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 * Compute the least common TSO limit.
590 if_hw_tsomax_common(struct ifnet *ifp, struct ifnet_hw_tsomax *pmax)
593 * 1) If there is no limit currently, take the limit from
594 * the network adapter.
596 * 2) If the network adapter has a limit below the current
599 if (pmax->tsomaxbytes == 0 || (ifp->if_hw_tsomax != 0 &&
600 ifp->if_hw_tsomax < pmax->tsomaxbytes)) {
601 pmax->tsomaxbytes = ifp->if_hw_tsomax;
603 if (pmax->tsomaxsegcount == 0 || (ifp->if_hw_tsomaxsegcount != 0 &&
604 ifp->if_hw_tsomaxsegcount < pmax->tsomaxsegcount)) {
605 pmax->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
607 if (pmax->tsomaxsegsize == 0 || (ifp->if_hw_tsomaxsegsize != 0 &&
608 ifp->if_hw_tsomaxsegsize < pmax->tsomaxsegsize)) {
609 pmax->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
614 * Update TSO limit of a network adapter.
616 * Returns zero if no change. Else non-zero.
619 if_hw_tsomax_update(struct ifnet *ifp, struct ifnet_hw_tsomax *pmax)
622 if (ifp->if_hw_tsomax != pmax->tsomaxbytes) {
623 ifp->if_hw_tsomax = pmax->tsomaxbytes;
626 if (ifp->if_hw_tsomaxsegsize != pmax->tsomaxsegsize) {
627 ifp->if_hw_tsomaxsegsize = pmax->tsomaxsegsize;
630 if (ifp->if_hw_tsomaxsegcount != pmax->tsomaxsegcount) {
631 ifp->if_hw_tsomaxsegcount = pmax->tsomaxsegcount;
638 if_attach_internal(struct ifnet *ifp, int vmove)
640 unsigned socksize, ifasize;
641 int namelen, masklen;
642 struct sockaddr_dl *sdl;
645 if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
646 panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
650 ifp->if_vnet = curvnet;
651 if (ifp->if_home_vnet == NULL)
652 ifp->if_home_vnet = curvnet;
655 if_addgroup(ifp, IFG_ALL);
657 getmicrotime(&ifp->if_lastchange);
658 ifp->if_data.ifi_epoch = time_uptime;
659 ifp->if_data.ifi_datalen = sizeof(struct if_data);
661 KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
662 (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
663 ("transmit and qflush must both either be set or both be NULL"));
664 if (ifp->if_transmit == NULL) {
665 ifp->if_transmit = if_transmit;
666 ifp->if_qflush = if_qflush;
671 mac_ifnet_create(ifp);
675 * Create a Link Level name for this device.
677 namelen = strlen(ifp->if_xname);
679 * Always save enough space for any possiable name so we
680 * can do a rename in place later.
682 masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
683 socksize = masklen + ifp->if_addrlen;
684 if (socksize < sizeof(*sdl))
685 socksize = sizeof(*sdl);
686 socksize = roundup2(socksize, sizeof(long));
687 ifasize = sizeof(*ifa) + 2 * socksize;
688 ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
690 sdl = (struct sockaddr_dl *)(ifa + 1);
691 sdl->sdl_len = socksize;
692 sdl->sdl_family = AF_LINK;
693 bcopy(ifp->if_xname, sdl->sdl_data, namelen);
694 sdl->sdl_nlen = namelen;
695 sdl->sdl_index = ifp->if_index;
696 sdl->sdl_type = ifp->if_type;
699 ifa->ifa_rtrequest = link_rtrequest;
700 ifa->ifa_addr = (struct sockaddr *)sdl;
701 sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
702 ifa->ifa_netmask = (struct sockaddr *)sdl;
703 sdl->sdl_len = masklen;
705 sdl->sdl_data[--namelen] = 0xff;
706 TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
707 /* Reliably crash if used uninitialized. */
708 ifp->if_broadcastaddr = NULL;
710 #if defined(INET) || defined(INET6)
711 /* Use defaults for TSO, if nothing is set */
712 if (ifp->if_hw_tsomax == 0 &&
713 ifp->if_hw_tsomaxsegcount == 0 &&
714 ifp->if_hw_tsomaxsegsize == 0) {
716 * The TSO defaults needs to be such that an
717 * NFS mbuf list of 35 mbufs totalling just
718 * below 64K works and that a chain of mbufs
719 * can be defragged into at most 32 segments:
721 ifp->if_hw_tsomax = min(IP_MAXPACKET, (32 * MCLBYTES) -
722 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
723 ifp->if_hw_tsomaxsegcount = 35;
724 ifp->if_hw_tsomaxsegsize = 2048; /* 2K */
726 /* XXX some drivers set IFCAP_TSO after ethernet attach */
727 if (ifp->if_capabilities & IFCAP_TSO) {
728 if_printf(ifp, "Using defaults for TSO: %u/%u/%u\n",
730 ifp->if_hw_tsomaxsegcount,
731 ifp->if_hw_tsomaxsegsize);
735 * If the "if_hw_tsomax" limit is set, check if it is
738 KASSERT(ifp->if_hw_tsomax == 0 ||
739 ifp->if_hw_tsomax >= (IP_MAXPACKET / 8),
740 ("%s: if_hw_tsomax is outside of range", __func__));
746 * Update the interface index in the link layer address
749 for (ifa = ifp->if_addr; ifa != NULL;
750 ifa = TAILQ_NEXT(ifa, ifa_link)) {
751 if (ifa->ifa_addr->sa_family == AF_LINK) {
752 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
753 sdl->sdl_index = ifp->if_index;
760 TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
762 curvnet->vnet_ifcnt++;
766 if (domain_init_status >= 2)
767 if_attachdomain1(ifp);
769 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
770 if (IS_DEFAULT_VNET(curvnet))
771 devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
773 /* Announce the interface. */
774 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
778 if_attachdomain(void *dummy)
782 TAILQ_FOREACH(ifp, &V_ifnet, if_link)
783 if_attachdomain1(ifp);
785 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
786 if_attachdomain, NULL);
789 if_attachdomain1(struct ifnet *ifp)
794 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
795 * cannot lock ifp->if_afdata initialization, entirely.
797 if (IF_AFDATA_TRYLOCK(ifp) == 0)
799 if (ifp->if_afdata_initialized >= domain_init_status) {
800 IF_AFDATA_UNLOCK(ifp);
801 log(LOG_WARNING, "%s called more than once on %s\n",
802 __func__, ifp->if_xname);
805 ifp->if_afdata_initialized = domain_init_status;
806 IF_AFDATA_UNLOCK(ifp);
808 /* address family dependent data region */
809 bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
810 for (dp = domains; dp; dp = dp->dom_next) {
811 if (dp->dom_ifattach)
812 ifp->if_afdata[dp->dom_family] =
813 (*dp->dom_ifattach)(ifp);
818 * Remove any unicast or broadcast network addresses from an interface.
821 if_purgeaddrs(struct ifnet *ifp)
823 struct ifaddr *ifa, *next;
825 TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
826 if (ifa->ifa_addr->sa_family == AF_LINK)
829 /* XXX: Ugly!! ad hoc just for INET */
830 if (ifa->ifa_addr->sa_family == AF_INET) {
831 struct ifaliasreq ifr;
833 bzero(&ifr, sizeof(ifr));
834 ifr.ifra_addr = *ifa->ifa_addr;
835 if (ifa->ifa_dstaddr)
836 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
837 if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
843 if (ifa->ifa_addr->sa_family == AF_INET6) {
845 /* ifp_addrhead is already updated */
849 TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
855 * Remove any multicast network addresses from an interface when an ifnet
859 if_purgemaddrs(struct ifnet *ifp)
861 struct ifmultiaddr *ifma;
862 struct ifmultiaddr *next;
865 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
866 if_delmulti_locked(ifp, ifma, 1);
867 IF_ADDR_WUNLOCK(ifp);
871 * Detach an interface, removing it from the list of "active" interfaces.
