sys/net/if.c

   1 /*
   2  * Copyright (c) 1980, 1986, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)if.c        8.3 (Berkeley) 1/4/94
  30  * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
  31  */
  32
  33 #include "opt_compat.h"
  34 #include "opt_inet6.h"
  35 #include "opt_inet.h"
  36 #include "opt_ifpoll.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/malloc.h>
  40 #include <sys/mbuf.h>
  41 #include <sys/systm.h>
  42 #include <sys/proc.h>
  43 #include <sys/priv.h>
  44 #include <sys/protosw.h>
  45 #include <sys/socket.h>
  46 #include <sys/socketvar.h>
  47 #include <sys/socketops.h>
  48 #include <sys/protosw.h>
  49 #include <sys/kernel.h>
  50 #include <sys/ktr.h>
  51 #include <sys/mutex.h>
  52 #include <sys/sockio.h>
  53 #include <sys/syslog.h>
  54 #include <sys/sysctl.h>
  55 #include <sys/domain.h>
  56 #include <sys/thread.h>
  57 #include <sys/serialize.h>
  58 #include <sys/bus.h>
  59
  60 #include <sys/thread2.h>
  61 #include <sys/msgport2.h>
  62 #include <sys/mutex2.h>
  63
  64 #include <net/if.h>
  65 #include <net/if_arp.h>
  66 #include <net/if_dl.h>
  67 #include <net/if_types.h>
  68 #include <net/if_var.h>
  69 #include <net/ifq_var.h>
  70 #include <net/radix.h>
  71 #include <net/route.h>
  72 #include <net/if_clone.h>
  73 #include <net/netisr2.h>
  74 #include <net/netmsg2.h>
  75
  76 #include <machine/atomic.h>
  77 #include <machine/stdarg.h>
  78 #include <machine/smp.h>
  79
  80 #if defined(INET) || defined(INET6)
  81 /*XXX*/
  82 #include <netinet/in.h>
  83 #include <netinet/in_var.h>
  84 #include <netinet/if_ether.h>
  85 #ifdef INET6
  86 #include <netinet6/in6_var.h>
  87 #include <netinet6/in6_ifattach.h>
  88 #endif
  89 #endif
  90
  91 #if defined(COMPAT_43)
  92 #include <emulation/43bsd/43bsd_socket.h>
  93 #endif /* COMPAT_43 */
  94
  95 struct netmsg_ifaddr {
  96         struct netmsg_base base;
  97         struct ifaddr   *ifa;
  98         struct ifnet    *ifp;
  99         int             tail;
 100 };
 101
 102 struct ifsubq_stage_head {
 103         TAILQ_HEAD(, ifsubq_stage)      stg_head;
 104 } __cachealign;
 105
 106 /*
 107  * System initialization
 108  */
 109 static void     if_attachdomain(void *);
 110 static void     if_attachdomain1(struct ifnet *);
 111 static int      ifconf(u_long, caddr_t, struct ucred *);
 112 static void     ifinit(void *);
 113 static void     ifnetinit(void *);
 114 static void     if_slowtimo(void *);
 115 static void     link_rtrequest(int, struct rtentry *);
 116 static int      if_rtdel(struct radix_node *, void *);
 117
 118 /* Helper functions */
 119 static void     ifsq_watchdog_reset(struct ifsubq_watchdog *);
 120
 121 #ifdef INET6
 122 /*
 123  * XXX: declare here to avoid to include many inet6 related files..
 124  * should be more generalized?
 125  */
 126 extern void     nd6_setmtu(struct ifnet *);
 127 #endif
 128
 129 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
 130 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
 131
 132 static int ifsq_stage_cntmax = 4;
 133 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
 134 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
 135     &ifsq_stage_cntmax, 0, "ifq staging packet count max");
 136
 137 static int if_stats_compat = 0;
 138 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
 139     &if_stats_compat, 0, "Compat the old ifnet stats");
 140
 141 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
 142 /* Must be after netisr_init */
 143 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
 144
 145 static  if_com_alloc_t *if_com_alloc[256];
 146 static  if_com_free_t *if_com_free[256];
 147
 148 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
 149 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
 150 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
 151
 152 int                     ifqmaxlen = IFQ_MAXLEN;
 153 struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
 154
 155 struct callout          if_slowtimo_timer;
 156
 157 int                     if_index = 0;
 158 struct ifnet            **ifindex2ifnet = NULL;
 159 static struct thread    ifnet_threads[MAXCPU];
 160
 161 static struct ifsubq_stage_head ifsubq_stage_heads[MAXCPU];
 162
 163 #ifdef notyet
 164 #define IFQ_KTR_STRING          "ifq=%p"
 165 #define IFQ_KTR_ARGS    struct ifaltq *ifq
 166 #ifndef KTR_IFQ
 167 #define KTR_IFQ                 KTR_ALL
 168 #endif
 169 KTR_INFO_MASTER(ifq);
 170 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
 171 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
 172 #define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)
 173
 174 #define IF_START_KTR_STRING     "ifp=%p"
 175 #define IF_START_KTR_ARGS       struct ifnet *ifp
 176 #ifndef KTR_IF_START
 177 #define KTR_IF_START            KTR_ALL
 178 #endif
 179 KTR_INFO_MASTER(if_start);
 180 KTR_INFO(KTR_IF_START, if_start, run, 0,
 181          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 182 KTR_INFO(KTR_IF_START, if_start, sched, 1,
 183          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 184 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
 185          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 186 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
 187          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 188 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
 189          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 190 #define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
 191 #endif
 192
 193 TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
 194
 195 /*
 196  * Network interface utility routines.
 197  *
 198  * Routines with ifa_ifwith* names take sockaddr *'s as
 199  * parameters.
 200  */
 201 /* ARGSUSED*/
 202 void
 203 ifinit(void *dummy)
 204 {
 205         struct ifnet *ifp;
 206
 207         callout_init(&if_slowtimo_timer);
 208
 209         crit_enter();
 210         TAILQ_FOREACH(ifp, &ifnet, if_link) {
 211                 if (ifp->if_snd.altq_maxlen == 0) {
 212                         if_printf(ifp, "XXX: driver didn't set altq_maxlen\n");
 213                         ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
 214                 }
 215         }
 216         crit_exit();
 217
 218         if_slowtimo(0);
 219 }
 220
 221 static void
 222 ifsq_ifstart_ipifunc(void *arg)
 223 {
 224         struct ifaltq_subque *ifsq = arg;
 225         struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
 226
 227         crit_enter();
 228         if (lmsg->ms_flags & MSGF_DONE)
 229                 lwkt_sendmsg(netisr_cpuport(mycpuid), lmsg);
 230         crit_exit();
 231 }
 232
 233 static __inline void
 234 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
 235 {
 236         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
 237         TAILQ_REMOVE(&head->stg_head, stage, stg_link);
 238         stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
 239         stage->stg_cnt = 0;
 240         stage->stg_len = 0;
 241 }
 242
 243 static __inline void
 244 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
 245 {
 246         KKASSERT((stage->stg_flags &
 247             (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
 248         stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
 249         TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
 250 }
 251
 252 /*
 253  * Schedule ifnet.if_start on the subqueue owner CPU
 254  */
 255 static void
 256 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
 257 {
 258         int cpu;
 259
 260         if (!force && curthread->td_type == TD_TYPE_NETISR &&
 261             ifsq_stage_cntmax > 0) {
 262                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 263
 264                 stage->stg_cnt = 0;
 265                 stage->stg_len = 0;
 266                 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
 267                         ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
 268                 stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
 269                 return;
 270         }
 271
 272         cpu = ifsq_get_cpuid(ifsq);
 273         if (cpu != mycpuid)
 274                 lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
 275         else
 276                 ifsq_ifstart_ipifunc(ifsq);
 277 }
 278
 279 /*
 280  * NOTE:
 281  * This function will release ifnet.if_start subqueue interlock,
 282  * if ifnet.if_start for the subqueue does not need to be scheduled
 283  */
 284 static __inline int
 285 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
 286 {
 287         if (!running || ifsq_is_empty(ifsq)
 288 #ifdef ALTQ
 289             || ifsq->ifsq_altq->altq_tbr != NULL
 290 #endif
 291         ) {
 292                 ALTQ_SQ_LOCK(ifsq);
 293                 /*
 294                  * ifnet.if_start subqueue interlock is released, if:
 295                  * 1) Hardware can not take any packets, due to
 296                  *    o  interface is marked down
 297                  *    o  hardware queue is full (ifsq_is_oactive)
 298                  *    Under the second situation, hardware interrupt
 299                  *    or polling(4) will call/schedule ifnet.if_start
 300                  *    on the subqueue when hardware queue is ready
 301                  * 2) There is no packet in the subqueue.
 302                  *    Further ifq_dispatch or ifq_handoff will call/
 303                  *    schedule ifnet.if_start on the subqueue.
 304                  * 3) TBR is used and it does not allow further
 305                  *    dequeueing.
 306                  *    TBR callout will call ifnet.if_start on the
 307                  *    subqueue.
 308                  */
 309                 if (!running || !ifsq_data_ready(ifsq)) {
 310                         ifsq_clr_started(ifsq);
 311                         ALTQ_SQ_UNLOCK(ifsq);
 312                         return 0;
 313                 }
 314                 ALTQ_SQ_UNLOCK(ifsq);
 315         }
 316         return 1;
 317 }
 318
 319 static void
 320 ifsq_ifstart_dispatch(netmsg_t msg)
 321 {
 322         struct lwkt_msg *lmsg = &msg->base.lmsg;
 323         struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
 324         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 325         int running = 0, need_sched;
 326
 327         crit_enter();
 328         lwkt_replymsg(lmsg, 0); /* reply ASAP */
 329         crit_exit();
 330
 331         if (mycpuid != ifsq_get_cpuid(ifsq)) {
 332                 /*
 333                  * We need to chase the subqueue owner CPU change.
 334                  */
 335                 ifsq_ifstart_schedule(ifsq, 1);
 336                 return;
 337         }
 338
 339         ifsq_serialize_hw(ifsq);
 340         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
 341                 ifp->if_start(ifp, ifsq);
 342                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 343                         running = 1;
 344         }
 345         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
 346         ifsq_deserialize_hw(ifsq);
 347
 348         if (need_sched) {
 349                 /*
 350                  * More data need to be transmitted, ifnet.if_start is
 351                  * scheduled on the subqueue owner CPU, and we keep going.
 352                  * NOTE: ifnet.if_start subqueue interlock is not released.
 353                  */
 354                 ifsq_ifstart_schedule(ifsq, 0);
 355         }
 356 }
 357
 358 /* Device driver ifnet.if_start helper function */
 359 void
 360 ifsq_devstart(struct ifaltq_subque *ifsq)
 361 {
 362         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 363         int running = 0;
 364
 365         ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
 366
 367         ALTQ_SQ_LOCK(ifsq);
 368         if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
 369                 ALTQ_SQ_UNLOCK(ifsq);
 370                 return;
 371         }
 372         ifsq_set_started(ifsq);
 373         ALTQ_SQ_UNLOCK(ifsq);
 374
 375         ifp->if_start(ifp, ifsq);
 376
 377         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 378                 running = 1;
 379
 380         if (ifsq_ifstart_need_schedule(ifsq, running)) {
 381                 /*
 382                  * More data need to be transmitted, ifnet.if_start is
 383                  * scheduled on ifnet's CPU, and we keep going.
 384                  * NOTE: ifnet.if_start interlock is not released.
