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 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 /*
1262  * Locate an interface based on a complete address.
1263  */
1264 struct ifaddr *
1265 ifa_ifwithaddr(struct sockaddr *addr)
1266 {
1267         const struct ifnet_array *arr;
1268         int i;
1269
1270         arr = ifnet_array_get();
1271         for (i = 0; i < arr->ifnet_count; ++i) {
1272                 struct ifnet *ifp = arr->ifnet_arr[i];
1273                 struct ifaddr_container *ifac;
1274
1275                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1276                         struct ifaddr *ifa = ifac->ifa;
1277
1278                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1279                                 continue;
1280                         if (sa_equal(addr, ifa->ifa_addr))
1281                                 return (ifa);
1282                         if ((ifp->if_flags & IFF_BROADCAST) &&
1283                             ifa->ifa_broadaddr &&
1284                             /* IPv6 doesn't have broadcast */
1285                             ifa->ifa_broadaddr->sa_len != 0 &&
1286                             sa_equal(ifa->ifa_broadaddr, addr))
1287                                 return (ifa);
1288                 }
1289         }
1290         return (NULL);
1291 }
1292 /*
1293  * Locate the point to point interface with a given destination address.
1294  */
1295 struct ifaddr *
1296 ifa_ifwithdstaddr(struct sockaddr *addr)
1297 {
1298         const struct ifnet_array *arr;
1299         int i;
1300
1301         arr = ifnet_array_get();
1302         for (i = 0; i < arr->ifnet_count; ++i) {
1303                 struct ifnet *ifp = arr->ifnet_arr[i];
1304                 struct ifaddr_container *ifac;
1305
1306                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1307                         continue;
1308
1309                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1310                         struct ifaddr *ifa = ifac->ifa;
1311
1312                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1313                                 continue;
1314                         if (ifa->ifa_dstaddr &&
1315                             sa_equal(addr, ifa->ifa_dstaddr))
1316                                 return (ifa);
1317                 }
1318         }
1319         return (NULL);
1320 }
1321
1322 /*
1323  * Find an interface on a specific network.  If many, choice
1324  * is most specific found.
1325  */
1326 struct ifaddr *
1327 ifa_ifwithnet(struct sockaddr *addr)
1328 {
1329         struct ifaddr *ifa_maybe = NULL;
1330         u_int af = addr->sa_family;
1331         char *addr_data = addr->sa_data, *cplim;
1332         const struct ifnet_array *arr;
1333         int i;
1334
1335         /*
1336          * AF_LINK addresses can be looked up directly by their index number,
1337          * so do that if we can.
1338          */
1339         if (af == AF_LINK) {
1340                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1341
1342                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1343                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1344         }
1345
1346         /*
1347          * Scan though each interface, looking for ones that have
1348          * addresses in this address family.
1349          */
1350         arr = ifnet_array_get();
1351         for (i = 0; i < arr->ifnet_count; ++i) {
1352                 struct ifnet *ifp = arr->ifnet_arr[i];
1353                 struct ifaddr_container *ifac;
1354
1355                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1356                         struct ifaddr *ifa = ifac->ifa;
1357                         char *cp, *cp2, *cp3;
1358
1359                         if (ifa->ifa_addr->sa_family != af)
1360 next:                           continue;
1361                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1362                                 /*
1363                                  * This is a bit broken as it doesn't
1364                                  * take into account that the remote end may
1365                                  * be a single node in the network we are
1366                                  * looking for.
1367                                  * The trouble is that we don't know the
1368                                  * netmask for the remote end.
1369                                  */
1370                                 if (ifa->ifa_dstaddr != NULL &&
1371                                     sa_equal(addr, ifa->ifa_dstaddr))
1372                                         return (ifa);
1373                         } else {
1374                                 /*
1375                                  * if we have a special address handler,
1376                                  * then use it instead of the generic one.
1377                                  */
1378                                 if (ifa->ifa_claim_addr) {
1379                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1380                                                 return (ifa);
1381                                         } else {
1382                                                 continue;
1383                                         }
1384                                 }
1385
1386                                 /*
1387                                  * Scan all the bits in the ifa's address.
1388                                  * If a bit dissagrees with what we are
1389                                  * looking for, mask it with the netmask
1390                                  * to see if it really matters.
1391                                  * (A byte at a time)
1392                                  */
1393                                 if (ifa->ifa_netmask == 0)
1394                                         continue;
1395                                 cp = addr_data;
1396                                 cp2 = ifa->ifa_addr->sa_data;
1397                                 cp3 = ifa->ifa_netmask->sa_data;
1398                                 cplim = ifa->ifa_netmask->sa_len +
1399                                         (char *)ifa->ifa_netmask;
1400                                 while (cp3 < cplim)
1401                                         if ((*cp++ ^ *cp2++) & *cp3++)
1402                                                 goto next; /* next address! */
1403                                 /*
1404                                  * If the netmask of what we just found
1405                                  * is more specific than what we had before
1406                                  * (if we had one) then remember the new one
1407                                  * before continuing to search
1408                                  * for an even better one.
1409                                  */
1410                                 if (ifa_maybe == NULL ||
1411                                     rn_refines((char *)ifa->ifa_netmask,
1412                                                (char *)ifa_maybe->ifa_netmask))
1413                                         ifa_maybe = ifa;
1414                         }
1415                 }
1416         }
1417         return (ifa_maybe);
1418 }
1419
1420 /*
1421  * Find an interface address specific to an interface best matching
1422  * a given address.
1423  */
1424 struct ifaddr *
1425 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1426 {
1427         struct ifaddr_container *ifac;
1428         char *cp, *cp2, *cp3;
1429         char *cplim;
1430         struct ifaddr *ifa_maybe = NULL;
1431         u_int af = addr->sa_family;
1432
1433         if (af >= AF_MAX)
1434                 return (0);
1435         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1436                 struct ifaddr *ifa = ifac->ifa;
1437
1438                 if (ifa->ifa_addr->sa_family != af)
1439                         continue;
1440                 if (ifa_maybe == NULL)
1441                         ifa_maybe = ifa;
1442                 if (ifa->ifa_netmask == NULL) {
1443                         if (sa_equal(addr, ifa->ifa_addr) ||
1444                             (ifa->ifa_dstaddr != NULL &&
1445                              sa_equal(addr, ifa->ifa_dstaddr)))
1446                                 return (ifa);
1447                         continue;
1448                 }
1449                 if (ifp->if_flags & IFF_POINTOPOINT) {
1450                         if (sa_equal(addr, ifa->ifa_dstaddr))
1451                                 return (ifa);
1452                 } else {
1453                         cp = addr->sa_data;
1454                         cp2 = ifa->ifa_addr->sa_data;
1455                         cp3 = ifa->ifa_netmask->sa_data;
1456                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1457                         for (; cp3 < cplim; cp3++)
1458                                 if ((*cp++ ^ *cp2++) & *cp3)
1459                                         break;
1460                         if (cp3 == cplim)
1461                                 return (ifa);
1462                 }
1463         }
1464         return (ifa_maybe);
1465 }
1466
1467 /*
1468  * Default action when installing a route with a Link Level gateway.
1469  * Lookup an appropriate real ifa to point to.
1470  * This should be moved to /sys/net/link.c eventually.
1471  */
1472 static void
1473 link_rtrequest(int cmd, struct rtentry *rt)
1474 {
1475         struct ifaddr *ifa;
1476         struct sockaddr *dst;
1477         struct ifnet *ifp;
1478
1479         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1480             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1481                 return;
1482         ifa = ifaof_ifpforaddr(dst, ifp);
1483         if (ifa != NULL) {
1484                 IFAFREE(rt->rt_ifa);
1485                 IFAREF(ifa);
1486                 rt->rt_ifa = ifa;
1487                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1488                         ifa->ifa_rtrequest(cmd, rt);
1489         }
1490 }
1491
1492 struct netmsg_ifroute {
1493         struct netmsg_base      base;
1494         struct ifnet            *ifp;
1495         int                     flag;
1496         int                     fam;
1497 };
1498
1499 /*
1500  * Mark an interface down and notify protocols of the transition.
1501  */
1502 static void
1503 if_unroute_dispatch(netmsg_t nmsg)
1504 {
1505         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1506         struct ifnet *ifp = msg->ifp;
1507         int flag = msg->flag, fam = msg->fam;
1508         struct ifaddr_container *ifac;
1509
1510         ifp->if_flags &= ~flag;
1511         getmicrotime(&ifp->if_lastchange);
1512         /*
1513          * The ifaddr processing in the following loop will block,
1514          * however, this function is called in netisr0, in which
1515          * ifaddr list changes happen, so we don't care about the
1516          * blockness of the ifaddr processing here.
