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                         IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
1701                 }
1702
1703                 if (ifp->if_timer == 0 || --ifp->if_timer) {
1704                         crit_exit_gd(gd);
1705                         continue;
1706                 }
1707                 if (ifp->if_watchdog) {
1708                         if (ifnet_tryserialize_all(ifp)) {
1709                                 (*ifp->if_watchdog)(ifp);
1710                                 ifnet_deserialize_all(ifp);
1711                         } else {
1712                                 /* try again next timeout */
1713                                 ++ifp->if_timer;
1714                         }
1715                 }
1716
1717                 crit_exit_gd(gd);
1718         }
1719
1720         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1721 }
1722
1723 static void
1724 if_slowtimo(void *arg __unused)
1725 {
1726         struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1727
1728         KASSERT(mycpuid == 0, ("not on cpu0"));
1729         crit_enter();
1730         if (lmsg->ms_flags & MSGF_DONE)
1731                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1732         crit_exit();
1733 }
1734
1735 /*
1736  * Map interface name to
1737  * interface structure pointer.
1738  */
1739 struct ifnet *
1740 ifunit(const char *name)
1741 {
1742         struct ifnet *ifp;
1743
1744         /*
1745          * Search all the interfaces for this name/number
1746          */
1747         KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1748
1749         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1750                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1751                         break;
1752         }
1753         return (ifp);
1754 }
1755
1756 struct ifnet *
1757 ifunit_netisr(const char *name)
1758 {
1759         const struct ifnet_array *arr;
1760         int i;
1761
1762         /*
1763          * Search all the interfaces for this name/number
1764          */
1765
1766         arr = ifnet_array_get();
1767         for (i = 0; i < arr->ifnet_count; ++i) {
1768                 struct ifnet *ifp = arr->ifnet_arr[i];
1769
1770                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1771                         return ifp;
1772         }
1773         return NULL;
1774 }
1775
1776 /*
1777  * Interface ioctls.
1778  */
1779 int
1780 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1781 {
1782         struct ifnet *ifp;
1783         struct ifreq *ifr;
1784         struct ifstat *ifs;
1785         int error;
1786         short oif_flags;
1787         int new_flags;
1788 #ifdef COMPAT_43
1789         int ocmd;
1790 #endif
1791         size_t namelen, onamelen;
1792         char new_name[IFNAMSIZ];
1793         struct ifaddr *ifa;
1794         struct sockaddr_dl *sdl;
1795
1796         switch (cmd) {
1797         case SIOCGIFCONF:
1798         case OSIOCGIFCONF:
1799                 return (ifconf(cmd, data, cred));
1800         default:
1801                 break;
1802         }
1803
1804         ifr = (struct ifreq *)data;
1805
1806         switch (cmd) {
1807         case SIOCIFCREATE:
1808         case SIOCIFCREATE2:
1809                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1810                         return (error);
1811                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1812                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1813         case SIOCIFDESTROY:
1814                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1815                         return (error);
1816                 return (if_clone_destroy(ifr->ifr_name));
1817         case SIOCIFGCLONERS:
1818                 return (if_clone_list((struct if_clonereq *)data));
1819         default:
1820                 break;
1821         }
1822
1823         /*
1824          * Nominal ioctl through interface, lookup the ifp and obtain a
1825          * lock to serialize the ifconfig ioctl operation.
1826          */
1827         ifnet_lock();
1828
1829         ifp = ifunit(ifr->ifr_name);
1830         if (ifp == NULL) {
1831                 ifnet_unlock();
1832                 return (ENXIO);
1833         }
1834         error = 0;
1835
1836         switch (cmd) {
1837         case SIOCGIFINDEX:
1838                 ifr->ifr_index = ifp->if_index;
1839                 break;
1840
1841         case SIOCGIFFLAGS:
1842                 ifr->ifr_flags = ifp->if_flags;
1843                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1844                 break;
1845
1846         case SIOCGIFCAP:
1847                 ifr->ifr_reqcap = ifp->if_capabilities;
1848                 ifr->ifr_curcap = ifp->if_capenable;
1849                 break;
1850
1851         case SIOCGIFMETRIC:
1852                 ifr->ifr_metric = ifp->if_metric;
1853                 break;
1854
1855         case SIOCGIFMTU:
1856                 ifr->ifr_mtu = ifp->if_mtu;
1857                 break;
1858
1859         case SIOCGIFTSOLEN:
1860                 ifr->ifr_tsolen = ifp->if_tsolen;
1861                 break;
1862
1863         case SIOCGIFDATA:
1864                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1865                                 sizeof(ifp->if_data));
1866                 break;
1867
1868         case SIOCGIFPHYS:
1869                 ifr->ifr_phys = ifp->if_physical;
1870                 break;
1871
1872         case SIOCGIFPOLLCPU:
1873                 ifr->ifr_pollcpu = -1;
1874                 break;
1875
1876         case SIOCSIFPOLLCPU:
1877                 break;
1878
1879         case SIOCSIFFLAGS:
1880                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1881                 if (error)
1882                         break;
1883                 new_flags = (ifr->ifr_flags & 0xffff) |
1884                     (ifr->ifr_flagshigh << 16);
1885                 if (ifp->if_flags & IFF_SMART) {
1886                         /* Smart drivers twiddle their own routes */
1887                 } else if (ifp->if_flags & IFF_UP &&
1888                     (new_flags & IFF_UP) == 0) {
1889                         crit_enter();
1890                         if_down(ifp);
1891                         crit_exit();
1892                 } else if (new_flags & IFF_UP &&
1893                     (ifp->if_flags & IFF_UP) == 0) {
1894                         crit_enter();
1895                         if_up(ifp);
1896                         crit_exit();
1897                 }
1898
1899 #ifdef IFPOLL_ENABLE
1900                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1901                         if (new_flags & IFF_NPOLLING)
1902                                 ifpoll_register(ifp);
1903                         else
1904                                 ifpoll_deregister(ifp);
1905                 }
1906 #endif
1907
1908                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1909                         (new_flags &~ IFF_CANTCHANGE);
1910                 if (new_flags & IFF_PPROMISC) {
1911                         /* Permanently promiscuous mode requested */
1912                         ifp->if_flags |= IFF_PROMISC;
1913                 } else if (ifp->if_pcount == 0) {
1914                         ifp->if_flags &= ~IFF_PROMISC;
1915                 }
1916                 if (ifp->if_ioctl) {
1917                         ifnet_serialize_all(ifp);
1918                         ifp->if_ioctl(ifp, cmd, data, cred);
1919                         ifnet_deserialize_all(ifp);
1920                 }
1921                 getmicrotime(&ifp->if_lastchange);
1922                 break;
1923
1924         case SIOCSIFCAP:
1925                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1926                 if (error)
1927                         break;
1928                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1929                         error = EINVAL;
1930                         break;
1931                 }
1932                 ifnet_serialize_all(ifp);
1933                 ifp->if_ioctl(ifp, cmd, data, cred);
1934                 ifnet_deserialize_all(ifp);
1935                 break;
1936
1937         case SIOCSIFNAME:
1938                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1939                 if (error)
1940                         break;
1941                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1942                 if (error)
1943                         break;
1944                 if (new_name[0] == '\0') {
1945                         error = EINVAL;
1946                         break;
1947                 }
1948                 if (ifunit(new_name) != NULL) {
1949                         error = EEXIST;
1950                         break;
1951                 }
1952
1953                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1954
1955                 /* Announce the departure of the interface. */
1956                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1957
1958                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1959                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1960                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1961                 namelen = strlen(new_name);
1962                 onamelen = sdl->sdl_nlen;
1963                 /*
1964                  * Move the address if needed.  This is safe because we
1965                  * allocate space for a name of length IFNAMSIZ when we
1966                  * create this in if_attach().
