Add ifpoll, which support hardware TX/RX queues based polling.
[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 1.84 2008/11/15 11:58:16 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 #include "opt_ifpoll.h"
43
44 #include <sys/param.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/systm.h>
48 #include <sys/proc.h>
49 #include <sys/priv.h>
50 #include <sys/protosw.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/socketops.h>
54 #include <sys/protosw.h>
55 #include <sys/kernel.h>
56 #include <sys/ktr.h>
57 #include <sys/sockio.h>
58 #include <sys/syslog.h>
59 #include <sys/sysctl.h>
60 #include <sys/domain.h>
61 #include <sys/thread.h>
62 #include <sys/thread2.h>
63 #include <sys/serialize.h>
64 #include <sys/msgport2.h>
65 #include <sys/bus.h>
66
67 #include <net/if.h>
68 #include <net/if_arp.h>
69 #include <net/if_dl.h>
70 #include <net/if_types.h>
71 #include <net/if_var.h>
72 #include <net/ifq_var.h>
73 #include <net/radix.h>
74 #include <net/route.h>
75 #include <net/if_clone.h>
76 #include <net/netisr.h>
77 #include <net/netmsg2.h>
78
79 #include <machine/atomic.h>
80 #include <machine/stdarg.h>
81 #include <machine/smp.h>
82
83 #if defined(INET) || defined(INET6)
84 /*XXX*/
85 #include <netinet/in.h>
86 #include <netinet/in_var.h>
87 #include <netinet/if_ether.h>
88 #ifdef INET6
89 #include <netinet6/in6_var.h>
90 #include <netinet6/in6_ifattach.h>
91 #endif
92 #endif
93
94 #if defined(COMPAT_43)
95 #include <emulation/43bsd/43bsd_socket.h>
96 #endif /* COMPAT_43 */
97
98 struct netmsg_ifaddr {
99         struct netmsg   netmsg;
100         struct ifaddr   *ifa;
101         struct ifnet    *ifp;
102         int             tail;
103 };
104
105 /*
106  * System initialization
107  */
108 static void     if_attachdomain(void *);
109 static void     if_attachdomain1(struct ifnet *);
110 static int      ifconf(u_long, caddr_t, struct ucred *);
111 static void     ifinit(void *);
112 static void     ifnetinit(void *);
113 static void     if_slowtimo(void *);
114 static void     link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
115 static int      if_rtdel(struct radix_node *, void *);
116
117 #ifdef INET6
118 /*
119  * XXX: declare here to avoid to include many inet6 related files..
120  * should be more generalized?
121  */
122 extern void     nd6_setmtu(struct ifnet *);
123 #endif
124
125 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
126 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
127
128 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
129 /* Must be after netisr_init */
130 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
131
132 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
133 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
134
135 int                     ifqmaxlen = IFQ_MAXLEN;
136 struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
137
138 /* In ifq_dispatch(), try to do direct ifnet.if_start first */
139 static int              ifq_dispatch_schedonly = 0;
140 SYSCTL_INT(_net_link_generic, OID_AUTO, ifq_dispatch_schedonly, CTLFLAG_RW,
141            &ifq_dispatch_schedonly, 0, "");
142
143 /* In ifq_dispatch(), schedule ifnet.if_start without checking ifnet.if_snd */
144 static int              ifq_dispatch_schednochk = 0;
145 SYSCTL_INT(_net_link_generic, OID_AUTO, ifq_dispatch_schednochk, CTLFLAG_RW,
146            &ifq_dispatch_schednochk, 0, "");
147
148 /* In if_devstart(), try to do direct ifnet.if_start first */
149 static int              if_devstart_schedonly = 0;
150 SYSCTL_INT(_net_link_generic, OID_AUTO, if_devstart_schedonly, CTLFLAG_RW,
151            &if_devstart_schedonly, 0, "");
152
153 /* In if_devstart(), schedule ifnet.if_start without checking ifnet.if_snd */
154 static int              if_devstart_schednochk = 0;
155 SYSCTL_INT(_net_link_generic, OID_AUTO, if_devstart_schednochk, CTLFLAG_RW,
156            &if_devstart_schednochk, 0, "");
157
158 #ifdef SMP
159 /* Schedule ifnet.if_start on the current CPU */
160 static int              if_start_oncpu_sched = 0;
161 SYSCTL_INT(_net_link_generic, OID_AUTO, if_start_oncpu_sched, CTLFLAG_RW,
162            &if_start_oncpu_sched, 0, "");
163 #endif
164
165 struct callout          if_slowtimo_timer;
166
167 int                     if_index = 0;
168 struct ifnet            **ifindex2ifnet = NULL;
169 static struct thread    ifnet_threads[MAXCPU];
170 static int              ifnet_mpsafe_thread = NETMSG_SERVICE_MPSAFE;
171
172 #define IFQ_KTR_STRING          "ifq=%p"
173 #define IFQ_KTR_ARG_SIZE        (sizeof(void *))
174 #ifndef KTR_IFQ
175 #define KTR_IFQ                 KTR_ALL
176 #endif
177 KTR_INFO_MASTER(ifq);
178 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARG_SIZE);
179 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARG_SIZE);
180 #define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)
181
182 #define IF_START_KTR_STRING     "ifp=%p"
183 #define IF_START_KTR_ARG_SIZE   (sizeof(void *))
184 #ifndef KTR_IF_START
185 #define KTR_IF_START            KTR_ALL
186 #endif
187 KTR_INFO_MASTER(if_start);
188 KTR_INFO(KTR_IF_START, if_start, run, 0,
189          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
190 KTR_INFO(KTR_IF_START, if_start, sched, 1,
191          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
192 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
193          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
194 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
195          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
196 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
197          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
198 #define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
199
200 /*
201  * Network interface utility routines.
202  *
203  * Routines with ifa_ifwith* names take sockaddr *'s as
204  * parameters.
205  */
206 /* ARGSUSED*/
207 void
208 ifinit(void *dummy)
209 {
210         struct ifnet *ifp;
211
212         callout_init(&if_slowtimo_timer);
213
214         crit_enter();
215         TAILQ_FOREACH(ifp, &ifnet, if_link) {
216                 if (ifp->if_snd.ifq_maxlen == 0) {
217                         if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
218                         ifp->if_snd.ifq_maxlen = ifqmaxlen;
219                 }
220         }
221         crit_exit();
222
223         if_slowtimo(0);
224 }
225
226 static int
227 if_start_cpuid(struct ifnet *ifp)
228 {
229         return ifp->if_cpuid;
230 }
231
232 #ifdef DEVICE_POLLING
233 static int
234 if_start_cpuid_poll(struct ifnet *ifp)
235 {
236         int poll_cpuid = ifp->if_poll_cpuid;
237
238         if (poll_cpuid >= 0)
239                 return poll_cpuid;
240         else
241                 return ifp->if_cpuid;
242 }
243 #endif
244
245 static void
246 if_start_ipifunc(void *arg)
247 {
248         struct ifnet *ifp = arg;
249         struct lwkt_msg *lmsg = &ifp->if_start_nmsg[mycpuid].nm_lmsg;
250
251         crit_enter();
252         if (lmsg->ms_flags & MSGF_DONE)
253                 lwkt_sendmsg(ifnet_portfn(mycpuid), lmsg);
254         crit_exit();
255 }
256
257 /*
258  * Schedule ifnet.if_start on ifnet's CPU
259  */
260 static void
261 if_start_schedule(struct ifnet *ifp)
262 {
263 #ifdef SMP
264         int cpu;
265
266         if (if_start_oncpu_sched)
267                 cpu = mycpuid;
268         else
269                 cpu = ifp->if_start_cpuid(ifp);
270
271         if (cpu != mycpuid)
272                 lwkt_send_ipiq(globaldata_find(cpu), if_start_ipifunc, ifp);
273         else
274 #endif
275         if_start_ipifunc(ifp);
276 }
277
278 /*
279  * NOTE:
280  * This function will release ifnet.if_start interlock,
281  * if ifnet.if_start does not need to be scheduled
282  */
283 static __inline int
284 if_start_need_schedule(struct ifaltq *ifq, int running)
285 {
286         if (!running || ifq_is_empty(ifq)
287 #ifdef ALTQ
288             || ifq->altq_tbr != NULL
289 #endif
290         ) {
291                 ALTQ_LOCK(ifq);
292                 /*
293                  * ifnet.if_start interlock is released, if:
294                  * 1) Hardware can not take any packets, due to
295                  *    o  interface is marked down
296                  *    o  hardware queue is full (IFF_OACTIVE)
297                  *    Under the second situation, hardware interrupt
298                  *    or polling(4) will call/schedule ifnet.if_start
299                  *    when hardware queue is ready
300                  * 2) There is not packet in the ifnet.if_snd.
301                  *    Further ifq_dispatch or ifq_handoff will call/
302                  *    schedule ifnet.if_start
303                  * 3) TBR is used and it does not allow further
304                  *    dequeueing.
305                  *    TBR callout will call ifnet.if_start
306                  */
307                 if (!running || !ifq_data_ready(ifq)) {
308                         ifq->altq_started = 0;
309                         ALTQ_UNLOCK(ifq);
310                         return 0;
311                 }
312                 ALTQ_UNLOCK(ifq);
313         }
314         return 1;
315 }
316
317 static void
318 if_start_dispatch(struct netmsg *nmsg)
319 {
320         struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
321         struct ifnet *ifp = lmsg->u.ms_resultp;
322         struct ifaltq *ifq = &ifp->if_snd;
323         int running = 0;
324
325         crit_enter();
326         lwkt_replymsg(lmsg, 0); /* reply ASAP */
327         crit_exit();
328
329 #ifdef SMP
330         if (!if_start_oncpu_sched && mycpuid != ifp->if_start_cpuid(ifp)) {
331                 /*
332                  * If the ifnet is still up, we need to
333                  * chase its CPU change.
