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 #if defined(INET) || defined(INET6)
134 struct ifnet *(*carp_forus_p)(struct ifnet *ifp, u_char *dhost);
135 int (*carp_output_p)(struct ifnet *ifp, struct mbuf *m,
136 struct sockaddr *sa, struct rtentry *rt);
139 int (*carp_iamatch_p)(struct ifnet *, struct in_ifaddr *, struct in_addr *,
143 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6);
144 caddr_t (*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m,
145 const struct in6_addr *taddr);
148 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
151 * XXX: Style; these should be sorted alphabetically, and unprototyped
152 * static functions should be prototyped. Currently they are sorted by
155 static void if_attachdomain(void *);
156 static void if_attachdomain1(struct ifnet *);
157 static int ifconf(u_long, caddr_t);
158 static void if_freemulti(struct ifmultiaddr *);
159 static void if_init(void *);
160 static void if_grow(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 int if_transmit(struct ifnet *ifp, struct mbuf *m);
164 static void if_unroute(struct ifnet *, int flag, int fam);
165 static void link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
166 static int if_rtdel(struct radix_node *, void *);
167 static int ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
168 static int if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
169 static void do_link_state_change(void *, int);
170 static int if_getgroup(struct ifgroupreq *, struct ifnet *);
171 static int if_getgroupmembers(struct ifgroupreq *);
172 static void if_delgroups(struct ifnet *);
173 static void if_attach_internal(struct ifnet *, int);
174 static void if_detach_internal(struct ifnet *, int);
178 * XXX: declare here to avoid to include many inet6 related files..
179 * should be more generalized?
181 extern void nd6_setmtu(struct ifnet *);
184 VNET_DEFINE(int, if_index);
185 int ifqmaxlen = IFQ_MAXLEN;
186 VNET_DEFINE(struct ifnethead, ifnet); /* depend on static init XXX */
187 VNET_DEFINE(struct ifgrouphead, ifg_head);
189 static VNET_DEFINE(int, if_indexlim) = 8;
191 /* Table of ifnet by index. */
192 VNET_DEFINE(struct ifindex_entry *, ifindex_table);
194 #define V_if_indexlim VNET(if_indexlim)
195 #define V_ifindex_table VNET(ifindex_table)
198 * The global network interface list (V_ifnet) and related state (such as
199 * if_index, if_indexlim, and ifindex_table) are protected by an sxlock and
200 * an rwlock. Either may be acquired shared to stablize the list, but both
201 * must be acquired writable to modify the list. This model allows us to
202 * both stablize the interface list during interrupt thread processing, but
203 * also to stablize it over long-running ioctls, without introducing priority
204 * inversions and deadlocks.
206 struct rwlock ifnet_rwlock;
207 struct sx ifnet_sxlock;
210 * The allocation of network interfaces is a rather non-atomic affair; we
211 * need to select an index before we are ready to expose the interface for
212 * use, so will use this pointer value to indicate reservation.
214 #define IFNET_HOLD (void *)(uintptr_t)(-1)
216 static if_com_alloc_t *if_com_alloc[256];
217 static if_com_free_t *if_com_free[256];
219 static MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
220 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
221 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
224 ifnet_byindex_locked(u_short idx)
227 if (idx > V_if_index)
229 if (V_ifindex_table[idx].ife_ifnet == IFNET_HOLD)
231 return (V_ifindex_table[idx].ife_ifnet);
235 ifnet_byindex(u_short idx)
239 IFNET_RLOCK_NOSLEEP();
240 ifp = ifnet_byindex_locked(idx);
241 IFNET_RUNLOCK_NOSLEEP();
246 ifnet_byindex_ref(u_short idx)
250 IFNET_RLOCK_NOSLEEP();
251 ifp = ifnet_byindex_locked(idx);
252 if (ifp == NULL || (ifp->if_flags & IFF_DYING)) {
253 IFNET_RUNLOCK_NOSLEEP();
257 IFNET_RUNLOCK_NOSLEEP();
262 * Allocate an ifindex array entry; return 0 on success or an error on
266 ifindex_alloc_locked(u_short *idxp)
270 IFNET_WLOCK_ASSERT();
274 * Try to find an empty slot below V_if_index. If we fail, take the
277 for (idx = 1; idx <= V_if_index; idx++) {
278 if (V_ifindex_table[idx].ife_ifnet == NULL)
282 /* Catch if_index overflow. */
285 if (idx >= V_if_indexlim) {
289 if (idx > V_if_index)
296 ifindex_free_locked(u_short idx)
299 IFNET_WLOCK_ASSERT();
301 V_ifindex_table[idx].ife_ifnet = NULL;
302 while (V_if_index > 0 &&
303 V_ifindex_table[V_if_index].ife_ifnet == NULL)
308 ifindex_free(u_short idx)
312 ifindex_free_locked(idx);
317 ifnet_setbyindex_locked(u_short idx, struct ifnet *ifp)
320 IFNET_WLOCK_ASSERT();
322 V_ifindex_table[idx].ife_ifnet = ifp;
326 ifnet_setbyindex(u_short idx, struct ifnet *ifp)
330 ifnet_setbyindex_locked(idx, ifp);
335 ifaddr_byindex(u_short idx)
339 IFNET_RLOCK_NOSLEEP();
340 ifa = ifnet_byindex_locked(idx)->if_addr;
343 IFNET_RUNLOCK_NOSLEEP();
348 * Network interface utility routines.
350 * Routines with ifa_ifwith* names take sockaddr *'s as
355 vnet_if_init(const void *unused __unused)
358 TAILQ_INIT(&V_ifnet);
359 TAILQ_INIT(&V_ifg_head);
361 if_grow(); /* create initial table */
363 vnet_if_clone_init();
365 VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init,
370 if_init(void *dummy __unused)
376 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
381 vnet_if_uninit(const void *unused __unused)
384 VNET_ASSERT(TAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p "
385 "not empty", __func__, __LINE__, &V_ifnet));
386 VNET_ASSERT(TAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p "
387 "not empty", __func__, __LINE__, &V_ifg_head));
389 free((caddr_t)V_ifindex_table, M_IFNET);
391 VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
392 vnet_if_uninit, NULL);
400 struct ifindex_entry *e;
402 IFNET_WLOCK_ASSERT();
403 oldlim = V_if_indexlim;
405 n = (oldlim << 1) * sizeof(*e);
406 e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
408 if (V_if_indexlim != oldlim) {
412 if (V_ifindex_table != NULL) {
413 memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2);
414 free((caddr_t)V_ifindex_table, M_IFNET);
421 * Allocate a struct ifnet and an index for an interface. A layer 2
422 * common structure will also be allocated if an allocation routine is
423 * registered for the passed type.
426 if_alloc(u_char type)
431 ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
433 if (ifindex_alloc_locked(&idx) != 0) {
438 ifnet_setbyindex_locked(idx, IFNET_HOLD);
442 ifp->if_alloctype = type;
443 if (if_com_alloc[type] != NULL) {
444 ifp->if_l2com = if_com_alloc[type](type, ifp);
445 if (ifp->if_l2com == NULL) {
452 IF_ADDR_LOCK_INIT(ifp);
453 TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
454 ifp->if_afdata_initialized = 0;
455 IF_AFDATA_LOCK_INIT(ifp);
456 TAILQ_INIT(&ifp->if_addrhead);
457 TAILQ_INIT(&ifp->if_prefixhead);
458 TAILQ_INIT(&ifp->if_multiaddrs);
459 TAILQ_INIT(&ifp->if_groups);
463 ifq_init(&ifp->if_snd, ifp);
465 refcount_init(&ifp->if_refcount, 1); /* Index reference. */
466 ifnet_setbyindex(ifp->if_index, ifp);
471 * Do the actual work of freeing a struct ifnet, and layer 2 common
472 * structure. This call is made when the last reference to an
473 * interface is released.
476 if_free_internal(struct ifnet *ifp)
479 KASSERT((ifp->if_flags & IFF_DYING),
480 ("if_free_internal: interface not dying"));
482 if (if_com_free[ifp->if_alloctype] != NULL)
483 if_com_free[ifp->if_alloctype](ifp->if_l2com,
487 mac_ifnet_destroy(ifp);
489 if (ifp->if_description != NULL)
490 free(ifp->if_description, M_IFDESCR);
491 IF_AFDATA_DESTROY(ifp);
492 IF_ADDR_LOCK_DESTROY(ifp);
493 ifq_delete(&ifp->if_snd);
498 * This version should only be called by intefaces that switch their type
499 * after calling if_alloc(). if_free_type() will go away again now that we
500 * have if_alloctype to cache the original allocation type. For now, assert
501 * that they match, since we require that in practice.
504 if_free_type(struct ifnet *ifp, u_char type)
507 KASSERT(ifp->if_alloctype == type,
508 ("if_free_type: type (%d) != alloctype (%d)", type,
511 ifp->if_flags |= IFF_DYING; /* XXX: Locking */
513 CURVNET_SET_QUIET(ifp->if_vnet);
515 KASSERT(ifp == ifnet_byindex_locked(ifp->if_index),
516 ("%s: freeing unallocated ifnet", ifp->if_xname));
518 ifindex_free_locked(ifp->if_index);
521 if (refcount_release(&ifp->if_refcount))
522 if_free_internal(ifp);
527 * This is the normal version of if_free(), used by device drivers to free a
528 * detached network interface. The contents of if_free_type() will move into
529 * here when if_free_type() goes away.
