f04f771474d7e55ca57bd50bad0bd945d892d82b
[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
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/priv.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/socketops.h>
53 #include <sys/protosw.h>
54 #include <sys/kernel.h>
55 #include <sys/ktr.h>
56 #include <sys/sockio.h>
57 #include <sys/syslog.h>
58 #include <sys/sysctl.h>
59 #include <sys/domain.h>
60 #include <sys/thread.h>
61 #include <sys/thread2.h>
62 #include <sys/serialize.h>
63 #include <sys/msgport2.h>
64 #include <sys/bus.h>
65
66 #include <net/if.h>
67 #include <net/if_arp.h>
68 #include <net/if_dl.h>
69 #include <net/if_types.h>
70 #include <net/if_var.h>
71 #include <net/ifq_var.h>
72 #include <net/radix.h>
73 #include <net/route.h>
74 #include <net/if_clone.h>
75 #include <net/netisr.h>
76 #include <net/netmsg2.h>
77
78 #include <machine/atomic.h>
79 #include <machine/stdarg.h>
80 #include <machine/smp.h>
81
82 #if defined(INET) || defined(INET6)
83 /*XXX*/
84 #include <netinet/in.h>
85 #include <netinet/in_var.h>
86 #include <netinet/if_ether.h>
87 #ifdef INET6
88 #include <netinet6/in6_var.h>
89 #include <netinet6/in6_ifattach.h>
90 #endif
91 #endif
92
93 #if defined(COMPAT_43)
94 #include <emulation/43bsd/43bsd_socket.h>
95 #endif /* COMPAT_43 */
96
97 struct netmsg_ifaddr {
98         struct netmsg   netmsg;
99         struct ifaddr   *ifa;
100         struct ifnet    *ifp;
101         int             tail;
102 };
103
104 /*
105  * System initialization
106  */
107 static void     if_attachdomain(void *);
108 static void     if_attachdomain1(struct ifnet *);
109 static int      ifconf(u_long, caddr_t, struct ucred *);
110 static void     ifinit(void *);
111 static void     ifnetinit(void *);
112 static void     if_slowtimo(void *);
113 static void     link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
114 static int      if_rtdel(struct radix_node *, void *);
115
116 #ifdef INET6
117 /*
118  * XXX: declare here to avoid to include many inet6 related files..
119  * should be more generalized?
120  */
121 extern void     nd6_setmtu(struct ifnet *);
122 #endif
123
124 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
125 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
126
127 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
128 /* Must be after netisr_init */
129 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
130
131 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
132 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
133
134 int                     ifqmaxlen = IFQ_MAXLEN;
135 struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
136
137 /* In ifq_dispatch(), try to do direct ifnet.if_start first */
138 static int              ifq_dispatch_schedonly = 0;
139 SYSCTL_INT(_net_link_generic, OID_AUTO, ifq_dispatch_schedonly, CTLFLAG_RW,
140            &ifq_dispatch_schedonly, 0, "");
141
142 /* In ifq_dispatch(), schedule ifnet.if_start without checking ifnet.if_snd */
143 static int              ifq_dispatch_schednochk = 0;
144 SYSCTL_INT(_net_link_generic, OID_AUTO, ifq_dispatch_schednochk, CTLFLAG_RW,
145            &ifq_dispatch_schednochk, 0, "");
146
147 /* In if_devstart(), try to do direct ifnet.if_start first */
148 static int              if_devstart_schedonly = 0;
149 SYSCTL_INT(_net_link_generic, OID_AUTO, if_devstart_schedonly, CTLFLAG_RW,
150            &if_devstart_schedonly, 0, "");
151
152 /* In if_devstart(), schedule ifnet.if_start without checking ifnet.if_snd */
153 static int              if_devstart_schednochk = 0;
154 SYSCTL_INT(_net_link_generic, OID_AUTO, if_devstart_schednochk, CTLFLAG_RW,
155            &if_devstart_schednochk, 0, "");
156
157 #ifdef SMP
158 /* Schedule ifnet.if_start on the current CPU */
159 static int              if_start_oncpu_sched = 0;
160 SYSCTL_INT(_net_link_generic, OID_AUTO, if_start_oncpu_sched, CTLFLAG_RW,
161            &if_start_oncpu_sched, 0, "");
162 #endif
163
164 struct callout          if_slowtimo_timer;
165
166 int                     if_index = 0;
167 struct ifnet            **ifindex2ifnet = NULL;
168 static struct thread    ifnet_threads[MAXCPU];
169 static int              ifnet_mpsafe_thread = NETMSG_SERVICE_MPSAFE;
170
171 #define IFQ_KTR_STRING          "ifq=%p"
172 #define IFQ_KTR_ARG_SIZE        (sizeof(void *))
173 #ifndef KTR_IFQ
174 #define KTR_IFQ                 KTR_ALL
175 #endif
176 KTR_INFO_MASTER(ifq);
177 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARG_SIZE);
178 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARG_SIZE);
179 #define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)
180
181 #define IF_START_KTR_STRING     "ifp=%p"
182 #define IF_START_KTR_ARG_SIZE   (sizeof(void *))
183 #ifndef KTR_IF_START
184 #define KTR_IF_START            KTR_ALL
185 #endif
186 KTR_INFO_MASTER(if_start);
187 KTR_INFO(KTR_IF_START, if_start, run, 0,
188          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
189 KTR_INFO(KTR_IF_START, if_start, sched, 1,
190          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
191 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
192          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
193 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
194          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
195 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
196          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
197 #define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
198
199 /*
200  * Network interface utility routines.
201  *
202  * Routines with ifa_ifwith* names take sockaddr *'s as
203  * parameters.
204  */
205 /* ARGSUSED*/
206 void
207 ifinit(void *dummy)
208 {
209         struct ifnet *ifp;
210
211         callout_init(&if_slowtimo_timer);
212
213         crit_enter();
214         TAILQ_FOREACH(ifp, &ifnet, if_link) {
215                 if (ifp->if_snd.ifq_maxlen == 0) {
216                         if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
217                         ifp->if_snd.ifq_maxlen = ifqmaxlen;
218                 }
219         }
220         crit_exit();
221
222         if_slowtimo(0);
223 }
224
225 static int
226 if_start_cpuid(struct ifnet *ifp)
227 {
228         return ifp->if_cpuid;
229 }
230
231 #ifdef DEVICE_POLLING
232 static int
233 if_start_cpuid_poll(struct ifnet *ifp)
234 {
235         int poll_cpuid = ifp->if_poll_cpuid;
236
237         if (poll_cpuid >= 0)
238                 return poll_cpuid;
239         else
240                 return ifp->if_cpuid;
241 }
242 #endif
243
244 static void
245 if_start_ipifunc(void *arg)
246 {
247         struct ifnet *ifp = arg;
248         struct lwkt_msg *lmsg = &ifp->if_start_nmsg[mycpuid].nm_lmsg;
249
250         crit_enter();
251         if (lmsg->ms_flags & MSGF_DONE)
252                 lwkt_sendmsg(ifnet_portfn(mycpuid), lmsg);
253         crit_exit();
254 }
255
256 /*
257  * Schedule ifnet.if_start on ifnet's CPU
258  */
259 static void
260 if_start_schedule(struct ifnet *ifp)
261 {
262 #ifdef SMP
263         int cpu;
264
265         if (if_start_oncpu_sched)
266                 cpu = mycpuid;
267         else
268                 cpu = ifp->if_start_cpuid(ifp);
269
270         if (cpu != mycpuid)
271                 lwkt_send_ipiq(globaldata_find(cpu), if_start_ipifunc, ifp);
272         else
273 #endif
274         if_start_ipifunc(ifp);
275 }
276
277 /*
278  * NOTE:
279  * This function will release ifnet.if_start interlock,
280  * if ifnet.if_start does not need to be scheduled
281  */
282 static __inline int
283 if_start_need_schedule(struct ifaltq *ifq, int running)
284 {
285         if (!running || ifq_is_empty(ifq)
286 #ifdef ALTQ
287             || ifq->altq_tbr != NULL
288 #endif
289         ) {
290                 ALTQ_LOCK(ifq);
291                 /*
292                  * ifnet.if_start interlock is released, if:
293                  * 1) Hardware can not take any packets, due to
294                  *    o  interface is marked down
295                  *    o  hardware queue is full (IFF_OACTIVE)
296                  *    Under the second situation, hardware interrupt
297                  *    or polling(4) will call/schedule ifnet.if_start
298                  *    when hardware queue is ready
299                  * 2) There is not packet in the ifnet.if_snd.
300                  *    Further ifq_dispatch or ifq_handoff will call/
301                  *    schedule ifnet.if_start
302                  * 3) TBR is used and it does not allow further
303                  *    dequeueing.
304                  *    TBR callout will call ifnet.if_start
305                  */
306                 if (!running || !ifq_data_ready(ifq)) {
307                         ifq->altq_started = 0;
308                         ALTQ_UNLOCK(ifq);
309                         return 0;
310                 }
311                 ALTQ_UNLOCK(ifq);
312         }
313         return 1;
314 }
315
316 static void
317 if_start_dispatch(struct netmsg *nmsg)
318 {
319         struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
320         struct ifnet *ifp = lmsg->u.ms_resultp;
321         struct ifaltq *ifq = &ifp->if_snd;
322         int running = 0;
323
324         crit_enter();
325         lwkt_replymsg(lmsg, 0); /* reply ASAP */
326         crit_exit();
327
328 #ifdef SMP
329         if (!if_start_oncpu_sched && mycpuid != ifp->if_start_cpuid(ifp)) {
330                 /*
331                  * If the ifnet is still up, we need to
332                  * chase its CPU change.
