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));
391 VNET_ASSERT(TAILQ_EMPTY(&V_ifg_head));
393 free((caddr_t)V_ifindex_table, M_IFNET);
395 VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
396 vnet_if_uninit, NULL);
404 struct ifindex_entry *e;
406 IFNET_WLOCK_ASSERT();
407 oldlim = V_if_indexlim;
409 n = (oldlim << 1) * sizeof(*e);
410 e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
412 if (V_if_indexlim != oldlim) {
416 if (V_ifindex_table != NULL) {
417 memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2);
418 free((caddr_t)V_ifindex_table, M_IFNET);
425 if_check(void *dummy __unused)
429 * If at least one interface added during boot uses
430 * if_watchdog then start the timer.
432 if (slowtimo_started)
437 * Allocate a struct ifnet and an index for an interface. A layer 2
438 * common structure will also be allocated if an allocation routine is
439 * registered for the passed type.
442 if_alloc(u_char type)
447 ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
449 if (ifindex_alloc_locked(&idx) != 0) {
454 ifnet_setbyindex_locked(idx, IFNET_HOLD);
458 ifp->if_alloctype = type;
459 if (if_com_alloc[type] != NULL) {
460 ifp->if_l2com = if_com_alloc[type](type, ifp);
461 if (ifp->if_l2com == NULL) {
468 IF_ADDR_LOCK_INIT(ifp);
469 TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
470 ifp->if_afdata_initialized = 0;
471 IF_AFDATA_LOCK_INIT(ifp);
472 TAILQ_INIT(&ifp->if_addrhead);
473 TAILQ_INIT(&ifp->if_prefixhead);
474 TAILQ_INIT(&ifp->if_multiaddrs);
475 TAILQ_INIT(&ifp->if_groups);
479 ifq_init(&ifp->if_snd, ifp);
481 refcount_init(&ifp->if_refcount, 1); /* Index reference. */
482 ifnet_setbyindex(ifp->if_index, ifp);
487 * Do the actual work of freeing a struct ifnet, and layer 2 common
488 * structure. This call is made when the last reference to an
489 * interface is released.
492 if_free_internal(struct ifnet *ifp)
495 KASSERT((ifp->if_flags & IFF_DYING),
496 ("if_free_internal: interface not dying"));
498 if (if_com_free[ifp->if_alloctype] != NULL)
499 if_com_free[ifp->if_alloctype](ifp->if_l2com,
503 mac_ifnet_destroy(ifp);
505 if (ifp->if_description != NULL)
506 free(ifp->if_description, M_IFDESCR);
507 IF_AFDATA_DESTROY(ifp);
508 IF_ADDR_LOCK_DESTROY(ifp);
509 ifq_delete(&ifp->if_snd);
514 * This version should only be called by intefaces that switch their type
515 * after calling if_alloc(). if_free_type() will go away again now that we
516 * have if_alloctype to cache the original allocation type. For now, assert
517 * that they match, since we require that in practice.
520 if_free_type(struct ifnet *ifp, u_char type)
523 KASSERT(ifp->if_alloctype == type,
524 ("if_free_type: type (%d) != alloctype (%d)", type,
527 ifp->if_flags |= IFF_DYING; /* XXX: Locking */
530 KASSERT(ifp == ifnet_byindex_locked(ifp->if_index),
531 ("%s: freeing unallocated ifnet", ifp->if_xname));
533 ifindex_free_locked(ifp->if_index);
536 if (!refcount_release(&ifp->if_refcount))
538 if_free_internal(ifp);
542 * This is the normal version of if_free(), used by device drivers to free a
543 * detached network interface. The contents of if_free_type() will move into
544 * here when if_free_type() goes away.
547 if_free(struct ifnet *ifp)
550 if_free_type(ifp, ifp->if_alloctype);
554 * Interfaces to keep an ifnet type-stable despite the possibility of the
555 * driver calling if_free(). If there are additional references, we defer
556 * freeing the underlying data structure.
559 if_ref(struct ifnet *ifp)
562 /* We don't assert the ifnet list lock here, but arguably should. */
563 refcount_acquire(&ifp->if_refcount);
567 if_rele(struct ifnet *ifp)
570 if (!refcount_release(&ifp->if_refcount))
572 if_free_internal(ifp);
576 ifq_init(struct ifaltq *ifq, struct ifnet *ifp)
579 mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
581 if (ifq->ifq_maxlen == 0)
582 ifq->ifq_maxlen = ifqmaxlen;
585 ifq->altq_disc = NULL;
586 ifq->altq_flags &= ALTQF_CANTCHANGE;
587 ifq->altq_tbr = NULL;
592 ifq_delete(struct ifaltq *ifq)
594 mtx_destroy(&ifq->ifq_mtx);
598 * Perform generic interface initalization tasks and attach the interface
599 * to the list of "active" interfaces. If vmove flag is set on entry
600 * to if_attach_internal(), perform only a limited subset of initialization
601 * tasks, given that we are moving from one vnet to another an ifnet which
602 * has already been fully initialized.
605 * - The decision to return void and thus require this function to
606 * succeed is questionable.
607 * - We should probably do more sanity checking. For instance we don't
608 * do anything to insure if_xname is unique or non-empty.
611 if_attach(struct ifnet *ifp)
614 if_attach_internal(ifp, 0);
618 if_attach_internal(struct ifnet *ifp, int vmove)
620 unsigned socksize, ifasize;
621 int namelen, masklen;
622 struct sockaddr_dl *sdl;
625 if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
626 panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
630 ifp->if_vnet = curvnet;
631 if (ifp->if_home_vnet == NULL)
632 ifp->if_home_vnet = curvnet;
635 if_addgroup(ifp, IFG_ALL);
637 getmicrotime(&ifp->if_lastchange);
638 ifp->if_data.ifi_epoch = time_uptime;
639 ifp->if_data.ifi_datalen = sizeof(struct if_data);
641 KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
642 (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
643 ("transmit and qflush must both either be set or both be NULL"));
644 if (ifp->if_transmit == NULL) {
645 ifp->if_transmit = if_transmit;
646 ifp->if_qflush = if_qflush;
651 mac_ifnet_create(ifp);
655 * Create a Link Level name for this device.
657 namelen = strlen(ifp->if_xname);
659 * Always save enough space for any possiable name so we
660 * can do a rename in place later.
662 masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
663 socksize = masklen + ifp->if_addrlen;
664 if (socksize < sizeof(*sdl))
665 socksize = sizeof(*sdl);
666 socksize = roundup2(socksize, sizeof(long));
667 ifasize = sizeof(*ifa) + 2 * socksize;
668 ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
670 sdl = (struct sockaddr_dl *)(ifa + 1);
671 sdl->sdl_len = socksize;
672 sdl->sdl_family = AF_LINK;
673 bcopy(ifp->if_xname, sdl->sdl_data, namelen);
674 sdl->sdl_nlen = namelen;
675 sdl->sdl_index = ifp->if_index;
676 sdl->sdl_type = ifp->if_type;
679 ifa->ifa_rtrequest = link_rtrequest;
680 ifa->ifa_addr = (struct sockaddr *)sdl;
681 sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
682 ifa->ifa_netmask = (struct sockaddr *)sdl;
683 sdl->sdl_len = masklen;
685 sdl->sdl_data[--namelen] = 0xff;
686 TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
687 /* Reliably crash if used uninitialized. */
688 ifp->if_broadcastaddr = NULL;
693 * Update the interface index in the link layer address
696 for (ifa = ifp->if_addr; ifa != NULL;
697 ifa = TAILQ_NEXT(ifa, ifa_link)) {
698 if (ifa->ifa_addr->sa_family == AF_LINK) {
699 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
700 sdl->sdl_index = ifp->if_index;
707 TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
709 curvnet->vnet_ifcnt++;
713 if (domain_init_status >= 2)
714 if_attachdomain1(ifp);
716 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
717 if (IS_DEFAULT_VNET(curvnet))
718 devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
720 /* Announce the interface. */
721 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
723 if (!vmove && ifp->if_watchdog != NULL) {
725 "WARNING: using obsoleted if_watchdog interface\n");
728 * Note that we need if_slowtimo(). If this happens after
729 * boot, then call if_slowtimo() directly.
