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 static void     if_slowtimo_dispatch(netmsg_t);
 117
 118 /* Helper functions */
 119 static void     ifsq_watchdog_reset(struct ifsubq_watchdog *);
 120 static int      if_delmulti_serialized(struct ifnet *, struct sockaddr *);
 121 static struct ifnet_array *ifnet_array_alloc(int);
 122 static void     ifnet_array_free(struct ifnet_array *);
 123 static struct ifnet_array *ifnet_array_add(struct ifnet *,
 124                     const struct ifnet_array *);
 125 static struct ifnet_array *ifnet_array_del(struct ifnet *,
 126                     const struct ifnet_array *);
 127
 128 #ifdef INET6
 129 /*
 130  * XXX: declare here to avoid to include many inet6 related files..
 131  * should be more generalized?
 132  */
 133 extern void     nd6_setmtu(struct ifnet *);
 134 #endif
 135
 136 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
 137 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
 138
 139 static int ifsq_stage_cntmax = 4;
 140 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
 141 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
 142     &ifsq_stage_cntmax, 0, "ifq staging packet count max");
 143
 144 static int if_stats_compat = 0;
 145 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
 146     &if_stats_compat, 0, "Compat the old ifnet stats");
 147
 148 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL);
 149 /* Must be after netisr_init */
 150 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL);
 151
 152 static  if_com_alloc_t *if_com_alloc[256];
 153 static  if_com_free_t *if_com_free[256];
 154
 155 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
 156 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
 157 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
 158
 159 int                     ifqmaxlen = IFQ_MAXLEN;
 160 struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
 161
 162 static struct ifnet_array       ifnet_array0;
 163 static struct ifnet_array       *ifnet_array = &ifnet_array0;
 164
 165 static struct callout           if_slowtimo_timer;
 166 static struct netmsg_base       if_slowtimo_netmsg;
 167
 168 int                     if_index = 0;
 169 struct ifnet            **ifindex2ifnet = NULL;
 170 static struct thread    ifnet_threads[MAXCPU];
 171 static struct mtx       ifnet_mtx = MTX_INITIALIZER("ifnet");
 172
 173 static struct ifsubq_stage_head ifsubq_stage_heads[MAXCPU];
 174
 175 #ifdef notyet
 176 #define IFQ_KTR_STRING          "ifq=%p"
 177 #define IFQ_KTR_ARGS    struct ifaltq *ifq
 178 #ifndef KTR_IFQ
 179 #define KTR_IFQ                 KTR_ALL
 180 #endif
 181 KTR_INFO_MASTER(ifq);
 182 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
 183 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
 184 #define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)
 185
 186 #define IF_START_KTR_STRING     "ifp=%p"
 187 #define IF_START_KTR_ARGS       struct ifnet *ifp
 188 #ifndef KTR_IF_START
 189 #define KTR_IF_START            KTR_ALL
 190 #endif
 191 KTR_INFO_MASTER(if_start);
 192 KTR_INFO(KTR_IF_START, if_start, run, 0,
 193          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 194 KTR_INFO(KTR_IF_START, if_start, sched, 1,
 195          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 196 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
 197          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 198 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
 199          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 200 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
 201          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 202 #define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
 203 #endif
 204
 205 TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
 206
 207 /*
 208  * Network interface utility routines.
 209  *
 210  * Routines with ifa_ifwith* names take sockaddr *'s as
 211  * parameters.
 212  */
 213 /* ARGSUSED*/
 214 static void
 215 ifinit(void *dummy)
 216 {
 217         struct ifnet *ifp;
 218
 219         callout_init_mp(&if_slowtimo_timer);
 220         netmsg_init(&if_slowtimo_netmsg, NULL, &netisr_adone_rport,
 221             MSGF_PRIORITY, if_slowtimo_dispatch);
 222
 223         /* XXX is this necessary? */
 224         ifnet_lock();
 225         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
 226                 if (ifp->if_snd.altq_maxlen == 0) {
 227                         if_printf(ifp, "XXX: driver didn't set altq_maxlen\n");
 228                         ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
 229                 }
 230         }
 231         ifnet_unlock();
 232
 233         /* Start if_slowtimo */
 234         lwkt_sendmsg(netisr_cpuport(0), &if_slowtimo_netmsg.lmsg);
 235 }
 236
 237 static void
 238 ifsq_ifstart_ipifunc(void *arg)
 239 {
 240         struct ifaltq_subque *ifsq = arg;
 241         struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
 242
 243         crit_enter();
 244         if (lmsg->ms_flags & MSGF_DONE)
 245                 lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg);
 246         crit_exit();
 247 }
 248
 249 static __inline void
 250 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
 251 {
 252         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
 253         TAILQ_REMOVE(&head->stg_head, stage, stg_link);
 254         stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
 255         stage->stg_cnt = 0;
 256         stage->stg_len = 0;
 257 }
 258
 259 static __inline void
 260 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
 261 {
 262         KKASSERT((stage->stg_flags &
 263             (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
 264         stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
 265         TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
 266 }
 267
 268 /*
 269  * Schedule ifnet.if_start on the subqueue owner CPU
 270  */
 271 static void
 272 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
 273 {
 274         int cpu;
 275
 276         if (!force && curthread->td_type == TD_TYPE_NETISR &&
 277             ifsq_stage_cntmax > 0) {
 278                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 279
 280                 stage->stg_cnt = 0;
 281                 stage->stg_len = 0;
 282                 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
 283                         ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
 284                 stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
 285                 return;
 286         }
 287
 288         cpu = ifsq_get_cpuid(ifsq);
 289         if (cpu != mycpuid)
 290                 lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
 291         else
 292                 ifsq_ifstart_ipifunc(ifsq);
 293 }
 294
 295 /*
 296  * NOTE:
 297  * This function will release ifnet.if_start subqueue interlock,
 298  * if ifnet.if_start for the subqueue does not need to be scheduled
 299  */
 300 static __inline int
 301 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
 302 {
 303         if (!running || ifsq_is_empty(ifsq)
 304 #ifdef ALTQ
 305             || ifsq->ifsq_altq->altq_tbr != NULL
 306 #endif
 307         ) {
 308                 ALTQ_SQ_LOCK(ifsq);
 309                 /*
 310                  * ifnet.if_start subqueue interlock is released, if:
 311                  * 1) Hardware can not take any packets, due to
 312                  *    o  interface is marked down
 313                  *    o  hardware queue is full (ifsq_is_oactive)
 314                  *    Under the second situation, hardware interrupt
 315                  *    or polling(4) will call/schedule ifnet.if_start
 316                  *    on the subqueue when hardware queue is ready
 317                  * 2) There is no packet in the subqueue.
 318                  *    Further ifq_dispatch or ifq_handoff will call/
 319                  *    schedule ifnet.if_start on the subqueue.
 320                  * 3) TBR is used and it does not allow further
 321                  *    dequeueing.
 322                  *    TBR callout will call ifnet.if_start on the
 323                  *    subqueue.
 324                  */
 325                 if (!running || !ifsq_data_ready(ifsq)) {
 326                         ifsq_clr_started(ifsq);
 327                         ALTQ_SQ_UNLOCK(ifsq);
 328                         return 0;
 329                 }
 330                 ALTQ_SQ_UNLOCK(ifsq);
 331         }
 332         return 1;
 333 }
 334
 335 static void
 336 ifsq_ifstart_dispatch(netmsg_t msg)
 337 {
 338         struct lwkt_msg *lmsg = &msg->base.lmsg;
 339         struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
 340         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 341         struct globaldata *gd = mycpu;
 342         int running = 0, need_sched;
 343
 344         crit_enter_gd(gd);
 345
 346         lwkt_replymsg(lmsg, 0); /* reply ASAP */
 347
 348         if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) {
 349                 /*
 350                  * We need to chase the subqueue owner CPU change.
 351                  */
 352                 ifsq_ifstart_schedule(ifsq, 1);
 353                 crit_exit_gd(gd);
 354                 return;
 355         }
 356
 357         ifsq_serialize_hw(ifsq);
 358         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
 359                 ifp->if_start(ifp, ifsq);
 360                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 361                         running = 1;
 362         }
 363         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
 364         ifsq_deserialize_hw(ifsq);
 365
 366         if (need_sched) {
 367                 /*
 368                  * More data need to be transmitted, ifnet.if_start is
 369                  * scheduled on the subqueue owner CPU, and we keep going.
 370                  * NOTE: ifnet.if_start subqueue interlock is not released.
 371                  */
 372                 ifsq_ifstart_schedule(ifsq, 0);
 373         }
 374
 375         crit_exit_gd(gd);
 376 }
 377
 378 /* Device driver ifnet.if_start helper function */
 379 void
 380 ifsq_devstart(struct ifaltq_subque *ifsq)
 381 {
 382         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 383         int running = 0;
 384
 385         ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
 386
 387         ALTQ_SQ_LOCK(ifsq);
 388         if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
 389                 ALTQ_SQ_UNLOCK(ifsq);
 390                 return;
 391         }
 392         ifsq_set_started(ifsq);
 393         ALTQ_SQ_UNLOCK(ifsq);
 394
 395         ifp->if_start(ifp, ifsq);
 396
 397         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 398                 running = 1;
 399
 400         if (ifsq_ifstart_need_schedule(ifsq, running)) {
 401                 /*
 402                  * More data need to be transmitted, ifnet.if_start is
 403                  * scheduled on ifnet's CPU, and we keep going.
 404                  * NOTE: ifnet.if_start interlock is not released.
 405                  */
 406                 ifsq_ifstart_schedule(ifsq, 0);
 407         }
 408 }
 409
 410 void
 411 if_devstart(struct ifnet *ifp)
 412 {
 413         ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
 414 }
 415
 416 /* Device driver ifnet.if_start schedule helper function */
 417 void
 418 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
 419 {
 420         ifsq_ifstart_schedule(ifsq, 1);
 421 }
 422
 423 void
 424 if_devstart_sched(struct ifnet *ifp)
 425 {
 426         ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
 427 }
 428
 429 static void
 430 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 431 {
 432         lwkt_serialize_enter(ifp->if_serializer);
 433 }
 434
 435 static void
 436 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 437 {
 438         lwkt_serialize_exit(ifp->if_serializer);
 439 }
 440
 441 static int
 442 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 443 {
 444         return lwkt_serialize_try(ifp->if_serializer);
 445 }
 446
 447 #ifdef INVARIANTS
 448 static void
 449 if_default_serialize_assert(struct ifnet *ifp,
 450                             enum ifnet_serialize slz __unused,
 451                             boolean_t serialized)
 452 {
 453         if (serialized)
 454                 ASSERT_SERIALIZED(ifp->if_serializer);
 455         else
 456                 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
 457 }
 458 #endif
 459
 460 /*
 461  * Attach an interface to the list of "active" interfaces.
 462  *
 463  * The serializer is optional.
 464  */
 465 void
 466 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
 467 {
 468         unsigned socksize;
 469         int namelen, masklen;
 470         struct sockaddr_dl *sdl, *sdl_addr;
 471         struct ifaddr *ifa;
 472         struct ifaltq *ifq;
 473         struct ifnet **old_ifindex2ifnet = NULL;
 474         struct ifnet_array *old_ifnet_array;
 475         int i, q;
 476
 477         static int if_indexlim = 8;
 478
 479         if (ifp->if_serialize != NULL) {
 480                 KASSERT(ifp->if_deserialize != NULL &&
 481                         ifp->if_tryserialize != NULL &&
 482                         ifp->if_serialize_assert != NULL,
 483                         ("serialize functions are partially setup"));
 484
 485                 /*
 486                  * If the device supplies serialize functions,
 487                  * then clear if_serializer to catch any invalid
 488                  * usage of this field.
 489                  */
 490                 KASSERT(serializer == NULL,
 491                         ("both serialize functions and default serializer "
 492                          "are supplied"));
 493                 ifp->if_serializer = NULL;
 494         } else {
 495                 KASSERT(ifp->if_deserialize == NULL &&
 496                         ifp->if_tryserialize == NULL &&
 497                         ifp->if_serialize_assert == NULL,
 498                         ("serialize functions are partially setup"));
 499                 ifp->if_serialize = if_default_serialize;
 500                 ifp->if_deserialize = if_default_deserialize;
 501                 ifp->if_tryserialize = if_default_tryserialize;
 502 #ifdef INVARIANTS
 503                 ifp->if_serialize_assert = if_default_serialize_assert;
 504 #endif
 505
 506                 /*
 507                  * The serializer can be passed in from the device,
 508                  * allowing the same serializer to be used for both
 509                  * the interrupt interlock and the device queue.
 510                  * If not specified, the netif structure will use an
 511                  * embedded serializer.
 512                  */
 513                 if (serializer == NULL) {
 514                         serializer = &ifp->if_default_serializer;
 515                         lwkt_serialize_init(serializer);
 516                 }
 517                 ifp->if_serializer = serializer;
 518         }
 519
 520         /*
 521          * XXX -
 522          * The old code would work if the interface passed a pre-existing
 523          * chain of ifaddrs to this code.  We don't trust our callers to
 524          * properly initialize the tailq, however, so we no longer allow
 525          * this unlikely case.
