if: Add power of 2 mask based CPUID to subqueue mapping
[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  */
36
37 #include "opt_compat.h"
38 #include "opt_inet6.h"
39 #include "opt_inet.h"
40 #include "opt_ifpoll.h"
41
42 #include <sys/param.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/systm.h>
46 #include <sys/proc.h>
47 #include <sys/priv.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/mutex.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/serialize.h>
62 #include <sys/bus.h>
63
64 #include <sys/thread2.h>
65 #include <sys/msgport2.h>
66 #include <sys/mutex2.h>
67
68 #include <net/if.h>
69 #include <net/if_arp.h>
70 #include <net/if_dl.h>
71 #include <net/if_types.h>
72 #include <net/if_var.h>
73 #include <net/ifq_var.h>
74 #include <net/radix.h>
75 #include <net/route.h>
76 #include <net/if_clone.h>
77 #include <net/netisr.h>
78 #include <net/netmsg2.h>
79
80 #include <machine/atomic.h>
81 #include <machine/stdarg.h>
82 #include <machine/smp.h>
83
84 #if defined(INET) || defined(INET6)
85 /*XXX*/
86 #include <netinet/in.h>
87 #include <netinet/in_var.h>
88 #include <netinet/if_ether.h>
89 #ifdef INET6
90 #include <netinet6/in6_var.h>
91 #include <netinet6/in6_ifattach.h>
92 #endif
93 #endif
94
95 #if defined(COMPAT_43)
96 #include <emulation/43bsd/43bsd_socket.h>
97 #endif /* COMPAT_43 */
98
99 struct netmsg_ifaddr {
100         struct netmsg_base base;
101         struct ifaddr   *ifa;
102         struct ifnet    *ifp;
103         int             tail;
104 };
105
106 struct ifsubq_stage_head {
107         TAILQ_HEAD(, ifsubq_stage)      stg_head;
108 } __cachealign;
109
110 /*
111  * System initialization
112  */
113 static void     if_attachdomain(void *);
114 static void     if_attachdomain1(struct ifnet *);
115 static int      ifconf(u_long, caddr_t, struct ucred *);
116 static void     ifinit(void *);
117 static void     ifnetinit(void *);
118 static void     if_slowtimo(void *);
119 static void     link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
120 static int      if_rtdel(struct radix_node *, void *);
121
122 /* Helper functions */
123 static void     ifsq_watchdog_reset(struct ifsubq_watchdog *);
124
125 #ifdef INET6
126 /*
127  * XXX: declare here to avoid to include many inet6 related files..
128  * should be more generalized?
129  */
130 extern void     nd6_setmtu(struct ifnet *);
131 #endif
132
133 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
134 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
135
136 static int ifsq_stage_cntmax = 4;
137 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
138 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
139     &ifsq_stage_cntmax, 0, "ifq staging packet count max");
140
141 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
142 /* Must be after netisr_init */
143 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
144
145 static  if_com_alloc_t *if_com_alloc[256];
146 static  if_com_free_t *if_com_free[256];
147
148 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
149 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
150 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
151
152 int                     ifqmaxlen = IFQ_MAXLEN;
153 struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
154
155 struct callout          if_slowtimo_timer;
156
157 int                     if_index = 0;
158 struct ifnet            **ifindex2ifnet = NULL;
159 static struct thread    ifnet_threads[MAXCPU];
160
161 static struct ifsubq_stage_head ifsubq_stage_heads[MAXCPU];
162
163 #ifdef notyet
164 #define IFQ_KTR_STRING          "ifq=%p"
165 #define IFQ_KTR_ARGS    struct ifaltq *ifq
166 #ifndef KTR_IFQ
167 #define KTR_IFQ                 KTR_ALL
168 #endif
169 KTR_INFO_MASTER(ifq);
170 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
171 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
172 #define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)
173
174 #define IF_START_KTR_STRING     "ifp=%p"
175 #define IF_START_KTR_ARGS       struct ifnet *ifp
176 #ifndef KTR_IF_START
177 #define KTR_IF_START            KTR_ALL
178 #endif
179 KTR_INFO_MASTER(if_start);
180 KTR_INFO(KTR_IF_START, if_start, run, 0,
181          IF_START_KTR_STRING, IF_START_KTR_ARGS);
182 KTR_INFO(KTR_IF_START, if_start, sched, 1,
183          IF_START_KTR_STRING, IF_START_KTR_ARGS);
184 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
185          IF_START_KTR_STRING, IF_START_KTR_ARGS);
186 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
187          IF_START_KTR_STRING, IF_START_KTR_ARGS);
188 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
189          IF_START_KTR_STRING, IF_START_KTR_ARGS);
190 #define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
191 #endif
192
193 TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
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.altq_maxlen == 0) {
212                         if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
213                         ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
214                 }
215         }
216         crit_exit();
217
218         if_slowtimo(0);
219 }
220
221 static void
222 ifsq_ifstart_ipifunc(void *arg)
223 {
224         struct ifaltq_subque *ifsq = arg;
225         struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
226
227         crit_enter();
228         if (lmsg->ms_flags & MSGF_DONE)
229                 lwkt_sendmsg(netisr_portfn(mycpuid), lmsg);
230         crit_exit();
231 }
232
233 static __inline void
234 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
235 {
236         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
237         TAILQ_REMOVE(&head->stg_head, stage, stg_link);
238         stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
239         stage->stg_cnt = 0;
240         stage->stg_len = 0;
241 }
242
243 static __inline void
244 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
245 {
246         KKASSERT((stage->stg_flags &
247             (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
248         stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
249         TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
250 }
251
252 /*
253  * Schedule ifnet.if_start on ifnet's CPU
254  */
255 static void
256 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
257 {
258         int cpu;
259
260         if (!force && curthread->td_type == TD_TYPE_NETISR &&
261             ifsq_stage_cntmax > 0) {
262                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
263
264                 stage->stg_cnt = 0;
265                 stage->stg_len = 0;
266                 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
267                         ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
268                 stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
269                 return;
270         }
271
272         cpu = ifsq_get_cpuid(ifsq);
273         if (cpu != mycpuid)
274                 lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
275         else
276                 ifsq_ifstart_ipifunc(ifsq);
277 }
278
279 /*
280  * NOTE:
281  * This function will release ifnet.if_start interlock,
282  * if ifnet.if_start does not need to be scheduled
283  */
284 static __inline int
285 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
286 {
287         if (!running || ifsq_is_empty(ifsq)
288 #ifdef ALTQ
289             || ifsq->ifsq_altq->altq_tbr != NULL
290 #endif
291         ) {
292                 ALTQ_SQ_LOCK(ifsq);
293                 /*
294                  * ifnet.if_start interlock is released, if:
295                  * 1) Hardware can not take any packets, due to
296                  *    o  interface is marked down
297                  *    o  hardware queue is full (ifq_is_oactive)
298                  *    Under the second situation, hardware interrupt
299                  *    or polling(4) will call/schedule ifnet.if_start
300                  *    when hardware queue is ready
301                  * 2) There is not packet in the ifnet.if_snd.
302                  *    Further ifq_dispatch or ifq_handoff will call/
303                  *    schedule ifnet.if_start
304                  * 3) TBR is used and it does not allow further
305                  *    dequeueing.
306                  *    TBR callout will call ifnet.if_start
307                  */
308                 if (!running || !ifsq_data_ready(ifsq)) {
309                         ifsq_clr_started(ifsq);
310                         ALTQ_SQ_UNLOCK(ifsq);
311                         return 0;
312                 }
313                 ALTQ_SQ_UNLOCK(ifsq);
314         }
315         return 1;
316 }
317
318 static void
319 ifsq_ifstart_dispatch(netmsg_t msg)
320 {
321         struct lwkt_msg *lmsg = &msg->base.lmsg;
322         struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
323         struct ifnet *ifp = ifsq_get_ifp(ifsq);
324         int running = 0, need_sched;
325
326         crit_enter();
327         lwkt_replymsg(lmsg, 0); /* reply ASAP */
328         crit_exit();
329
330         if (mycpuid != ifsq_get_cpuid(ifsq)) {
331                 /*
332                  * We need to chase the ifnet CPU change.
333                  */
334                 ifsq_ifstart_schedule(ifsq, 1);
335                 return;
336         }
337
338         ifnet_serialize_tx(ifp, ifsq);
339         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
340                 ifp->if_start(ifp, ifsq);
341                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
342                         running = 1;
343         }
344         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
345         ifnet_deserialize_tx(ifp, ifsq);
346
347         if (need_sched) {
348                 /*
349                  * More data need to be transmitted, ifnet.if_start is
350                  * scheduled on ifnet's CPU, and we keep going.
351                  * NOTE: ifnet.if_start interlock is not released.
352                  */
353                 ifsq_ifstart_schedule(ifsq, 0);
354         }
355 }
356
357 /* Device driver ifnet.if_start helper function */
358 void
359 ifsq_devstart(struct ifaltq_subque *ifsq)
360 {
361         struct ifnet *ifp = ifsq_get_ifp(ifsq);
362         int running = 0;
363
364         ASSERT_IFNET_SERIALIZED_TX(ifp, ifsq);
365
366         ALTQ_SQ_LOCK(ifsq);
367         if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
368                 ALTQ_SQ_UNLOCK(ifsq);
369                 return;
370         }
371         ifsq_set_started(ifsq);
372         ALTQ_SQ_UNLOCK(ifsq);
373
374         ifp->if_start(ifp, ifsq);
375
376         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
377                 running = 1;
378
379         if (ifsq_ifstart_need_schedule(ifsq, running)) {
380                 /*
381                  * More data need to be transmitted, ifnet.if_start is
382                  * scheduled on ifnet's CPU, and we keep going.
383                  * NOTE: ifnet.if_start interlock is not released.
384                  */
385                 ifsq_ifstart_schedule(ifsq, 0);
386         }
387 }
388
389 void
390 if_devstart(struct ifnet *ifp)
391 {
392         ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
393 }
394
395 /* Device driver ifnet.if_start schedule helper function */
396 void
397 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
398 {
399         ifsq_ifstart_schedule(ifsq, 1);
400 }
401
402 void
403 if_devstart_sched(struct ifnet *ifp)
404 {
405         ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
406 }
407
408 static void
409 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
410 {
411         lwkt_serialize_enter(ifp->if_serializer);
412 }
413
414 static void
415 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
416 {
417         lwkt_serialize_exit(ifp->if_serializer);
418 }
419
420 static int
421 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
422 {
423         return lwkt_serialize_try(ifp->if_serializer);
424 }
425
426 #ifdef INVARIANTS
427 static void
428 if_default_serialize_assert(struct ifnet *ifp,
429                             enum ifnet_serialize slz __unused,
430                             boolean_t serialized)
431 {
432         if (serialized)
433                 ASSERT_SERIALIZED(ifp->if_serializer);
434         else
435                 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
436 }
437 #endif
438
439 /*
440  * Attach an interface to the list of "active" interfaces.
441  *
442  * The serializer is optional.  If non-NULL access to the interface
443  * may be MPSAFE.