731 if (atomic_cmpset_int(&slowtimo_started, 0, 1) && !cold)
737 if_attachdomain(void *dummy)
743 TAILQ_FOREACH(ifp, &V_ifnet, if_link)
744 if_attachdomain1(ifp);
747 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
748 if_attachdomain, NULL);
751 if_attachdomain1(struct ifnet *ifp)
759 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
760 * cannot lock ifp->if_afdata initialization, entirely.
762 if (IF_AFDATA_TRYLOCK(ifp) == 0) {
766 if (ifp->if_afdata_initialized >= domain_init_status) {
767 IF_AFDATA_UNLOCK(ifp);
769 printf("if_attachdomain called more than once on %s\n",
773 ifp->if_afdata_initialized = domain_init_status;
774 IF_AFDATA_UNLOCK(ifp);
776 /* address family dependent data region */
777 bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
778 for (dp = domains; dp; dp = dp->dom_next) {
779 if (dp->dom_ifattach)
780 ifp->if_afdata[dp->dom_family] =
781 (*dp->dom_ifattach)(ifp);
788 * Remove any unicast or broadcast network addresses from an interface.
791 if_purgeaddrs(struct ifnet *ifp)
793 struct ifaddr *ifa, *next;
795 TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
796 if (ifa->ifa_addr->sa_family == AF_LINK)
799 /* XXX: Ugly!! ad hoc just for INET */
800 if (ifa->ifa_addr->sa_family == AF_INET) {
801 struct ifaliasreq ifr;
803 bzero(&ifr, sizeof(ifr));
804 ifr.ifra_addr = *ifa->ifa_addr;
805 if (ifa->ifa_dstaddr)
806 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
807 if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
813 if (ifa->ifa_addr->sa_family == AF_INET6) {
815 /* ifp_addrhead is already updated */
819 TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
825 * Remove any multicast network addresses from an interface when an ifnet
829 if_purgemaddrs(struct ifnet *ifp)
831 struct ifmultiaddr *ifma;
832 struct ifmultiaddr *next;
835 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
836 if_delmulti_locked(ifp, ifma, 1);
841 * Detach an interface, removing it from the list of "active" interfaces.
842 * If vmove flag is set on entry to if_detach_internal(), perform only a
843 * limited subset of cleanup tasks, given that we are moving an ifnet from
844 * one vnet to another, where it must be fully operational.
846 * XXXRW: There are some significant questions about event ordering, and
847 * how to prevent things from starting to use the interface during detach.
850 if_detach(struct ifnet *ifp)
853 if_detach_internal(ifp, 0);
857 if_detach_internal(struct ifnet *ifp, int vmove)
860 struct radix_node_head *rnh;
867 TAILQ_FOREACH(iter, &V_ifnet, if_link)
869 TAILQ_REMOVE(&V_ifnet, ifp, if_link);
875 curvnet->vnet_ifcnt--;
880 panic("%s: ifp=%p not on the ifnet tailq %p",
881 __func__, ifp, &V_ifnet);
883 return; /* XXX this should panic as well? */
887 * Remove/wait for pending events.
889 taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
892 * Remove routes and flush queues.
896 if (ALTQ_IS_ENABLED(&ifp->if_snd))
897 altq_disable(&ifp->if_snd);
898 if (ALTQ_IS_ATTACHED(&ifp->if_snd))
899 altq_detach(&ifp->if_snd);
910 * Remove all IPv6 kernel structs related to ifp. This should be done
911 * before removing routing entries below, since IPv6 interface direct
912 * routes are expected to be removed by the IPv6-specific kernel API.
913 * Otherwise, the kernel will detect some inconsistency and bark it.
921 * Prevent further calls into the device driver via ifnet.
926 * Remove link ifaddr pointer and maybe decrement if_index.
927 * Clean up all addresses.
931 /* We can now free link ifaddr. */
932 if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
933 ifa = TAILQ_FIRST(&ifp->if_addrhead);
934 TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
940 * Delete all remaining routes using this interface
941 * Unfortuneatly the only way to do this is to slog through
942 * the entire routing table looking for routes which point
943 * to this interface...oh well...
945 for (i = 1; i <= AF_MAX; i++) {
946 for (j = 0; j < rt_numfibs; j++) {
947 rnh = rt_tables_get_rnh(j, i);
950 RADIX_NODE_HEAD_LOCK(rnh);
951 (void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
952 RADIX_NODE_HEAD_UNLOCK(rnh);
956 /* Announce that the interface is gone. */
957 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
958 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
959 if (IS_DEFAULT_VNET(curvnet))
960 devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
964 * We cannot hold the lock over dom_ifdetach calls as they might
965 * sleep, for example trying to drain a callout, thus open up the
966 * theoretical race with re-attaching.
969 i = ifp->if_afdata_initialized;
970 ifp->if_afdata_initialized = 0;
971 IF_AFDATA_UNLOCK(ifp);
972 for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
973 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
974 (*dp->dom_ifdetach)(ifp,
975 ifp->if_afdata[dp->dom_family]);
981 * if_vmove() performs a limited version of if_detach() in current
982 * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
983 * An attempt is made to shrink if_index in current vnet, find an
984 * unused if_index in target vnet and calls if_grow() if necessary,
985 * and finally find an unused if_xname for the target vnet.
988 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
993 * Detach from current vnet, but preserve LLADDR info, do not
994 * mark as dead etc. so that the ifnet can be reattached later.
996 if_detach_internal(ifp, 1);
999 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
1000 * the if_index for that vnet if possible.
1002 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
1003 * or we'd lock on one vnet and unlock on another.
1006 ifindex_free_locked(ifp->if_index);
1010 * Perform interface-specific reassignment tasks, if provided by
1013 if (ifp->if_reassign != NULL)
1014 ifp->if_reassign(ifp, new_vnet, NULL);
1017 * Switch to the context of the target vnet.
1019 CURVNET_SET_QUIET(new_vnet);
1022 if (ifindex_alloc_locked(&idx) != 0) {
1024 panic("if_index overflow");
1026 ifp->if_index = idx;
1027 ifnet_setbyindex_locked(ifp->if_index, ifp);
1030 if_attach_internal(ifp, 1);
1036 * Move an ifnet to or from another child prison/vnet, specified by the jail id.