532 if_free(struct ifnet *ifp)
535 if_free_type(ifp, ifp->if_alloctype);
539 * Interfaces to keep an ifnet type-stable despite the possibility of the
540 * driver calling if_free(). If there are additional references, we defer
541 * freeing the underlying data structure.
544 if_ref(struct ifnet *ifp)
547 /* We don't assert the ifnet list lock here, but arguably should. */
548 refcount_acquire(&ifp->if_refcount);
552 if_rele(struct ifnet *ifp)
555 if (!refcount_release(&ifp->if_refcount))
557 if_free_internal(ifp);
561 ifq_init(struct ifaltq *ifq, struct ifnet *ifp)
564 mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
566 if (ifq->ifq_maxlen == 0)
567 ifq->ifq_maxlen = ifqmaxlen;
570 ifq->altq_disc = NULL;
571 ifq->altq_flags &= ALTQF_CANTCHANGE;
572 ifq->altq_tbr = NULL;
577 ifq_delete(struct ifaltq *ifq)
579 mtx_destroy(&ifq->ifq_mtx);
583 * Perform generic interface initalization tasks and attach the interface
584 * to the list of "active" interfaces. If vmove flag is set on entry
585 * to if_attach_internal(), perform only a limited subset of initialization
586 * tasks, given that we are moving from one vnet to another an ifnet which
587 * has already been fully initialized.
590 * - The decision to return void and thus require this function to
591 * succeed is questionable.
592 * - We should probably do more sanity checking. For instance we don't
593 * do anything to insure if_xname is unique or non-empty.
596 if_attach(struct ifnet *ifp)
599 if_attach_internal(ifp, 0);
603 * Compute the least common TSO limit.
606 if_hw_tsomax_common(struct ifnet *ifp, struct ifnet_hw_tsomax *pmax)
609 * 1) If there is no limit currently, take the limit from
610 * the network adapter.
612 * 2) If the network adapter has a limit below the current
615 if (pmax->tsomaxbytes == 0 || (ifp->if_hw_tsomax != 0 &&
616 ifp->if_hw_tsomax < pmax->tsomaxbytes)) {
617 pmax->tsomaxbytes = ifp->if_hw_tsomax;
619 if (pmax->tsomaxsegcount == 0 || (ifp->if_hw_tsomaxsegcount != 0 &&
620 ifp->if_hw_tsomaxsegcount < pmax->tsomaxsegcount)) {
621 pmax->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
623 if (pmax->tsomaxsegsize == 0 || (ifp->if_hw_tsomaxsegsize != 0 &&
624 ifp->if_hw_tsomaxsegsize < pmax->tsomaxsegsize)) {
625 pmax->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
630 * Update TSO limit of a network adapter.
632 * Returns zero if no change. Else non-zero.
635 if_hw_tsomax_update(struct ifnet *ifp, struct ifnet_hw_tsomax *pmax)
638 if (ifp->if_hw_tsomax != pmax->tsomaxbytes) {
639 ifp->if_hw_tsomax = pmax->tsomaxbytes;
642 if (ifp->if_hw_tsomaxsegsize != pmax->tsomaxsegsize) {
643 ifp->if_hw_tsomaxsegsize = pmax->tsomaxsegsize;
646 if (ifp->if_hw_tsomaxsegcount != pmax->tsomaxsegcount) {
647 ifp->if_hw_tsomaxsegcount = pmax->tsomaxsegcount;
654 if_attach_internal(struct ifnet *ifp, int vmove)
656 unsigned socksize, ifasize;
657 int namelen, masklen;
658 struct sockaddr_dl *sdl;
661 if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
662 panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
666 ifp->if_vnet = curvnet;
667 if (ifp->if_home_vnet == NULL)
668 ifp->if_home_vnet = curvnet;
671 if_addgroup(ifp, IFG_ALL);
673 getmicrotime(&ifp->if_lastchange);
674 ifp->if_data.ifi_epoch = time_uptime;
675 ifp->if_data.ifi_datalen = sizeof(struct if_data);
677 KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
678 (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
679 ("transmit and qflush must both either be set or both be NULL"));
680 if (ifp->if_transmit == NULL) {
681 ifp->if_transmit = if_transmit;
682 ifp->if_qflush = if_qflush;
687 mac_ifnet_create(ifp);
691 * Create a Link Level name for this device.
693 namelen = strlen(ifp->if_xname);
695 * Always save enough space for any possiable name so we
696 * can do a rename in place later.
698 masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
699 socksize = masklen + ifp->if_addrlen;
700 if (socksize < sizeof(*sdl))
701 socksize = sizeof(*sdl);
702 socksize = roundup2(socksize, sizeof(long));
703 ifasize = sizeof(*ifa) + 2 * socksize;
704 ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
706 sdl = (struct sockaddr_dl *)(ifa + 1);
707 sdl->sdl_len = socksize;
708 sdl->sdl_family = AF_LINK;
709 bcopy(ifp->if_xname, sdl->sdl_data, namelen);
710 sdl->sdl_nlen = namelen;
711 sdl->sdl_index = ifp->if_index;
712 sdl->sdl_type = ifp->if_type;
715 ifa->ifa_rtrequest = link_rtrequest;
716 ifa->ifa_addr = (struct sockaddr *)sdl;
717 sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
718 ifa->ifa_netmask = (struct sockaddr *)sdl;
719 sdl->sdl_len = masklen;
721 sdl->sdl_data[--namelen] = 0xff;
722 TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
723 /* Reliably crash if used uninitialized. */
724 ifp->if_broadcastaddr = NULL;
726 #if defined(INET) || defined(INET6)
727 /* Use defaults for TSO, if nothing is set */
728 if (ifp->if_hw_tsomax == 0 &&
729 ifp->if_hw_tsomaxsegcount == 0 &&
730 ifp->if_hw_tsomaxsegsize == 0) {
732 * The TSO defaults needs to be such that an
733 * NFS mbuf list of 35 mbufs totalling just
734 * below 64K works and that a chain of mbufs
735 * can be defragged into at most 32 segments:
737 ifp->if_hw_tsomax = min(IP_MAXPACKET, (32 * MCLBYTES) -
738 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
739 ifp->if_hw_tsomaxsegcount = 35;
740 ifp->if_hw_tsomaxsegsize = 2048; /* 2K */
742 /* XXX some drivers set IFCAP_TSO after ethernet attach */
743 if (ifp->if_capabilities & IFCAP_TSO) {
744 if_printf(ifp, "Using defaults for TSO: %u/%u/%u\n",
746 ifp->if_hw_tsomaxsegcount,
747 ifp->if_hw_tsomaxsegsize);
755 * Update the interface index in the link layer address
758 for (ifa = ifp->if_addr; ifa != NULL;
759 ifa = TAILQ_NEXT(ifa, ifa_link)) {
760 if (ifa->ifa_addr->sa_family == AF_LINK) {
761 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
762 sdl->sdl_index = ifp->if_index;
769 TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
771 curvnet->vnet_ifcnt++;
775 if (domain_init_status >= 2)
776 if_attachdomain1(ifp);
778 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
779 if (IS_DEFAULT_VNET(curvnet))
780 devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
782 /* Announce the interface. */
783 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
787 if_attachdomain(void *dummy)
793 TAILQ_FOREACH(ifp, &V_ifnet, if_link)
794 if_attachdomain1(ifp);
797 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
798 if_attachdomain, NULL);
801 if_attachdomain1(struct ifnet *ifp)
809 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
810 * cannot lock ifp->if_afdata initialization, entirely.
812 if (IF_AFDATA_TRYLOCK(ifp) == 0) {
816 if (ifp->if_afdata_initialized >= domain_init_status) {
817 IF_AFDATA_UNLOCK(ifp);
819 printf("if_attachdomain called more than once on %s\n",
823 ifp->if_afdata_initialized = domain_init_status;
824 IF_AFDATA_UNLOCK(ifp);
826 /* address family dependent data region */
827 bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
828 for (dp = domains; dp; dp = dp->dom_next) {
829 if (dp->dom_ifattach)
830 ifp->if_afdata[dp->dom_family] =
831 (*dp->dom_ifattach)(ifp);
838 * Remove any unicast or broadcast network addresses from an interface.
841 if_purgeaddrs(struct ifnet *ifp)
843 struct ifaddr *ifa, *next;
845 TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
846 if (ifa->ifa_addr->sa_family == AF_LINK)
849 /* XXX: Ugly!! ad hoc just for INET */
850 if (ifa->ifa_addr->sa_family == AF_INET) {
851 struct ifaliasreq ifr;
853 bzero(&ifr, sizeof(ifr));
854 ifr.ifra_addr = *ifa->ifa_addr;
855 if (ifa->ifa_dstaddr)
856 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
857 if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
863 if (ifa->ifa_addr->sa_family == AF_INET6) {
865 /* ifp_addrhead is already updated */
869 TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
875 * Remove any multicast network addresses from an interface when an ifnet
879 if_purgemaddrs(struct ifnet *ifp)
881 struct ifmultiaddr *ifma;
882 struct ifmultiaddr *next;
885 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
886 if_delmulti_locked(ifp, ifma, 1);
887 IF_ADDR_WUNLOCK(ifp);
891 * Detach an interface, removing it from the list of "active" interfaces.