444  */
445 void
446 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
447 {
448         unsigned socksize, ifasize;
449         int namelen, masklen;
450         struct sockaddr_dl *sdl;
451         struct ifaddr *ifa;
452         struct ifaltq *ifq;
453         int i, q;
454
455         static int if_indexlim = 8;
456
457         if (ifp->if_serialize != NULL) {
458                 KASSERT(ifp->if_deserialize != NULL &&
459                         ifp->if_tryserialize != NULL &&
460                         ifp->if_serialize_assert != NULL,
461                         ("serialize functions are partially setup"));
462
463                 /*
464                  * If the device supplies serialize functions,
465                  * then clear if_serializer to catch any invalid
466                  * usage of this field.
467                  */
468                 KASSERT(serializer == NULL,
469                         ("both serialize functions and default serializer "
470                          "are supplied"));
471                 ifp->if_serializer = NULL;
472         } else {
473                 KASSERT(ifp->if_deserialize == NULL &&
474                         ifp->if_tryserialize == NULL &&
475                         ifp->if_serialize_assert == NULL,
476                         ("serialize functions are partially setup"));
477                 ifp->if_serialize = if_default_serialize;
478                 ifp->if_deserialize = if_default_deserialize;
479                 ifp->if_tryserialize = if_default_tryserialize;
480 #ifdef INVARIANTS
481                 ifp->if_serialize_assert = if_default_serialize_assert;
482 #endif
483
484                 /*
485                  * The serializer can be passed in from the device,
486                  * allowing the same serializer to be used for both
487                  * the interrupt interlock and the device queue.
488                  * If not specified, the netif structure will use an
489                  * embedded serializer.
490                  */
491                 if (serializer == NULL) {
492                         serializer = &ifp->if_default_serializer;
493                         lwkt_serialize_init(serializer);
494                 }
495                 ifp->if_serializer = serializer;
496         }
497
498         mtx_init(&ifp->if_ioctl_mtx);
499         mtx_lock(&ifp->if_ioctl_mtx);
500
501         TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
502         ifp->if_index = ++if_index;
503
504         /*
505          * XXX -
506          * The old code would work if the interface passed a pre-existing
507          * chain of ifaddrs to this code.  We don't trust our callers to
508          * properly initialize the tailq, however, so we no longer allow
509          * this unlikely case.
510          */
511         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
512                                     M_IFADDR, M_WAITOK | M_ZERO);
513         for (i = 0; i < ncpus; ++i)
514                 TAILQ_INIT(&ifp->if_addrheads[i]);
515
516         TAILQ_INIT(&ifp->if_prefixhead);
517         TAILQ_INIT(&ifp->if_multiaddrs);
518         TAILQ_INIT(&ifp->if_groups);
519         getmicrotime(&ifp->if_lastchange);
520         if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
521                 unsigned int n;
522                 struct ifnet **q;
523
524                 if_indexlim <<= 1;
525
526                 /* grow ifindex2ifnet */
527                 n = if_indexlim * sizeof(*q);
528                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
529                 if (ifindex2ifnet) {
530                         bcopy(ifindex2ifnet, q, n/2);
531                         kfree(ifindex2ifnet, M_IFADDR);
532                 }
533                 ifindex2ifnet = q;
534         }
535
536         ifindex2ifnet[if_index] = ifp;
537
538         /*
539          * create a Link Level name for this device
540          */
541         namelen = strlen(ifp->if_xname);
542         masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
543         socksize = masklen + ifp->if_addrlen;
544 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
545         if (socksize < sizeof(*sdl))
546                 socksize = sizeof(*sdl);
547         socksize = ROUNDUP(socksize);
548 #undef ROUNDUP
549         ifasize = sizeof(struct ifaddr) + 2 * socksize;
550         ifa = ifa_create(ifasize, M_WAITOK);
551         sdl = (struct sockaddr_dl *)(ifa + 1);
552         sdl->sdl_len = socksize;
553         sdl->sdl_family = AF_LINK;
554         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
555         sdl->sdl_nlen = namelen;
556         sdl->sdl_index = ifp->if_index;
557         sdl->sdl_type = ifp->if_type;
558         ifp->if_lladdr = ifa;
559         ifa->ifa_ifp = ifp;
560         ifa->ifa_rtrequest = link_rtrequest;
561         ifa->ifa_addr = (struct sockaddr *)sdl;
562         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
563         ifa->ifa_netmask = (struct sockaddr *)sdl;
564         sdl->sdl_len = masklen;
565         while (namelen != 0)
566                 sdl->sdl_data[--namelen] = 0xff;
567         ifa_iflink(ifa, ifp, 0 /* Insert head */);
568
569         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
570         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
571
572         if (ifp->if_mapsubq == NULL)
573                 ifp->if_mapsubq = ifq_mapsubq_default;
574
575         ifq = &ifp->if_snd;
576         ifq->altq_type = 0;
577         ifq->altq_disc = NULL;
578         ifq->altq_flags &= ALTQF_CANTCHANGE;
579         ifq->altq_tbr = NULL;
580         ifq->altq_ifp = ifp;
581
582         if (ifq->altq_subq_cnt <= 0)
583                 ifq->altq_subq_cnt = 1;
584         ifq->altq_subq = kmalloc_cachealign(
585             ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
586             M_DEVBUF, M_WAITOK | M_ZERO);
587
588         if (ifq->altq_maxlen == 0) {
589                 if_printf(ifp, "driver didn't set ifq_maxlen\n");
590                 ifq_set_maxlen(ifq, ifqmaxlen);
591         }
592
593         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
594                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
595
596                 ALTQ_SQ_LOCK_INIT(ifsq);
597                 ifsq->ifsq_index = q;
598
599                 ifsq->ifsq_altq = ifq;
600                 ifsq->ifsq_ifp = ifp;
601
602                 ifsq->ifq_maxlen = ifq->altq_maxlen;
603                 ifsq->ifsq_prepended = NULL;
604                 ifsq->ifsq_started = 0;
605                 ifsq->ifsq_hw_oactive = 0;
606                 ifsq_set_cpuid(ifsq, 0);
607
608                 ifsq->ifsq_stage =
609                     kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
610                     M_DEVBUF, M_WAITOK | M_ZERO);
611                 for (i = 0; i < ncpus; ++i)
612                         ifsq->ifsq_stage[i].stg_subq = ifsq;
613
614                 ifsq->ifsq_ifstart_nmsg =
615                     kmalloc(ncpus * sizeof(struct netmsg_base),
616                     M_LWKTMSG, M_WAITOK);
617                 for (i = 0; i < ncpus; ++i) {
618                         netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
619                             &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
620                         ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
621                 }
622         }
623         ifq_set_classic(ifq);
624
625         if (!SLIST_EMPTY(&domains))
626                 if_attachdomain1(ifp);
627
628         /* Announce the interface. */
629         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
630
631         mtx_unlock(&ifp->if_ioctl_mtx);
632 }
633
634 static void
635 if_attachdomain(void *dummy)
636 {
637         struct ifnet *ifp;
638
639         crit_enter();
640         TAILQ_FOREACH(ifp, &ifnet, if_list)
641                 if_attachdomain1(ifp);
642         crit_exit();
643 }
644 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
645         if_attachdomain, NULL);
646
647 static void
648 if_attachdomain1(struct ifnet *ifp)
649 {
650         struct domain *dp;
651
652         crit_enter();
653
654         /* address family dependent data region */
655         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
656         SLIST_FOREACH(dp, &domains, dom_next)
657                 if (dp->dom_ifattach)
658                         ifp->if_afdata[dp->dom_family] =
659                                 (*dp->dom_ifattach)(ifp);
660         crit_exit();
661 }
662
663 /*
664  * Purge all addresses whose type is _not_ AF_LINK
665  */
666 void
667 if_purgeaddrs_nolink(struct ifnet *ifp)
668 {
669         struct ifaddr_container *ifac, *next;
670
671         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
672                               ifa_link, next) {
673                 struct ifaddr *ifa = ifac->ifa;
674
675                 /* Leave link ifaddr as it is */
676                 if (ifa->ifa_addr->sa_family == AF_LINK)
677                         continue;
678 #ifdef INET
679                 /* XXX: Ugly!! ad hoc just for INET */
680                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
681                         struct ifaliasreq ifr;
682 #ifdef IFADDR_DEBUG_VERBOSE
683                         int i;
684
685                         kprintf("purge in4 addr %p: ", ifa);
686                         for (i = 0; i < ncpus; ++i)
687                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
688                         kprintf("\n");
689 #endif
690
691                         bzero(&ifr, sizeof ifr);
692                         ifr.ifra_addr = *ifa->ifa_addr;
693                         if (ifa->ifa_dstaddr)
694                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
695                         if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
696                                        NULL) == 0)
697                                 continue;
698                 }
699 #endif /* INET */
700 #ifdef INET6
701                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
702 #ifdef IFADDR_DEBUG_VERBOSE
703                         int i;
704
705                         kprintf("purge in6 addr %p: ", ifa);
706                         for (i = 0; i < ncpus; ++i)
707                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
708                         kprintf("\n");
709 #endif
710
711                         in6_purgeaddr(ifa);
712                         /* ifp_addrhead is already updated */
713                         continue;
714                 }
715 #endif /* INET6 */
716                 ifa_ifunlink(ifa, ifp);
717                 ifa_destroy(ifa);
718         }
719 }
720
721 static void
722 ifq_stage_detach_handler(netmsg_t nmsg)
723 {
724         struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
725         int q;
726
727         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
728                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
729                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
730
731                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
732                         ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
733         }
734         lwkt_replymsg(&nmsg->lmsg, 0);
735 }
736
737 static void
738 ifq_stage_detach(struct ifaltq *ifq)
739 {
740         struct netmsg_base base;
741         int cpu;
742
743         netmsg_init(&base, NULL, &curthread->td_msgport, 0,
744             ifq_stage_detach_handler);
745         base.lmsg.u.ms_resultp = ifq;
746
747         for (cpu = 0; cpu < ncpus; ++cpu)
748                 lwkt_domsg(netisr_portfn(cpu), &base.lmsg, 0);
749 }
750
751 /*
752  * Detach an interface, removing it from the
753  * list of "active" interfaces.
754  */
755 void
756 if_detach(struct ifnet *ifp)
757 {
758         struct radix_node_head  *rnh;
759         int i, q;
760         int cpu, origcpu;
761         struct domain *dp;
762
763         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
764
765         /*
766          * Remove routes and flush queues.
767          */
768         crit_enter();
769 #ifdef IFPOLL_ENABLE
770         if (ifp->if_flags & IFF_NPOLLING)
771                 ifpoll_deregister(ifp);
772 #endif
773         if_down(ifp);
774
775 #ifdef ALTQ
776         if (ifq_is_enabled(&ifp->if_snd))
777                 altq_disable(&ifp->if_snd);
778         if (ifq_is_attached(&ifp->if_snd))
779                 altq_detach(&ifp->if_snd);
780 #endif
781
782         /*
783          * Clean up all addresses.
784          */
785         ifp->if_lladdr = NULL;
786
787         if_purgeaddrs_nolink(ifp);
788         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
789                 struct ifaddr *ifa;
790
791                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
792                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
793                         ("non-link ifaddr is left on if_addrheads"));
794
795                 ifa_ifunlink(ifa, ifp);
796                 ifa_destroy(ifa);
797                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
798                         ("there are still ifaddrs left on if_addrheads"));
799         }
800
801 #ifdef INET
802         /*
803          * Remove all IPv4 kernel structures related to ifp.
804          */
805         in_ifdetach(ifp);
806 #endif
807
808 #ifdef INET6
809         /*
810          * Remove all IPv6 kernel structs related to ifp.  This should be done
811          * before removing routing entries below, since IPv6 interface direct
812          * routes are expected to be removed by the IPv6-specific kernel API.