333                  */
334                 if (ifp->if_flags & IFF_UP) {
335                         logifstart(chase_sched, ifp);
336                         if_start_schedule(ifp);
337                         return;
338                 } else {
339                         goto check;
340                 }
341         }
342 #endif
343
344         if (ifp->if_flags & IFF_UP) {
345                 ifnet_serialize_tx(ifp); /* XXX try? */
346                 if ((ifp->if_flags & IFF_OACTIVE) == 0) {
347                         logifstart(run, ifp);
348                         ifp->if_start(ifp);
349                         if ((ifp->if_flags &
350                         (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
351                                 running = 1;
352                 }
353                 ifnet_deserialize_tx(ifp);
354         }
355 #ifdef SMP
356 check:
357 #endif
358         if (if_start_need_schedule(ifq, running)) {
359                 crit_enter();
360                 if (lmsg->ms_flags & MSGF_DONE) { /* XXX necessary? */
361                         logifstart(sched, ifp);
362                         lwkt_sendmsg(ifnet_portfn(mycpuid), lmsg);
363                 }
364                 crit_exit();
365         }
366 }
367
368 /* Device driver ifnet.if_start helper function */
369 void
370 if_devstart(struct ifnet *ifp)
371 {
372         struct ifaltq *ifq = &ifp->if_snd;
373         int running = 0;
374
375         ASSERT_IFNET_SERIALIZED_TX(ifp);
376
377         ALTQ_LOCK(ifq);
378         if (ifq->altq_started || !ifq_data_ready(ifq)) {
379                 logifstart(avoid, ifp);
380                 ALTQ_UNLOCK(ifq);
381                 return;
382         }
383         ifq->altq_started = 1;
384         ALTQ_UNLOCK(ifq);
385
386         if (if_devstart_schedonly) {
387                 /*
388                  * Always schedule ifnet.if_start on ifnet's CPU,
389                  * short circuit the rest of this function.
390                  */
391                 logifstart(sched, ifp);
392                 if_start_schedule(ifp);
393                 return;
394         }
395
396         logifstart(run, ifp);
397         ifp->if_start(ifp);
398
399         if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
400                 running = 1;
401
402         if (if_devstart_schednochk || if_start_need_schedule(ifq, running)) {
403                 /*
404                  * More data need to be transmitted, ifnet.if_start is
405                  * scheduled on ifnet's CPU, and we keep going.
406                  * NOTE: ifnet.if_start interlock is not released.
407                  */
408                 logifstart(sched, ifp);
409                 if_start_schedule(ifp);
410         }
411 }
412
413 static void
414 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
415 {
416         lwkt_serialize_enter(ifp->if_serializer);
417 }
418
419 static void
420 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
421 {
422         lwkt_serialize_exit(ifp->if_serializer);
423 }
424
425 static int
426 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
427 {
428         return lwkt_serialize_try(ifp->if_serializer);
429 }
430
431 #ifdef INVARIANTS
432 static void
433 if_default_serialize_assert(struct ifnet *ifp,
434                             enum ifnet_serialize slz __unused,
435                             boolean_t serialized)
436 {
437         if (serialized)
438                 ASSERT_SERIALIZED(ifp->if_serializer);
439         else
440                 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
441 }
442 #endif
443
444 /*
445  * Attach an interface to the list of "active" interfaces.
446  *
447  * The serializer is optional.  If non-NULL access to the interface
448  * may be MPSAFE.
449  */
450 void
451 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
452 {
453         unsigned socksize, ifasize;
454         int namelen, masklen;
455         struct sockaddr_dl *sdl;
456         struct ifaddr *ifa;
457         struct ifaltq *ifq;
458         int i;
459
460         static int if_indexlim = 8;
461
462         if (ifp->if_serialize != NULL) {
463                 KASSERT(ifp->if_deserialize != NULL &&
464                         ifp->if_tryserialize != NULL &&
465                         ifp->if_serialize_assert != NULL,
466                         ("serialize functions are partially setup\n"));
467
468                 /*
469                  * If the device supplies serialize functions,
470                  * then clear if_serializer to catch any invalid
471                  * usage of this field.
472                  */
473                 KASSERT(serializer == NULL,
474                         ("both serialize functions and default serializer "
475                          "are supplied\n"));
476                 ifp->if_serializer = NULL;
477         } else {
478                 KASSERT(ifp->if_deserialize == NULL &&
479                         ifp->if_tryserialize == NULL &&
480                         ifp->if_serialize_assert == NULL,
481                         ("serialize functions are partially setup\n"));
482                 ifp->if_serialize = if_default_serialize;
483                 ifp->if_deserialize = if_default_deserialize;
484                 ifp->if_tryserialize = if_default_tryserialize;
485 #ifdef INVARIANTS
486                 ifp->if_serialize_assert = if_default_serialize_assert;
487 #endif
488
489                 /*
490                  * The serializer can be passed in from the device,
491                  * allowing the same serializer to be used for both
492                  * the interrupt interlock and the device queue.
493                  * If not specified, the netif structure will use an
494                  * embedded serializer.
495                  */
496                 if (serializer == NULL) {
497                         serializer = &ifp->if_default_serializer;
498                         lwkt_serialize_init(serializer);
499                 }
500                 ifp->if_serializer = serializer;
501         }
502
503         ifp->if_start_cpuid = if_start_cpuid;
504         ifp->if_cpuid = 0;
505
506 #ifdef DEVICE_POLLING
507         /* Device is not in polling mode by default */
508         ifp->if_poll_cpuid = -1;
509         if (ifp->if_poll != NULL)
510                 ifp->if_start_cpuid = if_start_cpuid_poll;
511 #endif
512
513         ifp->if_start_nmsg = kmalloc(ncpus * sizeof(struct netmsg),
514                                      M_LWKTMSG, M_WAITOK);
515         for (i = 0; i < ncpus; ++i) {
516                 netmsg_init(&ifp->if_start_nmsg[i], &netisr_adone_rport, 0,
517                             if_start_dispatch);
518                 ifp->if_start_nmsg[i].nm_lmsg.u.ms_resultp = ifp;
519         }
520
521         TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
522         ifp->if_index = ++if_index;
523
524         /*
525          * XXX -
526          * The old code would work if the interface passed a pre-existing
527          * chain of ifaddrs to this code.  We don't trust our callers to
528          * properly initialize the tailq, however, so we no longer allow
529          * this unlikely case.