872 * If vmove flag is set on entry to if_detach_internal(), perform only a
873 * limited subset of cleanup tasks, given that we are moving an ifnet from
874 * one vnet to another, where it must be fully operational.
876 * XXXRW: There are some significant questions about event ordering, and
877 * how to prevent things from starting to use the interface during detach.
880 if_detach(struct ifnet *ifp)
883 CURVNET_SET_QUIET(ifp->if_vnet);
884 if_detach_internal(ifp, 0);
889 if_detach_internal(struct ifnet *ifp, int vmove)
892 struct radix_node_head *rnh;
899 TAILQ_FOREACH(iter, &V_ifnet, if_link)
901 TAILQ_REMOVE(&V_ifnet, ifp, if_link);
907 curvnet->vnet_ifcnt--;
912 panic("%s: ifp=%p not on the ifnet tailq %p",
913 __func__, ifp, &V_ifnet);
915 return; /* XXX this should panic as well? */
919 * Remove/wait for pending events.
921 taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
924 * Remove routes and flush queues.
928 if (ALTQ_IS_ENABLED(&ifp->if_snd))
929 altq_disable(&ifp->if_snd);
930 if (ALTQ_IS_ATTACHED(&ifp->if_snd))
931 altq_detach(&ifp->if_snd);
942 * Remove all IPv6 kernel structs related to ifp. This should be done
943 * before removing routing entries below, since IPv6 interface direct
944 * routes are expected to be removed by the IPv6-specific kernel API.
945 * Otherwise, the kernel will detect some inconsistency and bark it.
953 * Prevent further calls into the device driver via ifnet.
958 * Remove link ifaddr pointer and maybe decrement if_index.
959 * Clean up all addresses.
963 /* We can now free link ifaddr. */
964 if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
965 ifa = TAILQ_FIRST(&ifp->if_addrhead);
966 TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
972 * Delete all remaining routes using this interface
973 * Unfortuneatly the only way to do this is to slog through
974 * the entire routing table looking for routes which point
975 * to this interface...oh well...
977 for (i = 1; i <= AF_MAX; i++) {
978 for (j = 0; j < rt_numfibs; j++) {
979 rnh = rt_tables_get_rnh(j, i);
982 RADIX_NODE_HEAD_LOCK(rnh);
983 (void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
984 RADIX_NODE_HEAD_UNLOCK(rnh);
988 /* Announce that the interface is gone. */
989 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
990 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
991 if (IS_DEFAULT_VNET(curvnet))
992 devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
996 * We cannot hold the lock over dom_ifdetach calls as they might
997 * sleep, for example trying to drain a callout, thus open up the
998 * theoretical race with re-attaching.
1000 IF_AFDATA_LOCK(ifp);
1001 i = ifp->if_afdata_initialized;
1002 ifp->if_afdata_initialized = 0;
1003 IF_AFDATA_UNLOCK(ifp);
1004 for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
1005 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
1006 (*dp->dom_ifdetach)(ifp,
1007 ifp->if_afdata[dp->dom_family]);
1013 * if_vmove() performs a limited version of if_detach() in current
1014 * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
1015 * An attempt is made to shrink if_index in current vnet, find an
1016 * unused if_index in target vnet and calls if_grow() if necessary,
1017 * and finally find an unused if_xname for the target vnet.
1020 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
1025 * Detach from current vnet, but preserve LLADDR info, do not
1026 * mark as dead etc. so that the ifnet can be reattached later.
1028 if_detach_internal(ifp, 1);
1031 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
1032 * the if_index for that vnet if possible.
1034 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
1035 * or we'd lock on one vnet and unlock on another.
1038 ifindex_free_locked(ifp->if_index);
1042 * Perform interface-specific reassignment tasks, if provided by
1045 if (ifp->if_reassign != NULL)
1046 ifp->if_reassign(ifp, new_vnet, NULL);
1049 * Switch to the context of the target vnet.
1051 CURVNET_SET_QUIET(new_vnet);
1054 if (ifindex_alloc_locked(&idx) != 0) {
1056 panic("if_index overflow");
1058 ifp->if_index = idx;
1059 ifnet_setbyindex_locked(ifp->if_index, ifp);
1062 if_attach_internal(ifp, 1);
1068 * Move an ifnet to or from another child prison/vnet, specified by the jail id.
1071 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
1076 /* Try to find the prison within our visibility. */
1077 sx_slock(&allprison_lock);
1078 pr = prison_find_child(td->td_ucred->cr_prison, jid);
1079 sx_sunlock(&allprison_lock);
1082 prison_hold_locked(pr);
1083 mtx_unlock(&pr->pr_mtx);
1085 /* Do not try to move the iface from and to the same prison. */
1086 if (pr->pr_vnet == ifp->if_vnet) {
1091 /* Make sure the named iface does not exists in the dst. prison/vnet. */
1092 /* XXX Lock interfaces to avoid races. */
1093 CURVNET_SET_QUIET(pr->pr_vnet);
1094 difp = ifunit(ifname);
1101 /* Move the interface into the child jail/vnet. */
1102 if_vmove(ifp, pr->pr_vnet);
1104 /* Report the new if_xname back to the userland. */
1105 sprintf(ifname, "%s", ifp->if_xname);
1112 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1115 struct vnet *vnet_dst;
1118 /* Try to find the prison within our visibility. */
1119 sx_slock(&allprison_lock);
1120 pr = prison_find_child(td->td_ucred->cr_prison, jid);
1121 sx_sunlock(&allprison_lock);
1124 prison_hold_locked(pr);
1125 mtx_unlock(&pr->pr_mtx);
1127 /* Make sure the named iface exists in the source prison/vnet. */
1128 CURVNET_SET(pr->pr_vnet);
1129 ifp = ifunit(ifname); /* XXX Lock to avoid races. */
1136 /* Do not try to move the iface from and to the same prison. */
1137 vnet_dst = TD_TO_VNET(td);
1138 if (vnet_dst == ifp->if_vnet) {
1144 /* Get interface back from child jail/vnet. */
1145 if_vmove(ifp, vnet_dst);
1148 /* Report the new if_xname back to the userland. */
1149 sprintf(ifname, "%s", ifp->if_xname);
1157 * Add a group to an interface
1160 if_addgroup(struct ifnet *ifp, const char *groupname)
1162 struct ifg_list *ifgl;
1163 struct ifg_group *ifg = NULL;
1164 struct ifg_member *ifgm;
1167 if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1168 groupname[strlen(groupname) - 1] <= '9')
1172 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1173 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1178 if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1179 M_NOWAIT)) == NULL) {
1184 if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1185 M_TEMP, M_NOWAIT)) == NULL) {
1191 TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1192 if (!strcmp(ifg->ifg_group, groupname))
1196 if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1197 M_TEMP, M_NOWAIT)) == NULL) {
1203 strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1204 ifg->ifg_refcnt = 0;
1205 TAILQ_INIT(&ifg->ifg_members);
1206 TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1211 ifgl->ifgl_group = ifg;
1212 ifgm->ifgm_ifp = ifp;
1215 TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1216 TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1217 IF_ADDR_WUNLOCK(ifp);
1222 EVENTHANDLER_INVOKE(group_attach_event, ifg);
1223 EVENTHANDLER_INVOKE(group_change_event, groupname);
1229 * Remove a group from an interface
1232 if_delgroup(struct ifnet *ifp, const char *groupname)
1234 struct ifg_list *ifgl;
1235 struct ifg_member *ifgm;
1238 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1239 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1247 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1248 IF_ADDR_WUNLOCK(ifp);
1250 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1251 if (ifgm->ifgm_ifp == ifp)
1255 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1259 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1260 TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1262 EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1263 free(ifgl->ifgl_group, M_TEMP);