1967                  */
1968                 if (namelen != onamelen) {
1969                         bcopy(sdl->sdl_data + onamelen,
1970                             sdl->sdl_data + namelen, sdl->sdl_alen);
1971                 }
1972                 bcopy(new_name, sdl->sdl_data, namelen);
1973                 sdl->sdl_nlen = namelen;
1974                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1975                 bzero(sdl->sdl_data, onamelen);
1976                 while (namelen != 0)
1977                         sdl->sdl_data[--namelen] = 0xff;
1978
1979                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1980
1981                 /* Announce the return of the interface. */
1982                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1983                 break;
1984
1985         case SIOCSIFMETRIC:
1986                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1987                 if (error)
1988                         break;
1989                 ifp->if_metric = ifr->ifr_metric;
1990                 getmicrotime(&ifp->if_lastchange);
1991                 break;
1992
1993         case SIOCSIFPHYS:
1994                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1995                 if (error)
1996                         break;
1997                 if (ifp->if_ioctl == NULL) {
1998                         error = EOPNOTSUPP;
1999                         break;
2000                 }
2001                 ifnet_serialize_all(ifp);
2002                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2003                 ifnet_deserialize_all(ifp);
2004                 if (error == 0)
2005                         getmicrotime(&ifp->if_lastchange);
2006                 break;
2007
2008         case SIOCSIFMTU:
2009         {
2010                 u_long oldmtu = ifp->if_mtu;
2011
2012                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2013                 if (error)
2014                         break;
2015                 if (ifp->if_ioctl == NULL) {
2016                         error = EOPNOTSUPP;
2017                         break;
2018                 }
2019                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2020                         error = EINVAL;
2021                         break;
2022                 }
2023                 ifnet_serialize_all(ifp);
2024                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2025                 ifnet_deserialize_all(ifp);
2026                 if (error == 0) {
2027                         getmicrotime(&ifp->if_lastchange);
2028                         rt_ifmsg(ifp);
2029                 }
2030                 /*
2031                  * If the link MTU changed, do network layer specific procedure.
2032                  */
2033                 if (ifp->if_mtu != oldmtu) {
2034 #ifdef INET6
2035                         nd6_setmtu(ifp);
2036 #endif
2037                 }
2038                 break;
2039         }
2040
2041         case SIOCSIFTSOLEN:
2042                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2043                 if (error)
2044                         break;
2045
2046                 /* XXX need driver supplied upper limit */
2047                 if (ifr->ifr_tsolen <= 0) {
2048                         error = EINVAL;
2049                         break;
2050                 }
2051                 ifp->if_tsolen = ifr->ifr_tsolen;
2052                 break;
2053
2054         case SIOCADDMULTI:
2055         case SIOCDELMULTI:
2056                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2057                 if (error)
2058                         break;
2059
2060                 /* Don't allow group membership on non-multicast interfaces. */
2061                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2062                         error = EOPNOTSUPP;
2063                         break;
2064                 }
2065
2066                 /* Don't let users screw up protocols' entries. */
2067                 if (ifr->ifr_addr.sa_family != AF_LINK) {
2068                         error = EINVAL;
2069                         break;
2070                 }
2071
2072                 if (cmd == SIOCADDMULTI) {
2073                         struct ifmultiaddr *ifma;
2074                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2075                 } else {
2076                         error = if_delmulti(ifp, &ifr->ifr_addr);
2077                 }
2078                 if (error == 0)
2079                         getmicrotime(&ifp->if_lastchange);
2080                 break;
2081
2082         case SIOCSIFPHYADDR:
2083         case SIOCDIFPHYADDR:
2084 #ifdef INET6
2085         case SIOCSIFPHYADDR_IN6:
2086 #endif
2087         case SIOCSLIFPHYADDR:
2088         case SIOCSIFMEDIA:
2089         case SIOCSIFGENERIC:
2090                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2091                 if (error)
2092                         break;
2093                 if (ifp->if_ioctl == 0) {
2094                         error = EOPNOTSUPP;
2095                         break;
2096                 }
2097                 ifnet_serialize_all(ifp);
2098                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2099                 ifnet_deserialize_all(ifp);
2100                 if (error == 0)
2101                         getmicrotime(&ifp->if_lastchange);
2102                 break;
2103
2104         case SIOCGIFSTATUS:
2105                 ifs = (struct ifstat *)data;
2106                 ifs->ascii[0] = '\0';
2107                 /* fall through */
2108         case SIOCGIFPSRCADDR:
2109         case SIOCGIFPDSTADDR:
2110         case SIOCGLIFPHYADDR:
2111         case SIOCGIFMEDIA:
2112         case SIOCGIFGENERIC:
2113                 if (ifp->if_ioctl == NULL) {
2114                         error = EOPNOTSUPP;
2115                         break;
2116                 }
2117                 ifnet_serialize_all(ifp);
2118                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2119                 ifnet_deserialize_all(ifp);
2120                 break;
2121
2122         case SIOCSIFLLADDR:
2123                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2124                 if (error)
2125                         break;
2126                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2127                                      ifr->ifr_addr.sa_len);
2128                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2129                 break;
2130
2131         default:
2132                 oif_flags = ifp->if_flags;
2133                 if (so->so_proto == 0) {
2134                         error = EOPNOTSUPP;
2135                         break;
2136                 }
2137 #ifndef COMPAT_43
2138                 error = so_pru_control_direct(so, cmd, data, ifp);
2139 #else
2140                 ocmd = cmd;
2141
2142                 switch (cmd) {
2143                 case SIOCSIFDSTADDR:
2144                 case SIOCSIFADDR:
2145                 case SIOCSIFBRDADDR:
2146                 case SIOCSIFNETMASK:
2147 #if BYTE_ORDER != BIG_ENDIAN
2148                         if (ifr->ifr_addr.sa_family == 0 &&
2149                             ifr->ifr_addr.sa_len < 16) {
2150                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2151                                 ifr->ifr_addr.sa_len = 16;
2152                         }
2153 #else
2154                         if (ifr->ifr_addr.sa_len == 0)
2155                                 ifr->ifr_addr.sa_len = 16;
2156 #endif
2157                         break;
2158                 case OSIOCGIFADDR:
2159                         cmd = SIOCGIFADDR;
2160                         break;
2161                 case OSIOCGIFDSTADDR:
2162                         cmd = SIOCGIFDSTADDR;
2163                         break;
2164                 case OSIOCGIFBRDADDR:
2165                         cmd = SIOCGIFBRDADDR;
2166                         break;
2167                 case OSIOCGIFNETMASK:
2168                         cmd = SIOCGIFNETMASK;
2169                         break;
2170                 default:
2171                         break;
2172                 }
2173
2174                 error = so_pru_control_direct(so, cmd, data, ifp);
2175
2176                 switch (ocmd) {
2177                 case OSIOCGIFADDR:
2178                 case OSIOCGIFDSTADDR:
2179                 case OSIOCGIFBRDADDR:
2180                 case OSIOCGIFNETMASK:
2181                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2182                         break;
2183                 }
2184 #endif /* COMPAT_43 */
2185
2186                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2187 #ifdef INET6
2188                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
2189                         if (ifp->if_flags & IFF_UP) {
2190                                 crit_enter();
2191                                 in6_if_up(ifp);
2192                                 crit_exit();
2193                         }
2194 #endif
2195                 }
2196                 break;
2197         }
2198
2199         ifnet_unlock();
2200         return (error);
2201 }
2202
2203 /*
2204  * Set/clear promiscuous mode on interface ifp based on the truth value
2205  * of pswitch.  The calls are reference counted so that only the first
2206  * "on" request actually has an effect, as does the final "off" request.