1517          */
1518         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1519                 struct ifaddr *ifa = ifac->ifa;
1520
1521                 /* Ignore marker */
1522                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1523                         continue;
1524
1525                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1526                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1527         }
1528         ifq_purge_all(&ifp->if_snd);
1529         rt_ifmsg(ifp);
1530
1531         lwkt_replymsg(&nmsg->lmsg, 0);
1532 }
1533
1534 void
1535 if_unroute(struct ifnet *ifp, int flag, int fam)
1536 {
1537         struct netmsg_ifroute msg;
1538
1539         ASSERT_CANDOMSG_NETISR0(curthread);
1540
1541         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1542             if_unroute_dispatch);
1543         msg.ifp = ifp;
1544         msg.flag = flag;
1545         msg.fam = fam;
1546         lwkt_domsg(netisr_cpuport(0), &msg.base.lmsg, 0);
1547 }
1548
1549 /*
1550  * Mark an interface up and notify protocols of the transition.
1551  */
1552 static void
1553 if_route_dispatch(netmsg_t nmsg)
1554 {
1555         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1556         struct ifnet *ifp = msg->ifp;
1557         int flag = msg->flag, fam = msg->fam;
1558         struct ifaddr_container *ifac;
1559
1560         ifq_purge_all(&ifp->if_snd);
1561         ifp->if_flags |= flag;
1562         getmicrotime(&ifp->if_lastchange);
1563         /*
1564          * The ifaddr processing in the following loop will block,
1565          * however, this function is called in netisr0, in which
1566          * ifaddr list changes happen, so we don't care about the
1567          * blockness of the ifaddr processing here.
1568          */
1569         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1570                 struct ifaddr *ifa = ifac->ifa;
1571
1572                 /* Ignore marker */
1573                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1574                         continue;
1575
1576                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1577                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1578         }
1579         rt_ifmsg(ifp);
1580 #ifdef INET6
1581         in6_if_up(ifp);
1582 #endif
1583
1584         lwkt_replymsg(&nmsg->lmsg, 0);
1585 }
1586
1587 void
1588 if_route(struct ifnet *ifp, int flag, int fam)
1589 {
1590         struct netmsg_ifroute msg;
1591
1592         ASSERT_CANDOMSG_NETISR0(curthread);
1593
1594         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1595             if_route_dispatch);
1596         msg.ifp = ifp;
1597         msg.flag = flag;
1598         msg.fam = fam;
1599         lwkt_domsg(netisr_cpuport(0), &msg.base.lmsg, 0);
1600 }
1601
1602 /*
1603  * Mark an interface down and notify protocols of the transition.  An
1604  * interface going down is also considered to be a synchronizing event.
1605  * We must ensure that all packet processing related to the interface
1606  * has completed before we return so e.g. the caller can free the ifnet
1607  * structure that the mbufs may be referencing.
1608  *
1609  * NOTE: must be called at splnet or eqivalent.
1610  */
1611 void
1612 if_down(struct ifnet *ifp)
1613 {
1614         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1615         netmsg_service_sync();
1616 }
1617
1618 /*
1619  * Mark an interface up and notify protocols of
1620  * the transition.
1621  * NOTE: must be called at splnet or eqivalent.
1622  */
1623 void
1624 if_up(struct ifnet *ifp)
1625 {
1626         if_route(ifp, IFF_UP, AF_UNSPEC);
1627 }
1628
1629 /*
1630  * Process a link state change.
1631  * NOTE: must be called at splsoftnet or equivalent.
1632  */
1633 void
1634 if_link_state_change(struct ifnet *ifp)
1635 {
1636         int link_state = ifp->if_link_state;
1637
1638         rt_ifmsg(ifp);
1639         devctl_notify("IFNET", ifp->if_xname,
1640             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1641 }
1642
1643 /*
1644  * Handle interface watchdog timer routines.  Called
1645  * from softclock, we decrement timers (if set) and
1646  * call the appropriate interface routine on expiration.
1647  */
1648 static void
1649 if_slowtimo_dispatch(netmsg_t nmsg)
1650 {
1651         struct globaldata *gd = mycpu;
1652         const struct ifnet_array *arr;
1653         int i;
1654
1655         ASSERT_IN_NETISR(0);
1656
1657         crit_enter_gd(gd);
1658         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1659         crit_exit_gd(gd);
1660
1661         arr = ifnet_array_get();
1662         for (i = 0; i < arr->ifnet_count; ++i) {
1663                 struct ifnet *ifp = arr->ifnet_arr[i];
1664
1665                 crit_enter_gd(gd);
1666
1667                 if (if_stats_compat) {
1668                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1669                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1670                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1671                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1672                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1673                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1674                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1675                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1676                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1677                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1678                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1679                 }
1680
1681                 if (ifp->if_timer == 0 || --ifp->if_timer) {
1682                         crit_exit_gd(gd);
1683                         continue;
1684                 }
1685                 if (ifp->if_watchdog) {
1686                         if (ifnet_tryserialize_all(ifp)) {
1687                                 (*ifp->if_watchdog)(ifp);
1688                                 ifnet_deserialize_all(ifp);
1689                         } else {
1690                                 /* try again next timeout */
1691                                 ++ifp->if_timer;
1692                         }
1693                 }
1694
1695                 crit_exit_gd(gd);
1696         }
1697
1698         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1699 }
1700
1701 static void
1702 if_slowtimo(void *arg __unused)
1703 {
1704         struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1705
1706         KASSERT(mycpuid == 0, ("not on cpu0"));
1707         crit_enter();
1708         if (lmsg->ms_flags & MSGF_DONE)
1709                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1710         crit_exit();
1711 }
1712
1713 /*
1714  * Map interface name to
1715  * interface structure pointer.
1716  */
1717 struct ifnet *
1718 ifunit(const char *name)
1719 {
1720         struct ifnet *ifp;
1721
1722         /*
1723          * Search all the interfaces for this name/number
1724          */
1725         KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1726
1727         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1728                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1729                         break;
1730         }
1731         return (ifp);
1732 }
1733
1734 struct ifnet *
1735 ifunit_netisr(const char *name)
1736 {
1737         const struct ifnet_array *arr;
1738         int i;
1739
1740         /*
1741          * Search all the interfaces for this name/number
1742          */
1743
1744         arr = ifnet_array_get();
1745         for (i = 0; i < arr->ifnet_count; ++i) {
1746                 struct ifnet *ifp = arr->ifnet_arr[i];
1747
1748                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1749                         return ifp;
1750         }
1751         return NULL;
1752 }
1753
1754 /*
1755  * Interface ioctls.
1756  */
1757 int
1758 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1759 {
1760         struct ifnet *ifp;
1761         struct ifreq *ifr;
1762         struct ifstat *ifs;
1763         int error;
1764         short oif_flags;
1765         int new_flags;
1766 #ifdef COMPAT_43
1767         int ocmd;
1768 #endif
1769         size_t namelen, onamelen;
1770         char new_name[IFNAMSIZ];
1771         struct ifaddr *ifa;
1772         struct sockaddr_dl *sdl;
1773
1774         switch (cmd) {
1775         case SIOCGIFCONF:
1776         case OSIOCGIFCONF:
1777                 return (ifconf(cmd, data, cred));
1778         default:
1779                 break;
1780         }
1781
1782         ifr = (struct ifreq *)data;
1783
1784         switch (cmd) {
1785         case SIOCIFCREATE:
1786         case SIOCIFCREATE2:
1787                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1788                         return (error);
1789                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1790                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1791         case SIOCIFDESTROY:
1792                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1793                         return (error);
1794                 return (if_clone_destroy(ifr->ifr_name));
1795         case SIOCIFGCLONERS:
1796                 return (if_clone_list((struct if_clonereq *)data));
1797         default:
1798                 break;
1799         }
1800
1801         /*
1802          * Nominal ioctl through interface, lookup the ifp and obtain a
1803          * lock to serialize the ifconfig ioctl operation.