813          * Otherwise, the kernel will detect some inconsistency and bark it.
814          */
815         in6_ifdetach(ifp);
816 #endif
817
818         /*
819          * Delete all remaining routes using this interface
820          * Unfortuneatly the only way to do this is to slog through
821          * the entire routing table looking for routes which point
822          * to this interface...oh well...
823          */
824         origcpu = mycpuid;
825         for (cpu = 0; cpu < ncpus; cpu++) {
826                 lwkt_migratecpu(cpu);
827                 for (i = 1; i <= AF_MAX; i++) {
828                         if ((rnh = rt_tables[cpu][i]) == NULL)
829                                 continue;
830                         rnh->rnh_walktree(rnh, if_rtdel, ifp);
831                 }
832         }
833         lwkt_migratecpu(origcpu);
834
835         /* Announce that the interface is gone. */
836         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
837         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
838
839         SLIST_FOREACH(dp, &domains, dom_next)
840                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
841                         (*dp->dom_ifdetach)(ifp,
842                                 ifp->if_afdata[dp->dom_family]);
843
844         /*
845          * Remove interface from ifindex2ifp[] and maybe decrement if_index.
846          */
847         ifindex2ifnet[ifp->if_index] = NULL;
848         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
849                 if_index--;
850
851         TAILQ_REMOVE(&ifnet, ifp, if_link);
852         kfree(ifp->if_addrheads, M_IFADDR);
853
854         lwkt_synchronize_ipiqs("if_detach");
855         ifq_stage_detach(&ifp->if_snd);
856
857         for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
858                 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
859
860                 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
861                 kfree(ifsq->ifsq_stage, M_DEVBUF);
862         }
863         kfree(ifp->if_snd.altq_subq, M_DEVBUF);
864
865         crit_exit();
866 }
867
868 /*
869  * Create interface group without members
870  */
871 struct ifg_group *
872 if_creategroup(const char *groupname)
873 {
874         struct ifg_group        *ifg = NULL;
875
876         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
877             M_TEMP, M_NOWAIT)) == NULL)
878                 return (NULL);
879
880         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
881         ifg->ifg_refcnt = 0;
882         ifg->ifg_carp_demoted = 0;
883         TAILQ_INIT(&ifg->ifg_members);
884 #if NPF > 0
885         pfi_attach_ifgroup(ifg);
886 #endif
887         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
888
889         return (ifg);
890 }
891
892 /*
893  * Add a group to an interface
894  */
895 int
896 if_addgroup(struct ifnet *ifp, const char *groupname)
897 {
898         struct ifg_list         *ifgl;
899         struct ifg_group        *ifg = NULL;
900         struct ifg_member       *ifgm;
901
902         if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
903             groupname[strlen(groupname) - 1] <= '9')
904                 return (EINVAL);
905
906         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
907                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
908                         return (EEXIST);
909
910         if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
911                 return (ENOMEM);
912
913         if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
914                 kfree(ifgl, M_TEMP);
915                 return (ENOMEM);
916         }
917
918         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
919                 if (!strcmp(ifg->ifg_group, groupname))
920                         break;
921
922         if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
923                 kfree(ifgl, M_TEMP);
924                 kfree(ifgm, M_TEMP);
925                 return (ENOMEM);
926         }
927
928         ifg->ifg_refcnt++;
929         ifgl->ifgl_group = ifg;
930         ifgm->ifgm_ifp = ifp;
931
932         TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
933         TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
934
935 #if NPF > 0
936         pfi_group_change(groupname);
937 #endif
938
939         return (0);
940 }
941
942 /*
943  * Remove a group from an interface
944  */
945 int
946 if_delgroup(struct ifnet *ifp, const char *groupname)
947 {
948         struct ifg_list         *ifgl;
949         struct ifg_member       *ifgm;
950
951         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
952                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
953                         break;
954         if (ifgl == NULL)
955                 return (ENOENT);
956
957         TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
958
959         TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
960                 if (ifgm->ifgm_ifp == ifp)
961                         break;
962
963         if (ifgm != NULL) {
964                 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
965                 kfree(ifgm, M_TEMP);
966         }
967
968         if (--ifgl->ifgl_group->ifg_refcnt == 0) {
969                 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
970 #if NPF > 0
971                 pfi_detach_ifgroup(ifgl->ifgl_group);
972 #endif
973                 kfree(ifgl->ifgl_group, M_TEMP);
974         }
975
976         kfree(ifgl, M_TEMP);
977
978 #if NPF > 0
979         pfi_group_change(groupname);
980 #endif
981
982         return (0);
983 }
984
985 /*
986  * Stores all groups from an interface in memory pointed
987  * to by data
988  */
989 int
990 if_getgroup(caddr_t data, struct ifnet *ifp)
991 {
992         int                      len, error;
993         struct ifg_list         *ifgl;
994         struct ifg_req           ifgrq, *ifgp;
995         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
996
997         if (ifgr->ifgr_len == 0) {
998                 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
999                         ifgr->ifgr_len += sizeof(struct ifg_req);
1000                 return (0);
1001         }
1002
1003         len = ifgr->ifgr_len;
1004         ifgp = ifgr->ifgr_groups;
1005         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1006                 if (len < sizeof(ifgrq))
1007                         return (EINVAL);
1008                 bzero(&ifgrq, sizeof ifgrq);
1009                 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1010                     sizeof(ifgrq.ifgrq_group));
1011                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1012                     sizeof(struct ifg_req))))
1013                         return (error);
1014                 len -= sizeof(ifgrq);
1015                 ifgp++;
1016         }
1017
1018         return (0);
1019 }
1020
1021 /*
1022  * Stores all members of a group in memory pointed to by data
1023  */
1024 int
1025 if_getgroupmembers(caddr_t data)
1026 {
1027         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1028         struct ifg_group        *ifg;
1029         struct ifg_member       *ifgm;
1030         struct ifg_req           ifgrq, *ifgp;
1031         int                      len, error;
1032
1033         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1034                 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1035                         break;
1036         if (ifg == NULL)
1037                 return (ENOENT);
1038
1039         if (ifgr->ifgr_len == 0) {
1040                 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1041                         ifgr->ifgr_len += sizeof(ifgrq);
1042                 return (0);
1043         }
1044
1045         len = ifgr->ifgr_len;
1046         ifgp = ifgr->ifgr_groups;
1047         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1048                 if (len < sizeof(ifgrq))
1049                         return (EINVAL);
1050                 bzero(&ifgrq, sizeof ifgrq);
1051                 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1052                     sizeof(ifgrq.ifgrq_member));
1053                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1054                     sizeof(struct ifg_req))))
1055                         return (error);
1056                 len -= sizeof(ifgrq);
1057                 ifgp++;
1058         }
1059
1060         return (0);
1061 }
1062
1063 /*
1064  * Delete Routes for a Network Interface
1065  *
1066  * Called for each routing entry via the rnh->rnh_walktree() call above
1067  * to delete all route entries referencing a detaching network interface.
1068  *
1069  * Arguments:
1070  *      rn      pointer to node in the routing table
1071  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1072  *
1073  * Returns:
1074  *      0       successful
1075  *      errno   failed - reason indicated
1076  *
1077  */
1078 static int
1079 if_rtdel(struct radix_node *rn, void *arg)
1080 {
1081         struct rtentry  *rt = (struct rtentry *)rn;
1082         struct ifnet    *ifp = arg;
1083         int             err;
1084
1085         if (rt->rt_ifp == ifp) {
1086
1087                 /*
1088                  * Protect (sorta) against walktree recursion problems
1089                  * with cloned routes
1090                  */
1091                 if (!(rt->rt_flags & RTF_UP))
1092                         return (0);
1093
1094                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1095                                 rt_mask(rt), rt->rt_flags,
1096                                 NULL);
1097                 if (err) {
1098                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1099                 }
1100         }
1101
1102         return (0);
1103 }
1104
1105 /*
1106  * Locate an interface based on a complete address.
1107  */
1108 struct ifaddr *
1109 ifa_ifwithaddr(struct sockaddr *addr)
1110 {
1111         struct ifnet *ifp;
1112
1113         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1114                 struct ifaddr_container *ifac;
1115
1116                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1117                         struct ifaddr *ifa = ifac->ifa;
1118
1119                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1120                                 continue;
1121                         if (sa_equal(addr, ifa->ifa_addr))
1122                                 return (ifa);
1123                         if ((ifp->if_flags & IFF_BROADCAST) &&
1124                             ifa->ifa_broadaddr &&
1125                             /* IPv6 doesn't have broadcast */
1126                             ifa->ifa_broadaddr->sa_len != 0 &&
1127                             sa_equal(ifa->ifa_broadaddr, addr))
1128                                 return (ifa);
1129                 }
1130         }
1131         return (NULL);
1132 }
1133 /*
1134  * Locate the point to point interface with a given destination address.
1135  */
1136 struct ifaddr *
1137 ifa_ifwithdstaddr(struct sockaddr *addr)
1138 {
1139         struct ifnet *ifp;
1140
1141         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1142                 struct ifaddr_container *ifac;
1143
1144                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1145                         continue;
1146
1147                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1148                         struct ifaddr *ifa = ifac->ifa;
1149
1150                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1151                                 continue;
1152                         if (ifa->ifa_dstaddr &&
1153                             sa_equal(addr, ifa->ifa_dstaddr))
1154                                 return (ifa);
1155                 }
1156         }
1157         return (NULL);
1158 }
1159
1160 /*
1161  * Find an interface on a specific network.  If many, choice
1162  * is most specific found.
1163  */
1164 struct ifaddr *
1165 ifa_ifwithnet(struct sockaddr *addr)
1166 {
1167         struct ifnet *ifp;
1168         struct ifaddr *ifa_maybe = NULL;
1169         u_int af = addr->sa_family;
1170         char *addr_data = addr->sa_data, *cplim;
1171
1172         /*
1173          * AF_LINK addresses can be looked up directly by their index number,
1174          * so do that if we can.
1175          */
1176         if (af == AF_LINK) {
1177                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1178
1179                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1180                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1181         }
1182
1183         /*
1184          * Scan though each interface, looking for ones that have
1185          * addresses in this address family.
1186          */
1187         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1188                 struct ifaddr_container *ifac;
1189
1190                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1191                         struct ifaddr *ifa = ifac->ifa;
1192                         char *cp, *cp2, *cp3;
1193
1194                         if (ifa->ifa_addr->sa_family != af)
1195 next:                           continue;
1196                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1197                                 /*
1198                                  * This is a bit broken as it doesn't
1199                                  * take into account that the remote end may
1200                                  * be a single node in the network we are
1201                                  * looking for.
1202                                  * The trouble is that we don't know the
1203                                  * netmask for the remote end.
1204                                  */
1205                                 if (ifa->ifa_dstaddr != NULL &&
1206                                     sa_equal(addr, ifa->ifa_dstaddr))
1207                                         return (ifa);
1208                         } else {
1209                                 /*
1210                                  * if we have a special address handler,
1211                                  * then use it instead of the generic one.
1212                                  */
1213                                 if (ifa->ifa_claim_addr) {
1214                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1215                                                 return (ifa);
1216                                         } else {
1217                                                 continue;
1218                                         }
1219                                 }
1220
1221                                 /*
1222                                  * Scan all the bits in the ifa's address.
