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_prefixhead);
 521         TAILQ_INIT(&ifp->if_multiaddrs);
 522         TAILQ_INIT(&ifp->if_groups);
 523         getmicrotime(&ifp->if_lastchange);
 524         if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
 525                 unsigned int n;
 526                 struct ifnet **q;
 527
 528                 if_indexlim <<= 1;
 529
 530                 /* grow ifindex2ifnet */
 531                 n = if_indexlim * sizeof(*q);
 532                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
 533                 if (ifindex2ifnet) {
 534                         bcopy(ifindex2ifnet, q, n/2);
 535                         kfree(ifindex2ifnet, M_IFADDR);
 536                 }
 537                 ifindex2ifnet = q;
 538         }
 539
 540         ifindex2ifnet[if_index] = ifp;
 541
 542         /*
 543          * create a Link Level name for this device
 544          */
 545         namelen = strlen(ifp->if_xname);
 546         masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
 547         socksize = masklen + ifp->if_addrlen;
 548 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
 549         if (socksize < sizeof(*sdl))
 550                 socksize = sizeof(*sdl);
 551         socksize = ROUNDUP(socksize);
 552 #undef ROUNDUP
 553         ifasize = sizeof(struct ifaddr) + 2 * socksize;
 554         ifa = ifa_create(ifasize, M_WAITOK);
 555         sdl = (struct sockaddr_dl *)(ifa + 1);
 556         sdl->sdl_len = socksize;
 557         sdl->sdl_family = AF_LINK;
 558         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
 559         sdl->sdl_nlen = namelen;
 560         sdl->sdl_index = ifp->if_index;
 561         sdl->sdl_type = ifp->if_type;
 562         ifp->if_lladdr = ifa;
 563         ifa->ifa_ifp = ifp;
 564         ifa->ifa_rtrequest = link_rtrequest;
 565         ifa->ifa_addr = (struct sockaddr *)sdl;
 566         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
 567         ifa->ifa_netmask = (struct sockaddr *)sdl;
 568         sdl->sdl_len = masklen;
 569         while (namelen != 0)
 570                 sdl->sdl_data[--namelen] = 0xff;
 571         ifa_iflink(ifa, ifp, 0 /* Insert head */);
 572
 573         ifp->if_data_pcpu = kmalloc_cachealign(
 574             ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
 575
 576         if (ifp->if_mapsubq == NULL)
 577                 ifp->if_mapsubq = ifq_mapsubq_default;
 578
 579         ifq = &ifp->if_snd;
 580         ifq->altq_type = 0;
 581         ifq->altq_disc = NULL;
 582         ifq->altq_flags &= ALTQF_CANTCHANGE;
 583         ifq->altq_tbr = NULL;
 584         ifq->altq_ifp = ifp;
 585
 586         if (ifq->altq_subq_cnt <= 0)
 587                 ifq->altq_subq_cnt = 1;
 588         ifq->altq_subq = kmalloc_cachealign(
 589             ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
 590             M_DEVBUF, M_WAITOK | M_ZERO);
 591
 592         if (ifq->altq_maxlen == 0) {
 593                 if_printf(ifp, "driver didn't set altq_maxlen\n");
 594                 ifq_set_maxlen(ifq, ifqmaxlen);
 595         }
 596
 597         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 598                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 599
 600                 ALTQ_SQ_LOCK_INIT(ifsq);
 601                 ifsq->ifsq_index = q;
 602
 603                 ifsq->ifsq_altq = ifq;
 604                 ifsq->ifsq_ifp = ifp;
 605
 606                 ifsq->ifsq_maxlen = ifq->altq_maxlen;
 607                 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
 608                 ifsq->ifsq_prepended = NULL;
 609                 ifsq->ifsq_started = 0;
 610                 ifsq->ifsq_hw_oactive = 0;
 611                 ifsq_set_cpuid(ifsq, 0);
 612                 if (ifp->if_serializer != NULL)
 613                         ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
 614
 615                 ifsq->ifsq_stage =
 616                     kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
 617                     M_DEVBUF, M_WAITOK | M_ZERO);
 618                 for (i = 0; i < ncpus; ++i)
 619                         ifsq->ifsq_stage[i].stg_subq = ifsq;
 620
 621                 ifsq->ifsq_ifstart_nmsg =
 622                     kmalloc(ncpus * sizeof(struct netmsg_base),
 623                     M_LWKTMSG, M_WAITOK);
 624                 for (i = 0; i < ncpus; ++i) {
 625                         netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
 626                             &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
 627                         ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
 628                 }
 629         }
 630         ifq_set_classic(ifq);
 631
 632         if (!SLIST_EMPTY(&domains))
 633                 if_attachdomain1(ifp);
 634
 635         TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
 636         lwkt_reltoken(&ifnet_token);
 637
 638         /* Announce the interface. */
 639         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
 640         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
 641         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
 642
 643         mtx_unlock(&ifp->if_ioctl_mtx);
 644 }
 645
 646 static void
 647 if_attachdomain(void *dummy)
 648 {
 649         struct ifnet *ifp;
 650
 651         crit_enter();
 652         TAILQ_FOREACH(ifp, &ifnet, if_list)
 653                 if_attachdomain1(ifp);
 654         crit_exit();
 655 }
 656 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
 657         if_attachdomain, NULL);
 658
 659 static void
 660 if_attachdomain1(struct ifnet *ifp)
 661 {
 662         struct domain *dp;
 663
 664         crit_enter();
 665
 666         /* address family dependent data region */
 667         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
 668         SLIST_FOREACH(dp, &domains, dom_next)
 669                 if (dp->dom_ifattach)
 670                         ifp->if_afdata[dp->dom_family] =
 671                                 (*dp->dom_ifattach)(ifp);
 672         crit_exit();
 673 }
 674
 675 /*
 676  * Purge all addresses whose type is _not_ AF_LINK
 677  */
 678 void
 679 if_purgeaddrs_nolink(struct ifnet *ifp)
 680 {
 681         struct ifaddr_container *ifac, *next;
 682
 683         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
 684                               ifa_link, next) {
 685                 struct ifaddr *ifa = ifac->ifa;
 686
 687                 /* Leave link ifaddr as it is */
 688                 if (ifa->ifa_addr->sa_family == AF_LINK)
 689                         continue;
 690 #ifdef INET
 691                 /* XXX: Ugly!! ad hoc just for INET */
 692                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
 693                         struct ifaliasreq ifr;
 694 #ifdef IFADDR_DEBUG_VERBOSE
 695                         int i;
 696
 697                         kprintf("purge in4 addr %p: ", ifa);
 698                         for (i = 0; i < ncpus; ++i)
 699                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 700                         kprintf("\n");
 701 #endif
 702
 703                         bzero(&ifr, sizeof ifr);
 704                         ifr.ifra_addr = *ifa->ifa_addr;
 705                         if (ifa->ifa_dstaddr)
 706                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
 707                         if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
 708                                        NULL) == 0)
 709                                 continue;
 710                 }
 711 #endif /* INET */
 712 #ifdef INET6
 713                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
 714 #ifdef IFADDR_DEBUG_VERBOSE
 715                         int i;
 716
 717                         kprintf("purge in6 addr %p: ", ifa);
 718                         for (i = 0; i < ncpus; ++i)
 719                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 720                         kprintf("\n");
 721 #endif
 722
 723                         in6_purgeaddr(ifa);
 724                         /* ifp_addrhead is already updated */
 725                         continue;
 726                 }
 727 #endif /* INET6 */
 728                 ifa_ifunlink(ifa, ifp);
 729                 ifa_destroy(ifa);
 730         }
 731 }
 732
 733 static void
 734 ifq_stage_detach_handler(netmsg_t nmsg)
 735 {
 736         struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
 737         int q;
 738
 739         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 740                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 741                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 742
 743                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
 744                         ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
 745         }
 746         lwkt_replymsg(&nmsg->lmsg, 0);
 747 }
 748
 749 static void
 750 ifq_stage_detach(struct ifaltq *ifq)
 751 {
 752         struct netmsg_base base;
 753         int cpu;
 754
 755         netmsg_init(&base, NULL, &curthread->td_msgport, 0,
 756             ifq_stage_detach_handler);
 757         base.lmsg.u.ms_resultp = ifq;
 758
 759         for (cpu = 0; cpu < ncpus; ++cpu)
 760                 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
 761 }
 762
 763 struct netmsg_if_rtdel {
 764         struct netmsg_base      base;
 765         struct ifnet            *ifp;
 766 };
 767
 768 static void
 769 if_rtdel_dispatch(netmsg_t msg)
 770 {
 771         struct netmsg_if_rtdel *rmsg = (void *)msg;
 772         int i, nextcpu, cpu;
 773
 774         cpu = mycpuid;
 775         for (i = 1; i <= AF_MAX; i++) {
 776                 struct radix_node_head  *rnh;
 777
 778                 if ((rnh = rt_tables[cpu][i]) == NULL)
 779                         continue;
 780                 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
 781         }
 782
 783         nextcpu = cpu + 1;
 784         if (nextcpu < ncpus)
 785                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg);
 786         else
 787                 lwkt_replymsg(&rmsg->base.lmsg, 0);
 788 }
 789
 790 /*
 791  * Detach an interface, removing it from the
 792  * list of "active" interfaces.
 793  */
 794 void
 795 if_detach(struct ifnet *ifp)
 796 {
 797         struct netmsg_if_rtdel msg;
 798         struct domain *dp;
 799         int q;
 800
 801         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
 802
 803         /*
 804          * Remove routes and flush queues.
 805          */
 806         crit_enter();
 807 #ifdef IFPOLL_ENABLE
 808         if (ifp->if_flags & IFF_NPOLLING)
 809                 ifpoll_deregister(ifp);
 810 #endif
 811         if_down(ifp);
 812
 813 #ifdef ALTQ
 814         if (ifq_is_enabled(&ifp->if_snd))
 815                 altq_disable(&ifp->if_snd);
 816         if (ifq_is_attached(&ifp->if_snd))
 817                 altq_detach(&ifp->if_snd);
 818 #endif
 819
 820         /*
 821          * Clean up all addresses.