1039 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
1044 /* Try to find the prison within our visibility. */
1045 sx_slock(&allprison_lock);
1046 pr = prison_find_child(td->td_ucred->cr_prison, jid);
1047 sx_sunlock(&allprison_lock);
1050 prison_hold_locked(pr);
1051 mtx_unlock(&pr->pr_mtx);
1053 /* Do not try to move the iface from and to the same prison. */
1054 if (pr->pr_vnet == ifp->if_vnet) {
1059 /* Make sure the named iface does not exists in the dst. prison/vnet. */
1060 /* XXX Lock interfaces to avoid races. */
1061 CURVNET_SET_QUIET(pr->pr_vnet);
1062 difp = ifunit(ifname);
1069 /* Move the interface into the child jail/vnet. */
1070 if_vmove(ifp, pr->pr_vnet);
1072 /* Report the new if_xname back to the userland. */
1073 sprintf(ifname, "%s", ifp->if_xname);
1080 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1083 struct vnet *vnet_dst;
1086 /* Try to find the prison within our visibility. */
1087 sx_slock(&allprison_lock);
1088 pr = prison_find_child(td->td_ucred->cr_prison, jid);
1089 sx_sunlock(&allprison_lock);
1092 prison_hold_locked(pr);
1093 mtx_unlock(&pr->pr_mtx);
1095 /* Make sure the named iface exists in the source prison/vnet. */
1096 CURVNET_SET(pr->pr_vnet);
1097 ifp = ifunit(ifname); /* XXX Lock to avoid races. */
1104 /* Do not try to move the iface from and to the same prison. */
1105 vnet_dst = TD_TO_VNET(td);
1106 if (vnet_dst == ifp->if_vnet) {
1112 /* Get interface back from child jail/vnet. */
1113 if_vmove(ifp, vnet_dst);
1116 /* Report the new if_xname back to the userland. */
1117 sprintf(ifname, "%s", ifp->if_xname);
1125 * Add a group to an interface
1128 if_addgroup(struct ifnet *ifp, const char *groupname)
1130 struct ifg_list *ifgl;
1131 struct ifg_group *ifg = NULL;
1132 struct ifg_member *ifgm;
1134 if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1135 groupname[strlen(groupname) - 1] <= '9')
1139 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1140 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1145 if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1146 M_NOWAIT)) == NULL) {
1151 if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1152 M_TEMP, M_NOWAIT)) == NULL) {
1158 TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1159 if (!strcmp(ifg->ifg_group, groupname))
1163 if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1164 M_TEMP, M_NOWAIT)) == NULL) {
1170 strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1171 ifg->ifg_refcnt = 0;
1172 TAILQ_INIT(&ifg->ifg_members);
1173 EVENTHANDLER_INVOKE(group_attach_event, ifg);
1174 TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1178 ifgl->ifgl_group = ifg;
1179 ifgm->ifgm_ifp = ifp;
1182 TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1183 TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1184 IF_ADDR_UNLOCK(ifp);
1188 EVENTHANDLER_INVOKE(group_change_event, groupname);
1194 * Remove a group from an interface
1197 if_delgroup(struct ifnet *ifp, const char *groupname)
1199 struct ifg_list *ifgl;
1200 struct ifg_member *ifgm;
1203 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1204 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1212 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1213 IF_ADDR_UNLOCK(ifp);
1215 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1216 if (ifgm->ifgm_ifp == ifp)
1220 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1224 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1225 TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1226 EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1227 free(ifgl->ifgl_group, M_TEMP);
1233 EVENTHANDLER_INVOKE(group_change_event, groupname);
1239 * Remove an interface from all groups
1242 if_delgroups(struct ifnet *ifp)
1244 struct ifg_list *ifgl;
1245 struct ifg_member *ifgm;
1246 char groupname[IFNAMSIZ];
1249 while (!TAILQ_EMPTY(&ifp->if_groups)) {
1250 ifgl = TAILQ_FIRST(&ifp->if_groups);
1252 strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1255 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1256 IF_ADDR_UNLOCK(ifp);
1258 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1259 if (ifgm->ifgm_ifp == ifp)
1263 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1268 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1269 TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1270 EVENTHANDLER_INVOKE(group_detach_event,
1272 free(ifgl->ifgl_group, M_TEMP);
1278 EVENTHANDLER_INVOKE(group_change_event, groupname);
1286 * Stores all groups from an interface in memory pointed
1290 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1293 struct ifg_list *ifgl;
1294 struct ifg_req ifgrq, *ifgp;
1295 struct ifgroupreq *ifgr = data;
1297 if (ifgr->ifgr_len == 0) {
1299 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1300 ifgr->ifgr_len += sizeof(struct ifg_req);
1301 IF_ADDR_UNLOCK(ifp);
1305 len = ifgr->ifgr_len;
1306 ifgp = ifgr->ifgr_groups;
1309 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1310 if (len < sizeof(ifgrq)) {
1311 IF_ADDR_UNLOCK(ifp);
1314 bzero(&ifgrq, sizeof ifgrq);
1315 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1316 sizeof(ifgrq.ifgrq_group));
1317 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1318 IF_ADDR_UNLOCK(ifp);
1321 len -= sizeof(ifgrq);
1324 IF_ADDR_UNLOCK(ifp);
1330 * Stores all members of a group in memory pointed to by data
1333 if_getgroupmembers(struct ifgroupreq *data)
1335 struct ifgroupreq *ifgr = data;
1336 struct ifg_group *ifg;
1337 struct ifg_member *ifgm;
1338 struct ifg_req ifgrq, *ifgp;
1342 TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1343 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1350 if (ifgr->ifgr_len == 0) {
1351 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1352 ifgr->ifgr_len += sizeof(ifgrq);
1357 len = ifgr->ifgr_len;
1358 ifgp = ifgr->ifgr_groups;
1359 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1360 if (len < sizeof(ifgrq)) {
1364 bzero(&ifgrq, sizeof ifgrq);
1365 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1366 sizeof(ifgrq.ifgrq_member));
1367 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1371 len -= sizeof(ifgrq);
1380 * Delete Routes for a Network Interface
1382 * Called for each routing entry via the rnh->rnh_walktree() call above
1383 * to delete all route entries referencing a detaching network interface.
1386 * rn pointer to node in the routing table
1387 * arg argument passed to rnh->rnh_walktree() - detaching interface
1391 * errno failed - reason indicated
1395 if_rtdel(struct radix_node *rn, void *arg)
1397 struct rtentry *rt = (struct rtentry *)rn;
1398 struct ifnet *ifp = arg;
1401 if (rt->rt_ifp == ifp) {
1404 * Protect (sorta) against walktree recursion problems
1405 * with cloned routes
1407 if ((rt->rt_flags & RTF_UP) == 0)
1410 err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1411 rt_mask(rt), rt->rt_flags|RTF_RNH_LOCKED,
1412 (struct rtentry **) NULL, rt->rt_fibnum);
1414 log(LOG_WARNING, "if_rtdel: error %d\n", err);
1422 * Wrapper functions for struct ifnet address list locking macros. These are
1423 * used by kernel modules to avoid encoding programming interface or binary
1424 * interface assumptions that may be violated when kernel-internal locking
1425 * approaches change.
1428 if_addr_rlock(struct ifnet *ifp)
1435 if_addr_runlock(struct ifnet *ifp)
1438 IF_ADDR_UNLOCK(ifp);
1442 if_maddr_rlock(struct ifnet *ifp)
1449 if_maddr_runlock(struct ifnet *ifp)
1452 IF_ADDR_UNLOCK(ifp);
1456 * Reference count functions for ifaddrs.
1459 ifa_init(struct ifaddr *ifa)
1462 mtx_init(&ifa->ifa_mtx, "ifaddr", NULL, MTX_DEF);
1463 refcount_init(&ifa->ifa_refcnt, 1);
1467 ifa_ref(struct ifaddr *ifa)
1470 refcount_acquire(&ifa->ifa_refcnt);
1474 ifa_free(struct ifaddr *ifa)
1477 if (refcount_release(&ifa->ifa_refcnt)) {
1478 mtx_destroy(&ifa->ifa_mtx);
1479 free(ifa, M_IFADDR);
1484 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1487 struct rtentry *rt = NULL;
1488 struct rt_addrinfo info;
1489 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1491 bzero(&info, sizeof(info));
1492 info.rti_ifp = V_loif;
1493 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1494 info.rti_info[RTAX_DST] = ia;
1495 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1496 error = rtrequest1_fib(RTM_ADD, &info, &rt, 0);
1498 if (error == 0 && rt != NULL) {
1500 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1501 ifa->ifa_ifp->if_type;
1502 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1503 ifa->ifa_ifp->if_index;
1506 } else if (error != 0)
1507 log(LOG_INFO, "ifa_add_loopback_route: insertion failed\n");
1513 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1516 struct rt_addrinfo info;
1517 struct sockaddr_dl null_sdl;
1519 bzero(&null_sdl, sizeof(null_sdl));
1520 null_sdl.sdl_len = sizeof(null_sdl);
1521 null_sdl.sdl_family = AF_LINK;
1522 null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1523 null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1524 bzero(&info, sizeof(info));
1525 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1526 info.rti_info[RTAX_DST] = ia;
1527 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1528 error = rtrequest1_fib(RTM_DELETE, &info, NULL, 0);
1531 log(LOG_INFO, "ifa_del_loopback_route: deletion failed\n");
1537 * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1538 * structs used to represent other address families, it is necessary
1539 * to perform a different comparison.
1542 #define sa_equal(a1, a2) \
1543 (bcmp((a1), (a2), ((a1))->sa_len) == 0)
1545 #define sa_dl_equal(a1, a2) \
1546 ((((struct sockaddr_dl *)(a1))->sdl_len == \
1547 ((struct sockaddr_dl *)(a2))->sdl_len) && \
1548 (bcmp(LLADDR((struct sockaddr_dl *)(a1)), \
1549 LLADDR((struct sockaddr_dl *)(a2)), \
1550 ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1553 * Locate an interface based on a complete address.