 385                  */
 386                 ifsq_ifstart_schedule(ifsq, 0);
 387         }
 388 }
 389
 390 void
 391 if_devstart(struct ifnet *ifp)
 392 {
 393         ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
 394 }
 395
 396 /* Device driver ifnet.if_start schedule helper function */
 397 void
 398 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
 399 {
 400         ifsq_ifstart_schedule(ifsq, 1);
 401 }
 402
 403 void
 404 if_devstart_sched(struct ifnet *ifp)
 405 {
 406         ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
 407 }
 408
 409 static void
 410 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 411 {
 412         lwkt_serialize_enter(ifp->if_serializer);
 413 }
 414
 415 static void
 416 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 417 {
 418         lwkt_serialize_exit(ifp->if_serializer);
 419 }
 420
 421 static int
 422 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 423 {
 424         return lwkt_serialize_try(ifp->if_serializer);
 425 }
 426
 427 #ifdef INVARIANTS
 428 static void
 429 if_default_serialize_assert(struct ifnet *ifp,
 430                             enum ifnet_serialize slz __unused,
 431                             boolean_t serialized)
 432 {
 433         if (serialized)
 434                 ASSERT_SERIALIZED(ifp->if_serializer);
 435         else
 436                 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
 437 }
 438 #endif
 439
 440 /*
 441  * Attach an interface to the list of "active" interfaces.
 442  *
 443  * The serializer is optional.
 444  */
 445 void
 446 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
 447 {
 448         unsigned socksize, ifasize;
 449         int namelen, masklen;
 450         struct sockaddr_dl *sdl;
 451         struct ifaddr *ifa;
 452         struct ifaltq *ifq;
 453         int i, q;
 454
 455         static int if_indexlim = 8;
 456
 457         if (ifp->if_serialize != NULL) {
 458                 KASSERT(ifp->if_deserialize != NULL &&
 459                         ifp->if_tryserialize != NULL &&
 460                         ifp->if_serialize_assert != NULL,
 461                         ("serialize functions are partially setup"));
 462
 463                 /*
 464                  * If the device supplies serialize functions,
 465                  * then clear if_serializer to catch any invalid
 466                  * usage of this field.
 467                  */
 468                 KASSERT(serializer == NULL,
 469                         ("both serialize functions and default serializer "
 470                          "are supplied"));
 471                 ifp->if_serializer = NULL;
 472         } else {
 473                 KASSERT(ifp->if_deserialize == NULL &&
 474                         ifp->if_tryserialize == NULL &&
 475                         ifp->if_serialize_assert == NULL,
 476                         ("serialize functions are partially setup"));
 477                 ifp->if_serialize = if_default_serialize;
 478                 ifp->if_deserialize = if_default_deserialize;
 479                 ifp->if_tryserialize = if_default_tryserialize;
 480 #ifdef INVARIANTS
 481                 ifp->if_serialize_assert = if_default_serialize_assert;
 482 #endif
 483
 484                 /*
 485                  * The serializer can be passed in from the device,
 486                  * allowing the same serializer to be used for both
 487                  * the interrupt interlock and the device queue.
 488                  * If not specified, the netif structure will use an
 489                  * embedded serializer.
 490                  */
 491                 if (serializer == NULL) {
 492                         serializer = &ifp->if_default_serializer;
 493                         lwkt_serialize_init(serializer);
 494                 }
 495                 ifp->if_serializer = serializer;
 496         }
 497
 498         mtx_init(&ifp->if_ioctl_mtx);
 499         mtx_lock(&ifp->if_ioctl_mtx);
 500
 501         TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
 502         ifp->if_index = ++if_index;
 503
 504         /*
 505          * XXX -
 506          * The old code would work if the interface passed a pre-existing
 507          * chain of ifaddrs to this code.  We don't trust our callers to
 508          * properly initialize the tailq, however, so we no longer allow
 509          * this unlikely case.
 510          */
 511         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
 512                                     M_IFADDR, M_WAITOK | M_ZERO);
 513         for (i = 0; i < ncpus; ++i)
 514                 TAILQ_INIT(&ifp->if_addrheads[i]);
 515
 516         TAILQ_INIT(&ifp->if_prefixhead);
 517         TAILQ_INIT(&ifp->if_multiaddrs);
 518         TAILQ_INIT(&ifp->if_groups);
 519         getmicrotime(&ifp->if_lastchange);
 520         if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
 521                 unsigned int n;
 522                 struct ifnet **q;
 523
 524                 if_indexlim <<= 1;
 525
 526                 /* grow ifindex2ifnet */
 527                 n = if_indexlim * sizeof(*q);
 528                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
 529                 if (ifindex2ifnet) {
 530                         bcopy(ifindex2ifnet, q, n/2);
 531                         kfree(ifindex2ifnet, M_IFADDR);
 532                 }
 533                 ifindex2ifnet = q;
 534         }
 535
 536         ifindex2ifnet[if_index] = ifp;
 537
 538         /*
 539          * create a Link Level name for this device
 540          */
 541         namelen = strlen(ifp->if_xname);
 542         masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
 543         socksize = masklen + ifp->if_addrlen;
 544 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
 545         if (socksize < sizeof(*sdl))
 546                 socksize = sizeof(*sdl);
 547         socksize = ROUNDUP(socksize);
 548 #undef ROUNDUP
 549         ifasize = sizeof(struct ifaddr) + 2 * socksize;
 550         ifa = ifa_create(ifasize, M_WAITOK);
 551         sdl = (struct sockaddr_dl *)(ifa + 1);
 552         sdl->sdl_len = socksize;
 553         sdl->sdl_family = AF_LINK;
 554         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
 555         sdl->sdl_nlen = namelen;
 556         sdl->sdl_index = ifp->if_index;
 557         sdl->sdl_type = ifp->if_type;
 558         ifp->if_lladdr = ifa;
 559         ifa->ifa_ifp = ifp;
 560         ifa->ifa_rtrequest = link_rtrequest;
 561         ifa->ifa_addr = (struct sockaddr *)sdl;
 562         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
 563         ifa->ifa_netmask = (struct sockaddr *)sdl;
 564         sdl->sdl_len = masklen;
 565         while (namelen != 0)
 566                 sdl->sdl_data[--namelen] = 0xff;
 567         ifa_iflink(ifa, ifp, 0 /* Insert head */);
 568
 569         ifp->if_data_pcpu = kmalloc_cachealign(
 570             ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
 571
 572         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
 573         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
 574
 575         if (ifp->if_mapsubq == NULL)
 576                 ifp->if_mapsubq = ifq_mapsubq_default;
 577
 578         ifq = &ifp->if_snd;
 579         ifq->altq_type = 0;
 580         ifq->altq_disc = NULL;
 581         ifq->altq_flags &= ALTQF_CANTCHANGE;
 582         ifq->altq_tbr = NULL;
 583         ifq->altq_ifp = ifp;
 584
 585         if (ifq->altq_subq_cnt <= 0)
 586                 ifq->altq_subq_cnt = 1;
 587         ifq->altq_subq = kmalloc_cachealign(
 588             ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
 589             M_DEVBUF, M_WAITOK | M_ZERO);
 590
 591         if (ifq->altq_maxlen == 0) {
 592                 if_printf(ifp, "driver didn't set altq_maxlen\n");
 593                 ifq_set_maxlen(ifq, ifqmaxlen);
 594         }
 595
 596         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 597                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 598
 599                 ALTQ_SQ_LOCK_INIT(ifsq);
 600                 ifsq->ifsq_index = q;
 601
 602                 ifsq->ifsq_altq = ifq;
 603                 ifsq->ifsq_ifp = ifp;
 604
 605                 ifsq->ifsq_maxlen = ifq->altq_maxlen;
 606                 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
 607                 ifsq->ifsq_prepended = NULL;
 608                 ifsq->ifsq_started = 0;
 609                 ifsq->ifsq_hw_oactive = 0;
 610                 ifsq_set_cpuid(ifsq, 0);
 611                 if (ifp->if_serializer != NULL)
 612                         ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
 613
 614                 ifsq->ifsq_stage =
 615                     kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
 616                     M_DEVBUF, M_WAITOK | M_ZERO);
 617                 for (i = 0; i < ncpus; ++i)
 618                         ifsq->ifsq_stage[i].stg_subq = ifsq;
 619
 620                 ifsq->ifsq_ifstart_nmsg =
 621                     kmalloc(ncpus * sizeof(struct netmsg_base),
 622                     M_LWKTMSG, M_WAITOK);
 623                 for (i = 0; i < ncpus; ++i) {
 624                         netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
 625                             &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
 626                         ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
 627                 }
 628         }
 629         ifq_set_classic(ifq);
 630
 631         if (!SLIST_EMPTY(&domains))
 632                 if_attachdomain1(ifp);
 633
 634         /* Announce the interface. */
 635         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
 636
 637         mtx_unlock(&ifp->if_ioctl_mtx);
 638 }
 639
 640 static void
 641 if_attachdomain(void *dummy)
 642 {
 643         struct ifnet *ifp;
 644
 645         crit_enter();
 646         TAILQ_FOREACH(ifp, &ifnet, if_list)
 647                 if_attachdomain1(ifp);
 648         crit_exit();
 649 }
 650 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
 651         if_attachdomain, NULL);
 652
 653 static void
 654 if_attachdomain1(struct ifnet *ifp)
 655 {
 656         struct domain *dp;
 657
 658         crit_enter();
 659
 660         /* address family dependent data region */
 661         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
 662         SLIST_FOREACH(dp, &domains, dom_next)
 663                 if (dp->dom_ifattach)
 664                         ifp->if_afdata[dp->dom_family] =
 665                                 (*dp->dom_ifattach)(ifp);
 666         crit_exit();
 667 }
 668
 669 /*
 670  * Purge all addresses whose type is _not_ AF_LINK
 671  */
 672 void
 673 if_purgeaddrs_nolink(struct ifnet *ifp)
 674 {
 675         struct ifaddr_container *ifac, *next;
 676
 677         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
 678                               ifa_link, next) {
 679                 struct ifaddr *ifa = ifac->ifa;
 680
 681                 /* Leave link ifaddr as it is */
 682                 if (ifa->ifa_addr->sa_family == AF_LINK)
 683                         continue;
 684 #ifdef INET
 685                 /* XXX: Ugly!! ad hoc just for INET */
 686                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
 687                         struct ifaliasreq ifr;
 688 #ifdef IFADDR_DEBUG_VERBOSE
 689                         int i;
 690
 691                         kprintf("purge in4 addr %p: ", ifa);
 692                         for (i = 0; i < ncpus; ++i)
 693                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 694                         kprintf("\n");
 695 #endif
 696
 697                         bzero(&ifr, sizeof ifr);
 698                         ifr.ifra_addr = *ifa->ifa_addr;
 699                         if (ifa->ifa_dstaddr)
 700                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
 701                         if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
 702                                        NULL) == 0)
 703                                 continue;
 704                 }
 705 #endif /* INET */
 706 #ifdef INET6
 707                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
 708 #ifdef IFADDR_DEBUG_VERBOSE
 709                         int i;
 710
 711                         kprintf("purge in6 addr %p: ", ifa);
 712                         for (i = 0; i < ncpus; ++i)
 713                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 714                         kprintf("\n");
 715 #endif
 716
 717                         in6_purgeaddr(ifa);
 718                         /* ifp_addrhead is already updated */
 719                         continue;
 720                 }
 721 #endif /* INET6 */
 722                 ifa_ifunlink(ifa, ifp);
 723                 ifa_destroy(ifa);
 724         }
 725 }
 726
 727 static void
 728 ifq_stage_detach_handler(netmsg_t nmsg)
 729 {
 730         struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
 731         int q;
 732
 733         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 734                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 735                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 736
 737                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
 738                         ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
 739         }
 740         lwkt_replymsg(&nmsg->lmsg, 0);
 741 }
 742
 743 static void
 744 ifq_stage_detach(struct ifaltq *ifq)
 745 {
 746         struct netmsg_base base;
 747         int cpu;
 748
 749         netmsg_init(&base, NULL, &curthread->td_msgport, 0,
 750             ifq_stage_detach_handler);
 751         base.lmsg.u.ms_resultp = ifq;
 752
 753         for (cpu = 0; cpu < ncpus; ++cpu)
 754                 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
 755 }
 756
 757 struct netmsg_if_rtdel {
 758         struct netmsg_base      base;
 759         struct ifnet            *ifp;
 760 };
 761
 762 static void
 763 if_rtdel_dispatch(netmsg_t msg)
 764 {
 765         struct netmsg_if_rtdel *rmsg = (void *)msg;
 766         int i, nextcpu, cpu;
 767
 768         cpu = mycpuid;
 769         for (i = 1; i <= AF_MAX; i++) {
 770                 struct radix_node_head  *rnh;
 771
 772                 if ((rnh = rt_tables[cpu][i]) == NULL)
 773                         continue;
 774                 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
 775         }
 776
 777         nextcpu = cpu + 1;
 778         if (nextcpu < ncpus)
 779                 lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg);
 780         else
 781                 lwkt_replymsg(&rmsg->base.lmsg, 0);
 782 }
 783
 784 /*
 785  * Detach an interface, removing it from the
 786  * list of "active" interfaces.