2207  * Results are undefined if the "off" and "on" requests are not matched.
2208  */
2209 int
2210 ifpromisc(struct ifnet *ifp, int pswitch)
2211 {
2212         struct ifreq ifr;
2213         int error;
2214         int oldflags;
2215
2216         oldflags = ifp->if_flags;
2217         if (ifp->if_flags & IFF_PPROMISC) {
2218                 /* Do nothing if device is in permanently promiscuous mode */
2219                 ifp->if_pcount += pswitch ? 1 : -1;
2220                 return (0);
2221         }
2222         if (pswitch) {
2223                 /*
2224                  * If the device is not configured up, we cannot put it in
2225                  * promiscuous mode.
2226                  */
2227                 if ((ifp->if_flags & IFF_UP) == 0)
2228                         return (ENETDOWN);
2229                 if (ifp->if_pcount++ != 0)
2230                         return (0);
2231                 ifp->if_flags |= IFF_PROMISC;
2232                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
2233                     ifp->if_xname);
2234         } else {
2235                 if (--ifp->if_pcount > 0)
2236                         return (0);
2237                 ifp->if_flags &= ~IFF_PROMISC;
2238                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2239                     ifp->if_xname);
2240         }
2241         ifr.ifr_flags = ifp->if_flags;
2242         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2243         ifnet_serialize_all(ifp);
2244         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2245         ifnet_deserialize_all(ifp);
2246         if (error == 0)
2247                 rt_ifmsg(ifp);
2248         else
2249                 ifp->if_flags = oldflags;
2250         return error;
2251 }
2252
2253 /*
2254  * Return interface configuration
2255  * of system.  List may be used
2256  * in later ioctl's (above) to get
2257  * other information.
2258  */
2259 static int
2260 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2261 {
2262         struct ifconf *ifc = (struct ifconf *)data;
2263         struct ifnet *ifp;
2264         struct sockaddr *sa;
2265         struct ifreq ifr, *ifrp;
2266         int space = ifc->ifc_len, error = 0;
2267
2268         ifrp = ifc->ifc_req;
2269
2270         ifnet_lock();
2271         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2272                 struct ifaddr_container *ifac, *ifac_mark;
2273                 struct ifaddr_marker mark;
2274                 struct ifaddrhead *head;
2275                 int addrs;
2276
2277                 if (space <= sizeof ifr)
2278                         break;
2279
2280                 /*
2281                  * Zero the stack declared structure first to prevent
2282                  * memory disclosure.
2283                  */
2284                 bzero(&ifr, sizeof(ifr));
2285                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2286                     >= sizeof(ifr.ifr_name)) {
2287                         error = ENAMETOOLONG;
2288                         break;
2289                 }
2290
2291                 /*
2292                  * Add a marker, since copyout() could block and during that
2293                  * period the list could be changed.  Inserting the marker to
2294                  * the header of the list will not cause trouble for the code
2295                  * assuming that the first element of the list is AF_LINK; the
2296                  * marker will be moved to the next position w/o blocking.
2297                  */
2298                 ifa_marker_init(&mark, ifp);
2299                 ifac_mark = &mark.ifac;
2300                 head = &ifp->if_addrheads[mycpuid];
2301
2302                 addrs = 0;
2303                 TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2304                 while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2305                         struct ifaddr *ifa = ifac->ifa;
2306
2307                         TAILQ_REMOVE(head, ifac_mark, ifa_link);
2308                         TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2309
2310                         /* Ignore marker */
2311                         if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2312                                 continue;
2313
2314                         if (space <= sizeof ifr)
2315                                 break;
2316                         sa = ifa->ifa_addr;
2317                         if (cred->cr_prison &&
2318                             prison_if(cred, sa))
2319                                 continue;
2320                         addrs++;
2321                         /*
2322                          * Keep a reference on this ifaddr, so that it will
2323                          * not be destroyed when its address is copied to
2324                          * the userland, which could block.
2325                          */
2326                         IFAREF(ifa);
2327 #ifdef COMPAT_43
2328                         if (cmd == OSIOCGIFCONF) {
2329                                 struct osockaddr *osa =
2330                                          (struct osockaddr *)&ifr.ifr_addr;
2331                                 ifr.ifr_addr = *sa;
2332                                 osa->sa_family = sa->sa_family;
2333                                 error = copyout(&ifr, ifrp, sizeof ifr);
2334                                 ifrp++;
2335                         } else
2336 #endif
2337                         if (sa->sa_len <= sizeof(*sa)) {
2338                                 ifr.ifr_addr = *sa;
2339                                 error = copyout(&ifr, ifrp, sizeof ifr);
2340                                 ifrp++;
2341                         } else {
2342                                 if (space < (sizeof ifr) + sa->sa_len -
2343                                             sizeof(*sa)) {
2344                                         IFAFREE(ifa);
2345                                         break;
2346                                 }
2347                                 space -= sa->sa_len - sizeof(*sa);
2348                                 error = copyout(&ifr, ifrp,
2349                                                 sizeof ifr.ifr_name);
2350                                 if (error == 0)
2351                                         error = copyout(sa, &ifrp->ifr_addr,
2352                                                         sa->sa_len);
2353                                 ifrp = (struct ifreq *)
2354                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2355                         }
2356                         IFAFREE(ifa);
2357                         if (error)
2358                                 break;
2359                         space -= sizeof ifr;
2360                 }
2361                 TAILQ_REMOVE(head, ifac_mark, ifa_link);
2362                 if (error)
2363                         break;
2364                 if (!addrs) {
2365                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2366                         error = copyout(&ifr, ifrp, sizeof ifr);
2367                         if (error)
2368                                 break;
2369                         space -= sizeof ifr;
2370                         ifrp++;
2371                 }
2372         }
2373         ifnet_unlock();
2374
2375         ifc->ifc_len -= space;
2376         return (error);
2377 }
2378
2379 /*
2380  * Just like if_promisc(), but for all-multicast-reception mode.