1556 static struct ifaddr *
1557 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1562 IFNET_RLOCK_NOSLEEP();
1563 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1565 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1566 if (ifa->ifa_addr->sa_family != addr->sa_family)
1568 if (sa_equal(addr, ifa->ifa_addr)) {
1571 IF_ADDR_UNLOCK(ifp);
1574 /* IP6 doesn't have broadcast */
1575 if ((ifp->if_flags & IFF_BROADCAST) &&
1576 ifa->ifa_broadaddr &&
1577 ifa->ifa_broadaddr->sa_len != 0 &&
1578 sa_equal(ifa->ifa_broadaddr, addr)) {
1581 IF_ADDR_UNLOCK(ifp);
1585 IF_ADDR_UNLOCK(ifp);
1589 IFNET_RUNLOCK_NOSLEEP();
1594 ifa_ifwithaddr(struct sockaddr *addr)
1597 return (ifa_ifwithaddr_internal(addr, 1));
1601 ifa_ifwithaddr_check(struct sockaddr *addr)
1604 return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1608 * Locate an interface based on the broadcast address.
1612 ifa_ifwithbroadaddr(struct sockaddr *addr)
1617 IFNET_RLOCK_NOSLEEP();
1618 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1620 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1621 if (ifa->ifa_addr->sa_family != addr->sa_family)
1623 if ((ifp->if_flags & IFF_BROADCAST) &&
1624 ifa->ifa_broadaddr &&
1625 ifa->ifa_broadaddr->sa_len != 0 &&
1626 sa_equal(ifa->ifa_broadaddr, addr)) {
1628 IF_ADDR_UNLOCK(ifp);
1632 IF_ADDR_UNLOCK(ifp);
1636 IFNET_RUNLOCK_NOSLEEP();
1641 * Locate the point to point interface with a given destination address.
1645 ifa_ifwithdstaddr(struct sockaddr *addr)
1650 IFNET_RLOCK_NOSLEEP();
1651 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1652 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1655 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1656 if (ifa->ifa_addr->sa_family != addr->sa_family)
1658 if (ifa->ifa_dstaddr != NULL &&
1659 sa_equal(addr, ifa->ifa_dstaddr)) {
1661 IF_ADDR_UNLOCK(ifp);
1665 IF_ADDR_UNLOCK(ifp);
1669 IFNET_RUNLOCK_NOSLEEP();
1674 * Find an interface on a specific network. If many, choice
1675 * is most specific found.
1678 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp)
1682 struct ifaddr *ifa_maybe = NULL;
1683 u_int af = addr->sa_family;
1684 char *addr_data = addr->sa_data, *cplim;
1687 * AF_LINK addresses can be looked up directly by their index number,
1688 * so do that if we can.
1690 if (af == AF_LINK) {
1691 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1692 if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1693 return (ifaddr_byindex(sdl->sdl_index));
1697 * Scan though each interface, looking for ones that have addresses
1698 * in this address family. Maintain a reference on ifa_maybe once
1699 * we find one, as we release the IF_ADDR_LOCK() that kept it stable
1700 * when we move onto the next interface.
1702 IFNET_RLOCK_NOSLEEP();
1703 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1705 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1706 char *cp, *cp2, *cp3;
1708 if (ifa->ifa_addr->sa_family != af)
1710 if (af == AF_INET &&
1711 ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1713 * This is a bit broken as it doesn't
1714 * take into account that the remote end may
1715 * be a single node in the network we are
1717 * The trouble is that we don't know the
1718 * netmask for the remote end.
1720 if (ifa->ifa_dstaddr != NULL &&
1721 sa_equal(addr, ifa->ifa_dstaddr)) {
1723 IF_ADDR_UNLOCK(ifp);
1728 * if we have a special address handler,
1729 * then use it instead of the generic one.
1731 if (ifa->ifa_claim_addr) {
1732 if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1734 IF_ADDR_UNLOCK(ifp);
1741 * Scan all the bits in the ifa's address.
1742 * If a bit dissagrees with what we are
1743 * looking for, mask it with the netmask
1744 * to see if it really matters.
1745 * (A byte at a time)
1747 if (ifa->ifa_netmask == 0)
1750 cp2 = ifa->ifa_addr->sa_data;
1751 cp3 = ifa->ifa_netmask->sa_data;
1752 cplim = ifa->ifa_netmask->sa_len
1753 + (char *)ifa->ifa_netmask;
1755 if ((*cp++ ^ *cp2++) & *cp3++)
1756 goto next; /* next address! */
1758 * If the netmask of what we just found
1759 * is more specific than what we had before
1760 * (if we had one) then remember the new one
1761 * before continuing to search
1762 * for an even better one.
1764 if (ifa_maybe == NULL ||
1765 rn_refines((caddr_t)ifa->ifa_netmask,
1766 (caddr_t)ifa_maybe->ifa_netmask)) {
1767 if (ifa_maybe != NULL)
1768 ifa_free(ifa_maybe);
1774 IF_ADDR_UNLOCK(ifp);
1779 IFNET_RUNLOCK_NOSLEEP();
1780 if (ifa_maybe != NULL)
1781 ifa_free(ifa_maybe);
1786 * Find an interface address specific to an interface best matching
1790 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1793 char *cp, *cp2, *cp3;
1795 struct ifaddr *ifa_maybe = NULL;
1796 u_int af = addr->sa_family;
1801 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1802 if (ifa->ifa_addr->sa_family != af)
1804 if (ifa_maybe == NULL)
1806 if (ifa->ifa_netmask == 0) {
1807 if (sa_equal(addr, ifa->ifa_addr) ||
1808 (ifa->ifa_dstaddr &&
1809 sa_equal(addr, ifa->ifa_dstaddr)))
1813 if (ifp->if_flags & IFF_POINTOPOINT) {
1814 if (sa_equal(addr, ifa->ifa_dstaddr))
1818 cp2 = ifa->ifa_addr->sa_data;
1819 cp3 = ifa->ifa_netmask->sa_data;
1820 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1821 for (; cp3 < cplim; cp3++)
1822 if ((*cp++ ^ *cp2++) & *cp3)
1832 IF_ADDR_UNLOCK(ifp);
1836 #include <net/if_llatbl.h>
1839 * Default action when installing a route with a Link Level gateway.
1840 * Lookup an appropriate real ifa to point to.
1841 * This should be moved to /sys/net/link.c eventually.
1844 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1846 struct ifaddr *ifa, *oifa;
1847 struct sockaddr *dst;
1852 if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1853 ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1855 ifa = ifaof_ifpforaddr(dst, ifp);
1860 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1861 ifa->ifa_rtrequest(cmd, rt, info);
1866 * Mark an interface down and notify protocols of
1868 * NOTE: must be called at splnet or eqivalent.
1871 if_unroute(struct ifnet *ifp, int flag, int fam)
1875 KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1877 ifp->if_flags &= ~flag;
1878 getmicrotime(&ifp->if_lastchange);
1879 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1880 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1881 pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1882 ifp->if_qflush(ifp);
1885 (*carp_linkstate_p)(ifp);
1890 * Mark an interface up and notify protocols of
1892 * NOTE: must be called at splnet or eqivalent.
1895 if_route(struct ifnet *ifp, int flag, int fam)
1899 KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1901 ifp->if_flags |= flag;
1902 getmicrotime(&ifp->if_lastchange);
1903 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1904 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1905 pfctlinput(PRC_IFUP, ifa->ifa_addr);
1907 (*carp_linkstate_p)(ifp);
1914 void (*vlan_link_state_p)(struct ifnet *, int); /* XXX: private from if_vlan */
1915 void (*vlan_trunk_cap_p)(struct ifnet *); /* XXX: private from if_vlan */
1918 * Handle a change in the interface link state. To avoid LORs
1919 * between driver lock and upper layer locks, as well as possible
1920 * recursions, we post event to taskqueue, and all job
1921 * is done in static do_link_state_change().