892 * If vmove flag is set on entry to if_detach_internal(), perform only a
893 * limited subset of cleanup tasks, given that we are moving an ifnet from
894 * one vnet to another, where it must be fully operational.
896 * XXXRW: There are some significant questions about event ordering, and
897 * how to prevent things from starting to use the interface during detach.
900 if_detach(struct ifnet *ifp)
903 CURVNET_SET_QUIET(ifp->if_vnet);
904 if_detach_internal(ifp, 0);
909 if_detach_internal(struct ifnet *ifp, int vmove)
912 struct radix_node_head *rnh;
919 TAILQ_FOREACH(iter, &V_ifnet, if_link)
921 TAILQ_REMOVE(&V_ifnet, ifp, if_link);
927 curvnet->vnet_ifcnt--;
932 panic("%s: ifp=%p not on the ifnet tailq %p",
933 __func__, ifp, &V_ifnet);
935 return; /* XXX this should panic as well? */
939 * Remove/wait for pending events.
941 taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
944 * Remove routes and flush queues.
948 if (ALTQ_IS_ENABLED(&ifp->if_snd))
949 altq_disable(&ifp->if_snd);
950 if (ALTQ_IS_ATTACHED(&ifp->if_snd))
951 altq_detach(&ifp->if_snd);
962 * Remove all IPv6 kernel structs related to ifp. This should be done
963 * before removing routing entries below, since IPv6 interface direct
964 * routes are expected to be removed by the IPv6-specific kernel API.
965 * Otherwise, the kernel will detect some inconsistency and bark it.
973 * Prevent further calls into the device driver via ifnet.
978 * Remove link ifaddr pointer and maybe decrement if_index.
979 * Clean up all addresses.
983 /* We can now free link ifaddr. */
984 if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
985 ifa = TAILQ_FIRST(&ifp->if_addrhead);
986 TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
992 * Delete all remaining routes using this interface
993 * Unfortuneatly the only way to do this is to slog through
994 * the entire routing table looking for routes which point
995 * to this interface...oh well...
997 for (i = 1; i <= AF_MAX; i++) {
998 for (j = 0; j < rt_numfibs; j++) {
999 rnh = rt_tables_get_rnh(j, i);
1002 RADIX_NODE_HEAD_LOCK(rnh);
1003 (void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
1004 RADIX_NODE_HEAD_UNLOCK(rnh);
1008 /* Announce that the interface is gone. */
1009 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1010 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
1011 if (IS_DEFAULT_VNET(curvnet))
1012 devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
1016 * We cannot hold the lock over dom_ifdetach calls as they might
1017 * sleep, for example trying to drain a callout, thus open up the
1018 * theoretical race with re-attaching.
1020 IF_AFDATA_LOCK(ifp);
1021 i = ifp->if_afdata_initialized;
1022 ifp->if_afdata_initialized = 0;
1023 IF_AFDATA_UNLOCK(ifp);
1024 for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
1025 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
1026 (*dp->dom_ifdetach)(ifp,
1027 ifp->if_afdata[dp->dom_family]);
1033 * if_vmove() performs a limited version of if_detach() in current
1034 * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
1035 * An attempt is made to shrink if_index in current vnet, find an
1036 * unused if_index in target vnet and calls if_grow() if necessary,
1037 * and finally find an unused if_xname for the target vnet.
1040 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
1045 * Detach from current vnet, but preserve LLADDR info, do not
1046 * mark as dead etc. so that the ifnet can be reattached later.
1048 if_detach_internal(ifp, 1);
1051 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
1052 * the if_index for that vnet if possible.
1054 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
1055 * or we'd lock on one vnet and unlock on another.
1058 ifindex_free_locked(ifp->if_index);
1062 * Perform interface-specific reassignment tasks, if provided by
1065 if (ifp->if_reassign != NULL)
1066 ifp->if_reassign(ifp, new_vnet, NULL);
1069 * Switch to the context of the target vnet.
1071 CURVNET_SET_QUIET(new_vnet);
1074 if (ifindex_alloc_locked(&idx) != 0) {
1076 panic("if_index overflow");
1078 ifp->if_index = idx;
1079 ifnet_setbyindex_locked(ifp->if_index, ifp);
1082 if_attach_internal(ifp, 1);
1088 * Move an ifnet to or from another child prison/vnet, specified by the jail id.
1091 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
1096 /* Try to find the prison within our visibility. */
1097 sx_slock(&allprison_lock);
1098 pr = prison_find_child(td->td_ucred->cr_prison, jid);
1099 sx_sunlock(&allprison_lock);
1102 prison_hold_locked(pr);
1103 mtx_unlock(&pr->pr_mtx);
1105 /* Do not try to move the iface from and to the same prison. */
1106 if (pr->pr_vnet == ifp->if_vnet) {
1111 /* Make sure the named iface does not exists in the dst. prison/vnet. */
1112 /* XXX Lock interfaces to avoid races. */
1113 CURVNET_SET_QUIET(pr->pr_vnet);
1114 difp = ifunit(ifname);
1121 /* Move the interface into the child jail/vnet. */
1122 if_vmove(ifp, pr->pr_vnet);
1124 /* Report the new if_xname back to the userland. */
1125 sprintf(ifname, "%s", ifp->if_xname);
1132 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1135 struct vnet *vnet_dst;
1138 /* Try to find the prison within our visibility. */
1139 sx_slock(&allprison_lock);
1140 pr = prison_find_child(td->td_ucred->cr_prison, jid);
1141 sx_sunlock(&allprison_lock);
1144 prison_hold_locked(pr);
1145 mtx_unlock(&pr->pr_mtx);
1147 /* Make sure the named iface exists in the source prison/vnet. */
1148 CURVNET_SET(pr->pr_vnet);
1149 ifp = ifunit(ifname); /* XXX Lock to avoid races. */
1156 /* Do not try to move the iface from and to the same prison. */
1157 vnet_dst = TD_TO_VNET(td);
1158 if (vnet_dst == ifp->if_vnet) {
1164 /* Get interface back from child jail/vnet. */
1165 if_vmove(ifp, vnet_dst);
1168 /* Report the new if_xname back to the userland. */
1169 sprintf(ifname, "%s", ifp->if_xname);
1177 * Add a group to an interface
1180 if_addgroup(struct ifnet *ifp, const char *groupname)
1182 struct ifg_list *ifgl;
1183 struct ifg_group *ifg = NULL;
1184 struct ifg_member *ifgm;
1186 if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1187 groupname[strlen(groupname) - 1] <= '9')
1191 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1192 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1197 if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1198 M_NOWAIT)) == NULL) {
1203 if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1204 M_TEMP, M_NOWAIT)) == NULL) {
1210 TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1211 if (!strcmp(ifg->ifg_group, groupname))
1215 if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1216 M_TEMP, M_NOWAIT)) == NULL) {
1222 strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1223 ifg->ifg_refcnt = 0;
1224 TAILQ_INIT(&ifg->ifg_members);
1225 EVENTHANDLER_INVOKE(group_attach_event, ifg);
1226 TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1230 ifgl->ifgl_group = ifg;
1231 ifgm->ifgm_ifp = ifp;
1234 TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1235 TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1236 IF_ADDR_WUNLOCK(ifp);
1240 EVENTHANDLER_INVOKE(group_change_event, groupname);
1246 * Remove a group from an interface
1249 if_delgroup(struct ifnet *ifp, const char *groupname)
1251 struct ifg_list *ifgl;
1252 struct ifg_member *ifgm;
1255 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1256 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1264 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1265 IF_ADDR_WUNLOCK(ifp);
1267 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1268 if (ifgm->ifgm_ifp == ifp)
1272 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1276 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1277 TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1278 EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1279 free(ifgl->ifgl_group, M_TEMP);
1285 EVENTHANDLER_INVOKE(group_change_event, groupname);
1291 * Remove an interface from all groups
1294 if_delgroups(struct ifnet *ifp)
1296 struct ifg_list *ifgl;
1297 struct ifg_member *ifgm;
1298 char groupname[IFNAMSIZ];
1301 while (!TAILQ_EMPTY(&ifp->if_groups)) {
1302 ifgl = TAILQ_FIRST(&ifp->if_groups);
1304 strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1307 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1308 IF_ADDR_WUNLOCK(ifp);
1310 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1311 if (ifgm->ifgm_ifp == ifp)
1315 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1320 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1321 TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1322 EVENTHANDLER_INVOKE(group_detach_event,
1324 free(ifgl->ifgl_group, M_TEMP);
1330 EVENTHANDLER_INVOKE(group_change_event, groupname);
1338 * Stores all groups from an interface in memory pointed
1342 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1345 struct ifg_list *ifgl;
1346 struct ifg_req ifgrq, *ifgp;
1347 struct ifgroupreq *ifgr = data;
1349 if (ifgr->ifgr_len == 0) {
1351 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1352 ifgr->ifgr_len += sizeof(struct ifg_req);
1353 IF_ADDR_RUNLOCK(ifp);
1357 len = ifgr->ifgr_len;
1358 ifgp = ifgr->ifgr_groups;
1361 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1362 if (len < sizeof(ifgrq)) {
1363 IF_ADDR_RUNLOCK(ifp);
1366 bzero(&ifgrq, sizeof ifgrq);
1367 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1368 sizeof(ifgrq.ifgrq_group));
1369 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1370 IF_ADDR_RUNLOCK(ifp);
1373 len -= sizeof(ifgrq);
1376 IF_ADDR_RUNLOCK(ifp);
1382 * Stores all members of a group in memory pointed to by data
1385 if_getgroupmembers(struct ifgroupreq *data)
1387 struct ifgroupreq *ifgr = data;
1388 struct ifg_group *ifg;
1389 struct ifg_member *ifgm;
1390 struct ifg_req ifgrq, *ifgp;
1394 TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1395 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1402 if (ifgr->ifgr_len == 0) {
1403 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1404 ifgr->ifgr_len += sizeof(ifgrq);
1409 len = ifgr->ifgr_len;
1410 ifgp = ifgr->ifgr_groups;
1411 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1412 if (len < sizeof(ifgrq)) {
1416 bzero(&ifgrq, sizeof ifgrq);
1417 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1418 sizeof(ifgrq.ifgrq_member));
1419 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1423 len -= sizeof(ifgrq);
1432 * Delete Routes for a Network Interface
1434 * Called for each routing entry via the rnh->rnh_walktree() call above
1435 * to delete all route entries referencing a detaching network interface.