 526          */
 527         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
 528                                     M_IFADDR, M_WAITOK | M_ZERO);
 529         for (i = 0; i < ncpus; ++i)
 530                 TAILQ_INIT(&ifp->if_addrheads[i]);
 531
 532         TAILQ_INIT(&ifp->if_multiaddrs);
 533         TAILQ_INIT(&ifp->if_groups);
 534         getmicrotime(&ifp->if_lastchange);
 535
 536         /*
 537          * create a Link Level name for this device
 538          */
 539         namelen = strlen(ifp->if_xname);
 540         masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
 541         socksize = masklen + ifp->if_addrlen;
 542         if (socksize < sizeof(*sdl))
 543                 socksize = sizeof(*sdl);
 544         socksize = RT_ROUNDUP(socksize);
 545         ifa = ifa_create(sizeof(struct ifaddr) + 2 * socksize);
 546         sdl = sdl_addr = (struct sockaddr_dl *)(ifa + 1);
 547         sdl->sdl_len = socksize;
 548         sdl->sdl_family = AF_LINK;
 549         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
 550         sdl->sdl_nlen = namelen;
 551         sdl->sdl_type = ifp->if_type;
 552         ifp->if_lladdr = ifa;
 553         ifa->ifa_ifp = ifp;
 554         ifa->ifa_rtrequest = link_rtrequest;
 555         ifa->ifa_addr = (struct sockaddr *)sdl;
 556         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
 557         ifa->ifa_netmask = (struct sockaddr *)sdl;
 558         sdl->sdl_len = masklen;
 559         while (namelen != 0)
 560                 sdl->sdl_data[--namelen] = 0xff;
 561         ifa_iflink(ifa, ifp, 0 /* Insert head */);
 562
 563         ifp->if_data_pcpu = kmalloc_cachealign(
 564             ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
 565
 566         if (ifp->if_mapsubq == NULL)
 567                 ifp->if_mapsubq = ifq_mapsubq_default;
 568
 569         ifq = &ifp->if_snd;
 570         ifq->altq_type = 0;
 571         ifq->altq_disc = NULL;
 572         ifq->altq_flags &= ALTQF_CANTCHANGE;
 573         ifq->altq_tbr = NULL;
 574         ifq->altq_ifp = ifp;
 575
 576         if (ifq->altq_subq_cnt <= 0)
 577                 ifq->altq_subq_cnt = 1;
 578         ifq->altq_subq = kmalloc_cachealign(
 579             ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
 580             M_DEVBUF, M_WAITOK | M_ZERO);
 581
 582         if (ifq->altq_maxlen == 0) {
 583                 if_printf(ifp, "driver didn't set altq_maxlen\n");
 584                 ifq_set_maxlen(ifq, ifqmaxlen);
 585         }
 586
 587         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 588                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 589
 590                 ALTQ_SQ_LOCK_INIT(ifsq);
 591                 ifsq->ifsq_index = q;
 592
 593                 ifsq->ifsq_altq = ifq;
 594                 ifsq->ifsq_ifp = ifp;
 595
 596                 ifsq->ifsq_maxlen = ifq->altq_maxlen;
 597                 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
 598                 ifsq->ifsq_prepended = NULL;
 599                 ifsq->ifsq_started = 0;
 600                 ifsq->ifsq_hw_oactive = 0;
 601                 ifsq_set_cpuid(ifsq, 0);
 602                 if (ifp->if_serializer != NULL)
 603                         ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
 604
 605                 ifsq->ifsq_stage =
 606                     kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
 607                     M_DEVBUF, M_WAITOK | M_ZERO);
 608                 for (i = 0; i < ncpus; ++i)
 609                         ifsq->ifsq_stage[i].stg_subq = ifsq;
 610
 611                 ifsq->ifsq_ifstart_nmsg =
 612                     kmalloc(ncpus * sizeof(struct netmsg_base),
 613                     M_LWKTMSG, M_WAITOK);
 614                 for (i = 0; i < ncpus; ++i) {
 615                         netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
 616                             &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
 617                         ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
 618                 }
 619         }
 620         ifq_set_classic(ifq);
 621
 622         /*
 623          * Increase mbuf cluster/jcluster limits for the mbufs that
 624          * could sit on the device queues for quite some time.
 625          */
 626         if (ifp->if_nmbclusters > 0)
 627                 mcl_inclimit(ifp->if_nmbclusters);
 628         if (ifp->if_nmbjclusters > 0)
 629                 mjcl_inclimit(ifp->if_nmbjclusters);
 630
 631         /*
 632          * Install this ifp into ifindex2inet, ifnet queue and ifnet
 633          * array after it is setup.
 634          *
 635          * Protect ifindex2ifnet, ifnet queue and ifnet array changes
 636          * by ifnet lock, so that non-netisr threads could get a
 637          * consistent view.
 638          */
 639         ifnet_lock();
 640
 641         /* Don't update if_index until ifindex2ifnet is setup */
 642         ifp->if_index = if_index + 1;
 643         sdl_addr->sdl_index = ifp->if_index;
 644
 645         /*
 646          * Install this ifp into ifindex2ifnet
 647          */
 648         if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
 649                 unsigned int n;
 650                 struct ifnet **q;
 651
 652                 /*
 653                  * Grow ifindex2ifnet
 654                  */
 655                 if_indexlim <<= 1;
 656                 n = if_indexlim * sizeof(*q);
 657                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
 658                 if (ifindex2ifnet != NULL) {
 659                         bcopy(ifindex2ifnet, q, n/2);
 660                         /* Free old ifindex2ifnet after sync all netisrs */
 661                         old_ifindex2ifnet = ifindex2ifnet;
 662                 }
 663                 ifindex2ifnet = q;
 664         }
 665         ifindex2ifnet[ifp->if_index] = ifp;
 666         /*
 667          * Update if_index after this ifp is installed into ifindex2ifnet,
 668          * so that netisrs could get a consistent view of ifindex2ifnet.
 669          */
 670         cpu_sfence();
 671         if_index = ifp->if_index;
 672
 673         /*
 674          * Install this ifp into ifnet array.
 675          */
 676         /* Free old ifnet array after sync all netisrs */
 677         old_ifnet_array = ifnet_array;
 678         ifnet_array = ifnet_array_add(ifp, old_ifnet_array);
 679
 680         /*
 681          * Install this ifp into ifnet queue.
 682          */
 683         TAILQ_INSERT_TAIL(&ifnetlist, ifp, if_link);
 684
 685         ifnet_unlock();
 686
 687         /*
 688          * Sync all netisrs so that the old ifindex2ifnet and ifnet array
 689          * are no longer accessed and we can free them safely later on.
 690          */
 691         netmsg_service_sync();
 692         if (old_ifindex2ifnet != NULL)
 693                 kfree(old_ifindex2ifnet, M_IFADDR);
 694         ifnet_array_free(old_ifnet_array);
 695
 696         if (!SLIST_EMPTY(&domains))
 697                 if_attachdomain1(ifp);
 698
 699         /* Announce the interface. */
 700         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
 701         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
 702         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
 703 }
 704
 705 static void
 706 if_attachdomain(void *dummy)
 707 {
 708         struct ifnet *ifp;
 709
 710         ifnet_lock();
 711         TAILQ_FOREACH(ifp, &ifnetlist, if_list)
 712                 if_attachdomain1(ifp);
 713         ifnet_unlock();
 714 }
 715 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
 716         if_attachdomain, NULL);
 717
 718 static void
 719 if_attachdomain1(struct ifnet *ifp)
 720 {
 721         struct domain *dp;
 722
 723         crit_enter();
 724
 725         /* address family dependent data region */
 726         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
 727         SLIST_FOREACH(dp, &domains, dom_next)
 728                 if (dp->dom_ifattach)
 729                         ifp->if_afdata[dp->dom_family] =
 730                                 (*dp->dom_ifattach)(ifp);
 731         crit_exit();
 732 }
 733
 734 /*
 735  * Purge all addresses whose type is _not_ AF_LINK
 736  */
 737 static void
 738 if_purgeaddrs_nolink_dispatch(netmsg_t nmsg)
 739 {
 740         struct lwkt_msg *lmsg = &nmsg->lmsg;
 741         struct ifnet *ifp = lmsg->u.ms_resultp;
 742         struct ifaddr_container *ifac, *next;
 743
 744         ASSERT_IN_NETISR(0);
 745
 746         /*
 747          * The ifaddr processing in the following loop will block,
 748          * however, this function is called in netisr0, in which
 749          * ifaddr list changes happen, so we don't care about the
 750          * blockness of the ifaddr processing here.
 751          */
 752         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
 753                               ifa_link, next) {
 754                 struct ifaddr *ifa = ifac->ifa;
 755
 756                 /* Ignore marker */
 757                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
 758                         continue;
 759
 760                 /* Leave link ifaddr as it is */
 761                 if (ifa->ifa_addr->sa_family == AF_LINK)
 762                         continue;
 763 #ifdef INET
 764                 /* XXX: Ugly!! ad hoc just for INET */
 765                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
 766                         struct ifaliasreq ifr;
 767 #ifdef IFADDR_DEBUG_VERBOSE
 768                         int i;
 769
 770                         kprintf("purge in4 addr %p: ", ifa);
 771                         for (i = 0; i < ncpus; ++i)
 772                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 773                         kprintf("\n");
 774 #endif
 775
 776                         bzero(&ifr, sizeof ifr);
 777                         ifr.ifra_addr = *ifa->ifa_addr;
 778                         if (ifa->ifa_dstaddr)
 779                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
 780                         if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp,
 781                                        NULL) == 0)
 782                                 continue;
 783                 }
 784 #endif /* INET */
 785 #ifdef INET6
 786                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
 787 #ifdef IFADDR_DEBUG_VERBOSE
 788                         int i;
 789
 790                         kprintf("purge in6 addr %p: ", ifa);
 791                         for (i = 0; i < ncpus; ++i)
 792                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 793                         kprintf("\n");
 794 #endif
 795
 796                         in6_purgeaddr(ifa);
 797                         /* ifp_addrhead is already updated */
 798                         continue;
 799                 }
 800 #endif /* INET6 */
 801                 ifa_ifunlink(ifa, ifp);
 802                 ifa_destroy(ifa);
 803         }
 804
 805         lwkt_replymsg(lmsg, 0);
 806 }
 807
 808 void
 809 if_purgeaddrs_nolink(struct ifnet *ifp)
 810 {
 811         struct netmsg_base nmsg;
 812         struct lwkt_msg *lmsg = &nmsg.lmsg;
 813
 814         ASSERT_CANDOMSG_NETISR0(curthread);
 815
 816         netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
 817             if_purgeaddrs_nolink_dispatch);
 818         lmsg->u.ms_resultp = ifp;
 819         lwkt_domsg(netisr_cpuport(0), lmsg, 0);
 820 }
 821
 822 static void
 823 ifq_stage_detach_handler(netmsg_t nmsg)
 824 {
 825         struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
 826         int q;
 827
 828         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 829                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 830                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 831
 832                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
 833                         ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
 834         }
 835         lwkt_replymsg(&nmsg->lmsg, 0);
 836 }
 837
 838 static void
 839 ifq_stage_detach(struct ifaltq *ifq)
 840 {
 841         struct netmsg_base base;
 842         int cpu;
 843
 844         netmsg_init(&base, NULL, &curthread->td_msgport, 0,
 845             ifq_stage_detach_handler);
 846         base.lmsg.u.ms_resultp = ifq;
 847
 848         for (cpu = 0; cpu < ncpus; ++cpu)
 849                 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
 850 }
 851
 852 struct netmsg_if_rtdel {
 853         struct netmsg_base      base;
 854         struct ifnet            *ifp;
 855 };
 856
 857 static void
 858 if_rtdel_dispatch(netmsg_t msg)
 859 {
 860         struct netmsg_if_rtdel *rmsg = (void *)msg;
 861         int i, nextcpu, cpu;
 862
 863         cpu = mycpuid;
 864         for (i = 1; i <= AF_MAX; i++) {
 865                 struct radix_node_head  *rnh;
 866
 867                 if ((rnh = rt_tables[cpu][i]) == NULL)
 868                         continue;
 869                 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
 870         }
 871
 872         nextcpu = cpu + 1;
 873         if (nextcpu < ncpus)
 874                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg);
 875         else
 876                 lwkt_replymsg(&rmsg->base.lmsg, 0);
 877 }
 878
 879 /*
 880  * Detach an interface, removing it from the
 881  * list of "active" interfaces.
 882  */
 883 void
 884 if_detach(struct ifnet *ifp)
 885 {
 886         struct ifnet_array *old_ifnet_array;
 887         struct netmsg_if_rtdel msg;
 888         struct domain *dp;
 889         int q;
 890
 891         /* Announce that the interface is gone. */
 892         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
 893         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
 894         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
 895
 896         /*
 897          * Remove this ifp from ifindex2inet, ifnet queue and ifnet
 898          * array before it is whacked.
 899          *
 900          * Protect ifindex2ifnet, ifnet queue and ifnet array changes
 901          * by ifnet lock, so that non-netisr threads could get a
 902          * consistent view.
 903          */
 904         ifnet_lock();
 905
 906         /*
 907          * Remove this ifp from ifindex2ifnet and maybe decrement if_index.
 908          */
 909         ifindex2ifnet[ifp->if_index] = NULL;
 910         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
 911                 if_index--;
 912
 913         /*
 914          * Remove this ifp from ifnet queue.
 915          */
 916         TAILQ_REMOVE(&ifnetlist, ifp, if_link);
 917
 918         /*
 919          * Remove this ifp from ifnet array.
 920          */
 921         /* Free old ifnet array after sync all netisrs */
 922         old_ifnet_array = ifnet_array;
 923         ifnet_array = ifnet_array_del(ifp, old_ifnet_array);
 924
 925         ifnet_unlock();
 926
 927         /*
 928          * Sync all netisrs so that the old ifnet array is no longer
 929          * accessed and we can free it safely later on.