1269 EVENTHANDLER_INVOKE(group_change_event, groupname);
1275 * Remove an interface from all groups
1278 if_delgroups(struct ifnet *ifp)
1280 struct ifg_list *ifgl;
1281 struct ifg_member *ifgm;
1282 char groupname[IFNAMSIZ];
1285 while (!TAILQ_EMPTY(&ifp->if_groups)) {
1286 ifgl = TAILQ_FIRST(&ifp->if_groups);
1288 strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1291 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1292 IF_ADDR_WUNLOCK(ifp);
1294 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1295 if (ifgm->ifgm_ifp == ifp)
1299 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1304 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1305 TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1307 EVENTHANDLER_INVOKE(group_detach_event,
1309 free(ifgl->ifgl_group, M_TEMP);
1315 EVENTHANDLER_INVOKE(group_change_event, groupname);
1323 * Stores all groups from an interface in memory pointed
1327 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1330 struct ifg_list *ifgl;
1331 struct ifg_req ifgrq, *ifgp;
1332 struct ifgroupreq *ifgr = data;
1334 if (ifgr->ifgr_len == 0) {
1336 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1337 ifgr->ifgr_len += sizeof(struct ifg_req);
1338 IF_ADDR_RUNLOCK(ifp);
1342 len = ifgr->ifgr_len;
1343 ifgp = ifgr->ifgr_groups;
1346 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1347 if (len < sizeof(ifgrq)) {
1348 IF_ADDR_RUNLOCK(ifp);
1351 bzero(&ifgrq, sizeof ifgrq);
1352 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1353 sizeof(ifgrq.ifgrq_group));
1354 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1355 IF_ADDR_RUNLOCK(ifp);
1358 len -= sizeof(ifgrq);
1361 IF_ADDR_RUNLOCK(ifp);
1367 * Stores all members of a group in memory pointed to by data
1370 if_getgroupmembers(struct ifgroupreq *data)
1372 struct ifgroupreq *ifgr = data;
1373 struct ifg_group *ifg;
1374 struct ifg_member *ifgm;
1375 struct ifg_req ifgrq, *ifgp;
1379 TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1380 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1387 if (ifgr->ifgr_len == 0) {
1388 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1389 ifgr->ifgr_len += sizeof(ifgrq);
1394 len = ifgr->ifgr_len;
1395 ifgp = ifgr->ifgr_groups;
1396 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1397 if (len < sizeof(ifgrq)) {
1401 bzero(&ifgrq, sizeof ifgrq);
1402 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1403 sizeof(ifgrq.ifgrq_member));
1404 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1408 len -= sizeof(ifgrq);
1417 * Delete Routes for a Network Interface
1419 * Called for each routing entry via the rnh->rnh_walktree() call above
1420 * to delete all route entries referencing a detaching network interface.
1423 * rn pointer to node in the routing table
1424 * arg argument passed to rnh->rnh_walktree() - detaching interface
1428 * errno failed - reason indicated
1432 if_rtdel(struct radix_node *rn, void *arg)
1434 struct rtentry *rt = (struct rtentry *)rn;
1435 struct ifnet *ifp = arg;
1438 if (rt->rt_ifp == ifp) {
1441 * Protect (sorta) against walktree recursion problems
1442 * with cloned routes
1444 if ((rt->rt_flags & RTF_UP) == 0)
1447 err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1449 rt->rt_flags|RTF_RNH_LOCKED|RTF_PINNED,
1450 (struct rtentry **) NULL, rt->rt_fibnum);
1452 log(LOG_WARNING, "if_rtdel: error %d\n", err);
1460 * Wrapper functions for struct ifnet address list locking macros. These are
1461 * used by kernel modules to avoid encoding programming interface or binary
1462 * interface assumptions that may be violated when kernel-internal locking
1463 * approaches change.
1466 if_addr_rlock(struct ifnet *ifp)
1473 if_addr_runlock(struct ifnet *ifp)
1476 IF_ADDR_RUNLOCK(ifp);
1480 if_maddr_rlock(struct ifnet *ifp)
1487 if_maddr_runlock(struct ifnet *ifp)
1490 IF_ADDR_RUNLOCK(ifp);
1494 * Initialization, destruction and refcounting functions for ifaddrs.
1497 ifa_init(struct ifaddr *ifa)
1500 mtx_init(&ifa->ifa_mtx, "ifaddr", NULL, MTX_DEF);
1501 refcount_init(&ifa->ifa_refcnt, 1);
1502 ifa->if_data.ifi_datalen = sizeof(ifa->if_data);
1506 ifa_ref(struct ifaddr *ifa)
1509 refcount_acquire(&ifa->ifa_refcnt);
1513 ifa_free(struct ifaddr *ifa)
1516 if (refcount_release(&ifa->ifa_refcnt)) {
1517 mtx_destroy(&ifa->ifa_mtx);
1518 free(ifa, M_IFADDR);
1523 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1526 struct rtentry *rt = NULL;
1527 struct rt_addrinfo info;
1528 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1530 bzero(&info, sizeof(info));
1531 info.rti_ifp = V_loif;
1532 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1533 info.rti_info[RTAX_DST] = ia;
1534 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1535 error = rtrequest1_fib(RTM_ADD, &info, &rt, ifa->ifa_ifp->if_fib);
1537 if (error == 0 && rt != NULL) {
1539 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1540 ifa->ifa_ifp->if_type;
1541 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1542 ifa->ifa_ifp->if_index;
1545 } else if (error != 0)
1546 log(LOG_DEBUG, "%s: insertion failed: %u\n", __func__, error);
1552 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1555 struct rt_addrinfo info;
1556 struct sockaddr_dl null_sdl;
1558 bzero(&null_sdl, sizeof(null_sdl));
1559 null_sdl.sdl_len = sizeof(null_sdl);
1560 null_sdl.sdl_family = AF_LINK;
1561 null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1562 null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1563 bzero(&info, sizeof(info));
1564 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1565 info.rti_info[RTAX_DST] = ia;
1566 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1567 error = rtrequest1_fib(RTM_DELETE, &info, NULL, ifa->ifa_ifp->if_fib);
1570 log(LOG_DEBUG, "%s: deletion failed: %u\n", __func__, error);
1576 * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1577 * structs used to represent other address families, it is necessary
1578 * to perform a different comparison.
1581 #define sa_equal(a1, a2) \
1582 (bcmp((a1), (a2), ((a1))->sa_len) == 0)
1584 #define sa_dl_equal(a1, a2) \
1585 ((((struct sockaddr_dl *)(a1))->sdl_len == \
1586 ((struct sockaddr_dl *)(a2))->sdl_len) && \
1587 (bcmp(LLADDR((struct sockaddr_dl *)(a1)), \
1588 LLADDR((struct sockaddr_dl *)(a2)), \
1589 ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1592 * Locate an interface based on a complete address.
1595 static struct ifaddr *
1596 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1601 IFNET_RLOCK_NOSLEEP();
1602 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1604 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1605 if (ifa->ifa_addr->sa_family != addr->sa_family)
1607 if (sa_equal(addr, ifa->ifa_addr)) {
1610 IF_ADDR_RUNLOCK(ifp);
1613 /* IP6 doesn't have broadcast */
1614 if ((ifp->if_flags & IFF_BROADCAST) &&
1615 ifa->ifa_broadaddr &&
1616 ifa->ifa_broadaddr->sa_len != 0 &&
1617 sa_equal(ifa->ifa_broadaddr, addr)) {
1620 IF_ADDR_RUNLOCK(ifp);
1624 IF_ADDR_RUNLOCK(ifp);
1628 IFNET_RUNLOCK_NOSLEEP();
1633 ifa_ifwithaddr(struct sockaddr *addr)
1636 return (ifa_ifwithaddr_internal(addr, 1));
1640 ifa_ifwithaddr_check(struct sockaddr *addr)
1643 return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1647 * Locate an interface based on the broadcast address.
1651 ifa_ifwithbroadaddr(struct sockaddr *addr)
1656 IFNET_RLOCK_NOSLEEP();
1657 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1659 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1660 if (ifa->ifa_addr->sa_family != addr->sa_family)
1662 if ((ifp->if_flags & IFF_BROADCAST) &&
1663 ifa->ifa_broadaddr &&
1664 ifa->ifa_broadaddr->sa_len != 0 &&
1665 sa_equal(ifa->ifa_broadaddr, addr)) {
1667 IF_ADDR_RUNLOCK(ifp);
1671 IF_ADDR_RUNLOCK(ifp);
1675 IFNET_RUNLOCK_NOSLEEP();
1680 * Locate the point to point interface with a given destination address.
