sys/net/if.c

   1 /*
   2  * Copyright (c) 1980, 1986, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)if.c        8.3 (Berkeley) 1/4/94
  30  * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
  31  */
  32
  33 #include "opt_compat.h"
  34 #include "opt_inet6.h"
  35 #include "opt_inet.h"
  36 #include "opt_ifpoll.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/malloc.h>
  40 #include <sys/mbuf.h>
  41 #include <sys/systm.h>
  42 #include <sys/proc.h>
  43 #include <sys/priv.h>
  44 #include <sys/protosw.h>
  45 #include <sys/socket.h>
  46 #include <sys/socketvar.h>
  47 #include <sys/socketops.h>
  48 #include <sys/kernel.h>
  49 #include <sys/ktr.h>
  50 #include <sys/mutex.h>
  51 #include <sys/sockio.h>
  52 #include <sys/syslog.h>
  53 #include <sys/sysctl.h>
  54 #include <sys/domain.h>
  55 #include <sys/thread.h>
  56 #include <sys/serialize.h>
  57 #include <sys/bus.h>
  58
  59 #include <sys/thread2.h>
  60 #include <sys/msgport2.h>
  61 #include <sys/mutex2.h>
  62
  63 #include <net/if.h>
  64 #include <net/if_arp.h>
  65 #include <net/if_dl.h>
  66 #include <net/if_types.h>
  67 #include <net/if_var.h>
  68 #include <net/ifq_var.h>
  69 #include <net/radix.h>
  70 #include <net/route.h>
  71 #include <net/if_clone.h>
  72 #include <net/netisr2.h>
  73 #include <net/netmsg2.h>
  74
  75 #include <machine/atomic.h>
  76 #include <machine/stdarg.h>
  77 #include <machine/smp.h>
  78
  79 #if defined(INET) || defined(INET6)
  80 /*XXX*/
  81 #include <netinet/in.h>
  82 #include <netinet/in_var.h>
  83 #include <netinet/if_ether.h>
  84 #ifdef INET6
  85 #include <netinet6/in6_var.h>
  86 #include <netinet6/in6_ifattach.h>
  87 #endif
  88 #endif
  89
  90 #if defined(COMPAT_43)
  91 #include <emulation/43bsd/43bsd_socket.h>
  92 #endif /* COMPAT_43 */
  93
  94 struct netmsg_ifaddr {
  95         struct netmsg_base base;
  96         struct ifaddr   *ifa;
  97         struct ifnet    *ifp;
  98         int             tail;
  99 };
 100
 101 struct ifsubq_stage_head {
 102         TAILQ_HEAD(, ifsubq_stage)      stg_head;
 103 } __cachealign;
 104
 105 /*
 106  * System initialization
 107  */
 108 static void     if_attachdomain(void *);
 109 static void     if_attachdomain1(struct ifnet *);
 110 static int      ifconf(u_long, caddr_t, struct ucred *);
 111 static void     ifinit(void *);
 112 static void     ifnetinit(void *);
 113 static void     if_slowtimo(void *);
 114 static void     link_rtrequest(int, struct rtentry *);
 115 static int      if_rtdel(struct radix_node *, void *);
 116
 117 /* Helper functions */
 118 static void     ifsq_watchdog_reset(struct ifsubq_watchdog *);
 119 static int      if_delmulti_serialized(struct ifnet *, struct sockaddr *);
 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_oncpu(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         struct globaldata *gd = mycpu;
 326         int running = 0, need_sched;
 327
 328         crit_enter_gd(gd);
 329
 330         lwkt_replymsg(lmsg, 0); /* reply ASAP */
 331
 332         if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) {
 333                 /*
 334                  * We need to chase the subqueue owner CPU change.
 335                  */
 336                 ifsq_ifstart_schedule(ifsq, 1);
 337                 crit_exit_gd(gd);
 338                 return;
 339         }
 340
 341         ifsq_serialize_hw(ifsq);
 342         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
 343                 ifp->if_start(ifp, ifsq);
 344                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 345                         running = 1;
 346         }
 347         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
 348         ifsq_deserialize_hw(ifsq);
 349
 350         if (need_sched) {
 351                 /*
 352                  * More data need to be transmitted, ifnet.if_start is
 353                  * scheduled on the subqueue owner CPU, and we keep going.
 354                  * NOTE: ifnet.if_start subqueue interlock is not released.
 355                  */
 356                 ifsq_ifstart_schedule(ifsq, 0);
 357         }
 358
 359         crit_exit_gd(gd);
 360 }
 361
 362 /* Device driver ifnet.if_start helper function */
 363 void
 364 ifsq_devstart(struct ifaltq_subque *ifsq)
 365 {
 366         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 367         int running = 0;
 368
 369         ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
 370
 371         ALTQ_SQ_LOCK(ifsq);
 372         if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
 373                 ALTQ_SQ_UNLOCK(ifsq);
 374                 return;
 375         }
 376         ifsq_set_started(ifsq);
 377         ALTQ_SQ_UNLOCK(ifsq);
 378
 379         ifp->if_start(ifp, ifsq);
 380
 381         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 382                 running = 1;
 383
 384         if (ifsq_ifstart_need_schedule(ifsq, running)) {
 385                 /*
 386                  * More data need to be transmitted, ifnet.if_start is
 387                  * scheduled on ifnet's CPU, and we keep going.
 388                  * NOTE: ifnet.if_start interlock is not released.
 389                  */
 390                 ifsq_ifstart_schedule(ifsq, 0);
 391         }
 392 }
 393
 394 void
 395 if_devstart(struct ifnet *ifp)
 396 {
 397         ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
 398 }
 399
 400 /* Device driver ifnet.if_start schedule helper function */
 401 void
 402 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
 403 {
 404         ifsq_ifstart_schedule(ifsq, 1);
 405 }
 406
 407 void
 408 if_devstart_sched(struct ifnet *ifp)
 409 {
 410         ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
 411 }
 412
 413 static void
 414 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 415 {
 416         lwkt_serialize_enter(ifp->if_serializer);
 417 }
 418
 419 static void
 420 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 421 {
 422         lwkt_serialize_exit(ifp->if_serializer);
 423 }
 424
 425 static int
 426 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 427 {
 428         return lwkt_serialize_try(ifp->if_serializer);
 429 }
 430
 431 #ifdef INVARIANTS
 432 static void
 433 if_default_serialize_assert(struct ifnet *ifp,
 434                             enum ifnet_serialize slz __unused,
 435                             boolean_t serialized)
 436 {
 437         if (serialized)
 438                 ASSERT_SERIALIZED(ifp->if_serializer);
 439         else
 440                 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
 441 }
 442 #endif
 443
 444 /*
 445  * Attach an interface to the list of "active" interfaces.
 446  *
 447  * The serializer is optional.
 448  */
 449 void
 450 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
 451 {
 452         unsigned socksize, ifasize;
 453         int namelen, masklen;
 454         struct sockaddr_dl *sdl;
 455         struct ifaddr *ifa;
 456         struct ifaltq *ifq;
 457         int i, q;
 458
 459         static int if_indexlim = 8;
 460
 461         if (ifp->if_serialize != NULL) {
 462                 KASSERT(ifp->if_deserialize != NULL &&
 463                         ifp->if_tryserialize != NULL &&
 464                         ifp->if_serialize_assert != NULL,
 465                         ("serialize functions are partially setup"));
 466
 467                 /*
 468                  * If the device supplies serialize functions,
 469                  * then clear if_serializer to catch any invalid
 470                  * usage of this field.
 471                  */
 472                 KASSERT(serializer == NULL,
 473                         ("both serialize functions and default serializer "
 474                          "are supplied"));
 475                 ifp->if_serializer = NULL;
 476         } else {
 477                 KASSERT(ifp->if_deserialize == NULL &&
 478                         ifp->if_tryserialize == NULL &&
 479                         ifp->if_serialize_assert == NULL,
 480                         ("serialize functions are partially setup"));
 481                 ifp->if_serialize = if_default_serialize;
 482                 ifp->if_deserialize = if_default_deserialize;
 483                 ifp->if_tryserialize = if_default_tryserialize;
 484 #ifdef INVARIANTS
 485                 ifp->if_serialize_assert = if_default_serialize_assert;
 486 #endif
 487
 488                 /*
 489                  * The serializer can be passed in from the device,
 490                  * allowing the same serializer to be used for both
 491                  * the interrupt interlock and the device queue.
 492                  * If not specified, the netif structure will use an
 493                  * embedded serializer.
 494                  */
 495                 if (serializer == NULL) {
 496                         serializer = &ifp->if_default_serializer;
 497                         lwkt_serialize_init(serializer);
 498                 }
 499                 ifp->if_serializer = serializer;
 500         }
 501
 502         mtx_init(&ifp->if_ioctl_mtx);
 503         mtx_lock(&ifp->if_ioctl_mtx);
 504
 505         lwkt_gettoken(&ifnet_token);    /* protect if_index and ifnet tailq */
 506         ifp->if_index = ++if_index;
 507
 508         /*
 509          * XXX -
 510          * The old code would work if the interface passed a pre-existing
 511          * chain of ifaddrs to this code.  We don't trust our callers to
 512          * properly initialize the tailq, however, so we no longer allow
 513          * this unlikely case.