 930          */
 931         netmsg_service_sync();
 932         ifnet_array_free(old_ifnet_array);
 933
 934         /*
 935          * Remove routes and flush queues.
 936          */
 937         crit_enter();
 938 #ifdef IFPOLL_ENABLE
 939         if (ifp->if_flags & IFF_NPOLLING)
 940                 ifpoll_deregister(ifp);
 941 #endif
 942         if_down(ifp);
 943
 944         /* Decrease the mbuf clusters/jclusters limits increased by us */
 945         if (ifp->if_nmbclusters > 0)
 946                 mcl_inclimit(-ifp->if_nmbclusters);
 947         if (ifp->if_nmbjclusters > 0)
 948                 mjcl_inclimit(-ifp->if_nmbjclusters);
 949
 950 #ifdef ALTQ
 951         if (ifq_is_enabled(&ifp->if_snd))
 952                 altq_disable(&ifp->if_snd);
 953         if (ifq_is_attached(&ifp->if_snd))
 954                 altq_detach(&ifp->if_snd);
 955 #endif
 956
 957         /*
 958          * Clean up all addresses.
 959          */
 960         ifp->if_lladdr = NULL;
 961
 962         if_purgeaddrs_nolink(ifp);
 963         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
 964                 struct ifaddr *ifa;
 965
 966                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
 967                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
 968                         ("non-link ifaddr is left on if_addrheads"));
 969
 970                 ifa_ifunlink(ifa, ifp);
 971                 ifa_destroy(ifa);
 972                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
 973                         ("there are still ifaddrs left on if_addrheads"));
 974         }
 975
 976 #ifdef INET
 977         /*
 978          * Remove all IPv4 kernel structures related to ifp.
 979          */
 980         in_ifdetach(ifp);
 981 #endif
 982
 983 #ifdef INET6
 984         /*
 985          * Remove all IPv6 kernel structs related to ifp.  This should be done
 986          * before removing routing entries below, since IPv6 interface direct
 987          * routes are expected to be removed by the IPv6-specific kernel API.
 988          * Otherwise, the kernel will detect some inconsistency and bark it.
 989          */
 990         in6_ifdetach(ifp);
 991 #endif
 992
 993         /*
 994          * Delete all remaining routes using this interface
 995          */
 996         netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
 997             if_rtdel_dispatch);
 998         msg.ifp = ifp;
 999         rt_domsg_global(&msg.base);
1000
1001         SLIST_FOREACH(dp, &domains, dom_next)
1002                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
1003                         (*dp->dom_ifdetach)(ifp,
1004                                 ifp->if_afdata[dp->dom_family]);
1005
1006         kfree(ifp->if_addrheads, M_IFADDR);
1007
1008         lwkt_synchronize_ipiqs("if_detach");
1009         ifq_stage_detach(&ifp->if_snd);
1010
1011         for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
1012                 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
1013
1014                 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
1015                 kfree(ifsq->ifsq_stage, M_DEVBUF);
1016         }
1017         kfree(ifp->if_snd.altq_subq, M_DEVBUF);
1018
1019         kfree(ifp->if_data_pcpu, M_DEVBUF);
1020
1021         crit_exit();
1022 }
1023
1024 /*
1025  * Create interface group without members
1026  */
1027 struct ifg_group *
1028 if_creategroup(const char *groupname)
1029 {
1030         struct ifg_group        *ifg = NULL;
1031
1032         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
1033             M_TEMP, M_NOWAIT)) == NULL)
1034                 return (NULL);
1035
1036         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1037         ifg->ifg_refcnt = 0;
1038         ifg->ifg_carp_demoted = 0;
1039         TAILQ_INIT(&ifg->ifg_members);
1040 #if NPF > 0
1041         pfi_attach_ifgroup(ifg);
1042 #endif
1043         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
1044
1045         return (ifg);
1046 }
1047
1048 /*
1049  * Add a group to an interface
1050  */
1051 int
1052 if_addgroup(struct ifnet *ifp, const char *groupname)
1053 {
1054         struct ifg_list         *ifgl;
1055         struct ifg_group        *ifg = NULL;
1056         struct ifg_member       *ifgm;
1057
1058         if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1059             groupname[strlen(groupname) - 1] <= '9')
1060                 return (EINVAL);
1061
1062         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1063                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1064                         return (EEXIST);
1065
1066         if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
1067                 return (ENOMEM);
1068
1069         if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
1070                 kfree(ifgl, M_TEMP);
1071                 return (ENOMEM);
1072         }
1073
1074         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1075                 if (!strcmp(ifg->ifg_group, groupname))
1076                         break;
1077
1078         if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
1079                 kfree(ifgl, M_TEMP);
1080                 kfree(ifgm, M_TEMP);
1081                 return (ENOMEM);
1082         }
1083
1084         ifg->ifg_refcnt++;
1085         ifgl->ifgl_group = ifg;
1086         ifgm->ifgm_ifp = ifp;
1087
1088         TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1089         TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1090
1091 #if NPF > 0
1092         pfi_group_change(groupname);
1093 #endif
1094
1095         return (0);
1096 }
1097
1098 /*
1099  * Remove a group from an interface
1100  */
1101 int
1102 if_delgroup(struct ifnet *ifp, const char *groupname)
1103 {
1104         struct ifg_list         *ifgl;
1105         struct ifg_member       *ifgm;
1106
1107         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1108                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1109                         break;
1110         if (ifgl == NULL)
1111                 return (ENOENT);
1112
1113         TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1114
1115         TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1116                 if (ifgm->ifgm_ifp == ifp)
1117                         break;
1118
1119         if (ifgm != NULL) {
1120                 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1121                 kfree(ifgm, M_TEMP);
1122         }
1123
1124         if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1125                 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
1126 #if NPF > 0
1127                 pfi_detach_ifgroup(ifgl->ifgl_group);
1128 #endif
1129                 kfree(ifgl->ifgl_group, M_TEMP);
1130         }
1131
1132         kfree(ifgl, M_TEMP);
1133
1134 #if NPF > 0
1135         pfi_group_change(groupname);
1136 #endif
1137
1138         return (0);
1139 }
1140
1141 /*
1142  * Stores all groups from an interface in memory pointed
1143  * to by data
1144  */
1145 int
1146 if_getgroup(caddr_t data, struct ifnet *ifp)
1147 {
1148         int                      len, error;
1149         struct ifg_list         *ifgl;
1150         struct ifg_req           ifgrq, *ifgp;
1151         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1152
1153         if (ifgr->ifgr_len == 0) {
1154                 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1155                         ifgr->ifgr_len += sizeof(struct ifg_req);
1156                 return (0);
1157         }
1158
1159         len = ifgr->ifgr_len;
1160         ifgp = ifgr->ifgr_groups;
1161         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1162                 if (len < sizeof(ifgrq))
1163                         return (EINVAL);
1164                 bzero(&ifgrq, sizeof ifgrq);
1165                 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1166                     sizeof(ifgrq.ifgrq_group));
1167                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1168                     sizeof(struct ifg_req))))
1169                         return (error);
1170                 len -= sizeof(ifgrq);
1171                 ifgp++;
1172         }
1173
1174         return (0);
1175 }
1176
1177 /*
1178  * Stores all members of a group in memory pointed to by data
1179  */
1180 int
1181 if_getgroupmembers(caddr_t data)
1182 {
1183         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1184         struct ifg_group        *ifg;
1185         struct ifg_member       *ifgm;
1186         struct ifg_req           ifgrq, *ifgp;
1187         int                      len, error;
1188
1189         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1190                 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1191                         break;
1192         if (ifg == NULL)
1193                 return (ENOENT);
1194
1195         if (ifgr->ifgr_len == 0) {
1196                 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1197                         ifgr->ifgr_len += sizeof(ifgrq);
1198                 return (0);
1199         }
1200
1201         len = ifgr->ifgr_len;
1202         ifgp = ifgr->ifgr_groups;
1203         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1204                 if (len < sizeof(ifgrq))
1205                         return (EINVAL);
1206                 bzero(&ifgrq, sizeof ifgrq);
1207                 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1208                     sizeof(ifgrq.ifgrq_member));
1209                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1210                     sizeof(struct ifg_req))))
1211                         return (error);
1212                 len -= sizeof(ifgrq);
1213                 ifgp++;
1214         }
1215
1216         return (0);
1217 }
1218
1219 /*
1220  * Delete Routes for a Network Interface
1221  *
1222  * Called for each routing entry via the rnh->rnh_walktree() call above
1223  * to delete all route entries referencing a detaching network interface.
1224  *
1225  * Arguments:
1226  *      rn      pointer to node in the routing table
1227  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1228  *
1229  * Returns:
1230  *      0       successful
1231  *      errno   failed - reason indicated
1232  *
1233  */
1234 static int
1235 if_rtdel(struct radix_node *rn, void *arg)
1236 {
1237         struct rtentry  *rt = (struct rtentry *)rn;
1238         struct ifnet    *ifp = arg;
1239         int             err;
1240
1241         if (rt->rt_ifp == ifp) {
1242
1243                 /*
1244                  * Protect (sorta) against walktree recursion problems
1245                  * with cloned routes
1246                  */
1247                 if (!(rt->rt_flags & RTF_UP))
1248                         return (0);
1249
1250                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1251                                 rt_mask(rt), rt->rt_flags,
1252                                 NULL);
1253                 if (err) {
1254                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1255                 }
1256         }
1257
1258         return (0);
1259 }
1260
1261 static __inline boolean_t
1262 ifa_prefer(const struct ifaddr *cur_ifa, const struct ifaddr *old_ifa)
1263 {
1264         if (old_ifa == NULL)
1265                 return TRUE;
1266
1267         if ((old_ifa->ifa_ifp->if_flags & IFF_UP) == 0 &&
1268             (cur_ifa->ifa_ifp->if_flags & IFF_UP))
1269                 return TRUE;
1270         if ((old_ifa->ifa_flags & IFA_ROUTE) == 0 &&
1271             (cur_ifa->ifa_flags & IFA_ROUTE))
1272                 return TRUE;
1273         return FALSE;
1274 }
1275
1276 /*
1277  * Locate an interface based on a complete address.
1278  */
1279 struct ifaddr *
1280 ifa_ifwithaddr(struct sockaddr *addr)
1281 {
1282         const struct ifnet_array *arr;
1283         int i;
1284
1285         arr = ifnet_array_get();
1286         for (i = 0; i < arr->ifnet_count; ++i) {
1287                 struct ifnet *ifp = arr->ifnet_arr[i];
1288                 struct ifaddr_container *ifac;
1289
1290                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1291                         struct ifaddr *ifa = ifac->ifa;
1292
1293                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1294                                 continue;
1295                         if (sa_equal(addr, ifa->ifa_addr))
1296                                 return (ifa);
1297                         if ((ifp->if_flags & IFF_BROADCAST) &&
1298                             ifa->ifa_broadaddr &&
1299                             /* IPv6 doesn't have broadcast */
1300                             ifa->ifa_broadaddr->sa_len != 0 &&
1301                             sa_equal(ifa->ifa_broadaddr, addr))
1302                                 return (ifa);
1303                 }
1304         }
1305         return (NULL);
1306 }
1307
1308 /*
1309  * Locate the point to point interface with a given destination address.
1310  */
1311 struct ifaddr *
1312 ifa_ifwithdstaddr(struct sockaddr *addr)
1313 {
1314         const struct ifnet_array *arr;
1315         int i;
1316
1317         arr = ifnet_array_get();
1318         for (i = 0; i < arr->ifnet_count; ++i) {
1319                 struct ifnet *ifp = arr->ifnet_arr[i];
1320                 struct ifaddr_container *ifac;
1321
1322                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1323                         continue;
1324
1325                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1326                         struct ifaddr *ifa = ifac->ifa;
1327
1328                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1329                                 continue;
1330                         if (ifa->ifa_dstaddr &&
1331                             sa_equal(addr, ifa->ifa_dstaddr))
1332                                 return (ifa);
1333                 }
1334         }
1335         return (NULL);
1336 }
1337
1338 /*
1339  * Find an interface on a specific network.  If many, choice
1340  * is most specific found.
1341  */
1342 struct ifaddr *
1343 ifa_ifwithnet(struct sockaddr *addr)
1344 {
1345         struct ifaddr *ifa_maybe = NULL;
1346         u_int af = addr->sa_family;
1347         char *addr_data = addr->sa_data, *cplim;
1348         const struct ifnet_array *arr;
1349         int i;
1350
1351         /*
1352          * AF_LINK addresses can be looked up directly by their index number,
1353          * so do that if we can.
1354          */
1355         if (af == AF_LINK) {
1356                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1357
1358                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1359                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1360         }
1361
1362         /*
1363          * Scan though each interface, looking for ones that have
1364          * addresses in this address family.
1365          */
1366         arr = ifnet_array_get();
1367         for (i = 0; i < arr->ifnet_count; ++i) {
1368                 struct ifnet *ifp = arr->ifnet_arr[i];
1369                 struct ifaddr_container *ifac;
1370
1371                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1372                         struct ifaddr *ifa = ifac->ifa;
1373                         char *cp, *cp2, *cp3;
1374
1375                         if (ifa->ifa_addr->sa_family != af)
1376 next:                           continue;
1377                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1378                                 /*
1379                                  * This is a bit broken as it doesn't
1380                                  * take into account that the remote end may
1381                                  * be a single node in the network we are
1382                                  * looking for.
