registers writing aggregation could save ~20% CPU time when 18bytes UDP
2711 * datagrams are transmitted at 1.48Mpps. The performance improvement by
2712 * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
2713 * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
2715 * Subqueue packets staging is performed for two entry points into drivers'
2716 * transmission function:
2717 * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
2718 * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
2720 * Subqueue packets staging will be stopped upon any of the following
2722 * - If the count of packets enqueued on the current CPU is great than or
2723 * equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2724 * - If the total length of packets enqueued on the current CPU is great
2725 * than or equal to the hardware's MTU - max_protohdr. max_protohdr is
2726 * cut from the hardware's MTU mainly bacause a full TCP segment's size
2727 * is usually less than hardware's MTU.
2728 * - ifsq_ifstart_schedule() is not pending on the current CPU and
2729 * ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
2731 * - The if_start_rollup(), which is registered as low priority netisr
2732 * rollup function, is called; probably because no more work is pending
2736 * Currently subqueue packet staging is only performed in netisr threads.
2739 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2741 struct ifaltq *ifq = &ifp->if_snd;
2742 struct ifaltq_subque *ifsq;
2743 int error, start = 0, len, mcast = 0, avoid_start = 0;
2744 struct ifsubq_stage_head *head = NULL;
2745 struct ifsubq_stage *stage = NULL;
2747 ifsq = ifq_map_subq(ifq, mycpuid);
2748 ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
2750 len = m->m_pkthdr.len;
2751 if (m->m_flags & M_MCAST)
2754 if (curthread->td_type == TD_TYPE_NETISR) {
2755 head = &ifsubq_stage_heads[mycpuid];
2756 stage = ifsq_get_stage(ifsq, mycpuid);
2759 stage->stg_len += len;
2760 if (stage->stg_cnt < ifsq_stage_cntmax &&
2761 stage->stg_len < (ifp->if_mtu - max_protohdr))
2766 error = ifsq_enqueue_locked(ifsq, m, pa);
2768 if (!ifsq_data_ready(ifsq)) {
2769 ALTQ_SQ_UNLOCK(ifsq);
2774 if (!ifsq_is_started(ifsq)) {
2776 ALTQ_SQ_UNLOCK(ifsq);
2779 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
2780 ifsq_stage_insert(head, stage);
2782 IFNET_STAT_INC(ifp, obytes, len);
2784 IFNET_STAT_INC(ifp, omcasts, 1);
2789 * Hold the subqueue interlock of ifnet.if_start
2791 ifsq_set_started(ifsq);
2794 ALTQ_SQ_UNLOCK(ifsq);
2797 IFNET_STAT_INC(ifp, obytes, len);
2799 IFNET_STAT_INC(ifp, omcasts, 1);
2802 if (stage != NULL) {
2803 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
2804 KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
2806 ifsq_stage_remove(head, stage);
2807 ifsq_ifstart_schedule(ifsq, 1);
2812 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
2813 ifsq_stage_remove(head, stage);
2823 ifsq_ifstart_try(ifsq, 0);
2828 ifa_create(int size, int flags)
2833 KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
2835 ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2839 ifa->ifa_containers =
2840 kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
2841 M_IFADDR, M_WAITOK | M_ZERO);
2842 ifa->ifa_ncnt = ncpus;
2843 for (i = 0; i < ncpus; ++i) {
2844 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2846 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2848 ifac->ifa_refcnt = 1;
2851 kprintf("alloc ifa %p %d\n", ifa, size);
2857 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2859 struct ifaddr *ifa = ifac->ifa;
2861 KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2862 KKASSERT(ifac->ifa_refcnt == 0);
2863 KASSERT(ifac->ifa_listmask == 0,
2864 ("ifa is still on %#x lists", ifac->ifa_listmask));
2866 ifac->ifa_magic = IFA_CONTAINER_DEAD;
2868 #ifdef IFADDR_DEBUG_VERBOSE
2869 kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2872 KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2873 ("invalid # of ifac, %d", ifa->ifa_ncnt));
2874 if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2876 kprintf("free ifa %p\n", ifa);
2878 kfree(ifa->ifa_containers, M_IFADDR);
2879 kfree(ifa, M_IFADDR);
2884 ifa_iflink_dispatch(netmsg_t nmsg)
2886 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2887 struct ifaddr *ifa = msg->ifa;
2888 struct ifnet *ifp = msg->ifp;
2890 struct ifaddr_container *ifac;
2894 ifac = &ifa->ifa_containers[cpu];
2895 ASSERT_IFAC_VALID(ifac);
2896 KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2897 ("ifaddr is on if_addrheads"));
2899 ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2901 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2903 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2907 ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2911 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2913 struct netmsg_ifaddr msg;
2915 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2916 0, ifa_iflink_dispatch);
2921 ifa_domsg(&msg.base.lmsg, 0);
2925 ifa_ifunlink_dispatch(netmsg_t nmsg)
2927 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2928 struct ifaddr *ifa = msg->ifa;
2929 struct ifnet *ifp = msg->ifp;
2931 struct ifaddr_container *ifac;
2935 ifac = &ifa->ifa_containers[cpu];
2936 ASSERT_IFAC_VALID(ifac);
2937 KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2938 ("ifaddr is not on if_addrhead"));
2940 TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2941 ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2945 ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2949 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2951 struct netmsg_ifaddr msg;
2953 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2954 0, ifa_ifunlink_dispatch);
2958 ifa_domsg(&msg.base.lmsg, 0);
2962 ifa_destroy_dispatch(netmsg_t nmsg)
2964 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2967 ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2971 ifa_destroy(struct ifaddr *ifa)
2973 struct netmsg_ifaddr msg;
2975 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2976 0, ifa_destroy_dispatch);
2979 ifa_domsg(&msg.base.lmsg, 0);
2983 ifnet_portfn(int cpu)
2985 return &ifnet_threads[cpu].td_msgport;
2989 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2991 KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2993 if (next_cpu < ncpus)
2994 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2996 lwkt_replymsg(lmsg, 0);
3000 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
3002 KKASSERT(cpu < ncpus);
3003 return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
3007 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
3009 KKASSERT(cpu < ncpus);
3010 lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
3014 * Generic netmsg service loop. Some protocols may roll their own but all
3015 * must do the basic command dispatch function call done here.