334                  */
335                 if (ifp->if_flags & IFF_UP) {
336                         logifstart(chase_sched, ifp);
337                         if_start_schedule(ifp);
338                         return;
339                 } else {
340                         goto check;
341                 }
342         }
343 #endif
344
345         if (ifp->if_flags & IFF_UP) {
346                 ifnet_serialize_tx(ifp); /* XXX try? */
347                 if ((ifp->if_flags & IFF_OACTIVE) == 0) {
348                         logifstart(run, ifp);
349                         ifp->if_start(ifp);
350                         if ((ifp->if_flags &
351                         (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
352                                 running = 1;
353                 }
354                 ifnet_deserialize_tx(ifp);
355         }
356 #ifdef SMP
357 check:
358 #endif
359         if (if_start_need_schedule(ifq, running)) {
360                 crit_enter();
361                 if (lmsg->ms_flags & MSGF_DONE) { /* XXX necessary? */
362                         logifstart(sched, ifp);
363                         lwkt_sendmsg(ifnet_portfn(mycpuid), lmsg);
364                 }
365                 crit_exit();
366         }
367 }
368
369 /* Device driver ifnet.if_start helper function */
370 void
371 if_devstart(struct ifnet *ifp)
372 {
373         struct ifaltq *ifq = &ifp->if_snd;
374         int running = 0;
375
376         ASSERT_IFNET_SERIALIZED_TX(ifp);
377
378         ALTQ_LOCK(ifq);
379         if (ifq->altq_started || !ifq_data_ready(ifq)) {
380                 logifstart(avoid, ifp);
381                 ALTQ_UNLOCK(ifq);
382                 return;
383         }
384         ifq->altq_started = 1;
385         ALTQ_UNLOCK(ifq);
386
387         if (if_devstart_schedonly) {
388                 /*
389                  * Always schedule ifnet.if_start on ifnet's CPU,
390                  * short circuit the rest of this function.
391                  */
392                 logifstart(sched, ifp);
393                 if_start_schedule(ifp);
394                 return;
395         }
396
397         logifstart(run, ifp);
398         ifp->if_start(ifp);
399
400         if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
401                 running = 1;
402
403         if (if_devstart_schednochk || if_start_need_schedule(ifq, running)) {
404                 /*
405                  * More data need to be transmitted, ifnet.if_start is
406                  * scheduled on ifnet's CPU, and we keep going.
407                  * NOTE: ifnet.if_start interlock is not released.
408                  */
409                 logifstart(sched, ifp);
410                 if_start_schedule(ifp);
411         }
412 }
413
414 static void
415 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
416 {
417         lwkt_serialize_enter(ifp->if_serializer);
418 }
419
420 static void
421 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
422 {
423         lwkt_serialize_exit(ifp->if_serializer);
424 }
425
426 static int
427 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
428 {
429         return lwkt_serialize_try(ifp->if_serializer);
430 }
431
432 #ifdef INVARIANTS
433 static void
434 if_default_serialize_assert(struct ifnet *ifp,
435                             enum ifnet_serialize slz __unused,
436                             boolean_t serialized)
437 {
438         if (serialized)
439                 ASSERT_SERIALIZED(ifp->if_serializer);
440         else
441                 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
442 }
443 #endif
444
445 /*
446  * Attach an interface to the list of "active" interfaces.
447  *
448  * The serializer is optional.  If non-NULL access to the interface
449  * may be MPSAFE.
450  */
451 void
452 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
453 {
454         unsigned socksize, ifasize;
455         int namelen, masklen;
456         struct sockaddr_dl *sdl;
457         struct ifaddr *ifa;
458         struct ifaltq *ifq;
459         int i;
460
461         static int if_indexlim = 8;
462
463         if (ifp->if_serialize != NULL) {
464                 KASSERT(ifp->if_deserialize != NULL &&
465                         ifp->if_tryserialize != NULL &&
466                         ifp->if_serialize_assert != NULL,
467                         ("serialize functions are partially setup\n"));
468
469                 /*
470                  * If the device supplies serialize functions,
471                  * then clear if_serializer to catch any invalid
472                  * usage of this field.
473                  */
474                 KASSERT(serializer == NULL,
475                         ("both serialize functions and default serializer "
476                          "are supplied\n"));
477                 ifp->if_serializer = NULL;
478         } else {
479                 KASSERT(ifp->if_deserialize == NULL &&
480                         ifp->if_tryserialize == NULL &&
481                         ifp->if_serialize_assert == NULL,
482                         ("serialize functions are partially setup\n"));
483                 ifp->if_serialize = if_default_serialize;
484                 ifp->if_deserialize = if_default_deserialize;
485                 ifp->if_tryserialize = if_default_tryserialize;
486 #ifdef INVARIANTS
487                 ifp->if_serialize_assert = if_default_serialize_assert;
488 #endif
489
490                 /*
491                  * The serializer can be passed in from the device,
492                  * allowing the same serializer to be used for both
493                  * the interrupt interlock and the device queue.
494                  * If not specified, the netif structure will use an
495                  * embedded serializer.
496                  */
497                 if (serializer == NULL) {
498                         serializer = &ifp->if_default_serializer;
499                         lwkt_serialize_init(serializer);
500                 }
501                 ifp->if_serializer = serializer;
502         }
503
504         ifp->if_start_cpuid = if_start_cpuid;
505         ifp->if_cpuid = 0;
506
507 #ifdef DEVICE_POLLING
508         /* Device is not in polling mode by default */
509         ifp->if_poll_cpuid = -1;
510         if (ifp->if_poll != NULL)
511                 ifp->if_start_cpuid = if_start_cpuid_poll;
512 #endif
513
514         ifp->if_start_nmsg = kmalloc(ncpus * sizeof(struct netmsg),
515                                      M_LWKTMSG, M_WAITOK);
516         for (i = 0; i < ncpus; ++i) {
517                 netmsg_init(&ifp->if_start_nmsg[i], &netisr_adone_rport, 0,
518                             if_start_dispatch);
519                 ifp->if_start_nmsg[i].nm_lmsg.u.ms_resultp = ifp;
520         }
521
522         TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
523         ifp->if_index = ++if_index;
524
525         /*
526          * XXX -
527          * The old code would work if the interface passed a pre-existing
528          * chain of ifaddrs to this code.  We don't trust our callers to
529          * properly initialize the tailq, however, so we no longer allow
530          * this unlikely case.
