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