1804          */
1805         ifnet_lock();
1806
1807         ifp = ifunit(ifr->ifr_name);
1808         if (ifp == NULL) {
1809                 ifnet_unlock();
1810                 return (ENXIO);
1811         }
1812         error = 0;
1813
1814         switch (cmd) {
1815         case SIOCGIFINDEX:
1816                 ifr->ifr_index = ifp->if_index;
1817                 break;
1818
1819         case SIOCGIFFLAGS:
1820                 ifr->ifr_flags = ifp->if_flags;
1821                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1822                 break;
1823
1824         case SIOCGIFCAP:
1825                 ifr->ifr_reqcap = ifp->if_capabilities;
1826                 ifr->ifr_curcap = ifp->if_capenable;
1827                 break;
1828
1829         case SIOCGIFMETRIC:
1830                 ifr->ifr_metric = ifp->if_metric;
1831                 break;
1832
1833         case SIOCGIFMTU:
1834                 ifr->ifr_mtu = ifp->if_mtu;
1835                 break;
1836
1837         case SIOCGIFTSOLEN:
1838                 ifr->ifr_tsolen = ifp->if_tsolen;
1839                 break;
1840
1841         case SIOCGIFDATA:
1842                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1843                                 sizeof(ifp->if_data));
1844                 break;
1845
1846         case SIOCGIFPHYS:
1847                 ifr->ifr_phys = ifp->if_physical;
1848                 break;
1849
1850         case SIOCGIFPOLLCPU:
1851                 ifr->ifr_pollcpu = -1;
1852                 break;
1853
1854         case SIOCSIFPOLLCPU:
1855                 break;
1856
1857         case SIOCSIFFLAGS:
1858                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1859                 if (error)
1860                         break;
1861                 new_flags = (ifr->ifr_flags & 0xffff) |
1862                     (ifr->ifr_flagshigh << 16);
1863                 if (ifp->if_flags & IFF_SMART) {
1864                         /* Smart drivers twiddle their own routes */
1865                 } else if (ifp->if_flags & IFF_UP &&
1866                     (new_flags & IFF_UP) == 0) {
1867                         crit_enter();
1868                         if_down(ifp);
1869                         crit_exit();
1870                 } else if (new_flags & IFF_UP &&
1871                     (ifp->if_flags & IFF_UP) == 0) {
1872                         crit_enter();
1873                         if_up(ifp);
1874                         crit_exit();
1875                 }
1876
1877 #ifdef IFPOLL_ENABLE
1878                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1879                         if (new_flags & IFF_NPOLLING)
1880                                 ifpoll_register(ifp);
1881                         else
1882                                 ifpoll_deregister(ifp);
1883                 }
1884 #endif
1885
1886                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1887                         (new_flags &~ IFF_CANTCHANGE);
1888                 if (new_flags & IFF_PPROMISC) {
1889                         /* Permanently promiscuous mode requested */
1890                         ifp->if_flags |= IFF_PROMISC;
1891                 } else if (ifp->if_pcount == 0) {
1892                         ifp->if_flags &= ~IFF_PROMISC;
1893                 }
1894                 if (ifp->if_ioctl) {
1895                         ifnet_serialize_all(ifp);
1896                         ifp->if_ioctl(ifp, cmd, data, cred);
1897                         ifnet_deserialize_all(ifp);
1898                 }
1899                 getmicrotime(&ifp->if_lastchange);
1900                 break;
1901
1902         case SIOCSIFCAP:
1903                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1904                 if (error)
1905                         break;
1906                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1907                         error = EINVAL;
1908                         break;
1909                 }
1910                 ifnet_serialize_all(ifp);
1911                 ifp->if_ioctl(ifp, cmd, data, cred);
1912                 ifnet_deserialize_all(ifp);
1913                 break;
1914
1915         case SIOCSIFNAME:
1916                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1917                 if (error)
1918                         break;
1919                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1920                 if (error)
1921                         break;
1922                 if (new_name[0] == '\0') {
1923                         error = EINVAL;
1924                         break;
1925                 }
1926                 if (ifunit(new_name) != NULL) {
1927                         error = EEXIST;
1928                         break;
1929                 }
1930
1931                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1932
1933                 /* Announce the departure of the interface. */
1934                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1935
1936                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1937                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1938                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1939                 namelen = strlen(new_name);
1940                 onamelen = sdl->sdl_nlen;
1941                 /*
1942                  * Move the address if needed.  This is safe because we
1943                  * allocate space for a name of length IFNAMSIZ when we
1944                  * create this in if_attach().
1945                  */
1946                 if (namelen != onamelen) {
1947                         bcopy(sdl->sdl_data + onamelen,
1948                             sdl->sdl_data + namelen, sdl->sdl_alen);
1949                 }
1950                 bcopy(new_name, sdl->sdl_data, namelen);
1951                 sdl->sdl_nlen = namelen;
1952                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1953                 bzero(sdl->sdl_data, onamelen);
1954                 while (namelen != 0)
1955                         sdl->sdl_data[--namelen] = 0xff;
1956
1957                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1958
1959                 /* Announce the return of the interface. */
1960                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1961                 break;
1962
1963         case SIOCSIFMETRIC:
1964                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1965                 if (error)
1966                         break;
1967                 ifp->if_metric = ifr->ifr_metric;
1968                 getmicrotime(&ifp->if_lastchange);
1969                 break;
1970
1971         case SIOCSIFPHYS:
1972                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1973                 if (error)
1974                         break;
1975                 if (ifp->if_ioctl == NULL) {
1976                         error = EOPNOTSUPP;
1977                         break;
1978                 }
1979                 ifnet_serialize_all(ifp);
1980                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1981                 ifnet_deserialize_all(ifp);
1982                 if (error == 0)
1983                         getmicrotime(&ifp->if_lastchange);
1984                 break;
1985
1986         case SIOCSIFMTU:
1987         {
1988                 u_long oldmtu = ifp->if_mtu;
1989
1990                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1991                 if (error)
1992                         break;
1993                 if (ifp->if_ioctl == NULL) {
1994                         error = EOPNOTSUPP;
1995                         break;
1996                 }
1997                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
1998                         error = EINVAL;
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                         rt_ifmsg(ifp);
2007                 }
2008                 /*
2009                  * If the link MTU changed, do network layer specific procedure.
2010                  */
2011                 if (ifp->if_mtu != oldmtu) {
2012 #ifdef INET6
2013                         nd6_setmtu(ifp);
2014 #endif
2015                 }
2016                 break;
2017         }
2018
2019         case SIOCSIFTSOLEN:
2020                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2021                 if (error)
2022                         break;
2023
2024                 /* XXX need driver supplied upper limit */
2025                 if (ifr->ifr_tsolen <= 0) {
2026                         error = EINVAL;
2027                         break;
2028                 }
2029                 ifp->if_tsolen = ifr->ifr_tsolen;
2030                 break;
2031
2032         case SIOCADDMULTI:
2033         case SIOCDELMULTI:
2034                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2035                 if (error)
2036                         break;
2037
2038                 /* Don't allow group membership on non-multicast interfaces. */
2039                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2040                         error = EOPNOTSUPP;
2041                         break;
2042                 }
2043
2044                 /* Don't let users screw up protocols' entries. */
2045                 if (ifr->ifr_addr.sa_family != AF_LINK) {
2046                         error = EINVAL;
2047                         break;
2048                 }
2049
2050                 if (cmd == SIOCADDMULTI) {
2051                         struct ifmultiaddr *ifma;
2052                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2053                 } else {
2054                         error = if_delmulti(ifp, &ifr->ifr_addr);
2055                 }
2056                 if (error == 0)
2057                         getmicrotime(&ifp->if_lastchange);
2058                 break;
2059
2060         case SIOCSIFPHYADDR:
2061         case SIOCDIFPHYADDR:
2062 #ifdef INET6
2063         case SIOCSIFPHYADDR_IN6:
2064 #endif
2065         case SIOCSLIFPHYADDR:
2066         case SIOCSIFMEDIA:
2067         case SIOCSIFGENERIC:
2068                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2069                 if (error)
2070                         break;
2071                 if (ifp->if_ioctl == 0) {
2072                         error = EOPNOTSUPP;
2073                         break;
2074                 }
2075                 ifnet_serialize_all(ifp);
2076                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2077                 ifnet_deserialize_all(ifp);
2078                 if (error == 0)
2079                         getmicrotime(&ifp->if_lastchange);
2080                 break;
2081
2082         case SIOCGIFSTATUS:
2083                 ifs = (struct ifstat *)data;
2084                 ifs->ascii[0] = '\0';
2085                 /* fall through */
2086         case SIOCGIFPSRCADDR:
2087         case SIOCGIFPDSTADDR:
2088         case SIOCGLIFPHYADDR:
2089         case SIOCGIFMEDIA:
2090         case SIOCGIFGENERIC:
2091                 if (ifp->if_ioctl == NULL) {
2092                         error = EOPNOTSUPP;
2093                         break;
2094                 }
2095                 ifnet_serialize_all(ifp);
2096                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2097                 ifnet_deserialize_all(ifp);
2098                 break;
2099
2100         case SIOCSIFLLADDR:
2101                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2102                 if (error)
2103                         break;
2104                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2105                                      ifr->ifr_addr.sa_len);
2106                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2107                 break;
2108
2109         default:
2110                 oif_flags = ifp->if_flags;
2111                 if (so->so_proto == 0) {
2112                         error = EOPNOTSUPP;
2113                         break;
2114                 }
2115 #ifndef COMPAT_43
2116                 error = so_pru_control_direct(so, cmd, data, ifp);
2117 #else
2118                 ocmd = cmd;
2119
2120                 switch (cmd) {
2121                 case SIOCSIFDSTADDR:
2122                 case SIOCSIFADDR:
2123                 case SIOCSIFBRDADDR:
2124                 case SIOCSIFNETMASK:
2125 #if BYTE_ORDER != BIG_ENDIAN
2126                         if (ifr->ifr_addr.sa_family == 0 &&
2127                             ifr->ifr_addr.sa_len < 16) {
2128                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2129                                 ifr->ifr_addr.sa_len = 16;
2130                         }
2131 #else
2132                         if (ifr->ifr_addr.sa_len == 0)
2133                                 ifr->ifr_addr.sa_len = 16;
2134 #endif
2135                         break;
2136                 case OSIOCGIFADDR:
2137                         cmd = SIOCGIFADDR;
2138                         break;
2139                 case OSIOCGIFDSTADDR:
2140                         cmd = SIOCGIFDSTADDR;
2141                         break;
2142                 case OSIOCGIFBRDADDR:
2143                         cmd = SIOCGIFBRDADDR;
2144                         break;
2145                 case OSIOCGIFNETMASK:
2146                         cmd = SIOCGIFNETMASK;
2147                         break;
2148                 default:
2149                         break;
2150                 }
2151
2152                 error = so_pru_control_direct(so, cmd, data, ifp);
2153
2154                 switch (ocmd) {
2155                 case OSIOCGIFADDR:
2156                 case OSIOCGIFDSTADDR:
2157                 case OSIOCGIFBRDADDR:
2158                 case OSIOCGIFNETMASK:
2159                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2160                         break;
2161                 }
2162 #endif /* COMPAT_43 */
2163
2164                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2165 #ifdef INET6
2166                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
2167                         if (ifp->if_flags & IFF_UP) {
2168                                 crit_enter();
2169                                 in6_if_up(ifp);
2170                                 crit_exit();
2171                         }
2172 #endif
2173                 }
2174                 break;
2175         }
2176
2177         ifnet_unlock();
2178         return (error);
2179 }
2180
2181 /*
2182  * Set/clear promiscuous mode on interface ifp based on the truth value
2183  * of pswitch.  The calls are reference counted so that only the first
2184  * "on" request actually has an effect, as does the final "off" request.