2381  */
2382 int
2383 if_allmulti(struct ifnet *ifp, int onswitch)
2384 {
2385         int error = 0;
2386         struct ifreq ifr;
2387
2388         crit_enter();
2389
2390         if (onswitch) {
2391                 if (ifp->if_amcount++ == 0) {
2392                         ifp->if_flags |= IFF_ALLMULTI;
2393                         ifr.ifr_flags = ifp->if_flags;
2394                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2395                         ifnet_serialize_all(ifp);
2396                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2397                                               NULL);
2398                         ifnet_deserialize_all(ifp);
2399                 }
2400         } else {
2401                 if (ifp->if_amcount > 1) {
2402                         ifp->if_amcount--;
2403                 } else {
2404                         ifp->if_amcount = 0;
2405                         ifp->if_flags &= ~IFF_ALLMULTI;
2406                         ifr.ifr_flags = ifp->if_flags;
2407                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2408                         ifnet_serialize_all(ifp);
2409                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2410                                               NULL);
2411                         ifnet_deserialize_all(ifp);
2412                 }
2413         }
2414
2415         crit_exit();
2416
2417         if (error == 0)
2418                 rt_ifmsg(ifp);
2419         return error;
2420 }
2421
2422 /*
2423  * Add a multicast listenership to the interface in question.
2424  * The link layer provides a routine which converts
2425  */
2426 int
2427 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2428     struct ifmultiaddr **retifma)
2429 {
2430         struct sockaddr *llsa, *dupsa;
2431         int error;
2432         struct ifmultiaddr *ifma;
2433
2434         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2435
2436         /*
2437          * If the matching multicast address already exists
2438          * then don't add a new one, just add a reference
2439          */
2440         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2441                 if (sa_equal(sa, ifma->ifma_addr)) {
2442                         ifma->ifma_refcount++;
2443                         if (retifma)
2444                                 *retifma = ifma;
2445                         return 0;
2446                 }
2447         }
2448
2449         /*
2450          * Give the link layer a chance to accept/reject it, and also
2451          * find out which AF_LINK address this maps to, if it isn't one
2452          * already.
2453          */
2454         if (ifp->if_resolvemulti) {
2455                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2456                 if (error)
2457                         return error;
2458         } else {
2459                 llsa = NULL;
2460         }
2461
2462         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2463         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2464         bcopy(sa, dupsa, sa->sa_len);
2465
2466         ifma->ifma_addr = dupsa;
2467         ifma->ifma_lladdr = llsa;
2468         ifma->ifma_ifp = ifp;
2469         ifma->ifma_refcount = 1;
2470         ifma->ifma_protospec = NULL;
2471         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2472
2473         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2474         if (retifma)
2475                 *retifma = ifma;
2476
2477         if (llsa != NULL) {
2478                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2479                         if (sa_equal(ifma->ifma_addr, llsa))
2480                                 break;
2481                 }
2482                 if (ifma) {
2483                         ifma->ifma_refcount++;
2484                 } else {
2485                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2486                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2487                         bcopy(llsa, dupsa, llsa->sa_len);
2488                         ifma->ifma_addr = dupsa;
2489                         ifma->ifma_ifp = ifp;
2490                         ifma->ifma_refcount = 1;
2491                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2492                 }
2493         }
2494         /*
2495          * We are certain we have added something, so call down to the
2496          * interface to let them know about it.
2497          */
2498         if (ifp->if_ioctl)
2499                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2500
2501         return 0;
2502 }
2503
2504 int
2505 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2506     struct ifmultiaddr **retifma)
2507 {
2508         int error;
2509
2510         ifnet_serialize_all(ifp);
2511         error = if_addmulti_serialized(ifp, sa, retifma);
2512         ifnet_deserialize_all(ifp);
2513
2514         return error;
2515 }
2516
2517 /*
2518  * Remove a reference to a multicast address on this interface.  Yell
2519  * if the request does not match an existing membership.
2520  */
2521 static int
2522 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2523 {
2524         struct ifmultiaddr *ifma;
2525
2526         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2527
2528         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2529                 if (sa_equal(sa, ifma->ifma_addr))
2530                         break;
2531         if (ifma == NULL)
2532                 return ENOENT;
2533
2534         if (ifma->ifma_refcount > 1) {
2535                 ifma->ifma_refcount--;
2536                 return 0;
2537         }
2538
2539         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2540         sa = ifma->ifma_lladdr;
2541         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2542         /*
2543          * Make sure the interface driver is notified
2544          * in the case of a link layer mcast group being left.
2545          */
2546         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2547                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2548         kfree(ifma->ifma_addr, M_IFMADDR);
2549         kfree(ifma, M_IFMADDR);
2550         if (sa == NULL)
2551                 return 0;
2552
2553         /*
2554          * Now look for the link-layer address which corresponds to
2555          * this network address.  It had been squirreled away in
2556          * ifma->ifma_lladdr for this purpose (so we don't have
2557          * to call ifp->if_resolvemulti() again), and we saved that
2558          * value in sa above.  If some nasty deleted the
2559          * link-layer address out from underneath us, we can deal because
2560          * the address we stored was is not the same as the one which was
2561          * in the record for the link-layer address.  (So we don't complain
2562          * in that case.)
2563          */
2564         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2565                 if (sa_equal(sa, ifma->ifma_addr))
2566                         break;
2567         if (ifma == NULL)
2568                 return 0;
2569
2570         if (ifma->ifma_refcount > 1) {
2571                 ifma->ifma_refcount--;
2572                 return 0;
2573         }
2574
2575         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2576         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2577         kfree(ifma->ifma_addr, M_IFMADDR);
2578         kfree(sa, M_IFMADDR);
2579         kfree(ifma, M_IFMADDR);
2580
2581         return 0;
2582 }
2583
2584 int
2585 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2586 {
2587         int error;
2588
2589         ifnet_serialize_all(ifp);
2590         error = if_delmulti_serialized(ifp, sa);
2591         ifnet_deserialize_all(ifp);
2592
2593         return error;
2594 }
2595
2596 /*
2597  * Delete all multicast group membership for an interface.
2598  * Should be used to quickly flush all multicast filters.
2599  */
2600 void
2601 if_delallmulti_serialized(struct ifnet *ifp)
2602 {
2603         struct ifmultiaddr *ifma, mark;
2604         struct sockaddr sa;
2605
2606         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2607
2608         bzero(&sa, sizeof(sa));
2609         sa.sa_family = AF_UNSPEC;
2610         sa.sa_len = sizeof(sa);
2611
2612         bzero(&mark, sizeof(mark));
2613         mark.ifma_addr = &sa;
2614
2615         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2616         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2617                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2618                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2619                     ifma_link);
2620
2621                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2622                         continue;
2623
2624                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2625         }
2626         TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2627 }
2628
2629
2630 /*
2631  * Set the link layer address on an interface.
2632  *
2633  * At this time we only support certain types of interfaces,
2634  * and we don't allow the length of the address to change.
2635  */
2636 int
2637 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2638 {
2639         struct sockaddr_dl *sdl;
2640         struct ifreq ifr;
2641
2642         sdl = IF_LLSOCKADDR(ifp);
2643         if (sdl == NULL)
2644                 return (EINVAL);
2645         if (len != sdl->sdl_alen)       /* don't allow length to change */
2646                 return (EINVAL);
2647         switch (ifp->if_type) {
2648         case IFT_ETHER:                 /* these types use struct arpcom */
2649         case IFT_XETHER:
2650         case IFT_L2VLAN:
2651         case IFT_IEEE8023ADLAG:
2652                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2653                 bcopy(lladdr, LLADDR(sdl), len);
2654                 break;
2655         default:
2656                 return (ENODEV);
2657         }
2658         /*
2659          * If the interface is already up, we need
2660          * to re-init it in order to reprogram its
2661          * address filter.