1924 if_link_state_change(struct ifnet *ifp, int link_state)
1926 /* Return if state hasn't changed. */
1927 if (ifp->if_link_state == link_state)
1930 ifp->if_link_state = link_state;
1932 taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1936 do_link_state_change(void *arg, int pending)
1938 struct ifnet *ifp = (struct ifnet *)arg;
1939 int link_state = ifp->if_link_state;
1940 CURVNET_SET(ifp->if_vnet);
1942 /* Notify that the link state has changed. */
1944 if (ifp->if_vlantrunk != NULL)
1945 (*vlan_link_state_p)(ifp, 0);
1947 if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1948 IFP2AC(ifp)->ac_netgraph != NULL)
1949 (*ng_ether_link_state_p)(ifp, link_state);
1951 (*carp_linkstate_p)(ifp);
1952 if (ifp->if_bridge) {
1953 KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1954 (*bstp_linkstate_p)(ifp, link_state);
1957 KASSERT(lagg_linkstate_p != NULL,("if_lagg not loaded!"));
1958 (*lagg_linkstate_p)(ifp, link_state);
1961 if (IS_DEFAULT_VNET(curvnet))
1962 devctl_notify("IFNET", ifp->if_xname,
1963 (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
1966 if_printf(ifp, "%d link states coalesced\n", pending);
1967 if (log_link_state_change)
1968 log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1969 (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1974 * Mark an interface down and notify protocols of
1976 * NOTE: must be called at splnet or eqivalent.
1979 if_down(struct ifnet *ifp)
1982 if_unroute(ifp, IFF_UP, AF_UNSPEC);
1986 * Mark an interface up and notify protocols of
1988 * NOTE: must be called at splnet or eqivalent.
1991 if_up(struct ifnet *ifp)
1994 if_route(ifp, IFF_UP, AF_UNSPEC);
1998 * Flush an interface queue.
2001 if_qflush(struct ifnet *ifp)
2009 if (ALTQ_IS_ENABLED(ifq))
2013 while ((m = n) != 0) {
2024 * Handle interface watchdog timer routines. Called
2025 * from softclock, we decrement timers (if set) and
2026 * call the appropriate interface routine on expiration.
2028 * XXXRW: Note that because timeouts run with Giant, if_watchdog() is called
2032 if_slowtimo(void *arg)
2034 VNET_ITERATOR_DECL(vnet_iter);
2038 VNET_LIST_RLOCK_NOSLEEP();
2039 IFNET_RLOCK_NOSLEEP();
2040 VNET_FOREACH(vnet_iter) {
2041 CURVNET_SET(vnet_iter);
2042 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2043 if (ifp->if_timer == 0 || --ifp->if_timer)
2045 if (ifp->if_watchdog)
2046 (*ifp->if_watchdog)(ifp);
2050 IFNET_RUNLOCK_NOSLEEP();
2051 VNET_LIST_RUNLOCK_NOSLEEP();
2053 timeout(if_slowtimo, (void *)0, hz / IFNET_SLOWHZ);
2057 * Map interface name to interface structure pointer, with or without
2058 * returning a reference.
2061 ifunit_ref(const char *name)
2065 IFNET_RLOCK_NOSLEEP();
2066 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2067 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2068 !(ifp->if_flags & IFF_DYING))
2073 IFNET_RUNLOCK_NOSLEEP();
2078 ifunit(const char *name)
2082 IFNET_RLOCK_NOSLEEP();
2083 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2084 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2087 IFNET_RUNLOCK_NOSLEEP();
2092 * Hardware specific interface ioctls.
2095 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2100 int new_flags, temp_flags;
2101 size_t namelen, onamelen;
2103 char *descrbuf, *odescrbuf;
2104 char new_name[IFNAMSIZ];
2106 struct sockaddr_dl *sdl;
2108 ifr = (struct ifreq *)data;
2111 ifr->ifr_index = ifp->if_index;
2115 temp_flags = ifp->if_flags | ifp->if_drv_flags;
2116 ifr->ifr_flags = temp_flags & 0xffff;
2117 ifr->ifr_flagshigh = temp_flags >> 16;
2121 ifr->ifr_reqcap = ifp->if_capabilities;
2122 ifr->ifr_curcap = ifp->if_capenable;
2127 error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2132 ifr->ifr_metric = ifp->if_metric;
2136 ifr->ifr_mtu = ifp->if_mtu;
2140 ifr->ifr_phys = ifp->if_physical;
2145 sx_slock(&ifdescr_sx);
2146 if (ifp->if_description == NULL)
2149 /* space for terminating nul */
2150 descrlen = strlen(ifp->if_description) + 1;
2151 if (ifr->ifr_buffer.length < descrlen)
2152 ifr->ifr_buffer.buffer = NULL;
2154 error = copyout(ifp->if_description,
2155 ifr->ifr_buffer.buffer, descrlen);
2156 ifr->ifr_buffer.length = descrlen;
2158 sx_sunlock(&ifdescr_sx);
2162 error = priv_check(td, PRIV_NET_SETIFDESCR);
2167 * Copy only (length-1) bytes to make sure that
2168 * if_description is always nul terminated. The
2169 * length parameter is supposed to count the
2170 * terminating nul in.
2172 if (ifr->ifr_buffer.length > ifdescr_maxlen)
2173 return (ENAMETOOLONG);
2174 else if (ifr->ifr_buffer.length == 0)
2177 descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2179 error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2180 ifr->ifr_buffer.length - 1);
2182 free(descrbuf, M_IFDESCR);
2187 sx_xlock(&ifdescr_sx);
2188 odescrbuf = ifp->if_description;
2189 ifp->if_description = descrbuf;
2190 sx_xunlock(&ifdescr_sx);
2192 getmicrotime(&ifp->if_lastchange);
2193 free(odescrbuf, M_IFDESCR);
2197 error = priv_check(td, PRIV_NET_SETIFFLAGS);
2201 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2202 * check, so we don't need special handling here yet.
2204 new_flags = (ifr->ifr_flags & 0xffff) |
2205 (ifr->ifr_flagshigh << 16);
2206 if (ifp->if_flags & IFF_SMART) {
2207 /* Smart drivers twiddle their own routes */
2208 } else if (ifp->if_flags & IFF_UP &&
2209 (new_flags & IFF_UP) == 0) {
2213 } else if (new_flags & IFF_UP &&
2214 (ifp->if_flags & IFF_UP) == 0) {
2219 /* See if permanently promiscuous mode bit is about to flip */
2220 if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2221 if (new_flags & IFF_PPROMISC)
2222 ifp->if_flags |= IFF_PROMISC;
2223 else if (ifp->if_pcount == 0)
2224 ifp->if_flags &= ~IFF_PROMISC;
2225 log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2227 (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
2229 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2230 (new_flags &~ IFF_CANTCHANGE);
2231 if (ifp->if_ioctl) {
2232 (void) (*ifp->if_ioctl)(ifp, cmd, data);
2234 getmicrotime(&ifp->if_lastchange);
2238 error = priv_check(td, PRIV_NET_SETIFCAP);
2241 if (ifp->if_ioctl == NULL)
2242 return (EOPNOTSUPP);
2243 if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2245 error = (*ifp->if_ioctl)(ifp, cmd, data);
2247 getmicrotime(&ifp->if_lastchange);
2252 error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2257 error = priv_check(td, PRIV_NET_SETIFNAME);
2260 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2263 if (new_name[0] == '\0')
2265 if (ifunit(new_name) != NULL)
2269 * XXX: Locking. Nothing else seems to lock if_flags,
2270 * and there are numerous other races with the
2271 * ifunit() checks not being atomic with namespace
2272 * changes (renames, vmoves, if_attach, etc).
2274 ifp->if_flags |= IFF_RENAMING;
2276 /* Announce the departure of the interface. */
2277 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2278 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2280 log(LOG_INFO, "%s: changing name to '%s'\n",
2281 ifp->if_xname, new_name);
2283 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2286 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2287 namelen = strlen(new_name);
2288 onamelen = sdl->sdl_nlen;
2290 * Move the address if needed. This is safe because we
2291 * allocate space for a name of length IFNAMSIZ when we
2292 * create this in if_attach().
