3408a0279fb85bb2b7131e5ed3751f0b4d45d0db
[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 1.78 2008/09/17 11:25:16 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/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/socketops.h>
52 #include <sys/protosw.h>
53 #include <sys/kernel.h>
54 #include <sys/ktr.h>
55 #include <sys/sockio.h>
56 #include <sys/syslog.h>
57 #include <sys/sysctl.h>
58 #include <sys/domain.h>
59 #include <sys/thread.h>
60 #include <sys/thread2.h>
61 #include <sys/serialize.h>
62 #include <sys/msgport2.h>
63
64 #include <net/if.h>
65 #include <net/if_arp.h>
66 #include <net/if_dl.h>
67 #include <net/if_types.h>
68 #include <net/if_var.h>
69 #include <net/ifq_var.h>
70 #include <net/radix.h>
71 #include <net/route.h>
72 #include <net/if_clone.h>
73 #include <net/netisr.h>
74 #include <net/netmsg2.h>
75
76 #include <machine/stdarg.h>
77 #include <machine/smp.h>
78
79 #if defined(INET) || defined(INET6)
80 /*XXX*/
81 #include <netinet/in.h>
82 #include <netinet/in_var.h>
83 #include <netinet/if_ether.h>
84 #ifdef INET6
85 #include <netinet6/in6_var.h>
86 #include <netinet6/in6_ifattach.h>
87 #endif
88 #endif
89
90 #if defined(COMPAT_43)
91 #include <emulation/43bsd/43bsd_socket.h>
92 #endif /* COMPAT_43 */
93
94 struct netmsg_ifaddr {
95         struct netmsg   netmsg;
96         struct ifaddr   *ifa;
97         struct ifnet    *ifp;
98         int             tail;
99 };
100
101 /*
102  * System initialization
103  */
104 static void     if_attachdomain(void *);
105 static void     if_attachdomain1(struct ifnet *);
106 static int      ifconf(u_long, caddr_t, struct ucred *);
107 static void     ifinit(void *);
108 static void     ifnetinit(void *);
109 static void     if_slowtimo(void *);
110 static void     link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
111 static int      if_rtdel(struct radix_node *, void *);
112
113 #ifdef INET6
114 /*
115  * XXX: declare here to avoid to include many inet6 related files..
116  * should be more generalized?
117  */
118 extern void     nd6_setmtu(struct ifnet *);
119 #endif
120
121 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
122 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
123
124 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
125 /* Must be after netisr_init */
126 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
127
128 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
129 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
130
131 int                     ifqmaxlen = IFQ_MAXLEN;
132 struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
133
134 /* In ifq_dispatch(), try to do direct ifnet.if_start first */
135 static int              ifq_dispatch_schedonly = 0;
136 SYSCTL_INT(_net_link_generic, OID_AUTO, ifq_dispatch_schedonly, CTLFLAG_RW,
137            &ifq_dispatch_schedonly, 0, "");
138
139 /* In ifq_dispatch(), schedule ifnet.if_start without checking ifnet.if_snd */
140 static int              ifq_dispatch_schednochk = 0;
141 SYSCTL_INT(_net_link_generic, OID_AUTO, ifq_dispatch_schednochk, CTLFLAG_RW,
142            &ifq_dispatch_schednochk, 0, "");
143
144 /* In if_devstart(), try to do direct ifnet.if_start first */
145 static int              if_devstart_schedonly = 0;
146 SYSCTL_INT(_net_link_generic, OID_AUTO, if_devstart_schedonly, CTLFLAG_RW,
147            &if_devstart_schedonly, 0, "");
148
149 /* In if_devstart(), schedule ifnet.if_start without checking ifnet.if_snd */
150 static int              if_devstart_schednochk = 0;
151 SYSCTL_INT(_net_link_generic, OID_AUTO, if_devstart_schednochk, CTLFLAG_RW,
152            &if_devstart_schednochk, 0, "");
153
154 #ifdef SMP
155 /* Schedule ifnet.if_start on the current CPU */
156 static int              if_start_oncpu_sched = 0;
157 SYSCTL_INT(_net_link_generic, OID_AUTO, if_start_oncpu_sched, CTLFLAG_RW,
158            &if_start_oncpu_sched, 0, "");
159 #endif
160
161 struct callout          if_slowtimo_timer;
162
163 int                     if_index = 0;
164 struct ifnet            **ifindex2ifnet = NULL;
165 static struct thread    ifnet_threads[MAXCPU];
166
167 #define IFQ_KTR_STRING          "ifq=%p"
168 #define IFQ_KTR_ARG_SIZE        (sizeof(void *))
169 #ifndef KTR_IFQ
170 #define KTR_IFQ                 KTR_ALL
171 #endif
172 KTR_INFO_MASTER(ifq);
173 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARG_SIZE);
174 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARG_SIZE);
175 #define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)
176
177 #define IF_START_KTR_STRING     "ifp=%p"
178 #define IF_START_KTR_ARG_SIZE   (sizeof(void *))
179 #ifndef KTR_IF_START
180 #define KTR_IF_START            KTR_ALL
181 #endif
182 KTR_INFO_MASTER(if_start);
183 KTR_INFO(KTR_IF_START, if_start, run, 0,
184          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
185 KTR_INFO(KTR_IF_START, if_start, sched, 1,
186          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
187 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
188          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
189 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
190          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
191 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
192          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
193 #define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
194
195 /*
196  * Network interface utility routines.
197  *
198  * Routines with ifa_ifwith* names take sockaddr *'s as
199  * parameters.
200  */
201 /* ARGSUSED*/
202 void
203 ifinit(void *dummy)
204 {
205         struct ifnet *ifp;
206
207         callout_init(&if_slowtimo_timer);
208
209         crit_enter();
210         TAILQ_FOREACH(ifp, &ifnet, if_link) {
211                 if (ifp->if_snd.ifq_maxlen == 0) {
212                         if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
213                         ifp->if_snd.ifq_maxlen = ifqmaxlen;
214                 }
215         }
216         crit_exit();
217
218         if_slowtimo(0);
219 }
220
221 static int
222 if_start_cpuid(struct ifnet *ifp)
223 {
224         return ifp->if_cpuid;
225 }
226
227 #ifdef DEVICE_POLLING
228 static int
229 if_start_cpuid_poll(struct ifnet *ifp)
230 {
231         int poll_cpuid = ifp->if_poll_cpuid;
232
233         if (poll_cpuid >= 0)
234                 return poll_cpuid;
235         else
236                 return ifp->if_cpuid;
237 }
238 #endif
239
240 static void
241 if_start_ipifunc(void *arg)
242 {
243         struct ifnet *ifp = arg;
244         struct lwkt_msg *lmsg = &ifp->if_start_nmsg[mycpuid].nm_lmsg;
245
246         crit_enter();
247         if (lmsg->ms_flags & MSGF_DONE)
248                 lwkt_sendmsg(ifnet_portfn(mycpuid), lmsg);
249         crit_exit();
250 }
251
252 /*
253  * Schedule ifnet.if_start on ifnet's CPU
254  */
255 static void
256 if_start_schedule(struct ifnet *ifp)
257 {
258 #ifdef SMP
259         int cpu;
260
261         if (if_start_oncpu_sched)
262                 cpu = mycpuid;
263         else
264                 cpu = ifp->if_start_cpuid(ifp);
265
266         if (cpu != mycpuid)
267                 lwkt_send_ipiq(globaldata_find(cpu), if_start_ipifunc, ifp);
268         else
269 #endif
270         if_start_ipifunc(ifp);
271 }
272
273 /*
274  * NOTE:
275  * This function will release ifnet.if_start interlock,
276  * if ifnet.if_start does not need to be scheduled
277  */
278 static __inline int
279 if_start_need_schedule(struct ifaltq *ifq, int running)
280 {
281         if (!running || ifq_is_empty(ifq)
282 #ifdef ALTQ
283             || ifq->altq_tbr != NULL
284 #endif
285         ) {
286                 ALTQ_LOCK(ifq);
287                 /*
288                  * ifnet.if_start interlock is released, if:
289                  * 1) Hardware can not take any packets, due to
290                  *    o  interface is marked down
291                  *    o  hardware queue is full (IFF_OACTIVE)
292                  *    Under the second situation, hardware interrupt
293                  *    or polling(4) will call/schedule ifnet.if_start
294                  *    when hardware queue is ready
295                  * 2) There is not packet in the ifnet.if_snd.
296                  *    Further ifq_dispatch or ifq_handoff will call/
297                  *    schedule ifnet.if_start
298                  * 3) TBR is used and it does not allow further
299                  *    dequeueing.
300                  *    TBR callout will call ifnet.if_start
301                  */
302                 if (!running || !ifq_data_ready(ifq)) {
303                         ifq->altq_started = 0;
304                         ALTQ_UNLOCK(ifq);
305                         return 0;
306                 }
307                 ALTQ_UNLOCK(ifq);
308         }
309         return 1;
310 }
311
312 static void
313 if_start_dispatch(struct netmsg *nmsg)
314 {
315         struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
316         struct ifnet *ifp = lmsg->u.ms_resultp;
317         struct ifaltq *ifq = &ifp->if_snd;
318         int running = 0;
319
320         crit_enter();
321         lwkt_replymsg(lmsg, 0); /* reply ASAP */
322         crit_exit();
323
324 #ifdef SMP
325         if (!if_start_oncpu_sched && mycpuid != ifp->if_start_cpuid(ifp)) {
326                 /*
327                  * If the ifnet is still up, we need to
328                  * chase its CPU change.
