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. All advertising materials mentioning features or use of this software
  14  *    must display the following acknowledgement:
  15  *      This product includes software developed by the University of
  16  *      California, Berkeley and its contributors.
  17  * 4. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  *      @(#)if.c        8.3 (Berkeley) 1/4/94
  34  * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
  35  */
  36
  37 #include "opt_compat.h"
  38 #include "opt_inet6.h"
  39 #include "opt_inet.h"
  40 #include "opt_ifpoll.h"
  41
  42 #include <sys/param.h>
  43 #include <sys/malloc.h>
  44 #include <sys/mbuf.h>
  45 #include <sys/systm.h>
  46 #include <sys/proc.h>
  47 #include <sys/priv.h>
  48 #include <sys/protosw.h>
  49 #include <sys/socket.h>
  50 #include <sys/socketvar.h>
  51 #include <sys/socketops.h>
  52 #include <sys/protosw.h>
  53 #include <sys/kernel.h>
  54 #include <sys/ktr.h>
  55 #include <sys/mutex.h>
  56 #include <sys/sockio.h>
  57 #include <sys/syslog.h>
  58 #include <sys/sysctl.h>
  59 #include <sys/domain.h>
  60 #include <sys/thread.h>
  61 #include <sys/serialize.h>
  62 #include <sys/bus.h>
  63
  64 #include <sys/thread2.h>
  65 #include <sys/msgport2.h>
  66 #include <sys/mutex2.h>
  67
  68 #include <net/if.h>
  69 #include <net/if_arp.h>
  70 #include <net/if_dl.h>
  71 #include <net/if_types.h>
  72 #include <net/if_var.h>
  73 #include <net/ifq_var.h>
  74 #include <net/radix.h>
  75 #include <net/route.h>
  76 #include <net/if_clone.h>
  77 #include <net/netisr.h>
  78 #include <net/netmsg2.h>
  79
  80 #include <machine/atomic.h>
  81 #include <machine/stdarg.h>
  82 #include <machine/smp.h>
  83
  84 #if defined(INET) || defined(INET6)
  85 /*XXX*/
  86 #include <netinet/in.h>
  87 #include <netinet/in_var.h>
  88 #include <netinet/if_ether.h>
  89 #ifdef INET6
  90 #include <netinet6/in6_var.h>
  91 #include <netinet6/in6_ifattach.h>
  92 #endif
  93 #endif
  94
  95 #if defined(COMPAT_43)
  96 #include <emulation/43bsd/43bsd_socket.h>
  97 #endif /* COMPAT_43 */
  98
  99 struct netmsg_ifaddr {
 100         struct netmsg_base base;
 101         struct ifaddr   *ifa;
 102         struct ifnet    *ifp;
 103         int             tail;
 104 };
 105
 106 struct ifsubq_stage_head {
 107         TAILQ_HEAD(, ifsubq_stage)      stg_head;
 108 } __cachealign;
 109
 110 /*
 111  * System initialization
 112  */
 113 static void     if_attachdomain(void *);
 114 static void     if_attachdomain1(struct ifnet *);
 115 static int      ifconf(u_long, caddr_t, struct ucred *);
 116 static void     ifinit(void *);
 117 static void     ifnetinit(void *);
 118 static void     if_slowtimo(void *);
 119 static void     link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
 120 static int      if_rtdel(struct radix_node *, void *);
 121
 122 /* Helper functions */
 123 static void     ifsq_watchdog_reset(struct ifsubq_watchdog *);
 124
 125 #ifdef INET6
 126 /*
 127  * XXX: declare here to avoid to include many inet6 related files..
 128  * should be more generalized?
 129  */
 130 extern void     nd6_setmtu(struct ifnet *);
 131 #endif
 132
 133 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
 134 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
 135
 136 static int ifsq_stage_cntmax = 4;
 137 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
 138 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
 139     &ifsq_stage_cntmax, 0, "ifq staging packet count max");
 140
 141 static int if_stats_compat = 0;
 142 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
 143     &if_stats_compat, 0, "Compat the old ifnet stats");
 144
 145 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
 146 /* Must be after netisr_init */
 147 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifnetinit, NULL)
 148
 149 static  if_com_alloc_t *if_com_alloc[256];
 150 static  if_com_free_t *if_com_free[256];
 151
 152 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
 153 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
 154 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
 155
 156 int                     ifqmaxlen = IFQ_MAXLEN;
 157 struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
 158
 159 struct callout          if_slowtimo_timer;
 160
 161 int                     if_index = 0;
 162 struct ifnet            **ifindex2ifnet = NULL;
 163 static struct thread    ifnet_threads[MAXCPU];
 164
 165 static struct ifsubq_stage_head ifsubq_stage_heads[MAXCPU];
 166
 167 #ifdef notyet
 168 #define IFQ_KTR_STRING          "ifq=%p"
 169 #define IFQ_KTR_ARGS    struct ifaltq *ifq
 170 #ifndef KTR_IFQ
 171 #define KTR_IFQ                 KTR_ALL
 172 #endif
 173 KTR_INFO_MASTER(ifq);
 174 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
 175 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
 176 #define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)
 177
 178 #define IF_START_KTR_STRING     "ifp=%p"
 179 #define IF_START_KTR_ARGS       struct ifnet *ifp
 180 #ifndef KTR_IF_START
 181 #define KTR_IF_START            KTR_ALL
 182 #endif
 183 KTR_INFO_MASTER(if_start);
 184 KTR_INFO(KTR_IF_START, if_start, run, 0,
 185          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 186 KTR_INFO(KTR_IF_START, if_start, sched, 1,
 187          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 188 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
 189          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 190 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
 191          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 192 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
 193          IF_START_KTR_STRING, IF_START_KTR_ARGS);
 194 #define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
 195 #endif
 196
 197 TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
 198
 199 /*
 200  * Network interface utility routines.
 201  *
 202  * Routines with ifa_ifwith* names take sockaddr *'s as
 203  * parameters.
 204  */
 205 /* ARGSUSED*/
 206 void
 207 ifinit(void *dummy)
 208 {
 209         struct ifnet *ifp;
 210
 211         callout_init(&if_slowtimo_timer);
 212
 213         crit_enter();
 214         TAILQ_FOREACH(ifp, &ifnet, if_link) {
 215                 if (ifp->if_snd.altq_maxlen == 0) {
 216                         if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
 217                         ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
 218                 }
 219         }
 220         crit_exit();
 221
 222         if_slowtimo(0);
 223 }
 224
 225 static void
 226 ifsq_ifstart_ipifunc(void *arg)
 227 {
 228         struct ifaltq_subque *ifsq = arg;
 229         struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
 230
 231         crit_enter();
 232         if (lmsg->ms_flags & MSGF_DONE)
 233                 lwkt_sendmsg(netisr_portfn(mycpuid), lmsg);
 234         crit_exit();
 235 }
 236
 237 static __inline void
 238 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
 239 {
 240         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
 241         TAILQ_REMOVE(&head->stg_head, stage, stg_link);
 242         stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
 243         stage->stg_cnt = 0;
 244         stage->stg_len = 0;
 245 }
 246
 247 static __inline void
 248 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
 249 {
 250         KKASSERT((stage->stg_flags &
 251             (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
 252         stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
 253         TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
 254 }
 255
 256 /*
 257  * Schedule ifnet.if_start on ifnet's CPU
 258  */
 259 static void
 260 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
 261 {
 262         int cpu;
 263
 264         if (!force && curthread->td_type == TD_TYPE_NETISR &&
 265             ifsq_stage_cntmax > 0) {
 266                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 267
 268                 stage->stg_cnt = 0;
 269                 stage->stg_len = 0;
 270                 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
 271                         ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
 272                 stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
 273                 return;
 274         }
 275
 276         cpu = ifsq_get_cpuid(ifsq);
 277         if (cpu != mycpuid)
 278                 lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
 279         else
 280                 ifsq_ifstart_ipifunc(ifsq);
 281 }
 282
 283 /*
 284  * NOTE:
 285  * This function will release ifnet.if_start interlock,
 286  * if ifnet.if_start does not need to be scheduled
 287  */
 288 static __inline int
 289 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
 290 {
 291         if (!running || ifsq_is_empty(ifsq)
 292 #ifdef ALTQ
 293             || ifsq->ifsq_altq->altq_tbr != NULL
 294 #endif
 295         ) {
 296                 ALTQ_SQ_LOCK(ifsq);
 297                 /*
 298                  * ifnet.if_start interlock is released, if:
 299                  * 1) Hardware can not take any packets, due to
 300                  *    o  interface is marked down
 301                  *    o  hardware queue is full (ifq_is_oactive)
 302                  *    Under the second situation, hardware interrupt
 303                  *    or polling(4) will call/schedule ifnet.if_start
 304                  *    when hardware queue is ready
 305                  * 2) There is not packet in the ifnet.if_snd.
 306                  *    Further ifq_dispatch or ifq_handoff will call/
 307                  *    schedule ifnet.if_start
 308                  * 3) TBR is used and it does not allow further
 309                  *    dequeueing.
 310                  *    TBR callout will call ifnet.if_start
 311                  */
 312                 if (!running || !ifsq_data_ready(ifsq)) {
 313                         ifsq_clr_started(ifsq);
 314                         ALTQ_SQ_UNLOCK(ifsq);
 315                         return 0;
 316                 }
 317                 ALTQ_SQ_UNLOCK(ifsq);
 318         }
 319         return 1;
 320 }
 321
 322 static void
 323 ifsq_ifstart_dispatch(netmsg_t msg)
 324 {
 325         struct lwkt_msg *lmsg = &msg->base.lmsg;
 326         struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
 327         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 328         int running = 0, need_sched;
 329
 330         crit_enter();
 331         lwkt_replymsg(lmsg, 0); /* reply ASAP */
 332         crit_exit();
 333
 334         if (mycpuid != ifsq_get_cpuid(ifsq)) {
 335                 /*
 336                  * We need to chase the ifnet CPU change.
 337                  */
 338                 ifsq_ifstart_schedule(ifsq, 1);
 339                 return;
 340         }
 341
 342         ifsq_serialize_hw(ifsq);
 343         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
 344                 ifp->if_start(ifp, ifsq);
 345                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 346                         running = 1;
 347         }
 348         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
 349         ifsq_deserialize_hw(ifsq);
 350
 351         if (need_sched) {
 352                 /*
 353                  * More data need to be transmitted, ifnet.if_start is
 354                  * scheduled on ifnet's CPU, and we keep going.
 355                  * NOTE: ifnet.if_start interlock is not released.
 356                  */
 357                 ifsq_ifstart_schedule(ifsq, 0);
 358         }
 359 }
 360
 361 /* Device driver ifnet.if_start helper function */
 362 void
 363 ifsq_devstart(struct ifaltq_subque *ifsq)
 364 {
 365         struct ifnet *ifp = ifsq_get_ifp(ifsq);
 366         int running = 0;
 367
 368         ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
 369
 370         ALTQ_SQ_LOCK(ifsq);
 371         if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
 372                 ALTQ_SQ_UNLOCK(ifsq);
 373                 return;
 374         }
 375         ifsq_set_started(ifsq);
 376         ALTQ_SQ_UNLOCK(ifsq);
 377
 378         ifp->if_start(ifp, ifsq);
 379
 380         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
 381                 running = 1;
 382
 383         if (ifsq_ifstart_need_schedule(ifsq, running)) {
 384                 /*
 385                  * More data need to be transmitted, ifnet.if_start is
 386                  * scheduled on ifnet's CPU, and we keep going.
 387                  * NOTE: ifnet.if_start interlock is not released.