1383                                  * The trouble is that we don't know the
1384                                  * netmask for the remote end.
1385                                  */
1386                                 if (ifa->ifa_dstaddr != NULL &&
1387                                     sa_equal(addr, ifa->ifa_dstaddr))
1388                                         return (ifa);
1389                         } else {
1390                                 /*
1391                                  * if we have a special address handler,
1392                                  * then use it instead of the generic one.
1393                                  */
1394                                 if (ifa->ifa_claim_addr) {
1395                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1396                                                 return (ifa);
1397                                         } else {
1398                                                 continue;
1399                                         }
1400                                 }
1401
1402                                 /*
1403                                  * Scan all the bits in the ifa's address.
1404                                  * If a bit dissagrees with what we are
1405                                  * looking for, mask it with the netmask
1406                                  * to see if it really matters.
1407                                  * (A byte at a time)
1408                                  */
1409                                 if (ifa->ifa_netmask == 0)
1410                                         continue;
1411                                 cp = addr_data;
1412                                 cp2 = ifa->ifa_addr->sa_data;
1413                                 cp3 = ifa->ifa_netmask->sa_data;
1414                                 cplim = ifa->ifa_netmask->sa_len +
1415                                         (char *)ifa->ifa_netmask;
1416                                 while (cp3 < cplim)
1417                                         if ((*cp++ ^ *cp2++) & *cp3++)
1418                                                 goto next; /* next address! */
1419                                 /*
1420                                  * If the netmask of what we just found
1421                                  * is more specific than what we had before
1422                                  * (if we had one) then remember the new one
1423                                  * before continuing to search for an even
1424                                  * better one.  If the netmasks are equal,
1425                                  * we prefer the this ifa based on the result
1426                                  * of ifa_prefer().
1427                                  */
1428                                 if (ifa_maybe == NULL ||
1429                                     rn_refines((char *)ifa->ifa_netmask,
1430                                         (char *)ifa_maybe->ifa_netmask) ||
1431                                     (sa_equal(ifa_maybe->ifa_netmask,
1432                                         ifa->ifa_netmask) &&
1433                                      ifa_prefer(ifa, ifa_maybe)))
1434                                         ifa_maybe = ifa;
1435                         }
1436                 }
1437         }
1438         return (ifa_maybe);
1439 }
1440
1441 /*
1442  * Find an interface address specific to an interface best matching
1443  * a given address.
1444  */
1445 struct ifaddr *
1446 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1447 {
1448         struct ifaddr_container *ifac;
1449         char *cp, *cp2, *cp3;
1450         char *cplim;
1451         struct ifaddr *ifa_maybe = NULL;
1452         u_int af = addr->sa_family;
1453
1454         if (af >= AF_MAX)
1455                 return (0);
1456         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1457                 struct ifaddr *ifa = ifac->ifa;
1458
1459                 if (ifa->ifa_addr->sa_family != af)
1460                         continue;
1461                 if (ifa_maybe == NULL)
1462                         ifa_maybe = ifa;
1463                 if (ifa->ifa_netmask == NULL) {
1464                         if (sa_equal(addr, ifa->ifa_addr) ||
1465                             (ifa->ifa_dstaddr != NULL &&
1466                              sa_equal(addr, ifa->ifa_dstaddr)))
1467                                 return (ifa);
1468                         continue;
1469                 }
1470                 if (ifp->if_flags & IFF_POINTOPOINT) {
1471                         if (sa_equal(addr, ifa->ifa_dstaddr))
1472                                 return (ifa);
1473                 } else {
1474                         cp = addr->sa_data;
1475                         cp2 = ifa->ifa_addr->sa_data;
1476                         cp3 = ifa->ifa_netmask->sa_data;
1477                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1478                         for (; cp3 < cplim; cp3++)
1479                                 if ((*cp++ ^ *cp2++) & *cp3)
1480                                         break;
1481                         if (cp3 == cplim)
1482                                 return (ifa);
1483                 }
1484         }
1485         return (ifa_maybe);
1486 }
1487
1488 /*
1489  * Default action when installing a route with a Link Level gateway.
1490  * Lookup an appropriate real ifa to point to.
1491  * This should be moved to /sys/net/link.c eventually.
1492  */
1493 static void
1494 link_rtrequest(int cmd, struct rtentry *rt)
1495 {
1496         struct ifaddr *ifa;
1497         struct sockaddr *dst;
1498         struct ifnet *ifp;
1499
1500         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1501             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1502                 return;
1503         ifa = ifaof_ifpforaddr(dst, ifp);
1504         if (ifa != NULL) {
1505                 IFAFREE(rt->rt_ifa);
1506                 IFAREF(ifa);
1507                 rt->rt_ifa = ifa;
1508                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1509                         ifa->ifa_rtrequest(cmd, rt);
1510         }
1511 }
1512
1513 struct netmsg_ifroute {
1514         struct netmsg_base      base;
1515         struct ifnet            *ifp;
1516         int                     flag;
1517         int                     fam;
1518 };
1519
1520 /*
1521  * Mark an interface down and notify protocols of the transition.
1522  */
1523 static void
1524 if_unroute_dispatch(netmsg_t nmsg)
1525 {
1526         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1527         struct ifnet *ifp = msg->ifp;
1528         int flag = msg->flag, fam = msg->fam;
1529         struct ifaddr_container *ifac;
1530
1531         ifp->if_flags &= ~flag;
1532         getmicrotime(&ifp->if_lastchange);
1533         /*
1534          * The ifaddr processing in the following loop will block,
1535          * however, this function is called in netisr0, in which
1536          * ifaddr list changes happen, so we don't care about the
1537          * blockness of the ifaddr processing here.
1538          */
1539         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1540                 struct ifaddr *ifa = ifac->ifa;
1541
1542                 /* Ignore marker */
1543                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1544                         continue;
1545
1546                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1547                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1548         }
1549         ifq_purge_all(&ifp->if_snd);
1550         rt_ifmsg(ifp);
1551
1552         lwkt_replymsg(&nmsg->lmsg, 0);
1553 }
1554
1555 void
1556 if_unroute(struct ifnet *ifp, int flag, int fam)
1557 {
1558         struct netmsg_ifroute msg;
1559
1560         ASSERT_CANDOMSG_NETISR0(curthread);
1561
1562         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1563             if_unroute_dispatch);
1564         msg.ifp = ifp;
1565         msg.flag = flag;
1566         msg.fam = fam;
1567         lwkt_domsg(netisr_cpuport(0), &msg.base.lmsg, 0);
1568 }
1569
1570 /*
1571  * Mark an interface up and notify protocols of the transition.
1572  */
1573 static void
1574 if_route_dispatch(netmsg_t nmsg)
1575 {
1576         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1577         struct ifnet *ifp = msg->ifp;
1578         int flag = msg->flag, fam = msg->fam;
1579         struct ifaddr_container *ifac;
1580
1581         ifq_purge_all(&ifp->if_snd);
1582         ifp->if_flags |= flag;
1583         getmicrotime(&ifp->if_lastchange);
1584         /*
1585          * The ifaddr processing in the following loop will block,
1586          * however, this function is called in netisr0, in which
1587          * ifaddr list changes happen, so we don't care about the
1588          * blockness of the ifaddr processing here.
1589          */
1590         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1591                 struct ifaddr *ifa = ifac->ifa;
1592
1593                 /* Ignore marker */
1594                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1595                         continue;
1596
1597                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1598                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1599         }
1600         rt_ifmsg(ifp);
1601 #ifdef INET6
1602         in6_if_up(ifp);
1603 #endif
1604
1605         lwkt_replymsg(&nmsg->lmsg, 0);
1606 }
1607
1608 void
1609 if_route(struct ifnet *ifp, int flag, int fam)
1610 {
1611         struct netmsg_ifroute msg;
1612
1613         ASSERT_CANDOMSG_NETISR0(curthread);
1614
1615         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1616             if_route_dispatch);
1617         msg.ifp = ifp;
1618         msg.flag = flag;
1619         msg.fam = fam;
1620         lwkt_domsg(netisr_cpuport(0), &msg.base.lmsg, 0);
1621 }
1622
1623 /*
1624  * Mark an interface down and notify protocols of the transition.  An
1625  * interface going down is also considered to be a synchronizing event.
1626  * We must ensure that all packet processing related to the interface
1627  * has completed before we return so e.g. the caller can free the ifnet
1628  * structure that the mbufs may be referencing.
1629  *
1630  * NOTE: must be called at splnet or eqivalent.
1631  */
1632 void
1633 if_down(struct ifnet *ifp)
1634 {
1635         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1636         netmsg_service_sync();
1637 }
1638
1639 /*
1640  * Mark an interface up and notify protocols of
1641  * the transition.
1642  * NOTE: must be called at splnet or eqivalent.
1643  */
1644 void
1645 if_up(struct ifnet *ifp)
1646 {
1647         if_route(ifp, IFF_UP, AF_UNSPEC);
1648 }
1649
1650 /*
1651  * Process a link state change.
1652  * NOTE: must be called at splsoftnet or equivalent.
1653  */
1654 void
1655 if_link_state_change(struct ifnet *ifp)
1656 {
1657         int link_state = ifp->if_link_state;
1658
1659         rt_ifmsg(ifp);
1660         devctl_notify("IFNET", ifp->if_xname,
1661             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1662 }
1663
1664 /*
1665  * Handle interface watchdog timer routines.  Called
1666  * from softclock, we decrement timers (if set) and
1667  * call the appropriate interface routine on expiration.
1668  */
1669 static void
1670 if_slowtimo_dispatch(netmsg_t nmsg)
1671 {
1672         struct globaldata *gd = mycpu;
1673         const struct ifnet_array *arr;
1674         int i;
1675
1676         ASSERT_IN_NETISR(0);
1677
1678         crit_enter_gd(gd);
1679         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1680         crit_exit_gd(gd);
1681
1682         arr = ifnet_array_get();
1683         for (i = 0; i < arr->ifnet_count; ++i) {
1684                 struct ifnet *ifp = arr->ifnet_arr[i];
1685
1686                 crit_enter_gd(gd);
1687
1688                 if (if_stats_compat) {
1689                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1690                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1691                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1692                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1693                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1694                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1695                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1696                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1697                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1698                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1699                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1700                 }
1701
1702                 if (ifp->if_timer == 0 || --ifp->if_timer) {
1703                         crit_exit_gd(gd);
1704                         continue;
1705                 }
1706                 if (ifp->if_watchdog) {
1707                         if (ifnet_tryserialize_all(ifp)) {
1708                                 (*ifp->if_watchdog)(ifp);
1709                                 ifnet_deserialize_all(ifp);
1710                         } else {
1711                                 /* try again next timeout */
1712                                 ++ifp->if_timer;
1713                         }
1714                 }
1715
1716                 crit_exit_gd(gd);
1717         }
1718
1719         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1720 }
1721
1722 static void
1723 if_slowtimo(void *arg __unused)
1724 {
1725         struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1726
1727         KASSERT(mycpuid == 0, ("not on cpu0"));
1728         crit_enter();
1729         if (lmsg->ms_flags & MSGF_DONE)
1730                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1731         crit_exit();
1732 }
1733
1734 /*
1735  * Map interface name to
1736  * interface structure pointer.
1737  */
1738 struct ifnet *
1739 ifunit(const char *name)
1740 {
1741         struct ifnet *ifp;
1742
1743         /*
1744          * Search all the interfaces for this name/number
1745          */
1746         KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1747
1748         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1749                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1750                         break;
1751         }
1752         return (ifp);
1753 }
1754
1755 struct ifnet *
1756 ifunit_netisr(const char *name)
1757 {
1758         const struct ifnet_array *arr;
1759         int i;
1760
1761         /*
1762          * Search all the interfaces for this name/number
1763          */
1764
1765         arr = ifnet_array_get();
1766         for (i = 0; i < arr->ifnet_count; ++i) {
1767                 struct ifnet *ifp = arr->ifnet_arr[i];
1768
1769                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1770                         return ifp;
1771         }
1772         return NULL;
1773 }
1774
1775 /*
1776  * Interface ioctls.
1777  */
1778 int
1779 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1780 {
1781         struct ifnet *ifp;
1782         struct ifreq *ifr;
1783         struct ifstat *ifs;
1784         int error;
1785         short oif_flags;
1786         int new_flags;
1787 #ifdef COMPAT_43
1788         int ocmd;
1789 #endif
1790         size_t namelen, onamelen;
1791         char new_name[IFNAMSIZ];
1792         struct ifaddr *ifa;
1793         struct sockaddr_dl *sdl;
1794
1795         switch (cmd) {
1796         case SIOCGIFCONF:
1797         case OSIOCGIFCONF:
1798                 return (ifconf(cmd, data, cred));
1799         default:
1800                 break;
1801         }
1802
1803         ifr = (struct ifreq *)data;
1804
1805         switch (cmd) {
1806         case SIOCIFCREATE:
1807         case SIOCIFCREATE2:
1808                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1809                         return (error);
1810                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1811                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1812         case SIOCIFDESTROY:
1813                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1814                         return (error);
1815                 return (if_clone_destroy(ifr->ifr_name));
1816         case SIOCIFGCLONERS:
1817                 return (if_clone_list((struct if_clonereq *)data));
1818         default:
1819                 break;
1820         }
1821
1822         /*
1823          * Nominal ioctl through interface, lookup the ifp and obtain a
1824          * lock to serialize the ifconfig ioctl operation.