1438 * rn pointer to node in the routing table
1439 * arg argument passed to rnh->rnh_walktree() - detaching interface
1443 * errno failed - reason indicated
1447 if_rtdel(struct radix_node *rn, void *arg)
1449 struct rtentry *rt = (struct rtentry *)rn;
1450 struct ifnet *ifp = arg;
1453 if (rt->rt_ifp == ifp) {
1456 * Protect (sorta) against walktree recursion problems
1457 * with cloned routes
1459 if ((rt->rt_flags & RTF_UP) == 0)
1462 err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1464 rt->rt_flags|RTF_RNH_LOCKED|RTF_PINNED,
1465 (struct rtentry **) NULL, rt->rt_fibnum);
1467 log(LOG_WARNING, "if_rtdel: error %d\n", err);
1475 * Wrapper functions for struct ifnet address list locking macros. These are
1476 * used by kernel modules to avoid encoding programming interface or binary
1477 * interface assumptions that may be violated when kernel-internal locking
1478 * approaches change.
1481 if_addr_rlock(struct ifnet *ifp)
1488 if_addr_runlock(struct ifnet *ifp)
1491 IF_ADDR_RUNLOCK(ifp);
1495 if_maddr_rlock(struct ifnet *ifp)
1502 if_maddr_runlock(struct ifnet *ifp)
1505 IF_ADDR_RUNLOCK(ifp);
1509 * Reference count functions for ifaddrs.
1512 ifa_init(struct ifaddr *ifa)
1515 mtx_init(&ifa->ifa_mtx, "ifaddr", NULL, MTX_DEF);
1516 refcount_init(&ifa->ifa_refcnt, 1);
1520 ifa_ref(struct ifaddr *ifa)
1523 refcount_acquire(&ifa->ifa_refcnt);
1527 ifa_free(struct ifaddr *ifa)
1530 if (refcount_release(&ifa->ifa_refcnt)) {
1531 mtx_destroy(&ifa->ifa_mtx);
1532 free(ifa, M_IFADDR);
1537 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1540 struct rtentry *rt = NULL;
1541 struct rt_addrinfo info;
1542 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1544 bzero(&info, sizeof(info));
1545 info.rti_ifp = V_loif;
1546 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1547 info.rti_info[RTAX_DST] = ia;
1548 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1549 error = rtrequest1_fib(RTM_ADD, &info, &rt, ifa->ifa_ifp->if_fib);
1551 if (error == 0 && rt != NULL) {
1553 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1554 ifa->ifa_ifp->if_type;
1555 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1556 ifa->ifa_ifp->if_index;
1559 } else if (error != 0)
1560 log(LOG_INFO, "ifa_add_loopback_route: insertion failed\n");
1566 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1569 struct rt_addrinfo info;
1570 struct sockaddr_dl null_sdl;
1572 bzero(&null_sdl, sizeof(null_sdl));
1573 null_sdl.sdl_len = sizeof(null_sdl);
1574 null_sdl.sdl_family = AF_LINK;
1575 null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1576 null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1577 bzero(&info, sizeof(info));
1578 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1579 info.rti_info[RTAX_DST] = ia;
1580 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1581 error = rtrequest1_fib(RTM_DELETE, &info, NULL, ifa->ifa_ifp->if_fib);
1584 log(LOG_INFO, "ifa_del_loopback_route: deletion failed\n");
1590 * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1591 * structs used to represent other address families, it is necessary
1592 * to perform a different comparison.
1595 #define sa_equal(a1, a2) \
1596 (bcmp((a1), (a2), ((a1))->sa_len) == 0)
1598 #define sa_dl_equal(a1, a2) \
1599 ((((struct sockaddr_dl *)(a1))->sdl_len == \
1600 ((struct sockaddr_dl *)(a2))->sdl_len) && \
1601 (bcmp(LLADDR((struct sockaddr_dl *)(a1)), \
1602 LLADDR((struct sockaddr_dl *)(a2)), \
1603 ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1606 * Locate an interface based on a complete address.
1609 static struct ifaddr *
1610 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1615 IFNET_RLOCK_NOSLEEP();
1616 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1618 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1619 if (ifa->ifa_addr->sa_family != addr->sa_family)
1621 if (sa_equal(addr, ifa->ifa_addr)) {
1624 IF_ADDR_RUNLOCK(ifp);
1627 /* IP6 doesn't have broadcast */
1628 if ((ifp->if_flags & IFF_BROADCAST) &&
1629 ifa->ifa_broadaddr &&
1630 ifa->ifa_broadaddr->sa_len != 0 &&
1631 sa_equal(ifa->ifa_broadaddr, addr)) {
1634 IF_ADDR_RUNLOCK(ifp);
1638 IF_ADDR_RUNLOCK(ifp);
1642 IFNET_RUNLOCK_NOSLEEP();
1647 ifa_ifwithaddr(struct sockaddr *addr)
1650 return (ifa_ifwithaddr_internal(addr, 1));
1654 ifa_ifwithaddr_check(struct sockaddr *addr)
1657 return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1661 * Locate an interface based on the broadcast address.
1665 ifa_ifwithbroadaddr(struct sockaddr *addr)
1670 IFNET_RLOCK_NOSLEEP();
1671 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1673 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1674 if (ifa->ifa_addr->sa_family != addr->sa_family)
1676 if ((ifp->if_flags & IFF_BROADCAST) &&
1677 ifa->ifa_broadaddr &&
1678 ifa->ifa_broadaddr->sa_len != 0 &&
1679 sa_equal(ifa->ifa_broadaddr, addr)) {
1681 IF_ADDR_RUNLOCK(ifp);
1685 IF_ADDR_RUNLOCK(ifp);
1689 IFNET_RUNLOCK_NOSLEEP();
1694 * Locate the point to point interface with a given destination address.
1698 ifa_ifwithdstaddr_fib(struct sockaddr *addr, int fibnum)
1703 IFNET_RLOCK_NOSLEEP();
1704 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1705 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1707 if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum))
1710 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1711 if (ifa->ifa_addr->sa_family != addr->sa_family)
1713 if (ifa->ifa_dstaddr != NULL &&
1714 sa_equal(addr, ifa->ifa_dstaddr)) {
1716 IF_ADDR_RUNLOCK(ifp);
1720 IF_ADDR_RUNLOCK(ifp);
1724 IFNET_RUNLOCK_NOSLEEP();
1729 ifa_ifwithdstaddr(struct sockaddr *addr)
1732 return (ifa_ifwithdstaddr_fib(addr, RT_ALL_FIBS));
1736 * Find an interface on a specific network. If many, choice
1737 * is most specific found.
1740 ifa_ifwithnet_fib(struct sockaddr *addr, int ignore_ptp, int fibnum)
1744 struct ifaddr *ifa_maybe = NULL;
1745 u_int af = addr->sa_family;
1746 char *addr_data = addr->sa_data, *cplim;
1749 * AF_LINK addresses can be looked up directly by their index number,
1750 * so do that if we can.
1752 if (af == AF_LINK) {
1753 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1754 if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1755 return (ifaddr_byindex(sdl->sdl_index));
1759 * Scan though each interface, looking for ones that have addresses
1760 * in this address family and the requested fib. Maintain a reference
1761 * on ifa_maybe once we find one, as we release the IF_ADDR_RLOCK() that
1762 * kept it stable when we move onto the next interface.
1764 IFNET_RLOCK_NOSLEEP();
1765 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1766 if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum))
1769 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1770 char *cp, *cp2, *cp3;
1772 if (ifa->ifa_addr->sa_family != af)
1774 if (af == AF_INET &&
1775 ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1777 * This is a bit broken as it doesn't
1778 * take into account that the remote end may
1779 * be a single node in the network we are
1781 * The trouble is that we don't know the
1782 * netmask for the remote end.
