FreeBSD/Linux Kernel Cross Reference
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. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)if.c 8.3 (Berkeley) 1/4/94
30 * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
31 */
32
33 #include "opt_compat.h"
34 #include "opt_inet6.h"
35 #include "opt_inet.h"
36 #include "opt_ifpoll.h"
37
38 #include <sys/param.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/systm.h>
42 #include <sys/proc.h>
43 #include <sys/priv.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/socketops.h>
48 #include <sys/kernel.h>
49 #include <sys/ktr.h>
50 #include <sys/mutex.h>
51 #include <sys/sockio.h>
52 #include <sys/syslog.h>
53 #include <sys/sysctl.h>
54 #include <sys/domain.h>
55 #include <sys/thread.h>
56 #include <sys/serialize.h>
57 #include <sys/bus.h>
58
59 #include <sys/thread2.h>
60 #include <sys/msgport2.h>
61 #include <sys/mutex2.h>
62
63 #include <net/if.h>
64 #include <net/if_arp.h>
65 #include <net/if_dl.h>
66 #include <net/if_types.h>
67 #include <net/if_var.h>
68 #include <net/ifq_var.h>
69 #include <net/radix.h>
70 #include <net/route.h>
71 #include <net/if_clone.h>
72 #include <net/netisr2.h>
73 #include <net/netmsg2.h>
74
75 #include <machine/atomic.h>
76 #include <machine/stdarg.h>
77 #include <machine/smp.h>
78
79 #if defined(INET) || defined(INET6)
80 /*XXX*/
81 #include <netinet/in.h>
82 #include <netinet/in_var.h>
83 #include <netinet/if_ether.h>
84 #ifdef INET6
85 #include <netinet6/in6_var.h>
86 #include <netinet6/in6_ifattach.h>
87 #endif
88 #endif
89
90 #if defined(COMPAT_43)
91 #include <emulation/43bsd/43bsd_socket.h>
92 #endif /* COMPAT_43 */
93
94 struct netmsg_ifaddr {
95 struct netmsg_base base;
96 struct ifaddr *ifa;
97 struct ifnet *ifp;
98 int tail;
99 };
100
101 struct ifsubq_stage_head {
102 TAILQ_HEAD(, ifsubq_stage) stg_head;
103 } __cachealign;
104
105 /*
106 * System initialization
107 */
108 static void if_attachdomain(void *);
109 static void if_attachdomain1(struct ifnet *);
110 static int ifconf(u_long, caddr_t, struct ucred *);
111 static void ifinit(void *);
112 static void ifnetinit(void *);
113 static void if_slowtimo(void *);
114 static void link_rtrequest(int, struct rtentry *);
115 static int if_rtdel(struct radix_node *, void *);
116
117 /* Helper functions */
118 static void ifsq_watchdog_reset(struct ifsubq_watchdog *);
119
120 #ifdef INET6
121 /*
122 * XXX: declare here to avoid to include many inet6 related files..
123 * should be more generalized?
124 */
125 extern void nd6_setmtu(struct ifnet *);
126 #endif
127
128 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
129 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
130
131 static int ifsq_stage_cntmax = 4;
132 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
133 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
134 &ifsq_stage_cntmax, 0, "ifq staging packet count max");
135
136 static int if_stats_compat = 0;
137 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
138 &if_stats_compat, 0, "Compat the old ifnet stats");
139
140 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
141 /* Must be after netisr_init */
142 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
143
144 static if_com_alloc_t *if_com_alloc[256];
145 static if_com_free_t *if_com_free[256];
146
147 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
148 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
149 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
150
151 int ifqmaxlen = IFQ_MAXLEN;
152 struct ifnethead ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
153
154 struct callout if_slowtimo_timer;
155
156 int if_index = 0;
157 struct ifnet **ifindex2ifnet = NULL;
158 static struct thread ifnet_threads[MAXCPU];
159
160 static struct ifsubq_stage_head ifsubq_stage_heads[MAXCPU];
161
162 #ifdef notyet
163 #define IFQ_KTR_STRING "ifq=%p"
164 #define IFQ_KTR_ARGS struct ifaltq *ifq
165 #ifndef KTR_IFQ
166 #define KTR_IFQ KTR_ALL
167 #endif
168 KTR_INFO_MASTER(ifq);
169 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
170 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
171 #define logifq(name, arg) KTR_LOG(ifq_ ## name, arg)
172
173 #define IF_START_KTR_STRING "ifp=%p"
174 #define IF_START_KTR_ARGS struct ifnet *ifp
175 #ifndef KTR_IF_START
176 #define KTR_IF_START KTR_ALL
177 #endif
178 KTR_INFO_MASTER(if_start);
179 KTR_INFO(KTR_IF_START, if_start, run, 0,
180 IF_START_KTR_STRING, IF_START_KTR_ARGS);
181 KTR_INFO(KTR_IF_START, if_start, sched, 1,
182 IF_START_KTR_STRING, IF_START_KTR_ARGS);
183 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
184 IF_START_KTR_STRING, IF_START_KTR_ARGS);
185 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
186 IF_START_KTR_STRING, IF_START_KTR_ARGS);
187 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
188 IF_START_KTR_STRING, IF_START_KTR_ARGS);
189 #define logifstart(name, arg) KTR_LOG(if_start_ ## name, arg)
190 #endif
191
192 TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
193
194 /*
195 * Network interface utility routines.
196 *
197 * Routines with ifa_ifwith* names take sockaddr *'s as
198 * parameters.
199 */
200 /* ARGSUSED*/
201 void
202 ifinit(void *dummy)
203 {
204 struct ifnet *ifp;
205
206 callout_init(&if_slowtimo_timer);
207
208 crit_enter();
209 TAILQ_FOREACH(ifp, &ifnet, if_link) {
210 if (ifp->if_snd.altq_maxlen == 0) {
211 if_printf(ifp, "XXX: driver didn't set altq_maxlen\n");
212 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
213 }
214 }
215 crit_exit();
216
217 if_slowtimo(0);
218 }
219
220 static void
221 ifsq_ifstart_ipifunc(void *arg)
222 {
223 struct ifaltq_subque *ifsq = arg;
224 struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
225
226 crit_enter();
227 if (lmsg->ms_flags & MSGF_DONE)
228 lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg);
229 crit_exit();
230 }
231
232 static __inline void
233 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
234 {
235 KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
236 TAILQ_REMOVE(&head->stg_head, stage, stg_link);
237 stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
238 stage->stg_cnt = 0;
239 stage->stg_len = 0;
240 }
241
242 static __inline void
243 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
244 {
245 KKASSERT((stage->stg_flags &
246 (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
247 stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
248 TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
249 }
250
251 /*
252 * Schedule ifnet.if_start on the subqueue owner CPU
253 */
254 static void
255 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
256 {
257 int cpu;
258
259 if (!force && curthread->td_type == TD_TYPE_NETISR &&
260 ifsq_stage_cntmax > 0) {
261 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
262
263 stage->stg_cnt = 0;
264 stage->stg_len = 0;
265 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
266 ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
267 stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
268 return;
269 }
270
271 cpu = ifsq_get_cpuid(ifsq);
272 if (cpu != mycpuid)
273 lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
274 else
275 ifsq_ifstart_ipifunc(ifsq);
276 }
277
278 /*
279 * NOTE:
280 * This function will release ifnet.if_start subqueue interlock,
281 * if ifnet.if_start for the subqueue does not need to be scheduled
282 */
283 static __inline int
284 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
285 {
286 if (!running || ifsq_is_empty(ifsq)
287 #ifdef ALTQ
288 || ifsq->ifsq_altq->altq_tbr != NULL
289 #endif
290 ) {
291 ALTQ_SQ_LOCK(ifsq);
292 /*
293 * ifnet.if_start subqueue interlock is released, if:
294 * 1) Hardware can not take any packets, due to
295 * o interface is marked down
296 * o hardware queue is full (ifsq_is_oactive)
297 * Under the second situation, hardware interrupt
298 * or polling(4) will call/schedule ifnet.if_start
299 * on the subqueue when hardware queue is ready
300 * 2) There is no packet in the subqueue.
301 * Further ifq_dispatch or ifq_handoff will call/
302 * schedule ifnet.if_start on the subqueue.
303 * 3) TBR is used and it does not allow further
304 * dequeueing.
305 * TBR callout will call ifnet.if_start on the
306 * subqueue.
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 subqueue owner CPU change.
333 */
334 ifsq_ifstart_schedule(ifsq, 1);
335 return;
336 }
337
338 ifsq_serialize_hw(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 ifsq_deserialize_hw(ifsq);
346
347 if (need_sched) {
348 /*
349 * More data need to be transmitted, ifnet.if_start is
350 * scheduled on the subqueue owner CPU, and we keep going.
351 * NOTE: ifnet.if_start subqueue 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_ALTQ_SQ_SERIALIZED_HW(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.
443 */
444 void
445 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
446 {
447 unsigned socksize, ifasize;
448 int namelen, masklen;
449 struct sockaddr_dl *sdl;
450 struct ifaddr *ifa;
451 struct ifaltq *ifq;
452 int i, q;
453
454 static int if_indexlim = 8;
455
456 if (ifp->if_serialize != NULL) {
457 KASSERT(ifp->if_deserialize != NULL &&
458 ifp->if_tryserialize != NULL &&
459 ifp->if_serialize_assert != NULL,
460 ("serialize functions are partially setup"));
461
462 /*
463 * If the device supplies serialize functions,
464 * then clear if_serializer to catch any invalid
465 * usage of this field.
466 */
467 KASSERT(serializer == NULL,
468 ("both serialize functions and default serializer "
469 "are supplied"));
470 ifp->if_serializer = NULL;
471 } else {
472 KASSERT(ifp->if_deserialize == NULL &&
473 ifp->if_tryserialize == NULL &&
474 ifp->if_serialize_assert == NULL,
475 ("serialize functions are partially setup"));
476 ifp->if_serialize = if_default_serialize;
477 ifp->if_deserialize = if_default_deserialize;
478 ifp->if_tryserialize = if_default_tryserialize;
479 #ifdef INVARIANTS
480 ifp->if_serialize_assert = if_default_serialize_assert;
481 #endif
482
483 /*
484 * The serializer can be passed in from the device,
485 * allowing the same serializer to be used for both
486 * the interrupt interlock and the device queue.
487 * If not specified, the netif structure will use an
488 * embedded serializer.
489 */
490 if (serializer == NULL) {
491 serializer = &ifp->if_default_serializer;
492 lwkt_serialize_init(serializer);
493 }
494 ifp->if_serializer = serializer;
495 }
496
497 mtx_init(&ifp->if_ioctl_mtx);
498 mtx_lock(&ifp->if_ioctl_mtx);
499
500 lwkt_gettoken(&ifnet_token); /* protect if_index and ifnet tailq */
501 ifp->if_index = ++if_index;
502
503 /*
504 * XXX -
505 * The old code would work if the interface passed a pre-existing
506 * chain of ifaddrs to this code. We don't trust our callers to
507 * properly initialize the tailq, however, so we no longer allow
508 * this unlikely case.
