Record number of valid ifaddr containers in ifaddr, so we could use
[dragonfly.git] / sys / net / if.c
1 /*
2  * Copyright (c) 1980, 1986, 1993
3  *      The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
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  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *      This product includes software developed by the University of
16  *      California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *      @(#)if.c        8.3 (Berkeley) 1/4/94
34  * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
35  * $DragonFly: src/sys/net/if.c,v 1.82 2008/11/11 13:48:01 sephe Exp $
36  */
37
38 #include "opt_compat.h"
39 #include "opt_inet6.h"
40 #include "opt_inet.h"
41 #include "opt_polling.h"
42
43 #include <sys/param.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/systm.h>
47 #include <sys/proc.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/socketops.h>
52 #include <sys/protosw.h>
53 #include <sys/kernel.h>
54 #include <sys/ktr.h>
55 #include <sys/sockio.h>
56 #include <sys/syslog.h>
57 #include <sys/sysctl.h>
58 #include <sys/domain.h>
59 #include <sys/thread.h>
60 #include <sys/thread2.h>
61 #include <sys/serialize.h>
62 #include <sys/msgport2.h>
63 #include <sys/bus.h>
64
65 #include <net/if.h>
66 #include <net/if_arp.h>
67 #include <net/if_dl.h>
68 #include <net/if_types.h>
69 #include <net/if_var.h>
70 #include <net/ifq_var.h>
71 #include <net/radix.h>
72 #include <net/route.h>
73 #include <net/if_clone.h>
74 #include <net/netisr.h>
75 #include <net/netmsg2.h>
76
77 #include <machine/atomic.h>
78 #include <machine/stdarg.h>
79 #include <machine/smp.h>
80
81 #if defined(INET) || defined(INET6)
82 /*XXX*/
83 #include <netinet/in.h>
84 #include <netinet/in_var.h>
85 #include <netinet/if_ether.h>
86 #ifdef INET6
87 #include <netinet6/in6_var.h>
88 #include <netinet6/in6_ifattach.h>
89 #endif
90 #endif
91
92 #if defined(COMPAT_43)
93 #include <emulation/43bsd/43bsd_socket.h>
94 #endif /* COMPAT_43 */
95
96 struct netmsg_ifaddr {
97         struct netmsg   netmsg;
98         struct ifaddr   *ifa;
99         struct ifnet    *ifp;
100         int             tail;
101 };
102
103 /*
104  * System initialization
105  */
106 static void     if_attachdomain(void *);
107 static void     if_attachdomain1(struct ifnet *);
108 static int      ifconf(u_long, caddr_t, struct ucred *);
109 static void     ifinit(void *);
110 static void     ifnetinit(void *);
111 static void     if_slowtimo(void *);
112 static void     link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
113 static int      if_rtdel(struct radix_node *, void *);
114
115 #ifdef INET6
116 /*
117  * XXX: declare here to avoid to include many inet6 related files..
118  * should be more generalized?
119  */
120 extern void     nd6_setmtu(struct ifnet *);
121 #endif
122
123 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
124 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
125
126 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
127 /* Must be after netisr_init */
128 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
129
130 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
131 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
132
133 int                     ifqmaxlen = IFQ_MAXLEN;
134 struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
135
136 /* In ifq_dispatch(), try to do direct ifnet.if_start first */
137 static int              ifq_dispatch_schedonly = 0;
138 SYSCTL_INT(_net_link_generic, OID_AUTO, ifq_dispatch_schedonly, CTLFLAG_RW,
139            &ifq_dispatch_schedonly, 0, "");
140
141 /* In ifq_dispatch(), schedule ifnet.if_start without checking ifnet.if_snd */
142 static int              ifq_dispatch_schednochk = 0;
143 SYSCTL_INT(_net_link_generic, OID_AUTO, ifq_dispatch_schednochk, CTLFLAG_RW,
144            &ifq_dispatch_schednochk, 0, "");
145
146 /* In if_devstart(), try to do direct ifnet.if_start first */
147 static int              if_devstart_schedonly = 0;
148 SYSCTL_INT(_net_link_generic, OID_AUTO, if_devstart_schedonly, CTLFLAG_RW,
149            &if_devstart_schedonly, 0, "");
150
151 /* In if_devstart(), schedule ifnet.if_start without checking ifnet.if_snd */
152 static int              if_devstart_schednochk = 0;
153 SYSCTL_INT(_net_link_generic, OID_AUTO, if_devstart_schednochk, CTLFLAG_RW,
154            &if_devstart_schednochk, 0, "");
155
156 #ifdef SMP
157 /* Schedule ifnet.if_start on the current CPU */
158 static int              if_start_oncpu_sched = 0;
159 SYSCTL_INT(_net_link_generic, OID_AUTO, if_start_oncpu_sched, CTLFLAG_RW,
160            &if_start_oncpu_sched, 0, "");
161 #endif
162
163 struct callout          if_slowtimo_timer;
164
165 int                     if_index = 0;
166 struct ifnet            **ifindex2ifnet = NULL;
167 static struct thread    ifnet_threads[MAXCPU];
168 static int              ifnet_mpsafe_thread = NETMSG_SERVICE_MPSAFE;
169
170 #define IFQ_KTR_STRING          "ifq=%p"
171 #define IFQ_KTR_ARG_SIZE        (sizeof(void *))
172 #ifndef KTR_IFQ
173 #define KTR_IFQ                 KTR_ALL
174 #endif
175 KTR_INFO_MASTER(ifq);
176 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARG_SIZE);
177 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARG_SIZE);
178 #define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)
179
180 #define IF_START_KTR_STRING     "ifp=%p"
181 #define IF_START_KTR_ARG_SIZE   (sizeof(void *))
182 #ifndef KTR_IF_START
183 #define KTR_IF_START            KTR_ALL
184 #endif
185 KTR_INFO_MASTER(if_start);
186 KTR_INFO(KTR_IF_START, if_start, run, 0,
187          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
188 KTR_INFO(KTR_IF_START, if_start, sched, 1,
189          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
190 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
191          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
192 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
193          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
194 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
195          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
196 #define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
197
198 /*
199  * Network interface utility routines.
200  *
201  * Routines with ifa_ifwith* names take sockaddr *'s as
202  * parameters.
203  */
204 /* ARGSUSED*/
205 void
206 ifinit(void *dummy)
207 {
208         struct ifnet *ifp;
209
210         callout_init(&if_slowtimo_timer);
211
212         crit_enter();
213         TAILQ_FOREACH(ifp, &ifnet, if_link) {
214                 if (ifp->if_snd.ifq_maxlen == 0) {
215                         if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
216                         ifp->if_snd.ifq_maxlen = ifqmaxlen;
217                 }
218         }
219         crit_exit();
220
221         if_slowtimo(0);
222 }
223
224 static int
225 if_start_cpuid(struct ifnet *ifp)
226 {
227         return ifp->if_cpuid;
228 }
229
230 #ifdef DEVICE_POLLING
231 static int
232 if_start_cpuid_poll(struct ifnet *ifp)
233 {
234         int poll_cpuid = ifp->if_poll_cpuid;
235
236         if (poll_cpuid >= 0)
237                 return poll_cpuid;
238         else
239                 return ifp->if_cpuid;
240 }
241 #endif
242
243 static void
244 if_start_ipifunc(void *arg)
245 {
246         struct ifnet *ifp = arg;
247         struct lwkt_msg *lmsg = &ifp->if_start_nmsg[mycpuid].nm_lmsg;
248
249         crit_enter();
250         if (lmsg->ms_flags & MSGF_DONE)
251                 lwkt_sendmsg(ifnet_portfn(mycpuid), lmsg);
252         crit_exit();
253 }
254
255 /*
256  * Schedule ifnet.if_start on ifnet's CPU
257  */
258 static void
259 if_start_schedule(struct ifnet *ifp)
260 {
261 #ifdef SMP
262         int cpu;
263
264         if (if_start_oncpu_sched)
265                 cpu = mycpuid;
266         else
267                 cpu = ifp->if_start_cpuid(ifp);
268
269         if (cpu != mycpuid)
270                 lwkt_send_ipiq(globaldata_find(cpu), if_start_ipifunc, ifp);
271         else
272 #endif
273         if_start_ipifunc(ifp);
274 }
275
276 /*
277  * NOTE:
278  * This function will release ifnet.if_start interlock,
279  * if ifnet.if_start does not need to be scheduled
280  */
281 static __inline int
282 if_start_need_schedule(struct ifaltq *ifq, int running)
283 {
284         if (!running || ifq_is_empty(ifq)
285 #ifdef ALTQ
286             || ifq->altq_tbr != NULL
287 #endif
288         ) {
289                 ALTQ_LOCK(ifq);
290                 /*
291                  * ifnet.if_start interlock is released, if:
292                  * 1) Hardware can not take any packets, due to
293                  *    o  interface is marked down
294                  *    o  hardware queue is full (IFF_OACTIVE)
295                  *    Under the second situation, hardware interrupt
296                  *    or polling(4) will call/schedule ifnet.if_start
297                  *    when hardware queue is ready
298                  * 2) There is not packet in the ifnet.if_snd.
299                  *    Further ifq_dispatch or ifq_handoff will call/
300                  *    schedule ifnet.if_start
301                  * 3) TBR is used and it does not allow further
302                  *    dequeueing.