1825          */
1826         ifnet_lock();
1827
1828         ifp = ifunit(ifr->ifr_name);
1829         if (ifp == NULL) {
1830                 ifnet_unlock();
1831                 return (ENXIO);
1832         }
1833         error = 0;
1834
1835         switch (cmd) {
1836         case SIOCGIFINDEX:
1837                 ifr->ifr_index = ifp->if_index;
1838                 break;
1839
1840         case SIOCGIFFLAGS:
1841                 ifr->ifr_flags = ifp->if_flags;
1842                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1843                 break;
1844
1845         case SIOCGIFCAP:
1846                 ifr->ifr_reqcap = ifp->if_capabilities;
1847                 ifr->ifr_curcap = ifp->if_capenable;
1848                 break;
1849
1850         case SIOCGIFMETRIC:
1851                 ifr->ifr_metric = ifp->if_metric;
1852                 break;
1853
1854         case SIOCGIFMTU:
1855                 ifr->ifr_mtu = ifp->if_mtu;
1856                 break;
1857
1858         case SIOCGIFTSOLEN:
1859                 ifr->ifr_tsolen = ifp->if_tsolen;
1860                 break;
1861
1862         case SIOCGIFDATA:
1863                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1864                                 sizeof(ifp->if_data));
1865                 break;
1866
1867         case SIOCGIFPHYS:
1868                 ifr->ifr_phys = ifp->if_physical;
1869                 break;
1870
1871         case SIOCGIFPOLLCPU:
1872                 ifr->ifr_pollcpu = -1;
1873                 break;
1874
1875         case SIOCSIFPOLLCPU:
1876                 break;
1877
1878         case SIOCSIFFLAGS:
1879                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1880                 if (error)
1881                         break;
1882                 new_flags = (ifr->ifr_flags & 0xffff) |
1883                     (ifr->ifr_flagshigh << 16);
1884                 if (ifp->if_flags & IFF_SMART) {
1885                         /* Smart drivers twiddle their own routes */
1886                 } else if (ifp->if_flags & IFF_UP &&
1887                     (new_flags & IFF_UP) == 0) {
1888                         crit_enter();
1889                         if_down(ifp);
1890                         crit_exit();
1891                 } else if (new_flags & IFF_UP &&
1892                     (ifp->if_flags & IFF_UP) == 0) {
1893                         crit_enter();
1894                         if_up(ifp);
1895                         crit_exit();
1896                 }
1897
1898 #ifdef IFPOLL_ENABLE
1899                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1900                         if (new_flags & IFF_NPOLLING)
1901                                 ifpoll_register(ifp);
1902                         else
1903                                 ifpoll_deregister(ifp);
1904                 }
1905 #endif
1906
1907                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1908                         (new_flags &~ IFF_CANTCHANGE);
1909                 if (new_flags & IFF_PPROMISC) {
1910                         /* Permanently promiscuous mode requested */
1911                         ifp->if_flags |= IFF_PROMISC;
1912                 } else if (ifp->if_pcount == 0) {
1913                         ifp->if_flags &= ~IFF_PROMISC;
1914                 }
1915                 if (ifp->if_ioctl) {
1916                         ifnet_serialize_all(ifp);
1917                         ifp->if_ioctl(ifp, cmd, data, cred);
1918                         ifnet_deserialize_all(ifp);
1919                 }
1920                 getmicrotime(&ifp->if_lastchange);
1921                 break;
1922
1923         case SIOCSIFCAP:
1924                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1925                 if (error)
1926                         break;
1927                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1928                         error = EINVAL;
1929                         break;
1930                 }
1931                 ifnet_serialize_all(ifp);
1932                 ifp->if_ioctl(ifp, cmd, data, cred);
1933                 ifnet_deserialize_all(ifp);
1934                 break;
1935
1936         case SIOCSIFNAME:
1937                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1938                 if (error)
1939                         break;
1940                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1941                 if (error)
1942                         break;
1943                 if (new_name[0] == '\0') {
1944                         error = EINVAL;
1945                         break;
1946                 }
1947                 if (ifunit(new_name) != NULL) {
1948                         error = EEXIST;
1949                         break;
1950                 }
1951
1952                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1953
1954                 /* Announce the departure of the interface. */
1955                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1956
1957                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1958                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1959                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1960                 namelen = strlen(new_name);
1961                 onamelen = sdl->sdl_nlen;
1962                 /*
1963                  * Move the address if needed.  This is safe because we
1964                  * allocate space for a name of length IFNAMSIZ when we
1965                  * create this in if_attach().
1966                  */
1967                 if (namelen != onamelen) {
1968                         bcopy(sdl->sdl_data + onamelen,
1969                             sdl->sdl_data + namelen, sdl->sdl_alen);
1970                 }
1971                 bcopy(new_name, sdl->sdl_data, namelen);
1972                 sdl->sdl_nlen = namelen;
1973                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1974                 bzero(sdl->sdl_data, onamelen);
1975                 while (namelen != 0)
1976                         sdl->sdl_data[--namelen] = 0xff;
1977
1978                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1979
1980                 /* Announce the return of the interface. */
1981                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1982                 break;
1983
1984         case SIOCSIFMETRIC:
1985                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1986                 if (error)
1987                         break;
1988                 ifp->if_metric = ifr->ifr_metric;
1989                 getmicrotime(&ifp->if_lastchange);
1990                 break;
1991
1992         case SIOCSIFPHYS:
1993                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1994                 if (error)
1995                         break;
1996                 if (ifp->if_ioctl == NULL) {
1997                         error = EOPNOTSUPP;
1998                         break;
1999                 }
2000                 ifnet_serialize_all(ifp);
2001                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2002                 ifnet_deserialize_all(ifp);
2003                 if (error == 0)
2004                         getmicrotime(&ifp->if_lastchange);
2005                 break;
2006
2007         case SIOCSIFMTU:
2008         {
2009                 u_long oldmtu = ifp->if_mtu;
2010
2011                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2012                 if (error)
2013                         break;
2014                 if (ifp->if_ioctl == NULL) {
2015                         error = EOPNOTSUPP;
2016                         break;
2017                 }
2018                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2019                         error = EINVAL;
2020                         break;
2021                 }
2022                 ifnet_serialize_all(ifp);
2023                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2024                 ifnet_deserialize_all(ifp);
2025                 if (error == 0) {
2026                         getmicrotime(&ifp->if_lastchange);
2027                         rt_ifmsg(ifp);
2028                 }
2029                 /*
2030                  * If the link MTU changed, do network layer specific procedure.
2031                  */
2032                 if (ifp->if_mtu != oldmtu) {
2033 #ifdef INET6
2034                         nd6_setmtu(ifp);
2035 #endif
2036                 }
2037                 break;
2038         }
2039
2040         case SIOCSIFTSOLEN:
2041                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2042                 if (error)
2043                         break;
2044
2045                 /* XXX need driver supplied upper limit */
2046                 if (ifr->ifr_tsolen <= 0) {
2047                         error = EINVAL;
2048                         break;
2049                 }
2050                 ifp->if_tsolen = ifr->ifr_tsolen;
2051                 break;
2052
2053         case SIOCADDMULTI:
2054         case SIOCDELMULTI:
2055                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2056                 if (error)
2057                         break;
2058
2059                 /* Don't allow group membership on non-multicast interfaces. */
2060                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2061                         error = EOPNOTSUPP;
2062                         break;
2063                 }
2064
2065                 /* Don't let users screw up protocols' entries. */
2066                 if (ifr->ifr_addr.sa_family != AF_LINK) {
2067                         error = EINVAL;
2068                         break;
2069                 }
2070
2071                 if (cmd == SIOCADDMULTI) {
2072                         struct ifmultiaddr *ifma;
2073                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2074                 } else {
2075                         error = if_delmulti(ifp, &ifr->ifr_addr);
2076                 }
2077                 if (error == 0)
2078                         getmicrotime(&ifp->if_lastchange);
2079                 break;
2080
2081         case SIOCSIFPHYADDR:
2082         case SIOCDIFPHYADDR:
2083 #ifdef INET6
2084         case SIOCSIFPHYADDR_IN6:
2085 #endif
2086         case SIOCSLIFPHYADDR:
2087         case SIOCSIFMEDIA:
2088         case SIOCSIFGENERIC:
2089                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2090                 if (error)
2091                         break;
2092                 if (ifp->if_ioctl == 0) {
2093                         error = EOPNOTSUPP;
2094                         break;
2095                 }
2096                 ifnet_serialize_all(ifp);
2097                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2098                 ifnet_deserialize_all(ifp);
2099                 if (error == 0)
2100                         getmicrotime(&ifp->if_lastchange);
2101                 break;
2102
2103         case SIOCGIFSTATUS:
2104                 ifs = (struct ifstat *)data;
2105                 ifs->ascii[0] = '\0';
2106                 /* fall through */
2107         case SIOCGIFPSRCADDR:
2108         case SIOCGIFPDSTADDR:
2109         case SIOCGLIFPHYADDR:
2110         case SIOCGIFMEDIA:
2111         case SIOCGIFGENERIC:
2112                 if (ifp->if_ioctl == NULL) {
2113                         error = EOPNOTSUPP;
2114                         break;
2115                 }
2116                 ifnet_serialize_all(ifp);
2117                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2118                 ifnet_deserialize_all(ifp);
2119                 break;
2120
2121         case SIOCSIFLLADDR:
2122                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2123                 if (error)
2124                         break;
2125                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2126                                      ifr->ifr_addr.sa_len);
2127                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2128                 break;
2129
2130         default:
2131                 oif_flags = ifp->if_flags;
2132                 if (so->so_proto == 0) {
2133                         error = EOPNOTSUPP;
2134                         break;
2135                 }
2136 #ifndef COMPAT_43
2137                 error = so_pru_control_direct(so, cmd, data, ifp);
2138 #else
2139                 ocmd = cmd;
2140
2141                 switch (cmd) {
2142                 case SIOCSIFDSTADDR:
2143                 case SIOCSIFADDR:
2144                 case SIOCSIFBRDADDR:
2145                 case SIOCSIFNETMASK:
2146 #if BYTE_ORDER != BIG_ENDIAN
2147                         if (ifr->ifr_addr.sa_family == 0 &&
2148                             ifr->ifr_addr.sa_len < 16) {
2149                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2150                                 ifr->ifr_addr.sa_len = 16;
2151                         }
2152 #else
2153                         if (ifr->ifr_addr.sa_len == 0)
2154                                 ifr->ifr_addr.sa_len = 16;
2155 #endif
2156                         break;
2157                 case OSIOCGIFADDR:
2158                         cmd = SIOCGIFADDR;
2159                         break;
2160                 case OSIOCGIFDSTADDR:
2161                         cmd = SIOCGIFDSTADDR;
2162                         break;
2163                 case OSIOCGIFBRDADDR:
2164                         cmd = SIOCGIFBRDADDR;
2165                         break;
2166                 case OSIOCGIFNETMASK:
2167                         cmd = SIOCGIFNETMASK;
2168                         break;
2169                 default:
2170                         break;
2171                 }
2172
2173                 error = so_pru_control_direct(so, cmd, data, ifp);
2174
2175                 switch (ocmd) {
2176                 case OSIOCGIFADDR:
2177                 case OSIOCGIFDSTADDR:
2178                 case OSIOCGIFBRDADDR:
2179                 case OSIOCGIFNETMASK:
2180                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2181                         break;
2182                 }
2183 #endif /* COMPAT_43 */
2184
2185                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2186 #ifdef INET6
2187                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
2188                         if (ifp->if_flags & IFF_UP) {
2189                                 crit_enter();
2190                                 in6_if_up(ifp);
2191                                 crit_exit();
2192                         }
2193 #endif
2194                 }
2195                 break;
2196         }
2197
2198         ifnet_unlock();
2199         return (error);
2200 }
2201
2202 /*
2203  * Set/clear promiscuous mode on interface ifp based on the truth value
2204  * of pswitch.  The calls are reference counted so that only the first
2205  * "on" request actually has an effect, as does the final "off" request.
2206  * Results are undefined if the "off" and "on" requests are not matched.
2207  */
2208 int
2209 ifpromisc(struct ifnet *ifp, int pswitch)
2210 {
2211         struct ifreq ifr;
2212         int error;
2213         int oldflags;
2214
2215         oldflags = ifp->if_flags;
2216         if (ifp->if_flags & IFF_PPROMISC) {
2217                 /* Do nothing if device is in permanently promiscuous mode */
2218                 ifp->if_pcount += pswitch ? 1 : -1;
2219                 return (0);
2220         }
2221         if (pswitch) {
2222                 /*
2223                  * If the device is not configured up, we cannot put it in
2224                  * promiscuous mode.
2225                  */
2226                 if ((ifp->if_flags & IFF_UP) == 0)
2227                         return (ENETDOWN);
2228                 if (ifp->if_pcount++ != 0)
2229                         return (0);
2230                 ifp->if_flags |= IFF_PROMISC;
2231                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
2232                     ifp->if_xname);
2233         } else {
2234                 if (--ifp->if_pcount > 0)
2235                         return (0);
2236                 ifp->if_flags &= ~IFF_PROMISC;
2237                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2238                     ifp->if_xname);
2239         }
2240         ifr.ifr_flags = ifp->if_flags;
2241         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2242         ifnet_serialize_all(ifp);
2243         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2244         ifnet_deserialize_all(ifp);
2245         if (error == 0)
2246                 rt_ifmsg(ifp);
2247         else
2248                 ifp->if_flags = oldflags;
2249         return error;
2250 }
2251
2252 /*
2253  * Return interface configuration
2254  * of system.  List may be used
2255  * in later ioctl's (above) to get
2256  * other information.