 822          */
 823         ifp->if_lladdr = NULL;
 824
 825         if_purgeaddrs_nolink(ifp);
 826         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
 827                 struct ifaddr *ifa;
 828
 829                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
 830                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
 831                         ("non-link ifaddr is left on if_addrheads"));
 832
 833                 ifa_ifunlink(ifa, ifp);
 834                 ifa_destroy(ifa);
 835                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
 836                         ("there are still ifaddrs left on if_addrheads"));
 837         }
 838
 839 #ifdef INET
 840         /*
 841          * Remove all IPv4 kernel structures related to ifp.
 842          */
 843         in_ifdetach(ifp);
 844 #endif
 845
 846 #ifdef INET6
 847         /*
 848          * Remove all IPv6 kernel structs related to ifp.  This should be done
 849          * before removing routing entries below, since IPv6 interface direct
 850          * routes are expected to be removed by the IPv6-specific kernel API.
 851          * Otherwise, the kernel will detect some inconsistency and bark it.
 852          */
 853         in6_ifdetach(ifp);
 854 #endif
 855
 856         /*
 857          * Delete all remaining routes using this interface
 858          */
 859         netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
 860             if_rtdel_dispatch);
 861         msg.ifp = ifp;
 862         rt_domsg_global(&msg.base);
 863
 864         /* Announce that the interface is gone. */
 865         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
 866         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
 867
 868         SLIST_FOREACH(dp, &domains, dom_next)
 869                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
 870                         (*dp->dom_ifdetach)(ifp,
 871                                 ifp->if_afdata[dp->dom_family]);
 872
 873         /*
 874          * Remove interface from ifindex2ifp[] and maybe decrement if_index.
 875          */
 876         lwkt_gettoken(&ifnet_token);
 877         ifindex2ifnet[ifp->if_index] = NULL;
 878         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
 879                 if_index--;
 880         TAILQ_REMOVE(&ifnet, ifp, if_link);
 881         lwkt_reltoken(&ifnet_token);
 882
 883         kfree(ifp->if_addrheads, M_IFADDR);
 884
 885         lwkt_synchronize_ipiqs("if_detach");
 886         ifq_stage_detach(&ifp->if_snd);
 887
 888         for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
 889                 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
 890
 891                 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
 892                 kfree(ifsq->ifsq_stage, M_DEVBUF);
 893         }
 894         kfree(ifp->if_snd.altq_subq, M_DEVBUF);
 895
 896         kfree(ifp->if_data_pcpu, M_DEVBUF);
 897
 898         crit_exit();
 899 }
 900
 901 /*
 902  * Create interface group without members
 903  */
 904 struct ifg_group *
 905 if_creategroup(const char *groupname)
 906 {
 907         struct ifg_group        *ifg = NULL;
 908
 909         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
 910             M_TEMP, M_NOWAIT)) == NULL)
 911                 return (NULL);
 912
 913         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
 914         ifg->ifg_refcnt = 0;
 915         ifg->ifg_carp_demoted = 0;
 916         TAILQ_INIT(&ifg->ifg_members);
 917 #if NPF > 0
 918         pfi_attach_ifgroup(ifg);
 919 #endif
 920         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
 921
 922         return (ifg);
 923 }
 924
 925 /*
 926  * Add a group to an interface
 927  */
 928 int
 929 if_addgroup(struct ifnet *ifp, const char *groupname)
 930 {
 931         struct ifg_list         *ifgl;
 932         struct ifg_group        *ifg = NULL;
 933         struct ifg_member       *ifgm;
 934
 935         if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
 936             groupname[strlen(groupname) - 1] <= '9')
 937                 return (EINVAL);
 938
 939         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
 940                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
 941                         return (EEXIST);
 942
 943         if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
 944                 return (ENOMEM);
 945
 946         if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
 947                 kfree(ifgl, M_TEMP);
 948                 return (ENOMEM);
 949         }
 950
 951         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
 952                 if (!strcmp(ifg->ifg_group, groupname))
 953                         break;
 954
 955         if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
 956                 kfree(ifgl, M_TEMP);
 957                 kfree(ifgm, M_TEMP);
 958                 return (ENOMEM);
 959         }
 960
 961         ifg->ifg_refcnt++;
 962         ifgl->ifgl_group = ifg;
 963         ifgm->ifgm_ifp = ifp;
 964
 965         TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
 966         TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
 967
 968 #if NPF > 0
 969         pfi_group_change(groupname);
 970 #endif
 971
 972         return (0);
 973 }
 974
 975 /*
 976  * Remove a group from an interface
 977  */
 978 int
 979 if_delgroup(struct ifnet *ifp, const char *groupname)
 980 {
 981         struct ifg_list         *ifgl;
 982         struct ifg_member       *ifgm;
 983
 984         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
 985                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
 986                         break;
 987         if (ifgl == NULL)
 988                 return (ENOENT);
 989
 990         TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
 991
 992         TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
 993                 if (ifgm->ifgm_ifp == ifp)
 994                         break;
 995
 996         if (ifgm != NULL) {
 997                 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
 998                 kfree(ifgm, M_TEMP);
 999         }
1000
1001         if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1002                 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
1003 #if NPF > 0
1004                 pfi_detach_ifgroup(ifgl->ifgl_group);
1005 #endif
1006                 kfree(ifgl->ifgl_group, M_TEMP);
1007         }
1008
1009         kfree(ifgl, M_TEMP);
1010
1011 #if NPF > 0
1012         pfi_group_change(groupname);
1013 #endif
1014
1015         return (0);
1016 }
1017
1018 /*
1019  * Stores all groups from an interface in memory pointed
1020  * to by data
1021  */
1022 int
1023 if_getgroup(caddr_t data, struct ifnet *ifp)
1024 {
1025         int                      len, error;
1026         struct ifg_list         *ifgl;
1027         struct ifg_req           ifgrq, *ifgp;
1028         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1029
1030         if (ifgr->ifgr_len == 0) {
1031                 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1032                         ifgr->ifgr_len += sizeof(struct ifg_req);
1033                 return (0);
1034         }
1035
1036         len = ifgr->ifgr_len;
1037         ifgp = ifgr->ifgr_groups;
1038         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1039                 if (len < sizeof(ifgrq))
1040                         return (EINVAL);
1041                 bzero(&ifgrq, sizeof ifgrq);
1042                 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1043                     sizeof(ifgrq.ifgrq_group));
1044                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1045                     sizeof(struct ifg_req))))
1046                         return (error);
1047                 len -= sizeof(ifgrq);
1048                 ifgp++;
1049         }
1050
1051         return (0);
1052 }
1053
1054 /*
1055  * Stores all members of a group in memory pointed to by data
1056  */
1057 int
1058 if_getgroupmembers(caddr_t data)
1059 {
1060         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1061         struct ifg_group        *ifg;
1062         struct ifg_member       *ifgm;
1063         struct ifg_req           ifgrq, *ifgp;
1064         int                      len, error;
1065
1066         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1067                 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1068                         break;
1069         if (ifg == NULL)
1070                 return (ENOENT);
1071
1072         if (ifgr->ifgr_len == 0) {
1073                 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1074                         ifgr->ifgr_len += sizeof(ifgrq);
1075                 return (0);
1076         }
1077
1078         len = ifgr->ifgr_len;
1079         ifgp = ifgr->ifgr_groups;
1080         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1081                 if (len < sizeof(ifgrq))
1082                         return (EINVAL);
1083                 bzero(&ifgrq, sizeof ifgrq);
1084                 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1085                     sizeof(ifgrq.ifgrq_member));
1086                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1087                     sizeof(struct ifg_req))))
1088                         return (error);
1089                 len -= sizeof(ifgrq);
1090                 ifgp++;
1091         }
1092
1093         return (0);
1094 }
1095
1096 /*
1097  * Delete Routes for a Network Interface
1098  *
1099  * Called for each routing entry via the rnh->rnh_walktree() call above
1100  * to delete all route entries referencing a detaching network interface.
1101  *
1102  * Arguments:
1103  *      rn      pointer to node in the routing table
1104  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1105  *
1106  * Returns:
1107  *      0       successful
1108  *      errno   failed - reason indicated
1109  *
1110  */
1111 static int
1112 if_rtdel(struct radix_node *rn, void *arg)
1113 {
1114         struct rtentry  *rt = (struct rtentry *)rn;
1115         struct ifnet    *ifp = arg;
1116         int             err;
1117
1118         if (rt->rt_ifp == ifp) {
1119
1120                 /*
1121                  * Protect (sorta) against walktree recursion problems
1122                  * with cloned routes
1123                  */
1124                 if (!(rt->rt_flags & RTF_UP))
1125                         return (0);
1126
1127                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1128                                 rt_mask(rt), rt->rt_flags,
1129                                 NULL);
1130                 if (err) {
1131                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1132                 }
1133         }
1134
1135         return (0);
1136 }
1137
1138 /*
1139  * Locate an interface based on a complete address.
1140  */
1141 struct ifaddr *
1142 ifa_ifwithaddr(struct sockaddr *addr)
1143 {
1144         struct ifnet *ifp;
1145
1146         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1147                 struct ifaddr_container *ifac;
1148
1149                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1150                         struct ifaddr *ifa = ifac->ifa;
1151
1152                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1153                                 continue;
1154                         if (sa_equal(addr, ifa->ifa_addr))
1155                                 return (ifa);
1156                         if ((ifp->if_flags & IFF_BROADCAST) &&
1157                             ifa->ifa_broadaddr &&
1158                             /* IPv6 doesn't have broadcast */
1159                             ifa->ifa_broadaddr->sa_len != 0 &&
1160                             sa_equal(ifa->ifa_broadaddr, addr))
1161                                 return (ifa);
1162                 }
1163         }
1164         return (NULL);
1165 }
1166 /*
1167  * Locate the point to point interface with a given destination address.
1168  */
1169 struct ifaddr *
1170 ifa_ifwithdstaddr(struct sockaddr *addr)
1171 {
1172         struct ifnet *ifp;
1173
1174         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1175                 struct ifaddr_container *ifac;
1176
1177                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1178                         continue;
1179
1180                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1181                         struct ifaddr *ifa = ifac->ifa;
1182
1183                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1184                                 continue;
1185                         if (ifa->ifa_dstaddr &&
1186                             sa_equal(addr, ifa->ifa_dstaddr))
1187                                 return (ifa);
1188                 }
1189         }
1190         return (NULL);
1191 }
1192
1193 /*
1194  * Find an interface on a specific network.  If many, choice
1195  * is most specific found.
