sys/net/if.c

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