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