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 * 4. 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 8.5 (Berkeley) 1/9/95
33 #include "opt_compat.h"
34 #include "opt_inet6.h"
37 #include <sys/param.h>
38 #include <sys/types.h>
40 #include <sys/malloc.h>
44 #include <sys/systm.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/protosw.h>
50 #include <sys/kernel.h>
52 #include <sys/refcount.h>
53 #include <sys/module.h>
54 #include <sys/rwlock.h>
55 #include <sys/sockio.h>
56 #include <sys/syslog.h>
57 #include <sys/sysctl.h>
58 #include <sys/taskqueue.h>
59 #include <sys/domain.h>
61 #include <machine/stdarg.h>
65 #include <net/if_arp.h>
66 #include <net/if_clone.h>
67 #include <net/if_dl.h>
68 #include <net/if_types.h>
69 #include <net/if_var.h>
70 #include <net/radix.h>
71 #include <net/route.h>
74 #if defined(INET) || defined(INET6)
76 #include <netinet/in.h>
77 #include <netinet/in_var.h>
78 #include <netinet/ip_carp.h>
80 #include <netinet6/in6_var.h>
81 #include <netinet6/in6_ifattach.h>
85 #include <netinet/if_ether.h>
88 #include <security/mac/mac_framework.h>
90 #ifdef COMPAT_FREEBSD32
91 #include <sys/mount.h>
92 #include <compat/freebsd32/freebsd32.h>
95 struct ifindex_entry {
96 struct ifnet *ife_ifnet;
99 static int slowtimo_started;
101 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
102 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
104 TUNABLE_INT("net.link.ifqmaxlen", &ifqmaxlen);
105 SYSCTL_UINT(_net_link, OID_AUTO, ifqmaxlen, CTLFLAG_RDTUN,
106 &ifqmaxlen, 0, "max send queue size");
108 /* Log link state change events */
109 static int log_link_state_change = 1;
111 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
112 &log_link_state_change, 0,
113 "log interface link state change events");
115 /* Interface description */
116 static unsigned int ifdescr_maxlen = 1024;
117 SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW,
119 "administrative maximum length for interface description");
121 MALLOC_DEFINE(M_IFDESCR, "ifdescr", "ifnet descriptions");
123 /* global sx for non-critical path ifdescr */
124 static struct sx ifdescr_sx;
125 SX_SYSINIT(ifdescr_sx, &ifdescr_sx, "ifnet descr");
127 void (*bstp_linkstate_p)(struct ifnet *ifp, int state);
128 void (*ng_ether_link_state_p)(struct ifnet *ifp, int state);
129 void (*lagg_linkstate_p)(struct ifnet *ifp, int state);
130 /* These are external hooks for CARP. */
131 void (*carp_linkstate_p)(struct ifnet *ifp);
132 #if defined(INET) || defined(INET6)
133 struct ifnet *(*carp_forus_p)(struct ifnet *ifp, u_char *dhost);
134 int (*carp_output_p)(struct ifnet *ifp, struct mbuf *m,
135 struct sockaddr *sa, struct rtentry *rt);
138 int (*carp_iamatch_p)(struct ifnet *, struct in_ifaddr *, struct in_addr *,
142 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6);
143 caddr_t (*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m,
144 const struct in6_addr *taddr);
147 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
150 * XXX: Style; these should be sorted alphabetically, and unprototyped
151 * static functions should be prototyped. Currently they are sorted by
154 static void if_attachdomain(void *);
155 static void if_attachdomain1(struct ifnet *);
156 static int ifconf(u_long, caddr_t);
157 static void if_freemulti(struct ifmultiaddr *);
158 static void if_init(void *);
159 static void if_grow(void);
160 static void if_check(void *);
161 static void if_route(struct ifnet *, int flag, int fam);
162 static int if_setflag(struct ifnet *, int, int, int *, int);
163 static void if_slowtimo(void *);
164 static int if_transmit(struct ifnet *ifp, struct mbuf *m);
165 static void if_unroute(struct ifnet *, int flag, int fam);
166 static void link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
167 static int if_rtdel(struct radix_node *, void *);
168 static int ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
169 static int if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
170 static void do_link_state_change(void *, int);
171 static int if_getgroup(struct ifgroupreq *, struct ifnet *);
172 static int if_getgroupmembers(struct ifgroupreq *);
173 static void if_delgroups(struct ifnet *);
174 static void if_attach_internal(struct ifnet *, int);
175 static void if_detach_internal(struct ifnet *, int);
179 * XXX: declare here to avoid to include many inet6 related files..
180 * should be more generalized?
182 extern void nd6_setmtu(struct ifnet *);
185 VNET_DEFINE(int, if_index);
186 int ifqmaxlen = IFQ_MAXLEN;
187 VNET_DEFINE(struct ifnethead, ifnet); /* depend on static init XXX */
188 VNET_DEFINE(struct ifgrouphead, ifg_head);
190 static VNET_DEFINE(int, if_indexlim) = 8;
192 /* Table of ifnet by index. */
193 VNET_DEFINE(struct ifindex_entry *, ifindex_table);
195 #define V_if_indexlim VNET(if_indexlim)
196 #define V_ifindex_table VNET(ifindex_table)
199 * The global network interface list (V_ifnet) and related state (such as
200 * if_index, if_indexlim, and ifindex_table) are protected by an sxlock and
201 * an rwlock. Either may be acquired shared to stablize the list, but both
202 * must be acquired writable to modify the list. This model allows us to
203 * both stablize the interface list during interrupt thread processing, but
204 * also to stablize it over long-running ioctls, without introducing priority
205 * inversions and deadlocks.
207 struct rwlock ifnet_rwlock;
208 struct sx ifnet_sxlock;
211 * The allocation of network interfaces is a rather non-atomic affair; we
212 * need to select an index before we are ready to expose the interface for
213 * use, so will use this pointer value to indicate reservation.
215 #define IFNET_HOLD (void *)(uintptr_t)(-1)
217 static if_com_alloc_t *if_com_alloc[256];
218 static if_com_free_t *if_com_free[256];
221 * System initialization
223 SYSINIT(interface_check, SI_SUB_PROTO_IF, SI_ORDER_FIRST, if_check, NULL);
225 MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
226 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
227 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
230 ifnet_byindex_locked(u_short idx)
233 if (idx > V_if_index)
235 if (V_ifindex_table[idx].ife_ifnet == IFNET_HOLD)
237 return (V_ifindex_table[idx].ife_ifnet);
241 ifnet_byindex(u_short idx)
245 IFNET_RLOCK_NOSLEEP();
246 ifp = ifnet_byindex_locked(idx);
247 IFNET_RUNLOCK_NOSLEEP();
252 ifnet_byindex_ref(u_short idx)
256 IFNET_RLOCK_NOSLEEP();
257 ifp = ifnet_byindex_locked(idx);
258 if (ifp == NULL || (ifp->if_flags & IFF_DYING)) {
259 IFNET_RUNLOCK_NOSLEEP();
263 IFNET_RUNLOCK_NOSLEEP();
268 * Allocate an ifindex array entry; return 0 on success or an error on
272 ifindex_alloc_locked(u_short *idxp)
276 IFNET_WLOCK_ASSERT();
280 * Try to find an empty slot below V_if_index. If we fail, take the
283 for (idx = 1; idx <= V_if_index; idx++) {
284 if (V_ifindex_table[idx].ife_ifnet == NULL)
288 /* Catch if_index overflow. */
291 if (idx >= V_if_indexlim) {
295 if (idx > V_if_index)
302 ifindex_free_locked(u_short idx)
305 IFNET_WLOCK_ASSERT();
307 V_ifindex_table[idx].ife_ifnet = NULL;
308 while (V_if_index > 0 &&
309 V_ifindex_table[V_if_index].ife_ifnet == NULL)
314 ifindex_free(u_short idx)
318 ifindex_free_locked(idx);
323 ifnet_setbyindex_locked(u_short idx, struct ifnet *ifp)
326 IFNET_WLOCK_ASSERT();
328 V_ifindex_table[idx].ife_ifnet = ifp;
332 ifnet_setbyindex(u_short idx, struct ifnet *ifp)
336 ifnet_setbyindex_locked(idx, ifp);
341 ifaddr_byindex(u_short idx)
345 IFNET_RLOCK_NOSLEEP();
346 ifa = ifnet_byindex_locked(idx)->if_addr;
349 IFNET_RUNLOCK_NOSLEEP();
354 * Network interface utility routines.
356 * Routines with ifa_ifwith* names take sockaddr *'s as
361 vnet_if_init(const void *unused __unused)
364 TAILQ_INIT(&V_ifnet);
365 TAILQ_INIT(&V_ifg_head);
367 if_grow(); /* create initial table */
369 vnet_if_clone_init();
371 VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init,
376 if_init(void *dummy __unused)
382 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
387 vnet_if_uninit(const void *unused __unused)
390 VNET_ASSERT(TAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p "
391 "not empty", __func__, __LINE__, &V_ifnet));
392 VNET_ASSERT(TAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p "
393 "not empty", __func__, __LINE__, &V_ifg_head));
395 free((caddr_t)V_ifindex_table, M_IFNET);
397 VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
398 vnet_if_uninit, NULL);
406 struct ifindex_entry *e;
408 IFNET_WLOCK_ASSERT();
409 oldlim = V_if_indexlim;
411 n = (oldlim << 1) * sizeof(*e);
412 e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
414 if (V_if_indexlim != oldlim) {
418 if (V_ifindex_table != NULL) {
419 memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2);
420 free((caddr_t)V_ifindex_table, M_IFNET);
427 if_check(void *dummy __unused)
431 * If at least one interface added during boot uses
432 * if_watchdog then start the timer.