2294 if (namelen != onamelen) {
2295 bcopy(sdl->sdl_data + onamelen,
2296 sdl->sdl_data + namelen, sdl->sdl_alen);
2298 bcopy(new_name, sdl->sdl_data, namelen);
2299 sdl->sdl_nlen = namelen;
2300 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2301 bzero(sdl->sdl_data, onamelen);
2302 while (namelen != 0)
2303 sdl->sdl_data[--namelen] = 0xff;
2306 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2307 /* Announce the return of the interface. */
2308 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2310 ifp->if_flags &= ~IFF_RENAMING;
2315 error = priv_check(td, PRIV_NET_SETIFVNET);
2318 error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2323 error = priv_check(td, PRIV_NET_SETIFMETRIC);
2326 ifp->if_metric = ifr->ifr_metric;
2327 getmicrotime(&ifp->if_lastchange);
2331 error = priv_check(td, PRIV_NET_SETIFPHYS);
2334 if (ifp->if_ioctl == NULL)
2335 return (EOPNOTSUPP);
2336 error = (*ifp->if_ioctl)(ifp, cmd, data);
2338 getmicrotime(&ifp->if_lastchange);
2343 u_long oldmtu = ifp->if_mtu;
2345 error = priv_check(td, PRIV_NET_SETIFMTU);
2348 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2350 if (ifp->if_ioctl == NULL)
2351 return (EOPNOTSUPP);
2352 error = (*ifp->if_ioctl)(ifp, cmd, data);
2354 getmicrotime(&ifp->if_lastchange);
2358 * If the link MTU changed, do network layer specific procedure.
2360 if (ifp->if_mtu != oldmtu) {
2370 if (cmd == SIOCADDMULTI)
2371 error = priv_check(td, PRIV_NET_ADDMULTI);
2373 error = priv_check(td, PRIV_NET_DELMULTI);
2377 /* Don't allow group membership on non-multicast interfaces. */
2378 if ((ifp->if_flags & IFF_MULTICAST) == 0)
2379 return (EOPNOTSUPP);
2381 /* Don't let users screw up protocols' entries. */
2382 if (ifr->ifr_addr.sa_family != AF_LINK)
2385 if (cmd == SIOCADDMULTI) {
2386 struct ifmultiaddr *ifma;
2389 * Userland is only permitted to join groups once
2390 * via the if_addmulti() KPI, because it cannot hold
2391 * struct ifmultiaddr * between calls. It may also
2392 * lose a race while we check if the membership
2396 ifma = if_findmulti(ifp, &ifr->ifr_addr);
2397 IF_ADDR_UNLOCK(ifp);
2401 error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2403 error = if_delmulti(ifp, &ifr->ifr_addr);
2406 getmicrotime(&ifp->if_lastchange);
2409 case SIOCSIFPHYADDR:
2410 case SIOCDIFPHYADDR:
2412 case SIOCSIFPHYADDR_IN6:
2414 case SIOCSLIFPHYADDR:
2416 case SIOCSIFGENERIC:
2417 error = priv_check(td, PRIV_NET_HWIOCTL);
2420 if (ifp->if_ioctl == NULL)
2421 return (EOPNOTSUPP);
2422 error = (*ifp->if_ioctl)(ifp, cmd, data);
2424 getmicrotime(&ifp->if_lastchange);
2428 ifs = (struct ifstat *)data;
2429 ifs->ascii[0] = '\0';
2431 case SIOCGIFPSRCADDR:
2432 case SIOCGIFPDSTADDR:
2433 case SIOCGLIFPHYADDR:
2435 case SIOCGIFGENERIC:
2436 if (ifp->if_ioctl == NULL)
2437 return (EOPNOTSUPP);
2438 error = (*ifp->if_ioctl)(ifp, cmd, data);
2442 error = priv_check(td, PRIV_NET_SETLLADDR);
2445 error = if_setlladdr(ifp,
2446 ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2447 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2452 struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2454 error = priv_check(td, PRIV_NET_ADDIFGROUP);
2457 if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2463 if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2469 struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2471 error = priv_check(td, PRIV_NET_DELIFGROUP);
2474 if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2486 #ifdef COMPAT_FREEBSD32
2494 #define SIOCGIFCONF32 _IOWR('i', 36, struct ifconf32)
2501 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2511 return (ifconf(cmd, data));
2513 #ifdef COMPAT_FREEBSD32
2516 struct ifconf32 *ifc32;
2519 ifc32 = (struct ifconf32 *)data;
2520 ifc.ifc_len = ifc32->ifc_len;
2521 ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2523 return (ifconf(SIOCGIFCONF, (void *)&ifc));
2527 ifr = (struct ifreq *)data;
2532 error = priv_check(td, PRIV_NET_SETIFVNET);
2535 return (if_vmove_reclaim(td, ifr->ifr_name, ifr->ifr_jid));
2539 error = priv_check(td, PRIV_NET_IFCREATE);
2542 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
2543 cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
2545 error = priv_check(td, PRIV_NET_IFDESTROY);
2548 return if_clone_destroy(ifr->ifr_name);
2550 case SIOCIFGCLONERS:
2551 return (if_clone_list((struct if_clonereq *)data));
2553 return (if_getgroupmembers((struct ifgroupreq *)data));
2556 ifp = ifunit_ref(ifr->ifr_name);
2560 error = ifhwioctl(cmd, ifp, data, td);
2561 if (error != ENOIOCTL) {
2566 oif_flags = ifp->if_flags;
2567 if (so->so_proto == NULL) {
2569 return (EOPNOTSUPP);
2572 error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
2575 if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL)
2576 error = (*ifp->if_ioctl)(ifp, cmd, data);
2583 case SIOCSIFDSTADDR:
2585 case SIOCSIFBRDADDR:
2586 case SIOCSIFNETMASK:
2587 #if BYTE_ORDER != BIG_ENDIAN
2588 if (ifr->ifr_addr.sa_family == 0 &&
2589 ifr->ifr_addr.sa_len < 16) {
2590 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2591 ifr->ifr_addr.sa_len = 16;
2594 if (ifr->ifr_addr.sa_len == 0)
2595 ifr->ifr_addr.sa_len = 16;
2603 case OSIOCGIFDSTADDR:
2604 cmd = SIOCGIFDSTADDR;
2607 case OSIOCGIFBRDADDR:
2608 cmd = SIOCGIFBRDADDR;
2611 case OSIOCGIFNETMASK:
2612 cmd = SIOCGIFNETMASK;
2614 error = ((*so->so_proto->pr_usrreqs->pru_control)(so,
2618 if (error == EOPNOTSUPP && ifp != NULL &&
2619 ifp->if_ioctl != NULL)
2620 error = (*ifp->if_ioctl)(ifp, cmd, data);
2624 case OSIOCGIFDSTADDR:
2625 case OSIOCGIFBRDADDR:
2626 case OSIOCGIFNETMASK:
2627 *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2631 #endif /* COMPAT_43 */
2633 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2635 if (ifp->if_flags & IFF_UP) {
2647 * The code common to handling reference counted flags,
2648 * e.g., in ifpromisc() and if_allmulti().
2649 * The "pflag" argument can specify a permanent mode flag to check,
2650 * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2652 * Only to be used on stack-owned flags, not driver-owned flags.
2655 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2659 int oldflags, oldcount;
2661 /* Sanity checks to catch programming errors */
2662 KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2663 ("%s: setting driver-owned flag %d", __func__, flag));
2666 KASSERT(*refcount >= 0,
2667 ("%s: increment negative refcount %d for flag %d",
2668 __func__, *refcount, flag));
2670 KASSERT(*refcount > 0,
2671 ("%s: decrement non-positive refcount %d for flag %d",
2672 __func__, *refcount, flag));
2674 /* In case this mode is permanent, just touch refcount */
2675 if (ifp->if_flags & pflag) {
2676 *refcount += onswitch ? 1 : -1;
2680 /* Save ifnet parameters for if_ioctl() may fail */
2681 oldcount = *refcount;
2682 oldflags = ifp->if_flags;
2685 * See if we aren't the only and touching refcount is enough.