329                  */
330                 if (ifp->if_flags & IFF_UP) {
331                         logifstart(chase_sched, ifp);
332                         if_start_schedule(ifp);
333                         return;
334                 } else {
335                         goto check;
336                 }
337         }
338 #endif
339
340         if (ifp->if_flags & IFF_UP) {
341                 lwkt_serialize_enter(ifp->if_serializer); /* XXX try? */
342                 if ((ifp->if_flags & IFF_OACTIVE) == 0) {
343                         logifstart(run, ifp);
344                         ifp->if_start(ifp);
345                         if ((ifp->if_flags &
346                         (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
347                                 running = 1;
348                 }
349                 lwkt_serialize_exit(ifp->if_serializer);
350         }
351 #ifdef SMP
352 check:
353 #endif
354         if (if_start_need_schedule(ifq, running)) {
355                 crit_enter();
356                 if (lmsg->ms_flags & MSGF_DONE) { /* XXX necessary? */
357                         logifstart(sched, ifp);
358                         lwkt_sendmsg(ifnet_portfn(mycpuid), lmsg);
359                 }
360                 crit_exit();
361         }
362 }
363
364 /* Device driver ifnet.if_start helper function */
365 void
366 if_devstart(struct ifnet *ifp)
367 {
368         struct ifaltq *ifq = &ifp->if_snd;
369         int running = 0;
370
371         ASSERT_SERIALIZED(ifp->if_serializer);
372
373         ALTQ_LOCK(ifq);
374         if (ifq->altq_started || !ifq_data_ready(ifq)) {
375                 logifstart(avoid, ifp);
376                 ALTQ_UNLOCK(ifq);
377                 return;
378         }
379         ifq->altq_started = 1;
380         ALTQ_UNLOCK(ifq);
381
382         if (if_devstart_schedonly) {
383                 /*
384                  * Always schedule ifnet.if_start on ifnet's CPU,
385                  * short circuit the rest of this function.
386                  */
387                 logifstart(sched, ifp);
388                 if_start_schedule(ifp);
389                 return;
390         }
391
392         logifstart(run, ifp);
393         ifp->if_start(ifp);
394
395         if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
396                 running = 1;
397
398         if (if_devstart_schednochk || if_start_need_schedule(ifq, running)) {
399                 /*
400                  * More data need to be transmitted, ifnet.if_start is
401                  * scheduled on ifnet's CPU, and we keep going.
402                  * NOTE: ifnet.if_start interlock is not released.
403                  */
404                 logifstart(sched, ifp);
405                 if_start_schedule(ifp);
406         }
407 }
408
409 /*
410  * Attach an interface to the list of "active" interfaces.
411  *
412  * The serializer is optional.  If non-NULL access to the interface
413  * may be MPSAFE.
414  */
415 void
416 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
417 {
418         unsigned socksize, ifasize;
419         int namelen, masklen;
420         struct sockaddr_dl *sdl;
421         struct ifaddr *ifa;
422         struct ifaltq *ifq;
423         int i;
424
425         static int if_indexlim = 8;
426
427         /*
428          * The serializer can be passed in from the device, allowing the
429          * same serializer to be used for both the interrupt interlock and
430          * the device queue.  If not specified, the netif structure will
431          * use an embedded serializer.
432          */
433         if (serializer == NULL) {
434                 serializer = &ifp->if_default_serializer;
435                 lwkt_serialize_init(serializer);
436         }
437         ifp->if_serializer = serializer;
438
439         ifp->if_start_cpuid = if_start_cpuid;
440         ifp->if_cpuid = 0;
441
442 #ifdef DEVICE_POLLING
443         /* Device is not in polling mode by default */
444         ifp->if_poll_cpuid = -1;
445         if (ifp->if_poll != NULL)
446                 ifp->if_start_cpuid = if_start_cpuid_poll;
447 #endif
448
449         ifp->if_start_nmsg = kmalloc(ncpus * sizeof(struct netmsg),
450                                      M_IFADDR /* XXX */, M_WAITOK);
451         for (i = 0; i < ncpus; ++i) {
452                 netmsg_init(&ifp->if_start_nmsg[i], &netisr_adone_rport, 0,
453                             if_start_dispatch);
454                 ifp->if_start_nmsg[i].nm_lmsg.u.ms_resultp = ifp;
455         }
456
457         TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
458         ifp->if_index = ++if_index;
459
460         /*
461          * XXX -
462          * The old code would work if the interface passed a pre-existing
463          * chain of ifaddrs to this code.  We don't trust our callers to
464          * properly initialize the tailq, however, so we no longer allow
465          * this unlikely case.
466          */
467         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
468                                     M_IFADDR, M_WAITOK | M_ZERO);
469         for (i = 0; i < ncpus; ++i)
470                 TAILQ_INIT(&ifp->if_addrheads[i]);
471
472         TAILQ_INIT(&ifp->if_prefixhead);
473         LIST_INIT(&ifp->if_multiaddrs);
474         getmicrotime(&ifp->if_lastchange);
475         if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
476                 unsigned int n;
477                 struct ifnet **q;
478
479                 if_indexlim <<= 1;
480
481                 /* grow ifindex2ifnet */
482                 n = if_indexlim * sizeof(*q);
483                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
484                 if (ifindex2ifnet) {
485                         bcopy(ifindex2ifnet, q, n/2);
486                         kfree(ifindex2ifnet, M_IFADDR);
487                 }
488                 ifindex2ifnet = q;
489         }
490
491         ifindex2ifnet[if_index] = ifp;
492
493         /*
494          * create a Link Level name for this device
495          */
496         namelen = strlen(ifp->if_xname);
497 #define _offsetof(t, m) ((int)((caddr_t)&((t *)0)->m))
498         masklen = _offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
499         socksize = masklen + ifp->if_addrlen;
500 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
501         if (socksize < sizeof(*sdl))
502                 socksize = sizeof(*sdl);
503         socksize = ROUNDUP(socksize);
504         ifasize = sizeof(struct ifaddr) + 2 * socksize;
505         ifa = ifa_create(ifasize, M_WAITOK);
506         sdl = (struct sockaddr_dl *)(ifa + 1);
507         sdl->sdl_len = socksize;
508         sdl->sdl_family = AF_LINK;
509         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
510         sdl->sdl_nlen = namelen;
511         sdl->sdl_index = ifp->if_index;
512         sdl->sdl_type = ifp->if_type;
513         ifp->if_lladdr = ifa;
514         ifa->ifa_ifp = ifp;
515         ifa->ifa_rtrequest = link_rtrequest;
516         ifa->ifa_addr = (struct sockaddr *)sdl;
517         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
518         ifa->ifa_netmask = (struct sockaddr *)sdl;
519         sdl->sdl_len = masklen;
520         while (namelen != 0)
521                 sdl->sdl_data[--namelen] = 0xff;
522         ifa_iflink(ifa, ifp, 0 /* Insert head */);
523
524         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
525
526         ifq = &ifp->if_snd;
527         ifq->altq_type = 0;
528         ifq->altq_disc = NULL;
529         ifq->altq_flags &= ALTQF_CANTCHANGE;
530         ifq->altq_tbr = NULL;
531         ifq->altq_ifp = ifp;
532         ifq->altq_started = 0;
533         ifq->altq_prepended = NULL;
534         ALTQ_LOCK_INIT(ifq);
535         ifq_set_classic(ifq);
536
537         if (!SLIST_EMPTY(&domains))
538                 if_attachdomain1(ifp);
539
540         /* Announce the interface. */
541         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
542 }
543
544 static void
545 if_attachdomain(void *dummy)
546 {
547         struct ifnet *ifp;
548
549         crit_enter();
550         TAILQ_FOREACH(ifp, &ifnet, if_list)
551                 if_attachdomain1(ifp);
552         crit_exit();
553 }
554 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
555         if_attachdomain, NULL);
556
557 static void
558 if_attachdomain1(struct ifnet *ifp)
559 {
560         struct domain *dp;
561
562         crit_enter();
563
564         /* address family dependent data region */
565         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
566         SLIST_FOREACH(dp, &domains, dom_next)
567                 if (dp->dom_ifattach)
568                         ifp->if_afdata[dp->dom_family] =
569                                 (*dp->dom_ifattach)(ifp);
570         crit_exit();
571 }
572
573 /*
574  * Purge all addresses whose type is _not_ AF_LINK
575  */
576 void
577 if_purgeaddrs_nolink(struct ifnet *ifp)
578 {
579         struct ifaddr_container *ifac, *next;
580
581         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
582                               ifa_link, next) {
583                 struct ifaddr *ifa = ifac->ifa;
584
585                 /* Leave link ifaddr as it is */
586                 if (ifa->ifa_addr->sa_family == AF_LINK)
587                         continue;
588 #ifdef INET
589                 /* XXX: Ugly!! ad hoc just for INET */
590                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
591                         struct ifaliasreq ifr;
592 #ifdef IFADDR_DEBUG_VERBOSE
593                         int i;
594
595                         kprintf("purge in4 addr %p: ", ifa);
596                         for (i = 0; i < ncpus; ++i)
597                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
598                         kprintf("\n");
599 #endif
600
601                         bzero(&ifr, sizeof ifr);
602                         ifr.ifra_addr = *ifa->ifa_addr;
603                         if (ifa->ifa_dstaddr)
604                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
605                         if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
606                                        NULL) == 0)
607                                 continue;
608                 }
609 #endif /* INET */
610 #ifdef INET6
611                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
612 #ifdef IFADDR_DEBUG_VERBOSE
613                         int i;
614
615                         kprintf("purge in6 addr %p: ", ifa);
616                         for (i = 0; i < ncpus; ++i)
617                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
618                         kprintf("\n");
619 #endif
620
621                         in6_purgeaddr(ifa);
622                         /* ifp_addrhead is already updated */
623                         continue;
624                 }
625 #endif /* INET6 */
626                 ifa_ifunlink(ifa, ifp);
627                 ifa_destroy(ifa);
628         }
629 }
630
631 /*
632  * Detach an interface, removing it from the
633  * list of "active" interfaces.