434 if (slowtimo_started)
439 * Allocate a struct ifnet and an index for an interface. A layer 2
440 * common structure will also be allocated if an allocation routine is
441 * registered for the passed type.
444 if_alloc(u_char type)
449 ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
451 if (ifindex_alloc_locked(&idx) != 0) {
456 ifnet_setbyindex_locked(idx, IFNET_HOLD);
460 ifp->if_alloctype = type;
461 if (if_com_alloc[type] != NULL) {
462 ifp->if_l2com = if_com_alloc[type](type, ifp);
463 if (ifp->if_l2com == NULL) {
470 IF_ADDR_LOCK_INIT(ifp);
471 TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
472 ifp->if_afdata_initialized = 0;
473 IF_AFDATA_LOCK_INIT(ifp);
474 TAILQ_INIT(&ifp->if_addrhead);
475 TAILQ_INIT(&ifp->if_prefixhead);
476 TAILQ_INIT(&ifp->if_multiaddrs);
477 TAILQ_INIT(&ifp->if_groups);
481 ifq_init(&ifp->if_snd, ifp);
483 refcount_init(&ifp->if_refcount, 1); /* Index reference. */
484 ifnet_setbyindex(ifp->if_index, ifp);
489 * Do the actual work of freeing a struct ifnet, and layer 2 common
490 * structure. This call is made when the last reference to an
491 * interface is released.
494 if_free_internal(struct ifnet *ifp)
497 KASSERT((ifp->if_flags & IFF_DYING),
498 ("if_free_internal: interface not dying"));
500 if (if_com_free[ifp->if_alloctype] != NULL)
501 if_com_free[ifp->if_alloctype](ifp->if_l2com,
505 mac_ifnet_destroy(ifp);
507 if (ifp->if_description != NULL)
508 free(ifp->if_description, M_IFDESCR);
509 IF_AFDATA_DESTROY(ifp);
510 IF_ADDR_LOCK_DESTROY(ifp);
511 ifq_delete(&ifp->if_snd);
516 * This version should only be called by intefaces that switch their type
517 * after calling if_alloc(). if_free_type() will go away again now that we
518 * have if_alloctype to cache the original allocation type. For now, assert
519 * that they match, since we require that in practice.
522 if_free_type(struct ifnet *ifp, u_char type)
525 KASSERT(ifp->if_alloctype == type,
526 ("if_free_type: type (%d) != alloctype (%d)", type,
529 ifp->if_flags |= IFF_DYING; /* XXX: Locking */
532 KASSERT(ifp == ifnet_byindex_locked(ifp->if_index),
533 ("%s: freeing unallocated ifnet", ifp->if_xname));
535 ifindex_free_locked(ifp->if_index);
538 if (!refcount_release(&ifp->if_refcount))
540 if_free_internal(ifp);
544 * This is the normal version of if_free(), used by device drivers to free a
545 * detached network interface. The contents of if_free_type() will move into
546 * here when if_free_type() goes away.
549 if_free(struct ifnet *ifp)
552 if_free_type(ifp, ifp->if_alloctype);
556 * Interfaces to keep an ifnet type-stable despite the possibility of the
557 * driver calling if_free(). If there are additional references, we defer
558 * freeing the underlying data structure.
561 if_ref(struct ifnet *ifp)
564 /* We don't assert the ifnet list lock here, but arguably should. */
565 refcount_acquire(&ifp->if_refcount);
569 if_rele(struct ifnet *ifp)
572 if (!refcount_release(&ifp->if_refcount))
574 if_free_internal(ifp);
578 ifq_init(struct ifaltq *ifq, struct ifnet *ifp)
581 mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
583 if (ifq->ifq_maxlen == 0)
584 ifq->ifq_maxlen = ifqmaxlen;
587 ifq->altq_disc = NULL;
588 ifq->altq_flags &= ALTQF_CANTCHANGE;
589 ifq->altq_tbr = NULL;
594 ifq_delete(struct ifaltq *ifq)
596 mtx_destroy(&ifq->ifq_mtx);
600 * Perform generic interface initalization tasks and attach the interface
601 * to the list of "active" interfaces. If vmove flag is set on entry
602 * to if_attach_internal(), perform only a limited subset of initialization
603 * tasks, given that we are moving from one vnet to another an ifnet which
604 * has already been fully initialized.
607 * - The decision to return void and thus require this function to
608 * succeed is questionable.
609 * - We should probably do more sanity checking. For instance we don't
610 * do anything to insure if_xname is unique or non-empty.
613 if_attach(struct ifnet *ifp)
616 if_attach_internal(ifp, 0);
620 if_attach_internal(struct ifnet *ifp, int vmove)
622 unsigned socksize, ifasize;
623 int namelen, masklen;
624 struct sockaddr_dl *sdl;
627 if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
628 panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
632 ifp->if_vnet = curvnet;
633 if (ifp->if_home_vnet == NULL)
634 ifp->if_home_vnet = curvnet;
637 if_addgroup(ifp, IFG_ALL);
639 getmicrotime(&ifp->if_lastchange);
640 ifp->if_data.ifi_epoch = time_uptime;
641 ifp->if_data.ifi_datalen = sizeof(struct if_data);
643 KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
644 (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
645 ("transmit and qflush must both either be set or both be NULL"));
646 if (ifp->if_transmit == NULL) {
647 ifp->if_transmit = if_transmit;
648 ifp->if_qflush = if_qflush;
653 mac_ifnet_create(ifp);
657 * Create a Link Level name for this device.
659 namelen = strlen(ifp->if_xname);
661 * Always save enough space for any possiable name so we
662 * can do a rename in place later.
664 masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
665 socksize = masklen + ifp->if_addrlen;
666 if (socksize < sizeof(*sdl))
667 socksize = sizeof(*sdl);
668 socksize = roundup2(socksize, sizeof(long));
669 ifasize = sizeof(*ifa) + 2 * socksize;
670 ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
672 sdl = (struct sockaddr_dl *)(ifa + 1);
673 sdl->sdl_len = socksize;
674 sdl->sdl_family = AF_LINK;
675 bcopy(ifp->if_xname, sdl->sdl_data, namelen);
676 sdl->sdl_nlen = namelen;
677 sdl->sdl_index = ifp->if_index;
678 sdl->sdl_type = ifp->if_type;
681 ifa->ifa_rtrequest = link_rtrequest;
682 ifa->ifa_addr = (struct sockaddr *)sdl;
683 sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
684 ifa->ifa_netmask = (struct sockaddr *)sdl;
685 sdl->sdl_len = masklen;
687 sdl->sdl_data[--namelen] = 0xff;
688 TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
689 /* Reliably crash if used uninitialized. */
690 ifp->if_broadcastaddr = NULL;
695 * Update the interface index in the link layer address
698 for (ifa = ifp->if_addr; ifa != NULL;
699 ifa = TAILQ_NEXT(ifa, ifa_link)) {
700 if (ifa->ifa_addr->sa_family == AF_LINK) {
701 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
702 sdl->sdl_index = ifp->if_index;
709 TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
711 curvnet->vnet_ifcnt++;
715 if (domain_init_status >= 2)
716 if_attachdomain1(ifp);
718 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
719 if (IS_DEFAULT_VNET(curvnet))
720 devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
722 /* Announce the interface. */
723 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
725 if (!vmove && ifp->if_watchdog != NULL) {
727 "WARNING: using obsoleted if_watchdog interface\n");
730 * Note that we need if_slowtimo(). If this happens after
731 * boot, then call if_slowtimo() directly.
733 if (atomic_cmpset_int(&slowtimo_started, 0, 1) && !cold)
739 if_attachdomain(void *dummy)
745 TAILQ_FOREACH(ifp, &V_ifnet, if_link)
746 if_attachdomain1(ifp);
749 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
750 if_attachdomain, NULL);
753 if_attachdomain1(struct ifnet *ifp)
761 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
762 * cannot lock ifp->if_afdata initialization, entirely.
764 if (IF_AFDATA_TRYLOCK(ifp) == 0) {
768 if (ifp->if_afdata_initialized >= domain_init_status) {
769 IF_AFDATA_UNLOCK(ifp);
771 printf("if_attachdomain called more than once on %s\n",
775 ifp->if_afdata_initialized = domain_init_status;
776 IF_AFDATA_UNLOCK(ifp);
778 /* address family dependent data region */
779 bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
780 for (dp = domains; dp; dp = dp->dom_next) {
781 if (dp->dom_ifattach)
782 ifp->if_afdata[dp->dom_family] =
783 (*dp->dom_ifattach)(ifp);
790 * Remove any unicast or broadcast network addresses from an interface.
