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>
63 #include <machine/stdarg.h>
67 #include <net/if_arp.h>
68 #include <net/if_clone.h>
69 #include <net/if_dl.h>
70 #include <net/if_types.h>
71 #include <net/if_var.h>
72 #include <net/radix.h>
73 #include <net/route.h>
76 #if defined(INET) || defined(INET6)
77 #include <net/ethernet.h>
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 #include <netinet/ip.h>
81 #include <netinet/ip_carp.h>
83 #include <netinet/if_ether.h>
86 #include <netinet6/in6_var.h>
87 #include <netinet6/in6_ifattach.h>
89 #endif /* INET || INET6 */
91 #include <security/mac/mac_framework.h>
93 #ifdef COMPAT_FREEBSD32
94 #include <sys/mount.h>
95 #include <compat/freebsd32/freebsd32.h>
98 struct ifindex_entry {
99 struct ifnet *ife_ifnet;
102 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
103 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
105 TUNABLE_INT("net.link.ifqmaxlen", &ifqmaxlen);
106 SYSCTL_INT(_net_link, OID_AUTO, ifqmaxlen, CTLFLAG_RDTUN,
107 &ifqmaxlen, 0, "max send queue size");
109 /* Log link state change events */
110 static int log_link_state_change = 1;
112 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
113 &log_link_state_change, 0,
114 "log interface link state change events");
116 /* Interface description */
117 static unsigned int ifdescr_maxlen = 1024;
118 SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW,
120 "administrative maximum length for interface description");
122 static MALLOC_DEFINE(M_IFDESCR, "ifdescr", "ifnet descriptions");
124 /* global sx for non-critical path ifdescr */
125 static struct sx ifdescr_sx;
126 SX_SYSINIT(ifdescr_sx, &ifdescr_sx, "ifnet descr");
128 void (*bridge_linkstate_p)(struct ifnet *ifp);
129 void (*ng_ether_link_state_p)(struct ifnet *ifp, int state);
130 void (*lagg_linkstate_p)(struct ifnet *ifp, int state);
131 /* These are external hooks for CARP. */
132 void (*carp_linkstate_p)(struct ifnet *ifp);
133 void (*carp_demote_adj_p)(int, char *);
134 int (*carp_master_p)(struct ifaddr *);
135 #if defined(INET) || defined(INET6)
136 int (*carp_forus_p)(struct ifnet *ifp, u_char *dhost);
137 int (*carp_output_p)(struct ifnet *ifp, struct mbuf *m,
138 const struct sockaddr *sa);
139 int (*carp_ioctl_p)(struct ifreq *, u_long, struct thread *);
140 int (*carp_attach_p)(struct ifaddr *, int);
141 void (*carp_detach_p)(struct ifaddr *);
144 int (*carp_iamatch_p)(struct ifaddr *, uint8_t **);
147 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6);
148 caddr_t (*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m,
149 const struct in6_addr *taddr);
152 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
155 * XXX: Style; these should be sorted alphabetically, and unprototyped
156 * static functions should be prototyped. Currently they are sorted by
159 static void if_attachdomain(void *);
160 static void if_attachdomain1(struct ifnet *);
161 static int ifconf(u_long, caddr_t);
162 static void if_freemulti(struct ifmultiaddr *);
163 static void if_init(void *);
164 static void if_grow(void);
165 static void if_route(struct ifnet *, int flag, int fam);
166 static int if_setflag(struct ifnet *, int, int, int *, int);
167 static int if_transmit(struct ifnet *ifp, struct mbuf *m);
168 static void if_unroute(struct ifnet *, int flag, int fam);
169 static void link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
170 static int if_rtdel(struct radix_node *, void *);
171 static int ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
172 static int if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
173 static void do_link_state_change(void *, int);
174 static int if_getgroup(struct ifgroupreq *, struct ifnet *);
175 static int if_getgroupmembers(struct ifgroupreq *);
176 static void if_delgroups(struct ifnet *);
177 static void if_attach_internal(struct ifnet *, int);
178 static void if_detach_internal(struct ifnet *, int);
182 * XXX: declare here to avoid to include many inet6 related files..
183 * should be more generalized?
185 extern void nd6_setmtu(struct ifnet *);
188 VNET_DEFINE(int, if_index);
189 int ifqmaxlen = IFQ_MAXLEN;
190 VNET_DEFINE(struct ifnethead, ifnet); /* depend on static init XXX */
191 VNET_DEFINE(struct ifgrouphead, ifg_head);
193 static VNET_DEFINE(int, if_indexlim) = 8;
195 /* Table of ifnet by index. */
196 VNET_DEFINE(struct ifindex_entry *, ifindex_table);
198 #define V_if_indexlim VNET(if_indexlim)
199 #define V_ifindex_table VNET(ifindex_table)
202 * The global network interface list (V_ifnet) and related state (such as
203 * if_index, if_indexlim, and ifindex_table) are protected by an sxlock and
204 * an rwlock. Either may be acquired shared to stablize the list, but both
205 * must be acquired writable to modify the list. This model allows us to
206 * both stablize the interface list during interrupt thread processing, but
207 * also to stablize it over long-running ioctls, without introducing priority
208 * inversions and deadlocks.
210 struct rwlock ifnet_rwlock;
211 struct sx ifnet_sxlock;
214 * The allocation of network interfaces is a rather non-atomic affair; we
215 * need to select an index before we are ready to expose the interface for
216 * use, so will use this pointer value to indicate reservation.
218 #define IFNET_HOLD (void *)(uintptr_t)(-1)
220 static if_com_alloc_t *if_com_alloc[256];
221 static if_com_free_t *if_com_free[256];
223 static MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
224 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
225 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
228 ifnet_byindex_locked(u_short idx)
231 if (idx > V_if_index)
233 if (V_ifindex_table[idx].ife_ifnet == IFNET_HOLD)
235 return (V_ifindex_table[idx].ife_ifnet);
239 ifnet_byindex(u_short idx)
243 IFNET_RLOCK_NOSLEEP();
244 ifp = ifnet_byindex_locked(idx);
245 IFNET_RUNLOCK_NOSLEEP();
250 ifnet_byindex_ref(u_short idx)
254 IFNET_RLOCK_NOSLEEP();
255 ifp = ifnet_byindex_locked(idx);
256 if (ifp == NULL || (ifp->if_flags & IFF_DYING)) {
257 IFNET_RUNLOCK_NOSLEEP();
261 IFNET_RUNLOCK_NOSLEEP();
266 * Allocate an ifindex array entry; return 0 on success or an error on
270 ifindex_alloc_locked(u_short *idxp)
274 IFNET_WLOCK_ASSERT();
278 * Try to find an empty slot below V_if_index. If we fail, take the
281 for (idx = 1; idx <= V_if_index; idx++) {
282 if (V_ifindex_table[idx].ife_ifnet == NULL)
286 /* Catch if_index overflow. */
289 if (idx >= V_if_indexlim) {
293 if (idx > V_if_index)
300 ifindex_free_locked(u_short idx)
303 IFNET_WLOCK_ASSERT();
305 V_ifindex_table[idx].ife_ifnet = NULL;
306 while (V_if_index > 0 &&
307 V_ifindex_table[V_if_index].ife_ifnet == NULL)
312 ifindex_free(u_short idx)
316 ifindex_free_locked(idx);
321 ifnet_setbyindex_locked(u_short idx, struct ifnet *ifp)
324 IFNET_WLOCK_ASSERT();
326 V_ifindex_table[idx].ife_ifnet = ifp;
330 ifnet_setbyindex(u_short idx, struct ifnet *ifp)
334 ifnet_setbyindex_locked(idx, ifp);
339 ifaddr_byindex(u_short idx)
343 IFNET_RLOCK_NOSLEEP();
344 ifa = ifnet_byindex_locked(idx)->if_addr;
347 IFNET_RUNLOCK_NOSLEEP();
352 * Network interface utility routines.
354 * Routines with ifa_ifwith* names take sockaddr *'s as
359 vnet_if_init(const void *unused __unused)
362 TAILQ_INIT(&V_ifnet);
363 TAILQ_INIT(&V_ifg_head);
365 if_grow(); /* create initial table */
367 vnet_if_clone_init();
369 VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init,
374 if_init(void *dummy __unused)
380 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
385 vnet_if_uninit(const void *unused __unused)
388 VNET_ASSERT(TAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p "
389 "not empty", __func__, __LINE__, &V_ifnet));
390 VNET_ASSERT(TAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p "
391 "not empty", __func__, __LINE__, &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 * Allocate a struct ifnet and an index for an interface. A layer 2
426 * common structure will also be allocated if an allocation routine is
427 * registered for the passed type.