530          */
531         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
532                                     M_IFADDR, M_WAITOK | M_ZERO);
533         for (i = 0; i < ncpus; ++i)
534                 TAILQ_INIT(&ifp->if_addrheads[i]);
535
536         TAILQ_INIT(&ifp->if_prefixhead);
537         LIST_INIT(&ifp->if_multiaddrs);
538         getmicrotime(&ifp->if_lastchange);
539         if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
540                 unsigned int n;
541                 struct ifnet **q;
542
543                 if_indexlim <<= 1;
544
545                 /* grow ifindex2ifnet */
546                 n = if_indexlim * sizeof(*q);
547                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
548                 if (ifindex2ifnet) {
549                         bcopy(ifindex2ifnet, q, n/2);
550                         kfree(ifindex2ifnet, M_IFADDR);
551                 }
552                 ifindex2ifnet = q;
553         }
554
555         ifindex2ifnet[if_index] = ifp;
556
557         /*
558          * create a Link Level name for this device
559          */
560         namelen = strlen(ifp->if_xname);
561 #define _offsetof(t, m) ((int)((caddr_t)&((t *)0)->m))
562         masklen = _offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
563         socksize = masklen + ifp->if_addrlen;
564 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
565         if (socksize < sizeof(*sdl))
566                 socksize = sizeof(*sdl);
567         socksize = ROUNDUP(socksize);
568         ifasize = sizeof(struct ifaddr) + 2 * socksize;
569         ifa = ifa_create(ifasize, M_WAITOK);
570         sdl = (struct sockaddr_dl *)(ifa + 1);
571         sdl->sdl_len = socksize;
572         sdl->sdl_family = AF_LINK;
573         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
574         sdl->sdl_nlen = namelen;
575         sdl->sdl_index = ifp->if_index;
576         sdl->sdl_type = ifp->if_type;
577         ifp->if_lladdr = ifa;
578         ifa->ifa_ifp = ifp;
579         ifa->ifa_rtrequest = link_rtrequest;
580         ifa->ifa_addr = (struct sockaddr *)sdl;
581         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
582         ifa->ifa_netmask = (struct sockaddr *)sdl;
583         sdl->sdl_len = masklen;
584         while (namelen != 0)
585                 sdl->sdl_data[--namelen] = 0xff;
586         ifa_iflink(ifa, ifp, 0 /* Insert head */);
587
588         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
589         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
590
591         ifq = &ifp->if_snd;
592         ifq->altq_type = 0;
593         ifq->altq_disc = NULL;
594         ifq->altq_flags &= ALTQF_CANTCHANGE;
595         ifq->altq_tbr = NULL;
596         ifq->altq_ifp = ifp;
597         ifq->altq_started = 0;
598         ifq->altq_prepended = NULL;
599         ALTQ_LOCK_INIT(ifq);
600         ifq_set_classic(ifq);
601
602         if (!SLIST_EMPTY(&domains))
603                 if_attachdomain1(ifp);
604
605         /* Announce the interface. */
606         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
607 }
608
609 static void
610 if_attachdomain(void *dummy)
611 {
612         struct ifnet *ifp;
613
614         crit_enter();
615         TAILQ_FOREACH(ifp, &ifnet, if_list)
616                 if_attachdomain1(ifp);
617         crit_exit();
618 }
619 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
620         if_attachdomain, NULL);
621
622 static void
623 if_attachdomain1(struct ifnet *ifp)
624 {
625         struct domain *dp;
626
627         crit_enter();
628
629         /* address family dependent data region */
630         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
631         SLIST_FOREACH(dp, &domains, dom_next)
632                 if (dp->dom_ifattach)
633                         ifp->if_afdata[dp->dom_family] =
634                                 (*dp->dom_ifattach)(ifp);
635         crit_exit();
636 }
637
638 /*
639  * Purge all addresses whose type is _not_ AF_LINK
640  */
641 void
642 if_purgeaddrs_nolink(struct ifnet *ifp)
643 {
644         struct ifaddr_container *ifac, *next;
645
646         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
647                               ifa_link, next) {
648                 struct ifaddr *ifa = ifac->ifa;
649
650                 /* Leave link ifaddr as it is */
651                 if (ifa->ifa_addr->sa_family == AF_LINK)
652                         continue;
653 #ifdef INET
654                 /* XXX: Ugly!! ad hoc just for INET */
655                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
656                         struct ifaliasreq ifr;
657 #ifdef IFADDR_DEBUG_VERBOSE
658                         int i;
659
660                         kprintf("purge in4 addr %p: ", ifa);
661                         for (i = 0; i < ncpus; ++i)
662                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
663                         kprintf("\n");
664 #endif
665
666                         bzero(&ifr, sizeof ifr);
667                         ifr.ifra_addr = *ifa->ifa_addr;
668                         if (ifa->ifa_dstaddr)
669                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
670                         if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
671                                        NULL) == 0)
672                                 continue;
673                 }
674 #endif /* INET */
675 #ifdef INET6
676                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
677 #ifdef IFADDR_DEBUG_VERBOSE
678                         int i;
679
680                         kprintf("purge in6 addr %p: ", ifa);
681                         for (i = 0; i < ncpus; ++i)
682                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
683                         kprintf("\n");
684 #endif
685
686                         in6_purgeaddr(ifa);
687                         /* ifp_addrhead is already updated */
688                         continue;
689                 }
690 #endif /* INET6 */
691                 ifa_ifunlink(ifa, ifp);
692                 ifa_destroy(ifa);
693         }
694 }
695
696 /*
697  * Detach an interface, removing it from the
698  * list of "active" interfaces.
699  */
700 void
701 if_detach(struct ifnet *ifp)
702 {
703         struct radix_node_head  *rnh;
704         int i;
705         int cpu, origcpu;
706         struct domain *dp;
707
708         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
709
710         /*
711          * Remove routes and flush queues.
712          */
713         crit_enter();
714 #ifdef DEVICE_POLLING
715         if (ifp->if_flags & IFF_POLLING)
716                 ether_poll_deregister(ifp);
717 #endif
718         if_down(ifp);
719
720         if (ifq_is_enabled(&ifp->if_snd))
721                 altq_disable(&ifp->if_snd);
722         if (ifq_is_attached(&ifp->if_snd))
723                 altq_detach(&ifp->if_snd);
724
725         /*
726          * Clean up all addresses.
727          */
728         ifp->if_lladdr = NULL;
729
730         if_purgeaddrs_nolink(ifp);
731         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
732                 struct ifaddr *ifa;
733
734                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
735                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
736                         ("non-link ifaddr is left on if_addrheads"));
737
738                 ifa_ifunlink(ifa, ifp);
739                 ifa_destroy(ifa);
740                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
741                         ("there are still ifaddrs left on if_addrheads"));
742         }
743
744 #ifdef INET
745         /*
746          * Remove all IPv4 kernel structures related to ifp.
747          */
748         in_ifdetach(ifp);
749 #endif
750
751 #ifdef INET6
752         /*
753          * Remove all IPv6 kernel structs related to ifp.  This should be done
754          * before removing routing entries below, since IPv6 interface direct
755          * routes are expected to be removed by the IPv6-specific kernel API.
756          * Otherwise, the kernel will detect some inconsistency and bark it.
757          */
758         in6_ifdetach(ifp);
759 #endif
760
761         /*
762          * Delete all remaining routes using this interface
763          * Unfortuneatly the only way to do this is to slog through
764          * the entire routing table looking for routes which point
765          * to this interface...oh well...
766          */
767         origcpu = mycpuid;
768         for (cpu = 0; cpu < ncpus2; cpu++) {
769                 lwkt_migratecpu(cpu);
770                 for (i = 1; i <= AF_MAX; i++) {
771                         if ((rnh = rt_tables[cpu][i]) == NULL)
772                                 continue;
773                         rnh->rnh_walktree(rnh, if_rtdel, ifp);
774                 }
775         }
776         lwkt_migratecpu(origcpu);
777
778         /* Announce that the interface is gone. */
779         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
780         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
781
782         SLIST_FOREACH(dp, &domains, dom_next)
783                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
784                         (*dp->dom_ifdetach)(ifp,
785                                 ifp->if_afdata[dp->dom_family]);
786
787         /*
788          * Remove interface from ifindex2ifp[] and maybe decrement if_index.
789          */
790         ifindex2ifnet[ifp->if_index] = NULL;
791         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
792                 if_index--;
793
794         TAILQ_REMOVE(&ifnet, ifp, if_link);
795         kfree(ifp->if_addrheads, M_IFADDR);
796         kfree(ifp->if_start_nmsg, M_LWKTMSG);
797         crit_exit();
798 }
799
800 /*
801  * Delete Routes for a Network Interface
802  *
803  * Called for each routing entry via the rnh->rnh_walktree() call above
804  * to delete all route entries referencing a detaching network interface.
805  *
806  * Arguments:
807  *      rn      pointer to node in the routing table
808  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
809  *
810  * Returns:
811  *      0       successful
812  *      errno   failed - reason indicated
813  *
814  */
815 static int
816 if_rtdel(struct radix_node *rn, void *arg)
817 {
818         struct rtentry  *rt = (struct rtentry *)rn;
819         struct ifnet    *ifp = arg;
820         int             err;
821
822         if (rt->rt_ifp == ifp) {
823
824                 /*
825                  * Protect (sorta) against walktree recursion problems
826                  * with cloned routes
827                  */
828                 if (!(rt->rt_flags & RTF_UP))
829                         return (0);
830
831                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
832                                 rt_mask(rt), rt->rt_flags,
833                                 NULL);
834                 if (err) {
835                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
836                 }
837         }
838
839         return (0);
840 }
841
842 /*
843  * Locate an interface based on a complete address.
844  */
845 struct ifaddr *
846 ifa_ifwithaddr(struct sockaddr *addr)
847 {
848         struct ifnet *ifp;
849
850         TAILQ_FOREACH(ifp, &ifnet, if_link) {
851                 struct ifaddr_container *ifac;
852
853                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
854                         struct ifaddr *ifa = ifac->ifa;
855
856                         if (ifa->ifa_addr->sa_family != addr->sa_family)
857                                 continue;
858                         if (sa_equal(addr, ifa->ifa_addr))
859                                 return (ifa);
860                         if ((ifp->if_flags & IFF_BROADCAST) &&
861                             ifa->ifa_broadaddr &&
862                             /* IPv6 doesn't have broadcast */
863                             ifa->ifa_broadaddr->sa_len != 0 &&
864                             sa_equal(ifa->ifa_broadaddr, addr))
865                                 return (ifa);
866                 }
867         }
868         return (NULL);
869 }
870 /*
871  * Locate the point to point interface with a given destination address.
872  */
873 struct ifaddr *
874 ifa_ifwithdstaddr(struct sockaddr *addr)
875 {
876         struct ifnet *ifp;
877
878         TAILQ_FOREACH(ifp, &ifnet, if_link) {
879                 struct ifaddr_container *ifac;
880
881                 if (!(ifp->if_flags & IFF_POINTOPOINT))
882                         continue;
883
884                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
885                         struct ifaddr *ifa = ifac->ifa;
886
887                         if (ifa->ifa_addr->sa_family != addr->sa_family)
888                                 continue;
889                         if (ifa->ifa_dstaddr &&
890                             sa_equal(addr, ifa->ifa_dstaddr))
891                                 return (ifa);
892                 }
893         }
894         return (NULL);
895 }
896
897 /*
898  * Find an interface on a specific network.  If many, choice
899  * is most specific found.
900  */
901 struct ifaddr *
902 ifa_ifwithnet(struct sockaddr *addr)
903 {
904         struct ifnet *ifp;
905         struct ifaddr *ifa_maybe = NULL;
906         u_int af = addr->sa_family;
907         char *addr_data = addr->sa_data, *cplim;
908
909         /*
910          * AF_LINK addresses can be looked up directly by their index number,
911          * so do that if we can.
912          */
913         if (af == AF_LINK) {
914                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
915
916                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
917                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
918         }
919
920         /*
921          * Scan though each interface, looking for ones that have
922          * addresses in this address family.