1684 ifa_ifwithdstaddr_fib(struct sockaddr *addr, int fibnum)
1689 IFNET_RLOCK_NOSLEEP();
1690 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1691 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1693 if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum))
1696 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1697 if (ifa->ifa_addr->sa_family != addr->sa_family)
1699 if (ifa->ifa_dstaddr != NULL &&
1700 sa_equal(addr, ifa->ifa_dstaddr)) {
1702 IF_ADDR_RUNLOCK(ifp);
1706 IF_ADDR_RUNLOCK(ifp);
1710 IFNET_RUNLOCK_NOSLEEP();
1715 ifa_ifwithdstaddr(struct sockaddr *addr)
1718 return (ifa_ifwithdstaddr_fib(addr, RT_ALL_FIBS));
1722 * Find an interface on a specific network. If many, choice
1723 * is most specific found.
1726 ifa_ifwithnet_fib(struct sockaddr *addr, int ignore_ptp, int fibnum)
1730 struct ifaddr *ifa_maybe = NULL;
1731 u_int af = addr->sa_family;
1732 char *addr_data = addr->sa_data, *cplim;
1735 * AF_LINK addresses can be looked up directly by their index number,
1736 * so do that if we can.
1738 if (af == AF_LINK) {
1739 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1740 if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1741 return (ifaddr_byindex(sdl->sdl_index));
1745 * Scan though each interface, looking for ones that have addresses
1746 * in this address family and the requested fib. Maintain a reference
1747 * on ifa_maybe once we find one, as we release the IF_ADDR_RLOCK() that
1748 * kept it stable when we move onto the next interface.
1750 IFNET_RLOCK_NOSLEEP();
1751 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1752 if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum))
1755 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1756 char *cp, *cp2, *cp3;
1758 if (ifa->ifa_addr->sa_family != af)
1760 if (af == AF_INET &&
1761 ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1763 * This is a bit broken as it doesn't
1764 * take into account that the remote end may
1765 * be a single node in the network we are
1767 * The trouble is that we don't know the
1768 * netmask for the remote end.
1770 if (ifa->ifa_dstaddr != NULL &&
1771 sa_equal(addr, ifa->ifa_dstaddr)) {
1773 IF_ADDR_RUNLOCK(ifp);
1778 * if we have a special address handler,
1779 * then use it instead of the generic one.
1781 if (ifa->ifa_claim_addr) {
1782 if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1784 IF_ADDR_RUNLOCK(ifp);
1791 * Scan all the bits in the ifa's address.
1792 * If a bit dissagrees with what we are
1793 * looking for, mask it with the netmask
1794 * to see if it really matters.
1795 * (A byte at a time)
1797 if (ifa->ifa_netmask == 0)
1800 cp2 = ifa->ifa_addr->sa_data;
1801 cp3 = ifa->ifa_netmask->sa_data;
1802 cplim = ifa->ifa_netmask->sa_len
1803 + (char *)ifa->ifa_netmask;
1805 if ((*cp++ ^ *cp2++) & *cp3++)
1806 goto next; /* next address! */
1808 * If the netmask of what we just found
1809 * is more specific than what we had before
1810 * (if we had one), or if the virtual status
1811 * of new prefix is better than of the old one,
1812 * then remember the new one before continuing
1813 * to search for an even better one.
1815 if (ifa_maybe == NULL ||
1816 ifa_preferred(ifa_maybe, ifa) ||
1817 rn_refines((caddr_t)ifa->ifa_netmask,
1818 (caddr_t)ifa_maybe->ifa_netmask)) {
1819 if (ifa_maybe != NULL)
1820 ifa_free(ifa_maybe);
1826 IF_ADDR_RUNLOCK(ifp);
1831 IFNET_RUNLOCK_NOSLEEP();
1832 if (ifa_maybe != NULL)
1833 ifa_free(ifa_maybe);
1838 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp)
1841 return (ifa_ifwithnet_fib(addr, ignore_ptp, RT_ALL_FIBS));
1845 * Find an interface address specific to an interface best matching
1849 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1852 char *cp, *cp2, *cp3;
1854 struct ifaddr *ifa_maybe = NULL;
1855 u_int af = addr->sa_family;
1860 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1861 if (ifa->ifa_addr->sa_family != af)
1863 if (ifa_maybe == NULL)
1865 if (ifa->ifa_netmask == 0) {
1866 if (sa_equal(addr, ifa->ifa_addr) ||
1867 (ifa->ifa_dstaddr &&
1868 sa_equal(addr, ifa->ifa_dstaddr)))
1872 if (ifp->if_flags & IFF_POINTOPOINT) {
1873 if (sa_equal(addr, ifa->ifa_dstaddr))
1877 cp2 = ifa->ifa_addr->sa_data;
1878 cp3 = ifa->ifa_netmask->sa_data;
1879 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1880 for (; cp3 < cplim; cp3++)
1881 if ((*cp++ ^ *cp2++) & *cp3)
1891 IF_ADDR_RUNLOCK(ifp);
1896 * See whether new ifa is better than current one:
1897 * 1) A non-virtual one is preferred over virtual.
1898 * 2) A virtual in master state preferred over any other state.
1900 * Used in several address selecting functions.
1903 ifa_preferred(struct ifaddr *cur, struct ifaddr *next)
1906 return (cur->ifa_carp && (!next->ifa_carp ||
1907 ((*carp_master_p)(next) && !(*carp_master_p)(cur))));
1910 #include <net/if_llatbl.h>
1913 * Default action when installing a route with a Link Level gateway.
1914 * Lookup an appropriate real ifa to point to.
1915 * This should be moved to /sys/net/link.c eventually.
1918 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1920 struct ifaddr *ifa, *oifa;
1921 struct sockaddr *dst;
1926 if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1927 ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1929 ifa = ifaof_ifpforaddr(dst, ifp);
1934 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1935 ifa->ifa_rtrequest(cmd, rt, info);
1940 * Mark an interface down and notify protocols of
1944 if_unroute(struct ifnet *ifp, int flag, int fam)
1948 KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1950 ifp->if_flags &= ~flag;
1951 getmicrotime(&ifp->if_lastchange);
1952 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1953 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1954 pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1955 ifp->if_qflush(ifp);
1958 (*carp_linkstate_p)(ifp);
1963 * Mark an interface up and notify protocols of
1967 if_route(struct ifnet *ifp, int flag, int fam)
1971 KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1973 ifp->if_flags |= flag;
1974 getmicrotime(&ifp->if_lastchange);
1975 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1976 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1977 pfctlinput(PRC_IFUP, ifa->ifa_addr);
1979 (*carp_linkstate_p)(ifp);
1986 void (*vlan_link_state_p)(struct ifnet *); /* XXX: private from if_vlan */
1987 void (*vlan_trunk_cap_p)(struct ifnet *); /* XXX: private from if_vlan */
1988 struct ifnet *(*vlan_trunkdev_p)(struct ifnet *);
1989 struct ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t);
1990 int (*vlan_tag_p)(struct ifnet *, uint16_t *);
1991 int (*vlan_setcookie_p)(struct ifnet *, void *);
1992 void *(*vlan_cookie_p)(struct ifnet *);
1995 * Handle a change in the interface link state. To avoid LORs
1996 * between driver lock and upper layer locks, as well as possible
1997 * recursions, we post event to taskqueue, and all job
1998 * is done in static do_link_state_change().