430 if_alloc(u_char type)
435 ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
437 if (ifindex_alloc_locked(&idx) != 0) {
442 ifnet_setbyindex_locked(idx, IFNET_HOLD);
446 ifp->if_alloctype = type;
447 if (if_com_alloc[type] != NULL) {
448 ifp->if_l2com = if_com_alloc[type](type, ifp);
449 if (ifp->if_l2com == NULL) {
456 IF_ADDR_LOCK_INIT(ifp);
457 TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
458 ifp->if_afdata_initialized = 0;
459 IF_AFDATA_LOCK_INIT(ifp);
460 TAILQ_INIT(&ifp->if_addrhead);
461 TAILQ_INIT(&ifp->if_multiaddrs);
462 TAILQ_INIT(&ifp->if_groups);
466 ifq_init(&ifp->if_snd, ifp);
468 refcount_init(&ifp->if_refcount, 1); /* Index reference. */
469 ifnet_setbyindex(ifp->if_index, ifp);
474 * Do the actual work of freeing a struct ifnet, and layer 2 common
475 * structure. This call is made when the last reference to an
476 * interface is released.
479 if_free_internal(struct ifnet *ifp)
482 KASSERT((ifp->if_flags & IFF_DYING),
483 ("if_free_internal: interface not dying"));
485 if (if_com_free[ifp->if_alloctype] != NULL)
486 if_com_free[ifp->if_alloctype](ifp->if_l2com,
490 mac_ifnet_destroy(ifp);
492 if (ifp->if_description != NULL)
493 free(ifp->if_description, M_IFDESCR);
494 IF_AFDATA_DESTROY(ifp);
495 IF_ADDR_LOCK_DESTROY(ifp);
496 ifq_delete(&ifp->if_snd);
501 * Deregister an interface and free the associated storage.
504 if_free(struct ifnet *ifp)
507 ifp->if_flags |= IFF_DYING; /* XXX: Locking */
509 CURVNET_SET_QUIET(ifp->if_vnet);
511 KASSERT(ifp == ifnet_byindex_locked(ifp->if_index),
512 ("%s: freeing unallocated ifnet", ifp->if_xname));
514 ifindex_free_locked(ifp->if_index);
517 if (refcount_release(&ifp->if_refcount))
518 if_free_internal(ifp);
523 * Interfaces to keep an ifnet type-stable despite the possibility of the
524 * driver calling if_free(). If there are additional references, we defer
525 * freeing the underlying data structure.
528 if_ref(struct ifnet *ifp)
531 /* We don't assert the ifnet list lock here, but arguably should. */
532 refcount_acquire(&ifp->if_refcount);
536 if_rele(struct ifnet *ifp)
539 if (!refcount_release(&ifp->if_refcount))
541 if_free_internal(ifp);
545 ifq_init(struct ifaltq *ifq, struct ifnet *ifp)
548 mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
550 if (ifq->ifq_maxlen == 0)
551 ifq->ifq_maxlen = ifqmaxlen;
554 ifq->altq_disc = NULL;
555 ifq->altq_flags &= ALTQF_CANTCHANGE;
556 ifq->altq_tbr = NULL;
561 ifq_delete(struct ifaltq *ifq)
563 mtx_destroy(&ifq->ifq_mtx);
567 * Perform generic interface initalization tasks and attach the interface
568 * to the list of "active" interfaces. If vmove flag is set on entry
569 * to if_attach_internal(), perform only a limited subset of initialization
570 * tasks, given that we are moving from one vnet to another an ifnet which
571 * has already been fully initialized.
574 * - The decision to return void and thus require this function to
575 * succeed is questionable.
576 * - We should probably do more sanity checking. For instance we don't
577 * do anything to insure if_xname is unique or non-empty.
580 if_attach(struct ifnet *ifp)
583 if_attach_internal(ifp, 0);
587 if_attach_internal(struct ifnet *ifp, int vmove)
589 unsigned socksize, ifasize;
590 int namelen, masklen;
591 struct sockaddr_dl *sdl;
594 if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
595 panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
599 ifp->if_vnet = curvnet;
600 if (ifp->if_home_vnet == NULL)
601 ifp->if_home_vnet = curvnet;
604 if_addgroup(ifp, IFG_ALL);
606 getmicrotime(&ifp->if_lastchange);
607 ifp->if_data.ifi_epoch = time_uptime;
608 ifp->if_data.ifi_datalen = sizeof(struct if_data);
610 KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
611 (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
612 ("transmit and qflush must both either be set or both be NULL"));
613 if (ifp->if_transmit == NULL) {
614 ifp->if_transmit = if_transmit;
615 ifp->if_qflush = if_qflush;
620 mac_ifnet_create(ifp);
624 * Create a Link Level name for this device.
626 namelen = strlen(ifp->if_xname);
628 * Always save enough space for any possiable name so we
629 * can do a rename in place later.
631 masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
632 socksize = masklen + ifp->if_addrlen;
633 if (socksize < sizeof(*sdl))
634 socksize = sizeof(*sdl);
635 socksize = roundup2(socksize, sizeof(long));
636 ifasize = sizeof(*ifa) + 2 * socksize;
637 ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
639 sdl = (struct sockaddr_dl *)(ifa + 1);
640 sdl->sdl_len = socksize;
641 sdl->sdl_family = AF_LINK;
642 bcopy(ifp->if_xname, sdl->sdl_data, namelen);
643 sdl->sdl_nlen = namelen;
644 sdl->sdl_index = ifp->if_index;
645 sdl->sdl_type = ifp->if_type;
648 ifa->ifa_rtrequest = link_rtrequest;
649 ifa->ifa_addr = (struct sockaddr *)sdl;
650 sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
651 ifa->ifa_netmask = (struct sockaddr *)sdl;
652 sdl->sdl_len = masklen;
654 sdl->sdl_data[--namelen] = 0xff;
655 TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
656 /* Reliably crash if used uninitialized. */
657 ifp->if_broadcastaddr = NULL;
659 #if defined(INET) || defined(INET6)
660 /* Initialize to max value. */
661 if (ifp->if_hw_tsomax == 0)
662 ifp->if_hw_tsomax = min(IP_MAXPACKET, 32 * MCLBYTES -
663 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
664 KASSERT(ifp->if_hw_tsomax <= IP_MAXPACKET &&
665 ifp->if_hw_tsomax >= IP_MAXPACKET / 8,
666 ("%s: tsomax outside of range", __func__));
672 * Update the interface index in the link layer address
675 for (ifa = ifp->if_addr; ifa != NULL;
676 ifa = TAILQ_NEXT(ifa, ifa_link)) {
677 if (ifa->ifa_addr->sa_family == AF_LINK) {
678 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
679 sdl->sdl_index = ifp->if_index;
686 TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
688 curvnet->vnet_ifcnt++;
692 if (domain_init_status >= 2)
693 if_attachdomain1(ifp);
695 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
696 if (IS_DEFAULT_VNET(curvnet))
697 devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
699 /* Announce the interface. */
700 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
704 if_attachdomain(void *dummy)
708 TAILQ_FOREACH(ifp, &V_ifnet, if_link)
709 if_attachdomain1(ifp);
711 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
712 if_attachdomain, NULL);
715 if_attachdomain1(struct ifnet *ifp)
720 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
721 * cannot lock ifp->if_afdata initialization, entirely.
723 if (IF_AFDATA_TRYLOCK(ifp) == 0)
725 if (ifp->if_afdata_initialized >= domain_init_status) {
726 IF_AFDATA_UNLOCK(ifp);
727 log(LOG_WARNING, "%s called more than once on %s\n",
728 __func__, ifp->if_xname);
731 ifp->if_afdata_initialized = domain_init_status;
732 IF_AFDATA_UNLOCK(ifp);
734 /* address family dependent data region */
735 bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
736 for (dp = domains; dp; dp = dp->dom_next) {
737 if (dp->dom_ifattach)
738 ifp->if_afdata[dp->dom_family] =
739 (*dp->dom_ifattach)(ifp);
744 * Remove any unicast or broadcast network addresses from an interface.
747 if_purgeaddrs(struct ifnet *ifp)
749 struct ifaddr *ifa, *next;
751 TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
752 if (ifa->ifa_addr->sa_family == AF_LINK)
755 /* XXX: Ugly!! ad hoc just for INET */
756 if (ifa->ifa_addr->sa_family == AF_INET) {
757 struct ifaliasreq ifr;
759 bzero(&ifr, sizeof(ifr));
760 ifr.ifra_addr = *ifa->ifa_addr;
761 if (ifa->ifa_dstaddr)
762 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
763 if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
769 if (ifa->ifa_addr->sa_family == AF_INET6) {
771 /* ifp_addrhead is already updated */
775 TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
781 * Remove any multicast network addresses from an interface when an ifnet
785 if_purgemaddrs(struct ifnet *ifp)
787 struct ifmultiaddr *ifma;
788 struct ifmultiaddr *next;
791 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
792 if_delmulti_locked(ifp, ifma, 1);
793 IF_ADDR_WUNLOCK(ifp);
797 * Detach an interface, removing it from the list of "active" interfaces.
798 * If vmove flag is set on entry to if_detach_internal(), perform only a
799 * limited subset of cleanup tasks, given that we are moving an ifnet from
800 * one vnet to another, where it must be fully operational.
802 * XXXRW: There are some significant questions about event ordering, and
803 * how to prevent things from starting to use the interface during detach.