 388                  */
 389                 ifsq_ifstart_schedule(ifsq, 0);
 390         }
 391 }
 392
 393 void
 394 if_devstart(struct ifnet *ifp)
 395 {
 396         ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
 397 }
 398
 399 /* Device driver ifnet.if_start schedule helper function */
 400 void
 401 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
 402 {
 403         ifsq_ifstart_schedule(ifsq, 1);
 404 }
 405
 406 void
 407 if_devstart_sched(struct ifnet *ifp)
 408 {
 409         ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
 410 }
 411
 412 static void
 413 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 414 {
 415         lwkt_serialize_enter(ifp->if_serializer);
 416 }
 417
 418 static void
 419 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 420 {
 421         lwkt_serialize_exit(ifp->if_serializer);
 422 }
 423
 424 static int
 425 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
 426 {
 427         return lwkt_serialize_try(ifp->if_serializer);
 428 }
 429
 430 #ifdef INVARIANTS
 431 static void
 432 if_default_serialize_assert(struct ifnet *ifp,
 433                             enum ifnet_serialize slz __unused,
 434                             boolean_t serialized)
 435 {
 436         if (serialized)
 437                 ASSERT_SERIALIZED(ifp->if_serializer);
 438         else
 439                 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
 440 }
 441 #endif
 442
 443 /*
 444  * Attach an interface to the list of "active" interfaces.
 445  *
 446  * The serializer is optional.  If non-NULL access to the interface
 447  * may be MPSAFE.
 448  */
 449 void
 450 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
 451 {
 452         unsigned socksize, ifasize;
 453         int namelen, masklen;
 454         struct sockaddr_dl *sdl;
 455         struct ifaddr *ifa;
 456         struct ifaltq *ifq;
 457         int i, q;
 458
 459         static int if_indexlim = 8;
 460
 461         if (ifp->if_serialize != NULL) {
 462                 KASSERT(ifp->if_deserialize != NULL &&
 463                         ifp->if_tryserialize != NULL &&
 464                         ifp->if_serialize_assert != NULL,
 465                         ("serialize functions are partially setup"));
 466
 467                 /*
 468                  * If the device supplies serialize functions,
 469                  * then clear if_serializer to catch any invalid
 470                  * usage of this field.
 471                  */
 472                 KASSERT(serializer == NULL,
 473                         ("both serialize functions and default serializer "
 474                          "are supplied"));
 475                 ifp->if_serializer = NULL;
 476         } else {
 477                 KASSERT(ifp->if_deserialize == NULL &&
 478                         ifp->if_tryserialize == NULL &&
 479                         ifp->if_serialize_assert == NULL,
 480                         ("serialize functions are partially setup"));
 481                 ifp->if_serialize = if_default_serialize;
 482                 ifp->if_deserialize = if_default_deserialize;
 483                 ifp->if_tryserialize = if_default_tryserialize;
 484 #ifdef INVARIANTS
 485                 ifp->if_serialize_assert = if_default_serialize_assert;
 486 #endif
 487
 488                 /*
 489                  * The serializer can be passed in from the device,
 490                  * allowing the same serializer to be used for both
 491                  * the interrupt interlock and the device queue.
 492                  * If not specified, the netif structure will use an
 493                  * embedded serializer.
 494                  */
 495                 if (serializer == NULL) {
 496                         serializer = &ifp->if_default_serializer;
 497                         lwkt_serialize_init(serializer);
 498                 }
 499                 ifp->if_serializer = serializer;
 500         }
 501
 502         mtx_init(&ifp->if_ioctl_mtx);
 503         mtx_lock(&ifp->if_ioctl_mtx);
 504
 505         TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
 506         ifp->if_index = ++if_index;
 507
 508         /*
 509          * XXX -
 510          * The old code would work if the interface passed a pre-existing
 511          * chain of ifaddrs to this code.  We don't trust our callers to
 512          * properly initialize the tailq, however, so we no longer allow
 513          * this unlikely case.
 514          */
 515         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
 516                                     M_IFADDR, M_WAITOK | M_ZERO);
 517         for (i = 0; i < ncpus; ++i)
 518                 TAILQ_INIT(&ifp->if_addrheads[i]);
 519
 520         TAILQ_INIT(&ifp->if_prefixhead);
 521         TAILQ_INIT(&ifp->if_multiaddrs);
 522         TAILQ_INIT(&ifp->if_groups);
 523         getmicrotime(&ifp->if_lastchange);
 524         if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
 525                 unsigned int n;
 526                 struct ifnet **q;
 527
 528                 if_indexlim <<= 1;
 529
 530                 /* grow ifindex2ifnet */
 531                 n = if_indexlim * sizeof(*q);
 532                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
 533                 if (ifindex2ifnet) {
 534                         bcopy(ifindex2ifnet, q, n/2);
 535                         kfree(ifindex2ifnet, M_IFADDR);
 536                 }
 537                 ifindex2ifnet = q;
 538         }
 539
 540         ifindex2ifnet[if_index] = ifp;
 541
 542         /*
 543          * create a Link Level name for this device
 544          */
 545         namelen = strlen(ifp->if_xname);
 546         masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
 547         socksize = masklen + ifp->if_addrlen;
 548 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
 549         if (socksize < sizeof(*sdl))
 550                 socksize = sizeof(*sdl);
 551         socksize = ROUNDUP(socksize);
 552 #undef ROUNDUP
 553         ifasize = sizeof(struct ifaddr) + 2 * socksize;
 554         ifa = ifa_create(ifasize, M_WAITOK);
 555         sdl = (struct sockaddr_dl *)(ifa + 1);
 556         sdl->sdl_len = socksize;
 557         sdl->sdl_family = AF_LINK;
 558         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
 559         sdl->sdl_nlen = namelen;
 560         sdl->sdl_index = ifp->if_index;
 561         sdl->sdl_type = ifp->if_type;
 562         ifp->if_lladdr = ifa;
 563         ifa->ifa_ifp = ifp;
 564         ifa->ifa_rtrequest = link_rtrequest;
 565         ifa->ifa_addr = (struct sockaddr *)sdl;
 566         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
 567         ifa->ifa_netmask = (struct sockaddr *)sdl;
 568         sdl->sdl_len = masklen;
 569         while (namelen != 0)
 570                 sdl->sdl_data[--namelen] = 0xff;
 571         ifa_iflink(ifa, ifp, 0 /* Insert head */);
 572
 573         ifp->if_data_pcpu = kmalloc_cachealign(
 574             ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);
 575
 576         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
 577         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
 578
 579         if (ifp->if_mapsubq == NULL)
 580                 ifp->if_mapsubq = ifq_mapsubq_default;
 581
 582         ifq = &ifp->if_snd;
 583         ifq->altq_type = 0;
 584         ifq->altq_disc = NULL;
 585         ifq->altq_flags &= ALTQF_CANTCHANGE;
 586         ifq->altq_tbr = NULL;
 587         ifq->altq_ifp = ifp;
 588
 589         if (ifq->altq_subq_cnt <= 0)
 590                 ifq->altq_subq_cnt = 1;
 591         ifq->altq_subq = kmalloc_cachealign(
 592             ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
 593             M_DEVBUF, M_WAITOK | M_ZERO);
 594
 595         if (ifq->altq_maxlen == 0) {
 596                 if_printf(ifp, "driver didn't set ifq_maxlen\n");
 597                 ifq_set_maxlen(ifq, ifqmaxlen);
 598         }
 599
 600         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 601                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 602
 603                 ALTQ_SQ_LOCK_INIT(ifsq);
 604                 ifsq->ifsq_index = q;
 605
 606                 ifsq->ifsq_altq = ifq;
 607                 ifsq->ifsq_ifp = ifp;
 608
 609                 ifsq->ifq_maxlen = ifq->altq_maxlen;
 610                 ifsq->ifsq_prepended = NULL;
 611                 ifsq->ifsq_started = 0;
 612                 ifsq->ifsq_hw_oactive = 0;
 613                 ifsq_set_cpuid(ifsq, 0);
 614                 if (ifp->if_serializer != NULL)
 615                         ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
 616
 617                 ifsq->ifsq_stage =
 618                     kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
 619                     M_DEVBUF, M_WAITOK | M_ZERO);
 620                 for (i = 0; i < ncpus; ++i)
 621                         ifsq->ifsq_stage[i].stg_subq = ifsq;
 622
 623                 ifsq->ifsq_ifstart_nmsg =
 624                     kmalloc(ncpus * sizeof(struct netmsg_base),
 625                     M_LWKTMSG, M_WAITOK);
 626                 for (i = 0; i < ncpus; ++i) {
 627                         netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
 628                             &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
 629                         ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
 630                 }
 631         }
 632         ifq_set_classic(ifq);
 633
 634         if (!SLIST_EMPTY(&domains))
 635                 if_attachdomain1(ifp);
 636
 637         /* Announce the interface. */
 638         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
 639
 640         mtx_unlock(&ifp->if_ioctl_mtx);
 641 }
 642
 643 static void
 644 if_attachdomain(void *dummy)
 645 {
 646         struct ifnet *ifp;
 647
 648         crit_enter();
 649         TAILQ_FOREACH(ifp, &ifnet, if_list)
 650                 if_attachdomain1(ifp);
 651         crit_exit();
 652 }
 653 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
 654         if_attachdomain, NULL);
 655
 656 static void
 657 if_attachdomain1(struct ifnet *ifp)
 658 {
 659         struct domain *dp;
 660
 661         crit_enter();
 662
 663         /* address family dependent data region */
 664         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
 665         SLIST_FOREACH(dp, &domains, dom_next)
 666                 if (dp->dom_ifattach)
 667                         ifp->if_afdata[dp->dom_family] =
 668                                 (*dp->dom_ifattach)(ifp);
 669         crit_exit();
 670 }
 671
 672 /*
 673  * Purge all addresses whose type is _not_ AF_LINK
 674  */
 675 void
 676 if_purgeaddrs_nolink(struct ifnet *ifp)
 677 {
 678         struct ifaddr_container *ifac, *next;
 679
 680         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
 681                               ifa_link, next) {
 682                 struct ifaddr *ifa = ifac->ifa;
 683
 684                 /* Leave link ifaddr as it is */
 685                 if (ifa->ifa_addr->sa_family == AF_LINK)
 686                         continue;
 687 #ifdef INET
 688                 /* XXX: Ugly!! ad hoc just for INET */
 689                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
 690                         struct ifaliasreq ifr;
 691 #ifdef IFADDR_DEBUG_VERBOSE
 692                         int i;
 693
 694                         kprintf("purge in4 addr %p: ", ifa);
 695                         for (i = 0; i < ncpus; ++i)
 696                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 697                         kprintf("\n");
 698 #endif
 699
 700                         bzero(&ifr, sizeof ifr);
 701                         ifr.ifra_addr = *ifa->ifa_addr;
 702                         if (ifa->ifa_dstaddr)
 703                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
 704                         if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
 705                                        NULL) == 0)
 706                                 continue;
 707                 }
 708 #endif /* INET */
 709 #ifdef INET6
 710                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
 711 #ifdef IFADDR_DEBUG_VERBOSE
 712                         int i;
 713
 714                         kprintf("purge in6 addr %p: ", ifa);
 715                         for (i = 0; i < ncpus; ++i)
 716                                 kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
 717                         kprintf("\n");
 718 #endif
 719
 720                         in6_purgeaddr(ifa);
 721                         /* ifp_addrhead is already updated */
 722                         continue;
 723                 }
 724 #endif /* INET6 */
 725                 ifa_ifunlink(ifa, ifp);
 726                 ifa_destroy(ifa);
 727         }
 728 }
 729
 730 static void
 731 ifq_stage_detach_handler(netmsg_t nmsg)
 732 {
 733         struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
 734         int q;
 735
 736         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
 737                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
 738                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
 739
 740                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
 741                         ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
 742         }
 743         lwkt_replymsg(&nmsg->lmsg, 0);
 744 }
 745
 746 static void
 747 ifq_stage_detach(struct ifaltq *ifq)
 748 {
 749         struct netmsg_base base;
 750         int cpu;
 751
 752         netmsg_init(&base, NULL, &curthread->td_msgport, 0,
 753             ifq_stage_detach_handler);
 754         base.lmsg.u.ms_resultp = ifq;
 755
 756         for (cpu = 0; cpu < ncpus; ++cpu)
 757                 lwkt_domsg(netisr_portfn(cpu), &base.lmsg, 0);
 758 }
 759
 760 /*
 761  * Detach an interface, removing it from the
 762  * list of "active" interfaces.