1196  */
1197 struct ifaddr *
1198 ifa_ifwithnet(struct sockaddr *addr)
1199 {
1200         struct ifnet *ifp;
1201         struct ifaddr *ifa_maybe = NULL;
1202         u_int af = addr->sa_family;
1203         char *addr_data = addr->sa_data, *cplim;
1204
1205         /*
1206          * AF_LINK addresses can be looked up directly by their index number,
1207          * so do that if we can.
1208          */
1209         if (af == AF_LINK) {
1210                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1211
1212                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1213                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1214         }
1215
1216         /*
1217          * Scan though each interface, looking for ones that have
1218          * addresses in this address family.
1219          */
1220         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1221                 struct ifaddr_container *ifac;
1222
1223                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1224                         struct ifaddr *ifa = ifac->ifa;
1225                         char *cp, *cp2, *cp3;
1226
1227                         if (ifa->ifa_addr->sa_family != af)
1228 next:                           continue;
1229                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1230                                 /*
1231                                  * This is a bit broken as it doesn't
1232                                  * take into account that the remote end may
1233                                  * be a single node in the network we are
1234                                  * looking for.
1235                                  * The trouble is that we don't know the
1236                                  * netmask for the remote end.
1237                                  */
1238                                 if (ifa->ifa_dstaddr != NULL &&
1239                                     sa_equal(addr, ifa->ifa_dstaddr))
1240                                         return (ifa);
1241                         } else {
1242                                 /*
1243                                  * if we have a special address handler,
1244                                  * then use it instead of the generic one.
1245                                  */
1246                                 if (ifa->ifa_claim_addr) {
1247                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1248                                                 return (ifa);
1249                                         } else {
1250                                                 continue;
1251                                         }
1252                                 }
1253
1254                                 /*
1255                                  * Scan all the bits in the ifa's address.
1256                                  * If a bit dissagrees with what we are
1257                                  * looking for, mask it with the netmask
1258                                  * to see if it really matters.
1259                                  * (A byte at a time)
1260                                  */
1261                                 if (ifa->ifa_netmask == 0)
1262                                         continue;
1263                                 cp = addr_data;
1264                                 cp2 = ifa->ifa_addr->sa_data;
1265                                 cp3 = ifa->ifa_netmask->sa_data;
1266                                 cplim = ifa->ifa_netmask->sa_len +
1267                                         (char *)ifa->ifa_netmask;
1268                                 while (cp3 < cplim)
1269                                         if ((*cp++ ^ *cp2++) & *cp3++)
1270                                                 goto next; /* next address! */
1271                                 /*
1272                                  * If the netmask of what we just found
1273                                  * is more specific than what we had before
1274                                  * (if we had one) then remember the new one
1275                                  * before continuing to search
1276                                  * for an even better one.
1277                                  */
1278                                 if (ifa_maybe == NULL ||
1279                                     rn_refines((char *)ifa->ifa_netmask,
1280                                                (char *)ifa_maybe->ifa_netmask))
1281                                         ifa_maybe = ifa;
1282                         }
1283                 }
1284         }
1285         return (ifa_maybe);
1286 }
1287
1288 /*
1289  * Find an interface address specific to an interface best matching
1290  * a given address.
1291  */
1292 struct ifaddr *
1293 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1294 {
1295         struct ifaddr_container *ifac;
1296         char *cp, *cp2, *cp3;
1297         char *cplim;
1298         struct ifaddr *ifa_maybe = NULL;
1299         u_int af = addr->sa_family;
1300
1301         if (af >= AF_MAX)
1302                 return (0);
1303         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1304                 struct ifaddr *ifa = ifac->ifa;
1305
1306                 if (ifa->ifa_addr->sa_family != af)
1307                         continue;
1308                 if (ifa_maybe == NULL)
1309                         ifa_maybe = ifa;
1310                 if (ifa->ifa_netmask == NULL) {
1311                         if (sa_equal(addr, ifa->ifa_addr) ||
1312                             (ifa->ifa_dstaddr != NULL &&
1313                              sa_equal(addr, ifa->ifa_dstaddr)))
1314                                 return (ifa);
1315                         continue;
1316                 }
1317                 if (ifp->if_flags & IFF_POINTOPOINT) {
1318                         if (sa_equal(addr, ifa->ifa_dstaddr))
1319                                 return (ifa);
1320                 } else {
1321                         cp = addr->sa_data;
1322                         cp2 = ifa->ifa_addr->sa_data;
1323                         cp3 = ifa->ifa_netmask->sa_data;
1324                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1325                         for (; cp3 < cplim; cp3++)
1326                                 if ((*cp++ ^ *cp2++) & *cp3)
1327                                         break;
1328                         if (cp3 == cplim)
1329                                 return (ifa);
1330                 }
1331         }
1332         return (ifa_maybe);
1333 }
1334
1335 /*
1336  * Default action when installing a route with a Link Level gateway.
1337  * Lookup an appropriate real ifa to point to.
1338  * This should be moved to /sys/net/link.c eventually.
1339  */
1340 static void
1341 link_rtrequest(int cmd, struct rtentry *rt)
1342 {
1343         struct ifaddr *ifa;
1344         struct sockaddr *dst;
1345         struct ifnet *ifp;
1346
1347         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1348             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1349                 return;
1350         ifa = ifaof_ifpforaddr(dst, ifp);
1351         if (ifa != NULL) {
1352                 IFAFREE(rt->rt_ifa);
1353                 IFAREF(ifa);
1354                 rt->rt_ifa = ifa;
1355                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1356                         ifa->ifa_rtrequest(cmd, rt);
1357         }
1358 }
1359
1360 /*
1361  * Mark an interface down and notify protocols of
1362  * the transition.
1363  * NOTE: must be called at splnet or eqivalent.
1364  */
1365 void
1366 if_unroute(struct ifnet *ifp, int flag, int fam)
1367 {
1368         struct ifaddr_container *ifac;
1369
1370         ifp->if_flags &= ~flag;
1371         getmicrotime(&ifp->if_lastchange);
1372         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1373                 struct ifaddr *ifa = ifac->ifa;
1374
1375                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1376                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1377         }
1378         ifq_purge_all(&ifp->if_snd);
1379         rt_ifmsg(ifp);
1380 }
1381
1382 /*
1383  * Mark an interface up and notify protocols of
1384  * the transition.
1385  * NOTE: must be called at splnet or eqivalent.
1386  */
1387 void
1388 if_route(struct ifnet *ifp, int flag, int fam)
1389 {
1390         struct ifaddr_container *ifac;
1391
1392         ifq_purge_all(&ifp->if_snd);
1393         ifp->if_flags |= flag;
1394         getmicrotime(&ifp->if_lastchange);
1395         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1396                 struct ifaddr *ifa = ifac->ifa;
1397
1398                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1399                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1400         }
1401         rt_ifmsg(ifp);
1402 #ifdef INET6
1403         in6_if_up(ifp);
1404 #endif
1405 }
1406
1407 /*
1408  * Mark an interface down and notify protocols of the transition.  An
1409  * interface going down is also considered to be a synchronizing event.
1410  * We must ensure that all packet processing related to the interface
1411  * has completed before we return so e.g. the caller can free the ifnet
1412  * structure that the mbufs may be referencing.
1413  *
1414  * NOTE: must be called at splnet or eqivalent.
1415  */
1416 void
1417 if_down(struct ifnet *ifp)
1418 {
1419         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1420         netmsg_service_sync();
1421 }
1422
1423 /*
1424  * Mark an interface up and notify protocols of
1425  * the transition.
1426  * NOTE: must be called at splnet or eqivalent.
1427  */
1428 void
1429 if_up(struct ifnet *ifp)
1430 {
1431         if_route(ifp, IFF_UP, AF_UNSPEC);
1432 }
1433
1434 /*
1435  * Process a link state change.
1436  * NOTE: must be called at splsoftnet or equivalent.
1437  */
1438 void
1439 if_link_state_change(struct ifnet *ifp)
1440 {
1441         int link_state = ifp->if_link_state;
1442
1443         rt_ifmsg(ifp);
1444         devctl_notify("IFNET", ifp->if_xname,
1445             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1446 }
1447
1448 /*
1449  * Handle interface watchdog timer routines.  Called
1450  * from softclock, we decrement timers (if set) and
1451  * call the appropriate interface routine on expiration.
1452  */
1453 static void
1454 if_slowtimo(void *arg)
1455 {
1456         struct ifnet *ifp;
1457
1458         crit_enter();
1459
1460         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1461                 if (if_stats_compat) {
1462                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1463                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1464                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1465                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1466                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1467                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1468                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1469                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1470                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1471                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1472                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1473                 }
1474
1475                 if (ifp->if_timer == 0 || --ifp->if_timer)
1476                         continue;
1477                 if (ifp->if_watchdog) {
1478                         if (ifnet_tryserialize_all(ifp)) {
1479                                 (*ifp->if_watchdog)(ifp);
1480                                 ifnet_deserialize_all(ifp);
1481                         } else {
1482                                 /* try again next timeout */
1483                                 ++ifp->if_timer;
1484                         }
1485                 }
1486         }
1487
1488         crit_exit();
1489
1490         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1491 }
1492
1493 /*
1494  * Map interface name to
1495  * interface structure pointer.
1496  */
1497 struct ifnet *
1498 ifunit(const char *name)
1499 {
1500         struct ifnet *ifp;
1501
1502         /*
1503          * Search all the interfaces for this name/number
1504          */
1505
1506         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1507                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1508                         break;
1509         }
1510         return (ifp);
1511 }
1512
1513
1514 /*
1515  * Map interface name in a sockaddr_dl to
1516  * interface structure pointer.
1517  */
1518 struct ifnet *
1519 if_withname(struct sockaddr *sa)
1520 {
1521         char ifname[IFNAMSIZ+1];
1522         struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1523
1524         if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1525              (sdl->sdl_nlen > IFNAMSIZ) )
1526                 return NULL;
1527
1528         /*
1529          * ifunit wants a null-terminated name.  It may not be null-terminated
1530          * in the sockaddr.  We don't want to change the caller's sockaddr,
1531          * and there might not be room to put the trailing null anyway, so we
1532          * make a local copy that we know we can null terminate safely.