806 if_detach(struct ifnet *ifp)
809 CURVNET_SET_QUIET(ifp->if_vnet);
810 if_detach_internal(ifp, 0);
815 if_detach_internal(struct ifnet *ifp, int vmove)
818 struct radix_node_head *rnh;
825 TAILQ_FOREACH(iter, &V_ifnet, if_link)
827 TAILQ_REMOVE(&V_ifnet, ifp, if_link);
833 curvnet->vnet_ifcnt--;
838 panic("%s: ifp=%p not on the ifnet tailq %p",
839 __func__, ifp, &V_ifnet);
841 return; /* XXX this should panic as well? */
845 * Remove/wait for pending events.
847 taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
850 * Remove routes and flush queues.
854 if (ALTQ_IS_ENABLED(&ifp->if_snd))
855 altq_disable(&ifp->if_snd);
856 if (ALTQ_IS_ATTACHED(&ifp->if_snd))
857 altq_detach(&ifp->if_snd);
868 * Remove all IPv6 kernel structs related to ifp. This should be done
869 * before removing routing entries below, since IPv6 interface direct
870 * routes are expected to be removed by the IPv6-specific kernel API.
871 * Otherwise, the kernel will detect some inconsistency and bark it.
879 * Prevent further calls into the device driver via ifnet.
884 * Remove link ifaddr pointer and maybe decrement if_index.
885 * Clean up all addresses.
889 /* We can now free link ifaddr. */
890 if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
891 ifa = TAILQ_FIRST(&ifp->if_addrhead);
892 TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
898 * Delete all remaining routes using this interface
899 * Unfortuneatly the only way to do this is to slog through
900 * the entire routing table looking for routes which point
901 * to this interface...oh well...
903 for (i = 1; i <= AF_MAX; i++) {
904 for (j = 0; j < rt_numfibs; j++) {
905 rnh = rt_tables_get_rnh(j, i);
908 RADIX_NODE_HEAD_LOCK(rnh);
909 (void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
910 RADIX_NODE_HEAD_UNLOCK(rnh);
914 /* Announce that the interface is gone. */
915 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
916 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
917 if (IS_DEFAULT_VNET(curvnet))
918 devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
922 * We cannot hold the lock over dom_ifdetach calls as they might
923 * sleep, for example trying to drain a callout, thus open up the
924 * theoretical race with re-attaching.
927 i = ifp->if_afdata_initialized;
928 ifp->if_afdata_initialized = 0;
929 IF_AFDATA_UNLOCK(ifp);
930 for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
931 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
932 (*dp->dom_ifdetach)(ifp,
933 ifp->if_afdata[dp->dom_family]);
939 * if_vmove() performs a limited version of if_detach() in current
940 * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
941 * An attempt is made to shrink if_index in current vnet, find an
942 * unused if_index in target vnet and calls if_grow() if necessary,
943 * and finally find an unused if_xname for the target vnet.
946 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
951 * Detach from current vnet, but preserve LLADDR info, do not
952 * mark as dead etc. so that the ifnet can be reattached later.
954 if_detach_internal(ifp, 1);
957 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
958 * the if_index for that vnet if possible.
960 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
961 * or we'd lock on one vnet and unlock on another.
964 ifindex_free_locked(ifp->if_index);
968 * Perform interface-specific reassignment tasks, if provided by
971 if (ifp->if_reassign != NULL)
972 ifp->if_reassign(ifp, new_vnet, NULL);
975 * Switch to the context of the target vnet.
977 CURVNET_SET_QUIET(new_vnet);
980 if (ifindex_alloc_locked(&idx) != 0) {
982 panic("if_index overflow");
985 ifnet_setbyindex_locked(ifp->if_index, ifp);
988 if_attach_internal(ifp, 1);
994 * Move an ifnet to or from another child prison/vnet, specified by the jail id.
997 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
1002 /* Try to find the prison within our visibility. */
1003 sx_slock(&allprison_lock);
1004 pr = prison_find_child(td->td_ucred->cr_prison, jid);
1005 sx_sunlock(&allprison_lock);
1008 prison_hold_locked(pr);
1009 mtx_unlock(&pr->pr_mtx);
1011 /* Do not try to move the iface from and to the same prison. */
1012 if (pr->pr_vnet == ifp->if_vnet) {
1017 /* Make sure the named iface does not exists in the dst. prison/vnet. */
1018 /* XXX Lock interfaces to avoid races. */
1019 CURVNET_SET_QUIET(pr->pr_vnet);
1020 difp = ifunit(ifname);
1027 /* Move the interface into the child jail/vnet. */
1028 if_vmove(ifp, pr->pr_vnet);
1030 /* Report the new if_xname back to the userland. */
1031 sprintf(ifname, "%s", ifp->if_xname);
1038 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1041 struct vnet *vnet_dst;
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 /* Make sure the named iface exists in the source prison/vnet. */
1054 CURVNET_SET(pr->pr_vnet);
1055 ifp = ifunit(ifname); /* XXX Lock to avoid races. */
1062 /* Do not try to move the iface from and to the same prison. */
1063 vnet_dst = TD_TO_VNET(td);
1064 if (vnet_dst == ifp->if_vnet) {
1070 /* Get interface back from child jail/vnet. */
1071 if_vmove(ifp, vnet_dst);
1074 /* Report the new if_xname back to the userland. */
1075 sprintf(ifname, "%s", ifp->if_xname);
1083 * Add a group to an interface
1086 if_addgroup(struct ifnet *ifp, const char *groupname)
1088 struct ifg_list *ifgl;
1089 struct ifg_group *ifg = NULL;
1090 struct ifg_member *ifgm;
1093 if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1094 groupname[strlen(groupname) - 1] <= '9')
1098 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1099 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1104 if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1105 M_NOWAIT)) == NULL) {
1110 if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1111 M_TEMP, M_NOWAIT)) == NULL) {
1117 TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1118 if (!strcmp(ifg->ifg_group, groupname))
1122 if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1123 M_TEMP, M_NOWAIT)) == NULL) {
1129 strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1130 ifg->ifg_refcnt = 0;
1131 TAILQ_INIT(&ifg->ifg_members);
1132 TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1137 ifgl->ifgl_group = ifg;
1138 ifgm->ifgm_ifp = ifp;
1141 TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1142 TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1143 IF_ADDR_WUNLOCK(ifp);
1148 EVENTHANDLER_INVOKE(group_attach_event, ifg);
1149 EVENTHANDLER_INVOKE(group_change_event, groupname);
1155 * Remove a group from an interface
1158 if_delgroup(struct ifnet *ifp, const char *groupname)
1160 struct ifg_list *ifgl;
1161 struct ifg_member *ifgm;
1164 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1165 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1173 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1174 IF_ADDR_WUNLOCK(ifp);
1176 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1177 if (ifgm->ifgm_ifp == ifp)
1181 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1185 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1186 TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1188 EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1189 free(ifgl->ifgl_group, M_TEMP);
1195 EVENTHANDLER_INVOKE(group_change_event, groupname);
1201 * Remove an interface from all groups
1204 if_delgroups(struct ifnet *ifp)
1206 struct ifg_list *ifgl;
1207 struct ifg_member *ifgm;
1208 char groupname[IFNAMSIZ];
1211 while (!TAILQ_EMPTY(&ifp->if_groups)) {
1212 ifgl = TAILQ_FIRST(&ifp->if_groups);
1214 strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1217 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1218 IF_ADDR_WUNLOCK(ifp);
1220 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1221 if (ifgm->ifgm_ifp == ifp)
1225 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1230 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1231 TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1233 EVENTHANDLER_INVOKE(group_detach_event,
1235 free(ifgl->ifgl_group, M_TEMP);
1241 EVENTHANDLER_INVOKE(group_change_event, groupname);
1249 * Stores all groups from an interface in memory pointed
1253 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1256 struct ifg_list *ifgl;
1257 struct ifg_req ifgrq, *ifgp;
1258 struct ifgroupreq *ifgr = data;
1260 if (ifgr->ifgr_len == 0) {
1262 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1263 ifgr->ifgr_len += sizeof(struct ifg_req);
1264 IF_ADDR_RUNLOCK(ifp);
1268 len = ifgr->ifgr_len;
1269 ifgp = ifgr->ifgr_groups;
1272 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1273 if (len < sizeof(ifgrq)) {
1274 IF_ADDR_RUNLOCK(ifp);
1277 bzero(&ifgrq, sizeof ifgrq);
1278 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1279 sizeof(ifgrq.ifgrq_group));
1280 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1281 IF_ADDR_RUNLOCK(ifp);
1284 len -= sizeof(ifgrq);
1287 IF_ADDR_RUNLOCK(ifp);
1293 * Stores all members of a group in memory pointed to by data
1296 if_getgroupmembers(struct ifgroupreq *data)
1298 struct ifgroupreq *ifgr = data;
1299 struct ifg_group *ifg;
1300 struct ifg_member *ifgm;
1301 struct ifg_req ifgrq, *ifgp;
1305 TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1306 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1313 if (ifgr->ifgr_len == 0) {
1314 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1315 ifgr->ifgr_len += sizeof(ifgrq);
1320 len = ifgr->ifgr_len;
1321 ifgp = ifgr->ifgr_groups;
1322 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1323 if (len < sizeof(ifgrq)) {
1327 bzero(&ifgrq, sizeof ifgrq);
1328 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1329 sizeof(ifgrq.ifgrq_member));
1330 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1334 len -= sizeof(ifgrq);
1343 * Delete Routes for a Network Interface
1345 * Called for each routing entry via the rnh->rnh_walktree() call above
1346 * to delete all route entries referencing a detaching network interface.