 763  */
 764 void
 765 if_detach(struct ifnet *ifp)
 766 {
 767         struct radix_node_head  *rnh;
 768         int i, q;
 769         int cpu, origcpu;
 770         struct domain *dp;
 771
 772         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
 773
 774         /*
 775          * Remove routes and flush queues.
 776          */
 777         crit_enter();
 778 #ifdef IFPOLL_ENABLE
 779         if (ifp->if_flags & IFF_NPOLLING)
 780                 ifpoll_deregister(ifp);
 781 #endif
 782         if_down(ifp);
 783
 784 #ifdef ALTQ
 785         if (ifq_is_enabled(&ifp->if_snd))
 786                 altq_disable(&ifp->if_snd);
 787         if (ifq_is_attached(&ifp->if_snd))
 788                 altq_detach(&ifp->if_snd);
 789 #endif
 790
 791         /*
 792          * Clean up all addresses.
 793          */
 794         ifp->if_lladdr = NULL;
 795
 796         if_purgeaddrs_nolink(ifp);
 797         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
 798                 struct ifaddr *ifa;
 799
 800                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
 801                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
 802                         ("non-link ifaddr is left on if_addrheads"));
 803
 804                 ifa_ifunlink(ifa, ifp);
 805                 ifa_destroy(ifa);
 806                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
 807                         ("there are still ifaddrs left on if_addrheads"));
 808         }
 809
 810 #ifdef INET
 811         /*
 812          * Remove all IPv4 kernel structures related to ifp.
 813          */
 814         in_ifdetach(ifp);
 815 #endif
 816
 817 #ifdef INET6
 818         /*
 819          * Remove all IPv6 kernel structs related to ifp.  This should be done
 820          * before removing routing entries below, since IPv6 interface direct
 821          * routes are expected to be removed by the IPv6-specific kernel API.
 822          * Otherwise, the kernel will detect some inconsistency and bark it.
 823          */
 824         in6_ifdetach(ifp);
 825 #endif
 826
 827         /*
 828          * Delete all remaining routes using this interface
 829          * Unfortuneatly the only way to do this is to slog through
 830          * the entire routing table looking for routes which point
 831          * to this interface...oh well...
 832          */
 833         origcpu = mycpuid;
 834         for (cpu = 0; cpu < ncpus; cpu++) {
 835                 lwkt_migratecpu(cpu);
 836                 for (i = 1; i <= AF_MAX; i++) {
 837                         if ((rnh = rt_tables[cpu][i]) == NULL)
 838                                 continue;
 839                         rnh->rnh_walktree(rnh, if_rtdel, ifp);
 840                 }
 841         }
 842         lwkt_migratecpu(origcpu);
 843
 844         /* Announce that the interface is gone. */
 845         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
 846         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
 847
 848         SLIST_FOREACH(dp, &domains, dom_next)
 849                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
 850                         (*dp->dom_ifdetach)(ifp,
 851                                 ifp->if_afdata[dp->dom_family]);
 852
 853         /*
 854          * Remove interface from ifindex2ifp[] and maybe decrement if_index.
 855          */
 856         ifindex2ifnet[ifp->if_index] = NULL;
 857         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
 858                 if_index--;
 859
 860         TAILQ_REMOVE(&ifnet, ifp, if_link);
 861         kfree(ifp->if_addrheads, M_IFADDR);
 862
 863         lwkt_synchronize_ipiqs("if_detach");
 864         ifq_stage_detach(&ifp->if_snd);
 865
 866         for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
 867                 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
 868
 869                 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
 870                 kfree(ifsq->ifsq_stage, M_DEVBUF);
 871         }
 872         kfree(ifp->if_snd.altq_subq, M_DEVBUF);
 873
 874         kfree(ifp->if_data_pcpu, M_DEVBUF);
 875
 876         crit_exit();
 877 }
 878
 879 /*
 880  * Create interface group without members
 881  */
 882 struct ifg_group *
 883 if_creategroup(const char *groupname)
 884 {
 885         struct ifg_group        *ifg = NULL;
 886
 887         if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
 888             M_TEMP, M_NOWAIT)) == NULL)
 889                 return (NULL);
 890
 891         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
 892         ifg->ifg_refcnt = 0;
 893         ifg->ifg_carp_demoted = 0;
 894         TAILQ_INIT(&ifg->ifg_members);
 895 #if NPF > 0
 896         pfi_attach_ifgroup(ifg);
 897 #endif
 898         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
 899
 900         return (ifg);
 901 }
 902
 903 /*
 904  * Add a group to an interface
 905  */
 906 int
 907 if_addgroup(struct ifnet *ifp, const char *groupname)
 908 {
 909         struct ifg_list         *ifgl;
 910         struct ifg_group        *ifg = NULL;
 911         struct ifg_member       *ifgm;
 912
 913         if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
 914             groupname[strlen(groupname) - 1] <= '9')
 915                 return (EINVAL);
 916
 917         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
 918                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
 919                         return (EEXIST);
 920
 921         if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
 922                 return (ENOMEM);
 923
 924         if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
 925                 kfree(ifgl, M_TEMP);
 926                 return (ENOMEM);
 927         }
 928
 929         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
 930                 if (!strcmp(ifg->ifg_group, groupname))
 931                         break;
 932
 933         if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
 934                 kfree(ifgl, M_TEMP);
 935                 kfree(ifgm, M_TEMP);
 936                 return (ENOMEM);
 937         }
 938
 939         ifg->ifg_refcnt++;
 940         ifgl->ifgl_group = ifg;
 941         ifgm->ifgm_ifp = ifp;
 942
 943         TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
 944         TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
 945
 946 #if NPF > 0
 947         pfi_group_change(groupname);
 948 #endif
 949
 950         return (0);
 951 }
 952
 953 /*
 954  * Remove a group from an interface
 955  */
 956 int
 957 if_delgroup(struct ifnet *ifp, const char *groupname)
 958 {
 959         struct ifg_list         *ifgl;
 960         struct ifg_member       *ifgm;
 961
 962         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
 963                 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
 964                         break;
 965         if (ifgl == NULL)
 966                 return (ENOENT);
 967
 968         TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
 969
 970         TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
 971                 if (ifgm->ifgm_ifp == ifp)
 972                         break;
 973
 974         if (ifgm != NULL) {
 975                 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
 976                 kfree(ifgm, M_TEMP);
 977         }
 978
 979         if (--ifgl->ifgl_group->ifg_refcnt == 0) {
 980                 TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
 981 #if NPF > 0
 982                 pfi_detach_ifgroup(ifgl->ifgl_group);
 983 #endif
 984                 kfree(ifgl->ifgl_group, M_TEMP);
 985         }
 986
 987         kfree(ifgl, M_TEMP);
 988
 989 #if NPF > 0
 990         pfi_group_change(groupname);
 991 #endif
 992
 993         return (0);
 994 }
 995
 996 /*
 997  * Stores all groups from an interface in memory pointed
 998  * to by data
 999  */
1000 int
1001 if_getgroup(caddr_t data, struct ifnet *ifp)
1002 {
1003         int                      len, error;
1004         struct ifg_list         *ifgl;
1005         struct ifg_req           ifgrq, *ifgp;
1006         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1007
1008         if (ifgr->ifgr_len == 0) {
1009                 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1010                         ifgr->ifgr_len += sizeof(struct ifg_req);
1011                 return (0);
1012         }
1013
1014         len = ifgr->ifgr_len;
1015         ifgp = ifgr->ifgr_groups;
1016         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1017                 if (len < sizeof(ifgrq))
1018                         return (EINVAL);
1019                 bzero(&ifgrq, sizeof ifgrq);
1020                 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1021                     sizeof(ifgrq.ifgrq_group));
1022                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1023                     sizeof(struct ifg_req))))
1024                         return (error);
1025                 len -= sizeof(ifgrq);
1026                 ifgp++;
1027         }
1028
1029         return (0);
1030 }
1031
1032 /*
1033  * Stores all members of a group in memory pointed to by data
1034  */
1035 int
1036 if_getgroupmembers(caddr_t data)
1037 {
1038         struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
1039         struct ifg_group        *ifg;
1040         struct ifg_member       *ifgm;
1041         struct ifg_req           ifgrq, *ifgp;
1042         int                      len, error;
1043
1044         TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
1045                 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1046                         break;
1047         if (ifg == NULL)
1048                 return (ENOENT);
1049
1050         if (ifgr->ifgr_len == 0) {
1051                 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1052                         ifgr->ifgr_len += sizeof(ifgrq);
1053                 return (0);
1054         }
1055
1056         len = ifgr->ifgr_len;
1057         ifgp = ifgr->ifgr_groups;
1058         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1059                 if (len < sizeof(ifgrq))
1060                         return (EINVAL);
1061                 bzero(&ifgrq, sizeof ifgrq);
1062                 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1063                     sizeof(ifgrq.ifgrq_member));
1064                 if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
1065                     sizeof(struct ifg_req))))
1066                         return (error);
1067                 len -= sizeof(ifgrq);
1068                 ifgp++;
1069         }
1070
1071         return (0);
1072 }
1073
1074 /*
1075  * Delete Routes for a Network Interface
1076  *
1077  * Called for each routing entry via the rnh->rnh_walktree() call above
1078  * to delete all route entries referencing a detaching network interface.
1079  *
1080  * Arguments:
1081  *      rn      pointer to node in the routing table
1082  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1083  *
1084  * Returns:
1085  *      0       successful
1086  *      errno   failed - reason indicated
1087  *
1088  */
1089 static int
1090 if_rtdel(struct radix_node *rn, void *arg)
1091 {
1092         struct rtentry  *rt = (struct rtentry *)rn;
1093         struct ifnet    *ifp = arg;
1094         int             err;
1095
1096         if (rt->rt_ifp == ifp) {
1097
1098                 /*
1099                  * Protect (sorta) against walktree recursion problems
1100                  * with cloned routes
1101                  */
1102                 if (!(rt->rt_flags & RTF_UP))
1103                         return (0);
1104
1105                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1106                                 rt_mask(rt), rt->rt_flags,
1107                                 NULL);
1108                 if (err) {
1109                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1110                 }
1111         }
1112
1113         return (0);
1114 }
1115
1116 /*
1117  * Locate an interface based on a complete address.
1118  */
1119 struct ifaddr *
1120 ifa_ifwithaddr(struct sockaddr *addr)
1121 {
1122         struct ifnet *ifp;
1123
1124         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1125                 struct ifaddr_container *ifac;
1126
1127                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1128                         struct ifaddr *ifa = ifac->ifa;
1129
1130                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1131                                 continue;
1132                         if (sa_equal(addr, ifa->ifa_addr))
1133                                 return (ifa);
1134                         if ((ifp->if_flags & IFF_BROADCAST) &&
1135                             ifa->ifa_broadaddr &&
1136                             /* IPv6 doesn't have broadcast */
1137                             ifa->ifa_broadaddr->sa_len != 0 &&
1138                             sa_equal(ifa->ifa_broadaddr, addr))
1139                                 return (ifa);
1140                 }
1141         }
1142         return (NULL);
1143 }
1144 /*
1145  * Locate the point to point interface with a given destination address.
1146  */
1147 struct ifaddr *
1148 ifa_ifwithdstaddr(struct sockaddr *addr)
1149 {
1150         struct ifnet *ifp;
1151
1152         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1153                 struct ifaddr_container *ifac;
1154
1155                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1156                         continue;
1157
1158                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1159                         struct ifaddr *ifa = ifac->ifa;
1160
1161                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1162                                 continue;
1163                         if (ifa->ifa_dstaddr &&
1164                             sa_equal(addr, ifa->ifa_dstaddr))
1165                                 return (ifa);
1166                 }
1167         }
1168         return (NULL);
1169 }
1170
1171 /*
1172  * Find an interface on a specific network.  If many, choice
1173  * is most specific found.