634  */
635 void
636 if_detach(struct ifnet *ifp)
637 {
638         struct radix_node_head  *rnh;
639         int i;
640         int cpu, origcpu;
641         struct domain *dp;
642
643         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
644
645         /*
646          * Remove routes and flush queues.
647          */
648         crit_enter();
649 #ifdef DEVICE_POLLING
650         if (ifp->if_flags & IFF_POLLING)
651                 ether_poll_deregister(ifp);
652 #endif
653         if_down(ifp);
654
655         if (ifq_is_enabled(&ifp->if_snd))
656                 altq_disable(&ifp->if_snd);
657         if (ifq_is_attached(&ifp->if_snd))
658                 altq_detach(&ifp->if_snd);
659
660         /*
661          * Clean up all addresses.
662          */
663         ifp->if_lladdr = NULL;
664
665         if_purgeaddrs_nolink(ifp);
666         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
667                 struct ifaddr *ifa;
668
669                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
670                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
671                         ("non-link ifaddr is left on if_addrheads"));
672
673                 ifa_ifunlink(ifa, ifp);
674                 ifa_destroy(ifa);
675                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
676                         ("there are still ifaddrs left on if_addrheads"));
677         }
678
679 #ifdef INET
680         /*
681          * Remove all IPv4 kernel structures related to ifp.
682          */
683         in_ifdetach(ifp);
684 #endif
685
686 #ifdef INET6
687         /*
688          * Remove all IPv6 kernel structs related to ifp.  This should be done
689          * before removing routing entries below, since IPv6 interface direct
690          * routes are expected to be removed by the IPv6-specific kernel API.
691          * Otherwise, the kernel will detect some inconsistency and bark it.
692          */
693         in6_ifdetach(ifp);
694 #endif
695
696         /*
697          * Delete all remaining routes using this interface
698          * Unfortuneatly the only way to do this is to slog through
699          * the entire routing table looking for routes which point
700          * to this interface...oh well...
701          */
702         origcpu = mycpuid;
703         for (cpu = 0; cpu < ncpus2; cpu++) {
704                 lwkt_migratecpu(cpu);
705                 for (i = 1; i <= AF_MAX; i++) {
706                         if ((rnh = rt_tables[cpu][i]) == NULL)
707                                 continue;
708                         rnh->rnh_walktree(rnh, if_rtdel, ifp);
709                 }
710         }
711         lwkt_migratecpu(origcpu);
712
713         /* Announce that the interface is gone. */
714         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
715
716         SLIST_FOREACH(dp, &domains, dom_next)
717                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
718                         (*dp->dom_ifdetach)(ifp,
719                                 ifp->if_afdata[dp->dom_family]);
720
721         /*
722          * Remove interface from ifindex2ifp[] and maybe decrement if_index.
723          */
724         ifindex2ifnet[ifp->if_index] = NULL;
725         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
726                 if_index--;
727
728         TAILQ_REMOVE(&ifnet, ifp, if_link);
729         kfree(ifp->if_addrheads, M_IFADDR);
730         kfree(ifp->if_start_nmsg, M_IFADDR);
731         crit_exit();
732 }
733
734 /*
735  * Delete Routes for a Network Interface
736  *
737  * Called for each routing entry via the rnh->rnh_walktree() call above
738  * to delete all route entries referencing a detaching network interface.
739  *
740  * Arguments:
741  *      rn      pointer to node in the routing table
742  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
743  *
744  * Returns:
745  *      0       successful
746  *      errno   failed - reason indicated
747  *
748  */
749 static int
750 if_rtdel(struct radix_node *rn, void *arg)
751 {
752         struct rtentry  *rt = (struct rtentry *)rn;
753         struct ifnet    *ifp = arg;
754         int             err;
755
756         if (rt->rt_ifp == ifp) {
757
758                 /*
759                  * Protect (sorta) against walktree recursion problems
760                  * with cloned routes
761                  */
762                 if (!(rt->rt_flags & RTF_UP))
763                         return (0);
764
765                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
766                                 rt_mask(rt), rt->rt_flags,
767                                 (struct rtentry **) NULL);
768                 if (err) {
769                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
770                 }
771         }
772
773         return (0);
774 }
775
776 /*
777  * Locate an interface based on a complete address.
778  */
779 struct ifaddr *
780 ifa_ifwithaddr(struct sockaddr *addr)
781 {
782         struct ifnet *ifp;
783
784         TAILQ_FOREACH(ifp, &ifnet, if_link) {
785                 struct ifaddr_container *ifac;
786
787                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
788                         struct ifaddr *ifa = ifac->ifa;
789
790                         if (ifa->ifa_addr->sa_family != addr->sa_family)
791                                 continue;
792                         if (sa_equal(addr, ifa->ifa_addr))
793                                 return (ifa);
794                         if ((ifp->if_flags & IFF_BROADCAST) &&
795                             ifa->ifa_broadaddr &&
796                             /* IPv6 doesn't have broadcast */
797                             ifa->ifa_broadaddr->sa_len != 0 &&
798                             sa_equal(ifa->ifa_broadaddr, addr))
799                                 return (ifa);
800                 }
801         }
802         return (NULL);
803 }
804 /*
805  * Locate the point to point interface with a given destination address.
806  */
807 struct ifaddr *
808 ifa_ifwithdstaddr(struct sockaddr *addr)
809 {
810         struct ifnet *ifp;
811
812         TAILQ_FOREACH(ifp, &ifnet, if_link) {
813                 struct ifaddr_container *ifac;
814
815                 if (!(ifp->if_flags & IFF_POINTOPOINT))
816                         continue;
817
818                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
819                         struct ifaddr *ifa = ifac->ifa;
820
821                         if (ifa->ifa_addr->sa_family != addr->sa_family)
822                                 continue;
823                         if (ifa->ifa_dstaddr &&
824                             sa_equal(addr, ifa->ifa_dstaddr))
825                                 return (ifa);
826                 }
827         }
828         return (NULL);
829 }
830
831 /*
832  * Find an interface on a specific network.  If many, choice
833  * is most specific found.
834  */
835 struct ifaddr *
836 ifa_ifwithnet(struct sockaddr *addr)
837 {
838         struct ifnet *ifp;
839         struct ifaddr *ifa_maybe = NULL;
840         u_int af = addr->sa_family;
841         char *addr_data = addr->sa_data, *cplim;
842
843         /*
844          * AF_LINK addresses can be looked up directly by their index number,
845          * so do that if we can.
846          */
847         if (af == AF_LINK) {
848                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
849
850                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
851                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
852         }
853
854         /*
855          * Scan though each interface, looking for ones that have
856          * addresses in this address family.
857          */
858         TAILQ_FOREACH(ifp, &ifnet, if_link) {
859                 struct ifaddr_container *ifac;
860
861                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
862                         struct ifaddr *ifa = ifac->ifa;
863                         char *cp, *cp2, *cp3;
864
865                         if (ifa->ifa_addr->sa_family != af)
866 next:                           continue;
867                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
868                                 /*
869                                  * This is a bit broken as it doesn't
870                                  * take into account that the remote end may
871                                  * be a single node in the network we are
872                                  * looking for.
873                                  * The trouble is that we don't know the
874                                  * netmask for the remote end.
875                                  */
876                                 if (ifa->ifa_dstaddr != NULL &&
877                                     sa_equal(addr, ifa->ifa_dstaddr))
878                                         return (ifa);
879                         } else {
880                                 /*
881                                  * if we have a special address handler,
882                                  * then use it instead of the generic one.
883                                  */
884                                 if (ifa->ifa_claim_addr) {
885                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
886                                                 return (ifa);
887                                         } else {
888                                                 continue;
889                                         }
890                                 }
891
892                                 /*
893                                  * Scan all the bits in the ifa's address.
894                                  * If a bit dissagrees with what we are
895                                  * looking for, mask it with the netmask
896                                  * to see if it really matters.
897                                  * (A byte at a time)
898                                  */
899                                 if (ifa->ifa_netmask == 0)
900                                         continue;
901                                 cp = addr_data;
902                                 cp2 = ifa->ifa_addr->sa_data;
903                                 cp3 = ifa->ifa_netmask->sa_data;
904                                 cplim = ifa->ifa_netmask->sa_len +
905                                         (char *)ifa->ifa_netmask;
906                                 while (cp3 < cplim)
907                                         if ((*cp++ ^ *cp2++) & *cp3++)
908                                                 goto next; /* next address! */
909                                 /*
910                                  * If the netmask of what we just found
911                                  * is more specific than what we had before
912                                  * (if we had one) then remember the new one
913                                  * before continuing to search
914                                  * for an even better one.