 514          */
 515         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
 516                                     M_IFADDR, M_WAITOK | M_ZERO);
 517         for (i = 0; i < ncpus; ++i)
 518                 TAILQ_INIT(&ifp->if_addrheads[i]);
 519
 520         TAILQ_INIT(&ifp->if_multiaddrs);
 521         TAILQ_INIT(&ifp->if_groups);
 522         getmicrotime(&ifp->if_lastchange);
 523         if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
 524                 unsigned int n;
 525                 struct ifnet **q;
 526
 527                 if_indexlim <<= 1;
 528
 529                 /* grow ifindex2ifnet */
 530                 n = if_indexlim * sizeof(*q);
 531                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
 532                 if (ifindex2ifnet) {
 533                         bcopy(ifindex2ifnet, q, n/2);
 534                         kfree(ifindex2ifnet, M_IFADDR);
 535                 }
 536                 ifindex2ifnet = q;
 537         }
 538
 539         ifindex2ifnet[if_index] = ifp;
 540
 541         /*
 542          * create a Link Level name for this device
 543          */
 544         namelen = strlen(ifp->if_xname);
 545         masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
 546         socksize = masklen + ifp->if_addrlen;
 547         if (socksize < sizeof(*sdl))
 548                 socksize = sizeof(*sdl);
 549         socksize = RT_ROUNDUP(socksize);
 550         ifasize = sizeof(struct ifaddr) + 2 * socksize;
 551         ifa = ifa_create(ifasize, M_WAITOK);
 552         sdl = (struct sockaddr_dl *)(ifa + 1);
 553         sdl->sdl_len = socksize;
 554         sdl->sdl_family = AF_LINK;
 555         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
 556         sdl->sdl_nlen = namelen;
 557         sdl->sdl_index = ifp->if_index;
 558         sdl->sdl_type = ifp->if_type;
 559         ifp->if_lladdr = ifa;
 560         ifa->ifa_ifp = ifp;
 561         ifa->ifa_rtrequest = link_rtrequest;
 562         ifa->ifa_addr = (struct sockaddr *)sdl;
 563         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
 564         ifa->ifa_netmask = (struct sockaddr *)sdl;
 565         sdl->sdl_len = masklen;
 566         while (namelen != 0)
 567                 sdl->sdl_data[--namelen] = 0xff;
 568         ifa_iflink(ifa, ifp, 0 /* Insert head */);
 569
 570         ifp->if_data_pcpu = kmalloc_cachealign(
 571             ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
 572
 573         if (ifp->if_mapsubq == NULL)
 574                 ifp->if_mapsubq = ifq_mapsubq_default;
 575
 576         ifq = &ifp->if_snd;
 577         ifq->altq_type = 0;
 578         ifq->altq_disc = NULL;
 579         ifq->altq_flags &= ALTQF_CANTCHANGE;
 580         ifq->altq_tbr = NULL;
 581         ifq->altq_ifp = ifp;
 582
 583         if (ifq->altq_subq_cnt <= 0)
 584                 ifq->altq_subq_cnt = 1;
 585         ifq->altq_subq = kmalloc_cachealign(
 586             ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
 587             M_DEVBUF, M_WAITOK | M_ZERO);
 588
 589         if (ifq->altq_maxlen == 0) {
 590                 if_printf(ifp, "driver didn't set altq_maxlen\n");
 591                 ifq_set_maxlen(ifq, ifqmaxlen);
 592         }
 593
 594         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 595                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 596
 597                 ALTQ_SQ_LOCK_INIT(ifsq);
 598                 ifsq->ifsq_index = q;
 599
 600                 ifsq->ifsq_altq = ifq;
 601                 ifsq->ifsq_ifp = ifp;
 602
 603                 ifsq->ifsq_maxlen = ifq->altq_maxlen;
 604                 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
 605                 ifsq->ifsq_prepended = NULL;
 606                 ifsq->ifsq_started = 0;
 607                 ifsq->ifsq_hw_oactive = 0;
 608                 ifsq_set_cpuid(ifsq, 0);
 609                 if (ifp->if_serializer != NULL)
 610                         ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
 611
 612                 ifsq->ifsq_stage =
 613                     kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
 614                     M_DEVBUF, M_WAITOK | M_ZERO);
 615                 for (i = 0; i < ncpus; ++i)
 616                         ifsq->ifsq_stage[i].stg_subq = ifsq;
 617
 618                 ifsq->ifsq_ifstart_nmsg =
 619                     kmalloc(ncpus * sizeof(struct netmsg_base),
 620                     M_LWKTMSG, M_WAITOK);
 621                 for (i = 0; i < ncpus; ++i) {
 622                         netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
 623                             &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
 624                         ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
 625                 }
 626         }
 627         ifq_set_classic(ifq);
 628
 629         if (!SLIST_EMPTY(&domains))
 630                 if_attachdomain1(ifp);
 631
 632         TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
 633         lwkt_reltoken(&ifnet_token);
 634
 635         /* Announce the interface. */
 636         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
 637         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
 638         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
 639
 640         mtx_unlock(&ifp->if_ioctl_mtx);
 641 }
 642
 643 static void
 644 if_attachdomain(void *dummy)
 645 {
 646         struct ifnet *ifp;
 647
 648         crit_enter();
 649         TAILQ_FOREACH(ifp, &ifnet, if_list)
 650                 if_attachdomain1(ifp);
 651         crit_exit();
 652 }
 653 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
 654         if_attachdomain, NULL);
 655
 656 static void
 657 if_attachdomain1(struct ifnet *ifp)
 658 {
 659         struct domain *dp;
 660
 661         crit_enter();
 662
 663         /* address family dependent data region */
 664         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
 665         SLIST_FOREACH(dp, &domains, dom_next)
 666                 if (dp->dom_ifattach)
 667                         ifp->if_afdata[dp->dom_family] =
 668                                 (*dp->dom_ifattach)(ifp);
 669         crit_exit();
 670 }
 671
 672 /*
 673  * Purge all addresses whose type is _not_ AF_LINK
 674  */
 675 void
 676 if_purgeaddrs_nolink(struct ifnet *ifp)
 677 {
 678         struct ifaddr_container *ifac, *next;
 679
 680         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
 681                               ifa_link, next) {
 682                 struct ifaddr *ifa = ifac->ifa;
 683
 684                 /* Leave link ifaddr as it is */
 685                 if (ifa->ifa_addr->sa_family == AF_LINK)
 686                         continue;
 687 #ifdef INET
 688                 /* XXX: Ugly!! ad hoc just for INET */
 689                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
 690                         struct ifaliasreq ifr;
 691 #ifdef IFADDR_DEBUG_VERBOSE
 692                         int i;
 693
 694                         kprintf("purge in4 addr %p: ", ifa);
 695                         for (i = 0; i < ncpus; ++i)
 696                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 697                         kprintf("\n");
 698 #endif
 699
 700                         bzero(&ifr, sizeof ifr);
 701                         ifr.ifra_addr = *ifa->ifa_addr;
 702                         if (ifa->ifa_dstaddr)
 703                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
 704                         if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp,
 705                                        NULL) == 0)
 706                                 continue;
 707                 }
 708 #endif /* INET */
 709 #ifdef INET6
 710                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
 711 #ifdef IFADDR_DEBUG_VERBOSE
 712                         int i;
 713
 714                         kprintf("purge in6 addr %p: ", ifa);
 715                         for (i = 0; i < ncpus; ++i)
 716                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 717                         kprintf("\n");
 718 #endif
 719
 720                         in6_purgeaddr(ifa);
 721                         /* ifp_addrhead is already updated */
 722                         continue;
 723                 }
 724 #endif /* INET6 */
 725                 ifa_ifunlink(ifa, ifp);
 726                 ifa_destroy(ifa);
 727         }
 728 }
 729
 730 static void
 731 ifq_stage_detach_handler(netmsg_t nmsg)
 732 {
 733         struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
 734         int q;
 735
 736         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 737                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 738                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 739
 740                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
 741                         ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
 742         }
 743         lwkt_replymsg(&nmsg->lmsg, 0);
 744 }
 745
 746 static void
 747 ifq_stage_detach(struct ifaltq *ifq)
 748 {
 749         struct netmsg_base base;
 750         int cpu;
 751
 752         netmsg_init(&base, NULL, &curthread->td_msgport, 0,
 753             ifq_stage_detach_handler);
 754         base.lmsg.u.ms_resultp = ifq;
 755
 756         for (cpu = 0; cpu < ncpus; ++cpu)
 757                 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
 758 }
 759
 760 struct netmsg_if_rtdel {
 761         struct netmsg_base      base;
 762         struct ifnet            *ifp;
 763 };
 764
 765 static void
 766 if_rtdel_dispatch(netmsg_t msg)
 767 {
 768         struct netmsg_if_rtdel *rmsg = (void *)msg;
 769         int i, nextcpu, cpu;
 770
 771         cpu = mycpuid;
 772         for (i = 1; i <= AF_MAX; i++) {
 773                 struct radix_node_head  *rnh;
 774
 775                 if ((rnh = rt_tables[cpu][i]) == NULL)
 776                         continue;
 777                 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
 778         }
 779
 780         nextcpu = cpu + 1;
 781         if (nextcpu < ncpus)
 782                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg);
 783         else
 784                 lwkt_replymsg(&rmsg->base.lmsg, 0);
 785 }
 786
 787 /*
 788  * Detach an interface, removing it from the
 789  * list of "active" interfaces.
 790  */
 791 void
 792 if_detach(struct ifnet *ifp)
 793 {
 794         struct netmsg_if_rtdel msg;
 795         struct domain *dp;
 796         int q;
 797
 798         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
 799
 800         /*
 801          * Remove routes and flush queues.
 802          */
 803         crit_enter();
 804 #ifdef IFPOLL_ENABLE
 805         if (ifp->if_flags & IFF_NPOLLING)
 806                 ifpoll_deregister(ifp);
 807 #endif
 808         if_down(ifp);
 809
 810 #ifdef ALTQ
 811         if (ifq_is_enabled(&ifp->if_snd))
 812                 altq_disable(&ifp->if_snd);
 813         if (ifq_is_attached(&ifp->if_snd))
 814                 altq_detach(&ifp->if_snd);
 815 #endif
 816
 817         /*
 818          * Clean up all addresses.
 819          */
 820         ifp->if_lladdr = NULL;
 821
 822         if_purgeaddrs_nolink(ifp);
 823         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
 824                 struct ifaddr *ifa;
 825
 826                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
 827                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
 828                         ("non-link ifaddr is left on if_addrheads"));
 829
 830                 ifa_ifunlink(ifa, ifp);
 831                 ifa_destroy(ifa);
 832                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
 833                         ("there are still ifaddrs left on if_addrheads"));
 834         }
 835
 836 #ifdef INET
 837         /*
 838          * Remove all IPv4 kernel structures related to ifp.
 839          */
 840         in_ifdetach(ifp);
 841 #endif
 842
 843 #ifdef INET6
 844         /*
 845          * Remove all IPv6 kernel structs related to ifp.  This should be done
 846          * before removing routing entries below, since IPv6 interface direct
 847          * routes are expected to be removed by the IPv6-specific kernel API.
 848          * Otherwise, the kernel will detect some inconsistency and bark it.
 849          */
 850         in6_ifdetach(ifp);
 851 #endif
 852
 853         /*
 854          * Delete all remaining routes using this interface
 855          */
 856         netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
 857             if_rtdel_dispatch);
 858         msg.ifp = ifp;
 859         rt_domsg_global(&msg.base);
 860
 861         /* Announce that the interface is gone. */
 862         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
 863         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
 864
 865         SLIST_FOREACH(dp, &domains, dom_next)
 866                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
 867                         (*dp->dom_ifdetach)(ifp,
 868                                 ifp->if_afdata[dp->dom_family]);
 869
 870         /*
 871          * Remove interface from ifindex2ifp[] and maybe decrement if_index.