2001 if_link_state_change(struct ifnet *ifp, int link_state)
2003 /* Return if state hasn't changed. */
2004 if (ifp->if_link_state == link_state)
2007 ifp->if_link_state = link_state;
2009 taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
2013 do_link_state_change(void *arg, int pending)
2015 struct ifnet *ifp = (struct ifnet *)arg;
2016 int link_state = ifp->if_link_state;
2017 CURVNET_SET(ifp->if_vnet);
2019 /* Notify that the link state has changed. */
2021 if (ifp->if_vlantrunk != NULL)
2022 (*vlan_link_state_p)(ifp);
2024 if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
2025 IFP2AC(ifp)->ac_netgraph != NULL)
2026 (*ng_ether_link_state_p)(ifp, link_state);
2028 (*carp_linkstate_p)(ifp);
2030 (*bridge_linkstate_p)(ifp);
2032 (*lagg_linkstate_p)(ifp, link_state);
2034 if (IS_DEFAULT_VNET(curvnet))
2035 devctl_notify("IFNET", ifp->if_xname,
2036 (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
2039 if_printf(ifp, "%d link states coalesced\n", pending);
2040 if (log_link_state_change)
2041 log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
2042 (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
2043 EVENTHANDLER_INVOKE(ifnet_link_event, ifp, ifp->if_link_state);
2048 * Mark an interface down and notify protocols of
2052 if_down(struct ifnet *ifp)
2055 if_unroute(ifp, IFF_UP, AF_UNSPEC);
2059 * Mark an interface up and notify protocols of
2063 if_up(struct ifnet *ifp)
2066 if_route(ifp, IFF_UP, AF_UNSPEC);
2070 * Flush an interface queue.
2073 if_qflush(struct ifnet *ifp)
2081 if (ALTQ_IS_ENABLED(ifq))
2085 while ((m = n) != 0) {
2096 * Map interface name to interface structure pointer, with or without
2097 * returning a reference.
2100 ifunit_ref(const char *name)
2104 IFNET_RLOCK_NOSLEEP();
2105 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2106 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2107 !(ifp->if_flags & IFF_DYING))
2112 IFNET_RUNLOCK_NOSLEEP();
2117 ifunit(const char *name)
2121 IFNET_RLOCK_NOSLEEP();
2122 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2123 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2126 IFNET_RUNLOCK_NOSLEEP();
2131 * Hardware specific interface ioctls.
2134 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2139 int new_flags, temp_flags;
2140 size_t namelen, onamelen;
2142 char *descrbuf, *odescrbuf;
2143 char new_name[IFNAMSIZ];
2145 struct sockaddr_dl *sdl;
2147 ifr = (struct ifreq *)data;
2150 ifr->ifr_index = ifp->if_index;
2154 temp_flags = ifp->if_flags | ifp->if_drv_flags;
2155 ifr->ifr_flags = temp_flags & 0xffff;
2156 ifr->ifr_flagshigh = temp_flags >> 16;
2160 ifr->ifr_reqcap = ifp->if_capabilities;
2161 ifr->ifr_curcap = ifp->if_capenable;
2166 error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2171 ifr->ifr_metric = ifp->if_metric;
2175 ifr->ifr_mtu = ifp->if_mtu;
2179 ifr->ifr_phys = ifp->if_physical;
2184 sx_slock(&ifdescr_sx);
2185 if (ifp->if_description == NULL)
2188 /* space for terminating nul */
2189 descrlen = strlen(ifp->if_description) + 1;
2190 if (ifr->ifr_buffer.length < descrlen)
2191 ifr->ifr_buffer.buffer = NULL;
2193 error = copyout(ifp->if_description,
2194 ifr->ifr_buffer.buffer, descrlen);
2195 ifr->ifr_buffer.length = descrlen;
2197 sx_sunlock(&ifdescr_sx);
2201 error = priv_check(td, PRIV_NET_SETIFDESCR);
2206 * Copy only (length-1) bytes to make sure that
2207 * if_description is always nul terminated. The
2208 * length parameter is supposed to count the
2209 * terminating nul in.
2211 if (ifr->ifr_buffer.length > ifdescr_maxlen)
2212 return (ENAMETOOLONG);
2213 else if (ifr->ifr_buffer.length == 0)
2216 descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2218 error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2219 ifr->ifr_buffer.length - 1);
2221 free(descrbuf, M_IFDESCR);
2226 sx_xlock(&ifdescr_sx);
2227 odescrbuf = ifp->if_description;
2228 ifp->if_description = descrbuf;
2229 sx_xunlock(&ifdescr_sx);
2231 getmicrotime(&ifp->if_lastchange);
2232 free(odescrbuf, M_IFDESCR);
2236 ifr->ifr_fib = ifp->if_fib;
2240 error = priv_check(td, PRIV_NET_SETIFFIB);
2243 if (ifr->ifr_fib >= rt_numfibs)
2246 ifp->if_fib = ifr->ifr_fib;
2250 error = priv_check(td, PRIV_NET_SETIFFLAGS);
2254 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2255 * check, so we don't need special handling here yet.
2257 new_flags = (ifr->ifr_flags & 0xffff) |
2258 (ifr->ifr_flagshigh << 16);
2259 if (ifp->if_flags & IFF_SMART) {
2260 /* Smart drivers twiddle their own routes */
2261 } else if (ifp->if_flags & IFF_UP &&
2262 (new_flags & IFF_UP) == 0) {
2264 } else if (new_flags & IFF_UP &&
2265 (ifp->if_flags & IFF_UP) == 0) {
2268 /* See if permanently promiscuous mode bit is about to flip */
2269 if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2270 if (new_flags & IFF_PPROMISC)
2271 ifp->if_flags |= IFF_PROMISC;
2272 else if (ifp->if_pcount == 0)
2273 ifp->if_flags &= ~IFF_PROMISC;
2274 log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2276 (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
2278 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2279 (new_flags &~ IFF_CANTCHANGE);
2280 if (ifp->if_ioctl) {
2281 (void) (*ifp->if_ioctl)(ifp, cmd, data);
2283 getmicrotime(&ifp->if_lastchange);
2287 error = priv_check(td, PRIV_NET_SETIFCAP);
2290 if (ifp->if_ioctl == NULL)
2291 return (EOPNOTSUPP);
2292 if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2294 error = (*ifp->if_ioctl)(ifp, cmd, data);
2296 getmicrotime(&ifp->if_lastchange);
2301 error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2306 error = priv_check(td, PRIV_NET_SETIFNAME);
2309 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2312 if (new_name[0] == '\0')
2314 if (ifunit(new_name) != NULL)
2318 * XXX: Locking. Nothing else seems to lock if_flags,
2319 * and there are numerous other races with the
2320 * ifunit() checks not being atomic with namespace
2321 * changes (renames, vmoves, if_attach, etc).
2323 ifp->if_flags |= IFF_RENAMING;
2325 /* Announce the departure of the interface. */
2326 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2327 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2329 log(LOG_INFO, "%s: changing name to '%s'\n",
2330 ifp->if_xname, new_name);
2332 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2335 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2336 namelen = strlen(new_name);
2337 onamelen = sdl->sdl_nlen;
2339 * Move the address if needed. This is safe because we
2340 * allocate space for a name of length IFNAMSIZ when we
2341 * create this in if_attach().
2343 if (namelen != onamelen) {
2344 bcopy(sdl->sdl_data + onamelen,
2345 sdl->sdl_data + namelen, sdl->sdl_alen);
2347 bcopy(new_name, sdl->sdl_data, namelen);
2348 sdl->sdl_nlen = namelen;
2349 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2350 bzero(sdl->sdl_data, onamelen);
2351 while (namelen != 0)
2352 sdl->sdl_data[--namelen] = 0xff;
2355 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2356 /* Announce the return of the interface. */
2357 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2359 ifp->if_flags &= ~IFF_RENAMING;
2364 error = priv_check(td, PRIV_NET_SETIFVNET);
2367 error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2372 error = priv_check(td, PRIV_NET_SETIFMETRIC);
2375 ifp->if_metric = ifr->ifr_metric;
2376 getmicrotime(&ifp->if_lastchange);
2380 error = priv_check(td, PRIV_NET_SETIFPHYS);
2383 if (ifp->if_ioctl == NULL)
2384 return (EOPNOTSUPP);
2385 error = (*ifp->if_ioctl)(ifp, cmd, data);
2387 getmicrotime(&ifp->if_lastchange);
2392 u_long oldmtu = ifp->if_mtu;
2394 error = priv_check(td, PRIV_NET_SETIFMTU);
2397 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2399 if (ifp->if_ioctl == NULL)
2400 return (EOPNOTSUPP);
2401 error = (*ifp->if_ioctl)(ifp, cmd, data);
2403 getmicrotime(&ifp->if_lastchange);
2407 * If the link MTU changed, do network layer specific procedure.