923          */
924         TAILQ_FOREACH(ifp, &ifnet, if_link) {
925                 struct ifaddr_container *ifac;
926
927                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
928                         struct ifaddr *ifa = ifac->ifa;
929                         char *cp, *cp2, *cp3;
930
931                         if (ifa->ifa_addr->sa_family != af)
932 next:                           continue;
933                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
934                                 /*
935                                  * This is a bit broken as it doesn't
936                                  * take into account that the remote end may
937                                  * be a single node in the network we are
938                                  * looking for.
939                                  * The trouble is that we don't know the
940                                  * netmask for the remote end.
941                                  */
942                                 if (ifa->ifa_dstaddr != NULL &&
943                                     sa_equal(addr, ifa->ifa_dstaddr))
944                                         return (ifa);
945                         } else {
946                                 /*
947                                  * if we have a special address handler,
948                                  * then use it instead of the generic one.
949                                  */
950                                 if (ifa->ifa_claim_addr) {
951                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
952                                                 return (ifa);
953                                         } else {
954                                                 continue;
955                                         }
956                                 }
957
958                                 /*
959                                  * Scan all the bits in the ifa's address.
960                                  * If a bit dissagrees with what we are
961                                  * looking for, mask it with the netmask
962                                  * to see if it really matters.
963                                  * (A byte at a time)
964                                  */
965                                 if (ifa->ifa_netmask == 0)
966                                         continue;
967                                 cp = addr_data;
968                                 cp2 = ifa->ifa_addr->sa_data;
969                                 cp3 = ifa->ifa_netmask->sa_data;
970                                 cplim = ifa->ifa_netmask->sa_len +
971                                         (char *)ifa->ifa_netmask;
972                                 while (cp3 < cplim)
973                                         if ((*cp++ ^ *cp2++) & *cp3++)
974                                                 goto next; /* next address! */
975                                 /*
976                                  * If the netmask of what we just found
977                                  * is more specific than what we had before
978                                  * (if we had one) then remember the new one
979                                  * before continuing to search
980                                  * for an even better one.
981                                  */
982                                 if (ifa_maybe == 0 ||
983                                     rn_refines((char *)ifa->ifa_netmask,
984                                                (char *)ifa_maybe->ifa_netmask))
985                                         ifa_maybe = ifa;
986                         }
987                 }
988         }
989         return (ifa_maybe);
990 }
991
992 /*
993  * Find an interface address specific to an interface best matching
994  * a given address.
995  */
996 struct ifaddr *
997 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
998 {
999         struct ifaddr_container *ifac;
1000         char *cp, *cp2, *cp3;
1001         char *cplim;
1002         struct ifaddr *ifa_maybe = 0;
1003         u_int af = addr->sa_family;
1004
1005         if (af >= AF_MAX)
1006                 return (0);
1007         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1008                 struct ifaddr *ifa = ifac->ifa;
1009
1010                 if (ifa->ifa_addr->sa_family != af)
1011                         continue;
1012                 if (ifa_maybe == 0)
1013                         ifa_maybe = ifa;
1014                 if (ifa->ifa_netmask == NULL) {
1015                         if (sa_equal(addr, ifa->ifa_addr) ||
1016                             (ifa->ifa_dstaddr != NULL &&
1017                              sa_equal(addr, ifa->ifa_dstaddr)))
1018                                 return (ifa);
1019                         continue;
1020                 }
1021                 if (ifp->if_flags & IFF_POINTOPOINT) {
1022                         if (sa_equal(addr, ifa->ifa_dstaddr))
1023                                 return (ifa);
1024                 } else {
1025                         cp = addr->sa_data;
1026                         cp2 = ifa->ifa_addr->sa_data;
1027                         cp3 = ifa->ifa_netmask->sa_data;
1028                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1029                         for (; cp3 < cplim; cp3++)
1030                                 if ((*cp++ ^ *cp2++) & *cp3)
1031                                         break;
1032                         if (cp3 == cplim)
1033                                 return (ifa);
1034                 }
1035         }
1036         return (ifa_maybe);
1037 }
1038
1039 /*
1040  * Default action when installing a route with a Link Level gateway.
1041  * Lookup an appropriate real ifa to point to.
1042  * This should be moved to /sys/net/link.c eventually.
1043  */
1044 static void
1045 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1046 {
1047         struct ifaddr *ifa;
1048         struct sockaddr *dst;
1049         struct ifnet *ifp;
1050
1051         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1052             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1053                 return;
1054         ifa = ifaof_ifpforaddr(dst, ifp);
1055         if (ifa != NULL) {
1056                 IFAFREE(rt->rt_ifa);
1057                 IFAREF(ifa);
1058                 rt->rt_ifa = ifa;
1059                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1060                         ifa->ifa_rtrequest(cmd, rt, info);
1061         }
1062 }
1063
1064 /*
1065  * Mark an interface down and notify protocols of
1066  * the transition.
1067  * NOTE: must be called at splnet or eqivalent.
1068  */
1069 void
1070 if_unroute(struct ifnet *ifp, int flag, int fam)
1071 {
1072         struct ifaddr_container *ifac;
1073
1074         ifp->if_flags &= ~flag;
1075         getmicrotime(&ifp->if_lastchange);
1076         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1077                 struct ifaddr *ifa = ifac->ifa;
1078
1079                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1080                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1081         }
1082         ifq_purge(&ifp->if_snd);
1083         rt_ifmsg(ifp);
1084 }
1085
1086 /*
1087  * Mark an interface up and notify protocols of
1088  * the transition.
1089  * NOTE: must be called at splnet or eqivalent.
1090  */
1091 void
1092 if_route(struct ifnet *ifp, int flag, int fam)
1093 {
1094         struct ifaddr_container *ifac;
1095
1096         ifq_purge(&ifp->if_snd);
1097         ifp->if_flags |= flag;
1098         getmicrotime(&ifp->if_lastchange);
1099         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1100                 struct ifaddr *ifa = ifac->ifa;
1101
1102                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1103                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1104         }
1105         rt_ifmsg(ifp);
1106 #ifdef INET6
1107         in6_if_up(ifp);
1108 #endif
1109 }
1110
1111 /*
1112  * Mark an interface down and notify protocols of the transition.  An
1113  * interface going down is also considered to be a synchronizing event.
1114  * We must ensure that all packet processing related to the interface
1115  * has completed before we return so e.g. the caller can free the ifnet
1116  * structure that the mbufs may be referencing.
1117  *
1118  * NOTE: must be called at splnet or eqivalent.
1119  */
1120 void
1121 if_down(struct ifnet *ifp)
1122 {
1123         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1124         netmsg_service_sync();
1125 }
1126
1127 /*
1128  * Mark an interface up and notify protocols of
1129  * the transition.
1130  * NOTE: must be called at splnet or eqivalent.
1131  */
1132 void
1133 if_up(struct ifnet *ifp)
1134 {
1135         if_route(ifp, IFF_UP, AF_UNSPEC);
1136 }
1137
1138 /*
1139  * Process a link state change.
1140  * NOTE: must be called at splsoftnet or equivalent.
1141  */
1142 void
1143 if_link_state_change(struct ifnet *ifp)
1144 {
1145         int link_state = ifp->if_link_state;
1146
1147         rt_ifmsg(ifp);
1148         devctl_notify("IFNET", ifp->if_xname,
1149             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1150 }
1151
1152 /*
1153  * Handle interface watchdog timer routines.  Called
1154  * from softclock, we decrement timers (if set) and
1155  * call the appropriate interface routine on expiration.
1156  */
1157 static void
1158 if_slowtimo(void *arg)
1159 {
1160         struct ifnet *ifp;
1161
1162         crit_enter();
1163
1164         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1165                 if (ifp->if_timer == 0 || --ifp->if_timer)
1166                         continue;
1167                 if (ifp->if_watchdog) {
1168                         if (ifnet_tryserialize_all(ifp)) {
1169                                 (*ifp->if_watchdog)(ifp);
1170                                 ifnet_deserialize_all(ifp);
1171                         } else {
1172                                 /* try again next timeout */
1173                                 ++ifp->if_timer;
1174                         }
1175                 }
1176         }
1177
1178         crit_exit();
1179
1180         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1181 }
1182
1183 /*
1184  * Map interface name to
1185  * interface structure pointer.
1186  */
1187 struct ifnet *
1188 ifunit(const char *name)
1189 {
1190         struct ifnet *ifp;
1191
1192         /*
1193          * Search all the interfaces for this name/number
1194          */
1195
1196         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1197                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1198                         break;
1199         }
1200         return (ifp);
1201 }
1202
1203
1204 /*
1205  * Map interface name in a sockaddr_dl to
1206  * interface structure pointer.
1207  */
1208 struct ifnet *
1209 if_withname(struct sockaddr *sa)
1210 {
1211         char ifname[IFNAMSIZ+1];
1212         struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1213
1214         if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1215              (sdl->sdl_nlen > IFNAMSIZ) )
1216                 return NULL;
1217
1218         /*
1219          * ifunit wants a null-terminated name.  It may not be null-terminated
1220          * in the sockaddr.  We don't want to change the caller's sockaddr,
1221          * and there might not be room to put the trailing null anyway, so we
1222          * make a local copy that we know we can null terminate safely.
1223          */
1224
1225         bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1226         ifname[sdl->sdl_nlen] = '\0';
1227         return ifunit(ifname);
1228 }
1229
1230
1231 /*
1232  * Interface ioctls.