793 if_purgeaddrs(struct ifnet *ifp)
795 struct ifaddr *ifa, *next;
797 TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
798 if (ifa->ifa_addr->sa_family == AF_LINK)
801 /* XXX: Ugly!! ad hoc just for INET */
802 if (ifa->ifa_addr->sa_family == AF_INET) {
803 struct ifaliasreq ifr;
805 bzero(&ifr, sizeof(ifr));
806 ifr.ifra_addr = *ifa->ifa_addr;
807 if (ifa->ifa_dstaddr)
808 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
809 if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
815 if (ifa->ifa_addr->sa_family == AF_INET6) {
817 /* ifp_addrhead is already updated */
821 TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
827 * Remove any multicast network addresses from an interface when an ifnet
831 if_purgemaddrs(struct ifnet *ifp)
833 struct ifmultiaddr *ifma;
834 struct ifmultiaddr *next;
837 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
838 if_delmulti_locked(ifp, ifma, 1);
843 * Detach an interface, removing it from the list of "active" interfaces.
844 * If vmove flag is set on entry to if_detach_internal(), perform only a
845 * limited subset of cleanup tasks, given that we are moving an ifnet from
846 * one vnet to another, where it must be fully operational.
848 * XXXRW: There are some significant questions about event ordering, and
849 * how to prevent things from starting to use the interface during detach.
852 if_detach(struct ifnet *ifp)
855 if_detach_internal(ifp, 0);
859 if_detach_internal(struct ifnet *ifp, int vmove)
862 struct radix_node_head *rnh;
869 TAILQ_FOREACH(iter, &V_ifnet, if_link)
871 TAILQ_REMOVE(&V_ifnet, ifp, if_link);
877 curvnet->vnet_ifcnt--;
882 panic("%s: ifp=%p not on the ifnet tailq %p",
883 __func__, ifp, &V_ifnet);
885 return; /* XXX this should panic as well? */
889 * Remove/wait for pending events.
891 taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
894 * Remove routes and flush queues.
898 if (ALTQ_IS_ENABLED(&ifp->if_snd))
899 altq_disable(&ifp->if_snd);
900 if (ALTQ_IS_ATTACHED(&ifp->if_snd))
901 altq_detach(&ifp->if_snd);
912 * Remove all IPv6 kernel structs related to ifp. This should be done
913 * before removing routing entries below, since IPv6 interface direct
914 * routes are expected to be removed by the IPv6-specific kernel API.
915 * Otherwise, the kernel will detect some inconsistency and bark it.
923 * Prevent further calls into the device driver via ifnet.
928 * Remove link ifaddr pointer and maybe decrement if_index.
929 * Clean up all addresses.
933 /* We can now free link ifaddr. */
934 if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
935 ifa = TAILQ_FIRST(&ifp->if_addrhead);
936 TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
942 * Delete all remaining routes using this interface
943 * Unfortuneatly the only way to do this is to slog through
944 * the entire routing table looking for routes which point
945 * to this interface...oh well...
947 for (i = 1; i <= AF_MAX; i++) {
948 for (j = 0; j < rt_numfibs; j++) {
949 rnh = rt_tables_get_rnh(j, i);
952 RADIX_NODE_HEAD_LOCK(rnh);
953 (void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
954 RADIX_NODE_HEAD_UNLOCK(rnh);
958 /* Announce that the interface is gone. */
959 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
960 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
961 if (IS_DEFAULT_VNET(curvnet))
962 devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
966 * We cannot hold the lock over dom_ifdetach calls as they might
967 * sleep, for example trying to drain a callout, thus open up the
968 * theoretical race with re-attaching.
971 i = ifp->if_afdata_initialized;
972 ifp->if_afdata_initialized = 0;
973 IF_AFDATA_UNLOCK(ifp);
974 for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
975 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
976 (*dp->dom_ifdetach)(ifp,
977 ifp->if_afdata[dp->dom_family]);
983 * if_vmove() performs a limited version of if_detach() in current
984 * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
985 * An attempt is made to shrink if_index in current vnet, find an
986 * unused if_index in target vnet and calls if_grow() if necessary,
987 * and finally find an unused if_xname for the target vnet.
990 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
995 * Detach from current vnet, but preserve LLADDR info, do not
996 * mark as dead etc. so that the ifnet can be reattached later.
998 if_detach_internal(ifp, 1);
1001 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
1002 * the if_index for that vnet if possible.
1004 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
1005 * or we'd lock on one vnet and unlock on another.
1008 ifindex_free_locked(ifp->if_index);
1012 * Perform interface-specific reassignment tasks, if provided by
1015 if (ifp->if_reassign != NULL)
1016 ifp->if_reassign(ifp, new_vnet, NULL);
1019 * Switch to the context of the target vnet.
1021 CURVNET_SET_QUIET(new_vnet);
1024 if (ifindex_alloc_locked(&idx) != 0) {
1026 panic("if_index overflow");
1028 ifp->if_index = idx;
1029 ifnet_setbyindex_locked(ifp->if_index, ifp);
1032 if_attach_internal(ifp, 1);
1038 * Move an ifnet to or from another child prison/vnet, specified by the jail id.
1041 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
1046 /* Try to find the prison within our visibility. */
1047 sx_slock(&allprison_lock);
1048 pr = prison_find_child(td->td_ucred->cr_prison, jid);
1049 sx_sunlock(&allprison_lock);
1052 prison_hold_locked(pr);
1053 mtx_unlock(&pr->pr_mtx);
1055 /* Do not try to move the iface from and to the same prison. */
1056 if (pr->pr_vnet == ifp->if_vnet) {
1061 /* Make sure the named iface does not exists in the dst. prison/vnet. */
1062 /* XXX Lock interfaces to avoid races. */
1063 CURVNET_SET_QUIET(pr->pr_vnet);
1064 difp = ifunit(ifname);
1071 /* Move the interface into the child jail/vnet. */
1072 if_vmove(ifp, pr->pr_vnet);
1074 /* Report the new if_xname back to the userland. */
1075 sprintf(ifname, "%s", ifp->if_xname);
1082 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1085 struct vnet *vnet_dst;
1088 /* Try to find the prison within our visibility. */
1089 sx_slock(&allprison_lock);
1090 pr = prison_find_child(td->td_ucred->cr_prison, jid);
1091 sx_sunlock(&allprison_lock);
1094 prison_hold_locked(pr);
1095 mtx_unlock(&pr->pr_mtx);
1097 /* Make sure the named iface exists in the source prison/vnet. */
1098 CURVNET_SET(pr->pr_vnet);
1099 ifp = ifunit(ifname); /* XXX Lock to avoid races. */
1106 /* Do not try to move the iface from and to the same prison. */
1107 vnet_dst = TD_TO_VNET(td);
1108 if (vnet_dst == ifp->if_vnet) {
1114 /* Get interface back from child jail/vnet. */
1115 if_vmove(ifp, vnet_dst);
1118 /* Report the new if_xname back to the userland. */
1119 sprintf(ifname, "%s", ifp->if_xname);
1127 * Add a group to an interface
1130 if_addgroup(struct ifnet *ifp, const char *groupname)
1132 struct ifg_list *ifgl;
1133 struct ifg_group *ifg = NULL;
1134 struct ifg_member *ifgm;
1136 if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1137 groupname[strlen(groupname) - 1] <= '9')
1141 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1142 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1147 if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1148 M_NOWAIT)) == NULL) {
1153 if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1154 M_TEMP, M_NOWAIT)) == NULL) {
1160 TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1161 if (!strcmp(ifg->ifg_group, groupname))
1165 if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1166 M_TEMP, M_NOWAIT)) == NULL) {
1172 strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1173 ifg->ifg_refcnt = 0;
1174 TAILQ_INIT(&ifg->ifg_members);
1175 EVENTHANDLER_INVOKE(group_attach_event, ifg);
1176 TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1180 ifgl->ifgl_group = ifg;
1181 ifgm->ifgm_ifp = ifp;
1184 TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1185 TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1186 IF_ADDR_UNLOCK(ifp);
1190 EVENTHANDLER_INVOKE(group_change_event, groupname);
1196 * Remove a group from an interface
1199 if_delgroup(struct ifnet *ifp, const char *groupname)
1201 struct ifg_list *ifgl;
1202 struct ifg_member *ifgm;
1205 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1206 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1214 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1215 IF_ADDR_UNLOCK(ifp);
1217 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1218 if (ifgm->ifgm_ifp == ifp)
1222 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1226 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1227 TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1228 EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1229 free(ifgl->ifgl_group, M_TEMP);
1235 EVENTHANDLER_INVOKE(group_change_event, groupname);
1241 * Remove an interface from all groups
1244 if_delgroups(struct ifnet *ifp)
1246 struct ifg_list *ifgl;
1247 struct ifg_member *ifgm;
1248 char groupname[IFNAMSIZ];
1251 while (!TAILQ_EMPTY(&ifp->if_groups)) {
1252 ifgl = TAILQ_FIRST(&ifp->if_groups);
1254 strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1257 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1258 IF_ADDR_UNLOCK(ifp);
1260 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1261 if (ifgm->ifgm_ifp == ifp)
1265 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1270 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1271 TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1272 EVENTHANDLER_INVOKE(group_detach_event,
1274 free(ifgl->ifgl_group, M_TEMP);
1280 EVENTHANDLER_INVOKE(group_change_event, groupname);
1288 * Stores all groups from an interface in memory pointed
1292 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1295 struct ifg_list *ifgl;
1296 struct ifg_req ifgrq, *ifgp;
1297 struct ifgroupreq *ifgr = data;
1299 if (ifgr->ifgr_len == 0) {
1301 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1302 ifgr->ifgr_len += sizeof(struct ifg_req);
1303 IF_ADDR_UNLOCK(ifp);
1307 len = ifgr->ifgr_len;
1308 ifgp = ifgr->ifgr_groups;
1311 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1312 if (len < sizeof(ifgrq)) {
1313 IF_ADDR_UNLOCK(ifp);
1316 bzero(&ifgrq, sizeof ifgrq);
1317 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1318 sizeof(ifgrq.ifgrq_group));
1319 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1320 IF_ADDR_UNLOCK(ifp);
1323 len -= sizeof(ifgrq);
1326 IF_ADDR_UNLOCK(ifp);
1332 * Stores all members of a group in memory pointed to by data
1335 if_getgroupmembers(struct ifgroupreq *data)
1337 struct ifgroupreq *ifgr = data;
1338 struct ifg_group *ifg;
1339 struct ifg_member *ifgm;
1340 struct ifg_req ifgrq, *ifgp;
1344 TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1345 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1352 if (ifgr->ifgr_len == 0) {
1353 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1354 ifgr->ifgr_len += sizeof(ifgrq);
1359 len = ifgr->ifgr_len;
1360 ifgp = ifgr->ifgr_groups;
1361 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1362 if (len < sizeof(ifgrq)) {
1366 bzero(&ifgrq, sizeof ifgrq);
1367 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1368 sizeof(ifgrq.ifgrq_member));
1369 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1373 len -= sizeof(ifgrq);
1382 * Delete Routes for a Network Interface
1384 * Called for each routing entry via the rnh->rnh_walktree() call above
1385 * to delete all route entries referencing a detaching network interface.