2409 if (ifp->if_mtu != oldmtu) {
2419 if (cmd == SIOCADDMULTI)
2420 error = priv_check(td, PRIV_NET_ADDMULTI);
2422 error = priv_check(td, PRIV_NET_DELMULTI);
2426 /* Don't allow group membership on non-multicast interfaces. */
2427 if ((ifp->if_flags & IFF_MULTICAST) == 0)
2428 return (EOPNOTSUPP);
2430 /* Don't let users screw up protocols' entries. */
2431 if (ifr->ifr_addr.sa_family != AF_LINK)
2434 if (cmd == SIOCADDMULTI) {
2435 struct ifmultiaddr *ifma;
2438 * Userland is only permitted to join groups once
2439 * via the if_addmulti() KPI, because it cannot hold
2440 * struct ifmultiaddr * between calls. It may also
2441 * lose a race while we check if the membership
2445 ifma = if_findmulti(ifp, &ifr->ifr_addr);
2446 IF_ADDR_RUNLOCK(ifp);
2450 error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2452 error = if_delmulti(ifp, &ifr->ifr_addr);
2455 getmicrotime(&ifp->if_lastchange);
2458 case SIOCSIFPHYADDR:
2459 case SIOCDIFPHYADDR:
2461 case SIOCSIFPHYADDR_IN6:
2463 case SIOCSLIFPHYADDR:
2465 case SIOCSIFGENERIC:
2466 error = priv_check(td, PRIV_NET_HWIOCTL);
2469 if (ifp->if_ioctl == NULL)
2470 return (EOPNOTSUPP);
2471 error = (*ifp->if_ioctl)(ifp, cmd, data);
2473 getmicrotime(&ifp->if_lastchange);
2477 ifs = (struct ifstat *)data;
2478 ifs->ascii[0] = '\0';
2480 case SIOCGIFPSRCADDR:
2481 case SIOCGIFPDSTADDR:
2482 case SIOCGLIFPHYADDR:
2484 case SIOCGIFGENERIC:
2485 if (ifp->if_ioctl == NULL)
2486 return (EOPNOTSUPP);
2487 error = (*ifp->if_ioctl)(ifp, cmd, data);
2491 error = priv_check(td, PRIV_NET_SETLLADDR);
2494 error = if_setlladdr(ifp,
2495 ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2496 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2501 struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2503 error = priv_check(td, PRIV_NET_ADDIFGROUP);
2506 if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2512 if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2518 struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2520 error = priv_check(td, PRIV_NET_DELIFGROUP);
2523 if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2535 #ifdef COMPAT_FREEBSD32
2543 #define SIOCGIFCONF32 _IOWR('i', 36, struct ifconf32)
2550 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2557 CURVNET_SET(so->so_vnet);
2561 error = ifconf(cmd, data);
2565 #ifdef COMPAT_FREEBSD32
2568 struct ifconf32 *ifc32;
2571 ifc32 = (struct ifconf32 *)data;
2572 ifc.ifc_len = ifc32->ifc_len;
2573 ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2575 error = ifconf(SIOCGIFCONF, (void *)&ifc);
2578 ifc32->ifc_len = ifc.ifc_len;
2583 ifr = (struct ifreq *)data;
2588 error = priv_check(td, PRIV_NET_SETIFVNET);
2590 error = if_vmove_reclaim(td, ifr->ifr_name,
2597 error = priv_check(td, PRIV_NET_IFCREATE);
2599 error = if_clone_create(ifr->ifr_name,
2600 sizeof(ifr->ifr_name),
2601 cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
2605 error = priv_check(td, PRIV_NET_IFDESTROY);
2607 error = if_clone_destroy(ifr->ifr_name);
2611 case SIOCIFGCLONERS:
2612 error = if_clone_list((struct if_clonereq *)data);
2616 error = if_getgroupmembers((struct ifgroupreq *)data);
2619 #if defined(INET) || defined(INET6)
2622 if (carp_ioctl_p == NULL)
2623 error = EPROTONOSUPPORT;
2625 error = (*carp_ioctl_p)(ifr, cmd, td);
2631 ifp = ifunit_ref(ifr->ifr_name);
2637 error = ifhwioctl(cmd, ifp, data, td);
2638 if (error != ENOIOCTL) {
2644 oif_flags = ifp->if_flags;
2645 if (so->so_proto == NULL) {
2648 return (EOPNOTSUPP);
2652 * Pass the request on to the socket control method, and if the
2653 * latter returns EOPNOTSUPP, directly to the interface.
2655 * Make an exception for the legacy SIOCSIF* requests. Drivers
2656 * trust SIOCSIFADDR et al to come from an already privileged
2657 * layer, and do not perform any credentials checks or input
2661 error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
2664 if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL &&
2665 cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR &&
2666 cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK)
2667 error = (*ifp->if_ioctl)(ifp, cmd, data);
2674 case SIOCSIFDSTADDR:
2676 case SIOCSIFBRDADDR:
2677 case SIOCSIFNETMASK:
2678 #if BYTE_ORDER != BIG_ENDIAN
2679 if (ifr->ifr_addr.sa_family == 0 &&
2680 ifr->ifr_addr.sa_len < 16) {
2681 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2682 ifr->ifr_addr.sa_len = 16;
2685 if (ifr->ifr_addr.sa_len == 0)
2686 ifr->ifr_addr.sa_len = 16;
2694 case OSIOCGIFDSTADDR:
2695 cmd = SIOCGIFDSTADDR;
2698 case OSIOCGIFBRDADDR:
2699 cmd = SIOCGIFBRDADDR;
2702 case OSIOCGIFNETMASK:
2703 cmd = SIOCGIFNETMASK;
2705 error = ((*so->so_proto->pr_usrreqs->pru_control)(so,
2709 if (error == EOPNOTSUPP && ifp != NULL &&
2710 ifp->if_ioctl != NULL &&
2711 cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR &&
2712 cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK)
2713 error = (*ifp->if_ioctl)(ifp, cmd, data);
2717 case OSIOCGIFDSTADDR:
2718 case OSIOCGIFBRDADDR:
2719 case OSIOCGIFNETMASK:
2720 *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2724 #endif /* COMPAT_43 */
2726 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2728 if (ifp->if_flags & IFF_UP)
2738 * The code common to handling reference counted flags,
2739 * e.g., in ifpromisc() and if_allmulti().
2740 * The "pflag" argument can specify a permanent mode flag to check,
2741 * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2743 * Only to be used on stack-owned flags, not driver-owned flags.
2746 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2750 int oldflags, oldcount;
2752 /* Sanity checks to catch programming errors */
2753 KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2754 ("%s: setting driver-owned flag %d", __func__, flag));
2757 KASSERT(*refcount >= 0,
2758 ("%s: increment negative refcount %d for flag %d",
2759 __func__, *refcount, flag));
2761 KASSERT(*refcount > 0,
2762 ("%s: decrement non-positive refcount %d for flag %d",
2763 __func__, *refcount, flag));
2765 /* In case this mode is permanent, just touch refcount */
2766 if (ifp->if_flags & pflag) {
2767 *refcount += onswitch ? 1 : -1;
2771 /* Save ifnet parameters for if_ioctl() may fail */
2772 oldcount = *refcount;
2773 oldflags = ifp->if_flags;
2776 * See if we aren't the only and touching refcount is enough.
2777 * Actually toggle interface flag if we are the first or last.