 872          */
 873         lwkt_gettoken(&ifnet_token);
 874         ifindex2ifnet[ifp->if_index] = NULL;
 875         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
 876                 if_index--;
 877         TAILQ_REMOVE(&ifnet, ifp, if_link);
 878         lwkt_reltoken(&ifnet_token);
 879
 880         kfree(ifp->if_addrheads, M_IFADDR);
 881
 882         lwkt_synchronize_ipiqs("if_detach");
 883         ifq_stage_detach(&ifp->if_snd);
 884
 885         for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
 886                 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
 887
 888                 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
 889                 kfree(ifsq->ifsq_stage, M_DEVBUF);
 890         }
 891         kfree(ifp->if_snd.altq_subq, M_DEVBUF);
 892
 893         kfree(ifp->if_data_pcpu, M_DEVBUF);
 894
 895         crit_exit();
 896 }
 897
 898 /*
 899  * Create interface group without members
 900  */
 901 struct ifg_group *
 902 if_creategroup(const char *groupname)
 903 {
 904         struct ifg_group        *ifg = NULL;
 905
 906         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
 907             M_TEMP, M_NOWAIT)) == NULL)
 908                 return (NULL);
 909
 910         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
 911         ifg->ifg_refcnt = 0;
 912         ifg->ifg_carp_demoted = 0;
 913         TAILQ_INIT(&ifg->ifg_members);
 914 #if NPF > 0
 915         pfi_attach_ifgroup(ifg);
 916 #endif
 917         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
 918
 919         return (ifg);
 920 }
 921
 922 /*
 923  * Add a group to an interface
 924  */
 925 int
 926 if_addgroup(struct ifnet *ifp, const char *groupname)
 927 {
 928         struct ifg_list         *ifgl;
 929         struct ifg_group        *ifg = NULL;
 930         struct ifg_member       *ifgm;
 931
 932         if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
 933             groupname[strlen(groupname) - 1] <= '9')
 934                 return (EINVAL);
 935
 936         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
 937                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
 938                         return (EEXIST);
 939
 940         if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
 941                 return (ENOMEM);
 942
 943         if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
 944                 kfree(ifgl, M_TEMP);
 945                 return (ENOMEM);
 946         }
 947
 948         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
 949                 if (!strcmp(ifg->ifg_group, groupname))
 950                         break;
 951
 952         if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
 953                 kfree(ifgl, M_TEMP);
 954                 kfree(ifgm, M_TEMP);
 955                 return (ENOMEM);
 956         }
 957
 958         ifg->ifg_refcnt++;
 959         ifgl->ifgl_group = ifg;
 960         ifgm->ifgm_ifp = ifp;
 961
 962         TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
 963         TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
 964
 965 #if NPF > 0
 966         pfi_group_change(groupname);
 967 #endif
 968
 969         return (0);
 970 }
 971
 972 /*
 973  * Remove a group from an interface
 974  */
 975 int
 976 if_delgroup(struct ifnet *ifp, const char *groupname)
 977 {
 978         struct ifg_list         *ifgl;
 979         struct ifg_member       *ifgm;
 980
 981         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
 982                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
 983                         break;
 984         if (ifgl == NULL)
 985                 return (ENOENT);
 986
 987         TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
 988
 989         TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
 990                 if (ifgm->ifgm_ifp == ifp)
 991                         break;
 992
 993         if (ifgm != NULL) {
 994                 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
 995                 kfree(ifgm, M_TEMP);
 996         }
 997
 998         if (--ifgl->ifgl_group->ifg_refcnt == 0) {
 999                 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
1000 #if NPF > 0
1001                 pfi_detach_ifgroup(ifgl->ifgl_group);
1002 #endif
1003                 kfree(ifgl->ifgl_group, M_TEMP);
1004         }
1005
1006         kfree(ifgl, M_TEMP);
1007
1008 #if NPF > 0
1009         pfi_group_change(groupname);
1010 #endif
1011
1012         return (0);
1013 }
1014
1015 /*
1016  * Stores all groups from an interface in memory pointed
1017  * to by data
1018  */
1019 int
1020 if_getgroup(caddr_t data, struct ifnet *ifp)
1021 {
1022         int                      len, error;
1023         struct ifg_list         *ifgl;
1024         struct ifg_req           ifgrq, *ifgp;
1025         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1026
1027         if (ifgr->ifgr_len == 0) {
1028                 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1029                         ifgr->ifgr_len += sizeof(struct ifg_req);
1030                 return (0);
1031         }
1032
1033         len = ifgr->ifgr_len;
1034         ifgp = ifgr->ifgr_groups;
1035         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1036                 if (len < sizeof(ifgrq))
1037                         return (EINVAL);
1038                 bzero(&ifgrq, sizeof ifgrq);
1039                 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1040                     sizeof(ifgrq.ifgrq_group));
1041                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1042                     sizeof(struct ifg_req))))
1043                         return (error);
1044                 len -= sizeof(ifgrq);
1045                 ifgp++;
1046         }
1047
1048         return (0);
1049 }
1050
1051 /*
1052  * Stores all members of a group in memory pointed to by data
1053  */
1054 int
1055 if_getgroupmembers(caddr_t data)
1056 {
1057         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1058         struct ifg_group        *ifg;
1059         struct ifg_member       *ifgm;
1060         struct ifg_req           ifgrq, *ifgp;
1061         int                      len, error;
1062
1063         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1064                 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1065                         break;
1066         if (ifg == NULL)
1067                 return (ENOENT);
1068
1069         if (ifgr->ifgr_len == 0) {
1070                 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1071                         ifgr->ifgr_len += sizeof(ifgrq);
1072                 return (0);
1073         }
1074
1075         len = ifgr->ifgr_len;
1076         ifgp = ifgr->ifgr_groups;
1077         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1078                 if (len < sizeof(ifgrq))
1079                         return (EINVAL);
1080                 bzero(&ifgrq, sizeof ifgrq);
1081                 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1082                     sizeof(ifgrq.ifgrq_member));
1083                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1084                     sizeof(struct ifg_req))))
1085                         return (error);
1086                 len -= sizeof(ifgrq);
1087                 ifgp++;
1088         }
1089
1090         return (0);
1091 }
1092
1093 /*
1094  * Delete Routes for a Network Interface
1095  *
1096  * Called for each routing entry via the rnh->rnh_walktree() call above
1097  * to delete all route entries referencing a detaching network interface.
1098  *
1099  * Arguments:
1100  *      rn      pointer to node in the routing table
1101  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1102  *
1103  * Returns:
1104  *      0       successful
1105  *      errno   failed - reason indicated
1106  *
1107  */
1108 static int
1109 if_rtdel(struct radix_node *rn, void *arg)
1110 {
1111         struct rtentry  *rt = (struct rtentry *)rn;
1112         struct ifnet    *ifp = arg;
1113         int             err;
1114
1115         if (rt->rt_ifp == ifp) {
1116
1117                 /*
1118                  * Protect (sorta) against walktree recursion problems
1119                  * with cloned routes
1120                  */
1121                 if (!(rt->rt_flags & RTF_UP))
1122                         return (0);
1123
1124                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1125                                 rt_mask(rt), rt->rt_flags,
1126                                 NULL);
1127                 if (err) {
1128                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1129                 }
1130         }
1131
1132         return (0);
1133 }
1134
1135 /*
1136  * Locate an interface based on a complete address.
1137  */
1138 struct ifaddr *
1139 ifa_ifwithaddr(struct sockaddr *addr)
1140 {
1141         struct ifnet *ifp;
1142
1143         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1144                 struct ifaddr_container *ifac;
1145
1146                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1147                         struct ifaddr *ifa = ifac->ifa;
1148
1149                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1150                                 continue;
1151                         if (sa_equal(addr, ifa->ifa_addr))
1152                                 return (ifa);
1153                         if ((ifp->if_flags & IFF_BROADCAST) &&
1154                             ifa->ifa_broadaddr &&
1155                             /* IPv6 doesn't have broadcast */
1156                             ifa->ifa_broadaddr->sa_len != 0 &&
1157                             sa_equal(ifa->ifa_broadaddr, addr))
1158                                 return (ifa);
1159                 }
1160         }
1161         return (NULL);
1162 }
1163 /*
1164  * Locate the point to point interface with a given destination address.
1165  */
1166 struct ifaddr *
1167 ifa_ifwithdstaddr(struct sockaddr *addr)
1168 {
1169         struct ifnet *ifp;
1170
1171         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1172                 struct ifaddr_container *ifac;
1173
1174                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1175                         continue;
1176
1177                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1178                         struct ifaddr *ifa = ifac->ifa;
1179
1180                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1181                                 continue;
1182                         if (ifa->ifa_dstaddr &&
1183                             sa_equal(addr, ifa->ifa_dstaddr))
1184                                 return (ifa);
1185                 }
1186         }
1187         return (NULL);
1188 }
1189
1190 /*
1191  * Find an interface on a specific network.  If many, choice
1192  * is most specific found.
1193  */
1194 struct ifaddr *
1195 ifa_ifwithnet(struct sockaddr *addr)
1196 {
1197         struct ifnet *ifp;
1198         struct ifaddr *ifa_maybe = NULL;
1199         u_int af = addr->sa_family;
1200         char *addr_data = addr->sa_data, *cplim;
1201
1202         /*
1203          * AF_LINK addresses can be looked up directly by their index number,
1204          * so do that if we can.
1205          */
1206         if (af == AF_LINK) {
1207                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1208
1209                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1210                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1211         }
1212
1213         /*
1214          * Scan though each interface, looking for ones that have
1215          * addresses in this address family.
1216          */
1217         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1218                 struct ifaddr_container *ifac;
1219
1220                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1221                         struct ifaddr *ifa = ifac->ifa;
1222                         char *cp, *cp2, *cp3;
1223
1224                         if (ifa->ifa_addr->sa_family != af)
1225 next:                           continue;
1226                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1227                                 /*
1228                                  * This is a bit broken as it doesn't
1229                                  * take into account that the remote end may
1230                                  * be a single node in the network we are
1231                                  * looking for.
1232                                  * The trouble is that we don't know the
1233                                  * netmask for the remote end.
1234                                  */
1235                                 if (ifa->ifa_dstaddr != NULL &&
1236                                     sa_equal(addr, ifa->ifa_dstaddr))
1237                                         return (ifa);
1238                         } else {
1239                                 /*
1240                                  * if we have a special address handler,
1241                                  * then use it instead of the generic one.
1242                                  */
1243                                 if (ifa->ifa_claim_addr) {
1244                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1245                                                 return (ifa);
1246                                         } else {
1247                                                 continue;
1248                                         }
1249                                 }
1250
1251                                 /*
1252                                  * Scan all the bits in the ifa's address.
1253                                  * If a bit dissagrees with what we are
1254                                  * looking for, mask it with the netmask
1255                                  * to see if it really matters.
1256                                  * (A byte at a time)
1257                                  */
1258                                 if (ifa->ifa_netmask == 0)
1259                                         continue;
1260                                 cp = addr_data;
1261                                 cp2 = ifa->ifa_addr->sa_data;
1262                                 cp3 = ifa->ifa_netmask->sa_data;
1263                                 cplim = ifa->ifa_netmask->sa_len +
1264                                         (char *)ifa->ifa_netmask;
1265                                 while (cp3 < cplim)
1266                                         if ((*cp++ ^ *cp2++) & *cp3++)
1267                                                 goto next; /* next address! */
1268                                 /*
1269                                  * If the netmask of what we just found
1270                                  * is more specific than what we had before
1271                                  * (if we had one) then remember the new one
1272                                  * before continuing to search
1273                                  * for an even better one.
1274                                  */
1275                                 if (ifa_maybe == NULL ||
1276                                     rn_refines((char *)ifa->ifa_netmask,
1277                                                (char *)ifa_maybe->ifa_netmask))
1278                                         ifa_maybe = ifa;
1279                         }
1280                 }
1281         }
1282         return (ifa_maybe);
1283 }
1284
1285 /*
1286  * Find an interface address specific to an interface best matching
1287  * a given address.
1288  */
1289 struct ifaddr *
1290 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1291 {
1292         struct ifaddr_container *ifac;
1293         char *cp, *cp2, *cp3;
1294         char *cplim;
1295         struct ifaddr *ifa_maybe = NULL;
1296         u_int af = addr->sa_family;
1297
1298         if (af >= AF_MAX)
1299                 return (0);
1300         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1301                 struct ifaddr *ifa = ifac->ifa;
1302
1303                 if (ifa->ifa_addr->sa_family != af)
1304                         continue;
1305                 if (ifa_maybe == NULL)
1306                         ifa_maybe = ifa;
1307                 if (ifa->ifa_netmask == NULL) {
1308                         if (sa_equal(addr, ifa->ifa_addr) ||
1309                             (ifa->ifa_dstaddr != NULL &&
1310                              sa_equal(addr, ifa->ifa_dstaddr)))
1311                                 return (ifa);
1312                         continue;
1313                 }
1314                 if (ifp->if_flags & IFF_POINTOPOINT) {
1315                         if (sa_equal(addr, ifa->ifa_dstaddr))
1316                                 return (ifa);
1317                 } else {
1318                         cp = addr->sa_data;
1319                         cp2 = ifa->ifa_addr->sa_data;
1320                         cp3 = ifa->ifa_netmask->sa_data;
1321                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1322                         for (; cp3 < cplim; cp3++)
1323                                 if ((*cp++ ^ *cp2++) & *cp3)
1324                                         break;
1325                         if (cp3 == cplim)
1326                                 return (ifa);
1327                 }
1328         }
1329         return (ifa_maybe);
1330 }
1331
1332 /*
1333  * Default action when installing a route with a Link Level gateway.