 787  */
 788 void
 789 if_detach(struct ifnet *ifp)
 790 {
 791         struct netmsg_if_rtdel msg;
 792         struct domain *dp;
 793         int q;
 794
 795         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
 796
 797         /*
 798          * Remove routes and flush queues.
 799          */
 800         crit_enter();
 801 #ifdef IFPOLL_ENABLE
 802         if (ifp->if_flags & IFF_NPOLLING)
 803                 ifpoll_deregister(ifp);
 804 #endif
 805         if_down(ifp);
 806
 807 #ifdef ALTQ
 808         if (ifq_is_enabled(&ifp->if_snd))
 809                 altq_disable(&ifp->if_snd);
 810         if (ifq_is_attached(&ifp->if_snd))
 811                 altq_detach(&ifp->if_snd);
 812 #endif
 813
 814         /*
 815          * Clean up all addresses.
 816          */
 817         ifp->if_lladdr = NULL;
 818
 819         if_purgeaddrs_nolink(ifp);
 820         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
 821                 struct ifaddr *ifa;
 822
 823                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
 824                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
 825                         ("non-link ifaddr is left on if_addrheads"));
 826
 827                 ifa_ifunlink(ifa, ifp);
 828                 ifa_destroy(ifa);
 829                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
 830                         ("there are still ifaddrs left on if_addrheads"));
 831         }
 832
 833 #ifdef INET
 834         /*
 835          * Remove all IPv4 kernel structures related to ifp.
 836          */
 837         in_ifdetach(ifp);
 838 #endif
 839
 840 #ifdef INET6
 841         /*
 842          * Remove all IPv6 kernel structs related to ifp.  This should be done
 843          * before removing routing entries below, since IPv6 interface direct
 844          * routes are expected to be removed by the IPv6-specific kernel API.
 845          * Otherwise, the kernel will detect some inconsistency and bark it.
 846          */
 847         in6_ifdetach(ifp);
 848 #endif
 849
 850         /*
 851          * Delete all remaining routes using this interface
 852          */
 853         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
 854             if_rtdel_dispatch);
 855         msg.ifp = ifp;
 856         KASSERT(&curthread->td_msgport != rtable_portfn(0),
 857             ("if_detach in rtable thread"));
 858         lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0);
 859
 860         /* Announce that the interface is gone. */
 861         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
 862         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
 863
 864         SLIST_FOREACH(dp, &domains, dom_next)
 865                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
 866                         (*dp->dom_ifdetach)(ifp,
 867                                 ifp->if_afdata[dp->dom_family]);
 868
 869         /*
 870          * Remove interface from ifindex2ifp[] and maybe decrement if_index.
 871          */
 872         ifindex2ifnet[ifp->if_index] = NULL;
 873         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
 874                 if_index--;
 875
 876         TAILQ_REMOVE(&ifnet, ifp, if_link);
 877         kfree(ifp->if_addrheads, M_IFADDR);
 878
 879         lwkt_synchronize_ipiqs("if_detach");
 880         ifq_stage_detach(&ifp->if_snd);
 881
 882         for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
 883                 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
 884
 885                 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
 886                 kfree(ifsq->ifsq_stage, M_DEVBUF);
 887         }
 888         kfree(ifp->if_snd.altq_subq, M_DEVBUF);
 889
 890         kfree(ifp->if_data_pcpu, M_DEVBUF);
 891
 892         crit_exit();
 893 }
 894
 895 /*
 896  * Create interface group without members
 897  */
 898 struct ifg_group *
 899 if_creategroup(const char *groupname)
 900 {
 901         struct ifg_group        *ifg = NULL;
 902
 903         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
 904             M_TEMP, M_NOWAIT)) == NULL)
 905                 return (NULL);
 906
 907         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
 908         ifg->ifg_refcnt = 0;
 909         ifg->ifg_carp_demoted = 0;
 910         TAILQ_INIT(&ifg->ifg_members);
 911 #if NPF > 0
 912         pfi_attach_ifgroup(ifg);
 913 #endif
 914         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
 915
 916         return (ifg);
 917 }
 918
 919 /*
 920  * Add a group to an interface
 921  */
 922 int
 923 if_addgroup(struct ifnet *ifp, const char *groupname)
 924 {
 925         struct ifg_list         *ifgl;
 926         struct ifg_group        *ifg = NULL;
 927         struct ifg_member       *ifgm;
 928
 929         if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
 930             groupname[strlen(groupname) - 1] <= '9')
 931                 return (EINVAL);
 932
 933         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
 934                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
 935                         return (EEXIST);
 936
 937         if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
 938                 return (ENOMEM);
 939
 940         if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
 941                 kfree(ifgl, M_TEMP);
 942                 return (ENOMEM);
 943         }
 944
 945         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
 946                 if (!strcmp(ifg->ifg_group, groupname))
 947                         break;
 948
 949         if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
 950                 kfree(ifgl, M_TEMP);
 951                 kfree(ifgm, M_TEMP);
 952                 return (ENOMEM);
 953         }
 954
 955         ifg->ifg_refcnt++;
 956         ifgl->ifgl_group = ifg;
 957         ifgm->ifgm_ifp = ifp;
 958
 959         TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
 960         TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
 961
 962 #if NPF > 0
 963         pfi_group_change(groupname);
 964 #endif
 965
 966         return (0);
 967 }
 968
 969 /*
 970  * Remove a group from an interface
 971  */
 972 int
 973 if_delgroup(struct ifnet *ifp, const char *groupname)
 974 {
 975         struct ifg_list         *ifgl;
 976         struct ifg_member       *ifgm;
 977
 978         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
 979                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
 980                         break;
 981         if (ifgl == NULL)
 982                 return (ENOENT);
 983
 984         TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
 985
 986         TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
 987                 if (ifgm->ifgm_ifp == ifp)
 988                         break;
 989
 990         if (ifgm != NULL) {
 991                 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
 992                 kfree(ifgm, M_TEMP);
 993         }
 994
 995         if (--ifgl->ifgl_group->ifg_refcnt == 0) {
 996                 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
 997 #if NPF > 0
 998                 pfi_detach_ifgroup(ifgl->ifgl_group);
 999 #endif
1000                 kfree(ifgl->ifgl_group, M_TEMP);
1001         }
1002
1003         kfree(ifgl, M_TEMP);
1004
1005 #if NPF > 0
1006         pfi_group_change(groupname);
1007 #endif
1008
1009         return (0);
1010 }
1011
1012 /*
1013  * Stores all groups from an interface in memory pointed
1014  * to by data
1015  */
1016 int
1017 if_getgroup(caddr_t data, struct ifnet *ifp)
1018 {
1019         int                      len, error;
1020         struct ifg_list         *ifgl;
1021         struct ifg_req           ifgrq, *ifgp;
1022         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1023
1024         if (ifgr->ifgr_len == 0) {
1025                 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1026                         ifgr->ifgr_len += sizeof(struct ifg_req);
1027                 return (0);
1028         }
1029
1030         len = ifgr->ifgr_len;
1031         ifgp = ifgr->ifgr_groups;
1032         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1033                 if (len < sizeof(ifgrq))
1034                         return (EINVAL);
1035                 bzero(&ifgrq, sizeof ifgrq);
1036                 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1037                     sizeof(ifgrq.ifgrq_group));
1038                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1039                     sizeof(struct ifg_req))))
1040                         return (error);
1041                 len -= sizeof(ifgrq);
1042                 ifgp++;
1043         }
1044
1045         return (0);
1046 }
1047
1048 /*
1049  * Stores all members of a group in memory pointed to by data
1050  */
1051 int
1052 if_getgroupmembers(caddr_t data)
1053 {
1054         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1055         struct ifg_group        *ifg;
1056         struct ifg_member       *ifgm;
1057         struct ifg_req           ifgrq, *ifgp;
1058         int                      len, error;
1059
1060         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1061                 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1062                         break;
1063         if (ifg == NULL)
1064                 return (ENOENT);
1065
1066         if (ifgr->ifgr_len == 0) {
1067                 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1068                         ifgr->ifgr_len += sizeof(ifgrq);
1069                 return (0);
1070         }
1071
1072         len = ifgr->ifgr_len;
1073         ifgp = ifgr->ifgr_groups;
1074         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1075                 if (len < sizeof(ifgrq))
1076                         return (EINVAL);
1077                 bzero(&ifgrq, sizeof ifgrq);
1078                 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1079                     sizeof(ifgrq.ifgrq_member));
1080                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1081                     sizeof(struct ifg_req))))
1082                         return (error);
1083                 len -= sizeof(ifgrq);
1084                 ifgp++;
1085         }
1086
1087         return (0);
1088 }
1089
1090 /*
1091  * Delete Routes for a Network Interface
1092  *
1093  * Called for each routing entry via the rnh->rnh_walktree() call above
1094  * to delete all route entries referencing a detaching network interface.
1095  *
1096  * Arguments:
1097  *      rn      pointer to node in the routing table
1098  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1099  *
1100  * Returns:
1101  *      0       successful
1102  *      errno   failed - reason indicated
1103  *
1104  */
1105 static int
1106 if_rtdel(struct radix_node *rn, void *arg)
1107 {
1108         struct rtentry  *rt = (struct rtentry *)rn;
1109         struct ifnet    *ifp = arg;
1110         int             err;
1111
1112         if (rt->rt_ifp == ifp) {
1113
1114                 /*
1115                  * Protect (sorta) against walktree recursion problems
1116                  * with cloned routes
1117                  */
1118                 if (!(rt->rt_flags & RTF_UP))
1119                         return (0);
1120
1121                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1122                                 rt_mask(rt), rt->rt_flags,
1123                                 NULL);
1124                 if (err) {
1125                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1126                 }
1127         }
1128
1129         return (0);
1130 }
1131
1132 /*
1133  * Locate an interface based on a complete address.
1134  */
1135 struct ifaddr *
1136 ifa_ifwithaddr(struct sockaddr *addr)
1137 {
1138         struct ifnet *ifp;
1139
1140         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1141                 struct ifaddr_container *ifac;
1142
1143                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1144                         struct ifaddr *ifa = ifac->ifa;
1145
1146                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1147                                 continue;
1148                         if (sa_equal(addr, ifa->ifa_addr))
1149                                 return (ifa);
1150                         if ((ifp->if_flags & IFF_BROADCAST) &&
1151                             ifa->ifa_broadaddr &&
1152                             /* IPv6 doesn't have broadcast */
1153                             ifa->ifa_broadaddr->sa_len != 0 &&
1154                             sa_equal(ifa->ifa_broadaddr, addr))
1155                                 return (ifa);
1156                 }
1157         }
1158         return (NULL);
1159 }
1160 /*
1161  * Locate the point to point interface with a given destination address.