2782 ifp->if_flags |= flag;
2786 ifp->if_flags &= ~flag;
2789 /* Call down the driver since we've changed interface flags */
2790 if (ifp->if_ioctl == NULL) {
2794 ifr.ifr_flags = ifp->if_flags & 0xffff;
2795 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2796 error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2799 /* Notify userland that interface flags have changed */
2804 /* Recover after driver error */
2805 *refcount = oldcount;
2806 ifp->if_flags = oldflags;
2811 * Set/clear promiscuous mode on interface ifp based on the truth value
2812 * of pswitch. The calls are reference counted so that only the first
2813 * "on" request actually has an effect, as does the final "off" request.
2814 * Results are undefined if the "off" and "on" requests are not matched.
2817 ifpromisc(struct ifnet *ifp, int pswitch)
2820 int oldflags = ifp->if_flags;
2822 error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2823 &ifp->if_pcount, pswitch);
2824 /* If promiscuous mode status has changed, log a message */
2825 if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2826 log(LOG_INFO, "%s: promiscuous mode %s\n",
2828 (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2833 * Return interface configuration
2834 * of system. List may be used
2835 * in later ioctl's (above) to get
2836 * other information.
2840 ifconf(u_long cmd, caddr_t data)
2842 struct ifconf *ifc = (struct ifconf *)data;
2847 int error, full = 0, valid_len, max_len;
2849 /* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2850 max_len = MAXPHYS - 1;
2852 /* Prevent hostile input from being able to crash the system */
2853 if (ifc->ifc_len <= 0)
2857 if (ifc->ifc_len <= max_len) {
2858 max_len = ifc->ifc_len;
2861 sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2866 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2870 * Zero the ifr_name buffer to make sure we don't
2871 * disclose the contents of the stack.
2873 memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2875 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2876 >= sizeof(ifr.ifr_name)) {
2879 return (ENAMETOOLONG);
2884 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2885 struct sockaddr *sa = ifa->ifa_addr;
2887 if (prison_if(curthread->td_ucred, sa) != 0)
2891 if (cmd == OSIOCGIFCONF) {
2892 struct osockaddr *osa =
2893 (struct osockaddr *)&ifr.ifr_addr;
2895 osa->sa_family = sa->sa_family;
2896 sbuf_bcat(sb, &ifr, sizeof(ifr));
2897 max_len += sizeof(ifr);
2900 if (sa->sa_len <= sizeof(*sa)) {
2902 sbuf_bcat(sb, &ifr, sizeof(ifr));
2903 max_len += sizeof(ifr);
2906 offsetof(struct ifreq, ifr_addr));
2907 max_len += offsetof(struct ifreq, ifr_addr);
2908 sbuf_bcat(sb, sa, sa->sa_len);
2909 max_len += sa->sa_len;
2912 if (sbuf_error(sb) == 0)
2913 valid_len = sbuf_len(sb);
2915 IF_ADDR_RUNLOCK(ifp);
2917 bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2918 sbuf_bcat(sb, &ifr, sizeof(ifr));
2919 max_len += sizeof(ifr);
2921 if (sbuf_error(sb) == 0)
2922 valid_len = sbuf_len(sb);
2928 * If we didn't allocate enough space (uncommon), try again. If
2929 * we have already allocated as much space as we are allowed,
2930 * return what we've got.
2932 if (valid_len != max_len && !full) {
2937 ifc->ifc_len = valid_len;
2939 error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2945 * Just like ifpromisc(), but for all-multicast-reception mode.
2948 if_allmulti(struct ifnet *ifp, int onswitch)
2951 return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2954 struct ifmultiaddr *
2955 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2957 struct ifmultiaddr *ifma;
2959 IF_ADDR_LOCK_ASSERT(ifp);
2961 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2962 if (sa->sa_family == AF_LINK) {
2963 if (sa_dl_equal(ifma->ifma_addr, sa))
2966 if (sa_equal(ifma->ifma_addr, sa))
2975 * Allocate a new ifmultiaddr and initialize based on passed arguments. We
2976 * make copies of passed sockaddrs. The ifmultiaddr will not be added to
2977 * the ifnet multicast address list here, so the caller must do that and
2978 * other setup work (such as notifying the device driver). The reference
2979 * count is initialized to 1.
2981 static struct ifmultiaddr *
2982 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2985 struct ifmultiaddr *ifma;
2986 struct sockaddr *dupsa;
2988 ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2993 dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2994 if (dupsa == NULL) {
2995 free(ifma, M_IFMADDR);
2998 bcopy(sa, dupsa, sa->sa_len);
2999 ifma->ifma_addr = dupsa;
3001 ifma->ifma_ifp = ifp;
3002 ifma->ifma_refcount = 1;
3003 ifma->ifma_protospec = NULL;
3006 ifma->ifma_lladdr = NULL;
3010 dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
3011 if (dupsa == NULL) {
3012 free(ifma->ifma_addr, M_IFMADDR);
3013 free(ifma, M_IFMADDR);
3016 bcopy(llsa, dupsa, llsa->sa_len);
3017 ifma->ifma_lladdr = dupsa;
3023 * if_freemulti: free ifmultiaddr structure and possibly attached related
3024 * addresses. The caller is responsible for implementing reference
3025 * counting, notifying the driver, handling routing messages, and releasing
3026 * any dependent link layer state.
3029 if_freemulti(struct ifmultiaddr *ifma)
3032 KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
3033 ifma->ifma_refcount));
3034 KASSERT(ifma->ifma_protospec == NULL,
3035 ("if_freemulti: protospec not NULL"));
3037 if (ifma->ifma_lladdr != NULL)
3038 free(ifma->ifma_lladdr, M_IFMADDR);
3039 free(ifma->ifma_addr, M_IFMADDR);
3040 free(ifma, M_IFMADDR);
3044 * Register an additional multicast address with a network interface.
3046 * - If the address is already present, bump the reference count on the
3047 * address and return.
3048 * - If the address is not link-layer, look up a link layer address.
3049 * - Allocate address structures for one or both addresses, and attach to the
3050 * multicast address list on the interface. If automatically adding a link
3051 * layer address, the protocol address will own a reference to the link
3052 * layer address, to be freed when it is freed.
3053 * - Notify the network device driver of an addition to the multicast address
3056 * 'sa' points to caller-owned memory with the desired multicast address.
3058 * 'retifma' will be used to return a pointer to the resulting multicast
3059 * address reference, if desired.
3062 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
3063 struct ifmultiaddr **retifma)
3065 struct ifmultiaddr *ifma, *ll_ifma;
3066 struct sockaddr *llsa;
3070 * If the address is already present, return a new reference to it;
3071 * otherwise, allocate storage and set up a new address.
3074 ifma = if_findmulti(ifp, sa);
3076 ifma->ifma_refcount++;
3077 if (retifma != NULL)
3079 IF_ADDR_WUNLOCK(ifp);
3084 * The address isn't already present; resolve the protocol address
3085 * into a link layer address, and then look that up, bump its
3086 * refcount or allocate an ifma for that also. If 'llsa' was
3087 * returned, we will need to free it later.
3091 if (ifp->if_resolvemulti != NULL) {
3092 error = ifp->if_resolvemulti(ifp, &llsa, sa);
3098 * Allocate the new address. Don't hook it up yet, as we may also
3099 * need to allocate a link layer multicast address.
3101 ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
3108 * If a link layer address is found, we'll need to see if it's
3109 * already present in the address list, or allocate is as well.
3110 * When this block finishes, the link layer address will be on the
3114 ll_ifma = if_findmulti(ifp, llsa);
3115 if (ll_ifma == NULL) {
3116 ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
3117 if (ll_ifma == NULL) {
3118 --ifma->ifma_refcount;
3123 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
3126 ll_ifma->ifma_refcount++;
3127 ifma->ifma_llifma = ll_ifma;
3131 * We now have a new multicast address, ifma, and possibly a new or
3132 * referenced link layer address. Add the primary address to the
3133 * ifnet address list.
3135 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3137 if (retifma != NULL)
3141 * Must generate the message while holding the lock so that 'ifma'
3142 * pointer is still valid.