1334  * Lookup an appropriate real ifa to point to.
1335  * This should be moved to /sys/net/link.c eventually.
1336  */
1337 static void
1338 link_rtrequest(int cmd, struct rtentry *rt)
1339 {
1340         struct ifaddr *ifa;
1341         struct sockaddr *dst;
1342         struct ifnet *ifp;
1343
1344         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1345             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1346                 return;
1347         ifa = ifaof_ifpforaddr(dst, ifp);
1348         if (ifa != NULL) {
1349                 IFAFREE(rt->rt_ifa);
1350                 IFAREF(ifa);
1351                 rt->rt_ifa = ifa;
1352                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1353                         ifa->ifa_rtrequest(cmd, rt);
1354         }
1355 }
1356
1357 /*
1358  * Mark an interface down and notify protocols of
1359  * the transition.
1360  * NOTE: must be called at splnet or eqivalent.
1361  */
1362 void
1363 if_unroute(struct ifnet *ifp, int flag, int fam)
1364 {
1365         struct ifaddr_container *ifac;
1366
1367         ifp->if_flags &= ~flag;
1368         getmicrotime(&ifp->if_lastchange);
1369         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1370                 struct ifaddr *ifa = ifac->ifa;
1371
1372                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1373                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1374         }
1375         ifq_purge_all(&ifp->if_snd);
1376         rt_ifmsg(ifp);
1377 }
1378
1379 /*
1380  * Mark an interface up and notify protocols of
1381  * the transition.
1382  * NOTE: must be called at splnet or eqivalent.
1383  */
1384 void
1385 if_route(struct ifnet *ifp, int flag, int fam)
1386 {
1387         struct ifaddr_container *ifac;
1388
1389         ifq_purge_all(&ifp->if_snd);
1390         ifp->if_flags |= flag;
1391         getmicrotime(&ifp->if_lastchange);
1392         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1393                 struct ifaddr *ifa = ifac->ifa;
1394
1395                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1396                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1397         }
1398         rt_ifmsg(ifp);
1399 #ifdef INET6
1400         in6_if_up(ifp);
1401 #endif
1402 }
1403
1404 /*
1405  * Mark an interface down and notify protocols of the transition.  An
1406  * interface going down is also considered to be a synchronizing event.
1407  * We must ensure that all packet processing related to the interface
1408  * has completed before we return so e.g. the caller can free the ifnet
1409  * structure that the mbufs may be referencing.
1410  *
1411  * NOTE: must be called at splnet or eqivalent.
1412  */
1413 void
1414 if_down(struct ifnet *ifp)
1415 {
1416         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1417         netmsg_service_sync();
1418 }
1419
1420 /*
1421  * Mark an interface up and notify protocols of
1422  * the transition.
1423  * NOTE: must be called at splnet or eqivalent.
1424  */
1425 void
1426 if_up(struct ifnet *ifp)
1427 {
1428         if_route(ifp, IFF_UP, AF_UNSPEC);
1429 }
1430
1431 /*
1432  * Process a link state change.
1433  * NOTE: must be called at splsoftnet or equivalent.
1434  */
1435 void
1436 if_link_state_change(struct ifnet *ifp)
1437 {
1438         int link_state = ifp->if_link_state;
1439
1440         rt_ifmsg(ifp);
1441         devctl_notify("IFNET", ifp->if_xname,
1442             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1443 }
1444
1445 /*
1446  * Handle interface watchdog timer routines.  Called
1447  * from softclock, we decrement timers (if set) and
1448  * call the appropriate interface routine on expiration.
1449  */
1450 static void
1451 if_slowtimo(void *arg)
1452 {
1453         struct ifnet *ifp;
1454
1455         crit_enter();
1456
1457         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1458                 if (if_stats_compat) {
1459                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1460                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1461                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1462                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1463                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1464                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1465                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1466                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1467                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1468                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1469                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1470                 }
1471
1472                 if (ifp->if_timer == 0 || --ifp->if_timer)
1473                         continue;
1474                 if (ifp->if_watchdog) {
1475                         if (ifnet_tryserialize_all(ifp)) {
1476                                 (*ifp->if_watchdog)(ifp);
1477                                 ifnet_deserialize_all(ifp);
1478                         } else {
1479                                 /* try again next timeout */
1480                                 ++ifp->if_timer;
1481                         }
1482                 }
1483         }
1484
1485         crit_exit();
1486
1487         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1488 }
1489
1490 /*
1491  * Map interface name to
1492  * interface structure pointer.
1493  */
1494 struct ifnet *
1495 ifunit(const char *name)
1496 {
1497         struct ifnet *ifp;
1498
1499         /*
1500          * Search all the interfaces for this name/number
1501          */
1502
1503         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1504                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1505                         break;
1506         }
1507         return (ifp);
1508 }
1509
1510
1511 /*
1512  * Map interface name in a sockaddr_dl to
1513  * interface structure pointer.
1514  */
1515 struct ifnet *
1516 if_withname(struct sockaddr *sa)
1517 {
1518         char ifname[IFNAMSIZ+1];
1519         struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1520
1521         if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1522              (sdl->sdl_nlen > IFNAMSIZ) )
1523                 return NULL;
1524
1525         /*
1526          * ifunit wants a null-terminated name.  It may not be null-terminated
1527          * in the sockaddr.  We don't want to change the caller's sockaddr,
1528          * and there might not be room to put the trailing null anyway, so we
1529          * make a local copy that we know we can null terminate safely.
1530          */
1531
1532         bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1533         ifname[sdl->sdl_nlen] = '\0';
1534         return ifunit(ifname);
1535 }
1536
1537
1538 /*
1539  * Interface ioctls.
1540  */
1541 int
1542 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1543 {
1544         struct ifnet *ifp;
1545         struct ifreq *ifr;
1546         struct ifstat *ifs;
1547         int error;
1548         short oif_flags;
1549         int new_flags;
1550 #ifdef COMPAT_43
1551         int ocmd;
1552 #endif
1553         size_t namelen, onamelen;
1554         char new_name[IFNAMSIZ];
1555         struct ifaddr *ifa;
1556         struct sockaddr_dl *sdl;
1557
1558         switch (cmd) {
1559         case SIOCGIFCONF:
1560         case OSIOCGIFCONF:
1561                 return (ifconf(cmd, data, cred));
1562         default:
1563                 break;
1564         }
1565
1566         ifr = (struct ifreq *)data;
1567
1568         switch (cmd) {
1569         case SIOCIFCREATE:
1570         case SIOCIFCREATE2:
1571                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1572                         return (error);
1573                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1574                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1575         case SIOCIFDESTROY:
1576                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1577                         return (error);
1578                 return (if_clone_destroy(ifr->ifr_name));
1579         case SIOCIFGCLONERS:
1580                 return (if_clone_list((struct if_clonereq *)data));
1581         default:
1582                 break;
1583         }
1584
1585         /*
1586          * Nominal ioctl through interface, lookup the ifp and obtain a
1587          * lock to serialize the ifconfig ioctl operation.
1588          */
1589         ifp = ifunit(ifr->ifr_name);
1590         if (ifp == NULL)
1591                 return (ENXIO);
1592         error = 0;
1593         mtx_lock(&ifp->if_ioctl_mtx);
1594
1595         switch (cmd) {
1596         case SIOCGIFINDEX:
1597                 ifr->ifr_index = ifp->if_index;
1598                 break;
1599
1600         case SIOCGIFFLAGS:
1601                 ifr->ifr_flags = ifp->if_flags;
1602                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1603                 break;
1604
1605         case SIOCGIFCAP:
1606                 ifr->ifr_reqcap = ifp->if_capabilities;
1607                 ifr->ifr_curcap = ifp->if_capenable;
1608                 break;
1609
1610         case SIOCGIFMETRIC:
1611                 ifr->ifr_metric = ifp->if_metric;
1612                 break;
1613
1614         case SIOCGIFMTU:
1615                 ifr->ifr_mtu = ifp->if_mtu;
1616                 break;
1617
1618         case SIOCGIFTSOLEN:
1619                 ifr->ifr_tsolen = ifp->if_tsolen;
1620                 break;
1621
1622         case SIOCGIFDATA:
1623                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1624                                 sizeof(ifp->if_data));
1625                 break;
1626
1627         case SIOCGIFPHYS:
1628                 ifr->ifr_phys = ifp->if_physical;
1629                 break;
1630
1631         case SIOCGIFPOLLCPU:
1632                 ifr->ifr_pollcpu = -1;
1633                 break;
1634
1635         case SIOCSIFPOLLCPU:
1636                 break;
1637
1638         case SIOCSIFFLAGS:
1639                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1640                 if (error)
1641                         break;
1642                 new_flags = (ifr->ifr_flags & 0xffff) |
1643                     (ifr->ifr_flagshigh << 16);
1644                 if (ifp->if_flags & IFF_SMART) {
1645                         /* Smart drivers twiddle their own routes */
1646                 } else if (ifp->if_flags & IFF_UP &&
1647                     (new_flags & IFF_UP) == 0) {
1648                         crit_enter();
1649                         if_down(ifp);
1650                         crit_exit();
1651                 } else if (new_flags & IFF_UP &&
1652                     (ifp->if_flags & IFF_UP) == 0) {
1653                         crit_enter();
1654                         if_up(ifp);
1655                         crit_exit();
1656                 }
1657
1658 #ifdef IFPOLL_ENABLE
1659                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1660                         if (new_flags & IFF_NPOLLING)
1661                                 ifpoll_register(ifp);
1662                         else
1663                                 ifpoll_deregister(ifp);
1664                 }
1665 #endif
1666
1667                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1668                         (new_flags &~ IFF_CANTCHANGE);
1669                 if (new_flags & IFF_PPROMISC) {
1670                         /* Permanently promiscuous mode requested */
1671                         ifp->if_flags |= IFF_PROMISC;
1672                 } else if (ifp->if_pcount == 0) {
1673                         ifp->if_flags &= ~IFF_PROMISC;
1674                 }
1675                 if (ifp->if_ioctl) {
1676                         ifnet_serialize_all(ifp);
1677                         ifp->if_ioctl(ifp, cmd, data, cred);
1678                         ifnet_deserialize_all(ifp);
1679                 }
1680                 getmicrotime(&ifp->if_lastchange);
1681                 break;
1682
1683         case SIOCSIFCAP:
1684                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1685                 if (error)
1686                         break;
1687                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1688                         error = EINVAL;
1689                         break;
1690                 }
1691                 ifnet_serialize_all(ifp);
1692                 ifp->if_ioctl(ifp, cmd, data, cred);
1693                 ifnet_deserialize_all(ifp);
1694                 break;
1695
1696         case SIOCSIFNAME:
1697                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1698                 if (error)
1699                         break;
1700                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1701                 if (error)
1702                         break;
1703                 if (new_name[0] == '\0') {
1704                         error = EINVAL;
1705                         break;
1706                 }
1707                 if (ifunit(new_name) != NULL) {
1708                         error = EEXIST;
1709                         break;
1710                 }
1711
1712                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1713
1714                 /* Announce the departure of the interface. */
1715                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1716
1717                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1718                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1719                 /* XXX IFA_LOCK(ifa); */
1720                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1721                 namelen = strlen(new_name);
1722                 onamelen = sdl->sdl_nlen;
1723                 /*
1724                  * Move the address if needed.  This is safe because we
1725                  * allocate space for a name of length IFNAMSIZ when we
1726                  * create this in if_attach().