303                  *    TBR callout will call ifnet.if_start
304                  */
305                 if (!running || !ifq_data_ready(ifq)) {
306                         ifq->altq_started = 0;
307                         ALTQ_UNLOCK(ifq);
308                         return 0;
309                 }
310                 ALTQ_UNLOCK(ifq);
311         }
312         return 1;
313 }
314
315 static void
316 if_start_dispatch(struct netmsg *nmsg)
317 {
318         struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
319         struct ifnet *ifp = lmsg->u.ms_resultp;
320         struct ifaltq *ifq = &ifp->if_snd;
321         int running = 0;
322
323         crit_enter();
324         lwkt_replymsg(lmsg, 0); /* reply ASAP */
325         crit_exit();
326
327 #ifdef SMP
328         if (!if_start_oncpu_sched && mycpuid != ifp->if_start_cpuid(ifp)) {
329                 /*
330                  * If the ifnet is still up, we need to
331                  * chase its CPU change.
332                  */
333                 if (ifp->if_flags & IFF_UP) {
334                         logifstart(chase_sched, ifp);
335                         if_start_schedule(ifp);
336                         return;
337                 } else {
338                         goto check;
339                 }
340         }
341 #endif
342
343         if (ifp->if_flags & IFF_UP) {
344                 lwkt_serialize_enter(ifp->if_serializer); /* XXX try? */
345                 if ((ifp->if_flags & IFF_OACTIVE) == 0) {
346                         logifstart(run, ifp);
347                         ifp->if_start(ifp);
348                         if ((ifp->if_flags &
349                         (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
350                                 running = 1;
351                 }
352                 lwkt_serialize_exit(ifp->if_serializer);
353         }
354 #ifdef SMP
355 check:
356 #endif
357         if (if_start_need_schedule(ifq, running)) {
358                 crit_enter();
359                 if (lmsg->ms_flags & MSGF_DONE) { /* XXX necessary? */
360                         logifstart(sched, ifp);
361                         lwkt_sendmsg(ifnet_portfn(mycpuid), lmsg);
362                 }
363                 crit_exit();
364         }
365 }
366
367 /* Device driver ifnet.if_start helper function */
368 void
369 if_devstart(struct ifnet *ifp)
370 {
371         struct ifaltq *ifq = &ifp->if_snd;
372         int running = 0;
373
374         ASSERT_SERIALIZED(ifp->if_serializer);
375
376         ALTQ_LOCK(ifq);
377         if (ifq->altq_started || !ifq_data_ready(ifq)) {
378                 logifstart(avoid, ifp);
379                 ALTQ_UNLOCK(ifq);
380                 return;
381         }
382         ifq->altq_started = 1;
383         ALTQ_UNLOCK(ifq);
384
385         if (if_devstart_schedonly) {
386                 /*
387                  * Always schedule ifnet.if_start on ifnet's CPU,
388                  * short circuit the rest of this function.
389                  */
390                 logifstart(sched, ifp);
391                 if_start_schedule(ifp);
392                 return;
393         }
394
395         logifstart(run, ifp);
396         ifp->if_start(ifp);
397
398         if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
399                 running = 1;
400
401         if (if_devstart_schednochk || if_start_need_schedule(ifq, running)) {
402                 /*
403                  * More data need to be transmitted, ifnet.if_start is
404                  * scheduled on ifnet's CPU, and we keep going.
405                  * NOTE: ifnet.if_start interlock is not released.
406                  */
407                 logifstart(sched, ifp);
408                 if_start_schedule(ifp);
409         }
410 }
411
412 /*
413  * Attach an interface to the list of "active" interfaces.
414  *
415  * The serializer is optional.  If non-NULL access to the interface
416  * may be MPSAFE.
417  */
418 void
419 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
420 {
421         unsigned socksize, ifasize;
422         int namelen, masklen;
423         struct sockaddr_dl *sdl;
424         struct ifaddr *ifa;
425         struct ifaltq *ifq;
426         int i;
427
428         static int if_indexlim = 8;
429
430         /*
431          * The serializer can be passed in from the device, allowing the
432          * same serializer to be used for both the interrupt interlock and
433          * the device queue.  If not specified, the netif structure will
434          * use an embedded serializer.
435          */
436         if (serializer == NULL) {
437                 serializer = &ifp->if_default_serializer;
438                 lwkt_serialize_init(serializer);
439         }
440         ifp->if_serializer = serializer;
441
442         ifp->if_start_cpuid = if_start_cpuid;
443         ifp->if_cpuid = 0;
444
445 #ifdef DEVICE_POLLING
446         /* Device is not in polling mode by default */
447         ifp->if_poll_cpuid = -1;
448         if (ifp->if_poll != NULL)
449                 ifp->if_start_cpuid = if_start_cpuid_poll;
450 #endif
451
452         ifp->if_start_nmsg = kmalloc(ncpus * sizeof(struct netmsg),
453                                      M_IFADDR /* XXX */, M_WAITOK);
454         for (i = 0; i < ncpus; ++i) {
455                 netmsg_init(&ifp->if_start_nmsg[i], &netisr_adone_rport, 0,
456                             if_start_dispatch);
457                 ifp->if_start_nmsg[i].nm_lmsg.u.ms_resultp = ifp;
458         }
459
460         TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
461         ifp->if_index = ++if_index;
462
463         /*
464          * XXX -
465          * The old code would work if the interface passed a pre-existing
466          * chain of ifaddrs to this code.  We don't trust our callers to
467          * properly initialize the tailq, however, so we no longer allow
468          * this unlikely case.
469          */
470         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
471                                     M_IFADDR, M_WAITOK | M_ZERO);
472         for (i = 0; i < ncpus; ++i)
473                 TAILQ_INIT(&ifp->if_addrheads[i]);
474
475         TAILQ_INIT(&ifp->if_prefixhead);
476         LIST_INIT(&ifp->if_multiaddrs);
477         getmicrotime(&ifp->if_lastchange);
478         if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
479                 unsigned int n;
480                 struct ifnet **q;
481
482                 if_indexlim <<= 1;
483
484                 /* grow ifindex2ifnet */
485                 n = if_indexlim * sizeof(*q);
486                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
487                 if (ifindex2ifnet) {
488                         bcopy(ifindex2ifnet, q, n/2);
489                         kfree(ifindex2ifnet, M_IFADDR);
490                 }
491                 ifindex2ifnet = q;
492         }
493
494         ifindex2ifnet[if_index] = ifp;
495
496         /*
497          * create a Link Level name for this device
498          */
499         namelen = strlen(ifp->if_xname);
500 #define _offsetof(t, m) ((int)((caddr_t)&((t *)0)->m))
501         masklen = _offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
502         socksize = masklen + ifp->if_addrlen;
503 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
504         if (socksize < sizeof(*sdl))
505                 socksize = sizeof(*sdl);
506         socksize = ROUNDUP(socksize);
507         ifasize = sizeof(struct ifaddr) + 2 * socksize;
508         ifa = ifa_create(ifasize, M_WAITOK);
509         sdl = (struct sockaddr_dl *)(ifa + 1);
510         sdl->sdl_len = socksize;
511         sdl->sdl_family = AF_LINK;
512         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
513         sdl->sdl_nlen = namelen;
514         sdl->sdl_index = ifp->if_index;
515         sdl->sdl_type = ifp->if_type;
516         ifp->if_lladdr = ifa;
517         ifa->ifa_ifp = ifp;
518         ifa->ifa_rtrequest = link_rtrequest;
519         ifa->ifa_addr = (struct sockaddr *)sdl;
520         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
521         ifa->ifa_netmask = (struct sockaddr *)sdl;
522         sdl->sdl_len = masklen;
523         while (namelen != 0)
524                 sdl->sdl_data[--namelen] = 0xff;
525         ifa_iflink(ifa, ifp, 0 /* Insert head */);
526
527         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
528         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
529
530         ifq = &ifp->if_snd;
531         ifq->altq_type = 0;
532         ifq->altq_disc = NULL;
533         ifq->altq_flags &= ALTQF_CANTCHANGE;
534         ifq->altq_tbr = NULL;
535         ifq->altq_ifp = ifp;
536         ifq->altq_started = 0;
537         ifq->altq_prepended = NULL;
538         ALTQ_LOCK_INIT(ifq);
539         ifq_set_classic(ifq);
540
541         if (!SLIST_EMPTY(&domains))
542                 if_attachdomain1(ifp);
543
544         /* Announce the interface. */
545         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
546 }
547
548 static void
549 if_attachdomain(void *dummy)
550 {
551         struct ifnet *ifp;
552
553         crit_enter();
554         TAILQ_FOREACH(ifp, &ifnet, if_list)
555                 if_attachdomain1(ifp);
556         crit_exit();
557 }
558 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
559         if_attachdomain, NULL);
560
561 static void
562 if_attachdomain1(struct ifnet *ifp)
563 {
564         struct domain *dp;
565
566         crit_enter();
567
568         /* address family dependent data region */
569         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
570         SLIST_FOREACH(dp, &domains, dom_next)
571                 if (dp->dom_ifattach)
572                         ifp->if_afdata[dp->dom_family] =
573                                 (*dp->dom_ifattach)(ifp);
574         crit_exit();
575 }
576
577 /*
578  * Purge all addresses whose type is _not_ AF_LINK
579  */
580 void
581 if_purgeaddrs_nolink(struct ifnet *ifp)
582 {
583         struct ifaddr_container *ifac, *next;
584
585         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
586                               ifa_link, next) {
587                 struct ifaddr *ifa = ifac->ifa;
588
589                 /* Leave link ifaddr as it is */
590                 if (ifa->ifa_addr->sa_family == AF_LINK)
591                         continue;
592 #ifdef INET
593                 /* XXX: Ugly!! ad hoc just for INET */
594                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
595                         struct ifaliasreq ifr;
596 #ifdef IFADDR_DEBUG_VERBOSE
597                         int i;
598
599                         kprintf("purge in4 addr %p: ", ifa);
600                         for (i = 0; i < ncpus; ++i)
601                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
602                         kprintf("\n");
603 #endif
604
605                         bzero(&ifr, sizeof ifr);
606                         ifr.ifra_addr = *ifa->ifa_addr;
607                         if (ifa->ifa_dstaddr)
608                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
609                         if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
610                                        NULL) == 0)
611                                 continue;
612                 }
613 #endif /* INET */
614 #ifdef INET6
615                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
616 #ifdef IFADDR_DEBUG_VERBOSE
617                         int i;
618
619                         kprintf("purge in6 addr %p: ", ifa);
620                         for (i = 0; i < ncpus; ++i)
621                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
622                         kprintf("\n");
623 #endif
624
625                         in6_purgeaddr(ifa);
626                         /* ifp_addrhead is already updated */
627                         continue;
628                 }
629 #endif /* INET6 */
630                 ifa_ifunlink(ifa, ifp);
631                 ifa_destroy(ifa);
632         }
633 }
634
635 /*
636  * Detach an interface, removing it from the
637  * list of "active" interfaces.