2257  */
2258 static int
2259 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2260 {
2261         struct ifconf *ifc = (struct ifconf *)data;
2262         struct ifnet *ifp;
2263         struct sockaddr *sa;
2264         struct ifreq ifr, *ifrp;
2265         int space = ifc->ifc_len, error = 0;
2266
2267         ifrp = ifc->ifc_req;
2268
2269         ifnet_lock();
2270         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2271                 struct ifaddr_container *ifac, *ifac_mark;
2272                 struct ifaddr_marker mark;
2273                 struct ifaddrhead *head;
2274                 int addrs;
2275
2276                 if (space <= sizeof ifr)
2277                         break;
2278
2279                 /*
2280                  * Zero the stack declared structure first to prevent
2281                  * memory disclosure.
2282                  */
2283                 bzero(&ifr, sizeof(ifr));
2284                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2285                     >= sizeof(ifr.ifr_name)) {
2286                         error = ENAMETOOLONG;
2287                         break;
2288                 }
2289
2290                 /*
2291                  * Add a marker, since copyout() could block and during that
2292                  * period the list could be changed.  Inserting the marker to
2293                  * the header of the list will not cause trouble for the code
2294                  * assuming that the first element of the list is AF_LINK; the
2295                  * marker will be moved to the next position w/o blocking.
2296                  */
2297                 ifa_marker_init(&mark, ifp);
2298                 ifac_mark = &mark.ifac;
2299                 head = &ifp->if_addrheads[mycpuid];
2300
2301                 addrs = 0;
2302                 TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2303                 while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2304                         struct ifaddr *ifa = ifac->ifa;
2305
2306                         TAILQ_REMOVE(head, ifac_mark, ifa_link);
2307                         TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2308
2309                         /* Ignore marker */
2310                         if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2311                                 continue;
2312
2313                         if (space <= sizeof ifr)
2314                                 break;
2315                         sa = ifa->ifa_addr;
2316                         if (cred->cr_prison &&
2317                             prison_if(cred, sa))
2318                                 continue;
2319                         addrs++;
2320                         /*
2321                          * Keep a reference on this ifaddr, so that it will
2322                          * not be destroyed when its address is copied to
2323                          * the userland, which could block.
2324                          */
2325                         IFAREF(ifa);
2326 #ifdef COMPAT_43
2327                         if (cmd == OSIOCGIFCONF) {
2328                                 struct osockaddr *osa =
2329                                          (struct osockaddr *)&ifr.ifr_addr;
2330                                 ifr.ifr_addr = *sa;
2331                                 osa->sa_family = sa->sa_family;
2332                                 error = copyout(&ifr, ifrp, sizeof ifr);
2333                                 ifrp++;
2334                         } else
2335 #endif
2336                         if (sa->sa_len <= sizeof(*sa)) {
2337                                 ifr.ifr_addr = *sa;
2338                                 error = copyout(&ifr, ifrp, sizeof ifr);
2339                                 ifrp++;
2340                         } else {
2341                                 if (space < (sizeof ifr) + sa->sa_len -
2342                                             sizeof(*sa)) {
2343                                         IFAFREE(ifa);
2344                                         break;
2345                                 }
2346                                 space -= sa->sa_len - sizeof(*sa);
2347                                 error = copyout(&ifr, ifrp,
2348                                                 sizeof ifr.ifr_name);
2349                                 if (error == 0)
2350                                         error = copyout(sa, &ifrp->ifr_addr,
2351                                                         sa->sa_len);
2352                                 ifrp = (struct ifreq *)
2353                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2354                         }
2355                         IFAFREE(ifa);
2356                         if (error)
2357                                 break;
2358                         space -= sizeof ifr;
2359                 }
2360                 TAILQ_REMOVE(head, ifac_mark, ifa_link);
2361                 if (error)
2362                         break;
2363                 if (!addrs) {
2364                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2365                         error = copyout(&ifr, ifrp, sizeof ifr);
2366                         if (error)
2367                                 break;
2368                         space -= sizeof ifr;
2369                         ifrp++;
2370                 }
2371         }
2372         ifnet_unlock();
2373
2374         ifc->ifc_len -= space;
2375         return (error);
2376 }
2377
2378 /*
2379  * Just like if_promisc(), but for all-multicast-reception mode.
2380  */
2381 int
2382 if_allmulti(struct ifnet *ifp, int onswitch)
2383 {
2384         int error = 0;
2385         struct ifreq ifr;
2386
2387         crit_enter();
2388
2389         if (onswitch) {
2390                 if (ifp->if_amcount++ == 0) {
2391                         ifp->if_flags |= IFF_ALLMULTI;
2392                         ifr.ifr_flags = ifp->if_flags;
2393                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2394                         ifnet_serialize_all(ifp);
2395                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2396                                               NULL);
2397                         ifnet_deserialize_all(ifp);
2398                 }
2399         } else {
2400                 if (ifp->if_amcount > 1) {
2401                         ifp->if_amcount--;
2402                 } else {
2403                         ifp->if_amcount = 0;
2404                         ifp->if_flags &= ~IFF_ALLMULTI;
2405                         ifr.ifr_flags = ifp->if_flags;
2406                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2407                         ifnet_serialize_all(ifp);
2408                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2409                                               NULL);
2410                         ifnet_deserialize_all(ifp);
2411                 }
2412         }
2413
2414         crit_exit();
2415
2416         if (error == 0)
2417                 rt_ifmsg(ifp);
2418         return error;
2419 }
2420
2421 /*
2422  * Add a multicast listenership to the interface in question.
2423  * The link layer provides a routine which converts
2424  */
2425 int
2426 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2427     struct ifmultiaddr **retifma)
2428 {
2429         struct sockaddr *llsa, *dupsa;
2430         int error;
2431         struct ifmultiaddr *ifma;
2432
2433         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2434
2435         /*
2436          * If the matching multicast address already exists
2437          * then don't add a new one, just add a reference
2438          */
2439         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2440                 if (sa_equal(sa, ifma->ifma_addr)) {
2441                         ifma->ifma_refcount++;
2442                         if (retifma)
2443                                 *retifma = ifma;
2444                         return 0;
2445                 }
2446         }
2447
2448         /*
2449          * Give the link layer a chance to accept/reject it, and also
2450          * find out which AF_LINK address this maps to, if it isn't one
2451          * already.
2452          */
2453         if (ifp->if_resolvemulti) {
2454                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2455                 if (error)
2456                         return error;
2457         } else {
2458                 llsa = NULL;
2459         }
2460
2461         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2462         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2463         bcopy(sa, dupsa, sa->sa_len);
2464
2465         ifma->ifma_addr = dupsa;
2466         ifma->ifma_lladdr = llsa;
2467         ifma->ifma_ifp = ifp;
2468         ifma->ifma_refcount = 1;
2469         ifma->ifma_protospec = NULL;
2470         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2471
2472         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2473         if (retifma)
2474                 *retifma = ifma;
2475
2476         if (llsa != NULL) {
2477                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2478                         if (sa_equal(ifma->ifma_addr, llsa))
2479                                 break;
2480                 }
2481                 if (ifma) {
2482                         ifma->ifma_refcount++;
2483                 } else {
2484                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2485                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2486                         bcopy(llsa, dupsa, llsa->sa_len);
2487                         ifma->ifma_addr = dupsa;
2488                         ifma->ifma_ifp = ifp;
2489                         ifma->ifma_refcount = 1;
2490                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2491                 }
2492         }
2493         /*
2494          * We are certain we have added something, so call down to the
2495          * interface to let them know about it.
2496          */
2497         if (ifp->if_ioctl)
2498                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2499
2500         return 0;
2501 }
2502
2503 int
2504 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2505     struct ifmultiaddr **retifma)
2506 {
2507         int error;
2508
2509         ifnet_serialize_all(ifp);
2510         error = if_addmulti_serialized(ifp, sa, retifma);
2511         ifnet_deserialize_all(ifp);
2512
2513         return error;
2514 }
2515
2516 /*
2517  * Remove a reference to a multicast address on this interface.  Yell
2518  * if the request does not match an existing membership.
2519  */
2520 static int
2521 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2522 {
2523         struct ifmultiaddr *ifma;
2524
2525         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2526
2527         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2528                 if (sa_equal(sa, ifma->ifma_addr))
2529                         break;
2530         if (ifma == NULL)
2531                 return ENOENT;
2532
2533         if (ifma->ifma_refcount > 1) {
2534                 ifma->ifma_refcount--;
2535                 return 0;
2536         }
2537
2538         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2539         sa = ifma->ifma_lladdr;
2540         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2541         /*
2542          * Make sure the interface driver is notified
2543          * in the case of a link layer mcast group being left.
2544          */
2545         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2546                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2547         kfree(ifma->ifma_addr, M_IFMADDR);
2548         kfree(ifma, M_IFMADDR);
2549         if (sa == NULL)
2550                 return 0;
2551
2552         /*
2553          * Now look for the link-layer address which corresponds to
2554          * this network address.  It had been squirreled away in
2555          * ifma->ifma_lladdr for this purpose (so we don't have
2556          * to call ifp->if_resolvemulti() again), and we saved that
2557          * value in sa above.  If some nasty deleted the
2558          * link-layer address out from underneath us, we can deal because
2559          * the address we stored was is not the same as the one which was
2560          * in the record for the link-layer address.  (So we don't complain
2561          * in that case.)
2562          */
2563         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2564                 if (sa_equal(sa, ifma->ifma_addr))
2565                         break;
2566         if (ifma == NULL)
2567                 return 0;
2568
2569         if (ifma->ifma_refcount > 1) {
2570                 ifma->ifma_refcount--;
2571                 return 0;
2572         }
2573
2574         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2575         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2576         kfree(ifma->ifma_addr, M_IFMADDR);
2577         kfree(sa, M_IFMADDR);
2578         kfree(ifma, M_IFMADDR);
2579
2580         return 0;
2581 }
2582
2583 int
2584 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2585 {
2586         int error;
2587
2588         ifnet_serialize_all(ifp);
2589         error = if_delmulti_serialized(ifp, sa);
2590         ifnet_deserialize_all(ifp);
2591
2592         return error;
2593 }
2594
2595 /*
2596  * Delete all multicast group membership for an interface.
2597  * Should be used to quickly flush all multicast filters.
2598  */
2599 void
2600 if_delallmulti_serialized(struct ifnet *ifp)
2601 {
2602         struct ifmultiaddr *ifma, mark;
2603         struct sockaddr sa;
2604
2605         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2606
2607         bzero(&sa, sizeof(sa));
2608         sa.sa_family = AF_UNSPEC;
2609         sa.sa_len = sizeof(sa);
2610
2611         bzero(&mark, sizeof(mark));
2612         mark.ifma_addr = &sa;
2613
2614         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2615         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2616                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2617                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2618                     ifma_link);
2619
2620                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2621                         continue;
2622
2623                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2624         }
2625         TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2626 }
2627
2628
2629 /*
2630  * Set the link layer address on an interface.
2631  *
2632  * At this time we only support certain types of interfaces,
2633  * and we don't allow the length of the address to change.
2634  */
2635 int
2636 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2637 {
2638         struct sockaddr_dl *sdl;
2639         struct ifreq ifr;
2640
2641         sdl = IF_LLSOCKADDR(ifp);
2642         if (sdl == NULL)
2643                 return (EINVAL);
2644         if (len != sdl->sdl_alen)       /* don't allow length to change */
2645                 return (EINVAL);
2646         switch (ifp->if_type) {
2647         case IFT_ETHER:                 /* these types use struct arpcom */
2648         case IFT_XETHER:
2649         case IFT_L2VLAN:
2650         case IFT_IEEE8023ADLAG:
2651                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2652                 bcopy(lladdr, LLADDR(sdl), len);
2653                 break;
2654         default:
2655                 return (ENODEV);
2656         }
2657         /*
2658          * If the interface is already up, we need
2659          * to re-init it in order to reprogram its
2660          * address filter.
2661          */
2662         ifnet_serialize_all(ifp);
2663         if ((ifp->if_flags & IFF_UP) != 0) {
2664 #ifdef INET
2665                 struct ifaddr_container *ifac;
2666 #endif
2667
2668                 ifp->if_flags &= ~IFF_UP;
2669                 ifr.ifr_flags = ifp->if_flags;
2670                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2671                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2672                               NULL);
2673                 ifp->if_flags |= IFF_UP;
2674                 ifr.ifr_flags = ifp->if_flags;
2675                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2676                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2677                                  NULL);
2678 #ifdef INET
2679                 /*
2680                  * Also send gratuitous ARPs to notify other nodes about
2681                  * the address change.
2682                  */
2683                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2684                         struct ifaddr *ifa = ifac->ifa;
2685
2686                         if (ifa->ifa_addr != NULL &&
2687                             ifa->ifa_addr->sa_family == AF_INET)
2688                                 arp_gratuitous(ifp, ifa);
2689                 }
2690 #endif
2691         }
2692         ifnet_deserialize_all(ifp);
2693         return (0);
2694 }
2695
2696 struct ifmultiaddr *
2697 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2698 {
2699         struct ifmultiaddr *ifma;
2700
2701         /* TODO: need ifnet_serialize_main */
2702         ifnet_serialize_all(ifp);
2703         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2704                 if (sa_equal(ifma->ifma_addr, sa))
2705                         break;
2706         ifnet_deserialize_all(ifp);
2707
2708         return ifma;
2709 }
2710
2711 /*
2712  * This function locates the first real ethernet MAC from a network
2713  * card and loads it into node, returning 0 on success or ENOENT if
2714  * no suitable interfaces were found.  It is used by the uuid code to
2715  * generate a unique 6-byte number.