509 */
510 ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
511 M_IFADDR, M_WAITOK | M_ZERO);
512 for (i = 0; i < ncpus; ++i)
513 TAILQ_INIT(&ifp->if_addrheads[i]);
514
515 TAILQ_INIT(&ifp->if_prefixhead);
516 TAILQ_INIT(&ifp->if_multiaddrs);
517 TAILQ_INIT(&ifp->if_groups);
518 getmicrotime(&ifp->if_lastchange);
519 if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
520 unsigned int n;
521 struct ifnet **q;
522
523 if_indexlim <<= 1;
524
525 /* grow ifindex2ifnet */
526 n = if_indexlim * sizeof(*q);
527 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
528 if (ifindex2ifnet) {
529 bcopy(ifindex2ifnet, q, n/2);
530 kfree(ifindex2ifnet, M_IFADDR);
531 }
532 ifindex2ifnet = q;
533 }
534
535 ifindex2ifnet[if_index] = ifp;
536
537 /*
538 * create a Link Level name for this device
539 */
540 namelen = strlen(ifp->if_xname);
541 masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
542 socksize = masklen + ifp->if_addrlen;
543 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
544 if (socksize < sizeof(*sdl))
545 socksize = sizeof(*sdl);
546 socksize = ROUNDUP(socksize);
547 #undef ROUNDUP
548 ifasize = sizeof(struct ifaddr) + 2 * socksize;
549 ifa = ifa_create(ifasize, M_WAITOK);
550 sdl = (struct sockaddr_dl *)(ifa + 1);
551 sdl->sdl_len = socksize;
552 sdl->sdl_family = AF_LINK;
553 bcopy(ifp->if_xname, sdl->sdl_data, namelen);
554 sdl->sdl_nlen = namelen;
555 sdl->sdl_index = ifp->if_index;
556 sdl->sdl_type = ifp->if_type;
557 ifp->if_lladdr = ifa;
558 ifa->ifa_ifp = ifp;
559 ifa->ifa_rtrequest = link_rtrequest;
560 ifa->ifa_addr = (struct sockaddr *)sdl;
561 sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
562 ifa->ifa_netmask = (struct sockaddr *)sdl;
563 sdl->sdl_len = masklen;
564 while (namelen != 0)
565 sdl->sdl_data[--namelen] = 0xff;
566 ifa_iflink(ifa, ifp, 0 /* Insert head */);
567
568 ifp->if_data_pcpu = kmalloc_cachealign(
569 ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
570
571 if (ifp->if_mapsubq == NULL)
572 ifp->if_mapsubq = ifq_mapsubq_default;
573
574 ifq = &ifp->if_snd;
575 ifq->altq_type = 0;
576 ifq->altq_disc = NULL;
577 ifq->altq_flags &= ALTQF_CANTCHANGE;
578 ifq->altq_tbr = NULL;
579 ifq->altq_ifp = ifp;
580
581 if (ifq->altq_subq_cnt <= 0)
582 ifq->altq_subq_cnt = 1;
583 ifq->altq_subq = kmalloc_cachealign(
584 ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
585 M_DEVBUF, M_WAITOK | M_ZERO);
586
587 if (ifq->altq_maxlen == 0) {
588 if_printf(ifp, "driver didn't set altq_maxlen\n");
589 ifq_set_maxlen(ifq, ifqmaxlen);
590 }
591
592 for (q = 0; q < ifq->altq_subq_cnt; ++q) {
593 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
594
595 ALTQ_SQ_LOCK_INIT(ifsq);
596 ifsq->ifsq_index = q;
597
598 ifsq->ifsq_altq = ifq;
599 ifsq->ifsq_ifp = ifp;
600
601 ifsq->ifsq_maxlen = ifq->altq_maxlen;
602 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
603 ifsq->ifsq_prepended = NULL;
604 ifsq->ifsq_started = 0;
605 ifsq->ifsq_hw_oactive = 0;
606 ifsq_set_cpuid(ifsq, 0);
607 if (ifp->if_serializer != NULL)
608 ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
609
610 ifsq->ifsq_stage =
611 kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
612 M_DEVBUF, M_WAITOK | M_ZERO);
613 for (i = 0; i < ncpus; ++i)
614 ifsq->ifsq_stage[i].stg_subq = ifsq;
615
616 ifsq->ifsq_ifstart_nmsg =
617 kmalloc(ncpus * sizeof(struct netmsg_base),
618 M_LWKTMSG, M_WAITOK);
619 for (i = 0; i < ncpus; ++i) {
620 netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
621 &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
622 ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
623 }
624 }
625 ifq_set_classic(ifq);
626
627 if (!SLIST_EMPTY(&domains))
628 if_attachdomain1(ifp);
629
630 TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
631 lwkt_reltoken(&ifnet_token);
632
633 /* Announce the interface. */
634 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
635 devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
636 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
637
638 mtx_unlock(&ifp->if_ioctl_mtx);
639 }
640
641 static void
642 if_attachdomain(void *dummy)
643 {
644 struct ifnet *ifp;
645
646 crit_enter();
647 TAILQ_FOREACH(ifp, &ifnet, if_list)
648 if_attachdomain1(ifp);
649 crit_exit();
650 }
651 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
652 if_attachdomain, NULL);
653
654 static void
655 if_attachdomain1(struct ifnet *ifp)
656 {
657 struct domain *dp;
658
659 crit_enter();
660
661 /* address family dependent data region */
662 bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
663 SLIST_FOREACH(dp, &domains, dom_next)
664 if (dp->dom_ifattach)
665 ifp->if_afdata[dp->dom_family] =
666 (*dp->dom_ifattach)(ifp);
667 crit_exit();
668 }
669
670 /*
671 * Purge all addresses whose type is _not_ AF_LINK
672 */
673 void
674 if_purgeaddrs_nolink(struct ifnet *ifp)
675 {
676 struct ifaddr_container *ifac, *next;
677
678 TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
679 ifa_link, next) {
680 struct ifaddr *ifa = ifac->ifa;
681
682 /* Leave link ifaddr as it is */
683 if (ifa->ifa_addr->sa_family == AF_LINK)
684 continue;
685 #ifdef INET
686 /* XXX: Ugly!! ad hoc just for INET */
687 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
688 struct ifaliasreq ifr;
689 #ifdef IFADDR_DEBUG_VERBOSE
690 int i;
691
692 kprintf("purge in4 addr %p: ", ifa);
693 for (i = 0; i < ncpus; ++i)
694 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
695 kprintf("\n");
696 #endif
697
698 bzero(&ifr, sizeof ifr);
699 ifr.ifra_addr = *ifa->ifa_addr;
700 if (ifa->ifa_dstaddr)
701 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
702 if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
703 NULL) == 0)
704 continue;
705 }
706 #endif /* INET */
707 #ifdef INET6
708 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
709 #ifdef IFADDR_DEBUG_VERBOSE
710 int i;
711
712 kprintf("purge in6 addr %p: ", ifa);
713 for (i = 0; i < ncpus; ++i)
714 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
715 kprintf("\n");
716 #endif
717
718 in6_purgeaddr(ifa);
719 /* ifp_addrhead is already updated */
720 continue;
721 }
722 #endif /* INET6 */
723 ifa_ifunlink(ifa, ifp);
724 ifa_destroy(ifa);
725 }
726 }
727
728 static void
729 ifq_stage_detach_handler(netmsg_t nmsg)
730 {
731 struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
732 int q;
733
734 for (q = 0; q < ifq->altq_subq_cnt; ++q) {
735 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
736 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
737
738 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
739 ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
740 }
741 lwkt_replymsg(&nmsg->lmsg, 0);
742 }
743
744 static void
745 ifq_stage_detach(struct ifaltq *ifq)
746 {
747 struct netmsg_base base;
748 int cpu;
749
750 netmsg_init(&base, NULL, &curthread->td_msgport, 0,
751 ifq_stage_detach_handler);
752 base.lmsg.u.ms_resultp = ifq;
753
754 for (cpu = 0; cpu < ncpus; ++cpu)
755 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
756 }
757
758 struct netmsg_if_rtdel {
759 struct netmsg_base base;
760 struct ifnet *ifp;
761 };
762
763 static void
764 if_rtdel_dispatch(netmsg_t msg)
765 {
766 struct netmsg_if_rtdel *rmsg = (void *)msg;
767 int i, nextcpu, cpu;
768
769 cpu = mycpuid;
770 for (i = 1; i <= AF_MAX; i++) {
771 struct radix_node_head *rnh;
772
773 if ((rnh = rt_tables[cpu][i]) == NULL)
774 continue;
775 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
776 }
777
778 nextcpu = cpu + 1;
779 if (nextcpu < ncpus)
780 lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg);
781 else
782 lwkt_replymsg(&rmsg->base.lmsg, 0);
783 }
784
785 /*
786 * Detach an interface, removing it from the
787 * list of "active" interfaces.
788 */
789 void
790 if_detach(struct ifnet *ifp)
791 {
792 struct netmsg_if_rtdel msg;
793 struct domain *dp;
794 int q;
795
796 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
797
798 /*
799 * Remove routes and flush queues.
800 */
801 crit_enter();
802 #ifdef IFPOLL_ENABLE
803 if (ifp->if_flags & IFF_NPOLLING)
804 ifpoll_deregister(ifp);
805 #endif
806 if_down(ifp);
807
808 #ifdef ALTQ
809 if (ifq_is_enabled(&ifp->if_snd))
810 altq_disable(&ifp->if_snd);
811 if (ifq_is_attached(&ifp->if_snd))
812 altq_detach(&ifp->if_snd);
813 #endif
814
815 /*
816 * Clean up all addresses.
817 */
818 ifp->if_lladdr = NULL;
819
820 if_purgeaddrs_nolink(ifp);
821 if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
822 struct ifaddr *ifa;
823
824 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
825 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
826 ("non-link ifaddr is left on if_addrheads"));
827
828 ifa_ifunlink(ifa, ifp);
829 ifa_destroy(ifa);
830 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
831 ("there are still ifaddrs left on if_addrheads"));
832 }
833
834 #ifdef INET
835 /*
836 * Remove all IPv4 kernel structures related to ifp.
837 */
838 in_ifdetach(ifp);
839 #endif
840
841 #ifdef INET6
842 /*
843 * Remove all IPv6 kernel structs related to ifp. This should be done
844 * before removing routing entries below, since IPv6 interface direct
845 * routes are expected to be removed by the IPv6-specific kernel API.
846 * Otherwise, the kernel will detect some inconsistency and bark it.
847 */
848 in6_ifdetach(ifp);
849 #endif
850
851 /*
852 * Delete all remaining routes using this interface
853 */
854 netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
855 if_rtdel_dispatch);
856 msg.ifp = ifp;
857 rt_domsg_global(&msg.base);
858
859 /* Announce that the interface is gone. */
860 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
861 devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
862
863 SLIST_FOREACH(dp, &domains, dom_next)
864 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
865 (*dp->dom_ifdetach)(ifp,
866 ifp->if_afdata[dp->dom_family]);
867
868 /*
869 * Remove interface from ifindex2ifp[] and maybe decrement if_index.