915                                  */
916                                 if (ifa_maybe == 0 ||
917                                     rn_refines((char *)ifa->ifa_netmask,
918                                                (char *)ifa_maybe->ifa_netmask))
919                                         ifa_maybe = ifa;
920                         }
921                 }
922         }
923         return (ifa_maybe);
924 }
925
926 /*
927  * Find an interface address specific to an interface best matching
928  * a given address.
929  */
930 struct ifaddr *
931 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
932 {
933         struct ifaddr_container *ifac;
934         char *cp, *cp2, *cp3;
935         char *cplim;
936         struct ifaddr *ifa_maybe = 0;
937         u_int af = addr->sa_family;
938
939         if (af >= AF_MAX)
940                 return (0);
941         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
942                 struct ifaddr *ifa = ifac->ifa;
943
944                 if (ifa->ifa_addr->sa_family != af)
945                         continue;
946                 if (ifa_maybe == 0)
947                         ifa_maybe = ifa;
948                 if (ifa->ifa_netmask == NULL) {
949                         if (sa_equal(addr, ifa->ifa_addr) ||
950                             (ifa->ifa_dstaddr != NULL &&
951                              sa_equal(addr, ifa->ifa_dstaddr)))
952                                 return (ifa);
953                         continue;
954                 }
955                 if (ifp->if_flags & IFF_POINTOPOINT) {
956                         if (sa_equal(addr, ifa->ifa_dstaddr))
957                                 return (ifa);
958                 } else {
959                         cp = addr->sa_data;
960                         cp2 = ifa->ifa_addr->sa_data;
961                         cp3 = ifa->ifa_netmask->sa_data;
962                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
963                         for (; cp3 < cplim; cp3++)
964                                 if ((*cp++ ^ *cp2++) & *cp3)
965                                         break;
966                         if (cp3 == cplim)
967                                 return (ifa);
968                 }
969         }
970         return (ifa_maybe);
971 }
972
973 /*
974  * Default action when installing a route with a Link Level gateway.
975  * Lookup an appropriate real ifa to point to.
976  * This should be moved to /sys/net/link.c eventually.
977  */
978 static void
979 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
980 {
981         struct ifaddr *ifa;
982         struct sockaddr *dst;
983         struct ifnet *ifp;
984
985         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
986             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
987                 return;
988         ifa = ifaof_ifpforaddr(dst, ifp);
989         if (ifa != NULL) {
990                 IFAFREE(rt->rt_ifa);
991                 IFAREF(ifa);
992                 rt->rt_ifa = ifa;
993                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
994                         ifa->ifa_rtrequest(cmd, rt, info);
995         }
996 }
997
998 /*
999  * Mark an interface down and notify protocols of
1000  * the transition.
1001  * NOTE: must be called at splnet or eqivalent.
1002  */
1003 void
1004 if_unroute(struct ifnet *ifp, int flag, int fam)
1005 {
1006         struct ifaddr_container *ifac;
1007
1008         ifp->if_flags &= ~flag;
1009         getmicrotime(&ifp->if_lastchange);
1010         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1011                 struct ifaddr *ifa = ifac->ifa;
1012
1013                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1014                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1015         }
1016         ifq_purge(&ifp->if_snd);
1017         rt_ifmsg(ifp);
1018 }
1019
1020 /*
1021  * Mark an interface up and notify protocols of
1022  * the transition.
1023  * NOTE: must be called at splnet or eqivalent.
1024  */
1025 void
1026 if_route(struct ifnet *ifp, int flag, int fam)
1027 {
1028         struct ifaddr_container *ifac;
1029
1030         ifq_purge(&ifp->if_snd);
1031         ifp->if_flags |= flag;
1032         getmicrotime(&ifp->if_lastchange);
1033         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1034                 struct ifaddr *ifa = ifac->ifa;
1035
1036                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1037                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1038         }
1039         rt_ifmsg(ifp);
1040 #ifdef INET6
1041         in6_if_up(ifp);
1042 #endif
1043 }
1044
1045 /*
1046  * Mark an interface down and notify protocols of the transition.  An
1047  * interface going down is also considered to be a synchronizing event.
1048  * We must ensure that all packet processing related to the interface
1049  * has completed before we return so e.g. the caller can free the ifnet
1050  * structure that the mbufs may be referencing.
1051  *
1052  * NOTE: must be called at splnet or eqivalent.
1053  */
1054 void
1055 if_down(struct ifnet *ifp)
1056 {
1057         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1058         netmsg_service_sync();
1059 }
1060
1061 /*
1062  * Mark an interface up and notify protocols of
1063  * the transition.
1064  * NOTE: must be called at splnet or eqivalent.
1065  */
1066 void
1067 if_up(struct ifnet *ifp)
1068 {
1069         if_route(ifp, IFF_UP, AF_UNSPEC);
1070 }
1071
1072 /*
1073  * Process a link state change.
1074  * NOTE: must be called at splsoftnet or equivalent.
1075  */
1076 void
1077 if_link_state_change(struct ifnet *ifp)
1078 {
1079         rt_ifmsg(ifp);
1080 }
1081
1082 /*
1083  * Handle interface watchdog timer routines.  Called
1084  * from softclock, we decrement timers (if set) and
1085  * call the appropriate interface routine on expiration.
1086  */
1087 static void
1088 if_slowtimo(void *arg)
1089 {
1090         struct ifnet *ifp;
1091
1092         crit_enter();
1093
1094         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1095                 if (ifp->if_timer == 0 || --ifp->if_timer)
1096                         continue;
1097                 if (ifp->if_watchdog) {
1098                         if (lwkt_serialize_try(ifp->if_serializer)) {
1099                                 (*ifp->if_watchdog)(ifp);
1100                                 lwkt_serialize_exit(ifp->if_serializer);
1101                         } else {
1102                                 /* try again next timeout */
1103                                 ++ifp->if_timer;
1104                         }
1105                 }
1106         }
1107
1108         crit_exit();
1109
1110         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1111 }
1112
1113 /*
1114  * Map interface name to
1115  * interface structure pointer.
1116  */
1117 struct ifnet *
1118 ifunit(const char *name)
1119 {
1120         struct ifnet *ifp;
1121
1122         /*
1123          * Search all the interfaces for this name/number
1124          */
1125
1126         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1127                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1128                         break;
1129         }
1130         return (ifp);
1131 }
1132
1133
1134 /*
1135  * Map interface name in a sockaddr_dl to
1136  * interface structure pointer.
1137  */
1138 struct ifnet *
1139 if_withname(struct sockaddr *sa)
1140 {
1141         char ifname[IFNAMSIZ+1];
1142         struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1143
1144         if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1145              (sdl->sdl_nlen > IFNAMSIZ) )
1146                 return NULL;
1147
1148         /*
1149          * ifunit wants a null-terminated name.  It may not be null-terminated
1150          * in the sockaddr.  We don't want to change the caller's sockaddr,
1151          * and there might not be room to put the trailing null anyway, so we
1152          * make a local copy that we know we can null terminate safely.
1153          */
1154
1155         bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1156         ifname[sdl->sdl_nlen] = '\0';
1157         return ifunit(ifname);
1158 }
1159
1160
1161 /*
1162  * Interface ioctls.
1163  */
1164 int
1165 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1166 {
1167         struct ifnet *ifp;
1168         struct ifreq *ifr;
1169         struct ifstat *ifs;
1170         int error;
1171         short oif_flags;
1172         int new_flags;
1173         size_t namelen, onamelen;
1174         char new_name[IFNAMSIZ];
1175         struct ifaddr *ifa;
1176         struct sockaddr_dl *sdl;
1177
1178         switch (cmd) {
1179
1180         case SIOCGIFCONF:
1181         case OSIOCGIFCONF:
1182                 return (ifconf(cmd, data, cred));
1183         }
1184         ifr = (struct ifreq *)data;
1185
1186         switch (cmd) {
1187         case SIOCIFCREATE:
1188         case SIOCIFDESTROY:
1189                 if ((error = suser_cred(cred, 0)) != 0)
1190                         return (error);
1191                 return ((cmd == SIOCIFCREATE) ?