2185  * Results are undefined if the "off" and "on" requests are not matched.
2186  */
2187 int
2188 ifpromisc(struct ifnet *ifp, int pswitch)
2189 {
2190         struct ifreq ifr;
2191         int error;
2192         int oldflags;
2193
2194         oldflags = ifp->if_flags;
2195         if (ifp->if_flags & IFF_PPROMISC) {
2196                 /* Do nothing if device is in permanently promiscuous mode */
2197                 ifp->if_pcount += pswitch ? 1 : -1;
2198                 return (0);
2199         }
2200         if (pswitch) {
2201                 /*
2202                  * If the device is not configured up, we cannot put it in
2203                  * promiscuous mode.
2204                  */
2205                 if ((ifp->if_flags & IFF_UP) == 0)
2206                         return (ENETDOWN);
2207                 if (ifp->if_pcount++ != 0)
2208                         return (0);
2209                 ifp->if_flags |= IFF_PROMISC;
2210                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
2211                     ifp->if_xname);
2212         } else {
2213                 if (--ifp->if_pcount > 0)
2214                         return (0);
2215                 ifp->if_flags &= ~IFF_PROMISC;
2216                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2217                     ifp->if_xname);
2218         }
2219         ifr.ifr_flags = ifp->if_flags;
2220         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2221         ifnet_serialize_all(ifp);
2222         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2223         ifnet_deserialize_all(ifp);
2224         if (error == 0)
2225                 rt_ifmsg(ifp);
2226         else
2227                 ifp->if_flags = oldflags;
2228         return error;
2229 }
2230
2231 /*
2232  * Return interface configuration
2233  * of system.  List may be used
2234  * in later ioctl's (above) to get
2235  * other information.
2236  */
2237 static int
2238 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2239 {
2240         struct ifconf *ifc = (struct ifconf *)data;
2241         struct ifnet *ifp;
2242         struct sockaddr *sa;
2243         struct ifreq ifr, *ifrp;
2244         int space = ifc->ifc_len, error = 0;
2245
2246         ifrp = ifc->ifc_req;
2247
2248         ifnet_lock();
2249         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2250                 struct ifaddr_container *ifac, *ifac_mark;
2251                 struct ifaddr_marker mark;
2252                 struct ifaddrhead *head;
2253                 int addrs;
2254
2255                 if (space <= sizeof ifr)
2256                         break;
2257
2258                 /*
2259                  * Zero the stack declared structure first to prevent
2260                  * memory disclosure.
2261                  */
2262                 bzero(&ifr, sizeof(ifr));
2263                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2264                     >= sizeof(ifr.ifr_name)) {
2265                         error = ENAMETOOLONG;
2266                         break;
2267                 }
2268
2269                 /*
2270                  * Add a marker, since copyout() could block and during that
2271                  * period the list could be changed.  Inserting the marker to
2272                  * the header of the list will not cause trouble for the code
2273                  * assuming that the first element of the list is AF_LINK; the
2274                  * marker will be moved to the next position w/o blocking.
2275                  */
2276                 ifa_marker_init(&mark, ifp);
2277                 ifac_mark = &mark.ifac;
2278                 head = &ifp->if_addrheads[mycpuid];
2279
2280                 addrs = 0;
2281                 TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2282                 while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2283                         struct ifaddr *ifa = ifac->ifa;
2284
2285                         TAILQ_REMOVE(head, ifac_mark, ifa_link);
2286                         TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2287
2288                         /* Ignore marker */
2289                         if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2290                                 continue;
2291
2292                         if (space <= sizeof ifr)
2293                                 break;
2294                         sa = ifa->ifa_addr;
2295                         if (cred->cr_prison &&
2296                             prison_if(cred, sa))
2297                                 continue;
2298                         addrs++;
2299                         /*
2300                          * Keep a reference on this ifaddr, so that it will
2301                          * not be destroyed when its address is copied to
2302                          * the userland, which could block.
2303                          */
2304                         IFAREF(ifa);
2305 #ifdef COMPAT_43
2306                         if (cmd == OSIOCGIFCONF) {
2307                                 struct osockaddr *osa =
2308                                          (struct osockaddr *)&ifr.ifr_addr;
2309                                 ifr.ifr_addr = *sa;
2310                                 osa->sa_family = sa->sa_family;
2311                                 error = copyout(&ifr, ifrp, sizeof ifr);
2312                                 ifrp++;
2313                         } else
2314 #endif
2315                         if (sa->sa_len <= sizeof(*sa)) {
2316                                 ifr.ifr_addr = *sa;
2317                                 error = copyout(&ifr, ifrp, sizeof ifr);
2318                                 ifrp++;
2319                         } else {
2320                                 if (space < (sizeof ifr) + sa->sa_len -
2321                                             sizeof(*sa)) {
2322                                         IFAFREE(ifa);
2323                                         break;
2324                                 }
2325                                 space -= sa->sa_len - sizeof(*sa);
2326                                 error = copyout(&ifr, ifrp,
2327                                                 sizeof ifr.ifr_name);
2328                                 if (error == 0)
2329                                         error = copyout(sa, &ifrp->ifr_addr,
2330                                                         sa->sa_len);
2331                                 ifrp = (struct ifreq *)
2332                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2333                         }
2334                         IFAFREE(ifa);
2335                         if (error)
2336                                 break;
2337                         space -= sizeof ifr;
2338                 }
2339                 TAILQ_REMOVE(head, ifac_mark, ifa_link);
2340                 if (error)
2341                         break;
2342                 if (!addrs) {
2343                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2344                         error = copyout(&ifr, ifrp, sizeof ifr);
2345                         if (error)
2346                                 break;
2347                         space -= sizeof ifr;
2348                         ifrp++;
2349                 }
2350         }
2351         ifnet_unlock();
2352
2353         ifc->ifc_len -= space;
2354         return (error);
2355 }
2356
2357 /*
2358  * Just like if_promisc(), but for all-multicast-reception mode.
2359  */
2360 int
2361 if_allmulti(struct ifnet *ifp, int onswitch)
2362 {
2363         int error = 0;
2364         struct ifreq ifr;
2365
2366         crit_enter();
2367
2368         if (onswitch) {
2369                 if (ifp->if_amcount++ == 0) {
2370                         ifp->if_flags |= IFF_ALLMULTI;
2371                         ifr.ifr_flags = ifp->if_flags;
2372                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2373                         ifnet_serialize_all(ifp);
2374                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2375                                               NULL);
2376                         ifnet_deserialize_all(ifp);
2377                 }
2378         } else {
2379                 if (ifp->if_amcount > 1) {
2380                         ifp->if_amcount--;
2381                 } else {
2382                         ifp->if_amcount = 0;
2383                         ifp->if_flags &= ~IFF_ALLMULTI;
2384                         ifr.ifr_flags = ifp->if_flags;
2385                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2386                         ifnet_serialize_all(ifp);
2387                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2388                                               NULL);
2389                         ifnet_deserialize_all(ifp);
2390                 }
2391         }
2392
2393         crit_exit();
2394
2395         if (error == 0)
2396                 rt_ifmsg(ifp);
2397         return error;
2398 }
2399
2400 /*
2401  * Add a multicast listenership to the interface in question.