638  */
639 void
640 if_detach(struct ifnet *ifp)
641 {
642         struct radix_node_head  *rnh;
643         int i;
644         int cpu, origcpu;
645         struct domain *dp;
646
647         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
648
649         /*
650          * Remove routes and flush queues.
651          */
652         crit_enter();
653 #ifdef DEVICE_POLLING
654         if (ifp->if_flags & IFF_POLLING)
655                 ether_poll_deregister(ifp);
656 #endif
657         if_down(ifp);
658
659         if (ifq_is_enabled(&ifp->if_snd))
660                 altq_disable(&ifp->if_snd);
661         if (ifq_is_attached(&ifp->if_snd))
662                 altq_detach(&ifp->if_snd);
663
664         /*
665          * Clean up all addresses.
666          */
667         ifp->if_lladdr = NULL;
668
669         if_purgeaddrs_nolink(ifp);
670         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
671                 struct ifaddr *ifa;
672
673                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
674                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
675                         ("non-link ifaddr is left on if_addrheads"));
676
677                 ifa_ifunlink(ifa, ifp);
678                 ifa_destroy(ifa);
679                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
680                         ("there are still ifaddrs left on if_addrheads"));
681         }
682
683 #ifdef INET
684         /*
685          * Remove all IPv4 kernel structures related to ifp.
686          */
687         in_ifdetach(ifp);
688 #endif
689
690 #ifdef INET6
691         /*
692          * Remove all IPv6 kernel structs related to ifp.  This should be done
693          * before removing routing entries below, since IPv6 interface direct
694          * routes are expected to be removed by the IPv6-specific kernel API.
695          * Otherwise, the kernel will detect some inconsistency and bark it.
696          */
697         in6_ifdetach(ifp);
698 #endif
699
700         /*
701          * Delete all remaining routes using this interface
702          * Unfortuneatly the only way to do this is to slog through
703          * the entire routing table looking for routes which point
704          * to this interface...oh well...
705          */
706         origcpu = mycpuid;
707         for (cpu = 0; cpu < ncpus2; cpu++) {
708                 lwkt_migratecpu(cpu);
709                 for (i = 1; i <= AF_MAX; i++) {
710                         if ((rnh = rt_tables[cpu][i]) == NULL)
711                                 continue;
712                         rnh->rnh_walktree(rnh, if_rtdel, ifp);
713                 }
714         }
715         lwkt_migratecpu(origcpu);
716
717         /* Announce that the interface is gone. */
718         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
719         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
720
721         SLIST_FOREACH(dp, &domains, dom_next)
722                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
723                         (*dp->dom_ifdetach)(ifp,
724                                 ifp->if_afdata[dp->dom_family]);
725
726         /*
727          * Remove interface from ifindex2ifp[] and maybe decrement if_index.
728          */
729         ifindex2ifnet[ifp->if_index] = NULL;
730         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
731                 if_index--;
732
733         TAILQ_REMOVE(&ifnet, ifp, if_link);
734         kfree(ifp->if_addrheads, M_IFADDR);
735         kfree(ifp->if_start_nmsg, M_IFADDR);
736         crit_exit();
737 }
738
739 /*
740  * Delete Routes for a Network Interface
741  *
742  * Called for each routing entry via the rnh->rnh_walktree() call above
743  * to delete all route entries referencing a detaching network interface.
744  *
745  * Arguments:
746  *      rn      pointer to node in the routing table
747  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
748  *
749  * Returns:
750  *      0       successful
751  *      errno   failed - reason indicated
752  *
753  */
754 static int
755 if_rtdel(struct radix_node *rn, void *arg)
756 {
757         struct rtentry  *rt = (struct rtentry *)rn;
758         struct ifnet    *ifp = arg;
759         int             err;
760
761         if (rt->rt_ifp == ifp) {
762
763                 /*
764                  * Protect (sorta) against walktree recursion problems
765                  * with cloned routes
766                  */
767                 if (!(rt->rt_flags & RTF_UP))
768                         return (0);
769
770                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
771                                 rt_mask(rt), rt->rt_flags,
772                                 (struct rtentry **) NULL);
773                 if (err) {
774                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
775                 }
776         }
777
778         return (0);
779 }
780
781 /*
782  * Locate an interface based on a complete address.
783  */
784 struct ifaddr *
785 ifa_ifwithaddr(struct sockaddr *addr)
786 {
787         struct ifnet *ifp;
788
789         TAILQ_FOREACH(ifp, &ifnet, if_link) {
790                 struct ifaddr_container *ifac;
791
792                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
793                         struct ifaddr *ifa = ifac->ifa;
794
795                         if (ifa->ifa_addr->sa_family != addr->sa_family)
796                                 continue;
797                         if (sa_equal(addr, ifa->ifa_addr))
798                                 return (ifa);
799                         if ((ifp->if_flags & IFF_BROADCAST) &&
800                             ifa->ifa_broadaddr &&
801                             /* IPv6 doesn't have broadcast */
802                             ifa->ifa_broadaddr->sa_len != 0 &&
803                             sa_equal(ifa->ifa_broadaddr, addr))
804                                 return (ifa);
805                 }
806         }
807         return (NULL);
808 }
809 /*
810  * Locate the point to point interface with a given destination address.
811  */
812 struct ifaddr *
813 ifa_ifwithdstaddr(struct sockaddr *addr)
814 {
815         struct ifnet *ifp;
816
817         TAILQ_FOREACH(ifp, &ifnet, if_link) {
818                 struct ifaddr_container *ifac;
819
820                 if (!(ifp->if_flags & IFF_POINTOPOINT))
821                         continue;
822
823                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
824                         struct ifaddr *ifa = ifac->ifa;
825
826                         if (ifa->ifa_addr->sa_family != addr->sa_family)
827                                 continue;
828                         if (ifa->ifa_dstaddr &&
829                             sa_equal(addr, ifa->ifa_dstaddr))
830                                 return (ifa);
831                 }
832         }
833         return (NULL);
834 }
835
836 /*
837  * Find an interface on a specific network.  If many, choice
838  * is most specific found.
839  */
840 struct ifaddr *
841 ifa_ifwithnet(struct sockaddr *addr)
842 {
843         struct ifnet *ifp;
844         struct ifaddr *ifa_maybe = NULL;
845         u_int af = addr->sa_family;
846         char *addr_data = addr->sa_data, *cplim;
847
848         /*
849          * AF_LINK addresses can be looked up directly by their index number,
850          * so do that if we can.
851          */
852         if (af == AF_LINK) {
853                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
854
855                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
856                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
857         }
858
859         /*
860          * Scan though each interface, looking for ones that have
861          * addresses in this address family.
862          */
863         TAILQ_FOREACH(ifp, &ifnet, if_link) {
864                 struct ifaddr_container *ifac;
865
866                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
867                         struct ifaddr *ifa = ifac->ifa;
868                         char *cp, *cp2, *cp3;
869
870                         if (ifa->ifa_addr->sa_family != af)
871 next:                           continue;
872                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
873                                 /*
874                                  * This is a bit broken as it doesn't
875                                  * take into account that the remote end may
876                                  * be a single node in the network we are
877                                  * looking for.
878                                  * The trouble is that we don't know the
879                                  * netmask for the remote end.
880                                  */
881                                 if (ifa->ifa_dstaddr != NULL &&
882                                     sa_equal(addr, ifa->ifa_dstaddr))
883                                         return (ifa);
884                         } else {
885                                 /*
886                                  * if we have a special address handler,
887                                  * then use it instead of the generic one.
888                                  */
889                                 if (ifa->ifa_claim_addr) {
890                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
891                                                 return (ifa);
892                                         } else {
893                                                 continue;
894                                         }
895                                 }
896
897                                 /*
898                                  * Scan all the bits in the ifa's address.