1533          */
1534
1535         bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1536         ifname[sdl->sdl_nlen] = '\0';
1537         return ifunit(ifname);
1538 }
1539
1540
1541 /*
1542  * Interface ioctls.
1543  */
1544 int
1545 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1546 {
1547         struct ifnet *ifp;
1548         struct ifreq *ifr;
1549         struct ifstat *ifs;
1550         int error;
1551         short oif_flags;
1552         int new_flags;
1553 #ifdef COMPAT_43
1554         int ocmd;
1555 #endif
1556         size_t namelen, onamelen;
1557         char new_name[IFNAMSIZ];
1558         struct ifaddr *ifa;
1559         struct sockaddr_dl *sdl;
1560
1561         switch (cmd) {
1562         case SIOCGIFCONF:
1563         case OSIOCGIFCONF:
1564                 return (ifconf(cmd, data, cred));
1565         default:
1566                 break;
1567         }
1568
1569         ifr = (struct ifreq *)data;
1570
1571         switch (cmd) {
1572         case SIOCIFCREATE:
1573         case SIOCIFCREATE2:
1574                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1575                         return (error);
1576                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1577                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1578         case SIOCIFDESTROY:
1579                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1580                         return (error);
1581                 return (if_clone_destroy(ifr->ifr_name));
1582         case SIOCIFGCLONERS:
1583                 return (if_clone_list((struct if_clonereq *)data));
1584         default:
1585                 break;
1586         }
1587
1588         /*
1589          * Nominal ioctl through interface, lookup the ifp and obtain a
1590          * lock to serialize the ifconfig ioctl operation.
1591          */
1592         ifp = ifunit(ifr->ifr_name);
1593         if (ifp == NULL)
1594                 return (ENXIO);
1595         error = 0;
1596         mtx_lock(&ifp->if_ioctl_mtx);
1597
1598         switch (cmd) {
1599         case SIOCGIFINDEX:
1600                 ifr->ifr_index = ifp->if_index;
1601                 break;
1602
1603         case SIOCGIFFLAGS:
1604                 ifr->ifr_flags = ifp->if_flags;
1605                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1606                 break;
1607
1608         case SIOCGIFCAP:
1609                 ifr->ifr_reqcap = ifp->if_capabilities;
1610                 ifr->ifr_curcap = ifp->if_capenable;
1611                 break;
1612
1613         case SIOCGIFMETRIC:
1614                 ifr->ifr_metric = ifp->if_metric;
1615                 break;
1616
1617         case SIOCGIFMTU:
1618                 ifr->ifr_mtu = ifp->if_mtu;
1619                 break;
1620
1621         case SIOCGIFTSOLEN:
1622                 ifr->ifr_tsolen = ifp->if_tsolen;
1623                 break;
1624
1625         case SIOCGIFDATA:
1626                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1627                                 sizeof(ifp->if_data));
1628                 break;
1629
1630         case SIOCGIFPHYS:
1631                 ifr->ifr_phys = ifp->if_physical;
1632                 break;
1633
1634         case SIOCGIFPOLLCPU:
1635                 ifr->ifr_pollcpu = -1;
1636                 break;
1637
1638         case SIOCSIFPOLLCPU:
1639                 break;
1640
1641         case SIOCSIFFLAGS:
1642                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1643                 if (error)
1644                         break;
1645                 new_flags = (ifr->ifr_flags & 0xffff) |
1646                     (ifr->ifr_flagshigh << 16);
1647                 if (ifp->if_flags & IFF_SMART) {
1648                         /* Smart drivers twiddle their own routes */
1649                 } else if (ifp->if_flags & IFF_UP &&
1650                     (new_flags & IFF_UP) == 0) {
1651                         crit_enter();
1652                         if_down(ifp);
1653                         crit_exit();
1654                 } else if (new_flags & IFF_UP &&
1655                     (ifp->if_flags & IFF_UP) == 0) {
1656                         crit_enter();
1657                         if_up(ifp);
1658                         crit_exit();
1659                 }
1660
1661 #ifdef IFPOLL_ENABLE
1662                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1663                         if (new_flags & IFF_NPOLLING)
1664                                 ifpoll_register(ifp);
1665                         else
1666                                 ifpoll_deregister(ifp);
1667                 }
1668 #endif
1669
1670                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1671                         (new_flags &~ IFF_CANTCHANGE);
1672                 if (new_flags & IFF_PPROMISC) {
1673                         /* Permanently promiscuous mode requested */
1674                         ifp->if_flags |= IFF_PROMISC;
1675                 } else if (ifp->if_pcount == 0) {
1676                         ifp->if_flags &= ~IFF_PROMISC;
1677                 }
1678                 if (ifp->if_ioctl) {
1679                         ifnet_serialize_all(ifp);
1680                         ifp->if_ioctl(ifp, cmd, data, cred);
1681                         ifnet_deserialize_all(ifp);
1682                 }
1683                 getmicrotime(&ifp->if_lastchange);
1684                 break;
1685
1686         case SIOCSIFCAP:
1687                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1688                 if (error)
1689                         break;
1690                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1691                         error = EINVAL;
1692                         break;
1693                 }
1694                 ifnet_serialize_all(ifp);
1695                 ifp->if_ioctl(ifp, cmd, data, cred);
1696                 ifnet_deserialize_all(ifp);
1697                 break;
1698
1699         case SIOCSIFNAME:
1700                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1701                 if (error)
1702                         break;
1703                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1704                 if (error)
1705                         break;
1706                 if (new_name[0] == '\0') {
1707                         error = EINVAL;
1708                         break;
1709                 }
1710                 if (ifunit(new_name) != NULL) {
1711                         error = EEXIST;
1712                         break;
1713                 }
1714
1715                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1716
1717                 /* Announce the departure of the interface. */
1718                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1719
1720                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1721                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1722                 /* XXX IFA_LOCK(ifa); */
1723                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1724                 namelen = strlen(new_name);
1725                 onamelen = sdl->sdl_nlen;
1726                 /*
1727                  * Move the address if needed.  This is safe because we
1728                  * allocate space for a name of length IFNAMSIZ when we
1729                  * create this in if_attach().
1730                  */
1731                 if (namelen != onamelen) {
1732                         bcopy(sdl->sdl_data + onamelen,
1733                             sdl->sdl_data + namelen, sdl->sdl_alen);
1734                 }
1735                 bcopy(new_name, sdl->sdl_data, namelen);
1736                 sdl->sdl_nlen = namelen;
1737                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1738                 bzero(sdl->sdl_data, onamelen);
1739                 while (namelen != 0)
1740                         sdl->sdl_data[--namelen] = 0xff;
1741                 /* XXX IFA_UNLOCK(ifa) */
1742
1743                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1744
1745                 /* Announce the return of the interface. */
1746                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1747                 break;
1748
1749         case SIOCSIFMETRIC:
1750                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1751                 if (error)
1752                         break;
1753                 ifp->if_metric = ifr->ifr_metric;
1754                 getmicrotime(&ifp->if_lastchange);
1755                 break;
1756
1757         case SIOCSIFPHYS:
1758                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1759                 if (error)
1760                         break;
1761                 if (ifp->if_ioctl == NULL) {
1762                         error = EOPNOTSUPP;
1763                         break;
1764                 }
1765                 ifnet_serialize_all(ifp);
1766                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1767                 ifnet_deserialize_all(ifp);
1768                 if (error == 0)
1769                         getmicrotime(&ifp->if_lastchange);
1770                 break;
1771
1772         case SIOCSIFMTU:
1773         {
1774                 u_long oldmtu = ifp->if_mtu;
1775
1776                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1777                 if (error)
1778                         break;
1779                 if (ifp->if_ioctl == NULL) {
1780                         error = EOPNOTSUPP;
1781                         break;
1782                 }
1783                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
1784                         error = EINVAL;
1785                         break;
1786                 }
1787                 ifnet_serialize_all(ifp);
1788                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1789                 ifnet_deserialize_all(ifp);
1790                 if (error == 0) {
1791                         getmicrotime(&ifp->if_lastchange);
1792                         rt_ifmsg(ifp);
1793                 }
1794                 /*
1795                  * If the link MTU changed, do network layer specific procedure.