531          */
532         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
533                                     M_IFADDR, M_WAITOK | M_ZERO);
534         for (i = 0; i < ncpus; ++i)
535                 TAILQ_INIT(&ifp->if_addrheads[i]);
536
537         TAILQ_INIT(&ifp->if_prefixhead);
538         LIST_INIT(&ifp->if_multiaddrs);
539         getmicrotime(&ifp->if_lastchange);
540         if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
541                 unsigned int n;
542                 struct ifnet **q;
543
544                 if_indexlim <<= 1;
545
546                 /* grow ifindex2ifnet */
547                 n = if_indexlim * sizeof(*q);
548                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
549                 if (ifindex2ifnet) {
550                         bcopy(ifindex2ifnet, q, n/2);
551                         kfree(ifindex2ifnet, M_IFADDR);
552                 }
553                 ifindex2ifnet = q;
554         }
555
556         ifindex2ifnet[if_index] = ifp;
557
558         /*
559          * create a Link Level name for this device
560          */
561         namelen = strlen(ifp->if_xname);
562 #define _offsetof(t, m) ((int)((caddr_t)&((t *)0)->m))
563         masklen = _offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
564         socksize = masklen + ifp->if_addrlen;
565 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
566         if (socksize < sizeof(*sdl))
567                 socksize = sizeof(*sdl);
568         socksize = ROUNDUP(socksize);
569         ifasize = sizeof(struct ifaddr) + 2 * socksize;
570         ifa = ifa_create(ifasize, M_WAITOK);
571         sdl = (struct sockaddr_dl *)(ifa + 1);
572         sdl->sdl_len = socksize;
573         sdl->sdl_family = AF_LINK;
574         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
575         sdl->sdl_nlen = namelen;
576         sdl->sdl_index = ifp->if_index;
577         sdl->sdl_type = ifp->if_type;
578         ifp->if_lladdr = ifa;
579         ifa->ifa_ifp = ifp;
580         ifa->ifa_rtrequest = link_rtrequest;
581         ifa->ifa_addr = (struct sockaddr *)sdl;
582         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
583         ifa->ifa_netmask = (struct sockaddr *)sdl;
584         sdl->sdl_len = masklen;
585         while (namelen != 0)
586                 sdl->sdl_data[--namelen] = 0xff;
587         ifa_iflink(ifa, ifp, 0 /* Insert head */);
588
589         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
590         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
591
592         ifq = &ifp->if_snd;
593         ifq->altq_type = 0;
594         ifq->altq_disc = NULL;
595         ifq->altq_flags &= ALTQF_CANTCHANGE;
596         ifq->altq_tbr = NULL;
597         ifq->altq_ifp = ifp;
598         ifq->altq_started = 0;
599         ifq->altq_prepended = NULL;
600         ALTQ_LOCK_INIT(ifq);
601         ifq_set_classic(ifq);
602
603         if (!SLIST_EMPTY(&domains))
604                 if_attachdomain1(ifp);
605
606         /* Announce the interface. */
607         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
608 }
609
610 static void
611 if_attachdomain(void *dummy)
612 {
613         struct ifnet *ifp;
614
615         crit_enter();
616         TAILQ_FOREACH(ifp, &ifnet, if_list)
617                 if_attachdomain1(ifp);
618         crit_exit();
619 }
620 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
621         if_attachdomain, NULL);
622
623 static void
624 if_attachdomain1(struct ifnet *ifp)
625 {
626         struct domain *dp;
627
628         crit_enter();
629
630         /* address family dependent data region */
631         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
632         SLIST_FOREACH(dp, &domains, dom_next)
633                 if (dp->dom_ifattach)
634                         ifp->if_afdata[dp->dom_family] =
635                                 (*dp->dom_ifattach)(ifp);
636         crit_exit();
637 }
638
639 /*
640  * Purge all addresses whose type is _not_ AF_LINK
641  */
642 void
643 if_purgeaddrs_nolink(struct ifnet *ifp)
644 {
645         struct ifaddr_container *ifac, *next;
646
647         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
648                               ifa_link, next) {
649                 struct ifaddr *ifa = ifac->ifa;
650
651                 /* Leave link ifaddr as it is */
652                 if (ifa->ifa_addr->sa_family == AF_LINK)
653                         continue;
654 #ifdef INET
655                 /* XXX: Ugly!! ad hoc just for INET */
656                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
657                         struct ifaliasreq ifr;
658 #ifdef IFADDR_DEBUG_VERBOSE
659                         int i;
660
661                         kprintf("purge in4 addr %p: ", ifa);
662                         for (i = 0; i < ncpus; ++i)
663                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
664                         kprintf("\n");
665 #endif
666
667                         bzero(&ifr, sizeof ifr);
668                         ifr.ifra_addr = *ifa->ifa_addr;
669                         if (ifa->ifa_dstaddr)
670                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
671                         if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
672                                        NULL) == 0)
673                                 continue;
674                 }
675 #endif /* INET */
676 #ifdef INET6
677                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
678 #ifdef IFADDR_DEBUG_VERBOSE
679                         int i;
680
681                         kprintf("purge in6 addr %p: ", ifa);
682                         for (i = 0; i < ncpus; ++i)
683                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
684                         kprintf("\n");
685 #endif
686
687                         in6_purgeaddr(ifa);
688                         /* ifp_addrhead is already updated */
689                         continue;
690                 }
691 #endif /* INET6 */
692                 ifa_ifunlink(ifa, ifp);
693                 ifa_destroy(ifa);
694         }
695 }
696
697 /*
698  * Detach an interface, removing it from the
699  * list of "active" interfaces.
700  */
701 void
702 if_detach(struct ifnet *ifp)
703 {
704         struct radix_node_head  *rnh;
705         int i;
706         int cpu, origcpu;
707         struct domain *dp;
708
709         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
710
711         /*
712          * Remove routes and flush queues.
713          */
714         crit_enter();
715 #ifdef DEVICE_POLLING
716         if (ifp->if_flags & IFF_POLLING)
717                 ether_poll_deregister(ifp);
718 #endif
719 #ifdef IFPOLL_ENABLE
720         if (ifp->if_flags & IFF_NPOLLING)
721                 ifpoll_deregister(ifp);
722 #endif
723         if_down(ifp);
724
725         if (ifq_is_enabled(&ifp->if_snd))
726                 altq_disable(&ifp->if_snd);
727         if (ifq_is_attached(&ifp->if_snd))
728                 altq_detach(&ifp->if_snd);
729
730         /*
731          * Clean up all addresses.
732          */
733         ifp->if_lladdr = NULL;
734
735         if_purgeaddrs_nolink(ifp);
736         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
737                 struct ifaddr *ifa;
738
739                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
740                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
741                         ("non-link ifaddr is left on if_addrheads"));
742
743                 ifa_ifunlink(ifa, ifp);
744                 ifa_destroy(ifa);
745                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
746                         ("there are still ifaddrs left on if_addrheads"));
747         }
748
749 #ifdef INET
750         /*
751          * Remove all IPv4 kernel structures related to ifp.
752          */
753         in_ifdetach(ifp);
754 #endif
755
756 #ifdef INET6
757         /*
758          * Remove all IPv6 kernel structs related to ifp.  This should be done
759          * before removing routing entries below, since IPv6 interface direct
760          * routes are expected to be removed by the IPv6-specific kernel API.
761          * Otherwise, the kernel will detect some inconsistency and bark it.
762          */
763         in6_ifdetach(ifp);
764 #endif
765
766         /*
767          * Delete all remaining routes using this interface
768          * Unfortuneatly the only way to do this is to slog through
769          * the entire routing table looking for routes which point
770          * to this interface...oh well...
771          */
772         origcpu = mycpuid;
773         for (cpu = 0; cpu < ncpus2; cpu++) {
774                 lwkt_migratecpu(cpu);
775                 for (i = 1; i <= AF_MAX; i++) {
776                         if ((rnh = rt_tables[cpu][i]) == NULL)
777                                 continue;
778                         rnh->rnh_walktree(rnh, if_rtdel, ifp);
779                 }
780         }
781         lwkt_migratecpu(origcpu);
782
783         /* Announce that the interface is gone. */
784         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
785         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
786
787         SLIST_FOREACH(dp, &domains, dom_next)
788                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
789                         (*dp->dom_ifdetach)(ifp,
790                                 ifp->if_afdata[dp->dom_family]);
791
792         /*
793          * Remove interface from ifindex2ifp[] and maybe decrement if_index.
794          */
795         ifindex2ifnet[ifp->if_index] = NULL;
796         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
797                 if_index--;
798
799         TAILQ_REMOVE(&ifnet, ifp, if_link);
800         kfree(ifp->if_addrheads, M_IFADDR);
801         kfree(ifp->if_start_nmsg, M_LWKTMSG);
802         crit_exit();
803 }
804
805 /*
806  * Delete Routes for a Network Interface
807  *
808  * Called for each routing entry via the rnh->rnh_walktree() call above
809  * to delete all route entries referencing a detaching network interface.
810  *
811  * Arguments:
812  *      rn      pointer to node in the routing table
813  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
814  *
815  * Returns:
816  *      0       successful
817  *      errno   failed - reason indicated
818  *
819  */
820 static int
821 if_rtdel(struct radix_node *rn, void *arg)
822 {
823         struct rtentry  *rt = (struct rtentry *)rn;
824         struct ifnet    *ifp = arg;
825         int             err;
826
827         if (rt->rt_ifp == ifp) {
828
829                 /*
830                  * Protect (sorta) against walktree recursion problems
831                  * with cloned routes
832                  */
833                 if (!(rt->rt_flags & RTF_UP))
834                         return (0);
835
836                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
837                                 rt_mask(rt), rt->rt_flags,
838                                 NULL);
839                 if (err) {
840                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
841                 }
842         }
843
844         return (0);
845 }
846
847 /*
848  * Locate an interface based on a complete address.
849  */
850 struct ifaddr *
851 ifa_ifwithaddr(struct sockaddr *addr)
852 {
853         struct ifnet *ifp;
854
855         TAILQ_FOREACH(ifp, &ifnet, if_link) {
856                 struct ifaddr_container *ifac;
857
858                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
859                         struct ifaddr *ifa = ifac->ifa;
860
861                         if (ifa->ifa_addr->sa_family != addr->sa_family)
862                                 continue;
863                         if (sa_equal(addr, ifa->ifa_addr))
864                                 return (ifa);
865                         if ((ifp->if_flags & IFF_BROADCAST) &&
866                             ifa->ifa_broadaddr &&
867                             /* IPv6 doesn't have broadcast */
868                             ifa->ifa_broadaddr->sa_len != 0 &&
869                             sa_equal(ifa->ifa_broadaddr, addr))
870                                 return (ifa);
871                 }
872         }
873         return (NULL);
874 }
875 /*
876  * Locate the point to point interface with a given destination address.
877  */
878 struct ifaddr *
879 ifa_ifwithdstaddr(struct sockaddr *addr)
880 {
881         struct ifnet *ifp;
882
883         TAILQ_FOREACH(ifp, &ifnet, if_link) {
884                 struct ifaddr_container *ifac;
885
886                 if (!(ifp->if_flags & IFF_POINTOPOINT))
887                         continue;
888
889                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
890                         struct ifaddr *ifa = ifac->ifa;
891
892                         if (ifa->ifa_addr->sa_family != addr->sa_family)
893                                 continue;
894                         if (ifa->ifa_dstaddr &&
895                             sa_equal(addr, ifa->ifa_dstaddr))
896                                 return (ifa);
897                 }
898         }
899         return (NULL);
900 }
901
902 /*
903  * Find an interface on a specific network.  If many, choice
904  * is most specific found.
905  */
906 struct ifaddr *
907 ifa_ifwithnet(struct sockaddr *addr)
908 {
909         struct ifnet *ifp;
910         struct ifaddr *ifa_maybe = NULL;
911         u_int af = addr->sa_family;
912         char *addr_data = addr->sa_data, *cplim;
913
914         /*
915          * AF_LINK addresses can be looked up directly by their index number,
916          * so do that if we can.
917          */
918         if (af == AF_LINK) {
919                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
920
921                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
922                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
923         }
924
925         /*
926          * Scan though each interface, looking for ones that have
927          * addresses in this address family.