2716  */
2717 int
2718 if_getanyethermac(uint16_t *node, int minlen)
2719 {
2720         struct ifnet *ifp;
2721         struct sockaddr_dl *sdl;
2722
2723         ifnet_lock();
2724         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2725                 if (ifp->if_type != IFT_ETHER)
2726                         continue;
2727                 sdl = IF_LLSOCKADDR(ifp);
2728                 if (sdl->sdl_alen < minlen)
2729                         continue;
2730                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2731                       minlen);
2732                 ifnet_unlock();
2733                 return(0);
2734         }
2735         ifnet_unlock();
2736         return (ENOENT);
2737 }
2738
2739 /*
2740  * The name argument must be a pointer to storage which will last as
2741  * long as the interface does.  For physical devices, the result of
2742  * device_get_name(dev) is a good choice and for pseudo-devices a
2743  * static string works well.
2744  */
2745 void
2746 if_initname(struct ifnet *ifp, const char *name, int unit)
2747 {
2748         ifp->if_dname = name;
2749         ifp->if_dunit = unit;
2750         if (unit != IF_DUNIT_NONE)
2751                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2752         else
2753                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2754 }
2755
2756 int
2757 if_printf(struct ifnet *ifp, const char *fmt, ...)
2758 {
2759         __va_list ap;
2760         int retval;
2761
2762         retval = kprintf("%s: ", ifp->if_xname);
2763         __va_start(ap, fmt);
2764         retval += kvprintf(fmt, ap);
2765         __va_end(ap);
2766         return (retval);
2767 }
2768
2769 struct ifnet *
2770 if_alloc(uint8_t type)
2771 {
2772         struct ifnet *ifp;
2773         size_t size;
2774
2775         /*
2776          * XXX temporary hack until arpcom is setup in if_l2com
2777          */
2778         if (type == IFT_ETHER)
2779                 size = sizeof(struct arpcom);
2780         else
2781                 size = sizeof(struct ifnet);
2782
2783         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2784
2785         ifp->if_type = type;
2786
2787         if (if_com_alloc[type] != NULL) {
2788                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2789                 if (ifp->if_l2com == NULL) {
2790                         kfree(ifp, M_IFNET);
2791                         return (NULL);
2792                 }
2793         }
2794         return (ifp);
2795 }
2796
2797 void
2798 if_free(struct ifnet *ifp)
2799 {
2800         kfree(ifp, M_IFNET);
2801 }
2802
2803 void
2804 ifq_set_classic(struct ifaltq *ifq)
2805 {
2806         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2807             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2808 }
2809
2810 void
2811 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2812     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2813 {
2814         int q;
2815
2816         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2817         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2818         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2819         KASSERT(request != NULL, ("request is not specified"));
2820
2821         ifq->altq_mapsubq = mapsubq;
2822         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2823                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2824
2825                 ifsq->ifsq_enqueue = enqueue;
2826                 ifsq->ifsq_dequeue = dequeue;
2827                 ifsq->ifsq_request = request;
2828         }
2829 }
2830
2831 static void
2832 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2833 {
2834         m->m_nextpkt = NULL;
2835         if (ifsq->ifsq_norm_tail == NULL)
2836                 ifsq->ifsq_norm_head = m;
2837         else
2838                 ifsq->ifsq_norm_tail->m_nextpkt = m;
2839         ifsq->ifsq_norm_tail = m;
2840         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2841 }
2842
2843 static void
2844 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2845 {
2846         m->m_nextpkt = NULL;
2847         if (ifsq->ifsq_prio_tail == NULL)
2848                 ifsq->ifsq_prio_head = m;
2849         else
2850                 ifsq->ifsq_prio_tail->m_nextpkt = m;
2851         ifsq->ifsq_prio_tail = m;
2852         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2853         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2854 }
2855
2856 static struct mbuf *
2857 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2858 {
2859         struct mbuf *m;
2860
2861         m = ifsq->ifsq_norm_head;
2862         if (m != NULL) {
2863                 if ((ifsq->ifsq_norm_head = m->m_nextpkt) == NULL)
2864                         ifsq->ifsq_norm_tail = NULL;
2865                 m->m_nextpkt = NULL;
2866                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2867         }
2868         return m;
2869 }
2870
2871 static struct mbuf *
2872 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2873 {
2874         struct mbuf *m;
2875
2876         m = ifsq->ifsq_prio_head;
2877         if (m != NULL) {
2878                 if ((ifsq->ifsq_prio_head = m->m_nextpkt) == NULL)
2879                         ifsq->ifsq_prio_tail = NULL;
2880                 m->m_nextpkt = NULL;
2881                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2882                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2883         }
2884         return m;
2885 }
2886
2887 int
2888 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2889     struct altq_pktattr *pa __unused)
2890 {
2891         M_ASSERTPKTHDR(m);
2892         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2893             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2894                 if ((m->m_flags & M_PRIO) &&
2895                     ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen / 2) &&
2896                     ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt / 2)) {
2897                         struct mbuf *m_drop;
2898
2899                         /*
2900                          * Perform drop-head on normal queue
2901                          */
2902                         m_drop = ifsq_norm_dequeue(ifsq);
2903                         if (m_drop != NULL) {
2904                                 m_freem(m_drop);
2905                                 ifsq_prio_enqueue(ifsq, m);
2906                                 return 0;
2907                         }
2908                         /* XXX nothing could be dropped? */
2909                 }
2910                 m_freem(m);
2911                 return ENOBUFS;
2912         } else {
2913                 if (m->m_flags & M_PRIO)
2914                         ifsq_prio_enqueue(ifsq, m);
2915                 else
2916                         ifsq_norm_enqueue(ifsq, m);
2917                 return 0;
2918         }
2919 }
2920
2921 struct mbuf *
2922 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2923 {
2924         struct mbuf *m;
2925
2926         switch (op) {
2927         case ALTDQ_POLL:
2928                 m = ifsq->ifsq_prio_head;
2929                 if (m == NULL)
2930                         m = ifsq->ifsq_norm_head;
2931                 break;
2932
2933         case ALTDQ_REMOVE:
2934                 m = ifsq_prio_dequeue(ifsq);
2935                 if (m == NULL)
2936                         m = ifsq_norm_dequeue(ifsq);
2937                 break;
2938
2939         default:
2940                 panic("unsupported ALTQ dequeue op: %d", op);
2941         }
2942         return m;
2943 }
2944
2945 int
2946 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2947 {
2948         switch (req) {
2949         case ALTRQ_PURGE:
2950                 for (;;) {
2951                         struct mbuf *m;
2952
2953                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2954                         if (m == NULL)
2955                                 break;
2956                         m_freem(m);
2957                 }
2958                 break;
2959
2960         default:
2961                 panic("unsupported ALTQ request: %d", req);
2962         }
2963         return 0;
2964 }
2965
2966 static void
2967 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2968 {
2969         struct ifnet *ifp = ifsq_get_ifp(ifsq);
2970         int running = 0, need_sched;
2971
2972         /*
2973          * Try to do direct ifnet.if_start on the subqueue first, if there is
2974          * contention on the subqueue hardware serializer, ifnet.if_start on
2975          * the subqueue will be scheduled on the subqueue owner CPU.
2976          */
2977         if (!ifsq_tryserialize_hw(ifsq)) {
2978                 /*
2979                  * Subqueue hardware serializer contention happened,
2980                  * ifnet.if_start on the subqueue is scheduled on
2981                  * the subqueue owner CPU, and we keep going.
2982                  */
2983                 ifsq_ifstart_schedule(ifsq, 1);
2984                 return;
2985         }
2986
2987         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2988                 ifp->if_start(ifp, ifsq);
2989                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2990                         running = 1;
2991         }
2992         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2993
2994         ifsq_deserialize_hw(ifsq);
2995
2996         if (need_sched) {
2997                 /*
2998                  * More data need to be transmitted, ifnet.if_start on the
2999                  * subqueue is scheduled on the subqueue owner CPU, and we
3000                  * keep going.
3001                  * NOTE: ifnet.if_start subqueue interlock is not released.
3002                  */
3003                 ifsq_ifstart_schedule(ifsq, force_sched);
3004         }
3005 }
3006
3007 /*
3008  * Subqeue packets staging mechanism:
3009  *
3010  * The packets enqueued into the subqueue are staged to a certain amount
3011  * before the ifnet.if_start on the subqueue is called.  In this way, the
3012  * driver could avoid writing to hardware registers upon every packet,
3013  * instead, hardware registers could be written when certain amount of
3014  * packets are put onto hardware TX ring.  The measurement on several modern
3015  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
3016  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
3017  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
3018  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
3019  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
3020  *
3021  * Subqueue packets staging is performed for two entry points into drivers'
3022  * transmission function:
3023  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
3024  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
3025  *
3026  * Subqueue packets staging will be stopped upon any of the following
3027  * conditions:
3028  * - If the count of packets enqueued on the current CPU is great than or
3029  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
3030  * - If the total length of packets enqueued on the current CPU is great
3031  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
3032  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
3033  *   is usually less than hardware's MTU.
3034  * - ifsq_ifstart_schedule() is not pending on the current CPU and
3035  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
3036  *   released.
3037  * - The if_start_rollup(), which is registered as low priority netisr
3038  *   rollup function, is called; probably because no more work is pending
3039  *   for netisr.
3040  *
3041  * NOTE:
3042  * Currently subqueue packet staging is only performed in netisr threads.