1388 * rn pointer to node in the routing table
1389 * arg argument passed to rnh->rnh_walktree() - detaching interface
1393 * errno failed - reason indicated
1397 if_rtdel(struct radix_node *rn, void *arg)
1399 struct rtentry *rt = (struct rtentry *)rn;
1400 struct ifnet *ifp = arg;
1403 if (rt->rt_ifp == ifp) {
1406 * Protect (sorta) against walktree recursion problems
1407 * with cloned routes
1409 if ((rt->rt_flags & RTF_UP) == 0)
1412 err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1413 rt_mask(rt), rt->rt_flags|RTF_RNH_LOCKED,
1414 (struct rtentry **) NULL, rt->rt_fibnum);
1416 log(LOG_WARNING, "if_rtdel: error %d\n", err);
1424 * Wrapper functions for struct ifnet address list locking macros. These are
1425 * used by kernel modules to avoid encoding programming interface or binary
1426 * interface assumptions that may be violated when kernel-internal locking
1427 * approaches change.
1430 if_addr_rlock(struct ifnet *ifp)
1437 if_addr_runlock(struct ifnet *ifp)
1440 IF_ADDR_UNLOCK(ifp);
1444 if_maddr_rlock(struct ifnet *ifp)
1451 if_maddr_runlock(struct ifnet *ifp)
1454 IF_ADDR_UNLOCK(ifp);
1458 * Reference count functions for ifaddrs.
1461 ifa_init(struct ifaddr *ifa)
1464 mtx_init(&ifa->ifa_mtx, "ifaddr", NULL, MTX_DEF);
1465 refcount_init(&ifa->ifa_refcnt, 1);
1469 ifa_ref(struct ifaddr *ifa)
1472 refcount_acquire(&ifa->ifa_refcnt);
1476 ifa_free(struct ifaddr *ifa)
1479 if (refcount_release(&ifa->ifa_refcnt)) {
1480 mtx_destroy(&ifa->ifa_mtx);
1481 free(ifa, M_IFADDR);
1486 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1489 struct rtentry *rt = NULL;
1490 struct rt_addrinfo info;
1491 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1493 bzero(&info, sizeof(info));
1494 info.rti_ifp = V_loif;
1495 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1496 info.rti_info[RTAX_DST] = ia;
1497 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1498 error = rtrequest1_fib(RTM_ADD, &info, &rt, 0);
1500 if (error == 0 && rt != NULL) {
1502 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1503 ifa->ifa_ifp->if_type;
1504 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1505 ifa->ifa_ifp->if_index;
1508 } else if (error != 0)
1509 log(LOG_INFO, "ifa_add_loopback_route: insertion failed\n");
1515 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1518 struct rt_addrinfo info;
1519 struct sockaddr_dl null_sdl;
1521 bzero(&null_sdl, sizeof(null_sdl));
1522 null_sdl.sdl_len = sizeof(null_sdl);
1523 null_sdl.sdl_family = AF_LINK;
1524 null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1525 null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1526 bzero(&info, sizeof(info));
1527 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1528 info.rti_info[RTAX_DST] = ia;
1529 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1530 error = rtrequest1_fib(RTM_DELETE, &info, NULL, 0);
1533 log(LOG_INFO, "ifa_del_loopback_route: deletion failed\n");
1539 * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1540 * structs used to represent other address families, it is necessary
1541 * to perform a different comparison.
1544 #define sa_equal(a1, a2) \
1545 (bcmp((a1), (a2), ((a1))->sa_len) == 0)
1547 #define sa_dl_equal(a1, a2) \
1548 ((((struct sockaddr_dl *)(a1))->sdl_len == \
1549 ((struct sockaddr_dl *)(a2))->sdl_len) && \
1550 (bcmp(LLADDR((struct sockaddr_dl *)(a1)), \
1551 LLADDR((struct sockaddr_dl *)(a2)), \
1552 ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1555 * Locate an interface based on a complete address.
1558 static struct ifaddr *
1559 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1564 IFNET_RLOCK_NOSLEEP();
1565 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1567 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1568 if (ifa->ifa_addr->sa_family != addr->sa_family)
1570 if (sa_equal(addr, ifa->ifa_addr)) {
1573 IF_ADDR_UNLOCK(ifp);
1576 /* IP6 doesn't have broadcast */
1577 if ((ifp->if_flags & IFF_BROADCAST) &&
1578 ifa->ifa_broadaddr &&
1579 ifa->ifa_broadaddr->sa_len != 0 &&
1580 sa_equal(ifa->ifa_broadaddr, addr)) {
1583 IF_ADDR_UNLOCK(ifp);
1587 IF_ADDR_UNLOCK(ifp);
1591 IFNET_RUNLOCK_NOSLEEP();
1596 ifa_ifwithaddr(struct sockaddr *addr)
1599 return (ifa_ifwithaddr_internal(addr, 1));
1603 ifa_ifwithaddr_check(struct sockaddr *addr)
1606 return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1610 * Locate an interface based on the broadcast address.
1614 ifa_ifwithbroadaddr(struct sockaddr *addr)
1619 IFNET_RLOCK_NOSLEEP();
1620 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1622 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1623 if (ifa->ifa_addr->sa_family != addr->sa_family)
1625 if ((ifp->if_flags & IFF_BROADCAST) &&
1626 ifa->ifa_broadaddr &&
1627 ifa->ifa_broadaddr->sa_len != 0 &&
1628 sa_equal(ifa->ifa_broadaddr, addr)) {
1630 IF_ADDR_UNLOCK(ifp);
1634 IF_ADDR_UNLOCK(ifp);
1638 IFNET_RUNLOCK_NOSLEEP();
1643 * Locate the point to point interface with a given destination address.
1647 ifa_ifwithdstaddr(struct sockaddr *addr)
1652 IFNET_RLOCK_NOSLEEP();
1653 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1654 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1657 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1658 if (ifa->ifa_addr->sa_family != addr->sa_family)
1660 if (ifa->ifa_dstaddr != NULL &&
1661 sa_equal(addr, ifa->ifa_dstaddr)) {
1663 IF_ADDR_UNLOCK(ifp);
1667 IF_ADDR_UNLOCK(ifp);
1671 IFNET_RUNLOCK_NOSLEEP();
1676 * Find an interface on a specific network. If many, choice
1677 * is most specific found.
1680 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp)
1684 struct ifaddr *ifa_maybe = NULL;
1685 u_int af = addr->sa_family;
1686 char *addr_data = addr->sa_data, *cplim;
1689 * AF_LINK addresses can be looked up directly by their index number,
1690 * so do that if we can.
1692 if (af == AF_LINK) {
1693 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1694 if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1695 return (ifaddr_byindex(sdl->sdl_index));
1699 * Scan though each interface, looking for ones that have addresses
1700 * in this address family. Maintain a reference on ifa_maybe once
1701 * we find one, as we release the IF_ADDR_LOCK() that kept it stable
1702 * when we move onto the next interface.
1704 IFNET_RLOCK_NOSLEEP();
1705 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1707 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1708 char *cp, *cp2, *cp3;
1710 if (ifa->ifa_addr->sa_family != af)
1712 if (af == AF_INET &&
1713 ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1715 * This is a bit broken as it doesn't
1716 * take into account that the remote end may
1717 * be a single node in the network we are
1719 * The trouble is that we don't know the
1720 * netmask for the remote end.