928          */
929         TAILQ_FOREACH(ifp, &ifnet, if_link) {
930                 struct ifaddr_container *ifac;
931
932                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
933                         struct ifaddr *ifa = ifac->ifa;
934                         char *cp, *cp2, *cp3;
935
936                         if (ifa->ifa_addr->sa_family != af)
937 next:                           continue;
938                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
939                                 /*
940                                  * This is a bit broken as it doesn't
941                                  * take into account that the remote end may
942                                  * be a single node in the network we are
943                                  * looking for.
944                                  * The trouble is that we don't know the
945                                  * netmask for the remote end.
946                                  */
947                                 if (ifa->ifa_dstaddr != NULL &&
948                                     sa_equal(addr, ifa->ifa_dstaddr))
949                                         return (ifa);
950                         } else {
951                                 /*
952                                  * if we have a special address handler,
953                                  * then use it instead of the generic one.
954                                  */
955                                 if (ifa->ifa_claim_addr) {
956                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
957                                                 return (ifa);
958                                         } else {
959                                                 continue;
960                                         }
961                                 }
962
963                                 /*
964                                  * Scan all the bits in the ifa's address.
965                                  * If a bit dissagrees with what we are
966                                  * looking for, mask it with the netmask
967                                  * to see if it really matters.
968                                  * (A byte at a time)
969                                  */
970                                 if (ifa->ifa_netmask == 0)
971                                         continue;
972                                 cp = addr_data;
973                                 cp2 = ifa->ifa_addr->sa_data;
974                                 cp3 = ifa->ifa_netmask->sa_data;
975                                 cplim = ifa->ifa_netmask->sa_len +
976                                         (char *)ifa->ifa_netmask;
977                                 while (cp3 < cplim)
978                                         if ((*cp++ ^ *cp2++) & *cp3++)
979                                                 goto next; /* next address! */
980                                 /*
981                                  * If the netmask of what we just found
982                                  * is more specific than what we had before
983                                  * (if we had one) then remember the new one
984                                  * before continuing to search
985                                  * for an even better one.
986                                  */
987                                 if (ifa_maybe == 0 ||
988                                     rn_refines((char *)ifa->ifa_netmask,
989                                                (char *)ifa_maybe->ifa_netmask))
990                                         ifa_maybe = ifa;
991                         }
992                 }
993         }
994         return (ifa_maybe);
995 }
996
997 /*
998  * Find an interface address specific to an interface best matching
999  * a given address.
1000  */
1001 struct ifaddr *
1002 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1003 {
1004         struct ifaddr_container *ifac;
1005         char *cp, *cp2, *cp3;
1006         char *cplim;
1007         struct ifaddr *ifa_maybe = 0;
1008         u_int af = addr->sa_family;
1009
1010         if (af >= AF_MAX)
1011                 return (0);
1012         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1013                 struct ifaddr *ifa = ifac->ifa;
1014
1015                 if (ifa->ifa_addr->sa_family != af)
1016                         continue;
1017                 if (ifa_maybe == 0)
1018                         ifa_maybe = ifa;
1019                 if (ifa->ifa_netmask == NULL) {
1020                         if (sa_equal(addr, ifa->ifa_addr) ||
1021                             (ifa->ifa_dstaddr != NULL &&
1022                              sa_equal(addr, ifa->ifa_dstaddr)))
1023                                 return (ifa);
1024                         continue;
1025                 }
1026                 if (ifp->if_flags & IFF_POINTOPOINT) {
1027                         if (sa_equal(addr, ifa->ifa_dstaddr))
1028                                 return (ifa);
1029                 } else {
1030                         cp = addr->sa_data;
1031                         cp2 = ifa->ifa_addr->sa_data;
1032                         cp3 = ifa->ifa_netmask->sa_data;
1033                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1034                         for (; cp3 < cplim; cp3++)
1035                                 if ((*cp++ ^ *cp2++) & *cp3)
1036                                         break;
1037                         if (cp3 == cplim)
1038                                 return (ifa);
1039                 }
1040         }
1041         return (ifa_maybe);
1042 }
1043
1044 /*
1045  * Default action when installing a route with a Link Level gateway.
1046  * Lookup an appropriate real ifa to point to.
1047  * This should be moved to /sys/net/link.c eventually.
1048  */
1049 static void
1050 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1051 {
1052         struct ifaddr *ifa;
1053         struct sockaddr *dst;
1054         struct ifnet *ifp;
1055
1056         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1057             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1058                 return;
1059         ifa = ifaof_ifpforaddr(dst, ifp);
1060         if (ifa != NULL) {
1061                 IFAFREE(rt->rt_ifa);
1062                 IFAREF(ifa);
1063                 rt->rt_ifa = ifa;
1064                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1065                         ifa->ifa_rtrequest(cmd, rt, info);
1066         }
1067 }
1068
1069 /*
1070  * Mark an interface down and notify protocols of
1071  * the transition.
1072  * NOTE: must be called at splnet or eqivalent.
1073  */
1074 void
1075 if_unroute(struct ifnet *ifp, int flag, int fam)
1076 {
1077         struct ifaddr_container *ifac;
1078
1079         ifp->if_flags &= ~flag;
1080         getmicrotime(&ifp->if_lastchange);
1081         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1082                 struct ifaddr *ifa = ifac->ifa;
1083
1084                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1085                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1086         }
1087         ifq_purge(&ifp->if_snd);
1088         rt_ifmsg(ifp);
1089 }
1090
1091 /*
1092  * Mark an interface up and notify protocols of
1093  * the transition.
1094  * NOTE: must be called at splnet or eqivalent.
1095  */
1096 void
1097 if_route(struct ifnet *ifp, int flag, int fam)
1098 {
1099         struct ifaddr_container *ifac;
1100
1101         ifq_purge(&ifp->if_snd);
1102         ifp->if_flags |= flag;
1103         getmicrotime(&ifp->if_lastchange);
1104         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1105                 struct ifaddr *ifa = ifac->ifa;
1106
1107                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1108                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1109         }
1110         rt_ifmsg(ifp);
1111 #ifdef INET6
1112         in6_if_up(ifp);
1113 #endif
1114 }
1115
1116 /*
1117  * Mark an interface down and notify protocols of the transition.  An
1118  * interface going down is also considered to be a synchronizing event.
1119  * We must ensure that all packet processing related to the interface
1120  * has completed before we return so e.g. the caller can free the ifnet
1121  * structure that the mbufs may be referencing.
1122  *
1123  * NOTE: must be called at splnet or eqivalent.
1124  */
1125 void
1126 if_down(struct ifnet *ifp)
1127 {
1128         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1129         netmsg_service_sync();
1130 }
1131
1132 /*
1133  * Mark an interface up and notify protocols of
1134  * the transition.
1135  * NOTE: must be called at splnet or eqivalent.
1136  */
1137 void
1138 if_up(struct ifnet *ifp)
1139 {
1140         if_route(ifp, IFF_UP, AF_UNSPEC);
1141 }
1142
1143 /*
1144  * Process a link state change.
1145  * NOTE: must be called at splsoftnet or equivalent.
1146  */
1147 void
1148 if_link_state_change(struct ifnet *ifp)
1149 {
1150         int link_state = ifp->if_link_state;
1151
1152         rt_ifmsg(ifp);
1153         devctl_notify("IFNET", ifp->if_xname,
1154             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1155 }
1156
1157 /*
1158  * Handle interface watchdog timer routines.  Called
1159  * from softclock, we decrement timers (if set) and
1160  * call the appropriate interface routine on expiration.
1161  */
1162 static void
1163 if_slowtimo(void *arg)
1164 {
1165         struct ifnet *ifp;
1166
1167         crit_enter();
1168
1169         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1170                 if (ifp->if_timer == 0 || --ifp->if_timer)
1171                         continue;
1172                 if (ifp->if_watchdog) {
1173                         if (ifnet_tryserialize_all(ifp)) {
1174                                 (*ifp->if_watchdog)(ifp);
1175                                 ifnet_deserialize_all(ifp);
1176                         } else {
1177                                 /* try again next timeout */
1178                                 ++ifp->if_timer;
1179                         }
1180                 }
1181         }
1182
1183         crit_exit();
1184
1185         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1186 }
1187
1188 /*
1189  * Map interface name to
1190  * interface structure pointer.
1191  */
1192 struct ifnet *
1193 ifunit(const char *name)
1194 {
1195         struct ifnet *ifp;
1196
1197         /*
1198          * Search all the interfaces for this name/number
1199          */
1200
1201         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1202                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1203                         break;
1204         }
1205         return (ifp);
1206 }
1207
1208
1209 /*
1210  * Map interface name in a sockaddr_dl to
1211  * interface structure pointer.
1212  */
1213 struct ifnet *
1214 if_withname(struct sockaddr *sa)
1215 {
1216         char ifname[IFNAMSIZ+1];
1217         struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1218
1219         if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1220              (sdl->sdl_nlen > IFNAMSIZ) )
1221                 return NULL;
1222
1223         /*
1224          * ifunit wants a null-terminated name.  It may not be null-terminated
1225          * in the sockaddr.  We don't want to change the caller's sockaddr,
1226          * and there might not be room to put the trailing null anyway, so we
1227          * make a local copy that we know we can null terminate safely.
1228          */
1229
1230         bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1231         ifname[sdl->sdl_nlen] = '\0';
1232         return ifunit(ifname);
1233 }
1234
1235
1236 /*
1237  * Interface ioctls.