1796                  */
1797                 if (ifp->if_mtu != oldmtu) {
1798 #ifdef INET6
1799                         nd6_setmtu(ifp);
1800 #endif
1801                 }
1802                 break;
1803         }
1804
1805         case SIOCSIFTSOLEN:
1806                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1807                 if (error)
1808                         break;
1809
1810                 /* XXX need driver supplied upper limit */
1811                 if (ifr->ifr_tsolen <= 0) {
1812                         error = EINVAL;
1813                         break;
1814                 }
1815                 ifp->if_tsolen = ifr->ifr_tsolen;
1816                 break;
1817
1818         case SIOCADDMULTI:
1819         case SIOCDELMULTI:
1820                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1821                 if (error)
1822                         break;
1823
1824                 /* Don't allow group membership on non-multicast interfaces. */
1825                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1826                         error = EOPNOTSUPP;
1827                         break;
1828                 }
1829
1830                 /* Don't let users screw up protocols' entries. */
1831                 if (ifr->ifr_addr.sa_family != AF_LINK) {
1832                         error = EINVAL;
1833                         break;
1834                 }
1835
1836                 if (cmd == SIOCADDMULTI) {
1837                         struct ifmultiaddr *ifma;
1838                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1839                 } else {
1840                         error = if_delmulti(ifp, &ifr->ifr_addr);
1841                 }
1842                 if (error == 0)
1843                         getmicrotime(&ifp->if_lastchange);
1844                 break;
1845
1846         case SIOCSIFPHYADDR:
1847         case SIOCDIFPHYADDR:
1848 #ifdef INET6
1849         case SIOCSIFPHYADDR_IN6:
1850 #endif
1851         case SIOCSLIFPHYADDR:
1852         case SIOCSIFMEDIA:
1853         case SIOCSIFGENERIC:
1854                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1855                 if (error)
1856                         break;
1857                 if (ifp->if_ioctl == 0) {
1858                         error = EOPNOTSUPP;
1859                         break;
1860                 }
1861                 ifnet_serialize_all(ifp);
1862                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1863                 ifnet_deserialize_all(ifp);
1864                 if (error == 0)
1865                         getmicrotime(&ifp->if_lastchange);
1866                 break;
1867
1868         case SIOCGIFSTATUS:
1869                 ifs = (struct ifstat *)data;
1870                 ifs->ascii[0] = '\0';
1871                 /* fall through */
1872         case SIOCGIFPSRCADDR:
1873         case SIOCGIFPDSTADDR:
1874         case SIOCGLIFPHYADDR:
1875         case SIOCGIFMEDIA:
1876         case SIOCGIFGENERIC:
1877                 if (ifp->if_ioctl == NULL) {
1878                         error = EOPNOTSUPP;
1879                         break;
1880                 }
1881                 ifnet_serialize_all(ifp);
1882                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1883                 ifnet_deserialize_all(ifp);
1884                 break;
1885
1886         case SIOCSIFLLADDR:
1887                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1888                 if (error)
1889                         break;
1890                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
1891                                      ifr->ifr_addr.sa_len);
1892                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1893                 break;
1894
1895         default:
1896                 oif_flags = ifp->if_flags;
1897                 if (so->so_proto == 0) {
1898                         error = EOPNOTSUPP;
1899                         break;
1900                 }
1901 #ifndef COMPAT_43
1902                 error = so_pru_control_direct(so, cmd, data, ifp);
1903 #else
1904                 ocmd = cmd;
1905
1906                 switch (cmd) {
1907                 case SIOCSIFDSTADDR:
1908                 case SIOCSIFADDR:
1909                 case SIOCSIFBRDADDR:
1910                 case SIOCSIFNETMASK:
1911 #if BYTE_ORDER != BIG_ENDIAN
1912                         if (ifr->ifr_addr.sa_family == 0 &&
1913                             ifr->ifr_addr.sa_len < 16) {
1914                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1915                                 ifr->ifr_addr.sa_len = 16;
1916                         }
1917 #else
1918                         if (ifr->ifr_addr.sa_len == 0)
1919                                 ifr->ifr_addr.sa_len = 16;
1920 #endif
1921                         break;
1922                 case OSIOCGIFADDR:
1923                         cmd = SIOCGIFADDR;
1924                         break;
1925                 case OSIOCGIFDSTADDR:
1926                         cmd = SIOCGIFDSTADDR;
1927                         break;
1928                 case OSIOCGIFBRDADDR:
1929                         cmd = SIOCGIFBRDADDR;
1930                         break;
1931                 case OSIOCGIFNETMASK:
1932                         cmd = SIOCGIFNETMASK;
1933                         break;
1934                 default:
1935                         break;
1936                 }
1937
1938                 error = so_pru_control_direct(so, cmd, data, ifp);
1939
1940                 switch (ocmd) {
1941                 case OSIOCGIFADDR:
1942                 case OSIOCGIFDSTADDR:
1943                 case OSIOCGIFBRDADDR:
1944                 case OSIOCGIFNETMASK:
1945                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1946                         break;
1947                 }
1948 #endif /* COMPAT_43 */
1949
1950                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1951 #ifdef INET6
1952                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
1953                         if (ifp->if_flags & IFF_UP) {
1954                                 crit_enter();
1955                                 in6_if_up(ifp);
1956                                 crit_exit();
1957                         }
1958 #endif
1959                 }
1960                 break;
1961         }
1962
1963         mtx_unlock(&ifp->if_ioctl_mtx);
1964         return (error);
1965 }
1966
1967 /*
1968  * Set/clear promiscuous mode on interface ifp based on the truth value
1969  * of pswitch.  The calls are reference counted so that only the first
1970  * "on" request actually has an effect, as does the final "off" request.
1971  * Results are undefined if the "off" and "on" requests are not matched.
1972  */
1973 int
1974 ifpromisc(struct ifnet *ifp, int pswitch)
1975 {
1976         struct ifreq ifr;
1977         int error;
1978         int oldflags;
1979
1980         oldflags = ifp->if_flags;
1981         if (ifp->if_flags & IFF_PPROMISC) {
1982                 /* Do nothing if device is in permanently promiscuous mode */
1983                 ifp->if_pcount += pswitch ? 1 : -1;
1984                 return (0);
1985         }
1986         if (pswitch) {
1987                 /*
1988                  * If the device is not configured up, we cannot put it in
1989                  * promiscuous mode.
1990                  */
1991                 if ((ifp->if_flags & IFF_UP) == 0)
1992                         return (ENETDOWN);
1993                 if (ifp->if_pcount++ != 0)
1994                         return (0);
1995                 ifp->if_flags |= IFF_PROMISC;
1996                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1997                     ifp->if_xname);
1998         } else {
1999                 if (--ifp->if_pcount > 0)
2000                         return (0);
2001                 ifp->if_flags &= ~IFF_PROMISC;
2002                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2003                     ifp->if_xname);
2004         }
2005         ifr.ifr_flags = ifp->if_flags;
2006         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2007         ifnet_serialize_all(ifp);
2008         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2009         ifnet_deserialize_all(ifp);
2010         if (error == 0)
2011                 rt_ifmsg(ifp);
2012         else
2013                 ifp->if_flags = oldflags;
2014         return error;
2015 }
2016
2017 /*
2018  * Return interface configuration
2019  * of system.  List may be used
2020  * in later ioctl's (above) to get
2021  * other information.
2022  */
2023 static int
2024 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2025 {
2026         struct ifconf *ifc = (struct ifconf *)data;
2027         struct ifnet *ifp;
2028         struct sockaddr *sa;
2029         struct ifreq ifr, *ifrp;
2030         int space = ifc->ifc_len, error = 0;
2031
2032         ifrp = ifc->ifc_req;
2033         TAILQ_FOREACH(ifp, &ifnet, if_link) {
2034                 struct ifaddr_container *ifac;
2035                 int addrs;
2036
2037                 if (space <= sizeof ifr)
2038                         break;
2039
2040                 /*
2041                  * Zero the stack declared structure first to prevent
2042                  * memory disclosure.
2043                  */
2044                 bzero(&ifr, sizeof(ifr));
2045                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2046                     >= sizeof(ifr.ifr_name)) {
2047                         error = ENAMETOOLONG;
2048                         break;
2049                 }
2050
2051                 addrs = 0;
2052                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2053                         struct ifaddr *ifa = ifac->ifa;
2054
2055                         if (space <= sizeof ifr)
2056                                 break;
2057                         sa = ifa->ifa_addr;
2058                         if (cred->cr_prison &&
2059                             prison_if(cred, sa))
2060                                 continue;
2061                         addrs++;
2062 #ifdef COMPAT_43
2063                         if (cmd == OSIOCGIFCONF) {
2064                                 struct osockaddr *osa =
2065                                          (struct osockaddr *)&ifr.ifr_addr;
2066                                 ifr.ifr_addr = *sa;
2067                                 osa->sa_family = sa->sa_family;
2068                                 error = copyout(&ifr, ifrp, sizeof ifr);
2069                                 ifrp++;
2070                         } else
2071 #endif
2072                         if (sa->sa_len <= sizeof(*sa)) {
2073                                 ifr.ifr_addr = *sa;
2074                                 error = copyout(&ifr, ifrp, sizeof ifr);
2075                                 ifrp++;
2076                         } else {
2077                                 if (space < (sizeof ifr) + sa->sa_len -
2078                                             sizeof(*sa))
2079                                         break;
2080                                 space -= sa->sa_len - sizeof(*sa);
2081                                 error = copyout(&ifr, ifrp,
2082                                                 sizeof ifr.ifr_name);
2083                                 if (error == 0)
2084                                         error = copyout(sa, &ifrp->ifr_addr,
2085                                                         sa->sa_len);
2086                                 ifrp = (struct ifreq *)
2087                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2088                         }
2089                         if (error)
2090                                 break;
2091                         space -= sizeof ifr;
2092                 }
2093                 if (error)
2094                         break;
2095                 if (!addrs) {
2096                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2097                         error = copyout(&ifr, ifrp, sizeof ifr);
2098                         if (error)
2099                                 break;
2100                         space -= sizeof ifr;
2101                         ifrp++;
2102                 }
2103         }
2104         ifc->ifc_len -= space;
2105         return (error);
2106 }
2107
2108 /*
2109  * Just like if_promisc(), but for all-multicast-reception mode.
2110  */
2111 int
2112 if_allmulti(struct ifnet *ifp, int onswitch)
2113 {
2114         int error = 0;
2115         struct ifreq ifr;
2116
2117         crit_enter();
2118
2119         if (onswitch) {
2120                 if (ifp->if_amcount++ == 0) {
2121                         ifp->if_flags |= IFF_ALLMULTI;
2122                         ifr.ifr_flags = ifp->if_flags;
2123                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2124                         ifnet_serialize_all(ifp);
2125                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2126                                               NULL);
2127                         ifnet_deserialize_all(ifp);
2128                 }
2129         } else {
2130                 if (ifp->if_amcount > 1) {
2131                         ifp->if_amcount--;
2132                 } else {
2133                         ifp->if_amcount = 0;
2134                         ifp->if_flags &= ~IFF_ALLMULTI;
2135                         ifr.ifr_flags = ifp->if_flags;
2136                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2137                         ifnet_serialize_all(ifp);
2138                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2139                                               NULL);
2140                         ifnet_deserialize_all(ifp);
2141                 }
2142         }
2143
2144         crit_exit();
2145
2146         if (error == 0)
2147                 rt_ifmsg(ifp);
2148         return error;
2149 }
2150
2151 /*
2152  * Add a multicast listenership to the interface in question.
2153  * The link layer provides a routine which converts
2154  */
2155 int
2156 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2157     struct ifmultiaddr **retifma)
2158 {
2159         struct sockaddr *llsa, *dupsa;
2160         int error;
2161         struct ifmultiaddr *ifma;
2162
2163         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2164
2165         /*
2166          * If the matching multicast address already exists
2167          * then don't add a new one, just add a reference
2168          */
2169         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2170                 if (sa_equal(sa, ifma->ifma_addr)) {
2171                         ifma->ifma_refcount++;
2172                         if (retifma)
2173                                 *retifma = ifma;
2174                         return 0;
2175                 }
2176         }
2177
2178         /*
2179          * Give the link layer a chance to accept/reject it, and also
2180          * find out which AF_LINK address this maps to, if it isn't one
2181          * already.