1784 if (ifa->ifa_dstaddr != NULL &&
1785 sa_equal(addr, ifa->ifa_dstaddr)) {
1787 IF_ADDR_RUNLOCK(ifp);
1792 * if we have a special address handler,
1793 * then use it instead of the generic one.
1795 if (ifa->ifa_claim_addr) {
1796 if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1798 IF_ADDR_RUNLOCK(ifp);
1805 * Scan all the bits in the ifa's address.
1806 * If a bit dissagrees with what we are
1807 * looking for, mask it with the netmask
1808 * to see if it really matters.
1809 * (A byte at a time)
1811 if (ifa->ifa_netmask == 0)
1814 cp2 = ifa->ifa_addr->sa_data;
1815 cp3 = ifa->ifa_netmask->sa_data;
1816 cplim = ifa->ifa_netmask->sa_len
1817 + (char *)ifa->ifa_netmask;
1819 if ((*cp++ ^ *cp2++) & *cp3++)
1820 goto next; /* next address! */
1822 * If the netmask of what we just found
1823 * is more specific than what we had before
1824 * (if we had one) then remember the new one
1825 * before continuing to search
1826 * for an even better one.
1828 if (ifa_maybe == NULL ||
1829 rn_refines((caddr_t)ifa->ifa_netmask,
1830 (caddr_t)ifa_maybe->ifa_netmask)) {
1831 if (ifa_maybe != NULL)
1832 ifa_free(ifa_maybe);
1838 IF_ADDR_RUNLOCK(ifp);
1843 IFNET_RUNLOCK_NOSLEEP();
1844 if (ifa_maybe != NULL)
1845 ifa_free(ifa_maybe);
1850 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp)
1853 return (ifa_ifwithnet_fib(addr, ignore_ptp, RT_ALL_FIBS));
1857 * Find an interface address specific to an interface best matching
1861 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1864 char *cp, *cp2, *cp3;
1866 struct ifaddr *ifa_maybe = NULL;
1867 u_int af = addr->sa_family;
1872 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1873 if (ifa->ifa_addr->sa_family != af)
1875 if (ifa_maybe == NULL)
1877 if (ifa->ifa_netmask == 0) {
1878 if (sa_equal(addr, ifa->ifa_addr) ||
1879 (ifa->ifa_dstaddr &&
1880 sa_equal(addr, ifa->ifa_dstaddr)))
1884 if (ifp->if_flags & IFF_POINTOPOINT) {
1885 if (sa_equal(addr, ifa->ifa_dstaddr))
1889 cp2 = ifa->ifa_addr->sa_data;
1890 cp3 = ifa->ifa_netmask->sa_data;
1891 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1892 for (; cp3 < cplim; cp3++)
1893 if ((*cp++ ^ *cp2++) & *cp3)
1903 IF_ADDR_RUNLOCK(ifp);
1907 #include <net/if_llatbl.h>
1910 * Default action when installing a route with a Link Level gateway.
1911 * Lookup an appropriate real ifa to point to.
1912 * This should be moved to /sys/net/link.c eventually.
1915 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1917 struct ifaddr *ifa, *oifa;
1918 struct sockaddr *dst;
1923 if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1924 ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1926 ifa = ifaof_ifpforaddr(dst, ifp);
1931 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1932 ifa->ifa_rtrequest(cmd, rt, info);
1937 * Mark an interface down and notify protocols of
1939 * NOTE: must be called at splnet or eqivalent.
1942 if_unroute(struct ifnet *ifp, int flag, int fam)
1946 KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1948 ifp->if_flags &= ~flag;
1949 getmicrotime(&ifp->if_lastchange);
1950 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1951 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1952 pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1953 ifp->if_qflush(ifp);
1956 (*carp_linkstate_p)(ifp);
1961 * Mark an interface up and notify protocols of
1963 * NOTE: must be called at splnet or eqivalent.
1966 if_route(struct ifnet *ifp, int flag, int fam)
1970 KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1972 ifp->if_flags |= flag;
1973 getmicrotime(&ifp->if_lastchange);
1974 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1975 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1976 pfctlinput(PRC_IFUP, ifa->ifa_addr);
1978 (*carp_linkstate_p)(ifp);
1985 void (*vlan_link_state_p)(struct ifnet *); /* XXX: private from if_vlan */
1986 void (*vlan_trunk_cap_p)(struct ifnet *); /* XXX: private from if_vlan */
1987 struct ifnet *(*vlan_trunkdev_p)(struct ifnet *);
1988 struct ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t);
1989 int (*vlan_tag_p)(struct ifnet *, uint16_t *);
1990 int (*vlan_setcookie_p)(struct ifnet *, void *);
1991 void *(*vlan_cookie_p)(struct ifnet *);
1994 * Handle a change in the interface link state. To avoid LORs
1995 * between driver lock and upper layer locks, as well as possible
1996 * recursions, we post event to taskqueue, and all job
1997 * is done in static do_link_state_change().
2000 if_link_state_change(struct ifnet *ifp, int link_state)
2002 /* Return if state hasn't changed. */
2003 if (ifp->if_link_state == link_state)
2006 ifp->if_link_state = link_state;
2008 taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
2012 do_link_state_change(void *arg, int pending)
2014 struct ifnet *ifp = (struct ifnet *)arg;
2015 int link_state = ifp->if_link_state;
2016 CURVNET_SET(ifp->if_vnet);
2018 /* Notify that the link state has changed. */
2020 if (ifp->if_vlantrunk != NULL)
2021 (*vlan_link_state_p)(ifp);
2023 if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
2024 IFP2AC(ifp)->ac_netgraph != NULL)
2025 (*ng_ether_link_state_p)(ifp, link_state);
2027 (*carp_linkstate_p)(ifp);
2029 (*bridge_linkstate_p)(ifp);
2031 (*lagg_linkstate_p)(ifp, link_state);
2033 if (IS_DEFAULT_VNET(curvnet))
2034 devctl_notify("IFNET", ifp->if_xname,
2035 (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
2038 if_printf(ifp, "%d link states coalesced\n", pending);
2039 if (log_link_state_change)
2040 log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
2041 (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
2042 EVENTHANDLER_INVOKE(ifnet_link_event, ifp, ifp->if_link_state);
2047 * Mark an interface down and notify protocols of
2049 * NOTE: must be called at splnet or eqivalent.
2052 if_down(struct ifnet *ifp)
2055 if_unroute(ifp, IFF_UP, AF_UNSPEC);
2059 * Mark an interface up and notify protocols of
2061 * NOTE: must be called at splnet or eqivalent.
2064 if_up(struct ifnet *ifp)
2067 if_route(ifp, IFF_UP, AF_UNSPEC);
2071 * Flush an interface queue.
2074 if_qflush(struct ifnet *ifp)
2082 if (ALTQ_IS_ENABLED(ifq))
2086 while ((m = n) != 0) {
2097 * Map interface name to interface structure pointer, with or without
2098 * returning a reference.
2101 ifunit_ref(const char *name)
2105 IFNET_RLOCK_NOSLEEP();
2106 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2107 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2108 !(ifp->if_flags & IFF_DYING))
2113 IFNET_RUNLOCK_NOSLEEP();
2118 ifunit(const char *name)
2122 IFNET_RLOCK_NOSLEEP();
2123 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2124 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2127 IFNET_RUNLOCK_NOSLEEP();
2132 * Hardware specific interface ioctls.
2135 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2140 int new_flags, temp_flags;
2141 size_t namelen, onamelen;
2143 char *descrbuf, *odescrbuf;
2144 char new_name[IFNAMSIZ];
2146 struct sockaddr_dl *sdl;
2148 ifr = (struct ifreq *)data;
2151 ifr->ifr_index = ifp->if_index;
2155 temp_flags = ifp->if_flags | ifp->if_drv_flags;
2156 ifr->ifr_flags = temp_flags & 0xffff;
2157 ifr->ifr_flagshigh = temp_flags >> 16;
2161 ifr->ifr_reqcap = ifp->if_capabilities;
2162 ifr->ifr_curcap = ifp->if_capenable;
2167 error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2172 ifr->ifr_metric = ifp->if_metric;
2176 ifr->ifr_mtu = ifp->if_mtu;
2180 ifr->ifr_phys = ifp->if_physical;
2185 sx_slock(&ifdescr_sx);
2186 if (ifp->if_description == NULL)
2189 /* space for terminating nul */
2190 descrlen = strlen(ifp->if_description) + 1;
2191 if (ifr->ifr_buffer.length < descrlen)
2192 ifr->ifr_buffer.buffer = NULL;
2194 error = copyout(ifp->if_description,
2195 ifr->ifr_buffer.buffer, descrlen);
2196 ifr->ifr_buffer.length = descrlen;
2198 sx_sunlock(&ifdescr_sx);
2202 error = priv_check(td, PRIV_NET_SETIFDESCR);
2207 * Copy only (length-1) bytes to make sure that
2208 * if_description is always nul terminated. The
2209 * length parameter is supposed to count the
2210 * terminating nul in.