1238  */
1239 int
1240 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1241 {
1242         struct ifnet *ifp;
1243         struct ifreq *ifr;
1244         struct ifstat *ifs;
1245         int error;
1246         short oif_flags;
1247         int new_flags;
1248         size_t namelen, onamelen;
1249         char new_name[IFNAMSIZ];
1250         struct ifaddr *ifa;
1251         struct sockaddr_dl *sdl;
1252
1253         switch (cmd) {
1254
1255         case SIOCGIFCONF:
1256         case OSIOCGIFCONF:
1257                 return (ifconf(cmd, data, cred));
1258         }
1259         ifr = (struct ifreq *)data;
1260
1261         switch (cmd) {
1262         case SIOCIFCREATE:
1263         case SIOCIFDESTROY:
1264                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1265                         return (error);
1266                 return ((cmd == SIOCIFCREATE) ?
1267                         if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name)) :
1268                         if_clone_destroy(ifr->ifr_name));
1269
1270         case SIOCIFGCLONERS:
1271                 return (if_clone_list((struct if_clonereq *)data));
1272         }
1273
1274         ifp = ifunit(ifr->ifr_name);
1275         if (ifp == 0)
1276                 return (ENXIO);
1277         switch (cmd) {
1278
1279         case SIOCGIFINDEX:
1280                 ifr->ifr_index = ifp->if_index;
1281                 break;
1282
1283         case SIOCGIFFLAGS:
1284                 ifr->ifr_flags = ifp->if_flags;
1285                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1286                 break;
1287
1288         case SIOCGIFCAP:
1289                 ifr->ifr_reqcap = ifp->if_capabilities;
1290                 ifr->ifr_curcap = ifp->if_capenable;
1291                 break;
1292
1293         case SIOCGIFMETRIC:
1294                 ifr->ifr_metric = ifp->if_metric;
1295                 break;
1296
1297         case SIOCGIFMTU:
1298                 ifr->ifr_mtu = ifp->if_mtu;
1299                 break;
1300
1301         case SIOCGIFPHYS:
1302                 ifr->ifr_phys = ifp->if_physical;
1303                 break;
1304
1305         case SIOCGIFPOLLCPU:
1306 #ifdef DEVICE_POLLING
1307                 ifr->ifr_pollcpu = ifp->if_poll_cpuid;
1308 #else
1309                 ifr->ifr_pollcpu = -1;
1310 #endif
1311                 break;
1312
1313         case SIOCSIFPOLLCPU:
1314 #ifdef DEVICE_POLLING
1315                 if ((ifp->if_flags & IFF_POLLING) == 0)
1316                         ether_pollcpu_register(ifp, ifr->ifr_pollcpu);
1317 #endif
1318                 break;
1319
1320         case SIOCSIFFLAGS:
1321                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1322                 if (error)
1323                         return (error);
1324                 new_flags = (ifr->ifr_flags & 0xffff) |
1325                     (ifr->ifr_flagshigh << 16);
1326                 if (ifp->if_flags & IFF_SMART) {
1327                         /* Smart drivers twiddle their own routes */
1328                 } else if (ifp->if_flags & IFF_UP &&
1329                     (new_flags & IFF_UP) == 0) {
1330                         crit_enter();
1331                         if_down(ifp);
1332                         crit_exit();
1333                 } else if (new_flags & IFF_UP &&
1334                     (ifp->if_flags & IFF_UP) == 0) {
1335                         crit_enter();
1336                         if_up(ifp);
1337                         crit_exit();
1338                 }
1339
1340 #ifdef DEVICE_POLLING
1341                 if ((new_flags ^ ifp->if_flags) & IFF_POLLING) {
1342                         if (new_flags & IFF_POLLING) {
1343                                 ether_poll_register(ifp);
1344                         } else {
1345                                 ether_poll_deregister(ifp);
1346                         }
1347                 }
1348 #endif
1349 #ifdef IFPOLL_ENABLE
1350                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1351                         if (new_flags & IFF_NPOLLING)
1352                                 ifpoll_register(ifp);
1353                         else
1354                                 ifpoll_deregister(ifp);
1355                 }
1356 #endif
1357
1358                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1359                         (new_flags &~ IFF_CANTCHANGE);
1360                 if (new_flags & IFF_PPROMISC) {
1361                         /* Permanently promiscuous mode requested */
1362                         ifp->if_flags |= IFF_PROMISC;
1363                 } else if (ifp->if_pcount == 0) {
1364                         ifp->if_flags &= ~IFF_PROMISC;
1365                 }
1366                 if (ifp->if_ioctl) {
1367                         ifnet_serialize_all(ifp);
1368                         ifp->if_ioctl(ifp, cmd, data, cred);
1369                         ifnet_deserialize_all(ifp);
1370                 }
1371                 getmicrotime(&ifp->if_lastchange);
1372                 break;
1373
1374         case SIOCSIFCAP:
1375                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1376                 if (error)
1377                         return (error);
1378                 if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1379                         return (EINVAL);
1380                 ifnet_serialize_all(ifp);
1381                 ifp->if_ioctl(ifp, cmd, data, cred);
1382                 ifnet_deserialize_all(ifp);
1383                 break;
1384
1385         case SIOCSIFNAME:
1386                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1387                 if (error != 0)
1388                         return (error);
1389                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1390                 if (error != 0)
1391                         return (error);
1392                 if (new_name[0] == '\0')
1393                         return (EINVAL);
1394                 if (ifunit(new_name) != NULL)
1395                         return (EEXIST);
1396
1397                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1398
1399                 /* Announce the departure of the interface. */
1400                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1401
1402                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1403                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1404                 /* XXX IFA_LOCK(ifa); */
1405                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1406                 namelen = strlen(new_name);
1407                 onamelen = sdl->sdl_nlen;
1408                 /*
1409                  * Move the address if needed.  This is safe because we
1410                  * allocate space for a name of length IFNAMSIZ when we
1411                  * create this in if_attach().
1412                  */
1413                 if (namelen != onamelen) {
1414                         bcopy(sdl->sdl_data + onamelen,
1415                             sdl->sdl_data + namelen, sdl->sdl_alen);
1416                 }
1417                 bcopy(new_name, sdl->sdl_data, namelen);
1418                 sdl->sdl_nlen = namelen;
1419                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1420                 bzero(sdl->sdl_data, onamelen);
1421                 while (namelen != 0)
1422                         sdl->sdl_data[--namelen] = 0xff;
1423                 /* XXX IFA_UNLOCK(ifa) */
1424
1425                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1426
1427                 /* Announce the return of the interface. */
1428                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1429                 break;
1430
1431         case SIOCSIFMETRIC:
1432                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1433                 if (error)
1434                         return (error);
1435                 ifp->if_metric = ifr->ifr_metric;
1436                 getmicrotime(&ifp->if_lastchange);
1437                 break;
1438
1439         case SIOCSIFPHYS:
1440                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1441                 if (error)
1442                         return error;
1443                 if (!ifp->if_ioctl)
1444                         return EOPNOTSUPP;
1445                 ifnet_serialize_all(ifp);
1446                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1447                 ifnet_deserialize_all(ifp);
1448                 if (error == 0)
1449                         getmicrotime(&ifp->if_lastchange);
1450                 return (error);
1451
1452         case SIOCSIFMTU:
1453         {
1454                 u_long oldmtu = ifp->if_mtu;
1455
1456                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1457                 if (error)
1458                         return (error);
1459                 if (ifp->if_ioctl == NULL)
1460                         return (EOPNOTSUPP);
1461                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1462                         return (EINVAL);
1463                 ifnet_serialize_all(ifp);
1464                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1465                 ifnet_deserialize_all(ifp);
1466                 if (error == 0) {
1467                         getmicrotime(&ifp->if_lastchange);
1468                         rt_ifmsg(ifp);
1469                 }
1470                 /*
1471                  * If the link MTU changed, do network layer specific procedure.