899                                  * If a bit dissagrees with what we are
900                                  * looking for, mask it with the netmask
901                                  * to see if it really matters.
902                                  * (A byte at a time)
903                                  */
904                                 if (ifa->ifa_netmask == 0)
905                                         continue;
906                                 cp = addr_data;
907                                 cp2 = ifa->ifa_addr->sa_data;
908                                 cp3 = ifa->ifa_netmask->sa_data;
909                                 cplim = ifa->ifa_netmask->sa_len +
910                                         (char *)ifa->ifa_netmask;
911                                 while (cp3 < cplim)
912                                         if ((*cp++ ^ *cp2++) & *cp3++)
913                                                 goto next; /* next address! */
914                                 /*
915                                  * If the netmask of what we just found
916                                  * is more specific than what we had before
917                                  * (if we had one) then remember the new one
918                                  * before continuing to search
919                                  * for an even better one.
920                                  */
921                                 if (ifa_maybe == 0 ||
922                                     rn_refines((char *)ifa->ifa_netmask,
923                                                (char *)ifa_maybe->ifa_netmask))
924                                         ifa_maybe = ifa;
925                         }
926                 }
927         }
928         return (ifa_maybe);
929 }
930
931 /*
932  * Find an interface address specific to an interface best matching
933  * a given address.
934  */
935 struct ifaddr *
936 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
937 {
938         struct ifaddr_container *ifac;
939         char *cp, *cp2, *cp3;
940         char *cplim;
941         struct ifaddr *ifa_maybe = 0;
942         u_int af = addr->sa_family;
943
944         if (af >= AF_MAX)
945                 return (0);
946         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
947                 struct ifaddr *ifa = ifac->ifa;
948
949                 if (ifa->ifa_addr->sa_family != af)
950                         continue;
951                 if (ifa_maybe == 0)
952                         ifa_maybe = ifa;
953                 if (ifa->ifa_netmask == NULL) {
954                         if (sa_equal(addr, ifa->ifa_addr) ||
955                             (ifa->ifa_dstaddr != NULL &&
956                              sa_equal(addr, ifa->ifa_dstaddr)))
957                                 return (ifa);
958                         continue;
959                 }
960                 if (ifp->if_flags & IFF_POINTOPOINT) {
961                         if (sa_equal(addr, ifa->ifa_dstaddr))
962                                 return (ifa);
963                 } else {
964                         cp = addr->sa_data;
965                         cp2 = ifa->ifa_addr->sa_data;
966                         cp3 = ifa->ifa_netmask->sa_data;
967                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
968                         for (; cp3 < cplim; cp3++)
969                                 if ((*cp++ ^ *cp2++) & *cp3)
970                                         break;
971                         if (cp3 == cplim)
972                                 return (ifa);
973                 }
974         }
975         return (ifa_maybe);
976 }
977
978 /*
979  * Default action when installing a route with a Link Level gateway.
980  * Lookup an appropriate real ifa to point to.
981  * This should be moved to /sys/net/link.c eventually.
982  */
983 static void
984 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
985 {
986         struct ifaddr *ifa;
987         struct sockaddr *dst;
988         struct ifnet *ifp;
989
990         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
991             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
992                 return;
993         ifa = ifaof_ifpforaddr(dst, ifp);
994         if (ifa != NULL) {
995                 IFAFREE(rt->rt_ifa);
996                 IFAREF(ifa);
997                 rt->rt_ifa = ifa;
998                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
999                         ifa->ifa_rtrequest(cmd, rt, info);
1000         }
1001 }
1002
1003 /*
1004  * Mark an interface down and notify protocols of
1005  * the transition.
1006  * NOTE: must be called at splnet or eqivalent.
1007  */
1008 void
1009 if_unroute(struct ifnet *ifp, int flag, int fam)
1010 {
1011         struct ifaddr_container *ifac;
1012
1013         ifp->if_flags &= ~flag;
1014         getmicrotime(&ifp->if_lastchange);
1015         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1016                 struct ifaddr *ifa = ifac->ifa;
1017
1018                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1019                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1020         }
1021         ifq_purge(&ifp->if_snd);
1022         rt_ifmsg(ifp);
1023 }
1024
1025 /*
1026  * Mark an interface up and notify protocols of
1027  * the transition.
1028  * NOTE: must be called at splnet or eqivalent.
1029  */
1030 void
1031 if_route(struct ifnet *ifp, int flag, int fam)
1032 {
1033         struct ifaddr_container *ifac;
1034
1035         ifq_purge(&ifp->if_snd);
1036         ifp->if_flags |= flag;
1037         getmicrotime(&ifp->if_lastchange);
1038         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1039                 struct ifaddr *ifa = ifac->ifa;
1040
1041                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1042                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1043         }
1044         rt_ifmsg(ifp);
1045 #ifdef INET6
1046         in6_if_up(ifp);
1047 #endif
1048 }
1049
1050 /*
1051  * Mark an interface down and notify protocols of the transition.  An
1052  * interface going down is also considered to be a synchronizing event.
1053  * We must ensure that all packet processing related to the interface
1054  * has completed before we return so e.g. the caller can free the ifnet
1055  * structure that the mbufs may be referencing.
1056  *
1057  * NOTE: must be called at splnet or eqivalent.
1058  */
1059 void
1060 if_down(struct ifnet *ifp)
1061 {
1062         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1063         netmsg_service_sync();
1064 }
1065
1066 /*
1067  * Mark an interface up and notify protocols of
1068  * the transition.
1069  * NOTE: must be called at splnet or eqivalent.
1070  */
1071 void
1072 if_up(struct ifnet *ifp)
1073 {
1074         if_route(ifp, IFF_UP, AF_UNSPEC);
1075 }
1076
1077 /*
1078  * Process a link state change.
1079  * NOTE: must be called at splsoftnet or equivalent.
1080  */
1081 void
1082 if_link_state_change(struct ifnet *ifp)
1083 {
1084         int link_state = ifp->if_link_state;
1085
1086         rt_ifmsg(ifp);
1087         devctl_notify("IFNET", ifp->if_xname,
1088             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1089 }
1090
1091 /*
1092  * Handle interface watchdog timer routines.  Called
1093  * from softclock, we decrement timers (if set) and
1094  * call the appropriate interface routine on expiration.
1095  */
1096 static void
1097 if_slowtimo(void *arg)
1098 {
1099         struct ifnet *ifp;
1100
1101         crit_enter();
1102
1103         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1104                 if (ifp->if_timer == 0 || --ifp->if_timer)
1105                         continue;
1106                 if (ifp->if_watchdog) {
1107                         if (lwkt_serialize_try(ifp->if_serializer)) {
1108                                 (*ifp->if_watchdog)(ifp);
1109                                 lwkt_serialize_exit(ifp->if_serializer);
1110                         } else {
1111                                 /* try again next timeout */
1112                                 ++ifp->if_timer;
1113                         }
1114                 }
1115         }
1116
1117         crit_exit();
1118
1119         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1120 }
1121
1122 /*
1123  * Map interface name to
1124  * interface structure pointer.
1125  */
1126 struct ifnet *
1127 ifunit(const char *name)
1128 {
1129         struct ifnet *ifp;
1130
1131         /*
1132          * Search all the interfaces for this name/number
1133          */
1134
1135         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1136                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1137                         break;
1138         }
1139         return (ifp);
1140 }
1141
1142
1143 /*
1144  * Map interface name in a sockaddr_dl to
1145  * interface structure pointer.
1146  */
1147 struct ifnet *
1148 if_withname(struct sockaddr *sa)
1149 {
1150         char ifname[IFNAMSIZ+1];
1151         struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1152
1153         if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1154              (sdl->sdl_nlen > IFNAMSIZ) )
1155                 return NULL;
1156
1157         /*
1158          * ifunit wants a null-terminated name.  It may not be null-terminated
1159          * in the sockaddr.  We don't want to change the caller's sockaddr,
1160          * and there might not be room to put the trailing null anyway, so we
1161          * make a local copy that we know we can null terminate safely.
1162          */
1163
1164         bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1165         ifname[sdl->sdl_nlen] = '\0';
1166         return ifunit(ifname);
1167 }
1168
1169
1170 /*
1171  * Interface ioctls.
1172  */
1173 int
1174 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1175 {
1176         struct ifnet *ifp;
1177         struct ifreq *ifr;
1178         struct ifstat *ifs;
1179         int error;
1180         short oif_flags;
1181         int new_flags;
1182         size_t namelen, onamelen;
1183         char new_name[IFNAMSIZ];
1184         struct ifaddr *ifa;
1185         struct sockaddr_dl *sdl;
1186
1187         switch (cmd) {
1188
1189         case SIOCGIFCONF:
1190         case OSIOCGIFCONF:
1191                 return (ifconf(cmd, data, cred));
1192         }
1193         ifr = (struct ifreq *)data;
1194
1195         switch (cmd) {
1196         case SIOCIFCREATE:
1197         case SIOCIFDESTROY:
1198                 if ((error = suser_cred(cred, 0)) != 0)
1199                         return (error);
1200                 return ((cmd == SIOCIFCREATE) ?