1349 * rn pointer to node in the routing table
1350 * arg argument passed to rnh->rnh_walktree() - detaching interface
1354 * errno failed - reason indicated
1358 if_rtdel(struct radix_node *rn, void *arg)
1360 struct rtentry *rt = (struct rtentry *)rn;
1361 struct ifnet *ifp = arg;
1364 if (rt->rt_ifp == ifp) {
1367 * Protect (sorta) against walktree recursion problems
1368 * with cloned routes
1370 if ((rt->rt_flags & RTF_UP) == 0)
1373 err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1375 rt->rt_flags|RTF_RNH_LOCKED|RTF_PINNED,
1376 (struct rtentry **) NULL, rt->rt_fibnum);
1378 log(LOG_WARNING, "if_rtdel: error %d\n", err);
1386 * Wrapper functions for struct ifnet address list locking macros. These are
1387 * used by kernel modules to avoid encoding programming interface or binary
1388 * interface assumptions that may be violated when kernel-internal locking
1389 * approaches change.
1392 if_addr_rlock(struct ifnet *ifp)
1399 if_addr_runlock(struct ifnet *ifp)
1402 IF_ADDR_RUNLOCK(ifp);
1406 if_maddr_rlock(struct ifnet *ifp)
1413 if_maddr_runlock(struct ifnet *ifp)
1416 IF_ADDR_RUNLOCK(ifp);
1420 * Initialization, destruction and refcounting functions for ifaddrs.
1423 ifa_init(struct ifaddr *ifa)
1426 mtx_init(&ifa->ifa_mtx, "ifaddr", NULL, MTX_DEF);
1427 refcount_init(&ifa->ifa_refcnt, 1);
1428 ifa->if_data.ifi_datalen = sizeof(ifa->if_data);
1432 ifa_ref(struct ifaddr *ifa)
1435 refcount_acquire(&ifa->ifa_refcnt);
1439 ifa_free(struct ifaddr *ifa)
1442 if (refcount_release(&ifa->ifa_refcnt)) {
1443 mtx_destroy(&ifa->ifa_mtx);
1444 free(ifa, M_IFADDR);
1449 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1452 struct rtentry *rt = NULL;
1453 struct rt_addrinfo info;
1454 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1456 bzero(&info, sizeof(info));
1457 info.rti_ifp = V_loif;
1458 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1459 info.rti_info[RTAX_DST] = ia;
1460 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1461 error = rtrequest1_fib(RTM_ADD, &info, &rt, ifa->ifa_ifp->if_fib);
1463 if (error == 0 && rt != NULL) {
1465 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1466 ifa->ifa_ifp->if_type;
1467 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1468 ifa->ifa_ifp->if_index;
1471 } else if (error != 0)
1472 log(LOG_DEBUG, "%s: insertion failed: %u\n", __func__, error);
1478 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1481 struct rt_addrinfo info;
1482 struct sockaddr_dl null_sdl;
1484 bzero(&null_sdl, sizeof(null_sdl));
1485 null_sdl.sdl_len = sizeof(null_sdl);
1486 null_sdl.sdl_family = AF_LINK;
1487 null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1488 null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1489 bzero(&info, sizeof(info));
1490 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1491 info.rti_info[RTAX_DST] = ia;
1492 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1493 error = rtrequest1_fib(RTM_DELETE, &info, NULL, ifa->ifa_ifp->if_fib);
1496 log(LOG_DEBUG, "%s: deletion failed: %u\n", __func__, error);
1502 * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1503 * structs used to represent other address families, it is necessary
1504 * to perform a different comparison.
1507 #define sa_equal(a1, a2) \
1508 (bcmp((a1), (a2), ((a1))->sa_len) == 0)
1510 #define sa_dl_equal(a1, a2) \
1511 ((((struct sockaddr_dl *)(a1))->sdl_len == \
1512 ((struct sockaddr_dl *)(a2))->sdl_len) && \
1513 (bcmp(LLADDR((struct sockaddr_dl *)(a1)), \
1514 LLADDR((struct sockaddr_dl *)(a2)), \
1515 ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1518 * Locate an interface based on a complete address.
1521 static struct ifaddr *
1522 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1527 IFNET_RLOCK_NOSLEEP();
1528 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1530 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1531 if (ifa->ifa_addr->sa_family != addr->sa_family)
1533 if (sa_equal(addr, ifa->ifa_addr)) {
1536 IF_ADDR_RUNLOCK(ifp);
1539 /* IP6 doesn't have broadcast */
1540 if ((ifp->if_flags & IFF_BROADCAST) &&
1541 ifa->ifa_broadaddr &&
1542 ifa->ifa_broadaddr->sa_len != 0 &&
1543 sa_equal(ifa->ifa_broadaddr, addr)) {
1546 IF_ADDR_RUNLOCK(ifp);
1550 IF_ADDR_RUNLOCK(ifp);
1554 IFNET_RUNLOCK_NOSLEEP();
1559 ifa_ifwithaddr(struct sockaddr *addr)
1562 return (ifa_ifwithaddr_internal(addr, 1));
1566 ifa_ifwithaddr_check(struct sockaddr *addr)
1569 return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1573 * Locate an interface based on the broadcast address.
1577 ifa_ifwithbroadaddr(struct sockaddr *addr)
1582 IFNET_RLOCK_NOSLEEP();
1583 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1585 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1586 if (ifa->ifa_addr->sa_family != addr->sa_family)
1588 if ((ifp->if_flags & IFF_BROADCAST) &&
1589 ifa->ifa_broadaddr &&
1590 ifa->ifa_broadaddr->sa_len != 0 &&
1591 sa_equal(ifa->ifa_broadaddr, addr)) {
1593 IF_ADDR_RUNLOCK(ifp);
1597 IF_ADDR_RUNLOCK(ifp);
1601 IFNET_RUNLOCK_NOSLEEP();
1606 * Locate the point to point interface with a given destination address.
1610 ifa_ifwithdstaddr_fib(struct sockaddr *addr, int fibnum)
1615 IFNET_RLOCK_NOSLEEP();
1616 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1617 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1619 if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum))
1622 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1623 if (ifa->ifa_addr->sa_family != addr->sa_family)
1625 if (ifa->ifa_dstaddr != NULL &&
1626 sa_equal(addr, ifa->ifa_dstaddr)) {
1628 IF_ADDR_RUNLOCK(ifp);
1632 IF_ADDR_RUNLOCK(ifp);
1636 IFNET_RUNLOCK_NOSLEEP();
1641 ifa_ifwithdstaddr(struct sockaddr *addr)
1644 return (ifa_ifwithdstaddr_fib(addr, RT_ALL_FIBS));
1648 * Find an interface on a specific network. If many, choice
1649 * is most specific found.
1652 ifa_ifwithnet_fib(struct sockaddr *addr, int ignore_ptp, int fibnum)
1656 struct ifaddr *ifa_maybe = NULL;
1657 u_int af = addr->sa_family;
1658 char *addr_data = addr->sa_data, *cplim;
1661 * AF_LINK addresses can be looked up directly by their index number,
1662 * so do that if we can.
1664 if (af == AF_LINK) {
1665 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1666 if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1667 return (ifaddr_byindex(sdl->sdl_index));
1671 * Scan though each interface, looking for ones that have addresses
1672 * in this address family and the requested fib. Maintain a reference
1673 * on ifa_maybe once we find one, as we release the IF_ADDR_RLOCK() that
1674 * kept it stable when we move onto the next interface.
1676 IFNET_RLOCK_NOSLEEP();
1677 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1678 if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum))
1681 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1682 char *cp, *cp2, *cp3;
1684 if (ifa->ifa_addr->sa_family != af)
1686 if (af == AF_INET &&
1687 ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1689 * This is a bit broken as it doesn't
1690 * take into account that the remote end may
1691 * be a single node in the network we are
1693 * The trouble is that we don't know the
1694 * netmask for the remote end.
1696 if (ifa->ifa_dstaddr != NULL &&
1697 sa_equal(addr, ifa->ifa_dstaddr)) {
1699 IF_ADDR_RUNLOCK(ifp);
1704 * if we have a special address handler,
1705 * then use it instead of the generic one.
1707 if (ifa->ifa_claim_addr) {
1708 if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1710 IF_ADDR_RUNLOCK(ifp);
1717 * Scan all the bits in the ifa's address.
1718 * If a bit dissagrees with what we are
1719 * looking for, mask it with the netmask
1720 * to see if it really matters.