2686 * Actually toggle interface flag if we are the first or last.
2691 ifp->if_flags |= flag;
2695 ifp->if_flags &= ~flag;
2698 /* Call down the driver since we've changed interface flags */
2699 if (ifp->if_ioctl == NULL) {
2703 ifr.ifr_flags = ifp->if_flags & 0xffff;
2704 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2705 error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2708 /* Notify userland that interface flags have changed */
2713 /* Recover after driver error */
2714 *refcount = oldcount;
2715 ifp->if_flags = oldflags;
2720 * Set/clear promiscuous mode on interface ifp based on the truth value
2721 * of pswitch. The calls are reference counted so that only the first
2722 * "on" request actually has an effect, as does the final "off" request.
2723 * Results are undefined if the "off" and "on" requests are not matched.
2726 ifpromisc(struct ifnet *ifp, int pswitch)
2729 int oldflags = ifp->if_flags;
2731 error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2732 &ifp->if_pcount, pswitch);
2733 /* If promiscuous mode status has changed, log a message */
2734 if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2735 log(LOG_INFO, "%s: promiscuous mode %s\n",
2737 (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2742 * Return interface configuration
2743 * of system. List may be used
2744 * in later ioctl's (above) to get
2745 * other information.
2749 ifconf(u_long cmd, caddr_t data)
2751 struct ifconf *ifc = (struct ifconf *)data;
2756 int error, full = 0, valid_len, max_len;
2758 /* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2759 max_len = MAXPHYS - 1;
2761 /* Prevent hostile input from being able to crash the system */
2762 if (ifc->ifc_len <= 0)
2766 if (ifc->ifc_len <= max_len) {
2767 max_len = ifc->ifc_len;
2770 sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2775 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2779 * Zero the ifr_name buffer to make sure we don't
2780 * disclose the contents of the stack.
2782 memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2784 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2785 >= sizeof(ifr.ifr_name)) {
2788 return (ENAMETOOLONG);
2793 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2794 struct sockaddr *sa = ifa->ifa_addr;
2796 if (prison_if(curthread->td_ucred, sa) != 0)
2800 if (cmd == OSIOCGIFCONF) {
2801 struct osockaddr *osa =
2802 (struct osockaddr *)&ifr.ifr_addr;
2804 osa->sa_family = sa->sa_family;
2805 sbuf_bcat(sb, &ifr, sizeof(ifr));
2806 max_len += sizeof(ifr);
2809 if (sa->sa_len <= sizeof(*sa)) {
2811 sbuf_bcat(sb, &ifr, sizeof(ifr));
2812 max_len += sizeof(ifr);
2815 offsetof(struct ifreq, ifr_addr));
2816 max_len += offsetof(struct ifreq, ifr_addr);
2817 sbuf_bcat(sb, sa, sa->sa_len);
2818 max_len += sa->sa_len;
2821 if (!sbuf_overflowed(sb))
2822 valid_len = sbuf_len(sb);
2824 IF_ADDR_UNLOCK(ifp);
2826 bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2827 sbuf_bcat(sb, &ifr, sizeof(ifr));
2828 max_len += sizeof(ifr);
2830 if (!sbuf_overflowed(sb))
2831 valid_len = sbuf_len(sb);
2837 * If we didn't allocate enough space (uncommon), try again. If
2838 * we have already allocated as much space as we are allowed,
2839 * return what we've got.
2841 if (valid_len != max_len && !full) {
2846 ifc->ifc_len = valid_len;
2848 error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2854 * Just like ifpromisc(), but for all-multicast-reception mode.
2857 if_allmulti(struct ifnet *ifp, int onswitch)
2860 return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2863 struct ifmultiaddr *
2864 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2866 struct ifmultiaddr *ifma;
2868 IF_ADDR_LOCK_ASSERT(ifp);
2870 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2871 if (sa->sa_family == AF_LINK) {
2872 if (sa_dl_equal(ifma->ifma_addr, sa))
2875 if (sa_equal(ifma->ifma_addr, sa))
2884 * Allocate a new ifmultiaddr and initialize based on passed arguments. We
2885 * make copies of passed sockaddrs. The ifmultiaddr will not be added to
2886 * the ifnet multicast address list here, so the caller must do that and
2887 * other setup work (such as notifying the device driver). The reference
2888 * count is initialized to 1.
2890 static struct ifmultiaddr *
2891 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2894 struct ifmultiaddr *ifma;
2895 struct sockaddr *dupsa;
2897 ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2902 dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2903 if (dupsa == NULL) {
2904 free(ifma, M_IFMADDR);
2907 bcopy(sa, dupsa, sa->sa_len);
2908 ifma->ifma_addr = dupsa;
2910 ifma->ifma_ifp = ifp;
2911 ifma->ifma_refcount = 1;
2912 ifma->ifma_protospec = NULL;
2915 ifma->ifma_lladdr = NULL;
2919 dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
2920 if (dupsa == NULL) {
2921 free(ifma->ifma_addr, M_IFMADDR);
2922 free(ifma, M_IFMADDR);
2925 bcopy(llsa, dupsa, llsa->sa_len);
2926 ifma->ifma_lladdr = dupsa;
2932 * if_freemulti: free ifmultiaddr structure and possibly attached related
2933 * addresses. The caller is responsible for implementing reference
2934 * counting, notifying the driver, handling routing messages, and releasing
2935 * any dependent link layer state.
2938 if_freemulti(struct ifmultiaddr *ifma)
2941 KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2942 ifma->ifma_refcount));
2943 KASSERT(ifma->ifma_protospec == NULL,
2944 ("if_freemulti: protospec not NULL"));
2946 if (ifma->ifma_lladdr != NULL)
2947 free(ifma->ifma_lladdr, M_IFMADDR);
2948 free(ifma->ifma_addr, M_IFMADDR);
2949 free(ifma, M_IFMADDR);
2953 * Register an additional multicast address with a network interface.
2955 * - If the address is already present, bump the reference count on the
2956 * address and return.
2957 * - If the address is not link-layer, look up a link layer address.
2958 * - Allocate address structures for one or both addresses, and attach to the
2959 * multicast address list on the interface. If automatically adding a link
2960 * layer address, the protocol address will own a reference to the link
2961 * layer address, to be freed when it is freed.
2962 * - Notify the network device driver of an addition to the multicast address
2965 * 'sa' points to caller-owned memory with the desired multicast address.
2967 * 'retifma' will be used to return a pointer to the resulting multicast
2968 * address reference, if desired.
2971 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2972 struct ifmultiaddr **retifma)
2974 struct ifmultiaddr *ifma, *ll_ifma;
2975 struct sockaddr *llsa;
2979 * If the address is already present, return a new reference to it;
2980 * otherwise, allocate storage and set up a new address.
2983 ifma = if_findmulti(ifp, sa);
2985 ifma->ifma_refcount++;
2986 if (retifma != NULL)
2988 IF_ADDR_UNLOCK(ifp);
2993 * The address isn't already present; resolve the protocol address
2994 * into a link layer address, and then look that up, bump its
2995 * refcount or allocate an ifma for that also. If 'llsa' was
2996 * returned, we will need to free it later.
3000 if (ifp->if_resolvemulti != NULL) {
3001 error = ifp->if_resolvemulti(ifp, &llsa, sa);
3007 * Allocate the new address. Don't hook it up yet, as we may also
3008 * need to allocate a link layer multicast address.
3010 ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
3017 * If a link layer address is found, we'll need to see if it's
3018 * already present in the address list, or allocate is as well.
3019 * When this block finishes, the link layer address will be on the
3023 ll_ifma = if_findmulti(ifp, llsa);
3024 if (ll_ifma == NULL) {
3025 ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
3026 if (ll_ifma == NULL) {
3027 --ifma->ifma_refcount;
3032 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
3035 ll_ifma->ifma_refcount++;
3036 ifma->ifma_llifma = ll_ifma;
3040 * We now have a new multicast address, ifma, and possibly a new or
3041 * referenced link layer address. Add the primary address to the
3042 * ifnet address list.