1201                         if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name)) :
1202                         if_clone_destroy(ifr->ifr_name));
1203
1204         case SIOCIFGCLONERS:
1205                 return (if_clone_list((struct if_clonereq *)data));
1206         }
1207
1208         ifp = ifunit(ifr->ifr_name);
1209         if (ifp == 0)
1210                 return (ENXIO);
1211         switch (cmd) {
1212
1213         case SIOCGIFFLAGS:
1214                 ifr->ifr_flags = ifp->if_flags;
1215                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1216                 break;
1217
1218         case SIOCGIFCAP:
1219                 ifr->ifr_reqcap = ifp->if_capabilities;
1220                 ifr->ifr_curcap = ifp->if_capenable;
1221                 break;
1222
1223         case SIOCGIFMETRIC:
1224                 ifr->ifr_metric = ifp->if_metric;
1225                 break;
1226
1227         case SIOCGIFMTU:
1228                 ifr->ifr_mtu = ifp->if_mtu;
1229                 break;
1230
1231         case SIOCGIFPHYS:
1232                 ifr->ifr_phys = ifp->if_physical;
1233                 break;
1234
1235         case SIOCGIFPOLLCPU:
1236 #ifdef DEVICE_POLLING
1237                 ifr->ifr_pollcpu = ifp->if_poll_cpuid;
1238 #else
1239                 ifr->ifr_pollcpu = -1;
1240 #endif
1241                 break;
1242
1243         case SIOCSIFPOLLCPU:
1244 #ifdef DEVICE_POLLING
1245                 if ((ifp->if_flags & IFF_POLLING) == 0)
1246                         ether_pollcpu_register(ifp, ifr->ifr_pollcpu);
1247 #endif
1248                 break;
1249
1250         case SIOCSIFFLAGS:
1251                 error = suser_cred(cred, 0);
1252                 if (error)
1253                         return (error);
1254                 new_flags = (ifr->ifr_flags & 0xffff) |
1255                     (ifr->ifr_flagshigh << 16);
1256                 if (ifp->if_flags & IFF_SMART) {
1257                         /* Smart drivers twiddle their own routes */
1258                 } else if (ifp->if_flags & IFF_UP &&
1259                     (new_flags & IFF_UP) == 0) {
1260                         crit_enter();
1261                         if_down(ifp);
1262                         crit_exit();
1263                 } else if (new_flags & IFF_UP &&
1264                     (ifp->if_flags & IFF_UP) == 0) {
1265                         crit_enter();
1266                         if_up(ifp);
1267                         crit_exit();
1268                 }
1269
1270 #ifdef DEVICE_POLLING
1271                 if ((new_flags ^ ifp->if_flags) & IFF_POLLING) {
1272                         if (new_flags & IFF_POLLING) {
1273                                 ether_poll_register(ifp);
1274                         } else {
1275                                 ether_poll_deregister(ifp);
1276                         }
1277                 }
1278 #endif
1279
1280                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1281                         (new_flags &~ IFF_CANTCHANGE);
1282                 if (new_flags & IFF_PPROMISC) {
1283                         /* Permanently promiscuous mode requested */
1284                         ifp->if_flags |= IFF_PROMISC;
1285                 } else if (ifp->if_pcount == 0) {
1286                         ifp->if_flags &= ~IFF_PROMISC;
1287                 }
1288                 if (ifp->if_ioctl) {
1289                         lwkt_serialize_enter(ifp->if_serializer);
1290                         ifp->if_ioctl(ifp, cmd, data, cred);
1291                         lwkt_serialize_exit(ifp->if_serializer);
1292                 }
1293                 getmicrotime(&ifp->if_lastchange);
1294                 break;
1295
1296         case SIOCSIFCAP:
1297                 error = suser_cred(cred, 0);
1298                 if (error)
1299                         return (error);
1300                 if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1301                         return (EINVAL);
1302                 lwkt_serialize_enter(ifp->if_serializer);
1303                 ifp->if_ioctl(ifp, cmd, data, cred);
1304                 lwkt_serialize_exit(ifp->if_serializer);
1305                 break;
1306
1307         case SIOCSIFNAME:
1308                 error = suser_cred(cred, 0);
1309                 if (error != 0)
1310                         return (error);
1311                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1312                 if (error != 0)
1313                         return (error);
1314                 if (new_name[0] == '\0')
1315                         return (EINVAL);
1316                 if (ifunit(new_name) != NULL)
1317                         return (EEXIST);
1318
1319                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1320
1321                 /* Announce the departure of the interface. */
1322                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1323
1324                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1325                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1326                 /* XXX IFA_LOCK(ifa); */
1327                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1328                 namelen = strlen(new_name);
1329                 onamelen = sdl->sdl_nlen;
1330                 /*
1331                  * Move the address if needed.  This is safe because we
1332                  * allocate space for a name of length IFNAMSIZ when we
1333                  * create this in if_attach().
1334                  */
1335                 if (namelen != onamelen) {
1336                         bcopy(sdl->sdl_data + onamelen,
1337                             sdl->sdl_data + namelen, sdl->sdl_alen);
1338                 }
1339                 bcopy(new_name, sdl->sdl_data, namelen);
1340                 sdl->sdl_nlen = namelen;
1341                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1342                 bzero(sdl->sdl_data, onamelen);
1343                 while (namelen != 0)
1344                         sdl->sdl_data[--namelen] = 0xff;
1345                 /* XXX IFA_UNLOCK(ifa) */
1346
1347                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1348
1349                 /* Announce the return of the interface. */
1350                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1351                 break;
1352
1353         case SIOCSIFMETRIC:
1354                 error = suser_cred(cred, 0);
1355                 if (error)
1356                         return (error);
1357                 ifp->if_metric = ifr->ifr_metric;
1358                 getmicrotime(&ifp->if_lastchange);
1359                 break;
1360
1361         case SIOCSIFPHYS:
1362                 error = suser_cred(cred, 0);
1363                 if (error)
1364                         return error;
1365                 if (!ifp->if_ioctl)
1366                         return EOPNOTSUPP;
1367                 lwkt_serialize_enter(ifp->if_serializer);
1368                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1369                 lwkt_serialize_exit(ifp->if_serializer);
1370                 if (error == 0)
1371                         getmicrotime(&ifp->if_lastchange);
1372                 return (error);
1373
1374         case SIOCSIFMTU:
1375         {
1376                 u_long oldmtu = ifp->if_mtu;
1377
1378                 error = suser_cred(cred, 0);
1379                 if (error)
1380                         return (error);
1381                 if (ifp->if_ioctl == NULL)
1382                         return (EOPNOTSUPP);
1383                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1384                         return (EINVAL);
1385                 lwkt_serialize_enter(ifp->if_serializer);
1386                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1387                 lwkt_serialize_exit(ifp->if_serializer);
1388                 if (error == 0) {
1389                         getmicrotime(&ifp->if_lastchange);
1390                         rt_ifmsg(ifp);
1391                 }
1392                 /*
1393                  * If the link MTU changed, do network layer specific procedure.
1394                  */
1395                 if (ifp->if_mtu != oldmtu) {
1396 #ifdef INET6
1397                         nd6_setmtu(ifp);
1398 #endif
1399                 }
1400                 return (error);
1401         }
1402
1403         case SIOCADDMULTI:
1404         case SIOCDELMULTI:
1405                 error = suser_cred(cred, 0);
1406                 if (error)
1407                         return (error);
1408
1409                 /* Don't allow group membership on non-multicast interfaces. */
1410                 if ((ifp->if_flags & IFF_MULTICAST) == 0)
1411                         return EOPNOTSUPP;
1412
1413                 /* Don't let users screw up protocols' entries. */
1414                 if (ifr->ifr_addr.sa_family != AF_LINK)
1415                         return EINVAL;
1416
1417                 if (cmd == SIOCADDMULTI) {
1418                         struct ifmultiaddr *ifma;
1419                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1420                 } else {
1421                         error = if_delmulti(ifp, &ifr->ifr_addr);
1422                 }
1423                 if (error == 0)
1424                         getmicrotime(&ifp->if_lastchange);
1425                 return error;
1426
1427         case SIOCSIFPHYADDR:
1428         case SIOCDIFPHYADDR:
1429 #ifdef INET6
1430         case SIOCSIFPHYADDR_IN6:
1431 #endif
1432         case SIOCSLIFPHYADDR:
1433         case SIOCSIFMEDIA:
1434         case SIOCSIFGENERIC:
1435                 error = suser_cred(cred, 0);
1436                 if (error)
1437                         return (error);
1438                 if (ifp->if_ioctl == 0)
1439                         return (EOPNOTSUPP);
1440                 lwkt_serialize_enter(ifp->if_serializer);
1441                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1442                 lwkt_serialize_exit(ifp->if_serializer);
1443                 if (error == 0)
1444                         getmicrotime(&ifp->if_lastchange);
1445                 return error;
1446
1447         case SIOCGIFSTATUS:
1448                 ifs = (struct ifstat *)data;
1449                 ifs->ascii[0] = '\0';
1450
1451         case SIOCGIFPSRCADDR:
1452         case SIOCGIFPDSTADDR:
1453         case SIOCGLIFPHYADDR:
1454         case SIOCGIFMEDIA:
1455         case SIOCGIFGENERIC:
1456                 if (ifp->if_ioctl == NULL)
1457                         return (EOPNOTSUPP);
1458                 lwkt_serialize_enter(ifp->if_serializer);
1459                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1460                 lwkt_serialize_exit(ifp->if_serializer);
1461                 return (error);
1462
1463         case SIOCSIFLLADDR:
1464                 error = suser_cred(cred, 0);
1465                 if (error)
1466                         return (error);
1467                 return if_setlladdr(ifp,
1468                     ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1469
1470         default:
1471                 oif_flags = ifp->if_flags;
1472                 if (so->so_proto == 0)
1473                         return (EOPNOTSUPP);
1474 #ifndef COMPAT_43
1475                 error = so_pru_control(so, cmd, data, ifp);
1476 #else
1477             {
1478                 int ocmd = cmd;
1479
1480                 switch (cmd) {
1481
1482                 case SIOCSIFDSTADDR:
1483                 case SIOCSIFADDR:
1484                 case SIOCSIFBRDADDR:
1485                 case SIOCSIFNETMASK:
1486 #if BYTE_ORDER != BIG_ENDIAN
1487                         if (ifr->ifr_addr.sa_family == 0 &&
1488                             ifr->ifr_addr.sa_len < 16) {
1489                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1490                                 ifr->ifr_addr.sa_len = 16;
1491                         }
1492 #else
1493                         if (ifr->ifr_addr.sa_len == 0)
1494                                 ifr->ifr_addr.sa_len = 16;
1495 #endif
1496                         break;
1497
1498                 case OSIOCGIFADDR:
1499                         cmd = SIOCGIFADDR;
1500                         break;
1501
1502                 case OSIOCGIFDSTADDR:
1503                         cmd = SIOCGIFDSTADDR;
1504                         break;
1505
1506                 case OSIOCGIFBRDADDR:
1507                         cmd = SIOCGIFBRDADDR;
1508                         break;
1509
1510                 case OSIOCGIFNETMASK:
1511                         cmd = SIOCGIFNETMASK;
1512                 }
1513                 error =  so_pru_control(so, cmd, data, ifp);
1514                 switch (ocmd) {
1515
1516                 case OSIOCGIFADDR:
1517                 case OSIOCGIFDSTADDR:
1518                 case OSIOCGIFBRDADDR:
1519                 case OSIOCGIFNETMASK:
1520                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1521
1522                 }
1523             }
1524 #endif /* COMPAT_43 */
1525
1526                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1527 #ifdef INET6
1528                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
1529                         if (ifp->if_flags & IFF_UP) {
1530                                 crit_enter();
1531                                 in6_if_up(ifp);
1532                                 crit_exit();
1533                         }
1534 #endif
1535                 }
1536                 return (error);
1537
1538         }
1539         return (0);
1540 }
1541
1542 /*
1543  * Set/clear promiscuous mode on interface ifp based on the truth value
1544  * of pswitch.  The calls are reference counted so that only the first
1545  * "on" request actually has an effect, as does the final "off" request.