2662          */
2663         ifnet_serialize_all(ifp);
2664         if ((ifp->if_flags & IFF_UP) != 0) {
2665 #ifdef INET
2666                 struct ifaddr_container *ifac;
2667 #endif
2668
2669                 ifp->if_flags &= ~IFF_UP;
2670                 ifr.ifr_flags = ifp->if_flags;
2671                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2672                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2673                               NULL);
2674                 ifp->if_flags |= IFF_UP;
2675                 ifr.ifr_flags = ifp->if_flags;
2676                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2677                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2678                                  NULL);
2679 #ifdef INET
2680                 /*
2681                  * Also send gratuitous ARPs to notify other nodes about
2682                  * the address change.
2683                  */
2684                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2685                         struct ifaddr *ifa = ifac->ifa;
2686
2687                         if (ifa->ifa_addr != NULL &&
2688                             ifa->ifa_addr->sa_family == AF_INET)
2689                                 arp_gratuitous(ifp, ifa);
2690                 }
2691 #endif
2692         }
2693         ifnet_deserialize_all(ifp);
2694         return (0);
2695 }
2696
2697 struct ifmultiaddr *
2698 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2699 {
2700         struct ifmultiaddr *ifma;
2701
2702         /* TODO: need ifnet_serialize_main */
2703         ifnet_serialize_all(ifp);
2704         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2705                 if (sa_equal(ifma->ifma_addr, sa))
2706                         break;
2707         ifnet_deserialize_all(ifp);
2708
2709         return ifma;
2710 }
2711
2712 /*
2713  * This function locates the first real ethernet MAC from a network
2714  * card and loads it into node, returning 0 on success or ENOENT if
2715  * no suitable interfaces were found.  It is used by the uuid code to
2716  * generate a unique 6-byte number.
2717  */
2718 int
2719 if_getanyethermac(uint16_t *node, int minlen)
2720 {
2721         struct ifnet *ifp;
2722         struct sockaddr_dl *sdl;
2723
2724         ifnet_lock();
2725         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2726                 if (ifp->if_type != IFT_ETHER)
2727                         continue;
2728                 sdl = IF_LLSOCKADDR(ifp);
2729                 if (sdl->sdl_alen < minlen)
2730                         continue;
2731                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2732                       minlen);
2733                 ifnet_unlock();
2734                 return(0);
2735         }
2736         ifnet_unlock();
2737         return (ENOENT);
2738 }
2739
2740 /*
2741  * The name argument must be a pointer to storage which will last as
2742  * long as the interface does.  For physical devices, the result of
2743  * device_get_name(dev) is a good choice and for pseudo-devices a
2744  * static string works well.
2745  */
2746 void
2747 if_initname(struct ifnet *ifp, const char *name, int unit)
2748 {
2749         ifp->if_dname = name;
2750         ifp->if_dunit = unit;
2751         if (unit != IF_DUNIT_NONE)
2752                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2753         else
2754                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2755 }
2756
2757 int
2758 if_printf(struct ifnet *ifp, const char *fmt, ...)
2759 {
2760         __va_list ap;
2761         int retval;
2762
2763         retval = kprintf("%s: ", ifp->if_xname);
2764         __va_start(ap, fmt);
2765         retval += kvprintf(fmt, ap);
2766         __va_end(ap);
2767         return (retval);
2768 }
2769
2770 struct ifnet *
2771 if_alloc(uint8_t type)
2772 {
2773         struct ifnet *ifp;
2774         size_t size;
2775
2776         /*
2777          * XXX temporary hack until arpcom is setup in if_l2com
2778          */
2779         if (type == IFT_ETHER)
2780                 size = sizeof(struct arpcom);
2781         else
2782                 size = sizeof(struct ifnet);
2783
2784         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2785
2786         ifp->if_type = type;
2787
2788         if (if_com_alloc[type] != NULL) {
2789                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2790                 if (ifp->if_l2com == NULL) {
2791                         kfree(ifp, M_IFNET);
2792                         return (NULL);
2793                 }
2794         }
2795         return (ifp);
2796 }
2797
2798 void
2799 if_free(struct ifnet *ifp)
2800 {
2801         kfree(ifp, M_IFNET);
2802 }
2803
2804 void
2805 ifq_set_classic(struct ifaltq *ifq)
2806 {
2807         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2808             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2809 }
2810
2811 void
2812 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2813     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2814 {
2815         int q;
2816
2817         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2818         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2819         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2820         KASSERT(request != NULL, ("request is not specified"));
2821
2822         ifq->altq_mapsubq = mapsubq;
2823         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2824                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2825
2826                 ifsq->ifsq_enqueue = enqueue;
2827                 ifsq->ifsq_dequeue = dequeue;
2828                 ifsq->ifsq_request = request;
2829         }
2830 }
2831
2832 static void
2833 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2834 {
2835         m->m_nextpkt = NULL;
2836         if (ifsq->ifsq_norm_tail == NULL)
2837                 ifsq->ifsq_norm_head = m;
2838         else
2839                 ifsq->ifsq_norm_tail->m_nextpkt = m;
2840         ifsq->ifsq_norm_tail = m;
2841         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2842 }
2843
2844 static void
2845 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2846 {
2847         m->m_nextpkt = NULL;
2848         if (ifsq->ifsq_prio_tail == NULL)
2849                 ifsq->ifsq_prio_head = m;
2850         else
2851                 ifsq->ifsq_prio_tail->m_nextpkt = m;
2852         ifsq->ifsq_prio_tail = m;
2853         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2854         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2855 }
2856
2857 static struct mbuf *
2858 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2859 {
2860         struct mbuf *m;
2861
2862         m = ifsq->ifsq_norm_head;
2863         if (m != NULL) {
2864                 if ((ifsq->ifsq_norm_head = m->m_nextpkt) == NULL)
2865                         ifsq->ifsq_norm_tail = NULL;
2866                 m->m_nextpkt = NULL;
2867                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2868         }
2869         return m;
2870 }
2871
2872 static struct mbuf *
2873 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2874 {
2875         struct mbuf *m;
2876
2877         m = ifsq->ifsq_prio_head;
2878         if (m != NULL) {
2879                 if ((ifsq->ifsq_prio_head = m->m_nextpkt) == NULL)
2880                         ifsq->ifsq_prio_tail = NULL;
2881                 m->m_nextpkt = NULL;
2882                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2883                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2884         }
2885         return m;
2886 }
2887
2888 int
2889 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2890     struct altq_pktattr *pa __unused)
2891 {
2892         M_ASSERTPKTHDR(m);
2893         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2894             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2895                 if ((m->m_flags & M_PRIO) &&
2896                     ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen / 2) &&
2897                     ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt / 2)) {
2898                         struct mbuf *m_drop;
2899
2900                         /*
2901                          * Perform drop-head on normal queue
2902                          */
2903                         m_drop = ifsq_norm_dequeue(ifsq);
2904                         if (m_drop != NULL) {
2905                                 m_freem(m_drop);
2906                                 ifsq_prio_enqueue(ifsq, m);
2907                                 return 0;
2908                         }
2909                         /* XXX nothing could be dropped? */
2910                 }
2911                 m_freem(m);
2912                 return ENOBUFS;
2913         } else {
2914                 if (m->m_flags & M_PRIO)
2915                         ifsq_prio_enqueue(ifsq, m);
2916                 else
2917                         ifsq_norm_enqueue(ifsq, m);
2918                 return 0;
2919         }
2920 }
2921
2922 struct mbuf *
2923 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2924 {
2925         struct mbuf *m;
2926
2927         switch (op) {
2928         case ALTDQ_POLL:
2929                 m = ifsq->ifsq_prio_head;
2930                 if (m == NULL)
2931                         m = ifsq->ifsq_norm_head;
2932                 break;
2933
2934         case ALTDQ_REMOVE:
2935                 m = ifsq_prio_dequeue(ifsq);
2936                 if (m == NULL)
2937                         m = ifsq_norm_dequeue(ifsq);
2938                 break;
2939
2940         default:
2941                 panic("unsupported ALTQ dequeue op: %d", op);
2942         }
2943         return m;
2944 }
2945
2946 int
2947 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2948 {
2949         switch (req) {
2950         case ALTRQ_PURGE:
2951                 for (;;) {
2952                         struct mbuf *m;
2953
2954                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2955                         if (m == NULL)
2956                                 break;
2957                         m_freem(m);
2958                 }
2959                 break;
2960
2961         default:
2962                 panic("unsupported ALTQ request: %d", req);
2963         }
2964         return 0;
2965 }
2966
2967 static void
2968 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2969 {
2970         struct ifnet *ifp = ifsq_get_ifp(ifsq);
2971         int running = 0, need_sched;
2972
2973         /*
2974          * Try to do direct ifnet.if_start on the subqueue first, if there is
2975          * contention on the subqueue hardware serializer, ifnet.if_start on
2976          * the subqueue will be scheduled on the subqueue owner CPU.