1162  */
1163 struct ifaddr *
1164 ifa_ifwithdstaddr(struct sockaddr *addr)
1165 {
1166         struct ifnet *ifp;
1167
1168         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1169                 struct ifaddr_container *ifac;
1170
1171                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1172                         continue;
1173
1174                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1175                         struct ifaddr *ifa = ifac->ifa;
1176
1177                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1178                                 continue;
1179                         if (ifa->ifa_dstaddr &&
1180                             sa_equal(addr, ifa->ifa_dstaddr))
1181                                 return (ifa);
1182                 }
1183         }
1184         return (NULL);
1185 }
1186
1187 /*
1188  * Find an interface on a specific network.  If many, choice
1189  * is most specific found.
1190  */
1191 struct ifaddr *
1192 ifa_ifwithnet(struct sockaddr *addr)
1193 {
1194         struct ifnet *ifp;
1195         struct ifaddr *ifa_maybe = NULL;
1196         u_int af = addr->sa_family;
1197         char *addr_data = addr->sa_data, *cplim;
1198
1199         /*
1200          * AF_LINK addresses can be looked up directly by their index number,
1201          * so do that if we can.
1202          */
1203         if (af == AF_LINK) {
1204                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1205
1206                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1207                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1208         }
1209
1210         /*
1211          * Scan though each interface, looking for ones that have
1212          * addresses in this address family.
1213          */
1214         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1215                 struct ifaddr_container *ifac;
1216
1217                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1218                         struct ifaddr *ifa = ifac->ifa;
1219                         char *cp, *cp2, *cp3;
1220
1221                         if (ifa->ifa_addr->sa_family != af)
1222 next:                           continue;
1223                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1224                                 /*
1225                                  * This is a bit broken as it doesn't
1226                                  * take into account that the remote end may
1227                                  * be a single node in the network we are
1228                                  * looking for.
1229                                  * The trouble is that we don't know the
1230                                  * netmask for the remote end.
1231                                  */
1232                                 if (ifa->ifa_dstaddr != NULL &&
1233                                     sa_equal(addr, ifa->ifa_dstaddr))
1234                                         return (ifa);
1235                         } else {
1236                                 /*
1237                                  * if we have a special address handler,
1238                                  * then use it instead of the generic one.
1239                                  */
1240                                 if (ifa->ifa_claim_addr) {
1241                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1242                                                 return (ifa);
1243                                         } else {
1244                                                 continue;
1245                                         }
1246                                 }
1247
1248                                 /*
1249                                  * Scan all the bits in the ifa's address.
1250                                  * If a bit dissagrees with what we are
1251                                  * looking for, mask it with the netmask
1252                                  * to see if it really matters.
1253                                  * (A byte at a time)
1254                                  */
1255                                 if (ifa->ifa_netmask == 0)
1256                                         continue;
1257                                 cp = addr_data;
1258                                 cp2 = ifa->ifa_addr->sa_data;
1259                                 cp3 = ifa->ifa_netmask->sa_data;
1260                                 cplim = ifa->ifa_netmask->sa_len +
1261                                         (char *)ifa->ifa_netmask;
1262                                 while (cp3 < cplim)
1263                                         if ((*cp++ ^ *cp2++) & *cp3++)
1264                                                 goto next; /* next address! */
1265                                 /*
1266                                  * If the netmask of what we just found
1267                                  * is more specific than what we had before
1268                                  * (if we had one) then remember the new one
1269                                  * before continuing to search
1270                                  * for an even better one.
1271                                  */
1272                                 if (ifa_maybe == NULL ||
1273                                     rn_refines((char *)ifa->ifa_netmask,
1274                                                (char *)ifa_maybe->ifa_netmask))
1275                                         ifa_maybe = ifa;
1276                         }
1277                 }
1278         }
1279         return (ifa_maybe);
1280 }
1281
1282 /*
1283  * Find an interface address specific to an interface best matching
1284  * a given address.
1285  */
1286 struct ifaddr *
1287 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1288 {
1289         struct ifaddr_container *ifac;
1290         char *cp, *cp2, *cp3;
1291         char *cplim;
1292         struct ifaddr *ifa_maybe = NULL;
1293         u_int af = addr->sa_family;
1294
1295         if (af >= AF_MAX)
1296                 return (0);
1297         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1298                 struct ifaddr *ifa = ifac->ifa;
1299
1300                 if (ifa->ifa_addr->sa_family != af)
1301                         continue;
1302                 if (ifa_maybe == NULL)
1303                         ifa_maybe = ifa;
1304                 if (ifa->ifa_netmask == NULL) {
1305                         if (sa_equal(addr, ifa->ifa_addr) ||
1306                             (ifa->ifa_dstaddr != NULL &&
1307                              sa_equal(addr, ifa->ifa_dstaddr)))
1308                                 return (ifa);
1309                         continue;
1310                 }
1311                 if (ifp->if_flags & IFF_POINTOPOINT) {
1312                         if (sa_equal(addr, ifa->ifa_dstaddr))
1313                                 return (ifa);
1314                 } else {
1315                         cp = addr->sa_data;
1316                         cp2 = ifa->ifa_addr->sa_data;
1317                         cp3 = ifa->ifa_netmask->sa_data;
1318                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1319                         for (; cp3 < cplim; cp3++)
1320                                 if ((*cp++ ^ *cp2++) & *cp3)
1321                                         break;
1322                         if (cp3 == cplim)
1323                                 return (ifa);
1324                 }
1325         }
1326         return (ifa_maybe);
1327 }
1328
1329 /*
1330  * Default action when installing a route with a Link Level gateway.
1331  * Lookup an appropriate real ifa to point to.
1332  * This should be moved to /sys/net/link.c eventually.
1333  */
1334 static void
1335 link_rtrequest(int cmd, struct rtentry *rt)
1336 {
1337         struct ifaddr *ifa;
1338         struct sockaddr *dst;
1339         struct ifnet *ifp;
1340
1341         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1342             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1343                 return;
1344         ifa = ifaof_ifpforaddr(dst, ifp);
1345         if (ifa != NULL) {
1346                 IFAFREE(rt->rt_ifa);
1347                 IFAREF(ifa);
1348                 rt->rt_ifa = ifa;
1349                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1350                         ifa->ifa_rtrequest(cmd, rt);
1351         }
1352 }
1353
1354 /*
1355  * Mark an interface down and notify protocols of
1356  * the transition.
1357  * NOTE: must be called at splnet or eqivalent.
1358  */
1359 void
1360 if_unroute(struct ifnet *ifp, int flag, int fam)
1361 {
1362         struct ifaddr_container *ifac;
1363
1364         ifp->if_flags &= ~flag;
1365         getmicrotime(&ifp->if_lastchange);
1366         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1367                 struct ifaddr *ifa = ifac->ifa;
1368
1369                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1370                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1371         }
1372         ifq_purge_all(&ifp->if_snd);
1373         rt_ifmsg(ifp);
1374 }
1375
1376 /*
1377  * Mark an interface up and notify protocols of
1378  * the transition.
1379  * NOTE: must be called at splnet or eqivalent.
1380  */
1381 void
1382 if_route(struct ifnet *ifp, int flag, int fam)
1383 {
1384         struct ifaddr_container *ifac;
1385
1386         ifq_purge_all(&ifp->if_snd);
1387         ifp->if_flags |= flag;
1388         getmicrotime(&ifp->if_lastchange);
1389         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1390                 struct ifaddr *ifa = ifac->ifa;
1391
1392                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1393                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1394         }
1395         rt_ifmsg(ifp);
1396 #ifdef INET6
1397         in6_if_up(ifp);
1398 #endif
1399 }
1400
1401 /*
1402  * Mark an interface down and notify protocols of the transition.  An
1403  * interface going down is also considered to be a synchronizing event.
1404  * We must ensure that all packet processing related to the interface
1405  * has completed before we return so e.g. the caller can free the ifnet
1406  * structure that the mbufs may be referencing.
1407  *
1408  * NOTE: must be called at splnet or eqivalent.
1409  */
1410 void
1411 if_down(struct ifnet *ifp)
1412 {
1413         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1414         netmsg_service_sync();
1415 }
1416
1417 /*
1418  * Mark an interface up and notify protocols of
1419  * the transition.
1420  * NOTE: must be called at splnet or eqivalent.
1421  */
1422 void
1423 if_up(struct ifnet *ifp)
1424 {
1425         if_route(ifp, IFF_UP, AF_UNSPEC);
1426 }
1427
1428 /*
1429  * Process a link state change.
1430  * NOTE: must be called at splsoftnet or equivalent.
1431  */
1432 void
1433 if_link_state_change(struct ifnet *ifp)
1434 {
1435         int link_state = ifp->if_link_state;
1436
1437         rt_ifmsg(ifp);
1438         devctl_notify("IFNET", ifp->if_xname,
1439             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1440 }
1441
1442 /*
1443  * Handle interface watchdog timer routines.  Called
1444  * from softclock, we decrement timers (if set) and
1445  * call the appropriate interface routine on expiration.
1446  */
1447 static void
1448 if_slowtimo(void *arg)
1449 {
1450         struct ifnet *ifp;
1451
1452         crit_enter();
1453
1454         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1455                 if (if_stats_compat) {
1456                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1457                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1458                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1459                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1460                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1461                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1462                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1463                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1464                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1465                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1466                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1467                 }
1468
1469                 if (ifp->if_timer == 0 || --ifp->if_timer)
1470                         continue;
1471                 if (ifp->if_watchdog) {
1472                         if (ifnet_tryserialize_all(ifp)) {
1473                                 (*ifp->if_watchdog)(ifp);
1474                                 ifnet_deserialize_all(ifp);
1475                         } else {
1476                                 /* try again next timeout */
1477                                 ++ifp->if_timer;
1478                         }
1479                 }
1480         }
1481
1482         crit_exit();
1483
1484         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1485 }
1486
1487 /*
1488  * Map interface name to
1489  * interface structure pointer.
1490  */
1491 struct ifnet *
1492 ifunit(const char *name)
1493 {
1494         struct ifnet *ifp;
1495
1496         /*
1497          * Search all the interfaces for this name/number
1498          */
1499
1500         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1501                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1502                         break;
1503         }
1504         return (ifp);
1505 }
1506
1507
1508 /*
1509  * Map interface name in a sockaddr_dl to
1510  * interface structure pointer.
1511  */
1512 struct ifnet *
1513 if_withname(struct sockaddr *sa)
1514 {
1515         char ifname[IFNAMSIZ+1];
1516         struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1517
1518         if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1519              (sdl->sdl_nlen > IFNAMSIZ) )
1520                 return NULL;
1521
1522         /*
1523          * ifunit wants a null-terminated name.  It may not be null-terminated
1524          * in the sockaddr.  We don't want to change the caller's sockaddr,
1525          * and there might not be room to put the trailing null anyway, so we
1526          * make a local copy that we know we can null terminate safely.
1527          */
1528
1529         bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1530         ifname[sdl->sdl_nlen] = '\0';
1531         return ifunit(ifname);
1532 }
1533
1534
1535 /*
1536  * Interface ioctls.
1537  */
1538 int
1539 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1540 {
1541         struct ifnet *ifp;
1542         struct ifreq *ifr;
1543         struct ifstat *ifs;
1544         int error;
1545         short oif_flags;
1546         int new_flags;
1547 #ifdef COMPAT_43
1548         int ocmd;
1549 #endif
1550         size_t namelen, onamelen;
1551         char new_name[IFNAMSIZ];
1552         struct ifaddr *ifa;
1553         struct sockaddr_dl *sdl;
1554
1555         switch (cmd) {
1556         case SIOCGIFCONF:
1557         case OSIOCGIFCONF:
1558                 return (ifconf(cmd, data, cred));
1559         default:
1560                 break;
1561         }
1562
1563         ifr = (struct ifreq *)data;
1564
1565         switch (cmd) {
1566         case SIOCIFCREATE:
1567         case SIOCIFCREATE2:
1568                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1569                         return (error);
1570                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1571                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1572         case SIOCIFDESTROY:
1573                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1574                         return (error);
1575                 return (if_clone_destroy(ifr->ifr_name));
1576         case SIOCIFGCLONERS:
1577                 return (if_clone_list((struct if_clonereq *)data));
1578         default:
1579                 break;
1580         }
1581
1582         /*
1583          * Nominal ioctl through interface, lookup the ifp and obtain a
1584          * lock to serialize the ifconfig ioctl operation.