1472                  */
1473                 if (ifp->if_mtu != oldmtu) {
1474 #ifdef INET6
1475                         nd6_setmtu(ifp);
1476 #endif
1477                 }
1478                 return (error);
1479         }
1480
1481         case SIOCADDMULTI:
1482         case SIOCDELMULTI:
1483                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1484                 if (error)
1485                         return (error);
1486
1487                 /* Don't allow group membership on non-multicast interfaces. */
1488                 if ((ifp->if_flags & IFF_MULTICAST) == 0)
1489                         return EOPNOTSUPP;
1490
1491                 /* Don't let users screw up protocols' entries. */
1492                 if (ifr->ifr_addr.sa_family != AF_LINK)
1493                         return EINVAL;
1494
1495                 if (cmd == SIOCADDMULTI) {
1496                         struct ifmultiaddr *ifma;
1497                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1498                 } else {
1499                         error = if_delmulti(ifp, &ifr->ifr_addr);
1500                 }
1501                 if (error == 0)
1502                         getmicrotime(&ifp->if_lastchange);
1503                 return error;
1504
1505         case SIOCSIFPHYADDR:
1506         case SIOCDIFPHYADDR:
1507 #ifdef INET6
1508         case SIOCSIFPHYADDR_IN6:
1509 #endif
1510         case SIOCSLIFPHYADDR:
1511         case SIOCSIFMEDIA:
1512         case SIOCSIFGENERIC:
1513                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1514                 if (error)
1515                         return (error);
1516                 if (ifp->if_ioctl == 0)
1517                         return (EOPNOTSUPP);
1518                 ifnet_serialize_all(ifp);
1519                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1520                 ifnet_deserialize_all(ifp);
1521                 if (error == 0)
1522                         getmicrotime(&ifp->if_lastchange);
1523                 return error;
1524
1525         case SIOCGIFSTATUS:
1526                 ifs = (struct ifstat *)data;
1527                 ifs->ascii[0] = '\0';
1528
1529         case SIOCGIFPSRCADDR:
1530         case SIOCGIFPDSTADDR:
1531         case SIOCGLIFPHYADDR:
1532         case SIOCGIFMEDIA:
1533         case SIOCGIFGENERIC:
1534                 if (ifp->if_ioctl == NULL)
1535                         return (EOPNOTSUPP);
1536                 ifnet_serialize_all(ifp);
1537                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1538                 ifnet_deserialize_all(ifp);
1539                 return (error);
1540
1541         case SIOCSIFLLADDR:
1542                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1543                 if (error)
1544                         return (error);
1545                 return if_setlladdr(ifp,
1546                     ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1547
1548         default:
1549                 oif_flags = ifp->if_flags;
1550                 if (so->so_proto == 0)
1551                         return (EOPNOTSUPP);
1552 #ifndef COMPAT_43
1553                 error = so_pru_control(so, cmd, data, ifp);
1554 #else
1555             {
1556                 int ocmd = cmd;
1557
1558                 switch (cmd) {
1559
1560                 case SIOCSIFDSTADDR:
1561                 case SIOCSIFADDR:
1562                 case SIOCSIFBRDADDR:
1563                 case SIOCSIFNETMASK:
1564 #if BYTE_ORDER != BIG_ENDIAN
1565                         if (ifr->ifr_addr.sa_family == 0 &&
1566                             ifr->ifr_addr.sa_len < 16) {
1567                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1568                                 ifr->ifr_addr.sa_len = 16;
1569                         }
1570 #else
1571                         if (ifr->ifr_addr.sa_len == 0)
1572                                 ifr->ifr_addr.sa_len = 16;
1573 #endif
1574                         break;
1575
1576                 case OSIOCGIFADDR:
1577                         cmd = SIOCGIFADDR;
1578                         break;
1579
1580                 case OSIOCGIFDSTADDR:
1581                         cmd = SIOCGIFDSTADDR;
1582                         break;
1583
1584                 case OSIOCGIFBRDADDR:
1585                         cmd = SIOCGIFBRDADDR;
1586                         break;
1587
1588                 case OSIOCGIFNETMASK:
1589                         cmd = SIOCGIFNETMASK;
1590                 }
1591                 error =  so_pru_control(so, cmd, data, ifp);
1592                 switch (ocmd) {
1593
1594                 case OSIOCGIFADDR:
1595                 case OSIOCGIFDSTADDR:
1596                 case OSIOCGIFBRDADDR:
1597                 case OSIOCGIFNETMASK:
1598                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1599
1600                 }
1601             }
1602 #endif /* COMPAT_43 */
1603
1604                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1605 #ifdef INET6
1606                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
1607                         if (ifp->if_flags & IFF_UP) {
1608                                 crit_enter();
1609                                 in6_if_up(ifp);
1610                                 crit_exit();
1611                         }
1612 #endif
1613                 }
1614                 return (error);
1615
1616         }
1617         return (0);
1618 }
1619
1620 /*
1621  * Set/clear promiscuous mode on interface ifp based on the truth value
1622  * of pswitch.  The calls are reference counted so that only the first
1623  * "on" request actually has an effect, as does the final "off" request.
1624  * Results are undefined if the "off" and "on" requests are not matched.
1625  */
1626 int
1627 ifpromisc(struct ifnet *ifp, int pswitch)
1628 {
1629         struct ifreq ifr;
1630         int error;
1631         int oldflags;
1632
1633         oldflags = ifp->if_flags;
1634         if (ifp->if_flags & IFF_PPROMISC) {
1635                 /* Do nothing if device is in permanently promiscuous mode */
1636                 ifp->if_pcount += pswitch ? 1 : -1;
1637                 return (0);
1638         }
1639         if (pswitch) {
1640                 /*
1641                  * If the device is not configured up, we cannot put it in
1642                  * promiscuous mode.
1643                  */
1644                 if ((ifp->if_flags & IFF_UP) == 0)
1645                         return (ENETDOWN);
1646                 if (ifp->if_pcount++ != 0)
1647                         return (0);
1648                 ifp->if_flags |= IFF_PROMISC;
1649                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1650                     ifp->if_xname);
1651         } else {
1652                 if (--ifp->if_pcount > 0)
1653                         return (0);
1654                 ifp->if_flags &= ~IFF_PROMISC;
1655                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
1656                     ifp->if_xname);
1657         }
1658         ifr.ifr_flags = ifp->if_flags;
1659         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1660         ifnet_serialize_all(ifp);
1661         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
1662         ifnet_deserialize_all(ifp);
1663         if (error == 0)
1664                 rt_ifmsg(ifp);
1665         else
1666                 ifp->if_flags = oldflags;
1667         return error;
1668 }
1669
1670 /*
1671  * Return interface configuration
1672  * of system.  List may be used
1673  * in later ioctl's (above) to get
1674  * other information.
1675  */
1676 static int
1677 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
1678 {
1679         struct ifconf *ifc = (struct ifconf *)data;
1680         struct ifnet *ifp;
1681         struct sockaddr *sa;
1682         struct ifreq ifr, *ifrp;
1683         int space = ifc->ifc_len, error = 0;
1684
1685         ifrp = ifc->ifc_req;
1686         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1687                 struct ifaddr_container *ifac;
1688                 int addrs;
1689
1690                 if (space <= sizeof ifr)
1691                         break;
1692
1693                 /*
1694                  * Zero the stack declared structure first to prevent
1695                  * memory disclosure.
1696                  */
1697                 bzero(&ifr, sizeof(ifr));
1698                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
1699                     >= sizeof(ifr.ifr_name)) {
1700                         error = ENAMETOOLONG;
1701                         break;
1702                 }
1703
1704                 addrs = 0;
1705                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1706                         struct ifaddr *ifa = ifac->ifa;
1707
1708                         if (space <= sizeof ifr)
1709                                 break;
1710                         sa = ifa->ifa_addr;
1711                         if (cred->cr_prison &&
1712                             prison_if(cred, sa))
1713                                 continue;
1714                         addrs++;
1715 #ifdef COMPAT_43
1716                         if (cmd == OSIOCGIFCONF) {
1717                                 struct osockaddr *osa =
1718                                          (struct osockaddr *)&ifr.ifr_addr;
1719                                 ifr.ifr_addr = *sa;
1720                                 osa->sa_family = sa->sa_family;
1721                                 error = copyout(&ifr, ifrp, sizeof ifr);
1722                                 ifrp++;
1723                         } else
1724 #endif
1725                         if (sa->sa_len <= sizeof(*sa)) {
1726                                 ifr.ifr_addr = *sa;
1727                                 error = copyout(&ifr, ifrp, sizeof ifr);
1728                                 ifrp++;
1729                         } else {
1730                                 if (space < (sizeof ifr) + sa->sa_len -
1731                                             sizeof(*sa))
1732                                         break;
1733                                 space -= sa->sa_len - sizeof(*sa);
1734                                 error = copyout(&ifr, ifrp,
1735                                                 sizeof ifr.ifr_name);
1736                                 if (error == 0)
1737                                         error = copyout(sa, &ifrp->ifr_addr,
1738                                                         sa->sa_len);
1739                                 ifrp = (struct ifreq *)
1740                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
1741                         }
1742                         if (error)
1743                                 break;
1744                         space -= sizeof ifr;
1745                 }
1746                 if (error)
1747                         break;
1748                 if (!addrs) {
1749                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
1750                         error = copyout(&ifr, ifrp, sizeof ifr);
1751                         if (error)
1752                                 break;
1753                         space -= sizeof ifr;
1754                         ifrp++;
1755                 }
1756         }
1757         ifc->ifc_len -= space;
1758         return (error);
1759 }
1760
1761 /*
1762  * Just like if_promisc(), but for all-multicast-reception mode.
1763  */
1764 int
1765 if_allmulti(struct ifnet *ifp, int onswitch)
1766 {
1767         int error = 0;
1768         struct ifreq ifr;
1769
1770         crit_enter();
1771
1772         if (onswitch) {
1773                 if (ifp->if_amcount++ == 0) {
1774                         ifp->if_flags |= IFF_ALLMULTI;
1775                         ifr.ifr_flags = ifp->if_flags;
1776                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1777                         ifnet_serialize_all(ifp);
1778                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1779                                               NULL);
1780                         ifnet_deserialize_all(ifp);
1781                 }
1782         } else {
1783                 if (ifp->if_amcount > 1) {
1784                         ifp->if_amcount--;
1785                 } else {
1786                         ifp->if_amcount = 0;
1787                         ifp->if_flags &= ~IFF_ALLMULTI;
1788                         ifr.ifr_flags = ifp->if_flags;
1789                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1790                         ifnet_serialize_all(ifp);
1791                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1792                                               NULL);
1793                         ifnet_deserialize_all(ifp);
1794                 }
1795         }
1796
1797         crit_exit();
1798
1799         if (error == 0)
1800                 rt_ifmsg(ifp);
1801         return error;
1802 }
1803
1804 /*
1805  * Add a multicast listenership to the interface in question.
