sys/net/if.c

   1 /*
   2  * Copyright (c) 1980, 1986, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. All advertising materials mentioning features or use of this software
  14  *    must display the following acknowledgement:
  15  *      This product includes software developed by the University of
  16  *      California, Berkeley and its contributors.
  17  * 4. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  *      @(#)if.c        8.3 (Berkeley) 1/4/94
  34  * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
  35  * $DragonFly: src/sys/net/if.c,v 1.84 2008/11/15 11:58:16 sephe Exp $
  36  */
  37
  38 #include "opt_compat.h"
  39 #include "opt_inet6.h"
  40 #include "opt_inet.h"
  41 #include "opt_polling.h"
  42 #include "opt_ifpoll.h"
  43
  44 #include <sys/param.h>
  45 #include <sys/malloc.h>
  46 #include <sys/mbuf.h>
  47 #include <sys/systm.h>
  48 #include <sys/proc.h>
  49 #include <sys/priv.h>
  50 #include <sys/protosw.h>
  51 #include <sys/socket.h>
  52 #include <sys/socketvar.h>
  53 #include <sys/socketops.h>
  54 #include <sys/protosw.h>
  55 #include <sys/kernel.h>
  56 #include <sys/ktr.h>
  57 #include <sys/sockio.h>
  58 #include <sys/syslog.h>
  59 #include <sys/sysctl.h>
  60 #include <sys/domain.h>
  61 #include <sys/thread.h>
  62 #include <sys/thread2.h>
  63 #include <sys/serialize.h>
  64 #include <sys/msgport2.h>
  65 #include <sys/bus.h>
  66
  67 #include <net/if.h>
  68 #include <net/if_arp.h>
  69 #include <net/if_dl.h>
  70 #include <net/if_types.h>
  71 #include <net/if_var.h>
  72 #include <net/ifq_var.h>
  73 #include <net/radix.h>
  74 #include <net/route.h>
  75 #include <net/if_clone.h>
  76 #include <net/netisr.h>
  77 #include <net/netmsg2.h>
  78
  79 #include <machine/atomic.h>
  80 #include <machine/stdarg.h>
  81 #include <machine/smp.h>
  82
  83 #if defined(INET) || defined(INET6)
  84 /*XXX*/
  85 #include <netinet/in.h>
  86 #include <netinet/in_var.h>
  87 #include <netinet/if_ether.h>
  88 #ifdef INET6
  89 #include <netinet6/in6_var.h>
  90 #include <netinet6/in6_ifattach.h>
  91 #endif
  92 #endif
  93
  94 #if defined(COMPAT_43)
  95 #include <emulation/43bsd/43bsd_socket.h>
  96 #endif /* COMPAT_43 */
  97
  98 struct netmsg_ifaddr {
  99         struct netmsg_base base;
 100         struct ifaddr   *ifa;
 101         struct ifnet    *ifp;
 102         int             tail;
 103 };
 104
 105 /*
 106  * System initialization
 107  */
 108 static void     if_attachdomain(void *);
 109 static void     if_attachdomain1(struct ifnet *);
 110 static int      ifconf(u_long, caddr_t, struct ucred *);
 111 static void     ifinit(void *);
 112 static void     ifnetinit(void *);
 113 static void     if_slowtimo(void *);
 114 static void     link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
 115 static int      if_rtdel(struct radix_node *, void *);
 116
 117 #ifdef INET6
 118 /*
 119  * XXX: declare here to avoid to include many inet6 related files..
 120  * should be more generalized?
 121  */
 122 extern void     nd6_setmtu(struct ifnet *);
 123 #endif
 124
 125 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
 126 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
 127
 128 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
 129 /* Must be after netisr_init */
 130 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
 131
 132 static  if_com_alloc_t *if_com_alloc[256];
 133 static  if_com_free_t *if_com_free[256];
 134
 135 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
 136 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
 137 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
 138
 139 int                     ifqmaxlen = IFQ_MAXLEN;
 140 struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
 141
 142 /* In ifq_dispatch(), try to do direct ifnet.if_start first */
 143 static int              ifq_dispatch_schedonly = 0;
 144 SYSCTL_INT(_net_link_generic, OID_AUTO, ifq_dispatch_schedonly, CTLFLAG_RW,
 145            &ifq_dispatch_schedonly, 0, "");
 146
 147 /* In ifq_dispatch(), schedule ifnet.if_start without checking ifnet.if_snd */
 148 static int              ifq_dispatch_schednochk = 0;
 149 SYSCTL_INT(_net_link_generic, OID_AUTO, ifq_dispatch_schednochk, CTLFLAG_RW,
 150            &ifq_dispatch_schednochk, 0, "");
 151
 152 /* In if_devstart(), try to do direct ifnet.if_start first */
 153 static int              if_devstart_schedonly = 0;
 154 SYSCTL_INT(_net_link_generic, OID_AUTO, if_devstart_schedonly, CTLFLAG_RW,
 155            &if_devstart_schedonly, 0, "");
 156
 157 /* In if_devstart(), schedule ifnet.if_start without checking ifnet.if_snd */
 158 static int              if_devstart_schednochk = 0;
 159 SYSCTL_INT(_net_link_generic, OID_AUTO, if_devstart_schednochk, CTLFLAG_RW,
 160            &if_devstart_schednochk, 0, "");
 161
 162 #ifdef SMP
 163 /* Schedule ifnet.if_start on the current CPU */
 164 static int              if_start_oncpu_sched = 0;
 165 SYSCTL_INT(_net_link_generic, OID_AUTO, if_start_oncpu_sched, CTLFLAG_RW,
 166            &if_start_oncpu_sched, 0, "");
 167 #endif
 168
 169 struct callout          if_slowtimo_timer;
 170
 171 int                     if_index = 0;
 172 struct ifnet            **ifindex2ifnet = NULL;
 173 static struct thread    ifnet_threads[MAXCPU];
 174
 175 #define IFQ_KTR_STRING          "ifq=%p"
 176 #define IFQ_KTR_ARG_SIZE        (sizeof(void *))
 177 #ifndef KTR_IFQ
 178 #define KTR_IFQ                 KTR_ALL
 179 #endif
 180 KTR_INFO_MASTER(ifq);
 181 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARG_SIZE);
 182 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARG_SIZE);
 183 #define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)
 184
 185 #define IF_START_KTR_STRING     "ifp=%p"
 186 #define IF_START_KTR_ARG_SIZE   (sizeof(void *))
 187 #ifndef KTR_IF_START
 188 #define KTR_IF_START            KTR_ALL
 189 #endif
 190 KTR_INFO_MASTER(if_start);
 191 KTR_INFO(KTR_IF_START, if_start, run, 0,
 192          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
 193 KTR_INFO(KTR_IF_START, if_start, sched, 1,
 194          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
 195 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
 196          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
 197 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
 198          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
 199 #ifdef SMP
 200 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
 201          IF_START_KTR_STRING, IF_START_KTR_ARG_SIZE);
 202 #endif
 203 #define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
 204
 205 TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
 206
 207 /*
 208  * Network interface utility routines.
 209  *
 210  * Routines with ifa_ifwith* names take sockaddr *'s as
 211  * parameters.
 212  */
 213 /* ARGSUSED*/
 214 void
 215 ifinit(void *dummy)
 216 {
 217         struct ifnet *ifp;
 218
 219         callout_init(&if_slowtimo_timer);
 220
 221         crit_enter();
 222         TAILQ_FOREACH(ifp, &ifnet, if_link) {
 223                 if (ifp->if_snd.ifq_maxlen == 0) {
 224                         if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
 225                         ifp->if_snd.ifq_maxlen = ifqmaxlen;
 226                 }
 227         }
 228         crit_exit();
 229
 230         if_slowtimo(0);
 231 }
 232
 233 static int
 234 if_start_cpuid(struct ifnet *ifp)
 235 {
 236         return ifp->if_cpuid;
 237 }
 238
 239 #ifdef DEVICE_POLLING
 240 static int
 241 if_start_cpuid_poll(struct ifnet *ifp)
 242 {
 243         int poll_cpuid = ifp->if_poll_cpuid;
 244
 245         if (poll_cpuid >= 0)
 246                 return poll_cpuid;
 247         else
 248                 return ifp->if_cpuid;
 249 }
 250 #endif
 251
 252 static void
 253 if_start_ipifunc(void *arg)
 254 {
 255         struct ifnet *ifp = arg;
 256         struct lwkt_msg *lmsg = &ifp->if_start_nmsg[mycpuid].lmsg;
 257
 258         crit_enter();
 259         if (lmsg->ms_flags & MSGF_DONE)
 260                 lwkt_sendmsg(ifnet_portfn(mycpuid), lmsg);
 261         crit_exit();
 262 }
 263
 264 /*
 265  * Schedule ifnet.if_start on ifnet's CPU
 266  */
 267 static void
 268 if_start_schedule(struct ifnet *ifp)
 269 {
 270 #ifdef SMP
 271         int cpu;
 272
 273         if (if_start_oncpu_sched)
 274                 cpu = mycpuid;
 275         else
 276                 cpu = ifp->if_start_cpuid(ifp);
 277
 278         if (cpu != mycpuid)
 279                 lwkt_send_ipiq(globaldata_find(cpu), if_start_ipifunc, ifp);
 280         else
 281 #endif
 282         if_start_ipifunc(ifp);
 283 }
 284
 285 /*
 286  * NOTE:
 287  * This function will release ifnet.if_start interlock,
 288  * if ifnet.if_start does not need to be scheduled
 289  */
 290 static __inline int
 291 if_start_need_schedule(struct ifaltq *ifq, int running)
 292 {
 293         if (!running || ifq_is_empty(ifq)
 294 #ifdef ALTQ
 295             || ifq->altq_tbr != NULL
 296 #endif
 297         ) {
 298                 ALTQ_LOCK(ifq);
 299                 /*
 300                  * ifnet.if_start interlock is released, if:
 301                  * 1) Hardware can not take any packets, due to
 302                  *    o  interface is marked down
 303                  *    o  hardware queue is full (IFF_OACTIVE)
 304                  *    Under the second situation, hardware interrupt
 305                  *    or polling(4) will call/schedule ifnet.if_start
 306                  *    when hardware queue is ready
 307                  * 2) There is not packet in the ifnet.if_snd.
 308                  *    Further ifq_dispatch or ifq_handoff will call/
 309                  *    schedule ifnet.if_start
 310                  * 3) TBR is used and it does not allow further
 311                  *    dequeueing.