1546  * Results are undefined if the "off" and "on" requests are not matched.
1547  */
1548 int
1549 ifpromisc(struct ifnet *ifp, int pswitch)
1550 {
1551         struct ifreq ifr;
1552         int error;
1553         int oldflags;
1554
1555         oldflags = ifp->if_flags;
1556         if (ifp->if_flags & IFF_PPROMISC) {
1557                 /* Do nothing if device is in permanently promiscuous mode */
1558                 ifp->if_pcount += pswitch ? 1 : -1;
1559                 return (0);
1560         }
1561         if (pswitch) {
1562                 /*
1563                  * If the device is not configured up, we cannot put it in
1564                  * promiscuous mode.
1565                  */
1566                 if ((ifp->if_flags & IFF_UP) == 0)
1567                         return (ENETDOWN);
1568                 if (ifp->if_pcount++ != 0)
1569                         return (0);
1570                 ifp->if_flags |= IFF_PROMISC;
1571                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1572                     ifp->if_xname);
1573         } else {
1574                 if (--ifp->if_pcount > 0)
1575                         return (0);
1576                 ifp->if_flags &= ~IFF_PROMISC;
1577                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
1578                     ifp->if_xname);
1579         }
1580         ifr.ifr_flags = ifp->if_flags;
1581         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1582         lwkt_serialize_enter(ifp->if_serializer);
1583         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1584                                  (struct ucred *)NULL);
1585         lwkt_serialize_exit(ifp->if_serializer);
1586         if (error == 0)
1587                 rt_ifmsg(ifp);
1588         else
1589                 ifp->if_flags = oldflags;
1590         return error;
1591 }
1592
1593 /*
1594  * Return interface configuration
1595  * of system.  List may be used
1596  * in later ioctl's (above) to get
1597  * other information.
1598  */
1599 static int
1600 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
1601 {
1602         struct ifconf *ifc = (struct ifconf *)data;
1603         struct ifnet *ifp;
1604         struct sockaddr *sa;
1605         struct ifreq ifr, *ifrp;
1606         int space = ifc->ifc_len, error = 0;
1607
1608         ifrp = ifc->ifc_req;
1609         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1610                 struct ifaddr_container *ifac;
1611                 int addrs;
1612
1613                 if (space <= sizeof ifr)
1614                         break;
1615
1616                 /*
1617                  * Zero the stack declared structure first to prevent
1618                  * memory disclosure.
1619                  */
1620                 bzero(&ifr, sizeof(ifr));
1621                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
1622                     >= sizeof(ifr.ifr_name)) {
1623                         error = ENAMETOOLONG;
1624                         break;
1625                 }
1626
1627                 addrs = 0;
1628                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1629                         struct ifaddr *ifa = ifac->ifa;
1630
1631                         if (space <= sizeof ifr)
1632                                 break;
1633                         sa = ifa->ifa_addr;
1634                         if (cred->cr_prison &&
1635                             prison_if(cred, sa))
1636                                 continue;
1637                         addrs++;
1638 #ifdef COMPAT_43
1639                         if (cmd == OSIOCGIFCONF) {
1640                                 struct osockaddr *osa =
1641                                          (struct osockaddr *)&ifr.ifr_addr;
1642                                 ifr.ifr_addr = *sa;
1643                                 osa->sa_family = sa->sa_family;
1644                                 error = copyout(&ifr, ifrp, sizeof ifr);
1645                                 ifrp++;
1646                         } else
1647 #endif
1648                         if (sa->sa_len <= sizeof(*sa)) {
1649                                 ifr.ifr_addr = *sa;
1650                                 error = copyout(&ifr, ifrp, sizeof ifr);
1651                                 ifrp++;
1652                         } else {
1653                                 if (space < (sizeof ifr) + sa->sa_len -
1654                                             sizeof(*sa))
1655                                         break;
1656                                 space -= sa->sa_len - sizeof(*sa);
1657                                 error = copyout(&ifr, ifrp,
1658                                                 sizeof ifr.ifr_name);
1659                                 if (error == 0)
1660                                         error = copyout(sa, &ifrp->ifr_addr,
1661                                                         sa->sa_len);
1662                                 ifrp = (struct ifreq *)
1663                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
1664                         }
1665                         if (error)
1666                                 break;
1667                         space -= sizeof ifr;
1668                 }
1669                 if (error)
1670                         break;
1671                 if (!addrs) {
1672                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
1673                         error = copyout(&ifr, ifrp, sizeof ifr);
1674                         if (error)
1675                                 break;
1676                         space -= sizeof ifr;
1677                         ifrp++;
1678                 }
1679         }
1680         ifc->ifc_len -= space;
1681         return (error);
1682 }
1683
1684 /*
1685  * Just like if_promisc(), but for all-multicast-reception mode.
1686  */
1687 int
1688 if_allmulti(struct ifnet *ifp, int onswitch)
1689 {
1690         int error = 0;
1691         struct ifreq ifr;
1692
1693         crit_enter();
1694
1695         if (onswitch) {
1696                 if (ifp->if_amcount++ == 0) {
1697                         ifp->if_flags |= IFF_ALLMULTI;
1698                         ifr.ifr_flags = ifp->if_flags;
1699                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1700                         lwkt_serialize_enter(ifp->if_serializer);
1701                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1702                                               (struct ucred *)NULL);
1703                         lwkt_serialize_exit(ifp->if_serializer);
1704                 }
1705         } else {
1706                 if (ifp->if_amcount > 1) {
1707                         ifp->if_amcount--;
1708                 } else {
1709                         ifp->if_amcount = 0;
1710                         ifp->if_flags &= ~IFF_ALLMULTI;
1711                         ifr.ifr_flags = ifp->if_flags;
1712                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1713                         lwkt_serialize_enter(ifp->if_serializer);
1714                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1715                                               (struct ucred *)NULL);
1716                         lwkt_serialize_exit(ifp->if_serializer);
1717                 }
1718         }
1719
1720         crit_exit();
1721
1722         if (error == 0)
1723                 rt_ifmsg(ifp);
1724         return error;
1725 }
1726
1727 /*
1728  * Add a multicast listenership to the interface in question.
1729  * The link layer provides a routine which converts
1730  */
1731 int
1732 if_addmulti(
1733         struct ifnet *ifp,      /* interface to manipulate */
1734         struct sockaddr *sa,    /* address to add */
1735         struct ifmultiaddr **retifma)
1736 {
1737         struct sockaddr *llsa, *dupsa;
1738         int error;
1739         struct ifmultiaddr *ifma;
1740
1741         /*
1742          * If the matching multicast address already exists
1743          * then don't add a new one, just add a reference
1744          */
1745         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1746                 if (sa_equal(sa, ifma->ifma_addr)) {
1747                         ifma->ifma_refcount++;
1748                         if (retifma)
1749                                 *retifma = ifma;
1750                         return 0;
1751                 }
1752         }
1753
1754         /*
1755          * Give the link layer a chance to accept/reject it, and also
1756          * find out which AF_LINK address this maps to, if it isn't one
1757          * already.