1223                                  * If a bit dissagrees with what we are
1224                                  * looking for, mask it with the netmask
1225                                  * to see if it really matters.
1226                                  * (A byte at a time)
1227                                  */
1228                                 if (ifa->ifa_netmask == 0)
1229                                         continue;
1230                                 cp = addr_data;
1231                                 cp2 = ifa->ifa_addr->sa_data;
1232                                 cp3 = ifa->ifa_netmask->sa_data;
1233                                 cplim = ifa->ifa_netmask->sa_len +
1234                                         (char *)ifa->ifa_netmask;
1235                                 while (cp3 < cplim)
1236                                         if ((*cp++ ^ *cp2++) & *cp3++)
1237                                                 goto next; /* next address! */
1238                                 /*
1239                                  * If the netmask of what we just found
1240                                  * is more specific than what we had before
1241                                  * (if we had one) then remember the new one
1242                                  * before continuing to search
1243                                  * for an even better one.
1244                                  */
1245                                 if (ifa_maybe == NULL ||
1246                                     rn_refines((char *)ifa->ifa_netmask,
1247                                                (char *)ifa_maybe->ifa_netmask))
1248                                         ifa_maybe = ifa;
1249                         }
1250                 }
1251         }
1252         return (ifa_maybe);
1253 }
1254
1255 /*
1256  * Find an interface address specific to an interface best matching
1257  * a given address.
1258  */
1259 struct ifaddr *
1260 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1261 {
1262         struct ifaddr_container *ifac;
1263         char *cp, *cp2, *cp3;
1264         char *cplim;
1265         struct ifaddr *ifa_maybe = NULL;
1266         u_int af = addr->sa_family;
1267
1268         if (af >= AF_MAX)
1269                 return (0);
1270         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1271                 struct ifaddr *ifa = ifac->ifa;
1272
1273                 if (ifa->ifa_addr->sa_family != af)
1274                         continue;
1275                 if (ifa_maybe == NULL)
1276                         ifa_maybe = ifa;
1277                 if (ifa->ifa_netmask == NULL) {
1278                         if (sa_equal(addr, ifa->ifa_addr) ||
1279                             (ifa->ifa_dstaddr != NULL &&
1280                              sa_equal(addr, ifa->ifa_dstaddr)))
1281                                 return (ifa);
1282                         continue;
1283                 }
1284                 if (ifp->if_flags & IFF_POINTOPOINT) {
1285                         if (sa_equal(addr, ifa->ifa_dstaddr))
1286                                 return (ifa);
1287                 } else {
1288                         cp = addr->sa_data;
1289                         cp2 = ifa->ifa_addr->sa_data;
1290                         cp3 = ifa->ifa_netmask->sa_data;
1291                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1292                         for (; cp3 < cplim; cp3++)
1293                                 if ((*cp++ ^ *cp2++) & *cp3)
1294                                         break;
1295                         if (cp3 == cplim)
1296                                 return (ifa);
1297                 }
1298         }
1299         return (ifa_maybe);
1300 }
1301
1302 /*
1303  * Default action when installing a route with a Link Level gateway.
1304  * Lookup an appropriate real ifa to point to.
1305  * This should be moved to /sys/net/link.c eventually.
1306  */
1307 static void
1308 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1309 {
1310         struct ifaddr *ifa;
1311         struct sockaddr *dst;
1312         struct ifnet *ifp;
1313
1314         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1315             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1316                 return;
1317         ifa = ifaof_ifpforaddr(dst, ifp);
1318         if (ifa != NULL) {
1319                 IFAFREE(rt->rt_ifa);
1320                 IFAREF(ifa);
1321                 rt->rt_ifa = ifa;
1322                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1323                         ifa->ifa_rtrequest(cmd, rt, info);
1324         }
1325 }
1326
1327 /*
1328  * Mark an interface down and notify protocols of
1329  * the transition.
1330  * NOTE: must be called at splnet or eqivalent.
1331  */
1332 void
1333 if_unroute(struct ifnet *ifp, int flag, int fam)
1334 {
1335         struct ifaddr_container *ifac;
1336
1337         ifp->if_flags &= ~flag;
1338         getmicrotime(&ifp->if_lastchange);
1339         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1340                 struct ifaddr *ifa = ifac->ifa;
1341
1342                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1343                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1344         }
1345         ifq_purge_all(&ifp->if_snd);
1346         rt_ifmsg(ifp);
1347 }
1348
1349 /*
1350  * Mark an interface up and notify protocols of
1351  * the transition.
1352  * NOTE: must be called at splnet or eqivalent.
1353  */
1354 void
1355 if_route(struct ifnet *ifp, int flag, int fam)
1356 {
1357         struct ifaddr_container *ifac;
1358
1359         ifq_purge_all(&ifp->if_snd);
1360         ifp->if_flags |= flag;
1361         getmicrotime(&ifp->if_lastchange);
1362         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1363                 struct ifaddr *ifa = ifac->ifa;
1364
1365                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1366                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1367         }
1368         rt_ifmsg(ifp);
1369 #ifdef INET6
1370         in6_if_up(ifp);
1371 #endif
1372 }
1373
1374 /*
1375  * Mark an interface down and notify protocols of the transition.  An
1376  * interface going down is also considered to be a synchronizing event.
1377  * We must ensure that all packet processing related to the interface
1378  * has completed before we return so e.g. the caller can free the ifnet
1379  * structure that the mbufs may be referencing.
1380  *
1381  * NOTE: must be called at splnet or eqivalent.
1382  */
1383 void
1384 if_down(struct ifnet *ifp)
1385 {
1386         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1387         netmsg_service_sync();
1388 }
1389
1390 /*
1391  * Mark an interface up and notify protocols of
1392  * the transition.
1393  * NOTE: must be called at splnet or eqivalent.
1394  */
1395 void
1396 if_up(struct ifnet *ifp)
1397 {
1398         if_route(ifp, IFF_UP, AF_UNSPEC);
1399 }
1400
1401 /*
1402  * Process a link state change.
1403  * NOTE: must be called at splsoftnet or equivalent.
1404  */
1405 void
1406 if_link_state_change(struct ifnet *ifp)
1407 {
1408         int link_state = ifp->if_link_state;
1409
1410         rt_ifmsg(ifp);
1411         devctl_notify("IFNET", ifp->if_xname,
1412             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1413 }
1414
1415 /*
1416  * Handle interface watchdog timer routines.  Called
1417  * from softclock, we decrement timers (if set) and
1418  * call the appropriate interface routine on expiration.
1419  */
1420 static void
1421 if_slowtimo(void *arg)
1422 {
1423         struct ifnet *ifp;
1424
1425         crit_enter();
1426
1427         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1428                 if (ifp->if_timer == 0 || --ifp->if_timer)
1429                         continue;
1430                 if (ifp->if_watchdog) {
1431                         if (ifnet_tryserialize_all(ifp)) {
1432                                 (*ifp->if_watchdog)(ifp);
1433                                 ifnet_deserialize_all(ifp);
1434                         } else {
1435                                 /* try again next timeout */
1436                                 ++ifp->if_timer;
1437                         }
1438                 }
1439         }
1440
1441         crit_exit();
1442
1443         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1444 }
1445
1446 /*
1447  * Map interface name to
1448  * interface structure pointer.
1449  */
1450 struct ifnet *
1451 ifunit(const char *name)
1452 {
1453         struct ifnet *ifp;
1454
1455         /*
1456          * Search all the interfaces for this name/number
1457          */
1458
1459         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1460                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1461                         break;
1462         }
1463         return (ifp);
1464 }
1465
1466
1467 /*
1468  * Map interface name in a sockaddr_dl to
1469  * interface structure pointer.
1470  */
1471 struct ifnet *
1472 if_withname(struct sockaddr *sa)
1473 {
1474         char ifname[IFNAMSIZ+1];
1475         struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1476
1477         if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1478              (sdl->sdl_nlen > IFNAMSIZ) )
1479                 return NULL;
1480
1481         /*
1482          * ifunit wants a null-terminated name.  It may not be null-terminated
1483          * in the sockaddr.  We don't want to change the caller's sockaddr,
1484          * and there might not be room to put the trailing null anyway, so we
1485          * make a local copy that we know we can null terminate safely.
1486          */
1487
1488         bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1489         ifname[sdl->sdl_nlen] = '\0';
1490         return ifunit(ifname);
1491 }
1492
1493
1494 /*
1495  * Interface ioctls.
1496  */
1497 int
1498 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1499 {
1500         struct ifnet *ifp;
1501         struct ifreq *ifr;
1502         struct ifstat *ifs;
1503         int error;
1504         short oif_flags;
1505         int new_flags;
1506 #ifdef COMPAT_43
1507         int ocmd;
1508 #endif
1509         size_t namelen, onamelen;
1510         char new_name[IFNAMSIZ];
1511         struct ifaddr *ifa;
1512         struct sockaddr_dl *sdl;
1513
1514         switch (cmd) {
1515         case SIOCGIFCONF:
1516         case OSIOCGIFCONF:
1517                 return (ifconf(cmd, data, cred));
1518         default:
1519                 break;
1520         }
1521
1522         ifr = (struct ifreq *)data;
1523
1524         switch (cmd) {
1525         case SIOCIFCREATE:
1526         case SIOCIFCREATE2:
1527                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1528                         return (error);
1529                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1530                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1531         case SIOCIFDESTROY:
1532                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1533                         return (error);
1534                 return (if_clone_destroy(ifr->ifr_name));
1535         case SIOCIFGCLONERS:
1536                 return (if_clone_list((struct if_clonereq *)data));
1537         default:
1538                 break;
1539         }
1540
1541         /*
1542          * Nominal ioctl through interface, lookup the ifp and obtain a
1543          * lock to serialize the ifconfig ioctl operation.