 312                  *    TBR callout will call ifnet.if_start
 313                  */
 314                 if (!running || !ifq_data_ready(ifq)) {
 315                         ifq->altq_started = 0;
 316                         ALTQ_UNLOCK(ifq);
 317                         return 0;
 318                 }
 319                 ALTQ_UNLOCK(ifq);
 320         }
 321         return 1;
 322 }
 323
 324 static void
 325 if_start_dispatch(netmsg_t msg)
 326 {
 327         struct lwkt_msg *lmsg = &msg->base.lmsg;
 328         struct ifnet *ifp = lmsg->u.ms_resultp;
 329         struct ifaltq *ifq = &ifp->if_snd;
 330         int running = 0;
 331
 332         crit_enter();
 333         lwkt_replymsg(lmsg, 0); /* reply ASAP */
 334         crit_exit();
 335
 336 #ifdef SMP
 337         if (!if_start_oncpu_sched && mycpuid != ifp->if_start_cpuid(ifp)) {
 338                 /*
 339                  * If the ifnet is still up, we need to
 340                  * chase its CPU change.
 341                  */
 342                 if (ifp->if_flags & IFF_UP) {
 343                         logifstart(chase_sched, ifp);
 344                         if_start_schedule(ifp);
 345                         return;
 346                 } else {
 347                         goto check;
 348                 }
 349         }
 350 #endif
 351
 352         if (ifp->if_flags & IFF_UP) {
 353                 ifnet_serialize_tx(ifp); /* XXX try? */
 354                 if ((ifp->if_flags & IFF_OACTIVE) == 0) {
 355                         logifstart(run, ifp);
 356                         ifp->if_start(ifp);
 357                         if ((ifp->if_flags &
 358                         (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
 359                                 running = 1;
 360                 }
 361                 ifnet_deserialize_tx(ifp);
 362         }
 363 #ifdef SMP
 364 check:
 365 #endif
 366         if (if_start_need_schedule(ifq, running)) {
 367                 crit_enter();
 368                 if (lmsg->ms_flags & MSGF_DONE) { /* XXX necessary? */
 369                         logifstart(sched, ifp);
 370                         lwkt_sendmsg(ifnet_portfn(mycpuid), lmsg);
 371                 }
 372                 crit_exit();
 373         }
 374 }
 375
 376 /* Device driver ifnet.if_start helper function */
 377 void
 378 if_devstart(struct ifnet *ifp)
 379 {
 380         struct ifaltq *ifq = &ifp->if_snd;
 381         int running = 0;
 382
 383         ASSERT_IFNET_SERIALIZED_TX(ifp);
 384
 385         ALTQ_LOCK(ifq);
 386         if (ifq->altq_started || !ifq_data_ready(ifq)) {
 387                 logifstart(avoid, ifp);
 388                 ALTQ_UNLOCK(ifq);
 389                 return;
 390         }
 391         ifq->altq_started = 1;
 392         ALTQ_UNLOCK(ifq);
 393
 394         if (if_devstart_schedonly) {
 395                 /*
 396                  * Always schedule ifnet.if_start on ifnet's CPU,
 397                  * short circuit the rest of this function.
 398                  */
 399                 logifstart(sched, ifp);
 400                 if_start_schedule(ifp);
 401                 return;
 402         }
 403
 404         logifstart(run, ifp);
 405         ifp->if_start(ifp);
 406
 407         if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
 408                 running = 1;
 409
 410         if (if_devstart_schednochk || if_start_need_schedule(ifq, running)) {
 411                 /*
 412                  * More data need to be transmitted, ifnet.if_start is
 413                  * scheduled on ifnet's CPU, and we keep going.
 414                  * NOTE: ifnet.if_start interlock is not released.
 415                  */
 416                 logifstart(sched, ifp);
 417                 if_start_schedule(ifp);
 418         }
 419 }
 420
 421 static void
 422 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 423 {
 424         lwkt_serialize_enter(ifp->if_serializer);
 425 }
 426
 427 static void
 428 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 429 {
 430         lwkt_serialize_exit(ifp->if_serializer);
 431 }
 432
 433 static int
 434 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 435 {
 436         return lwkt_serialize_try(ifp->if_serializer);
 437 }
 438
 439 #ifdef INVARIANTS
 440 static void
 441 if_default_serialize_assert(struct ifnet *ifp,
 442                             enum ifnet_serialize slz __unused,
 443                             boolean_t serialized)
 444 {
 445         if (serialized)
 446                 ASSERT_SERIALIZED(ifp->if_serializer);
 447         else
 448                 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
 449 }
 450 #endif
 451
 452 /*
 453  * Attach an interface to the list of "active" interfaces.
 454  *
 455  * The serializer is optional.  If non-NULL access to the interface
 456  * may be MPSAFE.
 457  */
 458 void
 459 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
 460 {
 461         unsigned socksize, ifasize;
 462         int namelen, masklen;
 463         struct sockaddr_dl *sdl;
 464         struct ifaddr *ifa;
 465         struct ifaltq *ifq;
 466         int i;
 467
 468         static int if_indexlim = 8;
 469
 470         if (ifp->if_serialize != NULL) {
 471                 KASSERT(ifp->if_deserialize != NULL &&
 472                         ifp->if_tryserialize != NULL &&
 473                         ifp->if_serialize_assert != NULL,
 474                         ("serialize functions are partially setup\n"));
 475
 476                 /*
 477                  * If the device supplies serialize functions,
 478                  * then clear if_serializer to catch any invalid
 479                  * usage of this field.
 480                  */
 481                 KASSERT(serializer == NULL,
 482                         ("both serialize functions and default serializer "
 483                          "are supplied\n"));
 484                 ifp->if_serializer = NULL;
 485         } else {
 486                 KASSERT(ifp->if_deserialize == NULL &&
 487                         ifp->if_tryserialize == NULL &&
 488                         ifp->if_serialize_assert == NULL,
 489                         ("serialize functions are partially setup\n"));
 490                 ifp->if_serialize = if_default_serialize;
 491                 ifp->if_deserialize = if_default_deserialize;
 492                 ifp->if_tryserialize = if_default_tryserialize;
 493 #ifdef INVARIANTS
 494                 ifp->if_serialize_assert = if_default_serialize_assert;
 495 #endif
 496
 497                 /*
 498                  * The serializer can be passed in from the device,
 499                  * allowing the same serializer to be used for both
 500                  * the interrupt interlock and the device queue.
 501                  * If not specified, the netif structure will use an
 502                  * embedded serializer.
 503                  */
 504                 if (serializer == NULL) {
 505                         serializer = &ifp->if_default_serializer;
 506                         lwkt_serialize_init(serializer);
 507                 }
 508                 ifp->if_serializer = serializer;
 509         }
 510
 511         ifp->if_start_cpuid = if_start_cpuid;
 512         ifp->if_cpuid = 0;
 513
 514 #ifdef DEVICE_POLLING
 515         /* Device is not in polling mode by default */
 516         ifp->if_poll_cpuid = -1;
 517         if (ifp->if_poll != NULL)
 518                 ifp->if_start_cpuid = if_start_cpuid_poll;
 519 #endif
 520
 521         ifp->if_start_nmsg = kmalloc(ncpus * sizeof(*ifp->if_start_nmsg),
 522                                      M_LWKTMSG, M_WAITOK);
 523         for (i = 0; i < ncpus; ++i) {
 524                 netmsg_init(&ifp->if_start_nmsg[i], NULL, &netisr_adone_rport,
 525                             0, if_start_dispatch);
 526                 ifp->if_start_nmsg[i].lmsg.u.ms_resultp = ifp;
 527         }
 528
 529         TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
 530         ifp->if_index = ++if_index;
 531
 532         /*
 533          * XXX -
 534          * The old code would work if the interface passed a pre-existing
 535          * chain of ifaddrs to this code.  We don't trust our callers to
 536          * properly initialize the tailq, however, so we no longer allow
 537          * this unlikely case.
 538          */
 539         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
 540                                     M_IFADDR, M_WAITOK | M_ZERO);
 541         for (i = 0; i < ncpus; ++i)
 542                 TAILQ_INIT(&ifp->if_addrheads[i]);
 543
 544         TAILQ_INIT(&ifp->if_prefixhead);
 545         TAILQ_INIT(&ifp->if_multiaddrs);
 546         TAILQ_INIT(&ifp->if_groups);
 547         getmicrotime(&ifp->if_lastchange);
 548         if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
 549                 unsigned int n;
 550                 struct ifnet **q;
 551
 552                 if_indexlim <<= 1;
 553
 554                 /* grow ifindex2ifnet */
 555                 n = if_indexlim * sizeof(*q);
 556                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
 557                 if (ifindex2ifnet) {
 558                         bcopy(ifindex2ifnet, q, n/2);
 559                         kfree(ifindex2ifnet, M_IFADDR);
 560                 }
 561                 ifindex2ifnet = q;
 562         }
 563
 564         ifindex2ifnet[if_index] = ifp;
 565
 566         /*
 567          * create a Link Level name for this device
 568          */
 569         namelen = strlen(ifp->if_xname);
 570 #define _offsetof(t, m) ((int)((caddr_t)&((t *)0)->m))
 571         masklen = _offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
 572         socksize = masklen + ifp->if_addrlen;
 573 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
 574         if (socksize < sizeof(*sdl))
 575                 socksize = sizeof(*sdl);
 576         socksize = ROUNDUP(socksize);
 577         ifasize = sizeof(struct ifaddr) + 2 * socksize;
 578         ifa = ifa_create(ifasize, M_WAITOK);
 579         sdl = (struct sockaddr_dl *)(ifa + 1);
 580         sdl->sdl_len = socksize;
 581         sdl->sdl_family = AF_LINK;
 582         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
 583         sdl->sdl_nlen = namelen;
 584         sdl->sdl_index = ifp->if_index;
 585         sdl->sdl_type = ifp->if_type;
 586         ifp->if_lladdr = ifa;
 587         ifa->ifa_ifp = ifp;
 588         ifa->ifa_rtrequest = link_rtrequest;
 589         ifa->ifa_addr = (struct sockaddr *)sdl;
 590         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
 591         ifa->ifa_netmask = (struct sockaddr *)sdl;
 592         sdl->sdl_len = masklen;
 593         while (namelen != 0)
 594                 sdl->sdl_data[--namelen] = 0xff;
 595         ifa_iflink(ifa, ifp, 0 /* Insert head */);
 596
 597         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
 598         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
 599
 600         ifq = &ifp->if_snd;
 601         ifq->altq_type = 0;
 602         ifq->altq_disc = NULL;
 603         ifq->altq_flags &= ALTQF_CANTCHANGE;
 604         ifq->altq_tbr = NULL;
 605         ifq->altq_ifp = ifp;
 606         ifq->altq_started = 0;
 607         ifq->altq_prepended = NULL;
 608         ALTQ_LOCK_INIT(ifq);
 609         ifq_set_classic(ifq);
 610
 611         if (!SLIST_EMPTY(&domains))
 612                 if_attachdomain1(ifp);
 613
 614         /* Announce the interface. */
 615         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
 616 }
 617
 618 static void
 619 if_attachdomain(void *dummy)
 620 {
 621         struct ifnet *ifp;
 622
 623         crit_enter();
 624         TAILQ_FOREACH(ifp, &ifnet, if_list)
 625                 if_attachdomain1(ifp);
 626         crit_exit();
 627 }
 628 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
 629         if_attachdomain, NULL);
 630
 631 static void
 632 if_attachdomain1(struct ifnet *ifp)
 633 {
 634         struct domain *dp;
 635
 636         crit_enter();
 637
 638         /* address family dependent data region */
 639         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
 640         SLIST_FOREACH(dp, &domains, dom_next)
 641                 if (dp->dom_ifattach)
 642                         ifp->if_afdata[dp->dom_family] =
 643                                 (*dp->dom_ifattach)(ifp);
 644         crit_exit();
 645 }
 646
 647 /*
 648  * Purge all addresses whose type is _not_ AF_LINK
 649  */
 650 void
 651 if_purgeaddrs_nolink(struct ifnet *ifp)
 652 {
 653         struct ifaddr_container *ifac, *next;
 654
 655         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
 656                               ifa_link, next) {
 657                 struct ifaddr *ifa = ifac->ifa;
 658
 659                 /* Leave link ifaddr as it is */
 660                 if (ifa->ifa_addr->sa_family == AF_LINK)
 661                         continue;
 662 #ifdef INET
 663                 /* XXX: Ugly!! ad hoc just for INET */
 664                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
 665                         struct ifaliasreq ifr;
 666 #ifdef IFADDR_DEBUG_VERBOSE
 667                         int i;
 668
 669                         kprintf("purge in4 addr %p: ", ifa);
 670                         for (i = 0; i < ncpus; ++i)
 671                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 672                         kprintf("\n");
 673 #endif
 674
 675                         bzero(&ifr, sizeof ifr);
 676                         ifr.ifra_addr = *ifa->ifa_addr;
 677                         if (ifa->ifa_dstaddr)
 678                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
 679                         if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
 680                                        NULL) == 0)
 681                                 continue;
 682                 }
 683 #endif /* INET */
 684 #ifdef INET6
 685                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
 686 #ifdef IFADDR_DEBUG_VERBOSE
 687                         int i;
 688
 689                         kprintf("purge in6 addr %p: ", ifa);
 690                         for (i = 0; i < ncpus; ++i)
 691                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 692                         kprintf("\n");
 693 #endif
 694
 695                         in6_purgeaddr(ifa);
 696                         /* ifp_addrhead is already updated */
 697                         continue;
 698                 }
 699 #endif /* INET6 */
 700                 ifa_ifunlink(ifa, ifp);
 701                 ifa_destroy(ifa);
 702         }
 703 }
 704
 705 /*
 706  * Detach an interface, removing it from the
 707  * list of "active" interfaces.