1585          */
1586         ifp = ifunit(ifr->ifr_name);
1587         if (ifp == NULL)
1588                 return (ENXIO);
1589         error = 0;
1590         mtx_lock(&ifp->if_ioctl_mtx);
1591
1592         switch (cmd) {
1593         case SIOCGIFINDEX:
1594                 ifr->ifr_index = ifp->if_index;
1595                 break;
1596
1597         case SIOCGIFFLAGS:
1598                 ifr->ifr_flags = ifp->if_flags;
1599                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1600                 break;
1601
1602         case SIOCGIFCAP:
1603                 ifr->ifr_reqcap = ifp->if_capabilities;
1604                 ifr->ifr_curcap = ifp->if_capenable;
1605                 break;
1606
1607         case SIOCGIFMETRIC:
1608                 ifr->ifr_metric = ifp->if_metric;
1609                 break;
1610
1611         case SIOCGIFMTU:
1612                 ifr->ifr_mtu = ifp->if_mtu;
1613                 break;
1614
1615         case SIOCGIFTSOLEN:
1616                 ifr->ifr_tsolen = ifp->if_tsolen;
1617                 break;
1618
1619         case SIOCGIFDATA:
1620                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1621                                 sizeof(ifp->if_data));
1622                 break;
1623
1624         case SIOCGIFPHYS:
1625                 ifr->ifr_phys = ifp->if_physical;
1626                 break;
1627
1628         case SIOCGIFPOLLCPU:
1629                 ifr->ifr_pollcpu = -1;
1630                 break;
1631
1632         case SIOCSIFPOLLCPU:
1633                 break;
1634
1635         case SIOCSIFFLAGS:
1636                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1637                 if (error)
1638                         break;
1639                 new_flags = (ifr->ifr_flags & 0xffff) |
1640                     (ifr->ifr_flagshigh << 16);
1641                 if (ifp->if_flags & IFF_SMART) {
1642                         /* Smart drivers twiddle their own routes */
1643                 } else if (ifp->if_flags & IFF_UP &&
1644                     (new_flags & IFF_UP) == 0) {
1645                         crit_enter();
1646                         if_down(ifp);
1647                         crit_exit();
1648                 } else if (new_flags & IFF_UP &&
1649                     (ifp->if_flags & IFF_UP) == 0) {
1650                         crit_enter();
1651                         if_up(ifp);
1652                         crit_exit();
1653                 }
1654
1655 #ifdef IFPOLL_ENABLE
1656                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1657                         if (new_flags & IFF_NPOLLING)
1658                                 ifpoll_register(ifp);
1659                         else
1660                                 ifpoll_deregister(ifp);
1661                 }
1662 #endif
1663
1664                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1665                         (new_flags &~ IFF_CANTCHANGE);
1666                 if (new_flags & IFF_PPROMISC) {
1667                         /* Permanently promiscuous mode requested */
1668                         ifp->if_flags |= IFF_PROMISC;
1669                 } else if (ifp->if_pcount == 0) {
1670                         ifp->if_flags &= ~IFF_PROMISC;
1671                 }
1672                 if (ifp->if_ioctl) {
1673                         ifnet_serialize_all(ifp);
1674                         ifp->if_ioctl(ifp, cmd, data, cred);
1675                         ifnet_deserialize_all(ifp);
1676                 }
1677                 getmicrotime(&ifp->if_lastchange);
1678                 break;
1679
1680         case SIOCSIFCAP:
1681                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1682                 if (error)
1683                         break;
1684                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1685                         error = EINVAL;
1686                         break;
1687                 }
1688                 ifnet_serialize_all(ifp);
1689                 ifp->if_ioctl(ifp, cmd, data, cred);
1690                 ifnet_deserialize_all(ifp);
1691                 break;
1692
1693         case SIOCSIFNAME:
1694                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1695                 if (error)
1696                         break;
1697                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1698                 if (error)
1699                         break;
1700                 if (new_name[0] == '\0') {
1701                         error = EINVAL;
1702                         break;
1703                 }
1704                 if (ifunit(new_name) != NULL) {
1705                         error = EEXIST;
1706                         break;
1707                 }
1708
1709                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1710
1711                 /* Announce the departure of the interface. */
1712                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1713
1714                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1715                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1716                 /* XXX IFA_LOCK(ifa); */
1717                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1718                 namelen = strlen(new_name);
1719                 onamelen = sdl->sdl_nlen;
1720                 /*
1721                  * Move the address if needed.  This is safe because we
1722                  * allocate space for a name of length IFNAMSIZ when we
1723                  * create this in if_attach().
1724                  */
1725                 if (namelen != onamelen) {
1726                         bcopy(sdl->sdl_data + onamelen,
1727                             sdl->sdl_data + namelen, sdl->sdl_alen);
1728                 }
1729                 bcopy(new_name, sdl->sdl_data, namelen);
1730                 sdl->sdl_nlen = namelen;
1731                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1732                 bzero(sdl->sdl_data, onamelen);
1733                 while (namelen != 0)
1734                         sdl->sdl_data[--namelen] = 0xff;
1735                 /* XXX IFA_UNLOCK(ifa) */
1736
1737                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1738
1739                 /* Announce the return of the interface. */
1740                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1741                 break;
1742
1743         case SIOCSIFMETRIC:
1744                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1745                 if (error)
1746                         break;
1747                 ifp->if_metric = ifr->ifr_metric;
1748                 getmicrotime(&ifp->if_lastchange);
1749                 break;
1750
1751         case SIOCSIFPHYS:
1752                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1753                 if (error)
1754                         break;
1755                 if (ifp->if_ioctl == NULL) {
1756                         error = EOPNOTSUPP;
1757                         break;
1758                 }
1759                 ifnet_serialize_all(ifp);
1760                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1761                 ifnet_deserialize_all(ifp);
1762                 if (error == 0)
1763                         getmicrotime(&ifp->if_lastchange);
1764                 break;
1765
1766         case SIOCSIFMTU:
1767         {
1768                 u_long oldmtu = ifp->if_mtu;
1769
1770                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1771                 if (error)
1772                         break;
1773                 if (ifp->if_ioctl == NULL) {
1774                         error = EOPNOTSUPP;
1775                         break;
1776                 }
1777                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
1778                         error = EINVAL;
1779                         break;
1780                 }
1781                 ifnet_serialize_all(ifp);
1782                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1783                 ifnet_deserialize_all(ifp);
1784                 if (error == 0) {
1785                         getmicrotime(&ifp->if_lastchange);
1786                         rt_ifmsg(ifp);
1787                 }
1788                 /*
1789                  * If the link MTU changed, do network layer specific procedure.
1790                  */
1791                 if (ifp->if_mtu != oldmtu) {
1792 #ifdef INET6
1793                         nd6_setmtu(ifp);
1794 #endif
1795                 }
1796                 break;
1797         }
1798
1799         case SIOCSIFTSOLEN:
1800                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1801                 if (error)
1802                         break;
1803
1804                 /* XXX need driver supplied upper limit */
1805                 if (ifr->ifr_tsolen <= 0) {
1806                         error = EINVAL;
1807                         break;
1808                 }
1809                 ifp->if_tsolen = ifr->ifr_tsolen;
1810                 break;
1811
1812         case SIOCADDMULTI:
1813         case SIOCDELMULTI:
1814                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1815                 if (error)
1816                         break;
1817
1818                 /* Don't allow group membership on non-multicast interfaces. */
1819                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1820                         error = EOPNOTSUPP;
1821                         break;
1822                 }
1823
1824                 /* Don't let users screw up protocols' entries. */
1825                 if (ifr->ifr_addr.sa_family != AF_LINK) {
1826                         error = EINVAL;
1827                         break;
1828                 }
1829
1830                 if (cmd == SIOCADDMULTI) {
1831                         struct ifmultiaddr *ifma;
1832                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1833                 } else {
1834                         error = if_delmulti(ifp, &ifr->ifr_addr);
1835                 }
1836                 if (error == 0)
1837                         getmicrotime(&ifp->if_lastchange);
1838                 break;
1839
1840         case SIOCSIFPHYADDR:
1841         case SIOCDIFPHYADDR:
1842 #ifdef INET6
1843         case SIOCSIFPHYADDR_IN6:
1844 #endif
1845         case SIOCSLIFPHYADDR:
1846         case SIOCSIFMEDIA:
1847         case SIOCSIFGENERIC:
1848                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1849                 if (error)
1850                         break;
1851                 if (ifp->if_ioctl == 0) {
1852                         error = EOPNOTSUPP;
1853                         break;
1854                 }
1855                 ifnet_serialize_all(ifp);
1856                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1857                 ifnet_deserialize_all(ifp);
1858                 if (error == 0)
1859                         getmicrotime(&ifp->if_lastchange);
1860                 break;
1861
1862         case SIOCGIFSTATUS:
1863                 ifs = (struct ifstat *)data;
1864                 ifs->ascii[0] = '\0';
1865                 /* fall through */
1866         case SIOCGIFPSRCADDR:
1867         case SIOCGIFPDSTADDR:
1868         case SIOCGLIFPHYADDR:
1869         case SIOCGIFMEDIA:
1870         case SIOCGIFGENERIC:
1871                 if (ifp->if_ioctl == NULL) {
1872                         error = EOPNOTSUPP;
1873                         break;
1874                 }
1875                 ifnet_serialize_all(ifp);
1876                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1877                 ifnet_deserialize_all(ifp);
1878                 break;
1879
1880         case SIOCSIFLLADDR:
1881                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1882                 if (error)
1883                         break;
1884                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
1885                                      ifr->ifr_addr.sa_len);
1886                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1887                 break;
1888
1889         default:
1890                 oif_flags = ifp->if_flags;
1891                 if (so->so_proto == 0) {
1892                         error = EOPNOTSUPP;
1893                         break;
1894                 }
1895 #ifndef COMPAT_43
1896                 error = so_pru_control_direct(so, cmd, data, ifp);
1897 #else
1898                 ocmd = cmd;
1899
1900                 switch (cmd) {
1901                 case SIOCSIFDSTADDR:
1902                 case SIOCSIFADDR:
1903                 case SIOCSIFBRDADDR:
1904                 case SIOCSIFNETMASK:
1905 #if BYTE_ORDER != BIG_ENDIAN
1906                         if (ifr->ifr_addr.sa_family == 0 &&
1907                             ifr->ifr_addr.sa_len < 16) {
1908                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1909                                 ifr->ifr_addr.sa_len = 16;
1910                         }
1911 #else
1912                         if (ifr->ifr_addr.sa_len == 0)
1913                                 ifr->ifr_addr.sa_len = 16;
1914 #endif
1915                         break;
1916                 case OSIOCGIFADDR:
1917                         cmd = SIOCGIFADDR;
1918                         break;
1919                 case OSIOCGIFDSTADDR:
1920                         cmd = SIOCGIFDSTADDR;
1921                         break;
1922                 case OSIOCGIFBRDADDR:
1923                         cmd = SIOCGIFBRDADDR;
1924                         break;
1925                 case OSIOCGIFNETMASK:
1926                         cmd = SIOCGIFNETMASK;
1927                         break;
1928                 default:
1929                         break;
1930                 }
1931
1932                 error = so_pru_control_direct(so, cmd, data, ifp);
1933
1934                 switch (ocmd) {
1935                 case OSIOCGIFADDR:
1936                 case OSIOCGIFDSTADDR:
1937                 case OSIOCGIFBRDADDR:
1938                 case OSIOCGIFNETMASK:
1939                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1940                         break;
1941                 }
1942 #endif /* COMPAT_43 */
1943
1944                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1945 #ifdef INET6
1946                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
1947                         if (ifp->if_flags & IFF_UP) {
1948                                 crit_enter();
1949                                 in6_if_up(ifp);
1950                                 crit_exit();
1951                         }
1952 #endif
1953                 }
1954                 break;
1955         }
1956
1957         mtx_unlock(&ifp->if_ioctl_mtx);
1958         return (error);
1959 }
1960
1961 /*
1962  * Set/clear promiscuous mode on interface ifp based on the truth value
1963  * of pswitch.  The calls are reference counted so that only the first
1964  * "on" request actually has an effect, as does the final "off" request.