1192                         if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name)) :
1193                         if_clone_destroy(ifr->ifr_name));
1194
1195         case SIOCIFGCLONERS:
1196                 return (if_clone_list((struct if_clonereq *)data));
1197         }
1198
1199         ifp = ifunit(ifr->ifr_name);
1200         if (ifp == 0)
1201                 return (ENXIO);
1202         switch (cmd) {
1203
1204         case SIOCGIFFLAGS:
1205                 ifr->ifr_flags = ifp->if_flags;
1206                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1207                 break;
1208
1209         case SIOCGIFCAP:
1210                 ifr->ifr_reqcap = ifp->if_capabilities;
1211                 ifr->ifr_curcap = ifp->if_capenable;
1212                 break;
1213
1214         case SIOCGIFMETRIC:
1215                 ifr->ifr_metric = ifp->if_metric;
1216                 break;
1217
1218         case SIOCGIFMTU:
1219                 ifr->ifr_mtu = ifp->if_mtu;
1220                 break;
1221
1222         case SIOCGIFPHYS:
1223                 ifr->ifr_phys = ifp->if_physical;
1224                 break;
1225
1226         case SIOCGIFPOLLCPU:
1227 #ifdef DEVICE_POLLING
1228                 ifr->ifr_pollcpu = ifp->if_poll_cpuid;
1229 #else
1230                 ifr->ifr_pollcpu = -1;
1231 #endif
1232                 break;
1233
1234         case SIOCSIFPOLLCPU:
1235 #ifdef DEVICE_POLLING
1236                 if ((ifp->if_flags & IFF_POLLING) == 0)
1237                         ether_pollcpu_register(ifp, ifr->ifr_pollcpu);
1238 #endif
1239                 break;
1240
1241         case SIOCSIFFLAGS:
1242                 error = suser_cred(cred, 0);
1243                 if (error)
1244                         return (error);
1245                 new_flags = (ifr->ifr_flags & 0xffff) |
1246                     (ifr->ifr_flagshigh << 16);
1247                 if (ifp->if_flags & IFF_SMART) {
1248                         /* Smart drivers twiddle their own routes */
1249                 } else if (ifp->if_flags & IFF_UP &&
1250                     (new_flags & IFF_UP) == 0) {
1251                         crit_enter();
1252                         if_down(ifp);
1253                         crit_exit();
1254                 } else if (new_flags & IFF_UP &&
1255                     (ifp->if_flags & IFF_UP) == 0) {
1256                         crit_enter();
1257                         if_up(ifp);
1258                         crit_exit();
1259                 }
1260
1261 #ifdef DEVICE_POLLING
1262                 if ((new_flags ^ ifp->if_flags) & IFF_POLLING) {
1263                         if (new_flags & IFF_POLLING) {
1264                                 ether_poll_register(ifp);
1265                         } else {
1266                                 ether_poll_deregister(ifp);
1267                         }
1268                 }
1269 #endif
1270
1271                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1272                         (new_flags &~ IFF_CANTCHANGE);
1273                 if (new_flags & IFF_PPROMISC) {
1274                         /* Permanently promiscuous mode requested */
1275                         ifp->if_flags |= IFF_PROMISC;
1276                 } else if (ifp->if_pcount == 0) {
1277                         ifp->if_flags &= ~IFF_PROMISC;
1278                 }
1279                 if (ifp->if_ioctl) {
1280                         lwkt_serialize_enter(ifp->if_serializer);
1281                         ifp->if_ioctl(ifp, cmd, data, cred);
1282                         lwkt_serialize_exit(ifp->if_serializer);
1283                 }
1284                 getmicrotime(&ifp->if_lastchange);
1285                 break;
1286
1287         case SIOCSIFCAP:
1288                 error = suser_cred(cred, 0);
1289                 if (error)
1290                         return (error);
1291                 if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1292                         return (EINVAL);
1293                 lwkt_serialize_enter(ifp->if_serializer);
1294                 ifp->if_ioctl(ifp, cmd, data, cred);
1295                 lwkt_serialize_exit(ifp->if_serializer);
1296                 break;
1297
1298         case SIOCSIFNAME:
1299                 error = suser_cred(cred, 0);
1300                 if (error != 0)
1301                         return (error);
1302                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1303                 if (error != 0)
1304                         return (error);
1305                 if (new_name[0] == '\0')
1306                         return (EINVAL);
1307                 if (ifunit(new_name) != NULL)
1308                         return (EEXIST);
1309
1310                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1311
1312                 /* Announce the departure of the interface. */
1313                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1314
1315                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1316                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1317                 /* XXX IFA_LOCK(ifa); */
1318                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1319                 namelen = strlen(new_name);
1320                 onamelen = sdl->sdl_nlen;
1321                 /*
1322                  * Move the address if needed.  This is safe because we
1323                  * allocate space for a name of length IFNAMSIZ when we
1324                  * create this in if_attach().
1325                  */
1326                 if (namelen != onamelen) {
1327                         bcopy(sdl->sdl_data + onamelen,
1328                             sdl->sdl_data + namelen, sdl->sdl_alen);
1329                 }
1330                 bcopy(new_name, sdl->sdl_data, namelen);
1331                 sdl->sdl_nlen = namelen;
1332                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1333                 bzero(sdl->sdl_data, onamelen);
1334                 while (namelen != 0)
1335                         sdl->sdl_data[--namelen] = 0xff;
1336                 /* XXX IFA_UNLOCK(ifa) */
1337
1338                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1339
1340                 /* Announce the return of the interface. */
1341                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1342                 break;
1343
1344         case SIOCSIFMETRIC:
1345                 error = suser_cred(cred, 0);
1346                 if (error)
1347                         return (error);
1348                 ifp->if_metric = ifr->ifr_metric;
1349                 getmicrotime(&ifp->if_lastchange);
1350                 break;
1351
1352         case SIOCSIFPHYS:
1353                 error = suser_cred(cred, 0);
1354                 if (error)
1355                         return error;
1356                 if (!ifp->if_ioctl)
1357                         return EOPNOTSUPP;
1358                 lwkt_serialize_enter(ifp->if_serializer);
1359                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1360                 lwkt_serialize_exit(ifp->if_serializer);
1361                 if (error == 0)
1362                         getmicrotime(&ifp->if_lastchange);
1363                 return (error);
1364
1365         case SIOCSIFMTU:
1366         {
1367                 u_long oldmtu = ifp->if_mtu;
1368
1369                 error = suser_cred(cred, 0);
1370                 if (error)
1371                         return (error);
1372                 if (ifp->if_ioctl == NULL)
1373                         return (EOPNOTSUPP);
1374                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1375                         return (EINVAL);
1376                 lwkt_serialize_enter(ifp->if_serializer);
1377                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1378                 lwkt_serialize_exit(ifp->if_serializer);
1379                 if (error == 0) {
1380                         getmicrotime(&ifp->if_lastchange);
1381                         rt_ifmsg(ifp);
1382                 }
1383                 /*
1384                  * If the link MTU changed, do network layer specific procedure.
1385                  */
1386                 if (ifp->if_mtu != oldmtu) {
1387 #ifdef INET6
1388                         nd6_setmtu(ifp);
1389 #endif
1390                 }
1391                 return (error);
1392         }
1393
1394         case SIOCADDMULTI:
1395         case SIOCDELMULTI:
1396                 error = suser_cred(cred, 0);
1397                 if (error)
1398                         return (error);
1399
1400                 /* Don't allow group membership on non-multicast interfaces. */
1401                 if ((ifp->if_flags & IFF_MULTICAST) == 0)
1402                         return EOPNOTSUPP;
1403
1404                 /* Don't let users screw up protocols' entries. */
1405                 if (ifr->ifr_addr.sa_family != AF_LINK)
1406                         return EINVAL;
1407
1408                 if (cmd == SIOCADDMULTI) {
1409                         struct ifmultiaddr *ifma;
1410                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1411                 } else {
1412                         error = if_delmulti(ifp, &ifr->ifr_addr);
1413                 }
1414                 if (error == 0)
1415                         getmicrotime(&ifp->if_lastchange);
1416                 return error;
1417
1418         case SIOCSIFPHYADDR:
1419         case SIOCDIFPHYADDR:
1420 #ifdef INET6
1421         case SIOCSIFPHYADDR_IN6:
1422 #endif
1423         case SIOCSLIFPHYADDR:
1424         case SIOCSIFMEDIA:
1425         case SIOCSIFGENERIC:
1426                 error = suser_cred(cred, 0);
1427                 if (error)
1428                         return (error);
1429                 if (ifp->if_ioctl == 0)
1430                         return (EOPNOTSUPP);
1431                 lwkt_serialize_enter(ifp->if_serializer);
1432                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1433                 lwkt_serialize_exit(ifp->if_serializer);
1434                 if (error == 0)
1435                         getmicrotime(&ifp->if_lastchange);
1436                 return error;
1437
1438         case SIOCGIFSTATUS:
1439                 ifs = (struct ifstat *)data;
1440                 ifs->ascii[0] = '\0';
1441
1442         case SIOCGIFPSRCADDR:
1443         case SIOCGIFPDSTADDR:
1444         case SIOCGLIFPHYADDR:
1445         case SIOCGIFMEDIA:
1446         case SIOCGIFGENERIC:
1447                 if (ifp->if_ioctl == NULL)
1448                         return (EOPNOTSUPP);
1449                 lwkt_serialize_enter(ifp->if_serializer);
1450                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1451                 lwkt_serialize_exit(ifp->if_serializer);
1452                 return (error);
1453
1454         case SIOCSIFLLADDR:
1455                 error = suser_cred(cred, 0);
1456                 if (error)
1457                         return (error);
1458                 return if_setlladdr(ifp,
1459                     ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1460
1461         default:
1462                 oif_flags = ifp->if_flags;
1463                 if (so->so_proto == 0)
1464                         return (EOPNOTSUPP);
1465 #ifndef COMPAT_43
1466                 error = so_pru_control(so, cmd, data, ifp);
1467 #else
1468             {
1469                 int ocmd = cmd;
1470
1471                 switch (cmd) {
1472
1473                 case SIOCSIFDSTADDR:
1474                 case SIOCSIFADDR:
1475                 case SIOCSIFBRDADDR:
1476                 case SIOCSIFNETMASK:
1477 #if BYTE_ORDER != BIG_ENDIAN
1478                         if (ifr->ifr_addr.sa_family == 0 &&
1479                             ifr->ifr_addr.sa_len < 16) {
1480                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1481                                 ifr->ifr_addr.sa_len = 16;
1482                         }
1483 #else
1484                         if (ifr->ifr_addr.sa_len == 0)
1485                                 ifr->ifr_addr.sa_len = 16;
1486 #endif
1487                         break;
1488
1489                 case OSIOCGIFADDR:
1490                         cmd = SIOCGIFADDR;
1491                         break;
1492
1493                 case OSIOCGIFDSTADDR:
1494                         cmd = SIOCGIFDSTADDR;
1495                         break;
1496
1497                 case OSIOCGIFBRDADDR:
1498                         cmd = SIOCGIFBRDADDR;
1499                         break;
1500
1501                 case OSIOCGIFNETMASK:
1502                         cmd = SIOCGIFNETMASK;
1503                 }
1504                 error =  so_pru_control(so, cmd, data, ifp);
1505                 switch (ocmd) {
1506
1507                 case OSIOCGIFADDR:
1508                 case OSIOCGIFDSTADDR:
1509                 case OSIOCGIFBRDADDR:
1510                 case OSIOCGIFNETMASK:
1511                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1512
1513                 }
1514             }
1515 #endif /* COMPAT_43 */
1516
1517                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1518 #ifdef INET6
1519                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
1520                         if (ifp->if_flags & IFF_UP) {
1521                                 crit_enter();
1522                                 in6_if_up(ifp);
1523                                 crit_exit();
1524                         }
1525 #endif
1526                 }
1527                 return (error);
1528
1529         }
1530         return (0);
1531 }
1532
1533 /*
1534  * Set/clear promiscuous mode on interface ifp based on the truth value
1535  * of pswitch.  The calls are reference counted so that only the first
1536  * "on" request actually has an effect, as does the final "off" request.