2212 if (ifr->ifr_buffer.length > ifdescr_maxlen)
2213 return (ENAMETOOLONG);
2214 else if (ifr->ifr_buffer.length == 0)
2217 descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2219 error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2220 ifr->ifr_buffer.length - 1);
2222 free(descrbuf, M_IFDESCR);
2227 sx_xlock(&ifdescr_sx);
2228 odescrbuf = ifp->if_description;
2229 ifp->if_description = descrbuf;
2230 sx_xunlock(&ifdescr_sx);
2232 getmicrotime(&ifp->if_lastchange);
2233 free(odescrbuf, M_IFDESCR);
2237 ifr->ifr_fib = ifp->if_fib;
2241 error = priv_check(td, PRIV_NET_SETIFFIB);
2244 if (ifr->ifr_fib >= rt_numfibs)
2247 ifp->if_fib = ifr->ifr_fib;
2251 error = priv_check(td, PRIV_NET_SETIFFLAGS);
2255 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2256 * check, so we don't need special handling here yet.
2258 new_flags = (ifr->ifr_flags & 0xffff) |
2259 (ifr->ifr_flagshigh << 16);
2260 if (ifp->if_flags & IFF_SMART) {
2261 /* Smart drivers twiddle their own routes */
2262 } else if (ifp->if_flags & IFF_UP &&
2263 (new_flags & IFF_UP) == 0) {
2267 } else if (new_flags & IFF_UP &&
2268 (ifp->if_flags & IFF_UP) == 0) {
2273 /* See if permanently promiscuous mode bit is about to flip */
2274 if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2275 if (new_flags & IFF_PPROMISC)
2276 ifp->if_flags |= IFF_PROMISC;
2277 else if (ifp->if_pcount == 0)
2278 ifp->if_flags &= ~IFF_PROMISC;
2279 log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2281 (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
2283 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2284 (new_flags &~ IFF_CANTCHANGE);
2285 if (ifp->if_ioctl) {
2286 (void) (*ifp->if_ioctl)(ifp, cmd, data);
2288 getmicrotime(&ifp->if_lastchange);
2292 error = priv_check(td, PRIV_NET_SETIFCAP);
2295 if (ifp->if_ioctl == NULL)
2296 return (EOPNOTSUPP);
2297 if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2299 error = (*ifp->if_ioctl)(ifp, cmd, data);
2301 getmicrotime(&ifp->if_lastchange);
2306 error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2311 error = priv_check(td, PRIV_NET_SETIFNAME);
2314 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2317 if (new_name[0] == '\0')
2319 if (ifunit(new_name) != NULL)
2323 * XXX: Locking. Nothing else seems to lock if_flags,
2324 * and there are numerous other races with the
2325 * ifunit() checks not being atomic with namespace
2326 * changes (renames, vmoves, if_attach, etc).
2328 ifp->if_flags |= IFF_RENAMING;
2330 /* Announce the departure of the interface. */
2331 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2332 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2334 log(LOG_INFO, "%s: changing name to '%s'\n",
2335 ifp->if_xname, new_name);
2337 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2340 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2341 namelen = strlen(new_name);
2342 onamelen = sdl->sdl_nlen;
2344 * Move the address if needed. This is safe because we
2345 * allocate space for a name of length IFNAMSIZ when we
2346 * create this in if_attach().
2348 if (namelen != onamelen) {
2349 bcopy(sdl->sdl_data + onamelen,
2350 sdl->sdl_data + namelen, sdl->sdl_alen);
2352 bcopy(new_name, sdl->sdl_data, namelen);
2353 sdl->sdl_nlen = namelen;
2354 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2355 bzero(sdl->sdl_data, onamelen);
2356 while (namelen != 0)
2357 sdl->sdl_data[--namelen] = 0xff;
2360 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2361 /* Announce the return of the interface. */
2362 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2364 ifp->if_flags &= ~IFF_RENAMING;
2369 error = priv_check(td, PRIV_NET_SETIFVNET);
2372 error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2377 error = priv_check(td, PRIV_NET_SETIFMETRIC);
2380 ifp->if_metric = ifr->ifr_metric;
2381 getmicrotime(&ifp->if_lastchange);
2385 error = priv_check(td, PRIV_NET_SETIFPHYS);
2388 if (ifp->if_ioctl == NULL)
2389 return (EOPNOTSUPP);
2390 error = (*ifp->if_ioctl)(ifp, cmd, data);
2392 getmicrotime(&ifp->if_lastchange);
2397 u_long oldmtu = ifp->if_mtu;
2399 error = priv_check(td, PRIV_NET_SETIFMTU);
2402 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2404 if (ifp->if_ioctl == NULL)
2405 return (EOPNOTSUPP);
2406 error = (*ifp->if_ioctl)(ifp, cmd, data);
2408 getmicrotime(&ifp->if_lastchange);
2412 * If the link MTU changed, do network layer specific procedure.
2414 if (ifp->if_mtu != oldmtu) {
2424 if (cmd == SIOCADDMULTI)
2425 error = priv_check(td, PRIV_NET_ADDMULTI);
2427 error = priv_check(td, PRIV_NET_DELMULTI);
2431 /* Don't allow group membership on non-multicast interfaces. */
2432 if ((ifp->if_flags & IFF_MULTICAST) == 0)
2433 return (EOPNOTSUPP);
2435 /* Don't let users screw up protocols' entries. */
2436 if (ifr->ifr_addr.sa_family != AF_LINK)
2439 if (cmd == SIOCADDMULTI) {
2440 struct ifmultiaddr *ifma;
2443 * Userland is only permitted to join groups once
2444 * via the if_addmulti() KPI, because it cannot hold
2445 * struct ifmultiaddr * between calls. It may also
2446 * lose a race while we check if the membership
2450 ifma = if_findmulti(ifp, &ifr->ifr_addr);
2451 IF_ADDR_RUNLOCK(ifp);
2455 error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2457 error = if_delmulti(ifp, &ifr->ifr_addr);
2460 getmicrotime(&ifp->if_lastchange);
2463 case SIOCSIFPHYADDR:
2464 case SIOCDIFPHYADDR:
2466 case SIOCSIFPHYADDR_IN6:
2468 case SIOCSLIFPHYADDR:
2470 case SIOCSIFGENERIC:
2471 error = priv_check(td, PRIV_NET_HWIOCTL);
2474 if (ifp->if_ioctl == NULL)
2475 return (EOPNOTSUPP);
2476 error = (*ifp->if_ioctl)(ifp, cmd, data);
2478 getmicrotime(&ifp->if_lastchange);
2482 ifs = (struct ifstat *)data;
2483 ifs->ascii[0] = '\0';
2485 case SIOCGIFPSRCADDR:
2486 case SIOCGIFPDSTADDR:
2487 case SIOCGLIFPHYADDR:
2489 case SIOCGIFGENERIC:
2490 if (ifp->if_ioctl == NULL)
2491 return (EOPNOTSUPP);
2492 error = (*ifp->if_ioctl)(ifp, cmd, data);
2496 error = priv_check(td, PRIV_NET_SETLLADDR);
2499 error = if_setlladdr(ifp,
2500 ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2501 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2506 struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2508 error = priv_check(td, PRIV_NET_ADDIFGROUP);
2511 if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2517 if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2523 struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2525 error = priv_check(td, PRIV_NET_DELIFGROUP);
2528 if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2540 #ifdef COMPAT_FREEBSD32
2548 #define SIOCGIFCONF32 _IOWR('i', 36, struct ifconf32)
2555 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2562 CURVNET_SET(so->so_vnet);
2566 error = ifconf(cmd, data);
2570 #ifdef COMPAT_FREEBSD32
2573 struct ifconf32 *ifc32;
2576 ifc32 = (struct ifconf32 *)data;
2577 ifc.ifc_len = ifc32->ifc_len;
2578 ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2580 error = ifconf(SIOCGIFCONF, (void *)&ifc);
2583 ifc32->ifc_len = ifc.ifc_len;
2588 ifr = (struct ifreq *)data;
2593 error = priv_check(td, PRIV_NET_SETIFVNET);
2595 error = if_vmove_reclaim(td, ifr->ifr_name,
2602 error = priv_check(td, PRIV_NET_IFCREATE);
2604 error = if_clone_create(ifr->ifr_name,
2605 sizeof(ifr->ifr_name),
2606 cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
2610 error = priv_check(td, PRIV_NET_IFDESTROY);
2612 error = if_clone_destroy(ifr->ifr_name);
2616 case SIOCIFGCLONERS:
2617 error = if_clone_list((struct if_clonereq *)data);
2621 error = if_getgroupmembers((struct ifgroupreq *)data);
2626 ifp = ifunit_ref(ifr->ifr_name);
2632 error = ifhwioctl(cmd, ifp, data, td);
2633 if (error != ENOIOCTL) {
2639 oif_flags = ifp->if_flags;
2640 if (so->so_proto == NULL) {
2643 return (EOPNOTSUPP);
2647 * Pass the request on to the socket control method, and if the
2648 * latter returns EOPNOTSUPP, directly to the interface.