1965  * Results are undefined if the "off" and "on" requests are not matched.
1966  */
1967 int
1968 ifpromisc(struct ifnet *ifp, int pswitch)
1969 {
1970         struct ifreq ifr;
1971         int error;
1972         int oldflags;
1973
1974         oldflags = ifp->if_flags;
1975         if (ifp->if_flags & IFF_PPROMISC) {
1976                 /* Do nothing if device is in permanently promiscuous mode */
1977                 ifp->if_pcount += pswitch ? 1 : -1;
1978                 return (0);
1979         }
1980         if (pswitch) {
1981                 /*
1982                  * If the device is not configured up, we cannot put it in
1983                  * promiscuous mode.
1984                  */
1985                 if ((ifp->if_flags & IFF_UP) == 0)
1986                         return (ENETDOWN);
1987                 if (ifp->if_pcount++ != 0)
1988                         return (0);
1989                 ifp->if_flags |= IFF_PROMISC;
1990                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1991                     ifp->if_xname);
1992         } else {
1993                 if (--ifp->if_pcount > 0)
1994                         return (0);
1995                 ifp->if_flags &= ~IFF_PROMISC;
1996                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
1997                     ifp->if_xname);
1998         }
1999         ifr.ifr_flags = ifp->if_flags;
2000         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2001         ifnet_serialize_all(ifp);
2002         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2003         ifnet_deserialize_all(ifp);
2004         if (error == 0)
2005                 rt_ifmsg(ifp);
2006         else
2007                 ifp->if_flags = oldflags;
2008         return error;
2009 }
2010
2011 /*
2012  * Return interface configuration
2013  * of system.  List may be used
2014  * in later ioctl's (above) to get
2015  * other information.
2016  */
2017 static int
2018 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2019 {
2020         struct ifconf *ifc = (struct ifconf *)data;
2021         struct ifnet *ifp;
2022         struct sockaddr *sa;
2023         struct ifreq ifr, *ifrp;
2024         int space = ifc->ifc_len, error = 0;
2025
2026         ifrp = ifc->ifc_req;
2027         TAILQ_FOREACH(ifp, &ifnet, if_link) {
2028                 struct ifaddr_container *ifac;
2029                 int addrs;
2030
2031                 if (space <= sizeof ifr)
2032                         break;
2033
2034                 /*
2035                  * Zero the stack declared structure first to prevent
2036                  * memory disclosure.
2037                  */
2038                 bzero(&ifr, sizeof(ifr));
2039                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2040                     >= sizeof(ifr.ifr_name)) {
2041                         error = ENAMETOOLONG;
2042                         break;
2043                 }
2044
2045                 addrs = 0;
2046                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2047                         struct ifaddr *ifa = ifac->ifa;
2048
2049                         if (space <= sizeof ifr)
2050                                 break;
2051                         sa = ifa->ifa_addr;
2052                         if (cred->cr_prison &&
2053                             prison_if(cred, sa))
2054                                 continue;
2055                         addrs++;
2056 #ifdef COMPAT_43
2057                         if (cmd == OSIOCGIFCONF) {
2058                                 struct osockaddr *osa =
2059                                          (struct osockaddr *)&ifr.ifr_addr;
2060                                 ifr.ifr_addr = *sa;
2061                                 osa->sa_family = sa->sa_family;
2062                                 error = copyout(&ifr, ifrp, sizeof ifr);
2063                                 ifrp++;
2064                         } else
2065 #endif
2066                         if (sa->sa_len <= sizeof(*sa)) {
2067                                 ifr.ifr_addr = *sa;
2068                                 error = copyout(&ifr, ifrp, sizeof ifr);
2069                                 ifrp++;
2070                         } else {
2071                                 if (space < (sizeof ifr) + sa->sa_len -
2072                                             sizeof(*sa))
2073                                         break;
2074                                 space -= sa->sa_len - sizeof(*sa);
2075                                 error = copyout(&ifr, ifrp,
2076                                                 sizeof ifr.ifr_name);
2077                                 if (error == 0)
2078                                         error = copyout(sa, &ifrp->ifr_addr,
2079                                                         sa->sa_len);
2080                                 ifrp = (struct ifreq *)
2081                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2082                         }
2083                         if (error)
2084                                 break;
2085                         space -= sizeof ifr;
2086                 }
2087                 if (error)
2088                         break;
2089                 if (!addrs) {
2090                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2091                         error = copyout(&ifr, ifrp, sizeof ifr);
2092                         if (error)
2093                                 break;
2094                         space -= sizeof ifr;
2095                         ifrp++;
2096                 }
2097         }
2098         ifc->ifc_len -= space;
2099         return (error);
2100 }
2101
2102 /*
2103  * Just like if_promisc(), but for all-multicast-reception mode.
2104  */
2105 int
2106 if_allmulti(struct ifnet *ifp, int onswitch)
2107 {
2108         int error = 0;
2109         struct ifreq ifr;
2110
2111         crit_enter();
2112
2113         if (onswitch) {
2114                 if (ifp->if_amcount++ == 0) {
2115                         ifp->if_flags |= IFF_ALLMULTI;
2116                         ifr.ifr_flags = ifp->if_flags;
2117                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2118                         ifnet_serialize_all(ifp);
2119                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2120                                               NULL);
2121                         ifnet_deserialize_all(ifp);
2122                 }
2123         } else {
2124                 if (ifp->if_amcount > 1) {
2125                         ifp->if_amcount--;
2126                 } else {
2127                         ifp->if_amcount = 0;
2128                         ifp->if_flags &= ~IFF_ALLMULTI;
2129                         ifr.ifr_flags = ifp->if_flags;
2130                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2131                         ifnet_serialize_all(ifp);
2132                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2133                                               NULL);
2134                         ifnet_deserialize_all(ifp);
2135                 }
2136         }
2137
2138         crit_exit();
2139
2140         if (error == 0)
2141                 rt_ifmsg(ifp);
2142         return error;
2143 }
2144
2145 /*
2146  * Add a multicast listenership to the interface in question.
2147  * The link layer provides a routine which converts
2148  */
2149 int
2150 if_addmulti(
2151         struct ifnet *ifp,      /* interface to manipulate */
2152         struct sockaddr *sa,    /* address to add */
2153         struct ifmultiaddr **retifma)
2154 {
2155         struct sockaddr *llsa, *dupsa;
2156         int error;
2157         struct ifmultiaddr *ifma;
2158
2159         /*
2160          * If the matching multicast address already exists
2161          * then don't add a new one, just add a reference
2162          */
2163         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2164                 if (sa_equal(sa, ifma->ifma_addr)) {
2165                         ifma->ifma_refcount++;
2166                         if (retifma)
2167                                 *retifma = ifma;
2168                         return 0;
2169                 }
2170         }
2171
2172         /*
2173          * Give the link layer a chance to accept/reject it, and also
2174          * find out which AF_LINK address this maps to, if it isn't one
2175          * already.
2176          */
2177         if (ifp->if_resolvemulti) {
2178                 ifnet_serialize_all(ifp);
2179                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2180                 ifnet_deserialize_all(ifp);
2181                 if (error)
2182                         return error;
2183         } else {
2184                 llsa = NULL;
2185         }
2186
2187         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2188         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_WAITOK);
2189         bcopy(sa, dupsa, sa->sa_len);
2190
2191         ifma->ifma_addr = dupsa;
2192         ifma->ifma_lladdr = llsa;
2193         ifma->ifma_ifp = ifp;
2194         ifma->ifma_refcount = 1;
2195         ifma->ifma_protospec = 0;
2196         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2197
2198         /*
2199          * Some network interfaces can scan the address list at
2200          * interrupt time; lock them out.
2201          */
2202         crit_enter();
2203         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2204         crit_exit();
2205         if (retifma)
2206                 *retifma = ifma;
2207
2208         if (llsa != NULL) {
2209                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2210                         if (sa_equal(ifma->ifma_addr, llsa))
2211                                 break;
2212                 }
2213                 if (ifma) {
2214                         ifma->ifma_refcount++;
2215                 } else {
2216                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2217                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_WAITOK);
2218                         bcopy(llsa, dupsa, llsa->sa_len);
2219                         ifma->ifma_addr = dupsa;
2220                         ifma->ifma_ifp = ifp;
2221                         ifma->ifma_refcount = 1;
2222                         crit_enter();
2223                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2224                         crit_exit();
2225                 }
2226         }
2227         /*
2228          * We are certain we have added something, so call down to the
2229          * interface to let them know about it.
2230          */
2231         crit_enter();
2232         ifnet_serialize_all(ifp);
2233         if (ifp->if_ioctl)
2234                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2235         ifnet_deserialize_all(ifp);
2236         crit_exit();
2237
2238         return 0;
2239 }
2240
2241 /*
2242  * Remove a reference to a multicast address on this interface.  Yell
2243  * if the request does not match an existing membership.
2244  */
2245 int
2246 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2247 {
2248         struct ifmultiaddr *ifma;
2249
2250         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2251                 if (sa_equal(sa, ifma->ifma_addr))
2252                         break;
2253         if (ifma == NULL)
2254                 return ENOENT;
2255
2256         if (ifma->ifma_refcount > 1) {
2257                 ifma->ifma_refcount--;
2258                 return 0;
2259         }
2260
2261         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2262         sa = ifma->ifma_lladdr;
2263         crit_enter();
2264         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2265         /*
2266          * Make sure the interface driver is notified
2267          * in the case of a link layer mcast group being left.
2268          */
2269         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL) {
2270                 ifnet_serialize_all(ifp);
2271                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2272                 ifnet_deserialize_all(ifp);
2273         }
2274         crit_exit();
2275         kfree(ifma->ifma_addr, M_IFMADDR);
2276         kfree(ifma, M_IFMADDR);
2277         if (sa == NULL)
2278                 return 0;
2279
2280         /*
2281          * Now look for the link-layer address which corresponds to
2282          * this network address.  It had been squirreled away in
2283          * ifma->ifma_lladdr for this purpose (so we don't have
2284          * to call ifp->if_resolvemulti() again), and we saved that
2285          * value in sa above.  If some nasty deleted the
2286          * link-layer address out from underneath us, we can deal because
2287          * the address we stored was is not the same as the one which was
2288          * in the record for the link-layer address.  (So we don't complain
2289          * in that case.)
2290          */
2291         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2292                 if (sa_equal(sa, ifma->ifma_addr))
2293                         break;
2294         if (ifma == NULL)
2295                 return 0;
2296
2297         if (ifma->ifma_refcount > 1) {
2298                 ifma->ifma_refcount--;
2299                 return 0;
2300         }
2301
2302         crit_enter();
2303         ifnet_serialize_all(ifp);
2304         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2305         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2306         ifnet_deserialize_all(ifp);
2307         crit_exit();
2308         kfree(ifma->ifma_addr, M_IFMADDR);
2309         kfree(sa, M_IFMADDR);
2310         kfree(ifma, M_IFMADDR);
2311
2312         return 0;
2313 }
2314
2315 /*
2316  * Delete all multicast group membership for an interface.
2317  * Should be used to quickly flush all multicast filters.