1537  * Results are undefined if the "off" and "on" requests are not matched.
1538  */
1539 int
1540 ifpromisc(struct ifnet *ifp, int pswitch)
1541 {
1542         struct ifreq ifr;
1543         int error;
1544         int oldflags;
1545
1546         oldflags = ifp->if_flags;
1547         if (ifp->if_flags & IFF_PPROMISC) {
1548                 /* Do nothing if device is in permanently promiscuous mode */
1549                 ifp->if_pcount += pswitch ? 1 : -1;
1550                 return (0);
1551         }
1552         if (pswitch) {
1553                 /*
1554                  * If the device is not configured up, we cannot put it in
1555                  * promiscuous mode.
1556                  */
1557                 if ((ifp->if_flags & IFF_UP) == 0)
1558                         return (ENETDOWN);
1559                 if (ifp->if_pcount++ != 0)
1560                         return (0);
1561                 ifp->if_flags |= IFF_PROMISC;
1562                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1563                     ifp->if_xname);
1564         } else {
1565                 if (--ifp->if_pcount > 0)
1566                         return (0);
1567                 ifp->if_flags &= ~IFF_PROMISC;
1568                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
1569                     ifp->if_xname);
1570         }
1571         ifr.ifr_flags = ifp->if_flags;
1572         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1573         lwkt_serialize_enter(ifp->if_serializer);
1574         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1575                                  (struct ucred *)NULL);
1576         lwkt_serialize_exit(ifp->if_serializer);
1577         if (error == 0)
1578                 rt_ifmsg(ifp);
1579         else
1580                 ifp->if_flags = oldflags;
1581         return error;
1582 }
1583
1584 /*
1585  * Return interface configuration
1586  * of system.  List may be used
1587  * in later ioctl's (above) to get
1588  * other information.
1589  */
1590 static int
1591 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
1592 {
1593         struct ifconf *ifc = (struct ifconf *)data;
1594         struct ifnet *ifp;
1595         struct sockaddr *sa;
1596         struct ifreq ifr, *ifrp;
1597         int space = ifc->ifc_len, error = 0;
1598
1599         ifrp = ifc->ifc_req;
1600         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1601                 struct ifaddr_container *ifac;
1602                 int addrs;
1603
1604                 if (space <= sizeof ifr)
1605                         break;
1606
1607                 /*
1608                  * Zero the stack declared structure first to prevent
1609                  * memory disclosure.
1610                  */
1611                 bzero(&ifr, sizeof(ifr));
1612                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
1613                     >= sizeof(ifr.ifr_name)) {
1614                         error = ENAMETOOLONG;
1615                         break;
1616                 }
1617
1618                 addrs = 0;
1619                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1620                         struct ifaddr *ifa = ifac->ifa;
1621
1622                         if (space <= sizeof ifr)
1623                                 break;
1624                         sa = ifa->ifa_addr;
1625                         if (cred->cr_prison &&
1626                             prison_if(cred, sa))
1627                                 continue;
1628                         addrs++;
1629 #ifdef COMPAT_43
1630                         if (cmd == OSIOCGIFCONF) {
1631                                 struct osockaddr *osa =
1632                                          (struct osockaddr *)&ifr.ifr_addr;
1633                                 ifr.ifr_addr = *sa;
1634                                 osa->sa_family = sa->sa_family;
1635                                 error = copyout(&ifr, ifrp, sizeof ifr);
1636                                 ifrp++;
1637                         } else
1638 #endif
1639                         if (sa->sa_len <= sizeof(*sa)) {
1640                                 ifr.ifr_addr = *sa;
1641                                 error = copyout(&ifr, ifrp, sizeof ifr);
1642                                 ifrp++;
1643                         } else {
1644                                 if (space < (sizeof ifr) + sa->sa_len -
1645                                             sizeof(*sa))
1646                                         break;
1647                                 space -= sa->sa_len - sizeof(*sa);
1648                                 error = copyout(&ifr, ifrp,
1649                                                 sizeof ifr.ifr_name);
1650                                 if (error == 0)
1651                                         error = copyout(sa, &ifrp->ifr_addr,
1652                                                         sa->sa_len);
1653                                 ifrp = (struct ifreq *)
1654                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
1655                         }
1656                         if (error)
1657                                 break;
1658                         space -= sizeof ifr;
1659                 }
1660                 if (error)
1661                         break;
1662                 if (!addrs) {
1663                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
1664                         error = copyout(&ifr, ifrp, sizeof ifr);
1665                         if (error)
1666                                 break;
1667                         space -= sizeof ifr;
1668                         ifrp++;
1669                 }
1670         }
1671         ifc->ifc_len -= space;
1672         return (error);
1673 }
1674
1675 /*
1676  * Just like if_promisc(), but for all-multicast-reception mode.
1677  */
1678 int
1679 if_allmulti(struct ifnet *ifp, int onswitch)
1680 {
1681         int error = 0;
1682         struct ifreq ifr;
1683
1684         crit_enter();
1685
1686         if (onswitch) {
1687                 if (ifp->if_amcount++ == 0) {
1688                         ifp->if_flags |= IFF_ALLMULTI;
1689                         ifr.ifr_flags = ifp->if_flags;
1690                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1691                         lwkt_serialize_enter(ifp->if_serializer);
1692                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1693                                               (struct ucred *)NULL);
1694                         lwkt_serialize_exit(ifp->if_serializer);
1695                 }
1696         } else {
1697                 if (ifp->if_amcount > 1) {
1698                         ifp->if_amcount--;
1699                 } else {
1700                         ifp->if_amcount = 0;
1701                         ifp->if_flags &= ~IFF_ALLMULTI;
1702                         ifr.ifr_flags = ifp->if_flags;
1703                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1704                         lwkt_serialize_enter(ifp->if_serializer);
1705                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1706                                               (struct ucred *)NULL);
1707                         lwkt_serialize_exit(ifp->if_serializer);
1708                 }
1709         }
1710
1711         crit_exit();
1712
1713         if (error == 0)
1714                 rt_ifmsg(ifp);
1715         return error;
1716 }
1717
1718 /*
1719  * Add a multicast listenership to the interface in question.
1720  * The link layer provides a routine which converts
1721  */
1722 int
1723 if_addmulti(
1724         struct ifnet *ifp,      /* interface to manipulate */
1725         struct sockaddr *sa,    /* address to add */
1726         struct ifmultiaddr **retifma)
1727 {
1728         struct sockaddr *llsa, *dupsa;
1729         int error;
1730         struct ifmultiaddr *ifma;
1731
1732         /*
1733          * If the matching multicast address already exists
1734          * then don't add a new one, just add a reference
1735          */
1736         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1737                 if (sa_equal(sa, ifma->ifma_addr)) {
1738                         ifma->ifma_refcount++;
1739                         if (retifma)
1740                                 *retifma = ifma;
1741                         return 0;
1742                 }
1743         }
1744
1745         /*
1746          * Give the link layer a chance to accept/reject it, and also
1747          * find out which AF_LINK address this maps to, if it isn't one
1748          * already.
1749          */
1750         if (ifp->if_resolvemulti) {
1751                 lwkt_serialize_enter(ifp->if_serializer);
1752                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
1753                 lwkt_serialize_exit(ifp->if_serializer);
1754                 if (error) 
1755                         return error;
1756         } else {
1757                 llsa = 0;
1758         }
1759
1760         MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, M_WAITOK);
1761         MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, M_WAITOK);
1762         bcopy(sa, dupsa, sa->sa_len);
1763
1764         ifma->ifma_addr = dupsa;
1765         ifma->ifma_lladdr = llsa;
1766         ifma->ifma_ifp = ifp;
1767         ifma->ifma_refcount = 1;
1768         ifma->ifma_protospec = 0;
1769         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
1770
1771         /*
1772          * Some network interfaces can scan the address list at
1773          * interrupt time; lock them out.