1174  */
1175 struct ifaddr *
1176 ifa_ifwithnet(struct sockaddr *addr)
1177 {
1178         struct ifnet *ifp;
1179         struct ifaddr *ifa_maybe = NULL;
1180         u_int af = addr->sa_family;
1181         char *addr_data = addr->sa_data, *cplim;
1182
1183         /*
1184          * AF_LINK addresses can be looked up directly by their index number,
1185          * so do that if we can.
1186          */
1187         if (af == AF_LINK) {
1188                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1189
1190                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1191                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1192         }
1193
1194         /*
1195          * Scan though each interface, looking for ones that have
1196          * addresses in this address family.
1197          */
1198         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1199                 struct ifaddr_container *ifac;
1200
1201                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1202                         struct ifaddr *ifa = ifac->ifa;
1203                         char *cp, *cp2, *cp3;
1204
1205                         if (ifa->ifa_addr->sa_family != af)
1206 next:                           continue;
1207                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1208                                 /*
1209                                  * This is a bit broken as it doesn't
1210                                  * take into account that the remote end may
1211                                  * be a single node in the network we are
1212                                  * looking for.
1213                                  * The trouble is that we don't know the
1214                                  * netmask for the remote end.
1215                                  */
1216                                 if (ifa->ifa_dstaddr != NULL &&
1217                                     sa_equal(addr, ifa->ifa_dstaddr))
1218                                         return (ifa);
1219                         } else {
1220                                 /*
1221                                  * if we have a special address handler,
1222                                  * then use it instead of the generic one.
1223                                  */
1224                                 if (ifa->ifa_claim_addr) {
1225                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1226                                                 return (ifa);
1227                                         } else {
1228                                                 continue;
1229                                         }
1230                                 }
1231
1232                                 /*
1233                                  * Scan all the bits in the ifa's address.
1234                                  * If a bit dissagrees with what we are
1235                                  * looking for, mask it with the netmask
1236                                  * to see if it really matters.
1237                                  * (A byte at a time)
1238                                  */
1239                                 if (ifa->ifa_netmask == 0)
1240                                         continue;
1241                                 cp = addr_data;
1242                                 cp2 = ifa->ifa_addr->sa_data;
1243                                 cp3 = ifa->ifa_netmask->sa_data;
1244                                 cplim = ifa->ifa_netmask->sa_len +
1245                                         (char *)ifa->ifa_netmask;
1246                                 while (cp3 < cplim)
1247                                         if ((*cp++ ^ *cp2++) & *cp3++)
1248                                                 goto next; /* next address! */
1249                                 /*
1250                                  * If the netmask of what we just found
1251                                  * is more specific than what we had before
1252                                  * (if we had one) then remember the new one
1253                                  * before continuing to search
1254                                  * for an even better one.
1255                                  */
1256                                 if (ifa_maybe == NULL ||
1257                                     rn_refines((char *)ifa->ifa_netmask,
1258                                                (char *)ifa_maybe->ifa_netmask))
1259                                         ifa_maybe = ifa;
1260                         }
1261                 }
1262         }
1263         return (ifa_maybe);
1264 }
1265
1266 /*
1267  * Find an interface address specific to an interface best matching
1268  * a given address.
1269  */
1270 struct ifaddr *
1271 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1272 {
1273         struct ifaddr_container *ifac;
1274         char *cp, *cp2, *cp3;
1275         char *cplim;
1276         struct ifaddr *ifa_maybe = NULL;
1277         u_int af = addr->sa_family;
1278
1279         if (af >= AF_MAX)
1280                 return (0);
1281         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1282                 struct ifaddr *ifa = ifac->ifa;
1283
1284                 if (ifa->ifa_addr->sa_family != af)
1285                         continue;
1286                 if (ifa_maybe == NULL)
1287                         ifa_maybe = ifa;
1288                 if (ifa->ifa_netmask == NULL) {
1289                         if (sa_equal(addr, ifa->ifa_addr) ||
1290                             (ifa->ifa_dstaddr != NULL &&
1291                              sa_equal(addr, ifa->ifa_dstaddr)))
1292                                 return (ifa);
1293                         continue;
1294                 }
1295                 if (ifp->if_flags & IFF_POINTOPOINT) {
1296                         if (sa_equal(addr, ifa->ifa_dstaddr))
1297                                 return (ifa);
1298                 } else {
1299                         cp = addr->sa_data;
1300                         cp2 = ifa->ifa_addr->sa_data;
1301                         cp3 = ifa->ifa_netmask->sa_data;
1302                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1303                         for (; cp3 < cplim; cp3++)
1304                                 if ((*cp++ ^ *cp2++) & *cp3)
1305                                         break;
1306                         if (cp3 == cplim)
1307                                 return (ifa);
1308                 }
1309         }
1310         return (ifa_maybe);
1311 }
1312
1313 /*
1314  * Default action when installing a route with a Link Level gateway.
1315  * Lookup an appropriate real ifa to point to.
1316  * This should be moved to /sys/net/link.c eventually.
1317  */
1318 static void
1319 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1320 {
1321         struct ifaddr *ifa;
1322         struct sockaddr *dst;
1323         struct ifnet *ifp;
1324
1325         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1326             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1327                 return;
1328         ifa = ifaof_ifpforaddr(dst, ifp);
1329         if (ifa != NULL) {
1330                 IFAFREE(rt->rt_ifa);
1331                 IFAREF(ifa);
1332                 rt->rt_ifa = ifa;
1333                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1334                         ifa->ifa_rtrequest(cmd, rt, info);
1335         }
1336 }
1337
1338 /*
1339  * Mark an interface down and notify protocols of
1340  * the transition.
1341  * NOTE: must be called at splnet or eqivalent.
1342  */
1343 void
1344 if_unroute(struct ifnet *ifp, int flag, int fam)
1345 {
1346         struct ifaddr_container *ifac;
1347
1348         ifp->if_flags &= ~flag;
1349         getmicrotime(&ifp->if_lastchange);
1350         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1351                 struct ifaddr *ifa = ifac->ifa;
1352
1353                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1354                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1355         }
1356         ifq_purge_all(&ifp->if_snd);
1357         rt_ifmsg(ifp);
1358 }
1359
1360 /*
1361  * Mark an interface up and notify protocols of
1362  * the transition.
1363  * NOTE: must be called at splnet or eqivalent.
1364  */
1365 void
1366 if_route(struct ifnet *ifp, int flag, int fam)
1367 {
1368         struct ifaddr_container *ifac;
1369
1370         ifq_purge_all(&ifp->if_snd);
1371         ifp->if_flags |= flag;
1372         getmicrotime(&ifp->if_lastchange);
1373         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1374                 struct ifaddr *ifa = ifac->ifa;
1375
1376                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1377                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1378         }
1379         rt_ifmsg(ifp);
1380 #ifdef INET6
1381         in6_if_up(ifp);
1382 #endif
1383 }
1384
1385 /*
1386  * Mark an interface down and notify protocols of the transition.  An
1387  * interface going down is also considered to be a synchronizing event.
1388  * We must ensure that all packet processing related to the interface
1389  * has completed before we return so e.g. the caller can free the ifnet
1390  * structure that the mbufs may be referencing.
1391  *
1392  * NOTE: must be called at splnet or eqivalent.
1393  */
1394 void
1395 if_down(struct ifnet *ifp)
1396 {
1397         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1398         netmsg_service_sync();
1399 }
1400
1401 /*
1402  * Mark an interface up and notify protocols of
1403  * the transition.
1404  * NOTE: must be called at splnet or eqivalent.
1405  */
1406 void
1407 if_up(struct ifnet *ifp)
1408 {
1409         if_route(ifp, IFF_UP, AF_UNSPEC);
1410 }
1411
1412 /*
1413  * Process a link state change.
1414  * NOTE: must be called at splsoftnet or equivalent.
1415  */
1416 void
1417 if_link_state_change(struct ifnet *ifp)
1418 {
1419         int link_state = ifp->if_link_state;
1420
1421         rt_ifmsg(ifp);
1422         devctl_notify("IFNET", ifp->if_xname,
1423             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1424 }
1425
1426 /*
1427  * Handle interface watchdog timer routines.  Called
1428  * from softclock, we decrement timers (if set) and
1429  * call the appropriate interface routine on expiration.
1430  */
1431 static void
1432 if_slowtimo(void *arg)
1433 {
1434         struct ifnet *ifp;
1435
1436         crit_enter();
1437
1438         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1439                 if (if_stats_compat) {
1440                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1441                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1442                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1443                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1444                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1445                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1446                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1447                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1448                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1449                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1450                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1451                 }
1452
1453                 if (ifp->if_timer == 0 || --ifp->if_timer)
1454                         continue;
1455                 if (ifp->if_watchdog) {
1456                         if (ifnet_tryserialize_all(ifp)) {
1457                                 (*ifp->if_watchdog)(ifp);
1458                                 ifnet_deserialize_all(ifp);
1459                         } else {
1460                                 /* try again next timeout */
1461                                 ++ifp->if_timer;
1462                         }
1463                 }
1464         }
1465
1466         crit_exit();
1467
1468         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1469 }
1470
1471 /*
1472  * Map interface name to
1473  * interface structure pointer.
1474  */
1475 struct ifnet *
1476 ifunit(const char *name)
1477 {
1478         struct ifnet *ifp;
1479
1480         /*
1481          * Search all the interfaces for this name/number
1482          */
1483
1484         TAILQ_FOREACH(ifp, &ifnet, if_link) {
1485                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1486                         break;
1487         }
1488         return (ifp);
1489 }
1490
1491
1492 /*
1493  * Map interface name in a sockaddr_dl to
1494  * interface structure pointer.
1495  */
1496 struct ifnet *
1497 if_withname(struct sockaddr *sa)
1498 {
1499         char ifname[IFNAMSIZ+1];
1500         struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
1501
1502         if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
1503              (sdl->sdl_nlen > IFNAMSIZ) )
1504                 return NULL;
1505
1506         /*
1507          * ifunit wants a null-terminated name.  It may not be null-terminated
1508          * in the sockaddr.  We don't want to change the caller's sockaddr,
1509          * and there might not be room to put the trailing null anyway, so we
1510          * make a local copy that we know we can null terminate safely.
1511          */
1512
1513         bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
1514         ifname[sdl->sdl_nlen] = '\0';
1515         return ifunit(ifname);
1516 }
1517
1518
1519 /*
1520  * Interface ioctls.
1521  */
1522 int
1523 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1524 {
1525         struct ifnet *ifp;
1526         struct ifreq *ifr;
1527         struct ifstat *ifs;
1528         int error;
1529         short oif_flags;
1530         int new_flags;
1531 #ifdef COMPAT_43
1532         int ocmd;
1533 #endif
1534         size_t namelen, onamelen;
1535         char new_name[IFNAMSIZ];
1536         struct ifaddr *ifa;
1537         struct sockaddr_dl *sdl;
1538
1539         switch (cmd) {
1540         case SIOCGIFCONF:
1541         case OSIOCGIFCONF:
1542                 return (ifconf(cmd, data, cred));
1543         default:
1544                 break;
1545         }
1546
1547         ifr = (struct ifreq *)data;
1548
1549         switch (cmd) {
1550         case SIOCIFCREATE:
1551         case SIOCIFCREATE2:
1552                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1553                         return (error);
1554                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1555                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1556         case SIOCIFDESTROY:
1557                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1558                         return (error);
1559                 return (if_clone_destroy(ifr->ifr_name));
1560         case SIOCIFGCLONERS:
1561                 return (if_clone_list((struct if_clonereq *)data));
1562         default:
1563                 break;
1564         }
1565
1566         /*
1567          * Nominal ioctl through interface, lookup the ifp and obtain a
1568          * lock to serialize the ifconfig ioctl operation.