1727                  */
1728                 if (namelen != onamelen) {
1729                         bcopy(sdl->sdl_data + onamelen,
1730                             sdl->sdl_data + namelen, sdl->sdl_alen);
1731                 }
1732                 bcopy(new_name, sdl->sdl_data, namelen);
1733                 sdl->sdl_nlen = namelen;
1734                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1735                 bzero(sdl->sdl_data, onamelen);
1736                 while (namelen != 0)
1737                         sdl->sdl_data[--namelen] = 0xff;
1738                 /* XXX IFA_UNLOCK(ifa) */
1739
1740                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1741
1742                 /* Announce the return of the interface. */
1743                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1744                 break;
1745
1746         case SIOCSIFMETRIC:
1747                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1748                 if (error)
1749                         break;
1750                 ifp->if_metric = ifr->ifr_metric;
1751                 getmicrotime(&ifp->if_lastchange);
1752                 break;
1753
1754         case SIOCSIFPHYS:
1755                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1756                 if (error)
1757                         break;
1758                 if (ifp->if_ioctl == NULL) {
1759                         error = EOPNOTSUPP;
1760                         break;
1761                 }
1762                 ifnet_serialize_all(ifp);
1763                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1764                 ifnet_deserialize_all(ifp);
1765                 if (error == 0)
1766                         getmicrotime(&ifp->if_lastchange);
1767                 break;
1768
1769         case SIOCSIFMTU:
1770         {
1771                 u_long oldmtu = ifp->if_mtu;
1772
1773                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1774                 if (error)
1775                         break;
1776                 if (ifp->if_ioctl == NULL) {
1777                         error = EOPNOTSUPP;
1778                         break;
1779                 }
1780                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
1781                         error = EINVAL;
1782                         break;
1783                 }
1784                 ifnet_serialize_all(ifp);
1785                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1786                 ifnet_deserialize_all(ifp);
1787                 if (error == 0) {
1788                         getmicrotime(&ifp->if_lastchange);
1789                         rt_ifmsg(ifp);
1790                 }
1791                 /*
1792                  * If the link MTU changed, do network layer specific procedure.
1793                  */
1794                 if (ifp->if_mtu != oldmtu) {
1795 #ifdef INET6
1796                         nd6_setmtu(ifp);
1797 #endif
1798                 }
1799                 break;
1800         }
1801
1802         case SIOCSIFTSOLEN:
1803                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1804                 if (error)
1805                         break;
1806
1807                 /* XXX need driver supplied upper limit */
1808                 if (ifr->ifr_tsolen <= 0) {
1809                         error = EINVAL;
1810                         break;
1811                 }
1812                 ifp->if_tsolen = ifr->ifr_tsolen;
1813                 break;
1814
1815         case SIOCADDMULTI:
1816         case SIOCDELMULTI:
1817                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1818                 if (error)
1819                         break;
1820
1821                 /* Don't allow group membership on non-multicast interfaces. */
1822                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1823                         error = EOPNOTSUPP;
1824                         break;
1825                 }
1826
1827                 /* Don't let users screw up protocols' entries. */
1828                 if (ifr->ifr_addr.sa_family != AF_LINK) {
1829                         error = EINVAL;
1830                         break;
1831                 }
1832
1833                 if (cmd == SIOCADDMULTI) {
1834                         struct ifmultiaddr *ifma;
1835                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1836                 } else {
1837                         error = if_delmulti(ifp, &ifr->ifr_addr);
1838                 }
1839                 if (error == 0)
1840                         getmicrotime(&ifp->if_lastchange);
1841                 break;
1842
1843         case SIOCSIFPHYADDR:
1844         case SIOCDIFPHYADDR:
1845 #ifdef INET6
1846         case SIOCSIFPHYADDR_IN6:
1847 #endif
1848         case SIOCSLIFPHYADDR:
1849         case SIOCSIFMEDIA:
1850         case SIOCSIFGENERIC:
1851                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1852                 if (error)
1853                         break;
1854                 if (ifp->if_ioctl == 0) {
1855                         error = EOPNOTSUPP;
1856                         break;
1857                 }
1858                 ifnet_serialize_all(ifp);
1859                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1860                 ifnet_deserialize_all(ifp);
1861                 if (error == 0)
1862                         getmicrotime(&ifp->if_lastchange);
1863                 break;
1864
1865         case SIOCGIFSTATUS:
1866                 ifs = (struct ifstat *)data;
1867                 ifs->ascii[0] = '\0';
1868                 /* fall through */
1869         case SIOCGIFPSRCADDR:
1870         case SIOCGIFPDSTADDR:
1871         case SIOCGLIFPHYADDR:
1872         case SIOCGIFMEDIA:
1873         case SIOCGIFGENERIC:
1874                 if (ifp->if_ioctl == NULL) {
1875                         error = EOPNOTSUPP;
1876                         break;
1877                 }
1878                 ifnet_serialize_all(ifp);
1879                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1880                 ifnet_deserialize_all(ifp);
1881                 break;
1882
1883         case SIOCSIFLLADDR:
1884                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1885                 if (error)
1886                         break;
1887                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
1888                                      ifr->ifr_addr.sa_len);
1889                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1890                 break;
1891
1892         default:
1893                 oif_flags = ifp->if_flags;
1894                 if (so->so_proto == 0) {
1895                         error = EOPNOTSUPP;
1896                         break;
1897                 }
1898 #ifndef COMPAT_43
1899                 error = so_pru_control_direct(so, cmd, data, ifp);
1900 #else
1901                 ocmd = cmd;
1902
1903                 switch (cmd) {
1904                 case SIOCSIFDSTADDR:
1905                 case SIOCSIFADDR:
1906                 case SIOCSIFBRDADDR:
1907                 case SIOCSIFNETMASK:
1908 #if BYTE_ORDER != BIG_ENDIAN
1909                         if (ifr->ifr_addr.sa_family == 0 &&
1910                             ifr->ifr_addr.sa_len < 16) {
1911                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1912                                 ifr->ifr_addr.sa_len = 16;
1913                         }
1914 #else
1915                         if (ifr->ifr_addr.sa_len == 0)
1916                                 ifr->ifr_addr.sa_len = 16;
1917 #endif
1918                         break;
1919                 case OSIOCGIFADDR:
1920                         cmd = SIOCGIFADDR;
1921                         break;
1922                 case OSIOCGIFDSTADDR:
1923                         cmd = SIOCGIFDSTADDR;
1924                         break;
1925                 case OSIOCGIFBRDADDR:
1926                         cmd = SIOCGIFBRDADDR;
1927                         break;
1928                 case OSIOCGIFNETMASK:
1929                         cmd = SIOCGIFNETMASK;
1930                         break;
1931                 default:
1932                         break;
1933                 }
1934
1935                 error = so_pru_control_direct(so, cmd, data, ifp);
1936
1937                 switch (ocmd) {
1938                 case OSIOCGIFADDR:
1939                 case OSIOCGIFDSTADDR:
1940                 case OSIOCGIFBRDADDR:
1941                 case OSIOCGIFNETMASK:
1942                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1943                         break;
1944                 }
1945 #endif /* COMPAT_43 */
1946
1947                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1948 #ifdef INET6
1949                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
1950                         if (ifp->if_flags & IFF_UP) {
1951                                 crit_enter();
1952                                 in6_if_up(ifp);
1953                                 crit_exit();
1954                         }
1955 #endif
1956                 }
1957                 break;
1958         }
1959
1960         mtx_unlock(&ifp->if_ioctl_mtx);
1961         return (error);
1962 }
1963
1964 /*
1965  * Set/clear promiscuous mode on interface ifp based on the truth value
1966  * of pswitch.  The calls are reference counted so that only the first
1967  * "on" request actually has an effect, as does the final "off" request.
1968  * Results are undefined if the "off" and "on" requests are not matched.
1969  */
1970 int
1971 ifpromisc(struct ifnet *ifp, int pswitch)
1972 {
1973         struct ifreq ifr;
1974         int error;
1975         int oldflags;
1976
1977         oldflags = ifp->if_flags;
1978         if (ifp->if_flags & IFF_PPROMISC) {
1979                 /* Do nothing if device is in permanently promiscuous mode */
1980                 ifp->if_pcount += pswitch ? 1 : -1;
1981                 return (0);
1982         }
1983         if (pswitch) {
1984                 /*
1985                  * If the device is not configured up, we cannot put it in
1986                  * promiscuous mode.
1987                  */
1988                 if ((ifp->if_flags & IFF_UP) == 0)
1989                         return (ENETDOWN);
1990                 if (ifp->if_pcount++ != 0)
1991                         return (0);
1992                 ifp->if_flags |= IFF_PROMISC;
1993                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1994                     ifp->if_xname);
1995         } else {
1996                 if (--ifp->if_pcount > 0)
1997                         return (0);
1998                 ifp->if_flags &= ~IFF_PROMISC;
1999                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2000                     ifp->if_xname);
2001         }
2002         ifr.ifr_flags = ifp->if_flags;
2003         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2004         ifnet_serialize_all(ifp);
2005         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2006         ifnet_deserialize_all(ifp);
2007         if (error == 0)
2008                 rt_ifmsg(ifp);
2009         else
2010                 ifp->if_flags = oldflags;
2011         return error;
2012 }
2013
2014 /*
2015  * Return interface configuration
2016  * of system.  List may be used
2017  * in later ioctl's (above) to get
2018  * other information.
2019  */
2020 static int
2021 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2022 {
2023         struct ifconf *ifc = (struct ifconf *)data;
2024         struct ifnet *ifp;
2025         struct sockaddr *sa;
2026         struct ifreq ifr, *ifrp;
2027         int space = ifc->ifc_len, error = 0;
2028
2029         ifrp = ifc->ifc_req;
2030         TAILQ_FOREACH(ifp, &ifnet, if_link) {
2031                 struct ifaddr_container *ifac;
2032                 int addrs;
2033
2034                 if (space <= sizeof ifr)
2035                         break;
2036
2037                 /*
2038                  * Zero the stack declared structure first to prevent
2039                  * memory disclosure.