1544          */
1545         ifp = ifunit(ifr->ifr_name);
1546         if (ifp == NULL)
1547                 return (ENXIO);
1548         error = 0;
1549         mtx_lock(&ifp->if_ioctl_mtx);
1550
1551         switch (cmd) {
1552         case SIOCGIFINDEX:
1553                 ifr->ifr_index = ifp->if_index;
1554                 break;
1555
1556         case SIOCGIFFLAGS:
1557                 ifr->ifr_flags = ifp->if_flags;
1558                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1559                 break;
1560
1561         case SIOCGIFCAP:
1562                 ifr->ifr_reqcap = ifp->if_capabilities;
1563                 ifr->ifr_curcap = ifp->if_capenable;
1564                 break;
1565
1566         case SIOCGIFMETRIC:
1567                 ifr->ifr_metric = ifp->if_metric;
1568                 break;
1569
1570         case SIOCGIFMTU:
1571                 ifr->ifr_mtu = ifp->if_mtu;
1572                 break;
1573
1574         case SIOCGIFTSOLEN:
1575                 ifr->ifr_tsolen = ifp->if_tsolen;
1576                 break;
1577
1578         case SIOCGIFDATA:
1579                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1580                                 sizeof(ifp->if_data));
1581                 break;
1582
1583         case SIOCGIFPHYS:
1584                 ifr->ifr_phys = ifp->if_physical;
1585                 break;
1586
1587         case SIOCGIFPOLLCPU:
1588                 ifr->ifr_pollcpu = -1;
1589                 break;
1590
1591         case SIOCSIFPOLLCPU:
1592                 break;
1593
1594         case SIOCSIFFLAGS:
1595                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1596                 if (error)
1597                         break;
1598                 new_flags = (ifr->ifr_flags & 0xffff) |
1599                     (ifr->ifr_flagshigh << 16);
1600                 if (ifp->if_flags & IFF_SMART) {
1601                         /* Smart drivers twiddle their own routes */
1602                 } else if (ifp->if_flags & IFF_UP &&
1603                     (new_flags & IFF_UP) == 0) {
1604                         crit_enter();
1605                         if_down(ifp);
1606                         crit_exit();
1607                 } else if (new_flags & IFF_UP &&
1608                     (ifp->if_flags & IFF_UP) == 0) {
1609                         crit_enter();
1610                         if_up(ifp);
1611                         crit_exit();
1612                 }
1613
1614 #ifdef IFPOLL_ENABLE
1615                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1616                         if (new_flags & IFF_NPOLLING)
1617                                 ifpoll_register(ifp);
1618                         else
1619                                 ifpoll_deregister(ifp);
1620                 }
1621 #endif
1622
1623                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1624                         (new_flags &~ IFF_CANTCHANGE);
1625                 if (new_flags & IFF_PPROMISC) {
1626                         /* Permanently promiscuous mode requested */
1627                         ifp->if_flags |= IFF_PROMISC;
1628                 } else if (ifp->if_pcount == 0) {
1629                         ifp->if_flags &= ~IFF_PROMISC;
1630                 }
1631                 if (ifp->if_ioctl) {
1632                         ifnet_serialize_all(ifp);
1633                         ifp->if_ioctl(ifp, cmd, data, cred);
1634                         ifnet_deserialize_all(ifp);
1635                 }
1636                 getmicrotime(&ifp->if_lastchange);
1637                 break;
1638
1639         case SIOCSIFCAP:
1640                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1641                 if (error)
1642                         break;
1643                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1644                         error = EINVAL;
1645                         break;
1646                 }
1647                 ifnet_serialize_all(ifp);
1648                 ifp->if_ioctl(ifp, cmd, data, cred);
1649                 ifnet_deserialize_all(ifp);
1650                 break;
1651
1652         case SIOCSIFNAME:
1653                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1654                 if (error)
1655                         break;
1656                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1657                 if (error)
1658                         break;
1659                 if (new_name[0] == '\0') {
1660                         error = EINVAL;
1661                         break;
1662                 }
1663                 if (ifunit(new_name) != NULL) {
1664                         error = EEXIST;
1665                         break;
1666                 }
1667
1668                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1669
1670                 /* Announce the departure of the interface. */
1671                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1672
1673                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1674                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1675                 /* XXX IFA_LOCK(ifa); */
1676                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1677                 namelen = strlen(new_name);
1678                 onamelen = sdl->sdl_nlen;
1679                 /*
1680                  * Move the address if needed.  This is safe because we
1681                  * allocate space for a name of length IFNAMSIZ when we
1682                  * create this in if_attach().
1683                  */
1684                 if (namelen != onamelen) {
1685                         bcopy(sdl->sdl_data + onamelen,
1686                             sdl->sdl_data + namelen, sdl->sdl_alen);
1687                 }
1688                 bcopy(new_name, sdl->sdl_data, namelen);
1689                 sdl->sdl_nlen = namelen;
1690                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1691                 bzero(sdl->sdl_data, onamelen);
1692                 while (namelen != 0)
1693                         sdl->sdl_data[--namelen] = 0xff;
1694                 /* XXX IFA_UNLOCK(ifa) */
1695
1696                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1697
1698                 /* Announce the return of the interface. */
1699                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1700                 break;
1701
1702         case SIOCSIFMETRIC:
1703                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1704                 if (error)
1705                         break;
1706                 ifp->if_metric = ifr->ifr_metric;
1707                 getmicrotime(&ifp->if_lastchange);
1708                 break;
1709
1710         case SIOCSIFPHYS:
1711                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1712                 if (error)
1713                         break;
1714                 if (ifp->if_ioctl == NULL) {
1715                         error = EOPNOTSUPP;
1716                         break;
1717                 }
1718                 ifnet_serialize_all(ifp);
1719                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1720                 ifnet_deserialize_all(ifp);
1721                 if (error == 0)
1722                         getmicrotime(&ifp->if_lastchange);
1723                 break;
1724
1725         case SIOCSIFMTU:
1726         {
1727                 u_long oldmtu = ifp->if_mtu;
1728
1729                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1730                 if (error)
1731                         break;
1732                 if (ifp->if_ioctl == NULL) {
1733                         error = EOPNOTSUPP;
1734                         break;
1735                 }
1736                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
1737                         error = EINVAL;
1738                         break;
1739                 }
1740                 ifnet_serialize_all(ifp);
1741                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1742                 ifnet_deserialize_all(ifp);
1743                 if (error == 0) {
1744                         getmicrotime(&ifp->if_lastchange);
1745                         rt_ifmsg(ifp);
1746                 }
1747                 /*
1748                  * If the link MTU changed, do network layer specific procedure.
1749                  */
1750                 if (ifp->if_mtu != oldmtu) {
1751 #ifdef INET6
1752                         nd6_setmtu(ifp);
1753 #endif
1754                 }
1755                 break;
1756         }
1757
1758         case SIOCSIFTSOLEN:
1759                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1760                 if (error)
1761                         break;
1762
1763                 /* XXX need driver supplied upper limit */
1764                 if (ifr->ifr_tsolen <= 0) {
1765                         error = EINVAL;
1766                         break;
1767                 }
1768                 ifp->if_tsolen = ifr->ifr_tsolen;
1769                 break;
1770
1771         case SIOCADDMULTI:
1772         case SIOCDELMULTI:
1773                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1774                 if (error)
1775                         break;
1776
1777                 /* Don't allow group membership on non-multicast interfaces. */
1778                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1779                         error = EOPNOTSUPP;
1780                         break;
1781                 }
1782
1783                 /* Don't let users screw up protocols' entries. */
1784                 if (ifr->ifr_addr.sa_family != AF_LINK) {
1785                         error = EINVAL;
1786                         break;
1787                 }
1788
1789                 if (cmd == SIOCADDMULTI) {
1790                         struct ifmultiaddr *ifma;
1791                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1792                 } else {
1793                         error = if_delmulti(ifp, &ifr->ifr_addr);
1794                 }
1795                 if (error == 0)
1796                         getmicrotime(&ifp->if_lastchange);
1797                 break;
1798
1799         case SIOCSIFPHYADDR:
1800         case SIOCDIFPHYADDR:
1801 #ifdef INET6
1802         case SIOCSIFPHYADDR_IN6:
1803 #endif
1804         case SIOCSLIFPHYADDR:
1805         case SIOCSIFMEDIA:
1806         case SIOCSIFGENERIC:
1807                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1808                 if (error)
1809                         break;
1810                 if (ifp->if_ioctl == 0) {
1811                         error = EOPNOTSUPP;
1812                         break;
1813                 }
1814                 ifnet_serialize_all(ifp);
1815                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1816                 ifnet_deserialize_all(ifp);
1817                 if (error == 0)
1818                         getmicrotime(&ifp->if_lastchange);
1819                 break;
1820
1821         case SIOCGIFSTATUS:
1822                 ifs = (struct ifstat *)data;
1823                 ifs->ascii[0] = '\0';
1824                 /* fall through */
1825         case SIOCGIFPSRCADDR:
1826         case SIOCGIFPDSTADDR:
1827         case SIOCGLIFPHYADDR:
1828         case SIOCGIFMEDIA:
1829         case SIOCGIFGENERIC:
1830                 if (ifp->if_ioctl == NULL) {
1831                         error = EOPNOTSUPP;
1832                         break;
1833                 }
1834                 ifnet_serialize_all(ifp);
1835                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1836                 ifnet_deserialize_all(ifp);
1837                 break;
1838
1839         case SIOCSIFLLADDR:
1840                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1841                 if (error)
1842                         break;
1843                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
1844                                      ifr->ifr_addr.sa_len);
1845                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1846                 break;
1847
1848         default:
1849                 oif_flags = ifp->if_flags;
1850                 if (so->so_proto == 0) {
1851                         error = EOPNOTSUPP;
1852                         break;
1853                 }
1854 #ifndef COMPAT_43
1855                 error = so_pru_control_direct(so, cmd, data, ifp);
1856 #else
1857                 ocmd = cmd;
1858
1859                 switch (cmd) {
1860                 case SIOCSIFDSTADDR:
1861                 case SIOCSIFADDR:
1862                 case SIOCSIFBRDADDR:
1863                 case SIOCSIFNETMASK:
1864 #if BYTE_ORDER != BIG_ENDIAN
1865                         if (ifr->ifr_addr.sa_family == 0 &&
1866                             ifr->ifr_addr.sa_len < 16) {
1867                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1868                                 ifr->ifr_addr.sa_len = 16;
1869                         }
1870 #else
1871                         if (ifr->ifr_addr.sa_len == 0)
1872                                 ifr->ifr_addr.sa_len = 16;
1873 #endif
1874                         break;
1875                 case OSIOCGIFADDR:
1876                         cmd = SIOCGIFADDR;
1877                         break;
1878                 case OSIOCGIFDSTADDR:
1879                         cmd = SIOCGIFDSTADDR;
1880                         break;
1881                 case OSIOCGIFBRDADDR:
1882                         cmd = SIOCGIFBRDADDR;
1883                         break;
1884                 case OSIOCGIFNETMASK:
1885                         cmd = SIOCGIFNETMASK;
1886                         break;
1887                 default:
1888                         break;
1889                 }
1890
1891                 error = so_pru_control_direct(so, cmd, data, ifp);
1892
1893                 switch (ocmd) {
1894                 case OSIOCGIFADDR:
1895                 case OSIOCGIFDSTADDR:
1896                 case OSIOCGIFBRDADDR:
1897                 case OSIOCGIFNETMASK:
1898                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1899                         break;
1900                 }
1901 #endif /* COMPAT_43 */
1902
1903                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1904 #ifdef INET6
1905                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
1906                         if (ifp->if_flags & IFF_UP) {
1907                                 crit_enter();
1908                                 in6_if_up(ifp);
1909                                 crit_exit();
1910                         }
1911 #endif
1912                 }
1913                 break;
1914         }
1915
1916         mtx_unlock(&ifp->if_ioctl_mtx);
1917         return (error);
1918 }
1919
1920 /*
1921  * Set/clear promiscuous mode on interface ifp based on the truth value
1922  * of pswitch.  The calls are reference counted so that only the first
1923  * "on" request actually has an effect, as does the final "off" request.
1924  * Results are undefined if the "off" and "on" requests are not matched.
1925  */
1926 int
1927 ifpromisc(struct ifnet *ifp, int pswitch)
1928 {
1929         struct ifreq ifr;
1930         int error;
1931         int oldflags;
1932
1933         oldflags = ifp->if_flags;
1934         if (ifp->if_flags & IFF_PPROMISC) {
1935                 /* Do nothing if device is in permanently promiscuous mode */
1936                 ifp->if_pcount += pswitch ? 1 : -1;
1937                 return (0);
1938         }
1939         if (pswitch) {
1940                 /*
1941                  * If the device is not configured up, we cannot put it in
1942                  * promiscuous mode.