870 */
871 lwkt_gettoken(&ifnet_token);
872 ifindex2ifnet[ifp->if_index] = NULL;
873 while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
874 if_index--;
875 TAILQ_REMOVE(&ifnet, ifp, if_link);
876 lwkt_reltoken(&ifnet_token);
877
878 kfree(ifp->if_addrheads, M_IFADDR);
879
880 lwkt_synchronize_ipiqs("if_detach");
881 ifq_stage_detach(&ifp->if_snd);
882
883 for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
884 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
885
886 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
887 kfree(ifsq->ifsq_stage, M_DEVBUF);
888 }
889 kfree(ifp->if_snd.altq_subq, M_DEVBUF);
890
891 kfree(ifp->if_data_pcpu, M_DEVBUF);
892
893 crit_exit();
894 }
895
896 /*
897 * Create interface group without members
898 */
899 struct ifg_group *
900 if_creategroup(const char *groupname)
901 {
902 struct ifg_group *ifg = NULL;
903
904 if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
905 M_TEMP, M_NOWAIT)) == NULL)
906 return (NULL);
907
908 strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
909 ifg->ifg_refcnt = 0;
910 ifg->ifg_carp_demoted = 0;
911 TAILQ_INIT(&ifg->ifg_members);
912 #if NPF > 0
913 pfi_attach_ifgroup(ifg);
914 #endif
915 TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
916
917 return (ifg);
918 }
919
920 /*
921 * Add a group to an interface
922 */
923 int
924 if_addgroup(struct ifnet *ifp, const char *groupname)
925 {
926 struct ifg_list *ifgl;
927 struct ifg_group *ifg = NULL;
928 struct ifg_member *ifgm;
929
930 if (groupname[0] && groupname[strlen(groupname) - 1] >= '' &&
931 groupname[strlen(groupname) - 1] <= '9')
932 return (EINVAL);
933
934 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
935 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
936 return (EEXIST);
937
938 if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
939 return (ENOMEM);
940
941 if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
942 kfree(ifgl, M_TEMP);
943 return (ENOMEM);
944 }
945
946 TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
947 if (!strcmp(ifg->ifg_group, groupname))
948 break;
949
950 if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
951 kfree(ifgl, M_TEMP);
952 kfree(ifgm, M_TEMP);
953 return (ENOMEM);
954 }
955
956 ifg->ifg_refcnt++;
957 ifgl->ifgl_group = ifg;
958 ifgm->ifgm_ifp = ifp;
959
960 TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
961 TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
962
963 #if NPF > 0
964 pfi_group_change(groupname);
965 #endif
966
967 return (0);
968 }
969
970 /*
971 * Remove a group from an interface
972 */
973 int
974 if_delgroup(struct ifnet *ifp, const char *groupname)
975 {
976 struct ifg_list *ifgl;
977 struct ifg_member *ifgm;
978
979 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
980 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
981 break;
982 if (ifgl == NULL)
983 return (ENOENT);
984
985 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
986
987 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
988 if (ifgm->ifgm_ifp == ifp)
989 break;
990
991 if (ifgm != NULL) {
992 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
993 kfree(ifgm, M_TEMP);
994 }
995
996 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
997 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
998 #if NPF > 0
999 pfi_detach_ifgroup(ifgl->ifgl_group);
1000 #endif
1001 kfree(ifgl->ifgl_group, M_TEMP);
1002 }
1003
1004 kfree(ifgl, M_TEMP);
1005
1006 #if NPF > 0
1007 pfi_group_change(groupname);
1008 #endif
1009
1010 return (0);
1011 }
1012
1013 /*
1014 * Stores all groups from an interface in memory pointed
1015 * to by data
1016 */
1017 int
1018 if_getgroup(caddr_t data, struct ifnet *ifp)
1019 {
1020 int len, error;
1021 struct ifg_list *ifgl;
1022 struct ifg_req ifgrq, *ifgp;
1023 struct ifgroupreq *ifgr = (struct ifgroupreq *)data;
1024
1025 if (ifgr->ifgr_len == 0) {
1026 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1027 ifgr->ifgr_len += sizeof(struct ifg_req);
1028 return (0);
1029 }
1030
1031 len = ifgr->ifgr_len;
1032 ifgp = ifgr->ifgr_groups;
1033 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1034 if (len < sizeof(ifgrq))
1035 return (EINVAL);
1036 bzero(&ifgrq, sizeof ifgrq);
1037 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1038 sizeof(ifgrq.ifgrq_group));
1039 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1040 sizeof(struct ifg_req))))
1041 return (error);
1042 len -= sizeof(ifgrq);
1043 ifgp++;
1044 }
1045
1046 return (0);
1047 }
1048
1049 /*
1050 * Stores all members of a group in memory pointed to by data
1051 */
1052 int
1053 if_getgroupmembers(caddr_t data)
1054 {
1055 struct ifgroupreq *ifgr = (struct ifgroupreq *)data;
1056 struct ifg_group *ifg;
1057 struct ifg_member *ifgm;
1058 struct ifg_req ifgrq, *ifgp;
1059 int len, error;
1060
1061 TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1062 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1063 break;
1064 if (ifg == NULL)
1065 return (ENOENT);
1066
1067 if (ifgr->ifgr_len == 0) {
1068 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1069 ifgr->ifgr_len += sizeof(ifgrq);
1070 return (0);
1071 }
1072
1073 len = ifgr->ifgr_len;
1074 ifgp = ifgr->ifgr_groups;
1075 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1076 if (len < sizeof(ifgrq))
1077 return (EINVAL);
1078 bzero(&ifgrq, sizeof ifgrq);
1079 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1080 sizeof(ifgrq.ifgrq_member));
1081 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1082 sizeof(struct ifg_req))))
1083 return (error);
1084 len -= sizeof(ifgrq);
1085 ifgp++;
1086 }
1087
1088 return (0);
1089 }
1090
1091 /*
1092 * Delete Routes for a Network Interface
1093 *
1094 * Called for each routing entry via the rnh->rnh_walktree() call above
1095 * to delete all route entries referencing a detaching network interface.
1096 *
1097 * Arguments:
1098 * rn pointer to node in the routing table
1099 * arg argument passed to rnh->rnh_walktree() - detaching interface
1100 *
1101 * Returns:
1102 * 0 successful
1103 * errno failed - reason indicated
1104 *
1105 */
1106 static int
1107 if_rtdel(struct radix_node *rn, void *arg)
1108 {
1109 struct rtentry *rt = (struct rtentry *)rn;
1110 struct ifnet *ifp = arg;
1111 int err;
1112
1113 if (rt->rt_ifp == ifp) {
1114
1115 /*
1116 * Protect (sorta) against walktree recursion problems
1117 * with cloned routes
1118 */
1119 if (!(rt->rt_flags & RTF_UP))
1120 return (0);
1121
1122 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1123 rt_mask(rt), rt->rt_flags,
1124 NULL);
1125 if (err) {
1126 log(LOG_WARNING, "if_rtdel: error %d\n", err);
1127 }
1128 }
1129
1130 return (0);
1131 }
1132
1133 /*
1134 * Locate an interface based on a complete address.
1135 */
1136 struct ifaddr *
1137 ifa_ifwithaddr(struct sockaddr *addr)
1138 {
1139 struct ifnet *ifp;
1140
1141 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1142 struct ifaddr_container *ifac;
1143
1144 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1145 struct ifaddr *ifa = ifac->ifa;
1146
1147 if (ifa->ifa_addr->sa_family != addr->sa_family)
1148 continue;
1149 if (sa_equal(addr, ifa->ifa_addr))
1150 return (ifa);
1151 if ((ifp->if_flags & IFF_BROADCAST) &&
1152 ifa->ifa_broadaddr &&
1153 /* IPv6 doesn't have broadcast */
1154 ifa->ifa_broadaddr->sa_len != 0 &&
1155 sa_equal(ifa->ifa_broadaddr, addr))
1156 return (ifa);
1157 }
1158 }
1159 return (NULL);
1160 }
1161 /*
1162 * Locate the point to point interface with a given destination address.
1163 */
1164 struct ifaddr *
1165 ifa_ifwithdstaddr(struct sockaddr *addr)
1166 {
1167 struct ifnet *ifp;
1168
1169 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1170 struct ifaddr_container *ifac;
1171
1172 if (!(ifp->if_flags & IFF_POINTOPOINT))
1173 continue;
1174
1175 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1176 struct ifaddr *ifa = ifac->ifa;
1177
1178 if (ifa->ifa_addr->sa_family != addr->sa_family)
1179 continue;
1180 if (ifa->ifa_dstaddr &&
1181 sa_equal(addr, ifa->ifa_dstaddr))
1182 return (ifa);
1183 }
1184 }
1185 return (NULL);
1186 }
1187
1188 /*
1189 * Find an interface on a specific network. If many, choice
1190 * is most specific found.
1191 */
1192 struct ifaddr *
1193 ifa_ifwithnet(struct sockaddr *addr)
1194 {
1195 struct ifnet *ifp;
1196 struct ifaddr *ifa_maybe = NULL;
1197 u_int af = addr->sa_family;
1198 char *addr_data = addr->sa_data, *cplim;
1199
1200 /*
1201 * AF_LINK addresses can be looked up directly by their index number,
1202 * so do that if we can.
1203 */
1204 if (af == AF_LINK) {
1205 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1206
1207 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1208 return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1209 }
1210
1211 /*
1212 * Scan though each interface, looking for ones that have
1213 * addresses in this address family.
1214 */
1215 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1216 struct ifaddr_container *ifac;
1217
1218 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1219 struct ifaddr *ifa = ifac->ifa;
1220 char *cp, *cp2, *cp3;
1221
1222 if (ifa->ifa_addr->sa_family != af)
1223 next: continue;
1224 if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1225 /*
1226 * This is a bit broken as it doesn't
1227 * take into account that the remote end may
1228 * be a single node in the network we are
1229 * looking for.
1230 * The trouble is that we don't know the
1231 * netmask for the remote end.
1232 */
1233 if (ifa->ifa_dstaddr != NULL &&
1234 sa_equal(addr, ifa->ifa_dstaddr))
1235 return (ifa);
1236 } else {
1237 /*
1238 * if we have a special address handler,
1239 * then use it instead of the generic one.
1240 */
1241 if (ifa->ifa_claim_addr) {
1242 if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1243 return (ifa);
1244 } else {
1245 continue;
1246 }
1247 }
1248
1249 /*
1250 * Scan all the bits in the ifa's address.
1251 * If a bit dissagrees with what we are
1252 * looking for, mask it with the netmask
1253 * to see if it really matters.
1254 * (A byte at a time)
1255 */
1256 if (ifa->ifa_netmask == 0)
1257 continue;
1258 cp = addr_data;
1259 cp2 = ifa->ifa_addr->sa_data;
1260 cp3 = ifa->ifa_netmask->sa_data;
1261 cplim = ifa->ifa_netmask->sa_len +
1262 (char *)ifa->ifa_netmask;
1263 while (cp3 < cplim)
1264 if ((*cp++ ^ *cp2++) & *cp3++)
1265 goto next; /* next address! */
1266 /*
1267 * If the netmask of what we just found
1268 * is more specific than what we had before
1269 * (if we had one) then remember the new one
1270 * before continuing to search
1271 * for an even better one.
1272 */
1273 if (ifa_maybe == NULL ||
1274 rn_refines((char *)ifa->ifa_netmask,
1275 (char *)ifa_maybe->ifa_netmask))
1276 ifa_maybe = ifa;
1277 }
1278 }
1279 }
1280 return (ifa_maybe);
1281 }
1282
1283 /*
1284 * Find an interface address specific to an interface best matching
1285 * a given address.