1806  * The link layer provides a routine which converts
1807  */
1808 int
1809 if_addmulti(
1810         struct ifnet *ifp,      /* interface to manipulate */
1811         struct sockaddr *sa,    /* address to add */
1812         struct ifmultiaddr **retifma)
1813 {
1814         struct sockaddr *llsa, *dupsa;
1815         int error;
1816         struct ifmultiaddr *ifma;
1817
1818         /*
1819          * If the matching multicast address already exists
1820          * then don't add a new one, just add a reference
1821          */
1822         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1823                 if (sa_equal(sa, ifma->ifma_addr)) {
1824                         ifma->ifma_refcount++;
1825                         if (retifma)
1826                                 *retifma = ifma;
1827                         return 0;
1828                 }
1829         }
1830
1831         /*
1832          * Give the link layer a chance to accept/reject it, and also
1833          * find out which AF_LINK address this maps to, if it isn't one
1834          * already.
1835          */
1836         if (ifp->if_resolvemulti) {
1837                 ifnet_serialize_all(ifp);
1838                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
1839                 ifnet_deserialize_all(ifp);
1840                 if (error) 
1841                         return error;
1842         } else {
1843                 llsa = 0;
1844         }
1845
1846         MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, M_WAITOK);
1847         MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, M_WAITOK);
1848         bcopy(sa, dupsa, sa->sa_len);
1849
1850         ifma->ifma_addr = dupsa;
1851         ifma->ifma_lladdr = llsa;
1852         ifma->ifma_ifp = ifp;
1853         ifma->ifma_refcount = 1;
1854         ifma->ifma_protospec = 0;
1855         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
1856
1857         /*
1858          * Some network interfaces can scan the address list at
1859          * interrupt time; lock them out.
1860          */
1861         crit_enter();
1862         LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1863         crit_exit();
1864         *retifma = ifma;
1865
1866         if (llsa != 0) {
1867                 LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1868                         if (sa_equal(ifma->ifma_addr, llsa))
1869                                 break;
1870                 }
1871                 if (ifma) {
1872                         ifma->ifma_refcount++;
1873                 } else {
1874                         MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma,
1875                                M_IFMADDR, M_WAITOK);
1876                         MALLOC(dupsa, struct sockaddr *, llsa->sa_len,
1877                                M_IFMADDR, M_WAITOK);
1878                         bcopy(llsa, dupsa, llsa->sa_len);
1879                         ifma->ifma_addr = dupsa;
1880                         ifma->ifma_ifp = ifp;
1881                         ifma->ifma_refcount = 1;
1882                         crit_enter();
1883                         LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1884                         crit_exit();
1885                 }
1886         }
1887         /*
1888          * We are certain we have added something, so call down to the
1889          * interface to let them know about it.
1890          */
1891         crit_enter();
1892         ifnet_serialize_all(ifp);
1893         ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
1894         ifnet_deserialize_all(ifp);
1895         crit_exit();
1896
1897         return 0;
1898 }
1899
1900 /*
1901  * Remove a reference to a multicast address on this interface.  Yell
1902  * if the request does not match an existing membership.
1903  */
1904 int
1905 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
1906 {
1907         struct ifmultiaddr *ifma;
1908
1909         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1910                 if (sa_equal(sa, ifma->ifma_addr))
1911                         break;
1912         if (ifma == 0)
1913                 return ENOENT;
1914
1915         if (ifma->ifma_refcount > 1) {
1916                 ifma->ifma_refcount--;
1917                 return 0;
1918         }
1919
1920         rt_newmaddrmsg(RTM_DELMADDR, ifma);
1921         sa = ifma->ifma_lladdr;
1922         crit_enter();
1923         LIST_REMOVE(ifma, ifma_link);
1924         /*
1925          * Make sure the interface driver is notified
1926          * in the case of a link layer mcast group being left.
1927          */
1928         if (ifma->ifma_addr->sa_family == AF_LINK && sa == 0) {
1929                 ifnet_serialize_all(ifp);
1930                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
1931                 ifnet_deserialize_all(ifp);
1932         }
1933         crit_exit();
1934         kfree(ifma->ifma_addr, M_IFMADDR);
1935         kfree(ifma, M_IFMADDR);
1936         if (sa == 0)
1937                 return 0;
1938
1939         /*
1940          * Now look for the link-layer address which corresponds to
1941          * this network address.  It had been squirreled away in
1942          * ifma->ifma_lladdr for this purpose (so we don't have
1943          * to call ifp->if_resolvemulti() again), and we saved that
1944          * value in sa above.  If some nasty deleted the
1945          * link-layer address out from underneath us, we can deal because
1946          * the address we stored was is not the same as the one which was
1947          * in the record for the link-layer address.  (So we don't complain
1948          * in that case.)
1949          */
1950         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1951                 if (sa_equal(sa, ifma->ifma_addr))
1952                         break;
1953         if (ifma == 0)
1954                 return 0;
1955
1956         if (ifma->ifma_refcount > 1) {
1957                 ifma->ifma_refcount--;
1958                 return 0;
1959         }
1960
1961         crit_enter();
1962         ifnet_serialize_all(ifp);
1963         LIST_REMOVE(ifma, ifma_link);
1964         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
1965         ifnet_deserialize_all(ifp);
1966         crit_exit();
1967         kfree(ifma->ifma_addr, M_IFMADDR);
1968         kfree(sa, M_IFMADDR);
1969         kfree(ifma, M_IFMADDR);
1970
1971         return 0;
1972 }
1973
1974 /*
1975  * Set the link layer address on an interface.
1976  *
1977  * At this time we only support certain types of interfaces,
1978  * and we don't allow the length of the address to change.
1979  */
1980 int
1981 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
1982 {
1983         struct sockaddr_dl *sdl;
1984         struct ifreq ifr;
1985
1986         sdl = IF_LLSOCKADDR(ifp);
1987         if (sdl == NULL)
1988                 return (EINVAL);
1989         if (len != sdl->sdl_alen)       /* don't allow length to change */
1990                 return (EINVAL);
1991         switch (ifp->if_type) {
1992         case IFT_ETHER:                 /* these types use struct arpcom */
1993         case IFT_XETHER:
1994         case IFT_L2VLAN:
1995                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
1996                 bcopy(lladdr, LLADDR(sdl), len);
1997                 break;
1998         default:
1999                 return (ENODEV);
2000         }
2001         /*
2002          * If the interface is already up, we need
2003          * to re-init it in order to reprogram its
2004          * address filter.
2005          */
2006         ifnet_serialize_all(ifp);
2007         if ((ifp->if_flags & IFF_UP) != 0) {
2008                 struct ifaddr_container *ifac;
2009
2010                 ifp->if_flags &= ~IFF_UP;
2011                 ifr.ifr_flags = ifp->if_flags;
2012                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2013                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2014                               NULL);
2015                 ifp->if_flags |= IFF_UP;
2016                 ifr.ifr_flags = ifp->if_flags;
2017                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2018                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2019                                  NULL);
2020 #ifdef INET
2021                 /*
2022                  * Also send gratuitous ARPs to notify other nodes about
2023                  * the address change.
2024                  */
2025                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2026                         struct ifaddr *ifa = ifac->ifa;
2027
2028                         if (ifa->ifa_addr != NULL &&
2029                             ifa->ifa_addr->sa_family == AF_INET)
2030                                 arp_ifinit(ifp, ifa);
2031                 }
2032 #endif
2033         }
2034         ifnet_deserialize_all(ifp);
2035         return (0);
2036 }
2037
2038 struct ifmultiaddr *
2039 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2040 {
2041         struct ifmultiaddr *ifma;
2042
2043         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2044                 if (sa_equal(ifma->ifma_addr, sa))
2045                         break;
2046
2047         return ifma;
2048 }
2049
2050 /*
2051  * This function locates the first real ethernet MAC from a network
2052  * card and loads it into node, returning 0 on success or ENOENT if
2053  * no suitable interfaces were found.  It is used by the uuid code to
2054  * generate a unique 6-byte number.
2055  */
2056 int
2057 if_getanyethermac(uint16_t *node, int minlen)
2058 {
2059         struct ifnet *ifp;
2060         struct sockaddr_dl *sdl;
2061
2062         TAILQ_FOREACH(ifp, &ifnet, if_link) {
2063                 if (ifp->if_type != IFT_ETHER)
2064                         continue;
2065                 sdl = IF_LLSOCKADDR(ifp);
2066                 if (sdl->sdl_alen < minlen)
2067                         continue;
2068                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2069                       minlen);
2070                 return(0);
2071         }
2072         return (ENOENT);
2073 }
2074
2075 /*
2076  * The name argument must be a pointer to storage which will last as
2077  * long as the interface does.  For physical devices, the result of
2078  * device_get_name(dev) is a good choice and for pseudo-devices a
2079  * static string works well.
2080  */
2081 void
2082 if_initname(struct ifnet *ifp, const char *name, int unit)
2083 {
2084         ifp->if_dname = name;
2085         ifp->if_dunit = unit;
2086         if (unit != IF_DUNIT_NONE)
2087                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2088         else
2089                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2090 }
2091
2092 int
2093 if_printf(struct ifnet *ifp, const char *fmt, ...)