1569          */
1570         ifp = ifunit(ifr->ifr_name);
1571         if (ifp == NULL)
1572                 return (ENXIO);
1573         error = 0;
1574         mtx_lock(&ifp->if_ioctl_mtx);
1575
1576         switch (cmd) {
1577         case SIOCGIFINDEX:
1578                 ifr->ifr_index = ifp->if_index;
1579                 break;
1580
1581         case SIOCGIFFLAGS:
1582                 ifr->ifr_flags = ifp->if_flags;
1583                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1584                 break;
1585
1586         case SIOCGIFCAP:
1587                 ifr->ifr_reqcap = ifp->if_capabilities;
1588                 ifr->ifr_curcap = ifp->if_capenable;
1589                 break;
1590
1591         case SIOCGIFMETRIC:
1592                 ifr->ifr_metric = ifp->if_metric;
1593                 break;
1594
1595         case SIOCGIFMTU:
1596                 ifr->ifr_mtu = ifp->if_mtu;
1597                 break;
1598
1599         case SIOCGIFTSOLEN:
1600                 ifr->ifr_tsolen = ifp->if_tsolen;
1601                 break;
1602
1603         case SIOCGIFDATA:
1604                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
1605                                 sizeof(ifp->if_data));
1606                 break;
1607
1608         case SIOCGIFPHYS:
1609                 ifr->ifr_phys = ifp->if_physical;
1610                 break;
1611
1612         case SIOCGIFPOLLCPU:
1613                 ifr->ifr_pollcpu = -1;
1614                 break;
1615
1616         case SIOCSIFPOLLCPU:
1617                 break;
1618
1619         case SIOCSIFFLAGS:
1620                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1621                 if (error)
1622                         break;
1623                 new_flags = (ifr->ifr_flags & 0xffff) |
1624                     (ifr->ifr_flagshigh << 16);
1625                 if (ifp->if_flags & IFF_SMART) {
1626                         /* Smart drivers twiddle their own routes */
1627                 } else if (ifp->if_flags & IFF_UP &&
1628                     (new_flags & IFF_UP) == 0) {
1629                         crit_enter();
1630                         if_down(ifp);
1631                         crit_exit();
1632                 } else if (new_flags & IFF_UP &&
1633                     (ifp->if_flags & IFF_UP) == 0) {
1634                         crit_enter();
1635                         if_up(ifp);
1636                         crit_exit();
1637                 }
1638
1639 #ifdef IFPOLL_ENABLE
1640                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
1641                         if (new_flags & IFF_NPOLLING)
1642                                 ifpoll_register(ifp);
1643                         else
1644                                 ifpoll_deregister(ifp);
1645                 }
1646 #endif
1647
1648                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1649                         (new_flags &~ IFF_CANTCHANGE);
1650                 if (new_flags & IFF_PPROMISC) {
1651                         /* Permanently promiscuous mode requested */
1652                         ifp->if_flags |= IFF_PROMISC;
1653                 } else if (ifp->if_pcount == 0) {
1654                         ifp->if_flags &= ~IFF_PROMISC;
1655                 }
1656                 if (ifp->if_ioctl) {
1657                         ifnet_serialize_all(ifp);
1658                         ifp->if_ioctl(ifp, cmd, data, cred);
1659                         ifnet_deserialize_all(ifp);
1660                 }
1661                 getmicrotime(&ifp->if_lastchange);
1662                 break;
1663
1664         case SIOCSIFCAP:
1665                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1666                 if (error)
1667                         break;
1668                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
1669                         error = EINVAL;
1670                         break;
1671                 }
1672                 ifnet_serialize_all(ifp);
1673                 ifp->if_ioctl(ifp, cmd, data, cred);
1674                 ifnet_deserialize_all(ifp);
1675                 break;
1676
1677         case SIOCSIFNAME:
1678                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1679                 if (error)
1680                         break;
1681                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1682                 if (error)
1683                         break;
1684                 if (new_name[0] == '\0') {
1685                         error = EINVAL;
1686                         break;
1687                 }
1688                 if (ifunit(new_name) != NULL) {
1689                         error = EEXIST;
1690                         break;
1691                 }
1692
1693                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1694
1695                 /* Announce the departure of the interface. */
1696                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1697
1698                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1699                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1700                 /* XXX IFA_LOCK(ifa); */
1701                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1702                 namelen = strlen(new_name);
1703                 onamelen = sdl->sdl_nlen;
1704                 /*
1705                  * Move the address if needed.  This is safe because we
1706                  * allocate space for a name of length IFNAMSIZ when we
1707                  * create this in if_attach().
1708                  */
1709                 if (namelen != onamelen) {
1710                         bcopy(sdl->sdl_data + onamelen,
1711                             sdl->sdl_data + namelen, sdl->sdl_alen);
1712                 }
1713                 bcopy(new_name, sdl->sdl_data, namelen);
1714                 sdl->sdl_nlen = namelen;
1715                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1716                 bzero(sdl->sdl_data, onamelen);
1717                 while (namelen != 0)
1718                         sdl->sdl_data[--namelen] = 0xff;
1719                 /* XXX IFA_UNLOCK(ifa) */
1720
1721                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1722
1723                 /* Announce the return of the interface. */
1724                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1725                 break;
1726
1727         case SIOCSIFMETRIC:
1728                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1729                 if (error)
1730                         break;
1731                 ifp->if_metric = ifr->ifr_metric;
1732                 getmicrotime(&ifp->if_lastchange);
1733                 break;
1734
1735         case SIOCSIFPHYS:
1736                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1737                 if (error)
1738                         break;
1739                 if (ifp->if_ioctl == NULL) {
1740                         error = EOPNOTSUPP;
1741                         break;
1742                 }
1743                 ifnet_serialize_all(ifp);
1744                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1745                 ifnet_deserialize_all(ifp);
1746                 if (error == 0)
1747                         getmicrotime(&ifp->if_lastchange);
1748                 break;
1749
1750         case SIOCSIFMTU:
1751         {
1752                 u_long oldmtu = ifp->if_mtu;
1753
1754                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1755                 if (error)
1756                         break;
1757                 if (ifp->if_ioctl == NULL) {
1758                         error = EOPNOTSUPP;
1759                         break;
1760                 }
1761                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
1762                         error = EINVAL;
1763                         break;
1764                 }
1765                 ifnet_serialize_all(ifp);
1766                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1767                 ifnet_deserialize_all(ifp);
1768                 if (error == 0) {
1769                         getmicrotime(&ifp->if_lastchange);
1770                         rt_ifmsg(ifp);
1771                 }
1772                 /*
1773                  * If the link MTU changed, do network layer specific procedure.
1774                  */
1775                 if (ifp->if_mtu != oldmtu) {
1776 #ifdef INET6
1777                         nd6_setmtu(ifp);
1778 #endif
1779                 }
1780                 break;
1781         }
1782
1783         case SIOCSIFTSOLEN:
1784                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1785                 if (error)
1786                         break;
1787
1788                 /* XXX need driver supplied upper limit */
1789                 if (ifr->ifr_tsolen <= 0) {
1790                         error = EINVAL;
1791                         break;
1792                 }
1793                 ifp->if_tsolen = ifr->ifr_tsolen;
1794                 break;
1795
1796         case SIOCADDMULTI:
1797         case SIOCDELMULTI:
1798                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1799                 if (error)
1800                         break;
1801
1802                 /* Don't allow group membership on non-multicast interfaces. */
1803                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1804                         error = EOPNOTSUPP;
1805                         break;
1806                 }
1807
1808                 /* Don't let users screw up protocols' entries. */
1809                 if (ifr->ifr_addr.sa_family != AF_LINK) {
1810                         error = EINVAL;
1811                         break;
1812                 }
1813
1814                 if (cmd == SIOCADDMULTI) {
1815                         struct ifmultiaddr *ifma;
1816                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1817                 } else {
1818                         error = if_delmulti(ifp, &ifr->ifr_addr);
1819                 }
1820                 if (error == 0)
1821                         getmicrotime(&ifp->if_lastchange);
1822                 break;
1823
1824         case SIOCSIFPHYADDR:
1825         case SIOCDIFPHYADDR:
1826 #ifdef INET6
1827         case SIOCSIFPHYADDR_IN6:
1828 #endif
1829         case SIOCSLIFPHYADDR:
1830         case SIOCSIFMEDIA:
1831         case SIOCSIFGENERIC:
1832                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1833                 if (error)
1834                         break;
1835                 if (ifp->if_ioctl == 0) {
1836                         error = EOPNOTSUPP;
1837                         break;
1838                 }
1839                 ifnet_serialize_all(ifp);
1840                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1841                 ifnet_deserialize_all(ifp);
1842                 if (error == 0)
1843                         getmicrotime(&ifp->if_lastchange);
1844                 break;
1845
1846         case SIOCGIFSTATUS:
1847                 ifs = (struct ifstat *)data;
1848                 ifs->ascii[0] = '\0';
1849                 /* fall through */
1850         case SIOCGIFPSRCADDR:
1851         case SIOCGIFPDSTADDR:
1852         case SIOCGLIFPHYADDR:
1853         case SIOCGIFMEDIA:
1854         case SIOCGIFGENERIC:
1855                 if (ifp->if_ioctl == NULL) {
1856                         error = EOPNOTSUPP;
1857                         break;
1858                 }
1859                 ifnet_serialize_all(ifp);
1860                 error = ifp->if_ioctl(ifp, cmd, data, cred);
1861                 ifnet_deserialize_all(ifp);
1862                 break;
1863
1864         case SIOCSIFLLADDR:
1865                 error = priv_check_cred(cred, PRIV_ROOT, 0);
1866                 if (error)
1867                         break;
1868                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
1869                                      ifr->ifr_addr.sa_len);
1870                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
1871                 break;
1872
1873         default:
1874                 oif_flags = ifp->if_flags;
1875                 if (so->so_proto == 0) {
1876                         error = EOPNOTSUPP;
1877                         break;
1878                 }
1879 #ifndef COMPAT_43
1880                 error = so_pru_control_direct(so, cmd, data, ifp);
1881 #else
1882                 ocmd = cmd;
1883
1884                 switch (cmd) {
1885                 case SIOCSIFDSTADDR:
1886                 case SIOCSIFADDR:
1887                 case SIOCSIFBRDADDR:
1888                 case SIOCSIFNETMASK:
1889 #if BYTE_ORDER != BIG_ENDIAN
1890                         if (ifr->ifr_addr.sa_family == 0 &&
1891                             ifr->ifr_addr.sa_len < 16) {
1892                                 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1893                                 ifr->ifr_addr.sa_len = 16;
1894                         }
1895 #else
1896                         if (ifr->ifr_addr.sa_len == 0)
1897                                 ifr->ifr_addr.sa_len = 16;
1898 #endif
1899                         break;
1900                 case OSIOCGIFADDR:
1901                         cmd = SIOCGIFADDR;
1902                         break;
1903                 case OSIOCGIFDSTADDR:
1904                         cmd = SIOCGIFDSTADDR;
1905                         break;
1906                 case OSIOCGIFBRDADDR:
1907                         cmd = SIOCGIFBRDADDR;
1908                         break;
1909                 case OSIOCGIFNETMASK:
1910                         cmd = SIOCGIFNETMASK;
1911                         break;
1912                 default:
1913                         break;
1914                 }
1915
1916                 error = so_pru_control_direct(so, cmd, data, ifp);
1917
1918                 switch (ocmd) {
1919                 case OSIOCGIFADDR:
1920                 case OSIOCGIFDSTADDR:
1921                 case OSIOCGIFBRDADDR:
1922                 case OSIOCGIFNETMASK:
1923                         *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1924                         break;
1925                 }
1926 #endif /* COMPAT_43 */
1927
1928                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1929 #ifdef INET6
1930                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
1931                         if (ifp->if_flags & IFF_UP) {
1932                                 crit_enter();
1933                                 in6_if_up(ifp);
1934                                 crit_exit();
1935                         }
1936 #endif
1937                 }
1938                 break;
1939         }
1940
1941         mtx_unlock(&ifp->if_ioctl_mtx);
1942         return (error);
1943 }
1944
1945 /*
1946  * Set/clear promiscuous mode on interface ifp based on the truth value
1947  * of pswitch.  The calls are reference counted so that only the first
1948  * "on" request actually has an effect, as does the final "off" request.