1722 if (ifa->ifa_dstaddr != NULL &&
1723 sa_equal(addr, ifa->ifa_dstaddr)) {
1725 IF_ADDR_UNLOCK(ifp);
1730 * if we have a special address handler,
1731 * then use it instead of the generic one.
1733 if (ifa->ifa_claim_addr) {
1734 if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1736 IF_ADDR_UNLOCK(ifp);
1743 * Scan all the bits in the ifa's address.
1744 * If a bit dissagrees with what we are
1745 * looking for, mask it with the netmask
1746 * to see if it really matters.
1747 * (A byte at a time)
1749 if (ifa->ifa_netmask == 0)
1752 cp2 = ifa->ifa_addr->sa_data;
1753 cp3 = ifa->ifa_netmask->sa_data;
1754 cplim = ifa->ifa_netmask->sa_len
1755 + (char *)ifa->ifa_netmask;
1757 if ((*cp++ ^ *cp2++) & *cp3++)
1758 goto next; /* next address! */
1760 * If the netmask of what we just found
1761 * is more specific than what we had before
1762 * (if we had one) then remember the new one
1763 * before continuing to search
1764 * for an even better one.
1766 if (ifa_maybe == NULL ||
1767 rn_refines((caddr_t)ifa->ifa_netmask,
1768 (caddr_t)ifa_maybe->ifa_netmask)) {
1769 if (ifa_maybe != NULL)
1770 ifa_free(ifa_maybe);
1776 IF_ADDR_UNLOCK(ifp);
1781 IFNET_RUNLOCK_NOSLEEP();
1782 if (ifa_maybe != NULL)
1783 ifa_free(ifa_maybe);
1788 * Find an interface address specific to an interface best matching
1792 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1795 char *cp, *cp2, *cp3;
1797 struct ifaddr *ifa_maybe = NULL;
1798 u_int af = addr->sa_family;
1803 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1804 if (ifa->ifa_addr->sa_family != af)
1806 if (ifa_maybe == NULL)
1808 if (ifa->ifa_netmask == 0) {
1809 if (sa_equal(addr, ifa->ifa_addr) ||
1810 (ifa->ifa_dstaddr &&
1811 sa_equal(addr, ifa->ifa_dstaddr)))
1815 if (ifp->if_flags & IFF_POINTOPOINT) {
1816 if (sa_equal(addr, ifa->ifa_dstaddr))
1820 cp2 = ifa->ifa_addr->sa_data;
1821 cp3 = ifa->ifa_netmask->sa_data;
1822 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1823 for (; cp3 < cplim; cp3++)
1824 if ((*cp++ ^ *cp2++) & *cp3)
1834 IF_ADDR_UNLOCK(ifp);
1838 #include <net/if_llatbl.h>
1841 * Default action when installing a route with a Link Level gateway.
1842 * Lookup an appropriate real ifa to point to.
1843 * This should be moved to /sys/net/link.c eventually.
1846 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1848 struct ifaddr *ifa, *oifa;
1849 struct sockaddr *dst;
1854 if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1855 ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1857 ifa = ifaof_ifpforaddr(dst, ifp);
1862 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1863 ifa->ifa_rtrequest(cmd, rt, info);
1868 * Mark an interface down and notify protocols of
1870 * NOTE: must be called at splnet or eqivalent.
1873 if_unroute(struct ifnet *ifp, int flag, int fam)
1877 KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1879 ifp->if_flags &= ~flag;
1880 getmicrotime(&ifp->if_lastchange);
1881 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1882 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1883 pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1884 ifp->if_qflush(ifp);
1887 (*carp_linkstate_p)(ifp);
1892 * Mark an interface up and notify protocols of
1894 * NOTE: must be called at splnet or eqivalent.
1897 if_route(struct ifnet *ifp, int flag, int fam)
1901 KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1903 ifp->if_flags |= flag;
1904 getmicrotime(&ifp->if_lastchange);
1905 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1906 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1907 pfctlinput(PRC_IFUP, ifa->ifa_addr);
1909 (*carp_linkstate_p)(ifp);
1916 void (*vlan_link_state_p)(struct ifnet *, int); /* XXX: private from if_vlan */
1917 void (*vlan_trunk_cap_p)(struct ifnet *); /* XXX: private from if_vlan */
1920 * Handle a change in the interface link state. To avoid LORs
1921 * between driver lock and upper layer locks, as well as possible
1922 * recursions, we post event to taskqueue, and all job
1923 * is done in static do_link_state_change().
1926 if_link_state_change(struct ifnet *ifp, int link_state)
1928 /* Return if state hasn't changed. */
1929 if (ifp->if_link_state == link_state)
1932 ifp->if_link_state = link_state;
1934 taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1938 do_link_state_change(void *arg, int pending)
1940 struct ifnet *ifp = (struct ifnet *)arg;
1941 int link_state = ifp->if_link_state;
1942 CURVNET_SET(ifp->if_vnet);
1944 /* Notify that the link state has changed. */
1946 if (ifp->if_vlantrunk != NULL)
1947 (*vlan_link_state_p)(ifp, 0);
1949 if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1950 IFP2AC(ifp)->ac_netgraph != NULL)
1951 (*ng_ether_link_state_p)(ifp, link_state);
1953 (*carp_linkstate_p)(ifp);
1954 if (ifp->if_bridge) {
1955 KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1956 (*bstp_linkstate_p)(ifp, link_state);
1959 KASSERT(lagg_linkstate_p != NULL,("if_lagg not loaded!"));
1960 (*lagg_linkstate_p)(ifp, link_state);
1963 if (IS_DEFAULT_VNET(curvnet))
1964 devctl_notify("IFNET", ifp->if_xname,
1965 (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
1968 if_printf(ifp, "%d link states coalesced\n", pending);
1969 if (log_link_state_change)
1970 log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1971 (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1976 * Mark an interface down and notify protocols of
1978 * NOTE: must be called at splnet or eqivalent.
1981 if_down(struct ifnet *ifp)
1984 if_unroute(ifp, IFF_UP, AF_UNSPEC);
1988 * Mark an interface up and notify protocols of
1990 * NOTE: must be called at splnet or eqivalent.
1993 if_up(struct ifnet *ifp)
1996 if_route(ifp, IFF_UP, AF_UNSPEC);
2000 * Flush an interface queue.
2003 if_qflush(struct ifnet *ifp)
2011 if (ALTQ_IS_ENABLED(ifq))
2015 while ((m = n) != 0) {
2026 * Handle interface watchdog timer routines. Called
2027 * from softclock, we decrement timers (if set) and
2028 * call the appropriate interface routine on expiration.
2030 * XXXRW: Note that because timeouts run with Giant, if_watchdog() is called
2034 if_slowtimo(void *arg)
2036 VNET_ITERATOR_DECL(vnet_iter);
2040 VNET_LIST_RLOCK_NOSLEEP();
2041 IFNET_RLOCK_NOSLEEP();
2042 VNET_FOREACH(vnet_iter) {
2043 CURVNET_SET(vnet_iter);
2044 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2045 if (ifp->if_timer == 0 || --ifp->if_timer)
2047 if (ifp->if_watchdog)
2048 (*ifp->if_watchdog)(ifp);
2052 IFNET_RUNLOCK_NOSLEEP();
2053 VNET_LIST_RUNLOCK_NOSLEEP();
2055 timeout(if_slowtimo, (void *)0, hz / IFNET_SLOWHZ);
2059 * Map interface name to interface structure pointer, with or without
2060 * returning a reference.
2063 ifunit_ref(const char *name)
2067 IFNET_RLOCK_NOSLEEP();
2068 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2069 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2070 !(ifp->if_flags & IFF_DYING))
2075 IFNET_RUNLOCK_NOSLEEP();
2080 ifunit(const char *name)
2084 IFNET_RLOCK_NOSLEEP();
2085 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2086 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2089 IFNET_RUNLOCK_NOSLEEP();
2094 * Hardware specific interface ioctls.
2097 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2102 int new_flags, temp_flags;
2103 size_t namelen, onamelen;
2105 char *descrbuf, *odescrbuf;
2106 char new_name[IFNAMSIZ];
2108 struct sockaddr_dl *sdl;
2110 ifr = (struct ifreq *)data;
2113 ifr->ifr_index = ifp->if_index;
2117 temp_flags = ifp->if_flags | ifp->if_drv_flags;
2118 ifr->ifr_flags = temp_flags & 0xffff;
2119 ifr->ifr_flagshigh = temp_flags >> 16;
2123 ifr->ifr_reqcap = ifp->if_capabilities;
2124 ifr->ifr_curcap = ifp->if_capenable;
2129 error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2134 ifr->ifr_metric = ifp->if_metric;
2138 ifr->ifr_mtu = ifp->if_mtu;
2142 ifr->ifr_phys = ifp->if_physical;
2147 sx_slock(&ifdescr_sx);
2148 if (ifp->if_description == NULL)
2151 /* space for terminating nul */
2152 descrlen = strlen(ifp->if_description) + 1;
2153 if (ifr->ifr_buffer.length < descrlen)
2154 ifr->ifr_buffer.buffer = NULL;
2156 error = copyout(ifp->if_description,
2157 ifr->ifr_buffer.buffer, descrlen);
2158 ifr->ifr_buffer.length = descrlen;
2160 sx_sunlock(&ifdescr_sx);
2164 error = priv_check(td, PRIV_NET_SETIFDESCR);
2169 * Copy only (length-1) bytes to make sure that
2170 * if_description is always nul terminated. The
2171 * length parameter is supposed to count the
2172 * terminating nul in.