3044 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3046 if (retifma != NULL)
3050 * Must generate the message while holding the lock so that 'ifma'
3051 * pointer is still valid.
3053 rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3054 IF_ADDR_UNLOCK(ifp);
3057 * We are certain we have added something, so call down to the
3058 * interface to let them know about it.
3060 if (ifp->if_ioctl != NULL) {
3061 (void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3065 free(llsa, M_IFMADDR);
3071 free(llsa, M_IFMADDR);
3074 IF_ADDR_UNLOCK(ifp);
3079 * Delete a multicast group membership by network-layer group address.
3081 * Returns ENOENT if the entry could not be found. If ifp no longer
3082 * exists, results are undefined. This entry point should only be used
3083 * from subsystems which do appropriate locking to hold ifp for the
3084 * duration of the call.
3085 * Network-layer protocol domains must use if_delmulti_ifma().
3088 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3090 struct ifmultiaddr *ifma;
3095 IFNET_RLOCK_NOSLEEP();
3096 TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3101 IFNET_RUNLOCK_NOSLEEP();
3103 KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3110 ifma = if_findmulti(ifp, sa);
3112 lastref = if_delmulti_locked(ifp, ifma, 0);
3113 IF_ADDR_UNLOCK(ifp);
3118 if (lastref && ifp->if_ioctl != NULL) {
3119 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3126 * Delete all multicast group membership for an interface.
3127 * Should be used to quickly flush all multicast filters.
3130 if_delallmulti(struct ifnet *ifp)
3132 struct ifmultiaddr *ifma;
3133 struct ifmultiaddr *next;
3136 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3137 if_delmulti_locked(ifp, ifma, 0);
3138 IF_ADDR_UNLOCK(ifp);
3142 * Delete a multicast group membership by group membership pointer.
3143 * Network-layer protocol domains must use this routine.
3145 * It is safe to call this routine if the ifp disappeared.
3148 if_delmulti_ifma(struct ifmultiaddr *ifma)
3153 ifp = ifma->ifma_ifp;
3156 printf("%s: ifma_ifp seems to be detached\n", __func__);
3160 IFNET_RLOCK_NOSLEEP();
3161 TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3165 printf("%s: ifnet %p disappeared\n", __func__, ifp);
3168 IFNET_RUNLOCK_NOSLEEP();
3172 * If and only if the ifnet instance exists: Acquire the address lock.
3177 lastref = if_delmulti_locked(ifp, ifma, 0);
3181 * If and only if the ifnet instance exists:
3182 * Release the address lock.
3183 * If the group was left: update the hardware hash filter.
3185 IF_ADDR_UNLOCK(ifp);
3186 if (lastref && ifp->if_ioctl != NULL) {
3187 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3193 * Perform deletion of network-layer and/or link-layer multicast address.
3195 * Return 0 if the reference count was decremented.
3196 * Return 1 if the final reference was released, indicating that the
3197 * hardware hash filter should be reprogrammed.
3200 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3202 struct ifmultiaddr *ll_ifma;
3204 if (ifp != NULL && ifma->ifma_ifp != NULL) {
3205 KASSERT(ifma->ifma_ifp == ifp,
3206 ("%s: inconsistent ifp %p", __func__, ifp));
3207 IF_ADDR_LOCK_ASSERT(ifp);
3210 ifp = ifma->ifma_ifp;
3213 * If the ifnet is detaching, null out references to ifnet,
3214 * so that upper protocol layers will notice, and not attempt
3215 * to obtain locks for an ifnet which no longer exists. The
3216 * routing socket announcement must happen before the ifnet
3217 * instance is detached from the system.
3221 printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3224 * ifp may already be nulled out if we are being reentered
3225 * to delete the ll_ifma.
3228 rt_newmaddrmsg(RTM_DELMADDR, ifma);
3229 ifma->ifma_ifp = NULL;
3233 if (--ifma->ifma_refcount > 0)
3237 * If this ifma is a network-layer ifma, a link-layer ifma may
3238 * have been associated with it. Release it first if so.
3240 ll_ifma = ifma->ifma_llifma;
3241 if (ll_ifma != NULL) {
3242 KASSERT(ifma->ifma_lladdr != NULL,
3243 ("%s: llifma w/o lladdr", __func__));
3245 ll_ifma->ifma_ifp = NULL; /* XXX */
3246 if (--ll_ifma->ifma_refcount == 0) {
3248 TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3251 if_freemulti(ll_ifma);
3256 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3261 * The last reference to this instance of struct ifmultiaddr
3262 * was released; the hardware should be notified of this change.
3268 * Set the link layer address on an interface.
3270 * At this time we only support certain types of interfaces,
3271 * and we don't allow the length of the address to change.
3274 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3276 struct sockaddr_dl *sdl;
3283 IF_ADDR_UNLOCK(ifp);
3287 IF_ADDR_UNLOCK(ifp);
3288 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3293 if (len != sdl->sdl_alen) { /* don't allow length to change */
3297 switch (ifp->if_type) {
3305 case IFT_IEEE8023ADLAG:
3307 bcopy(lladdr, LLADDR(sdl), len);
3316 * If the interface is already up, we need
3317 * to re-init it in order to reprogram its
3320 if ((ifp->if_flags & IFF_UP) != 0) {
3321 if (ifp->if_ioctl) {
3322 ifp->if_flags &= ~IFF_UP;
3323 ifr.ifr_flags = ifp->if_flags & 0xffff;
3324 ifr.ifr_flagshigh = ifp->if_flags >> 16;
3325 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3326 ifp->if_flags |= IFF_UP;
3327 ifr.ifr_flags = ifp->if_flags & 0xffff;
3328 ifr.ifr_flagshigh = ifp->if_flags >> 16;
3329 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3333 * Also send gratuitous ARPs to notify other nodes about
3334 * the address change.
3336 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3337 if (ifa->ifa_addr->sa_family == AF_INET)
3338 arp_ifinit(ifp, ifa);
3346 * The name argument must be a pointer to storage which will last as
3347 * long as the interface does. For physical devices, the result of
3348 * device_get_name(dev) is a good choice and for pseudo-devices a
3349 * static string works well.
3352 if_initname(struct ifnet *ifp, const char *name, int unit)
3354 ifp->if_dname = name;
3355 ifp->if_dunit = unit;
3356 if (unit != IF_DUNIT_NONE)
3357 snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3359 strlcpy(ifp->if_xname, name, IFNAMSIZ);
3363 if_printf(struct ifnet *ifp, const char * fmt, ...)
3368 retval = printf("%s: ", ifp->if_xname);
3370 retval += vprintf(fmt, ap);
3376 if_start(struct ifnet *ifp)
3379 (*(ifp)->if_start)(ifp);
3383 * Backwards compatibility interface for drivers
3384 * that have not implemented it
3387 if_transmit(struct ifnet *ifp, struct mbuf *m)
3391 IFQ_HANDOFF(ifp, m, error);
3396 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3401 if (_IF_QFULL(ifq)) {
3408 ifp->if_obytes += m->m_pkthdr.len + adjust;
3409 if (m->m_flags & (M_BCAST|M_MCAST))
3411 active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3413 _IF_ENQUEUE(ifq, m);
3415 if (ifp != NULL && !active)
3416 (*(ifp)->if_start)(ifp);
3421 if_register_com_alloc(u_char type,
3422 if_com_alloc_t *a, if_com_free_t *f)
3425 KASSERT(if_com_alloc[type] == NULL,
3426 ("if_register_com_alloc: %d already registered", type));
3427 KASSERT(if_com_free[type] == NULL,
3428 ("if_register_com_alloc: %d free already registered", type));
3430 if_com_alloc[type] = a;
3431 if_com_free[type] = f;
3435 if_deregister_com_alloc(u_char type)
3438 KASSERT(if_com_alloc[type] != NULL,
3439 ("if_deregister_com_alloc: %d not registered", type));
3440 KASSERT(if_com_free[type] != NULL,
3441 ("if_deregister_com_alloc: %d free not registered", type));
3442 if_com_alloc[type] = NULL;
3443 if_com_free[type] = NULL;