1949  * Results are undefined if the "off" and "on" requests are not matched.
1950  */
1951 int
1952 ifpromisc(struct ifnet *ifp, int pswitch)
1953 {
1954         struct ifreq ifr;
1955         int error;
1956         int oldflags;
1957
1958         oldflags = ifp->if_flags;
1959         if (ifp->if_flags & IFF_PPROMISC) {
1960                 /* Do nothing if device is in permanently promiscuous mode */
1961                 ifp->if_pcount += pswitch ? 1 : -1;
1962                 return (0);
1963         }
1964         if (pswitch) {
1965                 /*
1966                  * If the device is not configured up, we cannot put it in
1967                  * promiscuous mode.
1968                  */
1969                 if ((ifp->if_flags & IFF_UP) == 0)
1970                         return (ENETDOWN);
1971                 if (ifp->if_pcount++ != 0)
1972                         return (0);
1973                 ifp->if_flags |= IFF_PROMISC;
1974                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
1975                     ifp->if_xname);
1976         } else {
1977                 if (--ifp->if_pcount > 0)
1978                         return (0);
1979                 ifp->if_flags &= ~IFF_PROMISC;
1980                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
1981                     ifp->if_xname);
1982         }
1983         ifr.ifr_flags = ifp->if_flags;
1984         ifr.ifr_flagshigh = ifp->if_flags >> 16;
1985         ifnet_serialize_all(ifp);
1986         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
1987         ifnet_deserialize_all(ifp);
1988         if (error == 0)
1989                 rt_ifmsg(ifp);
1990         else
1991                 ifp->if_flags = oldflags;
1992         return error;
1993 }
1994
1995 /*
1996  * Return interface configuration
1997  * of system.  List may be used
1998  * in later ioctl's (above) to get
1999  * other information.
2000  */
2001 static int
2002 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2003 {
2004         struct ifconf *ifc = (struct ifconf *)data;
2005         struct ifnet *ifp;
2006         struct sockaddr *sa;
2007         struct ifreq ifr, *ifrp;
2008         int space = ifc->ifc_len, error = 0;
2009
2010         ifrp = ifc->ifc_req;
2011         TAILQ_FOREACH(ifp, &ifnet, if_link) {
2012                 struct ifaddr_container *ifac;
2013                 int addrs;
2014
2015                 if (space <= sizeof ifr)
2016                         break;
2017
2018                 /*
2019                  * Zero the stack declared structure first to prevent
2020                  * memory disclosure.
2021                  */
2022                 bzero(&ifr, sizeof(ifr));
2023                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2024                     >= sizeof(ifr.ifr_name)) {
2025                         error = ENAMETOOLONG;
2026                         break;
2027                 }
2028
2029                 addrs = 0;
2030                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2031                         struct ifaddr *ifa = ifac->ifa;
2032
2033                         if (space <= sizeof ifr)
2034                                 break;
2035                         sa = ifa->ifa_addr;
2036                         if (cred->cr_prison &&
2037                             prison_if(cred, sa))
2038                                 continue;
2039                         addrs++;
2040 #ifdef COMPAT_43
2041                         if (cmd == OSIOCGIFCONF) {
2042                                 struct osockaddr *osa =
2043                                          (struct osockaddr *)&ifr.ifr_addr;
2044                                 ifr.ifr_addr = *sa;
2045                                 osa->sa_family = sa->sa_family;
2046                                 error = copyout(&ifr, ifrp, sizeof ifr);
2047                                 ifrp++;
2048                         } else
2049 #endif
2050                         if (sa->sa_len <= sizeof(*sa)) {
2051                                 ifr.ifr_addr = *sa;
2052                                 error = copyout(&ifr, ifrp, sizeof ifr);
2053                                 ifrp++;
2054                         } else {
2055                                 if (space < (sizeof ifr) + sa->sa_len -
2056                                             sizeof(*sa))
2057                                         break;
2058                                 space -= sa->sa_len - sizeof(*sa);
2059                                 error = copyout(&ifr, ifrp,
2060                                                 sizeof ifr.ifr_name);
2061                                 if (error == 0)
2062                                         error = copyout(sa, &ifrp->ifr_addr,
2063                                                         sa->sa_len);
2064                                 ifrp = (struct ifreq *)
2065                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2066                         }
2067                         if (error)
2068                                 break;
2069                         space -= sizeof ifr;
2070                 }
2071                 if (error)
2072                         break;
2073                 if (!addrs) {
2074                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2075                         error = copyout(&ifr, ifrp, sizeof ifr);
2076                         if (error)
2077                                 break;
2078                         space -= sizeof ifr;
2079                         ifrp++;
2080                 }
2081         }
2082         ifc->ifc_len -= space;
2083         return (error);
2084 }
2085
2086 /*
2087  * Just like if_promisc(), but for all-multicast-reception mode.
2088  */
2089 int
2090 if_allmulti(struct ifnet *ifp, int onswitch)
2091 {
2092         int error = 0;
2093         struct ifreq ifr;
2094
2095         crit_enter();
2096
2097         if (onswitch) {
2098                 if (ifp->if_amcount++ == 0) {
2099                         ifp->if_flags |= IFF_ALLMULTI;
2100                         ifr.ifr_flags = ifp->if_flags;
2101                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2102                         ifnet_serialize_all(ifp);
2103                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2104                                               NULL);
2105                         ifnet_deserialize_all(ifp);
2106                 }
2107         } else {
2108                 if (ifp->if_amcount > 1) {
2109                         ifp->if_amcount--;
2110                 } else {
2111                         ifp->if_amcount = 0;
2112                         ifp->if_flags &= ~IFF_ALLMULTI;
2113                         ifr.ifr_flags = ifp->if_flags;
2114                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2115                         ifnet_serialize_all(ifp);
2116                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2117                                               NULL);
2118                         ifnet_deserialize_all(ifp);
2119                 }
2120         }
2121
2122         crit_exit();
2123
2124         if (error == 0)
2125                 rt_ifmsg(ifp);
2126         return error;
2127 }
2128
2129 /*
2130  * Add a multicast listenership to the interface in question.
2131  * The link layer provides a routine which converts
2132  */
2133 int
2134 if_addmulti(
2135         struct ifnet *ifp,      /* interface to manipulate */
2136         struct sockaddr *sa,    /* address to add */
2137         struct ifmultiaddr **retifma)
2138 {
2139         struct sockaddr *llsa, *dupsa;
2140         int error;
2141         struct ifmultiaddr *ifma;
2142
2143         /*
2144          * If the matching multicast address already exists
2145          * then don't add a new one, just add a reference
2146          */
2147         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2148                 if (sa_equal(sa, ifma->ifma_addr)) {
2149                         ifma->ifma_refcount++;
2150                         if (retifma)
2151                                 *retifma = ifma;
2152                         return 0;
2153                 }
2154         }
2155
2156         /*
2157          * Give the link layer a chance to accept/reject it, and also
2158          * find out which AF_LINK address this maps to, if it isn't one
2159          * already.
2160          */
2161         if (ifp->if_resolvemulti) {
2162                 ifnet_serialize_all(ifp);
2163                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2164                 ifnet_deserialize_all(ifp);
2165                 if (error)
2166                         return error;
2167         } else {
2168                 llsa = NULL;
2169         }
2170
2171         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2172         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_WAITOK);
2173         bcopy(sa, dupsa, sa->sa_len);
2174
2175         ifma->ifma_addr = dupsa;
2176         ifma->ifma_lladdr = llsa;
2177         ifma->ifma_ifp = ifp;
2178         ifma->ifma_refcount = 1;
2179         ifma->ifma_protospec = 0;
2180         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2181
2182         /*
2183          * Some network interfaces can scan the address list at
2184          * interrupt time; lock them out.
2185          */
2186         crit_enter();
2187         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2188         crit_exit();
2189         if (retifma)
2190                 *retifma = ifma;
2191
2192         if (llsa != NULL) {
2193                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2194                         if (sa_equal(ifma->ifma_addr, llsa))
2195                                 break;
2196                 }
2197                 if (ifma) {
2198                         ifma->ifma_refcount++;
2199                 } else {
2200                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_WAITOK);
2201                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_WAITOK);
2202                         bcopy(llsa, dupsa, llsa->sa_len);
2203                         ifma->ifma_addr = dupsa;
2204                         ifma->ifma_ifp = ifp;
2205                         ifma->ifma_refcount = 1;
2206                         crit_enter();
2207                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2208                         crit_exit();
2209                 }
2210         }
2211         /*
2212          * We are certain we have added something, so call down to the
2213          * interface to let them know about it.
2214          */
2215         crit_enter();
2216         ifnet_serialize_all(ifp);
2217         if (ifp->if_ioctl)
2218                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2219         ifnet_deserialize_all(ifp);
2220         crit_exit();
2221
2222         return 0;
2223 }
2224
2225 /*
2226  * Remove a reference to a multicast address on this interface.  Yell
2227  * if the request does not match an existing membership.
2228  */
2229 int
2230 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2231 {
2232         struct ifmultiaddr *ifma;
2233
2234         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2235                 if (sa_equal(sa, ifma->ifma_addr))
2236                         break;
2237         if (ifma == NULL)
2238                 return ENOENT;
2239
2240         if (ifma->ifma_refcount > 1) {
2241                 ifma->ifma_refcount--;
2242                 return 0;
2243         }
2244
2245         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2246         sa = ifma->ifma_lladdr;
2247         crit_enter();
2248         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2249         /*
2250          * Make sure the interface driver is notified
2251          * in the case of a link layer mcast group being left.
2252          */
2253         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL) {
2254                 ifnet_serialize_all(ifp);
2255                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2256                 ifnet_deserialize_all(ifp);
2257         }
2258         crit_exit();
2259         kfree(ifma->ifma_addr, M_IFMADDR);
2260         kfree(ifma, M_IFMADDR);
2261         if (sa == NULL)
2262                 return 0;
2263
2264         /*
2265          * Now look for the link-layer address which corresponds to
2266          * this network address.  It had been squirreled away in
2267          * ifma->ifma_lladdr for this purpose (so we don't have
2268          * to call ifp->if_resolvemulti() again), and we saved that
2269          * value in sa above.  If some nasty deleted the
2270          * link-layer address out from underneath us, we can deal because
2271          * the address we stored was is not the same as the one which was
2272          * in the record for the link-layer address.  (So we don't complain
2273          * in that case.)
2274          */
2275         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2276                 if (sa_equal(sa, ifma->ifma_addr))
2277                         break;
2278         if (ifma == NULL)
2279                 return 0;
2280
2281         if (ifma->ifma_refcount > 1) {
2282                 ifma->ifma_refcount--;
2283                 return 0;
2284         }
2285
2286         crit_enter();
2287         ifnet_serialize_all(ifp);
2288         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2289         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2290         ifnet_deserialize_all(ifp);
2291         crit_exit();
2292         kfree(ifma->ifma_addr, M_IFMADDR);
2293         kfree(sa, M_IFMADDR);
2294         kfree(ifma, M_IFMADDR);
2295
2296         return 0;
2297 }
2298
2299 /*
2300  * Delete all multicast group membership for an interface.
2301  * Should be used to quickly flush all multicast filters.
2302  */
2303 void
2304 if_delallmulti(struct ifnet *ifp)
2305 {
2306         struct ifmultiaddr *ifma;
2307         struct ifmultiaddr *next;
2308
2309         TAILQ_FOREACH_MUTABLE(ifma, &ifp->if_multiaddrs, ifma_link, next)
2310                 if_delmulti(ifp, ifma->ifma_addr);
2311 }
2312
2313
2314 /*
2315  * Set the link layer address on an interface.
2316  *
2317  * At this time we only support certain types of interfaces,
2318  * and we don't allow the length of the address to change.