2182          */
2183         if (ifp->if_resolvemulti) {
2184                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2185                 if (error)
2186                         return error;
2187         } else {
2188                 llsa = NULL;
2189         }
2190
2191         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2192         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_WAITOK);
2193         bcopy(sa, dupsa, sa->sa_len);
2194
2195         ifma->ifma_addr = dupsa;
2196         ifma->ifma_lladdr = llsa;
2197         ifma->ifma_ifp = ifp;
2198         ifma->ifma_refcount = 1;
2199         ifma->ifma_protospec = NULL;
2200         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2201
2202         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2203         if (retifma)
2204                 *retifma = ifma;
2205
2206         if (llsa != NULL) {
2207                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2208                         if (sa_equal(ifma->ifma_addr, llsa))
2209                                 break;
2210                 }
2211                 if (ifma) {
2212                         ifma->ifma_refcount++;
2213                 } else {
2214                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2215                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_WAITOK);
2216                         bcopy(llsa, dupsa, llsa->sa_len);
2217                         ifma->ifma_addr = dupsa;
2218                         ifma->ifma_ifp = ifp;
2219                         ifma->ifma_refcount = 1;
2220                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2221                 }
2222         }
2223         /*
2224          * We are certain we have added something, so call down to the
2225          * interface to let them know about it.
2226          */
2227         if (ifp->if_ioctl)
2228                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2229
2230         return 0;
2231 }
2232
2233 int
2234 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2235     struct ifmultiaddr **retifma)
2236 {
2237         int error;
2238
2239         ifnet_serialize_all(ifp);
2240         error = if_addmulti_serialized(ifp, sa, retifma);
2241         ifnet_deserialize_all(ifp);
2242
2243         return error;
2244 }
2245
2246 /*
2247  * Remove a reference to a multicast address on this interface.  Yell
2248  * if the request does not match an existing membership.
2249  */
2250 static int
2251 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2252 {
2253         struct ifmultiaddr *ifma;
2254
2255         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2256
2257         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2258                 if (sa_equal(sa, ifma->ifma_addr))
2259                         break;
2260         if (ifma == NULL)
2261                 return ENOENT;
2262
2263         if (ifma->ifma_refcount > 1) {
2264                 ifma->ifma_refcount--;
2265                 return 0;
2266         }
2267
2268         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2269         sa = ifma->ifma_lladdr;
2270         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2271         /*
2272          * Make sure the interface driver is notified
2273          * in the case of a link layer mcast group being left.
2274          */
2275         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2276                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2277         kfree(ifma->ifma_addr, M_IFMADDR);
2278         kfree(ifma, M_IFMADDR);
2279         if (sa == NULL)
2280                 return 0;
2281
2282         /*
2283          * Now look for the link-layer address which corresponds to
2284          * this network address.  It had been squirreled away in
2285          * ifma->ifma_lladdr for this purpose (so we don't have
2286          * to call ifp->if_resolvemulti() again), and we saved that
2287          * value in sa above.  If some nasty deleted the
2288          * link-layer address out from underneath us, we can deal because
2289          * the address we stored was is not the same as the one which was
2290          * in the record for the link-layer address.  (So we don't complain
2291          * in that case.)
2292          */
2293         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2294                 if (sa_equal(sa, ifma->ifma_addr))
2295                         break;
2296         if (ifma == NULL)
2297                 return 0;
2298
2299         if (ifma->ifma_refcount > 1) {
2300                 ifma->ifma_refcount--;
2301                 return 0;
2302         }
2303
2304         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2305         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2306         kfree(ifma->ifma_addr, M_IFMADDR);
2307         kfree(sa, M_IFMADDR);
2308         kfree(ifma, M_IFMADDR);
2309
2310         return 0;
2311 }
2312
2313 int
2314 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2315 {
2316         int error;
2317
2318         ifnet_serialize_all(ifp);
2319         error = if_delmulti_serialized(ifp, sa);
2320         ifnet_deserialize_all(ifp);
2321
2322         return error;
2323 }
2324
2325 /*
2326  * Delete all multicast group membership for an interface.
2327  * Should be used to quickly flush all multicast filters.
2328  */
2329 void
2330 if_delallmulti_serialized(struct ifnet *ifp)
2331 {
2332         struct ifmultiaddr *ifma, mark;
2333         struct sockaddr sa;
2334
2335         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2336
2337         bzero(&sa, sizeof(sa));
2338         sa.sa_family = AF_UNSPEC;
2339         sa.sa_len = sizeof(sa);
2340
2341         bzero(&mark, sizeof(mark));
2342         mark.ifma_addr = &sa;
2343
2344         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2345
2346         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2347                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2348                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2349                     ifma_link);
2350
2351                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2352                         continue;
2353
2354                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2355         }
2356 }
2357
2358
2359 /*
2360  * Set the link layer address on an interface.
2361  *
2362  * At this time we only support certain types of interfaces,
2363  * and we don't allow the length of the address to change.
2364  */
2365 int
2366 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2367 {
2368         struct sockaddr_dl *sdl;
2369         struct ifreq ifr;
2370
2371         sdl = IF_LLSOCKADDR(ifp);
2372         if (sdl == NULL)
2373                 return (EINVAL);
2374         if (len != sdl->sdl_alen)       /* don't allow length to change */
2375                 return (EINVAL);
2376         switch (ifp->if_type) {
2377         case IFT_ETHER:                 /* these types use struct arpcom */
2378         case IFT_XETHER:
2379         case IFT_L2VLAN:
2380                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2381                 bcopy(lladdr, LLADDR(sdl), len);
2382                 break;
2383         default:
2384                 return (ENODEV);
2385         }
2386         /*
2387          * If the interface is already up, we need
2388          * to re-init it in order to reprogram its
2389          * address filter.
2390          */
2391         ifnet_serialize_all(ifp);
2392         if ((ifp->if_flags & IFF_UP) != 0) {
2393 #ifdef INET
2394                 struct ifaddr_container *ifac;
2395 #endif
2396
2397                 ifp->if_flags &= ~IFF_UP;
2398                 ifr.ifr_flags = ifp->if_flags;
2399                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2400                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2401                               NULL);
2402                 ifp->if_flags |= IFF_UP;
2403                 ifr.ifr_flags = ifp->if_flags;
2404                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2405                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2406                                  NULL);
2407 #ifdef INET
2408                 /*
2409                  * Also send gratuitous ARPs to notify other nodes about
2410                  * the address change.
2411                  */
2412                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2413                         struct ifaddr *ifa = ifac->ifa;
2414
2415                         if (ifa->ifa_addr != NULL &&
2416                             ifa->ifa_addr->sa_family == AF_INET)
2417                                 arp_gratuitous(ifp, ifa);
2418                 }
2419 #endif
2420         }
2421         ifnet_deserialize_all(ifp);
2422         return (0);
2423 }
2424
2425 struct ifmultiaddr *
2426 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2427 {
2428         struct ifmultiaddr *ifma;
2429
2430         /* TODO: need ifnet_serialize_main */
2431         ifnet_serialize_all(ifp);
2432         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2433                 if (sa_equal(ifma->ifma_addr, sa))
2434                         break;
2435         ifnet_deserialize_all(ifp);
2436
2437         return ifma;
2438 }
2439
2440 /*
2441  * This function locates the first real ethernet MAC from a network
2442  * card and loads it into node, returning 0 on success or ENOENT if
2443  * no suitable interfaces were found.  It is used by the uuid code to
2444  * generate a unique 6-byte number.
2445  */
2446 int
2447 if_getanyethermac(uint16_t *node, int minlen)
2448 {
2449         struct ifnet *ifp;
2450         struct sockaddr_dl *sdl;
2451
2452         TAILQ_FOREACH(ifp, &ifnet, if_link) {
2453                 if (ifp->if_type != IFT_ETHER)
2454                         continue;
2455                 sdl = IF_LLSOCKADDR(ifp);
2456                 if (sdl->sdl_alen < minlen)
2457                         continue;
2458                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2459                       minlen);
2460                 return(0);
2461         }
2462         return (ENOENT);
2463 }
2464
2465 /*
2466  * The name argument must be a pointer to storage which will last as
2467  * long as the interface does.  For physical devices, the result of
2468  * device_get_name(dev) is a good choice and for pseudo-devices a
2469  * static string works well.
2470  */
2471 void
2472 if_initname(struct ifnet *ifp, const char *name, int unit)
2473 {
2474         ifp->if_dname = name;
2475         ifp->if_dunit = unit;
2476         if (unit != IF_DUNIT_NONE)
2477                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2478         else
2479                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2480 }
2481
2482 int
2483 if_printf(struct ifnet *ifp, const char *fmt, ...)