3043  */
3044 int
3045 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
3046 {
3047         struct ifaltq *ifq = &ifp->if_snd;
3048         struct ifaltq_subque *ifsq;
3049         int error, start = 0, len, mcast = 0, avoid_start = 0;
3050         struct ifsubq_stage_head *head = NULL;
3051         struct ifsubq_stage *stage = NULL;
3052         struct globaldata *gd = mycpu;
3053         struct thread *td = gd->gd_curthread;
3054
3055         crit_enter_quick(td);
3056
3057         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
3058         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3059
3060         len = m->m_pkthdr.len;
3061         if (m->m_flags & M_MCAST)
3062                 mcast = 1;
3063
3064         if (td->td_type == TD_TYPE_NETISR) {
3065                 head = &ifsubq_stage_heads[mycpuid];
3066                 stage = ifsq_get_stage(ifsq, mycpuid);
3067
3068                 stage->stg_cnt++;
3069                 stage->stg_len += len;
3070                 if (stage->stg_cnt < ifsq_stage_cntmax &&
3071                     stage->stg_len < (ifp->if_mtu - max_protohdr))
3072                         avoid_start = 1;
3073         }
3074
3075         ALTQ_SQ_LOCK(ifsq);
3076         error = ifsq_enqueue_locked(ifsq, m, pa);
3077         if (error) {
3078                 if (!ifsq_data_ready(ifsq)) {
3079                         ALTQ_SQ_UNLOCK(ifsq);
3080                         crit_exit_quick(td);
3081                         return error;
3082                 }
3083                 avoid_start = 0;
3084         }
3085         if (!ifsq_is_started(ifsq)) {
3086                 if (avoid_start) {
3087                         ALTQ_SQ_UNLOCK(ifsq);
3088
3089                         KKASSERT(!error);
3090                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3091                                 ifsq_stage_insert(head, stage);
3092
3093                         IFNET_STAT_INC(ifp, obytes, len);
3094                         if (mcast)
3095                                 IFNET_STAT_INC(ifp, omcasts, 1);
3096                         crit_exit_quick(td);
3097                         return error;
3098                 }
3099
3100                 /*
3101                  * Hold the subqueue interlock of ifnet.if_start
3102                  */
3103                 ifsq_set_started(ifsq);
3104                 start = 1;
3105         }
3106         ALTQ_SQ_UNLOCK(ifsq);
3107
3108         if (!error) {
3109                 IFNET_STAT_INC(ifp, obytes, len);
3110                 if (mcast)
3111                         IFNET_STAT_INC(ifp, omcasts, 1);
3112         }
3113
3114         if (stage != NULL) {
3115                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3116                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3117                         if (!avoid_start) {
3118                                 ifsq_stage_remove(head, stage);
3119                                 ifsq_ifstart_schedule(ifsq, 1);
3120                         }
3121                         crit_exit_quick(td);
3122                         return error;
3123                 }
3124
3125                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3126                         ifsq_stage_remove(head, stage);
3127                 } else {
3128                         stage->stg_cnt = 0;
3129                         stage->stg_len = 0;
3130                 }
3131         }
3132
3133         if (!start) {
3134                 crit_exit_quick(td);
3135                 return error;
3136         }
3137
3138         ifsq_ifstart_try(ifsq, 0);
3139
3140         crit_exit_quick(td);
3141         return error;
3142 }
3143
3144 void *
3145 ifa_create(int size)
3146 {
3147         struct ifaddr *ifa;
3148         int i;
3149
3150         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3151
3152         ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3153         ifa->ifa_containers =
3154             kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
3155                 M_IFADDR, M_INTWAIT | M_ZERO);
3156
3157         ifa->ifa_ncnt = ncpus;
3158         for (i = 0; i < ncpus; ++i) {
3159                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3160
3161                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3162                 ifac->ifa = ifa;
3163                 ifac->ifa_refcnt = 1;
3164         }
3165 #ifdef IFADDR_DEBUG
3166         kprintf("alloc ifa %p %d\n", ifa, size);
3167 #endif
3168         return ifa;
3169 }
3170
3171 void
3172 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3173 {
3174         struct ifaddr *ifa = ifac->ifa;
3175
3176         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3177         KKASSERT(ifac->ifa_refcnt == 0);
3178         KASSERT(ifac->ifa_listmask == 0,
3179                 ("ifa is still on %#x lists", ifac->ifa_listmask));
3180
3181         ifac->ifa_magic = IFA_CONTAINER_DEAD;
3182
3183 #ifdef IFADDR_DEBUG_VERBOSE
3184         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3185 #endif
3186
3187         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3188                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
3189         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3190 #ifdef IFADDR_DEBUG
3191                 kprintf("free ifa %p\n", ifa);
3192 #endif
3193                 kfree(ifa->ifa_containers, M_IFADDR);
3194                 kfree(ifa, M_IFADDR);
3195         }
3196 }
3197
3198 static void
3199 ifa_iflink_dispatch(netmsg_t nmsg)
3200 {
3201         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3202         struct ifaddr *ifa = msg->ifa;
3203         struct ifnet *ifp = msg->ifp;
3204         int cpu = mycpuid;
3205         struct ifaddr_container *ifac;
3206
3207         crit_enter();
3208
3209         ifac = &ifa->ifa_containers[cpu];
3210         ASSERT_IFAC_VALID(ifac);
3211         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3212                 ("ifaddr is on if_addrheads"));
3213
3214         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3215         if (msg->tail)
3216                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3217         else
3218                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3219
3220         crit_exit();
3221
3222         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
3223 }
3224
3225 void
3226 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3227 {
3228         struct netmsg_ifaddr msg;
3229
3230         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3231                     0, ifa_iflink_dispatch);
3232         msg.ifa = ifa;
3233         msg.ifp = ifp;
3234         msg.tail = tail;
3235
3236         ifa_domsg(&msg.base.lmsg, 0);
3237 }
3238
3239 static void
3240 ifa_ifunlink_dispatch(netmsg_t nmsg)
3241 {
3242         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3243         struct ifaddr *ifa = msg->ifa;
3244         struct ifnet *ifp = msg->ifp;
3245         int cpu = mycpuid;
3246         struct ifaddr_container *ifac;
3247
3248         crit_enter();
3249
3250         ifac = &ifa->ifa_containers[cpu];
3251         ASSERT_IFAC_VALID(ifac);
3252         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3253                 ("ifaddr is not on if_addrhead"));
3254
3255         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3256         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3257
3258         crit_exit();
3259
3260         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
3261 }
3262
3263 void
3264 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3265 {
3266         struct netmsg_ifaddr msg;
3267
3268         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3269                     0, ifa_ifunlink_dispatch);
3270         msg.ifa = ifa;
3271         msg.ifp = ifp;
3272
3273         ifa_domsg(&msg.base.lmsg, 0);
3274 }
3275
3276 static void
3277 ifa_destroy_dispatch(netmsg_t nmsg)
3278 {
3279         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3280
3281         IFAFREE(msg->ifa);
3282         ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3283 }
3284
3285 void
3286 ifa_destroy(struct ifaddr *ifa)
3287 {
3288         struct netmsg_ifaddr msg;
3289
3290         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3291                     0, ifa_destroy_dispatch);
3292         msg.ifa = ifa;
3293
3294         ifa_domsg(&msg.base.lmsg, 0);
3295 }
3296
3297 struct lwkt_port *
3298 ifnet_portfn(int cpu)
3299 {
3300         return &ifnet_threads[cpu].td_msgport;
3301 }
3302
3303 void
3304 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
3305 {
3306         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
3307
3308         if (next_cpu < ncpus)
3309                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
3310         else
3311                 lwkt_replymsg(lmsg, 0);
3312 }
3313
3314 int
3315 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
3316 {
3317         KKASSERT(cpu < ncpus);
3318         return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
3319 }
3320
3321 void
3322 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
3323 {
3324         KKASSERT(cpu < ncpus);
3325         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
3326 }
3327
3328 /*
3329  * Generic netmsg service loop.  Some protocols may roll their own but all
3330  * must do the basic command dispatch function call done here.
3331  */
3332 static void
3333 ifnet_service_loop(void *arg __unused)
3334 {
3335         netmsg_t msg;
3336
3337         while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
3338                 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
3339                 msg->base.nm_dispatch(msg);
3340         }
3341 }
3342
3343 static void
3344 if_start_rollup(void)
3345 {
3346         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3347         struct ifsubq_stage *stage;
3348
3349         crit_enter();
3350
3351         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3352                 struct ifaltq_subque *ifsq = stage->stg_subq;
3353                 int is_sched = 0;
3354
3355                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3356                         is_sched = 1;
3357                 ifsq_stage_remove(head, stage);
3358
3359                 if (is_sched) {
3360                         ifsq_ifstart_schedule(ifsq, 1);
3361                 } else {
3362                         int start = 0;
3363
3364                         ALTQ_SQ_LOCK(ifsq);
3365                         if (!ifsq_is_started(ifsq)) {
3366                                 /*
3367                                  * Hold the subqueue interlock of
3368                                  * ifnet.if_start
3369                                  */
3370                                 ifsq_set_started(ifsq);
3371                                 start = 1;
3372                         }
3373                         ALTQ_SQ_UNLOCK(ifsq);
3374
3375                         if (start)
3376                                 ifsq_ifstart_try(ifsq, 1);
3377                 }
3378                 KKASSERT((stage->stg_flags &
3379                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3380         }
3381
3382         crit_exit();
3383 }
3384
3385 static void
3386 ifnetinit(void *dummy __unused)
3387 {
3388         int i;
3389
3390         for (i = 0; i < ncpus; ++i) {
3391                 struct thread *thr = &ifnet_threads[i];
3392
3393                 lwkt_create(ifnet_service_loop, NULL, NULL,
3394                             thr, TDF_NOSTART|TDF_FORCE_SPINPORT|TDF_FIXEDCPU,
3395                             i, "ifnet %d", i);
3396                 netmsg_service_port_init(&thr->td_msgport);
3397                 lwkt_schedule(thr);
3398         }
3399
3400         for (i = 0; i < ncpus; ++i)
3401                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3402         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3403 }
3404
3405 void
3406 if_register_com_alloc(u_char type,
3407     if_com_alloc_t *a, if_com_free_t *f)
3408 {
3409
3410         KASSERT(if_com_alloc[type] == NULL,
3411             ("if_register_com_alloc: %d already registered", type));
3412         KASSERT(if_com_free[type] == NULL,
3413             ("if_register_com_alloc: %d free already registered", type));
3414
3415         if_com_alloc[type] = a;
3416         if_com_free[type] = f;
3417 }
3418
3419 void
3420 if_deregister_com_alloc(u_char type)
3421 {
3422
3423         KASSERT(if_com_alloc[type] != NULL,
3424             ("if_deregister_com_alloc: %d not registered", type));
3425         KASSERT(if_com_free[type] != NULL,
3426             ("if_deregister_com_alloc: %d free not registered", type));
3427         if_com_alloc[type] = NULL;
3428         if_com_free[type] = NULL;
3429 }
3430
3431 int
3432 if_ring_count2(int cnt, int cnt_max)
3433 {
3434         int shift = 0;
3435
3436         KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3437             ("invalid ring count max %d", cnt_max));
3438
3439         if (cnt <= 0)
3440                 cnt = cnt_max;
3441         if (cnt > ncpus2)
3442                 cnt = ncpus2;
3443         if (cnt > cnt_max)
3444                 cnt = cnt_max;
3445
3446         while ((1 << (shift + 1)) <= cnt)
3447                 ++shift;
3448         cnt = 1 << shift;
3449
3450         KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3451             ("calculate cnt %d, ncpus2 %d, cnt max %d",
3452              cnt, ncpus2, cnt_max));
3453         return cnt;
3454 }
3455
3456 void
3457 ifq_set_maxlen(struct ifaltq *ifq, int len)
3458 {
3459         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3460 }
3461
3462 int
3463 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3464 {
3465         return ALTQ_SUBQ_INDEX_DEFAULT;
3466 }
3467
3468 int
3469 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3470 {
3471         return (cpuid & ifq->altq_subq_mask);
3472 }
3473
3474 static void
3475 ifsq_watchdog(void *arg)
3476 {
3477         struct ifsubq_watchdog *wd = arg;
3478         struct ifnet *ifp;
3479
3480         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3481                 goto done;
3482
3483         ifp = ifsq_get_ifp(wd->wd_subq);
3484         if (ifnet_tryserialize_all(ifp)) {
3485                 wd->wd_watchdog(wd->wd_subq);
3486                 ifnet_deserialize_all(ifp);
3487         } else {
3488                 /* try again next timeout */
3489                 wd->wd_timer = 1;
3490         }
3491 done:
3492         ifsq_watchdog_reset(wd);
3493 }
3494
3495 static void
3496 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3497 {
3498         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3499             ifsq_get_cpuid(wd->wd_subq));
3500 }
3501
3502 void
3503 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3504     ifsq_watchdog_t watchdog)
3505 {
3506         callout_init_mp(&wd->wd_callout);
3507         wd->wd_timer = 0;
3508         wd->wd_subq = ifsq;
3509         wd->wd_watchdog = watchdog;
3510 }
3511
3512 void
3513 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3514 {
3515         wd->wd_timer = 0;
3516         ifsq_watchdog_reset(wd);
3517 }
3518
3519 void
3520 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3521 {
3522         wd->wd_timer = 0;
3523         callout_stop(&wd->wd_callout);
3524 }
3525
3526 void
3527 ifnet_lock(void)
3528 {
3529         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3530             ("try holding ifnet lock in netisr"));
3531         mtx_lock(&ifnet_mtx);
3532 }
3533
3534 void
3535 ifnet_unlock(void)
3536 {
3537         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3538             ("try holding ifnet lock in netisr"));
3539         mtx_unlock(&ifnet_mtx);
3540 }
3541
3542 static struct ifnet_array *
3543 ifnet_array_alloc(int count)
3544 {
3545         struct ifnet_array *arr;
3546
3547         arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3548             M_IFNET, M_WAITOK);
3549         arr->ifnet_count = count;
3550
3551         return arr;
3552 }
3553
3554 static void
3555 ifnet_array_free(struct ifnet_array *arr)
3556 {
3557         if (arr == &ifnet_array0)
3558                 return;
3559         kfree(arr, M_IFNET);
3560 }
3561
3562 static struct ifnet_array *
3563 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3564 {
3565         struct ifnet_array *arr;
3566         int count, i;
3567
3568         KASSERT(old_arr->ifnet_count >= 0,
3569             ("invalid ifnet array count %d", old_arr->ifnet_count));
3570         count = old_arr->ifnet_count + 1;
3571         arr = ifnet_array_alloc(count);
3572
3573         /*
3574          * Save the old ifnet array and append this ifp to the end of
3575          * the new ifnet array.
3576          */
3577         for (i = 0; i < old_arr->ifnet_count; ++i) {
3578                 KASSERT(old_arr->ifnet_arr[i] != ifp,
3579                     ("%s is already in ifnet array", ifp->if_xname));
3580                 arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3581         }
3582         KASSERT(i == count - 1,
3583             ("add %s, ifnet array index mismatch, should be %d, but got %d",
3584              ifp->if_xname, count - 1, i));
3585         arr->ifnet_arr[i] = ifp;
3586
3587         return arr;
3588 }
3589
3590 static struct ifnet_array *
3591 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3592 {
3593         struct ifnet_array *arr;
3594         int count, i, idx, found = 0;
3595
3596         KASSERT(old_arr->ifnet_count > 0,
3597             ("invalid ifnet array count %d", old_arr->ifnet_count));
3598         count = old_arr->ifnet_count - 1;
3599         arr = ifnet_array_alloc(count);
3600
3601         /*
3602          * Save the old ifnet array, but skip this ifp.
3603          */
3604         idx = 0;
3605         for (i = 0; i < old_arr->ifnet_count; ++i) {
3606                 if (old_arr->ifnet_arr[i] == ifp) {
3607                         KASSERT(!found,
3608                             ("dup %s is in ifnet array", ifp->if_xname));
3609                         found = 1;
3610                         continue;
3611                 }
3612                 KASSERT(idx < count,
3613                     ("invalid ifnet array index %d, count %d", idx, count));
3614                 arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3615                 ++idx;
3616         }
3617         KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3618         KASSERT(idx == count,
3619             ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3620              ifp->if_xname, count, idx));
3621
3622         return arr;
3623 }
3624
3625 const struct ifnet_array *
3626 ifnet_array_get(void)
3627 {
3628         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3629         return ifnet_array;
3630 }
3631
3632 int
3633 ifnet_array_isempty(void)
3634 {
3635         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3636         if (ifnet_array->ifnet_count == 0)
3637                 return 1;
3638         else
3639                 return 0;
3640 }
3641
3642 void
3643 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3644 {
3645         struct ifaddr *ifa;
3646
3647         memset(mark, 0, sizeof(*mark));
3648         ifa = &mark->ifa;
3649
3650         mark->ifac.ifa = ifa;
3651
3652         ifa->ifa_addr = &mark->addr;
3653         ifa->ifa_dstaddr = &mark->dstaddr;
3654         ifa->ifa_netmask = &mark->netmask;
3655         ifa->ifa_ifp = ifp;
3656 }