2402  * The link layer provides a routine which converts
2403  */
2404 int
2405 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2406     struct ifmultiaddr **retifma)
2407 {
2408         struct sockaddr *llsa, *dupsa;
2409         int error;
2410         struct ifmultiaddr *ifma;
2411
2412         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2413
2414         /*
2415          * If the matching multicast address already exists
2416          * then don't add a new one, just add a reference
2417          */
2418         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2419                 if (sa_equal(sa, ifma->ifma_addr)) {
2420                         ifma->ifma_refcount++;
2421                         if (retifma)
2422                                 *retifma = ifma;
2423                         return 0;
2424                 }
2425         }
2426
2427         /*
2428          * Give the link layer a chance to accept/reject it, and also
2429          * find out which AF_LINK address this maps to, if it isn't one
2430          * already.
2431          */
2432         if (ifp->if_resolvemulti) {
2433                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2434                 if (error)
2435                         return error;
2436         } else {
2437                 llsa = NULL;
2438         }
2439
2440         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2441         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2442         bcopy(sa, dupsa, sa->sa_len);
2443
2444         ifma->ifma_addr = dupsa;
2445         ifma->ifma_lladdr = llsa;
2446         ifma->ifma_ifp = ifp;
2447         ifma->ifma_refcount = 1;
2448         ifma->ifma_protospec = NULL;
2449         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2450
2451         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2452         if (retifma)
2453                 *retifma = ifma;
2454
2455         if (llsa != NULL) {
2456                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2457                         if (sa_equal(ifma->ifma_addr, llsa))
2458                                 break;
2459                 }
2460                 if (ifma) {
2461                         ifma->ifma_refcount++;
2462                 } else {
2463                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2464                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2465                         bcopy(llsa, dupsa, llsa->sa_len);
2466                         ifma->ifma_addr = dupsa;
2467                         ifma->ifma_ifp = ifp;
2468                         ifma->ifma_refcount = 1;
2469                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2470                 }
2471         }
2472         /*
2473          * We are certain we have added something, so call down to the
2474          * interface to let them know about it.
2475          */
2476         if (ifp->if_ioctl)
2477                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2478
2479         return 0;
2480 }
2481
2482 int
2483 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2484     struct ifmultiaddr **retifma)
2485 {
2486         int error;
2487
2488         ifnet_serialize_all(ifp);
2489         error = if_addmulti_serialized(ifp, sa, retifma);
2490         ifnet_deserialize_all(ifp);
2491
2492         return error;
2493 }
2494
2495 /*
2496  * Remove a reference to a multicast address on this interface.  Yell
2497  * if the request does not match an existing membership.
2498  */
2499 static int
2500 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2501 {
2502         struct ifmultiaddr *ifma;
2503
2504         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2505
2506         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2507                 if (sa_equal(sa, ifma->ifma_addr))
2508                         break;
2509         if (ifma == NULL)
2510                 return ENOENT;
2511
2512         if (ifma->ifma_refcount > 1) {
2513                 ifma->ifma_refcount--;
2514                 return 0;
2515         }
2516
2517         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2518         sa = ifma->ifma_lladdr;
2519         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2520         /*
2521          * Make sure the interface driver is notified
2522          * in the case of a link layer mcast group being left.
2523          */
2524         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2525                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2526         kfree(ifma->ifma_addr, M_IFMADDR);
2527         kfree(ifma, M_IFMADDR);
2528         if (sa == NULL)
2529                 return 0;
2530
2531         /*
2532          * Now look for the link-layer address which corresponds to
2533          * this network address.  It had been squirreled away in
2534          * ifma->ifma_lladdr for this purpose (so we don't have
2535          * to call ifp->if_resolvemulti() again), and we saved that
2536          * value in sa above.  If some nasty deleted the
2537          * link-layer address out from underneath us, we can deal because
2538          * the address we stored was is not the same as the one which was
2539          * in the record for the link-layer address.  (So we don't complain
2540          * in that case.)
2541          */
2542         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2543                 if (sa_equal(sa, ifma->ifma_addr))
2544                         break;
2545         if (ifma == NULL)
2546                 return 0;
2547
2548         if (ifma->ifma_refcount > 1) {
2549                 ifma->ifma_refcount--;
2550                 return 0;
2551         }
2552
2553         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2554         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2555         kfree(ifma->ifma_addr, M_IFMADDR);
2556         kfree(sa, M_IFMADDR);
2557         kfree(ifma, M_IFMADDR);
2558
2559         return 0;
2560 }
2561
2562 int
2563 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2564 {
2565         int error;
2566
2567         ifnet_serialize_all(ifp);
2568         error = if_delmulti_serialized(ifp, sa);
2569         ifnet_deserialize_all(ifp);
2570
2571         return error;
2572 }
2573
2574 /*
2575  * Delete all multicast group membership for an interface.
2576  * Should be used to quickly flush all multicast filters.
2577  */
2578 void
2579 if_delallmulti_serialized(struct ifnet *ifp)
2580 {
2581         struct ifmultiaddr *ifma, mark;
2582         struct sockaddr sa;
2583
2584         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2585
2586         bzero(&sa, sizeof(sa));
2587         sa.sa_family = AF_UNSPEC;
2588         sa.sa_len = sizeof(sa);
2589
2590         bzero(&mark, sizeof(mark));
2591         mark.ifma_addr = &sa;
2592
2593         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2594         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2595                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2596                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2597                     ifma_link);
2598
2599                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2600                         continue;
2601
2602                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2603         }
2604         TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2605 }
2606
2607
2608 /*
2609  * Set the link layer address on an interface.
2610  *
2611  * At this time we only support certain types of interfaces,
2612  * and we don't allow the length of the address to change.
2613  */
2614 int
2615 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2616 {
2617         struct sockaddr_dl *sdl;
2618         struct ifreq ifr;
2619
2620         sdl = IF_LLSOCKADDR(ifp);
2621         if (sdl == NULL)
2622                 return (EINVAL);
2623         if (len != sdl->sdl_alen)       /* don't allow length to change */
2624                 return (EINVAL);
2625         switch (ifp->if_type) {
2626         case IFT_ETHER:                 /* these types use struct arpcom */
2627         case IFT_XETHER:
2628         case IFT_L2VLAN:
2629         case IFT_IEEE8023ADLAG:
2630                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2631                 bcopy(lladdr, LLADDR(sdl), len);
2632                 break;
2633         default:
2634                 return (ENODEV);
2635         }
2636         /*
2637          * If the interface is already up, we need
2638          * to re-init it in order to reprogram its
2639          * address filter.
2640          */
2641         ifnet_serialize_all(ifp);
2642         if ((ifp->if_flags & IFF_UP) != 0) {
2643 #ifdef INET
2644                 struct ifaddr_container *ifac;
2645 #endif
2646
2647                 ifp->if_flags &= ~IFF_UP;
2648                 ifr.ifr_flags = ifp->if_flags;
2649                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2650                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2651                               NULL);
2652                 ifp->if_flags |= IFF_UP;
2653                 ifr.ifr_flags = ifp->if_flags;
2654                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2655                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2656                                  NULL);
2657 #ifdef INET
2658                 /*
2659                  * Also send gratuitous ARPs to notify other nodes about
2660                  * the address change.
2661                  */
2662                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2663                         struct ifaddr *ifa = ifac->ifa;
2664
2665                         if (ifa->ifa_addr != NULL &&
2666                             ifa->ifa_addr->sa_family == AF_INET)
2667                                 arp_gratuitous(ifp, ifa);
2668                 }
2669 #endif
2670         }
2671         ifnet_deserialize_all(ifp);
2672         return (0);
2673 }
2674
2675 struct ifmultiaddr *
2676 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2677 {
2678         struct ifmultiaddr *ifma;
2679
2680         /* TODO: need ifnet_serialize_main */
2681         ifnet_serialize_all(ifp);
2682         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2683                 if (sa_equal(ifma->ifma_addr, sa))
2684                         break;
2685         ifnet_deserialize_all(ifp);
2686
2687         return ifma;
2688 }
2689
2690 /*
2691  * This function locates the first real ethernet MAC from a network
2692  * card and loads it into node, returning 0 on success or ENOENT if
2693  * no suitable interfaces were found.  It is used by the uuid code to
2694  * generate a unique 6-byte number.