1758          */
1759         if (ifp->if_resolvemulti) {
1760                 lwkt_serialize_enter(ifp->if_serializer);
1761                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
1762                 lwkt_serialize_exit(ifp->if_serializer);
1763                 if (error) 
1764                         return error;
1765         } else {
1766                 llsa = 0;
1767         }
1768
1769         MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, M_WAITOK);
1770         MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, M_WAITOK);
1771         bcopy(sa, dupsa, sa->sa_len);
1772
1773         ifma->ifma_addr = dupsa;
1774         ifma->ifma_lladdr = llsa;
1775         ifma->ifma_ifp = ifp;
1776         ifma->ifma_refcount = 1;
1777         ifma->ifma_protospec = 0;
1778         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
1779
1780         /*
1781          * Some network interfaces can scan the address list at
1782          * interrupt time; lock them out.
1783          */
1784         crit_enter();
1785         LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1786         crit_exit();
1787         *retifma = ifma;
1788
1789         if (llsa != 0) {
1790                 LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1791                         if (sa_equal(ifma->ifma_addr, llsa))
1792                                 break;
1793                 }
1794                 if (ifma) {
1795                         ifma->ifma_refcount++;
1796                 } else {
1797                         MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma,
1798                                M_IFMADDR, M_WAITOK);
1799                         MALLOC(dupsa, struct sockaddr *, llsa->sa_len,
1800                                M_IFMADDR, M_WAITOK);
1801                         bcopy(llsa, dupsa, llsa->sa_len);
1802                         ifma->ifma_addr = dupsa;
1803                         ifma->ifma_ifp = ifp;
1804                         ifma->ifma_refcount = 1;
1805                         crit_enter();
1806                         LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1807                         crit_exit();
1808                 }
1809         }
1810         /*
1811          * We are certain we have added something, so call down to the
1812          * interface to let them know about it.
1813          */
1814         crit_enter();
1815         lwkt_serialize_enter(ifp->if_serializer);
1816         ifp->if_ioctl(ifp, SIOCADDMULTI, 0, (struct ucred *)NULL);
1817         lwkt_serialize_exit(ifp->if_serializer);
1818         crit_exit();
1819
1820         return 0;
1821 }
1822
1823 /*
1824  * Remove a reference to a multicast address on this interface.  Yell
1825  * if the request does not match an existing membership.
1826  */
1827 int
1828 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
1829 {
1830         struct ifmultiaddr *ifma;
1831
1832         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1833                 if (sa_equal(sa, ifma->ifma_addr))
1834                         break;
1835         if (ifma == 0)
1836                 return ENOENT;
1837
1838         if (ifma->ifma_refcount > 1) {
1839                 ifma->ifma_refcount--;
1840                 return 0;
1841         }
1842
1843         rt_newmaddrmsg(RTM_DELMADDR, ifma);
1844         sa = ifma->ifma_lladdr;
1845         crit_enter();
1846         LIST_REMOVE(ifma, ifma_link);
1847         /*
1848          * Make sure the interface driver is notified
1849          * in the case of a link layer mcast group being left.
1850          */
1851         if (ifma->ifma_addr->sa_family == AF_LINK && sa == 0) {
1852                 lwkt_serialize_enter(ifp->if_serializer);
1853                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, (struct ucred *)NULL);
1854                 lwkt_serialize_exit(ifp->if_serializer);
1855         }
1856         crit_exit();
1857         kfree(ifma->ifma_addr, M_IFMADDR);
1858         kfree(ifma, M_IFMADDR);
1859         if (sa == 0)
1860                 return 0;
1861
1862         /*
1863          * Now look for the link-layer address which corresponds to
1864          * this network address.  It had been squirreled away in
1865          * ifma->ifma_lladdr for this purpose (so we don't have
1866          * to call ifp->if_resolvemulti() again), and we saved that
1867          * value in sa above.  If some nasty deleted the
1868          * link-layer address out from underneath us, we can deal because
1869          * the address we stored was is not the same as the one which was
1870          * in the record for the link-layer address.  (So we don't complain
1871          * in that case.)
1872          */
1873         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1874                 if (sa_equal(sa, ifma->ifma_addr))
1875                         break;
1876         if (ifma == 0)
1877                 return 0;
1878
1879         if (ifma->ifma_refcount > 1) {
1880                 ifma->ifma_refcount--;
1881                 return 0;
1882         }
1883
1884         crit_enter();
1885         lwkt_serialize_enter(ifp->if_serializer);
1886         LIST_REMOVE(ifma, ifma_link);
1887         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, (struct ucred *)NULL);
1888         lwkt_serialize_exit(ifp->if_serializer);
1889         crit_exit();
1890         kfree(ifma->ifma_addr, M_IFMADDR);
1891         kfree(sa, M_IFMADDR);
1892         kfree(ifma, M_IFMADDR);
1893
1894         return 0;
1895 }
1896
1897 /*
1898  * Set the link layer address on an interface.
1899  *
1900  * At this time we only support certain types of interfaces,
1901  * and we don't allow the length of the address to change.
1902  */
1903 int
1904 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
1905 {
1906         struct sockaddr_dl *sdl;
1907         struct ifreq ifr;
1908
1909         sdl = IF_LLSOCKADDR(ifp);
1910         if (sdl == NULL)
1911                 return (EINVAL);
1912         if (len != sdl->sdl_alen)       /* don't allow length to change */
1913                 return (EINVAL);
1914         switch (ifp->if_type) {
1915         case IFT_ETHER:                 /* these types use struct arpcom */
1916         case IFT_XETHER:
1917         case IFT_L2VLAN:
1918                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
1919                 bcopy(lladdr, LLADDR(sdl), len);
1920                 break;
1921         default:
1922                 return (ENODEV);
1923         }
1924         /*
1925          * If the interface is already up, we need
1926          * to re-init it in order to reprogram its
1927          * address filter.
1928          */
1929         lwkt_serialize_enter(ifp->if_serializer);
1930         if ((ifp->if_flags & IFF_UP) != 0) {
1931                 struct ifaddr_container *ifac;
1932
1933                 ifp->if_flags &= ~IFF_UP;
1934                 ifr.ifr_flags = ifp->if_flags;
1935                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
1936                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1937                               (struct ucred *)NULL);
1938                 ifp->if_flags |= IFF_UP;
1939                 ifr.ifr_flags = ifp->if_flags;
1940                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
1941                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1942                                  (struct ucred *)NULL);
1943 #ifdef INET
1944                 /*
1945                  * Also send gratuitous ARPs to notify other nodes about
1946                  * the address change.
1947                  */
1948                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1949                         struct ifaddr *ifa = ifac->ifa;
1950
1951                         if (ifa->ifa_addr != NULL &&
1952                             ifa->ifa_addr->sa_family == AF_INET)
1953                                 arp_ifinit(ifp, ifa);
1954                 }
1955 #endif
1956         }
1957         lwkt_serialize_exit(ifp->if_serializer);
1958         return (0);
1959 }
1960
1961 struct ifmultiaddr *
1962 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
1963 {
1964         struct ifmultiaddr *ifma;
1965
1966         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1967                 if (sa_equal(ifma->ifma_addr, sa))
1968                         break;
1969
1970         return ifma;
1971 }
1972
1973 /*
1974  * This function locates the first real ethernet MAC from a network
1975  * card and loads it into node, returning 0 on success or ENOENT if
1976  * no suitable interfaces were found.  It is used by the uuid code to
1977  * generate a unique 6-byte number.
1978  */
1979 int
1980 if_getanyethermac(uint16_t *node, int minlen)
1981 {
1982         struct ifnet *ifp;
1983         struct sockaddr_dl *sdl;
1984
1985         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1986                 if (ifp->if_type != IFT_ETHER)
1987                         continue;
1988                 sdl = IF_LLSOCKADDR(ifp);
1989                 if (sdl->sdl_alen < minlen)
1990                         continue;
1991                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
1992                       minlen);
1993                 return(0);
1994         }
1995         return (ENOENT);
1996 }
1997
1998 /*
1999  * The name argument must be a pointer to storage which will last as
2000  * long as the interface does.  For physical devices, the result of
2001  * device_get_name(dev) is a good choice and for pseudo-devices a
2002  * static string works well.
2003  */
2004 void
2005 if_initname(struct ifnet *ifp, const char *name, int unit)
2006 {
2007         ifp->if_dname = name;
2008         ifp->if_dunit = unit;
2009         if (unit != IF_DUNIT_NONE)
2010                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2011         else
2012                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2013 }
2014
2015 int
2016 if_printf(struct ifnet *ifp, const char *fmt, ...)