1774          */
1775         crit_enter();
1776         LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1777         crit_exit();
1778         *retifma = ifma;
1779
1780         if (llsa != 0) {
1781                 LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1782                         if (sa_equal(ifma->ifma_addr, llsa))
1783                                 break;
1784                 }
1785                 if (ifma) {
1786                         ifma->ifma_refcount++;
1787                 } else {
1788                         MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma,
1789                                M_IFMADDR, M_WAITOK);
1790                         MALLOC(dupsa, struct sockaddr *, llsa->sa_len,
1791                                M_IFMADDR, M_WAITOK);
1792                         bcopy(llsa, dupsa, llsa->sa_len);
1793                         ifma->ifma_addr = dupsa;
1794                         ifma->ifma_ifp = ifp;
1795                         ifma->ifma_refcount = 1;
1796                         crit_enter();
1797                         LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1798                         crit_exit();
1799                 }
1800         }
1801         /*
1802          * We are certain we have added something, so call down to the
1803          * interface to let them know about it.
1804          */
1805         crit_enter();
1806         lwkt_serialize_enter(ifp->if_serializer);
1807         ifp->if_ioctl(ifp, SIOCADDMULTI, 0, (struct ucred *)NULL);
1808         lwkt_serialize_exit(ifp->if_serializer);
1809         crit_exit();
1810
1811         return 0;
1812 }
1813
1814 /*
1815  * Remove a reference to a multicast address on this interface.  Yell
1816  * if the request does not match an existing membership.
1817  */
1818 int
1819 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
1820 {
1821         struct ifmultiaddr *ifma;
1822
1823         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1824                 if (sa_equal(sa, ifma->ifma_addr))
1825                         break;
1826         if (ifma == 0)
1827                 return ENOENT;
1828
1829         if (ifma->ifma_refcount > 1) {
1830                 ifma->ifma_refcount--;
1831                 return 0;
1832         }
1833
1834         rt_newmaddrmsg(RTM_DELMADDR, ifma);
1835         sa = ifma->ifma_lladdr;
1836         crit_enter();
1837         LIST_REMOVE(ifma, ifma_link);
1838         /*
1839          * Make sure the interface driver is notified
1840          * in the case of a link layer mcast group being left.
1841          */
1842         if (ifma->ifma_addr->sa_family == AF_LINK && sa == 0) {
1843                 lwkt_serialize_enter(ifp->if_serializer);
1844                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, (struct ucred *)NULL);
1845                 lwkt_serialize_exit(ifp->if_serializer);
1846         }
1847         crit_exit();
1848         kfree(ifma->ifma_addr, M_IFMADDR);
1849         kfree(ifma, M_IFMADDR);
1850         if (sa == 0)
1851                 return 0;
1852
1853         /*
1854          * Now look for the link-layer address which corresponds to
1855          * this network address.  It had been squirreled away in
1856          * ifma->ifma_lladdr for this purpose (so we don't have
1857          * to call ifp->if_resolvemulti() again), and we saved that
1858          * value in sa above.  If some nasty deleted the
1859          * link-layer address out from underneath us, we can deal because
1860          * the address we stored was is not the same as the one which was
1861          * in the record for the link-layer address.  (So we don't complain
1862          * in that case.)
1863          */
1864         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1865                 if (sa_equal(sa, ifma->ifma_addr))
1866                         break;
1867         if (ifma == 0)
1868                 return 0;
1869
1870         if (ifma->ifma_refcount > 1) {
1871                 ifma->ifma_refcount--;
1872                 return 0;
1873         }
1874
1875         crit_enter();
1876         lwkt_serialize_enter(ifp->if_serializer);
1877         LIST_REMOVE(ifma, ifma_link);
1878         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, (struct ucred *)NULL);
1879         lwkt_serialize_exit(ifp->if_serializer);
1880         crit_exit();
1881         kfree(ifma->ifma_addr, M_IFMADDR);
1882         kfree(sa, M_IFMADDR);
1883         kfree(ifma, M_IFMADDR);
1884
1885         return 0;
1886 }
1887
1888 /*
1889  * Set the link layer address on an interface.
1890  *
1891  * At this time we only support certain types of interfaces,
1892  * and we don't allow the length of the address to change.
1893  */
1894 int
1895 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
1896 {
1897         struct sockaddr_dl *sdl;
1898         struct ifreq ifr;
1899
1900         sdl = IF_LLSOCKADDR(ifp);
1901         if (sdl == NULL)
1902                 return (EINVAL);
1903         if (len != sdl->sdl_alen)       /* don't allow length to change */
1904                 return (EINVAL);
1905         switch (ifp->if_type) {
1906         case IFT_ETHER:                 /* these types use struct arpcom */
1907         case IFT_XETHER:
1908         case IFT_L2VLAN:
1909                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
1910                 bcopy(lladdr, LLADDR(sdl), len);
1911                 break;
1912         default:
1913                 return (ENODEV);
1914         }
1915         /*
1916          * If the interface is already up, we need
1917          * to re-init it in order to reprogram its
1918          * address filter.
1919          */
1920         lwkt_serialize_enter(ifp->if_serializer);
1921         if ((ifp->if_flags & IFF_UP) != 0) {
1922                 struct ifaddr_container *ifac;
1923
1924                 ifp->if_flags &= ~IFF_UP;
1925                 ifr.ifr_flags = ifp->if_flags;
1926                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
1927                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1928                               (struct ucred *)NULL);
1929                 ifp->if_flags |= IFF_UP;
1930                 ifr.ifr_flags = ifp->if_flags;
1931                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
1932                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1933                                  (struct ucred *)NULL);
1934 #ifdef INET
1935                 /*
1936                  * Also send gratuitous ARPs to notify other nodes about
1937                  * the address change.
1938                  */
1939                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1940                         struct ifaddr *ifa = ifac->ifa;
1941
1942                         if (ifa->ifa_addr != NULL &&
1943                             ifa->ifa_addr->sa_family == AF_INET)
1944                                 arp_ifinit(ifp, ifa);
1945                 }
1946 #endif
1947         }
1948         lwkt_serialize_exit(ifp->if_serializer);
1949         return (0);
1950 }
1951
1952 struct ifmultiaddr *
1953 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
1954 {
1955         struct ifmultiaddr *ifma;
1956
1957         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1958                 if (sa_equal(ifma->ifma_addr, sa))
1959                         break;
1960
1961         return ifma;
1962 }
1963
1964 /*
1965  * This function locates the first real ethernet MAC from a network
1966  * card and loads it into node, returning 0 on success or ENOENT if
1967  * no suitable interfaces were found.  It is used by the uuid code to
1968  * generate a unique 6-byte number.
1969  */
1970 int
1971 if_getanyethermac(uint16_t *node, int minlen)
1972 {
1973         struct ifnet *ifp;
1974         struct sockaddr_dl *sdl;
1975
1976         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1977                 if (ifp->if_type != IFT_ETHER)
1978                         continue;
1979                 sdl = IF_LLSOCKADDR(ifp);
1980                 if (sdl->sdl_alen < minlen)
1981                         continue;
1982                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
1983                       minlen);
1984                 return(0);
1985         }
1986         return (ENOENT);
1987 }
1988
1989 /*
1990  * The name argument must be a pointer to storage which will last as
1991  * long as the interface does.  For physical devices, the result of
1992  * device_get_name(dev) is a good choice and for pseudo-devices a
1993  * static string works well.
1994  */
1995 void
1996 if_initname(struct ifnet *ifp, const char *name, int unit)
1997 {
1998         ifp->if_dname = name;
1999         ifp->if_dunit = unit;
2000         if (unit != IF_DUNIT_NONE)
2001                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2002         else
2003                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2004 }
2005
2006 int
2007 if_printf(struct ifnet *ifp, const char *fmt, ...)