1233  */
1234 int
1235 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1236 {
1237         struct ifnet *ifp;
1238         struct ifreq *ifr;
1239         struct ifstat *ifs;
1240         int error;
1241         short oif_flags;
1242         int new_flags;
1243         size_t namelen, onamelen;
1244         char new_name[IFNAMSIZ];
1245         struct ifaddr *ifa;
1246         struct sockaddr_dl *sdl;
1247
1248         switch (cmd) {
1249
1250         case SIOCGIFCONF:
1251         case OSIOCGIFCONF:
1252                 return (ifconf(cmd, data, cred));
1253         }
1254         ifr = (struct ifreq *)data;
1255
1256         switch (cmd) {
1257         case SIOCIFCREATE:
1258         case SIOCIFDESTROY:
1259                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1260                         return (error);
1261                 return ((cmd == SIOCIFCREATE) ?
1262                         if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name)) :
1263                         if_clone_destroy(ifr->ifr_name));
1264
1265         case SIOCIFGCLONERS:
1266                 return (if_clone_list((struct if_clonereq *)data));
1267         }
1268
1269         ifp = ifunit(ifr->ifr_name);
1270         if (ifp == 0)
1271                 return (ENXIO);
1272         switch (cmd) {
1273
1274         case SIOCGIFINDEX:
1275                 ifr->ifr_index = ifp->if_index;
1276                 break;
1277
1278         case SIOCGIFFLAGS:
1279                 ifr->ifr_flags = ifp->if_flags;
1280                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1281                 break;
1282
1283         case SIOCGIFCAP:
1284                 ifr->ifr_reqcap = ifp->if_capabilities;
1285                 ifr->ifr_curcap = ifp->if_capenable;
1286                 break;
1287
1288         case SIOCGIFMETRIC:
1289                 ifr->ifr_metric = ifp->if_metric;
1290                 break;
1291
1292         case SIOCGIFMTU:
1293                 ifr->ifr_mtu = ifp->if_mtu;
1294                 break;
1295
1296         case SIOCGIFPHYS:
1297                 ifr->ifr_phys = ifp->if_physical;
1298                 break;
1299
1300         case SIOCGIFPOLLCPU:
1301 #ifdef DEVICE_POLLING
1302                 ifr->ifr_pollcpu = ifp->if_poll_cpuid;
1303 #else
1304                 ifr->ifr_pollcpu = -1;
1305 #endif
1306                 break;
1307
1308         case SIOCSIFPOLLCPU:
1309 #ifdef DEVICE_POLLING
1310                 if ((ifp->if_flags & IFF_POLLING) == 0)
1311                         ether_pollcpu_register(ifp, ifr->ifr_pollcpu);
1312 #endif
1313                 break;
1314
1315         case SIOCSIFFLAGS:
1316                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1317                 if (error)
1318                         return (error);
1319                 new_flags = (ifr->ifr_flags & 0xffff) |
1320                     (ifr->ifr_flagshigh << 16);
1321                 if (ifp->if_flags & IFF_SMART) {
1322                         /* Smart drivers twiddle their own routes */
1323                 } else if (ifp->if_flags & IFF_UP &&
1324                     (new_flags & IFF_UP) == 0) {
1325                         crit_enter();
1326                         if_down(ifp);
1327                         crit_exit();
1328                 } else if (new_flags & IFF_UP &&
1329                     (ifp->if_flags & IFF_UP) == 0) {
1330                         crit_enter();
1331                         if_up(ifp);
1332                         crit_exit();
1333                 }
1334
1335 #ifdef DEVICE_POLLING
1336                 if ((new_flags ^ ifp->if_flags) & IFF_POLLING) {
1337                         if (new_flags & IFF_POLLING) {
1338                                 ether_poll_register(ifp);
1339                         } else {
1340                                 ether_poll_deregister(ifp);
1341                         }
1342                 }
1343 #endif
1344
1345                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1346                         (new_flags &~ IFF_CANTCHANGE);
1347                 if (new_flags & IFF_PPROMISC) {
1348                         /* Permanently promiscuous mode requested */
1349                         ifp->if_flags |= IFF_PROMISC;
1350                 } else if (ifp->if_pcount == 0) {
1351                         ifp->if_flags &= ~IFF_PROMISC;
1352                 }
1353                 if (ifp->if_ioctl) {
1354                         ifnet_serialize_all(ifp);
1355                         ifp->if_ioctl(ifp, cmd, data, cred);
1356                         ifnet_deserialize_all(ifp);
1357                 }
1358                 getmicrotime(&ifp->if_lastchange);
1359                 break;
1360
1361         case SIOCSIFCAP:
1362                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1363                 if (error)
1364                         return (error);
1365                 if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1366                         return (EINVAL);
1367                 ifnet_serialize_all(ifp);
1368                 ifp->if_ioctl(ifp, cmd, data, cred);
1369                 ifnet_deserialize_all(ifp);
1370                 break;
1371
1372         case SIOCSIFNAME:
1373                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1374                 if (error != 0)
1375                         return (error);
1376                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1377                 if (error != 0)
1378                         return (error);
1379                 if (new_name[0] == '\0')
1380                         return (EINVAL);
1381                 if (ifunit(new_name) != NULL)
1382                         return (EEXIST);
1383
1384                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1385
1386                 /* Announce the departure of the interface. */
1387                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1388
1389                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1390                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1391                 /* XXX IFA_LOCK(ifa); */
1392                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1393                 namelen = strlen(new_name);
1394                 onamelen = sdl->sdl_nlen;
1395                 /*
1396                  * Move the address if needed.  This is safe because we
1397                  * allocate space for a name of length IFNAMSIZ when we
1398                  * create this in if_attach().
1399                  */
1400                 if (namelen != onamelen) {
1401                         bcopy(sdl->sdl_data + onamelen,
1402                             sdl->sdl_data + namelen, sdl->sdl_alen);
1403                 }
1404                 bcopy(new_name, sdl->sdl_data, namelen);
1405                 sdl->sdl_nlen = namelen;
1406                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1407                 bzero(sdl->sdl_data, onamelen);
1408                 while (namelen != 0)
1409                         sdl->sdl_data[--namelen] = 0xff;
1410                 /* XXX IFA_UNLOCK(ifa) */
1411
1412                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1413
1414                 /* Announce the return of the interface. */
1415                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1416                 break;
1417
1418         case SIOCSIFMETRIC:
1419                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1420                 if (error)
1421                         return (error);
1422                 ifp->if_metric = ifr->ifr_metric;
1423                 getmicrotime(&ifp->if_lastchange);
1424                 break;
1425
1426         case SIOCSIFPHYS:
1427                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1428                 if (error)
1429                         return error;
1430                 if (!ifp->if_ioctl)
1431                         return EOPNOTSUPP;
1432                 ifnet_serialize_all(ifp);
1433                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1434                 ifnet_deserialize_all(ifp);
1435                 if (error == 0)
1436                         getmicrotime(&ifp->if_lastchange);
1437                 return (error);
1438
1439         case SIOCSIFMTU:
1440         {
1441                 u_long oldmtu = ifp->if_mtu;
1442
1443                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1444                 if (error)
1445                         return (error);
1446                 if (ifp->if_ioctl == NULL)
1447                         return (EOPNOTSUPP);
1448                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1449                         return (EINVAL);
1450                 ifnet_serialize_all(ifp);
1451                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1452                 ifnet_deserialize_all(ifp);
1453                 if (error == 0) {
1454                         getmicrotime(&ifp->if_lastchange);
1455                         rt_ifmsg(ifp);
1456                 }
1457                 /*
1458                  * If the link MTU changed, do network layer specific procedure.