2977          */
2978         if (!ifsq_tryserialize_hw(ifsq)) {
2979                 /*
2980                  * Subqueue hardware serializer contention happened,
2981                  * ifnet.if_start on the subqueue is scheduled on
2982                  * the subqueue owner CPU, and we keep going.
2983                  */
2984                 ifsq_ifstart_schedule(ifsq, 1);
2985                 return;
2986         }
2987
2988         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2989                 ifp->if_start(ifp, ifsq);
2990                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2991                         running = 1;
2992         }
2993         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2994
2995         ifsq_deserialize_hw(ifsq);
2996
2997         if (need_sched) {
2998                 /*
2999                  * More data need to be transmitted, ifnet.if_start on the
3000                  * subqueue is scheduled on the subqueue owner CPU, and we
3001                  * keep going.
3002                  * NOTE: ifnet.if_start subqueue interlock is not released.
3003                  */
3004                 ifsq_ifstart_schedule(ifsq, force_sched);
3005         }
3006 }
3007
3008 /*
3009  * Subqeue packets staging mechanism:
3010  *
3011  * The packets enqueued into the subqueue are staged to a certain amount
3012  * before the ifnet.if_start on the subqueue is called.  In this way, the
3013  * driver could avoid writing to hardware registers upon every packet,
3014  * instead, hardware registers could be written when certain amount of
3015  * packets are put onto hardware TX ring.  The measurement on several modern
3016  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
3017  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
3018  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
3019  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
3020  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
3021  *
3022  * Subqueue packets staging is performed for two entry points into drivers'
3023  * transmission function:
3024  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
3025  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
3026  *
3027  * Subqueue packets staging will be stopped upon any of the following
3028  * conditions:
3029  * - If the count of packets enqueued on the current CPU is great than or
3030  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
3031  * - If the total length of packets enqueued on the current CPU is great
3032  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
3033  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
3034  *   is usually less than hardware's MTU.
3035  * - ifsq_ifstart_schedule() is not pending on the current CPU and
3036  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
3037  *   released.
3038  * - The if_start_rollup(), which is registered as low priority netisr
3039  *   rollup function, is called; probably because no more work is pending
3040  *   for netisr.
3041  *
3042  * NOTE:
3043  * Currently subqueue packet staging is only performed in netisr threads.
3044  */
3045 int
3046 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
3047 {
3048         struct ifaltq *ifq = &ifp->if_snd;
3049         struct ifaltq_subque *ifsq;
3050         int error, start = 0, len, mcast = 0, avoid_start = 0;
3051         struct ifsubq_stage_head *head = NULL;
3052         struct ifsubq_stage *stage = NULL;
3053         struct globaldata *gd = mycpu;
3054         struct thread *td = gd->gd_curthread;
3055
3056         crit_enter_quick(td);
3057
3058         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
3059         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3060
3061         len = m->m_pkthdr.len;
3062         if (m->m_flags & M_MCAST)
3063                 mcast = 1;
3064
3065         if (td->td_type == TD_TYPE_NETISR) {
3066                 head = &ifsubq_stage_heads[mycpuid];
3067                 stage = ifsq_get_stage(ifsq, mycpuid);
3068
3069                 stage->stg_cnt++;
3070                 stage->stg_len += len;
3071                 if (stage->stg_cnt < ifsq_stage_cntmax &&
3072                     stage->stg_len < (ifp->if_mtu - max_protohdr))
3073                         avoid_start = 1;
3074         }
3075
3076         ALTQ_SQ_LOCK(ifsq);
3077         error = ifsq_enqueue_locked(ifsq, m, pa);
3078         if (error) {
3079                 IFNET_STAT_INC(ifp, oqdrops, 1);
3080                 if (!ifsq_data_ready(ifsq)) {
3081                         ALTQ_SQ_UNLOCK(ifsq);
3082                         crit_exit_quick(td);
3083                         return error;
3084                 }
3085                 avoid_start = 0;
3086         }
3087         if (!ifsq_is_started(ifsq)) {
3088                 if (avoid_start) {
3089                         ALTQ_SQ_UNLOCK(ifsq);
3090
3091                         KKASSERT(!error);
3092                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3093                                 ifsq_stage_insert(head, stage);
3094
3095                         IFNET_STAT_INC(ifp, obytes, len);
3096                         if (mcast)
3097                                 IFNET_STAT_INC(ifp, omcasts, 1);
3098                         crit_exit_quick(td);
3099                         return error;
3100                 }
3101
3102                 /*
3103                  * Hold the subqueue interlock of ifnet.if_start
3104                  */
3105                 ifsq_set_started(ifsq);
3106                 start = 1;
3107         }
3108         ALTQ_SQ_UNLOCK(ifsq);
3109
3110         if (!error) {
3111                 IFNET_STAT_INC(ifp, obytes, len);
3112                 if (mcast)
3113                         IFNET_STAT_INC(ifp, omcasts, 1);
3114         }
3115
3116         if (stage != NULL) {
3117                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3118                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3119                         if (!avoid_start) {
3120                                 ifsq_stage_remove(head, stage);
3121                                 ifsq_ifstart_schedule(ifsq, 1);
3122                         }
3123                         crit_exit_quick(td);
3124                         return error;
3125                 }