2484 {
2485         __va_list ap;
2486         int retval;
2487
2488         retval = kprintf("%s: ", ifp->if_xname);
2489         __va_start(ap, fmt);
2490         retval += kvprintf(fmt, ap);
2491         __va_end(ap);
2492         return (retval);
2493 }
2494
2495 struct ifnet *
2496 if_alloc(uint8_t type)
2497 {
2498         struct ifnet *ifp;
2499         size_t size;
2500
2501         /*
2502          * XXX temporary hack until arpcom is setup in if_l2com
2503          */
2504         if (type == IFT_ETHER)
2505                 size = sizeof(struct arpcom);
2506         else
2507                 size = sizeof(struct ifnet);
2508
2509         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2510
2511         ifp->if_type = type;
2512
2513         if (if_com_alloc[type] != NULL) {
2514                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2515                 if (ifp->if_l2com == NULL) {
2516                         kfree(ifp, M_IFNET);
2517                         return (NULL);
2518                 }
2519         }
2520         return (ifp);
2521 }
2522
2523 void
2524 if_free(struct ifnet *ifp)
2525 {
2526         kfree(ifp, M_IFNET);
2527 }
2528
2529 void
2530 ifq_set_classic(struct ifaltq *ifq)
2531 {
2532         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2533             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2534 }
2535
2536 void
2537 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2538     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2539 {
2540         int q;
2541
2542         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2543         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2544         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2545         KASSERT(request != NULL, ("request is not specified"));
2546
2547         ifq->altq_mapsubq = mapsubq;
2548         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2549                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2550
2551                 ifsq->ifsq_enqueue = enqueue;
2552                 ifsq->ifsq_dequeue = dequeue;
2553                 ifsq->ifsq_request = request;
2554         }
2555 }
2556
2557 static void
2558 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2559 {
2560         m->m_nextpkt = NULL;
2561         if (ifsq->ifsq_norm_tail == NULL)
2562                 ifsq->ifsq_norm_head = m;
2563         else
2564                 ifsq->ifsq_norm_tail->m_nextpkt = m;
2565         ifsq->ifsq_norm_tail = m;
2566         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2567 }
2568
2569 static void
2570 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2571 {
2572         m->m_nextpkt = NULL;
2573         if (ifsq->ifsq_prio_tail == NULL)
2574                 ifsq->ifsq_prio_head = m;
2575         else
2576                 ifsq->ifsq_prio_tail->m_nextpkt = m;
2577         ifsq->ifsq_prio_tail = m;
2578         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2579         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2580 }
2581
2582 static struct mbuf *
2583 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2584 {
2585         struct mbuf *m;
2586
2587         m = ifsq->ifsq_norm_head;
2588         if (m != NULL) {
2589                 if ((ifsq->ifsq_norm_head = m->m_nextpkt) == NULL)
2590                         ifsq->ifsq_norm_tail = NULL;
2591                 m->m_nextpkt = NULL;
2592                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2593         }
2594         return m;
2595 }
2596
2597 static struct mbuf *
2598 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2599 {
2600         struct mbuf *m;
2601
2602         m = ifsq->ifsq_prio_head;
2603         if (m != NULL) {
2604                 if ((ifsq->ifsq_prio_head = m->m_nextpkt) == NULL)
2605                         ifsq->ifsq_prio_tail = NULL;
2606                 m->m_nextpkt = NULL;
2607                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2608                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2609         }
2610         return m;
2611 }
2612
2613 int
2614 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2615     struct altq_pktattr *pa __unused)
2616 {
2617         M_ASSERTPKTHDR(m);
2618         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2619             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2620                 if ((m->m_flags & M_PRIO) &&
2621                     ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen / 2) &&
2622                     ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt / 2)) {
2623                         struct mbuf *m_drop;
2624
2625                         /*
2626                          * Perform drop-head on normal queue
2627                          */
2628                         m_drop = ifsq_norm_dequeue(ifsq);
2629                         if (m_drop != NULL) {
2630                                 m_freem(m_drop);
2631                                 ifsq_prio_enqueue(ifsq, m);
2632                                 return 0;
2633                         }
2634                         /* XXX nothing could be dropped? */
2635                 }
2636                 m_freem(m);
2637                 return ENOBUFS;
2638         } else {
2639                 if (m->m_flags & M_PRIO)
2640                         ifsq_prio_enqueue(ifsq, m);
2641                 else
2642                         ifsq_norm_enqueue(ifsq, m);
2643                 return 0;
2644         }
2645 }
2646
2647 struct mbuf *
2648 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2649 {
2650         struct mbuf *m;
2651
2652         switch (op) {
2653         case ALTDQ_POLL:
2654                 m = ifsq->ifsq_prio_head;
2655                 if (m == NULL)
2656                         m = ifsq->ifsq_norm_head;
2657                 break;
2658
2659         case ALTDQ_REMOVE:
2660                 m = ifsq_prio_dequeue(ifsq);
2661                 if (m == NULL)
2662                         m = ifsq_norm_dequeue(ifsq);
2663                 break;
2664
2665         default:
2666                 panic("unsupported ALTQ dequeue op: %d", op);
2667         }
2668         return m;
2669 }
2670
2671 int
2672 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2673 {
2674         switch (req) {
2675         case ALTRQ_PURGE:
2676                 for (;;) {
2677                         struct mbuf *m;
2678
2679                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2680                         if (m == NULL)
2681                                 break;
2682                         m_freem(m);
2683                 }
2684                 break;
2685
2686         default:
2687                 panic("unsupported ALTQ request: %d", req);
2688         }
2689         return 0;
2690 }
2691
2692 static void
2693 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2694 {
2695         struct ifnet *ifp = ifsq_get_ifp(ifsq);
2696         int running = 0, need_sched;
2697
2698         /*
2699          * Try to do direct ifnet.if_start on the subqueue first, if there is
2700          * contention on the subqueue hardware serializer, ifnet.if_start on
2701          * the subqueue will be scheduled on the subqueue owner CPU.
2702          */
2703         if (!ifsq_tryserialize_hw(ifsq)) {
2704                 /*
2705                  * Subqueue hardware serializer contention happened,
2706                  * ifnet.if_start on the subqueue is scheduled on
2707                  * the subqueue owner CPU, and we keep going.
2708                  */
2709                 ifsq_ifstart_schedule(ifsq, 1);
2710                 return;
2711         }
2712
2713         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2714                 ifp->if_start(ifp, ifsq);
2715                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2716                         running = 1;
2717         }
2718         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2719
2720         ifsq_deserialize_hw(ifsq);
2721
2722         if (need_sched) {
2723                 /*
2724                  * More data need to be transmitted, ifnet.if_start on the
2725                  * subqueue is scheduled on the subqueue owner CPU, and we
2726                  * keep going.
2727                  * NOTE: ifnet.if_start subqueue interlock is not released.
2728                  */
2729                 ifsq_ifstart_schedule(ifsq, force_sched);
2730         }
2731 }
2732
2733 /*
2734  * Subqeue packets staging mechanism:
2735  *
2736  * The packets enqueued into the subqueue are staged to a certain amount
2737  * before the ifnet.if_start on the subqueue is called.  In this way, the
2738  * driver could avoid writing to hardware registers upon every packet,
2739  * instead, hardware registers could be written when certain amount of
2740  * packets are put onto hardware TX ring.  The measurement on several modern
2741  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
2742  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
2743  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
2744  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
2745  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
2746  *
2747  * Subqueue packets staging is performed for two entry points into drivers'
2748  * transmission function:
2749  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
2750  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
2751  *
2752  * Subqueue packets staging will be stopped upon any of the following
2753  * conditions:
2754  * - If the count of packets enqueued on the current CPU is great than or
2755  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2756  * - If the total length of packets enqueued on the current CPU is great
2757  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
2758  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
2759  *   is usually less than hardware's MTU.
2760  * - ifsq_ifstart_schedule() is not pending on the current CPU and
2761  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
2762  *   released.
2763  * - The if_start_rollup(), which is registered as low priority netisr
2764  *   rollup function, is called; probably because no more work is pending
2765  *   for netisr.
2766  *
2767  * NOTE:
2768  * Currently subqueue packet staging is only performed in netisr threads.
2769  */
2770 int
2771 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2772 {
2773         struct ifaltq *ifq = &ifp->if_snd;
2774         struct ifaltq_subque *ifsq;
2775         int error, start = 0, len, mcast = 0, avoid_start = 0;
2776         struct ifsubq_stage_head *head = NULL;
2777         struct ifsubq_stage *stage = NULL;
2778         struct globaldata *gd = mycpu;
2779         struct thread *td = gd->gd_curthread;
2780
2781         crit_enter_quick(td);
2782
2783         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
2784         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
2785
2786         len = m->m_pkthdr.len;
2787         if (m->m_flags & M_MCAST)
2788                 mcast = 1;
2789
2790         if (td->td_type == TD_TYPE_NETISR) {
2791                 head = &ifsubq_stage_heads[mycpuid];
2792                 stage = ifsq_get_stage(ifsq, mycpuid);
2793
2794                 stage->stg_cnt++;
2795                 stage->stg_len += len;
2796                 if (stage->stg_cnt < ifsq_stage_cntmax &&
2797                     stage->stg_len < (ifp->if_mtu - max_protohdr))
2798                         avoid_start = 1;
2799         }
2800
2801         ALTQ_SQ_LOCK(ifsq);
2802         error = ifsq_enqueue_locked(ifsq, m, pa);
2803         if (error) {
2804                 if (!ifsq_data_ready(ifsq)) {
2805                         ALTQ_SQ_UNLOCK(ifsq);
2806                         crit_exit_quick(td);
2807                         return error;
2808                 }
2809                 avoid_start = 0;
2810         }
2811         if (!ifsq_is_started(ifsq)) {
2812                 if (avoid_start) {
2813                         ALTQ_SQ_UNLOCK(ifsq);
2814
2815                         KKASSERT(!error);
2816                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
2817                                 ifsq_stage_insert(head, stage);
2818
2819                         IFNET_STAT_INC(ifp, obytes, len);
2820                         if (mcast)
2821                                 IFNET_STAT_INC(ifp, omcasts, 1);
2822                         crit_exit_quick(td);
2823                         return error;
2824                 }
2825
2826                 /*
2827                  * Hold the subqueue interlock of ifnet.if_start
2828                  */
2829                 ifsq_set_started(ifsq);
2830                 start = 1;
2831         }
2832         ALTQ_SQ_UNLOCK(ifsq);
2833
2834         if (!error) {
2835                 IFNET_STAT_INC(ifp, obytes, len);
2836                 if (mcast)
2837                         IFNET_STAT_INC(ifp, omcasts, 1);
2838         }
2839
2840         if (stage != NULL) {
2841                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
2842                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