1459                  */
1460                 if (ifp->if_mtu != oldmtu) {
1461 #ifdef INET6
1462                         nd6_setmtu(ifp);
1463 #endif
1464                 }
1465                 return (error);
1466         }
1467
1468         case SIOCADDMULTI:
1469         case SIOCDELMULTI:
1470                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1471                 if (error)
1472                         return (error);
1473
1474                 /* Don't allow group membership on non-multicast interfaces. */
1475                 if ((ifp->if_flags & IFF_MULTICAST) == 0)
1476                         return EOPNOTSUPP;
1477
1478                 /* Don't let users screw up protocols' entries. */
1479                 if (ifr->ifr_addr.sa_family != AF_LINK)
1480                         return EINVAL;
1481
1482                 if (cmd == SIOCADDMULTI) {
1483                         struct ifmultiaddr *ifma;
1484                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1485                 } else {
1486                         error = if_delmulti(ifp, &ifr->ifr_addr);
1487                 }
1488                 if (error == 0)
1489                         getmicrotime(&ifp->if_lastchange);
1490                 return error;
1491
1492         case SIOCSIFPHYADDR:
1493         case SIOCDIFPHYADDR:
1494 #ifdef INET6
1495         case SIOCSIFPHYADDR_IN6:
1496 #endif
1497         case SIOCSLIFPHYADDR:
1498         case SIOCSIFMEDIA:
1499         case SIOCSIFGENERIC:
1500                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1501                 if (error)
1502                         return (error);
1503                 if (ifp->if_ioctl == 0)
1504                         return (EOPNOTSUPP);
1505                 ifnet_serialize_all(ifp);
1506                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1507                 ifnet_deserialize_all(ifp);
1508                 if (error == 0)
1509                         getmicrotime(&ifp->if_lastchange);
1510                 return error;
1511
1512         case SIOCGIFSTATUS:
1513                 ifs = (struct ifstat *)data;
1514                 ifs->ascii[0] = '\0';
1515
1516         case SIOCGIFPSRCADDR:
1517         case SIOCGIFPDSTADDR:
1518         case SIOCGLIFPHYADDR:
1519         case SIOCGIFMEDIA:
1520         case SIOCGIFGENERIC:
1521                 if (ifp->if_ioctl == NULL)
1522                         return (EOPNOTSUPP);
1523                 ifnet_serialize_all(ifp);
1524                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1525                 ifnet_deserialize_all(ifp);
1526                 return (error);
1527
1528         case SIOCSIFLLADDR:
1529                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1530                 if (error)
1531                         return (error);
1532                 return if_setlladdr(ifp,
1533                     ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1534
1535         default:
1536                 oif_flags = ifp->if_flags;
1537                 if (so->so_proto == 0)
1538                         return (EOPNOTSUPP);
1539 #ifndef COMPAT_43
1540                 error = so_pru_control(so, cmd, data, ifp);
1541 #else
1542             {
1543                 int ocmd = cmd;
1544
1545                 switch (cmd) {
1546
1547                 case SIOCSIFDSTADDR:
1548                 case SIOCSIFADDR:
1549                 case SIOCSIFBRDADDR:
1550                 case SIOCSIFNETMASK:
1551 #if BYTE_ORDER != BIG_ENDIAN
1552                         if (ifr->ifr_addr.sa_family == 0 &&
1553                             ifr->ifr_addr.sa_len < 16) {
1554                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1555                                 ifr->ifr_addr.sa_len = 16;
1556                         }
1557 #else
1558                         if (ifr->ifr_addr.sa_len == 0)
1559                                 ifr->ifr_addr.sa_len = 16;
1560 #endif
1561                         break;
1562
1563                 case OSIOCGIFADDR:
1564                         cmd = SIOCGIFADDR;
1565                         break;
1566
1567                 case OSIOCGIFDSTADDR:
1568                         cmd = SIOCGIFDSTADDR;
1569                         break;
1570
1571                 case OSIOCGIFBRDADDR:
1572                         cmd = SIOCGIFBRDADDR;
1573                         break;
1574
1575                 case OSIOCGIFNETMASK:
1576                         cmd = SIOCGIFNETMASK;
1577                 }
1578                 error =  so_pru_control(so, cmd, data, ifp);
1579                 switch (ocmd) {
1580
1581                 case OSIOCGIFADDR:
1582                 case OSIOCGIFDSTADDR:
1583                 case OSIOCGIFBRDADDR:
1584                 case OSIOCGIFNETMASK:
1585                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1586
1587                 }
1588             }
1589 #endif /* COMPAT_43 */
1590
1591                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1592 #ifdef INET6
1593                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
1594                         if (ifp->if_flags & IFF_UP) {
1595                                 crit_enter();
1596                                 in6_if_up(ifp);
1597                                 crit_exit();
1598                         }
1599 #endif
1600                 }
1601                 return (error);
1602
1603         }
1604         return (0);
1605 }
1606
1607 /*
1608  * Set/clear promiscuous mode on interface ifp based on the truth value
1609  * of pswitch.  The calls are reference counted so that only the first
1610  * "on" request actually has an effect, as does the final "off" request.
1611  * Results are undefined if the "off" and "on" requests are not matched.
1612  */
1613 int
1614 ifpromisc(struct ifnet *ifp, int pswitch)
1615 {
1616         struct ifreq ifr;
1617         int error;
1618         int oldflags;
1619
1620         oldflags = ifp->if_flags;
1621         if (ifp->if_flags & IFF_PPROMISC) {
1622                 /* Do nothing if device is in permanently promiscuous mode */
1623                 ifp->if_pcount += pswitch ? 1 : -1;
1624                 return (0);
1625         }
1626         if (pswitch) {
1627                 /*
1628                  * If the device is not configured up, we cannot put it in
1629                  * promiscuous mode.
1630                  */
1631                 if ((ifp->if_flags & IFF_UP) == 0)
1632                         return (ENETDOWN);
1633                 if (ifp->if_pcount++ != 0)
1634                         return (0);
1635                 ifp->if_flags |= IFF_PROMISC;
1636                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1637                     ifp->if_xname);
1638         } else {
1639                 if (--ifp->if_pcount > 0)
1640                         return (0);
1641                 ifp->if_flags &= ~IFF_PROMISC;
1642                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
1643                     ifp->if_xname);
1644         }
1645         ifr.ifr_flags = ifp->if_flags;
1646         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1647         ifnet_serialize_all(ifp);
1648         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
1649         ifnet_deserialize_all(ifp);
1650         if (error == 0)
1651                 rt_ifmsg(ifp);
1652         else
1653                 ifp->if_flags = oldflags;
1654         return error;
1655 }
1656
1657 /*
1658  * Return interface configuration
1659  * of system.  List may be used
1660  * in later ioctl's (above) to get
1661  * other information.
1662  */
1663 static int
1664 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
1665 {
1666         struct ifconf *ifc = (struct ifconf *)data;
1667         struct ifnet *ifp;
1668         struct sockaddr *sa;
1669         struct ifreq ifr, *ifrp;
1670         int space = ifc->ifc_len, error = 0;
1671
1672         ifrp = ifc->ifc_req;
1673         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1674                 struct ifaddr_container *ifac;
1675                 int addrs;
1676
1677                 if (space <= sizeof ifr)
1678                         break;
1679
1680                 /*
1681                  * Zero the stack declared structure first to prevent
1682                  * memory disclosure.
1683                  */
1684                 bzero(&ifr, sizeof(ifr));
1685                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
1686                     >= sizeof(ifr.ifr_name)) {
1687                         error = ENAMETOOLONG;
1688                         break;
1689                 }
1690
1691                 addrs = 0;
1692                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1693                         struct ifaddr *ifa = ifac->ifa;
1694
1695                         if (space <= sizeof ifr)
1696                                 break;
1697                         sa = ifa->ifa_addr;
1698                         if (cred->cr_prison &&
1699                             prison_if(cred, sa))
1700                                 continue;
1701                         addrs++;
1702 #ifdef COMPAT_43
1703                         if (cmd == OSIOCGIFCONF) {
1704                                 struct osockaddr *osa =
1705                                          (struct osockaddr *)&ifr.ifr_addr;
1706                                 ifr.ifr_addr = *sa;
1707                                 osa->sa_family = sa->sa_family;
1708                                 error = copyout(&ifr, ifrp, sizeof ifr);
1709                                 ifrp++;
1710                         } else
1711 #endif
1712                         if (sa->sa_len <= sizeof(*sa)) {
1713                                 ifr.ifr_addr = *sa;
1714                                 error = copyout(&ifr, ifrp, sizeof ifr);
1715                                 ifrp++;
1716                         } else {
1717                                 if (space < (sizeof ifr) + sa->sa_len -
1718                                             sizeof(*sa))
1719                                         break;
1720                                 space -= sa->sa_len - sizeof(*sa);
1721                                 error = copyout(&ifr, ifrp,
1722                                                 sizeof ifr.ifr_name);
1723                                 if (error == 0)
1724                                         error = copyout(sa, &ifrp->ifr_addr,
1725                                                         sa->sa_len);
1726                                 ifrp = (struct ifreq *)
1727                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
1728                         }
1729                         if (error)
1730                                 break;
1731                         space -= sizeof ifr;
1732                 }
1733                 if (error)
1734                         break;
1735                 if (!addrs) {
1736                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
1737                         error = copyout(&ifr, ifrp, sizeof ifr);
1738                         if (error)
1739                                 break;
1740                         space -= sizeof ifr;
1741                         ifrp++;
1742                 }
1743         }
1744         ifc->ifc_len -= space;
1745         return (error);
1746 }
1747
1748 /*
1749  * Just like if_promisc(), but for all-multicast-reception mode.
1750  */
1751 int
1752 if_allmulti(struct ifnet *ifp, int onswitch)
1753 {
1754         int error = 0;
1755         struct ifreq ifr;
1756
1757         crit_enter();
1758
1759         if (onswitch) {
1760                 if (ifp->if_amcount++ == 0) {
1761                         ifp->if_flags |= IFF_ALLMULTI;
1762                         ifr.ifr_flags = ifp->if_flags;
1763                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1764                         ifnet_serialize_all(ifp);
1765                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1766                                               NULL);
1767                         ifnet_deserialize_all(ifp);
1768                 }
1769         } else {
1770                 if (ifp->if_amcount > 1) {
1771                         ifp->if_amcount--;
1772                 } else {
1773                         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         }
1783
1784         crit_exit();
1785
1786         if (error == 0)
1787                 rt_ifmsg(ifp);
1788         return error;
1789 }
1790
1791 /*
1792  * Add a multicast listenership to the interface in question.