1721 * (A byte at a time)
1723 if (ifa->ifa_netmask == 0)
1726 cp2 = ifa->ifa_addr->sa_data;
1727 cp3 = ifa->ifa_netmask->sa_data;
1728 cplim = ifa->ifa_netmask->sa_len
1729 + (char *)ifa->ifa_netmask;
1731 if ((*cp++ ^ *cp2++) & *cp3++)
1732 goto next; /* next address! */
1734 * If the netmask of what we just found
1735 * is more specific than what we had before
1736 * (if we had one), or if the virtual status
1737 * of new prefix is better than of the old one,
1738 * then remember the new one before continuing
1739 * to search for an even better one.
1741 if (ifa_maybe == NULL ||
1742 ifa_preferred(ifa_maybe, ifa) ||
1743 rn_refines((caddr_t)ifa->ifa_netmask,
1744 (caddr_t)ifa_maybe->ifa_netmask)) {
1745 if (ifa_maybe != NULL)
1746 ifa_free(ifa_maybe);
1752 IF_ADDR_RUNLOCK(ifp);
1757 IFNET_RUNLOCK_NOSLEEP();
1758 if (ifa_maybe != NULL)
1759 ifa_free(ifa_maybe);
1764 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp)
1767 return (ifa_ifwithnet_fib(addr, ignore_ptp, RT_ALL_FIBS));
1771 * Find an interface address specific to an interface best matching
1775 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1778 char *cp, *cp2, *cp3;
1780 struct ifaddr *ifa_maybe = NULL;
1781 u_int af = addr->sa_family;
1786 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1787 if (ifa->ifa_addr->sa_family != af)
1789 if (ifa_maybe == NULL)
1791 if (ifa->ifa_netmask == 0) {
1792 if (sa_equal(addr, ifa->ifa_addr) ||
1793 (ifa->ifa_dstaddr &&
1794 sa_equal(addr, ifa->ifa_dstaddr)))
1798 if (ifp->if_flags & IFF_POINTOPOINT) {
1799 if (sa_equal(addr, ifa->ifa_dstaddr))
1803 cp2 = ifa->ifa_addr->sa_data;
1804 cp3 = ifa->ifa_netmask->sa_data;
1805 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1806 for (; cp3 < cplim; cp3++)
1807 if ((*cp++ ^ *cp2++) & *cp3)
1817 IF_ADDR_RUNLOCK(ifp);
1822 * See whether new ifa is better than current one:
1823 * 1) A non-virtual one is preferred over virtual.
1824 * 2) A virtual in master state preferred over any other state.
1826 * Used in several address selecting functions.
1829 ifa_preferred(struct ifaddr *cur, struct ifaddr *next)
1832 return (cur->ifa_carp && (!next->ifa_carp ||
1833 ((*carp_master_p)(next) && !(*carp_master_p)(cur))));
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
1870 if_unroute(struct ifnet *ifp, int flag, int fam)
1874 KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1876 ifp->if_flags &= ~flag;
1877 getmicrotime(&ifp->if_lastchange);
1878 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1879 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1880 pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1881 ifp->if_qflush(ifp);
1884 (*carp_linkstate_p)(ifp);
1889 * Mark an interface up and notify protocols of
1893 if_route(struct ifnet *ifp, int flag, int fam)
1897 KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1899 ifp->if_flags |= flag;
1900 getmicrotime(&ifp->if_lastchange);
1901 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1902 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1903 pfctlinput(PRC_IFUP, ifa->ifa_addr);
1905 (*carp_linkstate_p)(ifp);
1912 void (*vlan_link_state_p)(struct ifnet *); /* XXX: private from if_vlan */
1913 void (*vlan_trunk_cap_p)(struct ifnet *); /* XXX: private from if_vlan */
1914 struct ifnet *(*vlan_trunkdev_p)(struct ifnet *);
1915 struct ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t);
1916 int (*vlan_tag_p)(struct ifnet *, uint16_t *);
1917 int (*vlan_setcookie_p)(struct ifnet *, void *);
1918 void *(*vlan_cookie_p)(struct ifnet *);
1921 * Handle a change in the interface link state. To avoid LORs
1922 * between driver lock and upper layer locks, as well as possible
1923 * recursions, we post event to taskqueue, and all job
1924 * is done in static do_link_state_change().
1927 if_link_state_change(struct ifnet *ifp, int link_state)
1929 /* Return if state hasn't changed. */
1930 if (ifp->if_link_state == link_state)
1933 ifp->if_link_state = link_state;
1935 taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1939 do_link_state_change(void *arg, int pending)
1941 struct ifnet *ifp = (struct ifnet *)arg;
1942 int link_state = ifp->if_link_state;
1943 CURVNET_SET(ifp->if_vnet);
1945 /* Notify that the link state has changed. */
1947 if (ifp->if_vlantrunk != NULL)
1948 (*vlan_link_state_p)(ifp);
1950 if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1951 IFP2AC(ifp)->ac_netgraph != NULL)
1952 (*ng_ether_link_state_p)(ifp, link_state);
1954 (*carp_linkstate_p)(ifp);
1956 (*bridge_linkstate_p)(ifp);
1958 (*lagg_linkstate_p)(ifp, link_state);
1960 if (IS_DEFAULT_VNET(curvnet))
1961 devctl_notify("IFNET", ifp->if_xname,
1962 (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
1965 if_printf(ifp, "%d link states coalesced\n", pending);
1966 if (log_link_state_change)
1967 log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1968 (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1969 EVENTHANDLER_INVOKE(ifnet_link_event, ifp, ifp->if_link_state);
1974 * Mark an interface down and notify protocols of
1978 if_down(struct ifnet *ifp)
1981 if_unroute(ifp, IFF_UP, AF_UNSPEC);
1985 * Mark an interface up and notify protocols of
1989 if_up(struct ifnet *ifp)
1992 if_route(ifp, IFF_UP, AF_UNSPEC);
1996 * Flush an interface queue.
1999 if_qflush(struct ifnet *ifp)
2007 if (ALTQ_IS_ENABLED(ifq))
2011 while ((m = n) != 0) {
2022 * Map interface name to interface structure pointer, with or without
2023 * returning a reference.
2026 ifunit_ref(const char *name)
2030 IFNET_RLOCK_NOSLEEP();
2031 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2032 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2033 !(ifp->if_flags & IFF_DYING))
2038 IFNET_RUNLOCK_NOSLEEP();
2043 ifunit(const char *name)
2047 IFNET_RLOCK_NOSLEEP();
2048 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2049 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2052 IFNET_RUNLOCK_NOSLEEP();
2057 * Hardware specific interface ioctls.
2060 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2065 int new_flags, temp_flags;
2066 size_t namelen, onamelen;
2068 char *descrbuf, *odescrbuf;
2069 char new_name[IFNAMSIZ];
2071 struct sockaddr_dl *sdl;
2073 ifr = (struct ifreq *)data;
2076 ifr->ifr_index = ifp->if_index;
2080 temp_flags = ifp->if_flags | ifp->if_drv_flags;
2081 ifr->ifr_flags = temp_flags & 0xffff;
2082 ifr->ifr_flagshigh = temp_flags >> 16;
2086 ifr->ifr_reqcap = ifp->if_capabilities;
2087 ifr->ifr_curcap = ifp->if_capenable;
2092 error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2097 ifr->ifr_metric = ifp->if_metric;
2101 ifr->ifr_mtu = ifp->if_mtu;
2105 ifr->ifr_phys = ifp->if_physical;
2110 sx_slock(&ifdescr_sx);
2111 if (ifp->if_description == NULL)
2114 /* space for terminating nul */
2115 descrlen = strlen(ifp->if_description) + 1;
2116 if (ifr->ifr_buffer.length < descrlen)
2117 ifr->ifr_buffer.buffer = NULL;
2119 error = copyout(ifp->if_description,
2120 ifr->ifr_buffer.buffer, descrlen);
2121 ifr->ifr_buffer.length = descrlen;
2123 sx_sunlock(&ifdescr_sx);
2127 error = priv_check(td, PRIV_NET_SETIFDESCR);
2132 * Copy only (length-1) bytes to make sure that
2133 * if_description is always nul terminated. The
2134 * length parameter is supposed to count the
2135 * terminating nul in.
2137 if (ifr->ifr_buffer.length > ifdescr_maxlen)
2138 return (ENAMETOOLONG);
2139 else if (ifr->ifr_buffer.length == 0)
2142 descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2144 error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2145 ifr->ifr_buffer.length - 1);
2147 free(descrbuf, M_IFDESCR);
2152 sx_xlock(&ifdescr_sx);
2153 odescrbuf = ifp->if_description;
2154 ifp->if_description = descrbuf;
2155 sx_xunlock(&ifdescr_sx);
2157 getmicrotime(&ifp->if_lastchange);
2158 free(odescrbuf, M_IFDESCR);
2162 ifr->ifr_fib = ifp->if_fib;
2166 error = priv_check(td, PRIV_NET_SETIFFIB);
2169 if (ifr->ifr_fib >= rt_numfibs)
2172 ifp->if_fib = ifr->ifr_fib;
2176 error = priv_check(td, PRIV_NET_SETIFFLAGS);
2180 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2181 * check, so we don't need special handling here yet.