1286 */
1287 struct ifaddr *
1288 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1289 {
1290 struct ifaddr_container *ifac;
1291 char *cp, *cp2, *cp3;
1292 char *cplim;
1293 struct ifaddr *ifa_maybe = NULL;
1294 u_int af = addr->sa_family;
1295
1296 if (af >= AF_MAX)
1297 return (0);
1298 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1299 struct ifaddr *ifa = ifac->ifa;
1300
1301 if (ifa->ifa_addr->sa_family != af)
1302 continue;
1303 if (ifa_maybe == NULL)
1304 ifa_maybe = ifa;
1305 if (ifa->ifa_netmask == NULL) {
1306 if (sa_equal(addr, ifa->ifa_addr) ||
1307 (ifa->ifa_dstaddr != NULL &&
1308 sa_equal(addr, ifa->ifa_dstaddr)))
1309 return (ifa);
1310 continue;
1311 }
1312 if (ifp->if_flags & IFF_POINTOPOINT) {
1313 if (sa_equal(addr, ifa->ifa_dstaddr))
1314 return (ifa);
1315 } else {
1316 cp = addr->sa_data;
1317 cp2 = ifa->ifa_addr->sa_data;
1318 cp3 = ifa->ifa_netmask->sa_data;
1319 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1320 for (; cp3 < cplim; cp3++)
1321 if ((*cp++ ^ *cp2++) & *cp3)
1322 break;
1323 if (cp3 == cplim)
1324 return (ifa);
1325 }
1326 }
1327 return (ifa_maybe);
1328 }
1329
1330 /*
1331 * Default action when installing a route with a Link Level gateway.
1332 * Lookup an appropriate real ifa to point to.
1333 * This should be moved to /sys/net/link.c eventually.
1334 */
1335 static void
1336 link_rtrequest(int cmd, struct rtentry *rt)
1337 {
1338 struct ifaddr *ifa;
1339 struct sockaddr *dst;
1340 struct ifnet *ifp;
1341
1342 if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1343 (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1344 return;
1345 ifa = ifaof_ifpforaddr(dst, ifp);
1346 if (ifa != NULL) {
1347 IFAFREE(rt->rt_ifa);
1348 IFAREF(ifa);
1349 rt->rt_ifa = ifa;
1350 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1351 ifa->ifa_rtrequest(cmd, rt);
1352 }
1353 }
1354
1355 /*
1356 * Mark an interface down and notify protocols of
1357 * the transition.
1358 * NOTE: must be called at splnet or eqivalent.
1359 */
1360 void
1361 if_unroute(struct ifnet *ifp, int flag, int fam)
1362 {
1363 struct ifaddr_container *ifac;
1364
1365 ifp->if_flags &= ~flag;
1366 getmicrotime(&ifp->if_lastchange);
1367 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1368 struct ifaddr *ifa = ifac->ifa;
1369
1370 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1371 kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1372 }
1373 ifq_purge_all(&ifp->if_snd);
1374 rt_ifmsg(ifp);
1375 }
1376
1377 /*
1378 * Mark an interface up and notify protocols of
1379 * the transition.
1380 * NOTE: must be called at splnet or eqivalent.
1381 */
1382 void
1383 if_route(struct ifnet *ifp, int flag, int fam)
1384 {
1385 struct ifaddr_container *ifac;
1386
1387 ifq_purge_all(&ifp->if_snd);
1388 ifp->if_flags |= flag;
1389 getmicrotime(&ifp->if_lastchange);
1390 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1391 struct ifaddr *ifa = ifac->ifa;
1392
1393 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1394 kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1395 }
1396 rt_ifmsg(ifp);
1397 #ifdef INET6
1398 in6_if_up(ifp);
1399 #endif
1400 }
1401
1402 /*
1403 * Mark an interface down and notify protocols of the transition. An
1404 * interface going down is also considered to be a synchronizing event.
1405 * We must ensure that all packet processing related to the interface
1406 * has completed before we return so e.g. the caller can free the ifnet
1407 * structure that the mbufs may be referencing.
1408 *
1409 * NOTE: must be called at splnet or eqivalent.
1410 */
1411 void
1412 if_down(struct ifnet *ifp)
1413 {
1414 if_unroute(ifp, IFF_UP, AF_UNSPEC);
1415 netmsg_service_sync();
1416 }
1417
1418 /*
1419 * Mark an interface up and notify protocols of
1420 * the transition.
1421 * NOTE: must be called at splnet or eqivalent.
1422 */
1423 void
1424 if_up(struct ifnet *ifp)
1425 {
1426 if_route(ifp, IFF_UP, AF_UNSPEC);
1427 }
1428
1429 /*
1430 * Process a link state change.
1431 * NOTE: must be called at splsoftnet or equivalent.
1432 */
1433 void
1434 if_link_state_change(struct ifnet *ifp)
1435 {
1436 int link_state = ifp->if_link_state;
1437
1438 rt_ifmsg(ifp);
1439 devctl_notify("IFNET", ifp->if_xname,
1440 (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1441 }
1442
1443 /*
1444 * Handle interface watchdog timer routines. Called
1445 * from softclock, we decrement timers (if set) and
1446 * call the appropriate interface routine on expiration.
1447 */
1448 static void
1449 if_slowtimo(void *arg)
1450 {
1451 struct ifnet *ifp;
1452
1453 crit_enter();
1454
1455 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1456 if (if_stats_compat) {
1457 IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1458 IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1459 IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1460 IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1461 IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1462 IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1463 IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1464 IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1465 IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1466 IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1467 IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1468 }
1469
1470 if (ifp->if_timer == 0 || --ifp->if_timer)
1471 continue;
1472 if (ifp->if_watchdog) {
1473 if (ifnet_tryserialize_all(ifp)) {
1474 (*ifp->if_watchdog)(ifp);
1475 ifnet_deserialize_all(ifp);
1476 } else {
1477 /* try again next timeout */
1478 ++ifp->if_timer;
1479 }
1480 }
1481 }
1482
1483 crit_exit();
1484
1485 callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1486 }
1487
1488 /*
1489 * Map interface name to
1490 * interface structure pointer.
1491 */
1492 struct ifnet *
1493 ifunit(const char *name)
1494 {
1495 struct ifnet *ifp;
1496
1497 /*
1498 * Search all the interfaces for this name/number
1499 */
1500
1501 TAILQ_FOREACH(ifp, &ifnet, if_link) {
1502 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1503 break;
1504 }
1505 return (ifp);
1506 }
1507
1508
1509 /*
1510 * Map interface name in a sockaddr_dl to
1511 * interface structure pointer.
1512 */
1513 struct ifnet *
1514 if_withname(struct sockaddr *sa)
1515 {
1516 char ifname[IFNAMSIZ+1];
1517 struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1518
1519 if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1520 (sdl->sdl_nlen > IFNAMSIZ) )
1521 return NULL;
1522
1523 /*
1524 * ifunit wants a null-terminated name. It may not be null-terminated
1525 * in the sockaddr. We don't want to change the caller's sockaddr,
1526 * and there might not be room to put the trailing null anyway, so we
1527 * make a local copy that we know we can null terminate safely.
1528 */
1529
1530 bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1531 ifname[sdl->sdl_nlen] = '\0';
1532 return ifunit(ifname);
1533 }
1534
1535
1536 /*
1537 * Interface ioctls.
1538 */
1539 int
1540 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1541 {
1542 struct ifnet *ifp;
1543 struct ifreq *ifr;
1544 struct ifstat *ifs;
1545 int error;
1546 short oif_flags;
1547 int new_flags;
1548 #ifdef COMPAT_43
1549 int ocmd;
1550 #endif
1551 size_t namelen, onamelen;
1552 char new_name[IFNAMSIZ];
1553 struct ifaddr *ifa;
1554 struct sockaddr_dl *sdl;
1555
1556 switch (cmd) {
1557 case SIOCGIFCONF:
1558 case OSIOCGIFCONF:
1559 return (ifconf(cmd, data, cred));
1560 default:
1561 break;
1562 }
1563
1564 ifr = (struct ifreq *)data;
1565
1566 switch (cmd) {
1567 case SIOCIFCREATE:
1568 case SIOCIFCREATE2:
1569 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1570 return (error);
1571 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1572 cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1573 case SIOCIFDESTROY:
1574 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1575 return (error);
1576 return (if_clone_destroy(ifr->ifr_name));
1577 case SIOCIFGCLONERS:
1578 return (if_clone_list((struct if_clonereq *)data));
1579 default:
1580 break;
1581 }
1582
1583 /*
1584 * Nominal ioctl through interface, lookup the ifp and obtain a
1585 * lock to serialize the ifconfig ioctl operation.
1586 */
1587 ifp = ifunit(ifr->ifr_name);
1588 if (ifp == NULL)
1589 return (ENXIO);
1590 error = 0;
1591 mtx_lock(&ifp->if_ioctl_mtx);
1592
1593 switch (cmd) {
1594 case SIOCGIFINDEX:
1595 ifr->ifr_index = ifp->if_index;
1596 break;
1597
1598 case SIOCGIFFLAGS:
1599 ifr->ifr_flags = ifp->if_flags;
1600 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1601 break;
1602
1603 case SIOCGIFCAP:
1604 ifr->ifr_reqcap = ifp->if_capabilities;
1605 ifr->ifr_curcap = ifp->if_capenable;
1606 break;
1607
1608 case SIOCGIFMETRIC:
1609 ifr->ifr_metric = ifp->if_metric;
1610 break;
1611
1612 case SIOCGIFMTU:
1613 ifr->ifr_mtu = ifp->if_mtu;
1614 break;
1615
1616 case SIOCGIFTSOLEN:
1617 ifr->ifr_tsolen = ifp->if_tsolen;
1618 break;
1619
1620 case SIOCGIFDATA:
1621 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1622 sizeof(ifp->if_data));
1623 break;
1624
1625 case SIOCGIFPHYS:
1626 ifr->ifr_phys = ifp->if_physical;
1627 break;
1628
1629 case SIOCGIFPOLLCPU:
1630 ifr->ifr_pollcpu = -1;
1631 break;
1632
1633 case SIOCSIFPOLLCPU:
1634 break;
1635
1636 case SIOCSIFFLAGS:
1637 error = priv_check_cred(cred, PRIV_ROOT, 0);
1638 if (error)
1639 break;
1640 new_flags = (ifr->ifr_flags & 0xffff) |
1641 (ifr->ifr_flagshigh << 16);
1642 if (ifp->if_flags & IFF_SMART) {
1643 /* Smart drivers twiddle their own routes */
1644 } else if (ifp->if_flags & IFF_UP &&
1645 (new_flags & IFF_UP) == 0) {
1646 crit_enter();
1647 if_down(ifp);
1648 crit_exit();
1649 } else if (new_flags & IFF_UP &&
1650 (ifp->if_flags & IFF_UP) == 0) {
1651 crit_enter();
1652 if_up(ifp);
1653 crit_exit();
1654 }
1655
1656 #ifdef IFPOLL_ENABLE
1657 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1658 if (new_flags & IFF_NPOLLING)
1659 ifpoll_register(ifp);
1660 else
1661 ifpoll_deregister(ifp);
1662 }
1663 #endif
1664
1665 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1666 (new_flags &~ IFF_CANTCHANGE);
1667 if (new_flags & IFF_PPROMISC) {
1668 /* Permanently promiscuous mode requested */
1669 ifp->if_flags |= IFF_PROMISC;
1670 } else if (ifp->if_pcount == 0) {
1671 ifp->if_flags &= ~IFF_PROMISC;
1672 }
1673 if (ifp->if_ioctl) {
1674 ifnet_serialize_all(ifp);
1675 ifp->if_ioctl(ifp, cmd, data, cred);
1676 ifnet_deserialize_all(ifp);
1677 }
1678 getmicrotime(&ifp->if_lastchange);
1679 break;
1680
1681 case SIOCSIFCAP:
1682 error = priv_check_cred(cred, PRIV_ROOT, 0);
1683 if (error)
1684 break;
1685 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1686 error = EINVAL;
1687 break;
1688 }
1689 ifnet_serialize_all(ifp);
1690 ifp->if_ioctl(ifp, cmd, data, cred);
1691 ifnet_deserialize_all(ifp);
1692 break;
1693
1694 case SIOCSIFNAME:
1695 error = priv_check_cred(cred, PRIV_ROOT, 0);
1696 if (error)
1697 break;
1698 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1699 if (error)
1700 break;
1701 if (new_name[0] == '\0') {
1702 error = EINVAL;
1703 break;
1704 }
1705 if (ifunit(new_name) != NULL) {
1706 error = EEXIST;
1707 break;
1708 }
1709
1710 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1711
1712 /* Announce the departure of the interface. */
1713 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1714
1715 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1716 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1717 /* XXX IFA_LOCK(ifa); */
1718 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1719 namelen = strlen(new_name);
1720 onamelen = sdl->sdl_nlen;
1721 /*
1722 * Move the address if needed. This is safe because we
1723 * allocate space for a name of length IFNAMSIZ when we
1724 * create this in if_attach().