2174 if (ifr->ifr_buffer.length > ifdescr_maxlen)
2175 return (ENAMETOOLONG);
2176 else if (ifr->ifr_buffer.length == 0)
2179 descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2181 error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2182 ifr->ifr_buffer.length - 1);
2184 free(descrbuf, M_IFDESCR);
2189 sx_xlock(&ifdescr_sx);
2190 odescrbuf = ifp->if_description;
2191 ifp->if_description = descrbuf;
2192 sx_xunlock(&ifdescr_sx);
2194 getmicrotime(&ifp->if_lastchange);
2195 free(odescrbuf, M_IFDESCR);
2199 error = priv_check(td, PRIV_NET_SETIFFLAGS);
2203 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2204 * check, so we don't need special handling here yet.
2206 new_flags = (ifr->ifr_flags & 0xffff) |
2207 (ifr->ifr_flagshigh << 16);
2208 if (ifp->if_flags & IFF_SMART) {
2209 /* Smart drivers twiddle their own routes */
2210 } else if (ifp->if_flags & IFF_UP &&
2211 (new_flags & IFF_UP) == 0) {
2215 } else if (new_flags & IFF_UP &&
2216 (ifp->if_flags & IFF_UP) == 0) {
2221 /* See if permanently promiscuous mode bit is about to flip */
2222 if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2223 if (new_flags & IFF_PPROMISC)
2224 ifp->if_flags |= IFF_PROMISC;
2225 else if (ifp->if_pcount == 0)
2226 ifp->if_flags &= ~IFF_PROMISC;
2227 log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2229 (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
2231 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2232 (new_flags &~ IFF_CANTCHANGE);
2233 if (ifp->if_ioctl) {
2234 (void) (*ifp->if_ioctl)(ifp, cmd, data);
2236 getmicrotime(&ifp->if_lastchange);
2240 error = priv_check(td, PRIV_NET_SETIFCAP);
2243 if (ifp->if_ioctl == NULL)
2244 return (EOPNOTSUPP);
2245 if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2247 error = (*ifp->if_ioctl)(ifp, cmd, data);
2249 getmicrotime(&ifp->if_lastchange);
2254 error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2259 error = priv_check(td, PRIV_NET_SETIFNAME);
2262 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2265 if (new_name[0] == '\0')
2267 if (ifunit(new_name) != NULL)
2271 * XXX: Locking. Nothing else seems to lock if_flags,
2272 * and there are numerous other races with the
2273 * ifunit() checks not being atomic with namespace
2274 * changes (renames, vmoves, if_attach, etc).
2276 ifp->if_flags |= IFF_RENAMING;
2278 /* Announce the departure of the interface. */
2279 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2280 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2282 log(LOG_INFO, "%s: changing name to '%s'\n",
2283 ifp->if_xname, new_name);
2285 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2288 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2289 namelen = strlen(new_name);
2290 onamelen = sdl->sdl_nlen;
2292 * Move the address if needed. This is safe because we
2293 * allocate space for a name of length IFNAMSIZ when we
2294 * create this in if_attach().
2296 if (namelen != onamelen) {
2297 bcopy(sdl->sdl_data + onamelen,
2298 sdl->sdl_data + namelen, sdl->sdl_alen);
2300 bcopy(new_name, sdl->sdl_data, namelen);
2301 sdl->sdl_nlen = namelen;
2302 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2303 bzero(sdl->sdl_data, onamelen);
2304 while (namelen != 0)
2305 sdl->sdl_data[--namelen] = 0xff;
2308 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2309 /* Announce the return of the interface. */
2310 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2312 ifp->if_flags &= ~IFF_RENAMING;
2317 error = priv_check(td, PRIV_NET_SETIFVNET);
2320 error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2325 error = priv_check(td, PRIV_NET_SETIFMETRIC);
2328 ifp->if_metric = ifr->ifr_metric;
2329 getmicrotime(&ifp->if_lastchange);
2333 error = priv_check(td, PRIV_NET_SETIFPHYS);
2336 if (ifp->if_ioctl == NULL)
2337 return (EOPNOTSUPP);
2338 error = (*ifp->if_ioctl)(ifp, cmd, data);
2340 getmicrotime(&ifp->if_lastchange);
2345 u_long oldmtu = ifp->if_mtu;
2347 error = priv_check(td, PRIV_NET_SETIFMTU);
2350 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2352 if (ifp->if_ioctl == NULL)
2353 return (EOPNOTSUPP);
2354 error = (*ifp->if_ioctl)(ifp, cmd, data);
2356 getmicrotime(&ifp->if_lastchange);
2360 * If the link MTU changed, do network layer specific procedure.
2362 if (ifp->if_mtu != oldmtu) {
2372 if (cmd == SIOCADDMULTI)
2373 error = priv_check(td, PRIV_NET_ADDMULTI);
2375 error = priv_check(td, PRIV_NET_DELMULTI);
2379 /* Don't allow group membership on non-multicast interfaces. */
2380 if ((ifp->if_flags & IFF_MULTICAST) == 0)
2381 return (EOPNOTSUPP);
2383 /* Don't let users screw up protocols' entries. */
2384 if (ifr->ifr_addr.sa_family != AF_LINK)
2387 if (cmd == SIOCADDMULTI) {
2388 struct ifmultiaddr *ifma;
2391 * Userland is only permitted to join groups once
2392 * via the if_addmulti() KPI, because it cannot hold
2393 * struct ifmultiaddr * between calls. It may also
2394 * lose a race while we check if the membership
2398 ifma = if_findmulti(ifp, &ifr->ifr_addr);
2399 IF_ADDR_UNLOCK(ifp);
2403 error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2405 error = if_delmulti(ifp, &ifr->ifr_addr);
2408 getmicrotime(&ifp->if_lastchange);
2411 case SIOCSIFPHYADDR:
2412 case SIOCDIFPHYADDR:
2414 case SIOCSIFPHYADDR_IN6:
2416 case SIOCSLIFPHYADDR:
2418 case SIOCSIFGENERIC:
2419 error = priv_check(td, PRIV_NET_HWIOCTL);
2422 if (ifp->if_ioctl == NULL)
2423 return (EOPNOTSUPP);
2424 error = (*ifp->if_ioctl)(ifp, cmd, data);
2426 getmicrotime(&ifp->if_lastchange);
2430 ifs = (struct ifstat *)data;
2431 ifs->ascii[0] = '\0';
2433 case SIOCGIFPSRCADDR:
2434 case SIOCGIFPDSTADDR:
2435 case SIOCGLIFPHYADDR:
2437 case SIOCGIFGENERIC:
2438 if (ifp->if_ioctl == NULL)
2439 return (EOPNOTSUPP);
2440 error = (*ifp->if_ioctl)(ifp, cmd, data);
2444 error = priv_check(td, PRIV_NET_SETLLADDR);
2447 error = if_setlladdr(ifp,
2448 ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2449 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2454 struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2456 error = priv_check(td, PRIV_NET_ADDIFGROUP);
2459 if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2465 if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2471 struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2473 error = priv_check(td, PRIV_NET_DELIFGROUP);
2476 if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2488 #ifdef COMPAT_FREEBSD32
2496 #define SIOCGIFCONF32 _IOWR('i', 36, struct ifconf32)
2503 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2510 CURVNET_SET(so->so_vnet);
2514 error = ifconf(cmd, data);
2518 #ifdef COMPAT_FREEBSD32
2521 struct ifconf32 *ifc32;
2524 ifc32 = (struct ifconf32 *)data;
2525 ifc.ifc_len = ifc32->ifc_len;
2526 ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2528 error = ifconf(SIOCGIFCONF, (void *)&ifc);
2531 ifc32->ifc_len = ifc.ifc_len;
2536 ifr = (struct ifreq *)data;
2541 error = priv_check(td, PRIV_NET_SETIFVNET);
2543 error = if_vmove_reclaim(td, ifr->ifr_name,
2550 error = priv_check(td, PRIV_NET_IFCREATE);
2552 error = if_clone_create(ifr->ifr_name,
2553 sizeof(ifr->ifr_name),
2554 cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
2558 error = priv_check(td, PRIV_NET_IFDESTROY);
2560 error = if_clone_destroy(ifr->ifr_name);
2564 case SIOCIFGCLONERS:
2565 error = if_clone_list((struct if_clonereq *)data);
2569 error = if_getgroupmembers((struct ifgroupreq *)data);
2574 ifp = ifunit_ref(ifr->ifr_name);
2580 error = ifhwioctl(cmd, ifp, data, td);
2581 if (error != ENOIOCTL) {
2587 oif_flags = ifp->if_flags;
2588 if (so->so_proto == NULL) {
2591 return (EOPNOTSUPP);
2594 error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
2597 if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL)
2598 error = (*ifp->if_ioctl)(ifp, cmd, data);
2605 case SIOCSIFDSTADDR:
2607 case SIOCSIFBRDADDR:
2608 case SIOCSIFNETMASK:
2609 #if BYTE_ORDER != BIG_ENDIAN
2610 if (ifr->ifr_addr.sa_family == 0 &&
2611 ifr->ifr_addr.sa_len < 16) {
2612 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2613 ifr->ifr_addr.sa_len = 16;
2616 if (ifr->ifr_addr.sa_len == 0)
2617 ifr->ifr_addr.sa_len = 16;
2625 case OSIOCGIFDSTADDR:
2626 cmd = SIOCGIFDSTADDR;
2629 case OSIOCGIFBRDADDR:
2630 cmd = SIOCGIFBRDADDR;
2633 case OSIOCGIFNETMASK:
2634 cmd = SIOCGIFNETMASK;
2636 error = ((*so->so_proto->pr_usrreqs->pru_control)(so,
2640 if (error == EOPNOTSUPP && ifp != NULL &&
2641 ifp->if_ioctl != NULL)
2642 error = (*ifp->if_ioctl)(ifp, cmd, data);
2646 case OSIOCGIFDSTADDR:
2647 case OSIOCGIFBRDADDR:
2648 case OSIOCGIFNETMASK:
2649 *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2653 #endif /* COMPAT_43 */
2655 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2657 if (ifp->if_flags & IFF_UP) {
2670 * The code common to handling reference counted flags,
2671 * e.g., in ifpromisc() and if_allmulti().