1943                  */
1944                 if ((ifp->if_flags & IFF_UP) == 0)
1945                         return (ENETDOWN);
1946                 if (ifp->if_pcount++ != 0)
1947                         return (0);
1948                 ifp->if_flags |= IFF_PROMISC;
1949                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1950                     ifp->if_xname);
1951         } else {
1952                 if (--ifp->if_pcount > 0)
1953                         return (0);
1954                 ifp->if_flags &= ~IFF_PROMISC;
1955                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
1956                     ifp->if_xname);
1957         }
1958         ifr.ifr_flags = ifp->if_flags;
1959         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1960         ifnet_serialize_all(ifp);
1961         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
1962         ifnet_deserialize_all(ifp);
1963         if (error == 0)
1964                 rt_ifmsg(ifp);
1965         else
1966                 ifp->if_flags = oldflags;
1967         return error;
1968 }
1969
1970 /*
1971  * Return interface configuration
1972  * of system.  List may be used
1973  * in later ioctl's (above) to get
1974  * other information.
1975  */
1976 static int
1977 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
1978 {
1979         struct ifconf *ifc = (struct ifconf *)data;
1980         struct ifnet *ifp;
1981         struct sockaddr *sa;
1982         struct ifreq ifr, *ifrp;
1983         int space = ifc->ifc_len, error = 0;
1984
1985         ifrp = ifc->ifc_req;
1986         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1987                 struct ifaddr_container *ifac;
1988                 int addrs;
1989
1990                 if (space <= sizeof ifr)
1991                         break;
1992
1993                 /*
1994                  * Zero the stack declared structure first to prevent
1995                  * memory disclosure.
1996                  */
1997                 bzero(&ifr, sizeof(ifr));
1998                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
1999                     >= sizeof(ifr.ifr_name)) {
2000                         error = ENAMETOOLONG;
2001                         break;
2002                 }
2003
2004                 addrs = 0;
2005                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2006                         struct ifaddr *ifa = ifac->ifa;
2007
2008                         if (space <= sizeof ifr)
2009                                 break;
2010                         sa = ifa->ifa_addr;
2011                         if (cred->cr_prison &&
2012                             prison_if(cred, sa))
2013                                 continue;
2014                         addrs++;
2015 #ifdef COMPAT_43
2016                         if (cmd == OSIOCGIFCONF) {
2017                                 struct osockaddr *osa =
2018                                          (struct osockaddr *)&ifr.ifr_addr;
2019                                 ifr.ifr_addr = *sa;
2020                                 osa->sa_family = sa->sa_family;
2021                                 error = copyout(&ifr, ifrp, sizeof ifr);
2022                                 ifrp++;
2023                         } else
2024 #endif
2025                         if (sa->sa_len <= sizeof(*sa)) {
2026                                 ifr.ifr_addr = *sa;
2027                                 error = copyout(&ifr, ifrp, sizeof ifr);
2028                                 ifrp++;
2029                         } else {
2030                                 if (space < (sizeof ifr) + sa->sa_len -
2031                                             sizeof(*sa))
2032                                         break;
2033                                 space -= sa->sa_len - sizeof(*sa);
2034                                 error = copyout(&ifr, ifrp,
2035                                                 sizeof ifr.ifr_name);
2036                                 if (error == 0)
2037                                         error = copyout(sa, &ifrp->ifr_addr,
2038                                                         sa->sa_len);
2039                                 ifrp = (struct ifreq *)
2040                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2041                         }
2042                         if (error)
2043                                 break;
2044                         space -= sizeof ifr;
2045                 }
2046                 if (error)
2047                         break;
2048                 if (!addrs) {
2049                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2050                         error = copyout(&ifr, ifrp, sizeof ifr);
2051                         if (error)
2052                                 break;
2053                         space -= sizeof ifr;
2054                         ifrp++;
2055                 }
2056         }
2057         ifc->ifc_len -= space;
2058         return (error);
2059 }
2060
2061 /*
2062  * Just like if_promisc(), but for all-multicast-reception mode.
2063  */
2064 int
2065 if_allmulti(struct ifnet *ifp, int onswitch)
2066 {
2067         int error = 0;
2068         struct ifreq ifr;
2069
2070         crit_enter();
2071
2072         if (onswitch) {
2073                 if (ifp->if_amcount++ == 0) {
2074                         ifp->if_flags |= IFF_ALLMULTI;
2075                         ifr.ifr_flags = ifp->if_flags;
2076                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2077                         ifnet_serialize_all(ifp);
2078                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2079                                               NULL);
2080                         ifnet_deserialize_all(ifp);
2081                 }
2082         } else {
2083                 if (ifp->if_amcount > 1) {
2084                         ifp->if_amcount--;
2085                 } else {
2086                         ifp->if_amcount = 0;
2087                         ifp->if_flags &= ~IFF_ALLMULTI;
2088                         ifr.ifr_flags = ifp->if_flags;
2089                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2090                         ifnet_serialize_all(ifp);
2091                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2092                                               NULL);
2093                         ifnet_deserialize_all(ifp);
2094                 }
2095         }
2096
2097         crit_exit();
2098
2099         if (error == 0)
2100                 rt_ifmsg(ifp);
2101         return error;
2102 }
2103
2104 /*
2105  * Add a multicast listenership to the interface in question.
2106  * The link layer provides a routine which converts
2107  */
2108 int
2109 if_addmulti(
2110         struct ifnet *ifp,      /* interface to manipulate */
2111         struct sockaddr *sa,    /* address to add */
2112         struct ifmultiaddr **retifma)
2113 {
2114         struct sockaddr *llsa, *dupsa;
2115         int error;
2116         struct ifmultiaddr *ifma;
2117
2118         /*
2119          * If the matching multicast address already exists
2120          * then don't add a new one, just add a reference
2121          */
2122         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2123                 if (sa_equal(sa, ifma->ifma_addr)) {
2124                         ifma->ifma_refcount++;
2125                         if (retifma)
2126                                 *retifma = ifma;
2127                         return 0;
2128                 }
2129         }
2130
2131         /*
2132          * Give the link layer a chance to accept/reject it, and also
2133          * find out which AF_LINK address this maps to, if it isn't one
2134          * already.
2135          */
2136         if (ifp->if_resolvemulti) {
2137                 ifnet_serialize_all(ifp);
2138                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2139                 ifnet_deserialize_all(ifp);
2140                 if (error) 
2141                         return error;
2142         } else {
2143                 llsa = NULL;
2144         }
2145
2146         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2147         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_WAITOK);
2148         bcopy(sa, dupsa, sa->sa_len);
2149
2150         ifma->ifma_addr = dupsa;
2151         ifma->ifma_lladdr = llsa;
2152         ifma->ifma_ifp = ifp;
2153         ifma->ifma_refcount = 1;
2154         ifma->ifma_protospec = 0;
2155         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2156
2157         /*
2158          * Some network interfaces can scan the address list at
2159          * interrupt time; lock them out.
2160          */
2161         crit_enter();
2162         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2163         crit_exit();
2164         if (retifma)
2165                 *retifma = ifma;
2166
2167         if (llsa != NULL) {
2168                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2169                         if (sa_equal(ifma->ifma_addr, llsa))
2170                                 break;
2171                 }
2172                 if (ifma) {
2173                         ifma->ifma_refcount++;
2174                 } else {
2175                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2176                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_WAITOK);
2177                         bcopy(llsa, dupsa, llsa->sa_len);
2178                         ifma->ifma_addr = dupsa;
2179                         ifma->ifma_ifp = ifp;
2180                         ifma->ifma_refcount = 1;
2181                         crit_enter();
2182                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2183                         crit_exit();
2184                 }
2185         }
2186         /*
2187          * We are certain we have added something, so call down to the
2188          * interface to let them know about it.
2189          */
2190         crit_enter();
2191         ifnet_serialize_all(ifp);
2192         if (ifp->if_ioctl)
2193                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2194         ifnet_deserialize_all(ifp);
2195         crit_exit();
2196
2197         return 0;
2198 }
2199
2200 /*
2201  * Remove a reference to a multicast address on this interface.  Yell
2202  * if the request does not match an existing membership.
2203  */
2204 int
2205 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2206 {
2207         struct ifmultiaddr *ifma;
2208
2209         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2210                 if (sa_equal(sa, ifma->ifma_addr))
2211                         break;
2212         if (ifma == NULL)
2213                 return ENOENT;
2214
2215         if (ifma->ifma_refcount > 1) {
2216                 ifma->ifma_refcount--;
2217                 return 0;
2218         }
2219
2220         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2221         sa = ifma->ifma_lladdr;
2222         crit_enter();
2223         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2224         /*
2225          * Make sure the interface driver is notified
2226          * in the case of a link layer mcast group being left.
2227          */
2228         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL) {
2229                 ifnet_serialize_all(ifp);
2230                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2231                 ifnet_deserialize_all(ifp);
2232         }
2233         crit_exit();
2234         kfree(ifma->ifma_addr, M_IFMADDR);
2235         kfree(ifma, M_IFMADDR);
2236         if (sa == NULL)
2237                 return 0;
2238
2239         /*
2240          * Now look for the link-layer address which corresponds to
2241          * this network address.  It had been squirreled away in
2242          * ifma->ifma_lladdr for this purpose (so we don't have
2243          * to call ifp->if_resolvemulti() again), and we saved that
2244          * value in sa above.  If some nasty deleted the
2245          * link-layer address out from underneath us, we can deal because
2246          * the address we stored was is not the same as the one which was
2247          * in the record for the link-layer address.  (So we don't complain
2248          * in that case.)
2249          */
2250         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2251                 if (sa_equal(sa, ifma->ifma_addr))
2252                         break;
2253         if (ifma == NULL)
2254                 return 0;
2255
2256         if (ifma->ifma_refcount > 1) {
2257                 ifma->ifma_refcount--;
2258                 return 0;
2259         }
2260
2261         crit_enter();
2262         ifnet_serialize_all(ifp);
2263         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2264         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2265         ifnet_deserialize_all(ifp);
2266         crit_exit();
2267         kfree(ifma->ifma_addr, M_IFMADDR);
2268         kfree(sa, M_IFMADDR);
2269         kfree(ifma, M_IFMADDR);
2270
2271         return 0;
2272 }
2273
2274 /*
2275  * Delete all multicast group membership for an interface.
2276  * Should be used to quickly flush all multicast filters.
2277  */
2278 void
2279 if_delallmulti(struct ifnet *ifp)
2280 {
2281         struct ifmultiaddr *ifma;
2282         struct ifmultiaddr *next;
2283
2284         TAILQ_FOREACH_MUTABLE(ifma, &ifp->if_multiaddrs, ifma_link, next)
2285                 if_delmulti(ifp, ifma->ifma_addr);
2286 }
2287
2288
2289 /*
2290  * Set the link layer address on an interface.
2291  *
2292  * At this time we only support certain types of interfaces,
2293  * and we don't allow the length of the address to change.
2294  */
2295 int
2296 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2297 {
2298         struct sockaddr_dl *sdl;
2299         struct ifreq ifr;
2300
2301         sdl = IF_LLSOCKADDR(ifp);
2302         if (sdl == NULL)
2303                 return (EINVAL);
2304         if (len != sdl->sdl_alen)       /* don't allow length to change */
2305                 return (EINVAL);
2306         switch (ifp->if_type) {
2307         case IFT_ETHER:                 /* these types use struct arpcom */
2308         case IFT_XETHER:
2309         case IFT_L2VLAN:
2310                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2311                 bcopy(lladdr, LLADDR(sdl), len);
2312                 break;
2313         default:
2314                 return (ENODEV);
2315         }
2316         /*
2317          * If the interface is already up, we need
2318          * to re-init it in order to reprogram its
2319          * address filter.