 708  */
 709 void
 710 if_detach(struct ifnet *ifp)
 711 {
 712         struct radix_node_head  *rnh;
 713         int i;
 714         int cpu, origcpu;
 715         struct domain *dp;
 716
 717         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
 718
 719         /*
 720          * Remove routes and flush queues.
 721          */
 722         crit_enter();
 723 #ifdef DEVICE_POLLING
 724         if (ifp->if_flags & IFF_POLLING)
 725                 ether_poll_deregister(ifp);
 726 #endif
 727 #ifdef IFPOLL_ENABLE
 728         if (ifp->if_flags & IFF_NPOLLING)
 729                 ifpoll_deregister(ifp);
 730 #endif
 731         if_down(ifp);
 732
 733 #ifdef ALTQ
 734         if (ifq_is_enabled(&ifp->if_snd))
 735                 altq_disable(&ifp->if_snd);
 736         if (ifq_is_attached(&ifp->if_snd))
 737                 altq_detach(&ifp->if_snd);
 738 #endif
 739
 740         /*
 741          * Clean up all addresses.
 742          */
 743         ifp->if_lladdr = NULL;
 744
 745         if_purgeaddrs_nolink(ifp);
 746         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
 747                 struct ifaddr *ifa;
 748
 749                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
 750                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
 751                         ("non-link ifaddr is left on if_addrheads"));
 752
 753                 ifa_ifunlink(ifa, ifp);
 754                 ifa_destroy(ifa);
 755                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
 756                         ("there are still ifaddrs left on if_addrheads"));
 757         }
 758
 759 #ifdef INET
 760         /*
 761          * Remove all IPv4 kernel structures related to ifp.
 762          */
 763         in_ifdetach(ifp);
 764 #endif
 765
 766 #ifdef INET6
 767         /*
 768          * Remove all IPv6 kernel structs related to ifp.  This should be done
 769          * before removing routing entries below, since IPv6 interface direct
 770          * routes are expected to be removed by the IPv6-specific kernel API.
 771          * Otherwise, the kernel will detect some inconsistency and bark it.
 772          */
 773         in6_ifdetach(ifp);
 774 #endif
 775
 776         /*
 777          * Delete all remaining routes using this interface
 778          * Unfortuneatly the only way to do this is to slog through
 779          * the entire routing table looking for routes which point
 780          * to this interface...oh well...
 781          */
 782         origcpu = mycpuid;
 783         for (cpu = 0; cpu < ncpus2; cpu++) {
 784                 lwkt_migratecpu(cpu);
 785                 for (i = 1; i <= AF_MAX; i++) {
 786                         if ((rnh = rt_tables[cpu][i]) == NULL)
 787                                 continue;
 788                         rnh->rnh_walktree(rnh, if_rtdel, ifp);
 789                 }
 790         }
 791         lwkt_migratecpu(origcpu);
 792
 793         /* Announce that the interface is gone. */
 794         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
 795         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
 796
 797         SLIST_FOREACH(dp, &domains, dom_next)
 798                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
 799                         (*dp->dom_ifdetach)(ifp,
 800                                 ifp->if_afdata[dp->dom_family]);
 801
 802         /*
 803          * Remove interface from ifindex2ifp[] and maybe decrement if_index.
 804          */
 805         ifindex2ifnet[ifp->if_index] = NULL;
 806         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
 807                 if_index--;
 808
 809         TAILQ_REMOVE(&ifnet, ifp, if_link);
 810         kfree(ifp->if_addrheads, M_IFADDR);
 811         kfree(ifp->if_start_nmsg, M_LWKTMSG);
 812         crit_exit();
 813 }
 814
 815 /*
 816  * Create interface group without members
 817  */
 818 struct ifg_group *
 819 if_creategroup(const char *groupname)
 820 {
 821         struct ifg_group        *ifg = NULL;
 822
 823         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
 824             M_TEMP, M_NOWAIT)) == NULL)
 825                 return (NULL);
 826
 827         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
 828         ifg->ifg_refcnt = 0;
 829         ifg->ifg_carp_demoted = 0;
 830         TAILQ_INIT(&ifg->ifg_members);
 831 #if NPF > 0
 832         pfi_attach_ifgroup(ifg);
 833 #endif
 834         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
 835
 836         return (ifg);
 837 }
 838
 839 /*
 840  * Add a group to an interface
 841  */
 842 int
 843 if_addgroup(struct ifnet *ifp, const char *groupname)
 844 {
 845         struct ifg_list         *ifgl;
 846         struct ifg_group        *ifg = NULL;
 847         struct ifg_member       *ifgm;
 848
 849         if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
 850             groupname[strlen(groupname) - 1] <= '9')
 851                 return (EINVAL);
 852
 853         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
 854                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
 855                         return (EEXIST);
 856
 857         if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
 858                 return (ENOMEM);
 859
 860         if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
 861                 kfree(ifgl, M_TEMP);
 862                 return (ENOMEM);
 863         }
 864
 865         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
 866                 if (!strcmp(ifg->ifg_group, groupname))
 867                         break;
 868
 869         if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
 870                 kfree(ifgl, M_TEMP);
 871                 kfree(ifgm, M_TEMP);
 872                 return (ENOMEM);
 873         }
 874
 875         ifg->ifg_refcnt++;
 876         ifgl->ifgl_group = ifg;
 877         ifgm->ifgm_ifp = ifp;
 878
 879         TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
 880         TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
 881
 882 #if NPF > 0
 883         pfi_group_change(groupname);
 884 #endif
 885
 886         return (0);
 887 }
 888
 889 /*
 890  * Remove a group from an interface
 891  */
 892 int
 893 if_delgroup(struct ifnet *ifp, const char *groupname)
 894 {
 895         struct ifg_list         *ifgl;
 896         struct ifg_member       *ifgm;
 897
 898         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
 899                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
 900                         break;
 901         if (ifgl == NULL)
 902                 return (ENOENT);
 903
 904         TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
 905
 906         TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
 907                 if (ifgm->ifgm_ifp == ifp)
 908                         break;
 909
 910         if (ifgm != NULL) {
 911                 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
 912                 kfree(ifgm, M_TEMP);
 913         }
 914
 915         if (--ifgl->ifgl_group->ifg_refcnt == 0) {
 916                 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
 917 #if NPF > 0
 918                 pfi_detach_ifgroup(ifgl->ifgl_group);
 919 #endif
 920                 kfree(ifgl->ifgl_group, M_TEMP);
 921         }
 922
 923         kfree(ifgl, M_TEMP);
 924
 925 #if NPF > 0
 926         pfi_group_change(groupname);
 927 #endif
 928
 929         return (0);
 930 }
 931
 932 /*
 933  * Stores all groups from an interface in memory pointed
 934  * to by data
 935  */
 936 int
 937 if_getgroup(caddr_t data, struct ifnet *ifp)
 938 {
 939         int                      len, error;
 940         struct ifg_list         *ifgl;
 941         struct ifg_req           ifgrq, *ifgp;
 942         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
 943
 944         if (ifgr->ifgr_len == 0) {
 945                 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
 946                         ifgr->ifgr_len += sizeof(struct ifg_req);
 947                 return (0);
 948         }
 949
 950         len = ifgr->ifgr_len;
 951         ifgp = ifgr->ifgr_groups;
 952         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
 953                 if (len < sizeof(ifgrq))
 954                         return (EINVAL);
 955                 bzero(&ifgrq, sizeof ifgrq);
 956                 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
 957                     sizeof(ifgrq.ifgrq_group));
 958                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
 959                     sizeof(struct ifg_req))))
 960                         return (error);
 961                 len -= sizeof(ifgrq);
 962                 ifgp++;
 963         }
 964
 965         return (0);
 966 }
 967
 968 /*
 969  * Stores all members of a group in memory pointed to by data
 970  */
 971 int
 972 if_getgroupmembers(caddr_t data)
 973 {
 974         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
 975         struct ifg_group        *ifg;
 976         struct ifg_member       *ifgm;
 977         struct ifg_req           ifgrq, *ifgp;
 978         int                      len, error;
 979
 980         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
 981                 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
 982                         break;
 983         if (ifg == NULL)
 984                 return (ENOENT);
 985
 986         if (ifgr->ifgr_len == 0) {
 987                 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
 988                         ifgr->ifgr_len += sizeof(ifgrq);
 989                 return (0);
 990         }
 991
 992         len = ifgr->ifgr_len;
 993         ifgp = ifgr->ifgr_groups;
 994         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
 995                 if (len < sizeof(ifgrq))
 996                         return (EINVAL);
 997                 bzero(&ifgrq, sizeof ifgrq);
 998                 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
 999                     sizeof(ifgrq.ifgrq_member));
1000                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1001                     sizeof(struct ifg_req))))
1002                         return (error);
1003                 len -= sizeof(ifgrq);
1004                 ifgp++;
1005         }
1006
1007         return (0);
1008 }
1009
1010 /*
1011  * Delete Routes for a Network Interface
1012  *
1013  * Called for each routing entry via the rnh->rnh_walktree() call above
1014  * to delete all route entries referencing a detaching network interface.