2183 new_flags = (ifr->ifr_flags & 0xffff) |
2184 (ifr->ifr_flagshigh << 16);
2185 if (ifp->if_flags & IFF_SMART) {
2186 /* Smart drivers twiddle their own routes */
2187 } else if (ifp->if_flags & IFF_UP &&
2188 (new_flags & IFF_UP) == 0) {
2190 } else if (new_flags & IFF_UP &&
2191 (ifp->if_flags & IFF_UP) == 0) {
2194 /* See if permanently promiscuous mode bit is about to flip */
2195 if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2196 if (new_flags & IFF_PPROMISC)
2197 ifp->if_flags |= IFF_PROMISC;
2198 else if (ifp->if_pcount == 0)
2199 ifp->if_flags &= ~IFF_PROMISC;
2200 log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2202 (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
2204 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2205 (new_flags &~ IFF_CANTCHANGE);
2206 if (ifp->if_ioctl) {
2207 (void) (*ifp->if_ioctl)(ifp, cmd, data);
2209 getmicrotime(&ifp->if_lastchange);
2213 error = priv_check(td, PRIV_NET_SETIFCAP);
2216 if (ifp->if_ioctl == NULL)
2217 return (EOPNOTSUPP);
2218 if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2220 error = (*ifp->if_ioctl)(ifp, cmd, data);
2222 getmicrotime(&ifp->if_lastchange);
2227 error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2232 error = priv_check(td, PRIV_NET_SETIFNAME);
2235 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2238 if (new_name[0] == '\0')
2240 if (ifunit(new_name) != NULL)
2244 * XXX: Locking. Nothing else seems to lock if_flags,
2245 * and there are numerous other races with the
2246 * ifunit() checks not being atomic with namespace
2247 * changes (renames, vmoves, if_attach, etc).
2249 ifp->if_flags |= IFF_RENAMING;
2251 /* Announce the departure of the interface. */
2252 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2253 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2255 log(LOG_INFO, "%s: changing name to '%s'\n",
2256 ifp->if_xname, new_name);
2258 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2261 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2262 namelen = strlen(new_name);
2263 onamelen = sdl->sdl_nlen;
2265 * Move the address if needed. This is safe because we
2266 * allocate space for a name of length IFNAMSIZ when we
2267 * create this in if_attach().
2269 if (namelen != onamelen) {
2270 bcopy(sdl->sdl_data + onamelen,
2271 sdl->sdl_data + namelen, sdl->sdl_alen);
2273 bcopy(new_name, sdl->sdl_data, namelen);
2274 sdl->sdl_nlen = namelen;
2275 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2276 bzero(sdl->sdl_data, onamelen);
2277 while (namelen != 0)
2278 sdl->sdl_data[--namelen] = 0xff;
2281 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2282 /* Announce the return of the interface. */
2283 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2285 ifp->if_flags &= ~IFF_RENAMING;
2290 error = priv_check(td, PRIV_NET_SETIFVNET);
2293 error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2298 error = priv_check(td, PRIV_NET_SETIFMETRIC);
2301 ifp->if_metric = ifr->ifr_metric;
2302 getmicrotime(&ifp->if_lastchange);
2306 error = priv_check(td, PRIV_NET_SETIFPHYS);
2309 if (ifp->if_ioctl == NULL)
2310 return (EOPNOTSUPP);
2311 error = (*ifp->if_ioctl)(ifp, cmd, data);
2313 getmicrotime(&ifp->if_lastchange);
2318 u_long oldmtu = ifp->if_mtu;
2320 error = priv_check(td, PRIV_NET_SETIFMTU);
2323 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2325 if (ifp->if_ioctl == NULL)
2326 return (EOPNOTSUPP);
2327 error = (*ifp->if_ioctl)(ifp, cmd, data);
2329 getmicrotime(&ifp->if_lastchange);
2333 * If the link MTU changed, do network layer specific procedure.
2335 if (ifp->if_mtu != oldmtu) {
2345 if (cmd == SIOCADDMULTI)
2346 error = priv_check(td, PRIV_NET_ADDMULTI);
2348 error = priv_check(td, PRIV_NET_DELMULTI);
2352 /* Don't allow group membership on non-multicast interfaces. */
2353 if ((ifp->if_flags & IFF_MULTICAST) == 0)
2354 return (EOPNOTSUPP);
2356 /* Don't let users screw up protocols' entries. */
2357 if (ifr->ifr_addr.sa_family != AF_LINK)
2360 if (cmd == SIOCADDMULTI) {
2361 struct ifmultiaddr *ifma;
2364 * Userland is only permitted to join groups once
2365 * via the if_addmulti() KPI, because it cannot hold
2366 * struct ifmultiaddr * between calls. It may also
2367 * lose a race while we check if the membership
2371 ifma = if_findmulti(ifp, &ifr->ifr_addr);
2372 IF_ADDR_RUNLOCK(ifp);
2376 error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2378 error = if_delmulti(ifp, &ifr->ifr_addr);
2381 getmicrotime(&ifp->if_lastchange);
2384 case SIOCSIFPHYADDR:
2385 case SIOCDIFPHYADDR:
2387 case SIOCSIFPHYADDR_IN6:
2389 case SIOCSLIFPHYADDR:
2391 case SIOCSIFGENERIC:
2392 error = priv_check(td, PRIV_NET_HWIOCTL);
2395 if (ifp->if_ioctl == NULL)
2396 return (EOPNOTSUPP);
2397 error = (*ifp->if_ioctl)(ifp, cmd, data);
2399 getmicrotime(&ifp->if_lastchange);
2403 ifs = (struct ifstat *)data;
2404 ifs->ascii[0] = '\0';
2406 case SIOCGIFPSRCADDR:
2407 case SIOCGIFPDSTADDR:
2408 case SIOCGLIFPHYADDR:
2410 case SIOCGIFGENERIC:
2411 if (ifp->if_ioctl == NULL)
2412 return (EOPNOTSUPP);
2413 error = (*ifp->if_ioctl)(ifp, cmd, data);
2417 error = priv_check(td, PRIV_NET_SETLLADDR);
2420 error = if_setlladdr(ifp,
2421 ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2422 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2427 struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2429 error = priv_check(td, PRIV_NET_ADDIFGROUP);
2432 if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2438 if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2444 struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2446 error = priv_check(td, PRIV_NET_DELIFGROUP);
2449 if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2461 #ifdef COMPAT_FREEBSD32
2469 #define SIOCGIFCONF32 _IOWR('i', 36, struct ifconf32)
2476 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2483 CURVNET_SET(so->so_vnet);
2487 error = ifconf(cmd, data);
2491 #ifdef COMPAT_FREEBSD32
2494 struct ifconf32 *ifc32;
2497 ifc32 = (struct ifconf32 *)data;
2498 ifc.ifc_len = ifc32->ifc_len;
2499 ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2501 error = ifconf(SIOCGIFCONF, (void *)&ifc);
2504 ifc32->ifc_len = ifc.ifc_len;
2509 ifr = (struct ifreq *)data;
2514 error = priv_check(td, PRIV_NET_SETIFVNET);
2516 error = if_vmove_reclaim(td, ifr->ifr_name,
2523 error = priv_check(td, PRIV_NET_IFCREATE);
2525 error = if_clone_create(ifr->ifr_name,
2526 sizeof(ifr->ifr_name),
2527 cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
2531 error = priv_check(td, PRIV_NET_IFDESTROY);
2533 error = if_clone_destroy(ifr->ifr_name);
2537 case SIOCIFGCLONERS:
2538 error = if_clone_list((struct if_clonereq *)data);
2542 error = if_getgroupmembers((struct ifgroupreq *)data);
2545 #if defined(INET) || defined(INET6)
2548 if (carp_ioctl_p == NULL)
2549 error = EPROTONOSUPPORT;
2551 error = (*carp_ioctl_p)(ifr, cmd, td);
2557 ifp = ifunit_ref(ifr->ifr_name);
2563 error = ifhwioctl(cmd, ifp, data, td);
2564 if (error != ENOIOCTL) {
2570 oif_flags = ifp->if_flags;
2571 if (so->so_proto == NULL) {
2574 return (EOPNOTSUPP);
2578 * Pass the request on to the socket control method, and if the
2579 * latter returns EOPNOTSUPP, directly to the interface.