2318  */
2319 void
2320 if_delallmulti(struct ifnet *ifp)
2321 {
2322         struct ifmultiaddr *ifma;
2323         struct ifmultiaddr *next;
2324
2325         TAILQ_FOREACH_MUTABLE(ifma, &ifp->if_multiaddrs, ifma_link, next)
2326                 if_delmulti(ifp, ifma->ifma_addr);
2327 }
2328
2329
2330 /*
2331  * Set the link layer address on an interface.
2332  *
2333  * At this time we only support certain types of interfaces,
2334  * and we don't allow the length of the address to change.
2335  */
2336 int
2337 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2338 {
2339         struct sockaddr_dl *sdl;
2340         struct ifreq ifr;
2341
2342         sdl = IF_LLSOCKADDR(ifp);
2343         if (sdl == NULL)
2344                 return (EINVAL);
2345         if (len != sdl->sdl_alen)       /* don't allow length to change */
2346                 return (EINVAL);
2347         switch (ifp->if_type) {
2348         case IFT_ETHER:                 /* these types use struct arpcom */
2349         case IFT_XETHER:
2350         case IFT_L2VLAN:
2351                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2352                 bcopy(lladdr, LLADDR(sdl), len);
2353                 break;
2354         default:
2355                 return (ENODEV);
2356         }
2357         /*
2358          * If the interface is already up, we need
2359          * to re-init it in order to reprogram its
2360          * address filter.
2361          */
2362         ifnet_serialize_all(ifp);
2363         if ((ifp->if_flags & IFF_UP) != 0) {
2364 #ifdef INET
2365                 struct ifaddr_container *ifac;
2366 #endif
2367
2368                 ifp->if_flags &= ~IFF_UP;
2369                 ifr.ifr_flags = ifp->if_flags;
2370                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2371                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2372                               NULL);
2373                 ifp->if_flags |= IFF_UP;
2374                 ifr.ifr_flags = ifp->if_flags;
2375                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2376                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2377                                  NULL);
2378 #ifdef INET
2379                 /*
2380                  * Also send gratuitous ARPs to notify other nodes about
2381                  * the address change.
2382                  */
2383                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2384                         struct ifaddr *ifa = ifac->ifa;
2385
2386                         if (ifa->ifa_addr != NULL &&
2387                             ifa->ifa_addr->sa_family == AF_INET)
2388                                 arp_gratuitous(ifp, ifa);
2389                 }
2390 #endif
2391         }
2392         ifnet_deserialize_all(ifp);
2393         return (0);
2394 }
2395
2396 struct ifmultiaddr *
2397 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2398 {
2399         struct ifmultiaddr *ifma;
2400
2401         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2402                 if (sa_equal(ifma->ifma_addr, sa))
2403                         break;
2404
2405         return ifma;
2406 }
2407
2408 /*
2409  * This function locates the first real ethernet MAC from a network
2410  * card and loads it into node, returning 0 on success or ENOENT if
2411  * no suitable interfaces were found.  It is used by the uuid code to
2412  * generate a unique 6-byte number.
2413  */
2414 int
2415 if_getanyethermac(uint16_t *node, int minlen)
2416 {
2417         struct ifnet *ifp;
2418         struct sockaddr_dl *sdl;
2419
2420         TAILQ_FOREACH(ifp, &ifnet, if_link) {
2421                 if (ifp->if_type != IFT_ETHER)
2422                         continue;
2423                 sdl = IF_LLSOCKADDR(ifp);
2424                 if (sdl->sdl_alen < minlen)
2425                         continue;
2426                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2427                       minlen);
2428                 return(0);
2429         }
2430         return (ENOENT);
2431 }
2432
2433 /*
2434  * The name argument must be a pointer to storage which will last as
2435  * long as the interface does.  For physical devices, the result of
2436  * device_get_name(dev) is a good choice and for pseudo-devices a
2437  * static string works well.
2438  */
2439 void
2440 if_initname(struct ifnet *ifp, const char *name, int unit)
2441 {
2442         ifp->if_dname = name;
2443         ifp->if_dunit = unit;
2444         if (unit != IF_DUNIT_NONE)
2445                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2446         else
2447                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2448 }
2449
2450 int
2451 if_printf(struct ifnet *ifp, const char *fmt, ...)
2452 {
2453         __va_list ap;
2454         int retval;
2455
2456         retval = kprintf("%s: ", ifp->if_xname);
2457         __va_start(ap, fmt);
2458         retval += kvprintf(fmt, ap);
2459         __va_end(ap);
2460         return (retval);
2461 }
2462
2463 struct ifnet *
2464 if_alloc(uint8_t type)
2465 {
2466         struct ifnet *ifp;
2467         size_t size;
2468
2469         /*
2470          * XXX temporary hack until arpcom is setup in if_l2com
2471          */
2472         if (type == IFT_ETHER)
2473                 size = sizeof(struct arpcom);
2474         else
2475                 size = sizeof(struct ifnet);
2476
2477         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2478
2479         ifp->if_type = type;
2480
2481         if (if_com_alloc[type] != NULL) {
2482                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2483                 if (ifp->if_l2com == NULL) {
2484                         kfree(ifp, M_IFNET);
2485                         return (NULL);
2486                 }
2487         }
2488         return (ifp);
2489 }
2490
2491 void
2492 if_free(struct ifnet *ifp)
2493 {
2494         kfree(ifp, M_IFNET);
2495 }
2496
2497 void
2498 ifq_set_classic(struct ifaltq *ifq)
2499 {
2500         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2501             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2502 }
2503
2504 void
2505 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2506     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2507 {
2508         int q;
2509
2510         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2511         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2512         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2513         KASSERT(request != NULL, ("request is not specified"));
2514
2515         ifq->altq_mapsubq = mapsubq;
2516         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2517                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2518
2519                 ifsq->ifsq_enqueue = enqueue;
2520                 ifsq->ifsq_dequeue = dequeue;
2521                 ifsq->ifsq_request = request;
2522         }
2523 }
2524
2525 int
2526 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2527     struct altq_pktattr *pa __unused)
2528 {
2529         M_ASSERTPKTHDR(m);
2530         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2531             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2532                 m_freem(m);
2533                 return ENOBUFS;
2534         } else {
2535                 m->m_nextpkt = NULL;
2536                 if (ifsq->ifsq_tail == NULL)
2537                         ifsq->ifsq_head = m;
2538                 else
2539                         ifsq->ifsq_tail->m_nextpkt = m;
2540                 ifsq->ifsq_tail = m;
2541                 ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2542                 return 0;
2543         }
2544 }
2545
2546 struct mbuf *
2547 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2548 {
2549         struct mbuf *m;
2550
2551         switch (op) {
2552         case ALTDQ_POLL:
2553                 m = ifsq->ifsq_head;
2554                 break;
2555
2556         case ALTDQ_REMOVE:
2557                 m = ifsq->ifsq_head;
2558                 if (m != NULL) {
2559                         if ((ifsq->ifsq_head = m->m_nextpkt) == NULL)
2560                                 ifsq->ifsq_tail = NULL;
2561                         m->m_nextpkt = NULL;
2562                         ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2563                 }
2564                 break;
2565
2566         default:
2567                 panic("unsupported ALTQ dequeue op: %d", op);
2568         }
2569         return m;
2570 }
2571
2572 int
2573 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2574 {
2575         switch (req) {
2576         case ALTRQ_PURGE:
2577                 for (;;) {
2578                         struct mbuf *m;
2579
2580                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2581                         if (m == NULL)
2582                                 break;
2583                         m_freem(m);
2584                 }
2585                 break;
2586
2587         default:
2588                 panic("unsupported ALTQ request: %d", req);
2589         }
2590         return 0;
2591 }
2592
2593 static void
2594 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2595 {
2596         struct ifnet *ifp = ifsq_get_ifp(ifsq);
2597         int running = 0, need_sched;
2598
2599         /*
2600          * Try to do direct ifnet.if_start on the subqueue first, if there is
2601          * contention on the subqueue hardware serializer, ifnet.if_start on
2602          * the subqueue will be scheduled on the subqueue owner CPU.
2603          */
2604         if (!ifsq_tryserialize_hw(ifsq)) {
2605                 /*
2606                  * Subqueue hardware serializer contention happened,
2607                  * ifnet.if_start on the subqueue is scheduled on
2608                  * the subqueue owner CPU, and we keep going.
2609                  */
2610                 ifsq_ifstart_schedule(ifsq, 1);
2611                 return;
2612         }
2613
2614         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2615                 ifp->if_start(ifp, ifsq);
2616                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2617                         running = 1;
2618         }
2619         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2620
2621         ifsq_deserialize_hw(ifsq);
2622
2623         if (need_sched) {
2624                 /*
2625                  * More data need to be transmitted, ifnet.if_start on the
2626                  * subqueue is scheduled on the subqueue owner CPU, and we
2627                  * keep going.
2628                  * NOTE: ifnet.if_start subqueue interlock is not released.
2629                  */
2630                 ifsq_ifstart_schedule(ifsq, force_sched);
2631         }
2632 }
2633
2634 /*
2635  * Subqeue packets staging mechanism:
2636  *
2637  * The packets enqueued into the subqueue are staged to a certain amount
2638  * before the ifnet.if_start on the subqueue is called.  In this way, the
2639  * driver could avoid writing to hardware registers upon every packet,
2640  * instead, hardware registers could be written when certain amount of
2641  * packets are put onto hardware TX ring.  The measurement on several modern
2642  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
2643  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
2644  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
2645  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
2646  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
2647  *
2648  * Subqueue packets staging is performed for two entry points into drivers'
2649  * transmission function:
2650  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
2651  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
2652  *
2653  * Subqueue packets staging will be stopped upon any of the following
2654  * conditions:
2655  * - If the count of packets enqueued on the current CPU is great than or
2656  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2657  * - If the total length of packets enqueued on the current CPU is great
2658  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
2659  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
2660  *   is usually less than hardware's MTU.
2661  * - ifsq_ifstart_schedule() is not pending on the current CPU and
2662  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
2663  *   released.
2664  * - The if_start_rollup(), which is registered as low priority netisr
2665  *   rollup function, is called; probably because no more work is pending
2666  *   for netisr.
2667  *
2668  * NOTE:
2669  * Currently subqueue packet staging is only performed in netisr threads.