1725 */
1726 if (namelen != onamelen) {
1727 bcopy(sdl->sdl_data + onamelen,
1728 sdl->sdl_data + namelen, sdl->sdl_alen);
1729 }
1730 bcopy(new_name, sdl->sdl_data, namelen);
1731 sdl->sdl_nlen = namelen;
1732 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1733 bzero(sdl->sdl_data, onamelen);
1734 while (namelen != 0)
1735 sdl->sdl_data[--namelen] = 0xff;
1736 /* XXX IFA_UNLOCK(ifa) */
1737
1738 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1739
1740 /* Announce the return of the interface. */
1741 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1742 break;
1743
1744 case SIOCSIFMETRIC:
1745 error = priv_check_cred(cred, PRIV_ROOT, 0);
1746 if (error)
1747 break;
1748 ifp->if_metric = ifr->ifr_metric;
1749 getmicrotime(&ifp->if_lastchange);
1750 break;
1751
1752 case SIOCSIFPHYS:
1753 error = priv_check_cred(cred, PRIV_ROOT, 0);
1754 if (error)
1755 break;
1756 if (ifp->if_ioctl == NULL) {
1757 error = EOPNOTSUPP;
1758 break;
1759 }
1760 ifnet_serialize_all(ifp);
1761 error = ifp->if_ioctl(ifp, cmd, data, cred);
1762 ifnet_deserialize_all(ifp);
1763 if (error == 0)
1764 getmicrotime(&ifp->if_lastchange);
1765 break;
1766
1767 case SIOCSIFMTU:
1768 {
1769 u_long oldmtu = ifp->if_mtu;
1770
1771 error = priv_check_cred(cred, PRIV_ROOT, 0);
1772 if (error)
1773 break;
1774 if (ifp->if_ioctl == NULL) {
1775 error = EOPNOTSUPP;
1776 break;
1777 }
1778 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
1779 error = EINVAL;
1780 break;
1781 }
1782 ifnet_serialize_all(ifp);
1783 error = ifp->if_ioctl(ifp, cmd, data, cred);
1784 ifnet_deserialize_all(ifp);
1785 if (error == 0) {
1786 getmicrotime(&ifp->if_lastchange);
1787 rt_ifmsg(ifp);
1788 }
1789 /*
1790 * If the link MTU changed, do network layer specific procedure.
1791 */
1792 if (ifp->if_mtu != oldmtu) {
1793 #ifdef INET6
1794 nd6_setmtu(ifp);
1795 #endif
1796 }
1797 break;
1798 }
1799
1800 case SIOCSIFTSOLEN:
1801 error = priv_check_cred(cred, PRIV_ROOT, 0);
1802 if (error)
1803 break;
1804
1805 /* XXX need driver supplied upper limit */
1806 if (ifr->ifr_tsolen <= 0) {
1807 error = EINVAL;
1808 break;
1809 }
1810 ifp->if_tsolen = ifr->ifr_tsolen;
1811 break;
1812
1813 case SIOCADDMULTI:
1814 case SIOCDELMULTI:
1815 error = priv_check_cred(cred, PRIV_ROOT, 0);
1816 if (error)
1817 break;
1818
1819 /* Don't allow group membership on non-multicast interfaces. */
1820 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1821 error = EOPNOTSUPP;
1822 break;
1823 }
1824
1825 /* Don't let users screw up protocols' entries. */
1826 if (ifr->ifr_addr.sa_family != AF_LINK) {
1827 error = EINVAL;
1828 break;
1829 }
1830
1831 if (cmd == SIOCADDMULTI) {
1832 struct ifmultiaddr *ifma;
1833 error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1834 } else {
1835 error = if_delmulti(ifp, &ifr->ifr_addr);
1836 }
1837 if (error == 0)
1838 getmicrotime(&ifp->if_lastchange);
1839 break;
1840
1841 case SIOCSIFPHYADDR:
1842 case SIOCDIFPHYADDR:
1843 #ifdef INET6
1844 case SIOCSIFPHYADDR_IN6:
1845 #endif
1846 case SIOCSLIFPHYADDR:
1847 case SIOCSIFMEDIA:
1848 case SIOCSIFGENERIC:
1849 error = priv_check_cred(cred, PRIV_ROOT, 0);
1850 if (error)
1851 break;
1852 if (ifp->if_ioctl == 0) {
1853 error = EOPNOTSUPP;
1854 break;
1855 }
1856 ifnet_serialize_all(ifp);
1857 error = ifp->if_ioctl(ifp, cmd, data, cred);
1858 ifnet_deserialize_all(ifp);
1859 if (error == 0)
1860 getmicrotime(&ifp->if_lastchange);
1861 break;
1862
1863 case SIOCGIFSTATUS:
1864 ifs = (struct ifstat *)data;
1865 ifs->ascii[0] = '\0';
1866 /* fall through */
1867 case SIOCGIFPSRCADDR:
1868 case SIOCGIFPDSTADDR:
1869 case SIOCGLIFPHYADDR:
1870 case SIOCGIFMEDIA:
1871 case SIOCGIFGENERIC:
1872 if (ifp->if_ioctl == NULL) {
1873 error = EOPNOTSUPP;
1874 break;
1875 }
1876 ifnet_serialize_all(ifp);
1877 error = ifp->if_ioctl(ifp, cmd, data, cred);
1878 ifnet_deserialize_all(ifp);
1879 break;
1880
1881 case SIOCSIFLLADDR:
1882 error = priv_check_cred(cred, PRIV_ROOT, 0);
1883 if (error)
1884 break;
1885 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
1886 ifr->ifr_addr.sa_len);
1887 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1888 break;
1889
1890 default:
1891 oif_flags = ifp->if_flags;
1892 if (so->so_proto == 0) {
1893 error = EOPNOTSUPP;
1894 break;
1895 }
1896 #ifndef COMPAT_43
1897 error = so_pru_control_direct(so, cmd, data, ifp);
1898 #else
1899 ocmd = cmd;
1900
1901 switch (cmd) {
1902 case SIOCSIFDSTADDR:
1903 case SIOCSIFADDR:
1904 case SIOCSIFBRDADDR:
1905 case SIOCSIFNETMASK:
1906 #if BYTE_ORDER != BIG_ENDIAN
1907 if (ifr->ifr_addr.sa_family == 0 &&
1908 ifr->ifr_addr.sa_len < 16) {
1909 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1910 ifr->ifr_addr.sa_len = 16;
1911 }
1912 #else
1913 if (ifr->ifr_addr.sa_len == 0)
1914 ifr->ifr_addr.sa_len = 16;
1915 #endif
1916 break;
1917 case OSIOCGIFADDR:
1918 cmd = SIOCGIFADDR;
1919 break;
1920 case OSIOCGIFDSTADDR:
1921 cmd = SIOCGIFDSTADDR;
1922 break;
1923 case OSIOCGIFBRDADDR:
1924 cmd = SIOCGIFBRDADDR;
1925 break;
1926 case OSIOCGIFNETMASK:
1927 cmd = SIOCGIFNETMASK;
1928 break;
1929 default:
1930 break;
1931 }
1932
1933 error = so_pru_control_direct(so, cmd, data, ifp);
1934
1935 switch (ocmd) {
1936 case OSIOCGIFADDR:
1937 case OSIOCGIFDSTADDR:
1938 case OSIOCGIFBRDADDR:
1939 case OSIOCGIFNETMASK:
1940 *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1941 break;
1942 }
1943 #endif /* COMPAT_43 */
1944
1945 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1946 #ifdef INET6
1947 DELAY(100);/* XXX: temporary workaround for fxp issue*/
1948 if (ifp->if_flags & IFF_UP) {
1949 crit_enter();
1950 in6_if_up(ifp);
1951 crit_exit();
1952 }
1953 #endif
1954 }
1955 break;
1956 }
1957
1958 mtx_unlock(&ifp->if_ioctl_mtx);
1959 return (error);
1960 }
1961
1962 /*
1963 * Set/clear promiscuous mode on interface ifp based on the truth value
1964 * of pswitch. The calls are reference counted so that only the first
1965 * "on" request actually has an effect, as does the final "off" request.
1966 * Results are undefined if the "off" and "on" requests are not matched.
1967 */
1968 int
1969 ifpromisc(struct ifnet *ifp, int pswitch)
1970 {
1971 struct ifreq ifr;
1972 int error;
1973 int oldflags;
1974
1975 oldflags = ifp->if_flags;
1976 if (ifp->if_flags & IFF_PPROMISC) {
1977 /* Do nothing if device is in permanently promiscuous mode */
1978 ifp->if_pcount += pswitch ? 1 : -1;
1979 return (0);
1980 }
1981 if (pswitch) {
1982 /*
1983 * If the device is not configured up, we cannot put it in
1984 * promiscuous mode.
1985 */
1986 if ((ifp->if_flags & IFF_UP) == 0)
1987 return (ENETDOWN);
1988 if (ifp->if_pcount++ != 0)
1989 return (0);
1990 ifp->if_flags |= IFF_PROMISC;
1991 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1992 ifp->if_xname);
1993 } else {
1994 if (--ifp->if_pcount > 0)
1995 return (0);
1996 ifp->if_flags &= ~IFF_PROMISC;
1997 log(LOG_INFO, "%s: promiscuous mode disabled\n",
1998 ifp->if_xname);
1999 }
2000 ifr.ifr_flags = ifp->if_flags;
2001 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2002 ifnet_serialize_all(ifp);
2003 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2004 ifnet_deserialize_all(ifp);
2005 if (error == 0)
2006 rt_ifmsg(ifp);
2007 else
2008 ifp->if_flags = oldflags;
2009 return error;
2010 }
2011
2012 /*
2013 * Return interface configuration
2014 * of system. List may be used
2015 * in later ioctl's (above) to get
2016 * other information.
2017 */
2018 static int
2019 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2020 {
2021 struct ifconf *ifc = (struct ifconf *)data;
2022 struct ifnet *ifp;
2023 struct sockaddr *sa;
2024 struct ifreq ifr, *ifrp;
2025 int space = ifc->ifc_len, error = 0;
2026
2027 ifrp = ifc->ifc_req;
2028 TAILQ_FOREACH(ifp, &ifnet, if_link) {
2029 struct ifaddr_container *ifac;
2030 int addrs;
2031
2032 if (space <= sizeof ifr)
2033 break;
2034
2035 /*
2036 * Zero the stack declared structure first to prevent
2037 * memory disclosure.