1015  *
1016  * Arguments:
1017  *      rn      pointer to node in the routing table
1018  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1019  *
1020  * Returns:
1021  *      0       successful
1022  *      errno   failed - reason indicated
1023  *
1024  */
1025 static int
1026 if_rtdel(struct radix_node *rn, void *arg)
1027 {
1028         struct rtentry  *rt = (struct rtentry *)rn;
1029         struct ifnet    *ifp = arg;
1030         int             err;
1031
1032         if (rt->rt_ifp == ifp) {
1033
1034                 /*
1035                  * Protect (sorta) against walktree recursion problems
1036                  * with cloned routes
1037                  */
1038                 if (!(rt->rt_flags & RTF_UP))
1039                         return (0);
1040
1041                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1042                                 rt_mask(rt), rt->rt_flags,
1043                                 NULL);
1044                 if (err) {
1045                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1046                 }
1047         }
1048
1049         return (0);
1050 }
1051
1052 /*
1053  * Locate an interface based on a complete address.
1054  */
1055 struct ifaddr *
1056 ifa_ifwithaddr(struct sockaddr *addr)
1057 {
1058         struct ifnet *ifp;
1059
1060         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1061                 struct ifaddr_container *ifac;
1062
1063                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1064                         struct ifaddr *ifa = ifac->ifa;
1065
1066                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1067                                 continue;
1068                         if (sa_equal(addr, ifa->ifa_addr))
1069                                 return (ifa);
1070                         if ((ifp->if_flags & IFF_BROADCAST) &&
1071                             ifa->ifa_broadaddr &&
1072                             /* IPv6 doesn't have broadcast */
1073                             ifa->ifa_broadaddr->sa_len != 0 &&
1074                             sa_equal(ifa->ifa_broadaddr, addr))
1075                                 return (ifa);
1076                 }
1077         }
1078         return (NULL);
1079 }
1080 /*
1081  * Locate the point to point interface with a given destination address.
1082  */
1083 struct ifaddr *
1084 ifa_ifwithdstaddr(struct sockaddr *addr)
1085 {
1086         struct ifnet *ifp;
1087
1088         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1089                 struct ifaddr_container *ifac;
1090
1091                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1092                         continue;
1093
1094                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1095                         struct ifaddr *ifa = ifac->ifa;
1096
1097                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1098                                 continue;
1099                         if (ifa->ifa_dstaddr &&
1100                             sa_equal(addr, ifa->ifa_dstaddr))
1101                                 return (ifa);
1102                 }
1103         }
1104         return (NULL);
1105 }
1106
1107 /*
1108  * Find an interface on a specific network.  If many, choice
1109  * is most specific found.
1110  */
1111 struct ifaddr *
1112 ifa_ifwithnet(struct sockaddr *addr)
1113 {
1114         struct ifnet *ifp;
1115         struct ifaddr *ifa_maybe = NULL;
1116         u_int af = addr->sa_family;
1117         char *addr_data = addr->sa_data, *cplim;
1118
1119         /*
1120          * AF_LINK addresses can be looked up directly by their index number,
1121          * so do that if we can.
1122          */
1123         if (af == AF_LINK) {
1124                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1125
1126                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1127                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1128         }
1129
1130         /*
1131          * Scan though each interface, looking for ones that have
1132          * addresses in this address family.
1133          */
1134         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1135                 struct ifaddr_container *ifac;
1136
1137                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1138                         struct ifaddr *ifa = ifac->ifa;
1139                         char *cp, *cp2, *cp3;
1140
1141                         if (ifa->ifa_addr->sa_family != af)
1142 next:                           continue;
1143                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1144                                 /*
1145                                  * This is a bit broken as it doesn't
1146                                  * take into account that the remote end may
1147                                  * be a single node in the network we are
1148                                  * looking for.
1149                                  * The trouble is that we don't know the
1150                                  * netmask for the remote end.
1151                                  */
1152                                 if (ifa->ifa_dstaddr != NULL &&
1153                                     sa_equal(addr, ifa->ifa_dstaddr))
1154                                         return (ifa);
1155                         } else {
1156                                 /*
1157                                  * if we have a special address handler,
1158                                  * then use it instead of the generic one.
1159                                  */
1160                                 if (ifa->ifa_claim_addr) {
1161                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1162                                                 return (ifa);
1163                                         } else {
1164                                                 continue;
1165                                         }
1166                                 }
1167
1168                                 /*
1169                                  * Scan all the bits in the ifa's address.
1170                                  * If a bit dissagrees with what we are
1171                                  * looking for, mask it with the netmask
1172                                  * to see if it really matters.
1173                                  * (A byte at a time)
1174                                  */
1175                                 if (ifa->ifa_netmask == 0)
1176                                         continue;
1177                                 cp = addr_data;
1178                                 cp2 = ifa->ifa_addr->sa_data;
1179                                 cp3 = ifa->ifa_netmask->sa_data;
1180                                 cplim = ifa->ifa_netmask->sa_len +
1181                                         (char *)ifa->ifa_netmask;
1182                                 while (cp3 < cplim)
1183                                         if ((*cp++ ^ *cp2++) & *cp3++)
1184                                                 goto next; /* next address! */
1185                                 /*
1186                                  * If the netmask of what we just found
1187                                  * is more specific than what we had before
1188                                  * (if we had one) then remember the new one
1189                                  * before continuing to search
1190                                  * for an even better one.
1191                                  */
1192                                 if (ifa_maybe == 0 ||
1193                                     rn_refines((char *)ifa->ifa_netmask,
1194                                                (char *)ifa_maybe->ifa_netmask))
1195                                         ifa_maybe = ifa;
1196                         }
1197                 }
1198         }
1199         return (ifa_maybe);
1200 }
1201
1202 /*
1203  * Find an interface address specific to an interface best matching
1204  * a given address.
1205  */
1206 struct ifaddr *
1207 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1208 {
1209         struct ifaddr_container *ifac;
1210         char *cp, *cp2, *cp3;
1211         char *cplim;
1212         struct ifaddr *ifa_maybe = 0;
1213         u_int af = addr->sa_family;
1214
1215         if (af >= AF_MAX)
1216                 return (0);
1217         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1218                 struct ifaddr *ifa = ifac->ifa;
1219
1220                 if (ifa->ifa_addr->sa_family != af)
1221                         continue;
1222                 if (ifa_maybe == 0)
1223                         ifa_maybe = ifa;
1224                 if (ifa->ifa_netmask == NULL) {
1225                         if (sa_equal(addr, ifa->ifa_addr) ||
1226                             (ifa->ifa_dstaddr != NULL &&
1227                              sa_equal(addr, ifa->ifa_dstaddr)))
1228                                 return (ifa);
1229                         continue;
1230                 }
1231                 if (ifp->if_flags & IFF_POINTOPOINT) {
1232                         if (sa_equal(addr, ifa->ifa_dstaddr))
1233                                 return (ifa);
1234                 } else {
1235                         cp = addr->sa_data;
1236                         cp2 = ifa->ifa_addr->sa_data;
1237                         cp3 = ifa->ifa_netmask->sa_data;
1238                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1239                         for (; cp3 < cplim; cp3++)
1240                                 if ((*cp++ ^ *cp2++) & *cp3)
1241                                         break;
1242                         if (cp3 == cplim)
1243                                 return (ifa);
1244                 }
1245         }
1246         return (ifa_maybe);
1247 }
1248
1249 /*
1250  * Default action when installing a route with a Link Level gateway.
1251  * Lookup an appropriate real ifa to point to.
1252  * This should be moved to /sys/net/link.c eventually.
1253  */
1254 static void
1255 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1256 {
1257         struct ifaddr *ifa;
1258         struct sockaddr *dst;
1259         struct ifnet *ifp;
1260
1261         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1262             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1263                 return;
1264         ifa = ifaof_ifpforaddr(dst, ifp);
1265         if (ifa != NULL) {
1266                 IFAFREE(rt->rt_ifa);
1267                 IFAREF(ifa);
1268                 rt->rt_ifa = ifa;
1269                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1270                         ifa->ifa_rtrequest(cmd, rt, info);
1271         }
1272 }
1273
1274 /*
1275  * Mark an interface down and notify protocols of
1276  * the transition.
1277  * NOTE: must be called at splnet or eqivalent.
1278  */
1279 void
1280 if_unroute(struct ifnet *ifp, int flag, int fam)
1281 {
1282         struct ifaddr_container *ifac;
1283
1284         ifp->if_flags &= ~flag;
1285         getmicrotime(&ifp->if_lastchange);
1286         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1287                 struct ifaddr *ifa = ifac->ifa;
1288
1289                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1290                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1291         }
1292         ifq_purge(&ifp->if_snd);
1293         rt_ifmsg(ifp);
1294 }
1295
1296 /*
1297  * Mark an interface up and notify protocols of
1298  * the transition.
1299  * NOTE: must be called at splnet or eqivalent.
1300  */
1301 void
1302 if_route(struct ifnet *ifp, int flag, int fam)
1303 {
1304         struct ifaddr_container *ifac;
1305
1306         ifq_purge(&ifp->if_snd);
1307         ifp->if_flags |= flag;
1308         getmicrotime(&ifp->if_lastchange);
1309         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1310                 struct ifaddr *ifa = ifac->ifa;
1311
1312                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1313                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1314         }
1315         rt_ifmsg(ifp);
1316 #ifdef INET6
1317         in6_if_up(ifp);
1318 #endif
1319 }
1320
1321 /*
1322  * Mark an interface down and notify protocols of the transition.  An
1323  * interface going down is also considered to be a synchronizing event.
1324  * We must ensure that all packet processing related to the interface
1325  * has completed before we return so e.g. the caller can free the ifnet
1326  * structure that the mbufs may be referencing.
1327  *
1328  * NOTE: must be called at splnet or eqivalent.
1329  */
1330 void
1331 if_down(struct ifnet *ifp)
1332 {
1333         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1334         netmsg_service_sync();
1335 }
1336
1337 /*
1338  * Mark an interface up and notify protocols of
1339  * the transition.
1340  * NOTE: must be called at splnet or eqivalent.
1341  */
1342 void
1343 if_up(struct ifnet *ifp)
1344 {
1345         if_route(ifp, IFF_UP, AF_UNSPEC);
1346 }
1347
1348 /*
1349  * Process a link state change.