2672 * The "pflag" argument can specify a permanent mode flag to check,
2673 * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2675 * Only to be used on stack-owned flags, not driver-owned flags.
2678 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2682 int oldflags, oldcount;
2684 /* Sanity checks to catch programming errors */
2685 KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2686 ("%s: setting driver-owned flag %d", __func__, flag));
2689 KASSERT(*refcount >= 0,
2690 ("%s: increment negative refcount %d for flag %d",
2691 __func__, *refcount, flag));
2693 KASSERT(*refcount > 0,
2694 ("%s: decrement non-positive refcount %d for flag %d",
2695 __func__, *refcount, flag));
2697 /* In case this mode is permanent, just touch refcount */
2698 if (ifp->if_flags & pflag) {
2699 *refcount += onswitch ? 1 : -1;
2703 /* Save ifnet parameters for if_ioctl() may fail */
2704 oldcount = *refcount;
2705 oldflags = ifp->if_flags;
2708 * See if we aren't the only and touching refcount is enough.
2709 * Actually toggle interface flag if we are the first or last.
2714 ifp->if_flags |= flag;
2718 ifp->if_flags &= ~flag;
2721 /* Call down the driver since we've changed interface flags */
2722 if (ifp->if_ioctl == NULL) {
2726 ifr.ifr_flags = ifp->if_flags & 0xffff;
2727 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2728 error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2731 /* Notify userland that interface flags have changed */
2736 /* Recover after driver error */
2737 *refcount = oldcount;
2738 ifp->if_flags = oldflags;
2743 * Set/clear promiscuous mode on interface ifp based on the truth value
2744 * of pswitch. The calls are reference counted so that only the first
2745 * "on" request actually has an effect, as does the final "off" request.
2746 * Results are undefined if the "off" and "on" requests are not matched.
2749 ifpromisc(struct ifnet *ifp, int pswitch)
2752 int oldflags = ifp->if_flags;
2754 error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2755 &ifp->if_pcount, pswitch);
2756 /* If promiscuous mode status has changed, log a message */
2757 if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2758 log(LOG_INFO, "%s: promiscuous mode %s\n",
2760 (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2765 * Return interface configuration
2766 * of system. List may be used
2767 * in later ioctl's (above) to get
2768 * other information.
2772 ifconf(u_long cmd, caddr_t data)
2774 struct ifconf *ifc = (struct ifconf *)data;
2779 int error, full = 0, valid_len, max_len;
2781 /* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2782 max_len = MAXPHYS - 1;
2784 /* Prevent hostile input from being able to crash the system */
2785 if (ifc->ifc_len <= 0)
2789 if (ifc->ifc_len <= max_len) {
2790 max_len = ifc->ifc_len;
2793 sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2798 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2802 * Zero the ifr_name buffer to make sure we don't
2803 * disclose the contents of the stack.
2805 memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2807 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2808 >= sizeof(ifr.ifr_name)) {
2811 return (ENAMETOOLONG);
2816 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2817 struct sockaddr *sa = ifa->ifa_addr;
2819 if (prison_if(curthread->td_ucred, sa) != 0)
2823 if (cmd == OSIOCGIFCONF) {
2824 struct osockaddr *osa =
2825 (struct osockaddr *)&ifr.ifr_addr;
2827 osa->sa_family = sa->sa_family;
2828 sbuf_bcat(sb, &ifr, sizeof(ifr));
2829 max_len += sizeof(ifr);
2832 if (sa->sa_len <= sizeof(*sa)) {
2834 sbuf_bcat(sb, &ifr, sizeof(ifr));
2835 max_len += sizeof(ifr);
2838 offsetof(struct ifreq, ifr_addr));
2839 max_len += offsetof(struct ifreq, ifr_addr);
2840 sbuf_bcat(sb, sa, sa->sa_len);
2841 max_len += sa->sa_len;
2844 if (!sbuf_overflowed(sb))
2845 valid_len = sbuf_len(sb);
2847 IF_ADDR_UNLOCK(ifp);
2849 bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2850 sbuf_bcat(sb, &ifr, sizeof(ifr));
2851 max_len += sizeof(ifr);
2853 if (!sbuf_overflowed(sb))
2854 valid_len = sbuf_len(sb);
2860 * If we didn't allocate enough space (uncommon), try again. If
2861 * we have already allocated as much space as we are allowed,
2862 * return what we've got.
2864 if (valid_len != max_len && !full) {
2869 ifc->ifc_len = valid_len;
2871 error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2877 * Just like ifpromisc(), but for all-multicast-reception mode.
2880 if_allmulti(struct ifnet *ifp, int onswitch)
2883 return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2886 struct ifmultiaddr *
2887 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2889 struct ifmultiaddr *ifma;
2891 IF_ADDR_LOCK_ASSERT(ifp);
2893 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2894 if (sa->sa_family == AF_LINK) {
2895 if (sa_dl_equal(ifma->ifma_addr, sa))
2898 if (sa_equal(ifma->ifma_addr, sa))
2907 * Allocate a new ifmultiaddr and initialize based on passed arguments. We
2908 * make copies of passed sockaddrs. The ifmultiaddr will not be added to
2909 * the ifnet multicast address list here, so the caller must do that and
2910 * other setup work (such as notifying the device driver). The reference
2911 * count is initialized to 1.
2913 static struct ifmultiaddr *
2914 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2917 struct ifmultiaddr *ifma;
2918 struct sockaddr *dupsa;
2920 ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2925 dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2926 if (dupsa == NULL) {
2927 free(ifma, M_IFMADDR);
2930 bcopy(sa, dupsa, sa->sa_len);
2931 ifma->ifma_addr = dupsa;
2933 ifma->ifma_ifp = ifp;
2934 ifma->ifma_refcount = 1;
2935 ifma->ifma_protospec = NULL;
2938 ifma->ifma_lladdr = NULL;
2942 dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
2943 if (dupsa == NULL) {
2944 free(ifma->ifma_addr, M_IFMADDR);
2945 free(ifma, M_IFMADDR);
2948 bcopy(llsa, dupsa, llsa->sa_len);
2949 ifma->ifma_lladdr = dupsa;
2955 * if_freemulti: free ifmultiaddr structure and possibly attached related
2956 * addresses. The caller is responsible for implementing reference
2957 * counting, notifying the driver, handling routing messages, and releasing
2958 * any dependent link layer state.
2961 if_freemulti(struct ifmultiaddr *ifma)
2964 KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2965 ifma->ifma_refcount));
2966 KASSERT(ifma->ifma_protospec == NULL,
2967 ("if_freemulti: protospec not NULL"));
2969 if (ifma->ifma_lladdr != NULL)
2970 free(ifma->ifma_lladdr, M_IFMADDR);
2971 free(ifma->ifma_addr, M_IFMADDR);
2972 free(ifma, M_IFMADDR);
2976 * Register an additional multicast address with a network interface.
2978 * - If the address is already present, bump the reference count on the
2979 * address and return.
2980 * - If the address is not link-layer, look up a link layer address.
2981 * - Allocate address structures for one or both addresses, and attach to the
2982 * multicast address list on the interface. If automatically adding a link
2983 * layer address, the protocol address will own a reference to the link
2984 * layer address, to be freed when it is freed.
2985 * - Notify the network device driver of an addition to the multicast address
2988 * 'sa' points to caller-owned memory with the desired multicast address.
2990 * 'retifma' will be used to return a pointer to the resulting multicast
2991 * address reference, if desired.
2994 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2995 struct ifmultiaddr **retifma)
2997 struct ifmultiaddr *ifma, *ll_ifma;
2998 struct sockaddr *llsa;
3002 * If the address is already present, return a new reference to it;
3003 * otherwise, allocate storage and set up a new address.