2040                  */
2041                 bzero(&ifr, sizeof(ifr));
2042                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2043                     >= sizeof(ifr.ifr_name)) {
2044                         error = ENAMETOOLONG;
2045                         break;
2046                 }
2047
2048                 addrs = 0;
2049                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2050                         struct ifaddr *ifa = ifac->ifa;
2051
2052                         if (space <= sizeof ifr)
2053                                 break;
2054                         sa = ifa->ifa_addr;
2055                         if (cred->cr_prison &&
2056                             prison_if(cred, sa))
2057                                 continue;
2058                         addrs++;
2059 #ifdef COMPAT_43
2060                         if (cmd == OSIOCGIFCONF) {
2061                                 struct osockaddr *osa =
2062                                          (struct osockaddr *)&ifr.ifr_addr;
2063                                 ifr.ifr_addr = *sa;
2064                                 osa->sa_family = sa->sa_family;
2065                                 error = copyout(&ifr, ifrp, sizeof ifr);
2066                                 ifrp++;
2067                         } else
2068 #endif
2069                         if (sa->sa_len <= sizeof(*sa)) {
2070                                 ifr.ifr_addr = *sa;
2071                                 error = copyout(&ifr, ifrp, sizeof ifr);
2072                                 ifrp++;
2073                         } else {
2074                                 if (space < (sizeof ifr) + sa->sa_len -
2075                                             sizeof(*sa))
2076                                         break;
2077                                 space -= sa->sa_len - sizeof(*sa);
2078                                 error = copyout(&ifr, ifrp,
2079                                                 sizeof ifr.ifr_name);
2080                                 if (error == 0)
2081                                         error = copyout(sa, &ifrp->ifr_addr,
2082                                                         sa->sa_len);
2083                                 ifrp = (struct ifreq *)
2084                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2085                         }
2086                         if (error)
2087                                 break;
2088                         space -= sizeof ifr;
2089                 }
2090                 if (error)
2091                         break;
2092                 if (!addrs) {
2093                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2094                         error = copyout(&ifr, ifrp, sizeof ifr);
2095                         if (error)
2096                                 break;
2097                         space -= sizeof ifr;
2098                         ifrp++;
2099                 }
2100         }
2101         ifc->ifc_len -= space;
2102         return (error);
2103 }
2104
2105 /*
2106  * Just like if_promisc(), but for all-multicast-reception mode.
2107  */
2108 int
2109 if_allmulti(struct ifnet *ifp, int onswitch)
2110 {
2111         int error = 0;
2112         struct ifreq ifr;
2113
2114         crit_enter();
2115
2116         if (onswitch) {
2117                 if (ifp->if_amcount++ == 0) {
2118                         ifp->if_flags |= IFF_ALLMULTI;
2119                         ifr.ifr_flags = ifp->if_flags;
2120                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2121                         ifnet_serialize_all(ifp);
2122                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2123                                               NULL);
2124                         ifnet_deserialize_all(ifp);
2125                 }
2126         } else {
2127                 if (ifp->if_amcount > 1) {
2128                         ifp->if_amcount--;
2129                 } else {
2130                         ifp->if_amcount = 0;
2131                         ifp->if_flags &= ~IFF_ALLMULTI;
2132                         ifr.ifr_flags = ifp->if_flags;
2133                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2134                         ifnet_serialize_all(ifp);
2135                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2136                                               NULL);
2137                         ifnet_deserialize_all(ifp);
2138                 }
2139         }
2140
2141         crit_exit();
2142
2143         if (error == 0)
2144                 rt_ifmsg(ifp);
2145         return error;
2146 }
2147
2148 /*
2149  * Add a multicast listenership to the interface in question.
2150  * The link layer provides a routine which converts
2151  */
2152 int
2153 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2154     struct ifmultiaddr **retifma)
2155 {
2156         struct sockaddr *llsa, *dupsa;
2157         int error;
2158         struct ifmultiaddr *ifma;
2159
2160         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2161
2162         /*
2163          * If the matching multicast address already exists
2164          * then don't add a new one, just add a reference
2165          */
2166         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2167                 if (sa_equal(sa, ifma->ifma_addr)) {
2168                         ifma->ifma_refcount++;
2169                         if (retifma)
2170                                 *retifma = ifma;
2171                         return 0;
2172                 }
2173         }
2174
2175         /*
2176          * Give the link layer a chance to accept/reject it, and also
2177          * find out which AF_LINK address this maps to, if it isn't one
2178          * already.
2179          */
2180         if (ifp->if_resolvemulti) {
2181                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2182                 if (error)
2183                         return error;
2184         } else {
2185                 llsa = NULL;
2186         }
2187
2188         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2189         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_WAITOK);
2190         bcopy(sa, dupsa, sa->sa_len);
2191
2192         ifma->ifma_addr = dupsa;
2193         ifma->ifma_lladdr = llsa;
2194         ifma->ifma_ifp = ifp;
2195         ifma->ifma_refcount = 1;
2196         ifma->ifma_protospec = NULL;
2197         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2198
2199         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2200         if (retifma)
2201                 *retifma = ifma;
2202
2203         if (llsa != NULL) {
2204                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2205                         if (sa_equal(ifma->ifma_addr, llsa))
2206                                 break;
2207                 }
2208                 if (ifma) {
2209                         ifma->ifma_refcount++;
2210                 } else {
2211                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2212                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_WAITOK);
2213                         bcopy(llsa, dupsa, llsa->sa_len);
2214                         ifma->ifma_addr = dupsa;
2215                         ifma->ifma_ifp = ifp;
2216                         ifma->ifma_refcount = 1;
2217                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2218                 }
2219         }
2220         /*
2221          * We are certain we have added something, so call down to the
2222          * interface to let them know about it.
2223          */
2224         if (ifp->if_ioctl)
2225                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2226
2227         return 0;
2228 }
2229
2230 int
2231 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2232     struct ifmultiaddr **retifma)
2233 {
2234         int error;
2235
2236         ifnet_serialize_all(ifp);
2237         error = if_addmulti_serialized(ifp, sa, retifma);
2238         ifnet_deserialize_all(ifp);
2239
2240         return error;
2241 }
2242
2243 /*
2244  * Remove a reference to a multicast address on this interface.  Yell
2245  * if the request does not match an existing membership.
2246  */
2247 static int
2248 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2249 {
2250         struct ifmultiaddr *ifma;
2251
2252         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2253
2254         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2255                 if (sa_equal(sa, ifma->ifma_addr))
2256                         break;
2257         if (ifma == NULL)
2258                 return ENOENT;
2259
2260         if (ifma->ifma_refcount > 1) {
2261                 ifma->ifma_refcount--;
2262                 return 0;
2263         }
2264
2265         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2266         sa = ifma->ifma_lladdr;
2267         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2268         /*
2269          * Make sure the interface driver is notified
2270          * in the case of a link layer mcast group being left.
2271          */
2272         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2273                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2274         kfree(ifma->ifma_addr, M_IFMADDR);
2275         kfree(ifma, M_IFMADDR);
2276         if (sa == NULL)
2277                 return 0;
2278
2279         /*
2280          * Now look for the link-layer address which corresponds to
2281          * this network address.  It had been squirreled away in
2282          * ifma->ifma_lladdr for this purpose (so we don't have
2283          * to call ifp->if_resolvemulti() again), and we saved that
2284          * value in sa above.  If some nasty deleted the
2285          * link-layer address out from underneath us, we can deal because
2286          * the address we stored was is not the same as the one which was
2287          * in the record for the link-layer address.  (So we don't complain
2288          * in that case.)
2289          */
2290         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2291                 if (sa_equal(sa, ifma->ifma_addr))
2292                         break;
2293         if (ifma == NULL)
2294                 return 0;
2295
2296         if (ifma->ifma_refcount > 1) {
2297                 ifma->ifma_refcount--;
2298                 return 0;
2299         }
2300
2301         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2302         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2303         kfree(ifma->ifma_addr, M_IFMADDR);
2304         kfree(sa, M_IFMADDR);
2305         kfree(ifma, M_IFMADDR);
2306
2307         return 0;
2308 }
2309
2310 int
2311 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2312 {
2313         int error;
2314
2315         ifnet_serialize_all(ifp);
2316         error = if_delmulti_serialized(ifp, sa);
2317         ifnet_deserialize_all(ifp);
2318
2319         return error;
2320 }
2321
2322 /*
2323  * Delete all multicast group membership for an interface.
2324  * Should be used to quickly flush all multicast filters.
2325  */
2326 void
2327 if_delallmulti_serialized(struct ifnet *ifp)
2328 {
2329         struct ifmultiaddr *ifma, mark;
2330         struct sockaddr sa;
2331
2332         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2333
2334         bzero(&sa, sizeof(sa));
2335         sa.sa_family = AF_UNSPEC;
2336         sa.sa_len = sizeof(sa);
2337
2338         bzero(&mark, sizeof(mark));
2339         mark.ifma_addr = &sa;
2340
2341         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2342         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2343                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2344                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2345                     ifma_link);
2346
2347                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2348                         continue;
2349
2350                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2351         }
2352         TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2353 }
2354
2355
2356 /*
2357  * Set the link layer address on an interface.
2358  *
2359  * At this time we only support certain types of interfaces,
2360  * and we don't allow the length of the address to change.
2361  */
2362 int
2363 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2364 {
2365         struct sockaddr_dl *sdl;
2366         struct ifreq ifr;
2367
2368         sdl = IF_LLSOCKADDR(ifp);
2369         if (sdl == NULL)
2370                 return (EINVAL);
2371         if (len != sdl->sdl_alen)       /* don't allow length to change */
2372                 return (EINVAL);
2373         switch (ifp->if_type) {
2374         case IFT_ETHER:                 /* these types use struct arpcom */
2375         case IFT_XETHER:
2376         case IFT_L2VLAN:
2377         case IFT_IEEE8023ADLAG:
2378                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2379                 bcopy(lladdr, LLADDR(sdl), len);
2380                 break;
2381         default:
2382                 return (ENODEV);
2383         }
2384         /*
2385          * If the interface is already up, we need
2386          * to re-init it in order to reprogram its
2387          * address filter.
2388          */
2389         ifnet_serialize_all(ifp);
2390         if ((ifp->if_flags & IFF_UP) != 0) {
2391 #ifdef INET
2392                 struct ifaddr_container *ifac;
2393 #endif
2394
2395                 ifp->if_flags &= ~IFF_UP;
2396                 ifr.ifr_flags = ifp->if_flags;
2397                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2398                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2399                               NULL);
2400                 ifp->if_flags |= IFF_UP;
2401                 ifr.ifr_flags = ifp->if_flags;
2402                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2403                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2404                                  NULL);
2405 #ifdef INET
2406                 /*
2407                  * Also send gratuitous ARPs to notify other nodes about
2408                  * the address change.
2409                  */
2410                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2411                         struct ifaddr *ifa = ifac->ifa;
2412
2413                         if (ifa->ifa_addr != NULL &&
2414                             ifa->ifa_addr->sa_family == AF_INET)
2415                                 arp_gratuitous(ifp, ifa);
2416                 }
2417 #endif
2418         }
2419         ifnet_deserialize_all(ifp);
2420         return (0);
2421 }
2422
2423 struct ifmultiaddr *
2424 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2425 {
2426         struct ifmultiaddr *ifma;
2427
2428         /* TODO: need ifnet_serialize_main */
2429         ifnet_serialize_all(ifp);
2430         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2431                 if (sa_equal(ifma->ifma_addr, sa))
2432                         break;
2433         ifnet_deserialize_all(ifp);
2434
2435         return ifma;
2436 }
2437
2438 /*
2439  * This function locates the first real ethernet MAC from a network
2440  * card and loads it into node, returning 0 on success or ENOENT if
2441  * no suitable interfaces were found.  It is used by the uuid code to
2442  * generate a unique 6-byte number.