3144 rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3145 IF_ADDR_WUNLOCK(ifp);
3148 * We are certain we have added something, so call down to the
3149 * interface to let them know about it.
3151 if (ifp->if_ioctl != NULL) {
3152 (void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3156 free(llsa, M_IFMADDR);
3162 free(llsa, M_IFMADDR);
3165 IF_ADDR_WUNLOCK(ifp);
3170 * Delete a multicast group membership by network-layer group address.
3172 * Returns ENOENT if the entry could not be found. If ifp no longer
3173 * exists, results are undefined. This entry point should only be used
3174 * from subsystems which do appropriate locking to hold ifp for the
3175 * duration of the call.
3176 * Network-layer protocol domains must use if_delmulti_ifma().
3179 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3181 struct ifmultiaddr *ifma;
3186 IFNET_RLOCK_NOSLEEP();
3187 TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3192 IFNET_RUNLOCK_NOSLEEP();
3194 KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3201 ifma = if_findmulti(ifp, sa);
3203 lastref = if_delmulti_locked(ifp, ifma, 0);
3204 IF_ADDR_WUNLOCK(ifp);
3209 if (lastref && ifp->if_ioctl != NULL) {
3210 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3217 * Delete all multicast group membership for an interface.
3218 * Should be used to quickly flush all multicast filters.
3221 if_delallmulti(struct ifnet *ifp)
3223 struct ifmultiaddr *ifma;
3224 struct ifmultiaddr *next;
3227 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3228 if_delmulti_locked(ifp, ifma, 0);
3229 IF_ADDR_WUNLOCK(ifp);
3233 * Delete a multicast group membership by group membership pointer.
3234 * Network-layer protocol domains must use this routine.
3236 * It is safe to call this routine if the ifp disappeared.
3239 if_delmulti_ifma(struct ifmultiaddr *ifma)
3244 ifp = ifma->ifma_ifp;
3247 printf("%s: ifma_ifp seems to be detached\n", __func__);
3251 IFNET_RLOCK_NOSLEEP();
3252 TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3256 printf("%s: ifnet %p disappeared\n", __func__, ifp);
3259 IFNET_RUNLOCK_NOSLEEP();
3263 * If and only if the ifnet instance exists: Acquire the address lock.
3268 lastref = if_delmulti_locked(ifp, ifma, 0);
3272 * If and only if the ifnet instance exists:
3273 * Release the address lock.
3274 * If the group was left: update the hardware hash filter.
3276 IF_ADDR_WUNLOCK(ifp);
3277 if (lastref && ifp->if_ioctl != NULL) {
3278 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3284 * Perform deletion of network-layer and/or link-layer multicast address.
3286 * Return 0 if the reference count was decremented.
3287 * Return 1 if the final reference was released, indicating that the
3288 * hardware hash filter should be reprogrammed.
3291 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3293 struct ifmultiaddr *ll_ifma;
3295 if (ifp != NULL && ifma->ifma_ifp != NULL) {
3296 KASSERT(ifma->ifma_ifp == ifp,
3297 ("%s: inconsistent ifp %p", __func__, ifp));
3298 IF_ADDR_WLOCK_ASSERT(ifp);
3301 ifp = ifma->ifma_ifp;
3304 * If the ifnet is detaching, null out references to ifnet,
3305 * so that upper protocol layers will notice, and not attempt
3306 * to obtain locks for an ifnet which no longer exists. The
3307 * routing socket announcement must happen before the ifnet
3308 * instance is detached from the system.
3312 printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3315 * ifp may already be nulled out if we are being reentered
3316 * to delete the ll_ifma.
3319 rt_newmaddrmsg(RTM_DELMADDR, ifma);
3320 ifma->ifma_ifp = NULL;
3324 if (--ifma->ifma_refcount > 0)
3328 * If this ifma is a network-layer ifma, a link-layer ifma may
3329 * have been associated with it. Release it first if so.
3331 ll_ifma = ifma->ifma_llifma;
3332 if (ll_ifma != NULL) {
3333 KASSERT(ifma->ifma_lladdr != NULL,
3334 ("%s: llifma w/o lladdr", __func__));
3336 ll_ifma->ifma_ifp = NULL; /* XXX */
3337 if (--ll_ifma->ifma_refcount == 0) {
3339 TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3342 if_freemulti(ll_ifma);
3347 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3352 * The last reference to this instance of struct ifmultiaddr
3353 * was released; the hardware should be notified of this change.
3359 * Set the link layer address on an interface.
3361 * At this time we only support certain types of interfaces,
3362 * and we don't allow the length of the address to change.
3365 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3367 struct sockaddr_dl *sdl;
3374 IF_ADDR_RUNLOCK(ifp);
3378 IF_ADDR_RUNLOCK(ifp);
3379 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3384 if (len != sdl->sdl_alen) { /* don't allow length to change */
3388 switch (ifp->if_type) {
3396 case IFT_IEEE8023ADLAG:
3398 bcopy(lladdr, LLADDR(sdl), len);
3407 * If the interface is already up, we need
3408 * to re-init it in order to reprogram its
3411 if ((ifp->if_flags & IFF_UP) != 0) {
3412 if (ifp->if_ioctl) {
3413 ifp->if_flags &= ~IFF_UP;
3414 ifr.ifr_flags = ifp->if_flags & 0xffff;
3415 ifr.ifr_flagshigh = ifp->if_flags >> 16;
3416 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3417 ifp->if_flags |= IFF_UP;
3418 ifr.ifr_flags = ifp->if_flags & 0xffff;
3419 ifr.ifr_flagshigh = ifp->if_flags >> 16;
3420 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3424 * Also send gratuitous ARPs to notify other nodes about
3425 * the address change.
3427 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3428 if (ifa->ifa_addr->sa_family == AF_INET)
3429 arp_ifinit(ifp, ifa);
3437 * The name argument must be a pointer to storage which will last as
3438 * long as the interface does. For physical devices, the result of
3439 * device_get_name(dev) is a good choice and for pseudo-devices a
3440 * static string works well.
3443 if_initname(struct ifnet *ifp, const char *name, int unit)
3445 ifp->if_dname = name;
3446 ifp->if_dunit = unit;
3447 if (unit != IF_DUNIT_NONE)
3448 snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3450 strlcpy(ifp->if_xname, name, IFNAMSIZ);
3454 if_printf(struct ifnet *ifp, const char * fmt, ...)
3459 retval = printf("%s: ", ifp->if_xname);
3461 retval += vprintf(fmt, ap);
3467 if_start(struct ifnet *ifp)
3470 (*(ifp)->if_start)(ifp);
3474 * Backwards compatibility interface for drivers
3475 * that have not implemented it
3478 if_transmit(struct ifnet *ifp, struct mbuf *m)
3482 IFQ_HANDOFF(ifp, m, error);
3487 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3492 if (_IF_QFULL(ifq)) {
3499 ifp->if_obytes += m->m_pkthdr.len + adjust;
3500 if (m->m_flags & (M_BCAST|M_MCAST))
3502 active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3504 _IF_ENQUEUE(ifq, m);
3506 if (ifp != NULL && !active)
3507 (*(ifp)->if_start)(ifp);
3512 if_register_com_alloc(u_char type,
3513 if_com_alloc_t *a, if_com_free_t *f)
3516 KASSERT(if_com_alloc[type] == NULL,
3517 ("if_register_com_alloc: %d already registered", type));
3518 KASSERT(if_com_free[type] == NULL,
3519 ("if_register_com_alloc: %d free already registered", type));
3521 if_com_alloc[type] = a;
3522 if_com_free[type] = f;
3526 if_deregister_com_alloc(u_char type)
3529 KASSERT(if_com_alloc[type] != NULL,
3530 ("if_deregister_com_alloc: %d not registered", type));
3531 KASSERT(if_com_free[type] != NULL,
3532 ("if_deregister_com_alloc: %d free not registered", type));
3533 if_com_alloc[type] = NULL;
3534 if_com_free[type] = NULL;