3126
3127                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3128                         ifsq_stage_remove(head, stage);
3129                 } else {
3130                         stage->stg_cnt = 0;
3131                         stage->stg_len = 0;
3132                 }
3133         }
3134
3135         if (!start) {
3136                 crit_exit_quick(td);
3137                 return error;
3138         }
3139
3140         ifsq_ifstart_try(ifsq, 0);
3141
3142         crit_exit_quick(td);
3143         return error;
3144 }
3145
3146 void *
3147 ifa_create(int size)
3148 {
3149         struct ifaddr *ifa;
3150         int i;
3151
3152         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3153
3154         ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3155         ifa->ifa_containers =
3156             kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
3157                 M_IFADDR, M_INTWAIT | M_ZERO);
3158
3159         ifa->ifa_ncnt = ncpus;
3160         for (i = 0; i < ncpus; ++i) {
3161                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3162
3163                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3164                 ifac->ifa = ifa;
3165                 ifac->ifa_refcnt = 1;
3166         }
3167 #ifdef IFADDR_DEBUG
3168         kprintf("alloc ifa %p %d\n", ifa, size);
3169 #endif
3170         return ifa;
3171 }
3172
3173 void
3174 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3175 {
3176         struct ifaddr *ifa = ifac->ifa;
3177
3178         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3179         KKASSERT(ifac->ifa_refcnt == 0);
3180         KASSERT(ifac->ifa_listmask == 0,
3181                 ("ifa is still on %#x lists", ifac->ifa_listmask));
3182
3183         ifac->ifa_magic = IFA_CONTAINER_DEAD;
3184
3185 #ifdef IFADDR_DEBUG_VERBOSE
3186         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3187 #endif
3188
3189         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3190                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
3191         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3192 #ifdef IFADDR_DEBUG
3193                 kprintf("free ifa %p\n", ifa);
3194 #endif
3195                 kfree(ifa->ifa_containers, M_IFADDR);
3196                 kfree(ifa, M_IFADDR);
3197         }
3198 }
3199
3200 static void
3201 ifa_iflink_dispatch(netmsg_t nmsg)
3202 {
3203         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3204         struct ifaddr *ifa = msg->ifa;
3205         struct ifnet *ifp = msg->ifp;
3206         int cpu = mycpuid;
3207         struct ifaddr_container *ifac;
3208
3209         crit_enter();
3210
3211         ifac = &ifa->ifa_containers[cpu];
3212         ASSERT_IFAC_VALID(ifac);
3213         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3214                 ("ifaddr is on if_addrheads"));
3215
3216         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3217         if (msg->tail)
3218                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3219         else
3220                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3221
3222         crit_exit();
3223
3224         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
3225 }
3226
3227 void
3228 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3229 {
3230         struct netmsg_ifaddr msg;
3231
3232         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3233                     0, ifa_iflink_dispatch);
3234         msg.ifa = ifa;
3235         msg.ifp = ifp;
3236         msg.tail = tail;
3237
3238         ifa_domsg(&msg.base.lmsg, 0);
3239 }
3240
3241 static void
3242 ifa_ifunlink_dispatch(netmsg_t nmsg)
3243 {
3244         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3245         struct ifaddr *ifa = msg->ifa;
3246         struct ifnet *ifp = msg->ifp;
3247         int cpu = mycpuid;
3248         struct ifaddr_container *ifac;
3249
3250         crit_enter();
3251
3252         ifac = &ifa->ifa_containers[cpu];
3253         ASSERT_IFAC_VALID(ifac);
3254         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3255                 ("ifaddr is not on if_addrhead"));
3256
3257         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3258         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3259
3260         crit_exit();
3261
3262         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
3263 }
3264
3265 void
3266 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3267 {
3268         struct netmsg_ifaddr msg;
3269
3270         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3271                     0, ifa_ifunlink_dispatch);
3272         msg.ifa = ifa;
3273         msg.ifp = ifp;
3274
3275         ifa_domsg(&msg.base.lmsg, 0);
3276 }
3277
3278 static void
3279 ifa_destroy_dispatch(netmsg_t nmsg)
3280 {
3281         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3282
3283         IFAFREE(msg->ifa);
3284         ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3285 }
3286
3287 void
3288 ifa_destroy(struct ifaddr *ifa)
3289 {
3290         struct netmsg_ifaddr msg;
3291
3292         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3293                     0, ifa_destroy_dispatch);
3294         msg.ifa = ifa;
3295
3296         ifa_domsg(&msg.base.lmsg, 0);
3297 }
3298
3299 struct lwkt_port *
3300 ifnet_portfn(int cpu)
3301 {
3302         return &ifnet_threads[cpu]->td_msgport;
3303 }
3304
3305 void
3306 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
3307 {
3308         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
3309
3310         if (next_cpu < ncpus)
3311                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
3312         else
3313                 lwkt_replymsg(lmsg, 0);
3314 }
3315
3316 int
3317 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
3318 {
3319         KKASSERT(cpu < ncpus);
3320         return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
3321 }
3322
3323 void
3324 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
3325 {
3326         KKASSERT(cpu < ncpus);
3327         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
3328 }
3329
3330 /*
3331  * Generic netmsg service loop.  Some protocols may roll their own but all
3332  * must do the basic command dispatch function call done here.