2443  */
2444 int
2445 if_getanyethermac(uint16_t *node, int minlen)
2446 {
2447         struct ifnet *ifp;
2448         struct sockaddr_dl *sdl;
2449
2450         TAILQ_FOREACH(ifp, &ifnet, if_link) {
2451                 if (ifp->if_type != IFT_ETHER)
2452                         continue;
2453                 sdl = IF_LLSOCKADDR(ifp);
2454                 if (sdl->sdl_alen < minlen)
2455                         continue;
2456                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2457                       minlen);
2458                 return(0);
2459         }
2460         return (ENOENT);
2461 }
2462
2463 /*
2464  * The name argument must be a pointer to storage which will last as
2465  * long as the interface does.  For physical devices, the result of
2466  * device_get_name(dev) is a good choice and for pseudo-devices a
2467  * static string works well.
2468  */
2469 void
2470 if_initname(struct ifnet *ifp, const char *name, int unit)
2471 {
2472         ifp->if_dname = name;
2473         ifp->if_dunit = unit;
2474         if (unit != IF_DUNIT_NONE)
2475                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2476         else
2477                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2478 }
2479
2480 int
2481 if_printf(struct ifnet *ifp, const char *fmt, ...)
2482 {
2483         __va_list ap;
2484         int retval;
2485
2486         retval = kprintf("%s: ", ifp->if_xname);
2487         __va_start(ap, fmt);
2488         retval += kvprintf(fmt, ap);
2489         __va_end(ap);
2490         return (retval);
2491 }
2492
2493 struct ifnet *
2494 if_alloc(uint8_t type)
2495 {
2496         struct ifnet *ifp;
2497         size_t size;
2498
2499         /*
2500          * XXX temporary hack until arpcom is setup in if_l2com
2501          */
2502         if (type == IFT_ETHER)
2503                 size = sizeof(struct arpcom);
2504         else
2505                 size = sizeof(struct ifnet);
2506
2507         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2508
2509         ifp->if_type = type;
2510
2511         if (if_com_alloc[type] != NULL) {
2512                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2513                 if (ifp->if_l2com == NULL) {
2514                         kfree(ifp, M_IFNET);
2515                         return (NULL);
2516                 }
2517         }
2518         return (ifp);
2519 }
2520
2521 void
2522 if_free(struct ifnet *ifp)
2523 {
2524         kfree(ifp, M_IFNET);
2525 }
2526
2527 void
2528 ifq_set_classic(struct ifaltq *ifq)
2529 {
2530         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2531             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2532 }
2533
2534 void
2535 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2536     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2537 {
2538         int q;
2539
2540         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2541         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2542         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2543         KASSERT(request != NULL, ("request is not specified"));
2544
2545         ifq->altq_mapsubq = mapsubq;
2546         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2547                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2548
2549                 ifsq->ifsq_enqueue = enqueue;
2550                 ifsq->ifsq_dequeue = dequeue;
2551                 ifsq->ifsq_request = request;
2552         }
2553 }
2554
2555 static void
2556 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2557 {
2558         m->m_nextpkt = NULL;
2559         if (ifsq->ifsq_norm_tail == NULL)
2560                 ifsq->ifsq_norm_head = m;
2561         else
2562                 ifsq->ifsq_norm_tail->m_nextpkt = m;
2563         ifsq->ifsq_norm_tail = m;
2564         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2565 }
2566
2567 static void
2568 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2569 {
2570         m->m_nextpkt = NULL;
2571         if (ifsq->ifsq_prio_tail == NULL)
2572                 ifsq->ifsq_prio_head = m;
2573         else
2574                 ifsq->ifsq_prio_tail->m_nextpkt = m;
2575         ifsq->ifsq_prio_tail = m;
2576         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2577         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2578 }
2579
2580 static struct mbuf *
2581 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2582 {
2583         struct mbuf *m;
2584
2585         m = ifsq->ifsq_norm_head;
2586         if (m != NULL) {
2587                 if ((ifsq->ifsq_norm_head = m->m_nextpkt) == NULL)
2588                         ifsq->ifsq_norm_tail = NULL;
2589                 m->m_nextpkt = NULL;
2590                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2591         }
2592         return m;
2593 }
2594
2595 static struct mbuf *
2596 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2597 {
2598         struct mbuf *m;
2599
2600         m = ifsq->ifsq_prio_head;
2601         if (m != NULL) {
2602                 if ((ifsq->ifsq_prio_head = m->m_nextpkt) == NULL)
2603                         ifsq->ifsq_prio_tail = NULL;
2604                 m->m_nextpkt = NULL;
2605                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2606                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2607         }
2608         return m;
2609 }
2610
2611 int
2612 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2613     struct altq_pktattr *pa __unused)
2614 {
2615         M_ASSERTPKTHDR(m);
2616         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2617             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2618                 if ((m->m_flags & M_PRIO) &&
2619                     ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen / 2) &&
2620                     ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt / 2)) {
2621                         struct mbuf *m_drop;
2622
2623                         /*
2624                          * Perform drop-head on normal queue
2625                          */
2626                         m_drop = ifsq_norm_dequeue(ifsq);
2627                         if (m_drop != NULL) {
2628                                 m_freem(m_drop);
2629                                 ifsq_prio_enqueue(ifsq, m);
2630                                 return 0;
2631                         }
2632                         /* XXX nothing could be dropped? */
2633                 }
2634                 m_freem(m);
2635                 return ENOBUFS;
2636         } else {
2637                 if (m->m_flags & M_PRIO)
2638                         ifsq_prio_enqueue(ifsq, m);
2639                 else
2640                         ifsq_norm_enqueue(ifsq, m);
2641                 return 0;
2642         }
2643 }
2644
2645 struct mbuf *
2646 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2647 {
2648         struct mbuf *m;
2649
2650         switch (op) {
2651         case ALTDQ_POLL:
2652                 m = ifsq->ifsq_prio_head;
2653                 if (m == NULL)
2654                         m = ifsq->ifsq_norm_head;
2655                 break;
2656
2657         case ALTDQ_REMOVE:
2658                 m = ifsq_prio_dequeue(ifsq);
2659                 if (m == NULL)
2660                         m = ifsq_norm_dequeue(ifsq);
2661                 break;
2662
2663         default:
2664                 panic("unsupported ALTQ dequeue op: %d", op);
2665         }
2666         return m;
2667 }
2668
2669 int
2670 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2671 {
2672         switch (req) {
2673         case ALTRQ_PURGE:
2674                 for (;;) {
2675                         struct mbuf *m;
2676
2677                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2678                         if (m == NULL)
2679                                 break;
2680                         m_freem(m);
2681                 }
2682                 break;
2683
2684         default:
2685                 panic("unsupported ALTQ request: %d", req);
2686         }
2687         return 0;
2688 }
2689
2690 static void
2691 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2692 {
2693         struct ifnet *ifp = ifsq_get_ifp(ifsq);
2694         int running = 0, need_sched;
2695
2696         /*
2697          * Try to do direct ifnet.if_start on the subqueue first, if there is
2698          * contention on the subqueue hardware serializer, ifnet.if_start on
2699          * the subqueue will be scheduled on the subqueue owner CPU.
2700          */
2701         if (!ifsq_tryserialize_hw(ifsq)) {
2702                 /*
2703                  * Subqueue hardware serializer contention happened,
2704                  * ifnet.if_start on the subqueue is scheduled on
2705                  * the subqueue owner CPU, and we keep going.
2706                  */
2707                 ifsq_ifstart_schedule(ifsq, 1);
2708                 return;
2709         }
2710
2711         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2712                 ifp->if_start(ifp, ifsq);
2713                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2714                         running = 1;
2715         }
2716         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2717
2718         ifsq_deserialize_hw(ifsq);
2719
2720         if (need_sched) {
2721                 /*
2722                  * More data need to be transmitted, ifnet.if_start on the
2723                  * subqueue is scheduled on the subqueue owner CPU, and we
2724                  * keep going.
2725                  * NOTE: ifnet.if_start subqueue interlock is not released.
2726                  */
2727                 ifsq_ifstart_schedule(ifsq, force_sched);
2728         }
2729 }
2730
2731 /*
2732  * Subqeue packets staging mechanism:
2733  *
2734  * The packets enqueued into the subqueue are staged to a certain amount
2735  * before the ifnet.if_start on the subqueue is called.  In this way, the
2736  * driver could avoid writing to hardware registers upon every packet,
2737  * instead, hardware registers could be written when certain amount of
2738  * packets are put onto hardware TX ring.  The measurement on several modern
2739  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
2740  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
2741  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
2742  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
2743  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
2744  *
2745  * Subqueue packets staging is performed for two entry points into drivers'
2746  * transmission function:
2747  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
2748  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
2749  *
2750  * Subqueue packets staging will be stopped upon any of the following
2751  * conditions:
2752  * - If the count of packets enqueued on the current CPU is great than or
2753  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2754  * - If the total length of packets enqueued on the current CPU is great
2755  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
2756  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
2757  *   is usually less than hardware's MTU.
2758  * - ifsq_ifstart_schedule() is not pending on the current CPU and
2759  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
2760  *   released.
2761  * - The if_start_rollup(), which is registered as low priority netisr
2762  *   rollup function, is called; probably because no more work is pending
2763  *   for netisr.
2764  *
2765  * NOTE:
2766  * Currently subqueue packet staging is only performed in netisr threads.