2843                         if (!avoid_start) {
2844                                 ifsq_stage_remove(head, stage);
2845                                 ifsq_ifstart_schedule(ifsq, 1);
2846                         }
2847                         crit_exit_quick(td);
2848                         return error;
2849                 }
2850
2851                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
2852                         ifsq_stage_remove(head, stage);
2853                 } else {
2854                         stage->stg_cnt = 0;
2855                         stage->stg_len = 0;
2856                 }
2857         }
2858
2859         if (!start) {
2860                 crit_exit_quick(td);
2861                 return error;
2862         }
2863
2864         ifsq_ifstart_try(ifsq, 0);
2865
2866         crit_exit_quick(td);
2867         return error;
2868 }
2869
2870 void *
2871 ifa_create(int size, int flags)
2872 {
2873         struct ifaddr *ifa;
2874         int i;
2875
2876         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
2877
2878         ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2879         if (ifa == NULL)
2880                 return NULL;
2881
2882         ifa->ifa_containers =
2883             kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
2884                 M_IFADDR, M_WAITOK | M_ZERO);
2885         ifa->ifa_ncnt = ncpus;
2886         for (i = 0; i < ncpus; ++i) {
2887                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2888
2889                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2890                 ifac->ifa = ifa;
2891                 ifac->ifa_refcnt = 1;
2892         }
2893 #ifdef IFADDR_DEBUG
2894         kprintf("alloc ifa %p %d\n", ifa, size);
2895 #endif
2896         return ifa;
2897 }
2898
2899 void
2900 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2901 {
2902         struct ifaddr *ifa = ifac->ifa;
2903
2904         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2905         KKASSERT(ifac->ifa_refcnt == 0);
2906         KASSERT(ifac->ifa_listmask == 0,
2907                 ("ifa is still on %#x lists", ifac->ifa_listmask));
2908
2909         ifac->ifa_magic = IFA_CONTAINER_DEAD;
2910
2911 #ifdef IFADDR_DEBUG_VERBOSE
2912         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2913 #endif
2914
2915         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2916                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
2917         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2918 #ifdef IFADDR_DEBUG
2919                 kprintf("free ifa %p\n", ifa);
2920 #endif
2921                 kfree(ifa->ifa_containers, M_IFADDR);
2922                 kfree(ifa, M_IFADDR);
2923         }
2924 }
2925
2926 static void
2927 ifa_iflink_dispatch(netmsg_t nmsg)
2928 {
2929         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2930         struct ifaddr *ifa = msg->ifa;
2931         struct ifnet *ifp = msg->ifp;
2932         int cpu = mycpuid;
2933         struct ifaddr_container *ifac;
2934
2935         crit_enter();
2936
2937         ifac = &ifa->ifa_containers[cpu];
2938         ASSERT_IFAC_VALID(ifac);
2939         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2940                 ("ifaddr is on if_addrheads"));
2941
2942         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2943         if (msg->tail)
2944                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2945         else
2946                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2947
2948         crit_exit();
2949
2950         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2951 }
2952
2953 void
2954 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2955 {
2956         struct netmsg_ifaddr msg;
2957
2958         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2959                     0, ifa_iflink_dispatch);
2960         msg.ifa = ifa;
2961         msg.ifp = ifp;
2962         msg.tail = tail;
2963
2964         ifa_domsg(&msg.base.lmsg, 0);
2965 }
2966
2967 static void
2968 ifa_ifunlink_dispatch(netmsg_t nmsg)
2969 {
2970         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2971         struct ifaddr *ifa = msg->ifa;
2972         struct ifnet *ifp = msg->ifp;
2973         int cpu = mycpuid;
2974         struct ifaddr_container *ifac;
2975
2976         crit_enter();
2977
2978         ifac = &ifa->ifa_containers[cpu];
2979         ASSERT_IFAC_VALID(ifac);
2980         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2981                 ("ifaddr is not on if_addrhead"));
2982
2983         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2984         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2985
2986         crit_exit();
2987
2988         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2989 }
2990
2991 void
2992 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2993 {
2994         struct netmsg_ifaddr msg;
2995
2996         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2997                     0, ifa_ifunlink_dispatch);
2998         msg.ifa = ifa;
2999         msg.ifp = ifp;
3000
3001         ifa_domsg(&msg.base.lmsg, 0);
3002 }
3003
3004 static void
3005 ifa_destroy_dispatch(netmsg_t nmsg)
3006 {
3007         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3008
3009         IFAFREE(msg->ifa);
3010         ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3011 }
3012
3013 void
3014 ifa_destroy(struct ifaddr *ifa)
3015 {
3016         struct netmsg_ifaddr msg;
3017
3018         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3019                     0, ifa_destroy_dispatch);
3020         msg.ifa = ifa;
3021
3022         ifa_domsg(&msg.base.lmsg, 0);
3023 }
3024
3025 struct lwkt_port *
3026 ifnet_portfn(int cpu)
3027 {
3028         return &ifnet_threads[cpu].td_msgport;
3029 }
3030
3031 void
3032 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
3033 {
3034         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
3035
3036         if (next_cpu < ncpus)
3037                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
3038         else
3039                 lwkt_replymsg(lmsg, 0);
3040 }
3041
3042 int
3043 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
3044 {
3045         KKASSERT(cpu < ncpus);
3046         return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
3047 }
3048
3049 void
3050 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
3051 {
3052         KKASSERT(cpu < ncpus);
3053         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
3054 }
3055
3056 /*
3057  * Generic netmsg service loop.  Some protocols may roll their own but all
3058  * must do the basic command dispatch function call done here.
3059  */
3060 static void
3061 ifnet_service_loop(void *arg __unused)
3062 {
3063         netmsg_t msg;
3064
3065         while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
3066                 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
3067                 msg->base.nm_dispatch(msg);
3068         }
3069 }
3070
3071 static void
3072 if_start_rollup(void)
3073 {
3074         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3075         struct ifsubq_stage *stage;
3076
3077         crit_enter();
3078
3079         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3080                 struct ifaltq_subque *ifsq = stage->stg_subq;
3081                 int is_sched = 0;
3082
3083                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3084                         is_sched = 1;
3085                 ifsq_stage_remove(head, stage);
3086
3087                 if (is_sched) {
3088                         ifsq_ifstart_schedule(ifsq, 1);
3089                 } else {
3090                         int start = 0;
3091
3092                         ALTQ_SQ_LOCK(ifsq);
3093                         if (!ifsq_is_started(ifsq)) {
3094                                 /*
3095                                  * Hold the subqueue interlock of
3096                                  * ifnet.if_start
3097                                  */
3098                                 ifsq_set_started(ifsq);
3099                                 start = 1;
3100                         }
3101                         ALTQ_SQ_UNLOCK(ifsq);
3102
3103                         if (start)
3104                                 ifsq_ifstart_try(ifsq, 1);
3105                 }
3106                 KKASSERT((stage->stg_flags &
3107                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3108         }
3109
3110         crit_exit();
3111 }
3112
3113 static void
3114 ifnetinit(void *dummy __unused)
3115 {
3116         int i;
3117
3118         for (i = 0; i < ncpus; ++i) {
3119                 struct thread *thr = &ifnet_threads[i];
3120
3121                 lwkt_create(ifnet_service_loop, NULL, NULL,
3122                             thr, TDF_NOSTART|TDF_FORCE_SPINPORT|TDF_FIXEDCPU,
3123                             i, "ifnet %d", i);
3124                 netmsg_service_port_init(&thr->td_msgport);
3125                 lwkt_schedule(thr);
3126         }
3127
3128         for (i = 0; i < ncpus; ++i)
3129                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3130         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3131 }
3132
3133 struct ifnet *
3134 ifnet_byindex(unsigned short idx)
3135 {
3136         if (idx > if_index)
3137                 return NULL;
3138         return ifindex2ifnet[idx];
3139 }
3140
3141 struct ifaddr *
3142 ifaddr_byindex(unsigned short idx)
3143 {
3144         struct ifnet *ifp;
3145
3146         ifp = ifnet_byindex(idx);
3147         if (!ifp)
3148                 return NULL;
3149         return TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
3150 }
3151
3152 void
3153 if_register_com_alloc(u_char type,
3154     if_com_alloc_t *a, if_com_free_t *f)
3155 {
3156
3157         KASSERT(if_com_alloc[type] == NULL,
3158             ("if_register_com_alloc: %d already registered", type));
3159         KASSERT(if_com_free[type] == NULL,
3160             ("if_register_com_alloc: %d free already registered", type));
3161
3162         if_com_alloc[type] = a;
3163         if_com_free[type] = f;
3164 }
3165
3166 void
3167 if_deregister_com_alloc(u_char type)
3168 {
3169
3170         KASSERT(if_com_alloc[type] != NULL,
3171             ("if_deregister_com_alloc: %d not registered", type));
3172         KASSERT(if_com_free[type] != NULL,
3173             ("if_deregister_com_alloc: %d free not registered", type));
3174         if_com_alloc[type] = NULL;
3175         if_com_free[type] = NULL;
3176 }
3177
3178 int
3179 if_ring_count2(int cnt, int cnt_max)
3180 {
3181         int shift = 0;
3182
3183         KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3184             ("invalid ring count max %d", cnt_max));
3185
3186         if (cnt <= 0)
3187                 cnt = cnt_max;
3188         if (cnt > ncpus2)
3189                 cnt = ncpus2;
3190         if (cnt > cnt_max)
3191                 cnt = cnt_max;
3192
3193         while ((1 << (shift + 1)) <= cnt)
3194                 ++shift;
3195         cnt = 1 << shift;
3196
3197         KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3198             ("calculate cnt %d, ncpus2 %d, cnt max %d",
3199              cnt, ncpus2, cnt_max));
3200         return cnt;
3201 }
3202
3203 void
3204 ifq_set_maxlen(struct ifaltq *ifq, int len)
3205 {
3206         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3207 }
3208
3209 int
3210 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3211 {
3212         return ALTQ_SUBQ_INDEX_DEFAULT;
3213 }
3214
3215 int
3216 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3217 {
3218         return (cpuid & ifq->altq_subq_mask);
3219 }
3220
3221 static void
3222 ifsq_watchdog(void *arg)
3223 {
3224         struct ifsubq_watchdog *wd = arg;
3225         struct ifnet *ifp;
3226
3227         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3228                 goto done;
3229
3230         ifp = ifsq_get_ifp(wd->wd_subq);
3231         if (ifnet_tryserialize_all(ifp)) {
3232                 wd->wd_watchdog(wd->wd_subq);
3233                 ifnet_deserialize_all(ifp);
3234         } else {
3235                 /* try again next timeout */
3236                 wd->wd_timer = 1;
3237         }
3238 done:
3239         ifsq_watchdog_reset(wd);
3240 }
3241
3242 static void
3243 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3244 {
3245         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3246             ifsq_get_cpuid(wd->wd_subq));
3247 }
3248
3249 void
3250 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3251     ifsq_watchdog_t watchdog)
3252 {
3253         callout_init_mp(&wd->wd_callout);
3254         wd->wd_timer = 0;
3255         wd->wd_subq = ifsq;
3256         wd->wd_watchdog = watchdog;
3257 }
3258
3259 void
3260 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3261 {
3262         wd->wd_timer = 0;
3263         ifsq_watchdog_reset(wd);
3264 }
3265
3266 void
3267 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3268 {
3269         wd->wd_timer = 0;
3270         callout_stop(&wd->wd_callout);
3271 }