2038 */
2039 bzero(&ifr, sizeof(ifr));
2040 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2041 >= sizeof(ifr.ifr_name)) {
2042 error = ENAMETOOLONG;
2043 break;
2044 }
2045
2046 addrs = 0;
2047 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2048 struct ifaddr *ifa = ifac->ifa;
2049
2050 if (space <= sizeof ifr)
2051 break;
2052 sa = ifa->ifa_addr;
2053 if (cred->cr_prison &&
2054 prison_if(cred, sa))
2055 continue;
2056 addrs++;
2057 #ifdef COMPAT_43
2058 if (cmd == OSIOCGIFCONF) {
2059 struct osockaddr *osa =
2060 (struct osockaddr *)&ifr.ifr_addr;
2061 ifr.ifr_addr = *sa;
2062 osa->sa_family = sa->sa_family;
2063 error = copyout(&ifr, ifrp, sizeof ifr);
2064 ifrp++;
2065 } else
2066 #endif
2067 if (sa->sa_len <= sizeof(*sa)) {
2068 ifr.ifr_addr = *sa;
2069 error = copyout(&ifr, ifrp, sizeof ifr);
2070 ifrp++;
2071 } else {
2072 if (space < (sizeof ifr) + sa->sa_len -
2073 sizeof(*sa))
2074 break;
2075 space -= sa->sa_len - sizeof(*sa);
2076 error = copyout(&ifr, ifrp,
2077 sizeof ifr.ifr_name);
2078 if (error == 0)
2079 error = copyout(sa, &ifrp->ifr_addr,
2080 sa->sa_len);
2081 ifrp = (struct ifreq *)
2082 (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2083 }
2084 if (error)
2085 break;
2086 space -= sizeof ifr;
2087 }
2088 if (error)
2089 break;
2090 if (!addrs) {
2091 bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2092 error = copyout(&ifr, ifrp, sizeof ifr);
2093 if (error)
2094 break;
2095 space -= sizeof ifr;
2096 ifrp++;
2097 }
2098 }
2099 ifc->ifc_len -= space;
2100 return (error);
2101 }
2102
2103 /*
2104 * Just like if_promisc(), but for all-multicast-reception mode.
2105 */
2106 int
2107 if_allmulti(struct ifnet *ifp, int onswitch)
2108 {
2109 int error = 0;
2110 struct ifreq ifr;
2111
2112 crit_enter();
2113
2114 if (onswitch) {
2115 if (ifp->if_amcount++ == 0) {
2116 ifp->if_flags |= IFF_ALLMULTI;
2117 ifr.ifr_flags = ifp->if_flags;
2118 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2119 ifnet_serialize_all(ifp);
2120 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2121 NULL);
2122 ifnet_deserialize_all(ifp);
2123 }
2124 } else {
2125 if (ifp->if_amcount > 1) {
2126 ifp->if_amcount--;
2127 } else {
2128 ifp->if_amcount = 0;
2129 ifp->if_flags &= ~IFF_ALLMULTI;
2130 ifr.ifr_flags = ifp->if_flags;
2131 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2132 ifnet_serialize_all(ifp);
2133 error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2134 NULL);
2135 ifnet_deserialize_all(ifp);
2136 }
2137 }
2138
2139 crit_exit();
2140
2141 if (error == 0)
2142 rt_ifmsg(ifp);
2143 return error;
2144 }
2145
2146 /*
2147 * Add a multicast listenership to the interface in question.
2148 * The link layer provides a routine which converts
2149 */
2150 int
2151 if_addmulti(
2152 struct ifnet *ifp, /* interface to manipulate */
2153 struct sockaddr *sa, /* address to add */
2154 struct ifmultiaddr **retifma)
2155 {
2156 struct sockaddr *llsa, *dupsa;
2157 int error;
2158 struct ifmultiaddr *ifma;
2159
2160 /*
2161 * If the matching multicast address already exists
2162 * then don't add a new one, just add a reference
2163 */
2164 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2165 if (sa_equal(sa, ifma->ifma_addr)) {
2166 ifma->ifma_refcount++;
2167 if (retifma)
2168 *retifma = ifma;
2169 return 0;
2170 }
2171 }
2172
2173 /*
2174 * Give the link layer a chance to accept/reject it, and also
2175 * find out which AF_LINK address this maps to, if it isn't one
2176 * already.
2177 */
2178 if (ifp->if_resolvemulti) {
2179 ifnet_serialize_all(ifp);
2180 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2181 ifnet_deserialize_all(ifp);
2182 if (error)
2183 return error;
2184 } else {
2185 llsa = NULL;
2186 }
2187
2188 ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2189 dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_WAITOK);
2190 bcopy(sa, dupsa, sa->sa_len);
2191
2192 ifma->ifma_addr = dupsa;
2193 ifma->ifma_lladdr = llsa;
2194 ifma->ifma_ifp = ifp;
2195 ifma->ifma_refcount = 1;
2196 ifma->ifma_protospec = 0;
2197 rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2198
2199 /*
2200 * Some network interfaces can scan the address list at
2201 * interrupt time; lock them out.
2202 */
2203 crit_enter();
2204 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2205 crit_exit();
2206 if (retifma)
2207 *retifma = ifma;
2208
2209 if (llsa != NULL) {
2210 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2211 if (sa_equal(ifma->ifma_addr, llsa))
2212 break;
2213 }
2214 if (ifma) {
2215 ifma->ifma_refcount++;
2216 } else {
2217 ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2218 dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_WAITOK);
2219 bcopy(llsa, dupsa, llsa->sa_len);
2220 ifma->ifma_addr = dupsa;
2221 ifma->ifma_ifp = ifp;
2222 ifma->ifma_refcount = 1;
2223 crit_enter();
2224 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2225 crit_exit();
2226 }
2227 }
2228 /*
2229 * We are certain we have added something, so call down to the
2230 * interface to let them know about it.
2231 */
2232 crit_enter();
2233 ifnet_serialize_all(ifp);
2234 if (ifp->if_ioctl)
2235 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2236 ifnet_deserialize_all(ifp);
2237 crit_exit();
2238
2239 return 0;
2240 }
2241
2242 /*
2243 * Remove a reference to a multicast address on this interface. Yell
2244 * if the request does not match an existing membership.
2245 */
2246 int
2247 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2248 {
2249 struct ifmultiaddr *ifma;
2250
2251 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2252 if (sa_equal(sa, ifma->ifma_addr))
2253 break;
2254 if (ifma == NULL)
2255 return ENOENT;
2256
2257 if (ifma->ifma_refcount > 1) {
2258 ifma->ifma_refcount--;
2259 return 0;
2260 }
2261
2262 rt_newmaddrmsg(RTM_DELMADDR, ifma);
2263 sa = ifma->ifma_lladdr;
2264 crit_enter();
2265 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2266 /*
2267 * Make sure the interface driver is notified
2268 * in the case of a link layer mcast group being left.
2269 */
2270 if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL) {
2271 ifnet_serialize_all(ifp);
2272 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2273 ifnet_deserialize_all(ifp);
2274 }
2275 crit_exit();
2276 kfree(ifma->ifma_addr, M_IFMADDR);
2277 kfree(ifma, M_IFMADDR);
2278 if (sa == NULL)
2279 return 0;
2280
2281 /*
2282 * Now look for the link-layer address which corresponds to
2283 * this network address. It had been squirreled away in
2284 * ifma->ifma_lladdr for this purpose (so we don't have
2285 * to call ifp->if_resolvemulti() again), and we saved that
2286 * value in sa above. If some nasty deleted the
2287 * link-layer address out from underneath us, we can deal because
2288 * the address we stored was is not the same as the one which was
2289 * in the record for the link-layer address. (So we don't complain
2290 * in that case.)
2291 */
2292 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2293 if (sa_equal(sa, ifma->ifma_addr))
2294 break;
2295 if (ifma == NULL)
2296 return 0;
2297
2298 if (ifma->ifma_refcount > 1) {
2299 ifma->ifma_refcount--;
2300 return 0;
2301 }
2302
2303 crit_enter();
2304 ifnet_serialize_all(ifp);
2305 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2306 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2307 ifnet_deserialize_all(ifp);
2308 crit_exit();
2309 kfree(ifma->ifma_addr, M_IFMADDR);
2310 kfree(sa, M_IFMADDR);
2311 kfree(ifma, M_IFMADDR);
2312
2313 return 0;
2314 }
2315
2316 /*
2317 * Delete all multicast group membership for an interface.
2318 * Should be used to quickly flush all multicast filters.
2319 */
2320 void
2321 if_delallmulti(struct ifnet *ifp)
2322 {
2323 struct ifmultiaddr *ifma;
2324 struct ifmultiaddr *next;
2325
2326 TAILQ_FOREACH_MUTABLE(ifma, &ifp->if_multiaddrs, ifma_link, next)
2327 if_delmulti(ifp, ifma->ifma_addr);
2328 }
2329
2330
2331 /*
2332 * Set the link layer address on an interface.
2333 *
2334 * At this time we only support certain types of interfaces,
2335 * and we don't allow the length of the address to change.
2336 */
2337 int
2338 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2339 {
2340 struct sockaddr_dl *sdl;
2341 struct ifreq ifr;
2342
2343 sdl = IF_LLSOCKADDR(ifp);
2344 if (sdl == NULL)
2345 return (EINVAL);
2346 if (len != sdl->sdl_alen) /* don't allow length to change */
2347 return (EINVAL);
2348 switch (ifp->if_type) {
2349 case IFT_ETHER: /* these types use struct arpcom */
2350 case IFT_XETHER:
2351 case IFT_L2VLAN:
2352 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2353 bcopy(lladdr, LLADDR(sdl), len);
2354 break;
2355 default:
2356 return (ENODEV);
2357 }
2358 /*
2359 * If the interface is already up, we need
2360 * to re-init it in order to reprogram its
2361 * address filter.
2362 */
2363 ifnet_serialize_all(ifp);
2364 if ((ifp->if_flags & IFF_UP) != 0) {
2365 #ifdef INET
2366 struct ifaddr_container *ifac;
2367 #endif
2368
2369 ifp->if_flags &= ~IFF_UP;
2370 ifr.ifr_flags = ifp->if_flags;
2371 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2372 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2373 NULL);
2374 ifp->if_flags |= IFF_UP;
2375 ifr.ifr_flags = ifp->if_flags;
2376 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2377 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2378 NULL);
2379 #ifdef INET
2380 /*
2381 * Also send gratuitous ARPs to notify other nodes about
2382 * the address change.
2383 */
2384 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2385 struct ifaddr *ifa = ifac->ifa;
2386
2387 if (ifa->ifa_addr != NULL &&
2388 ifa->ifa_addr->sa_family == AF_INET)
2389 arp_gratuitous(ifp, ifa);
2390 }
2391 #endif
2392 }
2393 ifnet_deserialize_all(ifp);
2394 return (0);
2395 }
2396
2397 struct ifmultiaddr *
2398 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2399 {
2400 struct ifmultiaddr *ifma;
2401
2402 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2403 if (sa_equal(ifma->ifma_addr, sa))
2404 break;
2405
2406 return ifma;
2407 }
2408
2409 /*
2410 * This function locates the first real ethernet MAC from a network
2411 * card and loads it into node, returning 0 on success or ENOENT if
2412 * no suitable interfaces were found. It is used by the uuid code to
2413 * generate a unique 6-byte number.