2767  */
2768 int
2769 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2770 {
2771         struct ifaltq *ifq = &ifp->if_snd;
2772         struct ifaltq_subque *ifsq;
2773         int error, start = 0, len, mcast = 0, avoid_start = 0;
2774         struct ifsubq_stage_head *head = NULL;
2775         struct ifsubq_stage *stage = NULL;
2776         struct globaldata *gd = mycpu;
2777         struct thread *td = gd->gd_curthread;
2778
2779         crit_enter_quick(td);
2780
2781         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
2782         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
2783
2784         len = m->m_pkthdr.len;
2785         if (m->m_flags & M_MCAST)
2786                 mcast = 1;
2787
2788         if (td->td_type == TD_TYPE_NETISR) {
2789                 head = &ifsubq_stage_heads[mycpuid];
2790                 stage = ifsq_get_stage(ifsq, mycpuid);
2791
2792                 stage->stg_cnt++;
2793                 stage->stg_len += len;
2794                 if (stage->stg_cnt < ifsq_stage_cntmax &&
2795                     stage->stg_len < (ifp->if_mtu - max_protohdr))
2796                         avoid_start = 1;
2797         }
2798
2799         ALTQ_SQ_LOCK(ifsq);
2800         error = ifsq_enqueue_locked(ifsq, m, pa);
2801         if (error) {
2802                 if (!ifsq_data_ready(ifsq)) {
2803                         ALTQ_SQ_UNLOCK(ifsq);
2804                         crit_exit_quick(td);
2805                         return error;
2806                 }
2807                 avoid_start = 0;
2808         }
2809         if (!ifsq_is_started(ifsq)) {
2810                 if (avoid_start) {
2811                         ALTQ_SQ_UNLOCK(ifsq);
2812
2813                         KKASSERT(!error);
2814                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
2815                                 ifsq_stage_insert(head, stage);
2816
2817                         IFNET_STAT_INC(ifp, obytes, len);
2818                         if (mcast)
2819                                 IFNET_STAT_INC(ifp, omcasts, 1);
2820                         crit_exit_quick(td);
2821                         return error;
2822                 }
2823
2824                 /*
2825                  * Hold the subqueue interlock of ifnet.if_start
2826                  */
2827                 ifsq_set_started(ifsq);
2828                 start = 1;
2829         }
2830         ALTQ_SQ_UNLOCK(ifsq);
2831
2832         if (!error) {
2833                 IFNET_STAT_INC(ifp, obytes, len);
2834                 if (mcast)
2835                         IFNET_STAT_INC(ifp, omcasts, 1);
2836         }
2837
2838         if (stage != NULL) {
2839                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
2840                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
2841                         if (!avoid_start) {
2842                                 ifsq_stage_remove(head, stage);
2843                                 ifsq_ifstart_schedule(ifsq, 1);
2844                         }
2845                         crit_exit_quick(td);
2846                         return error;
2847                 }
2848
2849                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
2850                         ifsq_stage_remove(head, stage);
2851                 } else {
2852                         stage->stg_cnt = 0;
2853                         stage->stg_len = 0;
2854                 }
2855         }
2856
2857         if (!start) {
2858                 crit_exit_quick(td);
2859                 return error;
2860         }
2861
2862         ifsq_ifstart_try(ifsq, 0);
2863
2864         crit_exit_quick(td);
2865         return error;
2866 }
2867
2868 void *
2869 ifa_create(int size, int flags)
2870 {
2871         struct ifaddr *ifa;
2872         int i;
2873
2874         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
2875
2876         ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2877         if (ifa == NULL)
2878                 return NULL;
2879
2880         ifa->ifa_containers =
2881             kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
2882                 M_IFADDR, M_WAITOK | M_ZERO);
2883         ifa->ifa_ncnt = ncpus;
2884         for (i = 0; i < ncpus; ++i) {
2885                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2886
2887                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2888                 ifac->ifa = ifa;
2889                 ifac->ifa_refcnt = 1;
2890         }
2891 #ifdef IFADDR_DEBUG
2892         kprintf("alloc ifa %p %d\n", ifa, size);
2893 #endif
2894         return ifa;
2895 }
2896
2897 void
2898 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2899 {
2900         struct ifaddr *ifa = ifac->ifa;
2901
2902         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2903         KKASSERT(ifac->ifa_refcnt == 0);
2904         KASSERT(ifac->ifa_listmask == 0,
2905                 ("ifa is still on %#x lists", ifac->ifa_listmask));
2906
2907         ifac->ifa_magic = IFA_CONTAINER_DEAD;
2908
2909 #ifdef IFADDR_DEBUG_VERBOSE
2910         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2911 #endif
2912
2913         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2914                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
2915         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2916 #ifdef IFADDR_DEBUG
2917                 kprintf("free ifa %p\n", ifa);
2918 #endif
2919                 kfree(ifa->ifa_containers, M_IFADDR);
2920                 kfree(ifa, M_IFADDR);
2921         }
2922 }
2923
2924 static void
2925 ifa_iflink_dispatch(netmsg_t nmsg)
2926 {
2927         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2928         struct ifaddr *ifa = msg->ifa;
2929         struct ifnet *ifp = msg->ifp;
2930         int cpu = mycpuid;
2931         struct ifaddr_container *ifac;
2932
2933         crit_enter();
2934
2935         ifac = &ifa->ifa_containers[cpu];
2936         ASSERT_IFAC_VALID(ifac);
2937         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2938                 ("ifaddr is on if_addrheads"));
2939
2940         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2941         if (msg->tail)
2942                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2943         else
2944                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2945
2946         crit_exit();
2947
2948         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2949 }
2950
2951 void
2952 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2953 {
2954         struct netmsg_ifaddr msg;
2955
2956         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2957                     0, ifa_iflink_dispatch);
2958         msg.ifa = ifa;
2959         msg.ifp = ifp;
2960         msg.tail = tail;
2961
2962         ifa_domsg(&msg.base.lmsg, 0);
2963 }
2964
2965 static void
2966 ifa_ifunlink_dispatch(netmsg_t nmsg)
2967 {
2968         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2969         struct ifaddr *ifa = msg->ifa;
2970         struct ifnet *ifp = msg->ifp;
2971         int cpu = mycpuid;
2972         struct ifaddr_container *ifac;
2973
2974         crit_enter();
2975
2976         ifac = &ifa->ifa_containers[cpu];
2977         ASSERT_IFAC_VALID(ifac);
2978         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2979                 ("ifaddr is not on if_addrhead"));
2980
2981         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2982         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2983
2984         crit_exit();
2985
2986         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2987 }
2988
2989 void
2990 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2991 {
2992         struct netmsg_ifaddr msg;
2993
2994         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2995                     0, ifa_ifunlink_dispatch);
2996         msg.ifa = ifa;
2997         msg.ifp = ifp;
2998
2999         ifa_domsg(&msg.base.lmsg, 0);
3000 }
3001
3002 static void
3003 ifa_destroy_dispatch(netmsg_t nmsg)
3004 {
3005         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3006
3007         IFAFREE(msg->ifa);
3008         ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3009 }
3010
3011 void
3012 ifa_destroy(struct ifaddr *ifa)
3013 {
3014         struct netmsg_ifaddr msg;
3015
3016         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3017                     0, ifa_destroy_dispatch);
3018         msg.ifa = ifa;
3019
3020         ifa_domsg(&msg.base.lmsg, 0);
3021 }
3022
3023 struct lwkt_port *
3024 ifnet_portfn(int cpu)
3025 {
3026         return &ifnet_threads[cpu].td_msgport;
3027 }
3028
3029 void
3030 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
3031 {
3032         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
3033
3034         if (next_cpu < ncpus)
3035                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
3036         else
3037                 lwkt_replymsg(lmsg, 0);
3038 }
3039
3040 int
3041 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
3042 {
3043         KKASSERT(cpu < ncpus);
3044         return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
3045 }
3046
3047 void
3048 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
3049 {
3050         KKASSERT(cpu < ncpus);
3051         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
3052 }
3053
3054 /*
3055  * Generic netmsg service loop.  Some protocols may roll their own but all
3056  * must do the basic command dispatch function call done here.
3057  */
3058 static void
3059 ifnet_service_loop(void *arg __unused)
3060 {
3061         netmsg_t msg;
3062
3063         while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
3064                 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
3065                 msg->base.nm_dispatch(msg);
3066         }
3067 }
3068
3069 static void
3070 if_start_rollup(void)
3071 {
3072         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3073         struct ifsubq_stage *stage;
3074
3075         crit_enter();
3076
3077         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3078                 struct ifaltq_subque *ifsq = stage->stg_subq;
3079                 int is_sched = 0;
3080
3081                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3082                         is_sched = 1;
3083                 ifsq_stage_remove(head, stage);
3084
3085                 if (is_sched) {
3086                         ifsq_ifstart_schedule(ifsq, 1);
3087                 } else {
3088                         int start = 0;
3089
3090                         ALTQ_SQ_LOCK(ifsq);
3091                         if (!ifsq_is_started(ifsq)) {
3092                                 /*
3093                                  * Hold the subqueue interlock of
3094                                  * ifnet.if_start
3095                                  */
3096                                 ifsq_set_started(ifsq);
3097                                 start = 1;
3098                         }
3099                         ALTQ_SQ_UNLOCK(ifsq);
3100
3101                         if (start)
3102                                 ifsq_ifstart_try(ifsq, 1);
3103                 }
3104                 KKASSERT((stage->stg_flags &
3105                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3106         }
3107
3108         crit_exit();
3109 }
3110
3111 static void
3112 ifnetinit(void *dummy __unused)
3113 {
3114         int i;
3115
3116         for (i = 0; i < ncpus; ++i) {
3117                 struct thread *thr = &ifnet_threads[i];
3118
3119                 lwkt_create(ifnet_service_loop, NULL, NULL,
3120                             thr, TDF_NOSTART|TDF_FORCE_SPINPORT|TDF_FIXEDCPU,
3121                             i, "ifnet %d", i);
3122                 netmsg_service_port_init(&thr->td_msgport);
3123                 lwkt_schedule(thr);
3124         }
3125
3126         for (i = 0; i < ncpus; ++i)
3127                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3128         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3129 }
3130
3131 void
3132 if_register_com_alloc(u_char type,
3133     if_com_alloc_t *a, if_com_free_t *f)
3134 {
3135
3136         KASSERT(if_com_alloc[type] == NULL,
3137             ("if_register_com_alloc: %d already registered", type));
3138         KASSERT(if_com_free[type] == NULL,
3139             ("if_register_com_alloc: %d free already registered", type));
3140
3141         if_com_alloc[type] = a;
3142         if_com_free[type] = f;
3143 }
3144
3145 void
3146 if_deregister_com_alloc(u_char type)
3147 {
3148
3149         KASSERT(if_com_alloc[type] != NULL,
3150             ("if_deregister_com_alloc: %d not registered", type));
3151         KASSERT(if_com_free[type] != NULL,
3152             ("if_deregister_com_alloc: %d free not registered", type));
3153         if_com_alloc[type] = NULL;
3154         if_com_free[type] = NULL;
3155 }
3156
3157 int
3158 if_ring_count2(int cnt, int cnt_max)
3159 {
3160         int shift = 0;
3161
3162         KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3163             ("invalid ring count max %d", cnt_max));
3164
3165         if (cnt <= 0)
3166                 cnt = cnt_max;
3167         if (cnt > ncpus2)
3168                 cnt = ncpus2;
3169         if (cnt > cnt_max)
3170                 cnt = cnt_max;
3171
3172         while ((1 << (shift + 1)) <= cnt)
3173                 ++shift;
3174         cnt = 1 << shift;
3175
3176         KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3177             ("calculate cnt %d, ncpus2 %d, cnt max %d",
3178              cnt, ncpus2, cnt_max));
3179         return cnt;
3180 }
3181
3182 void
3183 ifq_set_maxlen(struct ifaltq *ifq, int len)
3184 {
3185         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3186 }
3187
3188 int
3189 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3190 {
3191         return ALTQ_SUBQ_INDEX_DEFAULT;
3192 }
3193
3194 int
3195 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3196 {
3197         return (cpuid & ifq->altq_subq_mask);
3198 }
3199
3200 static void
3201 ifsq_watchdog(void *arg)
3202 {
3203         struct ifsubq_watchdog *wd = arg;
3204         struct ifnet *ifp;
3205
3206         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3207                 goto done;
3208
3209         ifp = ifsq_get_ifp(wd->wd_subq);
3210         if (ifnet_tryserialize_all(ifp)) {
3211                 wd->wd_watchdog(wd->wd_subq);
3212                 ifnet_deserialize_all(ifp);
3213         } else {
3214                 /* try again next timeout */
3215                 wd->wd_timer = 1;
3216         }
3217 done:
3218         ifsq_watchdog_reset(wd);
3219 }
3220
3221 static void
3222 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3223 {
3224         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3225             ifsq_get_cpuid(wd->wd_subq));
3226 }
3227
3228 void
3229 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3230     ifsq_watchdog_t watchdog)
3231 {
3232         callout_init_mp(&wd->wd_callout);
3233         wd->wd_timer = 0;
3234         wd->wd_subq = ifsq;
3235         wd->wd_watchdog = watchdog;
3236 }
3237
3238 void
3239 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3240 {
3241         wd->wd_timer = 0;
3242         ifsq_watchdog_reset(wd);
3243 }
3244
3245 void
3246 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3247 {
3248         wd->wd_timer = 0;
3249         callout_stop(&wd->wd_callout);
3250 }