2695  */
2696 int
2697 if_getanyethermac(uint16_t *node, int minlen)
2698 {
2699         struct ifnet *ifp;
2700         struct sockaddr_dl *sdl;
2701
2702         ifnet_lock();
2703         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2704                 if (ifp->if_type != IFT_ETHER)
2705                         continue;
2706                 sdl = IF_LLSOCKADDR(ifp);
2707                 if (sdl->sdl_alen < minlen)
2708                         continue;
2709                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2710                       minlen);
2711                 ifnet_unlock();
2712                 return(0);
2713         }
2714         ifnet_unlock();
2715         return (ENOENT);
2716 }
2717
2718 /*
2719  * The name argument must be a pointer to storage which will last as
2720  * long as the interface does.  For physical devices, the result of
2721  * device_get_name(dev) is a good choice and for pseudo-devices a
2722  * static string works well.
2723  */
2724 void
2725 if_initname(struct ifnet *ifp, const char *name, int unit)
2726 {
2727         ifp->if_dname = name;
2728         ifp->if_dunit = unit;
2729         if (unit != IF_DUNIT_NONE)
2730                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2731         else
2732                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2733 }
2734
2735 int
2736 if_printf(struct ifnet *ifp, const char *fmt, ...)
2737 {
2738         __va_list ap;
2739         int retval;
2740
2741         retval = kprintf("%s: ", ifp->if_xname);
2742         __va_start(ap, fmt);
2743         retval += kvprintf(fmt, ap);
2744         __va_end(ap);
2745         return (retval);
2746 }
2747
2748 struct ifnet *
2749 if_alloc(uint8_t type)
2750 {
2751         struct ifnet *ifp;
2752         size_t size;
2753
2754         /*
2755          * XXX temporary hack until arpcom is setup in if_l2com
2756          */
2757         if (type == IFT_ETHER)
2758                 size = sizeof(struct arpcom);
2759         else
2760                 size = sizeof(struct ifnet);
2761
2762         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2763
2764         ifp->if_type = type;
2765
2766         if (if_com_alloc[type] != NULL) {
2767                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2768                 if (ifp->if_l2com == NULL) {
2769                         kfree(ifp, M_IFNET);
2770                         return (NULL);
2771                 }
2772         }
2773         return (ifp);
2774 }
2775
2776 void
2777 if_free(struct ifnet *ifp)
2778 {
2779         kfree(ifp, M_IFNET);
2780 }
2781
2782 void
2783 ifq_set_classic(struct ifaltq *ifq)
2784 {
2785         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2786             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2787 }
2788
2789 void
2790 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2791     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2792 {
2793         int q;
2794
2795         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2796         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2797         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2798         KASSERT(request != NULL, ("request is not specified"));
2799
2800         ifq->altq_mapsubq = mapsubq;
2801         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2802                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2803
2804                 ifsq->ifsq_enqueue = enqueue;
2805                 ifsq->ifsq_dequeue = dequeue;
2806                 ifsq->ifsq_request = request;
2807         }
2808 }
2809
2810 static void
2811 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2812 {
2813         m->m_nextpkt = NULL;
2814         if (ifsq->ifsq_norm_tail == NULL)
2815                 ifsq->ifsq_norm_head = m;
2816         else
2817                 ifsq->ifsq_norm_tail->m_nextpkt = m;
2818         ifsq->ifsq_norm_tail = m;
2819         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2820 }
2821
2822 static void
2823 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2824 {
2825         m->m_nextpkt = NULL;
2826         if (ifsq->ifsq_prio_tail == NULL)
2827                 ifsq->ifsq_prio_head = m;
2828         else
2829                 ifsq->ifsq_prio_tail->m_nextpkt = m;
2830         ifsq->ifsq_prio_tail = m;
2831         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2832         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2833 }
2834
2835 static struct mbuf *
2836 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2837 {
2838         struct mbuf *m;
2839
2840         m = ifsq->ifsq_norm_head;
2841         if (m != NULL) {
2842                 if ((ifsq->ifsq_norm_head = m->m_nextpkt) == NULL)
2843                         ifsq->ifsq_norm_tail = NULL;
2844                 m->m_nextpkt = NULL;
2845                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2846         }
2847         return m;
2848 }
2849
2850 static struct mbuf *
2851 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2852 {
2853         struct mbuf *m;
2854
2855         m = ifsq->ifsq_prio_head;
2856         if (m != NULL) {
2857                 if ((ifsq->ifsq_prio_head = m->m_nextpkt) == NULL)
2858                         ifsq->ifsq_prio_tail = NULL;
2859                 m->m_nextpkt = NULL;
2860                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2861                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2862         }
2863         return m;
2864 }
2865
2866 int
2867 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2868     struct altq_pktattr *pa __unused)
2869 {
2870         M_ASSERTPKTHDR(m);
2871         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
2872             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
2873                 if ((m->m_flags & M_PRIO) &&
2874                     ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen / 2) &&
2875                     ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt / 2)) {
2876                         struct mbuf *m_drop;
2877
2878                         /*
2879                          * Perform drop-head on normal queue
2880                          */
2881                         m_drop = ifsq_norm_dequeue(ifsq);
2882                         if (m_drop != NULL) {
2883                                 m_freem(m_drop);
2884                                 ifsq_prio_enqueue(ifsq, m);
2885                                 return 0;
2886                         }
2887                         /* XXX nothing could be dropped? */
2888                 }
2889                 m_freem(m);
2890                 return ENOBUFS;
2891         } else {
2892                 if (m->m_flags & M_PRIO)
2893                         ifsq_prio_enqueue(ifsq, m);
2894                 else
2895                         ifsq_norm_enqueue(ifsq, m);
2896                 return 0;
2897         }
2898 }
2899
2900 struct mbuf *
2901 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
2902 {
2903         struct mbuf *m;
2904
2905         switch (op) {
2906         case ALTDQ_POLL:
2907                 m = ifsq->ifsq_prio_head;
2908                 if (m == NULL)
2909                         m = ifsq->ifsq_norm_head;
2910                 break;
2911
2912         case ALTDQ_REMOVE:
2913                 m = ifsq_prio_dequeue(ifsq);
2914                 if (m == NULL)
2915                         m = ifsq_norm_dequeue(ifsq);
2916                 break;
2917
2918         default:
2919                 panic("unsupported ALTQ dequeue op: %d", op);
2920         }
2921         return m;
2922 }
2923
2924 int
2925 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2926 {
2927         switch (req) {
2928         case ALTRQ_PURGE:
2929                 for (;;) {
2930                         struct mbuf *m;
2931
2932                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
2933                         if (m == NULL)
2934                                 break;
2935                         m_freem(m);
2936                 }
2937                 break;
2938
2939         default:
2940                 panic("unsupported ALTQ request: %d", req);
2941         }
2942         return 0;
2943 }
2944
2945 static void
2946 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2947 {
2948         struct ifnet *ifp = ifsq_get_ifp(ifsq);
2949         int running = 0, need_sched;
2950
2951         /*
2952          * Try to do direct ifnet.if_start on the subqueue first, if there is
2953          * contention on the subqueue hardware serializer, ifnet.if_start on
2954          * the subqueue will be scheduled on the subqueue owner CPU.
2955          */
2956         if (!ifsq_tryserialize_hw(ifsq)) {
2957                 /*
2958                  * Subqueue hardware serializer contention happened,
2959                  * ifnet.if_start on the subqueue is scheduled on
2960                  * the subqueue owner CPU, and we keep going.
2961                  */
2962                 ifsq_ifstart_schedule(ifsq, 1);
2963                 return;
2964         }
2965
2966         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2967                 ifp->if_start(ifp, ifsq);
2968                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2969                         running = 1;
2970         }
2971         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2972
2973         ifsq_deserialize_hw(ifsq);
2974
2975         if (need_sched) {
2976                 /*
2977                  * More data need to be transmitted, ifnet.if_start on the
2978                  * subqueue is scheduled on the subqueue owner CPU, and we
2979                  * keep going.
2980                  * NOTE: ifnet.if_start subqueue interlock is not released.
2981                  */
2982                 ifsq_ifstart_schedule(ifsq, force_sched);
2983         }
2984 }
2985
2986 /*
2987  * Subqeue packets staging mechanism:
2988  *
2989  * The packets enqueued into the subqueue are staged to a certain amount
2990  * before the ifnet.if_start on the subqueue is called.  In this way, the
2991  * driver could avoid writing to hardware registers upon every packet,
2992  * instead, hardware registers could be written when certain amount of
2993  * packets are put onto hardware TX ring.  The measurement on several modern
2994  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
2995  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
2996  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
2997  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
2998  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
2999  *
3000  * Subqueue packets staging is performed for two entry points into drivers'
3001  * transmission function:
3002  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
3003  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
3004  *
3005  * Subqueue packets staging will be stopped upon any of the following
3006  * conditions:
3007  * - If the count of packets enqueued on the current CPU is great than or
3008  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
3009  * - If the total length of packets enqueued on the current CPU is great
3010  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
3011  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
3012  *   is usually less than hardware's MTU.
3013  * - ifsq_ifstart_schedule() is not pending on the current CPU and
3014  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
3015  *   released.