2670  */
2671 int
2672 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2673 {
2674         struct ifaltq *ifq = &ifp->if_snd;
2675         struct ifaltq_subque *ifsq;
2676         int error, start = 0, len, mcast = 0, avoid_start = 0;
2677         struct ifsubq_stage_head *head = NULL;
2678         struct ifsubq_stage *stage = NULL;
2679
2680         ifsq = ifq_map_subq(ifq, mycpuid);
2681         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
2682
2683         len = m->m_pkthdr.len;
2684         if (m->m_flags & M_MCAST)
2685                 mcast = 1;
2686
2687         if (curthread->td_type == TD_TYPE_NETISR) {
2688                 head = &ifsubq_stage_heads[mycpuid];
2689                 stage = ifsq_get_stage(ifsq, mycpuid);
2690
2691                 stage->stg_cnt++;
2692                 stage->stg_len += len;
2693                 if (stage->stg_cnt < ifsq_stage_cntmax &&
2694                     stage->stg_len < (ifp->if_mtu - max_protohdr))
2695                         avoid_start = 1;
2696         }
2697
2698         ALTQ_SQ_LOCK(ifsq);
2699         error = ifsq_enqueue_locked(ifsq, m, pa);
2700         if (error) {
2701                 if (!ifsq_data_ready(ifsq)) {
2702                         ALTQ_SQ_UNLOCK(ifsq);
2703                         return error;
2704                 }
2705                 avoid_start = 0;
2706         }
2707         if (!ifsq_is_started(ifsq)) {
2708                 if (avoid_start) {
2709                         ALTQ_SQ_UNLOCK(ifsq);
2710
2711                         KKASSERT(!error);
2712                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
2713                                 ifsq_stage_insert(head, stage);
2714
2715                         IFNET_STAT_INC(ifp, obytes, len);
2716                         if (mcast)
2717                                 IFNET_STAT_INC(ifp, omcasts, 1);
2718                         return error;
2719                 }
2720
2721                 /*
2722                  * Hold the subqueue interlock of ifnet.if_start
2723                  */
2724                 ifsq_set_started(ifsq);
2725                 start = 1;
2726         }
2727         ALTQ_SQ_UNLOCK(ifsq);
2728
2729         if (!error) {
2730                 IFNET_STAT_INC(ifp, obytes, len);
2731                 if (mcast)
2732                         IFNET_STAT_INC(ifp, omcasts, 1);
2733         }
2734
2735         if (stage != NULL) {
2736                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
2737                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
2738                         if (!avoid_start) {
2739                                 ifsq_stage_remove(head, stage);
2740                                 ifsq_ifstart_schedule(ifsq, 1);
2741                         }
2742                         return error;
2743                 }
2744
2745                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
2746                         ifsq_stage_remove(head, stage);
2747                 } else {
2748                         stage->stg_cnt = 0;
2749                         stage->stg_len = 0;
2750                 }
2751         }
2752
2753         if (!start)
2754                 return error;
2755
2756         ifsq_ifstart_try(ifsq, 0);
2757         return error;
2758 }
2759
2760 void *
2761 ifa_create(int size, int flags)
2762 {
2763         struct ifaddr *ifa;
2764         int i;
2765
2766         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
2767
2768         ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2769         if (ifa == NULL)
2770                 return NULL;
2771
2772         ifa->ifa_containers =
2773             kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
2774                 M_IFADDR, M_WAITOK | M_ZERO);
2775         ifa->ifa_ncnt = ncpus;
2776         for (i = 0; i < ncpus; ++i) {
2777                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2778
2779                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2780                 ifac->ifa = ifa;
2781                 ifac->ifa_refcnt = 1;
2782         }
2783 #ifdef IFADDR_DEBUG
2784         kprintf("alloc ifa %p %d\n", ifa, size);
2785 #endif
2786         return ifa;
2787 }
2788
2789 void
2790 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2791 {
2792         struct ifaddr *ifa = ifac->ifa;
2793
2794         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2795         KKASSERT(ifac->ifa_refcnt == 0);
2796         KASSERT(ifac->ifa_listmask == 0,
2797                 ("ifa is still on %#x lists", ifac->ifa_listmask));
2798
2799         ifac->ifa_magic = IFA_CONTAINER_DEAD;
2800
2801 #ifdef IFADDR_DEBUG_VERBOSE
2802         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2803 #endif
2804
2805         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2806                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
2807         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2808 #ifdef IFADDR_DEBUG
2809                 kprintf("free ifa %p\n", ifa);
2810 #endif
2811                 kfree(ifa->ifa_containers, M_IFADDR);
2812                 kfree(ifa, M_IFADDR);
2813         }
2814 }
2815
2816 static void
2817 ifa_iflink_dispatch(netmsg_t nmsg)
2818 {
2819         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2820         struct ifaddr *ifa = msg->ifa;
2821         struct ifnet *ifp = msg->ifp;
2822         int cpu = mycpuid;
2823         struct ifaddr_container *ifac;
2824
2825         crit_enter();
2826
2827         ifac = &ifa->ifa_containers[cpu];
2828         ASSERT_IFAC_VALID(ifac);
2829         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2830                 ("ifaddr is on if_addrheads"));
2831
2832         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2833         if (msg->tail)
2834                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2835         else
2836                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2837
2838         crit_exit();
2839
2840         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2841 }
2842
2843 void
2844 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2845 {
2846         struct netmsg_ifaddr msg;
2847
2848         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2849                     0, ifa_iflink_dispatch);
2850         msg.ifa = ifa;
2851         msg.ifp = ifp;
2852         msg.tail = tail;
2853
2854         ifa_domsg(&msg.base.lmsg, 0);
2855 }
2856
2857 static void
2858 ifa_ifunlink_dispatch(netmsg_t nmsg)
2859 {
2860         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2861         struct ifaddr *ifa = msg->ifa;
2862         struct ifnet *ifp = msg->ifp;
2863         int cpu = mycpuid;
2864         struct ifaddr_container *ifac;
2865
2866         crit_enter();
2867
2868         ifac = &ifa->ifa_containers[cpu];
2869         ASSERT_IFAC_VALID(ifac);
2870         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2871                 ("ifaddr is not on if_addrhead"));
2872
2873         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2874         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2875
2876         crit_exit();
2877
2878         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2879 }
2880
2881 void
2882 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2883 {
2884         struct netmsg_ifaddr msg;
2885
2886         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2887                     0, ifa_ifunlink_dispatch);
2888         msg.ifa = ifa;
2889         msg.ifp = ifp;
2890
2891         ifa_domsg(&msg.base.lmsg, 0);
2892 }
2893
2894 static void
2895 ifa_destroy_dispatch(netmsg_t nmsg)
2896 {
2897         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2898
2899         IFAFREE(msg->ifa);
2900         ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2901 }
2902
2903 void
2904 ifa_destroy(struct ifaddr *ifa)
2905 {
2906         struct netmsg_ifaddr msg;
2907
2908         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2909                     0, ifa_destroy_dispatch);
2910         msg.ifa = ifa;
2911
2912         ifa_domsg(&msg.base.lmsg, 0);
2913 }
2914
2915 struct lwkt_port *
2916 ifnet_portfn(int cpu)
2917 {
2918         return &ifnet_threads[cpu].td_msgport;
2919 }
2920
2921 void
2922 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2923 {
2924         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2925
2926         if (next_cpu < ncpus)
2927                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2928         else
2929                 lwkt_replymsg(lmsg, 0);
2930 }
2931
2932 int
2933 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
2934 {
2935         KKASSERT(cpu < ncpus);
2936         return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
2937 }
2938
2939 void
2940 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
2941 {
2942         KKASSERT(cpu < ncpus);
2943         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
2944 }
2945
2946 /*
2947  * Generic netmsg service loop.  Some protocols may roll their own but all
2948  * must do the basic command dispatch function call done here.
2949  */
2950 static void
2951 ifnet_service_loop(void *arg __unused)
2952 {
2953         netmsg_t msg;
2954
2955         while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
2956                 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
2957                 msg->base.nm_dispatch(msg);
2958         }
2959 }
2960
2961 static void
2962 if_start_rollup(void)
2963 {
2964         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
2965         struct ifsubq_stage *stage;
2966
2967         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
2968                 struct ifaltq_subque *ifsq = stage->stg_subq;
2969                 int is_sched = 0;
2970
2971                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
2972                         is_sched = 1;
2973                 ifsq_stage_remove(head, stage);
2974
2975                 if (is_sched) {
2976                         ifsq_ifstart_schedule(ifsq, 1);
2977                 } else {
2978                         int start = 0;
2979
2980                         ALTQ_SQ_LOCK(ifsq);
2981                         if (!ifsq_is_started(ifsq)) {
2982                                 /*
2983                                  * Hold the subqueue interlock of
2984                                  * ifnet.if_start
2985                                  */
2986                                 ifsq_set_started(ifsq);
2987                                 start = 1;
2988                         }
2989                         ALTQ_SQ_UNLOCK(ifsq);
2990
2991                         if (start)
2992                                 ifsq_ifstart_try(ifsq, 1);
2993                 }
2994                 KKASSERT((stage->stg_flags &
2995                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
2996         }
2997 }
2998
2999 static void
3000 ifnetinit(void *dummy __unused)
3001 {
3002         int i;
3003
3004         for (i = 0; i < ncpus; ++i) {
3005                 struct thread *thr = &ifnet_threads[i];
3006
3007                 lwkt_create(ifnet_service_loop, NULL, NULL,
3008                             thr, TDF_NOSTART|TDF_FORCE_SPINPORT,
3009                             i, "ifnet %d", i);
3010                 netmsg_service_port_init(&thr->td_msgport);
3011                 lwkt_schedule(thr);
3012         }
3013
3014         for (i = 0; i < ncpus; ++i)
3015                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3016         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3017 }
3018
3019 struct ifnet *
3020 ifnet_byindex(unsigned short idx)
3021 {
3022         if (idx > if_index)
3023                 return NULL;
3024         return ifindex2ifnet[idx];
3025 }
3026
3027 struct ifaddr *
3028 ifaddr_byindex(unsigned short idx)
3029 {
3030         struct ifnet *ifp;
3031
3032         ifp = ifnet_byindex(idx);
3033         if (!ifp)
3034                 return NULL;
3035         return TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
3036 }
3037
3038 void
3039 if_register_com_alloc(u_char type,
3040     if_com_alloc_t *a, if_com_free_t *f)
3041 {
3042
3043         KASSERT(if_com_alloc[type] == NULL,
3044             ("if_register_com_alloc: %d already registered", type));
3045         KASSERT(if_com_free[type] == NULL,
3046             ("if_register_com_alloc: %d free already registered", type));
3047
3048         if_com_alloc[type] = a;
3049         if_com_free[type] = f;
3050 }
3051
3052 void
3053 if_deregister_com_alloc(u_char type)
3054 {
3055
3056         KASSERT(if_com_alloc[type] != NULL,
3057             ("if_deregister_com_alloc: %d not registered", type));
3058         KASSERT(if_com_free[type] != NULL,
3059             ("if_deregister_com_alloc: %d free not registered", type));
3060         if_com_alloc[type] = NULL;
3061         if_com_free[type] = NULL;
3062 }
3063
3064 int
3065 if_ring_count2(int cnt, int cnt_max)
3066 {
3067         int shift = 0;
3068
3069         KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3070             ("invalid ring count max %d", cnt_max));
3071
3072         if (cnt <= 0)
3073                 cnt = cnt_max;
3074         if (cnt > ncpus2)
3075                 cnt = ncpus2;
3076         if (cnt > cnt_max)
3077                 cnt = cnt_max;
3078
3079         while ((1 << (shift + 1)) <= cnt)
3080                 ++shift;
3081         cnt = 1 << shift;
3082
3083         KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3084             ("calculate cnt %d, ncpus2 %d, cnt max %d",
3085              cnt, ncpus2, cnt_max));
3086         return cnt;
3087 }
3088
3089 void
3090 ifq_set_maxlen(struct ifaltq *ifq, int len)
3091 {
3092         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3093 }
3094
3095 int
3096 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3097 {
3098         return ALTQ_SUBQ_INDEX_DEFAULT;
3099 }
3100
3101 int
3102 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3103 {
3104         return (cpuid & ifq->altq_subq_mask);
3105 }
3106
3107 static void
3108 ifsq_watchdog(void *arg)
3109 {
3110         struct ifsubq_watchdog *wd = arg;
3111         struct ifnet *ifp;
3112
3113         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3114                 goto done;
3115
3116         ifp = ifsq_get_ifp(wd->wd_subq);
3117         if (ifnet_tryserialize_all(ifp)) {
3118                 wd->wd_watchdog(wd->wd_subq);
3119                 ifnet_deserialize_all(ifp);
3120         } else {
3121                 /* try again next timeout */
3122                 wd->wd_timer = 1;
3123         }
3124 done:
3125         ifsq_watchdog_reset(wd);
3126 }
3127
3128 static void
3129 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3130 {
3131         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3132             ifsq_get_cpuid(wd->wd_subq));
3133 }
3134
3135 void
3136 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3137     ifsq_watchdog_t watchdog)
3138 {
3139         callout_init_mp(&wd->wd_callout);
3140         wd->wd_timer = 0;
3141         wd->wd_subq = ifsq;
3142         wd->wd_watchdog = watchdog;
3143 }
3144
3145 void
3146 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3147 {
3148         wd->wd_timer = 0;
3149         ifsq_watchdog_reset(wd);
3150 }
3151
3152 void
3153 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3154 {
3155         wd->wd_timer = 0;
3156         callout_stop(&wd->wd_callout);
3157 }