2319  */
2320 int
2321 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2322 {
2323         struct sockaddr_dl *sdl;
2324         struct ifreq ifr;
2325
2326         sdl = IF_LLSOCKADDR(ifp);
2327         if (sdl == NULL)
2328                 return (EINVAL);
2329         if (len != sdl->sdl_alen)       /* don't allow length to change */
2330                 return (EINVAL);
2331         switch (ifp->if_type) {
2332         case IFT_ETHER:                 /* these types use struct arpcom */
2333         case IFT_XETHER:
2334         case IFT_L2VLAN:
2335                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2336                 bcopy(lladdr, LLADDR(sdl), len);
2337                 break;
2338         default:
2339                 return (ENODEV);
2340         }
2341         /*
2342          * If the interface is already up, we need
2343          * to re-init it in order to reprogram its
2344          * address filter.
2345          */
2346         ifnet_serialize_all(ifp);
2347         if ((ifp->if_flags & IFF_UP) != 0) {
2348 #ifdef INET
2349                 struct ifaddr_container *ifac;
2350 #endif
2351
2352                 ifp->if_flags &= ~IFF_UP;
2353                 ifr.ifr_flags = ifp->if_flags;
2354                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2355                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2356                               NULL);
2357                 ifp->if_flags |= IFF_UP;
2358                 ifr.ifr_flags = ifp->if_flags;
2359                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2360                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2361                                  NULL);
2362 #ifdef INET
2363                 /*
2364                  * Also send gratuitous ARPs to notify other nodes about
2365                  * the address change.
2366                  */
2367                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2368                         struct ifaddr *ifa = ifac->ifa;
2369
2370                         if (ifa->ifa_addr != NULL &&
2371                             ifa->ifa_addr->sa_family == AF_INET)
2372                                 arp_gratuitous(ifp, ifa);
2373                 }
2374 #endif
2375         }
2376         ifnet_deserialize_all(ifp);
2377         return (0);
2378 }
2379
2380 struct ifmultiaddr *
2381 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2382 {
2383         struct ifmultiaddr *ifma;
2384
2385         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2386                 if (sa_equal(ifma->ifma_addr, sa))
2387                         break;
2388
2389         return ifma;
2390 }
2391
2392 /*
2393  * This function locates the first real ethernet MAC from a network
2394  * card and loads it into node, returning 0 on success or ENOENT if
2395  * no suitable interfaces were found.  It is used by the uuid code to
2396  * generate a unique 6-byte number.
2397  */
2398 int
2399 if_getanyethermac(uint16_t *node, int minlen)
2400 {
2401         struct ifnet *ifp;
2402         struct sockaddr_dl *sdl;
2403
2404         TAILQ_FOREACH(ifp, &ifnet, if_link) {
2405                 if (ifp->if_type != IFT_ETHER)
2406                         continue;
2407                 sdl = IF_LLSOCKADDR(ifp);
2408                 if (sdl->sdl_alen < minlen)
2409                         continue;
2410                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2411                       minlen);
2412                 return(0);
2413         }
2414         return (ENOENT);
2415 }
2416
2417 /*
2418  * The name argument must be a pointer to storage which will last as
2419  * long as the interface does.  For physical devices, the result of
2420  * device_get_name(dev) is a good choice and for pseudo-devices a
2421  * static string works well.
2422  */
2423 void
2424 if_initname(struct ifnet *ifp, const char *name, int unit)
2425 {
2426         ifp->if_dname = name;
2427         ifp->if_dunit = unit;
2428         if (unit != IF_DUNIT_NONE)
2429                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2430         else
2431                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2432 }
2433
2434 int
2435 if_printf(struct ifnet *ifp, const char *fmt, ...)
2436 {
2437         __va_list ap;
2438         int retval;
2439
2440         retval = kprintf("%s: ", ifp->if_xname);
2441         __va_start(ap, fmt);
2442         retval += kvprintf(fmt, ap);
2443         __va_end(ap);
2444         return (retval);
2445 }
2446
2447 struct ifnet *
2448 if_alloc(uint8_t type)
2449 {
2450         struct ifnet *ifp;
2451         size_t size;
2452
2453         /*
2454          * XXX temporary hack until arpcom is setup in if_l2com
2455          */
2456         if (type == IFT_ETHER)
2457                 size = sizeof(struct arpcom);
2458         else
2459                 size = sizeof(struct ifnet);
2460
2461         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2462
2463         ifp->if_type = type;
2464
2465         if (if_com_alloc[type] != NULL) {
2466                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2467                 if (ifp->if_l2com == NULL) {
2468                         kfree(ifp, M_IFNET);
2469                         return (NULL);
2470                 }
2471         }
2472         return (ifp);
2473 }
2474
2475 void
2476 if_free(struct ifnet *ifp)
2477 {
2478         kfree(ifp, M_IFNET);
2479 }
2480
2481 void
2482 ifq_set_classic(struct ifaltq *ifq)
2483 {
2484         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2485             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2486 }
2487
2488 void
2489 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2490     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2491 {
2492         int q;
2493
2494         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2495         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2496         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2497         KASSERT(request != NULL, ("request is not specified"));
2498
2499         ifq->altq_mapsubq = mapsubq;
2500         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2501                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2502
2503                 ifsq->ifsq_enqueue = enqueue;
2504                 ifsq->ifsq_dequeue = dequeue;
2505                 ifsq->ifsq_request = request;
2506         }
2507 }
2508
2509 int
2510 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2511     struct altq_pktattr *pa __unused)
2512 {
2513         if (IF_QFULL(ifsq)) {
2514                 m_freem(m);
2515                 return(ENOBUFS);
2516         } else {
2517                 IF_ENQUEUE(ifsq, m);
2518                 return(0);
2519         }
2520 }
2521
2522 struct mbuf *
2523 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, struct mbuf *mpolled, int op)
2524 {
2525         struct mbuf *m;
2526
2527         switch (op) {
2528         case ALTDQ_POLL:
2529                 IF_POLL(ifsq, m);
2530                 break;
2531         case ALTDQ_REMOVE:
2532                 IF_DEQUEUE(ifsq, m);
2533                 break;
2534         default:
2535                 panic("unsupported ALTQ dequeue op: %d", op);
2536         }
2537         KKASSERT(mpolled == NULL || mpolled == m);
2538         return(m);
2539 }
2540
2541 int
2542 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
2543 {
2544         switch (req) {
2545         case ALTRQ_PURGE:
2546                 IF_DRAIN(ifsq);
2547                 break;
2548         default:
2549                 panic("unsupported ALTQ request: %d", req);
2550         }
2551         return(0);
2552 }
2553
2554 static void
2555 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
2556 {
2557         struct ifnet *ifp = ifsq_get_ifp(ifsq);
2558         int running = 0, need_sched;
2559
2560         /*
2561          * Try to do direct ifnet.if_start first, if there is
2562          * contention on ifnet's serializer, ifnet.if_start will
2563          * be scheduled on ifnet's CPU.
2564          */
2565         if (!ifsq_tryserialize_hw(ifsq)) {
2566                 /*
2567                  * ifnet serializer contention happened,
2568                  * ifnet.if_start is scheduled on ifnet's
2569                  * CPU, and we keep going.
2570                  */
2571                 ifsq_ifstart_schedule(ifsq, 1);
2572                 return;
2573         }
2574
2575         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
2576                 ifp->if_start(ifp, ifsq);
2577                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
2578                         running = 1;
2579         }
2580         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
2581
2582         ifsq_deserialize_hw(ifsq);
2583
2584         if (need_sched) {
2585                 /*
2586                  * More data need to be transmitted, ifnet.if_start is
2587                  * scheduled on ifnet's CPU, and we keep going.
2588                  * NOTE: ifnet.if_start interlock is not released.
2589                  */
2590                 ifsq_ifstart_schedule(ifsq, force_sched);
2591         }
2592 }
2593
2594 /*
2595  * IFSUBQ packets staging mechanism:
2596  *
2597  * The packets enqueued into IFSUBQ are staged to a certain amount before the
2598  * ifnet's if_start is called.  In this way, the driver could avoid writing
2599  * to hardware registers upon every packet, instead, hardware registers
2600  * could be written when certain amount of packets are put onto hardware
2601  * TX ring.  The measurement on several modern NICs (emx(4), igb(4), bnx(4),
2602  * bge(4), jme(4)) shows that the hardware registers writing aggregation
2603  * could save ~20% CPU time when 18bytes UDP datagrams are transmitted at
2604  * 1.48Mpps.  The performance improvement by hardware registers writing
2605  * aggeregation is also mentioned by Luigi Rizzo's netmap paper
2606  * (http://info.iet.unipi.it/~luigi/netmap/).
2607  *
2608  * IFSUBQ packets staging is performed for two entry points into drivers's
2609  * transmission function:
2610  * - Direct ifnet's if_start calling, i.e. ifsq_ifstart_try()
2611  * - ifnet's if_start scheduling, i.e. ifsq_ifstart_schedule()
2612  *
2613  * IFSUBQ packets staging will be stopped upon any of the following conditions:
2614  * - If the count of packets enqueued on the current CPU is great than or
2615  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
2616  * - If the total length of packets enqueued on the current CPU is great
2617  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
2618  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
2619  *   is usually less than hardware's MTU.
2620  * - ifsq_ifstart_schedule() is not pending on the current CPU and if_start
2621  *   interlock (if_snd.altq_started) is not released.
2622  * - The if_start_rollup(), which is registered as low priority netisr
2623  *   rollup function, is called; probably because no more work is pending
2624  *   for netisr.
2625  *
2626  * NOTE:
2627  * Currently IFSUBQ packet staging is only performed in netisr threads.