2414 */
2415 int
2416 if_getanyethermac(uint16_t *node, int minlen)
2417 {
2418 struct ifnet *ifp;
2419 struct sockaddr_dl *sdl;
2420
2421 TAILQ_FOREACH(ifp, &ifnet, if_link) {
2422 if (ifp->if_type != IFT_ETHER)
2423 continue;
2424 sdl = IF_LLSOCKADDR(ifp);
2425 if (sdl->sdl_alen < minlen)
2426 continue;
2427 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2428 minlen);
2429 return(0);
2430 }
2431 return (ENOENT);
2432 }
2433
2434 /*
2435 * The name argument must be a pointer to storage which will last as
2436 * long as the interface does. For physical devices, the result of
2437 * device_get_name(dev) is a good choice and for pseudo-devices a
2438 * static string works well.
2439 */
2440 void
2441 if_initname(struct ifnet *ifp, const char *name, int unit)
2442 {
2443 ifp->if_dname = name;
2444 ifp->if_dunit = unit;
2445 if (unit != IF_DUNIT_NONE)
2446 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2447 else
2448 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2449 }
2450
2451 int
2452 if_printf(struct ifnet *ifp, const char *fmt, ...)
2453 {
2454 __va_list ap;
2455 int retval;
2456
2457 retval = kprintf("%s: ", ifp->if_xname);
2458 __va_start(ap, fmt);
2459 retval += kvprintf(fmt, ap);
2460 __va_end(ap);
2461 return (retval);
2462 }
2463
2464 struct ifnet *
2465 if_alloc(uint8_t type)
2466 {
2467 struct ifnet *ifp;
2468 size_t size;
2469
2470 /*
2471 * XXX temporary hack until arpcom is setup in if_l2com
2472 */
2473 if (type == IFT_ETHER)
2474 size = sizeof(struct arpcom);
2475 else
2476 size = sizeof(struct ifnet);
2477
2478 ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2479
2480 ifp->if_type = type;
2481
2482 if (if_com_alloc[type] != NULL) {
2483 ifp->if_l2com = if_com_alloc[type](type, ifp);
2484 if (ifp->if_l2com == NULL) {
2485 kfree(ifp, M_IFNET);
2486 return (NULL);
2487 }
2488 }
2489 return (ifp);
2490 }
2491
2492 void
2493 if_free(struct ifnet *ifp)
2494 {
2495 kfree(ifp, M_IFNET);
2496 }
2497
2498 void
2499 ifq_set_classic(struct ifaltq *ifq)
2500 {
2501 ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2502 ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2503 }
2504
2505 void
2506 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2507 ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2508 {
2509 int q;
2510
2511 KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2512 KASSERT(enqueue != NULL, ("enqueue is not specified"));
2513 KASSERT(dequeue != NULL, ("dequeue is not specified"));
2514 KASSERT(request != NULL, ("request is not specified"));
2515
2516 ifq->altq_mapsubq = mapsubq;
2517 for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2518 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2519
2520 ifsq->ifsq_enqueue = enqueue;
2521 ifsq->ifsq_dequeue = dequeue;
2522 ifsq->ifsq_request = request;
2523 }
2524 }
2525
2526 static void
2527 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2528 {
2529 m->m_nextpkt = NULL;
2530 if (ifsq->ifsq_norm_tail == NULL)
2531 ifsq->ifsq_norm_head = m;
2532 else
2533 ifsq->ifsq_norm_tail->m_nextpkt = m;
2534 ifsq->ifsq_norm_tail = m;
2535 ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2536 }
2537
2538 static void
2539 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2540 {
2541 m->m_nextpkt = NULL;
2542 if (ifsq->ifsq_prio_tail == NULL)
2543 ifsq->ifsq_prio_head = m;
2544 else
2545 ifsq->ifsq_prio_tail->m_nextpkt = m;
2546 ifsq->ifsq_prio_tail = m;
2547 ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2548 ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2549 }
2550
2551 static struct mbuf *
2552 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2553 {
2554 struct mbuf *m;
2555
2556 m = ifsq->ifsq_norm_head;
2557 if (m != NULL) {
2558 if ((ifsq->ifsq_norm_head = m->m_nextpkt) == NULL)
2559 ifsq->ifsq_norm_tail = NULL;
2560 m->m_nextpkt = NULL;
2561 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2562 }
2563 return m;
2564 }
2565
2566 static struct mbuf *
2567 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2568 {
2569 struct mbuf *m;
2570
2571 m = ifsq->ifsq_prio_head;
2572 if (m != NULL) {
2573 if ((ifsq->ifsq_prio_head = m->m_nextpkt) == NULL)
2574 ifsq->ifsq_prio_tail = NULL;
2575 m->m_nextpkt = NULL;
2576 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2577 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2578 }
2579 return m;
2580 }
2581
2582 int
2583 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2584 struct altq_pktattr *pa __unused)
2585 {
2586 M_ASSERTPKTHDR(m);
2587 if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2588 ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2589 if ((m->m_flags & M_PRIO) &&
2590 ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen / 2) &&
2591 ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt / 2)) {
2592 struct mbuf *m_drop;
2593
2594 /*
2595 * Perform drop-head on normal queue
2596 */
2597 m_drop = ifsq_norm_dequeue(ifsq);
2598 if (m_drop != NULL) {
2599 m_freem(m_drop);
2600 ifsq_prio_enqueue(ifsq, m);
2601 return 0;
2602 }
2603 /* XXX nothing could be dropped? */
2604 }
2605 m_freem(m);
2606 return ENOBUFS;
2607 } else {
2608 if (m->m_flags & M_PRIO)
2609 ifsq_prio_enqueue(ifsq, m);
2610 else
2611 ifsq_norm_enqueue(ifsq, m);
2612 return 0;
2613 }
2614 }
2615
2616 struct mbuf *
2617 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2618 {
2619 struct mbuf *m;
2620
2621 switch (op) {
2622 case ALTDQ_POLL:
2623 m = ifsq->ifsq_prio_head;
2624 if (m == NULL)
2625 m = ifsq->ifsq_norm_head;
2626 break;
2627
2628 case ALTDQ_REMOVE:
2629 m = ifsq_prio_dequeue(ifsq);
2630 if (m == NULL)
2631 m = ifsq_norm_dequeue(ifsq);
2632 break;
2633
2634 default:
2635 panic("unsupported ALTQ dequeue op: %d", op);
2636 }
2637 return m;
2638 }
2639
2640 int
2641 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2642 {
2643 switch (req) {
2644 case ALTRQ_PURGE:
2645 for (;;) {
2646 struct mbuf *m;
2647
2648 m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2649 if (m == NULL)
2650 break;
2651 m_freem(m);
2652 }
2653 break;
2654
2655 default:
2656 panic("unsupported ALTQ request: %d", req);
2657 }
2658 return 0;
2659 }
2660
2661 static void
2662 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2663 {
2664 struct ifnet *ifp = ifsq_get_ifp(ifsq);
2665 int running = 0, need_sched;
2666
2667 /*
2668 * Try to do direct ifnet.if_start on the subqueue first, if there is
2669 * contention on the subqueue hardware serializer, ifnet.if_start on
2670 * the subqueue will be scheduled on the subqueue owner CPU.
2671 */
2672 if (!ifsq_tryserialize_hw(ifsq)) {
2673 /*
2674 * Subqueue hardware serializer contention happened,
2675 * ifnet.if_start on the subqueue is scheduled on
2676 * the subqueue owner CPU, and we keep going.
2677 */
2678 ifsq_ifstart_schedule(ifsq, 1);
2679 return;
2680 }
2681
2682 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2683 ifp->if_start(ifp, ifsq);
2684 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2685 running = 1;
2686 }
2687 need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2688
2689 ifsq_deserialize_hw(ifsq);
2690
2691 if (need_sched) {
2692 /*
2693 * More data need to be transmitted, ifnet.if_start on the
2694 * subqueue is scheduled on the subqueue owner CPU, and we
2695 * keep going.
2696 * NOTE: ifnet.if_start subqueue interlock is not released.
2697 */
2698 ifsq_ifstart_schedule(ifsq, force_sched);
2699 }
2700 }
2701
2702 /*
2703 * Subqeue packets staging mechanism:
2704 *
2705 * The packets enqueued into the subqueue are staged to a certain amount
2706 * before the ifnet.if_start on the subqueue is called. In this way, the
2707 * driver could avoid writing to hardware registers upon every packet,
2708 * instead, hardware registers could be written when certain amount of
2709 * packets are put onto hardware TX ring. The measurement on several modern
2710 * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
2711 * registers writing aggregation could save ~20% CPU time when 18bytes UDP
2712 * datagrams are transmitted at 1.48Mpps. The performance improvement by
2713 * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
2714 * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
2715 *
2716 * Subqueue packets staging is performed for two entry points into drivers'
2717 * transmission function:
2718 * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
2719 * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
2720 *
2721 * Subqueue packets staging will be stopped upon any of the following
2722 * conditions:
2723 * - If the count of packets enqueued on the current CPU is great than or
2724 * equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2725 * - If the total length of packets enqueued on the current CPU is great
2726 * than or equal to the hardware's MTU - max_protohdr. max_protohdr is
2727 * cut from the hardware's MTU mainly bacause a full TCP segment's size
2728 * is usually less than hardware's MTU.
2729 * - ifsq_ifstart_schedule() is not pending on the current CPU and
2730 * ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
2731 * released.
2732 * - The if_start_rollup(), which is registered as low priority netisr
2733 * rollup function, is called; probably because no more work is pending
2734 * for netisr.
2735 *
2736 * NOTE:
2737 * Currently subqueue packet staging is only performed in netisr threads.