2650 * Make an exception for the legacy SIOCSIF* requests. Drivers
2651 * trust SIOCSIFADDR et al to come from an already privileged
2652 * layer, and do not perform any credentials checks or input
2656 error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
2659 if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL &&
2660 cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR &&
2661 cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK)
2662 error = (*ifp->if_ioctl)(ifp, cmd, data);
2669 case SIOCSIFDSTADDR:
2671 case SIOCSIFBRDADDR:
2672 case SIOCSIFNETMASK:
2673 #if BYTE_ORDER != BIG_ENDIAN
2674 if (ifr->ifr_addr.sa_family == 0 &&
2675 ifr->ifr_addr.sa_len < 16) {
2676 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2677 ifr->ifr_addr.sa_len = 16;
2680 if (ifr->ifr_addr.sa_len == 0)
2681 ifr->ifr_addr.sa_len = 16;
2689 case OSIOCGIFDSTADDR:
2690 cmd = SIOCGIFDSTADDR;
2693 case OSIOCGIFBRDADDR:
2694 cmd = SIOCGIFBRDADDR;
2697 case OSIOCGIFNETMASK:
2698 cmd = SIOCGIFNETMASK;
2700 error = ((*so->so_proto->pr_usrreqs->pru_control)(so,
2704 if (error == EOPNOTSUPP && ifp != NULL &&
2705 ifp->if_ioctl != NULL &&
2706 cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR &&
2707 cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK)
2708 error = (*ifp->if_ioctl)(ifp, cmd, data);
2712 case OSIOCGIFDSTADDR:
2713 case OSIOCGIFBRDADDR:
2714 case OSIOCGIFNETMASK:
2715 *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2719 #endif /* COMPAT_43 */
2721 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2723 if (ifp->if_flags & IFF_UP) {
2736 * The code common to handling reference counted flags,
2737 * e.g., in ifpromisc() and if_allmulti().
2738 * The "pflag" argument can specify a permanent mode flag to check,
2739 * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2741 * Only to be used on stack-owned flags, not driver-owned flags.
2744 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2748 int oldflags, oldcount;
2750 /* Sanity checks to catch programming errors */
2751 KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2752 ("%s: setting driver-owned flag %d", __func__, flag));
2755 KASSERT(*refcount >= 0,
2756 ("%s: increment negative refcount %d for flag %d",
2757 __func__, *refcount, flag));
2759 KASSERT(*refcount > 0,
2760 ("%s: decrement non-positive refcount %d for flag %d",
2761 __func__, *refcount, flag));
2763 /* In case this mode is permanent, just touch refcount */
2764 if (ifp->if_flags & pflag) {
2765 *refcount += onswitch ? 1 : -1;
2769 /* Save ifnet parameters for if_ioctl() may fail */
2770 oldcount = *refcount;
2771 oldflags = ifp->if_flags;
2774 * See if we aren't the only and touching refcount is enough.
2775 * Actually toggle interface flag if we are the first or last.
2780 ifp->if_flags |= flag;
2784 ifp->if_flags &= ~flag;
2787 /* Call down the driver since we've changed interface flags */
2788 if (ifp->if_ioctl == NULL) {
2792 ifr.ifr_flags = ifp->if_flags & 0xffff;
2793 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2794 error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2797 /* Notify userland that interface flags have changed */
2802 /* Recover after driver error */
2803 *refcount = oldcount;
2804 ifp->if_flags = oldflags;
2809 * Set/clear promiscuous mode on interface ifp based on the truth value
2810 * of pswitch. The calls are reference counted so that only the first
2811 * "on" request actually has an effect, as does the final "off" request.
2812 * Results are undefined if the "off" and "on" requests are not matched.
2815 ifpromisc(struct ifnet *ifp, int pswitch)
2818 int oldflags = ifp->if_flags;
2820 error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2821 &ifp->if_pcount, pswitch);
2822 /* If promiscuous mode status has changed, log a message */
2823 if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2824 log(LOG_INFO, "%s: promiscuous mode %s\n",
2826 (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2831 * Return interface configuration
2832 * of system. List may be used
2833 * in later ioctl's (above) to get
2834 * other information.
2838 ifconf(u_long cmd, caddr_t data)
2840 struct ifconf *ifc = (struct ifconf *)data;
2845 int error, full = 0, valid_len, max_len;
2847 /* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2848 max_len = MAXPHYS - 1;
2850 /* Prevent hostile input from being able to crash the system */
2851 if (ifc->ifc_len <= 0)
2855 if (ifc->ifc_len <= max_len) {
2856 max_len = ifc->ifc_len;
2859 sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2864 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2868 * Zero the ifr_name buffer to make sure we don't
2869 * disclose the contents of the stack.
2871 memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2873 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2874 >= sizeof(ifr.ifr_name)) {
2877 return (ENAMETOOLONG);
2882 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2883 struct sockaddr *sa = ifa->ifa_addr;
2885 if (prison_if(curthread->td_ucred, sa) != 0)
2889 if (cmd == OSIOCGIFCONF) {
2890 struct osockaddr *osa =
2891 (struct osockaddr *)&ifr.ifr_addr;
2893 osa->sa_family = sa->sa_family;
2894 sbuf_bcat(sb, &ifr, sizeof(ifr));
2895 max_len += sizeof(ifr);
2898 if (sa->sa_len <= sizeof(*sa)) {
2900 sbuf_bcat(sb, &ifr, sizeof(ifr));
2901 max_len += sizeof(ifr);
2904 offsetof(struct ifreq, ifr_addr));
2905 max_len += offsetof(struct ifreq, ifr_addr);
2906 sbuf_bcat(sb, sa, sa->sa_len);
2907 max_len += sa->sa_len;
2910 if (sbuf_error(sb) == 0)
2911 valid_len = sbuf_len(sb);
2913 IF_ADDR_RUNLOCK(ifp);
2915 bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2916 sbuf_bcat(sb, &ifr, sizeof(ifr));
2917 max_len += sizeof(ifr);
2919 if (sbuf_error(sb) == 0)
2920 valid_len = sbuf_len(sb);
2926 * If we didn't allocate enough space (uncommon), try again. If
2927 * we have already allocated as much space as we are allowed,
2928 * return what we've got.
2930 if (valid_len != max_len && !full) {
2935 ifc->ifc_len = valid_len;
2937 error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2943 * Just like ifpromisc(), but for all-multicast-reception mode.
2946 if_allmulti(struct ifnet *ifp, int onswitch)
2949 return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2952 struct ifmultiaddr *
2953 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2955 struct ifmultiaddr *ifma;
2957 IF_ADDR_LOCK_ASSERT(ifp);
2959 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2960 if (sa->sa_family == AF_LINK) {
2961 if (sa_dl_equal(ifma->ifma_addr, sa))
2964 if (sa_equal(ifma->ifma_addr, sa))
2973 * Allocate a new ifmultiaddr and initialize based on passed arguments. We
2974 * make copies of passed sockaddrs. The ifmultiaddr will not be added to
2975 * the ifnet multicast address list here, so the caller must do that and
2976 * other setup work (such as notifying the device driver). The reference
2977 * count is initialized to 1.
2979 static struct ifmultiaddr *
2980 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2983 struct ifmultiaddr *ifma;
2984 struct sockaddr *dupsa;
2986 ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2991 dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2992 if (dupsa == NULL) {
2993 free(ifma, M_IFMADDR);
2996 bcopy(sa, dupsa, sa->sa_len);
2997 ifma->ifma_addr = dupsa;
2999 ifma->ifma_ifp = ifp;
3000 ifma->ifma_refcount = 1;
3001 ifma->ifma_protospec = NULL;
3004 ifma->ifma_lladdr = NULL;
3008 dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
3009 if (dupsa == NULL) {
3010 free(ifma->ifma_addr, M_IFMADDR);
3011 free(ifma, M_IFMADDR);
3014 bcopy(llsa, dupsa, llsa->sa_len);
3015 ifma->ifma_lladdr = dupsa;
3021 * if_freemulti: free ifmultiaddr structure and possibly attached related
3022 * addresses. The caller is responsible for implementing reference
3023 * counting, notifying the driver, handling routing messages, and releasing
3024 * any dependent link layer state.
3027 if_freemulti(struct ifmultiaddr *ifma)
3030 KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
3031 ifma->ifma_refcount));
3032 KASSERT(ifma->ifma_protospec == NULL,
3033 ("if_freemulti: protospec not NULL"));
3035 if (ifma->ifma_lladdr != NULL)
3036 free(ifma->ifma_lladdr, M_IFMADDR);
3037 free(ifma->ifma_addr, M_IFMADDR);
3038 free(ifma, M_IFMADDR);
3042 * Register an additional multicast address with a network interface.
3044 * - If the address is already present, bump the reference count on the
3045 * address and return.
3046 * - If the address is not link-layer, look up a link layer address.
3047 * - Allocate address structures for one or both addresses, and attach to the
3048 * multicast address list on the interface. If automatically adding a link
3049 * layer address, the protocol address will own a reference to the link
3050 * layer address, to be freed when it is freed.
3051 * - Notify the network device driver of an addition to the multicast address
3054 * 'sa' points to caller-owned memory with the desired multicast address.
3056 * 'retifma' will be used to return a pointer to the resulting multicast
3057 * address reference, if desired.
3060 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
3061 struct ifmultiaddr **retifma)
3063 struct ifmultiaddr *ifma, *ll_ifma;
3064 struct sockaddr *llsa;
3068 * If the address is already present, return a new reference to it;
3069 * otherwise, allocate storage and set up a new address.