2581 * Make an exception for the legacy SIOCSIF* requests. Drivers
2582 * trust SIOCSIFADDR et al to come from an already privileged
2583 * layer, and do not perform any credentials checks or input
2587 error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
2590 if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL &&
2591 cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR &&
2592 cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK)
2593 error = (*ifp->if_ioctl)(ifp, cmd, data);
2600 case SIOCSIFDSTADDR:
2602 case SIOCSIFBRDADDR:
2603 case SIOCSIFNETMASK:
2604 #if BYTE_ORDER != BIG_ENDIAN
2605 if (ifr->ifr_addr.sa_family == 0 &&
2606 ifr->ifr_addr.sa_len < 16) {
2607 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2608 ifr->ifr_addr.sa_len = 16;
2611 if (ifr->ifr_addr.sa_len == 0)
2612 ifr->ifr_addr.sa_len = 16;
2620 case OSIOCGIFDSTADDR:
2621 cmd = SIOCGIFDSTADDR;
2624 case OSIOCGIFBRDADDR:
2625 cmd = SIOCGIFBRDADDR;
2628 case OSIOCGIFNETMASK:
2629 cmd = SIOCGIFNETMASK;
2631 error = ((*so->so_proto->pr_usrreqs->pru_control)(so,
2635 if (error == EOPNOTSUPP && ifp != NULL &&
2636 ifp->if_ioctl != NULL &&
2637 cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR &&
2638 cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK)
2639 error = (*ifp->if_ioctl)(ifp, cmd, data);
2643 case OSIOCGIFDSTADDR:
2644 case OSIOCGIFBRDADDR:
2645 case OSIOCGIFNETMASK:
2646 *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2650 #endif /* COMPAT_43 */
2652 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2654 if (ifp->if_flags & IFF_UP)
2664 * The code common to handling reference counted flags,
2665 * e.g., in ifpromisc() and if_allmulti().
2666 * The "pflag" argument can specify a permanent mode flag to check,
2667 * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2669 * Only to be used on stack-owned flags, not driver-owned flags.
2672 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2676 int oldflags, oldcount;
2678 /* Sanity checks to catch programming errors */
2679 KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2680 ("%s: setting driver-owned flag %d", __func__, flag));
2683 KASSERT(*refcount >= 0,
2684 ("%s: increment negative refcount %d for flag %d",
2685 __func__, *refcount, flag));
2687 KASSERT(*refcount > 0,
2688 ("%s: decrement non-positive refcount %d for flag %d",
2689 __func__, *refcount, flag));
2691 /* In case this mode is permanent, just touch refcount */
2692 if (ifp->if_flags & pflag) {
2693 *refcount += onswitch ? 1 : -1;
2697 /* Save ifnet parameters for if_ioctl() may fail */
2698 oldcount = *refcount;
2699 oldflags = ifp->if_flags;
2702 * See if we aren't the only and touching refcount is enough.
2703 * Actually toggle interface flag if we are the first or last.
2708 ifp->if_flags |= flag;
2712 ifp->if_flags &= ~flag;
2715 /* Call down the driver since we've changed interface flags */
2716 if (ifp->if_ioctl == NULL) {
2720 ifr.ifr_flags = ifp->if_flags & 0xffff;
2721 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2722 error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2725 /* Notify userland that interface flags have changed */
2730 /* Recover after driver error */
2731 *refcount = oldcount;
2732 ifp->if_flags = oldflags;
2737 * Set/clear promiscuous mode on interface ifp based on the truth value
2738 * of pswitch. The calls are reference counted so that only the first
2739 * "on" request actually has an effect, as does the final "off" request.
2740 * Results are undefined if the "off" and "on" requests are not matched.
2743 ifpromisc(struct ifnet *ifp, int pswitch)
2746 int oldflags = ifp->if_flags;
2748 error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2749 &ifp->if_pcount, pswitch);
2750 /* If promiscuous mode status has changed, log a message */
2751 if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2752 log(LOG_INFO, "%s: promiscuous mode %s\n",
2754 (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2759 * Return interface configuration
2760 * of system. List may be used
2761 * in later ioctl's (above) to get
2762 * other information.
2766 ifconf(u_long cmd, caddr_t data)
2768 struct ifconf *ifc = (struct ifconf *)data;
2773 int error, full = 0, valid_len, max_len;
2775 /* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2776 max_len = MAXPHYS - 1;
2778 /* Prevent hostile input from being able to crash the system */
2779 if (ifc->ifc_len <= 0)
2783 if (ifc->ifc_len <= max_len) {
2784 max_len = ifc->ifc_len;
2787 sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2792 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2796 * Zero the ifr_name buffer to make sure we don't
2797 * disclose the contents of the stack.
2799 memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2801 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2802 >= sizeof(ifr.ifr_name)) {
2805 return (ENAMETOOLONG);
2810 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2811 struct sockaddr *sa = ifa->ifa_addr;
2813 if (prison_if(curthread->td_ucred, sa) != 0)
2817 if (cmd == OSIOCGIFCONF) {
2818 struct osockaddr *osa =
2819 (struct osockaddr *)&ifr.ifr_addr;
2821 osa->sa_family = sa->sa_family;
2822 sbuf_bcat(sb, &ifr, sizeof(ifr));
2823 max_len += sizeof(ifr);
2826 if (sa->sa_len <= sizeof(*sa)) {
2828 sbuf_bcat(sb, &ifr, sizeof(ifr));
2829 max_len += sizeof(ifr);
2832 offsetof(struct ifreq, ifr_addr));
2833 max_len += offsetof(struct ifreq, ifr_addr);
2834 sbuf_bcat(sb, sa, sa->sa_len);
2835 max_len += sa->sa_len;
2838 if (sbuf_error(sb) == 0)
2839 valid_len = sbuf_len(sb);
2841 IF_ADDR_RUNLOCK(ifp);
2843 bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2844 sbuf_bcat(sb, &ifr, sizeof(ifr));
2845 max_len += sizeof(ifr);
2847 if (sbuf_error(sb) == 0)
2848 valid_len = sbuf_len(sb);
2854 * If we didn't allocate enough space (uncommon), try again. If
2855 * we have already allocated as much space as we are allowed,
2856 * return what we've got.
2858 if (valid_len != max_len && !full) {
2863 ifc->ifc_len = valid_len;
2865 error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2871 * Just like ifpromisc(), but for all-multicast-reception mode.
2874 if_allmulti(struct ifnet *ifp, int onswitch)
2877 return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2880 struct ifmultiaddr *
2881 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2883 struct ifmultiaddr *ifma;
2885 IF_ADDR_LOCK_ASSERT(ifp);
2887 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2888 if (sa->sa_family == AF_LINK) {
2889 if (sa_dl_equal(ifma->ifma_addr, sa))
2892 if (sa_equal(ifma->ifma_addr, sa))
2901 * Allocate a new ifmultiaddr and initialize based on passed arguments. We
2902 * make copies of passed sockaddrs. The ifmultiaddr will not be added to
2903 * the ifnet multicast address list here, so the caller must do that and
2904 * other setup work (such as notifying the device driver). The reference
2905 * count is initialized to 1.
2907 static struct ifmultiaddr *
2908 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2911 struct ifmultiaddr *ifma;
2912 struct sockaddr *dupsa;
2914 ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2919 dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2920 if (dupsa == NULL) {
2921 free(ifma, M_IFMADDR);
2924 bcopy(sa, dupsa, sa->sa_len);
2925 ifma->ifma_addr = dupsa;
2927 ifma->ifma_ifp = ifp;
2928 ifma->ifma_refcount = 1;
2929 ifma->ifma_protospec = NULL;
2932 ifma->ifma_lladdr = NULL;
2936 dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
2937 if (dupsa == NULL) {
2938 free(ifma->ifma_addr, M_IFMADDR);
2939 free(ifma, M_IFMADDR);
2942 bcopy(llsa, dupsa, llsa->sa_len);
2943 ifma->ifma_lladdr = dupsa;
2949 * if_freemulti: free ifmultiaddr structure and possibly attached related
2950 * addresses. The caller is responsible for implementing reference
2951 * counting, notifying the driver, handling routing messages, and releasing
2952 * any dependent link layer state.
2955 if_freemulti(struct ifmultiaddr *ifma)
2958 KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2959 ifma->ifma_refcount));
2960 KASSERT(ifma->ifma_protospec == NULL,
2961 ("if_freemulti: protospec not NULL"));
2963 if (ifma->ifma_lladdr != NULL)
2964 free(ifma->ifma_lladdr, M_IFMADDR);
2965 free(ifma->ifma_addr, M_IFMADDR);
2966 free(ifma, M_IFMADDR);
2970 * Register an additional multicast address with a network interface.
2972 * - If the address is already present, bump the reference count on the
2973 * address and return.
2974 * - If the address is not link-layer, look up a link layer address.
2975 * - Allocate address structures for one or both addresses, and attach to the
2976 * multicast address list on the interface. If automatically adding a link
2977 * layer address, the protocol address will own a reference to the link
2978 * layer address, to be freed when it is freed.
2979 * - Notify the network device driver of an addition to the multicast address
2982 * 'sa' points to caller-owned memory with the desired multicast address.
2984 * 'retifma' will be used to return a pointer to the resulting multicast
2985 * address reference, if desired.
2988 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2989 struct ifmultiaddr **retifma)
2991 struct ifmultiaddr *ifma, *ll_ifma;
2992 struct sockaddr *llsa;
2996 * If the address is already present, return a new reference to it;
2997 * otherwise, allocate storage and set up a new address.
3000 ifma = if_findmulti(ifp, sa);
3002 ifma->ifma_refcount++;
3003 if (retifma != NULL)
3005 IF_ADDR_WUNLOCK(ifp);
3010 * The address isn't already present; resolve the protocol address
3011 * into a link layer address, and then look that up, bump its
3012 * refcount or allocate an ifma for that also. If 'llsa' was
3013 * returned, we will need to free it later.