3006 ifma = if_findmulti(ifp, sa);
3008 ifma->ifma_refcount++;
3009 if (retifma != NULL)
3011 IF_ADDR_UNLOCK(ifp);
3016 * The address isn't already present; resolve the protocol address
3017 * into a link layer address, and then look that up, bump its
3018 * refcount or allocate an ifma for that also. If 'llsa' was
3019 * returned, we will need to free it later.
3023 if (ifp->if_resolvemulti != NULL) {
3024 error = ifp->if_resolvemulti(ifp, &llsa, sa);
3030 * Allocate the new address. Don't hook it up yet, as we may also
3031 * need to allocate a link layer multicast address.
3033 ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
3040 * If a link layer address is found, we'll need to see if it's
3041 * already present in the address list, or allocate is as well.
3042 * When this block finishes, the link layer address will be on the
3046 ll_ifma = if_findmulti(ifp, llsa);
3047 if (ll_ifma == NULL) {
3048 ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
3049 if (ll_ifma == NULL) {
3050 --ifma->ifma_refcount;
3055 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
3058 ll_ifma->ifma_refcount++;
3059 ifma->ifma_llifma = ll_ifma;
3063 * We now have a new multicast address, ifma, and possibly a new or
3064 * referenced link layer address. Add the primary address to the
3065 * ifnet address list.
3067 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3069 if (retifma != NULL)
3073 * Must generate the message while holding the lock so that 'ifma'
3074 * pointer is still valid.
3076 rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3077 IF_ADDR_UNLOCK(ifp);
3080 * We are certain we have added something, so call down to the
3081 * interface to let them know about it.
3083 if (ifp->if_ioctl != NULL) {
3084 (void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3088 free(llsa, M_IFMADDR);
3094 free(llsa, M_IFMADDR);
3097 IF_ADDR_UNLOCK(ifp);
3102 * Delete a multicast group membership by network-layer group address.
3104 * Returns ENOENT if the entry could not be found. If ifp no longer
3105 * exists, results are undefined. This entry point should only be used
3106 * from subsystems which do appropriate locking to hold ifp for the
3107 * duration of the call.
3108 * Network-layer protocol domains must use if_delmulti_ifma().
3111 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3113 struct ifmultiaddr *ifma;
3118 IFNET_RLOCK_NOSLEEP();
3119 TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3124 IFNET_RUNLOCK_NOSLEEP();
3126 KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3133 ifma = if_findmulti(ifp, sa);
3135 lastref = if_delmulti_locked(ifp, ifma, 0);
3136 IF_ADDR_UNLOCK(ifp);
3141 if (lastref && ifp->if_ioctl != NULL) {
3142 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3149 * Delete all multicast group membership for an interface.
3150 * Should be used to quickly flush all multicast filters.
3153 if_delallmulti(struct ifnet *ifp)
3155 struct ifmultiaddr *ifma;
3156 struct ifmultiaddr *next;
3159 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3160 if_delmulti_locked(ifp, ifma, 0);
3161 IF_ADDR_UNLOCK(ifp);
3165 * Delete a multicast group membership by group membership pointer.
3166 * Network-layer protocol domains must use this routine.
3168 * It is safe to call this routine if the ifp disappeared.
3171 if_delmulti_ifma(struct ifmultiaddr *ifma)
3176 ifp = ifma->ifma_ifp;
3179 printf("%s: ifma_ifp seems to be detached\n", __func__);
3183 IFNET_RLOCK_NOSLEEP();
3184 TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3188 printf("%s: ifnet %p disappeared\n", __func__, ifp);
3191 IFNET_RUNLOCK_NOSLEEP();
3195 * If and only if the ifnet instance exists: Acquire the address lock.
3200 lastref = if_delmulti_locked(ifp, ifma, 0);
3204 * If and only if the ifnet instance exists:
3205 * Release the address lock.
3206 * If the group was left: update the hardware hash filter.
3208 IF_ADDR_UNLOCK(ifp);
3209 if (lastref && ifp->if_ioctl != NULL) {
3210 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3216 * Perform deletion of network-layer and/or link-layer multicast address.
3218 * Return 0 if the reference count was decremented.
3219 * Return 1 if the final reference was released, indicating that the
3220 * hardware hash filter should be reprogrammed.
3223 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3225 struct ifmultiaddr *ll_ifma;
3227 if (ifp != NULL && ifma->ifma_ifp != NULL) {
3228 KASSERT(ifma->ifma_ifp == ifp,
3229 ("%s: inconsistent ifp %p", __func__, ifp));
3230 IF_ADDR_LOCK_ASSERT(ifp);
3233 ifp = ifma->ifma_ifp;
3236 * If the ifnet is detaching, null out references to ifnet,
3237 * so that upper protocol layers will notice, and not attempt
3238 * to obtain locks for an ifnet which no longer exists. The
3239 * routing socket announcement must happen before the ifnet
3240 * instance is detached from the system.
3244 printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3247 * ifp may already be nulled out if we are being reentered
3248 * to delete the ll_ifma.
3251 rt_newmaddrmsg(RTM_DELMADDR, ifma);
3252 ifma->ifma_ifp = NULL;
3256 if (--ifma->ifma_refcount > 0)
3260 * If this ifma is a network-layer ifma, a link-layer ifma may
3261 * have been associated with it. Release it first if so.
3263 ll_ifma = ifma->ifma_llifma;
3264 if (ll_ifma != NULL) {
3265 KASSERT(ifma->ifma_lladdr != NULL,
3266 ("%s: llifma w/o lladdr", __func__));
3268 ll_ifma->ifma_ifp = NULL; /* XXX */
3269 if (--ll_ifma->ifma_refcount == 0) {
3271 TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3274 if_freemulti(ll_ifma);
3279 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3284 * The last reference to this instance of struct ifmultiaddr
3285 * was released; the hardware should be notified of this change.
3291 * Set the link layer address on an interface.
3293 * At this time we only support certain types of interfaces,
3294 * and we don't allow the length of the address to change.
3297 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3299 struct sockaddr_dl *sdl;
3306 IF_ADDR_UNLOCK(ifp);
3310 IF_ADDR_UNLOCK(ifp);
3311 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3316 if (len != sdl->sdl_alen) { /* don't allow length to change */
3320 switch (ifp->if_type) {
3328 case IFT_IEEE8023ADLAG:
3330 bcopy(lladdr, LLADDR(sdl), len);
3339 * If the interface is already up, we need
3340 * to re-init it in order to reprogram its
3343 if ((ifp->if_flags & IFF_UP) != 0) {
3344 if (ifp->if_ioctl) {
3345 ifp->if_flags &= ~IFF_UP;
3346 ifr.ifr_flags = ifp->if_flags & 0xffff;
3347 ifr.ifr_flagshigh = ifp->if_flags >> 16;
3348 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3349 ifp->if_flags |= IFF_UP;
3350 ifr.ifr_flags = ifp->if_flags & 0xffff;
3351 ifr.ifr_flagshigh = ifp->if_flags >> 16;
3352 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3356 * Also send gratuitous ARPs to notify other nodes about
3357 * the address change.
3359 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3360 if (ifa->ifa_addr->sa_family == AF_INET)
3361 arp_ifinit(ifp, ifa);
3369 * The name argument must be a pointer to storage which will last as
3370 * long as the interface does. For physical devices, the result of
3371 * device_get_name(dev) is a good choice and for pseudo-devices a
3372 * static string works well.
3375 if_initname(struct ifnet *ifp, const char *name, int unit)
3377 ifp->if_dname = name;
3378 ifp->if_dunit = unit;
3379 if (unit != IF_DUNIT_NONE)
3380 snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3382 strlcpy(ifp->if_xname, name, IFNAMSIZ);
3386 if_printf(struct ifnet *ifp, const char * fmt, ...)
3391 retval = printf("%s: ", ifp->if_xname);
3393 retval += vprintf(fmt, ap);
3399 if_start(struct ifnet *ifp)
3402 (*(ifp)->if_start)(ifp);
3406 * Backwards compatibility interface for drivers
3407 * that have not implemented it
3410 if_transmit(struct ifnet *ifp, struct mbuf *m)
3414 IFQ_HANDOFF(ifp, m, error);
3419 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3424 if (_IF_QFULL(ifq)) {
3431 ifp->if_obytes += m->m_pkthdr.len + adjust;
3432 if (m->m_flags & (M_BCAST|M_MCAST))
3434 active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3436 _IF_ENQUEUE(ifq, m);
3438 if (ifp != NULL && !active)
3439 (*(ifp)->if_start)(ifp);
3444 if_register_com_alloc(u_char type,
3445 if_com_alloc_t *a, if_com_free_t *f)
3448 KASSERT(if_com_alloc[type] == NULL,
3449 ("if_register_com_alloc: %d already registered", type));
3450 KASSERT(if_com_free[type] == NULL,
3451 ("if_register_com_alloc: %d free already registered", type));
3453 if_com_alloc[type] = a;
3454 if_com_free[type] = f;
3458 if_deregister_com_alloc(u_char type)
3461 KASSERT(if_com_alloc[type] != NULL,
3462 ("if_deregister_com_alloc: %d not registered", type));
3463 KASSERT(if_com_free[type] != NULL,
3464 ("if_deregister_com_alloc: %d free not registered", type));
3465 if_com_alloc[type] = NULL;
3466 if_com_free[type] = NULL;