3016  * - The if_start_rollup(), which is registered as low priority netisr
3017  *   rollup function, is called; probably because no more work is pending
3018  *   for netisr.
3019  *
3020  * NOTE:
3021  * Currently subqueue packet staging is only performed in netisr threads.
3022  */
3023 int
3024 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
3025 {
3026         struct ifaltq *ifq = &ifp->if_snd;
3027         struct ifaltq_subque *ifsq;
3028         int error, start = 0, len, mcast = 0, avoid_start = 0;
3029         struct ifsubq_stage_head *head = NULL;
3030         struct ifsubq_stage *stage = NULL;
3031         struct globaldata *gd = mycpu;
3032         struct thread *td = gd->gd_curthread;
3033
3034         crit_enter_quick(td);
3035
3036         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
3037         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3038
3039         len = m->m_pkthdr.len;
3040         if (m->m_flags & M_MCAST)
3041                 mcast = 1;
3042
3043         if (td->td_type == TD_TYPE_NETISR) {
3044                 head = &ifsubq_stage_heads[mycpuid];
3045                 stage = ifsq_get_stage(ifsq, mycpuid);
3046
3047                 stage->stg_cnt++;
3048                 stage->stg_len += len;
3049                 if (stage->stg_cnt < ifsq_stage_cntmax &&
3050                     stage->stg_len < (ifp->if_mtu - max_protohdr))
3051                         avoid_start = 1;
3052         }
3053
3054         ALTQ_SQ_LOCK(ifsq);
3055         error = ifsq_enqueue_locked(ifsq, m, pa);
3056         if (error) {
3057                 if (!ifsq_data_ready(ifsq)) {
3058                         ALTQ_SQ_UNLOCK(ifsq);
3059                         crit_exit_quick(td);
3060                         return error;
3061                 }
3062                 avoid_start = 0;
3063         }
3064         if (!ifsq_is_started(ifsq)) {
3065                 if (avoid_start) {
3066                         ALTQ_SQ_UNLOCK(ifsq);
3067
3068                         KKASSERT(!error);
3069                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3070                                 ifsq_stage_insert(head, stage);
3071
3072                         IFNET_STAT_INC(ifp, obytes, len);
3073                         if (mcast)
3074                                 IFNET_STAT_INC(ifp, omcasts, 1);
3075                         crit_exit_quick(td);
3076                         return error;
3077                 }
3078
3079                 /*
3080                  * Hold the subqueue interlock of ifnet.if_start
3081                  */
3082                 ifsq_set_started(ifsq);
3083                 start = 1;
3084         }
3085         ALTQ_SQ_UNLOCK(ifsq);
3086
3087         if (!error) {
3088                 IFNET_STAT_INC(ifp, obytes, len);
3089                 if (mcast)
3090                         IFNET_STAT_INC(ifp, omcasts, 1);
3091         }
3092
3093         if (stage != NULL) {
3094                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3095                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3096                         if (!avoid_start) {
3097                                 ifsq_stage_remove(head, stage);
3098                                 ifsq_ifstart_schedule(ifsq, 1);
3099                         }
3100                         crit_exit_quick(td);
3101                         return error;
3102                 }
3103
3104                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3105                         ifsq_stage_remove(head, stage);
3106                 } else {
3107                         stage->stg_cnt = 0;
3108                         stage->stg_len = 0;
3109                 }
3110         }
3111
3112         if (!start) {
3113                 crit_exit_quick(td);
3114                 return error;
3115         }
3116
3117         ifsq_ifstart_try(ifsq, 0);
3118
3119         crit_exit_quick(td);
3120         return error;
3121 }
3122
3123 void *
3124 ifa_create(int size)
3125 {
3126         struct ifaddr *ifa;
3127         int i;
3128
3129         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3130
3131         ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3132         ifa->ifa_containers =
3133             kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
3134                 M_IFADDR, M_INTWAIT | M_ZERO);
3135
3136         ifa->ifa_ncnt = ncpus;
3137         for (i = 0; i < ncpus; ++i) {
3138                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3139
3140                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3141                 ifac->ifa = ifa;
3142                 ifac->ifa_refcnt = 1;
3143         }
3144 #ifdef IFADDR_DEBUG
3145         kprintf("alloc ifa %p %d\n", ifa, size);
3146 #endif
3147         return ifa;
3148 }
3149
3150 void
3151 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3152 {
3153         struct ifaddr *ifa = ifac->ifa;
3154
3155         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3156         KKASSERT(ifac->ifa_refcnt == 0);
3157         KASSERT(ifac->ifa_listmask == 0,
3158                 ("ifa is still on %#x lists", ifac->ifa_listmask));
3159
3160         ifac->ifa_magic = IFA_CONTAINER_DEAD;
3161
3162 #ifdef IFADDR_DEBUG_VERBOSE
3163         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3164 #endif
3165
3166         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3167                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
3168         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3169 #ifdef IFADDR_DEBUG
3170                 kprintf("free ifa %p\n", ifa);
3171 #endif
3172                 kfree(ifa->ifa_containers, M_IFADDR);
3173                 kfree(ifa, M_IFADDR);
3174         }
3175 }
3176
3177 static void
3178 ifa_iflink_dispatch(netmsg_t nmsg)
3179 {
3180         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3181         struct ifaddr *ifa = msg->ifa;
3182         struct ifnet *ifp = msg->ifp;
3183         int cpu = mycpuid;
3184         struct ifaddr_container *ifac;
3185
3186         crit_enter();
3187
3188         ifac = &ifa->ifa_containers[cpu];
3189         ASSERT_IFAC_VALID(ifac);
3190         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3191                 ("ifaddr is on if_addrheads"));
3192
3193         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3194         if (msg->tail)
3195                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3196         else
3197                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3198
3199         crit_exit();
3200
3201         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
3202 }
3203
3204 void
3205 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3206 {
3207         struct netmsg_ifaddr msg;
3208
3209         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3210                     0, ifa_iflink_dispatch);
3211         msg.ifa = ifa;
3212         msg.ifp = ifp;
3213         msg.tail = tail;
3214
3215         ifa_domsg(&msg.base.lmsg, 0);
3216 }
3217
3218 static void
3219 ifa_ifunlink_dispatch(netmsg_t nmsg)
3220 {
3221         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3222         struct ifaddr *ifa = msg->ifa;
3223         struct ifnet *ifp = msg->ifp;
3224         int cpu = mycpuid;
3225         struct ifaddr_container *ifac;
3226
3227         crit_enter();
3228
3229         ifac = &ifa->ifa_containers[cpu];
3230         ASSERT_IFAC_VALID(ifac);
3231         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3232                 ("ifaddr is not on if_addrhead"));
3233
3234         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3235         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3236
3237         crit_exit();
3238
3239         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
3240 }
3241
3242 void
3243 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3244 {
3245         struct netmsg_ifaddr msg;
3246
3247         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3248                     0, ifa_ifunlink_dispatch);
3249         msg.ifa = ifa;
3250         msg.ifp = ifp;
3251
3252         ifa_domsg(&msg.base.lmsg, 0);
3253 }
3254
3255 static void
3256 ifa_destroy_dispatch(netmsg_t nmsg)
3257 {
3258         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3259
3260         IFAFREE(msg->ifa);
3261         ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3262 }
3263
3264 void
3265 ifa_destroy(struct ifaddr *ifa)
3266 {
3267         struct netmsg_ifaddr msg;
3268
3269         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3270                     0, ifa_destroy_dispatch);
3271         msg.ifa = ifa;
3272
3273         ifa_domsg(&msg.base.lmsg, 0);
3274 }
3275
3276 struct lwkt_port *
3277 ifnet_portfn(int cpu)
3278 {
3279         return &ifnet_threads[cpu].td_msgport;
3280 }
3281
3282 void
3283 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
3284 {
3285         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
3286
3287         if (next_cpu < ncpus)
3288                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
3289         else
3290                 lwkt_replymsg(lmsg, 0);
3291 }
3292
3293 int
3294 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
3295 {
3296         KKASSERT(cpu < ncpus);
3297         return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
3298 }
3299
3300 void
3301 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
3302 {
3303         KKASSERT(cpu < ncpus);
3304         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
3305 }
3306
3307 /*
3308  * Generic netmsg service loop.  Some protocols may roll their own but all
3309  * must do the basic command dispatch function call done here.