3333  */
3334 static void
3335 ifnet_service_loop(void *arg __unused)
3336 {
3337         netmsg_t msg;
3338
3339         while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
3340                 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
3341                 msg->base.nm_dispatch(msg);
3342         }
3343 }
3344
3345 static void
3346 if_start_rollup(void)
3347 {
3348         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3349         struct ifsubq_stage *stage;
3350
3351         crit_enter();
3352
3353         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3354                 struct ifaltq_subque *ifsq = stage->stg_subq;
3355                 int is_sched = 0;
3356
3357                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3358                         is_sched = 1;
3359                 ifsq_stage_remove(head, stage);
3360
3361                 if (is_sched) {
3362                         ifsq_ifstart_schedule(ifsq, 1);
3363                 } else {
3364                         int start = 0;
3365
3366                         ALTQ_SQ_LOCK(ifsq);
3367                         if (!ifsq_is_started(ifsq)) {
3368                                 /*
3369                                  * Hold the subqueue interlock of
3370                                  * ifnet.if_start
3371                                  */
3372                                 ifsq_set_started(ifsq);
3373                                 start = 1;
3374                         }
3375                         ALTQ_SQ_UNLOCK(ifsq);
3376
3377                         if (start)
3378                                 ifsq_ifstart_try(ifsq, 1);
3379                 }
3380                 KKASSERT((stage->stg_flags &
3381                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3382         }
3383
3384         crit_exit();
3385 }
3386
3387 static void
3388 ifnetinit(void *dummy __unused)
3389 {
3390         int i;
3391
3392         for (i = 0; i < ncpus; ++i) {
3393                 struct thread **thr = &ifnet_threads[i];
3394
3395                 lwkt_create(ifnet_service_loop, NULL, thr, NULL,
3396                             TDF_NOSTART|TDF_FORCE_SPINPORT|TDF_FIXEDCPU,
3397                             i, "ifnet %d", i);
3398                 netmsg_service_port_init(&(*thr)->td_msgport);
3399                 lwkt_schedule(*thr);
3400         }
3401
3402         for (i = 0; i < ncpus; ++i)
3403                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3404         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3405 }
3406
3407 void
3408 if_register_com_alloc(u_char type,
3409     if_com_alloc_t *a, if_com_free_t *f)
3410 {
3411
3412         KASSERT(if_com_alloc[type] == NULL,
3413             ("if_register_com_alloc: %d already registered", type));
3414         KASSERT(if_com_free[type] == NULL,
3415             ("if_register_com_alloc: %d free already registered", type));
3416
3417         if_com_alloc[type] = a;
3418         if_com_free[type] = f;
3419 }
3420
3421 void
3422 if_deregister_com_alloc(u_char type)
3423 {
3424
3425         KASSERT(if_com_alloc[type] != NULL,
3426             ("if_deregister_com_alloc: %d not registered", type));
3427         KASSERT(if_com_free[type] != NULL,
3428             ("if_deregister_com_alloc: %d free not registered", type));
3429         if_com_alloc[type] = NULL;
3430         if_com_free[type] = NULL;
3431 }
3432
3433 int
3434 if_ring_count2(int cnt, int cnt_max)
3435 {
3436         int shift = 0;
3437
3438         KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3439             ("invalid ring count max %d", cnt_max));
3440
3441         if (cnt <= 0)
3442                 cnt = cnt_max;
3443         if (cnt > ncpus2)
3444                 cnt = ncpus2;
3445         if (cnt > cnt_max)
3446                 cnt = cnt_max;
3447
3448         while ((1 << (shift + 1)) <= cnt)
3449                 ++shift;
3450         cnt = 1 << shift;
3451
3452         KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3453             ("calculate cnt %d, ncpus2 %d, cnt max %d",
3454              cnt, ncpus2, cnt_max));
3455         return cnt;
3456 }
3457
3458 void
3459 ifq_set_maxlen(struct ifaltq *ifq, int len)
3460 {
3461         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3462 }
3463
3464 int
3465 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3466 {
3467         return ALTQ_SUBQ_INDEX_DEFAULT;
3468 }
3469
3470 int
3471 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3472 {
3473         return (cpuid & ifq->altq_subq_mask);
3474 }
3475
3476 static void
3477 ifsq_watchdog(void *arg)
3478 {
3479         struct ifsubq_watchdog *wd = arg;
3480         struct ifnet *ifp;
3481
3482         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3483                 goto done;
3484
3485         ifp = ifsq_get_ifp(wd->wd_subq);
3486         if (ifnet_tryserialize_all(ifp)) {
3487                 wd->wd_watchdog(wd->wd_subq);
3488                 ifnet_deserialize_all(ifp);
3489         } else {
3490                 /* try again next timeout */
3491                 wd->wd_timer = 1;
3492         }
3493 done:
3494         ifsq_watchdog_reset(wd);
3495 }
3496
3497 static void
3498 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3499 {
3500         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3501             ifsq_get_cpuid(wd->wd_subq));
3502 }
3503
3504 void
3505 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3506     ifsq_watchdog_t watchdog)
3507 {
3508         callout_init_mp(&wd->wd_callout);
3509         wd->wd_timer = 0;
3510         wd->wd_subq = ifsq;
3511         wd->wd_watchdog = watchdog;
3512 }
3513
3514 void
3515 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3516 {
3517         wd->wd_timer = 0;
3518         ifsq_watchdog_reset(wd);
3519 }
3520
3521 void
3522 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3523 {
3524         wd->wd_timer = 0;
3525         callout_stop(&wd->wd_callout);
3526 }
3527
3528 void
3529 ifnet_lock(void)
3530 {
3531         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3532             ("try holding ifnet lock in netisr"));
3533         mtx_lock(&ifnet_mtx);
3534 }
3535
3536 void
3537 ifnet_unlock(void)
3538 {
3539         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3540             ("try holding ifnet lock in netisr"));
3541         mtx_unlock(&ifnet_mtx);
3542 }
3543
3544 static struct ifnet_array *
3545 ifnet_array_alloc(int count)
3546 {
3547         struct ifnet_array *arr;
3548
3549         arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3550             M_IFNET, M_WAITOK);
3551         arr->ifnet_count = count;
3552
3553         return arr;
3554 }
3555
3556 static void
3557 ifnet_array_free(struct ifnet_array *arr)
3558 {
3559         if (arr == &ifnet_array0)
3560                 return;
3561         kfree(arr, M_IFNET);
3562 }
3563
3564 static struct ifnet_array *
3565 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3566 {
3567         struct ifnet_array *arr;
3568         int count, i;
3569
3570         KASSERT(old_arr->ifnet_count >= 0,
3571             ("invalid ifnet array count %d", old_arr->ifnet_count));
3572         count = old_arr->ifnet_count + 1;
3573         arr = ifnet_array_alloc(count);
3574
3575         /*
3576          * Save the old ifnet array and append this ifp to the end of
3577          * the new ifnet array.
3578          */
3579         for (i = 0; i < old_arr->ifnet_count; ++i) {
3580                 KASSERT(old_arr->ifnet_arr[i] != ifp,
3581                     ("%s is already in ifnet array", ifp->if_xname));
3582                 arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3583         }
3584         KASSERT(i == count - 1,
3585             ("add %s, ifnet array index mismatch, should be %d, but got %d",
3586              ifp->if_xname, count - 1, i));
3587         arr->ifnet_arr[i] = ifp;
3588
3589         return arr;
3590 }
3591
3592 static struct ifnet_array *
3593 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3594 {
3595         struct ifnet_array *arr;
3596         int count, i, idx, found = 0;
3597
3598         KASSERT(old_arr->ifnet_count > 0,
3599             ("invalid ifnet array count %d", old_arr->ifnet_count));
3600         count = old_arr->ifnet_count - 1;
3601         arr = ifnet_array_alloc(count);
3602
3603         /*
3604          * Save the old ifnet array, but skip this ifp.
3605          */
3606         idx = 0;
3607         for (i = 0; i < old_arr->ifnet_count; ++i) {
3608                 if (old_arr->ifnet_arr[i] == ifp) {
3609                         KASSERT(!found,
3610                             ("dup %s is in ifnet array", ifp->if_xname));
3611                         found = 1;
3612                         continue;
3613                 }
3614                 KASSERT(idx < count,
3615                     ("invalid ifnet array index %d, count %d", idx, count));
3616                 arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3617                 ++idx;
3618         }
3619         KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3620         KASSERT(idx == count,
3621             ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3622              ifp->if_xname, count, idx));
3623
3624         return arr;
3625 }
3626
3627 const struct ifnet_array *
3628 ifnet_array_get(void)
3629 {
3630         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3631         return ifnet_array;
3632 }
3633
3634 int
3635 ifnet_array_isempty(void)
3636 {
3637         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3638         if (ifnet_array->ifnet_count == 0)
3639                 return 1;
3640         else
3641                 return 0;
3642 }
3643
3644 void
3645 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3646 {
3647         struct ifaddr *ifa;
3648
3649         memset(mark, 0, sizeof(*mark));
3650         ifa = &mark->ifa;
3651
3652         mark->ifac.ifa = ifa;
3653
3654         ifa->ifa_addr = &mark->addr;
3655         ifa->ifa_dstaddr = &mark->dstaddr;
3656         ifa->ifa_netmask = &mark->netmask;
3657         ifa->ifa_ifp = ifp;
3658 }