2094 {
2095         __va_list ap;
2096         int retval;
2097
2098         retval = kprintf("%s: ", ifp->if_xname);
2099         __va_start(ap, fmt);
2100         retval += kvprintf(fmt, ap);
2101         __va_end(ap);
2102         return (retval);
2103 }
2104
2105 void
2106 ifq_set_classic(struct ifaltq *ifq)
2107 {
2108         ifq->altq_enqueue = ifq_classic_enqueue;
2109         ifq->altq_dequeue = ifq_classic_dequeue;
2110         ifq->altq_request = ifq_classic_request;
2111 }
2112
2113 int
2114 ifq_classic_enqueue(struct ifaltq *ifq, struct mbuf *m,
2115                     struct altq_pktattr *pa __unused)
2116 {
2117         logifq(enqueue, ifq);
2118         if (IF_QFULL(ifq)) {
2119                 m_freem(m);
2120                 return(ENOBUFS);
2121         } else {
2122                 IF_ENQUEUE(ifq, m);
2123                 return(0);
2124         }       
2125 }
2126
2127 struct mbuf *
2128 ifq_classic_dequeue(struct ifaltq *ifq, struct mbuf *mpolled, int op)
2129 {
2130         struct mbuf *m;
2131
2132         switch (op) {
2133         case ALTDQ_POLL:
2134                 IF_POLL(ifq, m);
2135                 break;
2136         case ALTDQ_REMOVE:
2137                 logifq(dequeue, ifq);
2138                 IF_DEQUEUE(ifq, m);
2139                 break;
2140         default:
2141                 panic("unsupported ALTQ dequeue op: %d", op);
2142         }
2143         KKASSERT(mpolled == NULL || mpolled == m);
2144         return(m);
2145 }
2146
2147 int
2148 ifq_classic_request(struct ifaltq *ifq, int req, void *arg)
2149 {
2150         switch (req) {
2151         case ALTRQ_PURGE:
2152                 IF_DRAIN(ifq);
2153                 break;
2154         default:
2155                 panic("unsupported ALTQ request: %d", req);
2156         }
2157         return(0);
2158 }
2159
2160 int
2161 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2162 {
2163         struct ifaltq *ifq = &ifp->if_snd;
2164         int running = 0, error, start = 0;
2165
2166         ASSERT_IFNET_NOT_SERIALIZED_TX(ifp);
2167
2168         ALTQ_LOCK(ifq);
2169         error = ifq_enqueue_locked(ifq, m, pa);
2170         if (error) {
2171                 ALTQ_UNLOCK(ifq);
2172                 return error;
2173         }
2174         if (!ifq->altq_started) {
2175                 /*
2176                  * Hold the interlock of ifnet.if_start
2177                  */
2178                 ifq->altq_started = 1;
2179                 start = 1;
2180         }
2181         ALTQ_UNLOCK(ifq);
2182
2183         ifp->if_obytes += m->m_pkthdr.len;
2184         if (m->m_flags & M_MCAST)
2185                 ifp->if_omcasts++;
2186
2187         if (!start) {
2188                 logifstart(avoid, ifp);
2189                 return 0;
2190         }
2191
2192         if (ifq_dispatch_schedonly) {
2193                 /*
2194                  * Always schedule ifnet.if_start on ifnet's CPU,
2195                  * short circuit the rest of this function.
2196                  */
2197                 logifstart(sched, ifp);
2198                 if_start_schedule(ifp);
2199                 return 0;
2200         }
2201
2202         /*
2203          * Try to do direct ifnet.if_start first, if there is
2204          * contention on ifnet's serializer, ifnet.if_start will
2205          * be scheduled on ifnet's CPU.
2206          */
2207         if (!ifnet_tryserialize_tx(ifp)) {
2208                 /*
2209                  * ifnet serializer contention happened,
2210                  * ifnet.if_start is scheduled on ifnet's
2211                  * CPU, and we keep going.
2212                  */
2213                 logifstart(contend_sched, ifp);
2214                 if_start_schedule(ifp);
2215                 return 0;
2216         }
2217
2218         if ((ifp->if_flags & IFF_OACTIVE) == 0) {
2219                 logifstart(run, ifp);
2220                 ifp->if_start(ifp);
2221                 if ((ifp->if_flags &
2222                      (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
2223                         running = 1;
2224         }
2225
2226         ifnet_deserialize_tx(ifp);
2227
2228         if (ifq_dispatch_schednochk || if_start_need_schedule(ifq, running)) {
2229                 /*
2230                  * More data need to be transmitted, ifnet.if_start is
2231                  * scheduled on ifnet's CPU, and we keep going.
2232                  * NOTE: ifnet.if_start interlock is not released.
2233                  */
2234                 logifstart(sched, ifp);
2235                 if_start_schedule(ifp);
2236         }
2237         return 0;
2238 }
2239
2240 void *
2241 ifa_create(int size, int flags)
2242 {
2243         struct ifaddr *ifa;
2244         int i;
2245
2246         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small\n"));
2247
2248         ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2249         if (ifa == NULL)
2250                 return NULL;
2251
2252         ifa->ifa_containers = kmalloc(ncpus * sizeof(struct ifaddr_container),
2253                                       M_IFADDR, M_WAITOK | M_ZERO);
2254         ifa->ifa_ncnt = ncpus;
2255         for (i = 0; i < ncpus; ++i) {
2256                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2257
2258                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2259                 ifac->ifa = ifa;
2260                 ifac->ifa_refcnt = 1;
2261         }
2262 #ifdef IFADDR_DEBUG
2263         kprintf("alloc ifa %p %d\n", ifa, size);
2264 #endif
2265         return ifa;
2266 }
2267
2268 void
2269 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2270 {
2271         struct ifaddr *ifa = ifac->ifa;
2272
2273         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2274         KKASSERT(ifac->ifa_refcnt == 0);
2275         KASSERT(ifac->ifa_listmask == 0,
2276                 ("ifa is still on %#x lists\n", ifac->ifa_listmask));
2277
2278         ifac->ifa_magic = IFA_CONTAINER_DEAD;
2279
2280 #ifdef IFADDR_DEBUG_VERBOSE
2281         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2282 #endif
2283
2284         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2285                 ("invalid # of ifac, %d\n", ifa->ifa_ncnt));
2286         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2287 #ifdef IFADDR_DEBUG
2288                 kprintf("free ifa %p\n", ifa);
2289 #endif
2290                 kfree(ifa->ifa_containers, M_IFADDR);
2291                 kfree(ifa, M_IFADDR);
2292         }
2293 }
2294
2295 static void
2296 ifa_iflink_dispatch(struct netmsg *nmsg)
2297 {
2298         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2299         struct ifaddr *ifa = msg->ifa;
2300         struct ifnet *ifp = msg->ifp;
2301         int cpu = mycpuid;
2302         struct ifaddr_container *ifac;
2303
2304         crit_enter();
2305
2306         ifac = &ifa->ifa_containers[cpu];
2307         ASSERT_IFAC_VALID(ifac);
2308         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2309                 ("ifaddr is on if_addrheads\n"));
2310
2311         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2312         if (msg->tail)
2313                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2314         else
2315                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2316
2317         crit_exit();
2318
2319         ifa_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
2320 }
2321
2322 void
2323 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2324 {
2325         struct netmsg_ifaddr msg;
2326
2327         netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2328                     ifa_iflink_dispatch);
2329         msg.ifa = ifa;
2330         msg.ifp = ifp;
2331         msg.tail = tail;
2332
2333         ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2334 }
2335
2336 static void
2337 ifa_ifunlink_dispatch(struct netmsg *nmsg)
2338 {
2339         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2340         struct ifaddr *ifa = msg->ifa;
2341         struct ifnet *ifp = msg->ifp;
2342         int cpu = mycpuid;
2343         struct ifaddr_container *ifac;
2344
2345         crit_enter();
2346
2347         ifac = &ifa->ifa_containers[cpu];
2348         ASSERT_IFAC_VALID(ifac);
2349         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2350                 ("ifaddr is not on if_addrhead\n"));
2351
2352         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2353         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2354
2355         crit_exit();
2356
2357         ifa_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
2358 }
2359
2360 void
2361 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2362 {
2363         struct netmsg_ifaddr msg;
2364
2365         netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2366                     ifa_ifunlink_dispatch);
2367         msg.ifa = ifa;
2368         msg.ifp = ifp;
2369
2370         ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2371 }
2372
2373 static void
2374 ifa_destroy_dispatch(struct netmsg *nmsg)
2375 {
2376         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2377
2378         IFAFREE(msg->ifa);
2379         ifa_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2380 }
2381
2382 void
2383 ifa_destroy(struct ifaddr *ifa)
2384 {
2385         struct netmsg_ifaddr msg;
2386
2387         netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2388                     ifa_destroy_dispatch);
2389         msg.ifa = ifa;
2390
2391         ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2392 }
2393
2394 struct lwkt_port *
2395 ifnet_portfn(int cpu)
2396 {
2397         return &ifnet_threads[cpu].td_msgport;
2398 }
2399
2400 void
2401 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2402 {
2403         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2404
2405         if (next_cpu < ncpus)
2406                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2407         else
2408                 lwkt_replymsg(lmsg, 0);
2409 }
2410
2411 int
2412 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
2413 {
2414         KKASSERT(cpu < ncpus);
2415         return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
2416 }
2417
2418 void
2419 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
2420 {
2421         KKASSERT(cpu < ncpus);
2422         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
2423 }
2424
2425 static void
2426 ifnetinit(void *dummy __unused)
2427 {
2428         int i;
2429
2430         for (i = 0; i < ncpus; ++i) {
2431                 struct thread *thr = &ifnet_threads[i];
2432
2433                 lwkt_create(netmsg_service_loop, &ifnet_mpsafe_thread, NULL,
2434                             thr, TDF_NETWORK | TDF_MPSAFE, i, "ifnet %d", i);
2435                 netmsg_service_port_init(&thr->td_msgport);
2436         }
2437 }