3310  */
3311 static void
3312 ifnet_service_loop(void *arg __unused)
3313 {
3314         netmsg_t msg;
3315
3316         while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
3317                 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
3318                 msg->base.nm_dispatch(msg);
3319         }
3320 }
3321
3322 static void
3323 if_start_rollup(void)
3324 {
3325         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3326         struct ifsubq_stage *stage;
3327
3328         crit_enter();
3329
3330         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3331                 struct ifaltq_subque *ifsq = stage->stg_subq;
3332                 int is_sched = 0;
3333
3334                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3335                         is_sched = 1;
3336                 ifsq_stage_remove(head, stage);
3337
3338                 if (is_sched) {
3339                         ifsq_ifstart_schedule(ifsq, 1);
3340                 } else {
3341                         int start = 0;
3342
3343                         ALTQ_SQ_LOCK(ifsq);
3344                         if (!ifsq_is_started(ifsq)) {
3345                                 /*
3346                                  * Hold the subqueue interlock of
3347                                  * ifnet.if_start
3348                                  */
3349                                 ifsq_set_started(ifsq);
3350                                 start = 1;
3351                         }
3352                         ALTQ_SQ_UNLOCK(ifsq);
3353
3354                         if (start)
3355                                 ifsq_ifstart_try(ifsq, 1);
3356                 }
3357                 KKASSERT((stage->stg_flags &
3358                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3359         }
3360
3361         crit_exit();
3362 }
3363
3364 static void
3365 ifnetinit(void *dummy __unused)
3366 {
3367         int i;
3368
3369         for (i = 0; i < ncpus; ++i) {
3370                 struct thread *thr = &ifnet_threads[i];
3371
3372                 lwkt_create(ifnet_service_loop, NULL, NULL,
3373                             thr, TDF_NOSTART|TDF_FORCE_SPINPORT|TDF_FIXEDCPU,
3374                             i, "ifnet %d", i);
3375                 netmsg_service_port_init(&thr->td_msgport);
3376                 lwkt_schedule(thr);
3377         }
3378
3379         for (i = 0; i < ncpus; ++i)
3380                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3381         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3382 }
3383
3384 void
3385 if_register_com_alloc(u_char type,
3386     if_com_alloc_t *a, if_com_free_t *f)
3387 {
3388
3389         KASSERT(if_com_alloc[type] == NULL,
3390             ("if_register_com_alloc: %d already registered", type));
3391         KASSERT(if_com_free[type] == NULL,
3392             ("if_register_com_alloc: %d free already registered", type));
3393
3394         if_com_alloc[type] = a;
3395         if_com_free[type] = f;
3396 }
3397
3398 void
3399 if_deregister_com_alloc(u_char type)
3400 {
3401
3402         KASSERT(if_com_alloc[type] != NULL,
3403             ("if_deregister_com_alloc: %d not registered", type));
3404         KASSERT(if_com_free[type] != NULL,
3405             ("if_deregister_com_alloc: %d free not registered", type));
3406         if_com_alloc[type] = NULL;
3407         if_com_free[type] = NULL;
3408 }
3409
3410 int
3411 if_ring_count2(int cnt, int cnt_max)
3412 {
3413         int shift = 0;
3414
3415         KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3416             ("invalid ring count max %d", cnt_max));
3417
3418         if (cnt <= 0)
3419                 cnt = cnt_max;
3420         if (cnt > ncpus2)
3421                 cnt = ncpus2;
3422         if (cnt > cnt_max)
3423                 cnt = cnt_max;
3424
3425         while ((1 << (shift + 1)) <= cnt)
3426                 ++shift;
3427         cnt = 1 << shift;
3428
3429         KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3430             ("calculate cnt %d, ncpus2 %d, cnt max %d",
3431              cnt, ncpus2, cnt_max));
3432         return cnt;
3433 }
3434
3435 void
3436 ifq_set_maxlen(struct ifaltq *ifq, int len)
3437 {
3438         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3439 }
3440
3441 int
3442 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3443 {
3444         return ALTQ_SUBQ_INDEX_DEFAULT;
3445 }
3446
3447 int
3448 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3449 {
3450         return (cpuid & ifq->altq_subq_mask);
3451 }
3452
3453 static void
3454 ifsq_watchdog(void *arg)
3455 {
3456         struct ifsubq_watchdog *wd = arg;
3457         struct ifnet *ifp;
3458
3459         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3460                 goto done;
3461
3462         ifp = ifsq_get_ifp(wd->wd_subq);
3463         if (ifnet_tryserialize_all(ifp)) {
3464                 wd->wd_watchdog(wd->wd_subq);
3465                 ifnet_deserialize_all(ifp);
3466         } else {
3467                 /* try again next timeout */
3468                 wd->wd_timer = 1;
3469         }
3470 done:
3471         ifsq_watchdog_reset(wd);
3472 }
3473
3474 static void
3475 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3476 {
3477         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3478             ifsq_get_cpuid(wd->wd_subq));
3479 }
3480
3481 void
3482 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3483     ifsq_watchdog_t watchdog)
3484 {
3485         callout_init_mp(&wd->wd_callout);
3486         wd->wd_timer = 0;
3487         wd->wd_subq = ifsq;
3488         wd->wd_watchdog = watchdog;
3489 }
3490
3491 void
3492 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3493 {
3494         wd->wd_timer = 0;
3495         ifsq_watchdog_reset(wd);
3496 }
3497
3498 void
3499 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3500 {
3501         wd->wd_timer = 0;
3502         callout_stop(&wd->wd_callout);
3503 }
3504
3505 void
3506 ifnet_lock(void)
3507 {
3508         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3509             ("try holding ifnet lock in netisr"));
3510         mtx_lock(&ifnet_mtx);
3511 }
3512
3513 void
3514 ifnet_unlock(void)
3515 {
3516         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3517             ("try holding ifnet lock in netisr"));
3518         mtx_unlock(&ifnet_mtx);
3519 }
3520
3521 static struct ifnet_array *
3522 ifnet_array_alloc(int count)
3523 {
3524         struct ifnet_array *arr;
3525
3526         arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3527             M_IFNET, M_WAITOK);
3528         arr->ifnet_count = count;
3529
3530         return arr;
3531 }
3532
3533 static void
3534 ifnet_array_free(struct ifnet_array *arr)
3535 {
3536         if (arr == &ifnet_array0)
3537                 return;
3538         kfree(arr, M_IFNET);
3539 }
3540
3541 static struct ifnet_array *
3542 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3543 {
3544         struct ifnet_array *arr;
3545         int count, i;
3546
3547         KASSERT(old_arr->ifnet_count >= 0,
3548             ("invalid ifnet array count %d", old_arr->ifnet_count));
3549         count = old_arr->ifnet_count + 1;
3550         arr = ifnet_array_alloc(count);
3551
3552         /*
3553          * Save the old ifnet array and append this ifp to the end of
3554          * the new ifnet array.
3555          */
3556         for (i = 0; i < old_arr->ifnet_count; ++i) {
3557                 KASSERT(old_arr->ifnet_arr[i] != ifp,
3558                     ("%s is already in ifnet array", ifp->if_xname));
3559                 arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3560         }
3561         KASSERT(i == count - 1,
3562             ("add %s, ifnet array index mismatch, should be %d, but got %d",
3563              ifp->if_xname, count - 1, i));
3564         arr->ifnet_arr[i] = ifp;
3565
3566         return arr;
3567 }
3568
3569 static struct ifnet_array *
3570 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3571 {
3572         struct ifnet_array *arr;
3573         int count, i, idx, found = 0;
3574
3575         KASSERT(old_arr->ifnet_count > 0,
3576             ("invalid ifnet array count %d", old_arr->ifnet_count));
3577         count = old_arr->ifnet_count - 1;
3578         arr = ifnet_array_alloc(count);
3579
3580         /*
3581          * Save the old ifnet array, but skip this ifp.
3582          */
3583         idx = 0;
3584         for (i = 0; i < old_arr->ifnet_count; ++i) {
3585                 if (old_arr->ifnet_arr[i] == ifp) {
3586                         KASSERT(!found,
3587                             ("dup %s is in ifnet array", ifp->if_xname));
3588                         found = 1;
3589                         continue;
3590                 }
3591                 KASSERT(idx < count,
3592                     ("invalid ifnet array index %d, count %d", idx, count));
3593                 arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3594                 ++idx;
3595         }
3596         KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3597         KASSERT(idx == count,
3598             ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3599              ifp->if_xname, count, idx));
3600
3601         return arr;
3602 }
3603
3604 const struct ifnet_array *
3605 ifnet_array_get(void)
3606 {
3607         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3608         return ifnet_array;
3609 }
3610
3611 int
3612 ifnet_array_isempty(void)
3613 {
3614         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3615         if (ifnet_array->ifnet_count == 0)
3616                 return 1;
3617         else
3618                 return 0;
3619 }
3620
3621 void
3622 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3623 {
3624         struct ifaddr *ifa;
3625
3626         memset(mark, 0, sizeof(*mark));
3627         ifa = &mark->ifa;
3628
3629         mark->ifac.ifa = ifa;
3630
3631         ifa->ifa_addr = &mark->addr;
3632         ifa->ifa_dstaddr = &mark->dstaddr;
3633         ifa->ifa_netmask = &mark->netmask;
3634         ifa->ifa_ifp = ifp;
3635 }