3017 if (ifp->if_resolvemulti != NULL) {
3018 error = ifp->if_resolvemulti(ifp, &llsa, sa);
3024 * Allocate the new address. Don't hook it up yet, as we may also
3025 * need to allocate a link layer multicast address.
3027 ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
3034 * If a link layer address is found, we'll need to see if it's
3035 * already present in the address list, or allocate is as well.
3036 * When this block finishes, the link layer address will be on the
3040 ll_ifma = if_findmulti(ifp, llsa);
3041 if (ll_ifma == NULL) {
3042 ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
3043 if (ll_ifma == NULL) {
3044 --ifma->ifma_refcount;
3049 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
3052 ll_ifma->ifma_refcount++;
3053 ifma->ifma_llifma = ll_ifma;
3057 * We now have a new multicast address, ifma, and possibly a new or
3058 * referenced link layer address. Add the primary address to the
3059 * ifnet address list.
3061 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3063 if (retifma != NULL)
3067 * Must generate the message while holding the lock so that 'ifma'
3068 * pointer is still valid.
3070 rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3071 IF_ADDR_WUNLOCK(ifp);
3074 * We are certain we have added something, so call down to the
3075 * interface to let them know about it.
3077 if (ifp->if_ioctl != NULL) {
3078 (void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3082 free(llsa, M_IFMADDR);
3088 free(llsa, M_IFMADDR);
3091 IF_ADDR_WUNLOCK(ifp);
3096 * Delete a multicast group membership by network-layer group address.
3098 * Returns ENOENT if the entry could not be found. If ifp no longer
3099 * exists, results are undefined. This entry point should only be used
3100 * from subsystems which do appropriate locking to hold ifp for the
3101 * duration of the call.
3102 * Network-layer protocol domains must use if_delmulti_ifma().
3105 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3107 struct ifmultiaddr *ifma;
3112 IFNET_RLOCK_NOSLEEP();
3113 TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3118 IFNET_RUNLOCK_NOSLEEP();
3120 KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3127 ifma = if_findmulti(ifp, sa);
3129 lastref = if_delmulti_locked(ifp, ifma, 0);
3130 IF_ADDR_WUNLOCK(ifp);
3135 if (lastref && ifp->if_ioctl != NULL) {
3136 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3143 * Delete all multicast group membership for an interface.
3144 * Should be used to quickly flush all multicast filters.
3147 if_delallmulti(struct ifnet *ifp)
3149 struct ifmultiaddr *ifma;
3150 struct ifmultiaddr *next;
3153 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3154 if_delmulti_locked(ifp, ifma, 0);
3155 IF_ADDR_WUNLOCK(ifp);
3159 * Delete a multicast group membership by group membership pointer.
3160 * Network-layer protocol domains must use this routine.
3162 * It is safe to call this routine if the ifp disappeared.
3165 if_delmulti_ifma(struct ifmultiaddr *ifma)
3170 ifp = ifma->ifma_ifp;
3173 printf("%s: ifma_ifp seems to be detached\n", __func__);
3177 IFNET_RLOCK_NOSLEEP();
3178 TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3182 printf("%s: ifnet %p disappeared\n", __func__, ifp);
3185 IFNET_RUNLOCK_NOSLEEP();
3189 * If and only if the ifnet instance exists: Acquire the address lock.
3194 lastref = if_delmulti_locked(ifp, ifma, 0);
3198 * If and only if the ifnet instance exists:
3199 * Release the address lock.
3200 * If the group was left: update the hardware hash filter.
3202 IF_ADDR_WUNLOCK(ifp);
3203 if (lastref && ifp->if_ioctl != NULL) {
3204 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3210 * Perform deletion of network-layer and/or link-layer multicast address.
3212 * Return 0 if the reference count was decremented.
3213 * Return 1 if the final reference was released, indicating that the
3214 * hardware hash filter should be reprogrammed.
3217 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3219 struct ifmultiaddr *ll_ifma;
3221 if (ifp != NULL && ifma->ifma_ifp != NULL) {
3222 KASSERT(ifma->ifma_ifp == ifp,
3223 ("%s: inconsistent ifp %p", __func__, ifp));
3224 IF_ADDR_WLOCK_ASSERT(ifp);
3227 ifp = ifma->ifma_ifp;
3230 * If the ifnet is detaching, null out references to ifnet,
3231 * so that upper protocol layers will notice, and not attempt
3232 * to obtain locks for an ifnet which no longer exists. The
3233 * routing socket announcement must happen before the ifnet
3234 * instance is detached from the system.
3238 printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3241 * ifp may already be nulled out if we are being reentered
3242 * to delete the ll_ifma.
3245 rt_newmaddrmsg(RTM_DELMADDR, ifma);
3246 ifma->ifma_ifp = NULL;
3250 if (--ifma->ifma_refcount > 0)
3254 * If this ifma is a network-layer ifma, a link-layer ifma may
3255 * have been associated with it. Release it first if so.
3257 ll_ifma = ifma->ifma_llifma;
3258 if (ll_ifma != NULL) {
3259 KASSERT(ifma->ifma_lladdr != NULL,
3260 ("%s: llifma w/o lladdr", __func__));
3262 ll_ifma->ifma_ifp = NULL; /* XXX */
3263 if (--ll_ifma->ifma_refcount == 0) {
3265 TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3268 if_freemulti(ll_ifma);
3273 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3278 * The last reference to this instance of struct ifmultiaddr
3279 * was released; the hardware should be notified of this change.
3285 * Set the link layer address on an interface.
3287 * At this time we only support certain types of interfaces,
3288 * and we don't allow the length of the address to change.
3291 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3293 struct sockaddr_dl *sdl;
3300 IF_ADDR_RUNLOCK(ifp);
3304 IF_ADDR_RUNLOCK(ifp);
3305 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3310 if (len != sdl->sdl_alen) { /* don't allow length to change */
3314 switch (ifp->if_type) {
3322 case IFT_IEEE8023ADLAG:
3324 bcopy(lladdr, LLADDR(sdl), len);
3333 * If the interface is already up, we need
3334 * to re-init it in order to reprogram its
3337 if ((ifp->if_flags & IFF_UP) != 0) {
3338 if (ifp->if_ioctl) {
3339 ifp->if_flags &= ~IFF_UP;
3340 ifr.ifr_flags = ifp->if_flags & 0xffff;
3341 ifr.ifr_flagshigh = ifp->if_flags >> 16;
3342 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3343 ifp->if_flags |= IFF_UP;
3344 ifr.ifr_flags = ifp->if_flags & 0xffff;
3345 ifr.ifr_flagshigh = ifp->if_flags >> 16;
3346 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3350 * Also send gratuitous ARPs to notify other nodes about
3351 * the address change.
3353 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3354 if (ifa->ifa_addr->sa_family == AF_INET)
3355 arp_ifinit(ifp, ifa);
3363 * The name argument must be a pointer to storage which will last as
3364 * long as the interface does. For physical devices, the result of
3365 * device_get_name(dev) is a good choice and for pseudo-devices a
3366 * static string works well.
3369 if_initname(struct ifnet *ifp, const char *name, int unit)
3371 ifp->if_dname = name;
3372 ifp->if_dunit = unit;
3373 if (unit != IF_DUNIT_NONE)
3374 snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3376 strlcpy(ifp->if_xname, name, IFNAMSIZ);
3380 if_printf(struct ifnet *ifp, const char * fmt, ...)
3385 retval = printf("%s: ", ifp->if_xname);
3387 retval += vprintf(fmt, ap);
3393 if_start(struct ifnet *ifp)
3396 (*(ifp)->if_start)(ifp);
3400 * Backwards compatibility interface for drivers
3401 * that have not implemented it
3404 if_transmit(struct ifnet *ifp, struct mbuf *m)
3408 IFQ_HANDOFF(ifp, m, error);
3413 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3418 if (_IF_QFULL(ifq)) {
3425 ifp->if_obytes += m->m_pkthdr.len + adjust;
3426 if (m->m_flags & (M_BCAST|M_MCAST))
3428 active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3430 _IF_ENQUEUE(ifq, m);
3432 if (ifp != NULL && !active)
3433 (*(ifp)->if_start)(ifp);
3438 if_register_com_alloc(u_char type,
3439 if_com_alloc_t *a, if_com_free_t *f)
3442 KASSERT(if_com_alloc[type] == NULL,
3443 ("if_register_com_alloc: %d already registered", type));
3444 KASSERT(if_com_free[type] == NULL,
3445 ("if_register_com_alloc: %d free already registered", type));
3447 if_com_alloc[type] = a;
3448 if_com_free[type] = f;
3452 if_deregister_com_alloc(u_char type)
3455 KASSERT(if_com_alloc[type] != NULL,
3456 ("if_deregister_com_alloc: %d not registered", type));
3457 KASSERT(if_com_free[type] != NULL,
3458 ("if_deregister_com_alloc: %d free not registered", type));
3459 if_com_alloc[type] = NULL;
3460 if_com_free[type] = NULL;