2008 {
2009         __va_list ap;
2010         int retval;
2011
2012         retval = kprintf("%s: ", ifp->if_xname);
2013         __va_start(ap, fmt);
2014         retval += kvprintf(fmt, ap);
2015         __va_end(ap);
2016         return (retval);
2017 }
2018
2019 void
2020 ifq_set_classic(struct ifaltq *ifq)
2021 {
2022         ifq->altq_enqueue = ifq_classic_enqueue;
2023         ifq->altq_dequeue = ifq_classic_dequeue;
2024         ifq->altq_request = ifq_classic_request;
2025 }
2026
2027 int
2028 ifq_classic_enqueue(struct ifaltq *ifq, struct mbuf *m,
2029                     struct altq_pktattr *pa __unused)
2030 {
2031         logifq(enqueue, ifq);
2032         if (IF_QFULL(ifq)) {
2033                 m_freem(m);
2034                 return(ENOBUFS);
2035         } else {
2036                 IF_ENQUEUE(ifq, m);
2037                 return(0);
2038         }       
2039 }
2040
2041 struct mbuf *
2042 ifq_classic_dequeue(struct ifaltq *ifq, struct mbuf *mpolled, int op)
2043 {
2044         struct mbuf *m;
2045
2046         switch (op) {
2047         case ALTDQ_POLL:
2048                 IF_POLL(ifq, m);
2049                 break;
2050         case ALTDQ_REMOVE:
2051                 logifq(dequeue, ifq);
2052                 IF_DEQUEUE(ifq, m);
2053                 break;
2054         default:
2055                 panic("unsupported ALTQ dequeue op: %d", op);
2056         }
2057         KKASSERT(mpolled == NULL || mpolled == m);
2058         return(m);
2059 }
2060
2061 int
2062 ifq_classic_request(struct ifaltq *ifq, int req, void *arg)
2063 {
2064         switch (req) {
2065         case ALTRQ_PURGE:
2066                 IF_DRAIN(ifq);
2067                 break;
2068         default:
2069                 panic("unsupported ALTQ request: %d", req);
2070         }
2071         return(0);
2072 }
2073
2074 int
2075 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2076 {
2077         struct ifaltq *ifq = &ifp->if_snd;
2078         int running = 0, error, start = 0;
2079
2080         ASSERT_NOT_SERIALIZED(ifp->if_serializer);
2081
2082         ALTQ_LOCK(ifq);
2083         error = ifq_enqueue_locked(ifq, m, pa);
2084         if (error) {
2085                 ALTQ_UNLOCK(ifq);
2086                 return error;
2087         }
2088         if (!ifq->altq_started) {
2089                 /*
2090                  * Hold the interlock of ifnet.if_start
2091                  */
2092                 ifq->altq_started = 1;
2093                 start = 1;
2094         }
2095         ALTQ_UNLOCK(ifq);
2096
2097         ifp->if_obytes += m->m_pkthdr.len;
2098         if (m->m_flags & M_MCAST)
2099                 ifp->if_omcasts++;
2100
2101         if (!start) {
2102                 logifstart(avoid, ifp);
2103                 return 0;
2104         }
2105
2106         if (ifq_dispatch_schedonly) {
2107                 /*
2108                  * Always schedule ifnet.if_start on ifnet's CPU,
2109                  * short circuit the rest of this function.
2110                  */
2111                 logifstart(sched, ifp);
2112                 if_start_schedule(ifp);
2113                 return 0;
2114         }
2115
2116         /*
2117          * Try to do direct ifnet.if_start first, if there is
2118          * contention on ifnet's serializer, ifnet.if_start will
2119          * be scheduled on ifnet's CPU.
2120          */
2121         if (!lwkt_serialize_try(ifp->if_serializer)) {
2122                 /*
2123                  * ifnet serializer contention happened,
2124                  * ifnet.if_start is scheduled on ifnet's
2125                  * CPU, and we keep going.
2126                  */
2127                 logifstart(contend_sched, ifp);
2128                 if_start_schedule(ifp);
2129                 return 0;
2130         }
2131
2132         if ((ifp->if_flags & IFF_OACTIVE) == 0) {
2133                 logifstart(run, ifp);
2134                 ifp->if_start(ifp);
2135                 if ((ifp->if_flags &
2136                      (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
2137                         running = 1;
2138         }
2139
2140         lwkt_serialize_exit(ifp->if_serializer);
2141
2142         if (ifq_dispatch_schednochk || if_start_need_schedule(ifq, running)) {
2143                 /*
2144                  * More data need to be transmitted, ifnet.if_start is
2145                  * scheduled on ifnet's CPU, and we keep going.
2146                  * NOTE: ifnet.if_start interlock is not released.
2147                  */
2148                 logifstart(sched, ifp);
2149                 if_start_schedule(ifp);
2150         }
2151         return 0;
2152 }
2153
2154 void *
2155 ifa_create(int size, int flags)
2156 {
2157         struct ifaddr *ifa;
2158         int i;
2159
2160         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small\n"));
2161
2162         ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2163         if (ifa == NULL)
2164                 return NULL;
2165
2166         ifa->ifa_containers = kmalloc(ncpus * sizeof(struct ifaddr_container),
2167                                       M_IFADDR, M_WAITOK | M_ZERO);
2168         ifa->ifa_cpumask = smp_active_mask;
2169         for (i = 0; i < ncpus; ++i) {
2170                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2171
2172                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2173                 ifac->ifa = ifa;
2174                 ifac->ifa_refcnt = 1;
2175         }
2176 #ifdef IFADDR_DEBUG
2177         kprintf("alloc ifa %p %d\n", ifa, size);
2178 #endif
2179         return ifa;
2180 }
2181
2182 static void
2183 ifac_free_dispatch(struct netmsg *nmsg)
2184 {
2185         struct netmsg_ifaddr_free *fmsg = (struct netmsg_ifaddr_free *)nmsg;
2186         struct ifaddr *ifa = fmsg->nm_ifaddr;
2187
2188         KKASSERT(ifa->ifa_cpumask & (1 << fmsg->nm_cpuid));
2189         ifa->ifa_cpumask &= ~(1 << fmsg->nm_cpuid);
2190         if (ifa->ifa_cpumask == 0) {
2191 #ifdef IFADDR_DEBUG
2192                 kprintf("free ifa %p\n", ifa);
2193 #endif
2194                 kfree(ifa->ifa_containers, M_IFADDR);
2195                 kfree(ifa, M_IFADDR);
2196         }
2197         /* Don't reply, 'nmsg' is embedded in ifaddr_container */
2198 }
2199
2200 void
2201 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2202 {
2203         struct netmsg_ifaddr_free *fmsg;
2204
2205         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2206         KKASSERT(ifac->ifa_refcnt == 0);
2207         KASSERT(ifac->ifa_listmask == 0,
2208                 ("ifa is still on %#x lists\n", ifac->ifa_listmask));
2209
2210         ifac->ifa_magic = IFA_CONTAINER_DEAD;
2211
2212 #ifdef IFADDR_DEBUG_VERBOSE
2213         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2214 #endif
2215
2216         fmsg = &ifac->ifa_freemsg;
2217         netmsg_init(&fmsg->nm_netmsg, &netisr_apanic_rport, 0,
2218                     ifac_free_dispatch);
2219         fmsg->nm_ifaddr = ifac->ifa;
2220         fmsg->nm_cpuid = cpu_id;
2221
2222         ifa_sendmsg(&fmsg->nm_netmsg.nm_lmsg, 0);
2223 }
2224
2225 static void
2226 ifa_iflink_dispatch(struct netmsg *nmsg)
2227 {
2228         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2229         struct ifaddr *ifa = msg->ifa;
2230         struct ifnet *ifp = msg->ifp;
2231         int cpu = mycpuid;
2232         struct ifaddr_container *ifac;
2233
2234         crit_enter();
2235
2236         ifac = &ifa->ifa_containers[cpu];
2237         ASSERT_IFAC_VALID(ifac);
2238         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2239                 ("ifaddr is on if_addrheads\n"));
2240
2241         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2242         if (msg->tail)
2243                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2244         else
2245                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2246
2247         crit_exit();
2248
2249         ifa_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
2250 }
2251
2252 void
2253 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2254 {
2255         struct netmsg_ifaddr msg;
2256
2257         netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2258                     ifa_iflink_dispatch);
2259         msg.ifa = ifa;
2260         msg.ifp = ifp;
2261         msg.tail = tail;
2262
2263         ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2264 }
2265
2266 static void
2267 ifa_ifunlink_dispatch(struct netmsg *nmsg)
2268 {
2269         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2270         struct ifaddr *ifa = msg->ifa;
2271         struct ifnet *ifp = msg->ifp;
2272         int cpu = mycpuid;
2273         struct ifaddr_container *ifac;
2274
2275         crit_enter();
2276
2277         ifac = &ifa->ifa_containers[cpu];
2278         ASSERT_IFAC_VALID(ifac);
2279         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2280                 ("ifaddr is not on if_addrhead\n"));
2281
2282         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2283         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2284
2285         crit_exit();
2286
2287         ifa_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
2288 }
2289
2290 void
2291 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2292 {
2293         struct netmsg_ifaddr msg;
2294
2295         netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2296                     ifa_ifunlink_dispatch);
2297         msg.ifa = ifa;
2298         msg.ifp = ifp;
2299
2300         ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2301 }
2302
2303 static void
2304 ifa_destroy_dispatch(struct netmsg *nmsg)
2305 {
2306         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2307
2308         IFAFREE(msg->ifa);
2309         ifa_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2310 }
2311
2312 void
2313 ifa_destroy(struct ifaddr *ifa)
2314 {
2315         struct netmsg_ifaddr msg;
2316
2317         netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
2318                     ifa_destroy_dispatch);
2319         msg.ifa = ifa;
2320
2321         ifa_domsg(&msg.netmsg.nm_lmsg, 0);
2322 }
2323
2324 struct lwkt_port *
2325 ifnet_portfn(int cpu)
2326 {
2327         return &ifnet_threads[cpu].td_msgport;
2328 }
2329
2330 void
2331 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2332 {
2333         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2334
2335         if (next_cpu < ncpus)
2336                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2337         else
2338                 lwkt_replymsg(lmsg, 0);
2339 }
2340
2341 void
2342 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
2343 {
2344         KKASSERT(cpu < ncpus);
2345         lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
2346 }
2347
2348 void
2349 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
2350 {
2351         KKASSERT(cpu < ncpus);
2352         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
2353 }
2354
2355 static void
2356 ifnetinit(void *dummy __unused)
2357 {
2358         int i;
2359
2360         for (i = 0; i < ncpus; ++i) {
2361                 struct thread *thr = &ifnet_threads[i];
2362
2363                 lwkt_create(netmsg_service_loop_mpsafe, NULL, NULL, thr, 0, i,
2364                             "ifnet %d", i);
2365                 netmsg_service_port_init(&thr->td_msgport);
2366         }
2367 }