2320          */
2321         ifnet_serialize_all(ifp);
2322         if ((ifp->if_flags & IFF_UP) != 0) {
2323 #ifdef INET
2324                 struct ifaddr_container *ifac;
2325 #endif
2326
2327                 ifp->if_flags &= ~IFF_UP;
2328                 ifr.ifr_flags = ifp->if_flags;
2329                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2330                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2331                               NULL);
2332                 ifp->if_flags |= IFF_UP;
2333                 ifr.ifr_flags = ifp->if_flags;
2334                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2335                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2336                                  NULL);
2337 #ifdef INET
2338                 /*
2339                  * Also send gratuitous ARPs to notify other nodes about
2340                  * the address change.
2341                  */
2342                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2343                         struct ifaddr *ifa = ifac->ifa;
2344
2345                         if (ifa->ifa_addr != NULL &&
2346                             ifa->ifa_addr->sa_family == AF_INET)
2347                                 arp_gratuitous(ifp, ifa);
2348                 }
2349 #endif
2350         }
2351         ifnet_deserialize_all(ifp);
2352         return (0);
2353 }
2354
2355 struct ifmultiaddr *
2356 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2357 {
2358         struct ifmultiaddr *ifma;
2359
2360         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2361                 if (sa_equal(ifma->ifma_addr, sa))
2362                         break;
2363
2364         return ifma;
2365 }
2366
2367 /*
2368  * This function locates the first real ethernet MAC from a network
2369  * card and loads it into node, returning 0 on success or ENOENT if
2370  * no suitable interfaces were found.  It is used by the uuid code to
2371  * generate a unique 6-byte number.
2372  */
2373 int
2374 if_getanyethermac(uint16_t *node, int minlen)
2375 {
2376         struct ifnet *ifp;
2377         struct sockaddr_dl *sdl;
2378
2379         TAILQ_FOREACH(ifp, &ifnet, if_link) {
2380                 if (ifp->if_type != IFT_ETHER)
2381                         continue;
2382                 sdl = IF_LLSOCKADDR(ifp);
2383                 if (sdl->sdl_alen < minlen)
2384                         continue;
2385                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2386                       minlen);
2387                 return(0);
2388         }
2389         return (ENOENT);
2390 }
2391
2392 /*
2393  * The name argument must be a pointer to storage which will last as
2394  * long as the interface does.  For physical devices, the result of
2395  * device_get_name(dev) is a good choice and for pseudo-devices a
2396  * static string works well.
2397  */
2398 void
2399 if_initname(struct ifnet *ifp, const char *name, int unit)
2400 {
2401         ifp->if_dname = name;
2402         ifp->if_dunit = unit;
2403         if (unit != IF_DUNIT_NONE)
2404                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2405         else
2406                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2407 }
2408
2409 int
2410 if_printf(struct ifnet *ifp, const char *fmt, ...)
2411 {
2412         __va_list ap;
2413         int retval;
2414
2415         retval = kprintf("%s: ", ifp->if_xname);
2416         __va_start(ap, fmt);
2417         retval += kvprintf(fmt, ap);
2418         __va_end(ap);
2419         return (retval);
2420 }
2421
2422 struct ifnet *
2423 if_alloc(uint8_t type)
2424 {
2425         struct ifnet *ifp;
2426         size_t size;
2427
2428         /*
2429          * XXX temporary hack until arpcom is setup in if_l2com
2430          */
2431         if (type == IFT_ETHER)
2432                 size = sizeof(struct arpcom);
2433         else
2434                 size = sizeof(struct ifnet);
2435
2436         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2437
2438         ifp->if_type = type;
2439
2440         if (if_com_alloc[type] != NULL) {
2441                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2442                 if (ifp->if_l2com == NULL) {
2443                         kfree(ifp, M_IFNET);
2444                         return (NULL);
2445                 }
2446         }
2447         return (ifp);
2448 }
2449
2450 void
2451 if_free(struct ifnet *ifp)
2452 {
2453         kfree(ifp, M_IFNET);
2454 }
2455
2456 void
2457 ifq_set_classic(struct ifaltq *ifq)
2458 {
2459         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2460             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2461 }
2462
2463 void
2464 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2465     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2466 {
2467         int q;
2468
2469         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2470         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2471         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2472         KASSERT(request != NULL, ("request is not specified"));
2473
2474         ifq->altq_mapsubq = mapsubq;
2475         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2476                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2477
2478                 ifsq->ifsq_enqueue = enqueue;
2479                 ifsq->ifsq_dequeue = dequeue;
2480                 ifsq->ifsq_request = request;
2481         }
2482 }
2483
2484 int
2485 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2486     struct altq_pktattr *pa __unused)
2487 {
2488         if (IF_QFULL(ifsq)) {
2489                 m_freem(m);
2490                 return(ENOBUFS);
2491         } else {
2492                 IF_ENQUEUE(ifsq, m);
2493                 return(0);
2494         }       
2495 }
2496
2497 struct mbuf *
2498 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, struct mbuf *mpolled, int op)
2499 {
2500         struct mbuf *m;
2501
2502         switch (op) {
2503         case ALTDQ_POLL:
2504                 IF_POLL(ifsq, m);
2505                 break;
2506         case ALTDQ_REMOVE:
2507                 IF_DEQUEUE(ifsq, m);
2508                 break;
2509         default:
2510                 panic("unsupported ALTQ dequeue op: %d", op);
2511         }
2512         KKASSERT(mpolled == NULL || mpolled == m);
2513         return(m);
2514 }
2515
2516 int
2517 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2518 {
2519         switch (req) {
2520         case ALTRQ_PURGE:
2521                 IF_DRAIN(ifsq);
2522                 break;
2523         default:
2524                 panic("unsupported ALTQ request: %d", req);
2525         }
2526         return(0);
2527 }
2528
2529 static void
2530 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2531 {
2532         struct ifnet *ifp = ifsq_get_ifp(ifsq);
2533         int running = 0, need_sched;
2534
2535         /*
2536          * Try to do direct ifnet.if_start first, if there is
2537          * contention on ifnet's serializer, ifnet.if_start will
2538          * be scheduled on ifnet's CPU.
2539          */
2540         if (!ifnet_tryserialize_tx(ifp, ifsq)) {
2541                 /*
2542                  * ifnet serializer contention happened,
2543                  * ifnet.if_start is scheduled on ifnet's
2544                  * CPU, and we keep going.
2545                  */
2546                 ifsq_ifstart_schedule(ifsq, 1);
2547                 return;
2548         }
2549
2550         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2551                 ifp->if_start(ifp, ifsq);
2552                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2553                         running = 1;
2554         }
2555         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2556
2557         ifnet_deserialize_tx(ifp, ifsq);
2558
2559         if (need_sched) {
2560                 /*
2561                  * More data need to be transmitted, ifnet.if_start is
2562                  * scheduled on ifnet's CPU, and we keep going.
2563                  * NOTE: ifnet.if_start interlock is not released.
2564                  */
2565                 ifsq_ifstart_schedule(ifsq, force_sched);
2566         }
2567 }
2568
2569 /*
2570  * IFSUBQ packets staging mechanism:
2571  *
2572  * The packets enqueued into IFSUBQ are staged to a certain amount before the
2573  * ifnet's if_start is called.  In this way, the driver could avoid writing
2574  * to hardware registers upon every packet, instead, hardware registers
2575  * could be written when certain amount of packets are put onto hardware
2576  * TX ring.  The measurement on several modern NICs (emx(4), igb(4), bnx(4),
2577  * bge(4), jme(4)) shows that the hardware registers writing aggregation
2578  * could save ~20% CPU time when 18bytes UDP datagrams are transmitted at
2579  * 1.48Mpps.  The performance improvement by hardware registers writing
2580  * aggeregation is also mentioned by Luigi Rizzo's netmap paper
2581  * (http://info.iet.unipi.it/~luigi/netmap/).
2582  *
2583  * IFSUBQ packets staging is performed for two entry points into drivers's
2584  * transmission function:
2585  * - Direct ifnet's if_start calling, i.e. ifsq_ifstart_try()
2586  * - ifnet's if_start scheduling, i.e. ifsq_ifstart_schedule()
2587  *
2588  * IFSUBQ packets staging will be stopped upon any of the following conditions:
2589  * - If the count of packets enqueued on the current CPU is great than or
2590  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2591  * - If the total length of packets enqueued on the current CPU is great
2592  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
2593  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
2594  *   is usually less than hardware's MTU.
2595  * - ifsq_ifstart_schedule() is not pending on the current CPU and if_start
2596  *   interlock (if_snd.altq_started) is not released.
2597  * - The if_start_rollup(), which is registered as low priority netisr
2598  *   rollup function, is called; probably because no more work is pending
2599  *   for netisr.
2600  *
2601  * NOTE:
2602  * Currently IFSUBQ packet staging is only performed in netisr threads.