1793  * The link layer provides a routine which converts
1794  */
1795 int
1796 if_addmulti(
1797         struct ifnet *ifp,      /* interface to manipulate */
1798         struct sockaddr *sa,    /* address to add */
1799         struct ifmultiaddr **retifma)
1800 {
1801         struct sockaddr *llsa, *dupsa;
1802         int error;
1803         struct ifmultiaddr *ifma;
1804
1805         /*
1806          * If the matching multicast address already exists
1807          * then don't add a new one, just add a reference
1808          */
1809         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1810                 if (sa_equal(sa, ifma->ifma_addr)) {
1811                         ifma->ifma_refcount++;
1812                         if (retifma)
1813                                 *retifma = ifma;
1814                         return 0;
1815                 }
1816         }
1817
1818         /*
1819          * Give the link layer a chance to accept/reject it, and also
1820          * find out which AF_LINK address this maps to, if it isn't one
1821          * already.
1822          */
1823         if (ifp->if_resolvemulti) {
1824                 ifnet_serialize_all(ifp);
1825                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
1826                 ifnet_deserialize_all(ifp);
1827                 if (error) 
1828                         return error;
1829         } else {
1830                 llsa = 0;
1831         }
1832
1833         MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, M_WAITOK);
1834         MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, M_WAITOK);
1835         bcopy(sa, dupsa, sa->sa_len);
1836
1837         ifma->ifma_addr = dupsa;
1838         ifma->ifma_lladdr = llsa;
1839         ifma->ifma_ifp = ifp;
1840         ifma->ifma_refcount = 1;
1841         ifma->ifma_protospec = 0;
1842         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
1843
1844         /*
1845          * Some network interfaces can scan the address list at
1846          * interrupt time; lock them out.
1847          */
1848         crit_enter();
1849         LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1850         crit_exit();
1851         *retifma = ifma;
1852
1853         if (llsa != 0) {
1854                 LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1855                         if (sa_equal(ifma->ifma_addr, llsa))
1856                                 break;
1857                 }
1858                 if (ifma) {
1859                         ifma->ifma_refcount++;
1860                 } else {
1861                         MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma,
1862                                M_IFMADDR, M_WAITOK);
1863                         MALLOC(dupsa, struct sockaddr *, llsa->sa_len,
1864                                M_IFMADDR, M_WAITOK);
1865                         bcopy(llsa, dupsa, llsa->sa_len);
1866                         ifma->ifma_addr = dupsa;
1867                         ifma->ifma_ifp = ifp;
1868                         ifma->ifma_refcount = 1;
1869                         crit_enter();
1870                         LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1871                         crit_exit();
1872                 }
1873         }
1874         /*
1875          * We are certain we have added something, so call down to the
1876          * interface to let them know about it.
1877          */
1878         crit_enter();
1879         ifnet_serialize_all(ifp);
1880         ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
1881         ifnet_deserialize_all(ifp);
1882         crit_exit();
1883
1884         return 0;
1885 }
1886
1887 /*
1888  * Remove a reference to a multicast address on this interface.  Yell
1889  * if the request does not match an existing membership.
1890  */
1891 int
1892 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
1893 {
1894         struct ifmultiaddr *ifma;
1895
1896         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1897                 if (sa_equal(sa, ifma->ifma_addr))
1898                         break;
1899         if (ifma == 0)
1900                 return ENOENT;
1901
1902         if (ifma->ifma_refcount > 1) {
1903                 ifma->ifma_refcount--;
1904                 return 0;
1905         }
1906
1907         rt_newmaddrmsg(RTM_DELMADDR, ifma);
1908         sa = ifma->ifma_lladdr;
1909         crit_enter();
1910         LIST_REMOVE(ifma, ifma_link);
1911         /*
1912          * Make sure the interface driver is notified
1913          * in the case of a link layer mcast group being left.
1914          */
1915         if (ifma->ifma_addr->sa_family == AF_LINK && sa == 0) {
1916                 ifnet_serialize_all(ifp);
1917                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
1918                 ifnet_deserialize_all(ifp);
1919         }
1920         crit_exit();
1921         kfree(ifma->ifma_addr, M_IFMADDR);
1922         kfree(ifma, M_IFMADDR);
1923         if (sa == 0)
1924                 return 0;
1925
1926         /*
1927          * Now look for the link-layer address which corresponds to
1928          * this network address.  It had been squirreled away in
1929          * ifma->ifma_lladdr for this purpose (so we don't have
1930          * to call ifp->if_resolvemulti() again), and we saved that
1931          * value in sa above.  If some nasty deleted the
1932          * link-layer address out from underneath us, we can deal because
1933          * the address we stored was is not the same as the one which was
1934          * in the record for the link-layer address.  (So we don't complain
1935          * in that case.)
1936          */
1937         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1938                 if (sa_equal(sa, ifma->ifma_addr))
1939                         break;
1940         if (ifma == 0)
1941                 return 0;
1942
1943         if (ifma->ifma_refcount > 1) {
1944                 ifma->ifma_refcount--;
1945                 return 0;
1946         }
1947
1948         crit_enter();
1949         ifnet_serialize_all(ifp);
1950         LIST_REMOVE(ifma, ifma_link);
1951         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
1952         ifnet_deserialize_all(ifp);
1953         crit_exit();
1954         kfree(ifma->ifma_addr, M_IFMADDR);
1955         kfree(sa, M_IFMADDR);
1956         kfree(ifma, M_IFMADDR);
1957
1958         return 0;
1959 }
1960
1961 /*
1962  * Set the link layer address on an interface.
1963  *
1964  * At this time we only support certain types of interfaces,
1965  * and we don't allow the length of the address to change.
1966  */
1967 int
1968 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
1969 {
1970         struct sockaddr_dl *sdl;
1971         struct ifreq ifr;
1972
1973         sdl = IF_LLSOCKADDR(ifp);
1974         if (sdl == NULL)
1975                 return (EINVAL);
1976         if (len != sdl->sdl_alen)       /* don't allow length to change */
1977                 return (EINVAL);
1978         switch (ifp->if_type) {
1979         case IFT_ETHER:                 /* these types use struct arpcom */
1980         case IFT_XETHER:
1981         case IFT_L2VLAN:
1982                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
1983                 bcopy(lladdr, LLADDR(sdl), len);
1984                 break;
1985         default:
1986                 return (ENODEV);
1987         }
1988         /*
1989          * If the interface is already up, we need
1990          * to re-init it in order to reprogram its
1991          * address filter.
1992          */
1993         ifnet_serialize_all(ifp);
1994         if ((ifp->if_flags & IFF_UP) != 0) {
1995                 struct ifaddr_container *ifac;
1996
1997                 ifp->if_flags &= ~IFF_UP;
1998                 ifr.ifr_flags = ifp->if_flags;
1999                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2000                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2001                               NULL);
2002                 ifp->if_flags |= IFF_UP;
2003                 ifr.ifr_flags = ifp->if_flags;
2004                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2005                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2006                                  NULL);
2007 #ifdef INET
2008                 /*
2009                  * Also send gratuitous ARPs to notify other nodes about
2010                  * the address change.
2011                  */
2012                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2013                         struct ifaddr *ifa = ifac->ifa;
2014
2015                         if (ifa->ifa_addr != NULL &&
2016                             ifa->ifa_addr->sa_family == AF_INET)
2017                                 arp_ifinit(ifp, ifa);
2018                 }
2019 #endif
2020         }
2021         ifnet_deserialize_all(ifp);
2022         return (0);
2023 }
2024
2025 struct ifmultiaddr *
2026 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2027 {
2028         struct ifmultiaddr *ifma;
2029
2030         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2031                 if (sa_equal(ifma->ifma_addr, sa))
2032                         break;
2033
2034         return ifma;
2035 }
2036
2037 /*
2038  * This function locates the first real ethernet MAC from a network
2039  * card and loads it into node, returning 0 on success or ENOENT if
2040  * no suitable interfaces were found.  It is used by the uuid code to
2041  * generate a unique 6-byte number.
2042  */
2043 int
2044 if_getanyethermac(uint16_t *node, int minlen)
2045 {
2046         struct ifnet *ifp;
2047         struct sockaddr_dl *sdl;
2048
2049         TAILQ_FOREACH(ifp, &ifnet, if_link) {
2050                 if (ifp->if_type != IFT_ETHER)
2051                         continue;
2052                 sdl = IF_LLSOCKADDR(ifp);
2053                 if (sdl->sdl_alen < minlen)
2054                         continue;
2055                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2056                       minlen);
2057                 return(0);
2058         }
2059         return (ENOENT);
2060 }
2061
2062 /*
2063  * The name argument must be a pointer to storage which will last as
2064  * long as the interface does.  For physical devices, the result of
2065  * device_get_name(dev) is a good choice and for pseudo-devices a
2066  * static string works well.
2067  */
2068 void
2069 if_initname(struct ifnet *ifp, const char *name, int unit)
2070 {
2071         ifp->if_dname = name;
2072         ifp->if_dunit = unit;
2073         if (unit != IF_DUNIT_NONE)
2074                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2075         else
2076                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2077 }
2078
2079 int
2080 if_printf(struct ifnet *ifp, const char *fmt, ...)