1350  * NOTE: must be called at splsoftnet or equivalent.
1351  */
1352 void
1353 if_link_state_change(struct ifnet *ifp)
1354 {
1355         int link_state = ifp->if_link_state;
1356
1357         rt_ifmsg(ifp);
1358         devctl_notify("IFNET", ifp->if_xname,
1359             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1360 }
1361
1362 /*
1363  * Handle interface watchdog timer routines.  Called
1364  * from softclock, we decrement timers (if set) and
1365  * call the appropriate interface routine on expiration.
1366  */
1367 static void
1368 if_slowtimo(void *arg)
1369 {
1370         struct ifnet *ifp;
1371
1372         crit_enter();
1373
1374         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1375                 if (ifp->if_timer == 0 || --ifp->if_timer)
1376                         continue;
1377                 if (ifp->if_watchdog) {
1378                         if (ifnet_tryserialize_all(ifp)) {
1379                                 (*ifp->if_watchdog)(ifp);
1380                                 ifnet_deserialize_all(ifp);
1381                         } else {
1382                                 /* try again next timeout */
1383                                 ++ifp->if_timer;
1384                         }
1385                 }
1386         }
1387
1388         crit_exit();
1389
1390         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1391 }
1392
1393 /*
1394  * Map interface name to
1395  * interface structure pointer.
1396  */
1397 struct ifnet *
1398 ifunit(const char *name)
1399 {
1400         struct ifnet *ifp;
1401
1402         /*
1403          * Search all the interfaces for this name/number
1404          */
1405
1406         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1407                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1408                         break;
1409         }
1410         return (ifp);
1411 }
1412
1413
1414 /*
1415  * Map interface name in a sockaddr_dl to
1416  * interface structure pointer.
1417  */
1418 struct ifnet *
1419 if_withname(struct sockaddr *sa)
1420 {
1421         char ifname[IFNAMSIZ+1];
1422         struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1423
1424         if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1425              (sdl->sdl_nlen > IFNAMSIZ) )
1426                 return NULL;
1427
1428         /*
1429          * ifunit wants a null-terminated name.  It may not be null-terminated
1430          * in the sockaddr.  We don't want to change the caller's sockaddr,
1431          * and there might not be room to put the trailing null anyway, so we
1432          * make a local copy that we know we can null terminate safely.
1433          */
1434
1435         bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1436         ifname[sdl->sdl_nlen] = '\0';
1437         return ifunit(ifname);
1438 }
1439
1440
1441 /*
1442  * Interface ioctls.
1443  */
1444 int
1445 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1446 {
1447         struct ifnet *ifp;
1448         struct ifreq *ifr;
1449         struct ifstat *ifs;
1450         int error;
1451         short oif_flags;
1452         int new_flags;
1453         size_t namelen, onamelen;
1454         char new_name[IFNAMSIZ];
1455         struct ifaddr *ifa;
1456         struct sockaddr_dl *sdl;
1457
1458         switch (cmd) {
1459
1460         case SIOCGIFCONF:
1461         case OSIOCGIFCONF:
1462                 return (ifconf(cmd, data, cred));
1463         }
1464         ifr = (struct ifreq *)data;
1465
1466         switch (cmd) {
1467         case SIOCIFCREATE:
1468         case SIOCIFCREATE2:
1469                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1470                         return (error);
1471                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1472                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1473         case SIOCIFDESTROY:
1474                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1475                         return (error);
1476                 return (if_clone_destroy(ifr->ifr_name));
1477
1478         case SIOCIFGCLONERS:
1479                 return (if_clone_list((struct if_clonereq *)data));
1480         }
1481
1482         ifp = ifunit(ifr->ifr_name);
1483         if (ifp == 0)
1484                 return (ENXIO);
1485         switch (cmd) {
1486
1487         case SIOCGIFINDEX:
1488                 ifr->ifr_index = ifp->if_index;
1489                 break;
1490
1491         case SIOCGIFFLAGS:
1492                 ifr->ifr_flags = ifp->if_flags;
1493                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1494                 break;
1495
1496         case SIOCGIFCAP:
1497                 ifr->ifr_reqcap = ifp->if_capabilities;
1498                 ifr->ifr_curcap = ifp->if_capenable;
1499                 break;
1500
1501         case SIOCGIFMETRIC:
1502                 ifr->ifr_metric = ifp->if_metric;
1503                 break;
1504
1505         case SIOCGIFMTU:
1506                 ifr->ifr_mtu = ifp->if_mtu;
1507                 break;
1508
1509         case SIOCGIFDATA:
1510                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1511                     sizeof(ifp->if_data));
1512                 break;
1513
1514
1515         case SIOCGIFPHYS:
1516                 ifr->ifr_phys = ifp->if_physical;
1517                 break;
1518
1519         case SIOCGIFPOLLCPU:
1520 #ifdef DEVICE_POLLING
1521                 ifr->ifr_pollcpu = ifp->if_poll_cpuid;
1522 #else
1523                 ifr->ifr_pollcpu = -1;
1524 #endif
1525                 break;
1526
1527         case SIOCSIFPOLLCPU:
1528 #ifdef DEVICE_POLLING
1529                 if ((ifp->if_flags & IFF_POLLING) == 0)
1530                         ether_pollcpu_register(ifp, ifr->ifr_pollcpu);
1531 #endif
1532                 break;
1533
1534         case SIOCSIFFLAGS:
1535                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1536                 if (error)
1537                         return (error);
1538                 new_flags = (ifr->ifr_flags & 0xffff) |
1539                     (ifr->ifr_flagshigh << 16);
1540                 if (ifp->if_flags & IFF_SMART) {
1541                         /* Smart drivers twiddle their own routes */
1542                 } else if (ifp->if_flags & IFF_UP &&
1543                     (new_flags & IFF_UP) == 0) {
1544                         crit_enter();
1545                         if_down(ifp);
1546                         crit_exit();
1547                 } else if (new_flags & IFF_UP &&
1548                     (ifp->if_flags & IFF_UP) == 0) {
1549                         crit_enter();
1550                         if_up(ifp);
1551                         crit_exit();
1552                 }
1553
1554 #ifdef DEVICE_POLLING
1555                 if ((new_flags ^ ifp->if_flags) & IFF_POLLING) {
1556                         if (new_flags & IFF_POLLING) {
1557                                 ether_poll_register(ifp);
1558                         } else {
1559                                 ether_poll_deregister(ifp);
1560                         }
1561                 }
1562 #endif
1563 #ifdef IFPOLL_ENABLE
1564                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1565                         if (new_flags & IFF_NPOLLING)
1566                                 ifpoll_register(ifp);
1567                         else
1568                                 ifpoll_deregister(ifp);
1569                 }
1570 #endif
1571
1572                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1573                         (new_flags &~ IFF_CANTCHANGE);
1574                 if (new_flags & IFF_PPROMISC) {
1575                         /* Permanently promiscuous mode requested */
1576                         ifp->if_flags |= IFF_PROMISC;
1577                 } else if (ifp->if_pcount == 0) {
1578                         ifp->if_flags &= ~IFF_PROMISC;
1579                 }
1580                 if (ifp->if_ioctl) {
1581                         ifnet_serialize_all(ifp);
1582                         ifp->if_ioctl(ifp, cmd, data, cred);
1583                         ifnet_deserialize_all(ifp);
1584                 }
1585                 getmicrotime(&ifp->if_lastchange);
1586                 break;
1587
1588         case SIOCSIFCAP:
1589                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1590                 if (error)
1591                         return (error);
1592                 if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1593                         return (EINVAL);
1594                 ifnet_serialize_all(ifp);
1595                 ifp->if_ioctl(ifp, cmd, data, cred);
1596                 ifnet_deserialize_all(ifp);
1597                 break;
1598
1599         case SIOCSIFNAME:
1600                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1601                 if (error != 0)
1602                         return (error);
1603                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1604                 if (error != 0)
1605                         return (error);
1606                 if (new_name[0] == '\0')
1607                         return (EINVAL);
1608                 if (ifunit(new_name) != NULL)
1609                         return (EEXIST);
1610
1611                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1612
1613                 /* Announce the departure of the interface. */
1614                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1615
1616                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1617                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1618                 /* XXX IFA_LOCK(ifa); */
1619                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1620                 namelen = strlen(new_name);
1621                 onamelen = sdl->sdl_nlen;
1622                 /*
1623                  * Move the address if needed.  This is safe because we
1624                  * allocate space for a name of length IFNAMSIZ when we
1625                  * create this in if_attach().
1626                  */
1627                 if (namelen != onamelen) {
1628                         bcopy(sdl->sdl_data + onamelen,
1629                             sdl->sdl_data + namelen, sdl->sdl_alen);
1630                 }
1631                 bcopy(new_name, sdl->sdl_data, namelen);
1632                 sdl->sdl_nlen = namelen;
1633                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1634                 bzero(sdl->sdl_data, onamelen);
1635                 while (namelen != 0)
1636                         sdl->sdl_data[--namelen] = 0xff;
1637                 /* XXX IFA_UNLOCK(ifa) */
1638
1639                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1640
1641                 /* Announce the return of the interface. */
1642                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1643                 break;
1644
1645         case SIOCSIFMETRIC:
1646                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1647                 if (error)
1648                         return (error);
1649                 ifp->if_metric = ifr->ifr_metric;
1650                 getmicrotime(&ifp->if_lastchange);
1651                 break;
1652
1653         case SIOCSIFPHYS:
1654                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1655                 if (error)
1656                         return error;
1657                 if (!ifp->if_ioctl)
1658                         return EOPNOTSUPP;
1659                 ifnet_serialize_all(ifp);
1660                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1661                 ifnet_deserialize_all(ifp);
1662                 if (error == 0)
1663                         getmicrotime(&ifp->if_lastchange);
1664                 return (error);
1665
1666         case SIOCSIFMTU:
1667         {
1668                 u_long oldmtu = ifp->if_mtu;
1669
1670                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1671                 if (error)
1672                         return (error);
1673                 if (ifp->if_ioctl == NULL)
1674                         return (EOPNOTSUPP);
1675                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1676                         return (EINVAL);
1677                 ifnet_serialize_all(ifp);
1678                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1679                 ifnet_deserialize_all(ifp);
1680                 if (error == 0) {
1681                         getmicrotime(&ifp->if_lastchange);
1682                         rt_ifmsg(ifp);
1683                 }
1684                 /*
1685                  * If the link MTU changed, do network layer specific procedure.