3018 ifnet_service_loop(void *arg __unused)
3022 while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
3023 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
3024 msg->base.nm_dispatch(msg);
3029 if_start_rollup(void)
3031 struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3032 struct ifsubq_stage *stage;
3034 while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3035 struct ifaltq_subque *ifsq = stage->stg_subq;
3038 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3040 ifsq_stage_remove(head, stage);
3043 ifsq_ifstart_schedule(ifsq, 1);
3048 if (!ifsq_is_started(ifsq)) {
3050 * Hold the subqueue interlock of
3053 ifsq_set_started(ifsq);
3056 ALTQ_SQ_UNLOCK(ifsq);
3059 ifsq_ifstart_try(ifsq, 1);
3061 KKASSERT((stage->stg_flags &
3062 (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3067 ifnetinit(void *dummy __unused)
3071 for (i = 0; i < ncpus; ++i) {
3072 struct thread *thr = &ifnet_threads[i];
3074 lwkt_create(ifnet_service_loop, NULL, NULL,
3075 thr, TDF_NOSTART|TDF_FORCE_SPINPORT,
3077 netmsg_service_port_init(&thr->td_msgport);
3081 for (i = 0; i < ncpus; ++i)
3082 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3083 netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3087 ifnet_byindex(unsigned short idx)
3091 return ifindex2ifnet[idx];
3095 ifaddr_byindex(unsigned short idx)
3099 ifp = ifnet_byindex(idx);
3102 return TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
3106 if_register_com_alloc(u_char type,
3107 if_com_alloc_t *a, if_com_free_t *f)
3110 KASSERT(if_com_alloc[type] == NULL,
3111 ("if_register_com_alloc: %d already registered", type));
3112 KASSERT(if_com_free[type] == NULL,
3113 ("if_register_com_alloc: %d free already registered", type));
3115 if_com_alloc[type] = a;
3116 if_com_free[type] = f;
3120 if_deregister_com_alloc(u_char type)
3123 KASSERT(if_com_alloc[type] != NULL,
3124 ("if_deregister_com_alloc: %d not registered", type));
3125 KASSERT(if_com_free[type] != NULL,
3126 ("if_deregister_com_alloc: %d free not registered", type));
3127 if_com_alloc[type] = NULL;
3128 if_com_free[type] = NULL;
3132 if_ring_count2(int cnt, int cnt_max)
3136 KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3137 ("invalid ring count max %d", cnt_max));
3146 while ((1 << (shift + 1)) <= cnt)
3150 KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3151 ("calculate cnt %d, ncpus2 %d, cnt max %d",
3152 cnt, ncpus2, cnt_max));
3157 ifq_set_maxlen(struct ifaltq *ifq, int len)
3159 ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3163 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3165 return ALTQ_SUBQ_INDEX_DEFAULT;
3169 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3171 return (cpuid & ifq->altq_subq_mask);
3175 ifsq_watchdog(void *arg)
3177 struct ifsubq_watchdog *wd = arg;
3180 if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3183 ifp = ifsq_get_ifp(wd->wd_subq);
3184 if (ifnet_tryserialize_all(ifp)) {
3185 wd->wd_watchdog(wd->wd_subq);
3186 ifnet_deserialize_all(ifp);
3188 /* try again next timeout */
3192 ifsq_watchdog_reset(wd);
3196 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3198 callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3199 ifsq_get_cpuid(wd->wd_subq));
3203 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3204 ifsq_watchdog_t watchdog)
3206 callout_init_mp(&wd->wd_callout);
3209 wd->wd_watchdog = watchdog;
3213 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3216 ifsq_watchdog_reset(wd);
3220 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3223 callout_stop(&wd->wd_callout);