2081 {
2082         __va_list ap;
2083         int retval;
2084
2085         retval = kprintf("%s: ", ifp->if_xname);
2086         __va_start(ap, fmt);
2087         retval += kvprintf(fmt, ap);
2088         __va_end(ap);
2089         return (retval);
2090 }
2091
2092 void
2093 ifq_set_classic(struct ifaltq *ifq)
2094 {
2095         ifq->altq_enqueue = ifq_classic_enqueue;
2096         ifq->altq_dequeue = ifq_classic_dequeue;
2097         ifq->altq_request = ifq_classic_request;
2098 }
2099
2100 int
2101 ifq_classic_enqueue(struct ifaltq *ifq, struct mbuf *m,
2102                     struct altq_pktattr *pa __unused)
2103 {
2104         logifq(enqueue, ifq);
2105         if (IF_QFULL(ifq)) {
2106                 m_freem(m);
2107                 return(ENOBUFS);
2108         } else {
2109                 IF_ENQUEUE(ifq, m);
2110                 return(0);
2111         }       
2112 }
2113
2114 struct mbuf *
2115 ifq_classic_dequeue(struct ifaltq *ifq, struct mbuf *mpolled, int op)
2116 {
2117         struct mbuf *m;
2118
2119         switch (op) {
2120         case ALTDQ_POLL:
2121                 IF_POLL(ifq, m);
2122                 break;
2123         case ALTDQ_REMOVE:
2124                 logifq(dequeue, ifq);
2125                 IF_DEQUEUE(ifq, m);
2126                 break;
2127         default:
2128                 panic("unsupported ALTQ dequeue op: %d", op);
2129         }
2130         KKASSERT(mpolled == NULL || mpolled == m);
2131         return(m);
2132 }
2133
2134 int
2135 ifq_classic_request(struct ifaltq *ifq, int req, void *arg)
2136 {
2137         switch (req) {
2138         case ALTRQ_PURGE:
2139                 IF_DRAIN(ifq);
2140                 break;
2141         default:
2142                 panic("unsupported ALTQ request: %d", req);
2143         }
2144         return(0);
2145 }
2146
2147 int
2148 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2149 {
2150         struct ifaltq *ifq = &ifp->if_snd;
2151         int running = 0, error, start = 0;
2152
2153         ASSERT_IFNET_NOT_SERIALIZED_TX(ifp);
2154
2155         ALTQ_LOCK(ifq);
2156         error = ifq_enqueue_locked(ifq, m, pa);
2157         if (error) {
2158                 ALTQ_UNLOCK(ifq);
2159                 return error;
2160         }
2161         if (!ifq->altq_started) {
2162                 /*
2163                  * Hold the interlock of ifnet.if_start
2164                  */
2165                 ifq->altq_started = 1;
2166                 start = 1;
2167         }
2168         ALTQ_UNLOCK(ifq);
2169
2170         ifp->if_obytes += m->m_pkthdr.len;
2171         if (m->m_flags & M_MCAST)
2172                 ifp->if_omcasts++;
2173
2174         if (!start) {
2175                 logifstart(avoid, ifp);
2176                 return 0;
2177         }
2178
2179         if (ifq_dispatch_schedonly) {
2180                 /*
2181                  * Always schedule ifnet.if_start on ifnet's CPU,
2182                  * short circuit the rest of this function.
2183                  */
2184                 logifstart(sched, ifp);
2185                 if_start_schedule(ifp);
2186                 return 0;
2187         }
2188
2189         /*
2190          * Try to do direct ifnet.if_start first, if there is
2191          * contention on ifnet's serializer, ifnet.if_start will
2192          * be scheduled on ifnet's CPU.
2193          */
2194         if (!ifnet_tryserialize_tx(ifp)) {
2195                 /*
2196                  * ifnet serializer contention happened,
2197                  * ifnet.if_start is scheduled on ifnet's
2198                  * CPU, and we keep going.
2199                  */
2200                 logifstart(contend_sched, ifp);
2201                 if_start_schedule(ifp);
2202                 return 0;
2203         }
2204
2205         if ((ifp->if_flags & IFF_OACTIVE) == 0) {
2206                 logifstart(run, ifp);
2207                 ifp->if_start(ifp);
2208                 if ((ifp->if_flags &
2209                      (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
2210                         running = 1;
2211         }
2212
2213         ifnet_deserialize_tx(ifp);
2214
2215         if (ifq_dispatch_schednochk || if_start_need_schedule(ifq, running)) {
2216                 /*
2217                  * More data need to be transmitted, ifnet.if_start is
2218                  * scheduled on ifnet's CPU, and we keep going.
2219                  * NOTE: ifnet.if_start interlock is not released.
2220                  */
2221                 logifstart(sched, ifp);
2222                 if_start_schedule(ifp);
2223         }
2224         return 0;
2225 }
2226
2227 void *
2228 ifa_create(int size, int flags)
2229 {
2230         struct ifaddr *ifa;
2231         int i;
2232
2233         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small\n"));
2234
2235         ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2236         if (ifa == NULL)
2237                 return NULL;
2238
2239         ifa->ifa_containers = kmalloc(ncpus * sizeof(struct ifaddr_container),
2240                                       M_IFADDR, M_WAITOK | M_ZERO);
2241         ifa->ifa_ncnt = ncpus;
2242         for (i = 0; i < ncpus; ++i) {
2243                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2244
2245                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2246                 ifac->ifa = ifa;
2247                 ifac->ifa_refcnt = 1;
2248         }
2249 #ifdef IFADDR_DEBUG
2250         kprintf("alloc ifa %p %d\n", ifa, size);
2251 #endif
2252         return ifa;
2253 }
2254
2255 void
2256 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2257 {
2258         struct ifaddr *ifa = ifac->ifa;
2259
2260         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2261         KKASSERT(ifac->ifa_refcnt == 0);
2262         KASSERT(ifac->ifa_listmask == 0,
2263                 ("ifa is still on %#x lists\n", ifac->ifa_listmask));
2264
2265         ifac->ifa_magic = IFA_CONTAINER_DEAD;
2266
2267 #ifdef IFADDR_DEBUG_VERBOSE
2268         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2269 #endif
2270
2271         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2272                 ("invalid # of ifac, %d\n", ifa->ifa_ncnt));
2273         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2274 #ifdef IFADDR_DEBUG
2275                 kprintf("free ifa %p\n", ifa);
2276 #endif
2277                 kfree(ifa->ifa_containers, M_IFADDR);
2278                 kfree(ifa, M_IFADDR);
2279         }
2280 }
2281
2282 static void
2283 ifa_iflink_dispatch(struct netmsg *nmsg)
2284 {
2285         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2286         struct ifaddr *ifa = msg->ifa;
2287         struct ifnet *ifp = msg->ifp;
2288         int cpu = mycpuid;
2289         struct ifaddr_container *ifac;
2290
2291         crit_enter();
2292
2293         ifac = &ifa->ifa_containers[cpu];
2294         ASSERT_IFAC_VALID(ifac);
2295         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2296                 ("ifaddr is on if_addrheads\n"));
2297
2298         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2299         if (msg->tail)
2300                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2301         else
2302                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2303
2304         crit_exit();
2305
2306         ifa_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
2307 }
2308
2309 void
2310 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2311 {
2312         struct netmsg_ifaddr msg;
2313
2314         netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2315                     ifa_iflink_dispatch);
2316         msg.ifa = ifa;
2317         msg.ifp = ifp;
2318         msg.tail = tail;
2319
2320         ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2321 }
2322
2323 static void
2324 ifa_ifunlink_dispatch(struct netmsg *nmsg)
2325 {
2326         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2327         struct ifaddr *ifa = msg->ifa;
2328         struct ifnet *ifp = msg->ifp;
2329         int cpu = mycpuid;
2330         struct ifaddr_container *ifac;
2331
2332         crit_enter();
2333
2334         ifac = &ifa->ifa_containers[cpu];
2335         ASSERT_IFAC_VALID(ifac);
2336         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2337                 ("ifaddr is not on if_addrhead\n"));
2338
2339         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2340         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2341
2342         crit_exit();
2343
2344         ifa_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
2345 }
2346
2347 void
2348 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2349 {
2350         struct netmsg_ifaddr msg;
2351
2352         netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2353                     ifa_ifunlink_dispatch);
2354         msg.ifa = ifa;
2355         msg.ifp = ifp;
2356
2357         ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2358 }
2359
2360 static void
2361 ifa_destroy_dispatch(struct netmsg *nmsg)
2362 {
2363         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2364
2365         IFAFREE(msg->ifa);
2366         ifa_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2367 }
2368
2369 void
2370 ifa_destroy(struct ifaddr *ifa)
2371 {
2372         struct netmsg_ifaddr msg;
2373
2374         netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2375                     ifa_destroy_dispatch);
2376         msg.ifa = ifa;
2377
2378         ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2379 }
2380
2381 struct lwkt_port *
2382 ifnet_portfn(int cpu)
2383 {
2384         return &ifnet_threads[cpu].td_msgport;
2385 }
2386
2387 void
2388 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2389 {
2390         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2391
2392         if (next_cpu < ncpus)
2393                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2394         else
2395                 lwkt_replymsg(lmsg, 0);
2396 }
2397
2398 int
2399 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
2400 {
2401         KKASSERT(cpu < ncpus);
2402         return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
2403 }
2404
2405 void
2406 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
2407 {
2408         KKASSERT(cpu < ncpus);
2409         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
2410 }
2411
2412 static void
2413 ifnetinit(void *dummy __unused)
2414 {
2415         int i;
2416
2417         for (i = 0; i < ncpus; ++i) {
2418                 struct thread *thr = &ifnet_threads[i];
2419
2420                 lwkt_create(netmsg_service_loop, &ifnet_mpsafe_thread, NULL,
2421                             thr, TDF_NETWORK | TDF_MPSAFE, i, "ifnet %d", i);
2422                 netmsg_service_port_init(&thr->td_msgport);
2423         }
2424 }