2603  */
2604 int
2605 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2606 {
2607         struct ifaltq *ifq = &ifp->if_snd;
2608         struct ifaltq_subque *ifsq;
2609         int error, start = 0, len, mcast = 0, avoid_start = 0;
2610         struct ifsubq_stage_head *head = NULL;
2611         struct ifsubq_stage *stage = NULL;
2612
2613         ifsq = ifq_map_subq(ifq, mycpuid);
2614         ASSERT_IFNET_NOT_SERIALIZED_TX(ifp, ifsq);
2615
2616         len = m->m_pkthdr.len;
2617         if (m->m_flags & M_MCAST)
2618                 mcast = 1;
2619
2620         if (curthread->td_type == TD_TYPE_NETISR) {
2621                 head = &ifsubq_stage_heads[mycpuid];
2622                 stage = ifsq_get_stage(ifsq, mycpuid);
2623
2624                 stage->stg_cnt++;
2625                 stage->stg_len += len;
2626                 if (stage->stg_cnt < ifsq_stage_cntmax &&
2627                     stage->stg_len < (ifp->if_mtu - max_protohdr))
2628                         avoid_start = 1;
2629         }
2630
2631         ALTQ_SQ_LOCK(ifsq);
2632         error = ifsq_enqueue_locked(ifsq, m, pa);
2633         if (error) {
2634                 if (!ifsq_data_ready(ifsq)) {
2635                         ALTQ_SQ_UNLOCK(ifsq);
2636                         return error;
2637                 }
2638                 avoid_start = 0;
2639         }
2640         if (!ifsq_is_started(ifsq)) {
2641                 if (avoid_start) {
2642                         ALTQ_SQ_UNLOCK(ifsq);
2643
2644                         KKASSERT(!error);
2645                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
2646                                 ifsq_stage_insert(head, stage);
2647
2648                         ifp->if_obytes += len;
2649                         if (mcast)
2650                                 ifp->if_omcasts++;
2651                         return error;
2652                 }
2653
2654                 /*
2655                  * Hold the interlock of ifnet.if_start
2656                  */
2657                 ifsq_set_started(ifsq);
2658                 start = 1;
2659         }
2660         ALTQ_SQ_UNLOCK(ifsq);
2661
2662         if (!error) {
2663                 ifp->if_obytes += len;
2664                 if (mcast)
2665                         ifp->if_omcasts++;
2666         }
2667
2668         if (stage != NULL) {
2669                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
2670                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
2671                         if (!avoid_start) {
2672                                 ifsq_stage_remove(head, stage);
2673                                 ifsq_ifstart_schedule(ifsq, 1);
2674                         }
2675                         return error;
2676                 }
2677
2678                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
2679                         ifsq_stage_remove(head, stage);
2680                 } else {
2681                         stage->stg_cnt = 0;
2682                         stage->stg_len = 0;
2683                 }
2684         }
2685
2686         if (!start)
2687                 return error;
2688
2689         ifsq_ifstart_try(ifsq, 0);
2690         return error;
2691 }
2692
2693 void *
2694 ifa_create(int size, int flags)
2695 {
2696         struct ifaddr *ifa;
2697         int i;
2698
2699         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
2700
2701         ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2702         if (ifa == NULL)
2703                 return NULL;
2704
2705         ifa->ifa_containers = kmalloc(ncpus * sizeof(struct ifaddr_container),
2706                                       M_IFADDR, M_WAITOK | M_ZERO);
2707         ifa->ifa_ncnt = ncpus;
2708         for (i = 0; i < ncpus; ++i) {
2709                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2710
2711                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2712                 ifac->ifa = ifa;
2713                 ifac->ifa_refcnt = 1;
2714         }
2715 #ifdef IFADDR_DEBUG
2716         kprintf("alloc ifa %p %d\n", ifa, size);
2717 #endif
2718         return ifa;
2719 }
2720
2721 void
2722 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2723 {
2724         struct ifaddr *ifa = ifac->ifa;
2725
2726         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2727         KKASSERT(ifac->ifa_refcnt == 0);
2728         KASSERT(ifac->ifa_listmask == 0,
2729                 ("ifa is still on %#x lists", ifac->ifa_listmask));
2730
2731         ifac->ifa_magic = IFA_CONTAINER_DEAD;
2732
2733 #ifdef IFADDR_DEBUG_VERBOSE
2734         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2735 #endif
2736
2737         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2738                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
2739         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2740 #ifdef IFADDR_DEBUG
2741                 kprintf("free ifa %p\n", ifa);
2742 #endif
2743                 kfree(ifa->ifa_containers, M_IFADDR);
2744                 kfree(ifa, M_IFADDR);
2745         }
2746 }
2747
2748 static void
2749 ifa_iflink_dispatch(netmsg_t nmsg)
2750 {
2751         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2752         struct ifaddr *ifa = msg->ifa;
2753         struct ifnet *ifp = msg->ifp;
2754         int cpu = mycpuid;
2755         struct ifaddr_container *ifac;
2756
2757         crit_enter();
2758
2759         ifac = &ifa->ifa_containers[cpu];
2760         ASSERT_IFAC_VALID(ifac);
2761         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2762                 ("ifaddr is on if_addrheads"));
2763
2764         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2765         if (msg->tail)
2766                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2767         else
2768                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2769
2770         crit_exit();
2771
2772         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2773 }
2774
2775 void
2776 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2777 {
2778         struct netmsg_ifaddr msg;
2779
2780         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2781                     0, ifa_iflink_dispatch);
2782         msg.ifa = ifa;
2783         msg.ifp = ifp;
2784         msg.tail = tail;
2785
2786         ifa_domsg(&msg.base.lmsg, 0);
2787 }
2788
2789 static void
2790 ifa_ifunlink_dispatch(netmsg_t nmsg)
2791 {
2792         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2793         struct ifaddr *ifa = msg->ifa;
2794         struct ifnet *ifp = msg->ifp;
2795         int cpu = mycpuid;
2796         struct ifaddr_container *ifac;
2797
2798         crit_enter();
2799
2800         ifac = &ifa->ifa_containers[cpu];
2801         ASSERT_IFAC_VALID(ifac);
2802         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2803                 ("ifaddr is not on if_addrhead"));
2804
2805         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2806         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2807
2808         crit_exit();
2809
2810         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2811 }
2812
2813 void
2814 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2815 {
2816         struct netmsg_ifaddr msg;
2817
2818         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2819                     0, ifa_ifunlink_dispatch);
2820         msg.ifa = ifa;
2821         msg.ifp = ifp;
2822
2823         ifa_domsg(&msg.base.lmsg, 0);
2824 }
2825
2826 static void
2827 ifa_destroy_dispatch(netmsg_t nmsg)
2828 {
2829         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2830
2831         IFAFREE(msg->ifa);
2832         ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2833 }
2834
2835 void
2836 ifa_destroy(struct ifaddr *ifa)
2837 {
2838         struct netmsg_ifaddr msg;
2839
2840         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2841                     0, ifa_destroy_dispatch);
2842         msg.ifa = ifa;
2843
2844         ifa_domsg(&msg.base.lmsg, 0);
2845 }
2846
2847 struct lwkt_port *
2848 ifnet_portfn(int cpu)
2849 {
2850         return &ifnet_threads[cpu].td_msgport;
2851 }
2852
2853 void
2854 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2855 {
2856         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2857
2858         if (next_cpu < ncpus)
2859                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2860         else
2861                 lwkt_replymsg(lmsg, 0);
2862 }
2863
2864 int
2865 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
2866 {
2867         KKASSERT(cpu < ncpus);
2868         return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
2869 }
2870
2871 void
2872 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
2873 {
2874         KKASSERT(cpu < ncpus);
2875         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
2876 }
2877
2878 /*
2879  * Generic netmsg service loop.  Some protocols may roll their own but all
2880  * must do the basic command dispatch function call done here.
2881  */
2882 static void
2883 ifnet_service_loop(void *arg __unused)
2884 {
2885         netmsg_t msg;
2886
2887         while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
2888                 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
2889                 msg->base.nm_dispatch(msg);
2890         }
2891 }
2892
2893 static void
2894 if_start_rollup(void)
2895 {
2896         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
2897         struct ifsubq_stage *stage;
2898
2899         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
2900                 struct ifaltq_subque *ifsq = stage->stg_subq;
2901                 int is_sched = 0;
2902
2903                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
2904                         is_sched = 1;
2905                 ifsq_stage_remove(head, stage);
2906
2907                 if (is_sched) {
2908                         ifsq_ifstart_schedule(ifsq, 1);
2909                 } else {
2910                         int start = 0;
2911
2912                         ALTQ_SQ_LOCK(ifsq);
2913                         if (!ifsq_is_started(ifsq)) {
2914                                 /*
2915                                  * Hold the interlock of ifnet.if_start
2916                                  */
2917                                 ifsq_set_started(ifsq);
2918                                 start = 1;
2919                         }
2920                         ALTQ_SQ_UNLOCK(ifsq);
2921
2922                         if (start)
2923                                 ifsq_ifstart_try(ifsq, 1);
2924                 }
2925                 KKASSERT((stage->stg_flags &
2926                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
2927         }
2928 }
2929
2930 static void
2931 ifnetinit(void *dummy __unused)
2932 {
2933         int i;
2934
2935         for (i = 0; i < ncpus; ++i) {
2936                 struct thread *thr = &ifnet_threads[i];
2937
2938                 lwkt_create(ifnet_service_loop, NULL, NULL,
2939                             thr, TDF_NOSTART|TDF_FORCE_SPINPORT,
2940                             i, "ifnet %d", i);
2941                 netmsg_service_port_init(&thr->td_msgport);
2942                 lwkt_schedule(thr);
2943         }
2944
2945         for (i = 0; i < ncpus; ++i)
2946                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
2947         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
2948 }
2949
2950 struct ifnet *
2951 ifnet_byindex(unsigned short idx)
2952 {
2953         if (idx > if_index)
2954                 return NULL;
2955         return ifindex2ifnet[idx];
2956 }
2957
2958 struct ifaddr *
2959 ifaddr_byindex(unsigned short idx)
2960 {
2961         struct ifnet *ifp;
2962
2963         ifp = ifnet_byindex(idx);
2964         if (!ifp)
2965                 return NULL;
2966         return TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
2967 }
2968
2969 void
2970 if_register_com_alloc(u_char type,
2971     if_com_alloc_t *a, if_com_free_t *f)
2972 {
2973
2974         KASSERT(if_com_alloc[type] == NULL,
2975             ("if_register_com_alloc: %d already registered", type));
2976         KASSERT(if_com_free[type] == NULL,
2977             ("if_register_com_alloc: %d free already registered", type));
2978
2979         if_com_alloc[type] = a;
2980         if_com_free[type] = f;
2981 }
2982
2983 void
2984 if_deregister_com_alloc(u_char type)
2985 {
2986
2987         KASSERT(if_com_alloc[type] != NULL,
2988             ("if_deregister_com_alloc: %d not registered", type));
2989         KASSERT(if_com_free[type] != NULL,
2990             ("if_deregister_com_alloc: %d free not registered", type));
2991         if_com_alloc[type] = NULL;
2992         if_com_free[type] = NULL;
2993 }
2994
2995 int
2996 if_ring_count2(int cnt, int cnt_max)
2997 {
2998         int shift = 0;
2999
3000         KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3001             ("invalid ring count max %d", cnt_max));
3002
3003         if (cnt <= 0)
3004                 cnt = cnt_max;
3005         if (cnt > ncpus2)
3006                 cnt = ncpus2;
3007         if (cnt > cnt_max)
3008                 cnt = cnt_max;
3009
3010         while ((1 << (shift + 1)) <= cnt)
3011                 ++shift;
3012         cnt = 1 << shift;
3013
3014         KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3015             ("calculate cnt %d, ncpus2 %d, cnt max %d",
3016              cnt, ncpus2, cnt_max));
3017         return cnt;
3018 }
3019
3020 void
3021 ifq_set_maxlen(struct ifaltq *ifq, int len)
3022 {
3023         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3024 }
3025
3026 int
3027 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3028 {
3029         return ALTQ_SUBQ_INDEX_DEFAULT;
3030 }
3031
3032 int
3033 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3034 {
3035         return (cpuid & ifq->altq_subq_mask);
3036 }
3037
3038 static void
3039 ifsq_watchdog(void *arg)
3040 {
3041         struct ifsubq_watchdog *wd = arg;
3042         struct ifnet *ifp;
3043
3044         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3045                 goto done;
3046
3047         ifp = ifsq_get_ifp(wd->wd_subq);
3048         if (ifnet_tryserialize_all(ifp)) {
3049                 wd->wd_watchdog(wd->wd_subq);
3050                 ifnet_deserialize_all(ifp);
3051         } else {
3052                 /* try again next timeout */
3053                 wd->wd_timer = 1;
3054         }
3055 done:
3056         ifsq_watchdog_reset(wd);
3057 }
3058
3059 static void
3060 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3061 {
3062         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3063             ifsq_get_cpuid(wd->wd_subq));
3064 }
3065
3066 void
3067 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3068     ifsq_watchdog_t watchdog)
3069 {
3070         callout_init_mp(&wd->wd_callout);
3071         wd->wd_timer = 0;
3072         wd->wd_subq = ifsq;
3073         wd->wd_watchdog = watchdog;
3074 }
3075
3076 void
3077 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3078 {
3079         wd->wd_timer = 0;
3080         ifsq_watchdog_reset(wd);
3081 }
3082
3083 void
3084 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3085 {
3086         wd->wd_timer = 0;
3087         callout_stop(&wd->wd_callout);
3088 }