2738 */
2739 int
2740 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2741 {
2742 struct ifaltq *ifq = &ifp->if_snd;
2743 struct ifaltq_subque *ifsq;
2744 int error, start = 0, len, mcast = 0, avoid_start = 0;
2745 struct ifsubq_stage_head *head = NULL;
2746 struct ifsubq_stage *stage = NULL;
2747
2748 ifsq = ifq_map_subq(ifq, mycpuid);
2749 ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
2750
2751 len = m->m_pkthdr.len;
2752 if (m->m_flags & M_MCAST)
2753 mcast = 1;
2754
2755 if (curthread->td_type == TD_TYPE_NETISR) {
2756 head = &ifsubq_stage_heads[mycpuid];
2757 stage = ifsq_get_stage(ifsq, mycpuid);
2758
2759 stage->stg_cnt++;
2760 stage->stg_len += len;
2761 if (stage->stg_cnt < ifsq_stage_cntmax &&
2762 stage->stg_len < (ifp->if_mtu - max_protohdr))
2763 avoid_start = 1;
2764 }
2765
2766 ALTQ_SQ_LOCK(ifsq);
2767 error = ifsq_enqueue_locked(ifsq, m, pa);
2768 if (error) {
2769 if (!ifsq_data_ready(ifsq)) {
2770 ALTQ_SQ_UNLOCK(ifsq);
2771 return error;
2772 }
2773 avoid_start = 0;
2774 }
2775 if (!ifsq_is_started(ifsq)) {
2776 if (avoid_start) {
2777 ALTQ_SQ_UNLOCK(ifsq);
2778
2779 KKASSERT(!error);
2780 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
2781 ifsq_stage_insert(head, stage);
2782
2783 IFNET_STAT_INC(ifp, obytes, len);
2784 if (mcast)
2785 IFNET_STAT_INC(ifp, omcasts, 1);
2786 return error;
2787 }
2788
2789 /*
2790 * Hold the subqueue interlock of ifnet.if_start
2791 */
2792 ifsq_set_started(ifsq);
2793 start = 1;
2794 }
2795 ALTQ_SQ_UNLOCK(ifsq);
2796
2797 if (!error) {
2798 IFNET_STAT_INC(ifp, obytes, len);
2799 if (mcast)
2800 IFNET_STAT_INC(ifp, omcasts, 1);
2801 }
2802
2803 if (stage != NULL) {
2804 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
2805 KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
2806 if (!avoid_start) {
2807 ifsq_stage_remove(head, stage);
2808 ifsq_ifstart_schedule(ifsq, 1);
2809 }
2810 return error;
2811 }
2812
2813 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
2814 ifsq_stage_remove(head, stage);
2815 } else {
2816 stage->stg_cnt = 0;
2817 stage->stg_len = 0;
2818 }
2819 }
2820
2821 if (!start)
2822 return error;
2823
2824 ifsq_ifstart_try(ifsq, 0);
2825 return error;
2826 }
2827
2828 void *
2829 ifa_create(int size, int flags)
2830 {
2831 struct ifaddr *ifa;
2832 int i;
2833
2834 KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
2835
2836 ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2837 if (ifa == NULL)
2838 return NULL;
2839
2840 ifa->ifa_containers =
2841 kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
2842 M_IFADDR, M_WAITOK | M_ZERO);
2843 ifa->ifa_ncnt = ncpus;
2844 for (i = 0; i < ncpus; ++i) {
2845 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2846
2847 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2848 ifac->ifa = ifa;
2849 ifac->ifa_refcnt = 1;
2850 }
2851 #ifdef IFADDR_DEBUG
2852 kprintf("alloc ifa %p %d\n", ifa, size);
2853 #endif
2854 return ifa;
2855 }
2856
2857 void
2858 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2859 {
2860 struct ifaddr *ifa = ifac->ifa;
2861
2862 KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2863 KKASSERT(ifac->ifa_refcnt == 0);
2864 KASSERT(ifac->ifa_listmask == 0,
2865 ("ifa is still on %#x lists", ifac->ifa_listmask));
2866
2867 ifac->ifa_magic = IFA_CONTAINER_DEAD;
2868
2869 #ifdef IFADDR_DEBUG_VERBOSE
2870 kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2871 #endif
2872
2873 KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2874 ("invalid # of ifac, %d", ifa->ifa_ncnt));
2875 if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2876 #ifdef IFADDR_DEBUG
2877 kprintf("free ifa %p\n", ifa);
2878 #endif
2879 kfree(ifa->ifa_containers, M_IFADDR);
2880 kfree(ifa, M_IFADDR);
2881 }
2882 }
2883
2884 static void
2885 ifa_iflink_dispatch(netmsg_t nmsg)
2886 {
2887 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2888 struct ifaddr *ifa = msg->ifa;
2889 struct ifnet *ifp = msg->ifp;
2890 int cpu = mycpuid;
2891 struct ifaddr_container *ifac;
2892
2893 crit_enter();
2894
2895 ifac = &ifa->ifa_containers[cpu];
2896 ASSERT_IFAC_VALID(ifac);
2897 KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2898 ("ifaddr is on if_addrheads"));
2899
2900 ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2901 if (msg->tail)
2902 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2903 else
2904 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2905
2906 crit_exit();
2907
2908 ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2909 }
2910
2911 void
2912 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2913 {
2914 struct netmsg_ifaddr msg;
2915
2916 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2917 0, ifa_iflink_dispatch);
2918 msg.ifa = ifa;
2919 msg.ifp = ifp;
2920 msg.tail = tail;
2921
2922 ifa_domsg(&msg.base.lmsg, 0);
2923 }
2924
2925 static void
2926 ifa_ifunlink_dispatch(netmsg_t nmsg)
2927 {
2928 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2929 struct ifaddr *ifa = msg->ifa;
2930 struct ifnet *ifp = msg->ifp;
2931 int cpu = mycpuid;
2932 struct ifaddr_container *ifac;
2933
2934 crit_enter();
2935
2936 ifac = &ifa->ifa_containers[cpu];
2937 ASSERT_IFAC_VALID(ifac);
2938 KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2939 ("ifaddr is not on if_addrhead"));
2940
2941 TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2942 ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2943
2944 crit_exit();
2945
2946 ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2947 }
2948
2949 void
2950 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2951 {
2952 struct netmsg_ifaddr msg;
2953
2954 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2955 0, ifa_ifunlink_dispatch);
2956 msg.ifa = ifa;
2957 msg.ifp = ifp;
2958
2959 ifa_domsg(&msg.base.lmsg, 0);
2960 }
2961
2962 static void
2963 ifa_destroy_dispatch(netmsg_t nmsg)
2964 {
2965 struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2966
2967 IFAFREE(msg->ifa);
2968 ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2969 }
2970
2971 void
2972 ifa_destroy(struct ifaddr *ifa)
2973 {
2974 struct netmsg_ifaddr msg;
2975
2976 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2977 0, ifa_destroy_dispatch);
2978 msg.ifa = ifa;
2979
2980 ifa_domsg(&msg.base.lmsg, 0);
2981 }
2982
2983 struct lwkt_port *
2984 ifnet_portfn(int cpu)
2985 {
2986 return &ifnet_threads[cpu].td_msgport;
2987 }
2988
2989 void
2990 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2991 {
2992 KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2993
2994 if (next_cpu < ncpus)
2995 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2996 else
2997 lwkt_replymsg(lmsg, 0);
2998 }
2999
3000 int
3001 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
3002 {
3003 KKASSERT(cpu < ncpus);
3004 return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
3005 }
3006
3007 void
3008 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
3009 {
3010 KKASSERT(cpu < ncpus);
3011 lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
3012 }
3013
3014 /*
3015 * Generic netmsg service loop. Some protocols may roll their own but all
3016 * must do the basic command dispatch function call done here.
3017 */
3018 static void
3019 ifnet_service_loop(void *arg __unused)
3020 {
3021 netmsg_t msg;
3022
3023 while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
3024 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
3025 msg->base.nm_dispatch(msg);
3026 }
3027 }
3028
3029 static void
3030 if_start_rollup(void)
3031 {
3032 struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3033 struct ifsubq_stage *stage;
3034
3035 while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3036 struct ifaltq_subque *ifsq = stage->stg_subq;
3037 int is_sched = 0;
3038
3039 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3040 is_sched = 1;
3041 ifsq_stage_remove(head, stage);
3042
3043 if (is_sched) {
3044 ifsq_ifstart_schedule(ifsq, 1);
3045 } else {
3046 int start = 0;
3047
3048 ALTQ_SQ_LOCK(ifsq);
3049 if (!ifsq_is_started(ifsq)) {
3050 /*
3051 * Hold the subqueue interlock of
3052 * ifnet.if_start
3053 */
3054 ifsq_set_started(ifsq);
3055 start = 1;
3056 }
3057 ALTQ_SQ_UNLOCK(ifsq);
3058
3059 if (start)
3060 ifsq_ifstart_try(ifsq, 1);
3061 }
3062 KKASSERT((stage->stg_flags &
3063 (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3064 }
3065 }
3066
3067 static void
3068 ifnetinit(void *dummy __unused)
3069 {
3070 int i;
3071
3072 for (i = 0; i < ncpus; ++i) {
3073 struct thread *thr = &ifnet_threads[i];
3074
3075 lwkt_create(ifnet_service_loop, NULL, NULL,
3076 thr, TDF_NOSTART|TDF_FORCE_SPINPORT|TDF_FIXEDCPU,
3077 i, "ifnet %d", i);
3078 netmsg_service_port_init(&thr->td_msgport);
3079 lwkt_schedule(thr);
3080 }
3081
3082 for (i = 0; i < ncpus; ++i)
3083 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3084 netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3085 }
3086
3087 struct ifnet *
3088 ifnet_byindex(unsigned short idx)
3089 {
3090 if (idx > if_index)
3091 return NULL;
3092 return ifindex2ifnet[idx];
3093 }
3094
3095 struct ifaddr *
3096 ifaddr_byindex(unsigned short idx)
3097 {
3098 struct ifnet *ifp;
3099
3100 ifp = ifnet_byindex(idx);
3101 if (!ifp)
3102 return NULL;
3103 return TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
3104 }
3105
3106 void
3107 if_register_com_alloc(u_char type,
3108 if_com_alloc_t *a, if_com_free_t *f)
3109 {
3110
3111 KASSERT(if_com_alloc[type] == NULL,
3112 ("if_register_com_alloc: %d already registered", type));
3113 KASSERT(if_com_free[type] == NULL,
3114 ("if_register_com_alloc: %d free already registered", type));
3115
3116 if_com_alloc[type] = a;
3117 if_com_free[type] = f;
3118 }
3119
3120 void
3121 if_deregister_com_alloc(u_char type)
3122 {
3123
3124 KASSERT(if_com_alloc[type] != NULL,
3125 ("if_deregister_com_alloc: %d not registered", type));
3126 KASSERT(if_com_free[type] != NULL,
3127 ("if_deregister_com_alloc: %d free not registered", type));
3128 if_com_alloc[type] = NULL;
3129 if_com_free[type] = NULL;
3130 }
3131
3132 int
3133 if_ring_count2(int cnt, int cnt_max)
3134 {
3135 int shift = 0;
3136
3137 KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3138 ("invalid ring count max %d", cnt_max));
3139
3140 if (cnt <= 0)
3141 cnt = cnt_max;
3142 if (cnt > ncpus2)
3143 cnt = ncpus2;
3144 if (cnt > cnt_max)
3145 cnt = cnt_max;
3146
3147 while ((1 << (shift + 1)) <= cnt)
3148 ++shift;
3149 cnt = 1 << shift;
3150
3151 KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3152 ("calculate cnt %d, ncpus2 %d, cnt max %d",
3153 cnt, ncpus2, cnt_max));
3154 return cnt;
3155 }
3156
3157 void
3158 ifq_set_maxlen(struct ifaltq *ifq, int len)
3159 {
3160 ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3161 }
3162
3163 int
3164 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3165 {
3166 return ALTQ_SUBQ_INDEX_DEFAULT;
3167 }
3168
3169 int
3170 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3171 {
3172 return (cpuid & ifq->altq_subq_mask);
3173 }
3174
3175 static void
3176 ifsq_watchdog(void *arg)
3177 {
3178 struct ifsubq_watchdog *wd = arg;
3179 struct ifnet *ifp;
3180
3181 if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3182 goto done;
3183
3184 ifp = ifsq_get_ifp(wd->wd_subq);
3185 if (ifnet_tryserialize_all(ifp)) {
3186 wd->wd_watchdog(wd->wd_subq);
3187 ifnet_deserialize_all(ifp);
3188 } else {
3189 /* try again next timeout */
3190 wd->wd_timer = 1;
3191 }
3192 done:
3193 ifsq_watchdog_reset(wd);
3194 }
3195
3196 static void
3197 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3198 {
3199 callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3200 ifsq_get_cpuid(wd->wd_subq));
3201 }
3202
3203 void
3204 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3205 ifsq_watchdog_t watchdog)
3206 {
3207 callout_init_mp(&wd->wd_callout);
3208 wd->wd_timer = 0;
3209 wd->wd_subq = ifsq;
3210 wd->wd_watchdog = watchdog;
3211 }
3212
3213 void
3214 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3215 {
3216 wd->wd_timer = 0;
3217 ifsq_watchdog_reset(wd);
3218 }
3219
3220 void
3221 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3222 {
3223 wd->wd_timer = 0;
3224 callout_stop(&wd->wd_callout);
3225 }
Cache object: 0b3ea36c5bf4cf35e601c5bee799aa8e
|