1686                  */
1687                 if (ifp->if_mtu != oldmtu) {
1688 #ifdef INET6
1689                         nd6_setmtu(ifp);
1690 #endif
1691                 }
1692                 return (error);
1693         }
1694
1695         case SIOCADDMULTI:
1696         case SIOCDELMULTI:
1697                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1698                 if (error)
1699                         return (error);
1700
1701                 /* Don't allow group membership on non-multicast interfaces. */
1702                 if ((ifp->if_flags & IFF_MULTICAST) == 0)
1703                         return EOPNOTSUPP;
1704
1705                 /* Don't let users screw up protocols' entries. */
1706                 if (ifr->ifr_addr.sa_family != AF_LINK)
1707                         return EINVAL;
1708
1709                 if (cmd == SIOCADDMULTI) {
1710                         struct ifmultiaddr *ifma;
1711                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1712                 } else {
1713                         error = if_delmulti(ifp, &ifr->ifr_addr);
1714                 }
1715                 if (error == 0)
1716                         getmicrotime(&ifp->if_lastchange);
1717                 return error;
1718
1719         case SIOCSIFPHYADDR:
1720         case SIOCDIFPHYADDR:
1721 #ifdef INET6
1722         case SIOCSIFPHYADDR_IN6:
1723 #endif
1724         case SIOCSLIFPHYADDR:
1725         case SIOCSIFMEDIA:
1726         case SIOCSIFGENERIC:
1727                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1728                 if (error)
1729                         return (error);
1730                 if (ifp->if_ioctl == 0)
1731                         return (EOPNOTSUPP);
1732                 ifnet_serialize_all(ifp);
1733                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1734                 ifnet_deserialize_all(ifp);
1735                 if (error == 0)
1736                         getmicrotime(&ifp->if_lastchange);
1737                 return error;
1738
1739         case SIOCGIFSTATUS:
1740                 ifs = (struct ifstat *)data;
1741                 ifs->ascii[0] = '\0';
1742
1743         case SIOCGIFPSRCADDR:
1744         case SIOCGIFPDSTADDR:
1745         case SIOCGLIFPHYADDR:
1746         case SIOCGIFMEDIA:
1747         case SIOCGIFGENERIC:
1748                 if (ifp->if_ioctl == NULL)
1749                         return (EOPNOTSUPP);
1750                 ifnet_serialize_all(ifp);
1751                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1752                 ifnet_deserialize_all(ifp);
1753                 return (error);
1754
1755         case SIOCSIFLLADDR:
1756                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1757                 if (error)
1758                         return (error);
1759                 error = if_setlladdr(ifp,
1760                     ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1761                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1762                 return (error);
1763
1764         default:
1765                 oif_flags = ifp->if_flags;
1766                 if (so->so_proto == 0)
1767                         return (EOPNOTSUPP);
1768 #ifndef COMPAT_43
1769                 error = so_pru_control_direct(so, cmd, data, ifp);
1770 #else
1771             {
1772                 int ocmd = cmd;
1773
1774                 switch (cmd) {
1775
1776                 case SIOCSIFDSTADDR:
1777                 case SIOCSIFADDR:
1778                 case SIOCSIFBRDADDR:
1779                 case SIOCSIFNETMASK:
1780 #if BYTE_ORDER != BIG_ENDIAN
1781                         if (ifr->ifr_addr.sa_family == 0 &&
1782                             ifr->ifr_addr.sa_len < 16) {
1783                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1784                                 ifr->ifr_addr.sa_len = 16;
1785                         }
1786 #else
1787                         if (ifr->ifr_addr.sa_len == 0)
1788                                 ifr->ifr_addr.sa_len = 16;
1789 #endif
1790                         break;
1791
1792                 case OSIOCGIFADDR:
1793                         cmd = SIOCGIFADDR;
1794                         break;
1795
1796                 case OSIOCGIFDSTADDR:
1797                         cmd = SIOCGIFDSTADDR;
1798                         break;
1799
1800                 case OSIOCGIFBRDADDR:
1801                         cmd = SIOCGIFBRDADDR;
1802                         break;
1803
1804                 case OSIOCGIFNETMASK:
1805                         cmd = SIOCGIFNETMASK;
1806                 }
1807
1808                 error = so_pru_control_direct(so, cmd, data, ifp);
1809
1810                 switch (ocmd) {
1811                 case OSIOCGIFADDR:
1812                 case OSIOCGIFDSTADDR:
1813                 case OSIOCGIFBRDADDR:
1814                 case OSIOCGIFNETMASK:
1815                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1816                         break;
1817                 }
1818             }
1819 #endif /* COMPAT_43 */
1820
1821                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1822 #ifdef INET6
1823                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
1824                         if (ifp->if_flags & IFF_UP) {
1825                                 crit_enter();
1826                                 in6_if_up(ifp);
1827                                 crit_exit();
1828                         }
1829 #endif
1830                 }
1831                 return (error);
1832
1833         }
1834         return (0);
1835 }
1836
1837 /*
1838  * Set/clear promiscuous mode on interface ifp based on the truth value
1839  * of pswitch.  The calls are reference counted so that only the first
1840  * "on" request actually has an effect, as does the final "off" request.
1841  * Results are undefined if the "off" and "on" requests are not matched.
1842  */
1843 int
1844 ifpromisc(struct ifnet *ifp, int pswitch)
1845 {
1846         struct ifreq ifr;
1847         int error;
1848         int oldflags;
1849
1850         oldflags = ifp->if_flags;
1851         if (ifp->if_flags & IFF_PPROMISC) {
1852                 /* Do nothing if device is in permanently promiscuous mode */
1853                 ifp->if_pcount += pswitch ? 1 : -1;
1854                 return (0);
1855         }
1856         if (pswitch) {
1857                 /*
1858                  * If the device is not configured up, we cannot put it in
1859                  * promiscuous mode.
1860                  */
1861                 if ((ifp->if_flags & IFF_UP) == 0)
1862                         return (ENETDOWN);
1863                 if (ifp->if_pcount++ != 0)
1864                         return (0);
1865                 ifp->if_flags |= IFF_PROMISC;
1866                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1867                     ifp->if_xname);
1868         } else {
1869                 if (--ifp->if_pcount > 0)
1870                         return (0);
1871                 ifp->if_flags &= ~IFF_PROMISC;
1872                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
1873                     ifp->if_xname);
1874         }
1875         ifr.ifr_flags = ifp->if_flags;
1876         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1877         ifnet_serialize_all(ifp);
1878         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
1879         ifnet_deserialize_all(ifp);
1880         if (error == 0)
1881                 rt_ifmsg(ifp);
1882         else
1883                 ifp->if_flags = oldflags;
1884         return error;
1885 }
1886
1887 /*
1888  * Return interface configuration
1889  * of system.  List may be used
1890  * in later ioctl's (above) to get
1891  * other information.
1892  */
1893 static int
1894 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
1895 {
1896         struct ifconf *ifc = (struct ifconf *)data;
1897         struct ifnet *ifp;
1898         struct sockaddr *sa;
1899         struct ifreq ifr, *ifrp;
1900         int space = ifc->ifc_len, error = 0;
1901
1902         ifrp = ifc->ifc_req;
1903         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1904                 struct ifaddr_container *ifac;
1905                 int addrs;
1906
1907                 if (space <= sizeof ifr)
1908                         break;
1909
1910                 /*
1911                  * Zero the stack declared structure first to prevent
1912                  * memory disclosure.
1913                  */
1914                 bzero(&ifr, sizeof(ifr));
1915                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
1916                     >= sizeof(ifr.ifr_name)) {
1917                         error = ENAMETOOLONG;
1918                         break;
1919                 }
1920
1921                 addrs = 0;
1922                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1923                         struct ifaddr *ifa = ifac->ifa;
1924
1925                         if (space <= sizeof ifr)
1926                                 break;
1927                         sa = ifa->ifa_addr;
1928                         if (cred->cr_prison &&
1929                             prison_if(cred, sa))
1930                                 continue;
1931                         addrs++;
1932 #ifdef COMPAT_43
1933                         if (cmd == OSIOCGIFCONF) {
1934                                 struct osockaddr *osa =
1935                                          (struct osockaddr *)&ifr.ifr_addr;
1936                                 ifr.ifr_addr = *sa;
1937                                 osa->sa_family = sa->sa_family;
1938                                 error = copyout(&ifr, ifrp, sizeof ifr);
1939                                 ifrp++;
1940                         } else
1941 #endif
1942                         if (sa->sa_len <= sizeof(*sa)) {
1943                                 ifr.ifr_addr = *sa;
1944                                 error = copyout(&ifr, ifrp, sizeof ifr);
1945                                 ifrp++;
1946                         } else {
1947                                 if (space < (sizeof ifr) + sa->sa_len -
1948                                             sizeof(*sa))
1949                                         break;
1950                                 space -= sa->sa_len - sizeof(*sa);
1951                                 error = copyout(&ifr, ifrp,
1952                                                 sizeof ifr.ifr_name);
1953                                 if (error == 0)
1954                                         error = copyout(sa, &ifrp->ifr_addr,
1955                                                         sa->sa_len);
1956                                 ifrp = (struct ifreq *)
1957                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
1958                         }
1959                         if (error)
1960                                 break;
1961                         space -= sizeof ifr;
1962                 }
1963                 if (error)
1964                         break;
1965                 if (!addrs) {
1966                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
1967                         error = copyout(&ifr, ifrp, sizeof ifr);
1968                         if (error)
1969                                 break;
1970                         space -= sizeof ifr;
1971                         ifrp++;
1972                 }
1973         }
1974         ifc->ifc_len -= space;
1975         return (error);
1976 }
1977
1978 /*
1979  * Just like if_promisc(), but for all-multicast-reception mode.
1980  */
1981 int
1982 if_allmulti(struct ifnet *ifp, int onswitch)
1983 {
1984         int error = 0;
1985         struct ifreq ifr;
1986
1987         crit_enter();
1988
1989         if (onswitch) {
1990                 if (ifp->if_amcount++ == 0) {
1991                         ifp->if_flags |= IFF_ALLMULTI;
1992                         ifr.ifr_flags = ifp->if_flags;
1993                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1994                         ifnet_serialize_all(ifp);
1995                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1996                                               NULL);
1997                         ifnet_deserialize_all(ifp);
1998                 }
1999         } else {
2000                 if (ifp->if_amcount > 1) {
2001                         ifp->if_amcount--;
2002                 } else {
2003                         ifp->if_amcount = 0;
2004                         ifp->if_flags &= ~IFF_ALLMULTI;
2005                         ifr.ifr_flags = ifp->if_flags;
2006                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2007                         ifnet_serialize_all(ifp);
2008                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2009                                               NULL);
2010                         ifnet_deserialize_all(ifp);
2011                 }
2012         }
2013
2014         crit_exit();
2015
2016         if (error == 0)
2017                 rt_ifmsg(ifp);
2018         return error;
2019 }
2020
2021 /*
2022  * Add a multicast listenership to the interface in question.
2023  * The link layer provides a routine which converts
2024  */
2025 int
2026 if_addmulti(
2027         struct ifnet *ifp,      /* interface to manipulate */
2028         struct sockaddr *sa,    /* address to add */
2029         struct ifmultiaddr **retifma)
2030 {
2031         struct sockaddr *llsa, *dupsa;
2032         int error;
2033         struct ifmultiaddr *ifma;
2034
2035         /*
2036          * If the matching multicast address already exists
2037          * then don't add a new one, just add a reference
2038          */
2039         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2040                 if (sa_equal(sa, ifma->ifma_addr)) {
2041                         ifma->ifma_refcount++;
2042                         if (retifma)
2043                                 *retifma = ifma;
2044                         return 0;
2045                 }
2046         }
2047
2048         /*
2049          * Give the link layer a chance to accept/reject it, and also
2050          * find out which AF_LINK address this maps to, if it isn't one
2051          * already.