2017 {
2018         __va_list ap;
2019         int retval;
2020
2021         retval = kprintf("%s: ", ifp->if_xname);
2022         __va_start(ap, fmt);
2023         retval += kvprintf(fmt, ap);
2024         __va_end(ap);
2025         return (retval);
2026 }
2027
2028 void
2029 ifq_set_classic(struct ifaltq *ifq)
2030 {
2031         ifq->altq_enqueue = ifq_classic_enqueue;
2032         ifq->altq_dequeue = ifq_classic_dequeue;
2033         ifq->altq_request = ifq_classic_request;
2034 }
2035
2036 int
2037 ifq_classic_enqueue(struct ifaltq *ifq, struct mbuf *m,
2038                     struct altq_pktattr *pa __unused)
2039 {
2040         logifq(enqueue, ifq);
2041         if (IF_QFULL(ifq)) {
2042                 m_freem(m);
2043                 return(ENOBUFS);
2044         } else {
2045                 IF_ENQUEUE(ifq, m);
2046                 return(0);
2047         }       
2048 }
2049
2050 struct mbuf *
2051 ifq_classic_dequeue(struct ifaltq *ifq, struct mbuf *mpolled, int op)
2052 {
2053         struct mbuf *m;
2054
2055         switch (op) {
2056         case ALTDQ_POLL:
2057                 IF_POLL(ifq, m);
2058                 break;
2059         case ALTDQ_REMOVE:
2060                 logifq(dequeue, ifq);
2061                 IF_DEQUEUE(ifq, m);
2062                 break;
2063         default:
2064                 panic("unsupported ALTQ dequeue op: %d", op);
2065         }
2066         KKASSERT(mpolled == NULL || mpolled == m);
2067         return(m);
2068 }
2069
2070 int
2071 ifq_classic_request(struct ifaltq *ifq, int req, void *arg)
2072 {
2073         switch (req) {
2074         case ALTRQ_PURGE:
2075                 IF_DRAIN(ifq);
2076                 break;
2077         default:
2078                 panic("unsupported ALTQ request: %d", req);
2079         }
2080         return(0);
2081 }
2082
2083 int
2084 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2085 {
2086         struct ifaltq *ifq = &ifp->if_snd;
2087         int running = 0, error, start = 0;
2088
2089         ASSERT_NOT_SERIALIZED(ifp->if_serializer);
2090
2091         ALTQ_LOCK(ifq);
2092         error = ifq_enqueue_locked(ifq, m, pa);
2093         if (error) {
2094                 ALTQ_UNLOCK(ifq);
2095                 return error;
2096         }
2097         if (!ifq->altq_started) {
2098                 /*
2099                  * Hold the interlock of ifnet.if_start
2100                  */
2101                 ifq->altq_started = 1;
2102                 start = 1;
2103         }
2104         ALTQ_UNLOCK(ifq);
2105
2106         ifp->if_obytes += m->m_pkthdr.len;
2107         if (m->m_flags & M_MCAST)
2108                 ifp->if_omcasts++;
2109
2110         if (!start) {
2111                 logifstart(avoid, ifp);
2112                 return 0;
2113         }
2114
2115         if (ifq_dispatch_schedonly) {
2116                 /*
2117                  * Always schedule ifnet.if_start on ifnet's CPU,
2118                  * short circuit the rest of this function.
2119                  */
2120                 logifstart(sched, ifp);
2121                 if_start_schedule(ifp);
2122                 return 0;
2123         }
2124
2125         /*
2126          * Try to do direct ifnet.if_start first, if there is
2127          * contention on ifnet's serializer, ifnet.if_start will
2128          * be scheduled on ifnet's CPU.
2129          */
2130         if (!lwkt_serialize_try(ifp->if_serializer)) {
2131                 /*
2132                  * ifnet serializer contention happened,
2133                  * ifnet.if_start is scheduled on ifnet's
2134                  * CPU, and we keep going.
2135                  */
2136                 logifstart(contend_sched, ifp);
2137                 if_start_schedule(ifp);
2138                 return 0;
2139         }
2140
2141         if ((ifp->if_flags & IFF_OACTIVE) == 0) {
2142                 logifstart(run, ifp);
2143                 ifp->if_start(ifp);
2144                 if ((ifp->if_flags &
2145                      (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
2146                         running = 1;
2147         }
2148
2149         lwkt_serialize_exit(ifp->if_serializer);
2150
2151         if (ifq_dispatch_schednochk || if_start_need_schedule(ifq, running)) {
2152                 /*
2153                  * More data need to be transmitted, ifnet.if_start is
2154                  * scheduled on ifnet's CPU, and we keep going.
2155                  * NOTE: ifnet.if_start interlock is not released.
2156                  */
2157                 logifstart(sched, ifp);
2158                 if_start_schedule(ifp);
2159         }
2160         return 0;
2161 }
2162
2163 void *
2164 ifa_create(int size, int flags)
2165 {
2166         struct ifaddr *ifa;
2167         int i;
2168
2169         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small\n"));
2170
2171         ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2172         if (ifa == NULL)
2173                 return NULL;
2174
2175         ifa->ifa_containers = kmalloc(ncpus * sizeof(struct ifaddr_container),
2176                                       M_IFADDR, M_WAITOK | M_ZERO);
2177         ifa->ifa_ncnt = ncpus;
2178         for (i = 0; i < ncpus; ++i) {
2179                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2180
2181                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2182                 ifac->ifa = ifa;
2183                 ifac->ifa_refcnt = 1;
2184         }
2185 #ifdef IFADDR_DEBUG
2186         kprintf("alloc ifa %p %d\n", ifa, size);
2187 #endif
2188         return ifa;
2189 }
2190
2191 void
2192 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2193 {
2194         struct ifaddr *ifa = ifac->ifa;
2195
2196         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2197         KKASSERT(ifac->ifa_refcnt == 0);
2198         KASSERT(ifac->ifa_listmask == 0,
2199                 ("ifa is still on %#x lists\n", ifac->ifa_listmask));
2200
2201         ifac->ifa_magic = IFA_CONTAINER_DEAD;
2202
2203 #ifdef IFADDR_DEBUG_VERBOSE
2204         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2205 #endif
2206
2207         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2208                 ("invalid # of ifac, %d\n", ifa->ifa_ncnt));
2209         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2210 #ifdef IFADDR_DEBUG
2211                 kprintf("free ifa %p\n", ifa);
2212 #endif
2213                 kfree(ifa->ifa_containers, M_IFADDR);
2214                 kfree(ifa, M_IFADDR);
2215         }
2216 }
2217
2218 static void
2219 ifa_iflink_dispatch(struct netmsg *nmsg)
2220 {
2221         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2222         struct ifaddr *ifa = msg->ifa;
2223         struct ifnet *ifp = msg->ifp;
2224         int cpu = mycpuid;
2225         struct ifaddr_container *ifac;
2226
2227         crit_enter();
2228
2229         ifac = &ifa->ifa_containers[cpu];
2230         ASSERT_IFAC_VALID(ifac);
2231         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2232                 ("ifaddr is on if_addrheads\n"));
2233
2234         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2235         if (msg->tail)
2236                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2237         else
2238                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2239
2240         crit_exit();
2241
2242         ifa_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
2243 }
2244
2245 void
2246 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2247 {
2248         struct netmsg_ifaddr msg;
2249
2250         netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2251                     ifa_iflink_dispatch);
2252         msg.ifa = ifa;
2253         msg.ifp = ifp;
2254         msg.tail = tail;
2255
2256         ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2257 }
2258
2259 static void
2260 ifa_ifunlink_dispatch(struct netmsg *nmsg)
2261 {
2262         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2263         struct ifaddr *ifa = msg->ifa;
2264         struct ifnet *ifp = msg->ifp;
2265         int cpu = mycpuid;
2266         struct ifaddr_container *ifac;
2267
2268         crit_enter();
2269
2270         ifac = &ifa->ifa_containers[cpu];
2271         ASSERT_IFAC_VALID(ifac);
2272         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2273                 ("ifaddr is not on if_addrhead\n"));
2274
2275         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2276         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2277
2278         crit_exit();
2279
2280         ifa_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
2281 }
2282
2283 void
2284 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2285 {
2286         struct netmsg_ifaddr msg;
2287
2288         netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2289                     ifa_ifunlink_dispatch);
2290         msg.ifa = ifa;
2291         msg.ifp = ifp;
2292
2293         ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2294 }
2295
2296 static void
2297 ifa_destroy_dispatch(struct netmsg *nmsg)
2298 {
2299         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2300
2301         IFAFREE(msg->ifa);
2302         ifa_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2303 }
2304
2305 void
2306 ifa_destroy(struct ifaddr *ifa)
2307 {
2308         struct netmsg_ifaddr msg;
2309
2310         netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2311                     ifa_destroy_dispatch);
2312         msg.ifa = ifa;
2313
2314         ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2315 }
2316
2317 struct lwkt_port *
2318 ifnet_portfn(int cpu)
2319 {
2320         return &ifnet_threads[cpu].td_msgport;
2321 }
2322
2323 void
2324 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2325 {
2326         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2327
2328         if (next_cpu < ncpus)
2329                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2330         else
2331                 lwkt_replymsg(lmsg, 0);
2332 }
2333
2334 void
2335 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
2336 {
2337         KKASSERT(cpu < ncpus);
2338         lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
2339 }
2340
2341 void
2342 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
2343 {
2344         KKASSERT(cpu < ncpus);
2345         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
2346 }
2347
2348 static void
2349 ifnetinit(void *dummy __unused)
2350 {
2351         int i;
2352
2353         for (i = 0; i < ncpus; ++i) {
2354                 struct thread *thr = &ifnet_threads[i];
2355
2356                 lwkt_create(netmsg_service_loop, &ifnet_mpsafe_thread, NULL,
2357                             thr, TDF_NETWORK | TDF_MPSAFE, i, "ifnet %d", i);
2358                 netmsg_service_port_init(&thr->td_msgport);
2359         }
2360 }