3072 ifma = if_findmulti(ifp, sa);
3074 ifma->ifma_refcount++;
3075 if (retifma != NULL)
3077 IF_ADDR_WUNLOCK(ifp);
3082 * The address isn't already present; resolve the protocol address
3083 * into a link layer address, and then look that up, bump its
3084 * refcount or allocate an ifma for that also. If 'llsa' was
3085 * returned, we will need to free it later.
3089 if (ifp->if_resolvemulti != NULL) {
3090 error = ifp->if_resolvemulti(ifp, &llsa, sa);
3096 * Allocate the new address. Don't hook it up yet, as we may also
3097 * need to allocate a link layer multicast address.
3099 ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
3106 * If a link layer address is found, we'll need to see if it's
3107 * already present in the address list, or allocate is as well.
3108 * When this block finishes, the link layer address will be on the
3112 ll_ifma = if_findmulti(ifp, llsa);
3113 if (ll_ifma == NULL) {
3114 ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
3115 if (ll_ifma == NULL) {
3116 --ifma->ifma_refcount;
3121 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
3124 ll_ifma->ifma_refcount++;
3125 ifma->ifma_llifma = ll_ifma;
3129 * We now have a new multicast address, ifma, and possibly a new or
3130 * referenced link layer address. Add the primary address to the
3131 * ifnet address list.
3133 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3135 if (retifma != NULL)
3139 * Must generate the message while holding the lock so that 'ifma'
3140 * pointer is still valid.
3142 rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3143 IF_ADDR_WUNLOCK(ifp);
3146 * We are certain we have added something, so call down to the
3147 * interface to let them know about it.
3149 if (ifp->if_ioctl != NULL) {
3150 (void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3154 free(llsa, M_IFMADDR);
3160 free(llsa, M_IFMADDR);
3163 IF_ADDR_WUNLOCK(ifp);
3168 * Delete a multicast group membership by network-layer group address.
3170 * Returns ENOENT if the entry could not be found. If ifp no longer
3171 * exists, results are undefined. This entry point should only be used
3172 * from subsystems which do appropriate locking to hold ifp for the
3173 * duration of the call.
3174 * Network-layer protocol domains must use if_delmulti_ifma().
3177 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3179 struct ifmultiaddr *ifma;
3184 IFNET_RLOCK_NOSLEEP();
3185 TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3190 IFNET_RUNLOCK_NOSLEEP();
3192 KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3199 ifma = if_findmulti(ifp, sa);
3201 lastref = if_delmulti_locked(ifp, ifma, 0);
3202 IF_ADDR_WUNLOCK(ifp);
3207 if (lastref && ifp->if_ioctl != NULL) {
3208 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3215 * Delete all multicast group membership for an interface.
3216 * Should be used to quickly flush all multicast filters.
3219 if_delallmulti(struct ifnet *ifp)
3221 struct ifmultiaddr *ifma;
3222 struct ifmultiaddr *next;
3225 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3226 if_delmulti_locked(ifp, ifma, 0);
3227 IF_ADDR_WUNLOCK(ifp);
3231 * Delete a multicast group membership by group membership pointer.
3232 * Network-layer protocol domains must use this routine.
3234 * It is safe to call this routine if the ifp disappeared.
3237 if_delmulti_ifma(struct ifmultiaddr *ifma)
3242 ifp = ifma->ifma_ifp;
3245 printf("%s: ifma_ifp seems to be detached\n", __func__);
3249 IFNET_RLOCK_NOSLEEP();
3250 TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3254 printf("%s: ifnet %p disappeared\n", __func__, ifp);
3257 IFNET_RUNLOCK_NOSLEEP();
3261 * If and only if the ifnet instance exists: Acquire the address lock.
3266 lastref = if_delmulti_locked(ifp, ifma, 0);
3270 * If and only if the ifnet instance exists:
3271 * Release the address lock.
3272 * If the group was left: update the hardware hash filter.
3274 IF_ADDR_WUNLOCK(ifp);
3275 if (lastref && ifp->if_ioctl != NULL) {
3276 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3282 * Perform deletion of network-layer and/or link-layer multicast address.
3284 * Return 0 if the reference count was decremented.
3285 * Return 1 if the final reference was released, indicating that the
3286 * hardware hash filter should be reprogrammed.
3289 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3291 struct ifmultiaddr *ll_ifma;
3293 if (ifp != NULL && ifma->ifma_ifp != NULL) {
3294 KASSERT(ifma->ifma_ifp == ifp,
3295 ("%s: inconsistent ifp %p", __func__, ifp));
3296 IF_ADDR_WLOCK_ASSERT(ifp);
3299 ifp = ifma->ifma_ifp;
3302 * If the ifnet is detaching, null out references to ifnet,
3303 * so that upper protocol layers will notice, and not attempt
3304 * to obtain locks for an ifnet which no longer exists. The
3305 * routing socket announcement must happen before the ifnet
3306 * instance is detached from the system.
3310 printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3313 * ifp may already be nulled out if we are being reentered
3314 * to delete the ll_ifma.
3317 rt_newmaddrmsg(RTM_DELMADDR, ifma);
3318 ifma->ifma_ifp = NULL;
3322 if (--ifma->ifma_refcount > 0)
3326 * If this ifma is a network-layer ifma, a link-layer ifma may
3327 * have been associated with it. Release it first if so.
3329 ll_ifma = ifma->ifma_llifma;
3330 if (ll_ifma != NULL) {
3331 KASSERT(ifma->ifma_lladdr != NULL,
3332 ("%s: llifma w/o lladdr", __func__));
3334 ll_ifma->ifma_ifp = NULL; /* XXX */
3335 if (--ll_ifma->ifma_refcount == 0) {
3337 TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3340 if_freemulti(ll_ifma);
3345 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3350 * The last reference to this instance of struct ifmultiaddr
3351 * was released; the hardware should be notified of this change.
3357 * Set the link layer address on an interface.
3359 * At this time we only support certain types of interfaces,
3360 * and we don't allow the length of the address to change.
3363 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3365 struct sockaddr_dl *sdl;
3372 IF_ADDR_RUNLOCK(ifp);
3376 IF_ADDR_RUNLOCK(ifp);
3377 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3382 if (len != sdl->sdl_alen) { /* don't allow length to change */
3386 switch (ifp->if_type) {
3394 case IFT_IEEE8023ADLAG:
3396 bcopy(lladdr, LLADDR(sdl), len);
3405 * If the interface is already up, we need
3406 * to re-init it in order to reprogram its
3409 if ((ifp->if_flags & IFF_UP) != 0) {
3410 if (ifp->if_ioctl) {
3411 ifp->if_flags &= ~IFF_UP;
3412 ifr.ifr_flags = ifp->if_flags & 0xffff;
3413 ifr.ifr_flagshigh = ifp->if_flags >> 16;
3414 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3415 ifp->if_flags |= IFF_UP;
3416 ifr.ifr_flags = ifp->if_flags & 0xffff;
3417 ifr.ifr_flagshigh = ifp->if_flags >> 16;
3418 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3422 * Also send gratuitous ARPs to notify other nodes about
3423 * the address change.
3425 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3426 if (ifa->ifa_addr->sa_family == AF_INET)
3427 arp_ifinit(ifp, ifa);
3435 * The name argument must be a pointer to storage which will last as
3436 * long as the interface does. For physical devices, the result of
3437 * device_get_name(dev) is a good choice and for pseudo-devices a
3438 * static string works well.
3441 if_initname(struct ifnet *ifp, const char *name, int unit)
3443 ifp->if_dname = name;
3444 ifp->if_dunit = unit;
3445 if (unit != IF_DUNIT_NONE)
3446 snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3448 strlcpy(ifp->if_xname, name, IFNAMSIZ);
3452 if_printf(struct ifnet *ifp, const char * fmt, ...)
3457 retval = printf("%s: ", ifp->if_xname);
3459 retval += vprintf(fmt, ap);
3465 if_start(struct ifnet *ifp)
3468 (*(ifp)->if_start)(ifp);
3472 * Backwards compatibility interface for drivers
3473 * that have not implemented it
3476 if_transmit(struct ifnet *ifp, struct mbuf *m)
3480 IFQ_HANDOFF(ifp, m, error);
3485 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3490 if (_IF_QFULL(ifq)) {
3497 ifp->if_obytes += m->m_pkthdr.len + adjust;
3498 if (m->m_flags & (M_BCAST|M_MCAST))
3500 active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3502 _IF_ENQUEUE(ifq, m);
3504 if (ifp != NULL && !active)
3505 (*(ifp)->if_start)(ifp);
3510 if_register_com_alloc(u_char type,
3511 if_com_alloc_t *a, if_com_free_t *f)
3514 KASSERT(if_com_alloc[type] == NULL,
3515 ("if_register_com_alloc: %d already registered", type));
3516 KASSERT(if_com_free[type] == NULL,
3517 ("if_register_com_alloc: %d free already registered", type));
3519 if_com_alloc[type] = a;
3520 if_com_free[type] = f;
3524 if_deregister_com_alloc(u_char type)
3527 KASSERT(if_com_alloc[type] != NULL,
3528 ("if_deregister_com_alloc: %d not registered", type));
3529 KASSERT(if_com_free[type] != NULL,
3530 ("if_deregister_com_alloc: %d free not registered", type));
3531 if_com_alloc[type] = NULL;
3532 if_com_free[type] = NULL;