2052          */
2053         if (ifp->if_resolvemulti) {
2054                 ifnet_serialize_all(ifp);
2055                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2056                 ifnet_deserialize_all(ifp);
2057                 if (error)
2058                         return error;
2059         } else {
2060                 llsa = 0;
2061         }
2062
2063         MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, M_WAITOK);
2064         MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, M_WAITOK);
2065         bcopy(sa, dupsa, sa->sa_len);
2066
2067         ifma->ifma_addr = dupsa;
2068         ifma->ifma_lladdr = llsa;
2069         ifma->ifma_ifp = ifp;
2070         ifma->ifma_refcount = 1;
2071         ifma->ifma_protospec = 0;
2072         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2073
2074         /*
2075          * Some network interfaces can scan the address list at
2076          * interrupt time; lock them out.
2077          */
2078         crit_enter();
2079         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2080         crit_exit();
2081         if (retifma)
2082                 *retifma = ifma;
2083
2084         if (llsa != 0) {
2085                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2086                         if (sa_equal(ifma->ifma_addr, llsa))
2087                                 break;
2088                 }
2089                 if (ifma) {
2090                         ifma->ifma_refcount++;
2091                 } else {
2092                         MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma,
2093                                M_IFMADDR, M_WAITOK);
2094                         MALLOC(dupsa, struct sockaddr *, llsa->sa_len,
2095                                M_IFMADDR, M_WAITOK);
2096                         bcopy(llsa, dupsa, llsa->sa_len);
2097                         ifma->ifma_addr = dupsa;
2098                         ifma->ifma_ifp = ifp;
2099                         ifma->ifma_refcount = 1;
2100                         crit_enter();
2101                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2102                         crit_exit();
2103                 }
2104         }
2105         /*
2106          * We are certain we have added something, so call down to the
2107          * interface to let them know about it.
2108          */
2109         crit_enter();
2110         ifnet_serialize_all(ifp);
2111         if (ifp->if_ioctl)
2112                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2113         ifnet_deserialize_all(ifp);
2114         crit_exit();
2115
2116         return 0;
2117 }
2118
2119 /*
2120  * Remove a reference to a multicast address on this interface.  Yell
2121  * if the request does not match an existing membership.
2122  */
2123 int
2124 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2125 {
2126         struct ifmultiaddr *ifma;
2127
2128         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2129                 if (sa_equal(sa, ifma->ifma_addr))
2130                         break;
2131         if (ifma == 0)
2132                 return ENOENT;
2133
2134         if (ifma->ifma_refcount > 1) {
2135                 ifma->ifma_refcount--;
2136                 return 0;
2137         }
2138
2139         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2140         sa = ifma->ifma_lladdr;
2141         crit_enter();
2142         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2143         /*
2144          * Make sure the interface driver is notified
2145          * in the case of a link layer mcast group being left.
2146          */
2147         if (ifma->ifma_addr->sa_family == AF_LINK && sa == 0) {
2148                 ifnet_serialize_all(ifp);
2149                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2150                 ifnet_deserialize_all(ifp);
2151         }
2152         crit_exit();
2153         kfree(ifma->ifma_addr, M_IFMADDR);
2154         kfree(ifma, M_IFMADDR);
2155         if (sa == 0)
2156                 return 0;
2157
2158         /*
2159          * Now look for the link-layer address which corresponds to
2160          * this network address.  It had been squirreled away in
2161          * ifma->ifma_lladdr for this purpose (so we don't have
2162          * to call ifp->if_resolvemulti() again), and we saved that
2163          * value in sa above.  If some nasty deleted the
2164          * link-layer address out from underneath us, we can deal because
2165          * the address we stored was is not the same as the one which was
2166          * in the record for the link-layer address.  (So we don't complain
2167          * in that case.)
2168          */
2169         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2170                 if (sa_equal(sa, ifma->ifma_addr))
2171                         break;
2172         if (ifma == 0)
2173                 return 0;
2174
2175         if (ifma->ifma_refcount > 1) {
2176                 ifma->ifma_refcount--;
2177                 return 0;
2178         }
2179
2180         crit_enter();
2181         ifnet_serialize_all(ifp);
2182         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2183         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2184         ifnet_deserialize_all(ifp);
2185         crit_exit();
2186         kfree(ifma->ifma_addr, M_IFMADDR);
2187         kfree(sa, M_IFMADDR);
2188         kfree(ifma, M_IFMADDR);
2189
2190         return 0;
2191 }
2192
2193 /*
2194  * Delete all multicast group membership for an interface.
2195  * Should be used to quickly flush all multicast filters.
2196  */
2197 void
2198 if_delallmulti(struct ifnet *ifp)
2199 {
2200         struct ifmultiaddr *ifma;
2201         struct ifmultiaddr *next;
2202
2203         TAILQ_FOREACH_MUTABLE(ifma, &ifp->if_multiaddrs, ifma_link, next)
2204                 if_delmulti(ifp, ifma->ifma_addr);
2205 }
2206
2207
2208 /*
2209  * Set the link layer address on an interface.
2210  *
2211  * At this time we only support certain types of interfaces,
2212  * and we don't allow the length of the address to change.
2213  */
2214 int
2215 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2216 {
2217         struct sockaddr_dl *sdl;
2218         struct ifreq ifr;
2219
2220         sdl = IF_LLSOCKADDR(ifp);
2221         if (sdl == NULL)
2222                 return (EINVAL);
2223         if (len != sdl->sdl_alen)       /* don't allow length to change */
2224                 return (EINVAL);
2225         switch (ifp->if_type) {
2226         case IFT_ETHER:                 /* these types use struct arpcom */
2227         case IFT_XETHER:
2228         case IFT_L2VLAN:
2229                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2230                 bcopy(lladdr, LLADDR(sdl), len);
2231                 break;
2232         default:
2233                 return (ENODEV);
2234         }
2235         /*
2236          * If the interface is already up, we need
2237          * to re-init it in order to reprogram its
2238          * address filter.
2239          */
2240         ifnet_serialize_all(ifp);
2241         if ((ifp->if_flags & IFF_UP) != 0) {
2242                 struct ifaddr_container *ifac;
2243
2244                 ifp->if_flags &= ~IFF_UP;
2245                 ifr.ifr_flags = ifp->if_flags;
2246                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2247                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2248                               NULL);
2249                 ifp->if_flags |= IFF_UP;
2250                 ifr.ifr_flags = ifp->if_flags;
2251                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2252                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2253                                  NULL);
2254 #ifdef INET
2255                 /*
2256                  * Also send gratuitous ARPs to notify other nodes about
2257                  * the address change.
2258                  */
2259                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2260                         struct ifaddr *ifa = ifac->ifa;
2261
2262                         if (ifa->ifa_addr != NULL &&
2263                             ifa->ifa_addr->sa_family == AF_INET)
2264                                 arp_ifinit(ifp, ifa);
2265                 }
2266 #endif
2267         }
2268         ifnet_deserialize_all(ifp);
2269         return (0);
2270 }
2271
2272 struct ifmultiaddr *
2273 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2274 {
2275         struct ifmultiaddr *ifma;
2276
2277         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2278                 if (sa_equal(ifma->ifma_addr, sa))
2279                         break;
2280
2281         return ifma;
2282 }
2283
2284 /*
2285  * This function locates the first real ethernet MAC from a network
2286  * card and loads it into node, returning 0 on success or ENOENT if
2287  * no suitable interfaces were found.  It is used by the uuid code to
2288  * generate a unique 6-byte number.
2289  */
2290 int
2291 if_getanyethermac(uint16_t *node, int minlen)
2292 {
2293         struct ifnet *ifp;
2294         struct sockaddr_dl *sdl;
2295
2296         TAILQ_FOREACH(ifp, &ifnet, if_link) {
2297                 if (ifp->if_type != IFT_ETHER)
2298                         continue;
2299                 sdl = IF_LLSOCKADDR(ifp);
2300                 if (sdl->sdl_alen < minlen)
2301                         continue;
2302                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2303                       minlen);
2304                 return(0);
2305         }
2306         return (ENOENT);
2307 }
2308
2309 /*
2310  * The name argument must be a pointer to storage which will last as
2311  * long as the interface does.  For physical devices, the result of
2312  * device_get_name(dev) is a good choice and for pseudo-devices a
2313  * static string works well.
2314  */
2315 void
2316 if_initname(struct ifnet *ifp, const char *name, int unit)
2317 {
2318         ifp->if_dname = name;
2319         ifp->if_dunit = unit;
2320         if (unit != IF_DUNIT_NONE)
2321                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2322         else
2323                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2324 }
2325
2326 int
2327 if_printf(struct ifnet *ifp, const char *fmt, ...)