2628  */
2629 int
2630 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
2631 {
2632         struct ifaltq *ifq = &ifp->if_snd;
2633         struct ifaltq_subque *ifsq;
2634         int error, start = 0, len, mcast = 0, avoid_start = 0;
2635         struct ifsubq_stage_head *head = NULL;
2636         struct ifsubq_stage *stage = NULL;
2637
2638         ifsq = ifq_map_subq(ifq, mycpuid);
2639         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
2640
2641         len = m->m_pkthdr.len;
2642         if (m->m_flags & M_MCAST)
2643                 mcast = 1;
2644
2645         if (curthread->td_type == TD_TYPE_NETISR) {
2646                 head = &ifsubq_stage_heads[mycpuid];
2647                 stage = ifsq_get_stage(ifsq, mycpuid);
2648
2649                 stage->stg_cnt++;
2650                 stage->stg_len += len;
2651                 if (stage->stg_cnt < ifsq_stage_cntmax &&
2652                     stage->stg_len < (ifp->if_mtu - max_protohdr))
2653                         avoid_start = 1;
2654         }
2655
2656         ALTQ_SQ_LOCK(ifsq);
2657         error = ifsq_enqueue_locked(ifsq, m, pa);
2658         if (error) {
2659                 if (!ifsq_data_ready(ifsq)) {
2660                         ALTQ_SQ_UNLOCK(ifsq);
2661                         return error;
2662                 }
2663                 avoid_start = 0;
2664         }
2665         if (!ifsq_is_started(ifsq)) {
2666                 if (avoid_start) {
2667                         ALTQ_SQ_UNLOCK(ifsq);
2668
2669                         KKASSERT(!error);
2670                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
2671                                 ifsq_stage_insert(head, stage);
2672
2673                         IFNET_STAT_INC(ifp, obytes, len);
2674                         if (mcast)
2675                                 IFNET_STAT_INC(ifp, omcasts, 1);
2676                         return error;
2677                 }
2678
2679                 /*
2680                  * Hold the interlock of ifnet.if_start
2681                  */
2682                 ifsq_set_started(ifsq);
2683                 start = 1;
2684         }
2685         ALTQ_SQ_UNLOCK(ifsq);
2686
2687         if (!error) {
2688                 IFNET_STAT_INC(ifp, obytes, len);
2689                 if (mcast)
2690                         IFNET_STAT_INC(ifp, omcasts, 1);
2691         }
2692
2693         if (stage != NULL) {
2694                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
2695                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
2696                         if (!avoid_start) {
2697                                 ifsq_stage_remove(head, stage);
2698                                 ifsq_ifstart_schedule(ifsq, 1);
2699                         }
2700                         return error;
2701                 }
2702
2703                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
2704                         ifsq_stage_remove(head, stage);
2705                 } else {
2706                         stage->stg_cnt = 0;
2707                         stage->stg_len = 0;
2708                 }
2709         }
2710
2711         if (!start)
2712                 return error;
2713
2714         ifsq_ifstart_try(ifsq, 0);
2715         return error;
2716 }
2717
2718 void *
2719 ifa_create(int size, int flags)
2720 {
2721         struct ifaddr *ifa;
2722         int i;
2723
2724         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
2725
2726         ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
2727         if (ifa == NULL)
2728                 return NULL;
2729
2730         ifa->ifa_containers =
2731             kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
2732                 M_IFADDR, M_WAITOK | M_ZERO);
2733         ifa->ifa_ncnt = ncpus;
2734         for (i = 0; i < ncpus; ++i) {
2735                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
2736
2737                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
2738                 ifac->ifa = ifa;
2739                 ifac->ifa_refcnt = 1;
2740         }
2741 #ifdef IFADDR_DEBUG
2742         kprintf("alloc ifa %p %d\n", ifa, size);
2743 #endif
2744         return ifa;
2745 }
2746
2747 void
2748 ifac_free(struct ifaddr_container *ifac, int cpu_id)
2749 {
2750         struct ifaddr *ifa = ifac->ifa;
2751
2752         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
2753         KKASSERT(ifac->ifa_refcnt == 0);
2754         KASSERT(ifac->ifa_listmask == 0,
2755                 ("ifa is still on %#x lists", ifac->ifa_listmask));
2756
2757         ifac->ifa_magic = IFA_CONTAINER_DEAD;
2758
2759 #ifdef IFADDR_DEBUG_VERBOSE
2760         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
2761 #endif
2762
2763         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
2764                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
2765         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
2766 #ifdef IFADDR_DEBUG
2767                 kprintf("free ifa %p\n", ifa);
2768 #endif
2769                 kfree(ifa->ifa_containers, M_IFADDR);
2770                 kfree(ifa, M_IFADDR);
2771         }
2772 }
2773
2774 static void
2775 ifa_iflink_dispatch(netmsg_t nmsg)
2776 {
2777         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2778         struct ifaddr *ifa = msg->ifa;
2779         struct ifnet *ifp = msg->ifp;
2780         int cpu = mycpuid;
2781         struct ifaddr_container *ifac;
2782
2783         crit_enter();
2784
2785         ifac = &ifa->ifa_containers[cpu];
2786         ASSERT_IFAC_VALID(ifac);
2787         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
2788                 ("ifaddr is on if_addrheads"));
2789
2790         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
2791         if (msg->tail)
2792                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
2793         else
2794                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
2795
2796         crit_exit();
2797
2798         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2799 }
2800
2801 void
2802 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
2803 {
2804         struct netmsg_ifaddr msg;
2805
2806         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2807                     0, ifa_iflink_dispatch);
2808         msg.ifa = ifa;
2809         msg.ifp = ifp;
2810         msg.tail = tail;
2811
2812         ifa_domsg(&msg.base.lmsg, 0);
2813 }
2814
2815 static void
2816 ifa_ifunlink_dispatch(netmsg_t nmsg)
2817 {
2818         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2819         struct ifaddr *ifa = msg->ifa;
2820         struct ifnet *ifp = msg->ifp;
2821         int cpu = mycpuid;
2822         struct ifaddr_container *ifac;
2823
2824         crit_enter();
2825
2826         ifac = &ifa->ifa_containers[cpu];
2827         ASSERT_IFAC_VALID(ifac);
2828         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
2829                 ("ifaddr is not on if_addrhead"));
2830
2831         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
2832         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
2833
2834         crit_exit();
2835
2836         ifa_forwardmsg(&nmsg->lmsg, cpu + 1);
2837 }
2838
2839 void
2840 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
2841 {
2842         struct netmsg_ifaddr msg;
2843
2844         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2845                     0, ifa_ifunlink_dispatch);
2846         msg.ifa = ifa;
2847         msg.ifp = ifp;
2848
2849         ifa_domsg(&msg.base.lmsg, 0);
2850 }
2851
2852 static void
2853 ifa_destroy_dispatch(netmsg_t nmsg)
2854 {
2855         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
2856
2857         IFAFREE(msg->ifa);
2858         ifa_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2859 }
2860
2861 void
2862 ifa_destroy(struct ifaddr *ifa)
2863 {
2864         struct netmsg_ifaddr msg;
2865
2866         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
2867                     0, ifa_destroy_dispatch);
2868         msg.ifa = ifa;
2869
2870         ifa_domsg(&msg.base.lmsg, 0);
2871 }
2872
2873 struct lwkt_port *
2874 ifnet_portfn(int cpu)
2875 {
2876         return &ifnet_threads[cpu].td_msgport;
2877 }
2878
2879 void
2880 ifnet_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
2881 {
2882         KKASSERT(next_cpu > mycpuid && next_cpu <= ncpus);
2883
2884         if (next_cpu < ncpus)
2885                 lwkt_forwardmsg(ifnet_portfn(next_cpu), lmsg);
2886         else
2887                 lwkt_replymsg(lmsg, 0);
2888 }
2889
2890 int
2891 ifnet_domsg(struct lwkt_msg *lmsg, int cpu)
2892 {
2893         KKASSERT(cpu < ncpus);
2894         return lwkt_domsg(ifnet_portfn(cpu), lmsg, 0);
2895 }
2896
2897 void
2898 ifnet_sendmsg(struct lwkt_msg *lmsg, int cpu)
2899 {
2900         KKASSERT(cpu < ncpus);
2901         lwkt_sendmsg(ifnet_portfn(cpu), lmsg);
2902 }
2903
2904 /*
2905  * Generic netmsg service loop.  Some protocols may roll their own but all
2906  * must do the basic command dispatch function call done here.
2907  */
2908 static void
2909 ifnet_service_loop(void *arg __unused)
2910 {
2911         netmsg_t msg;
2912
2913         while ((msg = lwkt_waitport(&curthread->td_msgport, 0))) {
2914                 KASSERT(msg->base.nm_dispatch, ("ifnet_service: badmsg"));
2915                 msg->base.nm_dispatch(msg);
2916         }
2917 }
2918
2919 static void
2920 if_start_rollup(void)
2921 {
2922         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
2923         struct ifsubq_stage *stage;
2924
2925         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
2926                 struct ifaltq_subque *ifsq = stage->stg_subq;
2927                 int is_sched = 0;
2928
2929                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
2930                         is_sched = 1;
2931                 ifsq_stage_remove(head, stage);
2932
2933                 if (is_sched) {
2934                         ifsq_ifstart_schedule(ifsq, 1);
2935                 } else {
2936                         int start = 0;
2937
2938                         ALTQ_SQ_LOCK(ifsq);
2939                         if (!ifsq_is_started(ifsq)) {
2940                                 /*
2941                                  * Hold the interlock of ifnet.if_start
2942                                  */
2943                                 ifsq_set_started(ifsq);
2944                                 start = 1;
2945                         }
2946                         ALTQ_SQ_UNLOCK(ifsq);
2947
2948                         if (start)
2949                                 ifsq_ifstart_try(ifsq, 1);
2950                 }
2951                 KKASSERT((stage->stg_flags &
2952                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
2953         }
2954 }
2955
2956 static void
2957 ifnetinit(void *dummy __unused)
2958 {
2959         int i;
2960
2961         for (i = 0; i < ncpus; ++i) {
2962                 struct thread *thr = &ifnet_threads[i];
2963
2964                 lwkt_create(ifnet_service_loop, NULL, NULL,
2965                             thr, TDF_NOSTART|TDF_FORCE_SPINPORT,
2966                             i, "ifnet %d", i);
2967                 netmsg_service_port_init(&thr->td_msgport);
2968                 lwkt_schedule(thr);
2969         }
2970
2971         for (i = 0; i < ncpus; ++i)
2972                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
2973         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
2974 }
2975
2976 struct ifnet *
2977 ifnet_byindex(unsigned short idx)
2978 {
2979         if (idx > if_index)
2980                 return NULL;
2981         return ifindex2ifnet[idx];
2982 }
2983
2984 struct ifaddr *
2985 ifaddr_byindex(unsigned short idx)
2986 {
2987         struct ifnet *ifp;
2988
2989         ifp = ifnet_byindex(idx);
2990         if (!ifp)
2991                 return NULL;
2992         return TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
2993 }
2994
2995 void
2996 if_register_com_alloc(u_char type,
2997     if_com_alloc_t *a, if_com_free_t *f)
2998 {
2999
3000         KASSERT(if_com_alloc[type] == NULL,
3001             ("if_register_com_alloc: %d already registered", type));
3002         KASSERT(if_com_free[type] == NULL,
3003             ("if_register_com_alloc: %d free already registered", type));
3004
3005         if_com_alloc[type] = a;
3006         if_com_free[type] = f;
3007 }
3008
3009 void
3010 if_deregister_com_alloc(u_char type)
3011 {
3012
3013         KASSERT(if_com_alloc[type] != NULL,
3014             ("if_deregister_com_alloc: %d not registered", type));
3015         KASSERT(if_com_free[type] != NULL,
3016             ("if_deregister_com_alloc: %d free not registered", type));
3017         if_com_alloc[type] = NULL;
3018         if_com_free[type] = NULL;
3019 }
3020
3021 int
3022 if_ring_count2(int cnt, int cnt_max)
3023 {
3024         int shift = 0;
3025
3026         KASSERT(cnt_max >= 1 && powerof2(cnt_max),
3027             ("invalid ring count max %d", cnt_max));
3028
3029         if (cnt <= 0)
3030                 cnt = cnt_max;
3031         if (cnt > ncpus2)
3032                 cnt = ncpus2;
3033         if (cnt > cnt_max)
3034                 cnt = cnt_max;
3035
3036         while ((1 << (shift + 1)) <= cnt)
3037                 ++shift;
3038         cnt = 1 << shift;
3039
3040         KASSERT(cnt >= 1 && cnt <= ncpus2 && cnt <= cnt_max,
3041             ("calculate cnt %d, ncpus2 %d, cnt max %d",
3042              cnt, ncpus2, cnt_max));
3043         return cnt;
3044 }
3045
3046 void
3047 ifq_set_maxlen(struct ifaltq *ifq, int len)
3048 {
3049         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3050 }
3051
3052 int
3053 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3054 {
3055         return ALTQ_SUBQ_INDEX_DEFAULT;
3056 }
3057
3058 int
3059 ifq_mapsubq_mask(struct ifaltq *ifq, int cpuid)
3060 {
3061         return (cpuid & ifq->altq_subq_mask);
3062 }
3063
3064 static void
3065 ifsq_watchdog(void *arg)
3066 {
3067         struct ifsubq_watchdog *wd = arg;
3068         struct ifnet *ifp;
3069
3070         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3071                 goto done;
3072
3073         ifp = ifsq_get_ifp(wd->wd_subq);
3074         if (ifnet_tryserialize_all(ifp)) {
3075                 wd->wd_watchdog(wd->wd_subq);
3076                 ifnet_deserialize_all(ifp);
3077         } else {
3078                 /* try again next timeout */
3079                 wd->wd_timer = 1;
3080         }
3081 done:
3082         ifsq_watchdog_reset(wd);
3083 }
3084
3085 static void
3086 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3087 {
3088         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3089             ifsq_get_cpuid(wd->wd_subq));
3090 }
3091
3092 void
3093 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3094     ifsq_watchdog_t watchdog)
3095 {
3096         callout_init_mp(&wd->wd_callout);
3097         wd->wd_timer = 0;
3098         wd->wd_subq = ifsq;
3099         wd->wd_watchdog = watchdog;
3100 }
3101
3102 void
3103 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3104 {
3105         wd->wd_timer = 0;
3106         ifsq_watchdog_reset(wd);
3107 }
3108
3109 void
3110 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3111 {
3112         wd->wd_timer = 0;
3113         callout_stop(&wd->wd_callout);
3114 }