2328 {
2329         __va_list ap;
2330         int retval;
2331
2332         retval = kprintf("%s: ", ifp->if_xname);
2333         __va_start(ap, fmt);
2334         retval += kvprintf(fmt, ap);
2335         __va_end(ap);
2336         return (retval);
2337 }
2338
2339 struct ifnet *
2340 if_alloc(uint8_t type)
2341 {
2342         struct ifnet *ifp;
2343         size_t size;
2344
2345         /*
2346          * XXX temporary hack until arpcom is setup in if_l2com
2347          */
2348         if (type == IFT_ETHER)
2349                 size = sizeof(struct arpcom);
2350         else
2351                 size = sizeof(struct ifnet);
2352
2353         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2354
2355         ifp->if_type = type;
2356
2357         if (if_com_alloc[type] != NULL) {
2358                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2359                 if (ifp->if_l2com == NULL) {
2360                         kfree(ifp, M_IFNET);
2361                         return (NULL);
2362                 }
2363         }
2364         return (ifp);
2365 }
2366
2367 void
2368 if_free(struct ifnet *ifp)
2369 {
2370         kfree(ifp, M_IFNET);
2371 }
2372
2373 void
2374 ifq_set_classic(struct ifaltq *ifq)
2375 {
2376         ifq->altq_enqueue = ifq_classic_enqueue;
2377         ifq->altq_dequeue = ifq_classic_dequeue;
2378         ifq->altq_request = ifq_classic_request;
2379 }
2380
2381 int
2382 ifq_classic_enqueue(struct ifaltq *ifq, struct mbuf *m,
2383                     struct altq_pktattr *pa __unused)
2384 {
2385         logifq(enqueue, ifq);
2386         if (IF_QFULL(ifq)) {
2387                 m_freem(m);
2388                 return(ENOBUFS);
2389         } else {
2390                 IF_ENQUEUE(ifq, m);
2391                 return(0);
2392         }
2393 }
2394
2395 struct mbuf *
2396 ifq_classic_dequeue(struct ifaltq *ifq, struct mbuf *mpolled, int op)
2397 {
2398         struct mbuf *m;
2399
2400         switch (op) {
2401         case ALTDQ_POLL:
2402                 IF_POLL(ifq, m);
2403                 break;
2404         case ALTDQ_REMOVE:
2405                 logifq(dequeue, ifq);
2406                 IF_DEQUEUE(ifq, m);
2407                 break;
2408         default:
2409                 panic("unsupported ALTQ dequeue op: %d", op);
2410         }
2411         KKASSERT(mpolled == NULL || mpolled == m);
2412         return(m);
2413 }
2414
2415 int
2416 ifq_classic_request(struct ifaltq *ifq, int req, void *arg)
2417 {
2418         switch (req) {
2419         case ALTRQ_PURGE:
2420                 IF_DRAIN(ifq);
2421                 break;
2422         default:
2423                 panic("unsupported ALTQ request: %d", req);
2424         }
2425         return(0);
2426 }
2427
2428 int
2429 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2430 {
2431         struct ifaltq *ifq = &ifp->if_snd;
2432         int running = 0, error, start = 0;
2433
2434         ASSERT_IFNET_NOT_SERIALIZED_TX(ifp);
2435
2436         ALTQ_LOCK(ifq);
2437         error = ifq_enqueue_locked(ifq, m, pa);
2438         if (error) {
2439                 ALTQ_UNLOCK(ifq);
2440                 return error;
2441         }
2442         if (!ifq->altq_started) {
2443                 /*
2444                  * Hold the interlock of ifnet.if_start
2445                  */
2446                 ifq->altq_started = 1;
2447                 start = 1;
2448         }
2449         ALTQ_UNLOCK(ifq);
2450
2451         ifp->if_obytes += m->m_pkthdr.len;
2452         if (m->m_flags & M_MCAST)
2453                 ifp->if_omcasts++;
2454
2455         if (!start) {
2456                 logifstart(avoid, ifp);
2457                 return 0;
2458         }
2459
2460         if (ifq_dispatch_schedonly) {
2461                 /*
2462                  * Always schedule ifnet.if_start on ifnet's CPU,
2463                  * short circuit the rest of this function.
2464                  */
2465                 logifstart(sched, ifp);
2466                 if_start_schedule(ifp);
2467                 return 0;
2468         }
2469
2470         /*
2471          * Try to do direct ifnet.if_start first, if there is
2472          * contention on ifnet's serializer, ifnet.if_start will
2473          * be scheduled on ifnet's CPU.
2474          */
2475         if (!ifnet_tryserialize_tx(ifp)) {
2476                 /*
2477                  * ifnet serializer contention happened,
2478                  * ifnet.if_start is scheduled on ifnet's
2479                  * CPU, and we keep going.
2480                  */
2481                 logifstart(contend_sched, ifp);
2482                 if_start_schedule(ifp);
2483                 return 0;
2484         }
2485
2486         if ((ifp->if_flags & IFF_OACTIVE) == 0) {
2487                 logifstart(run, ifp);
2488                 ifp->if_start(ifp);
2489                 if ((ifp->if_flags &
2490                      (IFF_OACTIVE | IFF_RUNNING)) == IFF_RUNNING)
2491                         running = 1;
2492         }
2493
2494         ifnet_deserialize_tx(ifp);
2495
2496         if (ifq_dispatch_schednochk || if_start_need_schedule(ifq, running)) {
2497                 /*
2498                  * More data need to be transmitted, ifnet.if_start is
2499                  * scheduled on ifnet's CPU, and we keep going.
2500                  * NOTE: ifnet.if_start interlock is not released.
2501                  */
2502                 logifstart(sched, ifp);
2503                 if_start_schedule(ifp);
2504         }
2505         return 0;
2506 }
2507
2508 void *
2509 ifa_create(int size, int flags)
2510 {
2511         struct ifaddr *ifa;
2512         int i;
2513
2514         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small\n"));
2515
2516         ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2517         if (ifa == NULL)
2518                 return NULL;
2519
2520         ifa->ifa_containers = kmalloc(ncpus * sizeof(struct ifaddr_container),
2521                                       M_IFADDR, M_WAITOK | M_ZERO);
2522         ifa->ifa_ncnt = ncpus;
2523         for (i = 0; i < ncpus; ++i) {
2524                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2525
2526                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2527                 ifac->ifa = ifa;
2528                 ifac->ifa_refcnt = 1;
2529         }
2530 #ifdef IFADDR_DEBUG
2531         kprintf("alloc ifa %p %d\n", ifa, size);
2532 #endif
2533         return ifa;
2534 }
2535
2536 void
2537 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2538 {
2539         struct ifaddr *ifa = ifac->ifa;
2540
2541         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2542         KKASSERT(ifac->ifa_refcnt == 0);
2543         KASSERT(ifac->ifa_listmask == 0,
2544                 ("ifa is still on %#x lists\n", ifac->ifa_listmask));
2545
2546         ifac->ifa_magic = IFA_CONTAINER_DEAD;
2547
2548 #ifdef IFADDR_DEBUG_VERBOSE
2549         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2550 #endif
2551
2552         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2553                 ("invalid # of ifac, %d\n", ifa->ifa_ncnt));
2554         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2555 #ifdef IFADDR_DEBUG
2556                 kprintf("free ifa %p\n", ifa);
2557 #endif
2558                 kfree(ifa->ifa_containers, M_IFADDR);
2559                 kfree(ifa, M_IFADDR);
2560         }
2561 }
2562
2563 static void
2564 ifa_iflink_dispatch(netmsg_t nmsg)
2565 {
2566         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2567         struct ifaddr *ifa = msg->ifa;
2568         struct ifnet *ifp = msg->ifp;
2569         int cpu = mycpuid;
2570         struct ifaddr_container *ifac;
2571
2572         crit_enter();
2573
2574         ifac = &ifa->ifa_containers[cpu];
2575         ASSERT_IFAC_VALID(ifac);
2576         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2577                 ("ifaddr is on if_addrheads\n"));
2578
2579         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2580         if (msg->tail)
2581                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2582         else
2583                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2584
2585         crit_exit();
2586
2587         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2588 }
2589
2590 void
2591 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2592 {
2593         struct netmsg_ifaddr msg;
2594
2595         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2596                     0, ifa_iflink_dispatch);
2597         msg.ifa = ifa;
2598         msg.ifp = ifp;
2599         msg.tail = tail;
2600
2601         ifa_domsg(&msg.base.lmsg, 0);
2602 }
2603
2604 static void
2605 ifa_ifunlink_dispatch(netmsg_t nmsg)
2606 {
2607         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2608         struct ifaddr *ifa = msg->ifa;
2609         struct ifnet *ifp = msg->ifp;
2610         int cpu = mycpuid;
2611         struct ifaddr_container *ifac;
2612
2613         crit_enter();
2614
2615         ifac = &ifa->ifa_containers[cpu];
2616         ASSERT_IFAC_VALID(ifac);
2617         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2618                 ("ifaddr is not on if_addrhead\n"));
2619
2620         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2621         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2622
2623         crit_exit();
2624
2625         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2626 }
2627
2628 void
2629 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2630 {
2631         struct netmsg_ifaddr msg;
2632
2633         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2634                     0, ifa_ifunlink_dispatch);
2635         msg.ifa = ifa;
2636         msg.ifp = ifp;
2637
2638         ifa_domsg(&msg.base.lmsg, 0);
2639 }
2640
2641 static void
2642 ifa_destroy_dispatch(netmsg_t nmsg)
2643 {
2644         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2645
2646         IFAFREE(msg->ifa);
2647         ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2648 }
2649
2650 void
2651 ifa_destroy(struct ifaddr *ifa)
2652 {
2653         struct netmsg_ifaddr msg;
2654
2655         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2656                     0, ifa_destroy_dispatch);
2657         msg.ifa = ifa;
2658
2659         ifa_domsg(&msg.base.lmsg, 0);
2660 }
2661
2662 struct lwkt_port *
2663 ifnet_portfn(int cpu)
2664 {
2665         return &ifnet_threads[cpu].td_msgport;
2666 }
2667
2668 void
2669 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2670 {
2671         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2672
2673         if (next_cpu < ncpus)
2674                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2675         else
2676                 lwkt_replymsg(lmsg, 0);
2677 }
2678
2679 int
2680 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
2681 {
2682         KKASSERT(cpu < ncpus);
2683         return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
2684 }
2685
2686 void
2687 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
2688 {
2689         KKASSERT(cpu < ncpus);
2690         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
2691 }
2692
2693 /*
2694  * Generic netmsg service loop.  Some protocols may roll their own but all
2695  * must do the basic command dispatch function call done here.
2696  */
2697 static void
2698 ifnet_service_loop(void *arg __unused)
2699 {
2700         netmsg_t msg;
2701
2702         while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
2703                 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
2704                 msg->base.nm_dispatch(msg);
2705         }
2706 }
2707
2708 static void
2709 ifnetinit(void *dummy __unused)
2710 {
2711         int i;
2712
2713         for (i = 0; i < ncpus; ++i) {
2714                 struct thread *thr = &ifnet_threads[i];
2715
2716                 lwkt_create(ifnet_service_loop, NULL, NULL,
2717                             thr, TDF_STOPREQ, i, "ifnet %d", i);
2718                 netmsg_service_port_init(&thr->td_msgport);
2719                 lwkt_schedule(thr);
2720         }
2721 }
2722
2723 struct ifnet *
2724 ifnet_byindex(unsigned short idx)
2725 {
2726         if (idx > if_index)
2727                 return NULL;
2728         return ifindex2ifnet[idx];
2729 }
2730
2731 struct ifaddr *
2732 ifaddr_byindex(unsigned short idx)
2733 {
2734         struct ifnet *ifp;
2735
2736         ifp = ifnet_byindex(idx);
2737         if (!ifp)
2738                 return NULL;
2739         return TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
2740 }
2741
2742 void
2743 if_register_com_alloc(u_char type,
2744     if_com_alloc_t *a, if_com_free_t *f)
2745 {
2746
2747         KASSERT(if_com_alloc[type] == NULL,
2748             ("if_register_com_alloc: %d already registered", type));
2749         KASSERT(if_com_free[type] == NULL,
2750             ("if_register_com_alloc: %d free already registered", type));
2751
2752         if_com_alloc[type] = a;
2753         if_com_free[type] = f;
2754 }
2755
2756 void
2757 if_deregister_com_alloc(u_char type)
2758 {
2759
2760         KASSERT(if_com_alloc[type] != NULL,
2761             ("if_deregister_com_alloc: %d not registered", type));
2762         KASSERT(if_com_free[type] != NULL,
2763             ("if_deregister_com_alloc: %d free not registered", type));
2764         if_com_alloc[type] = NULL;
2765         if_com_free[type] = NULL;
2766 }